xref: /titanic_44/usr/src/uts/sun4v/io/vdc.c (revision a31148363f598def767ac48c5d82e1572e44b935)
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 2010 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 /*
28  * LDoms virtual disk client (vdc) device driver
29  *
30  * This driver runs on a guest logical domain and communicates with the virtual
31  * disk server (vds) driver running on the service domain which is exporting
32  * virtualized "disks" to the guest logical domain.
33  *
34  * The driver can be divided into four sections:
35  *
36  * 1) generic device driver housekeeping
37  *	_init, _fini, attach, detach, ops structures, etc.
38  *
39  * 2) communication channel setup
40  *	Setup the communications link over the LDC channel that vdc uses to
41  *	talk to the vDisk server. Initialise the descriptor ring which
42  *	allows the LDC clients to transfer data via memory mappings.
43  *
44  * 3) Support exported to upper layers (filesystems, etc)
45  *	The upper layers call into vdc via strategy(9E) and DKIO(7I)
46  *	ioctl calls. vdc will copy the data to be written to the descriptor
47  *	ring or maps the buffer to store the data read by the vDisk
48  *	server into the descriptor ring. It then sends a message to the
49  *	vDisk server requesting it to complete the operation.
50  *
51  * 4) Handling responses from vDisk server.
52  *	The vDisk server will ACK some or all of the messages vdc sends to it
53  *	(this is configured during the handshake). Upon receipt of an ACK
54  *	vdc will check the descriptor ring and signal to the upper layer
55  *	code waiting on the IO.
56  */
57 
58 #include <sys/atomic.h>
59 #include <sys/conf.h>
60 #include <sys/disp.h>
61 #include <sys/ddi.h>
62 #include <sys/dkio.h>
63 #include <sys/efi_partition.h>
64 #include <sys/fcntl.h>
65 #include <sys/file.h>
66 #include <sys/kstat.h>
67 #include <sys/mach_descrip.h>
68 #include <sys/modctl.h>
69 #include <sys/mdeg.h>
70 #include <sys/note.h>
71 #include <sys/open.h>
72 #include <sys/random.h>
73 #include <sys/sdt.h>
74 #include <sys/stat.h>
75 #include <sys/sunddi.h>
76 #include <sys/types.h>
77 #include <sys/promif.h>
78 #include <sys/var.h>
79 #include <sys/vtoc.h>
80 #include <sys/archsystm.h>
81 #include <sys/sysmacros.h>
82 
83 #include <sys/cdio.h>
84 #include <sys/dktp/fdisk.h>
85 #include <sys/dktp/dadkio.h>
86 #include <sys/fs/dv_node.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 #define	VD_OLDVTOC_LIMIT	0x7fffffff
103 
104 /*
105  * function prototypes
106  */
107 
108 /* standard driver functions */
109 static int	vdc_open(dev_t *dev, int flag, int otyp, cred_t *cred);
110 static int	vdc_close(dev_t dev, int flag, int otyp, cred_t *cred);
111 static int	vdc_strategy(struct buf *buf);
112 static int	vdc_print(dev_t dev, char *str);
113 static int	vdc_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblk);
114 static int	vdc_read(dev_t dev, struct uio *uio, cred_t *cred);
115 static int	vdc_write(dev_t dev, struct uio *uio, cred_t *cred);
116 static int	vdc_ioctl(dev_t dev, int cmd, intptr_t arg, int mode,
117 			cred_t *credp, int *rvalp);
118 static int	vdc_aread(dev_t dev, struct aio_req *aio, cred_t *cred);
119 static int	vdc_awrite(dev_t dev, struct aio_req *aio, cred_t *cred);
120 
121 static int	vdc_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd,
122 			void *arg, void **resultp);
123 static int	vdc_attach(dev_info_t *dip, ddi_attach_cmd_t cmd);
124 static int	vdc_detach(dev_info_t *dip, ddi_detach_cmd_t cmd);
125 static int	vdc_prop_op(dev_t dev, dev_info_t *dip, ddi_prop_op_t prop_op,
126 		    int mod_flags, char *name, caddr_t valuep, int *lengthp);
127 
128 /* setup */
129 static void	vdc_min(struct buf *bufp);
130 static int	vdc_send(vdc_t *vdc, caddr_t pkt, size_t *msglen);
131 static int	vdc_do_ldc_init(vdc_t *vdc, vdc_server_t *srvr);
132 static int	vdc_start_ldc_connection(vdc_t *vdc);
133 static int	vdc_create_device_nodes(vdc_t *vdc);
134 static int	vdc_create_device_nodes_efi(vdc_t *vdc);
135 static int	vdc_create_device_nodes_vtoc(vdc_t *vdc);
136 static void	vdc_create_io_kstats(vdc_t *vdc);
137 static void	vdc_create_err_kstats(vdc_t *vdc);
138 static void	vdc_set_err_kstats(vdc_t *vdc);
139 static int	vdc_get_md_node(dev_info_t *dip, md_t **mdpp,
140 		    mde_cookie_t *vd_nodep);
141 static int	vdc_init_ports(vdc_t *vdc, md_t *mdp, mde_cookie_t vd_nodep);
142 static void	vdc_fini_ports(vdc_t *vdc);
143 static void	vdc_switch_server(vdc_t *vdcp);
144 static int	vdc_do_ldc_up(vdc_t *vdc);
145 static void	vdc_terminate_ldc(vdc_t *vdc, vdc_server_t *srvr);
146 static int	vdc_init_descriptor_ring(vdc_t *vdc);
147 static void	vdc_destroy_descriptor_ring(vdc_t *vdc);
148 static int	vdc_setup_devid(vdc_t *vdc);
149 static void	vdc_store_label_efi(vdc_t *, efi_gpt_t *, efi_gpe_t *);
150 static void	vdc_store_label_vtoc(vdc_t *, struct dk_geom *,
151 		    struct extvtoc *);
152 static void	vdc_store_label_unk(vdc_t *vdc);
153 static boolean_t vdc_is_opened(vdc_t *vdc);
154 static void	vdc_update_size(vdc_t *vdc, size_t, size_t, size_t);
155 static int	vdc_update_vio_bsize(vdc_t *vdc, uint32_t);
156 
157 /* handshake with vds */
158 static int		vdc_init_ver_negotiation(vdc_t *vdc, vio_ver_t ver);
159 static int		vdc_ver_negotiation(vdc_t *vdcp);
160 static int		vdc_init_attr_negotiation(vdc_t *vdc);
161 static int		vdc_attr_negotiation(vdc_t *vdcp);
162 static int		vdc_init_dring_negotiate(vdc_t *vdc);
163 static int		vdc_dring_negotiation(vdc_t *vdcp);
164 static int		vdc_send_rdx(vdc_t *vdcp);
165 static int		vdc_rdx_exchange(vdc_t *vdcp);
166 static boolean_t	vdc_is_supported_version(vio_ver_msg_t *ver_msg);
167 
168 /* processing incoming messages from vDisk server */
169 static void	vdc_process_msg_thread(vdc_t *vdc);
170 static int	vdc_recv(vdc_t *vdc, vio_msg_t *msgp, size_t *nbytesp);
171 
172 static uint_t	vdc_handle_cb(uint64_t event, caddr_t arg);
173 static int	vdc_process_data_msg(vdc_t *vdc, vio_msg_t *msg);
174 static int	vdc_handle_ver_msg(vdc_t *vdc, vio_ver_msg_t *ver_msg);
175 static int	vdc_handle_attr_msg(vdc_t *vdc, vd_attr_msg_t *attr_msg);
176 static int	vdc_handle_dring_reg_msg(vdc_t *vdc, vio_dring_reg_msg_t *msg);
177 static int 	vdc_send_request(vdc_t *vdcp, int operation,
178 		    caddr_t addr, size_t nbytes, int slice, diskaddr_t offset,
179 		    buf_t *bufp, vio_desc_direction_t dir, int flags);
180 static int	vdc_map_to_shared_dring(vdc_t *vdcp, int idx);
181 static int 	vdc_populate_descriptor(vdc_t *vdcp, int operation,
182 		    caddr_t addr, size_t nbytes, int slice, diskaddr_t offset,
183 		    buf_t *bufp, vio_desc_direction_t dir, int flags);
184 static int 	vdc_do_sync_op(vdc_t *vdcp, int operation, caddr_t addr,
185 		    size_t nbytes, int slice, diskaddr_t offset,
186 		    vio_desc_direction_t dir, boolean_t);
187 static int	vdc_do_op(vdc_t *vdc, int op, caddr_t addr, size_t nbytes,
188 		    int slice, diskaddr_t offset, struct buf *bufp,
189 		    vio_desc_direction_t dir, int flags);
190 
191 static int	vdc_wait_for_response(vdc_t *vdcp, vio_msg_t *msgp);
192 static int	vdc_drain_response(vdc_t *vdcp, struct buf *buf);
193 static int	vdc_depopulate_descriptor(vdc_t *vdc, uint_t idx);
194 static int	vdc_populate_mem_hdl(vdc_t *vdcp, vdc_local_desc_t *ldep);
195 static int	vdc_verify_seq_num(vdc_t *vdc, vio_dring_msg_t *dring_msg);
196 
197 /* dkio */
198 static int	vd_process_ioctl(dev_t dev, int cmd, caddr_t arg, int mode,
199 		    int *rvalp);
200 static int	vd_process_efi_ioctl(void *vdisk, int cmd, uintptr_t arg);
201 static void	vdc_create_fake_geometry(vdc_t *vdc);
202 static int	vdc_validate_geometry(vdc_t *vdc);
203 static void	vdc_validate(vdc_t *vdc);
204 static void	vdc_validate_task(void *arg);
205 static int	vdc_null_copy_func(vdc_t *vdc, void *from, void *to,
206 		    int mode, int dir);
207 static int	vdc_get_wce_convert(vdc_t *vdc, void *from, void *to,
208 		    int mode, int dir);
209 static int	vdc_set_wce_convert(vdc_t *vdc, void *from, void *to,
210 		    int mode, int dir);
211 static int	vdc_get_vtoc_convert(vdc_t *vdc, void *from, void *to,
212 		    int mode, int dir);
213 static int	vdc_set_vtoc_convert(vdc_t *vdc, void *from, void *to,
214 		    int mode, int dir);
215 static int	vdc_get_extvtoc_convert(vdc_t *vdc, void *from, void *to,
216 		    int mode, int dir);
217 static int	vdc_set_extvtoc_convert(vdc_t *vdc, void *from, void *to,
218 		    int mode, int dir);
219 static int	vdc_get_geom_convert(vdc_t *vdc, void *from, void *to,
220 		    int mode, int dir);
221 static int	vdc_set_geom_convert(vdc_t *vdc, void *from, void *to,
222 		    int mode, int dir);
223 static int	vdc_get_efi_convert(vdc_t *vdc, void *from, void *to,
224 		    int mode, int dir);
225 static int	vdc_set_efi_convert(vdc_t *vdc, void *from, void *to,
226 		    int mode, int dir);
227 
228 static void 	vdc_ownership_update(vdc_t *vdc, int ownership_flags);
229 static int	vdc_access_set(vdc_t *vdc, uint64_t flags);
230 static vdc_io_t	*vdc_eio_queue(vdc_t *vdc, int index);
231 static void	vdc_eio_unqueue(vdc_t *vdc, clock_t deadline,
232 		    boolean_t complete_io);
233 static int	vdc_eio_check(vdc_t *vdc, int flags);
234 static void	vdc_eio_thread(void *arg);
235 
236 /*
237  * Module variables
238  */
239 
240 /*
241  * Tunable variables to control how long vdc waits before timing out on
242  * various operations
243  */
244 static int	vdc_hshake_retries = 3;
245 
246 static int	vdc_timeout = 0; /* units: seconds */
247 static int 	vdc_ldcup_timeout = 1; /* units: seconds */
248 
249 static uint64_t vdc_hz_min_ldc_delay;
250 static uint64_t vdc_min_timeout_ldc = 1 * MILLISEC;
251 static uint64_t vdc_hz_max_ldc_delay;
252 static uint64_t vdc_max_timeout_ldc = 100 * MILLISEC;
253 
254 static uint64_t vdc_ldc_read_init_delay = 1 * MILLISEC;
255 static uint64_t vdc_ldc_read_max_delay = 100 * MILLISEC;
256 
257 /* values for dumping - need to run in a tighter loop */
258 static uint64_t	vdc_usec_timeout_dump = 100 * MILLISEC;	/* 0.1s units: ns */
259 static int	vdc_dump_retries = 100;
260 
261 static uint16_t	vdc_scsi_timeout = 60;	/* 60s units: seconds  */
262 
263 static uint64_t vdc_ownership_delay = 6 * MICROSEC; /* 6s units: usec */
264 
265 /* Count of the number of vdc instances attached */
266 static volatile uint32_t	vdc_instance_count = 0;
267 
268 /* Tunable to log all SCSI errors */
269 static boolean_t vdc_scsi_log_error = B_FALSE;
270 
271 /* Soft state pointer */
272 static void	*vdc_state;
273 
274 /*
275  * Controlling the verbosity of the error/debug messages
276  *
277  * vdc_msglevel - controls level of messages
278  * vdc_matchinst - 64-bit variable where each bit corresponds
279  *                 to the vdc instance the vdc_msglevel applies.
280  */
281 int		vdc_msglevel = 0x0;
282 uint64_t	vdc_matchinst = 0ull;
283 
284 /*
285  * Supported vDisk protocol version pairs.
286  *
287  * The first array entry is the latest and preferred version.
288  */
289 static const vio_ver_t	vdc_version[] = {{1, 1}};
290 
291 static struct cb_ops vdc_cb_ops = {
292 	vdc_open,	/* cb_open */
293 	vdc_close,	/* cb_close */
294 	vdc_strategy,	/* cb_strategy */
295 	vdc_print,	/* cb_print */
296 	vdc_dump,	/* cb_dump */
297 	vdc_read,	/* cb_read */
298 	vdc_write,	/* cb_write */
299 	vdc_ioctl,	/* cb_ioctl */
300 	nodev,		/* cb_devmap */
301 	nodev,		/* cb_mmap */
302 	nodev,		/* cb_segmap */
303 	nochpoll,	/* cb_chpoll */
304 	vdc_prop_op,	/* cb_prop_op */
305 	NULL,		/* cb_str */
306 	D_MP | D_64BIT,	/* cb_flag */
307 	CB_REV,		/* cb_rev */
308 	vdc_aread,	/* cb_aread */
309 	vdc_awrite	/* cb_awrite */
310 };
311 
312 static struct dev_ops vdc_ops = {
313 	DEVO_REV,	/* devo_rev */
314 	0,		/* devo_refcnt */
315 	vdc_getinfo,	/* devo_getinfo */
316 	nulldev,	/* devo_identify */
317 	nulldev,	/* devo_probe */
318 	vdc_attach,	/* devo_attach */
319 	vdc_detach,	/* devo_detach */
320 	nodev,		/* devo_reset */
321 	&vdc_cb_ops,	/* devo_cb_ops */
322 	NULL,		/* devo_bus_ops */
323 	nulldev,	/* devo_power */
324 	ddi_quiesce_not_needed,	/* devo_quiesce */
325 };
326 
327 static struct modldrv modldrv = {
328 	&mod_driverops,
329 	"virtual disk client",
330 	&vdc_ops,
331 };
332 
333 static struct modlinkage modlinkage = {
334 	MODREV_1,
335 	&modldrv,
336 	NULL
337 };
338 
339 /* -------------------------------------------------------------------------- */
340 
341 /*
342  * Device Driver housekeeping and setup
343  */
344 
345 int
346 _init(void)
347 {
348 	int	status;
349 
350 	if ((status = ddi_soft_state_init(&vdc_state, sizeof (vdc_t), 1)) != 0)
351 		return (status);
352 	if ((status = mod_install(&modlinkage)) != 0)
353 		ddi_soft_state_fini(&vdc_state);
354 	return (status);
355 }
356 
357 int
358 _info(struct modinfo *modinfop)
359 {
360 	return (mod_info(&modlinkage, modinfop));
361 }
362 
363 int
364 _fini(void)
365 {
366 	int	status;
367 
368 	if ((status = mod_remove(&modlinkage)) != 0)
369 		return (status);
370 	ddi_soft_state_fini(&vdc_state);
371 	return (0);
372 }
373 
374 static int
375 vdc_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd,  void *arg, void **resultp)
376 {
377 	_NOTE(ARGUNUSED(dip))
378 
379 	int	instance = VDCUNIT((dev_t)arg);
380 	vdc_t	*vdc = NULL;
381 
382 	switch (cmd) {
383 	case DDI_INFO_DEVT2DEVINFO:
384 		if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
385 			*resultp = NULL;
386 			return (DDI_FAILURE);
387 		}
388 		*resultp = vdc->dip;
389 		return (DDI_SUCCESS);
390 	case DDI_INFO_DEVT2INSTANCE:
391 		*resultp = (void *)(uintptr_t)instance;
392 		return (DDI_SUCCESS);
393 	default:
394 		*resultp = NULL;
395 		return (DDI_FAILURE);
396 	}
397 }
398 
399 static int
400 vdc_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
401 {
402 	kt_did_t eio_tid, ownership_tid;
403 	int	instance;
404 	int	rv;
405 	vdc_server_t *srvr;
406 	vdc_t	*vdc = NULL;
407 
408 	switch (cmd) {
409 	case DDI_DETACH:
410 		/* the real work happens below */
411 		break;
412 	case DDI_SUSPEND:
413 		/* nothing to do for this non-device */
414 		return (DDI_SUCCESS);
415 	default:
416 		return (DDI_FAILURE);
417 	}
418 
419 	ASSERT(cmd == DDI_DETACH);
420 	instance = ddi_get_instance(dip);
421 	DMSGX(1, "[%d] Entered\n", instance);
422 
423 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
424 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
425 		return (DDI_FAILURE);
426 	}
427 
428 	if (vdc_is_opened(vdc)) {
429 		DMSG(vdc, 0, "[%d] Cannot detach: device is open", instance);
430 		return (DDI_FAILURE);
431 	}
432 
433 	if (vdc->dkio_flush_pending) {
434 		DMSG(vdc, 0,
435 		    "[%d] Cannot detach: %d outstanding DKIO flushes\n",
436 		    instance, vdc->dkio_flush_pending);
437 		return (DDI_FAILURE);
438 	}
439 
440 	if (vdc->validate_pending) {
441 		DMSG(vdc, 0,
442 		    "[%d] Cannot detach: %d outstanding validate request\n",
443 		    instance, vdc->validate_pending);
444 		return (DDI_FAILURE);
445 	}
446 
447 	DMSG(vdc, 0, "[%d] proceeding...\n", instance);
448 
449 	/* If we took ownership, release ownership */
450 	mutex_enter(&vdc->ownership_lock);
451 	if (vdc->ownership & VDC_OWNERSHIP_GRANTED) {
452 		rv = vdc_access_set(vdc, VD_ACCESS_SET_CLEAR);
453 		if (rv == 0) {
454 			vdc_ownership_update(vdc, VDC_OWNERSHIP_NONE);
455 		}
456 	}
457 	mutex_exit(&vdc->ownership_lock);
458 
459 	/* mark instance as detaching */
460 	vdc->lifecycle	= VDC_LC_DETACHING;
461 
462 	/*
463 	 * Try and disable callbacks to prevent another handshake. We have to
464 	 * disable callbacks for all servers.
465 	 */
466 	for (srvr = vdc->server_list; srvr != NULL; srvr = srvr->next) {
467 		rv = ldc_set_cb_mode(srvr->ldc_handle, LDC_CB_DISABLE);
468 		DMSG(vdc, 0, "callback disabled (ldc=%lu, rv=%d)\n",
469 		    srvr->ldc_id, rv);
470 	}
471 
472 	if (vdc->initialized & VDC_THREAD) {
473 		mutex_enter(&vdc->read_lock);
474 		if ((vdc->read_state == VDC_READ_WAITING) ||
475 		    (vdc->read_state == VDC_READ_RESET)) {
476 			vdc->read_state = VDC_READ_RESET;
477 			cv_signal(&vdc->read_cv);
478 		}
479 
480 		mutex_exit(&vdc->read_lock);
481 
482 		/* wake up any thread waiting for connection to come online */
483 		mutex_enter(&vdc->lock);
484 		if (vdc->state == VDC_STATE_INIT_WAITING) {
485 			DMSG(vdc, 0,
486 			    "[%d] write reset - move to resetting state...\n",
487 			    instance);
488 			vdc->state = VDC_STATE_RESETTING;
489 			cv_signal(&vdc->initwait_cv);
490 		} else if (vdc->state == VDC_STATE_FAILED) {
491 			vdc->io_pending = B_TRUE;
492 			cv_signal(&vdc->io_pending_cv);
493 		}
494 		mutex_exit(&vdc->lock);
495 
496 		/* now wait until state transitions to VDC_STATE_DETACH */
497 		thread_join(vdc->msg_proc_thr->t_did);
498 		ASSERT(vdc->state == VDC_STATE_DETACH);
499 		DMSG(vdc, 0, "[%d] Reset thread exit and join ..\n",
500 		    vdc->instance);
501 	}
502 
503 	mutex_enter(&vdc->lock);
504 
505 	if (vdc->initialized & VDC_DRING)
506 		vdc_destroy_descriptor_ring(vdc);
507 
508 	vdc_fini_ports(vdc);
509 
510 	if (vdc->eio_thread) {
511 		eio_tid = vdc->eio_thread->t_did;
512 		vdc->failfast_interval = 0;
513 		ASSERT(vdc->num_servers == 0);
514 		cv_signal(&vdc->eio_cv);
515 	} else {
516 		eio_tid = 0;
517 	}
518 
519 	if (vdc->ownership & VDC_OWNERSHIP_WANTED) {
520 		ownership_tid = vdc->ownership_thread->t_did;
521 		vdc->ownership = VDC_OWNERSHIP_NONE;
522 		cv_signal(&vdc->ownership_cv);
523 	} else {
524 		ownership_tid = 0;
525 	}
526 
527 	mutex_exit(&vdc->lock);
528 
529 	if (eio_tid != 0)
530 		thread_join(eio_tid);
531 
532 	if (ownership_tid != 0)
533 		thread_join(ownership_tid);
534 
535 	if (vdc->initialized & VDC_MINOR)
536 		ddi_remove_minor_node(dip, NULL);
537 
538 	if (vdc->io_stats) {
539 		kstat_delete(vdc->io_stats);
540 		vdc->io_stats = NULL;
541 	}
542 
543 	if (vdc->err_stats) {
544 		kstat_delete(vdc->err_stats);
545 		vdc->err_stats = NULL;
546 	}
547 
548 	if (vdc->initialized & VDC_LOCKS) {
549 		mutex_destroy(&vdc->lock);
550 		mutex_destroy(&vdc->read_lock);
551 		mutex_destroy(&vdc->ownership_lock);
552 		cv_destroy(&vdc->initwait_cv);
553 		cv_destroy(&vdc->dring_free_cv);
554 		cv_destroy(&vdc->membind_cv);
555 		cv_destroy(&vdc->sync_blocked_cv);
556 		cv_destroy(&vdc->read_cv);
557 		cv_destroy(&vdc->running_cv);
558 		cv_destroy(&vdc->io_pending_cv);
559 		cv_destroy(&vdc->ownership_cv);
560 		cv_destroy(&vdc->eio_cv);
561 	}
562 
563 	if (vdc->minfo)
564 		kmem_free(vdc->minfo, sizeof (struct dk_minfo));
565 
566 	if (vdc->cinfo)
567 		kmem_free(vdc->cinfo, sizeof (struct dk_cinfo));
568 
569 	if (vdc->vtoc)
570 		kmem_free(vdc->vtoc, sizeof (struct extvtoc));
571 
572 	if (vdc->geom)
573 		kmem_free(vdc->geom, sizeof (struct dk_geom));
574 
575 	if (vdc->devid) {
576 		ddi_devid_unregister(dip);
577 		ddi_devid_free(vdc->devid);
578 	}
579 
580 	if (vdc->initialized & VDC_SOFT_STATE)
581 		ddi_soft_state_free(vdc_state, instance);
582 
583 	DMSG(vdc, 0, "[%d] End %p\n", instance, (void *)vdc);
584 
585 	return (DDI_SUCCESS);
586 }
587 
588 
589 static int
590 vdc_do_attach(dev_info_t *dip)
591 {
592 	int		instance;
593 	vdc_t		*vdc = NULL;
594 	int		status;
595 	md_t		*mdp;
596 	mde_cookie_t	vd_node;
597 
598 	ASSERT(dip != NULL);
599 
600 	instance = ddi_get_instance(dip);
601 	if (ddi_soft_state_zalloc(vdc_state, instance) != DDI_SUCCESS) {
602 		cmn_err(CE_NOTE, "[%d] Couldn't alloc state structure",
603 		    instance);
604 		return (DDI_FAILURE);
605 	}
606 
607 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
608 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
609 		return (DDI_FAILURE);
610 	}
611 
612 	/*
613 	 * We assign the value to initialized in this case to zero out the
614 	 * variable and then set bits in it to indicate what has been done
615 	 */
616 	vdc->initialized = VDC_SOFT_STATE;
617 
618 	vdc_hz_min_ldc_delay = drv_usectohz(vdc_min_timeout_ldc);
619 	vdc_hz_max_ldc_delay = drv_usectohz(vdc_max_timeout_ldc);
620 
621 	vdc->dip	= dip;
622 	vdc->instance	= instance;
623 	vdc->vdisk_type	= VD_DISK_TYPE_UNK;
624 	vdc->vdisk_label = VD_DISK_LABEL_UNK;
625 	vdc->state	= VDC_STATE_INIT;
626 	vdc->lifecycle	= VDC_LC_ATTACHING;
627 	vdc->session_id = 0;
628 	vdc->vdisk_bsize = DEV_BSIZE;
629 	vdc->vio_bmask = 0;
630 	vdc->vio_bshift = 0;
631 	vdc->max_xfer_sz = maxphys / vdc->vdisk_bsize;
632 
633 	/*
634 	 * We assume, for now, that the vDisk server will export 'read'
635 	 * operations to us at a minimum (this is needed because of checks
636 	 * in vdc for supported operations early in the handshake process).
637 	 * The vDisk server will return ENOTSUP if this is not the case.
638 	 * The value will be overwritten during the attribute exchange with
639 	 * the bitmask of operations exported by server.
640 	 */
641 	vdc->operations = VD_OP_MASK_READ;
642 
643 	vdc->vtoc = NULL;
644 	vdc->geom = NULL;
645 	vdc->cinfo = NULL;
646 	vdc->minfo = NULL;
647 
648 	mutex_init(&vdc->lock, NULL, MUTEX_DRIVER, NULL);
649 	cv_init(&vdc->initwait_cv, NULL, CV_DRIVER, NULL);
650 	cv_init(&vdc->dring_free_cv, NULL, CV_DRIVER, NULL);
651 	cv_init(&vdc->membind_cv, NULL, CV_DRIVER, NULL);
652 	cv_init(&vdc->running_cv, NULL, CV_DRIVER, NULL);
653 	cv_init(&vdc->io_pending_cv, NULL, CV_DRIVER, NULL);
654 
655 	vdc->io_pending = B_FALSE;
656 	vdc->threads_pending = 0;
657 	vdc->sync_op_blocked = B_FALSE;
658 	cv_init(&vdc->sync_blocked_cv, NULL, CV_DRIVER, NULL);
659 
660 	mutex_init(&vdc->ownership_lock, NULL, MUTEX_DRIVER, NULL);
661 	cv_init(&vdc->ownership_cv, NULL, CV_DRIVER, NULL);
662 	cv_init(&vdc->eio_cv, NULL, CV_DRIVER, NULL);
663 
664 	/* init blocking msg read functionality */
665 	mutex_init(&vdc->read_lock, NULL, MUTEX_DRIVER, NULL);
666 	cv_init(&vdc->read_cv, NULL, CV_DRIVER, NULL);
667 	vdc->read_state = VDC_READ_IDLE;
668 
669 	vdc->initialized |= VDC_LOCKS;
670 
671 	/* get device and port MD node for this disk instance */
672 	if (vdc_get_md_node(dip, &mdp, &vd_node) != 0) {
673 		cmn_err(CE_NOTE, "[%d] Could not get machine description node",
674 		    instance);
675 		return (DDI_FAILURE);
676 	}
677 
678 	if (vdc_init_ports(vdc, mdp, vd_node) != 0) {
679 		cmn_err(CE_NOTE, "[%d] Error initialising ports", instance);
680 		return (DDI_FAILURE);
681 	}
682 
683 	(void) md_fini_handle(mdp);
684 
685 	/* Create the kstats for saving the I/O statistics used by iostat(1M) */
686 	vdc_create_io_kstats(vdc);
687 	vdc_create_err_kstats(vdc);
688 
689 	/* Initialize remaining structures before starting the msg thread */
690 	vdc->vdisk_label = VD_DISK_LABEL_UNK;
691 	vdc->vtoc = kmem_zalloc(sizeof (struct extvtoc), KM_SLEEP);
692 	vdc->geom = kmem_zalloc(sizeof (struct dk_geom), KM_SLEEP);
693 	vdc->minfo = kmem_zalloc(sizeof (struct dk_minfo), KM_SLEEP);
694 
695 	/* initialize the thread responsible for managing state with server */
696 	vdc->msg_proc_thr = thread_create(NULL, 0, vdc_process_msg_thread,
697 	    vdc, 0, &p0, TS_RUN, minclsyspri);
698 	if (vdc->msg_proc_thr == NULL) {
699 		cmn_err(CE_NOTE, "[%d] Failed to create msg processing thread",
700 		    instance);
701 		return (DDI_FAILURE);
702 	}
703 
704 	/*
705 	 * If there are multiple servers then start the eio thread.
706 	 */
707 	if (vdc->num_servers > 1) {
708 		vdc->eio_thread = thread_create(NULL, 0, vdc_eio_thread, vdc, 0,
709 		    &p0, TS_RUN, v.v_maxsyspri - 2);
710 		if (vdc->eio_thread == NULL) {
711 			cmn_err(CE_NOTE, "[%d] Failed to create error "
712 			    "I/O thread", instance);
713 			return (DDI_FAILURE);
714 		}
715 	}
716 
717 	vdc->initialized |= VDC_THREAD;
718 
719 	atomic_inc_32(&vdc_instance_count);
720 
721 	/*
722 	 * Check the disk label. This will send requests and do the handshake.
723 	 * We don't really care about the disk label now. What we really need is
724 	 * the handshake do be done so that we know the type of the disk (slice
725 	 * or full disk) and the appropriate device nodes can be created.
726 	 */
727 
728 	mutex_enter(&vdc->lock);
729 	(void) vdc_validate_geometry(vdc);
730 	mutex_exit(&vdc->lock);
731 
732 	/*
733 	 * Now that we have the device info we can create the device nodes
734 	 */
735 	status = vdc_create_device_nodes(vdc);
736 	if (status) {
737 		DMSG(vdc, 0, "[%d] Failed to create device nodes",
738 		    instance);
739 		goto return_status;
740 	}
741 
742 	/*
743 	 * Fill in the fields of the error statistics kstat that were not
744 	 * available when creating the kstat
745 	 */
746 	vdc_set_err_kstats(vdc);
747 
748 	ddi_report_dev(dip);
749 	vdc->lifecycle	= VDC_LC_ONLINE;
750 	DMSG(vdc, 0, "[%d] Attach tasks successful\n", instance);
751 
752 return_status:
753 	DMSG(vdc, 0, "[%d] Attach completed\n", instance);
754 	return (status);
755 }
756 
757 static int
758 vdc_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
759 {
760 	int	status;
761 
762 	switch (cmd) {
763 	case DDI_ATTACH:
764 		if ((status = vdc_do_attach(dip)) != 0)
765 			(void) vdc_detach(dip, DDI_DETACH);
766 		return (status);
767 	case DDI_RESUME:
768 		/* nothing to do for this non-device */
769 		return (DDI_SUCCESS);
770 	default:
771 		return (DDI_FAILURE);
772 	}
773 }
774 
775 static int
776 vdc_do_ldc_init(vdc_t *vdc, vdc_server_t *srvr)
777 {
778 	int			status = 0;
779 	ldc_status_t		ldc_state;
780 	ldc_attr_t		ldc_attr;
781 
782 	ASSERT(vdc != NULL);
783 	ASSERT(srvr != NULL);
784 
785 	ldc_attr.devclass = LDC_DEV_BLK;
786 	ldc_attr.instance = vdc->instance;
787 	ldc_attr.mode = LDC_MODE_UNRELIABLE;	/* unreliable transport */
788 	ldc_attr.mtu = VD_LDC_MTU;
789 
790 	if ((srvr->state & VDC_LDC_INIT) == 0) {
791 		status = ldc_init(srvr->ldc_id, &ldc_attr,
792 		    &srvr->ldc_handle);
793 		if (status != 0) {
794 			DMSG(vdc, 0, "[%d] ldc_init(chan %ld) returned %d",
795 			    vdc->instance, srvr->ldc_id, status);
796 			return (status);
797 		}
798 		srvr->state |= VDC_LDC_INIT;
799 	}
800 	status = ldc_status(srvr->ldc_handle, &ldc_state);
801 	if (status != 0) {
802 		DMSG(vdc, 0, "[%d] Cannot discover LDC status [err=%d]",
803 		    vdc->instance, status);
804 		goto init_exit;
805 	}
806 	srvr->ldc_state = ldc_state;
807 
808 	if ((srvr->state & VDC_LDC_CB) == 0) {
809 		status = ldc_reg_callback(srvr->ldc_handle, vdc_handle_cb,
810 		    (caddr_t)srvr);
811 		if (status != 0) {
812 			DMSG(vdc, 0, "[%d] LDC callback reg. failed (%d)",
813 			    vdc->instance, status);
814 			goto init_exit;
815 		}
816 		srvr->state |= VDC_LDC_CB;
817 	}
818 
819 	/*
820 	 * At this stage we have initialised LDC, we will now try and open
821 	 * the connection.
822 	 */
823 	if (srvr->ldc_state == LDC_INIT) {
824 		status = ldc_open(srvr->ldc_handle);
825 		if (status != 0) {
826 			DMSG(vdc, 0, "[%d] ldc_open(chan %ld) returned %d",
827 			    vdc->instance, srvr->ldc_id, status);
828 			goto init_exit;
829 		}
830 		srvr->state |= VDC_LDC_OPEN;
831 	}
832 
833 init_exit:
834 	if (status) {
835 		vdc_terminate_ldc(vdc, srvr);
836 	}
837 
838 	return (status);
839 }
840 
841 static int
842 vdc_start_ldc_connection(vdc_t *vdc)
843 {
844 	int		status = 0;
845 
846 	ASSERT(vdc != NULL);
847 
848 	ASSERT(MUTEX_HELD(&vdc->lock));
849 
850 	status = vdc_do_ldc_up(vdc);
851 
852 	DMSG(vdc, 0, "[%d] Finished bringing up LDC\n", vdc->instance);
853 
854 	return (status);
855 }
856 
857 static int
858 vdc_stop_ldc_connection(vdc_t *vdcp)
859 {
860 	int	status;
861 
862 	ASSERT(vdcp != NULL);
863 
864 	ASSERT(MUTEX_HELD(&vdcp->lock));
865 
866 	DMSG(vdcp, 0, ": Resetting connection to vDisk server : state %d\n",
867 	    vdcp->state);
868 
869 	status = ldc_down(vdcp->curr_server->ldc_handle);
870 	DMSG(vdcp, 0, "ldc_down() = %d\n", status);
871 
872 	vdcp->initialized &= ~VDC_HANDSHAKE;
873 	DMSG(vdcp, 0, "initialized=%x\n", vdcp->initialized);
874 
875 	return (status);
876 }
877 
878 static void
879 vdc_create_io_kstats(vdc_t *vdc)
880 {
881 	if (vdc->io_stats != NULL) {
882 		DMSG(vdc, 0, "[%d] I/O kstat already exists\n", vdc->instance);
883 		return;
884 	}
885 
886 	vdc->io_stats = kstat_create(VDC_DRIVER_NAME, vdc->instance, NULL,
887 	    "disk", KSTAT_TYPE_IO, 1, KSTAT_FLAG_PERSISTENT);
888 	if (vdc->io_stats != NULL) {
889 		vdc->io_stats->ks_lock = &vdc->lock;
890 		kstat_install(vdc->io_stats);
891 	} else {
892 		cmn_err(CE_NOTE, "[%d] Failed to create kstat: I/O statistics"
893 		    " will not be gathered", vdc->instance);
894 	}
895 }
896 
897 static void
898 vdc_create_err_kstats(vdc_t *vdc)
899 {
900 	vd_err_stats_t	*stp;
901 	char	kstatmodule_err[KSTAT_STRLEN];
902 	char	kstatname[KSTAT_STRLEN];
903 	int	ndata = (sizeof (vd_err_stats_t) / sizeof (kstat_named_t));
904 	int	instance = vdc->instance;
905 
906 	if (vdc->err_stats != NULL) {
907 		DMSG(vdc, 0, "[%d] ERR kstat already exists\n", vdc->instance);
908 		return;
909 	}
910 
911 	(void) snprintf(kstatmodule_err, sizeof (kstatmodule_err),
912 	    "%serr", VDC_DRIVER_NAME);
913 	(void) snprintf(kstatname, sizeof (kstatname),
914 	    "%s%d,err", VDC_DRIVER_NAME, instance);
915 
916 	vdc->err_stats = kstat_create(kstatmodule_err, instance, kstatname,
917 	    "device_error", KSTAT_TYPE_NAMED, ndata, KSTAT_FLAG_PERSISTENT);
918 
919 	if (vdc->err_stats == NULL) {
920 		cmn_err(CE_NOTE, "[%d] Failed to create kstat: Error statistics"
921 		    " will not be gathered", instance);
922 		return;
923 	}
924 
925 	stp = (vd_err_stats_t *)vdc->err_stats->ks_data;
926 	kstat_named_init(&stp->vd_softerrs,	"Soft Errors",
927 	    KSTAT_DATA_UINT32);
928 	kstat_named_init(&stp->vd_transerrs,	"Transport Errors",
929 	    KSTAT_DATA_UINT32);
930 	kstat_named_init(&stp->vd_protoerrs,	"Protocol Errors",
931 	    KSTAT_DATA_UINT32);
932 	kstat_named_init(&stp->vd_vid,		"Vendor",
933 	    KSTAT_DATA_CHAR);
934 	kstat_named_init(&stp->vd_pid,		"Product",
935 	    KSTAT_DATA_CHAR);
936 	kstat_named_init(&stp->vd_capacity,	"Size",
937 	    KSTAT_DATA_ULONGLONG);
938 
939 	vdc->err_stats->ks_update  = nulldev;
940 
941 	kstat_install(vdc->err_stats);
942 }
943 
944 static void
945 vdc_set_err_kstats(vdc_t *vdc)
946 {
947 	vd_err_stats_t  *stp;
948 
949 	if (vdc->err_stats == NULL)
950 		return;
951 
952 	mutex_enter(&vdc->lock);
953 
954 	stp = (vd_err_stats_t *)vdc->err_stats->ks_data;
955 	ASSERT(stp != NULL);
956 
957 	stp->vd_capacity.value.ui64 = vdc->vdisk_size * vdc->vdisk_bsize;
958 	(void) strcpy(stp->vd_vid.value.c, "SUN");
959 	(void) strcpy(stp->vd_pid.value.c, "VDSK");
960 
961 	mutex_exit(&vdc->lock);
962 }
963 
964 static int
965 vdc_create_device_nodes_efi(vdc_t *vdc)
966 {
967 	ddi_remove_minor_node(vdc->dip, "h");
968 	ddi_remove_minor_node(vdc->dip, "h,raw");
969 
970 	if (ddi_create_minor_node(vdc->dip, "wd", S_IFBLK,
971 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
972 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
973 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'wd'",
974 		    vdc->instance);
975 		return (EIO);
976 	}
977 
978 	/* if any device node is created we set this flag */
979 	vdc->initialized |= VDC_MINOR;
980 
981 	if (ddi_create_minor_node(vdc->dip, "wd,raw", S_IFCHR,
982 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
983 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
984 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'wd,raw'",
985 		    vdc->instance);
986 		return (EIO);
987 	}
988 
989 	return (0);
990 }
991 
992 static int
993 vdc_create_device_nodes_vtoc(vdc_t *vdc)
994 {
995 	ddi_remove_minor_node(vdc->dip, "wd");
996 	ddi_remove_minor_node(vdc->dip, "wd,raw");
997 
998 	if (ddi_create_minor_node(vdc->dip, "h", S_IFBLK,
999 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
1000 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
1001 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'h'",
1002 		    vdc->instance);
1003 		return (EIO);
1004 	}
1005 
1006 	/* if any device node is created we set this flag */
1007 	vdc->initialized |= VDC_MINOR;
1008 
1009 	if (ddi_create_minor_node(vdc->dip, "h,raw", S_IFCHR,
1010 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
1011 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
1012 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'h,raw'",
1013 		    vdc->instance);
1014 		return (EIO);
1015 	}
1016 
1017 	return (0);
1018 }
1019 
1020 /*
1021  * Function:
1022  *	vdc_create_device_nodes
1023  *
1024  * Description:
1025  *	This function creates the block and character device nodes under
1026  *	/devices. It is called as part of the attach(9E) of the instance
1027  *	during the handshake with vds after vds has sent the attributes
1028  *	to vdc.
1029  *
1030  *	If the device is of type VD_DISK_TYPE_SLICE then the minor node
1031  *	of 2 is used in keeping with the Solaris convention that slice 2
1032  *	refers to a whole disk. Slices start at 'a'
1033  *
1034  * Parameters:
1035  *	vdc 		- soft state pointer
1036  *
1037  * Return Values
1038  *	0		- Success
1039  *	EIO		- Failed to create node
1040  */
1041 static int
1042 vdc_create_device_nodes(vdc_t *vdc)
1043 {
1044 	char		name[sizeof ("s,raw")];
1045 	dev_info_t	*dip = NULL;
1046 	int		instance, status;
1047 	int		num_slices = 1;
1048 	int		i;
1049 
1050 	ASSERT(vdc != NULL);
1051 
1052 	instance = vdc->instance;
1053 	dip = vdc->dip;
1054 
1055 	switch (vdc->vdisk_type) {
1056 	case VD_DISK_TYPE_DISK:
1057 	case VD_DISK_TYPE_UNK:
1058 		num_slices = V_NUMPAR;
1059 		break;
1060 	case VD_DISK_TYPE_SLICE:
1061 		num_slices = 1;
1062 		break;
1063 	default:
1064 		ASSERT(0);
1065 	}
1066 
1067 	/*
1068 	 * Minor nodes are different for EFI disks: EFI disks do not have
1069 	 * a minor node 'g' for the minor number corresponding to slice
1070 	 * VD_EFI_WD_SLICE (slice 7) instead they have a minor node 'wd'
1071 	 * representing the whole disk.
