xref: /titanic_51/usr/src/uts/sun4v/io/vdc.c (revision f808c858fa61e7769218966759510a8b1190dfcf)
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 2006 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/cm.h>
84 #include <sys/dktp/fdisk.h>
85 #include <sys/scsi/generic/sense.h>
86 #include <sys/scsi/impl/uscsi.h>	/* Needed for defn of USCSICMD ioctl */
87 #include <sys/scsi/targets/sddef.h>
88 
89 #include <sys/ldoms.h>
90 #include <sys/ldc.h>
91 #include <sys/vio_common.h>
92 #include <sys/vio_mailbox.h>
93 #include <sys/vdsk_common.h>
94 #include <sys/vdsk_mailbox.h>
95 #include <sys/vdc.h>
96 
97 /*
98  * function prototypes
99  */
100 
101 /* standard driver functions */
102 static int	vdc_open(dev_t *dev, int flag, int otyp, cred_t *cred);
103 static int	vdc_close(dev_t dev, int flag, int otyp, cred_t *cred);
104 static int	vdc_strategy(struct buf *buf);
105 static int	vdc_print(dev_t dev, char *str);
106 static int	vdc_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblk);
107 static int	vdc_read(dev_t dev, struct uio *uio, cred_t *cred);
108 static int	vdc_write(dev_t dev, struct uio *uio, cred_t *cred);
109 static int	vdc_ioctl(dev_t dev, int cmd, intptr_t arg, int mode,
110 			cred_t *credp, int *rvalp);
111 static int	vdc_aread(dev_t dev, struct aio_req *aio, cred_t *cred);
112 static int	vdc_awrite(dev_t dev, struct aio_req *aio, cred_t *cred);
113 
114 static int	vdc_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd,
115 			void *arg, void **resultp);
116 static int	vdc_attach(dev_info_t *dip, ddi_attach_cmd_t cmd);
117 static int	vdc_detach(dev_info_t *dip, ddi_detach_cmd_t cmd);
118 
119 /* setup */
120 static int	vdc_send(vdc_t *vdc, caddr_t pkt, size_t *msglen);
121 static int	vdc_do_ldc_init(vdc_t *vdc);
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_props(vdc_t *vdc);
125 static int	vdc_get_ldc_id(dev_info_t *dip, uint64_t *ldc_id);
126 static int	vdc_do_ldc_up(vdc_t *vdc);
127 static void	vdc_terminate_ldc(vdc_t *vdc);
128 static int	vdc_init_descriptor_ring(vdc_t *vdc);
129 static void	vdc_destroy_descriptor_ring(vdc_t *vdc);
130 
131 /* handshake with vds */
132 static void		vdc_init_handshake_negotiation(void *arg);
133 static int		vdc_init_ver_negotiation(vdc_t *vdc, vio_ver_t ver);
134 static int		vdc_init_attr_negotiation(vdc_t *vdc);
135 static int		vdc_init_dring_negotiate(vdc_t *vdc);
136 static void		vdc_reset_connection(vdc_t *vdc, boolean_t resetldc);
137 static boolean_t	vdc_is_able_to_tx_data(vdc_t *vdc, int flag);
138 static boolean_t	vdc_is_supported_version(vio_ver_msg_t *ver_msg);
139 
140 /* processing incoming messages from vDisk server */
141 static void	vdc_process_msg_thread(vdc_t *vdc);
142 static void	vdc_process_msg(void *arg);
143 static void	vdc_do_process_msg(vdc_t *vdc);
144 static uint_t	vdc_handle_cb(uint64_t event, caddr_t arg);
145 static int	vdc_process_ctrl_msg(vdc_t *vdc, vio_msg_t msg);
146 static int	vdc_process_data_msg(vdc_t *vdc, vio_msg_t msg);
147 static int	vdc_process_err_msg(vdc_t *vdc, vio_msg_t msg);
148 static int	vdc_handle_ver_msg(vdc_t *vdc, vio_ver_msg_t *ver_msg);
149 static int	vdc_handle_attr_msg(vdc_t *vdc, vd_attr_msg_t *attr_msg);
150 static int	vdc_handle_dring_reg_msg(vdc_t *vdc, vio_dring_reg_msg_t *msg);
151 static int	vdc_get_next_dring_entry_id(vdc_t *vdc, uint_t needed);
152 static int	vdc_populate_descriptor(vdc_t *vdc, caddr_t addr,
153 			size_t nbytes, int op, uint64_t arg, uint64_t slice);
154 static int	vdc_wait_for_descriptor_update(vdc_t *vdc, uint_t idx,
155 			vio_dring_msg_t dmsg);
156 static int	vdc_depopulate_descriptor(vdc_t *vdc, uint_t idx);
157 static int	vdc_populate_mem_hdl(vdc_t *vdc, uint_t idx,
158 			caddr_t addr, size_t nbytes, int operation);
159 static int	vdc_verify_seq_num(vdc_t *vdc, vio_dring_msg_t *dring_msg);
160 
161 /* dkio */
162 static int	vd_process_ioctl(dev_t dev, int cmd, caddr_t arg, int mode);
163 static int	vdc_create_fake_geometry(vdc_t *vdc);
164 static int	vdc_setup_disk_layout(vdc_t *vdc);
165 static int	vdc_null_copy_func(vdc_t *vdc, void *from, void *to,
166 		    int mode, int dir);
167 static int	vdc_get_vtoc_convert(vdc_t *vdc, void *from, void *to,
168 		    int mode, int dir);
169 static int	vdc_set_vtoc_convert(vdc_t *vdc, void *from, void *to,
170 		    int mode, int dir);
171 static int	vdc_get_geom_convert(vdc_t *vdc, void *from, void *to,
172 		    int mode, int dir);
173 static int	vdc_set_geom_convert(vdc_t *vdc, void *from, void *to,
174 		    int mode, int dir);
175 static int	vdc_uscsicmd_convert(vdc_t *vdc, void *from, void *to,
176 		    int mode, int dir);
177 
178 /*
179  * Module variables
180  */
181 
182 /*
183  * Tunable variables to control how long vdc waits before timing out on
184  * various operations
185  */
186 static int	vdc_retries = 10;
187 
188 /* calculated from 'vdc_usec_timeout' during attach */
189 static uint64_t	vdc_hz_timeout;				/* units: Hz */
190 static uint64_t	vdc_usec_timeout = 30 * MICROSEC;	/* 30s units: ns */
191 
192 static uint64_t	vdc_hz_timeout_ldc;			/* units: Hz */
193 static uint64_t	vdc_usec_timeout_ldc = 10 * MILLISEC;	/* 0.01s units: ns */
194 
195 /* values for dumping - need to run in a tighter loop */
196 static uint64_t	vdc_usec_timeout_dump = 100 * MILLISEC;	/* 0.1s units: ns */
197 static int	vdc_dump_retries = 100;
198 
199 /* Count of the number of vdc instances attached */
200 static volatile uint32_t	vdc_instance_count = 0;
201 
202 /* Soft state pointer */
203 static void	*vdc_state;
204 
205 /* variable level controlling the verbosity of the error/debug messages */
206 int	vdc_msglevel = 0;
207 
208 /*
209  * Supported vDisk protocol version pairs.
210  *
211  * The first array entry is the latest and preferred version.
212  */
213 static const vio_ver_t	vdc_version[] = {{1, 0}};
214 
215 static struct cb_ops vdc_cb_ops = {
216 	vdc_open,	/* cb_open */
217 	vdc_close,	/* cb_close */
218 	vdc_strategy,	/* cb_strategy */
219 	vdc_print,	/* cb_print */
220 	vdc_dump,	/* cb_dump */
221 	vdc_read,	/* cb_read */
222 	vdc_write,	/* cb_write */
223 	vdc_ioctl,	/* cb_ioctl */
224 	nodev,		/* cb_devmap */
225 	nodev,		/* cb_mmap */
226 	nodev,		/* cb_segmap */
227 	nochpoll,	/* cb_chpoll */
228 	ddi_prop_op,	/* cb_prop_op */
229 	NULL,		/* cb_str */
230 	D_MP | D_64BIT,	/* cb_flag */
231 	CB_REV,		/* cb_rev */
232 	vdc_aread,	/* cb_aread */
233 	vdc_awrite	/* cb_awrite */
234 };
235 
236 static struct dev_ops vdc_ops = {
237 	DEVO_REV,	/* devo_rev */
238 	0,		/* devo_refcnt */
239 	vdc_getinfo,	/* devo_getinfo */
240 	nulldev,	/* devo_identify */
241 	nulldev,	/* devo_probe */
242 	vdc_attach,	/* devo_attach */
243 	vdc_detach,	/* devo_detach */
244 	nodev,		/* devo_reset */
245 	&vdc_cb_ops,	/* devo_cb_ops */
246 	NULL,		/* devo_bus_ops */
247 	nulldev		/* devo_power */
248 };
249 
250 static struct modldrv modldrv = {
251 	&mod_driverops,
252 	"virtual disk client %I%",
253 	&vdc_ops,
254 };
255 
256 static struct modlinkage modlinkage = {
257 	MODREV_1,
258 	&modldrv,
259 	NULL
260 };
261 
262 /* -------------------------------------------------------------------------- */
263 
264 /*
265  * Device Driver housekeeping and setup
266  */
267 
268 int
269 _init(void)
270 {
271 	int	status;
272 
273 	if ((status = ddi_soft_state_init(&vdc_state, sizeof (vdc_t), 1)) != 0)
274 		return (status);
275 	if ((status = mod_install(&modlinkage)) != 0)
276 		ddi_soft_state_fini(&vdc_state);
277 	return (status);
278 }
279 
280 int
281 _info(struct modinfo *modinfop)
282 {
283 	return (mod_info(&modlinkage, modinfop));
284 }
285 
286 int
287 _fini(void)
288 {
289 	int	status;
290 
291 	if ((status = mod_remove(&modlinkage)) != 0)
292 		return (status);
293 	ddi_soft_state_fini(&vdc_state);
294 	return (0);
295 }
296 
297 static int
298 vdc_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd,  void *arg, void **resultp)
299 {
300 	_NOTE(ARGUNUSED(dip))
301 
302 	int	instance = SDUNIT(getminor((dev_t)arg));
303 	vdc_t	*vdc = NULL;
304 
305 	switch (cmd) {
306 	case DDI_INFO_DEVT2DEVINFO:
307 		if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
308 			*resultp = NULL;
309 			return (DDI_FAILURE);
310 		}
311 		*resultp = vdc->dip;
312 		return (DDI_SUCCESS);
313 	case DDI_INFO_DEVT2INSTANCE:
314 		*resultp = (void *)(uintptr_t)instance;
315 		return (DDI_SUCCESS);
316 	default:
317 		*resultp = NULL;
318 		return (DDI_FAILURE);
319 	}
320 }
321 
322 static int
323 vdc_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
324 {
325 	int	instance;
326 	int	rv;
327 	uint_t	retries = 0;
328 	vdc_t	*vdc = NULL;
329 
330 	switch (cmd) {
331 	case DDI_DETACH:
332 		/* the real work happens below */
333 		break;
334 	case DDI_SUSPEND:
335 		/* nothing to do for this non-device */
336 		return (DDI_SUCCESS);
337 	default:
338 		return (DDI_FAILURE);
339 	}
340 
341 	ASSERT(cmd == DDI_DETACH);
342 	instance = ddi_get_instance(dip);
343 	DMSG(1, "[%d] Entered\n", instance);
344 
345 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
346 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
347 		return (DDI_FAILURE);
348 	}
349 
350 	if (vdc->open) {
351 		DMSG(0, "[%d] Cannot detach: device is open", instance);
352 		return (DDI_FAILURE);
353 	}
354 
355 	DMSG(0, "[%d] proceeding...\n", instance);
356 
357 	/*
358 	 * try and disable callbacks to prevent another handshake
359 	 */
360 	rv = ldc_set_cb_mode(vdc->ldc_handle, LDC_CB_DISABLE);
361 	DMSG(0, "[%d] callback disabled (rv=%d)\n", instance, rv);
362 
363 	/*
364 	 * Prevent any more attempts to start a handshake with the vdisk
365 	 * server and tear down the existing connection.
366 	 */
367 	mutex_enter(&vdc->lock);
368 	vdc->initialized |= VDC_HANDSHAKE_STOP;
369 	vdc_reset_connection(vdc, B_TRUE);
370 	mutex_exit(&vdc->lock);
371 
372 	if (vdc->initialized & VDC_THREAD) {
373 		mutex_enter(&vdc->msg_proc_lock);
374 		vdc->msg_proc_thr_state = VDC_THR_STOP;
375 		vdc->msg_pending = B_TRUE;
376 		cv_signal(&vdc->msg_proc_cv);
377 
378 		while (vdc->msg_proc_thr_state != VDC_THR_DONE) {
379 			DMSG(0, "[%d] Waiting for thread to exit\n", instance);
380 			rv = cv_timedwait(&vdc->msg_proc_cv,
381 				&vdc->msg_proc_lock,
382 				VD_GET_TIMEOUT_HZ(vdc_hz_timeout, 1));
383 			if ((rv == -1) && (retries++ > vdc_retries))
384 				break;
385 		}
386 		mutex_exit(&vdc->msg_proc_lock);
387 	}
388 
389 	mutex_enter(&vdc->lock);
390 
391 	if (vdc->initialized & VDC_DRING)
392 		vdc_destroy_descriptor_ring(vdc);
393 
394 	if (vdc->initialized & VDC_LDC)
395 		vdc_terminate_ldc(vdc);
396 
397 	mutex_exit(&vdc->lock);
398 
399 	if (vdc->initialized & VDC_MINOR) {
400 		ddi_prop_remove_all(dip);
401 		ddi_remove_minor_node(dip, NULL);
402 	}
403 
404 	if (vdc->initialized & VDC_LOCKS) {
405 		mutex_destroy(&vdc->lock);
406 		mutex_destroy(&vdc->attach_lock);
407 		mutex_destroy(&vdc->msg_proc_lock);
408 		mutex_destroy(&vdc->dring_lock);
409 		cv_destroy(&vdc->cv);
410 		cv_destroy(&vdc->attach_cv);
411 		cv_destroy(&vdc->msg_proc_cv);
412 	}
413 
414 	if (vdc->minfo)
415 		kmem_free(vdc->minfo, sizeof (struct dk_minfo));
416 
417 	if (vdc->cinfo)
418 		kmem_free(vdc->cinfo, sizeof (struct dk_cinfo));
419 
420 	if (vdc->vtoc)
421 		kmem_free(vdc->vtoc, sizeof (struct vtoc));
422 
423 	if (vdc->label)
424 		kmem_free(vdc->label, DK_LABEL_SIZE);
425 
426 	if (vdc->initialized & VDC_SOFT_STATE)
427 		ddi_soft_state_free(vdc_state, instance);
428 
429 	DMSG(0, "[%d] End %p\n", instance, (void *)vdc);
430 
431 	return (DDI_SUCCESS);
432 }
433 
434 
435 static int
436 vdc_do_attach(dev_info_t *dip)
437 {
438 	int		instance;
439 	vdc_t		*vdc = NULL;
440 	int		status;
441 	uint_t		retries = 0;
442 
443 	ASSERT(dip != NULL);
444 
445 	instance = ddi_get_instance(dip);
446 	if (ddi_soft_state_zalloc(vdc_state, instance) != DDI_SUCCESS) {
447 		cmn_err(CE_NOTE, "[%d] Couldn't alloc state structure",
448 		    instance);
449 		return (DDI_FAILURE);
450 	}
451 
452 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
453 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
454 		return (DDI_FAILURE);
455 	}
456 
457 	/*
458 	 * We assign the value to initialized in this case to zero out the
459 	 * variable and then set bits in it to indicate what has been done
460 	 */
461 	vdc->initialized = VDC_SOFT_STATE;
462 
463 	vdc_hz_timeout = drv_usectohz(vdc_usec_timeout);
464 	vdc_hz_timeout_ldc = drv_usectohz(vdc_usec_timeout_ldc);
465 
466 	vdc->dip	= dip;
467 	vdc->instance	= instance;
468 	vdc->open	= 0;
469 	vdc->vdisk_type	= VD_DISK_TYPE_UNK;
470 	vdc->state	= VD_STATE_INIT;
471 	vdc->ldc_state	= 0;
472 	vdc->session_id = 0;
473 	vdc->block_size = DEV_BSIZE;
474 	vdc->max_xfer_sz = maxphys / DEV_BSIZE;
475 
476 	vdc->vtoc = NULL;
477 	vdc->cinfo = NULL;
478 	vdc->minfo = NULL;
479 
480 	mutex_init(&vdc->lock, NULL, MUTEX_DRIVER, NULL);
481 	mutex_init(&vdc->attach_lock, NULL, MUTEX_DRIVER, NULL);
482 	mutex_init(&vdc->msg_proc_lock, NULL, MUTEX_DRIVER, NULL);
483 	mutex_init(&vdc->dring_lock, NULL, MUTEX_DRIVER, NULL);
484 	cv_init(&vdc->cv, NULL, CV_DRIVER, NULL);
485 	cv_init(&vdc->attach_cv, NULL, CV_DRIVER, NULL);
486 	cv_init(&vdc->msg_proc_cv, NULL, CV_DRIVER, NULL);
487 	vdc->initialized |= VDC_LOCKS;
488 
489 	vdc->msg_pending = B_FALSE;
490 	vdc->msg_proc_thr_id = thread_create(NULL, 0, vdc_process_msg_thread,
491 		vdc, 0, &p0, TS_RUN, minclsyspri);
492 	if (vdc->msg_proc_thr_id == NULL) {
493 		cmn_err(CE_NOTE, "[%d] Failed to create msg processing thread",
494 				instance);
495 		return (DDI_FAILURE);
496 	}
497 	vdc->initialized |= VDC_THREAD;
498 
499 	/* initialise LDC channel which will be used to communicate with vds */
500 	if (vdc_do_ldc_init(vdc) != 0) {
501 		cmn_err(CE_NOTE, "[%d] Couldn't initialize LDC", instance);
502 		return (DDI_FAILURE);
503 	}
504 
505 	/* Bring up connection with vds via LDC */
506 	status = vdc_start_ldc_connection(vdc);
507 	if (status != 0) {
508 		cmn_err(CE_NOTE, "[%d] Could not start LDC", instance);
509 		return (DDI_FAILURE);
510 	}
511 
512 	/*
513 	 * We need to wait until the handshake has completed before leaving
514 	 * the attach(). If this is the first vdc device attached (i.e. the root
515 	 * filesystem) we will wait much longer in the hope that we can finally
516 	 * communicate with the vDisk server (the service domain may be
517 	 * rebooting, etc.). This wait is necessary so that the device node(s)
518 	 * are created before the attach(9E) return (otherwise the open(9E) will
519 	 * fail and and the root file system will not boot).
