xref: /titanic_51/usr/src/uts/sun4v/io/vdc.c (revision 8e50dcc9f00b393d43e6aa42b820bcbf1d3e1ce4)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #pragma ident	"%Z%%M%	%I%	%E% SMI"
28 
29 /*
30  * LDoms virtual disk client (vdc) device driver
31  *
32  * This driver runs on a guest logical domain and communicates with the virtual
33  * disk server (vds) driver running on the service domain which is exporting
34  * virtualized "disks" to the guest logical domain.
35  *
36  * The driver can be divided into four sections:
37  *
38  * 1) generic device driver housekeeping
39  *	_init, _fini, attach, detach, ops structures, etc.
40  *
41  * 2) communication channel setup
42  *	Setup the communications link over the LDC channel that vdc uses to
43  *	talk to the vDisk server. Initialise the descriptor ring which
44  *	allows the LDC clients to transfer data via memory mappings.
45  *
46  * 3) Support exported to upper layers (filesystems, etc)
47  *	The upper layers call into vdc via strategy(9E) and DKIO(7I)
48  *	ioctl calls. vdc will copy the data to be written to the descriptor
49  *	ring or maps the buffer to store the data read by the vDisk
50  *	server into the descriptor ring. It then sends a message to the
51  *	vDisk server requesting it to complete the operation.
52  *
53  * 4) Handling responses from vDisk server.
54  *	The vDisk server will ACK some or all of the messages vdc sends to it
55  *	(this is configured during the handshake). Upon receipt of an ACK
56  *	vdc will check the descriptor ring and signal to the upper layer
57  *	code waiting on the IO.
58  */
59 
60 #include <sys/atomic.h>
61 #include <sys/conf.h>
62 #include <sys/disp.h>
63 #include <sys/ddi.h>
64 #include <sys/dkio.h>
65 #include <sys/efi_partition.h>
66 #include <sys/fcntl.h>
67 #include <sys/file.h>
68 #include <sys/kstat.h>
69 #include <sys/mach_descrip.h>
70 #include <sys/modctl.h>
71 #include <sys/mdeg.h>
72 #include <sys/note.h>
73 #include <sys/open.h>
74 #include <sys/sdt.h>
75 #include <sys/stat.h>
76 #include <sys/sunddi.h>
77 #include <sys/types.h>
78 #include <sys/promif.h>
79 #include <sys/var.h>
80 #include <sys/vtoc.h>
81 #include <sys/archsystm.h>
82 #include <sys/sysmacros.h>
83 
84 #include <sys/cdio.h>
85 #include <sys/dktp/fdisk.h>
86 #include <sys/dktp/dadkio.h>
87 #include <sys/mhd.h>
88 #include <sys/scsi/generic/sense.h>
89 #include <sys/scsi/impl/uscsi.h>
90 #include <sys/scsi/impl/services.h>
91 #include <sys/scsi/targets/sddef.h>
92 
93 #include <sys/ldoms.h>
94 #include <sys/ldc.h>
95 #include <sys/vio_common.h>
96 #include <sys/vio_mailbox.h>
97 #include <sys/vio_util.h>
98 #include <sys/vdsk_common.h>
99 #include <sys/vdsk_mailbox.h>
100 #include <sys/vdc.h>
101 
102 /*
103  * function prototypes
104  */
105 
106 /* standard driver functions */
107 static int	vdc_open(dev_t *dev, int flag, int otyp, cred_t *cred);
108 static int	vdc_close(dev_t dev, int flag, int otyp, cred_t *cred);
109 static int	vdc_strategy(struct buf *buf);
110 static int	vdc_print(dev_t dev, char *str);
111 static int	vdc_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblk);
112 static int	vdc_read(dev_t dev, struct uio *uio, cred_t *cred);
113 static int	vdc_write(dev_t dev, struct uio *uio, cred_t *cred);
114 static int	vdc_ioctl(dev_t dev, int cmd, intptr_t arg, int mode,
115 			cred_t *credp, int *rvalp);
116 static int	vdc_aread(dev_t dev, struct aio_req *aio, cred_t *cred);
117 static int	vdc_awrite(dev_t dev, struct aio_req *aio, cred_t *cred);
118 
119 static int	vdc_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd,
120 			void *arg, void **resultp);
121 static int	vdc_attach(dev_info_t *dip, ddi_attach_cmd_t cmd);
122 static int	vdc_detach(dev_info_t *dip, ddi_detach_cmd_t cmd);
123 
124 /* setup */
125 static void	vdc_min(struct buf *bufp);
126 static int	vdc_send(vdc_t *vdc, caddr_t pkt, size_t *msglen);
127 static int	vdc_do_ldc_init(vdc_t *vdc, md_t *mdp, mde_cookie_t vd_node);
128 static int	vdc_start_ldc_connection(vdc_t *vdc);
129 static int	vdc_create_device_nodes(vdc_t *vdc);
130 static int	vdc_create_device_nodes_efi(vdc_t *vdc);
131 static int	vdc_create_device_nodes_vtoc(vdc_t *vdc);
132 static int	vdc_create_device_nodes_props(vdc_t *vdc);
133 static void	vdc_create_io_kstats(vdc_t *vdc);
134 static void	vdc_create_err_kstats(vdc_t *vdc);
135 static void	vdc_set_err_kstats(vdc_t *vdc);
136 static int	vdc_get_md_node(dev_info_t *dip, md_t **mdpp,
137 		    mde_cookie_t *vd_nodep, mde_cookie_t *vd_portp);
138 static int	vdc_get_ldc_id(md_t *, mde_cookie_t, uint64_t *);
139 static int	vdc_do_ldc_up(vdc_t *vdc);
140 static void	vdc_terminate_ldc(vdc_t *vdc);
141 static int	vdc_init_descriptor_ring(vdc_t *vdc);
142 static void	vdc_destroy_descriptor_ring(vdc_t *vdc);
143 static int	vdc_setup_devid(vdc_t *vdc);
144 static void	vdc_store_label_efi(vdc_t *, efi_gpt_t *, efi_gpe_t *);
145 static void	vdc_store_label_vtoc(vdc_t *, struct dk_geom *, struct vtoc *);
146 static void	vdc_store_label_unk(vdc_t *vdc);
147 static boolean_t vdc_is_opened(vdc_t *vdc);
148 
149 /* handshake with vds */
150 static int		vdc_init_ver_negotiation(vdc_t *vdc, vio_ver_t ver);
151 static int		vdc_ver_negotiation(vdc_t *vdcp);
152 static int		vdc_init_attr_negotiation(vdc_t *vdc);
153 static int		vdc_attr_negotiation(vdc_t *vdcp);
154 static int		vdc_init_dring_negotiate(vdc_t *vdc);
155 static int		vdc_dring_negotiation(vdc_t *vdcp);
156 static int		vdc_send_rdx(vdc_t *vdcp);
157 static int		vdc_rdx_exchange(vdc_t *vdcp);
158 static boolean_t	vdc_is_supported_version(vio_ver_msg_t *ver_msg);
159 
160 /* processing incoming messages from vDisk server */
161 static void	vdc_process_msg_thread(vdc_t *vdc);
162 static int	vdc_recv(vdc_t *vdc, vio_msg_t *msgp, size_t *nbytesp);
163 
164 static uint_t	vdc_handle_cb(uint64_t event, caddr_t arg);
165 static int	vdc_process_data_msg(vdc_t *vdc, vio_msg_t *msg);
166 static int	vdc_handle_ver_msg(vdc_t *vdc, vio_ver_msg_t *ver_msg);
167 static int	vdc_handle_attr_msg(vdc_t *vdc, vd_attr_msg_t *attr_msg);
168 static int	vdc_handle_dring_reg_msg(vdc_t *vdc, vio_dring_reg_msg_t *msg);
169 static int 	vdc_send_request(vdc_t *vdcp, int operation,
170 		    caddr_t addr, size_t nbytes, int slice, diskaddr_t offset,
171 		    int cb_type, void *cb_arg, vio_desc_direction_t dir);
172 static int	vdc_map_to_shared_dring(vdc_t *vdcp, int idx);
173 static int 	vdc_populate_descriptor(vdc_t *vdcp, int operation,
174 		    caddr_t addr, size_t nbytes, int slice, diskaddr_t offset,
175 		    int cb_type, void *cb_arg, vio_desc_direction_t dir);
176 static int 	vdc_do_sync_op(vdc_t *vdcp, int operation, caddr_t addr,
177 		    size_t nbytes, int slice, diskaddr_t offset, int cb_type,
178 		    void *cb_arg, vio_desc_direction_t dir, boolean_t);
179 
180 static int	vdc_wait_for_response(vdc_t *vdcp, vio_msg_t *msgp);
181 static int	vdc_drain_response(vdc_t *vdcp);
182 static int	vdc_depopulate_descriptor(vdc_t *vdc, uint_t idx);
183 static int	vdc_populate_mem_hdl(vdc_t *vdcp, vdc_local_desc_t *ldep);
184 static int	vdc_verify_seq_num(vdc_t *vdc, vio_dring_msg_t *dring_msg);
185 
186 /* dkio */
187 static int	vd_process_ioctl(dev_t dev, int cmd, caddr_t arg, int mode,
188 		    int *rvalp);
189 static int	vd_process_efi_ioctl(void *vdisk, int cmd, uintptr_t arg);
190 static void	vdc_create_fake_geometry(vdc_t *vdc);
191 static int	vdc_validate_geometry(vdc_t *vdc);
192 static void	vdc_validate(vdc_t *vdc);
193 static void	vdc_validate_task(void *arg);
194 static int	vdc_null_copy_func(vdc_t *vdc, void *from, void *to,
195 		    int mode, int dir);
196 static int	vdc_get_wce_convert(vdc_t *vdc, void *from, void *to,
197 		    int mode, int dir);
198 static int	vdc_set_wce_convert(vdc_t *vdc, void *from, void *to,
199 		    int mode, int dir);
200 static int	vdc_get_vtoc_convert(vdc_t *vdc, void *from, void *to,
201 		    int mode, int dir);
202 static int	vdc_set_vtoc_convert(vdc_t *vdc, void *from, void *to,
203 		    int mode, int dir);
204 static int	vdc_get_geom_convert(vdc_t *vdc, void *from, void *to,
205 		    int mode, int dir);
206 static int	vdc_set_geom_convert(vdc_t *vdc, void *from, void *to,
207 		    int mode, int dir);
208 static int	vdc_get_efi_convert(vdc_t *vdc, void *from, void *to,
209 		    int mode, int dir);
210 static int	vdc_set_efi_convert(vdc_t *vdc, void *from, void *to,
211 		    int mode, int dir);
212 
213 static void 	vdc_ownership_update(vdc_t *vdc, int ownership_flags);
214 static int	vdc_access_set(vdc_t *vdc, uint64_t flags, int mode);
215 static vdc_io_t	*vdc_failfast_io_queue(vdc_t *vdc, struct buf *buf);
216 static int	vdc_failfast_check_resv(vdc_t *vdc);
217 
218 /*
219  * Module variables
220  */
221 
222 /*
223  * Tunable variables to control how long vdc waits before timing out on
224  * various operations
225  */
226 static int	vdc_hshake_retries = 3;
227 
228 static int	vdc_timeout = 0; /* units: seconds */
229 
230 static uint64_t vdc_hz_min_ldc_delay;
231 static uint64_t vdc_min_timeout_ldc = 1 * MILLISEC;
232 static uint64_t vdc_hz_max_ldc_delay;
233 static uint64_t vdc_max_timeout_ldc = 100 * MILLISEC;
234 
235 static uint64_t vdc_ldc_read_init_delay = 1 * MILLISEC;
236 static uint64_t vdc_ldc_read_max_delay = 100 * MILLISEC;
237 
238 /* values for dumping - need to run in a tighter loop */
239 static uint64_t	vdc_usec_timeout_dump = 100 * MILLISEC;	/* 0.1s units: ns */
240 static int	vdc_dump_retries = 100;
241 
242 static uint16_t	vdc_scsi_timeout = 60;	/* 60s units: seconds  */
243 
244 static uint64_t vdc_ownership_delay = 6 * MICROSEC; /* 6s units: usec */
245 
246 /* Count of the number of vdc instances attached */
247 static volatile uint32_t	vdc_instance_count = 0;
248 
249 /* Tunable to log all SCSI errors */
250 static boolean_t vdc_scsi_log_error = B_FALSE;
251 
252 /* Soft state pointer */
253 static void	*vdc_state;
254 
255 /*
256  * Controlling the verbosity of the error/debug messages
257  *
258  * vdc_msglevel - controls level of messages
259  * vdc_matchinst - 64-bit variable where each bit corresponds
260  *                 to the vdc instance the vdc_msglevel applies.
261  */
262 int		vdc_msglevel = 0x0;
263 uint64_t	vdc_matchinst = 0ull;
264 
265 /*
266  * Supported vDisk protocol version pairs.
267  *
268  * The first array entry is the latest and preferred version.
269  */
270 static const vio_ver_t	vdc_version[] = {{1, 1}};
271 
272 static struct cb_ops vdc_cb_ops = {
273 	vdc_open,	/* cb_open */
274 	vdc_close,	/* cb_close */
275 	vdc_strategy,	/* cb_strategy */
276 	vdc_print,	/* cb_print */
277 	vdc_dump,	/* cb_dump */
278 	vdc_read,	/* cb_read */
279 	vdc_write,	/* cb_write */
280 	vdc_ioctl,	/* cb_ioctl */
281 	nodev,		/* cb_devmap */
282 	nodev,		/* cb_mmap */
283 	nodev,		/* cb_segmap */
284 	nochpoll,	/* cb_chpoll */
285 	ddi_prop_op,	/* cb_prop_op */
286 	NULL,		/* cb_str */
287 	D_MP | D_64BIT,	/* cb_flag */
288 	CB_REV,		/* cb_rev */
289 	vdc_aread,	/* cb_aread */
290 	vdc_awrite	/* cb_awrite */
291 };
292 
293 static struct dev_ops vdc_ops = {
294 	DEVO_REV,	/* devo_rev */
295 	0,		/* devo_refcnt */
296 	vdc_getinfo,	/* devo_getinfo */
297 	nulldev,	/* devo_identify */
298 	nulldev,	/* devo_probe */
299 	vdc_attach,	/* devo_attach */
300 	vdc_detach,	/* devo_detach */
301 	nodev,		/* devo_reset */
302 	&vdc_cb_ops,	/* devo_cb_ops */
303 	NULL,		/* devo_bus_ops */
304 	nulldev		/* devo_power */
305 };
306 
307 static struct modldrv modldrv = {
308 	&mod_driverops,
309 	"virtual disk client",
310 	&vdc_ops,
311 };
312 
313 static struct modlinkage modlinkage = {
314 	MODREV_1,
315 	&modldrv,
316 	NULL
317 };
318 
319 /* -------------------------------------------------------------------------- */
320 
321 /*
322  * Device Driver housekeeping and setup
323  */
324 
325 int
326 _init(void)
327 {
328 	int	status;
329 
330 	if ((status = ddi_soft_state_init(&vdc_state, sizeof (vdc_t), 1)) != 0)
331 		return (status);
332 	if ((status = mod_install(&modlinkage)) != 0)
333 		ddi_soft_state_fini(&vdc_state);
334 	return (status);
335 }
336 
337 int
338 _info(struct modinfo *modinfop)
339 {
340 	return (mod_info(&modlinkage, modinfop));
341 }
342 
343 int
344 _fini(void)
345 {
346 	int	status;
347 
348 	if ((status = mod_remove(&modlinkage)) != 0)
349 		return (status);
350 	ddi_soft_state_fini(&vdc_state);
351 	return (0);
352 }
353 
354 static int
355 vdc_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd,  void *arg, void **resultp)
356 {
357 	_NOTE(ARGUNUSED(dip))
358 
359 	int	instance = VDCUNIT((dev_t)arg);
360 	vdc_t	*vdc = NULL;
361 
362 	switch (cmd) {
363 	case DDI_INFO_DEVT2DEVINFO:
364 		if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
365 			*resultp = NULL;
366 			return (DDI_FAILURE);
367 		}
368 		*resultp = vdc->dip;
369 		return (DDI_SUCCESS);
370 	case DDI_INFO_DEVT2INSTANCE:
371 		*resultp = (void *)(uintptr_t)instance;
372 		return (DDI_SUCCESS);
373 	default:
374 		*resultp = NULL;
375 		return (DDI_FAILURE);
376 	}
377 }
378 
379 static int
380 vdc_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
381 {
382 	kt_did_t failfast_tid, ownership_tid;
383 	int	instance;
384 	int	rv;
385 	vdc_t	*vdc = NULL;
386 
387 	switch (cmd) {
388 	case DDI_DETACH:
389 		/* the real work happens below */
390 		break;
391 	case DDI_SUSPEND:
392 		/* nothing to do for this non-device */
393 		return (DDI_SUCCESS);
394 	default:
395 		return (DDI_FAILURE);
396 	}
397 
398 	ASSERT(cmd == DDI_DETACH);
399 	instance = ddi_get_instance(dip);
400 	DMSGX(1, "[%d] Entered\n", instance);
401 
402 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
403 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
404 		return (DDI_FAILURE);
405 	}
406 
407 	/*
408 	 * This function is called when vdc is detached or if it has failed to
409 	 * attach. In that case, the attach may have fail before the vdisk type
410 	 * has been set so we can't call vdc_is_opened(). However as the attach
411 	 * has failed, we know that the vdisk is not opened and we can safely
412 	 * detach.
413 	 */
414 	if (vdc->vdisk_type != VD_DISK_TYPE_UNK && vdc_is_opened(vdc)) {
415 		DMSG(vdc, 0, "[%d] Cannot detach: device is open", instance);
416 		return (DDI_FAILURE);
417 	}
418 
419 	if (vdc->dkio_flush_pending) {
420 		DMSG(vdc, 0,
421 		    "[%d] Cannot detach: %d outstanding DKIO flushes\n",
422 		    instance, vdc->dkio_flush_pending);
423 		return (DDI_FAILURE);
424 	}
425 
426 	if (vdc->validate_pending) {
427 		DMSG(vdc, 0,
428 		    "[%d] Cannot detach: %d outstanding validate request\n",
429 		    instance, vdc->validate_pending);
430 		return (DDI_FAILURE);
431 	}
432 
433 	DMSG(vdc, 0, "[%d] proceeding...\n", instance);
434 
435 	/* If we took ownership, release ownership */
436 	mutex_enter(&vdc->ownership_lock);
437 	if (vdc->ownership & VDC_OWNERSHIP_GRANTED) {
438 		rv = vdc_access_set(vdc, VD_ACCESS_SET_CLEAR, FKIOCTL);
439 		if (rv == 0) {
440 			vdc_ownership_update(vdc, VDC_OWNERSHIP_NONE);
441 		}
442 	}
443 	mutex_exit(&vdc->ownership_lock);
444 
445 	/* mark instance as detaching */
446 	vdc->lifecycle	= VDC_LC_DETACHING;
447 
448 	/*
449 	 * try and disable callbacks to prevent another handshake
450 	 */
451 	rv = ldc_set_cb_mode(vdc->ldc_handle, LDC_CB_DISABLE);
452 	DMSG(vdc, 0, "callback disabled (rv=%d)\n", rv);
453 
454 	if (vdc->initialized & VDC_THREAD) {
455 		mutex_enter(&vdc->read_lock);
456 		if ((vdc->read_state == VDC_READ_WAITING) ||
457 		    (vdc->read_state == VDC_READ_RESET)) {
458 			vdc->read_state = VDC_READ_RESET;
459 			cv_signal(&vdc->read_cv);
460 		}
461 
462 		mutex_exit(&vdc->read_lock);
463 
464 		/* wake up any thread waiting for connection to come online */
465 		mutex_enter(&vdc->lock);
466 		if (vdc->state == VDC_STATE_INIT_WAITING) {
467 			DMSG(vdc, 0,
468 			    "[%d] write reset - move to resetting state...\n",
469 			    instance);
470 			vdc->state = VDC_STATE_RESETTING;
471 			cv_signal(&vdc->initwait_cv);
472 		}
473 		mutex_exit(&vdc->lock);
474 
475 		/* now wait until state transitions to VDC_STATE_DETACH */
476 		thread_join(vdc->msg_proc_thr->t_did);
477 		ASSERT(vdc->state == VDC_STATE_DETACH);
478 		DMSG(vdc, 0, "[%d] Reset thread exit and join ..\n",
479 		    vdc->instance);
480 	}
481 
482 	mutex_enter(&vdc->lock);
483 
484 	if (vdc->initialized & VDC_DRING)
485 		vdc_destroy_descriptor_ring(vdc);
486 
487 	if (vdc->initialized & VDC_LDC)
488 		vdc_terminate_ldc(vdc);
489 
490 	if (vdc->failfast_thread) {
491 		failfast_tid = vdc->failfast_thread->t_did;
492 		vdc->failfast_interval = 0;
493 		cv_signal(&vdc->failfast_cv);
494 	} else {
495 		failfast_tid = 0;
496 	}
497 
498 	if (vdc->ownership & VDC_OWNERSHIP_WANTED) {
499 		ownership_tid = vdc->ownership_thread->t_did;
500 		vdc->ownership = VDC_OWNERSHIP_NONE;
501 		cv_signal(&vdc->ownership_cv);
502 	} else {
503 		ownership_tid = 0;
504 	}
505 
506 	mutex_exit(&vdc->lock);
507 
508 	if (failfast_tid != 0)
509 		thread_join(failfast_tid);
510 
511 	if (ownership_tid != 0)
512 		thread_join(ownership_tid);
513 
514 	if (vdc->initialized & VDC_MINOR) {
515 		ddi_prop_remove_all(dip);
516 		ddi_remove_minor_node(dip, NULL);
517 	}
518 
519 	if (vdc->io_stats) {
520 		kstat_delete(vdc->io_stats);
521 		vdc->io_stats = NULL;
522 	}
523 
524 	if (vdc->err_stats) {
525 		kstat_delete(vdc->err_stats);
526 		vdc->err_stats = NULL;
527 	}
528 
529 	if (vdc->initialized & VDC_LOCKS) {
530 		mutex_destroy(&vdc->lock);
531 		mutex_destroy(&vdc->read_lock);
532 		mutex_destroy(&vdc->ownership_lock);
533 		cv_destroy(&vdc->initwait_cv);
534 		cv_destroy(&vdc->dring_free_cv);
535 		cv_destroy(&vdc->membind_cv);
536 		cv_destroy(&vdc->sync_pending_cv);
537 		cv_destroy(&vdc->sync_blocked_cv);
538 		cv_destroy(&vdc->read_cv);
539 		cv_destroy(&vdc->running_cv);
540 		cv_destroy(&vdc->ownership_cv);
541 		cv_destroy(&vdc->failfast_cv);
542 		cv_destroy(&vdc->failfast_io_cv);
543 	}
544 
545 	if (vdc->minfo)
546 		kmem_free(vdc->minfo, sizeof (struct dk_minfo));
547 
548 	if (vdc->cinfo)
549 		kmem_free(vdc->cinfo, sizeof (struct dk_cinfo));
550 
551 	if (vdc->vtoc)
552 		kmem_free(vdc->vtoc, sizeof (struct vtoc));
553 
554 	if (vdc->geom)
555 		kmem_free(vdc->geom, sizeof (struct dk_geom));
556 
557 	if (vdc->devid) {
558 		ddi_devid_unregister(dip);
559 		ddi_devid_free(vdc->devid);
560 	}
561 
562 	if (vdc->initialized & VDC_SOFT_STATE)
563 		ddi_soft_state_free(vdc_state, instance);
564 
565 	DMSG(vdc, 0, "[%d] End %p\n", instance, (void *)vdc);
566 
567 	return (DDI_SUCCESS);
568 }
569 
570 
571 static int
572 vdc_do_attach(dev_info_t *dip)
573 {
574 	int		instance;
575 	vdc_t		*vdc = NULL;
576 	int		status;
577 	md_t		*mdp;
578 	mde_cookie_t	vd_node, vd_port;
579 
580 	ASSERT(dip != NULL);
581 
582 	instance = ddi_get_instance(dip);
583 	if (ddi_soft_state_zalloc(vdc_state, instance) != DDI_SUCCESS) {
584 		cmn_err(CE_NOTE, "[%d] Couldn't alloc state structure",
585 		    instance);
586 		return (DDI_FAILURE);
587 	}
588 
589 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
590 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
591 		return (DDI_FAILURE);
592 	}
593 
594 	/*
595 	 * We assign the value to initialized in this case to zero out the
596 	 * variable and then set bits in it to indicate what has been done
597 	 */
598 	vdc->initialized = VDC_SOFT_STATE;
599 
600 	vdc_hz_min_ldc_delay = drv_usectohz(vdc_min_timeout_ldc);
601 	vdc_hz_max_ldc_delay = drv_usectohz(vdc_max_timeout_ldc);
602 
603 	vdc->dip	= dip;
604 	vdc->instance	= instance;
605 	vdc->vdisk_type	= VD_DISK_TYPE_UNK;
606 	vdc->vdisk_label = VD_DISK_LABEL_UNK;
607 	vdc->state	= VDC_STATE_INIT;
608 	vdc->lifecycle	= VDC_LC_ATTACHING;
609 	vdc->ldc_state	= 0;
610 	vdc->session_id = 0;
611 	vdc->block_size = DEV_BSIZE;
612 	vdc->max_xfer_sz = maxphys / DEV_BSIZE;
613 
614 	/*
615 	 * We assume, for now, that the vDisk server will export 'read'
616 	 * operations to us at a minimum (this is needed because of checks
617 	 * in vdc for supported operations early in the handshake process).
618 	 * The vDisk server will return ENOTSUP if this is not the case.
619 	 * The value will be overwritten during the attribute exchange with
620 	 * the bitmask of operations exported by server.
621 	 */
622 	vdc->operations = VD_OP_MASK_READ;
623 
624 	vdc->vtoc = NULL;
625 	vdc->geom = NULL;
626 	vdc->cinfo = NULL;
627 	vdc->minfo = NULL;
628 
629 	mutex_init(&vdc->lock, NULL, MUTEX_DRIVER, NULL);
630 	cv_init(&vdc->initwait_cv, NULL, CV_DRIVER, NULL);
631 	cv_init(&vdc->dring_free_cv, NULL, CV_DRIVER, NULL);
632 	cv_init(&vdc->membind_cv, NULL, CV_DRIVER, NULL);
633 	cv_init(&vdc->running_cv, NULL, CV_DRIVER, NULL);
634 
635 	vdc->threads_pending = 0;
636 	vdc->sync_op_pending = B_FALSE;
637 	vdc->sync_op_blocked = B_FALSE;
638 	cv_init(&vdc->sync_pending_cv, NULL, CV_DRIVER, NULL);
639 	cv_init(&vdc->sync_blocked_cv, NULL, CV_DRIVER, NULL);
640 
641 	mutex_init(&vdc->ownership_lock, NULL, MUTEX_DRIVER, NULL);
642 	cv_init(&vdc->ownership_cv, NULL, CV_DRIVER, NULL);
643 	cv_init(&vdc->failfast_cv, NULL, CV_DRIVER, NULL);
644 	cv_init(&vdc->failfast_io_cv, NULL, CV_DRIVER, NULL);
645 
646 	/* init blocking msg read functionality */
647 	mutex_init(&vdc->read_lock, NULL, MUTEX_DRIVER, NULL);
648 	cv_init(&vdc->read_cv, NULL, CV_DRIVER, NULL);
649 	vdc->read_state = VDC_READ_IDLE;
650 
651 	vdc->initialized |= VDC_LOCKS;
652 
653 	/* get device and port MD node for this disk instance */
654 	if (vdc_get_md_node(dip, &mdp, &vd_node, &vd_port) != 0) {
655 		cmn_err(CE_NOTE, "[%d] Could not get machine description node",
656 		    instance);
657 		return (DDI_FAILURE);
658 	}
659 
660 	/* set the connection timeout */
661 	if (vd_port == NULL || (md_get_prop_val(mdp, vd_port,
662 	    VDC_MD_TIMEOUT, &vdc->ctimeout) != 0)) {
663 		vdc->ctimeout = 0;
664 	}
665 
666 	/* initialise LDC channel which will be used to communicate with vds */
667 	status = vdc_do_ldc_init(vdc, mdp, vd_node);
668 
669 	(void) md_fini_handle(mdp);
670 
671 	if (status != 0) {
672 		cmn_err(CE_NOTE, "[%d] Couldn't initialize LDC", instance);
673 		goto return_status;
674 	}
675 
676 	/* initialize the thread responsible for managing state with server */
677 	vdc->msg_proc_thr = thread_create(NULL, 0, vdc_process_msg_thread,
678 	    vdc, 0, &p0, TS_RUN, minclsyspri);
679 	if (vdc->msg_proc_thr == NULL) {
680 		cmn_err(CE_NOTE, "[%d] Failed to create msg processing thread",
681 		    instance);
682 		return (DDI_FAILURE);
683 	}
684 
685 	vdc->initialized |= VDC_THREAD;
686 
687 	/* Create the kstats for saving the I/O statistics used by iostat(1M) */
688 	vdc_create_io_kstats(vdc);
689 	vdc_create_err_kstats(vdc);
690 
691 	atomic_inc_32(&vdc_instance_count);
692 
693 	/*
694 	 * Check the disk label. This will send requests and do the handshake.
695 	 * We don't really care about the disk label now. What we really need is
696 	 * the handshake do be done so that we know the type of the disk (slice
697 	 * or full disk) and the appropriate device nodes can be created.
698 	 */
699 	vdc->vdisk_label = VD_DISK_LABEL_UNK;
700 	vdc->vtoc = kmem_zalloc(sizeof (struct vtoc), KM_SLEEP);
701 	vdc->geom = kmem_zalloc(sizeof (struct dk_geom), KM_SLEEP);
702 	vdc->minfo = kmem_zalloc(sizeof (struct dk_minfo), KM_SLEEP);
703 
704 	mutex_enter(&vdc->lock);
705 	(void) vdc_validate_geometry(vdc);
706 	mutex_exit(&vdc->lock);
707 
708 	/*
709 	 * Now that we have the device info we can create the
710 	 * device nodes and properties
711 	 */
712 	status = vdc_create_device_nodes(vdc);
713 	if (status) {
714 		DMSG(vdc, 0, "[%d] Failed to create device nodes",
715 		    instance);
716 		goto return_status;
717 	}
718 	status = vdc_create_device_nodes_props(vdc);
719 	if (status) {
720 		DMSG(vdc, 0, "[%d] Failed to create device nodes"
721 		    " properties (%d)", instance, status);
722 		goto return_status;
723 	}
724 
725 	/*
726 	 * Setup devid
727 	 */
728 	if (vdc_setup_devid(vdc)) {
729 		DMSG(vdc, 0, "[%d] No device id available\n", instance);
730 	}
731 
732 	/*
733 	 * Fill in the fields of the error statistics kstat that were not
734 	 * available when creating the kstat
735 	 */
736 	vdc_set_err_kstats(vdc);
737 
738 	ddi_report_dev(dip);
739 	vdc->lifecycle	= VDC_LC_ONLINE;
740 	DMSG(vdc, 0, "[%d] Attach tasks successful\n", instance);
741 
742 return_status:
743 	DMSG(vdc, 0, "[%d] Attach completed\n", instance);
744 	return (status);
745 }
746 
747 static int
748 vdc_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
749 {
750 	int	status;
751 
752 	switch (cmd) {
753 	case DDI_ATTACH:
754 		if ((status = vdc_do_attach(dip)) != 0)
755 			(void) vdc_detach(dip, DDI_DETACH);
756 		return (status);
757 	case DDI_RESUME:
758 		/* nothing to do for this non-device */
759 		return (DDI_SUCCESS);
760 	default:
761 		return (DDI_FAILURE);
762 	}
763 }
764 
765 static int
766 vdc_do_ldc_init(vdc_t *vdc, md_t *mdp, mde_cookie_t vd_node)
767 {
768 	int			status = 0;
769 	ldc_status_t		ldc_state;
770 	ldc_attr_t		ldc_attr;
771 	uint64_t		ldc_id = 0;
772 
773 	ASSERT(vdc != NULL);
774 
775 	vdc->initialized |= VDC_LDC;
776 
777 	if ((status = vdc_get_ldc_id(mdp, vd_node, &ldc_id)) != 0) {
778 		DMSG(vdc, 0, "[%d] Failed to get LDC channel ID property",
779 		    vdc->instance);
780 		return (EIO);
781 	}
782 
783 	DMSGX(0, "[%d] LDC id is 0x%lx\n", vdc->instance, ldc_id);
784 
785 	vdc->ldc_id = ldc_id;
786 
787 	ldc_attr.devclass = LDC_DEV_BLK;
788 	ldc_attr.instance = vdc->instance;
789 	ldc_attr.mode = LDC_MODE_UNRELIABLE;	/* unreliable transport */
790 	ldc_attr.mtu = VD_LDC_MTU;
791 
792 	if ((vdc->initialized & VDC_LDC_INIT) == 0) {
793 		status = ldc_init(ldc_id, &ldc_attr, &vdc->ldc_handle);
794 		if (status != 0) {
795 			DMSG(vdc, 0, "[%d] ldc_init(chan %ld) returned %d",
796 			    vdc->instance, ldc_id, status);
797 			return (status);
798 		}
799 		vdc->initialized |= VDC_LDC_INIT;
800 	}
801 	status = ldc_status(vdc->ldc_handle, &ldc_state);
802 	if (status != 0) {
803 		DMSG(vdc, 0, "[%d] Cannot discover LDC status [err=%d]",
804 		    vdc->instance, status);
805 		return (status);
806 	}
807 	vdc->ldc_state = ldc_state;
808 
809 	if ((vdc->initialized & VDC_LDC_CB) == 0) {
810 		status = ldc_reg_callback(vdc->ldc_handle, vdc_handle_cb,
811 		    (caddr_t)vdc);
812 		if (status != 0) {
813 			DMSG(vdc, 0, "[%d] LDC callback reg. failed (%d)",
814 			    vdc->instance, status);
815 			return (status);
816 		}
817 		vdc->initialized |= VDC_LDC_CB;
818 	}
819 
820 	vdc->initialized |= VDC_LDC;
821 
822 	/*
823 	 * At this stage we have initialised LDC, we will now try and open
824 	 * the connection.
825 	 */
826 	if (vdc->ldc_state == LDC_INIT) {
827 		status = ldc_open(vdc->ldc_handle);
828 		if (status != 0) {
829 			DMSG(vdc, 0, "[%d] ldc_open(chan %ld) returned %d",
830 			    vdc->instance, vdc->ldc_id, status);
831 			return (status);
832 		}
833 		vdc->initialized |= VDC_LDC_OPEN;
834 	}
835 
836 	return (status);
837 }
838 
839 static int
840 vdc_start_ldc_connection(vdc_t *vdc)
841 {
842 	int		status = 0;
843 
844 	ASSERT(vdc != NULL);
845 
846 	ASSERT(MUTEX_HELD(&vdc->lock));
847 
848 	status = vdc_do_ldc_up(vdc);
849 
850 	DMSG(vdc, 0, "[%d] Finished bringing up LDC\n", vdc->instance);
851 
852 	return (status);
853 }
854 
855 static int
856 vdc_stop_ldc_connection(vdc_t *vdcp)
857 {
858 	int	status;
859 
860 	DMSG(vdcp, 0, ": Resetting connection to vDisk server : state %d\n",
861 	    vdcp->state);
862 
863 	status = ldc_down(vdcp->ldc_handle);
864 	DMSG(vdcp, 0, "ldc_down() = %d\n", status);
865 
866 	vdcp->initialized &= ~VDC_HANDSHAKE;
867 	DMSG(vdcp, 0, "initialized=%x\n", vdcp->initialized);
868 
869 	return (status);
870 }
871 
872 static void
873 vdc_create_io_kstats(vdc_t *vdc)
874 {
875 	if (vdc->io_stats != NULL) {
876 		DMSG(vdc, 0, "[%d] I/O kstat already exists\n", vdc->instance);
877 		return;
878 	}
879 
880 	vdc->io_stats = kstat_create(VDC_DRIVER_NAME, vdc->instance, NULL,
881 	    "disk", KSTAT_TYPE_IO, 1, KSTAT_FLAG_PERSISTENT);
882 	if (vdc->io_stats != NULL) {
883 		vdc->io_stats->ks_lock = &vdc->lock;
884 		kstat_install(vdc->io_stats);
885 	} else {
886 		cmn_err(CE_NOTE, "[%d] Failed to create kstat: I/O statistics"
887 		    " will not be gathered", vdc->instance);
888 	}
889 }
890 
891 static void
892 vdc_create_err_kstats(vdc_t *vdc)
893 {
894 	vd_err_stats_t	*stp;
895 	char	kstatmodule_err[KSTAT_STRLEN];
896 	char	kstatname[KSTAT_STRLEN];
897 	int	ndata = (sizeof (vd_err_stats_t) / sizeof (kstat_named_t));
898 	int	instance = vdc->instance;
899 
900 	if (vdc->err_stats != NULL) {
901 		DMSG(vdc, 0, "[%d] ERR kstat already exists\n", vdc->instance);
902 		return;
903 	}
904 
905 	(void) snprintf(kstatmodule_err, sizeof (kstatmodule_err),
906 	    "%serr", VDC_DRIVER_NAME);
907 	(void) snprintf(kstatname, sizeof (kstatname),
908 	    "%s%d,err", VDC_DRIVER_NAME, instance);
909 
910 	vdc->err_stats = kstat_create(kstatmodule_err, instance, kstatname,
911 	    "device_error", KSTAT_TYPE_NAMED, ndata, KSTAT_FLAG_PERSISTENT);
912 
913 	if (vdc->err_stats == NULL) {
914 		cmn_err(CE_NOTE, "[%d] Failed to create kstat: Error statistics"
915 		    " will not be gathered", instance);
916 		return;
917 	}
918 
919 	stp = (vd_err_stats_t *)vdc->err_stats->ks_data;
920 	kstat_named_init(&stp->vd_softerrs,	"Soft Errors",
921 	    KSTAT_DATA_UINT32);
922 	kstat_named_init(&stp->vd_transerrs,	"Transport Errors",
923 	    KSTAT_DATA_UINT32);
924 	kstat_named_init(&stp->vd_protoerrs,	"Protocol Errors",
925 	    KSTAT_DATA_UINT32);
926 	kstat_named_init(&stp->vd_vid,		"Vendor",
927 	    KSTAT_DATA_CHAR);
928 	kstat_named_init(&stp->vd_pid,		"Product",
929 	    KSTAT_DATA_CHAR);
930 	kstat_named_init(&stp->vd_capacity,	"Size",
931 	    KSTAT_DATA_ULONGLONG);
932 
933 	vdc->err_stats->ks_update  = nulldev;
934 
935 	kstat_install(vdc->err_stats);
936 }
937 
938 static void
939 vdc_set_err_kstats(vdc_t *vdc)
940 {
941 	vd_err_stats_t  *stp;
942 
943 	if (vdc->err_stats == NULL)
944 		return;
945 
946 	mutex_enter(&vdc->lock);
947 
948 	stp = (vd_err_stats_t *)vdc->err_stats->ks_data;
949 	ASSERT(stp != NULL);
950 
951 	stp->vd_capacity.value.ui64 = vdc->vdisk_size * vdc->block_size;
952 	(void) strcpy(stp->vd_vid.value.c, "SUN");
953 	(void) strcpy(stp->vd_pid.value.c, "VDSK");
954 
955 	mutex_exit(&vdc->lock);
956 }
957 
958 static int
959 vdc_create_device_nodes_efi(vdc_t *vdc)
960 {
961 	ddi_remove_minor_node(vdc->dip, "h");
962 	ddi_remove_minor_node(vdc->dip, "h,raw");
963 
964 	if (ddi_create_minor_node(vdc->dip, "wd", S_IFBLK,
965 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
966 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
967 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'wd'",
968 		    vdc->instance);
969 		return (EIO);
970 	}
971 
972 	/* if any device node is created we set this flag */
973 	vdc->initialized |= VDC_MINOR;
974 
975 	if (ddi_create_minor_node(vdc->dip, "wd,raw", S_IFCHR,
976 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
977 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
978 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'wd,raw'",
979 		    vdc->instance);
980 		return (EIO);
981 	}
982 
983 	return (0);
984 }
985 
986 static int
987 vdc_create_device_nodes_vtoc(vdc_t *vdc)
988 {
989 	ddi_remove_minor_node(vdc->dip, "wd");
990 	ddi_remove_minor_node(vdc->dip, "wd,raw");
991 
992 	if (ddi_create_minor_node(vdc->dip, "h", S_IFBLK,
993 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
994 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
995 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'h'",
996 		    vdc->instance);
997 		return (EIO);
998 	}
999 
1000 	/* if any device node is created we set this flag */
1001 	vdc->initialized |= VDC_MINOR;
1002 
1003 	if (ddi_create_minor_node(vdc->dip, "h,raw", S_IFCHR,
1004 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
1005 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
1006 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'h,raw'",
1007 		    vdc->instance);
1008 		return (EIO);
1009 	}
1010 
1011 	return (0);
1012 }
1013 
1014 /*
1015  * Function:
1016  *	vdc_create_device_nodes
1017  *
1018  * Description:
1019  *	This function creates the block and character device nodes under
1020  *	/devices along with the node properties. It is called as part of
1021  *	the attach(9E) of the instance during the handshake with vds after
1022  *	vds has sent the attributes to vdc.
