xref: /illumos-gate/usr/src/uts/sun4v/io/vdc.c (revision 1a7b528f36ec61147fc237b7b9a23642fab7ad13)
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/vdsk_common.h>
93 #include <sys/vdsk_mailbox.h>
94 #include <sys/vdc.h>
95 
96 /*
97  * function prototypes
98  */
99 
100 /* standard driver functions */
101 static int	vdc_open(dev_t *dev, int flag, int otyp, cred_t *cred);
102 static int	vdc_close(dev_t dev, int flag, int otyp, cred_t *cred);
103 static int	vdc_strategy(struct buf *buf);
104 static int	vdc_print(dev_t dev, char *str);
105 static int	vdc_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblk);
106 static int	vdc_read(dev_t dev, struct uio *uio, cred_t *cred);
107 static int	vdc_write(dev_t dev, struct uio *uio, cred_t *cred);
108 static int	vdc_ioctl(dev_t dev, int cmd, intptr_t arg, int mode,
109 			cred_t *credp, int *rvalp);
110 static int	vdc_aread(dev_t dev, struct aio_req *aio, cred_t *cred);
111 static int	vdc_awrite(dev_t dev, struct aio_req *aio, cred_t *cred);
112 
113 static int	vdc_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd,
114 			void *arg, void **resultp);
115 static int	vdc_attach(dev_info_t *dip, ddi_attach_cmd_t cmd);
116 static int	vdc_detach(dev_info_t *dip, ddi_detach_cmd_t cmd);
117 
118 /* setup */
119 static void	vdc_min(struct buf *bufp);
120 static int	vdc_send(vdc_t *vdc, caddr_t pkt, size_t *msglen);
121 static int	vdc_do_ldc_init(vdc_t *vdc, md_t *mdp, mde_cookie_t vd_node);
122 static int	vdc_start_ldc_connection(vdc_t *vdc);
123 static int	vdc_create_device_nodes(vdc_t *vdc);
124 static int	vdc_create_device_nodes_efi(vdc_t *vdc);
125 static int	vdc_create_device_nodes_vtoc(vdc_t *vdc);
126 static int	vdc_create_device_nodes_props(vdc_t *vdc);
127 static int	vdc_get_md_node(dev_info_t *dip, md_t **mdpp,
128 		    mde_cookie_t *vd_nodep, mde_cookie_t *vd_portp);
129 static int	vdc_get_ldc_id(md_t *, mde_cookie_t, uint64_t *);
130 static int	vdc_do_ldc_up(vdc_t *vdc);
131 static void	vdc_terminate_ldc(vdc_t *vdc);
132 static int	vdc_init_descriptor_ring(vdc_t *vdc);
133 static void	vdc_destroy_descriptor_ring(vdc_t *vdc);
134 static int	vdc_setup_devid(vdc_t *vdc);
135 static void	vdc_store_label_efi(vdc_t *vdc, struct dk_gpt *efi);
136 static void	vdc_store_label_vtoc(vdc_t *, struct dk_geom *, struct vtoc *);
137 static void	vdc_store_label_unk(vdc_t *vdc);
138 static boolean_t vdc_is_opened(vdc_t *vdc);
139 
140 /* handshake with vds */
141 static int		vdc_init_ver_negotiation(vdc_t *vdc, vio_ver_t ver);
142 static int		vdc_ver_negotiation(vdc_t *vdcp);
143 static int		vdc_init_attr_negotiation(vdc_t *vdc);
144 static int		vdc_attr_negotiation(vdc_t *vdcp);
145 static int		vdc_init_dring_negotiate(vdc_t *vdc);
146 static int		vdc_dring_negotiation(vdc_t *vdcp);
147 static int		vdc_send_rdx(vdc_t *vdcp);
148 static int		vdc_rdx_exchange(vdc_t *vdcp);
149 static boolean_t	vdc_is_supported_version(vio_ver_msg_t *ver_msg);
150 
151 /* processing incoming messages from vDisk server */
152 static void	vdc_process_msg_thread(vdc_t *vdc);
153 static int	vdc_recv(vdc_t *vdc, vio_msg_t *msgp, size_t *nbytesp);
154 
155 static uint_t	vdc_handle_cb(uint64_t event, caddr_t arg);
156 static int	vdc_process_data_msg(vdc_t *vdc, vio_msg_t *msg);
157 static int	vdc_process_err_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_uscsicmd_convert(vdc_t *vdc, void *from, void *to,
199 		    int mode, int dir);
200 static int	vdc_get_efi_convert(vdc_t *vdc, void *from, void *to,
201 		    int mode, int dir);
202 static int	vdc_set_efi_convert(vdc_t *vdc, void *from, void *to,
203 		    int mode, int dir);
204 
205 /*
206  * Module variables
207  */
208 
209 /*
210  * Tunable variables to control how long vdc waits before timing out on
211  * various operations
212  */
213 static int	vdc_retries = 10;
214 static int	vdc_hshake_retries = 3;
215 
216 static int	vdc_timeout = 0; /* units: seconds */
217 
218 /* calculated from 'vdc_usec_timeout' during attach */
219 static uint64_t	vdc_hz_timeout;				/* units: Hz */
220 static uint64_t	vdc_usec_timeout = 30 * MICROSEC;	/* 30s units: ns */
221 
222 static uint64_t vdc_hz_min_ldc_delay;
223 static uint64_t vdc_min_timeout_ldc = 1 * MILLISEC;
224 static uint64_t vdc_hz_max_ldc_delay;
225 static uint64_t vdc_max_timeout_ldc = 100 * MILLISEC;
226 
227 static uint64_t vdc_ldc_read_init_delay = 1 * MILLISEC;
228 static uint64_t vdc_ldc_read_max_delay = 100 * MILLISEC;
229 
230 /* values for dumping - need to run in a tighter loop */
231 static uint64_t	vdc_usec_timeout_dump = 100 * MILLISEC;	/* 0.1s units: ns */
232 static int	vdc_dump_retries = 100;
233 
234 /* Count of the number of vdc instances attached */
235 static volatile uint32_t	vdc_instance_count = 0;
236 
237 /* Soft state pointer */
238 static void	*vdc_state;
239 
240 /*
241  * Controlling the verbosity of the error/debug messages
242  *
243  * vdc_msglevel - controls level of messages
244  * vdc_matchinst - 64-bit variable where each bit corresponds
245  *                 to the vdc instance the vdc_msglevel applies.
246  */
247 int		vdc_msglevel = 0x0;
248 uint64_t	vdc_matchinst = 0ull;
249 
250 /*
251  * Supported vDisk protocol version pairs.
252  *
253  * The first array entry is the latest and preferred version.
254  */
255 static const vio_ver_t	vdc_version[] = {{1, 0}};
256 
257 static struct cb_ops vdc_cb_ops = {
258 	vdc_open,	/* cb_open */
259 	vdc_close,	/* cb_close */
260 	vdc_strategy,	/* cb_strategy */
261 	vdc_print,	/* cb_print */
262 	vdc_dump,	/* cb_dump */
263 	vdc_read,	/* cb_read */
264 	vdc_write,	/* cb_write */
265 	vdc_ioctl,	/* cb_ioctl */
266 	nodev,		/* cb_devmap */
267 	nodev,		/* cb_mmap */
268 	nodev,		/* cb_segmap */
269 	nochpoll,	/* cb_chpoll */
270 	ddi_prop_op,	/* cb_prop_op */
271 	NULL,		/* cb_str */
272 	D_MP | D_64BIT,	/* cb_flag */
273 	CB_REV,		/* cb_rev */
274 	vdc_aread,	/* cb_aread */
275 	vdc_awrite	/* cb_awrite */
276 };
277 
278 static struct dev_ops vdc_ops = {
279 	DEVO_REV,	/* devo_rev */
280 	0,		/* devo_refcnt */
281 	vdc_getinfo,	/* devo_getinfo */
282 	nulldev,	/* devo_identify */
283 	nulldev,	/* devo_probe */
284 	vdc_attach,	/* devo_attach */
285 	vdc_detach,	/* devo_detach */
286 	nodev,		/* devo_reset */
287 	&vdc_cb_ops,	/* devo_cb_ops */
288 	NULL,		/* devo_bus_ops */
289 	nulldev		/* devo_power */
290 };
291 
292 static struct modldrv modldrv = {
293 	&mod_driverops,
294 	"virtual disk client",
295 	&vdc_ops,
296 };
297 
298 static struct modlinkage modlinkage = {
299 	MODREV_1,
300 	&modldrv,
301 	NULL
302 };
303 
304 /* -------------------------------------------------------------------------- */
305 
306 /*
307  * Device Driver housekeeping and setup
308  */
309 
310 int
311 _init(void)
312 {
313 	int	status;
314 
315 	if ((status = ddi_soft_state_init(&vdc_state, sizeof (vdc_t), 1)) != 0)
316 		return (status);
317 	if ((status = mod_install(&modlinkage)) != 0)
318 		ddi_soft_state_fini(&vdc_state);
319 	vdc_efi_init(vd_process_ioctl);
320 	return (status);
321 }
322 
323 int
324 _info(struct modinfo *modinfop)
325 {
326 	return (mod_info(&modlinkage, modinfop));
327 }
328 
329 int
330 _fini(void)
331 {
332 	int	status;
333 
334 	if ((status = mod_remove(&modlinkage)) != 0)
335 		return (status);
336 	vdc_efi_fini();
337 	ddi_soft_state_fini(&vdc_state);
338 	return (0);
339 }
340 
341 static int
342 vdc_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd,  void *arg, void **resultp)
343 {
344 	_NOTE(ARGUNUSED(dip))
345 
346 	int	instance = VDCUNIT((dev_t)arg);
347 	vdc_t	*vdc = NULL;
348 
349 	switch (cmd) {
350 	case DDI_INFO_DEVT2DEVINFO:
351 		if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
352 			*resultp = NULL;
353 			return (DDI_FAILURE);
354 		}
355 		*resultp = vdc->dip;
356 		return (DDI_SUCCESS);
357 	case DDI_INFO_DEVT2INSTANCE:
358 		*resultp = (void *)(uintptr_t)instance;
359 		return (DDI_SUCCESS);
360 	default:
361 		*resultp = NULL;
362 		return (DDI_FAILURE);
363 	}
364 }
365 
366 static int
367 vdc_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
368 {
369 	int	instance;
370 	int	rv;
371 	vdc_t	*vdc = NULL;
372 
373 	switch (cmd) {
374 	case DDI_DETACH:
375 		/* the real work happens below */
376 		break;
377 	case DDI_SUSPEND:
378 		/* nothing to do for this non-device */
379 		return (DDI_SUCCESS);
380 	default:
381 		return (DDI_FAILURE);
382 	}
383 
384 	ASSERT(cmd == DDI_DETACH);
385 	instance = ddi_get_instance(dip);
386 	DMSGX(1, "[%d] Entered\n", instance);
387 
388 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
389 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
390 		return (DDI_FAILURE);
391 	}
392 
393 	if (vdc_is_opened(vdc)) {
394 		DMSG(vdc, 0, "[%d] Cannot detach: device is open", instance);
395 		return (DDI_FAILURE);
396 	}
397 
398 	if (vdc->dkio_flush_pending) {
399 		DMSG(vdc, 0,
400 		    "[%d] Cannot detach: %d outstanding DKIO flushes\n",
401 		    instance, vdc->dkio_flush_pending);
402 		return (DDI_FAILURE);
403 	}
404 
405 	if (vdc->validate_pending) {
406 		DMSG(vdc, 0,
407 		    "[%d] Cannot detach: %d outstanding validate request\n",
408 		    instance, vdc->validate_pending);
409 		return (DDI_FAILURE);
410 	}
411 
412 	DMSG(vdc, 0, "[%d] proceeding...\n", instance);
413 
414 	/* mark instance as detaching */
415 	vdc->lifecycle	= VDC_LC_DETACHING;
416 
417 	/*
418 	 * try and disable callbacks to prevent another handshake
419 	 */
420 	rv = ldc_set_cb_mode(vdc->ldc_handle, LDC_CB_DISABLE);
421 	DMSG(vdc, 0, "callback disabled (rv=%d)\n", rv);
422 
423 	if (vdc->initialized & VDC_THREAD) {
424 		mutex_enter(&vdc->read_lock);
425 		if ((vdc->read_state == VDC_READ_WAITING) ||
426 		    (vdc->read_state == VDC_READ_RESET)) {
427 			vdc->read_state = VDC_READ_RESET;
428 			cv_signal(&vdc->read_cv);
429 		}
430 
431 		mutex_exit(&vdc->read_lock);
432 
433 		/* wake up any thread waiting for connection to come online */
434 		mutex_enter(&vdc->lock);
435 		if (vdc->state == VDC_STATE_INIT_WAITING) {
436 			DMSG(vdc, 0,
437 			    "[%d] write reset - move to resetting state...\n",
438 			    instance);
439 			vdc->state = VDC_STATE_RESETTING;
440 			cv_signal(&vdc->initwait_cv);
441 		}
442 		mutex_exit(&vdc->lock);
443 
444 		/* now wait until state transitions to VDC_STATE_DETACH */
445 		thread_join(vdc->msg_proc_thr->t_did);
446 		ASSERT(vdc->state == VDC_STATE_DETACH);
447 		DMSG(vdc, 0, "[%d] Reset thread exit and join ..\n",
448 		    vdc->instance);
449 	}
450 
451 	mutex_enter(&vdc->lock);
452 
453 	if (vdc->initialized & VDC_DRING)
454 		vdc_destroy_descriptor_ring(vdc);
455 
456 	if (vdc->initialized & VDC_LDC)
457 		vdc_terminate_ldc(vdc);
458 
459 	mutex_exit(&vdc->lock);
460 
461 	if (vdc->initialized & VDC_MINOR) {
462 		ddi_prop_remove_all(dip);
463 		ddi_remove_minor_node(dip, NULL);
464 	}
465 
466 	if (vdc->initialized & VDC_LOCKS) {
467 		mutex_destroy(&vdc->lock);
468 		mutex_destroy(&vdc->read_lock);
469 		cv_destroy(&vdc->initwait_cv);
470 		cv_destroy(&vdc->dring_free_cv);
471 		cv_destroy(&vdc->membind_cv);
472 		cv_destroy(&vdc->sync_pending_cv);
473 		cv_destroy(&vdc->sync_blocked_cv);
474 		cv_destroy(&vdc->read_cv);
475 		cv_destroy(&vdc->running_cv);
476 	}
477 
478 	if (vdc->minfo)
479 		kmem_free(vdc->minfo, sizeof (struct dk_minfo));
480 
481 	if (vdc->cinfo)
482 		kmem_free(vdc->cinfo, sizeof (struct dk_cinfo));
483 
484 	if (vdc->vtoc)
485 		kmem_free(vdc->vtoc, sizeof (struct vtoc));
486 
487 	if (vdc->geom)
488 		kmem_free(vdc->geom, sizeof (struct dk_geom));
489 
490 	if (vdc->devid) {
491 		ddi_devid_unregister(dip);
492 		ddi_devid_free(vdc->devid);
493 	}
494 
495 	if (vdc->initialized & VDC_SOFT_STATE)
496 		ddi_soft_state_free(vdc_state, instance);
497 
498 	DMSG(vdc, 0, "[%d] End %p\n", instance, (void *)vdc);
499 
500 	return (DDI_SUCCESS);
501 }
502 
503 
504 static int
505 vdc_do_attach(dev_info_t *dip)
506 {
507 	int		instance;
508 	vdc_t		*vdc = NULL;
509 	int		status;
510 	md_t		*mdp;
511 	mde_cookie_t	vd_node, vd_port;
512 
513 	ASSERT(dip != NULL);
514 
515 	instance = ddi_get_instance(dip);
516 	if (ddi_soft_state_zalloc(vdc_state, instance) != DDI_SUCCESS) {
517 		cmn_err(CE_NOTE, "[%d] Couldn't alloc state structure",
518 		    instance);
519 		return (DDI_FAILURE);
520 	}
521 
522 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
523 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
524 		return (DDI_FAILURE);
525 	}
526 
527 	/*
528 	 * We assign the value to initialized in this case to zero out the
529 	 * variable and then set bits in it to indicate what has been done
530 	 */
531 	vdc->initialized = VDC_SOFT_STATE;
532 
533 	vdc_hz_timeout = drv_usectohz(vdc_usec_timeout);
534 
535 	vdc_hz_min_ldc_delay = drv_usectohz(vdc_min_timeout_ldc);
536 	vdc_hz_max_ldc_delay = drv_usectohz(vdc_max_timeout_ldc);
537 
538 	vdc->dip	= dip;
539 	vdc->instance	= instance;
540 	vdc->vdisk_type	= VD_DISK_TYPE_UNK;
541 	vdc->vdisk_label = VD_DISK_LABEL_UNK;
542 	vdc->state	= VDC_STATE_INIT;
543 	vdc->lifecycle	= VDC_LC_ATTACHING;
544 	vdc->ldc_state	= 0;
545 	vdc->session_id = 0;
546 	vdc->block_size = DEV_BSIZE;
547 	vdc->max_xfer_sz = maxphys / DEV_BSIZE;
548 
549 	vdc->vtoc = NULL;
550 	vdc->geom = NULL;
551 	vdc->cinfo = NULL;
552 	vdc->minfo = NULL;
553 
554 	mutex_init(&vdc->lock, NULL, MUTEX_DRIVER, NULL);
555 	cv_init(&vdc->initwait_cv, NULL, CV_DRIVER, NULL);
556 	cv_init(&vdc->dring_free_cv, NULL, CV_DRIVER, NULL);
557 	cv_init(&vdc->membind_cv, NULL, CV_DRIVER, NULL);
558 	cv_init(&vdc->running_cv, NULL, CV_DRIVER, NULL);
559 
560 	vdc->threads_pending = 0;
561 	vdc->sync_op_pending = B_FALSE;
562 	vdc->sync_op_blocked = B_FALSE;
563 	cv_init(&vdc->sync_pending_cv, NULL, CV_DRIVER, NULL);
564 	cv_init(&vdc->sync_blocked_cv, NULL, CV_DRIVER, NULL);
565 
566 	/* init blocking msg read functionality */
567 	mutex_init(&vdc->read_lock, NULL, MUTEX_DRIVER, NULL);
568 	cv_init(&vdc->read_cv, NULL, CV_DRIVER, NULL);
569 	vdc->read_state = VDC_READ_IDLE;
570 
571 	vdc->initialized |= VDC_LOCKS;
572 
573 	/* get device and port MD node for this disk instance */
574 	if (vdc_get_md_node(dip, &mdp, &vd_node, &vd_port) != 0) {
575 		cmn_err(CE_NOTE, "[%d] Could not get machine description node",
576 		    instance);
577 		return (DDI_FAILURE);
578 	}
579 
580 	/* set the connection timeout */
581 	if (vd_port == NULL || (md_get_prop_val(mdp, vd_port,
582 	    VDC_MD_TIMEOUT, &vdc->ctimeout) != 0)) {
583 		vdc->ctimeout = 0;
584 	}
585 
586 	/* initialise LDC channel which will be used to communicate with vds */
587 	status = vdc_do_ldc_init(vdc, mdp, vd_node);
588 
589 	(void) md_fini_handle(mdp);
590 
591 	if (status != 0) {
592 		cmn_err(CE_NOTE, "[%d] Couldn't initialize LDC", instance);
593 		goto return_status;
594 	}
595 
596 	/* initialize the thread responsible for managing state with server */
597 	vdc->msg_proc_thr = thread_create(NULL, 0, vdc_process_msg_thread,
598 	    vdc, 0, &p0, TS_RUN, minclsyspri);
599 	if (vdc->msg_proc_thr == NULL) {
600 		cmn_err(CE_NOTE, "[%d] Failed to create msg processing thread",
601 		    instance);
602 		return (DDI_FAILURE);
603 	}
604 
605 	vdc->initialized |= VDC_THREAD;
606 
607 	atomic_inc_32(&vdc_instance_count);
608 
609 	/*
610 	 * Check the disk label. This will send requests and do the handshake.
611 	 * We don't really care about the disk label now. What we really need is
612 	 * the handshake do be done so that we know the type of the disk (slice
613 	 * or full disk) and the appropriate device nodes can be created.
614 	 */
615 	vdc->vdisk_label = VD_DISK_LABEL_UNK;
616 	vdc->vtoc = kmem_zalloc(sizeof (struct vtoc), KM_SLEEP);
617 	vdc->geom = kmem_zalloc(sizeof (struct dk_geom), KM_SLEEP);
618 
619 	mutex_enter(&vdc->lock);
620 	(void) vdc_validate_geometry(vdc);
621 	mutex_exit(&vdc->lock);
622 
623 	/*
624 	 * Now that we have the device info we can create the
625 	 * device nodes and properties
626 	 */
627 	status = vdc_create_device_nodes(vdc);
628 	if (status) {
629 		DMSG(vdc, 0, "[%d] Failed to create device nodes",
630 		    instance);
631 		goto return_status;
632 	}
633 	status = vdc_create_device_nodes_props(vdc);
634 	if (status) {
635 		DMSG(vdc, 0, "[%d] Failed to create device nodes"
636 		    " properties (%d)", instance, status);
637 		goto return_status;
638 	}
639 
640 	/*
641 	 * Setup devid
642 	 */
643 	if (vdc_setup_devid(vdc)) {
644 		DMSG(vdc, 0, "[%d] No device id available\n", instance);
645 	}
646 
647 	ddi_report_dev(dip);
648 	vdc->lifecycle	= VDC_LC_ONLINE;
649 	DMSG(vdc, 0, "[%d] Attach tasks successful\n", instance);
650 
651 return_status:
652 	DMSG(vdc, 0, "[%d] Attach completed\n", instance);
653 	return (status);
654 }
655 
656 static int
657 vdc_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
658 {
659 	int	status;
660 
661 	switch (cmd) {
662 	case DDI_ATTACH:
663 		if ((status = vdc_do_attach(dip)) != 0)
664 			(void) vdc_detach(dip, DDI_DETACH);
665 		return (status);
666 	case DDI_RESUME:
667 		/* nothing to do for this non-device */
668 		return (DDI_SUCCESS);
669 	default:
670 		return (DDI_FAILURE);
671 	}
672 }
673 
674 static int
675 vdc_do_ldc_init(vdc_t *vdc, md_t *mdp, mde_cookie_t vd_node)
676 {
677 	int			status = 0;
678 	ldc_status_t		ldc_state;
679 	ldc_attr_t		ldc_attr;
680 	uint64_t		ldc_id = 0;
681 
682 	ASSERT(vdc != NULL);
683 
684 	vdc->initialized |= VDC_LDC;
685 
686 	if ((status = vdc_get_ldc_id(mdp, vd_node, &ldc_id)) != 0) {
687 		DMSG(vdc, 0, "[%d] Failed to get LDC channel ID property",
688 		    vdc->instance);
689 		return (EIO);
690 	}
691 
692 	DMSGX(0, "[%d] LDC id is 0x%lx\n", vdc->instance, ldc_id);
693 
694 	vdc->ldc_id = ldc_id;
695 
696 	ldc_attr.devclass = LDC_DEV_BLK;
697 	ldc_attr.instance = vdc->instance;
698 	ldc_attr.mode = LDC_MODE_UNRELIABLE;	/* unreliable transport */
699 	ldc_attr.mtu = VD_LDC_MTU;
700 
701 	if ((vdc->initialized & VDC_LDC_INIT) == 0) {
702 		status = ldc_init(ldc_id, &ldc_attr, &vdc->ldc_handle);
703 		if (status != 0) {
704 			DMSG(vdc, 0, "[%d] ldc_init(chan %ld) returned %d",
705 			    vdc->instance, ldc_id, status);
706 			return (status);
707 		}
708 		vdc->initialized |= VDC_LDC_INIT;
709 	}
710 	status = ldc_status(vdc->ldc_handle, &ldc_state);
711 	if (status != 0) {
712 		DMSG(vdc, 0, "[%d] Cannot discover LDC status [err=%d]",
713 		    vdc->instance, status);
714 		return (status);
715 	}
716 	vdc->ldc_state = ldc_state;
717 
718 	if ((vdc->initialized & VDC_LDC_CB) == 0) {
719 		status = ldc_reg_callback(vdc->ldc_handle, vdc_handle_cb,
720 		    (caddr_t)vdc);
721 		if (status != 0) {
722 			DMSG(vdc, 0, "[%d] LDC callback reg. failed (%d)",
723 			    vdc->instance, status);
724 			return (status);
725 		}
726 		vdc->initialized |= VDC_LDC_CB;
727 	}
728 
729 	vdc->initialized |= VDC_LDC;
730 
731 	/*
732 	 * At this stage we have initialised LDC, we will now try and open
733 	 * the connection.
734 	 */
735 	if (vdc->ldc_state == LDC_INIT) {
736 		status = ldc_open(vdc->ldc_handle);
737 		if (status != 0) {
738 			DMSG(vdc, 0, "[%d] ldc_open(chan %ld) returned %d",
739 			    vdc->instance, vdc->ldc_id, status);
740 			return (status);
741 		}
742 		vdc->initialized |= VDC_LDC_OPEN;
743 	}
744 
745 	return (status);
746 }
747 
748 static int
749 vdc_start_ldc_connection(vdc_t *vdc)
750 {
751 	int		status = 0;
752 
753 	ASSERT(vdc != NULL);
754 
755 	ASSERT(MUTEX_HELD(&vdc->lock));
756 
757 	status = vdc_do_ldc_up(vdc);
758 
759 	DMSG(vdc, 0, "[%d] Finished bringing up LDC\n", vdc->instance);
760 
761 	return (status);
762 }
763 
764 static int
765 vdc_stop_ldc_connection(vdc_t *vdcp)
766 {
767 	int	status;
768 
769 	DMSG(vdcp, 0, ": Resetting connection to vDisk server : state %d\n",
770 	    vdcp->state);
771 
772 	status = ldc_down(vdcp->ldc_handle);
773 	DMSG(vdcp, 0, "ldc_down() = %d\n", status);
774 
775 	vdcp->initialized &= ~VDC_HANDSHAKE;
776 	DMSG(vdcp, 0, "initialized=%x\n", vdcp->initialized);
777 
778 	return (status);
779 }
780 
781 static int
782 vdc_create_device_nodes_efi(vdc_t *vdc)
783 {
784 	ddi_remove_minor_node(vdc->dip, "h");
785 	ddi_remove_minor_node(vdc->dip, "h,raw");
786 
787 	if (ddi_create_minor_node(vdc->dip, "wd", S_IFBLK,
788 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
789 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
790 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'wd'",
791 		    vdc->instance);
792 		return (EIO);
793 	}
794 
795 	/* if any device node is created we set this flag */
796 	vdc->initialized |= VDC_MINOR;
797 
798 	if (ddi_create_minor_node(vdc->dip, "wd,raw", S_IFCHR,
799 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
800 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
801 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'wd,raw'",
802 		    vdc->instance);
803 		return (EIO);
804 	}
805 
806 	return (0);
807 }
808 
809 static int
810 vdc_create_device_nodes_vtoc(vdc_t *vdc)
811 {
812 	ddi_remove_minor_node(vdc->dip, "wd");
813 	ddi_remove_minor_node(vdc->dip, "wd,raw");
814 
815 	if (ddi_create_minor_node(vdc->dip, "h", S_IFBLK,
816 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
817 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
818 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'h'",
819 		    vdc->instance);
820 		return (EIO);
821 	}
822 
823 	/* if any device node is created we set this flag */
824 	vdc->initialized |= VDC_MINOR;
825 
826 	if (ddi_create_minor_node(vdc->dip, "h,raw", S_IFCHR,
827 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
828 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
829 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'h,raw'",
830 		    vdc->instance);
831 		return (EIO);
832 	}
833 
834 	return (0);
835 }
836 
837 /*
838  * Function:
839  *	vdc_create_device_nodes
840  *
841  * Description:
842  *	This function creates the block and character device nodes under
843  *	/devices along with the node properties. It is called as part of
844  *	the attach(9E) of the instance during the handshake with vds after
845  *	vds has sent the attributes to vdc.
