xref: /titanic_41/usr/src/uts/sun4v/io/vds.c (revision f8919bdadda3ebb97bd55cc14a16e0271ed57615)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #pragma ident	"%Z%%M%	%I%	%E% SMI"
28 
29 /*
30  * Virtual disk server
31  */
32 
33 
34 #include <sys/types.h>
35 #include <sys/conf.h>
36 #include <sys/crc32.h>
37 #include <sys/ddi.h>
38 #include <sys/dkio.h>
39 #include <sys/file.h>
40 #include <sys/fs/hsfs_isospec.h>
41 #include <sys/mdeg.h>
42 #include <sys/mhd.h>
43 #include <sys/modhash.h>
44 #include <sys/note.h>
45 #include <sys/pathname.h>
46 #include <sys/sdt.h>
47 #include <sys/sunddi.h>
48 #include <sys/sunldi.h>
49 #include <sys/sysmacros.h>
50 #include <sys/vio_common.h>
51 #include <sys/vio_util.h>
52 #include <sys/vdsk_mailbox.h>
53 #include <sys/vdsk_common.h>
54 #include <sys/vtoc.h>
55 #include <sys/vfs.h>
56 #include <sys/stat.h>
57 #include <sys/scsi/impl/uscsi.h>
58 #include <vm/seg_map.h>
59 
60 /* Virtual disk server initialization flags */
61 #define	VDS_LDI			0x01
62 #define	VDS_MDEG		0x02
63 
64 /* Virtual disk server tunable parameters */
65 #define	VDS_RETRIES		5
66 #define	VDS_LDC_DELAY		1000 /* 1 msecs */
67 #define	VDS_DEV_DELAY		10000000 /* 10 secs */
68 #define	VDS_NCHAINS		32
69 
70 /* Identification parameters for MD, synthetic dkio(7i) structures, etc. */
71 #define	VDS_NAME		"virtual-disk-server"
72 
73 #define	VD_NAME			"vd"
74 #define	VD_VOLUME_NAME		"vdisk"
75 #define	VD_ASCIILABEL		"Virtual Disk"
76 
77 #define	VD_CHANNEL_ENDPOINT	"channel-endpoint"
78 #define	VD_ID_PROP		"id"
79 #define	VD_BLOCK_DEVICE_PROP	"vds-block-device"
80 #define	VD_BLOCK_DEVICE_OPTS	"vds-block-device-opts"
81 #define	VD_REG_PROP		"reg"
82 
83 /* Virtual disk initialization flags */
84 #define	VD_DISK_READY		0x01
85 #define	VD_LOCKING		0x02
86 #define	VD_LDC			0x04
87 #define	VD_DRING		0x08
88 #define	VD_SID			0x10
89 #define	VD_SEQ_NUM		0x20
90 #define	VD_SETUP_ERROR		0x40
91 
92 /* Flags for writing to a vdisk which is a file */
93 #define	VD_FILE_WRITE_FLAGS	SM_ASYNC
94 
95 /* Number of backup labels */
96 #define	VD_FILE_NUM_BACKUP	5
97 
98 /* Timeout for SCSI I/O */
99 #define	VD_SCSI_RDWR_TIMEOUT	30	/* 30 secs */
100 
101 /*
102  * By Solaris convention, slice/partition 2 represents the entire disk;
103  * unfortunately, this convention does not appear to be codified.
104  */
105 #define	VD_ENTIRE_DISK_SLICE	2
106 
107 /* Return a cpp token as a string */
108 #define	STRINGIZE(token)	#token
109 
110 /*
111  * Print a message prefixed with the current function name to the message log
112  * (and optionally to the console for verbose boots); these macros use cpp's
113  * concatenation of string literals and C99 variable-length-argument-list
114  * macros
115  */
116 #define	PRN(...)	_PRN("?%s():  "__VA_ARGS__, "")
117 #define	_PRN(format, ...)					\
118 	cmn_err(CE_CONT, format"%s", __func__, __VA_ARGS__)
119 
120 /* Return a pointer to the "i"th vdisk dring element */
121 #define	VD_DRING_ELEM(i)	((vd_dring_entry_t *)(void *)	\
122 	    (vd->dring + (i)*vd->descriptor_size))
123 
124 /* Return the virtual disk client's type as a string (for use in messages) */
125 #define	VD_CLIENT(vd)							\
126 	(((vd)->xfer_mode == VIO_DESC_MODE) ? "in-band client" :	\
127 	    (((vd)->xfer_mode == VIO_DRING_MODE) ? "dring client" :	\
128 		(((vd)->xfer_mode == 0) ? "null client" :		\
129 		    "unsupported client")))
130 
131 /* Read disk label from a disk on file */
132 #define	VD_FILE_LABEL_READ(vd, labelp) \
133 	vd_file_rw(vd, VD_SLICE_NONE, VD_OP_BREAD, (caddr_t)labelp, \
134 	    0, sizeof (struct dk_label))
135 
136 /* Write disk label to a disk on file */
137 #define	VD_FILE_LABEL_WRITE(vd, labelp)	\
138 	vd_file_rw(vd, VD_SLICE_NONE, VD_OP_BWRITE, (caddr_t)labelp, \
139 	    0, sizeof (struct dk_label))
140 
141 /* Message for disk access rights reset failure */
142 #define	VD_RESET_ACCESS_FAILURE_MSG \
143 	"Fail to reset disk access rights for disk %s"
144 
145 /*
146  * Specification of an MD node passed to the MDEG to filter any
147  * 'vport' nodes that do not belong to the specified node. This
148  * template is copied for each vds instance and filled in with
149  * the appropriate 'cfg-handle' value before being passed to the MDEG.
150  */
151 static mdeg_prop_spec_t	vds_prop_template[] = {
152 	{ MDET_PROP_STR,	"name",		VDS_NAME },
153 	{ MDET_PROP_VAL,	"cfg-handle",	NULL },
154 	{ MDET_LIST_END,	NULL, 		NULL }
155 };
156 
157 #define	VDS_SET_MDEG_PROP_INST(specp, val) (specp)[1].ps_val = (val);
158 
159 /*
160  * Matching criteria passed to the MDEG to register interest
161  * in changes to 'virtual-device-port' nodes identified by their
162  * 'id' property.
163  */
164 static md_prop_match_t	vd_prop_match[] = {
165 	{ MDET_PROP_VAL,	VD_ID_PROP },
166 	{ MDET_LIST_END,	NULL }
167 };
168 
169 static mdeg_node_match_t vd_match = {"virtual-device-port",
170 				    vd_prop_match};
171 
172 /*
173  * Options for the VD_BLOCK_DEVICE_OPTS property.
174  */
175 #define	VD_OPT_RDONLY		0x1	/* read-only  */
176 #define	VD_OPT_SLICE		0x2	/* single slice */
177 #define	VD_OPT_EXCLUSIVE	0x4	/* exclusive access */
178 
179 #define	VD_OPTION_NLEN	128
180 
181 typedef struct vd_option {
182 	char vdo_name[VD_OPTION_NLEN];
183 	uint64_t vdo_value;
184 } vd_option_t;
185 
186 vd_option_t vd_bdev_options[] = {
187 	{ "ro",		VD_OPT_RDONLY },
188 	{ "slice", 	VD_OPT_SLICE },
189 	{ "excl",	VD_OPT_EXCLUSIVE }
190 };
191 
192 /* Debugging macros */
193 #ifdef DEBUG
194 
195 static int	vd_msglevel = 0;
196 
197 #define	PR0 if (vd_msglevel > 0)	PRN
198 #define	PR1 if (vd_msglevel > 1)	PRN
199 #define	PR2 if (vd_msglevel > 2)	PRN
200 
201 #define	VD_DUMP_DRING_ELEM(elem)					\
202 	PR0("dst:%x op:%x st:%u nb:%lx addr:%lx ncook:%u\n",		\
203 	    elem->hdr.dstate,						\
204 	    elem->payload.operation,					\
205 	    elem->payload.status,					\
206 	    elem->payload.nbytes,					\
207 	    elem->payload.addr,						\
208 	    elem->payload.ncookies);
209 
210 char *
211 vd_decode_state(int state)
212 {
213 	char *str;
214 
215 #define	CASE_STATE(_s)	case _s: str = #_s; break;
216 
217 	switch (state) {
218 	CASE_STATE(VD_STATE_INIT)
219 	CASE_STATE(VD_STATE_VER)
220 	CASE_STATE(VD_STATE_ATTR)
221 	CASE_STATE(VD_STATE_DRING)
222 	CASE_STATE(VD_STATE_RDX)
223 	CASE_STATE(VD_STATE_DATA)
224 	default: str = "unknown"; break;
225 	}
226 
227 #undef CASE_STATE
228 
229 	return (str);
230 }
231 
232 void
233 vd_decode_tag(vio_msg_t *msg)
234 {
235 	char *tstr, *sstr, *estr;
236 
237 #define	CASE_TYPE(_s)	case _s: tstr = #_s; break;
238 
239 	switch (msg->tag.vio_msgtype) {
240 	CASE_TYPE(VIO_TYPE_CTRL)
241 	CASE_TYPE(VIO_TYPE_DATA)
242 	CASE_TYPE(VIO_TYPE_ERR)
243 	default: tstr = "unknown"; break;
244 	}
245 
246 #undef CASE_TYPE
247 
248 #define	CASE_SUBTYPE(_s) case _s: sstr = #_s; break;
249 
250 	switch (msg->tag.vio_subtype) {
251 	CASE_SUBTYPE(VIO_SUBTYPE_INFO)
252 	CASE_SUBTYPE(VIO_SUBTYPE_ACK)
253 	CASE_SUBTYPE(VIO_SUBTYPE_NACK)
254 	default: sstr = "unknown"; break;
255 	}
256 
257 #undef CASE_SUBTYPE
258 
259 #define	CASE_ENV(_s)	case _s: estr = #_s; break;
260 
261 	switch (msg->tag.vio_subtype_env) {
262 	CASE_ENV(VIO_VER_INFO)
263 	CASE_ENV(VIO_ATTR_INFO)
264 	CASE_ENV(VIO_DRING_REG)
265 	CASE_ENV(VIO_DRING_UNREG)
266 	CASE_ENV(VIO_RDX)
267 	CASE_ENV(VIO_PKT_DATA)
268 	CASE_ENV(VIO_DESC_DATA)
269 	CASE_ENV(VIO_DRING_DATA)
270 	default: estr = "unknown"; break;
271 	}
272 
273 #undef CASE_ENV
274 
275 	PR1("(%x/%x/%x) message : (%s/%s/%s)",
276 	    msg->tag.vio_msgtype, msg->tag.vio_subtype,
277 	    msg->tag.vio_subtype_env, tstr, sstr, estr);
278 }
279 
280 #else	/* !DEBUG */
281 
282 #define	PR0(...)
283 #define	PR1(...)
284 #define	PR2(...)
285 
286 #define	VD_DUMP_DRING_ELEM(elem)
287 
288 #define	vd_decode_state(_s)	(NULL)
289 #define	vd_decode_tag(_s)	(NULL)
290 
291 #endif	/* DEBUG */
292 
293 
294 /*
295  * Soft state structure for a vds instance
296  */
297 typedef struct vds {
298 	uint_t		initialized;	/* driver inst initialization flags */
299 	dev_info_t	*dip;		/* driver inst devinfo pointer */
300 	ldi_ident_t	ldi_ident;	/* driver's identifier for LDI */
301 	mod_hash_t	*vd_table;	/* table of virtual disks served */
302 	mdeg_node_spec_t *ispecp;	/* mdeg node specification */
303 	mdeg_handle_t	mdeg;		/* handle for MDEG operations  */
304 } vds_t;
305 
306 /*
307  * Types of descriptor-processing tasks
308  */
309 typedef enum vd_task_type {
310 	VD_NONFINAL_RANGE_TASK,	/* task for intermediate descriptor in range */
311 	VD_FINAL_RANGE_TASK,	/* task for last in a range of descriptors */
312 } vd_task_type_t;
313 
314 /*
315  * Structure describing the task for processing a descriptor
316  */
317 typedef struct vd_task {
318 	struct vd		*vd;		/* vd instance task is for */
319 	vd_task_type_t		type;		/* type of descriptor task */
320 	int			index;		/* dring elem index for task */
321 	vio_msg_t		*msg;		/* VIO message task is for */
322 	size_t			msglen;		/* length of message content */
323 	vd_dring_payload_t	*request;	/* request task will perform */
324 	struct buf		buf;		/* buf(9s) for I/O request */
325 	ldc_mem_handle_t	mhdl;		/* task memory handle */
326 	int			status;		/* status of processing task */
327 	int	(*completef)(struct vd_task *task); /* completion func ptr */
328 } vd_task_t;
329 
330 /*
331  * Soft state structure for a virtual disk instance
332  */
333 typedef struct vd {
334 	uint_t			initialized;	/* vdisk initialization flags */
335 	uint64_t		operations;	/* bitmask of VD_OPs exported */
336 	vio_ver_t		version;	/* ver negotiated with client */
337 	vds_t			*vds;		/* server for this vdisk */
338 	ddi_taskq_t		*startq;	/* queue for I/O start tasks */
339 	ddi_taskq_t		*completionq;	/* queue for completion tasks */
340 	ldi_handle_t		ldi_handle[V_NUMPAR];	/* LDI slice handles */
341 	char			device_path[MAXPATHLEN + 1]; /* vdisk device */
342 	dev_t			dev[V_NUMPAR];	/* dev numbers for slices */
343 	int			open_flags;	/* open flags */
344 	uint_t			nslices;	/* number of slices */
345 	size_t			vdisk_size;	/* number of blocks in vdisk */
346 	size_t			vdisk_block_size; /* size of each vdisk block */
347 	vd_disk_type_t		vdisk_type;	/* slice or entire disk */
348 	vd_disk_label_t		vdisk_label;	/* EFI or VTOC label */
349 	vd_media_t		vdisk_media;	/* media type of backing dev. */
350 	boolean_t		is_atapi_dev;	/* Is this an IDE CD-ROM dev? */
351 	ushort_t		max_xfer_sz;	/* max xfer size in DEV_BSIZE */
352 	size_t			block_size;	/* blk size of actual device */
353 	boolean_t		pseudo;		/* underlying pseudo dev */
354 	boolean_t		file;		/* is vDisk backed by a file? */
355 	boolean_t		scsi;		/* is vDisk backed by scsi? */
356 	vnode_t			*file_vnode;	/* file vnode */
357 	size_t			file_size;	/* file size */
358 	ddi_devid_t		file_devid;	/* devid for disk image */
359 	struct dk_efi		dk_efi;		/* synthetic for slice type */
360 	struct dk_geom		dk_geom;	/* synthetic for slice type */
361 	struct dk_minfo		dk_minfo;	/* synthetic for slice type */
362 	struct vtoc		vtoc;		/* synthetic for slice type */
363 	boolean_t		ownership;	/* disk ownership status */
364 	ldc_status_t		ldc_state;	/* LDC connection state */
365 	ldc_handle_t		ldc_handle;	/* handle for LDC comm */
366 	size_t			max_msglen;	/* largest LDC message len */
367 	vd_state_t		state;		/* client handshake state */
368 	uint8_t			xfer_mode;	/* transfer mode with client */
369 	uint32_t		sid;		/* client's session ID */
370 	uint64_t		seq_num;	/* message sequence number */
371 	uint64_t		dring_ident;	/* identifier of dring */
372 	ldc_dring_handle_t	dring_handle;	/* handle for dring ops */
373 	uint32_t		descriptor_size;	/* num bytes in desc */
374 	uint32_t		dring_len;	/* number of dring elements */
375 	caddr_t			dring;		/* address of dring */
376 	caddr_t			vio_msgp;	/* vio msg staging buffer */
377 	vd_task_t		inband_task;	/* task for inband descriptor */
378 	vd_task_t		*dring_task;	/* tasks dring elements */
379 
380 	kmutex_t		lock;		/* protects variables below */
381 	boolean_t		enabled;	/* is vdisk enabled? */
382 	boolean_t		reset_state;	/* reset connection state? */
383 	boolean_t		reset_ldc;	/* reset LDC channel? */
384 } vd_t;
385 
386 typedef struct vds_operation {
387 	char	*namep;
388 	uint8_t	operation;
389 	int	(*start)(vd_task_t *task);
390 	int	(*complete)(vd_task_t *task);
391 } vds_operation_t;
392 
393 typedef struct vd_ioctl {
394 	uint8_t		operation;		/* vdisk operation */
395 	const char	*operation_name;	/* vdisk operation name */
396 	size_t		nbytes;			/* size of operation buffer */
397 	int		cmd;			/* corresponding ioctl cmd */
398 	const char	*cmd_name;		/* ioctl cmd name */
399 	void		*arg;			/* ioctl cmd argument */
400 	/* convert input vd_buf to output ioctl_arg */
401 	int		(*copyin)(void *vd_buf, size_t, void *ioctl_arg);
402 	/* convert input ioctl_arg to output vd_buf */
403 	void		(*copyout)(void *ioctl_arg, void *vd_buf);
404 	/* write is true if the operation writes any data to the backend */
405 	boolean_t	write;
406 } vd_ioctl_t;
407 
408 /* Define trivial copyin/copyout conversion function flag */
409 #define	VD_IDENTITY_IN	((int (*)(void *, size_t, void *))-1)
410 #define	VD_IDENTITY_OUT	((void (*)(void *, void *))-1)
411 
412 
413 static int	vds_ldc_retries = VDS_RETRIES;
414 static int	vds_ldc_delay = VDS_LDC_DELAY;
415 static int	vds_dev_retries = VDS_RETRIES;
416 static int	vds_dev_delay = VDS_DEV_DELAY;
417 static void	*vds_state;
418 
419 static uint_t	vd_file_write_flags = VD_FILE_WRITE_FLAGS;
420 
421 static short	vd_scsi_rdwr_timeout = VD_SCSI_RDWR_TIMEOUT;
422 static int	vd_scsi_debug = USCSI_SILENT;
423 
424 /*
425  * Tunable to define the behavior of the service domain if the vdisk server
426  * fails to reset disk exclusive access when a LDC channel is reset. When a
427  * LDC channel is reset the vdisk server will try to reset disk exclusive
428  * access by releasing any SCSI-2 reservation or resetting the disk. If these
429  * actions fail then the default behavior (vd_reset_access_failure = 0) is to
430  * print a warning message. This default behavior can be changed by setting
431  * the vd_reset_access_failure variable to A_REBOOT (= 0x1) and that will
432  * cause the service domain to reboot, or A_DUMP (= 0x5) and that will cause
433  * the service domain to panic. In both cases, the reset of the service domain
434  * should trigger a reset SCSI buses and hopefully clear any SCSI-2 reservation.
435  */
436 static int 	vd_reset_access_failure = 0;
437 
438 /*
439  * Tunable for backward compatibility. When this variable is set to B_TRUE,
440  * all disk volumes (ZFS, SVM, VxvM volumes) will be exported as single
441  * slice disks whether or not they have the "slice" option set. This is
442  * to provide a simple backward compatibility mechanism when upgrading
443  * the vds driver and using a domain configuration created before the
444  * "slice" option was available.
445  */
446 static boolean_t vd_volume_force_slice = B_FALSE;
447 
448 /*
449  * Supported protocol version pairs, from highest (newest) to lowest (oldest)
450  *
451  * Each supported major version should appear only once, paired with (and only
452  * with) its highest supported minor version number (as the protocol requires
453  * supporting all lower minor version numbers as well)
454  */
455 static const vio_ver_t	vds_version[] = {{1, 1}};
456 static const size_t	vds_num_versions =
457     sizeof (vds_version)/sizeof (vds_version[0]);
458 
459 static void vd_free_dring_task(vd_t *vdp);
460 static int vd_setup_vd(vd_t *vd);
461 static int vd_setup_single_slice_disk(vd_t *vd);
462 static int vd_setup_mediainfo(vd_t *vd);
463 static boolean_t vd_enabled(vd_t *vd);
464 static ushort_t vd_lbl2cksum(struct dk_label *label);
465 static int vd_file_validate_geometry(vd_t *vd);
466 static boolean_t vd_file_is_iso_image(vd_t *vd);
467 static void vd_set_exported_operations(vd_t *vd);
468 static void vd_reset_access(vd_t *vd);
469 
470 /*
471  * Function:
472  *	vd_file_rw
473  *
474  * Description:
475  * 	Read or write to a disk on file.
476  *
477  * Parameters:
478  *	vd		- disk on which the operation is performed.
479  *	slice		- slice on which the operation is performed,
480  *			  VD_SLICE_NONE indicates that the operation
481  *			  is done using an absolute disk offset.
482  *	operation	- operation to execute: read (VD_OP_BREAD) or
483  *			  write (VD_OP_BWRITE).
484  *	data		- buffer where data are read to or written from.
485  *	blk		- starting block for the operation.
486  *	len		- number of bytes to read or write.
487  *
488  * Return Code:
489  *	n >= 0		- success, n indicates the number of bytes read
490  *			  or written.
491  *	-1		- error.
492  */
493 static ssize_t
494 vd_file_rw(vd_t *vd, int slice, int operation, caddr_t data, size_t blk,
495     size_t len)
496 {
497 	caddr_t	maddr;
498 	size_t offset, maxlen, moffset, mlen, n;
499 	uint_t smflags;
500 	enum seg_rw srw;
501 
502 	ASSERT(vd->file);
503 	ASSERT(len > 0);
504 
505 	/*
506 	 * If a file is exported as a slice then we don't care about the vtoc.
507 	 * In that case, the vtoc is a fake mainly to make newfs happy and we
508 	 * handle any I/O as a raw disk access so that we can have access to the
509 	 * entire backend.
510 	 */
511 	if (vd->vdisk_type == VD_DISK_TYPE_SLICE || slice == VD_SLICE_NONE) {
512 		/* raw disk access */
513 		offset = blk * DEV_BSIZE;
514 	} else {
515 		ASSERT(slice >= 0 && slice < V_NUMPAR);
516 
517 		/*
518 		 * v1.0 vDisk clients depended on the server not verifying
519 		 * the label of a unformatted disk.  This "feature" is
520 		 * maintained for backward compatibility but all versions
521 		 * from v1.1 onwards must do the right thing.
522 		 */
523 		if (vd->vdisk_label == VD_DISK_LABEL_UNK &&
524 		    vio_ver_is_supported(vd->version, 1, 1) &&
525 		    vd_file_validate_geometry(vd) != 0) {
526 			PR0("Unknown disk label, can't do I/O from slice %d",
527 			    slice);
528 			return (-1);
529 		}
530 
531 		if (blk >= vd->vtoc.v_part[slice].p_size) {
532 			/* address past the end of the slice */
533 			PR0("req_addr (0x%lx) > psize (0x%lx)",
534 			    blk, vd->vtoc.v_part[slice].p_size);
535 			return (0);
536 		}
537 
538 		offset = (vd->vtoc.v_part[slice].p_start + blk) * DEV_BSIZE;
539 
540 		/*
541 		 * If the requested size is greater than the size
542 		 * of the partition, truncate the read/write.
543 		 */
544 		maxlen = (vd->vtoc.v_part[slice].p_size - blk) * DEV_BSIZE;
545 
546 		if (len > maxlen) {
547 			PR0("I/O size truncated to %lu bytes from %lu bytes",
548 			    maxlen, len);
549 			len = maxlen;
550 		}
551 	}
552 
553 	/*
554 	 * We have to ensure that we are reading/writing into the mmap
555 	 * range. If we have a partial disk image (e.g. an image of
556 	 * s0 instead s2) the system can try to access slices that
557 	 * are not included into the disk image.
558 	 */
559 	if ((offset + len) >= vd->file_size) {
560 		PR0("offset + nbytes (0x%lx + 0x%lx) >= "
561 		    "file_size (0x%lx)", offset, len, vd->file_size);
562 		return (-1);
563 	}
564 
565 	srw = (operation == VD_OP_BREAD)? S_READ : S_WRITE;
566 	smflags = (operation == VD_OP_BREAD)? 0 :
567 	    (SM_WRITE | vd_file_write_flags);
568 	n = len;
569 
570 	do {
571 		/*
572 		 * segmap_getmapflt() returns a MAXBSIZE chunk which is
573 		 * MAXBSIZE aligned.
574 		 */
575 		moffset = offset & MAXBOFFSET;
576 		mlen = MIN(MAXBSIZE - moffset, n);
577 		maddr = segmap_getmapflt(segkmap, vd->file_vnode, offset,
578 		    mlen, 1, srw);
579 		/*
580 		 * Fault in the pages so we can check for error and ensure
581 		 * that we can safely used the mapped address.
582 		 */
583 		if (segmap_fault(kas.a_hat, segkmap, maddr, mlen,
584 		    F_SOFTLOCK, srw) != 0) {
585 			(void) segmap_release(segkmap, maddr, 0);
586 			return (-1);
587 		}
588 
589 		if (operation == VD_OP_BREAD)
590 			bcopy(maddr + moffset, data, mlen);
591 		else
592 			bcopy(data, maddr + moffset, mlen);
593 
594 		if (segmap_fault(kas.a_hat, segkmap, maddr, mlen,
595 		    F_SOFTUNLOCK, srw) != 0) {
596 			(void) segmap_release(segkmap, maddr, 0);
597 			return (-1);
598 		}
599 		if (segmap_release(segkmap, maddr, smflags) != 0)
600 			return (-1);
601 		n -= mlen;
602 		offset += mlen;
603 		data += mlen;
604 
605 	} while (n > 0);
606 
607 	return (len);
608 }
609 
610 /*
611  * Function:
612  *	vd_file_build_default_label
613  *
614  * Description:
615  *	Return a default label for the given disk. This is used when the disk
616  *	does not have a valid VTOC so that the user can get a valid default
617  *	configuration. The default label has all slice sizes set to 0 (except
618  *	slice 2 which is the entire disk) to force the user to write a valid
619  *	label onto the disk image.
620  *
621  * Parameters:
622  *	vd		- disk on which the operation is performed.
623  *	label		- the returned default label.
624  *
625  * Return Code:
626  *	none.
627  */
628 static void
629 vd_file_build_default_label(vd_t *vd, struct dk_label *label)
630 {
631 	size_t size;
632 	char prefix;
633 	int slice, nparts;
634 	uint16_t tag;
635 
636 	ASSERT(vd->file);
637 
638 	/*
639 	 * We must have a resonable number of cylinders and sectors so
640 	 * that newfs can run using default values.
641 	 *
642 	 * if (disk_size < 2MB)
643 	 * 	phys_cylinders = disk_size / 100K
644 	 * else
645 	 * 	phys_cylinders = disk_size / 300K
646 	 *
647 	 * phys_cylinders = (phys_cylinders == 0) ? 1 : phys_cylinders
648 	 * alt_cylinders = (phys_cylinders > 2) ? 2 : 0;
649 	 * data_cylinders = phys_cylinders - alt_cylinders
650 	 *
651 	 * sectors = disk_size / (phys_cylinders * blk_size)
652 	 *
653 	 * The file size test is an attempt to not have too few cylinders
654 	 * for a small file, or so many on a big file that you waste space
655 	 * for backup superblocks or cylinder group structures.
656 	 */
657 	if (vd->file_size < (2 * 1024 * 1024))
658 		label->dkl_pcyl = vd->file_size / (100 * 1024);
659 	else
660 		label->dkl_pcyl = vd->file_size / (300 * 1024);
661 
662 	if (label->dkl_pcyl == 0)
663 		label->dkl_pcyl = 1;
664 
665 	label->dkl_acyl = 0;
666 
667 	if (vd->vdisk_type == VD_DISK_TYPE_SLICE) {
668 		nparts = 1;
669 		slice = 0;
670 		tag = V_UNASSIGNED;
671 	} else {
672 		if (label->dkl_pcyl > 2)
673 			label->dkl_acyl = 2;
674 		nparts = V_NUMPAR;
675 		slice = VD_ENTIRE_DISK_SLICE;
676 		tag = V_BACKUP;
677 	}
678 
679 	label->dkl_nsect = vd->file_size /
680 	    (DEV_BSIZE * label->dkl_pcyl);
681 	label->dkl_ncyl = label->dkl_pcyl - label->dkl_acyl;
682 	label->dkl_nhead = 1;
683 	label->dkl_write_reinstruct = 0;
684 	label->dkl_read_reinstruct = 0;
685 	label->dkl_rpm = 7200;
686 	label->dkl_apc = 0;
687 	label->dkl_intrlv = 0;
688 
689 	PR0("requested disk size: %ld bytes\n", vd->file_size);
690 	PR0("setup: ncyl=%d nhead=%d nsec=%d\n", label->dkl_pcyl,
691 	    label->dkl_nhead, label->dkl_nsect);
692 	PR0("provided disk size: %ld bytes\n", (uint64_t)
693 	    (label->dkl_pcyl * label->dkl_nhead *
694 	    label->dkl_nsect * DEV_BSIZE));
695 
696 	if (vd->file_size < (1ULL << 20)) {
697 		size = vd->file_size >> 10;
698 		prefix = 'K'; /* Kilobyte */
699 	} else if (vd->file_size < (1ULL << 30)) {
700 		size = vd->file_size >> 20;
701 		prefix = 'M'; /* Megabyte */
702 	} else if (vd->file_size < (1ULL << 40)) {
703 		size = vd->file_size >> 30;
704 		prefix = 'G'; /* Gigabyte */
705 	} else {
706 		size = vd->file_size >> 40;
707 		prefix = 'T'; /* Terabyte */
708 	}
709 
710 	/*
711 	 * We must have a correct label name otherwise format(1m) will
712 	 * not recognized the disk as labeled.