520 	 */
521 	atomic_inc_32(&vdc_instance_count);
522 	mutex_enter(&vdc->attach_lock);
523 	while ((vdc->ldc_state != LDC_UP) || (vdc->state != VD_STATE_DATA)) {
524 
525 		DMSG(0, "[%d] handshake in progress [VD %d (LDC %d)]\n",
526 			instance, vdc->state, vdc->ldc_state);
527 
528 		status = cv_timedwait(&vdc->attach_cv, &vdc->attach_lock,
529 				VD_GET_TIMEOUT_HZ(vdc_hz_timeout, retries));
530 		if (status == -1) {
531 			/*
532 			 * If this is not the first instance attached or we
533 			 * have exceeeded the max number of retries we give
534 			 * up waiting and do not delay the attach any longer
535 			 */
536 			if ((vdc_instance_count != 1) ||
537 			    (retries >= vdc_retries)) {
538 				DMSG(0, "[%d] Giving up wait for handshake\n",
539 						instance);
540 				mutex_exit(&vdc->attach_lock);
541 				return (DDI_FAILURE);
542 			} else {
543 				DMSG(0, "[%d] Retry #%d for handshake.\n",
544 						instance, retries);
545 				vdc_init_handshake_negotiation(vdc);
546 				retries++;
547 			}
548 		}
549 	}
550 	mutex_exit(&vdc->attach_lock);
551 
552 	/*
553 	 * Once the handshake is complete, we can use the DRing to send
554 	 * requests to the vDisk server to calculate the geometry and
555 	 * VTOC of the "disk"
556 	 */
557 	status = vdc_setup_disk_layout(vdc);
558 	if (status != 0) {
559 		cmn_err(CE_NOTE, "[%d] Failed to discover disk layout (err%d)",
560 				vdc->instance, status);
561 	}
562 
563 	/*
564 	 * Now that we have the device info we can create the
565 	 * device nodes and properties
566 	 */
567 	status = vdc_create_device_nodes(vdc);
568 	if (status) {
569 		cmn_err(CE_NOTE, "[%d] Failed to create device nodes",
570 				instance);
571 		return (status);
572 	}
573 	status = vdc_create_device_nodes_props(vdc);
574 	if (status) {
575 		cmn_err(CE_NOTE, "[%d] Failed to create device nodes"
576 				" properties (%d)", instance, status);
577 		return (status);
578 	}
579 
580 	ddi_report_dev(dip);
581 
582 	DMSG(0, "[%d] Attach completed\n", instance);
583 	return (status);
584 }
585 
586 static int
587 vdc_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
588 {
589 	int	status;
590 
591 	switch (cmd) {
592 	case DDI_ATTACH:
593 		if ((status = vdc_do_attach(dip)) != 0)
594 			(void) vdc_detach(dip, DDI_DETACH);
595 		return (status);
596 	case DDI_RESUME:
597 		/* nothing to do for this non-device */
598 		return (DDI_SUCCESS);
599 	default:
600 		return (DDI_FAILURE);
601 	}
602 }
603 
604 static int
605 vdc_do_ldc_init(vdc_t *vdc)
606 {
607 	int			status = 0;
608 	ldc_status_t		ldc_state;
609 	ldc_attr_t		ldc_attr;
610 	uint64_t		ldc_id = 0;
611 	dev_info_t		*dip = NULL;
612 
613 	ASSERT(vdc != NULL);
614 
615 	dip = vdc->dip;
616 	vdc->initialized |= VDC_LDC;
617 
618 	if ((status = vdc_get_ldc_id(dip, &ldc_id)) != 0) {
619 		cmn_err(CE_NOTE, "[%d] Failed to get LDC channel ID property",
620 				vdc->instance);
621 		return (EIO);
622 	}
623 	vdc->ldc_id = ldc_id;
624 
625 	ldc_attr.devclass = LDC_DEV_BLK;
626 	ldc_attr.instance = vdc->instance;
627 	ldc_attr.mode = LDC_MODE_UNRELIABLE;	/* unreliable transport */
628 	ldc_attr.mtu = VD_LDC_MTU;
629 
630 	if ((vdc->initialized & VDC_LDC_INIT) == 0) {
631 		status = ldc_init(ldc_id, &ldc_attr, &vdc->ldc_handle);
632 		if (status != 0) {
633 			cmn_err(CE_NOTE, "[%d] ldc_init(chan %ld) returned %d",
634 					vdc->instance, ldc_id, status);
635 			return (status);
636 		}
637 		vdc->initialized |= VDC_LDC_INIT;
638 	}
639 	status = ldc_status(vdc->ldc_handle, &ldc_state);
640 	if (status != 0) {
641 		cmn_err(CE_NOTE, "[%d] Cannot discover LDC status [err=%d]",
642 				vdc->instance, status);
643 		return (status);
644 	}
645 	vdc->ldc_state = ldc_state;
646 
647 	if ((vdc->initialized & VDC_LDC_CB) == 0) {
648 		status = ldc_reg_callback(vdc->ldc_handle, vdc_handle_cb,
649 		    (caddr_t)vdc);
650 		if (status != 0) {
651 			cmn_err(CE_NOTE, "[%d] LDC callback reg. failed (%d)",
652 					vdc->instance, status);
653 			return (status);
654 		}
655 		vdc->initialized |= VDC_LDC_CB;
656 	}
657 
658 	vdc->initialized |= VDC_LDC;
659 
660 	/*
661 	 * At this stage we have initialised LDC, we will now try and open
662 	 * the connection.
663 	 */
664 	if (vdc->ldc_state == LDC_INIT) {
665 		status = ldc_open(vdc->ldc_handle);
666 		if (status != 0) {
667 			cmn_err(CE_NOTE, "[%d] ldc_open(chan %ld) returned %d",
668 					vdc->instance, vdc->ldc_id, status);
669 			return (status);
670 		}
671 		vdc->initialized |= VDC_LDC_OPEN;
672 	}
673 
674 	return (status);
675 }
676 
677 static int
678 vdc_start_ldc_connection(vdc_t *vdc)
679 {
680 	int		status = 0;
681 
682 	ASSERT(vdc != NULL);
683 
684 	mutex_enter(&vdc->lock);
685 
686 	if (vdc->ldc_state == LDC_UP) {
687 		DMSG(0, "[%d] LDC is already UP ..\n", vdc->instance);
688 		mutex_exit(&vdc->lock);
689 		return (0);
690 	}
691 
692 	status = vdc_do_ldc_up(vdc);
693 
694 	DMSG(0, "[%d] Finished bringing up LDC\n", vdc->instance);
695 
696 	mutex_exit(&vdc->lock);
697 
698 	return (status);
699 }
700 
701 
702 /*
703  * Function:
704  *	vdc_create_device_nodes
705  *
706  * Description:
707  *	This function creates the block and character device nodes under
708  *	/devices along with the node properties. It is called as part of
709  *	the attach(9E) of the instance during the handshake with vds after
710  *	vds has sent the attributes to vdc.
711  *
712  *	If the device is of type VD_DISK_TYPE_SLICE then the minor node
713  *	of 2 is used in keeping with the Solaris convention that slice 2
714  *	refers to a whole disk. Slices start at 'a'
715  *
716  * Parameters:
717  *	vdc 		- soft state pointer
718  *
719  * Return Values
720  *	0		- Success
721  *	EIO		- Failed to create node
722  *	EINVAL		- Unknown type of disk exported
723  */
724 static int
725 vdc_create_device_nodes(vdc_t *vdc)
726 {
727 	/* uses NNNN which is OK as long as # of disks <= 10000 */
728 	char		name[sizeof ("disk@NNNN:s,raw")];
729 	dev_info_t	*dip = NULL;
730 	int		instance;
731 	int		num_slices = 1;
732 	int		i;
733 
734 	ASSERT(vdc != NULL);
735 
736 	instance = vdc->instance;
737 	dip = vdc->dip;
738 
739 	switch (vdc->vdisk_type) {
740 	case VD_DISK_TYPE_DISK:
741 		num_slices = V_NUMPAR;
742 		break;
743 	case VD_DISK_TYPE_SLICE:
744 		num_slices = 1;
745 		break;
746 	case VD_DISK_TYPE_UNK:
747 	default:
748 		return (EINVAL);
749 	}
750 
751 	for (i = 0; i < num_slices; i++) {
752 		(void) snprintf(name, sizeof (name), "%c", 'a' + i);
753 		if (ddi_create_minor_node(dip, name, S_IFBLK,
754 		    VD_MAKE_DEV(instance, i), DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
755 			cmn_err(CE_NOTE, "[%d] Couldn't add block node '%s'",
756 				instance, name);
757 			return (EIO);
758 		}
759 
760 		/* if any device node is created we set this flag */
761 		vdc->initialized |= VDC_MINOR;
762 
763 		(void) snprintf(name, sizeof (name), "%c%s",
764 			'a' + i, ",raw");
765 		if (ddi_create_minor_node(dip, name, S_IFCHR,
766 		    VD_MAKE_DEV(instance, i), DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
767 			cmn_err(CE_NOTE, "[%d] Couldn't add raw node '%s'",
768 				instance, name);
769 			return (EIO);
770 		}
771 	}
772 
773 	return (0);
774 }
775 
776 /*
777  * Function:
778  *	vdc_create_device_nodes_props
779  *
780  * Description:
781  *	This function creates the block and character device nodes under
782  *	/devices along with the node properties. It is called as part of
783  *	the attach(9E) of the instance during the handshake with vds after
784  *	vds has sent the attributes to vdc.
785  *
786  * Parameters:
787  *	vdc 		- soft state pointer
788  *
789  * Return Values
790  *	0		- Success
791  *	EIO		- Failed to create device node property
792  *	EINVAL		- Unknown type of disk exported
793  */
794 static int
795 vdc_create_device_nodes_props(vdc_t *vdc)
796 {
797 	dev_info_t	*dip = NULL;
798 	int		instance;
799 	int		num_slices = 1;
800 	int64_t		size = 0;
801 	dev_t		dev;
802 	int		rv;
803 	int		i;
804 
805 	ASSERT(vdc != NULL);
806 
807 	instance = vdc->instance;
808 	dip = vdc->dip;
809 
810 	if ((vdc->vtoc == NULL) || (vdc->vtoc->v_sanity != VTOC_SANE)) {
811 		cmn_err(CE_NOTE, "![%d] Could not create device node property."
812 				" No VTOC available", instance);
813 		return (ENXIO);
814 	}
815 
816 	switch (vdc->vdisk_type) {
817 	case VD_DISK_TYPE_DISK:
818 		num_slices = V_NUMPAR;
819 		break;
820 	case VD_DISK_TYPE_SLICE:
821 		num_slices = 1;
822 		break;
823 	case VD_DISK_TYPE_UNK:
824 	default:
825 		return (EINVAL);
826 	}
827 
828 	for (i = 0; i < num_slices; i++) {
829 		dev = makedevice(ddi_driver_major(dip),
830 			VD_MAKE_DEV(instance, i));
831 
832 		size = vdc->vtoc->v_part[i].p_size * vdc->vtoc->v_sectorsz;
833 		DMSG(0, "[%d] sz %ld (%ld Mb)  p_size %lx\n",
834 				instance, size, size / (1024 * 1024),
835 				vdc->vtoc->v_part[i].p_size);
836 
837 		rv = ddi_prop_update_int64(dev, dip, VDC_SIZE_PROP_NAME, size);
838 		if (rv != DDI_PROP_SUCCESS) {
839 			cmn_err(CE_NOTE, "[%d] Couldn't add '%s' prop of [%ld]",
840 				instance, VDC_SIZE_PROP_NAME, size);
841 			return (EIO);
842 		}
843 
844 		rv = ddi_prop_update_int64(dev, dip, VDC_NBLOCKS_PROP_NAME,
845 			lbtodb(size));
846 		if (rv != DDI_PROP_SUCCESS) {
847 			cmn_err(CE_NOTE, "[%d] Couldn't add '%s' prop [%llu]",
848 				instance, VDC_NBLOCKS_PROP_NAME, lbtodb(size));
849 			return (EIO);
850 		}
851 	}
852 
853 	return (0);
854 }
855 
856 static int
857 vdc_open(dev_t *dev, int flag, int otyp, cred_t *cred)
858 {
859 	_NOTE(ARGUNUSED(cred))
860 
861 	int		instance;
862 	vdc_t		*vdc;
863 
864 	ASSERT(dev != NULL);
865 	instance = SDUNIT(getminor(*dev));
866 
867 	DMSG(0, "[%d] minor = %d flag = %x, otyp = %x\n",
868 			instance, getminor(*dev), flag, otyp);
869 
870 	if ((otyp != OTYP_CHR) && (otyp != OTYP_BLK))
871 		return (EINVAL);
872 
873 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
874 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
875 		return (ENXIO);
876 	}
877 
878 	/*
879 	 * Check to see if we can communicate with vds
880 	 */
881 	if (!vdc_is_able_to_tx_data(vdc, flag)) {
882 		DMSG(0, "[%d] Not ready to transmit data (flag=%x)\n",
883 				instance, flag);
884 		return (ENOLINK);
885 	}
886 
887 	mutex_enter(&vdc->lock);
888 	vdc->open++;
889 	mutex_exit(&vdc->lock);
890 
891 	return (0);
892 }
893 
894 static int
895 vdc_close(dev_t dev, int flag, int otyp, cred_t *cred)
896 {
897 	_NOTE(ARGUNUSED(cred))
898 
899 	int	instance;
900 	vdc_t	*vdc;
901 
902 	instance = SDUNIT(getminor(dev));
903 
904 	DMSG(0, "[%d] flag = %x, otyp = %x\n", instance, flag, otyp);
905 
906 	if ((otyp != OTYP_CHR) && (otyp != OTYP_BLK))
907 		return (EINVAL);
908 
909 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
910 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
911 		return (ENXIO);
912 	}
913 
914 	/*
915 	 * Check to see if we can communicate with vds
916 	 */
917 	if (!vdc_is_able_to_tx_data(vdc, 0)) {
918 		DMSG(0, "[%d] Not ready to transmit data (flag=%x)\n",
919 				instance, flag);
920 		return (ETIMEDOUT);
921 	}
922 
923 	if (vdc->dkio_flush_pending) {
924 		DMSG(0, "[%d] Cannot detach: %d outstanding DKIO flushes\n",
925 			instance, vdc->dkio_flush_pending);
926 		return (EBUSY);
927 	}
928 
929 	/*
930 	 * Should not need the mutex here, since the framework should protect
931 	 * against more opens on this device, but just in case.
932 	 */
933 	mutex_enter(&vdc->lock);
934 	vdc->open--;
935 	mutex_exit(&vdc->lock);
936 
937 	return (0);
938 }
939 
940 static int
941 vdc_ioctl(dev_t dev, int cmd, intptr_t arg, int mode, cred_t *credp, int *rvalp)
942 {
943 	_NOTE(ARGUNUSED(credp))
944 	_NOTE(ARGUNUSED(rvalp))
945 
946 	return (vd_process_ioctl(dev, cmd, (caddr_t)arg, mode));
947 }
948 
949 static int
950 vdc_print(dev_t dev, char *str)
951 {
952 	cmn_err(CE_NOTE, "vdc%d:  %s", SDUNIT(getminor(dev)), str);
953 	return (0);
954 }
955 
956 static int
957 vdc_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblk)
958 {
959 	buf_t	*buf;	/* BWRITE requests need to be in a buf_t structure */
960 	int	rv;
961 	size_t	nbytes = nblk * DEV_BSIZE;
962 	int	instance = SDUNIT(getminor(dev));
963 	vdc_t	*vdc = NULL;
964 
965 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
966 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
967 		return (ENXIO);
968 	}
969 
970 	buf = kmem_alloc(sizeof (buf_t), KM_SLEEP);
971 	bioinit(buf);
972 	buf->b_un.b_addr = addr;
973 	buf->b_bcount = nbytes;
974 	buf->b_flags = B_BUSY | B_WRITE;
975 	buf->b_dev = dev;
976 	rv = vdc_populate_descriptor(vdc, (caddr_t)buf, nbytes,
977 			VD_OP_BWRITE, blkno, SDPART(getminor(dev)));
978 
979 	/*
980 	 * If the OS instance is panicking, the call above will ensure that
981 	 * the descriptor is done before returning. This should always be
982 	 * case when coming through this function but we check just in case
983 	 * and wait if necessary for the vDisk server to ACK and trigger
984 	 * the biodone.
985 	 */
986 	if (!ddi_in_panic())
987 		rv = biowait(buf);
988 
989 	biofini(buf);
990 	kmem_free(buf, sizeof (buf_t));
991 
992 	DMSG(1, "[%d] status=%d\n", instance, rv);
993 
994 	return (rv);
995 }
996 
997 /* -------------------------------------------------------------------------- */
998 
999 /*
1000  * Disk access routines
1001  *
1002  */
1003 
1004 /*
1005  * vdc_strategy()
1006  *
1007  * Return Value:
1008  *	0:	As per strategy(9E), the strategy() function must return 0
1009  *		[ bioerror(9f) sets b_flags to the proper error code ]
1010  */
1011 static int
1012 vdc_strategy(struct buf *buf)
1013 {
1014 	int		rv = -1;
1015 	vdc_t		*vdc = NULL;
1016 	int		instance = SDUNIT(getminor(buf->b_edev));
1017 	int	op = (buf->b_flags & B_READ) ? VD_OP_BREAD : VD_OP_BWRITE;
1018 
1019 	DMSG(2, "[%d] %s %ld bytes at block %llx : b_addr=0x%p",
1020 	    instance, (buf->b_flags & B_READ) ? "Read" : "Write",
1021 	    buf->b_bcount, buf->b_lblkno, (void *)buf->b_un.b_addr);
1022 
1023 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1024 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1025 		bioerror(buf, ENXIO);
1026 		biodone(buf);
1027 		return (0);
1028 	}
1029 
1030 	DTRACE_IO2(vstart, buf_t *, buf, vdc_t *, vdc);
1031 
1032 	if (!vdc_is_able_to_tx_data(vdc, O_NONBLOCK)) {
1033 		DMSG(0, "[%d] Not ready to transmit data\n", instance);
1034 		bioerror(buf, ENXIO);
1035 		biodone(buf);
1036 		return (0);
1037 	}
1038 	bp_mapin(buf);
1039 
1040 	rv = vdc_populate_descriptor(vdc, (caddr_t)buf, buf->b_bcount, op,
1041 			buf->b_lblkno, SDPART(getminor(buf->b_edev)));
1042 
1043 	/*
1044 	 * If the request was successfully sent, the strategy call returns and
1045 	 * the ACK handler calls the bioxxx functions when the vDisk server is
1046 	 * done.
1047 	 */
1048 	if (rv) {
1049 		DMSG(0, "[%d] Failed to read/write (err=%d)\n", instance, rv);
1050 		bioerror(buf, rv);
1051 		biodone(buf);
1052 	}
1053 
1054 	return (0);
1055 }
1056 
1057 
1058 static int
1059 vdc_read(dev_t dev, struct uio *uio, cred_t *cred)
1060 {
1061 	_NOTE(ARGUNUSED(cred))
1062 
1063 	DMSG(1, "[%d] Entered", SDUNIT(getminor(dev)));
1064 	return (physio(vdc_strategy, NULL, dev, B_READ, minphys, uio));
1065 }
1066 
1067 static int
1068 vdc_write(dev_t dev, struct uio *uio, cred_t *cred)
1069 {
1070 	_NOTE(ARGUNUSED(cred))
1071 
1072 	DMSG(1, "[%d] Entered", SDUNIT(getminor(dev)));
1073 	return (physio(vdc_strategy, NULL, dev, B_WRITE, minphys, uio));
1074 }
1075 
1076 static int
1077 vdc_aread(dev_t dev, struct aio_req *aio, cred_t *cred)
1078 {
1079 	_NOTE(ARGUNUSED(cred))
1080 
1081 	DMSG(1, "[%d] Entered", SDUNIT(getminor(dev)));
1082 	return (aphysio(vdc_strategy, anocancel, dev, B_READ, minphys, aio));
1083 }
1084 
1085 static int
1086 vdc_awrite(dev_t dev, struct aio_req *aio, cred_t *cred)
1087 {
1088 	_NOTE(ARGUNUSED(cred))
1089 
1090 	DMSG(1, "[%d] Entered", SDUNIT(getminor(dev)));
1091 	return (aphysio(vdc_strategy, anocancel, dev, B_WRITE, minphys, aio));
1092 }
1093 
1094 
1095 /* -------------------------------------------------------------------------- */
1096 
1097 /*
1098  * Handshake support
1099  */
1100 
1101 /*
1102  * vdc_init_handshake_negotiation
1103  *
1104  * Description:
1105  *	This function is called to trigger the handshake negotiations between
1106  *	the client (vdc) and the server (vds). It may be called multiple times.
1107  *
1108  * Parameters:
1109  *	vdc - soft state pointer
1110  */
1111 static void
1112 vdc_init_handshake_negotiation(void *arg)
1113 {
1114 	vdc_t		*vdc = (vdc_t *)(void *)arg;
1115 	ldc_status_t	ldc_state;
1116 	vd_state_t	state;
1117 	int		status;
1118 
1119 	ASSERT(vdc != NULL);
1120 
1121 	DMSG(0, "[%d] Initializing vdc<->vds handshake\n", vdc->instance);
1122 
1123 	/* get LDC state */
1124 	status = ldc_status(vdc->ldc_handle, &ldc_state);
1125 	if (status != 0) {
1126 		cmn_err(CE_NOTE, "[%d] Couldn't get LDC status (err=%d)",
1127 				vdc->instance, status);
1128 		return;
1129 	}
1130 
1131 	/*
1132 	 * If the LDC connection is not UP we bring it up now and return.
1133 	 * The handshake will be started again when the callback is
1134 	 * triggered due to the UP event.
1135 	 */
1136 	if (ldc_state != LDC_UP) {
1137 		DMSG(0, "[%d] Triggering LDC_UP & returning\n", vdc->instance);
1138 		(void) vdc_do_ldc_up(vdc);
1139 		return;
1140 	}
1141 
1142 	mutex_enter(&vdc->lock);
1143 	/*
1144 	 * Do not continue if another thread has triggered a handshake which
1145 	 * has not been reset or detach() has stopped further handshakes.
1146 	 */
1147 	if (vdc->initialized & (VDC_HANDSHAKE | VDC_HANDSHAKE_STOP)) {
1148 		DMSG(0, "[%d] Negotiation not triggered. [init=%x]\n",
1149 			vdc->instance, vdc->initialized);
1150 		mutex_exit(&vdc->lock);
1151 		return;
1152 	}
1153 
1154 	if (vdc->hshake_cnt++ > vdc_retries) {
1155 		cmn_err(CE_NOTE, "[%d] Failed repeatedly to complete handshake"
1156 				"with vDisk server", vdc->instance);
1157 		mutex_exit(&vdc->lock);
1158 		return;
1159 	}
1160 
1161 	vdc->initialized |= VDC_HANDSHAKE;
1162 	vdc->ldc_state = ldc_state;
1163 
1164 	state = vdc->state;
1165 
1166 	if (state == VD_STATE_INIT) {
1167 		/*
1168 		 * Set the desired version parameter to the first entry in the
1169 		 * version array. If this specific version is not supported,
1170 		 * the response handling code will step down the version number
1171 		 * to the next array entry and deal with it accordingly.
