xref: /titanic_41/usr/src/uts/sun4v/io/vdc.c (revision ac19272f7eb4a433cfccf2fdccc769cca5528169)
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 2007 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #pragma ident	"%Z%%M%	%I%	%E% SMI"
28 
29 /*
30  * LDoms virtual disk client (vdc) device driver
31  *
32  * This driver runs on a guest logical domain and communicates with the virtual
33  * disk server (vds) driver running on the service domain which is exporting
34  * virtualized "disks" to the guest logical domain.
35  *
36  * The driver can be divided into four sections:
37  *
38  * 1) generic device driver housekeeping
39  *	_init, _fini, attach, detach, ops structures, etc.
40  *
41  * 2) communication channel setup
42  *	Setup the communications link over the LDC channel that vdc uses to
43  *	talk to the vDisk server. Initialise the descriptor ring which
44  *	allows the LDC clients to transfer data via memory mappings.
45  *
46  * 3) Support exported to upper layers (filesystems, etc)
47  *	The upper layers call into vdc via strategy(9E) and DKIO(7I)
48  *	ioctl calls. vdc will copy the data to be written to the descriptor
49  *	ring or maps the buffer to store the data read by the vDisk
50  *	server into the descriptor ring. It then sends a message to the
51  *	vDisk server requesting it to complete the operation.
52  *
53  * 4) Handling responses from vDisk server.
54  *	The vDisk server will ACK some or all of the messages vdc sends to it
55  *	(this is configured during the handshake). Upon receipt of an ACK
56  *	vdc will check the descriptor ring and signal to the upper layer
57  *	code waiting on the IO.
58  */
59 
60 #include <sys/atomic.h>
61 #include <sys/conf.h>
62 #include <sys/disp.h>
63 #include <sys/ddi.h>
64 #include <sys/dkio.h>
65 #include <sys/efi_partition.h>
66 #include <sys/fcntl.h>
67 #include <sys/file.h>
68 #include <sys/mach_descrip.h>
69 #include <sys/modctl.h>
70 #include <sys/mdeg.h>
71 #include <sys/note.h>
72 #include <sys/open.h>
73 #include <sys/sdt.h>
74 #include <sys/stat.h>
75 #include <sys/sunddi.h>
76 #include <sys/types.h>
77 #include <sys/promif.h>
78 #include <sys/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/scsi/generic/sense.h>
86 #include <sys/scsi/impl/uscsi.h>	/* Needed for defn of USCSICMD ioctl */
87 
88 #include <sys/ldoms.h>
89 #include <sys/ldc.h>
90 #include <sys/vio_common.h>
91 #include <sys/vio_mailbox.h>
92 #include <sys/vio_util.h>
93 #include <sys/vdsk_common.h>
94 #include <sys/vdsk_mailbox.h>
95 #include <sys/vdc.h>
96 
97 /*
98  * function prototypes
99  */
100 
101 /* standard driver functions */
102 static int	vdc_open(dev_t *dev, int flag, int otyp, cred_t *cred);
103 static int	vdc_close(dev_t dev, int flag, int otyp, cred_t *cred);
104 static int	vdc_strategy(struct buf *buf);
105 static int	vdc_print(dev_t dev, char *str);
106 static int	vdc_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblk);
107 static int	vdc_read(dev_t dev, struct uio *uio, cred_t *cred);
108 static int	vdc_write(dev_t dev, struct uio *uio, cred_t *cred);
109 static int	vdc_ioctl(dev_t dev, int cmd, intptr_t arg, int mode,
110 			cred_t *credp, int *rvalp);
111 static int	vdc_aread(dev_t dev, struct aio_req *aio, cred_t *cred);
112 static int	vdc_awrite(dev_t dev, struct aio_req *aio, cred_t *cred);
113 
114 static int	vdc_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd,
115 			void *arg, void **resultp);
116 static int	vdc_attach(dev_info_t *dip, ddi_attach_cmd_t cmd);
117 static int	vdc_detach(dev_info_t *dip, ddi_detach_cmd_t cmd);
118 
119 /* setup */
120 static void	vdc_min(struct buf *bufp);
121 static int	vdc_send(vdc_t *vdc, caddr_t pkt, size_t *msglen);
122 static int	vdc_do_ldc_init(vdc_t *vdc, md_t *mdp, mde_cookie_t vd_node);
123 static int	vdc_start_ldc_connection(vdc_t *vdc);
124 static int	vdc_create_device_nodes(vdc_t *vdc);
125 static int	vdc_create_device_nodes_efi(vdc_t *vdc);
126 static int	vdc_create_device_nodes_vtoc(vdc_t *vdc);
127 static int	vdc_create_device_nodes_props(vdc_t *vdc);
128 static int	vdc_get_md_node(dev_info_t *dip, md_t **mdpp,
129 		    mde_cookie_t *vd_nodep, mde_cookie_t *vd_portp);
130 static int	vdc_get_ldc_id(md_t *, mde_cookie_t, uint64_t *);
131 static int	vdc_do_ldc_up(vdc_t *vdc);
132 static void	vdc_terminate_ldc(vdc_t *vdc);
133 static int	vdc_init_descriptor_ring(vdc_t *vdc);
134 static void	vdc_destroy_descriptor_ring(vdc_t *vdc);
135 static int	vdc_setup_devid(vdc_t *vdc);
136 static void	vdc_store_label_efi(vdc_t *vdc, struct dk_gpt *efi);
137 static void	vdc_store_label_vtoc(vdc_t *, struct dk_geom *, struct vtoc *);
138 static void	vdc_store_label_unk(vdc_t *vdc);
139 static boolean_t vdc_is_opened(vdc_t *vdc);
140 
141 /* handshake with vds */
142 static int		vdc_init_ver_negotiation(vdc_t *vdc, vio_ver_t ver);
143 static int		vdc_ver_negotiation(vdc_t *vdcp);
144 static int		vdc_init_attr_negotiation(vdc_t *vdc);
145 static int		vdc_attr_negotiation(vdc_t *vdcp);
146 static int		vdc_init_dring_negotiate(vdc_t *vdc);
147 static int		vdc_dring_negotiation(vdc_t *vdcp);
148 static int		vdc_send_rdx(vdc_t *vdcp);
149 static int		vdc_rdx_exchange(vdc_t *vdcp);
150 static boolean_t	vdc_is_supported_version(vio_ver_msg_t *ver_msg);
151 
152 /* processing incoming messages from vDisk server */
153 static void	vdc_process_msg_thread(vdc_t *vdc);
154 static int	vdc_recv(vdc_t *vdc, vio_msg_t *msgp, size_t *nbytesp);
155 
156 static uint_t	vdc_handle_cb(uint64_t event, caddr_t arg);
157 static int	vdc_process_data_msg(vdc_t *vdc, vio_msg_t *msg);
158 static int	vdc_handle_ver_msg(vdc_t *vdc, vio_ver_msg_t *ver_msg);
159 static int	vdc_handle_attr_msg(vdc_t *vdc, vd_attr_msg_t *attr_msg);
160 static int	vdc_handle_dring_reg_msg(vdc_t *vdc, vio_dring_reg_msg_t *msg);
161 static int 	vdc_send_request(vdc_t *vdcp, int operation,
162 		    caddr_t addr, size_t nbytes, int slice, diskaddr_t offset,
163 		    int cb_type, void *cb_arg, vio_desc_direction_t dir);
164 static int	vdc_map_to_shared_dring(vdc_t *vdcp, int idx);
165 static int 	vdc_populate_descriptor(vdc_t *vdcp, int operation,
166 		    caddr_t addr, size_t nbytes, int slice, diskaddr_t offset,
167 		    int cb_type, void *cb_arg, vio_desc_direction_t dir);
168 static int 	vdc_do_sync_op(vdc_t *vdcp, int operation,
169 		    caddr_t addr, size_t nbytes, int slice, diskaddr_t offset,
170 		    int cb_type, void *cb_arg, vio_desc_direction_t dir);
171 
172 static int	vdc_wait_for_response(vdc_t *vdcp, vio_msg_t *msgp);
173 static int	vdc_drain_response(vdc_t *vdcp);
174 static int	vdc_depopulate_descriptor(vdc_t *vdc, uint_t idx);
175 static int	vdc_populate_mem_hdl(vdc_t *vdcp, vdc_local_desc_t *ldep);
176 static int	vdc_verify_seq_num(vdc_t *vdc, vio_dring_msg_t *dring_msg);
177 
178 /* dkio */
179 static int	vd_process_ioctl(dev_t dev, int cmd, caddr_t arg, int mode);
180 static void	vdc_create_fake_geometry(vdc_t *vdc);
181 static int	vdc_validate_geometry(vdc_t *vdc);
182 static void	vdc_validate(vdc_t *vdc);
183 static void	vdc_validate_task(void *arg);
184 static int	vdc_null_copy_func(vdc_t *vdc, void *from, void *to,
185 		    int mode, int dir);
186 static int	vdc_get_wce_convert(vdc_t *vdc, void *from, void *to,
187 		    int mode, int dir);
188 static int	vdc_set_wce_convert(vdc_t *vdc, void *from, void *to,
189 		    int mode, int dir);
190 static int	vdc_get_vtoc_convert(vdc_t *vdc, void *from, void *to,
191 		    int mode, int dir);
192 static int	vdc_set_vtoc_convert(vdc_t *vdc, void *from, void *to,
193 		    int mode, int dir);
194 static int	vdc_get_geom_convert(vdc_t *vdc, void *from, void *to,
195 		    int mode, int dir);
196 static int	vdc_set_geom_convert(vdc_t *vdc, void *from, void *to,
197 		    int mode, int dir);
198 static int	vdc_get_efi_convert(vdc_t *vdc, void *from, void *to,
199 		    int mode, int dir);
200 static int	vdc_set_efi_convert(vdc_t *vdc, void *from, void *to,
201 		    int mode, int dir);
202 
203 /*
204  * Module variables
205  */
206 
207 /*
208  * Tunable variables to control how long vdc waits before timing out on
209  * various operations
210  */
211 static int	vdc_hshake_retries = 3;
212 
213 static int	vdc_timeout = 0; /* units: seconds */
214 
215 static uint64_t vdc_hz_min_ldc_delay;
216 static uint64_t vdc_min_timeout_ldc = 1 * MILLISEC;
217 static uint64_t vdc_hz_max_ldc_delay;
218 static uint64_t vdc_max_timeout_ldc = 100 * MILLISEC;
219 
220 static uint64_t vdc_ldc_read_init_delay = 1 * MILLISEC;
221 static uint64_t vdc_ldc_read_max_delay = 100 * MILLISEC;
222 
223 /* values for dumping - need to run in a tighter loop */
224 static uint64_t	vdc_usec_timeout_dump = 100 * MILLISEC;	/* 0.1s units: ns */
225 static int	vdc_dump_retries = 100;
226 
227 /* Count of the number of vdc instances attached */
228 static volatile uint32_t	vdc_instance_count = 0;
229 
230 /* Soft state pointer */
231 static void	*vdc_state;
232 
233 /*
234  * Controlling the verbosity of the error/debug messages
235  *
236  * vdc_msglevel - controls level of messages
237  * vdc_matchinst - 64-bit variable where each bit corresponds
238  *                 to the vdc instance the vdc_msglevel applies.
239  */
240 int		vdc_msglevel = 0x0;
241 uint64_t	vdc_matchinst = 0ull;
242 
243 /*
244  * Supported vDisk protocol version pairs.
245  *
246  * The first array entry is the latest and preferred version.
247  */
248 static const vio_ver_t	vdc_version[] = {{1, 1}};
249 
250 static struct cb_ops vdc_cb_ops = {
251 	vdc_open,	/* cb_open */
252 	vdc_close,	/* cb_close */
253 	vdc_strategy,	/* cb_strategy */
254 	vdc_print,	/* cb_print */
255 	vdc_dump,	/* cb_dump */
256 	vdc_read,	/* cb_read */
257 	vdc_write,	/* cb_write */
258 	vdc_ioctl,	/* cb_ioctl */
259 	nodev,		/* cb_devmap */
260 	nodev,		/* cb_mmap */
261 	nodev,		/* cb_segmap */
262 	nochpoll,	/* cb_chpoll */
263 	ddi_prop_op,	/* cb_prop_op */
264 	NULL,		/* cb_str */
265 	D_MP | D_64BIT,	/* cb_flag */
266 	CB_REV,		/* cb_rev */
267 	vdc_aread,	/* cb_aread */
268 	vdc_awrite	/* cb_awrite */
269 };
270 
271 static struct dev_ops vdc_ops = {
272 	DEVO_REV,	/* devo_rev */
273 	0,		/* devo_refcnt */
274 	vdc_getinfo,	/* devo_getinfo */
275 	nulldev,	/* devo_identify */
276 	nulldev,	/* devo_probe */
277 	vdc_attach,	/* devo_attach */
278 	vdc_detach,	/* devo_detach */
279 	nodev,		/* devo_reset */
280 	&vdc_cb_ops,	/* devo_cb_ops */
281 	NULL,		/* devo_bus_ops */
282 	nulldev		/* devo_power */
283 };
284 
285 static struct modldrv modldrv = {
286 	&mod_driverops,
287 	"virtual disk client",
288 	&vdc_ops,
289 };
290 
291 static struct modlinkage modlinkage = {
292 	MODREV_1,
293 	&modldrv,
294 	NULL
295 };
296 
297 /* -------------------------------------------------------------------------- */
298 
299 /*
300  * Device Driver housekeeping and setup
301  */
302 
303 int
304 _init(void)
305 {
306 	int	status;
307 
308 	if ((status = ddi_soft_state_init(&vdc_state, sizeof (vdc_t), 1)) != 0)
309 		return (status);
310 	if ((status = mod_install(&modlinkage)) != 0)
311 		ddi_soft_state_fini(&vdc_state);
312 	vdc_efi_init(vd_process_ioctl);
313 	return (status);
314 }
315 
316 int
317 _info(struct modinfo *modinfop)
318 {
319 	return (mod_info(&modlinkage, modinfop));
320 }
321 
322 int
323 _fini(void)
324 {
325 	int	status;
326 
327 	if ((status = mod_remove(&modlinkage)) != 0)
328 		return (status);
329 	vdc_efi_fini();
330 	ddi_soft_state_fini(&vdc_state);
331 	return (0);
332 }
333 
334 static int
335 vdc_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd,  void *arg, void **resultp)
336 {
337 	_NOTE(ARGUNUSED(dip))
338 
339 	int	instance = VDCUNIT((dev_t)arg);
340 	vdc_t	*vdc = NULL;
341 
342 	switch (cmd) {
343 	case DDI_INFO_DEVT2DEVINFO:
344 		if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
345 			*resultp = NULL;
346 			return (DDI_FAILURE);
347 		}
348 		*resultp = vdc->dip;
349 		return (DDI_SUCCESS);
350 	case DDI_INFO_DEVT2INSTANCE:
351 		*resultp = (void *)(uintptr_t)instance;
352 		return (DDI_SUCCESS);
353 	default:
354 		*resultp = NULL;
355 		return (DDI_FAILURE);
356 	}
357 }
358 
359 static int
360 vdc_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
361 {
362 	int	instance;
363 	int	rv;
364 	vdc_t	*vdc = NULL;
365 
366 	switch (cmd) {
367 	case DDI_DETACH:
368 		/* the real work happens below */
369 		break;
370 	case DDI_SUSPEND:
371 		/* nothing to do for this non-device */
372 		return (DDI_SUCCESS);
373 	default:
374 		return (DDI_FAILURE);
375 	}
376 
377 	ASSERT(cmd == DDI_DETACH);
378 	instance = ddi_get_instance(dip);
379 	DMSGX(1, "[%d] Entered\n", instance);
380 
381 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
382 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
383 		return (DDI_FAILURE);
384 	}
385 
386 	if (vdc_is_opened(vdc)) {
387 		DMSG(vdc, 0, "[%d] Cannot detach: device is open", instance);
388 		return (DDI_FAILURE);
389 	}
390 
391 	if (vdc->dkio_flush_pending) {
392 		DMSG(vdc, 0,
393 		    "[%d] Cannot detach: %d outstanding DKIO flushes\n",
394 		    instance, vdc->dkio_flush_pending);
395 		return (DDI_FAILURE);
396 	}
397 
398 	if (vdc->validate_pending) {
399 		DMSG(vdc, 0,
400 		    "[%d] Cannot detach: %d outstanding validate request\n",
401 		    instance, vdc->validate_pending);
402 		return (DDI_FAILURE);
403 	}
404 
405 	DMSG(vdc, 0, "[%d] proceeding...\n", instance);
406 
407 	/* mark instance as detaching */
408 	vdc->lifecycle	= VDC_LC_DETACHING;
409 
410 	/*
411 	 * try and disable callbacks to prevent another handshake
412 	 */
413 	rv = ldc_set_cb_mode(vdc->ldc_handle, LDC_CB_DISABLE);
414 	DMSG(vdc, 0, "callback disabled (rv=%d)\n", rv);
415 
416 	if (vdc->initialized & VDC_THREAD) {
417 		mutex_enter(&vdc->read_lock);
418 		if ((vdc->read_state == VDC_READ_WAITING) ||
419 		    (vdc->read_state == VDC_READ_RESET)) {
420 			vdc->read_state = VDC_READ_RESET;
421 			cv_signal(&vdc->read_cv);
422 		}
423 
424 		mutex_exit(&vdc->read_lock);
425 
426 		/* wake up any thread waiting for connection to come online */
427 		mutex_enter(&vdc->lock);
428 		if (vdc->state == VDC_STATE_INIT_WAITING) {
429 			DMSG(vdc, 0,
430 			    "[%d] write reset - move to resetting state...\n",
431 			    instance);
432 			vdc->state = VDC_STATE_RESETTING;
433 			cv_signal(&vdc->initwait_cv);
434 		}
435 		mutex_exit(&vdc->lock);
436 
437 		/* now wait until state transitions to VDC_STATE_DETACH */
438 		thread_join(vdc->msg_proc_thr->t_did);
439 		ASSERT(vdc->state == VDC_STATE_DETACH);
440 		DMSG(vdc, 0, "[%d] Reset thread exit and join ..\n",
441 		    vdc->instance);
442 	}
443 
444 	mutex_enter(&vdc->lock);
445 
446 	if (vdc->initialized & VDC_DRING)
447 		vdc_destroy_descriptor_ring(vdc);
448 
449 	if (vdc->initialized & VDC_LDC)
450 		vdc_terminate_ldc(vdc);
451 
452 	mutex_exit(&vdc->lock);
453 
454 	if (vdc->initialized & VDC_MINOR) {
455 		ddi_prop_remove_all(dip);
456 		ddi_remove_minor_node(dip, NULL);
457 	}
458 
459 	if (vdc->initialized & VDC_LOCKS) {
460 		mutex_destroy(&vdc->lock);
461 		mutex_destroy(&vdc->read_lock);
462 		cv_destroy(&vdc->initwait_cv);
463 		cv_destroy(&vdc->dring_free_cv);
464 		cv_destroy(&vdc->membind_cv);
465 		cv_destroy(&vdc->sync_pending_cv);
466 		cv_destroy(&vdc->sync_blocked_cv);
467 		cv_destroy(&vdc->read_cv);
468 		cv_destroy(&vdc->running_cv);
469 	}
470 
471 	if (vdc->minfo)
472 		kmem_free(vdc->minfo, sizeof (struct dk_minfo));
473 
474 	if (vdc->cinfo)
475 		kmem_free(vdc->cinfo, sizeof (struct dk_cinfo));
476 
477 	if (vdc->vtoc)
478 		kmem_free(vdc->vtoc, sizeof (struct vtoc));
479 
480 	if (vdc->geom)
481 		kmem_free(vdc->geom, sizeof (struct dk_geom));
482 
483 	if (vdc->devid) {
484 		ddi_devid_unregister(dip);
485 		ddi_devid_free(vdc->devid);
486 	}
487 
488 	if (vdc->initialized & VDC_SOFT_STATE)
489 		ddi_soft_state_free(vdc_state, instance);
490 
491 	DMSG(vdc, 0, "[%d] End %p\n", instance, (void *)vdc);
492 
493 	return (DDI_SUCCESS);
494 }
495 
496 
497 static int
498 vdc_do_attach(dev_info_t *dip)
499 {
500 	int		instance;
501 	vdc_t		*vdc = NULL;
502 	int		status;
503 	md_t		*mdp;
504 	mde_cookie_t	vd_node, vd_port;
505 
506 	ASSERT(dip != NULL);
507 
508 	instance = ddi_get_instance(dip);
509 	if (ddi_soft_state_zalloc(vdc_state, instance) != DDI_SUCCESS) {
510 		cmn_err(CE_NOTE, "[%d] Couldn't alloc state structure",
511 		    instance);
512 		return (DDI_FAILURE);
513 	}
514 
515 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
516 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
517 		return (DDI_FAILURE);
518 	}
519 
520 	/*
521 	 * We assign the value to initialized in this case to zero out the
522 	 * variable and then set bits in it to indicate what has been done
523 	 */
524 	vdc->initialized = VDC_SOFT_STATE;
525 
526 	vdc_hz_min_ldc_delay = drv_usectohz(vdc_min_timeout_ldc);
527 	vdc_hz_max_ldc_delay = drv_usectohz(vdc_max_timeout_ldc);
528 
529 	vdc->dip	= dip;
530 	vdc->instance	= instance;
531 	vdc->vdisk_type	= VD_DISK_TYPE_UNK;
532 	vdc->vdisk_label = VD_DISK_LABEL_UNK;
533 	vdc->state	= VDC_STATE_INIT;
534 	vdc->lifecycle	= VDC_LC_ATTACHING;
535 	vdc->ldc_state	= 0;
536 	vdc->session_id = 0;
537 	vdc->block_size = DEV_BSIZE;
538 	vdc->max_xfer_sz = maxphys / DEV_BSIZE;
539 
540 	/*
541 	 * We assume, for now, that the vDisk server will export 'read'
542 	 * operations to us at a minimum (this is needed because of checks
543 	 * in vdc for supported operations early in the handshake process).
544 	 * The vDisk server will return ENOTSUP if this is not the case.
545 	 * The value will be overwritten during the attribute exchange with
546 	 * the bitmask of operations exported by server.
547 	 */
548 	vdc->operations = VD_OP_MASK_READ;
549 
550 	vdc->vtoc = NULL;
551 	vdc->geom = NULL;
552 	vdc->cinfo = NULL;
553 	vdc->minfo = NULL;
554 
555 	mutex_init(&vdc->lock, NULL, MUTEX_DRIVER, NULL);
556 	cv_init(&vdc->initwait_cv, NULL, CV_DRIVER, NULL);
557 	cv_init(&vdc->dring_free_cv, NULL, CV_DRIVER, NULL);
558 	cv_init(&vdc->membind_cv, NULL, CV_DRIVER, NULL);
559 	cv_init(&vdc->running_cv, NULL, CV_DRIVER, NULL);
560 
561 	vdc->threads_pending = 0;
562 	vdc->sync_op_pending = B_FALSE;
563 	vdc->sync_op_blocked = B_FALSE;
564 	cv_init(&vdc->sync_pending_cv, NULL, CV_DRIVER, NULL);
565 	cv_init(&vdc->sync_blocked_cv, NULL, CV_DRIVER, NULL);
566 
567 	/* init blocking msg read functionality */
568 	mutex_init(&vdc->read_lock, NULL, MUTEX_DRIVER, NULL);
569 	cv_init(&vdc->read_cv, NULL, CV_DRIVER, NULL);
570 	vdc->read_state = VDC_READ_IDLE;
571 
572 	vdc->initialized |= VDC_LOCKS;
573 
574 	/* get device and port MD node for this disk instance */
575 	if (vdc_get_md_node(dip, &mdp, &vd_node, &vd_port) != 0) {
576 		cmn_err(CE_NOTE, "[%d] Could not get machine description node",
577 		    instance);
578 		return (DDI_FAILURE);
579 	}
580 
581 	/* set the connection timeout */
582 	if (vd_port == NULL || (md_get_prop_val(mdp, vd_port,
583 	    VDC_MD_TIMEOUT, &vdc->ctimeout) != 0)) {
584 		vdc->ctimeout = 0;
585 	}
586 
587 	/* initialise LDC channel which will be used to communicate with vds */
588 	status = vdc_do_ldc_init(vdc, mdp, vd_node);
589 
590 	(void) md_fini_handle(mdp);
591 
592 	if (status != 0) {
593 		cmn_err(CE_NOTE, "[%d] Couldn't initialize LDC", instance);
594 		goto return_status;
595 	}
596 
597 	/* initialize the thread responsible for managing state with server */
598 	vdc->msg_proc_thr = thread_create(NULL, 0, vdc_process_msg_thread,
599 	    vdc, 0, &p0, TS_RUN, minclsyspri);
600 	if (vdc->msg_proc_thr == NULL) {
601 		cmn_err(CE_NOTE, "[%d] Failed to create msg processing thread",
602 		    instance);
603 		return (DDI_FAILURE);
604 	}
605 
606 	vdc->initialized |= VDC_THREAD;
607 
608 	atomic_inc_32(&vdc_instance_count);
609 
610 	/*
611 	 * Check the disk label. This will send requests and do the handshake.
612 	 * We don't really care about the disk label now. What we really need is
613 	 * the handshake do be done so that we know the type of the disk (slice
614 	 * or full disk) and the appropriate device nodes can be created.
615 	 */
616 	vdc->vdisk_label = VD_DISK_LABEL_UNK;
617 	vdc->vtoc = kmem_zalloc(sizeof (struct vtoc), KM_SLEEP);
618 	vdc->geom = kmem_zalloc(sizeof (struct dk_geom), KM_SLEEP);
619 	vdc->minfo = kmem_zalloc(sizeof (struct dk_minfo), KM_SLEEP);
620 
621 	mutex_enter(&vdc->lock);
622 	(void) vdc_validate_geometry(vdc);
623 	mutex_exit(&vdc->lock);
624 
625 	/*
626 	 * Now that we have the device info we can create the
627 	 * device nodes and properties
628 	 */
629 	status = vdc_create_device_nodes(vdc);
630 	if (status) {
631 		DMSG(vdc, 0, "[%d] Failed to create device nodes",
632 		    instance);
633 		goto return_status;
634 	}
635 	status = vdc_create_device_nodes_props(vdc);
636 	if (status) {
637 		DMSG(vdc, 0, "[%d] Failed to create device nodes"
638 		    " properties (%d)", instance, status);
639 		goto return_status;
640 	}
641 
642 	/*
643 	 * Setup devid
644 	 */
645 	if (vdc_setup_devid(vdc)) {
646 		DMSG(vdc, 0, "[%d] No device id available\n", instance);
647 	}
648 
649 	ddi_report_dev(dip);
650 	vdc->lifecycle	= VDC_LC_ONLINE;
651 	DMSG(vdc, 0, "[%d] Attach tasks successful\n", instance);
652 
653 return_status:
654 	DMSG(vdc, 0, "[%d] Attach completed\n", instance);
655 	return (status);
656 }
657 
658 static int
659 vdc_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
660 {
661 	int	status;
662 
663 	switch (cmd) {
664 	case DDI_ATTACH:
665 		if ((status = vdc_do_attach(dip)) != 0)
666 			(void) vdc_detach(dip, DDI_DETACH);
667 		return (status);
668 	case DDI_RESUME:
669 		/* nothing to do for this non-device */
670 		return (DDI_SUCCESS);
671 	default:
672 		return (DDI_FAILURE);
673 	}
674 }
675 
676 static int
677 vdc_do_ldc_init(vdc_t *vdc, md_t *mdp, mde_cookie_t vd_node)
678 {
679 	int			status = 0;
680 	ldc_status_t		ldc_state;
681 	ldc_attr_t		ldc_attr;
682 	uint64_t		ldc_id = 0;
683 
684 	ASSERT(vdc != NULL);
685 
686 	vdc->initialized |= VDC_LDC;
687 
688 	if ((status = vdc_get_ldc_id(mdp, vd_node, &ldc_id)) != 0) {
689 		DMSG(vdc, 0, "[%d] Failed to get LDC channel ID property",
690 		    vdc->instance);
691 		return (EIO);
692 	}
693 
694 	DMSGX(0, "[%d] LDC id is 0x%lx\n", vdc->instance, ldc_id);
695 
696 	vdc->ldc_id = ldc_id;
697 
698 	ldc_attr.devclass = LDC_DEV_BLK;
699 	ldc_attr.instance = vdc->instance;
700 	ldc_attr.mode = LDC_MODE_UNRELIABLE;	/* unreliable transport */
701 	ldc_attr.mtu = VD_LDC_MTU;
702 
703 	if ((vdc->initialized & VDC_LDC_INIT) == 0) {
704 		status = ldc_init(ldc_id, &ldc_attr, &vdc->ldc_handle);
705 		if (status != 0) {
706 			DMSG(vdc, 0, "[%d] ldc_init(chan %ld) returned %d",
707 			    vdc->instance, ldc_id, status);
708 			return (status);
709 		}
710 		vdc->initialized |= VDC_LDC_INIT;
711 	}
712 	status = ldc_status(vdc->ldc_handle, &ldc_state);
713 	if (status != 0) {
714 		DMSG(vdc, 0, "[%d] Cannot discover LDC status [err=%d]",
715 		    vdc->instance, status);
716 		return (status);
717 	}
718 	vdc->ldc_state = ldc_state;
719 
720 	if ((vdc->initialized & VDC_LDC_CB) == 0) {
721 		status = ldc_reg_callback(vdc->ldc_handle, vdc_handle_cb,
722 		    (caddr_t)vdc);
723 		if (status != 0) {
724 			DMSG(vdc, 0, "[%d] LDC callback reg. failed (%d)",
725 			    vdc->instance, status);
726 			return (status);
727 		}
728 		vdc->initialized |= VDC_LDC_CB;
729 	}
730 
731 	vdc->initialized |= VDC_LDC;
732 
733 	/*
734 	 * At this stage we have initialised LDC, we will now try and open
735 	 * the connection.
