xref: /illumos-gate/usr/src/uts/common/io/cmlb.c (revision dcbf3bd6a1f1360fc1afcee9e22c6dcff7844bf2)
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 2012 DEY Storage Systems, Inc.  All rights reserved.
24  * Copyright 2010 Sun Microsystems, Inc.  All rights reserved.
25  * Use is subject to license terms.
26  * Copyright 2016 Toomas Soome <tsoome@me.com>
27  */
28 
29 /*
30  * This module provides support for labeling operations for target
31  * drivers.
32  */
33 
34 #include <sys/scsi/scsi.h>
35 #include <sys/sunddi.h>
36 #include <sys/dklabel.h>
37 #include <sys/dkio.h>
38 #include <sys/vtoc.h>
39 #include <sys/dktp/fdisk.h>
40 #include <sys/vtrace.h>
41 #include <sys/efi_partition.h>
42 #include <sys/cmlb.h>
43 #include <sys/cmlb_impl.h>
44 #if defined(__i386) || defined(__amd64)
45 #include <sys/fs/dv_node.h>
46 #endif
47 #include <sys/ddi_impldefs.h>
48 
49 /*
50  * Driver minor node structure and data table
51  */
52 struct driver_minor_data {
53 	char	*name;
54 	minor_t	minor;
55 	int	type;
56 };
57 
58 static struct driver_minor_data dk_minor_data[] = {
59 	{"a", 0, S_IFBLK},
60 	{"b", 1, S_IFBLK},
61 	{"c", 2, S_IFBLK},
62 	{"d", 3, S_IFBLK},
63 	{"e", 4, S_IFBLK},
64 	{"f", 5, S_IFBLK},
65 	{"g", 6, S_IFBLK},
66 	{"h", 7, S_IFBLK},
67 #if defined(_SUNOS_VTOC_16)
68 	{"i", 8, S_IFBLK},
69 	{"j", 9, S_IFBLK},
70 	{"k", 10, S_IFBLK},
71 	{"l", 11, S_IFBLK},
72 	{"m", 12, S_IFBLK},
73 	{"n", 13, S_IFBLK},
74 	{"o", 14, S_IFBLK},
75 	{"p", 15, S_IFBLK},
76 #endif			/* defined(_SUNOS_VTOC_16) */
77 #if defined(_FIRMWARE_NEEDS_FDISK)
78 	{"q", 16, S_IFBLK},
79 	{"r", 17, S_IFBLK},
80 	{"s", 18, S_IFBLK},
81 	{"t", 19, S_IFBLK},
82 	{"u", 20, S_IFBLK},
83 #endif			/* defined(_FIRMWARE_NEEDS_FDISK) */
84 	{"a,raw", 0, S_IFCHR},
85 	{"b,raw", 1, S_IFCHR},
86 	{"c,raw", 2, S_IFCHR},
87 	{"d,raw", 3, S_IFCHR},
88 	{"e,raw", 4, S_IFCHR},
89 	{"f,raw", 5, S_IFCHR},
90 	{"g,raw", 6, S_IFCHR},
91 	{"h,raw", 7, S_IFCHR},
92 #if defined(_SUNOS_VTOC_16)
93 	{"i,raw", 8, S_IFCHR},
94 	{"j,raw", 9, S_IFCHR},
95 	{"k,raw", 10, S_IFCHR},
96 	{"l,raw", 11, S_IFCHR},
97 	{"m,raw", 12, S_IFCHR},
98 	{"n,raw", 13, S_IFCHR},
99 	{"o,raw", 14, S_IFCHR},
100 	{"p,raw", 15, S_IFCHR},
101 #endif			/* defined(_SUNOS_VTOC_16) */
102 #if defined(_FIRMWARE_NEEDS_FDISK)
103 	{"q,raw", 16, S_IFCHR},
104 	{"r,raw", 17, S_IFCHR},
105 	{"s,raw", 18, S_IFCHR},
106 	{"t,raw", 19, S_IFCHR},
107 	{"u,raw", 20, S_IFCHR},
108 #endif			/* defined(_FIRMWARE_NEEDS_FDISK) */
109 	{0}
110 };
111 
112 #if defined(__i386) || defined(__amd64)
113 #if defined(_FIRMWARE_NEEDS_FDISK)
114 static struct driver_minor_data dk_ext_minor_data[] = {
115 	{"p5", 21, S_IFBLK},
116 	{"p6", 22, S_IFBLK},
117 	{"p7", 23, S_IFBLK},
118 	{"p8", 24, S_IFBLK},
119 	{"p9", 25, S_IFBLK},
120 	{"p10", 26, S_IFBLK},
121 	{"p11", 27, S_IFBLK},
122 	{"p12", 28, S_IFBLK},
123 	{"p13", 29, S_IFBLK},