846  *
847  *	If the device is of type VD_DISK_TYPE_SLICE then the minor node
848  *	of 2 is used in keeping with the Solaris convention that slice 2
849  *	refers to a whole disk. Slices start at 'a'
850  *
851  * Parameters:
852  *	vdc 		- soft state pointer
853  *
854  * Return Values
855  *	0		- Success
856  *	EIO		- Failed to create node
857  *	EINVAL		- Unknown type of disk exported
858  */
859 static int
860 vdc_create_device_nodes(vdc_t *vdc)
861 {
862 	char		name[sizeof ("s,raw")];
863 	dev_info_t	*dip = NULL;
864 	int		instance, status;
865 	int		num_slices = 1;
866 	int		i;
867 
868 	ASSERT(vdc != NULL);
869 
870 	instance = vdc->instance;
871 	dip = vdc->dip;
872 
873 	switch (vdc->vdisk_type) {
874 	case VD_DISK_TYPE_DISK:
875 		num_slices = V_NUMPAR;
876 		break;
877 	case VD_DISK_TYPE_SLICE:
878 		num_slices = 1;
879 		break;
880 	case VD_DISK_TYPE_UNK:
881 	default:
882 		return (EINVAL);
883 	}
884 
885 	/*
886 	 * Minor nodes are different for EFI disks: EFI disks do not have
887 	 * a minor node 'g' for the minor number corresponding to slice
888 	 * VD_EFI_WD_SLICE (slice 7) instead they have a minor node 'wd'
889 	 * representing the whole disk.
890 	 */
891 	for (i = 0; i < num_slices; i++) {
892 
893 		if (i == VD_EFI_WD_SLICE) {
894 			if (vdc->vdisk_label == VD_DISK_LABEL_EFI)
895 				status = vdc_create_device_nodes_efi(vdc);
896 			else
897 				status = vdc_create_device_nodes_vtoc(vdc);
898 			if (status != 0)
899 				return (status);
900 			continue;
901 		}
902 
903 		(void) snprintf(name, sizeof (name), "%c", 'a' + i);
904 		if (ddi_create_minor_node(dip, name, S_IFBLK,
905 		    VD_MAKE_DEV(instance, i), DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
906 			cmn_err(CE_NOTE, "[%d] Couldn't add block node '%s'",
907 			    instance, name);
908 			return (EIO);
909 		}
910 
911 		/* if any device node is created we set this flag */
912 		vdc->initialized |= VDC_MINOR;
913 
914 		(void) snprintf(name, sizeof (name), "%c%s", 'a' + i, ",raw");
915 
916 		if (ddi_create_minor_node(dip, name, S_IFCHR,
917 		    VD_MAKE_DEV(instance, i), DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
918 			cmn_err(CE_NOTE, "[%d] Couldn't add raw node '%s'",
919 			    instance, name);
920 			return (EIO);
921 		}
922 	}
923 
924 	return (0);
925 }
926 
927 /*
928  * Function:
929  *	vdc_create_device_nodes_props
930  *
931  * Description:
932  *	This function creates the block and character device nodes under
933  *	/devices along with the node properties. It is called as part of
934  *	the attach(9E) of the instance during the handshake with vds after
935  *	vds has sent the attributes to vdc.
936  *
937  * Parameters:
938  *	vdc 		- soft state pointer
939  *
940  * Return Values
941  *	0		- Success
942  *	EIO		- Failed to create device node property
943  *	EINVAL		- Unknown type of disk exported
944  */
945 static int
946 vdc_create_device_nodes_props(vdc_t *vdc)
947 {
948 	dev_info_t	*dip = NULL;
949 	int		instance;
950 	int		num_slices = 1;
951 	int64_t		size = 0;
952 	dev_t		dev;
953 	int		rv;
954 	int		i;
955 
956 	ASSERT(vdc != NULL);
957 	ASSERT(vdc->vtoc != NULL);
958 
959 	instance = vdc->instance;
960 	dip = vdc->dip;
961 
962 	switch (vdc->vdisk_type) {
963 	case VD_DISK_TYPE_DISK:
964 		num_slices = V_NUMPAR;
965 		break;
966 	case VD_DISK_TYPE_SLICE:
967 		num_slices = 1;
968 		break;
969 	case VD_DISK_TYPE_UNK:
970 	default:
971 		return (EINVAL);
972 	}
973 
974 	if (vdc->vdisk_label == VD_DISK_LABEL_UNK) {
975 		/* remove all properties */
976 		for (i = 0; i < num_slices; i++) {
977 			dev = makedevice(ddi_driver_major(dip),
978 			    VD_MAKE_DEV(instance, i));
979 			(void) ddi_prop_remove(dev, dip, VDC_SIZE_PROP_NAME);
980 			(void) ddi_prop_remove(dev, dip, VDC_NBLOCKS_PROP_NAME);
981 		}
982 		return (0);
983 	}
984 
985 	for (i = 0; i < num_slices; i++) {
986 		dev = makedevice(ddi_driver_major(dip),
987 		    VD_MAKE_DEV(instance, i));
988 
989 		size = vdc->vtoc->v_part[i].p_size * vdc->vtoc->v_sectorsz;
990 		DMSG(vdc, 0, "[%d] sz %ld (%ld Mb)  p_size %lx\n",
991 		    instance, size, size / (1024 * 1024),
992 		    vdc->vtoc->v_part[i].p_size);
993 
994 		rv = ddi_prop_update_int64(dev, dip, VDC_SIZE_PROP_NAME, size);
995 		if (rv != DDI_PROP_SUCCESS) {
996 			cmn_err(CE_NOTE, "[%d] Couldn't add '%s' prop of [%ld]",
997 			    instance, VDC_SIZE_PROP_NAME, size);
998 			return (EIO);
999 		}
1000 
1001 		rv = ddi_prop_update_int64(dev, dip, VDC_NBLOCKS_PROP_NAME,
1002 		    lbtodb(size));
1003 		if (rv != DDI_PROP_SUCCESS) {
1004 			cmn_err(CE_NOTE, "[%d] Couldn't add '%s' prop [%llu]",
1005 			    instance, VDC_NBLOCKS_PROP_NAME, lbtodb(size));
1006 			return (EIO);
1007 		}
1008 	}
1009 
1010 	return (0);
1011 }
1012 
1013 /*
1014  * Function:
1015  *	vdc_is_opened
1016  *
1017  * Description:
1018  *	This function checks if any slice of a given virtual disk is
1019  *	currently opened.
1020  *
1021  * Parameters:
1022  *	vdc 		- soft state pointer
1023  *
1024  * Return Values
1025  *	B_TRUE		- at least one slice is opened.
1026  *	B_FALSE		- no slice is opened.
1027  */
1028 static boolean_t
1029 vdc_is_opened(vdc_t *vdc)
1030 {
1031 	int i, nslices;
1032 
1033 	switch (vdc->vdisk_type) {
1034 	case VD_DISK_TYPE_DISK:
1035 		nslices = V_NUMPAR;
1036 		break;
1037 	case VD_DISK_TYPE_SLICE:
1038 		nslices = 1;
1039 		break;
1040 	case VD_DISK_TYPE_UNK:
1041 	default:
1042 		ASSERT(0);
1043 	}
1044 
1045 	/* check if there's any layered open */
1046 	for (i = 0; i < nslices; i++) {
1047 		if (vdc->open_lyr[i] > 0)
1048 			return (B_TRUE);
1049 	}
1050 
1051 	/* check if there is any other kind of open */
1052 	for (i = 0; i < OTYPCNT; i++) {
1053 		if (vdc->open[i] != 0)
1054 			return (B_TRUE);
1055 	}
1056 
1057 	return (B_FALSE);
1058 }
1059 
1060 static int
1061 vdc_mark_opened(vdc_t *vdc, int slice, int flag, int otyp)
1062 {
1063 	uint8_t slicemask;
1064 	int i;
1065 
1066 	ASSERT(otyp < OTYPCNT);
1067 	ASSERT(slice < V_NUMPAR);
1068 	ASSERT(MUTEX_HELD(&vdc->lock));
1069 
1070 	slicemask = 1 << slice;
1071 
1072 	/* check if slice is already exclusively opened */
1073 	if (vdc->open_excl & slicemask)
1074 		return (EBUSY);
1075 
1076 	/* if open exclusive, check if slice is already opened */
1077 	if (flag & FEXCL) {
1078 		if (vdc->open_lyr[slice] > 0)
1079 			return (EBUSY);
1080 		for (i = 0; i < OTYPCNT; i++) {
1081 			if (vdc->open[i] & slicemask)
1082 				return (EBUSY);
1083 		}
1084 		vdc->open_excl |= slicemask;
1085 	}
1086 
1087 	/* mark slice as opened */
1088 	if (otyp == OTYP_LYR) {
1089 		vdc->open_lyr[slice]++;
1090 	} else {
1091 		vdc->open[otyp] |= slicemask;
1092 	}
1093 
1094 	return (0);
1095 }
1096 
1097 static void
1098 vdc_mark_closed(vdc_t *vdc, int slice, int flag, int otyp)
1099 {
1100 	uint8_t slicemask;
1101 
1102 	ASSERT(otyp < OTYPCNT);
1103 	ASSERT(slice < V_NUMPAR);
1104 	ASSERT(MUTEX_HELD(&vdc->lock));
1105 
1106 	slicemask = 1 << slice;
1107 
1108 	if (otyp == OTYP_LYR) {
1109 		ASSERT(vdc->open_lyr[slice] > 0);
1110 		vdc->open_lyr[slice]--;
1111 	} else {
1112 		vdc->open[otyp] &= ~slicemask;
1113 	}
1114 
1115 	if (flag & FEXCL)
1116 		vdc->open_excl &= ~slicemask;
1117 }
1118 
1119 static int
1120 vdc_open(dev_t *dev, int flag, int otyp, cred_t *cred)
1121 {
1122 	_NOTE(ARGUNUSED(cred))
1123 
1124 	int	instance;
1125 	int	slice, status = 0;
1126 	vdc_t	*vdc;
1127 
1128 	ASSERT(dev != NULL);
1129 	instance = VDCUNIT(*dev);
1130 
1131 	if (otyp >= OTYPCNT)
1132 		return (EINVAL);
1133 
1134 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1135 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1136 		return (ENXIO);
1137 	}
1138 
1139 	DMSG(vdc, 0, "minor = %d flag = %x, otyp = %x\n",
1140 	    getminor(*dev), flag, otyp);
1141 
1142 	slice = VDCPART(*dev);
1143 
1144 	mutex_enter(&vdc->lock);
1145 
1146 	status = vdc_mark_opened(vdc, slice, flag, otyp);
1147 
1148 	if (status != 0) {
1149 		mutex_exit(&vdc->lock);
1150 		return (status);
1151 	}
1152 
1153 	if (flag & (FNDELAY | FNONBLOCK)) {
1154 
1155 		/* don't resubmit a validate request if there's already one */
1156 		if (vdc->validate_pending > 0) {
1157 			mutex_exit(&vdc->lock);
1158 			return (0);
1159 		}
1160 
1161 		/* call vdc_validate() asynchronously to avoid blocking */
1162 		if (taskq_dispatch(system_taskq, vdc_validate_task,
1163 		    (void *)vdc, TQ_NOSLEEP) == NULL) {
1164 			vdc_mark_closed(vdc, slice, flag, otyp);
1165 			mutex_exit(&vdc->lock);
1166 			return (ENXIO);
1167 		}
1168 
1169 		vdc->validate_pending++;
1170 		mutex_exit(&vdc->lock);
1171 		return (0);
1172 	}
1173 
1174 	mutex_exit(&vdc->lock);
1175 
1176 	vdc_validate(vdc);
1177 
1178 	mutex_enter(&vdc->lock);
1179 
1180 	if (vdc->vdisk_label == VD_DISK_LABEL_UNK ||
1181 	    vdc->vtoc->v_part[slice].p_size == 0) {
1182 		vdc_mark_closed(vdc, slice, flag, otyp);
1183 		status = EIO;
1184 	}
1185 
1186 	mutex_exit(&vdc->lock);
1187 
1188 	return (status);
1189 }
1190 
1191 static int
1192 vdc_close(dev_t dev, int flag, int otyp, cred_t *cred)
1193 {
1194 	_NOTE(ARGUNUSED(cred))
1195 
1196 	int	instance;
1197 	int	slice;
1198 	vdc_t	*vdc;
1199 
1200 	instance = VDCUNIT(dev);
1201 
1202 	if (otyp >= OTYPCNT)
1203 		return (EINVAL);
1204 
1205 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1206 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1207 		return (ENXIO);
1208 	}
1209 
1210 	DMSG(vdc, 0, "[%d] flag = %x, otyp = %x\n", instance, flag, otyp);
1211 
1212 	slice = VDCPART(dev);
1213 
1214 	mutex_enter(&vdc->lock);
1215 	vdc_mark_closed(vdc, slice, flag, otyp);
1216 	mutex_exit(&vdc->lock);
1217 
1218 	return (0);
1219 }
1220 
1221 static int
1222 vdc_ioctl(dev_t dev, int cmd, intptr_t arg, int mode, cred_t *credp, int *rvalp)
1223 {
1224 	_NOTE(ARGUNUSED(credp))
1225 	_NOTE(ARGUNUSED(rvalp))
1226 
1227 	return (vd_process_ioctl(dev, cmd, (caddr_t)arg, mode));
1228 }
1229 
1230 static int
1231 vdc_print(dev_t dev, char *str)
1232 {
1233 	cmn_err(CE_NOTE, "vdc%d:  %s", VDCUNIT(dev), str);
1234 	return (0);
1235 }
1236 
1237 static int
1238 vdc_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblk)
1239 {
1240 	int	rv;
1241 	size_t	nbytes = nblk * DEV_BSIZE;
1242 	int	instance = VDCUNIT(dev);
1243 	vdc_t	*vdc = NULL;
1244 
1245 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1246 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1247 		return (ENXIO);
1248 	}
1249 
1250 	DMSG(vdc, 2, "[%d] dump %ld bytes at block 0x%lx : addr=0x%p\n",
1251 	    instance, nbytes, blkno, (void *)addr);
1252 	rv = vdc_send_request(vdc, VD_OP_BWRITE, addr, nbytes,
1253 	    VDCPART(dev), blkno, CB_STRATEGY, 0, VIO_write_dir);
1254 	if (rv) {
1255 		DMSG(vdc, 0, "Failed to do a disk dump (err=%d)\n", rv);
1256 		return (rv);
1257 	}
1258 
1259 	if (ddi_in_panic())
1260 		(void) vdc_drain_response(vdc);
1261 
1262 	DMSG(vdc, 0, "[%d] End\n", instance);
1263 
1264 	return (0);
1265 }
1266 
1267 /* -------------------------------------------------------------------------- */
1268 
1269 /*
1270  * Disk access routines
1271  *
1272  */
1273 
1274 /*
1275  * vdc_strategy()
1276  *
1277  * Return Value:
1278  *	0:	As per strategy(9E), the strategy() function must return 0
1279  *		[ bioerror(9f) sets b_flags to the proper error code ]
1280  */
1281 static int
1282 vdc_strategy(struct buf *buf)
1283 {
1284 	int	rv = -1;
1285 	vdc_t	*vdc = NULL;
1286 	int	instance = VDCUNIT(buf->b_edev);
1287 	int	op = (buf->b_flags & B_READ) ? VD_OP_BREAD : VD_OP_BWRITE;
1288 	int	slice;
1289 
1290 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1291 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1292 		bioerror(buf, ENXIO);
1293 		biodone(buf);
1294 		return (0);
1295 	}
1296 
1297 	DMSG(vdc, 2, "[%d] %s %ld bytes at block %llx : b_addr=0x%p\n",
1298 	    instance, (buf->b_flags & B_READ) ? "Read" : "Write",
1299 	    buf->b_bcount, buf->b_lblkno, (void *)buf->b_un.b_addr);
1300 	DTRACE_IO2(vstart, buf_t *, buf, vdc_t *, vdc);
1301 
1302 	bp_mapin(buf);
1303 
1304 	if ((long)buf->b_private == VD_SLICE_NONE) {
1305 		/* I/O using an absolute disk offset */
1306 		slice = VD_SLICE_NONE;
1307 	} else {
1308 		slice = VDCPART(buf->b_edev);
1309 	}
1310 
1311 	rv = vdc_send_request(vdc, op, (caddr_t)buf->b_un.b_addr,
1312 	    buf->b_bcount, slice, buf->b_lblkno,
1313 	    CB_STRATEGY, buf, (op == VD_OP_BREAD) ? VIO_read_dir :
1314 	    VIO_write_dir);
1315 
1316 	/*
1317 	 * If the request was successfully sent, the strategy call returns and
1318 	 * the ACK handler calls the bioxxx functions when the vDisk server is
1319 	 * done.
1320 	 */
1321 	if (rv) {
1322 		DMSG(vdc, 0, "Failed to read/write (err=%d)\n", rv);
1323 		bioerror(buf, rv);
1324 		biodone(buf);
1325 	}
1326 
1327 	return (0);
1328 }
1329 
1330 /*
1331  * Function:
1332  *	vdc_min
1333  *
1334  * Description:
1335  *	Routine to limit the size of a data transfer. Used in
1336  *	conjunction with physio(9F).
1337  *
1338  * Arguments:
1339  *	bp - pointer to the indicated buf(9S) struct.
1340  *
1341  */
1342 static void
1343 vdc_min(struct buf *bufp)
1344 {
1345 	vdc_t	*vdc = NULL;
1346 	int	instance = VDCUNIT(bufp->b_edev);
1347 
1348 	vdc = ddi_get_soft_state(vdc_state, instance);
1349 	VERIFY(vdc != NULL);
1350 
1351 	if (bufp->b_bcount > (vdc->max_xfer_sz * vdc->block_size)) {
1352 		bufp->b_bcount = vdc->max_xfer_sz * vdc->block_size;
1353 	}
1354 }
1355 
1356 static int
1357 vdc_read(dev_t dev, struct uio *uio, cred_t *cred)
1358 {
1359 	_NOTE(ARGUNUSED(cred))
1360 
1361 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1362 	return (physio(vdc_strategy, NULL, dev, B_READ, vdc_min, uio));
1363 }
1364 
1365 static int
1366 vdc_write(dev_t dev, struct uio *uio, cred_t *cred)
1367 {
1368 	_NOTE(ARGUNUSED(cred))
1369 
1370 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1371 	return (physio(vdc_strategy, NULL, dev, B_WRITE, vdc_min, uio));
1372 }
1373 
1374 static int
1375 vdc_aread(dev_t dev, struct aio_req *aio, cred_t *cred)
1376 {
1377 	_NOTE(ARGUNUSED(cred))
1378 
1379 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1380 	return (aphysio(vdc_strategy, anocancel, dev, B_READ, vdc_min, aio));
1381 }
1382 
1383 static int
1384 vdc_awrite(dev_t dev, struct aio_req *aio, cred_t *cred)
1385 {
1386 	_NOTE(ARGUNUSED(cred))
1387 
1388 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1389 	return (aphysio(vdc_strategy, anocancel, dev, B_WRITE, vdc_min, aio));
1390 }
1391 
1392 
1393 /* -------------------------------------------------------------------------- */
1394 
1395 /*
1396  * Handshake support
1397  */
1398 
1399 
1400 /*
1401  * Function:
1402  *	vdc_init_ver_negotiation()
1403  *
1404  * Description:
1405  *
1406  * Arguments:
1407  *	vdc	- soft state pointer for this instance of the device driver.
1408  *
1409  * Return Code:
1410  *	0	- Success
1411  */
1412 static int
1413 vdc_init_ver_negotiation(vdc_t *vdc, vio_ver_t ver)
1414 {
1415 	vio_ver_msg_t	pkt;
1416 	size_t		msglen = sizeof (pkt);
1417 	int		status = -1;
1418 
1419 	ASSERT(vdc != NULL);
1420 	ASSERT(mutex_owned(&vdc->lock));
1421 
1422 	DMSG(vdc, 0, "[%d] Entered.\n", vdc->instance);
1423 
1424 	/*
1425 	 * set the Session ID to a unique value
1426 	 * (the lower 32 bits of the clock tick)
1427 	 */
1428 	vdc->session_id = ((uint32_t)gettick() & 0xffffffff);
1429 	DMSG(vdc, 0, "[%d] Set SID to 0x%lx\n", vdc->instance, vdc->session_id);
1430 
1431 	pkt.tag.vio_msgtype = VIO_TYPE_CTRL;
1432 	pkt.tag.vio_subtype = VIO_SUBTYPE_INFO;
1433 	pkt.tag.vio_subtype_env = VIO_VER_INFO;
1434 	pkt.tag.vio_sid = vdc->session_id;
1435 	pkt.dev_class = VDEV_DISK;
1436 	pkt.ver_major = ver.major;
1437 	pkt.ver_minor = ver.minor;
1438 
1439 	status = vdc_send(vdc, (caddr_t)&pkt, &msglen);
1440 	DMSG(vdc, 0, "[%d] Ver info sent (status = %d)\n",
1441 	    vdc->instance, status);
1442 	if ((status != 0) || (msglen != sizeof (vio_ver_msg_t))) {
1443 		DMSG(vdc, 0, "[%d] Failed to send Ver negotiation info: "
1444 		    "id(%lx) rv(%d) size(%ld)", vdc->instance, vdc->ldc_handle,
1445 		    status, msglen);
1446 		if (msglen != sizeof (vio_ver_msg_t))
1447 			status = ENOMSG;
1448 	}
1449 
1450 	return (status);
1451 }
1452 
1453 /*
1454  * Function:
1455  *	vdc_ver_negotiation()
1456  *
1457  * Description:
1458  *
1459  * Arguments:
1460  *	vdcp	- soft state pointer for this instance of the device driver.
1461  *
1462  * Return Code:
1463  *	0	- Success
1464  */
1465 static int
1466 vdc_ver_negotiation(vdc_t *vdcp)
1467 {
1468 	vio_msg_t vio_msg;
1469 	int status;
1470 
1471 	if (status = vdc_init_ver_negotiation(vdcp, vdc_version[0]))
1472 		return (status);
1473 
1474 	/* release lock and wait for response */
1475 	mutex_exit(&vdcp->lock);
1476 	status = vdc_wait_for_response(vdcp, &vio_msg);
1477 	mutex_enter(&vdcp->lock);
1478 	if (status) {
1479 		DMSG(vdcp, 0,
1480 		    "[%d] Failed waiting for Ver negotiation response, rv(%d)",
1481 		    vdcp->instance, status);
1482 		return (status);
1483 	}
1484 
1485 	/* check type and sub_type ... */
1486 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1487 	    vio_msg.tag.vio_subtype == VIO_SUBTYPE_INFO) {
1488 		DMSG(vdcp, 0, "[%d] Invalid ver negotiation response\n",
1489 		    vdcp->instance);
1490 		return (EPROTO);
1491 	}
1492 
1493 	return (vdc_handle_ver_msg(vdcp, (vio_ver_msg_t *)&vio_msg));
1494 }
1495 
1496 /*
1497  * Function:
1498  *	vdc_init_attr_negotiation()
1499  *
1500  * Description:
1501  *
1502  * Arguments:
1503  *	vdc	- soft state pointer for this instance of the device driver.
1504  *
1505  * Return Code:
1506  *	0	- Success
1507  */
1508 static int
1509 vdc_init_attr_negotiation(vdc_t *vdc)
1510 {
1511 	vd_attr_msg_t	pkt;
1512 	size_t		msglen = sizeof (pkt);
1513 	int		status;
1514 
1515 	ASSERT(vdc != NULL);
1516 	ASSERT(mutex_owned(&vdc->lock));
1517 
1518 	DMSG(vdc, 0, "[%d] entered\n", vdc->instance);
1519 
1520 	/* fill in tag */
1521 	pkt.tag.vio_msgtype = VIO_TYPE_CTRL;
1522 	pkt.tag.vio_subtype = VIO_SUBTYPE_INFO;
1523 	pkt.tag.vio_subtype_env = VIO_ATTR_INFO;
1524 	pkt.tag.vio_sid = vdc->session_id;
1525 	/* fill in payload */
1526 	pkt.max_xfer_sz = vdc->max_xfer_sz;
1527 	pkt.vdisk_block_size = vdc->block_size;
1528 	pkt.xfer_mode = VIO_DRING_MODE;
1529 	pkt.operations = 0;	/* server will set bits of valid operations */
1530 	pkt.vdisk_type = 0;	/* server will set to valid device type */
1531 	pkt.vdisk_size = 0;	/* server will set to valid size */
1532 
1533 	status = vdc_send(vdc, (caddr_t)&pkt, &msglen);
1534 	DMSG(vdc, 0, "Attr info sent (status = %d)\n", status);
1535 
1536 	if ((status != 0) || (msglen != sizeof (vio_ver_msg_t))) {
1537 		DMSG(vdc, 0, "[%d] Failed to send Attr negotiation info: "
1538 		    "id(%lx) rv(%d) size(%ld)", vdc->instance, vdc->ldc_handle,
1539 		    status, msglen);
1540 		if (msglen != sizeof (vio_ver_msg_t))
1541 			status = ENOMSG;
1542 	}
1543 
1544 	return (status);
1545 }
1546 
1547 /*
1548  * Function:
1549  *	vdc_attr_negotiation()
1550  *
1551  * Description:
1552  *
1553  * Arguments:
1554  *	vdc	- soft state pointer for this instance of the device driver.
1555  *
1556  * Return Code:
1557  *	0	- Success
1558  */
1559 static int
1560 vdc_attr_negotiation(vdc_t *vdcp)
1561 {
1562 	int status;
1563 	vio_msg_t vio_msg;
1564 
1565 	if (status = vdc_init_attr_negotiation(vdcp))
1566 		return (status);
1567 
1568 	/* release lock and wait for response */
1569 	mutex_exit(&vdcp->lock);
1570 	status = vdc_wait_for_response(vdcp, &vio_msg);
1571 	mutex_enter(&vdcp->lock);
1572 	if (status) {
1573 		DMSG(vdcp, 0,
1574 		    "[%d] Failed waiting for Attr negotiation response, rv(%d)",
1575 		    vdcp->instance, status);
1576 		return (status);
1577 	}
1578 
1579 	/* check type and sub_type ... */
1580 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1581 	    vio_msg.tag.vio_subtype == VIO_SUBTYPE_INFO) {
1582 		DMSG(vdcp, 0, "[%d] Invalid attr negotiation response\n",
1583 		    vdcp->instance);
1584 		return (EPROTO);
1585 	}
1586 
1587 	return (vdc_handle_attr_msg(vdcp, (vd_attr_msg_t *)&vio_msg));
1588 }
1589 
1590 
1591 /*
1592  * Function:
1593  *	vdc_init_dring_negotiate()
1594  *
1595  * Description:
1596  *
1597  * Arguments:
1598  *	vdc	- soft state pointer for this instance of the device driver.