713 	 */
714 	(void) snprintf(label->dkl_asciilabel, LEN_DKL_ASCII,
715 	    "SUN-DiskImage-%ld%cB cyl %d alt %d hd %d sec %d",
716 	    size, prefix,
717 	    label->dkl_ncyl, label->dkl_acyl, label->dkl_nhead,
718 	    label->dkl_nsect);
719 
720 	/* default VTOC */
721 	label->dkl_vtoc.v_version = V_VERSION;
722 	label->dkl_vtoc.v_nparts = nparts;
723 	label->dkl_vtoc.v_sanity = VTOC_SANE;
724 	label->dkl_vtoc.v_part[slice].p_tag = tag;
725 	label->dkl_map[slice].dkl_cylno = 0;
726 	label->dkl_map[slice].dkl_nblk = label->dkl_ncyl *
727 	    label->dkl_nhead * label->dkl_nsect;
728 	label->dkl_cksum = vd_lbl2cksum(label);
729 }
730 
731 /*
732  * Function:
733  *	vd_file_set_vtoc
734  *
735  * Description:
736  *	Set the vtoc of a disk image by writing the label and backup
737  *	labels into the disk image backend.
738  *
739  * Parameters:
740  *	vd		- disk on which the operation is performed.
741  *	label		- the data to be written.
742  *
743  * Return Code:
744  *	0		- success.
745  *	n > 0		- error, n indicates the errno code.
746  */
747 static int
748 vd_file_set_vtoc(vd_t *vd, struct dk_label *label)
749 {
750 	int blk, sec, cyl, head, cnt;
751 
752 	ASSERT(vd->file);
753 
754 	if (VD_FILE_LABEL_WRITE(vd, label) < 0) {
755 		PR0("fail to write disk label");
756 		return (EIO);
757 	}
758 
759 	/*
760 	 * Backup labels are on the last alternate cylinder's
761 	 * first five odd sectors.
762 	 */
763 	if (label->dkl_acyl == 0) {
764 		PR0("no alternate cylinder, can not store backup labels");
765 		return (0);
766 	}
767 
768 	cyl = label->dkl_ncyl  + label->dkl_acyl - 1;
769 	head = label->dkl_nhead - 1;
770 
771 	blk = (cyl * ((label->dkl_nhead * label->dkl_nsect) - label->dkl_apc)) +
772 	    (head * label->dkl_nsect);
773 
774 	/*
775 	 * Write the backup labels. Make sure we don't try to write past
776 	 * the last cylinder.
777 	 */
778 	sec = 1;
779 
780 	for (cnt = 0; cnt < VD_FILE_NUM_BACKUP; cnt++) {
781 
782 		if (sec >= label->dkl_nsect) {
783 			PR0("not enough sector to store all backup labels");
784 			return (0);
785 		}
786 
787 		if (vd_file_rw(vd, VD_SLICE_NONE, VD_OP_BWRITE, (caddr_t)label,
788 		    blk + sec, sizeof (struct dk_label)) < 0) {
789 			PR0("error writing backup label at block %d\n",
790 			    blk + sec);
791 			return (EIO);
792 		}
793 
794 		PR1("wrote backup label at block %d\n", blk + sec);
795 
796 		sec += 2;
797 	}
798 
799 	return (0);
800 }
801 
802 /*
803  * Function:
804  *	vd_file_get_devid_block
805  *
806  * Description:
807  *	Return the block number where the device id is stored.
808  *
809  * Parameters:
810  *	vd		- disk on which the operation is performed.
811  *	blkp		- pointer to the block number
812  *
813  * Return Code:
814  *	0		- success
815  *	ENOSPC		- disk has no space to store a device id
816  */
817 static int
818 vd_file_get_devid_block(vd_t *vd, size_t *blkp)
819 {
820 	diskaddr_t spc, head, cyl;
821 
822 	ASSERT(vd->file);
823 	ASSERT(vd->vdisk_label == VD_DISK_LABEL_VTOC);
824 
825 	/* this geometry doesn't allow us to have a devid */
826 	if (vd->dk_geom.dkg_acyl < 2) {
827 		PR0("not enough alternate cylinder available for devid "
828 		    "(acyl=%u)", vd->dk_geom.dkg_acyl);
829 		return (ENOSPC);
830 	}
831 
832 	/* the devid is in on the track next to the last cylinder */
833 	cyl = vd->dk_geom.dkg_ncyl + vd->dk_geom.dkg_acyl - 2;
834 	spc = vd->dk_geom.dkg_nhead * vd->dk_geom.dkg_nsect;
835 	head = vd->dk_geom.dkg_nhead - 1;
836 
837 	*blkp = (cyl * (spc - vd->dk_geom.dkg_apc)) +
838 	    (head * vd->dk_geom.dkg_nsect) + 1;
839 
840 	return (0);
841 }
842 
843 /*
844  * Return the checksum of a disk block containing an on-disk devid.
845  */
846 static uint_t
847 vd_dkdevid2cksum(struct dk_devid *dkdevid)
848 {
849 	uint_t chksum, *ip;
850 	int i;
851 
852 	chksum = 0;
853 	ip = (uint_t *)dkdevid;
854 	for (i = 0; i < ((DEV_BSIZE - sizeof (int)) / sizeof (int)); i++)
855 		chksum ^= ip[i];
856 
857 	return (chksum);
858 }
859 
860 /*
861  * Function:
862  *	vd_file_read_devid
863  *
864  * Description:
865  *	Read the device id stored on a disk image.
866  *
867  * Parameters:
868  *	vd		- disk on which the operation is performed.
869  *	devid		- the return address of the device ID.
870  *
871  * Return Code:
872  *	0		- success
873  *	EIO		- I/O error while trying to access the disk image
874  *	EINVAL		- no valid device id was found
875  *	ENOSPC		- disk has no space to store a device id
876  */
877 static int
878 vd_file_read_devid(vd_t *vd, ddi_devid_t *devid)
879 {
880 	struct dk_devid *dkdevid;
881 	size_t blk;
882 	uint_t chksum;
883 	int status, sz;
884 
885 	if ((status = vd_file_get_devid_block(vd, &blk)) != 0)
886 		return (status);
887 
888 	dkdevid = kmem_zalloc(DEV_BSIZE, KM_SLEEP);
889 
890 	/* get the devid */
891 	if ((vd_file_rw(vd, VD_SLICE_NONE, VD_OP_BREAD, (caddr_t)dkdevid, blk,
892 	    DEV_BSIZE)) < 0) {
893 		PR0("error reading devid block at %lu", blk);
894 		status = EIO;
895 		goto done;
896 	}
897 
898 	/* validate the revision */
899 	if ((dkdevid->dkd_rev_hi != DK_DEVID_REV_MSB) ||
900 	    (dkdevid->dkd_rev_lo != DK_DEVID_REV_LSB)) {
901 		PR0("invalid devid found at block %lu (bad revision)", blk);
902 		status = EINVAL;
903 		goto done;
904 	}
905 
906 	/* compute checksum */
907 	chksum = vd_dkdevid2cksum(dkdevid);
908 
909 	/* compare the checksums */
910 	if (DKD_GETCHKSUM(dkdevid) != chksum) {
911 		PR0("invalid devid found at block %lu (bad checksum)", blk);
912 		status = EINVAL;
913 		goto done;
914 	}
915 
916 	/* validate the device id */
917 	if (ddi_devid_valid((ddi_devid_t)&dkdevid->dkd_devid) != DDI_SUCCESS) {
918 		PR0("invalid devid found at block %lu", blk);
919 		status = EINVAL;
920 		goto done;
921 	}
922 
923 	PR1("devid read at block %lu", blk);
924 
925 	sz = ddi_devid_sizeof((ddi_devid_t)&dkdevid->dkd_devid);
926 	*devid = kmem_alloc(sz, KM_SLEEP);
927 	bcopy(&dkdevid->dkd_devid, *devid, sz);
928 
929 done:
930 	kmem_free(dkdevid, DEV_BSIZE);
931 	return (status);
932 
933 }
934 
935 /*
936  * Function:
937  *	vd_file_write_devid
938  *
939  * Description:
940  *	Write a device id into disk image.
941  *
942  * Parameters:
943  *	vd		- disk on which the operation is performed.
944  *	devid		- the device ID to store.
945  *
946  * Return Code:
947  *	0		- success
948  *	EIO		- I/O error while trying to access the disk image
949  *	ENOSPC		- disk has no space to store a device id
950  */
951 static int
952 vd_file_write_devid(vd_t *vd, ddi_devid_t devid)
953 {
954 	struct dk_devid *dkdevid;
955 	uint_t chksum;
956 	size_t blk;
957 	int status;
958 
959 	if ((status = vd_file_get_devid_block(vd, &blk)) != 0)
960 		return (status);
961 
962 	dkdevid = kmem_zalloc(DEV_BSIZE, KM_SLEEP);
963 
964 	/* set revision */
965 	dkdevid->dkd_rev_hi = DK_DEVID_REV_MSB;
966 	dkdevid->dkd_rev_lo = DK_DEVID_REV_LSB;
967 
968 	/* copy devid */
969 	bcopy(devid, &dkdevid->dkd_devid, ddi_devid_sizeof(devid));
970 
971 	/* compute checksum */
972 	chksum = vd_dkdevid2cksum(dkdevid);
973 
974 	/* set checksum */
975 	DKD_FORMCHKSUM(chksum, dkdevid);
976 
977 	/* store the devid */
978 	if ((status = vd_file_rw(vd, VD_SLICE_NONE, VD_OP_BWRITE,
979 	    (caddr_t)dkdevid, blk, DEV_BSIZE)) < 0) {
980 		PR0("Error writing devid block at %lu", blk);
981 		status = EIO;
982 	} else {
983 		PR1("devid written at block %lu", blk);
984 		status = 0;
985 	}
986 
987 	kmem_free(dkdevid, DEV_BSIZE);
988 	return (status);
989 }
990 
991 /*
992  * Function:
993  *	vd_do_scsi_rdwr
994  *
995  * Description:
996  * 	Read or write to a SCSI disk using an absolute disk offset.
997  *
998  * Parameters:
999  *	vd		- disk on which the operation is performed.
1000  *	operation	- operation to execute: read (VD_OP_BREAD) or
1001  *			  write (VD_OP_BWRITE).
1002  *	data		- buffer where data are read to or written from.
1003  *	blk		- starting block for the operation.
1004  *	len		- number of bytes to read or write.
1005  *
1006  * Return Code:
1007  *	0		- success
1008  *	n != 0		- error.
1009  */
1010 static int
1011 vd_do_scsi_rdwr(vd_t *vd, int operation, caddr_t data, size_t blk, size_t len)
1012 {
1013 	struct uscsi_cmd ucmd;
1014 	union scsi_cdb cdb;
1015 	int nsectors, nblk;
1016 	int max_sectors;
1017 	int status, rval;
1018 
1019 	ASSERT(!vd->file);
1020 	ASSERT(vd->vdisk_block_size > 0);
1021 
1022 	max_sectors = vd->max_xfer_sz;
1023 	nblk = (len / vd->vdisk_block_size);
1024 
1025 	if (len % vd->vdisk_block_size != 0)
1026 		return (EINVAL);
1027 
1028 	/*
1029 	 * Build and execute the uscsi ioctl.  We build a group0, group1
1030 	 * or group4 command as necessary, since some targets
1031 	 * do not support group1 commands.
1032 	 */
1033 	while (nblk) {
1034 
1035 		bzero(&ucmd, sizeof (ucmd));
1036 		bzero(&cdb, sizeof (cdb));
1037 
1038 		nsectors = (max_sectors < nblk) ? max_sectors : nblk;
1039 
1040 		/*
1041 		 * Some of the optical drives on sun4v machines are ATAPI
1042 		 * devices which use Group 1 Read/Write commands so we need
1043 		 * to explicitly check a flag which is set when a domain
1044 		 * is bound.
1045 		 */
1046 		if (blk < (2 << 20) && nsectors <= 0xff && !vd->is_atapi_dev) {
1047 			FORMG0ADDR(&cdb, blk);
1048 			FORMG0COUNT(&cdb, nsectors);
1049 			ucmd.uscsi_cdblen = CDB_GROUP0;
1050 		} else if (blk > 0xffffffff) {
1051 			FORMG4LONGADDR(&cdb, blk);
1052 			FORMG4COUNT(&cdb, nsectors);
1053 			ucmd.uscsi_cdblen = CDB_GROUP4;
1054 			cdb.scc_cmd |= SCMD_GROUP4;
1055 		} else {
1056 			FORMG1ADDR(&cdb, blk);
1057 			FORMG1COUNT(&cdb, nsectors);
1058 			ucmd.uscsi_cdblen = CDB_GROUP1;
1059 			cdb.scc_cmd |= SCMD_GROUP1;
1060 		}
1061 		ucmd.uscsi_cdb = (caddr_t)&cdb;
1062 		ucmd.uscsi_bufaddr = data;
1063 		ucmd.uscsi_buflen = nsectors * vd->block_size;
1064 		ucmd.uscsi_timeout = vd_scsi_rdwr_timeout;
1065 		/*
1066 		 * Set flags so that the command is isolated from normal
1067 		 * commands and no error message is printed.
1068 		 */
1069 		ucmd.uscsi_flags = USCSI_ISOLATE | USCSI_SILENT;
1070 
1071 		if (operation == VD_OP_BREAD) {
1072 			cdb.scc_cmd |= SCMD_READ;
1073 			ucmd.uscsi_flags |= USCSI_READ;
1074 		} else {
1075 			cdb.scc_cmd |= SCMD_WRITE;
1076 		}
1077 
1078 		status = ldi_ioctl(vd->ldi_handle[VD_ENTIRE_DISK_SLICE],
1079 		    USCSICMD, (intptr_t)&ucmd, (vd->open_flags | FKIOCTL),
1080 		    kcred, &rval);
1081 
1082 		if (status == 0)
1083 			status = ucmd.uscsi_status;
1084 
1085 		if (status != 0)
1086 			break;
1087 
1088 		/*
1089 		 * Check if partial DMA breakup is required. If so, reduce
1090 		 * the request size by half and retry the last request.
1091 		 */
1092 		if (ucmd.uscsi_resid == ucmd.uscsi_buflen) {
1093 			max_sectors >>= 1;
1094 			if (max_sectors <= 0) {
1095 				status = EIO;
1096 				break;
1097 			}
1098 			continue;
1099 		}
1100 
1101 		if (ucmd.uscsi_resid != 0) {
1102 			status = EIO;
1103 			break;
1104 		}
1105 
1106 		blk += nsectors;
1107 		nblk -= nsectors;
1108 		data += nsectors * vd->vdisk_block_size; /* SECSIZE */
1109 	}
1110 
1111 	return (status);
1112 }
1113 
1114 /*
1115  * Function:
1116  *	vd_scsi_rdwr
1117  *
1118  * Description:
1119  * 	Wrapper function to read or write to a SCSI disk using an absolute
1120  *	disk offset. It checks the blocksize of the underlying device and,
1121  *	if necessary, adjusts the buffers accordingly before calling
1122  *	vd_do_scsi_rdwr() to do the actual read or write.
1123  *
1124  * Parameters:
1125  *	vd		- disk on which the operation is performed.
1126  *	operation	- operation to execute: read (VD_OP_BREAD) or
1127  *			  write (VD_OP_BWRITE).
1128  *	data		- buffer where data are read to or written from.
1129  *	blk		- starting block for the operation.
1130  *	len		- number of bytes to read or write.
1131  *
1132  * Return Code:
1133  *	0		- success
1134  *	n != 0		- error.
1135  */
1136 static int
1137 vd_scsi_rdwr(vd_t *vd, int operation, caddr_t data, size_t vblk, size_t vlen)
1138 {
1139 	int	rv;
1140 
1141 	size_t	pblk;	/* physical device block number of data on device */
1142 	size_t	delta;	/* relative offset between pblk and vblk */
1143 	size_t	pnblk;	/* number of physical blocks to be read from device */
1144 	size_t	plen;	/* length of data to be read from physical device */
1145 	char	*buf;	/* buffer area to fit physical device's block size */
1146 
1147 	if (vd->block_size == 0) {
1148 		/*
1149 		 * The block size was not available during the attach,
1150 		 * try to update it now.
1151 		 */
1152 		if (vd_setup_mediainfo(vd) != 0)
1153 			return (EIO);
1154 	}
1155 
1156 	/*
1157 	 * If the vdisk block size and the block size of the underlying device
1158 	 * match we can skip straight to vd_do_scsi_rdwr(), otherwise we need
1159 	 * to create a buffer large enough to handle the device's block size
1160 	 * and adjust the block to be read from and the amount of data to
1161 	 * read to correspond with the device's block size.
1162 	 */
1163 	if (vd->vdisk_block_size == vd->block_size)
1164 		return (vd_do_scsi_rdwr(vd, operation, data, vblk, vlen));
1165 
1166 	if (vd->vdisk_block_size > vd->block_size)
1167 		return (EINVAL);
1168 
1169 	/*
1170 	 * Writing of physical block sizes larger than the virtual block size
1171 	 * is not supported. This would be added if/when support for guests
1172 	 * writing to DVDs is implemented.
1173 	 */
1174 	if (operation == VD_OP_BWRITE)
1175 		return (ENOTSUP);
1176 
1177 	/* BEGIN CSTYLED */
1178 	/*
1179 	 * Below is a diagram showing the relationship between the physical
1180 	 * and virtual blocks. If the virtual blocks marked by 'X' below are
1181 	 * requested, then the physical blocks denoted by 'Y' are read.
1182 	 *
1183 	 *           vblk
1184 	 *             |      vlen
1185 	 *             |<--------------->|
1186 	 *             v                 v
1187 	 *  --+--+--+--+--+--+--+--+--+--+--+--+--+--+--+-   virtual disk:
1188 	 *    |  |  |  |XX|XX|XX|XX|XX|XX|  |  |  |  |  |  } block size is
1189 	 *  --+--+--+--+--+--+--+--+--+--+--+--+--+--+--+-  vd->vdisk_block_size
1190 	 *          :  :                 :  :
1191 	 *         >:==:< delta          :  :
1192 	 *          :  :                 :  :
1193 	 *  --+-----+-----+-----+-----+-----+-----+-----+--   physical disk:
1194 	 *    |     |YY:YY|YYYYY|YYYYY|YY:YY|     |     |   } block size is
1195 	 *  --+-----+-----+-----+-----+-----+-----+-----+--   vd->block_size
1196 	 *          ^                       ^
1197 	 *          |<--------------------->|
1198 	 *          |         plen
1199 	 *         pblk
1200 	 */
1201 	/* END CSTYLED */
1202 	pblk = (vblk * vd->vdisk_block_size) / vd->block_size;
1203 	delta = (vblk * vd->vdisk_block_size) - (pblk * vd->block_size);
1204 	pnblk = ((delta + vlen - 1) / vd->block_size) + 1;
1205 	plen = pnblk * vd->block_size;
1206 
1207 	PR2("vblk %lx:pblk %lx: vlen %ld:plen %ld", vblk, pblk, vlen, plen);
1208 
1209 	buf = kmem_zalloc(sizeof (caddr_t) * plen, KM_SLEEP);
1210 	rv = vd_do_scsi_rdwr(vd, operation, (caddr_t)buf, pblk, plen);
1211 	bcopy(buf + delta, data, vlen);
1212 
1213 	kmem_free(buf, sizeof (caddr_t) * plen);
1214 
1215 	return (rv);
1216 }
1217 
1218 /*
1219  * Return Values
1220  *	EINPROGRESS	- operation was successfully started
1221  *	EIO		- encountered LDC (aka. task error)
1222  *	0		- operation completed successfully
1223  *
1224  * Side Effect
1225  *     sets request->status = <disk operation status>
1226  */
1227 static int
1228 vd_start_bio(vd_task_t *task)
1229 {
1230 	int			rv, status = 0;
1231 	vd_t			*vd		= task->vd;
1232 	vd_dring_payload_t	*request	= task->request;
1233 	struct buf		*buf		= &task->buf;
1234 	uint8_t			mtype;
1235 	int 			slice;
1236 	char			*bufaddr = 0;
1237 	size_t			buflen;
1238 
1239 	ASSERT(vd != NULL);
1240 	ASSERT(request != NULL);
1241 
1242 	slice = request->slice;
1243 
1244 	ASSERT(slice == VD_SLICE_NONE || slice < vd->nslices);
1245 	ASSERT((request->operation == VD_OP_BREAD) ||
1246 	    (request->operation == VD_OP_BWRITE));
1247 
1248 	if (request->nbytes == 0) {
1249 		/* no service for trivial requests */
1250 		request->status = EINVAL;
1251 		return (0);
1252 	}
1253 
1254 	PR1("%s %lu bytes at block %lu",
1255 	    (request->operation == VD_OP_BREAD) ? "Read" : "Write",
1256 	    request->nbytes, request->addr);
1257 
1258 	/*
1259 	 * We have to check the open flags because the functions processing
1260 	 * the read/write request will not do it.
1261 	 */
1262 	if (request->operation == VD_OP_BWRITE && !(vd->open_flags & FWRITE)) {
1263 		PR0("write fails because backend is opened read-only");
1264 		request->nbytes = 0;
1265 		request->status = EROFS;
1266 		return (0);
1267 	}
1268 
1269 	mtype = (&vd->inband_task == task) ? LDC_SHADOW_MAP : LDC_DIRECT_MAP;
1270 
1271 	/* Map memory exported by client */
1272 	status = ldc_mem_map(task->mhdl, request->cookie, request->ncookies,
1273 	    mtype, (request->operation == VD_OP_BREAD) ? LDC_MEM_W : LDC_MEM_R,
1274 	    &bufaddr, NULL);
1275 	if (status != 0) {
1276 		PR0("ldc_mem_map() returned err %d ", status);
1277 		return (EIO);
1278 	}
1279 
1280 	buflen = request->nbytes;
1281 
1282 	status = ldc_mem_acquire(task->mhdl, 0, buflen);
1283 	if (status != 0) {
1284 		(void) ldc_mem_unmap(task->mhdl);
1285 		PR0("ldc_mem_acquire() returned err %d ", status);
1286 		return (EIO);
1287 	}
1288 
1289 	/* Start the block I/O */
1290 	if (vd->file) {
1291 		rv = vd_file_rw(vd, slice, request->operation, bufaddr,
1292 		    request->addr, request->nbytes);
1293 		if (rv < 0) {
1294 			request->nbytes = 0;
1295 			request->status = EIO;
1296 		} else {
1297 			request->nbytes = rv;
1298 			request->status = 0;
1299 		}
1300 	} else {
1301 		if (slice == VD_SLICE_NONE) {
1302 			/*
1303 			 * This is not a disk image so it is a real disk. We
1304 			 * assume that the underlying device driver supports
1305 			 * USCSICMD ioctls. This is the case of all SCSI devices
1306 			 * (sd, ssd...).
1307 			 *
1308 			 * In the future if we have non-SCSI disks we would need
1309 			 * to invoke the appropriate function to do I/O using an
1310 			 * absolute disk offset (for example using DIOCTL_RWCMD
1311 			 * for IDE disks).
1312 			 */
1313 			rv = vd_scsi_rdwr(vd, request->operation, bufaddr,
1314 			    request->addr, request->nbytes);
1315 			if (rv != 0) {
1316 				request->nbytes = 0;
1317 				request->status = EIO;
1318 			} else {
1319 				request->status = 0;
1320 			}
1321 		} else {
1322 			bioinit(buf);
1323 			buf->b_flags	= B_BUSY;
1324 			buf->b_bcount	= request->nbytes;
1325 			buf->b_lblkno	= request->addr;
1326 			buf->b_edev 	= vd->dev[slice];
1327 			buf->b_un.b_addr = bufaddr;
1328 			buf->b_flags 	|= (request->operation == VD_OP_BREAD)?
1329 			    B_READ : B_WRITE;
1330 
1331 			request->status =
1332 			    ldi_strategy(vd->ldi_handle[slice], buf);
1333 
1334 			/*
1335 			 * This is to indicate to the caller that the request
1336 			 * needs to be finished by vd_complete_bio() by calling
1337 			 * biowait() there and waiting for that to return before
1338 			 * triggering the notification of the vDisk client.
1339 			 *
1340 			 * This is necessary when writing to real disks as
1341 			 * otherwise calls to ldi_strategy() would be serialized
1342 			 * behind the calls to biowait() and performance would
1343 			 * suffer.
1344 			 */
1345 			if (request->status == 0)
1346 				return (EINPROGRESS);
1347 
1348 			biofini(buf);
1349 		}
1350 	}
1351 
1352 	/* Clean up after error */
1353 	rv = ldc_mem_release(task->mhdl, 0, buflen);
1354 	if (rv) {
1355 		PR0("ldc_mem_release() returned err %d ", rv);
1356 		status = EIO;
1357 	}
1358 	rv = ldc_mem_unmap(task->mhdl);
1359 	if (rv) {
1360 		PR0("ldc_mem_unmap() returned err %d ", rv);
1361 		status = EIO;
1362 	}
1363 
1364 	return (status);
1365 }
1366 
1367 /*
1368  * This function should only be called from vd_notify to ensure that requests
1369  * are responded to in the order that they are received.
1370  */
1371 static int
1372 send_msg(ldc_handle_t ldc_handle, void *msg, size_t msglen)
1373 {
1374 	int	status;
1375 	size_t	nbytes;
1376 
1377 	do {
1378 		nbytes = msglen;
1379 		status = ldc_write(ldc_handle, msg, &nbytes);
1380 		if (status != EWOULDBLOCK)
1381 			break;
1382 		drv_usecwait(vds_ldc_delay);
1383 	} while (status == EWOULDBLOCK);
1384 
1385 	if (status != 0) {
1386 		if (status != ECONNRESET)
1387 			PR0("ldc_write() returned errno %d", status);
1388 		return (status);
1389 	} else if (nbytes != msglen) {
1390 		PR0("ldc_write() performed only partial write");
1391 		return (EIO);
1392 	}
1393 
1394 	PR1("SENT %lu bytes", msglen);
1395 	return (0);
1396 }
1397 
1398 static void
1399 vd_need_reset(vd_t *vd, boolean_t reset_ldc)
1400 {
1401 	mutex_enter(&vd->lock);
1402 	vd->reset_state	= B_TRUE;
1403 	vd->reset_ldc	= reset_ldc;
1404 	mutex_exit(&vd->lock);
1405 }
1406 
1407 /*
1408  * Reset the state of the connection with a client, if needed; reset the LDC
1409  * transport as well, if needed.  This function should only be called from the
1410  * "vd_recv_msg", as it waits for tasks - otherwise a deadlock can occur.