124 	{"p14", 30, S_IFBLK},
125 	{"p15", 31, S_IFBLK},
126 	{"p16", 32, S_IFBLK},
127 	{"p17", 33, S_IFBLK},
128 	{"p18", 34, S_IFBLK},
129 	{"p19", 35, S_IFBLK},
130 	{"p20", 36, S_IFBLK},
131 	{"p21", 37, S_IFBLK},
132 	{"p22", 38, S_IFBLK},
133 	{"p23", 39, S_IFBLK},
134 	{"p24", 40, S_IFBLK},
135 	{"p25", 41, S_IFBLK},
136 	{"p26", 42, S_IFBLK},
137 	{"p27", 43, S_IFBLK},
138 	{"p28", 44, S_IFBLK},
139 	{"p29", 45, S_IFBLK},
140 	{"p30", 46, S_IFBLK},
141 	{"p31", 47, S_IFBLK},
142 	{"p32", 48, S_IFBLK},
143 	{"p33", 49, S_IFBLK},
144 	{"p34", 50, S_IFBLK},
145 	{"p35", 51, S_IFBLK},
146 	{"p36", 52, S_IFBLK},
147 	{"p5,raw", 21, S_IFCHR},
148 	{"p6,raw", 22, S_IFCHR},
149 	{"p7,raw", 23, S_IFCHR},
150 	{"p8,raw", 24, S_IFCHR},
151 	{"p9,raw", 25, S_IFCHR},
152 	{"p10,raw", 26, S_IFCHR},
153 	{"p11,raw", 27, S_IFCHR},
154 	{"p12,raw", 28, S_IFCHR},
155 	{"p13,raw", 29, S_IFCHR},
156 	{"p14,raw", 30, S_IFCHR},
157 	{"p15,raw", 31, S_IFCHR},
158 	{"p16,raw", 32, S_IFCHR},
159 	{"p17,raw", 33, S_IFCHR},
160 	{"p18,raw", 34, S_IFCHR},
161 	{"p19,raw", 35, S_IFCHR},
162 	{"p20,raw", 36, S_IFCHR},
163 	{"p21,raw", 37, S_IFCHR},
164 	{"p22,raw", 38, S_IFCHR},
165 	{"p23,raw", 39, S_IFCHR},
166 	{"p24,raw", 40, S_IFCHR},
167 	{"p25,raw", 41, S_IFCHR},
168 	{"p26,raw", 42, S_IFCHR},
169 	{"p27,raw", 43, S_IFCHR},
170 	{"p28,raw", 44, S_IFCHR},
171 	{"p29,raw", 45, S_IFCHR},
172 	{"p30,raw", 46, S_IFCHR},
173 	{"p31,raw", 47, S_IFCHR},
174 	{"p32,raw", 48, S_IFCHR},
175 	{"p33,raw", 49, S_IFCHR},
176 	{"p34,raw", 50, S_IFCHR},
177 	{"p35,raw", 51, S_IFCHR},
178 	{"p36,raw", 52, S_IFCHR},
179 	{0}
180 };
181 #endif			/* defined(_FIRMWARE_NEEDS_FDISK) */
182 #endif			/* if defined(__i386) || defined(__amd64) */
183 
184 static struct driver_minor_data dk_minor_data_efi[] = {
185 	{"a", 0, S_IFBLK},
186 	{"b", 1, S_IFBLK},
187 	{"c", 2, S_IFBLK},
188 	{"d", 3, S_IFBLK},
189 	{"e", 4, S_IFBLK},
190 	{"f", 5, S_IFBLK},
191 	{"g", 6, S_IFBLK},
192 	{"wd", 7, S_IFBLK},
193 #if defined(_SUNOS_VTOC_16)
194 	{"i", 8, S_IFBLK},
195 	{"j", 9, S_IFBLK},
196 	{"k", 10, S_IFBLK},
197 	{"l", 11, S_IFBLK},
198 	{"m", 12, S_IFBLK},
199 	{"n", 13, S_IFBLK},
200 	{"o", 14, S_IFBLK},
201 	{"p", 15, S_IFBLK},
202 #endif			/* defined(_SUNOS_VTOC_16) */
203 #if defined(_FIRMWARE_NEEDS_FDISK)
204 	{"q", 16, S_IFBLK},
205 	{"r", 17, S_IFBLK},
206 	{"s", 18, S_IFBLK},
207 	{"t", 19, S_IFBLK},
208 	{"u", 20, S_IFBLK},
209 #endif			/* defined(_FIRMWARE_NEEDS_FDISK) */
210 	{"a,raw", 0, S_IFCHR},
211 	{"b,raw", 1, S_IFCHR},
212 	{"c,raw", 2, S_IFCHR},
213 	{"d,raw", 3, S_IFCHR},
214 	{"e,raw", 4, S_IFCHR},
215 	{"f,raw", 5, S_IFCHR},
216 	{"g,raw", 6, S_IFCHR},
217 	{"wd,raw", 7, S_IFCHR},
218 #if defined(_SUNOS_VTOC_16)
219 	{"i,raw", 8, S_IFCHR},
220 	{"j,raw", 9, S_IFCHR},
221 	{"k,raw", 10, S_IFCHR},
222 	{"l,raw", 11, S_IFCHR},
223 	{"m,raw", 12, S_IFCHR},