1172 		 */
1173 		(void) vdc_init_ver_negotiation(vdc, vdc_version[0]);
1174 	} else if (state == VD_STATE_VER) {
1175 		(void) vdc_init_attr_negotiation(vdc);
1176 	} else if (state == VD_STATE_ATTR) {
1177 		(void) vdc_init_dring_negotiate(vdc);
1178 	} else if (state == VD_STATE_DATA) {
1179 		/*
1180 		 * nothing to do - we have already completed the negotiation
1181 		 * and we can transmit data when ready.
1182 		 */
1183 		DMSG(0, "[%d] Negotiation triggered after handshake completed",
1184 			vdc->instance);
1185 	}
1186 
1187 	mutex_exit(&vdc->lock);
1188 }
1189 
1190 /*
1191  * Function:
1192  *	vdc_init_ver_negotiation()
1193  *
1194  * Description:
1195  *
1196  * Arguments:
1197  *	vdc	- soft state pointer for this instance of the device driver.
1198  *
1199  * Return Code:
1200  *	0	- Success
1201  */
1202 static int
1203 vdc_init_ver_negotiation(vdc_t *vdc, vio_ver_t ver)
1204 {
1205 	vio_ver_msg_t	pkt;
1206 	size_t		msglen = sizeof (pkt);
1207 	int		status = -1;
1208 
1209 	ASSERT(vdc != NULL);
1210 	ASSERT(mutex_owned(&vdc->lock));
1211 
1212 	DMSG(0, "[%d] Entered.\n", vdc->instance);
1213 
1214 	/*
1215 	 * set the Session ID to a unique value
1216 	 * (the lower 32 bits of the clock tick)
1217 	 */
1218 	vdc->session_id = ((uint32_t)gettick() & 0xffffffff);
1219 	DMSG(0, "[%d] Set SID to 0x%lx\n", vdc->instance, vdc->session_id);
1220 
1221 	pkt.tag.vio_msgtype = VIO_TYPE_CTRL;
1222 	pkt.tag.vio_subtype = VIO_SUBTYPE_INFO;
1223 	pkt.tag.vio_subtype_env = VIO_VER_INFO;
1224 	pkt.tag.vio_sid = vdc->session_id;
1225 	pkt.dev_class = VDEV_DISK;
1226 	pkt.ver_major = ver.major;
1227 	pkt.ver_minor = ver.minor;
1228 
1229 	status = vdc_send(vdc, (caddr_t)&pkt, &msglen);
1230 	DMSG(0, "[%d] Ver info sent (status = %d)\n", vdc->instance, status);
1231 
1232 	if ((status != 0) || (msglen != sizeof (vio_ver_msg_t))) {
1233 		cmn_err(CE_NOTE, "[%d] Failed to send Ver negotiation info: "
1234 				"id(%lx) rv(%d) size(%ld)",
1235 				vdc->instance, vdc->ldc_handle,
1236 				status, msglen);
1237 		if (msglen != sizeof (vio_ver_msg_t))
1238 			status = ENOMSG;
1239 	}
1240 
1241 	return (status);
1242 }
1243 
1244 /*
1245  * Function:
1246  *	vdc_init_attr_negotiation()
1247  *
1248  * Description:
1249  *
1250  * Arguments:
1251  *	vdc	- soft state pointer for this instance of the device driver.
1252  *
1253  * Return Code:
1254  *	0	- Success
1255  */
1256 static int
1257 vdc_init_attr_negotiation(vdc_t *vdc)
1258 {
1259 	vd_attr_msg_t	pkt;
1260 	size_t		msglen = sizeof (pkt);
1261 	int		status;
1262 
1263 	ASSERT(vdc != NULL);
1264 	ASSERT(mutex_owned(&vdc->lock));
1265 
1266 	DMSG(0, "[%d] entered\n", vdc->instance);
1267 
1268 	/* fill in tag */
1269 	pkt.tag.vio_msgtype = VIO_TYPE_CTRL;
1270 	pkt.tag.vio_subtype = VIO_SUBTYPE_INFO;
1271 	pkt.tag.vio_subtype_env = VIO_ATTR_INFO;
1272 	pkt.tag.vio_sid = vdc->session_id;
1273 	/* fill in payload */
1274 	pkt.max_xfer_sz = vdc->max_xfer_sz;
1275 	pkt.vdisk_block_size = vdc->block_size;
1276 	pkt.xfer_mode = VIO_DRING_MODE;
1277 	pkt.operations = 0;	/* server will set bits of valid operations */
1278 	pkt.vdisk_type = 0;	/* server will set to valid device type */
1279 	pkt.vdisk_size = 0;	/* server will set to valid size */
1280 
1281 	status = vdc_send(vdc, (caddr_t)&pkt, &msglen);
1282 	DMSG(0, "[%d] Attr info sent (status = %d)\n", vdc->instance, status);
1283 
1284 	if ((status != 0) || (msglen != sizeof (vio_ver_msg_t))) {
1285 		cmn_err(CE_NOTE, "[%d] Failed to send Attr negotiation info: "
1286 				"id(%lx) rv(%d) size(%ld)",
1287 				vdc->instance, vdc->ldc_handle,
1288 				status, msglen);
1289 		if (msglen != sizeof (vio_ver_msg_t))
1290 			status = ENOMSG;
1291 	}
1292 
1293 	return (status);
1294 }
1295 
1296 /*
1297  * Function:
1298  *	vdc_init_dring_negotiate()
1299  *
1300  * Description:
1301  *
1302  * Arguments:
1303  *	vdc	- soft state pointer for this instance of the device driver.
1304  *
1305  * Return Code:
1306  *	0	- Success
1307  */
1308 static int
1309 vdc_init_dring_negotiate(vdc_t *vdc)
1310 {
1311 	vio_dring_reg_msg_t	pkt;
1312 	size_t			msglen = sizeof (pkt);
1313 	int			status = -1;
1314 
1315 	ASSERT(vdc != NULL);
1316 	ASSERT(mutex_owned(&vdc->lock));
1317 
1318 	status = vdc_init_descriptor_ring(vdc);
1319 	if (status != 0) {
1320 		cmn_err(CE_CONT, "[%d] Failed to init DRing (status = %d)\n",
1321 				vdc->instance, status);
1322 		vdc_destroy_descriptor_ring(vdc);
1323 		vdc_reset_connection(vdc, B_TRUE);
1324 		return (status);
1325 	}
1326 	DMSG(0, "[%d] Init of descriptor ring completed (status = %d)\n",
1327 			vdc->instance, status);
1328 
1329 	/* fill in tag */
1330 	pkt.tag.vio_msgtype = VIO_TYPE_CTRL;
1331 	pkt.tag.vio_subtype = VIO_SUBTYPE_INFO;
1332 	pkt.tag.vio_subtype_env = VIO_DRING_REG;
1333 	pkt.tag.vio_sid = vdc->session_id;
1334 	/* fill in payload */
1335 	pkt.dring_ident = 0;
1336 	pkt.num_descriptors = vdc->dring_len;
1337 	pkt.descriptor_size = vdc->dring_entry_size;
1338 	pkt.options = (VIO_TX_DRING | VIO_RX_DRING);
1339 	pkt.ncookies = vdc->dring_cookie_count;
1340 	pkt.cookie[0] = vdc->dring_cookie[0];	/* for now just one cookie */
1341 
1342 	status = vdc_send(vdc, (caddr_t)&pkt, &msglen);
1343 	if (status != 0) {
1344 		cmn_err(CE_NOTE, "[%d] Failed to register DRing (err = %d)",
1345 				vdc->instance, status);
1346 		vdc_reset_connection(vdc, B_TRUE);
1347 	}
1348 
1349 	return (status);
1350 }
1351 
1352 
1353 /* -------------------------------------------------------------------------- */
1354 
1355 /*
1356  * LDC helper routines
1357  */
1358 
1359 /*
1360  * Function:
1361  *	vdc_send()
1362  *
1363  * Description:
1364  *	The function encapsulates the call to write a message using LDC.
1365  *	If LDC indicates that the call failed due to the queue being full,
1366  *	we retry the ldc_write() [ up to 'vdc_retries' time ], otherwise
1367  *	we return the error returned by LDC.
1368  *
1369  * Arguments:
1370  *	ldc_handle	- LDC handle for the channel this instance of vdc uses
1371  *	pkt		- address of LDC message to be sent
1372  *	msglen		- the size of the message being sent. When the function
1373  *			  returns, this contains the number of bytes written.
1374  *
1375  * Return Code:
1376  *	0		- Success.
1377  *	EINVAL		- pkt or msglen were NULL
1378  *	ECONNRESET	- The connection was not up.
1379  *	EWOULDBLOCK	- LDC queue is full
1380  *	xxx		- other error codes returned by ldc_write
1381  */
1382 static int
1383 vdc_send(vdc_t *vdc, caddr_t pkt, size_t *msglen)
1384 {
1385 	size_t	size = 0;
1386 	int	retries = 0;
1387 	int	status = 0;
1388 
1389 	ASSERT(vdc != NULL);
1390 	ASSERT(mutex_owned(&vdc->lock));
1391 	ASSERT(msglen != NULL);
1392 	ASSERT(*msglen != 0);
1393 
1394 	do {
1395 		size = *msglen;
1396 		status = ldc_write(vdc->ldc_handle, pkt, &size);
1397 		if (status == EWOULDBLOCK)
1398 			delay(vdc_hz_timeout_ldc);
1399 	} while (status == EWOULDBLOCK && retries++ < vdc_retries);
1400 
1401 	/* if LDC had serious issues --- reset vdc state */
1402 	if (status == EIO || status == ECONNRESET) {
1403 		vdc_reset_connection(vdc, B_TRUE);
1404 	}
1405 
1406 	/* return the last size written */
1407 	*msglen = size;
1408 
1409 	return (status);
1410 }
1411 
1412 /*
1413  * Function:
1414  *	vdc_get_ldc_id()
1415  *
1416  * Description:
1417  *	This function gets the 'ldc-id' for this particular instance of vdc.
1418  *	The id returned is the guest domain channel endpoint LDC uses for
1419  *	communication with vds.
1420  *
1421  * Arguments:
1422  *	dip	- dev info pointer for this instance of the device driver.
1423  *	ldc_id	- pointer to variable used to return the 'ldc-id' found.
1424  *
1425  * Return Code:
1426  *	0	- Success.
1427  *	ENOENT	- Expected node or property did not exist.
1428  *	ENXIO	- Unexpected error communicating with MD framework
1429  */
1430 static int
1431 vdc_get_ldc_id(dev_info_t *dip, uint64_t *ldc_id)
1432 {
1433 	int		status = ENOENT;
1434 	char		*node_name = NULL;
1435 	md_t		*mdp = NULL;
1436 	int		num_nodes;
1437 	int		num_vdevs;
1438 	int		num_chans;
1439 	mde_cookie_t	rootnode;
1440 	mde_cookie_t	*listp = NULL;
1441 	mde_cookie_t	*chanp = NULL;
1442 	boolean_t	found_inst = B_FALSE;
1443 	int		listsz;
1444 	int		idx;
1445 	uint64_t	md_inst;
1446 	int		obp_inst;
1447 	int		instance = ddi_get_instance(dip);
1448 
1449 	ASSERT(ldc_id != NULL);
1450 	*ldc_id = 0;
1451 
1452 	/*
1453 	 * Get the OBP instance number for comparison with the MD instance
1454 	 *
1455 	 * The "cfg-handle" property of a vdc node in an MD contains the MD's
1456 	 * notion of "instance", or unique identifier, for that node; OBP
1457 	 * stores the value of the "cfg-handle" MD property as the value of
1458 	 * the "reg" property on the node in the device tree it builds from
1459 	 * the MD and passes to Solaris.  Thus, we look up the devinfo node's
1460 	 * "reg" property value to uniquely identify this device instance.
1461 	 * If the "reg" property cannot be found, the device tree state is
1462 	 * presumably so broken that there is no point in continuing.
1463 	 */
1464 	if (!ddi_prop_exists(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS, OBP_REG)) {
1465 		cmn_err(CE_WARN, "'%s' property does not exist", OBP_REG);
1466 		return (ENOENT);
1467 	}
1468 	obp_inst = ddi_prop_get_int(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS,
1469 			OBP_REG, -1);
1470 	DMSG(1, "[%d] OBP inst=%d\n", instance, obp_inst);
1471 
1472 	/*
1473 	 * We now walk the MD nodes and if an instance of a vdc node matches
1474 	 * the instance got from OBP we get the ldc-id property.
1475 	 */
1476 	if ((mdp = md_get_handle()) == NULL) {
1477 		cmn_err(CE_WARN, "unable to init machine description");
1478 		return (ENXIO);
1479 	}
1480 
1481 	num_nodes = md_node_count(mdp);
1482 	ASSERT(num_nodes > 0);
1483 
1484 	listsz = num_nodes * sizeof (mde_cookie_t);
1485 
1486 	/* allocate memory for nodes */
1487 	listp = kmem_zalloc(listsz, KM_SLEEP);
1488 	chanp = kmem_zalloc(listsz, KM_SLEEP);
1489 
1490 	rootnode = md_root_node(mdp);
1491 	ASSERT(rootnode != MDE_INVAL_ELEM_COOKIE);
1492 
1493 	/*
1494 	 * Search for all the virtual devices, we will then check to see which
1495 	 * ones are disk nodes.
1496 	 */
1497 	num_vdevs = md_scan_dag(mdp, rootnode,
1498 			md_find_name(mdp, VDC_MD_VDEV_NAME),
1499 			md_find_name(mdp, "fwd"), listp);
1500 
1501 	if (num_vdevs <= 0) {
1502 		cmn_err(CE_NOTE, "No '%s' node found", VDC_MD_VDEV_NAME);
1503 		status = ENOENT;
1504 		goto done;
1505 	}
1506 
1507 	DMSG(1, "[%d] num_vdevs=%d\n", instance, num_vdevs);
1508 	for (idx = 0; idx < num_vdevs; idx++) {
1509 		status = md_get_prop_str(mdp, listp[idx], "name", &node_name);
1510 		if ((status != 0) || (node_name == NULL)) {
1511 			cmn_err(CE_NOTE, "Unable to get name of node type '%s'"
1512 					": err %d", VDC_MD_VDEV_NAME, status);
1513 			continue;
1514 		}
1515 
1516 		DMSG(1, "[%d] Found node '%s'\n", instance, node_name);
1517 		if (strcmp(VDC_MD_DISK_NAME, node_name) == 0) {
1518 			status = md_get_prop_val(mdp, listp[idx],
1519 					VDC_MD_CFG_HDL, &md_inst);
1520 			DMSG(1, "[%d] vdc inst in MD=%lx\n", instance, md_inst);
1521 			if ((status == 0) && (md_inst == obp_inst)) {
1522 				found_inst = B_TRUE;
1523 				break;
1524 			}
1525 		}
1526 	}
1527 
1528 	if (!found_inst) {
1529 		cmn_err(CE_NOTE, "Unable to find correct '%s' node",
1530 				VDC_MD_DISK_NAME);
1531 		status = ENOENT;
1532 		goto done;
1533 	}
1534 	DMSG(0, "[%d] MD inst=%lx\n", instance, md_inst);
1535 
1536 	/* get the channels for this node */
1537 	num_chans = md_scan_dag(mdp, listp[idx],
1538 			md_find_name(mdp, VDC_MD_CHAN_NAME),
1539 			md_find_name(mdp, "fwd"), chanp);
1540 
1541 	/* expecting at least one channel */
1542 	if (num_chans <= 0) {
1543 		cmn_err(CE_NOTE, "No '%s' node for '%s' port",
1544 				VDC_MD_CHAN_NAME, VDC_MD_VDEV_NAME);
1545 		status = ENOENT;
1546 		goto done;
1547 
1548 	} else if (num_chans != 1) {
1549 		DMSG(0, "[%d] Expected 1 '%s' node for '%s' port, found %d\n",
1550 			instance, VDC_MD_CHAN_NAME, VDC_MD_VDEV_NAME,
1551 			num_chans);
1552 	}
1553 
1554 	/*
1555 	 * We use the first channel found (index 0), irrespective of how
1556 	 * many are there in total.
1557 	 */
1558 	if (md_get_prop_val(mdp, chanp[0], VDC_ID_PROP, ldc_id) != 0) {
1559 		cmn_err(CE_NOTE, "Channel '%s' property not found",
1560 				VDC_ID_PROP);
1561 		status = ENOENT;
1562 	}
1563 
1564 	DMSG(0, "[%d] LDC id is 0x%lx\n", instance, *ldc_id);
1565 
1566 done:
1567 	if (chanp)
1568 		kmem_free(chanp, listsz);
1569 	if (listp)
1570 		kmem_free(listp, listsz);
1571 
1572 	(void) md_fini_handle(mdp);
1573 
1574 	return (status);
1575 }
1576 
1577 static int
1578 vdc_do_ldc_up(vdc_t *vdc)
1579 {
1580 	int	status;
1581 
1582 	DMSG(0, "[%d] Bringing up channel %lx\n", vdc->instance, vdc->ldc_id);
1583 
1584 	if ((status = ldc_up(vdc->ldc_handle)) != 0) {
1585 		switch (status) {
1586 		case ECONNREFUSED:	/* listener not ready at other end */
1587 			DMSG(0, "[%d] ldc_up(%lx,...) return %d\n",
1588 					vdc->instance, vdc->ldc_id, status);
1589 			status = 0;
1590 			break;
1591 		default:
1592 			cmn_err(CE_NOTE, "[%d] Failed to bring up LDC: "
1593 					"channel=%ld, err=%d",
1594 					vdc->instance, vdc->ldc_id, status);
1595 		}
1596 	}
1597 
1598 	return (status);
1599 }
1600 
1601 
1602 /*
1603  * vdc_is_able_to_tx_data()
1604  *
1605  * Description:
1606  *	This function checks if we are able to send data to the
1607  *	vDisk server (vds). The LDC connection needs to be up and
1608  *	vdc & vds need to have completed the handshake negotiation.
1609  *
1610  * Parameters:
1611  *	vdc 		- soft state pointer
1612  *	flag		- flag to indicate if we can block or not
1613  *			  [ If O_NONBLOCK or O_NDELAY (which are defined in
1614  *			    open(2)) are set then do not block)
1615  *
1616  * Return Values
1617  *	B_TRUE		- can talk to vds
1618  *	B_FALSE		- unable to talk to vds
1619  */
1620 static boolean_t
1621 vdc_is_able_to_tx_data(vdc_t *vdc, int flag)
1622 {
1623 	vd_state_t	state;
1624 	uint32_t	ldc_state;
1625 	uint_t		retries = 0;
1626 	int		rv = -1;
1627 
1628 	ASSERT(vdc != NULL);
1629 
1630 	mutex_enter(&vdc->lock);
1631 	state = vdc->state;
1632 	ldc_state = vdc->ldc_state;
1633 	mutex_exit(&vdc->lock);
1634 
1635 	if ((state == VD_STATE_DATA) && (ldc_state == LDC_UP))
1636 		return (B_TRUE);
1637 
1638 	if ((flag & O_NONBLOCK) || (flag & O_NDELAY)) {
1639 		DMSG(0, "[%d] Not ready to tx - state %d LDC state %d\n",
1640 			vdc->instance, state, ldc_state);
1641 		return (B_FALSE);
1642 	}
1643 
1644 	/*
1645 	 * We want to check and see if any negotiations triggered earlier
1646 	 * have succeeded. We are prepared to wait a little while in case
1647 	 * they are still in progress.
1648 	 */
1649 	mutex_enter(&vdc->lock);
1650 	while ((vdc->ldc_state != LDC_UP) || (vdc->state != VD_STATE_DATA)) {
1651 		DMSG(0, "[%d] Waiting for connection. (state %d : LDC %d)\n",
1652 			vdc->instance, vdc->state, vdc->ldc_state);
1653 
1654 		rv = cv_timedwait(&vdc->cv, &vdc->lock,
1655 			VD_GET_TIMEOUT_HZ(vdc_hz_timeout, retries));
1656 
1657 		/*
1658 		 * An rv of -1 indicates that we timed out without the LDC
1659 		 * state changing so it looks like the other side (vdc) is
1660 		 * not yet ready/responding.
1661 		 *
1662 		 * Any other value of rv indicates that the LDC triggered an
1663 		 * interrupt so we just loop again, check the handshake state
1664 		 * and keep waiting if necessary.