1072 	 */
1073 	for (i = 0; i < num_slices; i++) {
1074 
1075 		if (i == VD_EFI_WD_SLICE) {
1076 			if (vdc->vdisk_label == VD_DISK_LABEL_EFI)
1077 				status = vdc_create_device_nodes_efi(vdc);
1078 			else
1079 				status = vdc_create_device_nodes_vtoc(vdc);
1080 			if (status != 0)
1081 				return (status);
1082 			continue;
1083 		}
1084 
1085 		(void) snprintf(name, sizeof (name), "%c", 'a' + i);
1086 		if (ddi_create_minor_node(dip, name, S_IFBLK,
1087 		    VD_MAKE_DEV(instance, i), DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
1088 			cmn_err(CE_NOTE, "[%d] Couldn't add block node '%s'",
1089 			    instance, name);
1090 			return (EIO);
1091 		}
1092 
1093 		/* if any device node is created we set this flag */
1094 		vdc->initialized |= VDC_MINOR;
1095 
1096 		(void) snprintf(name, sizeof (name), "%c%s", 'a' + i, ",raw");
1097 
1098 		if (ddi_create_minor_node(dip, name, S_IFCHR,
1099 		    VD_MAKE_DEV(instance, i), DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
1100 			cmn_err(CE_NOTE, "[%d] Couldn't add raw node '%s'",
1101 			    instance, name);
1102 			return (EIO);
1103 		}
1104 	}
1105 
1106 	return (0);
1107 }
1108 
1109 /*
1110  * Driver prop_op(9e) entry point function. Return the number of blocks for
1111  * the partition in question or forward the request to the property facilities.
1112  */
1113 static int
1114 vdc_prop_op(dev_t dev, dev_info_t *dip, ddi_prop_op_t prop_op, int mod_flags,
1115     char *name, caddr_t valuep, int *lengthp)
1116 {
1117 	int instance = ddi_get_instance(dip);
1118 	vdc_t *vdc;
1119 	uint64_t nblocks;
1120 	uint_t blksize;
1121 
1122 	vdc = ddi_get_soft_state(vdc_state, instance);
1123 
1124 	if (dev == DDI_DEV_T_ANY || vdc == NULL) {
1125 		return (ddi_prop_op(dev, dip, prop_op, mod_flags,
1126 		    name, valuep, lengthp));
1127 	}
1128 
1129 	mutex_enter(&vdc->lock);
1130 	(void) vdc_validate_geometry(vdc);
1131 	if (vdc->vdisk_label == VD_DISK_LABEL_UNK) {
1132 		mutex_exit(&vdc->lock);
1133 		return (ddi_prop_op(dev, dip, prop_op, mod_flags,
1134 		    name, valuep, lengthp));
1135 	}
1136 	nblocks = vdc->slice[VDCPART(dev)].nblocks;
1137 	blksize = vdc->vdisk_bsize;
1138 	mutex_exit(&vdc->lock);
1139 
1140 	return (ddi_prop_op_nblocks_blksize(dev, dip, prop_op, mod_flags,
1141 	    name, valuep, lengthp, nblocks, blksize));
1142 }
1143 
1144 /*
1145  * Function:
1146  *	vdc_is_opened
1147  *
1148  * Description:
1149  *	This function checks if any slice of a given virtual disk is
1150  *	currently opened.
1151  *
1152  * Parameters:
1153  *	vdc 		- soft state pointer
1154  *
1155  * Return Values
1156  *	B_TRUE		- at least one slice is opened.
1157  *	B_FALSE		- no slice is opened.
1158  */
1159 static boolean_t
1160 vdc_is_opened(vdc_t *vdc)
1161 {
1162 	int i;
1163 
1164 	/* check if there's any layered open */
1165 	for (i = 0; i < V_NUMPAR; i++) {
1166 		if (vdc->open_lyr[i] > 0)
1167 			return (B_TRUE);
1168 	}
1169 
1170 	/* check if there is any other kind of open */
1171 	for (i = 0; i < OTYPCNT; i++) {
1172 		if (vdc->open[i] != 0)
1173 			return (B_TRUE);
1174 	}
1175 
1176 	return (B_FALSE);
1177 }
1178 
1179 static int
1180 vdc_mark_opened(vdc_t *vdc, int slice, int flag, int otyp)
1181 {
1182 	uint8_t slicemask;
1183 	int i;
1184 
1185 	ASSERT(otyp < OTYPCNT);
1186 	ASSERT(slice < V_NUMPAR);
1187 	ASSERT(MUTEX_HELD(&vdc->lock));
1188 
1189 	slicemask = 1 << slice;
1190 
1191 	/*
1192 	 * If we have a single-slice disk which was unavailable during the
1193 	 * attach then a device was created for each 8 slices. Now that
1194 	 * the type is known, we prevent opening any slice other than 0
1195 	 * even if a device still exists.
1196 	 */
1197 	if (vdc->vdisk_type == VD_DISK_TYPE_SLICE && slice != 0)
1198 		return (EIO);
1199 
1200 	/* check if slice is already exclusively opened */
1201 	if (vdc->open_excl & slicemask)
1202 		return (EBUSY);
1203 
1204 	/* if open exclusive, check if slice is already opened */
1205 	if (flag & FEXCL) {
1206 		if (vdc->open_lyr[slice] > 0)
1207 			return (EBUSY);
1208 		for (i = 0; i < OTYPCNT; i++) {
1209 			if (vdc->open[i] & slicemask)
1210 				return (EBUSY);
1211 		}
1212 		vdc->open_excl |= slicemask;
1213 	}
1214 
1215 	/* mark slice as opened */
1216 	if (otyp == OTYP_LYR) {
1217 		vdc->open_lyr[slice]++;
1218 	} else {
1219 		vdc->open[otyp] |= slicemask;
1220 	}
1221 
1222 	return (0);
1223 }
1224 
1225 static void
1226 vdc_mark_closed(vdc_t *vdc, int slice, int flag, int otyp)
1227 {
1228 	uint8_t slicemask;
1229 
1230 	ASSERT(otyp < OTYPCNT);
1231 	ASSERT(slice < V_NUMPAR);
1232 	ASSERT(MUTEX_HELD(&vdc->lock));
1233 
1234 	slicemask = 1 << slice;
1235 
1236 	if (otyp == OTYP_LYR) {
1237 		ASSERT(vdc->open_lyr[slice] > 0);
1238 		vdc->open_lyr[slice]--;
1239 	} else {
1240 		vdc->open[otyp] &= ~slicemask;
1241 	}
1242 
1243 	if (flag & FEXCL)
1244 		vdc->open_excl &= ~slicemask;
1245 }
1246 
1247 static int
1248 vdc_open(dev_t *dev, int flag, int otyp, cred_t *cred)
1249 {
1250 	_NOTE(ARGUNUSED(cred))
1251 
1252 	int	instance, nodelay;
1253 	int	slice, status = 0;
1254 	vdc_t	*vdc;
1255 
1256 	ASSERT(dev != NULL);
1257 	instance = VDCUNIT(*dev);
1258 
1259 	if (otyp >= OTYPCNT)
1260 		return (EINVAL);
1261 
1262 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1263 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1264 		return (ENXIO);
1265 	}
1266 
1267 	DMSG(vdc, 0, "minor = %d flag = %x, otyp = %x\n",
1268 	    getminor(*dev), flag, otyp);
1269 
1270 	slice = VDCPART(*dev);
1271 
1272 	nodelay = flag & (FNDELAY | FNONBLOCK);
1273 
1274 	if ((flag & FWRITE) && (!nodelay) &&
1275 	    !(VD_OP_SUPPORTED(vdc->operations, VD_OP_BWRITE))) {
1276 		return (EROFS);
1277 	}
1278 
1279 	mutex_enter(&vdc->lock);
1280 
1281 	status = vdc_mark_opened(vdc, slice, flag, otyp);
1282 
1283 	if (status != 0) {
1284 		mutex_exit(&vdc->lock);
1285 		return (status);
1286 	}
1287 
1288 	/*
1289 	 * If the disk type is unknown then we have to wait for the
1290 	 * handshake to complete because we don't know if the slice
1291 	 * device we are opening effectively exists.
1292 	 */
1293 	if (vdc->vdisk_type != VD_DISK_TYPE_UNK && nodelay) {
1294 
1295 		/* don't resubmit a validate request if there's already one */
1296 		if (vdc->validate_pending > 0) {
1297 			mutex_exit(&vdc->lock);
1298 			return (0);
1299 		}
1300 
1301 		/* call vdc_validate() asynchronously to avoid blocking */
1302 		if (taskq_dispatch(system_taskq, vdc_validate_task,
1303 		    (void *)vdc, TQ_NOSLEEP) == NULL) {
1304 			vdc_mark_closed(vdc, slice, flag, otyp);
1305 			mutex_exit(&vdc->lock);
1306 			return (ENXIO);
1307 		}
1308 
1309 		vdc->validate_pending++;
1310 		mutex_exit(&vdc->lock);
1311 		return (0);
1312 	}
1313 
1314 	mutex_exit(&vdc->lock);
1315 
1316 	vdc_validate(vdc);
1317 
1318 	mutex_enter(&vdc->lock);
1319 
1320 	if (vdc->vdisk_type == VD_DISK_TYPE_UNK ||
1321 	    (vdc->vdisk_type == VD_DISK_TYPE_SLICE && slice != 0) ||
1322 	    (!nodelay && (vdc->vdisk_label == VD_DISK_LABEL_UNK ||
1323 	    vdc->slice[slice].nblocks == 0))) {
1324 		vdc_mark_closed(vdc, slice, flag, otyp);
1325 		status = EIO;
1326 	}
1327 
1328 	mutex_exit(&vdc->lock);
1329 
1330 	return (status);
1331 }
1332 
1333 static int
1334 vdc_close(dev_t dev, int flag, int otyp, cred_t *cred)
1335 {
1336 	_NOTE(ARGUNUSED(cred))
1337 
1338 	int	instance;
1339 	int	slice;
1340 	int	rv, rval;
1341 	vdc_t	*vdc;
1342 
1343 	instance = VDCUNIT(dev);
1344 
1345 	if (otyp >= OTYPCNT)
1346 		return (EINVAL);
1347 
1348 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1349 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1350 		return (ENXIO);
1351 	}
1352 
1353 	DMSG(vdc, 0, "[%d] flag = %x, otyp = %x\n", instance, flag, otyp);
1354 
1355 	slice = VDCPART(dev);
1356 
1357 	/*
1358 	 * Attempt to flush the W$ on a close operation. If this is
1359 	 * not a supported IOCTL command or the backing device is read-only
1360 	 * do not fail the close operation.
1361 	 */
1362 	rv = vd_process_ioctl(dev, DKIOCFLUSHWRITECACHE, NULL, FKIOCTL, &rval);
1363 
1364 	if (rv != 0 && rv != ENOTSUP && rv != ENOTTY && rv != EROFS) {
1365 		DMSG(vdc, 0, "[%d] flush failed with error %d on close\n",
1366 		    instance, rv);
1367 		return (EIO);
1368 	}
1369 
1370 	mutex_enter(&vdc->lock);
1371 	vdc_mark_closed(vdc, slice, flag, otyp);
1372 	mutex_exit(&vdc->lock);
1373 
1374 	return (0);
1375 }
1376 
1377 static int
1378 vdc_ioctl(dev_t dev, int cmd, intptr_t arg, int mode, cred_t *credp, int *rvalp)
1379 {
1380 	_NOTE(ARGUNUSED(credp))
1381 
1382 	return (vd_process_ioctl(dev, cmd, (caddr_t)arg, mode, rvalp));
1383 }
1384 
1385 static int
1386 vdc_print(dev_t dev, char *str)
1387 {
1388 	cmn_err(CE_NOTE, "vdc%d:  %s", VDCUNIT(dev), str);
1389 	return (0);
1390 }
1391 
1392 static int
1393 vdc_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblk)
1394 {
1395 	int	rv, flags;
1396 	size_t	nbytes = nblk * DEV_BSIZE;
1397 	int	instance = VDCUNIT(dev);
1398 	vdc_t	*vdc = NULL;
1399 	diskaddr_t vio_blkno;
1400 
1401 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1402 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1403 		return (ENXIO);
1404 	}
1405 
1406 	DMSG(vdc, 2, "[%d] dump %ld bytes at block 0x%lx : addr=0x%p\n",
1407 	    instance, nbytes, blkno, (void *)addr);
1408 
1409 	/* convert logical block to vio block */
1410 	if ((blkno & vdc->vio_bmask) != 0) {
1411 		DMSG(vdc, 0, "Misaligned block number (%lu)\n", blkno);
1412 		return (EINVAL);
1413 	}
1414 	vio_blkno = blkno >> vdc->vio_bshift;
1415 
1416 	/*
1417 	 * If we are panicking, we need the state to be "running" so that we
1418 	 * can submit I/Os, but we don't want to check for any backend error.
1419 	 */
1420 	flags = (ddi_in_panic())? VDC_OP_STATE_RUNNING : VDC_OP_NORMAL;
1421 
1422 	rv = vdc_do_op(vdc, VD_OP_BWRITE, addr, nbytes, VDCPART(dev),
1423 	    vio_blkno, NULL, VIO_write_dir, flags);
1424 
1425 	if (rv) {
1426 		DMSG(vdc, 0, "Failed to do a disk dump (err=%d)\n", rv);
1427 		return (rv);
1428 	}
1429 
1430 	DMSG(vdc, 0, "[%d] End\n", instance);
1431 
1432 	return (0);
1433 }
1434 
1435 /* -------------------------------------------------------------------------- */
1436 
1437 /*
1438  * Disk access routines
1439  *
1440  */
1441 
1442 /*
1443  * vdc_strategy()
1444  *
1445  * Return Value:
1446  *	0:	As per strategy(9E), the strategy() function must return 0
1447  *		[ bioerror(9f) sets b_flags to the proper error code ]
1448  */
1449 static int
1450 vdc_strategy(struct buf *buf)
1451 {
1452 	diskaddr_t vio_blkno;
1453 	vdc_t	*vdc = NULL;
1454 	int	instance = VDCUNIT(buf->b_edev);
1455 	int	op = (buf->b_flags & B_READ) ? VD_OP_BREAD : VD_OP_BWRITE;
1456 	int	slice;
1457 
1458 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1459 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1460 		bioerror(buf, ENXIO);
1461 		biodone(buf);
1462 		return (0);
1463 	}
1464 
1465 	DMSG(vdc, 2, "[%d] %s %ld bytes at block %llx : b_addr=0x%p\n",
1466 	    instance, (buf->b_flags & B_READ) ? "Read" : "Write",
1467 	    buf->b_bcount, buf->b_lblkno, (void *)buf->b_un.b_addr);
1468 
1469 	bp_mapin(buf);
1470 
1471 	if ((long)buf->b_private == VD_SLICE_NONE) {
1472 		/* I/O using an absolute disk offset */
1473 		slice = VD_SLICE_NONE;
1474 	} else {
1475 		slice = VDCPART(buf->b_edev);
1476 	}
1477 
1478 	/*
1479 	 * In the buf structure, b_lblkno represents a logical block number
1480 	 * using a block size of 512 bytes. For the VIO request, this block
1481 	 * number has to be converted to be represented with the block size
1482 	 * used by the VIO protocol.
1483 	 */
1484 	if ((buf->b_lblkno & vdc->vio_bmask) != 0) {
1485 		bioerror(buf, EINVAL);
1486 		biodone(buf);
1487 		return (0);
1488 	}
1489 	vio_blkno = buf->b_lblkno >> vdc->vio_bshift;
1490 
1491 	/* submit the I/O, any error will be reported in the buf structure */
1492 	(void) vdc_do_op(vdc, op, (caddr_t)buf->b_un.b_addr,
1493 	    buf->b_bcount, slice, vio_blkno,
1494 	    buf, (op == VD_OP_BREAD) ? VIO_read_dir : VIO_write_dir,
1495 	    VDC_OP_NORMAL);
1496 
1497 	return (0);
1498 }
1499 
1500 /*
1501  * Function:
1502  *	vdc_min
1503  *
1504  * Description:
1505  *	Routine to limit the size of a data transfer. Used in
1506  *	conjunction with physio(9F).
1507  *
1508  * Arguments:
1509  *	bp - pointer to the indicated buf(9S) struct.
1510  *
1511  */
1512 static void
1513 vdc_min(struct buf *bufp)
1514 {
1515 	vdc_t	*vdc = NULL;
1516 	int	instance = VDCUNIT(bufp->b_edev);
1517 
1518 	vdc = ddi_get_soft_state(vdc_state, instance);
1519 	VERIFY(vdc != NULL);
1520 
1521 	if (bufp->b_bcount > (vdc->max_xfer_sz * vdc->vdisk_bsize)) {
1522 		bufp->b_bcount = vdc->max_xfer_sz * vdc->vdisk_bsize;
1523 	}
1524 }
1525 
1526 static int
1527 vdc_read(dev_t dev, struct uio *uio, cred_t *cred)
1528 {
1529 	_NOTE(ARGUNUSED(cred))
1530 
1531 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1532 	return (physio(vdc_strategy, NULL, dev, B_READ, vdc_min, uio));
1533 }
1534 
1535 static int
1536 vdc_write(dev_t dev, struct uio *uio, cred_t *cred)
1537 {
1538 	_NOTE(ARGUNUSED(cred))
1539 
1540 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1541 	return (physio(vdc_strategy, NULL, dev, B_WRITE, vdc_min, uio));
1542 }
1543 
1544 static int
1545 vdc_aread(dev_t dev, struct aio_req *aio, cred_t *cred)
1546 {
1547 	_NOTE(ARGUNUSED(cred))
1548 
1549 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1550 	return (aphysio(vdc_strategy, anocancel, dev, B_READ, vdc_min, aio));
1551 }
1552 
1553 static int
1554 vdc_awrite(dev_t dev, struct aio_req *aio, cred_t *cred)
1555 {
1556 	_NOTE(ARGUNUSED(cred))
1557 
1558 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1559 	return (aphysio(vdc_strategy, anocancel, dev, B_WRITE, vdc_min, aio));
1560 }
1561 
1562 
1563 /* -------------------------------------------------------------------------- */
1564 
1565 /*
1566  * Handshake support
1567  */
1568 
1569 
1570 /*
1571  * Function:
1572  *	vdc_init_ver_negotiation()
1573  *
1574  * Description:
1575  *
1576  * Arguments:
1577  *	vdc	- soft state pointer for this instance of the device driver.
1578  *
1579  * Return Code:
1580  *	0	- Success
1581  */
1582 static int
1583 vdc_init_ver_negotiation(vdc_t *vdc, vio_ver_t ver)
1584 {
1585 	vio_ver_msg_t	pkt;
1586 	size_t		msglen = sizeof (pkt);
1587 	int		status = -1;
1588 
1589 	ASSERT(vdc != NULL);
1590 	ASSERT(mutex_owned(&vdc->lock));
1591 
1592 	DMSG(vdc, 0, "[%d] Entered.\n", vdc->instance);
1593 
1594 	/*
1595 	 * set the Session ID to a unique value
1596 	 * (the lower 32 bits of the clock tick)
1597 	 */
1598 	vdc->session_id = ((uint32_t)gettick() & 0xffffffff);
1599 	DMSG(vdc, 0, "[%d] Set SID to 0x%lx\n", vdc->instance, vdc->session_id);
1600 
1601 	pkt.tag.vio_msgtype = VIO_TYPE_CTRL;
1602 	pkt.tag.vio_subtype = VIO_SUBTYPE_INFO;
1603 	pkt.tag.vio_subtype_env = VIO_VER_INFO;
1604 	pkt.tag.vio_sid = vdc->session_id;
1605 	pkt.dev_class = VDEV_DISK;
1606 	pkt.ver_major = ver.major;
1607 	pkt.ver_minor = ver.minor;
1608 
1609 	status = vdc_send(vdc, (caddr_t)&pkt, &msglen);
1610 	DMSG(vdc, 0, "[%d] Ver info sent (status = %d)\n",
1611 	    vdc->instance, status);
1612 	if ((status != 0) || (msglen != sizeof (vio_ver_msg_t))) {
1613 		DMSG(vdc, 0, "[%d] Failed to send Ver negotiation info: "
1614 		    "id(%lx) rv(%d) size(%ld)", vdc->instance,
1615 		    vdc->curr_server->ldc_handle, status, msglen);
1616 		if (msglen != sizeof (vio_ver_msg_t))
1617 			status = ENOMSG;
1618 	}
1619 
1620 	return (status);
1621 }
1622 
1623 /*
1624  * Function:
1625  *	vdc_ver_negotiation()
1626  *
1627  * Description:
1628  *
1629  * Arguments:
1630  *	vdcp	- soft state pointer for this instance of the device driver.
1631  *
1632  * Return Code:
1633  *	0	- Success
1634  */
1635 static int
1636 vdc_ver_negotiation(vdc_t *vdcp)
1637 {
1638 	vio_msg_t vio_msg;
1639 	int status;
1640 
1641 	if (status = vdc_init_ver_negotiation(vdcp, vdc_version[0]))
1642 		return (status);
1643 
1644 	/* release lock and wait for response */
1645 	mutex_exit(&vdcp->lock);
1646 	status = vdc_wait_for_response(vdcp, &vio_msg);
1647 	mutex_enter(&vdcp->lock);
1648 	if (status) {
1649 		DMSG(vdcp, 0,
1650 		    "[%d] Failed waiting for Ver negotiation response, rv(%d)",
1651 		    vdcp->instance, status);
1652 		return (status);
1653 	}
1654 
1655 	/* check type and sub_type ... */
1656 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1657 	    vio_msg.tag.vio_subtype == VIO_SUBTYPE_INFO) {
1658 		DMSG(vdcp, 0, "[%d] Invalid ver negotiation response\n",
1659 		    vdcp->instance);
1660 		return (EPROTO);
1661 	}
1662 
1663 	return (vdc_handle_ver_msg(vdcp, (vio_ver_msg_t *)&vio_msg));
1664 }
1665 
1666 /*
1667  * Function:
1668  *	vdc_init_attr_negotiation()
1669  *
1670  * Description:
1671  *
1672  * Arguments:
1673  *	vdc	- soft state pointer for this instance of the device driver.
1674  *
1675  * Return Code:
1676  *	0	- Success
1677  */
1678 static int
1679 vdc_init_attr_negotiation(vdc_t *vdc)
1680 {
1681 	vd_attr_msg_t	pkt;
1682 	size_t		msglen = sizeof (pkt);
1683 	int		status;
1684 
1685 	ASSERT(vdc != NULL);
1686 	ASSERT(mutex_owned(&vdc->lock));
1687 
1688 	DMSG(vdc, 0, "[%d] entered\n", vdc->instance);
1689 
1690 	/* fill in tag */
1691 	pkt.tag.vio_msgtype = VIO_TYPE_CTRL;
1692 	pkt.tag.vio_subtype = VIO_SUBTYPE_INFO;
1693 	pkt.tag.vio_subtype_env = VIO_ATTR_INFO;
1694 	pkt.tag.vio_sid = vdc->session_id;
1695 	/* fill in payload */
1696 	pkt.max_xfer_sz = vdc->max_xfer_sz;
1697 	pkt.vdisk_block_size = vdc->vdisk_bsize;
1698 	pkt.xfer_mode = VIO_DRING_MODE_V1_0;
1699 	pkt.operations = 0;	/* server will set bits of valid operations */
1700 	pkt.vdisk_type = 0;	/* server will set to valid device type */
1701 	pkt.vdisk_media = 0;	/* server will set to valid media type */
1702 	pkt.vdisk_size = 0;	/* server will set to valid size */
1703 
1704 	status = vdc_send(vdc, (caddr_t)&pkt, &msglen);
1705 	DMSG(vdc, 0, "Attr info sent (status = %d)\n", status);
1706 
1707 	if ((status != 0) || (msglen != sizeof (vd_attr_msg_t))) {
1708 		DMSG(vdc, 0, "[%d] Failed to send Attr negotiation info: "
1709 		    "id(%lx) rv(%d) size(%ld)", vdc->instance,
1710 		    vdc->curr_server->ldc_handle, status, msglen);
1711 		if (msglen != sizeof (vd_attr_msg_t))
1712 			status = ENOMSG;
1713 	}
1714 
1715 	return (status);
1716 }
1717 
1718 /*
1719  * Function:
1720  *	vdc_attr_negotiation()
1721  *
1722  * Description:
1723  *
1724  * Arguments:
1725  *	vdc	- soft state pointer for this instance of the device driver.
1726  *
1727  * Return Code:
1728  *	0	- Success
1729  */
1730 static int
1731 vdc_attr_negotiation(vdc_t *vdcp)
1732 {
1733 	int status;
1734 	vio_msg_t vio_msg;
1735 
1736 	if (status = vdc_init_attr_negotiation(vdcp))
1737 		return (status);
1738 
1739 	/* release lock and wait for response */
1740 	mutex_exit(&vdcp->lock);
1741 	status = vdc_wait_for_response(vdcp, &vio_msg);
1742 	mutex_enter(&vdcp->lock);
1743 	if (status) {
1744 		DMSG(vdcp, 0,
1745 		    "[%d] Failed waiting for Attr negotiation response, rv(%d)",
1746 		    vdcp->instance, status);
1747 		return (status);
1748 	}
1749 
1750 	/* check type and sub_type ... */
1751 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1752 	    vio_msg.tag.vio_subtype == VIO_SUBTYPE_INFO) {
1753 		DMSG(vdcp, 0, "[%d] Invalid attr negotiation response\n",
1754 		    vdcp->instance);
1755 		return (EPROTO);
1756 	}
1757 
1758 	return (vdc_handle_attr_msg(vdcp, (vd_attr_msg_t *)&vio_msg));
1759 }
1760 
1761 
1762 /*
1763  * Function:
1764  *	vdc_init_dring_negotiate()
1765  *
1766  * Description:
1767  *
1768  * Arguments:
1769  *	vdc	- soft state pointer for this instance of the device driver.
1770  *
1771  * Return Code:
1772  *	0	- Success
1773  */
1774 static int
1775 vdc_init_dring_negotiate(vdc_t *vdc)
1776 {
1777 	vio_dring_reg_msg_t	pkt;
1778 	size_t			msglen = sizeof (pkt);
1779 	int			status = -1;
1780 	int			retry;
1781 	int			nretries = 10;
1782 
1783 	ASSERT(vdc != NULL);
1784 	ASSERT(mutex_owned(&vdc->lock));
1785 
1786 	for (retry = 0; retry < nretries; retry++) {
1787 		status = vdc_init_descriptor_ring(vdc);
1788 		if (status != EAGAIN)
1789 			break;
1790 		drv_usecwait(vdc_min_timeout_ldc);
1791 	}
1792 
1793 	if (status != 0) {
1794 		DMSG(vdc, 0, "[%d] Failed to init DRing (status = %d)\n",
1795 		    vdc->instance, status);
1796 		return (status);
1797 	}
1798 
1799 	DMSG(vdc, 0, "[%d] Init of descriptor ring completed (status = %d)\n",
1800 	    vdc->instance, status);
1801 
1802 	/* fill in tag */
1803 	pkt.tag.vio_msgtype = VIO_TYPE_CTRL;
1804 	pkt.tag.vio_subtype = VIO_SUBTYPE_INFO;
1805 	pkt.tag.vio_subtype_env = VIO_DRING_REG;
1806 	pkt.tag.vio_sid = vdc->session_id;
1807 	/* fill in payload */
1808 	pkt.dring_ident = 0;
1809 	pkt.num_descriptors = vdc->dring_len;
1810 	pkt.descriptor_size = vdc->dring_entry_size;
1811 	pkt.options = (VIO_TX_DRING | VIO_RX_DRING);
1812 	pkt.ncookies = vdc->dring_cookie_count;
1813 	pkt.cookie[0] = vdc->dring_cookie[0];	/* for now just one cookie */
1814 
1815 	status = vdc_send(vdc, (caddr_t)&pkt, &msglen);
1816 	if (status != 0) {
1817 		DMSG(vdc, 0, "[%d] Failed to register DRing (err = %d)",
1818 		    vdc->instance, status);
1819 	}
1820 
1821 	return (status);
1822 }
1823 
1824 
1825 /*
1826  * Function:
1827  *	vdc_dring_negotiation()
1828  *
1829  * Description:
1830  *
1831  * Arguments:
1832  *	vdc	- soft state pointer for this instance of the device driver.
1833  *
1834  * Return Code:
1835  *	0	- Success
1836  */
1837 static int
1838 vdc_dring_negotiation(vdc_t *vdcp)
1839 {
1840 	int status;
1841 	vio_msg_t vio_msg;
1842 
1843 	if (status = vdc_init_dring_negotiate(vdcp))
1844 		return (status);
1845 
1846 	/* release lock and wait for response */
1847 	mutex_exit(&vdcp->lock);
1848 	status = vdc_wait_for_response(vdcp, &vio_msg);
1849 	mutex_enter(&vdcp->lock);
1850 	if (status) {
1851 		DMSG(vdcp, 0,
1852 		    "[%d] Failed waiting for Dring negotiation response,"
1853 		    " rv(%d)", vdcp->instance, status);
1854 		return (status);
1855 	}
1856 
1857 	/* check type and sub_type ... */
1858 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1859 	    vio_msg.tag.vio_subtype == VIO_SUBTYPE_INFO) {
1860 		DMSG(vdcp, 0, "[%d] Invalid Dring negotiation response\n",
1861 		    vdcp->instance);
1862 		return (EPROTO);
1863 	}
1864 
1865 	return (vdc_handle_dring_reg_msg(vdcp,
1866 	    (vio_dring_reg_msg_t *)&vio_msg));
1867 }
1868 
1869 
1870 /*
1871  * Function:
1872  *	vdc_send_rdx()
1873  *
1874  * Description:
1875  *
1876  * Arguments:
1877  *	vdc	- soft state pointer for this instance of the device driver.
1878  *
1879  * Return Code:
1880  *	0	- Success
1881  */
1882 static int
1883 vdc_send_rdx(vdc_t *vdcp)
1884 {
1885 	vio_msg_t	msg;
1886 	size_t		msglen = sizeof (vio_msg_t);
1887 	int		status;
1888 
1889 	/*
1890 	 * Send an RDX message to vds to indicate we are ready
1891 	 * to send data
1892 	 */
1893 	msg.tag.vio_msgtype = VIO_TYPE_CTRL;
1894 	msg.tag.vio_subtype = VIO_SUBTYPE_INFO;
1895 	msg.tag.vio_subtype_env = VIO_RDX;
1896 	msg.tag.vio_sid = vdcp->session_id;
1897 	status = vdc_send(vdcp, (caddr_t)&msg, &msglen);
1898 	if (status != 0) {
1899 		DMSG(vdcp, 0, "[%d] Failed to send RDX message (%d)",
1900 		    vdcp->instance, status);
1901 	}
1902 
1903 	return (status);
1904 }
1905 
1906 /*
1907  * Function:
1908  *	vdc_handle_rdx()
1909  *
1910  * Description:
1911  *
1912  * Arguments:
1913  *	vdc	- soft state pointer for this instance of the device driver.
1914  *	msgp	- received msg
1915  *
1916  * Return Code:
1917  *	0	- Success
1918  */
1919 static int
1920 vdc_handle_rdx(vdc_t *vdcp, vio_rdx_msg_t *msgp)
1921 {
1922 	_NOTE(ARGUNUSED(vdcp))
1923 	_NOTE(ARGUNUSED(msgp))
1924 
1925 	ASSERT(msgp->tag.vio_msgtype == VIO_TYPE_CTRL);
1926 	ASSERT(msgp->tag.vio_subtype == VIO_SUBTYPE_ACK);
1927 	ASSERT(msgp->tag.vio_subtype_env == VIO_RDX);
1928 
1929 	DMSG(vdcp, 1, "[%d] Got an RDX msg", vdcp->instance);
1930 
1931 	return (0);
1932 }
1933 
1934 /*
1935  * Function:
1936  *	vdc_rdx_exchange()
1937  *
1938  * Description:
1939  *
1940  * Arguments:
1941  *	vdc	- soft state pointer for this instance of the device driver.
1942  *
1943  * Return Code:
1944  *	0	- Success
1945  */
1946 static int
1947 vdc_rdx_exchange(vdc_t *vdcp)
1948 {
1949 	int status;
1950 	vio_msg_t vio_msg;
1951 
1952 	if (status = vdc_send_rdx(vdcp))
1953 		return (status);
1954 
1955 	/* release lock and wait for response */
1956 	mutex_exit(&vdcp->lock);
1957 	status = vdc_wait_for_response(vdcp, &vio_msg);
1958 	mutex_enter(&vdcp->lock);
1959 	if (status) {
1960 		DMSG(vdcp, 0, "[%d] Failed waiting for RDX response, rv(%d)",
1961 		    vdcp->instance, status);
1962 		return (status);
1963 	}
1964 
1965 	/* check type and sub_type ... */
1966 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1967 	    vio_msg.tag.vio_subtype != VIO_SUBTYPE_ACK) {
1968 		DMSG(vdcp, 0, "[%d] Invalid RDX response\n", vdcp->instance);
1969 		return (EPROTO);
1970 	}
1971 
1972 	return (vdc_handle_rdx(vdcp, (vio_rdx_msg_t *)&vio_msg));
1973 }
1974 
1975 
1976 /* -------------------------------------------------------------------------- */
1977 
1978 /*
1979  * LDC helper routines
1980  */
1981 
1982 static int
1983 vdc_recv(vdc_t *vdc, vio_msg_t *msgp, size_t *nbytesp)
1984 {
1985 	int		status;
1986 	uint64_t	delay_time;
1987 	size_t		len;
1988 
1989 	/*
1990 	 * Until we get a blocking ldc read we have to retry until the entire
1991 	 * LDC message has arrived before ldc_read() will return that message.
1992 	 * If ldc_read() succeed but returns a zero length message then that
1993 	 * means that the LDC queue is empty and we have to wait for a
1994 	 * notification from the LDC callback which will set the read_state to
1995 	 * VDC_READ_PENDING. Note we also bail out if the channel is reset or
1996 	 * goes away.
1997 	 */
1998 	delay_time = vdc_ldc_read_init_delay;
1999 
2000 	for (;;) {
2001 
2002 		len = *nbytesp;
2003 		/*
2004 		 * vdc->curr_server is protected by vdc->lock but to avoid
2005 		 * contentions we don't take the lock here. We can do this
2006 		 * safely because vdc_recv() is only called from thread
2007 		 * process_msg_thread() which is also the only thread that
2008 		 * can change vdc->curr_server.
2009 		 */
2010 		status = ldc_read(vdc->curr_server->ldc_handle,
2011 		    (caddr_t)msgp, &len);
2012 
2013 		if (status == EAGAIN) {
2014 			delay_time *= 2;
2015 			if (delay_time >= vdc_ldc_read_max_delay)
2016 				delay_time = vdc_ldc_read_max_delay;
2017 			delay(delay_time);
2018 			continue;
2019 		}
2020 
2021 		if (status != 0) {
2022 			DMSG(vdc, 0, "ldc_read returned %d\n", status);
2023 			break;
2024 		}
2025 
2026 		if (len != 0) {
2027 			*nbytesp = len;
2028 			break;
2029 		}
2030 
2031 		mutex_enter(&vdc->read_lock);
2032 
2033 		while (vdc->read_state != VDC_READ_PENDING) {
2034 
2035 			/* detect if the connection has been reset */
2036 			if (vdc->read_state == VDC_READ_RESET) {
2037 				mutex_exit(&vdc->read_lock);
2038 				return (ECONNRESET);
2039 			}
2040 
2041 			vdc->read_state = VDC_READ_WAITING;
2042 			cv_wait(&vdc->read_cv, &vdc->read_lock);
2043 		}
2044 
2045 		vdc->read_state = VDC_READ_IDLE;
2046 		mutex_exit(&vdc->read_lock);
2047 
2048 		delay_time = vdc_ldc_read_init_delay;
2049 	}
2050 
2051 	return (status);
2052 }
2053 
2054 
2055 
2056 #ifdef DEBUG
2057 void
2058 vdc_decode_tag(vdc_t *vdcp, vio_msg_t *msg)
2059 {
2060 	char *ms, *ss, *ses;
2061 	switch (msg->tag.vio_msgtype) {
2062 #define	Q(_s)	case _s : ms = #_s; break;
2063 	Q(VIO_TYPE_CTRL)
2064 	Q(VIO_TYPE_DATA)
2065 	Q(VIO_TYPE_ERR)
2066 #undef Q
2067 	default: ms = "unknown"; break;
2068 	}
2069 
2070 	switch (msg->tag.vio_subtype) {
2071 #define	Q(_s)	case _s : ss = #_s; break;
2072 	Q(VIO_SUBTYPE_INFO)
2073 	Q(VIO_SUBTYPE_ACK)
2074 	Q(VIO_SUBTYPE_NACK)
2075 #undef Q
2076 	default: ss = "unknown"; break;
2077 	}
2078 
2079 	switch (msg->tag.vio_subtype_env) {
2080 #define	Q(_s)	case _s : ses = #_s; break;
2081 	Q(VIO_VER_INFO)
2082 	Q(VIO_ATTR_INFO)
2083 	Q(VIO_DRING_REG)
2084 	Q(VIO_DRING_UNREG)
2085 	Q(VIO_RDX)
2086 	Q(VIO_PKT_DATA)
2087 	Q(VIO_DESC_DATA)
2088 	Q(VIO_DRING_DATA)
2089 #undef Q
2090 	default: ses = "unknown"; break;
2091 	}
2092 
2093 	DMSG(vdcp, 3, "(%x/%x/%x) message : (%s/%s/%s)\n",
2094 	    msg->tag.vio_msgtype, msg->tag.vio_subtype,
2095 	    msg->tag.vio_subtype_env, ms, ss, ses);
2096 }
2097 #endif
2098 
2099 /*
2100  * Function:
2101  *	vdc_send()
2102  *
2103  * Description:
2104  *	The function encapsulates the call to write a message using LDC.
2105  *	If LDC indicates that the call failed due to the queue being full,
2106  *	we retry the ldc_write(), otherwise we return the error returned by LDC.
2107  *
2108  * Arguments:
2109  *	ldc_handle	- LDC handle for the channel this instance of vdc uses
2110  *	pkt		- address of LDC message to be sent
2111  *	msglen		- the size of the message being sent. When the function
2112  *			  returns, this contains the number of bytes written.
2113  *
2114  * Return Code:
2115  *	0		- Success.
2116  *	EINVAL		- pkt or msglen were NULL
2117  *	ECONNRESET	- The connection was not up.
2118  *	EWOULDBLOCK	- LDC queue is full
2119  *	xxx		- other error codes returned by ldc_write
2120  */
2121 static int
2122 vdc_send(vdc_t *vdc, caddr_t pkt, size_t *msglen)
2123 {
2124 	size_t	size = 0;
2125 	int	status = 0;
2126 	clock_t delay_ticks;
2127 
2128 	ASSERT(vdc != NULL);
2129 	ASSERT(mutex_owned(&vdc->lock));
2130 	ASSERT(msglen != NULL);
2131 	ASSERT(*msglen != 0);
2132 
2133 #ifdef DEBUG
2134 	vdc_decode_tag(vdc, (vio_msg_t *)(uintptr_t)pkt);
2135 #endif
2136 	/*
2137 	 * Wait indefinitely to send if channel
2138 	 * is busy, but bail out if we succeed or
2139 	 * if the channel closes or is reset.
2140 	 */
2141 	delay_ticks = vdc_hz_min_ldc_delay;
2142 	do {
2143 		size = *msglen;
2144 		status = ldc_write(vdc->curr_server->ldc_handle, pkt, &size);
2145 		if (status == EWOULDBLOCK) {
2146 			delay(delay_ticks);
2147 			/* geometric backoff */
2148 			delay_ticks *= 2;
2149 			if (delay_ticks > vdc_hz_max_ldc_delay)
2150 				delay_ticks = vdc_hz_max_ldc_delay;
2151 		}
2152 	} while (status == EWOULDBLOCK);
2153 
2154 	/* if LDC had serious issues --- reset vdc state */
2155 	if (status == EIO || status == ECONNRESET) {
2156 		/* LDC had serious issues --- reset vdc state */
2157 		mutex_enter(&vdc->read_lock);
2158 		if ((vdc->read_state == VDC_READ_WAITING) ||
2159 		    (vdc->read_state == VDC_READ_RESET))
2160 			cv_signal(&vdc->read_cv);
2161 		vdc->read_state = VDC_READ_RESET;
2162 		mutex_exit(&vdc->read_lock);
2163 
2164 		/* wake up any waiters in the reset thread */
2165 		if (vdc->state == VDC_STATE_INIT_WAITING) {
2166 			DMSG(vdc, 0, "[%d] write reset - "
2167 			    "vdc is resetting ..\n", vdc->instance);
2168 			vdc->state = VDC_STATE_RESETTING;
2169 			cv_signal(&vdc->initwait_cv);
2170 		}
2171 
2172 		return (ECONNRESET);
2173 	}
2174 
2175 	/* return the last size written */
2176 	*msglen = size;
2177 
2178 	return (status);
2179 }
2180 
2181 /*
2182  * Function:
2183  *	vdc_get_md_node
2184  *
2185  * Description:
2186  *	Get the MD, the device node for the given disk instance. The
2187  *	caller is responsible for cleaning up the reference to the
2188  *	returned MD (mdpp) by calling md_fini_handle().
2189  *
2190  * Arguments:
2191  *	dip	- dev info pointer for this instance of the device driver.
2192  *	mdpp	- the returned MD.
2193  *	vd_nodep - the returned device node.
2194  *
2195  * Return Code:
2196  *	0	- Success.
2197  *	ENOENT	- Expected node or property did not exist.
