xref: /illumos-gate/usr/src/uts/sun4v/io/vdc.c (revision e4f7ccbe1e17940fdbde623116bcc8cd761d9ffc)
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 (c) 2006, 2010, Oracle and/or its affiliates. All rights reserved.
24  */
25 
26 /*
27  * LDoms virtual disk client (vdc) device driver
28  *
29  * This driver runs on a guest logical domain and communicates with the virtual
30  * disk server (vds) driver running on the service domain which is exporting
31  * virtualized "disks" to the guest logical domain.
32  *
33  * The driver can be divided into four sections:
34  *
35  * 1) generic device driver housekeeping
36  *	_init, _fini, attach, detach, ops structures, etc.
37  *
38  * 2) communication channel setup
39  *	Setup the communications link over the LDC channel that vdc uses to
40  *	talk to the vDisk server. Initialise the descriptor ring which
41  *	allows the LDC clients to transfer data via memory mappings.
42  *
43  * 3) Support exported to upper layers (filesystems, etc)
44  *	The upper layers call into vdc via strategy(9E) and dkio(4I)
45  *	ioctl calls. vdc will copy the data to be written to the descriptor
46  *	ring or maps the buffer to store the data read by the vDisk
47  *	server into the descriptor ring. It then sends a message to the
48  *	vDisk server requesting it to complete the operation.
49  *
50  * 4) Handling responses from vDisk server.
51  *	The vDisk server will ACK some or all of the messages vdc sends to it
52  *	(this is configured during the handshake). Upon receipt of an ACK
53  *	vdc will check the descriptor ring and signal to the upper layer
54  *	code waiting on the IO.
55  */
56 
57 #include <sys/atomic.h>
58 #include <sys/conf.h>
59 #include <sys/disp.h>
60 #include <sys/ddi.h>
61 #include <sys/dkio.h>
62 #include <sys/efi_partition.h>
63 #include <sys/fcntl.h>
64 #include <sys/file.h>
65 #include <sys/kstat.h>
66 #include <sys/mach_descrip.h>
67 #include <sys/modctl.h>
68 #include <sys/mdeg.h>
69 #include <sys/note.h>
70 #include <sys/open.h>
71 #include <sys/random.h>
72 #include <sys/sdt.h>
73 #include <sys/stat.h>
74 #include <sys/sunddi.h>
75 #include <sys/types.h>
76 #include <sys/promif.h>
77 #include <sys/var.h>
78 #include <sys/vtoc.h>
79 #include <sys/archsystm.h>
80 #include <sys/sysmacros.h>
81 
82 #include <sys/cdio.h>
83 #include <sys/dktp/fdisk.h>
84 #include <sys/dktp/dadkio.h>
85 #include <sys/fs/dv_node.h>
86 #include <sys/mhd.h>
87 #include <sys/scsi/generic/sense.h>
88 #include <sys/scsi/impl/uscsi.h>
89 #include <sys/scsi/impl/services.h>
90 #include <sys/scsi/targets/sddef.h>
91 
92 #include <sys/ldoms.h>
93 #include <sys/ldc.h>
94 #include <sys/vio_common.h>
95 #include <sys/vio_mailbox.h>
96 #include <sys/vio_util.h>
97 #include <sys/vdsk_common.h>
98 #include <sys/vdsk_mailbox.h>
99 #include <sys/vdc.h>
100 
101 #define	VD_OLDVTOC_LIMIT	0x7fffffff
102 
103 /*
104  * function prototypes
105  */
106 
107 /* standard driver functions */
108 static int	vdc_open(dev_t *dev, int flag, int otyp, cred_t *cred);
109 static int	vdc_close(dev_t dev, int flag, int otyp, cred_t *cred);
110 static int	vdc_strategy(struct buf *buf);
111 static int	vdc_print(dev_t dev, char *str);
112 static int	vdc_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblk);
113 static int	vdc_read(dev_t dev, struct uio *uio, cred_t *cred);
114 static int	vdc_write(dev_t dev, struct uio *uio, cred_t *cred);
115 static int	vdc_ioctl(dev_t dev, int cmd, intptr_t arg, int mode,
116 			cred_t *credp, int *rvalp);
117 static int	vdc_aread(dev_t dev, struct aio_req *aio, cred_t *cred);
118 static int	vdc_awrite(dev_t dev, struct aio_req *aio, cred_t *cred);
119 
120 static int	vdc_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd,
121 			void *arg, void **resultp);
122 static int	vdc_attach(dev_info_t *dip, ddi_attach_cmd_t cmd);
123 static int	vdc_detach(dev_info_t *dip, ddi_detach_cmd_t cmd);
124 static int	vdc_prop_op(dev_t dev, dev_info_t *dip, ddi_prop_op_t prop_op,
125 		    int mod_flags, char *name, caddr_t valuep, int *lengthp);
126 
127 /* setup */
128 static void	vdc_min(struct buf *bufp);
129 static int	vdc_send(vdc_t *vdc, caddr_t pkt, size_t *msglen);
130 static int	vdc_do_ldc_init(vdc_t *vdc, vdc_server_t *srvr);
131 static int	vdc_start_ldc_connection(vdc_t *vdc);
132 static int	vdc_create_device_nodes(vdc_t *vdc);
133 static int	vdc_create_device_nodes_efi(vdc_t *vdc);
134 static int	vdc_create_device_nodes_vtoc(vdc_t *vdc);
135 static void	vdc_create_io_kstats(vdc_t *vdc);
136 static void	vdc_create_err_kstats(vdc_t *vdc);
137 static void	vdc_set_err_kstats(vdc_t *vdc);
138 static int	vdc_get_md_node(dev_info_t *dip, md_t **mdpp,
139 		    mde_cookie_t *vd_nodep);
140 static int	vdc_init_ports(vdc_t *vdc, md_t *mdp, mde_cookie_t vd_nodep);
141 static void	vdc_fini_ports(vdc_t *vdc);
142 static void	vdc_switch_server(vdc_t *vdcp);
143 static int	vdc_do_ldc_up(vdc_t *vdc);
144 static void	vdc_terminate_ldc(vdc_t *vdc, vdc_server_t *srvr);
145 static int	vdc_init_descriptor_ring(vdc_t *vdc);
146 static void	vdc_destroy_descriptor_ring(vdc_t *vdc);
147 static int	vdc_setup_devid(vdc_t *vdc);
148 static void	vdc_store_label_efi(vdc_t *, efi_gpt_t *, efi_gpe_t *);
149 static void	vdc_store_label_vtoc(vdc_t *, struct dk_geom *,
150 		    struct extvtoc *);
151 static void	vdc_store_label_unk(vdc_t *vdc);
152 static boolean_t vdc_is_opened(vdc_t *vdc);
153 static void	vdc_update_size(vdc_t *vdc, size_t, size_t, size_t);
154 static int	vdc_update_vio_bsize(vdc_t *vdc, uint32_t);
155 
156 /* handshake with vds */
157 static int		vdc_init_ver_negotiation(vdc_t *vdc, vio_ver_t ver);
158 static int		vdc_ver_negotiation(vdc_t *vdcp);
159 static int		vdc_init_attr_negotiation(vdc_t *vdc);
160 static int		vdc_attr_negotiation(vdc_t *vdcp);
161 static int		vdc_init_dring_negotiate(vdc_t *vdc);
162 static int		vdc_dring_negotiation(vdc_t *vdcp);
163 static int		vdc_send_rdx(vdc_t *vdcp);
164 static int		vdc_rdx_exchange(vdc_t *vdcp);
165 static boolean_t	vdc_is_supported_version(vio_ver_msg_t *ver_msg);
166 
167 /* processing incoming messages from vDisk server */
168 static void	vdc_process_msg_thread(vdc_t *vdc);
169 static int	vdc_recv(vdc_t *vdc, vio_msg_t *msgp, size_t *nbytesp);
170 
171 static uint_t	vdc_handle_cb(uint64_t event, caddr_t arg);
172 static int	vdc_process_data_msg(vdc_t *vdc, vio_msg_t *msg);
173 static int	vdc_handle_ver_msg(vdc_t *vdc, vio_ver_msg_t *ver_msg);
174 static int	vdc_handle_attr_msg(vdc_t *vdc, vd_attr_msg_t *attr_msg);
175 static int	vdc_handle_dring_reg_msg(vdc_t *vdc, vio_dring_reg_msg_t *msg);
176 static int	vdc_send_request(vdc_t *vdcp, int operation,
177 		    caddr_t addr, size_t nbytes, int slice, diskaddr_t offset,
178 		    buf_t *bufp, vio_desc_direction_t dir, int flags);
179 static int	vdc_map_to_shared_dring(vdc_t *vdcp, int idx);
180 static int	vdc_populate_descriptor(vdc_t *vdcp, int operation,
181 		    caddr_t addr, size_t nbytes, int slice, diskaddr_t offset,
182 		    buf_t *bufp, vio_desc_direction_t dir, int flags);
183 static int	vdc_do_sync_op(vdc_t *vdcp, int operation, caddr_t addr,
184 		    size_t nbytes, int slice, diskaddr_t offset,
185 		    vio_desc_direction_t dir, boolean_t);
186 static int	vdc_do_op(vdc_t *vdc, int op, caddr_t addr, size_t nbytes,
187 		    int slice, diskaddr_t offset, struct buf *bufp,
188 		    vio_desc_direction_t dir, int flags);
189 
190 static int	vdc_wait_for_response(vdc_t *vdcp, vio_msg_t *msgp);
191 static int	vdc_drain_response(vdc_t *vdcp, struct buf *buf);
192 static int	vdc_depopulate_descriptor(vdc_t *vdc, uint_t idx);
193 static int	vdc_populate_mem_hdl(vdc_t *vdcp, vdc_local_desc_t *ldep);
194 static int	vdc_verify_seq_num(vdc_t *vdc, vio_dring_msg_t *dring_msg);
195 
196 /* dkio */
197 static int	vd_process_ioctl(dev_t dev, int cmd, caddr_t arg, int mode,
198 		    int *rvalp);
199 static int	vd_process_efi_ioctl(void *vdisk, int cmd, uintptr_t arg);
200 static void	vdc_create_fake_geometry(vdc_t *vdc);
201 static int	vdc_validate_geometry(vdc_t *vdc);
202 static void	vdc_validate(vdc_t *vdc);
203 static void	vdc_validate_task(void *arg);
204 static int	vdc_null_copy_func(vdc_t *vdc, void *from, void *to,
205 		    int mode, int dir);
206 static int	vdc_get_wce_convert(vdc_t *vdc, void *from, void *to,
207 		    int mode, int dir);
208 static int	vdc_set_wce_convert(vdc_t *vdc, void *from, void *to,
209 		    int mode, int dir);
210 static int	vdc_get_vtoc_convert(vdc_t *vdc, void *from, void *to,
211 		    int mode, int dir);
212 static int	vdc_set_vtoc_convert(vdc_t *vdc, void *from, void *to,
213 		    int mode, int dir);
214 static int	vdc_get_extvtoc_convert(vdc_t *vdc, void *from, void *to,
215 		    int mode, int dir);
216 static int	vdc_set_extvtoc_convert(vdc_t *vdc, void *from, void *to,
217 		    int mode, int dir);
218 static int	vdc_get_geom_convert(vdc_t *vdc, void *from, void *to,
219 		    int mode, int dir);
220 static int	vdc_set_geom_convert(vdc_t *vdc, void *from, void *to,
221 		    int mode, int dir);
222 static int	vdc_get_efi_convert(vdc_t *vdc, void *from, void *to,
223 		    int mode, int dir);
224 static int	vdc_set_efi_convert(vdc_t *vdc, void *from, void *to,
225 		    int mode, int dir);
226 
227 static void	vdc_ownership_update(vdc_t *vdc, int ownership_flags);
228 static int	vdc_access_set(vdc_t *vdc, uint64_t flags);
229 static vdc_io_t	*vdc_eio_queue(vdc_t *vdc, int index);
230 static void	vdc_eio_unqueue(vdc_t *vdc, clock_t deadline,
231 		    boolean_t complete_io);
232 static int	vdc_eio_check(vdc_t *vdc, int flags);
233 static void	vdc_eio_thread(void *arg);
234 
235 /*
236  * Module variables
237  */
238 
239 /*
240  * Number of handshake retries with the current server before switching to
241  * a different server. These retries are done so that we stick with the same
242  * server if vdc receives a LDC reset event during the initiation of the
243  * handshake. This can happen if vdc reset the LDC channel and then immediately
244  * retry a connexion before it has received the LDC reset event.
245  *
246  * If there is only one server then we "switch" to the same server. We also
247  * switch if the handshake has reached the attribute negotiate step whatever
248  * the number of handshake retries might be.
249  */
250 static uint_t vdc_hshake_retries = VDC_HSHAKE_RETRIES;
251 
252 /*
253  * If the handshake done during the attach fails then the two following
254  * variables will also be used to control the number of retries for the
255  * next handshakes. In that case, when a handshake is done after the
256  * attach (i.e. the vdc lifecycle is VDC_ONLINE_PENDING) then the handshake
257  * will be retried until we have done an attribution negotiation with each
258  * server, with a specified minimum total number of negotations (the value
259  * of the vdc_hattr_min_initial or vdc_hattr_min variable).
260  *
261  * This prevents new I/Os on a newly used vdisk to block forever if the
262  * attribute negotiations can not be done, and to limit the amount of time
263  * before I/Os will fail. Basically, attribute negotiations will fail when
264  * the service is up but the backend does not exist. In that case, vds will
265  * typically retry to access the backend during 50 seconds. So I/Os will fail
266  * after the following amount of time:
267  *
268  *	50 seconds x max(number of servers, vdc->hattr_min)
269  *
270  * After that the handshake done during the attach has failed then the next
271  * handshake will use vdc_attr_min_initial. This handshake will correspond to
272  * the very first I/O to the device. If this handshake also fails then
273  * vdc_hattr_min will be used for subsequent handshakes. We typically allow
274  * more retries for the first handshake (VDC_HATTR_MIN_INITIAL = 3) to give more
275  * time for the backend to become available (50s x VDC_HATTR_MIN_INITIAL = 150s)
276  * in case this is a critical vdisk (e.g. vdisk access during boot). Then we use
277  * a smaller value (VDC_HATTR_MIN = 1) to avoid waiting too long for each I/O.
278  */
279 static uint_t vdc_hattr_min_initial = VDC_HATTR_MIN_INITIAL;
280 static uint_t vdc_hattr_min = VDC_HATTR_MIN;
281 
282 /*
283  * Tunable variables to control how long vdc waits before timing out on
284  * various operations
285  */
286 static int	vdc_timeout = 0; /* units: seconds */
287 static int	vdc_ldcup_timeout = 1; /* units: seconds */
288 
289 static uint64_t vdc_hz_min_ldc_delay;
290 static uint64_t vdc_min_timeout_ldc = 1 * MILLISEC;
291 static uint64_t vdc_hz_max_ldc_delay;
292 static uint64_t vdc_max_timeout_ldc = 100 * MILLISEC;
293 
294 static uint64_t vdc_ldc_read_init_delay = 1 * MILLISEC;
295 static uint64_t vdc_ldc_read_max_delay = 100 * MILLISEC;
296 
297 /* values for dumping - need to run in a tighter loop */
298 static uint64_t	vdc_usec_timeout_dump = 100 * MILLISEC;	/* 0.1s units: ns */
299 static int	vdc_dump_retries = 100;
300 
301 static uint16_t	vdc_scsi_timeout = 60;	/* 60s units: seconds  */
302 
303 static uint64_t vdc_ownership_delay = 6 * MICROSEC; /* 6s units: usec */
304 
305 /* Count of the number of vdc instances attached */
306 static volatile uint32_t	vdc_instance_count = 0;
307 
308 /* Tunable to log all SCSI errors */
309 static boolean_t vdc_scsi_log_error = B_FALSE;
310 
311 /* Soft state pointer */
312 static void	*vdc_state;
313 
314 /*
315  * Controlling the verbosity of the error/debug messages
316  *
317  * vdc_msglevel - controls level of messages
318  * vdc_matchinst - 64-bit variable where each bit corresponds
319  *                 to the vdc instance the vdc_msglevel applies.
320  */
321 int		vdc_msglevel = 0x0;
322 uint64_t	vdc_matchinst = 0ull;
323 
324 /*
325  * Supported vDisk protocol version pairs.
326  *
327  * The first array entry is the latest and preferred version.
328  */
329 static const vio_ver_t	vdc_version[] = {{1, 1}};
330 
331 static struct cb_ops vdc_cb_ops = {
332 	vdc_open,	/* cb_open */
333 	vdc_close,	/* cb_close */
334 	vdc_strategy,	/* cb_strategy */
335 	vdc_print,	/* cb_print */
336 	vdc_dump,	/* cb_dump */
337 	vdc_read,	/* cb_read */
338 	vdc_write,	/* cb_write */
339 	vdc_ioctl,	/* cb_ioctl */
340 	nodev,		/* cb_devmap */
341 	nodev,		/* cb_mmap */
342 	nodev,		/* cb_segmap */
343 	nochpoll,	/* cb_chpoll */
344 	vdc_prop_op,	/* cb_prop_op */
345 	NULL,		/* cb_str */
346 	D_MP | D_64BIT,	/* cb_flag */
347 	CB_REV,		/* cb_rev */
348 	vdc_aread,	/* cb_aread */
349 	vdc_awrite	/* cb_awrite */
350 };
351 
352 static struct dev_ops vdc_ops = {
353 	DEVO_REV,	/* devo_rev */
354 	0,		/* devo_refcnt */
355 	vdc_getinfo,	/* devo_getinfo */
356 	nulldev,	/* devo_identify */
357 	nulldev,	/* devo_probe */
358 	vdc_attach,	/* devo_attach */
359 	vdc_detach,	/* devo_detach */
360 	nodev,		/* devo_reset */
361 	&vdc_cb_ops,	/* devo_cb_ops */
362 	NULL,		/* devo_bus_ops */
363 	nulldev,	/* devo_power */
364 	ddi_quiesce_not_needed,	/* devo_quiesce */
365 };
366 
367 static struct modldrv modldrv = {
368 	&mod_driverops,
369 	"virtual disk client",
370 	&vdc_ops,
371 };
372 
373 static struct modlinkage modlinkage = {
374 	MODREV_1,
375 	&modldrv,
376 	NULL
377 };
378 
379 /* -------------------------------------------------------------------------- */
380 
381 /*
382  * Device Driver housekeeping and setup
383  */
384 
385 int
386 _init(void)
387 {
388 	int	status;
389 
390 	if ((status = ddi_soft_state_init(&vdc_state, sizeof (vdc_t), 1)) != 0)
391 		return (status);
392 	if ((status = mod_install(&modlinkage)) != 0)
393 		ddi_soft_state_fini(&vdc_state);
394 	return (status);
395 }
396 
397 int
398 _info(struct modinfo *modinfop)
399 {
400 	return (mod_info(&modlinkage, modinfop));
401 }
402 
403 int
404 _fini(void)
405 {
406 	int	status;
407 
408 	if ((status = mod_remove(&modlinkage)) != 0)
409 		return (status);
410 	ddi_soft_state_fini(&vdc_state);
411 	return (0);
412 }
413 
414 static int
415 vdc_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd,  void *arg, void **resultp)
416 {
417 	_NOTE(ARGUNUSED(dip))
418 
419 	int	instance = VDCUNIT((dev_t)arg);
420 	vdc_t	*vdc = NULL;
421 
422 	switch (cmd) {
423 	case DDI_INFO_DEVT2DEVINFO:
424 		if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
425 			*resultp = NULL;
426 			return (DDI_FAILURE);
427 		}
428 		*resultp = vdc->dip;
429 		return (DDI_SUCCESS);
430 	case DDI_INFO_DEVT2INSTANCE:
431 		*resultp = (void *)(uintptr_t)instance;
432 		return (DDI_SUCCESS);
433 	default:
434 		*resultp = NULL;
435 		return (DDI_FAILURE);
436 	}
437 }
438 
439 static int
440 vdc_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
441 {
442 	kt_did_t eio_tid, ownership_tid;
443 	int	instance;
444 	int	rv;
445 	vdc_server_t *srvr;
446 	vdc_t	*vdc = NULL;
447 
448 	switch (cmd) {
449 	case DDI_DETACH:
450 		/* the real work happens below */
451 		break;
452 	case DDI_SUSPEND:
453 		/* nothing to do for this non-device */
454 		return (DDI_SUCCESS);
455 	default:
456 		return (DDI_FAILURE);
457 	}
458 
459 	ASSERT(cmd == DDI_DETACH);
460 	instance = ddi_get_instance(dip);
461 	DMSGX(1, "[%d] Entered\n", instance);
462 
463 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
464 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
465 		return (DDI_FAILURE);
466 	}
467 
468 	if (vdc_is_opened(vdc)) {
469 		DMSG(vdc, 0, "[%d] Cannot detach: device is open", instance);
470 		return (DDI_FAILURE);
471 	}
472 
473 	if (vdc->dkio_flush_pending) {
474 		DMSG(vdc, 0,
475 		    "[%d] Cannot detach: %d outstanding DKIO flushes\n",
476 		    instance, vdc->dkio_flush_pending);
477 		return (DDI_FAILURE);
478 	}
479 
480 	if (vdc->validate_pending) {
481 		DMSG(vdc, 0,
482 		    "[%d] Cannot detach: %d outstanding validate request\n",
483 		    instance, vdc->validate_pending);
484 		return (DDI_FAILURE);
485 	}
486 
487 	DMSG(vdc, 0, "[%d] proceeding...\n", instance);
488 
489 	/* If we took ownership, release ownership */
490 	mutex_enter(&vdc->ownership_lock);
491 	if (vdc->ownership & VDC_OWNERSHIP_GRANTED) {
492 		rv = vdc_access_set(vdc, VD_ACCESS_SET_CLEAR);
493 		if (rv == 0) {
494 			vdc_ownership_update(vdc, VDC_OWNERSHIP_NONE);
495 		}
496 	}
497 	mutex_exit(&vdc->ownership_lock);
498 
499 	/* mark instance as detaching */
500 	mutex_enter(&vdc->lock);
501 	vdc->lifecycle	= VDC_LC_DETACHING;
502 	mutex_exit(&vdc->lock);
503 
504 	/*
505 	 * Try and disable callbacks to prevent another handshake. We have to
506 	 * disable callbacks for all servers.
507 	 */
508 	for (srvr = vdc->server_list; srvr != NULL; srvr = srvr->next) {
509 		rv = ldc_set_cb_mode(srvr->ldc_handle, LDC_CB_DISABLE);
510 		DMSG(vdc, 0, "callback disabled (ldc=%lu, rv=%d)\n",
511 		    srvr->ldc_id, rv);
512 	}
513 
514 	if (vdc->initialized & VDC_THREAD) {
515 		mutex_enter(&vdc->read_lock);
516 		if ((vdc->read_state == VDC_READ_WAITING) ||
517 		    (vdc->read_state == VDC_READ_RESET)) {
518 			vdc->read_state = VDC_READ_RESET;
519 			cv_signal(&vdc->read_cv);
520 		}
521 
522 		mutex_exit(&vdc->read_lock);
523 
524 		/* wake up any thread waiting for connection to come online */
525 		mutex_enter(&vdc->lock);
526 		if (vdc->state == VDC_STATE_INIT_WAITING) {
527 			DMSG(vdc, 0,
528 			    "[%d] write reset - move to resetting state...\n",
529 			    instance);
530 			vdc->state = VDC_STATE_RESETTING;
531 			cv_signal(&vdc->initwait_cv);
532 		} else if (vdc->state == VDC_STATE_FAILED) {
533 			vdc->io_pending = B_TRUE;
534 			cv_signal(&vdc->io_pending_cv);
535 		}
536 		mutex_exit(&vdc->lock);
537 
538 		/* now wait until state transitions to VDC_STATE_DETACH */
539 		thread_join(vdc->msg_proc_thr->t_did);
540 		ASSERT(vdc->state == VDC_STATE_DETACH);
541 		DMSG(vdc, 0, "[%d] Reset thread exit and join ..\n",
542 		    vdc->instance);
543 	}
544 
545 	mutex_enter(&vdc->lock);
546 
547 	if (vdc->initialized & VDC_DRING)
548 		vdc_destroy_descriptor_ring(vdc);
549 
550 	vdc_fini_ports(vdc);
551 
552 	if (vdc->eio_thread) {
553 		eio_tid = vdc->eio_thread->t_did;
554 		vdc->failfast_interval = 0;
555 		ASSERT(vdc->num_servers == 0);
556 		cv_signal(&vdc->eio_cv);
557 	} else {
558 		eio_tid = 0;
559 	}
560 
561 	if (vdc->ownership & VDC_OWNERSHIP_WANTED) {
562 		ownership_tid = vdc->ownership_thread->t_did;
563 		vdc->ownership = VDC_OWNERSHIP_NONE;
564 		cv_signal(&vdc->ownership_cv);
565 	} else {
566 		ownership_tid = 0;
567 	}
568 
569 	mutex_exit(&vdc->lock);
570 
571 	if (eio_tid != 0)
572 		thread_join(eio_tid);
573 
574 	if (ownership_tid != 0)
575 		thread_join(ownership_tid);
576 
577 	if (vdc->initialized & VDC_MINOR)
578 		ddi_remove_minor_node(dip, NULL);
579 
580 	if (vdc->io_stats) {
581 		kstat_delete(vdc->io_stats);
582 		vdc->io_stats = NULL;
583 	}
584 
585 	if (vdc->err_stats) {
586 		kstat_delete(vdc->err_stats);
587 		vdc->err_stats = NULL;
588 	}
589 
590 	if (vdc->initialized & VDC_LOCKS) {
591 		mutex_destroy(&vdc->lock);
592 		mutex_destroy(&vdc->read_lock);
593 		mutex_destroy(&vdc->ownership_lock);
594 		cv_destroy(&vdc->initwait_cv);
595 		cv_destroy(&vdc->dring_free_cv);
596 		cv_destroy(&vdc->membind_cv);
597 		cv_destroy(&vdc->sync_blocked_cv);
598 		cv_destroy(&vdc->read_cv);
599 		cv_destroy(&vdc->running_cv);
600 		cv_destroy(&vdc->io_pending_cv);
601 		cv_destroy(&vdc->ownership_cv);
602 		cv_destroy(&vdc->eio_cv);
603 	}
604 
605 	if (vdc->minfo)
606 		kmem_free(vdc->minfo, sizeof (struct dk_minfo));
607 
608 	if (vdc->cinfo)
609 		kmem_free(vdc->cinfo, sizeof (struct dk_cinfo));
610 
611 	if (vdc->vtoc)
612 		kmem_free(vdc->vtoc, sizeof (struct extvtoc));
613 
614 	if (vdc->geom)
615 		kmem_free(vdc->geom, sizeof (struct dk_geom));
616 
617 	if (vdc->devid) {
618 		ddi_devid_unregister(dip);
619 		ddi_devid_free(vdc->devid);
620 	}
621 
622 	if (vdc->initialized & VDC_SOFT_STATE)
623 		ddi_soft_state_free(vdc_state, instance);
624 
625 	DMSG(vdc, 0, "[%d] End %p\n", instance, (void *)vdc);
626 
627 	return (DDI_SUCCESS);
628 }
629 
630 
631 static int
632 vdc_do_attach(dev_info_t *dip)
633 {
634 	int		instance;
635 	vdc_t		*vdc = NULL;
636 	int		status;
637 	md_t		*mdp;
638 	mde_cookie_t	vd_node;
639 
640 	ASSERT(dip != NULL);
641 
642 	instance = ddi_get_instance(dip);
643 	if (ddi_soft_state_zalloc(vdc_state, instance) != DDI_SUCCESS) {
644 		cmn_err(CE_NOTE, "[%d] Couldn't alloc state structure",
645 		    instance);
646 		return (DDI_FAILURE);
647 	}
648 
649 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
650 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
651 		return (DDI_FAILURE);
652 	}
653 
654 	/*
655 	 * We assign the value to initialized in this case to zero out the
656 	 * variable and then set bits in it to indicate what has been done
657 	 */
658 	vdc->initialized = VDC_SOFT_STATE;
659 
660 	vdc_hz_min_ldc_delay = drv_usectohz(vdc_min_timeout_ldc);
661 	vdc_hz_max_ldc_delay = drv_usectohz(vdc_max_timeout_ldc);
662 
663 	vdc->dip	= dip;
664 	vdc->instance	= instance;
665 	vdc->vdisk_type	= VD_DISK_TYPE_UNK;
666 	vdc->vdisk_label = VD_DISK_LABEL_UNK;
667 	vdc->state	= VDC_STATE_INIT;
668 	vdc->lifecycle	= VDC_LC_ATTACHING;
669 	vdc->session_id = 0;
670 	vdc->vdisk_bsize = DEV_BSIZE;
671 	vdc->vio_bmask = 0;
672 	vdc->vio_bshift = 0;
673 	vdc->max_xfer_sz = maxphys / vdc->vdisk_bsize;
674 
675 	/*
676 	 * We assume, for now, that the vDisk server will export 'read'
677 	 * operations to us at a minimum (this is needed because of checks
678 	 * in vdc for supported operations early in the handshake process).
679 	 * The vDisk server will return ENOTSUP if this is not the case.
680 	 * The value will be overwritten during the attribute exchange with
681 	 * the bitmask of operations exported by server.
682 	 */
683 	vdc->operations = VD_OP_MASK_READ;
684 
685 	vdc->vtoc = NULL;
686 	vdc->geom = NULL;
687 	vdc->cinfo = NULL;
688 	vdc->minfo = NULL;
689 
690 	mutex_init(&vdc->lock, NULL, MUTEX_DRIVER, NULL);
691 	cv_init(&vdc->initwait_cv, NULL, CV_DRIVER, NULL);
692 	cv_init(&vdc->dring_free_cv, NULL, CV_DRIVER, NULL);
693 	cv_init(&vdc->membind_cv, NULL, CV_DRIVER, NULL);
694 	cv_init(&vdc->running_cv, NULL, CV_DRIVER, NULL);
695 	cv_init(&vdc->io_pending_cv, NULL, CV_DRIVER, NULL);
696 
697 	vdc->io_pending = B_FALSE;
698 	vdc->threads_pending = 0;
699 	vdc->sync_op_blocked = B_FALSE;
700 	cv_init(&vdc->sync_blocked_cv, NULL, CV_DRIVER, NULL);
701 
702 	mutex_init(&vdc->ownership_lock, NULL, MUTEX_DRIVER, NULL);
703 	cv_init(&vdc->ownership_cv, NULL, CV_DRIVER, NULL);
704 	cv_init(&vdc->eio_cv, NULL, CV_DRIVER, NULL);
705 
706 	/* init blocking msg read functionality */
707 	mutex_init(&vdc->read_lock, NULL, MUTEX_DRIVER, NULL);
708 	cv_init(&vdc->read_cv, NULL, CV_DRIVER, NULL);
709 	vdc->read_state = VDC_READ_IDLE;
710 
711 	vdc->initialized |= VDC_LOCKS;
712 
713 	/* get device and port MD node for this disk instance */
714 	if (vdc_get_md_node(dip, &mdp, &vd_node) != 0) {
715 		cmn_err(CE_NOTE, "[%d] Could not get machine description node",
716 		    instance);
717 		return (DDI_FAILURE);
718 	}
719 
720 	if (vdc_init_ports(vdc, mdp, vd_node) != 0) {
721 		cmn_err(CE_NOTE, "[%d] Error initialising ports", instance);
722 		return (DDI_FAILURE);
723 	}
724 
725 	(void) md_fini_handle(mdp);
726 
727 	/* Create the kstats for saving the I/O statistics used by iostat(8) */
728 	vdc_create_io_kstats(vdc);
729 	vdc_create_err_kstats(vdc);
730 
731 	/* Initialize remaining structures before starting the msg thread */
732 	vdc->vdisk_label = VD_DISK_LABEL_UNK;
733 	vdc->vtoc = kmem_zalloc(sizeof (struct extvtoc), KM_SLEEP);
734 	vdc->geom = kmem_zalloc(sizeof (struct dk_geom), KM_SLEEP);
735 	vdc->minfo = kmem_zalloc(sizeof (struct dk_minfo), KM_SLEEP);
736 
737 	/* initialize the thread responsible for managing state with server */
738 	vdc->msg_proc_thr = thread_create(NULL, 0, vdc_process_msg_thread,
739 	    vdc, 0, &p0, TS_RUN, minclsyspri);
740 	if (vdc->msg_proc_thr == NULL) {
741 		cmn_err(CE_NOTE, "[%d] Failed to create msg processing thread",
742 		    instance);
743 		return (DDI_FAILURE);
744 	}
745 
746 	/*
747 	 * If there are multiple servers then start the eio thread.
748 	 */
749 	if (vdc->num_servers > 1) {
750 		vdc->eio_thread = thread_create(NULL, 0, vdc_eio_thread, vdc, 0,
751 		    &p0, TS_RUN, v.v_maxsyspri - 2);
752 		if (vdc->eio_thread == NULL) {
753 			cmn_err(CE_NOTE, "[%d] Failed to create error "
754 			    "I/O thread", instance);
755 			return (DDI_FAILURE);
756 		}
757 	}
758 
759 	vdc->initialized |= VDC_THREAD;
760 
761 	atomic_inc_32(&vdc_instance_count);
762 
763 	/*
764 	 * Check the disk label. This will send requests and do the handshake.
765 	 * We don't really care about the disk label now. What we really need is
766 	 * the handshake do be done so that we know the type of the disk (slice
767 	 * or full disk) and the appropriate device nodes can be created.
768 	 */
769 
770 	mutex_enter(&vdc->lock);
771 	(void) vdc_validate_geometry(vdc);
772 	mutex_exit(&vdc->lock);
773 
774 	/*
775 	 * Now that we have the device info we can create the device nodes
776 	 */
777 	status = vdc_create_device_nodes(vdc);
778 	if (status) {
779 		DMSG(vdc, 0, "[%d] Failed to create device nodes",
780 		    instance);
781 		goto return_status;
782 	}
783 
784 	/*
785 	 * Fill in the fields of the error statistics kstat that were not
786 	 * available when creating the kstat
787 	 */
788 	vdc_set_err_kstats(vdc);
789 	ddi_report_dev(dip);
790 	ASSERT(vdc->lifecycle == VDC_LC_ONLINE ||
791 	    vdc->lifecycle == VDC_LC_ONLINE_PENDING);
792 	DMSG(vdc, 0, "[%d] Attach tasks successful\n", instance);
793 
794 return_status:
795 	DMSG(vdc, 0, "[%d] Attach completed\n", instance);
796 	return (status);
797 }
798 
799 static int
800 vdc_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
801 {
802 	int	status;
803 
804 	switch (cmd) {
805 	case DDI_ATTACH:
806 		if ((status = vdc_do_attach(dip)) != 0)
807 			(void) vdc_detach(dip, DDI_DETACH);
808 		return (status);
809 	case DDI_RESUME:
810 		/* nothing to do for this non-device */
811 		return (DDI_SUCCESS);
812 	default:
813 		return (DDI_FAILURE);
814 	}
815 }
816 
817 static int
818 vdc_do_ldc_init(vdc_t *vdc, vdc_server_t *srvr)
819 {
820 	int			status = 0;
821 	ldc_status_t		ldc_state;
822 	ldc_attr_t		ldc_attr;
823 
824 	ASSERT(vdc != NULL);
825 	ASSERT(srvr != NULL);
826 
827 	ldc_attr.devclass = LDC_DEV_BLK;
828 	ldc_attr.instance = vdc->instance;
829 	ldc_attr.mode = LDC_MODE_UNRELIABLE;	/* unreliable transport */
830 	ldc_attr.mtu = VD_LDC_MTU;
831 
832 	if ((srvr->state & VDC_LDC_INIT) == 0) {
833 		status = ldc_init(srvr->ldc_id, &ldc_attr,
834 		    &srvr->ldc_handle);
835 		if (status != 0) {
836 			DMSG(vdc, 0, "[%d] ldc_init(chan %ld) returned %d",
837 			    vdc->instance, srvr->ldc_id, status);
838 			return (status);
839 		}
840 		srvr->state |= VDC_LDC_INIT;
841 	}
842 	status = ldc_status(srvr->ldc_handle, &ldc_state);
843 	if (status != 0) {
844 		DMSG(vdc, 0, "[%d] Cannot discover LDC status [err=%d]",
845 		    vdc->instance, status);
846 		goto init_exit;
847 	}
848 	srvr->ldc_state = ldc_state;
849 
850 	if ((srvr->state & VDC_LDC_CB) == 0) {
851 		status = ldc_reg_callback(srvr->ldc_handle, vdc_handle_cb,
852 		    (caddr_t)srvr);
853 		if (status != 0) {
854 			DMSG(vdc, 0, "[%d] LDC callback reg. failed (%d)",
855 			    vdc->instance, status);
856 			goto init_exit;
857 		}
858 		srvr->state |= VDC_LDC_CB;
859 	}
860 
861 	/*
862 	 * At this stage we have initialised LDC, we will now try and open
863 	 * the connection.
864 	 */
865 	if (srvr->ldc_state == LDC_INIT) {
866 		status = ldc_open(srvr->ldc_handle);
867 		if (status != 0) {
868 			DMSG(vdc, 0, "[%d] ldc_open(chan %ld) returned %d",
869 			    vdc->instance, srvr->ldc_id, status);
870 			goto init_exit;
871 		}
872 		srvr->state |= VDC_LDC_OPEN;
873 	}
874 
875 init_exit:
876 	if (status) {
877 		vdc_terminate_ldc(vdc, srvr);
878 	}
879 
880 	return (status);
881 }
882 
883 static int
884 vdc_start_ldc_connection(vdc_t *vdc)
885 {
886 	int		status = 0;
887 
888 	ASSERT(vdc != NULL);
889 
890 	ASSERT(MUTEX_HELD(&vdc->lock));
891 
892 	status = vdc_do_ldc_up(vdc);
893 
894 	DMSG(vdc, 0, "[%d] Finished bringing up LDC\n", vdc->instance);
895 
896 	return (status);
897 }
898 
899 static int
900 vdc_stop_ldc_connection(vdc_t *vdcp)
901 {
902 	int	status;
903 
904 	ASSERT(vdcp != NULL);
905 
906 	ASSERT(MUTEX_HELD(&vdcp->lock));
907 
908 	DMSG(vdcp, 0, ": Resetting connection to vDisk server : state %d\n",
909 	    vdcp->state);
910 
911 	status = ldc_down(vdcp->curr_server->ldc_handle);
912 	DMSG(vdcp, 0, "ldc_down() = %d\n", status);
913 
914 	vdcp->initialized &= ~VDC_HANDSHAKE;
915 	DMSG(vdcp, 0, "initialized=%x\n", vdcp->initialized);
916 
917 	return (status);
918 }
919 
920 static void
921 vdc_create_io_kstats(vdc_t *vdc)
922 {
923 	if (vdc->io_stats != NULL) {
924 		DMSG(vdc, 0, "[%d] I/O kstat already exists\n", vdc->instance);
925 		return;
926 	}
927 
928 	vdc->io_stats = kstat_create(VDC_DRIVER_NAME, vdc->instance, NULL,
929 	    "disk", KSTAT_TYPE_IO, 1, KSTAT_FLAG_PERSISTENT);
930 	if (vdc->io_stats != NULL) {
931 		vdc->io_stats->ks_lock = &vdc->lock;
932 		kstat_install(vdc->io_stats);
933 	} else {
934 		cmn_err(CE_NOTE, "[%d] Failed to create kstat: I/O statistics"
935 		    " will not be gathered", vdc->instance);
936 	}
937 }
938 
939 static void
940 vdc_create_err_kstats(vdc_t *vdc)
941 {
942 	vd_err_stats_t	*stp;
943 	char	kstatmodule_err[KSTAT_STRLEN];
944 	char	kstatname[KSTAT_STRLEN];
945 	int	ndata = (sizeof (vd_err_stats_t) / sizeof (kstat_named_t));
946 	int	instance = vdc->instance;
947 
948 	if (vdc->err_stats != NULL) {
949 		DMSG(vdc, 0, "[%d] ERR kstat already exists\n", vdc->instance);
950 		return;
951 	}
952 
953 	(void) snprintf(kstatmodule_err, sizeof (kstatmodule_err),
954 	    "%serr", VDC_DRIVER_NAME);
955 	(void) snprintf(kstatname, sizeof (kstatname),
956 	    "%s%d,err", VDC_DRIVER_NAME, instance);
957 
958 	vdc->err_stats = kstat_create(kstatmodule_err, instance, kstatname,
959 	    "device_error", KSTAT_TYPE_NAMED, ndata, KSTAT_FLAG_PERSISTENT);
960 
961 	if (vdc->err_stats == NULL) {
962 		cmn_err(CE_NOTE, "[%d] Failed to create kstat: Error statistics"
963 		    " will not be gathered", instance);
964 		return;
965 	}
966 
967 	stp = (vd_err_stats_t *)vdc->err_stats->ks_data;
968 	kstat_named_init(&stp->vd_softerrs,	"Soft Errors",
969 	    KSTAT_DATA_UINT32);
970 	kstat_named_init(&stp->vd_transerrs,	"Transport Errors",
971 	    KSTAT_DATA_UINT32);
972 	kstat_named_init(&stp->vd_protoerrs,	"Protocol Errors",
973 	    KSTAT_DATA_UINT32);
974 	kstat_named_init(&stp->vd_vid,		"Vendor",
975 	    KSTAT_DATA_CHAR);
976 	kstat_named_init(&stp->vd_pid,		"Product",
977 	    KSTAT_DATA_CHAR);
978 	kstat_named_init(&stp->vd_capacity,	"Size",
979 	    KSTAT_DATA_ULONGLONG);
980 
981 	vdc->err_stats->ks_update  = nulldev;
982 
983 	kstat_install(vdc->err_stats);
984 }
985 
986 static void
987 vdc_set_err_kstats(vdc_t *vdc)
988 {
989 	vd_err_stats_t  *stp;
990 
991 	if (vdc->err_stats == NULL)
992 		return;
993 
994 	mutex_enter(&vdc->lock);
995 
996 	stp = (vd_err_stats_t *)vdc->err_stats->ks_data;
997 	ASSERT(stp != NULL);
998 
999 	stp->vd_capacity.value.ui64 = vdc->vdisk_size * vdc->vdisk_bsize;
1000 	(void) strcpy(stp->vd_vid.value.c, "SUN");
1001 	(void) strcpy(stp->vd_pid.value.c, "VDSK");
1002 
1003 	mutex_exit(&vdc->lock);
1004 }
1005 
1006 static int
1007 vdc_create_device_nodes_efi(vdc_t *vdc)
1008 {
1009 	ddi_remove_minor_node(vdc->dip, "h");
1010 	ddi_remove_minor_node(vdc->dip, "h,raw");
1011 
1012 	if (ddi_create_minor_node(vdc->dip, "wd", S_IFBLK,
1013 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
1014 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
1015 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'wd'",
1016 		    vdc->instance);
1017 		return (EIO);
1018 	}
1019 
1020 	/* if any device node is created we set this flag */
1021 	vdc->initialized |= VDC_MINOR;
1022 
1023 	if (ddi_create_minor_node(vdc->dip, "wd,raw", S_IFCHR,
1024 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
1025 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
1026 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'wd,raw'",
1027 		    vdc->instance);
1028 		return (EIO);
1029 	}
1030 
1031 	return (0);
1032 }
1033 
1034 static int
1035 vdc_create_device_nodes_vtoc(vdc_t *vdc)
1036 {
1037 	ddi_remove_minor_node(vdc->dip, "wd");
1038 	ddi_remove_minor_node(vdc->dip, "wd,raw");
1039 
1040 	if (ddi_create_minor_node(vdc->dip, "h", S_IFBLK,
1041 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
1042 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
1043 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'h'",
1044 		    vdc->instance);
1045 		return (EIO);
1046 	}
1047 
1048 	/* if any device node is created we set this flag */
1049 	vdc->initialized |= VDC_MINOR;
1050 
1051 	if (ddi_create_minor_node(vdc->dip, "h,raw", S_IFCHR,
1052 	    VD_MAKE_DEV(vdc->instance, VD_EFI_WD_SLICE),
1053 	    DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
1054 		cmn_err(CE_NOTE, "[%d] Couldn't add block node 'h,raw'",
1055 		    vdc->instance);
1056 		return (EIO);
1057 	}
1058 
1059 	return (0);
1060 }
1061 
1062 /*
1063  * Function:
1064  *	vdc_create_device_nodes
1065  *
1066  * Description:
1067  *	This function creates the block and character device nodes under
1068  *	/devices. It is called as part of the attach(9E) of the instance
1069  *	during the handshake with vds after vds has sent the attributes
1070  *	to vdc.