1599  *
1600  * Return Code:
1601  *	0	- Success
1602  */
1603 static int
1604 vdc_init_dring_negotiate(vdc_t *vdc)
1605 {
1606 	vio_dring_reg_msg_t	pkt;
1607 	size_t			msglen = sizeof (pkt);
1608 	int			status = -1;
1609 	int			retry;
1610 	int			nretries = 10;
1611 
1612 	ASSERT(vdc != NULL);
1613 	ASSERT(mutex_owned(&vdc->lock));
1614 
1615 	for (retry = 0; retry < nretries; retry++) {
1616 		status = vdc_init_descriptor_ring(vdc);
1617 		if (status != EAGAIN)
1618 			break;
1619 		drv_usecwait(vdc_min_timeout_ldc);
1620 	}
1621 
1622 	if (status != 0) {
1623 		DMSG(vdc, 0, "[%d] Failed to init DRing (status = %d)\n",
1624 		    vdc->instance, status);
1625 		return (status);
1626 	}
1627 
1628 	DMSG(vdc, 0, "[%d] Init of descriptor ring completed (status = %d)\n",
1629 	    vdc->instance, status);
1630 
1631 	/* fill in tag */
1632 	pkt.tag.vio_msgtype = VIO_TYPE_CTRL;
1633 	pkt.tag.vio_subtype = VIO_SUBTYPE_INFO;
1634 	pkt.tag.vio_subtype_env = VIO_DRING_REG;
1635 	pkt.tag.vio_sid = vdc->session_id;
1636 	/* fill in payload */
1637 	pkt.dring_ident = 0;
1638 	pkt.num_descriptors = vdc->dring_len;
1639 	pkt.descriptor_size = vdc->dring_entry_size;
1640 	pkt.options = (VIO_TX_DRING | VIO_RX_DRING);
1641 	pkt.ncookies = vdc->dring_cookie_count;
1642 	pkt.cookie[0] = vdc->dring_cookie[0];	/* for now just one cookie */
1643 
1644 	status = vdc_send(vdc, (caddr_t)&pkt, &msglen);
1645 	if (status != 0) {
1646 		DMSG(vdc, 0, "[%d] Failed to register DRing (err = %d)",
1647 		    vdc->instance, status);
1648 	}
1649 
1650 	return (status);
1651 }
1652 
1653 
1654 /*
1655  * Function:
1656  *	vdc_dring_negotiation()
1657  *
1658  * Description:
1659  *
1660  * Arguments:
1661  *	vdc	- soft state pointer for this instance of the device driver.
1662  *
1663  * Return Code:
1664  *	0	- Success
1665  */
1666 static int
1667 vdc_dring_negotiation(vdc_t *vdcp)
1668 {
1669 	int status;
1670 	vio_msg_t vio_msg;
1671 
1672 	if (status = vdc_init_dring_negotiate(vdcp))
1673 		return (status);
1674 
1675 	/* release lock and wait for response */
1676 	mutex_exit(&vdcp->lock);
1677 	status = vdc_wait_for_response(vdcp, &vio_msg);
1678 	mutex_enter(&vdcp->lock);
1679 	if (status) {
1680 		DMSG(vdcp, 0,
1681 		    "[%d] Failed waiting for Dring negotiation response,"
1682 		    " rv(%d)", vdcp->instance, status);
1683 		return (status);
1684 	}
1685 
1686 	/* check type and sub_type ... */
1687 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1688 	    vio_msg.tag.vio_subtype == VIO_SUBTYPE_INFO) {
1689 		DMSG(vdcp, 0, "[%d] Invalid Dring negotiation response\n",
1690 		    vdcp->instance);
1691 		return (EPROTO);
1692 	}
1693 
1694 	return (vdc_handle_dring_reg_msg(vdcp,
1695 	    (vio_dring_reg_msg_t *)&vio_msg));
1696 }
1697 
1698 
1699 /*
1700  * Function:
1701  *	vdc_send_rdx()
1702  *
1703  * Description:
1704  *
1705  * Arguments:
1706  *	vdc	- soft state pointer for this instance of the device driver.
1707  *
1708  * Return Code:
1709  *	0	- Success
1710  */
1711 static int
1712 vdc_send_rdx(vdc_t *vdcp)
1713 {
1714 	vio_msg_t	msg;
1715 	size_t		msglen = sizeof (vio_msg_t);
1716 	int		status;
1717 
1718 	/*
1719 	 * Send an RDX message to vds to indicate we are ready
1720 	 * to send data
1721 	 */
1722 	msg.tag.vio_msgtype = VIO_TYPE_CTRL;
1723 	msg.tag.vio_subtype = VIO_SUBTYPE_INFO;
1724 	msg.tag.vio_subtype_env = VIO_RDX;
1725 	msg.tag.vio_sid = vdcp->session_id;
1726 	status = vdc_send(vdcp, (caddr_t)&msg, &msglen);
1727 	if (status != 0) {
1728 		DMSG(vdcp, 0, "[%d] Failed to send RDX message (%d)",
1729 		    vdcp->instance, status);
1730 	}
1731 
1732 	return (status);
1733 }
1734 
1735 /*
1736  * Function:
1737  *	vdc_handle_rdx()
1738  *
1739  * Description:
1740  *
1741  * Arguments:
1742  *	vdc	- soft state pointer for this instance of the device driver.
1743  *	msgp	- received msg
1744  *
1745  * Return Code:
1746  *	0	- Success
1747  */
1748 static int
1749 vdc_handle_rdx(vdc_t *vdcp, vio_rdx_msg_t *msgp)
1750 {
1751 	_NOTE(ARGUNUSED(vdcp))
1752 	_NOTE(ARGUNUSED(msgp))
1753 
1754 	ASSERT(msgp->tag.vio_msgtype == VIO_TYPE_CTRL);
1755 	ASSERT(msgp->tag.vio_subtype == VIO_SUBTYPE_ACK);
1756 	ASSERT(msgp->tag.vio_subtype_env == VIO_RDX);
1757 
1758 	DMSG(vdcp, 1, "[%d] Got an RDX msg", vdcp->instance);
1759 
1760 	return (0);
1761 }
1762 
1763 /*
1764  * Function:
1765  *	vdc_rdx_exchange()
1766  *
1767  * Description:
1768  *
1769  * Arguments:
1770  *	vdc	- soft state pointer for this instance of the device driver.
1771  *
1772  * Return Code:
1773  *	0	- Success
1774  */
1775 static int
1776 vdc_rdx_exchange(vdc_t *vdcp)
1777 {
1778 	int status;
1779 	vio_msg_t vio_msg;
1780 
1781 	if (status = vdc_send_rdx(vdcp))
1782 		return (status);
1783 
1784 	/* release lock and wait for response */
1785 	mutex_exit(&vdcp->lock);
1786 	status = vdc_wait_for_response(vdcp, &vio_msg);
1787 	mutex_enter(&vdcp->lock);
1788 	if (status) {
1789 		DMSG(vdcp, 0, "[%d] Failed waiting for RDX response, rv(%d)",
1790 		    vdcp->instance, status);
1791 		return (status);
1792 	}
1793 
1794 	/* check type and sub_type ... */
1795 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1796 	    vio_msg.tag.vio_subtype != VIO_SUBTYPE_ACK) {
1797 		DMSG(vdcp, 0, "[%d] Invalid RDX response\n", vdcp->instance);
1798 		return (EPROTO);
1799 	}
1800 
1801 	return (vdc_handle_rdx(vdcp, (vio_rdx_msg_t *)&vio_msg));
1802 }
1803 
1804 
1805 /* -------------------------------------------------------------------------- */
1806 
1807 /*
1808  * LDC helper routines
1809  */
1810 
1811 static int
1812 vdc_recv(vdc_t *vdc, vio_msg_t *msgp, size_t *nbytesp)
1813 {
1814 	int		status;
1815 	boolean_t	q_has_pkts = B_FALSE;
1816 	int		delay_time;
1817 	size_t		len;
1818 
1819 	mutex_enter(&vdc->read_lock);
1820 
1821 	if (vdc->read_state == VDC_READ_IDLE)
1822 		vdc->read_state = VDC_READ_WAITING;
1823 
1824 	while (vdc->read_state != VDC_READ_PENDING) {
1825 
1826 		/* detect if the connection has been reset */
1827 		if (vdc->read_state == VDC_READ_RESET) {
1828 			status = ECONNRESET;
1829 			goto done;
1830 		}
1831 
1832 		cv_wait(&vdc->read_cv, &vdc->read_lock);
1833 	}
1834 
1835 	/*
1836 	 * Until we get a blocking ldc read we have to retry
1837 	 * until the entire LDC message has arrived before
1838 	 * ldc_read() will succeed. Note we also bail out if
1839 	 * the channel is reset or goes away.
1840 	 */
1841 	delay_time = vdc_ldc_read_init_delay;
1842 loop:
1843 	len = *nbytesp;
1844 	status = ldc_read(vdc->ldc_handle, (caddr_t)msgp, &len);
1845 	switch (status) {
1846 	case EAGAIN:
1847 		delay_time *= 2;
1848 		if (delay_time >= vdc_ldc_read_max_delay)
1849 			delay_time = vdc_ldc_read_max_delay;
1850 		delay(delay_time);
1851 		goto loop;
1852 
1853 	case 0:
1854 		if (len == 0) {
1855 			DMSG(vdc, 0, "[%d] ldc_read returned 0 bytes with "
1856 			    "no error!\n", vdc->instance);
1857 			goto loop;
1858 		}
1859 
1860 		*nbytesp = len;
1861 
1862 		/*
1863 		 * If there are pending messages, leave the
1864 		 * read state as pending. Otherwise, set the state
1865 		 * back to idle.
1866 		 */
1867 		status = ldc_chkq(vdc->ldc_handle, &q_has_pkts);
1868 		if (status == 0 && !q_has_pkts)
1869 			vdc->read_state = VDC_READ_IDLE;
1870 
1871 		break;
1872 	default:
1873 		DMSG(vdc, 0, "ldc_read returned %d\n", status);
1874 		break;
1875 	}
1876 
1877 done:
1878 	mutex_exit(&vdc->read_lock);
1879 
1880 	return (status);
1881 }
1882 
1883 
1884 
1885 #ifdef DEBUG
1886 void
1887 vdc_decode_tag(vdc_t *vdcp, vio_msg_t *msg)
1888 {
1889 	char *ms, *ss, *ses;
1890 	switch (msg->tag.vio_msgtype) {
1891 #define	Q(_s)	case _s : ms = #_s; break;
1892 	Q(VIO_TYPE_CTRL)
1893 	Q(VIO_TYPE_DATA)
1894 	Q(VIO_TYPE_ERR)
1895 #undef Q
1896 	default: ms = "unknown"; break;
1897 	}
1898 
1899 	switch (msg->tag.vio_subtype) {
1900 #define	Q(_s)	case _s : ss = #_s; break;
1901 	Q(VIO_SUBTYPE_INFO)
1902 	Q(VIO_SUBTYPE_ACK)
1903 	Q(VIO_SUBTYPE_NACK)
1904 #undef Q
1905 	default: ss = "unknown"; break;
1906 	}
1907 
1908 	switch (msg->tag.vio_subtype_env) {
1909 #define	Q(_s)	case _s : ses = #_s; break;
1910 	Q(VIO_VER_INFO)
1911 	Q(VIO_ATTR_INFO)
1912 	Q(VIO_DRING_REG)
1913 	Q(VIO_DRING_UNREG)
1914 	Q(VIO_RDX)
1915 	Q(VIO_PKT_DATA)
1916 	Q(VIO_DESC_DATA)
1917 	Q(VIO_DRING_DATA)
1918 #undef Q
1919 	default: ses = "unknown"; break;
1920 	}
1921 
1922 	DMSG(vdcp, 3, "(%x/%x/%x) message : (%s/%s/%s)\n",
1923 	    msg->tag.vio_msgtype, msg->tag.vio_subtype,
1924 	    msg->tag.vio_subtype_env, ms, ss, ses);
1925 }
1926 #endif
1927 
1928 /*
1929  * Function:
1930  *	vdc_send()
1931  *
1932  * Description:
1933  *	The function encapsulates the call to write a message using LDC.
1934  *	If LDC indicates that the call failed due to the queue being full,
1935  *	we retry the ldc_write() [ up to 'vdc_retries' time ], otherwise
1936  *	we return the error returned by LDC.
1937  *
1938  * Arguments:
1939  *	ldc_handle	- LDC handle for the channel this instance of vdc uses
1940  *	pkt		- address of LDC message to be sent
1941  *	msglen		- the size of the message being sent. When the function
1942  *			  returns, this contains the number of bytes written.
1943  *
1944  * Return Code:
1945  *	0		- Success.
1946  *	EINVAL		- pkt or msglen were NULL
1947  *	ECONNRESET	- The connection was not up.
1948  *	EWOULDBLOCK	- LDC queue is full
1949  *	xxx		- other error codes returned by ldc_write
1950  */
1951 static int
1952 vdc_send(vdc_t *vdc, caddr_t pkt, size_t *msglen)
1953 {
1954 	size_t	size = 0;
1955 	int	status = 0;
1956 	clock_t delay_ticks;
1957 
1958 	ASSERT(vdc != NULL);
1959 	ASSERT(mutex_owned(&vdc->lock));
1960 	ASSERT(msglen != NULL);
1961 	ASSERT(*msglen != 0);
1962 
1963 #ifdef DEBUG
1964 	vdc_decode_tag(vdc, (vio_msg_t *)pkt);
1965 #endif
1966 	/*
1967 	 * Wait indefinitely to send if channel
1968 	 * is busy, but bail out if we succeed or
1969 	 * if the channel closes or is reset.
1970 	 */
1971 	delay_ticks = vdc_hz_min_ldc_delay;
1972 	do {
1973 		size = *msglen;
1974 		status = ldc_write(vdc->ldc_handle, pkt, &size);
1975 		if (status == EWOULDBLOCK) {
1976 			delay(delay_ticks);
1977 			/* geometric backoff */
1978 			delay_ticks *= 2;
1979 			if (delay_ticks > vdc_hz_max_ldc_delay)
1980 				delay_ticks = vdc_hz_max_ldc_delay;
1981 		}
1982 	} while (status == EWOULDBLOCK);
1983 
1984 	/* if LDC had serious issues --- reset vdc state */
1985 	if (status == EIO || status == ECONNRESET) {
1986 		/* LDC had serious issues --- reset vdc state */
1987 		mutex_enter(&vdc->read_lock);
1988 		if ((vdc->read_state == VDC_READ_WAITING) ||
1989 		    (vdc->read_state == VDC_READ_RESET))
1990 			cv_signal(&vdc->read_cv);
1991 		vdc->read_state = VDC_READ_RESET;
1992 		mutex_exit(&vdc->read_lock);
1993 
1994 		/* wake up any waiters in the reset thread */
1995 		if (vdc->state == VDC_STATE_INIT_WAITING) {
1996 			DMSG(vdc, 0, "[%d] write reset - "
1997 			    "vdc is resetting ..\n", vdc->instance);
1998 			vdc->state = VDC_STATE_RESETTING;
1999 			cv_signal(&vdc->initwait_cv);
2000 		}
2001 
2002 		return (ECONNRESET);
2003 	}
2004 
2005 	/* return the last size written */
2006 	*msglen = size;
2007 
2008 	return (status);
2009 }
2010 
2011 /*
2012  * Function:
2013  *	vdc_get_md_node
2014  *
2015  * Description:
2016  *	Get the MD, the device node and the port node for the given
2017  *	disk instance. The caller is responsible for cleaning up the
2018  *	reference to the returned MD (mdpp) by calling md_fini_handle().
2019  *
2020  * Arguments:
2021  *	dip	- dev info pointer for this instance of the device driver.
2022  *	mdpp	- the returned MD.
2023  *	vd_nodep - the returned device node.
2024  *	vd_portp - the returned port node. The returned port node is NULL
2025  *		   if no port node is found.
2026  *
2027  * Return Code:
2028  *	0	- Success.
2029  *	ENOENT	- Expected node or property did not exist.
2030  *	ENXIO	- Unexpected error communicating with MD framework
2031  */
2032 static int
2033 vdc_get_md_node(dev_info_t *dip, md_t **mdpp, mde_cookie_t *vd_nodep,
2034     mde_cookie_t *vd_portp)
2035 {
2036 	int		status = ENOENT;
2037 	char		*node_name = NULL;
2038 	md_t		*mdp = NULL;
2039 	int		num_nodes;
2040 	int		num_vdevs;
2041 	int		num_vports;
2042 	mde_cookie_t	rootnode;
2043 	mde_cookie_t	*listp = NULL;
2044 	boolean_t	found_inst = B_FALSE;
2045 	int		listsz;
2046 	int		idx;
2047 	uint64_t	md_inst;
2048 	int		obp_inst;
2049 	int		instance = ddi_get_instance(dip);
2050 
2051 	/*
2052 	 * Get the OBP instance number for comparison with the MD instance
2053 	 *
2054 	 * The "cfg-handle" property of a vdc node in an MD contains the MD's
2055 	 * notion of "instance", or unique identifier, for that node; OBP
2056 	 * stores the value of the "cfg-handle" MD property as the value of
2057 	 * the "reg" property on the node in the device tree it builds from
2058 	 * the MD and passes to Solaris.  Thus, we look up the devinfo node's
2059 	 * "reg" property value to uniquely identify this device instance.
2060 	 * If the "reg" property cannot be found, the device tree state is
2061 	 * presumably so broken that there is no point in continuing.
2062 	 */
2063 	if (!ddi_prop_exists(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS, OBP_REG)) {
2064 		cmn_err(CE_WARN, "'%s' property does not exist", OBP_REG);
2065 		return (ENOENT);
2066 	}
2067 	obp_inst = ddi_prop_get_int(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS,
2068 	    OBP_REG, -1);
2069 	DMSGX(1, "[%d] OBP inst=%d\n", instance, obp_inst);
2070 
2071 	/*
2072 	 * We now walk the MD nodes to find the node for this vdisk.
2073 	 */
2074 	if ((mdp = md_get_handle()) == NULL) {
2075 		cmn_err(CE_WARN, "unable to init machine description");
2076 		return (ENXIO);
2077 	}
2078 
2079 	num_nodes = md_node_count(mdp);
2080 	ASSERT(num_nodes > 0);
2081 
2082 	listsz = num_nodes * sizeof (mde_cookie_t);
2083 
2084 	/* allocate memory for nodes */
2085 	listp = kmem_zalloc(listsz, KM_SLEEP);
2086 
2087 	rootnode = md_root_node(mdp);
2088 	ASSERT(rootnode != MDE_INVAL_ELEM_COOKIE);
2089 
2090 	/*
2091 	 * Search for all the virtual devices, we will then check to see which
2092 	 * ones are disk nodes.
2093 	 */
2094 	num_vdevs = md_scan_dag(mdp, rootnode,
2095 	    md_find_name(mdp, VDC_MD_VDEV_NAME),
2096 	    md_find_name(mdp, "fwd"), listp);
2097 
2098 	if (num_vdevs <= 0) {
2099 		cmn_err(CE_NOTE, "No '%s' node found", VDC_MD_VDEV_NAME);
2100 		status = ENOENT;
2101 		goto done;
2102 	}
2103 
2104 	DMSGX(1, "[%d] num_vdevs=%d\n", instance, num_vdevs);
2105 	for (idx = 0; idx < num_vdevs; idx++) {
2106 		status = md_get_prop_str(mdp, listp[idx], "name", &node_name);
2107 		if ((status != 0) || (node_name == NULL)) {
2108 			cmn_err(CE_NOTE, "Unable to get name of node type '%s'"
2109 			    ": err %d", VDC_MD_VDEV_NAME, status);
2110 			continue;
2111 		}
2112 
2113 		DMSGX(1, "[%d] Found node '%s'\n", instance, node_name);
2114 		if (strcmp(VDC_MD_DISK_NAME, node_name) == 0) {
2115 			status = md_get_prop_val(mdp, listp[idx],
2116 			    VDC_MD_CFG_HDL, &md_inst);
2117 			DMSGX(1, "[%d] vdc inst in MD=%lx\n",
2118 			    instance, md_inst);
2119 			if ((status == 0) && (md_inst == obp_inst)) {
2120 				found_inst = B_TRUE;
2121 				break;
2122 			}
2123 		}
2124 	}
2125 
2126 	if (!found_inst) {
2127 		DMSGX(0, "Unable to find correct '%s' node", VDC_MD_DISK_NAME);
2128 		status = ENOENT;
2129 		goto done;
2130 	}
2131 	DMSGX(0, "[%d] MD inst=%lx\n", instance, md_inst);
2132 
2133 	*vd_nodep = listp[idx];
2134 	*mdpp = mdp;
2135 
2136 	num_vports = md_scan_dag(mdp, *vd_nodep,
2137 	    md_find_name(mdp, VDC_MD_PORT_NAME),
2138 	    md_find_name(mdp, "fwd"), listp);
2139 
2140 	if (num_vports != 1) {
2141 		DMSGX(0, "Expected 1 '%s' node for '%s' port, found %d\n",
2142 		    VDC_MD_PORT_NAME, VDC_MD_VDEV_NAME, num_vports);
2143 	}
2144 
2145 	*vd_portp = (num_vports == 0)? NULL: listp[0];
2146 
2147 done:
2148 	kmem_free(listp, listsz);
2149 	return (status);
2150 }
2151 
2152 /*
2153  * Function:
2154  *	vdc_get_ldc_id()
2155  *
2156  * Description:
2157  *	This function gets the 'ldc-id' for this particular instance of vdc.
2158  *	The id returned is the guest domain channel endpoint LDC uses for
2159  *	communication with vds.
2160  *
2161  * Arguments:
2162  *	mdp	- pointer to the machine description.
2163  *	vd_node	- the vdisk element from the MD.
2164  *	ldc_id	- pointer to variable used to return the 'ldc-id' found.
2165  *
2166  * Return Code:
2167  *	0	- Success.
2168  *	ENOENT	- Expected node or property did not exist.
2169  */
2170 static int
2171 vdc_get_ldc_id(md_t *mdp, mde_cookie_t vd_node, uint64_t *ldc_id)
2172 {
2173 	mde_cookie_t	*chanp = NULL;
2174 	int		listsz;
2175 	int		num_chans;
2176 	int		num_nodes;
2177 	int		status = 0;
2178 
2179 	num_nodes = md_node_count(mdp);
2180 	ASSERT(num_nodes > 0);
2181 
2182 	listsz = num_nodes * sizeof (mde_cookie_t);
2183 
2184 	/* allocate memory for nodes */
2185 	chanp = kmem_zalloc(listsz, KM_SLEEP);
2186 
2187 	/* get the channels for this node */
2188 	num_chans = md_scan_dag(mdp, vd_node,
2189 	    md_find_name(mdp, VDC_MD_CHAN_NAME),
2190 	    md_find_name(mdp, "fwd"), chanp);
2191 
2192 	/* expecting at least one channel */
2193 	if (num_chans <= 0) {
2194 		cmn_err(CE_NOTE, "No '%s' node for '%s' port",
2195 		    VDC_MD_CHAN_NAME, VDC_MD_VDEV_NAME);
2196 		status = ENOENT;
2197 		goto done;
2198 
2199 	} else if (num_chans != 1) {
2200 		DMSGX(0, "Expected 1 '%s' node for '%s' port, found %d\n",
2201 		    VDC_MD_CHAN_NAME, VDC_MD_VDEV_NAME, num_chans);
2202 	}
2203 
2204 	/*
2205 	 * We use the first channel found (index 0), irrespective of how
2206 	 * many are there in total.
2207 	 */
2208 	if (md_get_prop_val(mdp, chanp[0], VDC_MD_ID, ldc_id) != 0) {
2209 		cmn_err(CE_NOTE, "Channel '%s' property not found", VDC_MD_ID);
2210 		status = ENOENT;
2211 	}
2212 
2213 done:
2214 	kmem_free(chanp, listsz);
2215 	return (status);
2216 }
2217 
2218 static int
2219 vdc_do_ldc_up(vdc_t *vdc)
2220 {
2221 	int		status;
2222 	ldc_status_t	ldc_state;
2223 
2224 	DMSG(vdc, 0, "[%d] Bringing up channel %lx\n",
2225 	    vdc->instance, vdc->ldc_id);
2226 
2227 	if (vdc->lifecycle == VDC_LC_DETACHING)
2228 		return (EINVAL);
2229 
2230 	if ((status = ldc_up(vdc->ldc_handle)) != 0) {
2231 		switch (status) {
2232 		case ECONNREFUSED:	/* listener not ready at other end */
2233 			DMSG(vdc, 0, "[%d] ldc_up(%lx,...) return %d\n",
2234 			    vdc->instance, vdc->ldc_id, status);
2235 			status = 0;
2236 			break;
2237 		default:
2238 			DMSG(vdc, 0, "[%d] Failed to bring up LDC: "
2239 			    "channel=%ld, err=%d", vdc->instance, vdc->ldc_id,
2240 			    status);
2241 			break;
2242 		}
2243 	}
2244 
2245 	if (ldc_status(vdc->ldc_handle, &ldc_state) == 0) {
2246 		vdc->ldc_state = ldc_state;
2247 		if (ldc_state == LDC_UP) {
2248 			DMSG(vdc, 0, "[%d] LDC channel already up\n",
2249 			    vdc->instance);
2250 			vdc->seq_num = 1;
2251 			vdc->seq_num_reply = 0;
2252 		}
2253 	}
2254 
2255 	return (status);
2256 }
2257 
2258 /*
2259  * Function:
2260  *	vdc_terminate_ldc()
2261  *
2262  * Description:
2263  *
2264  * Arguments:
2265  *	vdc	- soft state pointer for this instance of the device driver.
2266  *
2267  * Return Code:
2268  *	None
2269  */
2270 static void
2271 vdc_terminate_ldc(vdc_t *vdc)
2272 {
2273 	int	instance = ddi_get_instance(vdc->dip);
2274 
2275 	ASSERT(vdc != NULL);
2276 	ASSERT(mutex_owned(&vdc->lock));
2277 
2278 	DMSG(vdc, 0, "[%d] initialized=%x\n", instance, vdc->initialized);
2279 
2280 	if (vdc->initialized & VDC_LDC_OPEN) {
2281 		DMSG(vdc, 0, "[%d] ldc_close()\n", instance);
2282 		(void) ldc_close(vdc->ldc_handle);
2283 	}
2284 	if (vdc->initialized & VDC_LDC_CB) {
2285 		DMSG(vdc, 0, "[%d] ldc_unreg_callback()\n", instance);
2286 		(void) ldc_unreg_callback(vdc->ldc_handle);
2287 	}
2288 	if (vdc->initialized & VDC_LDC) {
2289 		DMSG(vdc, 0, "[%d] ldc_fini()\n", instance);
2290 		(void) ldc_fini(vdc->ldc_handle);
2291 		vdc->ldc_handle = NULL;
2292 	}
2293 
2294 	vdc->initialized &= ~(VDC_LDC | VDC_LDC_CB | VDC_LDC_OPEN);
2295 }
2296 
2297 /* -------------------------------------------------------------------------- */
2298 
2299 /*
2300  * Descriptor Ring helper routines
2301  */
2302 
2303 /*
2304  * Function:
2305  *	vdc_init_descriptor_ring()
2306  *
2307  * Description:
2308  *
2309  * Arguments:
2310  *	vdc	- soft state pointer for this instance of the device driver.
2311  *
2312  * Return Code:
2313  *	0	- Success
2314  */
2315 static int
2316 vdc_init_descriptor_ring(vdc_t *vdc)
2317 {
2318 	vd_dring_entry_t	*dep = NULL;	/* DRing Entry pointer */
2319 	int	status = 0;
2320 	int	i;
2321 
2322 	DMSG(vdc, 0, "[%d] initialized=%x\n", vdc->instance, vdc->initialized);
2323 
2324 	ASSERT(vdc != NULL);
2325 	ASSERT(mutex_owned(&vdc->lock));
2326 	ASSERT(vdc->ldc_handle != NULL);
2327 
2328 	/* ensure we have enough room to store max sized block */
2329 	ASSERT(maxphys <= VD_MAX_BLOCK_SIZE);
2330 
2331 	if ((vdc->initialized & VDC_DRING_INIT) == 0) {
2332 		DMSG(vdc, 0, "[%d] ldc_mem_dring_create\n", vdc->instance);
2333 		/*
2334 		 * Calculate the maximum block size we can transmit using one
2335 		 * Descriptor Ring entry from the attributes returned by the
2336 		 * vDisk server. This is subject to a minimum of 'maxphys'
2337 		 * as we do not have the capability to split requests over
2338 		 * multiple DRing entries.