1411  */
1412 static void
1413 vd_reset_if_needed(vd_t *vd)
1414 {
1415 	int	status = 0;
1416 
1417 	mutex_enter(&vd->lock);
1418 	if (!vd->reset_state) {
1419 		ASSERT(!vd->reset_ldc);
1420 		mutex_exit(&vd->lock);
1421 		return;
1422 	}
1423 	mutex_exit(&vd->lock);
1424 
1425 	PR0("Resetting connection state with %s", VD_CLIENT(vd));
1426 
1427 	/*
1428 	 * Let any asynchronous I/O complete before possibly pulling the rug
1429 	 * out from under it; defer checking vd->reset_ldc, as one of the
1430 	 * asynchronous tasks might set it
1431 	 */
1432 	ddi_taskq_wait(vd->completionq);
1433 
1434 	if (vd->file) {
1435 		status = VOP_FSYNC(vd->file_vnode, FSYNC, kcred, NULL);
1436 		if (status) {
1437 			PR0("VOP_FSYNC returned errno %d", status);
1438 		}
1439 	}
1440 
1441 	if ((vd->initialized & VD_DRING) &&
1442 	    ((status = ldc_mem_dring_unmap(vd->dring_handle)) != 0))
1443 		PR0("ldc_mem_dring_unmap() returned errno %d", status);
1444 
1445 	vd_free_dring_task(vd);
1446 
1447 	/* Free the staging buffer for msgs */
1448 	if (vd->vio_msgp != NULL) {
1449 		kmem_free(vd->vio_msgp, vd->max_msglen);
1450 		vd->vio_msgp = NULL;
1451 	}
1452 
1453 	/* Free the inband message buffer */
1454 	if (vd->inband_task.msg != NULL) {
1455 		kmem_free(vd->inband_task.msg, vd->max_msglen);
1456 		vd->inband_task.msg = NULL;
1457 	}
1458 
1459 	mutex_enter(&vd->lock);
1460 
1461 	if (vd->reset_ldc)
1462 		PR0("taking down LDC channel");
1463 	if (vd->reset_ldc && ((status = ldc_down(vd->ldc_handle)) != 0))
1464 		PR0("ldc_down() returned errno %d", status);
1465 
1466 	/* Reset exclusive access rights */
1467 	vd_reset_access(vd);
1468 
1469 	vd->initialized	&= ~(VD_SID | VD_SEQ_NUM | VD_DRING);
1470 	vd->state	= VD_STATE_INIT;
1471 	vd->max_msglen	= sizeof (vio_msg_t);	/* baseline vio message size */
1472 
1473 	/* Allocate the staging buffer */
1474 	vd->vio_msgp = kmem_alloc(vd->max_msglen, KM_SLEEP);
1475 
1476 	PR0("calling ldc_up\n");
1477 	(void) ldc_up(vd->ldc_handle);
1478 
1479 	vd->reset_state	= B_FALSE;
1480 	vd->reset_ldc	= B_FALSE;
1481 
1482 	mutex_exit(&vd->lock);
1483 }
1484 
1485 static void vd_recv_msg(void *arg);
1486 
1487 static void
1488 vd_mark_in_reset(vd_t *vd)
1489 {
1490 	int status;
1491 
1492 	PR0("vd_mark_in_reset: marking vd in reset\n");
1493 
1494 	vd_need_reset(vd, B_FALSE);
1495 	status = ddi_taskq_dispatch(vd->startq, vd_recv_msg, vd, DDI_SLEEP);
1496 	if (status == DDI_FAILURE) {
1497 		PR0("cannot schedule task to recv msg\n");
1498 		vd_need_reset(vd, B_TRUE);
1499 		return;
1500 	}
1501 }
1502 
1503 static int
1504 vd_mark_elem_done(vd_t *vd, int idx, int elem_status, int elem_nbytes)
1505 {
1506 	boolean_t		accepted;
1507 	int			status;
1508 	vd_dring_entry_t	*elem = VD_DRING_ELEM(idx);
1509 
1510 	if (vd->reset_state)
1511 		return (0);
1512 
1513 	/* Acquire the element */
1514 	if (!vd->reset_state &&
1515 	    (status = ldc_mem_dring_acquire(vd->dring_handle, idx, idx)) != 0) {
1516 		if (status == ECONNRESET) {
1517 			vd_mark_in_reset(vd);
1518 			return (0);
1519 		} else {
1520 			PR0("ldc_mem_dring_acquire() returned errno %d",
1521 			    status);
1522 			return (status);
1523 		}
1524 	}
1525 
1526 	/* Set the element's status and mark it done */
1527 	accepted = (elem->hdr.dstate == VIO_DESC_ACCEPTED);
1528 	if (accepted) {
1529 		elem->payload.nbytes	= elem_nbytes;
1530 		elem->payload.status	= elem_status;
1531 		elem->hdr.dstate	= VIO_DESC_DONE;
1532 	} else {
1533 		/* Perhaps client timed out waiting for I/O... */
1534 		PR0("element %u no longer \"accepted\"", idx);
1535 		VD_DUMP_DRING_ELEM(elem);
1536 	}
1537 	/* Release the element */
1538 	if (!vd->reset_state &&
1539 	    (status = ldc_mem_dring_release(vd->dring_handle, idx, idx)) != 0) {
1540 		if (status == ECONNRESET) {
1541 			vd_mark_in_reset(vd);
1542 			return (0);
1543 		} else {
1544 			PR0("ldc_mem_dring_release() returned errno %d",
1545 			    status);
1546 			return (status);
1547 		}
1548 	}
1549 
1550 	return (accepted ? 0 : EINVAL);
1551 }
1552 
1553 /*
1554  * Return Values
1555  *	0	- operation completed successfully
1556  *	EIO	- encountered LDC / task error
1557  *
1558  * Side Effect
1559  *	sets request->status = <disk operation status>
1560  */
1561 static int
1562 vd_complete_bio(vd_task_t *task)
1563 {
1564 	int			status		= 0;
1565 	int			rv		= 0;
1566 	vd_t			*vd		= task->vd;
1567 	vd_dring_payload_t	*request	= task->request;
1568 	struct buf		*buf		= &task->buf;
1569 
1570 
1571 	ASSERT(vd != NULL);
1572 	ASSERT(request != NULL);
1573 	ASSERT(task->msg != NULL);
1574 	ASSERT(task->msglen >= sizeof (*task->msg));
1575 	ASSERT(!vd->file);
1576 	ASSERT(request->slice != VD_SLICE_NONE);
1577 
1578 	/* Wait for the I/O to complete [ call to ldi_strategy(9f) ] */
1579 	request->status = biowait(buf);
1580 
1581 	/* return back the number of bytes read/written */
1582 	request->nbytes = buf->b_bcount - buf->b_resid;
1583 
1584 	/* Release the buffer */
1585 	if (!vd->reset_state)
1586 		status = ldc_mem_release(task->mhdl, 0, buf->b_bcount);
1587 	if (status) {
1588 		PR0("ldc_mem_release() returned errno %d copying to "
1589 		    "client", status);
1590 		if (status == ECONNRESET) {
1591 			vd_mark_in_reset(vd);
1592 		}
1593 		rv = EIO;
1594 	}
1595 
1596 	/* Unmap the memory, even if in reset */
1597 	status = ldc_mem_unmap(task->mhdl);
1598 	if (status) {
1599 		PR0("ldc_mem_unmap() returned errno %d copying to client",
1600 		    status);
1601 		if (status == ECONNRESET) {
1602 			vd_mark_in_reset(vd);
1603 		}
1604 		rv = EIO;
1605 	}
1606 
1607 	biofini(buf);
1608 
1609 	return (rv);
1610 }
1611 
1612 /*
1613  * Description:
1614  *	This function is called by the two functions called by a taskq
1615  *	[ vd_complete_notify() and vd_serial_notify()) ] to send the
1616  *	message to the client.
1617  *
1618  * Parameters:
1619  *	arg 	- opaque pointer to structure containing task to be completed
1620  *
1621  * Return Values
1622  *	None
1623  */
1624 static void
1625 vd_notify(vd_task_t *task)
1626 {
1627 	int	status;
1628 
1629 	ASSERT(task != NULL);
1630 	ASSERT(task->vd != NULL);
1631 
1632 	if (task->vd->reset_state)
1633 		return;
1634 
1635 	/*
1636 	 * Send the "ack" or "nack" back to the client; if sending the message
1637 	 * via LDC fails, arrange to reset both the connection state and LDC
1638 	 * itself
1639 	 */
1640 	PR2("Sending %s",
1641 	    (task->msg->tag.vio_subtype == VIO_SUBTYPE_ACK) ? "ACK" : "NACK");
1642 
1643 	status = send_msg(task->vd->ldc_handle, task->msg, task->msglen);
1644 	switch (status) {
1645 	case 0:
1646 		break;
1647 	case ECONNRESET:
1648 		vd_mark_in_reset(task->vd);
1649 		break;
1650 	default:
1651 		PR0("initiating full reset");
1652 		vd_need_reset(task->vd, B_TRUE);
1653 		break;
1654 	}
1655 
1656 	DTRACE_PROBE1(task__end, vd_task_t *, task);
1657 }
1658 
1659 /*
1660  * Description:
1661  *	Mark the Dring entry as Done and (if necessary) send an ACK/NACK to
1662  *	the vDisk client
1663  *
1664  * Parameters:
1665  *	task 		- structure containing the request sent from client
1666  *
1667  * Return Values
1668  *	None
1669  */
1670 static void
1671 vd_complete_notify(vd_task_t *task)
1672 {
1673 	int			status		= 0;
1674 	vd_t			*vd		= task->vd;
1675 	vd_dring_payload_t	*request	= task->request;
1676 
1677 	/* Update the dring element for a dring client */
1678 	if (!vd->reset_state && (vd->xfer_mode == VIO_DRING_MODE)) {
1679 		status = vd_mark_elem_done(vd, task->index,
1680 		    request->status, request->nbytes);
1681 		if (status == ECONNRESET)
1682 			vd_mark_in_reset(vd);
1683 	}
1684 
1685 	/*
1686 	 * If a transport error occurred while marking the element done or
1687 	 * previously while executing the task, arrange to "nack" the message
1688 	 * when the final task in the descriptor element range completes
1689 	 */
1690 	if ((status != 0) || (task->status != 0))
1691 		task->msg->tag.vio_subtype = VIO_SUBTYPE_NACK;
1692 
1693 	/*
1694 	 * Only the final task for a range of elements will respond to and
1695 	 * free the message
1696 	 */
1697 	if (task->type == VD_NONFINAL_RANGE_TASK) {
1698 		return;
1699 	}
1700 
1701 	vd_notify(task);
1702 }
1703 
1704 /*
1705  * Description:
1706  *	This is the basic completion function called to handle inband data
1707  *	requests and handshake messages. All it needs to do is trigger a
1708  *	message to the client that the request is completed.
1709  *
1710  * Parameters:
1711  *	arg 	- opaque pointer to structure containing task to be completed
1712  *
1713  * Return Values
1714  *	None
1715  */
1716 static void
1717 vd_serial_notify(void *arg)
1718 {
1719 	vd_task_t		*task = (vd_task_t *)arg;
1720 
1721 	ASSERT(task != NULL);
1722 	vd_notify(task);
1723 }
1724 
1725 /* ARGSUSED */
1726 static int
1727 vd_geom2dk_geom(void *vd_buf, size_t vd_buf_len, void *ioctl_arg)
1728 {
1729 	VD_GEOM2DK_GEOM((vd_geom_t *)vd_buf, (struct dk_geom *)ioctl_arg);
1730 	return (0);
1731 }
1732 
1733 /* ARGSUSED */
1734 static int
1735 vd_vtoc2vtoc(void *vd_buf, size_t vd_buf_len, void *ioctl_arg)
1736 {
1737 	VD_VTOC2VTOC((vd_vtoc_t *)vd_buf, (struct vtoc *)ioctl_arg);
1738 	return (0);
1739 }
1740 
1741 static void
1742 dk_geom2vd_geom(void *ioctl_arg, void *vd_buf)
1743 {
1744 	DK_GEOM2VD_GEOM((struct dk_geom *)ioctl_arg, (vd_geom_t *)vd_buf);
1745 }
1746 
1747 static void
1748 vtoc2vd_vtoc(void *ioctl_arg, void *vd_buf)
1749 {
1750 	VTOC2VD_VTOC((struct vtoc *)ioctl_arg, (vd_vtoc_t *)vd_buf);
1751 }
1752 
1753 static int
1754 vd_get_efi_in(void *vd_buf, size_t vd_buf_len, void *ioctl_arg)
1755 {
1756 	vd_efi_t *vd_efi = (vd_efi_t *)vd_buf;
1757 	dk_efi_t *dk_efi = (dk_efi_t *)ioctl_arg;
1758 	size_t data_len;
1759 
1760 	data_len = vd_buf_len - (sizeof (vd_efi_t) - sizeof (uint64_t));
1761 	if (vd_efi->length > data_len)
1762 		return (EINVAL);
1763 
1764 	dk_efi->dki_lba = vd_efi->lba;
1765 	dk_efi->dki_length = vd_efi->length;
1766 	dk_efi->dki_data = kmem_zalloc(vd_efi->length, KM_SLEEP);
1767 	return (0);
1768 }
1769 
1770 static void
1771 vd_get_efi_out(void *ioctl_arg, void *vd_buf)
1772 {
1773 	int len;
1774 	vd_efi_t *vd_efi = (vd_efi_t *)vd_buf;
1775 	dk_efi_t *dk_efi = (dk_efi_t *)ioctl_arg;
1776 
1777 	len = vd_efi->length;
1778 	DK_EFI2VD_EFI(dk_efi, vd_efi);
1779 	kmem_free(dk_efi->dki_data, len);
1780 }
1781 
1782 static int
1783 vd_set_efi_in(void *vd_buf, size_t vd_buf_len, void *ioctl_arg)
1784 {
1785 	vd_efi_t *vd_efi = (vd_efi_t *)vd_buf;
1786 	dk_efi_t *dk_efi = (dk_efi_t *)ioctl_arg;
1787 	size_t data_len;
1788 
1789 	data_len = vd_buf_len - (sizeof (vd_efi_t) - sizeof (uint64_t));
1790 	if (vd_efi->length > data_len)
1791 		return (EINVAL);
1792 
1793 	dk_efi->dki_data = kmem_alloc(vd_efi->length, KM_SLEEP);
1794 	VD_EFI2DK_EFI(vd_efi, dk_efi);
1795 	return (0);
1796 }
1797 
1798 static void
1799 vd_set_efi_out(void *ioctl_arg, void *vd_buf)
1800 {
1801 	vd_efi_t *vd_efi = (vd_efi_t *)vd_buf;
1802 	dk_efi_t *dk_efi = (dk_efi_t *)ioctl_arg;
1803 
1804 	kmem_free(dk_efi->dki_data, vd_efi->length);
1805 }
1806 
1807 static int
1808 vd_scsicmd_in(void *vd_buf, size_t vd_buf_len, void *ioctl_arg)
1809 {
1810 	size_t vd_scsi_len;
1811 	vd_scsi_t *vd_scsi = (vd_scsi_t *)vd_buf;
1812 	struct uscsi_cmd *uscsi = (struct uscsi_cmd *)ioctl_arg;
1813 
1814 	/* check buffer size */
1815 	vd_scsi_len = VD_SCSI_SIZE;
1816 	vd_scsi_len += P2ROUNDUP(vd_scsi->cdb_len, sizeof (uint64_t));
1817 	vd_scsi_len += P2ROUNDUP(vd_scsi->sense_len, sizeof (uint64_t));
1818 	vd_scsi_len += P2ROUNDUP(vd_scsi->datain_len, sizeof (uint64_t));
1819 	vd_scsi_len += P2ROUNDUP(vd_scsi->dataout_len, sizeof (uint64_t));
1820 
1821 	ASSERT(vd_scsi_len % sizeof (uint64_t) == 0);
1822 
1823 	if (vd_buf_len < vd_scsi_len)
1824 		return (EINVAL);
1825 
1826 	/* set flags */
1827 	uscsi->uscsi_flags = vd_scsi_debug;
1828 
1829 	if (vd_scsi->options & VD_SCSI_OPT_NORETRY) {
1830 		uscsi->uscsi_flags |= USCSI_ISOLATE;
1831 		uscsi->uscsi_flags |= USCSI_DIAGNOSE;
1832 	}
1833 
1834 	/* task attribute */
1835 	switch (vd_scsi->task_attribute) {
1836 	case VD_SCSI_TASK_ACA:
1837 		uscsi->uscsi_flags |= USCSI_HEAD;
1838 		break;
1839 	case VD_SCSI_TASK_HQUEUE:
1840 		uscsi->uscsi_flags |= USCSI_HTAG;
1841 		break;
1842 	case VD_SCSI_TASK_ORDERED:
1843 		uscsi->uscsi_flags |= USCSI_OTAG;
1844 		break;
1845 	default:
1846 		uscsi->uscsi_flags |= USCSI_NOTAG;
1847 		break;
1848 	}
1849 
1850 	/* timeout */
1851 	uscsi->uscsi_timeout = vd_scsi->timeout;
1852 
1853 	/* cdb data */
1854 	uscsi->uscsi_cdb = (caddr_t)VD_SCSI_DATA_CDB(vd_scsi);
1855 	uscsi->uscsi_cdblen = vd_scsi->cdb_len;
1856 
1857 	/* sense buffer */
1858 	if (vd_scsi->sense_len != 0) {
1859 		uscsi->uscsi_flags |= USCSI_RQENABLE;
1860 		uscsi->uscsi_rqbuf = (caddr_t)VD_SCSI_DATA_SENSE(vd_scsi);
1861 		uscsi->uscsi_rqlen = vd_scsi->sense_len;
1862 	}
1863 
1864 	if (vd_scsi->datain_len != 0 && vd_scsi->dataout_len != 0) {
1865 		/* uscsi does not support read/write request */
1866 		return (EINVAL);
1867 	}
1868 
1869 	/* request data-in */
1870 	if (vd_scsi->datain_len != 0) {
1871 		uscsi->uscsi_flags |= USCSI_READ;
1872 		uscsi->uscsi_buflen = vd_scsi->datain_len;
1873 		uscsi->uscsi_bufaddr = (char *)VD_SCSI_DATA_IN(vd_scsi);
1874 	}
1875 
1876 	/* request data-out */
1877 	if (vd_scsi->dataout_len != 0) {
1878 		uscsi->uscsi_buflen = vd_scsi->dataout_len;
1879 		uscsi->uscsi_bufaddr = (char *)VD_SCSI_DATA_OUT(vd_scsi);
1880 	}
1881 
1882 	return (0);
1883 }
1884 
1885 static void
1886 vd_scsicmd_out(void *ioctl_arg, void *vd_buf)
1887 {
1888 	vd_scsi_t *vd_scsi = (vd_scsi_t *)vd_buf;
1889 	struct uscsi_cmd *uscsi = (struct uscsi_cmd *)ioctl_arg;
1890 
1891 	/* output fields */
1892 	vd_scsi->cmd_status = uscsi->uscsi_status;
1893 
1894 	/* sense data */
1895 	if ((uscsi->uscsi_flags & USCSI_RQENABLE) &&
1896 	    (uscsi->uscsi_status == STATUS_CHECK ||
1897 	    uscsi->uscsi_status == STATUS_TERMINATED)) {
1898 		vd_scsi->sense_status = uscsi->uscsi_rqstatus;
1899 		if (uscsi->uscsi_rqstatus == STATUS_GOOD)
1900 			vd_scsi->sense_len -= uscsi->uscsi_resid;
1901 		else
1902 			vd_scsi->sense_len = 0;
1903 	} else {
1904 		vd_scsi->sense_len = 0;
1905 	}
1906 
1907 	if (uscsi->uscsi_status != STATUS_GOOD) {
1908 		vd_scsi->dataout_len = 0;
1909 		vd_scsi->datain_len = 0;
1910 		return;
1911 	}
1912 
1913 	if (uscsi->uscsi_flags & USCSI_READ) {
1914 		/* request data (read) */
1915 		vd_scsi->datain_len -= uscsi->uscsi_resid;
1916 		vd_scsi->dataout_len = 0;
1917 	} else {
1918 		/* request data (write) */
1919 		vd_scsi->datain_len = 0;
1920 		vd_scsi->dataout_len -= uscsi->uscsi_resid;
1921 	}
1922 }
1923 
1924 static vd_disk_label_t
1925 vd_read_vtoc(vd_t *vd, struct vtoc *vtoc)
1926 {
1927 	int status, rval;
1928 	struct dk_gpt *efi;
1929 	size_t efi_len;
1930 
1931 	ASSERT(vd->ldi_handle[0] != NULL);
1932 
1933 	status = ldi_ioctl(vd->ldi_handle[0], DKIOCGVTOC, (intptr_t)vtoc,
1934 	    (vd->open_flags | FKIOCTL), kcred, &rval);
1935 
1936 	if (status == 0) {
1937 		return (VD_DISK_LABEL_VTOC);
1938 	} else if (status != ENOTSUP) {
1939 		PR0("ldi_ioctl(DKIOCGVTOC) returned error %d", status);
1940 		return (VD_DISK_LABEL_UNK);
1941 	}
1942 
1943 	status = vds_efi_alloc_and_read(vd->ldi_handle[0], &efi, &efi_len);
1944 
1945 	if (status) {
1946 		PR0("vds_efi_alloc_and_read returned error %d", status);
1947 		return (VD_DISK_LABEL_UNK);
1948 	}
1949 
1950 	vd_efi_to_vtoc(efi, vtoc);
1951 	vd_efi_free(efi, efi_len);
1952 
1953 	return (VD_DISK_LABEL_EFI);
1954 }
1955 
1956 static ushort_t
1957 vd_lbl2cksum(struct dk_label *label)
1958 {
1959 	int	count;
1960 	ushort_t sum, *sp;
1961 
1962 	count =	(sizeof (struct dk_label)) / (sizeof (short)) - 1;
1963 	sp = (ushort_t *)label;
1964 	sum = 0;
1965 	while (count--) {
1966 		sum ^= *sp++;
1967 	}
1968 
1969 	return (sum);
1970 }
1971 
1972 /*
1973  * Handle ioctls to a disk slice.
1974  *
1975  * Return Values
1976  *	0	- Indicates that there are no errors in disk operations
1977  *	ENOTSUP	- Unknown disk label type or unsupported DKIO ioctl
1978  *	EINVAL	- Not enough room to copy the EFI label
1979  *
1980  */
1981 static int
1982 vd_do_slice_ioctl(vd_t *vd, int cmd, void *ioctl_arg)
1983 {
1984 	dk_efi_t *dk_ioc;
1985 
1986 	switch (vd->vdisk_label) {
1987 
1988 	/* ioctls for a slice from a disk with a VTOC label */
1989 	case VD_DISK_LABEL_VTOC:
1990 
1991 		switch (cmd) {
1992 		case DKIOCGGEOM:
1993 			ASSERT(ioctl_arg != NULL);
1994 			bcopy(&vd->dk_geom, ioctl_arg, sizeof (vd->dk_geom));
1995 			return (0);
1996 		case DKIOCGVTOC:
1997 			ASSERT(ioctl_arg != NULL);
1998 			bcopy(&vd->vtoc, ioctl_arg, sizeof (vd->vtoc));
1999 			return (0);
2000 		default:
2001 			return (ENOTSUP);
2002 		}
2003 
2004 	/* ioctls for a slice from a disk with an EFI label */
2005 	case VD_DISK_LABEL_EFI:
2006 
2007 		switch (cmd) {
2008 		case DKIOCGETEFI:
2009 			ASSERT(ioctl_arg != NULL);
2010 			dk_ioc = (dk_efi_t *)ioctl_arg;
2011 			if (dk_ioc->dki_length < vd->dk_efi.dki_length)
2012 				return (EINVAL);
2013 			bcopy(vd->dk_efi.dki_data, dk_ioc->dki_data,
2014 			    vd->dk_efi.dki_length);
2015 			return (0);
2016 		default:
2017 			return (ENOTSUP);
2018 		}
2019 
2020 	default:
2021 		/* Unknown disk label type */
2022 		return (ENOTSUP);
2023 	}
2024 }
2025 
2026 /*
2027  * Function:
2028  *	vd_file_validate_geometry
2029  *
2030  * Description:
2031  *	Read the label and validate the geometry of a disk image. The driver
2032  *	label, vtoc and geometry information are updated according to the
2033  *	label read from the disk image.
2034  *
2035  *	If no valid label is found, the label is set to unknown and the
2036  *	function returns EINVAL, but a default vtoc and geometry are provided
2037  *	to the driver.
2038  *
2039  * Parameters:
2040  *	vd	- disk on which the operation is performed.
2041  *
2042  * Return Code:
2043  *	0	- success.
2044  *	EIO	- error reading the label from the disk image.
2045  *	EINVAL	- unknown disk label.
2046  */
2047 static int
2048 vd_file_validate_geometry(vd_t *vd)
2049 {
2050 	struct dk_label label;
2051 	struct dk_geom *geom = &vd->dk_geom;
2052 	struct vtoc *vtoc = &vd->vtoc;
2053 	int i;
2054 	int status = 0;
2055 
2056 	ASSERT(vd->file);
2057 
2058 	if (vd->vdisk_type == VD_DISK_TYPE_SLICE) {
2059 		/*
2060 		 * For single slice disk we always fake the geometry, and we
2061 		 * only need to do it once because the geometry will never
2062 		 * change.
2063 		 */
2064 		if (vd->vdisk_label == VD_DISK_LABEL_VTOC)
2065 			/* geometry was already validated */
2066 			return (0);
2067 
2068 		ASSERT(vd->vdisk_label == VD_DISK_LABEL_UNK);
2069 		vd_file_build_default_label(vd, &label);
2070 		vd->vdisk_label = VD_DISK_LABEL_VTOC;
2071 	} else {
2072 		if (VD_FILE_LABEL_READ(vd, &label) < 0)
2073 			return (EIO);
2074 
2075 		if (label.dkl_magic != DKL_MAGIC ||
2076 		    label.dkl_cksum != vd_lbl2cksum(&label) ||
2077 		    label.dkl_vtoc.v_sanity != VTOC_SANE ||
2078 		    label.dkl_vtoc.v_nparts != V_NUMPAR) {
2079 			vd->vdisk_label = VD_DISK_LABEL_UNK;
2080 			vd_file_build_default_label(vd, &label);
2081 			status = EINVAL;
2082 		} else {
2083 			vd->vdisk_label = VD_DISK_LABEL_VTOC;
2084 		}
2085 	}
2086 
2087 	/* Update the driver geometry */
2088 	bzero(geom, sizeof (struct dk_geom));
2089 
2090 	geom->dkg_ncyl = label.dkl_ncyl;
2091 	geom->dkg_acyl = label.dkl_acyl;
2092 	geom->dkg_nhead = label.dkl_nhead;
2093 	geom->dkg_nsect = label.dkl_nsect;
2094 	geom->dkg_intrlv = label.dkl_intrlv;
2095 	geom->dkg_apc = label.dkl_apc;
2096 	geom->dkg_rpm = label.dkl_rpm;
2097 	geom->dkg_pcyl = label.dkl_pcyl;
2098 	geom->dkg_write_reinstruct = label.dkl_write_reinstruct;
2099 	geom->dkg_read_reinstruct = label.dkl_read_reinstruct;
2100 
2101 	/* Update the driver vtoc */
2102 	bzero(vtoc, sizeof (struct vtoc));
2103 
2104 	vtoc->v_sanity = label.dkl_vtoc.v_sanity;
2105 	vtoc->v_version = label.dkl_vtoc.v_version;
2106 	vtoc->v_sectorsz = DEV_BSIZE;
2107 	vtoc->v_nparts = label.dkl_vtoc.v_nparts;
2108 
2109 	for (i = 0; i < vtoc->v_nparts; i++) {
2110 		vtoc->v_part[i].p_tag =
2111 		    label.dkl_vtoc.v_part[i].p_tag;
2112 		vtoc->v_part[i].p_flag =
2113 		    label.dkl_vtoc.v_part[i].p_flag;
2114 		vtoc->v_part[i].p_start =
2115 		    label.dkl_map[i].dkl_cylno *
2116 		    (label.dkl_nhead * label.dkl_nsect);
2117 		vtoc->v_part[i].p_size = label.dkl_map[i].dkl_nblk;
2118 		vtoc->timestamp[i] =
2119 		    label.dkl_vtoc.v_timestamp[i];
2120 	}
2121 	/*
2122 	 * The bootinfo array can not be copied with bcopy() because
2123 	 * elements are of type long in vtoc (so 64-bit) and of type
2124 	 * int in dk_vtoc (so 32-bit).
2125 	 */
2126 	vtoc->v_bootinfo[0] = label.dkl_vtoc.v_bootinfo[0];
2127 	vtoc->v_bootinfo[1] = label.dkl_vtoc.v_bootinfo[1];
2128 	vtoc->v_bootinfo[2] = label.dkl_vtoc.v_bootinfo[2];
2129 	bcopy(label.dkl_asciilabel, vtoc->v_asciilabel,
2130 	    LEN_DKL_ASCII);
2131 	bcopy(label.dkl_vtoc.v_volume, vtoc->v_volume,
2132 	    LEN_DKL_VVOL);
2133 
2134 	return (status);
2135 }
2136 
2137 /*
2138  * Handle ioctls to a disk image (file-based).
2139  *
2140  * Return Values
2141  *	0	- Indicates that there are no errors
2142  *	!= 0	- Disk operation returned an error
2143  */
2144 static int
2145 vd_do_file_ioctl(vd_t *vd, int cmd, void *ioctl_arg)
2146 {
2147 	struct dk_label label;
2148 	struct dk_geom *geom;
2149 	struct vtoc *vtoc;
2150 	int i, rc;
2151 
2152 	ASSERT(vd->file);
2153 
2154 	switch (cmd) {
2155 
2156 	case DKIOCGGEOM:
2157 		ASSERT(ioctl_arg != NULL);
2158 		geom = (struct dk_geom *)ioctl_arg;
2159 
2160 		rc = vd_file_validate_geometry(vd);
2161 		if (rc != 0 && rc != EINVAL) {
2162 			ASSERT(vd->vdisk_type != VD_DISK_TYPE_SLICE);
2163 			return (rc);
2164 		}
2165 
2166 		bcopy(&vd->dk_geom, geom, sizeof (struct dk_geom));
2167 		return (0);
2168 
2169 	case DKIOCGVTOC:
2170 		ASSERT(ioctl_arg != NULL);
2171 		vtoc = (struct vtoc *)ioctl_arg;
2172 
2173 		rc = vd_file_validate_geometry(vd);
2174 		if (rc != 0 && rc != EINVAL) {
2175 			ASSERT(vd->vdisk_type != VD_DISK_TYPE_SLICE);
2176 			return (rc);
2177 		}
2178 
2179 		bcopy(&vd->vtoc, vtoc, sizeof (struct vtoc));
2180 		return (0);
2181 
2182 	case DKIOCSGEOM:
2183 		ASSERT(ioctl_arg != NULL);
2184 		geom = (struct dk_geom *)ioctl_arg;
2185 
2186 		/* geometry can only be changed for full disk */
2187 		if (vd->vdisk_type != VD_DISK_TYPE_DISK)
2188 			return (ENOTSUP);
2189 
2190 		if (geom->dkg_nhead == 0 || geom->dkg_nsect == 0)
2191 			return (EINVAL);
2192 
2193 		/*
2194 		 * The current device geometry is not updated, just the driver
2195 		 * "notion" of it. The device geometry will be effectively
2196 		 * updated when a label is written to the device during a next
2197 		 * DKIOCSVTOC.
2198 		 */
2199 		bcopy(ioctl_arg, &vd->dk_geom, sizeof (vd->dk_geom));
2200 		return (0);
2201 
2202 	case DKIOCSVTOC:
2203 		ASSERT(ioctl_arg != NULL);
2204 		ASSERT(vd->dk_geom.dkg_nhead != 0 &&
2205 		    vd->dk_geom.dkg_nsect != 0);
2206 		vtoc = (struct vtoc *)ioctl_arg;
2207 
2208 		/* vtoc can only be changed for full disk */
2209 		if (vd->vdisk_type != VD_DISK_TYPE_DISK)
2210 			return (ENOTSUP);
2211 
2212 		if (vtoc->v_sanity != VTOC_SANE ||
2213 		    vtoc->v_sectorsz != DEV_BSIZE ||
2214 		    vtoc->v_nparts != V_NUMPAR)
2215 			return (EINVAL);
2216 
2217 		bzero(&label, sizeof (label));
2218 		label.dkl_ncyl = vd->dk_geom.dkg_ncyl;
2219 		label.dkl_acyl = vd->dk_geom.dkg_acyl;
2220 		label.dkl_pcyl = vd->dk_geom.dkg_pcyl;
2221 		label.dkl_nhead = vd->dk_geom.dkg_nhead;
2222 		label.dkl_nsect = vd->dk_geom.dkg_nsect;
2223 		label.dkl_intrlv = vd->dk_geom.dkg_intrlv;
2224 		label.dkl_apc = vd->dk_geom.dkg_apc;
2225 		label.dkl_rpm = vd->dk_geom.dkg_rpm;
2226 		label.dkl_write_reinstruct = vd->dk_geom.dkg_write_reinstruct;
2227 		label.dkl_read_reinstruct = vd->dk_geom.dkg_read_reinstruct;
2228 
2229 		label.dkl_vtoc.v_nparts = V_NUMPAR;
2230 		label.dkl_vtoc.v_sanity = VTOC_SANE;
2231 		label.dkl_vtoc.v_version = vtoc->v_version;
2232 		for (i = 0; i < V_NUMPAR; i++) {
2233 			label.dkl_vtoc.v_timestamp[i] =
2234 			    vtoc->timestamp[i];
2235 			label.dkl_vtoc.v_part[i].p_tag =
2236 			    vtoc->v_part[i].p_tag;
2237 			label.dkl_vtoc.v_part[i].p_flag =
2238 			    vtoc->v_part[i].p_flag;
2239 			label.dkl_map[i].dkl_cylno =
2240 			    vtoc->v_part[i].p_start /
2241 			    (label.dkl_nhead * label.dkl_nsect);
2242 			label.dkl_map[i].dkl_nblk =
2243 			    vtoc->v_part[i].p_size;
2244 		}
2245 		/*
2246 		 * The bootinfo array can not be copied with bcopy() because
2247 		 * elements are of type long in vtoc (so 64-bit) and of type
2248 		 * int in dk_vtoc (so 32-bit).