1071  *
1072  *	If the device is of type VD_DISK_TYPE_SLICE then the minor node
1073  *	of 2 is used in keeping with the Solaris convention that slice 2
1074  *	refers to a whole disk. Slices start at 'a'
1075  *
1076  * Parameters:
1077  *	vdc		- soft state pointer
1078  *
1079  * Return Values
1080  *	0		- Success
1081  *	EIO		- Failed to create node
1082  */
1083 static int
1084 vdc_create_device_nodes(vdc_t *vdc)
1085 {
1086 	char		name[sizeof ("s,raw")];
1087 	dev_info_t	*dip = NULL;
1088 	int		instance, status;
1089 	int		num_slices = 1;
1090 	int		i;
1091 
1092 	ASSERT(vdc != NULL);
1093 
1094 	instance = vdc->instance;
1095 	dip = vdc->dip;
1096 
1097 	switch (vdc->vdisk_type) {
1098 	case VD_DISK_TYPE_DISK:
1099 	case VD_DISK_TYPE_UNK:
1100 		num_slices = V_NUMPAR;
1101 		break;
1102 	case VD_DISK_TYPE_SLICE:
1103 		num_slices = 1;
1104 		break;
1105 	default:
1106 		ASSERT(0);
1107 	}
1108 
1109 	/*
1110 	 * Minor nodes are different for EFI disks: EFI disks do not have
1111 	 * a minor node 'g' for the minor number corresponding to slice
1112 	 * VD_EFI_WD_SLICE (slice 7) instead they have a minor node 'wd'
1113 	 * representing the whole disk.
1114 	 */
1115 	for (i = 0; i < num_slices; i++) {
1116 
1117 		if (i == VD_EFI_WD_SLICE) {
1118 			if (vdc->vdisk_label == VD_DISK_LABEL_EFI)
1119 				status = vdc_create_device_nodes_efi(vdc);
1120 			else
1121 				status = vdc_create_device_nodes_vtoc(vdc);
1122 			if (status != 0)
1123 				return (status);
1124 			continue;
1125 		}
1126 
1127 		(void) snprintf(name, sizeof (name), "%c", 'a' + i);
1128 		if (ddi_create_minor_node(dip, name, S_IFBLK,
1129 		    VD_MAKE_DEV(instance, i), DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
1130 			cmn_err(CE_NOTE, "[%d] Couldn't add block node '%s'",
1131 			    instance, name);
1132 			return (EIO);
1133 		}
1134 
1135 		/* if any device node is created we set this flag */
1136 		vdc->initialized |= VDC_MINOR;
1137 
1138 		(void) snprintf(name, sizeof (name), "%c%s", 'a' + i, ",raw");
1139 
1140 		if (ddi_create_minor_node(dip, name, S_IFCHR,
1141 		    VD_MAKE_DEV(instance, i), DDI_NT_BLOCK, 0) != DDI_SUCCESS) {
1142 			cmn_err(CE_NOTE, "[%d] Couldn't add raw node '%s'",
1143 			    instance, name);
1144 			return (EIO);
1145 		}
1146 	}
1147 
1148 	return (0);
1149 }
1150 
1151 /*
1152  * Driver prop_op(9e) entry point function. Return the number of blocks for
1153  * the partition in question or forward the request to the property facilities.
1154  */
1155 static int
1156 vdc_prop_op(dev_t dev, dev_info_t *dip, ddi_prop_op_t prop_op, int mod_flags,
1157     char *name, caddr_t valuep, int *lengthp)
1158 {
1159 	int instance = ddi_get_instance(dip);
1160 	vdc_t *vdc;
1161 	uint64_t nblocks;
1162 	uint_t blksize;
1163 
1164 	vdc = ddi_get_soft_state(vdc_state, instance);
1165 
1166 	if (dev == DDI_DEV_T_ANY || vdc == NULL) {
1167 		return (ddi_prop_op(dev, dip, prop_op, mod_flags,
1168 		    name, valuep, lengthp));
1169 	}
1170 
1171 	mutex_enter(&vdc->lock);
1172 	(void) vdc_validate_geometry(vdc);
1173 	if (vdc->vdisk_label == VD_DISK_LABEL_UNK) {
1174 		mutex_exit(&vdc->lock);
1175 		return (ddi_prop_op(dev, dip, prop_op, mod_flags,
1176 		    name, valuep, lengthp));
1177 	}
1178 	nblocks = vdc->slice[VDCPART(dev)].nblocks;
1179 	blksize = vdc->vdisk_bsize;
1180 	mutex_exit(&vdc->lock);
1181 
1182 	return (ddi_prop_op_nblocks_blksize(dev, dip, prop_op, mod_flags,
1183 	    name, valuep, lengthp, nblocks, blksize));
1184 }
1185 
1186 /*
1187  * Function:
1188  *	vdc_is_opened
1189  *
1190  * Description:
1191  *	This function checks if any slice of a given virtual disk is
1192  *	currently opened.
1193  *
1194  * Parameters:
1195  *	vdc		- soft state pointer
1196  *
1197  * Return Values
1198  *	B_TRUE		- at least one slice is opened.
1199  *	B_FALSE		- no slice is opened.
1200  */
1201 static boolean_t
1202 vdc_is_opened(vdc_t *vdc)
1203 {
1204 	int i;
1205 
1206 	/* check if there's any layered open */
1207 	for (i = 0; i < V_NUMPAR; i++) {
1208 		if (vdc->open_lyr[i] > 0)
1209 			return (B_TRUE);
1210 	}
1211 
1212 	/* check if there is any other kind of open */
1213 	for (i = 0; i < OTYPCNT; i++) {
1214 		if (vdc->open[i] != 0)
1215 			return (B_TRUE);
1216 	}
1217 
1218 	return (B_FALSE);
1219 }
1220 
1221 static int
1222 vdc_mark_opened(vdc_t *vdc, int slice, int flag, int otyp)
1223 {
1224 	uint8_t slicemask;
1225 	int i;
1226 
1227 	ASSERT(otyp < OTYPCNT);
1228 	ASSERT(slice < V_NUMPAR);
1229 	ASSERT(MUTEX_HELD(&vdc->lock));
1230 
1231 	slicemask = 1 << slice;
1232 
1233 	/*
1234 	 * If we have a single-slice disk which was unavailable during the
1235 	 * attach then a device was created for each 8 slices. Now that
1236 	 * the type is known, we prevent opening any slice other than 0
1237 	 * even if a device still exists.
1238 	 */
1239 	if (vdc->vdisk_type == VD_DISK_TYPE_SLICE && slice != 0)
1240 		return (EIO);
1241 
1242 	/* check if slice is already exclusively opened */
1243 	if (vdc->open_excl & slicemask)
1244 		return (EBUSY);
1245 
1246 	/* if open exclusive, check if slice is already opened */
1247 	if (flag & FEXCL) {
1248 		if (vdc->open_lyr[slice] > 0)
1249 			return (EBUSY);
1250 		for (i = 0; i < OTYPCNT; i++) {
1251 			if (vdc->open[i] & slicemask)
1252 				return (EBUSY);
1253 		}
1254 		vdc->open_excl |= slicemask;
1255 	}
1256 
1257 	/* mark slice as opened */
1258 	if (otyp == OTYP_LYR) {
1259 		vdc->open_lyr[slice]++;
1260 	} else {
1261 		vdc->open[otyp] |= slicemask;
1262 	}
1263 
1264 	return (0);
1265 }
1266 
1267 static void
1268 vdc_mark_closed(vdc_t *vdc, int slice, int flag, int otyp)
1269 {
1270 	uint8_t slicemask;
1271 
1272 	ASSERT(otyp < OTYPCNT);
1273 	ASSERT(slice < V_NUMPAR);
1274 	ASSERT(MUTEX_HELD(&vdc->lock));
1275 
1276 	slicemask = 1 << slice;
1277 
1278 	if (otyp == OTYP_LYR) {
1279 		ASSERT(vdc->open_lyr[slice] > 0);
1280 		vdc->open_lyr[slice]--;
1281 	} else {
1282 		vdc->open[otyp] &= ~slicemask;
1283 	}
1284 
1285 	if (flag & FEXCL)
1286 		vdc->open_excl &= ~slicemask;
1287 }
1288 
1289 static int
1290 vdc_open(dev_t *dev, int flag, int otyp, cred_t *cred)
1291 {
1292 	_NOTE(ARGUNUSED(cred))
1293 
1294 	int	instance, nodelay;
1295 	int	slice, status = 0;
1296 	vdc_t	*vdc;
1297 
1298 	ASSERT(dev != NULL);
1299 	instance = VDCUNIT(*dev);
1300 
1301 	if (otyp >= OTYPCNT)
1302 		return (EINVAL);
1303 
1304 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1305 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1306 		return (ENXIO);
1307 	}
1308 
1309 	DMSG(vdc, 0, "minor = %d flag = %x, otyp = %x\n",
1310 	    getminor(*dev), flag, otyp);
1311 
1312 	slice = VDCPART(*dev);
1313 
1314 	nodelay = flag & (FNDELAY | FNONBLOCK);
1315 
1316 	if ((flag & FWRITE) && (!nodelay) &&
1317 	    !(VD_OP_SUPPORTED(vdc->operations, VD_OP_BWRITE))) {
1318 		return (EROFS);
1319 	}
1320 
1321 	mutex_enter(&vdc->lock);
1322 
1323 	status = vdc_mark_opened(vdc, slice, flag, otyp);
1324 
1325 	if (status != 0) {
1326 		mutex_exit(&vdc->lock);
1327 		return (status);
1328 	}
1329 
1330 	/*
1331 	 * If the disk type is unknown then we have to wait for the
1332 	 * handshake to complete because we don't know if the slice
1333 	 * device we are opening effectively exists.
1334 	 */
1335 	if (vdc->vdisk_type != VD_DISK_TYPE_UNK && nodelay) {
1336 
1337 		/* don't resubmit a validate request if there's already one */
1338 		if (vdc->validate_pending > 0) {
1339 			mutex_exit(&vdc->lock);
1340 			return (0);
1341 		}
1342 
1343 		/* call vdc_validate() asynchronously to avoid blocking */
1344 		if (taskq_dispatch(system_taskq, vdc_validate_task,
1345 		    (void *)vdc, TQ_NOSLEEP) == TASKQID_INVALID) {
1346 			vdc_mark_closed(vdc, slice, flag, otyp);
1347 			mutex_exit(&vdc->lock);
1348 			return (ENXIO);
1349 		}
1350 
1351 		vdc->validate_pending++;
1352 		mutex_exit(&vdc->lock);
1353 		return (0);
1354 	}
1355 
1356 	mutex_exit(&vdc->lock);
1357 
1358 	vdc_validate(vdc);
1359 
1360 	mutex_enter(&vdc->lock);
1361 
1362 	if (vdc->vdisk_type == VD_DISK_TYPE_UNK ||
1363 	    (vdc->vdisk_type == VD_DISK_TYPE_SLICE && slice != 0) ||
1364 	    (!nodelay && (vdc->vdisk_label == VD_DISK_LABEL_UNK ||
1365 	    vdc->slice[slice].nblocks == 0))) {
1366 		vdc_mark_closed(vdc, slice, flag, otyp);
1367 		status = EIO;
1368 	}
1369 
1370 	mutex_exit(&vdc->lock);
1371 
1372 	return (status);
1373 }
1374 
1375 static int
1376 vdc_close(dev_t dev, int flag, int otyp, cred_t *cred)
1377 {
1378 	_NOTE(ARGUNUSED(cred))
1379 
1380 	int	instance;
1381 	int	slice;
1382 	int	rv, rval;
1383 	vdc_t	*vdc;
1384 
1385 	instance = VDCUNIT(dev);
1386 
1387 	if (otyp >= OTYPCNT)
1388 		return (EINVAL);
1389 
1390 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1391 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1392 		return (ENXIO);
1393 	}
1394 
1395 	DMSG(vdc, 0, "[%d] flag = %x, otyp = %x\n", instance, flag, otyp);
1396 
1397 	slice = VDCPART(dev);
1398 
1399 	/*
1400 	 * Attempt to flush the W$ on a close operation. If this is
1401 	 * not a supported IOCTL command or the backing device is read-only
1402 	 * do not fail the close operation.
1403 	 */
1404 	rv = vd_process_ioctl(dev, DKIOCFLUSHWRITECACHE, NULL, FKIOCTL, &rval);
1405 
1406 	if (rv != 0 && rv != ENOTSUP && rv != ENOTTY && rv != EROFS) {
1407 		DMSG(vdc, 0, "[%d] flush failed with error %d on close\n",
1408 		    instance, rv);
1409 		return (EIO);
1410 	}
1411 
1412 	mutex_enter(&vdc->lock);
1413 	vdc_mark_closed(vdc, slice, flag, otyp);
1414 	mutex_exit(&vdc->lock);
1415 
1416 	return (0);
1417 }
1418 
1419 static int
1420 vdc_ioctl(dev_t dev, int cmd, intptr_t arg, int mode, cred_t *credp, int *rvalp)
1421 {
1422 	_NOTE(ARGUNUSED(credp))
1423 
1424 	return (vd_process_ioctl(dev, cmd, (caddr_t)arg, mode, rvalp));
1425 }
1426 
1427 static int
1428 vdc_print(dev_t dev, char *str)
1429 {
1430 	cmn_err(CE_NOTE, "vdc%d:  %s", VDCUNIT(dev), str);
1431 	return (0);
1432 }
1433 
1434 static int
1435 vdc_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblk)
1436 {
1437 	int	rv, flags;
1438 	size_t	nbytes = nblk * DEV_BSIZE;
1439 	int	instance = VDCUNIT(dev);
1440 	vdc_t	*vdc = NULL;
1441 	diskaddr_t vio_blkno;
1442 
1443 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1444 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1445 		return (ENXIO);
1446 	}
1447 
1448 	DMSG(vdc, 2, "[%d] dump %ld bytes at block 0x%lx : addr=0x%p\n",
1449 	    instance, nbytes, blkno, (void *)addr);
1450 
1451 	/* convert logical block to vio block */
1452 	if ((blkno & vdc->vio_bmask) != 0) {
1453 		DMSG(vdc, 0, "Misaligned block number (%lu)\n", blkno);
1454 		return (EINVAL);
1455 	}
1456 	vio_blkno = blkno >> vdc->vio_bshift;
1457 
1458 	/*
1459 	 * If we are panicking, we need the state to be "running" so that we
1460 	 * can submit I/Os, but we don't want to check for any backend error.
1461 	 */
1462 	flags = (ddi_in_panic())? VDC_OP_STATE_RUNNING : VDC_OP_NORMAL;
1463 
1464 	rv = vdc_do_op(vdc, VD_OP_BWRITE, addr, nbytes, VDCPART(dev),
1465 	    vio_blkno, NULL, VIO_write_dir, flags);
1466 
1467 	if (rv) {
1468 		DMSG(vdc, 0, "Failed to do a disk dump (err=%d)\n", rv);
1469 		return (rv);
1470 	}
1471 
1472 	DMSG(vdc, 0, "[%d] End\n", instance);
1473 
1474 	return (0);
1475 }
1476 
1477 /* -------------------------------------------------------------------------- */
1478 
1479 /*
1480  * Disk access routines
1481  *
1482  */
1483 
1484 /*
1485  * vdc_strategy()
1486  *
1487  * Return Value:
1488  *	0:	As per strategy(9E), the strategy() function must return 0
1489  *		[ bioerror(9f) sets b_flags to the proper error code ]
1490  */
1491 static int
1492 vdc_strategy(struct buf *buf)
1493 {
1494 	diskaddr_t vio_blkno;
1495 	vdc_t	*vdc = NULL;
1496 	int	instance = VDCUNIT(buf->b_edev);
1497 	int	op = (buf->b_flags & B_READ) ? VD_OP_BREAD : VD_OP_BWRITE;
1498 	int	slice;
1499 
1500 	if ((vdc = ddi_get_soft_state(vdc_state, instance)) == NULL) {
1501 		cmn_err(CE_NOTE, "[%d] Couldn't get state structure", instance);
1502 		bioerror(buf, ENXIO);
1503 		biodone(buf);
1504 		return (0);
1505 	}
1506 
1507 	DMSG(vdc, 2, "[%d] %s %ld bytes at block %llx : b_addr=0x%p\n",
1508 	    instance, (buf->b_flags & B_READ) ? "Read" : "Write",
1509 	    buf->b_bcount, buf->b_lblkno, (void *)buf->b_un.b_addr);
1510 
1511 	bp_mapin(buf);
1512 
1513 	if ((long)buf->b_private == VD_SLICE_NONE) {
1514 		/* I/O using an absolute disk offset */
1515 		slice = VD_SLICE_NONE;
1516 	} else {
1517 		slice = VDCPART(buf->b_edev);
1518 	}
1519 
1520 	/*
1521 	 * In the buf structure, b_lblkno represents a logical block number
1522 	 * using a block size of 512 bytes. For the VIO request, this block
1523 	 * number has to be converted to be represented with the block size
1524 	 * used by the VIO protocol.
1525 	 */
1526 	if ((buf->b_lblkno & vdc->vio_bmask) != 0) {
1527 		bioerror(buf, EINVAL);
1528 		biodone(buf);
1529 		return (0);
1530 	}
1531 	vio_blkno = buf->b_lblkno >> vdc->vio_bshift;
1532 
1533 	/* submit the I/O, any error will be reported in the buf structure */
1534 	(void) vdc_do_op(vdc, op, (caddr_t)buf->b_un.b_addr,
1535 	    buf->b_bcount, slice, vio_blkno,
1536 	    buf, (op == VD_OP_BREAD) ? VIO_read_dir : VIO_write_dir,
1537 	    VDC_OP_NORMAL);
1538 
1539 	return (0);
1540 }
1541 
1542 /*
1543  * Function:
1544  *	vdc_min
1545  *
1546  * Description:
1547  *	Routine to limit the size of a data transfer. Used in
1548  *	conjunction with physio(9F).
1549  *
1550  * Arguments:
1551  *	bp - pointer to the indicated buf(9S) struct.
1552  *
1553  */
1554 static void
1555 vdc_min(struct buf *bufp)
1556 {
1557 	vdc_t	*vdc = NULL;
1558 	int	instance = VDCUNIT(bufp->b_edev);
1559 
1560 	vdc = ddi_get_soft_state(vdc_state, instance);
1561 	VERIFY(vdc != NULL);
1562 
1563 	if (bufp->b_bcount > (vdc->max_xfer_sz * vdc->vdisk_bsize)) {
1564 		bufp->b_bcount = vdc->max_xfer_sz * vdc->vdisk_bsize;
1565 	}
1566 }
1567 
1568 static int
1569 vdc_read(dev_t dev, struct uio *uio, cred_t *cred)
1570 {
1571 	_NOTE(ARGUNUSED(cred))
1572 
1573 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1574 	return (physio(vdc_strategy, NULL, dev, B_READ, vdc_min, uio));
1575 }
1576 
1577 static int
1578 vdc_write(dev_t dev, struct uio *uio, cred_t *cred)
1579 {
1580 	_NOTE(ARGUNUSED(cred))
1581 
1582 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1583 	return (physio(vdc_strategy, NULL, dev, B_WRITE, vdc_min, uio));
1584 }
1585 
1586 static int
1587 vdc_aread(dev_t dev, struct aio_req *aio, cred_t *cred)
1588 {
1589 	_NOTE(ARGUNUSED(cred))
1590 
1591 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1592 	return (aphysio(vdc_strategy, anocancel, dev, B_READ, vdc_min, aio));
1593 }
1594 
1595 static int
1596 vdc_awrite(dev_t dev, struct aio_req *aio, cred_t *cred)
1597 {
1598 	_NOTE(ARGUNUSED(cred))
1599 
1600 	DMSGX(1, "[%d] Entered", VDCUNIT(dev));
1601 	return (aphysio(vdc_strategy, anocancel, dev, B_WRITE, vdc_min, aio));
1602 }
1603 
1604 
1605 /* -------------------------------------------------------------------------- */
1606 
1607 /*
1608  * Handshake support
1609  */
1610 
1611 
1612 /*
1613  * Function:
1614  *	vdc_init_ver_negotiation()
1615  *
1616  * Description:
1617  *
1618  * Arguments:
1619  *	vdc	- soft state pointer for this instance of the device driver.
1620  *
1621  * Return Code:
1622  *	0	- Success
1623  */
1624 static int
1625 vdc_init_ver_negotiation(vdc_t *vdc, vio_ver_t ver)
1626 {
1627 	vio_ver_msg_t	pkt;
1628 	size_t		msglen = sizeof (pkt);
1629 	int		status = -1;
1630 
1631 	ASSERT(vdc != NULL);
1632 	ASSERT(mutex_owned(&vdc->lock));
1633 
1634 	DMSG(vdc, 0, "[%d] Entered.\n", vdc->instance);
1635 
1636 	/*
1637 	 * set the Session ID to a unique value
1638 	 * (the lower 32 bits of the clock tick)
1639 	 */
1640 	vdc->session_id = ((uint32_t)gettick() & 0xffffffff);
1641 	DMSG(vdc, 0, "[%d] Set SID to 0x%lx\n", vdc->instance, vdc->session_id);
1642 
1643 	pkt.tag.vio_msgtype = VIO_TYPE_CTRL;
1644 	pkt.tag.vio_subtype = VIO_SUBTYPE_INFO;
1645 	pkt.tag.vio_subtype_env = VIO_VER_INFO;
1646 	pkt.tag.vio_sid = vdc->session_id;
1647 	pkt.dev_class = VDEV_DISK;
1648 	pkt.ver_major = ver.major;
1649 	pkt.ver_minor = ver.minor;
1650 
1651 	status = vdc_send(vdc, (caddr_t)&pkt, &msglen);
1652 	DMSG(vdc, 0, "[%d] Ver info sent (status = %d)\n",
1653 	    vdc->instance, status);
1654 	if ((status != 0) || (msglen != sizeof (vio_ver_msg_t))) {
1655 		DMSG(vdc, 0, "[%d] Failed to send Ver negotiation info: "
1656 		    "id(%lx) rv(%d) size(%ld)", vdc->instance,
1657 		    vdc->curr_server->ldc_handle, status, msglen);
1658 		if (msglen != sizeof (vio_ver_msg_t))
1659 			status = ENOMSG;
1660 	}
1661 
1662 	return (status);
1663 }
1664 
1665 /*
1666  * Function:
1667  *	vdc_ver_negotiation()
1668  *
1669  * Description:
1670  *
1671  * Arguments:
1672  *	vdcp	- soft state pointer for this instance of the device driver.
1673  *
1674  * Return Code:
1675  *	0	- Success
1676  */
1677 static int
1678 vdc_ver_negotiation(vdc_t *vdcp)
1679 {
1680 	vio_msg_t vio_msg;
1681 	int status;
1682 
1683 	if (status = vdc_init_ver_negotiation(vdcp, vdc_version[0]))
1684 		return (status);
1685 
1686 	/* release lock and wait for response */
1687 	mutex_exit(&vdcp->lock);
1688 	status = vdc_wait_for_response(vdcp, &vio_msg);
1689 	mutex_enter(&vdcp->lock);
1690 	if (status) {
1691 		DMSG(vdcp, 0,
1692 		    "[%d] Failed waiting for Ver negotiation response, rv(%d)",
1693 		    vdcp->instance, status);
1694 		return (status);
1695 	}
1696 
1697 	/* check type and sub_type ... */
1698 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1699 	    vio_msg.tag.vio_subtype == VIO_SUBTYPE_INFO) {
1700 		DMSG(vdcp, 0, "[%d] Invalid ver negotiation response\n",
1701 		    vdcp->instance);
1702 		return (EPROTO);
1703 	}
1704 
1705 	return (vdc_handle_ver_msg(vdcp, (vio_ver_msg_t *)&vio_msg));
1706 }
1707 
1708 /*
1709  * Function:
1710  *	vdc_init_attr_negotiation()
1711  *
1712  * Description:
1713  *
1714  * Arguments:
1715  *	vdc	- soft state pointer for this instance of the device driver.
1716  *
1717  * Return Code:
1718  *	0	- Success
1719  */
1720 static int
1721 vdc_init_attr_negotiation(vdc_t *vdc)
1722 {
1723 	vd_attr_msg_t	pkt;
1724 	size_t		msglen = sizeof (pkt);
1725 	int		status;
1726 
1727 	ASSERT(vdc != NULL);
1728 	ASSERT(mutex_owned(&vdc->lock));
1729 
1730 	DMSG(vdc, 0, "[%d] entered\n", vdc->instance);
1731 
1732 	/* fill in tag */
1733 	pkt.tag.vio_msgtype = VIO_TYPE_CTRL;
1734 	pkt.tag.vio_subtype = VIO_SUBTYPE_INFO;
1735 	pkt.tag.vio_subtype_env = VIO_ATTR_INFO;
1736 	pkt.tag.vio_sid = vdc->session_id;
1737 	/* fill in payload */
1738 	pkt.max_xfer_sz = vdc->max_xfer_sz;
1739 	pkt.vdisk_block_size = vdc->vdisk_bsize;
1740 	pkt.xfer_mode = VIO_DRING_MODE_V1_0;
1741 	pkt.operations = 0;	/* server will set bits of valid operations */
1742 	pkt.vdisk_type = 0;	/* server will set to valid device type */
1743 	pkt.vdisk_media = 0;	/* server will set to valid media type */
1744 	pkt.vdisk_size = 0;	/* server will set to valid size */
1745 
1746 	status = vdc_send(vdc, (caddr_t)&pkt, &msglen);
1747 	DMSG(vdc, 0, "Attr info sent (status = %d)\n", status);
1748 
1749 	if ((status != 0) || (msglen != sizeof (vd_attr_msg_t))) {
1750 		DMSG(vdc, 0, "[%d] Failed to send Attr negotiation info: "
1751 		    "id(%lx) rv(%d) size(%ld)", vdc->instance,
1752 		    vdc->curr_server->ldc_handle, status, msglen);
1753 		if (msglen != sizeof (vd_attr_msg_t))
1754 			status = ENOMSG;
1755 	}
1756 
1757 	return (status);
1758 }
1759 
1760 /*
1761  * Function:
1762  *	vdc_attr_negotiation()
1763  *
1764  * Description:
1765  *
1766  * Arguments:
1767  *	vdc	- soft state pointer for this instance of the device driver.
1768  *
1769  * Return Code:
1770  *	0	- Success
1771  */
1772 static int
1773 vdc_attr_negotiation(vdc_t *vdcp)
1774 {
1775 	int status;
1776 	vio_msg_t vio_msg;
1777 
1778 	if (status = vdc_init_attr_negotiation(vdcp))
1779 		return (status);
1780 
1781 	/* release lock and wait for response */
1782 	mutex_exit(&vdcp->lock);
1783 	status = vdc_wait_for_response(vdcp, &vio_msg);
1784 	mutex_enter(&vdcp->lock);
1785 	if (status) {
1786 		DMSG(vdcp, 0,
1787 		    "[%d] Failed waiting for Attr negotiation response, rv(%d)",
1788 		    vdcp->instance, status);
1789 		return (status);
1790 	}
1791 
1792 	/* check type and sub_type ... */
1793 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1794 	    vio_msg.tag.vio_subtype == VIO_SUBTYPE_INFO) {
1795 		DMSG(vdcp, 0, "[%d] Invalid attr negotiation response\n",
1796 		    vdcp->instance);
1797 		return (EPROTO);
1798 	}
1799 
1800 	return (vdc_handle_attr_msg(vdcp, (vd_attr_msg_t *)&vio_msg));
1801 }
1802 
1803 
1804 /*
1805  * Function:
1806  *	vdc_init_dring_negotiate()
1807  *
1808  * Description:
1809  *
1810  * Arguments:
1811  *	vdc	- soft state pointer for this instance of the device driver.
1812  *
1813  * Return Code:
1814  *	0	- Success
1815  */
1816 static int
1817 vdc_init_dring_negotiate(vdc_t *vdc)
1818 {
1819 	vio_dring_reg_msg_t	pkt;
1820 	size_t			msglen = sizeof (pkt);
1821 	int			status = -1;
1822 	int			retry;
1823 	int			nretries = 10;
1824 
1825 	ASSERT(vdc != NULL);
1826 	ASSERT(mutex_owned(&vdc->lock));
1827 
1828 	for (retry = 0; retry < nretries; retry++) {
1829 		status = vdc_init_descriptor_ring(vdc);
1830 		if (status != EAGAIN)
1831 			break;
1832 		drv_usecwait(vdc_min_timeout_ldc);
1833 	}
1834 
1835 	if (status != 0) {
1836 		DMSG(vdc, 0, "[%d] Failed to init DRing (status = %d)\n",
1837 		    vdc->instance, status);
1838 		return (status);
1839 	}
1840 
1841 	DMSG(vdc, 0, "[%d] Init of descriptor ring completed (status = %d)\n",
1842 	    vdc->instance, status);
1843 
1844 	/* fill in tag */
1845 	pkt.tag.vio_msgtype = VIO_TYPE_CTRL;
1846 	pkt.tag.vio_subtype = VIO_SUBTYPE_INFO;
1847 	pkt.tag.vio_subtype_env = VIO_DRING_REG;
1848 	pkt.tag.vio_sid = vdc->session_id;
1849 	/* fill in payload */
1850 	pkt.dring_ident = 0;
1851 	pkt.num_descriptors = vdc->dring_len;
1852 	pkt.descriptor_size = vdc->dring_entry_size;
1853 	pkt.options = (VIO_TX_DRING | VIO_RX_DRING);
1854 	pkt.ncookies = vdc->dring_cookie_count;
1855 	pkt.cookie[0] = vdc->dring_cookie[0];	/* for now just one cookie */
1856 
1857 	status = vdc_send(vdc, (caddr_t)&pkt, &msglen);
1858 	if (status != 0) {
1859 		DMSG(vdc, 0, "[%d] Failed to register DRing (err = %d)",
1860 		    vdc->instance, status);
1861 	}
1862 
1863 	return (status);
1864 }
1865 
1866 
1867 /*
1868  * Function:
1869  *	vdc_dring_negotiation()
1870  *
1871  * Description:
1872  *
1873  * Arguments:
1874  *	vdc	- soft state pointer for this instance of the device driver.
1875  *
1876  * Return Code:
1877  *	0	- Success
1878  */
1879 static int
1880 vdc_dring_negotiation(vdc_t *vdcp)
1881 {
1882 	int status;
1883 	vio_msg_t vio_msg;
1884 
1885 	if (status = vdc_init_dring_negotiate(vdcp))
1886 		return (status);
1887 
1888 	/* release lock and wait for response */
1889 	mutex_exit(&vdcp->lock);
1890 	status = vdc_wait_for_response(vdcp, &vio_msg);
1891 	mutex_enter(&vdcp->lock);
1892 	if (status) {
1893 		DMSG(vdcp, 0,
1894 		    "[%d] Failed waiting for Dring negotiation response,"
1895 		    " rv(%d)", vdcp->instance, status);
1896 		return (status);
1897 	}
1898 
1899 	/* check type and sub_type ... */
1900 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
1901 	    vio_msg.tag.vio_subtype == VIO_SUBTYPE_INFO) {
1902 		DMSG(vdcp, 0, "[%d] Invalid Dring negotiation response\n",
1903 		    vdcp->instance);
1904 		return (EPROTO);
1905 	}
1906 
1907 	return (vdc_handle_dring_reg_msg(vdcp,
1908 	    (vio_dring_reg_msg_t *)&vio_msg));
1909 }
1910 
1911 
1912 /*
1913  * Function:
1914  *	vdc_send_rdx()
1915  *
1916  * Description:
1917  *
1918  * Arguments:
1919  *	vdc	- soft state pointer for this instance of the device driver.
1920  *
1921  * Return Code:
1922  *	0	- Success
1923  */
1924 static int
1925 vdc_send_rdx(vdc_t *vdcp)
1926 {
1927 	vio_msg_t	msg;
1928 	size_t		msglen = sizeof (vio_msg_t);
1929 	int		status;
1930 
1931 	/*
1932 	 * Send an RDX message to vds to indicate we are ready
1933 	 * to send data
1934 	 */
1935 	msg.tag.vio_msgtype = VIO_TYPE_CTRL;
1936 	msg.tag.vio_subtype = VIO_SUBTYPE_INFO;
1937 	msg.tag.vio_subtype_env = VIO_RDX;
1938 	msg.tag.vio_sid = vdcp->session_id;
1939 	status = vdc_send(vdcp, (caddr_t)&msg, &msglen);
1940 	if (status != 0) {
1941 		DMSG(vdcp, 0, "[%d] Failed to send RDX message (%d)",
1942 		    vdcp->instance, status);
1943 	}
1944 
1945 	return (status);
1946 }
1947 
1948 /*
1949  * Function:
1950  *	vdc_handle_rdx()
1951  *
1952  * Description:
1953  *
1954  * Arguments:
1955  *	vdc	- soft state pointer for this instance of the device driver.
1956  *	msgp	- received msg
1957  *
1958  * Return Code:
1959  *	0	- Success
1960  */
1961 static int
1962 vdc_handle_rdx(vdc_t *vdcp, vio_rdx_msg_t *msgp)
1963 {
1964 	_NOTE(ARGUNUSED(vdcp))
1965 	_NOTE(ARGUNUSED(msgp))
1966 
1967 	ASSERT(msgp->tag.vio_msgtype == VIO_TYPE_CTRL);
1968 	ASSERT(msgp->tag.vio_subtype == VIO_SUBTYPE_ACK);
1969 	ASSERT(msgp->tag.vio_subtype_env == VIO_RDX);
1970 
1971 	DMSG(vdcp, 1, "[%d] Got an RDX msg", vdcp->instance);
1972 
1973 	return (0);
1974 }
1975 
1976 /*
1977  * Function:
1978  *	vdc_rdx_exchange()
1979  *
1980  * Description:
1981  *
1982  * Arguments:
1983  *	vdc	- soft state pointer for this instance of the device driver.
1984  *
1985  * Return Code:
1986  *	0	- Success
1987  */
1988 static int
1989 vdc_rdx_exchange(vdc_t *vdcp)
1990 {
1991 	int status;
1992 	vio_msg_t vio_msg;
1993 
1994 	if (status = vdc_send_rdx(vdcp))
1995 		return (status);
1996 
1997 	/* release lock and wait for response */
1998 	mutex_exit(&vdcp->lock);
1999 	status = vdc_wait_for_response(vdcp, &vio_msg);
2000 	mutex_enter(&vdcp->lock);
2001 	if (status) {
2002 		DMSG(vdcp, 0, "[%d] Failed waiting for RDX response, rv(%d)",
2003 		    vdcp->instance, status);
2004 		return (status);
2005 	}
2006 
2007 	/* check type and sub_type ... */
2008 	if (vio_msg.tag.vio_msgtype != VIO_TYPE_CTRL ||
2009 	    vio_msg.tag.vio_subtype != VIO_SUBTYPE_ACK) {
2010 		DMSG(vdcp, 0, "[%d] Invalid RDX response\n", vdcp->instance);
2011 		return (EPROTO);
2012 	}
2013 
2014 	return (vdc_handle_rdx(vdcp, (vio_rdx_msg_t *)&vio_msg));
2015 }
2016 
2017 
2018 /* -------------------------------------------------------------------------- */
2019 
2020 /*
2021  * LDC helper routines
2022  */
2023 
2024 static int
2025 vdc_recv(vdc_t *vdc, vio_msg_t *msgp, size_t *nbytesp)
2026 {
2027 	int		status;
2028 	uint64_t	delay_time;
2029 	size_t		len;
2030 
2031 	/*
2032 	 * Until we get a blocking ldc read we have to retry until the entire
2033 	 * LDC message has arrived before ldc_read() will return that message.
2034 	 * If ldc_read() succeed but returns a zero length message then that
2035 	 * means that the LDC queue is empty and we have to wait for a
2036 	 * notification from the LDC callback which will set the read_state to
2037 	 * VDC_READ_PENDING. Note we also bail out if the channel is reset or
2038 	 * goes away.
2039 	 */
2040 	delay_time = vdc_ldc_read_init_delay;
2041 
2042 	for (;;) {
2043 
2044 		len = *nbytesp;
2045 		/*
2046 		 * vdc->curr_server is protected by vdc->lock but to avoid
2047 		 * contentions we don't take the lock here. We can do this
2048 		 * safely because vdc_recv() is only called from thread
2049 		 * process_msg_thread() which is also the only thread that
2050 		 * can change vdc->curr_server.
2051 		 */
2052 		status = ldc_read(vdc->curr_server->ldc_handle,
2053 		    (caddr_t)msgp, &len);
2054 
2055 		if (status == EAGAIN) {
2056 			delay_time *= 2;
2057 			if (delay_time >= vdc_ldc_read_max_delay)
2058 				delay_time = vdc_ldc_read_max_delay;
2059 			delay(delay_time);
2060 			continue;
2061 		}
2062 
2063 		if (status != 0) {
2064 			DMSG(vdc, 0, "ldc_read returned %d\n", status);
2065 			break;
2066 		}
2067 
2068 		if (len != 0) {
2069 			*nbytesp = len;
2070 			break;
2071 		}
2072 
2073 		mutex_enter(&vdc->read_lock);
2074 
2075 		while (vdc->read_state != VDC_READ_PENDING) {
2076 
2077 			/* detect if the connection has been reset */
2078 			if (vdc->read_state == VDC_READ_RESET) {
2079 				mutex_exit(&vdc->read_lock);
2080 				return (ECONNRESET);
2081 			}
2082 
2083 			vdc->read_state = VDC_READ_WAITING;
2084 			cv_wait(&vdc->read_cv, &vdc->read_lock);
2085 		}
2086 
2087 		vdc->read_state = VDC_READ_IDLE;
2088 		mutex_exit(&vdc->read_lock);
2089 
2090 		delay_time = vdc_ldc_read_init_delay;
2091 	}
2092 
2093 	return (status);
2094 }
2095 
2096 
2097 
2098 #ifdef DEBUG
2099 void
2100 vdc_decode_tag(vdc_t *vdcp, vio_msg_t *msg)
2101 {
2102 	char *ms, *ss, *ses;
2103 	switch (msg->tag.vio_msgtype) {
2104 #define	Q(_s)	case _s : ms = #_s; break;
2105 	Q(VIO_TYPE_CTRL)
2106 	Q(VIO_TYPE_DATA)
2107 	Q(VIO_TYPE_ERR)
2108 #undef Q
2109 	default: ms = "unknown"; break;
2110 	}
2111 
2112 	switch (msg->tag.vio_subtype) {
2113 #define	Q(_s)	case _s : ss = #_s; break;
2114 	Q(VIO_SUBTYPE_INFO)
2115 	Q(VIO_SUBTYPE_ACK)
2116 	Q(VIO_SUBTYPE_NACK)
2117 #undef Q
2118 	default: ss = "unknown"; break;
2119 	}
2120 
2121 	switch (msg->tag.vio_subtype_env) {
2122 #define	Q(_s)	case _s : ses = #_s; break;
2123 	Q(VIO_VER_INFO)
2124 	Q(VIO_ATTR_INFO)
2125 	Q(VIO_DRING_REG)
2126 	Q(VIO_DRING_UNREG)
2127 	Q(VIO_RDX)
2128 	Q(VIO_PKT_DATA)
2129 	Q(VIO_DESC_DATA)
2130 	Q(VIO_DRING_DATA)
2131 #undef Q
2132 	default: ses = "unknown"; break;
2133 	}
2134 
2135 	DMSG(vdcp, 3, "(%x/%x/%x) message : (%s/%s/%s)\n",
2136 	    msg->tag.vio_msgtype, msg->tag.vio_subtype,
2137 	    msg->tag.vio_subtype_env, ms, ss, ses);
2138 }
2139 #endif
2140 
2141 /*
2142  * Function:
2143  *	vdc_send()
2144  *
2145  * Description:
2146  *	The function encapsulates the call to write a message using LDC.
2147  *	If LDC indicates that the call failed due to the queue being full,
2148  *	we retry the ldc_write(), otherwise we return the error returned by LDC.
2149  *
2150  * Arguments:
2151  *	ldc_handle	- LDC handle for the channel this instance of vdc uses
2152  *	pkt		- address of LDC message to be sent
2153  *	msglen		- the size of the message being sent. When the function
2154  *			  returns, this contains the number of bytes written.
2155  *
2156  * Return Code:
2157  *	0		- Success.
2158  *	EINVAL		- pkt or msglen were NULL
2159  *	ECONNRESET	- The connection was not up.
2160  *	EWOULDBLOCK	- LDC queue is full
2161  *	xxx		- other error codes returned by ldc_write
2162  */
2163 static int
2164 vdc_send(vdc_t *vdc, caddr_t pkt, size_t *msglen)
2165 {
2166 	size_t	size = 0;
2167 	int	status = 0;
2168 	clock_t delay_ticks;
2169 
2170 	ASSERT(vdc != NULL);
2171 	ASSERT(mutex_owned(&vdc->lock));
2172 	ASSERT(msglen != NULL);
2173 	ASSERT(*msglen != 0);
2174 
2175 #ifdef DEBUG
2176 	vdc_decode_tag(vdc, (vio_msg_t *)(uintptr_t)pkt);
2177 #endif
2178 	/*
2179 	 * Wait indefinitely to send if channel
2180 	 * is busy, but bail out if we succeed or
2181 	 * if the channel closes or is reset.
2182 	 */
2183 	delay_ticks = vdc_hz_min_ldc_delay;
2184 	do {
2185 		size = *msglen;
2186 		status = ldc_write(vdc->curr_server->ldc_handle, pkt, &size);
2187 		if (status == EWOULDBLOCK) {
2188 			delay(delay_ticks);
2189 			/* geometric backoff */
2190 			delay_ticks *= 2;
2191 			if (delay_ticks > vdc_hz_max_ldc_delay)
2192 				delay_ticks = vdc_hz_max_ldc_delay;
2193 		}
2194 	} while (status == EWOULDBLOCK);
2195 
2196 	/* if LDC had serious issues --- reset vdc state */
2197 	if (status == EIO || status == ECONNRESET) {
2198 		/* LDC had serious issues --- reset vdc state */
2199 		mutex_enter(&vdc->read_lock);
2200 		if ((vdc->read_state == VDC_READ_WAITING) ||
2201 		    (vdc->read_state == VDC_READ_RESET))
2202 			cv_signal(&vdc->read_cv);
2203 		vdc->read_state = VDC_READ_RESET;
2204 		mutex_exit(&vdc->read_lock);
2205 
2206 		/* wake up any waiters in the reset thread */
2207 		if (vdc->state == VDC_STATE_INIT_WAITING) {
2208 			DMSG(vdc, 0, "[%d] write reset - "
2209 			    "vdc is resetting ..\n", vdc->instance);
2210 			vdc->state = VDC_STATE_RESETTING;
2211 			cv_signal(&vdc->initwait_cv);
2212 		}
2213 
2214 		return (ECONNRESET);
2215 	}
2216 
2217 	/* return the last size written */
2218 	*msglen = size;
2219 
2220 	return (status);
2221 }
2222 
2223 /*
2224  * Function:
2225  *	vdc_get_md_node
2226  *
2227  * Description:
2228  *	Get the MD, the device node for the given disk instance. The
2229  *	caller is responsible for cleaning up the reference to the
2230  *	returned MD (mdpp) by calling md_fini_handle().