2198  *	ENXIO	- Unexpected error communicating with MD framework
2199  */
2200 static int
2201 vdc_get_md_node(dev_info_t *dip, md_t **mdpp, mde_cookie_t *vd_nodep)
2202 {
2203 	int		status = ENOENT;
2204 	char		*node_name = NULL;
2205 	md_t		*mdp = NULL;
2206 	int		num_nodes;
2207 	int		num_vdevs;
2208 	mde_cookie_t	rootnode;
2209 	mde_cookie_t	*listp = NULL;
2210 	boolean_t	found_inst = B_FALSE;
2211 	int		listsz;
2212 	int		idx;
2213 	uint64_t	md_inst;
2214 	int		obp_inst;
2215 	int		instance = ddi_get_instance(dip);
2216 
2217 	/*
2218 	 * Get the OBP instance number for comparison with the MD instance
2219 	 *
2220 	 * The "cfg-handle" property of a vdc node in an MD contains the MD's
2221 	 * notion of "instance", or unique identifier, for that node; OBP
2222 	 * stores the value of the "cfg-handle" MD property as the value of
2223 	 * the "reg" property on the node in the device tree it builds from
2224 	 * the MD and passes to Solaris.  Thus, we look up the devinfo node's
2225 	 * "reg" property value to uniquely identify this device instance.
2226 	 * If the "reg" property cannot be found, the device tree state is
2227 	 * presumably so broken that there is no point in continuing.
2228 	 */
2229 	if (!ddi_prop_exists(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS, OBP_REG)) {
2230 		cmn_err(CE_WARN, "'%s' property does not exist", OBP_REG);
2231 		return (ENOENT);
2232 	}
2233 	obp_inst = ddi_prop_get_int(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS,
2234 	    OBP_REG, -1);
2235 	DMSGX(1, "[%d] OBP inst=%d\n", instance, obp_inst);
2236 
2237 	/*
2238 	 * We now walk the MD nodes to find the node for this vdisk.
2239 	 */
2240 	if ((mdp = md_get_handle()) == NULL) {
2241 		cmn_err(CE_WARN, "unable to init machine description");
2242 		return (ENXIO);
2243 	}
2244 
2245 	num_nodes = md_node_count(mdp);
2246 	ASSERT(num_nodes > 0);
2247 
2248 	listsz = num_nodes * sizeof (mde_cookie_t);
2249 
2250 	/* allocate memory for nodes */
2251 	listp = kmem_zalloc(listsz, KM_SLEEP);
2252 
2253 	rootnode = md_root_node(mdp);
2254 	ASSERT(rootnode != MDE_INVAL_ELEM_COOKIE);
2255 
2256 	/*
2257 	 * Search for all the virtual devices, we will then check to see which
2258 	 * ones are disk nodes.
2259 	 */
2260 	num_vdevs = md_scan_dag(mdp, rootnode,
2261 	    md_find_name(mdp, VDC_MD_VDEV_NAME),
2262 	    md_find_name(mdp, "fwd"), listp);
2263 
2264 	if (num_vdevs <= 0) {
2265 		cmn_err(CE_NOTE, "No '%s' node found", VDC_MD_VDEV_NAME);
2266 		status = ENOENT;
2267 		goto done;
2268 	}
2269 
2270 	DMSGX(1, "[%d] num_vdevs=%d\n", instance, num_vdevs);
2271 	for (idx = 0; idx < num_vdevs; idx++) {
2272 		status = md_get_prop_str(mdp, listp[idx], "name", &node_name);
2273 		if ((status != 0) || (node_name == NULL)) {
2274 			cmn_err(CE_NOTE, "Unable to get name of node type '%s'"
2275 			    ": err %d", VDC_MD_VDEV_NAME, status);
2276 			continue;
2277 		}
2278 
2279 		DMSGX(1, "[%d] Found node '%s'\n", instance, node_name);
2280 		if (strcmp(VDC_MD_DISK_NAME, node_name) == 0) {
2281 			status = md_get_prop_val(mdp, listp[idx],
2282 			    VDC_MD_CFG_HDL, &md_inst);
2283 			DMSGX(1, "[%d] vdc inst in MD=%lx\n",
2284 			    instance, md_inst);
2285 			if ((status == 0) && (md_inst == obp_inst)) {
2286 				found_inst = B_TRUE;
2287 				break;
2288 			}
2289 		}
2290 	}
2291 
2292 	if (!found_inst) {
2293 		DMSGX(0, "Unable to find correct '%s' node", VDC_MD_DISK_NAME);
2294 		status = ENOENT;
2295 		goto done;
2296 	}
2297 	DMSGX(0, "[%d] MD inst=%lx\n", instance, md_inst);
2298 
2299 	*vd_nodep = listp[idx];
2300 	*mdpp = mdp;
2301 done:
2302 	kmem_free(listp, listsz);
2303 	return (status);
2304 }
2305 
2306 /*
2307  * Function:
2308  *	vdc_init_ports
2309  *
2310  * Description:
2311  *	Initialize all the ports for this vdisk instance.
2312  *
2313  * Arguments:
2314  *	vdc	- soft state pointer for this instance of the device driver.
2315  *	mdp	- md pointer
2316  *	vd_nodep - device md node.
2317  *
2318  * Return Code:
2319  *	0	- Success.
2320  *	ENOENT	- Expected node or property did not exist.
2321  */
2322 static int
2323 vdc_init_ports(vdc_t *vdc, md_t *mdp, mde_cookie_t vd_nodep)
2324 {
2325 	int		status = 0;
2326 	int		idx;
2327 	int		num_nodes;
2328 	int		num_vports;
2329 	int		num_chans;
2330 	int		listsz;
2331 	mde_cookie_t	vd_port;
2332 	mde_cookie_t	*chanp = NULL;
2333 	mde_cookie_t	*portp = NULL;
2334 	vdc_server_t	*srvr;
2335 	vdc_server_t	*prev_srvr = NULL;
2336 
2337 	/*
2338 	 * We now walk the MD nodes to find the port nodes for this vdisk.
2339 	 */
2340 	num_nodes = md_node_count(mdp);
2341 	ASSERT(num_nodes > 0);
2342 
2343 	listsz = num_nodes * sizeof (mde_cookie_t);
2344 
2345 	/* allocate memory for nodes */
2346 	portp = kmem_zalloc(listsz, KM_SLEEP);
2347 	chanp = kmem_zalloc(listsz, KM_SLEEP);
2348 
2349 	num_vports = md_scan_dag(mdp, vd_nodep,
2350 	    md_find_name(mdp, VDC_MD_PORT_NAME),
2351 	    md_find_name(mdp, "fwd"), portp);
2352 	if (num_vports == 0) {
2353 		DMSGX(0, "Found no '%s' node for '%s' port\n",
2354 		    VDC_MD_PORT_NAME, VDC_MD_VDEV_NAME);
2355 		status = ENOENT;
2356 		goto done;
2357 	}
2358 
2359 	DMSGX(1, "Found %d '%s' node(s) for '%s' port\n",
2360 	    num_vports, VDC_MD_PORT_NAME, VDC_MD_VDEV_NAME);
2361 
2362 	vdc->num_servers = 0;
2363 	for (idx = 0; idx < num_vports; idx++) {
2364 
2365 		/* initialize this port */
2366 		vd_port = portp[idx];
2367 		srvr = kmem_zalloc(sizeof (vdc_server_t), KM_SLEEP);
2368 		srvr->vdcp = vdc;
2369 		srvr->svc_state = VDC_SERVICE_OFFLINE;
2370 		srvr->log_state = VDC_SERVICE_NONE;
2371 
2372 		/* get port id */
2373 		if (md_get_prop_val(mdp, vd_port, VDC_MD_ID, &srvr->id) != 0) {
2374 			cmn_err(CE_NOTE, "vDisk port '%s' property not found",
2375 			    VDC_MD_ID);
2376 			kmem_free(srvr, sizeof (vdc_server_t));
2377 			continue;
2378 		}
2379 
2380 		/* set the connection timeout */
2381 		if (md_get_prop_val(mdp, vd_port, VDC_MD_TIMEOUT,
2382 		    &srvr->ctimeout) != 0) {
2383 			srvr->ctimeout = 0;
2384 		}
2385 
2386 		/* get the ldc id */
2387 		num_chans = md_scan_dag(mdp, vd_port,
2388 		    md_find_name(mdp, VDC_MD_CHAN_NAME),
2389 		    md_find_name(mdp, "fwd"), chanp);
2390 
2391 		/* expecting at least one channel */
2392 		if (num_chans <= 0) {
2393 			cmn_err(CE_NOTE, "No '%s' node for '%s' port",
2394 			    VDC_MD_CHAN_NAME, VDC_MD_VDEV_NAME);
2395 			kmem_free(srvr, sizeof (vdc_server_t));
2396 			continue;
2397 		} else if (num_chans != 1) {
2398 			DMSGX(0, "Expected 1 '%s' node for '%s' port, "
2399 			    "found %d\n", VDC_MD_CHAN_NAME, VDC_MD_VDEV_NAME,
2400 			    num_chans);
2401 		}
2402 
2403 		/*
2404 		 * We use the first channel found (index 0), irrespective of how
2405 		 * many are there in total.
2406 		 */
2407 		if (md_get_prop_val(mdp, chanp[0], VDC_MD_ID,
2408 		    &srvr->ldc_id) != 0) {
2409 			cmn_err(CE_NOTE, "Channel '%s' property not found",
2410 			    VDC_MD_ID);
2411 			kmem_free(srvr, sizeof (vdc_server_t));
2412 			continue;
2413 		}
2414 
2415 		/*
2416 		 * now initialise LDC channel which will be used to
2417 		 * communicate with this server
2418 		 */
2419 		if (vdc_do_ldc_init(vdc, srvr) != 0) {
2420 			kmem_free(srvr, sizeof (vdc_server_t));
2421 			continue;
2422 		}
2423 
2424 		/* add server to list */
2425 		if (prev_srvr)
2426 			prev_srvr->next = srvr;
2427 		else
2428 			vdc->server_list = srvr;
2429 
2430 		prev_srvr = srvr;
2431 
2432 		/* inc numbers of servers */
2433 		vdc->num_servers++;
2434 	}
2435 
2436 	/*
2437 	 * Adjust the max number of handshake retries to match
2438 	 * the number of vdisk servers.
2439 	 */
2440 	if (vdc_hshake_retries < vdc->num_servers)
2441 		vdc_hshake_retries = vdc->num_servers;
2442 
2443 	/* pick first server as current server */
2444 	if (vdc->server_list != NULL) {
2445 		vdc->curr_server = vdc->server_list;
2446 		status = 0;
2447 	} else {
2448 		status = ENOENT;
2449 	}
2450 
2451 done:
2452 	kmem_free(chanp, listsz);
2453 	kmem_free(portp, listsz);
2454 	return (status);
2455 }
2456 
2457 
2458 /*
2459  * Function:
2460  *	vdc_do_ldc_up
2461  *
2462  * Description:
2463  *	Bring the channel for the current server up.
2464  *
2465  * Arguments:
2466  *	vdc	- soft state pointer for this instance of the device driver.
2467  *
2468  * Return Code:
2469  *	0		- Success.
2470  *	EINVAL		- Driver is detaching / LDC error
2471  *	ECONNREFUSED	- Other end is not listening
2472  */
2473 static int
2474 vdc_do_ldc_up(vdc_t *vdc)
2475 {
2476 	int		status;
2477 	ldc_status_t	ldc_state;
2478 
2479 	ASSERT(MUTEX_HELD(&vdc->lock));
2480 
2481 	DMSG(vdc, 0, "[%d] Bringing up channel %lx\n",
2482 	    vdc->instance, vdc->curr_server->ldc_id);
2483 
2484 	if (vdc->lifecycle == VDC_LC_DETACHING)
2485 		return (EINVAL);
2486 
2487 	if ((status = ldc_up(vdc->curr_server->ldc_handle)) != 0) {
2488 		switch (status) {
2489 		case ECONNREFUSED:	/* listener not ready at other end */
2490 			DMSG(vdc, 0, "[%d] ldc_up(%lx,...) return %d\n",
2491 			    vdc->instance, vdc->curr_server->ldc_id, status);
2492 			status = 0;
2493 			break;
2494 		default:
2495 			DMSG(vdc, 0, "[%d] Failed to bring up LDC: "
2496 			    "channel=%ld, err=%d", vdc->instance,
2497 			    vdc->curr_server->ldc_id, status);
2498 			break;
2499 		}
2500 	}
2501 
2502 	if (ldc_status(vdc->curr_server->ldc_handle, &ldc_state) == 0) {
2503 		vdc->curr_server->ldc_state = ldc_state;
2504 		if (ldc_state == LDC_UP) {
2505 			DMSG(vdc, 0, "[%d] LDC channel already up\n",
2506 			    vdc->instance);
2507 			vdc->seq_num = 1;
2508 			vdc->seq_num_reply = 0;
2509 		}
2510 	}
2511 
2512 	return (status);
2513 }
2514 
2515 /*
2516  * Function:
2517  *	vdc_terminate_ldc()
2518  *
2519  * Description:
2520  *
2521  * Arguments:
2522  *	vdc	- soft state pointer for this instance of the device driver.
2523  *	srvr	- vdc per-server info structure
2524  *
2525  * Return Code:
2526  *	None
2527  */
2528 static void
2529 vdc_terminate_ldc(vdc_t *vdc, vdc_server_t *srvr)
2530 {
2531 	int	instance = ddi_get_instance(vdc->dip);
2532 
2533 	if (srvr->state & VDC_LDC_OPEN) {
2534 		DMSG(vdc, 0, "[%d] ldc_close()\n", instance);
2535 		(void) ldc_close(srvr->ldc_handle);
2536 	}
2537 	if (srvr->state & VDC_LDC_CB) {
2538 		DMSG(vdc, 0, "[%d] ldc_unreg_callback()\n", instance);
2539 		(void) ldc_unreg_callback(srvr->ldc_handle);
2540 	}
2541 	if (srvr->state & VDC_LDC_INIT) {
2542 		DMSG(vdc, 0, "[%d] ldc_fini()\n", instance);
2543 		(void) ldc_fini(srvr->ldc_handle);
2544 		srvr->ldc_handle = NULL;
2545 	}
2546 
2547 	srvr->state &= ~(VDC_LDC_INIT | VDC_LDC_CB | VDC_LDC_OPEN);
2548 }
2549 
2550 /*
2551  * Function:
2552  *	vdc_fini_ports()
2553  *
2554  * Description:
2555  *	Finalize all ports by closing the channel associated with each
2556  *	port and also freeing the server structure.
2557  *
2558  * Arguments:
2559  *	vdc	- soft state pointer for this instance of the device driver.
2560  *
2561  * Return Code:
2562  *	None
2563  */
2564 static void
2565 vdc_fini_ports(vdc_t *vdc)
2566 {
2567 	int		instance = ddi_get_instance(vdc->dip);
2568 	vdc_server_t	*srvr, *prev_srvr;
2569 
2570 	ASSERT(vdc != NULL);
2571 	ASSERT(mutex_owned(&vdc->lock));
2572 
2573 	DMSG(vdc, 0, "[%d] initialized=%x\n", instance, vdc->initialized);
2574 
2575 	srvr = vdc->server_list;
2576 
2577 	while (srvr) {
2578 
2579 		vdc_terminate_ldc(vdc, srvr);
2580 
2581 		/* next server */
2582 		prev_srvr = srvr;
2583 		srvr = srvr->next;
2584 
2585 		/* free server */
2586 		kmem_free(prev_srvr, sizeof (vdc_server_t));
2587 	}
2588 
2589 	vdc->server_list = NULL;
2590 	vdc->num_servers = 0;
2591 }
2592 
2593 /* -------------------------------------------------------------------------- */
2594 
2595 /*
2596  * Descriptor Ring helper routines
2597  */
2598 
2599 /*
2600  * Function:
2601  *	vdc_init_descriptor_ring()
2602  *
2603  * Description:
2604  *
2605  * Arguments:
2606  *	vdc	- soft state pointer for this instance of the device driver.
2607  *
2608  * Return Code:
2609  *	0	- Success
2610  */
2611 static int
2612 vdc_init_descriptor_ring(vdc_t *vdc)
2613 {
2614 	vd_dring_entry_t	*dep = NULL;	/* DRing Entry pointer */
2615 	int	status = 0;
2616 	int	i;
2617 
2618 	DMSG(vdc, 0, "[%d] initialized=%x\n", vdc->instance, vdc->initialized);
2619 
2620 	ASSERT(vdc != NULL);
2621 	ASSERT(mutex_owned(&vdc->lock));
2622 
2623 	/* ensure we have enough room to store max sized block */
2624 	ASSERT(maxphys <= VD_MAX_BLOCK_SIZE);
2625 
2626 	if ((vdc->initialized & VDC_DRING_INIT) == 0) {
2627 		DMSG(vdc, 0, "[%d] ldc_mem_dring_create\n", vdc->instance);
2628 		/*
2629 		 * Calculate the maximum block size we can transmit using one
2630 		 * Descriptor Ring entry from the attributes returned by the
2631 		 * vDisk server. This is subject to a minimum of 'maxphys'
2632 		 * as we do not have the capability to split requests over
2633 		 * multiple DRing entries.
2634 		 */
2635 		if ((vdc->max_xfer_sz * vdc->vdisk_bsize) < maxphys) {
2636 			DMSG(vdc, 0, "[%d] using minimum DRing size\n",
2637 			    vdc->instance);
2638 			vdc->dring_max_cookies = maxphys / PAGESIZE;
2639 		} else {
2640 			vdc->dring_max_cookies =
2641 			    (vdc->max_xfer_sz * vdc->vdisk_bsize) / PAGESIZE;
2642 		}
2643 		vdc->dring_entry_size = (sizeof (vd_dring_entry_t) +
2644 		    (sizeof (ldc_mem_cookie_t) *
2645 		    (vdc->dring_max_cookies - 1)));
2646 		vdc->dring_len = VD_DRING_LEN;
2647 
2648 		status = ldc_mem_dring_create(vdc->dring_len,
2649 		    vdc->dring_entry_size, &vdc->dring_hdl);
2650 		if ((vdc->dring_hdl == NULL) || (status != 0)) {
2651 			DMSG(vdc, 0, "[%d] Descriptor ring creation failed",
2652 			    vdc->instance);
2653 			return (status);
2654 		}
2655 		vdc->initialized |= VDC_DRING_INIT;
2656 	}
2657 
2658 	if ((vdc->initialized & VDC_DRING_BOUND) == 0) {
2659 		DMSG(vdc, 0, "[%d] ldc_mem_dring_bind\n", vdc->instance);
2660 		vdc->dring_cookie =
2661 		    kmem_zalloc(sizeof (ldc_mem_cookie_t), KM_SLEEP);
2662 
2663 		status = ldc_mem_dring_bind(vdc->curr_server->ldc_handle,
2664 		    vdc->dring_hdl,
2665 		    LDC_SHADOW_MAP|LDC_DIRECT_MAP, LDC_MEM_RW,
2666 		    &vdc->dring_cookie[0],
2667 		    &vdc->dring_cookie_count);
2668 		if (status != 0) {
2669 			DMSG(vdc, 0, "[%d] Failed to bind descriptor ring "
2670 			    "(%lx) to channel (%lx) status=%d\n",
2671 			    vdc->instance, vdc->dring_hdl,
2672 			    vdc->curr_server->ldc_handle, status);
2673 			return (status);
2674 		}
2675 		ASSERT(vdc->dring_cookie_count == 1);
2676 		vdc->initialized |= VDC_DRING_BOUND;
2677 	}
2678 
2679 	status = ldc_mem_dring_info(vdc->dring_hdl, &vdc->dring_mem_info);
2680 	if (status != 0) {
2681 		DMSG(vdc, 0,
2682 		    "[%d] Failed to get info for descriptor ring (%lx)\n",
2683 		    vdc->instance, vdc->dring_hdl);
2684 		return (status);
2685 	}
2686 
2687 	if ((vdc->initialized & VDC_DRING_LOCAL) == 0) {
2688 		DMSG(vdc, 0, "[%d] local dring\n", vdc->instance);
2689 
2690 		/* Allocate the local copy of this dring */
2691 		vdc->local_dring =
2692 		    kmem_zalloc(vdc->dring_len * sizeof (vdc_local_desc_t),
2693 		    KM_SLEEP);
2694 		vdc->initialized |= VDC_DRING_LOCAL;
2695 	}
2696 
2697 	/*
2698 	 * Mark all DRing entries as free and initialize the private
2699 	 * descriptor's memory handles. If any entry is initialized,
2700 	 * we need to free it later so we set the bit in 'initialized'
2701 	 * at the start.
2702 	 */
2703 	vdc->initialized |= VDC_DRING_ENTRY;
2704 	for (i = 0; i < vdc->dring_len; i++) {
2705 		dep = VDC_GET_DRING_ENTRY_PTR(vdc, i);
2706 		dep->hdr.dstate = VIO_DESC_FREE;
2707 
2708 		status = ldc_mem_alloc_handle(vdc->curr_server->ldc_handle,
2709 		    &vdc->local_dring[i].desc_mhdl);
2710 		if (status != 0) {
2711 			DMSG(vdc, 0, "![%d] Failed to alloc mem handle for"
2712 			    " descriptor %d", vdc->instance, i);
2713 			return (status);
2714 		}
2715 		vdc->local_dring[i].is_free = B_TRUE;
2716 		vdc->local_dring[i].dep = dep;
2717 	}
2718 
2719 	/* Initialize the starting index */
2720 	vdc->dring_curr_idx = VDC_DRING_FIRST_ENTRY;
2721 
2722 	return (status);
2723 }
2724 
2725 /*
2726  * Function:
2727  *	vdc_destroy_descriptor_ring()
2728  *
2729  * Description:
2730  *
2731  * Arguments:
2732  *	vdc	- soft state pointer for this instance of the device driver.
2733  *
2734  * Return Code:
2735  *	None
2736  */
2737 static void
2738 vdc_destroy_descriptor_ring(vdc_t *vdc)
2739 {
2740 	vdc_local_desc_t	*ldep = NULL;	/* Local Dring Entry Pointer */
2741 	ldc_mem_handle_t	mhdl = NULL;
2742 	ldc_mem_info_t		minfo;
2743 	int			status = -1;
2744 	int			i;	/* loop */
2745 
2746 	ASSERT(vdc != NULL);
2747 	ASSERT(mutex_owned(&vdc->lock));
2748 
2749 	DMSG(vdc, 0, "[%d] Entered\n", vdc->instance);
2750 
2751 	if (vdc->initialized & VDC_DRING_ENTRY) {
2752 		DMSG(vdc, 0,
2753 		    "[%d] Removing Local DRing entries\n", vdc->instance);
2754 		for (i = 0; i < vdc->dring_len; i++) {
2755 			ldep = &vdc->local_dring[i];
2756 			mhdl = ldep->desc_mhdl;
2757 
2758 			if (mhdl == NULL)
2759 				continue;
2760 
2761 			if ((status = ldc_mem_info(mhdl, &minfo)) != 0) {
2762 				DMSG(vdc, 0,
2763 				    "ldc_mem_info returned an error: %d\n",
2764 				    status);
2765 
2766 				/*
2767 				 * This must mean that the mem handle
2768 				 * is not valid. Clear it out so that
2769 				 * no one tries to use it.
2770 				 */
2771 				ldep->desc_mhdl = NULL;
2772 				continue;
2773 			}
2774 
2775 			if (minfo.status == LDC_BOUND) {
2776 				(void) ldc_mem_unbind_handle(mhdl);
2777 			}
2778 
2779 			(void) ldc_mem_free_handle(mhdl);
2780 
2781 			ldep->desc_mhdl = NULL;
2782 		}
2783 		vdc->initialized &= ~VDC_DRING_ENTRY;
2784 	}
2785 
2786 	if (vdc->initialized & VDC_DRING_LOCAL) {
2787 		DMSG(vdc, 0, "[%d] Freeing Local DRing\n", vdc->instance);
2788 		kmem_free(vdc->local_dring,
2789 		    vdc->dring_len * sizeof (vdc_local_desc_t));
2790 		vdc->initialized &= ~VDC_DRING_LOCAL;
2791 	}
2792 
2793 	if (vdc->initialized & VDC_DRING_BOUND) {
2794 		DMSG(vdc, 0, "[%d] Unbinding DRing\n", vdc->instance);
2795 		status = ldc_mem_dring_unbind(vdc->dring_hdl);
2796 		if (status == 0) {
2797 			vdc->initialized &= ~VDC_DRING_BOUND;
2798 		} else {
2799 			DMSG(vdc, 0, "[%d] Error %d unbinding DRing %lx",
2800 			    vdc->instance, status, vdc->dring_hdl);
2801 		}
2802 		kmem_free(vdc->dring_cookie, sizeof (ldc_mem_cookie_t));
2803 	}
2804 
2805 	if (vdc->initialized & VDC_DRING_INIT) {
2806 		DMSG(vdc, 0, "[%d] Destroying DRing\n", vdc->instance);
2807 		status = ldc_mem_dring_destroy(vdc->dring_hdl);
2808 		if (status == 0) {
2809 			vdc->dring_hdl = NULL;
2810 			bzero(&vdc->dring_mem_info, sizeof (ldc_mem_info_t));
2811 			vdc->initialized &= ~VDC_DRING_INIT;
2812 		} else {
2813 			DMSG(vdc, 0, "[%d] Error %d destroying DRing (%lx)",
2814 			    vdc->instance, status, vdc->dring_hdl);
2815 		}
2816 	}
2817 }
2818 
2819 /*
2820  * Function:
2821  *	vdc_map_to_shared_dring()
2822  *
2823  * Description:
2824  *	Copy contents of the local descriptor to the shared
2825  *	memory descriptor.
2826  *
2827  * Arguments:
2828  *	vdcp	- soft state pointer for this instance of the device driver.
2829  *	idx	- descriptor ring index
2830  *
2831  * Return Code:
2832  *	None
2833  */
2834 static int
2835 vdc_map_to_shared_dring(vdc_t *vdcp, int idx)
2836 {
2837 	vdc_local_desc_t	*ldep;
2838 	vd_dring_entry_t	*dep;
2839 	int			rv;
2840 
2841 	ldep = &(vdcp->local_dring[idx]);
2842 
2843 	/* for now leave in the old pop_mem_hdl stuff */
2844 	if (ldep->nbytes > 0) {
2845 		rv = vdc_populate_mem_hdl(vdcp, ldep);
2846 		if (rv) {
2847 			DMSG(vdcp, 0, "[%d] Cannot populate mem handle\n",
2848 			    vdcp->instance);
2849 			return (rv);
2850 		}
2851 	}
2852 
2853 	/*
2854 	 * fill in the data details into the DRing
2855 	 */
2856 	dep = ldep->dep;
2857 	ASSERT(dep != NULL);
2858 
2859 	dep->payload.req_id = VDC_GET_NEXT_REQ_ID(vdcp);
2860 	dep->payload.operation = ldep->operation;
2861 	dep->payload.addr = ldep->offset;
2862 	dep->payload.nbytes = ldep->nbytes;
2863 	dep->payload.status = (uint32_t)-1;	/* vds will set valid value */
2864 	dep->payload.slice = ldep->slice;
2865 	dep->hdr.dstate = VIO_DESC_READY;
2866 	dep->hdr.ack = 1;		/* request an ACK for every message */
2867 
2868 	return (0);
2869 }
2870 
2871 /*
2872  * Function:
2873  *	vdc_send_request
2874  *
2875  * Description:
2876  *	This routine writes the data to be transmitted to vds into the
2877  *	descriptor, notifies vds that the ring has been updated and
2878  *	then waits for the request to be processed.
2879  *
2880  * Arguments:
2881  *	vdcp	  - the soft state pointer
2882  *	operation - operation we want vds to perform (VD_OP_XXX)
2883  *	addr	  - address of data buf to be read/written.
2884  *	nbytes	  - number of bytes to read/write
2885  *	slice	  - the disk slice this request is for
2886  *	offset	  - relative disk offset
2887  *	bufp	  - buf of operation
2888  *	dir	  - direction of operation (READ/WRITE/BOTH)
2889  *
2890  * Return Codes:
2891  *	0
2892  *	ENXIO
2893  */
2894 static int
2895 vdc_send_request(vdc_t *vdcp, int operation, caddr_t addr,
2896     size_t nbytes, int slice, diskaddr_t offset, buf_t *bufp,
2897     vio_desc_direction_t dir, int flags)
2898 {
2899 	int	rv = 0;
2900 
2901 	ASSERT(vdcp != NULL);
2902 	ASSERT(slice == VD_SLICE_NONE || slice < V_NUMPAR);
2903 
2904 	mutex_enter(&vdcp->lock);
2905 
2906 	/*
2907 	 * If this is a block read/write operation we update the I/O statistics
2908 	 * to indicate that the request is being put on the waitq to be
2909 	 * serviced.
2910 	 *
2911 	 * We do it here (a common routine for both synchronous and strategy
2912 	 * calls) for performance reasons - we are already holding vdc->lock
2913 	 * so there is no extra locking overhead. We would have to explicitly
2914 	 * grab the 'lock' mutex to update the stats if we were to do this
2915 	 * higher up the stack in vdc_strategy() et. al.
2916 	 */
2917 	if ((operation == VD_OP_BREAD) || (operation == VD_OP_BWRITE)) {
2918 		DTRACE_IO1(start, buf_t *, bufp);
2919 		VD_KSTAT_WAITQ_ENTER(vdcp);
2920 	}
2921 
2922 	/*
2923 	 * If the request does not expect the state to be VDC_STATE_RUNNING
2924 	 * then we just try to populate the descriptor ring once.
2925 	 */
2926 	if (!(flags & VDC_OP_STATE_RUNNING)) {
2927 		rv = vdc_populate_descriptor(vdcp, operation, addr,
2928 		    nbytes, slice, offset, bufp, dir, flags);
2929 		goto done;
2930 	}
2931 
2932 	do {
2933 		while (vdcp->state != VDC_STATE_RUNNING) {
2934 
2935 			/* return error if detaching */
2936 			if (vdcp->state == VDC_STATE_DETACH) {
2937 				rv = ENXIO;
2938 				goto done;
2939 			}
2940 
2941 			/*
2942 			 * If we are panicking and the disk is not ready then
2943 			 * we can't send any request because we can't complete
2944 			 * the handshake now.
2945 			 */
2946 			if (ddi_in_panic()) {
2947 				rv = EIO;
2948 				goto done;
2949 			}
2950 
2951 			/*
2952 			 * If the state is faulted, notify that a new I/O is
2953 			 * being submitted to force the system to check if any
2954 			 * server has recovered.
2955 			 */
2956 			if (vdcp->state == VDC_STATE_FAILED) {
2957 				vdcp->io_pending = B_TRUE;
2958 				cv_signal(&vdcp->io_pending_cv);
2959 			}
2960 
2961 			cv_wait(&vdcp->running_cv, &vdcp->lock);
2962 
2963 			/* if service is still faulted then fail the request */
2964 			if (vdcp->state == VDC_STATE_FAILED) {
2965 				rv = EIO;
2966 				goto done;
2967 			}
2968 		}
2969 
2970 	} while (vdc_populate_descriptor(vdcp, operation, addr,
2971 	    nbytes, slice, offset, bufp, dir, flags));
2972 
2973 done:
2974 	/*
2975 	 * If this is a block read/write we update the I/O statistics kstat
2976 	 * to indicate that this request has been placed on the queue for
2977 	 * processing (i.e sent to the vDisk server) - iostat(1M) will
2978 	 * report the time waiting for the vDisk server under the %b column
2979 	 * In the case of an error we simply take it off the wait queue.
2980 	 */
2981 	if ((operation == VD_OP_BREAD) || (operation == VD_OP_BWRITE)) {
2982 		if (rv == 0) {
2983 			VD_KSTAT_WAITQ_TO_RUNQ(vdcp);
2984 			DTRACE_PROBE1(send, buf_t *, bufp);
2985 		} else {
2986 			VD_UPDATE_ERR_STATS(vdcp, vd_transerrs);
2987 			VD_KSTAT_WAITQ_EXIT(vdcp);
2988 			DTRACE_IO1(done, buf_t *, bufp);
2989 		}
2990 	}
2991 
2992 	mutex_exit(&vdcp->lock);
2993 
2994 	return (rv);
2995 }
2996 
2997 
2998 /*
2999  * Function:
3000  *	vdc_populate_descriptor
3001  *
3002  * Description:
3003  *	This routine writes the data to be transmitted to vds into the
3004  *	descriptor, notifies vds that the ring has been updated and
3005  *	then waits for the request to be processed.
3006  *
3007  * Arguments:
3008  *	vdcp	  - the soft state pointer
3009  *	operation - operation we want vds to perform (VD_OP_XXX)
3010  *	addr	  - address of data buf to be read/written.
3011  *	nbytes	  - number of bytes to read/write
3012  *	slice	  - the disk slice this request is for
3013  *	offset	  - relative disk offset
3014  *	bufp	  - buf of operation
3015  *	dir	  - direction of operation (READ/WRITE/BOTH)
3016  *
3017  * Return Codes:
3018  *	0
3019  *	EAGAIN
3020  *	ECONNRESET
3021  *	ENXIO
3022  */
3023 static int
3024 vdc_populate_descriptor(vdc_t *vdcp, int operation, caddr_t addr,
3025     size_t nbytes, int slice, diskaddr_t offset,
3026     buf_t *bufp, vio_desc_direction_t dir, int flags)
3027 {
3028 	vdc_local_desc_t	*local_dep = NULL; /* Local Dring Pointer */
3029 	int			idx;		/* Index of DRing entry used */
3030 	int			next_idx;
3031 	vio_dring_msg_t		dmsg;
3032 	size_t			msglen;
3033 	int			rv;
3034 
3035 	ASSERT(MUTEX_HELD(&vdcp->lock));
3036 	vdcp->threads_pending++;
3037 loop:
3038 	DMSG(vdcp, 2, ": dring_curr_idx = %d\n", vdcp->dring_curr_idx);
3039 
3040 	if (flags & VDC_OP_DRING_RESERVED) {
3041 		/* use D-Ring reserved entry */
3042 		idx = VDC_DRING_FIRST_RESV;
3043 		local_dep = &(vdcp->local_dring[idx]);
3044 	} else {
3045 		/* Get next available D-Ring entry */
3046 		idx = vdcp->dring_curr_idx;
3047 		local_dep = &(vdcp->local_dring[idx]);
3048 
3049 		if (!local_dep->is_free) {
3050 			DMSG(vdcp, 2, "[%d]: dring full - waiting for space\n",
3051 			    vdcp->instance);
3052 			cv_wait(&vdcp->dring_free_cv, &vdcp->lock);
3053 			if (vdcp->state == VDC_STATE_RUNNING ||
3054 			    vdcp->state == VDC_STATE_HANDLE_PENDING) {
3055 				goto loop;
3056 			}
3057 			vdcp->threads_pending--;
3058 			return (ECONNRESET);
3059 		}
3060 
3061 		next_idx = idx + 1;
3062 		if (next_idx >= vdcp->dring_len)
3063 			next_idx = VDC_DRING_FIRST_ENTRY;
3064 		vdcp->dring_curr_idx = next_idx;
3065 	}
3066 
3067 	ASSERT(local_dep->is_free);
3068 
3069 	local_dep->operation = operation;
3070 	local_dep->addr = addr;
3071 	local_dep->nbytes = nbytes;
3072 	local_dep->slice = slice;
3073 	local_dep->offset = offset;
3074 	local_dep->buf = bufp;
3075 	local_dep->dir = dir;
3076 	local_dep->flags = flags;
3077 
3078 	local_dep->is_free = B_FALSE;
3079 
3080 	rv = vdc_map_to_shared_dring(vdcp, idx);
3081 	if (rv) {
3082 		if (flags & VDC_OP_DRING_RESERVED) {
3083 			DMSG(vdcp, 0, "[%d]: cannot bind memory - error\n",
3084 			    vdcp->instance);
3085 			/*
3086 			 * We can't wait if we are using reserved slot.
3087 			 * Free the descriptor and return.
3088 			 */
3089 			local_dep->is_free = B_TRUE;
3090 			vdcp->threads_pending--;
3091 			return (rv);
3092 		}
3093 		DMSG(vdcp, 0, "[%d]: cannot bind memory - waiting ..\n",
3094 		    vdcp->instance);
3095 		/* free the descriptor */
3096 		local_dep->is_free = B_TRUE;
3097 		vdcp->dring_curr_idx = idx;
3098 		cv_wait(&vdcp->membind_cv, &vdcp->lock);
3099 		if (vdcp->state == VDC_STATE_RUNNING ||
3100 		    vdcp->state == VDC_STATE_HANDLE_PENDING) {
3101 			goto loop;
3102 		}
3103 		vdcp->threads_pending--;
3104 		return (ECONNRESET);
3105 	}
3106 
3107 	/*
3108 	 * Send a msg with the DRing details to vds
3109 	 */
3110 	VIO_INIT_DRING_DATA_TAG(dmsg);
3111 	VDC_INIT_DRING_DATA_MSG_IDS(dmsg, vdcp);
3112 	dmsg.dring_ident = vdcp->dring_ident;
3113 	dmsg.start_idx = idx;
3114 	dmsg.end_idx = idx;
3115 	vdcp->seq_num++;
3116 
3117 	DTRACE_PROBE2(populate, int, vdcp->instance,
3118 	    vdc_local_desc_t *, local_dep);
3119 	DMSG(vdcp, 2, "ident=0x%lx, st=%u, end=%u, seq=%ld\n",
3120 	    vdcp->dring_ident, dmsg.start_idx, dmsg.end_idx, dmsg.seq_num);
3121 
3122 	/*
3123 	 * note we're still holding the lock here to
3124 	 * make sure the message goes out in order !!!...
3125 	 */
3126 	msglen = sizeof (dmsg);
3127 	rv = vdc_send(vdcp, (caddr_t)&dmsg, &msglen);
3128 	switch (rv) {
3129 	case ECONNRESET:
3130 		/*
3131 		 * vdc_send initiates the reset on failure.
3132 		 * Since the transaction has already been put
3133 		 * on the local dring, it will automatically get
3134 		 * retried when the channel is reset. Given that,
3135 		 * it is ok to just return success even though the
3136 		 * send failed.
3137 		 */
3138 		rv = 0;
3139 		break;
3140 
3141 	case 0: /* EOK */
3142 		DMSG(vdcp, 1, "sent via LDC: rv=%d\n", rv);
3143 		break;
3144 
3145 	default:
3146 		DMSG(vdcp, 0, "unexpected error, rv=%d\n", rv);
3147 		rv = ENXIO;
3148 		break;
3149 	}
3150 
3151 	vdcp->threads_pending--;
3152 	return (rv);
3153 }
3154 
3155 /*
3156  * Function:
3157  *	vdc_do_op
3158  *
3159  * Description:
3160  * 	Wrapper around vdc_submit_request(). Each request is associated with a
3161  *	buf structure. If a buf structure is provided (bufp != NULL) then the
3162  *	request will be submitted with that buf, and the caller can wait for
3163  *	completion of the request with biowait(). If a buf structure is not
3164  *	provided (bufp == NULL) then a buf structure is created and the function
3165  *	waits for the completion of the request.
3166  *
3167  *	If the flag VD_OP_STATE_RUNNING is set then vdc_submit_request() will
3168  *	submit the request only when the vdisk is in state VD_STATE_RUNNING.
3169  *	If the vdisk is not in that state then the vdc_submit_request() will
3170  *	wait for that state to be reached. After the request is submitted, the
3171  *	reply will be processed asynchronously by the vdc_process_msg_thread()
3172  *	thread.
3173  *
3174  *	If the flag VD_OP_STATE_RUNNING is not set then vdc_submit_request()
3175  *	submit the request whatever the state of the vdisk is. Then vdc_do_op()
3176  *	will wait for a reply message, process the reply and complete the
3177  *	request.
3178  *
3179  * Arguments:
3180  *	vdc	- the soft state pointer
3181  *	op	- operation we want vds to perform (VD_OP_XXX)
3182  *	addr	- address of data buf to be read/written.
3183  *	nbytes	- number of bytes to read/write
3184  *	slice	- the disk slice this request is for
3185  *	offset	- relative disk offset
3186  *	bufp	- buf structure associated with the request (can be NULL).
3187  *	dir	- direction of operation (READ/WRITE/BOTH)
3188  *	flags	- flags for the request.
3189  *
3190  * Return Codes:
3191  *	0	- the request has been succesfully submitted and completed.
3192  *	!= 0	- the request has failed. In that case, if a buf structure
3193  *		  was provided (bufp != NULL) then the B_ERROR flag is set
3194  *		  and the b_error field of the buf structure is set to EIO.
3195  */
3196 static int
3197 vdc_do_op(vdc_t *vdc, int op, caddr_t addr, size_t nbytes, int slice,
3198     diskaddr_t offset, struct buf *bufp, vio_desc_direction_t dir, int flags)
3199 {
3200 	vio_msg_t vio_msg;
3201 	struct buf buf;
3202 	int rv;
3203 
3204 	if (bufp == NULL) {
3205 		/*
3206 		 * We use buf just as a convenient way to get a notification
3207 		 * that the request is completed, so we initialize buf to the
3208 		 * minimum we need.
3209 		 */
3210 		bioinit(&buf);
3211 		buf.b_bcount = nbytes;
3212 		buf.b_flags = B_BUSY;
3213 		bufp = &buf;
3214 	}
3215 
3216 	rv = vdc_send_request(vdc, op, addr, nbytes, slice, offset, bufp,
3217 	    dir, flags);
3218 
3219 	if (rv != 0)
3220 		goto done;
3221 
3222 	/*
3223 	 * If the request should be done in VDC_STATE_RUNNING state then the
3224 	 * reply will be received and processed by vdc_process_msg_thread()
3225 	 * and we just have to handle the panic case. Otherwise we have to
3226 	 * wait for the reply message and process it.
3227 	 */
3228 	if (flags & VDC_OP_STATE_RUNNING) {
3229 
3230 		if (ddi_in_panic()) {
3231 			rv = vdc_drain_response(vdc, bufp);
3232 			goto done;
3233 		}
3234 
3235 	} else {
3236 		/* wait for the response message */
3237 		rv = vdc_wait_for_response(vdc, &vio_msg);
3238 		if (rv) {
3239 			/*
3240 			 * If this is a block read/write we update the I/O
3241 			 * statistics kstat to take it off the run queue.