736 	 */
737 	if (vdc->ldc_state == LDC_INIT) {
738 		status = ldc_open(vdc->ldc_handle);
739 		if (status != 0) {
740 			DMSG(vdc, 0, "[%d] ldc_open(chan %ld) returned %d",
741 			    vdc->instance, vdc->ldc_id, status);
742 			return (status);
743 		}
744 		vdc->initialized |= VDC_LDC_OPEN;
745 	}
746 
747 	return (status);
748 }
749 
750 static int
751 vdc_start_ldc_connection(vdc_t *vdc)
752 {
753 	int		status = 0;
754 
755 	ASSERT(vdc != NULL);
756 
757 	ASSERT(MUTEX_HELD(&vdc->lock));
758 
759 	status = vdc_do_ldc_up(vdc);
760 
761 	DMSG(vdc, 0, "[%d] Finished bringing up LDC\n", vdc->instance);
762 
763 	return (status);
764 }
765 
766 static int
767 vdc_stop_ldc_connection(vdc_t *vdcp)
768 {
769 	int	status;
770 
771 	DMSG(vdcp, 0, ": Resetting connection to vDisk server : state %d\n",
772 	    vdcp->state);
773 
774 	status = ldc_down(vdcp->ldc_handle);
775 	DMSG(vdcp, 0, "ldc_down() = %d\n", status);
776 
777 	vdcp->initialized &= ~VDC_HANDSHAKE;
778 	DMSG(vdcp, 0, "initialized=%x\n", vdcp->initialized);
779 
780 	return (status);
781 }
782 
783 static int
784 vdc_create_device_nodes_efi(vdc_t *vdc)
785 {
786 	ddi_remove_minor_node(vdc->dip, "h");
787 	ddi_remove_minor_node(vdc->dip, "h,raw");
788 
789 	if (ddi_create_minor_node(vdc->dip, "wd", S_IFBLK,
790 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
791 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
792 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'wd'",
793 		    vdc->instance);
794 		return (EIO);
795 	}
796 
797 	/* if any device node is created we set this flag */
798 	vdc->initialized |= VDC_MINOR;
799 
800 	if (ddi_create_minor_node(vdc->dip, "wd,raw", S_IFCHR,
801 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
802 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
803 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'wd,raw'",
804 		    vdc->instance);
805 		return (EIO);
806 	}
807 
808 	return (0);
809 }
810 
811 static int
812 vdc_create_device_nodes_vtoc(vdc_t *vdc)
813 {
814 	ddi_remove_minor_node(vdc->dip, "wd");
815 	ddi_remove_minor_node(vdc->dip, "wd,raw");
816 
817 	if (ddi_create_minor_node(vdc->dip, "h", S_IFBLK,
818 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
819 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
820 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'h'",
821 		    vdc->instance);
822 		return (EIO);
823 	}
824 
825 	/* if any device node is created we set this flag */
826 	vdc->initialized |= VDC_MINOR;
827 
828 	if (ddi_create_minor_node(vdc->dip, "h,raw", S_IFCHR,
829 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
830 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
831 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'h,raw'",
832 		    vdc->instance);
833 		return (EIO);
834 	}
835 
836 	return (0);
837 }
838 
839 /*
840  * Function:
841  *	vdc_create_device_nodes
842  *
843  * Description:
844  *	This function creates the block and character device nodes under
845  *	/devices along with the node properties. It is called as part of
846  *	the attach(9E) of the instance during the handshake with vds after
847  *	vds has sent the attributes to vdc.
848  *
849  *	If the device is of type VD_DISK_TYPE_SLICE then the minor node
850  *	of 2 is used in keeping with the Solaris convention that slice 2
851  *	refers to a whole disk. Slices start at 'a'
852  *
853  * Parameters:
854  *	vdc 		- soft state pointer
855  *
856  * Return Values
857  *	0		- Success
858  *	EIO		- Failed to create node
859  *	EINVAL		- Unknown type of disk exported
860  */
861 static int
862 vdc_create_device_nodes(vdc_t *vdc)
863 {
864 	char		name[sizeof ("s,raw")];
865 	dev_info_t	*dip = NULL;
866 	int		instance, status;
867 	int		num_slices = 1;
868 	int		i;
869 
870 	ASSERT(vdc != NULL);
871 
872 	instance = vdc->instance;
873 	dip = vdc->dip;
874 
875 	switch (vdc->vdisk_type) {
876 	case VD_DISK_TYPE_DISK:
877 		num_slices = V_NUMPAR;
878 		break;
879 	case VD_DISK_TYPE_SLICE:
880 		num_slices = 1;
881 		break;
882 	case VD_DISK_TYPE_UNK:
883 	default:
884 		return (EINVAL);
885 	}
886 
887 	/*
888 	 * Minor nodes are different for EFI disks: EFI disks do not have
889 	 * a minor node 'g' for the minor number corresponding to slice
890 	 * VD_EFI_WD_SLICE (slice 7) instead they have a minor node 'wd'
891 	 * representing the whole disk.
892 	 */
893 	for (i = 0; i < num_slices; i++) {
894 
895 		if (i == VD_EFI_WD_SLICE) {
896 			if (vdc->vdisk_label == VD_DISK_LABEL_EFI)
897 				status = vdc_create_device_nodes_efi(vdc);
898 			else
899 				status = vdc_create_device_nodes_vtoc(vdc);
900 			if (status != 0)
901 				return (status);
902 			continue;
903 		}
904 
905 		(void) snprintf(name, sizeof (name), "%c", 'a' + i);
906 		if (ddi_create_minor_node(dip, name, S_IFBLK,
907 		    VD_MAKE_DEV(instance, i), DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
908 			cmn_err(CE_NOTE, "[%d] Couldn't add block node '%s'",
909 			    instance, name);
910 			return (EIO);
911 		}
912 
913 		/* if any device node is created we set this flag */
914 		vdc->initialized |= VDC_MINOR;
915 
916 		(void) snprintf(name, sizeof (name), "%c%s", 'a' + i, ",raw");
917 
918 		if (ddi_create_minor_node(dip, name, S_IFCHR,
919 		    VD_MAKE_DEV(instance, i), DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
920 			cmn_err(CE_NOTE, "[%d] Couldn't add raw node '%s'",
921 			    instance, name);
922 			return (EIO);
923 		}
924 	}
925 
926 	return (0);
927 }
928 
929 /*
930  * Function:
931  *	vdc_create_device_nodes_props
932  *
933  * Description:
934  *	This function creates the block and character device nodes under
935  *	/devices along with the node properties. It is called as part of
936  *	the attach(9E) of the instance during the handshake with vds after
937  *	vds has sent the attributes to vdc.
938  *
939  * Parameters:
940  *	vdc 		- soft state pointer
941  *
942  * Return Values
943  *	0		- Success
944  *	EIO		- Failed to create device node property
945  *	EINVAL		- Unknown type of disk exported
946  */
947 static int
948 vdc_create_device_nodes_props(vdc_t *vdc)
949 {
950 	dev_info_t	*dip = NULL;
951 	int		instance;
952 	int		num_slices = 1;
953 	int64_t		size = 0;
954 	dev_t		dev;
955 	int		rv;
956 	int		i;
957 
958 	ASSERT(vdc != NULL);
959 	ASSERT(vdc->vtoc != NULL);
960 
961 	instance = vdc->instance;
962 	dip = vdc->dip;
963 
964 	switch (vdc->vdisk_type) {
965 	case VD_DISK_TYPE_DISK:
966 		num_slices = V_NUMPAR;
967 		break;
968 	case VD_DISK_TYPE_SLICE:
969 		num_slices = 1;
970 		break;
971 	case VD_DISK_TYPE_UNK:
972 	default:
973 		return (EINVAL);
974 	}
975 
976 	if (vdc->vdisk_label == VD_DISK_LABEL_UNK) {
977 		/* remove all properties */
978 		for (i = 0; i < num_slices; i++) {
979 			dev = makedevice(ddi_driver_major(dip),
980 			    VD_MAKE_DEV(instance, i));
981 			(void) ddi_prop_remove(dev, dip, VDC_SIZE_PROP_NAME);
982 			(void) ddi_prop_remove(dev, dip, VDC_NBLOCKS_PROP_NAME);
983 		}
984 		return (0);
985 	}
986 
987 	for (i = 0; i < num_slices; i++) {
988 		dev = makedevice(ddi_driver_major(dip),
989 		    VD_MAKE_DEV(instance, i));
990 
991 		size = vdc->vtoc->v_part[i].p_size * vdc->vtoc->v_sectorsz;
992 		DMSG(vdc, 0, "[%d] sz %ld (%ld Mb)  p_size %lx\n",
993 		    instance, size, size / (1024 * 1024),
994 		    vdc->vtoc->v_part[i].p_size);
995 
996 		rv = ddi_prop_update_int64(dev, dip, VDC_SIZE_PROP_NAME, size);
997 		if (rv != DDI_PROP_SUCCESS) {
998 			cmn_err(CE_NOTE, "[%d] Couldn't add '%s' prop of [%ld]",
999 			    instance, VDC_SIZE_PROP_NAME, size);
1000 			return (EIO);
1001 		}
1002 
1003 		rv = ddi_prop_update_int64(dev, dip, VDC_NBLOCKS_PROP_NAME,
1004 		    lbtodb(size));
1005 		if (rv != DDI_PROP_SUCCESS) {
1006 			cmn_err(CE_NOTE, "[%d] Couldn't add '%s' prop [%llu]",
1007 			    instance, VDC_NBLOCKS_PROP_NAME, lbtodb(size));
1008 			return (EIO);
1009 		}
1010 	}
1011 
1012 	return (0);
1013 }
1014 
1015 /*
1016  * Function:
1017  *	vdc_is_opened
1018  *
1019  * Description:
1020  *	This function checks if any slice of a given virtual disk is
1021  *	currently opened.
1022  *
1023  * Parameters:
1024  *	vdc 		- soft state pointer
1025  *
1026  * Return Values
1027  *	B_TRUE		- at least one slice is opened.
1028  *	B_FALSE		- no slice is opened.
1029  */
1030 static boolean_t
1031 vdc_is_opened(vdc_t *vdc)
1032 {
1033 	int i, nslices;
1034 
1035 	switch (vdc->vdisk_type) {
1036 	case VD_DISK_TYPE_DISK:
1037 		nslices = V_NUMPAR;
1038 		break;
1039 	case VD_DISK_TYPE_SLICE:
1040 		nslices = 1;
1041 		break;
1042 	case VD_DISK_TYPE_UNK:
1043 	default:
1044 		ASSERT(0);
1045 	}
1046 
1047 	/* check if there's any layered open */
1048 	for (i = 0; i < nslices; i++) {
1049 		if (vdc->open_lyr[i] > 0)
1050 			return (B_TRUE);
1051 	}
1052 
1053 	/* check if there is any other kind of open */
1054 	for (i = 0; i < OTYPCNT; i++) {
1055 		if (vdc->open[i] != 0)
1056 			return (B_TRUE);
1057 	}
1058 
1059 	return (B_FALSE);
1060 }
1061 
1062 static int
1063 vdc_mark_opened(vdc_t *vdc, int slice, int flag, int otyp)
1064 {
1065 	uint8_t slicemask;
1066 	int i;
1067 
1068 	ASSERT(otyp < OTYPCNT);
1069 	ASSERT(slice < V_NUMPAR);
1070 	ASSERT(MUTEX_HELD(&vdc->lock));
1071 
1072 	slicemask = 1 << slice;
1073 
1074 	/* check if slice is already exclusively opened */
1075 	if (vdc->open_excl & slicemask)
1076 		return (EBUSY);
1077 
1078 	/* if open exclusive, check if slice is already opened */
1079 	if (flag & FEXCL) {
1080 		if (vdc->open_lyr[slice] > 0)
1081 			return (EBUSY);
1082 		for (i = 0; i < OTYPCNT; i++) {
1083 			if (vdc->open[i] & slicemask)
1084 				return (EBUSY);
1085 		}
1086 		vdc->open_excl |= slicemask;
1087 	}
1088 
1089 	/* mark slice as opened */
1090 	if (otyp == OTYP_LYR) {
1091 		vdc->open_lyr[slice]++;
1092 	} else {
1093 		vdc->open[otyp] |= slicemask;
1094 	}
1095 
1096 	return (0);
1097 }
1098 
1099 static void
1100 vdc_mark_closed(vdc_t *vdc, int slice, int flag, int otyp)
1101 {
1102 	uint8_t slicemask;
1103 
1104 	ASSERT(otyp < OTYPCNT);
1105 	ASSERT(slice < V_NUMPAR);
1106 	ASSERT(MUTEX_HELD(&vdc->lock));
1107 
1108 	slicemask = 1 << slice;
1109 
1110 	if (otyp == OTYP_LYR) {
1111 		ASSERT(vdc->open_lyr[slice] > 0);
1112 		vdc->open_lyr[slice]--;
1113 	} else {
1114 		vdc->open[otyp] &= ~slicemask;
1115 	}
1116 
1117 	if (flag & FEXCL)
1118 		vdc->open_excl &= ~slicemask;
1119 }
1120 
1121 static int
1122 vdc_open(dev_t *dev, int flag, int otyp, cred_t *cred)
1123 {
1124 	_NOTE(ARGUNUSED(cred))
1125 
1126 	int	instance;
1127 	int	slice, status = 0;
1128 	vdc_t	*vdc;
1129 
1130 	ASSERT(dev != NULL);
1131 	instance = VDCUNIT(*dev);
1132 
1133 	if (otyp >= OTYPCNT)
1134 		return (EINVAL);
1135 
1136 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1137 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1138 		return (ENXIO);
1139 	}
1140 
1141 	DMSG(vdc, 0, "minor = %d flag = %x, otyp = %x\n",
1142 	    getminor(*dev), flag, otyp);
1143 
1144 	slice = VDCPART(*dev);
1145 
1146 	mutex_enter(&vdc->lock);
1147 
1148 	status = vdc_mark_opened(vdc, slice, flag, otyp);
1149 
1150 	if (status != 0) {
1151 		mutex_exit(&vdc->lock);
1152 		return (status);
1153 	}
1154 
1155 	if (flag & (FNDELAY | FNONBLOCK)) {
1156 
1157 		/* don't resubmit a validate request if there's already one */
1158 		if (vdc->validate_pending > 0) {
1159 			mutex_exit(&vdc->lock);
1160 			return (0);
1161 		}
1162 
1163 		/* call vdc_validate() asynchronously to avoid blocking */
1164 		if (taskq_dispatch(system_taskq, vdc_validate_task,
1165 		    (void *)vdc, TQ_NOSLEEP) == NULL) {
1166 			vdc_mark_closed(vdc, slice, flag, otyp);
1167 			mutex_exit(&vdc->lock);
1168 			return (ENXIO);
1169 		}
1170 
1171 		vdc->validate_pending++;
1172 		mutex_exit(&vdc->lock);
1173 		return (0);
1174 	}
1175 
1176 	mutex_exit(&vdc->lock);
1177 
1178 	vdc_validate(vdc);
1179 
1180 	mutex_enter(&vdc->lock);
1181 
1182 	if (vdc->vdisk_label == VD_DISK_LABEL_UNK ||
1183 	    vdc->vtoc->v_part[slice].p_size == 0) {
1184 		vdc_mark_closed(vdc, slice, flag, otyp);
1185 		status = EIO;
1186 	}
1187 
1188 	mutex_exit(&vdc->lock);
1189 
1190 	return (status);
1191 }
1192 
1193 static int
1194 vdc_close(dev_t dev, int flag, int otyp, cred_t *cred)
1195 {
1196 	_NOTE(ARGUNUSED(cred))
1197 
1198 	int	instance;
1199 	int	slice;
1200 	int	rv;
1201 	vdc_t	*vdc;
1202 
1203 	instance = VDCUNIT(dev);
1204 
1205 	if (otyp >= OTYPCNT)
1206 		return (EINVAL);
1207 
1208 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1209 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1210 		return (ENXIO);
1211 	}
1212 
1213 	DMSG(vdc, 0, "[%d] flag = %x, otyp = %x\n", instance, flag, otyp);
1214 
1215 	slice = VDCPART(dev);
1216 
1217 	/*
1218 	 * Attempt to flush the W$ on a close operation. If this is
1219 	 * not a supported IOCTL command or the backing device is read-only
1220 	 * do not fail the close operation.
1221 	 */
1222 	rv = vd_process_ioctl(dev, DKIOCFLUSHWRITECACHE, NULL, FKIOCTL);
1223 
1224 	if (rv != 0 && rv != ENOTSUP && rv != ENOTTY && rv != EROFS) {
1225 		DMSG(vdc, 0, "[%d] flush failed with error %d on close\n",
1226 		    instance, rv);
1227 		return (EIO);
1228 	}
1229 
1230 	mutex_enter(&vdc->lock);
1231 	vdc_mark_closed(vdc, slice, flag, otyp);
1232 	mutex_exit(&vdc->lock);
1233 
1234 	return (0);
1235 }
1236 
1237 static int
1238 vdc_ioctl(dev_t dev, int cmd, intptr_t arg, int mode, cred_t *credp, int *rvalp)
1239 {
1240 	_NOTE(ARGUNUSED(credp))
1241 	_NOTE(ARGUNUSED(rvalp))
1242 
1243 	return (vd_process_ioctl(dev, cmd, (caddr_t)arg, mode));
1244 }
1245 
1246 static int
1247 vdc_print(dev_t dev, char *str)
1248 {
1249 	cmn_err(CE_NOTE, "vdc%d:  %s", VDCUNIT(dev), str);
1250 	return (0);
1251 }
1252 
1253 static int
1254 vdc_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblk)
1255 {
1256 	int	rv;
1257 	size_t	nbytes = nblk * DEV_BSIZE;
1258 	int	instance = VDCUNIT(dev);
1259 	vdc_t	*vdc = NULL;
1260 
1261 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1262 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1263 		return (ENXIO);
1264 	}
1265 
1266 	DMSG(vdc, 2, "[%d] dump %ld bytes at block 0x%lx : addr=0x%p\n",
1267 	    instance, nbytes, blkno, (void *)addr);
1268 	rv = vdc_send_request(vdc, VD_OP_BWRITE, addr, nbytes,
1269 	    VDCPART(dev), blkno, CB_STRATEGY, 0, VIO_write_dir);
1270 	if (rv) {
1271 		DMSG(vdc, 0, "Failed to do a disk dump (err=%d)\n", rv);
1272 		return (rv);
1273 	}
1274 
1275 	if (ddi_in_panic())
1276 		(void) vdc_drain_response(vdc);
1277 
1278 	DMSG(vdc, 0, "[%d] End\n", instance);
1279 
1280 	return (0);
1281 }
1282 
1283 /* -------------------------------------------------------------------------- */
1284 
1285 /*
1286  * Disk access routines
1287  *
1288  */
1289 
1290 /*
1291  * vdc_strategy()
1292  *
1293  * Return Value:
1294  *	0:	As per strategy(9E), the strategy() function must return 0
1295  *		[ bioerror(9f) sets b_flags to the proper error code ]
1296  */
1297 static int
1298 vdc_strategy(struct buf *buf)
1299 {
1300 	int	rv = -1;
1301 	vdc_t	*vdc = NULL;
1302 	int	instance = VDCUNIT(buf->b_edev);
1303 	int	op = (buf->b_flags & B_READ) ? VD_OP_BREAD : VD_OP_BWRITE;
1304 	int	slice;
1305 
1306 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1307 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1308 		bioerror(buf, ENXIO);
1309 		biodone(buf);
1310 		return (0);
1311 	}
1312 
1313 	DMSG(vdc, 2, "[%d] %s %ld bytes at block %llx : b_addr=0x%p\n",
1314 	    instance, (buf->b_flags & B_READ) ? "Read" : "Write",
1315 	    buf->b_bcount, buf->b_lblkno, (void *)buf->b_un.b_addr);
1316 	DTRACE_IO2(vstart, buf_t *, buf, vdc_t *, vdc);
1317 
1318 	bp_mapin(buf);
1319 
1320 	if ((long)buf->b_private == VD_SLICE_NONE) {
1321 		/* I/O using an absolute disk offset */
1322 		slice = VD_SLICE_NONE;
1323 	} else {
1324 		slice = VDCPART(buf->b_edev);
1325 	}
1326 
1327 	rv = vdc_send_request(vdc, op, (caddr_t)buf->b_un.b_addr,
1328 	    buf->b_bcount, slice, buf->b_lblkno,
1329 	    CB_STRATEGY, buf, (op == VD_OP_BREAD) ? VIO_read_dir :
1330 	    VIO_write_dir);
1331 
1332 	/*
1333 	 * If the request was successfully sent, the strategy call returns and
1334 	 * the ACK handler calls the bioxxx functions when the vDisk server is
1335 	 * done.
1336 	 */
1337 	if (rv) {
1338 		DMSG(vdc, 0, "Failed to read/write (err=%d)\n", rv);
1339 		bioerror(buf, rv);
1340 		biodone(buf);
1341 	}
1342 
1343 	return (0);
1344 }
1345 
1346 /*
1347  * Function:
1348  *	vdc_min
1349  *
1350  * Description:
1351  *	Routine to limit the size of a data transfer. Used in
1352  *	conjunction with physio(9F).
1353  *
1354  * Arguments:
1355  *	bp - pointer to the indicated buf(9S) struct.
1356  *
1357  */
1358 static void
1359 vdc_min(struct buf *bufp)
1360 {
1361 	vdc_t	*vdc = NULL;
1362 	int	instance = VDCUNIT(bufp->b_edev);
1363 
1364 	vdc = ddi_get_soft_state(vdc_state, instance);
1365 	VERIFY(vdc != NULL);
1366 
1367 	if (bufp->b_bcount > (vdc->max_xfer_sz * vdc->block_size)) {
1368 		bufp->b_bcount = vdc->max_xfer_sz * vdc->block_size;
1369 	}
1370 }
1371 
1372 static int
1373 vdc_read(dev_t dev, struct uio *uio, cred_t *cred)
1374 {
1375 	_NOTE(ARGUNUSED(cred))
1376 
1377 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1378 	return (physio(vdc_strategy, NULL, dev, B_READ, vdc_min, uio));
1379 }
1380 
1381 static int
1382 vdc_write(dev_t dev, struct uio *uio, cred_t *cred)
1383 {
1384 	_NOTE(ARGUNUSED(cred))
1385 
1386 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1387 	return (physio(vdc_strategy, NULL, dev, B_WRITE, vdc_min, uio));
1388 }
1389 
1390 static int
1391 vdc_aread(dev_t dev, struct aio_req *aio, cred_t *cred)
1392 {
1393 	_NOTE(ARGUNUSED(cred))
1394 
1395 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1396 	return (aphysio(vdc_strategy, anocancel, dev, B_READ, vdc_min, aio));
1397 }
1398 
1399 static int
1400 vdc_awrite(dev_t dev, struct aio_req *aio, cred_t *cred)
1401 {
1402 	_NOTE(ARGUNUSED(cred))
1403 
1404 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1405 	return (aphysio(vdc_strategy, anocancel, dev, B_WRITE, vdc_min, aio));
1406 }
1407 
1408 
1409 /* -------------------------------------------------------------------------- */
1410 
1411 /*
1412  * Handshake support
1413  */
1414 
1415 
1416 /*
1417  * Function:
1418  *	vdc_init_ver_negotiation()
1419  *
1420  * Description:
1421  *
1422  * Arguments:
1423  *	vdc	- soft state pointer for this instance of the device driver.
1424  *
1425  * Return Code:
1426  *	0	- Success
1427  */
1428 static int
1429 vdc_init_ver_negotiation(vdc_t *vdc, vio_ver_t ver)
1430 {
1431 	vio_ver_msg_t	pkt;
1432 	size_t		msglen = sizeof (pkt);
1433 	int		status = -1;
1434 
1435 	ASSERT(vdc != NULL);
1436 	ASSERT(mutex_owned(&vdc->lock));
1437 
1438 	DMSG(vdc, 0, "[%d] Entered.\n", vdc->instance);
1439 
1440 	/*
1441 	 * set the Session ID to a unique value
1442 	 * (the lower 32 bits of the clock tick)
1443 	 */
1444 	vdc->session_id = ((uint32_t)gettick() & 0xffffffff);
1445 	DMSG(vdc, 0, "[%d] Set SID to 0x%lx\n", vdc->instance, vdc->session_id);
1446 
1447 	pkt.tag.vio_msgtype = VIO_TYPE_CTRL;
1448 	pkt.tag.vio_subtype = VIO_SUBTYPE_INFO;
1449 	pkt.tag.vio_subtype_env = VIO_VER_INFO;
1450 	pkt.tag.vio_sid = vdc->session_id;
1451 	pkt.dev_class = VDEV_DISK;
1452 	pkt.ver_major = ver.major;
1453 	pkt.ver_minor = ver.minor;
1454 
1455 	status = vdc_send(vdc, (caddr_t)&pkt, &msglen);
1456 	DMSG(vdc, 0, "[%d] Ver info sent (status = %d)\n",
1457 	    vdc->instance, status);
1458 	if ((status != 0) || (msglen != sizeof (vio_ver_msg_t))) {
1459 		DMSG(vdc, 0, "[%d] Failed to send Ver negotiation info: "
1460 		    "id(%lx) rv(%d) size(%ld)", vdc->instance, vdc->ldc_handle,
1461 		    status, msglen);
1462 		if (msglen != sizeof (vio_ver_msg_t))
1463 			status = ENOMSG;
1464 	}
1465 
1466 	return (status);
1467 }
1468 
1469 /*
1470  * Function:
1471  *	vdc_ver_negotiation()
1472  *
1473  * Description:
1474  *
1475  * Arguments:
1476  *	vdcp	- soft state pointer for this instance of the device driver.
1477  *
1478  * Return Code:
1479  *	0	- Success
1480  */
1481 static int
1482 vdc_ver_negotiation(vdc_t *vdcp)
1483 {
1484 	vio_msg_t vio_msg;
1485 	int status;
1486 
1487 	if (status = vdc_init_ver_negotiation(vdcp, vdc_version[0]))
1488 		return (status);
1489 
1490 	/* release lock and wait for response */
1491 	mutex_exit(&vdcp->lock);
1492 	status = vdc_wait_for_response(vdcp, &vio_msg);
1493 	mutex_enter(&vdcp->lock);
1494 	if (status) {
1495 		DMSG(vdcp, 0,
1496 		    "[%d] Failed waiting for Ver negotiation response, rv(%d)",
1497 		    vdcp->instance, status);
1498 		return (status);
1499 	}
1500 
1501 	/* check type and sub_type ... */
1502 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1503 	    vio_msg.tag.vio_subtype == VIO_SUBTYPE_INFO) {
1504 		DMSG(vdcp, 0, "[%d] Invalid ver negotiation response\n",
1505 		    vdcp->instance);
1506 		return (EPROTO);
1507 	}
1508 
1509 	return (vdc_handle_ver_msg(vdcp, (vio_ver_msg_t *)&vio_msg));
1510 }
1511 
1512 /*
1513  * Function:
1514  *	vdc_init_attr_negotiation()
1515  *
1516  * Description:
1517  *
1518  * Arguments:
1519  *	vdc	- soft state pointer for this instance of the device driver.