2231  *
2232  * Arguments:
2233  *	dip	- dev info pointer for this instance of the device driver.
2234  *	mdpp	- the returned MD.
2235  *	vd_nodep - the returned device node.
2236  *
2237  * Return Code:
2238  *	0	- Success.
2239  *	ENOENT	- Expected node or property did not exist.
2240  *	ENXIO	- Unexpected error communicating with MD framework
2241  */
2242 static int
2243 vdc_get_md_node(dev_info_t *dip, md_t **mdpp, mde_cookie_t *vd_nodep)
2244 {
2245 	int		status = ENOENT;
2246 	char		*node_name = NULL;
2247 	md_t		*mdp = NULL;
2248 	int		num_nodes;
2249 	int		num_vdevs;
2250 	mde_cookie_t	rootnode;
2251 	mde_cookie_t	*listp = NULL;
2252 	boolean_t	found_inst = B_FALSE;
2253 	int		listsz;
2254 	int		idx;
2255 	uint64_t	md_inst;
2256 	int		obp_inst;
2257 	int		instance = ddi_get_instance(dip);
2258 
2259 	/*
2260 	 * Get the OBP instance number for comparison with the MD instance
2261 	 *
2262 	 * The "cfg-handle" property of a vdc node in an MD contains the MD's
2263 	 * notion of "instance", or unique identifier, for that node; OBP
2264 	 * stores the value of the "cfg-handle" MD property as the value of
2265 	 * the "reg" property on the node in the device tree it builds from
2266 	 * the MD and passes to Solaris.  Thus, we look up the devinfo node's
2267 	 * "reg" property value to uniquely identify this device instance.
2268 	 * If the "reg" property cannot be found, the device tree state is
2269 	 * presumably so broken that there is no point in continuing.
2270 	 */
2271 	if (!ddi_prop_exists(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS, OBP_REG)) {
2272 		cmn_err(CE_WARN, "'%s' property does not exist", OBP_REG);
2273 		return (ENOENT);
2274 	}
2275 	obp_inst = ddi_prop_get_int(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS,
2276 	    OBP_REG, -1);
2277 	DMSGX(1, "[%d] OBP inst=%d\n", instance, obp_inst);
2278 
2279 	/*
2280 	 * We now walk the MD nodes to find the node for this vdisk.
2281 	 */
2282 	if ((mdp = md_get_handle()) == NULL) {
2283 		cmn_err(CE_WARN, "unable to init machine description");
2284 		return (ENXIO);
2285 	}
2286 
2287 	num_nodes = md_node_count(mdp);
2288 	ASSERT(num_nodes > 0);
2289 
2290 	listsz = num_nodes * sizeof (mde_cookie_t);
2291 
2292 	/* allocate memory for nodes */
2293 	listp = kmem_zalloc(listsz, KM_SLEEP);
2294 
2295 	rootnode = md_root_node(mdp);
2296 	ASSERT(rootnode != MDE_INVAL_ELEM_COOKIE);
2297 
2298 	/*
2299 	 * Search for all the virtual devices, we will then check to see which
2300 	 * ones are disk nodes.
2301 	 */
2302 	num_vdevs = md_scan_dag(mdp, rootnode,
2303 	    md_find_name(mdp, VDC_MD_VDEV_NAME),
2304 	    md_find_name(mdp, "fwd"), listp);
2305 
2306 	if (num_vdevs <= 0) {
2307 		cmn_err(CE_NOTE, "No '%s' node found", VDC_MD_VDEV_NAME);
2308 		status = ENOENT;
2309 		goto done;
2310 	}
2311 
2312 	DMSGX(1, "[%d] num_vdevs=%d\n", instance, num_vdevs);
2313 	for (idx = 0; idx < num_vdevs; idx++) {
2314 		status = md_get_prop_str(mdp, listp[idx], "name", &node_name);
2315 		if ((status != 0) || (node_name == NULL)) {
2316 			cmn_err(CE_NOTE, "Unable to get name of node type '%s'"
2317 			    ": err %d", VDC_MD_VDEV_NAME, status);
2318 			continue;
2319 		}
2320 
2321 		DMSGX(1, "[%d] Found node '%s'\n", instance, node_name);
2322 		if (strcmp(VDC_MD_DISK_NAME, node_name) == 0) {
2323 			status = md_get_prop_val(mdp, listp[idx],
2324 			    VDC_MD_CFG_HDL, &md_inst);
2325 			DMSGX(1, "[%d] vdc inst in MD=%lx\n",
2326 			    instance, md_inst);
2327 			if ((status == 0) && (md_inst == obp_inst)) {
2328 				found_inst = B_TRUE;
2329 				break;
2330 			}
2331 		}
2332 	}
2333 
2334 	if (!found_inst) {
2335 		DMSGX(0, "Unable to find correct '%s' node", VDC_MD_DISK_NAME);
2336 		status = ENOENT;
2337 		goto done;
2338 	}
2339 	DMSGX(0, "[%d] MD inst=%lx\n", instance, md_inst);
2340 
2341 	*vd_nodep = listp[idx];
2342 	*mdpp = mdp;
2343 done:
2344 	kmem_free(listp, listsz);
2345 	return (status);
2346 }
2347 
2348 /*
2349  * Function:
2350  *	vdc_init_ports
2351  *
2352  * Description:
2353  *	Initialize all the ports for this vdisk instance.
2354  *
2355  * Arguments:
2356  *	vdc	- soft state pointer for this instance of the device driver.
2357  *	mdp	- md pointer
2358  *	vd_nodep - device md node.
2359  *
2360  * Return Code:
2361  *	0	- Success.
2362  *	ENOENT	- Expected node or property did not exist.
2363  */
2364 static int
2365 vdc_init_ports(vdc_t *vdc, md_t *mdp, mde_cookie_t vd_nodep)
2366 {
2367 	int		status = 0;
2368 	int		idx;
2369 	int		num_nodes;
2370 	int		num_vports;
2371 	int		num_chans;
2372 	int		listsz;
2373 	mde_cookie_t	vd_port;
2374 	mde_cookie_t	*chanp = NULL;
2375 	mde_cookie_t	*portp = NULL;
2376 	vdc_server_t	*srvr;
2377 	vdc_server_t	*prev_srvr = NULL;
2378 
2379 	/*
2380 	 * We now walk the MD nodes to find the port nodes for this vdisk.
2381 	 */
2382 	num_nodes = md_node_count(mdp);
2383 	ASSERT(num_nodes > 0);
2384 
2385 	listsz = num_nodes * sizeof (mde_cookie_t);
2386 
2387 	/* allocate memory for nodes */
2388 	portp = kmem_zalloc(listsz, KM_SLEEP);
2389 	chanp = kmem_zalloc(listsz, KM_SLEEP);
2390 
2391 	num_vports = md_scan_dag(mdp, vd_nodep,
2392 	    md_find_name(mdp, VDC_MD_PORT_NAME),
2393 	    md_find_name(mdp, "fwd"), portp);
2394 	if (num_vports == 0) {
2395 		DMSGX(0, "Found no '%s' node for '%s' port\n",
2396 		    VDC_MD_PORT_NAME, VDC_MD_VDEV_NAME);
2397 		status = ENOENT;
2398 		goto done;
2399 	}
2400 
2401 	DMSGX(1, "Found %d '%s' node(s) for '%s' port\n",
2402 	    num_vports, VDC_MD_PORT_NAME, VDC_MD_VDEV_NAME);
2403 
2404 	vdc->num_servers = 0;
2405 	for (idx = 0; idx < num_vports; idx++) {
2406 
2407 		/* initialize this port */
2408 		vd_port = portp[idx];
2409 		srvr = kmem_zalloc(sizeof (vdc_server_t), KM_SLEEP);
2410 		srvr->vdcp = vdc;
2411 		srvr->svc_state = VDC_SERVICE_OFFLINE;
2412 		srvr->log_state = VDC_SERVICE_NONE;
2413 
2414 		/* get port id */
2415 		if (md_get_prop_val(mdp, vd_port, VDC_MD_ID, &srvr->id) != 0) {
2416 			cmn_err(CE_NOTE, "vDisk port '%s' property not found",
2417 			    VDC_MD_ID);
2418 			kmem_free(srvr, sizeof (vdc_server_t));
2419 			continue;
2420 		}
2421 
2422 		/* set the connection timeout */
2423 		if (md_get_prop_val(mdp, vd_port, VDC_MD_TIMEOUT,
2424 		    &srvr->ctimeout) != 0) {
2425 			srvr->ctimeout = 0;
2426 		}
2427 
2428 		/* get the ldc id */
2429 		num_chans = md_scan_dag(mdp, vd_port,
2430 		    md_find_name(mdp, VDC_MD_CHAN_NAME),
2431 		    md_find_name(mdp, "fwd"), chanp);
2432 
2433 		/* expecting at least one channel */
2434 		if (num_chans <= 0) {
2435 			cmn_err(CE_NOTE, "No '%s' node for '%s' port",
2436 			    VDC_MD_CHAN_NAME, VDC_MD_VDEV_NAME);
2437 			kmem_free(srvr, sizeof (vdc_server_t));
2438 			continue;
2439 		} else if (num_chans != 1) {
2440 			DMSGX(0, "Expected 1 '%s' node for '%s' port, "
2441 			    "found %d\n", VDC_MD_CHAN_NAME, VDC_MD_VDEV_NAME,
2442 			    num_chans);
2443 		}
2444 
2445 		/*
2446 		 * We use the first channel found (index 0), irrespective of how
2447 		 * many are there in total.
2448 		 */
2449 		if (md_get_prop_val(mdp, chanp[0], VDC_MD_ID,
2450 		    &srvr->ldc_id) != 0) {
2451 			cmn_err(CE_NOTE, "Channel '%s' property not found",
2452 			    VDC_MD_ID);
2453 			kmem_free(srvr, sizeof (vdc_server_t));
2454 			continue;
2455 		}
2456 
2457 		/*
2458 		 * now initialise LDC channel which will be used to
2459 		 * communicate with this server
2460 		 */
2461 		if (vdc_do_ldc_init(vdc, srvr) != 0) {
2462 			kmem_free(srvr, sizeof (vdc_server_t));
2463 			continue;
2464 		}
2465 
2466 		/* add server to list */
2467 		if (prev_srvr)
2468 			prev_srvr->next = srvr;
2469 		else
2470 			vdc->server_list = srvr;
2471 
2472 		prev_srvr = srvr;
2473 
2474 		/* inc numbers of servers */
2475 		vdc->num_servers++;
2476 	}
2477 
2478 	/* pick first server as current server */
2479 	if (vdc->server_list != NULL) {
2480 		vdc->curr_server = vdc->server_list;
2481 		status = 0;
2482 	} else {
2483 		status = ENOENT;
2484 	}
2485 
2486 done:
2487 	kmem_free(chanp, listsz);
2488 	kmem_free(portp, listsz);
2489 	return (status);
2490 }
2491 
2492 
2493 /*
2494  * Function:
2495  *	vdc_do_ldc_up
2496  *
2497  * Description:
2498  *	Bring the channel for the current server up.
2499  *
2500  * Arguments:
2501  *	vdc	- soft state pointer for this instance of the device driver.
2502  *
2503  * Return Code:
2504  *	0		- Success.
2505  *	EINVAL		- Driver is detaching / LDC error
2506  *	ECONNREFUSED	- Other end is not listening
2507  */
2508 static int
2509 vdc_do_ldc_up(vdc_t *vdc)
2510 {
2511 	int		status;
2512 	ldc_status_t	ldc_state;
2513 
2514 	ASSERT(MUTEX_HELD(&vdc->lock));
2515 
2516 	DMSG(vdc, 0, "[%d] Bringing up channel %lx\n",
2517 	    vdc->instance, vdc->curr_server->ldc_id);
2518 
2519 	if (vdc->lifecycle == VDC_LC_DETACHING)
2520 		return (EINVAL);
2521 
2522 	if ((status = ldc_up(vdc->curr_server->ldc_handle)) != 0) {
2523 		switch (status) {
2524 		case ECONNREFUSED:	/* listener not ready at other end */
2525 			DMSG(vdc, 0, "[%d] ldc_up(%lx,...) return %d\n",
2526 			    vdc->instance, vdc->curr_server->ldc_id, status);
2527 			status = 0;
2528 			break;
2529 		default:
2530 			DMSG(vdc, 0, "[%d] Failed to bring up LDC: "
2531 			    "channel=%ld, err=%d", vdc->instance,
2532 			    vdc->curr_server->ldc_id, status);
2533 			break;
2534 		}
2535 	}
2536 
2537 	if (ldc_status(vdc->curr_server->ldc_handle, &ldc_state) == 0) {
2538 		vdc->curr_server->ldc_state = ldc_state;
2539 		if (ldc_state == LDC_UP) {
2540 			DMSG(vdc, 0, "[%d] LDC channel already up\n",
2541 			    vdc->instance);
2542 			vdc->seq_num = 1;
2543 			vdc->seq_num_reply = 0;
2544 		}
2545 	}
2546 
2547 	return (status);
2548 }
2549 
2550 /*
2551  * Function:
2552  *	vdc_terminate_ldc()
2553  *
2554  * Description:
2555  *
2556  * Arguments:
2557  *	vdc	- soft state pointer for this instance of the device driver.
2558  *	srvr	- vdc per-server info structure
2559  *
2560  * Return Code:
2561  *	None
2562  */
2563 static void
2564 vdc_terminate_ldc(vdc_t *vdc, vdc_server_t *srvr)
2565 {
2566 	int	instance = ddi_get_instance(vdc->dip);
2567 
2568 	if (srvr->state & VDC_LDC_OPEN) {
2569 		DMSG(vdc, 0, "[%d] ldc_close()\n", instance);
2570 		(void) ldc_close(srvr->ldc_handle);
2571 	}
2572 	if (srvr->state & VDC_LDC_CB) {
2573 		DMSG(vdc, 0, "[%d] ldc_unreg_callback()\n", instance);
2574 		(void) ldc_unreg_callback(srvr->ldc_handle);
2575 	}
2576 	if (srvr->state & VDC_LDC_INIT) {
2577 		DMSG(vdc, 0, "[%d] ldc_fini()\n", instance);
2578 		(void) ldc_fini(srvr->ldc_handle);
2579 		srvr->ldc_handle = 0;
2580 	}
2581 
2582 	srvr->state &= ~(VDC_LDC_INIT | VDC_LDC_CB | VDC_LDC_OPEN);
2583 }
2584 
2585 /*
2586  * Function:
2587  *	vdc_fini_ports()
2588  *
2589  * Description:
2590  *	Finalize all ports by closing the channel associated with each
2591  *	port and also freeing the server structure.
2592  *
2593  * Arguments:
2594  *	vdc	- soft state pointer for this instance of the device driver.
2595  *
2596  * Return Code:
2597  *	None
2598  */
2599 static void
2600 vdc_fini_ports(vdc_t *vdc)
2601 {
2602 	int		instance = ddi_get_instance(vdc->dip);
2603 	vdc_server_t	*srvr, *prev_srvr;
2604 
2605 	ASSERT(vdc != NULL);
2606 	ASSERT(mutex_owned(&vdc->lock));
2607 
2608 	DMSG(vdc, 0, "[%d] initialized=%x\n", instance, vdc->initialized);
2609 
2610 	srvr = vdc->server_list;
2611 
2612 	while (srvr) {
2613 
2614 		vdc_terminate_ldc(vdc, srvr);
2615 
2616 		/* next server */
2617 		prev_srvr = srvr;
2618 		srvr = srvr->next;
2619 
2620 		/* free server */
2621 		kmem_free(prev_srvr, sizeof (vdc_server_t));
2622 	}
2623 
2624 	vdc->server_list = NULL;
2625 	vdc->num_servers = 0;
2626 }
2627 
2628 /* -------------------------------------------------------------------------- */
2629 
2630 /*
2631  * Descriptor Ring helper routines
2632  */
2633 
2634 /*
2635  * Function:
2636  *	vdc_init_descriptor_ring()
2637  *
2638  * Description:
2639  *
2640  * Arguments:
2641  *	vdc	- soft state pointer for this instance of the device driver.
2642  *
2643  * Return Code:
2644  *	0	- Success
2645  */
2646 static int
2647 vdc_init_descriptor_ring(vdc_t *vdc)
2648 {
2649 	vd_dring_entry_t	*dep = NULL;	/* DRing Entry pointer */
2650 	int	status = 0;
2651 	int	i;
2652 
2653 	DMSG(vdc, 0, "[%d] initialized=%x\n", vdc->instance, vdc->initialized);
2654 
2655 	ASSERT(vdc != NULL);
2656 	ASSERT(mutex_owned(&vdc->lock));
2657 
2658 	/* ensure we have enough room to store max sized block */
2659 	ASSERT(maxphys <= VD_MAX_BLOCK_SIZE);
2660 
2661 	if ((vdc->initialized & VDC_DRING_INIT) == 0) {
2662 		DMSG(vdc, 0, "[%d] ldc_mem_dring_create\n", vdc->instance);
2663 		/*
2664 		 * Calculate the maximum block size we can transmit using one
2665 		 * Descriptor Ring entry from the attributes returned by the
2666 		 * vDisk server. This is subject to a minimum of 'maxphys'
2667 		 * as we do not have the capability to split requests over
2668 		 * multiple DRing entries.
2669 		 */
2670 		if ((vdc->max_xfer_sz * vdc->vdisk_bsize) < maxphys) {
2671 			DMSG(vdc, 0, "[%d] using minimum DRing size\n",
2672 			    vdc->instance);
2673 			vdc->dring_max_cookies = maxphys / PAGESIZE;
2674 		} else {
2675 			vdc->dring_max_cookies =
2676 			    (vdc->max_xfer_sz * vdc->vdisk_bsize) / PAGESIZE;
2677 		}
2678 		vdc->dring_entry_size = (sizeof (vd_dring_entry_t) +
2679 		    (sizeof (ldc_mem_cookie_t) *
2680 		    (vdc->dring_max_cookies - 1)));
2681 		vdc->dring_len = VD_DRING_LEN;
2682 
2683 		status = ldc_mem_dring_create(vdc->dring_len,
2684 		    vdc->dring_entry_size, &vdc->dring_hdl);
2685 		if ((vdc->dring_hdl == 0) || (status != 0)) {
2686 			DMSG(vdc, 0, "[%d] Descriptor ring creation failed",
2687 			    vdc->instance);
2688 			return (status);
2689 		}
2690 		vdc->initialized |= VDC_DRING_INIT;
2691 	}
2692 
2693 	if ((vdc->initialized & VDC_DRING_BOUND) == 0) {
2694 		DMSG(vdc, 0, "[%d] ldc_mem_dring_bind\n", vdc->instance);
2695 		vdc->dring_cookie =
2696 		    kmem_zalloc(sizeof (ldc_mem_cookie_t), KM_SLEEP);
2697 
2698 		status = ldc_mem_dring_bind(vdc->curr_server->ldc_handle,
2699 		    vdc->dring_hdl,
2700 		    LDC_SHADOW_MAP|LDC_DIRECT_MAP, LDC_MEM_RW,
2701 		    &vdc->dring_cookie[0],
2702 		    &vdc->dring_cookie_count);
2703 		if (status != 0) {
2704 			DMSG(vdc, 0, "[%d] Failed to bind descriptor ring "
2705 			    "(%lx) to channel (%lx) status=%d\n",
2706 			    vdc->instance, vdc->dring_hdl,
2707 			    vdc->curr_server->ldc_handle, status);
2708 			return (status);
2709 		}
2710 		ASSERT(vdc->dring_cookie_count == 1);
2711 		vdc->initialized |= VDC_DRING_BOUND;
2712 	}
2713 
2714 	status = ldc_mem_dring_info(vdc->dring_hdl, &vdc->dring_mem_info);
2715 	if (status != 0) {
2716 		DMSG(vdc, 0,
2717 		    "[%d] Failed to get info for descriptor ring (%lx)\n",
2718 		    vdc->instance, vdc->dring_hdl);
2719 		return (status);
2720 	}
2721 
2722 	if ((vdc->initialized & VDC_DRING_LOCAL) == 0) {
2723 		DMSG(vdc, 0, "[%d] local dring\n", vdc->instance);
2724 
2725 		/* Allocate the local copy of this dring */
2726 		vdc->local_dring =
2727 		    kmem_zalloc(vdc->dring_len * sizeof (vdc_local_desc_t),
2728 		    KM_SLEEP);
2729 		vdc->initialized |= VDC_DRING_LOCAL;
2730 	}
2731 
2732 	/*
2733 	 * Mark all DRing entries as free and initialize the private
2734 	 * descriptor's memory handles. If any entry is initialized,
2735 	 * we need to free it later so we set the bit in 'initialized'
2736 	 * at the start.
2737 	 */
2738 	vdc->initialized |= VDC_DRING_ENTRY;
2739 	for (i = 0; i < vdc->dring_len; i++) {
2740 		dep = VDC_GET_DRING_ENTRY_PTR(vdc, i);
2741 		dep->hdr.dstate = VIO_DESC_FREE;
2742 
2743 		status = ldc_mem_alloc_handle(vdc->curr_server->ldc_handle,
2744 		    &vdc->local_dring[i].desc_mhdl);
2745 		if (status != 0) {
2746 			DMSG(vdc, 0, "![%d] Failed to alloc mem handle for"
2747 			    " descriptor %d", vdc->instance, i);
2748 			return (status);
2749 		}
2750 		vdc->local_dring[i].is_free = B_TRUE;
2751 		vdc->local_dring[i].dep = dep;
2752 	}
2753 
2754 	/* Initialize the starting index */
2755 	vdc->dring_curr_idx = VDC_DRING_FIRST_ENTRY;
2756 
2757 	return (status);
2758 }
2759 
2760 /*
2761  * Function:
2762  *	vdc_destroy_descriptor_ring()
2763  *
2764  * Description:
2765  *
2766  * Arguments:
2767  *	vdc	- soft state pointer for this instance of the device driver.
2768  *
2769  * Return Code:
2770  *	None
2771  */
2772 static void
2773 vdc_destroy_descriptor_ring(vdc_t *vdc)
2774 {
2775 	vdc_local_desc_t	*ldep = NULL;	/* Local Dring Entry Pointer */
2776 	ldc_mem_handle_t	mhdl = 0;
2777 	ldc_mem_info_t		minfo;
2778 	int			status = -1;
2779 	int			i;	/* loop */
2780 
2781 	ASSERT(vdc != NULL);
2782 	ASSERT(mutex_owned(&vdc->lock));
2783 
2784 	DMSG(vdc, 0, "[%d] Entered\n", vdc->instance);
2785 
2786 	if (vdc->initialized & VDC_DRING_ENTRY) {
2787 		DMSG(vdc, 0,
2788 		    "[%d] Removing Local DRing entries\n", vdc->instance);
2789 		for (i = 0; i < vdc->dring_len; i++) {
2790 			ldep = &vdc->local_dring[i];
2791 			mhdl = ldep->desc_mhdl;
2792 
2793 			if (mhdl == 0)
2794 				continue;
2795 
2796 			if ((status = ldc_mem_info(mhdl, &minfo)) != 0) {
2797 				DMSG(vdc, 0,
2798 				    "ldc_mem_info returned an error: %d\n",
2799 				    status);
2800 
2801 				/*
2802 				 * This must mean that the mem handle
2803 				 * is not valid. Clear it out so that
2804 				 * no one tries to use it.
2805 				 */
2806 				ldep->desc_mhdl = 0;
2807 				continue;
2808 			}
2809 
2810 			if (minfo.status == LDC_BOUND) {
2811 				(void) ldc_mem_unbind_handle(mhdl);
2812 			}
2813 
2814 			(void) ldc_mem_free_handle(mhdl);
2815 
2816 			ldep->desc_mhdl = 0;
2817 		}
2818 		vdc->initialized &= ~VDC_DRING_ENTRY;
2819 	}
2820 
2821 	if (vdc->initialized & VDC_DRING_LOCAL) {
2822 		DMSG(vdc, 0, "[%d] Freeing Local DRing\n", vdc->instance);
2823 		kmem_free(vdc->local_dring,
2824 		    vdc->dring_len * sizeof (vdc_local_desc_t));
2825 		vdc->initialized &= ~VDC_DRING_LOCAL;
2826 	}
2827 
2828 	if (vdc->initialized & VDC_DRING_BOUND) {
2829 		DMSG(vdc, 0, "[%d] Unbinding DRing\n", vdc->instance);
2830 		status = ldc_mem_dring_unbind(vdc->dring_hdl);
2831 		if (status == 0) {
2832 			vdc->initialized &= ~VDC_DRING_BOUND;
2833 		} else {
2834 			DMSG(vdc, 0, "[%d] Error %d unbinding DRing %lx",
2835 			    vdc->instance, status, vdc->dring_hdl);
2836 		}
2837 		kmem_free(vdc->dring_cookie, sizeof (ldc_mem_cookie_t));
2838 	}
2839 
2840 	if (vdc->initialized & VDC_DRING_INIT) {
2841 		DMSG(vdc, 0, "[%d] Destroying DRing\n", vdc->instance);
2842 		status = ldc_mem_dring_destroy(vdc->dring_hdl);
2843 		if (status == 0) {
2844 			vdc->dring_hdl = 0;
2845 			bzero(&vdc->dring_mem_info, sizeof (ldc_mem_info_t));
2846 			vdc->initialized &= ~VDC_DRING_INIT;
2847 		} else {
2848 			DMSG(vdc, 0, "[%d] Error %d destroying DRing (%lx)",
2849 			    vdc->instance, status, vdc->dring_hdl);
2850 		}
2851 	}
2852 }
2853 
2854 /*
2855  * Function:
2856  *	vdc_map_to_shared_dring()
2857  *
2858  * Description:
2859  *	Copy contents of the local descriptor to the shared
2860  *	memory descriptor.
2861  *
2862  * Arguments:
2863  *	vdcp	- soft state pointer for this instance of the device driver.
2864  *	idx	- descriptor ring index
2865  *
2866  * Return Code:
2867  *	None
2868  */
2869 static int
2870 vdc_map_to_shared_dring(vdc_t *vdcp, int idx)
2871 {
2872 	vdc_local_desc_t	*ldep;
2873 	vd_dring_entry_t	*dep;
2874 	int			rv;
2875 
2876 	ldep = &(vdcp->local_dring[idx]);
2877 
2878 	/* for now leave in the old pop_mem_hdl stuff */
2879 	if (ldep->nbytes > 0) {
2880 		rv = vdc_populate_mem_hdl(vdcp, ldep);
2881 		if (rv) {
2882 			DMSG(vdcp, 0, "[%d] Cannot populate mem handle\n",
2883 			    vdcp->instance);
2884 			return (rv);
2885 		}
2886 	}
2887 
2888 	/*
2889 	 * fill in the data details into the DRing
2890 	 */
2891 	dep = ldep->dep;
2892 	ASSERT(dep != NULL);
2893 
2894 	dep->payload.req_id = VDC_GET_NEXT_REQ_ID(vdcp);
2895 	dep->payload.operation = ldep->operation;
2896 	dep->payload.addr = ldep->offset;
2897 	dep->payload.nbytes = ldep->nbytes;
2898 	dep->payload.status = (uint32_t)-1;	/* vds will set valid value */
2899 	dep->payload.slice = ldep->slice;
2900 	dep->hdr.dstate = VIO_DESC_READY;
2901 	dep->hdr.ack = 1;		/* request an ACK for every message */
2902 
2903 	return (0);
2904 }
2905 
2906 /*
2907  * Function:
2908  *	vdc_send_request
2909  *
2910  * Description:
2911  *	This routine writes the data to be transmitted to vds into the
2912  *	descriptor, notifies vds that the ring has been updated and
2913  *	then waits for the request to be processed.
2914  *
2915  * Arguments:
2916  *	vdcp	  - the soft state pointer
2917  *	operation - operation we want vds to perform (VD_OP_XXX)
2918  *	addr	  - address of data buf to be read/written.
2919  *	nbytes	  - number of bytes to read/write
2920  *	slice	  - the disk slice this request is for
2921  *	offset	  - relative disk offset
2922  *	bufp	  - buf of operation
2923  *	dir	  - direction of operation (READ/WRITE/BOTH)
2924  *
2925  * Return Codes:
2926  *	0
2927  *	ENXIO
2928  */
2929 static int
2930 vdc_send_request(vdc_t *vdcp, int operation, caddr_t addr,
2931     size_t nbytes, int slice, diskaddr_t offset, buf_t *bufp,
2932     vio_desc_direction_t dir, int flags)
2933 {
2934 	int	rv = 0;
2935 
2936 	ASSERT(vdcp != NULL);
2937 	ASSERT(slice == VD_SLICE_NONE || slice < V_NUMPAR);
2938 
2939 	mutex_enter(&vdcp->lock);
2940 
2941 	/*
2942 	 * If this is a block read/write operation we update the I/O statistics
2943 	 * to indicate that the request is being put on the waitq to be
2944 	 * serviced. Operations which are resubmitted are already in the waitq.
2945 	 *
2946 	 * We do it here (a common routine for both synchronous and strategy
2947 	 * calls) for performance reasons - we are already holding vdc->lock
2948 	 * so there is no extra locking overhead. We would have to explicitly
2949 	 * grab the 'lock' mutex to update the stats if we were to do this
2950 	 * higher up the stack in vdc_strategy() et. al.
2951 	 */
2952 	if (((operation == VD_OP_BREAD) || (operation == VD_OP_BWRITE)) &&
2953 	    !(flags & VDC_OP_RESUBMIT)) {
2954 		DTRACE_IO1(start, buf_t *, bufp);
2955 		VD_KSTAT_WAITQ_ENTER(vdcp);
2956 	}
2957 
2958 	/*
2959 	 * If the request does not expect the state to be VDC_STATE_RUNNING
2960 	 * then we just try to populate the descriptor ring once.
2961 	 */
2962 	if (!(flags & VDC_OP_STATE_RUNNING)) {
2963 		rv = vdc_populate_descriptor(vdcp, operation, addr,
2964 		    nbytes, slice, offset, bufp, dir, flags);
2965 		goto done;
2966 	}
2967 
2968 	do {
2969 		while (vdcp->state != VDC_STATE_RUNNING) {
2970 
2971 			/* return error if detaching */
2972 			if (vdcp->state == VDC_STATE_DETACH) {
2973 				rv = ENXIO;
2974 				goto done;
2975 			}
2976 
2977 			/*
2978 			 * If we are panicking and the disk is not ready then
2979 			 * we can't send any request because we can't complete
2980 			 * the handshake now.
2981 			 */
2982 			if (ddi_in_panic()) {
2983 				rv = EIO;
2984 				goto done;
2985 			}
2986 
2987 			/*
2988 			 * If the state is faulted, notify that a new I/O is
2989 			 * being submitted to force the system to check if any
2990 			 * server has recovered.
2991 			 */
2992 			if (vdcp->state == VDC_STATE_FAILED) {
2993 				vdcp->io_pending = B_TRUE;
2994 				cv_signal(&vdcp->io_pending_cv);
2995 			}
2996 
2997 			cv_wait(&vdcp->running_cv, &vdcp->lock);
2998 
2999 			/* if service is still faulted then fail the request */
3000 			if (vdcp->state == VDC_STATE_FAILED) {
3001 				rv = EIO;
3002 				goto done;
3003 			}
3004 		}
3005 
3006 	} while (vdc_populate_descriptor(vdcp, operation, addr,
3007 	    nbytes, slice, offset, bufp, dir, flags & ~VDC_OP_RESUBMIT));
3008 
3009 done:
3010 	/*
3011 	 * If this is a block read/write we update the I/O statistics kstat
3012 	 * to indicate that this request has been placed on the queue for
3013 	 * processing (i.e sent to the vDisk server) - iostat(8) will
3014 	 * report the time waiting for the vDisk server under the %b column
3015 	 *
3016 	 * In the case of an error we take it off the wait queue only if
3017 	 * the I/O was not resubmited.
3018 	 */
3019 	if ((operation == VD_OP_BREAD) || (operation == VD_OP_BWRITE)) {
3020 		if (rv == 0) {
3021 			VD_KSTAT_WAITQ_TO_RUNQ(vdcp);
3022 			DTRACE_PROBE1(send, buf_t *, bufp);
3023 		} else {
3024 			VD_UPDATE_ERR_STATS(vdcp, vd_transerrs);
3025 			if (!(flags & VDC_OP_RESUBMIT)) {
3026 				VD_KSTAT_WAITQ_EXIT(vdcp);
3027 				DTRACE_IO1(done, buf_t *, bufp);
3028 			}
3029 		}
3030 	}
3031 
3032 	mutex_exit(&vdcp->lock);
3033 
3034 	return (rv);
3035 }
3036 
3037 
3038 /*
3039  * Function:
3040  *	vdc_populate_descriptor
3041  *
3042  * Description:
3043  *	This routine writes the data to be transmitted to vds into the
3044  *	descriptor, notifies vds that the ring has been updated and
3045  *	then waits for the request to be processed.
3046  *
3047  * Arguments:
3048  *	vdcp	  - the soft state pointer
3049  *	operation - operation we want vds to perform (VD_OP_XXX)
3050  *	addr	  - address of data buf to be read/written.
3051  *	nbytes	  - number of bytes to read/write
3052  *	slice	  - the disk slice this request is for
3053  *	offset	  - relative disk offset
3054  *	bufp	  - buf of operation
3055  *	dir	  - direction of operation (READ/WRITE/BOTH)
3056  *
3057  * Return Codes:
3058  *	0
3059  *	EAGAIN
3060  *	ECONNRESET
3061  *	ENXIO
3062  */
3063 static int
3064 vdc_populate_descriptor(vdc_t *vdcp, int operation, caddr_t addr,
3065     size_t nbytes, int slice, diskaddr_t offset,
3066     buf_t *bufp, vio_desc_direction_t dir, int flags)
3067 {
3068 	vdc_local_desc_t	*local_dep = NULL; /* Local Dring Pointer */
3069 	int			idx;		/* Index of DRing entry used */
3070 	int			next_idx;
3071 	vio_dring_msg_t		dmsg;
3072 	size_t			msglen;
3073 	int			rv;
3074 
3075 	ASSERT(MUTEX_HELD(&vdcp->lock));
3076 	vdcp->threads_pending++;
3077 loop:
3078 	DMSG(vdcp, 2, ": dring_curr_idx = %d\n", vdcp->dring_curr_idx);
3079 
3080 	if (flags & VDC_OP_DRING_RESERVED) {
3081 		/* use D-Ring reserved entry */
3082 		idx = VDC_DRING_FIRST_RESV;
3083 		local_dep = &(vdcp->local_dring[idx]);
3084 	} else {
3085 		/* Get next available D-Ring entry */
3086 		idx = vdcp->dring_curr_idx;
3087 		local_dep = &(vdcp->local_dring[idx]);
3088 
3089 		if (!local_dep->is_free) {
3090 			DMSG(vdcp, 2, "[%d]: dring full - waiting for space\n",
3091 			    vdcp->instance);
3092 			cv_wait(&vdcp->dring_free_cv, &vdcp->lock);
3093 			if (vdcp->state == VDC_STATE_RUNNING ||
3094 			    vdcp->state == VDC_STATE_HANDLE_PENDING) {
3095 				goto loop;
3096 			}
3097 			vdcp->threads_pending--;
3098 			return (ECONNRESET);
3099 		}
3100 
3101 		next_idx = idx + 1;
3102 		if (next_idx >= vdcp->dring_len)
3103 			next_idx = VDC_DRING_FIRST_ENTRY;
3104 		vdcp->dring_curr_idx = next_idx;
3105 	}
3106 
3107 	ASSERT(local_dep->is_free);
3108 
3109 	local_dep->operation = operation;
3110 	local_dep->addr = addr;
3111 	local_dep->nbytes = nbytes;
3112 	local_dep->slice = slice;
3113 	local_dep->offset = offset;
3114 	local_dep->buf = bufp;
3115 	local_dep->dir = dir;
3116 	local_dep->flags = flags;
3117 
3118 	local_dep->is_free = B_FALSE;
3119 
3120 	rv = vdc_map_to_shared_dring(vdcp, idx);
3121 	if (rv) {
3122 		if (flags & VDC_OP_DRING_RESERVED) {
3123 			DMSG(vdcp, 0, "[%d]: cannot bind memory - error\n",
3124 			    vdcp->instance);
3125 			/*
3126 			 * We can't wait if we are using reserved slot.
3127 			 * Free the descriptor and return.
3128 			 */
3129 			local_dep->is_free = B_TRUE;
3130 			vdcp->threads_pending--;
3131 			return (rv);
3132 		}
3133 		DMSG(vdcp, 0, "[%d]: cannot bind memory - waiting ..\n",
3134 		    vdcp->instance);
3135 		/* free the descriptor */
3136 		local_dep->is_free = B_TRUE;
3137 		vdcp->dring_curr_idx = idx;
3138 		cv_wait(&vdcp->membind_cv, &vdcp->lock);
3139 		if (vdcp->state == VDC_STATE_RUNNING ||
3140 		    vdcp->state == VDC_STATE_HANDLE_PENDING) {
3141 			goto loop;
3142 		}
3143 		vdcp->threads_pending--;
3144 		return (ECONNRESET);
3145 	}
3146 
3147 	/*
3148 	 * Send a msg with the DRing details to vds
3149 	 */
3150 	VIO_INIT_DRING_DATA_TAG(dmsg);
3151 	VDC_INIT_DRING_DATA_MSG_IDS(dmsg, vdcp);
3152 	dmsg.dring_ident = vdcp->dring_ident;
3153 	dmsg.start_idx = idx;
3154 	dmsg.end_idx = idx;
3155 	vdcp->seq_num++;
3156 
3157 	DTRACE_PROBE2(populate, int, vdcp->instance,
3158 	    vdc_local_desc_t *, local_dep);
3159 	DMSG(vdcp, 2, "ident=0x%lx, st=%u, end=%u, seq=%ld\n",
3160 	    vdcp->dring_ident, dmsg.start_idx, dmsg.end_idx, dmsg.seq_num);
3161 
3162 	/*
3163 	 * note we're still holding the lock here to
3164 	 * make sure the message goes out in order !!!...
3165 	 */
3166 	msglen = sizeof (dmsg);
3167 	rv = vdc_send(vdcp, (caddr_t)&dmsg, &msglen);
3168 	switch (rv) {
3169 	case ECONNRESET:
3170 		/*
3171 		 * vdc_send initiates the reset on failure.
3172 		 * Since the transaction has already been put
3173 		 * on the local dring, it will automatically get
3174 		 * retried when the channel is reset. Given that,
3175 		 * it is ok to just return success even though the
3176 		 * send failed.
3177 		 */
3178 		rv = 0;
3179 		break;
3180 
3181 	case 0: /* EOK */
3182 		DMSG(vdcp, 1, "sent via LDC: rv=%d\n", rv);
3183 		break;
3184 
3185 	default:
3186 		DMSG(vdcp, 0, "unexpected error, rv=%d\n", rv);
3187 		rv = ENXIO;
3188 		break;
3189 	}
3190 
3191 	vdcp->threads_pending--;
3192 	return (rv);
3193 }
3194 
3195 /*
3196  * Function:
3197  *	vdc_do_op
3198  *
3199  * Description:
3200  *	Wrapper around vdc_submit_request(). Each request is associated with a
3201  *	buf structure. If a buf structure is provided (bufp != NULL) then the
3202  *	request will be submitted with that buf, and the caller can wait for
3203  *	completion of the request with biowait(). If a buf structure is not
3204  *	provided (bufp == NULL) then a buf structure is created and the function
3205  *	waits for the completion of the request.
3206  *
3207  *	If the flag VD_OP_STATE_RUNNING is set then vdc_submit_request() will
3208  *	submit the request only when the vdisk is in state VD_STATE_RUNNING.
3209  *	If the vdisk is not in that state then the vdc_submit_request() will
3210  *	wait for that state to be reached. After the request is submitted, the
3211  *	reply will be processed asynchronously by the vdc_process_msg_thread()
3212  *	thread.
3213  *
3214  *	If the flag VD_OP_STATE_RUNNING is not set then vdc_submit_request()
3215  *	submit the request whatever the state of the vdisk is. Then vdc_do_op()
3216  *	will wait for a reply message, process the reply and complete the
3217  *	request.
3218  *
3219  * Arguments:
3220  *	vdc	- the soft state pointer
3221  *	op	- operation we want vds to perform (VD_OP_XXX)
3222  *	addr	- address of data buf to be read/written.
3223  *	nbytes	- number of bytes to read/write
3224  *	slice	- the disk slice this request is for
3225  *	offset	- relative disk offset
3226  *	bufp	- buf structure associated with the request (can be NULL).
3227  *	dir	- direction of operation (READ/WRITE/BOTH)
3228  *	flags	- flags for the request.
3229  *
3230  * Return Codes:
3231  *	0	- the request has been succesfully submitted and completed.
3232  *	!= 0	- the request has failed. In that case, if a buf structure
3233  *		  was provided (bufp != NULL) then the B_ERROR flag is set
3234  *		  and the b_error field of the buf structure is set to EIO.
3235  */
3236 static int
3237 vdc_do_op(vdc_t *vdc, int op, caddr_t addr, size_t nbytes, int slice,
3238     diskaddr_t offset, struct buf *bufp, vio_desc_direction_t dir, int flags)
3239 {
3240 	vio_msg_t vio_msg;
3241 	struct buf buf;
3242 	int rv;
3243 
3244 	if (bufp == NULL) {
3245 		/*
3246 		 * We use buf just as a convenient way to get a notification
3247 		 * that the request is completed, so we initialize buf to the
3248 		 * minimum we need.
3249 		 */
3250 		bioinit(&buf);
3251 		buf.b_bcount = nbytes;
3252 		buf.b_flags = B_BUSY;
3253 		bufp = &buf;
3254 	}
3255 
3256 	rv = vdc_send_request(vdc, op, addr, nbytes, slice, offset, bufp,
3257 	    dir, flags);
3258 
3259 	if (rv != 0)
3260 		goto done;
3261 
3262 	/*
3263 	 * If the request should be done in VDC_STATE_RUNNING state then the
3264 	 * reply will be received and processed by vdc_process_msg_thread()
3265 	 * and we just have to handle the panic case. Otherwise we have to
3266 	 * wait for the reply message and process it.