2249 		 */
2250 		label.dkl_vtoc.v_bootinfo[0] = vtoc->v_bootinfo[0];
2251 		label.dkl_vtoc.v_bootinfo[1] = vtoc->v_bootinfo[1];
2252 		label.dkl_vtoc.v_bootinfo[2] = vtoc->v_bootinfo[2];
2253 		bcopy(vtoc->v_asciilabel, label.dkl_asciilabel,
2254 		    LEN_DKL_ASCII);
2255 		bcopy(vtoc->v_volume, label.dkl_vtoc.v_volume,
2256 		    LEN_DKL_VVOL);
2257 
2258 		/* re-compute checksum */
2259 		label.dkl_magic = DKL_MAGIC;
2260 		label.dkl_cksum = vd_lbl2cksum(&label);
2261 
2262 		/* write label to the disk image */
2263 		if ((rc = vd_file_set_vtoc(vd, &label)) != 0)
2264 			return (rc);
2265 
2266 		/* check the geometry and update the driver info */
2267 		if ((rc = vd_file_validate_geometry(vd)) != 0)
2268 			return (rc);
2269 
2270 		/*
2271 		 * The disk geometry may have changed, so we need to write
2272 		 * the devid (if there is one) so that it is stored at the
2273 		 * right location.
2274 		 */
2275 		if (vd->file_devid != NULL &&
2276 		    vd_file_write_devid(vd, vd->file_devid) != 0) {
2277 			PR0("Fail to write devid");
2278 		}
2279 
2280 		return (0);
2281 
2282 	case DKIOCFLUSHWRITECACHE:
2283 		return (VOP_FSYNC(vd->file_vnode, FSYNC, kcred, NULL));
2284 
2285 	default:
2286 		return (ENOTSUP);
2287 	}
2288 }
2289 
2290 /*
2291  * Description:
2292  *	This is the function that processes the ioctl requests (farming it
2293  *	out to functions that handle slices, files or whole disks)
2294  *
2295  * Return Values
2296  *     0		- ioctl operation completed successfully
2297  *     != 0		- The LDC error value encountered
2298  *			  (propagated back up the call stack as a task error)
2299  *
2300  * Side Effect
2301  *     sets request->status to the return value of the ioctl function.
2302  */
2303 static int
2304 vd_do_ioctl(vd_t *vd, vd_dring_payload_t *request, void* buf, vd_ioctl_t *ioctl)
2305 {
2306 	int	rval = 0, status = 0;
2307 	size_t	nbytes = request->nbytes;	/* modifiable copy */
2308 
2309 
2310 	ASSERT(request->slice < vd->nslices);
2311 	PR0("Performing %s", ioctl->operation_name);
2312 
2313 	/* Get data from client and convert, if necessary */
2314 	if (ioctl->copyin != NULL)  {
2315 		ASSERT(nbytes != 0 && buf != NULL);
2316 		PR1("Getting \"arg\" data from client");
2317 		if ((status = ldc_mem_copy(vd->ldc_handle, buf, 0, &nbytes,
2318 		    request->cookie, request->ncookies,
2319 		    LDC_COPY_IN)) != 0) {
2320 			PR0("ldc_mem_copy() returned errno %d "
2321 			    "copying from client", status);
2322 			return (status);
2323 		}
2324 
2325 		/* Convert client's data, if necessary */
2326 		if (ioctl->copyin == VD_IDENTITY_IN) {
2327 			/* use client buffer */
2328 			ioctl->arg = buf;
2329 		} else {
2330 			/* convert client vdisk operation data to ioctl data */
2331 			status = (ioctl->copyin)(buf, nbytes,
2332 			    (void *)ioctl->arg);
2333 			if (status != 0) {
2334 				request->status = status;
2335 				return (0);
2336 			}
2337 		}
2338 	}
2339 
2340 	if (ioctl->operation == VD_OP_SCSICMD) {
2341 		struct uscsi_cmd *uscsi = (struct uscsi_cmd *)ioctl->arg;
2342 
2343 		/* check write permission */
2344 		if (!(vd->open_flags & FWRITE) &&
2345 		    !(uscsi->uscsi_flags & USCSI_READ)) {
2346 			PR0("uscsi fails because backend is opened read-only");
2347 			request->status = EROFS;
2348 			return (0);
2349 		}
2350 	}
2351 
2352 	/*
2353 	 * Handle single-slice block devices internally; otherwise, have the
2354 	 * real driver perform the ioctl()
2355 	 */
2356 	if (vd->file) {
2357 		request->status =
2358 		    vd_do_file_ioctl(vd, ioctl->cmd, (void *)ioctl->arg);
2359 
2360 	} else if (vd->vdisk_type == VD_DISK_TYPE_SLICE && !vd->pseudo) {
2361 		request->status =
2362 		    vd_do_slice_ioctl(vd, ioctl->cmd, (void *)ioctl->arg);
2363 
2364 	} else {
2365 		request->status = ldi_ioctl(vd->ldi_handle[request->slice],
2366 		    ioctl->cmd, (intptr_t)ioctl->arg, vd->open_flags | FKIOCTL,
2367 		    kcred, &rval);
2368 
2369 #ifdef DEBUG
2370 		if (rval != 0) {
2371 			PR0("%s set rval = %d, which is not being returned to"
2372 			    " client", ioctl->cmd_name, rval);
2373 		}
2374 #endif /* DEBUG */
2375 	}
2376 
2377 	if (request->status != 0) {
2378 		PR0("ioctl(%s) = errno %d", ioctl->cmd_name, request->status);
2379 		if (ioctl->operation == VD_OP_SCSICMD &&
2380 		    ((struct uscsi_cmd *)ioctl->arg)->uscsi_status != 0)
2381 			/*
2382 			 * USCSICMD has reported an error and the uscsi_status
2383 			 * field is not zero. This means that the SCSI command
2384 			 * has completed but it has an error. So we should
2385 			 * mark the VD operation has succesfully completed
2386 			 * and clients can check the SCSI status field for
2387 			 * SCSI errors.
2388 			 */
2389 			request->status = 0;
2390 		else
2391 			return (0);
2392 	}
2393 
2394 	/* Convert data and send to client, if necessary */
2395 	if (ioctl->copyout != NULL)  {
2396 		ASSERT(nbytes != 0 && buf != NULL);
2397 		PR1("Sending \"arg\" data to client");
2398 
2399 		/* Convert ioctl data to vdisk operation data, if necessary */
2400 		if (ioctl->copyout != VD_IDENTITY_OUT)
2401 			(ioctl->copyout)((void *)ioctl->arg, buf);
2402 
2403 		if ((status = ldc_mem_copy(vd->ldc_handle, buf, 0, &nbytes,
2404 		    request->cookie, request->ncookies,
2405 		    LDC_COPY_OUT)) != 0) {
2406 			PR0("ldc_mem_copy() returned errno %d "
2407 			    "copying to client", status);
2408 			return (status);
2409 		}
2410 	}
2411 
2412 	return (status);
2413 }
2414 
2415 #define	RNDSIZE(expr) P2ROUNDUP(sizeof (expr), sizeof (uint64_t))
2416 
2417 /*
2418  * Description:
2419  *	This generic function is called by the task queue to complete
2420  *	the processing of the tasks. The specific completion function
2421  *	is passed in as a field in the task pointer.
2422  *
2423  * Parameters:
2424  *	arg 	- opaque pointer to structure containing task to be completed
2425  *
2426  * Return Values
2427  *	None
2428  */
2429 static void
2430 vd_complete(void *arg)
2431 {
2432 	vd_task_t	*task = (vd_task_t *)arg;
2433 
2434 	ASSERT(task != NULL);
2435 	ASSERT(task->status == EINPROGRESS);
2436 	ASSERT(task->completef != NULL);
2437 
2438 	task->status = task->completef(task);
2439 	if (task->status)
2440 		PR0("%s: Error %d completing task", __func__, task->status);
2441 
2442 	/* Now notify the vDisk client */
2443 	vd_complete_notify(task);
2444 }
2445 
2446 static int
2447 vd_ioctl(vd_task_t *task)
2448 {
2449 	int			i, status;
2450 	void			*buf = NULL;
2451 	struct dk_geom		dk_geom = {0};
2452 	struct vtoc		vtoc = {0};
2453 	struct dk_efi		dk_efi = {0};
2454 	struct uscsi_cmd	uscsi = {0};
2455 	vd_t			*vd		= task->vd;
2456 	vd_dring_payload_t	*request	= task->request;
2457 	vd_ioctl_t		ioctl[] = {
2458 		/* Command (no-copy) operations */
2459 		{VD_OP_FLUSH, STRINGIZE(VD_OP_FLUSH), 0,
2460 		    DKIOCFLUSHWRITECACHE, STRINGIZE(DKIOCFLUSHWRITECACHE),
2461 		    NULL, NULL, NULL, B_TRUE},
2462 
2463 		/* "Get" (copy-out) operations */
2464 		{VD_OP_GET_WCE, STRINGIZE(VD_OP_GET_WCE), RNDSIZE(int),
2465 		    DKIOCGETWCE, STRINGIZE(DKIOCGETWCE),
2466 		    NULL, VD_IDENTITY_IN, VD_IDENTITY_OUT, B_FALSE},
2467 		{VD_OP_GET_DISKGEOM, STRINGIZE(VD_OP_GET_DISKGEOM),
2468 		    RNDSIZE(vd_geom_t),
2469 		    DKIOCGGEOM, STRINGIZE(DKIOCGGEOM),
2470 		    &dk_geom, NULL, dk_geom2vd_geom, B_FALSE},
2471 		{VD_OP_GET_VTOC, STRINGIZE(VD_OP_GET_VTOC), RNDSIZE(vd_vtoc_t),
2472 		    DKIOCGVTOC, STRINGIZE(DKIOCGVTOC),
2473 		    &vtoc, NULL, vtoc2vd_vtoc, B_FALSE},
2474 		{VD_OP_GET_EFI, STRINGIZE(VD_OP_GET_EFI), RNDSIZE(vd_efi_t),
2475 		    DKIOCGETEFI, STRINGIZE(DKIOCGETEFI),
2476 		    &dk_efi, vd_get_efi_in, vd_get_efi_out, B_FALSE},
2477 
2478 		/* "Set" (copy-in) operations */
2479 		{VD_OP_SET_WCE, STRINGIZE(VD_OP_SET_WCE), RNDSIZE(int),
2480 		    DKIOCSETWCE, STRINGIZE(DKIOCSETWCE),
2481 		    NULL, VD_IDENTITY_IN, VD_IDENTITY_OUT, B_TRUE},
2482 		{VD_OP_SET_DISKGEOM, STRINGIZE(VD_OP_SET_DISKGEOM),
2483 		    RNDSIZE(vd_geom_t),
2484 		    DKIOCSGEOM, STRINGIZE(DKIOCSGEOM),
2485 		    &dk_geom, vd_geom2dk_geom, NULL, B_TRUE},
2486 		{VD_OP_SET_VTOC, STRINGIZE(VD_OP_SET_VTOC), RNDSIZE(vd_vtoc_t),
2487 		    DKIOCSVTOC, STRINGIZE(DKIOCSVTOC),
2488 		    &vtoc, vd_vtoc2vtoc, NULL, B_TRUE},
2489 		{VD_OP_SET_EFI, STRINGIZE(VD_OP_SET_EFI), RNDSIZE(vd_efi_t),
2490 		    DKIOCSETEFI, STRINGIZE(DKIOCSETEFI),
2491 		    &dk_efi, vd_set_efi_in, vd_set_efi_out, B_TRUE},
2492 
2493 		{VD_OP_SCSICMD, STRINGIZE(VD_OP_SCSICMD), RNDSIZE(vd_scsi_t),
2494 		    USCSICMD, STRINGIZE(USCSICMD),
2495 		    &uscsi, vd_scsicmd_in, vd_scsicmd_out, B_FALSE},
2496 	};
2497 	size_t		nioctls = (sizeof (ioctl))/(sizeof (ioctl[0]));
2498 
2499 
2500 	ASSERT(vd != NULL);
2501 	ASSERT(request != NULL);
2502 	ASSERT(request->slice < vd->nslices);
2503 
2504 	/*
2505 	 * Determine ioctl corresponding to caller's "operation" and
2506 	 * validate caller's "nbytes"
2507 	 */
2508 	for (i = 0; i < nioctls; i++) {
2509 		if (request->operation == ioctl[i].operation) {
2510 			/* LDC memory operations require 8-byte multiples */
2511 			ASSERT(ioctl[i].nbytes % sizeof (uint64_t) == 0);
2512 
2513 			if (request->operation == VD_OP_GET_EFI ||
2514 			    request->operation == VD_OP_SET_EFI ||
2515 			    request->operation == VD_OP_SCSICMD) {
2516 				if (request->nbytes >= ioctl[i].nbytes)
2517 					break;
2518 				PR0("%s:  Expected at least nbytes = %lu, "
2519 				    "got %lu", ioctl[i].operation_name,
2520 				    ioctl[i].nbytes, request->nbytes);
2521 				return (EINVAL);
2522 			}
2523 
2524 			if (request->nbytes != ioctl[i].nbytes) {
2525 				PR0("%s:  Expected nbytes = %lu, got %lu",
2526 				    ioctl[i].operation_name, ioctl[i].nbytes,
2527 				    request->nbytes);
2528 				return (EINVAL);
2529 			}
2530 
2531 			break;
2532 		}
2533 	}
2534 	ASSERT(i < nioctls);	/* because "operation" already validated */
2535 
2536 	if (!(vd->open_flags & FWRITE) && ioctl[i].write) {
2537 		PR0("%s fails because backend is opened read-only",
2538 		    ioctl[i].operation_name);
2539 		request->status = EROFS;
2540 		return (0);
2541 	}
2542 
2543 	if (request->nbytes)
2544 		buf = kmem_zalloc(request->nbytes, KM_SLEEP);
2545 	status = vd_do_ioctl(vd, request, buf, &ioctl[i]);
2546 	if (request->nbytes)
2547 		kmem_free(buf, request->nbytes);
2548 
2549 	return (status);
2550 }
2551 
2552 static int
2553 vd_get_devid(vd_task_t *task)
2554 {
2555 	vd_t *vd = task->vd;
2556 	vd_dring_payload_t *request = task->request;
2557 	vd_devid_t *vd_devid;
2558 	impl_devid_t *devid;
2559 	int status, bufid_len, devid_len, len, sz;
2560 	int bufbytes;
2561 
2562 	PR1("Get Device ID, nbytes=%ld", request->nbytes);
2563 
2564 	if (vd->file) {
2565 		if (vd->file_devid == NULL) {
2566 			PR2("No Device ID");
2567 			request->status = ENOENT;
2568 			return (0);
2569 		} else {
2570 			sz = ddi_devid_sizeof(vd->file_devid);
2571 			devid = kmem_alloc(sz, KM_SLEEP);
2572 			bcopy(vd->file_devid, devid, sz);
2573 		}
2574 	} else {
2575 		if (ddi_lyr_get_devid(vd->dev[request->slice],
2576 		    (ddi_devid_t *)&devid) != DDI_SUCCESS) {
2577 			PR2("No Device ID");
2578 			request->status = ENOENT;
2579 			return (0);
2580 		}
2581 	}
2582 
2583 	bufid_len = request->nbytes - sizeof (vd_devid_t) + 1;
2584 	devid_len = DEVID_GETLEN(devid);
2585 
2586 	/*
2587 	 * Save the buffer size here for use in deallocation.
2588 	 * The actual number of bytes copied is returned in
2589 	 * the 'nbytes' field of the request structure.
2590 	 */
2591 	bufbytes = request->nbytes;
2592 
2593 	vd_devid = kmem_zalloc(bufbytes, KM_SLEEP);
2594 	vd_devid->length = devid_len;
2595 	vd_devid->type = DEVID_GETTYPE(devid);
2596 
2597 	len = (devid_len > bufid_len)? bufid_len : devid_len;
2598 
2599 	bcopy(devid->did_id, vd_devid->id, len);
2600 
2601 	request->status = 0;
2602 
2603 	/* LDC memory operations require 8-byte multiples */
2604 	ASSERT(request->nbytes % sizeof (uint64_t) == 0);
2605 
2606 	if ((status = ldc_mem_copy(vd->ldc_handle, (caddr_t)vd_devid, 0,
2607 	    &request->nbytes, request->cookie, request->ncookies,
2608 	    LDC_COPY_OUT)) != 0) {
2609 		PR0("ldc_mem_copy() returned errno %d copying to client",
2610 		    status);
2611 	}
2612 	PR1("post mem_copy: nbytes=%ld", request->nbytes);
2613 
2614 	kmem_free(vd_devid, bufbytes);
2615 	ddi_devid_free((ddi_devid_t)devid);
2616 
2617 	return (status);
2618 }
2619 
2620 static int
2621 vd_scsi_reset(vd_t *vd)
2622 {
2623 	int rval, status;
2624 	struct uscsi_cmd uscsi = { 0 };
2625 
2626 	uscsi.uscsi_flags = vd_scsi_debug | USCSI_RESET;
2627 	uscsi.uscsi_timeout = vd_scsi_rdwr_timeout;
2628 
2629 	status = ldi_ioctl(vd->ldi_handle[0], USCSICMD, (intptr_t)&uscsi,
2630 	    (vd->open_flags | FKIOCTL), kcred, &rval);
2631 
2632 	return (status);
2633 }
2634 
2635 static int
2636 vd_reset(vd_task_t *task)
2637 {
2638 	vd_t *vd = task->vd;
2639 	vd_dring_payload_t *request = task->request;
2640 
2641 	ASSERT(request->operation == VD_OP_RESET);
2642 	ASSERT(vd->scsi);
2643 
2644 	PR0("Performing VD_OP_RESET");
2645 
2646 	if (request->nbytes != 0) {
2647 		PR0("VD_OP_RESET:  Expected nbytes = 0, got %lu",
2648 		    request->nbytes);
2649 		return (EINVAL);
2650 	}
2651 
2652 	request->status = vd_scsi_reset(vd);
2653 
2654 	return (0);
2655 }
2656 
2657 static int
2658 vd_get_capacity(vd_task_t *task)
2659 {
2660 	int rv;
2661 	size_t nbytes;
2662 	vd_t *vd = task->vd;
2663 	vd_dring_payload_t *request = task->request;
2664 	vd_capacity_t vd_cap = { 0 };
2665 
2666 	ASSERT(request->operation == VD_OP_GET_CAPACITY);
2667 	ASSERT(vd->scsi);
2668 
2669 	PR0("Performing VD_OP_GET_CAPACITY");
2670 
2671 	nbytes = request->nbytes;
2672 
2673 	if (nbytes != RNDSIZE(vd_capacity_t)) {
2674 		PR0("VD_OP_GET_CAPACITY:  Expected nbytes = %lu, got %lu",
2675 		    RNDSIZE(vd_capacity_t), nbytes);
2676 		return (EINVAL);
2677 	}
2678 
2679 	if (vd->vdisk_size == VD_SIZE_UNKNOWN) {
2680 		if (vd_setup_mediainfo(vd) != 0)
2681 			ASSERT(vd->vdisk_size == VD_SIZE_UNKNOWN);
2682 	}
2683 
2684 	ASSERT(vd->vdisk_size != 0);
2685 
2686 	request->status = 0;
2687 
2688 	vd_cap.vdisk_block_size = vd->vdisk_block_size;
2689 	vd_cap.vdisk_size = vd->vdisk_size;
2690 
2691 	if ((rv = ldc_mem_copy(vd->ldc_handle, (char *)&vd_cap, 0, &nbytes,
2692 	    request->cookie, request->ncookies, LDC_COPY_OUT)) != 0) {
2693 		PR0("ldc_mem_copy() returned errno %d copying to client", rv);
2694 		return (rv);
2695 	}
2696 
2697 	return (0);
2698 }
2699 
2700 static int
2701 vd_get_access(vd_task_t *task)
2702 {
2703 	uint64_t access;
2704 	int rv, rval = 0;
2705 	size_t nbytes;
2706 	vd_t *vd = task->vd;
2707 	vd_dring_payload_t *request = task->request;
2708 
2709 	ASSERT(request->operation == VD_OP_GET_ACCESS);
2710 	ASSERT(vd->scsi);
2711 
2712 	PR0("Performing VD_OP_GET_ACCESS");
2713 
2714 	nbytes = request->nbytes;
2715 
2716 	if (nbytes != sizeof (uint64_t)) {
2717 		PR0("VD_OP_GET_ACCESS:  Expected nbytes = %lu, got %lu",
2718 		    sizeof (uint64_t), nbytes);
2719 		return (EINVAL);
2720 	}
2721 
2722 	request->status = ldi_ioctl(vd->ldi_handle[request->slice], MHIOCSTATUS,
2723 	    NULL, (vd->open_flags | FKIOCTL), kcred, &rval);
2724 
2725 	if (request->status != 0)
2726 		return (0);
2727 
2728 	access = (rval == 0)? VD_ACCESS_ALLOWED : VD_ACCESS_DENIED;
2729 
2730 	if ((rv = ldc_mem_copy(vd->ldc_handle, (char *)&access, 0, &nbytes,
2731 	    request->cookie, request->ncookies, LDC_COPY_OUT)) != 0) {
2732 		PR0("ldc_mem_copy() returned errno %d copying to client", rv);
2733 		return (rv);
2734 	}
2735 
2736 	return (0);
2737 }
2738 
2739 static int
2740 vd_set_access(vd_task_t *task)
2741 {
2742 	uint64_t flags;
2743 	int rv, rval;
2744 	size_t nbytes;
2745 	vd_t *vd = task->vd;
2746 	vd_dring_payload_t *request = task->request;
2747 
2748 	ASSERT(request->operation == VD_OP_SET_ACCESS);
2749 	ASSERT(vd->scsi);
2750 
2751 	nbytes = request->nbytes;
2752 
2753 	if (nbytes != sizeof (uint64_t)) {
2754 		PR0("VD_OP_SET_ACCESS:  Expected nbytes = %lu, got %lu",
2755 		    sizeof (uint64_t), nbytes);
2756 		return (EINVAL);
2757 	}
2758 
2759 	if ((rv = ldc_mem_copy(vd->ldc_handle, (char *)&flags, 0, &nbytes,
2760 	    request->cookie, request->ncookies, LDC_COPY_IN)) != 0) {
2761 		PR0("ldc_mem_copy() returned errno %d copying from client", rv);
2762 		return (rv);
2763 	}
2764 
2765 	if (flags == VD_ACCESS_SET_CLEAR) {
2766 		PR0("Performing VD_OP_SET_ACCESS (CLEAR)");
2767 		request->status = ldi_ioctl(vd->ldi_handle[request->slice],
2768 		    MHIOCRELEASE, NULL, (vd->open_flags | FKIOCTL), kcred,
2769 		    &rval);
2770 		if (request->status == 0)
2771 			vd->ownership = B_FALSE;
2772 		return (0);
2773 	}
2774 
2775 	/*
2776 	 * As per the VIO spec, the PREEMPT and PRESERVE flags are only valid
2777 	 * when the EXCLUSIVE flag is set.
2778 	 */
2779 	if (!(flags & VD_ACCESS_SET_EXCLUSIVE)) {
2780 		PR0("Invalid VD_OP_SET_ACCESS flags: 0x%lx", flags);
2781 		request->status = EINVAL;
2782 		return (0);
2783 	}
2784 
2785 	switch (flags & (VD_ACCESS_SET_PREEMPT | VD_ACCESS_SET_PRESERVE)) {
2786 
2787 	case VD_ACCESS_SET_PREEMPT | VD_ACCESS_SET_PRESERVE:
2788 		/*
2789 		 * Flags EXCLUSIVE and PREEMPT and PRESERVE. We have to
2790 		 * acquire exclusive access rights, preserve them and we
2791 		 * can use preemption. So we can use the MHIOCTKNOWN ioctl.
2792 		 */
2793 		PR0("Performing VD_OP_SET_ACCESS (EXCLUSIVE|PREEMPT|PRESERVE)");
2794 		request->status = ldi_ioctl(vd->ldi_handle[request->slice],
2795 		    MHIOCTKOWN, NULL, (vd->open_flags | FKIOCTL), kcred, &rval);
2796 		break;
2797 
2798 	case VD_ACCESS_SET_PRESERVE:
2799 		/*
2800 		 * Flags EXCLUSIVE and PRESERVE. We have to acquire exclusive
2801 		 * access rights and preserve them, but not preempt any other
2802 		 * host. So we need to use the MHIOCTKOWN ioctl to enable the
2803 		 * "preserve" feature but we can not called it directly
2804 		 * because it uses preemption. So before that, we use the
2805 		 * MHIOCQRESERVE ioctl to ensure we can get exclusive rights
2806 		 * without preempting anyone.
2807 		 */
2808 		PR0("Performing VD_OP_SET_ACCESS (EXCLUSIVE|PRESERVE)");
2809 		request->status = ldi_ioctl(vd->ldi_handle[request->slice],
2810 		    MHIOCQRESERVE, NULL, (vd->open_flags | FKIOCTL), kcred,
2811 		    &rval);
2812 		if (request->status != 0)
2813 			break;
2814 		request->status = ldi_ioctl(vd->ldi_handle[request->slice],
2815 		    MHIOCTKOWN, NULL, (vd->open_flags | FKIOCTL), kcred, &rval);
2816 		break;
2817 
2818 	case VD_ACCESS_SET_PREEMPT:
2819 		/*
2820 		 * Flags EXCLUSIVE and PREEMPT. We have to acquire exclusive
2821 		 * access rights and we can use preemption. So we try to do
2822 		 * a SCSI reservation, if it fails we reset the disk to clear
2823 		 * any reservation and we try to reserve again.
2824 		 */
2825 		PR0("Performing VD_OP_SET_ACCESS (EXCLUSIVE|PREEMPT)");
2826 		request->status = ldi_ioctl(vd->ldi_handle[request->slice],
2827 		    MHIOCQRESERVE, NULL, (vd->open_flags | FKIOCTL), kcred,
2828 		    &rval);
2829 		if (request->status == 0)
2830 			break;
2831 
2832 		/* reset the disk */
2833 		(void) vd_scsi_reset(vd);
2834 
2835 		/* try again even if the reset has failed */
2836 		request->status = ldi_ioctl(vd->ldi_handle[request->slice],
2837 		    MHIOCQRESERVE, NULL, (vd->open_flags | FKIOCTL), kcred,
2838 		    &rval);
2839 		break;
2840 
2841 	case 0:
2842 		/* Flag EXCLUSIVE only. Just issue a SCSI reservation */
2843 		PR0("Performing VD_OP_SET_ACCESS (EXCLUSIVE)");
2844 		request->status = ldi_ioctl(vd->ldi_handle[request->slice],
2845 		    MHIOCQRESERVE, NULL, (vd->open_flags | FKIOCTL), kcred,
2846 		    &rval);
2847 		break;
2848 	}
2849 
2850 	if (request->status == 0)
2851 		vd->ownership = B_TRUE;
2852 	else
2853 		PR0("VD_OP_SET_ACCESS: error %d", request->status);
2854 
2855 	return (0);
2856 }
2857 
2858 static void
2859 vd_reset_access(vd_t *vd)
2860 {
2861 	int status, rval;
2862 
2863 	if (vd->file || !vd->ownership)
2864 		return;
2865 
2866 	PR0("Releasing disk ownership");
2867 	status = ldi_ioctl(vd->ldi_handle[0], MHIOCRELEASE, NULL,
2868 	    (vd->open_flags | FKIOCTL), kcred, &rval);
2869 
2870 	/*
2871 	 * An EACCES failure means that there is a reservation conflict,
2872 	 * so we are not the owner of the disk anymore.
2873 	 */
2874 	if (status == 0 || status == EACCES) {
2875 		vd->ownership = B_FALSE;
2876 		return;
2877 	}
2878 
2879 	PR0("Fail to release ownership, error %d", status);
2880 
2881 	/*
2882 	 * We have failed to release the ownership, try to reset the disk
2883 	 * to release reservations.
2884 	 */
2885 	PR0("Resetting disk");
2886 	status = vd_scsi_reset(vd);
2887 
2888 	if (status != 0)
2889 		PR0("Fail to reset disk, error %d", status);
2890 
2891 	/* whatever the result of the reset is, we try the release again */
2892 	status = ldi_ioctl(vd->ldi_handle[0], MHIOCRELEASE, NULL,
2893 	    (vd->open_flags | FKIOCTL), kcred, &rval);
2894 
2895 	if (status == 0 || status == EACCES) {
2896 		vd->ownership = B_FALSE;
2897 		return;
2898 	}
2899 
2900 	PR0("Fail to release ownership, error %d", status);
2901 
2902 	/*
2903 	 * At this point we have done our best to try to reset the
2904 	 * access rights to the disk and we don't know if we still
2905 	 * own a reservation and if any mechanism to preserve the
2906 	 * ownership is still in place. The ultimate solution would
2907 	 * be to reset the system but this is usually not what we
2908 	 * want to happen.