1023  *
1024  *	If the device is of type VD_DISK_TYPE_SLICE then the minor node
1025  *	of 2 is used in keeping with the Solaris convention that slice 2
1026  *	refers to a whole disk. Slices start at 'a'
1027  *
1028  * Parameters:
1029  *	vdc 		- soft state pointer
1030  *
1031  * Return Values
1032  *	0		- Success
1033  *	EIO		- Failed to create node
1034  *	EINVAL		- Unknown type of disk exported
1035  */
1036 static int
1037 vdc_create_device_nodes(vdc_t *vdc)
1038 {
1039 	char		name[sizeof ("s,raw")];
1040 	dev_info_t	*dip = NULL;
1041 	int		instance, status;
1042 	int		num_slices = 1;
1043 	int		i;
1044 
1045 	ASSERT(vdc != NULL);
1046 
1047 	instance = vdc->instance;
1048 	dip = vdc->dip;
1049 
1050 	switch (vdc->vdisk_type) {
1051 	case VD_DISK_TYPE_DISK:
1052 		num_slices = V_NUMPAR;
1053 		break;
1054 	case VD_DISK_TYPE_SLICE:
1055 		num_slices = 1;
1056 		break;
1057 	case VD_DISK_TYPE_UNK:
1058 	default:
1059 		return (EINVAL);
1060 	}
1061 
1062 	/*
1063 	 * Minor nodes are different for EFI disks: EFI disks do not have
1064 	 * a minor node 'g' for the minor number corresponding to slice
1065 	 * VD_EFI_WD_SLICE (slice 7) instead they have a minor node 'wd'
1066 	 * representing the whole disk.
1067 	 */
1068 	for (i = 0; i < num_slices; i++) {
1069 
1070 		if (i == VD_EFI_WD_SLICE) {
1071 			if (vdc->vdisk_label == VD_DISK_LABEL_EFI)
1072 				status = vdc_create_device_nodes_efi(vdc);
1073 			else
1074 				status = vdc_create_device_nodes_vtoc(vdc);
1075 			if (status != 0)
1076 				return (status);
1077 			continue;
1078 		}
1079 
1080 		(void) snprintf(name, sizeof (name), "%c", 'a' + i);
1081 		if (ddi_create_minor_node(dip, name, S_IFBLK,
1082 		    VD_MAKE_DEV(instance, i), DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
1083 			cmn_err(CE_NOTE, "[%d] Couldn't add block node '%s'",
1084 			    instance, name);
1085 			return (EIO);
1086 		}
1087 
1088 		/* if any device node is created we set this flag */
1089 		vdc->initialized |= VDC_MINOR;
1090 
1091 		(void) snprintf(name, sizeof (name), "%c%s", 'a' + i, ",raw");
1092 
1093 		if (ddi_create_minor_node(dip, name, S_IFCHR,
1094 		    VD_MAKE_DEV(instance, i), DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
1095 			cmn_err(CE_NOTE, "[%d] Couldn't add raw node '%s'",
1096 			    instance, name);
1097 			return (EIO);
1098 		}
1099 	}
1100 
1101 	return (0);
1102 }
1103 
1104 /*
1105  * Function:
1106  *	vdc_create_device_nodes_props
1107  *
1108  * Description:
1109  *	This function creates the block and character device nodes under
1110  *	/devices along with the node properties. It is called as part of
1111  *	the attach(9E) of the instance during the handshake with vds after
1112  *	vds has sent the attributes to vdc.
1113  *
1114  * Parameters:
1115  *	vdc 		- soft state pointer
1116  *
1117  * Return Values
1118  *	0		- Success
1119  *	EIO		- Failed to create device node property
1120  *	EINVAL		- Unknown type of disk exported
1121  */
1122 static int
1123 vdc_create_device_nodes_props(vdc_t *vdc)
1124 {
1125 	dev_info_t	*dip = NULL;
1126 	int		instance;
1127 	int		num_slices = 1;
1128 	int64_t		size = 0;
1129 	dev_t		dev;
1130 	int		rv;
1131 	int		i;
1132 
1133 	ASSERT(vdc != NULL);
1134 
1135 	instance = vdc->instance;
1136 	dip = vdc->dip;
1137 
1138 	switch (vdc->vdisk_type) {
1139 	case VD_DISK_TYPE_DISK:
1140 		num_slices = V_NUMPAR;
1141 		break;
1142 	case VD_DISK_TYPE_SLICE:
1143 		num_slices = 1;
1144 		break;
1145 	case VD_DISK_TYPE_UNK:
1146 	default:
1147 		return (EINVAL);
1148 	}
1149 
1150 	if (vdc->vdisk_label == VD_DISK_LABEL_UNK) {
1151 		/* remove all properties */
1152 		for (i = 0; i < num_slices; i++) {
1153 			dev = makedevice(ddi_driver_major(dip),
1154 			    VD_MAKE_DEV(instance, i));
1155 			(void) ddi_prop_remove(dev, dip, VDC_SIZE_PROP_NAME);
1156 			(void) ddi_prop_remove(dev, dip, VDC_NBLOCKS_PROP_NAME);
1157 		}
1158 		return (0);
1159 	}
1160 
1161 	for (i = 0; i < num_slices; i++) {
1162 		dev = makedevice(ddi_driver_major(dip),
1163 		    VD_MAKE_DEV(instance, i));
1164 
1165 		size = vdc->slice[i].nblocks * vdc->block_size;
1166 		DMSG(vdc, 0, "[%d] sz %ld (%ld Mb)  p_size %lx\n",
1167 		    instance, size, size / (1024 * 1024),
1168 		    vdc->slice[i].nblocks);
1169 
1170 		rv = ddi_prop_update_int64(dev, dip, VDC_SIZE_PROP_NAME, size);
1171 		if (rv != DDI_PROP_SUCCESS) {
1172 			cmn_err(CE_NOTE, "[%d] Couldn't add '%s' prop of [%ld]",
1173 			    instance, VDC_SIZE_PROP_NAME, size);
1174 			return (EIO);
1175 		}
1176 
1177 		rv = ddi_prop_update_int64(dev, dip, VDC_NBLOCKS_PROP_NAME,
1178 		    lbtodb(size));
1179 		if (rv != DDI_PROP_SUCCESS) {
1180 			cmn_err(CE_NOTE, "[%d] Couldn't add '%s' prop [%llu]",
1181 			    instance, VDC_NBLOCKS_PROP_NAME, lbtodb(size));
1182 			return (EIO);
1183 		}
1184 	}
1185 
1186 	return (0);
1187 }
1188 
1189 /*
1190  * Function:
1191  *	vdc_is_opened
1192  *
1193  * Description:
1194  *	This function checks if any slice of a given virtual disk is
1195  *	currently opened.
1196  *
1197  * Parameters:
1198  *	vdc 		- soft state pointer
1199  *
1200  * Return Values
1201  *	B_TRUE		- at least one slice is opened.
1202  *	B_FALSE		- no slice is opened.
1203  */
1204 static boolean_t
1205 vdc_is_opened(vdc_t *vdc)
1206 {
1207 	int i, nslices;
1208 
1209 	switch (vdc->vdisk_type) {
1210 	case VD_DISK_TYPE_DISK:
1211 		nslices = V_NUMPAR;
1212 		break;
1213 	case VD_DISK_TYPE_SLICE:
1214 		nslices = 1;
1215 		break;
1216 	case VD_DISK_TYPE_UNK:
1217 	default:
1218 		ASSERT(0);
1219 	}
1220 
1221 	/* check if there's any layered open */
1222 	for (i = 0; i < nslices; i++) {
1223 		if (vdc->open_lyr[i] > 0)
1224 			return (B_TRUE);
1225 	}
1226 
1227 	/* check if there is any other kind of open */
1228 	for (i = 0; i < OTYPCNT; i++) {
1229 		if (vdc->open[i] != 0)
1230 			return (B_TRUE);
1231 	}
1232 
1233 	return (B_FALSE);
1234 }
1235 
1236 static int
1237 vdc_mark_opened(vdc_t *vdc, int slice, int flag, int otyp)
1238 {
1239 	uint8_t slicemask;
1240 	int i;
1241 
1242 	ASSERT(otyp < OTYPCNT);
1243 	ASSERT(slice < V_NUMPAR);
1244 	ASSERT(MUTEX_HELD(&vdc->lock));
1245 
1246 	slicemask = 1 << slice;
1247 
1248 	/* check if slice is already exclusively opened */
1249 	if (vdc->open_excl & slicemask)
1250 		return (EBUSY);
1251 
1252 	/* if open exclusive, check if slice is already opened */
1253 	if (flag & FEXCL) {
1254 		if (vdc->open_lyr[slice] > 0)
1255 			return (EBUSY);
1256 		for (i = 0; i < OTYPCNT; i++) {
1257 			if (vdc->open[i] & slicemask)
1258 				return (EBUSY);
1259 		}
1260 		vdc->open_excl |= slicemask;
1261 	}
1262 
1263 	/* mark slice as opened */
1264 	if (otyp == OTYP_LYR) {
1265 		vdc->open_lyr[slice]++;
1266 	} else {
1267 		vdc->open[otyp] |= slicemask;
1268 	}
1269 
1270 	return (0);
1271 }
1272 
1273 static void
1274 vdc_mark_closed(vdc_t *vdc, int slice, int flag, int otyp)
1275 {
1276 	uint8_t slicemask;
1277 
1278 	ASSERT(otyp < OTYPCNT);
1279 	ASSERT(slice < V_NUMPAR);
1280 	ASSERT(MUTEX_HELD(&vdc->lock));
1281 
1282 	slicemask = 1 << slice;
1283 
1284 	if (otyp == OTYP_LYR) {
1285 		ASSERT(vdc->open_lyr[slice] > 0);
1286 		vdc->open_lyr[slice]--;
1287 	} else {
1288 		vdc->open[otyp] &= ~slicemask;
1289 	}
1290 
1291 	if (flag & FEXCL)
1292 		vdc->open_excl &= ~slicemask;
1293 }
1294 
1295 static int
1296 vdc_open(dev_t *dev, int flag, int otyp, cred_t *cred)
1297 {
1298 	_NOTE(ARGUNUSED(cred))
1299 
1300 	int	instance, nodelay;
1301 	int	slice, status = 0;
1302 	vdc_t	*vdc;
1303 
1304 	ASSERT(dev != NULL);
1305 	instance = VDCUNIT(*dev);
1306 
1307 	if (otyp >= OTYPCNT)
1308 		return (EINVAL);
1309 
1310 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1311 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1312 		return (ENXIO);
1313 	}
1314 
1315 	DMSG(vdc, 0, "minor = %d flag = %x, otyp = %x\n",
1316 	    getminor(*dev), flag, otyp);
1317 
1318 	slice = VDCPART(*dev);
1319 
1320 	nodelay = flag & (FNDELAY | FNONBLOCK);
1321 
1322 	if ((flag & FWRITE) && (!nodelay) &&
1323 	    !(VD_OP_SUPPORTED(vdc->operations, VD_OP_BWRITE))) {
1324 		return (EROFS);
1325 	}
1326 
1327 	mutex_enter(&vdc->lock);
1328 
1329 	status = vdc_mark_opened(vdc, slice, flag, otyp);
1330 
1331 	if (status != 0) {
1332 		mutex_exit(&vdc->lock);
1333 		return (status);
1334 	}
1335 
1336 	if (nodelay) {
1337 
1338 		/* don't resubmit a validate request if there's already one */
1339 		if (vdc->validate_pending > 0) {
1340 			mutex_exit(&vdc->lock);
1341 			return (0);
1342 		}
1343 
1344 		/* call vdc_validate() asynchronously to avoid blocking */
1345 		if (taskq_dispatch(system_taskq, vdc_validate_task,
1346 		    (void *)vdc, TQ_NOSLEEP) == NULL) {
1347 			vdc_mark_closed(vdc, slice, flag, otyp);
1348 			mutex_exit(&vdc->lock);
1349 			return (ENXIO);
1350 		}
1351 
1352 		vdc->validate_pending++;
1353 		mutex_exit(&vdc->lock);
1354 		return (0);
1355 	}
1356 
1357 	mutex_exit(&vdc->lock);
1358 
1359 	vdc_validate(vdc);
1360 
1361 	mutex_enter(&vdc->lock);
1362 
1363 	if (vdc->vdisk_label == VD_DISK_LABEL_UNK ||
1364 	    vdc->slice[slice].nblocks == 0) {
1365 		vdc_mark_closed(vdc, slice, flag, otyp);
1366 		status = EIO;
1367 	}
1368 
1369 	mutex_exit(&vdc->lock);
1370 
1371 	return (status);
1372 }
1373 
1374 static int
1375 vdc_close(dev_t dev, int flag, int otyp, cred_t *cred)
1376 {
1377 	_NOTE(ARGUNUSED(cred))
1378 
1379 	int	instance;
1380 	int	slice;
1381 	int	rv, rval;
1382 	vdc_t	*vdc;
1383 
1384 	instance = VDCUNIT(dev);
1385 
1386 	if (otyp >= OTYPCNT)
1387 		return (EINVAL);
1388 
1389 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1390 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1391 		return (ENXIO);
1392 	}
1393 
1394 	DMSG(vdc, 0, "[%d] flag = %x, otyp = %x\n", instance, flag, otyp);
1395 
1396 	slice = VDCPART(dev);
1397 
1398 	/*
1399 	 * Attempt to flush the W$ on a close operation. If this is
1400 	 * not a supported IOCTL command or the backing device is read-only
1401 	 * do not fail the close operation.
1402 	 */
1403 	rv = vd_process_ioctl(dev, DKIOCFLUSHWRITECACHE, NULL, FKIOCTL, &rval);
1404 
1405 	if (rv != 0 && rv != ENOTSUP && rv != ENOTTY && rv != EROFS) {
1406 		DMSG(vdc, 0, "[%d] flush failed with error %d on close\n",
1407 		    instance, rv);
1408 		return (EIO);
1409 	}
1410 
1411 	mutex_enter(&vdc->lock);
1412 	vdc_mark_closed(vdc, slice, flag, otyp);
1413 	mutex_exit(&vdc->lock);
1414 
1415 	return (0);
1416 }
1417 
1418 static int
1419 vdc_ioctl(dev_t dev, int cmd, intptr_t arg, int mode, cred_t *credp, int *rvalp)
1420 {
1421 	_NOTE(ARGUNUSED(credp))
1422 
1423 	return (vd_process_ioctl(dev, cmd, (caddr_t)arg, mode, rvalp));
1424 }
1425 
1426 static int
1427 vdc_print(dev_t dev, char *str)
1428 {
1429 	cmn_err(CE_NOTE, "vdc%d:  %s", VDCUNIT(dev), str);
1430 	return (0);
1431 }
1432 
1433 static int
1434 vdc_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblk)
1435 {
1436 	int	rv;
1437 	size_t	nbytes = nblk * DEV_BSIZE;
1438 	int	instance = VDCUNIT(dev);
1439 	vdc_t	*vdc = NULL;
1440 
1441 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1442 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1443 		return (ENXIO);
1444 	}
1445 
1446 	DMSG(vdc, 2, "[%d] dump %ld bytes at block 0x%lx : addr=0x%p\n",
1447 	    instance, nbytes, blkno, (void *)addr);
1448 	rv = vdc_send_request(vdc, VD_OP_BWRITE, addr, nbytes,
1449 	    VDCPART(dev), blkno, CB_STRATEGY, 0, VIO_write_dir);
1450 	if (rv) {
1451 		DMSG(vdc, 0, "Failed to do a disk dump (err=%d)\n", rv);
1452 		return (rv);
1453 	}
1454 
1455 	if (ddi_in_panic())
1456 		(void) vdc_drain_response(vdc);
1457 
1458 	DMSG(vdc, 0, "[%d] End\n", instance);
1459 
1460 	return (0);
1461 }
1462 
1463 /* -------------------------------------------------------------------------- */
1464 
1465 /*
1466  * Disk access routines
1467  *
1468  */
1469 
1470 /*
1471  * vdc_strategy()
1472  *
1473  * Return Value:
1474  *	0:	As per strategy(9E), the strategy() function must return 0
1475  *		[ bioerror(9f) sets b_flags to the proper error code ]
1476  */
1477 static int
1478 vdc_strategy(struct buf *buf)
1479 {
1480 	int	rv = -1;
1481 	vdc_t	*vdc = NULL;
1482 	int	instance = VDCUNIT(buf->b_edev);
1483 	int	op = (buf->b_flags & B_READ) ? VD_OP_BREAD : VD_OP_BWRITE;
1484 	int	slice;
1485 
1486 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1487 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1488 		bioerror(buf, ENXIO);
1489 		biodone(buf);
1490 		return (0);
1491 	}
1492 
1493 	DMSG(vdc, 2, "[%d] %s %ld bytes at block %llx : b_addr=0x%p\n",
1494 	    instance, (buf->b_flags & B_READ) ? "Read" : "Write",
1495 	    buf->b_bcount, buf->b_lblkno, (void *)buf->b_un.b_addr);
1496 
1497 	bp_mapin(buf);
1498 
1499 	if ((long)buf->b_private == VD_SLICE_NONE) {
1500 		/* I/O using an absolute disk offset */
1501 		slice = VD_SLICE_NONE;
1502 	} else {
1503 		slice = VDCPART(buf->b_edev);
1504 	}
1505 
1506 	rv = vdc_send_request(vdc, op, (caddr_t)buf->b_un.b_addr,
1507 	    buf->b_bcount, slice, buf->b_lblkno,
1508 	    CB_STRATEGY, buf, (op == VD_OP_BREAD) ? VIO_read_dir :
1509 	    VIO_write_dir);
1510 
1511 	/*
1512 	 * If the request was successfully sent, the strategy call returns and
1513 	 * the ACK handler calls the bioxxx functions when the vDisk server is
1514 	 * done otherwise we handle the error here.
1515 	 */
1516 	if (rv) {
1517 		DMSG(vdc, 0, "Failed to read/write (err=%d)\n", rv);
1518 		bioerror(buf, rv);
1519 		biodone(buf);
1520 	}
1521 
1522 	return (0);
1523 }
1524 
1525 /*
1526  * Function:
1527  *	vdc_min
1528  *
1529  * Description:
1530  *	Routine to limit the size of a data transfer. Used in
1531  *	conjunction with physio(9F).
1532  *
1533  * Arguments:
1534  *	bp - pointer to the indicated buf(9S) struct.
1535  *
1536  */
1537 static void
1538 vdc_min(struct buf *bufp)
1539 {
1540 	vdc_t	*vdc = NULL;
1541 	int	instance = VDCUNIT(bufp->b_edev);
1542 
1543 	vdc = ddi_get_soft_state(vdc_state, instance);
1544 	VERIFY(vdc != NULL);
1545 
1546 	if (bufp->b_bcount > (vdc->max_xfer_sz * vdc->block_size)) {
1547 		bufp->b_bcount = vdc->max_xfer_sz * vdc->block_size;
1548 	}
1549 }
1550 
1551 static int
1552 vdc_read(dev_t dev, struct uio *uio, cred_t *cred)
1553 {
1554 	_NOTE(ARGUNUSED(cred))
1555 
1556 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1557 	return (physio(vdc_strategy, NULL, dev, B_READ, vdc_min, uio));
1558 }
1559 
1560 static int
1561 vdc_write(dev_t dev, struct uio *uio, cred_t *cred)
1562 {
1563 	_NOTE(ARGUNUSED(cred))
1564 
1565 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1566 	return (physio(vdc_strategy, NULL, dev, B_WRITE, vdc_min, uio));
1567 }
1568 
1569 static int
1570 vdc_aread(dev_t dev, struct aio_req *aio, cred_t *cred)
1571 {
1572 	_NOTE(ARGUNUSED(cred))
1573 
1574 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1575 	return (aphysio(vdc_strategy, anocancel, dev, B_READ, vdc_min, aio));
1576 }
1577 
1578 static int
1579 vdc_awrite(dev_t dev, struct aio_req *aio, cred_t *cred)
1580 {
1581 	_NOTE(ARGUNUSED(cred))
1582 
1583 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1584 	return (aphysio(vdc_strategy, anocancel, dev, B_WRITE, vdc_min, aio));
1585 }
1586 
1587 
1588 /* -------------------------------------------------------------------------- */
1589 
1590 /*
1591  * Handshake support
1592  */
1593 
1594 
1595 /*
1596  * Function:
1597  *	vdc_init_ver_negotiation()
1598  *
1599  * Description:
1600  *
1601  * Arguments:
1602  *	vdc	- soft state pointer for this instance of the device driver.
1603  *
1604  * Return Code:
1605  *	0	- Success
1606  */
1607 static int
1608 vdc_init_ver_negotiation(vdc_t *vdc, vio_ver_t ver)
1609 {
1610 	vio_ver_msg_t	pkt;
1611 	size_t		msglen = sizeof (pkt);
1612 	int		status = -1;
1613 
1614 	ASSERT(vdc != NULL);
1615 	ASSERT(mutex_owned(&vdc->lock));
1616 
1617 	DMSG(vdc, 0, "[%d] Entered.\n", vdc->instance);
1618 
1619 	/*
1620 	 * set the Session ID to a unique value
1621 	 * (the lower 32 bits of the clock tick)
1622 	 */
1623 	vdc->session_id = ((uint32_t)gettick() & 0xffffffff);
1624 	DMSG(vdc, 0, "[%d] Set SID to 0x%lx\n", vdc->instance, vdc->session_id);
1625 
1626 	pkt.tag.vio_msgtype = VIO_TYPE_CTRL;
1627 	pkt.tag.vio_subtype = VIO_SUBTYPE_INFO;
1628 	pkt.tag.vio_subtype_env = VIO_VER_INFO;
1629 	pkt.tag.vio_sid = vdc->session_id;
1630 	pkt.dev_class = VDEV_DISK;
1631 	pkt.ver_major = ver.major;
1632 	pkt.ver_minor = ver.minor;
1633 
1634 	status = vdc_send(vdc, (caddr_t)&pkt, &msglen);
1635 	DMSG(vdc, 0, "[%d] Ver info sent (status = %d)\n",
1636 	    vdc->instance, status);
1637 	if ((status != 0) || (msglen != sizeof (vio_ver_msg_t))) {
1638 		DMSG(vdc, 0, "[%d] Failed to send Ver negotiation info: "
1639 		    "id(%lx) rv(%d) size(%ld)", vdc->instance, vdc->ldc_handle,
1640 		    status, msglen);
1641 		if (msglen != sizeof (vio_ver_msg_t))
1642 			status = ENOMSG;
1643 	}
1644 
1645 	return (status);
1646 }
1647 
1648 /*
1649  * Function:
1650  *	vdc_ver_negotiation()
1651  *
1652  * Description:
1653  *
1654  * Arguments:
1655  *	vdcp	- soft state pointer for this instance of the device driver.
1656  *
1657  * Return Code:
1658  *	0	- Success
1659  */
1660 static int
1661 vdc_ver_negotiation(vdc_t *vdcp)
1662 {
1663 	vio_msg_t vio_msg;
1664 	int status;
1665 
1666 	if (status = vdc_init_ver_negotiation(vdcp, vdc_version[0]))
1667 		return (status);
1668 
1669 	/* release lock and wait for response */
1670 	mutex_exit(&vdcp->lock);
1671 	status = vdc_wait_for_response(vdcp, &vio_msg);
1672 	mutex_enter(&vdcp->lock);
1673 	if (status) {
1674 		DMSG(vdcp, 0,
1675 		    "[%d] Failed waiting for Ver negotiation response, rv(%d)",
1676 		    vdcp->instance, status);
1677 		return (status);
1678 	}
1679 
1680 	/* check type and sub_type ... */
1681 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1682 	    vio_msg.tag.vio_subtype == VIO_SUBTYPE_INFO) {
1683 		DMSG(vdcp, 0, "[%d] Invalid ver negotiation response\n",
1684 		    vdcp->instance);
1685 		return (EPROTO);
1686 	}
1687 
1688 	return (vdc_handle_ver_msg(vdcp, (vio_ver_msg_t *)&vio_msg));
1689 }
1690 
1691 /*
1692  * Function:
1693  *	vdc_init_attr_negotiation()
1694  *
1695  * Description:
1696  *
1697  * Arguments:
1698  *	vdc	- soft state pointer for this instance of the device driver.
1699  *
1700  * Return Code:
1701  *	0	- Success
1702  */
1703 static int
1704 vdc_init_attr_negotiation(vdc_t *vdc)
1705 {
1706 	vd_attr_msg_t	pkt;
1707 	size_t		msglen = sizeof (pkt);
1708 	int		status;
1709 
1710 	ASSERT(vdc != NULL);
1711 	ASSERT(mutex_owned(&vdc->lock));
1712 
1713 	DMSG(vdc, 0, "[%d] entered\n", vdc->instance);
1714 
1715 	/* fill in tag */
1716 	pkt.tag.vio_msgtype = VIO_TYPE_CTRL;
1717 	pkt.tag.vio_subtype = VIO_SUBTYPE_INFO;
1718 	pkt.tag.vio_subtype_env = VIO_ATTR_INFO;
1719 	pkt.tag.vio_sid = vdc->session_id;
1720 	/* fill in payload */
1721 	pkt.max_xfer_sz = vdc->max_xfer_sz;
1722 	pkt.vdisk_block_size = vdc->block_size;
1723 	pkt.xfer_mode = VIO_DRING_MODE_V1_0;
1724 	pkt.operations = 0;	/* server will set bits of valid operations */
1725 	pkt.vdisk_type = 0;	/* server will set to valid device type */
1726 	pkt.vdisk_media = 0;	/* server will set to valid media type */
1727 	pkt.vdisk_size = 0;	/* server will set to valid size */
1728 
1729 	status = vdc_send(vdc, (caddr_t)&pkt, &msglen);
1730 	DMSG(vdc, 0, "Attr info sent (status = %d)\n", status);
1731 
1732 	if ((status != 0) || (msglen != sizeof (vio_ver_msg_t))) {
1733 		DMSG(vdc, 0, "[%d] Failed to send Attr negotiation info: "
1734 		    "id(%lx) rv(%d) size(%ld)", vdc->instance, vdc->ldc_handle,
1735 		    status, msglen);
1736 		if (msglen != sizeof (vio_ver_msg_t))
1737 			status = ENOMSG;
1738 	}
1739 
1740 	return (status);
1741 }
1742 
1743 /*
1744  * Function:
1745  *	vdc_attr_negotiation()
1746  *
1747  * Description:
1748  *
1749  * Arguments:
1750  *	vdc	- soft state pointer for this instance of the device driver.
1751  *
1752  * Return Code:
1753  *	0	- Success
1754  */
1755 static int
1756 vdc_attr_negotiation(vdc_t *vdcp)
1757 {
1758 	int status;
1759 	vio_msg_t vio_msg;
1760 
1761 	if (status = vdc_init_attr_negotiation(vdcp))
1762 		return (status);
1763 
1764 	/* release lock and wait for response */
1765 	mutex_exit(&vdcp->lock);
1766 	status = vdc_wait_for_response(vdcp, &vio_msg);
1767 	mutex_enter(&vdcp->lock);
1768 	if (status) {
1769 		DMSG(vdcp, 0,
1770 		    "[%d] Failed waiting for Attr negotiation response, rv(%d)",
1771 		    vdcp->instance, status);
1772 		return (status);
1773 	}
1774 
1775 	/* check type and sub_type ... */
1776 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1777 	    vio_msg.tag.vio_subtype == VIO_SUBTYPE_INFO) {
1778 		DMSG(vdcp, 0, "[%d] Invalid attr negotiation response\n",
1779 		    vdcp->instance);
1780 		return (EPROTO);
1781 	}
1782 
1783 	return (vdc_handle_attr_msg(vdcp, (vd_attr_msg_t *)&vio_msg));
1784 }
1785 
1786 
1787 /*
1788  * Function:
1789  *	vdc_init_dring_negotiate()
1790  *
1791  * Description:
1792  *
1793  * Arguments:
1794  *	vdc	- soft state pointer for this instance of the device driver.
1795  *
1796  * Return Code:
1797  *	0	- Success
1798  */
1799 static int
1800 vdc_init_dring_negotiate(vdc_t *vdc)
1801 {
1802 	vio_dring_reg_msg_t	pkt;
1803 	size_t			msglen = sizeof (pkt);
1804 	int			status = -1;
1805 	int			retry;
1806 	int			nretries = 10;
1807 
1808 	ASSERT(vdc != NULL);
1809 	ASSERT(mutex_owned(&vdc->lock));
1810 
1811 	for (retry = 0; retry < nretries; retry++) {
1812 		status = vdc_init_descriptor_ring(vdc);
1813 		if (status != EAGAIN)
1814 			break;
1815 		drv_usecwait(vdc_min_timeout_ldc);
1816 	}
1817 
1818 	if (status != 0) {
1819 		DMSG(vdc, 0, "[%d] Failed to init DRing (status = %d)\n",
1820 		    vdc->instance, status);
1821 		return (status);
1822 	}
1823 
1824 	DMSG(vdc, 0, "[%d] Init of descriptor ring completed (status = %d)\n",
1825 	    vdc->instance, status);
1826 
1827 	/* fill in tag */
1828 	pkt.tag.vio_msgtype = VIO_TYPE_CTRL;
1829 	pkt.tag.vio_subtype = VIO_SUBTYPE_INFO;
1830 	pkt.tag.vio_subtype_env = VIO_DRING_REG;
1831 	pkt.tag.vio_sid = vdc->session_id;
1832 	/* fill in payload */
1833 	pkt.dring_ident = 0;
1834 	pkt.num_descriptors = vdc->dring_len;
1835 	pkt.descriptor_size = vdc->dring_entry_size;
1836 	pkt.options = (VIO_TX_DRING | VIO_RX_DRING);
1837 	pkt.ncookies = vdc->dring_cookie_count;
1838 	pkt.cookie[0] = vdc->dring_cookie[0];	/* for now just one cookie */
1839 
1840 	status = vdc_send(vdc, (caddr_t)&pkt, &msglen);
1841 	if (status != 0) {
1842 		DMSG(vdc, 0, "[%d] Failed to register DRing (err = %d)",
1843 		    vdc->instance, status);
1844 	}
1845 
1846 	return (status);
1847 }
1848 
1849 
1850 /*
1851  * Function:
1852  *	vdc_dring_negotiation()
1853  *
1854  * Description:
1855  *
1856  * Arguments:
1857  *	vdc	- soft state pointer for this instance of the device driver.
1858  *
1859  * Return Code:
1860  *	0	- Success
1861  */
1862 static int
1863 vdc_dring_negotiation(vdc_t *vdcp)
1864 {
1865 	int status;
1866 	vio_msg_t vio_msg;
1867 
1868 	if (status = vdc_init_dring_negotiate(vdcp))
1869 		return (status);
1870 
1871 	/* release lock and wait for response */
1872 	mutex_exit(&vdcp->lock);
1873 	status = vdc_wait_for_response(vdcp, &vio_msg);
1874 	mutex_enter(&vdcp->lock);
1875 	if (status) {
1876 		DMSG(vdcp, 0,
1877 		    "[%d] Failed waiting for Dring negotiation response,"
1878 		    " rv(%d)", vdcp->instance, status);
1879 		return (status);
1880 	}
1881 
1882 	/* check type and sub_type ... */
1883 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1884 	    vio_msg.tag.vio_subtype == VIO_SUBTYPE_INFO) {
1885 		DMSG(vdcp, 0, "[%d] Invalid Dring negotiation response\n",
1886 		    vdcp->instance);
1887 		return (EPROTO);
1888 	}
1889 
1890 	return (vdc_handle_dring_reg_msg(vdcp,
1891 	    (vio_dring_reg_msg_t *)&vio_msg));
1892 }
1893 
1894 
1895 /*
1896  * Function:
1897  *	vdc_send_rdx()
1898  *
1899  * Description:
1900  *
1901  * Arguments:
1902  *	vdc	- soft state pointer for this instance of the device driver.
1903  *
1904  * Return Code:
1905  *	0	- Success
1906  */
1907 static int
1908 vdc_send_rdx(vdc_t *vdcp)
1909 {
1910 	vio_msg_t	msg;
1911 	size_t		msglen = sizeof (vio_msg_t);
1912 	int		status;
1913 
1914 	/*
1915 	 * Send an RDX message to vds to indicate we are ready
1916 	 * to send data
1917 	 */
1918 	msg.tag.vio_msgtype = VIO_TYPE_CTRL;
1919 	msg.tag.vio_subtype = VIO_SUBTYPE_INFO;
1920 	msg.tag.vio_subtype_env = VIO_RDX;
1921 	msg.tag.vio_sid = vdcp->session_id;
1922 	status = vdc_send(vdcp, (caddr_t)&msg, &msglen);
1923 	if (status != 0) {
1924 		DMSG(vdcp, 0, "[%d] Failed to send RDX message (%d)",
1925 		    vdcp->instance, status);
1926 	}
1927 
1928 	return (status);
1929 }
1930 
1931 /*
1932  * Function:
1933  *	vdc_handle_rdx()
1934  *
1935  * Description:
1936  *
1937  * Arguments:
1938  *	vdc	- soft state pointer for this instance of the device driver.
1939  *	msgp	- received msg
1940  *
1941  * Return Code:
1942  *	0	- Success
1943  */
1944 static int
1945 vdc_handle_rdx(vdc_t *vdcp, vio_rdx_msg_t *msgp)
1946 {
1947 	_NOTE(ARGUNUSED(vdcp))
1948 	_NOTE(ARGUNUSED(msgp))
1949 
1950 	ASSERT(msgp->tag.vio_msgtype == VIO_TYPE_CTRL);
1951 	ASSERT(msgp->tag.vio_subtype == VIO_SUBTYPE_ACK);
1952 	ASSERT(msgp->tag.vio_subtype_env == VIO_RDX);
1953 
1954 	DMSG(vdcp, 1, "[%d] Got an RDX msg", vdcp->instance);
1955 
1956 	return (0);
1957 }
1958 
1959 /*
1960  * Function:
1961  *	vdc_rdx_exchange()
1962  *
1963  * Description:
1964  *
1965  * Arguments:
1966  *	vdc	- soft state pointer for this instance of the device driver.
1967  *
1968  * Return Code:
1969  *	0	- Success
1970  */
1971 static int
1972 vdc_rdx_exchange(vdc_t *vdcp)
1973 {
1974 	int status;
1975 	vio_msg_t vio_msg;
1976 
1977 	if (status = vdc_send_rdx(vdcp))
1978 		return (status);
1979 
1980 	/* release lock and wait for response */
1981 	mutex_exit(&vdcp->lock);
1982 	status = vdc_wait_for_response(vdcp, &vio_msg);
1983 	mutex_enter(&vdcp->lock);
1984 	if (status) {
1985 		DMSG(vdcp, 0, "[%d] Failed waiting for RDX response, rv(%d)",
1986 		    vdcp->instance, status);
1987 		return (status);
1988 	}
1989 
1990 	/* check type and sub_type ... */
1991 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1992 	    vio_msg.tag.vio_subtype != VIO_SUBTYPE_ACK) {
1993 		DMSG(vdcp, 0, "[%d] Invalid RDX response\n", vdcp->instance);
1994 		return (EPROTO);
1995 	}
1996 
1997 	return (vdc_handle_rdx(vdcp, (vio_rdx_msg_t *)&vio_msg));
1998 }
1999 
2000 
2001 /* -------------------------------------------------------------------------- */
2002 
2003 /*
2004  * LDC helper routines
2005  */
2006 
2007 static int
2008 vdc_recv(vdc_t *vdc, vio_msg_t *msgp, size_t *nbytesp)
2009 {
2010 	int		status;
2011 	boolean_t	q_has_pkts = B_FALSE;
2012 	uint64_t	delay_time;
2013 	size_t		len;
2014 
2015 	mutex_enter(&vdc->read_lock);
2016 
2017 	if (vdc->read_state == VDC_READ_IDLE)
2018 		vdc->read_state = VDC_READ_WAITING;
2019 
2020 	while (vdc->read_state != VDC_READ_PENDING) {
2021 
2022 		/* detect if the connection has been reset */
2023 		if (vdc->read_state == VDC_READ_RESET) {
2024 			status = ECONNRESET;
2025 			goto done;
2026 		}
2027 
2028 		cv_wait(&vdc->read_cv, &vdc->read_lock);
2029 	}
2030 
2031 	/*
2032 	 * Until we get a blocking ldc read we have to retry
2033 	 * until the entire LDC message has arrived before
2034 	 * ldc_read() will succeed. Note we also bail out if
2035 	 * the channel is reset or goes away.