3267 	 */
3268 	if (flags & VDC_OP_STATE_RUNNING) {
3269 
3270 		if (ddi_in_panic()) {
3271 			rv = vdc_drain_response(vdc, bufp);
3272 			goto done;
3273 		}
3274 
3275 	} else {
3276 		/* wait for the response message */
3277 		rv = vdc_wait_for_response(vdc, &vio_msg);
3278 
3279 		if (rv == 0)
3280 			rv = vdc_process_data_msg(vdc, &vio_msg);
3281 
3282 		if (rv) {
3283 			/*
3284 			 * If this is a block read/write we update the I/O
3285 			 * statistics kstat to take it off the run queue.
3286 			 * If it is a resubmit then it needs to stay in
3287 			 * in the waitq, and it will be removed when the
3288 			 * I/O is eventually completed or cancelled.
3289 			 */
3290 			mutex_enter(&vdc->lock);
3291 			if (op == VD_OP_BREAD || op == VD_OP_BWRITE) {
3292 				if (flags & VDC_OP_RESUBMIT) {
3293 					VD_KSTAT_RUNQ_BACK_TO_WAITQ(vdc);
3294 				} else {
3295 					VD_KSTAT_RUNQ_EXIT(vdc);
3296 					DTRACE_IO1(done, buf_t *, bufp);
3297 				}
3298 			}
3299 			mutex_exit(&vdc->lock);
3300 			goto done;
3301 		}
3302 
3303 	}
3304 
3305 	if (bufp == &buf)
3306 		rv = biowait(bufp);
3307 
3308 done:
3309 	if (bufp == &buf) {
3310 		biofini(bufp);
3311 	} else if (rv != 0) {
3312 		bioerror(bufp, EIO);
3313 		biodone(bufp);
3314 	}
3315 
3316 	return (rv);
3317 }
3318 
3319 /*
3320  * Function:
3321  *	vdc_do_sync_op
3322  *
3323  * Description:
3324  *	Wrapper around vdc_do_op that serializes requests.
3325  *
3326  * Arguments:
3327  *	vdcp	  - the soft state pointer
3328  *	operation - operation we want vds to perform (VD_OP_XXX)
3329  *	addr	  - address of data buf to be read/written.
3330  *	nbytes	  - number of bytes to read/write
3331  *	slice	  - the disk slice this request is for
3332  *	offset	  - relative disk offset
3333  *	dir	  - direction of operation (READ/WRITE/BOTH)
3334  *	rconflict - check for reservation conflict in case of failure
3335  *
3336  * rconflict should be set to B_TRUE by most callers. Callers invoking the
3337  * VD_OP_SCSICMD operation can set rconflict to B_FALSE if they check the
3338  * result of a successful operation with vdc_scsi_status().
3339  *
3340  * Return Codes:
3341  *	0
3342  *	EAGAIN
3343  *	EFAULT
3344  *	ENXIO
3345  *	EIO
3346  */
3347 static int
3348 vdc_do_sync_op(vdc_t *vdcp, int operation, caddr_t addr, size_t nbytes,
3349     int slice, diskaddr_t offset, vio_desc_direction_t dir, boolean_t rconflict)
3350 {
3351 	int status;
3352 	int flags = VDC_OP_NORMAL;
3353 
3354 	/*
3355 	 * Grab the lock, if blocked wait until the server
3356 	 * response causes us to wake up again.
3357 	 */
3358 	mutex_enter(&vdcp->lock);
3359 	vdcp->sync_op_cnt++;
3360 	while (vdcp->sync_op_blocked && vdcp->state != VDC_STATE_DETACH) {
3361 		if (ddi_in_panic()) {
3362 			/* don't block if we are panicking */
3363 			vdcp->sync_op_cnt--;
3364 			mutex_exit(&vdcp->lock);
3365 			return (EIO);
3366 		} else {
3367 			cv_wait(&vdcp->sync_blocked_cv, &vdcp->lock);
3368 		}
3369 	}
3370 
3371 	if (vdcp->state == VDC_STATE_DETACH) {
3372 		cv_broadcast(&vdcp->sync_blocked_cv);
3373 		vdcp->sync_op_cnt--;
3374 		mutex_exit(&vdcp->lock);
3375 		return (ENXIO);
3376 	}
3377 
3378 	/* now block anyone other thread entering after us */
3379 	vdcp->sync_op_blocked = B_TRUE;
3380 
3381 	mutex_exit(&vdcp->lock);
3382 
3383 	if (!rconflict)
3384 		flags &= ~VDC_OP_ERRCHK_CONFLICT;
3385 
3386 	status = vdc_do_op(vdcp, operation, addr, nbytes, slice, offset,
3387 	    NULL, dir, flags);
3388 
3389 	mutex_enter(&vdcp->lock);
3390 
3391 	DMSG(vdcp, 2, ": operation returned %d\n", status);
3392 
3393 	if (vdcp->state == VDC_STATE_DETACH) {
3394 		status = ENXIO;
3395 	}
3396 
3397 	vdcp->sync_op_blocked = B_FALSE;
3398 	vdcp->sync_op_cnt--;
3399 
3400 	/* signal the next waiting thread */
3401 	cv_signal(&vdcp->sync_blocked_cv);
3402 
3403 	mutex_exit(&vdcp->lock);
3404 
3405 	return (status);
3406 }
3407 
3408 
3409 /*
3410  * Function:
3411  *	vdc_drain_response()
3412  *
3413  * Description:
3414  *	When a guest is panicking, the completion of requests needs to be
3415  *	handled differently because interrupts are disabled and vdc
3416  *	will not get messages. We have to poll for the messages instead.
3417  *
3418  *	Note: since we are panicking we don't implement	the io:::done
3419  *	DTrace probe or update the I/O statistics kstats.
3420  *
3421  * Arguments:
3422  *	vdc	- soft state pointer for this instance of the device driver.
3423  *	buf	- if buf is NULL then we drain all responses, otherwise we
3424  *		  poll until we receive a ACK/NACK for the specific I/O
3425  *		  described by buf.
3426  *
3427  * Return Code:
3428  *	0	- Success. If we were expecting a response to a particular
3429  *		  request then this means that a response has been received.
3430  */
3431 static int
3432 vdc_drain_response(vdc_t *vdc, struct buf *buf)
3433 {
3434 	int			rv, idx, retries;
3435 	size_t			msglen;
3436 	vdc_local_desc_t	*ldep = NULL;	/* Local Dring Entry Pointer */
3437 	vio_dring_msg_t		dmsg;
3438 	struct buf		*mbuf;
3439 	boolean_t		ack;
3440 
3441 	mutex_enter(&vdc->lock);
3442 
3443 	retries = 0;
3444 	for (;;) {
3445 		msglen = sizeof (dmsg);
3446 		rv = ldc_read(vdc->curr_server->ldc_handle, (caddr_t)&dmsg,
3447 		    &msglen);
3448 		if (rv) {
3449 			rv = EINVAL;
3450 			break;
3451 		}
3452 
3453 		/*
3454 		 * if there are no packets wait and check again
3455 		 */
3456 		if ((rv == 0) && (msglen == 0)) {
3457 			if (retries++ > vdc_dump_retries) {
3458 				rv = EAGAIN;
3459 				break;
3460 			}
3461 
3462 			drv_usecwait(vdc_usec_timeout_dump);
3463 			continue;
3464 		}
3465 
3466 		/*
3467 		 * Ignore all messages that are not ACKs/NACKs to
3468 		 * DRing requests.
3469 		 */
3470 		if ((dmsg.tag.vio_msgtype != VIO_TYPE_DATA) ||
3471 		    (dmsg.tag.vio_subtype_env != VIO_DRING_DATA)) {
3472 			DMSG(vdc, 0, "discard pkt: type=%d sub=%d env=%d\n",
3473 			    dmsg.tag.vio_msgtype,
3474 			    dmsg.tag.vio_subtype,
3475 			    dmsg.tag.vio_subtype_env);
3476 			continue;
3477 		}
3478 
3479 		/*
3480 		 * Record if the packet was ACK'ed or not. If the packet was not
3481 		 * ACK'ed then we will just mark the request as failed; we don't
3482 		 * want to reset the connection at this point.
3483 		 */
3484 		switch (dmsg.tag.vio_subtype) {
3485 		case VIO_SUBTYPE_ACK:
3486 			ack = B_TRUE;
3487 			break;
3488 		case VIO_SUBTYPE_NACK:
3489 			ack = B_FALSE;
3490 			break;
3491 		default:
3492 			continue;
3493 		}
3494 
3495 		idx = dmsg.start_idx;
3496 		if (idx >= vdc->dring_len) {
3497 			DMSG(vdc, 0, "[%d] Bogus ack data : start %d\n",
3498 			    vdc->instance, idx);
3499 			continue;
3500 		}
3501 		ldep = &vdc->local_dring[idx];
3502 		if (ldep->dep->hdr.dstate != VIO_DESC_DONE) {
3503 			DMSG(vdc, 0, "[%d] Entry @ %d - state !DONE %d\n",
3504 			    vdc->instance, idx, ldep->dep->hdr.dstate);
3505 			continue;
3506 		}
3507 
3508 		mbuf = ldep->buf;
3509 		ASSERT(mbuf != NULL);
3510 		mbuf->b_resid = mbuf->b_bcount - ldep->dep->payload.nbytes;
3511 		bioerror(mbuf, ack ? ldep->dep->payload.status : EIO);
3512 		biodone(mbuf);
3513 
3514 		rv = vdc_depopulate_descriptor(vdc, idx);
3515 		if (buf != NULL && buf == mbuf) {
3516 			rv = 0;
3517 			goto done;
3518 		}
3519 
3520 		/* if this is the last descriptor - break out of loop */
3521 		if ((idx + 1) % vdc->dring_len == vdc->dring_curr_idx) {
3522 			/*
3523 			 * If we were expecting a response for a particular
3524 			 * request then we return with an error otherwise we
3525 			 * have successfully completed the drain.
3526 			 */
3527 			rv = (buf != NULL)? ESRCH: 0;
3528 			break;
3529 		}
3530 	}
3531 
3532 done:
3533 	mutex_exit(&vdc->lock);
3534 	DMSG(vdc, 0, "End idx=%d\n", idx);
3535 
3536 	return (rv);
3537 }
3538 
3539 
3540 /*
3541  * Function:
3542  *	vdc_depopulate_descriptor()
3543  *
3544  * Description:
3545  *
3546  * Arguments:
3547  *	vdc	- soft state pointer for this instance of the device driver.
3548  *	idx	- Index of the Descriptor Ring entry being modified
3549  *
3550  * Return Code:
3551  *	0	- Success
3552  */
3553 static int
3554 vdc_depopulate_descriptor(vdc_t *vdc, uint_t idx)
3555 {
3556 	vd_dring_entry_t *dep = NULL;		/* Dring Entry Pointer */
3557 	vdc_local_desc_t *ldep = NULL;		/* Local Dring Entry Pointer */
3558 	int		status = ENXIO;
3559 	int		rv = 0;
3560 
3561 	ASSERT(vdc != NULL);
3562 	ASSERT(idx < vdc->dring_len);
3563 	ldep = &vdc->local_dring[idx];
3564 	ASSERT(ldep != NULL);
3565 	ASSERT(MUTEX_HELD(&vdc->lock));
3566 
3567 	DTRACE_PROBE2(depopulate, int, vdc->instance, vdc_local_desc_t *, ldep);
3568 	DMSG(vdc, 2, ": idx = %d\n", idx);
3569 
3570 	dep = ldep->dep;
3571 	ASSERT(dep != NULL);
3572 	ASSERT((dep->hdr.dstate == VIO_DESC_DONE) ||
3573 	    (dep->payload.status == ECANCELED));
3574 
3575 	VDC_MARK_DRING_ENTRY_FREE(vdc, idx);
3576 
3577 	ldep->is_free = B_TRUE;
3578 	status = dep->payload.status;
3579 	DMSG(vdc, 2, ": is_free = %d : status = %d\n", ldep->is_free, status);
3580 
3581 	/*
3582 	 * If no buffers were used to transfer information to the server when
3583 	 * populating the descriptor then no memory handles need to be unbound
3584 	 * and we can return now.
3585 	 */
3586 	if (ldep->nbytes == 0) {
3587 		cv_signal(&vdc->dring_free_cv);
3588 		return (status);
3589 	}
3590 
3591 	/*
3592 	 * If the upper layer passed in a misaligned address we copied the
3593 	 * data into an aligned buffer before sending it to LDC - we now
3594 	 * copy it back to the original buffer.
3595 	 */
3596 	if (ldep->align_addr) {
3597 		ASSERT(ldep->addr != NULL);
3598 
3599 		if (dep->payload.nbytes > 0)
3600 			bcopy(ldep->align_addr, ldep->addr,
3601 			    dep->payload.nbytes);
3602 		kmem_free(ldep->align_addr,
3603 		    sizeof (caddr_t) * P2ROUNDUP(ldep->nbytes, 8));
3604 		ldep->align_addr = NULL;
3605 	}
3606 
3607 	rv = ldc_mem_unbind_handle(ldep->desc_mhdl);
3608 	if (rv != 0) {
3609 		DMSG(vdc, 0, "?[%d] unbind mhdl 0x%lx @ idx %d failed (%d)",
3610 		    vdc->instance, ldep->desc_mhdl, idx, rv);
3611 		/*
3612 		 * The error returned by the vDisk server is more informative
3613 		 * and thus has a higher priority but if it isn't set we ensure
3614 		 * that this function returns an error.
3615 		 */
3616 		if (status == 0)
3617 			status = EINVAL;
3618 	}
3619 
3620 	cv_signal(&vdc->membind_cv);
3621 	cv_signal(&vdc->dring_free_cv);
3622 
3623 	return (status);
3624 }
3625 
3626 /*
3627  * Function:
3628  *	vdc_populate_mem_hdl()
3629  *
3630  * Description:
3631  *
3632  * Arguments:
3633  *	vdc	- soft state pointer for this instance of the device driver.
3634  *	idx	- Index of the Descriptor Ring entry being modified
3635  *	addr	- virtual address being mapped in
3636  *	nybtes	- number of bytes in 'addr'
3637  *	operation - the vDisk operation being performed (VD_OP_xxx)
3638  *
3639  * Return Code:
3640  *	0	- Success
3641  */
3642 static int
3643 vdc_populate_mem_hdl(vdc_t *vdcp, vdc_local_desc_t *ldep)
3644 {
3645 	vd_dring_entry_t	*dep = NULL;
3646 	ldc_mem_handle_t	mhdl;
3647 	caddr_t			vaddr;
3648 	size_t			nbytes;
3649 	uint8_t			perm = LDC_MEM_RW;
3650 	uint8_t			maptype;
3651 	int			rv = 0;
3652 	int			i;
3653 
3654 	ASSERT(vdcp != NULL);
3655 
3656 	dep = ldep->dep;
3657 	mhdl = ldep->desc_mhdl;
3658 
3659 	switch (ldep->dir) {
3660 	case VIO_read_dir:
3661 		perm = LDC_MEM_W;
3662 		break;
3663 
3664 	case VIO_write_dir:
3665 		perm = LDC_MEM_R;
3666 		break;
3667 
3668 	case VIO_both_dir:
3669 		perm = LDC_MEM_RW;
3670 		break;
3671 
3672 	default:
3673 		ASSERT(0);	/* catch bad programming in vdc */
3674 	}
3675 
3676 	/*
3677 	 * LDC expects any addresses passed in to be 8-byte aligned. We need
3678 	 * to copy the contents of any misaligned buffers to a newly allocated
3679 	 * buffer and bind it instead (and copy the the contents back to the
3680 	 * original buffer passed in when depopulating the descriptor)
3681 	 */
3682 	vaddr = ldep->addr;
3683 	nbytes = ldep->nbytes;
3684 	if (((uint64_t)vaddr & 0x7) != 0) {
3685 		ASSERT(ldep->align_addr == NULL);
3686 		ldep->align_addr =
3687 		    kmem_alloc(sizeof (caddr_t) *
3688 		    P2ROUNDUP(nbytes, 8), KM_SLEEP);
3689 		DMSG(vdcp, 0, "[%d] Misaligned address %p reallocating "
3690 		    "(buf=%p nb=%ld op=%d)\n",
3691 		    vdcp->instance, (void *)vaddr, (void *)ldep->align_addr,
3692 		    nbytes, ldep->operation);
3693 		if (perm != LDC_MEM_W)
3694 			bcopy(vaddr, ldep->align_addr, nbytes);
3695 		vaddr = ldep->align_addr;
3696 	}
3697 
3698 	maptype = LDC_IO_MAP|LDC_SHADOW_MAP;
3699 	rv = ldc_mem_bind_handle(mhdl, vaddr, P2ROUNDUP(nbytes, 8),
3700 	    maptype, perm, &dep->payload.cookie[0], &dep->payload.ncookies);
3701 	DMSG(vdcp, 2, "[%d] bound mem handle; ncookies=%d\n",
3702 	    vdcp->instance, dep->payload.ncookies);
3703 	if (rv != 0) {
3704 		DMSG(vdcp, 0, "[%d] Failed to bind LDC memory handle "
3705 		    "(mhdl=%p, buf=%p, err=%d)\n",
3706 		    vdcp->instance, (void *)mhdl, (void *)vaddr, rv);
3707 		if (ldep->align_addr) {
3708 			kmem_free(ldep->align_addr,
3709 			    sizeof (caddr_t) * P2ROUNDUP(nbytes, 8));
3710 			ldep->align_addr = NULL;
3711 		}
3712 		return (EAGAIN);
3713 	}
3714 
3715 	/*
3716 	 * Get the other cookies (if any).
3717 	 */
3718 	for (i = 1; i < dep->payload.ncookies; i++) {
3719 		rv = ldc_mem_nextcookie(mhdl, &dep->payload.cookie[i]);
3720 		if (rv != 0) {
3721 			(void) ldc_mem_unbind_handle(mhdl);
3722 			DMSG(vdcp, 0, "?[%d] Failed to get next cookie "
3723 			    "(mhdl=%lx cnum=%d), err=%d",
3724 			    vdcp->instance, mhdl, i, rv);
3725 			if (ldep->align_addr) {
3726 				kmem_free(ldep->align_addr,
3727 				    sizeof (caddr_t) * ldep->nbytes);
3728 				ldep->align_addr = NULL;
3729 			}
3730 			return (EAGAIN);
3731 		}
3732 	}
3733 
3734 	return (rv);
3735 }
3736 
3737 /*
3738  * Interrupt handlers for messages from LDC
3739  */
3740 
3741 /*
3742  * Function:
3743  *	vdc_handle_cb()
3744  *
3745  * Description:
3746  *
3747  * Arguments:
3748  *	event	- Type of event (LDC_EVT_xxx) that triggered the callback
3749  *	arg	- soft state pointer for this instance of the device driver.
3750  *
3751  * Return Code:
3752  *	0	- Success
3753  */
3754 static uint_t
3755 vdc_handle_cb(uint64_t event, caddr_t arg)
3756 {
3757 	ldc_status_t	ldc_state;
3758 	int		rv = 0;
3759 	vdc_server_t	*srvr = (vdc_server_t *)(void *)arg;
3760 	vdc_t		*vdc = srvr->vdcp;
3761 
3762 	ASSERT(vdc != NULL);
3763 
3764 	DMSG(vdc, 1, "evt=%lx seqID=%ld\n", event, vdc->seq_num);
3765 
3766 	/* If callback is not for the current server, ignore it */
3767 	mutex_enter(&vdc->lock);
3768 
3769 	if (vdc->curr_server != srvr) {
3770 		DMSG(vdc, 0, "[%d] Ignoring event 0x%lx for port@%ld\n",
3771 		    vdc->instance, event, srvr->id);
3772 		mutex_exit(&vdc->lock);
3773 		return (LDC_SUCCESS);
3774 	}
3775 
3776 	/*
3777 	 * Depending on the type of event that triggered this callback,
3778 	 * we modify the handshake state or read the data.
3779 	 *
3780 	 * NOTE: not done as a switch() as event could be triggered by
3781 	 * a state change and a read request. Also the ordering	of the
3782 	 * check for the event types is deliberate.
3783 	 */
3784 	if (event & LDC_EVT_UP) {
3785 		DMSG(vdc, 0, "[%d] Received LDC_EVT_UP\n", vdc->instance);
3786 
3787 		/* get LDC state */
3788 		rv = ldc_status(srvr->ldc_handle, &ldc_state);
3789 		if (rv != 0) {
3790 			DMSG(vdc, 0, "[%d] Couldn't get LDC status %d",
3791 			    vdc->instance, rv);
3792 			mutex_exit(&vdc->lock);
3793 			return (LDC_SUCCESS);
3794 		}
3795 		if (srvr->ldc_state != LDC_UP &&
3796 		    ldc_state == LDC_UP) {
3797 			/*
3798 			 * Reset the transaction sequence numbers when
3799 			 * LDC comes up. We then kick off the handshake
3800 			 * negotiation with the vDisk server.
3801 			 */
3802 			vdc->seq_num = 1;
3803 			vdc->seq_num_reply = 0;
3804 			vdc->io_pending = B_TRUE;
3805 			srvr->ldc_state = ldc_state;
3806 			cv_signal(&vdc->initwait_cv);
3807 			cv_signal(&vdc->io_pending_cv);
3808 		}
3809 	}
3810 
3811 	if (event & LDC_EVT_READ) {
3812 		DMSG(vdc, 1, "[%d] Received LDC_EVT_READ\n", vdc->instance);
3813 		mutex_enter(&vdc->read_lock);
3814 		cv_signal(&vdc->read_cv);
3815 		vdc->read_state = VDC_READ_PENDING;
3816 		mutex_exit(&vdc->read_lock);
3817 		mutex_exit(&vdc->lock);
3818 
3819 		/* that's all we have to do - no need to handle DOWN/RESET */
3820 		return (LDC_SUCCESS);
3821 	}
3822 
3823 	if (event & (LDC_EVT_RESET|LDC_EVT_DOWN)) {
3824 
3825 		DMSG(vdc, 0, "[%d] Received LDC RESET event\n", vdc->instance);
3826 
3827 		/*
3828 		 * Need to wake up any readers so they will
3829 		 * detect that a reset has occurred.
3830 		 */
3831 		mutex_enter(&vdc->read_lock);
3832 		if ((vdc->read_state == VDC_READ_WAITING) ||
3833 		    (vdc->read_state == VDC_READ_RESET))
3834 			cv_signal(&vdc->read_cv);
3835 		vdc->read_state = VDC_READ_RESET;
3836 		mutex_exit(&vdc->read_lock);
3837 
3838 		/* wake up any threads waiting for connection to come up */
3839 		if (vdc->state == VDC_STATE_INIT_WAITING) {
3840 			vdc->state = VDC_STATE_RESETTING;
3841 			cv_signal(&vdc->initwait_cv);
3842 		} else if (vdc->state == VDC_STATE_FAILED) {
3843 			vdc->io_pending = B_TRUE;
3844 			cv_signal(&vdc->io_pending_cv);
3845 		}
3846 
3847 	}
3848 
3849 	mutex_exit(&vdc->lock);
3850 
3851 	if (event & ~(LDC_EVT_UP | LDC_EVT_RESET | LDC_EVT_DOWN | LDC_EVT_READ))
3852 		DMSG(vdc, 0, "![%d] Unexpected LDC event (%lx) received",
3853 		    vdc->instance, event);
3854 
3855 	return (LDC_SUCCESS);
3856 }
3857 
3858 /*
3859  * Function:
3860  *	vdc_wait_for_response()
3861  *
3862  * Description:
3863  *	Block waiting for a response from the server. If there is
3864  *	no data the thread block on the read_cv that is signalled
3865  *	by the callback when an EVT_READ occurs.
3866  *
3867  * Arguments:
3868  *	vdcp	- soft state pointer for this instance of the device driver.
3869  *
3870  * Return Code:
3871  *	0	- Success
3872  */
3873 static int
3874 vdc_wait_for_response(vdc_t *vdcp, vio_msg_t *msgp)
3875 {
3876 	size_t		nbytes = sizeof (*msgp);
3877 	int		status;
3878 
3879 	ASSERT(vdcp != NULL);
3880 
3881 	DMSG(vdcp, 1, "[%d] Entered\n", vdcp->instance);
3882 
3883 	status = vdc_recv(vdcp, msgp, &nbytes);
3884 	DMSG(vdcp, 3, "vdc_read() done.. status=0x%x size=0x%x\n",
3885 	    status, (int)nbytes);
3886 	if (status) {
3887 		DMSG(vdcp, 0, "?[%d] Error %d reading LDC msg\n",
3888 		    vdcp->instance, status);
3889 		return (status);
3890 	}
3891 
3892 	if (nbytes < sizeof (vio_msg_tag_t)) {
3893 		DMSG(vdcp, 0, "?[%d] Expect %lu bytes; recv'd %lu\n",
3894 		    vdcp->instance, sizeof (vio_msg_tag_t), nbytes);
3895 		return (ENOMSG);
3896 	}
3897 
3898 	DMSG(vdcp, 2, "[%d] (%x/%x/%x)\n", vdcp->instance,
3899 	    msgp->tag.vio_msgtype,
3900 	    msgp->tag.vio_subtype,
3901 	    msgp->tag.vio_subtype_env);
3902 
3903 	/*
3904 	 * Verify the Session ID of the message
3905 	 *
3906 	 * Every message after the Version has been negotiated should
3907 	 * have the correct session ID set.
3908 	 */
3909 	if ((msgp->tag.vio_sid != vdcp->session_id) &&
3910 	    (msgp->tag.vio_subtype_env != VIO_VER_INFO)) {
3911 		DMSG(vdcp, 0, "[%d] Invalid SID: received 0x%x, "
3912 		    "expected 0x%lx [seq num %lx @ %d]",
3913 		    vdcp->instance, msgp->tag.vio_sid,
3914 		    vdcp->session_id,
3915 		    ((vio_dring_msg_t *)msgp)->seq_num,
3916 		    ((vio_dring_msg_t *)msgp)->start_idx);
3917 		return (ENOMSG);
3918 	}
3919 	return (0);
3920 }
3921 
3922 
3923 /*
3924  * Function:
3925  *	vdc_resubmit_backup_dring()
3926  *
3927  * Description:
3928  *	Resubmit each descriptor in the backed up dring to
3929  *	vDisk server. The Dring was backed up during connection
3930  *	reset.
3931  *
3932  * Arguments:
3933  *	vdcp	- soft state pointer for this instance of the device driver.
3934  *
3935  * Return Code:
3936  *	0	- Success
3937  */
3938 static int
3939 vdc_resubmit_backup_dring(vdc_t *vdcp)
3940 {
3941 	int		processed = 0;
3942 	int		count;
3943 	int		b_idx;
3944 	int		rv = 0;
3945 	int		dring_size;
3946 	vdc_local_desc_t	*curr_ldep;
3947 
3948 	ASSERT(MUTEX_NOT_HELD(&vdcp->lock));
3949 	ASSERT(vdcp->state == VDC_STATE_HANDLE_PENDING);
3950 
3951 	if (vdcp->local_dring_backup == NULL) {
3952 		/* the pending requests have already been processed */
3953 		return (0);
3954 	}
3955 
3956 	DMSG(vdcp, 1, "restoring pending dring entries (len=%d, tail=%d)\n",
3957 	    vdcp->local_dring_backup_len, vdcp->local_dring_backup_tail);
3958 
3959 	/*
3960 	 * Walk the backup copy of the local descriptor ring and
3961 	 * resubmit all the outstanding transactions.
3962 	 */
3963 	b_idx = vdcp->local_dring_backup_tail;
3964 	for (count = 0; count < vdcp->local_dring_backup_len; count++) {
3965 
3966 		curr_ldep = &(vdcp->local_dring_backup[b_idx]);
3967 
3968 		/* only resubmit outstanding transactions */
3969 		if (!curr_ldep->is_free) {
3970 
3971 			DMSG(vdcp, 1, "resubmitting entry idx=%x\n", b_idx);
3972 
3973 			rv = vdc_do_op(vdcp, curr_ldep->operation,
3974 			    curr_ldep->addr, curr_ldep->nbytes,
3975 			    curr_ldep->slice, curr_ldep->offset,
3976 			    curr_ldep->buf, curr_ldep->dir,
3977 			    (curr_ldep->flags & ~VDC_OP_STATE_RUNNING) |
3978 			    VDC_OP_RESUBMIT);
3979 
3980 			if (rv) {
3981 				DMSG(vdcp, 1, "[%d] resubmit entry %d failed\n",
3982 				    vdcp->instance, b_idx);
3983 				goto done;
3984 			}
3985 
3986 			/*
3987 			 * Mark this entry as free so that we will not resubmit
3988 			 * this "done" request again, if we were to use the same
3989 			 * backup_dring again in future. This could happen when
3990 			 * a reset happens while processing the backup_dring.
3991 			 */
3992 			curr_ldep->is_free = B_TRUE;
3993 			processed++;
3994 		}
3995 
3996 		/* get the next element to submit */
3997 		if (++b_idx >= vdcp->local_dring_backup_len)
3998 			b_idx = 0;
3999 	}
4000 
4001 	/* all done - now clear up pending dring copy */
4002 	dring_size = vdcp->local_dring_backup_len *
4003 	    sizeof (vdcp->local_dring_backup[0]);
4004 
4005 	(void) kmem_free(vdcp->local_dring_backup, dring_size);
4006 
4007 	vdcp->local_dring_backup = NULL;
4008 
4009 done:
4010 	DTRACE_PROBE2(processed, int, processed, vdc_t *, vdcp);
4011 
4012 	return (rv);
4013 }
4014 
4015 /*
4016  * Function:
4017  *	vdc_cancel_backup_dring
4018  *
4019  * Description:
4020  *	Cancel each descriptor in the backed up dring to vDisk server.
4021  *	The Dring was backed up during connection reset.
4022  *
4023  * Arguments:
4024  *	vdcp	- soft state pointer for this instance of the device driver.
4025  *
4026  * Return Code:
4027  *	None
4028  */
4029 void
4030 vdc_cancel_backup_dring(vdc_t *vdcp)
4031 {
4032 	vdc_local_desc_t *ldep;
4033 	struct buf	*bufp;
4034 	int		count;
4035 	int		b_idx;
4036 	int		dring_size;
4037 	int		cancelled = 0;
4038 
4039 	ASSERT(MUTEX_HELD(&vdcp->lock));
4040 	ASSERT(vdcp->state == VDC_STATE_FAILED);
4041 
4042 	if (vdcp->local_dring_backup == NULL) {
4043 		/* the pending requests have already been processed */
4044 		return;
4045 	}
4046 
4047 	DMSG(vdcp, 1, "cancelling pending dring entries (len=%d, tail=%d)\n",
4048 	    vdcp->local_dring_backup_len, vdcp->local_dring_backup_tail);
4049 
4050 	/*
4051 	 * Walk the backup copy of the local descriptor ring and
4052 	 * cancel all the outstanding transactions.
4053 	 */
4054 	b_idx = vdcp->local_dring_backup_tail;
4055 	for (count = 0; count < vdcp->local_dring_backup_len; count++) {
4056 
4057 		ldep = &(vdcp->local_dring_backup[b_idx]);
4058 
4059 		/* only cancel outstanding transactions */
4060 		if (!ldep->is_free) {
4061 
4062 			DMSG(vdcp, 1, "cancelling entry idx=%x\n", b_idx);
4063 			cancelled++;
4064 
4065 			/*
4066 			 * All requests have already been cleared from the
4067 			 * local descriptor ring and the LDC channel has been
4068 			 * reset so we will never get any reply for these
4069 			 * requests. Now we just have to notify threads waiting
4070 			 * for replies that the request has failed.
4071 			 */
4072 			bufp = ldep->buf;
4073 			ASSERT(bufp != NULL);
4074 			bufp->b_resid = bufp->b_bcount;
4075 			if (ldep->operation == VD_OP_BREAD ||
4076 			    ldep->operation == VD_OP_BWRITE) {
4077 				VD_UPDATE_ERR_STATS(vdcp, vd_softerrs);
4078 				VD_KSTAT_WAITQ_EXIT(vdcp);
4079 				DTRACE_IO1(done, buf_t *, bufp);
4080 			}
4081 			bioerror(bufp, EIO);
4082 			biodone(bufp);
4083 		}
4084 
4085 		/* get the next element to cancel */
4086 		if (++b_idx >= vdcp->local_dring_backup_len)
4087 			b_idx = 0;
4088 	}
4089 
4090 	/* all done - now clear up pending dring copy */
4091 	dring_size = vdcp->local_dring_backup_len *
4092 	    sizeof (vdcp->local_dring_backup[0]);
4093 
4094 	(void) kmem_free(vdcp->local_dring_backup, dring_size);
4095 
4096 	vdcp->local_dring_backup = NULL;
4097 
4098 	DTRACE_PROBE2(cancelled, int, cancelled, vdc_t *, vdcp);
4099 }
4100 
4101 /*
4102  * Function:
4103  *	vdc_connection_timeout
4104  *
4105  * Description:
4106  *	This function is invoked if the timeout set to establish the connection
4107  *	with vds expires. This will happen if we spend too much time in the
4108  *	VDC_STATE_INIT_WAITING, VDC_STATE_NEGOTIATE or VDC_STATE_HANDLE_PENDING
4109  *	states.
4110  *
4111  * Arguments:
4112  *	arg	- argument of the timeout function actually a soft state
4113  *		  pointer for the instance of the device driver.
4114  *
4115  * Return Code:
4116  *	None
4117  */
4118 void
4119 vdc_connection_timeout(void *arg)
4120 {
4121 	vdc_t		*vdcp = (vdc_t *)arg;
4122 
4123 	mutex_enter(&vdcp->lock);
4124 
4125 	vdcp->ctimeout_reached = B_TRUE;
4126 
4127 	mutex_exit(&vdcp->lock);
4128 }
4129 
4130 /*
4131  * Function:
4132  *	vdc_backup_local_dring()
4133  *
4134  * Description:
4135  *	Backup the current dring in the event of a reset. The Dring
4136  *	transactions will be resubmitted to the server when the
4137  *	connection is restored.
4138  *
4139  * Arguments:
4140  *	vdcp	- soft state pointer for this instance of the device driver.
4141  *
4142  * Return Code:
4143  *	NONE
4144  */
4145 static void
4146 vdc_backup_local_dring(vdc_t *vdcp)
4147 {
4148 	int b_idx, count, dring_size;
4149 	vdc_local_desc_t *curr_ldep;
4150 
4151 	ASSERT(MUTEX_HELD(&vdcp->lock));
4152 	ASSERT(vdcp->state == VDC_STATE_RESETTING);
4153 
4154 	/*
4155 	 * If the backup dring is stil around, it means
4156 	 * that the last restore did not complete. However,
4157 	 * since we never got back into the running state,
4158 	 * the backup copy we have is still valid.
4159 	 */
4160 	if (vdcp->local_dring_backup != NULL) {
4161 		DMSG(vdcp, 1, "reusing local descriptor ring backup "
4162 		    "(len=%d, tail=%d)\n", vdcp->local_dring_backup_len,
4163 		    vdcp->local_dring_backup_tail);
4164 		return;
4165 	}
4166 
4167 	/*
4168 	 * The backup dring can be NULL and the local dring may not be
4169 	 * initialized. This can happen if we had a reset while establishing
4170 	 * a new connection but after the connection has timed out. In that
4171 	 * case the backup dring is NULL because the requests have been
4172 	 * cancelled and the request occured before the local dring is
4173 	 * initialized.
4174 	 */
4175 	if (!(vdcp->initialized & VDC_DRING_LOCAL))
4176 		return;
4177 
4178 	DMSG(vdcp, 1, "backing up the local descriptor ring (len=%d, "
4179 	    "tail=%d)\n", vdcp->dring_len, vdcp->dring_curr_idx);
4180 
4181 	dring_size = vdcp->dring_len * sizeof (vdcp->local_dring[0]);
4182 
4183 	vdcp->local_dring_backup = kmem_alloc(dring_size, KM_SLEEP);
4184 	bcopy(vdcp->local_dring, vdcp->local_dring_backup, dring_size);
4185 
4186 	vdcp->local_dring_backup_tail = vdcp->dring_curr_idx;
4187 	vdcp->local_dring_backup_len = vdcp->dring_len;
4188 
4189 	/*
4190 	 * At this point, pending read or write I/Os are recorded in the
4191 	 * runq. We update the I/O statistics to indicate that they are now
4192 	 * back in the waitq.
4193 	 */
4194 	b_idx = vdcp->local_dring_backup_tail;
4195 	for (count = 0; count < vdcp->local_dring_backup_len; count++) {
4196 
4197 		curr_ldep = &(vdcp->local_dring_backup[b_idx]);
4198 
4199 		if (!curr_ldep->is_free &&
4200 		    (curr_ldep->operation == VD_OP_BREAD ||
4201 		    curr_ldep->operation == VD_OP_BWRITE)) {
4202 			VD_KSTAT_RUNQ_BACK_TO_WAITQ(vdcp);
4203 		}
4204 
4205 		/* get the next element */
4206 		if (++b_idx >= vdcp->local_dring_backup_len)
4207 			b_idx = 0;
4208 	}
4209 
4210 }
4211 
4212 static void
4213 vdc_switch_server(vdc_t *vdcp)
4214 {
4215 	int		rv;
4216 	vdc_server_t	*curr_server, *new_server;
4217 
4218 	ASSERT(MUTEX_HELD(&vdcp->lock));
4219 
4220 	/* if there is only one server return back */
4221 	if (vdcp->num_servers == 1) {
4222 		return;
4223 	}
4224 
4225 	/* Get current and next server */
4226 	curr_server = vdcp->curr_server;
4227 	new_server =
4228 	    (curr_server->next) ? curr_server->next : vdcp->server_list;
4229 	ASSERT(curr_server != new_server);
4230 
4231 	/* bring current server's channel down */
4232 	rv = ldc_down(curr_server->ldc_handle);
4233 	if (rv) {
4234 		DMSG(vdcp, 0, "[%d] Cannot bring channel down, port %ld\n",
4235 		    vdcp->instance, curr_server->id);
4236 		return;
4237 	}
4238 
4239 	/* switch the server */
4240 	vdcp->curr_server = new_server;
4241 
4242 	DMSG(vdcp, 0, "[%d] Switched to next vdisk server, port@%ld, ldc@%ld\n",
4243 	    vdcp->instance, vdcp->curr_server->id, vdcp->curr_server->ldc_id);
4244 }
4245 
4246 static void
4247 vdc_print_svc_status(vdc_t *vdcp)
4248 {
4249 	int instance;
4250 	uint64_t ldc_id, port_id;
4251 	vdc_service_state_t svc_state;
4252 
4253 	ASSERT(mutex_owned(&vdcp->lock));
4254 
4255 	svc_state = vdcp->curr_server->svc_state;
4256 
4257 	if (vdcp->curr_server->log_state == svc_state)
4258 		return;
4259 
4260 	instance = vdcp->instance;
4261 	ldc_id = vdcp->curr_server->ldc_id;
4262 	port_id = vdcp->curr_server->id;
4263 
4264 	switch (svc_state) {
4265 
4266 	case VDC_SERVICE_OFFLINE:
4267 		cmn_err(CE_CONT, "?vdisk@%d is offline\n", instance);
4268 		break;
4269 
4270 	case VDC_SERVICE_CONNECTED:
4271 		cmn_err(CE_CONT, "?vdisk@%d is connected using ldc@%ld,%ld\n",
4272 		    instance, ldc_id, port_id);
4273 		break;
4274 
4275 	case VDC_SERVICE_ONLINE:
4276 		cmn_err(CE_CONT, "?vdisk@%d is online using ldc@%ld,%ld\n",
4277 		    instance, ldc_id, port_id);
4278 		break;
4279 
4280 	case VDC_SERVICE_FAILED:
4281 		cmn_err(CE_CONT, "?vdisk@%d access to service failed "
4282 		    "using ldc@%ld,%ld\n", instance, ldc_id, port_id);
4283 		break;
4284 
4285 	case VDC_SERVICE_FAULTED:
4286 		cmn_err(CE_CONT, "?vdisk@%d access to backend failed "
4287 		    "using ldc@%ld,%ld\n", instance, ldc_id, port_id);
4288 		break;
4289 
4290 	default:
4291 		ASSERT(0);
4292 		break;
4293 	}
4294 
4295 	vdcp->curr_server->log_state = svc_state;
4296 }
4297 
4298 /*
4299  * Function:
4300  *	vdc_handshake_retry
4301  *
4302  * Description:
4303  *	This function indicates if the handshake should be retried or not.
4304  *	This depends on the lifecycle of the driver:
4305  *
4306  *	VDC_LC_ATTACHING: the handshake is retried until we have tried
4307  *	a handshake with each server. We don't care how far each handshake
4308  *	went, the goal is just to try the handshake. We want to minimize the
4309  *	the time spent doing the attach because this is locking the device
4310  *	tree.
4311  *
4312  *	VDC_LC_ONLINE_PENDING: the handshake is retried while we haven't done
4313  *	consecutive attribute negotiations with each server, and we haven't
4314  *	reached a minimum total of consecutive negotiations (hattr_min). The
4315  *	number of attribution negotiations determines the time spent before
4316  *	failing	pending I/Os if the handshake is not successful.
4317  *
4318  *	VDC_LC_ONLINE: the handshake is always retried, until we have a
4319  *	successful handshake with a server.
4320  *
4321  *	VDC_LC_DETACHING: N/A
4322  *
4323  * Arguments:
4324  *	hshake_cnt	- number of handshake attempts
4325  *	hattr_cnt	- number of attribute negotiation attempts
4326  *
4327  * Return Code:
4328  *	B_TRUE		- handshake should be retried
4329  *	B_FALSE		- handshake should not be retried
4330  */
4331 static boolean_t
4332 vdc_handshake_retry(vdc_t *vdcp, int hshake_cnt, int hattr_cnt)
4333 {
4334 	int		hattr_total = 0;
4335 	vdc_server_t	*srvr;
4336 
4337 	ASSERT(vdcp->lifecycle != VDC_LC_DETACHING);
4338 
4339 	/* update handshake counters */
4340 	vdcp->curr_server->hshake_cnt = hshake_cnt;
4341 	vdcp->curr_server->hattr_cnt = hattr_cnt;
4342 
4343 	/*
4344 	 * If no attribute negotiation was done then we reset the total
4345 	 *  number otherwise we cumulate the number.
4346 	 */
4347 	if (hattr_cnt == 0)
4348 		vdcp->curr_server->hattr_total = 0;
4349 	else
4350 		vdcp->curr_server->hattr_total += hattr_cnt;
4351 
4352 	/*
4353 	 * If we are online (i.e. at least one handshake was successfully
4354 	 * completed) then we always retry the handshake.
4355 	 */
4356 	if (vdcp->lifecycle == VDC_LC_ONLINE)
4357 		return (B_TRUE);
4358 
4359 	/*
4360 	 * If we are attaching then we retry the handshake only if we haven't
4361 	 * tried with all servers.
4362 	 */
4363 	if (vdcp->lifecycle == VDC_LC_ATTACHING) {
4364 
4365 		for (srvr = vdcp->server_list; srvr != NULL;
4366 		    srvr = srvr->next) {
4367 			if (srvr->hshake_cnt == 0) {
4368 				return (B_TRUE);
4369 			}
4370 		}
4371 
4372 		return (B_FALSE);
4373 	}
4374 
4375 	/*
4376 	 * Here we are in the case where we haven't completed any handshake
4377 	 * successfully yet.
4378 	 */
4379 	ASSERT(vdcp->lifecycle == VDC_LC_ONLINE_PENDING);
4380 
4381 	/*
4382 	 * We retry the handshake if we haven't done an attribute negotiation
4383 	 * with each server. This is to handle the case where one service domain
4384 	 * is down.