2339 		 */
2340 		if ((vdc->max_xfer_sz * vdc->block_size) < maxphys) {
2341 			DMSG(vdc, 0, "[%d] using minimum DRing size\n",
2342 			    vdc->instance);
2343 			vdc->dring_max_cookies = maxphys / PAGESIZE;
2344 		} else {
2345 			vdc->dring_max_cookies =
2346 			    (vdc->max_xfer_sz * vdc->block_size) / PAGESIZE;
2347 		}
2348 		vdc->dring_entry_size = (sizeof (vd_dring_entry_t) +
2349 		    (sizeof (ldc_mem_cookie_t) *
2350 		    (vdc->dring_max_cookies - 1)));
2351 		vdc->dring_len = VD_DRING_LEN;
2352 
2353 		status = ldc_mem_dring_create(vdc->dring_len,
2354 		    vdc->dring_entry_size, &vdc->ldc_dring_hdl);
2355 		if ((vdc->ldc_dring_hdl == NULL) || (status != 0)) {
2356 			DMSG(vdc, 0, "[%d] Descriptor ring creation failed",
2357 			    vdc->instance);
2358 			return (status);
2359 		}
2360 		vdc->initialized |= VDC_DRING_INIT;
2361 	}
2362 
2363 	if ((vdc->initialized & VDC_DRING_BOUND) == 0) {
2364 		DMSG(vdc, 0, "[%d] ldc_mem_dring_bind\n", vdc->instance);
2365 		vdc->dring_cookie =
2366 		    kmem_zalloc(sizeof (ldc_mem_cookie_t), KM_SLEEP);
2367 
2368 		status = ldc_mem_dring_bind(vdc->ldc_handle, vdc->ldc_dring_hdl,
2369 		    LDC_SHADOW_MAP|LDC_DIRECT_MAP, LDC_MEM_RW,
2370 		    &vdc->dring_cookie[0],
2371 		    &vdc->dring_cookie_count);
2372 		if (status != 0) {
2373 			DMSG(vdc, 0, "[%d] Failed to bind descriptor ring "
2374 			    "(%lx) to channel (%lx) status=%d\n",
2375 			    vdc->instance, vdc->ldc_dring_hdl,
2376 			    vdc->ldc_handle, status);
2377 			return (status);
2378 		}
2379 		ASSERT(vdc->dring_cookie_count == 1);
2380 		vdc->initialized |= VDC_DRING_BOUND;
2381 	}
2382 
2383 	status = ldc_mem_dring_info(vdc->ldc_dring_hdl, &vdc->dring_mem_info);
2384 	if (status != 0) {
2385 		DMSG(vdc, 0,
2386 		    "[%d] Failed to get info for descriptor ring (%lx)\n",
2387 		    vdc->instance, vdc->ldc_dring_hdl);
2388 		return (status);
2389 	}
2390 
2391 	if ((vdc->initialized & VDC_DRING_LOCAL) == 0) {
2392 		DMSG(vdc, 0, "[%d] local dring\n", vdc->instance);
2393 
2394 		/* Allocate the local copy of this dring */
2395 		vdc->local_dring =
2396 		    kmem_zalloc(vdc->dring_len * sizeof (vdc_local_desc_t),
2397 		    KM_SLEEP);
2398 		vdc->initialized |= VDC_DRING_LOCAL;
2399 	}
2400 
2401 	/*
2402 	 * Mark all DRing entries as free and initialize the private
2403 	 * descriptor's memory handles. If any entry is initialized,
2404 	 * we need to free it later so we set the bit in 'initialized'
2405 	 * at the start.
2406 	 */
2407 	vdc->initialized |= VDC_DRING_ENTRY;
2408 	for (i = 0; i < vdc->dring_len; i++) {
2409 		dep = VDC_GET_DRING_ENTRY_PTR(vdc, i);
2410 		dep->hdr.dstate = VIO_DESC_FREE;
2411 
2412 		status = ldc_mem_alloc_handle(vdc->ldc_handle,
2413 		    &vdc->local_dring[i].desc_mhdl);
2414 		if (status != 0) {
2415 			DMSG(vdc, 0, "![%d] Failed to alloc mem handle for"
2416 			    " descriptor %d", vdc->instance, i);
2417 			return (status);
2418 		}
2419 		vdc->local_dring[i].is_free = B_TRUE;
2420 		vdc->local_dring[i].dep = dep;
2421 	}
2422 
2423 	/* Initialize the starting index */
2424 	vdc->dring_curr_idx = 0;
2425 
2426 	return (status);
2427 }
2428 
2429 /*
2430  * Function:
2431  *	vdc_destroy_descriptor_ring()
2432  *
2433  * Description:
2434  *
2435  * Arguments:
2436  *	vdc	- soft state pointer for this instance of the device driver.
2437  *
2438  * Return Code:
2439  *	None
2440  */
2441 static void
2442 vdc_destroy_descriptor_ring(vdc_t *vdc)
2443 {
2444 	vdc_local_desc_t	*ldep = NULL;	/* Local Dring Entry Pointer */
2445 	ldc_mem_handle_t	mhdl = NULL;
2446 	ldc_mem_info_t		minfo;
2447 	int			status = -1;
2448 	int			i;	/* loop */
2449 
2450 	ASSERT(vdc != NULL);
2451 	ASSERT(mutex_owned(&vdc->lock));
2452 
2453 	DMSG(vdc, 0, "[%d] Entered\n", vdc->instance);
2454 
2455 	if (vdc->initialized & VDC_DRING_ENTRY) {
2456 		DMSG(vdc, 0,
2457 		    "[%d] Removing Local DRing entries\n", vdc->instance);
2458 		for (i = 0; i < vdc->dring_len; i++) {
2459 			ldep = &vdc->local_dring[i];
2460 			mhdl = ldep->desc_mhdl;
2461 
2462 			if (mhdl == NULL)
2463 				continue;
2464 
2465 			if ((status = ldc_mem_info(mhdl, &minfo)) != 0) {
2466 				DMSG(vdc, 0,
2467 				    "ldc_mem_info returned an error: %d\n",
2468 				    status);
2469 
2470 				/*
2471 				 * This must mean that the mem handle
2472 				 * is not valid. Clear it out so that
2473 				 * no one tries to use it.
2474 				 */
2475 				ldep->desc_mhdl = NULL;
2476 				continue;
2477 			}
2478 
2479 			if (minfo.status == LDC_BOUND) {
2480 				(void) ldc_mem_unbind_handle(mhdl);
2481 			}
2482 
2483 			(void) ldc_mem_free_handle(mhdl);
2484 
2485 			ldep->desc_mhdl = NULL;
2486 		}
2487 		vdc->initialized &= ~VDC_DRING_ENTRY;
2488 	}
2489 
2490 	if (vdc->initialized & VDC_DRING_LOCAL) {
2491 		DMSG(vdc, 0, "[%d] Freeing Local DRing\n", vdc->instance);
2492 		kmem_free(vdc->local_dring,
2493 		    vdc->dring_len * sizeof (vdc_local_desc_t));
2494 		vdc->initialized &= ~VDC_DRING_LOCAL;
2495 	}
2496 
2497 	if (vdc->initialized & VDC_DRING_BOUND) {
2498 		DMSG(vdc, 0, "[%d] Unbinding DRing\n", vdc->instance);
2499 		status = ldc_mem_dring_unbind(vdc->ldc_dring_hdl);
2500 		if (status == 0) {
2501 			vdc->initialized &= ~VDC_DRING_BOUND;
2502 		} else {
2503 			DMSG(vdc, 0, "[%d] Error %d unbinding DRing %lx",
2504 			    vdc->instance, status, vdc->ldc_dring_hdl);
2505 		}
2506 		kmem_free(vdc->dring_cookie, sizeof (ldc_mem_cookie_t));
2507 	}
2508 
2509 	if (vdc->initialized & VDC_DRING_INIT) {
2510 		DMSG(vdc, 0, "[%d] Destroying DRing\n", vdc->instance);
2511 		status = ldc_mem_dring_destroy(vdc->ldc_dring_hdl);
2512 		if (status == 0) {
2513 			vdc->ldc_dring_hdl = NULL;
2514 			bzero(&vdc->dring_mem_info, sizeof (ldc_mem_info_t));
2515 			vdc->initialized &= ~VDC_DRING_INIT;
2516 		} else {
2517 			DMSG(vdc, 0, "[%d] Error %d destroying DRing (%lx)",
2518 			    vdc->instance, status, vdc->ldc_dring_hdl);
2519 		}
2520 	}
2521 }
2522 
2523 /*
2524  * Function:
2525  *	vdc_map_to_shared_ring()
2526  *
2527  * Description:
2528  *	Copy contents of the local descriptor to the shared
2529  *	memory descriptor.
2530  *
2531  * Arguments:
2532  *	vdcp	- soft state pointer for this instance of the device driver.
2533  *	idx	- descriptor ring index
2534  *
2535  * Return Code:
2536  *	None
2537  */
2538 static int
2539 vdc_map_to_shared_dring(vdc_t *vdcp, int idx)
2540 {
2541 	vdc_local_desc_t	*ldep;
2542 	vd_dring_entry_t	*dep;
2543 	int			rv;
2544 
2545 	ldep = &(vdcp->local_dring[idx]);
2546 
2547 	/* for now leave in the old pop_mem_hdl stuff */
2548 	if (ldep->nbytes > 0) {
2549 		rv = vdc_populate_mem_hdl(vdcp, ldep);
2550 		if (rv) {
2551 			DMSG(vdcp, 0, "[%d] Cannot populate mem handle\n",
2552 			    vdcp->instance);
2553 			return (rv);
2554 		}
2555 	}
2556 
2557 	/*
2558 	 * fill in the data details into the DRing
2559 	 */
2560 	dep = ldep->dep;
2561 	ASSERT(dep != NULL);
2562 
2563 	dep->payload.req_id = VDC_GET_NEXT_REQ_ID(vdcp);
2564 	dep->payload.operation = ldep->operation;
2565 	dep->payload.addr = ldep->offset;
2566 	dep->payload.nbytes = ldep->nbytes;
2567 	dep->payload.status = (uint32_t)-1;	/* vds will set valid value */
2568 	dep->payload.slice = ldep->slice;
2569 	dep->hdr.dstate = VIO_DESC_READY;
2570 	dep->hdr.ack = 1;		/* request an ACK for every message */
2571 
2572 	return (0);
2573 }
2574 
2575 /*
2576  * Function:
2577  *	vdc_send_request
2578  *
2579  * Description:
2580  *	This routine writes the data to be transmitted to vds into the
2581  *	descriptor, notifies vds that the ring has been updated and
2582  *	then waits for the request to be processed.
2583  *
2584  * Arguments:
2585  *	vdcp	  - the soft state pointer
2586  *	operation - operation we want vds to perform (VD_OP_XXX)
2587  *	addr	  - address of data buf to be read/written.
2588  *	nbytes	  - number of bytes to read/write
2589  *	slice	  - the disk slice this request is for
2590  *	offset	  - relative disk offset
2591  *	cb_type   - type of call - STRATEGY or SYNC
2592  *	cb_arg	  - parameter to be sent to server (depends on VD_OP_XXX type)
2593  *			. mode for ioctl(9e)
2594  *			. LP64 diskaddr_t (block I/O)
2595  *	dir	  - direction of operation (READ/WRITE/BOTH)
2596  *
2597  * Return Codes:
2598  *	0
2599  *	ENXIO
2600  */
2601 static int
2602 vdc_send_request(vdc_t *vdcp, int operation, caddr_t addr,
2603     size_t nbytes, int slice, diskaddr_t offset, int cb_type,
2604     void *cb_arg, vio_desc_direction_t dir)
2605 {
2606 	ASSERT(vdcp != NULL);
2607 	ASSERT(slice == VD_SLICE_NONE || slice < V_NUMPAR);
2608 
2609 	mutex_enter(&vdcp->lock);
2610 
2611 	do {
2612 		while (vdcp->state != VDC_STATE_RUNNING) {
2613 
2614 			/* return error if detaching */
2615 			if (vdcp->state == VDC_STATE_DETACH) {
2616 				mutex_exit(&vdcp->lock);
2617 				return (ENXIO);
2618 			}
2619 
2620 			/* fail request if connection timeout is reached */
2621 			if (vdcp->ctimeout_reached) {
2622 				mutex_exit(&vdcp->lock);
2623 				return (EIO);
2624 			}
2625 
2626 			cv_wait(&vdcp->running_cv, &vdcp->lock);
2627 		}
2628 
2629 	} while (vdc_populate_descriptor(vdcp, operation, addr,
2630 	    nbytes, slice, offset, cb_type, cb_arg, dir));
2631 
2632 	mutex_exit(&vdcp->lock);
2633 	return (0);
2634 }
2635 
2636 
2637 /*
2638  * Function:
2639  *	vdc_populate_descriptor
2640  *
2641  * Description:
2642  *	This routine writes the data to be transmitted to vds into the
2643  *	descriptor, notifies vds that the ring has been updated and
2644  *	then waits for the request to be processed.
2645  *
2646  * Arguments:
2647  *	vdcp	  - the soft state pointer
2648  *	operation - operation we want vds to perform (VD_OP_XXX)
2649  *	addr	  - address of data buf to be read/written.
2650  *	nbytes	  - number of bytes to read/write
2651  *	slice	  - the disk slice this request is for
2652  *	offset	  - relative disk offset
2653  *	cb_type   - type of call - STRATEGY or SYNC
2654  *	cb_arg	  - parameter to be sent to server (depends on VD_OP_XXX type)
2655  *			. mode for ioctl(9e)
2656  *			. LP64 diskaddr_t (block I/O)
2657  *	dir	  - direction of operation (READ/WRITE/BOTH)
2658  *
2659  * Return Codes:
2660  *	0
2661  *	EAGAIN
2662  *		EFAULT
2663  *		ENXIO
2664  *		EIO
2665  */
2666 static int
2667 vdc_populate_descriptor(vdc_t *vdcp, int operation, caddr_t addr,
2668     size_t nbytes, int slice, diskaddr_t offset, int cb_type,
2669     void *cb_arg, vio_desc_direction_t dir)
2670 {
2671 	vdc_local_desc_t	*local_dep = NULL; /* Local Dring Pointer */
2672 	int			idx;		/* Index of DRing entry used */
2673 	int			next_idx;
2674 	vio_dring_msg_t		dmsg;
2675 	size_t			msglen;
2676 	int			rv;
2677 
2678 	ASSERT(MUTEX_HELD(&vdcp->lock));
2679 	vdcp->threads_pending++;
2680 loop:
2681 	DMSG(vdcp, 2, ": dring_curr_idx = %d\n", vdcp->dring_curr_idx);
2682 
2683 	/* Get next available D-Ring entry */
2684 	idx = vdcp->dring_curr_idx;
2685 	local_dep = &(vdcp->local_dring[idx]);
2686 
2687 	if (!local_dep->is_free) {
2688 		DMSG(vdcp, 2, "[%d]: dring full - waiting for space\n",
2689 		    vdcp->instance);
2690 		cv_wait(&vdcp->dring_free_cv, &vdcp->lock);
2691 		if (vdcp->state == VDC_STATE_RUNNING ||
2692 		    vdcp->state == VDC_STATE_HANDLE_PENDING) {
2693 			goto loop;
2694 		}
2695 		vdcp->threads_pending--;
2696 		return (ECONNRESET);
2697 	}
2698 
2699 	next_idx = idx + 1;
2700 	if (next_idx >= vdcp->dring_len)
2701 		next_idx = 0;
2702 	vdcp->dring_curr_idx = next_idx;
2703 
2704 	ASSERT(local_dep->is_free);
2705 
2706 	local_dep->operation = operation;
2707 	local_dep->addr = addr;
2708 	local_dep->nbytes = nbytes;
2709 	local_dep->slice = slice;
2710 	local_dep->offset = offset;
2711 	local_dep->cb_type = cb_type;
2712 	local_dep->cb_arg = cb_arg;
2713 	local_dep->dir = dir;
2714 
2715 	local_dep->is_free = B_FALSE;
2716 
2717 	rv = vdc_map_to_shared_dring(vdcp, idx);
2718 	if (rv) {
2719 		DMSG(vdcp, 0, "[%d]: cannot bind memory - waiting ..\n",
2720 		    vdcp->instance);
2721 		/* free the descriptor */
2722 		local_dep->is_free = B_TRUE;
2723 		vdcp->dring_curr_idx = idx;
2724 		cv_wait(&vdcp->membind_cv, &vdcp->lock);
2725 		if (vdcp->state == VDC_STATE_RUNNING ||
2726 		    vdcp->state == VDC_STATE_HANDLE_PENDING) {
2727 			goto loop;
2728 		}
2729 		vdcp->threads_pending--;
2730 		return (ECONNRESET);
2731 	}
2732 
2733 	/*
2734 	 * Send a msg with the DRing details to vds
2735 	 */
2736 	VIO_INIT_DRING_DATA_TAG(dmsg);
2737 	VDC_INIT_DRING_DATA_MSG_IDS(dmsg, vdcp);
2738 	dmsg.dring_ident = vdcp->dring_ident;
2739 	dmsg.start_idx = idx;
2740 	dmsg.end_idx = idx;
2741 	vdcp->seq_num++;
2742 
2743 	DTRACE_IO2(send, vio_dring_msg_t *, &dmsg, vdc_t *, vdcp);
2744 
2745 	DMSG(vdcp, 2, "ident=0x%lx, st=%u, end=%u, seq=%ld\n",
2746 	    vdcp->dring_ident, dmsg.start_idx, dmsg.end_idx, dmsg.seq_num);
2747 
2748 	/*
2749 	 * note we're still holding the lock here to
2750 	 * make sure the message goes out in order !!!...
2751 	 */
2752 	msglen = sizeof (dmsg);
2753 	rv = vdc_send(vdcp, (caddr_t)&dmsg, &msglen);
2754 	switch (rv) {
2755 	case ECONNRESET:
2756 		/*
2757 		 * vdc_send initiates the reset on failure.
2758 		 * Since the transaction has already been put
2759 		 * on the local dring, it will automatically get
2760 		 * retried when the channel is reset. Given that,
2761 		 * it is ok to just return success even though the
2762 		 * send failed.
2763 		 */
2764 		rv = 0;
2765 		break;
2766 
2767 	case 0: /* EOK */
2768 		DMSG(vdcp, 1, "sent via LDC: rv=%d\n", rv);
2769 		break;
2770 
2771 	default:
2772 		goto cleanup_and_exit;
2773 	}
2774 
2775 	vdcp->threads_pending--;
2776 	return (rv);
2777 
2778 cleanup_and_exit:
2779 	DMSG(vdcp, 0, "unexpected error, rv=%d\n", rv);
2780 	return (ENXIO);
2781 }
2782 
2783 /*
2784  * Function:
2785  *	vdc_do_sync_op
2786  *
2787  * Description:
2788  * 	Wrapper around vdc_populate_descriptor that blocks until the
2789  * 	response to the message is available.
2790  *
2791  * Arguments:
2792  *	vdcp	  - the soft state pointer
2793  *	operation - operation we want vds to perform (VD_OP_XXX)
2794  *	addr	  - address of data buf to be read/written.
2795  *	nbytes	  - number of bytes to read/write
2796  *	slice	  - the disk slice this request is for
2797  *	offset	  - relative disk offset
2798  *	cb_type   - type of call - STRATEGY or SYNC
2799  *	cb_arg	  - parameter to be sent to server (depends on VD_OP_XXX type)
2800  *			. mode for ioctl(9e)
2801  *			. LP64 diskaddr_t (block I/O)
2802  *	dir	  - direction of operation (READ/WRITE/BOTH)
2803  *
2804  * Return Codes:
2805  *	0
2806  *	EAGAIN
2807  *		EFAULT
2808  *		ENXIO
2809  *		EIO
2810  */
2811 static int
2812 vdc_do_sync_op(vdc_t *vdcp, int operation, caddr_t addr, size_t nbytes,
2813     int slice, diskaddr_t offset, int cb_type, void *cb_arg,
2814     vio_desc_direction_t dir)
2815 {
2816 	int status;
2817 
2818 	ASSERT(cb_type == CB_SYNC);
2819 
2820 	/*
2821 	 * Grab the lock, if blocked wait until the server
2822 	 * response causes us to wake up again.
2823 	 */
2824 	mutex_enter(&vdcp->lock);
2825 	vdcp->sync_op_cnt++;
2826 	while (vdcp->sync_op_blocked && vdcp->state != VDC_STATE_DETACH)
2827 		cv_wait(&vdcp->sync_blocked_cv, &vdcp->lock);
2828 
2829 	if (vdcp->state == VDC_STATE_DETACH) {
2830 		cv_broadcast(&vdcp->sync_blocked_cv);
2831 		vdcp->sync_op_cnt--;
2832 		mutex_exit(&vdcp->lock);
2833 		return (ENXIO);
2834 	}
2835 
2836 	/* now block anyone other thread entering after us */
2837 	vdcp->sync_op_blocked = B_TRUE;
2838 	vdcp->sync_op_pending = B_TRUE;
2839 	mutex_exit(&vdcp->lock);
2840 
2841 	status = vdc_send_request(vdcp, operation, addr,
2842 	    nbytes, slice, offset, cb_type, cb_arg, dir);
2843 
2844 	mutex_enter(&vdcp->lock);
2845 
2846 	if (status != 0) {
2847 		vdcp->sync_op_pending = B_FALSE;
2848 	} else {
2849 		/*
2850 		 * block until our transaction completes.
2851 		 * Also anyone else waiting also gets to go next.
2852 		 */
2853 		while (vdcp->sync_op_pending && vdcp->state != VDC_STATE_DETACH)
2854 			cv_wait(&vdcp->sync_pending_cv, &vdcp->lock);
2855 
2856 		DMSG(vdcp, 2, ": operation returned %d\n",
2857 		    vdcp->sync_op_status);
2858 		if (vdcp->state == VDC_STATE_DETACH) {
2859 			vdcp->sync_op_pending = B_FALSE;
2860 			status = ENXIO;
2861 		} else {
2862 			status = vdcp->sync_op_status;
2863 		}
2864 	}
2865 
2866 	vdcp->sync_op_status = 0;
2867 	vdcp->sync_op_blocked = B_FALSE;
2868 	vdcp->sync_op_cnt--;
2869 
2870 	/* signal the next waiting thread */
2871 	cv_signal(&vdcp->sync_blocked_cv);
2872 	mutex_exit(&vdcp->lock);
2873 
2874 	return (status);
2875 }
2876 
2877 
2878 /*
2879  * Function:
2880  *	vdc_drain_response()
2881  *
2882  * Description:
2883  * 	When a guest is panicking, the completion of requests needs to be
2884  * 	handled differently because interrupts are disabled and vdc
2885  * 	will not get messages. We have to poll for the messages instead.
2886  *
2887  * Arguments:
2888  *	vdc	- soft state pointer for this instance of the device driver.
2889  *
2890  * Return Code:
2891  *	0	- Success
2892  */
2893 static int
2894 vdc_drain_response(vdc_t *vdc)
2895 {
2896 	int 			rv, idx, retries;
2897 	size_t			msglen;
2898 	vdc_local_desc_t 	*ldep = NULL;	/* Local Dring Entry Pointer */
2899 	vio_dring_msg_t		dmsg;
2900 
2901 	mutex_enter(&vdc->lock);
2902 
2903 	retries = 0;
2904 	for (;;) {
2905 		msglen = sizeof (dmsg);
2906 		rv = ldc_read(vdc->ldc_handle, (caddr_t)&dmsg, &msglen);
2907 		if (rv) {
2908 			rv = EINVAL;
2909 			break;
2910 		}
2911 
2912 		/*
2913 		 * if there are no packets wait and check again
2914 		 */
2915 		if ((rv == 0) && (msglen == 0)) {
2916 			if (retries++ > vdc_dump_retries) {
2917 				rv = EAGAIN;
2918 				break;
2919 			}
2920 
2921 			drv_usecwait(vdc_usec_timeout_dump);
2922 			continue;
2923 		}
2924 
2925 		/*
2926 		 * Ignore all messages that are not ACKs/NACKs to
2927 		 * DRing requests.
2928 		 */
2929 		if ((dmsg.tag.vio_msgtype != VIO_TYPE_DATA) ||
2930 		    (dmsg.tag.vio_subtype_env != VIO_DRING_DATA)) {
2931 			DMSG(vdc, 0, "discard pkt: type=%d sub=%d env=%d\n",
2932 			    dmsg.tag.vio_msgtype,
2933 			    dmsg.tag.vio_subtype,
2934 			    dmsg.tag.vio_subtype_env);
2935 			continue;
2936 		}
2937 
2938 		/*
2939 		 * set the appropriate return value for the current request.
2940 		 */
2941 		switch (dmsg.tag.vio_subtype) {
2942 		case VIO_SUBTYPE_ACK:
2943 			rv = 0;
2944 			break;
2945 		case VIO_SUBTYPE_NACK:
2946 			rv = EAGAIN;
2947 			break;
2948 		default:
2949 			continue;
2950 		}
2951 
2952 		idx = dmsg.start_idx;
2953 		if (idx >= vdc->dring_len) {
2954 			DMSG(vdc, 0, "[%d] Bogus ack data : start %d\n",
2955 			    vdc->instance, idx);
2956 			continue;
2957 		}
2958 		ldep = &vdc->local_dring[idx];
2959 		if (ldep->dep->hdr.dstate != VIO_DESC_DONE) {
2960 			DMSG(vdc, 0, "[%d] Entry @ %d - state !DONE %d\n",
2961 			    vdc->instance, idx, ldep->dep->hdr.dstate);
2962 			continue;
2963 		}
2964 
2965 		DMSG(vdc, 1, "[%d] Depopulating idx=%d state=%d\n",
2966 		    vdc->instance, idx, ldep->dep->hdr.dstate);
2967 		rv = vdc_depopulate_descriptor(vdc, idx);
2968 		if (rv) {
2969 			DMSG(vdc, 0,
2970 			    "[%d] Entry @ %d - depopulate failed ..\n",
2971 			    vdc->instance, idx);
2972 		}
2973 
2974 		/* if this is the last descriptor - break out of loop */
2975 		if ((idx + 1) % vdc->dring_len == vdc->dring_curr_idx)
2976 			break;
2977 	}
2978 
2979 	mutex_exit(&vdc->lock);
2980 	DMSG(vdc, 0, "End idx=%d\n", idx);
2981 
2982 	return (rv);
2983 }
2984 
2985 
2986 /*
2987  * Function:
2988  *	vdc_depopulate_descriptor()
2989  *
2990  * Description:
2991  *
2992  * Arguments:
2993  *	vdc	- soft state pointer for this instance of the device driver.
2994  *	idx	- Index of the Descriptor Ring entry being modified
2995  *
2996  * Return Code:
2997  *	0	- Success
2998  */
2999 static int
3000 vdc_depopulate_descriptor(vdc_t *vdc, uint_t idx)
3001 {
3002 	vd_dring_entry_t *dep = NULL;		/* Dring Entry Pointer */
3003 	vdc_local_desc_t *ldep = NULL;		/* Local Dring Entry Pointer */
3004 	int		status = ENXIO;
3005 	int		rv = 0;
3006 
3007 	ASSERT(vdc != NULL);
3008 	ASSERT(idx < vdc->dring_len);
3009 	ldep = &vdc->local_dring[idx];
3010 	ASSERT(ldep != NULL);
3011 	ASSERT(MUTEX_HELD(&vdc->lock));
3012 
3013 	DMSG(vdc, 2, ": idx = %d\n", idx);
3014 	dep = ldep->dep;
3015 	ASSERT(dep != NULL);
3016 	ASSERT((dep->hdr.dstate == VIO_DESC_DONE) ||
3017 	    (dep->payload.status == ECANCELED));
3018 
3019 	VDC_MARK_DRING_ENTRY_FREE(vdc, idx);
3020 
3021 	ldep->is_free = B_TRUE;
3022 	status = dep->payload.status;
3023 	DMSG(vdc, 2, ": is_free = %d : status = %d\n", ldep->is_free, status);
3024 
3025 	/*
3026 	 * If no buffers were used to transfer information to the server when
3027 	 * populating the descriptor then no memory handles need to be unbound
3028 	 * and we can return now.
3029 	 */
3030 	if (ldep->nbytes == 0) {
3031 		cv_signal(&vdc->dring_free_cv);
3032 		return (status);
3033 	}
3034 
3035 	/*
3036 	 * If the upper layer passed in a misaligned address we copied the
3037 	 * data into an aligned buffer before sending it to LDC - we now
3038 	 * copy it back to the original buffer.