2036 	 */
2037 	delay_time = vdc_ldc_read_init_delay;
2038 loop:
2039 	len = *nbytesp;
2040 	status = ldc_read(vdc->ldc_handle, (caddr_t)msgp, &len);
2041 	switch (status) {
2042 	case EAGAIN:
2043 		delay_time *= 2;
2044 		if (delay_time >= vdc_ldc_read_max_delay)
2045 			delay_time = vdc_ldc_read_max_delay;
2046 		delay(delay_time);
2047 		goto loop;
2048 
2049 	case 0:
2050 		if (len == 0) {
2051 			DMSG(vdc, 1, "[%d] ldc_read returned 0 bytes with "
2052 			    "no error!\n", vdc->instance);
2053 			goto loop;
2054 		}
2055 
2056 		*nbytesp = len;
2057 
2058 		/*
2059 		 * If there are pending messages, leave the
2060 		 * read state as pending. Otherwise, set the state
2061 		 * back to idle.
2062 		 */
2063 		status = ldc_chkq(vdc->ldc_handle, &q_has_pkts);
2064 		if (status == 0 && !q_has_pkts)
2065 			vdc->read_state = VDC_READ_IDLE;
2066 
2067 		break;
2068 	default:
2069 		DMSG(vdc, 0, "ldc_read returned %d\n", status);
2070 		break;
2071 	}
2072 
2073 done:
2074 	mutex_exit(&vdc->read_lock);
2075 
2076 	return (status);
2077 }
2078 
2079 
2080 
2081 #ifdef DEBUG
2082 void
2083 vdc_decode_tag(vdc_t *vdcp, vio_msg_t *msg)
2084 {
2085 	char *ms, *ss, *ses;
2086 	switch (msg->tag.vio_msgtype) {
2087 #define	Q(_s)	case _s : ms = #_s; break;
2088 	Q(VIO_TYPE_CTRL)
2089 	Q(VIO_TYPE_DATA)
2090 	Q(VIO_TYPE_ERR)
2091 #undef Q
2092 	default: ms = "unknown"; break;
2093 	}
2094 
2095 	switch (msg->tag.vio_subtype) {
2096 #define	Q(_s)	case _s : ss = #_s; break;
2097 	Q(VIO_SUBTYPE_INFO)
2098 	Q(VIO_SUBTYPE_ACK)
2099 	Q(VIO_SUBTYPE_NACK)
2100 #undef Q
2101 	default: ss = "unknown"; break;
2102 	}
2103 
2104 	switch (msg->tag.vio_subtype_env) {
2105 #define	Q(_s)	case _s : ses = #_s; break;
2106 	Q(VIO_VER_INFO)
2107 	Q(VIO_ATTR_INFO)
2108 	Q(VIO_DRING_REG)
2109 	Q(VIO_DRING_UNREG)
2110 	Q(VIO_RDX)
2111 	Q(VIO_PKT_DATA)
2112 	Q(VIO_DESC_DATA)
2113 	Q(VIO_DRING_DATA)
2114 #undef Q
2115 	default: ses = "unknown"; break;
2116 	}
2117 
2118 	DMSG(vdcp, 3, "(%x/%x/%x) message : (%s/%s/%s)\n",
2119 	    msg->tag.vio_msgtype, msg->tag.vio_subtype,
2120 	    msg->tag.vio_subtype_env, ms, ss, ses);
2121 }
2122 #endif
2123 
2124 /*
2125  * Function:
2126  *	vdc_send()
2127  *
2128  * Description:
2129  *	The function encapsulates the call to write a message using LDC.
2130  *	If LDC indicates that the call failed due to the queue being full,
2131  *	we retry the ldc_write(), otherwise we return the error returned by LDC.
2132  *
2133  * Arguments:
2134  *	ldc_handle	- LDC handle for the channel this instance of vdc uses
2135  *	pkt		- address of LDC message to be sent
2136  *	msglen		- the size of the message being sent. When the function
2137  *			  returns, this contains the number of bytes written.
2138  *
2139  * Return Code:
2140  *	0		- Success.
2141  *	EINVAL		- pkt or msglen were NULL
2142  *	ECONNRESET	- The connection was not up.
2143  *	EWOULDBLOCK	- LDC queue is full
2144  *	xxx		- other error codes returned by ldc_write
2145  */
2146 static int
2147 vdc_send(vdc_t *vdc, caddr_t pkt, size_t *msglen)
2148 {
2149 	size_t	size = 0;
2150 	int	status = 0;
2151 	clock_t delay_ticks;
2152 
2153 	ASSERT(vdc != NULL);
2154 	ASSERT(mutex_owned(&vdc->lock));
2155 	ASSERT(msglen != NULL);
2156 	ASSERT(*msglen != 0);
2157 
2158 #ifdef DEBUG
2159 	vdc_decode_tag(vdc, (vio_msg_t *)(uintptr_t)pkt);
2160 #endif
2161 	/*
2162 	 * Wait indefinitely to send if channel
2163 	 * is busy, but bail out if we succeed or
2164 	 * if the channel closes or is reset.
2165 	 */
2166 	delay_ticks = vdc_hz_min_ldc_delay;
2167 	do {
2168 		size = *msglen;
2169 		status = ldc_write(vdc->ldc_handle, pkt, &size);
2170 		if (status == EWOULDBLOCK) {
2171 			delay(delay_ticks);
2172 			/* geometric backoff */
2173 			delay_ticks *= 2;
2174 			if (delay_ticks > vdc_hz_max_ldc_delay)
2175 				delay_ticks = vdc_hz_max_ldc_delay;
2176 		}
2177 	} while (status == EWOULDBLOCK);
2178 
2179 	/* if LDC had serious issues --- reset vdc state */
2180 	if (status == EIO || status == ECONNRESET) {
2181 		/* LDC had serious issues --- reset vdc state */
2182 		mutex_enter(&vdc->read_lock);
2183 		if ((vdc->read_state == VDC_READ_WAITING) ||
2184 		    (vdc->read_state == VDC_READ_RESET))
2185 			cv_signal(&vdc->read_cv);
2186 		vdc->read_state = VDC_READ_RESET;
2187 		mutex_exit(&vdc->read_lock);
2188 
2189 		/* wake up any waiters in the reset thread */
2190 		if (vdc->state == VDC_STATE_INIT_WAITING) {
2191 			DMSG(vdc, 0, "[%d] write reset - "
2192 			    "vdc is resetting ..\n", vdc->instance);
2193 			vdc->state = VDC_STATE_RESETTING;
2194 			cv_signal(&vdc->initwait_cv);
2195 		}
2196 
2197 		return (ECONNRESET);
2198 	}
2199 
2200 	/* return the last size written */
2201 	*msglen = size;
2202 
2203 	return (status);
2204 }
2205 
2206 /*
2207  * Function:
2208  *	vdc_get_md_node
2209  *
2210  * Description:
2211  *	Get the MD, the device node and the port node for the given
2212  *	disk instance. The caller is responsible for cleaning up the
2213  *	reference to the returned MD (mdpp) by calling md_fini_handle().
2214  *
2215  * Arguments:
2216  *	dip	- dev info pointer for this instance of the device driver.
2217  *	mdpp	- the returned MD.
2218  *	vd_nodep - the returned device node.
2219  *	vd_portp - the returned port node. The returned port node is NULL
2220  *		   if no port node is found.
2221  *
2222  * Return Code:
2223  *	0	- Success.
2224  *	ENOENT	- Expected node or property did not exist.
2225  *	ENXIO	- Unexpected error communicating with MD framework
2226  */
2227 static int
2228 vdc_get_md_node(dev_info_t *dip, md_t **mdpp, mde_cookie_t *vd_nodep,
2229     mde_cookie_t *vd_portp)
2230 {
2231 	int		status = ENOENT;
2232 	char		*node_name = NULL;
2233 	md_t		*mdp = NULL;
2234 	int		num_nodes;
2235 	int		num_vdevs;
2236 	int		num_vports;
2237 	mde_cookie_t	rootnode;
2238 	mde_cookie_t	*listp = NULL;
2239 	boolean_t	found_inst = B_FALSE;
2240 	int		listsz;
2241 	int		idx;
2242 	uint64_t	md_inst;
2243 	int		obp_inst;
2244 	int		instance = ddi_get_instance(dip);
2245 
2246 	/*
2247 	 * Get the OBP instance number for comparison with the MD instance
2248 	 *
2249 	 * The "cfg-handle" property of a vdc node in an MD contains the MD's
2250 	 * notion of "instance", or unique identifier, for that node; OBP
2251 	 * stores the value of the "cfg-handle" MD property as the value of
2252 	 * the "reg" property on the node in the device tree it builds from
2253 	 * the MD and passes to Solaris.  Thus, we look up the devinfo node's
2254 	 * "reg" property value to uniquely identify this device instance.
2255 	 * If the "reg" property cannot be found, the device tree state is
2256 	 * presumably so broken that there is no point in continuing.
2257 	 */
2258 	if (!ddi_prop_exists(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS, OBP_REG)) {
2259 		cmn_err(CE_WARN, "'%s' property does not exist", OBP_REG);
2260 		return (ENOENT);
2261 	}
2262 	obp_inst = ddi_prop_get_int(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS,
2263 	    OBP_REG, -1);
2264 	DMSGX(1, "[%d] OBP inst=%d\n", instance, obp_inst);
2265 
2266 	/*
2267 	 * We now walk the MD nodes to find the node for this vdisk.
2268 	 */
2269 	if ((mdp = md_get_handle()) == NULL) {
2270 		cmn_err(CE_WARN, "unable to init machine description");
2271 		return (ENXIO);
2272 	}
2273 
2274 	num_nodes = md_node_count(mdp);
2275 	ASSERT(num_nodes > 0);
2276 
2277 	listsz = num_nodes * sizeof (mde_cookie_t);
2278 
2279 	/* allocate memory for nodes */
2280 	listp = kmem_zalloc(listsz, KM_SLEEP);
2281 
2282 	rootnode = md_root_node(mdp);
2283 	ASSERT(rootnode != MDE_INVAL_ELEM_COOKIE);
2284 
2285 	/*
2286 	 * Search for all the virtual devices, we will then check to see which
2287 	 * ones are disk nodes.
2288 	 */
2289 	num_vdevs = md_scan_dag(mdp, rootnode,
2290 	    md_find_name(mdp, VDC_MD_VDEV_NAME),
2291 	    md_find_name(mdp, "fwd"), listp);
2292 
2293 	if (num_vdevs <= 0) {
2294 		cmn_err(CE_NOTE, "No '%s' node found", VDC_MD_VDEV_NAME);
2295 		status = ENOENT;
2296 		goto done;
2297 	}
2298 
2299 	DMSGX(1, "[%d] num_vdevs=%d\n", instance, num_vdevs);
2300 	for (idx = 0; idx < num_vdevs; idx++) {
2301 		status = md_get_prop_str(mdp, listp[idx], "name", &node_name);
2302 		if ((status != 0) || (node_name == NULL)) {
2303 			cmn_err(CE_NOTE, "Unable to get name of node type '%s'"
2304 			    ": err %d", VDC_MD_VDEV_NAME, status);
2305 			continue;
2306 		}
2307 
2308 		DMSGX(1, "[%d] Found node '%s'\n", instance, node_name);
2309 		if (strcmp(VDC_MD_DISK_NAME, node_name) == 0) {
2310 			status = md_get_prop_val(mdp, listp[idx],
2311 			    VDC_MD_CFG_HDL, &md_inst);
2312 			DMSGX(1, "[%d] vdc inst in MD=%lx\n",
2313 			    instance, md_inst);
2314 			if ((status == 0) && (md_inst == obp_inst)) {
2315 				found_inst = B_TRUE;
2316 				break;
2317 			}
2318 		}
2319 	}
2320 
2321 	if (!found_inst) {
2322 		DMSGX(0, "Unable to find correct '%s' node", VDC_MD_DISK_NAME);
2323 		status = ENOENT;
2324 		goto done;
2325 	}
2326 	DMSGX(0, "[%d] MD inst=%lx\n", instance, md_inst);
2327 
2328 	*vd_nodep = listp[idx];
2329 	*mdpp = mdp;
2330 
2331 	num_vports = md_scan_dag(mdp, *vd_nodep,
2332 	    md_find_name(mdp, VDC_MD_PORT_NAME),
2333 	    md_find_name(mdp, "fwd"), listp);
2334 
2335 	if (num_vports != 1) {
2336 		DMSGX(0, "Expected 1 '%s' node for '%s' port, found %d\n",
2337 		    VDC_MD_PORT_NAME, VDC_MD_VDEV_NAME, num_vports);
2338 	}
2339 
2340 	*vd_portp = (num_vports == 0)? NULL: listp[0];
2341 
2342 done:
2343 	kmem_free(listp, listsz);
2344 	return (status);
2345 }
2346 
2347 /*
2348  * Function:
2349  *	vdc_get_ldc_id()
2350  *
2351  * Description:
2352  *	This function gets the 'ldc-id' for this particular instance of vdc.
2353  *	The id returned is the guest domain channel endpoint LDC uses for
2354  *	communication with vds.
2355  *
2356  * Arguments:
2357  *	mdp	- pointer to the machine description.
2358  *	vd_node	- the vdisk element from the MD.
2359  *	ldc_id	- pointer to variable used to return the 'ldc-id' found.
2360  *
2361  * Return Code:
2362  *	0	- Success.
2363  *	ENOENT	- Expected node or property did not exist.
2364  */
2365 static int
2366 vdc_get_ldc_id(md_t *mdp, mde_cookie_t vd_node, uint64_t *ldc_id)
2367 {
2368 	mde_cookie_t	*chanp = NULL;
2369 	int		listsz;
2370 	int		num_chans;
2371 	int		num_nodes;
2372 	int		status = 0;
2373 
2374 	num_nodes = md_node_count(mdp);
2375 	ASSERT(num_nodes > 0);
2376 
2377 	listsz = num_nodes * sizeof (mde_cookie_t);
2378 
2379 	/* allocate memory for nodes */
2380 	chanp = kmem_zalloc(listsz, KM_SLEEP);
2381 
2382 	/* get the channels for this node */
2383 	num_chans = md_scan_dag(mdp, vd_node,
2384 	    md_find_name(mdp, VDC_MD_CHAN_NAME),
2385 	    md_find_name(mdp, "fwd"), chanp);
2386 
2387 	/* expecting at least one channel */
2388 	if (num_chans <= 0) {
2389 		cmn_err(CE_NOTE, "No '%s' node for '%s' port",
2390 		    VDC_MD_CHAN_NAME, VDC_MD_VDEV_NAME);
2391 		status = ENOENT;
2392 		goto done;
2393 
2394 	} else if (num_chans != 1) {
2395 		DMSGX(0, "Expected 1 '%s' node for '%s' port, found %d\n",
2396 		    VDC_MD_CHAN_NAME, VDC_MD_VDEV_NAME, num_chans);
2397 	}
2398 
2399 	/*
2400 	 * We use the first channel found (index 0), irrespective of how
2401 	 * many are there in total.
2402 	 */
2403 	if (md_get_prop_val(mdp, chanp[0], VDC_MD_ID, ldc_id) != 0) {
2404 		cmn_err(CE_NOTE, "Channel '%s' property not found", VDC_MD_ID);
2405 		status = ENOENT;
2406 	}
2407 
2408 done:
2409 	kmem_free(chanp, listsz);
2410 	return (status);
2411 }
2412 
2413 static int
2414 vdc_do_ldc_up(vdc_t *vdc)
2415 {
2416 	int		status;
2417 	ldc_status_t	ldc_state;
2418 
2419 	DMSG(vdc, 0, "[%d] Bringing up channel %lx\n",
2420 	    vdc->instance, vdc->ldc_id);
2421 
2422 	if (vdc->lifecycle == VDC_LC_DETACHING)
2423 		return (EINVAL);
2424 
2425 	if ((status = ldc_up(vdc->ldc_handle)) != 0) {
2426 		switch (status) {
2427 		case ECONNREFUSED:	/* listener not ready at other end */
2428 			DMSG(vdc, 0, "[%d] ldc_up(%lx,...) return %d\n",
2429 			    vdc->instance, vdc->ldc_id, status);
2430 			status = 0;
2431 			break;
2432 		default:
2433 			DMSG(vdc, 0, "[%d] Failed to bring up LDC: "
2434 			    "channel=%ld, err=%d", vdc->instance, vdc->ldc_id,
2435 			    status);
2436 			break;
2437 		}
2438 	}
2439 
2440 	if (ldc_status(vdc->ldc_handle, &ldc_state) == 0) {
2441 		vdc->ldc_state = ldc_state;
2442 		if (ldc_state == LDC_UP) {
2443 			DMSG(vdc, 0, "[%d] LDC channel already up\n",
2444 			    vdc->instance);
2445 			vdc->seq_num = 1;
2446 			vdc->seq_num_reply = 0;
2447 		}
2448 	}
2449 
2450 	return (status);
2451 }
2452 
2453 /*
2454  * Function:
2455  *	vdc_terminate_ldc()
2456  *
2457  * Description:
2458  *
2459  * Arguments:
2460  *	vdc	- soft state pointer for this instance of the device driver.
2461  *
2462  * Return Code:
2463  *	None
2464  */
2465 static void
2466 vdc_terminate_ldc(vdc_t *vdc)
2467 {
2468 	int	instance = ddi_get_instance(vdc->dip);
2469 
2470 	ASSERT(vdc != NULL);
2471 	ASSERT(mutex_owned(&vdc->lock));
2472 
2473 	DMSG(vdc, 0, "[%d] initialized=%x\n", instance, vdc->initialized);
2474 
2475 	if (vdc->initialized & VDC_LDC_OPEN) {
2476 		DMSG(vdc, 0, "[%d] ldc_close()\n", instance);
2477 		(void) ldc_close(vdc->ldc_handle);
2478 	}
2479 	if (vdc->initialized & VDC_LDC_CB) {
2480 		DMSG(vdc, 0, "[%d] ldc_unreg_callback()\n", instance);
2481 		(void) ldc_unreg_callback(vdc->ldc_handle);
2482 	}
2483 	if (vdc->initialized & VDC_LDC) {
2484 		DMSG(vdc, 0, "[%d] ldc_fini()\n", instance);
2485 		(void) ldc_fini(vdc->ldc_handle);
2486 		vdc->ldc_handle = NULL;
2487 	}
2488 
2489 	vdc->initialized &= ~(VDC_LDC | VDC_LDC_CB | VDC_LDC_OPEN);
2490 }
2491 
2492 /* -------------------------------------------------------------------------- */
2493 
2494 /*
2495  * Descriptor Ring helper routines
2496  */
2497 
2498 /*
2499  * Function:
2500  *	vdc_init_descriptor_ring()
2501  *
2502  * Description:
2503  *
2504  * Arguments:
2505  *	vdc	- soft state pointer for this instance of the device driver.
2506  *
2507  * Return Code:
2508  *	0	- Success
2509  */
2510 static int
2511 vdc_init_descriptor_ring(vdc_t *vdc)
2512 {
2513 	vd_dring_entry_t	*dep = NULL;	/* DRing Entry pointer */
2514 	int	status = 0;
2515 	int	i;
2516 
2517 	DMSG(vdc, 0, "[%d] initialized=%x\n", vdc->instance, vdc->initialized);
2518 
2519 	ASSERT(vdc != NULL);
2520 	ASSERT(mutex_owned(&vdc->lock));
2521 	ASSERT(vdc->ldc_handle != NULL);
2522 
2523 	/* ensure we have enough room to store max sized block */
2524 	ASSERT(maxphys <= VD_MAX_BLOCK_SIZE);
2525 
2526 	if ((vdc->initialized & VDC_DRING_INIT) == 0) {
2527 		DMSG(vdc, 0, "[%d] ldc_mem_dring_create\n", vdc->instance);
2528 		/*
2529 		 * Calculate the maximum block size we can transmit using one
2530 		 * Descriptor Ring entry from the attributes returned by the
2531 		 * vDisk server. This is subject to a minimum of 'maxphys'
2532 		 * as we do not have the capability to split requests over
2533 		 * multiple DRing entries.
2534 		 */
2535 		if ((vdc->max_xfer_sz * vdc->block_size) < maxphys) {
2536 			DMSG(vdc, 0, "[%d] using minimum DRing size\n",
2537 			    vdc->instance);
2538 			vdc->dring_max_cookies = maxphys / PAGESIZE;
2539 		} else {
2540 			vdc->dring_max_cookies =
2541 			    (vdc->max_xfer_sz * vdc->block_size) / PAGESIZE;
2542 		}
2543 		vdc->dring_entry_size = (sizeof (vd_dring_entry_t) +
2544 		    (sizeof (ldc_mem_cookie_t) *
2545 		    (vdc->dring_max_cookies - 1)));
2546 		vdc->dring_len = VD_DRING_LEN;
2547 
2548 		status = ldc_mem_dring_create(vdc->dring_len,
2549 		    vdc->dring_entry_size, &vdc->ldc_dring_hdl);
2550 		if ((vdc->ldc_dring_hdl == NULL) || (status != 0)) {
2551 			DMSG(vdc, 0, "[%d] Descriptor ring creation failed",
2552 			    vdc->instance);
2553 			return (status);
2554 		}
2555 		vdc->initialized |= VDC_DRING_INIT;
2556 	}
2557 
2558 	if ((vdc->initialized & VDC_DRING_BOUND) == 0) {
2559 		DMSG(vdc, 0, "[%d] ldc_mem_dring_bind\n", vdc->instance);
2560 		vdc->dring_cookie =
2561 		    kmem_zalloc(sizeof (ldc_mem_cookie_t), KM_SLEEP);
2562 
2563 		status = ldc_mem_dring_bind(vdc->ldc_handle, vdc->ldc_dring_hdl,
2564 		    LDC_SHADOW_MAP|LDC_DIRECT_MAP, LDC_MEM_RW,
2565 		    &vdc->dring_cookie[0],
2566 		    &vdc->dring_cookie_count);
2567 		if (status != 0) {
2568 			DMSG(vdc, 0, "[%d] Failed to bind descriptor ring "
2569 			    "(%lx) to channel (%lx) status=%d\n",
2570 			    vdc->instance, vdc->ldc_dring_hdl,
2571 			    vdc->ldc_handle, status);
2572 			return (status);
2573 		}
2574 		ASSERT(vdc->dring_cookie_count == 1);
2575 		vdc->initialized |= VDC_DRING_BOUND;
2576 	}
2577 
2578 	status = ldc_mem_dring_info(vdc->ldc_dring_hdl, &vdc->dring_mem_info);
2579 	if (status != 0) {
2580 		DMSG(vdc, 0,
2581 		    "[%d] Failed to get info for descriptor ring (%lx)\n",
2582 		    vdc->instance, vdc->ldc_dring_hdl);
2583 		return (status);
2584 	}
2585 
2586 	if ((vdc->initialized & VDC_DRING_LOCAL) == 0) {
2587 		DMSG(vdc, 0, "[%d] local dring\n", vdc->instance);
2588 
2589 		/* Allocate the local copy of this dring */
2590 		vdc->local_dring =
2591 		    kmem_zalloc(vdc->dring_len * sizeof (vdc_local_desc_t),
2592 		    KM_SLEEP);
2593 		vdc->initialized |= VDC_DRING_LOCAL;
2594 	}
2595 
2596 	/*
2597 	 * Mark all DRing entries as free and initialize the private
2598 	 * descriptor's memory handles. If any entry is initialized,
2599 	 * we need to free it later so we set the bit in 'initialized'
2600 	 * at the start.
2601 	 */
2602 	vdc->initialized |= VDC_DRING_ENTRY;
2603 	for (i = 0; i < vdc->dring_len; i++) {
2604 		dep = VDC_GET_DRING_ENTRY_PTR(vdc, i);
2605 		dep->hdr.dstate = VIO_DESC_FREE;
2606 
2607 		status = ldc_mem_alloc_handle(vdc->ldc_handle,
2608 		    &vdc->local_dring[i].desc_mhdl);
2609 		if (status != 0) {
2610 			DMSG(vdc, 0, "![%d] Failed to alloc mem handle for"
2611 			    " descriptor %d", vdc->instance, i);
2612 			return (status);
2613 		}
2614 		vdc->local_dring[i].is_free = B_TRUE;
2615 		vdc->local_dring[i].dep = dep;
2616 	}
2617 
2618 	/* Initialize the starting index */
2619 	vdc->dring_curr_idx = 0;
2620 
2621 	return (status);
2622 }
2623 
2624 /*
2625  * Function:
2626  *	vdc_destroy_descriptor_ring()
2627  *
2628  * Description:
2629  *
2630  * Arguments:
2631  *	vdc	- soft state pointer for this instance of the device driver.
2632  *
2633  * Return Code:
2634  *	None
2635  */
2636 static void
2637 vdc_destroy_descriptor_ring(vdc_t *vdc)
2638 {
2639 	vdc_local_desc_t	*ldep = NULL;	/* Local Dring Entry Pointer */
2640 	ldc_mem_handle_t	mhdl = NULL;
2641 	ldc_mem_info_t		minfo;
2642 	int			status = -1;
2643 	int			i;	/* loop */
2644 
2645 	ASSERT(vdc != NULL);
2646 	ASSERT(mutex_owned(&vdc->lock));
2647 
2648 	DMSG(vdc, 0, "[%d] Entered\n", vdc->instance);
2649 
2650 	if (vdc->initialized & VDC_DRING_ENTRY) {
2651 		DMSG(vdc, 0,
2652 		    "[%d] Removing Local DRing entries\n", vdc->instance);
2653 		for (i = 0; i < vdc->dring_len; i++) {
2654 			ldep = &vdc->local_dring[i];
2655 			mhdl = ldep->desc_mhdl;
2656 
2657 			if (mhdl == NULL)
2658 				continue;
2659 
2660 			if ((status = ldc_mem_info(mhdl, &minfo)) != 0) {
2661 				DMSG(vdc, 0,
2662 				    "ldc_mem_info returned an error: %d\n",
2663 				    status);
2664 
2665 				/*
2666 				 * This must mean that the mem handle
2667 				 * is not valid. Clear it out so that
2668 				 * no one tries to use it.
2669 				 */
2670 				ldep->desc_mhdl = NULL;
2671 				continue;
2672 			}
2673 
2674 			if (minfo.status == LDC_BOUND) {
2675 				(void) ldc_mem_unbind_handle(mhdl);
2676 			}
2677 
2678 			(void) ldc_mem_free_handle(mhdl);
2679 
2680 			ldep->desc_mhdl = NULL;
2681 		}
2682 		vdc->initialized &= ~VDC_DRING_ENTRY;
2683 	}
2684 
2685 	if (vdc->initialized & VDC_DRING_LOCAL) {
2686 		DMSG(vdc, 0, "[%d] Freeing Local DRing\n", vdc->instance);
2687 		kmem_free(vdc->local_dring,
2688 		    vdc->dring_len * sizeof (vdc_local_desc_t));
2689 		vdc->initialized &= ~VDC_DRING_LOCAL;
2690 	}
2691 
2692 	if (vdc->initialized & VDC_DRING_BOUND) {
2693 		DMSG(vdc, 0, "[%d] Unbinding DRing\n", vdc->instance);
2694 		status = ldc_mem_dring_unbind(vdc->ldc_dring_hdl);
2695 		if (status == 0) {
2696 			vdc->initialized &= ~VDC_DRING_BOUND;
2697 		} else {
2698 			DMSG(vdc, 0, "[%d] Error %d unbinding DRing %lx",
2699 			    vdc->instance, status, vdc->ldc_dring_hdl);
2700 		}
2701 		kmem_free(vdc->dring_cookie, sizeof (ldc_mem_cookie_t));
2702 	}
2703 
2704 	if (vdc->initialized & VDC_DRING_INIT) {
2705 		DMSG(vdc, 0, "[%d] Destroying DRing\n", vdc->instance);
2706 		status = ldc_mem_dring_destroy(vdc->ldc_dring_hdl);
2707 		if (status == 0) {
2708 			vdc->ldc_dring_hdl = NULL;
2709 			bzero(&vdc->dring_mem_info, sizeof (ldc_mem_info_t));
2710 			vdc->initialized &= ~VDC_DRING_INIT;
2711 		} else {
2712 			DMSG(vdc, 0, "[%d] Error %d destroying DRing (%lx)",
2713 			    vdc->instance, status, vdc->ldc_dring_hdl);
2714 		}
2715 	}
2716 }
2717 
2718 /*
2719  * Function:
2720  *	vdc_map_to_shared_dring()
2721  *
2722  * Description:
2723  *	Copy contents of the local descriptor to the shared
2724  *	memory descriptor.
2725  *
2726  * Arguments:
2727  *	vdcp	- soft state pointer for this instance of the device driver.
2728  *	idx	- descriptor ring index
2729  *
2730  * Return Code:
2731  *	None
2732  */
2733 static int
2734 vdc_map_to_shared_dring(vdc_t *vdcp, int idx)
2735 {
2736 	vdc_local_desc_t	*ldep;
2737 	vd_dring_entry_t	*dep;
2738 	int			rv;
2739 
2740 	ldep = &(vdcp->local_dring[idx]);
2741 
2742 	/* for now leave in the old pop_mem_hdl stuff */
2743 	if (ldep->nbytes > 0) {
2744 		rv = vdc_populate_mem_hdl(vdcp, ldep);
2745 		if (rv) {
2746 			DMSG(vdcp, 0, "[%d] Cannot populate mem handle\n",
2747 			    vdcp->instance);
2748 			return (rv);
2749 		}
2750 	}
2751 
2752 	/*
2753 	 * fill in the data details into the DRing
2754 	 */
2755 	dep = ldep->dep;
2756 	ASSERT(dep != NULL);
2757 
2758 	dep->payload.req_id = VDC_GET_NEXT_REQ_ID(vdcp);
2759 	dep->payload.operation = ldep->operation;
2760 	dep->payload.addr = ldep->offset;
2761 	dep->payload.nbytes = ldep->nbytes;
2762 	dep->payload.status = (uint32_t)-1;	/* vds will set valid value */
2763 	dep->payload.slice = ldep->slice;
2764 	dep->hdr.dstate = VIO_DESC_READY;
2765 	dep->hdr.ack = 1;		/* request an ACK for every message */
2766 
2767 	return (0);
2768 }
2769 
2770 /*
2771  * Function:
2772  *	vdc_send_request
2773  *
2774  * Description:
2775  *	This routine writes the data to be transmitted to vds into the
2776  *	descriptor, notifies vds that the ring has been updated and
2777  *	then waits for the request to be processed.
2778  *
2779  * Arguments:
2780  *	vdcp	  - the soft state pointer
2781  *	operation - operation we want vds to perform (VD_OP_XXX)
2782  *	addr	  - address of data buf to be read/written.
2783  *	nbytes	  - number of bytes to read/write
2784  *	slice	  - the disk slice this request is for
2785  *	offset	  - relative disk offset
2786  *	cb_type   - type of call - STRATEGY or SYNC
2787  *	cb_arg	  - parameter to be sent to server (depends on VD_OP_XXX type)
2788  *			. mode for ioctl(9e)
2789  *			. LP64 diskaddr_t (block I/O)
2790  *	dir	  - direction of operation (READ/WRITE/BOTH)
2791  *
2792  * Return Codes:
2793  *	0
2794  *	ENXIO
2795  */
2796 static int
2797 vdc_send_request(vdc_t *vdcp, int operation, caddr_t addr,
2798     size_t nbytes, int slice, diskaddr_t offset, int cb_type,
2799     void *cb_arg, vio_desc_direction_t dir)
2800 {
2801 	int	rv = 0;
2802 
2803 	ASSERT(vdcp != NULL);
2804 	ASSERT(slice == VD_SLICE_NONE || slice < V_NUMPAR);
2805 
2806 	mutex_enter(&vdcp->lock);
2807 
2808 	/*
2809 	 * If this is a block read/write operation we update the I/O statistics
2810 	 * to indicate that the request is being put on the waitq to be
2811 	 * serviced.
2812 	 *
2813 	 * We do it here (a common routine for both synchronous and strategy
2814 	 * calls) for performance reasons - we are already holding vdc->lock
2815 	 * so there is no extra locking overhead. We would have to explicitly
2816 	 * grab the 'lock' mutex to update the stats if we were to do this
2817 	 * higher up the stack in vdc_strategy() et. al.
2818 	 */
2819 	if ((operation == VD_OP_BREAD) || (operation == VD_OP_BWRITE)) {
2820 		DTRACE_IO1(start, buf_t *, cb_arg);
2821 		VD_KSTAT_WAITQ_ENTER(vdcp);
2822 	}
2823 
2824 	do {
2825 		while (vdcp->state != VDC_STATE_RUNNING) {
2826 
2827 			/* return error if detaching */
2828 			if (vdcp->state == VDC_STATE_DETACH) {
2829 				rv = ENXIO;
2830 				goto done;
2831 			}
2832 
2833 			/* fail request if connection timeout is reached */
2834 			if (vdcp->ctimeout_reached) {
2835 				rv = EIO;
2836 				goto done;
2837 			}
2838 
2839 			/*
2840 			 * If we are panicking and the disk is not ready then
2841 			 * we can't send any request because we can't complete
2842 			 * the handshake now.
2843 			 */
2844 			if (ddi_in_panic()) {
2845 				rv = EIO;
2846 				goto done;
2847 			}
2848 
2849 			cv_wait(&vdcp->running_cv, &vdcp->lock);
2850 		}
2851 
2852 	} while (vdc_populate_descriptor(vdcp, operation, addr,
2853 	    nbytes, slice, offset, cb_type, cb_arg, dir));
2854 
2855 done:
2856 	/*
2857 	 * If this is a block read/write we update the I/O statistics kstat
2858 	 * to indicate that this request has been placed on the queue for
2859 	 * processing (i.e sent to the vDisk server) - iostat(1M) will
2860 	 * report the time waiting for the vDisk server under the %b column
2861 	 * In the case of an error we simply take it off the wait queue.
2862 	 */
2863 	if ((operation == VD_OP_BREAD) || (operation == VD_OP_BWRITE)) {
2864 		if (rv == 0) {
2865 			VD_KSTAT_WAITQ_TO_RUNQ(vdcp);
2866 			DTRACE_PROBE1(send, buf_t *, cb_arg);
2867 		} else {
2868 			VD_UPDATE_ERR_STATS(vdcp, vd_transerrs);
2869 			VD_KSTAT_WAITQ_EXIT(vdcp);
2870 			DTRACE_IO1(done, buf_t *, cb_arg);
2871 		}
2872 	}
2873 
2874 	mutex_exit(&vdcp->lock);
2875 
2876 	return (rv);
2877 }
2878 
2879 
2880 /*
2881  * Function:
2882  *	vdc_populate_descriptor
2883  *
2884  * Description:
2885  *	This routine writes the data to be transmitted to vds into the
2886  *	descriptor, notifies vds that the ring has been updated and
2887  *	then waits for the request to be processed.
2888  *
2889  * Arguments:
2890  *	vdcp	  - the soft state pointer
2891  *	operation - operation we want vds to perform (VD_OP_XXX)
2892  *	addr	  - address of data buf to be read/written.
2893  *	nbytes	  - number of bytes to read/write
2894  *	slice	  - the disk slice this request is for
2895  *	offset	  - relative disk offset
2896  *	cb_type   - type of call - STRATEGY or SYNC
2897  *	cb_arg	  - parameter to be sent to server (depends on VD_OP_XXX type)
2898  *			. mode for ioctl(9e)
2899  *			. LP64 diskaddr_t (block I/O)
2900  *	dir	  - direction of operation (READ/WRITE/BOTH)
2901  *
2902  * Return Codes:
2903  *	0
2904  *	EAGAIN
2905  *	ECONNRESET
2906  *	ENXIO
2907  */
2908 static int
2909 vdc_populate_descriptor(vdc_t *vdcp, int operation, caddr_t addr,
2910     size_t nbytes, int slice, diskaddr_t offset, int cb_type,
2911     void *cb_arg, vio_desc_direction_t dir)
2912 {
2913 	vdc_local_desc_t	*local_dep = NULL; /* Local Dring Pointer */
2914 	int			idx;		/* Index of DRing entry used */
2915 	int			next_idx;
2916 	vio_dring_msg_t		dmsg;
2917 	size_t			msglen;
2918 	int			rv;
2919 
2920 	ASSERT(MUTEX_HELD(&vdcp->lock));
2921 	vdcp->threads_pending++;
2922 loop:
2923 	DMSG(vdcp, 2, ": dring_curr_idx = %d\n", vdcp->dring_curr_idx);
2924 
2925 	/* Get next available D-Ring entry */
2926 	idx = vdcp->dring_curr_idx;
2927 	local_dep = &(vdcp->local_dring[idx]);
2928 
2929 	if (!local_dep->is_free) {
2930 		DMSG(vdcp, 2, "[%d]: dring full - waiting for space\n",
2931 		    vdcp->instance);
2932 		cv_wait(&vdcp->dring_free_cv, &vdcp->lock);
2933 		if (vdcp->state == VDC_STATE_RUNNING ||
2934 		    vdcp->state == VDC_STATE_HANDLE_PENDING) {
2935 			goto loop;
2936 		}
2937 		vdcp->threads_pending--;
2938 		return (ECONNRESET);
2939 	}
2940 
2941 	next_idx = idx + 1;
2942 	if (next_idx >= vdcp->dring_len)
2943 		next_idx = 0;
2944 	vdcp->dring_curr_idx = next_idx;
2945 
2946 	ASSERT(local_dep->is_free);
2947 
2948 	local_dep->operation = operation;
2949 	local_dep->addr = addr;
2950 	local_dep->nbytes = nbytes;
2951 	local_dep->slice = slice;
2952 	local_dep->offset = offset;
2953 	local_dep->cb_type = cb_type;
2954 	local_dep->cb_arg = cb_arg;
2955 	local_dep->dir = dir;
2956 
2957 	local_dep->is_free = B_FALSE;
2958 
2959 	rv = vdc_map_to_shared_dring(vdcp, idx);
2960 	if (rv) {
2961 		DMSG(vdcp, 0, "[%d]: cannot bind memory - waiting ..\n",
2962 		    vdcp->instance);
2963 		/* free the descriptor */
2964 		local_dep->is_free = B_TRUE;
2965 		vdcp->dring_curr_idx = idx;
2966 		cv_wait(&vdcp->membind_cv, &vdcp->lock);
2967 		if (vdcp->state == VDC_STATE_RUNNING ||
2968 		    vdcp->state == VDC_STATE_HANDLE_PENDING) {
2969 			goto loop;
2970 		}
2971 		vdcp->threads_pending--;
2972 		return (ECONNRESET);
2973 	}
2974 
2975 	/*
2976 	 * Send a msg with the DRing details to vds
2977 	 */
2978 	VIO_INIT_DRING_DATA_TAG(dmsg);
2979 	VDC_INIT_DRING_DATA_MSG_IDS(dmsg, vdcp);
2980 	dmsg.dring_ident = vdcp->dring_ident;
2981 	dmsg.start_idx = idx;
2982 	dmsg.end_idx = idx;
2983 	vdcp->seq_num++;
2984 
2985 	DTRACE_PROBE2(populate, int, vdcp->instance,
2986 	    vdc_local_desc_t *, local_dep);
2987 	DMSG(vdcp, 2, "ident=0x%lx, st=%u, end=%u, seq=%ld\n",
2988 	    vdcp->dring_ident, dmsg.start_idx, dmsg.end_idx, dmsg.seq_num);
2989 
2990 	/*
2991 	 * note we're still holding the lock here to
2992 	 * make sure the message goes out in order !!!...