1520  *
1521  * Return Code:
1522  *	0	- Success
1523  */
1524 static int
1525 vdc_init_attr_negotiation(vdc_t *vdc)
1526 {
1527 	vd_attr_msg_t	pkt;
1528 	size_t		msglen = sizeof (pkt);
1529 	int		status;
1530 
1531 	ASSERT(vdc != NULL);
1532 	ASSERT(mutex_owned(&vdc->lock));
1533 
1534 	DMSG(vdc, 0, "[%d] entered\n", vdc->instance);
1535 
1536 	/* fill in tag */
1537 	pkt.tag.vio_msgtype = VIO_TYPE_CTRL;
1538 	pkt.tag.vio_subtype = VIO_SUBTYPE_INFO;
1539 	pkt.tag.vio_subtype_env = VIO_ATTR_INFO;
1540 	pkt.tag.vio_sid = vdc->session_id;
1541 	/* fill in payload */
1542 	pkt.max_xfer_sz = vdc->max_xfer_sz;
1543 	pkt.vdisk_block_size = vdc->block_size;
1544 	pkt.xfer_mode = VIO_DRING_MODE;
1545 	pkt.operations = 0;	/* server will set bits of valid operations */
1546 	pkt.vdisk_type = 0;	/* server will set to valid device type */
1547 	pkt.vdisk_media = 0;	/* server will set to valid media type */
1548 	pkt.vdisk_size = 0;	/* server will set to valid size */
1549 
1550 	status = vdc_send(vdc, (caddr_t)&pkt, &msglen);
1551 	DMSG(vdc, 0, "Attr info sent (status = %d)\n", status);
1552 
1553 	if ((status != 0) || (msglen != sizeof (vio_ver_msg_t))) {
1554 		DMSG(vdc, 0, "[%d] Failed to send Attr negotiation info: "
1555 		    "id(%lx) rv(%d) size(%ld)", vdc->instance, vdc->ldc_handle,
1556 		    status, msglen);
1557 		if (msglen != sizeof (vio_ver_msg_t))
1558 			status = ENOMSG;
1559 	}
1560 
1561 	return (status);
1562 }
1563 
1564 /*
1565  * Function:
1566  *	vdc_attr_negotiation()
1567  *
1568  * Description:
1569  *
1570  * Arguments:
1571  *	vdc	- soft state pointer for this instance of the device driver.
1572  *
1573  * Return Code:
1574  *	0	- Success
1575  */
1576 static int
1577 vdc_attr_negotiation(vdc_t *vdcp)
1578 {
1579 	int status;
1580 	vio_msg_t vio_msg;
1581 
1582 	if (status = vdc_init_attr_negotiation(vdcp))
1583 		return (status);
1584 
1585 	/* release lock and wait for response */
1586 	mutex_exit(&vdcp->lock);
1587 	status = vdc_wait_for_response(vdcp, &vio_msg);
1588 	mutex_enter(&vdcp->lock);
1589 	if (status) {
1590 		DMSG(vdcp, 0,
1591 		    "[%d] Failed waiting for Attr negotiation response, rv(%d)",
1592 		    vdcp->instance, status);
1593 		return (status);
1594 	}
1595 
1596 	/* check type and sub_type ... */
1597 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1598 	    vio_msg.tag.vio_subtype == VIO_SUBTYPE_INFO) {
1599 		DMSG(vdcp, 0, "[%d] Invalid attr negotiation response\n",
1600 		    vdcp->instance);
1601 		return (EPROTO);
1602 	}
1603 
1604 	return (vdc_handle_attr_msg(vdcp, (vd_attr_msg_t *)&vio_msg));
1605 }
1606 
1607 
1608 /*
1609  * Function:
1610  *	vdc_init_dring_negotiate()
1611  *
1612  * Description:
1613  *
1614  * Arguments:
1615  *	vdc	- soft state pointer for this instance of the device driver.
1616  *
1617  * Return Code:
1618  *	0	- Success
1619  */
1620 static int
1621 vdc_init_dring_negotiate(vdc_t *vdc)
1622 {
1623 	vio_dring_reg_msg_t	pkt;
1624 	size_t			msglen = sizeof (pkt);
1625 	int			status = -1;
1626 	int			retry;
1627 	int			nretries = 10;
1628 
1629 	ASSERT(vdc != NULL);
1630 	ASSERT(mutex_owned(&vdc->lock));
1631 
1632 	for (retry = 0; retry < nretries; retry++) {
1633 		status = vdc_init_descriptor_ring(vdc);
1634 		if (status != EAGAIN)
1635 			break;
1636 		drv_usecwait(vdc_min_timeout_ldc);
1637 	}
1638 
1639 	if (status != 0) {
1640 		DMSG(vdc, 0, "[%d] Failed to init DRing (status = %d)\n",
1641 		    vdc->instance, status);
1642 		return (status);
1643 	}
1644 
1645 	DMSG(vdc, 0, "[%d] Init of descriptor ring completed (status = %d)\n",
1646 	    vdc->instance, status);
1647 
1648 	/* fill in tag */
1649 	pkt.tag.vio_msgtype = VIO_TYPE_CTRL;
1650 	pkt.tag.vio_subtype = VIO_SUBTYPE_INFO;
1651 	pkt.tag.vio_subtype_env = VIO_DRING_REG;
1652 	pkt.tag.vio_sid = vdc->session_id;
1653 	/* fill in payload */
1654 	pkt.dring_ident = 0;
1655 	pkt.num_descriptors = vdc->dring_len;
1656 	pkt.descriptor_size = vdc->dring_entry_size;
1657 	pkt.options = (VIO_TX_DRING | VIO_RX_DRING);
1658 	pkt.ncookies = vdc->dring_cookie_count;
1659 	pkt.cookie[0] = vdc->dring_cookie[0];	/* for now just one cookie */
1660 
1661 	status = vdc_send(vdc, (caddr_t)&pkt, &msglen);
1662 	if (status != 0) {
1663 		DMSG(vdc, 0, "[%d] Failed to register DRing (err = %d)",
1664 		    vdc->instance, status);
1665 	}
1666 
1667 	return (status);
1668 }
1669 
1670 
1671 /*
1672  * Function:
1673  *	vdc_dring_negotiation()
1674  *
1675  * Description:
1676  *
1677  * Arguments:
1678  *	vdc	- soft state pointer for this instance of the device driver.
1679  *
1680  * Return Code:
1681  *	0	- Success
1682  */
1683 static int
1684 vdc_dring_negotiation(vdc_t *vdcp)
1685 {
1686 	int status;
1687 	vio_msg_t vio_msg;
1688 
1689 	if (status = vdc_init_dring_negotiate(vdcp))
1690 		return (status);
1691 
1692 	/* release lock and wait for response */
1693 	mutex_exit(&vdcp->lock);
1694 	status = vdc_wait_for_response(vdcp, &vio_msg);
1695 	mutex_enter(&vdcp->lock);
1696 	if (status) {
1697 		DMSG(vdcp, 0,
1698 		    "[%d] Failed waiting for Dring negotiation response,"
1699 		    " rv(%d)", vdcp->instance, status);
1700 		return (status);
1701 	}
1702 
1703 	/* check type and sub_type ... */
1704 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1705 	    vio_msg.tag.vio_subtype == VIO_SUBTYPE_INFO) {
1706 		DMSG(vdcp, 0, "[%d] Invalid Dring negotiation response\n",
1707 		    vdcp->instance);
1708 		return (EPROTO);
1709 	}
1710 
1711 	return (vdc_handle_dring_reg_msg(vdcp,
1712 	    (vio_dring_reg_msg_t *)&vio_msg));
1713 }
1714 
1715 
1716 /*
1717  * Function:
1718  *	vdc_send_rdx()
1719  *
1720  * Description:
1721  *
1722  * Arguments:
1723  *	vdc	- soft state pointer for this instance of the device driver.
1724  *
1725  * Return Code:
1726  *	0	- Success
1727  */
1728 static int
1729 vdc_send_rdx(vdc_t *vdcp)
1730 {
1731 	vio_msg_t	msg;
1732 	size_t		msglen = sizeof (vio_msg_t);
1733 	int		status;
1734 
1735 	/*
1736 	 * Send an RDX message to vds to indicate we are ready
1737 	 * to send data
1738 	 */
1739 	msg.tag.vio_msgtype = VIO_TYPE_CTRL;
1740 	msg.tag.vio_subtype = VIO_SUBTYPE_INFO;
1741 	msg.tag.vio_subtype_env = VIO_RDX;
1742 	msg.tag.vio_sid = vdcp->session_id;
1743 	status = vdc_send(vdcp, (caddr_t)&msg, &msglen);
1744 	if (status != 0) {
1745 		DMSG(vdcp, 0, "[%d] Failed to send RDX message (%d)",
1746 		    vdcp->instance, status);
1747 	}
1748 
1749 	return (status);
1750 }
1751 
1752 /*
1753  * Function:
1754  *	vdc_handle_rdx()
1755  *
1756  * Description:
1757  *
1758  * Arguments:
1759  *	vdc	- soft state pointer for this instance of the device driver.
1760  *	msgp	- received msg
1761  *
1762  * Return Code:
1763  *	0	- Success
1764  */
1765 static int
1766 vdc_handle_rdx(vdc_t *vdcp, vio_rdx_msg_t *msgp)
1767 {
1768 	_NOTE(ARGUNUSED(vdcp))
1769 	_NOTE(ARGUNUSED(msgp))
1770 
1771 	ASSERT(msgp->tag.vio_msgtype == VIO_TYPE_CTRL);
1772 	ASSERT(msgp->tag.vio_subtype == VIO_SUBTYPE_ACK);
1773 	ASSERT(msgp->tag.vio_subtype_env == VIO_RDX);
1774 
1775 	DMSG(vdcp, 1, "[%d] Got an RDX msg", vdcp->instance);
1776 
1777 	return (0);
1778 }
1779 
1780 /*
1781  * Function:
1782  *	vdc_rdx_exchange()
1783  *
1784  * Description:
1785  *
1786  * Arguments:
1787  *	vdc	- soft state pointer for this instance of the device driver.
1788  *
1789  * Return Code:
1790  *	0	- Success
1791  */
1792 static int
1793 vdc_rdx_exchange(vdc_t *vdcp)
1794 {
1795 	int status;
1796 	vio_msg_t vio_msg;
1797 
1798 	if (status = vdc_send_rdx(vdcp))
1799 		return (status);
1800 
1801 	/* release lock and wait for response */
1802 	mutex_exit(&vdcp->lock);
1803 	status = vdc_wait_for_response(vdcp, &vio_msg);
1804 	mutex_enter(&vdcp->lock);
1805 	if (status) {
1806 		DMSG(vdcp, 0, "[%d] Failed waiting for RDX response, rv(%d)",
1807 		    vdcp->instance, status);
1808 		return (status);
1809 	}
1810 
1811 	/* check type and sub_type ... */
1812 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1813 	    vio_msg.tag.vio_subtype != VIO_SUBTYPE_ACK) {
1814 		DMSG(vdcp, 0, "[%d] Invalid RDX response\n", vdcp->instance);
1815 		return (EPROTO);
1816 	}
1817 
1818 	return (vdc_handle_rdx(vdcp, (vio_rdx_msg_t *)&vio_msg));
1819 }
1820 
1821 
1822 /* -------------------------------------------------------------------------- */
1823 
1824 /*
1825  * LDC helper routines
1826  */
1827 
1828 static int
1829 vdc_recv(vdc_t *vdc, vio_msg_t *msgp, size_t *nbytesp)
1830 {
1831 	int		status;
1832 	boolean_t	q_has_pkts = B_FALSE;
1833 	uint64_t	delay_time;
1834 	size_t		len;
1835 
1836 	mutex_enter(&vdc->read_lock);
1837 
1838 	if (vdc->read_state == VDC_READ_IDLE)
1839 		vdc->read_state = VDC_READ_WAITING;
1840 
1841 	while (vdc->read_state != VDC_READ_PENDING) {
1842 
1843 		/* detect if the connection has been reset */
1844 		if (vdc->read_state == VDC_READ_RESET) {
1845 			status = ECONNRESET;
1846 			goto done;
1847 		}
1848 
1849 		cv_wait(&vdc->read_cv, &vdc->read_lock);
1850 	}
1851 
1852 	/*
1853 	 * Until we get a blocking ldc read we have to retry
1854 	 * until the entire LDC message has arrived before
1855 	 * ldc_read() will succeed. Note we also bail out if
1856 	 * the channel is reset or goes away.
1857 	 */
1858 	delay_time = vdc_ldc_read_init_delay;
1859 loop:
1860 	len = *nbytesp;
1861 	status = ldc_read(vdc->ldc_handle, (caddr_t)msgp, &len);
1862 	switch (status) {
1863 	case EAGAIN:
1864 		delay_time *= 2;
1865 		if (delay_time >= vdc_ldc_read_max_delay)
1866 			delay_time = vdc_ldc_read_max_delay;
1867 		delay(delay_time);
1868 		goto loop;
1869 
1870 	case 0:
1871 		if (len == 0) {
1872 			DMSG(vdc, 1, "[%d] ldc_read returned 0 bytes with "
1873 			    "no error!\n", vdc->instance);
1874 			goto loop;
1875 		}
1876 
1877 		*nbytesp = len;
1878 
1879 		/*
1880 		 * If there are pending messages, leave the
1881 		 * read state as pending. Otherwise, set the state
1882 		 * back to idle.
1883 		 */
1884 		status = ldc_chkq(vdc->ldc_handle, &q_has_pkts);
1885 		if (status == 0 && !q_has_pkts)
1886 			vdc->read_state = VDC_READ_IDLE;
1887 
1888 		break;
1889 	default:
1890 		DMSG(vdc, 0, "ldc_read returned %d\n", status);
1891 		break;
1892 	}
1893 
1894 done:
1895 	mutex_exit(&vdc->read_lock);
1896 
1897 	return (status);
1898 }
1899 
1900 
1901 
1902 #ifdef DEBUG
1903 void
1904 vdc_decode_tag(vdc_t *vdcp, vio_msg_t *msg)
1905 {
1906 	char *ms, *ss, *ses;
1907 	switch (msg->tag.vio_msgtype) {
1908 #define	Q(_s)	case _s : ms = #_s; break;
1909 	Q(VIO_TYPE_CTRL)
1910 	Q(VIO_TYPE_DATA)
1911 	Q(VIO_TYPE_ERR)
1912 #undef Q
1913 	default: ms = "unknown"; break;
1914 	}
1915 
1916 	switch (msg->tag.vio_subtype) {
1917 #define	Q(_s)	case _s : ss = #_s; break;
1918 	Q(VIO_SUBTYPE_INFO)
1919 	Q(VIO_SUBTYPE_ACK)
1920 	Q(VIO_SUBTYPE_NACK)
1921 #undef Q
1922 	default: ss = "unknown"; break;
1923 	}
1924 
1925 	switch (msg->tag.vio_subtype_env) {
1926 #define	Q(_s)	case _s : ses = #_s; break;
1927 	Q(VIO_VER_INFO)
1928 	Q(VIO_ATTR_INFO)
1929 	Q(VIO_DRING_REG)
1930 	Q(VIO_DRING_UNREG)
1931 	Q(VIO_RDX)
1932 	Q(VIO_PKT_DATA)
1933 	Q(VIO_DESC_DATA)
1934 	Q(VIO_DRING_DATA)
1935 #undef Q
1936 	default: ses = "unknown"; break;
1937 	}
1938 
1939 	DMSG(vdcp, 3, "(%x/%x/%x) message : (%s/%s/%s)\n",
1940 	    msg->tag.vio_msgtype, msg->tag.vio_subtype,
1941 	    msg->tag.vio_subtype_env, ms, ss, ses);
1942 }
1943 #endif
1944 
1945 /*
1946  * Function:
1947  *	vdc_send()
1948  *
1949  * Description:
1950  *	The function encapsulates the call to write a message using LDC.
1951  *	If LDC indicates that the call failed due to the queue being full,
1952  *	we retry the ldc_write(), otherwise we return the error returned by LDC.
1953  *
1954  * Arguments:
1955  *	ldc_handle	- LDC handle for the channel this instance of vdc uses
1956  *	pkt		- address of LDC message to be sent
1957  *	msglen		- the size of the message being sent. When the function
1958  *			  returns, this contains the number of bytes written.
1959  *
1960  * Return Code:
1961  *	0		- Success.
1962  *	EINVAL		- pkt or msglen were NULL
1963  *	ECONNRESET	- The connection was not up.
1964  *	EWOULDBLOCK	- LDC queue is full
1965  *	xxx		- other error codes returned by ldc_write
1966  */
1967 static int
1968 vdc_send(vdc_t *vdc, caddr_t pkt, size_t *msglen)
1969 {
1970 	size_t	size = 0;
1971 	int	status = 0;
1972 	clock_t delay_ticks;
1973 
1974 	ASSERT(vdc != NULL);
1975 	ASSERT(mutex_owned(&vdc->lock));
1976 	ASSERT(msglen != NULL);
1977 	ASSERT(*msglen != 0);
1978 
1979 #ifdef DEBUG
1980 	vdc_decode_tag(vdc, (vio_msg_t *)(uintptr_t)pkt);
1981 #endif
1982 	/*
1983 	 * Wait indefinitely to send if channel
1984 	 * is busy, but bail out if we succeed or
1985 	 * if the channel closes or is reset.
1986 	 */
1987 	delay_ticks = vdc_hz_min_ldc_delay;
1988 	do {
1989 		size = *msglen;
1990 		status = ldc_write(vdc->ldc_handle, pkt, &size);
1991 		if (status == EWOULDBLOCK) {
1992 			delay(delay_ticks);
1993 			/* geometric backoff */
1994 			delay_ticks *= 2;
1995 			if (delay_ticks > vdc_hz_max_ldc_delay)
1996 				delay_ticks = vdc_hz_max_ldc_delay;
1997 		}
1998 	} while (status == EWOULDBLOCK);
1999 
2000 	/* if LDC had serious issues --- reset vdc state */
2001 	if (status == EIO || status == ECONNRESET) {
2002 		/* LDC had serious issues --- reset vdc state */
2003 		mutex_enter(&vdc->read_lock);
2004 		if ((vdc->read_state == VDC_READ_WAITING) ||
2005 		    (vdc->read_state == VDC_READ_RESET))
2006 			cv_signal(&vdc->read_cv);
2007 		vdc->read_state = VDC_READ_RESET;
2008 		mutex_exit(&vdc->read_lock);
2009 
2010 		/* wake up any waiters in the reset thread */
2011 		if (vdc->state == VDC_STATE_INIT_WAITING) {
2012 			DMSG(vdc, 0, "[%d] write reset - "
2013 			    "vdc is resetting ..\n", vdc->instance);
2014 			vdc->state = VDC_STATE_RESETTING;
2015 			cv_signal(&vdc->initwait_cv);
2016 		}
2017 
2018 		return (ECONNRESET);
2019 	}
2020 
2021 	/* return the last size written */
2022 	*msglen = size;
2023 
2024 	return (status);
2025 }
2026 
2027 /*
2028  * Function:
2029  *	vdc_get_md_node
2030  *
2031  * Description:
2032  *	Get the MD, the device node and the port node for the given
2033  *	disk instance. The caller is responsible for cleaning up the
2034  *	reference to the returned MD (mdpp) by calling md_fini_handle().
2035  *
2036  * Arguments:
2037  *	dip	- dev info pointer for this instance of the device driver.
2038  *	mdpp	- the returned MD.
2039  *	vd_nodep - the returned device node.
2040  *	vd_portp - the returned port node. The returned port node is NULL
2041  *		   if no port node is found.
2042  *
2043  * Return Code:
2044  *	0	- Success.
2045  *	ENOENT	- Expected node or property did not exist.
2046  *	ENXIO	- Unexpected error communicating with MD framework
2047  */
2048 static int
2049 vdc_get_md_node(dev_info_t *dip, md_t **mdpp, mde_cookie_t *vd_nodep,
2050     mde_cookie_t *vd_portp)
2051 {
2052 	int		status = ENOENT;
2053 	char		*node_name = NULL;
2054 	md_t		*mdp = NULL;
2055 	int		num_nodes;
2056 	int		num_vdevs;
2057 	int		num_vports;
2058 	mde_cookie_t	rootnode;
2059 	mde_cookie_t	*listp = NULL;
2060 	boolean_t	found_inst = B_FALSE;
2061 	int		listsz;
2062 	int		idx;
2063 	uint64_t	md_inst;
2064 	int		obp_inst;
2065 	int		instance = ddi_get_instance(dip);
2066 
2067 	/*
2068 	 * Get the OBP instance number for comparison with the MD instance
2069 	 *
2070 	 * The "cfg-handle" property of a vdc node in an MD contains the MD's
2071 	 * notion of "instance", or unique identifier, for that node; OBP
2072 	 * stores the value of the "cfg-handle" MD property as the value of
2073 	 * the "reg" property on the node in the device tree it builds from
2074 	 * the MD and passes to Solaris.  Thus, we look up the devinfo node's
2075 	 * "reg" property value to uniquely identify this device instance.
2076 	 * If the "reg" property cannot be found, the device tree state is
2077 	 * presumably so broken that there is no point in continuing.
2078 	 */
2079 	if (!ddi_prop_exists(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS, OBP_REG)) {
2080 		cmn_err(CE_WARN, "'%s' property does not exist", OBP_REG);
2081 		return (ENOENT);
2082 	}
2083 	obp_inst = ddi_prop_get_int(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS,
2084 	    OBP_REG, -1);
2085 	DMSGX(1, "[%d] OBP inst=%d\n", instance, obp_inst);
2086 
2087 	/*
2088 	 * We now walk the MD nodes to find the node for this vdisk.
2089 	 */
2090 	if ((mdp = md_get_handle()) == NULL) {
2091 		cmn_err(CE_WARN, "unable to init machine description");
2092 		return (ENXIO);
2093 	}
2094 
2095 	num_nodes = md_node_count(mdp);
2096 	ASSERT(num_nodes > 0);
2097 
2098 	listsz = num_nodes * sizeof (mde_cookie_t);
2099 
2100 	/* allocate memory for nodes */
2101 	listp = kmem_zalloc(listsz, KM_SLEEP);
2102 
2103 	rootnode = md_root_node(mdp);
2104 	ASSERT(rootnode != MDE_INVAL_ELEM_COOKIE);
2105 
2106 	/*
2107 	 * Search for all the virtual devices, we will then check to see which
2108 	 * ones are disk nodes.
2109 	 */
2110 	num_vdevs = md_scan_dag(mdp, rootnode,
2111 	    md_find_name(mdp, VDC_MD_VDEV_NAME),
2112 	    md_find_name(mdp, "fwd"), listp);
2113 
2114 	if (num_vdevs <= 0) {
2115 		cmn_err(CE_NOTE, "No '%s' node found", VDC_MD_VDEV_NAME);
2116 		status = ENOENT;
2117 		goto done;
2118 	}
2119 
2120 	DMSGX(1, "[%d] num_vdevs=%d\n", instance, num_vdevs);
2121 	for (idx = 0; idx < num_vdevs; idx++) {
2122 		status = md_get_prop_str(mdp, listp[idx], "name", &node_name);
2123 		if ((status != 0) || (node_name == NULL)) {
2124 			cmn_err(CE_NOTE, "Unable to get name of node type '%s'"
2125 			    ": err %d", VDC_MD_VDEV_NAME, status);
2126 			continue;
2127 		}
2128 
2129 		DMSGX(1, "[%d] Found node '%s'\n", instance, node_name);
2130 		if (strcmp(VDC_MD_DISK_NAME, node_name) == 0) {
2131 			status = md_get_prop_val(mdp, listp[idx],
2132 			    VDC_MD_CFG_HDL, &md_inst);
2133 			DMSGX(1, "[%d] vdc inst in MD=%lx\n",
2134 			    instance, md_inst);
2135 			if ((status == 0) && (md_inst == obp_inst)) {
2136 				found_inst = B_TRUE;
2137 				break;
2138 			}
2139 		}
2140 	}
2141 
2142 	if (!found_inst) {
2143 		DMSGX(0, "Unable to find correct '%s' node", VDC_MD_DISK_NAME);
2144 		status = ENOENT;
2145 		goto done;
2146 	}
2147 	DMSGX(0, "[%d] MD inst=%lx\n", instance, md_inst);
2148 
2149 	*vd_nodep = listp[idx];
2150 	*mdpp = mdp;
2151 
2152 	num_vports = md_scan_dag(mdp, *vd_nodep,
2153 	    md_find_name(mdp, VDC_MD_PORT_NAME),
2154 	    md_find_name(mdp, "fwd"), listp);
2155 
2156 	if (num_vports != 1) {
2157 		DMSGX(0, "Expected 1 '%s' node for '%s' port, found %d\n",
2158 		    VDC_MD_PORT_NAME, VDC_MD_VDEV_NAME, num_vports);
2159 	}
2160 
2161 	*vd_portp = (num_vports == 0)? NULL: listp[0];
2162 
2163 done:
2164 	kmem_free(listp, listsz);
2165 	return (status);
2166 }
2167 
2168 /*
2169  * Function:
2170  *	vdc_get_ldc_id()
2171  *
2172  * Description:
2173  *	This function gets the 'ldc-id' for this particular instance of vdc.
2174  *	The id returned is the guest domain channel endpoint LDC uses for
2175  *	communication with vds.
2176  *
2177  * Arguments:
2178  *	mdp	- pointer to the machine description.
2179  *	vd_node	- the vdisk element from the MD.
2180  *	ldc_id	- pointer to variable used to return the 'ldc-id' found.
2181  *
2182  * Return Code:
2183  *	0	- Success.
2184  *	ENOENT	- Expected node or property did not exist.
2185  */
2186 static int
2187 vdc_get_ldc_id(md_t *mdp, mde_cookie_t vd_node, uint64_t *ldc_id)
2188 {
2189 	mde_cookie_t	*chanp = NULL;
2190 	int		listsz;
2191 	int		num_chans;
2192 	int		num_nodes;
2193 	int		status = 0;
2194 
2195 	num_nodes = md_node_count(mdp);
2196 	ASSERT(num_nodes > 0);
2197 
2198 	listsz = num_nodes * sizeof (mde_cookie_t);
2199 
2200 	/* allocate memory for nodes */
2201 	chanp = kmem_zalloc(listsz, KM_SLEEP);
2202 
2203 	/* get the channels for this node */
2204 	num_chans = md_scan_dag(mdp, vd_node,
2205 	    md_find_name(mdp, VDC_MD_CHAN_NAME),
2206 	    md_find_name(mdp, "fwd"), chanp);
2207 
2208 	/* expecting at least one channel */
2209 	if (num_chans <= 0) {
2210 		cmn_err(CE_NOTE, "No '%s' node for '%s' port",
2211 		    VDC_MD_CHAN_NAME, VDC_MD_VDEV_NAME);
2212 		status = ENOENT;
2213 		goto done;
2214 
2215 	} else if (num_chans != 1) {
2216 		DMSGX(0, "Expected 1 '%s' node for '%s' port, found %d\n",
2217 		    VDC_MD_CHAN_NAME, VDC_MD_VDEV_NAME, num_chans);
2218 	}
2219 
2220 	/*
2221 	 * We use the first channel found (index 0), irrespective of how
2222 	 * many are there in total.
2223 	 */
2224 	if (md_get_prop_val(mdp, chanp[0], VDC_MD_ID, ldc_id) != 0) {
2225 		cmn_err(CE_NOTE, "Channel '%s' property not found", VDC_MD_ID);
2226 		status = ENOENT;
2227 	}
2228 
2229 done:
2230 	kmem_free(chanp, listsz);
2231 	return (status);
2232 }
2233 
2234 static int
2235 vdc_do_ldc_up(vdc_t *vdc)
2236 {
2237 	int		status;
2238 	ldc_status_t	ldc_state;
2239 
2240 	DMSG(vdc, 0, "[%d] Bringing up channel %lx\n",
2241 	    vdc->instance, vdc->ldc_id);
2242 
2243 	if (vdc->lifecycle == VDC_LC_DETACHING)
2244 		return (EINVAL);
2245 
2246 	if ((status = ldc_up(vdc->ldc_handle)) != 0) {
2247 		switch (status) {
2248 		case ECONNREFUSED:	/* listener not ready at other end */
2249 			DMSG(vdc, 0, "[%d] ldc_up(%lx,...) return %d\n",
2250 			    vdc->instance, vdc->ldc_id, status);
2251 			status = 0;
2252 			break;
2253 		default:
2254 			DMSG(vdc, 0, "[%d] Failed to bring up LDC: "
2255 			    "channel=%ld, err=%d", vdc->instance, vdc->ldc_id,
2256 			    status);
2257 			break;
2258 		}
2259 	}
2260 
2261 	if (ldc_status(vdc->ldc_handle, &ldc_state) == 0) {
2262 		vdc->ldc_state = ldc_state;
2263 		if (ldc_state == LDC_UP) {
2264 			DMSG(vdc, 0, "[%d] LDC channel already up\n",
2265 			    vdc->instance);
2266 			vdc->seq_num = 1;
2267 			vdc->seq_num_reply = 0;
2268 		}
2269 	}
2270 
2271 	return (status);
2272 }
2273 
2274 /*
2275  * Function:
2276  *	vdc_terminate_ldc()
2277  *
2278  * Description:
2279  *
2280  * Arguments:
2281  *	vdc	- soft state pointer for this instance of the device driver.