224 	{"n,raw", 13, S_IFCHR},
225 	{"o,raw", 14, S_IFCHR},
226 	{"p,raw", 15, S_IFCHR},
227 #endif			/* defined(_SUNOS_VTOC_16) */
228 #if defined(_FIRMWARE_NEEDS_FDISK)
229 	{"q,raw", 16, S_IFCHR},
230 	{"r,raw", 17, S_IFCHR},
231 	{"s,raw", 18, S_IFCHR},
232 	{"t,raw", 19, S_IFCHR},
233 	{"u,raw", 20, S_IFCHR},
234 #endif			/* defined(_FIRMWARE_NEEDS_FDISK) */
235 	{0}
236 };
237 
238 /*
239  * Declare the dynamic properties implemented in prop_op(9E) implementation
240  * that we want to have show up in a di_init(3DEVINFO) device tree snapshot
241  * of drivers that call cmlb_attach().
242  */
243 static i_ddi_prop_dyn_t cmlb_prop_dyn[] = {
244 	{"Nblocks",		DDI_PROP_TYPE_INT64,	S_IFBLK},
245 	{"Size",		DDI_PROP_TYPE_INT64,	S_IFCHR},
246 	{"device-nblocks",	DDI_PROP_TYPE_INT64},
247 	{"device-blksize",	DDI_PROP_TYPE_INT},
248 	{"device-solid-state",	DDI_PROP_TYPE_INT},
249 	{NULL}
250 };
251 
252 /*
253  * This implies an upper limit of 8192 GPT partitions
254  * in one transfer for GUID Partition Entry Array.
255  */
256 len_t cmlb_tg_max_efi_xfer = 1024 * 1024;
257 
258 /*
259  * External kernel interfaces
260  */
261 extern struct mod_ops mod_miscops;
262 
263 extern int ddi_create_internal_pathname(dev_info_t *dip, char *name,
264     int spec_type, minor_t minor_num);
265 
266 /*
267  * Global buffer and mutex for debug logging
268  */
269 static char	cmlb_log_buffer[1024];
270 static kmutex_t	cmlb_log_mutex;
271 
272 
273 struct cmlb_lun *cmlb_debug_cl = NULL;
274 uint_t cmlb_level_mask = 0x0;
275 
276 int cmlb_rot_delay = 4;	/* default rotational delay */
277 
278 static struct modlmisc modlmisc = {
279 	&mod_miscops,   /* Type of module */
280 	"Common Labeling module"
281 };
282 
283 static struct modlinkage modlinkage = {
284 	MODREV_1, (void *)&modlmisc, NULL
285 };
286 
287 /* Local function prototypes */
288 static dev_t cmlb_make_device(struct cmlb_lun *cl);
289 static int cmlb_validate_geometry(struct cmlb_lun *cl, boolean_t forcerevalid,
290     int flags, void *tg_cookie);
291 static void cmlb_resync_geom_caches(struct cmlb_lun *cl, diskaddr_t capacity,
292     void *tg_cookie);
293 static int cmlb_read_fdisk(struct cmlb_lun *cl, diskaddr_t capacity,
294     void *tg_cookie);
295 static void cmlb_swap_efi_gpt(efi_gpt_t *e);
296 static void cmlb_swap_efi_gpe(int nparts, efi_gpe_t *p);
297 static int cmlb_validate_efi(efi_gpt_t *labp);
298 static int cmlb_use_efi(struct cmlb_lun *cl, diskaddr_t capacity, int flags,
299     void *tg_cookie);
300 static void cmlb_build_default_label(struct cmlb_lun *cl, void *tg_cookie);
301 static int  cmlb_uselabel(struct cmlb_lun *cl,  struct dk_label *l, int flags);
302 #if defined(_SUNOS_VTOC_8)
303 static void cmlb_build_user_vtoc(struct cmlb_lun *cl, struct vtoc *user_vtoc);
304 #endif
305 static int cmlb_build_label_vtoc(struct cmlb_lun *cl, struct vtoc *user_vtoc);