2993 	 */
2994 	msglen = sizeof (dmsg);
2995 	rv = vdc_send(vdcp, (caddr_t)&dmsg, &msglen);
2996 	switch (rv) {
2997 	case ECONNRESET:
2998 		/*
2999 		 * vdc_send initiates the reset on failure.
3000 		 * Since the transaction has already been put
3001 		 * on the local dring, it will automatically get
3002 		 * retried when the channel is reset. Given that,
3003 		 * it is ok to just return success even though the
3004 		 * send failed.
3005 		 */
3006 		rv = 0;
3007 		break;
3008 
3009 	case 0: /* EOK */
3010 		DMSG(vdcp, 1, "sent via LDC: rv=%d\n", rv);
3011 		break;
3012 
3013 	default:
3014 		goto cleanup_and_exit;
3015 	}
3016 
3017 	vdcp->threads_pending--;
3018 	return (rv);
3019 
3020 cleanup_and_exit:
3021 	DMSG(vdcp, 0, "unexpected error, rv=%d\n", rv);
3022 	return (ENXIO);
3023 }
3024 
3025 /*
3026  * Function:
3027  *	vdc_do_sync_op
3028  *
3029  * Description:
3030  * 	Wrapper around vdc_populate_descriptor that blocks until the
3031  * 	response to the message is available.
3032  *
3033  * Arguments:
3034  *	vdcp	  - the soft state pointer
3035  *	operation - operation we want vds to perform (VD_OP_XXX)
3036  *	addr	  - address of data buf to be read/written.
3037  *	nbytes	  - number of bytes to read/write
3038  *	slice	  - the disk slice this request is for
3039  *	offset	  - relative disk offset
3040  *	cb_type   - type of call - STRATEGY or SYNC
3041  *	cb_arg	  - parameter to be sent to server (depends on VD_OP_XXX type)
3042  *			. mode for ioctl(9e)
3043  *			. LP64 diskaddr_t (block I/O)
3044  *	dir	  - direction of operation (READ/WRITE/BOTH)
3045  *	rconflict - check for reservation conflict in case of failure
3046  *
3047  * rconflict should be set to B_TRUE by most callers. Callers invoking the
3048  * VD_OP_SCSICMD operation can set rconflict to B_FALSE if they check the
3049  * result of a successful operation with vd_scsi_status().
3050  *
3051  * Return Codes:
3052  *	0
3053  *	EAGAIN
3054  *	EFAULT
3055  *	ENXIO
3056  *	EIO
3057  */
3058 static int
3059 vdc_do_sync_op(vdc_t *vdcp, int operation, caddr_t addr, size_t nbytes,
3060     int slice, diskaddr_t offset, int cb_type, void *cb_arg,
3061     vio_desc_direction_t dir, boolean_t rconflict)
3062 {
3063 	int status;
3064 	vdc_io_t *vio;
3065 	boolean_t check_resv_conflict = B_FALSE;
3066 
3067 	ASSERT(cb_type == CB_SYNC);
3068 
3069 	/*
3070 	 * Grab the lock, if blocked wait until the server
3071 	 * response causes us to wake up again.
3072 	 */
3073 	mutex_enter(&vdcp->lock);
3074 	vdcp->sync_op_cnt++;
3075 	while (vdcp->sync_op_blocked && vdcp->state != VDC_STATE_DETACH)
3076 		cv_wait(&vdcp->sync_blocked_cv, &vdcp->lock);
3077 
3078 	if (vdcp->state == VDC_STATE_DETACH) {
3079 		cv_broadcast(&vdcp->sync_blocked_cv);
3080 		vdcp->sync_op_cnt--;
3081 		mutex_exit(&vdcp->lock);
3082 		return (ENXIO);
3083 	}
3084 
3085 	/* now block anyone other thread entering after us */
3086 	vdcp->sync_op_blocked = B_TRUE;
3087 	vdcp->sync_op_pending = B_TRUE;
3088 	mutex_exit(&vdcp->lock);
3089 
3090 	status = vdc_send_request(vdcp, operation, addr,
3091 	    nbytes, slice, offset, cb_type, cb_arg, dir);
3092 
3093 	mutex_enter(&vdcp->lock);
3094 
3095 	if (status != 0) {
3096 		vdcp->sync_op_pending = B_FALSE;
3097 	} else {
3098 		/*
3099 		 * block until our transaction completes.
3100 		 * Also anyone else waiting also gets to go next.
3101 		 */
3102 		while (vdcp->sync_op_pending && vdcp->state != VDC_STATE_DETACH)
3103 			cv_wait(&vdcp->sync_pending_cv, &vdcp->lock);
3104 
3105 		DMSG(vdcp, 2, ": operation returned %d\n",
3106 		    vdcp->sync_op_status);
3107 		if (vdcp->state == VDC_STATE_DETACH) {
3108 			vdcp->sync_op_pending = B_FALSE;
3109 			status = ENXIO;
3110 		} else {
3111 			status = vdcp->sync_op_status;
3112 			if (status != 0 && vdcp->failfast_interval != 0) {
3113 				/*
3114 				 * Operation has failed and failfast is enabled.
3115 				 * We need to check if the failure is due to a
3116 				 * reservation conflict if this was requested.
3117 				 */
3118 				check_resv_conflict = rconflict;
3119 			}
3120 
3121 		}
3122 	}
3123 
3124 	vdcp->sync_op_status = 0;
3125 	vdcp->sync_op_blocked = B_FALSE;
3126 	vdcp->sync_op_cnt--;
3127 
3128 	/* signal the next waiting thread */
3129 	cv_signal(&vdcp->sync_blocked_cv);
3130 
3131 	/*
3132 	 * We have to check for reservation conflict after unblocking sync
3133 	 * operations because some sync operations will be used to do this
3134 	 * check.
3135 	 */
3136 	if (check_resv_conflict) {
3137 		vio = vdc_failfast_io_queue(vdcp, NULL);
3138 		while (vio->vio_qtime != 0)
3139 			cv_wait(&vdcp->failfast_io_cv, &vdcp->lock);
3140 		kmem_free(vio, sizeof (vdc_io_t));
3141 	}
3142 
3143 	mutex_exit(&vdcp->lock);
3144 
3145 	return (status);
3146 }
3147 
3148 
3149 /*
3150  * Function:
3151  *	vdc_drain_response()
3152  *
3153  * Description:
3154  * 	When a guest is panicking, the completion of requests needs to be
3155  * 	handled differently because interrupts are disabled and vdc
3156  * 	will not get messages. We have to poll for the messages instead.
3157  *
3158  *	Note: since we don't have a buf_t available we cannot implement
3159  *	the io:::done DTrace probe in this specific case.
3160  *
3161  * Arguments:
3162  *	vdc	- soft state pointer for this instance of the device driver.
3163  *
3164  * Return Code:
3165  *	0	- Success
3166  */
3167 static int
3168 vdc_drain_response(vdc_t *vdc)
3169 {
3170 	int 			rv, idx, retries;
3171 	size_t			msglen;
3172 	vdc_local_desc_t 	*ldep = NULL;	/* Local Dring Entry Pointer */
3173 	vio_dring_msg_t		dmsg;
3174 
3175 	mutex_enter(&vdc->lock);
3176 
3177 	retries = 0;
3178 	for (;;) {
3179 		msglen = sizeof (dmsg);
3180 		rv = ldc_read(vdc->ldc_handle, (caddr_t)&dmsg, &msglen);
3181 		if (rv) {
3182 			rv = EINVAL;
3183 			break;
3184 		}
3185 
3186 		/*
3187 		 * if there are no packets wait and check again
3188 		 */
3189 		if ((rv == 0) && (msglen == 0)) {
3190 			if (retries++ > vdc_dump_retries) {
3191 				rv = EAGAIN;
3192 				break;
3193 			}
3194 
3195 			drv_usecwait(vdc_usec_timeout_dump);
3196 			continue;
3197 		}
3198 
3199 		/*
3200 		 * Ignore all messages that are not ACKs/NACKs to
3201 		 * DRing requests.
3202 		 */
3203 		if ((dmsg.tag.vio_msgtype != VIO_TYPE_DATA) ||
3204 		    (dmsg.tag.vio_subtype_env != VIO_DRING_DATA)) {
3205 			DMSG(vdc, 0, "discard pkt: type=%d sub=%d env=%d\n",
3206 			    dmsg.tag.vio_msgtype,
3207 			    dmsg.tag.vio_subtype,
3208 			    dmsg.tag.vio_subtype_env);
3209 			continue;
3210 		}
3211 
3212 		/*
3213 		 * set the appropriate return value for the current request.
3214 		 */
3215 		switch (dmsg.tag.vio_subtype) {
3216 		case VIO_SUBTYPE_ACK:
3217 			rv = 0;
3218 			break;
3219 		case VIO_SUBTYPE_NACK:
3220 			rv = EAGAIN;
3221 			break;
3222 		default:
3223 			continue;
3224 		}
3225 
3226 		idx = dmsg.start_idx;
3227 		if (idx >= vdc->dring_len) {
3228 			DMSG(vdc, 0, "[%d] Bogus ack data : start %d\n",
3229 			    vdc->instance, idx);
3230 			continue;
3231 		}
3232 		ldep = &vdc->local_dring[idx];
3233 		if (ldep->dep->hdr.dstate != VIO_DESC_DONE) {
3234 			DMSG(vdc, 0, "[%d] Entry @ %d - state !DONE %d\n",
3235 			    vdc->instance, idx, ldep->dep->hdr.dstate);
3236 			continue;
3237 		}
3238 
3239 		DMSG(vdc, 1, "[%d] Depopulating idx=%d state=%d\n",
3240 		    vdc->instance, idx, ldep->dep->hdr.dstate);
3241 
3242 		rv = vdc_depopulate_descriptor(vdc, idx);
3243 		if (rv) {
3244 			DMSG(vdc, 0,
3245 			    "[%d] Entry @ %d - depopulate failed ..\n",
3246 			    vdc->instance, idx);
3247 		}
3248 
3249 		/* if this is the last descriptor - break out of loop */
3250 		if ((idx + 1) % vdc->dring_len == vdc->dring_curr_idx)
3251 			break;
3252 	}
3253 
3254 	mutex_exit(&vdc->lock);
3255 	DMSG(vdc, 0, "End idx=%d\n", idx);
3256 
3257 	return (rv);
3258 }
3259 
3260 
3261 /*
3262  * Function:
3263  *	vdc_depopulate_descriptor()
3264  *
3265  * Description:
3266  *
3267  * Arguments:
3268  *	vdc	- soft state pointer for this instance of the device driver.
3269  *	idx	- Index of the Descriptor Ring entry being modified
3270  *
3271  * Return Code:
3272  *	0	- Success
3273  */
3274 static int
3275 vdc_depopulate_descriptor(vdc_t *vdc, uint_t idx)
3276 {
3277 	vd_dring_entry_t *dep = NULL;		/* Dring Entry Pointer */
3278 	vdc_local_desc_t *ldep = NULL;		/* Local Dring Entry Pointer */
3279 	int		status = ENXIO;
3280 	int		rv = 0;
3281 
3282 	ASSERT(vdc != NULL);
3283 	ASSERT(idx < vdc->dring_len);
3284 	ldep = &vdc->local_dring[idx];
3285 	ASSERT(ldep != NULL);
3286 	ASSERT(MUTEX_HELD(&vdc->lock));
3287 
3288 	DTRACE_PROBE2(depopulate, int, vdc->instance, vdc_local_desc_t *, ldep);
3289 	DMSG(vdc, 2, ": idx = %d\n", idx);
3290 
3291 	dep = ldep->dep;
3292 	ASSERT(dep != NULL);
3293 	ASSERT((dep->hdr.dstate == VIO_DESC_DONE) ||
3294 	    (dep->payload.status == ECANCELED));
3295 
3296 	VDC_MARK_DRING_ENTRY_FREE(vdc, idx);
3297 
3298 	ldep->is_free = B_TRUE;
3299 	status = dep->payload.status;
3300 	DMSG(vdc, 2, ": is_free = %d : status = %d\n", ldep->is_free, status);
3301 
3302 	/*
3303 	 * If no buffers were used to transfer information to the server when
3304 	 * populating the descriptor then no memory handles need to be unbound
3305 	 * and we can return now.
3306 	 */
3307 	if (ldep->nbytes == 0) {
3308 		cv_signal(&vdc->dring_free_cv);
3309 		return (status);
3310 	}
3311 
3312 	/*
3313 	 * If the upper layer passed in a misaligned address we copied the
3314 	 * data into an aligned buffer before sending it to LDC - we now
3315 	 * copy it back to the original buffer.
3316 	 */
3317 	if (ldep->align_addr) {
3318 		ASSERT(ldep->addr != NULL);
3319 
3320 		if (dep->payload.nbytes > 0)
3321 			bcopy(ldep->align_addr, ldep->addr,
3322 			    dep->payload.nbytes);
3323 		kmem_free(ldep->align_addr,
3324 		    sizeof (caddr_t) * P2ROUNDUP(ldep->nbytes, 8));
3325 		ldep->align_addr = NULL;
3326 	}
3327 
3328 	rv = ldc_mem_unbind_handle(ldep->desc_mhdl);
3329 	if (rv != 0) {
3330 		DMSG(vdc, 0, "?[%d] unbind mhdl 0x%lx @ idx %d failed (%d)",
3331 		    vdc->instance, ldep->desc_mhdl, idx, rv);
3332 		/*
3333 		 * The error returned by the vDisk server is more informative
3334 		 * and thus has a higher priority but if it isn't set we ensure
3335 		 * that this function returns an error.
3336 		 */
3337 		if (status == 0)
3338 			status = EINVAL;
3339 	}
3340 
3341 	cv_signal(&vdc->membind_cv);
3342 	cv_signal(&vdc->dring_free_cv);
3343 
3344 	return (status);
3345 }
3346 
3347 /*
3348  * Function:
3349  *	vdc_populate_mem_hdl()
3350  *
3351  * Description:
3352  *
3353  * Arguments:
3354  *	vdc	- soft state pointer for this instance of the device driver.
3355  *	idx	- Index of the Descriptor Ring entry being modified
3356  *	addr	- virtual address being mapped in
3357  *	nybtes	- number of bytes in 'addr'
3358  *	operation - the vDisk operation being performed (VD_OP_xxx)
3359  *
3360  * Return Code:
3361  *	0	- Success
3362  */
3363 static int
3364 vdc_populate_mem_hdl(vdc_t *vdcp, vdc_local_desc_t *ldep)
3365 {
3366 	vd_dring_entry_t	*dep = NULL;
3367 	ldc_mem_handle_t	mhdl;
3368 	caddr_t			vaddr;
3369 	size_t			nbytes;
3370 	uint8_t			perm = LDC_MEM_RW;
3371 	uint8_t			maptype;
3372 	int			rv = 0;
3373 	int			i;
3374 
3375 	ASSERT(vdcp != NULL);
3376 
3377 	dep = ldep->dep;
3378 	mhdl = ldep->desc_mhdl;
3379 
3380 	switch (ldep->dir) {
3381 	case VIO_read_dir:
3382 		perm = LDC_MEM_W;
3383 		break;
3384 
3385 	case VIO_write_dir:
3386 		perm = LDC_MEM_R;
3387 		break;
3388 
3389 	case VIO_both_dir:
3390 		perm = LDC_MEM_RW;
3391 		break;
3392 
3393 	default:
3394 		ASSERT(0);	/* catch bad programming in vdc */
3395 	}
3396 
3397 	/*
3398 	 * LDC expects any addresses passed in to be 8-byte aligned. We need
3399 	 * to copy the contents of any misaligned buffers to a newly allocated
3400 	 * buffer and bind it instead (and copy the the contents back to the
3401 	 * original buffer passed in when depopulating the descriptor)
3402 	 */
3403 	vaddr = ldep->addr;
3404 	nbytes = ldep->nbytes;
3405 	if (((uint64_t)vaddr & 0x7) != 0) {
3406 		ASSERT(ldep->align_addr == NULL);
3407 		ldep->align_addr =
3408 		    kmem_alloc(sizeof (caddr_t) *
3409 		    P2ROUNDUP(nbytes, 8), KM_SLEEP);
3410 		DMSG(vdcp, 0, "[%d] Misaligned address %p reallocating "
3411 		    "(buf=%p nb=%ld op=%d)\n",
3412 		    vdcp->instance, (void *)vaddr, (void *)ldep->align_addr,
3413 		    nbytes, ldep->operation);
3414 		if (perm != LDC_MEM_W)
3415 			bcopy(vaddr, ldep->align_addr, nbytes);
3416 		vaddr = ldep->align_addr;
3417 	}
3418 
3419 	maptype = LDC_IO_MAP|LDC_SHADOW_MAP|LDC_DIRECT_MAP;
3420 	rv = ldc_mem_bind_handle(mhdl, vaddr, P2ROUNDUP(nbytes, 8),
3421 	    maptype, perm, &dep->payload.cookie[0], &dep->payload.ncookies);
3422 	DMSG(vdcp, 2, "[%d] bound mem handle; ncookies=%d\n",
3423 	    vdcp->instance, dep->payload.ncookies);
3424 	if (rv != 0) {
3425 		DMSG(vdcp, 0, "[%d] Failed to bind LDC memory handle "
3426 		    "(mhdl=%p, buf=%p, err=%d)\n",
3427 		    vdcp->instance, (void *)mhdl, (void *)vaddr, rv);
3428 		if (ldep->align_addr) {
3429 			kmem_free(ldep->align_addr,
3430 			    sizeof (caddr_t) * P2ROUNDUP(nbytes, 8));
3431 			ldep->align_addr = NULL;
3432 		}
3433 		return (EAGAIN);
3434 	}
3435 
3436 	/*
3437 	 * Get the other cookies (if any).
3438 	 */
3439 	for (i = 1; i < dep->payload.ncookies; i++) {
3440 		rv = ldc_mem_nextcookie(mhdl, &dep->payload.cookie[i]);
3441 		if (rv != 0) {
3442 			(void) ldc_mem_unbind_handle(mhdl);
3443 			DMSG(vdcp, 0, "?[%d] Failed to get next cookie "
3444 			    "(mhdl=%lx cnum=%d), err=%d",
3445 			    vdcp->instance, mhdl, i, rv);
3446 			if (ldep->align_addr) {
3447 				kmem_free(ldep->align_addr,
3448 				    sizeof (caddr_t) * ldep->nbytes);
3449 				ldep->align_addr = NULL;
3450 			}
3451 			return (EAGAIN);
3452 		}
3453 	}
3454 
3455 	return (rv);
3456 }
3457 
3458 /*
3459  * Interrupt handlers for messages from LDC
3460  */
3461 
3462 /*
3463  * Function:
3464  *	vdc_handle_cb()
3465  *
3466  * Description:
3467  *
3468  * Arguments:
3469  *	event	- Type of event (LDC_EVT_xxx) that triggered the callback
3470  *	arg	- soft state pointer for this instance of the device driver.
3471  *
3472  * Return Code:
3473  *	0	- Success
3474  */
3475 static uint_t
3476 vdc_handle_cb(uint64_t event, caddr_t arg)
3477 {
3478 	ldc_status_t	ldc_state;
3479 	int		rv = 0;
3480 
3481 	vdc_t	*vdc = (vdc_t *)(void *)arg;
3482 
3483 	ASSERT(vdc != NULL);
3484 
3485 	DMSG(vdc, 1, "evt=%lx seqID=%ld\n", event, vdc->seq_num);
3486 
3487 	/*
3488 	 * Depending on the type of event that triggered this callback,
3489 	 * we modify the handshake state or read the data.
3490 	 *
3491 	 * NOTE: not done as a switch() as event could be triggered by
3492 	 * a state change and a read request. Also the ordering	of the
3493 	 * check for the event types is deliberate.
3494 	 */
3495 	if (event & LDC_EVT_UP) {
3496 		DMSG(vdc, 0, "[%d] Received LDC_EVT_UP\n", vdc->instance);
3497 
3498 		mutex_enter(&vdc->lock);
3499 
3500 		/* get LDC state */
3501 		rv = ldc_status(vdc->ldc_handle, &ldc_state);
3502 		if (rv != 0) {
3503 			DMSG(vdc, 0, "[%d] Couldn't get LDC status %d",
3504 			    vdc->instance, rv);
3505 			return (LDC_SUCCESS);
3506 		}
3507 		if (vdc->ldc_state != LDC_UP && ldc_state == LDC_UP) {
3508 			/*
3509 			 * Reset the transaction sequence numbers when
3510 			 * LDC comes up. We then kick off the handshake
3511 			 * negotiation with the vDisk server.
3512 			 */
3513 			vdc->seq_num = 1;
3514 			vdc->seq_num_reply = 0;
3515 			vdc->ldc_state = ldc_state;
3516 			cv_signal(&vdc->initwait_cv);
3517 		}
3518 
3519 		mutex_exit(&vdc->lock);
3520 	}
3521 
3522 	if (event & LDC_EVT_READ) {
3523 		DMSG(vdc, 1, "[%d] Received LDC_EVT_READ\n", vdc->instance);
3524 		mutex_enter(&vdc->read_lock);
3525 		cv_signal(&vdc->read_cv);
3526 		vdc->read_state = VDC_READ_PENDING;
3527 		mutex_exit(&vdc->read_lock);
3528 
3529 		/* that's all we have to do - no need to handle DOWN/RESET */
3530 		return (LDC_SUCCESS);
3531 	}
3532 
3533 	if (event & (LDC_EVT_RESET|LDC_EVT_DOWN)) {
3534 
3535 		DMSG(vdc, 0, "[%d] Received LDC RESET event\n", vdc->instance);
3536 
3537 		mutex_enter(&vdc->lock);
3538 		/*
3539 		 * Need to wake up any readers so they will
3540 		 * detect that a reset has occurred.
3541 		 */
3542 		mutex_enter(&vdc->read_lock);
3543 		if ((vdc->read_state == VDC_READ_WAITING) ||
3544 		    (vdc->read_state == VDC_READ_RESET))
3545 			cv_signal(&vdc->read_cv);
3546 		vdc->read_state = VDC_READ_RESET;
3547 		mutex_exit(&vdc->read_lock);
3548 
3549 		/* wake up any threads waiting for connection to come up */
3550 		if (vdc->state == VDC_STATE_INIT_WAITING) {
3551 			vdc->state = VDC_STATE_RESETTING;
3552 			cv_signal(&vdc->initwait_cv);
3553 		}
3554 
3555 		mutex_exit(&vdc->lock);
3556 	}
3557 
3558 	if (event & ~(LDC_EVT_UP | LDC_EVT_RESET | LDC_EVT_DOWN | LDC_EVT_READ))
3559 		DMSG(vdc, 0, "![%d] Unexpected LDC event (%lx) received",
3560 		    vdc->instance, event);
3561 
3562 	return (LDC_SUCCESS);
3563 }
3564 
3565 /*
3566  * Function:
3567  *	vdc_wait_for_response()
3568  *
3569  * Description:
3570  *	Block waiting for a response from the server. If there is
3571  *	no data the thread block on the read_cv that is signalled
3572  *	by the callback when an EVT_READ occurs.
3573  *
3574  * Arguments:
3575  *	vdcp	- soft state pointer for this instance of the device driver.
3576  *
3577  * Return Code:
3578  *	0	- Success
3579  */
3580 static int
3581 vdc_wait_for_response(vdc_t *vdcp, vio_msg_t *msgp)
3582 {
3583 	size_t		nbytes = sizeof (*msgp);
3584 	int		status;
3585 
3586 	ASSERT(vdcp != NULL);
3587 
3588 	DMSG(vdcp, 1, "[%d] Entered\n", vdcp->instance);
3589 
3590 	status = vdc_recv(vdcp, msgp, &nbytes);
3591 	DMSG(vdcp, 3, "vdc_read() done.. status=0x%x size=0x%x\n",
3592 	    status, (int)nbytes);
3593 	if (status) {
3594 		DMSG(vdcp, 0, "?[%d] Error %d reading LDC msg\n",
3595 		    vdcp->instance, status);
3596 		return (status);
3597 	}
3598 
3599 	if (nbytes < sizeof (vio_msg_tag_t)) {
3600 		DMSG(vdcp, 0, "?[%d] Expect %lu bytes; recv'd %lu\n",
3601 		    vdcp->instance, sizeof (vio_msg_tag_t), nbytes);
3602 		return (ENOMSG);
3603 	}
3604 
3605 	DMSG(vdcp, 2, "[%d] (%x/%x/%x)\n", vdcp->instance,
3606 	    msgp->tag.vio_msgtype,
3607 	    msgp->tag.vio_subtype,
3608 	    msgp->tag.vio_subtype_env);
3609 
3610 	/*
3611 	 * Verify the Session ID of the message
3612 	 *
3613 	 * Every message after the Version has been negotiated should
3614 	 * have the correct session ID set.
3615 	 */
3616 	if ((msgp->tag.vio_sid != vdcp->session_id) &&
3617 	    (msgp->tag.vio_subtype_env != VIO_VER_INFO)) {
3618 		DMSG(vdcp, 0, "[%d] Invalid SID: received 0x%x, "
3619 		    "expected 0x%lx [seq num %lx @ %d]",
3620 		    vdcp->instance, msgp->tag.vio_sid,
3621 		    vdcp->session_id,
3622 		    ((vio_dring_msg_t *)msgp)->seq_num,
3623 		    ((vio_dring_msg_t *)msgp)->start_idx);
3624 		return (ENOMSG);
3625 	}
3626 	return (0);
3627 }
3628 
3629 
3630 /*
3631  * Function:
3632  *	vdc_resubmit_backup_dring()
3633  *
3634  * Description:
3635  *	Resubmit each descriptor in the backed up dring to
3636  * 	vDisk server. The Dring was backed up during connection
3637  *	reset.
3638  *
3639  * Arguments:
3640  *	vdcp	- soft state pointer for this instance of the device driver.
3641  *
3642  * Return Code:
3643  *	0	- Success
3644  */
3645 static int
3646 vdc_resubmit_backup_dring(vdc_t *vdcp)
3647 {
3648 	int		processed = 0;
3649 	int		count;
3650 	int		b_idx;
3651 	int		rv = 0;
3652 	int		dring_size;
3653 	int		op;
3654 	vio_msg_t	vio_msg;
3655 	vdc_local_desc_t	*curr_ldep;
3656 
3657 	ASSERT(MUTEX_NOT_HELD(&vdcp->lock));
3658 	ASSERT(vdcp->state == VDC_STATE_HANDLE_PENDING);
3659 
3660 	if (vdcp->local_dring_backup == NULL) {
3661 		/* the pending requests have already been processed */
3662 		return (0);
3663 	}
3664 
3665 	DMSG(vdcp, 1, "restoring pending dring entries (len=%d, tail=%d)\n",
3666 	    vdcp->local_dring_backup_len, vdcp->local_dring_backup_tail);
3667 
3668 	/*
3669 	 * Walk the backup copy of the local descriptor ring and
3670 	 * resubmit all the outstanding transactions.
3671 	 */
3672 	b_idx = vdcp->local_dring_backup_tail;
3673 	for (count = 0; count < vdcp->local_dring_backup_len; count++) {
3674 
3675 		curr_ldep = &(vdcp->local_dring_backup[b_idx]);
3676 
3677 		/* only resubmit outstanding transactions */
3678 		if (!curr_ldep->is_free) {
3679 			/*
3680 			 * If we are retrying a block read/write operation we
3681 			 * need to update the I/O statistics to indicate that
3682 			 * the request is being put back on the waitq to be
3683 			 * serviced (it will have been taken off after the
3684 			 * error was reported).
3685 			 */
3686 			mutex_enter(&vdcp->lock);
3687 			op = curr_ldep->operation;
3688 			if ((op == VD_OP_BREAD) || (op == VD_OP_BWRITE)) {
3689 				DTRACE_IO1(start, buf_t *, curr_ldep->cb_arg);
3690 				VD_KSTAT_WAITQ_ENTER(vdcp);
3691 			}
3692 
3693 			DMSG(vdcp, 1, "resubmitting entry idx=%x\n", b_idx);
3694 			rv = vdc_populate_descriptor(vdcp, op,
3695 			    curr_ldep->addr, curr_ldep->nbytes,
3696 			    curr_ldep->slice, curr_ldep->offset,
3697 			    curr_ldep->cb_type, curr_ldep->cb_arg,
3698 			    curr_ldep->dir);
3699 
3700 			if (rv) {
3701 				if (op == VD_OP_BREAD || op == VD_OP_BWRITE) {
3702 					VD_UPDATE_ERR_STATS(vdcp, vd_transerrs);
3703 					VD_KSTAT_WAITQ_EXIT(vdcp);
3704 					DTRACE_IO1(done, buf_t *,
3705 					    curr_ldep->cb_arg);
3706 				}
3707 				DMSG(vdcp, 1, "[%d] cannot resubmit entry %d\n",
3708 				    vdcp->instance, b_idx);
3709 				mutex_exit(&vdcp->lock);
3710 				goto done;
3711 			}
3712 
3713 			/*
3714 			 * If this is a block read/write we update the I/O
3715 			 * statistics kstat to indicate that the request
3716 			 * has been sent back to the vDisk server and should
3717 			 * now be put on the run queue.
3718 			 */
3719 			if ((op == VD_OP_BREAD) || (op == VD_OP_BWRITE)) {
3720 				DTRACE_PROBE1(send, buf_t *, curr_ldep->cb_arg);
3721 				VD_KSTAT_WAITQ_TO_RUNQ(vdcp);
3722 			}
3723 			mutex_exit(&vdcp->lock);
3724 
3725 			/* Wait for the response message. */
3726 			DMSG(vdcp, 1, "waiting for response to idx=%x\n",
3727 			    b_idx);
3728 			rv = vdc_wait_for_response(vdcp, &vio_msg);
3729 			if (rv) {
3730 				/*
3731 				 * If this is a block read/write we update
3732 				 * the I/O statistics kstat to take it
3733 				 * off the run queue.
3734 				 */
3735 				mutex_enter(&vdcp->lock);
3736 				if (op == VD_OP_BREAD || op == VD_OP_BWRITE) {
3737 					VD_UPDATE_ERR_STATS(vdcp, vd_transerrs);
3738 					VD_KSTAT_RUNQ_EXIT(vdcp);
3739 					DTRACE_IO1(done, buf_t *,
3740 					    curr_ldep->cb_arg);
3741 				}
3742 				DMSG(vdcp, 1, "[%d] wait_for_response "
3743 				    "returned err=%d\n", vdcp->instance,
3744 				    rv);
3745 				mutex_exit(&vdcp->lock);
3746 				goto done;
3747 			}
3748 
3749 			DMSG(vdcp, 1, "processing msg for idx=%x\n", b_idx);
3750 			rv = vdc_process_data_msg(vdcp, &vio_msg);
3751 			if (rv) {
3752 				DMSG(vdcp, 1, "[%d] process_data_msg "
3753 				    "returned err=%d\n", vdcp->instance,
3754 				    rv);
3755 				goto done;
3756 			}
3757 			processed++;
3758 		}
3759 
3760 		/* get the next element to submit */
3761 		if (++b_idx >= vdcp->local_dring_backup_len)
3762 			b_idx = 0;
3763 	}
3764 
3765 	/* all done - now clear up pending dring copy */
3766 	dring_size = vdcp->local_dring_backup_len *
3767 	    sizeof (vdcp->local_dring_backup[0]);
3768 
3769 	(void) kmem_free(vdcp->local_dring_backup, dring_size);
3770 
3771 	vdcp->local_dring_backup = NULL;
3772 
3773 done:
3774 	DTRACE_PROBE2(processed, int, processed, vdc_t *, vdcp);
3775 
3776 	return (rv);
3777 }
3778 
3779 /*
3780  * Function:
3781  *	vdc_cancel_backup_dring
3782  *
3783  * Description:
3784  *	Cancel each descriptor in the backed up dring to vDisk server.
3785  *	The Dring was backed up during connection reset.
3786  *
3787  * Arguments:
3788  *	vdcp	- soft state pointer for this instance of the device driver.
3789  *
3790  * Return Code:
3791  *	None
3792  */
3793 void
3794 vdc_cancel_backup_dring(vdc_t *vdcp)
3795 {
3796 	vdc_local_desc_t *ldep;
3797 	struct buf 	*bufp;
3798 	int		count;
3799 	int		b_idx;
3800 	int		dring_size;
3801 	int		cancelled = 0;
3802 
3803 	ASSERT(MUTEX_HELD(&vdcp->lock));
3804 	ASSERT(vdcp->state == VDC_STATE_INIT ||
3805 	    vdcp->state == VDC_STATE_INIT_WAITING ||
3806 	    vdcp->state == VDC_STATE_NEGOTIATE ||
3807 	    vdcp->state == VDC_STATE_RESETTING);
3808 
3809 	if (vdcp->local_dring_backup == NULL) {
3810 		/* the pending requests have already been processed */
3811 		return;
3812 	}
3813 
3814 	DMSG(vdcp, 1, "cancelling pending dring entries (len=%d, tail=%d)\n",
3815 	    vdcp->local_dring_backup_len, vdcp->local_dring_backup_tail);
3816 
3817 	/*
3818 	 * Walk the backup copy of the local descriptor ring and
3819 	 * cancel all the outstanding transactions.
3820 	 */
3821 	b_idx = vdcp->local_dring_backup_tail;
3822 	for (count = 0; count < vdcp->local_dring_backup_len; count++) {
3823 
3824 		ldep = &(vdcp->local_dring_backup[b_idx]);
3825 
3826 		/* only cancel outstanding transactions */
3827 		if (!ldep->is_free) {
3828 
3829 			DMSG(vdcp, 1, "cancelling entry idx=%x\n", b_idx);
3830 			cancelled++;
3831 
3832 			/*
3833 			 * All requests have already been cleared from the
3834 			 * local descriptor ring and the LDC channel has been
3835 			 * reset so we will never get any reply for these
3836 			 * requests. Now we just have to notify threads waiting
3837 			 * for replies that the request has failed.
3838 			 */
3839 			switch (ldep->cb_type) {
3840 			case CB_SYNC:
3841 				ASSERT(vdcp->sync_op_pending);
3842 				vdcp->sync_op_status = EIO;
3843 				vdcp->sync_op_pending = B_FALSE;
3844 				cv_signal(&vdcp->sync_pending_cv);
3845 				break;
3846 
3847 			case CB_STRATEGY:
3848 				bufp = ldep->cb_arg;
3849 				ASSERT(bufp != NULL);
3850 				bufp->b_resid = bufp->b_bcount;
3851 				VD_UPDATE_ERR_STATS(vdcp, vd_softerrs);
3852 				VD_KSTAT_RUNQ_EXIT(vdcp);
3853 				DTRACE_IO1(done, buf_t *, bufp);
3854 				bioerror(bufp, EIO);
3855 				biodone(bufp);
3856 				break;
3857 
3858 			default:
3859 				ASSERT(0);
3860 			}
3861 
3862 		}
3863 
3864 		/* get the next element to cancel */
3865 		if (++b_idx >= vdcp->local_dring_backup_len)
3866 			b_idx = 0;
3867 	}
3868 
3869 	/* all done - now clear up pending dring copy */
3870 	dring_size = vdcp->local_dring_backup_len *
3871 	    sizeof (vdcp->local_dring_backup[0]);
3872 
3873 	(void) kmem_free(vdcp->local_dring_backup, dring_size);
3874 
3875 	vdcp->local_dring_backup = NULL;
3876 
3877 	DTRACE_PROBE2(cancelled, int, cancelled, vdc_t *, vdcp);
3878 }
3879 
3880 /*
3881  * Function:
3882  *	vdc_connection_timeout
3883  *
3884  * Description:
3885  *	This function is invoked if the timeout set to establish the connection
3886  *	with vds expires. This will happen if we spend too much time in the
3887  *	VDC_STATE_INIT_WAITING or VDC_STATE_NEGOTIATE states. Then we will
3888  *	cancel any pending request and mark them as failed.
3889  *
3890  *	If the timeout does not expire, it will be cancelled when we reach the
3891  *	VDC_STATE_HANDLE_PENDING or VDC_STATE_RESETTING state. This function can
3892  *	be invoked while we are in the VDC_STATE_HANDLE_PENDING or
3893  *	VDC_STATE_RESETTING state in which case we do nothing because the
3894  *	timeout is being cancelled.
3895  *
3896  * Arguments:
3897  *	arg	- argument of the timeout function actually a soft state
3898  *		  pointer for the instance of the device driver.
3899  *
3900  * Return Code:
3901  *	None
3902  */
3903 void
3904 vdc_connection_timeout(void *arg)
3905 {
3906 	vdc_t 		*vdcp = (vdc_t *)arg;
3907 
3908 	mutex_enter(&vdcp->lock);
3909 
3910 	if (vdcp->state == VDC_STATE_HANDLE_PENDING ||
3911 	    vdcp->state == VDC_STATE_DETACH) {
3912 		/*
3913 		 * The connection has just been re-established or
3914 		 * we are detaching.
3915 		 */
3916 		vdcp->ctimeout_reached = B_FALSE;
3917 		mutex_exit(&vdcp->lock);
3918 		return;
3919 	}
3920 
3921 	vdcp->ctimeout_reached = B_TRUE;
3922 
3923 	/* notify requests waiting for sending */
3924 	cv_broadcast(&vdcp->running_cv);
3925 
3926 	/* cancel requests waiting for a result */
3927 	vdc_cancel_backup_dring(vdcp);
3928 
3929 	mutex_exit(&vdcp->lock);
3930 
3931 	cmn_err(CE_NOTE, "[%d] connection to service domain timeout",
3932 	    vdcp->instance);
3933 }
3934 
3935 /*
3936  * Function:
3937  *	vdc_backup_local_dring()
3938  *
3939  * Description:
3940  *	Backup the current dring in the event of a reset. The Dring
3941  *	transactions will be resubmitted to the server when the
3942  *	connection is restored.
3943  *
3944  * Arguments:
3945  *	vdcp	- soft state pointer for this instance of the device driver.