3039 	 */
3040 	if (ldep->align_addr) {
3041 		ASSERT(ldep->addr != NULL);
3042 
3043 		if (dep->payload.nbytes > 0)
3044 			bcopy(ldep->align_addr, ldep->addr,
3045 			    dep->payload.nbytes);
3046 		kmem_free(ldep->align_addr,
3047 		    sizeof (caddr_t) * P2ROUNDUP(ldep->nbytes, 8));
3048 		ldep->align_addr = NULL;
3049 	}
3050 
3051 	rv = ldc_mem_unbind_handle(ldep->desc_mhdl);
3052 	if (rv != 0) {
3053 		DMSG(vdc, 0, "?[%d] unbind mhdl 0x%lx @ idx %d failed (%d)",
3054 		    vdc->instance, ldep->desc_mhdl, idx, rv);
3055 		/*
3056 		 * The error returned by the vDisk server is more informative
3057 		 * and thus has a higher priority but if it isn't set we ensure
3058 		 * that this function returns an error.
3059 		 */
3060 		if (status == 0)
3061 			status = EINVAL;
3062 	}
3063 
3064 	cv_signal(&vdc->membind_cv);
3065 	cv_signal(&vdc->dring_free_cv);
3066 
3067 	return (status);
3068 }
3069 
3070 /*
3071  * Function:
3072  *	vdc_populate_mem_hdl()
3073  *
3074  * Description:
3075  *
3076  * Arguments:
3077  *	vdc	- soft state pointer for this instance of the device driver.
3078  *	idx	- Index of the Descriptor Ring entry being modified
3079  *	addr	- virtual address being mapped in
3080  *	nybtes	- number of bytes in 'addr'
3081  *	operation - the vDisk operation being performed (VD_OP_xxx)
3082  *
3083  * Return Code:
3084  *	0	- Success
3085  */
3086 static int
3087 vdc_populate_mem_hdl(vdc_t *vdcp, vdc_local_desc_t *ldep)
3088 {
3089 	vd_dring_entry_t	*dep = NULL;
3090 	ldc_mem_handle_t	mhdl;
3091 	caddr_t			vaddr;
3092 	size_t			nbytes;
3093 	uint8_t			perm = LDC_MEM_RW;
3094 	uint8_t			maptype;
3095 	int			rv = 0;
3096 	int			i;
3097 
3098 	ASSERT(vdcp != NULL);
3099 
3100 	dep = ldep->dep;
3101 	mhdl = ldep->desc_mhdl;
3102 
3103 	switch (ldep->dir) {
3104 	case VIO_read_dir:
3105 		perm = LDC_MEM_W;
3106 		break;
3107 
3108 	case VIO_write_dir:
3109 		perm = LDC_MEM_R;
3110 		break;
3111 
3112 	case VIO_both_dir:
3113 		perm = LDC_MEM_RW;
3114 		break;
3115 
3116 	default:
3117 		ASSERT(0);	/* catch bad programming in vdc */
3118 	}
3119 
3120 	/*
3121 	 * LDC expects any addresses passed in to be 8-byte aligned. We need
3122 	 * to copy the contents of any misaligned buffers to a newly allocated
3123 	 * buffer and bind it instead (and copy the the contents back to the
3124 	 * original buffer passed in when depopulating the descriptor)
3125 	 */
3126 	vaddr = ldep->addr;
3127 	nbytes = ldep->nbytes;
3128 	if (((uint64_t)vaddr & 0x7) != 0) {
3129 		ASSERT(ldep->align_addr == NULL);
3130 		ldep->align_addr =
3131 		    kmem_alloc(sizeof (caddr_t) *
3132 		    P2ROUNDUP(nbytes, 8), KM_SLEEP);
3133 		DMSG(vdcp, 0, "[%d] Misaligned address %p reallocating "
3134 		    "(buf=%p nb=%ld op=%d)\n",
3135 		    vdcp->instance, (void *)vaddr, (void *)ldep->align_addr,
3136 		    nbytes, ldep->operation);
3137 		if (perm != LDC_MEM_W)
3138 			bcopy(vaddr, ldep->align_addr, nbytes);
3139 		vaddr = ldep->align_addr;
3140 	}
3141 
3142 	maptype = LDC_IO_MAP|LDC_SHADOW_MAP|LDC_DIRECT_MAP;
3143 	rv = ldc_mem_bind_handle(mhdl, vaddr, P2ROUNDUP(nbytes, 8),
3144 	    maptype, perm, &dep->payload.cookie[0], &dep->payload.ncookies);
3145 	DMSG(vdcp, 2, "[%d] bound mem handle; ncookies=%d\n",
3146 	    vdcp->instance, dep->payload.ncookies);
3147 	if (rv != 0) {
3148 		DMSG(vdcp, 0, "[%d] Failed to bind LDC memory handle "
3149 		    "(mhdl=%p, buf=%p, err=%d)\n",
3150 		    vdcp->instance, (void *)mhdl, (void *)vaddr, rv);
3151 		if (ldep->align_addr) {
3152 			kmem_free(ldep->align_addr,
3153 			    sizeof (caddr_t) * P2ROUNDUP(nbytes, 8));
3154 			ldep->align_addr = NULL;
3155 		}
3156 		return (EAGAIN);
3157 	}
3158 
3159 	/*
3160 	 * Get the other cookies (if any).
3161 	 */
3162 	for (i = 1; i < dep->payload.ncookies; i++) {
3163 		rv = ldc_mem_nextcookie(mhdl, &dep->payload.cookie[i]);
3164 		if (rv != 0) {
3165 			(void) ldc_mem_unbind_handle(mhdl);
3166 			DMSG(vdcp, 0, "?[%d] Failed to get next cookie "
3167 			    "(mhdl=%lx cnum=%d), err=%d",
3168 			    vdcp->instance, mhdl, i, rv);
3169 			if (ldep->align_addr) {
3170 				kmem_free(ldep->align_addr,
3171 				    sizeof (caddr_t) * ldep->nbytes);
3172 				ldep->align_addr = NULL;
3173 			}
3174 			return (EAGAIN);
3175 		}
3176 	}
3177 
3178 	return (rv);
3179 }
3180 
3181 /*
3182  * Interrupt handlers for messages from LDC
3183  */
3184 
3185 /*
3186  * Function:
3187  *	vdc_handle_cb()
3188  *
3189  * Description:
3190  *
3191  * Arguments:
3192  *	event	- Type of event (LDC_EVT_xxx) that triggered the callback
3193  *	arg	- soft state pointer for this instance of the device driver.
3194  *
3195  * Return Code:
3196  *	0	- Success
3197  */
3198 static uint_t
3199 vdc_handle_cb(uint64_t event, caddr_t arg)
3200 {
3201 	ldc_status_t	ldc_state;
3202 	int		rv = 0;
3203 
3204 	vdc_t	*vdc = (vdc_t *)(void *)arg;
3205 
3206 	ASSERT(vdc != NULL);
3207 
3208 	DMSG(vdc, 1, "evt=%lx seqID=%ld\n", event, vdc->seq_num);
3209 
3210 	/*
3211 	 * Depending on the type of event that triggered this callback,
3212 	 * we modify the handshake state or read the data.
3213 	 *
3214 	 * NOTE: not done as a switch() as event could be triggered by
3215 	 * a state change and a read request. Also the ordering	of the
3216 	 * check for the event types is deliberate.
3217 	 */
3218 	if (event & LDC_EVT_UP) {
3219 		DMSG(vdc, 0, "[%d] Received LDC_EVT_UP\n", vdc->instance);
3220 
3221 		mutex_enter(&vdc->lock);
3222 
3223 		/* get LDC state */
3224 		rv = ldc_status(vdc->ldc_handle, &ldc_state);
3225 		if (rv != 0) {
3226 			DMSG(vdc, 0, "[%d] Couldn't get LDC status %d",
3227 			    vdc->instance, rv);
3228 			return (LDC_SUCCESS);
3229 		}
3230 		if (vdc->ldc_state != LDC_UP && ldc_state == LDC_UP) {
3231 			/*
3232 			 * Reset the transaction sequence numbers when
3233 			 * LDC comes up. We then kick off the handshake
3234 			 * negotiation with the vDisk server.
3235 			 */
3236 			vdc->seq_num = 1;
3237 			vdc->seq_num_reply = 0;
3238 			vdc->ldc_state = ldc_state;
3239 			cv_signal(&vdc->initwait_cv);
3240 		}
3241 
3242 		mutex_exit(&vdc->lock);
3243 	}
3244 
3245 	if (event & LDC_EVT_READ) {
3246 		DMSG(vdc, 0, "[%d] Received LDC_EVT_READ\n", vdc->instance);
3247 		mutex_enter(&vdc->read_lock);
3248 		cv_signal(&vdc->read_cv);
3249 		vdc->read_state = VDC_READ_PENDING;
3250 		mutex_exit(&vdc->read_lock);
3251 
3252 		/* that's all we have to do - no need to handle DOWN/RESET */
3253 		return (LDC_SUCCESS);
3254 	}
3255 
3256 	if (event & (LDC_EVT_RESET|LDC_EVT_DOWN)) {
3257 
3258 		DMSG(vdc, 0, "[%d] Received LDC RESET event\n", vdc->instance);
3259 
3260 		mutex_enter(&vdc->lock);
3261 		/*
3262 		 * Need to wake up any readers so they will
3263 		 * detect that a reset has occurred.
3264 		 */
3265 		mutex_enter(&vdc->read_lock);
3266 		if ((vdc->read_state == VDC_READ_WAITING) ||
3267 		    (vdc->read_state == VDC_READ_RESET))
3268 			cv_signal(&vdc->read_cv);
3269 		vdc->read_state = VDC_READ_RESET;
3270 		mutex_exit(&vdc->read_lock);
3271 
3272 		/* wake up any threads waiting for connection to come up */
3273 		if (vdc->state == VDC_STATE_INIT_WAITING) {
3274 			vdc->state = VDC_STATE_RESETTING;
3275 			cv_signal(&vdc->initwait_cv);
3276 		}
3277 
3278 		mutex_exit(&vdc->lock);
3279 	}
3280 
3281 	if (event & ~(LDC_EVT_UP | LDC_EVT_RESET | LDC_EVT_DOWN | LDC_EVT_READ))
3282 		DMSG(vdc, 0, "![%d] Unexpected LDC event (%lx) received",
3283 		    vdc->instance, event);
3284 
3285 	return (LDC_SUCCESS);
3286 }
3287 
3288 /*
3289  * Function:
3290  *	vdc_wait_for_response()
3291  *
3292  * Description:
3293  *	Block waiting for a response from the server. If there is
3294  *	no data the thread block on the read_cv that is signalled
3295  *	by the callback when an EVT_READ occurs.
3296  *
3297  * Arguments:
3298  *	vdcp	- soft state pointer for this instance of the device driver.
3299  *
3300  * Return Code:
3301  *	0	- Success
3302  */
3303 static int
3304 vdc_wait_for_response(vdc_t *vdcp, vio_msg_t *msgp)
3305 {
3306 	size_t		nbytes = sizeof (*msgp);
3307 	int		status;
3308 
3309 	ASSERT(vdcp != NULL);
3310 
3311 	DMSG(vdcp, 1, "[%d] Entered\n", vdcp->instance);
3312 
3313 	status = vdc_recv(vdcp, msgp, &nbytes);
3314 	DMSG(vdcp, 3, "vdc_read() done.. status=0x%x size=0x%x\n",
3315 	    status, (int)nbytes);
3316 	if (status) {
3317 		DMSG(vdcp, 0, "?[%d] Error %d reading LDC msg\n",
3318 		    vdcp->instance, status);
3319 		return (status);
3320 	}
3321 
3322 	if (nbytes < sizeof (vio_msg_tag_t)) {
3323 		DMSG(vdcp, 0, "?[%d] Expect %lu bytes; recv'd %lu\n",
3324 		    vdcp->instance, sizeof (vio_msg_tag_t), nbytes);
3325 		return (ENOMSG);
3326 	}
3327 
3328 	DMSG(vdcp, 2, "[%d] (%x/%x/%x)\n", vdcp->instance,
3329 	    msgp->tag.vio_msgtype,
3330 	    msgp->tag.vio_subtype,
3331 	    msgp->tag.vio_subtype_env);
3332 
3333 	/*
3334 	 * Verify the Session ID of the message
3335 	 *
3336 	 * Every message after the Version has been negotiated should
3337 	 * have the correct session ID set.
3338 	 */
3339 	if ((msgp->tag.vio_sid != vdcp->session_id) &&
3340 	    (msgp->tag.vio_subtype_env != VIO_VER_INFO)) {
3341 		DMSG(vdcp, 0, "[%d] Invalid SID: received 0x%x, "
3342 		    "expected 0x%lx [seq num %lx @ %d]",
3343 		    vdcp->instance, msgp->tag.vio_sid,
3344 		    vdcp->session_id,
3345 		    ((vio_dring_msg_t *)msgp)->seq_num,
3346 		    ((vio_dring_msg_t *)msgp)->start_idx);
3347 		return (ENOMSG);
3348 	}
3349 	return (0);
3350 }
3351 
3352 
3353 /*
3354  * Function:
3355  *	vdc_resubmit_backup_dring()
3356  *
3357  * Description:
3358  *	Resubmit each descriptor in the backed up dring to
3359  * 	vDisk server. The Dring was backed up during connection
3360  *	reset.
3361  *
3362  * Arguments:
3363  *	vdcp	- soft state pointer for this instance of the device driver.
3364  *
3365  * Return Code:
3366  *	0	- Success
3367  */
3368 static int
3369 vdc_resubmit_backup_dring(vdc_t *vdcp)
3370 {
3371 	int		count;
3372 	int		b_idx;
3373 	int		rv;
3374 	int		dring_size;
3375 	int		status;
3376 	vio_msg_t	vio_msg;
3377 	vdc_local_desc_t	*curr_ldep;
3378 
3379 	ASSERT(MUTEX_NOT_HELD(&vdcp->lock));
3380 	ASSERT(vdcp->state == VDC_STATE_HANDLE_PENDING);
3381 
3382 	if (vdcp->local_dring_backup == NULL) {
3383 		/* the pending requests have already been processed */
3384 		return (0);
3385 	}
3386 
3387 	DMSG(vdcp, 1, "restoring pending dring entries (len=%d, tail=%d)\n",
3388 	    vdcp->local_dring_backup_len, vdcp->local_dring_backup_tail);
3389 
3390 	/*
3391 	 * Walk the backup copy of the local descriptor ring and
3392 	 * resubmit all the outstanding transactions.
3393 	 */
3394 	b_idx = vdcp->local_dring_backup_tail;
3395 	for (count = 0; count < vdcp->local_dring_backup_len; count++) {
3396 
3397 		curr_ldep = &(vdcp->local_dring_backup[b_idx]);
3398 
3399 		/* only resubmit outstanding transactions */
3400 		if (!curr_ldep->is_free) {
3401 
3402 			DMSG(vdcp, 1, "resubmitting entry idx=%x\n", b_idx);
3403 			mutex_enter(&vdcp->lock);
3404 			rv = vdc_populate_descriptor(vdcp, curr_ldep->operation,
3405 			    curr_ldep->addr, curr_ldep->nbytes,
3406 			    curr_ldep->slice, curr_ldep->offset,
3407 			    curr_ldep->cb_type, curr_ldep->cb_arg,
3408 			    curr_ldep->dir);
3409 			mutex_exit(&vdcp->lock);
3410 			if (rv) {
3411 				DMSG(vdcp, 1, "[%d] cannot resubmit entry %d\n",
3412 				    vdcp->instance, b_idx);
3413 				return (rv);
3414 			}
3415 
3416 			/* Wait for the response message. */
3417 			DMSG(vdcp, 1, "waiting for response to idx=%x\n",
3418 			    b_idx);
3419 			status = vdc_wait_for_response(vdcp, &vio_msg);
3420 			if (status) {
3421 				DMSG(vdcp, 1, "[%d] wait_for_response "
3422 				    "returned err=%d\n", vdcp->instance,
3423 				    status);
3424 				return (status);
3425 			}
3426 
3427 			DMSG(vdcp, 1, "processing msg for idx=%x\n", b_idx);
3428 			status = vdc_process_data_msg(vdcp, &vio_msg);
3429 			if (status) {
3430 				DMSG(vdcp, 1, "[%d] process_data_msg "
3431 				    "returned err=%d\n", vdcp->instance,
3432 				    status);
3433 				return (status);
3434 			}
3435 		}
3436 
3437 		/* get the next element to submit */
3438 		if (++b_idx >= vdcp->local_dring_backup_len)
3439 			b_idx = 0;
3440 	}
3441 
3442 	/* all done - now clear up pending dring copy */
3443 	dring_size = vdcp->local_dring_backup_len *
3444 	    sizeof (vdcp->local_dring_backup[0]);
3445 
3446 	(void) kmem_free(vdcp->local_dring_backup, dring_size);
3447 
3448 	vdcp->local_dring_backup = NULL;
3449 
3450 	return (0);
3451 }
3452 
3453 /*
3454  * Function:
3455  *	vdc_cancel_backup_dring
3456  *
3457  * Description:
3458  *	Cancel each descriptor in the backed up dring to vDisk server.
3459  *	The Dring was backed up during connection reset.
3460  *
3461  * Arguments:
3462  *	vdcp	- soft state pointer for this instance of the device driver.
3463  *
3464  * Return Code:
3465  *	None
3466  */
3467 void
3468 vdc_cancel_backup_ring(vdc_t *vdcp)
3469 {
3470 	vdc_local_desc_t *ldep;
3471 	struct buf 	*bufp;
3472 	int		count;
3473 	int		b_idx;
3474 	int		dring_size;
3475 
3476 	ASSERT(MUTEX_HELD(&vdcp->lock));
3477 	ASSERT(vdcp->state == VDC_STATE_INIT ||
3478 	    vdcp->state == VDC_STATE_INIT_WAITING ||
3479 	    vdcp->state == VDC_STATE_NEGOTIATE ||
3480 	    vdcp->state == VDC_STATE_RESETTING);
3481 
3482 	if (vdcp->local_dring_backup == NULL) {
3483 		/* the pending requests have already been processed */
3484 		return;
3485 	}
3486 
3487 	DMSG(vdcp, 1, "cancelling pending dring entries (len=%d, tail=%d)\n",
3488 	    vdcp->local_dring_backup_len, vdcp->local_dring_backup_tail);
3489 
3490 	/*
3491 	 * Walk the backup copy of the local descriptor ring and
3492 	 * cancel all the outstanding transactions.
3493 	 */
3494 	b_idx = vdcp->local_dring_backup_tail;
3495 	for (count = 0; count < vdcp->local_dring_backup_len; count++) {
3496 
3497 		ldep = &(vdcp->local_dring_backup[b_idx]);
3498 
3499 		/* only cancel outstanding transactions */
3500 		if (!ldep->is_free) {
3501 
3502 			DMSG(vdcp, 1, "cancelling entry idx=%x\n", b_idx);
3503 
3504 			/*
3505 			 * All requests have already been cleared from the
3506 			 * local descriptor ring and the LDC channel has been
3507 			 * reset so we will never get any reply for these
3508 			 * requests. Now we just have to notify threads waiting
3509 			 * for replies that the request has failed.
3510 			 */
3511 			switch (ldep->cb_type) {
3512 			case CB_SYNC:
3513 				ASSERT(vdcp->sync_op_pending);
3514 				vdcp->sync_op_status = EIO;
3515 				vdcp->sync_op_pending = B_FALSE;
3516 				cv_signal(&vdcp->sync_pending_cv);
3517 				break;
3518 
3519 			case CB_STRATEGY:
3520 				bufp = ldep->cb_arg;
3521 				ASSERT(bufp != NULL);
3522 				bufp->b_resid = bufp->b_bcount;
3523 				bioerror(bufp, EIO);
3524 				biodone(bufp);
3525 				break;
3526 
3527 			default:
3528 				ASSERT(0);
3529 			}
3530 
3531 		}
3532 
3533 		/* get the next element to cancel */
3534 		if (++b_idx >= vdcp->local_dring_backup_len)
3535 			b_idx = 0;
3536 	}
3537 
3538 	/* all done - now clear up pending dring copy */
3539 	dring_size = vdcp->local_dring_backup_len *
3540 	    sizeof (vdcp->local_dring_backup[0]);
3541 
3542 	(void) kmem_free(vdcp->local_dring_backup, dring_size);
3543 
3544 	vdcp->local_dring_backup = NULL;
3545 
3546 	DTRACE_IO2(processed, int, count, vdc_t *, vdcp);
3547 }
3548 
3549 /*
3550  * Function:
3551  *	vdc_connection_timeout
3552  *
3553  * Description:
3554  *	This function is invoked if the timeout set to establish the connection
3555  *	with vds expires. This will happen if we spend too much time in the
3556  *	VDC_STATE_INIT_WAITING or VDC_STATE_NEGOTIATE states. Then we will
3557  *	cancel any pending request and mark them as failed.
3558  *
3559  *	If the timeout does not expire, it will be cancelled when we reach the
3560  *	VDC_STATE_HANDLE_PENDING or VDC_STATE_RESETTING state. This function can
3561  *	be invoked while we are in the VDC_STATE_HANDLE_PENDING or
3562  *	VDC_STATE_RESETTING state in which case we do nothing because the
3563  *	timeout is being cancelled.
3564  *
3565  * Arguments:
3566  *	arg	- argument of the timeout function actually a soft state
3567  *		  pointer for the instance of the device driver.
3568  *
3569  * Return Code:
3570  *	None
3571  */
3572 void
3573 vdc_connection_timeout(void *arg)
3574 {
3575 	vdc_t 		*vdcp = (vdc_t *)arg;
3576 
3577 	mutex_enter(&vdcp->lock);
3578 
3579 	if (vdcp->state == VDC_STATE_HANDLE_PENDING ||
3580 	    vdcp->state == VDC_STATE_DETACH) {
3581 		/*
3582 		 * The connection has just been re-established or
3583 		 * we are detaching.
3584 		 */
3585 		vdcp->ctimeout_reached = B_FALSE;
3586 		mutex_exit(&vdcp->lock);
3587 		return;
3588 	}
3589 
3590 	vdcp->ctimeout_reached = B_TRUE;
3591 
3592 	/* notify requests waiting for sending */
3593 	cv_broadcast(&vdcp->running_cv);
3594 
3595 	/* cancel requests waiting for a result */
3596 	vdc_cancel_backup_ring(vdcp);
3597 
3598 	mutex_exit(&vdcp->lock);
3599 
3600 	cmn_err(CE_NOTE, "[%d] connection to service domain timeout",
3601 	    vdcp->instance);
3602 }
3603 
3604 /*
3605  * Function:
3606  *	vdc_backup_local_dring()
3607  *
3608  * Description:
3609  *	Backup the current dring in the event of a reset. The Dring
3610  *	transactions will be resubmitted to the server when the
3611  *	connection is restored.
3612  *
3613  * Arguments:
3614  *	vdcp	- soft state pointer for this instance of the device driver.
3615  *
3616  * Return Code:
3617  *	NONE
3618  */
3619 static void
3620 vdc_backup_local_dring(vdc_t *vdcp)
3621 {
3622 	int dring_size;
3623 
3624 	ASSERT(MUTEX_HELD(&vdcp->lock));
3625 	ASSERT(vdcp->state == VDC_STATE_RESETTING);
3626 
3627 	/*
3628 	 * If the backup dring is stil around, it means
3629 	 * that the last restore did not complete. However,
3630 	 * since we never got back into the running state,
3631 	 * the backup copy we have is still valid.
3632 	 */
3633 	if (vdcp->local_dring_backup != NULL) {
3634 		DMSG(vdcp, 1, "reusing local descriptor ring backup "
3635 		    "(len=%d, tail=%d)\n", vdcp->local_dring_backup_len,
3636 		    vdcp->local_dring_backup_tail);
3637 		return;
3638 	}
3639 
3640 	/*
3641 	 * The backup dring can be NULL and the local dring may not be
3642 	 * initialized. This can happen if we had a reset while establishing
3643 	 * a new connection but after the connection has timed out. In that
3644 	 * case the backup dring is NULL because the requests have been
3645 	 * cancelled and the request occured before the local dring is
3646 	 * initialized.
3647 	 */
3648 	if (!(vdcp->initialized & VDC_DRING_LOCAL))
3649 		return;
3650 
3651 	DMSG(vdcp, 1, "backing up the local descriptor ring (len=%d, "
3652 	    "tail=%d)\n", vdcp->dring_len, vdcp->dring_curr_idx);
3653 
3654 	dring_size = vdcp->dring_len * sizeof (vdcp->local_dring[0]);
3655 
3656 	vdcp->local_dring_backup = kmem_alloc(dring_size, KM_SLEEP);
3657 	bcopy(vdcp->local_dring, vdcp->local_dring_backup, dring_size);
3658 
3659 	vdcp->local_dring_backup_tail = vdcp->dring_curr_idx;
3660 	vdcp->local_dring_backup_len = vdcp->dring_len;
3661 }
3662 
3663 /* -------------------------------------------------------------------------- */
3664 
3665 /*
3666  * The following functions process the incoming messages from vds
3667  */
3668 
3669 /*
3670  * Function:
3671  *      vdc_process_msg_thread()
3672  *
3673  * Description:
3674  *
3675  *	Main VDC message processing thread. Each vDisk instance
3676  * 	consists of a copy of this thread. This thread triggers
3677  * 	all the handshakes and data exchange with the server. It
3678  * 	also handles all channel resets
3679  *
3680  * Arguments:
3681  *      vdc     - soft state pointer for this instance of the device driver.
3682  *
3683  * Return Code:
3684  *      None
3685  */
3686 static void
3687 vdc_process_msg_thread(vdc_t *vdcp)
3688 {
3689 	int	status;
3690 	int	ctimeout;
3691 	timeout_id_t tmid = 0;
3692 
3693 	mutex_enter(&vdcp->lock);
3694 
3695 	for (;;) {
3696 
3697 #define	Q(_s)	(vdcp->state == _s) ? #_s :
3698 		DMSG(vdcp, 3, "state = %d (%s)\n", vdcp->state,
3699 		    Q(VDC_STATE_INIT)
3700 		    Q(VDC_STATE_INIT_WAITING)
3701 		    Q(VDC_STATE_NEGOTIATE)
3702 		    Q(VDC_STATE_HANDLE_PENDING)
3703 		    Q(VDC_STATE_RUNNING)
3704 		    Q(VDC_STATE_RESETTING)
3705 		    Q(VDC_STATE_DETACH)
3706 		    "UNKNOWN");
3707 
3708 		switch (vdcp->state) {
3709 		case VDC_STATE_INIT:
3710 
3711 			/*
3712 			 * If requested, start a timeout to check if the
3713 			 * connection with vds is established in the
3714 			 * specified delay. If the timeout expires, we
3715 			 * will cancel any pending request.
3716 			 *
3717 			 * If some reset have occurred while establishing
3718 			 * the connection, we already have a timeout armed
3719 			 * and in that case we don't need to arm a new one.
3720 			 */
3721 			ctimeout = (vdc_timeout != 0)?