4385 	 */
4386 	for (srvr = vdcp->server_list; srvr != NULL; srvr = srvr->next) {
4387 		if (srvr->hattr_cnt == 0) {
4388 			return (B_TRUE);
4389 		}
4390 		hattr_total += srvr->hattr_total;
4391 	}
4392 
4393 	/*
4394 	 * We retry the handshake if we haven't reached the minimum number of
4395 	 * attribute negotiation.
4396 	 */
4397 	return (hattr_total < vdcp->hattr_min);
4398 }
4399 
4400 /* -------------------------------------------------------------------------- */
4401 
4402 /*
4403  * The following functions process the incoming messages from vds
4404  */
4405 
4406 /*
4407  * Function:
4408  *      vdc_process_msg_thread()
4409  *
4410  * Description:
4411  *
4412  *	Main VDC message processing thread. Each vDisk instance
4413  *	consists of a copy of this thread. This thread triggers
4414  *	all the handshakes and data exchange with the server. It
4415  *	also handles all channel resets
4416  *
4417  * Arguments:
4418  *      vdc     - soft state pointer for this instance of the device driver.
4419  *
4420  * Return Code:
4421  *      None
4422  */
4423 static void
4424 vdc_process_msg_thread(vdc_t *vdcp)
4425 {
4426 	boolean_t	failure_msg = B_FALSE;
4427 	int		status;
4428 	int		ctimeout;
4429 	timeout_id_t	tmid = 0;
4430 	clock_t		ldcup_timeout = 0;
4431 	vdc_server_t	*srvr;
4432 	vdc_service_state_t svc_state;
4433 	int		hshake_cnt = 0;
4434 	int		hattr_cnt = 0;
4435 
4436 	mutex_enter(&vdcp->lock);
4437 
4438 	ASSERT(vdcp->lifecycle == VDC_LC_ATTACHING);
4439 
4440 	for (;;) {
4441 
4442 #define	Q(_s)	(vdcp->state == _s) ? #_s :
4443 		DMSG(vdcp, 3, "state = %d (%s)\n", vdcp->state,
4444 		    Q(VDC_STATE_INIT)
4445 		    Q(VDC_STATE_INIT_WAITING)
4446 		    Q(VDC_STATE_NEGOTIATE)
4447 		    Q(VDC_STATE_HANDLE_PENDING)
4448 		    Q(VDC_STATE_FAULTED)
4449 		    Q(VDC_STATE_FAILED)
4450 		    Q(VDC_STATE_RUNNING)
4451 		    Q(VDC_STATE_RESETTING)
4452 		    Q(VDC_STATE_DETACH)
4453 		    "UNKNOWN");
4454 #undef Q
4455 
4456 		switch (vdcp->state) {
4457 		case VDC_STATE_INIT:
4458 
4459 			/*
4460 			 * If requested, start a timeout to check if the
4461 			 * connection with vds is established in the
4462 			 * specified delay. If the timeout expires, we
4463 			 * will cancel any pending request.
4464 			 *
4465 			 * If some reset have occurred while establishing
4466 			 * the connection, we already have a timeout armed
4467 			 * and in that case we don't need to arm a new one.
4468 			 *
4469 			 * The same rule applies when there are multiple vds'.
4470 			 * If either a connection cannot be established or
4471 			 * the handshake times out, the connection thread will
4472 			 * try another server. The 'ctimeout' will report
4473 			 * back an error after it expires irrespective of
4474 			 * whether the vdisk is trying to connect to just
4475 			 * one or multiple servers.
4476 			 */
4477 			ctimeout = (vdc_timeout != 0)?
4478 			    vdc_timeout : vdcp->curr_server->ctimeout;
4479 
4480 			if (ctimeout != 0 && tmid == 0) {
4481 				tmid = timeout(vdc_connection_timeout, vdcp,
4482 				    ctimeout * drv_usectohz(MICROSEC));
4483 			}
4484 
4485 			/* Switch to STATE_DETACH if drv is detaching */
4486 			if (vdcp->lifecycle == VDC_LC_DETACHING) {
4487 				vdcp->state = VDC_STATE_DETACH;
4488 				break;
4489 			}
4490 
4491 			/* Check if the timeout has been reached */
4492 			if (vdcp->ctimeout_reached) {
4493 				ASSERT(tmid != 0);
4494 				tmid = 0;
4495 				vdcp->state = VDC_STATE_FAILED;
4496 				break;
4497 			}
4498 
4499 			/*
4500 			 * Switch to another server when we reach the limit of
4501 			 * the number of handshake per server or if we have done
4502 			 * an attribute negotiation.
4503 			 */
4504 			if (hshake_cnt >= vdc_hshake_retries || hattr_cnt > 0) {
4505 
4506 				if (!vdc_handshake_retry(vdcp, hshake_cnt,
4507 				    hattr_cnt)) {
4508 					DMSG(vdcp, 0, "[%d] too many "
4509 					    "handshakes", vdcp->instance);
4510 					vdcp->state = VDC_STATE_FAILED;
4511 					break;
4512 				}
4513 
4514 				vdc_switch_server(vdcp);
4515 
4516 				hshake_cnt = 0;
4517 				hattr_cnt = 0;
4518 			}
4519 
4520 			hshake_cnt++;
4521 
4522 			/* Bring up connection with vds via LDC */
4523 			status = vdc_start_ldc_connection(vdcp);
4524 			if (status != EINVAL) {
4525 				vdcp->state = VDC_STATE_INIT_WAITING;
4526 			} else {
4527 				vdcp->curr_server->svc_state =
4528 				    VDC_SERVICE_FAILED;
4529 				vdc_print_svc_status(vdcp);
4530 			}
4531 			break;
4532 
4533 		case VDC_STATE_INIT_WAITING:
4534 
4535 			/* if channel is UP, start negotiation */
4536 			if (vdcp->curr_server->ldc_state == LDC_UP) {
4537 				vdcp->state = VDC_STATE_NEGOTIATE;
4538 				break;
4539 			}
4540 
4541 			/*
4542 			 * Wait for LDC_UP. If it times out and we have multiple
4543 			 * servers then we will retry using a different server.
4544 			 */
4545 			ldcup_timeout = ddi_get_lbolt() + (vdc_ldcup_timeout *
4546 			    drv_usectohz(MICROSEC));
4547 			status = cv_timedwait(&vdcp->initwait_cv, &vdcp->lock,
4548 			    ldcup_timeout);
4549 			if (status == -1 &&
4550 			    vdcp->state == VDC_STATE_INIT_WAITING &&
4551 			    vdcp->curr_server->ldc_state != LDC_UP) {
4552 				/* timed out & still waiting */
4553 				vdcp->curr_server->svc_state =
4554 				    VDC_SERVICE_FAILED;
4555 				vdc_print_svc_status(vdcp);
4556 				vdcp->state = VDC_STATE_INIT;
4557 				break;
4558 			}
4559 
4560 			if (vdcp->state != VDC_STATE_INIT_WAITING) {
4561 				DMSG(vdcp, 0,
4562 				    "state moved to %d out from under us...\n",
4563 				    vdcp->state);
4564 			}
4565 			break;
4566 
4567 		case VDC_STATE_NEGOTIATE:
4568 			switch (status = vdc_ver_negotiation(vdcp)) {
4569 			case 0:
4570 				break;
4571 			default:
4572 				DMSG(vdcp, 0, "ver negotiate failed (%d)..\n",
4573 				    status);
4574 				goto reset;
4575 			}
4576 
4577 			hattr_cnt++;
4578 
4579 			switch (status = vdc_attr_negotiation(vdcp)) {
4580 			case 0:
4581 				break;
4582 			default:
4583 				DMSG(vdcp, 0, "attr negotiate failed (%d)..\n",
4584 				    status);
4585 				goto reset;
4586 			}
4587 
4588 			switch (status = vdc_dring_negotiation(vdcp)) {
4589 			case 0:
4590 				break;
4591 			default:
4592 				DMSG(vdcp, 0, "dring negotiate failed (%d)..\n",
4593 				    status);
4594 				goto reset;
4595 			}
4596 
4597 			switch (status = vdc_rdx_exchange(vdcp)) {
4598 			case 0:
4599 				vdcp->state = VDC_STATE_HANDLE_PENDING;
4600 				goto done;
4601 			default:
4602 				DMSG(vdcp, 0, "RDX xchg failed ..(%d)\n",
4603 				    status);
4604 				goto reset;
4605 			}
4606 reset:
4607 			DMSG(vdcp, 0, "negotiation failed: resetting (%d)\n",
4608 			    status);
4609 			vdcp->state = VDC_STATE_RESETTING;
4610 			vdcp->self_reset = B_TRUE;
4611 			vdcp->curr_server->svc_state = VDC_SERVICE_FAILED;
4612 			vdc_print_svc_status(vdcp);
4613 done:
4614 			DMSG(vdcp, 0, "negotiation complete (state=0x%x)...\n",
4615 			    vdcp->state);
4616 			break;
4617 
4618 		case VDC_STATE_HANDLE_PENDING:
4619 
4620 			DMSG(vdcp, 0, "[%d] connection to service domain is up",
4621 			    vdcp->instance);
4622 			vdcp->curr_server->svc_state = VDC_SERVICE_CONNECTED;
4623 
4624 			mutex_exit(&vdcp->lock);
4625 
4626 			/*
4627 			 * If we have multiple servers, check that the backend
4628 			 * is effectively available before resubmitting any IO.
4629 			 */
4630 			if (vdcp->num_servers > 1 &&
4631 			    vdc_eio_check(vdcp, 0) != 0) {
4632 				mutex_enter(&vdcp->lock);
4633 				vdcp->curr_server->svc_state =
4634 				    VDC_SERVICE_FAULTED;
4635 				vdcp->state = VDC_STATE_FAULTED;
4636 				break;
4637 			}
4638 
4639 			if (tmid != 0) {
4640 				(void) untimeout(tmid);
4641 				tmid = 0;
4642 				vdcp->ctimeout_reached = B_FALSE;
4643 			}
4644 
4645 			/*
4646 			 * Setup devid
4647 			 */
4648 			(void) vdc_setup_devid(vdcp);
4649 
4650 			status = vdc_resubmit_backup_dring(vdcp);
4651 
4652 			mutex_enter(&vdcp->lock);
4653 
4654 			if (status) {
4655 				vdcp->state = VDC_STATE_RESETTING;
4656 				vdcp->self_reset = B_TRUE;
4657 				vdcp->curr_server->svc_state =
4658 				    VDC_SERVICE_FAILED;
4659 				vdc_print_svc_status(vdcp);
4660 			} else {
4661 				vdcp->state = VDC_STATE_RUNNING;
4662 			}
4663 			break;
4664 
4665 		case VDC_STATE_FAULTED:
4666 			/*
4667 			 * Server is faulted because the backend is unavailable.
4668 			 * If all servers are faulted then we mark the service
4669 			 * as failed, otherwise we reset to switch to another
4670 			 * server.
4671 			 */
4672 			vdc_print_svc_status(vdcp);
4673 
4674 			/* check if all servers are faulted */
4675 			for (srvr = vdcp->server_list; srvr != NULL;
4676 			    srvr = srvr->next) {
4677 				svc_state = srvr->svc_state;
4678 				if (svc_state != VDC_SERVICE_FAULTED)
4679 					break;
4680 			}
4681 
4682 			if (srvr != NULL) {
4683 				vdcp->state = VDC_STATE_RESETTING;
4684 				vdcp->self_reset = B_TRUE;
4685 			} else {
4686 				vdcp->state = VDC_STATE_FAILED;
4687 			}
4688 			break;
4689 
4690 		case VDC_STATE_FAILED:
4691 			/*
4692 			 * We reach this state when we are unable to access the
4693 			 * backend from any server, either because of a maximum
4694 			 * connection retries or timeout, or because the backend
4695 			 * is unavailable.
4696 			 *
4697 			 * Then we cancel the backup DRing so that errors get
4698 			 * reported and we wait for a new I/O before attempting
4699 			 * another connection.
4700 			 */
4701 
4702 			cmn_err(CE_NOTE, "vdisk@%d disk access failed",
4703 			    vdcp->instance);
4704 			failure_msg = B_TRUE;
4705 
4706 			if (vdcp->lifecycle == VDC_LC_ATTACHING) {
4707 				vdcp->lifecycle = VDC_LC_ONLINE_PENDING;
4708 				vdcp->hattr_min = vdc_hattr_min_initial;
4709 			} else {
4710 				vdcp->hattr_min = vdc_hattr_min;
4711 			}
4712 
4713 			/* cancel any timeout */
4714 			if (tmid != 0) {
4715 				(void) untimeout(tmid);
4716 				tmid = 0;
4717 			}
4718 
4719 			/* cancel pending I/Os */
4720 			cv_broadcast(&vdcp->running_cv);
4721 			vdc_cancel_backup_dring(vdcp);
4722 
4723 			/* wait for new I/O */
4724 			while (!vdcp->io_pending)
4725 				cv_wait(&vdcp->io_pending_cv, &vdcp->lock);
4726 
4727 			/*
4728 			 * There's a new IO pending. Try to re-establish a
4729 			 * connection. Mark all services as offline, so that
4730 			 * we don't stop again before having retried all
4731 			 * servers.
4732 			 */
4733 			for (srvr = vdcp->server_list; srvr != NULL;
4734 			    srvr = srvr->next) {
4735 				srvr->svc_state = VDC_SERVICE_OFFLINE;
4736 				srvr->hshake_cnt = 0;
4737 				srvr->hattr_cnt = 0;
4738 				srvr->hattr_total = 0;
4739 			}
4740 
4741 			/* reset variables */
4742 			hshake_cnt = 0;
4743 			hattr_cnt = 0;
4744 			vdcp->ctimeout_reached = B_FALSE;
4745 
4746 			vdcp->state = VDC_STATE_RESETTING;
4747 			vdcp->self_reset = B_TRUE;
4748 			break;
4749 
4750 		/* enter running state */
4751 		case VDC_STATE_RUNNING:
4752 
4753 			if (vdcp->lifecycle == VDC_LC_DETACHING) {
4754 				vdcp->state = VDC_STATE_DETACH;
4755 				break;
4756 			}
4757 
4758 			vdcp->lifecycle = VDC_LC_ONLINE;
4759 
4760 			if (failure_msg) {
4761 				cmn_err(CE_NOTE, "vdisk@%d disk access "
4762 				    "recovered", vdcp->instance);
4763 				failure_msg = B_FALSE;
4764 			}
4765 
4766 			/*
4767 			 * Signal anyone waiting for the connection
4768 			 * to come on line.
4769 			 */
4770 			cv_broadcast(&vdcp->running_cv);
4771 
4772 			/* backend has to be checked after reset */
4773 			if (vdcp->failfast_interval != 0 ||
4774 			    vdcp->num_servers > 1)
4775 				cv_signal(&vdcp->eio_cv);
4776 
4777 			/* ownership is lost during reset */
4778 			if (vdcp->ownership & VDC_OWNERSHIP_WANTED)
4779 				vdcp->ownership |= VDC_OWNERSHIP_RESET;
4780 			cv_signal(&vdcp->ownership_cv);
4781 
4782 			vdcp->curr_server->svc_state = VDC_SERVICE_ONLINE;
4783 			vdc_print_svc_status(vdcp);
4784 
4785 			mutex_exit(&vdcp->lock);
4786 
4787 			for (;;) {
4788 				vio_msg_t msg;
4789 				status = vdc_wait_for_response(vdcp, &msg);
4790 				if (status) break;
4791 
4792 				DMSG(vdcp, 1, "[%d] new pkt(s) available\n",
4793 				    vdcp->instance);
4794 				status = vdc_process_data_msg(vdcp, &msg);
4795 				if (status) {
4796 					DMSG(vdcp, 1, "[%d] process_data_msg "
4797 					    "returned err=%d\n", vdcp->instance,
4798 					    status);
4799 					break;
4800 				}
4801 
4802 			}
4803 
4804 			mutex_enter(&vdcp->lock);
4805 
4806 			/* all servers are now offline */
4807 			for (srvr = vdcp->server_list; srvr != NULL;
4808 			    srvr = srvr->next) {
4809 				srvr->svc_state = VDC_SERVICE_OFFLINE;
4810 				srvr->log_state = VDC_SERVICE_NONE;
4811 				srvr->hshake_cnt = 0;
4812 				srvr->hattr_cnt = 0;
4813 				srvr->hattr_total = 0;
4814 			}
4815 
4816 			hshake_cnt = 0;
4817 			hattr_cnt = 0;
4818 
4819 			vdc_print_svc_status(vdcp);
4820 
4821 			vdcp->state = VDC_STATE_RESETTING;
4822 			vdcp->self_reset = B_TRUE;
4823 			break;
4824 
4825 		case VDC_STATE_RESETTING:
4826 			/*
4827 			 * When we reach this state, we either come from the
4828 			 * VDC_STATE_RUNNING state and we can have pending
4829 			 * request but no timeout is armed; or we come from
4830 			 * the VDC_STATE_INIT_WAITING, VDC_NEGOTIATE or
4831 			 * VDC_HANDLE_PENDING state and there is no pending
4832 			 * request or pending requests have already been copied
4833 			 * into the backup dring. So we can safely keep the
4834 			 * connection timeout armed while we are in this state.
4835 			 */
4836 
4837 			DMSG(vdcp, 0, "Initiating channel reset "
4838 			    "(pending = %d)\n", (int)vdcp->threads_pending);
4839 
4840 			if (vdcp->self_reset) {
4841 				DMSG(vdcp, 0,
4842 				    "[%d] calling stop_ldc_connection.\n",
4843 				    vdcp->instance);
4844 				status = vdc_stop_ldc_connection(vdcp);
4845 				vdcp->self_reset = B_FALSE;
4846 			}
4847 
4848 			/*
4849 			 * Wait for all threads currently waiting
4850 			 * for a free dring entry to use.
4851 			 */
4852 			while (vdcp->threads_pending) {
4853 				cv_broadcast(&vdcp->membind_cv);
4854 				cv_broadcast(&vdcp->dring_free_cv);
4855 				mutex_exit(&vdcp->lock);
4856 				/* give the waiters enough time to wake up */
4857 				delay(vdc_hz_min_ldc_delay);
4858 				mutex_enter(&vdcp->lock);
4859 			}
4860 
4861 			ASSERT(vdcp->threads_pending == 0);
4862 
4863 			/* Sanity check that no thread is receiving */
4864 			ASSERT(vdcp->read_state != VDC_READ_WAITING);
4865 
4866 			vdcp->read_state = VDC_READ_IDLE;
4867 			vdcp->io_pending = B_FALSE;
4868 
4869 			/*
4870 			 * Cleanup any pending eio. These I/Os are going to
4871 			 * be resubmitted.
4872 			 */
4873 			vdc_eio_unqueue(vdcp, 0, B_FALSE);
4874 
4875 			vdc_backup_local_dring(vdcp);
4876 
4877 			/* cleanup the old d-ring */
4878 			vdc_destroy_descriptor_ring(vdcp);
4879 
4880 			/* go and start again */
4881 			vdcp->state = VDC_STATE_INIT;
4882 
4883 			break;
4884 
4885 		case VDC_STATE_DETACH:
4886 			DMSG(vdcp, 0, "[%d] Reset thread exit cleanup ..\n",
4887 			    vdcp->instance);
4888 
4889 			/* cancel any pending timeout */
4890 			mutex_exit(&vdcp->lock);
4891 			if (tmid != 0) {
4892 				(void) untimeout(tmid);
4893 				tmid = 0;
4894 			}
4895 			mutex_enter(&vdcp->lock);
4896 
4897 			/*
4898 			 * Signal anyone waiting for connection
4899 			 * to come online
4900 			 */
4901 			cv_broadcast(&vdcp->running_cv);
4902 
4903 			while (vdcp->sync_op_cnt > 0) {
4904 				cv_broadcast(&vdcp->sync_blocked_cv);
4905 				mutex_exit(&vdcp->lock);
4906 				/* give the waiters enough time to wake up */
4907 				delay(vdc_hz_min_ldc_delay);
4908 				mutex_enter(&vdcp->lock);
4909 			}
4910 
4911 			mutex_exit(&vdcp->lock);
4912 
4913 			DMSG(vdcp, 0, "[%d] Msg processing thread exiting ..\n",
4914 			    vdcp->instance);
4915 			thread_exit();
4916 			break;
4917 		}
4918 	}
4919 }
4920 
4921 
4922 /*
4923  * Function:
4924  *	vdc_process_data_msg()
4925  *
4926  * Description:
4927  *	This function is called by the message processing thread each time
4928  *	a message with a msgtype of VIO_TYPE_DATA is received. It will either
4929  *	be an ACK or NACK from vds[1] which vdc handles as follows.
4930  *		ACK	- wake up the waiting thread
4931  *		NACK	- resend any messages necessary
4932  *
4933  *	[1] Although the message format allows it, vds should not send a
4934  *	    VIO_SUBTYPE_INFO message to vdc asking it to read data; if for
4935  *	    some bizarre reason it does, vdc will reset the connection.
4936  *
4937  * Arguments:
4938  *	vdc	- soft state pointer for this instance of the device driver.
4939  *	msg	- the LDC message sent by vds
4940  *
4941  * Return Code:
4942  *	0	- Success.
4943  *	> 0	- error value returned by LDC
4944  */
4945 static int
4946 vdc_process_data_msg(vdc_t *vdcp, vio_msg_t *msg)
4947 {
4948 	int			status = 0;
4949 	vio_dring_msg_t		*dring_msg;
4950 	vdc_local_desc_t	*ldep = NULL;
4951 	int			start, end;
4952 	int			idx;
4953 	int			op;
4954 
4955 	dring_msg = (vio_dring_msg_t *)msg;
4956 
4957 	ASSERT(msg->tag.vio_msgtype == VIO_TYPE_DATA);
4958 	ASSERT(vdcp != NULL);
4959 
4960 	mutex_enter(&vdcp->lock);
4961 
4962 	/*
4963 	 * Check to see if the message has bogus data
4964 	 */
4965 	idx = start = dring_msg->start_idx;
4966 	end = dring_msg->end_idx;
4967 	if ((start >= vdcp->dring_len) ||
4968 	    (end >= vdcp->dring_len) || (end < -1)) {
4969 		/*
4970 		 * Update the I/O statistics to indicate that an error ocurred.
4971 		 * No need to update the wait/run queues as no specific read or
4972 		 * write request is being completed in response to this 'msg'.
4973 		 */
4974 		VD_UPDATE_ERR_STATS(vdcp, vd_softerrs);
4975 		DMSG(vdcp, 0, "[%d] Bogus ACK data : start %d, end %d\n",
4976 		    vdcp->instance, start, end);
4977 		mutex_exit(&vdcp->lock);
4978 		return (EINVAL);
4979 	}
4980 
4981 	/*
4982 	 * Verify that the sequence number is what vdc expects.
4983 	 */
4984 	switch (vdc_verify_seq_num(vdcp, dring_msg)) {
4985 	case VDC_SEQ_NUM_TODO:
4986 		break;	/* keep processing this message */
4987 	case VDC_SEQ_NUM_SKIP:
4988 		mutex_exit(&vdcp->lock);
4989 		return (0);
4990 	case VDC_SEQ_NUM_INVALID:
4991 		/*
4992 		 * Update the I/O statistics to indicate that an error ocurred.
4993 		 * No need to update the wait/run queues as no specific read or
4994 		 * write request is being completed in response to this 'msg'.
4995 		 */
4996 		VD_UPDATE_ERR_STATS(vdcp, vd_softerrs);
4997 		DMSG(vdcp, 0, "[%d] invalid seqno\n", vdcp->instance);
4998 		mutex_exit(&vdcp->lock);
4999 		return (ENXIO);
5000 	}
5001 
5002 	if (msg->tag.vio_subtype == VIO_SUBTYPE_NACK) {
5003 		/*
5004 		 * Update the I/O statistics to indicate that an error ocurred.
5005 		 * No need to update the wait/run queues, this will be done by
5006 		 * the thread calling this function.
5007 		 */
5008 		VD_UPDATE_ERR_STATS(vdcp, vd_softerrs);
5009 		VDC_DUMP_DRING_MSG(dring_msg);
5010 		DMSG(vdcp, 0, "[%d] DATA NACK\n", vdcp->instance);
5011 		mutex_exit(&vdcp->lock);
5012 		return (EIO);
5013 
5014 	} else if (msg->tag.vio_subtype == VIO_SUBTYPE_INFO) {
5015 		/*
5016 		 * Update the I/O statistics to indicate that an error occurred.
5017 		 * No need to update the wait/run queues as no specific read or
5018 		 * write request is being completed in response to this 'msg'.
5019 		 */
5020 		VD_UPDATE_ERR_STATS(vdcp, vd_protoerrs);
5021 		mutex_exit(&vdcp->lock);
5022 		return (EPROTO);
5023 	}
5024 
5025 	DMSG(vdcp, 1, ": start %d end %d\n", start, end);
5026 	ASSERT(start == end);
5027 
5028 	ldep = &vdcp->local_dring[idx];
5029 
5030 	DMSG(vdcp, 1, ": state 0x%x\n", ldep->dep->hdr.dstate);
5031 
5032 	if (ldep->dep->hdr.dstate == VIO_DESC_DONE) {
5033 		struct buf *bufp;
5034 
5035 		status = ldep->dep->payload.status;
5036 
5037 		bufp = ldep->buf;
5038 		ASSERT(bufp != NULL);
5039 
5040 		bufp->b_resid = bufp->b_bcount - ldep->dep->payload.nbytes;
5041 		bioerror(bufp, status);
5042 
5043 		if (status != 0) {
5044 			DMSG(vdcp, 1, "I/O status=%d\n", status);
5045 		}
5046 
5047 		DMSG(vdcp, 1,
5048 		    "I/O complete req=%ld bytes resp=%ld bytes\n",
5049 		    bufp->b_bcount, ldep->dep->payload.nbytes);
5050 
5051 		/*
5052 		 * If the request has failed and we have multiple servers or
5053 		 * failfast is enabled then we will have to defer the completion
5054 		 * of the request until we have checked that the vdisk backend
5055 		 * is effectively available (if multiple server) or that there
5056 		 * is no reservation conflict (if failfast).
5057 		 */
5058 		if (status != 0 &&
5059 		    ((vdcp->num_servers > 1 &&
5060 		    (ldep->flags & VDC_OP_ERRCHK_BACKEND)) ||
5061 		    (vdcp->failfast_interval != 0 &&
5062 		    (ldep->flags & VDC_OP_ERRCHK_CONFLICT)))) {
5063 			/*
5064 			 * The I/O has failed and we need to check the error.
5065 			 */
5066 			(void) vdc_eio_queue(vdcp, idx);
5067 		} else {
5068 			op = ldep->operation;
5069 			if (op == VD_OP_BREAD || op == VD_OP_BWRITE) {
5070 				if (status == 0) {
5071 					VD_UPDATE_IO_STATS(vdcp, op,
5072 					    ldep->dep->payload.nbytes);
5073 				} else {
5074 					VD_UPDATE_ERR_STATS(vdcp, vd_softerrs);
5075 				}
5076 				VD_KSTAT_RUNQ_EXIT(vdcp);
5077 				DTRACE_IO1(done, buf_t *, bufp);
5078 			}
5079 			(void) vdc_depopulate_descriptor(vdcp, idx);
5080 			biodone(bufp);
5081 		}
5082 	}
5083 
5084 	/* let the arrival signal propogate */
5085 	mutex_exit(&vdcp->lock);
5086 
5087 	/* probe gives the count of how many entries were processed */
5088 	DTRACE_PROBE2(processed, int, 1, vdc_t *, vdcp);
5089 
5090 	return (0);
5091 }
5092 
5093 
5094 /*
5095  * Function:
5096  *	vdc_handle_ver_msg()
5097  *
5098  * Description:
5099  *
5100  * Arguments:
5101  *	vdc	- soft state pointer for this instance of the device driver.
5102  *	ver_msg	- LDC message sent by vDisk server
5103  *
5104  * Return Code:
5105  *	0	- Success
5106  */
5107 static int
5108 vdc_handle_ver_msg(vdc_t *vdc, vio_ver_msg_t *ver_msg)
5109 {
5110 	int status = 0;
5111 
5112 	ASSERT(vdc != NULL);
5113 	ASSERT(mutex_owned(&vdc->lock));
5114 
5115 	if (ver_msg->tag.vio_subtype_env != VIO_VER_INFO) {
5116 		return (EPROTO);
5117 	}
5118 
5119 	if (ver_msg->dev_class != VDEV_DISK_SERVER) {
5120 		return (EINVAL);
5121 	}
5122 
5123 	switch (ver_msg->tag.vio_subtype) {
5124 	case VIO_SUBTYPE_ACK:
5125 		/*
5126 		 * We check to see if the version returned is indeed supported
5127 		 * (The server may have also adjusted the minor number downwards
5128 		 * and if so 'ver_msg' will contain the actual version agreed)
5129 		 */
5130 		if (vdc_is_supported_version(ver_msg)) {
5131 			vdc->ver.major = ver_msg->ver_major;
5132 			vdc->ver.minor = ver_msg->ver_minor;
5133 			ASSERT(vdc->ver.major > 0);
5134 		} else {
5135 			status = EPROTO;
5136 		}
5137 		break;
5138 
5139 	case VIO_SUBTYPE_NACK:
5140 		/*
5141 		 * call vdc_is_supported_version() which will return the next
5142 		 * supported version (if any) in 'ver_msg'
5143 		 */
5144 		(void) vdc_is_supported_version(ver_msg);
5145 		if (ver_msg->ver_major > 0) {
5146 			size_t len = sizeof (*ver_msg);
5147 
5148 			ASSERT(vdc->ver.major > 0);
5149 
5150 			/* reset the necessary fields and resend */
5151 			ver_msg->tag.vio_subtype = VIO_SUBTYPE_INFO;
5152 			ver_msg->dev_class = VDEV_DISK;
5153 
5154 			status = vdc_send(vdc, (caddr_t)ver_msg, &len);
5155 			DMSG(vdc, 0, "[%d] Resend VER info (LDC status = %d)\n",
5156 			    vdc->instance, status);
5157 			if (len != sizeof (*ver_msg))
5158 				status = EBADMSG;
5159 		} else {
5160 			DMSG(vdc, 0, "[%d] No common version with vDisk server",
5161 			    vdc->instance);
5162 			status = ENOTSUP;
5163 		}
5164 
5165 		break;
5166 	case VIO_SUBTYPE_INFO:
5167 		/*
5168 		 * Handle the case where vds starts handshake
5169 		 * (for now only vdc is the instigator)
5170 		 */
5171 		status = ENOTSUP;
5172 		break;
5173 
5174 	default:
5175 		status = EINVAL;
5176 		break;
5177 	}
5178 
5179 	return (status);
5180 }
5181 
5182 /*
5183  * Function:
5184  *	vdc_handle_attr_msg()
5185  *
5186  * Description:
5187  *
5188  * Arguments:
5189  *	vdc	- soft state pointer for this instance of the device driver.
5190  *	attr_msg	- LDC message sent by vDisk server
5191  *
5192  * Return Code:
5193  *	0	- Success
5194  */
5195 static int
5196 vdc_handle_attr_msg(vdc_t *vdc, vd_attr_msg_t *attr_msg)
5197 {
5198 	int status = 0;
5199 	vd_disk_type_t old_type;
5200 
5201 	ASSERT(vdc != NULL);
5202 	ASSERT(mutex_owned(&vdc->lock));
5203 
5204 	if (attr_msg->tag.vio_subtype_env != VIO_ATTR_INFO) {
5205 		return (EPROTO);
5206 	}
5207 
5208 	switch (attr_msg->tag.vio_subtype) {
5209 	case VIO_SUBTYPE_ACK:
5210 		/*
5211 		 * We now verify the attributes sent by vds.
5212 		 */
5213 		if (attr_msg->vdisk_size == 0) {
5214 			DMSG(vdc, 0, "[%d] Invalid disk size from vds",
5215 			    vdc->instance);
5216 			status = EINVAL;
5217 			break;
5218 		}
5219 
5220 		if (attr_msg->max_xfer_sz == 0) {
5221 			DMSG(vdc, 0, "[%d] Invalid transfer size from vds",
5222 			    vdc->instance);
5223 			status = EINVAL;
5224 			break;
5225 		}
5226 
5227 		if (attr_msg->vdisk_size == VD_SIZE_UNKNOWN) {
5228 			DMSG(vdc, 0, "[%d] Unknown disk size from vds",
5229 			    vdc->instance);
5230 			attr_msg->vdisk_size = 0;
5231 		}
5232 
5233 		/* update the VIO block size */
5234 		if (attr_msg->vdisk_block_size > 0 &&
5235 		    vdc_update_vio_bsize(vdc,
5236 		    attr_msg->vdisk_block_size) != 0) {
5237 			DMSG(vdc, 0, "[%d] Invalid block size (%u) from vds",
5238 			    vdc->instance, attr_msg->vdisk_block_size);
5239 			status = EINVAL;
5240 			break;
5241 		}
5242 
5243 		/* update disk, block and transfer sizes */
5244 		old_type = vdc->vdisk_type;
5245 		vdc_update_size(vdc, attr_msg->vdisk_size,
5246 		    attr_msg->vdisk_block_size, attr_msg->max_xfer_sz);
5247 		vdc->vdisk_type = attr_msg->vdisk_type;
5248 		vdc->operations = attr_msg->operations;
5249 		if (vio_ver_is_supported(vdc->ver, 1, 1))
5250 			vdc->vdisk_media = attr_msg->vdisk_media;
5251 		else
5252 			vdc->vdisk_media = 0;
5253 
5254 		DMSG(vdc, 0, "[%d] max_xfer_sz: sent %lx acked %lx\n",
5255 		    vdc->instance, vdc->max_xfer_sz, attr_msg->max_xfer_sz);
5256 		DMSG(vdc, 0, "[%d] vdisk_block_size: sent %lx acked %x\n",
5257 		    vdc->instance, vdc->vdisk_bsize,
5258 		    attr_msg->vdisk_block_size);
5259 
5260 		if ((attr_msg->xfer_mode != VIO_DRING_MODE_V1_0) ||
5261 		    (attr_msg->vdisk_size > INT64_MAX) ||
5262 		    (attr_msg->operations == 0) ||
5263 		    (attr_msg->vdisk_type > VD_DISK_TYPE_DISK)) {
5264 			DMSG(vdc, 0, "[%d] Invalid attributes from vds",
5265 			    vdc->instance);
5266 			status = EINVAL;
5267 			break;
5268 		}
5269 
5270 		/*
5271 		 * Now that we have received all attributes we can create a
5272 		 * fake geometry for the disk.
5273 		 */
5274 		vdc_create_fake_geometry(vdc);
5275 
5276 		/*
5277 		 * If the disk type was previously unknown and device nodes
5278 		 * were created then the driver would have created 8 device
5279 		 * nodes. If we now find out that this is a single-slice disk
5280 		 * then we need to re-create the appropriate device nodes.
5281 		 */
5282 		if (old_type == VD_DISK_TYPE_UNK &&
5283 		    (vdc->initialized & VDC_MINOR) &&
5284 		    vdc->vdisk_type == VD_DISK_TYPE_SLICE) {
5285 			ddi_remove_minor_node(vdc->dip, NULL);
5286 			(void) devfs_clean(ddi_get_parent(vdc->dip),
5287 			    NULL, DV_CLEAN_FORCE);
5288 			if (vdc_create_device_nodes(vdc) != 0) {
5289 				DMSG(vdc, 0, "![%d] Failed to update "
5290 				    "device nodes", vdc->instance);
5291 			}
5292 		}
5293 
5294 		break;
5295 
5296 	case VIO_SUBTYPE_NACK:
5297 		/*
5298 		 * vds could not handle the attributes we sent so we
5299 		 * stop negotiating.
5300 		 */
5301 		status = EPROTO;
5302 		break;
5303 
5304 	case VIO_SUBTYPE_INFO:
5305 		/*
5306 		 * Handle the case where vds starts the handshake
5307 		 * (for now; vdc is the only supported instigatior)
5308 		 */
5309 		status = ENOTSUP;
5310 		break;
5311 
5312 	default:
5313 		status = ENOTSUP;
5314 		break;
5315 	}
5316 
5317 	return (status);
5318 }
5319 
5320 /*
5321  * Function:
5322  *	vdc_handle_dring_reg_msg()
5323  *
5324  * Description:
5325  *
5326  * Arguments:
5327  *	vdc		- soft state pointer for this instance of the driver.
5328  *	dring_msg	- LDC message sent by vDisk server
5329  *
5330  * Return Code:
5331  *	0	- Success
5332  */
5333 static int
5334 vdc_handle_dring_reg_msg(vdc_t *vdc, vio_dring_reg_msg_t *dring_msg)
5335 {
5336 	int		status = 0;
5337 
5338 	ASSERT(vdc != NULL);
5339 	ASSERT(mutex_owned(&vdc->lock));
5340 
5341 	if (dring_msg->tag.vio_subtype_env != VIO_DRING_REG) {
5342 		return (EPROTO);
5343 	}
5344 
5345 	switch (dring_msg->tag.vio_subtype) {
5346 	case VIO_SUBTYPE_ACK:
5347 		/* save the received dring_ident */
5348 		vdc->dring_ident = dring_msg->dring_ident;
5349 		DMSG(vdc, 0, "[%d] Received dring ident=0x%lx\n",
5350 		    vdc->instance, vdc->dring_ident);
5351 		break;
5352 
5353 	case VIO_SUBTYPE_NACK:
5354 		/*
5355 		 * vds could not handle the DRing info we sent so we
5356 		 * stop negotiating.
5357 		 */
5358 		DMSG(vdc, 0, "[%d] server could not register DRing\n",
5359 		    vdc->instance);
5360 		status = EPROTO;
5361 		break;
5362 
5363 	case VIO_SUBTYPE_INFO:
5364 		/*
5365 		 * Handle the case where vds starts handshake
5366 		 * (for now only vdc is the instigatior)
5367 		 */
5368 		status = ENOTSUP;
5369 		break;
5370 	default:
5371 		status = ENOTSUP;
5372 	}
5373 
5374 	return (status);
5375 }
5376 
5377 /*
5378  * Function:
5379  *	vdc_verify_seq_num()
5380  *
5381  * Description:
5382  *	This functions verifies that the sequence number sent back by the vDisk
5383  *	server with the latest message is what is expected (i.e. it is greater
5384  *	than the last seq num sent by the vDisk server and less than or equal
5385  *	to the last seq num generated by vdc).
5386  *
5387  *	It then checks the request ID to see if any requests need processing
5388  *	in the DRing.
5389  *
5390  * Arguments:
5391  *	vdc		- soft state pointer for this instance of the driver.
5392  *	dring_msg	- pointer to the LDC message sent by vds
5393  *
5394  * Return Code:
5395  *	VDC_SEQ_NUM_TODO	- Message needs to be processed
5396  *	VDC_SEQ_NUM_SKIP	- Message has already been processed
5397  *	VDC_SEQ_NUM_INVALID	- The seq numbers are so out of sync,
5398  *				  vdc cannot deal with them
5399  */
5400 static int
5401 vdc_verify_seq_num(vdc_t *vdc, vio_dring_msg_t *dring_msg)
5402 {
5403 	ASSERT(vdc != NULL);
5404 	ASSERT(dring_msg != NULL);
5405 	ASSERT(mutex_owned(&vdc->lock));
5406 
5407 	/*
5408 	 * Check to see if the messages were responded to in the correct
5409 	 * order by vds.
5410 	 */
5411 	if ((dring_msg->seq_num <= vdc->seq_num_reply) ||
5412 	    (dring_msg->seq_num > vdc->seq_num)) {
5413 		DMSG(vdc, 0, "?[%d] Bogus sequence_number %lu: "
5414 		    "%lu > expected <= %lu (last proc req %lu sent %lu)\n",
5415 		    vdc->instance, dring_msg->seq_num,
5416 		    vdc->seq_num_reply, vdc->seq_num,
5417 		    vdc->req_id_proc, vdc->req_id);
5418 		return (VDC_SEQ_NUM_INVALID);
5419 	}
5420 	vdc->seq_num_reply = dring_msg->seq_num;
5421 
5422 	if (vdc->req_id_proc < vdc->req_id)
5423 		return (VDC_SEQ_NUM_TODO);
5424 	else
5425 		return (VDC_SEQ_NUM_SKIP);
5426 }
5427 
5428 
5429 /*
5430  * Function:
5431  *	vdc_is_supported_version()
5432  *
5433  * Description:
5434  *	This routine checks if the major/minor version numbers specified in
5435  *	'ver_msg' are supported. If not it finds the next version that is
5436  *	in the supported version list 'vdc_version[]' and sets the fields in
5437  *	'ver_msg' to those values
5438  *
5439  * Arguments:
5440  *	ver_msg	- LDC message sent by vDisk server
5441  *
5442  * Return Code:
5443  *	B_TRUE	- Success
5444  *	B_FALSE	- Version not supported
5445  */
5446 static boolean_t
5447 vdc_is_supported_version(vio_ver_msg_t *ver_msg)
5448 {
5449 	int vdc_num_versions = sizeof (vdc_version) / sizeof (vdc_version[0]);
5450 
5451 	for (int i = 0; i < vdc_num_versions; i++) {
5452 		ASSERT(vdc_version[i].major > 0);
5453 		ASSERT((i == 0) ||
5454 		    (vdc_version[i].major < vdc_version[i-1].major));
5455 
5456 		/*
5457 		 * If the major versions match, adjust the minor version, if
5458 		 * necessary, down to the highest value supported by this
5459 		 * client. The server should support all minor versions lower
5460 		 * than the value it sent
5461 		 */
5462 		if (ver_msg->ver_major == vdc_version[i].major) {
5463 			if (ver_msg->ver_minor > vdc_version[i].minor) {
5464 				DMSGX(0,
5465 				    "Adjusting minor version from %u to %u",
5466 				    ver_msg->ver_minor, vdc_version[i].minor);
5467 				ver_msg->ver_minor = vdc_version[i].minor;
5468 			}
5469 			return (B_TRUE);
5470 		}
5471 
5472 		/*
5473 		 * If the message contains a higher major version number, set
5474 		 * the message's major/minor versions to the current values
5475 		 * and return false, so this message will get resent with
5476 		 * these values, and the server will potentially try again
5477 		 * with the same or a lower version
5478 		 */
5479 		if (ver_msg->ver_major > vdc_version[i].major) {
5480 			ver_msg->ver_major = vdc_version[i].major;
5481 			ver_msg->ver_minor = vdc_version[i].minor;
5482 			DMSGX(0, "Suggesting major/minor (0x%x/0x%x)\n",
5483 			    ver_msg->ver_major, ver_msg->ver_minor);
5484 
5485 			return (B_FALSE);
5486 		}
5487 
5488 		/*
5489 		 * Otherwise, the message's major version is less than the
5490 		 * current major version, so continue the loop to the next
5491 		 * (lower) supported version
5492 		 */
5493 	}
5494 
5495 	/*
5496 	 * No common version was found; "ground" the version pair in the
5497 	 * message to terminate negotiation
5498 	 */
5499 	ver_msg->ver_major = 0;
5500 	ver_msg->ver_minor = 0;
5501 
5502 	return (B_FALSE);
5503 }
5504 /* -------------------------------------------------------------------------- */
5505 
5506 /*
5507  * dkio(4I) support
5508  */
5509 
5510 typedef struct vdc_dk_arg {
5511 	struct dk_callback	dkc;
5512 	int			mode;
5513 	dev_t			dev;
5514 	vdc_t			*vdc;
5515 } vdc_dk_arg_t;
5516 
5517 /*
5518  * Function:
5519  *	vdc_dkio_flush_cb()
5520  *
5521  * Description:
5522  *	This routine is a callback for DKIOCFLUSHWRITECACHE which can be called
5523  *	by kernel code.