1665 		 */
1666 		if (rv == -1) {
1667 			if (retries >= vdc_retries) {
1668 				DMSG(0, "[%d] handshake wait timed out\n",
1669 						vdc->instance);
1670 				mutex_exit(&vdc->lock);
1671 				return (B_FALSE);
1672 			} else {
1673 				DMSG(1, "[%d] Handshake retry #%d timed out\n",
1674 					vdc->instance, retries);
1675 				retries++;
1676 			}
1677 		}
1678 	}
1679 
1680 	ASSERT(vdc->ldc_state == LDC_UP);
1681 	ASSERT(vdc->state == VD_STATE_DATA);
1682 
1683 	mutex_exit(&vdc->lock);
1684 
1685 	return (B_TRUE);
1686 }
1687 
1688 
1689 /*
1690  * Function:
1691  *	vdc_terminate_ldc()
1692  *
1693  * Description:
1694  *
1695  * Arguments:
1696  *	vdc	- soft state pointer for this instance of the device driver.
1697  *
1698  * Return Code:
1699  *	None
1700  */
1701 static void
1702 vdc_terminate_ldc(vdc_t *vdc)
1703 {
1704 	int	instance = ddi_get_instance(vdc->dip);
1705 
1706 	ASSERT(vdc != NULL);
1707 	ASSERT(mutex_owned(&vdc->lock));
1708 
1709 	DMSG(0, "[%d] initialized=%x\n", instance, vdc->initialized);
1710 
1711 	if (vdc->initialized & VDC_LDC_OPEN) {
1712 		DMSG(0, "[%d] ldc_close()\n", instance);
1713 		(void) ldc_close(vdc->ldc_handle);
1714 	}
1715 	if (vdc->initialized & VDC_LDC_CB) {
1716 		DMSG(0, "[%d] ldc_unreg_callback()\n", instance);
1717 		(void) ldc_unreg_callback(vdc->ldc_handle);
1718 	}
1719 	if (vdc->initialized & VDC_LDC) {
1720 		DMSG(0, "[%d] ldc_fini()\n", instance);
1721 		(void) ldc_fini(vdc->ldc_handle);
1722 		vdc->ldc_handle = NULL;
1723 	}
1724 
1725 	vdc->initialized &= ~(VDC_LDC | VDC_LDC_CB | VDC_LDC_OPEN);
1726 }
1727 
1728 /*
1729  * Function:
1730  *	vdc_reset_connection()
1731  *
1732  * Description:
1733  *
1734  * Arguments:
1735  *	vdc	- soft state pointer for this instance of the device driver.
1736  *	reset_ldc - Flag whether or not to reset the LDC connection also.
1737  *
1738  * Return Code:
1739  *	None
1740  */
1741 static void
1742 vdc_reset_connection(vdc_t *vdc, boolean_t reset_ldc)
1743 {
1744 	int	status;
1745 
1746 	ASSERT(vdc != NULL);
1747 	ASSERT(mutex_owned(&vdc->lock));
1748 
1749 	cmn_err(CE_CONT, "?[%d] Resetting connection to vDisk server\n",
1750 			vdc->instance);
1751 
1752 	vdc->state = VD_STATE_INIT;
1753 
1754 	if (reset_ldc) {
1755 		status = ldc_down(vdc->ldc_handle);
1756 		DMSG(0, "[%d]  ldc_down() = %d\n", vdc->instance, status);
1757 	}
1758 
1759 	vdc->initialized &= ~VDC_HANDSHAKE;
1760 	DMSG(0, "[%d] initialized=%x\n", vdc->instance, vdc->initialized);
1761 }
1762 
1763 /* -------------------------------------------------------------------------- */
1764 
1765 /*
1766  * Descriptor Ring helper routines
1767  */
1768 
1769 /*
1770  * Function:
1771  *	vdc_init_descriptor_ring()
1772  *
1773  * Description:
1774  *
1775  * Arguments:
1776  *	vdc	- soft state pointer for this instance of the device driver.
1777  *
1778  * Return Code:
1779  *	0	- Success
1780  */
1781 static int
1782 vdc_init_descriptor_ring(vdc_t *vdc)
1783 {
1784 	vd_dring_entry_t	*dep = NULL;	/* DRing Entry pointer */
1785 	int	status = 0;
1786 	int	i;
1787 
1788 	DMSG(0, "[%d] initialized=%x\n", vdc->instance, vdc->initialized);
1789 
1790 	ASSERT(vdc != NULL);
1791 	ASSERT(mutex_owned(&vdc->lock));
1792 	ASSERT(vdc->ldc_handle != NULL);
1793 
1794 	/* ensure we have enough room to store max sized block */
1795 	ASSERT(maxphys <= VD_MAX_BLOCK_SIZE);
1796 
1797 	if ((vdc->initialized & VDC_DRING_INIT) == 0) {
1798 		DMSG(0, "[%d] ldc_mem_dring_create\n", vdc->instance);
1799 		/*
1800 		 * Calculate the maximum block size we can transmit using one
1801 		 * Descriptor Ring entry from the attributes returned by the
1802 		 * vDisk server. This is subject to a minimum of 'maxphys'
1803 		 * as we do not have the capability to split requests over
1804 		 * multiple DRing entries.
1805 		 */
1806 		if ((vdc->max_xfer_sz * vdc->block_size) < maxphys) {
1807 			DMSG(0, "[%d] using minimum DRing size\n",
1808 					vdc->instance);
1809 			vdc->dring_max_cookies = maxphys / PAGESIZE;
1810 		} else {
1811 			vdc->dring_max_cookies =
1812 				(vdc->max_xfer_sz * vdc->block_size) / PAGESIZE;
1813 		}
1814 		vdc->dring_entry_size = (sizeof (vd_dring_entry_t) +
1815 				(sizeof (ldc_mem_cookie_t) *
1816 					(vdc->dring_max_cookies - 1)));
1817 		vdc->dring_len = VD_DRING_LEN;
1818 
1819 		status = ldc_mem_dring_create(vdc->dring_len,
1820 				vdc->dring_entry_size, &vdc->ldc_dring_hdl);
1821 		if ((vdc->ldc_dring_hdl == NULL) || (status != 0)) {
1822 			cmn_err(CE_NOTE, "[%d] Descriptor ring creation failed",
1823 					vdc->instance);
1824 			return (status);
1825 		}
1826 		vdc->initialized |= VDC_DRING_INIT;
1827 	}
1828 
1829 	if ((vdc->initialized & VDC_DRING_BOUND) == 0) {
1830 		DMSG(0, "[%d] ldc_mem_dring_bind\n", vdc->instance);
1831 		vdc->dring_cookie =
1832 			kmem_zalloc(sizeof (ldc_mem_cookie_t), KM_SLEEP);
1833 
1834 		status = ldc_mem_dring_bind(vdc->ldc_handle, vdc->ldc_dring_hdl,
1835 				LDC_SHADOW_MAP, LDC_MEM_RW,
1836 				&vdc->dring_cookie[0],
1837 				&vdc->dring_cookie_count);
1838 		if (status != 0) {
1839 			cmn_err(CE_NOTE, "[%d] Failed to bind descriptor ring "
1840 				"(%lx) to channel (%lx)\n", vdc->instance,
1841 				vdc->ldc_dring_hdl, vdc->ldc_handle);
1842 			return (status);
1843 		}
1844 		ASSERT(vdc->dring_cookie_count == 1);
1845 		vdc->initialized |= VDC_DRING_BOUND;
1846 	}
1847 
1848 	status = ldc_mem_dring_info(vdc->ldc_dring_hdl, &vdc->dring_mem_info);
1849 	if (status != 0) {
1850 		DMSG(0, "[%d] Failed to get info for descriptor ring (%lx)\n",
1851 			vdc->instance, vdc->ldc_dring_hdl);
1852 		return (status);
1853 	}
1854 
1855 	if ((vdc->initialized & VDC_DRING_LOCAL) == 0) {
1856 		DMSG(0, "[%d] local dring\n", vdc->instance);
1857 
1858 		/* Allocate the local copy of this dring */
1859 		vdc->local_dring =
1860 			kmem_zalloc(vdc->dring_len * sizeof (vdc_local_desc_t),
1861 						KM_SLEEP);
1862 		vdc->initialized |= VDC_DRING_LOCAL;
1863 	}
1864 
1865 	/*
1866 	 * Mark all DRing entries as free and initialize the private
1867 	 * descriptor's memory handles. If any entry is initialized,
1868 	 * we need to free it later so we set the bit in 'initialized'
1869 	 * at the start.
1870 	 */
1871 	vdc->initialized |= VDC_DRING_ENTRY;
1872 	for (i = 0; i < vdc->dring_len; i++) {
1873 		dep = VDC_GET_DRING_ENTRY_PTR(vdc, i);
1874 		dep->hdr.dstate = VIO_DESC_FREE;
1875 
1876 		status = ldc_mem_alloc_handle(vdc->ldc_handle,
1877 				&vdc->local_dring[i].desc_mhdl);
1878 		if (status != 0) {
1879 			cmn_err(CE_NOTE, "![%d] Failed to alloc mem handle for"
1880 					" descriptor %d", vdc->instance, i);
1881 			return (status);
1882 		}
1883 		vdc->local_dring[i].flags = VIO_DESC_FREE;
1884 		vdc->local_dring[i].dep = dep;
1885 
1886 		mutex_init(&vdc->local_dring[i].lock, NULL, MUTEX_DRIVER, NULL);
1887 		cv_init(&vdc->local_dring[i].cv, NULL, CV_DRIVER, NULL);
1888 	}
1889 
1890 	/*
1891 	 * We init the index of the last DRing entry used. Since the code to
1892 	 * get the next available entry increments it before selecting one,
1893 	 * we set it to the last DRing entry so that it wraps around to zero
1894 	 * for the 1st entry to be used.
1895 	 */
1896 	vdc->dring_curr_idx = vdc->dring_len - 1;
1897 
1898 	vdc->dring_notify_server = B_TRUE;
1899 
1900 	return (status);
1901 }
1902 
1903 /*
1904  * Function:
1905  *	vdc_destroy_descriptor_ring()
1906  *
1907  * Description:
1908  *
1909  * Arguments:
1910  *	vdc	- soft state pointer for this instance of the device driver.
1911  *
1912  * Return Code:
1913  *	None
1914  */
1915 static void
1916 vdc_destroy_descriptor_ring(vdc_t *vdc)
1917 {
1918 	vdc_local_desc_t	*ldep = NULL;	/* Local Dring Entry Pointer */
1919 	ldc_mem_handle_t	mhdl = NULL;
1920 	int			status = -1;
1921 	int			i;	/* loop */
1922 
1923 	ASSERT(vdc != NULL);
1924 	ASSERT(mutex_owned(&vdc->lock));
1925 	ASSERT(vdc->state == VD_STATE_INIT);
1926 
1927 	DMSG(0, "[%d] Entered\n", vdc->instance);
1928 
1929 	if (vdc->initialized & VDC_DRING_ENTRY) {
1930 		DMSG(0, "[%d] Removing Local DRing entries\n", vdc->instance);
1931 		for (i = 0; i < vdc->dring_len; i++) {
1932 			ldep = &vdc->local_dring[i];
1933 			mhdl = ldep->desc_mhdl;
1934 
1935 			if (mhdl == NULL)
1936 				continue;
1937 
1938 			(void) ldc_mem_free_handle(mhdl);
1939 			mutex_destroy(&ldep->lock);
1940 			cv_destroy(&ldep->cv);
1941 		}
1942 		vdc->initialized &= ~VDC_DRING_ENTRY;
1943 	}
1944 
1945 	if (vdc->initialized & VDC_DRING_LOCAL) {
1946 		DMSG(0, "[%d] Freeing Local DRing\n", vdc->instance);
1947 		kmem_free(vdc->local_dring,
1948 				vdc->dring_len * sizeof (vdc_local_desc_t));
1949 		vdc->initialized &= ~VDC_DRING_LOCAL;
1950 	}
1951 
1952 	if (vdc->initialized & VDC_DRING_BOUND) {
1953 		DMSG(0, "[%d] Unbinding DRing\n", vdc->instance);
1954 		status = ldc_mem_dring_unbind(vdc->ldc_dring_hdl);
1955 		if (status == 0) {
1956 			vdc->initialized &= ~VDC_DRING_BOUND;
1957 		} else {
1958 			cmn_err(CE_NOTE, "[%d] Error %d unbinding DRing %lx",
1959 				vdc->instance, status, vdc->ldc_dring_hdl);
1960 		}
1961 	}
1962 
1963 	if (vdc->initialized & VDC_DRING_INIT) {
1964 		DMSG(0, "[%d] Destroying DRing\n", vdc->instance);
1965 		status = ldc_mem_dring_destroy(vdc->ldc_dring_hdl);
1966 		if (status == 0) {
1967 			vdc->ldc_dring_hdl = NULL;
1968 			bzero(&vdc->dring_mem_info, sizeof (ldc_mem_info_t));
1969 			vdc->initialized &= ~VDC_DRING_INIT;
1970 		} else {
1971 			cmn_err(CE_NOTE, "[%d] Error %d destroying DRing (%lx)",
1972 				vdc->instance, status, vdc->ldc_dring_hdl);
1973 		}
1974 	}
1975 }
1976 
1977 /*
1978  * vdc_get_next_dring_entry_idx()
1979  *
1980  * Description:
1981  *	This function gets the index of the next Descriptor Ring entry available
1982  *	If the ring is full, it will back off and wait for the next entry to be
1983  *	freed (the ACK handler will signal).
1984  *
1985  * Return Value:
1986  *	0 <= rv < vdc->dring_len		Next available slot
1987  *	-1 				DRing is full
1988  */
1989 static int
1990 vdc_get_next_dring_entry_idx(vdc_t *vdc, uint_t num_slots_needed)
1991 {
1992 	_NOTE(ARGUNUSED(num_slots_needed))
1993 
1994 	vd_dring_entry_t	*dep = NULL;	/* DRing Entry Pointer */
1995 	vdc_local_desc_t	*ldep = NULL;	/* Local DRing Entry Pointer */
1996 	int			idx = -1;
1997 
1998 	ASSERT(vdc != NULL);
1999 	ASSERT(vdc->dring_len == vdc->dring_len);
2000 	ASSERT(vdc->dring_curr_idx >= 0);
2001 	ASSERT(vdc->dring_curr_idx < vdc->dring_len);
2002 	ASSERT(mutex_owned(&vdc->dring_lock));
2003 
2004 	/* pick the next descriptor after the last one used */
2005 	idx = (vdc->dring_curr_idx + 1) % vdc->dring_len;
2006 	ldep = &vdc->local_dring[idx];
2007 	ASSERT(ldep != NULL);
2008 	dep = ldep->dep;
2009 	ASSERT(dep != NULL);
2010 
2011 	mutex_enter(&ldep->lock);
2012 	if (dep->hdr.dstate == VIO_DESC_FREE) {
2013 		vdc->dring_curr_idx = idx;
2014 	} else {
2015 		DTRACE_PROBE(full);
2016 		(void) cv_timedwait(&ldep->cv, &ldep->lock,
2017 					VD_GET_TIMEOUT_HZ(vdc_hz_timeout, 1));
2018 		if (dep->hdr.dstate == VIO_DESC_FREE) {
2019 			vdc->dring_curr_idx = idx;
2020 		} else {
2021 			DMSG(0, "[%d] Entry %d unavailable still in state %d\n",
2022 					vdc->instance, idx, dep->hdr.dstate);
2023 			idx = -1; /* indicate that the ring is full */
2024 		}
2025 	}
2026 	mutex_exit(&ldep->lock);
2027 
2028 	return (idx);
2029 }
2030 
2031 /*
2032  * Function:
2033  *	vdc_populate_descriptor
2034  *
2035  * Description:
2036  *	This routine writes the data to be transmitted to vds into the
2037  *	descriptor, notifies vds that the ring has been updated and
2038  *	then waits for the request to be processed.
2039  *
2040  * Arguments:
2041  *	vdc	- the soft state pointer
2042  *	addr	- address of structure to be written. In the case of block
2043  *		  reads and writes this structure will be a buf_t and the
2044  *		  address of the data to be written will be in the b_un.b_addr
2045  *		  field. Otherwise the value of addr will be the address
2046  *		  to be written.
2047  *	nbytes	- number of bytes to read/write
2048  *	operation - operation we want vds to perform (VD_OP_XXX)
2049  *	arg	- parameter to be sent to server (depends on VD_OP_XXX type)
2050  *			. mode for ioctl(9e)
2051  *			. LP64 diskaddr_t (block I/O)
2052  *	slice	- the disk slice this request is for
2053  *
2054  * Return Codes:
2055  *	0
2056  *	EAGAIN
2057  *		EFAULT
2058  *		ENXIO
2059  *		EIO
2060  */
2061 static int
2062 vdc_populate_descriptor(vdc_t *vdc, caddr_t addr, size_t nbytes, int operation,
2063 				uint64_t arg, uint64_t slice)
2064 {
2065 	vdc_local_desc_t *local_dep = NULL;	/* Local Dring Entry Pointer */
2066 	vd_dring_entry_t *dep = NULL;		/* Dring Entry Pointer */
2067 	int			idx = 0;	/* Index of DRing entry used */
2068 	vio_dring_msg_t		dmsg;
2069 	size_t			msglen = sizeof (dmsg);
2070 	int			retries = 0;
2071 	int			rv;
2072 
2073 	ASSERT(vdc != NULL);
2074 	ASSERT(slice < V_NUMPAR);
2075 
2076 	/*
2077 	 * Get next available DRing entry.
2078 	 */
2079 	mutex_enter(&vdc->dring_lock);
2080 	idx = vdc_get_next_dring_entry_idx(vdc, 1);
2081 	if (idx == -1) {
2082 		mutex_exit(&vdc->dring_lock);
2083 		DMSG(0, "[%d] no descriptor ring entry avail, last seq=%ld\n",
2084 				vdc->instance, vdc->seq_num - 1);
2085 
2086 		/*
2087 		 * Since strategy should not block we don't wait for the DRing
2088 		 * to empty and instead return
2089 		 */
2090 		return (EAGAIN);
2091 	}
2092 
2093 	ASSERT(idx < vdc->dring_len);
2094 	local_dep = &vdc->local_dring[idx];
2095 	dep = local_dep->dep;
2096 	ASSERT(dep != NULL);
2097 
2098 	/*
2099 	 * We now get the lock for this descriptor before dropping the overall
2100 	 * DRing lock. This prevents a race condition where another vdc thread
2101 	 * could grab the descriptor we selected.
2102 	 */
2103 	ASSERT(MUTEX_NOT_HELD(&local_dep->lock));
2104 	mutex_enter(&local_dep->lock);
2105 	mutex_exit(&vdc->dring_lock);
2106 
2107 	switch (operation) {
2108 	case VD_OP_BREAD:
2109 	case VD_OP_BWRITE:
2110 		local_dep->buf = (struct buf *)addr;
2111 		local_dep->addr = local_dep->buf->b_un.b_addr;
2112 		DMSG(2, "[%d] buf=%p, block=%lx, nbytes=%lu\n",
2113 				vdc->instance, (void *)addr, arg, nbytes);
2114 		dep->payload.addr = (diskaddr_t)arg;
2115 		rv = vdc_populate_mem_hdl(vdc, idx, local_dep->addr,
2116 						nbytes, operation);
2117 		break;
2118 
2119 	case VD_OP_GET_VTOC:
2120 	case VD_OP_SET_VTOC:
2121 	case VD_OP_GET_DISKGEOM:
2122 	case VD_OP_SET_DISKGEOM:
2123 	case VD_OP_SCSICMD:
2124 		local_dep->addr = addr;
2125 		if (nbytes > 0) {
2126 			rv = vdc_populate_mem_hdl(vdc, idx, addr, nbytes,
2127 							operation);
2128 		}
2129 		break;
2130 
2131 	case VD_OP_FLUSH:
2132 	case VD_OP_GET_WCE:
2133 	case VD_OP_SET_WCE:
2134 		rv = 0;		/* nothing to bind */
2135 		break;
2136 
2137 	default:
2138 		cmn_err(CE_CONT, "?[%d] Unsupported vDisk operation [%d]\n",
2139 				vdc->instance, operation);
2140 		rv = EINVAL;
2141 	}
2142 
2143 	if (rv != 0) {
2144 		mutex_exit(&local_dep->lock);
2145 		return (rv);
2146 	}
2147 
2148 	/*
2149 	 * fill in the data details into the DRing
2150 	 */
2151 	dep->payload.req_id = VDC_GET_NEXT_REQ_ID(vdc);
2152 	dep->payload.operation = operation;
2153 	dep->payload.nbytes = nbytes;
2154 	dep->payload.status = -1;	/* vds will set valid value */
2155 	dep->payload.slice = slice;
2156 	dep->hdr.dstate = VIO_DESC_READY;
2157 	dep->hdr.ack = 1;		/* request an ACK for every message */
2158 
2159 	local_dep->flags = VIO_DESC_READY;
2160 
2161 	/*
2162 	 * Send a msg with the DRing details to vds
2163 	 */
2164 	mutex_enter(&vdc->lock);
2165 	VIO_INIT_DRING_DATA_TAG(dmsg);
2166 	VDC_INIT_DRING_DATA_MSG_IDS(dmsg, vdc);
2167 	dmsg.dring_ident = vdc->dring_ident;
2168 	dmsg.start_idx = idx;
2169 	dmsg.end_idx = idx;
2170 
2171 	DTRACE_IO2(send, vio_dring_msg_t *, &dmsg, vdc_t *, vdc);
2172 
2173 	DMSG(2, "[%d] ident=0x%lx, st=%u, end=%u, seq=%ld req=%ld dep=%p\n",
2174 			vdc->instance, vdc->dring_ident,
2175 			dmsg.start_idx, dmsg.end_idx,
2176 			dmsg.seq_num, dep->payload.req_id, (void *)dep);
2177 
2178 	rv = vdc_send(vdc, (caddr_t)&dmsg, &msglen);
2179 	DMSG(1, "[%d] send via LDC: rv=%d\n", vdc->instance, rv);
2180 	if (rv != 0) {
2181 		cmn_err(CE_NOTE, "[%d] err (%d) sending DRing data msg via LDC",
2182 				vdc->instance, rv);
2183 
2184 		/* Clear the DRing entry */
2185 		rv = vdc_depopulate_descriptor(vdc, idx);
2186 
2187 		mutex_exit(&vdc->lock);
2188 		mutex_exit(&local_dep->lock);
2189 
2190 		return (rv ? rv : EAGAIN);
2191 	}
2192 
2193 	/*
2194 	 * If the message was successfully sent, we increment the sequence
2195 	 * number to be used by the next message
2196 	 */
2197 	vdc->seq_num++;
2198 	mutex_exit(&vdc->lock);
2199 
2200 	/*
2201 	 * When a guest is panicking, the completion of requests needs to be
2202 	 * handled differently because interrupts are disabled and vdc
2203 	 * will not get messages. We have to poll for the messages instead.