306 static int cmlb_write_label(struct cmlb_lun *cl, void *tg_cookie);
307 static int cmlb_set_vtoc(struct cmlb_lun *cl, struct dk_label *dkl,
308     void *tg_cookie);
309 static void cmlb_clear_efi(struct cmlb_lun *cl, void *tg_cookie);
310 static void cmlb_clear_vtoc(struct cmlb_lun *cl, void *tg_cookie);
311 static void cmlb_setup_default_geometry(struct cmlb_lun *cl, void *tg_cookie);
312 static int cmlb_create_minor_nodes(struct cmlb_lun *cl);
313 static int cmlb_check_update_blockcount(struct cmlb_lun *cl, void *tg_cookie);
314 static boolean_t cmlb_check_efi_mbr(uchar_t *buf, boolean_t *is_mbr);
315 
316 #if defined(__i386) || defined(__amd64)
317 static int cmlb_update_fdisk_and_vtoc(struct cmlb_lun *cl, void *tg_cookie);
318 #endif
319 
320 #if defined(_FIRMWARE_NEEDS_FDISK)
321 static boolean_t  cmlb_has_max_chs_vals(struct ipart *fdp);
322 #endif
323 
324 #if defined(_SUNOS_VTOC_16)
325 static void cmlb_convert_geometry(struct cmlb_lun *cl, diskaddr_t capacity,
326     struct dk_geom *cl_g, void *tg_cookie);
327 #endif
328 
329 static int cmlb_dkio_get_geometry(struct cmlb_lun *cl, caddr_t arg, int flag,
330     void *tg_cookie);
331 static int cmlb_dkio_set_geometry(struct cmlb_lun *cl, caddr_t arg, int flag);
332 static int cmlb_dkio_get_partition(struct cmlb_lun *cl, caddr_t arg, int flag,
333     void *tg_cookie);
334 static int cmlb_dkio_set_partition(struct cmlb_lun *cl, caddr_t arg, int flag);
335 static int cmlb_dkio_get_efi(struct cmlb_lun *cl, caddr_t arg, int flag,
336     void *tg_cookie);
337 static int cmlb_dkio_set_efi(struct cmlb_lun *cl, dev_t dev, caddr_t arg,
338     int flag, void *tg_cookie);
339 static int cmlb_dkio_get_vtoc(struct cmlb_lun *cl, caddr_t arg, int flag,
340     void *tg_cookie);
341 static int cmlb_dkio_get_extvtoc(struct cmlb_lun *cl, caddr_t arg, int flag,
342     void *tg_cookie);
343 static int cmlb_dkio_set_vtoc(struct cmlb_lun *cl, dev_t dev, caddr_t arg,
344     int flag, void *tg_cookie);
345 static int cmlb_dkio_set_extvtoc(struct cmlb_lun *cl, dev_t dev, caddr_t arg,
346     int flag, void *tg_cookie);
347 static int cmlb_dkio_get_mboot(struct cmlb_lun *cl, caddr_t arg, int flag,
348     void *tg_cookie);
349 static int cmlb_dkio_set_mboot(struct cmlb_lun *cl, caddr_t arg, int flag,
350     void *tg_cookie);
351 static int cmlb_dkio_partition(struct cmlb_lun *cl, caddr_t arg, int flag,
352     void *tg_cookie);
353 
354 #if defined(__i386) || defined(__amd64)
355 static int cmlb_dkio_set_ext_part(struct cmlb_lun *cl, caddr_t arg, int flag,
356     void *tg_cookie);
357 static int cmlb_validate_ext_part(struct cmlb_lun *cl, int part, int epart,
358     uint32_t start, uint32_t size);
359 static int cmlb_is_linux_swap(struct cmlb_lun *cl, uint32_t part_start,
360     void *tg_cookie);
361 static int cmlb_dkio_get_virtgeom(struct cmlb_lun *cl, caddr_t arg, int flag);
362 static int cmlb_dkio_get_phygeom(struct cmlb_lun *cl, caddr_t  arg, int flag,