3242 			 */
3243 			mutex_enter(&vdc->lock);
3244 			if (op == VD_OP_BREAD || op == VD_OP_BWRITE) {
3245 				VD_UPDATE_ERR_STATS(vdc, vd_transerrs);
3246 				VD_KSTAT_RUNQ_EXIT(vdc);
3247 				DTRACE_IO1(done, buf_t *, bufp);
3248 			}
3249 			mutex_exit(&vdc->lock);
3250 			goto done;
3251 		}
3252 
3253 		rv = vdc_process_data_msg(vdc, &vio_msg);
3254 		if (rv)
3255 			goto done;
3256 	}
3257 
3258 	if (bufp == &buf)
3259 		rv = biowait(bufp);
3260 
3261 done:
3262 	if (bufp == &buf) {
3263 		biofini(bufp);
3264 	} else if (rv != 0) {
3265 		bioerror(bufp, EIO);
3266 		biodone(bufp);
3267 	}
3268 
3269 	return (rv);
3270 }
3271 
3272 /*
3273  * Function:
3274  *	vdc_do_sync_op
3275  *
3276  * Description:
3277  * 	Wrapper around vdc_do_op that serializes requests.
3278  *
3279  * Arguments:
3280  *	vdcp	  - the soft state pointer
3281  *	operation - operation we want vds to perform (VD_OP_XXX)
3282  *	addr	  - address of data buf to be read/written.
3283  *	nbytes	  - number of bytes to read/write
3284  *	slice	  - the disk slice this request is for
3285  *	offset	  - relative disk offset
3286  *	dir	  - direction of operation (READ/WRITE/BOTH)
3287  *	rconflict - check for reservation conflict in case of failure
3288  *
3289  * rconflict should be set to B_TRUE by most callers. Callers invoking the
3290  * VD_OP_SCSICMD operation can set rconflict to B_FALSE if they check the
3291  * result of a successful operation with vdc_scsi_status().
3292  *
3293  * Return Codes:
3294  *	0
3295  *	EAGAIN
3296  *	EFAULT
3297  *	ENXIO
3298  *	EIO
3299  */
3300 static int
3301 vdc_do_sync_op(vdc_t *vdcp, int operation, caddr_t addr, size_t nbytes,
3302     int slice, diskaddr_t offset, vio_desc_direction_t dir, boolean_t rconflict)
3303 {
3304 	int status;
3305 	int flags = VDC_OP_NORMAL;
3306 
3307 	/*
3308 	 * Grab the lock, if blocked wait until the server
3309 	 * response causes us to wake up again.
3310 	 */
3311 	mutex_enter(&vdcp->lock);
3312 	vdcp->sync_op_cnt++;
3313 	while (vdcp->sync_op_blocked && vdcp->state != VDC_STATE_DETACH) {
3314 		if (ddi_in_panic()) {
3315 			/* don't block if we are panicking */
3316 			vdcp->sync_op_cnt--;
3317 			mutex_exit(&vdcp->lock);
3318 			return (EIO);
3319 		} else {
3320 			cv_wait(&vdcp->sync_blocked_cv, &vdcp->lock);
3321 		}
3322 	}
3323 
3324 	if (vdcp->state == VDC_STATE_DETACH) {
3325 		cv_broadcast(&vdcp->sync_blocked_cv);
3326 		vdcp->sync_op_cnt--;
3327 		mutex_exit(&vdcp->lock);
3328 		return (ENXIO);
3329 	}
3330 
3331 	/* now block anyone other thread entering after us */
3332 	vdcp->sync_op_blocked = B_TRUE;
3333 
3334 	mutex_exit(&vdcp->lock);
3335 
3336 	if (!rconflict)
3337 		flags &= ~VDC_OP_ERRCHK_CONFLICT;
3338 
3339 	status = vdc_do_op(vdcp, operation, addr, nbytes, slice, offset,
3340 	    NULL, dir, flags);
3341 
3342 	mutex_enter(&vdcp->lock);
3343 
3344 	DMSG(vdcp, 2, ": operation returned %d\n", status);
3345 
3346 	if (vdcp->state == VDC_STATE_DETACH) {
3347 		status = ENXIO;
3348 	}
3349 
3350 	vdcp->sync_op_blocked = B_FALSE;
3351 	vdcp->sync_op_cnt--;
3352 
3353 	/* signal the next waiting thread */
3354 	cv_signal(&vdcp->sync_blocked_cv);
3355 
3356 	mutex_exit(&vdcp->lock);
3357 
3358 	return (status);
3359 }
3360 
3361 
3362 /*
3363  * Function:
3364  *	vdc_drain_response()
3365  *
3366  * Description:
3367  * 	When a guest is panicking, the completion of requests needs to be
3368  * 	handled differently because interrupts are disabled and vdc
3369  * 	will not get messages. We have to poll for the messages instead.
3370  *
3371  *	Note: since we are panicking we don't implement	the io:::done
3372  *	DTrace probe or update the I/O statistics kstats.
3373  *
3374  * Arguments:
3375  *	vdc	- soft state pointer for this instance of the device driver.
3376  *	buf	- if buf is NULL then we drain all responses, otherwise we
3377  *		  poll until we receive a ACK/NACK for the specific I/O
3378  *		  described by buf.
3379  *
3380  * Return Code:
3381  *	0	- Success. If we were expecting a response to a particular
3382  *		  request then this means that a response has been received.
3383  */
3384 static int
3385 vdc_drain_response(vdc_t *vdc, struct buf *buf)
3386 {
3387 	int 			rv, idx, retries;
3388 	size_t			msglen;
3389 	vdc_local_desc_t 	*ldep = NULL;	/* Local Dring Entry Pointer */
3390 	vio_dring_msg_t		dmsg;
3391 	struct buf		*mbuf;
3392 	boolean_t		ack;
3393 
3394 	mutex_enter(&vdc->lock);
3395 
3396 	retries = 0;
3397 	for (;;) {
3398 		msglen = sizeof (dmsg);
3399 		rv = ldc_read(vdc->curr_server->ldc_handle, (caddr_t)&dmsg,
3400 		    &msglen);
3401 		if (rv) {
3402 			rv = EINVAL;
3403 			break;
3404 		}
3405 
3406 		/*
3407 		 * if there are no packets wait and check again
3408 		 */
3409 		if ((rv == 0) && (msglen == 0)) {
3410 			if (retries++ > vdc_dump_retries) {
3411 				rv = EAGAIN;
3412 				break;
3413 			}
3414 
3415 			drv_usecwait(vdc_usec_timeout_dump);
3416 			continue;
3417 		}
3418 
3419 		/*
3420 		 * Ignore all messages that are not ACKs/NACKs to
3421 		 * DRing requests.
3422 		 */
3423 		if ((dmsg.tag.vio_msgtype != VIO_TYPE_DATA) ||
3424 		    (dmsg.tag.vio_subtype_env != VIO_DRING_DATA)) {
3425 			DMSG(vdc, 0, "discard pkt: type=%d sub=%d env=%d\n",
3426 			    dmsg.tag.vio_msgtype,
3427 			    dmsg.tag.vio_subtype,
3428 			    dmsg.tag.vio_subtype_env);
3429 			continue;
3430 		}
3431 
3432 		/*
3433 		 * Record if the packet was ACK'ed or not. If the packet was not
3434 		 * ACK'ed then we will just mark the request as failed; we don't
3435 		 * want to reset the connection at this point.
3436 		 */
3437 		switch (dmsg.tag.vio_subtype) {
3438 		case VIO_SUBTYPE_ACK:
3439 			ack = B_TRUE;
3440 			break;
3441 		case VIO_SUBTYPE_NACK:
3442 			ack = B_FALSE;
3443 			break;
3444 		default:
3445 			continue;
3446 		}
3447 
3448 		idx = dmsg.start_idx;
3449 		if (idx >= vdc->dring_len) {
3450 			DMSG(vdc, 0, "[%d] Bogus ack data : start %d\n",
3451 			    vdc->instance, idx);
3452 			continue;
3453 		}
3454 		ldep = &vdc->local_dring[idx];
3455 		if (ldep->dep->hdr.dstate != VIO_DESC_DONE) {
3456 			DMSG(vdc, 0, "[%d] Entry @ %d - state !DONE %d\n",
3457 			    vdc->instance, idx, ldep->dep->hdr.dstate);
3458 			continue;
3459 		}
3460 
3461 		mbuf = ldep->buf;
3462 		ASSERT(mbuf != NULL);
3463 		mbuf->b_resid = mbuf->b_bcount - ldep->dep->payload.nbytes;
3464 		bioerror(mbuf, ack ? ldep->dep->payload.status : EIO);
3465 		biodone(mbuf);
3466 
3467 		rv = vdc_depopulate_descriptor(vdc, idx);
3468 		if (buf != NULL && buf == mbuf) {
3469 			rv = 0;
3470 			goto done;
3471 		}
3472 
3473 		/* if this is the last descriptor - break out of loop */
3474 		if ((idx + 1) % vdc->dring_len == vdc->dring_curr_idx) {
3475 			/*
3476 			 * If we were expecting a response for a particular
3477 			 * request then we return with an error otherwise we
3478 			 * have successfully completed the drain.
3479 			 */
3480 			rv = (buf != NULL)? ESRCH: 0;
3481 			break;
3482 		}
3483 	}
3484 
3485 done:
3486 	mutex_exit(&vdc->lock);
3487 	DMSG(vdc, 0, "End idx=%d\n", idx);
3488 
3489 	return (rv);
3490 }
3491 
3492 
3493 /*
3494  * Function:
3495  *	vdc_depopulate_descriptor()
3496  *
3497  * Description:
3498  *
3499  * Arguments:
3500  *	vdc	- soft state pointer for this instance of the device driver.
3501  *	idx	- Index of the Descriptor Ring entry being modified
3502  *
3503  * Return Code:
3504  *	0	- Success
3505  */
3506 static int
3507 vdc_depopulate_descriptor(vdc_t *vdc, uint_t idx)
3508 {
3509 	vd_dring_entry_t *dep = NULL;		/* Dring Entry Pointer */
3510 	vdc_local_desc_t *ldep = NULL;		/* Local Dring Entry Pointer */
3511 	int		status = ENXIO;
3512 	int		rv = 0;
3513 
3514 	ASSERT(vdc != NULL);
3515 	ASSERT(idx < vdc->dring_len);
3516 	ldep = &vdc->local_dring[idx];
3517 	ASSERT(ldep != NULL);
3518 	ASSERT(MUTEX_HELD(&vdc->lock));
3519 
3520 	DTRACE_PROBE2(depopulate, int, vdc->instance, vdc_local_desc_t *, ldep);
3521 	DMSG(vdc, 2, ": idx = %d\n", idx);
3522 
3523 	dep = ldep->dep;
3524 	ASSERT(dep != NULL);
3525 	ASSERT((dep->hdr.dstate == VIO_DESC_DONE) ||
3526 	    (dep->payload.status == ECANCELED));
3527 
3528 	VDC_MARK_DRING_ENTRY_FREE(vdc, idx);
3529 
3530 	ldep->is_free = B_TRUE;
3531 	status = dep->payload.status;
3532 	DMSG(vdc, 2, ": is_free = %d : status = %d\n", ldep->is_free, status);
3533 
3534 	/*
3535 	 * If no buffers were used to transfer information to the server when
3536 	 * populating the descriptor then no memory handles need to be unbound
3537 	 * and we can return now.
3538 	 */
3539 	if (ldep->nbytes == 0) {
3540 		cv_signal(&vdc->dring_free_cv);
3541 		return (status);
3542 	}
3543 
3544 	/*
3545 	 * If the upper layer passed in a misaligned address we copied the
3546 	 * data into an aligned buffer before sending it to LDC - we now
3547 	 * copy it back to the original buffer.
3548 	 */
3549 	if (ldep->align_addr) {
3550 		ASSERT(ldep->addr != NULL);
3551 
3552 		if (dep->payload.nbytes > 0)
3553 			bcopy(ldep->align_addr, ldep->addr,
3554 			    dep->payload.nbytes);
3555 		kmem_free(ldep->align_addr,
3556 		    sizeof (caddr_t) * P2ROUNDUP(ldep->nbytes, 8));
3557 		ldep->align_addr = NULL;
3558 	}
3559 
3560 	rv = ldc_mem_unbind_handle(ldep->desc_mhdl);
3561 	if (rv != 0) {
3562 		DMSG(vdc, 0, "?[%d] unbind mhdl 0x%lx @ idx %d failed (%d)",
3563 		    vdc->instance, ldep->desc_mhdl, idx, rv);
3564 		/*
3565 		 * The error returned by the vDisk server is more informative
3566 		 * and thus has a higher priority but if it isn't set we ensure
3567 		 * that this function returns an error.
3568 		 */
3569 		if (status == 0)
3570 			status = EINVAL;
3571 	}
3572 
3573 	cv_signal(&vdc->membind_cv);
3574 	cv_signal(&vdc->dring_free_cv);
3575 
3576 	return (status);
3577 }
3578 
3579 /*
3580  * Function:
3581  *	vdc_populate_mem_hdl()
3582  *
3583  * Description:
3584  *
3585  * Arguments:
3586  *	vdc	- soft state pointer for this instance of the device driver.
3587  *	idx	- Index of the Descriptor Ring entry being modified
3588  *	addr	- virtual address being mapped in
3589  *	nybtes	- number of bytes in 'addr'
3590  *	operation - the vDisk operation being performed (VD_OP_xxx)
3591  *
3592  * Return Code:
3593  *	0	- Success
3594  */
3595 static int
3596 vdc_populate_mem_hdl(vdc_t *vdcp, vdc_local_desc_t *ldep)
3597 {
3598 	vd_dring_entry_t	*dep = NULL;
3599 	ldc_mem_handle_t	mhdl;
3600 	caddr_t			vaddr;
3601 	size_t			nbytes;
3602 	uint8_t			perm = LDC_MEM_RW;
3603 	uint8_t			maptype;
3604 	int			rv = 0;
3605 	int			i;
3606 
3607 	ASSERT(vdcp != NULL);
3608 
3609 	dep = ldep->dep;
3610 	mhdl = ldep->desc_mhdl;
3611 
3612 	switch (ldep->dir) {
3613 	case VIO_read_dir:
3614 		perm = LDC_MEM_W;
3615 		break;
3616 
3617 	case VIO_write_dir:
3618 		perm = LDC_MEM_R;
3619 		break;
3620 
3621 	case VIO_both_dir:
3622 		perm = LDC_MEM_RW;
3623 		break;
3624 
3625 	default:
3626 		ASSERT(0);	/* catch bad programming in vdc */
3627 	}
3628 
3629 	/*
3630 	 * LDC expects any addresses passed in to be 8-byte aligned. We need
3631 	 * to copy the contents of any misaligned buffers to a newly allocated
3632 	 * buffer and bind it instead (and copy the the contents back to the
3633 	 * original buffer passed in when depopulating the descriptor)
3634 	 */
3635 	vaddr = ldep->addr;
3636 	nbytes = ldep->nbytes;
3637 	if (((uint64_t)vaddr & 0x7) != 0) {
3638 		ASSERT(ldep->align_addr == NULL);
3639 		ldep->align_addr =
3640 		    kmem_alloc(sizeof (caddr_t) *
3641 		    P2ROUNDUP(nbytes, 8), KM_SLEEP);
3642 		DMSG(vdcp, 0, "[%d] Misaligned address %p reallocating "
3643 		    "(buf=%p nb=%ld op=%d)\n",
3644 		    vdcp->instance, (void *)vaddr, (void *)ldep->align_addr,
3645 		    nbytes, ldep->operation);
3646 		if (perm != LDC_MEM_W)
3647 			bcopy(vaddr, ldep->align_addr, nbytes);
3648 		vaddr = ldep->align_addr;
3649 	}
3650 
3651 	maptype = LDC_IO_MAP|LDC_SHADOW_MAP|LDC_DIRECT_MAP;
3652 	rv = ldc_mem_bind_handle(mhdl, vaddr, P2ROUNDUP(nbytes, 8),
3653 	    maptype, perm, &dep->payload.cookie[0], &dep->payload.ncookies);
3654 	DMSG(vdcp, 2, "[%d] bound mem handle; ncookies=%d\n",
3655 	    vdcp->instance, dep->payload.ncookies);
3656 	if (rv != 0) {
3657 		DMSG(vdcp, 0, "[%d] Failed to bind LDC memory handle "
3658 		    "(mhdl=%p, buf=%p, err=%d)\n",
3659 		    vdcp->instance, (void *)mhdl, (void *)vaddr, rv);
3660 		if (ldep->align_addr) {
3661 			kmem_free(ldep->align_addr,
3662 			    sizeof (caddr_t) * P2ROUNDUP(nbytes, 8));
3663 			ldep->align_addr = NULL;
3664 		}
3665 		return (EAGAIN);
3666 	}
3667 
3668 	/*
3669 	 * Get the other cookies (if any).
3670 	 */
3671 	for (i = 1; i < dep->payload.ncookies; i++) {
3672 		rv = ldc_mem_nextcookie(mhdl, &dep->payload.cookie[i]);
3673 		if (rv != 0) {
3674 			(void) ldc_mem_unbind_handle(mhdl);
3675 			DMSG(vdcp, 0, "?[%d] Failed to get next cookie "
3676 			    "(mhdl=%lx cnum=%d), err=%d",
3677 			    vdcp->instance, mhdl, i, rv);
3678 			if (ldep->align_addr) {
3679 				kmem_free(ldep->align_addr,
3680 				    sizeof (caddr_t) * ldep->nbytes);
3681 				ldep->align_addr = NULL;
3682 			}
3683 			return (EAGAIN);
3684 		}
3685 	}
3686 
3687 	return (rv);
3688 }
3689 
3690 /*
3691  * Interrupt handlers for messages from LDC
3692  */
3693 
3694 /*
3695  * Function:
3696  *	vdc_handle_cb()
3697  *
3698  * Description:
3699  *
3700  * Arguments:
3701  *	event	- Type of event (LDC_EVT_xxx) that triggered the callback
3702  *	arg	- soft state pointer for this instance of the device driver.
3703  *
3704  * Return Code:
3705  *	0	- Success
3706  */
3707 static uint_t
3708 vdc_handle_cb(uint64_t event, caddr_t arg)
3709 {
3710 	ldc_status_t	ldc_state;
3711 	int		rv = 0;
3712 	vdc_server_t	*srvr = (vdc_server_t *)(void *)arg;
3713 	vdc_t		*vdc = srvr->vdcp;
3714 
3715 	ASSERT(vdc != NULL);
3716 
3717 	DMSG(vdc, 1, "evt=%lx seqID=%ld\n", event, vdc->seq_num);
3718 
3719 	/* If callback is not for the current server, ignore it */
3720 	mutex_enter(&vdc->lock);
3721 
3722 	if (vdc->curr_server != srvr) {
3723 		DMSG(vdc, 0, "[%d] Ignoring event 0x%lx for port@%ld\n",
3724 		    vdc->instance, event, srvr->id);
3725 		mutex_exit(&vdc->lock);
3726 		return (LDC_SUCCESS);
3727 	}
3728 
3729 	/*
3730 	 * Depending on the type of event that triggered this callback,
3731 	 * we modify the handshake state or read the data.
3732 	 *
3733 	 * NOTE: not done as a switch() as event could be triggered by
3734 	 * a state change and a read request. Also the ordering	of the
3735 	 * check for the event types is deliberate.
3736 	 */
3737 	if (event & LDC_EVT_UP) {
3738 		DMSG(vdc, 0, "[%d] Received LDC_EVT_UP\n", vdc->instance);
3739 
3740 		/* get LDC state */
3741 		rv = ldc_status(srvr->ldc_handle, &ldc_state);
3742 		if (rv != 0) {
3743 			DMSG(vdc, 0, "[%d] Couldn't get LDC status %d",
3744 			    vdc->instance, rv);
3745 			mutex_exit(&vdc->lock);
3746 			return (LDC_SUCCESS);
3747 		}
3748 		if (srvr->ldc_state != LDC_UP &&
3749 		    ldc_state == LDC_UP) {
3750 			/*
3751 			 * Reset the transaction sequence numbers when
3752 			 * LDC comes up. We then kick off the handshake
3753 			 * negotiation with the vDisk server.
3754 			 */
3755 			vdc->seq_num = 1;
3756 			vdc->seq_num_reply = 0;
3757 			vdc->io_pending = B_TRUE;
3758 			srvr->ldc_state = ldc_state;
3759 			cv_signal(&vdc->initwait_cv);
3760 			cv_signal(&vdc->io_pending_cv);
3761 		}
3762 	}
3763 
3764 	if (event & LDC_EVT_READ) {
3765 		DMSG(vdc, 1, "[%d] Received LDC_EVT_READ\n", vdc->instance);
3766 		mutex_enter(&vdc->read_lock);
3767 		cv_signal(&vdc->read_cv);
3768 		vdc->read_state = VDC_READ_PENDING;
3769 		mutex_exit(&vdc->read_lock);
3770 		mutex_exit(&vdc->lock);
3771 
3772 		/* that's all we have to do - no need to handle DOWN/RESET */
3773 		return (LDC_SUCCESS);
3774 	}
3775 
3776 	if (event & (LDC_EVT_RESET|LDC_EVT_DOWN)) {
3777 
3778 		DMSG(vdc, 0, "[%d] Received LDC RESET event\n", vdc->instance);
3779 
3780 		/*
3781 		 * Need to wake up any readers so they will
3782 		 * detect that a reset has occurred.
3783 		 */
3784 		mutex_enter(&vdc->read_lock);
3785 		if ((vdc->read_state == VDC_READ_WAITING) ||
3786 		    (vdc->read_state == VDC_READ_RESET))
3787 			cv_signal(&vdc->read_cv);
3788 		vdc->read_state = VDC_READ_RESET;
3789 		mutex_exit(&vdc->read_lock);
3790 
3791 		/* wake up any threads waiting for connection to come up */
3792 		if (vdc->state == VDC_STATE_INIT_WAITING) {
3793 			vdc->state = VDC_STATE_RESETTING;
3794 			cv_signal(&vdc->initwait_cv);
3795 		} else if (vdc->state == VDC_STATE_FAILED) {
3796 			vdc->io_pending = B_TRUE;
3797 			cv_signal(&vdc->io_pending_cv);
3798 		}
3799 
3800 	}
3801 
3802 	mutex_exit(&vdc->lock);
3803 
3804 	if (event & ~(LDC_EVT_UP | LDC_EVT_RESET | LDC_EVT_DOWN | LDC_EVT_READ))
3805 		DMSG(vdc, 0, "![%d] Unexpected LDC event (%lx) received",
3806 		    vdc->instance, event);
3807 
3808 	return (LDC_SUCCESS);
3809 }
3810 
3811 /*
3812  * Function:
3813  *	vdc_wait_for_response()
3814  *
3815  * Description:
3816  *	Block waiting for a response from the server. If there is
3817  *	no data the thread block on the read_cv that is signalled
3818  *	by the callback when an EVT_READ occurs.
3819  *
3820  * Arguments:
3821  *	vdcp	- soft state pointer for this instance of the device driver.
3822  *
3823  * Return Code:
3824  *	0	- Success
3825  */
3826 static int
3827 vdc_wait_for_response(vdc_t *vdcp, vio_msg_t *msgp)
3828 {
3829 	size_t		nbytes = sizeof (*msgp);
3830 	int		status;
3831 
3832 	ASSERT(vdcp != NULL);
3833 
3834 	DMSG(vdcp, 1, "[%d] Entered\n", vdcp->instance);
3835 
3836 	status = vdc_recv(vdcp, msgp, &nbytes);
3837 	DMSG(vdcp, 3, "vdc_read() done.. status=0x%x size=0x%x\n",
3838 	    status, (int)nbytes);
3839 	if (status) {
3840 		DMSG(vdcp, 0, "?[%d] Error %d reading LDC msg\n",
3841 		    vdcp->instance, status);
3842 		return (status);
3843 	}
3844 
3845 	if (nbytes < sizeof (vio_msg_tag_t)) {
3846 		DMSG(vdcp, 0, "?[%d] Expect %lu bytes; recv'd %lu\n",
3847 		    vdcp->instance, sizeof (vio_msg_tag_t), nbytes);
3848 		return (ENOMSG);
3849 	}
3850 
3851 	DMSG(vdcp, 2, "[%d] (%x/%x/%x)\n", vdcp->instance,
3852 	    msgp->tag.vio_msgtype,
3853 	    msgp->tag.vio_subtype,
3854 	    msgp->tag.vio_subtype_env);
3855 
3856 	/*
3857 	 * Verify the Session ID of the message
3858 	 *
3859 	 * Every message after the Version has been negotiated should
3860 	 * have the correct session ID set.
3861 	 */
3862 	if ((msgp->tag.vio_sid != vdcp->session_id) &&
3863 	    (msgp->tag.vio_subtype_env != VIO_VER_INFO)) {
3864 		DMSG(vdcp, 0, "[%d] Invalid SID: received 0x%x, "
3865 		    "expected 0x%lx [seq num %lx @ %d]",
3866 		    vdcp->instance, msgp->tag.vio_sid,
3867 		    vdcp->session_id,
3868 		    ((vio_dring_msg_t *)msgp)->seq_num,
3869 		    ((vio_dring_msg_t *)msgp)->start_idx);
3870 		return (ENOMSG);
3871 	}
3872 	return (0);
3873 }
3874 
3875 
3876 /*
3877  * Function:
3878  *	vdc_resubmit_backup_dring()
3879  *
3880  * Description:
3881  *	Resubmit each descriptor in the backed up dring to
3882  * 	vDisk server. The Dring was backed up during connection
3883  *	reset.
3884  *
3885  * Arguments:
3886  *	vdcp	- soft state pointer for this instance of the device driver.
3887  *
3888  * Return Code:
3889  *	0	- Success
3890  */
3891 static int
3892 vdc_resubmit_backup_dring(vdc_t *vdcp)
3893 {
3894 	int		processed = 0;
3895 	int		count;
3896 	int		b_idx;
3897 	int		rv = 0;
3898 	int		dring_size;
3899 	vdc_local_desc_t	*curr_ldep;
3900 
3901 	ASSERT(MUTEX_NOT_HELD(&vdcp->lock));
3902 	ASSERT(vdcp->state == VDC_STATE_HANDLE_PENDING);
3903 
3904 	if (vdcp->local_dring_backup == NULL) {
3905 		/* the pending requests have already been processed */
3906 		return (0);
3907 	}
3908 
3909 	DMSG(vdcp, 1, "restoring pending dring entries (len=%d, tail=%d)\n",
3910 	    vdcp->local_dring_backup_len, vdcp->local_dring_backup_tail);
3911 
3912 	/*
3913 	 * Walk the backup copy of the local descriptor ring and
3914 	 * resubmit all the outstanding transactions.
3915 	 */
3916 	b_idx = vdcp->local_dring_backup_tail;
3917 	for (count = 0; count < vdcp->local_dring_backup_len; count++) {
3918 
3919 		curr_ldep = &(vdcp->local_dring_backup[b_idx]);
3920 
3921 		/* only resubmit outstanding transactions */
3922 		if (!curr_ldep->is_free) {
3923 
3924 			DMSG(vdcp, 1, "resubmitting entry idx=%x\n", b_idx);
3925 
3926 			rv = vdc_do_op(vdcp, curr_ldep->operation,
3927 			    curr_ldep->addr, curr_ldep->nbytes,
3928 			    curr_ldep->slice, curr_ldep->offset,
3929 			    curr_ldep->buf, curr_ldep->dir,
3930 			    curr_ldep->flags & ~VDC_OP_STATE_RUNNING);
3931 
3932 			if (rv) {
3933 				DMSG(vdcp, 1, "[%d] resubmit entry %d failed\n",
3934 				    vdcp->instance, b_idx);
3935 				goto done;
3936 			}
3937 
3938 			/*
3939 			 * Mark this entry as free so that we will not resubmit
3940 			 * this "done" request again, if we were to use the same
3941 			 * backup_dring again in future. This could happen when
3942 			 * a reset happens while processing the backup_dring.
3943 			 */
3944 			curr_ldep->is_free = B_TRUE;
3945 			processed++;
3946 		}
3947 
3948 		/* get the next element to submit */
3949 		if (++b_idx >= vdcp->local_dring_backup_len)
3950 			b_idx = 0;
3951 	}
3952 
3953 	/* all done - now clear up pending dring copy */
3954 	dring_size = vdcp->local_dring_backup_len *
3955 	    sizeof (vdcp->local_dring_backup[0]);
3956 
3957 	(void) kmem_free(vdcp->local_dring_backup, dring_size);
3958 
3959 	vdcp->local_dring_backup = NULL;
3960 
3961 done:
3962 	DTRACE_PROBE2(processed, int, processed, vdc_t *, vdcp);
3963 
3964 	return (rv);
3965 }
3966 
3967 /*
3968  * Function:
3969  *	vdc_cancel_backup_dring
3970  *
3971  * Description:
3972  *	Cancel each descriptor in the backed up dring to vDisk server.
3973  *	The Dring was backed up during connection reset.
3974  *
3975  * Arguments:
3976  *	vdcp	- soft state pointer for this instance of the device driver.
3977  *
3978  * Return Code:
3979  *	None
3980  */
3981 void
3982 vdc_cancel_backup_dring(vdc_t *vdcp)
3983 {
3984 	vdc_local_desc_t *ldep;
3985 	struct buf 	*bufp;
3986 	int		count;
3987 	int		b_idx;
3988 	int		dring_size;
3989 	int		cancelled = 0;
3990 
3991 	ASSERT(MUTEX_HELD(&vdcp->lock));
3992 	ASSERT(vdcp->state == VDC_STATE_FAILED);
3993 
3994 	if (vdcp->local_dring_backup == NULL) {
3995 		/* the pending requests have already been processed */
3996 		return;
3997 	}
3998 
3999 	DMSG(vdcp, 1, "cancelling pending dring entries (len=%d, tail=%d)\n",
4000 	    vdcp->local_dring_backup_len, vdcp->local_dring_backup_tail);
4001 
4002 	/*
4003 	 * Walk the backup copy of the local descriptor ring and
4004 	 * cancel all the outstanding transactions.
4005 	 */
4006 	b_idx = vdcp->local_dring_backup_tail;
4007 	for (count = 0; count < vdcp->local_dring_backup_len; count++) {
4008 
4009 		ldep = &(vdcp->local_dring_backup[b_idx]);
4010 
4011 		/* only cancel outstanding transactions */
4012 		if (!ldep->is_free) {
4013 
4014 			DMSG(vdcp, 1, "cancelling entry idx=%x\n", b_idx);
4015 			cancelled++;
4016 
4017 			/*
4018 			 * All requests have already been cleared from the
4019 			 * local descriptor ring and the LDC channel has been
4020 			 * reset so we will never get any reply for these
4021 			 * requests. Now we just have to notify threads waiting
4022 			 * for replies that the request has failed.
4023 			 */
4024 			bufp = ldep->buf;
4025 			ASSERT(bufp != NULL);
4026 			bufp->b_resid = bufp->b_bcount;
4027 			if (ldep->operation == VD_OP_BREAD ||
4028 			    ldep->operation == VD_OP_BWRITE) {
4029 				VD_UPDATE_ERR_STATS(vdcp, vd_softerrs);
4030 				VD_KSTAT_RUNQ_EXIT(vdcp);
4031 				DTRACE_IO1(done, buf_t *, bufp);
4032 			}
4033 			bioerror(bufp, EIO);
4034 			biodone(bufp);
4035 		}
4036 
4037 		/* get the next element to cancel */
4038 		if (++b_idx >= vdcp->local_dring_backup_len)
4039 			b_idx = 0;
4040 	}
4041 
4042 	/* all done - now clear up pending dring copy */
4043 	dring_size = vdcp->local_dring_backup_len *
4044 	    sizeof (vdcp->local_dring_backup[0]);
4045 
4046 	(void) kmem_free(vdcp->local_dring_backup, dring_size);
4047 
4048 	vdcp->local_dring_backup = NULL;
4049 
4050 	DTRACE_PROBE2(cancelled, int, cancelled, vdc_t *, vdcp);
4051 }
4052 
4053 /*
4054  * Function:
4055  *	vdc_connection_timeout
4056  *
4057  * Description:
4058  *	This function is invoked if the timeout set to establish the connection
4059  *	with vds expires. This will happen if we spend too much time in the
4060  *	VDC_STATE_INIT_WAITING or VDC_STATE_NEGOTIATE states.
4061  *
4062  *	If the timeout does not expire, it will be cancelled when we reach the
4063  *	VDC_STATE_HANDLE_PENDING, VDC_STATE_FAILED or VDC_STATE_DETACH state.
4064  *	This function can also be invoked while we are in those states, in
4065  *	which case we do nothing because the timeout is being cancelled.
4066  *
4067  * Arguments:
4068  *	arg	- argument of the timeout function actually a soft state
4069  *		  pointer for the instance of the device driver.
4070  *
4071  * Return Code:
4072  *	None
4073  */
4074 void
4075 vdc_connection_timeout(void *arg)
4076 {
4077 	vdc_t 		*vdcp = (vdc_t *)arg;
4078 
4079 	mutex_enter(&vdcp->lock);
4080 
4081 	if (vdcp->state == VDC_STATE_HANDLE_PENDING ||
4082 	    vdcp->state == VDC_STATE_DETACH ||
4083 	    vdcp->state == VDC_STATE_FAILED) {
4084 		/*
4085 		 * The connection has just been re-established, has failed or
4086 		 * we are detaching.
4087 		 */
4088 		vdcp->ctimeout_reached = B_FALSE;
4089 	} else {
4090 		vdcp->ctimeout_reached = B_TRUE;
4091 	}
4092 
4093 	mutex_exit(&vdcp->lock);
4094 }
4095 
4096 /*
4097  * Function:
4098  *	vdc_backup_local_dring()
4099  *
4100  * Description:
4101  *	Backup the current dring in the event of a reset. The Dring
4102  *	transactions will be resubmitted to the server when the
4103  *	connection is restored.
4104  *
4105  * Arguments:
4106  *	vdcp	- soft state pointer for this instance of the device driver.
4107  *
4108  * Return Code:
4109  *	NONE
4110  */
4111 static void
4112 vdc_backup_local_dring(vdc_t *vdcp)
4113 {
4114 	int dring_size;
4115 
4116 	ASSERT(MUTEX_HELD(&vdcp->lock));
4117 	ASSERT(vdcp->state == VDC_STATE_RESETTING);
4118 
4119 	/*
4120 	 * If the backup dring is stil around, it means
4121 	 * that the last restore did not complete. However,
4122 	 * since we never got back into the running state,
4123 	 * the backup copy we have is still valid.
4124 	 */
4125 	if (vdcp->local_dring_backup != NULL) {
4126 		DMSG(vdcp, 1, "reusing local descriptor ring backup "
4127 		    "(len=%d, tail=%d)\n", vdcp->local_dring_backup_len,
4128 		    vdcp->local_dring_backup_tail);
4129 		return;
4130 	}
4131 
4132 	/*
4133 	 * The backup dring can be NULL and the local dring may not be
4134 	 * initialized. This can happen if we had a reset while establishing
4135 	 * a new connection but after the connection has timed out. In that
4136 	 * case the backup dring is NULL because the requests have been
4137 	 * cancelled and the request occured before the local dring is
4138 	 * initialized.
4139 	 */
4140 	if (!(vdcp->initialized & VDC_DRING_LOCAL))
4141 		return;
4142 
4143 	DMSG(vdcp, 1, "backing up the local descriptor ring (len=%d, "
4144 	    "tail=%d)\n", vdcp->dring_len, vdcp->dring_curr_idx);
4145 
4146 	dring_size = vdcp->dring_len * sizeof (vdcp->local_dring[0]);
4147 
4148 	vdcp->local_dring_backup = kmem_alloc(dring_size, KM_SLEEP);
4149 	bcopy(vdcp->local_dring, vdcp->local_dring_backup, dring_size);
4150 
4151 	vdcp->local_dring_backup_tail = vdcp->dring_curr_idx;
4152 	vdcp->local_dring_backup_len = vdcp->dring_len;
4153 }
4154 
4155 static void
4156 vdc_switch_server(vdc_t *vdcp)
4157 {
4158 	int		rv;
4159 	vdc_server_t 	*curr_server, *new_server;
4160 
4161 	ASSERT(MUTEX_HELD(&vdcp->lock));
4162 
4163 	/* if there is only one server return back */
4164 	if (vdcp->num_servers == 1) {
4165 		return;
4166 	}
4167 
4168 	/* Get current and next server */
4169 	curr_server = vdcp->curr_server;
4170 	new_server =
4171 	    (curr_server->next) ? curr_server->next : vdcp->server_list;
4172 	ASSERT(curr_server != new_server);
4173 
4174 	/* bring current server's channel down */
4175 	rv = ldc_down(curr_server->ldc_handle);
4176 	if (rv) {
4177 		DMSG(vdcp, 0, "[%d] Cannot bring channel down, port %ld\n",
4178 		    vdcp->instance, curr_server->id);
4179 		return;
4180 	}
4181 
4182 	/* switch the server */
4183 	vdcp->curr_server = new_server;
4184 
4185 	DMSG(vdcp, 0, "[%d] Switched to next vdisk server, port@%ld, ldc@%ld\n",
4186 	    vdcp->instance, vdcp->curr_server->id, vdcp->curr_server->ldc_id);
4187 }
4188 
4189 static void
4190 vdc_print_svc_status(vdc_t *vdcp)
4191 {
4192 	int instance;
4193 	uint64_t ldc_id, port_id;
4194 	vdc_service_state_t svc_state;
4195 
4196 	ASSERT(mutex_owned(&vdcp->lock));
4197 
4198 	svc_state = vdcp->curr_server->svc_state;
4199 
4200 	if (vdcp->curr_server->log_state == svc_state)
4201 		return;
4202 
4203 	instance = vdcp->instance;
4204 	ldc_id = vdcp->curr_server->ldc_id;
4205 	port_id = vdcp->curr_server->id;
4206 
4207 	switch (svc_state) {
4208 
4209 	case VDC_SERVICE_OFFLINE:
4210 		cmn_err(CE_CONT, "?vdisk@%d is offline\n", instance);
4211 		break;
4212 
4213 	case VDC_SERVICE_CONNECTED:
4214 		cmn_err(CE_CONT, "?vdisk@%d is connected using ldc@%ld,%ld\n",
4215 		    instance, ldc_id, port_id);
4216 		break;
4217 
4218 	case VDC_SERVICE_ONLINE:
4219 		cmn_err(CE_CONT, "?vdisk@%d is online using ldc@%ld,%ld\n",
4220 		    instance, ldc_id, port_id);
4221 		break;
4222 
4223 	case VDC_SERVICE_FAILED:
4224 		cmn_err(CE_CONT, "?vdisk@%d access to service failed "
4225 		    "using ldc@%ld,%ld\n", instance, ldc_id, port_id);
4226 		break;
4227 
4228 	case VDC_SERVICE_FAULTED:
4229 		cmn_err(CE_CONT, "?vdisk@%d access to backend failed "
4230 		    "using ldc@%ld,%ld\n", instance, ldc_id, port_id);
4231 		break;
4232 
4233 	default:
4234 		ASSERT(0);
4235 		break;
4236 	}
4237 
4238 	vdcp->curr_server->log_state = svc_state;
4239 }
4240 
4241 /* -------------------------------------------------------------------------- */
4242 
4243 /*
4244  * The following functions process the incoming messages from vds
4245  */
4246 
4247 /*
4248  * Function:
4249  *      vdc_process_msg_thread()
4250  *
4251  * Description:
4252  *
4253  *	Main VDC message processing thread. Each vDisk instance
4254  * 	consists of a copy of this thread. This thread triggers
4255  * 	all the handshakes and data exchange with the server. It
4256  * 	also handles all channel resets
4257  *
4258  * Arguments:
4259  *      vdc     - soft state pointer for this instance of the device driver.
4260  *
4261  * Return Code:
4262  *      None
4263  */
4264 static void
4265 vdc_process_msg_thread(vdc_t *vdcp)
4266 {
4267 	int		status;
4268 	int		ctimeout;
4269 	timeout_id_t	tmid = 0;
4270 	clock_t		ldcup_timeout = 0;
4271 	vdc_server_t	*srvr;
4272 	vdc_service_state_t svc_state;
4273 
4274 	mutex_enter(&vdcp->lock);
4275 
4276 	for (;;) {
4277 
4278 #define	Q(_s)	(vdcp->state == _s) ? #_s :
4279 		DMSG(vdcp, 3, "state = %d (%s)\n", vdcp->state,
4280 		    Q(VDC_STATE_INIT)
4281 		    Q(VDC_STATE_INIT_WAITING)
4282 		    Q(VDC_STATE_NEGOTIATE)
4283 		    Q(VDC_STATE_HANDLE_PENDING)
4284 		    Q(VDC_STATE_FAULTED)
4285 		    Q(VDC_STATE_FAILED)
4286 		    Q(VDC_STATE_RUNNING)
4287 		    Q(VDC_STATE_RESETTING)
4288 		    Q(VDC_STATE_DETACH)
4289 		    "UNKNOWN");
4290 
4291 		switch (vdcp->state) {
4292 		case VDC_STATE_INIT:
4293 
4294 			/*
4295 			 * If requested, start a timeout to check if the
4296 			 * connection with vds is established in the
4297 			 * specified delay. If the timeout expires, we
4298 			 * will cancel any pending request.
4299 			 *
4300 			 * If some reset have occurred while establishing
4301 			 * the connection, we already have a timeout armed
4302 			 * and in that case we don't need to arm a new one.
4303 			 *
4304 			 * The same rule applies when there are multiple vds'.
4305 			 * If either a connection cannot be established or
4306 			 * the handshake times out, the connection thread will
4307 			 * try another server. The 'ctimeout' will report
4308 			 * back an error after it expires irrespective of
4309 			 * whether the vdisk is trying to connect to just
4310 			 * one or multiple servers.
4311 			 */
4312 			ctimeout = (vdc_timeout != 0)?