2204 	 */
2205 	if (ddi_in_panic()) {
2206 		int start = 0;
2207 		retries = 0;
2208 		for (;;) {
2209 			msglen = sizeof (dmsg);
2210 			rv = ldc_read(vdc->ldc_handle, (caddr_t)&dmsg,
2211 					&msglen);
2212 			if (rv) {
2213 				rv = EINVAL;
2214 				break;
2215 			}
2216 
2217 			/*
2218 			 * if there are no packets wait and check again
2219 			 */
2220 			if ((rv == 0) && (msglen == 0)) {
2221 				if (retries++ > vdc_dump_retries) {
2222 					DMSG(0, "[%d] Stopping wait, idx %d\n",
2223 							vdc->instance, idx);
2224 					rv = EAGAIN;
2225 					break;
2226 				}
2227 
2228 				DMSG(1, "Waiting for next packet @ %d\n", idx);
2229 				drv_usecwait(vdc_usec_timeout_dump);
2230 				continue;
2231 			}
2232 
2233 			/*
2234 			 * Ignore all messages that are not ACKs/NACKs to
2235 			 * DRing requests.
2236 			 */
2237 			if ((dmsg.tag.vio_msgtype != VIO_TYPE_DATA) ||
2238 			    (dmsg.tag.vio_subtype_env != VIO_DRING_DATA)) {
2239 				DMSG(0, "discard pkt: type=%d sub=%d env=%d\n",
2240 					dmsg.tag.vio_msgtype,
2241 					dmsg.tag.vio_subtype,
2242 					dmsg.tag.vio_subtype_env);
2243 				continue;
2244 			}
2245 
2246 			/*
2247 			 * set the appropriate return value for the
2248 			 * current request.
2249 			 */
2250 			switch (dmsg.tag.vio_subtype) {
2251 			case VIO_SUBTYPE_ACK:
2252 				rv = 0;
2253 				break;
2254 			case VIO_SUBTYPE_NACK:
2255 				rv = EAGAIN;
2256 				break;
2257 			default:
2258 				continue;
2259 			}
2260 
2261 			start = dmsg.start_idx;
2262 			if (start >= vdc->dring_len) {
2263 				DMSG(0, "[%d] Bogus ack data : start %d\n",
2264 					vdc->instance, start);
2265 				continue;
2266 			}
2267 
2268 			dep = VDC_GET_DRING_ENTRY_PTR(vdc, start);
2269 
2270 			DMSG(1, "[%d] Dumping start=%d idx=%d state=%d\n",
2271 				vdc->instance, start, idx, dep->hdr.dstate);
2272 
2273 			if (dep->hdr.dstate != VIO_DESC_DONE) {
2274 				DMSG(0, "[%d] Entry @ %d - state !DONE %d\n",
2275 					vdc->instance, start, dep->hdr.dstate);
2276 				continue;
2277 			}
2278 
2279 			(void) vdc_depopulate_descriptor(vdc, start);
2280 
2281 			/*
2282 			 * We want to process all Dring entries up to
2283 			 * the current one so that we can return an
2284 			 * error with the correct request.
2285 			 */
2286 			if (idx > start) {
2287 				DMSG(0, "[%d] Looping: start %d, idx %d\n",
2288 						vdc->instance, idx, start);
2289 				continue;
2290 			}
2291 
2292 			/* exit - all outstanding requests are completed */
2293 			break;
2294 		}
2295 
2296 		mutex_exit(&local_dep->lock);
2297 
2298 		return (rv);
2299 	}
2300 
2301 	/*
2302 	 * In the case of calls from strategy and dump (in the non-panic case),
2303 	 * instead of waiting for a response from the vDisk server return now.
2304 	 * They will be processed asynchronously and the vdc ACK handling code
2305 	 * will trigger the biodone(9F)
2306 	 */
2307 	if ((operation == VD_OP_BREAD) || (operation == VD_OP_BWRITE)) {
2308 		mutex_exit(&local_dep->lock);
2309 		return (rv);
2310 	}
2311 
2312 	/*
2313 	 * In the case of synchronous calls we watch the DRing entries we
2314 	 * modified and await the response from vds.
2315 	 */
2316 	rv = vdc_wait_for_descriptor_update(vdc, idx, dmsg);
2317 	if (rv == ETIMEDOUT) {
2318 		/* debug info when dumping state on vds side */
2319 		dep->payload.status = ECANCELED;
2320 	}
2321 
2322 	rv = vdc_depopulate_descriptor(vdc, idx);
2323 	DMSG(0, "[%d] Exiting: status=%d\n", vdc->instance, rv);
2324 
2325 	mutex_exit(&local_dep->lock);
2326 
2327 	return (rv);
2328 }
2329 
2330 /*
2331  * Function:
2332  *	vdc_wait_for_descriptor_update()
2333  *
2334  * Description:
2335  *
2336  * Arguments:
2337  *	vdc	- soft state pointer for this instance of the device driver.
2338  *	idx	- Index of the Descriptor Ring entry being modified
2339  *	dmsg	- LDC message sent by vDisk server
2340  *
2341  * Return Code:
2342  *	0	- Success
2343  */
2344 static int
2345 vdc_wait_for_descriptor_update(vdc_t *vdc, uint_t idx, vio_dring_msg_t dmsg)
2346 {
2347 	vd_dring_entry_t *dep = NULL;		/* Dring Entry Pointer */
2348 	vdc_local_desc_t *local_dep = NULL;	/* Local Dring Entry Pointer */
2349 	size_t	msglen = sizeof (dmsg);
2350 	int	retries = 0;
2351 	int	status = 0;
2352 	int	rv = 0;
2353 
2354 	ASSERT(vdc != NULL);
2355 	ASSERT(idx < vdc->dring_len);
2356 	local_dep = &vdc->local_dring[idx];
2357 	ASSERT(local_dep != NULL);
2358 	ASSERT(MUTEX_HELD(&local_dep->lock));
2359 	dep = local_dep->dep;
2360 	ASSERT(dep != NULL);
2361 
2362 	while (dep->hdr.dstate != VIO_DESC_DONE) {
2363 		rv = cv_timedwait(&local_dep->cv, &local_dep->lock,
2364 			VD_GET_TIMEOUT_HZ(vdc_hz_timeout, retries));
2365 		if (rv == -1) {
2366 			/*
2367 			 * If they persist in ignoring us we'll storm off in a
2368 			 * huff and return ETIMEDOUT to the upper layers.
2369 			 */
2370 			if (retries >= vdc_retries) {
2371 				DMSG(0, "[%d] Finished waiting on entry %d\n",
2372 					vdc->instance, idx);
2373 				status = ETIMEDOUT;
2374 				break;
2375 			} else {
2376 				retries++;
2377 				DMSG(0, "[%d] Timeout #%d on entry %d "
2378 				    "[seq %lu][req %lu]\n", vdc->instance,
2379 				    retries, idx, dmsg.seq_num,
2380 				    dep->payload.req_id);
2381 			}
2382 
2383 			if (dep->hdr.dstate & VIO_DESC_ACCEPTED) {
2384 				DMSG(0, "[%d] entry %d ACCEPTED [seq %lu]"
2385 				    "[req %lu] but not ACK'ed by vds yet\n",
2386 				    vdc->instance, idx, dmsg.seq_num,
2387 				    dep->payload.req_id);
2388 				continue;
2389 			}
2390 
2391 			/*
2392 			 * we resend the message as it may have been dropped
2393 			 * and have never made it to the other side (vds).
2394 			 * (We reuse the original message but update seq ID)
2395 			 */
2396 			mutex_enter(&vdc->lock);
2397 			VDC_INIT_DRING_DATA_MSG_IDS(dmsg, vdc);
2398 			retries = 0;
2399 			status = vdc_send(vdc, (caddr_t)&dmsg, &msglen);
2400 			if (status != 0) {
2401 				mutex_exit(&vdc->lock);
2402 				cmn_err(CE_NOTE, "[%d] Error (%d) while sending"
2403 						" after timeout",
2404 						vdc->instance, status);
2405 				status = ETIMEDOUT;
2406 				break;
2407 			}
2408 			/*
2409 			 * If the message was successfully sent, we increment
2410 			 * the sequence number to be used by the next message.
2411 			 */
2412 			vdc->seq_num++;
2413 			mutex_exit(&vdc->lock);
2414 		}
2415 	}
2416 
2417 	return (status);
2418 }
2419 
2420 
2421 /*
2422  * Function:
2423  *	vdc_depopulate_descriptor()
2424  *
2425  * Description:
2426  *
2427  * Arguments:
2428  *	vdc	- soft state pointer for this instance of the device driver.
2429  *	idx	- Index of the Descriptor Ring entry being modified
2430  *
2431  * Return Code:
2432  *	0	- Success
2433  */
2434 static int
2435 vdc_depopulate_descriptor(vdc_t *vdc, uint_t idx)
2436 {
2437 	vd_dring_entry_t *dep = NULL;		/* Dring Entry Pointer */
2438 	vdc_local_desc_t *ldep = NULL;		/* Local Dring Entry Pointer */
2439 	int		status = ENXIO;
2440 	int		operation;
2441 	int		rv = 0;
2442 
2443 	ASSERT(vdc != NULL);
2444 	ASSERT(idx < vdc->dring_len);
2445 	ldep = &vdc->local_dring[idx];
2446 	ASSERT(ldep != NULL);
2447 	ASSERT(MUTEX_HELD(&ldep->lock));
2448 	dep = ldep->dep;
2449 	ASSERT(dep != NULL);
2450 	ASSERT((dep->hdr.dstate == VIO_DESC_DONE) ||
2451 			(dep->payload.status == ECANCELED));
2452 
2453 	VDC_MARK_DRING_ENTRY_FREE(vdc, idx);
2454 	VIO_SET_DESC_STATE(ldep->flags, VIO_DESC_FREE);
2455 	status = dep->payload.status;
2456 	operation = dep->payload.operation;
2457 
2458 	/* the DKIO W$ operations never bind handles so we can return now */
2459 	if ((operation == VD_OP_FLUSH) ||
2460 	    (operation == VD_OP_GET_WCE) ||
2461 	    (operation == VD_OP_SET_WCE))
2462 		return (status);
2463 
2464 	/*
2465 	 * If the upper layer passed in a misaligned address we copied the
2466 	 * data into an aligned buffer before sending it to LDC - we now
2467 	 * copy it back to the original buffer.
2468 	 */
2469 	if (ldep->align_addr) {
2470 		ASSERT(ldep->addr != NULL);
2471 		ASSERT(dep->payload.nbytes > 0);
2472 
2473 		bcopy(ldep->align_addr, ldep->addr, dep->payload.nbytes);
2474 		kmem_free(ldep->align_addr,
2475 			sizeof (caddr_t) * P2ROUNDUP(dep->payload.nbytes, 8));
2476 		ldep->align_addr = NULL;
2477 	}
2478 
2479 	rv = ldc_mem_unbind_handle(ldep->desc_mhdl);
2480 	if (rv != 0) {
2481 		cmn_err(CE_CONT, "?[%d] unbind mhdl 0x%lx @ idx %d failed (%d)",
2482 				vdc->instance, ldep->desc_mhdl, idx, rv);
2483 		/*
2484 		 * The error returned by the vDisk server is more informative
2485 		 * and thus has a higher priority but if it isn't set we ensure
2486 		 * that this function returns an error.
2487 		 */
2488 		if (status == 0)
2489 			status = EINVAL;
2490 	}
2491 
2492 	return (status);
2493 }
2494 
2495 /*
2496  * Function:
2497  *	vdc_populate_mem_hdl()
2498  *
2499  * Description:
2500  *
2501  * Arguments:
2502  *	vdc	- soft state pointer for this instance of the device driver.
2503  *	idx	- Index of the Descriptor Ring entry being modified
2504  *	addr	- virtual address being mapped in
2505  *	nybtes	- number of bytes in 'addr'
2506  *	operation - the vDisk operation being performed (VD_OP_xxx)
2507  *
2508  * Return Code:
2509  *	0	- Success
2510  */
2511 static int
2512 vdc_populate_mem_hdl(vdc_t *vdc, uint_t idx, caddr_t addr, size_t nbytes,
2513 			int operation)
2514 {
2515 	vd_dring_entry_t	*dep = NULL;
2516 	vdc_local_desc_t	*ldep = NULL;
2517 	ldc_mem_handle_t	mhdl;
2518 	caddr_t			vaddr;
2519 	int			perm = LDC_MEM_RW;
2520 	int			rv = 0;
2521 	int			i;
2522 
2523 	ASSERT(vdc != NULL);
2524 	ASSERT(idx < vdc->dring_len);
2525 
2526 	dep = VDC_GET_DRING_ENTRY_PTR(vdc, idx);
2527 	ldep = &vdc->local_dring[idx];
2528 	mhdl = ldep->desc_mhdl;
2529 
2530 	switch (operation) {
2531 	case VD_OP_BREAD:
2532 		perm = LDC_MEM_W;
2533 		break;
2534 
2535 	case VD_OP_BWRITE:
2536 		perm = LDC_MEM_R;
2537 		break;
2538 
2539 	case VD_OP_GET_VTOC:
2540 	case VD_OP_SET_VTOC:
2541 	case VD_OP_GET_DISKGEOM:
2542 	case VD_OP_SET_DISKGEOM:
2543 	case VD_OP_SCSICMD:
2544 		perm = LDC_MEM_RW;
2545 		break;
2546 
2547 	default:
2548 		ASSERT(0);	/* catch bad programming in vdc */
2549 	}
2550 
2551 	/*
2552 	 * LDC expects any addresses passed in to be 8-byte aligned. We need
2553 	 * to copy the contents of any misaligned buffers to a newly allocated
2554 	 * buffer and bind it instead (and copy the the contents back to the
2555 	 * original buffer passed in when depopulating the descriptor)
2556 	 */
2557 	vaddr = addr;
2558 	if (((uint64_t)addr & 0x7) != 0) {
2559 		ASSERT(ldep->align_addr == NULL);
2560 		ldep->align_addr =
2561 			kmem_zalloc(sizeof (caddr_t) * P2ROUNDUP(nbytes, 8),
2562 					KM_SLEEP);
2563 		DMSG(0, "[%d] Misaligned address %p reallocating "
2564 		    "(buf=%p nb=%ld op=%d entry=%d)\n",
2565 		    vdc->instance, (void *)addr, (void *)ldep->align_addr,
2566 		    nbytes, operation, idx);
2567 		bcopy(addr, ldep->align_addr, nbytes);
2568 		vaddr = ldep->align_addr;
2569 	}
2570 
2571 	rv = ldc_mem_bind_handle(mhdl, vaddr, P2ROUNDUP(nbytes, 8),
2572 		LDC_SHADOW_MAP, perm, &dep->payload.cookie[0],
2573 		&dep->payload.ncookies);
2574 	DMSG(2, "[%d] bound mem handle; ncookies=%d\n",
2575 			vdc->instance, dep->payload.ncookies);
2576 	if (rv != 0) {
2577 		cmn_err(CE_CONT, "?[%d] Failed to bind LDC memory handle "
2578 		    "(mhdl=%p, buf=%p entry=%u err=%d)\n",
2579 		    vdc->instance, (void *)mhdl, (void *)addr, idx, rv);
2580 		if (ldep->align_addr) {
2581 			kmem_free(ldep->align_addr,
2582 				sizeof (caddr_t) * P2ROUNDUP(nbytes, 8));
2583 			ldep->align_addr = NULL;
2584 		}
2585 		return (EAGAIN);
2586 	}
2587 
2588 	/*
2589 	 * Get the other cookies (if any).
2590 	 */
2591 	for (i = 1; i < dep->payload.ncookies; i++) {
2592 		rv = ldc_mem_nextcookie(mhdl, &dep->payload.cookie[i]);
2593 		if (rv != 0) {
2594 			(void) ldc_mem_unbind_handle(mhdl);
2595 			cmn_err(CE_CONT, "?[%d] Failed to get next cookie "
2596 					"(mhdl=%lx cnum=%d), err=%d",
2597 					vdc->instance, mhdl, i, rv);
2598 			if (ldep->align_addr) {
2599 				kmem_free(ldep->align_addr,
2600 					sizeof (caddr_t) * dep->payload.nbytes);
2601 				ldep->align_addr = NULL;
2602 			}
2603 			return (EAGAIN);
2604 		}
2605 	}
2606 
2607 	return (rv);
2608 }
2609 
2610 /*
2611  * Interrupt handlers for messages from LDC
2612  */
2613 
2614 /*
2615  * Function:
2616  *	vdc_handle_cb()
2617  *
2618  * Description:
2619  *
2620  * Arguments:
2621  *	event	- Type of event (LDC_EVT_xxx) that triggered the callback
2622  *	arg	- soft state pointer for this instance of the device driver.
2623  *
2624  * Return Code:
2625  *	0	- Success
2626  */
2627 static uint_t
2628 vdc_handle_cb(uint64_t event, caddr_t arg)
2629 {
2630 	ldc_status_t	ldc_state;
2631 	int		rv = 0;
2632 
2633 	vdc_t	*vdc = (vdc_t *)(void *)arg;
2634 
2635 	ASSERT(vdc != NULL);
2636 
2637 	DMSG(1, "[%d] evt=%lx seqID=%ld\n", vdc->instance, event, vdc->seq_num);
2638 
2639 	/*
2640 	 * Depending on the type of event that triggered this callback,
2641 	 * we modify the handhske state or read the data.
2642 	 *
2643 	 * NOTE: not done as a switch() as event could be triggered by
2644 	 * a state change and a read request. Also the ordering	of the
2645 	 * check for the event types is deliberate.
2646 	 */
2647 	if (event & LDC_EVT_UP) {
2648 		DMSG(0, "[%d] Received LDC_EVT_UP\n", vdc->instance);
2649 
2650 		/* get LDC state */
2651 		rv = ldc_status(vdc->ldc_handle, &ldc_state);
2652 		if (rv != 0) {
2653 			cmn_err(CE_NOTE, "[%d] Couldn't get LDC status %d",
2654 					vdc->instance, rv);
2655 			mutex_enter(&vdc->lock);
2656 			vdc_reset_connection(vdc, B_TRUE);
2657 			mutex_exit(&vdc->lock);
2658 			return (LDC_SUCCESS);
2659 		}
2660 
2661 		/*
2662 		 * Reset the transaction sequence numbers when LDC comes up.
2663 		 * We then kick off the handshake negotiation with the vDisk
2664 		 * server.