3946  *
3947  * Return Code:
3948  *	NONE
3949  */
3950 static void
3951 vdc_backup_local_dring(vdc_t *vdcp)
3952 {
3953 	int dring_size;
3954 
3955 	ASSERT(MUTEX_HELD(&vdcp->lock));
3956 	ASSERT(vdcp->state == VDC_STATE_RESETTING);
3957 
3958 	/*
3959 	 * If the backup dring is stil around, it means
3960 	 * that the last restore did not complete. However,
3961 	 * since we never got back into the running state,
3962 	 * the backup copy we have is still valid.
3963 	 */
3964 	if (vdcp->local_dring_backup != NULL) {
3965 		DMSG(vdcp, 1, "reusing local descriptor ring backup "
3966 		    "(len=%d, tail=%d)\n", vdcp->local_dring_backup_len,
3967 		    vdcp->local_dring_backup_tail);
3968 		return;
3969 	}
3970 
3971 	/*
3972 	 * The backup dring can be NULL and the local dring may not be
3973 	 * initialized. This can happen if we had a reset while establishing
3974 	 * a new connection but after the connection has timed out. In that
3975 	 * case the backup dring is NULL because the requests have been
3976 	 * cancelled and the request occured before the local dring is
3977 	 * initialized.
3978 	 */
3979 	if (!(vdcp->initialized & VDC_DRING_LOCAL))
3980 		return;
3981 
3982 	DMSG(vdcp, 1, "backing up the local descriptor ring (len=%d, "
3983 	    "tail=%d)\n", vdcp->dring_len, vdcp->dring_curr_idx);
3984 
3985 	dring_size = vdcp->dring_len * sizeof (vdcp->local_dring[0]);
3986 
3987 	vdcp->local_dring_backup = kmem_alloc(dring_size, KM_SLEEP);
3988 	bcopy(vdcp->local_dring, vdcp->local_dring_backup, dring_size);
3989 
3990 	vdcp->local_dring_backup_tail = vdcp->dring_curr_idx;
3991 	vdcp->local_dring_backup_len = vdcp->dring_len;
3992 }
3993 
3994 /* -------------------------------------------------------------------------- */
3995 
3996 /*
3997  * The following functions process the incoming messages from vds
3998  */
3999 
4000 /*
4001  * Function:
4002  *      vdc_process_msg_thread()
4003  *
4004  * Description:
4005  *
4006  *	Main VDC message processing thread. Each vDisk instance
4007  * 	consists of a copy of this thread. This thread triggers
4008  * 	all the handshakes and data exchange with the server. It
4009  * 	also handles all channel resets
4010  *
4011  * Arguments:
4012  *      vdc     - soft state pointer for this instance of the device driver.
4013  *
4014  * Return Code:
4015  *      None
4016  */
4017 static void
4018 vdc_process_msg_thread(vdc_t *vdcp)
4019 {
4020 	int	status;
4021 	int	ctimeout;
4022 	timeout_id_t tmid = 0;
4023 
4024 	mutex_enter(&vdcp->lock);
4025 
4026 	for (;;) {
4027 
4028 #define	Q(_s)	(vdcp->state == _s) ? #_s :
4029 		DMSG(vdcp, 3, "state = %d (%s)\n", vdcp->state,
4030 		    Q(VDC_STATE_INIT)
4031 		    Q(VDC_STATE_INIT_WAITING)
4032 		    Q(VDC_STATE_NEGOTIATE)
4033 		    Q(VDC_STATE_HANDLE_PENDING)
4034 		    Q(VDC_STATE_RUNNING)
4035 		    Q(VDC_STATE_RESETTING)
4036 		    Q(VDC_STATE_DETACH)
4037 		    "UNKNOWN");
4038 
4039 		switch (vdcp->state) {
4040 		case VDC_STATE_INIT:
4041 
4042 			/*
4043 			 * If requested, start a timeout to check if the
4044 			 * connection with vds is established in the
4045 			 * specified delay. If the timeout expires, we
4046 			 * will cancel any pending request.
4047 			 *
4048 			 * If some reset have occurred while establishing
4049 			 * the connection, we already have a timeout armed
4050 			 * and in that case we don't need to arm a new one.
4051 			 */
4052 			ctimeout = (vdc_timeout != 0)?
4053 			    vdc_timeout : vdcp->ctimeout;
4054 
4055 			if (ctimeout != 0 && tmid == 0) {
4056 				tmid = timeout(vdc_connection_timeout, vdcp,
4057 				    ctimeout * drv_usectohz(1000000));
4058 			}
4059 
4060 			/* Check if have re-initializing repeatedly */
4061 			if (vdcp->hshake_cnt++ > vdc_hshake_retries &&
4062 			    vdcp->lifecycle != VDC_LC_ONLINE) {
4063 				cmn_err(CE_NOTE, "[%d] disk access failed.\n",
4064 				    vdcp->instance);
4065 				vdcp->state = VDC_STATE_DETACH;
4066 				break;
4067 			}
4068 
4069 			/* Bring up connection with vds via LDC */
4070 			status = vdc_start_ldc_connection(vdcp);
4071 			if (status == EINVAL) {
4072 				DMSG(vdcp, 0, "[%d] Could not start LDC",
4073 				    vdcp->instance);
4074 				vdcp->state = VDC_STATE_DETACH;
4075 			} else {
4076 				vdcp->state = VDC_STATE_INIT_WAITING;
4077 			}
4078 			break;
4079 
4080 		case VDC_STATE_INIT_WAITING:
4081 
4082 			/*
4083 			 * Let the callback event move us on
4084 			 * when channel is open to server
4085 			 */
4086 			while (vdcp->ldc_state != LDC_UP) {
4087 				cv_wait(&vdcp->initwait_cv, &vdcp->lock);
4088 				if (vdcp->state != VDC_STATE_INIT_WAITING) {
4089 					DMSG(vdcp, 0,
4090 				"state moved to %d out from under us...\n",
4091 					    vdcp->state);
4092 
4093 					break;
4094 				}
4095 			}
4096 			if (vdcp->state == VDC_STATE_INIT_WAITING &&
4097 			    vdcp->ldc_state == LDC_UP) {
4098 				vdcp->state = VDC_STATE_NEGOTIATE;
4099 			}
4100 			break;
4101 
4102 		case VDC_STATE_NEGOTIATE:
4103 			switch (status = vdc_ver_negotiation(vdcp)) {
4104 			case 0:
4105 				break;
4106 			default:
4107 				DMSG(vdcp, 0, "ver negotiate failed (%d)..\n",
4108 				    status);
4109 				goto reset;
4110 			}
4111 
4112 			switch (status = vdc_attr_negotiation(vdcp)) {
4113 			case 0:
4114 				break;
4115 			default:
4116 				DMSG(vdcp, 0, "attr negotiate failed (%d)..\n",
4117 				    status);
4118 				goto reset;
4119 			}
4120 
4121 			switch (status = vdc_dring_negotiation(vdcp)) {
4122 			case 0:
4123 				break;
4124 			default:
4125 				DMSG(vdcp, 0, "dring negotiate failed (%d)..\n",
4126 				    status);
4127 				goto reset;
4128 			}
4129 
4130 			switch (status = vdc_rdx_exchange(vdcp)) {
4131 			case 0:
4132 				vdcp->state = VDC_STATE_HANDLE_PENDING;
4133 				goto done;
4134 			default:
4135 				DMSG(vdcp, 0, "RDX xchg failed ..(%d)\n",
4136 				    status);
4137 				goto reset;
4138 			}
4139 reset:
4140 			DMSG(vdcp, 0, "negotiation failed: resetting (%d)\n",
4141 			    status);
4142 			vdcp->state = VDC_STATE_RESETTING;
4143 			vdcp->self_reset = B_TRUE;
4144 done:
4145 			DMSG(vdcp, 0, "negotiation complete (state=0x%x)...\n",
4146 			    vdcp->state);
4147 			break;
4148 
4149 		case VDC_STATE_HANDLE_PENDING:
4150 
4151 			if (vdcp->ctimeout_reached) {
4152 				/*
4153 				 * The connection timeout had been reached so
4154 				 * pending requests have been cancelled. Now
4155 				 * that the connection is back we can reset
4156 				 * the timeout.
4157 				 */
4158 				ASSERT(vdcp->local_dring_backup == NULL);
4159 				ASSERT(tmid != 0);
4160 				tmid = 0;
4161 				vdcp->ctimeout_reached = B_FALSE;
4162 				vdcp->state = VDC_STATE_RUNNING;
4163 				DMSG(vdcp, 0, "[%d] connection to service "
4164 				    "domain is up", vdcp->instance);
4165 				break;
4166 			}
4167 
4168 			mutex_exit(&vdcp->lock);
4169 			if (tmid != 0) {
4170 				(void) untimeout(tmid);
4171 				tmid = 0;
4172 			}
4173 			status = vdc_resubmit_backup_dring(vdcp);
4174 			mutex_enter(&vdcp->lock);
4175 
4176 			if (status)
4177 				vdcp->state = VDC_STATE_RESETTING;
4178 			else
4179 				vdcp->state = VDC_STATE_RUNNING;
4180 
4181 			break;
4182 
4183 		/* enter running state */
4184 		case VDC_STATE_RUNNING:
4185 			/*
4186 			 * Signal anyone waiting for the connection
4187 			 * to come on line.
4188 			 */
4189 			vdcp->hshake_cnt = 0;
4190 			cv_broadcast(&vdcp->running_cv);
4191 
4192 			/* failfast has to been checked after reset */
4193 			cv_signal(&vdcp->failfast_cv);
4194 
4195 			/* ownership is lost during reset */
4196 			if (vdcp->ownership & VDC_OWNERSHIP_WANTED)
4197 				vdcp->ownership |= VDC_OWNERSHIP_RESET;
4198 			cv_signal(&vdcp->ownership_cv);
4199 
4200 			mutex_exit(&vdcp->lock);
4201 
4202 			for (;;) {
4203 				vio_msg_t msg;
4204 				status = vdc_wait_for_response(vdcp, &msg);
4205 				if (status) break;
4206 
4207 				DMSG(vdcp, 1, "[%d] new pkt(s) available\n",
4208 				    vdcp->instance);
4209 				status = vdc_process_data_msg(vdcp, &msg);
4210 				if (status) {
4211 					DMSG(vdcp, 1, "[%d] process_data_msg "
4212 					    "returned err=%d\n", vdcp->instance,
4213 					    status);
4214 					break;
4215 				}
4216 
4217 			}
4218 
4219 			mutex_enter(&vdcp->lock);
4220 
4221 			vdcp->state = VDC_STATE_RESETTING;
4222 			vdcp->self_reset = B_TRUE;
4223 			break;
4224 
4225 		case VDC_STATE_RESETTING:
4226 			/*
4227 			 * When we reach this state, we either come from the
4228 			 * VDC_STATE_RUNNING state and we can have pending
4229 			 * request but no timeout is armed; or we come from
4230 			 * the VDC_STATE_INIT_WAITING, VDC_NEGOTIATE or
4231 			 * VDC_HANDLE_PENDING state and there is no pending
4232 			 * request or pending requests have already been copied
4233 			 * into the backup dring. So we can safely keep the
4234 			 * connection timeout armed while we are in this state.
4235 			 */
4236 
4237 			DMSG(vdcp, 0, "Initiating channel reset "
4238 			    "(pending = %d)\n", (int)vdcp->threads_pending);
4239 
4240 			if (vdcp->self_reset) {
4241 				DMSG(vdcp, 0,
4242 				    "[%d] calling stop_ldc_connection.\n",
4243 				    vdcp->instance);
4244 				status = vdc_stop_ldc_connection(vdcp);
4245 				vdcp->self_reset = B_FALSE;
4246 			}
4247 
4248 			/*
4249 			 * Wait for all threads currently waiting
4250 			 * for a free dring entry to use.
4251 			 */
4252 			while (vdcp->threads_pending) {
4253 				cv_broadcast(&vdcp->membind_cv);
4254 				cv_broadcast(&vdcp->dring_free_cv);
4255 				mutex_exit(&vdcp->lock);
4256 				/* give the waiters enough time to wake up */
4257 				delay(vdc_hz_min_ldc_delay);
4258 				mutex_enter(&vdcp->lock);
4259 			}
4260 
4261 			ASSERT(vdcp->threads_pending == 0);
4262 
4263 			/* Sanity check that no thread is receiving */
4264 			ASSERT(vdcp->read_state != VDC_READ_WAITING);
4265 
4266 			vdcp->read_state = VDC_READ_IDLE;
4267 
4268 			vdc_backup_local_dring(vdcp);
4269 
4270 			/* cleanup the old d-ring */
4271 			vdc_destroy_descriptor_ring(vdcp);
4272 
4273 			/* go and start again */
4274 			vdcp->state = VDC_STATE_INIT;
4275 
4276 			break;
4277 
4278 		case VDC_STATE_DETACH:
4279 			DMSG(vdcp, 0, "[%d] Reset thread exit cleanup ..\n",
4280 			    vdcp->instance);
4281 
4282 			/* cancel any pending timeout */
4283 			mutex_exit(&vdcp->lock);
4284 			if (tmid != 0) {
4285 				(void) untimeout(tmid);
4286 				tmid = 0;
4287 			}
4288 			mutex_enter(&vdcp->lock);
4289 
4290 			/*
4291 			 * Signal anyone waiting for connection
4292 			 * to come online
4293 			 */
4294 			cv_broadcast(&vdcp->running_cv);
4295 
4296 			while (vdcp->sync_op_pending) {
4297 				cv_signal(&vdcp->sync_pending_cv);
4298 				cv_signal(&vdcp->sync_blocked_cv);
4299 				mutex_exit(&vdcp->lock);
4300 				/* give the waiters enough time to wake up */
4301 				delay(vdc_hz_min_ldc_delay);
4302 				mutex_enter(&vdcp->lock);
4303 			}
4304 
4305 			mutex_exit(&vdcp->lock);
4306 
4307 			DMSG(vdcp, 0, "[%d] Msg processing thread exiting ..\n",
4308 			    vdcp->instance);
4309 			thread_exit();
4310 			break;
4311 		}
4312 	}
4313 }
4314 
4315 
4316 /*
4317  * Function:
4318  *	vdc_process_data_msg()
4319  *
4320  * Description:
4321  *	This function is called by the message processing thread each time
4322  *	a message with a msgtype of VIO_TYPE_DATA is received. It will either
4323  *	be an ACK or NACK from vds[1] which vdc handles as follows.
4324  *		ACK	- wake up the waiting thread
4325  *		NACK	- resend any messages necessary
4326  *
4327  *	[1] Although the message format allows it, vds should not send a
4328  *	    VIO_SUBTYPE_INFO message to vdc asking it to read data; if for
4329  *	    some bizarre reason it does, vdc will reset the connection.
4330  *
4331  * Arguments:
4332  *	vdc	- soft state pointer for this instance of the device driver.
4333  *	msg	- the LDC message sent by vds
4334  *
4335  * Return Code:
4336  *	0	- Success.
4337  *	> 0	- error value returned by LDC
4338  */
4339 static int
4340 vdc_process_data_msg(vdc_t *vdcp, vio_msg_t *msg)
4341 {
4342 	int			status = 0;
4343 	vio_dring_msg_t		*dring_msg;
4344 	vdc_local_desc_t	*ldep = NULL;
4345 	int			start, end;
4346 	int			idx;
4347 	int			op;
4348 
4349 	dring_msg = (vio_dring_msg_t *)msg;
4350 
4351 	ASSERT(msg->tag.vio_msgtype == VIO_TYPE_DATA);
4352 	ASSERT(vdcp != NULL);
4353 
4354 	mutex_enter(&vdcp->lock);
4355 
4356 	/*
4357 	 * Check to see if the message has bogus data
4358 	 */
4359 	idx = start = dring_msg->start_idx;
4360 	end = dring_msg->end_idx;
4361 	if ((start >= vdcp->dring_len) ||
4362 	    (end >= vdcp->dring_len) || (end < -1)) {
4363 		/*
4364 		 * Update the I/O statistics to indicate that an error ocurred.
4365 		 * No need to update the wait/run queues as no specific read or
4366 		 * write request is being completed in response to this 'msg'.
4367 		 */
4368 		VD_UPDATE_ERR_STATS(vdcp, vd_softerrs);
4369 		DMSG(vdcp, 0, "[%d] Bogus ACK data : start %d, end %d\n",
4370 		    vdcp->instance, start, end);
4371 		mutex_exit(&vdcp->lock);
4372 		return (EINVAL);
4373 	}
4374 
4375 	/*
4376 	 * Verify that the sequence number is what vdc expects.
4377 	 */
4378 	switch (vdc_verify_seq_num(vdcp, dring_msg)) {
4379 	case VDC_SEQ_NUM_TODO:
4380 		break;	/* keep processing this message */
4381 	case VDC_SEQ_NUM_SKIP:
4382 		mutex_exit(&vdcp->lock);
4383 		return (0);
4384 	case VDC_SEQ_NUM_INVALID:
4385 		/*
4386 		 * Update the I/O statistics to indicate that an error ocurred.
4387 		 * No need to update the wait/run queues as no specific read or
4388 		 * write request is being completed in response to this 'msg'.
4389 		 */
4390 		VD_UPDATE_ERR_STATS(vdcp, vd_softerrs);
4391 		DMSG(vdcp, 0, "[%d] invalid seqno\n", vdcp->instance);
4392 		mutex_exit(&vdcp->lock);
4393 		return (ENXIO);
4394 	}
4395 
4396 	if (msg->tag.vio_subtype == VIO_SUBTYPE_NACK) {
4397 		/*
4398 		 * Update the I/O statistics to indicate that an error ocurred.
4399 		 *
4400 		 * We need to update the run queue if a read or write request
4401 		 * is being NACKed - otherwise there will appear to be an
4402 		 * indefinite outstanding request and statistics reported by
4403 		 * iostat(1M) will be incorrect. The transaction will be
4404 		 * resubmitted from the backup DRing following the reset
4405 		 * and the wait/run queues will be entered again.
4406 		 */
4407 		ldep = &vdcp->local_dring[idx];
4408 		op = ldep->operation;
4409 		if ((op == VD_OP_BREAD) || (op == VD_OP_BWRITE)) {
4410 			DTRACE_IO1(done, buf_t *, ldep->cb_arg);
4411 			VD_KSTAT_RUNQ_EXIT(vdcp);
4412 		}
4413 		VD_UPDATE_ERR_STATS(vdcp, vd_softerrs);
4414 		VDC_DUMP_DRING_MSG(dring_msg);
4415 		DMSG(vdcp, 0, "[%d] DATA NACK\n", vdcp->instance);
4416 		mutex_exit(&vdcp->lock);
4417 		return (EIO);
4418 
4419 	} else if (msg->tag.vio_subtype == VIO_SUBTYPE_INFO) {
4420 		/*
4421 		 * Update the I/O statistics to indicate that an error occurred.
4422 		 * No need to update the wait/run queues as no specific read or
4423 		 * write request is being completed in response to this 'msg'.
4424 		 */
4425 		VD_UPDATE_ERR_STATS(vdcp, vd_protoerrs);
4426 		mutex_exit(&vdcp->lock);
4427 		return (EPROTO);
4428 	}
4429 
4430 	DMSG(vdcp, 1, ": start %d end %d\n", start, end);
4431 	ASSERT(start == end);
4432 
4433 	ldep = &vdcp->local_dring[idx];
4434 
4435 	DMSG(vdcp, 1, ": state 0x%x - cb_type 0x%x\n",
4436 	    ldep->dep->hdr.dstate, ldep->cb_type);
4437 
4438 	if (ldep->dep->hdr.dstate == VIO_DESC_DONE) {
4439 		struct buf *bufp;
4440 
4441 		switch (ldep->cb_type) {
4442 		case CB_SYNC:
4443 			ASSERT(vdcp->sync_op_pending);
4444 
4445 			status = vdc_depopulate_descriptor(vdcp, idx);
4446 			vdcp->sync_op_status = status;
4447 			vdcp->sync_op_pending = B_FALSE;
4448 			cv_signal(&vdcp->sync_pending_cv);
4449 			break;
4450 
4451 		case CB_STRATEGY:
4452 			bufp = ldep->cb_arg;
4453 			ASSERT(bufp != NULL);
4454 			bufp->b_resid =
4455 			    bufp->b_bcount - ldep->dep->payload.nbytes;
4456 			status = ldep->dep->payload.status; /* Future:ntoh */
4457 			if (status != 0) {
4458 				DMSG(vdcp, 1, "strategy status=%d\n", status);
4459 				VD_UPDATE_ERR_STATS(vdcp, vd_softerrs);
4460 				bioerror(bufp, status);
4461 			}
4462 
4463 			(void) vdc_depopulate_descriptor(vdcp, idx);
4464 
4465 			DMSG(vdcp, 1,
4466 			    "strategy complete req=%ld bytes resp=%ld bytes\n",
4467 			    bufp->b_bcount, ldep->dep->payload.nbytes);
4468 
4469 			if (status != 0 && vdcp->failfast_interval != 0) {
4470 				/*
4471 				 * The I/O has failed and failfast is enabled.
4472 				 * We need the failfast thread to check if the
4473 				 * failure is due to a reservation conflict.
4474 				 */
4475 				(void) vdc_failfast_io_queue(vdcp, bufp);
4476 			} else {
4477 				if (status == 0) {
4478 					op = (bufp->b_flags & B_READ) ?
4479 					    VD_OP_BREAD : VD_OP_BWRITE;
4480 					VD_UPDATE_IO_STATS(vdcp, op,
4481 					    ldep->dep->payload.nbytes);
4482 				}
4483 				VD_KSTAT_RUNQ_EXIT(vdcp);
4484 				DTRACE_IO1(done, buf_t *, bufp);
4485 				biodone(bufp);
4486 			}
4487 			break;
4488 
4489 		default:
4490 			ASSERT(0);
4491 		}
4492 	}
4493 
4494 	/* let the arrival signal propogate */
4495 	mutex_exit(&vdcp->lock);
4496 
4497 	/* probe gives the count of how many entries were processed */
4498 	DTRACE_PROBE2(processed, int, 1, vdc_t *, vdcp);
4499 
4500 	return (0);
4501 }
4502 
4503 
4504 /*
4505  * Function:
4506  *	vdc_handle_ver_msg()
4507  *
4508  * Description:
4509  *
4510  * Arguments:
4511  *	vdc	- soft state pointer for this instance of the device driver.
4512  *	ver_msg	- LDC message sent by vDisk server
4513  *
4514  * Return Code:
4515  *	0	- Success
4516  */
4517 static int
4518 vdc_handle_ver_msg(vdc_t *vdc, vio_ver_msg_t *ver_msg)
4519 {
4520 	int status = 0;
4521 
4522 	ASSERT(vdc != NULL);
4523 	ASSERT(mutex_owned(&vdc->lock));
4524 
4525 	if (ver_msg->tag.vio_subtype_env != VIO_VER_INFO) {
4526 		return (EPROTO);
4527 	}
4528 
4529 	if (ver_msg->dev_class != VDEV_DISK_SERVER) {
4530 		return (EINVAL);
4531 	}
4532 
4533 	switch (ver_msg->tag.vio_subtype) {
4534 	case VIO_SUBTYPE_ACK:
4535 		/*
4536 		 * We check to see if the version returned is indeed supported
4537 		 * (The server may have also adjusted the minor number downwards
4538 		 * and if so 'ver_msg' will contain the actual version agreed)
4539 		 */
4540 		if (vdc_is_supported_version(ver_msg)) {
4541 			vdc->ver.major = ver_msg->ver_major;
4542 			vdc->ver.minor = ver_msg->ver_minor;
4543 			ASSERT(vdc->ver.major > 0);
4544 		} else {
4545 			status = EPROTO;
4546 		}
4547 		break;
4548 
4549 	case VIO_SUBTYPE_NACK:
4550 		/*
4551 		 * call vdc_is_supported_version() which will return the next
4552 		 * supported version (if any) in 'ver_msg'
4553 		 */
4554 		(void) vdc_is_supported_version(ver_msg);
4555 		if (ver_msg->ver_major > 0) {
4556 			size_t len = sizeof (*ver_msg);
4557 
4558 			ASSERT(vdc->ver.major > 0);
4559 
4560 			/* reset the necessary fields and resend */
4561 			ver_msg->tag.vio_subtype = VIO_SUBTYPE_INFO;
4562 			ver_msg->dev_class = VDEV_DISK;
4563 
4564 			status = vdc_send(vdc, (caddr_t)ver_msg, &len);
4565 			DMSG(vdc, 0, "[%d] Resend VER info (LDC status = %d)\n",
4566 			    vdc->instance, status);
4567 			if (len != sizeof (*ver_msg))
4568 				status = EBADMSG;
4569 		} else {
4570 			DMSG(vdc, 0, "[%d] No common version with vDisk server",
4571 			    vdc->instance);
4572 			status = ENOTSUP;
4573 		}
4574 
4575 		break;
4576 	case VIO_SUBTYPE_INFO:
4577 		/*
4578 		 * Handle the case where vds starts handshake
4579 		 * (for now only vdc is the instigator)
4580 		 */
4581 		status = ENOTSUP;
4582 		break;
4583 
4584 	default:
4585 		status = EINVAL;
4586 		break;
4587 	}
4588 
4589 	return (status);
4590 }
4591 
4592 /*
4593  * Function:
4594  *	vdc_handle_attr_msg()
4595  *
4596  * Description:
4597  *
4598  * Arguments:
4599  *	vdc	- soft state pointer for this instance of the device driver.
4600  *	attr_msg	- LDC message sent by vDisk server
4601  *
4602  * Return Code:
4603  *	0	- Success
4604  */
4605 static int
4606 vdc_handle_attr_msg(vdc_t *vdc, vd_attr_msg_t *attr_msg)
4607 {
4608 	int status = 0;
4609 
4610 	ASSERT(vdc != NULL);
4611 	ASSERT(mutex_owned(&vdc->lock));
4612 
4613 	if (attr_msg->tag.vio_subtype_env != VIO_ATTR_INFO) {
4614 		return (EPROTO);
4615 	}
4616 
4617 	switch (attr_msg->tag.vio_subtype) {
4618 	case VIO_SUBTYPE_ACK:
4619 		/*
4620 		 * We now verify the attributes sent by vds.
4621 		 */
4622 		if (attr_msg->vdisk_size == 0) {
4623 			DMSG(vdc, 0, "[%d] Invalid disk size from vds",
4624 			    vdc->instance);
4625 			status = EINVAL;
4626 			break;
4627 		}
4628 
4629 		if (attr_msg->max_xfer_sz == 0) {
4630 			DMSG(vdc, 0, "[%d] Invalid transfer size from vds",
4631 			    vdc->instance);
4632 			status = EINVAL;
4633 			break;
4634 		}
4635 
4636 		if (attr_msg->vdisk_size == VD_SIZE_UNKNOWN) {
4637 			DMSG(vdc, 0, "[%d] Unknown disk size from vds",
4638 			    vdc->instance);
4639 			attr_msg->vdisk_size = 0;
4640 		}
4641 
4642 		/*
4643 		 * If the disk size is already set check that it hasn't changed.
4644 		 */
4645 		if ((vdc->vdisk_size != 0) && (attr_msg->vdisk_size != 0) &&
4646 		    (vdc->vdisk_size != attr_msg->vdisk_size)) {
4647 			DMSG(vdc, 0, "[%d] Different disk size from vds "
4648 			    "(old=0x%lx - new=0x%lx", vdc->instance,
4649 			    vdc->vdisk_size, attr_msg->vdisk_size)
4650 			status = EINVAL;
4651 			break;
4652 		}
4653 
4654 		vdc->vdisk_size = attr_msg->vdisk_size;
4655 		vdc->vdisk_type = attr_msg->vdisk_type;
4656 		vdc->operations = attr_msg->operations;
4657 		if (vio_ver_is_supported(vdc->ver, 1, 1))
4658 			vdc->vdisk_media = attr_msg->vdisk_media;
4659 		else
4660 			vdc->vdisk_media = 0;
4661 
4662 		DMSG(vdc, 0, "[%d] max_xfer_sz: sent %lx acked %lx\n",
4663 		    vdc->instance, vdc->max_xfer_sz, attr_msg->max_xfer_sz);
4664 		DMSG(vdc, 0, "[%d] vdisk_block_size: sent %lx acked %x\n",
4665 		    vdc->instance, vdc->block_size,
4666 		    attr_msg->vdisk_block_size);
4667 
4668 		/*
4669 		 * We don't know at compile time what the vDisk server will
4670 		 * think are good values but we apply a large (arbitrary)
4671 		 * upper bound to prevent memory exhaustion in vdc if it was
4672 		 * allocating a DRing based of huge values sent by the server.
4673 		 * We probably will never exceed this except if the message
4674 		 * was garbage.
4675 		 */
4676 		if ((attr_msg->max_xfer_sz * attr_msg->vdisk_block_size) <=
4677 		    (PAGESIZE * DEV_BSIZE)) {
4678 			vdc->max_xfer_sz = attr_msg->max_xfer_sz;
4679 			vdc->block_size = attr_msg->vdisk_block_size;
4680 		} else {
4681 			DMSG(vdc, 0, "[%d] vds block transfer size too big;"
4682 			    " using max supported by vdc", vdc->instance);
4683 		}
4684 
4685 		if ((attr_msg->xfer_mode != VIO_DRING_MODE_V1_0) ||
4686 		    (attr_msg->vdisk_size > INT64_MAX) ||
4687 		    (attr_msg->operations == 0) ||
4688 		    (attr_msg->vdisk_type > VD_DISK_TYPE_DISK)) {
4689 			DMSG(vdc, 0, "[%d] Invalid attributes from vds",
4690 			    vdc->instance);
4691 			status = EINVAL;
4692 			break;
4693 		}
4694 
4695 		/*
4696 		 * Now that we have received all attributes we can create a
4697 		 * fake geometry for the disk.
4698 		 */
4699 		vdc_create_fake_geometry(vdc);
4700 		break;
4701 
4702 	case VIO_SUBTYPE_NACK:
4703 		/*
4704 		 * vds could not handle the attributes we sent so we
4705 		 * stop negotiating.
4706 		 */
4707 		status = EPROTO;
4708 		break;
4709 
4710 	case VIO_SUBTYPE_INFO:
4711 		/*
4712 		 * Handle the case where vds starts the handshake
4713 		 * (for now; vdc is the only supported instigatior)
4714 		 */
4715 		status = ENOTSUP;
4716 		break;
4717 
4718 	default:
4719 		status = ENOTSUP;
4720 		break;
4721 	}
4722 
4723 	return (status);
4724 }
4725 
4726 /*
4727  * Function:
4728  *	vdc_handle_dring_reg_msg()
4729  *
4730  * Description:
4731  *
4732  * Arguments:
4733  *	vdc		- soft state pointer for this instance of the driver.
4734  *	dring_msg	- LDC message sent by vDisk server
4735  *
4736  * Return Code:
4737  *	0	- Success
4738  */
4739 static int
4740 vdc_handle_dring_reg_msg(vdc_t *vdc, vio_dring_reg_msg_t *dring_msg)
4741 {
4742 	int		status = 0;
4743 
4744 	ASSERT(vdc != NULL);
4745 	ASSERT(mutex_owned(&vdc->lock));
4746 
4747 	if (dring_msg->tag.vio_subtype_env != VIO_DRING_REG) {
4748 		return (EPROTO);
4749 	}
4750 
4751 	switch (dring_msg->tag.vio_subtype) {
4752 	case VIO_SUBTYPE_ACK:
4753 		/* save the received dring_ident */
4754 		vdc->dring_ident = dring_msg->dring_ident;
4755 		DMSG(vdc, 0, "[%d] Received dring ident=0x%lx\n",
4756 		    vdc->instance, vdc->dring_ident);
4757 		break;
4758 
4759 	case VIO_SUBTYPE_NACK:
4760 		/*
4761 		 * vds could not handle the DRing info we sent so we
4762 		 * stop negotiating.
4763 		 */
4764 		DMSG(vdc, 0, "[%d] server could not register DRing\n",
4765 		    vdc->instance);
4766 		status = EPROTO;
4767 		break;
4768 
4769 	case VIO_SUBTYPE_INFO:
4770 		/*
4771 		 * Handle the case where vds starts handshake
4772 		 * (for now only vdc is the instigatior)
4773 		 */
4774 		status = ENOTSUP;
4775 		break;
4776 	default:
4777 		status = ENOTSUP;
4778 	}
4779 
4780 	return (status);
4781 }
4782 
4783 /*
4784  * Function:
4785  *	vdc_verify_seq_num()
4786  *
4787  * Description:
4788  *	This functions verifies that the sequence number sent back by the vDisk
4789  *	server with the latest message is what is expected (i.e. it is greater
4790  *	than the last seq num sent by the vDisk server and less than or equal
4791  *	to the last seq num generated by vdc).
4792  *
4793  *	It then checks the request ID to see if any requests need processing
4794  *	in the DRing.
4795  *
4796  * Arguments:
4797  *	vdc		- soft state pointer for this instance of the driver.
4798  *	dring_msg	- pointer to the LDC message sent by vds
4799  *
4800  * Return Code:
4801  *	VDC_SEQ_NUM_TODO	- Message needs to be processed
4802  *	VDC_SEQ_NUM_SKIP	- Message has already been processed
4803  *	VDC_SEQ_NUM_INVALID	- The seq numbers are so out of sync,
4804  *				  vdc cannot deal with them
4805  */
4806 static int
4807 vdc_verify_seq_num(vdc_t *vdc, vio_dring_msg_t *dring_msg)
4808 {
4809 	ASSERT(vdc != NULL);
4810 	ASSERT(dring_msg != NULL);
4811 	ASSERT(mutex_owned(&vdc->lock));
4812 
4813 	/*
4814 	 * Check to see if the messages were responded to in the correct
4815 	 * order by vds.
4816 	 */
4817 	if ((dring_msg->seq_num <= vdc->seq_num_reply) ||
4818 	    (dring_msg->seq_num > vdc->seq_num)) {
4819 		DMSG(vdc, 0, "?[%d] Bogus sequence_number %lu: "
4820 		    "%lu > expected <= %lu (last proc req %lu sent %lu)\n",
4821 		    vdc->instance, dring_msg->seq_num,
4822 		    vdc->seq_num_reply, vdc->seq_num,
4823 		    vdc->req_id_proc, vdc->req_id);
4824 		return (VDC_SEQ_NUM_INVALID);
4825 	}
4826 	vdc->seq_num_reply = dring_msg->seq_num;
4827 
4828 	if (vdc->req_id_proc < vdc->req_id)
4829 		return (VDC_SEQ_NUM_TODO);
4830 	else
4831 		return (VDC_SEQ_NUM_SKIP);
4832 }
4833 
4834 
4835 /*
4836  * Function:
4837  *	vdc_is_supported_version()
4838  *
4839  * Description:
4840  *	This routine checks if the major/minor version numbers specified in
4841  *	'ver_msg' are supported. If not it finds the next version that is
4842  *	in the supported version list 'vdc_version[]' and sets the fields in
4843  *	'ver_msg' to those values
4844  *
4845  * Arguments:
4846  *	ver_msg	- LDC message sent by vDisk server
4847  *
4848  * Return Code:
4849  *	B_TRUE	- Success
4850  *	B_FALSE	- Version not supported
4851  */
4852 static boolean_t
4853 vdc_is_supported_version(vio_ver_msg_t *ver_msg)
4854 {
4855 	int vdc_num_versions = sizeof (vdc_version) / sizeof (vdc_version[0]);
4856 
4857 	for (int i = 0; i < vdc_num_versions; i++) {
4858 		ASSERT(vdc_version[i].major > 0);
4859 		ASSERT((i == 0) ||
4860 		    (vdc_version[i].major < vdc_version[i-1].major));
4861 
4862 		/*
4863 		 * If the major versions match, adjust the minor version, if
4864 		 * necessary, down to the highest value supported by this
4865 		 * client. The server should support all minor versions lower
4866 		 * than the value it sent
4867 		 */
4868 		if (ver_msg->ver_major == vdc_version[i].major) {
4869 			if (ver_msg->ver_minor > vdc_version[i].minor) {
4870 				DMSGX(0,
4871 				    "Adjusting minor version from %u to %u",
4872 				    ver_msg->ver_minor, vdc_version[i].minor);
4873 				ver_msg->ver_minor = vdc_version[i].minor;
4874 			}
4875 			return (B_TRUE);
4876 		}
4877 
4878 		/*
4879 		 * If the message contains a higher major version number, set
4880 		 * the message's major/minor versions to the current values
4881 		 * and return false, so this message will get resent with
4882 		 * these values, and the server will potentially try again
4883 		 * with the same or a lower version
4884 		 */
4885 		if (ver_msg->ver_major > vdc_version[i].major) {
4886 			ver_msg->ver_major = vdc_version[i].major;
4887 			ver_msg->ver_minor = vdc_version[i].minor;
4888 			DMSGX(0, "Suggesting major/minor (0x%x/0x%x)\n",
4889 			    ver_msg->ver_major, ver_msg->ver_minor);
4890 
4891 			return (B_FALSE);
4892 		}
4893 
4894 		/*
4895 		 * Otherwise, the message's major version is less than the
4896 		 * current major version, so continue the loop to the next
4897 		 * (lower) supported version
4898 		 */
4899 	}
4900 
4901 	/*
4902 	 * No common version was found; "ground" the version pair in the
4903 	 * message to terminate negotiation
4904 	 */
4905 	ver_msg->ver_major = 0;
4906 	ver_msg->ver_minor = 0;
4907 
4908 	return (B_FALSE);
4909 }
4910 /* -------------------------------------------------------------------------- */
4911 
4912 /*
4913  * DKIO(7) support
4914  */
4915 
4916 typedef struct vdc_dk_arg {
4917 	struct dk_callback	dkc;
4918 	int			mode;
4919 	dev_t			dev;
4920 	vdc_t			*vdc;
4921 } vdc_dk_arg_t;
4922 
4923 /*
4924  * Function:
4925  * 	vdc_dkio_flush_cb()
4926  *
4927  * Description:
4928  *	This routine is a callback for DKIOCFLUSHWRITECACHE which can be called
4929  *	by kernel code.
4930  *
4931  * Arguments:
4932  *	arg	- a pointer to a vdc_dk_arg_t structure.