2909 	 */
2910 
2911 	if (vd_reset_access_failure == A_REBOOT) {
2912 		cmn_err(CE_WARN, VD_RESET_ACCESS_FAILURE_MSG
2913 		    ", rebooting the system", vd->device_path);
2914 		(void) uadmin(A_SHUTDOWN, AD_BOOT, NULL);
2915 	} else if (vd_reset_access_failure == A_DUMP) {
2916 		panic(VD_RESET_ACCESS_FAILURE_MSG, vd->device_path);
2917 	}
2918 
2919 	cmn_err(CE_WARN, VD_RESET_ACCESS_FAILURE_MSG, vd->device_path);
2920 }
2921 
2922 /*
2923  * Define the supported operations once the functions for performing them have
2924  * been defined
2925  */
2926 static const vds_operation_t	vds_operation[] = {
2927 #define	X(_s)	#_s, _s
2928 	{X(VD_OP_BREAD),	vd_start_bio,	vd_complete_bio},
2929 	{X(VD_OP_BWRITE),	vd_start_bio,	vd_complete_bio},
2930 	{X(VD_OP_FLUSH),	vd_ioctl,	NULL},
2931 	{X(VD_OP_GET_WCE),	vd_ioctl,	NULL},
2932 	{X(VD_OP_SET_WCE),	vd_ioctl,	NULL},
2933 	{X(VD_OP_GET_VTOC),	vd_ioctl,	NULL},
2934 	{X(VD_OP_SET_VTOC),	vd_ioctl,	NULL},
2935 	{X(VD_OP_GET_DISKGEOM),	vd_ioctl,	NULL},
2936 	{X(VD_OP_SET_DISKGEOM),	vd_ioctl,	NULL},
2937 	{X(VD_OP_GET_EFI),	vd_ioctl,	NULL},
2938 	{X(VD_OP_SET_EFI),	vd_ioctl,	NULL},
2939 	{X(VD_OP_GET_DEVID),	vd_get_devid,	NULL},
2940 	{X(VD_OP_SCSICMD),	vd_ioctl,	NULL},
2941 	{X(VD_OP_RESET),	vd_reset,	NULL},
2942 	{X(VD_OP_GET_CAPACITY),	vd_get_capacity, NULL},
2943 	{X(VD_OP_SET_ACCESS),	vd_set_access,	NULL},
2944 	{X(VD_OP_GET_ACCESS),	vd_get_access,	NULL},
2945 #undef	X
2946 };
2947 
2948 static const size_t	vds_noperations =
2949 	(sizeof (vds_operation))/(sizeof (vds_operation[0]));
2950 
2951 /*
2952  * Process a task specifying a client I/O request
2953  *
2954  * Parameters:
2955  *	task 		- structure containing the request sent from client
2956  *
2957  * Return Value
2958  *	0	- success
2959  *	ENOTSUP	- Unknown/Unsupported VD_OP_XXX operation
2960  *	EINVAL	- Invalid disk slice
2961  *	!= 0	- some other non-zero return value from start function
2962  */
2963 static int
2964 vd_do_process_task(vd_task_t *task)
2965 {
2966 	int			i;
2967 	vd_t			*vd		= task->vd;
2968 	vd_dring_payload_t	*request	= task->request;
2969 
2970 	ASSERT(vd != NULL);
2971 	ASSERT(request != NULL);
2972 
2973 	/* Find the requested operation */
2974 	for (i = 0; i < vds_noperations; i++) {
2975 		if (request->operation == vds_operation[i].operation) {
2976 			/* all operations should have a start func */
2977 			ASSERT(vds_operation[i].start != NULL);
2978 
2979 			task->completef = vds_operation[i].complete;
2980 			break;
2981 		}
2982 	}
2983 
2984 	/*
2985 	 * We need to check that the requested operation is permitted
2986 	 * for the particular client that sent it or that the loop above
2987 	 * did not complete without finding the operation type (indicating
2988 	 * that the requested operation is unknown/unimplemented)
2989 	 */
2990 	if ((VD_OP_SUPPORTED(vd->operations, request->operation) == B_FALSE) ||
2991 	    (i == vds_noperations)) {
2992 		PR0("Unsupported operation %u", request->operation);
2993 		request->status = ENOTSUP;
2994 		return (0);
2995 	}
2996 
2997 	/* Range-check slice */
2998 	if (request->slice >= vd->nslices &&
2999 	    (vd->vdisk_type != VD_DISK_TYPE_DISK ||
3000 	    request->slice != VD_SLICE_NONE)) {
3001 		PR0("Invalid \"slice\" %u (max %u) for virtual disk",
3002 		    request->slice, (vd->nslices - 1));
3003 		return (EINVAL);
3004 	}
3005 
3006 	/*
3007 	 * Call the function pointer that starts the operation.
3008 	 */
3009 	return (vds_operation[i].start(task));
3010 }
3011 
3012 /*
3013  * Description:
3014  *	This function is called by both the in-band and descriptor ring
3015  *	message processing functions paths to actually execute the task
3016  *	requested by the vDisk client. It in turn calls its worker
3017  *	function, vd_do_process_task(), to carry our the request.
3018  *
3019  *	Any transport errors (e.g. LDC errors, vDisk protocol errors) are
3020  *	saved in the 'status' field of the task and are propagated back
3021  *	up the call stack to trigger a NACK
3022  *
3023  *	Any request errors (e.g. ENOTTY from an ioctl) are saved in
3024  *	the 'status' field of the request and result in an ACK being sent
3025  *	by the completion handler.
3026  *
3027  * Parameters:
3028  *	task 		- structure containing the request sent from client
3029  *
3030  * Return Value
3031  *	0		- successful synchronous request.
3032  *	!= 0		- transport error (e.g. LDC errors, vDisk protocol)
3033  *	EINPROGRESS	- task will be finished in a completion handler
3034  */
3035 static int
3036 vd_process_task(vd_task_t *task)
3037 {
3038 	vd_t	*vd = task->vd;
3039 	int	status;
3040 
3041 	DTRACE_PROBE1(task__start, vd_task_t *, task);
3042 
3043 	task->status =  vd_do_process_task(task);
3044 
3045 	/*
3046 	 * If the task processing function returned EINPROGRESS indicating
3047 	 * that the task needs completing then schedule a taskq entry to
3048 	 * finish it now.
3049 	 *
3050 	 * Otherwise the task processing function returned either zero
3051 	 * indicating that the task was finished in the start function (and we
3052 	 * don't need to wait in a completion function) or the start function
3053 	 * returned an error - in both cases all that needs to happen is the
3054 	 * notification to the vDisk client higher up the call stack.
3055 	 * If the task was using a Descriptor Ring, we need to mark it as done
3056 	 * at this stage.
3057 	 */
3058 	if (task->status == EINPROGRESS) {
3059 		/* Queue a task to complete the operation */
3060 		(void) ddi_taskq_dispatch(vd->completionq, vd_complete,
3061 		    task, DDI_SLEEP);
3062 
3063 	} else if (!vd->reset_state && (vd->xfer_mode == VIO_DRING_MODE)) {
3064 		/* Update the dring element if it's a dring client */
3065 		status = vd_mark_elem_done(vd, task->index,
3066 		    task->request->status, task->request->nbytes);
3067 		if (status == ECONNRESET)
3068 			vd_mark_in_reset(vd);
3069 	}
3070 
3071 	return (task->status);
3072 }
3073 
3074 /*
3075  * Return true if the "type", "subtype", and "env" fields of the "tag" first
3076  * argument match the corresponding remaining arguments; otherwise, return false
3077  */
3078 boolean_t
3079 vd_msgtype(vio_msg_tag_t *tag, int type, int subtype, int env)
3080 {
3081 	return ((tag->vio_msgtype == type) &&
3082 	    (tag->vio_subtype == subtype) &&
3083 	    (tag->vio_subtype_env == env)) ? B_TRUE : B_FALSE;
3084 }
3085 
3086 /*
3087  * Check whether the major/minor version specified in "ver_msg" is supported
3088  * by this server.
3089  */
3090 static boolean_t
3091 vds_supported_version(vio_ver_msg_t *ver_msg)
3092 {
3093 	for (int i = 0; i < vds_num_versions; i++) {
3094 		ASSERT(vds_version[i].major > 0);
3095 		ASSERT((i == 0) ||
3096 		    (vds_version[i].major < vds_version[i-1].major));
3097 
3098 		/*
3099 		 * If the major versions match, adjust the minor version, if
3100 		 * necessary, down to the highest value supported by this
3101 		 * server and return true so this message will get "ack"ed;
3102 		 * the client should also support all minor versions lower
3103 		 * than the value it sent
3104 		 */
3105 		if (ver_msg->ver_major == vds_version[i].major) {
3106 			if (ver_msg->ver_minor > vds_version[i].minor) {
3107 				PR0("Adjusting minor version from %u to %u",
3108 				    ver_msg->ver_minor, vds_version[i].minor);
3109 				ver_msg->ver_minor = vds_version[i].minor;
3110 			}
3111 			return (B_TRUE);
3112 		}
3113 
3114 		/*
3115 		 * If the message contains a higher major version number, set
3116 		 * the message's major/minor versions to the current values
3117 		 * and return false, so this message will get "nack"ed with
3118 		 * these values, and the client will potentially try again
3119 		 * with the same or a lower version
3120 		 */
3121 		if (ver_msg->ver_major > vds_version[i].major) {
3122 			ver_msg->ver_major = vds_version[i].major;
3123 			ver_msg->ver_minor = vds_version[i].minor;
3124 			return (B_FALSE);
3125 		}
3126 
3127 		/*
3128 		 * Otherwise, the message's major version is less than the
3129 		 * current major version, so continue the loop to the next
3130 		 * (lower) supported version
3131 		 */
3132 	}
3133 
3134 	/*
3135 	 * No common version was found; "ground" the version pair in the
3136 	 * message to terminate negotiation
3137 	 */
3138 	ver_msg->ver_major = 0;
3139 	ver_msg->ver_minor = 0;
3140 	return (B_FALSE);
3141 }
3142 
3143 /*
3144  * Process a version message from a client.  vds expects to receive version
3145  * messages from clients seeking service, but never issues version messages
3146  * itself; therefore, vds can ACK or NACK client version messages, but does
3147  * not expect to receive version-message ACKs or NACKs (and will treat such
3148  * messages as invalid).
3149  */
3150 static int
3151 vd_process_ver_msg(vd_t *vd, vio_msg_t *msg, size_t msglen)
3152 {
3153 	vio_ver_msg_t	*ver_msg = (vio_ver_msg_t *)msg;
3154 
3155 
3156 	ASSERT(msglen >= sizeof (msg->tag));
3157 
3158 	if (!vd_msgtype(&msg->tag, VIO_TYPE_CTRL, VIO_SUBTYPE_INFO,
3159 	    VIO_VER_INFO)) {
3160 		return (ENOMSG);	/* not a version message */
3161 	}
3162 
3163 	if (msglen != sizeof (*ver_msg)) {
3164 		PR0("Expected %lu-byte version message; "
3165 		    "received %lu bytes", sizeof (*ver_msg), msglen);
3166 		return (EBADMSG);
3167 	}
3168 
3169 	if (ver_msg->dev_class != VDEV_DISK) {
3170 		PR0("Expected device class %u (disk); received %u",
3171 		    VDEV_DISK, ver_msg->dev_class);
3172 		return (EBADMSG);
3173 	}
3174 
3175 	/*
3176 	 * We're talking to the expected kind of client; set our device class
3177 	 * for "ack/nack" back to the client
3178 	 */
3179 	ver_msg->dev_class = VDEV_DISK_SERVER;
3180 
3181 	/*
3182 	 * Check whether the (valid) version message specifies a version
3183 	 * supported by this server.  If the version is not supported, return
3184 	 * EBADMSG so the message will get "nack"ed; vds_supported_version()
3185 	 * will have updated the message with a supported version for the
3186 	 * client to consider
3187 	 */
3188 	if (!vds_supported_version(ver_msg))
3189 		return (EBADMSG);
3190 
3191 
3192 	/*
3193 	 * A version has been agreed upon; use the client's SID for
3194 	 * communication on this channel now
3195 	 */
3196 	ASSERT(!(vd->initialized & VD_SID));
3197 	vd->sid = ver_msg->tag.vio_sid;
3198 	vd->initialized |= VD_SID;
3199 
3200 	/*
3201 	 * Store the negotiated major and minor version values in the "vd" data
3202 	 * structure so that we can check if certain operations are supported
3203 	 * by the client.
3204 	 */
3205 	vd->version.major = ver_msg->ver_major;
3206 	vd->version.minor = ver_msg->ver_minor;
3207 
3208 	PR0("Using major version %u, minor version %u",
3209 	    ver_msg->ver_major, ver_msg->ver_minor);
3210 	return (0);
3211 }
3212 
3213 static void
3214 vd_set_exported_operations(vd_t *vd)
3215 {
3216 	vd->operations = 0;	/* clear field */
3217 
3218 	/*
3219 	 * We need to check from the highest version supported to the
3220 	 * lowest because versions with a higher minor number implicitly
3221 	 * support versions with a lower minor number.
3222 	 */
3223 	if (vio_ver_is_supported(vd->version, 1, 1)) {
3224 		ASSERT(vd->open_flags & FREAD);
3225 		vd->operations |= VD_OP_MASK_READ;
3226 
3227 		if (vd->open_flags & FWRITE)
3228 			vd->operations |= VD_OP_MASK_WRITE;
3229 
3230 		if (vd->scsi)
3231 			vd->operations |= VD_OP_MASK_SCSI;
3232 
3233 		if (vd->file && vd_file_is_iso_image(vd)) {
3234 			/*
3235 			 * can't write to ISO images, make sure that write
3236 			 * support is not set in case administrator did not
3237 			 * use "options=ro" when doing an ldm add-vdsdev
3238 			 */
3239 			vd->operations &= ~VD_OP_MASK_WRITE;
3240 		}
3241 	} else if (vio_ver_is_supported(vd->version, 1, 0)) {
3242 		vd->operations = VD_OP_MASK_READ | VD_OP_MASK_WRITE;
3243 	}
3244 
3245 	/* we should have already agreed on a version */
3246 	ASSERT(vd->operations != 0);
3247 }
3248 
3249 static int
3250 vd_process_attr_msg(vd_t *vd, vio_msg_t *msg, size_t msglen)
3251 {
3252 	vd_attr_msg_t	*attr_msg = (vd_attr_msg_t *)msg;
3253 	int		status, retry = 0;
3254 
3255 
3256 	ASSERT(msglen >= sizeof (msg->tag));
3257 
3258 	if (!vd_msgtype(&msg->tag, VIO_TYPE_CTRL, VIO_SUBTYPE_INFO,
3259 	    VIO_ATTR_INFO)) {
3260 		PR0("Message is not an attribute message");
3261 		return (ENOMSG);
3262 	}
3263 
3264 	if (msglen != sizeof (*attr_msg)) {
3265 		PR0("Expected %lu-byte attribute message; "
3266 		    "received %lu bytes", sizeof (*attr_msg), msglen);
3267 		return (EBADMSG);
3268 	}
3269 
3270 	if (attr_msg->max_xfer_sz == 0) {
3271 		PR0("Received maximum transfer size of 0 from client");
3272 		return (EBADMSG);
3273 	}
3274 
3275 	if ((attr_msg->xfer_mode != VIO_DESC_MODE) &&
3276 	    (attr_msg->xfer_mode != VIO_DRING_MODE)) {
3277 		PR0("Client requested unsupported transfer mode");
3278 		return (EBADMSG);
3279 	}
3280 
3281 	/*
3282 	 * check if the underlying disk is ready, if not try accessing
3283 	 * the device again. Open the vdisk device and extract info
3284 	 * about it, as this is needed to respond to the attr info msg
3285 	 */
3286 	if ((vd->initialized & VD_DISK_READY) == 0) {
3287 		PR0("Retry setting up disk (%s)", vd->device_path);
3288 		do {
3289 			status = vd_setup_vd(vd);
3290 			if (status != EAGAIN || ++retry > vds_dev_retries)
3291 				break;
3292 
3293 			/* incremental delay */
3294 			delay(drv_usectohz(vds_dev_delay));
3295 
3296 			/* if vdisk is no longer enabled - return error */
3297 			if (!vd_enabled(vd))
3298 				return (ENXIO);
3299 
3300 		} while (status == EAGAIN);
3301 
3302 		if (status)
3303 			return (ENXIO);
3304 
3305 		vd->initialized |= VD_DISK_READY;
3306 		ASSERT(vd->nslices > 0 && vd->nslices <= V_NUMPAR);
3307 		PR0("vdisk_type = %s, pseudo = %s, file = %s, nslices = %u",
3308 		    ((vd->vdisk_type == VD_DISK_TYPE_DISK) ? "disk" : "slice"),
3309 		    (vd->pseudo ? "yes" : "no"),
3310 		    (vd->file ? "yes" : "no"),
3311 		    vd->nslices);
3312 	}
3313 
3314 	/* Success:  valid message and transfer mode */
3315 	vd->xfer_mode = attr_msg->xfer_mode;
3316 
3317 	if (vd->xfer_mode == VIO_DESC_MODE) {
3318 
3319 		/*
3320 		 * The vd_dring_inband_msg_t contains one cookie; need room
3321 		 * for up to n-1 more cookies, where "n" is the number of full
3322 		 * pages plus possibly one partial page required to cover
3323 		 * "max_xfer_sz".  Add room for one more cookie if
3324 		 * "max_xfer_sz" isn't an integral multiple of the page size.
3325 		 * Must first get the maximum transfer size in bytes.
3326 		 */
3327 		size_t	max_xfer_bytes = attr_msg->vdisk_block_size ?
3328 		    attr_msg->vdisk_block_size*attr_msg->max_xfer_sz :
3329 		    attr_msg->max_xfer_sz;
3330 		size_t	max_inband_msglen =
3331 		    sizeof (vd_dring_inband_msg_t) +
3332 		    ((max_xfer_bytes/PAGESIZE +
3333 		    ((max_xfer_bytes % PAGESIZE) ? 1 : 0))*
3334 		    (sizeof (ldc_mem_cookie_t)));
3335 
3336 		/*
3337 		 * Set the maximum expected message length to
3338 		 * accommodate in-band-descriptor messages with all
3339 		 * their cookies
3340 		 */
3341 		vd->max_msglen = MAX(vd->max_msglen, max_inband_msglen);
3342 
3343 		/*
3344 		 * Initialize the data structure for processing in-band I/O
3345 		 * request descriptors
3346 		 */
3347 		vd->inband_task.vd	= vd;
3348 		vd->inband_task.msg	= kmem_alloc(vd->max_msglen, KM_SLEEP);
3349 		vd->inband_task.index	= 0;
3350 		vd->inband_task.type	= VD_FINAL_RANGE_TASK;	/* range == 1 */
3351 	}
3352 
3353 	/* Return the device's block size and max transfer size to the client */
3354 	attr_msg->vdisk_block_size	= vd->vdisk_block_size;
3355 	attr_msg->max_xfer_sz		= vd->max_xfer_sz;
3356 
3357 	attr_msg->vdisk_size = vd->vdisk_size;
3358 	attr_msg->vdisk_type = vd->vdisk_type;
3359 	attr_msg->vdisk_media = vd->vdisk_media;
3360 
3361 	/* Discover and save the list of supported VD_OP_XXX operations */
3362 	vd_set_exported_operations(vd);
3363 	attr_msg->operations = vd->operations;
3364 
3365 	PR0("%s", VD_CLIENT(vd));
3366 
3367 	ASSERT(vd->dring_task == NULL);
3368 
3369 	return (0);
3370 }
3371 
3372 static int
3373 vd_process_dring_reg_msg(vd_t *vd, vio_msg_t *msg, size_t msglen)
3374 {
3375 	int			status;
3376 	size_t			expected;
3377 	ldc_mem_info_t		dring_minfo;
3378 	vio_dring_reg_msg_t	*reg_msg = (vio_dring_reg_msg_t *)msg;
3379 
3380 
3381 	ASSERT(msglen >= sizeof (msg->tag));
3382 
3383 	if (!vd_msgtype(&msg->tag, VIO_TYPE_CTRL, VIO_SUBTYPE_INFO,
3384 	    VIO_DRING_REG)) {
3385 		PR0("Message is not a register-dring message");
3386 		return (ENOMSG);
3387 	}
3388 
3389 	if (msglen < sizeof (*reg_msg)) {
3390 		PR0("Expected at least %lu-byte register-dring message; "
3391 		    "received %lu bytes", sizeof (*reg_msg), msglen);
3392 		return (EBADMSG);
3393 	}
3394 
3395 	expected = sizeof (*reg_msg) +
3396 	    (reg_msg->ncookies - 1)*(sizeof (reg_msg->cookie[0]));
3397 	if (msglen != expected) {
3398 		PR0("Expected %lu-byte register-dring message; "
3399 		    "received %lu bytes", expected, msglen);
3400 		return (EBADMSG);
3401 	}
3402 
3403 	if (vd->initialized & VD_DRING) {
3404 		PR0("A dring was previously registered; only support one");
3405 		return (EBADMSG);
3406 	}
3407 
3408 	if (reg_msg->num_descriptors > INT32_MAX) {
3409 		PR0("reg_msg->num_descriptors = %u; must be <= %u (%s)",
3410 		    reg_msg->ncookies, INT32_MAX, STRINGIZE(INT32_MAX));
3411 		return (EBADMSG);
3412 	}
3413 
3414 	if (reg_msg->ncookies != 1) {
3415 		/*
3416 		 * In addition to fixing the assertion in the success case
3417 		 * below, supporting drings which require more than one
3418 		 * "cookie" requires increasing the value of vd->max_msglen
3419 		 * somewhere in the code path prior to receiving the message
3420 		 * which results in calling this function.  Note that without
3421 		 * making this change, the larger message size required to
3422 		 * accommodate multiple cookies cannot be successfully
3423 		 * received, so this function will not even get called.
3424 		 * Gracefully accommodating more dring cookies might
3425 		 * reasonably demand exchanging an additional attribute or
3426 		 * making a minor protocol adjustment
3427 		 */
3428 		PR0("reg_msg->ncookies = %u != 1", reg_msg->ncookies);
3429 		return (EBADMSG);
3430 	}
3431 
3432 	status = ldc_mem_dring_map(vd->ldc_handle, reg_msg->cookie,
3433 	    reg_msg->ncookies, reg_msg->num_descriptors,
3434 	    reg_msg->descriptor_size, LDC_DIRECT_MAP, &vd->dring_handle);
3435 	if (status != 0) {
3436 		PR0("ldc_mem_dring_map() returned errno %d", status);
3437 		return (status);
3438 	}
3439 
3440 	/*
3441 	 * To remove the need for this assertion, must call
3442 	 * ldc_mem_dring_nextcookie() successfully ncookies-1 times after a
3443 	 * successful call to ldc_mem_dring_map()
3444 	 */
3445 	ASSERT(reg_msg->ncookies == 1);
3446 
3447 	if ((status =
3448 	    ldc_mem_dring_info(vd->dring_handle, &dring_minfo)) != 0) {
3449 		PR0("ldc_mem_dring_info() returned errno %d", status);
3450 		if ((status = ldc_mem_dring_unmap(vd->dring_handle)) != 0)
3451 			PR0("ldc_mem_dring_unmap() returned errno %d", status);
3452 		return (status);
3453 	}
3454 
3455 	if (dring_minfo.vaddr == NULL) {
3456 		PR0("Descriptor ring virtual address is NULL");
3457 		return (ENXIO);
3458 	}
3459 
3460 
3461 	/* Initialize for valid message and mapped dring */
3462 	PR1("descriptor size = %u, dring length = %u",
3463 	    vd->descriptor_size, vd->dring_len);
3464 	vd->initialized |= VD_DRING;
3465 	vd->dring_ident = 1;	/* "There Can Be Only One" */
3466 	vd->dring = dring_minfo.vaddr;
3467 	vd->descriptor_size = reg_msg->descriptor_size;
3468 	vd->dring_len = reg_msg->num_descriptors;
3469 	reg_msg->dring_ident = vd->dring_ident;
3470 
3471 	/*
3472 	 * Allocate and initialize a "shadow" array of data structures for
3473 	 * tasks to process I/O requests in dring elements
3474 	 */
3475 	vd->dring_task =
3476 	    kmem_zalloc((sizeof (*vd->dring_task)) * vd->dring_len, KM_SLEEP);
3477 	for (int i = 0; i < vd->dring_len; i++) {
3478 		vd->dring_task[i].vd		= vd;
3479 		vd->dring_task[i].index		= i;
3480 		vd->dring_task[i].request	= &VD_DRING_ELEM(i)->payload;
3481 
3482 		status = ldc_mem_alloc_handle(vd->ldc_handle,
3483 		    &(vd->dring_task[i].mhdl));
3484 		if (status) {
3485 			PR0("ldc_mem_alloc_handle() returned err %d ", status);
3486 			return (ENXIO);
3487 		}
3488 
3489 		vd->dring_task[i].msg = kmem_alloc(vd->max_msglen, KM_SLEEP);
3490 	}
3491 
3492 	return (0);
3493 }
3494 
3495 static int
3496 vd_process_dring_unreg_msg(vd_t *vd, vio_msg_t *msg, size_t msglen)
3497 {
3498 	vio_dring_unreg_msg_t	*unreg_msg = (vio_dring_unreg_msg_t *)msg;
3499 
3500 
3501 	ASSERT(msglen >= sizeof (msg->tag));
3502 
3503 	if (!vd_msgtype(&msg->tag, VIO_TYPE_CTRL, VIO_SUBTYPE_INFO,
3504 	    VIO_DRING_UNREG)) {
3505 		PR0("Message is not an unregister-dring message");
3506 		return (ENOMSG);
3507 	}
3508 
3509 	if (msglen != sizeof (*unreg_msg)) {
3510 		PR0("Expected %lu-byte unregister-dring message; "
3511 		    "received %lu bytes", sizeof (*unreg_msg), msglen);
3512 		return (EBADMSG);
3513 	}
3514 
3515 	if (unreg_msg->dring_ident != vd->dring_ident) {
3516 		PR0("Expected dring ident %lu; received %lu",
3517 		    vd->dring_ident, unreg_msg->dring_ident);
3518 		return (EBADMSG);
3519 	}
3520 
3521 	return (0);
3522 }
3523 
3524 static int
3525 process_rdx_msg(vio_msg_t *msg, size_t msglen)
3526 {
3527 	ASSERT(msglen >= sizeof (msg->tag));
3528 
3529 	if (!vd_msgtype(&msg->tag, VIO_TYPE_CTRL, VIO_SUBTYPE_INFO, VIO_RDX)) {
3530 		PR0("Message is not an RDX message");
3531 		return (ENOMSG);
3532 	}
3533 
3534 	if (msglen != sizeof (vio_rdx_msg_t)) {
3535 		PR0("Expected %lu-byte RDX message; received %lu bytes",
3536 		    sizeof (vio_rdx_msg_t), msglen);
3537 		return (EBADMSG);
3538 	}
3539 
3540 	PR0("Valid RDX message");
3541 	return (0);
3542 }
3543 
3544 static int
3545 vd_check_seq_num(vd_t *vd, uint64_t seq_num)
3546 {
3547 	if ((vd->initialized & VD_SEQ_NUM) && (seq_num != vd->seq_num + 1)) {
3548 		PR0("Received seq_num %lu; expected %lu",
3549 		    seq_num, (vd->seq_num + 1));
3550 		PR0("initiating soft reset");
3551 		vd_need_reset(vd, B_FALSE);
3552 		return (1);
3553 	}
3554 
3555 	vd->seq_num = seq_num;
3556 	vd->initialized |= VD_SEQ_NUM;	/* superfluous after first time... */
3557 	return (0);
3558 }
3559 
3560 /*
3561  * Return the expected size of an inband-descriptor message with all the
3562  * cookies it claims to include
3563  */
3564 static size_t
3565 expected_inband_size(vd_dring_inband_msg_t *msg)
3566 {
3567 	return ((sizeof (*msg)) +
3568 	    (msg->payload.ncookies - 1)*(sizeof (msg->payload.cookie[0])));
3569 }
3570 
3571 /*
3572  * Process an in-band descriptor message:  used with clients like OBP, with
3573  * which vds exchanges descriptors within VIO message payloads, rather than
3574  * operating on them within a descriptor ring
3575  */
3576 static int
3577 vd_process_desc_msg(vd_t *vd, vio_msg_t *msg, size_t msglen)
3578 {
3579 	size_t			expected;
3580 	vd_dring_inband_msg_t	*desc_msg = (vd_dring_inband_msg_t *)msg;
3581 
3582 
3583 	ASSERT(msglen >= sizeof (msg->tag));
3584 
3585 	if (!vd_msgtype(&msg->tag, VIO_TYPE_DATA, VIO_SUBTYPE_INFO,
3586 	    VIO_DESC_DATA)) {
3587 		PR1("Message is not an in-band-descriptor message");
3588 		return (ENOMSG);
3589 	}
3590 
3591 	if (msglen < sizeof (*desc_msg)) {
3592 		PR0("Expected at least %lu-byte descriptor message; "
3593 		    "received %lu bytes", sizeof (*desc_msg), msglen);
3594 		return (EBADMSG);
3595 	}
3596 
3597 	if (msglen != (expected = expected_inband_size(desc_msg))) {
3598 		PR0("Expected %lu-byte descriptor message; "
3599 		    "received %lu bytes", expected, msglen);
3600 		return (EBADMSG);
3601 	}
3602 
3603 	if (vd_check_seq_num(vd, desc_msg->hdr.seq_num) != 0)
3604 		return (EBADMSG);
3605 
3606 	/*
3607 	 * Valid message:  Set up the in-band descriptor task and process the
3608 	 * request.  Arrange to acknowledge the client's message, unless an
3609 	 * error processing the descriptor task results in setting
3610 	 * VIO_SUBTYPE_NACK
3611 	 */
3612 	PR1("Valid in-band-descriptor message");
3613 	msg->tag.vio_subtype = VIO_SUBTYPE_ACK;
3614 
3615 	ASSERT(vd->inband_task.msg != NULL);
3616 
3617 	bcopy(msg, vd->inband_task.msg, msglen);
3618 	vd->inband_task.msglen	= msglen;
3619 
3620 	/*
3621 	 * The task request is now the payload of the message
3622 	 * that was just copied into the body of the task.