3722 			    vdc_timeout : vdcp->ctimeout;
3723 
3724 			if (ctimeout != 0 && tmid == 0) {
3725 				tmid = timeout(vdc_connection_timeout, vdcp,
3726 				    ctimeout * drv_usectohz(1000000));
3727 			}
3728 
3729 			/* Check if have re-initializing repeatedly */
3730 			if (vdcp->hshake_cnt++ > vdc_hshake_retries &&
3731 			    vdcp->lifecycle != VDC_LC_ONLINE) {
3732 				cmn_err(CE_NOTE, "[%d] disk access failed.\n",
3733 				    vdcp->instance);
3734 				vdcp->state = VDC_STATE_DETACH;
3735 				break;
3736 			}
3737 
3738 			/* Bring up connection with vds via LDC */
3739 			status = vdc_start_ldc_connection(vdcp);
3740 			if (status == EINVAL) {
3741 				DMSG(vdcp, 0, "[%d] Could not start LDC",
3742 				    vdcp->instance);
3743 				vdcp->state = VDC_STATE_DETACH;
3744 			} else {
3745 				vdcp->state = VDC_STATE_INIT_WAITING;
3746 			}
3747 			break;
3748 
3749 		case VDC_STATE_INIT_WAITING:
3750 
3751 			/*
3752 			 * Let the callback event move us on
3753 			 * when channel is open to server
3754 			 */
3755 			while (vdcp->ldc_state != LDC_UP) {
3756 				cv_wait(&vdcp->initwait_cv, &vdcp->lock);
3757 				if (vdcp->state != VDC_STATE_INIT_WAITING) {
3758 					DMSG(vdcp, 0,
3759 				"state moved to %d out from under us...\n",
3760 					    vdcp->state);
3761 
3762 					break;
3763 				}
3764 			}
3765 			if (vdcp->state == VDC_STATE_INIT_WAITING &&
3766 			    vdcp->ldc_state == LDC_UP) {
3767 				vdcp->state = VDC_STATE_NEGOTIATE;
3768 			}
3769 			break;
3770 
3771 		case VDC_STATE_NEGOTIATE:
3772 			switch (status = vdc_ver_negotiation(vdcp)) {
3773 			case 0:
3774 				break;
3775 			default:
3776 				DMSG(vdcp, 0, "ver negotiate failed (%d)..\n",
3777 				    status);
3778 				goto reset;
3779 			}
3780 
3781 			switch (status = vdc_attr_negotiation(vdcp)) {
3782 			case 0:
3783 				break;
3784 			default:
3785 				DMSG(vdcp, 0, "attr negotiate failed (%d)..\n",
3786 				    status);
3787 				goto reset;
3788 			}
3789 
3790 			switch (status = vdc_dring_negotiation(vdcp)) {
3791 			case 0:
3792 				break;
3793 			default:
3794 				DMSG(vdcp, 0, "dring negotiate failed (%d)..\n",
3795 				    status);
3796 				goto reset;
3797 			}
3798 
3799 			switch (status = vdc_rdx_exchange(vdcp)) {
3800 			case 0:
3801 				vdcp->state = VDC_STATE_HANDLE_PENDING;
3802 				goto done;
3803 			default:
3804 				DMSG(vdcp, 0, "RDX xchg failed ..(%d)\n",
3805 				    status);
3806 				goto reset;
3807 			}
3808 reset:
3809 			DMSG(vdcp, 0, "negotiation failed: resetting (%d)\n",
3810 			    status);
3811 			vdcp->state = VDC_STATE_RESETTING;
3812 			vdcp->self_reset = B_TRUE;
3813 done:
3814 			DMSG(vdcp, 0, "negotiation complete (state=0x%x)...\n",
3815 			    vdcp->state);
3816 			break;
3817 
3818 		case VDC_STATE_HANDLE_PENDING:
3819 
3820 			if (vdcp->ctimeout_reached) {
3821 				/*
3822 				 * The connection timeout had been reached so
3823 				 * pending requests have been cancelled. Now
3824 				 * that the connection is back we can reset
3825 				 * the timeout.
3826 				 */
3827 				ASSERT(vdcp->local_dring_backup == NULL);
3828 				ASSERT(tmid != 0);
3829 				tmid = 0;
3830 				vdcp->ctimeout_reached = B_FALSE;
3831 				vdcp->state = VDC_STATE_RUNNING;
3832 				DMSG(vdcp, 0, "[%d] connection to service "
3833 				    "domain is up", vdcp->instance);
3834 				break;
3835 			}
3836 
3837 			mutex_exit(&vdcp->lock);
3838 			if (tmid != 0) {
3839 				(void) untimeout(tmid);
3840 				tmid = 0;
3841 			}
3842 			status = vdc_resubmit_backup_dring(vdcp);
3843 			mutex_enter(&vdcp->lock);
3844 
3845 			if (status)
3846 				vdcp->state = VDC_STATE_RESETTING;
3847 			else
3848 				vdcp->state = VDC_STATE_RUNNING;
3849 
3850 			break;
3851 
3852 		/* enter running state */
3853 		case VDC_STATE_RUNNING:
3854 			/*
3855 			 * Signal anyone waiting for the connection
3856 			 * to come on line.
3857 			 */
3858 			vdcp->hshake_cnt = 0;
3859 			cv_broadcast(&vdcp->running_cv);
3860 			mutex_exit(&vdcp->lock);
3861 
3862 			for (;;) {
3863 				vio_msg_t msg;
3864 				status = vdc_wait_for_response(vdcp, &msg);
3865 				if (status) break;
3866 
3867 				DMSG(vdcp, 1, "[%d] new pkt(s) available\n",
3868 				    vdcp->instance);
3869 				status = vdc_process_data_msg(vdcp, &msg);
3870 				if (status) {
3871 					DMSG(vdcp, 1, "[%d] process_data_msg "
3872 					    "returned err=%d\n", vdcp->instance,
3873 					    status);
3874 					break;
3875 				}
3876 
3877 			}
3878 
3879 			mutex_enter(&vdcp->lock);
3880 
3881 			vdcp->state = VDC_STATE_RESETTING;
3882 			vdcp->self_reset = B_TRUE;
3883 			break;
3884 
3885 		case VDC_STATE_RESETTING:
3886 			/*
3887 			 * When we reach this state, we either come from the
3888 			 * VDC_STATE_RUNNING state and we can have pending
3889 			 * request but no timeout is armed; or we come from
3890 			 * the VDC_STATE_INIT_WAITING, VDC_NEGOTIATE or
3891 			 * VDC_HANDLE_PENDING state and there is no pending
3892 			 * request or pending requests have already been copied
3893 			 * into the backup dring. So we can safely keep the
3894 			 * connection timeout armed while we are in this state.
3895 			 */
3896 
3897 			DMSG(vdcp, 0, "Initiating channel reset "
3898 			    "(pending = %d)\n", (int)vdcp->threads_pending);
3899 
3900 			if (vdcp->self_reset) {
3901 				DMSG(vdcp, 0,
3902 				    "[%d] calling stop_ldc_connection.\n",
3903 				    vdcp->instance);
3904 				status = vdc_stop_ldc_connection(vdcp);
3905 				vdcp->self_reset = B_FALSE;
3906 			}
3907 
3908 			/*
3909 			 * Wait for all threads currently waiting
3910 			 * for a free dring entry to use.
3911 			 */
3912 			while (vdcp->threads_pending) {
3913 				cv_broadcast(&vdcp->membind_cv);
3914 				cv_broadcast(&vdcp->dring_free_cv);
3915 				mutex_exit(&vdcp->lock);
3916 				/* give the waiters enough time to wake up */
3917 				delay(vdc_hz_min_ldc_delay);
3918 				mutex_enter(&vdcp->lock);
3919 			}
3920 
3921 			ASSERT(vdcp->threads_pending == 0);
3922 
3923 			/* Sanity check that no thread is receiving */
3924 			ASSERT(vdcp->read_state != VDC_READ_WAITING);
3925 
3926 			vdcp->read_state = VDC_READ_IDLE;
3927 
3928 			vdc_backup_local_dring(vdcp);
3929 
3930 			/* cleanup the old d-ring */
3931 			vdc_destroy_descriptor_ring(vdcp);
3932 
3933 			/* go and start again */
3934 			vdcp->state = VDC_STATE_INIT;
3935 
3936 			break;
3937 
3938 		case VDC_STATE_DETACH:
3939 			DMSG(vdcp, 0, "[%d] Reset thread exit cleanup ..\n",
3940 			    vdcp->instance);
3941 
3942 			/* cancel any pending timeout */
3943 			mutex_exit(&vdcp->lock);
3944 			if (tmid != 0) {
3945 				(void) untimeout(tmid);
3946 				tmid = 0;
3947 			}
3948 			mutex_enter(&vdcp->lock);
3949 
3950 			/*
3951 			 * Signal anyone waiting for connection
3952 			 * to come online
3953 			 */
3954 			cv_broadcast(&vdcp->running_cv);
3955 
3956 			while (vdcp->sync_op_pending) {
3957 				cv_signal(&vdcp->sync_pending_cv);
3958 				cv_signal(&vdcp->sync_blocked_cv);
3959 				mutex_exit(&vdcp->lock);
3960 				/* give the waiters enough time to wake up */
3961 				delay(vdc_hz_min_ldc_delay);
3962 				mutex_enter(&vdcp->lock);
3963 			}
3964 
3965 			mutex_exit(&vdcp->lock);
3966 
3967 			DMSG(vdcp, 0, "[%d] Msg processing thread exiting ..\n",
3968 			    vdcp->instance);
3969 			thread_exit();
3970 			break;
3971 		}
3972 	}
3973 }
3974 
3975 
3976 /*
3977  * Function:
3978  *	vdc_process_data_msg()
3979  *
3980  * Description:
3981  *	This function is called by the message processing thread each time
3982  *	a message with a msgtype of VIO_TYPE_DATA is received. It will either
3983  *	be an ACK or NACK from vds[1] which vdc handles as follows.
3984  *		ACK	- wake up the waiting thread
3985  *		NACK	- resend any messages necessary
3986  *
3987  *	[1] Although the message format allows it, vds should not send a
3988  *	    VIO_SUBTYPE_INFO message to vdc asking it to read data; if for
3989  *	    some bizarre reason it does, vdc will reset the connection.
3990  *
3991  * Arguments:
3992  *	vdc	- soft state pointer for this instance of the device driver.
3993  *	msg	- the LDC message sent by vds
3994  *
3995  * Return Code:
3996  *	0	- Success.
3997  *	> 0	- error value returned by LDC
3998  */
3999 static int
4000 vdc_process_data_msg(vdc_t *vdcp, vio_msg_t *msg)
4001 {
4002 	int			status = 0;
4003 	vio_dring_msg_t		*dring_msg;
4004 	vdc_local_desc_t	*ldep = NULL;
4005 	int			start, end;
4006 	int			idx;
4007 
4008 	dring_msg = (vio_dring_msg_t *)msg;
4009 
4010 	ASSERT(msg->tag.vio_msgtype == VIO_TYPE_DATA);
4011 	ASSERT(vdcp != NULL);
4012 
4013 	mutex_enter(&vdcp->lock);
4014 
4015 	/*
4016 	 * Check to see if the message has bogus data
4017 	 */
4018 	idx = start = dring_msg->start_idx;
4019 	end = dring_msg->end_idx;
4020 	if ((start >= vdcp->dring_len) ||
4021 	    (end >= vdcp->dring_len) || (end < -1)) {
4022 		DMSG(vdcp, 0, "[%d] Bogus ACK data : start %d, end %d\n",
4023 		    vdcp->instance, start, end);
4024 		mutex_exit(&vdcp->lock);
4025 		return (EINVAL);
4026 	}
4027 
4028 	/*
4029 	 * Verify that the sequence number is what vdc expects.
4030 	 */
4031 	switch (vdc_verify_seq_num(vdcp, dring_msg)) {
4032 	case VDC_SEQ_NUM_TODO:
4033 		break;	/* keep processing this message */
4034 	case VDC_SEQ_NUM_SKIP:
4035 		mutex_exit(&vdcp->lock);
4036 		return (0);
4037 	case VDC_SEQ_NUM_INVALID:
4038 		mutex_exit(&vdcp->lock);
4039 		DMSG(vdcp, 0, "[%d] invalid seqno\n", vdcp->instance);
4040 		return (ENXIO);
4041 	}
4042 
4043 	if (msg->tag.vio_subtype == VIO_SUBTYPE_NACK) {
4044 		DMSG(vdcp, 0, "[%d] DATA NACK\n", vdcp->instance);
4045 		VDC_DUMP_DRING_MSG(dring_msg);
4046 		mutex_exit(&vdcp->lock);
4047 		return (EIO);
4048 
4049 	} else if (msg->tag.vio_subtype == VIO_SUBTYPE_INFO) {
4050 		mutex_exit(&vdcp->lock);
4051 		return (EPROTO);
4052 	}
4053 
4054 	DTRACE_IO2(recv, vio_dring_msg_t, dring_msg, vdc_t *, vdcp);
4055 	DMSG(vdcp, 1, ": start %d end %d\n", start, end);
4056 	ASSERT(start == end);
4057 
4058 	ldep = &vdcp->local_dring[idx];
4059 
4060 	DMSG(vdcp, 1, ": state 0x%x - cb_type 0x%x\n",
4061 	    ldep->dep->hdr.dstate, ldep->cb_type);
4062 
4063 	if (ldep->dep->hdr.dstate == VIO_DESC_DONE) {
4064 		struct buf *bufp;
4065 
4066 		switch (ldep->cb_type) {
4067 		case CB_SYNC:
4068 			ASSERT(vdcp->sync_op_pending);
4069 
4070 			status = vdc_depopulate_descriptor(vdcp, idx);
4071 			vdcp->sync_op_status = status;
4072 			vdcp->sync_op_pending = B_FALSE;
4073 			cv_signal(&vdcp->sync_pending_cv);
4074 			break;
4075 
4076 		case CB_STRATEGY:
4077 			bufp = ldep->cb_arg;
4078 			ASSERT(bufp != NULL);
4079 			bufp->b_resid =
4080 			    bufp->b_bcount - ldep->dep->payload.nbytes;
4081 			status = ldep->dep->payload.status; /* Future:ntoh */
4082 			if (status != 0) {
4083 				DMSG(vdcp, 1, "strategy status=%d\n", status);
4084 				bioerror(bufp, status);
4085 			}
4086 			status = vdc_depopulate_descriptor(vdcp, idx);
4087 			biodone(bufp);
4088 
4089 			DMSG(vdcp, 1,
4090 			    "strategy complete req=%ld bytes resp=%ld bytes\n",
4091 			    bufp->b_bcount, ldep->dep->payload.nbytes);
4092 			break;
4093 
4094 		default:
4095 			ASSERT(0);
4096 		}
4097 	}
4098 
4099 	/* let the arrival signal propogate */
4100 	mutex_exit(&vdcp->lock);
4101 
4102 	/* probe gives the count of how many entries were processed */
4103 	DTRACE_IO2(processed, int, 1, vdc_t *, vdcp);
4104 
4105 	return (0);
4106 }
4107 
4108 /*
4109  * Function:
4110  *	vdc_process_err_msg()
4111  *
4112  * NOTE: No error messages are used as part of the vDisk protocol
4113  */
4114 static int
4115 vdc_process_err_msg(vdc_t *vdc, vio_msg_t msg)
4116 {
4117 	_NOTE(ARGUNUSED(vdc))
4118 	_NOTE(ARGUNUSED(msg))
4119 
4120 	ASSERT(msg.tag.vio_msgtype == VIO_TYPE_ERR);
4121 	DMSG(vdc, 1, "[%d] Got an ERR msg", vdc->instance);
4122 
4123 	return (ENOTSUP);
4124 }
4125 
4126 /*
4127  * Function:
4128  *	vdc_handle_ver_msg()
4129  *
4130  * Description:
4131  *
4132  * Arguments:
4133  *	vdc	- soft state pointer for this instance of the device driver.
4134  *	ver_msg	- LDC message sent by vDisk server
4135  *
4136  * Return Code:
4137  *	0	- Success
4138  */
4139 static int
4140 vdc_handle_ver_msg(vdc_t *vdc, vio_ver_msg_t *ver_msg)
4141 {
4142 	int status = 0;
4143 
4144 	ASSERT(vdc != NULL);
4145 	ASSERT(mutex_owned(&vdc->lock));
4146 
4147 	if (ver_msg->tag.vio_subtype_env != VIO_VER_INFO) {
4148 		return (EPROTO);
4149 	}
4150 
4151 	if (ver_msg->dev_class != VDEV_DISK_SERVER) {
4152 		return (EINVAL);
4153 	}
4154 
4155 	switch (ver_msg->tag.vio_subtype) {
4156 	case VIO_SUBTYPE_ACK:
4157 		/*
4158 		 * We check to see if the version returned is indeed supported
4159 		 * (The server may have also adjusted the minor number downwards
4160 		 * and if so 'ver_msg' will contain the actual version agreed)
4161 		 */
4162 		if (vdc_is_supported_version(ver_msg)) {
4163 			vdc->ver.major = ver_msg->ver_major;
4164 			vdc->ver.minor = ver_msg->ver_minor;
4165 			ASSERT(vdc->ver.major > 0);
4166 		} else {
4167 			status = EPROTO;
4168 		}
4169 		break;
4170 
4171 	case VIO_SUBTYPE_NACK:
4172 		/*
4173 		 * call vdc_is_supported_version() which will return the next
4174 		 * supported version (if any) in 'ver_msg'
4175 		 */
4176 		(void) vdc_is_supported_version(ver_msg);
4177 		if (ver_msg->ver_major > 0) {
4178 			size_t len = sizeof (*ver_msg);
4179 
4180 			ASSERT(vdc->ver.major > 0);
4181 
4182 			/* reset the necessary fields and resend */
4183 			ver_msg->tag.vio_subtype = VIO_SUBTYPE_INFO;
4184 			ver_msg->dev_class = VDEV_DISK;
4185 
4186 			status = vdc_send(vdc, (caddr_t)ver_msg, &len);
4187 			DMSG(vdc, 0, "[%d] Resend VER info (LDC status = %d)\n",
4188 			    vdc->instance, status);
4189 			if (len != sizeof (*ver_msg))
4190 				status = EBADMSG;
4191 		} else {
4192 			DMSG(vdc, 0, "[%d] No common version with vDisk server",
4193 			    vdc->instance);
4194 			status = ENOTSUP;
4195 		}
4196 
4197 		break;
4198 	case VIO_SUBTYPE_INFO:
4199 		/*
4200 		 * Handle the case where vds starts handshake
4201 		 * (for now only vdc is the instigator)
4202 		 */
4203 		status = ENOTSUP;
4204 		break;
4205 
4206 	default:
4207 		status = EINVAL;
4208 		break;
4209 	}
4210 
4211 	return (status);
4212 }
4213 
4214 /*
4215  * Function:
4216  *	vdc_handle_attr_msg()
4217  *
4218  * Description:
4219  *
4220  * Arguments:
4221  *	vdc	- soft state pointer for this instance of the device driver.
4222  *	attr_msg	- LDC message sent by vDisk server
4223  *
4224  * Return Code:
4225  *	0	- Success
4226  */
4227 static int
4228 vdc_handle_attr_msg(vdc_t *vdc, vd_attr_msg_t *attr_msg)
4229 {
4230 	int status = 0;
4231 
4232 	ASSERT(vdc != NULL);
4233 	ASSERT(mutex_owned(&vdc->lock));
4234 
4235 	if (attr_msg->tag.vio_subtype_env != VIO_ATTR_INFO) {
4236 		return (EPROTO);
4237 	}
4238 
4239 	switch (attr_msg->tag.vio_subtype) {
4240 	case VIO_SUBTYPE_ACK:
4241 		/*
4242 		 * We now verify the attributes sent by vds.
4243 		 */
4244 		if (attr_msg->vdisk_size == 0) {
4245 			DMSG(vdc, 0, "[%d] Invalid disk size from vds",
4246 			    vdc->instance);
4247 			status = EINVAL;
4248 			break;
4249 		}
4250 
4251 		if (attr_msg->max_xfer_sz == 0) {
4252 			DMSG(vdc, 0, "[%d] Invalid transfer size from vds",
4253 			    vdc->instance);
4254 			status = EINVAL;
4255 			break;
4256 		}
4257 
4258 		/*
4259 		 * If the disk size is already set check that it hasn't changed.
4260 		 */
4261 		if ((vdc->vdisk_size != 0) &&
4262 		    (vdc->vdisk_size != attr_msg->vdisk_size)) {
4263 			DMSG(vdc, 0, "[%d] Different disk size from vds "
4264 			    "(old=0x%lx - new=0x%lx", vdc->instance,
4265 			    vdc->vdisk_size, attr_msg->vdisk_size)
4266 			status = EINVAL;
4267 			break;
4268 		}
4269 
4270 		vdc->vdisk_size = attr_msg->vdisk_size;
4271 		vdc->vdisk_type = attr_msg->vdisk_type;
4272 
4273 		DMSG(vdc, 0, "[%d] max_xfer_sz: sent %lx acked %lx\n",
4274 		    vdc->instance, vdc->max_xfer_sz, attr_msg->max_xfer_sz);
4275 		DMSG(vdc, 0, "[%d] vdisk_block_size: sent %lx acked %x\n",
4276 		    vdc->instance, vdc->block_size,
4277 		    attr_msg->vdisk_block_size);
4278 
4279 		/*
4280 		 * We don't know at compile time what the vDisk server will
4281 		 * think are good values but we apply an large (arbitrary)
4282 		 * upper bound to prevent memory exhaustion in vdc if it was
4283 		 * allocating a DRing based of huge values sent by the server.
4284 		 * We probably will never exceed this except if the message
4285 		 * was garbage.
4286 		 */
4287 		if ((attr_msg->max_xfer_sz * attr_msg->vdisk_block_size) <=
4288 		    (PAGESIZE * DEV_BSIZE)) {
4289 			vdc->max_xfer_sz = attr_msg->max_xfer_sz;
4290 			vdc->block_size = attr_msg->vdisk_block_size;
4291 		} else {
4292 			DMSG(vdc, 0, "[%d] vds block transfer size too big;"
4293 			    " using max supported by vdc", vdc->instance);
4294 		}
4295 
4296 		if ((attr_msg->xfer_mode != VIO_DRING_MODE) ||
4297 		    (attr_msg->vdisk_size > INT64_MAX) ||
4298 		    (attr_msg->vdisk_type > VD_DISK_TYPE_DISK)) {
4299 			DMSG(vdc, 0, "[%d] Invalid attributes from vds",
4300 			    vdc->instance);
4301 			status = EINVAL;
4302 			break;
4303 		}
4304 
4305 		/*
4306 		 * Now that we have received all attributes we can create a
4307 		 * fake geometry for the disk.
4308 		 */
4309 		vdc_create_fake_geometry(vdc);
4310 		break;
4311 
4312 	case VIO_SUBTYPE_NACK:
4313 		/*
4314 		 * vds could not handle the attributes we sent so we
4315 		 * stop negotiating.
4316 		 */
4317 		status = EPROTO;
4318 		break;
4319 
4320 	case VIO_SUBTYPE_INFO:
4321 		/*
4322 		 * Handle the case where vds starts the handshake
4323 		 * (for now; vdc is the only supported instigatior)
4324 		 */
4325 		status = ENOTSUP;
4326 		break;
4327 
4328 	default:
4329 		status = ENOTSUP;
4330 		break;
4331 	}
4332 
4333 	return (status);
4334 }
4335 
4336 /*
4337  * Function:
4338  *	vdc_handle_dring_reg_msg()
4339  *
4340  * Description:
4341  *
4342  * Arguments:
4343  *	vdc		- soft state pointer for this instance of the driver.
4344  *	dring_msg	- LDC message sent by vDisk server
4345  *
4346  * Return Code:
4347  *	0	- Success
4348  */
4349 static int
4350 vdc_handle_dring_reg_msg(vdc_t *vdc, vio_dring_reg_msg_t *dring_msg)
4351 {
4352 	int		status = 0;
4353 
4354 	ASSERT(vdc != NULL);
4355 	ASSERT(mutex_owned(&vdc->lock));
4356 
4357 	if (dring_msg->tag.vio_subtype_env != VIO_DRING_REG) {
4358 		return (EPROTO);
4359 	}
4360 
4361 	switch (dring_msg->tag.vio_subtype) {
4362 	case VIO_SUBTYPE_ACK:
4363 		/* save the received dring_ident */
4364 		vdc->dring_ident = dring_msg->dring_ident;
4365 		DMSG(vdc, 0, "[%d] Received dring ident=0x%lx\n",
4366 		    vdc->instance, vdc->dring_ident);
4367 		break;
4368 
4369 	case VIO_SUBTYPE_NACK:
4370 		/*
4371 		 * vds could not handle the DRing info we sent so we
4372 		 * stop negotiating.
4373 		 */
4374 		DMSG(vdc, 0, "[%d] server could not register DRing\n",
4375 		    vdc->instance);
4376 		status = EPROTO;
4377 		break;
4378 
4379 	case VIO_SUBTYPE_INFO:
4380 		/*
4381 		 * Handle the case where vds starts handshake
4382 		 * (for now only vdc is the instigatior)
4383 		 */
4384 		status = ENOTSUP;
4385 		break;
4386 	default:
4387 		status = ENOTSUP;
4388 	}
4389 
4390 	return (status);
4391 }
4392 
4393 /*
4394  * Function:
4395  *	vdc_verify_seq_num()
4396  *
4397  * Description:
4398  *	This functions verifies that the sequence number sent back by the vDisk
4399  *	server with the latest message is what is expected (i.e. it is greater
4400  *	than the last seq num sent by the vDisk server and less than or equal
4401  *	to the last seq num generated by vdc).
4402  *
4403  *	It then checks the request ID to see if any requests need processing
4404  *	in the DRing.
4405  *
4406  * Arguments:
4407  *	vdc		- soft state pointer for this instance of the driver.
4408  *	dring_msg	- pointer to the LDC message sent by vds
4409  *
4410  * Return Code:
4411  *	VDC_SEQ_NUM_TODO	- Message needs to be processed
4412  *	VDC_SEQ_NUM_SKIP	- Message has already been processed
4413  *	VDC_SEQ_NUM_INVALID	- The seq numbers are so out of sync,
4414  *				  vdc cannot deal with them
4415  */
4416 static int
4417 vdc_verify_seq_num(vdc_t *vdc, vio_dring_msg_t *dring_msg)
4418 {
4419 	ASSERT(vdc != NULL);
4420 	ASSERT(dring_msg != NULL);
4421 	ASSERT(mutex_owned(&vdc->lock));
4422 
4423 	/*
4424 	 * Check to see if the messages were responded to in the correct
4425 	 * order by vds.