2282  *
2283  * Return Code:
2284  *	None
2285  */
2286 static void
2287 vdc_terminate_ldc(vdc_t *vdc)
2288 {
2289 	int	instance = ddi_get_instance(vdc->dip);
2290 
2291 	ASSERT(vdc != NULL);
2292 	ASSERT(mutex_owned(&vdc->lock));
2293 
2294 	DMSG(vdc, 0, "[%d] initialized=%x\n", instance, vdc->initialized);
2295 
2296 	if (vdc->initialized & VDC_LDC_OPEN) {
2297 		DMSG(vdc, 0, "[%d] ldc_close()\n", instance);
2298 		(void) ldc_close(vdc->ldc_handle);
2299 	}
2300 	if (vdc->initialized & VDC_LDC_CB) {
2301 		DMSG(vdc, 0, "[%d] ldc_unreg_callback()\n", instance);
2302 		(void) ldc_unreg_callback(vdc->ldc_handle);
2303 	}
2304 	if (vdc->initialized & VDC_LDC) {
2305 		DMSG(vdc, 0, "[%d] ldc_fini()\n", instance);
2306 		(void) ldc_fini(vdc->ldc_handle);
2307 		vdc->ldc_handle = NULL;
2308 	}
2309 
2310 	vdc->initialized &= ~(VDC_LDC | VDC_LDC_CB | VDC_LDC_OPEN);
2311 }
2312 
2313 /* -------------------------------------------------------------------------- */
2314 
2315 /*
2316  * Descriptor Ring helper routines
2317  */
2318 
2319 /*
2320  * Function:
2321  *	vdc_init_descriptor_ring()
2322  *
2323  * Description:
2324  *
2325  * Arguments:
2326  *	vdc	- soft state pointer for this instance of the device driver.
2327  *
2328  * Return Code:
2329  *	0	- Success
2330  */
2331 static int
2332 vdc_init_descriptor_ring(vdc_t *vdc)
2333 {
2334 	vd_dring_entry_t	*dep = NULL;	/* DRing Entry pointer */
2335 	int	status = 0;
2336 	int	i;
2337 
2338 	DMSG(vdc, 0, "[%d] initialized=%x\n", vdc->instance, vdc->initialized);
2339 
2340 	ASSERT(vdc != NULL);
2341 	ASSERT(mutex_owned(&vdc->lock));
2342 	ASSERT(vdc->ldc_handle != NULL);
2343 
2344 	/* ensure we have enough room to store max sized block */
2345 	ASSERT(maxphys <= VD_MAX_BLOCK_SIZE);
2346 
2347 	if ((vdc->initialized & VDC_DRING_INIT) == 0) {
2348 		DMSG(vdc, 0, "[%d] ldc_mem_dring_create\n", vdc->instance);
2349 		/*
2350 		 * Calculate the maximum block size we can transmit using one
2351 		 * Descriptor Ring entry from the attributes returned by the
2352 		 * vDisk server. This is subject to a minimum of 'maxphys'
2353 		 * as we do not have the capability to split requests over
2354 		 * multiple DRing entries.
2355 		 */
2356 		if ((vdc->max_xfer_sz * vdc->block_size) < maxphys) {
2357 			DMSG(vdc, 0, "[%d] using minimum DRing size\n",
2358 			    vdc->instance);
2359 			vdc->dring_max_cookies = maxphys / PAGESIZE;
2360 		} else {
2361 			vdc->dring_max_cookies =
2362 			    (vdc->max_xfer_sz * vdc->block_size) / PAGESIZE;
2363 		}
2364 		vdc->dring_entry_size = (sizeof (vd_dring_entry_t) +
2365 		    (sizeof (ldc_mem_cookie_t) *
2366 		    (vdc->dring_max_cookies - 1)));
2367 		vdc->dring_len = VD_DRING_LEN;
2368 
2369 		status = ldc_mem_dring_create(vdc->dring_len,
2370 		    vdc->dring_entry_size, &vdc->ldc_dring_hdl);
2371 		if ((vdc->ldc_dring_hdl == NULL) || (status != 0)) {
2372 			DMSG(vdc, 0, "[%d] Descriptor ring creation failed",
2373 			    vdc->instance);
2374 			return (status);
2375 		}
2376 		vdc->initialized |= VDC_DRING_INIT;
2377 	}
2378 
2379 	if ((vdc->initialized & VDC_DRING_BOUND) == 0) {
2380 		DMSG(vdc, 0, "[%d] ldc_mem_dring_bind\n", vdc->instance);
2381 		vdc->dring_cookie =
2382 		    kmem_zalloc(sizeof (ldc_mem_cookie_t), KM_SLEEP);
2383 
2384 		status = ldc_mem_dring_bind(vdc->ldc_handle, vdc->ldc_dring_hdl,
2385 		    LDC_SHADOW_MAP|LDC_DIRECT_MAP, LDC_MEM_RW,
2386 		    &vdc->dring_cookie[0],
2387 		    &vdc->dring_cookie_count);
2388 		if (status != 0) {
2389 			DMSG(vdc, 0, "[%d] Failed to bind descriptor ring "
2390 			    "(%lx) to channel (%lx) status=%d\n",
2391 			    vdc->instance, vdc->ldc_dring_hdl,
2392 			    vdc->ldc_handle, status);
2393 			return (status);
2394 		}
2395 		ASSERT(vdc->dring_cookie_count == 1);
2396 		vdc->initialized |= VDC_DRING_BOUND;
2397 	}
2398 
2399 	status = ldc_mem_dring_info(vdc->ldc_dring_hdl, &vdc->dring_mem_info);
2400 	if (status != 0) {
2401 		DMSG(vdc, 0,
2402 		    "[%d] Failed to get info for descriptor ring (%lx)\n",
2403 		    vdc->instance, vdc->ldc_dring_hdl);
2404 		return (status);
2405 	}
2406 
2407 	if ((vdc->initialized & VDC_DRING_LOCAL) == 0) {
2408 		DMSG(vdc, 0, "[%d] local dring\n", vdc->instance);
2409 
2410 		/* Allocate the local copy of this dring */
2411 		vdc->local_dring =
2412 		    kmem_zalloc(vdc->dring_len * sizeof (vdc_local_desc_t),
2413 		    KM_SLEEP);
2414 		vdc->initialized |= VDC_DRING_LOCAL;
2415 	}
2416 
2417 	/*
2418 	 * Mark all DRing entries as free and initialize the private
2419 	 * descriptor's memory handles. If any entry is initialized,
2420 	 * we need to free it later so we set the bit in 'initialized'
2421 	 * at the start.
2422 	 */
2423 	vdc->initialized |= VDC_DRING_ENTRY;
2424 	for (i = 0; i < vdc->dring_len; i++) {
2425 		dep = VDC_GET_DRING_ENTRY_PTR(vdc, i);
2426 		dep->hdr.dstate = VIO_DESC_FREE;
2427 
2428 		status = ldc_mem_alloc_handle(vdc->ldc_handle,
2429 		    &vdc->local_dring[i].desc_mhdl);
2430 		if (status != 0) {
2431 			DMSG(vdc, 0, "![%d] Failed to alloc mem handle for"
2432 			    " descriptor %d", vdc->instance, i);
2433 			return (status);
2434 		}
2435 		vdc->local_dring[i].is_free = B_TRUE;
2436 		vdc->local_dring[i].dep = dep;
2437 	}
2438 
2439 	/* Initialize the starting index */
2440 	vdc->dring_curr_idx = 0;
2441 
2442 	return (status);
2443 }
2444 
2445 /*
2446  * Function:
2447  *	vdc_destroy_descriptor_ring()
2448  *
2449  * Description:
2450  *
2451  * Arguments:
2452  *	vdc	- soft state pointer for this instance of the device driver.
2453  *
2454  * Return Code:
2455  *	None
2456  */
2457 static void
2458 vdc_destroy_descriptor_ring(vdc_t *vdc)
2459 {
2460 	vdc_local_desc_t	*ldep = NULL;	/* Local Dring Entry Pointer */
2461 	ldc_mem_handle_t	mhdl = NULL;
2462 	ldc_mem_info_t		minfo;
2463 	int			status = -1;
2464 	int			i;	/* loop */
2465 
2466 	ASSERT(vdc != NULL);
2467 	ASSERT(mutex_owned(&vdc->lock));
2468 
2469 	DMSG(vdc, 0, "[%d] Entered\n", vdc->instance);
2470 
2471 	if (vdc->initialized & VDC_DRING_ENTRY) {
2472 		DMSG(vdc, 0,
2473 		    "[%d] Removing Local DRing entries\n", vdc->instance);
2474 		for (i = 0; i < vdc->dring_len; i++) {
2475 			ldep = &vdc->local_dring[i];
2476 			mhdl = ldep->desc_mhdl;
2477 
2478 			if (mhdl == NULL)
2479 				continue;
2480 
2481 			if ((status = ldc_mem_info(mhdl, &minfo)) != 0) {
2482 				DMSG(vdc, 0,
2483 				    "ldc_mem_info returned an error: %d\n",
2484 				    status);
2485 
2486 				/*
2487 				 * This must mean that the mem handle
2488 				 * is not valid. Clear it out so that
2489 				 * no one tries to use it.
2490 				 */
2491 				ldep->desc_mhdl = NULL;
2492 				continue;
2493 			}
2494 
2495 			if (minfo.status == LDC_BOUND) {
2496 				(void) ldc_mem_unbind_handle(mhdl);
2497 			}
2498 
2499 			(void) ldc_mem_free_handle(mhdl);
2500 
2501 			ldep->desc_mhdl = NULL;
2502 		}
2503 		vdc->initialized &= ~VDC_DRING_ENTRY;
2504 	}
2505 
2506 	if (vdc->initialized & VDC_DRING_LOCAL) {
2507 		DMSG(vdc, 0, "[%d] Freeing Local DRing\n", vdc->instance);
2508 		kmem_free(vdc->local_dring,
2509 		    vdc->dring_len * sizeof (vdc_local_desc_t));
2510 		vdc->initialized &= ~VDC_DRING_LOCAL;
2511 	}
2512 
2513 	if (vdc->initialized & VDC_DRING_BOUND) {
2514 		DMSG(vdc, 0, "[%d] Unbinding DRing\n", vdc->instance);
2515 		status = ldc_mem_dring_unbind(vdc->ldc_dring_hdl);
2516 		if (status == 0) {
2517 			vdc->initialized &= ~VDC_DRING_BOUND;
2518 		} else {
2519 			DMSG(vdc, 0, "[%d] Error %d unbinding DRing %lx",
2520 			    vdc->instance, status, vdc->ldc_dring_hdl);
2521 		}
2522 		kmem_free(vdc->dring_cookie, sizeof (ldc_mem_cookie_t));
2523 	}
2524 
2525 	if (vdc->initialized & VDC_DRING_INIT) {
2526 		DMSG(vdc, 0, "[%d] Destroying DRing\n", vdc->instance);
2527 		status = ldc_mem_dring_destroy(vdc->ldc_dring_hdl);
2528 		if (status == 0) {
2529 			vdc->ldc_dring_hdl = NULL;
2530 			bzero(&vdc->dring_mem_info, sizeof (ldc_mem_info_t));
2531 			vdc->initialized &= ~VDC_DRING_INIT;
2532 		} else {
2533 			DMSG(vdc, 0, "[%d] Error %d destroying DRing (%lx)",
2534 			    vdc->instance, status, vdc->ldc_dring_hdl);
2535 		}
2536 	}
2537 }
2538 
2539 /*
2540  * Function:
2541  *	vdc_map_to_shared_ring()
2542  *
2543  * Description:
2544  *	Copy contents of the local descriptor to the shared
2545  *	memory descriptor.
2546  *
2547  * Arguments:
2548  *	vdcp	- soft state pointer for this instance of the device driver.
2549  *	idx	- descriptor ring index
2550  *
2551  * Return Code:
2552  *	None
2553  */
2554 static int
2555 vdc_map_to_shared_dring(vdc_t *vdcp, int idx)
2556 {
2557 	vdc_local_desc_t	*ldep;
2558 	vd_dring_entry_t	*dep;
2559 	int			rv;
2560 
2561 	ldep = &(vdcp->local_dring[idx]);
2562 
2563 	/* for now leave in the old pop_mem_hdl stuff */
2564 	if (ldep->nbytes > 0) {
2565 		rv = vdc_populate_mem_hdl(vdcp, ldep);
2566 		if (rv) {
2567 			DMSG(vdcp, 0, "[%d] Cannot populate mem handle\n",
2568 			    vdcp->instance);
2569 			return (rv);
2570 		}
2571 	}
2572 
2573 	/*
2574 	 * fill in the data details into the DRing
2575 	 */
2576 	dep = ldep->dep;
2577 	ASSERT(dep != NULL);
2578 
2579 	dep->payload.req_id = VDC_GET_NEXT_REQ_ID(vdcp);
2580 	dep->payload.operation = ldep->operation;
2581 	dep->payload.addr = ldep->offset;
2582 	dep->payload.nbytes = ldep->nbytes;
2583 	dep->payload.status = (uint32_t)-1;	/* vds will set valid value */
2584 	dep->payload.slice = ldep->slice;
2585 	dep->hdr.dstate = VIO_DESC_READY;
2586 	dep->hdr.ack = 1;		/* request an ACK for every message */
2587 
2588 	return (0);
2589 }
2590 
2591 /*
2592  * Function:
2593  *	vdc_send_request
2594  *
2595  * Description:
2596  *	This routine writes the data to be transmitted to vds into the
2597  *	descriptor, notifies vds that the ring has been updated and
2598  *	then waits for the request to be processed.
2599  *
2600  * Arguments:
2601  *	vdcp	  - the soft state pointer
2602  *	operation - operation we want vds to perform (VD_OP_XXX)
2603  *	addr	  - address of data buf to be read/written.
2604  *	nbytes	  - number of bytes to read/write
2605  *	slice	  - the disk slice this request is for
2606  *	offset	  - relative disk offset
2607  *	cb_type   - type of call - STRATEGY or SYNC
2608  *	cb_arg	  - parameter to be sent to server (depends on VD_OP_XXX type)
2609  *			. mode for ioctl(9e)
2610  *			. LP64 diskaddr_t (block I/O)
2611  *	dir	  - direction of operation (READ/WRITE/BOTH)
2612  *
2613  * Return Codes:
2614  *	0
2615  *	ENXIO
2616  */
2617 static int
2618 vdc_send_request(vdc_t *vdcp, int operation, caddr_t addr,
2619     size_t nbytes, int slice, diskaddr_t offset, int cb_type,
2620     void *cb_arg, vio_desc_direction_t dir)
2621 {
2622 	ASSERT(vdcp != NULL);
2623 	ASSERT(slice == VD_SLICE_NONE || slice < V_NUMPAR);
2624 
2625 	mutex_enter(&vdcp->lock);
2626 
2627 	do {
2628 		while (vdcp->state != VDC_STATE_RUNNING) {
2629 
2630 			/* return error if detaching */
2631 			if (vdcp->state == VDC_STATE_DETACH) {
2632 				mutex_exit(&vdcp->lock);
2633 				return (ENXIO);
2634 			}
2635 
2636 			/* fail request if connection timeout is reached */
2637 			if (vdcp->ctimeout_reached) {
2638 				mutex_exit(&vdcp->lock);
2639 				return (EIO);
2640 			}
2641 
2642 			cv_wait(&vdcp->running_cv, &vdcp->lock);
2643 		}
2644 
2645 	} while (vdc_populate_descriptor(vdcp, operation, addr,
2646 	    nbytes, slice, offset, cb_type, cb_arg, dir));
2647 
2648 	mutex_exit(&vdcp->lock);
2649 	return (0);
2650 }
2651 
2652 
2653 /*
2654  * Function:
2655  *	vdc_populate_descriptor
2656  *
2657  * Description:
2658  *	This routine writes the data to be transmitted to vds into the
2659  *	descriptor, notifies vds that the ring has been updated and
2660  *	then waits for the request to be processed.
2661  *
2662  * Arguments:
2663  *	vdcp	  - the soft state pointer
2664  *	operation - operation we want vds to perform (VD_OP_XXX)
2665  *	addr	  - address of data buf to be read/written.
2666  *	nbytes	  - number of bytes to read/write
2667  *	slice	  - the disk slice this request is for
2668  *	offset	  - relative disk offset
2669  *	cb_type   - type of call - STRATEGY or SYNC
2670  *	cb_arg	  - parameter to be sent to server (depends on VD_OP_XXX type)
2671  *			. mode for ioctl(9e)
2672  *			. LP64 diskaddr_t (block I/O)
2673  *	dir	  - direction of operation (READ/WRITE/BOTH)
2674  *
2675  * Return Codes:
2676  *	0
2677  *	EAGAIN
2678  *	ECONNRESET
2679  *	ENXIO
2680  */
2681 static int
2682 vdc_populate_descriptor(vdc_t *vdcp, int operation, caddr_t addr,
2683     size_t nbytes, int slice, diskaddr_t offset, int cb_type,
2684     void *cb_arg, vio_desc_direction_t dir)
2685 {
2686 	vdc_local_desc_t	*local_dep = NULL; /* Local Dring Pointer */
2687 	int			idx;		/* Index of DRing entry used */
2688 	int			next_idx;
2689 	vio_dring_msg_t		dmsg;
2690 	size_t			msglen;
2691 	int			rv;
2692 
2693 	ASSERT(MUTEX_HELD(&vdcp->lock));
2694 	vdcp->threads_pending++;
2695 loop:
2696 	DMSG(vdcp, 2, ": dring_curr_idx = %d\n", vdcp->dring_curr_idx);
2697 
2698 	/* Get next available D-Ring entry */
2699 	idx = vdcp->dring_curr_idx;
2700 	local_dep = &(vdcp->local_dring[idx]);
2701 
2702 	if (!local_dep->is_free) {
2703 		DMSG(vdcp, 2, "[%d]: dring full - waiting for space\n",
2704 		    vdcp->instance);
2705 		cv_wait(&vdcp->dring_free_cv, &vdcp->lock);
2706 		if (vdcp->state == VDC_STATE_RUNNING ||
2707 		    vdcp->state == VDC_STATE_HANDLE_PENDING) {
2708 			goto loop;
2709 		}
2710 		vdcp->threads_pending--;
2711 		return (ECONNRESET);
2712 	}
2713 
2714 	next_idx = idx + 1;
2715 	if (next_idx >= vdcp->dring_len)
2716 		next_idx = 0;
2717 	vdcp->dring_curr_idx = next_idx;
2718 
2719 	ASSERT(local_dep->is_free);
2720 
2721 	local_dep->operation = operation;
2722 	local_dep->addr = addr;
2723 	local_dep->nbytes = nbytes;
2724 	local_dep->slice = slice;
2725 	local_dep->offset = offset;
2726 	local_dep->cb_type = cb_type;
2727 	local_dep->cb_arg = cb_arg;
2728 	local_dep->dir = dir;
2729 
2730 	local_dep->is_free = B_FALSE;
2731 
2732 	rv = vdc_map_to_shared_dring(vdcp, idx);
2733 	if (rv) {
2734 		DMSG(vdcp, 0, "[%d]: cannot bind memory - waiting ..\n",
2735 		    vdcp->instance);
2736 		/* free the descriptor */
2737 		local_dep->is_free = B_TRUE;
2738 		vdcp->dring_curr_idx = idx;
2739 		cv_wait(&vdcp->membind_cv, &vdcp->lock);
2740 		if (vdcp->state == VDC_STATE_RUNNING ||
2741 		    vdcp->state == VDC_STATE_HANDLE_PENDING) {
2742 			goto loop;
2743 		}
2744 		vdcp->threads_pending--;
2745 		return (ECONNRESET);
2746 	}
2747 
2748 	/*
2749 	 * Send a msg with the DRing details to vds
2750 	 */
2751 	VIO_INIT_DRING_DATA_TAG(dmsg);
2752 	VDC_INIT_DRING_DATA_MSG_IDS(dmsg, vdcp);
2753 	dmsg.dring_ident = vdcp->dring_ident;
2754 	dmsg.start_idx = idx;
2755 	dmsg.end_idx = idx;
2756 	vdcp->seq_num++;
2757 
2758 	DTRACE_IO2(send, vio_dring_msg_t *, &dmsg, vdc_t *, vdcp);
2759 
2760 	DMSG(vdcp, 2, "ident=0x%lx, st=%u, end=%u, seq=%ld\n",
2761 	    vdcp->dring_ident, dmsg.start_idx, dmsg.end_idx, dmsg.seq_num);
2762 
2763 	/*
2764 	 * note we're still holding the lock here to
2765 	 * make sure the message goes out in order !!!...
2766 	 */
2767 	msglen = sizeof (dmsg);
2768 	rv = vdc_send(vdcp, (caddr_t)&dmsg, &msglen);
2769 	switch (rv) {
2770 	case ECONNRESET:
2771 		/*
2772 		 * vdc_send initiates the reset on failure.
2773 		 * Since the transaction has already been put
2774 		 * on the local dring, it will automatically get
2775 		 * retried when the channel is reset. Given that,
2776 		 * it is ok to just return success even though the
2777 		 * send failed.
2778 		 */
2779 		rv = 0;
2780 		break;
2781 
2782 	case 0: /* EOK */
2783 		DMSG(vdcp, 1, "sent via LDC: rv=%d\n", rv);
2784 		break;
2785 
2786 	default:
2787 		goto cleanup_and_exit;
2788 	}
2789 
2790 	vdcp->threads_pending--;
2791 	return (rv);
2792 
2793 cleanup_and_exit:
2794 	DMSG(vdcp, 0, "unexpected error, rv=%d\n", rv);
2795 	return (ENXIO);
2796 }
2797 
2798 /*
2799  * Function:
2800  *	vdc_do_sync_op
2801  *
2802  * Description:
2803  * 	Wrapper around vdc_populate_descriptor that blocks until the
2804  * 	response to the message is available.
2805  *
2806  * Arguments:
2807  *	vdcp	  - the soft state pointer
2808  *	operation - operation we want vds to perform (VD_OP_XXX)
2809  *	addr	  - address of data buf to be read/written.
2810  *	nbytes	  - number of bytes to read/write
2811  *	slice	  - the disk slice this request is for
2812  *	offset	  - relative disk offset
2813  *	cb_type   - type of call - STRATEGY or SYNC
2814  *	cb_arg	  - parameter to be sent to server (depends on VD_OP_XXX type)
2815  *			. mode for ioctl(9e)
2816  *			. LP64 diskaddr_t (block I/O)
2817  *	dir	  - direction of operation (READ/WRITE/BOTH)
2818  *
2819  * Return Codes:
2820  *	0
2821  *	EAGAIN
2822  *		EFAULT
2823  *		ENXIO
2824  *		EIO
2825  */
2826 static int
2827 vdc_do_sync_op(vdc_t *vdcp, int operation, caddr_t addr, size_t nbytes,
2828     int slice, diskaddr_t offset, int cb_type, void *cb_arg,
2829     vio_desc_direction_t dir)
2830 {
2831 	int status;
2832 
2833 	ASSERT(cb_type == CB_SYNC);
2834 
2835 	/*
2836 	 * Grab the lock, if blocked wait until the server
2837 	 * response causes us to wake up again.
2838 	 */
2839 	mutex_enter(&vdcp->lock);
2840 	vdcp->sync_op_cnt++;
2841 	while (vdcp->sync_op_blocked && vdcp->state != VDC_STATE_DETACH)
2842 		cv_wait(&vdcp->sync_blocked_cv, &vdcp->lock);
2843 
2844 	if (vdcp->state == VDC_STATE_DETACH) {
2845 		cv_broadcast(&vdcp->sync_blocked_cv);
2846 		vdcp->sync_op_cnt--;
2847 		mutex_exit(&vdcp->lock);
2848 		return (ENXIO);
2849 	}
2850 
2851 	/* now block anyone other thread entering after us */
2852 	vdcp->sync_op_blocked = B_TRUE;
2853 	vdcp->sync_op_pending = B_TRUE;
2854 	mutex_exit(&vdcp->lock);
2855 
2856 	status = vdc_send_request(vdcp, operation, addr,
2857 	    nbytes, slice, offset, cb_type, cb_arg, dir);
2858 
2859 	mutex_enter(&vdcp->lock);
2860 
2861 	if (status != 0) {
2862 		vdcp->sync_op_pending = B_FALSE;
2863 	} else {
2864 		/*
2865 		 * block until our transaction completes.
2866 		 * Also anyone else waiting also gets to go next.
2867 		 */
2868 		while (vdcp->sync_op_pending && vdcp->state != VDC_STATE_DETACH)
2869 			cv_wait(&vdcp->sync_pending_cv, &vdcp->lock);
2870 
2871 		DMSG(vdcp, 2, ": operation returned %d\n",
2872 		    vdcp->sync_op_status);
2873 		if (vdcp->state == VDC_STATE_DETACH) {
2874 			vdcp->sync_op_pending = B_FALSE;
2875 			status = ENXIO;
2876 		} else {
2877 			status = vdcp->sync_op_status;
2878 		}
2879 	}
2880 
2881 	vdcp->sync_op_status = 0;
2882 	vdcp->sync_op_blocked = B_FALSE;
2883 	vdcp->sync_op_cnt--;
2884 
2885 	/* signal the next waiting thread */
2886 	cv_signal(&vdcp->sync_blocked_cv);
2887 	mutex_exit(&vdcp->lock);
2888 
2889 	return (status);
2890 }
2891 
2892 
2893 /*
2894  * Function:
2895  *	vdc_drain_response()
2896  *
2897  * Description:
2898  * 	When a guest is panicking, the completion of requests needs to be
2899  * 	handled differently because interrupts are disabled and vdc
2900  * 	will not get messages. We have to poll for the messages instead.
2901  *
2902  * Arguments:
2903  *	vdc	- soft state pointer for this instance of the device driver.
2904  *
2905  * Return Code:
2906  *	0	- Success
2907  */
2908 static int
2909 vdc_drain_response(vdc_t *vdc)
2910 {
2911 	int 			rv, idx, retries;
2912 	size_t			msglen;
2913 	vdc_local_desc_t 	*ldep = NULL;	/* Local Dring Entry Pointer */
2914 	vio_dring_msg_t		dmsg;
2915 
2916 	mutex_enter(&vdc->lock);
2917 
2918 	retries = 0;
2919 	for (;;) {
2920 		msglen = sizeof (dmsg);
2921 		rv = ldc_read(vdc->ldc_handle, (caddr_t)&dmsg, &msglen);
2922 		if (rv) {
2923 			rv = EINVAL;
2924 			break;
2925 		}
2926 
2927 		/*
2928 		 * if there are no packets wait and check again
2929 		 */
2930 		if ((rv == 0) && (msglen == 0)) {
2931 			if (retries++ > vdc_dump_retries) {
2932 				rv = EAGAIN;
2933 				break;
2934 			}
2935 
2936 			drv_usecwait(vdc_usec_timeout_dump);
2937 			continue;
2938 		}
2939 
2940 		/*
2941 		 * Ignore all messages that are not ACKs/NACKs to
2942 		 * DRing requests.
2943 		 */
2944 		if ((dmsg.tag.vio_msgtype != VIO_TYPE_DATA) ||
2945 		    (dmsg.tag.vio_subtype_env != VIO_DRING_DATA)) {
2946 			DMSG(vdc, 0, "discard pkt: type=%d sub=%d env=%d\n",
2947 			    dmsg.tag.vio_msgtype,
2948 			    dmsg.tag.vio_subtype,
2949 			    dmsg.tag.vio_subtype_env);
2950 			continue;
2951 		}
2952 
2953 		/*
2954 		 * set the appropriate return value for the current request.
2955 		 */
2956 		switch (dmsg.tag.vio_subtype) {
2957 		case VIO_SUBTYPE_ACK:
2958 			rv = 0;
2959 			break;
2960 		case VIO_SUBTYPE_NACK:
2961 			rv = EAGAIN;
2962 			break;
2963 		default:
2964 			continue;
2965 		}
2966 
2967 		idx = dmsg.start_idx;
2968 		if (idx >= vdc->dring_len) {
2969 			DMSG(vdc, 0, "[%d] Bogus ack data : start %d\n",
2970 			    vdc->instance, idx);
2971 			continue;
2972 		}
2973 		ldep = &vdc->local_dring[idx];
2974 		if (ldep->dep->hdr.dstate != VIO_DESC_DONE) {
2975 			DMSG(vdc, 0, "[%d] Entry @ %d - state !DONE %d\n",
2976 			    vdc->instance, idx, ldep->dep->hdr.dstate);
2977 			continue;
2978 		}
2979 
2980 		DMSG(vdc, 1, "[%d] Depopulating idx=%d state=%d\n",
2981 		    vdc->instance, idx, ldep->dep->hdr.dstate);
2982 		rv = vdc_depopulate_descriptor(vdc, idx);
2983 		if (rv) {
2984 			DMSG(vdc, 0,
2985 			    "[%d] Entry @ %d - depopulate failed ..\n",
2986 			    vdc->instance, idx);
2987 		}
2988 
2989 		/* if this is the last descriptor - break out of loop */
2990 		if ((idx + 1) % vdc->dring_len == vdc->dring_curr_idx)
2991 			break;
2992 	}
2993 
2994 	mutex_exit(&vdc->lock);
2995 	DMSG(vdc, 0, "End idx=%d\n", idx);
2996 
2997 	return (rv);
2998 }
2999 
3000 
3001 /*
3002  * Function:
3003  *	vdc_depopulate_descriptor()
3004  *
3005  * Description:
3006  *
3007  * Arguments:
3008  *	vdc	- soft state pointer for this instance of the device driver.