3623 	 */
3624 	desc_msg = (vd_dring_inband_msg_t *)vd->inband_task.msg;
3625 	vd->inband_task.request	= &desc_msg->payload;
3626 
3627 	return (vd_process_task(&vd->inband_task));
3628 }
3629 
3630 static int
3631 vd_process_element(vd_t *vd, vd_task_type_t type, uint32_t idx,
3632     vio_msg_t *msg, size_t msglen)
3633 {
3634 	int			status;
3635 	boolean_t		ready;
3636 	vd_dring_entry_t	*elem = VD_DRING_ELEM(idx);
3637 
3638 
3639 	/* Accept the updated dring element */
3640 	if ((status = ldc_mem_dring_acquire(vd->dring_handle, idx, idx)) != 0) {
3641 		PR0("ldc_mem_dring_acquire() returned errno %d", status);
3642 		return (status);
3643 	}
3644 	ready = (elem->hdr.dstate == VIO_DESC_READY);
3645 	if (ready) {
3646 		elem->hdr.dstate = VIO_DESC_ACCEPTED;
3647 	} else {
3648 		PR0("descriptor %u not ready", idx);
3649 		VD_DUMP_DRING_ELEM(elem);
3650 	}
3651 	if ((status = ldc_mem_dring_release(vd->dring_handle, idx, idx)) != 0) {
3652 		PR0("ldc_mem_dring_release() returned errno %d", status);
3653 		return (status);
3654 	}
3655 	if (!ready)
3656 		return (EBUSY);
3657 
3658 
3659 	/* Initialize a task and process the accepted element */
3660 	PR1("Processing dring element %u", idx);
3661 	vd->dring_task[idx].type	= type;
3662 
3663 	/* duplicate msg buf for cookies etc. */
3664 	bcopy(msg, vd->dring_task[idx].msg, msglen);
3665 
3666 	vd->dring_task[idx].msglen	= msglen;
3667 	return (vd_process_task(&vd->dring_task[idx]));
3668 }
3669 
3670 static int
3671 vd_process_element_range(vd_t *vd, int start, int end,
3672     vio_msg_t *msg, size_t msglen)
3673 {
3674 	int		i, n, nelem, status = 0;
3675 	boolean_t	inprogress = B_FALSE;
3676 	vd_task_type_t	type;
3677 
3678 
3679 	ASSERT(start >= 0);
3680 	ASSERT(end >= 0);
3681 
3682 	/*
3683 	 * Arrange to acknowledge the client's message, unless an error
3684 	 * processing one of the dring elements results in setting
3685 	 * VIO_SUBTYPE_NACK
3686 	 */
3687 	msg->tag.vio_subtype = VIO_SUBTYPE_ACK;
3688 
3689 	/*
3690 	 * Process the dring elements in the range
3691 	 */
3692 	nelem = ((end < start) ? end + vd->dring_len : end) - start + 1;
3693 	for (i = start, n = nelem; n > 0; i = (i + 1) % vd->dring_len, n--) {
3694 		((vio_dring_msg_t *)msg)->end_idx = i;
3695 		type = (n == 1) ? VD_FINAL_RANGE_TASK : VD_NONFINAL_RANGE_TASK;
3696 		status = vd_process_element(vd, type, i, msg, msglen);
3697 		if (status == EINPROGRESS)
3698 			inprogress = B_TRUE;
3699 		else if (status != 0)
3700 			break;
3701 	}
3702 
3703 	/*
3704 	 * If some, but not all, operations of a multi-element range are in
3705 	 * progress, wait for other operations to complete before returning
3706 	 * (which will result in "ack" or "nack" of the message).  Note that
3707 	 * all outstanding operations will need to complete, not just the ones
3708 	 * corresponding to the current range of dring elements; howevever, as
3709 	 * this situation is an error case, performance is less critical.
3710 	 */
3711 	if ((nelem > 1) && (status != EINPROGRESS) && inprogress)
3712 		ddi_taskq_wait(vd->completionq);
3713 
3714 	return (status);
3715 }
3716 
3717 static int
3718 vd_process_dring_msg(vd_t *vd, vio_msg_t *msg, size_t msglen)
3719 {
3720 	vio_dring_msg_t	*dring_msg = (vio_dring_msg_t *)msg;
3721 
3722 
3723 	ASSERT(msglen >= sizeof (msg->tag));
3724 
3725 	if (!vd_msgtype(&msg->tag, VIO_TYPE_DATA, VIO_SUBTYPE_INFO,
3726 	    VIO_DRING_DATA)) {
3727 		PR1("Message is not a dring-data message");
3728 		return (ENOMSG);
3729 	}
3730 
3731 	if (msglen != sizeof (*dring_msg)) {
3732 		PR0("Expected %lu-byte dring message; received %lu bytes",
3733 		    sizeof (*dring_msg), msglen);
3734 		return (EBADMSG);
3735 	}
3736 
3737 	if (vd_check_seq_num(vd, dring_msg->seq_num) != 0)
3738 		return (EBADMSG);
3739 
3740 	if (dring_msg->dring_ident != vd->dring_ident) {
3741 		PR0("Expected dring ident %lu; received ident %lu",
3742 		    vd->dring_ident, dring_msg->dring_ident);
3743 		return (EBADMSG);
3744 	}
3745 
3746 	if (dring_msg->start_idx >= vd->dring_len) {
3747 		PR0("\"start_idx\" = %u; must be less than %u",
3748 		    dring_msg->start_idx, vd->dring_len);
3749 		return (EBADMSG);
3750 	}
3751 
3752 	if ((dring_msg->end_idx < 0) ||
3753 	    (dring_msg->end_idx >= vd->dring_len)) {
3754 		PR0("\"end_idx\" = %u; must be >= 0 and less than %u",
3755 		    dring_msg->end_idx, vd->dring_len);
3756 		return (EBADMSG);
3757 	}
3758 
3759 	/* Valid message; process range of updated dring elements */
3760 	PR1("Processing descriptor range, start = %u, end = %u",
3761 	    dring_msg->start_idx, dring_msg->end_idx);
3762 	return (vd_process_element_range(vd, dring_msg->start_idx,
3763 	    dring_msg->end_idx, msg, msglen));
3764 }
3765 
3766 static int
3767 recv_msg(ldc_handle_t ldc_handle, void *msg, size_t *nbytes)
3768 {
3769 	int	retry, status;
3770 	size_t	size = *nbytes;
3771 
3772 
3773 	for (retry = 0, status = ETIMEDOUT;
3774 	    retry < vds_ldc_retries && status == ETIMEDOUT;
3775 	    retry++) {
3776 		PR1("ldc_read() attempt %d", (retry + 1));
3777 		*nbytes = size;
3778 		status = ldc_read(ldc_handle, msg, nbytes);
3779 	}
3780 
3781 	if (status) {
3782 		PR0("ldc_read() returned errno %d", status);
3783 		if (status != ECONNRESET)
3784 			return (ENOMSG);
3785 		return (status);
3786 	} else if (*nbytes == 0) {
3787 		PR1("ldc_read() returned 0 and no message read");
3788 		return (ENOMSG);
3789 	}
3790 
3791 	PR1("RCVD %lu-byte message", *nbytes);
3792 	return (0);
3793 }
3794 
3795 static int
3796 vd_do_process_msg(vd_t *vd, vio_msg_t *msg, size_t msglen)
3797 {
3798 	int		status;
3799 
3800 
3801 	PR1("Processing (%x/%x/%x) message", msg->tag.vio_msgtype,
3802 	    msg->tag.vio_subtype, msg->tag.vio_subtype_env);
3803 #ifdef	DEBUG
3804 	vd_decode_tag(msg);
3805 #endif
3806 
3807 	/*
3808 	 * Validate session ID up front, since it applies to all messages
3809 	 * once set
3810 	 */
3811 	if ((msg->tag.vio_sid != vd->sid) && (vd->initialized & VD_SID)) {
3812 		PR0("Expected SID %u, received %u", vd->sid,
3813 		    msg->tag.vio_sid);
3814 		return (EBADMSG);
3815 	}
3816 
3817 	PR1("\tWhile in state %d (%s)", vd->state, vd_decode_state(vd->state));
3818 
3819 	/*
3820 	 * Process the received message based on connection state
3821 	 */
3822 	switch (vd->state) {
3823 	case VD_STATE_INIT:	/* expect version message */
3824 		if ((status = vd_process_ver_msg(vd, msg, msglen)) != 0)
3825 			return (status);
3826 
3827 		/* Version negotiated, move to that state */
3828 		vd->state = VD_STATE_VER;
3829 		return (0);
3830 
3831 	case VD_STATE_VER:	/* expect attribute message */
3832 		if ((status = vd_process_attr_msg(vd, msg, msglen)) != 0)
3833 			return (status);
3834 
3835 		/* Attributes exchanged, move to that state */
3836 		vd->state = VD_STATE_ATTR;
3837 		return (0);
3838 
3839 	case VD_STATE_ATTR:
3840 		switch (vd->xfer_mode) {
3841 		case VIO_DESC_MODE:	/* expect RDX message */
3842 			if ((status = process_rdx_msg(msg, msglen)) != 0)
3843 				return (status);
3844 
3845 			/* Ready to receive in-band descriptors */
3846 			vd->state = VD_STATE_DATA;
3847 			return (0);
3848 
3849 		case VIO_DRING_MODE:	/* expect register-dring message */
3850 			if ((status =
3851 			    vd_process_dring_reg_msg(vd, msg, msglen)) != 0)
3852 				return (status);
3853 
3854 			/* One dring negotiated, move to that state */
3855 			vd->state = VD_STATE_DRING;
3856 			return (0);
3857 
3858 		default:
3859 			ASSERT("Unsupported transfer mode");
3860 			PR0("Unsupported transfer mode");
3861 			return (ENOTSUP);
3862 		}
3863 
3864 	case VD_STATE_DRING:	/* expect RDX, register-dring, or unreg-dring */
3865 		if ((status = process_rdx_msg(msg, msglen)) == 0) {
3866 			/* Ready to receive data */
3867 			vd->state = VD_STATE_DATA;
3868 			return (0);
3869 		} else if (status != ENOMSG) {
3870 			return (status);
3871 		}
3872 
3873 
3874 		/*
3875 		 * If another register-dring message is received, stay in
3876 		 * dring state in case the client sends RDX; although the
3877 		 * protocol allows multiple drings, this server does not
3878 		 * support using more than one
3879 		 */
3880 		if ((status =
3881 		    vd_process_dring_reg_msg(vd, msg, msglen)) != ENOMSG)
3882 			return (status);
3883 
3884 		/*
3885 		 * Acknowledge an unregister-dring message, but reset the
3886 		 * connection anyway:  Although the protocol allows
3887 		 * unregistering drings, this server cannot serve a vdisk
3888 		 * without its only dring
3889 		 */
3890 		status = vd_process_dring_unreg_msg(vd, msg, msglen);
3891 		return ((status == 0) ? ENOTSUP : status);
3892 
3893 	case VD_STATE_DATA:
3894 		switch (vd->xfer_mode) {
3895 		case VIO_DESC_MODE:	/* expect in-band-descriptor message */
3896 			return (vd_process_desc_msg(vd, msg, msglen));
3897 
3898 		case VIO_DRING_MODE:	/* expect dring-data or unreg-dring */
3899 			/*
3900 			 * Typically expect dring-data messages, so handle
3901 			 * them first
3902 			 */
3903 			if ((status = vd_process_dring_msg(vd, msg,
3904 			    msglen)) != ENOMSG)
3905 				return (status);
3906 
3907 			/*
3908 			 * Acknowledge an unregister-dring message, but reset
3909 			 * the connection anyway:  Although the protocol
3910 			 * allows unregistering drings, this server cannot
3911 			 * serve a vdisk without its only dring
3912 			 */
3913 			status = vd_process_dring_unreg_msg(vd, msg, msglen);
3914 			return ((status == 0) ? ENOTSUP : status);
3915 
3916 		default:
3917 			ASSERT("Unsupported transfer mode");
3918 			PR0("Unsupported transfer mode");
3919 			return (ENOTSUP);
3920 		}
3921 
3922 	default:
3923 		ASSERT("Invalid client connection state");
3924 		PR0("Invalid client connection state");
3925 		return (ENOTSUP);
3926 	}
3927 }
3928 
3929 static int
3930 vd_process_msg(vd_t *vd, vio_msg_t *msg, size_t msglen)
3931 {
3932 	int		status;
3933 	boolean_t	reset_ldc = B_FALSE;
3934 	vd_task_t	task;
3935 
3936 	/*
3937 	 * Check that the message is at least big enough for a "tag", so that
3938 	 * message processing can proceed based on tag-specified message type
3939 	 */
3940 	if (msglen < sizeof (vio_msg_tag_t)) {
3941 		PR0("Received short (%lu-byte) message", msglen);
3942 		/* Can't "nack" short message, so drop the big hammer */
3943 		PR0("initiating full reset");
3944 		vd_need_reset(vd, B_TRUE);
3945 		return (EBADMSG);
3946 	}
3947 
3948 	/*
3949 	 * Process the message
3950 	 */
3951 	switch (status = vd_do_process_msg(vd, msg, msglen)) {
3952 	case 0:
3953 		/* "ack" valid, successfully-processed messages */
3954 		msg->tag.vio_subtype = VIO_SUBTYPE_ACK;
3955 		break;
3956 
3957 	case EINPROGRESS:
3958 		/* The completion handler will "ack" or "nack" the message */
3959 		return (EINPROGRESS);
3960 	case ENOMSG:
3961 		PR0("Received unexpected message");
3962 		_NOTE(FALLTHROUGH);
3963 	case EBADMSG:
3964 	case ENOTSUP:
3965 		/* "transport" error will cause NACK of invalid messages */
3966 		msg->tag.vio_subtype = VIO_SUBTYPE_NACK;
3967 		break;
3968 
3969 	default:
3970 		/* "transport" error will cause NACK of invalid messages */
3971 		msg->tag.vio_subtype = VIO_SUBTYPE_NACK;
3972 		/* An LDC error probably occurred, so try resetting it */
3973 		reset_ldc = B_TRUE;
3974 		break;
3975 	}
3976 
3977 	PR1("\tResulting in state %d (%s)", vd->state,
3978 	    vd_decode_state(vd->state));
3979 
3980 	/* populate the task so we can dispatch it on the taskq */
3981 	task.vd = vd;
3982 	task.msg = msg;
3983 	task.msglen = msglen;
3984 
3985 	/*
3986 	 * Queue a task to send the notification that the operation completed.
3987 	 * We need to ensure that requests are responded to in the correct
3988 	 * order and since the taskq is processed serially this ordering
3989 	 * is maintained.
3990 	 */
3991 	(void) ddi_taskq_dispatch(vd->completionq, vd_serial_notify,
3992 	    &task, DDI_SLEEP);
3993 
3994 	/*
3995 	 * To ensure handshake negotiations do not happen out of order, such
3996 	 * requests that come through this path should not be done in parallel
3997 	 * so we need to wait here until the response is sent to the client.
3998 	 */
3999 	ddi_taskq_wait(vd->completionq);
4000 
4001 	/* Arrange to reset the connection for nack'ed or failed messages */
4002 	if ((status != 0) || reset_ldc) {
4003 		PR0("initiating %s reset",
4004 		    (reset_ldc) ? "full" : "soft");
4005 		vd_need_reset(vd, reset_ldc);
4006 	}
4007 
4008 	return (status);
4009 }
4010 
4011 static boolean_t
4012 vd_enabled(vd_t *vd)
4013 {
4014 	boolean_t	enabled;
4015 
4016 	mutex_enter(&vd->lock);
4017 	enabled = vd->enabled;
4018 	mutex_exit(&vd->lock);
4019 	return (enabled);
4020 }
4021 
4022 static void
4023 vd_recv_msg(void *arg)
4024 {
4025 	vd_t	*vd = (vd_t *)arg;
4026 	int	rv = 0, status = 0;
4027 
4028 	ASSERT(vd != NULL);
4029 
4030 	PR2("New task to receive incoming message(s)");
4031 
4032 
4033 	while (vd_enabled(vd) && status == 0) {
4034 		size_t		msglen, msgsize;
4035 		ldc_status_t	lstatus;
4036 
4037 		/*
4038 		 * Receive and process a message
4039 		 */
4040 		vd_reset_if_needed(vd);	/* can change vd->max_msglen */
4041 
4042 		/*
4043 		 * check if channel is UP - else break out of loop
4044 		 */
4045 		status = ldc_status(vd->ldc_handle, &lstatus);
4046 		if (lstatus != LDC_UP) {
4047 			PR0("channel not up (status=%d), exiting recv loop\n",
4048 			    lstatus);
4049 			break;
4050 		}
4051 
4052 		ASSERT(vd->max_msglen != 0);
4053 
4054 		msgsize = vd->max_msglen; /* stable copy for alloc/free */
4055 		msglen	= msgsize;	  /* actual len after recv_msg() */
4056 
4057 		status = recv_msg(vd->ldc_handle, vd->vio_msgp, &msglen);
4058 		switch (status) {
4059 		case 0:
4060 			rv = vd_process_msg(vd, (vio_msg_t *)vd->vio_msgp,
4061 			    msglen);
4062 			/* check if max_msglen changed */
4063 			if (msgsize != vd->max_msglen) {
4064 				PR0("max_msglen changed 0x%lx to 0x%lx bytes\n",
4065 				    msgsize, vd->max_msglen);
4066 				kmem_free(vd->vio_msgp, msgsize);
4067 				vd->vio_msgp =
4068 				    kmem_alloc(vd->max_msglen, KM_SLEEP);
4069 			}
4070 			if (rv == EINPROGRESS)
4071 				continue;
4072 			break;
4073 
4074 		case ENOMSG:
4075 			break;
4076 
4077 		case ECONNRESET:
4078 			PR0("initiating soft reset (ECONNRESET)\n");
4079 			vd_need_reset(vd, B_FALSE);
4080 			status = 0;
4081 			break;
4082 
4083 		default:
4084 			/* Probably an LDC failure; arrange to reset it */
4085 			PR0("initiating full reset (status=0x%x)", status);
4086 			vd_need_reset(vd, B_TRUE);
4087 			break;
4088 		}
4089 	}
4090 
4091 	PR2("Task finished");
4092 }
4093 
4094 static uint_t
4095 vd_handle_ldc_events(uint64_t event, caddr_t arg)
4096 {
4097 	vd_t	*vd = (vd_t *)(void *)arg;
4098 	int	status;
4099 
4100 	ASSERT(vd != NULL);
4101 
4102 	if (!vd_enabled(vd))
4103 		return (LDC_SUCCESS);
4104 
4105 	if (event & LDC_EVT_DOWN) {
4106 		PR0("LDC_EVT_DOWN: LDC channel went down");
4107 
4108 		vd_need_reset(vd, B_TRUE);
4109 		status = ddi_taskq_dispatch(vd->startq, vd_recv_msg, vd,
4110 		    DDI_SLEEP);
4111 		if (status == DDI_FAILURE) {
4112 			PR0("cannot schedule task to recv msg\n");
4113 			vd_need_reset(vd, B_TRUE);
4114 		}
4115 	}
4116 
4117 	if (event & LDC_EVT_RESET) {
4118 		PR0("LDC_EVT_RESET: LDC channel was reset");
4119 
4120 		if (vd->state != VD_STATE_INIT) {
4121 			PR0("scheduling full reset");
4122 			vd_need_reset(vd, B_FALSE);
4123 			status = ddi_taskq_dispatch(vd->startq, vd_recv_msg,
4124 			    vd, DDI_SLEEP);
4125 			if (status == DDI_FAILURE) {
4126 				PR0("cannot schedule task to recv msg\n");
4127 				vd_need_reset(vd, B_TRUE);
4128 			}
4129 
4130 		} else {
4131 			PR0("channel already reset, ignoring...\n");
4132 			PR0("doing ldc up...\n");
4133 			(void) ldc_up(vd->ldc_handle);
4134 		}
4135 
4136 		return (LDC_SUCCESS);
4137 	}
4138 
4139 	if (event & LDC_EVT_UP) {
4140 		PR0("EVT_UP: LDC is up\nResetting client connection state");
4141 		PR0("initiating soft reset");
4142 		vd_need_reset(vd, B_FALSE);
4143 		status = ddi_taskq_dispatch(vd->startq, vd_recv_msg,
4144 		    vd, DDI_SLEEP);
4145 		if (status == DDI_FAILURE) {
4146 			PR0("cannot schedule task to recv msg\n");
4147 			vd_need_reset(vd, B_TRUE);
4148 			return (LDC_SUCCESS);
4149 		}
4150 	}
4151 
4152 	if (event & LDC_EVT_READ) {
4153 		int	status;
4154 
4155 		PR1("New data available");
4156 		/* Queue a task to receive the new data */
4157 		status = ddi_taskq_dispatch(vd->startq, vd_recv_msg, vd,
4158 		    DDI_SLEEP);
4159 
4160 		if (status == DDI_FAILURE) {
4161 			PR0("cannot schedule task to recv msg\n");
4162 			vd_need_reset(vd, B_TRUE);
4163 		}
4164 	}
4165 
4166 	return (LDC_SUCCESS);
4167 }
4168 
4169 static uint_t
4170 vds_check_for_vd(mod_hash_key_t key, mod_hash_val_t *val, void *arg)
4171 {
4172 	_NOTE(ARGUNUSED(key, val))
4173 	(*((uint_t *)arg))++;
4174 	return (MH_WALK_TERMINATE);
4175 }
4176 
4177 
4178 static int
4179 vds_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
4180 {
4181 	uint_t	vd_present = 0;
4182 	minor_t	instance;
4183 	vds_t	*vds;
4184 
4185 
4186 	switch (cmd) {
4187 	case DDI_DETACH:
4188 		/* the real work happens below */
4189 		break;
4190 	case DDI_SUSPEND:
4191 		PR0("No action required for DDI_SUSPEND");
4192 		return (DDI_SUCCESS);
4193 	default:
4194 		PR0("Unrecognized \"cmd\"");
4195 		return (DDI_FAILURE);
4196 	}
4197 
4198 	ASSERT(cmd == DDI_DETACH);
4199 	instance = ddi_get_instance(dip);
4200 	if ((vds = ddi_get_soft_state(vds_state, instance)) == NULL) {
4201 		PR0("Could not get state for instance %u", instance);
4202 		ddi_soft_state_free(vds_state, instance);
4203 		return (DDI_FAILURE);
4204 	}
4205 
4206 	/* Do no detach when serving any vdisks */
4207 	mod_hash_walk(vds->vd_table, vds_check_for_vd, &vd_present);
4208 	if (vd_present) {
4209 		PR0("Not detaching because serving vdisks");
4210 		return (DDI_FAILURE);
4211 	}
4212 
4213 	PR0("Detaching");
4214 	if (vds->initialized & VDS_MDEG) {
4215 		(void) mdeg_unregister(vds->mdeg);
4216 		kmem_free(vds->ispecp->specp, sizeof (vds_prop_template));
4217 		kmem_free(vds->ispecp, sizeof (mdeg_node_spec_t));
4218 		vds->ispecp = NULL;
4219 		vds->mdeg = NULL;
4220 	}
4221 
4222 	if (vds->initialized & VDS_LDI)
4223 		(void) ldi_ident_release(vds->ldi_ident);
4224 	mod_hash_destroy_hash(vds->vd_table);
4225 	ddi_soft_state_free(vds_state, instance);
4226 	return (DDI_SUCCESS);
4227 }
4228 
4229 static boolean_t
4230 is_pseudo_device(dev_info_t *dip)
4231 {
4232 	dev_info_t	*parent, *root = ddi_root_node();
4233 
4234 
4235 	for (parent = ddi_get_parent(dip); (parent != NULL) && (parent != root);
4236 	    parent = ddi_get_parent(parent)) {
4237 		if (strcmp(ddi_get_name(parent), DEVI_PSEUDO_NEXNAME) == 0)
4238 			return (B_TRUE);
4239 	}
4240 
4241 	return (B_FALSE);
4242 }
4243 
4244 /*
4245  * Description:
4246  *	This function checks to see if the file being used as a
4247  *	virtual disk is an ISO image. An ISO image is a special
4248  *	case which can be booted/installed from like a CD/DVD
4249  *
4250  * Parameters:
4251  *	vd		- disk on which the operation is performed.
4252  *
4253  * Return Code:
4254  *	B_TRUE		- The file is an ISO 9660 compliant image
4255  *	B_FALSE		- just a regular disk image file
4256  */
4257 static boolean_t
4258 vd_file_is_iso_image(vd_t *vd)
4259 {
4260 	char	iso_buf[ISO_SECTOR_SIZE];
4261 	int	i, rv;
4262 	uint_t	sec;
4263 
4264 	ASSERT(vd->file);
4265 
4266 	/*
4267 	 * If we have already discovered and saved this info we can
4268 	 * short-circuit the check and avoid reading the file.
4269 	 */
4270 	if (vd->vdisk_media == VD_MEDIA_DVD || vd->vdisk_media == VD_MEDIA_CD)
4271 		return (B_TRUE);
4272 
4273 	/*
4274 	 * We wish to read the sector that should contain the 2nd ISO volume
4275 	 * descriptor. The second field in this descriptor is called the
4276 	 * Standard Identifier and is set to CD001 for a CD-ROM compliant
4277 	 * to the ISO 9660 standard.
4278 	 */
4279 	sec = (ISO_VOLDESC_SEC * ISO_SECTOR_SIZE) / vd->vdisk_block_size;
4280 	rv = vd_file_rw(vd, VD_SLICE_NONE, VD_OP_BREAD, (caddr_t)iso_buf,
4281 	    sec, ISO_SECTOR_SIZE);
4282 
4283 	if (rv < 0)
4284 		return (B_FALSE);
4285 
4286 	for (i = 0; i < ISO_ID_STRLEN; i++) {
4287 		if (ISO_STD_ID(iso_buf)[i] != ISO_ID_STRING[i])
4288 			return (B_FALSE);
4289 	}
4290 
4291 	return (B_TRUE);
4292 }
4293 
4294 /*
4295  * Description:
4296  *	This function checks to see if the virtual device is an ATAPI
4297  *	device. ATAPI devices use Group 1 Read/Write commands, so
4298  *	any USCSI calls vds makes need to take this into account.
4299  *
4300  * Parameters:
4301  *	vd		- disk on which the operation is performed.
4302  *
4303  * Return Code:
4304  *	B_TRUE		- The virtual disk is backed by an ATAPI device
4305  *	B_FALSE		- not an ATAPI device (presumably SCSI)
4306  */
4307 static boolean_t
4308 vd_is_atapi_device(vd_t *vd)
4309 {
4310 	boolean_t	is_atapi = B_FALSE;
4311 	char		*variantp;
4312 	int		rv;
4313 
4314 	ASSERT(vd->ldi_handle[0] != NULL);
4315 	ASSERT(!vd->file);
4316 
4317 	rv = ldi_prop_lookup_string(vd->ldi_handle[0],
4318 	    (LDI_DEV_T_ANY | DDI_PROP_DONTPASS), "variant", &variantp);
4319 	if (rv == DDI_PROP_SUCCESS) {
4320 		PR0("'variant' property exists for %s", vd->device_path);
4321 		if (strcmp(variantp, "atapi") == 0)
4322 			is_atapi = B_TRUE;
4323 		ddi_prop_free(variantp);
4324 	}
4325 
4326 	rv = ldi_prop_exists(vd->ldi_handle[0], LDI_DEV_T_ANY, "atapi");
4327 	if (rv) {
4328 		PR0("'atapi' property exists for %s", vd->device_path);
4329 		is_atapi = B_TRUE;
4330 	}
4331 
4332 	return (is_atapi);
4333 }
4334 
4335 static int
4336 vd_setup_mediainfo(vd_t *vd)
4337 {
4338 	int status, rval;
4339 	struct dk_minfo	dk_minfo;
4340 
4341 	ASSERT(vd->ldi_handle[0] != NULL);
4342 	ASSERT(vd->vdisk_block_size != 0);
4343 
4344 	if ((status = ldi_ioctl(vd->ldi_handle[0], DKIOCGMEDIAINFO,
4345 	    (intptr_t)&dk_minfo, (vd->open_flags | FKIOCTL),
4346 	    kcred, &rval)) != 0)
4347 		return (status);
4348 
4349 	ASSERT(dk_minfo.dki_lbsize % vd->vdisk_block_size == 0);
4350 
4351 	vd->block_size = dk_minfo.dki_lbsize;
4352 	vd->vdisk_size = (dk_minfo.dki_capacity * dk_minfo.dki_lbsize) /
4353 	    vd->vdisk_block_size;
4354 	vd->vdisk_media = DK_MEDIATYPE2VD_MEDIATYPE(dk_minfo.dki_media_type);
4355 	return (0);
4356 }
4357 
4358 static int
4359 vd_setup_full_disk(vd_t *vd)
4360 {
4361 	int		status;
4362 	major_t		major = getmajor(vd->dev[0]);
4363 	minor_t		minor = getminor(vd->dev[0]) - VD_ENTIRE_DISK_SLICE;
4364 
4365 	ASSERT(vd->vdisk_type == VD_DISK_TYPE_DISK);
4366 
4367 	vd->vdisk_block_size = DEV_BSIZE;
4368 
4369 	/*
4370 	 * At this point, vdisk_size is set to the size of partition 2 but
4371 	 * this does not represent the size of the disk because partition 2
4372 	 * may not cover the entire disk and its size does not include reserved
4373 	 * blocks. So we call vd_get_mediainfo to udpate this information and
4374 	 * set the block size and the media type of the disk.