4933  */
4934 void
4935 vdc_dkio_flush_cb(void *arg)
4936 {
4937 	struct vdc_dk_arg	*dk_arg = (struct vdc_dk_arg *)arg;
4938 	struct dk_callback	*dkc = NULL;
4939 	vdc_t			*vdc = NULL;
4940 	int			rv;
4941 
4942 	if (dk_arg == NULL) {
4943 		cmn_err(CE_NOTE, "?[Unk] DKIOCFLUSHWRITECACHE arg is NULL\n");
4944 		return;
4945 	}
4946 	dkc = &dk_arg->dkc;
4947 	vdc = dk_arg->vdc;
4948 	ASSERT(vdc != NULL);
4949 
4950 	rv = vdc_do_sync_op(vdc, VD_OP_FLUSH, NULL, 0,
4951 	    VDCPART(dk_arg->dev), 0, CB_SYNC, 0, VIO_both_dir, B_TRUE);
4952 	if (rv != 0) {
4953 		DMSG(vdc, 0, "[%d] DKIOCFLUSHWRITECACHE failed %d : model %x\n",
4954 		    vdc->instance, rv,
4955 		    ddi_model_convert_from(dk_arg->mode & FMODELS));
4956 	}
4957 
4958 	/*
4959 	 * Trigger the call back to notify the caller the the ioctl call has
4960 	 * been completed.
4961 	 */
4962 	if ((dk_arg->mode & FKIOCTL) &&
4963 	    (dkc != NULL) &&
4964 	    (dkc->dkc_callback != NULL)) {
4965 		ASSERT(dkc->dkc_cookie != NULL);
4966 		(*dkc->dkc_callback)(dkc->dkc_cookie, rv);
4967 	}
4968 
4969 	/* Indicate that one less DKIO write flush is outstanding */
4970 	mutex_enter(&vdc->lock);
4971 	vdc->dkio_flush_pending--;
4972 	ASSERT(vdc->dkio_flush_pending >= 0);
4973 	mutex_exit(&vdc->lock);
4974 
4975 	/* free the mem that was allocated when the callback was dispatched */
4976 	kmem_free(arg, sizeof (vdc_dk_arg_t));
4977 }
4978 
4979 /*
4980  * Function:
4981  * 	vdc_dkio_gapart()
4982  *
4983  * Description:
4984  *	This function implements the DKIOCGAPART ioctl.
4985  *
4986  * Arguments:
4987  *	vdc	- soft state pointer
4988  *	arg	- a pointer to a dk_map[NDKMAP] or dk_map32[NDKMAP] structure
4989  *	flag	- ioctl flags
4990  */
4991 static int
4992 vdc_dkio_gapart(vdc_t *vdc, caddr_t arg, int flag)
4993 {
4994 	struct dk_geom *geom;
4995 	struct vtoc *vtoc;
4996 	union {
4997 		struct dk_map map[NDKMAP];
4998 		struct dk_map32 map32[NDKMAP];
4999 	} data;
5000 	int i, rv, size;
5001 
5002 	mutex_enter(&vdc->lock);
5003 
5004 	if ((rv = vdc_validate_geometry(vdc)) != 0) {
5005 		mutex_exit(&vdc->lock);
5006 		return (rv);
5007 	}
5008 
5009 	vtoc = vdc->vtoc;
5010 	geom = vdc->geom;
5011 
5012 	if (ddi_model_convert_from(flag & FMODELS) == DDI_MODEL_ILP32) {
5013 
5014 		for (i = 0; i < vtoc->v_nparts; i++) {
5015 			data.map32[i].dkl_cylno = vtoc->v_part[i].p_start /
5016 			    (geom->dkg_nhead * geom->dkg_nsect);
5017 			data.map32[i].dkl_nblk = vtoc->v_part[i].p_size;
5018 		}
5019 		size = NDKMAP * sizeof (struct dk_map32);
5020 
5021 	} else {
5022 
5023 		for (i = 0; i < vtoc->v_nparts; i++) {
5024 			data.map[i].dkl_cylno = vtoc->v_part[i].p_start /
5025 			    (geom->dkg_nhead * geom->dkg_nsect);
5026 			data.map[i].dkl_nblk = vtoc->v_part[i].p_size;
5027 		}
5028 		size = NDKMAP * sizeof (struct dk_map);
5029 
5030 	}
5031 
5032 	mutex_exit(&vdc->lock);
5033 
5034 	if (ddi_copyout(&data, arg, size, flag) != 0)
5035 		return (EFAULT);
5036 
5037 	return (0);
5038 }
5039 
5040 /*
5041  * Function:
5042  * 	vdc_dkio_partition()
5043  *
5044  * Description:
5045  *	This function implements the DKIOCPARTITION ioctl.
5046  *
5047  * Arguments:
5048  *	vdc	- soft state pointer
5049  *	arg	- a pointer to a struct partition64 structure
5050  *	flag	- ioctl flags
5051  */
5052 static int
5053 vdc_dkio_partition(vdc_t *vdc, caddr_t arg, int flag)
5054 {
5055 	struct partition64 p64;
5056 	efi_gpt_t *gpt;
5057 	efi_gpe_t *gpe;
5058 	vd_efi_dev_t edev;
5059 	uint_t partno;
5060 	int rv;
5061 
5062 	if (ddi_copyin(arg, &p64, sizeof (struct partition64), flag)) {
5063 		return (EFAULT);
5064 	}
5065 
5066 	VD_EFI_DEV_SET(edev, vdc, vd_process_efi_ioctl);
5067 
5068 	if ((rv = vd_efi_alloc_and_read(&edev, &gpt, &gpe)) != 0) {
5069 		return (rv);
5070 	}
5071 
5072 	partno = p64.p_partno;
5073 
5074 	if (partno >= gpt->efi_gpt_NumberOfPartitionEntries) {
5075 		vd_efi_free(&edev, gpt, gpe);
5076 		return (ESRCH);
5077 	}
5078 
5079 	bcopy(&gpe[partno].efi_gpe_PartitionTypeGUID, &p64.p_type,
5080 	    sizeof (struct uuid));
5081 	p64.p_start = gpe[partno].efi_gpe_StartingLBA;
5082 	p64.p_size = gpe[partno].efi_gpe_EndingLBA - p64.p_start + 1;
5083 
5084 	if (ddi_copyout(&p64, arg, sizeof (struct partition64), flag)) {
5085 		vd_efi_free(&edev, gpt, gpe);
5086 		return (EFAULT);
5087 	}
5088 
5089 	vd_efi_free(&edev, gpt, gpe);
5090 	return (0);
5091 }
5092 
5093 /*
5094  * Function:
5095  * 	vdc_dioctl_rwcmd()
5096  *
5097  * Description:
5098  *	This function implements the DIOCTL_RWCMD ioctl. This ioctl is used
5099  *	for DKC_DIRECT disks to read or write at an absolute disk offset.
5100  *
5101  * Arguments:
5102  *	dev	- device
5103  *	arg	- a pointer to a dadkio_rwcmd or dadkio_rwcmd32 structure
5104  *	flag	- ioctl flags
5105  */
5106 static int
5107 vdc_dioctl_rwcmd(dev_t dev, caddr_t arg, int flag)
5108 {
5109 	struct dadkio_rwcmd32 rwcmd32;
5110 	struct dadkio_rwcmd rwcmd;
5111 	struct iovec aiov;
5112 	struct uio auio;
5113 	int rw, status;
5114 	struct buf *buf;
5115 
5116 	if (ddi_model_convert_from(flag & FMODELS) == DDI_MODEL_ILP32) {
5117 		if (ddi_copyin((caddr_t)arg, (caddr_t)&rwcmd32,
5118 		    sizeof (struct dadkio_rwcmd32), flag)) {
5119 			return (EFAULT);
5120 		}
5121 		rwcmd.cmd = rwcmd32.cmd;
5122 		rwcmd.flags = rwcmd32.flags;
5123 		rwcmd.blkaddr = (daddr_t)rwcmd32.blkaddr;
5124 		rwcmd.buflen = rwcmd32.buflen;
5125 		rwcmd.bufaddr = (caddr_t)(uintptr_t)rwcmd32.bufaddr;
5126 	} else {
5127 		if (ddi_copyin((caddr_t)arg, (caddr_t)&rwcmd,
5128 		    sizeof (struct dadkio_rwcmd), flag)) {
5129 			return (EFAULT);
5130 		}
5131 	}
5132 
5133 	switch (rwcmd.cmd) {
5134 	case DADKIO_RWCMD_READ:
5135 		rw = B_READ;
5136 		break;
5137 	case DADKIO_RWCMD_WRITE:
5138 		rw = B_WRITE;
5139 		break;
5140 	default:
5141 		return (EINVAL);
5142 	}
5143 
5144 	bzero((caddr_t)&aiov, sizeof (struct iovec));
5145 	aiov.iov_base   = rwcmd.bufaddr;
5146 	aiov.iov_len    = rwcmd.buflen;
5147 
5148 	bzero((caddr_t)&auio, sizeof (struct uio));
5149 	auio.uio_iov    = &aiov;
5150 	auio.uio_iovcnt = 1;
5151 	auio.uio_loffset = rwcmd.blkaddr * DEV_BSIZE;
5152 	auio.uio_resid  = rwcmd.buflen;
5153 	auio.uio_segflg = flag & FKIOCTL ? UIO_SYSSPACE : UIO_USERSPACE;
5154 
5155 	buf = kmem_alloc(sizeof (buf_t), KM_SLEEP);
5156 	bioinit(buf);
5157 	/*
5158 	 * We use the private field of buf to specify that this is an
5159 	 * I/O using an absolute offset.
5160 	 */
5161 	buf->b_private = (void *)VD_SLICE_NONE;
5162 
5163 	status = physio(vdc_strategy, buf, dev, rw, vdc_min, &auio);
5164 
5165 	biofini(buf);
5166 	kmem_free(buf, sizeof (buf_t));
5167 
5168 	return (status);
5169 }
5170 
5171 /*
5172  * Allocate a buffer for a VD_OP_SCSICMD operation. The size of the allocated
5173  * buffer is returned in alloc_len.
5174  */
5175 static vd_scsi_t *
5176 vdc_scsi_alloc(int cdb_len, int sense_len, int datain_len, int dataout_len,
5177     int *alloc_len)
5178 {
5179 	vd_scsi_t *vd_scsi;
5180 	int vd_scsi_len = VD_SCSI_SIZE;
5181 
5182 	vd_scsi_len += P2ROUNDUP(cdb_len, sizeof (uint64_t));
5183 	vd_scsi_len += P2ROUNDUP(sense_len, sizeof (uint64_t));
5184 	vd_scsi_len += P2ROUNDUP(datain_len, sizeof (uint64_t));
5185 	vd_scsi_len += P2ROUNDUP(dataout_len, sizeof (uint64_t));
5186 
5187 	ASSERT(vd_scsi_len % sizeof (uint64_t) == 0);
5188 
5189 	vd_scsi = kmem_zalloc(vd_scsi_len, KM_SLEEP);
5190 
5191 	vd_scsi->cdb_len = cdb_len;
5192 	vd_scsi->sense_len = sense_len;
5193 	vd_scsi->datain_len = datain_len;
5194 	vd_scsi->dataout_len = dataout_len;
5195 
5196 	*alloc_len = vd_scsi_len;
5197 
5198 	return (vd_scsi);
5199 }
5200 
5201 /*
5202  * Convert the status of a SCSI command to a Solaris return code.
5203  *
5204  * Arguments:
5205  *	vd_scsi		- The SCSI operation buffer.
5206  *	log_error	- indicate if an error message should be logged.
5207  *
5208  * Note that our SCSI error messages are rather primitive for the moment
5209  * and could be improved by decoding some data like the SCSI command and
5210  * the sense key.
5211  *
5212  * Return value:
5213  *	0		- Status is good.
5214  *	EACCES		- Status reports a reservation conflict.
5215  *	ENOTSUP		- Status reports a check condition and sense key
5216  *			  reports an illegal request.
5217  *	EIO		- Any other status.
5218  */
5219 static int
5220 vdc_scsi_status(vdc_t *vdc, vd_scsi_t *vd_scsi, boolean_t log_error)
5221 {
5222 	int rv;
5223 	char path_str[MAXPATHLEN];
5224 	char panic_str[VDC_RESV_CONFLICT_FMT_LEN + MAXPATHLEN];
5225 	union scsi_cdb *cdb;
5226 	struct scsi_extended_sense *sense;
5227 
5228 	if (vd_scsi->cmd_status == STATUS_GOOD)
5229 		/* no error */
5230 		return (0);
5231 
5232 	/* when the tunable vdc_scsi_log_error is true we log all errors */
5233 	if (vdc_scsi_log_error)
5234 		log_error = B_TRUE;
5235 
5236 	if (log_error) {
5237 		cmn_err(CE_WARN, "%s (vdc%d):\tError for Command: 0x%x)\n",
5238 		    ddi_pathname(vdc->dip, path_str), vdc->instance,
5239 		    GETCMD(VD_SCSI_DATA_CDB(vd_scsi)));
5240 	}
5241 
5242 	/* default returned value */
5243 	rv = EIO;
5244 
5245 	switch (vd_scsi->cmd_status) {
5246 
5247 	case STATUS_CHECK:
5248 	case STATUS_TERMINATED:
5249 		if (log_error)
5250 			cmn_err(CE_CONT, "\tCheck Condition Error\n");
5251 
5252 		/* check sense buffer */
5253 		if (vd_scsi->sense_len == 0 ||
5254 		    vd_scsi->sense_status != STATUS_GOOD) {
5255 			if (log_error)
5256 				cmn_err(CE_CONT, "\tNo Sense Data Available\n");
5257 			break;
5258 		}
5259 
5260 		sense = VD_SCSI_DATA_SENSE(vd_scsi);
5261 
5262 		if (log_error) {
5263 			cmn_err(CE_CONT, "\tSense Key:  0x%x\n"
5264 			    "\tASC: 0x%x, ASCQ: 0x%x\n",
5265 			    scsi_sense_key((uint8_t *)sense),
5266 			    scsi_sense_asc((uint8_t *)sense),
5267 			    scsi_sense_ascq((uint8_t *)sense));
5268 		}
5269 
5270 		if (scsi_sense_key((uint8_t *)sense) == KEY_ILLEGAL_REQUEST)
5271 			rv = ENOTSUP;
5272 		break;
5273 
5274 	case STATUS_BUSY:
5275 		if (log_error)
5276 			cmn_err(CE_NOTE, "\tDevice Busy\n");
5277 		break;
5278 
5279 	case STATUS_RESERVATION_CONFLICT:
5280 		/*
5281 		 * If the command was PERSISTENT_RESERVATION_[IN|OUT] then
5282 		 * reservation conflict could be due to various reasons like
5283 		 * incorrect keys, not registered or not reserved etc. So,
5284 		 * we should not panic in that case.
5285 		 */
5286 		cdb = VD_SCSI_DATA_CDB(vd_scsi);
5287 		if (vdc->failfast_interval != 0 &&
5288 		    cdb->scc_cmd != SCMD_PERSISTENT_RESERVE_IN &&
5289 		    cdb->scc_cmd != SCMD_PERSISTENT_RESERVE_OUT) {
5290 			/* failfast is enabled so we have to panic */
5291 			(void) snprintf(panic_str, sizeof (panic_str),
5292 			    VDC_RESV_CONFLICT_FMT_STR "%s",
5293 			    ddi_pathname(vdc->dip, path_str));
5294 			panic(panic_str);
5295 		}
5296 		if (log_error)
5297 			cmn_err(CE_NOTE, "\tReservation Conflict\n");
5298 		rv = EACCES;
5299 		break;
5300 
5301 	case STATUS_QFULL:
5302 		if (log_error)
5303 			cmn_err(CE_NOTE, "\tQueue Full\n");
5304 		break;
5305 
5306 	case STATUS_MET:
5307 	case STATUS_INTERMEDIATE:
5308 	case STATUS_SCSI2:
5309 	case STATUS_INTERMEDIATE_MET:
5310 	case STATUS_ACA_ACTIVE:
5311 		if (log_error)
5312 			cmn_err(CE_CONT,
5313 			    "\tUnexpected SCSI status received: 0x%x\n",
5314 			    vd_scsi->cmd_status);
5315 		break;
5316 
5317 	default:
5318 		if (log_error)
5319 			cmn_err(CE_CONT,
5320 			    "\tInvalid SCSI status received: 0x%x\n",
5321 			    vd_scsi->cmd_status);
5322 		break;
5323 	}
5324 
5325 	return (rv);
5326 }
5327 
5328 /*
5329  * Implemented the USCSICMD uscsi(7I) ioctl. This ioctl is converted to
5330  * a VD_OP_SCSICMD operation which is sent to the vdisk server. If a SCSI
5331  * reset is requested (i.e. a flag USCSI_RESET* is set) then the ioctl is
5332  * converted to a VD_OP_RESET operation.
5333  */
5334 static int
5335 vdc_uscsi_cmd(vdc_t *vdc, caddr_t arg, int mode)
5336 {
5337 	struct uscsi_cmd 	uscsi;
5338 	struct uscsi_cmd32	uscsi32;
5339 	vd_scsi_t 		*vd_scsi;
5340 	int 			vd_scsi_len;
5341 	union scsi_cdb		*cdb;
5342 	struct scsi_extended_sense *sense;
5343 	char 			*datain, *dataout;
5344 	size_t			cdb_len, datain_len, dataout_len, sense_len;
5345 	int 			rv;
5346 
5347 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5348 		if (ddi_copyin(arg, &uscsi32, sizeof (struct uscsi_cmd32),
5349 		    mode) != 0)
5350 			return (EFAULT);
5351 		uscsi_cmd32touscsi_cmd((&uscsi32), (&uscsi));
5352 	} else {
5353 		if (ddi_copyin(arg, &uscsi, sizeof (struct uscsi_cmd),
5354 		    mode) != 0)
5355 			return (EFAULT);
5356 	}
5357 
5358 	/* a uscsi reset is converted to a VD_OP_RESET operation */
5359 	if (uscsi.uscsi_flags & (USCSI_RESET | USCSI_RESET_LUN |
5360 	    USCSI_RESET_ALL)) {
5361 		rv = vdc_do_sync_op(vdc, VD_OP_RESET, NULL, 0, 0, 0, CB_SYNC,
5362 		    (void *)(uint64_t)mode, VIO_both_dir, B_TRUE);
5363 		return (rv);
5364 	}
5365 
5366 	/* cdb buffer length */
5367 	cdb_len = uscsi.uscsi_cdblen;
5368 
5369 	/* data in and out buffers length */
5370 	if (uscsi.uscsi_flags & USCSI_READ) {
5371 		datain_len = uscsi.uscsi_buflen;
5372 		dataout_len = 0;
5373 	} else {
5374 		datain_len = 0;
5375 		dataout_len = uscsi.uscsi_buflen;
5376 	}
5377 
5378 	/* sense buffer length */
5379 	if (uscsi.uscsi_flags & USCSI_RQENABLE)
5380 		sense_len = uscsi.uscsi_rqlen;
5381 	else
5382 		sense_len = 0;
5383 
5384 	/* allocate buffer for the VD_SCSICMD_OP operation */
5385 	vd_scsi = vdc_scsi_alloc(cdb_len, sense_len, datain_len, dataout_len,
5386 	    &vd_scsi_len);
5387 
5388 	/*
5389 	 * The documentation of USCSI_ISOLATE and USCSI_DIAGNOSE is very vague,
5390 	 * but basically they prevent a SCSI command from being retried in case
5391 	 * of an error.
5392 	 */
5393 	if ((uscsi.uscsi_flags & USCSI_ISOLATE) ||
5394 	    (uscsi.uscsi_flags & USCSI_DIAGNOSE))
5395 		vd_scsi->options |= VD_SCSI_OPT_NORETRY;
5396 
5397 	/* set task attribute */
5398 	if (uscsi.uscsi_flags & USCSI_NOTAG) {
5399 		vd_scsi->task_attribute = 0;
5400 	} else {
5401 		if (uscsi.uscsi_flags & USCSI_HEAD)
5402 			vd_scsi->task_attribute = VD_SCSI_TASK_ACA;
5403 		else if (uscsi.uscsi_flags & USCSI_HTAG)
5404 			vd_scsi->task_attribute = VD_SCSI_TASK_HQUEUE;
5405 		else if (uscsi.uscsi_flags & USCSI_OTAG)
5406 			vd_scsi->task_attribute = VD_SCSI_TASK_ORDERED;
5407 		else
5408 			vd_scsi->task_attribute = 0;
5409 	}
5410 
5411 	/* set timeout */
5412 	vd_scsi->timeout = uscsi.uscsi_timeout;
5413 
5414 	/* copy-in cdb data */
5415 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
5416 	if (ddi_copyin(uscsi.uscsi_cdb, cdb, cdb_len, mode) != 0) {
5417 		rv = EFAULT;
5418 		goto done;
5419 	}
5420 
5421 	/* keep a pointer to the sense buffer */
5422 	sense = VD_SCSI_DATA_SENSE(vd_scsi);
5423 
5424 	/* keep a pointer to the data-in buffer */
5425 	datain = (char *)VD_SCSI_DATA_IN(vd_scsi);
5426 
5427 	/* copy-in request data to the data-out buffer */
5428 	dataout = (char *)VD_SCSI_DATA_OUT(vd_scsi);
5429 	if (!(uscsi.uscsi_flags & USCSI_READ)) {
5430 		if (ddi_copyin(uscsi.uscsi_bufaddr, dataout, dataout_len,
5431 		    mode)) {
5432 			rv = EFAULT;
5433 			goto done;
5434 		}
5435 	}
5436 
5437 	/* submit the request */
5438 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
5439 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
5440 
5441 	if (rv != 0)
5442 		goto done;
5443 
5444 	/* update scsi status */
5445 	uscsi.uscsi_status = vd_scsi->cmd_status;
5446 
5447 	/* update sense data */
5448 	if ((uscsi.uscsi_flags & USCSI_RQENABLE) &&
5449 	    (uscsi.uscsi_status == STATUS_CHECK ||
5450 	    uscsi.uscsi_status == STATUS_TERMINATED)) {
5451 
5452 		uscsi.uscsi_rqstatus = vd_scsi->sense_status;
5453 
5454 		if (uscsi.uscsi_rqstatus == STATUS_GOOD) {
5455 			uscsi.uscsi_rqresid = uscsi.uscsi_rqlen -
5456 			    vd_scsi->sense_len;
5457 			if (ddi_copyout(sense, uscsi.uscsi_rqbuf,
5458 			    vd_scsi->sense_len, mode) != 0) {
5459 				rv = EFAULT;
5460 				goto done;
5461 			}
5462 		}
5463 	}
5464 
5465 	/* update request data */
5466 	if (uscsi.uscsi_status == STATUS_GOOD) {
5467 		if (uscsi.uscsi_flags & USCSI_READ) {
5468 			uscsi.uscsi_resid = uscsi.uscsi_buflen -
5469 			    vd_scsi->datain_len;
5470 			if (ddi_copyout(datain, uscsi.uscsi_bufaddr,
5471 			    vd_scsi->datain_len, mode) != 0) {
5472 				rv = EFAULT;
5473 				goto done;
5474 			}
5475 		} else {
5476 			uscsi.uscsi_resid = uscsi.uscsi_buflen -
5477 			    vd_scsi->dataout_len;
5478 		}
5479 	}
5480 
5481 	/* copy-out result */
5482 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5483 		uscsi_cmdtouscsi_cmd32((&uscsi), (&uscsi32));
5484 		if (ddi_copyout(&uscsi32, arg, sizeof (struct uscsi_cmd32),
5485 		    mode) != 0) {
5486 			rv = EFAULT;
5487 			goto done;
5488 		}
5489 	} else {
5490 		if (ddi_copyout(&uscsi, arg, sizeof (struct uscsi_cmd),
5491 		    mode) != 0) {
5492 			rv = EFAULT;
5493 			goto done;
5494 		}
5495 	}
5496 
5497 	/* get the return code from the SCSI command status */
5498 	rv = vdc_scsi_status(vdc, vd_scsi,
5499 	    !(uscsi.uscsi_flags & USCSI_SILENT));
5500 
5501 done:
5502 	kmem_free(vd_scsi, vd_scsi_len);
5503 	return (rv);
5504 }
5505 
5506 /*
5507  * Create a VD_OP_SCSICMD buffer for a SCSI PERSISTENT IN command.
5508  *
5509  * Arguments:
5510  *	cmd		- SCSI PERSISTENT IN command
5511  *	len		- length of the SCSI input buffer
5512  *	vd_scsi_len	- return the length of the allocated buffer
5513  *
5514  * Returned Value:
5515  *	a pointer to the allocated VD_OP_SCSICMD buffer.
5516  */
5517 static vd_scsi_t *
5518 vdc_scsi_alloc_persistent_in(uchar_t cmd, int len, int *vd_scsi_len)
5519 {
5520 	int cdb_len, sense_len, datain_len, dataout_len;
5521 	vd_scsi_t *vd_scsi;
5522 	union scsi_cdb *cdb;
5523 
5524 	cdb_len = CDB_GROUP1;
5525 	sense_len = sizeof (struct scsi_extended_sense);
5526 	datain_len = len;
5527 	dataout_len = 0;
5528 
5529 	vd_scsi = vdc_scsi_alloc(cdb_len, sense_len, datain_len, dataout_len,
5530 	    vd_scsi_len);
5531 
5532 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
5533 
5534 	/* set cdb */
5535 	cdb->scc_cmd = SCMD_PERSISTENT_RESERVE_IN;
5536 	cdb->cdb_opaque[1] = cmd;
5537 	FORMG1COUNT(cdb, datain_len);
5538 
5539 	vd_scsi->timeout = vdc_scsi_timeout;
5540 
5541 	return (vd_scsi);
5542 }
5543 
5544 /*
5545  * Create a VD_OP_SCSICMD buffer for a SCSI PERSISTENT OUT command.
5546  *
5547  * Arguments:
5548  *	cmd		- SCSI PERSISTENT OUT command
5549  *	len		- length of the SCSI output buffer
5550  *	vd_scsi_len	- return the length of the allocated buffer
5551  *
5552  * Returned Code:
5553  *	a pointer to the allocated VD_OP_SCSICMD buffer.
5554  */
5555 static vd_scsi_t *
5556 vdc_scsi_alloc_persistent_out(uchar_t cmd, int len, int *vd_scsi_len)
5557 {
5558 	int cdb_len, sense_len, datain_len, dataout_len;
5559 	vd_scsi_t *vd_scsi;
5560 	union scsi_cdb *cdb;
5561 
5562 	cdb_len = CDB_GROUP1;
5563 	sense_len = sizeof (struct scsi_extended_sense);
5564 	datain_len = 0;
5565 	dataout_len = len;
5566 
5567 	vd_scsi = vdc_scsi_alloc(cdb_len, sense_len, datain_len, dataout_len,
5568 	    vd_scsi_len);
5569 
5570 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
5571 
5572 	/* set cdb */
5573 	cdb->scc_cmd = SCMD_PERSISTENT_RESERVE_OUT;
5574 	cdb->cdb_opaque[1] = cmd;
5575 	FORMG1COUNT(cdb, dataout_len);
5576 
5577 	vd_scsi->timeout = vdc_scsi_timeout;
5578 
5579 	return (vd_scsi);
5580 }
5581 
5582 /*
5583  * Implement the MHIOCGRP_INKEYS mhd(7i) ioctl. The ioctl is converted
5584  * to a SCSI PERSISTENT IN READ KEYS command which is sent to the vdisk
5585  * server with a VD_OP_SCSICMD operation.
5586  */
5587 static int
5588 vdc_mhd_inkeys(vdc_t *vdc, caddr_t arg, int mode)
5589 {
5590 	vd_scsi_t *vd_scsi;
5591 	mhioc_inkeys_t inkeys;
5592 	mhioc_key_list_t klist;
5593 	struct mhioc_inkeys32 inkeys32;
5594 	struct mhioc_key_list32 klist32;
5595 	sd_prin_readkeys_t *scsi_keys;
5596 	void *user_keys;
5597 	int vd_scsi_len;
5598 	int listsize, listlen, rv;
5599 
5600 	/* copyin arguments */
5601 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5602 		rv = ddi_copyin(arg, &inkeys32, sizeof (inkeys32), mode);
5603 		if (rv != 0)
5604 			return (EFAULT);
5605 
5606 		rv = ddi_copyin((caddr_t)(uintptr_t)inkeys32.li, &klist32,
5607 		    sizeof (klist32), mode);
5608 		if (rv != 0)
5609 			return (EFAULT);
5610 
5611 		listsize = klist32.listsize;
5612 	} else {
5613 		rv = ddi_copyin(arg, &inkeys, sizeof (inkeys), mode);
5614 		if (rv != 0)
5615 			return (EFAULT);
5616 
5617 		rv = ddi_copyin(inkeys.li, &klist, sizeof (klist), mode);
5618 		if (rv != 0)
5619 			return (EFAULT);
5620 
5621 		listsize = klist.listsize;
5622 	}
5623 
5624 	/* build SCSI VD_OP request */
5625 	vd_scsi = vdc_scsi_alloc_persistent_in(SD_READ_KEYS,
5626 	    sizeof (sd_prin_readkeys_t) - sizeof (caddr_t) +
5627 	    (sizeof (mhioc_resv_key_t) * listsize), &vd_scsi_len);
5628 
5629 	scsi_keys = (sd_prin_readkeys_t *)VD_SCSI_DATA_IN(vd_scsi);
5630 
5631 	/* submit the request */
5632 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
5633 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
5634 
5635 	if (rv != 0)
5636 		goto done;
5637 
5638 	listlen = scsi_keys->len / MHIOC_RESV_KEY_SIZE;
5639 
5640 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5641 		inkeys32.generation = scsi_keys->generation;
5642 		rv = ddi_copyout(&inkeys32, arg, sizeof (inkeys32), mode);
5643 		if (rv != 0) {
5644 			rv = EFAULT;
5645 			goto done;
5646 		}
5647 
5648 		klist32.listlen = listlen;
5649 		rv = ddi_copyout(&klist32, (caddr_t)(uintptr_t)inkeys32.li,
5650 		    sizeof (klist32), mode);
5651 		if (rv != 0) {
5652 			rv = EFAULT;
5653 			goto done;
5654 		}
5655 
5656 		user_keys = (caddr_t)(uintptr_t)klist32.list;
5657 	} else {
5658 		inkeys.generation = scsi_keys->generation;
5659 		rv = ddi_copyout(&inkeys, arg, sizeof (inkeys), mode);
5660 		if (rv != 0) {
5661 			rv = EFAULT;
5662 			goto done;
5663 		}
5664 
5665 		klist.listlen = listlen;
5666 		rv = ddi_copyout(&klist, inkeys.li, sizeof (klist), mode);
5667 		if (rv != 0) {
5668 			rv = EFAULT;
5669 			goto done;
5670 		}
5671 
5672 		user_keys = klist.list;
5673 	}
5674 
5675 	/* copy out keys */
5676 	if (listlen > 0 && listsize > 0) {
5677 		if (listsize < listlen)
5678 			listlen = listsize;
5679 		rv = ddi_copyout(&scsi_keys->keylist, user_keys,
5680 		    listlen * MHIOC_RESV_KEY_SIZE, mode);
5681 		if (rv != 0)
5682 			rv = EFAULT;
5683 	}
5684 
5685 	if (rv == 0)
5686 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
5687 
5688 done:
5689 	kmem_free(vd_scsi, vd_scsi_len);
5690 
5691 	return (rv);
5692 }
5693 
5694 /*
5695  * Implement the MHIOCGRP_INRESV mhd(7i) ioctl. The ioctl is converted
5696  * to a SCSI PERSISTENT IN READ RESERVATION command which is sent to
5697  * the vdisk server with a VD_OP_SCSICMD operation.
5698  */
5699 static int
5700 vdc_mhd_inresv(vdc_t *vdc, caddr_t arg, int mode)
5701 {
5702 	vd_scsi_t *vd_scsi;
5703 	mhioc_inresvs_t inresv;
5704 	mhioc_resv_desc_list_t rlist;
5705 	struct mhioc_inresvs32 inresv32;
5706 	struct mhioc_resv_desc_list32 rlist32;
5707 	mhioc_resv_desc_t mhd_resv;
5708 	sd_prin_readresv_t *scsi_resv;
5709 	sd_readresv_desc_t *resv;
5710 	mhioc_resv_desc_t *user_resv;
5711 	int vd_scsi_len;
5712 	int listsize, listlen, i, rv;
5713 
5714 	/* copyin arguments */
5715 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5716 		rv = ddi_copyin(arg, &inresv32, sizeof (inresv32), mode);
5717 		if (rv != 0)
5718 			return (EFAULT);
5719 
5720 		rv = ddi_copyin((caddr_t)(uintptr_t)inresv32.li, &rlist32,
5721 		    sizeof (rlist32), mode);
5722 		if (rv != 0)
5723 			return (EFAULT);
5724 
5725 		listsize = rlist32.listsize;
5726 	} else {
5727 		rv = ddi_copyin(arg, &inresv, sizeof (inresv), mode);
5728 		if (rv != 0)
5729 			return (EFAULT);
5730 
5731 		rv = ddi_copyin(inresv.li, &rlist, sizeof (rlist), mode);
5732 		if (rv != 0)
5733 			return (EFAULT);
5734 
5735 		listsize = rlist.listsize;
5736 	}
5737 
5738 	/* build SCSI VD_OP request */
5739 	vd_scsi = vdc_scsi_alloc_persistent_in(SD_READ_RESV,
5740 	    sizeof (sd_prin_readresv_t) - sizeof (caddr_t) +
5741 	    (SCSI3_RESV_DESC_LEN * listsize), &vd_scsi_len);
5742 
5743 	scsi_resv = (sd_prin_readresv_t *)VD_SCSI_DATA_IN(vd_scsi);
5744 
5745 	/* submit the request */
5746 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
5747 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
5748 
5749 	if (rv != 0)
5750 		goto done;
5751 
5752 	listlen = scsi_resv->len / SCSI3_RESV_DESC_LEN;
5753 
5754 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5755 		inresv32.generation = scsi_resv->generation;
5756 		rv = ddi_copyout(&inresv32, arg, sizeof (inresv32), mode);
5757 		if (rv != 0) {
5758 			rv = EFAULT;
5759 			goto done;
5760 		}
5761 
5762 		rlist32.listlen = listlen;
5763 		rv = ddi_copyout(&rlist32, (caddr_t)(uintptr_t)inresv32.li,
5764 		    sizeof (rlist32), mode);
5765 		if (rv != 0) {
5766 			rv = EFAULT;
5767 			goto done;
5768 		}
5769 
5770 		user_resv = (mhioc_resv_desc_t *)(uintptr_t)rlist32.list;
5771 	} else {
5772 		inresv.generation = scsi_resv->generation;
5773 		rv = ddi_copyout(&inresv, arg, sizeof (inresv), mode);
5774 		if (rv != 0) {
5775 			rv = EFAULT;
5776 			goto done;
5777 		}
5778 
5779 		rlist.listlen = listlen;
5780 		rv = ddi_copyout(&rlist, inresv.li, sizeof (rlist), mode);
5781 		if (rv != 0) {
5782 			rv = EFAULT;
5783 			goto done;
5784 		}
5785 
5786 		user_resv = rlist.list;
5787 	}
5788 
5789 	/* copy out reservations */
5790 	if (listsize > 0 && listlen > 0) {
5791 		if (listsize < listlen)
5792 			listlen = listsize;
5793 		resv = (sd_readresv_desc_t *)&scsi_resv->readresv_desc;
5794 
5795 		for (i = 0; i < listlen; i++) {
5796 			mhd_resv.type = resv->type;
5797 			mhd_resv.scope = resv->scope;
5798 			mhd_resv.scope_specific_addr =
5799 			    BE_32(resv->scope_specific_addr);
5800 			bcopy(&resv->resvkey, &mhd_resv.key,
5801 			    MHIOC_RESV_KEY_SIZE);
5802 
5803 			rv = ddi_copyout(&mhd_resv, user_resv,
5804 			    sizeof (mhd_resv), mode);
5805 			if (rv != 0) {
5806 				rv = EFAULT;
5807 				goto done;
5808 			}
5809 			resv++;
5810 			user_resv++;
5811 		}
5812 	}
5813 
5814 	if (rv == 0)
5815 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
5816 
5817 done:
5818 	kmem_free(vd_scsi, vd_scsi_len);
5819 	return (rv);
5820 }
5821 
5822 /*
5823  * Implement the MHIOCGRP_REGISTER mhd(7i) ioctl. The ioctl is converted
5824  * to a SCSI PERSISTENT OUT REGISTER command which is sent to the vdisk
5825  * server with a VD_OP_SCSICMD operation.
5826  */
5827 static int
5828 vdc_mhd_register(vdc_t *vdc, caddr_t arg, int mode)
5829 {
5830 	vd_scsi_t *vd_scsi;
5831 	sd_prout_t *scsi_prout;
5832 	mhioc_register_t mhd_reg;
5833 	int vd_scsi_len, rv;
5834 
5835 	/* copyin arguments */
5836 	rv = ddi_copyin(arg, &mhd_reg, sizeof (mhd_reg), mode);
5837 	if (rv != 0)
5838 		return (EFAULT);
5839 
5840 	/* build SCSI VD_OP request */
5841 	vd_scsi = vdc_scsi_alloc_persistent_out(SD_SCSI3_REGISTER,
5842 	    sizeof (sd_prout_t), &vd_scsi_len);
5843 
5844 	/* set parameters */
5845 	scsi_prout = (sd_prout_t *)VD_SCSI_DATA_OUT(vd_scsi);
5846 	bcopy(mhd_reg.oldkey.key, scsi_prout->res_key, MHIOC_RESV_KEY_SIZE);
5847 	bcopy(mhd_reg.newkey.key, scsi_prout->service_key, MHIOC_RESV_KEY_SIZE);
5848 	scsi_prout->aptpl = (uchar_t)mhd_reg.aptpl;
5849 
5850 	/* submit the request */
5851 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
5852 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
5853 
5854 	if (rv == 0)
5855 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
5856 
5857 	kmem_free(vd_scsi, vd_scsi_len);
5858 	return (rv);
5859 }
5860 
5861 /*
5862  * Implement the MHIOCGRP_RESERVE mhd(7i) ioctl. The ioctl is converted
5863  * to a SCSI PERSISTENT OUT RESERVE command which is sent to the vdisk
5864  * server with a VD_OP_SCSICMD operation.
5865  */
5866 static int
5867 vdc_mhd_reserve(vdc_t *vdc, caddr_t arg, int mode)
5868 {
5869 	union scsi_cdb *cdb;
5870 	vd_scsi_t *vd_scsi;
5871 	sd_prout_t *scsi_prout;
5872 	mhioc_resv_desc_t mhd_resv;
5873 	int vd_scsi_len, rv;
5874 
5875 	/* copyin arguments */
5876 	rv = ddi_copyin(arg, &mhd_resv, sizeof (mhd_resv), mode);
5877 	if (rv != 0)
5878 		return (EFAULT);
5879 
5880 	/* build SCSI VD_OP request */
5881 	vd_scsi = vdc_scsi_alloc_persistent_out(SD_SCSI3_RESERVE,
5882 	    sizeof (sd_prout_t), &vd_scsi_len);
5883 
5884 	/* set parameters */
5885 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
5886 	scsi_prout = (sd_prout_t *)VD_SCSI_DATA_OUT(vd_scsi);
5887 	bcopy(mhd_resv.key.key, scsi_prout->res_key, MHIOC_RESV_KEY_SIZE);
5888 	scsi_prout->scope_address = mhd_resv.scope_specific_addr;
5889 	cdb->cdb_opaque[2] = mhd_resv.type;
5890 
5891 	/* submit the request */
5892 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
5893 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
5894 
5895 	if (rv == 0)
5896 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
5897 
5898 	kmem_free(vd_scsi, vd_scsi_len);
5899 	return (rv);
5900 }
5901 
5902 /*
5903  * Implement the MHIOCGRP_PREEMPTANDABORT mhd(7i) ioctl. The ioctl is
5904  * converted to a SCSI PERSISTENT OUT PREEMPT AND ABORT command which
5905  * is sent to the vdisk server with a VD_OP_SCSICMD operation.