2665 		 */
2666 		mutex_enter(&vdc->lock);
2667 		vdc->seq_num = 1;
2668 		vdc->seq_num_reply = 0;
2669 		vdc->ldc_state = ldc_state;
2670 		ASSERT(ldc_state == LDC_UP);
2671 		mutex_exit(&vdc->lock);
2672 
2673 		vdc_init_handshake_negotiation(vdc);
2674 
2675 		ASSERT((event & (LDC_EVT_RESET | LDC_EVT_DOWN)) == 0);
2676 	}
2677 
2678 	if (event & LDC_EVT_READ) {
2679 		/*
2680 		 * Wake up the worker thread to process the message
2681 		 */
2682 		mutex_enter(&vdc->msg_proc_lock);
2683 		vdc->msg_pending = B_TRUE;
2684 		cv_signal(&vdc->msg_proc_cv);
2685 		mutex_exit(&vdc->msg_proc_lock);
2686 
2687 		ASSERT((event & (LDC_EVT_RESET | LDC_EVT_DOWN)) == 0);
2688 
2689 		/* that's all we have to do - no need to handle DOWN/RESET */
2690 		return (LDC_SUCCESS);
2691 	}
2692 
2693 	if (event & LDC_EVT_RESET) {
2694 		DMSG(0, "[%d] Received LDC RESET event\n", vdc->instance);
2695 
2696 		/* get LDC state */
2697 		rv = ldc_status(vdc->ldc_handle, &ldc_state);
2698 		if (rv != 0) {
2699 			cmn_err(CE_NOTE, "[%d] Couldn't get LDC status %d",
2700 					vdc->instance, rv);
2701 			ldc_state = LDC_OPEN;
2702 		}
2703 		mutex_enter(&vdc->lock);
2704 		vdc->ldc_state = ldc_state;
2705 		vdc_reset_connection(vdc, B_TRUE);
2706 		mutex_exit(&vdc->lock);
2707 
2708 		vdc_init_handshake_negotiation(vdc);
2709 	}
2710 
2711 	if (event & LDC_EVT_DOWN) {
2712 		DMSG(0, "[%d] Received LDC DOWN event\n", vdc->instance);
2713 
2714 		/* get LDC state */
2715 		rv = ldc_status(vdc->ldc_handle, &ldc_state);
2716 		if (rv != 0) {
2717 			cmn_err(CE_NOTE, "[%d] Couldn't get LDC status %d",
2718 					vdc->instance, rv);
2719 			ldc_state = LDC_OPEN;
2720 		}
2721 		mutex_enter(&vdc->lock);
2722 		vdc->ldc_state = ldc_state;
2723 		vdc_reset_connection(vdc, B_TRUE);
2724 		mutex_exit(&vdc->lock);
2725 	}
2726 
2727 	if (event & ~(LDC_EVT_UP | LDC_EVT_RESET | LDC_EVT_DOWN | LDC_EVT_READ))
2728 		cmn_err(CE_NOTE, "![%d] Unexpected LDC event (%lx) received",
2729 				vdc->instance, event);
2730 
2731 	return (LDC_SUCCESS);
2732 }
2733 
2734 /* -------------------------------------------------------------------------- */
2735 
2736 /*
2737  * The following functions process the incoming messages from vds
2738  */
2739 
2740 
2741 /*
2742  * Function:
2743  *	vdc_process_msg_thread()
2744  *
2745  * Description:
2746  *
2747  * Arguments:
2748  *	vdc	- soft state pointer for this instance of the device driver.
2749  *
2750  * Return Code:
2751  *	None
2752  */
2753 static void
2754 vdc_process_msg_thread(vdc_t *vdc)
2755 {
2756 	int		status = 0;
2757 	boolean_t	q_has_pkts = B_FALSE;
2758 
2759 	ASSERT(vdc != NULL);
2760 
2761 	mutex_enter(&vdc->msg_proc_lock);
2762 	DMSG(0, "[%d] Starting\n", vdc->instance);
2763 
2764 	vdc->msg_proc_thr_state = VDC_THR_RUNNING;
2765 
2766 	while (vdc->msg_proc_thr_state == VDC_THR_RUNNING) {
2767 
2768 		DMSG(2, "[%d] Waiting\n", vdc->instance);
2769 		while (!vdc->msg_pending)
2770 			cv_wait(&vdc->msg_proc_cv, &vdc->msg_proc_lock);
2771 
2772 		DMSG(2, "[%d] Message Received\n", vdc->instance);
2773 
2774 		/* check if there is data */
2775 		status = ldc_chkq(vdc->ldc_handle, &q_has_pkts);
2776 		if ((status != 0) &&
2777 		    (vdc->msg_proc_thr_state == VDC_THR_RUNNING)) {
2778 			cmn_err(CE_NOTE, "[%d] Unable to communicate with vDisk"
2779 					" server. Cannot check LDC queue: %d",
2780 					vdc->instance, status);
2781 			mutex_enter(&vdc->lock);
2782 			vdc_reset_connection(vdc, B_TRUE);
2783 			mutex_exit(&vdc->lock);
2784 			vdc->msg_proc_thr_state = VDC_THR_STOP;
2785 			continue;
2786 		}
2787 
2788 		if (q_has_pkts) {
2789 			DMSG(2, "[%d] new pkt(s) available\n", vdc->instance);
2790 			vdc_process_msg(vdc);
2791 		}
2792 
2793 		vdc->msg_pending = B_FALSE;
2794 	}
2795 
2796 	DMSG(0, "[%d] Message processing thread stopped\n", vdc->instance);
2797 	vdc->msg_pending = B_FALSE;
2798 	vdc->msg_proc_thr_state = VDC_THR_DONE;
2799 	cv_signal(&vdc->msg_proc_cv);
2800 	mutex_exit(&vdc->msg_proc_lock);
2801 	thread_exit();
2802 }
2803 
2804 
2805 /*
2806  * Function:
2807  *	vdc_process_msg()
2808  *
2809  * Description:
2810  *	This function is called by the message processing thread each time it
2811  *	is triggered when LDC sends an interrupt to indicate that there are
2812  *	more packets on the queue. When it is called it will continue to loop
2813  *	and read the messages until there are no more left of the queue. If it
2814  *	encounters an invalid sized message it will drop it and check the next
2815  *	message.
2816  *
2817  * Arguments:
2818  *	arg	- soft state pointer for this instance of the device driver.
2819  *
2820  * Return Code:
2821  *	None.
2822  */
2823 static void
2824 vdc_process_msg(void *arg)
2825 {
2826 	vdc_t		*vdc = (vdc_t *)(void *)arg;
2827 	vio_msg_t	vio_msg;
2828 	size_t		nbytes = sizeof (vio_msg);
2829 	int		status;
2830 
2831 	ASSERT(vdc != NULL);
2832 
2833 	mutex_enter(&vdc->lock);
2834 
2835 	DMSG(1, "[%d]\n", vdc->instance);
2836 
2837 	for (;;) {
2838 
2839 		/* read all messages - until no more left */
2840 		status = ldc_read(vdc->ldc_handle, (caddr_t)&vio_msg, &nbytes);
2841 
2842 		if (status) {
2843 			cmn_err(CE_CONT, "?[%d] Error %d reading LDC msg\n",
2844 					vdc->instance, status);
2845 
2846 			/* if status is ECONNRESET --- reset vdc state */
2847 			if (status == EIO || status == ECONNRESET) {
2848 				vdc_reset_connection(vdc, B_TRUE);
2849 			}
2850 
2851 			mutex_exit(&vdc->lock);
2852 			return;
2853 		}
2854 
2855 		if ((nbytes > 0) && (nbytes < sizeof (vio_msg_tag_t))) {
2856 			cmn_err(CE_CONT, "?[%d] Expect %lu bytes; recv'd %lu\n",
2857 				vdc->instance, sizeof (vio_msg_tag_t), nbytes);
2858 			mutex_exit(&vdc->lock);
2859 			return;
2860 		}
2861 
2862 		if (nbytes == 0) {
2863 			DMSG(3, "[%d] ldc_read() done..\n", vdc->instance);
2864 			mutex_exit(&vdc->lock);
2865 			return;
2866 		}
2867 
2868 		DMSG(2, "[%d] (%x/%x/%x)\n", vdc->instance,
2869 		    vio_msg.tag.vio_msgtype,
2870 		    vio_msg.tag.vio_subtype,
2871 		    vio_msg.tag.vio_subtype_env);
2872 
2873 		/*
2874 		 * Verify the Session ID of the message
2875 		 *
2876 		 * Every message after the Version has been negotiated should
2877 		 * have the correct session ID set.
2878 		 */
2879 		if ((vio_msg.tag.vio_sid != vdc->session_id) &&
2880 		    (vio_msg.tag.vio_subtype_env != VIO_VER_INFO)) {
2881 			cmn_err(CE_NOTE, "[%d] Invalid SID: received 0x%x, "
2882 					"expected 0x%lx [seq num %lx @ %d]",
2883 				vdc->instance, vio_msg.tag.vio_sid,
2884 				vdc->session_id,
2885 				((vio_dring_msg_t *)&vio_msg)->seq_num,
2886 				((vio_dring_msg_t *)&vio_msg)->start_idx);
2887 			vdc_reset_connection(vdc, B_TRUE);
2888 			mutex_exit(&vdc->lock);
2889 			return;
2890 		}
2891 
2892 		switch (vio_msg.tag.vio_msgtype) {
2893 		case VIO_TYPE_CTRL:
2894 			status = vdc_process_ctrl_msg(vdc, vio_msg);
2895 			break;
2896 		case VIO_TYPE_DATA:
2897 			status = vdc_process_data_msg(vdc, vio_msg);
2898 			break;
2899 		case VIO_TYPE_ERR:
2900 			status = vdc_process_err_msg(vdc, vio_msg);
2901 			break;
2902 		default:
2903 			cmn_err(CE_NOTE, "[%d] Unknown VIO message type",
2904 					vdc->instance);
2905 			status = EINVAL;
2906 			break;
2907 		}
2908 
2909 		if (status != 0) {
2910 			DMSG(0, "[%d] Error (%d) occurred processing req %lu\n",
2911 					vdc->instance, status,
2912 					vdc->req_id_proc);
2913 			vdc_reset_connection(vdc, B_TRUE);
2914 
2915 			/* we need to drop the lock to trigger the handshake */
2916 			mutex_exit(&vdc->lock);
2917 			vdc_init_handshake_negotiation(vdc);
2918 			mutex_enter(&vdc->lock);
2919 		}
2920 	}
2921 	_NOTE(NOTREACHED)
2922 }
2923 
2924 /*
2925  * Function:
2926  *	vdc_process_ctrl_msg()
2927  *
2928  * Description:
2929  *	This function is called by the message processing thread each time
2930  *	an LDC message with a msgtype of VIO_TYPE_CTRL is received.
2931  *
2932  * Arguments:
2933  *	vdc	- soft state pointer for this instance of the device driver.
2934  *	msg	- the LDC message sent by vds
2935  *
2936  * Return Codes:
2937  *	0	- Success.
2938  *	EPROTO	- A message was received which shouldn't have happened according
2939  *		  to the protocol
2940  *	ENOTSUP	- An action which is allowed according to the protocol but which
2941  *		  isn't (or doesn't need to be) implemented yet.
2942  *	EINVAL	- An invalid value was returned as part of a message.
2943  */
2944 static int
2945 vdc_process_ctrl_msg(vdc_t *vdc, vio_msg_t msg)
2946 {
2947 	int			status = -1;
2948 
2949 	ASSERT(msg.tag.vio_msgtype == VIO_TYPE_CTRL);
2950 	ASSERT(vdc != NULL);
2951 	ASSERT(mutex_owned(&vdc->lock));
2952 
2953 	/* Depending on which state we are in; process the message */
2954 	switch (vdc->state) {
2955 	case VD_STATE_INIT:
2956 		status = vdc_handle_ver_msg(vdc, (vio_ver_msg_t *)&msg);
2957 		break;
2958 
2959 	case VD_STATE_VER:
2960 		status = vdc_handle_attr_msg(vdc, (vd_attr_msg_t *)&msg);
2961 		break;
2962 
2963 	case VD_STATE_ATTR:
2964 		status = vdc_handle_dring_reg_msg(vdc,
2965 				(vio_dring_reg_msg_t *)&msg);
2966 		break;
2967 
2968 	case VD_STATE_RDX:
2969 		if (msg.tag.vio_subtype_env != VIO_RDX) {
2970 			status = EPROTO;
2971 			break;
2972 		}
2973 
2974 		DMSG(0, "[%d] Received RDX: handshake done\n", vdc->instance);
2975 
2976 		vdc->hshake_cnt = 0;	/* reset failed handshake count */
2977 		status = 0;
2978 		vdc->state = VD_STATE_DATA;
2979 
2980 		cv_broadcast(&vdc->attach_cv);
2981 		break;
2982 
2983 	case VD_STATE_DATA:
2984 	default:
2985 		cmn_err(CE_NOTE, "[%d] Unexpected handshake state %d",
2986 				vdc->instance, vdc->state);
2987 		status = EPROTO;
2988 		break;
2989 	}
2990 
2991 	return (status);
2992 }
2993 
2994 
2995 /*
2996  * Function:
2997  *	vdc_process_data_msg()
2998  *
2999  * Description:
3000  *	This function is called by the message processing thread each time
3001  *	a message with a msgtype of VIO_TYPE_DATA is received. It will either
3002  *	be an ACK or NACK from vds[1] which vdc handles as follows.
3003  *		ACK	- wake up the waiting thread
3004  *		NACK	- resend any messages necessary
3005  *
3006  *	[1] Although the message format allows it, vds should not send a
3007  *	    VIO_SUBTYPE_INFO message to vdc asking it to read data; if for
3008  *	    some bizarre reason it does, vdc will reset the connection.
3009  *
3010  * Arguments:
3011  *	vdc	- soft state pointer for this instance of the device driver.
3012  *	msg	- the LDC message sent by vds
3013  *
3014  * Return Code:
3015  *	0	- Success.
3016  *	> 0	- error value returned by LDC
3017  */
3018 static int
3019 vdc_process_data_msg(vdc_t *vdc, vio_msg_t msg)
3020 {
3021 	int			status = 0;
3022 	vdc_local_desc_t	*ldep = NULL;
3023 	vio_dring_msg_t		*dring_msg = NULL;
3024 	uint_t			start;
3025 	int			end;
3026 	uint_t			count = 0;
3027 	uint_t			operation;
3028 	uint_t			idx;
3029 
3030 	ASSERT(msg.tag.vio_msgtype == VIO_TYPE_DATA);
3031 	ASSERT(vdc != NULL);
3032 	ASSERT(mutex_owned(&vdc->lock));
3033 
3034 	dring_msg = (vio_dring_msg_t *)&msg;
3035 
3036 	/*
3037 	 * Check to see if the message has bogus data
3038 	 */
3039 	idx = start = dring_msg->start_idx;
3040 	end = dring_msg->end_idx;
3041 	if ((start >= vdc->dring_len) ||
3042 	    (end >= vdc->dring_len) || (end < -1)) {
3043 		cmn_err(CE_CONT, "?[%d] Bogus ACK data : start %d, end %d\n",
3044 			vdc->instance, start, end);
3045 		return (EINVAL);
3046 	}
3047 
3048 	DTRACE_IO2(recv, vio_dring_msg_t, dring_msg, vdc_t *, vdc);
3049 
3050 	/*
3051 	 * Verify that the sequence number is what vdc expects.
3052 	 */
3053 	switch (vdc_verify_seq_num(vdc, dring_msg)) {
3054 	case VDC_SEQ_NUM_TODO:
3055 		break;	/* keep processing this message */
3056 	case VDC_SEQ_NUM_SKIP:
3057 		return (0);
3058 	case VDC_SEQ_NUM_INVALID:
3059 		return (ENXIO);
3060 	}
3061 
3062 	if (msg.tag.vio_subtype == VIO_SUBTYPE_NACK) {
3063 		DMSG(0, "[%d] DATA NACK\n", vdc->instance);
3064 		VDC_DUMP_DRING_MSG(dring_msg);
3065 		return (EIO);
3066 
3067 	} else if (msg.tag.vio_subtype == VIO_SUBTYPE_INFO) {
3068 		return (EPROTO);
3069 	}
3070 
3071 	ldep = &vdc->local_dring[start];
3072 	if (ldep->dep->hdr.dstate == VIO_DESC_DONE) {
3073 		mutex_enter(&ldep->lock);
3074 		operation = ldep->dep->payload.operation;
3075 		vdc->req_id_proc = ldep->dep->payload.req_id;
3076 		vdc->dring_proc_idx = idx;
3077 		ASSERT(ldep->dep->hdr.dstate == VIO_DESC_DONE);
3078 
3079 		if ((operation == VD_OP_BREAD) || (operation == VD_OP_BWRITE)) {
3080 			bioerror(ldep->buf, ldep->dep->payload.status);
3081 			biodone(ldep->buf);
3082 
3083 			DTRACE_IO2(vdone, buf_t *, ldep->buf, vdc_t *, vdc);
3084 
3085 			/* Clear the DRing entry */
3086 			status = vdc_depopulate_descriptor(vdc, idx);
3087 		}
3088 		cv_signal(&ldep->cv);
3089 		mutex_exit(&ldep->lock);
3090 	}
3091 
3092 	/* probe gives the count of how many entries were processed */
3093 	DTRACE_IO2(processed, int, count, vdc_t *, vdc);
3094 
3095 	return (status);
3096 }
3097 
3098 /*
3099  * Function:
3100  *	vdc_process_err_msg()
3101  *
3102  * NOTE: No error messages are used as part of the vDisk protocol
3103  */
3104 static int
3105 vdc_process_err_msg(vdc_t *vdc, vio_msg_t msg)
3106 {
3107 	_NOTE(ARGUNUSED(vdc))
3108 	_NOTE(ARGUNUSED(msg))
3109 
3110 	ASSERT(msg.tag.vio_msgtype == VIO_TYPE_ERR);
3111 	cmn_err(CE_NOTE, "[%d] Got an ERR msg", vdc->instance);
3112 
3113 	return (ENOTSUP);
3114 }
3115 
3116 /*
3117  * Function:
3118  *	vdc_handle_ver_msg()
3119  *
3120  * Description:
3121  *
3122  * Arguments:
3123  *	vdc	- soft state pointer for this instance of the device driver.
3124  *	ver_msg	- LDC message sent by vDisk server
3125  *
3126  * Return Code:
3127  *	0	- Success
3128  */
3129 static int
3130 vdc_handle_ver_msg(vdc_t *vdc, vio_ver_msg_t *ver_msg)
3131 {
3132 	int status = 0;
3133 
3134 	ASSERT(vdc != NULL);
3135 	ASSERT(mutex_owned(&vdc->lock));
3136 
3137 	if (ver_msg->tag.vio_subtype_env != VIO_VER_INFO) {
3138 		return (EPROTO);
3139 	}
3140 
3141 	if (ver_msg->dev_class != VDEV_DISK_SERVER) {
3142 		return (EINVAL);
3143 	}
3144 
3145 	switch (ver_msg->tag.vio_subtype) {
3146 	case VIO_SUBTYPE_ACK:
3147 		/*
3148 		 * We check to see if the version returned is indeed supported
3149 		 * (The server may have also adjusted the minor number downwards
3150 		 * and if so 'ver_msg' will contain the actual version agreed)
3151 		 */
3152 		if (vdc_is_supported_version(ver_msg)) {
3153 			vdc->ver.major = ver_msg->ver_major;
3154 			vdc->ver.minor = ver_msg->ver_minor;
3155 			ASSERT(vdc->ver.major > 0);
3156 
3157 			vdc->state = VD_STATE_VER;
3158 			status = vdc_init_attr_negotiation(vdc);
3159 		} else {
3160 			status = EPROTO;
3161 		}
3162 		break;
3163 
3164 	case VIO_SUBTYPE_NACK:
3165 		/*
3166 		 * call vdc_is_supported_version() which will return the next
3167 		 * supported version (if any) in 'ver_msg'
3168 		 */
3169 		(void) vdc_is_supported_version(ver_msg);
3170 		if (ver_msg->ver_major > 0) {
3171 			size_t len = sizeof (*ver_msg);
3172 
3173 			ASSERT(vdc->ver.major > 0);
3174 
3175 			/* reset the necessary fields and resend */
3176 			ver_msg->tag.vio_subtype = VIO_SUBTYPE_INFO;
3177 			ver_msg->dev_class = VDEV_DISK;
3178 
3179 			status = vdc_send(vdc, (caddr_t)ver_msg, &len);
3180 			DMSG(0, "[%d] Resend VER info (LDC status = %d)\n",
3181 					vdc->instance, status);
3182 			if (len != sizeof (*ver_msg))
3183 				status = EBADMSG;
3184 		} else {
3185 			cmn_err(CE_NOTE, "[%d] No common version with "
3186 					"vDisk server", vdc->instance);
3187 			status = ENOTSUP;
3188 		}
3189 
3190 		break;
3191 	case VIO_SUBTYPE_INFO:
3192 		/*
3193 		 * Handle the case where vds starts handshake
3194 		 * (for now only vdc is the instigatior)
3195 		 */
3196 		status = ENOTSUP;
3197 		break;
3198 
3199 	default:
3200 		status = EINVAL;
3201 		break;
3202 	}
3203 
3204 	return (status);
3205 }
3206 
3207 /*
3208  * Function:
3209  *	vdc_handle_attr_msg()
3210  *
3211  * Description:
3212  *
3213  * Arguments:
3214  *	vdc	- soft state pointer for this instance of the device driver.