4375 	 */
4376 	status = vd_setup_mediainfo(vd);
4377 
4378 	if (status != 0) {
4379 		if (!vd->scsi) {
4380 			/* unexpected failure */
4381 			PRN("ldi_ioctl(DKIOCGMEDIAINFO) returned errno %d",
4382 			    status);
4383 			return (status);
4384 		}
4385 
4386 		/*
4387 		 * The function can fail for SCSI disks which are present but
4388 		 * reserved by another system. In that case, we don't know the
4389 		 * size of the disk and the block size.
4390 		 */
4391 		vd->vdisk_size = VD_SIZE_UNKNOWN;
4392 		vd->block_size = 0;
4393 		vd->vdisk_media = VD_MEDIA_FIXED;
4394 	}
4395 
4396 	/* Move dev number and LDI handle to entire-disk-slice array elements */
4397 	vd->dev[VD_ENTIRE_DISK_SLICE]		= vd->dev[0];
4398 	vd->dev[0]				= 0;
4399 	vd->ldi_handle[VD_ENTIRE_DISK_SLICE]	= vd->ldi_handle[0];
4400 	vd->ldi_handle[0]			= NULL;
4401 
4402 	/* Initialize device numbers for remaining slices and open them */
4403 	for (int slice = 0; slice < vd->nslices; slice++) {
4404 		/*
4405 		 * Skip the entire-disk slice, as it's already open and its
4406 		 * device known
4407 		 */
4408 		if (slice == VD_ENTIRE_DISK_SLICE)
4409 			continue;
4410 		ASSERT(vd->dev[slice] == 0);
4411 		ASSERT(vd->ldi_handle[slice] == NULL);
4412 
4413 		/*
4414 		 * Construct the device number for the current slice
4415 		 */
4416 		vd->dev[slice] = makedevice(major, (minor + slice));
4417 
4418 		/*
4419 		 * Open all slices of the disk to serve them to the client.
4420 		 * Slices are opened exclusively to prevent other threads or
4421 		 * processes in the service domain from performing I/O to
4422 		 * slices being accessed by a client.  Failure to open a slice
4423 		 * results in vds not serving this disk, as the client could
4424 		 * attempt (and should be able) to access any slice immediately.
4425 		 * Any slices successfully opened before a failure will get
4426 		 * closed by vds_destroy_vd() as a result of the error returned
4427 		 * by this function.
4428 		 *
4429 		 * We need to do the open with FNDELAY so that opening an empty
4430 		 * slice does not fail.
4431 		 */
4432 		PR0("Opening device major %u, minor %u = slice %u",
4433 		    major, minor, slice);
4434 
4435 		/*
4436 		 * Try to open the device. This can fail for example if we are
4437 		 * opening an empty slice. So in case of a failure, we try the
4438 		 * open again but this time with the FNDELAY flag.
4439 		 */
4440 		status = ldi_open_by_dev(&vd->dev[slice], OTYP_BLK,
4441 		    vd->open_flags, kcred, &vd->ldi_handle[slice],
4442 		    vd->vds->ldi_ident);
4443 
4444 		if (status != 0) {
4445 			status = ldi_open_by_dev(&vd->dev[slice], OTYP_BLK,
4446 			    vd->open_flags | FNDELAY, kcred,
4447 			    &vd->ldi_handle[slice], vd->vds->ldi_ident);
4448 		}
4449 
4450 		if (status != 0) {
4451 			PRN("ldi_open_by_dev() returned errno %d "
4452 			    "for slice %u", status, slice);
4453 			/* vds_destroy_vd() will close any open slices */
4454 			vd->ldi_handle[slice] = NULL;
4455 			return (status);
4456 		}
4457 	}
4458 
4459 	return (0);
4460 }
4461 
4462 static int
4463 vd_setup_partition_vtoc(vd_t *vd)
4464 {
4465 	int rval, status;
4466 	char *device_path = vd->device_path;
4467 
4468 	status = ldi_ioctl(vd->ldi_handle[0], DKIOCGGEOM,
4469 	    (intptr_t)&vd->dk_geom, (vd->open_flags | FKIOCTL), kcred, &rval);
4470 
4471 	if (status != 0) {
4472 		PRN("ldi_ioctl(DKIOCGEOM) returned errno %d for %s",
4473 		    status, device_path);
4474 		return (status);
4475 	}
4476 
4477 	/* Initialize dk_geom structure for single-slice device */
4478 	if (vd->dk_geom.dkg_nsect == 0) {
4479 		PRN("%s geometry claims 0 sectors per track", device_path);
4480 		return (EIO);
4481 	}
4482 	if (vd->dk_geom.dkg_nhead == 0) {
4483 		PRN("%s geometry claims 0 heads", device_path);
4484 		return (EIO);
4485 	}
4486 	vd->dk_geom.dkg_ncyl = vd->vdisk_size / vd->dk_geom.dkg_nsect /
4487 	    vd->dk_geom.dkg_nhead;
4488 	vd->dk_geom.dkg_acyl = 0;
4489 	vd->dk_geom.dkg_pcyl = vd->dk_geom.dkg_ncyl + vd->dk_geom.dkg_acyl;
4490 
4491 
4492 	/* Initialize vtoc structure for single-slice device */
4493 	bcopy(VD_VOLUME_NAME, vd->vtoc.v_volume,
4494 	    MIN(sizeof (VD_VOLUME_NAME), sizeof (vd->vtoc.v_volume)));
4495 	bzero(vd->vtoc.v_part, sizeof (vd->vtoc.v_part));
4496 	vd->vtoc.v_nparts = 1;
4497 	vd->vtoc.v_part[0].p_tag = V_UNASSIGNED;
4498 	vd->vtoc.v_part[0].p_flag = 0;
4499 	vd->vtoc.v_part[0].p_start = 0;
4500 	vd->vtoc.v_part[0].p_size = vd->vdisk_size;
4501 	bcopy(VD_ASCIILABEL, vd->vtoc.v_asciilabel,
4502 	    MIN(sizeof (VD_ASCIILABEL), sizeof (vd->vtoc.v_asciilabel)));
4503 
4504 	return (0);
4505 }
4506 
4507 static int
4508 vd_setup_partition_efi(vd_t *vd)
4509 {
4510 	efi_gpt_t *gpt;
4511 	efi_gpe_t *gpe;
4512 	struct uuid uuid = EFI_RESERVED;
4513 	uint32_t crc;
4514 	int length;
4515 
4516 	length = sizeof (efi_gpt_t) + sizeof (efi_gpe_t);
4517 
4518 	gpt = kmem_zalloc(length, KM_SLEEP);
4519 	gpe = (efi_gpe_t *)(gpt + 1);
4520 
4521 	gpt->efi_gpt_Signature = LE_64(EFI_SIGNATURE);
4522 	gpt->efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT);
4523 	gpt->efi_gpt_HeaderSize = LE_32(sizeof (efi_gpt_t));
4524 	gpt->efi_gpt_FirstUsableLBA = LE_64(0ULL);
4525 	gpt->efi_gpt_LastUsableLBA = LE_64(vd->vdisk_size - 1);
4526 	gpt->efi_gpt_NumberOfPartitionEntries = LE_32(1);
4527 	gpt->efi_gpt_SizeOfPartitionEntry = LE_32(sizeof (efi_gpe_t));
4528 
4529 	UUID_LE_CONVERT(gpe->efi_gpe_PartitionTypeGUID, uuid);
4530 	gpe->efi_gpe_StartingLBA = gpt->efi_gpt_FirstUsableLBA;
4531 	gpe->efi_gpe_EndingLBA = gpt->efi_gpt_LastUsableLBA;
4532 
4533 	CRC32(crc, gpe, sizeof (efi_gpe_t), -1U, crc32_table);
4534 	gpt->efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc);
4535 
4536 	CRC32(crc, gpt, sizeof (efi_gpt_t), -1U, crc32_table);
4537 	gpt->efi_gpt_HeaderCRC32 = LE_32(~crc);
4538 
4539 	vd->dk_efi.dki_lba = 0;
4540 	vd->dk_efi.dki_length = length;
4541 	vd->dk_efi.dki_data = gpt;
4542 
4543 	return (0);
4544 }
4545 
4546 /*
4547  * Setup for a virtual disk whose backend is a file (exported as a single slice
4548  * or as a full disk) or a pseudo device (for example a ZFS, SVM or VxVM volume)
4549  * exported as a full disk. In these cases, the backend is accessed using the
4550  * vnode interface.
4551  */
4552 static int
4553 vd_setup_backend_vnode(vd_t *vd)
4554 {
4555 	int 		rval, status;
4556 	vattr_t		vattr;
4557 	dev_t		dev;
4558 	char		*file_path = vd->device_path;
4559 	char		dev_path[MAXPATHLEN + 1];
4560 	ldi_handle_t	lhandle;
4561 	struct dk_cinfo	dk_cinfo;
4562 
4563 	if ((status = vn_open(file_path, UIO_SYSSPACE, vd->open_flags | FOFFMAX,
4564 	    0, &vd->file_vnode, 0, 0)) != 0) {
4565 		PRN("vn_open(%s) = errno %d", file_path, status);
4566 		return (status);
4567 	}
4568 
4569 	/*
4570 	 * We set vd->file now so that vds_destroy_vd will take care of
4571 	 * closing the file and releasing the vnode in case of an error.
4572 	 */
4573 	vd->file = B_TRUE;
4574 
4575 	vattr.va_mask = AT_SIZE;
4576 	if ((status = VOP_GETATTR(vd->file_vnode, &vattr, 0, kcred, NULL))
4577 	    != 0) {
4578 		PRN("VOP_GETATTR(%s) = errno %d", file_path, status);
4579 		return (EIO);
4580 	}
4581 
4582 	vd->file_size = vattr.va_size;
4583 	/* size should be at least sizeof(dk_label) */
4584 	if (vd->file_size < sizeof (struct dk_label)) {
4585 		PRN("Size of file has to be at least %ld bytes",
4586 		    sizeof (struct dk_label));
4587 		return (EIO);
4588 	}
4589 
4590 	if (vd->file_vnode->v_flag & VNOMAP) {
4591 		PRN("File %s cannot be mapped", file_path);
4592 		return (EIO);
4593 	}
4594 
4595 	/*
4596 	 * Find and validate the geometry of a disk image. For a single slice
4597 	 * disk image, this will build a fake geometry and vtoc.
4598 	 */
4599 	status = vd_file_validate_geometry(vd);
4600 	if (status != 0 && status != EINVAL) {
4601 		PRN("Failed to read label from %s", file_path);
4602 		return (EIO);
4603 	}
4604 
4605 	/* sector size = block size = DEV_BSIZE */
4606 	vd->block_size = DEV_BSIZE;
4607 	vd->vdisk_block_size = DEV_BSIZE;
4608 	vd->vdisk_size = vd->file_size / DEV_BSIZE;
4609 	vd->max_xfer_sz = maxphys / DEV_BSIZE; /* default transfer size */
4610 
4611 	if (vd_file_is_iso_image(vd)) {
4612 		/*
4613 		 * Indicate whether to call this a CD or DVD from the size
4614 		 * of the ISO image (images for both drive types are stored
4615 		 * in the ISO-9600 format). CDs can store up to just under 1Gb
4616 		 */
4617 		if ((vd->vdisk_size * vd->vdisk_block_size) >
4618 		    (1024 * 1024 * 1024))
4619 			vd->vdisk_media = VD_MEDIA_DVD;
4620 		else
4621 			vd->vdisk_media = VD_MEDIA_CD;
4622 	} else {
4623 		vd->vdisk_media = VD_MEDIA_FIXED;
4624 	}
4625 
4626 	/*
4627 	 * Get max_xfer_sz from the device where the file is or from the device
4628 	 * itself if we have a pseudo device.
4629 	 */
4630 	dev_path[0] = '\0';
4631 
4632 	if (vd->pseudo) {
4633 		status = ldi_open_by_name(file_path, FREAD, kcred, &lhandle,
4634 		    vd->vds->ldi_ident);
4635 	} else {
4636 		dev = vd->file_vnode->v_vfsp->vfs_dev;
4637 		if (ddi_dev_pathname(dev, S_IFBLK, dev_path) == DDI_SUCCESS) {
4638 			PR0("underlying device = %s\n", dev_path);
4639 		}
4640 
4641 		status = ldi_open_by_dev(&dev, OTYP_BLK, FREAD, kcred, &lhandle,
4642 		    vd->vds->ldi_ident);
4643 	}
4644 
4645 	if (status != 0) {
4646 		PR0("ldi_open() returned errno %d for device %s",
4647 		    status, (dev_path[0] == '\0')? file_path : dev_path);
4648 	} else {
4649 		if ((status = ldi_ioctl(lhandle, DKIOCINFO,
4650 		    (intptr_t)&dk_cinfo, (vd->open_flags | FKIOCTL), kcred,
4651 		    &rval)) != 0) {
4652 			PR0("ldi_ioctl(DKIOCINFO) returned errno %d for %s",
4653 			    status, dev_path);
4654 		} else {
4655 			/*
4656 			 * Store the device's max transfer size for
4657 			 * return to the client
4658 			 */
4659 			vd->max_xfer_sz = dk_cinfo.dki_maxtransfer;
4660 		}
4661 
4662 		PR0("close the device %s", dev_path);
4663 		(void) ldi_close(lhandle, FREAD, kcred);
4664 	}
4665 
4666 	PR0("using file %s, dev %s, max_xfer = %u blks",
4667 	    file_path, dev_path, vd->max_xfer_sz);
4668 
4669 	/* Setup devid for the disk image */
4670 
4671 	if (vd->vdisk_type == VD_DISK_TYPE_SLICE)
4672 		return (0);
4673 
4674 	if (vd->vdisk_label != VD_DISK_LABEL_UNK) {
4675 
4676 		status = vd_file_read_devid(vd, &vd->file_devid);
4677 
4678 		if (status == 0) {
4679 			/* a valid devid was found */
4680 			return (0);
4681 		}
4682 
4683 		if (status != EINVAL) {
4684 			/*
4685 			 * There was an error while trying to read the devid.
4686 			 * So this disk image may have a devid but we are
4687 			 * unable to read it.
4688 			 */
4689 			PR0("can not read devid for %s", file_path);
4690 			vd->file_devid = NULL;
4691 			return (0);
4692 		}
4693 	}
4694 
4695 	/*
4696 	 * No valid device id was found so we create one. Note that a failure
4697 	 * to create a device id is not fatal and does not prevent the disk
4698 	 * image from being attached.
4699 	 */
4700 	PR1("creating devid for %s", file_path);
4701 
4702 	if (ddi_devid_init(vd->vds->dip, DEVID_FAB, NULL, 0,
4703 	    &vd->file_devid) != DDI_SUCCESS) {
4704 		PR0("fail to create devid for %s", file_path);
4705 		vd->file_devid = NULL;
4706 		return (0);
4707 	}
4708 
4709 	/*
4710 	 * Write devid to the disk image. The devid is stored into the disk
4711 	 * image if we have a valid label; otherwise the devid will be stored
4712 	 * when the user writes a valid label.
4713 	 */
4714 	if (vd->vdisk_label != VD_DISK_LABEL_UNK) {
4715 		if (vd_file_write_devid(vd, vd->file_devid) != 0) {
4716 			PR0("fail to write devid for %s", file_path);
4717 			ddi_devid_free(vd->file_devid);
4718 			vd->file_devid = NULL;
4719 		}
4720 	}
4721 
4722 	return (0);
4723 }
4724 
4725 
4726 /*
4727  * Description:
4728  *	Open a device using its device path (supplied by ldm(1m))
4729  *
4730  * Parameters:
4731  *	vd 	- pointer to structure containing the vDisk info
4732  *
4733  * Return Value
4734  *	0	- success
4735  *	EIO	- Invalid number of partitions
4736  *	!= 0	- some other non-zero return value from ldi(9F) functions
4737  */
4738 static int
4739 vd_open_using_ldi_by_name(vd_t *vd)
4740 {
4741 	int		rval, status, open_flags;
4742 	struct dk_cinfo	dk_cinfo;
4743 	char		*device_path = vd->device_path;
4744 
4745 	/*
4746 	 * Try to open the device. If the flags indicate that the device should
4747 	 * be opened write-enabled, we first we try to open it "read-only"
4748 	 * to see if we have an optical device such as a CD-ROM which, for
4749 	 * now, we do not permit writes to and thus should not export write
4750 	 * operations to the client.
4751 	 *
4752 	 * Future: if/when we implement support for guest domains writing to
4753 	 * optical devices we will need to do further checking of the media type
4754 	 * to distinguish between read-only and writable discs.
4755 	 */
4756 	if (vd->open_flags & FWRITE) {
4757 		open_flags = vd->open_flags & ~FWRITE;
4758 		status = ldi_open_by_name(device_path, open_flags, kcred,
4759 		    &vd->ldi_handle[0], vd->vds->ldi_ident);
4760 
4761 		if (status == 0) {
4762 			/* Verify backing device supports dk_cinfo */
4763 			status = ldi_ioctl(vd->ldi_handle[0], DKIOCINFO,
4764 			    (intptr_t)&dk_cinfo, (open_flags | FKIOCTL),
4765 			    kcred, &rval);
4766 			if (status != 0) {
4767 				PRN("ldi_ioctl(DKIOCINFO) returned errno %d for"
4768 				    " %s opened as RO", status, device_path);
4769 				return (status);
4770 			}
4771 
4772 			if (dk_cinfo.dki_partition >= V_NUMPAR) {
4773 				PRN("slice %u >= maximum slice %u for %s",
4774 				    dk_cinfo.dki_partition, V_NUMPAR,
4775 				    device_path);
4776 				return (EIO);
4777 			}
4778 
4779 			/*
4780 			 * If this is an optical device then we disable
4781 			 * write access and return, otherwise we close
4782 			 * the device and try again with writes enabled.
4783 			 */
4784 			if (dk_cinfo.dki_ctype == DKC_CDROM) {
4785 				vd->open_flags = open_flags;
4786 				return (0);
4787 			} else {
4788 				(void) ldi_close(vd->ldi_handle[0],
4789 				    open_flags, kcred);
4790 			}
4791 		}
4792 	}
4793 
4794 	/* Attempt to (re)open device */
4795 	status = ldi_open_by_name(device_path, open_flags, kcred,
4796 	    &vd->ldi_handle[0], vd->vds->ldi_ident);
4797 
4798 	/*
4799 	 * The open can fail for example if we are opening an empty slice.
4800 	 * In case of a failure, we try the open again but this time with
4801 	 * the FNDELAY flag.
4802 	 */
4803 	if (status != 0)
4804 		status = ldi_open_by_name(device_path, vd->open_flags | FNDELAY,
4805 		    kcred, &vd->ldi_handle[0], vd->vds->ldi_ident);
4806 
4807 	if (status != 0) {
4808 		PR0("ldi_open_by_name(%s) = errno %d", device_path, status);
4809 		vd->ldi_handle[0] = NULL;
4810 		return (status);
4811 	}
4812 
4813 	/* Verify backing device supports dk_cinfo */
4814 	if ((status = ldi_ioctl(vd->ldi_handle[0], DKIOCINFO,
4815 	    (intptr_t)&dk_cinfo, (vd->open_flags | FKIOCTL), kcred,
4816 	    &rval)) != 0) {
4817 		PRN("ldi_ioctl(DKIOCINFO) returned errno %d for %s",
4818 		    status, device_path);
4819 		return (status);
4820 	}
4821 	if (dk_cinfo.dki_partition >= V_NUMPAR) {
4822 		PRN("slice %u >= maximum slice %u for %s",
4823 		    dk_cinfo.dki_partition, V_NUMPAR, device_path);
4824 		return (EIO);
4825 	}
4826 
4827 	return (0);
4828 }
4829 
4830 
4831 /*
4832  * Setup for a virtual disk which backend is a device (a physical disk,
4833  * slice or pseudo device) that is directly exported either as a full disk
4834  * for a physical disk or as a slice for a pseudo device or a disk slice.
4835  * In these cases, the backend is accessed using the LDI interface.
4836  */
4837 static int
4838 vd_setup_backend_ldi(vd_t *vd)
4839 {
4840 	int		rval, status;
4841 	struct dk_cinfo	dk_cinfo;
4842 	char		*device_path = vd->device_path;
4843 
4844 	status = vd_open_using_ldi_by_name(vd);
4845 	if (status != 0) {
4846 		PR0("Failed to open (%s) = errno %d", device_path, status);
4847 		return (status);
4848 	}
4849 
4850 	vd->file = B_FALSE;
4851 
4852 	/* Get device number of backing device */
4853 	if ((status = ldi_get_dev(vd->ldi_handle[0], &vd->dev[0])) != 0) {
4854 		PRN("ldi_get_dev() returned errno %d for %s",
4855 		    status, device_path);
4856 		return (status);
4857 	}
4858 
4859 	/* Verify backing device supports dk_cinfo */
4860 	if ((status = ldi_ioctl(vd->ldi_handle[0], DKIOCINFO,
4861 	    (intptr_t)&dk_cinfo, (vd->open_flags | FKIOCTL), kcred,
4862 	    &rval)) != 0) {
4863 		PRN("ldi_ioctl(DKIOCINFO) returned errno %d for %s",
4864 		    status, device_path);
4865 		return (status);
4866 	}
4867 	if (dk_cinfo.dki_partition >= V_NUMPAR) {
4868 		PRN("slice %u >= maximum slice %u for %s",
4869 		    dk_cinfo.dki_partition, V_NUMPAR, device_path);
4870 		return (EIO);
4871 	}
4872 
4873 	vd->vdisk_label = vd_read_vtoc(vd, &vd->vtoc);
4874 
4875 	/* Store the device's max transfer size for return to the client */
4876 	vd->max_xfer_sz = dk_cinfo.dki_maxtransfer;
4877 
4878 	/*
4879 	 * We need to work out if it's an ATAPI (IDE CD-ROM) or SCSI device so
4880 	 * that we can use the correct CDB group when sending USCSI commands.
4881 	 */
4882 	vd->is_atapi_dev = vd_is_atapi_device(vd);
4883 
4884 	/*
4885 	 * Export a full disk.
4886 	 *
4887 	 * When we use the LDI interface, we export a device as a full disk
4888 	 * if we have an entire disk slice (slice 2) and if this slice is
4889 	 * exported as a full disk and not as a single slice disk.
4890 	 * Similarly, we want to use LDI if we are accessing a CD or DVD
4891 	 * device (even if it isn't s2)
4892 	 *
4893 	 * Note that pseudo devices are exported as full disks using the vnode
4894 	 * interface, not the LDI interface.
4895 	 */
4896 	if ((dk_cinfo.dki_partition == VD_ENTIRE_DISK_SLICE &&
4897 	    vd->vdisk_type == VD_DISK_TYPE_DISK) ||
4898 	    dk_cinfo.dki_ctype == DKC_CDROM) {
4899 		ASSERT(!vd->pseudo);
4900 		if (dk_cinfo.dki_ctype == DKC_SCSI_CCS)
4901 			vd->scsi = B_TRUE;
4902 		return (vd_setup_full_disk(vd));
4903 	}
4904 
4905 	/*
4906 	 * Export a single slice disk.
4907 	 *
4908 	 * The exported device can be either a pseudo device or a disk slice. If
4909 	 * it is a disk slice different from slice 2 then it is always exported
4910 	 * as a single slice disk even if the "slice" option is not specified.
4911 	 * If it is disk slice 2 or a pseudo device then it is exported as a
4912 	 * single slice disk only if the "slice" option is specified.
4913 	 */
4914 	return (vd_setup_single_slice_disk(vd));
4915 }
4916 
4917 static int
4918 vd_setup_single_slice_disk(vd_t *vd)
4919 {
4920 	int status;
4921 	char *device_path = vd->device_path;
4922 
4923 	/* Get size of backing device */
4924 	if (ldi_get_size(vd->ldi_handle[0], &vd->vdisk_size) != DDI_SUCCESS) {
4925 		PRN("ldi_get_size() failed for %s", device_path);
4926 		return (EIO);
4927 	}
4928 	vd->vdisk_size = lbtodb(vd->vdisk_size);	/* convert to blocks */
4929 	vd->block_size = DEV_BSIZE;
4930 	vd->vdisk_block_size = DEV_BSIZE;
4931 	vd->vdisk_media = VD_MEDIA_FIXED;
4932 
4933 	if (vd->pseudo) {
4934 
4935 		ASSERT(vd->vdisk_type == VD_DISK_TYPE_SLICE);
4936 
4937 		/*
4938 		 * Currently we only support exporting pseudo devices which
4939 		 * provide a valid disk label.
4940 		 */
4941 		if (vd->vdisk_label == VD_DISK_LABEL_UNK) {
4942 			PRN("%s is a pseudo device with an invalid disk "
4943 			    "label\n", device_path);
4944 			return (EINVAL);
4945 		}
4946 		return (0);	/* ...and we're done */
4947 	}
4948 
4949 	/* We can only export a slice if the disk has a valid label */
4950 	if (vd->vdisk_label == VD_DISK_LABEL_UNK) {
4951 		PRN("%s is a slice from a disk with an unknown disk label\n",
4952 		    device_path);
4953 		return (EINVAL);
4954 	}
4955 
4956 	/*
4957 	 * We export the slice as a single slice disk even if the "slice"
4958 	 * option was not specified.
4959 	 */
4960 	vd->vdisk_type	= VD_DISK_TYPE_SLICE;
4961 	vd->nslices	= 1;
4962 
4963 	if (vd->vdisk_label == VD_DISK_LABEL_EFI) {
4964 		/* Slice from a disk with an EFI label */
4965 		status = vd_setup_partition_efi(vd);
4966 	} else {
4967 		/* Slice from a disk with a VTOC label */
4968 		ASSERT(vd->vdisk_label == VD_DISK_LABEL_VTOC);
4969 		status = vd_setup_partition_vtoc(vd);
4970 	}
4971 
4972 	return (status);
4973 }
4974 
4975 static int
4976 vd_setup_vd(vd_t *vd)
4977 {
4978 	int		status;
4979 	dev_info_t	*dip;
4980 	vnode_t 	*vnp;
4981 	char		*path = vd->device_path;
4982 
4983 	/* make sure the vdisk backend is valid */
4984 	if ((status = lookupname(path, UIO_SYSSPACE,
4985 	    FOLLOW, NULLVPP, &vnp)) != 0) {
4986 		PR0("Cannot lookup %s errno %d", path, status);
4987 		goto done;
4988 	}
4989 
4990 	switch (vnp->v_type) {
4991 	case VREG:
4992 		/*
4993 		 * Backend is a file so it is exported as a full disk or as a
4994 		 * single slice disk using the vnode interface.
4995 		 */
4996 		VN_RELE(vnp);
4997 		vd->pseudo = B_FALSE;
4998 		status = vd_setup_backend_vnode(vd);
4999 		break;
5000 
5001 	case VBLK:
5002 	case VCHR:
5003 		/*
5004 		 * Backend is a device. The way it is exported depends on the
5005 		 * type of the device.
5006 		 *
5007 		 * - A pseudo device is exported as a full disk using the vnode
5008 		 *   interface or as a single slice disk using the LDI
5009 		 *   interface.
5010 		 *
5011 		 * - A disk (represented by the slice 2 of that disk) is
5012 		 *   exported as a full disk using the LDI interface.
5013 		 *
5014 		 * - A disk slice (different from slice 2) is always exported
5015 		 *   as a single slice disk using the LDI interface.
5016 		 *
5017 		 * - The slice 2 of a disk is exported as a single slice disk
5018 		 *   if the "slice" option is specified, otherwise the entire
5019 		 *   disk will be exported. In any case, the LDI interface is
5020 		 *   used.
5021 		 */
5022 
5023 		/* check if this is a pseudo device */
5024 		if ((dip = ddi_hold_devi_by_instance(getmajor(vnp->v_rdev),
5025 		    dev_to_instance(vnp->v_rdev), 0))  == NULL) {
5026 			PRN("%s is no longer accessible", path);
5027 			VN_RELE(vnp);
5028 			status = EIO;
5029 			break;
5030 		}
5031 		vd->pseudo = is_pseudo_device(dip);
5032 		ddi_release_devi(dip);
5033 		VN_RELE(vnp);
5034 
5035 		if (!vd->pseudo) {
5036 			status = vd_setup_backend_ldi(vd);
5037 			break;
5038 		}
5039 
5040 		/*
5041 		 * If this is a pseudo device then its usage depends if the
5042 		 * "slice" option is set or not. If the "slice" option is set
5043 		 * then the pseudo device will be exported as a single slice,
5044 		 * otherwise it will be exported as a full disk.
5045 		 *
5046 		 * For backward compatibility, if vd_volume_force_slice is set
5047 		 * then we always export pseudo devices as slices.
5048 		 */
5049 		if (vd_volume_force_slice) {
5050 			vd->vdisk_type = VD_DISK_TYPE_SLICE;
5051 			vd->nslices = 1;
5052 		}
5053 
5054 		if (vd->vdisk_type == VD_DISK_TYPE_DISK)
5055 			status = vd_setup_backend_vnode(vd);
5056 		else
5057 			status = vd_setup_backend_ldi(vd);
5058 		break;
5059 
5060 	default:
5061 		PRN("Unsupported vdisk backend %s", path);
5062 		VN_RELE(vnp);
5063 		status = EBADF;
5064 	}
5065 
5066 done:
5067 	if (status != 0) {
5068 		/*
5069 		 * If the error is retryable print an error message only
5070 		 * during the first try.