3009  *	idx	- Index of the Descriptor Ring entry being modified
3010  *
3011  * Return Code:
3012  *	0	- Success
3013  */
3014 static int
3015 vdc_depopulate_descriptor(vdc_t *vdc, uint_t idx)
3016 {
3017 	vd_dring_entry_t *dep = NULL;		/* Dring Entry Pointer */
3018 	vdc_local_desc_t *ldep = NULL;		/* Local Dring Entry Pointer */
3019 	int		status = ENXIO;
3020 	int		rv = 0;
3021 
3022 	ASSERT(vdc != NULL);
3023 	ASSERT(idx < vdc->dring_len);
3024 	ldep = &vdc->local_dring[idx];
3025 	ASSERT(ldep != NULL);
3026 	ASSERT(MUTEX_HELD(&vdc->lock));
3027 
3028 	DMSG(vdc, 2, ": idx = %d\n", idx);
3029 	dep = ldep->dep;
3030 	ASSERT(dep != NULL);
3031 	ASSERT((dep->hdr.dstate == VIO_DESC_DONE) ||
3032 	    (dep->payload.status == ECANCELED));
3033 
3034 	VDC_MARK_DRING_ENTRY_FREE(vdc, idx);
3035 
3036 	ldep->is_free = B_TRUE;
3037 	status = dep->payload.status;
3038 	DMSG(vdc, 2, ": is_free = %d : status = %d\n", ldep->is_free, status);
3039 
3040 	/*
3041 	 * If no buffers were used to transfer information to the server when
3042 	 * populating the descriptor then no memory handles need to be unbound
3043 	 * and we can return now.
3044 	 */
3045 	if (ldep->nbytes == 0) {
3046 		cv_signal(&vdc->dring_free_cv);
3047 		return (status);
3048 	}
3049 
3050 	/*
3051 	 * If the upper layer passed in a misaligned address we copied the
3052 	 * data into an aligned buffer before sending it to LDC - we now
3053 	 * copy it back to the original buffer.
3054 	 */
3055 	if (ldep->align_addr) {
3056 		ASSERT(ldep->addr != NULL);
3057 
3058 		if (dep->payload.nbytes > 0)
3059 			bcopy(ldep->align_addr, ldep->addr,
3060 			    dep->payload.nbytes);
3061 		kmem_free(ldep->align_addr,
3062 		    sizeof (caddr_t) * P2ROUNDUP(ldep->nbytes, 8));
3063 		ldep->align_addr = NULL;
3064 	}
3065 
3066 	rv = ldc_mem_unbind_handle(ldep->desc_mhdl);
3067 	if (rv != 0) {
3068 		DMSG(vdc, 0, "?[%d] unbind mhdl 0x%lx @ idx %d failed (%d)",
3069 		    vdc->instance, ldep->desc_mhdl, idx, rv);
3070 		/*
3071 		 * The error returned by the vDisk server is more informative
3072 		 * and thus has a higher priority but if it isn't set we ensure
3073 		 * that this function returns an error.
3074 		 */
3075 		if (status == 0)
3076 			status = EINVAL;
3077 	}
3078 
3079 	cv_signal(&vdc->membind_cv);
3080 	cv_signal(&vdc->dring_free_cv);
3081 
3082 	return (status);
3083 }
3084 
3085 /*
3086  * Function:
3087  *	vdc_populate_mem_hdl()
3088  *
3089  * Description:
3090  *
3091  * Arguments:
3092  *	vdc	- soft state pointer for this instance of the device driver.
3093  *	idx	- Index of the Descriptor Ring entry being modified
3094  *	addr	- virtual address being mapped in
3095  *	nybtes	- number of bytes in 'addr'
3096  *	operation - the vDisk operation being performed (VD_OP_xxx)
3097  *
3098  * Return Code:
3099  *	0	- Success
3100  */
3101 static int
3102 vdc_populate_mem_hdl(vdc_t *vdcp, vdc_local_desc_t *ldep)
3103 {
3104 	vd_dring_entry_t	*dep = NULL;
3105 	ldc_mem_handle_t	mhdl;
3106 	caddr_t			vaddr;
3107 	size_t			nbytes;
3108 	uint8_t			perm = LDC_MEM_RW;
3109 	uint8_t			maptype;
3110 	int			rv = 0;
3111 	int			i;
3112 
3113 	ASSERT(vdcp != NULL);
3114 
3115 	dep = ldep->dep;
3116 	mhdl = ldep->desc_mhdl;
3117 
3118 	switch (ldep->dir) {
3119 	case VIO_read_dir:
3120 		perm = LDC_MEM_W;
3121 		break;
3122 
3123 	case VIO_write_dir:
3124 		perm = LDC_MEM_R;
3125 		break;
3126 
3127 	case VIO_both_dir:
3128 		perm = LDC_MEM_RW;
3129 		break;
3130 
3131 	default:
3132 		ASSERT(0);	/* catch bad programming in vdc */
3133 	}
3134 
3135 	/*
3136 	 * LDC expects any addresses passed in to be 8-byte aligned. We need
3137 	 * to copy the contents of any misaligned buffers to a newly allocated
3138 	 * buffer and bind it instead (and copy the the contents back to the
3139 	 * original buffer passed in when depopulating the descriptor)
3140 	 */
3141 	vaddr = ldep->addr;
3142 	nbytes = ldep->nbytes;
3143 	if (((uint64_t)vaddr & 0x7) != 0) {
3144 		ASSERT(ldep->align_addr == NULL);
3145 		ldep->align_addr =
3146 		    kmem_alloc(sizeof (caddr_t) *
3147 		    P2ROUNDUP(nbytes, 8), KM_SLEEP);
3148 		DMSG(vdcp, 0, "[%d] Misaligned address %p reallocating "
3149 		    "(buf=%p nb=%ld op=%d)\n",
3150 		    vdcp->instance, (void *)vaddr, (void *)ldep->align_addr,
3151 		    nbytes, ldep->operation);
3152 		if (perm != LDC_MEM_W)
3153 			bcopy(vaddr, ldep->align_addr, nbytes);
3154 		vaddr = ldep->align_addr;
3155 	}
3156 
3157 	maptype = LDC_IO_MAP|LDC_SHADOW_MAP|LDC_DIRECT_MAP;
3158 	rv = ldc_mem_bind_handle(mhdl, vaddr, P2ROUNDUP(nbytes, 8),
3159 	    maptype, perm, &dep->payload.cookie[0], &dep->payload.ncookies);
3160 	DMSG(vdcp, 2, "[%d] bound mem handle; ncookies=%d\n",
3161 	    vdcp->instance, dep->payload.ncookies);
3162 	if (rv != 0) {
3163 		DMSG(vdcp, 0, "[%d] Failed to bind LDC memory handle "
3164 		    "(mhdl=%p, buf=%p, err=%d)\n",
3165 		    vdcp->instance, (void *)mhdl, (void *)vaddr, rv);
3166 		if (ldep->align_addr) {
3167 			kmem_free(ldep->align_addr,
3168 			    sizeof (caddr_t) * P2ROUNDUP(nbytes, 8));
3169 			ldep->align_addr = NULL;
3170 		}
3171 		return (EAGAIN);
3172 	}
3173 
3174 	/*
3175 	 * Get the other cookies (if any).
3176 	 */
3177 	for (i = 1; i < dep->payload.ncookies; i++) {
3178 		rv = ldc_mem_nextcookie(mhdl, &dep->payload.cookie[i]);
3179 		if (rv != 0) {
3180 			(void) ldc_mem_unbind_handle(mhdl);
3181 			DMSG(vdcp, 0, "?[%d] Failed to get next cookie "
3182 			    "(mhdl=%lx cnum=%d), err=%d",
3183 			    vdcp->instance, mhdl, i, rv);
3184 			if (ldep->align_addr) {
3185 				kmem_free(ldep->align_addr,
3186 				    sizeof (caddr_t) * ldep->nbytes);
3187 				ldep->align_addr = NULL;
3188 			}
3189 			return (EAGAIN);
3190 		}
3191 	}
3192 
3193 	return (rv);
3194 }
3195 
3196 /*
3197  * Interrupt handlers for messages from LDC
3198  */
3199 
3200 /*
3201  * Function:
3202  *	vdc_handle_cb()
3203  *
3204  * Description:
3205  *
3206  * Arguments:
3207  *	event	- Type of event (LDC_EVT_xxx) that triggered the callback
3208  *	arg	- soft state pointer for this instance of the device driver.
3209  *
3210  * Return Code:
3211  *	0	- Success
3212  */
3213 static uint_t
3214 vdc_handle_cb(uint64_t event, caddr_t arg)
3215 {
3216 	ldc_status_t	ldc_state;
3217 	int		rv = 0;
3218 
3219 	vdc_t	*vdc = (vdc_t *)(void *)arg;
3220 
3221 	ASSERT(vdc != NULL);
3222 
3223 	DMSG(vdc, 1, "evt=%lx seqID=%ld\n", event, vdc->seq_num);
3224 
3225 	/*
3226 	 * Depending on the type of event that triggered this callback,
3227 	 * we modify the handshake state or read the data.
3228 	 *
3229 	 * NOTE: not done as a switch() as event could be triggered by
3230 	 * a state change and a read request. Also the ordering	of the
3231 	 * check for the event types is deliberate.
3232 	 */
3233 	if (event & LDC_EVT_UP) {
3234 		DMSG(vdc, 0, "[%d] Received LDC_EVT_UP\n", vdc->instance);
3235 
3236 		mutex_enter(&vdc->lock);
3237 
3238 		/* get LDC state */
3239 		rv = ldc_status(vdc->ldc_handle, &ldc_state);
3240 		if (rv != 0) {
3241 			DMSG(vdc, 0, "[%d] Couldn't get LDC status %d",
3242 			    vdc->instance, rv);
3243 			return (LDC_SUCCESS);
3244 		}
3245 		if (vdc->ldc_state != LDC_UP && ldc_state == LDC_UP) {
3246 			/*
3247 			 * Reset the transaction sequence numbers when
3248 			 * LDC comes up. We then kick off the handshake
3249 			 * negotiation with the vDisk server.
3250 			 */
3251 			vdc->seq_num = 1;
3252 			vdc->seq_num_reply = 0;
3253 			vdc->ldc_state = ldc_state;
3254 			cv_signal(&vdc->initwait_cv);
3255 		}
3256 
3257 		mutex_exit(&vdc->lock);
3258 	}
3259 
3260 	if (event & LDC_EVT_READ) {
3261 		DMSG(vdc, 1, "[%d] Received LDC_EVT_READ\n", vdc->instance);
3262 		mutex_enter(&vdc->read_lock);
3263 		cv_signal(&vdc->read_cv);
3264 		vdc->read_state = VDC_READ_PENDING;
3265 		mutex_exit(&vdc->read_lock);
3266 
3267 		/* that's all we have to do - no need to handle DOWN/RESET */
3268 		return (LDC_SUCCESS);
3269 	}
3270 
3271 	if (event & (LDC_EVT_RESET|LDC_EVT_DOWN)) {
3272 
3273 		DMSG(vdc, 0, "[%d] Received LDC RESET event\n", vdc->instance);
3274 
3275 		mutex_enter(&vdc->lock);
3276 		/*
3277 		 * Need to wake up any readers so they will
3278 		 * detect that a reset has occurred.
3279 		 */
3280 		mutex_enter(&vdc->read_lock);
3281 		if ((vdc->read_state == VDC_READ_WAITING) ||
3282 		    (vdc->read_state == VDC_READ_RESET))
3283 			cv_signal(&vdc->read_cv);
3284 		vdc->read_state = VDC_READ_RESET;
3285 		mutex_exit(&vdc->read_lock);
3286 
3287 		/* wake up any threads waiting for connection to come up */
3288 		if (vdc->state == VDC_STATE_INIT_WAITING) {
3289 			vdc->state = VDC_STATE_RESETTING;
3290 			cv_signal(&vdc->initwait_cv);
3291 		}
3292 
3293 		mutex_exit(&vdc->lock);
3294 	}
3295 
3296 	if (event & ~(LDC_EVT_UP | LDC_EVT_RESET | LDC_EVT_DOWN | LDC_EVT_READ))
3297 		DMSG(vdc, 0, "![%d] Unexpected LDC event (%lx) received",
3298 		    vdc->instance, event);
3299 
3300 	return (LDC_SUCCESS);
3301 }
3302 
3303 /*
3304  * Function:
3305  *	vdc_wait_for_response()
3306  *
3307  * Description:
3308  *	Block waiting for a response from the server. If there is
3309  *	no data the thread block on the read_cv that is signalled
3310  *	by the callback when an EVT_READ occurs.
3311  *
3312  * Arguments:
3313  *	vdcp	- soft state pointer for this instance of the device driver.
3314  *
3315  * Return Code:
3316  *	0	- Success
3317  */
3318 static int
3319 vdc_wait_for_response(vdc_t *vdcp, vio_msg_t *msgp)
3320 {
3321 	size_t		nbytes = sizeof (*msgp);
3322 	int		status;
3323 
3324 	ASSERT(vdcp != NULL);
3325 
3326 	DMSG(vdcp, 1, "[%d] Entered\n", vdcp->instance);
3327 
3328 	status = vdc_recv(vdcp, msgp, &nbytes);
3329 	DMSG(vdcp, 3, "vdc_read() done.. status=0x%x size=0x%x\n",
3330 	    status, (int)nbytes);
3331 	if (status) {
3332 		DMSG(vdcp, 0, "?[%d] Error %d reading LDC msg\n",
3333 		    vdcp->instance, status);
3334 		return (status);
3335 	}
3336 
3337 	if (nbytes < sizeof (vio_msg_tag_t)) {
3338 		DMSG(vdcp, 0, "?[%d] Expect %lu bytes; recv'd %lu\n",
3339 		    vdcp->instance, sizeof (vio_msg_tag_t), nbytes);
3340 		return (ENOMSG);
3341 	}
3342 
3343 	DMSG(vdcp, 2, "[%d] (%x/%x/%x)\n", vdcp->instance,
3344 	    msgp->tag.vio_msgtype,
3345 	    msgp->tag.vio_subtype,
3346 	    msgp->tag.vio_subtype_env);
3347 
3348 	/*
3349 	 * Verify the Session ID of the message
3350 	 *
3351 	 * Every message after the Version has been negotiated should
3352 	 * have the correct session ID set.
3353 	 */
3354 	if ((msgp->tag.vio_sid != vdcp->session_id) &&
3355 	    (msgp->tag.vio_subtype_env != VIO_VER_INFO)) {
3356 		DMSG(vdcp, 0, "[%d] Invalid SID: received 0x%x, "
3357 		    "expected 0x%lx [seq num %lx @ %d]",
3358 		    vdcp->instance, msgp->tag.vio_sid,
3359 		    vdcp->session_id,
3360 		    ((vio_dring_msg_t *)msgp)->seq_num,
3361 		    ((vio_dring_msg_t *)msgp)->start_idx);
3362 		return (ENOMSG);
3363 	}
3364 	return (0);
3365 }
3366 
3367 
3368 /*
3369  * Function:
3370  *	vdc_resubmit_backup_dring()
3371  *
3372  * Description:
3373  *	Resubmit each descriptor in the backed up dring to
3374  * 	vDisk server. The Dring was backed up during connection
3375  *	reset.
3376  *
3377  * Arguments:
3378  *	vdcp	- soft state pointer for this instance of the device driver.
3379  *
3380  * Return Code:
3381  *	0	- Success
3382  */
3383 static int
3384 vdc_resubmit_backup_dring(vdc_t *vdcp)
3385 {
3386 	int		count;
3387 	int		b_idx;
3388 	int		rv;
3389 	int		dring_size;
3390 	int		status;
3391 	vio_msg_t	vio_msg;
3392 	vdc_local_desc_t	*curr_ldep;
3393 
3394 	ASSERT(MUTEX_NOT_HELD(&vdcp->lock));
3395 	ASSERT(vdcp->state == VDC_STATE_HANDLE_PENDING);
3396 
3397 	if (vdcp->local_dring_backup == NULL) {
3398 		/* the pending requests have already been processed */
3399 		return (0);
3400 	}
3401 
3402 	DMSG(vdcp, 1, "restoring pending dring entries (len=%d, tail=%d)\n",
3403 	    vdcp->local_dring_backup_len, vdcp->local_dring_backup_tail);
3404 
3405 	/*
3406 	 * Walk the backup copy of the local descriptor ring and
3407 	 * resubmit all the outstanding transactions.
3408 	 */
3409 	b_idx = vdcp->local_dring_backup_tail;
3410 	for (count = 0; count < vdcp->local_dring_backup_len; count++) {
3411 
3412 		curr_ldep = &(vdcp->local_dring_backup[b_idx]);
3413 
3414 		/* only resubmit outstanding transactions */
3415 		if (!curr_ldep->is_free) {
3416 
3417 			DMSG(vdcp, 1, "resubmitting entry idx=%x\n", b_idx);
3418 			mutex_enter(&vdcp->lock);
3419 			rv = vdc_populate_descriptor(vdcp, curr_ldep->operation,
3420 			    curr_ldep->addr, curr_ldep->nbytes,
3421 			    curr_ldep->slice, curr_ldep->offset,
3422 			    curr_ldep->cb_type, curr_ldep->cb_arg,
3423 			    curr_ldep->dir);
3424 			mutex_exit(&vdcp->lock);
3425 			if (rv) {
3426 				DMSG(vdcp, 1, "[%d] cannot resubmit entry %d\n",
3427 				    vdcp->instance, b_idx);
3428 				return (rv);
3429 			}
3430 
3431 			/* Wait for the response message. */
3432 			DMSG(vdcp, 1, "waiting for response to idx=%x\n",
3433 			    b_idx);
3434 			status = vdc_wait_for_response(vdcp, &vio_msg);
3435 			if (status) {
3436 				DMSG(vdcp, 1, "[%d] wait_for_response "
3437 				    "returned err=%d\n", vdcp->instance,
3438 				    status);
3439 				return (status);
3440 			}
3441 
3442 			DMSG(vdcp, 1, "processing msg for idx=%x\n", b_idx);
3443 			status = vdc_process_data_msg(vdcp, &vio_msg);
3444 			if (status) {
3445 				DMSG(vdcp, 1, "[%d] process_data_msg "
3446 				    "returned err=%d\n", vdcp->instance,
3447 				    status);
3448 				return (status);
3449 			}
3450 		}
3451 
3452 		/* get the next element to submit */
3453 		if (++b_idx >= vdcp->local_dring_backup_len)
3454 			b_idx = 0;
3455 	}
3456 
3457 	/* all done - now clear up pending dring copy */
3458 	dring_size = vdcp->local_dring_backup_len *
3459 	    sizeof (vdcp->local_dring_backup[0]);
3460 
3461 	(void) kmem_free(vdcp->local_dring_backup, dring_size);
3462 
3463 	vdcp->local_dring_backup = NULL;
3464 
3465 	return (0);
3466 }
3467 
3468 /*
3469  * Function:
3470  *	vdc_cancel_backup_dring
3471  *
3472  * Description:
3473  *	Cancel each descriptor in the backed up dring to vDisk server.
3474  *	The Dring was backed up during connection reset.
3475  *
3476  * Arguments:
3477  *	vdcp	- soft state pointer for this instance of the device driver.
3478  *
3479  * Return Code:
3480  *	None
3481  */
3482 void
3483 vdc_cancel_backup_ring(vdc_t *vdcp)
3484 {
3485 	vdc_local_desc_t *ldep;
3486 	struct buf 	*bufp;
3487 	int		count;
3488 	int		b_idx;
3489 	int		dring_size;
3490 
3491 	ASSERT(MUTEX_HELD(&vdcp->lock));
3492 	ASSERT(vdcp->state == VDC_STATE_INIT ||
3493 	    vdcp->state == VDC_STATE_INIT_WAITING ||
3494 	    vdcp->state == VDC_STATE_NEGOTIATE ||
3495 	    vdcp->state == VDC_STATE_RESETTING);
3496 
3497 	if (vdcp->local_dring_backup == NULL) {
3498 		/* the pending requests have already been processed */
3499 		return;
3500 	}
3501 
3502 	DMSG(vdcp, 1, "cancelling pending dring entries (len=%d, tail=%d)\n",
3503 	    vdcp->local_dring_backup_len, vdcp->local_dring_backup_tail);
3504 
3505 	/*
3506 	 * Walk the backup copy of the local descriptor ring and
3507 	 * cancel all the outstanding transactions.
3508 	 */
3509 	b_idx = vdcp->local_dring_backup_tail;
3510 	for (count = 0; count < vdcp->local_dring_backup_len; count++) {
3511 
3512 		ldep = &(vdcp->local_dring_backup[b_idx]);
3513 
3514 		/* only cancel outstanding transactions */
3515 		if (!ldep->is_free) {
3516 
3517 			DMSG(vdcp, 1, "cancelling entry idx=%x\n", b_idx);
3518 
3519 			/*
3520 			 * All requests have already been cleared from the
3521 			 * local descriptor ring and the LDC channel has been
3522 			 * reset so we will never get any reply for these
3523 			 * requests. Now we just have to notify threads waiting
3524 			 * for replies that the request has failed.
3525 			 */
3526 			switch (ldep->cb_type) {
3527 			case CB_SYNC:
3528 				ASSERT(vdcp->sync_op_pending);
3529 				vdcp->sync_op_status = EIO;
3530 				vdcp->sync_op_pending = B_FALSE;
3531 				cv_signal(&vdcp->sync_pending_cv);
3532 				break;
3533 
3534 			case CB_STRATEGY:
3535 				bufp = ldep->cb_arg;
3536 				ASSERT(bufp != NULL);
3537 				bufp->b_resid = bufp->b_bcount;
3538 				bioerror(bufp, EIO);
3539 				biodone(bufp);
3540 				break;
3541 
3542 			default:
3543 				ASSERT(0);
3544 			}
3545 
3546 		}
3547 
3548 		/* get the next element to cancel */
3549 		if (++b_idx >= vdcp->local_dring_backup_len)
3550 			b_idx = 0;
3551 	}
3552 
3553 	/* all done - now clear up pending dring copy */
3554 	dring_size = vdcp->local_dring_backup_len *
3555 	    sizeof (vdcp->local_dring_backup[0]);
3556 
3557 	(void) kmem_free(vdcp->local_dring_backup, dring_size);
3558 
3559 	vdcp->local_dring_backup = NULL;
3560 
3561 	DTRACE_IO2(processed, int, count, vdc_t *, vdcp);
3562 }
3563 
3564 /*
3565  * Function:
3566  *	vdc_connection_timeout
3567  *
3568  * Description:
3569  *	This function is invoked if the timeout set to establish the connection
3570  *	with vds expires. This will happen if we spend too much time in the
3571  *	VDC_STATE_INIT_WAITING or VDC_STATE_NEGOTIATE states. Then we will
3572  *	cancel any pending request and mark them as failed.
3573  *
3574  *	If the timeout does not expire, it will be cancelled when we reach the
3575  *	VDC_STATE_HANDLE_PENDING or VDC_STATE_RESETTING state. This function can
3576  *	be invoked while we are in the VDC_STATE_HANDLE_PENDING or
3577  *	VDC_STATE_RESETTING state in which case we do nothing because the
3578  *	timeout is being cancelled.
3579  *
3580  * Arguments:
3581  *	arg	- argument of the timeout function actually a soft state
3582  *		  pointer for the instance of the device driver.
3583  *
3584  * Return Code:
3585  *	None
3586  */
3587 void
3588 vdc_connection_timeout(void *arg)
3589 {
3590 	vdc_t 		*vdcp = (vdc_t *)arg;
3591 
3592 	mutex_enter(&vdcp->lock);
3593 
3594 	if (vdcp->state == VDC_STATE_HANDLE_PENDING ||
3595 	    vdcp->state == VDC_STATE_DETACH) {
3596 		/*
3597 		 * The connection has just been re-established or
3598 		 * we are detaching.
3599 		 */
3600 		vdcp->ctimeout_reached = B_FALSE;
3601 		mutex_exit(&vdcp->lock);
3602 		return;
3603 	}
3604 
3605 	vdcp->ctimeout_reached = B_TRUE;
3606 
3607 	/* notify requests waiting for sending */
3608 	cv_broadcast(&vdcp->running_cv);
3609 
3610 	/* cancel requests waiting for a result */
3611 	vdc_cancel_backup_ring(vdcp);
3612 
3613 	mutex_exit(&vdcp->lock);
3614 
3615 	cmn_err(CE_NOTE, "[%d] connection to service domain timeout",
3616 	    vdcp->instance);
3617 }
3618 
3619 /*
3620  * Function:
3621  *	vdc_backup_local_dring()
3622  *
3623  * Description:
3624  *	Backup the current dring in the event of a reset. The Dring
3625  *	transactions will be resubmitted to the server when the
3626  *	connection is restored.
3627  *
3628  * Arguments:
3629  *	vdcp	- soft state pointer for this instance of the device driver.
3630  *
3631  * Return Code:
3632  *	NONE
3633  */
3634 static void
3635 vdc_backup_local_dring(vdc_t *vdcp)
3636 {
3637 	int dring_size;
3638 
3639 	ASSERT(MUTEX_HELD(&vdcp->lock));
3640 	ASSERT(vdcp->state == VDC_STATE_RESETTING);
3641 
3642 	/*
3643 	 * If the backup dring is stil around, it means
3644 	 * that the last restore did not complete. However,
3645 	 * since we never got back into the running state,
3646 	 * the backup copy we have is still valid.
3647 	 */
3648 	if (vdcp->local_dring_backup != NULL) {
3649 		DMSG(vdcp, 1, "reusing local descriptor ring backup "
3650 		    "(len=%d, tail=%d)\n", vdcp->local_dring_backup_len,
3651 		    vdcp->local_dring_backup_tail);
3652 		return;
3653 	}
3654 
3655 	/*
3656 	 * The backup dring can be NULL and the local dring may not be
3657 	 * initialized. This can happen if we had a reset while establishing
3658 	 * a new connection but after the connection has timed out. In that
3659 	 * case the backup dring is NULL because the requests have been
3660 	 * cancelled and the request occured before the local dring is
3661 	 * initialized.
3662 	 */
3663 	if (!(vdcp->initialized & VDC_DRING_LOCAL))
3664 		return;
3665 
3666 	DMSG(vdcp, 1, "backing up the local descriptor ring (len=%d, "
3667 	    "tail=%d)\n", vdcp->dring_len, vdcp->dring_curr_idx);
3668 
3669 	dring_size = vdcp->dring_len * sizeof (vdcp->local_dring[0]);
3670 
3671 	vdcp->local_dring_backup = kmem_alloc(dring_size, KM_SLEEP);
3672 	bcopy(vdcp->local_dring, vdcp->local_dring_backup, dring_size);
3673 
3674 	vdcp->local_dring_backup_tail = vdcp->dring_curr_idx;
3675 	vdcp->local_dring_backup_len = vdcp->dring_len;
3676 }
3677 
3678 /* -------------------------------------------------------------------------- */
3679 
3680 /*
3681  * The following functions process the incoming messages from vds
3682  */
3683 
3684 /*
3685  * Function:
3686  *      vdc_process_msg_thread()
3687  *
3688  * Description:
3689  *
3690  *	Main VDC message processing thread. Each vDisk instance
3691  * 	consists of a copy of this thread. This thread triggers
3692  * 	all the handshakes and data exchange with the server. It
3693  * 	also handles all channel resets
3694  *
3695  * Arguments:
3696  *      vdc     - soft state pointer for this instance of the device driver.