5906  */
5907 static int
5908 vdc_mhd_preemptabort(vdc_t *vdc, caddr_t arg, int mode)
5909 {
5910 	union scsi_cdb *cdb;
5911 	vd_scsi_t *vd_scsi;
5912 	sd_prout_t *scsi_prout;
5913 	mhioc_preemptandabort_t mhd_preempt;
5914 	int vd_scsi_len, rv;
5915 
5916 	/* copyin arguments */
5917 	rv = ddi_copyin(arg, &mhd_preempt, sizeof (mhd_preempt), mode);
5918 	if (rv != 0)
5919 		return (EFAULT);
5920 
5921 	/* build SCSI VD_OP request */
5922 	vd_scsi = vdc_scsi_alloc_persistent_out(SD_SCSI3_PREEMPTANDABORT,
5923 	    sizeof (sd_prout_t), &vd_scsi_len);
5924 
5925 	/* set parameters */
5926 	vd_scsi->task_attribute = VD_SCSI_TASK_ACA;
5927 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
5928 	scsi_prout = (sd_prout_t *)VD_SCSI_DATA_OUT(vd_scsi);
5929 	bcopy(mhd_preempt.resvdesc.key.key, scsi_prout->res_key,
5930 	    MHIOC_RESV_KEY_SIZE);
5931 	bcopy(mhd_preempt.victim_key.key, scsi_prout->service_key,
5932 	    MHIOC_RESV_KEY_SIZE);
5933 	scsi_prout->scope_address = mhd_preempt.resvdesc.scope_specific_addr;
5934 	cdb->cdb_opaque[2] = mhd_preempt.resvdesc.type;
5935 
5936 	/* submit the request */
5937 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
5938 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
5939 
5940 	if (rv == 0)
5941 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
5942 
5943 	kmem_free(vd_scsi, vd_scsi_len);
5944 	return (rv);
5945 }
5946 
5947 /*
5948  * Implement the MHIOCGRP_REGISTERANDIGNOREKEY mhd(7i) ioctl. The ioctl
5949  * is converted to a SCSI PERSISTENT OUT REGISTER AND IGNORE EXISTING KEY
5950  * command which is sent to the vdisk server with a VD_OP_SCSICMD operation.
5951  */
5952 static int
5953 vdc_mhd_registerignore(vdc_t *vdc, caddr_t arg, int mode)
5954 {
5955 	vd_scsi_t *vd_scsi;
5956 	sd_prout_t *scsi_prout;
5957 	mhioc_registerandignorekey_t mhd_regi;
5958 	int vd_scsi_len, rv;
5959 
5960 	/* copyin arguments */
5961 	rv = ddi_copyin(arg, &mhd_regi, sizeof (mhd_regi), mode);
5962 	if (rv != 0)
5963 		return (EFAULT);
5964 
5965 	/* build SCSI VD_OP request */
5966 	vd_scsi = vdc_scsi_alloc_persistent_out(SD_SCSI3_REGISTERANDIGNOREKEY,
5967 	    sizeof (sd_prout_t), &vd_scsi_len);
5968 
5969 	/* set parameters */
5970 	scsi_prout = (sd_prout_t *)VD_SCSI_DATA_OUT(vd_scsi);
5971 	bcopy(mhd_regi.newkey.key, scsi_prout->service_key,
5972 	    MHIOC_RESV_KEY_SIZE);
5973 	scsi_prout->aptpl = (uchar_t)mhd_regi.aptpl;
5974 
5975 	/* submit the request */
5976 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
5977 	    0, 0, CB_SYNC, (void *)(uint64_t)mode, VIO_both_dir, B_FALSE);
5978 
5979 	if (rv == 0)
5980 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
5981 
5982 	kmem_free(vd_scsi, vd_scsi_len);
5983 	return (rv);
5984 }
5985 
5986 /*
5987  * This function is used by the failfast mechanism to send a SCSI command
5988  * to check for reservation conflict.
5989  */
5990 static int
5991 vdc_failfast_scsi_cmd(vdc_t *vdc, uchar_t scmd)
5992 {
5993 	int cdb_len, sense_len, vd_scsi_len;
5994 	vd_scsi_t *vd_scsi;
5995 	union scsi_cdb *cdb;
5996 	int rv;
5997 
5998 	ASSERT(scmd == SCMD_TEST_UNIT_READY || scmd == SCMD_WRITE_G1);
5999 
6000 	if (scmd == SCMD_WRITE_G1)
6001 		cdb_len = CDB_GROUP1;
6002 	else
6003 		cdb_len = CDB_GROUP0;
6004 
6005 	sense_len = sizeof (struct scsi_extended_sense);
6006 
6007 	vd_scsi = vdc_scsi_alloc(cdb_len, sense_len, 0, 0, &vd_scsi_len);
6008 
6009 	/* set cdb */
6010 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
6011 	cdb->scc_cmd = scmd;
6012 
6013 	vd_scsi->timeout = vdc_scsi_timeout;
6014 
6015 	/*
6016 	 * Submit the request. The last argument has to be B_FALSE so that
6017 	 * vdc_do_sync_op does not loop checking for reservation conflict if
6018 	 * the operation returns an error.
6019 	 */
6020 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
6021 	    0, 0, CB_SYNC, (void *)(uint64_t)FKIOCTL, VIO_both_dir, B_FALSE);
6022 
6023 	if (rv == 0)
6024 		(void) vdc_scsi_status(vdc, vd_scsi, B_FALSE);
6025 
6026 	kmem_free(vd_scsi, vd_scsi_len);
6027 	return (rv);
6028 }
6029 
6030 /*
6031  * This function is used by the failfast mechanism to check for reservation
6032  * conflict. It sends some SCSI commands which will fail with a reservation
6033  * conflict error if the system does not have access to the disk and this
6034  * will panic the system.
6035  *
6036  * Returned Code:
6037  *	0	- disk is accessible without reservation conflict error
6038  *	!= 0	- unable to check if disk is accessible
6039  */
6040 int
6041 vdc_failfast_check_resv(vdc_t *vdc)
6042 {
6043 	int failure = 0;
6044 
6045 	/*
6046 	 * Send a TEST UNIT READY command. The command will panic
6047 	 * the system if it fails with a reservation conflict.
6048 	 */
6049 	if (vdc_failfast_scsi_cmd(vdc, SCMD_TEST_UNIT_READY) != 0)
6050 		failure++;
6051 
6052 	/*
6053 	 * With SPC-3 compliant devices TEST UNIT READY will succeed on
6054 	 * a reserved device, so we also do a WRITE(10) of zero byte in
6055 	 * order to provoke a Reservation Conflict status on those newer
6056 	 * devices.
6057 	 */
6058 	if (vdc_failfast_scsi_cmd(vdc, SCMD_WRITE_G1) != 0)
6059 		failure++;
6060 
6061 	return (failure);
6062 }
6063 
6064 /*
6065  * Add a pending I/O to the failfast I/O queue. An I/O is added to this
6066  * queue when it has failed and failfast is enabled. Then we have to check
6067  * if it has failed because of a reservation conflict in which case we have
6068  * to panic the system.
6069  *
6070  * Async I/O should be queued with their block I/O data transfer structure
6071  * (buf). Sync I/O should be queued with buf = NULL.
6072  */
6073 static vdc_io_t *
6074 vdc_failfast_io_queue(vdc_t *vdc, struct buf *buf)
6075 {
6076 	vdc_io_t *vio;
6077 
6078 	ASSERT(MUTEX_HELD(&vdc->lock));
6079 
6080 	vio = kmem_alloc(sizeof (vdc_io_t), KM_SLEEP);
6081 	vio->vio_next = vdc->failfast_io_queue;
6082 	vio->vio_buf = buf;
6083 	vio->vio_qtime = ddi_get_lbolt();
6084 
6085 	vdc->failfast_io_queue = vio;
6086 
6087 	/* notify the failfast thread that a new I/O is queued */
6088 	cv_signal(&vdc->failfast_cv);
6089 
6090 	return (vio);
6091 }
6092 
6093 /*
6094  * Remove and complete I/O in the failfast I/O queue which have been
6095  * added after the indicated deadline. A deadline of 0 means that all
6096  * I/O have to be unqueued and marked as completed.
6097  */
6098 static void
6099 vdc_failfast_io_unqueue(vdc_t *vdc, clock_t deadline)
6100 {
6101 	vdc_io_t *vio, *vio_tmp;
6102 
6103 	ASSERT(MUTEX_HELD(&vdc->lock));
6104 
6105 	vio_tmp = NULL;
6106 	vio = vdc->failfast_io_queue;
6107 
6108 	if (deadline != 0) {
6109 		/*
6110 		 * Skip any io queued after the deadline. The failfast
6111 		 * I/O queue is ordered starting with the last I/O added
6112 		 * to the queue.
6113 		 */
6114 		while (vio != NULL && vio->vio_qtime > deadline) {
6115 			vio_tmp = vio;
6116 			vio = vio->vio_next;
6117 		}
6118 	}
6119 
6120 	if (vio == NULL)
6121 		/* nothing to unqueue */
6122 		return;
6123 
6124 	/* update the queue */
6125 	if (vio_tmp == NULL)
6126 		vdc->failfast_io_queue = NULL;
6127 	else
6128 		vio_tmp->vio_next = NULL;
6129 
6130 	/*
6131 	 * Complete unqueued I/O. Async I/O have a block I/O data transfer
6132 	 * structure (buf) and they are completed by calling biodone(). Sync
6133 	 * I/O do not have a buf and they are completed by setting the
6134 	 * vio_qtime to zero and signaling failfast_io_cv. In that case, the
6135 	 * thread waiting for the I/O to complete is responsible for freeing
6136 	 * the vio structure.
6137 	 */
6138 	while (vio != NULL) {
6139 		vio_tmp = vio->vio_next;
6140 		if (vio->vio_buf != NULL) {
6141 			VD_KSTAT_RUNQ_EXIT(vdc);
6142 			DTRACE_IO1(done, buf_t *, vio->vio_buf);
6143 			biodone(vio->vio_buf);
6144 			kmem_free(vio, sizeof (vdc_io_t));
6145 		} else {
6146 			vio->vio_qtime = 0;
6147 		}
6148 		vio = vio_tmp;
6149 	}
6150 
6151 	cv_broadcast(&vdc->failfast_io_cv);
6152 }
6153 
6154 /*
6155  * Failfast Thread.
6156  *
6157  * While failfast is enabled, the failfast thread sends a TEST UNIT READY
6158  * and a zero size WRITE(10) SCSI commands on a regular basis to check that
6159  * we still have access to the disk. If a command fails with a RESERVATION
6160  * CONFLICT error then the system will immediatly panic.
6161  *
6162  * The failfast thread is also woken up when an I/O has failed. It then check
6163  * the access to the disk to ensure that the I/O failure was not due to a
6164  * reservation conflict.
6165  *
6166  * There is one failfast thread for each virtual disk for which failfast is
6167  * enabled. We could have only one thread sending requests for all disks but
6168  * this would need vdc to send asynchronous requests and to have callbacks to
6169  * process replies.
6170  */
6171 static void
6172 vdc_failfast_thread(void *arg)
6173 {
6174 	int status;
6175 	vdc_t *vdc = (vdc_t *)arg;
6176 	clock_t timeout, starttime;
6177 
6178 	mutex_enter(&vdc->lock);
6179 
6180 	while (vdc->failfast_interval != 0) {
6181 
6182 		starttime = ddi_get_lbolt();
6183 
6184 		mutex_exit(&vdc->lock);
6185 
6186 		/* check for reservation conflict */
6187 		status = vdc_failfast_check_resv(vdc);
6188 
6189 		mutex_enter(&vdc->lock);
6190 		/*
6191 		 * We have dropped the lock to send the SCSI command so we have
6192 		 * to check that failfast is still enabled.
6193 		 */
6194 		if (vdc->failfast_interval == 0)
6195 			break;
6196 
6197 		/*
6198 		 * If we have successfully check the disk access and there was
6199 		 * no reservation conflict then we can complete any I/O queued
6200 		 * before the last check.
6201 		 */
6202 		if (status == 0)
6203 			vdc_failfast_io_unqueue(vdc, starttime);
6204 
6205 		/* proceed again if some I/O are still in the queue */
6206 		if (vdc->failfast_io_queue != NULL)
6207 			continue;
6208 
6209 		timeout = ddi_get_lbolt() +
6210 		    drv_usectohz(vdc->failfast_interval);
6211 		(void) cv_timedwait(&vdc->failfast_cv, &vdc->lock, timeout);
6212 	}
6213 
6214 	/*
6215 	 * Failfast is being stop so we can complete any queued I/O.
6216 	 */
6217 	vdc_failfast_io_unqueue(vdc, 0);
6218 	vdc->failfast_thread = NULL;
6219 	mutex_exit(&vdc->lock);
6220 	thread_exit();
6221 }
6222 
6223 /*
6224  * Implement the MHIOCENFAILFAST mhd(7i) ioctl.
6225  */
6226 static int
6227 vdc_failfast(vdc_t *vdc, caddr_t arg, int mode)
6228 {
6229 	unsigned int mh_time;
6230 
6231 	if (ddi_copyin((void *)arg, &mh_time, sizeof (int), mode))
6232 		return (EFAULT);
6233 
6234 	mutex_enter(&vdc->lock);
6235 	if (mh_time != 0 && vdc->failfast_thread == NULL) {
6236 		vdc->failfast_thread = thread_create(NULL, 0,
6237 		    vdc_failfast_thread, vdc, 0, &p0, TS_RUN,
6238 		    v.v_maxsyspri - 2);
6239 	}
6240 
6241 	vdc->failfast_interval = mh_time * 1000;
6242 	cv_signal(&vdc->failfast_cv);
6243 	mutex_exit(&vdc->lock);
6244 
6245 	return (0);
6246 }
6247 
6248 /*
6249  * Implement the MHIOCTKOWN and MHIOCRELEASE mhd(7i) ioctls. These ioctls are
6250  * converted to VD_OP_SET_ACCESS operations.
6251  */
6252 static int
6253 vdc_access_set(vdc_t *vdc, uint64_t flags, int mode)
6254 {
6255 	int rv;
6256 
6257 	/* submit owership command request */
6258 	rv = vdc_do_sync_op(vdc, VD_OP_SET_ACCESS, (caddr_t)&flags,
6259 	    sizeof (uint64_t), 0, 0, CB_SYNC, (void *)(uint64_t)mode,
6260 	    VIO_both_dir, B_TRUE);
6261 
6262 	return (rv);
6263 }
6264 
6265 /*
6266  * Implement the MHIOCSTATUS mhd(7i) ioctl. This ioctl is converted to a
6267  * VD_OP_GET_ACCESS operation.
6268  */
6269 static int
6270 vdc_access_get(vdc_t *vdc, uint64_t *status, int mode)
6271 {
6272 	int rv;
6273 
6274 	/* submit owership command request */
6275 	rv = vdc_do_sync_op(vdc, VD_OP_GET_ACCESS, (caddr_t)status,
6276 	    sizeof (uint64_t), 0, 0, CB_SYNC, (void *)(uint64_t)mode,
6277 	    VIO_both_dir, B_TRUE);
6278 
6279 	return (rv);
6280 }
6281 
6282 /*
6283  * Disk Ownership Thread.
6284  *
6285  * When we have taken the ownership of a disk, this thread waits to be
6286  * notified when the LDC channel is reset so that it can recover the
6287  * ownership.
6288  *
6289  * Note that the thread handling the LDC reset (vdc_process_msg_thread())
6290  * can not be used to do the ownership recovery because it has to be
6291  * running to handle the reply message to the ownership operation.
6292  */
6293 static void
6294 vdc_ownership_thread(void *arg)
6295 {
6296 	vdc_t *vdc = (vdc_t *)arg;
6297 	clock_t timeout;
6298 	uint64_t status;
6299 
6300 	mutex_enter(&vdc->ownership_lock);
6301 	mutex_enter(&vdc->lock);
6302 
6303 	while (vdc->ownership & VDC_OWNERSHIP_WANTED) {
6304 
6305 		if ((vdc->ownership & VDC_OWNERSHIP_RESET) ||
6306 		    !(vdc->ownership & VDC_OWNERSHIP_GRANTED)) {
6307 			/*
6308 			 * There was a reset so the ownership has been lost,
6309 			 * try to recover. We do this without using the preempt
6310 			 * option so that we don't steal the ownership from
6311 			 * someone who has preempted us.
6312 			 */
6313 			DMSG(vdc, 0, "[%d] Ownership lost, recovering",
6314 			    vdc->instance);
6315 
6316 			vdc->ownership &= ~(VDC_OWNERSHIP_RESET |
6317 			    VDC_OWNERSHIP_GRANTED);
6318 
6319 			mutex_exit(&vdc->lock);
6320 
6321 			status = vdc_access_set(vdc, VD_ACCESS_SET_EXCLUSIVE |
6322 			    VD_ACCESS_SET_PRESERVE, FKIOCTL);
6323 
6324 			mutex_enter(&vdc->lock);
6325 
6326 			if (status == 0) {
6327 				DMSG(vdc, 0, "[%d] Ownership recovered",
6328 				    vdc->instance);
6329 				vdc->ownership |= VDC_OWNERSHIP_GRANTED;
6330 			} else {
6331 				DMSG(vdc, 0, "[%d] Fail to recover ownership",
6332 				    vdc->instance);
6333 			}
6334 
6335 		}
6336 
6337 		/*
6338 		 * If we have the ownership then we just wait for an event
6339 		 * to happen (LDC reset), otherwise we will retry to recover
6340 		 * after a delay.
6341 		 */
6342 		if (vdc->ownership & VDC_OWNERSHIP_GRANTED)
6343 			timeout = 0;
6344 		else
6345 			timeout = ddi_get_lbolt() +
6346 			    drv_usectohz(vdc_ownership_delay);
6347 
6348 		/* Release the ownership_lock and wait on the vdc lock */
6349 		mutex_exit(&vdc->ownership_lock);
6350 
6351 		if (timeout == 0)
6352 			(void) cv_wait(&vdc->ownership_cv, &vdc->lock);
6353 		else
6354 			(void) cv_timedwait(&vdc->ownership_cv,
6355 			    &vdc->lock, timeout);
6356 
6357 		mutex_exit(&vdc->lock);
6358 
6359 		mutex_enter(&vdc->ownership_lock);
6360 		mutex_enter(&vdc->lock);
6361 	}
6362 
6363 	vdc->ownership_thread = NULL;
6364 	mutex_exit(&vdc->lock);
6365 	mutex_exit(&vdc->ownership_lock);
6366 
6367 	thread_exit();
6368 }
6369 
6370 static void
6371 vdc_ownership_update(vdc_t *vdc, int ownership_flags)
6372 {
6373 	ASSERT(MUTEX_HELD(&vdc->ownership_lock));
6374 
6375 	mutex_enter(&vdc->lock);
6376 	vdc->ownership = ownership_flags;
6377 	if ((vdc->ownership & VDC_OWNERSHIP_WANTED) &&
6378 	    vdc->ownership_thread == NULL) {
6379 		/* start ownership thread */
6380 		vdc->ownership_thread = thread_create(NULL, 0,
6381 		    vdc_ownership_thread, vdc, 0, &p0, TS_RUN,
6382 		    v.v_maxsyspri - 2);
6383 	} else {
6384 		/* notify the ownership thread */
6385 		cv_signal(&vdc->ownership_cv);
6386 	}
6387 	mutex_exit(&vdc->lock);
6388 }
6389 
6390 /*
6391  * Get the size and the block size of a virtual disk from the vdisk server.
6392  * We need to use this operation when the vdisk_size attribute was not
6393  * available during the handshake with the vdisk server.
6394  */
6395 static int
6396 vdc_check_capacity(vdc_t *vdc)
6397 {
6398 	int rv = 0;
6399 	size_t alloc_len;
6400 	vd_capacity_t *vd_cap;
6401 
6402 	if (vdc->vdisk_size != 0)
6403 		return (0);
6404 
6405 	alloc_len = P2ROUNDUP(sizeof (vd_capacity_t), sizeof (uint64_t));
6406 
6407 	vd_cap = kmem_zalloc(alloc_len, KM_SLEEP);
6408 
6409 	rv = vdc_do_sync_op(vdc, VD_OP_GET_CAPACITY, (caddr_t)vd_cap, alloc_len,
6410 	    0, 0, CB_SYNC, (void *)(uint64_t)FKIOCTL, VIO_both_dir, B_TRUE);
6411 
6412 	if (rv == 0) {
6413 		if (vd_cap->vdisk_block_size != vdc->block_size ||
6414 		    vd_cap->vdisk_size == VD_SIZE_UNKNOWN ||
6415 		    vd_cap->vdisk_size == 0)
6416 			rv = EINVAL;
6417 		else
6418 			vdc->vdisk_size = vd_cap->vdisk_size;
6419 	}
6420 
6421 	kmem_free(vd_cap, alloc_len);
6422 	return (rv);
6423 }
6424 
6425 /*
6426  * This structure is used in the DKIO(7I) array below.
6427  */
6428 typedef struct vdc_dk_ioctl {
6429 	uint8_t		op;		/* VD_OP_XXX value */
6430 	int		cmd;		/* Solaris ioctl operation number */
6431 	size_t		nbytes;		/* size of structure to be copied */
6432 
6433 	/* function to convert between vDisk and Solaris structure formats */
6434 	int	(*convert)(vdc_t *vdc, void *vd_buf, void *ioctl_arg,
6435 	    int mode, int dir);
6436 } vdc_dk_ioctl_t;
6437 
6438 /*
6439  * Subset of DKIO(7I) operations currently supported
6440  */
6441 static vdc_dk_ioctl_t	dk_ioctl[] = {
6442 	{VD_OP_FLUSH,		DKIOCFLUSHWRITECACHE,	0,
6443 		vdc_null_copy_func},
6444 	{VD_OP_GET_WCE,		DKIOCGETWCE,		sizeof (int),
6445 		vdc_get_wce_convert},
6446 	{VD_OP_SET_WCE,		DKIOCSETWCE,		sizeof (int),
6447 		vdc_set_wce_convert},
6448 	{VD_OP_GET_VTOC,	DKIOCGVTOC,		sizeof (vd_vtoc_t),
6449 		vdc_get_vtoc_convert},
6450 	{VD_OP_SET_VTOC,	DKIOCSVTOC,		sizeof (vd_vtoc_t),
6451 		vdc_set_vtoc_convert},
6452 	{VD_OP_GET_DISKGEOM,	DKIOCGGEOM,		sizeof (vd_geom_t),
6453 		vdc_get_geom_convert},
6454 	{VD_OP_GET_DISKGEOM,	DKIOCG_PHYGEOM,		sizeof (vd_geom_t),
6455 		vdc_get_geom_convert},
6456 	{VD_OP_GET_DISKGEOM, 	DKIOCG_VIRTGEOM,	sizeof (vd_geom_t),
6457 		vdc_get_geom_convert},
6458 	{VD_OP_SET_DISKGEOM,	DKIOCSGEOM,		sizeof (vd_geom_t),
6459 		vdc_set_geom_convert},
6460 	{VD_OP_GET_EFI,		DKIOCGETEFI,		0,
6461 		vdc_get_efi_convert},
6462 	{VD_OP_SET_EFI,		DKIOCSETEFI,		0,
6463 		vdc_set_efi_convert},
6464 
6465 	/* DIOCTL_RWCMD is converted to a read or a write */
6466 	{0, DIOCTL_RWCMD,  sizeof (struct dadkio_rwcmd), NULL},
6467 
6468 	/* mhd(7I) non-shared multihost disks ioctls */
6469 	{0, MHIOCTKOWN,				0, vdc_null_copy_func},
6470 	{0, MHIOCRELEASE,			0, vdc_null_copy_func},
6471 	{0, MHIOCSTATUS,			0, vdc_null_copy_func},
6472 	{0, MHIOCQRESERVE,			0, vdc_null_copy_func},
6473 
6474 	/* mhd(7I) shared multihost disks ioctls */
6475 	{0, MHIOCGRP_INKEYS,			0, vdc_null_copy_func},
6476 	{0, MHIOCGRP_INRESV,			0, vdc_null_copy_func},
6477 	{0, MHIOCGRP_REGISTER,			0, vdc_null_copy_func},
6478 	{0, MHIOCGRP_RESERVE, 			0, vdc_null_copy_func},
6479 	{0, MHIOCGRP_PREEMPTANDABORT,		0, vdc_null_copy_func},
6480 	{0, MHIOCGRP_REGISTERANDIGNOREKEY,	0, vdc_null_copy_func},
6481 
6482 	/* mhd(7I) failfast ioctl */
6483 	{0, MHIOCENFAILFAST,			0, vdc_null_copy_func},
6484 
6485 	/*
6486 	 * These particular ioctls are not sent to the server - vdc fakes up
6487 	 * the necessary info.
6488 	 */
6489 	{0, DKIOCINFO, sizeof (struct dk_cinfo), vdc_null_copy_func},
6490 	{0, DKIOCGMEDIAINFO, sizeof (struct dk_minfo), vdc_null_copy_func},
6491 	{0, USCSICMD,	sizeof (struct uscsi_cmd), vdc_null_copy_func},
6492 	{0, DKIOCPARTITION, 0, vdc_null_copy_func },
6493 	{0, DKIOCGAPART, 0, vdc_null_copy_func },
6494 	{0, DKIOCREMOVABLE, 0, vdc_null_copy_func},
6495 	{0, CDROMREADOFFSET, 0, vdc_null_copy_func}
6496 };
6497 
6498 /*
6499  * This function handles ioctl requests from the vd_efi_alloc_and_read()
6500  * function and forward them to the vdisk.
6501  */
6502 static int
6503 vd_process_efi_ioctl(void *vdisk, int cmd, uintptr_t arg)
6504 {
6505 	vdc_t *vdc = (vdc_t *)vdisk;
6506 	dev_t dev;
6507 	int rval;
6508 
6509 	dev = makedevice(ddi_driver_major(vdc->dip),
6510 	    VD_MAKE_DEV(vdc->instance, 0));
6511 
6512 	return (vd_process_ioctl(dev, cmd, (caddr_t)arg, FKIOCTL, &rval));
6513 }
6514 
6515 /*
6516  * Function:
6517  *	vd_process_ioctl()
6518  *
6519  * Description:
6520  *	This routine processes disk specific ioctl calls
6521  *
6522  * Arguments:
6523  *	dev	- the device number
6524  *	cmd	- the operation [dkio(7I)] to be processed
6525  *	arg	- pointer to user provided structure
6526  *		  (contains data to be set or reference parameter for get)
6527  *	mode	- bit flag, indicating open settings, 32/64 bit type, etc
6528  *	rvalp	- pointer to return value for calling process.
6529  *
6530  * Return Code:
6531  *	0
6532  *	EFAULT
6533  *	ENXIO
6534  *	EIO
6535  *	ENOTSUP
6536  */
6537 static int
6538 vd_process_ioctl(dev_t dev, int cmd, caddr_t arg, int mode, int *rvalp)
6539 {
6540 	int		instance = VDCUNIT(dev);
6541 	vdc_t		*vdc = NULL;
6542 	int		rv = -1;
6543 	int		idx = 0;		/* index into dk_ioctl[] */
6544 	size_t		len = 0;		/* #bytes to send to vds */
6545 	size_t		alloc_len = 0;		/* #bytes to allocate mem for */
6546 	caddr_t		mem_p = NULL;
6547 	size_t		nioctls = (sizeof (dk_ioctl)) / (sizeof (dk_ioctl[0]));
6548 	vdc_dk_ioctl_t	*iop;
6549 
6550 	vdc = ddi_get_soft_state(vdc_state, instance);
6551 	if (vdc == NULL) {
6552 		cmn_err(CE_NOTE, "![%d] Could not get soft state structure",
6553 		    instance);
6554 		return (ENXIO);
6555 	}
6556 
6557 	DMSG(vdc, 0, "[%d] Processing ioctl(%x) for dev %lx : model %x\n",
6558 	    instance, cmd, dev, ddi_model_convert_from(mode & FMODELS));
6559 
6560 	if (rvalp != NULL) {
6561 		/* the return value of the ioctl is 0 by default */
6562 		*rvalp = 0;
6563 	}
6564 
6565 	/*
6566 	 * Validate the ioctl operation to be performed.
6567 	 *
6568 	 * If we have looped through the array without finding a match then we
6569 	 * don't support this ioctl.
6570 	 */
6571 	for (idx = 0; idx < nioctls; idx++) {
6572 		if (cmd == dk_ioctl[idx].cmd)
6573 			break;
6574 	}
6575 
6576 	if (idx >= nioctls) {
6577 		DMSG(vdc, 0, "[%d] Unsupported ioctl (0x%x)\n",
6578 		    vdc->instance, cmd);
6579 		return (ENOTSUP);
6580 	}
6581 
6582 	iop = &(dk_ioctl[idx]);
6583 
6584 	if (cmd == DKIOCGETEFI || cmd == DKIOCSETEFI) {
6585 		/* size is not fixed for EFI ioctls, it depends on ioctl arg */
6586 		dk_efi_t	dk_efi;
6587 
6588 		rv = ddi_copyin(arg, &dk_efi, sizeof (dk_efi_t), mode);
6589 		if (rv != 0)
6590 			return (EFAULT);
6591 
6592 		len = sizeof (vd_efi_t) - 1 + dk_efi.dki_length;
6593 	} else {
6594 		len = iop->nbytes;
6595 	}
6596 
6597 	/* check if the ioctl is applicable */
6598 	switch (cmd) {
6599 	case CDROMREADOFFSET:
6600 	case DKIOCREMOVABLE:
6601 		return (ENOTTY);
6602 
6603 	case USCSICMD:
6604 	case MHIOCTKOWN:
6605 	case MHIOCSTATUS:
6606 	case MHIOCQRESERVE:
6607 	case MHIOCRELEASE:
6608 	case MHIOCGRP_INKEYS:
6609 	case MHIOCGRP_INRESV:
6610 	case MHIOCGRP_REGISTER:
6611 	case MHIOCGRP_RESERVE:
6612 	case MHIOCGRP_PREEMPTANDABORT:
6613 	case MHIOCGRP_REGISTERANDIGNOREKEY:
6614 	case MHIOCENFAILFAST:
6615 		if (vdc->cinfo == NULL)
6616 			return (ENXIO);
6617 		if (vdc->cinfo->dki_ctype != DKC_SCSI_CCS)
6618 			return (ENOTTY);
6619 		break;
6620 
6621 	case DIOCTL_RWCMD:
6622 		if (vdc->cinfo == NULL)
6623 			return (ENXIO);
6624 		if (vdc->cinfo->dki_ctype != DKC_DIRECT)
6625 			return (ENOTTY);
6626 		break;
6627 
6628 	case DKIOCINFO:
6629 		if (vdc->cinfo == NULL)
6630 			return (ENXIO);
6631 		break;
6632 
6633 	case DKIOCGMEDIAINFO:
6634 		if (vdc->minfo == NULL)
6635 			return (ENXIO);
6636 		if (vdc_check_capacity(vdc) != 0)
6637 			/* disk capacity is not available */
6638 			return (EIO);
6639 		break;
6640 	}
6641 
6642 	/*
6643 	 * Deal with ioctls which require a processing different than
6644 	 * converting ioctl arguments and sending a corresponding
6645 	 * VD operation.
6646 	 */
6647 	switch (cmd) {
6648 
6649 	case USCSICMD:
6650 	{
6651 		return (vdc_uscsi_cmd(vdc, arg, mode));
6652 	}
6653 
6654 	case MHIOCTKOWN:
6655 	{
6656 		mutex_enter(&vdc->ownership_lock);
6657 		/*
6658 		 * We have to set VDC_OWNERSHIP_WANTED now so that the ownership
6659 		 * can be flagged with VDC_OWNERSHIP_RESET if the LDC is reset
6660 		 * while we are processing the ioctl.
6661 		 */
6662 		vdc_ownership_update(vdc, VDC_OWNERSHIP_WANTED);
6663 
6664 		rv = vdc_access_set(vdc, VD_ACCESS_SET_EXCLUSIVE |
6665 		    VD_ACCESS_SET_PREEMPT | VD_ACCESS_SET_PRESERVE, mode);
6666 		if (rv == 0) {
6667 			vdc_ownership_update(vdc, VDC_OWNERSHIP_WANTED |
6668 			    VDC_OWNERSHIP_GRANTED);
6669 		} else {
6670 			vdc_ownership_update(vdc, VDC_OWNERSHIP_NONE);
6671 		}
6672 		mutex_exit(&vdc->ownership_lock);
6673 		return (rv);
6674 	}
6675 
6676 	case MHIOCRELEASE:
6677 	{
6678 		mutex_enter(&vdc->ownership_lock);
6679 		rv = vdc_access_set(vdc, VD_ACCESS_SET_CLEAR, mode);
6680 		if (rv == 0) {
6681 			vdc_ownership_update(vdc, VDC_OWNERSHIP_NONE);
6682 		}
6683 		mutex_exit(&vdc->ownership_lock);
6684 		return (rv);
6685 	}
6686 
6687 	case MHIOCSTATUS:
6688 	{
6689 		uint64_t status;
6690 
6691 		rv = vdc_access_get(vdc, &status, mode);
6692 		if (rv == 0 && rvalp != NULL)
6693 			*rvalp = (status & VD_ACCESS_ALLOWED)? 0 : 1;
6694 		return (rv);
6695 	}
6696 
6697 	case MHIOCQRESERVE:
6698 	{
6699 		rv = vdc_access_set(vdc, VD_ACCESS_SET_EXCLUSIVE, mode);
6700 		return (rv);
6701 	}
6702 
6703 	case MHIOCGRP_INKEYS:
6704 	{
6705 		return (vdc_mhd_inkeys(vdc, arg, mode));
6706 	}
6707 
6708 	case MHIOCGRP_INRESV:
6709 	{
6710 		return (vdc_mhd_inresv(vdc, arg, mode));
6711 	}
6712 
6713 	case MHIOCGRP_REGISTER:
6714 	{
6715 		return (vdc_mhd_register(vdc, arg, mode));
6716 	}
6717 
6718 	case MHIOCGRP_RESERVE:
6719 	{
6720 		return (vdc_mhd_reserve(vdc, arg, mode));
6721 	}
6722 
6723 	case MHIOCGRP_PREEMPTANDABORT:
6724 	{
6725 		return (vdc_mhd_preemptabort(vdc, arg, mode));
6726 	}
6727 
6728 	case MHIOCGRP_REGISTERANDIGNOREKEY:
6729 	{
6730 		return (vdc_mhd_registerignore(vdc, arg, mode));
6731 	}
6732 
6733 	case MHIOCENFAILFAST:
6734 	{
6735 		rv = vdc_failfast(vdc, arg, mode);
6736 		return (rv);
6737 	}
6738 
6739 	case DIOCTL_RWCMD:
6740 	{
6741 		return (vdc_dioctl_rwcmd(dev, arg, mode));
6742 	}
6743 
6744 	case DKIOCGAPART:
6745 	{
6746 		return (vdc_dkio_gapart(vdc, arg, mode));
6747 	}
6748 
6749 	case DKIOCPARTITION:
6750 	{
6751 		return (vdc_dkio_partition(vdc, arg, mode));
6752 	}
6753 
6754 	case DKIOCINFO:
6755 	{
6756 		struct dk_cinfo	cinfo;
6757 
6758 		bcopy(vdc->cinfo, &cinfo, sizeof (struct dk_cinfo));
6759 		cinfo.dki_partition = VDCPART(dev);
6760 
6761 		rv = ddi_copyout(&cinfo, (void *)arg,
6762 		    sizeof (struct dk_cinfo), mode);
6763 		if (rv != 0)
6764 			return (EFAULT);
6765 
6766 		return (0);
6767 	}
6768 
6769 	case DKIOCGMEDIAINFO:
6770 	{
6771 		ASSERT(vdc->vdisk_size != 0);
6772 		if (vdc->minfo->dki_capacity == 0)
6773 			vdc->minfo->dki_capacity = vdc->vdisk_size;
6774 		rv = ddi_copyout(vdc->minfo, (void *)arg,
6775 		    sizeof (struct dk_minfo), mode);
6776 		if (rv != 0)
6777 			return (EFAULT);
6778 
6779 		return (0);
6780 	}
6781 
6782 	case DKIOCFLUSHWRITECACHE:
6783 		{
6784 			struct dk_callback *dkc =
6785 			    (struct dk_callback *)(uintptr_t)arg;
6786 			vdc_dk_arg_t	*dkarg = NULL;
6787 
6788 			DMSG(vdc, 1, "[%d] Flush W$: mode %x\n",
6789 			    instance, mode);
6790 
6791 			/*
6792 			 * If arg is NULL, then there is no callback function
6793 			 * registered and the call operates synchronously; we
6794 			 * break and continue with the rest of the function and
6795 			 * wait for vds to return (i.e. after the request to
6796 			 * vds returns successfully, all writes completed prior
6797 			 * to the ioctl will have been flushed from the disk
6798 			 * write cache to persistent media.
6799 			 *
6800 			 * If a callback function is registered, we dispatch
6801 			 * the request on a task queue and return immediately.
6802 			 * The callback will deal with informing the calling
6803 			 * thread that the flush request is completed.