5071 		 */
5072 		if (status == ENXIO || status == ENODEV ||
5073 		    status == ENOENT || status == EROFS) {
5074 			if (!(vd->initialized & VD_SETUP_ERROR)) {
5075 				PRN("%s is currently inaccessible (error %d)",
5076 				    path, status);
5077 			}
5078 			status = EAGAIN;
5079 		} else {
5080 			PRN("%s can not be exported as a virtual disk "
5081 			    "(error %d)", path, status);
5082 		}
5083 		vd->initialized |= VD_SETUP_ERROR;
5084 
5085 	} else if (vd->initialized & VD_SETUP_ERROR) {
5086 		/* print a message only if we previously had an error */
5087 		PRN("%s is now online", path);
5088 		vd->initialized &= ~VD_SETUP_ERROR;
5089 	}
5090 
5091 	return (status);
5092 }
5093 
5094 static int
5095 vds_do_init_vd(vds_t *vds, uint64_t id, char *device_path, uint64_t options,
5096     uint64_t ldc_id, vd_t **vdp)
5097 {
5098 	char			tq_name[TASKQ_NAMELEN];
5099 	int			status;
5100 	ddi_iblock_cookie_t	iblock = NULL;
5101 	ldc_attr_t		ldc_attr;
5102 	vd_t			*vd;
5103 
5104 
5105 	ASSERT(vds != NULL);
5106 	ASSERT(device_path != NULL);
5107 	ASSERT(vdp != NULL);
5108 	PR0("Adding vdisk for %s", device_path);
5109 
5110 	if ((vd = kmem_zalloc(sizeof (*vd), KM_NOSLEEP)) == NULL) {
5111 		PRN("No memory for virtual disk");
5112 		return (EAGAIN);
5113 	}
5114 	*vdp = vd;	/* assign here so vds_destroy_vd() can cleanup later */
5115 	vd->vds = vds;
5116 	(void) strncpy(vd->device_path, device_path, MAXPATHLEN);
5117 
5118 	/* Setup open flags */
5119 	vd->open_flags = FREAD;
5120 
5121 	if (!(options & VD_OPT_RDONLY))
5122 		vd->open_flags |= FWRITE;
5123 
5124 	if (options & VD_OPT_EXCLUSIVE)
5125 		vd->open_flags |= FEXCL;
5126 
5127 	/* Setup disk type */
5128 	if (options & VD_OPT_SLICE) {
5129 		vd->vdisk_type = VD_DISK_TYPE_SLICE;
5130 		vd->nslices = 1;
5131 	} else {
5132 		vd->vdisk_type = VD_DISK_TYPE_DISK;
5133 		vd->nslices = V_NUMPAR;
5134 	}
5135 
5136 	/* default disk label */
5137 	vd->vdisk_label = VD_DISK_LABEL_UNK;
5138 
5139 	/* Open vdisk and initialize parameters */
5140 	if ((status = vd_setup_vd(vd)) == 0) {
5141 		vd->initialized |= VD_DISK_READY;
5142 
5143 		ASSERT(vd->nslices > 0 && vd->nslices <= V_NUMPAR);
5144 		PR0("vdisk_type = %s, pseudo = %s, file = %s, nslices = %u",
5145 		    ((vd->vdisk_type == VD_DISK_TYPE_DISK) ? "disk" : "slice"),
5146 		    (vd->pseudo ? "yes" : "no"), (vd->file ? "yes" : "no"),
5147 		    vd->nslices);
5148 	} else {
5149 		if (status != EAGAIN)
5150 			return (status);
5151 	}
5152 
5153 	/* Initialize locking */
5154 	if (ddi_get_soft_iblock_cookie(vds->dip, DDI_SOFTINT_MED,
5155 	    &iblock) != DDI_SUCCESS) {
5156 		PRN("Could not get iblock cookie.");
5157 		return (EIO);
5158 	}
5159 
5160 	mutex_init(&vd->lock, NULL, MUTEX_DRIVER, iblock);
5161 	vd->initialized |= VD_LOCKING;
5162 
5163 
5164 	/* Create start and completion task queues for the vdisk */
5165 	(void) snprintf(tq_name, sizeof (tq_name), "vd_startq%lu", id);
5166 	PR1("tq_name = %s", tq_name);
5167 	if ((vd->startq = ddi_taskq_create(vds->dip, tq_name, 1,
5168 	    TASKQ_DEFAULTPRI, 0)) == NULL) {
5169 		PRN("Could not create task queue");
5170 		return (EIO);
5171 	}
5172 	(void) snprintf(tq_name, sizeof (tq_name), "vd_completionq%lu", id);
5173 	PR1("tq_name = %s", tq_name);
5174 	if ((vd->completionq = ddi_taskq_create(vds->dip, tq_name, 1,
5175 	    TASKQ_DEFAULTPRI, 0)) == NULL) {
5176 		PRN("Could not create task queue");
5177 		return (EIO);
5178 	}
5179 	vd->enabled = 1;	/* before callback can dispatch to startq */
5180 
5181 
5182 	/* Bring up LDC */
5183 	ldc_attr.devclass	= LDC_DEV_BLK_SVC;
5184 	ldc_attr.instance	= ddi_get_instance(vds->dip);
5185 	ldc_attr.mode		= LDC_MODE_UNRELIABLE;
5186 	ldc_attr.mtu		= VD_LDC_MTU;
5187 	if ((status = ldc_init(ldc_id, &ldc_attr, &vd->ldc_handle)) != 0) {
5188 		PRN("Could not initialize LDC channel %lx, "
5189 		    "init failed with error %d", ldc_id, status);
5190 		return (status);
5191 	}
5192 	vd->initialized |= VD_LDC;
5193 
5194 	if ((status = ldc_reg_callback(vd->ldc_handle, vd_handle_ldc_events,
5195 	    (caddr_t)vd)) != 0) {
5196 		PRN("Could not initialize LDC channel %lu,"
5197 		    "reg_callback failed with error %d", ldc_id, status);
5198 		return (status);
5199 	}
5200 
5201 	if ((status = ldc_open(vd->ldc_handle)) != 0) {
5202 		PRN("Could not initialize LDC channel %lu,"
5203 		    "open failed with error %d", ldc_id, status);
5204 		return (status);
5205 	}
5206 
5207 	if ((status = ldc_up(vd->ldc_handle)) != 0) {
5208 		PR0("ldc_up() returned errno %d", status);
5209 	}
5210 
5211 	/* Allocate the inband task memory handle */
5212 	status = ldc_mem_alloc_handle(vd->ldc_handle, &(vd->inband_task.mhdl));
5213 	if (status) {
5214 		PRN("Could not initialize LDC channel %lu,"
5215 		    "alloc_handle failed with error %d", ldc_id, status);
5216 		return (ENXIO);
5217 	}
5218 
5219 	/* Add the successfully-initialized vdisk to the server's table */
5220 	if (mod_hash_insert(vds->vd_table, (mod_hash_key_t)id, vd) != 0) {
5221 		PRN("Error adding vdisk ID %lu to table", id);
5222 		return (EIO);
5223 	}
5224 
5225 	/* Allocate the staging buffer */
5226 	vd->max_msglen	= sizeof (vio_msg_t);	/* baseline vio message size */
5227 	vd->vio_msgp = kmem_alloc(vd->max_msglen, KM_SLEEP);
5228 
5229 	/* store initial state */
5230 	vd->state = VD_STATE_INIT;
5231 
5232 	return (0);
5233 }
5234 
5235 static void
5236 vd_free_dring_task(vd_t *vdp)
5237 {
5238 	if (vdp->dring_task != NULL) {
5239 		ASSERT(vdp->dring_len != 0);
5240 		/* Free all dring_task memory handles */
5241 		for (int i = 0; i < vdp->dring_len; i++) {
5242 			(void) ldc_mem_free_handle(vdp->dring_task[i].mhdl);
5243 			kmem_free(vdp->dring_task[i].msg, vdp->max_msglen);
5244 			vdp->dring_task[i].msg = NULL;
5245 		}
5246 		kmem_free(vdp->dring_task,
5247 		    (sizeof (*vdp->dring_task)) * vdp->dring_len);
5248 		vdp->dring_task = NULL;
5249 	}
5250 }
5251 
5252 /*
5253  * Destroy the state associated with a virtual disk
5254  */
5255 static void
5256 vds_destroy_vd(void *arg)
5257 {
5258 	vd_t	*vd = (vd_t *)arg;
5259 	int	retry = 0, rv;
5260 
5261 	if (vd == NULL)
5262 		return;
5263 
5264 	PR0("Destroying vdisk state");
5265 
5266 	if (vd->dk_efi.dki_data != NULL)
5267 		kmem_free(vd->dk_efi.dki_data, vd->dk_efi.dki_length);
5268 
5269 	/* Disable queuing requests for the vdisk */
5270 	if (vd->initialized & VD_LOCKING) {
5271 		mutex_enter(&vd->lock);
5272 		vd->enabled = 0;
5273 		mutex_exit(&vd->lock);
5274 	}
5275 
5276 	/* Drain and destroy start queue (*before* destroying completionq) */
5277 	if (vd->startq != NULL)
5278 		ddi_taskq_destroy(vd->startq);	/* waits for queued tasks */
5279 
5280 	/* Drain and destroy completion queue (*before* shutting down LDC) */
5281 	if (vd->completionq != NULL)
5282 		ddi_taskq_destroy(vd->completionq);	/* waits for tasks */
5283 
5284 	vd_free_dring_task(vd);
5285 
5286 	/* Free the inband task memory handle */
5287 	(void) ldc_mem_free_handle(vd->inband_task.mhdl);
5288 
5289 	/* Shut down LDC */
5290 	if (vd->initialized & VD_LDC) {
5291 		/* unmap the dring */
5292 		if (vd->initialized & VD_DRING)
5293 			(void) ldc_mem_dring_unmap(vd->dring_handle);
5294 
5295 		/* close LDC channel - retry on EAGAIN */
5296 		while ((rv = ldc_close(vd->ldc_handle)) == EAGAIN) {
5297 			if (++retry > vds_ldc_retries) {
5298 				PR0("Timed out closing channel");
5299 				break;
5300 			}
5301 			drv_usecwait(vds_ldc_delay);
5302 		}
5303 		if (rv == 0) {
5304 			(void) ldc_unreg_callback(vd->ldc_handle);
5305 			(void) ldc_fini(vd->ldc_handle);
5306 		} else {
5307 			/*
5308 			 * Closing the LDC channel has failed. Ideally we should
5309 			 * fail here but there is no Zeus level infrastructure
5310 			 * to handle this. The MD has already been changed and
5311 			 * we have to do the close. So we try to do as much
5312 			 * clean up as we can.
5313 			 */
5314 			(void) ldc_set_cb_mode(vd->ldc_handle, LDC_CB_DISABLE);
5315 			while (ldc_unreg_callback(vd->ldc_handle) == EAGAIN)
5316 				drv_usecwait(vds_ldc_delay);
5317 		}
5318 	}
5319 
5320 	/* Free the staging buffer for msgs */
5321 	if (vd->vio_msgp != NULL) {
5322 		kmem_free(vd->vio_msgp, vd->max_msglen);
5323 		vd->vio_msgp = NULL;
5324 	}
5325 
5326 	/* Free the inband message buffer */
5327 	if (vd->inband_task.msg != NULL) {
5328 		kmem_free(vd->inband_task.msg, vd->max_msglen);
5329 		vd->inband_task.msg = NULL;
5330 	}
5331 
5332 	if (vd->file) {
5333 		/* Close file */
5334 		(void) VOP_CLOSE(vd->file_vnode, vd->open_flags, 1,
5335 		    0, kcred, NULL);
5336 		VN_RELE(vd->file_vnode);
5337 		if (vd->file_devid != NULL)
5338 			ddi_devid_free(vd->file_devid);
5339 	} else {
5340 		/* Close any open backing-device slices */
5341 		for (uint_t slice = 0; slice < vd->nslices; slice++) {
5342 			if (vd->ldi_handle[slice] != NULL) {
5343 				PR0("Closing slice %u", slice);
5344 				(void) ldi_close(vd->ldi_handle[slice],
5345 				    vd->open_flags, kcred);
5346 			}
5347 		}
5348 	}
5349 
5350 	/* Free lock */
5351 	if (vd->initialized & VD_LOCKING)
5352 		mutex_destroy(&vd->lock);
5353 
5354 	/* Finally, free the vdisk structure itself */
5355 	kmem_free(vd, sizeof (*vd));
5356 }
5357 
5358 static int
5359 vds_init_vd(vds_t *vds, uint64_t id, char *device_path, uint64_t options,
5360     uint64_t ldc_id)
5361 {
5362 	int	status;
5363 	vd_t	*vd = NULL;
5364 
5365 
5366 	if ((status = vds_do_init_vd(vds, id, device_path, options,
5367 	    ldc_id, &vd)) != 0)
5368 		vds_destroy_vd(vd);
5369 
5370 	return (status);
5371 }
5372 
5373 static int
5374 vds_do_get_ldc_id(md_t *md, mde_cookie_t vd_node, mde_cookie_t *channel,
5375     uint64_t *ldc_id)
5376 {
5377 	int	num_channels;
5378 
5379 
5380 	/* Look for channel endpoint child(ren) of the vdisk MD node */
5381 	if ((num_channels = md_scan_dag(md, vd_node,
5382 	    md_find_name(md, VD_CHANNEL_ENDPOINT),
5383 	    md_find_name(md, "fwd"), channel)) <= 0) {
5384 		PRN("No \"%s\" found for virtual disk", VD_CHANNEL_ENDPOINT);
5385 		return (-1);
5386 	}
5387 
5388 	/* Get the "id" value for the first channel endpoint node */
5389 	if (md_get_prop_val(md, channel[0], VD_ID_PROP, ldc_id) != 0) {
5390 		PRN("No \"%s\" property found for \"%s\" of vdisk",
5391 		    VD_ID_PROP, VD_CHANNEL_ENDPOINT);
5392 		return (-1);
5393 	}
5394 
5395 	if (num_channels > 1) {
5396 		PRN("Using ID of first of multiple channels for this vdisk");
5397 	}
5398 
5399 	return (0);
5400 }
5401 
5402 static int
5403 vds_get_ldc_id(md_t *md, mde_cookie_t vd_node, uint64_t *ldc_id)
5404 {
5405 	int		num_nodes, status;
5406 	size_t		size;
5407 	mde_cookie_t	*channel;
5408 
5409 
5410 	if ((num_nodes = md_node_count(md)) <= 0) {
5411 		PRN("Invalid node count in Machine Description subtree");
5412 		return (-1);
5413 	}
5414 	size = num_nodes*(sizeof (*channel));
5415 	channel = kmem_zalloc(size, KM_SLEEP);
5416 	status = vds_do_get_ldc_id(md, vd_node, channel, ldc_id);
5417 	kmem_free(channel, size);
5418 
5419 	return (status);
5420 }
5421 
5422 /*
5423  * Function:
5424  *	vds_get_options
5425  *
5426  * Description:
5427  * 	Parse the options of a vds node. Options are defined as an array
5428  *	of strings in the vds-block-device-opts property of the vds node
5429  *	in the machine description. Options are returned as a bitmask. The
5430  *	mapping between the bitmask options and the options strings from the
5431  *	machine description is defined in the vd_bdev_options[] array.
5432  *
5433  *	The vds-block-device-opts property is optional. If a vds has no such
5434  *	property then no option is defined.
5435  *
5436  * Parameters:
5437  *	md		- machine description.
5438  *	vd_node		- vds node in the machine description for which
5439  *			  options have to be parsed.
5440  *	options		- the returned options.
5441  *
5442  * Return Code:
5443  *	none.
5444  */
5445 static void
5446 vds_get_options(md_t *md, mde_cookie_t vd_node, uint64_t *options)
5447 {
5448 	char	*optstr, *opt;
5449 	int	len, n, i;
5450 
5451 	*options = 0;
5452 
5453 	if (md_get_prop_data(md, vd_node, VD_BLOCK_DEVICE_OPTS,
5454 	    (uint8_t **)&optstr, &len) != 0) {
5455 		PR0("No options found");
5456 		return;
5457 	}
5458 
5459 	/* parse options */
5460 	opt = optstr;
5461 	n = sizeof (vd_bdev_options) / sizeof (vd_option_t);
5462 
5463 	while (opt < optstr + len) {
5464 		for (i = 0; i < n; i++) {
5465 			if (strncmp(vd_bdev_options[i].vdo_name,
5466 			    opt, VD_OPTION_NLEN) == 0) {
5467 				*options |= vd_bdev_options[i].vdo_value;
5468 				break;
5469 			}
5470 		}
5471 
5472 		if (i < n) {
5473 			PR0("option: %s", opt);
5474 		} else {
5475 			PRN("option %s is unknown or unsupported", opt);
5476 		}
5477 
5478 		opt += strlen(opt) + 1;
5479 	}
5480 }
5481 
5482 static void
5483 vds_add_vd(vds_t *vds, md_t *md, mde_cookie_t vd_node)
5484 {
5485 	char		*device_path = NULL;
5486 	uint64_t	id = 0, ldc_id = 0, options = 0;
5487 
5488 	if (md_get_prop_val(md, vd_node, VD_ID_PROP, &id) != 0) {
5489 		PRN("Error getting vdisk \"%s\"", VD_ID_PROP);
5490 		return;
5491 	}
5492 	PR0("Adding vdisk ID %lu", id);
5493 	if (md_get_prop_str(md, vd_node, VD_BLOCK_DEVICE_PROP,
5494 	    &device_path) != 0) {
5495 		PRN("Error getting vdisk \"%s\"", VD_BLOCK_DEVICE_PROP);
5496 		return;
5497 	}
5498 
5499 	vds_get_options(md, vd_node, &options);
5500 
5501 	if (vds_get_ldc_id(md, vd_node, &ldc_id) != 0) {
5502 		PRN("Error getting LDC ID for vdisk %lu", id);
5503 		return;
5504 	}
5505 
5506 	if (vds_init_vd(vds, id, device_path, options, ldc_id) != 0) {
5507 		PRN("Failed to add vdisk ID %lu", id);
5508 		if (mod_hash_destroy(vds->vd_table, (mod_hash_key_t)id) != 0)
5509 			PRN("No vDisk entry found for vdisk ID %lu", id);
5510 		return;
5511 	}
5512 }
5513 
5514 static void
5515 vds_remove_vd(vds_t *vds, md_t *md, mde_cookie_t vd_node)
5516 {
5517 	uint64_t	id = 0;
5518 
5519 
5520 	if (md_get_prop_val(md, vd_node, VD_ID_PROP, &id) != 0) {
5521 		PRN("Unable to get \"%s\" property from vdisk's MD node",
5522 		    VD_ID_PROP);
5523 		return;
5524 	}
5525 	PR0("Removing vdisk ID %lu", id);
5526 	if (mod_hash_destroy(vds->vd_table, (mod_hash_key_t)id) != 0)
5527 		PRN("No vdisk entry found for vdisk ID %lu", id);
5528 }
5529 
5530 static void
5531 vds_change_vd(vds_t *vds, md_t *prev_md, mde_cookie_t prev_vd_node,
5532     md_t *curr_md, mde_cookie_t curr_vd_node)
5533 {
5534 	char		*curr_dev, *prev_dev;
5535 	uint64_t	curr_id = 0, curr_ldc_id = 0, curr_options = 0;
5536 	uint64_t	prev_id = 0, prev_ldc_id = 0, prev_options = 0;
5537 	size_t		len;
5538 
5539 
5540 	/* Validate that vdisk ID has not changed */
5541 	if (md_get_prop_val(prev_md, prev_vd_node, VD_ID_PROP, &prev_id) != 0) {
5542 		PRN("Error getting previous vdisk \"%s\" property",
5543 		    VD_ID_PROP);
5544 		return;
5545 	}
5546 	if (md_get_prop_val(curr_md, curr_vd_node, VD_ID_PROP, &curr_id) != 0) {
5547 		PRN("Error getting current vdisk \"%s\" property", VD_ID_PROP);
5548 		return;
5549 	}
5550 	if (curr_id != prev_id) {
5551 		PRN("Not changing vdisk:  ID changed from %lu to %lu",
5552 		    prev_id, curr_id);
5553 		return;
5554 	}
5555 
5556 	/* Validate that LDC ID has not changed */
5557 	if (vds_get_ldc_id(prev_md, prev_vd_node, &prev_ldc_id) != 0) {
5558 		PRN("Error getting LDC ID for vdisk %lu", prev_id);
5559 		return;
5560 	}
5561 
5562 	if (vds_get_ldc_id(curr_md, curr_vd_node, &curr_ldc_id) != 0) {
5563 		PRN("Error getting LDC ID for vdisk %lu", curr_id);
5564 		return;
5565 	}
5566 	if (curr_ldc_id != prev_ldc_id) {
5567 		_NOTE(NOTREACHED);	/* lint is confused */
5568 		PRN("Not changing vdisk:  "
5569 		    "LDC ID changed from %lu to %lu", prev_ldc_id, curr_ldc_id);
5570 		return;
5571 	}
5572 
5573 	/* Determine whether device path has changed */
5574 	if (md_get_prop_str(prev_md, prev_vd_node, VD_BLOCK_DEVICE_PROP,
5575 	    &prev_dev) != 0) {
5576 		PRN("Error getting previous vdisk \"%s\"",
5577 		    VD_BLOCK_DEVICE_PROP);
5578 		return;
5579 	}
5580 	if (md_get_prop_str(curr_md, curr_vd_node, VD_BLOCK_DEVICE_PROP,
5581 	    &curr_dev) != 0) {
5582 		PRN("Error getting current vdisk \"%s\"", VD_BLOCK_DEVICE_PROP);
5583 		return;
5584 	}
5585 	if (((len = strlen(curr_dev)) == strlen(prev_dev)) &&
5586 	    (strncmp(curr_dev, prev_dev, len) == 0))
5587 		return;	/* no relevant (supported) change */
5588 
5589 	/* Validate that options have not changed */
5590 	vds_get_options(prev_md, prev_vd_node, &prev_options);
5591 	vds_get_options(curr_md, curr_vd_node, &curr_options);
5592 	if (prev_options != curr_options) {
5593 		PRN("Not changing vdisk:  options changed from %lx to %lx",
5594 		    prev_options, curr_options);
5595 		return;
5596 	}
5597 
5598 	PR0("Changing vdisk ID %lu", prev_id);
5599 
5600 	/* Remove old state, which will close vdisk and reset */
5601 	if (mod_hash_destroy(vds->vd_table, (mod_hash_key_t)prev_id) != 0)
5602 		PRN("No entry found for vdisk ID %lu", prev_id);
5603 
5604 	/* Re-initialize vdisk with new state */
5605 	if (vds_init_vd(vds, curr_id, curr_dev, curr_options,
5606 	    curr_ldc_id) != 0) {
5607 		PRN("Failed to change vdisk ID %lu", curr_id);
5608 		return;
5609 	}
5610 }
5611 
5612 static int
5613 vds_process_md(void *arg, mdeg_result_t *md)
5614 {
5615 	int	i;
5616 	vds_t	*vds = arg;
5617 
5618 
5619 	if (md == NULL)
5620 		return (MDEG_FAILURE);
5621 	ASSERT(vds != NULL);
5622 
5623 	for (i = 0; i < md->removed.nelem; i++)
5624 		vds_remove_vd(vds, md->removed.mdp, md->removed.mdep[i]);
5625 	for (i = 0; i < md->match_curr.nelem; i++)
5626 		vds_change_vd(vds, md->match_prev.mdp, md->match_prev.mdep[i],
5627 		    md->match_curr.mdp, md->match_curr.mdep[i]);
5628 	for (i = 0; i < md->added.nelem; i++)
5629 		vds_add_vd(vds, md->added.mdp, md->added.mdep[i]);
5630 
5631 	return (MDEG_SUCCESS);
5632 }
5633 
5634 
5635 static int
5636 vds_do_attach(dev_info_t *dip)
5637 {
5638 	int			status, sz;
5639 	int			cfg_handle;
5640 	minor_t			instance = ddi_get_instance(dip);
5641 	vds_t			*vds;
5642 	mdeg_prop_spec_t	*pspecp;
5643 	mdeg_node_spec_t	*ispecp;
5644 
5645 	/*
5646 	 * The "cfg-handle" property of a vds node in an MD contains the MD's
5647 	 * notion of "instance", or unique identifier, for that node; OBP
5648 	 * stores the value of the "cfg-handle" MD property as the value of
5649 	 * the "reg" property on the node in the device tree it builds from
5650 	 * the MD and passes to Solaris.  Thus, we look up the devinfo node's
5651 	 * "reg" property value to uniquely identify this device instance when
5652 	 * registering with the MD event-generation framework.  If the "reg"
5653 	 * property cannot be found, the device tree state is presumably so
5654 	 * broken that there is no point in continuing.
5655 	 */
5656 	if (!ddi_prop_exists(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS,
5657 	    VD_REG_PROP)) {
5658 		PRN("vds \"%s\" property does not exist", VD_REG_PROP);
5659 		return (DDI_FAILURE);
5660 	}
5661 
5662 	/* Get the MD instance for later MDEG registration */
5663 	cfg_handle = ddi_prop_get_int(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS,
5664 	    VD_REG_PROP, -1);
5665 
5666 	if (ddi_soft_state_zalloc(vds_state, instance) != DDI_SUCCESS) {
5667 		PRN("Could not allocate state for instance %u", instance);
5668 		return (DDI_FAILURE);
5669 	}
5670 
5671 	if ((vds = ddi_get_soft_state(vds_state, instance)) == NULL) {
5672 		PRN("Could not get state for instance %u", instance);
5673 		ddi_soft_state_free(vds_state, instance);
5674 		return (DDI_FAILURE);
5675 	}
5676 
5677 	vds->dip	= dip;
5678 	vds->vd_table	= mod_hash_create_ptrhash("vds_vd_table", VDS_NCHAINS,
5679 	    vds_destroy_vd, sizeof (void *));
5680 
5681 	ASSERT(vds->vd_table != NULL);
5682 
5683 	if ((status = ldi_ident_from_dip(dip, &vds->ldi_ident)) != 0) {
5684 		PRN("ldi_ident_from_dip() returned errno %d", status);
5685 		return (DDI_FAILURE);
5686 	}
5687 	vds->initialized |= VDS_LDI;
5688 
5689 	/* Register for MD updates */
5690 	sz = sizeof (vds_prop_template);
5691 	pspecp = kmem_alloc(sz, KM_SLEEP);
5692 	bcopy(vds_prop_template, pspecp, sz);
5693 
5694 	VDS_SET_MDEG_PROP_INST(pspecp, cfg_handle);
5695 
5696 	/* initialize the complete prop spec structure */
5697 	ispecp = kmem_zalloc(sizeof (mdeg_node_spec_t), KM_SLEEP);
5698 	ispecp->namep = "virtual-device";
5699 	ispecp->specp = pspecp;
5700 
5701 	if (mdeg_register(ispecp, &vd_match, vds_process_md, vds,
5702 	    &vds->mdeg) != MDEG_SUCCESS) {
5703 		PRN("Unable to register for MD updates");
5704 		kmem_free(ispecp, sizeof (mdeg_node_spec_t));
5705 		kmem_free(pspecp, sz);
5706 		return (DDI_FAILURE);
5707 	}
5708 
5709 	vds->ispecp = ispecp;
5710 	vds->initialized |= VDS_MDEG;
5711 
5712 	/* Prevent auto-detaching so driver is available whenever MD changes */
5713 	if (ddi_prop_update_int(DDI_DEV_T_NONE, dip, DDI_NO_AUTODETACH, 1) !=
5714 	    DDI_PROP_SUCCESS) {
5715 		PRN("failed to set \"%s\" property for instance %u",
5716 		    DDI_NO_AUTODETACH, instance);
5717 	}
5718 
5719 	ddi_report_dev(dip);
5720 	return (DDI_SUCCESS);
5721 }
5722 
5723 static int
5724 vds_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
5725 {
5726 	int	status;
5727 
5728 	switch (cmd) {
5729 	case DDI_ATTACH:
5730 		PR0("Attaching");
5731 		if ((status = vds_do_attach(dip)) != DDI_SUCCESS)
5732 			(void) vds_detach(dip, DDI_DETACH);
5733 		return (status);
5734 	case DDI_RESUME:
5735 		PR0("No action required for DDI_RESUME");
5736 		return (DDI_SUCCESS);
5737 	default:
5738 		return (DDI_FAILURE);
5739 	}
5740 }
5741 
5742 static struct dev_ops vds_ops = {
5743 	DEVO_REV,	/* devo_rev */
5744 	0,		/* devo_refcnt */
5745 	ddi_no_info,	/* devo_getinfo */
5746 	nulldev,	/* devo_identify */
5747 	nulldev,	/* devo_probe */
5748 	vds_attach,	/* devo_attach */
5749 	vds_detach,	/* devo_detach */
5750 	nodev,		/* devo_reset */
5751 	NULL,		/* devo_cb_ops */
5752 	NULL,		/* devo_bus_ops */
5753 	nulldev		/* devo_power */
5754 };
5755 
5756 static struct modldrv modldrv = {
5757 	&mod_driverops,
5758 	"virtual disk server",
5759 	&vds_ops,
5760 };
5761 
5762 static struct modlinkage modlinkage = {
5763 	MODREV_1,
5764 	&modldrv,
5765 	NULL
5766 };
5767 
5768 
5769 int
5770 _init(void)
5771 {
5772 	int		status;
5773 
5774 	if ((status = ddi_soft_state_init(&vds_state, sizeof (vds_t), 1)) != 0)
5775 		return (status);
5776 
5777 	if ((status = mod_install(&modlinkage)) != 0) {
5778 		ddi_soft_state_fini(&vds_state);
5779 		return (status);
5780 	}
5781 
5782 	return (0);
5783 }
5784 
5785 int
5786 _info(struct modinfo *modinfop)
5787 {
5788 	return (mod_info(&modlinkage, modinfop));
5789 }
5790 
5791 int
5792 _fini(void)
5793 {
5794 	int	status;
5795 
5796 	if ((status = mod_remove(&modlinkage)) != 0)
5797 		return (status);
5798 	ddi_soft_state_fini(&vds_state);
5799 	return (0);
5800 }
5801