3215  *	attr_msg	- LDC message sent by vDisk server
3216  *
3217  * Return Code:
3218  *	0	- Success
3219  */
3220 static int
3221 vdc_handle_attr_msg(vdc_t *vdc, vd_attr_msg_t *attr_msg)
3222 {
3223 	int status = 0;
3224 
3225 	ASSERT(vdc != NULL);
3226 	ASSERT(mutex_owned(&vdc->lock));
3227 
3228 	if (attr_msg->tag.vio_subtype_env != VIO_ATTR_INFO) {
3229 		return (EPROTO);
3230 	}
3231 
3232 	switch (attr_msg->tag.vio_subtype) {
3233 	case VIO_SUBTYPE_ACK:
3234 		/*
3235 		 * We now verify the attributes sent by vds.
3236 		 */
3237 		vdc->vdisk_size = attr_msg->vdisk_size;
3238 		vdc->vdisk_type = attr_msg->vdisk_type;
3239 
3240 		DMSG(0, "[%d] max_xfer_sz: sent %lx acked %lx\n",
3241 			vdc->instance, vdc->max_xfer_sz, attr_msg->max_xfer_sz);
3242 		DMSG(0, "[%d] vdisk_block_size: sent %lx acked %x\n",
3243 			vdc->instance, vdc->block_size,
3244 			attr_msg->vdisk_block_size);
3245 
3246 		/*
3247 		 * We don't know at compile time what the vDisk server will
3248 		 * think are good values but we apply an large (arbitrary)
3249 		 * upper bound to prevent memory exhaustion in vdc if it was
3250 		 * allocating a DRing based of huge values sent by the server.
3251 		 * We probably will never exceed this except if the message
3252 		 * was garbage.
3253 		 */
3254 		if ((attr_msg->max_xfer_sz * attr_msg->vdisk_block_size) <=
3255 				(PAGESIZE * DEV_BSIZE)) {
3256 			vdc->max_xfer_sz = attr_msg->max_xfer_sz;
3257 			vdc->block_size = attr_msg->vdisk_block_size;
3258 		} else {
3259 			cmn_err(CE_NOTE, "[%d] vds block transfer size too big;"
3260 				" using max supported by vdc", vdc->instance);
3261 		}
3262 
3263 		if ((attr_msg->xfer_mode != VIO_DRING_MODE) ||
3264 		    (attr_msg->vdisk_size > INT64_MAX) ||
3265 		    (attr_msg->vdisk_type > VD_DISK_TYPE_DISK)) {
3266 			cmn_err(CE_NOTE, "[%d] Invalid attributes from vds",
3267 					vdc->instance);
3268 			status = EINVAL;
3269 			break;
3270 		}
3271 
3272 		vdc->state = VD_STATE_ATTR;
3273 		status = vdc_init_dring_negotiate(vdc);
3274 		break;
3275 
3276 	case VIO_SUBTYPE_NACK:
3277 		/*
3278 		 * vds could not handle the attributes we sent so we
3279 		 * stop negotiating.
3280 		 */
3281 		status = EPROTO;
3282 		break;
3283 
3284 	case VIO_SUBTYPE_INFO:
3285 		/*
3286 		 * Handle the case where vds starts the handshake
3287 		 * (for now; vdc is the only supported instigatior)
3288 		 */
3289 		status = ENOTSUP;
3290 		break;
3291 
3292 	default:
3293 		status = ENOTSUP;
3294 		break;
3295 	}
3296 
3297 	return (status);
3298 }
3299 
3300 /*
3301  * Function:
3302  *	vdc_handle_dring_reg_msg()
3303  *
3304  * Description:
3305  *
3306  * Arguments:
3307  *	vdc		- soft state pointer for this instance of the driver.
3308  *	dring_msg	- LDC message sent by vDisk server
3309  *
3310  * Return Code:
3311  *	0	- Success
3312  */
3313 static int
3314 vdc_handle_dring_reg_msg(vdc_t *vdc, vio_dring_reg_msg_t *dring_msg)
3315 {
3316 	int		status = 0;
3317 	vio_rdx_msg_t	msg = {0};
3318 	size_t		msglen = sizeof (msg);
3319 
3320 	ASSERT(vdc != NULL);
3321 	ASSERT(mutex_owned(&vdc->lock));
3322 
3323 	if (dring_msg->tag.vio_subtype_env != VIO_DRING_REG) {
3324 		return (EPROTO);
3325 	}
3326 
3327 	switch (dring_msg->tag.vio_subtype) {
3328 	case VIO_SUBTYPE_ACK:
3329 		/* save the received dring_ident */
3330 		vdc->dring_ident = dring_msg->dring_ident;
3331 		DMSG(0, "[%d] Received dring ident=0x%lx\n",
3332 			vdc->instance, vdc->dring_ident);
3333 
3334 		/*
3335 		 * Send an RDX message to vds to indicate we are ready
3336 		 * to send data
3337 		 */
3338 		msg.tag.vio_msgtype = VIO_TYPE_CTRL;
3339 		msg.tag.vio_subtype = VIO_SUBTYPE_INFO;
3340 		msg.tag.vio_subtype_env = VIO_RDX;
3341 		msg.tag.vio_sid = vdc->session_id;
3342 		status = vdc_send(vdc, (caddr_t)&msg, &msglen);
3343 		if (status != 0) {
3344 			cmn_err(CE_NOTE, "[%d] Failed to send RDX"
3345 				" message (%d)", vdc->instance, status);
3346 			break;
3347 		}
3348 
3349 		vdc->state = VD_STATE_RDX;
3350 		break;
3351 
3352 	case VIO_SUBTYPE_NACK:
3353 		/*
3354 		 * vds could not handle the DRing info we sent so we
3355 		 * stop negotiating.
3356 		 */
3357 		cmn_err(CE_CONT, "server could not register DRing\n");
3358 		vdc_reset_connection(vdc, B_TRUE);
3359 		vdc_destroy_descriptor_ring(vdc);
3360 		status = EPROTO;
3361 		break;
3362 
3363 	case VIO_SUBTYPE_INFO:
3364 		/*
3365 		 * Handle the case where vds starts handshake
3366 		 * (for now only vdc is the instigatior)
3367 		 */
3368 		status = ENOTSUP;
3369 		break;
3370 	default:
3371 		status = ENOTSUP;
3372 	}
3373 
3374 	return (status);
3375 }
3376 
3377 /*
3378  * Function:
3379  *	vdc_verify_seq_num()
3380  *
3381  * Description:
3382  *	This functions verifies that the sequence number sent back by the vDisk
3383  *	server with the latest message is what is expected (i.e. it is greater
3384  *	than the last seq num sent by the vDisk server and less than or equal
3385  *	to the last seq num generated by vdc).
3386  *
3387  *	It then checks the request ID to see if any requests need processing
3388  *	in the DRing.
3389  *
3390  * Arguments:
3391  *	vdc		- soft state pointer for this instance of the driver.
3392  *	dring_msg	- pointer to the LDC message sent by vds
3393  *
3394  * Return Code:
3395  *	VDC_SEQ_NUM_TODO	- Message needs to be processed
3396  *	VDC_SEQ_NUM_SKIP	- Message has already been processed
3397  *	VDC_SEQ_NUM_INVALID	- The seq numbers are so out of sync,
3398  *				  vdc cannot deal with them
3399  */
3400 static int
3401 vdc_verify_seq_num(vdc_t *vdc, vio_dring_msg_t *dring_msg)
3402 {
3403 	ASSERT(vdc != NULL);
3404 	ASSERT(dring_msg != NULL);
3405 	ASSERT(mutex_owned(&vdc->lock));
3406 
3407 	/*
3408 	 * Check to see if the messages were responded to in the correct
3409 	 * order by vds.
3410 	 */
3411 	if ((dring_msg->seq_num <= vdc->seq_num_reply) ||
3412 	    (dring_msg->seq_num > vdc->seq_num)) {
3413 		cmn_err(CE_CONT, "?[%d] Bogus sequence_number %lu: "
3414 			"%lu > expected <= %lu (last proc req %lu sent %lu)\n",
3415 				vdc->instance, dring_msg->seq_num,
3416 				vdc->seq_num_reply, vdc->seq_num,
3417 				vdc->req_id_proc, vdc->req_id);
3418 		return (VDC_SEQ_NUM_INVALID);
3419 	}
3420 	vdc->seq_num_reply = dring_msg->seq_num;
3421 
3422 	if (vdc->req_id_proc < vdc->req_id)
3423 		return (VDC_SEQ_NUM_TODO);
3424 	else
3425 		return (VDC_SEQ_NUM_SKIP);
3426 }
3427 
3428 
3429 /*
3430  * Function:
3431  *	vdc_is_supported_version()
3432  *
3433  * Description:
3434  *	This routine checks if the major/minor version numbers specified in
3435  *	'ver_msg' are supported. If not it finds the next version that is
3436  *	in the supported version list 'vdc_version[]' and sets the fields in
3437  *	'ver_msg' to those values
3438  *
3439  * Arguments:
3440  *	ver_msg	- LDC message sent by vDisk server
3441  *
3442  * Return Code:
3443  *	B_TRUE	- Success
3444  *	B_FALSE	- Version not supported
3445  */
3446 static boolean_t
3447 vdc_is_supported_version(vio_ver_msg_t *ver_msg)
3448 {
3449 	int vdc_num_versions = sizeof (vdc_version) / sizeof (vdc_version[0]);
3450 
3451 	for (int i = 0; i < vdc_num_versions; i++) {
3452 		ASSERT(vdc_version[i].major > 0);
3453 		ASSERT((i == 0) ||
3454 		    (vdc_version[i].major < vdc_version[i-1].major));
3455 
3456 		/*
3457 		 * If the major versions match, adjust the minor version, if
3458 		 * necessary, down to the highest value supported by this
3459 		 * client. The server should support all minor versions lower
3460 		 * than the value it sent
3461 		 */
3462 		if (ver_msg->ver_major == vdc_version[i].major) {
3463 			if (ver_msg->ver_minor > vdc_version[i].minor) {
3464 				DMSG(0, "Adjusting minor version from %u to %u",
3465 				    ver_msg->ver_minor, vdc_version[i].minor);
3466 				ver_msg->ver_minor = vdc_version[i].minor;
3467 			}
3468 			return (B_TRUE);
3469 		}
3470 
3471 		/*
3472 		 * If the message contains a higher major version number, set
3473 		 * the message's major/minor versions to the current values
3474 		 * and return false, so this message will get resent with
3475 		 * these values, and the server will potentially try again
3476 		 * with the same or a lower version
3477 		 */
3478 		if (ver_msg->ver_major > vdc_version[i].major) {
3479 			ver_msg->ver_major = vdc_version[i].major;
3480 			ver_msg->ver_minor = vdc_version[i].minor;
3481 			DMSG(0, "Suggesting major/minor (0x%x/0x%x)\n",
3482 				ver_msg->ver_major, ver_msg->ver_minor);
3483 
3484 			return (B_FALSE);
3485 		}
3486 
3487 		/*
3488 		 * Otherwise, the message's major version is less than the
3489 		 * current major version, so continue the loop to the next
3490 		 * (lower) supported version
3491 		 */
3492 	}
3493 
3494 	/*
3495 	 * No common version was found; "ground" the version pair in the
3496 	 * message to terminate negotiation
3497 	 */
3498 	ver_msg->ver_major = 0;
3499 	ver_msg->ver_minor = 0;
3500 
3501 	return (B_FALSE);
3502 }
3503 /* -------------------------------------------------------------------------- */
3504 
3505 /*
3506  * DKIO(7) support
3507  */
3508 
3509 typedef struct vdc_dk_arg {
3510 	struct dk_callback	dkc;
3511 	int			mode;
3512 	dev_t			dev;
3513 	vdc_t			*vdc;
3514 } vdc_dk_arg_t;
3515 
3516 /*
3517  * Function:
3518  * 	vdc_dkio_flush_cb()
3519  *
3520  * Description:
3521  *	This routine is a callback for DKIOCFLUSHWRITECACHE which can be called
3522  *	by kernel code.
3523  *
3524  * Arguments:
3525  *	arg	- a pointer to a vdc_dk_arg_t structure.
3526  */
3527 void
3528 vdc_dkio_flush_cb(void *arg)
3529 {
3530 	struct vdc_dk_arg	*dk_arg = (struct vdc_dk_arg *)arg;
3531 	struct dk_callback	*dkc = NULL;
3532 	vdc_t			*vdc = NULL;
3533 	int			rv;
3534 
3535 	if (dk_arg == NULL) {
3536 		cmn_err(CE_CONT, "?[Unk] DKIOCFLUSHWRITECACHE arg is NULL\n");
3537 		return;
3538 	}
3539 	dkc = &dk_arg->dkc;
3540 	vdc = dk_arg->vdc;
3541 	ASSERT(vdc != NULL);
3542 
3543 	rv = vdc_populate_descriptor(vdc, NULL, 0, VD_OP_FLUSH,
3544 		dk_arg->mode, SDPART(getminor(dk_arg->dev)));
3545 	if (rv != 0) {
3546 		DMSG(0, "[%d] DKIOCFLUSHWRITECACHE failed %d : model %x\n",
3547 			vdc->instance, rv,
3548 			ddi_model_convert_from(dk_arg->mode & FMODELS));
3549 	}
3550 
3551 	/*
3552 	 * Trigger the call back to notify the caller the the ioctl call has
3553 	 * been completed.
3554 	 */
3555 	if ((dk_arg->mode & FKIOCTL) &&
3556 	    (dkc != NULL) &&
3557 	    (dkc->dkc_callback != NULL)) {
3558 		ASSERT(dkc->dkc_cookie != NULL);
3559 		(*dkc->dkc_callback)(dkc->dkc_cookie, rv);
3560 	}
3561 
3562 	/* Indicate that one less DKIO write flush is outstanding */
3563 	mutex_enter(&vdc->lock);
3564 	vdc->dkio_flush_pending--;
3565 	ASSERT(vdc->dkio_flush_pending >= 0);
3566 	mutex_exit(&vdc->lock);
3567 
3568 	/* free the mem that was allocated when the callback was dispatched */
3569 	kmem_free(arg, sizeof (vdc_dk_arg_t));
3570 }
3571 
3572 /*
3573  * This structure is used in the DKIO(7I) array below.
3574  */
3575 typedef struct vdc_dk_ioctl {
3576 	uint8_t		op;		/* VD_OP_XXX value */
3577 	int		cmd;		/* Solaris ioctl operation number */
3578 	size_t		nbytes;		/* size of structure to be copied */
3579 
3580 	/* function to convert between vDisk and Solaris structure formats */
3581 	int	(*convert)(vdc_t *vdc, void *vd_buf, void *ioctl_arg,
3582 	    int mode, int dir);
3583 } vdc_dk_ioctl_t;
3584 
3585 /*
3586  * Subset of DKIO(7I) operations currently supported
3587  */
3588 static vdc_dk_ioctl_t	dk_ioctl[] = {
3589 	{VD_OP_FLUSH,		DKIOCFLUSHWRITECACHE,	sizeof (int),
3590 		vdc_null_copy_func},
3591 	{VD_OP_GET_WCE,		DKIOCGETWCE,		sizeof (int),
3592 		vdc_null_copy_func},
3593 	{VD_OP_SET_WCE,		DKIOCSETWCE,		sizeof (int),
3594 		vdc_null_copy_func},
3595 	{VD_OP_GET_VTOC,	DKIOCGVTOC,		sizeof (vd_vtoc_t),
3596 		vdc_get_vtoc_convert},
3597 	{VD_OP_SET_VTOC,	DKIOCSVTOC,		sizeof (vd_vtoc_t),
3598 		vdc_set_vtoc_convert},
3599 	{VD_OP_GET_DISKGEOM,	DKIOCGGEOM,		sizeof (vd_geom_t),
3600 		vdc_get_geom_convert},
3601 	{VD_OP_GET_DISKGEOM,	DKIOCG_PHYGEOM,		sizeof (vd_geom_t),
3602 		vdc_get_geom_convert},
3603 	{VD_OP_GET_DISKGEOM, 	DKIOCG_VIRTGEOM,	sizeof (vd_geom_t),
3604 		vdc_get_geom_convert},
3605 	{VD_OP_SET_DISKGEOM,	DKIOCSGEOM,		sizeof (vd_geom_t),
3606 		vdc_set_geom_convert},
3607 
3608 	/*
3609 	 * These particular ioctls are not sent to the server - vdc fakes up
3610 	 * the necessary info.
3611 	 */
3612 	{0, DKIOCINFO, sizeof (struct dk_cinfo), vdc_null_copy_func},
3613 	{0, DKIOCGMEDIAINFO, sizeof (struct dk_minfo), vdc_null_copy_func},
3614 	{0, USCSICMD,	sizeof (struct uscsi_cmd), vdc_null_copy_func},
3615 	{0, DKIOCREMOVABLE, 0, vdc_null_copy_func},
3616 	{0, CDROMREADOFFSET, 0, vdc_null_copy_func}
3617 };
3618 
3619 /*
3620  * Function:
3621  *	vd_process_ioctl()
3622  *
3623  * Description:
3624  *	This routine processes disk specific ioctl calls
3625  *
3626  * Arguments:
3627  *	dev	- the device number
3628  *	cmd	- the operation [dkio(7I)] to be processed
3629  *	arg	- pointer to user provided structure
3630  *		  (contains data to be set or reference parameter for get)
3631  *	mode	- bit flag, indicating open settings, 32/64 bit type, etc
3632  *
3633  * Return Code:
3634  *	0
3635  *	EFAULT
3636  *	ENXIO
3637  *	EIO
3638  *	ENOTSUP
3639  */
3640 static int
3641 vd_process_ioctl(dev_t dev, int cmd, caddr_t arg, int mode)
3642 {
3643 	int		instance = SDUNIT(getminor(dev));
3644 	vdc_t		*vdc = NULL;
3645 	int		rv = -1;
3646 	int		idx = 0;		/* index into dk_ioctl[] */
3647 	size_t		len = 0;		/* #bytes to send to vds */
3648 	size_t		alloc_len = 0;		/* #bytes to allocate mem for */
3649 	caddr_t		mem_p = NULL;
3650 	size_t		nioctls = (sizeof (dk_ioctl)) / (sizeof (dk_ioctl[0]));
3651 	struct vtoc	vtoc_saved;
3652 
3653 	DMSG(0, "[%d] Processing ioctl(%x) for dev %lx : model %x\n",
3654 		instance, cmd, dev, ddi_model_convert_from(mode & FMODELS));
3655 
3656 	vdc = ddi_get_soft_state(vdc_state, instance);
3657 	if (vdc == NULL) {
3658 		cmn_err(CE_NOTE, "![%d] Could not get soft state structure",
3659 		    instance);
3660 		return (ENXIO);
3661 	}
3662 
3663 	/*
3664 	 * Check to see if we can communicate with the vDisk server
3665 	 */
3666 	if (!vdc_is_able_to_tx_data(vdc, O_NONBLOCK)) {
3667 		DMSG(0, "[%d] Not ready to transmit data\n", instance);
3668 		return (ENOLINK);
3669 	}
3670 
3671 	/*
3672 	 * Validate the ioctl operation to be performed.
3673 	 *
3674 	 * If we have looped through the array without finding a match then we
3675 	 * don't support this ioctl.
3676 	 */
3677 	for (idx = 0; idx < nioctls; idx++) {
3678 		if (cmd == dk_ioctl[idx].cmd)
3679 			break;
3680 	}
3681 
3682 	if (idx >= nioctls) {
3683 		cmn_err(CE_CONT, "?[%d] Unsupported ioctl (0x%x)\n",
3684 				vdc->instance, cmd);
3685 		return (ENOTSUP);
3686 	}
3687 
3688 	len = dk_ioctl[idx].nbytes;
3689 
3690 	/*
3691 	 * Deal with the ioctls which the server does not provide. vdc can
3692 	 * fake these up and return immediately
3693 	 */
3694 	switch (cmd) {
3695 	case CDROMREADOFFSET:
3696 	case DKIOCREMOVABLE:
3697 	case USCSICMD:
3698 		return (ENOTTY);
3699 
3700 	case DKIOCINFO:
3701 		{
3702 			struct dk_cinfo	cinfo;
3703 			if (vdc->cinfo == NULL)
3704 				return (ENXIO);
3705 
3706 			bcopy(vdc->cinfo, &cinfo, sizeof (struct dk_cinfo));
3707 			cinfo.dki_partition = SDPART(getminor(dev));
3708 
3709 			rv = ddi_copyout(&cinfo, (void *)arg,
3710 					sizeof (struct dk_cinfo), mode);
3711 			if (rv != 0)
3712 				return (EFAULT);
3713 
3714 			return (0);
3715 		}
3716 
3717 	case DKIOCGMEDIAINFO:
3718 		{
3719 			if (vdc->minfo == NULL)
3720 				return (ENXIO);
3721 
3722 			rv = ddi_copyout(vdc->minfo, (void *)arg,
3723 					sizeof (struct dk_minfo), mode);
3724 			if (rv != 0)
3725 				return (EFAULT);
3726 
3727 			return (0);
3728 		}
3729 
3730 	case DKIOCFLUSHWRITECACHE:
3731 		{
3732 			struct dk_callback *dkc = (struct dk_callback *)arg;
3733 			vdc_dk_arg_t	*dkarg = NULL;
3734 
3735 			DMSG(1, "[%d] Flush W$: mode %x\n", instance, mode);
3736 
3737 			/*
3738 			 * If the backing device is not a 'real' disk then the
3739 			 * W$ operation request to the vDisk server will fail
3740 			 * so we might as well save the cycles and return now.