6804 			 */
6805 			if (dkc == NULL)
6806 				break;
6807 
6808 			/*
6809 			 * the asynchronous callback is only supported if
6810 			 * invoked from within the kernel
6811 			 */
6812 			if ((mode & FKIOCTL) == 0)
6813 				return (ENOTSUP);
6814 
6815 			dkarg = kmem_zalloc(sizeof (vdc_dk_arg_t), KM_SLEEP);
6816 
6817 			dkarg->mode = mode;
6818 			dkarg->dev = dev;
6819 			bcopy(dkc, &dkarg->dkc, sizeof (*dkc));
6820 
6821 			mutex_enter(&vdc->lock);
6822 			vdc->dkio_flush_pending++;
6823 			dkarg->vdc = vdc;
6824 			mutex_exit(&vdc->lock);
6825 
6826 			/* put the request on a task queue */
6827 			rv = taskq_dispatch(system_taskq, vdc_dkio_flush_cb,
6828 			    (void *)dkarg, DDI_SLEEP);
6829 			if (rv == NULL) {
6830 				/* clean up if dispatch fails */
6831 				mutex_enter(&vdc->lock);
6832 				vdc->dkio_flush_pending--;
6833 				mutex_exit(&vdc->lock);
6834 				kmem_free(dkarg, sizeof (vdc_dk_arg_t));
6835 			}
6836 
6837 			return (rv == NULL ? ENOMEM : 0);
6838 		}
6839 	}
6840 
6841 	/* catch programming error in vdc - should be a VD_OP_XXX ioctl */
6842 	ASSERT(iop->op != 0);
6843 
6844 	/* check if the vDisk server handles the operation for this vDisk */
6845 	if (VD_OP_SUPPORTED(vdc->operations, iop->op) == B_FALSE) {
6846 		DMSG(vdc, 0, "[%d] Unsupported VD_OP operation (0x%x)\n",
6847 		    vdc->instance, iop->op);
6848 		return (ENOTSUP);
6849 	}
6850 
6851 	/* LDC requires that the memory being mapped is 8-byte aligned */
6852 	alloc_len = P2ROUNDUP(len, sizeof (uint64_t));
6853 	DMSG(vdc, 1, "[%d] struct size %ld alloc %ld\n",
6854 	    instance, len, alloc_len);
6855 
6856 	if (alloc_len > 0)
6857 		mem_p = kmem_zalloc(alloc_len, KM_SLEEP);
6858 
6859 	/*
6860 	 * Call the conversion function for this ioctl which, if necessary,
6861 	 * converts from the Solaris format to the format ARC'ed
6862 	 * as part of the vDisk protocol (FWARC 2006/195)
6863 	 */
6864 	ASSERT(iop->convert != NULL);
6865 	rv = (iop->convert)(vdc, arg, mem_p, mode, VD_COPYIN);
6866 	if (rv != 0) {
6867 		DMSG(vdc, 0, "[%d] convert func returned %d for ioctl 0x%x\n",
6868 		    instance, rv, cmd);
6869 		if (mem_p != NULL)
6870 			kmem_free(mem_p, alloc_len);
6871 		return (rv);
6872 	}
6873 
6874 	/*
6875 	 * send request to vds to service the ioctl.
6876 	 */
6877 	rv = vdc_do_sync_op(vdc, iop->op, mem_p, alloc_len,
6878 	    VDCPART(dev), 0, CB_SYNC, (void *)(uint64_t)mode,
6879 	    VIO_both_dir, B_TRUE);
6880 
6881 	if (rv != 0) {
6882 		/*
6883 		 * This is not necessarily an error. The ioctl could
6884 		 * be returning a value such as ENOTTY to indicate
6885 		 * that the ioctl is not applicable.
6886 		 */
6887 		DMSG(vdc, 0, "[%d] vds returned %d for ioctl 0x%x\n",
6888 		    instance, rv, cmd);
6889 		if (mem_p != NULL)
6890 			kmem_free(mem_p, alloc_len);
6891 
6892 		return (rv);
6893 	}
6894 
6895 	/*
6896 	 * Call the conversion function (if it exists) for this ioctl
6897 	 * which converts from the format ARC'ed as part of the vDisk
6898 	 * protocol (FWARC 2006/195) back to a format understood by
6899 	 * the rest of Solaris.
6900 	 */
6901 	rv = (iop->convert)(vdc, mem_p, arg, mode, VD_COPYOUT);
6902 	if (rv != 0) {
6903 		DMSG(vdc, 0, "[%d] convert func returned %d for ioctl 0x%x\n",
6904 		    instance, rv, cmd);
6905 		if (mem_p != NULL)
6906 			kmem_free(mem_p, alloc_len);
6907 		return (rv);
6908 	}
6909 
6910 	if (mem_p != NULL)
6911 		kmem_free(mem_p, alloc_len);
6912 
6913 	return (rv);
6914 }
6915 
6916 /*
6917  * Function:
6918  *
6919  * Description:
6920  *	This is an empty conversion function used by ioctl calls which
6921  *	do not need to convert the data being passed in/out to userland
6922  */
6923 static int
6924 vdc_null_copy_func(vdc_t *vdc, void *from, void *to, int mode, int dir)
6925 {
6926 	_NOTE(ARGUNUSED(vdc))
6927 	_NOTE(ARGUNUSED(from))
6928 	_NOTE(ARGUNUSED(to))
6929 	_NOTE(ARGUNUSED(mode))
6930 	_NOTE(ARGUNUSED(dir))
6931 
6932 	return (0);
6933 }
6934 
6935 static int
6936 vdc_get_wce_convert(vdc_t *vdc, void *from, void *to,
6937     int mode, int dir)
6938 {
6939 	_NOTE(ARGUNUSED(vdc))
6940 
6941 	if (dir == VD_COPYIN)
6942 		return (0);		/* nothing to do */
6943 
6944 	if (ddi_copyout(from, to, sizeof (int), mode) != 0)
6945 		return (EFAULT);
6946 
6947 	return (0);
6948 }
6949 
6950 static int
6951 vdc_set_wce_convert(vdc_t *vdc, void *from, void *to,
6952     int mode, int dir)
6953 {
6954 	_NOTE(ARGUNUSED(vdc))
6955 
6956 	if (dir == VD_COPYOUT)
6957 		return (0);		/* nothing to do */
6958 
6959 	if (ddi_copyin(from, to, sizeof (int), mode) != 0)
6960 		return (EFAULT);
6961 
6962 	return (0);
6963 }
6964 
6965 /*
6966  * Function:
6967  *	vdc_get_vtoc_convert()
6968  *
6969  * Description:
6970  *	This routine performs the necessary convertions from the DKIOCGVTOC
6971  *	Solaris structure to the format defined in FWARC 2006/195.
6972  *
6973  *	In the struct vtoc definition, the timestamp field is marked as not
6974  *	supported so it is not part of vDisk protocol (FWARC 2006/195).
6975  *	However SVM uses that field to check it can write into the VTOC,
6976  *	so we fake up the info of that field.
6977  *
6978  * Arguments:
6979  *	vdc	- the vDisk client
6980  *	from	- the buffer containing the data to be copied from
6981  *	to	- the buffer to be copied to
6982  *	mode	- flags passed to ioctl() call
6983  *	dir	- the "direction" of the copy - VD_COPYIN or VD_COPYOUT
6984  *
6985  * Return Code:
6986  *	0	- Success
6987  *	ENXIO	- incorrect buffer passed in.
6988  *	EFAULT	- ddi_copyout routine encountered an error.
6989  */
6990 static int
6991 vdc_get_vtoc_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
6992 {
6993 	int		i;
6994 	void		*tmp_mem = NULL;
6995 	void		*tmp_memp;
6996 	struct vtoc	vt;
6997 	struct vtoc32	vt32;
6998 	int		copy_len = 0;
6999 	int		rv = 0;
7000 
7001 	if (dir != VD_COPYOUT)
7002 		return (0);	/* nothing to do */
7003 
7004 	if ((from == NULL) || (to == NULL))
7005 		return (ENXIO);
7006 
7007 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32)
7008 		copy_len = sizeof (struct vtoc32);
7009 	else
7010 		copy_len = sizeof (struct vtoc);
7011 
7012 	tmp_mem = kmem_alloc(copy_len, KM_SLEEP);
7013 
7014 	VD_VTOC2VTOC((vd_vtoc_t *)from, &vt);
7015 
7016 	/* fake the VTOC timestamp field */
7017 	for (i = 0; i < V_NUMPAR; i++) {
7018 		vt.timestamp[i] = vdc->vtoc->timestamp[i];
7019 	}
7020 
7021 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
7022 		/* LINTED E_ASSIGN_NARROW_CONV */
7023 		vtoctovtoc32(vt, vt32);
7024 		tmp_memp = &vt32;
7025 	} else {
7026 		tmp_memp = &vt;
7027 	}
7028 	rv = ddi_copyout(tmp_memp, to, copy_len, mode);
7029 	if (rv != 0)
7030 		rv = EFAULT;
7031 
7032 	kmem_free(tmp_mem, copy_len);
7033 	return (rv);
7034 }
7035 
7036 /*
7037  * Function:
7038  *	vdc_set_vtoc_convert()
7039  *
7040  * Description:
7041  *	This routine performs the necessary convertions from the DKIOCSVTOC
7042  *	Solaris structure to the format defined in FWARC 2006/195.
7043  *
7044  * Arguments:
7045  *	vdc	- the vDisk client
7046  *	from	- Buffer with data
7047  *	to	- Buffer where data is to be copied to
7048  *	mode	- flags passed to ioctl
7049  *	dir	- direction of copy (in or out)
7050  *
7051  * Return Code:
7052  *	0	- Success
7053  *	ENXIO	- Invalid buffer passed in
7054  *	EFAULT	- ddi_copyin of data failed
7055  */
7056 static int
7057 vdc_set_vtoc_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7058 {
7059 	_NOTE(ARGUNUSED(vdc))
7060 
7061 	void		*tmp_mem = NULL, *uvtoc;
7062 	struct vtoc	vt;
7063 	struct vtoc	*vtp = &vt;
7064 	vd_vtoc_t	vtvd;
7065 	int		copy_len = 0;
7066 	int		i, rv = 0;
7067 
7068 	if ((from == NULL) || (to == NULL))
7069 		return (ENXIO);
7070 
7071 	if (dir == VD_COPYIN)
7072 		uvtoc = from;
7073 	else
7074 		uvtoc = to;
7075 
7076 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32)
7077 		copy_len = sizeof (struct vtoc32);
7078 	else
7079 		copy_len = sizeof (struct vtoc);
7080 
7081 	tmp_mem = kmem_alloc(copy_len, KM_SLEEP);
7082 
7083 	rv = ddi_copyin(uvtoc, tmp_mem, copy_len, mode);
7084 	if (rv != 0) {
7085 		kmem_free(tmp_mem, copy_len);
7086 		return (EFAULT);
7087 	}
7088 
7089 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
7090 		vtoc32tovtoc((*(struct vtoc32 *)tmp_mem), vt);
7091 	} else {
7092 		vtp = tmp_mem;
7093 	}
7094 
7095 	if (dir == VD_COPYOUT) {
7096 		/*
7097 		 * The disk label may have changed. Revalidate the disk
7098 		 * geometry. This will also update the device nodes and
7099 		 * properties.
7100 		 */
7101 		vdc_validate(vdc);
7102 
7103 		/*
7104 		 * We also need to keep track of the timestamp fields.
7105 		 */
7106 		for (i = 0; i < V_NUMPAR; i++) {
7107 			vdc->vtoc->timestamp[i] = vtp->timestamp[i];
7108 		}
7109 
7110 		return (0);
7111 	}
7112 
7113 	VTOC2VD_VTOC(vtp, &vtvd);
7114 	bcopy(&vtvd, to, sizeof (vd_vtoc_t));
7115 	kmem_free(tmp_mem, copy_len);
7116 
7117 	return (0);
7118 }
7119 
7120 /*
7121  * Function:
7122  *	vdc_get_geom_convert()
7123  *
7124  * Description:
7125  *	This routine performs the necessary convertions from the DKIOCGGEOM,
7126  *	DKIOCG_PHYSGEOM and DKIOG_VIRTGEOM Solaris structures to the format
7127  *	defined in FWARC 2006/195
7128  *
7129  * Arguments:
7130  *	vdc	- the vDisk client
7131  *	from	- Buffer with data
7132  *	to	- Buffer where data is to be copied to
7133  *	mode	- flags passed to ioctl
7134  *	dir	- direction of copy (in or out)
7135  *
7136  * Return Code:
7137  *	0	- Success
7138  *	ENXIO	- Invalid buffer passed in
7139  *	EFAULT	- ddi_copyout of data failed
7140  */
7141 static int
7142 vdc_get_geom_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7143 {
7144 	_NOTE(ARGUNUSED(vdc))
7145 
7146 	struct dk_geom	geom;
7147 	int	copy_len = sizeof (struct dk_geom);
7148 	int	rv = 0;
7149 
7150 	if (dir != VD_COPYOUT)
7151 		return (0);	/* nothing to do */
7152 
7153 	if ((from == NULL) || (to == NULL))
7154 		return (ENXIO);
7155 
7156 	VD_GEOM2DK_GEOM((vd_geom_t *)from, &geom);
7157 	rv = ddi_copyout(&geom, to, copy_len, mode);
7158 	if (rv != 0)
7159 		rv = EFAULT;
7160 
7161 	return (rv);
7162 }
7163 
7164 /*
7165  * Function:
7166  *	vdc_set_geom_convert()
7167  *
7168  * Description:
7169  *	This routine performs the necessary convertions from the DKIOCSGEOM
7170  *	Solaris structure to the format defined in FWARC 2006/195.
7171  *
7172  * Arguments:
7173  *	vdc	- the vDisk client
7174  *	from	- Buffer with data
7175  *	to	- Buffer where data is to be copied to
7176  *	mode	- flags passed to ioctl
7177  *	dir	- direction of copy (in or out)
7178  *
7179  * Return Code:
7180  *	0	- Success
7181  *	ENXIO	- Invalid buffer passed in
7182  *	EFAULT	- ddi_copyin of data failed
7183  */
7184 static int
7185 vdc_set_geom_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7186 {
7187 	_NOTE(ARGUNUSED(vdc))
7188 
7189 	vd_geom_t	vdgeom;
7190 	void		*tmp_mem = NULL;
7191 	int		copy_len = sizeof (struct dk_geom);
7192 	int		rv = 0;
7193 
7194 	if (dir != VD_COPYIN)
7195 		return (0);	/* nothing to do */
7196 
7197 	if ((from == NULL) || (to == NULL))
7198 		return (ENXIO);
7199 
7200 	tmp_mem = kmem_alloc(copy_len, KM_SLEEP);
7201 
7202 	rv = ddi_copyin(from, tmp_mem, copy_len, mode);
7203 	if (rv != 0) {
7204 		kmem_free(tmp_mem, copy_len);
7205 		return (EFAULT);
7206 	}
7207 	DK_GEOM2VD_GEOM((struct dk_geom *)tmp_mem, &vdgeom);
7208 	bcopy(&vdgeom, to, sizeof (vdgeom));
7209 	kmem_free(tmp_mem, copy_len);
7210 
7211 	return (0);
7212 }
7213 
7214 static int
7215 vdc_get_efi_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7216 {
7217 	_NOTE(ARGUNUSED(vdc))
7218 
7219 	vd_efi_t	*vd_efi;
7220 	dk_efi_t	dk_efi;
7221 	int		rv = 0;
7222 	void		*uaddr;
7223 
7224 	if ((from == NULL) || (to == NULL))
7225 		return (ENXIO);
7226 
7227 	if (dir == VD_COPYIN) {
7228 
7229 		vd_efi = (vd_efi_t *)to;
7230 
7231 		rv = ddi_copyin(from, &dk_efi, sizeof (dk_efi_t), mode);
7232 		if (rv != 0)
7233 			return (EFAULT);
7234 
7235 		vd_efi->lba = dk_efi.dki_lba;
7236 		vd_efi->length = dk_efi.dki_length;
7237 		bzero(vd_efi->data, vd_efi->length);
7238 
7239 	} else {
7240 
7241 		rv = ddi_copyin(to, &dk_efi, sizeof (dk_efi_t), mode);
7242 		if (rv != 0)
7243 			return (EFAULT);
7244 
7245 		uaddr = dk_efi.dki_data;
7246 
7247 		dk_efi.dki_data = kmem_alloc(dk_efi.dki_length, KM_SLEEP);
7248 
7249 		VD_EFI2DK_EFI((vd_efi_t *)from, &dk_efi);
7250 
7251 		rv = ddi_copyout(dk_efi.dki_data, uaddr, dk_efi.dki_length,
7252 		    mode);
7253 		if (rv != 0)
7254 			return (EFAULT);
7255 
7256 		kmem_free(dk_efi.dki_data, dk_efi.dki_length);
7257 	}
7258 
7259 	return (0);
7260 }
7261 
7262 static int
7263 vdc_set_efi_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7264 {
7265 	_NOTE(ARGUNUSED(vdc))
7266 
7267 	dk_efi_t	dk_efi;
7268 	void		*uaddr;
7269 
7270 	if (dir == VD_COPYOUT) {
7271 		/*
7272 		 * The disk label may have changed. Revalidate the disk
7273 		 * geometry. This will also update the device nodes and
7274 		 * properties.
7275 		 */
7276 		vdc_validate(vdc);
7277 		return (0);
7278 	}
7279 
7280 	if ((from == NULL) || (to == NULL))
7281 		return (ENXIO);
7282 
7283 	if (ddi_copyin(from, &dk_efi, sizeof (dk_efi_t), mode) != 0)
7284 		return (EFAULT);
7285 
7286 	uaddr = dk_efi.dki_data;
7287 
7288 	dk_efi.dki_data = kmem_alloc(dk_efi.dki_length, KM_SLEEP);
7289 
7290 	if (ddi_copyin(uaddr, dk_efi.dki_data, dk_efi.dki_length, mode) != 0)
7291 		return (EFAULT);
7292 
7293 	DK_EFI2VD_EFI(&dk_efi, (vd_efi_t *)to);
7294 
7295 	kmem_free(dk_efi.dki_data, dk_efi.dki_length);
7296 
7297 	return (0);
7298 }
7299 
7300 
7301 /* -------------------------------------------------------------------------- */
7302 
7303 /*
7304  * Function:
7305  *	vdc_create_fake_geometry()
7306  *
7307  * Description:
7308  *	This routine fakes up the disk info needed for some DKIO ioctls such
7309  *	as DKIOCINFO and DKIOCGMEDIAINFO [just like lofi(7D) and ramdisk(7D) do]
7310  *
7311  *	Note: This function must not be called until the vDisk attributes have
7312  *	been exchanged as part of the handshake with the vDisk server.
7313  *
7314  * Arguments:
7315  *	vdc	- soft state pointer for this instance of the device driver.
7316  *
7317  * Return Code:
7318  *	none.
7319  */
7320 static void
7321 vdc_create_fake_geometry(vdc_t *vdc)
7322 {
7323 	ASSERT(vdc != NULL);
7324 	ASSERT(vdc->max_xfer_sz != 0);
7325 
7326 	/*
7327 	 * DKIOCINFO support
7328 	 */
7329 	if (vdc->cinfo == NULL)
7330 		vdc->cinfo = kmem_zalloc(sizeof (struct dk_cinfo), KM_SLEEP);
7331 
7332 	(void) strcpy(vdc->cinfo->dki_cname, VDC_DRIVER_NAME);
7333 	(void) strcpy(vdc->cinfo->dki_dname, VDC_DRIVER_NAME);
7334 	/* max_xfer_sz is #blocks so we don't need to divide by DEV_BSIZE */
7335 	vdc->cinfo->dki_maxtransfer = vdc->max_xfer_sz;
7336 
7337 	/*
7338 	 * We set the controller type to DKC_SCSI_CCS only if the VD_OP_SCSICMD
7339 	 * operation is supported, otherwise the controller type is DKC_DIRECT.
7340 	 * Version 1.0 does not support the VD_OP_SCSICMD operation, so the
7341 	 * controller type is always DKC_DIRECT in that case.
7342 	 *
7343 	 * If the virtual disk is backed by a physical CD/DVD device or
7344 	 * an ISO image, modify the controller type to indicate this
7345 	 */
7346 	switch (vdc->vdisk_media) {
7347 	case VD_MEDIA_CD:
7348 	case VD_MEDIA_DVD:
7349 		vdc->cinfo->dki_ctype = DKC_CDROM;
7350 		break;
7351 	case VD_MEDIA_FIXED:
7352 		if (VD_OP_SUPPORTED(vdc->operations, VD_OP_SCSICMD))
7353 			vdc->cinfo->dki_ctype = DKC_SCSI_CCS;
7354 		else
7355 			vdc->cinfo->dki_ctype = DKC_DIRECT;
7356 		break;
7357 	default:
7358 		/* in the case of v1.0 we default to a fixed disk */
7359 		vdc->cinfo->dki_ctype = DKC_DIRECT;
7360 		break;
7361 	}
7362 	vdc->cinfo->dki_flags = DKI_FMTVOL;
7363 	vdc->cinfo->dki_cnum = 0;
7364 	vdc->cinfo->dki_addr = 0;
7365 	vdc->cinfo->dki_space = 0;
7366 	vdc->cinfo->dki_prio = 0;
7367 	vdc->cinfo->dki_vec = 0;
7368 	vdc->cinfo->dki_unit = vdc->instance;
7369 	vdc->cinfo->dki_slave = 0;
7370 	/*
7371 	 * The partition number will be created on the fly depending on the
7372 	 * actual slice (i.e. minor node) that is used to request the data.
7373 	 */
7374 	vdc->cinfo->dki_partition = 0;
7375 
7376 	/*
7377 	 * DKIOCGMEDIAINFO support
7378 	 */
7379 	if (vdc->minfo == NULL)
7380 		vdc->minfo = kmem_zalloc(sizeof (struct dk_minfo), KM_SLEEP);
7381 
7382 	if (vio_ver_is_supported(vdc->ver, 1, 1)) {
7383 		vdc->minfo->dki_media_type =
7384 		    VD_MEDIATYPE2DK_MEDIATYPE(vdc->vdisk_media);
7385 	} else {
7386 		vdc->minfo->dki_media_type = DK_FIXED_DISK;
7387 	}
7388 
7389 	vdc->minfo->dki_capacity = vdc->vdisk_size;
7390 	vdc->minfo->dki_lbsize = vdc->block_size;
7391 }
7392 
7393 static ushort_t
7394 vdc_lbl2cksum(struct dk_label *label)
7395 {
7396 	int	count;
7397 	ushort_t sum, *sp;
7398 
7399 	count =	(sizeof (struct dk_label)) / (sizeof (short)) - 1;
7400 	sp = (ushort_t *)label;
7401 	sum = 0;
7402 	while (count--) {
7403 		sum ^= *sp++;
7404 	}
7405 
7406 	return (sum);
7407 }
7408 
7409 /*
7410  * Function:
7411  *	vdc_validate_geometry
7412  *
7413  * Description:
7414  *	This routine discovers the label and geometry of the disk. It stores
7415  *	the disk label and related information in the vdc structure. If it
7416  *	fails to validate the geometry or to discover the disk label then
7417  *	the label is marked as unknown (VD_DISK_LABEL_UNK).
7418  *
7419  * Arguments:
7420  *	vdc	- soft state pointer for this instance of the device driver.
7421  *
7422  * Return Code:
7423  *	0	- success.
7424  *	EINVAL	- unknown disk label.
7425  *	ENOTSUP	- geometry not applicable (EFI label).
7426  *	EIO	- error accessing the disk.
7427  */
7428 static int
7429 vdc_validate_geometry(vdc_t *vdc)
7430 {
7431 	buf_t	*buf;	/* BREAD requests need to be in a buf_t structure */
7432 	dev_t	dev;
7433 	int	rv, rval;
7434 	struct dk_label label;
7435 	struct dk_geom geom;
7436 	struct vtoc vtoc;
7437 	efi_gpt_t *gpt;
7438 	efi_gpe_t *gpe;
7439 	vd_efi_dev_t edev;
7440 
7441 	ASSERT(vdc != NULL);
7442 	ASSERT(vdc->vtoc != NULL && vdc->geom != NULL);
7443 	ASSERT(MUTEX_HELD(&vdc->lock));
7444 
7445 	mutex_exit(&vdc->lock);
7446 
7447 	dev = makedevice(ddi_driver_major(vdc->dip),
7448 	    VD_MAKE_DEV(vdc->instance, 0));
7449 
7450 	rv = vd_process_ioctl(dev, DKIOCGGEOM, (caddr_t)&geom, FKIOCTL, &rval);
7451 	if (rv == 0)
7452 		rv = vd_process_ioctl(dev, DKIOCGVTOC, (caddr_t)&vtoc,
7453 		    FKIOCTL, &rval);
7454 
7455 	if (rv == ENOTSUP) {
7456 		/*
7457 		 * If the device does not support VTOC then we try
7458 		 * to read an EFI label.
7459 		 *
7460 		 * We need to know the block size and the disk size to
7461 		 * be able to read an EFI label.
7462 		 */
7463 		if (vdc->vdisk_size == 0) {
7464 			if ((rv = vdc_check_capacity(vdc)) != 0) {
7465 				mutex_enter(&vdc->lock);
7466 				vdc_store_label_unk(vdc);
7467 				return (rv);
7468 			}
7469 		}
7470 
7471 		VD_EFI_DEV_SET(edev, vdc, vd_process_efi_ioctl);
7472 
7473 		rv = vd_efi_alloc_and_read(&edev, &gpt, &gpe);
7474 
7475 		if (rv) {
7476 			DMSG(vdc, 0, "[%d] Failed to get EFI (err=%d)",
7477 			    vdc->instance, rv);
7478 			mutex_enter(&vdc->lock);
7479 			vdc_store_label_unk(vdc);
7480 			return (EIO);
7481 		}
7482 
7483 		mutex_enter(&vdc->lock);
7484 		vdc_store_label_efi(vdc, gpt, gpe);
7485 		vd_efi_free(&edev, gpt, gpe);
7486 		return (ENOTSUP);
7487 	}
7488 
7489 	if (rv != 0) {
7490 		DMSG(vdc, 0, "[%d] Failed to get VTOC (err=%d)",
7491 		    vdc->instance, rv);
7492 		mutex_enter(&vdc->lock);
7493 		vdc_store_label_unk(vdc);
7494 		if (rv != EINVAL)
7495 			rv = EIO;
7496 		return (rv);
7497 	}
7498 
7499 	/* check that geometry and vtoc are valid */
7500 	if (geom.dkg_nhead == 0 || geom.dkg_nsect == 0 ||
7501 	    vtoc.v_sanity != VTOC_SANE) {
7502 		mutex_enter(&vdc->lock);
7503 		vdc_store_label_unk(vdc);
7504 		return (EINVAL);
7505 	}
7506 
7507 	/*
7508 	 * We have a disk and a valid VTOC. However this does not mean
7509 	 * that the disk currently have a VTOC label. The returned VTOC may
7510 	 * be a default VTOC to be used for configuring the disk (this is
7511 	 * what is done for disk image). So we read the label from the
7512 	 * beginning of the disk to ensure we really have a VTOC label.
7513 	 *
7514 	 * FUTURE: This could be the default way for reading the VTOC
7515 	 * from the disk as opposed to sending the VD_OP_GET_VTOC
7516 	 * to the server. This will be the default if vdc is implemented
7517 	 * ontop of cmlb.
7518 	 */
7519 
7520 	/*
7521 	 * Single slice disk does not support read using an absolute disk
7522 	 * offset so we just rely on the DKIOCGVTOC ioctl in that case.
7523 	 */
7524 	if (vdc->vdisk_type == VD_DISK_TYPE_SLICE) {
7525 		mutex_enter(&vdc->lock);
7526 		if (vtoc.v_nparts != 1) {
7527 			vdc_store_label_unk(vdc);
7528 			return (EINVAL);
7529 		}
7530 		vdc_store_label_vtoc(vdc, &geom, &vtoc);
7531 		return (0);
7532 	}
7533 
7534 	if (vtoc.v_nparts != V_NUMPAR) {
7535 		mutex_enter(&vdc->lock);
7536 		vdc_store_label_unk(vdc);
7537 		return (EINVAL);
7538 	}
7539 
7540 	/*
7541 	 * Read disk label from start of disk
7542 	 */
7543 	buf = kmem_alloc(sizeof (buf_t), KM_SLEEP);
7544 	bioinit(buf);
7545 	buf->b_un.b_addr = (caddr_t)&label;
7546 	buf->b_bcount = DK_LABEL_SIZE;
7547 	buf->b_flags = B_BUSY | B_READ;
7548 	buf->b_dev = cmpdev(dev);
7549 	rv = vdc_send_request(vdc, VD_OP_BREAD, (caddr_t)&label,
7550 	    DK_LABEL_SIZE, VD_SLICE_NONE, 0, CB_STRATEGY, buf, VIO_read_dir);
7551 	if (rv) {
7552 		DMSG(vdc, 1, "[%d] Failed to read disk block 0\n",
7553 		    vdc->instance);
7554 	} else {
7555 		rv = biowait(buf);
7556 		biofini(buf);
7557 	}
7558 	kmem_free(buf, sizeof (buf_t));
7559 
7560 	if (rv != 0 || label.dkl_magic != DKL_MAGIC ||
7561 	    label.dkl_cksum != vdc_lbl2cksum(&label)) {
7562 		DMSG(vdc, 1, "[%d] Got VTOC with invalid label\n",
7563 		    vdc->instance);
7564 		mutex_enter(&vdc->lock);
7565 		vdc_store_label_unk(vdc);
7566 		return (EINVAL);
7567 	}
7568 
7569 	mutex_enter(&vdc->lock);
7570 	vdc_store_label_vtoc(vdc, &geom, &vtoc);
7571 	return (0);
7572 }
7573 
7574 /*
7575  * Function:
7576  *	vdc_validate
7577  *
7578  * Description:
7579  *	This routine discovers the label of the disk and create the
7580  *	appropriate device nodes if the label has changed.
7581  *
7582  * Arguments:
7583  *	vdc	- soft state pointer for this instance of the device driver.
7584  *
7585  * Return Code:
7586  *	none.
7587  */
7588 static void
7589 vdc_validate(vdc_t *vdc)
7590 {
7591 	vd_disk_label_t old_label;
7592 	vd_slice_t old_slice[V_NUMPAR];
7593 	int rv;
7594 
7595 	ASSERT(!MUTEX_HELD(&vdc->lock));
7596 
7597 	mutex_enter(&vdc->lock);
7598 
7599 	/* save the current label and vtoc */
7600 	old_label = vdc->vdisk_label;
7601 	bcopy(vdc->slice, &old_slice, sizeof (vd_slice_t) * V_NUMPAR);
7602 
7603 	/* check the geometry */
7604 	(void) vdc_validate_geometry(vdc);
7605 
7606 	/* if the disk label has changed, update device nodes */
7607 	if (vdc->vdisk_label != old_label) {
7608 
7609 		if (vdc->vdisk_label == VD_DISK_LABEL_EFI)
7610 			rv = vdc_create_device_nodes_efi(vdc);
7611 		else
7612 			rv = vdc_create_device_nodes_vtoc(vdc);
7613 
7614 		if (rv != 0) {
7615 			DMSG(vdc, 0, "![%d] Failed to update device nodes",
7616 			    vdc->instance);
7617 		}
7618 	}
7619 
7620 	/* if the vtoc has changed, update device nodes properties */
7621 	if (bcmp(vdc->slice, &old_slice, sizeof (vd_slice_t) * V_NUMPAR) != 0) {
7622 
7623 		if (vdc_create_device_nodes_props(vdc) != 0) {
7624 			DMSG(vdc, 0, "![%d] Failed to update device nodes"
7625 			    " properties", vdc->instance);
7626 		}
7627 	}
7628 
7629 	mutex_exit(&vdc->lock);
7630 }
7631 
7632 static void
7633 vdc_validate_task(void *arg)
7634 {
7635 	vdc_t *vdc = (vdc_t *)arg;
7636 
7637 	vdc_validate(vdc);
7638 
7639 	mutex_enter(&vdc->lock);
7640 	ASSERT(vdc->validate_pending > 0);
7641 	vdc->validate_pending--;
7642 	mutex_exit(&vdc->lock);
7643 }
7644 
7645 /*
7646  * Function:
7647  *	vdc_setup_devid()
7648  *
7649  * Description:
7650  *	This routine discovers the devid of a vDisk. It requests the devid of
7651  *	the underlying device from the vDisk server, builds an encapsulated
7652  *	devid based on the retrieved devid and registers that new devid to
7653  *	the vDisk.
7654  *
7655  * Arguments:
7656  *	vdc	- soft state pointer for this instance of the device driver.
7657  *
7658  * Return Code:
7659  *	0	- A devid was succesfully registered for the vDisk
7660  */
7661 static int
7662 vdc_setup_devid(vdc_t *vdc)
7663 {
7664 	int rv;
7665 	vd_devid_t *vd_devid;
7666 	size_t bufsize, bufid_len;
7667 
7668 	/*
7669 	 * At first sight, we don't know the size of the devid that the
7670 	 * server will return but this size will be encoded into the
7671 	 * reply. So we do a first request using a default size then we
7672 	 * check if this size was large enough. If not then we do a second
7673 	 * request with the correct size returned by the server. Note that
7674 	 * ldc requires size to be 8-byte aligned.
7675 	 */
7676 	bufsize = P2ROUNDUP(VD_DEVID_SIZE(VD_DEVID_DEFAULT_LEN),
7677 	    sizeof (uint64_t));
7678 	vd_devid = kmem_zalloc(bufsize, KM_SLEEP);
7679 	bufid_len = bufsize - sizeof (vd_efi_t) - 1;
7680 
7681 	rv = vdc_do_sync_op(vdc, VD_OP_GET_DEVID, (caddr_t)vd_devid,
7682 	    bufsize, 0, 0, CB_SYNC, 0, VIO_both_dir, B_TRUE);
7683 
7684 	DMSG(vdc, 2, "sync_op returned %d\n", rv);
7685 
7686 	if (rv) {
7687 		kmem_free(vd_devid, bufsize);
7688 		return (rv);
7689 	}
7690 
7691 	if (vd_devid->length > bufid_len) {
7692 		/*
7693 		 * The returned devid is larger than the buffer used. Try again
7694 		 * with a buffer with the right size.
7695 		 */
7696 		kmem_free(vd_devid, bufsize);
7697 		bufsize = P2ROUNDUP(VD_DEVID_SIZE(vd_devid->length),
7698 		    sizeof (uint64_t));
7699 		vd_devid = kmem_zalloc(bufsize, KM_SLEEP);
7700 		bufid_len = bufsize - sizeof (vd_efi_t) - 1;
7701 
7702 		rv = vdc_do_sync_op(vdc, VD_OP_GET_DEVID,
7703 		    (caddr_t)vd_devid, bufsize, 0, 0, CB_SYNC, 0,
7704 		    VIO_both_dir, B_TRUE);
7705 
7706 		if (rv) {
7707 			kmem_free(vd_devid, bufsize);
7708 			return (rv);
7709 		}
7710 	}
7711 
7712 	/*
7713 	 * The virtual disk should have the same device id as the one associated
7714 	 * with the physical disk it is mapped on, otherwise sharing a disk
7715 	 * between a LDom and a non-LDom may not work (for example for a shared
7716 	 * SVM disk set).
7717 	 *
7718 	 * The DDI framework does not allow creating a device id with any
7719 	 * type so we first create a device id of type DEVID_ENCAP and then
7720 	 * we restore the orignal type of the physical device.
7721 	 */
7722 
7723 	DMSG(vdc, 2, ": devid length = %d\n", vd_devid->length);
7724 
7725 	/* build an encapsulated devid based on the returned devid */
7726 	if (ddi_devid_init(vdc->dip, DEVID_ENCAP, vd_devid->length,
7727 	    vd_devid->id, &vdc->devid) != DDI_SUCCESS) {
7728 		DMSG(vdc, 1, "[%d] Fail to created devid\n", vdc->instance);
7729 		kmem_free(vd_devid, bufsize);
7730 		return (1);
7731 	}
7732 
7733 	DEVID_FORMTYPE((impl_devid_t *)vdc->devid, vd_devid->type);
7734 
7735 	ASSERT(ddi_devid_valid(vdc->devid) == DDI_SUCCESS);
7736 
7737 	kmem_free(vd_devid, bufsize);
7738 
7739 	if (ddi_devid_register(vdc->dip, vdc->devid) != DDI_SUCCESS) {
7740 		DMSG(vdc, 1, "[%d] Fail to register devid\n", vdc->instance);
7741 		return (1);
7742 	}
7743 
7744 	return (0);
7745 }
7746 
7747 static void
7748 vdc_store_label_efi(vdc_t *vdc, efi_gpt_t *gpt, efi_gpe_t *gpe)
7749 {
7750 	int i, nparts;
7751 
7752 	ASSERT(MUTEX_HELD(&vdc->lock));
7753 
7754 	vdc->vdisk_label = VD_DISK_LABEL_EFI;
7755 	bzero(vdc->vtoc, sizeof (struct vtoc));
7756 	bzero(vdc->geom, sizeof (struct dk_geom));
7757 	bzero(vdc->slice, sizeof (vd_slice_t) * V_NUMPAR);
7758 
7759 	nparts = gpt->efi_gpt_NumberOfPartitionEntries;
7760 
7761 	for (i = 0; i < nparts && i < VD_EFI_WD_SLICE; i++) {
7762 
7763 		if (gpe[i].efi_gpe_StartingLBA == 0 ||
7764 		    gpe[i].efi_gpe_EndingLBA == 0) {
7765 			continue;
7766 		}
7767 
7768 		vdc->slice[i].start = gpe[i].efi_gpe_StartingLBA;
7769 		vdc->slice[i].nblocks = gpe[i].efi_gpe_EndingLBA -
7770 		    gpe[i].efi_gpe_StartingLBA + 1;
7771 	}
7772 
7773 	ASSERT(vdc->vdisk_size != 0);
7774 	vdc->slice[VD_EFI_WD_SLICE].start = 0;
7775 	vdc->slice[VD_EFI_WD_SLICE].nblocks = vdc->vdisk_size;
7776 
7777 }
7778 
7779 static void
7780 vdc_store_label_vtoc(vdc_t *vdc, struct dk_geom *geom, struct vtoc *vtoc)
7781 {
7782 	int i;
7783 
7784 	ASSERT(MUTEX_HELD(&vdc->lock));
7785 	ASSERT(vdc->block_size == vtoc->v_sectorsz);
7786 
7787 	vdc->vdisk_label = VD_DISK_LABEL_VTOC;
7788 	bcopy(vtoc, vdc->vtoc, sizeof (struct vtoc));
7789 	bcopy(geom, vdc->geom, sizeof (struct dk_geom));
7790 	bzero(vdc->slice, sizeof (vd_slice_t) * V_NUMPAR);
7791 
7792 	for (i = 0; i < vtoc->v_nparts; i++) {
7793 		vdc->slice[i].start = vtoc->v_part[i].p_start;
7794 		vdc->slice[i].nblocks = vtoc->v_part[i].p_size;
7795 	}
7796 }
7797 
7798 static void
7799 vdc_store_label_unk(vdc_t *vdc)
7800 {
7801 	ASSERT(MUTEX_HELD(&vdc->lock));
7802 
7803 	vdc->vdisk_label = VD_DISK_LABEL_UNK;
7804 	bzero(vdc->vtoc, sizeof (struct vtoc));
7805 	bzero(vdc->geom, sizeof (struct dk_geom));
7806 	bzero(vdc->slice, sizeof (vd_slice_t) * V_NUMPAR);
7807 }
7808