5524  *
5525  * Arguments:
5526  *	arg	- a pointer to a vdc_dk_arg_t structure.
5527  */
5528 void
5529 vdc_dkio_flush_cb(void *arg)
5530 {
5531 	struct vdc_dk_arg	*dk_arg = (struct vdc_dk_arg *)arg;
5532 	struct dk_callback	*dkc = NULL;
5533 	vdc_t			*vdc = NULL;
5534 	int			rv;
5535 
5536 	if (dk_arg == NULL) {
5537 		cmn_err(CE_NOTE, "?[Unk] DKIOCFLUSHWRITECACHE arg is NULL\n");
5538 		return;
5539 	}
5540 	dkc = &dk_arg->dkc;
5541 	vdc = dk_arg->vdc;
5542 	ASSERT(vdc != NULL);
5543 
5544 	rv = vdc_do_sync_op(vdc, VD_OP_FLUSH, NULL, 0,
5545 	    VDCPART(dk_arg->dev), 0, VIO_both_dir, B_TRUE);
5546 	if (rv != 0) {
5547 		DMSG(vdc, 0, "[%d] DKIOCFLUSHWRITECACHE failed %d : model %x\n",
5548 		    vdc->instance, rv,
5549 		    ddi_model_convert_from(dk_arg->mode & FMODELS));
5550 	}
5551 
5552 	/*
5553 	 * Trigger the call back to notify the caller the the ioctl call has
5554 	 * been completed.
5555 	 */
5556 	if ((dk_arg->mode & FKIOCTL) &&
5557 	    (dkc != NULL) &&
5558 	    (dkc->dkc_callback != NULL)) {
5559 		ASSERT(dkc->dkc_cookie != NULL);
5560 		(*dkc->dkc_callback)(dkc->dkc_cookie, rv);
5561 	}
5562 
5563 	/* Indicate that one less DKIO write flush is outstanding */
5564 	mutex_enter(&vdc->lock);
5565 	vdc->dkio_flush_pending--;
5566 	ASSERT(vdc->dkio_flush_pending >= 0);
5567 	mutex_exit(&vdc->lock);
5568 
5569 	/* free the mem that was allocated when the callback was dispatched */
5570 	kmem_free(arg, sizeof (vdc_dk_arg_t));
5571 }
5572 
5573 /*
5574  * Function:
5575  *	vdc_dkio_gapart()
5576  *
5577  * Description:
5578  *	This function implements the DKIOCGAPART ioctl.
5579  *
5580  * Arguments:
5581  *	vdc	- soft state pointer
5582  *	arg	- a pointer to a dk_map[NDKMAP] or dk_map32[NDKMAP] structure
5583  *	flag	- ioctl flags
5584  */
5585 static int
5586 vdc_dkio_gapart(vdc_t *vdc, caddr_t arg, int flag)
5587 {
5588 	struct dk_geom *geom;
5589 	struct extvtoc *vtoc;
5590 	union {
5591 		struct dk_map map[NDKMAP];
5592 		struct dk_map32 map32[NDKMAP];
5593 	} data;
5594 	int i, rv, size;
5595 
5596 	mutex_enter(&vdc->lock);
5597 
5598 	if ((rv = vdc_validate_geometry(vdc)) != 0) {
5599 		mutex_exit(&vdc->lock);
5600 		return (rv);
5601 	}
5602 
5603 	if (vdc->vdisk_size > VD_OLDVTOC_LIMIT) {
5604 		mutex_exit(&vdc->lock);
5605 		return (EOVERFLOW);
5606 	}
5607 
5608 	vtoc = vdc->vtoc;
5609 	geom = vdc->geom;
5610 
5611 	if (ddi_model_convert_from(flag & FMODELS) == DDI_MODEL_ILP32) {
5612 
5613 		for (i = 0; i < vtoc->v_nparts; i++) {
5614 			data.map32[i].dkl_cylno = vtoc->v_part[i].p_start /
5615 			    (geom->dkg_nhead * geom->dkg_nsect);
5616 			data.map32[i].dkl_nblk = vtoc->v_part[i].p_size;
5617 		}
5618 		size = NDKMAP * sizeof (struct dk_map32);
5619 
5620 	} else {
5621 
5622 		for (i = 0; i < vtoc->v_nparts; i++) {
5623 			data.map[i].dkl_cylno = vtoc->v_part[i].p_start /
5624 			    (geom->dkg_nhead * geom->dkg_nsect);
5625 			data.map[i].dkl_nblk = vtoc->v_part[i].p_size;
5626 		}
5627 		size = NDKMAP * sizeof (struct dk_map);
5628 
5629 	}
5630 
5631 	mutex_exit(&vdc->lock);
5632 
5633 	if (ddi_copyout(&data, arg, size, flag) != 0)
5634 		return (EFAULT);
5635 
5636 	return (0);
5637 }
5638 
5639 /*
5640  * Function:
5641  *	vdc_dkio_partition()
5642  *
5643  * Description:
5644  *	This function implements the DKIOCPARTITION ioctl.
5645  *
5646  * Arguments:
5647  *	vdc	- soft state pointer
5648  *	arg	- a pointer to a struct partition64 structure
5649  *	flag	- ioctl flags
5650  */
5651 static int
5652 vdc_dkio_partition(vdc_t *vdc, caddr_t arg, int flag)
5653 {
5654 	struct partition64 p64;
5655 	efi_gpt_t *gpt;
5656 	efi_gpe_t *gpe;
5657 	vd_efi_dev_t edev;
5658 	uint_t partno;
5659 	int rv;
5660 
5661 	if (ddi_copyin(arg, &p64, sizeof (struct partition64), flag)) {
5662 		return (EFAULT);
5663 	}
5664 
5665 	VDC_EFI_DEV_SET(edev, vdc, vd_process_efi_ioctl);
5666 
5667 	if ((rv = vd_efi_alloc_and_read(&edev, &gpt, &gpe)) != 0) {
5668 		return (rv);
5669 	}
5670 
5671 	partno = p64.p_partno;
5672 
5673 	if (partno >= gpt->efi_gpt_NumberOfPartitionEntries) {
5674 		vd_efi_free(&edev, gpt, gpe);
5675 		return (ESRCH);
5676 	}
5677 
5678 	bcopy(&gpe[partno].efi_gpe_PartitionTypeGUID, &p64.p_type,
5679 	    sizeof (struct uuid));
5680 	p64.p_start = gpe[partno].efi_gpe_StartingLBA;
5681 	p64.p_size = gpe[partno].efi_gpe_EndingLBA - p64.p_start + 1;
5682 
5683 	if (ddi_copyout(&p64, arg, sizeof (struct partition64), flag)) {
5684 		vd_efi_free(&edev, gpt, gpe);
5685 		return (EFAULT);
5686 	}
5687 
5688 	vd_efi_free(&edev, gpt, gpe);
5689 	return (0);
5690 }
5691 
5692 /*
5693  * Function:
5694  *	vdc_dioctl_rwcmd()
5695  *
5696  * Description:
5697  *	This function implements the DIOCTL_RWCMD ioctl. This ioctl is used
5698  *	for DKC_DIRECT disks to read or write at an absolute disk offset.
5699  *
5700  * Arguments:
5701  *	dev	- device
5702  *	arg	- a pointer to a dadkio_rwcmd or dadkio_rwcmd32 structure
5703  *	flag	- ioctl flags
5704  */
5705 static int
5706 vdc_dioctl_rwcmd(vdc_t *vdc, caddr_t arg, int flag)
5707 {
5708 	struct dadkio_rwcmd32 rwcmd32;
5709 	struct dadkio_rwcmd rwcmd;
5710 	struct iovec aiov;
5711 	struct uio auio;
5712 	int rw, status;
5713 	struct buf *buf;
5714 
5715 	if (ddi_model_convert_from(flag & FMODELS) == DDI_MODEL_ILP32) {
5716 		if (ddi_copyin((caddr_t)arg, (caddr_t)&rwcmd32,
5717 		    sizeof (struct dadkio_rwcmd32), flag)) {
5718 			return (EFAULT);
5719 		}
5720 		rwcmd.cmd = rwcmd32.cmd;
5721 		rwcmd.flags = rwcmd32.flags;
5722 		rwcmd.blkaddr = (daddr_t)rwcmd32.blkaddr;
5723 		rwcmd.buflen = rwcmd32.buflen;
5724 		rwcmd.bufaddr = (caddr_t)(uintptr_t)rwcmd32.bufaddr;
5725 	} else {
5726 		if (ddi_copyin((caddr_t)arg, (caddr_t)&rwcmd,
5727 		    sizeof (struct dadkio_rwcmd), flag)) {
5728 			return (EFAULT);
5729 		}
5730 	}
5731 
5732 	switch (rwcmd.cmd) {
5733 	case DADKIO_RWCMD_READ:
5734 		rw = B_READ;
5735 		break;
5736 	case DADKIO_RWCMD_WRITE:
5737 		rw = B_WRITE;
5738 		break;
5739 	default:
5740 		return (EINVAL);
5741 	}
5742 
5743 	bzero((caddr_t)&aiov, sizeof (struct iovec));
5744 	aiov.iov_base   = rwcmd.bufaddr;
5745 	aiov.iov_len    = rwcmd.buflen;
5746 
5747 	bzero((caddr_t)&auio, sizeof (struct uio));
5748 	auio.uio_iov    = &aiov;
5749 	auio.uio_iovcnt = 1;
5750 	auio.uio_loffset = rwcmd.blkaddr * vdc->vdisk_bsize;
5751 	auio.uio_resid  = rwcmd.buflen;
5752 	auio.uio_segflg = flag & FKIOCTL ? UIO_SYSSPACE : UIO_USERSPACE;
5753 
5754 	buf = kmem_alloc(sizeof (buf_t), KM_SLEEP);
5755 	bioinit(buf);
5756 	/*
5757 	 * We use the private field of buf to specify that this is an
5758 	 * I/O using an absolute offset.
5759 	 */
5760 	buf->b_private = (void *)VD_SLICE_NONE;
5761 
5762 	status = physio(vdc_strategy, buf, VD_MAKE_DEV(vdc->instance, 0),
5763 	    rw, vdc_min, &auio);
5764 
5765 	biofini(buf);
5766 	kmem_free(buf, sizeof (buf_t));
5767 
5768 	return (status);
5769 }
5770 
5771 /*
5772  * Allocate a buffer for a VD_OP_SCSICMD operation. The size of the allocated
5773  * buffer is returned in alloc_len.
5774  */
5775 static vd_scsi_t *
5776 vdc_scsi_alloc(int cdb_len, int sense_len, int datain_len, int dataout_len,
5777     int *alloc_len)
5778 {
5779 	vd_scsi_t *vd_scsi;
5780 	int vd_scsi_len = VD_SCSI_SIZE;
5781 
5782 	vd_scsi_len += P2ROUNDUP(cdb_len, sizeof (uint64_t));
5783 	vd_scsi_len += P2ROUNDUP(sense_len, sizeof (uint64_t));
5784 	vd_scsi_len += P2ROUNDUP(datain_len, sizeof (uint64_t));
5785 	vd_scsi_len += P2ROUNDUP(dataout_len, sizeof (uint64_t));
5786 
5787 	ASSERT(vd_scsi_len % sizeof (uint64_t) == 0);
5788 
5789 	vd_scsi = kmem_zalloc(vd_scsi_len, KM_SLEEP);
5790 
5791 	vd_scsi->cdb_len = cdb_len;
5792 	vd_scsi->sense_len = sense_len;
5793 	vd_scsi->datain_len = datain_len;
5794 	vd_scsi->dataout_len = dataout_len;
5795 
5796 	*alloc_len = vd_scsi_len;
5797 
5798 	return (vd_scsi);
5799 }
5800 
5801 /*
5802  * Convert the status of a SCSI command to a Solaris return code.
5803  *
5804  * Arguments:
5805  *	vd_scsi		- The SCSI operation buffer.
5806  *	log_error	- indicate if an error message should be logged.
5807  *
5808  * Note that our SCSI error messages are rather primitive for the moment
5809  * and could be improved by decoding some data like the SCSI command and
5810  * the sense key.
5811  *
5812  * Return value:
5813  *	0		- Status is good.
5814  *	EACCES		- Status reports a reservation conflict.
5815  *	ENOTSUP		- Status reports a check condition and sense key
5816  *			  reports an illegal request.
5817  *	EIO		- Any other status.
5818  */
5819 static int
5820 vdc_scsi_status(vdc_t *vdc, vd_scsi_t *vd_scsi, boolean_t log_error)
5821 {
5822 	int rv;
5823 	char path_str[MAXPATHLEN];
5824 	char panic_str[VDC_RESV_CONFLICT_FMT_LEN + MAXPATHLEN];
5825 	union scsi_cdb *cdb;
5826 	struct scsi_extended_sense *sense;
5827 
5828 	if (vd_scsi->cmd_status == STATUS_GOOD)
5829 		/* no error */
5830 		return (0);
5831 
5832 	/* when the tunable vdc_scsi_log_error is true we log all errors */
5833 	if (vdc_scsi_log_error)
5834 		log_error = B_TRUE;
5835 
5836 	if (log_error) {
5837 		cmn_err(CE_WARN, "%s (vdc%d):\tError for Command: 0x%x)\n",
5838 		    ddi_pathname(vdc->dip, path_str), vdc->instance,
5839 		    GETCMD(VD_SCSI_DATA_CDB(vd_scsi)));
5840 	}
5841 
5842 	/* default returned value */
5843 	rv = EIO;
5844 
5845 	switch (vd_scsi->cmd_status) {
5846 
5847 	case STATUS_CHECK:
5848 	case STATUS_TERMINATED:
5849 		if (log_error)
5850 			cmn_err(CE_CONT, "\tCheck Condition Error\n");
5851 
5852 		/* check sense buffer */
5853 		if (vd_scsi->sense_len == 0 ||
5854 		    vd_scsi->sense_status != STATUS_GOOD) {
5855 			if (log_error)
5856 				cmn_err(CE_CONT, "\tNo Sense Data Available\n");
5857 			break;
5858 		}
5859 
5860 		sense = VD_SCSI_DATA_SENSE(vd_scsi);
5861 
5862 		if (log_error) {
5863 			cmn_err(CE_CONT, "\tSense Key:  0x%x\n"
5864 			    "\tASC: 0x%x, ASCQ: 0x%x\n",
5865 			    scsi_sense_key((uint8_t *)sense),
5866 			    scsi_sense_asc((uint8_t *)sense),
5867 			    scsi_sense_ascq((uint8_t *)sense));
5868 		}
5869 
5870 		if (scsi_sense_key((uint8_t *)sense) == KEY_ILLEGAL_REQUEST)
5871 			rv = ENOTSUP;
5872 		break;
5873 
5874 	case STATUS_BUSY:
5875 		if (log_error)
5876 			cmn_err(CE_NOTE, "\tDevice Busy\n");
5877 		break;
5878 
5879 	case STATUS_RESERVATION_CONFLICT:
5880 		/*
5881 		 * If the command was PERSISTENT_RESERVATION_[IN|OUT] then
5882 		 * reservation conflict could be due to various reasons like
5883 		 * incorrect keys, not registered or not reserved etc. So,
5884 		 * we should not panic in that case.
5885 		 */
5886 		cdb = VD_SCSI_DATA_CDB(vd_scsi);
5887 		if (vdc->failfast_interval != 0 &&
5888 		    cdb->scc_cmd != SCMD_PERSISTENT_RESERVE_IN &&
5889 		    cdb->scc_cmd != SCMD_PERSISTENT_RESERVE_OUT) {
5890 			/* failfast is enabled so we have to panic */
5891 			(void) snprintf(panic_str, sizeof (panic_str),
5892 			    VDC_RESV_CONFLICT_FMT_STR "%s",
5893 			    ddi_pathname(vdc->dip, path_str));
5894 			panic(panic_str);
5895 		}
5896 		if (log_error)
5897 			cmn_err(CE_NOTE, "\tReservation Conflict\n");
5898 		rv = EACCES;
5899 		break;
5900 
5901 	case STATUS_QFULL:
5902 		if (log_error)
5903 			cmn_err(CE_NOTE, "\tQueue Full\n");
5904 		break;
5905 
5906 	case STATUS_MET:
5907 	case STATUS_INTERMEDIATE:
5908 	case STATUS_SCSI2:
5909 	case STATUS_INTERMEDIATE_MET:
5910 	case STATUS_ACA_ACTIVE:
5911 		if (log_error)
5912 			cmn_err(CE_CONT,
5913 			    "\tUnexpected SCSI status received: 0x%x\n",
5914 			    vd_scsi->cmd_status);
5915 		break;
5916 
5917 	default:
5918 		if (log_error)
5919 			cmn_err(CE_CONT,
5920 			    "\tInvalid SCSI status received: 0x%x\n",
5921 			    vd_scsi->cmd_status);
5922 		break;
5923 	}
5924 
5925 	return (rv);
5926 }
5927 
5928 /*
5929  * Implemented the USCSICMD uscsi(4I) ioctl. This ioctl is converted to
5930  * a VD_OP_SCSICMD operation which is sent to the vdisk server. If a SCSI
5931  * reset is requested (i.e. a flag USCSI_RESET* is set) then the ioctl is
5932  * converted to a VD_OP_RESET operation.
5933  */
5934 static int
5935 vdc_uscsi_cmd(vdc_t *vdc, caddr_t arg, int mode)
5936 {
5937 	struct uscsi_cmd	uscsi;
5938 	struct uscsi_cmd32	uscsi32;
5939 	vd_scsi_t		*vd_scsi;
5940 	int			vd_scsi_len;
5941 	union scsi_cdb		*cdb;
5942 	struct scsi_extended_sense *sense;
5943 	char			*datain, *dataout;
5944 	size_t			cdb_len, datain_len, dataout_len, sense_len;
5945 	int			rv;
5946 
5947 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
5948 		if (ddi_copyin(arg, &uscsi32, sizeof (struct uscsi_cmd32),
5949 		    mode) != 0)
5950 			return (EFAULT);
5951 		uscsi_cmd32touscsi_cmd((&uscsi32), (&uscsi));
5952 	} else {
5953 		if (ddi_copyin(arg, &uscsi, sizeof (struct uscsi_cmd),
5954 		    mode) != 0)
5955 			return (EFAULT);
5956 	}
5957 
5958 	/* a uscsi reset is converted to a VD_OP_RESET operation */
5959 	if (uscsi.uscsi_flags & (USCSI_RESET | USCSI_RESET_LUN |
5960 	    USCSI_RESET_ALL)) {
5961 		rv = vdc_do_sync_op(vdc, VD_OP_RESET, NULL, 0, 0, 0,
5962 		    VIO_both_dir, B_TRUE);
5963 		return (rv);
5964 	}
5965 
5966 	/* cdb buffer length */
5967 	cdb_len = uscsi.uscsi_cdblen;
5968 
5969 	/* data in and out buffers length */
5970 	if (uscsi.uscsi_flags & USCSI_READ) {
5971 		datain_len = uscsi.uscsi_buflen;
5972 		dataout_len = 0;
5973 	} else {
5974 		datain_len = 0;
5975 		dataout_len = uscsi.uscsi_buflen;
5976 	}
5977 
5978 	/* sense buffer length */
5979 	if (uscsi.uscsi_flags & USCSI_RQENABLE)
5980 		sense_len = uscsi.uscsi_rqlen;
5981 	else
5982 		sense_len = 0;
5983 
5984 	/* allocate buffer for the VD_SCSICMD_OP operation */
5985 	vd_scsi = vdc_scsi_alloc(cdb_len, sense_len, datain_len, dataout_len,
5986 	    &vd_scsi_len);
5987 
5988 	/*
5989 	 * The documentation of USCSI_ISOLATE and USCSI_DIAGNOSE is very vague,
5990 	 * but basically they prevent a SCSI command from being retried in case
5991 	 * of an error.
5992 	 */
5993 	if ((uscsi.uscsi_flags & USCSI_ISOLATE) ||
5994 	    (uscsi.uscsi_flags & USCSI_DIAGNOSE))
5995 		vd_scsi->options |= VD_SCSI_OPT_NORETRY;
5996 
5997 	/* set task attribute */
5998 	if (uscsi.uscsi_flags & USCSI_NOTAG) {
5999 		vd_scsi->task_attribute = 0;
6000 	} else {
6001 		if (uscsi.uscsi_flags & USCSI_HEAD)
6002 			vd_scsi->task_attribute = VD_SCSI_TASK_ACA;
6003 		else if (uscsi.uscsi_flags & USCSI_HTAG)
6004 			vd_scsi->task_attribute = VD_SCSI_TASK_HQUEUE;
6005 		else if (uscsi.uscsi_flags & USCSI_OTAG)
6006 			vd_scsi->task_attribute = VD_SCSI_TASK_ORDERED;
6007 		else
6008 			vd_scsi->task_attribute = 0;
6009 	}
6010 
6011 	/* set timeout */
6012 	vd_scsi->timeout = uscsi.uscsi_timeout;
6013 
6014 	/* copy-in cdb data */
6015 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
6016 	if (ddi_copyin(uscsi.uscsi_cdb, cdb, cdb_len, mode) != 0) {
6017 		rv = EFAULT;
6018 		goto done;
6019 	}
6020 
6021 	/* keep a pointer to the sense buffer */
6022 	sense = VD_SCSI_DATA_SENSE(vd_scsi);
6023 
6024 	/* keep a pointer to the data-in buffer */
6025 	datain = (char *)VD_SCSI_DATA_IN(vd_scsi);
6026 
6027 	/* copy-in request data to the data-out buffer */
6028 	dataout = (char *)VD_SCSI_DATA_OUT(vd_scsi);
6029 	if (!(uscsi.uscsi_flags & USCSI_READ)) {
6030 		if (ddi_copyin(uscsi.uscsi_bufaddr, dataout, dataout_len,
6031 		    mode)) {
6032 			rv = EFAULT;
6033 			goto done;
6034 		}
6035 	}
6036 
6037 	/* submit the request */
6038 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
6039 	    0, 0, VIO_both_dir, B_FALSE);
6040 
6041 	if (rv != 0)
6042 		goto done;
6043 
6044 	/* update scsi status */
6045 	uscsi.uscsi_status = vd_scsi->cmd_status;
6046 
6047 	/* update sense data */
6048 	if ((uscsi.uscsi_flags & USCSI_RQENABLE) &&
6049 	    (uscsi.uscsi_status == STATUS_CHECK ||
6050 	    uscsi.uscsi_status == STATUS_TERMINATED)) {
6051 
6052 		uscsi.uscsi_rqstatus = vd_scsi->sense_status;
6053 
6054 		if (uscsi.uscsi_rqstatus == STATUS_GOOD) {
6055 			uscsi.uscsi_rqresid = uscsi.uscsi_rqlen -
6056 			    vd_scsi->sense_len;
6057 			if (ddi_copyout(sense, uscsi.uscsi_rqbuf,
6058 			    vd_scsi->sense_len, mode) != 0) {
6059 				rv = EFAULT;
6060 				goto done;
6061 			}
6062 		}
6063 	}
6064 
6065 	/* update request data */
6066 	if (uscsi.uscsi_status == STATUS_GOOD) {
6067 		if (uscsi.uscsi_flags & USCSI_READ) {
6068 			uscsi.uscsi_resid = uscsi.uscsi_buflen -
6069 			    vd_scsi->datain_len;
6070 			if (ddi_copyout(datain, uscsi.uscsi_bufaddr,
6071 			    vd_scsi->datain_len, mode) != 0) {
6072 				rv = EFAULT;
6073 				goto done;
6074 			}
6075 		} else {
6076 			uscsi.uscsi_resid = uscsi.uscsi_buflen -
6077 			    vd_scsi->dataout_len;
6078 		}
6079 	}
6080 
6081 	/* copy-out result */
6082 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
6083 		uscsi_cmdtouscsi_cmd32((&uscsi), (&uscsi32));
6084 		if (ddi_copyout(&uscsi32, arg, sizeof (struct uscsi_cmd32),
6085 		    mode) != 0) {
6086 			rv = EFAULT;
6087 			goto done;
6088 		}
6089 	} else {
6090 		if (ddi_copyout(&uscsi, arg, sizeof (struct uscsi_cmd),
6091 		    mode) != 0) {
6092 			rv = EFAULT;
6093 			goto done;
6094 		}
6095 	}
6096 
6097 	/* get the return code from the SCSI command status */
6098 	rv = vdc_scsi_status(vdc, vd_scsi,
6099 	    !(uscsi.uscsi_flags & USCSI_SILENT));
6100 
6101 done:
6102 	kmem_free(vd_scsi, vd_scsi_len);
6103 	return (rv);
6104 }
6105 
6106 /*
6107  * Create a VD_OP_SCSICMD buffer for a SCSI PERSISTENT IN command.
6108  *
6109  * Arguments:
6110  *	cmd		- SCSI PERSISTENT IN command
6111  *	len		- length of the SCSI input buffer
6112  *	vd_scsi_len	- return the length of the allocated buffer
6113  *
6114  * Returned Value:
6115  *	a pointer to the allocated VD_OP_SCSICMD buffer.
6116  */
6117 static vd_scsi_t *
6118 vdc_scsi_alloc_persistent_in(uchar_t cmd, int len, int *vd_scsi_len)
6119 {
6120 	int cdb_len, sense_len, datain_len, dataout_len;
6121 	vd_scsi_t *vd_scsi;
6122 	union scsi_cdb *cdb;
6123 
6124 	cdb_len = CDB_GROUP1;
6125 	sense_len = sizeof (struct scsi_extended_sense);
6126 	datain_len = len;
6127 	dataout_len = 0;
6128 
6129 	vd_scsi = vdc_scsi_alloc(cdb_len, sense_len, datain_len, dataout_len,
6130 	    vd_scsi_len);
6131 
6132 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
6133 
6134 	/* set cdb */
6135 	cdb->scc_cmd = SCMD_PERSISTENT_RESERVE_IN;
6136 	cdb->cdb_opaque[1] = cmd;
6137 	FORMG1COUNT(cdb, datain_len);
6138 
6139 	vd_scsi->timeout = vdc_scsi_timeout;
6140 
6141 	return (vd_scsi);
6142 }
6143 
6144 /*
6145  * Create a VD_OP_SCSICMD buffer for a SCSI PERSISTENT OUT command.
6146  *
6147  * Arguments:
6148  *	cmd		- SCSI PERSISTENT OUT command
6149  *	len		- length of the SCSI output buffer
6150  *	vd_scsi_len	- return the length of the allocated buffer
6151  *
6152  * Returned Code:
6153  *	a pointer to the allocated VD_OP_SCSICMD buffer.
6154  */
6155 static vd_scsi_t *
6156 vdc_scsi_alloc_persistent_out(uchar_t cmd, int len, int *vd_scsi_len)
6157 {
6158 	int cdb_len, sense_len, datain_len, dataout_len;
6159 	vd_scsi_t *vd_scsi;
6160 	union scsi_cdb *cdb;
6161 
6162 	cdb_len = CDB_GROUP1;
6163 	sense_len = sizeof (struct scsi_extended_sense);
6164 	datain_len = 0;
6165 	dataout_len = len;
6166 
6167 	vd_scsi = vdc_scsi_alloc(cdb_len, sense_len, datain_len, dataout_len,
6168 	    vd_scsi_len);
6169 
6170 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
6171 
6172 	/* set cdb */
6173 	cdb->scc_cmd = SCMD_PERSISTENT_RESERVE_OUT;
6174 	cdb->cdb_opaque[1] = cmd;
6175 	FORMG1COUNT(cdb, dataout_len);
6176 
6177 	vd_scsi->timeout = vdc_scsi_timeout;
6178 
6179 	return (vd_scsi);
6180 }
6181 
6182 /*
6183  * Implement the MHIOCGRP_INKEYS mhd(4I) ioctl. The ioctl is converted
6184  * to a SCSI PERSISTENT IN READ KEYS command which is sent to the vdisk
6185  * server with a VD_OP_SCSICMD operation.
6186  */
6187 static int
6188 vdc_mhd_inkeys(vdc_t *vdc, caddr_t arg, int mode)
6189 {
6190 	vd_scsi_t *vd_scsi;
6191 	mhioc_inkeys_t inkeys;
6192 	mhioc_key_list_t klist;
6193 	struct mhioc_inkeys32 inkeys32;
6194 	struct mhioc_key_list32 klist32;
6195 	sd_prin_readkeys_t *scsi_keys;
6196 	void *user_keys;
6197 	int vd_scsi_len;
6198 	int listsize, listlen, rv;
6199 
6200 	/* copyin arguments */
6201 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
6202 		rv = ddi_copyin(arg, &inkeys32, sizeof (inkeys32), mode);
6203 		if (rv != 0)
6204 			return (EFAULT);
6205 
6206 		rv = ddi_copyin((caddr_t)(uintptr_t)inkeys32.li, &klist32,
6207 		    sizeof (klist32), mode);
6208 		if (rv != 0)
6209 			return (EFAULT);
6210 
6211 		listsize = klist32.listsize;
6212 	} else {
6213 		rv = ddi_copyin(arg, &inkeys, sizeof (inkeys), mode);
6214 		if (rv != 0)
6215 			return (EFAULT);
6216 
6217 		rv = ddi_copyin(inkeys.li, &klist, sizeof (klist), mode);
6218 		if (rv != 0)
6219 			return (EFAULT);
6220 
6221 		listsize = klist.listsize;
6222 	}
6223 
6224 	/* build SCSI VD_OP request */
6225 	vd_scsi = vdc_scsi_alloc_persistent_in(SD_READ_KEYS,
6226 	    sizeof (sd_prin_readkeys_t) - sizeof (caddr_t) +
6227 	    (sizeof (mhioc_resv_key_t) * listsize), &vd_scsi_len);
6228 
6229 	scsi_keys = (sd_prin_readkeys_t *)VD_SCSI_DATA_IN(vd_scsi);
6230 
6231 	/* submit the request */
6232 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
6233 	    0, 0, VIO_both_dir, B_FALSE);
6234 
6235 	if (rv != 0)
6236 		goto done;
6237 
6238 	listlen = scsi_keys->len / MHIOC_RESV_KEY_SIZE;
6239 
6240 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
6241 		inkeys32.generation = scsi_keys->generation;
6242 		rv = ddi_copyout(&inkeys32, arg, sizeof (inkeys32), mode);
6243 		if (rv != 0) {
6244 			rv = EFAULT;
6245 			goto done;
6246 		}
6247 
6248 		klist32.listlen = listlen;
6249 		rv = ddi_copyout(&klist32, (caddr_t)(uintptr_t)inkeys32.li,
6250 		    sizeof (klist32), mode);
6251 		if (rv != 0) {
6252 			rv = EFAULT;
6253 			goto done;
6254 		}
6255 
6256 		user_keys = (caddr_t)(uintptr_t)klist32.list;
6257 	} else {
6258 		inkeys.generation = scsi_keys->generation;
6259 		rv = ddi_copyout(&inkeys, arg, sizeof (inkeys), mode);
6260 		if (rv != 0) {
6261 			rv = EFAULT;
6262 			goto done;
6263 		}
6264 
6265 		klist.listlen = listlen;
6266 		rv = ddi_copyout(&klist, inkeys.li, sizeof (klist), mode);
6267 		if (rv != 0) {
6268 			rv = EFAULT;
6269 			goto done;
6270 		}
6271 
6272 		user_keys = klist.list;
6273 	}
6274 
6275 	/* copy out keys */
6276 	if (listlen > 0 && listsize > 0) {
6277 		if (listsize < listlen)
6278 			listlen = listsize;
6279 		rv = ddi_copyout(&scsi_keys->keylist, user_keys,
6280 		    listlen * MHIOC_RESV_KEY_SIZE, mode);
6281 		if (rv != 0)
6282 			rv = EFAULT;
6283 	}
6284 
6285 	if (rv == 0)
6286 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
6287 
6288 done:
6289 	kmem_free(vd_scsi, vd_scsi_len);
6290 
6291 	return (rv);
6292 }
6293 
6294 /*
6295  * Implement the MHIOCGRP_INRESV mhd(4I) ioctl. The ioctl is converted
6296  * to a SCSI PERSISTENT IN READ RESERVATION command which is sent to
6297  * the vdisk server with a VD_OP_SCSICMD operation.
6298  */
6299 static int
6300 vdc_mhd_inresv(vdc_t *vdc, caddr_t arg, int mode)
6301 {
6302 	vd_scsi_t *vd_scsi;
6303 	mhioc_inresvs_t inresv;
6304 	mhioc_resv_desc_list_t rlist;
6305 	struct mhioc_inresvs32 inresv32;
6306 	struct mhioc_resv_desc_list32 rlist32;
6307 	mhioc_resv_desc_t mhd_resv;
6308 	sd_prin_readresv_t *scsi_resv;
6309 	sd_readresv_desc_t *resv;
6310 	mhioc_resv_desc_t *user_resv;
6311 	int vd_scsi_len;
6312 	int listsize, listlen, i, rv;
6313 
6314 	/* copyin arguments */
6315 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
6316 		rv = ddi_copyin(arg, &inresv32, sizeof (inresv32), mode);
6317 		if (rv != 0)
6318 			return (EFAULT);
6319 
6320 		rv = ddi_copyin((caddr_t)(uintptr_t)inresv32.li, &rlist32,
6321 		    sizeof (rlist32), mode);
6322 		if (rv != 0)
6323 			return (EFAULT);
6324 
6325 		listsize = rlist32.listsize;
6326 	} else {
6327 		rv = ddi_copyin(arg, &inresv, sizeof (inresv), mode);
6328 		if (rv != 0)
6329 			return (EFAULT);
6330 
6331 		rv = ddi_copyin(inresv.li, &rlist, sizeof (rlist), mode);
6332 		if (rv != 0)
6333 			return (EFAULT);
6334 
6335 		listsize = rlist.listsize;
6336 	}
6337 
6338 	/* build SCSI VD_OP request */
6339 	vd_scsi = vdc_scsi_alloc_persistent_in(SD_READ_RESV,
6340 	    sizeof (sd_prin_readresv_t) - sizeof (caddr_t) +
6341 	    (SCSI3_RESV_DESC_LEN * listsize), &vd_scsi_len);
6342 
6343 	scsi_resv = (sd_prin_readresv_t *)VD_SCSI_DATA_IN(vd_scsi);
6344 
6345 	/* submit the request */
6346 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
6347 	    0, 0, VIO_both_dir, B_FALSE);
6348 
6349 	if (rv != 0)
6350 		goto done;
6351 
6352 	listlen = scsi_resv->len / SCSI3_RESV_DESC_LEN;
6353 
6354 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
6355 		inresv32.generation = scsi_resv->generation;
6356 		rv = ddi_copyout(&inresv32, arg, sizeof (inresv32), mode);
6357 		if (rv != 0) {
6358 			rv = EFAULT;
6359 			goto done;
6360 		}
6361 
6362 		rlist32.listlen = listlen;
6363 		rv = ddi_copyout(&rlist32, (caddr_t)(uintptr_t)inresv32.li,
6364 		    sizeof (rlist32), mode);
6365 		if (rv != 0) {
6366 			rv = EFAULT;
6367 			goto done;
6368 		}
6369 
6370 		user_resv = (mhioc_resv_desc_t *)(uintptr_t)rlist32.list;
6371 	} else {
6372 		inresv.generation = scsi_resv->generation;
6373 		rv = ddi_copyout(&inresv, arg, sizeof (inresv), mode);
6374 		if (rv != 0) {
6375 			rv = EFAULT;
6376 			goto done;
6377 		}
6378 
6379 		rlist.listlen = listlen;
6380 		rv = ddi_copyout(&rlist, inresv.li, sizeof (rlist), mode);
6381 		if (rv != 0) {
6382 			rv = EFAULT;
6383 			goto done;
6384 		}
6385 
6386 		user_resv = rlist.list;
6387 	}
6388 
6389 	/* copy out reservations */
6390 	if (listsize > 0 && listlen > 0) {
6391 		if (listsize < listlen)
6392 			listlen = listsize;
6393 		resv = (sd_readresv_desc_t *)&scsi_resv->readresv_desc;
6394 
6395 		for (i = 0; i < listlen; i++) {
6396 			mhd_resv.type = resv->type;
6397 			mhd_resv.scope = resv->scope;
6398 			mhd_resv.scope_specific_addr =
6399 			    BE_32(resv->scope_specific_addr);
6400 			bcopy(&resv->resvkey, &mhd_resv.key,
6401 			    MHIOC_RESV_KEY_SIZE);
6402 
6403 			rv = ddi_copyout(&mhd_resv, user_resv,
6404 			    sizeof (mhd_resv), mode);
6405 			if (rv != 0) {
6406 				rv = EFAULT;
6407 				goto done;
6408 			}
6409 			resv++;
6410 			user_resv++;
6411 		}
6412 	}
6413 
6414 	if (rv == 0)
6415 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
6416 
6417 done:
6418 	kmem_free(vd_scsi, vd_scsi_len);
6419 	return (rv);
6420 }
6421 
6422 /*
6423  * Implement the MHIOCGRP_REGISTER mhd(4I) ioctl. The ioctl is converted
6424  * to a SCSI PERSISTENT OUT REGISTER command which is sent to the vdisk
6425  * server with a VD_OP_SCSICMD operation.
6426  */
6427 static int
6428 vdc_mhd_register(vdc_t *vdc, caddr_t arg, int mode)
6429 {
6430 	vd_scsi_t *vd_scsi;
6431 	sd_prout_t *scsi_prout;
6432 	mhioc_register_t mhd_reg;
6433 	int vd_scsi_len, rv;
6434 
6435 	/* copyin arguments */
6436 	rv = ddi_copyin(arg, &mhd_reg, sizeof (mhd_reg), mode);
6437 	if (rv != 0)
6438 		return (EFAULT);
6439 
6440 	/* build SCSI VD_OP request */
6441 	vd_scsi = vdc_scsi_alloc_persistent_out(SD_SCSI3_REGISTER,
6442 	    sizeof (sd_prout_t), &vd_scsi_len);
6443 
6444 	/* set parameters */
6445 	scsi_prout = (sd_prout_t *)VD_SCSI_DATA_OUT(vd_scsi);
6446 	bcopy(mhd_reg.oldkey.key, scsi_prout->res_key, MHIOC_RESV_KEY_SIZE);
6447 	bcopy(mhd_reg.newkey.key, scsi_prout->service_key, MHIOC_RESV_KEY_SIZE);
6448 	scsi_prout->aptpl = (uchar_t)mhd_reg.aptpl;
6449 
6450 	/* submit the request */
6451 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
6452 	    0, 0, VIO_both_dir, B_FALSE);
6453 
6454 	if (rv == 0)
6455 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
6456 
6457 	kmem_free(vd_scsi, vd_scsi_len);
6458 	return (rv);
6459 }
6460 
6461 /*
6462  * Implement the MHIOCGRP_RESERVE mhd(4I) ioctl. The ioctl is converted
6463  * to a SCSI PERSISTENT OUT RESERVE command which is sent to the vdisk
6464  * server with a VD_OP_SCSICMD operation.
6465  */
6466 static int
6467 vdc_mhd_reserve(vdc_t *vdc, caddr_t arg, int mode)
6468 {
6469 	union scsi_cdb *cdb;
6470 	vd_scsi_t *vd_scsi;
6471 	sd_prout_t *scsi_prout;
6472 	mhioc_resv_desc_t mhd_resv;
6473 	int vd_scsi_len, rv;
6474 
6475 	/* copyin arguments */
6476 	rv = ddi_copyin(arg, &mhd_resv, sizeof (mhd_resv), mode);
6477 	if (rv != 0)
6478 		return (EFAULT);
6479 
6480 	/* build SCSI VD_OP request */
6481 	vd_scsi = vdc_scsi_alloc_persistent_out(SD_SCSI3_RESERVE,
6482 	    sizeof (sd_prout_t), &vd_scsi_len);
6483 
6484 	/* set parameters */
6485 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
6486 	scsi_prout = (sd_prout_t *)VD_SCSI_DATA_OUT(vd_scsi);
6487 	bcopy(mhd_resv.key.key, scsi_prout->res_key, MHIOC_RESV_KEY_SIZE);
6488 	scsi_prout->scope_address = mhd_resv.scope_specific_addr;
6489 	cdb->cdb_opaque[2] = mhd_resv.type;
6490 
6491 	/* submit the request */
6492 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
6493 	    0, 0, VIO_both_dir, B_FALSE);
6494 
6495 	if (rv == 0)
6496 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
6497 
6498 	kmem_free(vd_scsi, vd_scsi_len);
6499 	return (rv);
6500 }
6501 
6502 /*
6503  * Implement the MHIOCGRP_PREEMPTANDABORT mhd(4I) ioctl. The ioctl is
6504  * converted to a SCSI PERSISTENT OUT PREEMPT AND ABORT command which
6505  * is sent to the vdisk server with a VD_OP_SCSICMD operation.
6506  */
6507 static int
6508 vdc_mhd_preemptabort(vdc_t *vdc, caddr_t arg, int mode)
6509 {
6510 	union scsi_cdb *cdb;
6511 	vd_scsi_t *vd_scsi;
6512 	sd_prout_t *scsi_prout;
6513 	mhioc_preemptandabort_t mhd_preempt;
6514 	int vd_scsi_len, rv;
6515 
6516 	/* copyin arguments */
6517 	rv = ddi_copyin(arg, &mhd_preempt, sizeof (mhd_preempt), mode);
6518 	if (rv != 0)
6519 		return (EFAULT);
6520 
6521 	/* build SCSI VD_OP request */
6522 	vd_scsi = vdc_scsi_alloc_persistent_out(SD_SCSI3_PREEMPTANDABORT,
6523 	    sizeof (sd_prout_t), &vd_scsi_len);
6524 
6525 	/* set parameters */
6526 	vd_scsi->task_attribute = VD_SCSI_TASK_ACA;
6527 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
6528 	scsi_prout = (sd_prout_t *)VD_SCSI_DATA_OUT(vd_scsi);
6529 	bcopy(mhd_preempt.resvdesc.key.key, scsi_prout->res_key,
6530 	    MHIOC_RESV_KEY_SIZE);
6531 	bcopy(mhd_preempt.victim_key.key, scsi_prout->service_key,
6532 	    MHIOC_RESV_KEY_SIZE);
6533 	scsi_prout->scope_address = mhd_preempt.resvdesc.scope_specific_addr;
6534 	cdb->cdb_opaque[2] = mhd_preempt.resvdesc.type;
6535 
6536 	/* submit the request */
6537 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
6538 	    0, 0, VIO_both_dir, B_FALSE);
6539 
6540 	if (rv == 0)
6541 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
6542 
6543 	kmem_free(vd_scsi, vd_scsi_len);
6544 	return (rv);
6545 }
6546 
6547 /*
6548  * Implement the MHIOCGRP_REGISTERANDIGNOREKEY mhd(4I) ioctl. The ioctl
6549  * is converted to a SCSI PERSISTENT OUT REGISTER AND IGNORE EXISTING KEY
6550  * command which is sent to the vdisk server with a VD_OP_SCSICMD operation.