4313 			    vdc_timeout : vdcp->curr_server->ctimeout;
4314 
4315 			if (ctimeout != 0 && tmid == 0) {
4316 				tmid = timeout(vdc_connection_timeout, vdcp,
4317 				    ctimeout * drv_usectohz(MICROSEC));
4318 			}
4319 
4320 			/* Switch to STATE_DETACH if drv is detaching */
4321 			if (vdcp->lifecycle == VDC_LC_DETACHING) {
4322 				vdcp->state = VDC_STATE_DETACH;
4323 				break;
4324 			}
4325 
4326 			/* Check if the timeout has been reached */
4327 			if (vdcp->ctimeout_reached) {
4328 				ASSERT(tmid != 0);
4329 				tmid = 0;
4330 				vdcp->state = VDC_STATE_FAILED;
4331 				break;
4332 			}
4333 
4334 			/* Check if we are re-initializing repeatedly */
4335 			if (vdcp->hshake_cnt > vdc_hshake_retries &&
4336 			    vdcp->lifecycle != VDC_LC_ONLINE) {
4337 
4338 				DMSG(vdcp, 0, "[%d] too many handshakes,cnt=%d",
4339 				    vdcp->instance, vdcp->hshake_cnt);
4340 				vdcp->state = VDC_STATE_FAILED;
4341 				break;
4342 			}
4343 
4344 			/* Switch server */
4345 			if (vdcp->hshake_cnt > 0)
4346 				vdc_switch_server(vdcp);
4347 			vdcp->hshake_cnt++;
4348 
4349 			/* Bring up connection with vds via LDC */
4350 			status = vdc_start_ldc_connection(vdcp);
4351 			if (status != EINVAL) {
4352 				vdcp->state = VDC_STATE_INIT_WAITING;
4353 			} else {
4354 				vdcp->curr_server->svc_state =
4355 				    VDC_SERVICE_FAILED;
4356 				vdc_print_svc_status(vdcp);
4357 			}
4358 			break;
4359 
4360 		case VDC_STATE_INIT_WAITING:
4361 
4362 			/* if channel is UP, start negotiation */
4363 			if (vdcp->curr_server->ldc_state == LDC_UP) {
4364 				vdcp->state = VDC_STATE_NEGOTIATE;
4365 				break;
4366 			}
4367 
4368 			/*
4369 			 * Wait for LDC_UP. If it times out and we have multiple
4370 			 * servers then we will retry using a different server.
4371 			 */
4372 			ldcup_timeout = ddi_get_lbolt() + (vdc_ldcup_timeout *
4373 			    drv_usectohz(MICROSEC));
4374 			status = cv_timedwait(&vdcp->initwait_cv, &vdcp->lock,
4375 			    ldcup_timeout);
4376 			if (status == -1 &&
4377 			    vdcp->state == VDC_STATE_INIT_WAITING &&
4378 			    vdcp->curr_server->ldc_state != LDC_UP) {
4379 				/* timed out & still waiting */
4380 				vdcp->curr_server->svc_state =
4381 				    VDC_SERVICE_FAILED;
4382 				vdc_print_svc_status(vdcp);
4383 				vdcp->state = VDC_STATE_INIT;
4384 				break;
4385 			}
4386 
4387 			if (vdcp->state != VDC_STATE_INIT_WAITING) {
4388 				DMSG(vdcp, 0,
4389 				    "state moved to %d out from under us...\n",
4390 				    vdcp->state);
4391 			}
4392 			break;
4393 
4394 		case VDC_STATE_NEGOTIATE:
4395 			switch (status = vdc_ver_negotiation(vdcp)) {
4396 			case 0:
4397 				break;
4398 			default:
4399 				DMSG(vdcp, 0, "ver negotiate failed (%d)..\n",
4400 				    status);
4401 				goto reset;
4402 			}
4403 
4404 			switch (status = vdc_attr_negotiation(vdcp)) {
4405 			case 0:
4406 				break;
4407 			default:
4408 				DMSG(vdcp, 0, "attr negotiate failed (%d)..\n",
4409 				    status);
4410 				goto reset;
4411 			}
4412 
4413 			switch (status = vdc_dring_negotiation(vdcp)) {
4414 			case 0:
4415 				break;
4416 			default:
4417 				DMSG(vdcp, 0, "dring negotiate failed (%d)..\n",
4418 				    status);
4419 				goto reset;
4420 			}
4421 
4422 			switch (status = vdc_rdx_exchange(vdcp)) {
4423 			case 0:
4424 				vdcp->state = VDC_STATE_HANDLE_PENDING;
4425 				goto done;
4426 			default:
4427 				DMSG(vdcp, 0, "RDX xchg failed ..(%d)\n",
4428 				    status);
4429 				goto reset;
4430 			}
4431 reset:
4432 			DMSG(vdcp, 0, "negotiation failed: resetting (%d)\n",
4433 			    status);
4434 			vdcp->state = VDC_STATE_RESETTING;
4435 			vdcp->self_reset = B_TRUE;
4436 			vdcp->curr_server->svc_state = VDC_SERVICE_FAILED;
4437 			vdc_print_svc_status(vdcp);
4438 done:
4439 			DMSG(vdcp, 0, "negotiation complete (state=0x%x)...\n",
4440 			    vdcp->state);
4441 			break;
4442 
4443 		case VDC_STATE_HANDLE_PENDING:
4444 
4445 			DMSG(vdcp, 0, "[%d] connection to service domain is up",
4446 			    vdcp->instance);
4447 			vdcp->curr_server->svc_state = VDC_SERVICE_CONNECTED;
4448 
4449 			mutex_exit(&vdcp->lock);
4450 
4451 			/*
4452 			 * If we have multiple servers, check that the backend
4453 			 * is effectively available before resubmitting any IO.
4454 			 */
4455 			if (vdcp->num_servers > 1 &&
4456 			    vdc_eio_check(vdcp, 0) != 0) {
4457 				mutex_enter(&vdcp->lock);
4458 				vdcp->curr_server->svc_state =
4459 				    VDC_SERVICE_FAULTED;
4460 				vdcp->state = VDC_STATE_FAULTED;
4461 				break;
4462 			}
4463 
4464 			if (tmid != 0) {
4465 				(void) untimeout(tmid);
4466 				tmid = 0;
4467 				vdcp->ctimeout_reached = B_FALSE;
4468 			}
4469 
4470 			/*
4471 			 * Setup devid
4472 			 */
4473 			(void) vdc_setup_devid(vdcp);
4474 
4475 			status = vdc_resubmit_backup_dring(vdcp);
4476 
4477 			mutex_enter(&vdcp->lock);
4478 
4479 			if (status) {
4480 				vdcp->state = VDC_STATE_RESETTING;
4481 				vdcp->self_reset = B_TRUE;
4482 				vdcp->curr_server->svc_state =
4483 				    VDC_SERVICE_FAILED;
4484 				vdc_print_svc_status(vdcp);
4485 			} else {
4486 				vdcp->state = VDC_STATE_RUNNING;
4487 			}
4488 			break;
4489 
4490 		case VDC_STATE_FAULTED:
4491 			/*
4492 			 * Server is faulted because the backend is unavailable.
4493 			 * If all servers are faulted then we mark the service
4494 			 * as failed, otherwise we reset to switch to another
4495 			 * server.
4496 			 */
4497 			vdc_print_svc_status(vdcp);
4498 
4499 			/* check if all servers are faulted */
4500 			for (srvr = vdcp->server_list; srvr != NULL;
4501 			    srvr = srvr->next) {
4502 				svc_state = srvr->svc_state;
4503 				if (svc_state != VDC_SERVICE_FAULTED)
4504 					break;
4505 			}
4506 
4507 			if (srvr != NULL) {
4508 				vdcp->state = VDC_STATE_RESETTING;
4509 				vdcp->self_reset = B_TRUE;
4510 			} else {
4511 				vdcp->state = VDC_STATE_FAILED;
4512 			}
4513 			break;
4514 
4515 		case VDC_STATE_FAILED:
4516 			/*
4517 			 * We reach this state when we are unable to access the
4518 			 * backend from any server, either because of a maximum
4519 			 * connection retries or timeout, or because the backend
4520 			 * is unavailable.
4521 			 *
4522 			 * Then we cancel the backup DRing so that errors get
4523 			 * reported and we wait for a new I/O before attempting
4524 			 * another connection.
4525 			 */
4526 			cmn_err(CE_NOTE, "vdisk@%d disk access failed",
4527 			    vdcp->instance);
4528 
4529 			/* cancel any timeout */
4530 			if (tmid != 0) {
4531 				(void) untimeout(tmid);
4532 				tmid = 0;
4533 			}
4534 
4535 			/* cancel pending I/Os */
4536 			cv_broadcast(&vdcp->running_cv);
4537 			vdc_cancel_backup_dring(vdcp);
4538 
4539 			/* wait for new I/O */
4540 			while (!vdcp->io_pending)
4541 				cv_wait(&vdcp->io_pending_cv, &vdcp->lock);
4542 
4543 			/*
4544 			 * There's a new IO pending. Try to re-establish a
4545 			 * connection. Mark all services as offline, so that
4546 			 * we don't stop again before having retried all
4547 			 * servers.
4548 			 */
4549 			for (srvr = vdcp->server_list; srvr != NULL;
4550 			    srvr = srvr->next) {
4551 				srvr->svc_state = VDC_SERVICE_OFFLINE;
4552 			}
4553 
4554 			/* reset variables */
4555 			vdcp->hshake_cnt = 0;
4556 			vdcp->ctimeout_reached = B_FALSE;
4557 
4558 			vdcp->state = VDC_STATE_RESETTING;
4559 			vdcp->self_reset = B_TRUE;
4560 			break;
4561 
4562 		/* enter running state */
4563 		case VDC_STATE_RUNNING:
4564 			/*
4565 			 * Signal anyone waiting for the connection
4566 			 * to come on line.
4567 			 */
4568 			vdcp->hshake_cnt = 0;
4569 			cv_broadcast(&vdcp->running_cv);
4570 
4571 			/* backend has to be checked after reset */
4572 			if (vdcp->failfast_interval != 0 ||
4573 			    vdcp->num_servers > 1)
4574 				cv_signal(&vdcp->eio_cv);
4575 
4576 			/* ownership is lost during reset */
4577 			if (vdcp->ownership & VDC_OWNERSHIP_WANTED)
4578 				vdcp->ownership |= VDC_OWNERSHIP_RESET;
4579 			cv_signal(&vdcp->ownership_cv);
4580 
4581 			vdcp->curr_server->svc_state = VDC_SERVICE_ONLINE;
4582 			vdc_print_svc_status(vdcp);
4583 
4584 			mutex_exit(&vdcp->lock);
4585 
4586 			for (;;) {
4587 				vio_msg_t msg;
4588 				status = vdc_wait_for_response(vdcp, &msg);
4589 				if (status) break;
4590 
4591 				DMSG(vdcp, 1, "[%d] new pkt(s) available\n",
4592 				    vdcp->instance);
4593 				status = vdc_process_data_msg(vdcp, &msg);
4594 				if (status) {
4595 					DMSG(vdcp, 1, "[%d] process_data_msg "
4596 					    "returned err=%d\n", vdcp->instance,
4597 					    status);
4598 					break;
4599 				}
4600 
4601 			}
4602 
4603 			mutex_enter(&vdcp->lock);
4604 
4605 			/* all servers are now offline */
4606 			for (srvr = vdcp->server_list; srvr != NULL;
4607 			    srvr = srvr->next) {
4608 				srvr->svc_state = VDC_SERVICE_OFFLINE;
4609 				srvr->log_state = VDC_SERVICE_NONE;
4610 			}
4611 
4612 			vdc_print_svc_status(vdcp);
4613 
4614 			vdcp->state = VDC_STATE_RESETTING;
4615 			vdcp->self_reset = B_TRUE;
4616 			break;
4617 
4618 		case VDC_STATE_RESETTING:
4619 			/*
4620 			 * When we reach this state, we either come from the
4621 			 * VDC_STATE_RUNNING state and we can have pending
4622 			 * request but no timeout is armed; or we come from
4623 			 * the VDC_STATE_INIT_WAITING, VDC_NEGOTIATE or
4624 			 * VDC_HANDLE_PENDING state and there is no pending
4625 			 * request or pending requests have already been copied
4626 			 * into the backup dring. So we can safely keep the
4627 			 * connection timeout armed while we are in this state.
4628 			 */
4629 
4630 			DMSG(vdcp, 0, "Initiating channel reset "
4631 			    "(pending = %d)\n", (int)vdcp->threads_pending);
4632 
4633 			if (vdcp->self_reset) {
4634 				DMSG(vdcp, 0,
4635 				    "[%d] calling stop_ldc_connection.\n",
4636 				    vdcp->instance);
4637 				status = vdc_stop_ldc_connection(vdcp);
4638 				vdcp->self_reset = B_FALSE;
4639 			}
4640 
4641 			/*
4642 			 * Wait for all threads currently waiting
4643 			 * for a free dring entry to use.
4644 			 */
4645 			while (vdcp->threads_pending) {
4646 				cv_broadcast(&vdcp->membind_cv);
4647 				cv_broadcast(&vdcp->dring_free_cv);
4648 				mutex_exit(&vdcp->lock);
4649 				/* give the waiters enough time to wake up */
4650 				delay(vdc_hz_min_ldc_delay);
4651 				mutex_enter(&vdcp->lock);
4652 			}
4653 
4654 			ASSERT(vdcp->threads_pending == 0);
4655 
4656 			/* Sanity check that no thread is receiving */
4657 			ASSERT(vdcp->read_state != VDC_READ_WAITING);
4658 
4659 			vdcp->read_state = VDC_READ_IDLE;
4660 			vdcp->io_pending = B_FALSE;
4661 
4662 			/*
4663 			 * Cleanup any pending eio. These I/Os are going to
4664 			 * be resubmitted.
4665 			 */
4666 			vdc_eio_unqueue(vdcp, 0, B_FALSE);
4667 
4668 			vdc_backup_local_dring(vdcp);
4669 
4670 			/* cleanup the old d-ring */
4671 			vdc_destroy_descriptor_ring(vdcp);
4672 
4673 			/* go and start again */
4674 			vdcp->state = VDC_STATE_INIT;
4675 
4676 			break;
4677 
4678 		case VDC_STATE_DETACH:
4679 			DMSG(vdcp, 0, "[%d] Reset thread exit cleanup ..\n",
4680 			    vdcp->instance);
4681 
4682 			/* cancel any pending timeout */
4683 			mutex_exit(&vdcp->lock);
4684 			if (tmid != 0) {
4685 				(void) untimeout(tmid);
4686 				tmid = 0;
4687 			}
4688 			mutex_enter(&vdcp->lock);
4689 
4690 			/*
4691 			 * Signal anyone waiting for connection
4692 			 * to come online
4693 			 */
4694 			cv_broadcast(&vdcp->running_cv);
4695 
4696 			while (vdcp->sync_op_cnt > 0) {
4697 				cv_broadcast(&vdcp->sync_blocked_cv);
4698 				mutex_exit(&vdcp->lock);
4699 				/* give the waiters enough time to wake up */
4700 				delay(vdc_hz_min_ldc_delay);
4701 				mutex_enter(&vdcp->lock);
4702 			}
4703 
4704 			mutex_exit(&vdcp->lock);
4705 
4706 			DMSG(vdcp, 0, "[%d] Msg processing thread exiting ..\n",
4707 			    vdcp->instance);
4708 			thread_exit();
4709 			break;
4710 		}
4711 	}
4712 }
4713 
4714 
4715 /*
4716  * Function:
4717  *	vdc_process_data_msg()
4718  *
4719  * Description:
4720  *	This function is called by the message processing thread each time
4721  *	a message with a msgtype of VIO_TYPE_DATA is received. It will either
4722  *	be an ACK or NACK from vds[1] which vdc handles as follows.
4723  *		ACK	- wake up the waiting thread
4724  *		NACK	- resend any messages necessary
4725  *
4726  *	[1] Although the message format allows it, vds should not send a
4727  *	    VIO_SUBTYPE_INFO message to vdc asking it to read data; if for
4728  *	    some bizarre reason it does, vdc will reset the connection.
4729  *
4730  * Arguments:
4731  *	vdc	- soft state pointer for this instance of the device driver.
4732  *	msg	- the LDC message sent by vds
4733  *
4734  * Return Code:
4735  *	0	- Success.
4736  *	> 0	- error value returned by LDC
4737  */
4738 static int
4739 vdc_process_data_msg(vdc_t *vdcp, vio_msg_t *msg)
4740 {
4741 	int			status = 0;
4742 	vio_dring_msg_t		*dring_msg;
4743 	vdc_local_desc_t	*ldep = NULL;
4744 	int			start, end;
4745 	int			idx;
4746 	int			op;
4747 
4748 	dring_msg = (vio_dring_msg_t *)msg;
4749 
4750 	ASSERT(msg->tag.vio_msgtype == VIO_TYPE_DATA);
4751 	ASSERT(vdcp != NULL);
4752 
4753 	mutex_enter(&vdcp->lock);
4754 
4755 	/*
4756 	 * Check to see if the message has bogus data
4757 	 */
4758 	idx = start = dring_msg->start_idx;
4759 	end = dring_msg->end_idx;
4760 	if ((start >= vdcp->dring_len) ||
4761 	    (end >= vdcp->dring_len) || (end < -1)) {
4762 		/*
4763 		 * Update the I/O statistics to indicate that an error ocurred.
4764 		 * No need to update the wait/run queues as no specific read or
4765 		 * write request is being completed in response to this 'msg'.
4766 		 */
4767 		VD_UPDATE_ERR_STATS(vdcp, vd_softerrs);
4768 		DMSG(vdcp, 0, "[%d] Bogus ACK data : start %d, end %d\n",
4769 		    vdcp->instance, start, end);
4770 		mutex_exit(&vdcp->lock);
4771 		return (EINVAL);
4772 	}
4773 
4774 	/*
4775 	 * Verify that the sequence number is what vdc expects.
4776 	 */
4777 	switch (vdc_verify_seq_num(vdcp, dring_msg)) {
4778 	case VDC_SEQ_NUM_TODO:
4779 		break;	/* keep processing this message */
4780 	case VDC_SEQ_NUM_SKIP:
4781 		mutex_exit(&vdcp->lock);
4782 		return (0);
4783 	case VDC_SEQ_NUM_INVALID:
4784 		/*
4785 		 * Update the I/O statistics to indicate that an error ocurred.
4786 		 * No need to update the wait/run queues as no specific read or
4787 		 * write request is being completed in response to this 'msg'.
4788 		 */
4789 		VD_UPDATE_ERR_STATS(vdcp, vd_softerrs);
4790 		DMSG(vdcp, 0, "[%d] invalid seqno\n", vdcp->instance);
4791 		mutex_exit(&vdcp->lock);
4792 		return (ENXIO);
4793 	}
4794 
4795 	if (msg->tag.vio_subtype == VIO_SUBTYPE_NACK) {
4796 		/*
4797 		 * Update the I/O statistics to indicate that an error ocurred.
4798 		 *
4799 		 * We need to update the run queue if a read or write request
4800 		 * is being NACKed - otherwise there will appear to be an
4801 		 * indefinite outstanding request and statistics reported by
4802 		 * iostat(1M) will be incorrect. The transaction will be
4803 		 * resubmitted from the backup DRing following the reset
4804 		 * and the wait/run queues will be entered again.
4805 		 */
4806 		ldep = &vdcp->local_dring[idx];
4807 		op = ldep->operation;
4808 		if ((op == VD_OP_BREAD) || (op == VD_OP_BWRITE)) {
4809 			DTRACE_IO1(done, buf_t *, ldep->buf);
4810 			VD_KSTAT_RUNQ_EXIT(vdcp);
4811 		}
4812 		VD_UPDATE_ERR_STATS(vdcp, vd_softerrs);
4813 		VDC_DUMP_DRING_MSG(dring_msg);
4814 		DMSG(vdcp, 0, "[%d] DATA NACK\n", vdcp->instance);
4815 		mutex_exit(&vdcp->lock);
4816 		return (EIO);
4817 
4818 	} else if (msg->tag.vio_subtype == VIO_SUBTYPE_INFO) {
4819 		/*
4820 		 * Update the I/O statistics to indicate that an error occurred.
4821 		 * No need to update the wait/run queues as no specific read or
4822 		 * write request is being completed in response to this 'msg'.
4823 		 */
4824 		VD_UPDATE_ERR_STATS(vdcp, vd_protoerrs);
4825 		mutex_exit(&vdcp->lock);
4826 		return (EPROTO);
4827 	}
4828 
4829 	DMSG(vdcp, 1, ": start %d end %d\n", start, end);
4830 	ASSERT(start == end);
4831 
4832 	ldep = &vdcp->local_dring[idx];
4833 
4834 	DMSG(vdcp, 1, ": state 0x%x\n", ldep->dep->hdr.dstate);
4835 
4836 	if (ldep->dep->hdr.dstate == VIO_DESC_DONE) {
4837 		struct buf *bufp;
4838 
4839 		status = ldep->dep->payload.status;
4840 
4841 		bufp = ldep->buf;
4842 		ASSERT(bufp != NULL);
4843 
4844 		bufp->b_resid = bufp->b_bcount - ldep->dep->payload.nbytes;
4845 		bioerror(bufp, status);
4846 
4847 		if (status != 0) {
4848 			DMSG(vdcp, 1, "I/O status=%d\n", status);
4849 		}
4850 
4851 		DMSG(vdcp, 1,
4852 		    "I/O complete req=%ld bytes resp=%ld bytes\n",
4853 		    bufp->b_bcount, ldep->dep->payload.nbytes);
4854 
4855 		/*
4856 		 * If the request has failed and we have multiple servers or
4857 		 * failfast is enabled then we will have to defer the completion
4858 		 * of the request until we have checked that the vdisk backend
4859 		 * is effectively available (if multiple server) or that there
4860 		 * is no reservation conflict (if failfast).
4861 		 */
4862 		if (status != 0 &&
4863 		    ((vdcp->num_servers > 1 &&
4864 		    (ldep->flags & VDC_OP_ERRCHK_BACKEND)) ||
4865 		    (vdcp->failfast_interval != 0 &&
4866 		    (ldep->flags & VDC_OP_ERRCHK_CONFLICT)))) {
4867 			/*
4868 			 * The I/O has failed and we need to check the error.
4869 			 */
4870 			(void) vdc_eio_queue(vdcp, idx);
4871 		} else {
4872 			op = ldep->operation;
4873 			if (op == VD_OP_BREAD || op == VD_OP_BWRITE) {
4874 				if (status == 0) {
4875 					VD_UPDATE_IO_STATS(vdcp, op,
4876 					    ldep->dep->payload.nbytes);
4877 				} else {
4878 					VD_UPDATE_ERR_STATS(vdcp, vd_softerrs);
4879 				}
4880 				VD_KSTAT_RUNQ_EXIT(vdcp);
4881 				DTRACE_IO1(done, buf_t *, bufp);
4882 			}
4883 			(void) vdc_depopulate_descriptor(vdcp, idx);
4884 			biodone(bufp);
4885 		}
4886 	}
4887 
4888 	/* let the arrival signal propogate */
4889 	mutex_exit(&vdcp->lock);
4890 
4891 	/* probe gives the count of how many entries were processed */
4892 	DTRACE_PROBE2(processed, int, 1, vdc_t *, vdcp);
4893 
4894 	return (0);
4895 }
4896 
4897 
4898 /*
4899  * Function:
4900  *	vdc_handle_ver_msg()
4901  *
4902  * Description:
4903  *
4904  * Arguments:
4905  *	vdc	- soft state pointer for this instance of the device driver.
4906  *	ver_msg	- LDC message sent by vDisk server
4907  *
4908  * Return Code:
4909  *	0	- Success
4910  */
4911 static int
4912 vdc_handle_ver_msg(vdc_t *vdc, vio_ver_msg_t *ver_msg)
4913 {
4914 	int status = 0;
4915 
4916 	ASSERT(vdc != NULL);
4917 	ASSERT(mutex_owned(&vdc->lock));
4918 
4919 	if (ver_msg->tag.vio_subtype_env != VIO_VER_INFO) {
4920 		return (EPROTO);
4921 	}
4922 
4923 	if (ver_msg->dev_class != VDEV_DISK_SERVER) {
4924 		return (EINVAL);
4925 	}
4926 
4927 	switch (ver_msg->tag.vio_subtype) {
4928 	case VIO_SUBTYPE_ACK:
4929 		/*
4930 		 * We check to see if the version returned is indeed supported
4931 		 * (The server may have also adjusted the minor number downwards
4932 		 * and if so 'ver_msg' will contain the actual version agreed)
4933 		 */
4934 		if (vdc_is_supported_version(ver_msg)) {
4935 			vdc->ver.major = ver_msg->ver_major;
4936 			vdc->ver.minor = ver_msg->ver_minor;
4937 			ASSERT(vdc->ver.major > 0);
4938 		} else {
4939 			status = EPROTO;
4940 		}
4941 		break;
4942 
4943 	case VIO_SUBTYPE_NACK:
4944 		/*
4945 		 * call vdc_is_supported_version() which will return the next
4946 		 * supported version (if any) in 'ver_msg'
4947 		 */
4948 		(void) vdc_is_supported_version(ver_msg);
4949 		if (ver_msg->ver_major > 0) {
4950 			size_t len = sizeof (*ver_msg);
4951 
4952 			ASSERT(vdc->ver.major > 0);
4953 
4954 			/* reset the necessary fields and resend */
4955 			ver_msg->tag.vio_subtype = VIO_SUBTYPE_INFO;
4956 			ver_msg->dev_class = VDEV_DISK;
4957 
4958 			status = vdc_send(vdc, (caddr_t)ver_msg, &len);
4959 			DMSG(vdc, 0, "[%d] Resend VER info (LDC status = %d)\n",
4960 			    vdc->instance, status);
4961 			if (len != sizeof (*ver_msg))
4962 				status = EBADMSG;
4963 		} else {
4964 			DMSG(vdc, 0, "[%d] No common version with vDisk server",
4965 			    vdc->instance);
4966 			status = ENOTSUP;
4967 		}
4968 
4969 		break;
4970 	case VIO_SUBTYPE_INFO:
4971 		/*
4972 		 * Handle the case where vds starts handshake
4973 		 * (for now only vdc is the instigator)
4974 		 */
4975 		status = ENOTSUP;
4976 		break;
4977 
4978 	default:
4979 		status = EINVAL;
4980 		break;
4981 	}
4982 
4983 	return (status);
4984 }
4985 
4986 /*
4987  * Function:
4988  *	vdc_handle_attr_msg()
4989  *
4990  * Description:
4991  *
4992  * Arguments:
4993  *	vdc	- soft state pointer for this instance of the device driver.
4994  *	attr_msg	- LDC message sent by vDisk server
4995  *
4996  * Return Code:
4997  *	0	- Success
4998  */
4999 static int
5000 vdc_handle_attr_msg(vdc_t *vdc, vd_attr_msg_t *attr_msg)
5001 {
5002 	int status = 0;
5003 	vd_disk_type_t old_type;
5004 
5005 	ASSERT(vdc != NULL);
5006 	ASSERT(mutex_owned(&vdc->lock));
5007 
5008 	if (attr_msg->tag.vio_subtype_env != VIO_ATTR_INFO) {
5009 		return (EPROTO);
5010 	}
5011 
5012 	switch (attr_msg->tag.vio_subtype) {
5013 	case VIO_SUBTYPE_ACK:
5014 		/*
5015 		 * We now verify the attributes sent by vds.
5016 		 */
5017 		if (attr_msg->vdisk_size == 0) {
5018 			DMSG(vdc, 0, "[%d] Invalid disk size from vds",
5019 			    vdc->instance);
5020 			status = EINVAL;
5021 			break;
5022 		}
5023 
5024 		if (attr_msg->max_xfer_sz == 0) {
5025 			DMSG(vdc, 0, "[%d] Invalid transfer size from vds",
5026 			    vdc->instance);
5027 			status = EINVAL;
5028 			break;
5029 		}
5030 
5031 		if (attr_msg->vdisk_size == VD_SIZE_UNKNOWN) {
5032 			DMSG(vdc, 0, "[%d] Unknown disk size from vds",
5033 			    vdc->instance);
5034 			attr_msg->vdisk_size = 0;
5035 		}
5036 
5037 		/* update the VIO block size */
5038 		if (attr_msg->vdisk_block_size > 0 &&
5039 		    vdc_update_vio_bsize(vdc,
5040 		    attr_msg->vdisk_block_size) != 0) {
5041 			DMSG(vdc, 0, "[%d] Invalid block size (%u) from vds",
5042 			    vdc->instance, attr_msg->vdisk_block_size);
5043 			status = EINVAL;
5044 			break;
5045 		}
5046 
5047 		/* update disk, block and transfer sizes */
5048 		old_type = vdc->vdisk_type;
5049 		vdc_update_size(vdc, attr_msg->vdisk_size,
5050 		    attr_msg->vdisk_block_size, attr_msg->max_xfer_sz);
5051 		vdc->vdisk_type = attr_msg->vdisk_type;
5052 		vdc->operations = attr_msg->operations;
5053 		if (vio_ver_is_supported(vdc->ver, 1, 1))
5054 			vdc->vdisk_media = attr_msg->vdisk_media;
5055 		else
5056 			vdc->vdisk_media = 0;
5057 
5058 		DMSG(vdc, 0, "[%d] max_xfer_sz: sent %lx acked %lx\n",
5059 		    vdc->instance, vdc->max_xfer_sz, attr_msg->max_xfer_sz);
5060 		DMSG(vdc, 0, "[%d] vdisk_block_size: sent %lx acked %x\n",
5061 		    vdc->instance, vdc->vdisk_bsize,
5062 		    attr_msg->vdisk_block_size);
5063 
5064 		if ((attr_msg->xfer_mode != VIO_DRING_MODE_V1_0) ||
5065 		    (attr_msg->vdisk_size > INT64_MAX) ||
5066 		    (attr_msg->operations == 0) ||
5067 		    (attr_msg->vdisk_type > VD_DISK_TYPE_DISK)) {
5068 			DMSG(vdc, 0, "[%d] Invalid attributes from vds",
5069 			    vdc->instance);
5070 			status = EINVAL;
5071 			break;
5072 		}
5073 
5074 		/*
5075 		 * Now that we have received all attributes we can create a
5076 		 * fake geometry for the disk.
5077 		 */
5078 		vdc_create_fake_geometry(vdc);
5079 
5080 		/*
5081 		 * If the disk type was previously unknown and device nodes
5082 		 * were created then the driver would have created 8 device
5083 		 * nodes. If we now find out that this is a single-slice disk
5084 		 * then we need to re-create the appropriate device nodes.
5085 		 */
5086 		if (old_type == VD_DISK_TYPE_UNK &&
5087 		    (vdc->initialized & VDC_MINOR) &&
5088 		    vdc->vdisk_type == VD_DISK_TYPE_SLICE) {
5089 			ddi_remove_minor_node(vdc->dip, NULL);
5090 			(void) devfs_clean(ddi_get_parent(vdc->dip),
5091 			    NULL, DV_CLEAN_FORCE);
5092 			if (vdc_create_device_nodes(vdc) != 0) {
5093 				DMSG(vdc, 0, "![%d] Failed to update "
5094 				    "device nodes", vdc->instance);
5095 			}
5096 		}
5097 
5098 		break;
5099 
5100 	case VIO_SUBTYPE_NACK:
5101 		/*
5102 		 * vds could not handle the attributes we sent so we
5103 		 * stop negotiating.
5104 		 */
5105 		status = EPROTO;
5106 		break;
5107 
5108 	case VIO_SUBTYPE_INFO:
5109 		/*
5110 		 * Handle the case where vds starts the handshake
5111 		 * (for now; vdc is the only supported instigatior)
5112 		 */
5113 		status = ENOTSUP;
5114 		break;
5115 
5116 	default:
5117 		status = ENOTSUP;
5118 		break;
5119 	}
5120 
5121 	return (status);
5122 }
5123 
5124 /*
5125  * Function:
5126  *	vdc_handle_dring_reg_msg()
5127  *
5128  * Description:
5129  *
5130  * Arguments:
5131  *	vdc		- soft state pointer for this instance of the driver.
5132  *	dring_msg	- LDC message sent by vDisk server
5133  *
5134  * Return Code:
5135  *	0	- Success
5136  */
5137 static int
5138 vdc_handle_dring_reg_msg(vdc_t *vdc, vio_dring_reg_msg_t *dring_msg)
5139 {
5140 	int		status = 0;
5141 
5142 	ASSERT(vdc != NULL);
5143 	ASSERT(mutex_owned(&vdc->lock));
5144 
5145 	if (dring_msg->tag.vio_subtype_env != VIO_DRING_REG) {
5146 		return (EPROTO);
5147 	}
5148 
5149 	switch (dring_msg->tag.vio_subtype) {
5150 	case VIO_SUBTYPE_ACK:
5151 		/* save the received dring_ident */
5152 		vdc->dring_ident = dring_msg->dring_ident;
5153 		DMSG(vdc, 0, "[%d] Received dring ident=0x%lx\n",
5154 		    vdc->instance, vdc->dring_ident);
5155 		break;
5156 
5157 	case VIO_SUBTYPE_NACK:
5158 		/*
5159 		 * vds could not handle the DRing info we sent so we
5160 		 * stop negotiating.
5161 		 */
5162 		DMSG(vdc, 0, "[%d] server could not register DRing\n",
5163 		    vdc->instance);
5164 		status = EPROTO;
5165 		break;
5166 
5167 	case VIO_SUBTYPE_INFO:
5168 		/*
5169 		 * Handle the case where vds starts handshake
5170 		 * (for now only vdc is the instigatior)
5171 		 */
5172 		status = ENOTSUP;
5173 		break;
5174 	default:
5175 		status = ENOTSUP;
5176 	}
5177 
5178 	return (status);
5179 }
5180 
5181 /*
5182  * Function:
5183  *	vdc_verify_seq_num()
5184  *
5185  * Description:
5186  *	This functions verifies that the sequence number sent back by the vDisk
5187  *	server with the latest message is what is expected (i.e. it is greater
5188  *	than the last seq num sent by the vDisk server and less than or equal
5189  *	to the last seq num generated by vdc).
5190  *
5191  *	It then checks the request ID to see if any requests need processing
5192  *	in the DRing.
5193  *
5194  * Arguments:
5195  *	vdc		- soft state pointer for this instance of the driver.
5196  *	dring_msg	- pointer to the LDC message sent by vds
5197  *
5198  * Return Code:
5199  *	VDC_SEQ_NUM_TODO	- Message needs to be processed
5200  *	VDC_SEQ_NUM_SKIP	- Message has already been processed
5201  *	VDC_SEQ_NUM_INVALID	- The seq numbers are so out of sync,
5202  *				  vdc cannot deal with them
5203  */
5204 static int
5205 vdc_verify_seq_num(vdc_t *vdc, vio_dring_msg_t *dring_msg)
5206 {
5207 	ASSERT(vdc != NULL);
5208 	ASSERT(dring_msg != NULL);
5209 	ASSERT(mutex_owned(&vdc->lock));
5210 
5211 	/*
5212 	 * Check to see if the messages were responded to in the correct
5213 	 * order by vds.
5214 	 */
5215 	if ((dring_msg->seq_num <= vdc->seq_num_reply) ||
5216 	    (dring_msg->seq_num > vdc->seq_num)) {
5217 		DMSG(vdc, 0, "?[%d] Bogus sequence_number %lu: "
5218 		    "%lu > expected <= %lu (last proc req %lu sent %lu)\n",
5219 		    vdc->instance, dring_msg->seq_num,
5220 		    vdc->seq_num_reply, vdc->seq_num,
5221 		    vdc->req_id_proc, vdc->req_id);
5222 		return (VDC_SEQ_NUM_INVALID);
5223 	}
5224 	vdc->seq_num_reply = dring_msg->seq_num;
5225 
5226 	if (vdc->req_id_proc < vdc->req_id)
5227 		return (VDC_SEQ_NUM_TODO);
5228 	else
5229 		return (VDC_SEQ_NUM_SKIP);
5230 }
5231 
5232 
5233 /*
5234  * Function:
5235  *	vdc_is_supported_version()
5236  *
5237  * Description:
5238  *	This routine checks if the major/minor version numbers specified in
5239  *	'ver_msg' are supported. If not it finds the next version that is
5240  *	in the supported version list 'vdc_version[]' and sets the fields in
5241  *	'ver_msg' to those values
5242  *
5243  * Arguments:
5244  *	ver_msg	- LDC message sent by vDisk server
5245  *
5246  * Return Code:
5247  *	B_TRUE	- Success
5248  *	B_FALSE	- Version not supported
5249  */
5250 static boolean_t
5251 vdc_is_supported_version(vio_ver_msg_t *ver_msg)
5252 {
5253 	int vdc_num_versions = sizeof (vdc_version) / sizeof (vdc_version[0]);
5254 
5255 	for (int i = 0; i < vdc_num_versions; i++) {
5256 		ASSERT(vdc_version[i].major > 0);
5257 		ASSERT((i == 0) ||
5258 		    (vdc_version[i].major < vdc_version[i-1].major));
5259 
5260 		/*
5261 		 * If the major versions match, adjust the minor version, if
5262 		 * necessary, down to the highest value supported by this
5263 		 * client. The server should support all minor versions lower
5264 		 * than the value it sent
5265 		 */
5266 		if (ver_msg->ver_major == vdc_version[i].major) {
5267 			if (ver_msg->ver_minor > vdc_version[i].minor) {
5268 				DMSGX(0,
5269 				    "Adjusting minor version from %u to %u",
5270 				    ver_msg->ver_minor, vdc_version[i].minor);
5271 				ver_msg->ver_minor = vdc_version[i].minor;
5272 			}
5273 			return (B_TRUE);
5274 		}
5275 
5276 		/*
5277 		 * If the message contains a higher major version number, set
5278 		 * the message's major/minor versions to the current values
5279 		 * and return false, so this message will get resent with
5280 		 * these values, and the server will potentially try again
5281 		 * with the same or a lower version
5282 		 */
5283 		if (ver_msg->ver_major > vdc_version[i].major) {
5284 			ver_msg->ver_major = vdc_version[i].major;
5285 			ver_msg->ver_minor = vdc_version[i].minor;
5286 			DMSGX(0, "Suggesting major/minor (0x%x/0x%x)\n",
5287 			    ver_msg->ver_major, ver_msg->ver_minor);
5288 
5289 			return (B_FALSE);
5290 		}
5291 
5292 		/*
5293 		 * Otherwise, the message's major version is less than the
5294 		 * current major version, so continue the loop to the next
5295 		 * (lower) supported version
5296 		 */
5297 	}
5298 
5299 	/*
5300 	 * No common version was found; "ground" the version pair in the
5301 	 * message to terminate negotiation
5302 	 */
5303 	ver_msg->ver_major = 0;
5304 	ver_msg->ver_minor = 0;
5305 
5306 	return (B_FALSE);
5307 }
5308 /* -------------------------------------------------------------------------- */
5309 
5310 /*
5311  * DKIO(7) support
5312  */
5313 
5314 typedef struct vdc_dk_arg {
5315 	struct dk_callback	dkc;
5316 	int			mode;
5317 	dev_t			dev;
5318 	vdc_t			*vdc;
5319 } vdc_dk_arg_t;
5320 
5321 /*
5322  * Function:
5323  * 	vdc_dkio_flush_cb()
5324  *
5325  * Description:
5326  *	This routine is a callback for DKIOCFLUSHWRITECACHE which can be called
5327  *	by kernel code.
5328  *
5329  * Arguments:
5330  *	arg	- a pointer to a vdc_dk_arg_t structure.
5331  */
5332 void
5333 vdc_dkio_flush_cb(void *arg)
5334 {
5335 	struct vdc_dk_arg	*dk_arg = (struct vdc_dk_arg *)arg;
5336 	struct dk_callback	*dkc = NULL;
5337 	vdc_t			*vdc = NULL;
5338 	int			rv;
5339 
5340 	if (dk_arg == NULL) {
5341 		cmn_err(CE_NOTE, "?[Unk] DKIOCFLUSHWRITECACHE arg is NULL\n");
5342 		return;
5343 	}
5344 	dkc = &dk_arg->dkc;
5345 	vdc = dk_arg->vdc;
5346 	ASSERT(vdc != NULL);
5347 
5348 	rv = vdc_do_sync_op(vdc, VD_OP_FLUSH, NULL, 0,
5349 	    VDCPART(dk_arg->dev), 0, VIO_both_dir, B_TRUE);
5350 	if (rv != 0) {
5351 		DMSG(vdc, 0, "[%d] DKIOCFLUSHWRITECACHE failed %d : model %x\n",
5352 		    vdc->instance, rv,
5353 		    ddi_model_convert_from(dk_arg->mode & FMODELS));
5354 	}
5355 
5356 	/*
5357 	 * Trigger the call back to notify the caller the the ioctl call has
5358 	 * been completed.
5359 	 */
5360 	if ((dk_arg->mode & FKIOCTL) &&
5361 	    (dkc != NULL) &&
5362 	    (dkc->dkc_callback != NULL)) {
5363 		ASSERT(dkc->dkc_cookie != NULL);
5364 		(*dkc->dkc_callback)(dkc->dkc_cookie, rv);
5365 	}
5366 
5367 	/* Indicate that one less DKIO write flush is outstanding */
5368 	mutex_enter(&vdc->lock);
5369 	vdc->dkio_flush_pending--;
5370 	ASSERT(vdc->dkio_flush_pending >= 0);
5371 	mutex_exit(&vdc->lock);
5372 
5373 	/* free the mem that was allocated when the callback was dispatched */
5374 	kmem_free(arg, sizeof (vdc_dk_arg_t));
5375 }
5376 
5377 /*
5378  * Function:
5379  * 	vdc_dkio_gapart()
5380  *
5381  * Description:
5382  *	This function implements the DKIOCGAPART ioctl.