3697  *
3698  * Return Code:
3699  *      None
3700  */
3701 static void
3702 vdc_process_msg_thread(vdc_t *vdcp)
3703 {
3704 	int	status;
3705 	int	ctimeout;
3706 	timeout_id_t tmid = 0;
3707 
3708 	mutex_enter(&vdcp->lock);
3709 
3710 	for (;;) {
3711 
3712 #define	Q(_s)	(vdcp->state == _s) ? #_s :
3713 		DMSG(vdcp, 3, "state = %d (%s)\n", vdcp->state,
3714 		    Q(VDC_STATE_INIT)
3715 		    Q(VDC_STATE_INIT_WAITING)
3716 		    Q(VDC_STATE_NEGOTIATE)
3717 		    Q(VDC_STATE_HANDLE_PENDING)
3718 		    Q(VDC_STATE_RUNNING)
3719 		    Q(VDC_STATE_RESETTING)
3720 		    Q(VDC_STATE_DETACH)
3721 		    "UNKNOWN");
3722 
3723 		switch (vdcp->state) {
3724 		case VDC_STATE_INIT:
3725 
3726 			/*
3727 			 * If requested, start a timeout to check if the
3728 			 * connection with vds is established in the
3729 			 * specified delay. If the timeout expires, we
3730 			 * will cancel any pending request.
3731 			 *
3732 			 * If some reset have occurred while establishing
3733 			 * the connection, we already have a timeout armed
3734 			 * and in that case we don't need to arm a new one.
3735 			 */
3736 			ctimeout = (vdc_timeout != 0)?
3737 			    vdc_timeout : vdcp->ctimeout;
3738 
3739 			if (ctimeout != 0 && tmid == 0) {
3740 				tmid = timeout(vdc_connection_timeout, vdcp,
3741 				    ctimeout * drv_usectohz(1000000));
3742 			}
3743 
3744 			/* Check if have re-initializing repeatedly */
3745 			if (vdcp->hshake_cnt++ > vdc_hshake_retries &&
3746 			    vdcp->lifecycle != VDC_LC_ONLINE) {
3747 				cmn_err(CE_NOTE, "[%d] disk access failed.\n",
3748 				    vdcp->instance);
3749 				vdcp->state = VDC_STATE_DETACH;
3750 				break;
3751 			}
3752 
3753 			/* Bring up connection with vds via LDC */
3754 			status = vdc_start_ldc_connection(vdcp);
3755 			if (status == EINVAL) {
3756 				DMSG(vdcp, 0, "[%d] Could not start LDC",
3757 				    vdcp->instance);
3758 				vdcp->state = VDC_STATE_DETACH;
3759 			} else {
3760 				vdcp->state = VDC_STATE_INIT_WAITING;
3761 			}
3762 			break;
3763 
3764 		case VDC_STATE_INIT_WAITING:
3765 
3766 			/*
3767 			 * Let the callback event move us on
3768 			 * when channel is open to server
3769 			 */
3770 			while (vdcp->ldc_state != LDC_UP) {
3771 				cv_wait(&vdcp->initwait_cv, &vdcp->lock);
3772 				if (vdcp->state != VDC_STATE_INIT_WAITING) {
3773 					DMSG(vdcp, 0,
3774 				"state moved to %d out from under us...\n",
3775 					    vdcp->state);
3776 
3777 					break;
3778 				}
3779 			}
3780 			if (vdcp->state == VDC_STATE_INIT_WAITING &&
3781 			    vdcp->ldc_state == LDC_UP) {
3782 				vdcp->state = VDC_STATE_NEGOTIATE;
3783 			}
3784 			break;
3785 
3786 		case VDC_STATE_NEGOTIATE:
3787 			switch (status = vdc_ver_negotiation(vdcp)) {
3788 			case 0:
3789 				break;
3790 			default:
3791 				DMSG(vdcp, 0, "ver negotiate failed (%d)..\n",
3792 				    status);
3793 				goto reset;
3794 			}
3795 
3796 			switch (status = vdc_attr_negotiation(vdcp)) {
3797 			case 0:
3798 				break;
3799 			default:
3800 				DMSG(vdcp, 0, "attr negotiate failed (%d)..\n",
3801 				    status);
3802 				goto reset;
3803 			}
3804 
3805 			switch (status = vdc_dring_negotiation(vdcp)) {
3806 			case 0:
3807 				break;
3808 			default:
3809 				DMSG(vdcp, 0, "dring negotiate failed (%d)..\n",
3810 				    status);
3811 				goto reset;
3812 			}
3813 
3814 			switch (status = vdc_rdx_exchange(vdcp)) {
3815 			case 0:
3816 				vdcp->state = VDC_STATE_HANDLE_PENDING;
3817 				goto done;
3818 			default:
3819 				DMSG(vdcp, 0, "RDX xchg failed ..(%d)\n",
3820 				    status);
3821 				goto reset;
3822 			}
3823 reset:
3824 			DMSG(vdcp, 0, "negotiation failed: resetting (%d)\n",
3825 			    status);
3826 			vdcp->state = VDC_STATE_RESETTING;
3827 			vdcp->self_reset = B_TRUE;
3828 done:
3829 			DMSG(vdcp, 0, "negotiation complete (state=0x%x)...\n",
3830 			    vdcp->state);
3831 			break;
3832 
3833 		case VDC_STATE_HANDLE_PENDING:
3834 
3835 			if (vdcp->ctimeout_reached) {
3836 				/*
3837 				 * The connection timeout had been reached so
3838 				 * pending requests have been cancelled. Now
3839 				 * that the connection is back we can reset
3840 				 * the timeout.
3841 				 */
3842 				ASSERT(vdcp->local_dring_backup == NULL);
3843 				ASSERT(tmid != 0);
3844 				tmid = 0;
3845 				vdcp->ctimeout_reached = B_FALSE;
3846 				vdcp->state = VDC_STATE_RUNNING;
3847 				DMSG(vdcp, 0, "[%d] connection to service "
3848 				    "domain is up", vdcp->instance);
3849 				break;
3850 			}
3851 
3852 			mutex_exit(&vdcp->lock);
3853 			if (tmid != 0) {
3854 				(void) untimeout(tmid);
3855 				tmid = 0;
3856 			}
3857 			status = vdc_resubmit_backup_dring(vdcp);
3858 			mutex_enter(&vdcp->lock);
3859 
3860 			if (status)
3861 				vdcp->state = VDC_STATE_RESETTING;
3862 			else
3863 				vdcp->state = VDC_STATE_RUNNING;
3864 
3865 			break;
3866 
3867 		/* enter running state */
3868 		case VDC_STATE_RUNNING:
3869 			/*
3870 			 * Signal anyone waiting for the connection
3871 			 * to come on line.
3872 			 */
3873 			vdcp->hshake_cnt = 0;
3874 			cv_broadcast(&vdcp->running_cv);
3875 			mutex_exit(&vdcp->lock);
3876 
3877 			for (;;) {
3878 				vio_msg_t msg;
3879 				status = vdc_wait_for_response(vdcp, &msg);
3880 				if (status) break;
3881 
3882 				DMSG(vdcp, 1, "[%d] new pkt(s) available\n",
3883 				    vdcp->instance);
3884 				status = vdc_process_data_msg(vdcp, &msg);
3885 				if (status) {
3886 					DMSG(vdcp, 1, "[%d] process_data_msg "
3887 					    "returned err=%d\n", vdcp->instance,
3888 					    status);
3889 					break;
3890 				}
3891 
3892 			}
3893 
3894 			mutex_enter(&vdcp->lock);
3895 
3896 			vdcp->state = VDC_STATE_RESETTING;
3897 			vdcp->self_reset = B_TRUE;
3898 			break;
3899 
3900 		case VDC_STATE_RESETTING:
3901 			/*
3902 			 * When we reach this state, we either come from the
3903 			 * VDC_STATE_RUNNING state and we can have pending
3904 			 * request but no timeout is armed; or we come from
3905 			 * the VDC_STATE_INIT_WAITING, VDC_NEGOTIATE or
3906 			 * VDC_HANDLE_PENDING state and there is no pending
3907 			 * request or pending requests have already been copied
3908 			 * into the backup dring. So we can safely keep the
3909 			 * connection timeout armed while we are in this state.
3910 			 */
3911 
3912 			DMSG(vdcp, 0, "Initiating channel reset "
3913 			    "(pending = %d)\n", (int)vdcp->threads_pending);
3914 
3915 			if (vdcp->self_reset) {
3916 				DMSG(vdcp, 0,
3917 				    "[%d] calling stop_ldc_connection.\n",
3918 				    vdcp->instance);
3919 				status = vdc_stop_ldc_connection(vdcp);
3920 				vdcp->self_reset = B_FALSE;
3921 			}
3922 
3923 			/*
3924 			 * Wait for all threads currently waiting
3925 			 * for a free dring entry to use.
3926 			 */
3927 			while (vdcp->threads_pending) {
3928 				cv_broadcast(&vdcp->membind_cv);
3929 				cv_broadcast(&vdcp->dring_free_cv);
3930 				mutex_exit(&vdcp->lock);
3931 				/* give the waiters enough time to wake up */
3932 				delay(vdc_hz_min_ldc_delay);
3933 				mutex_enter(&vdcp->lock);
3934 			}
3935 
3936 			ASSERT(vdcp->threads_pending == 0);
3937 
3938 			/* Sanity check that no thread is receiving */
3939 			ASSERT(vdcp->read_state != VDC_READ_WAITING);
3940 
3941 			vdcp->read_state = VDC_READ_IDLE;
3942 
3943 			vdc_backup_local_dring(vdcp);
3944 
3945 			/* cleanup the old d-ring */
3946 			vdc_destroy_descriptor_ring(vdcp);
3947 
3948 			/* go and start again */
3949 			vdcp->state = VDC_STATE_INIT;
3950 
3951 			break;
3952 
3953 		case VDC_STATE_DETACH:
3954 			DMSG(vdcp, 0, "[%d] Reset thread exit cleanup ..\n",
3955 			    vdcp->instance);
3956 
3957 			/* cancel any pending timeout */
3958 			mutex_exit(&vdcp->lock);
3959 			if (tmid != 0) {
3960 				(void) untimeout(tmid);
3961 				tmid = 0;
3962 			}
3963 			mutex_enter(&vdcp->lock);
3964 
3965 			/*
3966 			 * Signal anyone waiting for connection
3967 			 * to come online
3968 			 */
3969 			cv_broadcast(&vdcp->running_cv);
3970 
3971 			while (vdcp->sync_op_pending) {
3972 				cv_signal(&vdcp->sync_pending_cv);
3973 				cv_signal(&vdcp->sync_blocked_cv);
3974 				mutex_exit(&vdcp->lock);
3975 				/* give the waiters enough time to wake up */
3976 				delay(vdc_hz_min_ldc_delay);
3977 				mutex_enter(&vdcp->lock);
3978 			}
3979 
3980 			mutex_exit(&vdcp->lock);
3981 
3982 			DMSG(vdcp, 0, "[%d] Msg processing thread exiting ..\n",
3983 			    vdcp->instance);
3984 			thread_exit();
3985 			break;
3986 		}
3987 	}
3988 }
3989 
3990 
3991 /*
3992  * Function:
3993  *	vdc_process_data_msg()
3994  *
3995  * Description:
3996  *	This function is called by the message processing thread each time
3997  *	a message with a msgtype of VIO_TYPE_DATA is received. It will either
3998  *	be an ACK or NACK from vds[1] which vdc handles as follows.
3999  *		ACK	- wake up the waiting thread
4000  *		NACK	- resend any messages necessary
4001  *
4002  *	[1] Although the message format allows it, vds should not send a
4003  *	    VIO_SUBTYPE_INFO message to vdc asking it to read data; if for
4004  *	    some bizarre reason it does, vdc will reset the connection.
4005  *
4006  * Arguments:
4007  *	vdc	- soft state pointer for this instance of the device driver.
4008  *	msg	- the LDC message sent by vds
4009  *
4010  * Return Code:
4011  *	0	- Success.
4012  *	> 0	- error value returned by LDC
4013  */
4014 static int
4015 vdc_process_data_msg(vdc_t *vdcp, vio_msg_t *msg)
4016 {
4017 	int			status = 0;
4018 	vio_dring_msg_t		*dring_msg;
4019 	vdc_local_desc_t	*ldep = NULL;
4020 	int			start, end;
4021 	int			idx;
4022 
4023 	dring_msg = (vio_dring_msg_t *)msg;
4024 
4025 	ASSERT(msg->tag.vio_msgtype == VIO_TYPE_DATA);
4026 	ASSERT(vdcp != NULL);
4027 
4028 	mutex_enter(&vdcp->lock);
4029 
4030 	/*
4031 	 * Check to see if the message has bogus data
4032 	 */
4033 	idx = start = dring_msg->start_idx;
4034 	end = dring_msg->end_idx;
4035 	if ((start >= vdcp->dring_len) ||
4036 	    (end >= vdcp->dring_len) || (end < -1)) {
4037 		DMSG(vdcp, 0, "[%d] Bogus ACK data : start %d, end %d\n",
4038 		    vdcp->instance, start, end);
4039 		mutex_exit(&vdcp->lock);
4040 		return (EINVAL);
4041 	}
4042 
4043 	/*
4044 	 * Verify that the sequence number is what vdc expects.
4045 	 */
4046 	switch (vdc_verify_seq_num(vdcp, dring_msg)) {
4047 	case VDC_SEQ_NUM_TODO:
4048 		break;	/* keep processing this message */
4049 	case VDC_SEQ_NUM_SKIP:
4050 		mutex_exit(&vdcp->lock);
4051 		return (0);
4052 	case VDC_SEQ_NUM_INVALID:
4053 		mutex_exit(&vdcp->lock);
4054 		DMSG(vdcp, 0, "[%d] invalid seqno\n", vdcp->instance);
4055 		return (ENXIO);
4056 	}
4057 
4058 	if (msg->tag.vio_subtype == VIO_SUBTYPE_NACK) {
4059 		DMSG(vdcp, 0, "[%d] DATA NACK\n", vdcp->instance);
4060 		VDC_DUMP_DRING_MSG(dring_msg);
4061 		mutex_exit(&vdcp->lock);
4062 		return (EIO);
4063 
4064 	} else if (msg->tag.vio_subtype == VIO_SUBTYPE_INFO) {
4065 		mutex_exit(&vdcp->lock);
4066 		return (EPROTO);
4067 	}
4068 
4069 	DTRACE_IO2(recv, vio_dring_msg_t, dring_msg, vdc_t *, vdcp);
4070 	DMSG(vdcp, 1, ": start %d end %d\n", start, end);
4071 	ASSERT(start == end);
4072 
4073 	ldep = &vdcp->local_dring[idx];
4074 
4075 	DMSG(vdcp, 1, ": state 0x%x - cb_type 0x%x\n",
4076 	    ldep->dep->hdr.dstate, ldep->cb_type);
4077 
4078 	if (ldep->dep->hdr.dstate == VIO_DESC_DONE) {
4079 		struct buf *bufp;
4080 
4081 		switch (ldep->cb_type) {
4082 		case CB_SYNC:
4083 			ASSERT(vdcp->sync_op_pending);
4084 
4085 			status = vdc_depopulate_descriptor(vdcp, idx);
4086 			vdcp->sync_op_status = status;
4087 			vdcp->sync_op_pending = B_FALSE;
4088 			cv_signal(&vdcp->sync_pending_cv);
4089 			break;
4090 
4091 		case CB_STRATEGY:
4092 			bufp = ldep->cb_arg;
4093 			ASSERT(bufp != NULL);
4094 			bufp->b_resid =
4095 			    bufp->b_bcount - ldep->dep->payload.nbytes;
4096 			status = ldep->dep->payload.status; /* Future:ntoh */
4097 			if (status != 0) {
4098 				DMSG(vdcp, 1, "strategy status=%d\n", status);
4099 				bioerror(bufp, status);
4100 			}
4101 			status = vdc_depopulate_descriptor(vdcp, idx);
4102 			biodone(bufp);
4103 
4104 			DMSG(vdcp, 1,
4105 			    "strategy complete req=%ld bytes resp=%ld bytes\n",
4106 			    bufp->b_bcount, ldep->dep->payload.nbytes);
4107 			break;
4108 
4109 		default:
4110 			ASSERT(0);
4111 		}
4112 	}
4113 
4114 	/* let the arrival signal propogate */
4115 	mutex_exit(&vdcp->lock);
4116 
4117 	/* probe gives the count of how many entries were processed */
4118 	DTRACE_IO2(processed, int, 1, vdc_t *, vdcp);
4119 
4120 	return (0);
4121 }
4122 
4123 
4124 /*
4125  * Function:
4126  *	vdc_handle_ver_msg()
4127  *
4128  * Description:
4129  *
4130  * Arguments:
4131  *	vdc	- soft state pointer for this instance of the device driver.
4132  *	ver_msg	- LDC message sent by vDisk server
4133  *
4134  * Return Code:
4135  *	0	- Success
4136  */
4137 static int
4138 vdc_handle_ver_msg(vdc_t *vdc, vio_ver_msg_t *ver_msg)
4139 {
4140 	int status = 0;
4141 
4142 	ASSERT(vdc != NULL);
4143 	ASSERT(mutex_owned(&vdc->lock));
4144 
4145 	if (ver_msg->tag.vio_subtype_env != VIO_VER_INFO) {
4146 		return (EPROTO);
4147 	}
4148 
4149 	if (ver_msg->dev_class != VDEV_DISK_SERVER) {
4150 		return (EINVAL);
4151 	}
4152 
4153 	switch (ver_msg->tag.vio_subtype) {
4154 	case VIO_SUBTYPE_ACK:
4155 		/*
4156 		 * We check to see if the version returned is indeed supported
4157 		 * (The server may have also adjusted the minor number downwards
4158 		 * and if so 'ver_msg' will contain the actual version agreed)
4159 		 */
4160 		if (vdc_is_supported_version(ver_msg)) {
4161 			vdc->ver.major = ver_msg->ver_major;
4162 			vdc->ver.minor = ver_msg->ver_minor;
4163 			ASSERT(vdc->ver.major > 0);
4164 		} else {
4165 			status = EPROTO;
4166 		}
4167 		break;
4168 
4169 	case VIO_SUBTYPE_NACK:
4170 		/*
4171 		 * call vdc_is_supported_version() which will return the next
4172 		 * supported version (if any) in 'ver_msg'
4173 		 */
4174 		(void) vdc_is_supported_version(ver_msg);
4175 		if (ver_msg->ver_major > 0) {
4176 			size_t len = sizeof (*ver_msg);
4177 
4178 			ASSERT(vdc->ver.major > 0);
4179 
4180 			/* reset the necessary fields and resend */
4181 			ver_msg->tag.vio_subtype = VIO_SUBTYPE_INFO;
4182 			ver_msg->dev_class = VDEV_DISK;
4183 
4184 			status = vdc_send(vdc, (caddr_t)ver_msg, &len);
4185 			DMSG(vdc, 0, "[%d] Resend VER info (LDC status = %d)\n",
4186 			    vdc->instance, status);
4187 			if (len != sizeof (*ver_msg))
4188 				status = EBADMSG;
4189 		} else {
4190 			DMSG(vdc, 0, "[%d] No common version with vDisk server",
4191 			    vdc->instance);
4192 			status = ENOTSUP;
4193 		}
4194 
4195 		break;
4196 	case VIO_SUBTYPE_INFO:
4197 		/*
4198 		 * Handle the case where vds starts handshake
4199 		 * (for now only vdc is the instigator)
4200 		 */
4201 		status = ENOTSUP;
4202 		break;
4203 
4204 	default:
4205 		status = EINVAL;
4206 		break;
4207 	}
4208 
4209 	return (status);
4210 }
4211 
4212 /*
4213  * Function:
4214  *	vdc_handle_attr_msg()
4215  *
4216  * Description:
4217  *
4218  * Arguments:
4219  *	vdc	- soft state pointer for this instance of the device driver.
4220  *	attr_msg	- LDC message sent by vDisk server
4221  *
4222  * Return Code:
4223  *	0	- Success
4224  */
4225 static int
4226 vdc_handle_attr_msg(vdc_t *vdc, vd_attr_msg_t *attr_msg)
4227 {
4228 	int status = 0;
4229 
4230 	ASSERT(vdc != NULL);
4231 	ASSERT(mutex_owned(&vdc->lock));
4232 
4233 	if (attr_msg->tag.vio_subtype_env != VIO_ATTR_INFO) {
4234 		return (EPROTO);
4235 	}
4236 
4237 	switch (attr_msg->tag.vio_subtype) {
4238 	case VIO_SUBTYPE_ACK:
4239 		/*
4240 		 * We now verify the attributes sent by vds.
4241 		 */
4242 		if (attr_msg->vdisk_size == 0) {
4243 			DMSG(vdc, 0, "[%d] Invalid disk size from vds",
4244 			    vdc->instance);
4245 			status = EINVAL;
4246 			break;
4247 		}
4248 
4249 		if (attr_msg->max_xfer_sz == 0) {
4250 			DMSG(vdc, 0, "[%d] Invalid transfer size from vds",
4251 			    vdc->instance);
4252 			status = EINVAL;
4253 			break;
4254 		}
4255 
4256 		/*
4257 		 * If the disk size is already set check that it hasn't changed.
4258 		 */
4259 		if ((vdc->vdisk_size != 0) &&
4260 		    (vdc->vdisk_size != attr_msg->vdisk_size)) {
4261 			DMSG(vdc, 0, "[%d] Different disk size from vds "
4262 			    "(old=0x%lx - new=0x%lx", vdc->instance,
4263 			    vdc->vdisk_size, attr_msg->vdisk_size)
4264 			status = EINVAL;
4265 			break;
4266 		}
4267 
4268 		vdc->vdisk_size = attr_msg->vdisk_size;
4269 		vdc->vdisk_type = attr_msg->vdisk_type;
4270 		vdc->operations = attr_msg->operations;
4271 		if (vio_ver_is_supported(vdc->ver, 1, 1))
4272 			vdc->vdisk_media = attr_msg->vdisk_media;
4273 		else
4274 			vdc->vdisk_media = 0;
4275 
4276 		DMSG(vdc, 0, "[%d] max_xfer_sz: sent %lx acked %lx\n",
4277 		    vdc->instance, vdc->max_xfer_sz, attr_msg->max_xfer_sz);
4278 		DMSG(vdc, 0, "[%d] vdisk_block_size: sent %lx acked %x\n",
4279 		    vdc->instance, vdc->block_size,
4280 		    attr_msg->vdisk_block_size);
4281 
4282 		/*
4283 		 * We don't know at compile time what the vDisk server will
4284 		 * think are good values but we apply a large (arbitrary)
4285 		 * upper bound to prevent memory exhaustion in vdc if it was
4286 		 * allocating a DRing based of huge values sent by the server.
4287 		 * We probably will never exceed this except if the message
4288 		 * was garbage.
4289 		 */
4290 		if ((attr_msg->max_xfer_sz * attr_msg->vdisk_block_size) <=
4291 		    (PAGESIZE * DEV_BSIZE)) {
4292 			vdc->max_xfer_sz = attr_msg->max_xfer_sz;
4293 			vdc->block_size = attr_msg->vdisk_block_size;
4294 		} else {
4295 			DMSG(vdc, 0, "[%d] vds block transfer size too big;"
4296 			    " using max supported by vdc", vdc->instance);
4297 		}
4298 
4299 		if ((attr_msg->xfer_mode != VIO_DRING_MODE) ||
4300 		    (attr_msg->vdisk_size > INT64_MAX) ||
4301 		    (attr_msg->operations == 0) ||
4302 		    (attr_msg->vdisk_type > VD_DISK_TYPE_DISK)) {
4303 			DMSG(vdc, 0, "[%d] Invalid attributes from vds",
4304 			    vdc->instance);
4305 			status = EINVAL;
4306 			break;
4307 		}
4308 
4309 		/*
4310 		 * Now that we have received all attributes we can create a
4311 		 * fake geometry for the disk.
4312 		 */
4313 		vdc_create_fake_geometry(vdc);
4314 		break;
4315 
4316 	case VIO_SUBTYPE_NACK:
4317 		/*
4318 		 * vds could not handle the attributes we sent so we
4319 		 * stop negotiating.
4320 		 */
4321 		status = EPROTO;
4322 		break;
4323 
4324 	case VIO_SUBTYPE_INFO:
4325 		/*
4326 		 * Handle the case where vds starts the handshake
4327 		 * (for now; vdc is the only supported instigatior)
4328 		 */
4329 		status = ENOTSUP;
4330 		break;
4331 
4332 	default:
4333 		status = ENOTSUP;
4334 		break;
4335 	}
4336 
4337 	return (status);
4338 }
4339 
4340 /*
4341  * Function:
4342  *	vdc_handle_dring_reg_msg()
4343  *
4344  * Description:
4345  *
4346  * Arguments:
4347  *	vdc		- soft state pointer for this instance of the driver.
4348  *	dring_msg	- LDC message sent by vDisk server
4349  *
4350  * Return Code:
4351  *	0	- Success
4352  */
4353 static int
4354 vdc_handle_dring_reg_msg(vdc_t *vdc, vio_dring_reg_msg_t *dring_msg)
4355 {
4356 	int		status = 0;
4357 
4358 	ASSERT(vdc != NULL);
4359 	ASSERT(mutex_owned(&vdc->lock));
4360 
4361 	if (dring_msg->tag.vio_subtype_env != VIO_DRING_REG) {
4362 		return (EPROTO);
4363 	}
4364 
4365 	switch (dring_msg->tag.vio_subtype) {
4366 	case VIO_SUBTYPE_ACK:
4367 		/* save the received dring_ident */
4368 		vdc->dring_ident = dring_msg->dring_ident;
4369 		DMSG(vdc, 0, "[%d] Received dring ident=0x%lx\n",
4370 		    vdc->instance, vdc->dring_ident);
4371 		break;
4372 
4373 	case VIO_SUBTYPE_NACK:
4374 		/*
4375 		 * vds could not handle the DRing info we sent so we
4376 		 * stop negotiating.
4377 		 */
4378 		DMSG(vdc, 0, "[%d] server could not register DRing\n",
4379 		    vdc->instance);
4380 		status = EPROTO;
4381 		break;
4382 
4383 	case VIO_SUBTYPE_INFO:
4384 		/*
4385 		 * Handle the case where vds starts handshake
4386 		 * (for now only vdc is the instigatior)
4387 		 */
4388 		status = ENOTSUP;
4389 		break;
4390 	default:
4391 		status = ENOTSUP;
4392 	}
4393 
4394 	return (status);
4395 }
4396 
4397 /*
4398  * Function:
4399  *	vdc_verify_seq_num()
4400  *
4401  * Description:
4402  *	This functions verifies that the sequence number sent back by the vDisk
4403  *	server with the latest message is what is expected (i.e. it is greater
4404  *	than the last seq num sent by the vDisk server and less than or equal
4405  *	to the last seq num generated by vdc).
4406  *
4407  *	It then checks the request ID to see if any requests need processing
4408  *	in the DRing.
4409  *
4410  * Arguments:
4411  *	vdc		- soft state pointer for this instance of the driver.
4412  *	dring_msg	- pointer to the LDC message sent by vds
4413  *
4414  * Return Code:
4415  *	VDC_SEQ_NUM_TODO	- Message needs to be processed
4416  *	VDC_SEQ_NUM_SKIP	- Message has already been processed
4417  *	VDC_SEQ_NUM_INVALID	- The seq numbers are so out of sync,
4418  *				  vdc cannot deal with them
4419  */
4420 static int
4421 vdc_verify_seq_num(vdc_t *vdc, vio_dring_msg_t *dring_msg)
4422 {
4423 	ASSERT(vdc != NULL);
4424 	ASSERT(dring_msg != NULL);
4425 	ASSERT(mutex_owned(&vdc->lock));
4426 
4427 	/*
4428 	 * Check to see if the messages were responded to in the correct
4429 	 * order by vds.