3741 			 */
3742 			if (vdc->vdisk_type != VD_DISK_TYPE_DISK)
3743 				return (ENOTTY);
3744 
3745 			/*
3746 			 * If arg is NULL, then there is no callback function
3747 			 * registered and the call operates synchronously; we
3748 			 * break and continue with the rest of the function and
3749 			 * wait for vds to return (i.e. after the request to
3750 			 * vds returns successfully, all writes completed prior
3751 			 * to the ioctl will have been flushed from the disk
3752 			 * write cache to persistent media.
3753 			 *
3754 			 * If a callback function is registered, we dispatch
3755 			 * the request on a task queue and return immediately.
3756 			 * The callback will deal with informing the calling
3757 			 * thread that the flush request is completed.
3758 			 */
3759 			if (dkc == NULL)
3760 				break;
3761 
3762 			dkarg = kmem_zalloc(sizeof (vdc_dk_arg_t), KM_SLEEP);
3763 
3764 			dkarg->mode = mode;
3765 			dkarg->dev = dev;
3766 			bcopy(dkc, &dkarg->dkc, sizeof (*dkc));
3767 
3768 			mutex_enter(&vdc->lock);
3769 			vdc->dkio_flush_pending++;
3770 			dkarg->vdc = vdc;
3771 			mutex_exit(&vdc->lock);
3772 
3773 			/* put the request on a task queue */
3774 			rv = taskq_dispatch(system_taskq, vdc_dkio_flush_cb,
3775 				(void *)dkarg, DDI_SLEEP);
3776 
3777 			return (rv == NULL ? ENOMEM : 0);
3778 		}
3779 	}
3780 
3781 	/* catch programming error in vdc - should be a VD_OP_XXX ioctl */
3782 	ASSERT(dk_ioctl[idx].op != 0);
3783 
3784 	/* LDC requires that the memory being mapped is 8-byte aligned */
3785 	alloc_len = P2ROUNDUP(len, sizeof (uint64_t));
3786 	DMSG(1, "[%d] struct size %ld alloc %ld\n", instance, len, alloc_len);
3787 
3788 	ASSERT(alloc_len != 0);	/* sanity check */
3789 	mem_p = kmem_zalloc(alloc_len, KM_SLEEP);
3790 
3791 	if (cmd == DKIOCSVTOC) {
3792 		/*
3793 		 * Save a copy of the current VTOC so that we can roll back
3794 		 * if the setting of the new VTOC fails.
3795 		 */
3796 		bcopy(vdc->vtoc, &vtoc_saved, sizeof (struct vtoc));
3797 	}
3798 
3799 	/*
3800 	 * Call the conversion function for this ioctl whhich if necessary
3801 	 * converts from the Solaris format to the format ARC'ed
3802 	 * as part of the vDisk protocol (FWARC 2006/195)
3803 	 */
3804 	ASSERT(dk_ioctl[idx].convert != NULL);
3805 	rv = (dk_ioctl[idx].convert)(vdc, arg, mem_p, mode, VD_COPYIN);
3806 	if (rv != 0) {
3807 		DMSG(0, "[%d] convert func returned %d for ioctl 0x%x\n",
3808 				instance, rv, cmd);
3809 		if (mem_p != NULL)
3810 			kmem_free(mem_p, alloc_len);
3811 		return (rv);
3812 	}
3813 
3814 	/*
3815 	 * send request to vds to service the ioctl.
3816 	 */
3817 	rv = vdc_populate_descriptor(vdc, mem_p, alloc_len, dk_ioctl[idx].op,
3818 			mode, SDPART((getminor(dev))));
3819 	if (rv != 0) {
3820 		/*
3821 		 * This is not necessarily an error. The ioctl could
3822 		 * be returning a value such as ENOTTY to indicate
3823 		 * that the ioctl is not applicable.
3824 		 */
3825 		DMSG(0, "[%d] vds returned %d for ioctl 0x%x\n",
3826 			instance, rv, cmd);
3827 		if (mem_p != NULL)
3828 			kmem_free(mem_p, alloc_len);
3829 
3830 		if (cmd == DKIOCSVTOC) {
3831 			/* update of the VTOC has failed, roll back */
3832 			bcopy(&vtoc_saved, vdc->vtoc, sizeof (struct vtoc));
3833 		}
3834 
3835 		return (rv);
3836 	}
3837 
3838 	if (cmd == DKIOCSVTOC) {
3839 		/*
3840 		 * The VTOC has been changed, try and update the device
3841 		 * node properties. Failing to set the properties should
3842 		 * not cause an error to be return the caller though.
3843 		 */
3844 		if (vdc_create_device_nodes_props(vdc)) {
3845 			cmn_err(CE_NOTE, "![%d] Failed to update device nodes"
3846 			    " properties", vdc->instance);
3847 		}
3848 	}
3849 
3850 	/*
3851 	 * Call the conversion function (if it exists) for this ioctl
3852 	 * which converts from the format ARC'ed as part of the vDisk
3853 	 * protocol (FWARC 2006/195) back to a format understood by
3854 	 * the rest of Solaris.
3855 	 */
3856 	rv = (dk_ioctl[idx].convert)(vdc, mem_p, arg, mode, VD_COPYOUT);
3857 	if (rv != 0) {
3858 		DMSG(0, "[%d] convert func returned %d for ioctl 0x%x\n",
3859 				instance, rv, cmd);
3860 		if (mem_p != NULL)
3861 			kmem_free(mem_p, alloc_len);
3862 		return (rv);
3863 	}
3864 
3865 	if (mem_p != NULL)
3866 		kmem_free(mem_p, alloc_len);
3867 
3868 	return (rv);
3869 }
3870 
3871 /*
3872  * Function:
3873  *
3874  * Description:
3875  *	This is an empty conversion function used by ioctl calls which
3876  *	do not need to convert the data being passed in/out to userland
3877  */
3878 static int
3879 vdc_null_copy_func(vdc_t *vdc, void *from, void *to, int mode, int dir)
3880 {
3881 	_NOTE(ARGUNUSED(vdc))
3882 	_NOTE(ARGUNUSED(from))
3883 	_NOTE(ARGUNUSED(to))
3884 	_NOTE(ARGUNUSED(mode))
3885 	_NOTE(ARGUNUSED(dir))
3886 
3887 	return (0);
3888 }
3889 
3890 /*
3891  * Function:
3892  *	vdc_get_vtoc_convert()
3893  *
3894  * Description:
3895  *	This routine performs the necessary convertions from the DKIOCGVTOC
3896  *	Solaris structure to the format defined in FWARC 2006/195.
3897  *
3898  *	In the struct vtoc definition, the timestamp field is marked as not
3899  *	supported so it is not part of vDisk protocol (FWARC 2006/195).
3900  *	However SVM uses that field to check it can write into the VTOC,
3901  *	so we fake up the info of that field.
3902  *
3903  * Arguments:
3904  *	vdc	- the vDisk client
3905  *	from	- the buffer containing the data to be copied from
3906  *	to	- the buffer to be copied to
3907  *	mode	- flags passed to ioctl() call
3908  *	dir	- the "direction" of the copy - VD_COPYIN or VD_COPYOUT
3909  *
3910  * Return Code:
3911  *	0	- Success
3912  *	ENXIO	- incorrect buffer passed in.
3913  *	EFAULT	- ddi_copyout routine encountered an error.
3914  */
3915 static int
3916 vdc_get_vtoc_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
3917 {
3918 	int		i;
3919 	void		*tmp_mem = NULL;
3920 	void		*tmp_memp;
3921 	struct vtoc	vt;
3922 	struct vtoc32	vt32;
3923 	int		copy_len = 0;
3924 	int		rv = 0;
3925 
3926 	if (dir != VD_COPYOUT)
3927 		return (0);	/* nothing to do */
3928 
3929 	if ((from == NULL) || (to == NULL))
3930 		return (ENXIO);
3931 
3932 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32)
3933 		copy_len = sizeof (struct vtoc32);
3934 	else
3935 		copy_len = sizeof (struct vtoc);
3936 
3937 	tmp_mem = kmem_alloc(copy_len, KM_SLEEP);
3938 
3939 	VD_VTOC2VTOC((vd_vtoc_t *)from, &vt);
3940 
3941 	/* fake the VTOC timestamp field */
3942 	for (i = 0; i < V_NUMPAR; i++) {
3943 		vt.timestamp[i] = vdc->vtoc->timestamp[i];
3944 	}
3945 
3946 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
3947 		vtoctovtoc32(vt, vt32);
3948 		tmp_memp = &vt32;
3949 	} else {
3950 		tmp_memp = &vt;
3951 	}
3952 	rv = ddi_copyout(tmp_memp, to, copy_len, mode);
3953 	if (rv != 0)
3954 		rv = EFAULT;
3955 
3956 	kmem_free(tmp_mem, copy_len);
3957 	return (rv);
3958 }
3959 
3960 /*
3961  * Function:
3962  *	vdc_set_vtoc_convert()
3963  *
3964  * Description:
3965  *	This routine performs the necessary convertions from the DKIOCSVTOC
3966  *	Solaris structure to the format defined in FWARC 2006/195.
3967  *
3968  * Arguments:
3969  *	vdc	- the vDisk client
3970  *	from	- Buffer with data
3971  *	to	- Buffer where data is to be copied to
3972  *	mode	- flags passed to ioctl
3973  *	dir	- direction of copy (in or out)
3974  *
3975  * Return Code:
3976  *	0	- Success
3977  *	ENXIO	- Invalid buffer passed in
3978  *	EFAULT	- ddi_copyin of data failed
3979  */
3980 static int
3981 vdc_set_vtoc_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
3982 {
3983 	void		*tmp_mem = NULL;
3984 	struct vtoc	vt;
3985 	struct vtoc	*vtp = &vt;
3986 	vd_vtoc_t	vtvd;
3987 	int		copy_len = 0;
3988 	int		rv = 0;
3989 
3990 	if (dir != VD_COPYIN)
3991 		return (0);	/* nothing to do */
3992 
3993 	if ((from == NULL) || (to == NULL))
3994 		return (ENXIO);
3995 
3996 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32)
3997 		copy_len = sizeof (struct vtoc32);
3998 	else
3999 		copy_len = sizeof (struct vtoc);
4000 
4001 	tmp_mem = kmem_alloc(copy_len, KM_SLEEP);
4002 
4003 	rv = ddi_copyin(from, tmp_mem, copy_len, mode);
4004 	if (rv != 0) {
4005 		kmem_free(tmp_mem, copy_len);
4006 		return (EFAULT);
4007 	}
4008 
4009 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
4010 		vtoc32tovtoc((*(struct vtoc32 *)tmp_mem), vt);
4011 	} else {
4012 		vtp = tmp_mem;
4013 	}
4014 
4015 	/*
4016 	 * The VTOC is being changed, then vdc needs to update the copy
4017 	 * it saved in the soft state structure.
4018 	 */
4019 	bcopy(vtp, vdc->vtoc, sizeof (struct vtoc));
4020 
4021 	VTOC2VD_VTOC(vtp, &vtvd);
4022 	bcopy(&vtvd, to, sizeof (vd_vtoc_t));
4023 	kmem_free(tmp_mem, copy_len);
4024 
4025 	return (0);
4026 }
4027 
4028 /*
4029  * Function:
4030  *	vdc_get_geom_convert()
4031  *
4032  * Description:
4033  *	This routine performs the necessary convertions from the DKIOCGGEOM,
4034  *	DKIOCG_PHYSGEOM and DKIOG_VIRTGEOM Solaris structures to the format
4035  *	defined in FWARC 2006/195
4036  *
4037  * Arguments:
4038  *	vdc	- the vDisk client
4039  *	from	- Buffer with data
4040  *	to	- Buffer where data is to be copied to
4041  *	mode	- flags passed to ioctl
4042  *	dir	- direction of copy (in or out)
4043  *
4044  * Return Code:
4045  *	0	- Success
4046  *	ENXIO	- Invalid buffer passed in
4047  *	EFAULT	- ddi_copyout of data failed
4048  */
4049 static int
4050 vdc_get_geom_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
4051 {
4052 	_NOTE(ARGUNUSED(vdc))
4053 
4054 	struct dk_geom	geom;
4055 	int	copy_len = sizeof (struct dk_geom);
4056 	int	rv = 0;
4057 
4058 	if (dir != VD_COPYOUT)
4059 		return (0);	/* nothing to do */
4060 
4061 	if ((from == NULL) || (to == NULL))
4062 		return (ENXIO);
4063 
4064 	VD_GEOM2DK_GEOM((vd_geom_t *)from, &geom);
4065 	rv = ddi_copyout(&geom, to, copy_len, mode);
4066 	if (rv != 0)
4067 		rv = EFAULT;
4068 
4069 	return (rv);
4070 }
4071 
4072 /*
4073  * Function:
4074  *	vdc_set_geom_convert()
4075  *
4076  * Description:
4077  *	This routine performs the necessary convertions from the DKIOCSGEOM
4078  *	Solaris structure to the format defined in FWARC 2006/195.
4079  *
4080  * Arguments:
4081  *	vdc	- the vDisk client
4082  *	from	- Buffer with data
4083  *	to	- Buffer where data is to be copied to
4084  *	mode	- flags passed to ioctl
4085  *	dir	- direction of copy (in or out)
4086  *
4087  * Return Code:
4088  *	0	- Success
4089  *	ENXIO	- Invalid buffer passed in
4090  *	EFAULT	- ddi_copyin of data failed
4091  */
4092 static int
4093 vdc_set_geom_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
4094 {
4095 	_NOTE(ARGUNUSED(vdc))
4096 
4097 	vd_geom_t	vdgeom;
4098 	void		*tmp_mem = NULL;
4099 	int		copy_len = sizeof (struct dk_geom);
4100 	int		rv = 0;
4101 
4102 	if (dir != VD_COPYIN)
4103 		return (0);	/* nothing to do */
4104 
4105 	if ((from == NULL) || (to == NULL))
4106 		return (ENXIO);
4107 
4108 	tmp_mem = kmem_alloc(copy_len, KM_SLEEP);
4109 
4110 	rv = ddi_copyin(from, tmp_mem, copy_len, mode);
4111 	if (rv != 0) {
4112 		kmem_free(tmp_mem, copy_len);
4113 		return (EFAULT);
4114 	}
4115 	DK_GEOM2VD_GEOM((struct dk_geom *)tmp_mem, &vdgeom);
4116 	bcopy(&vdgeom, to, sizeof (vdgeom));
4117 	kmem_free(tmp_mem, copy_len);
4118 
4119 	return (0);
4120 }
4121 
4122 /*
4123  * Function:
4124  *	vdc_create_fake_geometry()
4125  *
4126  * Description:
4127  *	This routine fakes up the disk info needed for some DKIO ioctls.
4128  *		- DKIOCINFO
4129  *		- DKIOCGMEDIAINFO
4130  *
4131  *	[ just like lofi(7D) and ramdisk(7D) ]
4132  *
4133  * Arguments:
4134  *	vdc	- soft state pointer for this instance of the device driver.
4135  *
4136  * Return Code:
4137  *	0	- Success
4138  */
4139 static int
4140 vdc_create_fake_geometry(vdc_t *vdc)
4141 {
4142 	int	rv = 0;
4143 
4144 	ASSERT(vdc != NULL);
4145 
4146 	/*
4147 	 * DKIOCINFO support
4148 	 */
4149 	vdc->cinfo = kmem_zalloc(sizeof (struct dk_cinfo), KM_SLEEP);
4150 
4151 	(void) strcpy(vdc->cinfo->dki_cname, VDC_DRIVER_NAME);
4152 	(void) strcpy(vdc->cinfo->dki_dname, VDC_DRIVER_NAME);
4153 	/* max_xfer_sz is #blocks so we don't need to divide by DEV_BSIZE */
4154 	vdc->cinfo->dki_maxtransfer = vdc->max_xfer_sz;
4155 	vdc->cinfo->dki_ctype = DKC_SCSI_CCS;
4156 	vdc->cinfo->dki_flags = DKI_FMTVOL;
4157 	vdc->cinfo->dki_cnum = 0;
4158 	vdc->cinfo->dki_addr = 0;
4159 	vdc->cinfo->dki_space = 0;
4160 	vdc->cinfo->dki_prio = 0;
4161 	vdc->cinfo->dki_vec = 0;
4162 	vdc->cinfo->dki_unit = vdc->instance;
4163 	vdc->cinfo->dki_slave = 0;
4164 	/*
4165 	 * The partition number will be created on the fly depending on the
4166 	 * actual slice (i.e. minor node) that is used to request the data.
4167 	 */
4168 	vdc->cinfo->dki_partition = 0;
4169 
4170 	/*
4171 	 * DKIOCGMEDIAINFO support
4172 	 */
4173 	if (vdc->minfo == NULL)
4174 		vdc->minfo = kmem_zalloc(sizeof (struct dk_minfo), KM_SLEEP);
4175 	vdc->minfo->dki_media_type = DK_FIXED_DISK;
4176 	vdc->minfo->dki_capacity = 1;
4177 	vdc->minfo->dki_lbsize = DEV_BSIZE;
4178 
4179 	return (rv);
4180 }
4181 
4182 /*
4183  * Function:
4184  *	vdc_setup_disk_layout()
4185  *
4186  * Description:
4187  *	This routine discovers all the necessary details about the "disk"
4188  *	by requesting the data that is available from the vDisk server and by
4189  *	faking up the rest of the data.
4190  *
4191  * Arguments:
4192  *	vdc	- soft state pointer for this instance of the device driver.
4193  *
4194  * Return Code:
4195  *	0	- Success
4196  */
4197 static int
4198 vdc_setup_disk_layout(vdc_t *vdc)
4199 {
4200 	buf_t	*buf;	/* BREAD requests need to be in a buf_t structure */
4201 	dev_t	dev;
4202 	int	slice = 0;
4203 	int	rv;
4204 
4205 	ASSERT(vdc != NULL);
4206 
4207 	rv = vdc_create_fake_geometry(vdc);
4208 	if (rv != 0) {
4209 		cmn_err(CE_NOTE, "[%d] Failed to create disk geometry (err%d)",
4210 				vdc->instance, rv);
4211 	}
4212 
4213 	if (vdc->vtoc == NULL)
4214 		vdc->vtoc = kmem_zalloc(sizeof (struct vtoc), KM_SLEEP);
4215 
4216 	dev = makedevice(ddi_driver_major(vdc->dip),
4217 				VD_MAKE_DEV(vdc->instance, 0));
4218 	rv = vd_process_ioctl(dev, DKIOCGVTOC, (caddr_t)vdc->vtoc, FKIOCTL);
4219 	if (rv) {
4220 		cmn_err(CE_NOTE, "[%d] Failed to get VTOC (err=%d)",
4221 				vdc->instance, rv);
4222 		return (rv);
4223 	}
4224 
4225 	/*
4226 	 * find the slice that represents the entire "disk" and use that to
4227 	 * read the disk label. The convention in Solaris is that slice 2
4228 	 * represents the whole disk so we check that it is, otherwise we
4229 	 * default to slice 0
4230 	 */
4231 	if ((vdc->vdisk_type == VD_DISK_TYPE_DISK) &&
4232 	    (vdc->vtoc->v_part[2].p_tag == V_BACKUP)) {
4233 		slice = 2;
4234 	} else {
4235 		slice = 0;
4236 	}
4237 
4238 	/*
4239 	 * Read disk label from start of disk
4240 	 */
4241 	vdc->label = kmem_zalloc(DK_LABEL_SIZE, KM_SLEEP);
4242 	buf = kmem_alloc(sizeof (buf_t), KM_SLEEP);
4243 	bioinit(buf);
4244 	buf->b_un.b_addr = (caddr_t)vdc->label;
4245 	buf->b_bcount = DK_LABEL_SIZE;
4246 	buf->b_flags = B_BUSY | B_READ;
4247 	buf->b_dev = dev;
4248 	rv = vdc_populate_descriptor(vdc, (caddr_t)buf, DK_LABEL_SIZE,
4249 			VD_OP_BREAD, 0, slice);
4250 	rv = biowait(buf);
4251 	biofini(buf);
4252 	kmem_free(buf, sizeof (buf_t));
4253 
4254 	return (rv);
4255 }
4256