5383  *
5384  * Arguments:
5385  *	vdc	- soft state pointer
5386  *	arg	- a pointer to a dk_map[NDKMAP] or dk_map32[NDKMAP] structure
5387  *	flag	- ioctl flags
5388  */
5389 static int
5390 vdc_dkio_gapart(vdc_t *vdc, caddr_t arg, int flag)
5391 {
5392 	struct dk_geom *geom;
5393 	struct extvtoc *vtoc;
5394 	union {
5395 		struct dk_map map[NDKMAP];
5396 		struct dk_map32 map32[NDKMAP];
5397 	} data;
5398 	int i, rv, size;
5399 
5400 	mutex_enter(&vdc->lock);
5401 
5402 	if ((rv = vdc_validate_geometry(vdc)) != 0) {
5403 		mutex_exit(&vdc->lock);
5404 		return (rv);
5405 	}
5406 
5407 	if (vdc->vdisk_size > VD_OLDVTOC_LIMIT) {
5408 		mutex_exit(&vdc->lock);
5409 		return (EOVERFLOW);
5410 	}
5411 
5412 	vtoc = vdc->vtoc;
5413 	geom = vdc->geom;
5414 
5415 	if (ddi_model_convert_from(flag & FMODELS) == DDI_MODEL_ILP32) {
5416 
5417 		for (i = 0; i < vtoc->v_nparts; i++) {
5418 			data.map32[i].dkl_cylno = vtoc->v_part[i].p_start /
5419 			    (geom->dkg_nhead * geom->dkg_nsect);
5420 			data.map32[i].dkl_nblk = vtoc->v_part[i].p_size;
5421 		}
5422 		size = NDKMAP * sizeof (struct dk_map32);
5423 
5424 	} else {
5425 
5426 		for (i = 0; i < vtoc->v_nparts; i++) {
5427 			data.map[i].dkl_cylno = vtoc->v_part[i].p_start /
5428 			    (geom->dkg_nhead * geom->dkg_nsect);
5429 			data.map[i].dkl_nblk = vtoc->v_part[i].p_size;
5430 		}
5431 		size = NDKMAP * sizeof (struct dk_map);
5432 
5433 	}
5434 
5435 	mutex_exit(&vdc->lock);
5436 
5437 	if (ddi_copyout(&data, arg, size, flag) != 0)
5438 		return (EFAULT);
5439 
5440 	return (0);
5441 }
5442 
5443 /*
5444  * Function:
5445  * 	vdc_dkio_partition()
5446  *
5447  * Description:
5448  *	This function implements the DKIOCPARTITION ioctl.
5449  *
5450  * Arguments:
5451  *	vdc	- soft state pointer
5452  *	arg	- a pointer to a struct partition64 structure
5453  *	flag	- ioctl flags
5454  */
5455 static int
5456 vdc_dkio_partition(vdc_t *vdc, caddr_t arg, int flag)
5457 {
5458 	struct partition64 p64;
5459 	efi_gpt_t *gpt;
5460 	efi_gpe_t *gpe;
5461 	vd_efi_dev_t edev;
5462 	uint_t partno;
5463 	int rv;
5464 
5465 	if (ddi_copyin(arg, &p64, sizeof (struct partition64), flag)) {
5466 		return (EFAULT);
5467 	}
5468 
5469 	VDC_EFI_DEV_SET(edev, vdc, vd_process_efi_ioctl);
5470 
5471 	if ((rv = vd_efi_alloc_and_read(&edev, &gpt, &gpe)) != 0) {
5472 		return (rv);
5473 	}
5474 
5475 	partno = p64.p_partno;
5476 
5477 	if (partno >= gpt->efi_gpt_NumberOfPartitionEntries) {
5478 		vd_efi_free(&edev, gpt, gpe);
5479 		return (ESRCH);
5480 	}
5481 
5482 	bcopy(&gpe[partno].efi_gpe_PartitionTypeGUID, &p64.p_type,
5483 	    sizeof (struct uuid));
5484 	p64.p_start = gpe[partno].efi_gpe_StartingLBA;
5485 	p64.p_size = gpe[partno].efi_gpe_EndingLBA - p64.p_start + 1;
5486 
5487 	if (ddi_copyout(&p64, arg, sizeof (struct partition64), flag)) {
5488 		vd_efi_free(&edev, gpt, gpe);
5489 		return (EFAULT);
5490 	}
5491 
5492 	vd_efi_free(&edev, gpt, gpe);
5493 	return (0);
5494 }
5495 
5496 /*
5497  * Function:
5498  * 	vdc_dioctl_rwcmd()
5499  *
5500  * Description:
5501  *	This function implements the DIOCTL_RWCMD ioctl. This ioctl is used
5502  *	for DKC_DIRECT disks to read or write at an absolute disk offset.
5503  *
5504  * Arguments:
5505  *	dev	- device
5506  *	arg	- a pointer to a dadkio_rwcmd or dadkio_rwcmd32 structure
5507  *	flag	- ioctl flags
5508  */
5509 static int
5510 vdc_dioctl_rwcmd(vdc_t *vdc, caddr_t arg, int flag)
5511 {
5512 	struct dadkio_rwcmd32 rwcmd32;
5513 	struct dadkio_rwcmd rwcmd;
5514 	struct iovec aiov;
5515 	struct uio auio;
5516 	int rw, status;
5517 	struct buf *buf;
5518 
5519 	if (ddi_model_convert_from(flag & FMODELS) == DDI_MODEL_ILP32) {
5520 		if (ddi_copyin((caddr_t)arg, (caddr_t)&rwcmd32,
5521 		    sizeof (struct dadkio_rwcmd32), flag)) {
5522 			return (EFAULT);
5523 		}
5524 		rwcmd.cmd = rwcmd32.cmd;
5525 		rwcmd.flags = rwcmd32.flags;
5526 		rwcmd.blkaddr = (daddr_t)rwcmd32.blkaddr;
5527 		rwcmd.buflen = rwcmd32.buflen;
5528 		rwcmd.bufaddr = (caddr_t)(uintptr_t)rwcmd32.bufaddr;
5529 	} else {
5530 		if (ddi_copyin((caddr_t)arg, (caddr_t)&rwcmd,
5531 		    sizeof (struct dadkio_rwcmd), flag)) {
5532 			return (EFAULT);
5533 		}
5534 	}
5535 
5536 	switch (rwcmd.cmd) {
5537 	case DADKIO_RWCMD_READ:
5538 		rw = B_READ;
5539 		break;
5540 	case DADKIO_RWCMD_WRITE:
5541 		rw = B_WRITE;
5542 		break;
5543 	default:
5544 		return (EINVAL);
5545 	}
5546 
5547 	bzero((caddr_t)&aiov, sizeof (struct iovec));
5548 	aiov.iov_base   = rwcmd.bufaddr;
5549 	aiov.iov_len    = rwcmd.buflen;
5550 
5551 	bzero((caddr_t)&auio, sizeof (struct uio));
5552 	auio.uio_iov    = &aiov;
5553 	auio.uio_iovcnt = 1;
5554 	auio.uio_loffset = rwcmd.blkaddr * vdc->vdisk_bsize;
5555 	auio.uio_resid  = rwcmd.buflen;
5556 	auio.uio_segflg = flag & FKIOCTL ? UIO_SYSSPACE : UIO_USERSPACE;
5557 
5558 	buf = kmem_alloc(sizeof (buf_t), KM_SLEEP);
5559 	bioinit(buf);
5560 	/*
5561 	 * We use the private field of buf to specify that this is an
5562 	 * I/O using an absolute offset.
5563 	 */
5564 	buf->b_private = (void *)VD_SLICE_NONE;
5565 
5566 	status = physio(vdc_strategy, buf, VD_MAKE_DEV(vdc->instance, 0),
5567 	    rw, vdc_min, &auio);
5568 
5569 	biofini(buf);
5570 	kmem_free(buf, sizeof (buf_t));
5571 
5572 	return (status);
5573 }
5574 
5575 /*
5576  * Allocate a buffer for a VD_OP_SCSICMD operation. The size of the allocated
5577  * buffer is returned in alloc_len.
5578  */
5579 static vd_scsi_t *
5580 vdc_scsi_alloc(int cdb_len, int sense_len, int datain_len, int dataout_len,
5581     int *alloc_len)
5582 {
5583 	vd_scsi_t *vd_scsi;
5584 	int vd_scsi_len = VD_SCSI_SIZE;
5585 
5586 	vd_scsi_len += P2ROUNDUP(cdb_len, sizeof (uint64_t));
5587 	vd_scsi_len += P2ROUNDUP(sense_len, sizeof (uint64_t));
5588 	vd_scsi_len += P2ROUNDUP(datain_len, sizeof (uint64_t));
5589 	vd_scsi_len += P2ROUNDUP(dataout_len, sizeof (uint64_t));
5590 
5591 	ASSERT(vd_scsi_len % sizeof (uint64_t) == 0);
5592 
5593 	vd_scsi = kmem_zalloc(vd_scsi_len, KM_SLEEP);
5594 
5595 	vd_scsi->cdb_len = cdb_len;
5596 	vd_scsi->sense_len = sense_len;
5597 	vd_scsi->datain_len = datain_len;
5598 	vd_scsi->dataout_len = dataout_len;
5599 
5600 	*alloc_len = vd_scsi_len;
5601 
5602 	return (vd_scsi);
5603 }
5604 
5605 /*
5606  * Convert the status of a SCSI command to a Solaris return code.
5607  *
5608  * Arguments:
5609  *	vd_scsi		- The SCSI operation buffer.
5610  *	log_error	- indicate if an error message should be logged.
5611  *
5612  * Note that our SCSI error messages are rather primitive for the moment
5613  * and could be improved by decoding some data like the SCSI command and
5614  * the sense key.
5615  *
5616  * Return value:
5617  *	0		- Status is good.
5618  *	EACCES		- Status reports a reservation conflict.
5619  *	ENOTSUP		- Status reports a check condition and sense key
5620  *			  reports an illegal request.
5621  *	EIO		- Any other status.
5622  */
5623 static int
5624 vdc_scsi_status(vdc_t *vdc, vd_scsi_t *vd_scsi, boolean_t log_error)
5625 {
5626 	int rv;
5627 	char path_str[MAXPATHLEN];
5628 	char panic_str[VDC_RESV_CONFLICT_FMT_LEN + MAXPATHLEN];
5629 	union scsi_cdb *cdb;
5630 	struct scsi_extended_sense *sense;
5631 
5632 	if (vd_scsi->cmd_status == STATUS_GOOD)
5633 		/* no error */
5634 		return (0);
5635 
5636 	/* when the tunable vdc_scsi_log_error is true we log all errors */
5637 	if (vdc_scsi_log_error)
5638 		log_error = B_TRUE;
5639 
5640 	if (log_error) {
5641 		cmn_err(CE_WARN, "%s (vdc%d):\tError for Command: 0x%x)\n",
5642 		    ddi_pathname(vdc->dip, path_str), vdc->instance,
5643 		    GETCMD(VD_SCSI_DATA_CDB(vd_scsi)));
5644 	}
5645 
5646 	/* default returned value */
5647 	rv = EIO;
5648 
5649 	switch (vd_scsi->cmd_status) {
5650 
5651 	case STATUS_CHECK:
5652 	case STATUS_TERMINATED:
5653 		if (log_error)
5654 			cmn_err(CE_CONT, "\tCheck Condition Error\n");
5655 
5656 		/* check sense buffer */
5657 		if (vd_scsi->sense_len == 0 ||
5658 		    vd_scsi->sense_status != STATUS_GOOD) {
5659 			if (log_error)
5660 				cmn_err(CE_CONT, "\tNo Sense Data Available\n");
5661 			break;
5662 		}
5663 
5664 		sense = VD_SCSI_DATA_SENSE(vd_scsi);
5665 
5666 		if (log_error) {
5667 			cmn_err(CE_CONT, "\tSense Key:  0x%x\n"
5668 			    "\tASC: 0x%x, ASCQ: 0x%x\n",
5669 			    scsi_sense_key((uint8_t *)sense),
5670 			    scsi_sense_asc((uint8_t *)sense),
5671 			    scsi_sense_ascq((uint8_t *)sense));
5672 		}
5673 
5674 		if (scsi_sense_key((uint8_t *)sense) == KEY_ILLEGAL_REQUEST)
5675 			rv = ENOTSUP;
5676 		break;
5677 
5678 	case STATUS_BUSY:
5679 		if (log_error)
5680 			cmn_err(CE_NOTE, "\tDevice Busy\n");
5681 		break;
5682 
5683 	case STATUS_RESERVATION_CONFLICT:
5684 		/*
5685 		 * If the command was PERSISTENT_RESERVATION_[IN|OUT] then
5686 		 * reservation conflict could be due to various reasons like
5687 		 * incorrect keys, not registered or not reserved etc. So,
5688 		 * we should not panic in that case.
5689 		 */
5690 		cdb = VD_SCSI_DATA_CDB(vd_scsi);
5691 		if (vdc->failfast_interval != 0 &&
5692 		    cdb->scc_cmd != SCMD_PERSISTENT_RESERVE_IN &&
5693 		    cdb->scc_cmd != SCMD_PERSISTENT_RESERVE_OUT) {
5694 			/* failfast is enabled so we have to panic */
5695 			(void) snprintf(panic_str, sizeof (panic_str),
5696 			    VDC_RESV_CONFLICT_FMT_STR "%s",
5697 			    ddi_pathname(vdc->dip, path_str));
5698 			panic(panic_str);
5699 		}
5700 		if (log_error)
5701 			cmn_err(CE_NOTE, "\tReservation Conflict\n");
5702 		rv = EACCES;
5703 		break;
5704 
5705 	case STATUS_QFULL:
5706 		if (log_error)
5707 			cmn_err(CE_NOTE, "\tQueue Full\n");
5708 		break;
5709 
5710 	case STATUS_MET:
5711 	case STATUS_INTERMEDIATE:
5712 	case STATUS_SCSI2:
5713 	case STATUS_INTERMEDIATE_MET:
5714 	case STATUS_ACA_ACTIVE:
5715 		if (log_error)
5716 			cmn_err(CE_CONT,
5717 			    "\tUnexpected SCSI status received: 0x%x\n",
5718 			    vd_scsi->cmd_status);
5719 		break;
5720 
5721 	default:
5722 		if (log_error)
5723 			cmn_err(CE_CONT,
5724 			    "\tInvalid SCSI status received: 0x%x\n",
5725 			    vd_scsi->cmd_status);
5726 		break;
5727 	}
5728 
5729 	return (rv);
5730 }
5731 
5732 /*
5733  * Implemented the USCSICMD uscsi(7I) ioctl. This ioctl is converted to
5734  * a VD_OP_SCSICMD operation which is sent to the vdisk server. If a SCSI
5735  * reset is requested (i.e. a flag USCSI_RESET* is set) then the ioctl is
5736  * converted to a VD_OP_RESET operation.
5737  */
5738 static int
5739 vdc_uscsi_cmd(vdc_t *vdc, caddr_t arg, int mode)
5740 {
5741 	struct uscsi_cmd 	uscsi;
5742 	struct uscsi_cmd32	uscsi32;
5743 	vd_scsi_t 		*vd_scsi;
5744 	int 			vd_scsi_len;
5745 	union scsi_cdb		*cdb;
5746 	struct scsi_extended_sense *sense;
5747 	char 			*datain, *dataout;
5748 	size_t			cdb_len, datain_len, dataout_len, sense_len;
5749 	int 			rv;
5750 
5751 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5752 		if (ddi_copyin(arg, &uscsi32, sizeof (struct uscsi_cmd32),
5753 		    mode) != 0)
5754 			return (EFAULT);
5755 		uscsi_cmd32touscsi_cmd((&uscsi32), (&uscsi));
5756 	} else {
5757 		if (ddi_copyin(arg, &uscsi, sizeof (struct uscsi_cmd),
5758 		    mode) != 0)
5759 			return (EFAULT);
5760 	}
5761 
5762 	/* a uscsi reset is converted to a VD_OP_RESET operation */
5763 	if (uscsi.uscsi_flags & (USCSI_RESET | USCSI_RESET_LUN |
5764 	    USCSI_RESET_ALL)) {
5765 		rv = vdc_do_sync_op(vdc, VD_OP_RESET, NULL, 0, 0, 0,
5766 		    VIO_both_dir, B_TRUE);
5767 		return (rv);
5768 	}
5769 
5770 	/* cdb buffer length */
5771 	cdb_len = uscsi.uscsi_cdblen;
5772 
5773 	/* data in and out buffers length */
5774 	if (uscsi.uscsi_flags & USCSI_READ) {
5775 		datain_len = uscsi.uscsi_buflen;
5776 		dataout_len = 0;
5777 	} else {
5778 		datain_len = 0;
5779 		dataout_len = uscsi.uscsi_buflen;
5780 	}
5781 
5782 	/* sense buffer length */
5783 	if (uscsi.uscsi_flags & USCSI_RQENABLE)
5784 		sense_len = uscsi.uscsi_rqlen;
5785 	else
5786 		sense_len = 0;
5787 
5788 	/* allocate buffer for the VD_SCSICMD_OP operation */
5789 	vd_scsi = vdc_scsi_alloc(cdb_len, sense_len, datain_len, dataout_len,
5790 	    &vd_scsi_len);
5791 
5792 	/*
5793 	 * The documentation of USCSI_ISOLATE and USCSI_DIAGNOSE is very vague,
5794 	 * but basically they prevent a SCSI command from being retried in case
5795 	 * of an error.
5796 	 */
5797 	if ((uscsi.uscsi_flags & USCSI_ISOLATE) ||
5798 	    (uscsi.uscsi_flags & USCSI_DIAGNOSE))
5799 		vd_scsi->options |= VD_SCSI_OPT_NORETRY;
5800 
5801 	/* set task attribute */
5802 	if (uscsi.uscsi_flags & USCSI_NOTAG) {
5803 		vd_scsi->task_attribute = 0;
5804 	} else {
5805 		if (uscsi.uscsi_flags & USCSI_HEAD)
5806 			vd_scsi->task_attribute = VD_SCSI_TASK_ACA;
5807 		else if (uscsi.uscsi_flags & USCSI_HTAG)
5808 			vd_scsi->task_attribute = VD_SCSI_TASK_HQUEUE;
5809 		else if (uscsi.uscsi_flags & USCSI_OTAG)
5810 			vd_scsi->task_attribute = VD_SCSI_TASK_ORDERED;
5811 		else
5812 			vd_scsi->task_attribute = 0;
5813 	}
5814 
5815 	/* set timeout */
5816 	vd_scsi->timeout = uscsi.uscsi_timeout;
5817 
5818 	/* copy-in cdb data */
5819 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
5820 	if (ddi_copyin(uscsi.uscsi_cdb, cdb, cdb_len, mode) != 0) {
5821 		rv = EFAULT;
5822 		goto done;
5823 	}
5824 
5825 	/* keep a pointer to the sense buffer */
5826 	sense = VD_SCSI_DATA_SENSE(vd_scsi);
5827 
5828 	/* keep a pointer to the data-in buffer */
5829 	datain = (char *)VD_SCSI_DATA_IN(vd_scsi);
5830 
5831 	/* copy-in request data to the data-out buffer */
5832 	dataout = (char *)VD_SCSI_DATA_OUT(vd_scsi);
5833 	if (!(uscsi.uscsi_flags & USCSI_READ)) {
5834 		if (ddi_copyin(uscsi.uscsi_bufaddr, dataout, dataout_len,
5835 		    mode)) {
5836 			rv = EFAULT;
5837 			goto done;
5838 		}
5839 	}
5840 
5841 	/* submit the request */
5842 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
5843 	    0, 0, VIO_both_dir, B_FALSE);
5844 
5845 	if (rv != 0)
5846 		goto done;
5847 
5848 	/* update scsi status */
5849 	uscsi.uscsi_status = vd_scsi->cmd_status;
5850 
5851 	/* update sense data */
5852 	if ((uscsi.uscsi_flags & USCSI_RQENABLE) &&
5853 	    (uscsi.uscsi_status == STATUS_CHECK ||
5854 	    uscsi.uscsi_status == STATUS_TERMINATED)) {
5855 
5856 		uscsi.uscsi_rqstatus = vd_scsi->sense_status;
5857 
5858 		if (uscsi.uscsi_rqstatus == STATUS_GOOD) {
5859 			uscsi.uscsi_rqresid = uscsi.uscsi_rqlen -
5860 			    vd_scsi->sense_len;
5861 			if (ddi_copyout(sense, uscsi.uscsi_rqbuf,
5862 			    vd_scsi->sense_len, mode) != 0) {
5863 				rv = EFAULT;
5864 				goto done;
5865 			}
5866 		}
5867 	}
5868 
5869 	/* update request data */
5870 	if (uscsi.uscsi_status == STATUS_GOOD) {
5871 		if (uscsi.uscsi_flags & USCSI_READ) {
5872 			uscsi.uscsi_resid = uscsi.uscsi_buflen -
5873 			    vd_scsi->datain_len;
5874 			if (ddi_copyout(datain, uscsi.uscsi_bufaddr,
5875 			    vd_scsi->datain_len, mode) != 0) {
5876 				rv = EFAULT;
5877 				goto done;
5878 			}
5879 		} else {
5880 			uscsi.uscsi_resid = uscsi.uscsi_buflen -
5881 			    vd_scsi->dataout_len;
5882 		}
5883 	}
5884 
5885 	/* copy-out result */
5886 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5887 		uscsi_cmdtouscsi_cmd32((&uscsi), (&uscsi32));
5888 		if (ddi_copyout(&uscsi32, arg, sizeof (struct uscsi_cmd32),
5889 		    mode) != 0) {
5890 			rv = EFAULT;
5891 			goto done;
5892 		}
5893 	} else {
5894 		if (ddi_copyout(&uscsi, arg, sizeof (struct uscsi_cmd),
5895 		    mode) != 0) {
5896 			rv = EFAULT;
5897 			goto done;
5898 		}
5899 	}
5900 
5901 	/* get the return code from the SCSI command status */
5902 	rv = vdc_scsi_status(vdc, vd_scsi,
5903 	    !(uscsi.uscsi_flags & USCSI_SILENT));
5904 
5905 done:
5906 	kmem_free(vd_scsi, vd_scsi_len);
5907 	return (rv);
5908 }
5909 
5910 /*
5911  * Create a VD_OP_SCSICMD buffer for a SCSI PERSISTENT IN command.
5912  *
5913  * Arguments:
5914  *	cmd		- SCSI PERSISTENT IN command
5915  *	len		- length of the SCSI input buffer
5916  *	vd_scsi_len	- return the length of the allocated buffer
5917  *
5918  * Returned Value:
5919  *	a pointer to the allocated VD_OP_SCSICMD buffer.
5920  */
5921 static vd_scsi_t *
5922 vdc_scsi_alloc_persistent_in(uchar_t cmd, int len, int *vd_scsi_len)
5923 {
5924 	int cdb_len, sense_len, datain_len, dataout_len;
5925 	vd_scsi_t *vd_scsi;
5926 	union scsi_cdb *cdb;
5927 
5928 	cdb_len = CDB_GROUP1;
5929 	sense_len = sizeof (struct scsi_extended_sense);
5930 	datain_len = len;
5931 	dataout_len = 0;
5932 
5933 	vd_scsi = vdc_scsi_alloc(cdb_len, sense_len, datain_len, dataout_len,
5934 	    vd_scsi_len);
5935 
5936 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
5937 
5938 	/* set cdb */
5939 	cdb->scc_cmd = SCMD_PERSISTENT_RESERVE_IN;
5940 	cdb->cdb_opaque[1] = cmd;
5941 	FORMG1COUNT(cdb, datain_len);
5942 
5943 	vd_scsi->timeout = vdc_scsi_timeout;
5944 
5945 	return (vd_scsi);
5946 }
5947 
5948 /*
5949  * Create a VD_OP_SCSICMD buffer for a SCSI PERSISTENT OUT command.
5950  *
5951  * Arguments:
5952  *	cmd		- SCSI PERSISTENT OUT command
5953  *	len		- length of the SCSI output buffer
5954  *	vd_scsi_len	- return the length of the allocated buffer
5955  *
5956  * Returned Code:
5957  *	a pointer to the allocated VD_OP_SCSICMD buffer.
5958  */
5959 static vd_scsi_t *
5960 vdc_scsi_alloc_persistent_out(uchar_t cmd, int len, int *vd_scsi_len)
5961 {
5962 	int cdb_len, sense_len, datain_len, dataout_len;
5963 	vd_scsi_t *vd_scsi;
5964 	union scsi_cdb *cdb;
5965 
5966 	cdb_len = CDB_GROUP1;
5967 	sense_len = sizeof (struct scsi_extended_sense);
5968 	datain_len = 0;
5969 	dataout_len = len;
5970 
5971 	vd_scsi = vdc_scsi_alloc(cdb_len, sense_len, datain_len, dataout_len,
5972 	    vd_scsi_len);
5973 
5974 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
5975 
5976 	/* set cdb */
5977 	cdb->scc_cmd = SCMD_PERSISTENT_RESERVE_OUT;
5978 	cdb->cdb_opaque[1] = cmd;
5979 	FORMG1COUNT(cdb, dataout_len);
5980 
5981 	vd_scsi->timeout = vdc_scsi_timeout;
5982 
5983 	return (vd_scsi);
5984 }
5985 
5986 /*
5987  * Implement the MHIOCGRP_INKEYS mhd(7i) ioctl. The ioctl is converted
5988  * to a SCSI PERSISTENT IN READ KEYS command which is sent to the vdisk
5989  * server with a VD_OP_SCSICMD operation.
5990  */
5991 static int
5992 vdc_mhd_inkeys(vdc_t *vdc, caddr_t arg, int mode)
5993 {
5994 	vd_scsi_t *vd_scsi;
5995 	mhioc_inkeys_t inkeys;
5996 	mhioc_key_list_t klist;
5997 	struct mhioc_inkeys32 inkeys32;
5998 	struct mhioc_key_list32 klist32;
5999 	sd_prin_readkeys_t *scsi_keys;
6000 	void *user_keys;
6001 	int vd_scsi_len;
6002 	int listsize, listlen, rv;
6003 
6004 	/* copyin arguments */
6005 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
6006 		rv = ddi_copyin(arg, &inkeys32, sizeof (inkeys32), mode);
6007 		if (rv != 0)
6008 			return (EFAULT);
6009 
6010 		rv = ddi_copyin((caddr_t)(uintptr_t)inkeys32.li, &klist32,
6011 		    sizeof (klist32), mode);
6012 		if (rv != 0)
6013 			return (EFAULT);
6014 
6015 		listsize = klist32.listsize;
6016 	} else {
6017 		rv = ddi_copyin(arg, &inkeys, sizeof (inkeys), mode);
6018 		if (rv != 0)
6019 			return (EFAULT);
6020 
6021 		rv = ddi_copyin(inkeys.li, &klist, sizeof (klist), mode);
6022 		if (rv != 0)
6023 			return (EFAULT);
6024 
6025 		listsize = klist.listsize;
6026 	}
6027 
6028 	/* build SCSI VD_OP request */
6029 	vd_scsi = vdc_scsi_alloc_persistent_in(SD_READ_KEYS,
6030 	    sizeof (sd_prin_readkeys_t) - sizeof (caddr_t) +
6031 	    (sizeof (mhioc_resv_key_t) * listsize), &vd_scsi_len);
6032 
6033 	scsi_keys = (sd_prin_readkeys_t *)VD_SCSI_DATA_IN(vd_scsi);
6034 
6035 	/* submit the request */
6036 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
6037 	    0, 0, VIO_both_dir, B_FALSE);
6038 
6039 	if (rv != 0)
6040 		goto done;
6041 
6042 	listlen = scsi_keys->len / MHIOC_RESV_KEY_SIZE;
6043 
6044 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
6045 		inkeys32.generation = scsi_keys->generation;
6046 		rv = ddi_copyout(&inkeys32, arg, sizeof (inkeys32), mode);
6047 		if (rv != 0) {
6048 			rv = EFAULT;
6049 			goto done;
6050 		}
6051 
6052 		klist32.listlen = listlen;
6053 		rv = ddi_copyout(&klist32, (caddr_t)(uintptr_t)inkeys32.li,
6054 		    sizeof (klist32), mode);
6055 		if (rv != 0) {
6056 			rv = EFAULT;
6057 			goto done;
6058 		}
6059 
6060 		user_keys = (caddr_t)(uintptr_t)klist32.list;
6061 	} else {
6062 		inkeys.generation = scsi_keys->generation;
6063 		rv = ddi_copyout(&inkeys, arg, sizeof (inkeys), mode);
6064 		if (rv != 0) {
6065 			rv = EFAULT;
6066 			goto done;
6067 		}
6068 
6069 		klist.listlen = listlen;
6070 		rv = ddi_copyout(&klist, inkeys.li, sizeof (klist), mode);
6071 		if (rv != 0) {
6072 			rv = EFAULT;
6073 			goto done;
6074 		}
6075 
6076 		user_keys = klist.list;
6077 	}
6078 
6079 	/* copy out keys */
6080 	if (listlen > 0 && listsize > 0) {
6081 		if (listsize < listlen)
6082 			listlen = listsize;
6083 		rv = ddi_copyout(&scsi_keys->keylist, user_keys,
6084 		    listlen * MHIOC_RESV_KEY_SIZE, mode);
6085 		if (rv != 0)
6086 			rv = EFAULT;
6087 	}
6088 
6089 	if (rv == 0)
6090 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
6091 
6092 done:
6093 	kmem_free(vd_scsi, vd_scsi_len);
6094 
6095 	return (rv);
6096 }
6097 
6098 /*
6099  * Implement the MHIOCGRP_INRESV mhd(7i) ioctl. The ioctl is converted
6100  * to a SCSI PERSISTENT IN READ RESERVATION command which is sent to
6101  * the vdisk server with a VD_OP_SCSICMD operation.
6102  */
6103 static int
6104 vdc_mhd_inresv(vdc_t *vdc, caddr_t arg, int mode)
6105 {
6106 	vd_scsi_t *vd_scsi;
6107 	mhioc_inresvs_t inresv;
6108 	mhioc_resv_desc_list_t rlist;
6109 	struct mhioc_inresvs32 inresv32;
6110 	struct mhioc_resv_desc_list32 rlist32;
6111 	mhioc_resv_desc_t mhd_resv;
6112 	sd_prin_readresv_t *scsi_resv;
6113 	sd_readresv_desc_t *resv;
6114 	mhioc_resv_desc_t *user_resv;
6115 	int vd_scsi_len;
6116 	int listsize, listlen, i, rv;
6117 
6118 	/* copyin arguments */
6119 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
6120 		rv = ddi_copyin(arg, &inresv32, sizeof (inresv32), mode);
6121 		if (rv != 0)
6122 			return (EFAULT);
6123 
6124 		rv = ddi_copyin((caddr_t)(uintptr_t)inresv32.li, &rlist32,
6125 		    sizeof (rlist32), mode);
6126 		if (rv != 0)
6127 			return (EFAULT);
6128 
6129 		listsize = rlist32.listsize;
6130 	} else {
6131 		rv = ddi_copyin(arg, &inresv, sizeof (inresv), mode);
6132 		if (rv != 0)
6133 			return (EFAULT);
6134 
6135 		rv = ddi_copyin(inresv.li, &rlist, sizeof (rlist), mode);
6136 		if (rv != 0)
6137 			return (EFAULT);
6138 
6139 		listsize = rlist.listsize;
6140 	}
6141 
6142 	/* build SCSI VD_OP request */
6143 	vd_scsi = vdc_scsi_alloc_persistent_in(SD_READ_RESV,
6144 	    sizeof (sd_prin_readresv_t) - sizeof (caddr_t) +
6145 	    (SCSI3_RESV_DESC_LEN * listsize), &vd_scsi_len);
6146 
6147 	scsi_resv = (sd_prin_readresv_t *)VD_SCSI_DATA_IN(vd_scsi);
6148 
6149 	/* submit the request */
6150 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
6151 	    0, 0, VIO_both_dir, B_FALSE);
6152 
6153 	if (rv != 0)
6154 		goto done;
6155 
6156 	listlen = scsi_resv->len / SCSI3_RESV_DESC_LEN;
6157 
6158 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
6159 		inresv32.generation = scsi_resv->generation;
6160 		rv = ddi_copyout(&inresv32, arg, sizeof (inresv32), mode);
6161 		if (rv != 0) {
6162 			rv = EFAULT;
6163 			goto done;
6164 		}
6165 
6166 		rlist32.listlen = listlen;
6167 		rv = ddi_copyout(&rlist32, (caddr_t)(uintptr_t)inresv32.li,
6168 		    sizeof (rlist32), mode);
6169 		if (rv != 0) {
6170 			rv = EFAULT;
6171 			goto done;
6172 		}
6173 
6174 		user_resv = (mhioc_resv_desc_t *)(uintptr_t)rlist32.list;
6175 	} else {
6176 		inresv.generation = scsi_resv->generation;
6177 		rv = ddi_copyout(&inresv, arg, sizeof (inresv), mode);
6178 		if (rv != 0) {
6179 			rv = EFAULT;
6180 			goto done;
6181 		}
6182 
6183 		rlist.listlen = listlen;
6184 		rv = ddi_copyout(&rlist, inresv.li, sizeof (rlist), mode);
6185 		if (rv != 0) {
6186 			rv = EFAULT;
6187 			goto done;
6188 		}
6189 
6190 		user_resv = rlist.list;
6191 	}
6192 
6193 	/* copy out reservations */
6194 	if (listsize > 0 && listlen > 0) {
6195 		if (listsize < listlen)
6196 			listlen = listsize;
6197 		resv = (sd_readresv_desc_t *)&scsi_resv->readresv_desc;
6198 
6199 		for (i = 0; i < listlen; i++) {
6200 			mhd_resv.type = resv->type;
6201 			mhd_resv.scope = resv->scope;
6202 			mhd_resv.scope_specific_addr =
6203 			    BE_32(resv->scope_specific_addr);
6204 			bcopy(&resv->resvkey, &mhd_resv.key,
6205 			    MHIOC_RESV_KEY_SIZE);
6206 
6207 			rv = ddi_copyout(&mhd_resv, user_resv,
6208 			    sizeof (mhd_resv), mode);
6209 			if (rv != 0) {
6210 				rv = EFAULT;
6211 				goto done;
6212 			}
6213 			resv++;
6214 			user_resv++;
6215 		}
6216 	}
6217 
6218 	if (rv == 0)
6219 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
6220 
6221 done:
6222 	kmem_free(vd_scsi, vd_scsi_len);
6223 	return (rv);
6224 }
6225 
6226 /*
6227  * Implement the MHIOCGRP_REGISTER mhd(7i) ioctl. The ioctl is converted
6228  * to a SCSI PERSISTENT OUT REGISTER command which is sent to the vdisk
6229  * server with a VD_OP_SCSICMD operation.
6230  */
6231 static int
6232 vdc_mhd_register(vdc_t *vdc, caddr_t arg, int mode)
6233 {
6234 	vd_scsi_t *vd_scsi;
6235 	sd_prout_t *scsi_prout;
6236 	mhioc_register_t mhd_reg;
6237 	int vd_scsi_len, rv;
6238 
6239 	/* copyin arguments */
6240 	rv = ddi_copyin(arg, &mhd_reg, sizeof (mhd_reg), mode);
6241 	if (rv != 0)
6242 		return (EFAULT);
6243 
6244 	/* build SCSI VD_OP request */
6245 	vd_scsi = vdc_scsi_alloc_persistent_out(SD_SCSI3_REGISTER,
6246 	    sizeof (sd_prout_t), &vd_scsi_len);
6247 
6248 	/* set parameters */
6249 	scsi_prout = (sd_prout_t *)VD_SCSI_DATA_OUT(vd_scsi);
6250 	bcopy(mhd_reg.oldkey.key, scsi_prout->res_key, MHIOC_RESV_KEY_SIZE);
6251 	bcopy(mhd_reg.newkey.key, scsi_prout->service_key, MHIOC_RESV_KEY_SIZE);
6252 	scsi_prout->aptpl = (uchar_t)mhd_reg.aptpl;
6253 
6254 	/* submit the request */
6255 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
6256 	    0, 0, VIO_both_dir, B_FALSE);
6257 
6258 	if (rv == 0)
6259 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
6260 
6261 	kmem_free(vd_scsi, vd_scsi_len);
6262 	return (rv);
6263 }
6264 
6265 /*
6266  * Implement the MHIOCGRP_RESERVE mhd(7i) ioctl. The ioctl is converted
6267  * to a SCSI PERSISTENT OUT RESERVE command which is sent to the vdisk
6268  * server with a VD_OP_SCSICMD operation.
6269  */
6270 static int
6271 vdc_mhd_reserve(vdc_t *vdc, caddr_t arg, int mode)
6272 {
6273 	union scsi_cdb *cdb;
6274 	vd_scsi_t *vd_scsi;
6275 	sd_prout_t *scsi_prout;
6276 	mhioc_resv_desc_t mhd_resv;
6277 	int vd_scsi_len, rv;
6278 
6279 	/* copyin arguments */
6280 	rv = ddi_copyin(arg, &mhd_resv, sizeof (mhd_resv), mode);
6281 	if (rv != 0)
6282 		return (EFAULT);
6283 
6284 	/* build SCSI VD_OP request */
6285 	vd_scsi = vdc_scsi_alloc_persistent_out(SD_SCSI3_RESERVE,
6286 	    sizeof (sd_prout_t), &vd_scsi_len);
6287 
6288 	/* set parameters */
6289 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
6290 	scsi_prout = (sd_prout_t *)VD_SCSI_DATA_OUT(vd_scsi);
6291 	bcopy(mhd_resv.key.key, scsi_prout->res_key, MHIOC_RESV_KEY_SIZE);
6292 	scsi_prout->scope_address = mhd_resv.scope_specific_addr;
6293 	cdb->cdb_opaque[2] = mhd_resv.type;
6294 
6295 	/* submit the request */
6296 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
6297 	    0, 0, VIO_both_dir, B_FALSE);
6298 
6299 	if (rv == 0)
6300 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
6301 
6302 	kmem_free(vd_scsi, vd_scsi_len);
6303 	return (rv);
6304 }
6305 
6306 /*
6307  * Implement the MHIOCGRP_PREEMPTANDABORT mhd(7i) ioctl. The ioctl is
6308  * converted to a SCSI PERSISTENT OUT PREEMPT AND ABORT command which
6309  * is sent to the vdisk server with a VD_OP_SCSICMD operation.
6310  */
6311 static int
6312 vdc_mhd_preemptabort(vdc_t *vdc, caddr_t arg, int mode)
6313 {
6314 	union scsi_cdb *cdb;
6315 	vd_scsi_t *vd_scsi;
6316 	sd_prout_t *scsi_prout;
6317 	mhioc_preemptandabort_t mhd_preempt;
6318 	int vd_scsi_len, rv;
6319 
6320 	/* copyin arguments */
6321 	rv = ddi_copyin(arg, &mhd_preempt, sizeof (mhd_preempt), mode);
6322 	if (rv != 0)
6323 		return (EFAULT);
6324 
6325 	/* build SCSI VD_OP request */
6326 	vd_scsi = vdc_scsi_alloc_persistent_out(SD_SCSI3_PREEMPTANDABORT,
6327 	    sizeof (sd_prout_t), &vd_scsi_len);
6328 
6329 	/* set parameters */
6330 	vd_scsi->task_attribute = VD_SCSI_TASK_ACA;
6331 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
6332 	scsi_prout = (sd_prout_t *)VD_SCSI_DATA_OUT(vd_scsi);
6333 	bcopy(mhd_preempt.resvdesc.key.key, scsi_prout->res_key,
6334 	    MHIOC_RESV_KEY_SIZE);
6335 	bcopy(mhd_preempt.victim_key.key, scsi_prout->service_key,
6336 	    MHIOC_RESV_KEY_SIZE);
6337 	scsi_prout->scope_address = mhd_preempt.resvdesc.scope_specific_addr;
6338 	cdb->cdb_opaque[2] = mhd_preempt.resvdesc.type;
6339 
6340 	/* submit the request */
6341 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
6342 	    0, 0, VIO_both_dir, B_FALSE);
6343 
6344 	if (rv == 0)
6345 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
6346 
6347 	kmem_free(vd_scsi, vd_scsi_len);
6348 	return (rv);
6349 }
6350 
6351 /*
6352  * Implement the MHIOCGRP_REGISTERANDIGNOREKEY mhd(7i) ioctl. The ioctl
6353  * is converted to a SCSI PERSISTENT OUT REGISTER AND IGNORE EXISTING KEY
6354  * command which is sent to the vdisk server with a VD_OP_SCSICMD operation.
6355  */
6356 static int
6357 vdc_mhd_registerignore(vdc_t *vdc, caddr_t arg, int mode)
6358 {
6359 	vd_scsi_t *vd_scsi;
6360 	sd_prout_t *scsi_prout;
6361 	mhioc_registerandignorekey_t mhd_regi;
6362 	int vd_scsi_len, rv;
6363 
6364 	/* copyin arguments */
6365 	rv = ddi_copyin(arg, &mhd_regi, sizeof (mhd_regi), mode);
6366 	if (rv != 0)
6367 		return (EFAULT);
6368 
6369 	/* build SCSI VD_OP request */
6370 	vd_scsi = vdc_scsi_alloc_persistent_out(SD_SCSI3_REGISTERANDIGNOREKEY,
6371 	    sizeof (sd_prout_t), &vd_scsi_len);
6372 
6373 	/* set parameters */
6374 	scsi_prout = (sd_prout_t *)VD_SCSI_DATA_OUT(vd_scsi);
6375 	bcopy(mhd_regi.newkey.key, scsi_prout->service_key,
6376 	    MHIOC_RESV_KEY_SIZE);
6377 	scsi_prout->aptpl = (uchar_t)mhd_regi.aptpl;
6378 
6379 	/* submit the request */
6380 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
6381 	    0, 0, VIO_both_dir, B_FALSE);
6382 
6383 	if (rv == 0)
6384 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
6385 
6386 	kmem_free(vd_scsi, vd_scsi_len);
6387 	return (rv);
6388 }
6389 
6390 /*
6391  * This function is used to send a (simple) SCSI command and check errors.
6392  */
6393 static int
6394 vdc_eio_scsi_cmd(vdc_t *vdc, uchar_t scmd, int flags)
6395 {
6396 	int cdb_len, sense_len, vd_scsi_len;
6397 	vd_scsi_t *vd_scsi;
6398 	union scsi_cdb *cdb;
6399 	int rv;
6400 
6401 	ASSERT(scmd == SCMD_TEST_UNIT_READY || scmd == SCMD_WRITE_G1);
6402 
6403 	if (scmd == SCMD_WRITE_G1)
6404 		cdb_len = CDB_GROUP1;
6405 	else
6406 		cdb_len = CDB_GROUP0;
6407 
6408 	sense_len = sizeof (struct scsi_extended_sense);
6409 
6410 	vd_scsi = vdc_scsi_alloc(cdb_len, sense_len, 0, 0, &vd_scsi_len);
6411 
6412 	/* set cdb */
6413 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
6414 	cdb->scc_cmd = scmd;
6415 
6416 	vd_scsi->timeout = vdc_scsi_timeout;
6417 
6418 	/*
6419 	 * Submit the request. Note the operation should not request that any
6420 	 * error is checked because this function is precisely called when
6421 	 * checking errors.