4430 	 */
4431 	if ((dring_msg->seq_num <= vdc->seq_num_reply) ||
4432 	    (dring_msg->seq_num > vdc->seq_num)) {
4433 		DMSG(vdc, 0, "?[%d] Bogus sequence_number %lu: "
4434 		    "%lu > expected <= %lu (last proc req %lu sent %lu)\n",
4435 		    vdc->instance, dring_msg->seq_num,
4436 		    vdc->seq_num_reply, vdc->seq_num,
4437 		    vdc->req_id_proc, vdc->req_id);
4438 		return (VDC_SEQ_NUM_INVALID);
4439 	}
4440 	vdc->seq_num_reply = dring_msg->seq_num;
4441 
4442 	if (vdc->req_id_proc < vdc->req_id)
4443 		return (VDC_SEQ_NUM_TODO);
4444 	else
4445 		return (VDC_SEQ_NUM_SKIP);
4446 }
4447 
4448 
4449 /*
4450  * Function:
4451  *	vdc_is_supported_version()
4452  *
4453  * Description:
4454  *	This routine checks if the major/minor version numbers specified in
4455  *	'ver_msg' are supported. If not it finds the next version that is
4456  *	in the supported version list 'vdc_version[]' and sets the fields in
4457  *	'ver_msg' to those values
4458  *
4459  * Arguments:
4460  *	ver_msg	- LDC message sent by vDisk server
4461  *
4462  * Return Code:
4463  *	B_TRUE	- Success
4464  *	B_FALSE	- Version not supported
4465  */
4466 static boolean_t
4467 vdc_is_supported_version(vio_ver_msg_t *ver_msg)
4468 {
4469 	int vdc_num_versions = sizeof (vdc_version) / sizeof (vdc_version[0]);
4470 
4471 	for (int i = 0; i < vdc_num_versions; i++) {
4472 		ASSERT(vdc_version[i].major > 0);
4473 		ASSERT((i == 0) ||
4474 		    (vdc_version[i].major < vdc_version[i-1].major));
4475 
4476 		/*
4477 		 * If the major versions match, adjust the minor version, if
4478 		 * necessary, down to the highest value supported by this
4479 		 * client. The server should support all minor versions lower
4480 		 * than the value it sent
4481 		 */
4482 		if (ver_msg->ver_major == vdc_version[i].major) {
4483 			if (ver_msg->ver_minor > vdc_version[i].minor) {
4484 				DMSGX(0,
4485 				    "Adjusting minor version from %u to %u",
4486 				    ver_msg->ver_minor, vdc_version[i].minor);
4487 				ver_msg->ver_minor = vdc_version[i].minor;
4488 			}
4489 			return (B_TRUE);
4490 		}
4491 
4492 		/*
4493 		 * If the message contains a higher major version number, set
4494 		 * the message's major/minor versions to the current values
4495 		 * and return false, so this message will get resent with
4496 		 * these values, and the server will potentially try again
4497 		 * with the same or a lower version
4498 		 */
4499 		if (ver_msg->ver_major > vdc_version[i].major) {
4500 			ver_msg->ver_major = vdc_version[i].major;
4501 			ver_msg->ver_minor = vdc_version[i].minor;
4502 			DMSGX(0, "Suggesting major/minor (0x%x/0x%x)\n",
4503 			    ver_msg->ver_major, ver_msg->ver_minor);
4504 
4505 			return (B_FALSE);
4506 		}
4507 
4508 		/*
4509 		 * Otherwise, the message's major version is less than the
4510 		 * current major version, so continue the loop to the next
4511 		 * (lower) supported version
4512 		 */
4513 	}
4514 
4515 	/*
4516 	 * No common version was found; "ground" the version pair in the
4517 	 * message to terminate negotiation
4518 	 */
4519 	ver_msg->ver_major = 0;
4520 	ver_msg->ver_minor = 0;
4521 
4522 	return (B_FALSE);
4523 }
4524 /* -------------------------------------------------------------------------- */
4525 
4526 /*
4527  * DKIO(7) support
4528  */
4529 
4530 typedef struct vdc_dk_arg {
4531 	struct dk_callback	dkc;
4532 	int			mode;
4533 	dev_t			dev;
4534 	vdc_t			*vdc;
4535 } vdc_dk_arg_t;
4536 
4537 /*
4538  * Function:
4539  * 	vdc_dkio_flush_cb()
4540  *
4541  * Description:
4542  *	This routine is a callback for DKIOCFLUSHWRITECACHE which can be called
4543  *	by kernel code.
4544  *
4545  * Arguments:
4546  *	arg	- a pointer to a vdc_dk_arg_t structure.
4547  */
4548 void
4549 vdc_dkio_flush_cb(void *arg)
4550 {
4551 	struct vdc_dk_arg	*dk_arg = (struct vdc_dk_arg *)arg;
4552 	struct dk_callback	*dkc = NULL;
4553 	vdc_t			*vdc = NULL;
4554 	int			rv;
4555 
4556 	if (dk_arg == NULL) {
4557 		cmn_err(CE_NOTE, "?[Unk] DKIOCFLUSHWRITECACHE arg is NULL\n");
4558 		return;
4559 	}
4560 	dkc = &dk_arg->dkc;
4561 	vdc = dk_arg->vdc;
4562 	ASSERT(vdc != NULL);
4563 
4564 	rv = vdc_do_sync_op(vdc, VD_OP_FLUSH, NULL, 0,
4565 	    VDCPART(dk_arg->dev), 0, CB_SYNC, 0, VIO_both_dir);
4566 	if (rv != 0) {
4567 		DMSG(vdc, 0, "[%d] DKIOCFLUSHWRITECACHE failed %d : model %x\n",
4568 		    vdc->instance, rv,
4569 		    ddi_model_convert_from(dk_arg->mode & FMODELS));
4570 	}
4571 
4572 	/*
4573 	 * Trigger the call back to notify the caller the the ioctl call has
4574 	 * been completed.
4575 	 */
4576 	if ((dk_arg->mode & FKIOCTL) &&
4577 	    (dkc != NULL) &&
4578 	    (dkc->dkc_callback != NULL)) {
4579 		ASSERT(dkc->dkc_cookie != NULL);
4580 		(*dkc->dkc_callback)(dkc->dkc_cookie, rv);
4581 	}
4582 
4583 	/* Indicate that one less DKIO write flush is outstanding */
4584 	mutex_enter(&vdc->lock);
4585 	vdc->dkio_flush_pending--;
4586 	ASSERT(vdc->dkio_flush_pending >= 0);
4587 	mutex_exit(&vdc->lock);
4588 
4589 	/* free the mem that was allocated when the callback was dispatched */
4590 	kmem_free(arg, sizeof (vdc_dk_arg_t));
4591 }
4592 
4593 /*
4594  * Function:
4595  * 	vdc_dkio_get_partition()
4596  *
4597  * Description:
4598  *	This function implements the DKIOCGAPART ioctl.
4599  *
4600  * Arguments:
4601  *	vdc	- soft state pointer
4602  *	arg	- a pointer to a dk_map[NDKMAP] or dk_map32[NDKMAP] structure
4603  *	flag	- ioctl flags
4604  */
4605 static int
4606 vdc_dkio_get_partition(vdc_t *vdc, caddr_t arg, int flag)
4607 {
4608 	struct dk_geom *geom;
4609 	struct vtoc *vtoc;
4610 	union {
4611 		struct dk_map map[NDKMAP];
4612 		struct dk_map32 map32[NDKMAP];
4613 	} data;
4614 	int i, rv, size;
4615 
4616 	mutex_enter(&vdc->lock);
4617 
4618 	if ((rv = vdc_validate_geometry(vdc)) != 0) {
4619 		mutex_exit(&vdc->lock);
4620 		return (rv);
4621 	}
4622 
4623 	vtoc = vdc->vtoc;
4624 	geom = vdc->geom;
4625 
4626 	if (ddi_model_convert_from(flag & FMODELS) == DDI_MODEL_ILP32) {
4627 
4628 		for (i = 0; i < vtoc->v_nparts; i++) {
4629 			data.map32[i].dkl_cylno = vtoc->v_part[i].p_start /
4630 			    (geom->dkg_nhead * geom->dkg_nsect);
4631 			data.map32[i].dkl_nblk = vtoc->v_part[i].p_size;
4632 		}
4633 		size = NDKMAP * sizeof (struct dk_map32);
4634 
4635 	} else {
4636 
4637 		for (i = 0; i < vtoc->v_nparts; i++) {
4638 			data.map[i].dkl_cylno = vtoc->v_part[i].p_start /
4639 			    (geom->dkg_nhead * geom->dkg_nsect);
4640 			data.map[i].dkl_nblk = vtoc->v_part[i].p_size;
4641 		}
4642 		size = NDKMAP * sizeof (struct dk_map);
4643 
4644 	}
4645 
4646 	mutex_exit(&vdc->lock);
4647 
4648 	if (ddi_copyout(&data, arg, size, flag) != 0)
4649 		return (EFAULT);
4650 
4651 	return (0);
4652 }
4653 
4654 /*
4655  * Function:
4656  * 	vdc_dioctl_rwcmd()
4657  *
4658  * Description:
4659  *	This function implements the DIOCTL_RWCMD ioctl. This ioctl is used
4660  *	for DKC_DIRECT disks to read or write at an absolute disk offset.
4661  *
4662  * Arguments:
4663  *	dev	- device
4664  *	arg	- a pointer to a dadkio_rwcmd or dadkio_rwcmd32 structure
4665  *	flag	- ioctl flags
4666  */
4667 static int
4668 vdc_dioctl_rwcmd(dev_t dev, caddr_t arg, int flag)
4669 {
4670 	struct dadkio_rwcmd32 rwcmd32;
4671 	struct dadkio_rwcmd rwcmd;
4672 	struct iovec aiov;
4673 	struct uio auio;
4674 	int rw, status;
4675 	struct buf *buf;
4676 
4677 	if (ddi_model_convert_from(flag & FMODELS) == DDI_MODEL_ILP32) {
4678 		if (ddi_copyin((caddr_t)arg, (caddr_t)&rwcmd32,
4679 		    sizeof (struct dadkio_rwcmd32), flag)) {
4680 			return (EFAULT);
4681 		}
4682 		rwcmd.cmd = rwcmd32.cmd;
4683 		rwcmd.flags = rwcmd32.flags;
4684 		rwcmd.blkaddr = (daddr_t)rwcmd32.blkaddr;
4685 		rwcmd.buflen = rwcmd32.buflen;
4686 		rwcmd.bufaddr = (caddr_t)(uintptr_t)rwcmd32.bufaddr;
4687 	} else {
4688 		if (ddi_copyin((caddr_t)arg, (caddr_t)&rwcmd,
4689 		    sizeof (struct dadkio_rwcmd), flag)) {
4690 			return (EFAULT);
4691 		}
4692 	}
4693 
4694 	switch (rwcmd.cmd) {
4695 	case DADKIO_RWCMD_READ:
4696 		rw = B_READ;
4697 		break;
4698 	case DADKIO_RWCMD_WRITE:
4699 		rw = B_WRITE;
4700 		break;
4701 	default:
4702 		return (EINVAL);
4703 	}
4704 
4705 	bzero((caddr_t)&aiov, sizeof (struct iovec));
4706 	aiov.iov_base   = rwcmd.bufaddr;
4707 	aiov.iov_len    = rwcmd.buflen;
4708 
4709 	bzero((caddr_t)&auio, sizeof (struct uio));
4710 	auio.uio_iov    = &aiov;
4711 	auio.uio_iovcnt = 1;
4712 	auio.uio_loffset = rwcmd.blkaddr * DEV_BSIZE;
4713 	auio.uio_resid  = rwcmd.buflen;
4714 	auio.uio_segflg = flag & FKIOCTL ? UIO_SYSSPACE : UIO_USERSPACE;
4715 
4716 	buf = kmem_alloc(sizeof (buf_t), KM_SLEEP);
4717 	bioinit(buf);
4718 	/*
4719 	 * We use the private field of buf to specify that this is an
4720 	 * I/O using an absolute offset.
4721 	 */
4722 	buf->b_private = (void *)VD_SLICE_NONE;
4723 
4724 	status = physio(vdc_strategy, buf, dev, rw, vdc_min, &auio);
4725 
4726 	biofini(buf);
4727 	kmem_free(buf, sizeof (buf_t));
4728 
4729 	return (status);
4730 }
4731 
4732 /*
4733  * This structure is used in the DKIO(7I) array below.
4734  */
4735 typedef struct vdc_dk_ioctl {
4736 	uint8_t		op;		/* VD_OP_XXX value */
4737 	int		cmd;		/* Solaris ioctl operation number */
4738 	size_t		nbytes;		/* size of structure to be copied */
4739 
4740 	/* function to convert between vDisk and Solaris structure formats */
4741 	int	(*convert)(vdc_t *vdc, void *vd_buf, void *ioctl_arg,
4742 	    int mode, int dir);
4743 } vdc_dk_ioctl_t;
4744 
4745 /*
4746  * Subset of DKIO(7I) operations currently supported
4747  */
4748 static vdc_dk_ioctl_t	dk_ioctl[] = {
4749 	{VD_OP_FLUSH,		DKIOCFLUSHWRITECACHE,	0,
4750 		vdc_null_copy_func},
4751 	{VD_OP_GET_WCE,		DKIOCGETWCE,		sizeof (int),
4752 		vdc_get_wce_convert},
4753 	{VD_OP_SET_WCE,		DKIOCSETWCE,		sizeof (int),
4754 		vdc_set_wce_convert},
4755 	{VD_OP_GET_VTOC,	DKIOCGVTOC,		sizeof (vd_vtoc_t),
4756 		vdc_get_vtoc_convert},
4757 	{VD_OP_SET_VTOC,	DKIOCSVTOC,		sizeof (vd_vtoc_t),
4758 		vdc_set_vtoc_convert},
4759 	{VD_OP_GET_DISKGEOM,	DKIOCGGEOM,		sizeof (vd_geom_t),
4760 		vdc_get_geom_convert},
4761 	{VD_OP_GET_DISKGEOM,	DKIOCG_PHYGEOM,		sizeof (vd_geom_t),
4762 		vdc_get_geom_convert},
4763 	{VD_OP_GET_DISKGEOM, 	DKIOCG_VIRTGEOM,	sizeof (vd_geom_t),
4764 		vdc_get_geom_convert},
4765 	{VD_OP_SET_DISKGEOM,	DKIOCSGEOM,		sizeof (vd_geom_t),
4766 		vdc_set_geom_convert},
4767 	{VD_OP_GET_EFI,		DKIOCGETEFI,		0,
4768 		vdc_get_efi_convert},
4769 	{VD_OP_SET_EFI,		DKIOCSETEFI,		0,
4770 		vdc_set_efi_convert},
4771 
4772 	/* DIOCTL_RWCMD is converted to a read or a write */
4773 	{0, DIOCTL_RWCMD,  sizeof (struct dadkio_rwcmd), NULL},
4774 
4775 	/*
4776 	 * These particular ioctls are not sent to the server - vdc fakes up
4777 	 * the necessary info.
4778 	 */
4779 	{0, DKIOCINFO, sizeof (struct dk_cinfo), vdc_null_copy_func},
4780 	{0, DKIOCGMEDIAINFO, sizeof (struct dk_minfo), vdc_null_copy_func},
4781 	{0, USCSICMD,	sizeof (struct uscsi_cmd), vdc_null_copy_func},
4782 	{0, DKIOCGAPART, 0, vdc_null_copy_func },
4783 	{0, DKIOCREMOVABLE, 0, vdc_null_copy_func},
4784 	{0, CDROMREADOFFSET, 0, vdc_null_copy_func}
4785 };
4786 
4787 /*
4788  * Function:
4789  *	vd_process_ioctl()
4790  *
4791  * Description:
4792  *	This routine processes disk specific ioctl calls
4793  *
4794  * Arguments:
4795  *	dev	- the device number
4796  *	cmd	- the operation [dkio(7I)] to be processed
4797  *	arg	- pointer to user provided structure
4798  *		  (contains data to be set or reference parameter for get)
4799  *	mode	- bit flag, indicating open settings, 32/64 bit type, etc
4800  *
4801  * Return Code:
4802  *	0
4803  *	EFAULT
4804  *	ENXIO
4805  *	EIO
4806  *	ENOTSUP
4807  */
4808 static int
4809 vd_process_ioctl(dev_t dev, int cmd, caddr_t arg, int mode)
4810 {
4811 	int		instance = VDCUNIT(dev);
4812 	vdc_t		*vdc = NULL;
4813 	int		rv = -1;
4814 	int		idx = 0;		/* index into dk_ioctl[] */
4815 	size_t		len = 0;		/* #bytes to send to vds */
4816 	size_t		alloc_len = 0;		/* #bytes to allocate mem for */
4817 	caddr_t		mem_p = NULL;
4818 	size_t		nioctls = (sizeof (dk_ioctl)) / (sizeof (dk_ioctl[0]));
4819 	vdc_dk_ioctl_t	*iop;
4820 
4821 	vdc = ddi_get_soft_state(vdc_state, instance);
4822 	if (vdc == NULL) {
4823 		cmn_err(CE_NOTE, "![%d] Could not get soft state structure",
4824 		    instance);
4825 		return (ENXIO);
4826 	}
4827 
4828 	DMSG(vdc, 0, "[%d] Processing ioctl(%x) for dev %lx : model %x\n",
4829 	    instance, cmd, dev, ddi_model_convert_from(mode & FMODELS));
4830 
4831 	/*
4832 	 * Validate the ioctl operation to be performed.
4833 	 *
4834 	 * If we have looped through the array without finding a match then we
4835 	 * don't support this ioctl.
4836 	 */
4837 	for (idx = 0; idx < nioctls; idx++) {
4838 		if (cmd == dk_ioctl[idx].cmd)
4839 			break;
4840 	}
4841 
4842 	if (idx >= nioctls) {
4843 		DMSG(vdc, 0, "[%d] Unsupported ioctl (0x%x)\n",
4844 		    vdc->instance, cmd);
4845 		return (ENOTSUP);
4846 	}
4847 
4848 	iop = &(dk_ioctl[idx]);
4849 
4850 	if (cmd == DKIOCGETEFI || cmd == DKIOCSETEFI) {
4851 		/* size is not fixed for EFI ioctls, it depends on ioctl arg */
4852 		dk_efi_t	dk_efi;
4853 
4854 		rv = ddi_copyin(arg, &dk_efi, sizeof (dk_efi_t), mode);
4855 		if (rv != 0)
4856 			return (EFAULT);
4857 
4858 		len = sizeof (vd_efi_t) - 1 + dk_efi.dki_length;
4859 	} else {
4860 		len = iop->nbytes;
4861 	}
4862 
4863 	/*
4864 	 * Deal with the ioctls which the server does not provide. vdc can
4865 	 * fake these up and return immediately
4866 	 */
4867 	switch (cmd) {
4868 	case CDROMREADOFFSET:
4869 	case DKIOCREMOVABLE:
4870 	case USCSICMD:
4871 		return (ENOTTY);
4872 
4873 	case DIOCTL_RWCMD:
4874 		{
4875 			if (vdc->cinfo == NULL)
4876 				return (ENXIO);
4877 
4878 			if (vdc->cinfo->dki_ctype != DKC_DIRECT)
4879 				return (ENOTTY);
4880 
4881 			return (vdc_dioctl_rwcmd(dev, arg, mode));
4882 		}
4883 
4884 	case DKIOCGAPART:
4885 		{
4886 			return (vdc_dkio_get_partition(vdc, arg, mode));
4887 		}
4888 
4889 	case DKIOCINFO:
4890 		{
4891 			struct dk_cinfo	cinfo;
4892 			if (vdc->cinfo == NULL)
4893 				return (ENXIO);
4894 
4895 			bcopy(vdc->cinfo, &cinfo, sizeof (struct dk_cinfo));
4896 			cinfo.dki_partition = VDCPART(dev);
4897 
4898 			rv = ddi_copyout(&cinfo, (void *)arg,
4899 			    sizeof (struct dk_cinfo), mode);
4900 			if (rv != 0)
4901 				return (EFAULT);
4902 
4903 			return (0);
4904 		}
4905 
4906 	case DKIOCGMEDIAINFO:
4907 		{
4908 			if (vdc->minfo == NULL)
4909 				return (ENXIO);
4910 
4911 			rv = ddi_copyout(vdc->minfo, (void *)arg,
4912 			    sizeof (struct dk_minfo), mode);
4913 			if (rv != 0)
4914 				return (EFAULT);
4915 
4916 			return (0);
4917 		}
4918 
4919 	case DKIOCFLUSHWRITECACHE:
4920 		{
4921 			struct dk_callback *dkc =
4922 			    (struct dk_callback *)(uintptr_t)arg;
4923 			vdc_dk_arg_t	*dkarg = NULL;
4924 
4925 			DMSG(vdc, 1, "[%d] Flush W$: mode %x\n",
4926 			    instance, mode);
4927 
4928 			/*
4929 			 * If arg is NULL, then there is no callback function
4930 			 * registered and the call operates synchronously; we
4931 			 * break and continue with the rest of the function and
4932 			 * wait for vds to return (i.e. after the request to
4933 			 * vds returns successfully, all writes completed prior
4934 			 * to the ioctl will have been flushed from the disk
4935 			 * write cache to persistent media.
4936 			 *
4937 			 * If a callback function is registered, we dispatch
4938 			 * the request on a task queue and return immediately.
4939 			 * The callback will deal with informing the calling
4940 			 * thread that the flush request is completed.
4941 			 */
4942 			if (dkc == NULL)
4943 				break;
4944 
4945 			/*
4946 			 * the asynchronous callback is only supported if
4947 			 * invoked from within the kernel
4948 			 */
4949 			if ((mode & FKIOCTL) == 0)
4950 				return (ENOTSUP);
4951 
4952 			dkarg = kmem_zalloc(sizeof (vdc_dk_arg_t), KM_SLEEP);
4953 
4954 			dkarg->mode = mode;
4955 			dkarg->dev = dev;
4956 			bcopy(dkc, &dkarg->dkc, sizeof (*dkc));
4957 
4958 			mutex_enter(&vdc->lock);
4959 			vdc->dkio_flush_pending++;
4960 			dkarg->vdc = vdc;
4961 			mutex_exit(&vdc->lock);
4962 
4963 			/* put the request on a task queue */
4964 			rv = taskq_dispatch(system_taskq, vdc_dkio_flush_cb,
4965 			    (void *)dkarg, DDI_SLEEP);
4966 			if (rv == NULL) {
4967 				/* clean up if dispatch fails */
4968 				mutex_enter(&vdc->lock);
4969 				vdc->dkio_flush_pending--;
4970 				mutex_exit(&vdc->lock);
4971 				kmem_free(dkarg, sizeof (vdc_dk_arg_t));
4972 			}
4973 
4974 			return (rv == NULL ? ENOMEM : 0);
4975 		}
4976 	}
4977 
4978 	/* catch programming error in vdc - should be a VD_OP_XXX ioctl */
4979 	ASSERT(iop->op != 0);
4980 
4981 	/* check if the vDisk server handles the operation for this vDisk */
4982 	if (VD_OP_SUPPORTED(vdc->operations, iop->op) == B_FALSE) {
4983 		DMSG(vdc, 0, "[%d] Unsupported VD_OP operation (0x%x)\n",
4984 		    vdc->instance, iop->op);
4985 		return (ENOTSUP);
4986 	}
4987 
4988 	/* LDC requires that the memory being mapped is 8-byte aligned */
4989 	alloc_len = P2ROUNDUP(len, sizeof (uint64_t));
4990 	DMSG(vdc, 1, "[%d] struct size %ld alloc %ld\n",
4991 	    instance, len, alloc_len);
4992 
4993 	if (alloc_len > 0)
4994 		mem_p = kmem_zalloc(alloc_len, KM_SLEEP);
4995 
4996 	/*
4997 	 * Call the conversion function for this ioctl which, if necessary,
4998 	 * converts from the Solaris format to the format ARC'ed
4999 	 * as part of the vDisk protocol (FWARC 2006/195)
5000 	 */
5001 	ASSERT(iop->convert != NULL);
5002 	rv = (iop->convert)(vdc, arg, mem_p, mode, VD_COPYIN);
5003 	if (rv != 0) {
5004 		DMSG(vdc, 0, "[%d] convert func returned %d for ioctl 0x%x\n",
5005 		    instance, rv, cmd);
5006 		if (mem_p != NULL)
5007 			kmem_free(mem_p, alloc_len);
5008 		return (rv);
5009 	}
5010 
5011 	/*
5012 	 * send request to vds to service the ioctl.
5013 	 */
5014 	rv = vdc_do_sync_op(vdc, iop->op, mem_p, alloc_len,
5015 	    VDCPART(dev), 0, CB_SYNC, (void *)(uint64_t)mode,
5016 	    VIO_both_dir);
5017 
5018 	if (cmd == DKIOCSVTOC || cmd == DKIOCSETEFI) {
5019 		/*
5020 		 * The disk label may have changed. Revalidate the disk
5021 		 * geometry. This will also update the device nodes and
5022 		 * properties.
5023 		 */
5024 		vdc_validate(vdc);
5025 	}
5026 
5027 	if (rv != 0) {
5028 		/*
5029 		 * This is not necessarily an error. The ioctl could
5030 		 * be returning a value such as ENOTTY to indicate
5031 		 * that the ioctl is not applicable.
5032 		 */
5033 		DMSG(vdc, 0, "[%d] vds returned %d for ioctl 0x%x\n",
5034 		    instance, rv, cmd);
5035 		if (mem_p != NULL)
5036 			kmem_free(mem_p, alloc_len);
5037 
5038 		return (rv);
5039 	}
5040 
5041 	/*
5042 	 * Call the conversion function (if it exists) for this ioctl
5043 	 * which converts from the format ARC'ed as part of the vDisk
5044 	 * protocol (FWARC 2006/195) back to a format understood by
5045 	 * the rest of Solaris.
5046 	 */
5047 	rv = (iop->convert)(vdc, mem_p, arg, mode, VD_COPYOUT);
5048 	if (rv != 0) {
5049 		DMSG(vdc, 0, "[%d] convert func returned %d for ioctl 0x%x\n",
5050 		    instance, rv, cmd);
5051 		if (mem_p != NULL)
5052 			kmem_free(mem_p, alloc_len);
5053 		return (rv);
5054 	}
5055 
5056 	if (mem_p != NULL)
5057 		kmem_free(mem_p, alloc_len);
5058 
5059 	return (rv);
5060 }
5061 
5062 /*
5063  * Function:
5064  *
5065  * Description:
5066  *	This is an empty conversion function used by ioctl calls which
5067  *	do not need to convert the data being passed in/out to userland
5068  */
5069 static int
5070 vdc_null_copy_func(vdc_t *vdc, void *from, void *to, int mode, int dir)
5071 {
5072 	_NOTE(ARGUNUSED(vdc))
5073 	_NOTE(ARGUNUSED(from))
5074 	_NOTE(ARGUNUSED(to))
5075 	_NOTE(ARGUNUSED(mode))
5076 	_NOTE(ARGUNUSED(dir))
5077 
5078 	return (0);
5079 }
5080 
5081 static int
5082 vdc_get_wce_convert(vdc_t *vdc, void *from, void *to,
5083     int mode, int dir)
5084 {
5085 	_NOTE(ARGUNUSED(vdc))
5086 
5087 	if (dir == VD_COPYIN)
5088 		return (0);		/* nothing to do */
5089 
5090 	if (ddi_copyout(from, to, sizeof (int), mode) != 0)
5091 		return (EFAULT);
5092 
5093 	return (0);
5094 }
5095 
5096 static int
5097 vdc_set_wce_convert(vdc_t *vdc, void *from, void *to,
5098     int mode, int dir)
5099 {
5100 	_NOTE(ARGUNUSED(vdc))
5101 
5102 	if (dir == VD_COPYOUT)
5103 		return (0);		/* nothing to do */
5104 
5105 	if (ddi_copyin(from, to, sizeof (int), mode) != 0)
5106 		return (EFAULT);
5107 
5108 	return (0);
5109 }
5110 
5111 /*
5112  * Function:
5113  *	vdc_get_vtoc_convert()
5114  *
5115  * Description:
5116  *	This routine performs the necessary convertions from the DKIOCGVTOC
5117  *	Solaris structure to the format defined in FWARC 2006/195.
5118  *
5119  *	In the struct vtoc definition, the timestamp field is marked as not
5120  *	supported so it is not part of vDisk protocol (FWARC 2006/195).
5121  *	However SVM uses that field to check it can write into the VTOC,
5122  *	so we fake up the info of that field.
5123  *
5124  * Arguments:
5125  *	vdc	- the vDisk client
5126  *	from	- the buffer containing the data to be copied from
5127  *	to	- the buffer to be copied to
5128  *	mode	- flags passed to ioctl() call
5129  *	dir	- the "direction" of the copy - VD_COPYIN or VD_COPYOUT
5130  *
5131  * Return Code:
5132  *	0	- Success
5133  *	ENXIO	- incorrect buffer passed in.
5134  *	EFAULT	- ddi_copyout routine encountered an error.
5135  */
5136 static int
5137 vdc_get_vtoc_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
5138 {
5139 	int		i;
5140 	void		*tmp_mem = NULL;
5141 	void		*tmp_memp;
5142 	struct vtoc	vt;
5143 	struct vtoc32	vt32;
5144 	int		copy_len = 0;
5145 	int		rv = 0;
5146 
5147 	if (dir != VD_COPYOUT)
5148 		return (0);	/* nothing to do */
5149 
5150 	if ((from == NULL) || (to == NULL))
5151 		return (ENXIO);
5152 
5153 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32)
5154 		copy_len = sizeof (struct vtoc32);
5155 	else
5156 		copy_len = sizeof (struct vtoc);
5157 
5158 	tmp_mem = kmem_alloc(copy_len, KM_SLEEP);
5159 
5160 	VD_VTOC2VTOC((vd_vtoc_t *)from, &vt);
5161 
5162 	/* fake the VTOC timestamp field */
5163 	for (i = 0; i < V_NUMPAR; i++) {
5164 		vt.timestamp[i] = vdc->vtoc->timestamp[i];
5165 	}
5166 
5167 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5168 		/* LINTED E_ASSIGN_NARROW_CONV */
5169 		vtoctovtoc32(vt, vt32);
5170 		tmp_memp = &vt32;
5171 	} else {
5172 		tmp_memp = &vt;
5173 	}
5174 	rv = ddi_copyout(tmp_memp, to, copy_len, mode);
5175 	if (rv != 0)
5176 		rv = EFAULT;
5177 
5178 	kmem_free(tmp_mem, copy_len);
5179 	return (rv);
5180 }
5181 
5182 /*
5183  * Function:
5184  *	vdc_set_vtoc_convert()
5185  *
5186  * Description:
5187  *	This routine performs the necessary convertions from the DKIOCSVTOC
5188  *	Solaris structure to the format defined in FWARC 2006/195.