4426 	 */
4427 	if ((dring_msg->seq_num <= vdc->seq_num_reply) ||
4428 	    (dring_msg->seq_num > vdc->seq_num)) {
4429 		DMSG(vdc, 0, "?[%d] Bogus sequence_number %lu: "
4430 		    "%lu > expected <= %lu (last proc req %lu sent %lu)\n",
4431 		    vdc->instance, dring_msg->seq_num,
4432 		    vdc->seq_num_reply, vdc->seq_num,
4433 		    vdc->req_id_proc, vdc->req_id);
4434 		return (VDC_SEQ_NUM_INVALID);
4435 	}
4436 	vdc->seq_num_reply = dring_msg->seq_num;
4437 
4438 	if (vdc->req_id_proc < vdc->req_id)
4439 		return (VDC_SEQ_NUM_TODO);
4440 	else
4441 		return (VDC_SEQ_NUM_SKIP);
4442 }
4443 
4444 
4445 /*
4446  * Function:
4447  *	vdc_is_supported_version()
4448  *
4449  * Description:
4450  *	This routine checks if the major/minor version numbers specified in
4451  *	'ver_msg' are supported. If not it finds the next version that is
4452  *	in the supported version list 'vdc_version[]' and sets the fields in
4453  *	'ver_msg' to those values
4454  *
4455  * Arguments:
4456  *	ver_msg	- LDC message sent by vDisk server
4457  *
4458  * Return Code:
4459  *	B_TRUE	- Success
4460  *	B_FALSE	- Version not supported
4461  */
4462 static boolean_t
4463 vdc_is_supported_version(vio_ver_msg_t *ver_msg)
4464 {
4465 	int vdc_num_versions = sizeof (vdc_version) / sizeof (vdc_version[0]);
4466 
4467 	for (int i = 0; i < vdc_num_versions; i++) {
4468 		ASSERT(vdc_version[i].major > 0);
4469 		ASSERT((i == 0) ||
4470 		    (vdc_version[i].major < vdc_version[i-1].major));
4471 
4472 		/*
4473 		 * If the major versions match, adjust the minor version, if
4474 		 * necessary, down to the highest value supported by this
4475 		 * client. The server should support all minor versions lower
4476 		 * than the value it sent
4477 		 */
4478 		if (ver_msg->ver_major == vdc_version[i].major) {
4479 			if (ver_msg->ver_minor > vdc_version[i].minor) {
4480 				DMSGX(0,
4481 				    "Adjusting minor version from %u to %u",
4482 				    ver_msg->ver_minor, vdc_version[i].minor);
4483 				ver_msg->ver_minor = vdc_version[i].minor;
4484 			}
4485 			return (B_TRUE);
4486 		}
4487 
4488 		/*
4489 		 * If the message contains a higher major version number, set
4490 		 * the message's major/minor versions to the current values
4491 		 * and return false, so this message will get resent with
4492 		 * these values, and the server will potentially try again
4493 		 * with the same or a lower version
4494 		 */
4495 		if (ver_msg->ver_major > vdc_version[i].major) {
4496 			ver_msg->ver_major = vdc_version[i].major;
4497 			ver_msg->ver_minor = vdc_version[i].minor;
4498 			DMSGX(0, "Suggesting major/minor (0x%x/0x%x)\n",
4499 			    ver_msg->ver_major, ver_msg->ver_minor);
4500 
4501 			return (B_FALSE);
4502 		}
4503 
4504 		/*
4505 		 * Otherwise, the message's major version is less than the
4506 		 * current major version, so continue the loop to the next
4507 		 * (lower) supported version
4508 		 */
4509 	}
4510 
4511 	/*
4512 	 * No common version was found; "ground" the version pair in the
4513 	 * message to terminate negotiation
4514 	 */
4515 	ver_msg->ver_major = 0;
4516 	ver_msg->ver_minor = 0;
4517 
4518 	return (B_FALSE);
4519 }
4520 /* -------------------------------------------------------------------------- */
4521 
4522 /*
4523  * DKIO(7) support
4524  */
4525 
4526 typedef struct vdc_dk_arg {
4527 	struct dk_callback	dkc;
4528 	int			mode;
4529 	dev_t			dev;
4530 	vdc_t			*vdc;
4531 } vdc_dk_arg_t;
4532 
4533 /*
4534  * Function:
4535  * 	vdc_dkio_flush_cb()
4536  *
4537  * Description:
4538  *	This routine is a callback for DKIOCFLUSHWRITECACHE which can be called
4539  *	by kernel code.
4540  *
4541  * Arguments:
4542  *	arg	- a pointer to a vdc_dk_arg_t structure.
4543  */
4544 void
4545 vdc_dkio_flush_cb(void *arg)
4546 {
4547 	struct vdc_dk_arg	*dk_arg = (struct vdc_dk_arg *)arg;
4548 	struct dk_callback	*dkc = NULL;
4549 	vdc_t			*vdc = NULL;
4550 	int			rv;
4551 
4552 	if (dk_arg == NULL) {
4553 		cmn_err(CE_NOTE, "?[Unk] DKIOCFLUSHWRITECACHE arg is NULL\n");
4554 		return;
4555 	}
4556 	dkc = &dk_arg->dkc;
4557 	vdc = dk_arg->vdc;
4558 	ASSERT(vdc != NULL);
4559 
4560 	rv = vdc_do_sync_op(vdc, VD_OP_FLUSH, NULL, 0,
4561 	    VDCPART(dk_arg->dev), 0, CB_SYNC, 0, VIO_both_dir);
4562 	if (rv != 0) {
4563 		DMSG(vdc, 0, "[%d] DKIOCFLUSHWRITECACHE failed %d : model %x\n",
4564 		    vdc->instance, rv,
4565 		    ddi_model_convert_from(dk_arg->mode & FMODELS));
4566 	}
4567 
4568 	/*
4569 	 * Trigger the call back to notify the caller the the ioctl call has
4570 	 * been completed.
4571 	 */
4572 	if ((dk_arg->mode & FKIOCTL) &&
4573 	    (dkc != NULL) &&
4574 	    (dkc->dkc_callback != NULL)) {
4575 		ASSERT(dkc->dkc_cookie != NULL);
4576 		(*dkc->dkc_callback)(dkc->dkc_cookie, rv);
4577 	}
4578 
4579 	/* Indicate that one less DKIO write flush is outstanding */
4580 	mutex_enter(&vdc->lock);
4581 	vdc->dkio_flush_pending--;
4582 	ASSERT(vdc->dkio_flush_pending >= 0);
4583 	mutex_exit(&vdc->lock);
4584 
4585 	/* free the mem that was allocated when the callback was dispatched */
4586 	kmem_free(arg, sizeof (vdc_dk_arg_t));
4587 }
4588 
4589 /*
4590  * Function:
4591  * 	vdc_dkio_get_partition()
4592  *
4593  * Description:
4594  *	This function implements the DKIOCGAPART ioctl.
4595  *
4596  * Arguments:
4597  *	vdc	- soft state pointer
4598  *	arg	- a pointer to a dk_map[NDKMAP] or dk_map32[NDKMAP] structure
4599  *	flag	- ioctl flags
4600  */
4601 static int
4602 vdc_dkio_get_partition(vdc_t *vdc, caddr_t arg, int flag)
4603 {
4604 	struct dk_geom *geom;
4605 	struct vtoc *vtoc;
4606 	union {
4607 		struct dk_map map[NDKMAP];
4608 		struct dk_map32 map32[NDKMAP];
4609 	} data;
4610 	int i, rv, size;
4611 
4612 	mutex_enter(&vdc->lock);
4613 
4614 	if ((rv = vdc_validate_geometry(vdc)) != 0) {
4615 		mutex_exit(&vdc->lock);
4616 		return (rv);
4617 	}
4618 
4619 	vtoc = vdc->vtoc;
4620 	geom = vdc->geom;
4621 
4622 	if (ddi_model_convert_from(flag & FMODELS) == DDI_MODEL_ILP32) {
4623 
4624 		for (i = 0; i < vtoc->v_nparts; i++) {
4625 			data.map32[i].dkl_cylno = vtoc->v_part[i].p_start /
4626 			    (geom->dkg_nhead * geom->dkg_nsect);
4627 			data.map32[i].dkl_nblk = vtoc->v_part[i].p_size;
4628 		}
4629 		size = NDKMAP * sizeof (struct dk_map32);
4630 
4631 	} else {
4632 
4633 		for (i = 0; i < vtoc->v_nparts; i++) {
4634 			data.map[i].dkl_cylno = vtoc->v_part[i].p_start /
4635 			    (geom->dkg_nhead * geom->dkg_nsect);
4636 			data.map[i].dkl_nblk = vtoc->v_part[i].p_size;
4637 		}
4638 		size = NDKMAP * sizeof (struct dk_map);
4639 
4640 	}
4641 
4642 	mutex_exit(&vdc->lock);
4643 
4644 	if (ddi_copyout(&data, arg, size, flag) != 0)
4645 		return (EFAULT);
4646 
4647 	return (0);
4648 }
4649 
4650 /*
4651  * Function:
4652  * 	vdc_dioctl_rwcmd()
4653  *
4654  * Description:
4655  *	This function implements the DIOCTL_RWCMD ioctl. This ioctl is used
4656  *	for DKC_DIRECT disks to read or write at an absolute disk offset.
4657  *
4658  * Arguments:
4659  *	dev	- device
4660  *	arg	- a pointer to a dadkio_rwcmd or dadkio_rwcmd32 structure
4661  *	flag	- ioctl flags
4662  */
4663 static int
4664 vdc_dioctl_rwcmd(dev_t dev, caddr_t arg, int flag)
4665 {
4666 	struct dadkio_rwcmd32 rwcmd32;
4667 	struct dadkio_rwcmd rwcmd;
4668 	struct iovec aiov;
4669 	struct uio auio;
4670 	int rw, status;
4671 	struct buf *buf;
4672 
4673 	if (ddi_model_convert_from(flag & FMODELS) == DDI_MODEL_ILP32) {
4674 		if (ddi_copyin((caddr_t)arg, (caddr_t)&rwcmd32,
4675 		    sizeof (struct dadkio_rwcmd32), flag)) {
4676 			return (EFAULT);
4677 		}
4678 		rwcmd.cmd = rwcmd32.cmd;
4679 		rwcmd.flags = rwcmd32.flags;
4680 		rwcmd.blkaddr = (daddr_t)rwcmd32.blkaddr;
4681 		rwcmd.buflen = rwcmd32.buflen;
4682 		rwcmd.bufaddr = (caddr_t)(uintptr_t)rwcmd32.bufaddr;
4683 	} else {
4684 		if (ddi_copyin((caddr_t)arg, (caddr_t)&rwcmd,
4685 		    sizeof (struct dadkio_rwcmd), flag)) {
4686 			return (EFAULT);
4687 		}
4688 	}
4689 
4690 	switch (rwcmd.cmd) {
4691 	case DADKIO_RWCMD_READ:
4692 		rw = B_READ;
4693 		break;
4694 	case DADKIO_RWCMD_WRITE:
4695 		rw = B_WRITE;
4696 		break;
4697 	default:
4698 		return (EINVAL);
4699 	}
4700 
4701 	bzero((caddr_t)&aiov, sizeof (struct iovec));
4702 	aiov.iov_base   = rwcmd.bufaddr;
4703 	aiov.iov_len    = rwcmd.buflen;
4704 
4705 	bzero((caddr_t)&auio, sizeof (struct uio));
4706 	auio.uio_iov    = &aiov;
4707 	auio.uio_iovcnt = 1;
4708 	auio.uio_loffset = rwcmd.blkaddr * DEV_BSIZE;
4709 	auio.uio_resid  = rwcmd.buflen;
4710 	auio.uio_segflg = flag & FKIOCTL ? UIO_SYSSPACE : UIO_USERSPACE;
4711 
4712 	buf = kmem_alloc(sizeof (buf_t), KM_SLEEP);
4713 	bioinit(buf);
4714 	/*
4715 	 * We use the private field of buf to specify that this is an
4716 	 * I/O using an absolute offset.
4717 	 */
4718 	buf->b_private = (void *)VD_SLICE_NONE;
4719 
4720 	status = physio(vdc_strategy, buf, dev, rw, vdc_min, &auio);
4721 
4722 	biofini(buf);
4723 	kmem_free(buf, sizeof (buf_t));
4724 
4725 	return (status);
4726 }
4727 
4728 /*
4729  * This structure is used in the DKIO(7I) array below.
4730  */
4731 typedef struct vdc_dk_ioctl {
4732 	uint8_t		op;		/* VD_OP_XXX value */
4733 	int		cmd;		/* Solaris ioctl operation number */
4734 	size_t		nbytes;		/* size of structure to be copied */
4735 
4736 	/* function to convert between vDisk and Solaris structure formats */
4737 	int	(*convert)(vdc_t *vdc, void *vd_buf, void *ioctl_arg,
4738 	    int mode, int dir);
4739 } vdc_dk_ioctl_t;
4740 
4741 /*
4742  * Subset of DKIO(7I) operations currently supported
4743  */
4744 static vdc_dk_ioctl_t	dk_ioctl[] = {
4745 	{VD_OP_FLUSH,		DKIOCFLUSHWRITECACHE,	0,
4746 		vdc_null_copy_func},
4747 	{VD_OP_GET_WCE,		DKIOCGETWCE,		sizeof (int),
4748 		vdc_get_wce_convert},
4749 	{VD_OP_SET_WCE,		DKIOCSETWCE,		sizeof (int),
4750 		vdc_set_wce_convert},
4751 	{VD_OP_GET_VTOC,	DKIOCGVTOC,		sizeof (vd_vtoc_t),
4752 		vdc_get_vtoc_convert},
4753 	{VD_OP_SET_VTOC,	DKIOCSVTOC,		sizeof (vd_vtoc_t),
4754 		vdc_set_vtoc_convert},
4755 	{VD_OP_GET_DISKGEOM,	DKIOCGGEOM,		sizeof (vd_geom_t),
4756 		vdc_get_geom_convert},
4757 	{VD_OP_GET_DISKGEOM,	DKIOCG_PHYGEOM,		sizeof (vd_geom_t),
4758 		vdc_get_geom_convert},
4759 	{VD_OP_GET_DISKGEOM, 	DKIOCG_VIRTGEOM,	sizeof (vd_geom_t),
4760 		vdc_get_geom_convert},
4761 	{VD_OP_SET_DISKGEOM,	DKIOCSGEOM,		sizeof (vd_geom_t),
4762 		vdc_set_geom_convert},
4763 	{VD_OP_GET_EFI,		DKIOCGETEFI,		0,
4764 		vdc_get_efi_convert},
4765 	{VD_OP_SET_EFI,		DKIOCSETEFI,		0,
4766 		vdc_set_efi_convert},
4767 
4768 	/* DIOCTL_RWCMD is converted to a read or a write */
4769 	{0, DIOCTL_RWCMD,  sizeof (struct dadkio_rwcmd), NULL},
4770 
4771 	/*
4772 	 * These particular ioctls are not sent to the server - vdc fakes up
4773 	 * the necessary info.
4774 	 */
4775 	{0, DKIOCINFO, sizeof (struct dk_cinfo), vdc_null_copy_func},
4776 	{0, DKIOCGMEDIAINFO, sizeof (struct dk_minfo), vdc_null_copy_func},
4777 	{0, USCSICMD,	sizeof (struct uscsi_cmd), vdc_null_copy_func},
4778 	{0, DKIOCGAPART, 0, vdc_null_copy_func },
4779 	{0, DKIOCREMOVABLE, 0, vdc_null_copy_func},
4780 	{0, CDROMREADOFFSET, 0, vdc_null_copy_func}
4781 };
4782 
4783 /*
4784  * Function:
4785  *	vd_process_ioctl()
4786  *
4787  * Description:
4788  *	This routine processes disk specific ioctl calls
4789  *
4790  * Arguments:
4791  *	dev	- the device number
4792  *	cmd	- the operation [dkio(7I)] to be processed
4793  *	arg	- pointer to user provided structure
4794  *		  (contains data to be set or reference parameter for get)
4795  *	mode	- bit flag, indicating open settings, 32/64 bit type, etc
4796  *
4797  * Return Code:
4798  *	0
4799  *	EFAULT
4800  *	ENXIO
4801  *	EIO
4802  *	ENOTSUP
4803  */
4804 static int
4805 vd_process_ioctl(dev_t dev, int cmd, caddr_t arg, int mode)
4806 {
4807 	int		instance = VDCUNIT(dev);
4808 	vdc_t		*vdc = NULL;
4809 	int		rv = -1;
4810 	int		idx = 0;		/* index into dk_ioctl[] */
4811 	size_t		len = 0;		/* #bytes to send to vds */
4812 	size_t		alloc_len = 0;		/* #bytes to allocate mem for */
4813 	caddr_t		mem_p = NULL;
4814 	size_t		nioctls = (sizeof (dk_ioctl)) / (sizeof (dk_ioctl[0]));
4815 	vdc_dk_ioctl_t	*iop;
4816 
4817 	vdc = ddi_get_soft_state(vdc_state, instance);
4818 	if (vdc == NULL) {
4819 		cmn_err(CE_NOTE, "![%d] Could not get soft state structure",
4820 		    instance);
4821 		return (ENXIO);
4822 	}
4823 
4824 	DMSG(vdc, 0, "[%d] Processing ioctl(%x) for dev %lx : model %x\n",
4825 	    instance, cmd, dev, ddi_model_convert_from(mode & FMODELS));
4826 
4827 	/*
4828 	 * Validate the ioctl operation to be performed.
4829 	 *
4830 	 * If we have looped through the array without finding a match then we
4831 	 * don't support this ioctl.
4832 	 */
4833 	for (idx = 0; idx < nioctls; idx++) {
4834 		if (cmd == dk_ioctl[idx].cmd)
4835 			break;
4836 	}
4837 
4838 	if (idx >= nioctls) {
4839 		DMSG(vdc, 0, "[%d] Unsupported ioctl (0x%x)\n",
4840 		    vdc->instance, cmd);
4841 		return (ENOTSUP);
4842 	}
4843 
4844 	iop = &(dk_ioctl[idx]);
4845 
4846 	if (cmd == DKIOCGETEFI || cmd == DKIOCSETEFI) {
4847 		/* size is not fixed for EFI ioctls, it depends on ioctl arg */
4848 		dk_efi_t	dk_efi;
4849 
4850 		rv = ddi_copyin(arg, &dk_efi, sizeof (dk_efi_t), mode);
4851 		if (rv != 0)
4852 			return (EFAULT);
4853 
4854 		len = sizeof (vd_efi_t) - 1 + dk_efi.dki_length;
4855 	} else {
4856 		len = iop->nbytes;
4857 	}
4858 
4859 	/*
4860 	 * Deal with the ioctls which the server does not provide. vdc can
4861 	 * fake these up and return immediately
4862 	 */
4863 	switch (cmd) {
4864 	case CDROMREADOFFSET:
4865 	case DKIOCREMOVABLE:
4866 	case USCSICMD:
4867 		return (ENOTTY);
4868 
4869 	case DIOCTL_RWCMD:
4870 		{
4871 			if (vdc->cinfo == NULL)
4872 				return (ENXIO);
4873 
4874 			if (vdc->cinfo->dki_ctype != DKC_DIRECT)
4875 				return (ENOTTY);
4876 
4877 			return (vdc_dioctl_rwcmd(dev, arg, mode));
4878 		}
4879 
4880 	case DKIOCGAPART:
4881 		{
4882 			return (vdc_dkio_get_partition(vdc, arg, mode));
4883 		}
4884 
4885 	case DKIOCINFO:
4886 		{
4887 			struct dk_cinfo	cinfo;
4888 			if (vdc->cinfo == NULL)
4889 				return (ENXIO);
4890 
4891 			bcopy(vdc->cinfo, &cinfo, sizeof (struct dk_cinfo));
4892 			cinfo.dki_partition = VDCPART(dev);
4893 
4894 			rv = ddi_copyout(&cinfo, (void *)arg,
4895 			    sizeof (struct dk_cinfo), mode);
4896 			if (rv != 0)
4897 				return (EFAULT);
4898 
4899 			return (0);
4900 		}
4901 
4902 	case DKIOCGMEDIAINFO:
4903 		{
4904 			if (vdc->minfo == NULL)
4905 				return (ENXIO);
4906 
4907 			rv = ddi_copyout(vdc->minfo, (void *)arg,
4908 			    sizeof (struct dk_minfo), mode);
4909 			if (rv != 0)
4910 				return (EFAULT);
4911 
4912 			return (0);
4913 		}
4914 
4915 	case DKIOCFLUSHWRITECACHE:
4916 		{
4917 			struct dk_callback *dkc = (struct dk_callback *)arg;
4918 			vdc_dk_arg_t	*dkarg = NULL;
4919 
4920 			DMSG(vdc, 1, "[%d] Flush W$: mode %x\n",
4921 			    instance, mode);
4922 
4923 			/*
4924 			 * If the backing device is not a 'real' disk then the
4925 			 * W$ operation request to the vDisk server will fail
4926 			 * so we might as well save the cycles and return now.
4927 			 */
4928 			if (vdc->vdisk_type != VD_DISK_TYPE_DISK)
4929 				return (ENOTTY);
4930 
4931 			/*
4932 			 * If arg is NULL, then there is no callback function
4933 			 * registered and the call operates synchronously; we
4934 			 * break and continue with the rest of the function and
4935 			 * wait for vds to return (i.e. after the request to
4936 			 * vds returns successfully, all writes completed prior
4937 			 * to the ioctl will have been flushed from the disk
4938 			 * write cache to persistent media.
4939 			 *
4940 			 * If a callback function is registered, we dispatch
4941 			 * the request on a task queue and return immediately.
4942 			 * The callback will deal with informing the calling
4943 			 * thread that the flush request is completed.
4944 			 */
4945 			if (dkc == NULL)
4946 				break;
4947 
4948 			/*
4949 			 * the asynchronous callback is only supported if
4950 			 * invoked from within the kernel
4951 			 */
4952 			if ((mode & FKIOCTL) == 0)
4953 				return (ENOTSUP);
4954 
4955 			dkarg = kmem_zalloc(sizeof (vdc_dk_arg_t), KM_SLEEP);
4956 
4957 			dkarg->mode = mode;
4958 			dkarg->dev = dev;
4959 			bcopy(dkc, &dkarg->dkc, sizeof (*dkc));
4960 
4961 			mutex_enter(&vdc->lock);
4962 			vdc->dkio_flush_pending++;
4963 			dkarg->vdc = vdc;
4964 			mutex_exit(&vdc->lock);
4965 
4966 			/* put the request on a task queue */
4967 			rv = taskq_dispatch(system_taskq, vdc_dkio_flush_cb,
4968 			    (void *)dkarg, DDI_SLEEP);
4969 			if (rv == NULL) {
4970 				/* clean up if dispatch fails */
4971 				mutex_enter(&vdc->lock);
4972 				vdc->dkio_flush_pending--;
4973 				mutex_exit(&vdc->lock);
4974 				kmem_free(dkarg, sizeof (vdc_dk_arg_t));
4975 			}
4976 
4977 			return (rv == NULL ? ENOMEM : 0);
4978 		}
4979 	}
4980 
4981 	/* catch programming error in vdc - should be a VD_OP_XXX ioctl */
4982 	ASSERT(iop->op != 0);
4983 
4984 	/* LDC requires that the memory being mapped is 8-byte aligned */
4985 	alloc_len = P2ROUNDUP(len, sizeof (uint64_t));
4986 	DMSG(vdc, 1, "[%d] struct size %ld alloc %ld\n",
4987 	    instance, len, alloc_len);
4988 
4989 	ASSERT(alloc_len >= 0); /* sanity check */
4990 	if (alloc_len > 0)
4991 		mem_p = kmem_zalloc(alloc_len, KM_SLEEP);
4992 
4993 	/*
4994 	 * Call the conversion function for this ioctl which, if necessary,
4995 	 * converts from the Solaris format to the format ARC'ed
4996 	 * as part of the vDisk protocol (FWARC 2006/195)
4997 	 */
4998 	ASSERT(iop->convert != NULL);
4999 	rv = (iop->convert)(vdc, arg, mem_p, mode, VD_COPYIN);
5000 	if (rv != 0) {
5001 		DMSG(vdc, 0, "[%d] convert func returned %d for ioctl 0x%x\n",
5002 		    instance, rv, cmd);
5003 		if (mem_p != NULL)
5004 			kmem_free(mem_p, alloc_len);
5005 		return (rv);
5006 	}
5007 
5008 	/*
5009 	 * send request to vds to service the ioctl.
5010 	 */
5011 	rv = vdc_do_sync_op(vdc, iop->op, mem_p, alloc_len,
5012 	    VDCPART(dev), 0, CB_SYNC, (void *)(uint64_t)mode,
5013 	    VIO_both_dir);
5014 
5015 	if (cmd == DKIOCSVTOC || cmd == DKIOCSETEFI) {
5016 		/*
5017 		 * The disk label may have changed. Revalidate the disk
5018 		 * geometry. This will also update the device nodes and
5019 		 * properties.
5020 		 */
5021 		vdc_validate(vdc);
5022 	}
5023 
5024 	if (rv != 0) {
5025 		/*
5026 		 * This is not necessarily an error. The ioctl could
5027 		 * be returning a value such as ENOTTY to indicate
5028 		 * that the ioctl is not applicable.
5029 		 */
5030 		DMSG(vdc, 0, "[%d] vds returned %d for ioctl 0x%x\n",
5031 		    instance, rv, cmd);
5032 		if (mem_p != NULL)
5033 			kmem_free(mem_p, alloc_len);
5034 
5035 		return (rv);
5036 	}
5037 
5038 	/*
5039 	 * Call the conversion function (if it exists) for this ioctl
5040 	 * which converts from the format ARC'ed as part of the vDisk
5041 	 * protocol (FWARC 2006/195) back to a format understood by
5042 	 * the rest of Solaris.
5043 	 */
5044 	rv = (iop->convert)(vdc, mem_p, arg, mode, VD_COPYOUT);
5045 	if (rv != 0) {
5046 		DMSG(vdc, 0, "[%d] convert func returned %d for ioctl 0x%x\n",
5047 		    instance, rv, cmd);
5048 		if (mem_p != NULL)
5049 			kmem_free(mem_p, alloc_len);
5050 		return (rv);
5051 	}
5052 
5053 	if (mem_p != NULL)
5054 		kmem_free(mem_p, alloc_len);
5055 
5056 	return (rv);
5057 }
5058 
5059 /*
5060  * Function:
5061  *
5062  * Description:
5063  *	This is an empty conversion function used by ioctl calls which
5064  *	do not need to convert the data being passed in/out to userland
5065  */
5066 static int
5067 vdc_null_copy_func(vdc_t *vdc, void *from, void *to, int mode, int dir)
5068 {
5069 	_NOTE(ARGUNUSED(vdc))
5070 	_NOTE(ARGUNUSED(from))
5071 	_NOTE(ARGUNUSED(to))
5072 	_NOTE(ARGUNUSED(mode))
5073 	_NOTE(ARGUNUSED(dir))
5074 
5075 	return (0);
5076 }
5077 
5078 static int
5079 vdc_get_wce_convert(vdc_t *vdc, void *from, void *to,
5080     int mode, int dir)
5081 {
5082 	_NOTE(ARGUNUSED(vdc))
5083 
5084 	if (dir == VD_COPYIN)
5085 		return (0);		/* nothing to do */
5086 
5087 	if (ddi_copyout(from, to, sizeof (int), mode) != 0)
5088 		return (EFAULT);
5089 
5090 	return (0);
5091 }
5092 
5093 static int
5094 vdc_set_wce_convert(vdc_t *vdc, void *from, void *to,
5095     int mode, int dir)
5096 {
5097 	_NOTE(ARGUNUSED(vdc))
5098 
5099 	if (dir == VD_COPYOUT)
5100 		return (0);		/* nothing to do */
5101 
5102 	if (ddi_copyin(from, to, sizeof (int), mode) != 0)
5103 		return (EFAULT);
5104 
5105 	return (0);
5106 }
5107 
5108 /*
5109  * Function:
5110  *	vdc_get_vtoc_convert()
5111  *
5112  * Description:
5113  *	This routine performs the necessary convertions from the DKIOCGVTOC
5114  *	Solaris structure to the format defined in FWARC 2006/195.
5115  *
5116  *	In the struct vtoc definition, the timestamp field is marked as not
5117  *	supported so it is not part of vDisk protocol (FWARC 2006/195).
5118  *	However SVM uses that field to check it can write into the VTOC,
5119  *	so we fake up the info of that field.
5120  *
5121  * Arguments:
5122  *	vdc	- the vDisk client
5123  *	from	- the buffer containing the data to be copied from
5124  *	to	- the buffer to be copied to
5125  *	mode	- flags passed to ioctl() call
5126  *	dir	- the "direction" of the copy - VD_COPYIN or VD_COPYOUT
5127  *
5128  * Return Code:
5129  *	0	- Success
5130  *	ENXIO	- incorrect buffer passed in.
5131  *	EFAULT	- ddi_copyout routine encountered an error.
5132  */
5133 static int
5134 vdc_get_vtoc_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
5135 {
5136 	int		i;
5137 	void		*tmp_mem = NULL;
5138 	void		*tmp_memp;
5139 	struct vtoc	vt;
5140 	struct vtoc32	vt32;
5141 	int		copy_len = 0;
5142 	int		rv = 0;
5143 
5144 	if (dir != VD_COPYOUT)
5145 		return (0);	/* nothing to do */
5146 
5147 	if ((from == NULL) || (to == NULL))
5148 		return (ENXIO);
5149 
5150 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32)
5151 		copy_len = sizeof (struct vtoc32);
5152 	else
5153 		copy_len = sizeof (struct vtoc);
5154 
5155 	tmp_mem = kmem_alloc(copy_len, KM_SLEEP);
5156 
5157 	VD_VTOC2VTOC((vd_vtoc_t *)from, &vt);
5158 
5159 	/* fake the VTOC timestamp field */
5160 	for (i = 0; i < V_NUMPAR; i++) {
5161 		vt.timestamp[i] = vdc->vtoc->timestamp[i];
5162 	}
5163 
5164 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5165 		vtoctovtoc32(vt, vt32);
5166 		tmp_memp = &vt32;
5167 	} else {
5168 		tmp_memp = &vt;
5169 	}
5170 	rv = ddi_copyout(tmp_memp, to, copy_len, mode);
5171 	if (rv != 0)
5172 		rv = EFAULT;
5173 
5174 	kmem_free(tmp_mem, copy_len);
5175 	return (rv);
5176 }
5177 
5178 /*
5179  * Function:
5180  *	vdc_set_vtoc_convert()
5181  *
5182  * Description:
5183  *	This routine performs the necessary convertions from the DKIOCSVTOC
5184  *	Solaris structure to the format defined in FWARC 2006/195.