6422 	 */
6423 	ASSERT((flags & VDC_OP_ERRCHK) == 0);
6424 
6425 	rv = vdc_do_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
6426 	    0, 0, NULL, VIO_both_dir, flags);
6427 
6428 	if (rv == 0)
6429 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
6430 
6431 	kmem_free(vd_scsi, vd_scsi_len);
6432 	return (rv);
6433 }
6434 
6435 /*
6436  * This function is used to check if a SCSI backend is accessible. It will
6437  * also detect reservation conflict if failfast is enabled, and panic the
6438  * system in that case.
6439  *
6440  * Returned Code:
6441  *	0	- disk is accessible
6442  *	!= 0	- disk is inaccessible or unable to check if disk is accessible
6443  */
6444 static int
6445 vdc_eio_scsi_check(vdc_t *vdc, int flags)
6446 {
6447 	int failure = 0;
6448 	int rv;
6449 
6450 	/*
6451 	 * Send a TEST UNIT READY command. The command will panic
6452 	 * the system if it fails with a reservation conflict and
6453 	 * failfast is enabled. If there is a reservation conflict
6454 	 * and failfast is not enabled then the function will return
6455 	 * EACCES. In that case, there's no problem with accessing
6456 	 * the backend, it is just reserved.
6457 	 */
6458 	rv = vdc_eio_scsi_cmd(vdc, SCMD_TEST_UNIT_READY, flags);
6459 	if (rv != 0 && rv != EACCES)
6460 		failure++;
6461 
6462 	/* we don't need to do more checking if failfast is not enabled */
6463 	if (vdc->failfast_interval == 0)
6464 		return (failure);
6465 
6466 	/*
6467 	 * With SPC-3 compliant devices TEST UNIT READY will succeed on
6468 	 * a reserved device, so we also do a WRITE(10) of zero byte in
6469 	 * order to provoke a Reservation Conflict status on those newer
6470 	 * devices.
6471 	 */
6472 	if (vdc_eio_scsi_cmd(vdc, SCMD_WRITE_G1, flags) != 0)
6473 		failure++;
6474 
6475 	return (failure);
6476 }
6477 
6478 /*
6479  * This function is used to check if a backend is effectively accessible.
6480  *
6481  * Returned Code:
6482  *	0	- disk is accessible
6483  *	!= 0	- disk is inaccessible or unable to check if disk is accessible
6484  */
6485 static int
6486 vdc_eio_check(vdc_t *vdc, int flags)
6487 {
6488 	char *buffer;
6489 	diskaddr_t blkno;
6490 	int rv;
6491 
6492 	ASSERT((flags & VDC_OP_ERRCHK) == 0);
6493 
6494 	flags |= VDC_OP_DRING_RESERVED;
6495 
6496 	if (VD_OP_SUPPORTED(vdc->operations, VD_OP_SCSICMD))
6497 		return (vdc_eio_scsi_check(vdc, flags));
6498 
6499 	ASSERT(vdc->failfast_interval == 0);
6500 
6501 	/*
6502 	 * If the backend does not support SCSI operations then we simply
6503 	 * check if the backend is accessible by reading some data blocks.
6504 	 * We first try to read a random block, to try to avoid getting
6505 	 * a block that might have been cached on the service domain. Then
6506 	 * we try the last block, and finally the first block.
6507 	 *
6508 	 * We return success as soon as we are able to read any block.
6509 	 */
6510 	buffer = kmem_alloc(vdc->vdisk_bsize, KM_SLEEP);
6511 
6512 	if (vdc->vdisk_size > 0) {
6513 
6514 		/* try a random block */
6515 		(void) random_get_pseudo_bytes((uint8_t *)&blkno,
6516 		    sizeof (diskaddr_t));
6517 		blkno = blkno % vdc->vdisk_size;
6518 		rv = vdc_do_op(vdc, VD_OP_BREAD, (caddr_t)buffer,
6519 		    vdc->vdisk_bsize, VD_SLICE_NONE, blkno, NULL,
6520 		    VIO_read_dir, flags);
6521 
6522 		if (rv == 0)
6523 			goto done;
6524 
6525 		/* try the last block */
6526 		blkno = vdc->vdisk_size - 1;
6527 		rv = vdc_do_op(vdc, VD_OP_BREAD, (caddr_t)buffer,
6528 		    vdc->vdisk_bsize, VD_SLICE_NONE, blkno, NULL,
6529 		    VIO_read_dir, flags);
6530 
6531 		if (rv == 0)
6532 			goto done;
6533 	}
6534 
6535 	/* try block 0 */
6536 	blkno = 0;
6537 	rv = vdc_do_op(vdc, VD_OP_BREAD, (caddr_t)buffer, vdc->vdisk_bsize,
6538 	    VD_SLICE_NONE, blkno, NULL, VIO_read_dir, flags);
6539 
6540 done:
6541 	kmem_free(buffer, vdc->vdisk_bsize);
6542 	return (rv);
6543 }
6544 
6545 /*
6546  * Add a pending I/O to the eio queue. An I/O is added to this queue
6547  * when it has failed and failfast is enabled or the vdisk has multiple
6548  * servers. It will then be handled by the eio thread (vdc_eio_thread).
6549  * The eio queue is ordered starting with the most recent I/O added.
6550  */
6551 static vdc_io_t *
6552 vdc_eio_queue(vdc_t *vdc, int index)
6553 {
6554 	vdc_io_t *vio;
6555 
6556 	ASSERT(MUTEX_HELD(&vdc->lock));
6557 
6558 	vio = kmem_alloc(sizeof (vdc_io_t), KM_SLEEP);
6559 	vio->vio_next = vdc->eio_queue;
6560 	vio->vio_index = index;
6561 	vio->vio_qtime = ddi_get_lbolt();
6562 
6563 	vdc->eio_queue = vio;
6564 
6565 	/* notify the eio thread that a new I/O is queued */
6566 	cv_signal(&vdc->eio_cv);
6567 
6568 	return (vio);
6569 }
6570 
6571 /*
6572  * Remove I/Os added before the indicated deadline from the eio queue. A
6573  * deadline of 0 means that all I/Os have to be unqueued. The complete_io
6574  * boolean specifies if unqueued I/Os should be marked as completed or not.
6575  */
6576 static void
6577 vdc_eio_unqueue(vdc_t *vdc, clock_t deadline, boolean_t complete_io)
6578 {
6579 	struct buf *buf;
6580 	vdc_io_t *vio, *vio_tmp;
6581 	int index, op;
6582 
6583 	ASSERT(MUTEX_HELD(&vdc->lock));
6584 
6585 	vio_tmp = NULL;
6586 	vio = vdc->eio_queue;
6587 
6588 	if (deadline != 0) {
6589 		/*
6590 		 * Skip any io queued after the deadline. The eio queue is
6591 		 * ordered starting with the last I/O added to the queue.
6592 		 */
6593 		while (vio != NULL && vio->vio_qtime > deadline) {
6594 			vio_tmp = vio;
6595 			vio = vio->vio_next;
6596 		}
6597 	}
6598 
6599 	if (vio == NULL)
6600 		/* nothing to unqueue */
6601 		return;
6602 
6603 	/* update the queue */
6604 	if (vio_tmp == NULL)
6605 		vdc->eio_queue = NULL;
6606 	else
6607 		vio_tmp->vio_next = NULL;
6608 
6609 	/*
6610 	 * Free and complete unqueued I/Os if this was requested. All I/Os
6611 	 * have a block I/O data transfer structure (buf) and they are
6612 	 * completed by calling biodone().
6613 	 */
6614 	while (vio != NULL) {
6615 		vio_tmp = vio->vio_next;
6616 
6617 		if (complete_io) {
6618 			index = vio->vio_index;
6619 			op = vdc->local_dring[index].operation;
6620 			buf = vdc->local_dring[index].buf;
6621 			(void) vdc_depopulate_descriptor(vdc, index);
6622 			ASSERT(buf->b_flags & B_ERROR);
6623 			if (op == VD_OP_BREAD || op == VD_OP_BWRITE) {
6624 				VD_UPDATE_ERR_STATS(vdc, vd_softerrs);
6625 				VD_KSTAT_RUNQ_EXIT(vdc);
6626 				DTRACE_IO1(done, buf_t *, buf);
6627 			}
6628 			biodone(buf);
6629 		}
6630 
6631 		kmem_free(vio, sizeof (vdc_io_t));
6632 		vio = vio_tmp;
6633 	}
6634 }
6635 
6636 /*
6637  * Error I/O Thread.  There is one eio thread for each virtual disk that
6638  * has multiple servers or for which failfast is enabled. Failfast can only
6639  * be enabled for vdisk supporting SCSI commands.
6640  *
6641  * While failfast is enabled, the eio thread sends a TEST UNIT READY
6642  * and a zero size WRITE(10) SCSI commands on a regular basis to check that
6643  * we still have access to the disk. If a command fails with a RESERVATION
6644  * CONFLICT error then the system will immediatly panic.
6645  *
6646  * The eio thread is also woken up when an I/O has failed. It then checks
6647  * the access to the disk to ensure that the I/O failure was not due to a
6648  * reservation conflict or to the backend been inaccessible.
6649  *
6650  */
6651 static void
6652 vdc_eio_thread(void *arg)
6653 {
6654 	int status;
6655 	vdc_t *vdc = (vdc_t *)arg;
6656 	clock_t starttime, timeout = drv_usectohz(vdc->failfast_interval);
6657 
6658 	mutex_enter(&vdc->lock);
6659 
6660 	while (vdc->failfast_interval != 0 || vdc->num_servers > 1) {
6661 		/*
6662 		 * Wait if there is nothing in the eio queue or if the state
6663 		 * is not VDC_STATE_RUNNING.
6664 		 */
6665 		if (vdc->eio_queue == NULL || vdc->state != VDC_STATE_RUNNING) {
6666 			if (vdc->failfast_interval != 0) {
6667 				timeout = ddi_get_lbolt() +
6668 				    drv_usectohz(vdc->failfast_interval);
6669 				(void) cv_timedwait(&vdc->eio_cv, &vdc->lock,
6670 				    timeout);
6671 			} else {
6672 				ASSERT(vdc->num_servers > 1);
6673 				(void) cv_wait(&vdc->eio_cv, &vdc->lock);
6674 			}
6675 
6676 			if (vdc->state != VDC_STATE_RUNNING)
6677 				continue;
6678 		}
6679 
6680 		mutex_exit(&vdc->lock);
6681 
6682 		starttime = ddi_get_lbolt();
6683 
6684 		/* check error */
6685 		status = vdc_eio_check(vdc, VDC_OP_STATE_RUNNING);
6686 
6687 		mutex_enter(&vdc->lock);
6688 		/*
6689 		 * We have dropped the lock to check the backend so we have
6690 		 * to check that the eio thread is still enabled.
6691 		 */
6692 		if (vdc->failfast_interval == 0 && vdc->num_servers <= 1)
6693 			break;
6694 
6695 		/*
6696 		 * If the eio queue is empty or we are not in running state
6697 		 * anymore then there is nothing to do.
6698 		 */
6699 		if (vdc->state != VDC_STATE_RUNNING || vdc->eio_queue == NULL)
6700 			continue;
6701 
6702 		if (status == 0) {
6703 			/*
6704 			 * The backend access has been successfully checked,
6705 			 * we can complete any I/O queued before the last check.
6706 			 */
6707 			vdc_eio_unqueue(vdc, starttime, B_TRUE);
6708 
6709 		} else if (vdc->num_servers > 1) {
6710 			/*
6711 			 * The backend is inaccessible for a disk with multiple
6712 			 * servers. So we force a reset to switch to another
6713 			 * server. The reset will also clear the eio queue and
6714 			 * resubmit all pending I/Os.
6715 			 */
6716 			mutex_enter(&vdc->read_lock);
6717 			vdc->read_state = VDC_READ_RESET;
6718 			cv_signal(&vdc->read_cv);
6719 			mutex_exit(&vdc->read_lock);
6720 		} else {
6721 			/*
6722 			 * There is only one path and the backend is not
6723 			 * accessible, so I/Os are actually failing because
6724 			 * of that. So we can complete I/O queued before the
6725 			 * last check.
6726 			 */
6727 			vdc_eio_unqueue(vdc, starttime, B_TRUE);
6728 		}
6729 	}
6730 
6731 	/*
6732 	 * The thread is being stopped so we can complete any queued I/O.
6733 	 */
6734 	vdc_eio_unqueue(vdc, 0, B_TRUE);
6735 	vdc->eio_thread = NULL;
6736 	mutex_exit(&vdc->lock);
6737 	thread_exit();
6738 }
6739 
6740 /*
6741  * Implement the MHIOCENFAILFAST mhd(7i) ioctl.
6742  */
6743 static int
6744 vdc_failfast(vdc_t *vdc, caddr_t arg, int mode)
6745 {
6746 	unsigned int mh_time;
6747 
6748 	if (ddi_copyin((void *)arg, &mh_time, sizeof (int), mode))
6749 		return (EFAULT);
6750 
6751 	mutex_enter(&vdc->lock);
6752 	if (mh_time != 0 && vdc->eio_thread == NULL) {
6753 		vdc->eio_thread = thread_create(NULL, 0,
6754 		    vdc_eio_thread, vdc, 0, &p0, TS_RUN,
6755 		    v.v_maxsyspri - 2);
6756 	}
6757 
6758 	vdc->failfast_interval = ((long)mh_time) * MILLISEC;
6759 	cv_signal(&vdc->eio_cv);
6760 	mutex_exit(&vdc->lock);
6761 
6762 	return (0);
6763 }
6764 
6765 /*
6766  * Implement the MHIOCTKOWN and MHIOCRELEASE mhd(7i) ioctls. These ioctls are
6767  * converted to VD_OP_SET_ACCESS operations.
6768  */
6769 static int
6770 vdc_access_set(vdc_t *vdc, uint64_t flags)
6771 {
6772 	int rv;
6773 
6774 	/* submit owership command request */
6775 	rv = vdc_do_sync_op(vdc, VD_OP_SET_ACCESS, (caddr_t)&flags,
6776 	    sizeof (uint64_t), 0, 0, VIO_both_dir, B_TRUE);
6777 
6778 	return (rv);
6779 }
6780 
6781 /*
6782  * Implement the MHIOCSTATUS mhd(7i) ioctl. This ioctl is converted to a
6783  * VD_OP_GET_ACCESS operation.
6784  */
6785 static int
6786 vdc_access_get(vdc_t *vdc, uint64_t *status)
6787 {
6788 	int rv;
6789 
6790 	/* submit owership command request */
6791 	rv = vdc_do_sync_op(vdc, VD_OP_GET_ACCESS, (caddr_t)status,
6792 	    sizeof (uint64_t), 0, 0, VIO_both_dir, B_TRUE);
6793 
6794 	return (rv);
6795 }
6796 
6797 /*
6798  * Disk Ownership Thread.
6799  *
6800  * When we have taken the ownership of a disk, this thread waits to be
6801  * notified when the LDC channel is reset so that it can recover the
6802  * ownership.
6803  *
6804  * Note that the thread handling the LDC reset (vdc_process_msg_thread())
6805  * can not be used to do the ownership recovery because it has to be
6806  * running to handle the reply message to the ownership operation.
6807  */
6808 static void
6809 vdc_ownership_thread(void *arg)
6810 {
6811 	vdc_t *vdc = (vdc_t *)arg;
6812 	clock_t timeout;
6813 	uint64_t status;
6814 
6815 	mutex_enter(&vdc->ownership_lock);
6816 	mutex_enter(&vdc->lock);
6817 
6818 	while (vdc->ownership & VDC_OWNERSHIP_WANTED) {
6819 
6820 		if ((vdc->ownership & VDC_OWNERSHIP_RESET) ||
6821 		    !(vdc->ownership & VDC_OWNERSHIP_GRANTED)) {
6822 			/*
6823 			 * There was a reset so the ownership has been lost,
6824 			 * try to recover. We do this without using the preempt
6825 			 * option so that we don't steal the ownership from
6826 			 * someone who has preempted us.
6827 			 */
6828 			DMSG(vdc, 0, "[%d] Ownership lost, recovering",
6829 			    vdc->instance);
6830 
6831 			vdc->ownership &= ~(VDC_OWNERSHIP_RESET |
6832 			    VDC_OWNERSHIP_GRANTED);
6833 
6834 			mutex_exit(&vdc->lock);
6835 
6836 			status = vdc_access_set(vdc, VD_ACCESS_SET_EXCLUSIVE |
6837 			    VD_ACCESS_SET_PRESERVE);
6838 
6839 			mutex_enter(&vdc->lock);
6840 
6841 			if (status == 0) {
6842 				DMSG(vdc, 0, "[%d] Ownership recovered",
6843 				    vdc->instance);
6844 				vdc->ownership |= VDC_OWNERSHIP_GRANTED;
6845 			} else {
6846 				DMSG(vdc, 0, "[%d] Fail to recover ownership",
6847 				    vdc->instance);
6848 			}
6849 
6850 		}
6851 
6852 		/*
6853 		 * If we have the ownership then we just wait for an event
6854 		 * to happen (LDC reset), otherwise we will retry to recover
6855 		 * after a delay.
6856 		 */
6857 		if (vdc->ownership & VDC_OWNERSHIP_GRANTED)
6858 			timeout = 0;
6859 		else
6860 			timeout = drv_usectohz(vdc_ownership_delay);
6861 
6862 		/* Release the ownership_lock and wait on the vdc lock */
6863 		mutex_exit(&vdc->ownership_lock);
6864 
6865 		if (timeout == 0)
6866 			(void) cv_wait(&vdc->ownership_cv, &vdc->lock);
6867 		else
6868 			(void) cv_reltimedwait(&vdc->ownership_cv, &vdc->lock,
6869 			    timeout, TR_CLOCK_TICK);
6870 
6871 		mutex_exit(&vdc->lock);
6872 
6873 		mutex_enter(&vdc->ownership_lock);
6874 		mutex_enter(&vdc->lock);
6875 	}
6876 
6877 	vdc->ownership_thread = NULL;
6878 	mutex_exit(&vdc->lock);
6879 	mutex_exit(&vdc->ownership_lock);
6880 
6881 	thread_exit();
6882 }
6883 
6884 static void
6885 vdc_ownership_update(vdc_t *vdc, int ownership_flags)
6886 {
6887 	ASSERT(MUTEX_HELD(&vdc->ownership_lock));
6888 
6889 	mutex_enter(&vdc->lock);
6890 	vdc->ownership = ownership_flags;
6891 	if ((vdc->ownership & VDC_OWNERSHIP_WANTED) &&
6892 	    vdc->ownership_thread == NULL) {
6893 		/* start ownership thread */
6894 		vdc->ownership_thread = thread_create(NULL, 0,
6895 		    vdc_ownership_thread, vdc, 0, &p0, TS_RUN,
6896 		    v.v_maxsyspri - 2);
6897 	} else {
6898 		/* notify the ownership thread */
6899 		cv_signal(&vdc->ownership_cv);
6900 	}
6901 	mutex_exit(&vdc->lock);
6902 }
6903 
6904 /*
6905  * Get the size and the block size of a virtual disk from the vdisk server.
6906  */
6907 static int
6908 vdc_get_capacity(vdc_t *vdc, size_t *dsk_size, size_t *blk_size)
6909 {
6910 	int rv = 0;
6911 	size_t alloc_len;
6912 	vd_capacity_t *vd_cap;
6913 
6914 	ASSERT(MUTEX_NOT_HELD(&vdc->lock));
6915 
6916 	alloc_len = P2ROUNDUP(sizeof (vd_capacity_t), sizeof (uint64_t));
6917 
6918 	vd_cap = kmem_zalloc(alloc_len, KM_SLEEP);
6919 
6920 	rv = vdc_do_sync_op(vdc, VD_OP_GET_CAPACITY, (caddr_t)vd_cap, alloc_len,
6921 	    0, 0, VIO_both_dir, B_TRUE);
6922 
6923 	*dsk_size = vd_cap->vdisk_size;
6924 	*blk_size = vd_cap->vdisk_block_size;
6925 
6926 	kmem_free(vd_cap, alloc_len);
6927 	return (rv);
6928 }
6929 
6930 /*
6931  * Check the disk capacity. Disk size information is updated if size has
6932  * changed.
6933  *
6934  * Return 0 if the disk capacity is available, or non-zero if it is not.
6935  */
6936 static int
6937 vdc_check_capacity(vdc_t *vdc)
6938 {
6939 	size_t dsk_size, blk_size;
6940 	int rv;
6941 
6942 	/*
6943 	 * If the vdisk does not support the VD_OP_GET_CAPACITY operation
6944 	 * then the disk capacity has been retrieved during the handshake
6945 	 * and there's nothing more to do here.
6946 	 */
6947 	if (!VD_OP_SUPPORTED(vdc->operations, VD_OP_GET_CAPACITY))
6948 		return (0);
6949 
6950 	if ((rv = vdc_get_capacity(vdc, &dsk_size, &blk_size)) != 0)
6951 		return (rv);
6952 
6953 	if (dsk_size == VD_SIZE_UNKNOWN || dsk_size == 0 || blk_size == 0)
6954 		return (EINVAL);
6955 
6956 	mutex_enter(&vdc->lock);
6957 	/*
6958 	 * First try to update the VIO block size (which is the same as the
6959 	 * vdisk block size). If this returns an error then that means that
6960 	 * we can not use that block size so basically the vdisk is unusable
6961 	 * and we return an error.
6962 	 */
6963 	rv = vdc_update_vio_bsize(vdc, blk_size);
6964 	if (rv == 0)
6965 		vdc_update_size(vdc, dsk_size, blk_size, vdc->max_xfer_sz);
6966 
6967 	mutex_exit(&vdc->lock);
6968 
6969 	return (rv);
6970 }
6971 
6972 /*
6973  * This structure is used in the DKIO(7I) array below.
6974  */
6975 typedef struct vdc_dk_ioctl {
6976 	uint8_t		op;		/* VD_OP_XXX value */
6977 	int		cmd;		/* Solaris ioctl operation number */
6978 	size_t		nbytes;		/* size of structure to be copied */
6979 
6980 	/* function to convert between vDisk and Solaris structure formats */
6981 	int	(*convert)(vdc_t *vdc, void *vd_buf, void *ioctl_arg,
6982 	    int mode, int dir);
6983 } vdc_dk_ioctl_t;
6984 
6985 /*
6986  * Subset of DKIO(7I) operations currently supported
6987  */
6988 static vdc_dk_ioctl_t	dk_ioctl[] = {
6989 	{VD_OP_FLUSH,		DKIOCFLUSHWRITECACHE,	0,
6990 		vdc_null_copy_func},
6991 	{VD_OP_GET_WCE,		DKIOCGETWCE,		sizeof (int),
6992 		vdc_get_wce_convert},
6993 	{VD_OP_SET_WCE,		DKIOCSETWCE,		sizeof (int),
6994 		vdc_set_wce_convert},
6995 	{VD_OP_GET_VTOC,	DKIOCGVTOC,		sizeof (vd_vtoc_t),
6996 		vdc_get_vtoc_convert},
6997 	{VD_OP_SET_VTOC,	DKIOCSVTOC,		sizeof (vd_vtoc_t),
6998 		vdc_set_vtoc_convert},
6999 	{VD_OP_GET_VTOC,	DKIOCGEXTVTOC,		sizeof (vd_vtoc_t),
7000 		vdc_get_extvtoc_convert},
7001 	{VD_OP_SET_VTOC,	DKIOCSEXTVTOC,		sizeof (vd_vtoc_t),
7002 		vdc_set_extvtoc_convert},
7003 	{VD_OP_GET_DISKGEOM,	DKIOCGGEOM,		sizeof (vd_geom_t),
7004 		vdc_get_geom_convert},
7005 	{VD_OP_GET_DISKGEOM,	DKIOCG_PHYGEOM,		sizeof (vd_geom_t),
7006 		vdc_get_geom_convert},
7007 	{VD_OP_GET_DISKGEOM, 	DKIOCG_VIRTGEOM,	sizeof (vd_geom_t),
7008 		vdc_get_geom_convert},
7009 	{VD_OP_SET_DISKGEOM,	DKIOCSGEOM,		sizeof (vd_geom_t),
7010 		vdc_set_geom_convert},
7011 	{VD_OP_GET_EFI,		DKIOCGETEFI,		0,
7012 		vdc_get_efi_convert},
7013 	{VD_OP_SET_EFI,		DKIOCSETEFI,		0,
7014 		vdc_set_efi_convert},
7015 
7016 	/* DIOCTL_RWCMD is converted to a read or a write */
7017 	{0, DIOCTL_RWCMD,  sizeof (struct dadkio_rwcmd), NULL},
7018 
7019 	/* mhd(7I) non-shared multihost disks ioctls */
7020 	{0, MHIOCTKOWN,				0, vdc_null_copy_func},
7021 	{0, MHIOCRELEASE,			0, vdc_null_copy_func},
7022 	{0, MHIOCSTATUS,			0, vdc_null_copy_func},
7023 	{0, MHIOCQRESERVE,			0, vdc_null_copy_func},
7024 
7025 	/* mhd(7I) shared multihost disks ioctls */
7026 	{0, MHIOCGRP_INKEYS,			0, vdc_null_copy_func},
7027 	{0, MHIOCGRP_INRESV,			0, vdc_null_copy_func},
7028 	{0, MHIOCGRP_REGISTER,			0, vdc_null_copy_func},
7029 	{0, MHIOCGRP_RESERVE, 			0, vdc_null_copy_func},
7030 	{0, MHIOCGRP_PREEMPTANDABORT,		0, vdc_null_copy_func},
7031 	{0, MHIOCGRP_REGISTERANDIGNOREKEY,	0, vdc_null_copy_func},
7032 
7033 	/* mhd(7I) failfast ioctl */
7034 	{0, MHIOCENFAILFAST,			0, vdc_null_copy_func},
7035 
7036 	/*
7037 	 * These particular ioctls are not sent to the server - vdc fakes up
7038 	 * the necessary info.
7039 	 */
7040 	{0, DKIOCINFO, sizeof (struct dk_cinfo), vdc_null_copy_func},
7041 	{0, DKIOCGMEDIAINFO, sizeof (struct dk_minfo), vdc_null_copy_func},
7042 	{0, USCSICMD,	sizeof (struct uscsi_cmd), vdc_null_copy_func},
7043 	{0, DKIOCPARTITION, 0, vdc_null_copy_func },
7044 	{0, DKIOCGAPART, 0, vdc_null_copy_func },
7045 	{0, DKIOCREMOVABLE, 0, vdc_null_copy_func},
7046 	{0, CDROMREADOFFSET, 0, vdc_null_copy_func}
7047 };
7048 
7049 /*
7050  * This function handles ioctl requests from the vd_efi_alloc_and_read()
7051  * function and forward them to the vdisk.
7052  */
7053 static int
7054 vd_process_efi_ioctl(void *vdisk, int cmd, uintptr_t arg)
7055 {
7056 	vdc_t *vdc = (vdc_t *)vdisk;
7057 	dev_t dev;
7058 	int rval;
7059 
7060 	dev = makedevice(ddi_driver_major(vdc->dip),
7061 	    VD_MAKE_DEV(vdc->instance, 0));
7062 
7063 	return (vd_process_ioctl(dev, cmd, (caddr_t)arg, FKIOCTL, &rval));
7064 }
7065 
7066 /*
7067  * Function:
7068  *	vd_process_ioctl()
7069  *
7070  * Description:
7071  *	This routine processes disk specific ioctl calls
7072  *
7073  * Arguments:
7074  *	dev	- the device number
7075  *	cmd	- the operation [dkio(7I)] to be processed
7076  *	arg	- pointer to user provided structure
7077  *		  (contains data to be set or reference parameter for get)
7078  *	mode	- bit flag, indicating open settings, 32/64 bit type, etc
7079  *	rvalp	- pointer to return value for calling process.
7080  *
7081  * Return Code:
7082  *	0
7083  *	EFAULT
7084  *	ENXIO
7085  *	EIO
7086  *	ENOTSUP
7087  */
7088 static int
7089 vd_process_ioctl(dev_t dev, int cmd, caddr_t arg, int mode, int *rvalp)
7090 {
7091 	int		instance = VDCUNIT(dev);
7092 	vdc_t		*vdc = NULL;
7093 	int		rv = -1;
7094 	int		idx = 0;		/* index into dk_ioctl[] */
7095 	size_t		len = 0;		/* #bytes to send to vds */
7096 	size_t		alloc_len = 0;		/* #bytes to allocate mem for */
7097 	caddr_t		mem_p = NULL;
7098 	size_t		nioctls = (sizeof (dk_ioctl)) / (sizeof (dk_ioctl[0]));
7099 	vdc_dk_ioctl_t	*iop;
7100 
7101 	vdc = ddi_get_soft_state(vdc_state, instance);
7102 	if (vdc == NULL) {
7103 		cmn_err(CE_NOTE, "![%d] Could not get soft state structure",
7104 		    instance);
7105 		return (ENXIO);
7106 	}
7107 
7108 	DMSG(vdc, 0, "[%d] Processing ioctl(%x) for dev %lx : model %x\n",
7109 	    instance, cmd, dev, ddi_model_convert_from(mode & FMODELS));
7110 
7111 	if (rvalp != NULL) {
7112 		/* the return value of the ioctl is 0 by default */
7113 		*rvalp = 0;
7114 	}
7115 
7116 	/*
7117 	 * Validate the ioctl operation to be performed.
7118 	 *
7119 	 * If we have looped through the array without finding a match then we
7120 	 * don't support this ioctl.
7121 	 */
7122 	for (idx = 0; idx < nioctls; idx++) {
7123 		if (cmd == dk_ioctl[idx].cmd)
7124 			break;
7125 	}
7126 
7127 	if (idx >= nioctls) {
7128 		DMSG(vdc, 0, "[%d] Unsupported ioctl (0x%x)\n",
7129 		    vdc->instance, cmd);
7130 		return (ENOTSUP);
7131 	}
7132 
7133 	iop = &(dk_ioctl[idx]);
7134 
7135 	if (cmd == DKIOCGETEFI || cmd == DKIOCSETEFI) {
7136 		/* size is not fixed for EFI ioctls, it depends on ioctl arg */
7137 		dk_efi_t	dk_efi;
7138 
7139 		rv = ddi_copyin(arg, &dk_efi, sizeof (dk_efi_t), mode);
7140 		if (rv != 0)
7141 			return (EFAULT);
7142 
7143 		len = sizeof (vd_efi_t) - 1 + dk_efi.dki_length;
7144 	} else {
7145 		len = iop->nbytes;
7146 	}
7147 
7148 	/* check if the ioctl is applicable */
7149 	switch (cmd) {
7150 	case CDROMREADOFFSET:
7151 	case DKIOCREMOVABLE:
7152 		return (ENOTTY);
7153 
7154 	case USCSICMD:
7155 	case MHIOCTKOWN:
7156 	case MHIOCSTATUS:
7157 	case MHIOCQRESERVE:
7158 	case MHIOCRELEASE:
7159 	case MHIOCGRP_INKEYS:
7160 	case MHIOCGRP_INRESV:
7161 	case MHIOCGRP_REGISTER:
7162 	case MHIOCGRP_RESERVE:
7163 	case MHIOCGRP_PREEMPTANDABORT:
7164 	case MHIOCGRP_REGISTERANDIGNOREKEY:
7165 	case MHIOCENFAILFAST:
7166 		if (vdc->cinfo == NULL)
7167 			return (ENXIO);
7168 		if (vdc->cinfo->dki_ctype != DKC_SCSI_CCS)
7169 			return (ENOTTY);
7170 		break;
7171 
7172 	case DIOCTL_RWCMD:
7173 		if (vdc->cinfo == NULL)
7174 			return (ENXIO);
7175 		if (vdc->cinfo->dki_ctype != DKC_DIRECT)
7176 			return (ENOTTY);
7177 		break;
7178 
7179 	case DKIOCINFO:
7180 		if (vdc->cinfo == NULL)
7181 			return (ENXIO);
7182 		break;
7183 
7184 	case DKIOCGMEDIAINFO:
7185 		if (vdc->minfo == NULL)
7186 			return (ENXIO);
7187 		if (vdc_check_capacity(vdc) != 0)
7188 			/* disk capacity is not available */
7189 			return (EIO);
7190 		break;
7191 	}
7192 
7193 	/*
7194 	 * Deal with ioctls which require a processing different than
7195 	 * converting ioctl arguments and sending a corresponding
7196 	 * VD operation.
7197 	 */
7198 	switch (cmd) {
7199 
7200 	case USCSICMD:
7201 	{
7202 		return (vdc_uscsi_cmd(vdc, arg, mode));
7203 	}
7204 
7205 	case MHIOCTKOWN:
7206 	{
7207 		mutex_enter(&vdc->ownership_lock);
7208 		/*
7209 		 * We have to set VDC_OWNERSHIP_WANTED now so that the ownership
7210 		 * can be flagged with VDC_OWNERSHIP_RESET if the LDC is reset
7211 		 * while we are processing the ioctl.
7212 		 */
7213 		vdc_ownership_update(vdc, VDC_OWNERSHIP_WANTED);
7214 
7215 		rv = vdc_access_set(vdc, VD_ACCESS_SET_EXCLUSIVE |
7216 		    VD_ACCESS_SET_PREEMPT | VD_ACCESS_SET_PRESERVE);
7217 		if (rv == 0) {
7218 			vdc_ownership_update(vdc, VDC_OWNERSHIP_WANTED |
7219 			    VDC_OWNERSHIP_GRANTED);
7220 		} else {
7221 			vdc_ownership_update(vdc, VDC_OWNERSHIP_NONE);
7222 		}
7223 		mutex_exit(&vdc->ownership_lock);
7224 		return (rv);
7225 	}
7226 
7227 	case MHIOCRELEASE:
7228 	{
7229 		mutex_enter(&vdc->ownership_lock);
7230 		rv = vdc_access_set(vdc, VD_ACCESS_SET_CLEAR);
7231 		if (rv == 0) {
7232 			vdc_ownership_update(vdc, VDC_OWNERSHIP_NONE);
7233 		}
7234 		mutex_exit(&vdc->ownership_lock);
7235 		return (rv);
7236 	}
7237 
7238 	case MHIOCSTATUS:
7239 	{
7240 		uint64_t status;
7241 
7242 		rv = vdc_access_get(vdc, &status);
7243 		if (rv == 0 && rvalp != NULL)
7244 			*rvalp = (status & VD_ACCESS_ALLOWED)? 0 : 1;
7245 		return (rv);
7246 	}
7247 
7248 	case MHIOCQRESERVE:
7249 	{
7250 		rv = vdc_access_set(vdc, VD_ACCESS_SET_EXCLUSIVE);
7251 		return (rv);
7252 	}
7253 
7254 	case MHIOCGRP_INKEYS:
7255 	{
7256 		return (vdc_mhd_inkeys(vdc, arg, mode));
7257 	}
7258 
7259 	case MHIOCGRP_INRESV:
7260 	{
7261 		return (vdc_mhd_inresv(vdc, arg, mode));
7262 	}
7263 
7264 	case MHIOCGRP_REGISTER:
7265 	{
7266 		return (vdc_mhd_register(vdc, arg, mode));
7267 	}
7268 
7269 	case MHIOCGRP_RESERVE:
7270 	{
7271 		return (vdc_mhd_reserve(vdc, arg, mode));
7272 	}
7273 
7274 	case MHIOCGRP_PREEMPTANDABORT:
7275 	{
7276 		return (vdc_mhd_preemptabort(vdc, arg, mode));
7277 	}
7278 
7279 	case MHIOCGRP_REGISTERANDIGNOREKEY:
7280 	{
7281 		return (vdc_mhd_registerignore(vdc, arg, mode));
7282 	}
7283 
7284 	case MHIOCENFAILFAST:
7285 	{
7286 		rv = vdc_failfast(vdc, arg, mode);
7287 		return (rv);
7288 	}
7289 
7290 	case DIOCTL_RWCMD:
7291 	{
7292 		return (vdc_dioctl_rwcmd(vdc, arg, mode));
7293 	}
7294 
7295 	case DKIOCGAPART:
7296 	{
7297 		return (vdc_dkio_gapart(vdc, arg, mode));
7298 	}
7299 
7300 	case DKIOCPARTITION:
7301 	{
7302 		return (vdc_dkio_partition(vdc, arg, mode));
7303 	}
7304 
7305 	case DKIOCINFO:
7306 	{
7307 		struct dk_cinfo	cinfo;
7308 
7309 		bcopy(vdc->cinfo, &cinfo, sizeof (struct dk_cinfo));
7310 		cinfo.dki_partition = VDCPART(dev);
7311 
7312 		rv = ddi_copyout(&cinfo, (void *)arg,
7313 		    sizeof (struct dk_cinfo), mode);
7314 		if (rv != 0)
7315 			return (EFAULT);
7316 
7317 		return (0);
7318 	}
7319 
7320 	case DKIOCGMEDIAINFO:
7321 	{
7322 		ASSERT(vdc->vdisk_size != 0);
7323 		ASSERT(vdc->minfo->dki_capacity != 0);
7324 		rv = ddi_copyout(vdc->minfo, (void *)arg,
7325 		    sizeof (struct dk_minfo), mode);
7326 		if (rv != 0)
7327 			return (EFAULT);
7328 
7329 		return (0);
7330 	}
7331 
7332 	case DKIOCFLUSHWRITECACHE:
7333 		{
7334 			struct dk_callback *dkc =
7335 			    (struct dk_callback *)(uintptr_t)arg;
7336 			vdc_dk_arg_t	*dkarg = NULL;
7337 
7338 			DMSG(vdc, 1, "[%d] Flush W$: mode %x\n",
7339 			    instance, mode);
7340 
7341 			/*
7342 			 * If arg is NULL, then there is no callback function
7343 			 * registered and the call operates synchronously; we
7344 			 * break and continue with the rest of the function and
7345 			 * wait for vds to return (i.e. after the request to
7346 			 * vds returns successfully, all writes completed prior
7347 			 * to the ioctl will have been flushed from the disk
7348 			 * write cache to persistent media.
7349 			 *
7350 			 * If a callback function is registered, we dispatch
7351 			 * the request on a task queue and return immediately.
7352 			 * The callback will deal with informing the calling
7353 			 * thread that the flush request is completed.
7354 			 */
7355 			if (dkc == NULL)
7356 				break;
7357 
7358 			/*
7359 			 * the asynchronous callback is only supported if
7360 			 * invoked from within the kernel
7361 			 */
7362 			if ((mode & FKIOCTL) == 0)
7363 				return (ENOTSUP);
7364 
7365 			dkarg = kmem_zalloc(sizeof (vdc_dk_arg_t), KM_SLEEP);
7366 
7367 			dkarg->mode = mode;
7368 			dkarg->dev = dev;
7369 			bcopy(dkc, &dkarg->dkc, sizeof (*dkc));
7370 
7371 			mutex_enter(&vdc->lock);
7372 			vdc->dkio_flush_pending++;
7373 			dkarg->vdc = vdc;
7374 			mutex_exit(&vdc->lock);
7375 
7376 			/* put the request on a task queue */
7377 			rv = taskq_dispatch(system_taskq, vdc_dkio_flush_cb,
7378 			    (void *)dkarg, DDI_SLEEP);
7379 			if (rv == NULL) {
7380 				/* clean up if dispatch fails */
7381 				mutex_enter(&vdc->lock);
7382 				vdc->dkio_flush_pending--;
7383 				mutex_exit(&vdc->lock);
7384 				kmem_free(dkarg, sizeof (vdc_dk_arg_t));
7385 			}
7386 
7387 			return (rv == NULL ? ENOMEM : 0);
7388 		}
7389 	}
7390 
7391 	/* catch programming error in vdc - should be a VD_OP_XXX ioctl */
7392 	ASSERT(iop->op != 0);
7393 
7394 	/* check if the vDisk server handles the operation for this vDisk */
7395 	if (VD_OP_SUPPORTED(vdc->operations, iop->op) == B_FALSE) {
7396 		DMSG(vdc, 0, "[%d] Unsupported VD_OP operation (0x%x)\n",
7397 		    vdc->instance, iop->op);
7398 		return (ENOTSUP);
7399 	}
7400 
7401 	/* LDC requires that the memory being mapped is 8-byte aligned */
7402 	alloc_len = P2ROUNDUP(len, sizeof (uint64_t));
7403 	DMSG(vdc, 1, "[%d] struct size %ld alloc %ld\n",
7404 	    instance, len, alloc_len);
7405 
7406 	if (alloc_len > 0)
7407 		mem_p = kmem_zalloc(alloc_len, KM_SLEEP);
7408 
7409 	/*
7410 	 * Call the conversion function for this ioctl which, if necessary,
7411 	 * converts from the Solaris format to the format ARC'ed
7412 	 * as part of the vDisk protocol (FWARC 2006/195)
7413 	 */
7414 	ASSERT(iop->convert != NULL);
7415 	rv = (iop->convert)(vdc, arg, mem_p, mode, VD_COPYIN);
7416 	if (rv != 0) {
7417 		DMSG(vdc, 0, "[%d] convert func returned %d for ioctl 0x%x\n",
7418 		    instance, rv, cmd);
7419 		if (mem_p != NULL)
7420 			kmem_free(mem_p, alloc_len);
7421 		return (rv);
7422 	}
7423 
7424 	/*
7425 	 * send request to vds to service the ioctl.
7426 	 */
7427 	rv = vdc_do_sync_op(vdc, iop->op, mem_p, alloc_len,
7428 	    VDCPART(dev), 0, VIO_both_dir, B_TRUE);
7429 
7430 	if (rv != 0) {
7431 		/*
7432 		 * This is not necessarily an error. The ioctl could
7433 		 * be returning a value such as ENOTTY to indicate
7434 		 * that the ioctl is not applicable.