5189  *
5190  * Arguments:
5191  *	vdc	- the vDisk client
5192  *	from	- Buffer with data
5193  *	to	- Buffer where data is to be copied to
5194  *	mode	- flags passed to ioctl
5195  *	dir	- direction of copy (in or out)
5196  *
5197  * Return Code:
5198  *	0	- Success
5199  *	ENXIO	- Invalid buffer passed in
5200  *	EFAULT	- ddi_copyin of data failed
5201  */
5202 static int
5203 vdc_set_vtoc_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
5204 {
5205 	_NOTE(ARGUNUSED(vdc))
5206 
5207 	void		*tmp_mem = NULL;
5208 	struct vtoc	vt;
5209 	struct vtoc	*vtp = &vt;
5210 	vd_vtoc_t	vtvd;
5211 	int		copy_len = 0;
5212 	int		rv = 0;
5213 
5214 	if (dir != VD_COPYIN)
5215 		return (0);	/* nothing to do */
5216 
5217 	if ((from == NULL) || (to == NULL))
5218 		return (ENXIO);
5219 
5220 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32)
5221 		copy_len = sizeof (struct vtoc32);
5222 	else
5223 		copy_len = sizeof (struct vtoc);
5224 
5225 	tmp_mem = kmem_alloc(copy_len, KM_SLEEP);
5226 
5227 	rv = ddi_copyin(from, tmp_mem, copy_len, mode);
5228 	if (rv != 0) {
5229 		kmem_free(tmp_mem, copy_len);
5230 		return (EFAULT);
5231 	}
5232 
5233 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5234 		vtoc32tovtoc((*(struct vtoc32 *)tmp_mem), vt);
5235 	} else {
5236 		vtp = tmp_mem;
5237 	}
5238 
5239 	VTOC2VD_VTOC(vtp, &vtvd);
5240 	bcopy(&vtvd, to, sizeof (vd_vtoc_t));
5241 	kmem_free(tmp_mem, copy_len);
5242 
5243 	return (0);
5244 }
5245 
5246 /*
5247  * Function:
5248  *	vdc_get_geom_convert()
5249  *
5250  * Description:
5251  *	This routine performs the necessary convertions from the DKIOCGGEOM,
5252  *	DKIOCG_PHYSGEOM and DKIOG_VIRTGEOM Solaris structures to the format
5253  *	defined in FWARC 2006/195
5254  *
5255  * Arguments:
5256  *	vdc	- the vDisk client
5257  *	from	- Buffer with data
5258  *	to	- Buffer where data is to be copied to
5259  *	mode	- flags passed to ioctl
5260  *	dir	- direction of copy (in or out)
5261  *
5262  * Return Code:
5263  *	0	- Success
5264  *	ENXIO	- Invalid buffer passed in
5265  *	EFAULT	- ddi_copyout of data failed
5266  */
5267 static int
5268 vdc_get_geom_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
5269 {
5270 	_NOTE(ARGUNUSED(vdc))
5271 
5272 	struct dk_geom	geom;
5273 	int	copy_len = sizeof (struct dk_geom);
5274 	int	rv = 0;
5275 
5276 	if (dir != VD_COPYOUT)
5277 		return (0);	/* nothing to do */
5278 
5279 	if ((from == NULL) || (to == NULL))
5280 		return (ENXIO);
5281 
5282 	VD_GEOM2DK_GEOM((vd_geom_t *)from, &geom);
5283 	rv = ddi_copyout(&geom, to, copy_len, mode);
5284 	if (rv != 0)
5285 		rv = EFAULT;
5286 
5287 	return (rv);
5288 }
5289 
5290 /*
5291  * Function:
5292  *	vdc_set_geom_convert()
5293  *
5294  * Description:
5295  *	This routine performs the necessary convertions from the DKIOCSGEOM
5296  *	Solaris structure to the format defined in FWARC 2006/195.
5297  *
5298  * Arguments:
5299  *	vdc	- the vDisk client
5300  *	from	- Buffer with data
5301  *	to	- Buffer where data is to be copied to
5302  *	mode	- flags passed to ioctl
5303  *	dir	- direction of copy (in or out)
5304  *
5305  * Return Code:
5306  *	0	- Success
5307  *	ENXIO	- Invalid buffer passed in
5308  *	EFAULT	- ddi_copyin of data failed
5309  */
5310 static int
5311 vdc_set_geom_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
5312 {
5313 	_NOTE(ARGUNUSED(vdc))
5314 
5315 	vd_geom_t	vdgeom;
5316 	void		*tmp_mem = NULL;
5317 	int		copy_len = sizeof (struct dk_geom);
5318 	int		rv = 0;
5319 
5320 	if (dir != VD_COPYIN)
5321 		return (0);	/* nothing to do */
5322 
5323 	if ((from == NULL) || (to == NULL))
5324 		return (ENXIO);
5325 
5326 	tmp_mem = kmem_alloc(copy_len, KM_SLEEP);
5327 
5328 	rv = ddi_copyin(from, tmp_mem, copy_len, mode);
5329 	if (rv != 0) {
5330 		kmem_free(tmp_mem, copy_len);
5331 		return (EFAULT);
5332 	}
5333 	DK_GEOM2VD_GEOM((struct dk_geom *)tmp_mem, &vdgeom);
5334 	bcopy(&vdgeom, to, sizeof (vdgeom));
5335 	kmem_free(tmp_mem, copy_len);
5336 
5337 	return (0);
5338 }
5339 
5340 static int
5341 vdc_get_efi_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
5342 {
5343 	_NOTE(ARGUNUSED(vdc))
5344 
5345 	vd_efi_t	*vd_efi;
5346 	dk_efi_t	dk_efi;
5347 	int		rv = 0;
5348 	void		*uaddr;
5349 
5350 	if ((from == NULL) || (to == NULL))
5351 		return (ENXIO);
5352 
5353 	if (dir == VD_COPYIN) {
5354 
5355 		vd_efi = (vd_efi_t *)to;
5356 
5357 		rv = ddi_copyin(from, &dk_efi, sizeof (dk_efi_t), mode);
5358 		if (rv != 0)
5359 			return (EFAULT);
5360 
5361 		vd_efi->lba = dk_efi.dki_lba;
5362 		vd_efi->length = dk_efi.dki_length;
5363 		bzero(vd_efi->data, vd_efi->length);
5364 
5365 	} else {
5366 
5367 		rv = ddi_copyin(to, &dk_efi, sizeof (dk_efi_t), mode);
5368 		if (rv != 0)
5369 			return (EFAULT);
5370 
5371 		uaddr = dk_efi.dki_data;
5372 
5373 		dk_efi.dki_data = kmem_alloc(dk_efi.dki_length, KM_SLEEP);
5374 
5375 		VD_EFI2DK_EFI((vd_efi_t *)from, &dk_efi);
5376 
5377 		rv = ddi_copyout(dk_efi.dki_data, uaddr, dk_efi.dki_length,
5378 		    mode);
5379 		if (rv != 0)
5380 			return (EFAULT);
5381 
5382 		kmem_free(dk_efi.dki_data, dk_efi.dki_length);
5383 	}
5384 
5385 	return (0);
5386 }
5387 
5388 static int
5389 vdc_set_efi_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
5390 {
5391 	_NOTE(ARGUNUSED(vdc))
5392 
5393 	dk_efi_t	dk_efi;
5394 	void		*uaddr;
5395 
5396 	if (dir == VD_COPYOUT)
5397 		return (0);	/* nothing to do */
5398 
5399 	if ((from == NULL) || (to == NULL))
5400 		return (ENXIO);
5401 
5402 	if (ddi_copyin(from, &dk_efi, sizeof (dk_efi_t), mode) != 0)
5403 		return (EFAULT);
5404 
5405 	uaddr = dk_efi.dki_data;
5406 
5407 	dk_efi.dki_data = kmem_alloc(dk_efi.dki_length, KM_SLEEP);
5408 
5409 	if (ddi_copyin(uaddr, dk_efi.dki_data, dk_efi.dki_length, mode) != 0)
5410 		return (EFAULT);
5411 
5412 	DK_EFI2VD_EFI(&dk_efi, (vd_efi_t *)to);
5413 
5414 	kmem_free(dk_efi.dki_data, dk_efi.dki_length);
5415 
5416 	return (0);
5417 }
5418 
5419 
5420 /* -------------------------------------------------------------------------- */
5421 
5422 /*
5423  * Function:
5424  *	vdc_create_fake_geometry()
5425  *
5426  * Description:
5427  *	This routine fakes up the disk info needed for some DKIO ioctls such
5428  *	as DKIOCINFO and DKIOCGMEDIAINFO [just like lofi(7D) and ramdisk(7D) do]
5429  *
5430  *	Note: This function must not be called until the vDisk attributes have
5431  *	been exchanged as part of the handshake with the vDisk server.
5432  *
5433  * Arguments:
5434  *	vdc	- soft state pointer for this instance of the device driver.
5435  *
5436  * Return Code:
5437  *	none.
5438  */
5439 static void
5440 vdc_create_fake_geometry(vdc_t *vdc)
5441 {
5442 	ASSERT(vdc != NULL);
5443 	ASSERT(vdc->vdisk_size != 0);
5444 	ASSERT(vdc->max_xfer_sz != 0);
5445 
5446 	/*
5447 	 * DKIOCINFO support
5448 	 */
5449 	if (vdc->cinfo == NULL)
5450 		vdc->cinfo = kmem_zalloc(sizeof (struct dk_cinfo), KM_SLEEP);
5451 
5452 	(void) strcpy(vdc->cinfo->dki_cname, VDC_DRIVER_NAME);
5453 	(void) strcpy(vdc->cinfo->dki_dname, VDC_DRIVER_NAME);
5454 	/* max_xfer_sz is #blocks so we don't need to divide by DEV_BSIZE */
5455 	vdc->cinfo->dki_maxtransfer = vdc->max_xfer_sz;
5456 	/*
5457 	 * We currently set the controller type to DKC_DIRECT for any disk.
5458 	 * When SCSI support is implemented, we will eventually change this
5459 	 * type to DKC_SCSI_CCS for disks supporting the SCSI protocol.
5460 	 * If the virtual disk is backed by a physical CD/DVD device or
5461 	 * an ISO image, modify the controller type to indicate this
5462 	 */
5463 	switch (vdc->vdisk_media) {
5464 	case VD_MEDIA_CD:
5465 	case VD_MEDIA_DVD:
5466 		vdc->cinfo->dki_ctype = DKC_CDROM;
5467 		break;
5468 	case VD_MEDIA_FIXED:
5469 		vdc->cinfo->dki_ctype = DKC_DIRECT;
5470 		break;
5471 	default:
5472 		/* in the case of v1.0 we default to a fixed disk */
5473 		vdc->cinfo->dki_ctype = DKC_DIRECT;
5474 		break;
5475 	}
5476 	vdc->cinfo->dki_flags = DKI_FMTVOL;
5477 	vdc->cinfo->dki_cnum = 0;
5478 	vdc->cinfo->dki_addr = 0;
5479 	vdc->cinfo->dki_space = 0;
5480 	vdc->cinfo->dki_prio = 0;
5481 	vdc->cinfo->dki_vec = 0;
5482 	vdc->cinfo->dki_unit = vdc->instance;
5483 	vdc->cinfo->dki_slave = 0;
5484 	/*
5485 	 * The partition number will be created on the fly depending on the
5486 	 * actual slice (i.e. minor node) that is used to request the data.
5487 	 */
5488 	vdc->cinfo->dki_partition = 0;
5489 
5490 	/*
5491 	 * DKIOCGMEDIAINFO support
5492 	 */
5493 	if (vdc->minfo == NULL)
5494 		vdc->minfo = kmem_zalloc(sizeof (struct dk_minfo), KM_SLEEP);
5495 
5496 	if (vio_ver_is_supported(vdc->ver, 1, 1)) {
5497 		vdc->minfo->dki_media_type =
5498 		    VD_MEDIATYPE2DK_MEDIATYPE(vdc->vdisk_media);
5499 	} else {
5500 		vdc->minfo->dki_media_type = DK_FIXED_DISK;
5501 	}
5502 
5503 	vdc->minfo->dki_capacity = vdc->vdisk_size;
5504 	vdc->minfo->dki_lbsize = vdc->block_size;
5505 }
5506 
5507 static ushort_t
5508 vdc_lbl2cksum(struct dk_label *label)
5509 {
5510 	int	count;
5511 	ushort_t sum, *sp;
5512 
5513 	count =	(sizeof (struct dk_label)) / (sizeof (short)) - 1;
5514 	sp = (ushort_t *)label;
5515 	sum = 0;
5516 	while (count--) {
5517 		sum ^= *sp++;
5518 	}
5519 
5520 	return (sum);
5521 }
5522 
5523 /*
5524  * Function:
5525  *	vdc_validate_geometry
5526  *
5527  * Description:
5528  *	This routine discovers the label and geometry of the disk. It stores
5529  *	the disk label and related information in the vdc structure. If it
5530  *	fails to validate the geometry or to discover the disk label then
5531  *	the label is marked as unknown (VD_DISK_LABEL_UNK).
5532  *
5533  * Arguments:
5534  *	vdc	- soft state pointer for this instance of the device driver.
5535  *
5536  * Return Code:
5537  *	0	- success.
5538  *	EINVAL	- unknown disk label.
5539  *	ENOTSUP	- geometry not applicable (EFI label).
5540  *	EIO	- error accessing the disk.
5541  */
5542 static int
5543 vdc_validate_geometry(vdc_t *vdc)
5544 {
5545 	buf_t	*buf;	/* BREAD requests need to be in a buf_t structure */
5546 	dev_t	dev;
5547 	int	rv;
5548 	struct dk_label label;
5549 	struct dk_geom geom;
5550 	struct vtoc vtoc;
5551 
5552 	ASSERT(vdc != NULL);
5553 	ASSERT(vdc->vtoc != NULL && vdc->geom != NULL);
5554 	ASSERT(MUTEX_HELD(&vdc->lock));
5555 
5556 	mutex_exit(&vdc->lock);
5557 
5558 	dev = makedevice(ddi_driver_major(vdc->dip),
5559 	    VD_MAKE_DEV(vdc->instance, 0));
5560 
5561 	rv = vd_process_ioctl(dev, DKIOCGGEOM, (caddr_t)&geom, FKIOCTL);
5562 	if (rv == 0)
5563 		rv = vd_process_ioctl(dev, DKIOCGVTOC, (caddr_t)&vtoc, FKIOCTL);
5564 
5565 	if (rv == ENOTSUP) {
5566 		/*
5567 		 * If the device does not support VTOC then we try
5568 		 * to read an EFI label.
5569 		 */
5570 		struct dk_gpt *efi;
5571 		size_t efi_len;
5572 
5573 		rv = vdc_efi_alloc_and_read(dev, &efi, &efi_len);
5574 
5575 		if (rv) {
5576 			DMSG(vdc, 0, "[%d] Failed to get EFI (err=%d)",
5577 			    vdc->instance, rv);
5578 			mutex_enter(&vdc->lock);
5579 			vdc_store_label_unk(vdc);
5580 			return (EIO);
5581 		}
5582 
5583 		mutex_enter(&vdc->lock);
5584 		vdc_store_label_efi(vdc, efi);
5585 		vd_efi_free(efi, efi_len);
5586 		return (ENOTSUP);
5587 	}
5588 
5589 	if (rv != 0) {
5590 		DMSG(vdc, 0, "[%d] Failed to get VTOC (err=%d)",
5591 		    vdc->instance, rv);
5592 		mutex_enter(&vdc->lock);
5593 		vdc_store_label_unk(vdc);
5594 		if (rv != EINVAL)
5595 			rv = EIO;
5596 		return (rv);
5597 	}
5598 
5599 	/* check that geometry and vtoc are valid */
5600 	if (geom.dkg_nhead == 0 || geom.dkg_nsect == 0 ||
5601 	    vtoc.v_sanity != VTOC_SANE) {
5602 		mutex_enter(&vdc->lock);
5603 		vdc_store_label_unk(vdc);
5604 		return (EINVAL);
5605 	}
5606 
5607 	/*
5608 	 * We have a disk and a valid VTOC. However this does not mean
5609 	 * that the disk currently have a VTOC label. The returned VTOC may
5610 	 * be a default VTOC to be used for configuring the disk (this is
5611 	 * what is done for disk image). So we read the label from the
5612 	 * beginning of the disk to ensure we really have a VTOC label.
5613 	 *
5614 	 * FUTURE: This could be the default way for reading the VTOC
5615 	 * from the disk as opposed to sending the VD_OP_GET_VTOC
5616 	 * to the server. This will be the default if vdc is implemented
5617 	 * ontop of cmlb.
5618 	 */
5619 
5620 	/*
5621 	 * Single slice disk does not support read using an absolute disk
5622 	 * offset so we just rely on the DKIOCGVTOC ioctl in that case.
5623 	 */
5624 	if (vdc->vdisk_type == VD_DISK_TYPE_SLICE) {
5625 		mutex_enter(&vdc->lock);
5626 		if (vtoc.v_nparts != 1) {
5627 			vdc_store_label_unk(vdc);
5628 			return (EINVAL);
5629 		}
5630 		vdc_store_label_vtoc(vdc, &geom, &vtoc);
5631 		return (0);
5632 	}
5633 
5634 	if (vtoc.v_nparts != V_NUMPAR) {
5635 		mutex_enter(&vdc->lock);
5636 		vdc_store_label_unk(vdc);
5637 		return (EINVAL);
5638 	}
5639 
5640 	/*
5641 	 * Read disk label from start of disk
5642 	 */
5643 	buf = kmem_alloc(sizeof (buf_t), KM_SLEEP);
5644 	bioinit(buf);
5645 	buf->b_un.b_addr = (caddr_t)&label;
5646 	buf->b_bcount = DK_LABEL_SIZE;
5647 	buf->b_flags = B_BUSY | B_READ;
5648 	buf->b_dev = cmpdev(dev);
5649 	rv = vdc_send_request(vdc, VD_OP_BREAD, (caddr_t)&label,
5650 	    DK_LABEL_SIZE, VD_SLICE_NONE, 0, CB_STRATEGY, buf, VIO_read_dir);
5651 	if (rv) {
5652 		DMSG(vdc, 1, "[%d] Failed to read disk block 0\n",
5653 		    vdc->instance);
5654 	} else {
5655 		rv = biowait(buf);
5656 		biofini(buf);
5657 	}
5658 	kmem_free(buf, sizeof (buf_t));
5659 
5660 	if (rv != 0 || label.dkl_magic != DKL_MAGIC ||
5661 	    label.dkl_cksum != vdc_lbl2cksum(&label)) {
5662 		DMSG(vdc, 1, "[%d] Got VTOC with invalid label\n",
5663 		    vdc->instance);
5664 		mutex_enter(&vdc->lock);
5665 		vdc_store_label_unk(vdc);
5666 		return (EINVAL);
5667 	}
5668 
5669 	mutex_enter(&vdc->lock);
5670 	vdc_store_label_vtoc(vdc, &geom, &vtoc);
5671 	return (0);
5672 }
5673 
5674 /*
5675  * Function:
5676  *	vdc_validate
5677  *
5678  * Description:
5679  *	This routine discovers the label of the disk and create the
5680  *	appropriate device nodes if the label has changed.
5681  *
5682  * Arguments:
5683  *	vdc	- soft state pointer for this instance of the device driver.
5684  *
5685  * Return Code:
5686  *	none.
5687  */
5688 static void
5689 vdc_validate(vdc_t *vdc)
5690 {
5691 	vd_disk_label_t old_label;
5692 	struct vtoc old_vtoc;
5693 	int rv;
5694 
5695 	ASSERT(!MUTEX_HELD(&vdc->lock));
5696 
5697 	mutex_enter(&vdc->lock);
5698 
5699 	/* save the current label and vtoc */
5700 	old_label = vdc->vdisk_label;
5701 	bcopy(vdc->vtoc, &old_vtoc, sizeof (struct vtoc));
5702 
5703 	/* check the geometry */
5704 	(void) vdc_validate_geometry(vdc);
5705 
5706 	/* if the disk label has changed, update device nodes */
5707 	if (vdc->vdisk_label != old_label) {
5708 
5709 		if (vdc->vdisk_label == VD_DISK_LABEL_EFI)
5710 			rv = vdc_create_device_nodes_efi(vdc);
5711 		else
5712 			rv = vdc_create_device_nodes_vtoc(vdc);
5713 
5714 		if (rv != 0) {
5715 			DMSG(vdc, 0, "![%d] Failed to update device nodes",
5716 			    vdc->instance);
5717 		}
5718 	}
5719 
5720 	/* if the vtoc has changed, update device nodes properties */
5721 	if (bcmp(vdc->vtoc, &old_vtoc, sizeof (struct vtoc)) != 0) {
5722 
5723 		if (vdc_create_device_nodes_props(vdc) != 0) {
5724 			DMSG(vdc, 0, "![%d] Failed to update device nodes"
5725 			    " properties", vdc->instance);
5726 		}
5727 	}
5728 
5729 	mutex_exit(&vdc->lock);
5730 }
5731 
5732 static void
5733 vdc_validate_task(void *arg)
5734 {
5735 	vdc_t *vdc = (vdc_t *)arg;
5736 
5737 	vdc_validate(vdc);
5738 
5739 	mutex_enter(&vdc->lock);
5740 	ASSERT(vdc->validate_pending > 0);
5741 	vdc->validate_pending--;
5742 	mutex_exit(&vdc->lock);
5743 }
5744 
5745 /*
5746  * Function:
5747  *	vdc_setup_devid()
5748  *
5749  * Description:
5750  *	This routine discovers the devid of a vDisk. It requests the devid of
5751  *	the underlying device from the vDisk server, builds an encapsulated
5752  *	devid based on the retrieved devid and registers that new devid to
5753  *	the vDisk.
5754  *
5755  * Arguments:
5756  *	vdc	- soft state pointer for this instance of the device driver.
5757  *
5758  * Return Code:
5759  *	0	- A devid was succesfully registered for the vDisk
5760  */
5761 static int
5762 vdc_setup_devid(vdc_t *vdc)
5763 {
5764 	int rv;
5765 	vd_devid_t *vd_devid;
5766 	size_t bufsize, bufid_len;
5767 
5768 	/*
5769 	 * At first sight, we don't know the size of the devid that the
5770 	 * server will return but this size will be encoded into the
5771 	 * reply. So we do a first request using a default size then we
5772 	 * check if this size was large enough. If not then we do a second
5773 	 * request with the correct size returned by the server. Note that
5774 	 * ldc requires size to be 8-byte aligned.
5775 	 */
5776 	bufsize = P2ROUNDUP(VD_DEVID_SIZE(VD_DEVID_DEFAULT_LEN),
5777 	    sizeof (uint64_t));
5778 	vd_devid = kmem_zalloc(bufsize, KM_SLEEP);
5779 	bufid_len = bufsize - sizeof (vd_efi_t) - 1;
5780 
5781 	rv = vdc_do_sync_op(vdc, VD_OP_GET_DEVID, (caddr_t)vd_devid,
5782 	    bufsize, 0, 0, CB_SYNC, 0, VIO_both_dir);
5783 
5784 	DMSG(vdc, 2, "sync_op returned %d\n", rv);
5785 
5786 	if (rv) {
5787 		kmem_free(vd_devid, bufsize);
5788 		return (rv);
5789 	}
5790 
5791 	if (vd_devid->length > bufid_len) {
5792 		/*
5793 		 * The returned devid is larger than the buffer used. Try again
5794 		 * with a buffer with the right size.
5795 		 */
5796 		kmem_free(vd_devid, bufsize);
5797 		bufsize = P2ROUNDUP(VD_DEVID_SIZE(vd_devid->length),
5798 		    sizeof (uint64_t));
5799 		vd_devid = kmem_zalloc(bufsize, KM_SLEEP);
5800 		bufid_len = bufsize - sizeof (vd_efi_t) - 1;
5801 
5802 		rv = vdc_do_sync_op(vdc, VD_OP_GET_DEVID,
5803 		    (caddr_t)vd_devid, bufsize, 0, 0, CB_SYNC, 0,
5804 		    VIO_both_dir);
5805 
5806 		if (rv) {
5807 			kmem_free(vd_devid, bufsize);
5808 			return (rv);
5809 		}
5810 	}
5811 
5812 	/*
5813 	 * The virtual disk should have the same device id as the one associated
5814 	 * with the physical disk it is mapped on, otherwise sharing a disk
5815 	 * between a LDom and a non-LDom may not work (for example for a shared
5816 	 * SVM disk set).
5817 	 *
5818 	 * The DDI framework does not allow creating a device id with any
5819 	 * type so we first create a device id of type DEVID_ENCAP and then
5820 	 * we restore the orignal type of the physical device.
5821 	 */
5822 
5823 	DMSG(vdc, 2, ": devid length = %d\n", vd_devid->length);
5824 
5825 	/* build an encapsulated devid based on the returned devid */
5826 	if (ddi_devid_init(vdc->dip, DEVID_ENCAP, vd_devid->length,
5827 	    vd_devid->id, &vdc->devid) != DDI_SUCCESS) {
5828 		DMSG(vdc, 1, "[%d] Fail to created devid\n", vdc->instance);
5829 		kmem_free(vd_devid, bufsize);
5830 		return (1);
5831 	}
5832 
5833 	DEVID_FORMTYPE((impl_devid_t *)vdc->devid, vd_devid->type);
5834 
5835 	ASSERT(ddi_devid_valid(vdc->devid) == DDI_SUCCESS);
5836 
5837 	kmem_free(vd_devid, bufsize);
5838 
5839 	if (ddi_devid_register(vdc->dip, vdc->devid) != DDI_SUCCESS) {
5840 		DMSG(vdc, 1, "[%d] Fail to register devid\n", vdc->instance);
5841 		return (1);
5842 	}
5843 
5844 	return (0);
5845 }
5846 
5847 static void
5848 vdc_store_label_efi(vdc_t *vdc, struct dk_gpt *efi)
5849 {
5850 	struct vtoc *vtoc = vdc->vtoc;
5851 
5852 	ASSERT(MUTEX_HELD(&vdc->lock));
5853 
5854 	vdc->vdisk_label = VD_DISK_LABEL_EFI;
5855 	bzero(vdc->geom, sizeof (struct dk_geom));
5856 	vd_efi_to_vtoc(efi, vtoc);
5857 	if (vdc->vdisk_type == VD_DISK_TYPE_SLICE) {
5858 		/*
5859 		 * vd_efi_to_vtoc() will store information about the EFI Sun
5860 		 * reserved partition (representing the entire disk) into
5861 		 * partition 7. However single-slice device will only have
5862 		 * that single partition and the vdc driver expects to find
5863 		 * information about that partition in slice 0. So we need
5864 		 * to copy information from slice 7 to slice 0.
5865 		 */
5866 		vtoc->v_part[0].p_tag = vtoc->v_part[VD_EFI_WD_SLICE].p_tag;
5867 		vtoc->v_part[0].p_flag = vtoc->v_part[VD_EFI_WD_SLICE].p_flag;
5868 		vtoc->v_part[0].p_start = vtoc->v_part[VD_EFI_WD_SLICE].p_start;
5869 		vtoc->v_part[0].p_size =  vtoc->v_part[VD_EFI_WD_SLICE].p_size;
5870 	}
5871 }
5872 
5873 static void
5874 vdc_store_label_vtoc(vdc_t *vdc, struct dk_geom *geom, struct vtoc *vtoc)
5875 {
5876 	ASSERT(MUTEX_HELD(&vdc->lock));
5877 
5878 	vdc->vdisk_label = VD_DISK_LABEL_VTOC;
5879 	bcopy(vtoc, vdc->vtoc, sizeof (struct vtoc));
5880 	bcopy(geom, vdc->geom, sizeof (struct dk_geom));
5881 }
5882 
5883 static void
5884 vdc_store_label_unk(vdc_t *vdc)
5885 {
5886 	ASSERT(MUTEX_HELD(&vdc->lock));
5887 
5888 	vdc->vdisk_label = VD_DISK_LABEL_UNK;
5889 	bzero(vdc->vtoc, sizeof (struct vtoc));
5890 	bzero(vdc->geom, sizeof (struct dk_geom));
5891 }
5892