6551  */
6552 static int
6553 vdc_mhd_registerignore(vdc_t *vdc, caddr_t arg, int mode)
6554 {
6555 	vd_scsi_t *vd_scsi;
6556 	sd_prout_t *scsi_prout;
6557 	mhioc_registerandignorekey_t mhd_regi;
6558 	int vd_scsi_len, rv;
6559 
6560 	/* copyin arguments */
6561 	rv = ddi_copyin(arg, &mhd_regi, sizeof (mhd_regi), mode);
6562 	if (rv != 0)
6563 		return (EFAULT);
6564 
6565 	/* build SCSI VD_OP request */
6566 	vd_scsi = vdc_scsi_alloc_persistent_out(SD_SCSI3_REGISTERANDIGNOREKEY,
6567 	    sizeof (sd_prout_t), &vd_scsi_len);
6568 
6569 	/* set parameters */
6570 	scsi_prout = (sd_prout_t *)VD_SCSI_DATA_OUT(vd_scsi);
6571 	bcopy(mhd_regi.newkey.key, scsi_prout->service_key,
6572 	    MHIOC_RESV_KEY_SIZE);
6573 	scsi_prout->aptpl = (uchar_t)mhd_regi.aptpl;
6574 
6575 	/* submit the request */
6576 	rv = vdc_do_sync_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
6577 	    0, 0, VIO_both_dir, B_FALSE);
6578 
6579 	if (rv == 0)
6580 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
6581 
6582 	kmem_free(vd_scsi, vd_scsi_len);
6583 	return (rv);
6584 }
6585 
6586 /*
6587  * This function is used to send a (simple) SCSI command and check errors.
6588  */
6589 static int
6590 vdc_eio_scsi_cmd(vdc_t *vdc, uchar_t scmd, int flags)
6591 {
6592 	int cdb_len, sense_len, vd_scsi_len;
6593 	vd_scsi_t *vd_scsi;
6594 	union scsi_cdb *cdb;
6595 	int rv;
6596 
6597 	ASSERT(scmd == SCMD_TEST_UNIT_READY || scmd == SCMD_WRITE_G1);
6598 
6599 	if (scmd == SCMD_WRITE_G1)
6600 		cdb_len = CDB_GROUP1;
6601 	else
6602 		cdb_len = CDB_GROUP0;
6603 
6604 	sense_len = sizeof (struct scsi_extended_sense);
6605 
6606 	vd_scsi = vdc_scsi_alloc(cdb_len, sense_len, 0, 0, &vd_scsi_len);
6607 
6608 	/* set cdb */
6609 	cdb = VD_SCSI_DATA_CDB(vd_scsi);
6610 	cdb->scc_cmd = scmd;
6611 
6612 	vd_scsi->timeout = vdc_scsi_timeout;
6613 
6614 	/*
6615 	 * Submit the request. Note the operation should not request that any
6616 	 * error is checked because this function is precisely called when
6617 	 * checking errors.
6618 	 */
6619 	ASSERT((flags & VDC_OP_ERRCHK) == 0);
6620 
6621 	rv = vdc_do_op(vdc, VD_OP_SCSICMD, (caddr_t)vd_scsi, vd_scsi_len,
6622 	    0, 0, NULL, VIO_both_dir, flags);
6623 
6624 	if (rv == 0)
6625 		rv = vdc_scsi_status(vdc, vd_scsi, B_FALSE);
6626 
6627 	kmem_free(vd_scsi, vd_scsi_len);
6628 	return (rv);
6629 }
6630 
6631 /*
6632  * This function is used to check if a SCSI backend is accessible. It will
6633  * also detect reservation conflict if failfast is enabled, and panic the
6634  * system in that case.
6635  *
6636  * Returned Code:
6637  *	0	- disk is accessible
6638  *	!= 0	- disk is inaccessible or unable to check if disk is accessible
6639  */
6640 static int
6641 vdc_eio_scsi_check(vdc_t *vdc, int flags)
6642 {
6643 	int failure = 0;
6644 	int rv;
6645 
6646 	/*
6647 	 * Send a TEST UNIT READY command. The command will panic
6648 	 * the system if it fails with a reservation conflict and
6649 	 * failfast is enabled. If there is a reservation conflict
6650 	 * and failfast is not enabled then the function will return
6651 	 * EACCES. In that case, there's no problem with accessing
6652 	 * the backend, it is just reserved.
6653 	 */
6654 	rv = vdc_eio_scsi_cmd(vdc, SCMD_TEST_UNIT_READY, flags);
6655 	if (rv != 0 && rv != EACCES)
6656 		failure++;
6657 
6658 	/* we don't need to do more checking if failfast is not enabled */
6659 	if (vdc->failfast_interval == 0)
6660 		return (failure);
6661 
6662 	/*
6663 	 * With SPC-3 compliant devices TEST UNIT READY will succeed on
6664 	 * a reserved device, so we also do a WRITE(10) of zero byte in
6665 	 * order to provoke a Reservation Conflict status on those newer
6666 	 * devices.
6667 	 */
6668 	if (vdc_eio_scsi_cmd(vdc, SCMD_WRITE_G1, flags) != 0)
6669 		failure++;
6670 
6671 	return (failure);
6672 }
6673 
6674 /*
6675  * This function is used to check if a backend is effectively accessible.
6676  *
6677  * Returned Code:
6678  *	0	- disk is accessible
6679  *	!= 0	- disk is inaccessible or unable to check if disk is accessible
6680  */
6681 static int
6682 vdc_eio_check(vdc_t *vdc, int flags)
6683 {
6684 	char *buffer;
6685 	diskaddr_t blkno;
6686 	int rv;
6687 
6688 	ASSERT((flags & VDC_OP_ERRCHK) == 0);
6689 
6690 	flags |= VDC_OP_DRING_RESERVED;
6691 
6692 	if (VD_OP_SUPPORTED(vdc->operations, VD_OP_SCSICMD))
6693 		return (vdc_eio_scsi_check(vdc, flags));
6694 
6695 	ASSERT(vdc->failfast_interval == 0);
6696 
6697 	/*
6698 	 * If the backend does not support SCSI operations then we simply
6699 	 * check if the backend is accessible by reading some data blocks.
6700 	 * We first try to read a random block, to try to avoid getting
6701 	 * a block that might have been cached on the service domain. Then
6702 	 * we try the last block, and finally the first block.
6703 	 *
6704 	 * We return success as soon as we are able to read any block.
6705 	 */
6706 	buffer = kmem_alloc(vdc->vdisk_bsize, KM_SLEEP);
6707 
6708 	if (vdc->vdisk_size > 0) {
6709 
6710 		/* try a random block */
6711 		(void) random_get_pseudo_bytes((uint8_t *)&blkno,
6712 		    sizeof (diskaddr_t));
6713 		blkno = blkno % vdc->vdisk_size;
6714 		rv = vdc_do_op(vdc, VD_OP_BREAD, (caddr_t)buffer,
6715 		    vdc->vdisk_bsize, VD_SLICE_NONE, blkno, NULL,
6716 		    VIO_read_dir, flags);
6717 
6718 		if (rv == 0)
6719 			goto done;
6720 
6721 		/* try the last block */
6722 		blkno = vdc->vdisk_size - 1;
6723 		rv = vdc_do_op(vdc, VD_OP_BREAD, (caddr_t)buffer,
6724 		    vdc->vdisk_bsize, VD_SLICE_NONE, blkno, NULL,
6725 		    VIO_read_dir, flags);
6726 
6727 		if (rv == 0)
6728 			goto done;
6729 	}
6730 
6731 	/* try block 0 */
6732 	blkno = 0;
6733 	rv = vdc_do_op(vdc, VD_OP_BREAD, (caddr_t)buffer, vdc->vdisk_bsize,
6734 	    VD_SLICE_NONE, blkno, NULL, VIO_read_dir, flags);
6735 
6736 done:
6737 	kmem_free(buffer, vdc->vdisk_bsize);
6738 	return (rv);
6739 }
6740 
6741 /*
6742  * Add a pending I/O to the eio queue. An I/O is added to this queue
6743  * when it has failed and failfast is enabled or the vdisk has multiple
6744  * servers. It will then be handled by the eio thread (vdc_eio_thread).
6745  * The eio queue is ordered starting with the most recent I/O added.
6746  */
6747 static vdc_io_t *
6748 vdc_eio_queue(vdc_t *vdc, int index)
6749 {
6750 	vdc_io_t *vio;
6751 
6752 	ASSERT(MUTEX_HELD(&vdc->lock));
6753 
6754 	vio = kmem_alloc(sizeof (vdc_io_t), KM_SLEEP);
6755 	vio->vio_next = vdc->eio_queue;
6756 	vio->vio_index = index;
6757 	vio->vio_qtime = ddi_get_lbolt();
6758 
6759 	vdc->eio_queue = vio;
6760 
6761 	/* notify the eio thread that a new I/O is queued */
6762 	cv_signal(&vdc->eio_cv);
6763 
6764 	return (vio);
6765 }
6766 
6767 /*
6768  * Remove I/Os added before the indicated deadline from the eio queue. A
6769  * deadline of 0 means that all I/Os have to be unqueued. The complete_io
6770  * boolean specifies if unqueued I/Os should be marked as completed or not.
6771  */
6772 static void
6773 vdc_eio_unqueue(vdc_t *vdc, clock_t deadline, boolean_t complete_io)
6774 {
6775 	struct buf *buf;
6776 	vdc_io_t *vio, *vio_tmp;
6777 	int index, op;
6778 
6779 	ASSERT(MUTEX_HELD(&vdc->lock));
6780 
6781 	vio_tmp = NULL;
6782 	vio = vdc->eio_queue;
6783 
6784 	if (deadline != 0) {
6785 		/*
6786 		 * Skip any io queued after the deadline. The eio queue is
6787 		 * ordered starting with the last I/O added to the queue.
6788 		 */
6789 		while (vio != NULL && vio->vio_qtime > deadline) {
6790 			vio_tmp = vio;
6791 			vio = vio->vio_next;
6792 		}
6793 	}
6794 
6795 	if (vio == NULL)
6796 		/* nothing to unqueue */
6797 		return;
6798 
6799 	/* update the queue */
6800 	if (vio_tmp == NULL)
6801 		vdc->eio_queue = NULL;
6802 	else
6803 		vio_tmp->vio_next = NULL;
6804 
6805 	/*
6806 	 * Free and complete unqueued I/Os if this was requested. All I/Os
6807 	 * have a block I/O data transfer structure (buf) and they are
6808 	 * completed by calling biodone().
6809 	 */
6810 	while (vio != NULL) {
6811 		vio_tmp = vio->vio_next;
6812 
6813 		if (complete_io) {
6814 			index = vio->vio_index;
6815 			op = vdc->local_dring[index].operation;
6816 			buf = vdc->local_dring[index].buf;
6817 			(void) vdc_depopulate_descriptor(vdc, index);
6818 			ASSERT(buf->b_flags & B_ERROR);
6819 			if (op == VD_OP_BREAD || op == VD_OP_BWRITE) {
6820 				VD_UPDATE_ERR_STATS(vdc, vd_softerrs);
6821 				VD_KSTAT_RUNQ_EXIT(vdc);
6822 				DTRACE_IO1(done, buf_t *, buf);
6823 			}
6824 			biodone(buf);
6825 		}
6826 
6827 		kmem_free(vio, sizeof (vdc_io_t));
6828 		vio = vio_tmp;
6829 	}
6830 }
6831 
6832 /*
6833  * Error I/O Thread.  There is one eio thread for each virtual disk that
6834  * has multiple servers or for which failfast is enabled. Failfast can only
6835  * be enabled for vdisk supporting SCSI commands.
6836  *
6837  * While failfast is enabled, the eio thread sends a TEST UNIT READY
6838  * and a zero size WRITE(10) SCSI commands on a regular basis to check that
6839  * we still have access to the disk. If a command fails with a RESERVATION
6840  * CONFLICT error then the system will immediatly panic.
6841  *
6842  * The eio thread is also woken up when an I/O has failed. It then checks
6843  * the access to the disk to ensure that the I/O failure was not due to a
6844  * reservation conflict or to the backend been inaccessible.
6845  *
6846  */
6847 static void
6848 vdc_eio_thread(void *arg)
6849 {
6850 	int status;
6851 	vdc_t *vdc = (vdc_t *)arg;
6852 	clock_t starttime, timeout = drv_usectohz(vdc->failfast_interval);
6853 
6854 	mutex_enter(&vdc->lock);
6855 
6856 	while (vdc->failfast_interval != 0 || vdc->num_servers > 1) {
6857 		/*
6858 		 * Wait if there is nothing in the eio queue or if the state
6859 		 * is not VDC_STATE_RUNNING.
6860 		 */
6861 		if (vdc->eio_queue == NULL || vdc->state != VDC_STATE_RUNNING) {
6862 			if (vdc->failfast_interval != 0) {
6863 				timeout = ddi_get_lbolt() +
6864 				    drv_usectohz(vdc->failfast_interval);
6865 				(void) cv_timedwait(&vdc->eio_cv, &vdc->lock,
6866 				    timeout);
6867 			} else {
6868 				ASSERT(vdc->num_servers > 1);
6869 				(void) cv_wait(&vdc->eio_cv, &vdc->lock);
6870 			}
6871 
6872 			if (vdc->state != VDC_STATE_RUNNING)
6873 				continue;
6874 		}
6875 
6876 		mutex_exit(&vdc->lock);
6877 
6878 		starttime = ddi_get_lbolt();
6879 
6880 		/* check error */
6881 		status = vdc_eio_check(vdc, VDC_OP_STATE_RUNNING);
6882 
6883 		mutex_enter(&vdc->lock);
6884 		/*
6885 		 * We have dropped the lock to check the backend so we have
6886 		 * to check that the eio thread is still enabled.
6887 		 */
6888 		if (vdc->failfast_interval == 0 && vdc->num_servers <= 1)
6889 			break;
6890 
6891 		/*
6892 		 * If the eio queue is empty or we are not in running state
6893 		 * anymore then there is nothing to do.
6894 		 */
6895 		if (vdc->state != VDC_STATE_RUNNING || vdc->eio_queue == NULL)
6896 			continue;
6897 
6898 		if (status == 0) {
6899 			/*
6900 			 * The backend access has been successfully checked,
6901 			 * we can complete any I/O queued before the last check.
6902 			 */
6903 			vdc_eio_unqueue(vdc, starttime, B_TRUE);
6904 
6905 		} else if (vdc->num_servers > 1) {
6906 			/*
6907 			 * The backend is inaccessible for a disk with multiple
6908 			 * servers. So we force a reset to switch to another
6909 			 * server. The reset will also clear the eio queue and
6910 			 * resubmit all pending I/Os.
6911 			 */
6912 			mutex_enter(&vdc->read_lock);
6913 			vdc->read_state = VDC_READ_RESET;
6914 			cv_signal(&vdc->read_cv);
6915 			mutex_exit(&vdc->read_lock);
6916 		} else {
6917 			/*
6918 			 * There is only one path and the backend is not
6919 			 * accessible, so I/Os are actually failing because
6920 			 * of that. So we can complete I/O queued before the
6921 			 * last check.
6922 			 */
6923 			vdc_eio_unqueue(vdc, starttime, B_TRUE);
6924 		}
6925 	}
6926 
6927 	/*
6928 	 * The thread is being stopped so we can complete any queued I/O.
6929 	 */
6930 	vdc_eio_unqueue(vdc, 0, B_TRUE);
6931 	vdc->eio_thread = NULL;
6932 	mutex_exit(&vdc->lock);
6933 	thread_exit();
6934 }
6935 
6936 /*
6937  * Implement the MHIOCENFAILFAST mhd(4I) ioctl.
6938  */
6939 static int
6940 vdc_failfast(vdc_t *vdc, caddr_t arg, int mode)
6941 {
6942 	unsigned int mh_time;
6943 
6944 	if (ddi_copyin((void *)arg, &mh_time, sizeof (int), mode))
6945 		return (EFAULT);
6946 
6947 	mutex_enter(&vdc->lock);
6948 	if (mh_time != 0 && vdc->eio_thread == NULL) {
6949 		vdc->eio_thread = thread_create(NULL, 0,
6950 		    vdc_eio_thread, vdc, 0, &p0, TS_RUN,
6951 		    v.v_maxsyspri - 2);
6952 	}
6953 
6954 	vdc->failfast_interval = ((long)mh_time) * MILLISEC;
6955 	cv_signal(&vdc->eio_cv);
6956 	mutex_exit(&vdc->lock);
6957 
6958 	return (0);
6959 }
6960 
6961 /*
6962  * Implement the MHIOCTKOWN and MHIOCRELEASE mhd(4I) ioctls. These ioctls are
6963  * converted to VD_OP_SET_ACCESS operations.
6964  */
6965 static int
6966 vdc_access_set(vdc_t *vdc, uint64_t flags)
6967 {
6968 	int rv;
6969 
6970 	/* submit owership command request */
6971 	rv = vdc_do_sync_op(vdc, VD_OP_SET_ACCESS, (caddr_t)&flags,
6972 	    sizeof (uint64_t), 0, 0, VIO_both_dir, B_TRUE);
6973 
6974 	return (rv);
6975 }
6976 
6977 /*
6978  * Implement the MHIOCSTATUS mhd(4I) ioctl. This ioctl is converted to a
6979  * VD_OP_GET_ACCESS operation.
6980  */
6981 static int
6982 vdc_access_get(vdc_t *vdc, uint64_t *status)
6983 {
6984 	int rv;
6985 
6986 	/* submit owership command request */
6987 	rv = vdc_do_sync_op(vdc, VD_OP_GET_ACCESS, (caddr_t)status,
6988 	    sizeof (uint64_t), 0, 0, VIO_both_dir, B_TRUE);
6989 
6990 	return (rv);
6991 }
6992 
6993 /*
6994  * Disk Ownership Thread.
6995  *
6996  * When we have taken the ownership of a disk, this thread waits to be
6997  * notified when the LDC channel is reset so that it can recover the
6998  * ownership.
6999  *
7000  * Note that the thread handling the LDC reset (vdc_process_msg_thread())
7001  * can not be used to do the ownership recovery because it has to be
7002  * running to handle the reply message to the ownership operation.
7003  */
7004 static void
7005 vdc_ownership_thread(void *arg)
7006 {
7007 	vdc_t *vdc = (vdc_t *)arg;
7008 	clock_t timeout;
7009 	uint64_t status;
7010 
7011 	mutex_enter(&vdc->ownership_lock);
7012 	mutex_enter(&vdc->lock);
7013 
7014 	while (vdc->ownership & VDC_OWNERSHIP_WANTED) {
7015 
7016 		if ((vdc->ownership & VDC_OWNERSHIP_RESET) ||
7017 		    !(vdc->ownership & VDC_OWNERSHIP_GRANTED)) {
7018 			/*
7019 			 * There was a reset so the ownership has been lost,
7020 			 * try to recover. We do this without using the preempt
7021 			 * option so that we don't steal the ownership from
7022 			 * someone who has preempted us.
7023 			 */
7024 			DMSG(vdc, 0, "[%d] Ownership lost, recovering",
7025 			    vdc->instance);
7026 
7027 			vdc->ownership &= ~(VDC_OWNERSHIP_RESET |
7028 			    VDC_OWNERSHIP_GRANTED);
7029 
7030 			mutex_exit(&vdc->lock);
7031 
7032 			status = vdc_access_set(vdc, VD_ACCESS_SET_EXCLUSIVE |
7033 			    VD_ACCESS_SET_PRESERVE);
7034 
7035 			mutex_enter(&vdc->lock);
7036 
7037 			if (status == 0) {
7038 				DMSG(vdc, 0, "[%d] Ownership recovered",
7039 				    vdc->instance);
7040 				vdc->ownership |= VDC_OWNERSHIP_GRANTED;
7041 			} else {
7042 				DMSG(vdc, 0, "[%d] Fail to recover ownership",
7043 				    vdc->instance);
7044 			}
7045 
7046 		}
7047 
7048 		/*
7049 		 * If we have the ownership then we just wait for an event
7050 		 * to happen (LDC reset), otherwise we will retry to recover
7051 		 * after a delay.
7052 		 */
7053 		if (vdc->ownership & VDC_OWNERSHIP_GRANTED)
7054 			timeout = 0;
7055 		else
7056 			timeout = drv_usectohz(vdc_ownership_delay);
7057 
7058 		/* Release the ownership_lock and wait on the vdc lock */
7059 		mutex_exit(&vdc->ownership_lock);
7060 
7061 		if (timeout == 0)
7062 			(void) cv_wait(&vdc->ownership_cv, &vdc->lock);
7063 		else
7064 			(void) cv_reltimedwait(&vdc->ownership_cv, &vdc->lock,
7065 			    timeout, TR_CLOCK_TICK);
7066 
7067 		mutex_exit(&vdc->lock);
7068 
7069 		mutex_enter(&vdc->ownership_lock);
7070 		mutex_enter(&vdc->lock);
7071 	}
7072 
7073 	vdc->ownership_thread = NULL;
7074 	mutex_exit(&vdc->lock);
7075 	mutex_exit(&vdc->ownership_lock);
7076 
7077 	thread_exit();
7078 }
7079 
7080 static void
7081 vdc_ownership_update(vdc_t *vdc, int ownership_flags)
7082 {
7083 	ASSERT(MUTEX_HELD(&vdc->ownership_lock));
7084 
7085 	mutex_enter(&vdc->lock);
7086 	vdc->ownership = ownership_flags;
7087 	if ((vdc->ownership & VDC_OWNERSHIP_WANTED) &&
7088 	    vdc->ownership_thread == NULL) {
7089 		/* start ownership thread */
7090 		vdc->ownership_thread = thread_create(NULL, 0,
7091 		    vdc_ownership_thread, vdc, 0, &p0, TS_RUN,
7092 		    v.v_maxsyspri - 2);
7093 	} else {
7094 		/* notify the ownership thread */
7095 		cv_signal(&vdc->ownership_cv);
7096 	}
7097 	mutex_exit(&vdc->lock);
7098 }
7099 
7100 /*
7101  * Get the size and the block size of a virtual disk from the vdisk server.
7102  */
7103 static int
7104 vdc_get_capacity(vdc_t *vdc, size_t *dsk_size, size_t *blk_size)
7105 {
7106 	int rv = 0;
7107 	size_t alloc_len;
7108 	vd_capacity_t *vd_cap;
7109 
7110 	ASSERT(MUTEX_NOT_HELD(&vdc->lock));
7111 
7112 	alloc_len = P2ROUNDUP(sizeof (vd_capacity_t), sizeof (uint64_t));
7113 
7114 	vd_cap = kmem_zalloc(alloc_len, KM_SLEEP);
7115 
7116 	rv = vdc_do_sync_op(vdc, VD_OP_GET_CAPACITY, (caddr_t)vd_cap, alloc_len,
7117 	    0, 0, VIO_both_dir, B_TRUE);
7118 
7119 	*dsk_size = vd_cap->vdisk_size;
7120 	*blk_size = vd_cap->vdisk_block_size;
7121 
7122 	kmem_free(vd_cap, alloc_len);
7123 	return (rv);
7124 }
7125 
7126 /*
7127  * Check the disk capacity. Disk size information is updated if size has
7128  * changed.
7129  *
7130  * Return 0 if the disk capacity is available, or non-zero if it is not.
7131  */
7132 static int
7133 vdc_check_capacity(vdc_t *vdc)
7134 {
7135 	size_t dsk_size, blk_size;
7136 	int rv;
7137 
7138 	/*
7139 	 * If the vdisk does not support the VD_OP_GET_CAPACITY operation
7140 	 * then the disk capacity has been retrieved during the handshake
7141 	 * and there's nothing more to do here.
7142 	 */
7143 	if (!VD_OP_SUPPORTED(vdc->operations, VD_OP_GET_CAPACITY))
7144 		return (0);
7145 
7146 	if ((rv = vdc_get_capacity(vdc, &dsk_size, &blk_size)) != 0)
7147 		return (rv);
7148 
7149 	if (dsk_size == VD_SIZE_UNKNOWN || dsk_size == 0 || blk_size == 0)
7150 		return (EINVAL);
7151 
7152 	mutex_enter(&vdc->lock);
7153 	/*
7154 	 * First try to update the VIO block size (which is the same as the
7155 	 * vdisk block size). If this returns an error then that means that
7156 	 * we can not use that block size so basically the vdisk is unusable
7157 	 * and we return an error.
7158 	 */
7159 	rv = vdc_update_vio_bsize(vdc, blk_size);
7160 	if (rv == 0)
7161 		vdc_update_size(vdc, dsk_size, blk_size, vdc->max_xfer_sz);
7162 
7163 	mutex_exit(&vdc->lock);
7164 
7165 	return (rv);
7166 }
7167 
7168 /*
7169  * This structure is used in the dkio(4I) array below.
7170  */
7171 typedef struct vdc_dk_ioctl {
7172 	uint8_t		op;		/* VD_OP_XXX value */
7173 	int		cmd;		/* Solaris ioctl operation number */
7174 	size_t		nbytes;		/* size of structure to be copied */
7175 
7176 	/* function to convert between vDisk and Solaris structure formats */
7177 	int	(*convert)(vdc_t *vdc, void *vd_buf, void *ioctl_arg,
7178 	    int mode, int dir);
7179 } vdc_dk_ioctl_t;
7180 
7181 /*
7182  * Subset of dkio(4I) operations currently supported
7183  */
7184 static vdc_dk_ioctl_t	dk_ioctl[] = {
7185 	{VD_OP_FLUSH,		DKIOCFLUSHWRITECACHE,	0,
7186 		vdc_null_copy_func},
7187 	{VD_OP_GET_WCE,		DKIOCGETWCE,		sizeof (int),
7188 		vdc_get_wce_convert},
7189 	{VD_OP_SET_WCE,		DKIOCSETWCE,		sizeof (int),
7190 		vdc_set_wce_convert},
7191 	{VD_OP_GET_VTOC,	DKIOCGVTOC,		sizeof (vd_vtoc_t),
7192 		vdc_get_vtoc_convert},
7193 	{VD_OP_SET_VTOC,	DKIOCSVTOC,		sizeof (vd_vtoc_t),
7194 		vdc_set_vtoc_convert},
7195 	{VD_OP_GET_VTOC,	DKIOCGEXTVTOC,		sizeof (vd_vtoc_t),
7196 		vdc_get_extvtoc_convert},
7197 	{VD_OP_SET_VTOC,	DKIOCSEXTVTOC,		sizeof (vd_vtoc_t),
7198 		vdc_set_extvtoc_convert},
7199 	{VD_OP_GET_DISKGEOM,	DKIOCGGEOM,		sizeof (vd_geom_t),
7200 		vdc_get_geom_convert},
7201 	{VD_OP_GET_DISKGEOM,	DKIOCG_PHYGEOM,		sizeof (vd_geom_t),
7202 		vdc_get_geom_convert},
7203 	{VD_OP_GET_DISKGEOM,	DKIOCG_VIRTGEOM,	sizeof (vd_geom_t),
7204 		vdc_get_geom_convert},
7205 	{VD_OP_SET_DISKGEOM,	DKIOCSGEOM,		sizeof (vd_geom_t),
7206 		vdc_set_geom_convert},
7207 	{VD_OP_GET_EFI,		DKIOCGETEFI,		0,
7208 		vdc_get_efi_convert},
7209 	{VD_OP_SET_EFI,		DKIOCSETEFI,		0,
7210 		vdc_set_efi_convert},
7211 
7212 	/* DIOCTL_RWCMD is converted to a read or a write */
7213 	{0, DIOCTL_RWCMD,  sizeof (struct dadkio_rwcmd), NULL},
7214 
7215 	/* mhd(4I) non-shared multihost disks ioctls */
7216 	{0, MHIOCTKOWN,				0, vdc_null_copy_func},
7217 	{0, MHIOCRELEASE,			0, vdc_null_copy_func},
7218 	{0, MHIOCSTATUS,			0, vdc_null_copy_func},
7219 	{0, MHIOCQRESERVE,			0, vdc_null_copy_func},
7220 
7221 	/* mhd(4I) shared multihost disks ioctls */
7222 	{0, MHIOCGRP_INKEYS,			0, vdc_null_copy_func},
7223 	{0, MHIOCGRP_INRESV,			0, vdc_null_copy_func},
7224 	{0, MHIOCGRP_REGISTER,			0, vdc_null_copy_func},
7225 	{0, MHIOCGRP_RESERVE,			0, vdc_null_copy_func},
7226 	{0, MHIOCGRP_PREEMPTANDABORT,		0, vdc_null_copy_func},
7227 	{0, MHIOCGRP_REGISTERANDIGNOREKEY,	0, vdc_null_copy_func},
7228 
7229 	/* mhd(4I) failfast ioctl */
7230 	{0, MHIOCENFAILFAST,			0, vdc_null_copy_func},
7231 
7232 	/*
7233 	 * These particular ioctls are not sent to the server - vdc fakes up
7234 	 * the necessary info.
7235 	 */
7236 	{0, DKIOCINFO, sizeof (struct dk_cinfo), vdc_null_copy_func},
7237 	{0, DKIOCGMEDIAINFO, sizeof (struct dk_minfo), vdc_null_copy_func},
7238 	{0, USCSICMD,	sizeof (struct uscsi_cmd), vdc_null_copy_func},
7239 	{0, DKIOCPARTITION, 0, vdc_null_copy_func },
7240 	{0, DKIOCGAPART, 0, vdc_null_copy_func },
7241 	{0, DKIOCREMOVABLE, 0, vdc_null_copy_func},
7242 	{0, CDROMREADOFFSET, 0, vdc_null_copy_func}
7243 };
7244 
7245 /*
7246  * This function handles ioctl requests from the vd_efi_alloc_and_read()
7247  * function and forward them to the vdisk.
7248  */
7249 static int
7250 vd_process_efi_ioctl(void *vdisk, int cmd, uintptr_t arg)
7251 {
7252 	vdc_t *vdc = (vdc_t *)vdisk;
7253 	dev_t dev;
7254 	int rval;
7255 
7256 	dev = makedevice(ddi_driver_major(vdc->dip),
7257 	    VD_MAKE_DEV(vdc->instance, 0));
7258 
7259 	return (vd_process_ioctl(dev, cmd, (caddr_t)arg, FKIOCTL, &rval));
7260 }
7261 
7262 /*
7263  * Function:
7264  *	vd_process_ioctl()
7265  *
7266  * Description:
7267  *	This routine processes disk specific ioctl calls
7268  *
7269  * Arguments:
7270  *	dev	- the device number
7271  *	cmd	- the operation [dkio(4I)] to be processed
7272  *	arg	- pointer to user provided structure
7273  *		  (contains data to be set or reference parameter for get)
7274  *	mode	- bit flag, indicating open settings, 32/64 bit type, etc
7275  *	rvalp	- pointer to return value for calling process.
7276  *
7277  * Return Code:
7278  *	0
7279  *	EFAULT
7280  *	ENXIO
7281  *	EIO
7282  *	ENOTSUP
7283  */
7284 static int
7285 vd_process_ioctl(dev_t dev, int cmd, caddr_t arg, int mode, int *rvalp)
7286 {
7287 	int		instance = VDCUNIT(dev);
7288 	vdc_t		*vdc = NULL;
7289 	int		rv = -1;
7290 	int		idx = 0;		/* index into dk_ioctl[] */
7291 	size_t		len = 0;		/* #bytes to send to vds */
7292 	size_t		alloc_len = 0;		/* #bytes to allocate mem for */
7293 	caddr_t		mem_p = NULL;
7294 	size_t		nioctls = (sizeof (dk_ioctl)) / (sizeof (dk_ioctl[0]));
7295 	vdc_dk_ioctl_t	*iop;
7296 
7297 	vdc = ddi_get_soft_state(vdc_state, instance);
7298 	if (vdc == NULL) {
7299 		cmn_err(CE_NOTE, "![%d] Could not get soft state structure",
7300 		    instance);
7301 		return (ENXIO);
7302 	}
7303 
7304 	DMSG(vdc, 0, "[%d] Processing ioctl(%x) for dev %lx : model %x\n",
7305 	    instance, cmd, dev, ddi_model_convert_from(mode & FMODELS));
7306 
7307 	if (rvalp != NULL) {
7308 		/* the return value of the ioctl is 0 by default */
7309 		*rvalp = 0;
7310 	}
7311 
7312 	/*
7313 	 * Validate the ioctl operation to be performed.
7314 	 *
7315 	 * If we have looped through the array without finding a match then we
7316 	 * don't support this ioctl.
7317 	 */
7318 	for (idx = 0; idx < nioctls; idx++) {
7319 		if (cmd == dk_ioctl[idx].cmd)
7320 			break;
7321 	}
7322 
7323 	if (idx >= nioctls) {
7324 		DMSG(vdc, 0, "[%d] Unsupported ioctl (0x%x)\n",
7325 		    vdc->instance, cmd);
7326 		return (ENOTSUP);
7327 	}
7328 
7329 	iop = &(dk_ioctl[idx]);
7330 
7331 	if (cmd == DKIOCGETEFI || cmd == DKIOCSETEFI) {
7332 		/* size is not fixed for EFI ioctls, it depends on ioctl arg */
7333 		dk_efi_t	dk_efi;
7334 
7335 		rv = ddi_copyin(arg, &dk_efi, sizeof (dk_efi_t), mode);
7336 		if (rv != 0)
7337 			return (EFAULT);
7338 
7339 		len = sizeof (vd_efi_t) - 1 + dk_efi.dki_length;
7340 	} else {
7341 		len = iop->nbytes;
7342 	}
7343 
7344 	/* check if the ioctl is applicable */
7345 	switch (cmd) {
7346 	case CDROMREADOFFSET:
7347 	case DKIOCREMOVABLE:
7348 		return (ENOTTY);
7349 
7350 	case USCSICMD:
7351 	case MHIOCTKOWN:
7352 	case MHIOCSTATUS:
7353 	case MHIOCQRESERVE:
7354 	case MHIOCRELEASE:
7355 	case MHIOCGRP_INKEYS:
7356 	case MHIOCGRP_INRESV:
7357 	case MHIOCGRP_REGISTER:
7358 	case MHIOCGRP_RESERVE:
7359 	case MHIOCGRP_PREEMPTANDABORT:
7360 	case MHIOCGRP_REGISTERANDIGNOREKEY:
7361 	case MHIOCENFAILFAST:
7362 		if (vdc->cinfo == NULL)
7363 			return (ENXIO);
7364 		if (vdc->cinfo->dki_ctype != DKC_SCSI_CCS)
7365 			return (ENOTTY);
7366 		break;
7367 
7368 	case DIOCTL_RWCMD:
7369 		if (vdc->cinfo == NULL)
7370 			return (ENXIO);
7371 		if (vdc->cinfo->dki_ctype != DKC_DIRECT)
7372 			return (ENOTTY);
7373 		break;
7374 
7375 	case DKIOCINFO:
7376 		if (vdc->cinfo == NULL)
7377 			return (ENXIO);
7378 		break;
7379 
7380 	case DKIOCGMEDIAINFO:
7381 		if (vdc->minfo == NULL)
7382 			return (ENXIO);
7383 		if (vdc_check_capacity(vdc) != 0)
7384 			/* disk capacity is not available */
7385 			return (EIO);
7386 		break;
7387 	}
7388 
7389 	/*
7390 	 * Deal with ioctls which require a processing different than
7391 	 * converting ioctl arguments and sending a corresponding
7392 	 * VD operation.
7393 	 */
7394 	switch (cmd) {
7395 
7396 	case USCSICMD:
7397 	{
7398 		return (vdc_uscsi_cmd(vdc, arg, mode));
7399 	}
7400 
7401 	case MHIOCTKOWN:
7402 	{
7403 		mutex_enter(&vdc->ownership_lock);
7404 		/*
7405 		 * We have to set VDC_OWNERSHIP_WANTED now so that the ownership
7406 		 * can be flagged with VDC_OWNERSHIP_RESET if the LDC is reset
7407 		 * while we are processing the ioctl.
7408 		 */
7409 		vdc_ownership_update(vdc, VDC_OWNERSHIP_WANTED);
7410 
7411 		rv = vdc_access_set(vdc, VD_ACCESS_SET_EXCLUSIVE |
7412 		    VD_ACCESS_SET_PREEMPT | VD_ACCESS_SET_PRESERVE);
7413 		if (rv == 0) {
7414 			vdc_ownership_update(vdc, VDC_OWNERSHIP_WANTED |
7415 			    VDC_OWNERSHIP_GRANTED);
7416 		} else {
7417 			vdc_ownership_update(vdc, VDC_OWNERSHIP_NONE);
7418 		}
7419 		mutex_exit(&vdc->ownership_lock);
7420 		return (rv);
7421 	}
7422 
7423 	case MHIOCRELEASE:
7424 	{
7425 		mutex_enter(&vdc->ownership_lock);
7426 		rv = vdc_access_set(vdc, VD_ACCESS_SET_CLEAR);
7427 		if (rv == 0) {
7428 			vdc_ownership_update(vdc, VDC_OWNERSHIP_NONE);
7429 		}
7430 		mutex_exit(&vdc->ownership_lock);
7431 		return (rv);
7432 	}
7433 
7434 	case MHIOCSTATUS:
7435 	{
7436 		uint64_t status;
7437 
7438 		rv = vdc_access_get(vdc, &status);
7439 		if (rv == 0 && rvalp != NULL)
7440 			*rvalp = (status & VD_ACCESS_ALLOWED)? 0 : 1;
7441 		return (rv);
7442 	}
7443 
7444 	case MHIOCQRESERVE:
7445 	{
7446 		rv = vdc_access_set(vdc, VD_ACCESS_SET_EXCLUSIVE);
7447 		return (rv);
7448 	}
7449 
7450 	case MHIOCGRP_INKEYS:
7451 	{
7452 		return (vdc_mhd_inkeys(vdc, arg, mode));
7453 	}
7454 
7455 	case MHIOCGRP_INRESV:
7456 	{
7457 		return (vdc_mhd_inresv(vdc, arg, mode));
7458 	}
7459 
7460 	case MHIOCGRP_REGISTER:
7461 	{
7462 		return (vdc_mhd_register(vdc, arg, mode));
7463 	}
7464 
7465 	case MHIOCGRP_RESERVE:
7466 	{
7467 		return (vdc_mhd_reserve(vdc, arg, mode));
7468 	}
7469 
7470 	case MHIOCGRP_PREEMPTANDABORT:
7471 	{
7472 		return (vdc_mhd_preemptabort(vdc, arg, mode));
7473 	}
7474 
7475 	case MHIOCGRP_REGISTERANDIGNOREKEY:
7476 	{
7477 		return (vdc_mhd_registerignore(vdc, arg, mode));
7478 	}
7479 
7480 	case MHIOCENFAILFAST:
7481 	{
7482 		rv = vdc_failfast(vdc, arg, mode);
7483 		return (rv);
7484 	}
7485 
7486 	case DIOCTL_RWCMD:
7487 	{
7488 		return (vdc_dioctl_rwcmd(vdc, arg, mode));
7489 	}
7490 
7491 	case DKIOCGAPART:
7492 	{
7493 		return (vdc_dkio_gapart(vdc, arg, mode));
7494 	}
7495 
7496 	case DKIOCPARTITION:
7497 	{
7498 		return (vdc_dkio_partition(vdc, arg, mode));
7499 	}
7500 
7501 	case DKIOCINFO:
7502 	{
7503 		struct dk_cinfo	cinfo;
7504 
7505 		bcopy(vdc->cinfo, &cinfo, sizeof (struct dk_cinfo));
7506 		cinfo.dki_partition = VDCPART(dev);
7507 
7508 		rv = ddi_copyout(&cinfo, (void *)arg,
7509 		    sizeof (struct dk_cinfo), mode);
7510 		if (rv != 0)
7511 			return (EFAULT);
7512 
7513 		return (0);
7514 	}
7515 
7516 	case DKIOCGMEDIAINFO:
7517 	{
7518 		ASSERT(vdc->vdisk_size != 0);
7519 		ASSERT(vdc->minfo->dki_capacity != 0);
7520 		rv = ddi_copyout(vdc->minfo, (void *)arg,
7521 		    sizeof (struct dk_minfo), mode);
7522 		if (rv != 0)
7523 			return (EFAULT);
7524 
7525 		return (0);
7526 	}
7527 
7528 	case DKIOCFLUSHWRITECACHE:
7529 		{
7530 			struct dk_callback *dkc =
7531 			    (struct dk_callback *)(uintptr_t)arg;
7532 			vdc_dk_arg_t	*dkarg = NULL;
7533 
7534 			DMSG(vdc, 1, "[%d] Flush W$: mode %x\n",
7535 			    instance, mode);
7536 
7537 			/*
7538 			 * If arg is NULL, then there is no callback function
7539 			 * registered and the call operates synchronously; we
7540 			 * break and continue with the rest of the function and
7541 			 * wait for vds to return (i.e. after the request to
7542 			 * vds returns successfully, all writes completed prior
7543 			 * to the ioctl will have been flushed from the disk
7544 			 * write cache to persistent media.
7545 			 *
7546 			 * If a callback function is registered, we dispatch
7547 			 * the request on a task queue and return immediately.
7548 			 * The callback will deal with informing the calling
7549 			 * thread that the flush request is completed.
7550 			 */
7551 			if (dkc == NULL)
7552 				break;
7553 
7554 			/*
7555 			 * the asynchronous callback is only supported if
7556 			 * invoked from within the kernel
7557 			 */
7558 			if ((mode & FKIOCTL) == 0)
7559 				return (ENOTSUP);
7560 
7561 			dkarg = kmem_zalloc(sizeof (vdc_dk_arg_t), KM_SLEEP);
7562 
7563 			dkarg->mode = mode;
7564 			dkarg->dev = dev;
7565 			bcopy(dkc, &dkarg->dkc, sizeof (*dkc));
7566 
7567 			mutex_enter(&vdc->lock);
7568 			vdc->dkio_flush_pending++;
7569 			dkarg->vdc = vdc;
7570 			mutex_exit(&vdc->lock);
7571 
7572 			/* put the request on a task queue */
7573 			rv = taskq_dispatch(system_taskq, vdc_dkio_flush_cb,
7574 			    (void *)dkarg, DDI_SLEEP);
7575 			if (rv == TASKQID_INVALID) {
7576 				/* clean up if dispatch fails */
7577 				mutex_enter(&vdc->lock);
7578 				vdc->dkio_flush_pending--;
7579 				mutex_exit(&vdc->lock);
7580 				kmem_free(dkarg, sizeof (vdc_dk_arg_t));
7581 				return (ENOMEM);
7582 			}
7583 
7584 			return (0);
7585 		}
7586 	}
7587 
7588 	/* catch programming error in vdc - should be a VD_OP_XXX ioctl */
7589 	ASSERT(iop->op != 0);
7590 
7591 	/* check if the vDisk server handles the operation for this vDisk */
7592 	if (VD_OP_SUPPORTED(vdc->operations, iop->op) == B_FALSE) {
7593 		DMSG(vdc, 0, "[%d] Unsupported VD_OP operation (0x%x)\n",
7594 		    vdc->instance, iop->op);
7595 		return (ENOTSUP);
7596 	}
7597 
7598 	/* LDC requires that the memory being mapped is 8-byte aligned */
7599 	alloc_len = P2ROUNDUP(len, sizeof (uint64_t));
7600 	DMSG(vdc, 1, "[%d] struct size %ld alloc %ld\n",
7601 	    instance, len, alloc_len);
7602 
7603 	if (alloc_len > 0)
7604 		mem_p = kmem_zalloc(alloc_len, KM_SLEEP);
7605 
7606 	/*
7607 	 * Call the conversion function for this ioctl which, if necessary,
7608 	 * converts from the Solaris format to the format ARC'ed
7609 	 * as part of the vDisk protocol (FWARC 2006/195)
7610 	 */
7611 	ASSERT(iop->convert != NULL);
7612 	rv = (iop->convert)(vdc, arg, mem_p, mode, VD_COPYIN);
7613 	if (rv != 0) {
7614 		DMSG(vdc, 0, "[%d] convert func returned %d for ioctl 0x%x\n",
7615 		    instance, rv, cmd);
7616 		if (mem_p != NULL)
7617 			kmem_free(mem_p, alloc_len);
7618 		return (rv);
7619 	}
7620 
7621 	/*
7622 	 * send request to vds to service the ioctl.