5185  *
5186  * Arguments:
5187  *	vdc	- the vDisk client
5188  *	from	- Buffer with data
5189  *	to	- Buffer where data is to be copied to
5190  *	mode	- flags passed to ioctl
5191  *	dir	- direction of copy (in or out)
5192  *
5193  * Return Code:
5194  *	0	- Success
5195  *	ENXIO	- Invalid buffer passed in
5196  *	EFAULT	- ddi_copyin of data failed
5197  */
5198 static int
5199 vdc_set_vtoc_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
5200 {
5201 	_NOTE(ARGUNUSED(vdc))
5202 
5203 	void		*tmp_mem = NULL;
5204 	struct vtoc	vt;
5205 	struct vtoc	*vtp = &vt;
5206 	vd_vtoc_t	vtvd;
5207 	int		copy_len = 0;
5208 	int		rv = 0;
5209 
5210 	if (dir != VD_COPYIN)
5211 		return (0);	/* nothing to do */
5212 
5213 	if ((from == NULL) || (to == NULL))
5214 		return (ENXIO);
5215 
5216 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32)
5217 		copy_len = sizeof (struct vtoc32);
5218 	else
5219 		copy_len = sizeof (struct vtoc);
5220 
5221 	tmp_mem = kmem_alloc(copy_len, KM_SLEEP);
5222 
5223 	rv = ddi_copyin(from, tmp_mem, copy_len, mode);
5224 	if (rv != 0) {
5225 		kmem_free(tmp_mem, copy_len);
5226 		return (EFAULT);
5227 	}
5228 
5229 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5230 		vtoc32tovtoc((*(struct vtoc32 *)tmp_mem), vt);
5231 	} else {
5232 		vtp = tmp_mem;
5233 	}
5234 
5235 	VTOC2VD_VTOC(vtp, &vtvd);
5236 	bcopy(&vtvd, to, sizeof (vd_vtoc_t));
5237 	kmem_free(tmp_mem, copy_len);
5238 
5239 	return (0);
5240 }
5241 
5242 /*
5243  * Function:
5244  *	vdc_get_geom_convert()
5245  *
5246  * Description:
5247  *	This routine performs the necessary convertions from the DKIOCGGEOM,
5248  *	DKIOCG_PHYSGEOM and DKIOG_VIRTGEOM Solaris structures to the format
5249  *	defined in FWARC 2006/195
5250  *
5251  * Arguments:
5252  *	vdc	- the vDisk client
5253  *	from	- Buffer with data
5254  *	to	- Buffer where data is to be copied to
5255  *	mode	- flags passed to ioctl
5256  *	dir	- direction of copy (in or out)
5257  *
5258  * Return Code:
5259  *	0	- Success
5260  *	ENXIO	- Invalid buffer passed in
5261  *	EFAULT	- ddi_copyout of data failed
5262  */
5263 static int
5264 vdc_get_geom_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
5265 {
5266 	_NOTE(ARGUNUSED(vdc))
5267 
5268 	struct dk_geom	geom;
5269 	int	copy_len = sizeof (struct dk_geom);
5270 	int	rv = 0;
5271 
5272 	if (dir != VD_COPYOUT)
5273 		return (0);	/* nothing to do */
5274 
5275 	if ((from == NULL) || (to == NULL))
5276 		return (ENXIO);
5277 
5278 	VD_GEOM2DK_GEOM((vd_geom_t *)from, &geom);
5279 	rv = ddi_copyout(&geom, to, copy_len, mode);
5280 	if (rv != 0)
5281 		rv = EFAULT;
5282 
5283 	return (rv);
5284 }
5285 
5286 /*
5287  * Function:
5288  *	vdc_set_geom_convert()
5289  *
5290  * Description:
5291  *	This routine performs the necessary convertions from the DKIOCSGEOM
5292  *	Solaris structure to the format defined in FWARC 2006/195.
5293  *
5294  * Arguments:
5295  *	vdc	- the vDisk client
5296  *	from	- Buffer with data
5297  *	to	- Buffer where data is to be copied to
5298  *	mode	- flags passed to ioctl
5299  *	dir	- direction of copy (in or out)
5300  *
5301  * Return Code:
5302  *	0	- Success
5303  *	ENXIO	- Invalid buffer passed in
5304  *	EFAULT	- ddi_copyin of data failed
5305  */
5306 static int
5307 vdc_set_geom_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
5308 {
5309 	_NOTE(ARGUNUSED(vdc))
5310 
5311 	vd_geom_t	vdgeom;
5312 	void		*tmp_mem = NULL;
5313 	int		copy_len = sizeof (struct dk_geom);
5314 	int		rv = 0;
5315 
5316 	if (dir != VD_COPYIN)
5317 		return (0);	/* nothing to do */
5318 
5319 	if ((from == NULL) || (to == NULL))
5320 		return (ENXIO);
5321 
5322 	tmp_mem = kmem_alloc(copy_len, KM_SLEEP);
5323 
5324 	rv = ddi_copyin(from, tmp_mem, copy_len, mode);
5325 	if (rv != 0) {
5326 		kmem_free(tmp_mem, copy_len);
5327 		return (EFAULT);
5328 	}
5329 	DK_GEOM2VD_GEOM((struct dk_geom *)tmp_mem, &vdgeom);
5330 	bcopy(&vdgeom, to, sizeof (vdgeom));
5331 	kmem_free(tmp_mem, copy_len);
5332 
5333 	return (0);
5334 }
5335 
5336 static int
5337 vdc_get_efi_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
5338 {
5339 	_NOTE(ARGUNUSED(vdc))
5340 
5341 	vd_efi_t	*vd_efi;
5342 	dk_efi_t	dk_efi;
5343 	int		rv = 0;
5344 	void		*uaddr;
5345 
5346 	if ((from == NULL) || (to == NULL))
5347 		return (ENXIO);
5348 
5349 	if (dir == VD_COPYIN) {
5350 
5351 		vd_efi = (vd_efi_t *)to;
5352 
5353 		rv = ddi_copyin(from, &dk_efi, sizeof (dk_efi_t), mode);
5354 		if (rv != 0)
5355 			return (EFAULT);
5356 
5357 		vd_efi->lba = dk_efi.dki_lba;
5358 		vd_efi->length = dk_efi.dki_length;
5359 		bzero(vd_efi->data, vd_efi->length);
5360 
5361 	} else {
5362 
5363 		rv = ddi_copyin(to, &dk_efi, sizeof (dk_efi_t), mode);
5364 		if (rv != 0)
5365 			return (EFAULT);
5366 
5367 		uaddr = dk_efi.dki_data;
5368 
5369 		dk_efi.dki_data = kmem_alloc(dk_efi.dki_length, KM_SLEEP);
5370 
5371 		VD_EFI2DK_EFI((vd_efi_t *)from, &dk_efi);
5372 
5373 		rv = ddi_copyout(dk_efi.dki_data, uaddr, dk_efi.dki_length,
5374 		    mode);
5375 		if (rv != 0)
5376 			return (EFAULT);
5377 
5378 		kmem_free(dk_efi.dki_data, dk_efi.dki_length);
5379 	}
5380 
5381 	return (0);
5382 }
5383 
5384 static int
5385 vdc_set_efi_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
5386 {
5387 	_NOTE(ARGUNUSED(vdc))
5388 
5389 	dk_efi_t	dk_efi;
5390 	void		*uaddr;
5391 
5392 	if (dir == VD_COPYOUT)
5393 		return (0);	/* nothing to do */
5394 
5395 	if ((from == NULL) || (to == NULL))
5396 		return (ENXIO);
5397 
5398 	if (ddi_copyin(from, &dk_efi, sizeof (dk_efi_t), mode) != 0)
5399 		return (EFAULT);
5400 
5401 	uaddr = dk_efi.dki_data;
5402 
5403 	dk_efi.dki_data = kmem_alloc(dk_efi.dki_length, KM_SLEEP);
5404 
5405 	if (ddi_copyin(uaddr, dk_efi.dki_data, dk_efi.dki_length, mode) != 0)
5406 		return (EFAULT);
5407 
5408 	DK_EFI2VD_EFI(&dk_efi, (vd_efi_t *)to);
5409 
5410 	kmem_free(dk_efi.dki_data, dk_efi.dki_length);
5411 
5412 	return (0);
5413 }
5414 
5415 /*
5416  * Function:
5417  *	vdc_create_fake_geometry()
5418  *
5419  * Description:
5420  *	This routine fakes up the disk info needed for some DKIO ioctls.
5421  *		- DKIOCINFO
5422  *		- DKIOCGMEDIAINFO
5423  *
5424  *	[ just like lofi(7D) and ramdisk(7D) ]
5425  *
5426  * Arguments:
5427  *	vdc	- soft state pointer for this instance of the device driver.
5428  *
5429  * Return Code:
5430  *	none.
5431  */
5432 static void
5433 vdc_create_fake_geometry(vdc_t *vdc)
5434 {
5435 	ASSERT(vdc != NULL);
5436 	ASSERT(vdc->vdisk_size != 0);
5437 	ASSERT(vdc->max_xfer_sz != 0);
5438 
5439 	/*
5440 	 * DKIOCINFO support
5441 	 */
5442 	if (vdc->cinfo == NULL)
5443 		vdc->cinfo = kmem_zalloc(sizeof (struct dk_cinfo), KM_SLEEP);
5444 
5445 	(void) strcpy(vdc->cinfo->dki_cname, VDC_DRIVER_NAME);
5446 	(void) strcpy(vdc->cinfo->dki_dname, VDC_DRIVER_NAME);
5447 	/* max_xfer_sz is #blocks so we don't need to divide by DEV_BSIZE */
5448 	vdc->cinfo->dki_maxtransfer = vdc->max_xfer_sz;
5449 	/*
5450 	 * We currently set the controller type to DKC_DIRECT for any disk.
5451 	 * When SCSI support is implemented, we will eventually change this
5452 	 * type to DKC_SCSI_CCS for disks supporting the SCSI protocol.
5453 	 */
5454 	vdc->cinfo->dki_ctype = DKC_DIRECT;
5455 	vdc->cinfo->dki_flags = DKI_FMTVOL;
5456 	vdc->cinfo->dki_cnum = 0;
5457 	vdc->cinfo->dki_addr = 0;
5458 	vdc->cinfo->dki_space = 0;
5459 	vdc->cinfo->dki_prio = 0;
5460 	vdc->cinfo->dki_vec = 0;
5461 	vdc->cinfo->dki_unit = vdc->instance;
5462 	vdc->cinfo->dki_slave = 0;
5463 	/*
5464 	 * The partition number will be created on the fly depending on the
5465 	 * actual slice (i.e. minor node) that is used to request the data.
5466 	 */
5467 	vdc->cinfo->dki_partition = 0;
5468 
5469 	/*
5470 	 * DKIOCGMEDIAINFO support
5471 	 */
5472 	if (vdc->minfo == NULL)
5473 		vdc->minfo = kmem_zalloc(sizeof (struct dk_minfo), KM_SLEEP);
5474 	vdc->minfo->dki_media_type = DK_FIXED_DISK;
5475 	vdc->minfo->dki_capacity = vdc->vdisk_size;
5476 	vdc->minfo->dki_lbsize = DEV_BSIZE;
5477 }
5478 
5479 static ushort_t
5480 vdc_lbl2cksum(struct dk_label *label)
5481 {
5482 	int	count;
5483 	ushort_t sum, *sp;
5484 
5485 	count =	(sizeof (struct dk_label)) / (sizeof (short)) - 1;
5486 	sp = (ushort_t *)label;
5487 	sum = 0;
5488 	while (count--) {
5489 		sum ^= *sp++;
5490 	}
5491 
5492 	return (sum);
5493 }
5494 
5495 /*
5496  * Function:
5497  *	vdc_validate_geometry
5498  *
5499  * Description:
5500  *	This routine discovers the label and geometry of the disk. It stores
5501  *	the disk label and related information in the vdc structure. If it
5502  *	fails to validate the geometry or to discover the disk label then
5503  *	the label is marked as unknown (VD_DISK_LABEL_UNK).
5504  *
5505  * Arguments:
5506  *	vdc	- soft state pointer for this instance of the device driver.
5507  *
5508  * Return Code:
5509  *	0	- success.
5510  *	EINVAL	- unknown disk label.
5511  *	ENOTSUP	- geometry not applicable (EFI label).
5512  *	EIO	- error accessing the disk.
5513  */
5514 static int
5515 vdc_validate_geometry(vdc_t *vdc)
5516 {
5517 	buf_t	*buf;	/* BREAD requests need to be in a buf_t structure */
5518 	dev_t	dev;
5519 	int	rv;
5520 	struct dk_label label;
5521 	struct dk_geom geom;
5522 	struct vtoc vtoc;
5523 
5524 	ASSERT(vdc != NULL);
5525 	ASSERT(vdc->vtoc != NULL && vdc->geom != NULL);
5526 	ASSERT(MUTEX_HELD(&vdc->lock));
5527 
5528 	mutex_exit(&vdc->lock);
5529 
5530 	dev = makedevice(ddi_driver_major(vdc->dip),
5531 	    VD_MAKE_DEV(vdc->instance, 0));
5532 
5533 	rv = vd_process_ioctl(dev, DKIOCGGEOM, (caddr_t)&geom, FKIOCTL);
5534 	if (rv == 0)
5535 		rv = vd_process_ioctl(dev, DKIOCGVTOC, (caddr_t)&vtoc, FKIOCTL);
5536 
5537 	if (rv == ENOTSUP) {
5538 		/*
5539 		 * If the device does not support VTOC then we try
5540 		 * to read an EFI label.
5541 		 */
5542 		struct dk_gpt *efi;
5543 		size_t efi_len;
5544 
5545 		rv = vdc_efi_alloc_and_read(dev, &efi, &efi_len);
5546 
5547 		if (rv) {
5548 			DMSG(vdc, 0, "[%d] Failed to get EFI (err=%d)",
5549 			    vdc->instance, rv);
5550 			mutex_enter(&vdc->lock);
5551 			vdc_store_label_unk(vdc);
5552 			return (EIO);
5553 		}
5554 
5555 		mutex_enter(&vdc->lock);
5556 		vdc_store_label_efi(vdc, efi);
5557 		vd_efi_free(efi, efi_len);
5558 		return (ENOTSUP);
5559 	}
5560 
5561 	if (rv != 0) {
5562 		DMSG(vdc, 0, "[%d] Failed to get VTOC (err=%d)",
5563 		    vdc->instance, rv);
5564 		mutex_enter(&vdc->lock);
5565 		vdc_store_label_unk(vdc);
5566 		if (rv != EINVAL)
5567 			rv = EIO;
5568 		return (rv);
5569 	}
5570 
5571 	/* check that geometry and vtoc are valid */
5572 	if (geom.dkg_nhead == 0 || geom.dkg_nsect == 0 ||
5573 	    vtoc.v_sanity != VTOC_SANE) {
5574 		mutex_enter(&vdc->lock);
5575 		vdc_store_label_unk(vdc);
5576 		return (EINVAL);
5577 	}
5578 
5579 	/*
5580 	 * We have a disk and a valid VTOC. However this does not mean
5581 	 * that the disk currently have a VTOC label. The returned VTOC may
5582 	 * be a default VTOC to be used for configuring the disk (this is
5583 	 * what is done for disk image). So we read the label from the
5584 	 * beginning of the disk to ensure we really have a VTOC label.
5585 	 *
5586 	 * FUTURE: This could be the default way for reading the VTOC
5587 	 * from the disk as opposed to sending the VD_OP_GET_VTOC
5588 	 * to the server. This will be the default if vdc is implemented
5589 	 * ontop of cmlb.
5590 	 */
5591 
5592 	/*
5593 	 * Single slice disk does not support read using an absolute disk
5594 	 * offset so we just rely on the DKIOCGVTOC ioctl in that case.
5595 	 */
5596 	if (vdc->vdisk_type == VD_DISK_TYPE_SLICE) {
5597 		mutex_enter(&vdc->lock);
5598 		if (vtoc.v_nparts != 1) {
5599 			vdc_store_label_unk(vdc);
5600 			return (EINVAL);
5601 		}
5602 		vdc_store_label_vtoc(vdc, &geom, &vtoc);
5603 		return (0);
5604 	}
5605 
5606 	if (vtoc.v_nparts != V_NUMPAR) {
5607 		mutex_enter(&vdc->lock);
5608 		vdc_store_label_unk(vdc);
5609 		return (EINVAL);
5610 	}
5611 
5612 	/*
5613 	 * Read disk label from start of disk
5614 	 */
5615 	buf = kmem_alloc(sizeof (buf_t), KM_SLEEP);
5616 	bioinit(buf);
5617 	buf->b_un.b_addr = (caddr_t)&label;
5618 	buf->b_bcount = DK_LABEL_SIZE;
5619 	buf->b_flags = B_BUSY | B_READ;
5620 	buf->b_dev = dev;
5621 	rv = vdc_send_request(vdc, VD_OP_BREAD, (caddr_t)&label,
5622 	    DK_LABEL_SIZE, VD_SLICE_NONE, 0, CB_STRATEGY, buf, VIO_read_dir);
5623 	if (rv) {
5624 		DMSG(vdc, 1, "[%d] Failed to read disk block 0\n",
5625 		    vdc->instance);
5626 	} else {
5627 		rv = biowait(buf);
5628 		biofini(buf);
5629 	}
5630 	kmem_free(buf, sizeof (buf_t));
5631 
5632 	if (rv != 0 || label.dkl_magic != DKL_MAGIC ||
5633 	    label.dkl_cksum != vdc_lbl2cksum(&label)) {
5634 		DMSG(vdc, 1, "[%d] Got VTOC with invalid label\n",
5635 		    vdc->instance);
5636 		mutex_enter(&vdc->lock);
5637 		vdc_store_label_unk(vdc);
5638 		return (EINVAL);
5639 	}
5640 
5641 	mutex_enter(&vdc->lock);
5642 	vdc_store_label_vtoc(vdc, &geom, &vtoc);
5643 	return (0);
5644 }
5645 
5646 /*
5647  * Function:
5648  *	vdc_validate
5649  *
5650  * Description:
5651  *	This routine discovers the label of the disk and create the
5652  *	appropriate device nodes if the label has changed.
5653  *
5654  * Arguments:
5655  *	vdc	- soft state pointer for this instance of the device driver.
5656  *
5657  * Return Code:
5658  *	none.
5659  */
5660 static void
5661 vdc_validate(vdc_t *vdc)
5662 {
5663 	vd_disk_label_t old_label;
5664 	struct vtoc old_vtoc;
5665 	int rv;
5666 
5667 	ASSERT(!MUTEX_HELD(&vdc->lock));
5668 
5669 	mutex_enter(&vdc->lock);
5670 
5671 	/* save the current label and vtoc */
5672 	old_label = vdc->vdisk_label;
5673 	bcopy(vdc->vtoc, &old_vtoc, sizeof (struct vtoc));
5674 
5675 	/* check the geometry */
5676 	(void) vdc_validate_geometry(vdc);
5677 
5678 	/* if the disk label has changed, update device nodes */
5679 	if (vdc->vdisk_label != old_label) {
5680 
5681 		if (vdc->vdisk_label == VD_DISK_LABEL_EFI)
5682 			rv = vdc_create_device_nodes_efi(vdc);
5683 		else
5684 			rv = vdc_create_device_nodes_vtoc(vdc);
5685 
5686 		if (rv != 0) {
5687 			DMSG(vdc, 0, "![%d] Failed to update device nodes",
5688 			    vdc->instance);
5689 		}
5690 	}
5691 
5692 	/* if the vtoc has changed, update device nodes properties */
5693 	if (bcmp(vdc->vtoc, &old_vtoc, sizeof (struct vtoc)) != 0) {
5694 
5695 		if (vdc_create_device_nodes_props(vdc) != 0) {
5696 			DMSG(vdc, 0, "![%d] Failed to update device nodes"
5697 			    " properties", vdc->instance);
5698 		}
5699 	}
5700 
5701 	mutex_exit(&vdc->lock);
5702 }
5703 
5704 static void
5705 vdc_validate_task(void *arg)
5706 {
5707 	vdc_t *vdc = (vdc_t *)arg;
5708 
5709 	vdc_validate(vdc);
5710 
5711 	mutex_enter(&vdc->lock);
5712 	ASSERT(vdc->validate_pending > 0);
5713 	vdc->validate_pending--;
5714 	mutex_exit(&vdc->lock);
5715 }
5716 
5717 /*
5718  * Function:
5719  *	vdc_setup_devid()
5720  *
5721  * Description:
5722  *	This routine discovers the devid of a vDisk. It requests the devid of
5723  *	the underlying device from the vDisk server, builds an encapsulated
5724  *	devid based on the retrieved devid and registers that new devid to
5725  *	the vDisk.
5726  *
5727  * Arguments:
5728  *	vdc	- soft state pointer for this instance of the device driver.
5729  *
5730  * Return Code:
5731  *	0	- A devid was succesfully registered for the vDisk
5732  */
5733 static int
5734 vdc_setup_devid(vdc_t *vdc)
5735 {
5736 	int rv;
5737 	vd_devid_t *vd_devid;
5738 	size_t bufsize, bufid_len;
5739 
5740 	/*
5741 	 * At first sight, we don't know the size of the devid that the
5742 	 * server will return but this size will be encoded into the
5743 	 * reply. So we do a first request using a default size then we
5744 	 * check if this size was large enough. If not then we do a second
5745 	 * request with the correct size returned by the server. Note that
5746 	 * ldc requires size to be 8-byte aligned.
5747 	 */
5748 	bufsize = P2ROUNDUP(VD_DEVID_SIZE(VD_DEVID_DEFAULT_LEN),
5749 	    sizeof (uint64_t));
5750 	vd_devid = kmem_zalloc(bufsize, KM_SLEEP);
5751 	bufid_len = bufsize - sizeof (vd_efi_t) - 1;
5752 
5753 	rv = vdc_do_sync_op(vdc, VD_OP_GET_DEVID, (caddr_t)vd_devid,
5754 	    bufsize, 0, 0, CB_SYNC, 0, VIO_both_dir);
5755 
5756 	DMSG(vdc, 2, "sync_op returned %d\n", rv);
5757 
5758 	if (rv) {
5759 		kmem_free(vd_devid, bufsize);
5760 		return (rv);
5761 	}
5762 
5763 	if (vd_devid->length > bufid_len) {
5764 		/*
5765 		 * The returned devid is larger than the buffer used. Try again
5766 		 * with a buffer with the right size.
5767 		 */
5768 		kmem_free(vd_devid, bufsize);
5769 		bufsize = P2ROUNDUP(VD_DEVID_SIZE(vd_devid->length),
5770 		    sizeof (uint64_t));
5771 		vd_devid = kmem_zalloc(bufsize, KM_SLEEP);
5772 		bufid_len = bufsize - sizeof (vd_efi_t) - 1;
5773 
5774 		rv = vdc_do_sync_op(vdc, VD_OP_GET_DEVID,
5775 		    (caddr_t)vd_devid, bufsize, 0, 0, CB_SYNC, 0,
5776 		    VIO_both_dir);
5777 
5778 		if (rv) {
5779 			kmem_free(vd_devid, bufsize);
5780 			return (rv);
5781 		}
5782 	}
5783 
5784 	/*
5785 	 * The virtual disk should have the same device id as the one associated
5786 	 * with the physical disk it is mapped on, otherwise sharing a disk
5787 	 * between a LDom and a non-LDom may not work (for example for a shared
5788 	 * SVM disk set).
5789 	 *
5790 	 * The DDI framework does not allow creating a device id with any
5791 	 * type so we first create a device id of type DEVID_ENCAP and then
5792 	 * we restore the orignal type of the physical device.
5793 	 */
5794 
5795 	DMSG(vdc, 2, ": devid length = %d\n", vd_devid->length);
5796 
5797 	/* build an encapsulated devid based on the returned devid */
5798 	if (ddi_devid_init(vdc->dip, DEVID_ENCAP, vd_devid->length,
5799 	    vd_devid->id, &vdc->devid) != DDI_SUCCESS) {
5800 		DMSG(vdc, 1, "[%d] Fail to created devid\n", vdc->instance);
5801 		kmem_free(vd_devid, bufsize);
5802 		return (1);
5803 	}
5804 
5805 	DEVID_FORMTYPE((impl_devid_t *)vdc->devid, vd_devid->type);
5806 
5807 	ASSERT(ddi_devid_valid(vdc->devid) == DDI_SUCCESS);
5808 
5809 	kmem_free(vd_devid, bufsize);
5810 
5811 	if (ddi_devid_register(vdc->dip, vdc->devid) != DDI_SUCCESS) {
5812 		DMSG(vdc, 1, "[%d] Fail to register devid\n", vdc->instance);
5813 		return (1);
5814 	}
5815 
5816 	return (0);
5817 }
5818 
5819 static void
5820 vdc_store_label_efi(vdc_t *vdc, struct dk_gpt *efi)
5821 {
5822 	struct vtoc *vtoc = vdc->vtoc;
5823 
5824 	ASSERT(MUTEX_HELD(&vdc->lock));
5825 
5826 	vdc->vdisk_label = VD_DISK_LABEL_EFI;
5827 	bzero(vdc->geom, sizeof (struct dk_geom));
5828 	vd_efi_to_vtoc(efi, vtoc);
5829 	if (vdc->vdisk_type == VD_DISK_TYPE_SLICE) {
5830 		/*
5831 		 * vd_efi_to_vtoc() will store information about the EFI Sun
5832 		 * reserved partition (representing the entire disk) into
5833 		 * partition 7. However single-slice device will only have
5834 		 * that single partition and the vdc driver expects to find
5835 		 * information about that partition in slice 0. So we need
5836 		 * to copy information from slice 7 to slice 0.
5837 		 */
5838 		vtoc->v_part[0].p_tag = vtoc->v_part[VD_EFI_WD_SLICE].p_tag;
5839 		vtoc->v_part[0].p_flag = vtoc->v_part[VD_EFI_WD_SLICE].p_flag;
5840 		vtoc->v_part[0].p_start = vtoc->v_part[VD_EFI_WD_SLICE].p_start;
5841 		vtoc->v_part[0].p_size =  vtoc->v_part[VD_EFI_WD_SLICE].p_size;
5842 	}
5843 }
5844 
5845 static void
5846 vdc_store_label_vtoc(vdc_t *vdc, struct dk_geom *geom, struct vtoc *vtoc)
5847 {
5848 	ASSERT(MUTEX_HELD(&vdc->lock));
5849 
5850 	vdc->vdisk_label = VD_DISK_LABEL_VTOC;
5851 	bcopy(vtoc, vdc->vtoc, sizeof (struct vtoc));
5852 	bcopy(geom, vdc->geom, sizeof (struct dk_geom));
5853 }
5854 
5855 static void
5856 vdc_store_label_unk(vdc_t *vdc)
5857 {
5858 	ASSERT(MUTEX_HELD(&vdc->lock));
5859 
5860 	vdc->vdisk_label = VD_DISK_LABEL_UNK;
5861 	bzero(vdc->vtoc, sizeof (struct vtoc));
5862 	bzero(vdc->geom, sizeof (struct dk_geom));
5863 }
5864