7435 		 */
7436 		DMSG(vdc, 0, "[%d] vds returned %d for ioctl 0x%x\n",
7437 		    instance, rv, cmd);
7438 		if (mem_p != NULL)
7439 			kmem_free(mem_p, alloc_len);
7440 
7441 		return (rv);
7442 	}
7443 
7444 	/*
7445 	 * Call the conversion function (if it exists) for this ioctl
7446 	 * which converts from the format ARC'ed as part of the vDisk
7447 	 * protocol (FWARC 2006/195) back to a format understood by
7448 	 * the rest of Solaris.
7449 	 */
7450 	rv = (iop->convert)(vdc, mem_p, arg, mode, VD_COPYOUT);
7451 	if (rv != 0) {
7452 		DMSG(vdc, 0, "[%d] convert func returned %d for ioctl 0x%x\n",
7453 		    instance, rv, cmd);
7454 		if (mem_p != NULL)
7455 			kmem_free(mem_p, alloc_len);
7456 		return (rv);
7457 	}
7458 
7459 	if (mem_p != NULL)
7460 		kmem_free(mem_p, alloc_len);
7461 
7462 	return (rv);
7463 }
7464 
7465 /*
7466  * Function:
7467  *
7468  * Description:
7469  *	This is an empty conversion function used by ioctl calls which
7470  *	do not need to convert the data being passed in/out to userland
7471  */
7472 static int
7473 vdc_null_copy_func(vdc_t *vdc, void *from, void *to, int mode, int dir)
7474 {
7475 	_NOTE(ARGUNUSED(vdc))
7476 	_NOTE(ARGUNUSED(from))
7477 	_NOTE(ARGUNUSED(to))
7478 	_NOTE(ARGUNUSED(mode))
7479 	_NOTE(ARGUNUSED(dir))
7480 
7481 	return (0);
7482 }
7483 
7484 static int
7485 vdc_get_wce_convert(vdc_t *vdc, void *from, void *to,
7486     int mode, int dir)
7487 {
7488 	_NOTE(ARGUNUSED(vdc))
7489 
7490 	if (dir == VD_COPYIN)
7491 		return (0);		/* nothing to do */
7492 
7493 	if (ddi_copyout(from, to, sizeof (int), mode) != 0)
7494 		return (EFAULT);
7495 
7496 	return (0);
7497 }
7498 
7499 static int
7500 vdc_set_wce_convert(vdc_t *vdc, void *from, void *to,
7501     int mode, int dir)
7502 {
7503 	_NOTE(ARGUNUSED(vdc))
7504 
7505 	if (dir == VD_COPYOUT)
7506 		return (0);		/* nothing to do */
7507 
7508 	if (ddi_copyin(from, to, sizeof (int), mode) != 0)
7509 		return (EFAULT);
7510 
7511 	return (0);
7512 }
7513 
7514 /*
7515  * Function:
7516  *	vdc_get_vtoc_convert()
7517  *
7518  * Description:
7519  *	This routine performs the necessary convertions from the DKIOCGVTOC
7520  *	Solaris structure to the format defined in FWARC 2006/195.
7521  *
7522  *	In the struct vtoc definition, the timestamp field is marked as not
7523  *	supported so it is not part of vDisk protocol (FWARC 2006/195).
7524  *	However SVM uses that field to check it can write into the VTOC,
7525  *	so we fake up the info of that field.
7526  *
7527  * Arguments:
7528  *	vdc	- the vDisk client
7529  *	from	- the buffer containing the data to be copied from
7530  *	to	- the buffer to be copied to
7531  *	mode	- flags passed to ioctl() call
7532  *	dir	- the "direction" of the copy - VD_COPYIN or VD_COPYOUT
7533  *
7534  * Return Code:
7535  *	0	- Success
7536  *	ENXIO	- incorrect buffer passed in.
7537  *	EFAULT	- ddi_copyout routine encountered an error.
7538  */
7539 static int
7540 vdc_get_vtoc_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7541 {
7542 	int		i;
7543 	struct vtoc	vtoc;
7544 	struct vtoc32	vtoc32;
7545 	struct extvtoc	evtoc;
7546 	int		rv;
7547 
7548 	if (dir != VD_COPYOUT)
7549 		return (0);	/* nothing to do */
7550 
7551 	if ((from == NULL) || (to == NULL))
7552 		return (ENXIO);
7553 
7554 	if (vdc->vdisk_size > VD_OLDVTOC_LIMIT)
7555 		return (EOVERFLOW);
7556 
7557 	VD_VTOC2VTOC((vd_vtoc_t *)from, &evtoc);
7558 
7559 	/* fake the VTOC timestamp field */
7560 	for (i = 0; i < V_NUMPAR; i++) {
7561 		evtoc.timestamp[i] = vdc->vtoc->timestamp[i];
7562 	}
7563 
7564 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
7565 		/* LINTED E_ASSIGN_NARROW_CONV */
7566 		extvtoctovtoc32(evtoc, vtoc32);
7567 		rv = ddi_copyout(&vtoc32, to, sizeof (vtoc32), mode);
7568 		if (rv != 0)
7569 			rv = EFAULT;
7570 	} else {
7571 		extvtoctovtoc(evtoc, vtoc);
7572 		rv = ddi_copyout(&vtoc, to, sizeof (vtoc), mode);
7573 		if (rv != 0)
7574 			rv = EFAULT;
7575 	}
7576 
7577 	return (rv);
7578 }
7579 
7580 /*
7581  * Function:
7582  *	vdc_set_vtoc_convert()
7583  *
7584  * Description:
7585  *	This routine performs the necessary convertions from the DKIOCSVTOC
7586  *	Solaris structure to the format defined in FWARC 2006/195.
7587  *
7588  * Arguments:
7589  *	vdc	- the vDisk client
7590  *	from	- Buffer with data
7591  *	to	- Buffer where data is to be copied to
7592  *	mode	- flags passed to ioctl
7593  *	dir	- direction of copy (in or out)
7594  *
7595  * Return Code:
7596  *	0	- Success
7597  *	ENXIO	- Invalid buffer passed in
7598  *	EFAULT	- ddi_copyin of data failed
7599  */
7600 static int
7601 vdc_set_vtoc_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7602 {
7603 	void		*uvtoc;
7604 	struct vtoc	vtoc;
7605 	struct vtoc32	vtoc32;
7606 	struct extvtoc	evtoc;
7607 	int		i, rv;
7608 
7609 	if ((from == NULL) || (to == NULL))
7610 		return (ENXIO);
7611 
7612 	if (vdc->vdisk_size > VD_OLDVTOC_LIMIT)
7613 		return (EOVERFLOW);
7614 
7615 	uvtoc = (dir == VD_COPYIN)? from : to;
7616 
7617 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
7618 		rv = ddi_copyin(uvtoc, &vtoc32, sizeof (vtoc32), mode);
7619 		if (rv != 0)
7620 			return (EFAULT);
7621 		vtoc32toextvtoc(vtoc32, evtoc);
7622 	} else {
7623 		rv = ddi_copyin(uvtoc, &vtoc, sizeof (vtoc), mode);
7624 		if (rv != 0)
7625 			return (EFAULT);
7626 		vtoctoextvtoc(vtoc, evtoc);
7627 	}
7628 
7629 	if (dir == VD_COPYOUT) {
7630 		/*
7631 		 * The disk label may have changed. Revalidate the disk
7632 		 * geometry. This will also update the device nodes.
7633 		 */
7634 		vdc_validate(vdc);
7635 
7636 		/*
7637 		 * We also need to keep track of the timestamp fields.
7638 		 */
7639 		for (i = 0; i < V_NUMPAR; i++) {
7640 			vdc->vtoc->timestamp[i] = evtoc.timestamp[i];
7641 		}
7642 
7643 	} else {
7644 		VTOC2VD_VTOC(&evtoc, (vd_vtoc_t *)to);
7645 	}
7646 
7647 	return (0);
7648 }
7649 
7650 static int
7651 vdc_get_extvtoc_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7652 {
7653 	int		i, rv;
7654 	struct extvtoc	evtoc;
7655 
7656 	if (dir != VD_COPYOUT)
7657 		return (0);	/* nothing to do */
7658 
7659 	if ((from == NULL) || (to == NULL))
7660 		return (ENXIO);
7661 
7662 	VD_VTOC2VTOC((vd_vtoc_t *)from, &evtoc);
7663 
7664 	/* fake the VTOC timestamp field */
7665 	for (i = 0; i < V_NUMPAR; i++) {
7666 		evtoc.timestamp[i] = vdc->vtoc->timestamp[i];
7667 	}
7668 
7669 	rv = ddi_copyout(&evtoc, to, sizeof (struct extvtoc), mode);
7670 	if (rv != 0)
7671 		rv = EFAULT;
7672 
7673 	return (rv);
7674 }
7675 
7676 static int
7677 vdc_set_extvtoc_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7678 {
7679 	void		*uvtoc;
7680 	struct extvtoc	evtoc;
7681 	int		i, rv;
7682 
7683 	if ((from == NULL) || (to == NULL))
7684 		return (ENXIO);
7685 
7686 	uvtoc = (dir == VD_COPYIN)? from : to;
7687 
7688 	rv = ddi_copyin(uvtoc, &evtoc, sizeof (struct extvtoc), mode);
7689 	if (rv != 0)
7690 		return (EFAULT);
7691 
7692 	if (dir == VD_COPYOUT) {
7693 		/*
7694 		 * The disk label may have changed. Revalidate the disk
7695 		 * geometry. This will also update the device nodes.
7696 		 */
7697 		vdc_validate(vdc);
7698 
7699 		/*
7700 		 * We also need to keep track of the timestamp fields.
7701 		 */
7702 		for (i = 0; i < V_NUMPAR; i++) {
7703 			vdc->vtoc->timestamp[i] = evtoc.timestamp[i];
7704 		}
7705 
7706 	} else {
7707 		VTOC2VD_VTOC(&evtoc, (vd_vtoc_t *)to);
7708 	}
7709 
7710 	return (0);
7711 }
7712 
7713 /*
7714  * Function:
7715  *	vdc_get_geom_convert()
7716  *
7717  * Description:
7718  *	This routine performs the necessary convertions from the DKIOCGGEOM,
7719  *	DKIOCG_PHYSGEOM and DKIOG_VIRTGEOM Solaris structures to the format
7720  *	defined in FWARC 2006/195
7721  *
7722  * Arguments:
7723  *	vdc	- the vDisk client
7724  *	from	- Buffer with data
7725  *	to	- Buffer where data is to be copied to
7726  *	mode	- flags passed to ioctl
7727  *	dir	- direction of copy (in or out)
7728  *
7729  * Return Code:
7730  *	0	- Success
7731  *	ENXIO	- Invalid buffer passed in
7732  *	EFAULT	- ddi_copyout of data failed
7733  */
7734 static int
7735 vdc_get_geom_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7736 {
7737 	_NOTE(ARGUNUSED(vdc))
7738 
7739 	struct dk_geom	geom;
7740 	int	copy_len = sizeof (struct dk_geom);
7741 	int	rv = 0;
7742 
7743 	if (dir != VD_COPYOUT)
7744 		return (0);	/* nothing to do */
7745 
7746 	if ((from == NULL) || (to == NULL))
7747 		return (ENXIO);
7748 
7749 	VD_GEOM2DK_GEOM((vd_geom_t *)from, &geom);
7750 	rv = ddi_copyout(&geom, to, copy_len, mode);
7751 	if (rv != 0)
7752 		rv = EFAULT;
7753 
7754 	return (rv);
7755 }
7756 
7757 /*
7758  * Function:
7759  *	vdc_set_geom_convert()
7760  *
7761  * Description:
7762  *	This routine performs the necessary convertions from the DKIOCSGEOM
7763  *	Solaris structure to the format defined in FWARC 2006/195.
7764  *
7765  * Arguments:
7766  *	vdc	- the vDisk client
7767  *	from	- Buffer with data
7768  *	to	- Buffer where data is to be copied to
7769  *	mode	- flags passed to ioctl
7770  *	dir	- direction of copy (in or out)
7771  *
7772  * Return Code:
7773  *	0	- Success
7774  *	ENXIO	- Invalid buffer passed in
7775  *	EFAULT	- ddi_copyin of data failed
7776  */
7777 static int
7778 vdc_set_geom_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7779 {
7780 	_NOTE(ARGUNUSED(vdc))
7781 
7782 	vd_geom_t	vdgeom;
7783 	void		*tmp_mem = NULL;
7784 	int		copy_len = sizeof (struct dk_geom);
7785 	int		rv = 0;
7786 
7787 	if (dir != VD_COPYIN)
7788 		return (0);	/* nothing to do */
7789 
7790 	if ((from == NULL) || (to == NULL))
7791 		return (ENXIO);
7792 
7793 	tmp_mem = kmem_alloc(copy_len, KM_SLEEP);
7794 
7795 	rv = ddi_copyin(from, tmp_mem, copy_len, mode);
7796 	if (rv != 0) {
7797 		kmem_free(tmp_mem, copy_len);
7798 		return (EFAULT);
7799 	}
7800 	DK_GEOM2VD_GEOM((struct dk_geom *)tmp_mem, &vdgeom);
7801 	bcopy(&vdgeom, to, sizeof (vdgeom));
7802 	kmem_free(tmp_mem, copy_len);
7803 
7804 	return (0);
7805 }
7806 
7807 static int
7808 vdc_get_efi_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7809 {
7810 	_NOTE(ARGUNUSED(vdc))
7811 
7812 	vd_efi_t	*vd_efi;
7813 	dk_efi_t	dk_efi;
7814 	int		rv = 0;
7815 	void		*uaddr;
7816 
7817 	if ((from == NULL) || (to == NULL))
7818 		return (ENXIO);
7819 
7820 	if (dir == VD_COPYIN) {
7821 
7822 		vd_efi = (vd_efi_t *)to;
7823 
7824 		rv = ddi_copyin(from, &dk_efi, sizeof (dk_efi_t), mode);
7825 		if (rv != 0)
7826 			return (EFAULT);
7827 
7828 		vd_efi->lba = dk_efi.dki_lba;
7829 		vd_efi->length = dk_efi.dki_length;
7830 		bzero(vd_efi->data, vd_efi->length);
7831 
7832 	} else {
7833 
7834 		rv = ddi_copyin(to, &dk_efi, sizeof (dk_efi_t), mode);
7835 		if (rv != 0)
7836 			return (EFAULT);
7837 
7838 		uaddr = dk_efi.dki_data;
7839 
7840 		dk_efi.dki_data = kmem_alloc(dk_efi.dki_length, KM_SLEEP);
7841 
7842 		VD_EFI2DK_EFI((vd_efi_t *)from, &dk_efi);
7843 
7844 		rv = ddi_copyout(dk_efi.dki_data, uaddr, dk_efi.dki_length,
7845 		    mode);
7846 		if (rv != 0)
7847 			return (EFAULT);
7848 
7849 		kmem_free(dk_efi.dki_data, dk_efi.dki_length);
7850 	}
7851 
7852 	return (0);
7853 }
7854 
7855 static int
7856 vdc_set_efi_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7857 {
7858 	_NOTE(ARGUNUSED(vdc))
7859 
7860 	dk_efi_t	dk_efi;
7861 	void		*uaddr;
7862 
7863 	if (dir == VD_COPYOUT) {
7864 		/*
7865 		 * The disk label may have changed. Revalidate the disk
7866 		 * geometry. This will also update the device nodes.
7867 		 */
7868 		vdc_validate(vdc);
7869 		return (0);
7870 	}
7871 
7872 	if ((from == NULL) || (to == NULL))
7873 		return (ENXIO);
7874 
7875 	if (ddi_copyin(from, &dk_efi, sizeof (dk_efi_t), mode) != 0)
7876 		return (EFAULT);
7877 
7878 	uaddr = dk_efi.dki_data;
7879 
7880 	dk_efi.dki_data = kmem_alloc(dk_efi.dki_length, KM_SLEEP);
7881 
7882 	if (ddi_copyin(uaddr, dk_efi.dki_data, dk_efi.dki_length, mode) != 0)
7883 		return (EFAULT);
7884 
7885 	DK_EFI2VD_EFI(&dk_efi, (vd_efi_t *)to);
7886 
7887 	kmem_free(dk_efi.dki_data, dk_efi.dki_length);
7888 
7889 	return (0);
7890 }
7891 
7892 
7893 /* -------------------------------------------------------------------------- */
7894 
7895 /*
7896  * Function:
7897  *	vdc_create_fake_geometry()
7898  *
7899  * Description:
7900  *	This routine fakes up the disk info needed for some DKIO ioctls such
7901  *	as DKIOCINFO and DKIOCGMEDIAINFO [just like lofi(7D) and ramdisk(7D) do]
7902  *
7903  *	Note: This function must not be called until the vDisk attributes have
7904  *	been exchanged as part of the handshake with the vDisk server.
7905  *
7906  * Arguments:
7907  *	vdc	- soft state pointer for this instance of the device driver.
7908  *
7909  * Return Code:
7910  *	none.
7911  */
7912 static void
7913 vdc_create_fake_geometry(vdc_t *vdc)
7914 {
7915 	ASSERT(vdc != NULL);
7916 	ASSERT(vdc->max_xfer_sz != 0);
7917 
7918 	/*
7919 	 * DKIOCINFO support
7920 	 */
7921 	if (vdc->cinfo == NULL)
7922 		vdc->cinfo = kmem_zalloc(sizeof (struct dk_cinfo), KM_SLEEP);
7923 
7924 	(void) strcpy(vdc->cinfo->dki_cname, VDC_DRIVER_NAME);
7925 	(void) strcpy(vdc->cinfo->dki_dname, VDC_DRIVER_NAME);
7926 	/* max_xfer_sz is #blocks so we don't need to divide by vdisk_bsize */
7927 	vdc->cinfo->dki_maxtransfer = vdc->max_xfer_sz;
7928 
7929 	/*
7930 	 * We set the controller type to DKC_SCSI_CCS only if the VD_OP_SCSICMD
7931 	 * operation is supported, otherwise the controller type is DKC_DIRECT.
7932 	 * Version 1.0 does not support the VD_OP_SCSICMD operation, so the
7933 	 * controller type is always DKC_DIRECT in that case.
7934 	 *
7935 	 * If the virtual disk is backed by a physical CD/DVD device or
7936 	 * an ISO image, modify the controller type to indicate this
7937 	 */
7938 	switch (vdc->vdisk_media) {
7939 	case VD_MEDIA_CD:
7940 	case VD_MEDIA_DVD:
7941 		vdc->cinfo->dki_ctype = DKC_CDROM;
7942 		break;
7943 	case VD_MEDIA_FIXED:
7944 		if (VD_OP_SUPPORTED(vdc->operations, VD_OP_SCSICMD))
7945 			vdc->cinfo->dki_ctype = DKC_SCSI_CCS;
7946 		else
7947 			vdc->cinfo->dki_ctype = DKC_DIRECT;
7948 		break;
7949 	default:
7950 		/* in the case of v1.0 we default to a fixed disk */
7951 		vdc->cinfo->dki_ctype = DKC_DIRECT;
7952 		break;
7953 	}
7954 	vdc->cinfo->dki_flags = DKI_FMTVOL;
7955 	vdc->cinfo->dki_cnum = 0;
7956 	vdc->cinfo->dki_addr = 0;
7957 	vdc->cinfo->dki_space = 0;
7958 	vdc->cinfo->dki_prio = 0;
7959 	vdc->cinfo->dki_vec = 0;
7960 	vdc->cinfo->dki_unit = vdc->instance;
7961 	vdc->cinfo->dki_slave = 0;
7962 	/*
7963 	 * The partition number will be created on the fly depending on the
7964 	 * actual slice (i.e. minor node) that is used to request the data.
7965 	 */
7966 	vdc->cinfo->dki_partition = 0;
7967 
7968 	/*
7969 	 * DKIOCGMEDIAINFO support
7970 	 */
7971 	if (vdc->minfo == NULL)
7972 		vdc->minfo = kmem_zalloc(sizeof (struct dk_minfo), KM_SLEEP);
7973 
7974 	if (vio_ver_is_supported(vdc->ver, 1, 1)) {
7975 		vdc->minfo->dki_media_type =
7976 		    VD_MEDIATYPE2DK_MEDIATYPE(vdc->vdisk_media);
7977 	} else {
7978 		vdc->minfo->dki_media_type = DK_FIXED_DISK;
7979 	}
7980 
7981 	vdc->minfo->dki_capacity = vdc->vdisk_size;
7982 	vdc->minfo->dki_lbsize = vdc->vdisk_bsize;
7983 }
7984 
7985 static ushort_t
7986 vdc_lbl2cksum(struct dk_label *label)
7987 {
7988 	int	count;
7989 	ushort_t sum, *sp;
7990 
7991 	count =	(sizeof (struct dk_label)) / (sizeof (short)) - 1;
7992 	sp = (ushort_t *)label;
7993 	sum = 0;
7994 	while (count--) {
7995 		sum ^= *sp++;
7996 	}
7997 
7998 	return (sum);
7999 }
8000 
8001 static void
8002 vdc_update_size(vdc_t *vdc, size_t dsk_size, size_t blk_size, size_t xfr_size)
8003 {
8004 	vd_err_stats_t  *stp;
8005 
8006 	ASSERT(MUTEX_HELD(&vdc->lock));
8007 	ASSERT(xfr_size != 0);
8008 
8009 	/*
8010 	 * If the disk size is unknown or sizes are unchanged then don't
8011 	 * update anything.
8012 	 */
8013 	if (dsk_size == VD_SIZE_UNKNOWN || dsk_size == 0 ||
8014 	    (blk_size == vdc->vdisk_bsize && dsk_size == vdc->vdisk_size &&
8015 	    xfr_size == vdc->max_xfer_sz))
8016 		return;
8017 
8018 	/*
8019 	 * We don't know at compile time what the vDisk server will think
8020 	 * are good values but we apply a large (arbitrary) upper bound to
8021 	 * prevent memory exhaustion in vdc if it was allocating a DRing
8022 	 * based of huge values sent by the server. We probably will never
8023 	 * exceed this except if the message was garbage.
8024 	 */
8025 	if ((xfr_size * blk_size) > (PAGESIZE * DEV_BSIZE)) {
8026 		DMSG(vdc, 0, "[%d] vds block transfer size too big;"
8027 		    " using max supported by vdc", vdc->instance);
8028 		xfr_size = maxphys / blk_size;
8029 	}
8030 
8031 	vdc->max_xfer_sz = xfr_size;
8032 	vdc->vdisk_bsize = blk_size;
8033 	vdc->vdisk_size = dsk_size;
8034 
8035 	stp = (vd_err_stats_t *)vdc->err_stats->ks_data;
8036 	stp->vd_capacity.value.ui64 = dsk_size * blk_size;
8037 
8038 	vdc->minfo->dki_capacity = dsk_size;
8039 	vdc->minfo->dki_lbsize = (uint_t)blk_size;
8040 }
8041 
8042 /*
8043  * Update information about the VIO block size. The VIO block size is the
8044  * same as the vdisk block size which is stored in vdc->vdisk_bsize so we
8045  * do not store that information again.
8046  *
8047  * However, buf structures will always use a logical block size of 512 bytes
8048  * (DEV_BSIZE) and we will need to convert logical block numbers to VIO block
8049  * numbers for each read or write operation using vdc_strategy(). To speed up
8050  * this conversion, we expect the VIO block size to be a power of 2 and a
8051  * multiple 512 bytes (DEV_BSIZE), and we cache some useful information.
8052  *
8053  * The function return EINVAL if the new VIO block size (blk_size) is not a
8054  * power of 2 or not a multiple of 512 bytes, otherwise it returns 0.
8055  */
8056 static int
8057 vdc_update_vio_bsize(vdc_t *vdc, uint32_t blk_size)
8058 {
8059 	uint32_t ratio, n;
8060 	int nshift = 0;
8061 
8062 	vdc->vio_bmask = 0;
8063 	vdc->vio_bshift = 0;
8064 
8065 	ASSERT(blk_size > 0);
8066 
8067 	if ((blk_size % DEV_BSIZE) != 0)
8068 		return (EINVAL);
8069 
8070 	ratio = blk_size / DEV_BSIZE;
8071 
8072 	for (n = ratio; n > 1; n >>= 1) {
8073 		if ((n & 0x1) != 0) {
8074 			/* blk_size is not a power of 2 */
8075 			return (EINVAL);
8076 		}
8077 		nshift++;
8078 	}
8079 
8080 	vdc->vio_bshift = nshift;
8081 	vdc->vio_bmask = ratio - 1;
8082 
8083 	return (0);
8084 }
8085 
8086 /*
8087  * Function:
8088  *	vdc_validate_geometry
8089  *
8090  * Description:
8091  *	This routine discovers the label and geometry of the disk. It stores
8092  *	the disk label and related information in the vdc structure. If it
8093  *	fails to validate the geometry or to discover the disk label then
8094  *	the label is marked as unknown (VD_DISK_LABEL_UNK).
8095  *
8096  * Arguments:
8097  *	vdc	- soft state pointer for this instance of the device driver.
8098  *
8099  * Return Code:
8100  *	0	- success.
8101  *	EINVAL	- unknown disk label.
8102  *	ENOTSUP	- geometry not applicable (EFI label).
8103  *	EIO	- error accessing the disk.
8104  */
8105 static int
8106 vdc_validate_geometry(vdc_t *vdc)
8107 {
8108 	dev_t	dev;
8109 	int	rv, rval;
8110 	struct dk_label *label;
8111 	struct dk_geom geom;
8112 	struct extvtoc vtoc;
8113 	efi_gpt_t *gpt;
8114 	efi_gpe_t *gpe;
8115 	vd_efi_dev_t edev;
8116 
8117 	ASSERT(vdc != NULL);
8118 	ASSERT(vdc->vtoc != NULL && vdc->geom != NULL);
8119 	ASSERT(MUTEX_HELD(&vdc->lock));
8120 
8121 	mutex_exit(&vdc->lock);
8122 	/*
8123 	 * Check the disk capacity in case it has changed. If that fails then
8124 	 * we proceed and we will be using the disk size we currently have.
8125 	 */
8126 	(void) vdc_check_capacity(vdc);
8127 	dev = makedevice(ddi_driver_major(vdc->dip),
8128 	    VD_MAKE_DEV(vdc->instance, 0));
8129 
8130 	rv = vd_process_ioctl(dev, DKIOCGGEOM, (caddr_t)&geom, FKIOCTL, &rval);
8131 	if (rv == 0)
8132 		rv = vd_process_ioctl(dev, DKIOCGEXTVTOC, (caddr_t)&vtoc,
8133 		    FKIOCTL, &rval);
8134 
8135 	if (rv == ENOTSUP) {
8136 		/*
8137 		 * If the device does not support VTOC then we try
8138 		 * to read an EFI label.
8139 		 *
8140 		 * We need to know the block size and the disk size to
8141 		 * be able to read an EFI label.
8142 		 */
8143 		if (vdc->vdisk_size == 0) {
8144 			mutex_enter(&vdc->lock);
8145 			vdc_store_label_unk(vdc);
8146 			return (EIO);
8147 		}
8148 
8149 		VDC_EFI_DEV_SET(edev, vdc, vd_process_efi_ioctl);
8150 
8151 		rv = vd_efi_alloc_and_read(&edev, &gpt, &gpe);
8152 
8153 		if (rv) {
8154 			DMSG(vdc, 0, "[%d] Failed to get EFI (err=%d)",
8155 			    vdc->instance, rv);
8156 			mutex_enter(&vdc->lock);
8157 			vdc_store_label_unk(vdc);
8158 			return (EIO);
8159 		}
8160 
8161 		mutex_enter(&vdc->lock);
8162 		vdc_store_label_efi(vdc, gpt, gpe);
8163 		vd_efi_free(&edev, gpt, gpe);
8164 		return (ENOTSUP);
8165 	}
8166 
8167 	if (rv != 0) {
8168 		DMSG(vdc, 0, "[%d] Failed to get VTOC (err=%d)",
8169 		    vdc->instance, rv);
8170 		mutex_enter(&vdc->lock);
8171 		vdc_store_label_unk(vdc);
8172 		if (rv != EINVAL)
8173 			rv = EIO;
8174 		return (rv);
8175 	}
8176 
8177 	/* check that geometry and vtoc are valid */
8178 	if (geom.dkg_nhead == 0 || geom.dkg_nsect == 0 ||
8179 	    vtoc.v_sanity != VTOC_SANE) {
8180 		mutex_enter(&vdc->lock);
8181 		vdc_store_label_unk(vdc);
8182 		return (EINVAL);
8183 	}
8184 
8185 	/*
8186 	 * We have a disk and a valid VTOC. However this does not mean
8187 	 * that the disk currently have a VTOC label. The returned VTOC may
8188 	 * be a default VTOC to be used for configuring the disk (this is
8189 	 * what is done for disk image). So we read the label from the
8190 	 * beginning of the disk to ensure we really have a VTOC label.
8191 	 *
8192 	 * FUTURE: This could be the default way for reading the VTOC
8193 	 * from the disk as opposed to sending the VD_OP_GET_VTOC
8194 	 * to the server. This will be the default if vdc is implemented
8195 	 * ontop of cmlb.
8196 	 */
8197 
8198 	/*
8199 	 * Single slice disk does not support read using an absolute disk
8200 	 * offset so we just rely on the DKIOCGVTOC ioctl in that case.
8201 	 */
8202 	if (vdc->vdisk_type == VD_DISK_TYPE_SLICE) {
8203 		mutex_enter(&vdc->lock);
8204 		if (vtoc.v_nparts != 1) {
8205 			vdc_store_label_unk(vdc);
8206 			return (EINVAL);
8207 		}
8208 		vdc_store_label_vtoc(vdc, &geom, &vtoc);
8209 		return (0);
8210 	}
8211 
8212 	if (vtoc.v_nparts != V_NUMPAR) {
8213 		mutex_enter(&vdc->lock);
8214 		vdc_store_label_unk(vdc);
8215 		return (EINVAL);
8216 	}
8217 
8218 	/*
8219 	 * Most CD/DVDs do not have a disk label and the label is
8220 	 * generated by the disk driver.  So the on-disk label check
8221 	 * below may fail and we return now to avoid this problem.
8222 	 */
8223 	if (vdc->vdisk_media == VD_MEDIA_CD ||
8224 	    vdc->vdisk_media == VD_MEDIA_DVD) {
8225 		mutex_enter(&vdc->lock);
8226 		vdc_store_label_vtoc(vdc, &geom, &vtoc);
8227 		return (0);
8228 	}
8229 
8230 	/*
8231 	 * Read disk label from start of disk
8232 	 */
8233 	label = kmem_alloc(vdc->vdisk_bsize, KM_SLEEP);
8234 
8235 	rv = vdc_do_op(vdc, VD_OP_BREAD, (caddr_t)label, vdc->vdisk_bsize,
8236 	    VD_SLICE_NONE, 0, NULL, VIO_read_dir, VDC_OP_NORMAL);
8237 
8238 	if (rv != 0 || label->dkl_magic != DKL_MAGIC ||
8239 	    label->dkl_cksum != vdc_lbl2cksum(label)) {
8240 		DMSG(vdc, 1, "[%d] Got VTOC with invalid label\n",
8241 		    vdc->instance);
8242 		kmem_free(label, vdc->vdisk_bsize);
8243 		mutex_enter(&vdc->lock);
8244 		vdc_store_label_unk(vdc);
8245 		return (EINVAL);
8246 	}
8247 
8248 	kmem_free(label, vdc->vdisk_bsize);
8249 	mutex_enter(&vdc->lock);
8250 	vdc_store_label_vtoc(vdc, &geom, &vtoc);
8251 	return (0);
8252 }
8253 
8254 /*
8255  * Function:
8256  *	vdc_validate
8257  *
8258  * Description:
8259  *	This routine discovers the label of the disk and create the
8260  *	appropriate device nodes if the label has changed.
8261  *
8262  * Arguments:
8263  *	vdc	- soft state pointer for this instance of the device driver.
8264  *
8265  * Return Code:
8266  *	none.
8267  */
8268 static void
8269 vdc_validate(vdc_t *vdc)
8270 {
8271 	vd_disk_label_t old_label;
8272 	vd_slice_t old_slice[V_NUMPAR];
8273 	int rv;
8274 
8275 	ASSERT(!MUTEX_HELD(&vdc->lock));
8276 
8277 	mutex_enter(&vdc->lock);
8278 
8279 	/* save the current label and vtoc */
8280 	old_label = vdc->vdisk_label;
8281 	bcopy(vdc->slice, &old_slice, sizeof (vd_slice_t) * V_NUMPAR);
8282 
8283 	/* check the geometry */
8284 	(void) vdc_validate_geometry(vdc);
8285 
8286 	/* if the disk label has changed, update device nodes */
8287 	if (vdc->vdisk_type == VD_DISK_TYPE_DISK &&
8288 	    vdc->vdisk_label != old_label) {
8289 
8290 		if (vdc->vdisk_label == VD_DISK_LABEL_EFI)
8291 			rv = vdc_create_device_nodes_efi(vdc);
8292 		else
8293 			rv = vdc_create_device_nodes_vtoc(vdc);
8294 
8295 		if (rv != 0) {
8296 			DMSG(vdc, 0, "![%d] Failed to update device nodes",
8297 			    vdc->instance);
8298 		}
8299 	}
8300 
8301 	mutex_exit(&vdc->lock);
8302 }
8303 
8304 static void
8305 vdc_validate_task(void *arg)
8306 {
8307 	vdc_t *vdc = (vdc_t *)arg;
8308 
8309 	vdc_validate(vdc);
8310 
8311 	mutex_enter(&vdc->lock);
8312 	ASSERT(vdc->validate_pending > 0);
8313 	vdc->validate_pending--;
8314 	mutex_exit(&vdc->lock);
8315 }
8316 
8317 /*
8318  * Function:
8319  *	vdc_setup_devid()
8320  *
8321  * Description:
8322  *	This routine discovers the devid of a vDisk. It requests the devid of
8323  *	the underlying device from the vDisk server, builds an encapsulated
8324  *	devid based on the retrieved devid and registers that new devid to
8325  *	the vDisk.
8326  *
8327  * Arguments:
8328  *	vdc	- soft state pointer for this instance of the device driver.
8329  *
8330  * Return Code:
8331  *	0	- A devid was succesfully registered for the vDisk
8332  */
8333 static int
8334 vdc_setup_devid(vdc_t *vdc)
8335 {
8336 	int rv;
8337 	vd_devid_t *vd_devid;
8338 	size_t bufsize, bufid_len;
8339 	ddi_devid_t vdisk_devid;
8340 	char *devid_str;
8341 
8342 	/*
8343 	 * At first sight, we don't know the size of the devid that the
8344 	 * server will return but this size will be encoded into the
8345 	 * reply. So we do a first request using a default size then we
8346 	 * check if this size was large enough. If not then we do a second
8347 	 * request with the correct size returned by the server. Note that
8348 	 * ldc requires size to be 8-byte aligned.
8349 	 */
8350 	bufsize = P2ROUNDUP(VD_DEVID_SIZE(VD_DEVID_DEFAULT_LEN),
8351 	    sizeof (uint64_t));
8352 	vd_devid = kmem_zalloc(bufsize, KM_SLEEP);
8353 	bufid_len = bufsize - sizeof (vd_efi_t) - 1;
8354 
8355 	rv = vdc_do_op(vdc, VD_OP_GET_DEVID, (caddr_t)vd_devid,
8356 	    bufsize, 0, 0, NULL, VIO_both_dir, 0);
8357 
8358 	DMSG(vdc, 2, "do_op returned %d\n", rv);
8359 
8360 	if (rv) {
8361 		kmem_free(vd_devid, bufsize);
8362 		return (rv);
8363 	}
8364 
8365 	if (vd_devid->length > bufid_len) {
8366 		/*
8367 		 * The returned devid is larger than the buffer used. Try again
8368 		 * with a buffer with the right size.
8369 		 */
8370 		kmem_free(vd_devid, bufsize);
8371 		bufsize = P2ROUNDUP(VD_DEVID_SIZE(vd_devid->length),
8372 		    sizeof (uint64_t));
8373 		vd_devid = kmem_zalloc(bufsize, KM_SLEEP);
8374 		bufid_len = bufsize - sizeof (vd_efi_t) - 1;
8375 
8376 		rv = vdc_do_sync_op(vdc, VD_OP_GET_DEVID, (caddr_t)vd_devid,
8377 		    bufsize, 0, 0, VIO_both_dir, B_TRUE);
8378 
8379 		if (rv) {
8380 			kmem_free(vd_devid, bufsize);
8381 			return (rv);
8382 		}
8383 	}
8384 
8385 	/*
8386 	 * The virtual disk should have the same device id as the one associated
8387 	 * with the physical disk it is mapped on, otherwise sharing a disk
8388 	 * between a LDom and a non-LDom may not work (for example for a shared
8389 	 * SVM disk set).
8390 	 *
8391 	 * The DDI framework does not allow creating a device id with any
8392 	 * type so we first create a device id of type DEVID_ENCAP and then
8393 	 * we restore the orignal type of the physical device.
8394 	 */
8395 
8396 	DMSG(vdc, 2, ": devid length = %d\n", vd_devid->length);
8397 
8398 	/* build an encapsulated devid based on the returned devid */
8399 	if (ddi_devid_init(vdc->dip, DEVID_ENCAP, vd_devid->length,
8400 	    vd_devid->id, &vdisk_devid) != DDI_SUCCESS) {
8401 		DMSG(vdc, 1, "[%d] Fail to created devid\n", vdc->instance);
8402 		kmem_free(vd_devid, bufsize);
8403 		return (1);
8404 	}
8405 
8406 	DEVID_FORMTYPE((impl_devid_t *)vdisk_devid, vd_devid->type);
8407 
8408 	ASSERT(ddi_devid_valid(vdisk_devid) == DDI_SUCCESS);
8409 
8410 	kmem_free(vd_devid, bufsize);
8411 
8412 	if (vdc->devid != NULL) {
8413 		/* check that the devid hasn't changed */
8414 		if (ddi_devid_compare(vdisk_devid, vdc->devid) == 0) {
8415 			ddi_devid_free(vdisk_devid);
8416 			return (0);
8417 		}
8418 
8419 		cmn_err(CE_WARN, "vdisk@%d backend devid has changed",
8420 		    vdc->instance);
8421 
8422 		devid_str = ddi_devid_str_encode(vdc->devid, NULL);
8423 
8424 		cmn_err(CE_CONT, "vdisk@%d backend initial devid: %s",
8425 		    vdc->instance,
8426 		    (devid_str)? devid_str : "<encoding error>");
8427 
8428 		if (devid_str)
8429 			ddi_devid_str_free(devid_str);
8430 
8431 		devid_str = ddi_devid_str_encode(vdisk_devid, NULL);
8432 
8433 		cmn_err(CE_CONT, "vdisk@%d backend current devid: %s",
8434 		    vdc->instance,
8435 		    (devid_str)? devid_str : "<encoding error>");
8436 
8437 		if (devid_str)
8438 			ddi_devid_str_free(devid_str);
8439 
8440 		ddi_devid_free(vdisk_devid);
8441 		return (1);
8442 	}
8443 
8444 	if (ddi_devid_register(vdc->dip, vdisk_devid) != DDI_SUCCESS) {
8445 		DMSG(vdc, 1, "[%d] Fail to register devid\n", vdc->instance);
8446 		ddi_devid_free(vdisk_devid);
8447 		return (1);
8448 	}
8449 
8450 	vdc->devid = vdisk_devid;
8451 
8452 	return (0);
8453 }
8454 
8455 static void
8456 vdc_store_label_efi(vdc_t *vdc, efi_gpt_t *gpt, efi_gpe_t *gpe)
8457 {
8458 	int i, nparts;
8459 
8460 	ASSERT(MUTEX_HELD(&vdc->lock));
8461 
8462 	vdc->vdisk_label = VD_DISK_LABEL_EFI;
8463 	bzero(vdc->vtoc, sizeof (struct extvtoc));
8464 	bzero(vdc->geom, sizeof (struct dk_geom));
8465 	bzero(vdc->slice, sizeof (vd_slice_t) * V_NUMPAR);
8466 
8467 	nparts = gpt->efi_gpt_NumberOfPartitionEntries;
8468 
8469 	for (i = 0; i < nparts && i < VD_EFI_WD_SLICE; i++) {
8470 
8471 		if (gpe[i].efi_gpe_StartingLBA == 0 &&
8472 		    gpe[i].efi_gpe_EndingLBA == 0) {
8473 			continue;
8474 		}
8475 
8476 		vdc->slice[i].start = gpe[i].efi_gpe_StartingLBA;
8477 		vdc->slice[i].nblocks = gpe[i].efi_gpe_EndingLBA -
8478 		    gpe[i].efi_gpe_StartingLBA + 1;
8479 	}
8480 
8481 	ASSERT(vdc->vdisk_size != 0);
8482 	vdc->slice[VD_EFI_WD_SLICE].start = 0;
8483 	vdc->slice[VD_EFI_WD_SLICE].nblocks = vdc->vdisk_size;
8484 
8485 }
8486 
8487 static void
8488 vdc_store_label_vtoc(vdc_t *vdc, struct dk_geom *geom, struct extvtoc *vtoc)
8489 {
8490 	int i;
8491 
8492 	ASSERT(MUTEX_HELD(&vdc->lock));
8493 	ASSERT(vdc->vdisk_bsize == vtoc->v_sectorsz);
8494 
8495 	vdc->vdisk_label = VD_DISK_LABEL_VTOC;
8496 	bcopy(vtoc, vdc->vtoc, sizeof (struct extvtoc));
8497 	bcopy(geom, vdc->geom, sizeof (struct dk_geom));
8498 	bzero(vdc->slice, sizeof (vd_slice_t) * V_NUMPAR);
8499 
8500 	for (i = 0; i < vtoc->v_nparts; i++) {
8501 		vdc->slice[i].start = vtoc->v_part[i].p_start;
8502 		vdc->slice[i].nblocks = vtoc->v_part[i].p_size;
8503 	}
8504 }
8505 
8506 static void
8507 vdc_store_label_unk(vdc_t *vdc)
8508 {
8509 	ASSERT(MUTEX_HELD(&vdc->lock));
8510 
8511 	vdc->vdisk_label = VD_DISK_LABEL_UNK;
8512 	bzero(vdc->vtoc, sizeof (struct extvtoc));
8513 	bzero(vdc->geom, sizeof (struct dk_geom));
8514 	bzero(vdc->slice, sizeof (vd_slice_t) * V_NUMPAR);
8515 }
8516