7623 	 */
7624 	rv = vdc_do_sync_op(vdc, iop->op, mem_p, alloc_len,
7625 	    VDCPART(dev), 0, VIO_both_dir, B_TRUE);
7626 
7627 	if (rv != 0) {
7628 		/*
7629 		 * This is not necessarily an error. The ioctl could
7630 		 * be returning a value such as ENOTTY to indicate
7631 		 * that the ioctl is not applicable.
7632 		 */
7633 		DMSG(vdc, 0, "[%d] vds returned %d for ioctl 0x%x\n",
7634 		    instance, rv, cmd);
7635 		if (mem_p != NULL)
7636 			kmem_free(mem_p, alloc_len);
7637 
7638 		return (rv);
7639 	}
7640 
7641 	/*
7642 	 * Call the conversion function (if it exists) for this ioctl
7643 	 * which converts from the format ARC'ed as part of the vDisk
7644 	 * protocol (FWARC 2006/195) back to a format understood by
7645 	 * the rest of Solaris.
7646 	 */
7647 	rv = (iop->convert)(vdc, mem_p, arg, mode, VD_COPYOUT);
7648 	if (rv != 0) {
7649 		DMSG(vdc, 0, "[%d] convert func returned %d for ioctl 0x%x\n",
7650 		    instance, rv, cmd);
7651 		if (mem_p != NULL)
7652 			kmem_free(mem_p, alloc_len);
7653 		return (rv);
7654 	}
7655 
7656 	if (mem_p != NULL)
7657 		kmem_free(mem_p, alloc_len);
7658 
7659 	return (rv);
7660 }
7661 
7662 /*
7663  * Function:
7664  *
7665  * Description:
7666  *	This is an empty conversion function used by ioctl calls which
7667  *	do not need to convert the data being passed in/out to userland
7668  */
7669 static int
7670 vdc_null_copy_func(vdc_t *vdc, void *from, void *to, int mode, int dir)
7671 {
7672 	_NOTE(ARGUNUSED(vdc))
7673 	_NOTE(ARGUNUSED(from))
7674 	_NOTE(ARGUNUSED(to))
7675 	_NOTE(ARGUNUSED(mode))
7676 	_NOTE(ARGUNUSED(dir))
7677 
7678 	return (0);
7679 }
7680 
7681 static int
7682 vdc_get_wce_convert(vdc_t *vdc, void *from, void *to,
7683     int mode, int dir)
7684 {
7685 	_NOTE(ARGUNUSED(vdc))
7686 
7687 	if (dir == VD_COPYIN)
7688 		return (0);		/* nothing to do */
7689 
7690 	if (ddi_copyout(from, to, sizeof (int), mode) != 0)
7691 		return (EFAULT);
7692 
7693 	return (0);
7694 }
7695 
7696 static int
7697 vdc_set_wce_convert(vdc_t *vdc, void *from, void *to,
7698     int mode, int dir)
7699 {
7700 	_NOTE(ARGUNUSED(vdc))
7701 
7702 	if (dir == VD_COPYOUT)
7703 		return (0);		/* nothing to do */
7704 
7705 	if (ddi_copyin(from, to, sizeof (int), mode) != 0)
7706 		return (EFAULT);
7707 
7708 	return (0);
7709 }
7710 
7711 /*
7712  * Function:
7713  *	vdc_get_vtoc_convert()
7714  *
7715  * Description:
7716  *	This routine performs the necessary convertions from the DKIOCGVTOC
7717  *	Solaris structure to the format defined in FWARC 2006/195.
7718  *
7719  *	In the struct vtoc definition, the timestamp field is marked as not
7720  *	supported so it is not part of vDisk protocol (FWARC 2006/195).
7721  *	However SVM uses that field to check it can write into the VTOC,
7722  *	so we fake up the info of that field.
7723  *
7724  * Arguments:
7725  *	vdc	- the vDisk client
7726  *	from	- the buffer containing the data to be copied from
7727  *	to	- the buffer to be copied to
7728  *	mode	- flags passed to ioctl() call
7729  *	dir	- the "direction" of the copy - VD_COPYIN or VD_COPYOUT
7730  *
7731  * Return Code:
7732  *	0	- Success
7733  *	ENXIO	- incorrect buffer passed in.
7734  *	EFAULT	- ddi_copyout routine encountered an error.
7735  */
7736 static int
7737 vdc_get_vtoc_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7738 {
7739 	int		i;
7740 	struct vtoc	vtoc;
7741 	struct vtoc32	vtoc32;
7742 	struct extvtoc	evtoc;
7743 	int		rv;
7744 
7745 	if (dir != VD_COPYOUT)
7746 		return (0);	/* nothing to do */
7747 
7748 	if ((from == NULL) || (to == NULL))
7749 		return (ENXIO);
7750 
7751 	if (vdc->vdisk_size > VD_OLDVTOC_LIMIT)
7752 		return (EOVERFLOW);
7753 
7754 	VD_VTOC2VTOC((vd_vtoc_t *)from, &evtoc);
7755 
7756 	/* fake the VTOC timestamp field */
7757 	for (i = 0; i < V_NUMPAR; i++) {
7758 		evtoc.timestamp[i] = vdc->vtoc->timestamp[i];
7759 	}
7760 
7761 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
7762 		/* LINTED E_ASSIGN_NARROW_CONV */
7763 		extvtoctovtoc32(evtoc, vtoc32);
7764 		rv = ddi_copyout(&vtoc32, to, sizeof (vtoc32), mode);
7765 		if (rv != 0)
7766 			rv = EFAULT;
7767 	} else {
7768 		extvtoctovtoc(evtoc, vtoc);
7769 		rv = ddi_copyout(&vtoc, to, sizeof (vtoc), mode);
7770 		if (rv != 0)
7771 			rv = EFAULT;
7772 	}
7773 
7774 	return (rv);
7775 }
7776 
7777 /*
7778  * Function:
7779  *	vdc_set_vtoc_convert()
7780  *
7781  * Description:
7782  *	This routine performs the necessary convertions from the DKIOCSVTOC
7783  *	Solaris structure to the format defined in FWARC 2006/195.
7784  *
7785  * Arguments:
7786  *	vdc	- the vDisk client
7787  *	from	- Buffer with data
7788  *	to	- Buffer where data is to be copied to
7789  *	mode	- flags passed to ioctl
7790  *	dir	- direction of copy (in or out)
7791  *
7792  * Return Code:
7793  *	0	- Success
7794  *	ENXIO	- Invalid buffer passed in
7795  *	EFAULT	- ddi_copyin of data failed
7796  */
7797 static int
7798 vdc_set_vtoc_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7799 {
7800 	void		*uvtoc;
7801 	struct vtoc	vtoc;
7802 	struct vtoc32	vtoc32;
7803 	struct extvtoc	evtoc;
7804 	int		i, rv;
7805 
7806 	if ((from == NULL) || (to == NULL))
7807 		return (ENXIO);
7808 
7809 	if (vdc->vdisk_size > VD_OLDVTOC_LIMIT)
7810 		return (EOVERFLOW);
7811 
7812 	uvtoc = (dir == VD_COPYIN)? from : to;
7813 
7814 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
7815 		rv = ddi_copyin(uvtoc, &vtoc32, sizeof (vtoc32), mode);
7816 		if (rv != 0)
7817 			return (EFAULT);
7818 		vtoc32toextvtoc(vtoc32, evtoc);
7819 	} else {
7820 		rv = ddi_copyin(uvtoc, &vtoc, sizeof (vtoc), mode);
7821 		if (rv != 0)
7822 			return (EFAULT);
7823 		vtoctoextvtoc(vtoc, evtoc);
7824 	}
7825 
7826 	if (dir == VD_COPYOUT) {
7827 		/*
7828 		 * The disk label may have changed. Revalidate the disk
7829 		 * geometry. This will also update the device nodes.
7830 		 */
7831 		vdc_validate(vdc);
7832 
7833 		/*
7834 		 * We also need to keep track of the timestamp fields.
7835 		 */
7836 		for (i = 0; i < V_NUMPAR; i++) {
7837 			vdc->vtoc->timestamp[i] = evtoc.timestamp[i];
7838 		}
7839 
7840 	} else {
7841 		VTOC2VD_VTOC(&evtoc, (vd_vtoc_t *)to);
7842 	}
7843 
7844 	return (0);
7845 }
7846 
7847 static int
7848 vdc_get_extvtoc_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7849 {
7850 	int		i, rv;
7851 	struct extvtoc	evtoc;
7852 
7853 	if (dir != VD_COPYOUT)
7854 		return (0);	/* nothing to do */
7855 
7856 	if ((from == NULL) || (to == NULL))
7857 		return (ENXIO);
7858 
7859 	VD_VTOC2VTOC((vd_vtoc_t *)from, &evtoc);
7860 
7861 	/* fake the VTOC timestamp field */
7862 	for (i = 0; i < V_NUMPAR; i++) {
7863 		evtoc.timestamp[i] = vdc->vtoc->timestamp[i];
7864 	}
7865 
7866 	rv = ddi_copyout(&evtoc, to, sizeof (struct extvtoc), mode);
7867 	if (rv != 0)
7868 		rv = EFAULT;
7869 
7870 	return (rv);
7871 }
7872 
7873 static int
7874 vdc_set_extvtoc_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7875 {
7876 	void		*uvtoc;
7877 	struct extvtoc	evtoc;
7878 	int		i, rv;
7879 
7880 	if ((from == NULL) || (to == NULL))
7881 		return (ENXIO);
7882 
7883 	uvtoc = (dir == VD_COPYIN)? from : to;
7884 
7885 	rv = ddi_copyin(uvtoc, &evtoc, sizeof (struct extvtoc), mode);
7886 	if (rv != 0)
7887 		return (EFAULT);
7888 
7889 	if (dir == VD_COPYOUT) {
7890 		/*
7891 		 * The disk label may have changed. Revalidate the disk
7892 		 * geometry. This will also update the device nodes.
7893 		 */
7894 		vdc_validate(vdc);
7895 
7896 		/*
7897 		 * We also need to keep track of the timestamp fields.
7898 		 */
7899 		for (i = 0; i < V_NUMPAR; i++) {
7900 			vdc->vtoc->timestamp[i] = evtoc.timestamp[i];
7901 		}
7902 
7903 	} else {
7904 		VTOC2VD_VTOC(&evtoc, (vd_vtoc_t *)to);
7905 	}
7906 
7907 	return (0);
7908 }
7909 
7910 /*
7911  * Function:
7912  *	vdc_get_geom_convert()
7913  *
7914  * Description:
7915  *	This routine performs the necessary convertions from the DKIOCGGEOM,
7916  *	DKIOCG_PHYSGEOM and DKIOG_VIRTGEOM Solaris structures to the format
7917  *	defined in FWARC 2006/195
7918  *
7919  * Arguments:
7920  *	vdc	- the vDisk client
7921  *	from	- Buffer with data
7922  *	to	- Buffer where data is to be copied to
7923  *	mode	- flags passed to ioctl
7924  *	dir	- direction of copy (in or out)
7925  *
7926  * Return Code:
7927  *	0	- Success
7928  *	ENXIO	- Invalid buffer passed in
7929  *	EFAULT	- ddi_copyout of data failed
7930  */
7931 static int
7932 vdc_get_geom_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7933 {
7934 	_NOTE(ARGUNUSED(vdc))
7935 
7936 	struct dk_geom	geom;
7937 	int	copy_len = sizeof (struct dk_geom);
7938 	int	rv = 0;
7939 
7940 	if (dir != VD_COPYOUT)
7941 		return (0);	/* nothing to do */
7942 
7943 	if ((from == NULL) || (to == NULL))
7944 		return (ENXIO);
7945 
7946 	VD_GEOM2DK_GEOM((vd_geom_t *)from, &geom);
7947 	rv = ddi_copyout(&geom, to, copy_len, mode);
7948 	if (rv != 0)
7949 		rv = EFAULT;
7950 
7951 	return (rv);
7952 }
7953 
7954 /*
7955  * Function:
7956  *	vdc_set_geom_convert()
7957  *
7958  * Description:
7959  *	This routine performs the necessary convertions from the DKIOCSGEOM
7960  *	Solaris structure to the format defined in FWARC 2006/195.
7961  *
7962  * Arguments:
7963  *	vdc	- the vDisk client
7964  *	from	- Buffer with data
7965  *	to	- Buffer where data is to be copied to
7966  *	mode	- flags passed to ioctl
7967  *	dir	- direction of copy (in or out)
7968  *
7969  * Return Code:
7970  *	0	- Success
7971  *	ENXIO	- Invalid buffer passed in
7972  *	EFAULT	- ddi_copyin of data failed
7973  */
7974 static int
7975 vdc_set_geom_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
7976 {
7977 	_NOTE(ARGUNUSED(vdc))
7978 
7979 	vd_geom_t	vdgeom;
7980 	void		*tmp_mem = NULL;
7981 	int		copy_len = sizeof (struct dk_geom);
7982 	int		rv = 0;
7983 
7984 	if (dir != VD_COPYIN)
7985 		return (0);	/* nothing to do */
7986 
7987 	if ((from == NULL) || (to == NULL))
7988 		return (ENXIO);
7989 
7990 	tmp_mem = kmem_alloc(copy_len, KM_SLEEP);
7991 
7992 	rv = ddi_copyin(from, tmp_mem, copy_len, mode);
7993 	if (rv != 0) {
7994 		kmem_free(tmp_mem, copy_len);
7995 		return (EFAULT);
7996 	}
7997 	DK_GEOM2VD_GEOM((struct dk_geom *)tmp_mem, &vdgeom);
7998 	bcopy(&vdgeom, to, sizeof (vdgeom));
7999 	kmem_free(tmp_mem, copy_len);
8000 
8001 	return (0);
8002 }
8003 
8004 static int
8005 vdc_get_efi_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
8006 {
8007 	_NOTE(ARGUNUSED(vdc))
8008 
8009 	vd_efi_t	*vd_efi;
8010 	dk_efi_t	dk_efi;
8011 	int		rv = 0;
8012 	void		*uaddr;
8013 
8014 	if ((from == NULL) || (to == NULL))
8015 		return (ENXIO);
8016 
8017 	if (dir == VD_COPYIN) {
8018 
8019 		vd_efi = (vd_efi_t *)to;
8020 
8021 		rv = ddi_copyin(from, &dk_efi, sizeof (dk_efi_t), mode);
8022 		if (rv != 0)
8023 			return (EFAULT);
8024 
8025 		vd_efi->lba = dk_efi.dki_lba;
8026 		vd_efi->length = dk_efi.dki_length;
8027 		bzero(vd_efi->data, vd_efi->length);
8028 
8029 	} else {
8030 
8031 		rv = ddi_copyin(to, &dk_efi, sizeof (dk_efi_t), mode);
8032 		if (rv != 0)
8033 			return (EFAULT);
8034 
8035 		uaddr = dk_efi.dki_data;
8036 
8037 		dk_efi.dki_data = kmem_alloc(dk_efi.dki_length, KM_SLEEP);
8038 
8039 		VD_EFI2DK_EFI((vd_efi_t *)from, &dk_efi);
8040 
8041 		rv = ddi_copyout(dk_efi.dki_data, uaddr, dk_efi.dki_length,
8042 		    mode);
8043 		if (rv != 0)
8044 			return (EFAULT);
8045 
8046 		kmem_free(dk_efi.dki_data, dk_efi.dki_length);
8047 	}
8048 
8049 	return (0);
8050 }
8051 
8052 static int
8053 vdc_set_efi_convert(vdc_t *vdc, void *from, void *to, int mode, int dir)
8054 {
8055 	_NOTE(ARGUNUSED(vdc))
8056 
8057 	dk_efi_t	dk_efi;
8058 	void		*uaddr;
8059 
8060 	if (dir == VD_COPYOUT) {
8061 		/*
8062 		 * The disk label may have changed. Revalidate the disk
8063 		 * geometry. This will also update the device nodes.
8064 		 */
8065 		vdc_validate(vdc);
8066 		return (0);
8067 	}
8068 
8069 	if ((from == NULL) || (to == NULL))
8070 		return (ENXIO);
8071 
8072 	if (ddi_copyin(from, &dk_efi, sizeof (dk_efi_t), mode) != 0)
8073 		return (EFAULT);
8074 
8075 	uaddr = dk_efi.dki_data;
8076 
8077 	dk_efi.dki_data = kmem_alloc(dk_efi.dki_length, KM_SLEEP);
8078 
8079 	if (ddi_copyin(uaddr, dk_efi.dki_data, dk_efi.dki_length, mode) != 0)
8080 		return (EFAULT);
8081 
8082 	DK_EFI2VD_EFI(&dk_efi, (vd_efi_t *)to);
8083 
8084 	kmem_free(dk_efi.dki_data, dk_efi.dki_length);
8085 
8086 	return (0);
8087 }
8088 
8089 
8090 /* -------------------------------------------------------------------------- */
8091 
8092 /*
8093  * Function:
8094  *	vdc_create_fake_geometry()
8095  *
8096  * Description:
8097  *	This routine fakes up the disk info needed for some DKIO ioctls such
8098  *	as DKIOCINFO and DKIOCGMEDIAINFO [just like lofi(4D) and ramdisk(4D) do]
8099  *
8100  *	Note: This function must not be called until the vDisk attributes have
8101  *	been exchanged as part of the handshake with the vDisk server.
8102  *
8103  * Arguments:
8104  *	vdc	- soft state pointer for this instance of the device driver.
8105  *
8106  * Return Code:
8107  *	none.
8108  */
8109 static void
8110 vdc_create_fake_geometry(vdc_t *vdc)
8111 {
8112 	ASSERT(vdc != NULL);
8113 	ASSERT(vdc->max_xfer_sz != 0);
8114 
8115 	/*
8116 	 * DKIOCINFO support
8117 	 */
8118 	if (vdc->cinfo == NULL)
8119 		vdc->cinfo = kmem_zalloc(sizeof (struct dk_cinfo), KM_SLEEP);
8120 
8121 	(void) strcpy(vdc->cinfo->dki_cname, VDC_DRIVER_NAME);
8122 	(void) strcpy(vdc->cinfo->dki_dname, VDC_DRIVER_NAME);
8123 	/* max_xfer_sz is #blocks so we don't need to divide by vdisk_bsize */
8124 	vdc->cinfo->dki_maxtransfer = vdc->max_xfer_sz;
8125 
8126 	/*
8127 	 * We set the controller type to DKC_SCSI_CCS only if the VD_OP_SCSICMD
8128 	 * operation is supported, otherwise the controller type is DKC_DIRECT.
8129 	 * Version 1.0 does not support the VD_OP_SCSICMD operation, so the
8130 	 * controller type is always DKC_DIRECT in that case.
8131 	 *
8132 	 * If the virtual disk is backed by a physical CD/DVD device or
8133 	 * an ISO image, modify the controller type to indicate this
8134 	 */
8135 	switch (vdc->vdisk_media) {
8136 	case VD_MEDIA_CD:
8137 	case VD_MEDIA_DVD:
8138 		vdc->cinfo->dki_ctype = DKC_CDROM;
8139 		break;
8140 	case VD_MEDIA_FIXED:
8141 		if (VD_OP_SUPPORTED(vdc->operations, VD_OP_SCSICMD))
8142 			vdc->cinfo->dki_ctype = DKC_SCSI_CCS;
8143 		else
8144 			vdc->cinfo->dki_ctype = DKC_DIRECT;
8145 		break;
8146 	default:
8147 		/* in the case of v1.0 we default to a fixed disk */
8148 		vdc->cinfo->dki_ctype = DKC_DIRECT;
8149 		break;
8150 	}
8151 	vdc->cinfo->dki_flags = DKI_FMTVOL;
8152 	vdc->cinfo->dki_cnum = 0;
8153 	vdc->cinfo->dki_addr = 0;
8154 	vdc->cinfo->dki_space = 0;
8155 	vdc->cinfo->dki_prio = 0;
8156 	vdc->cinfo->dki_vec = 0;
8157 	vdc->cinfo->dki_unit = vdc->instance;
8158 	vdc->cinfo->dki_slave = 0;
8159 	/*
8160 	 * The partition number will be created on the fly depending on the
8161 	 * actual slice (i.e. minor node) that is used to request the data.
8162 	 */
8163 	vdc->cinfo->dki_partition = 0;
8164 
8165 	/*
8166 	 * DKIOCGMEDIAINFO support
8167 	 */
8168 	if (vdc->minfo == NULL)
8169 		vdc->minfo = kmem_zalloc(sizeof (struct dk_minfo), KM_SLEEP);
8170 
8171 	if (vio_ver_is_supported(vdc->ver, 1, 1)) {
8172 		vdc->minfo->dki_media_type =
8173 		    VD_MEDIATYPE2DK_MEDIATYPE(vdc->vdisk_media);
8174 	} else {
8175 		vdc->minfo->dki_media_type = DK_FIXED_DISK;
8176 	}
8177 
8178 	vdc->minfo->dki_capacity = vdc->vdisk_size;
8179 	vdc->minfo->dki_lbsize = vdc->vdisk_bsize;
8180 }
8181 
8182 static ushort_t
8183 vdc_lbl2cksum(struct dk_label *label)
8184 {
8185 	int	count;
8186 	ushort_t sum, *sp;
8187 
8188 	count =	(sizeof (struct dk_label)) / (sizeof (short)) - 1;
8189 	sp = (ushort_t *)label;
8190 	sum = 0;
8191 	while (count--) {
8192 		sum ^= *sp++;
8193 	}
8194 
8195 	return (sum);
8196 }
8197 
8198 static void
8199 vdc_update_size(vdc_t *vdc, size_t dsk_size, size_t blk_size, size_t xfr_size)
8200 {
8201 	vd_err_stats_t  *stp;
8202 
8203 	ASSERT(MUTEX_HELD(&vdc->lock));
8204 	ASSERT(xfr_size != 0);
8205 
8206 	/*
8207 	 * If the disk size is unknown or sizes are unchanged then don't
8208 	 * update anything.
8209 	 */
8210 	if (dsk_size == VD_SIZE_UNKNOWN || dsk_size == 0 ||
8211 	    (blk_size == vdc->vdisk_bsize && dsk_size == vdc->vdisk_size &&
8212 	    xfr_size == vdc->max_xfer_sz))
8213 		return;
8214 
8215 	/*
8216 	 * We don't know at compile time what the vDisk server will think
8217 	 * are good values but we apply a large (arbitrary) upper bound to
8218 	 * prevent memory exhaustion in vdc if it was allocating a DRing
8219 	 * based of huge values sent by the server. We probably will never
8220 	 * exceed this except if the message was garbage.
8221 	 */
8222 	if ((xfr_size * blk_size) > (PAGESIZE * DEV_BSIZE)) {
8223 		DMSG(vdc, 0, "[%d] vds block transfer size too big;"
8224 		    " using max supported by vdc", vdc->instance);
8225 		xfr_size = maxphys / blk_size;
8226 	}
8227 
8228 	vdc->max_xfer_sz = xfr_size;
8229 	vdc->vdisk_bsize = blk_size;
8230 	vdc->vdisk_size = dsk_size;
8231 
8232 	stp = (vd_err_stats_t *)vdc->err_stats->ks_data;
8233 	stp->vd_capacity.value.ui64 = dsk_size * blk_size;
8234 
8235 	vdc->minfo->dki_capacity = dsk_size;
8236 	vdc->minfo->dki_lbsize = (uint_t)blk_size;
8237 }
8238 
8239 /*
8240  * Update information about the VIO block size. The VIO block size is the
8241  * same as the vdisk block size which is stored in vdc->vdisk_bsize so we
8242  * do not store that information again.
8243  *
8244  * However, buf structures will always use a logical block size of 512 bytes
8245  * (DEV_BSIZE) and we will need to convert logical block numbers to VIO block
8246  * numbers for each read or write operation using vdc_strategy(). To speed up
8247  * this conversion, we expect the VIO block size to be a power of 2 and a
8248  * multiple 512 bytes (DEV_BSIZE), and we cache some useful information.
8249  *
8250  * The function return EINVAL if the new VIO block size (blk_size) is not a
8251  * power of 2 or not a multiple of 512 bytes, otherwise it returns 0.
8252  */
8253 static int
8254 vdc_update_vio_bsize(vdc_t *vdc, uint32_t blk_size)
8255 {
8256 	uint32_t ratio, n;
8257 	int nshift = 0;
8258 
8259 	vdc->vio_bmask = 0;
8260 	vdc->vio_bshift = 0;
8261 
8262 	ASSERT(blk_size > 0);
8263 
8264 	if ((blk_size % DEV_BSIZE) != 0)
8265 		return (EINVAL);
8266 
8267 	ratio = blk_size / DEV_BSIZE;
8268 
8269 	for (n = ratio; n > 1; n >>= 1) {
8270 		if ((n & 0x1) != 0) {
8271 			/* blk_size is not a power of 2 */
8272 			return (EINVAL);
8273 		}
8274 		nshift++;
8275 	}
8276 
8277 	vdc->vio_bshift = nshift;
8278 	vdc->vio_bmask = ratio - 1;
8279 
8280 	return (0);
8281 }
8282 
8283 /*
8284  * Function:
8285  *	vdc_validate_geometry
8286  *
8287  * Description:
8288  *	This routine discovers the label and geometry of the disk. It stores
8289  *	the disk label and related information in the vdc structure. If it
8290  *	fails to validate the geometry or to discover the disk label then
8291  *	the label is marked as unknown (VD_DISK_LABEL_UNK).
8292  *
8293  * Arguments:
8294  *	vdc	- soft state pointer for this instance of the device driver.
8295  *
8296  * Return Code:
8297  *	0	- success.
8298  *	EINVAL	- unknown disk label.
8299  *	ENOTSUP	- geometry not applicable (EFI label).
8300  *	EIO	- error accessing the disk.
8301  */
8302 static int
8303 vdc_validate_geometry(vdc_t *vdc)
8304 {
8305 	dev_t	dev;
8306 	int	rv, rval;
8307 	struct dk_label *label;
8308 	struct dk_geom geom;
8309 	struct extvtoc vtoc;
8310 	efi_gpt_t *gpt;
8311 	efi_gpe_t *gpe;
8312 	vd_efi_dev_t edev;
8313 
8314 	ASSERT(vdc != NULL);
8315 	ASSERT(vdc->vtoc != NULL && vdc->geom != NULL);
8316 	ASSERT(MUTEX_HELD(&vdc->lock));
8317 
8318 	mutex_exit(&vdc->lock);
8319 	/*
8320 	 * Check the disk capacity in case it has changed. If that fails then
8321 	 * we proceed and we will be using the disk size we currently have.
8322 	 */
8323 	(void) vdc_check_capacity(vdc);
8324 	dev = makedevice(ddi_driver_major(vdc->dip),
8325 	    VD_MAKE_DEV(vdc->instance, 0));
8326 
8327 	rv = vd_process_ioctl(dev, DKIOCGGEOM, (caddr_t)&geom, FKIOCTL, &rval);
8328 	if (rv == 0)
8329 		rv = vd_process_ioctl(dev, DKIOCGEXTVTOC, (caddr_t)&vtoc,
8330 		    FKIOCTL, &rval);
8331 
8332 	if (rv == ENOTSUP) {
8333 		/*
8334 		 * If the device does not support VTOC then we try
8335 		 * to read an EFI label.
8336 		 *
8337 		 * We need to know the block size and the disk size to
8338 		 * be able to read an EFI label.
8339 		 */
8340 		if (vdc->vdisk_size == 0) {
8341 			mutex_enter(&vdc->lock);
8342 			vdc_store_label_unk(vdc);
8343 			return (EIO);
8344 		}
8345 
8346 		VDC_EFI_DEV_SET(edev, vdc, vd_process_efi_ioctl);
8347 
8348 		rv = vd_efi_alloc_and_read(&edev, &gpt, &gpe);
8349 
8350 		if (rv) {
8351 			DMSG(vdc, 0, "[%d] Failed to get EFI (err=%d)",
8352 			    vdc->instance, rv);
8353 			mutex_enter(&vdc->lock);
8354 			vdc_store_label_unk(vdc);
8355 			return (EIO);
8356 		}
8357 
8358 		mutex_enter(&vdc->lock);
8359 		vdc_store_label_efi(vdc, gpt, gpe);
8360 		vd_efi_free(&edev, gpt, gpe);
8361 		return (ENOTSUP);
8362 	}
8363 
8364 	if (rv != 0) {
8365 		DMSG(vdc, 0, "[%d] Failed to get VTOC (err=%d)",
8366 		    vdc->instance, rv);
8367 		mutex_enter(&vdc->lock);
8368 		vdc_store_label_unk(vdc);
8369 		if (rv != EINVAL)
8370 			rv = EIO;
8371 		return (rv);
8372 	}
8373 
8374 	/* check that geometry and vtoc are valid */
8375 	if (geom.dkg_nhead == 0 || geom.dkg_nsect == 0 ||
8376 	    vtoc.v_sanity != VTOC_SANE) {
8377 		mutex_enter(&vdc->lock);
8378 		vdc_store_label_unk(vdc);
8379 		return (EINVAL);
8380 	}
8381 
8382 	/*
8383 	 * We have a disk and a valid VTOC. However this does not mean
8384 	 * that the disk currently have a VTOC label. The returned VTOC may
8385 	 * be a default VTOC to be used for configuring the disk (this is
8386 	 * what is done for disk image). So we read the label from the
8387 	 * beginning of the disk to ensure we really have a VTOC label.
8388 	 *
8389 	 * FUTURE: This could be the default way for reading the VTOC
8390 	 * from the disk as opposed to sending the VD_OP_GET_VTOC
8391 	 * to the server. This will be the default if vdc is implemented
8392 	 * ontop of cmlb.
8393 	 */
8394 
8395 	/*
8396 	 * Single slice disk does not support read using an absolute disk
8397 	 * offset so we just rely on the DKIOCGVTOC ioctl in that case.
8398 	 */
8399 	if (vdc->vdisk_type == VD_DISK_TYPE_SLICE) {
8400 		mutex_enter(&vdc->lock);
8401 		if (vtoc.v_nparts != 1) {
8402 			vdc_store_label_unk(vdc);
8403 			return (EINVAL);
8404 		}
8405 		vdc_store_label_vtoc(vdc, &geom, &vtoc);
8406 		return (0);
8407 	}
8408 
8409 	if (vtoc.v_nparts != V_NUMPAR) {
8410 		mutex_enter(&vdc->lock);
8411 		vdc_store_label_unk(vdc);
8412 		return (EINVAL);
8413 	}
8414 
8415 	/*
8416 	 * Most CD/DVDs do not have a disk label and the label is
8417 	 * generated by the disk driver.  So the on-disk label check
8418 	 * below may fail and we return now to avoid this problem.
8419 	 */
8420 	if (vdc->vdisk_media == VD_MEDIA_CD ||
8421 	    vdc->vdisk_media == VD_MEDIA_DVD) {
8422 		mutex_enter(&vdc->lock);
8423 		vdc_store_label_vtoc(vdc, &geom, &vtoc);
8424 		return (0);
8425 	}
8426 
8427 	/*
8428 	 * Read disk label from start of disk
8429 	 */
8430 	label = kmem_alloc(vdc->vdisk_bsize, KM_SLEEP);
8431 
8432 	rv = vdc_do_op(vdc, VD_OP_BREAD, (caddr_t)label, vdc->vdisk_bsize,
8433 	    VD_SLICE_NONE, 0, NULL, VIO_read_dir, VDC_OP_NORMAL);
8434 
8435 	if (rv != 0 || label->dkl_magic != DKL_MAGIC ||
8436 	    label->dkl_cksum != vdc_lbl2cksum(label)) {
8437 		DMSG(vdc, 1, "[%d] Got VTOC with invalid label\n",
8438 		    vdc->instance);
8439 		kmem_free(label, vdc->vdisk_bsize);
8440 		mutex_enter(&vdc->lock);
8441 		vdc_store_label_unk(vdc);
8442 		return (EINVAL);
8443 	}
8444 
8445 	kmem_free(label, vdc->vdisk_bsize);
8446 	mutex_enter(&vdc->lock);
8447 	vdc_store_label_vtoc(vdc, &geom, &vtoc);
8448 	return (0);
8449 }
8450 
8451 /*
8452  * Function:
8453  *	vdc_validate
8454  *
8455  * Description:
8456  *	This routine discovers the label of the disk and create the
8457  *	appropriate device nodes if the label has changed.
8458  *
8459  * Arguments:
8460  *	vdc	- soft state pointer for this instance of the device driver.
8461  *
8462  * Return Code:
8463  *	none.
8464  */
8465 static void
8466 vdc_validate(vdc_t *vdc)
8467 {
8468 	vd_disk_label_t old_label;
8469 	vd_slice_t old_slice[V_NUMPAR];
8470 	int rv;
8471 
8472 	ASSERT(!MUTEX_HELD(&vdc->lock));
8473 
8474 	mutex_enter(&vdc->lock);
8475 
8476 	/* save the current label and vtoc */
8477 	old_label = vdc->vdisk_label;
8478 	bcopy(vdc->slice, &old_slice, sizeof (vd_slice_t) * V_NUMPAR);
8479 
8480 	/* check the geometry */
8481 	(void) vdc_validate_geometry(vdc);
8482 
8483 	/* if the disk label has changed, update device nodes */
8484 	if (vdc->vdisk_type == VD_DISK_TYPE_DISK &&
8485 	    vdc->vdisk_label != old_label) {
8486 
8487 		if (vdc->vdisk_label == VD_DISK_LABEL_EFI)
8488 			rv = vdc_create_device_nodes_efi(vdc);
8489 		else
8490 			rv = vdc_create_device_nodes_vtoc(vdc);
8491 
8492 		if (rv != 0) {
8493 			DMSG(vdc, 0, "![%d] Failed to update device nodes",
8494 			    vdc->instance);
8495 		}
8496 	}
8497 
8498 	mutex_exit(&vdc->lock);
8499 }
8500 
8501 static void
8502 vdc_validate_task(void *arg)
8503 {
8504 	vdc_t *vdc = (vdc_t *)arg;
8505 
8506 	vdc_validate(vdc);
8507 
8508 	mutex_enter(&vdc->lock);
8509 	ASSERT(vdc->validate_pending > 0);
8510 	vdc->validate_pending--;
8511 	mutex_exit(&vdc->lock);
8512 }
8513 
8514 /*
8515  * Function:
8516  *	vdc_setup_devid()
8517  *
8518  * Description:
8519  *	This routine discovers the devid of a vDisk. It requests the devid of
8520  *	the underlying device from the vDisk server, builds an encapsulated
8521  *	devid based on the retrieved devid and registers that new devid to
8522  *	the vDisk.
8523  *
8524  * Arguments:
8525  *	vdc	- soft state pointer for this instance of the device driver.
8526  *
8527  * Return Code:
8528  *	0	- A devid was succesfully registered for the vDisk
8529  */
8530 static int
8531 vdc_setup_devid(vdc_t *vdc)
8532 {
8533 	int rv;
8534 	vd_devid_t *vd_devid;
8535 	size_t bufsize, bufid_len;
8536 	ddi_devid_t vdisk_devid;
8537 	char *devid_str;
8538 
8539 	/*
8540 	 * At first sight, we don't know the size of the devid that the
8541 	 * server will return but this size will be encoded into the
8542 	 * reply. So we do a first request using a default size then we
8543 	 * check if this size was large enough. If not then we do a second
8544 	 * request with the correct size returned by the server. Note that
8545 	 * ldc requires size to be 8-byte aligned.
8546 	 */
8547 	bufsize = P2ROUNDUP(VD_DEVID_SIZE(VD_DEVID_DEFAULT_LEN),
8548 	    sizeof (uint64_t));
8549 	vd_devid = kmem_zalloc(bufsize, KM_SLEEP);
8550 	bufid_len = bufsize - sizeof (vd_efi_t) - 1;
8551 
8552 	rv = vdc_do_op(vdc, VD_OP_GET_DEVID, (caddr_t)vd_devid,
8553 	    bufsize, 0, 0, NULL, VIO_both_dir, 0);
8554 
8555 	DMSG(vdc, 2, "do_op returned %d\n", rv);
8556 
8557 	if (rv) {
8558 		kmem_free(vd_devid, bufsize);
8559 		return (rv);
8560 	}
8561 
8562 	if (vd_devid->length > bufid_len) {
8563 		/*
8564 		 * The returned devid is larger than the buffer used. Try again
8565 		 * with a buffer with the right size.
8566 		 */
8567 		kmem_free(vd_devid, bufsize);
8568 		bufsize = P2ROUNDUP(VD_DEVID_SIZE(vd_devid->length),
8569 		    sizeof (uint64_t));
8570 		vd_devid = kmem_zalloc(bufsize, KM_SLEEP);
8571 		bufid_len = bufsize - sizeof (vd_efi_t) - 1;
8572 
8573 		rv = vdc_do_sync_op(vdc, VD_OP_GET_DEVID, (caddr_t)vd_devid,
8574 		    bufsize, 0, 0, VIO_both_dir, B_TRUE);
8575 
8576 		if (rv) {
8577 			kmem_free(vd_devid, bufsize);
8578 			return (rv);
8579 		}
8580 	}
8581 
8582 	/*
8583 	 * The virtual disk should have the same device id as the one associated
8584 	 * with the physical disk it is mapped on, otherwise sharing a disk
8585 	 * between a LDom and a non-LDom may not work (for example for a shared
8586 	 * SVM disk set).
8587 	 *
8588 	 * The DDI framework does not allow creating a device id with any
8589 	 * type so we first create a device id of type DEVID_ENCAP and then
8590 	 * we restore the orignal type of the physical device.
8591 	 */
8592 
8593 	DMSG(vdc, 2, ": devid length = %d\n", vd_devid->length);
8594 
8595 	/* build an encapsulated devid based on the returned devid */
8596 	if (ddi_devid_init(vdc->dip, DEVID_ENCAP, vd_devid->length,
8597 	    vd_devid->id, &vdisk_devid) != DDI_SUCCESS) {
8598 		DMSG(vdc, 1, "[%d] Fail to created devid\n", vdc->instance);
8599 		kmem_free(vd_devid, bufsize);
8600 		return (1);
8601 	}
8602 
8603 	DEVID_FORMTYPE((impl_devid_t *)vdisk_devid, vd_devid->type);
8604 
8605 	ASSERT(ddi_devid_valid(vdisk_devid) == DDI_SUCCESS);
8606 
8607 	kmem_free(vd_devid, bufsize);
8608 
8609 	if (vdc->devid != NULL) {
8610 		/* check that the devid hasn't changed */
8611 		if (ddi_devid_compare(vdisk_devid, vdc->devid) == 0) {
8612 			ddi_devid_free(vdisk_devid);
8613 			return (0);
8614 		}
8615 
8616 		cmn_err(CE_WARN, "vdisk@%d backend devid has changed",
8617 		    vdc->instance);
8618 
8619 		devid_str = ddi_devid_str_encode(vdc->devid, NULL);
8620 
8621 		cmn_err(CE_CONT, "vdisk@%d backend initial devid: %s",
8622 		    vdc->instance,
8623 		    (devid_str)? devid_str : "<encoding error>");
8624 
8625 		if (devid_str)
8626 			ddi_devid_str_free(devid_str);
8627 
8628 		devid_str = ddi_devid_str_encode(vdisk_devid, NULL);
8629 
8630 		cmn_err(CE_CONT, "vdisk@%d backend current devid: %s",
8631 		    vdc->instance,
8632 		    (devid_str)? devid_str : "<encoding error>");
8633 
8634 		if (devid_str)
8635 			ddi_devid_str_free(devid_str);
8636 
8637 		ddi_devid_free(vdisk_devid);
8638 		return (1);
8639 	}
8640 
8641 	if (ddi_devid_register(vdc->dip, vdisk_devid) != DDI_SUCCESS) {
8642 		DMSG(vdc, 1, "[%d] Fail to register devid\n", vdc->instance);
8643 		ddi_devid_free(vdisk_devid);
8644 		return (1);
8645 	}
8646 
8647 	vdc->devid = vdisk_devid;
8648 
8649 	return (0);
8650 }
8651 
8652 static void
8653 vdc_store_label_efi(vdc_t *vdc, efi_gpt_t *gpt, efi_gpe_t *gpe)
8654 {
8655 	int i, nparts;
8656 
8657 	ASSERT(MUTEX_HELD(&vdc->lock));
8658 
8659 	vdc->vdisk_label = VD_DISK_LABEL_EFI;
8660 	bzero(vdc->vtoc, sizeof (struct extvtoc));
8661 	bzero(vdc->geom, sizeof (struct dk_geom));
8662 	bzero(vdc->slice, sizeof (vd_slice_t) * V_NUMPAR);
8663 
8664 	nparts = gpt->efi_gpt_NumberOfPartitionEntries;
8665 
8666 	for (i = 0; i < nparts && i < VD_EFI_WD_SLICE; i++) {
8667 
8668 		if (gpe[i].efi_gpe_StartingLBA == 0 &&
8669 		    gpe[i].efi_gpe_EndingLBA == 0) {
8670 			continue;
8671 		}
8672 
8673 		vdc->slice[i].start = gpe[i].efi_gpe_StartingLBA;
8674 		vdc->slice[i].nblocks = gpe[i].efi_gpe_EndingLBA -
8675 		    gpe[i].efi_gpe_StartingLBA + 1;
8676 	}
8677 
8678 	ASSERT(vdc->vdisk_size != 0);
8679 	vdc->slice[VD_EFI_WD_SLICE].start = 0;
8680 	vdc->slice[VD_EFI_WD_SLICE].nblocks = vdc->vdisk_size;
8681 
8682 }
8683 
8684 static void
8685 vdc_store_label_vtoc(vdc_t *vdc, struct dk_geom *geom, struct extvtoc *vtoc)
8686 {
8687 	int i;
8688 
8689 	ASSERT(MUTEX_HELD(&vdc->lock));
8690 	ASSERT(vdc->vdisk_bsize == vtoc->v_sectorsz);
8691 
8692 	vdc->vdisk_label = VD_DISK_LABEL_VTOC;
8693 	bcopy(vtoc, vdc->vtoc, sizeof (struct extvtoc));
8694 	bcopy(geom, vdc->geom, sizeof (struct dk_geom));
8695 	bzero(vdc->slice, sizeof (vd_slice_t) * V_NUMPAR);
8696 
8697 	for (i = 0; i < vtoc->v_nparts; i++) {
8698 		vdc->slice[i].start = vtoc->v_part[i].p_start;
8699 		vdc->slice[i].nblocks = vtoc->v_part[i].p_size;
8700 	}
8701 }
8702 
8703 static void
8704 vdc_store_label_unk(vdc_t *vdc)
8705 {
8706 	ASSERT(MUTEX_HELD(&vdc->lock));
8707 
8708 	vdc->vdisk_label = VD_DISK_LABEL_UNK;
8709 	bzero(vdc->vtoc, sizeof (struct extvtoc));
8710 	bzero(vdc->geom, sizeof (struct dk_geom));
8711 	bzero(vdc->slice, sizeof (vd_slice_t) * V_NUMPAR);
8712 }
8713