363     void *tg_cookie);
364 static int cmlb_dkio_partinfo(struct cmlb_lun *cl, dev_t dev, caddr_t arg,
365     int flag);
366 static int cmlb_dkio_extpartinfo(struct cmlb_lun *cl, dev_t dev, caddr_t arg,
367     int flag);
368 #endif
369 
370 static void cmlb_dbg(uint_t comp, struct cmlb_lun *cl, const char *fmt, ...);
371 static void cmlb_v_log(dev_info_t *dev, const char *label, uint_t level,
372     const char *fmt, va_list ap);
373 static void cmlb_log(dev_info_t *dev, const char *label, uint_t level,
374     const char *fmt, ...);
375 
376 int
377 _init(void)
378 {
379 	mutex_init(&cmlb_log_mutex, NULL, MUTEX_DRIVER, NULL);
380 	return (mod_install(&modlinkage));
381 }
382 
383 int
384 _info(struct modinfo *modinfop)
385 {
386 	return (mod_info(&modlinkage, modinfop));
387 }
388 
389 int
390 _fini(void)
391 {
392 	int err;
393 
394 	if ((err = mod_remove(&modlinkage)) != 0) {
395 		return (err);
396 	}
397 
398 	mutex_destroy(&cmlb_log_mutex);
399 	return (err);
400 }
401 
402 /*
403  * cmlb_dbg is used for debugging to log additional info
404  * Level of output is controlled via cmlb_level_mask setting.
405  */
406 static void
407 cmlb_dbg(uint_t comp, struct cmlb_lun *cl, const char *fmt, ...)
408 {
409 	va_list		ap;
410 	dev_info_t	*dev;
411 	uint_t		level_mask = 0;
412 
413 	ASSERT(cl != NULL);
414 	dev = CMLB_DEVINFO(cl);
415 	ASSERT(dev != NULL);
416 	/*
417 	 * Filter messages based on the global component and level masks,
418 	 * also print if cl matches the value of cmlb_debug_cl, or if
419 	 * cmlb_debug_cl is set to NULL.
420 	 */
421 	if (comp & CMLB_TRACE)
422 		level_mask |= CMLB_LOGMASK_TRACE;
423 
424 	if (comp & CMLB_INFO)
425 		level_mask |= CMLB_LOGMASK_INFO;
426 
427 	if (comp & CMLB_ERROR)
428 		level_mask |= CMLB_LOGMASK_ERROR;
429 
430 	if ((cmlb_level_mask & level_mask) &&
431 	    ((cmlb_debug_cl == NULL) || (cmlb_debug_cl == cl))) {
432 		va_start(ap, fmt);
433 		cmlb_v_log(dev, CMLB_LABEL(cl), CE_CONT, fmt, ap);
434 		va_end(ap);
435 	}
436 }
437 
438 /*
439  * cmlb_log is basically a duplicate of scsi_log. It is redefined here
440  * so that this module does not depend on scsi module.
441  */
442 static void
443 cmlb_log(dev_info_t *dev, const char *label, uint_t level, const char *fmt, ...)
444 {
445 	va_list		ap;
446 
447 	va_start(ap, fmt);
448 	cmlb_v_log(dev, label, level, fmt, ap);
449 	va_end(ap);
450 }
451 
452 static void
453 cmlb_v_log(dev_info_t *dev, const char *label, uint_t level, const char *fmt,
454     va_list ap)
455 {
456 	static char 	name[256];
457 	int 		log_only = 0;
458 	int 		boot_only = 0;
459 	int 		console_only = 0;
460 
461 	mutex_enter(&cmlb_log_mutex);
462 
463 	if (dev) {
464 		if (level == CE_PANIC || level == CE_WARN ||
465 		    level == CE_NOTE) {
466 			(void) sprintf(name, "%s (%s%d):\n",
467 			    ddi_pathname(dev, cmlb_log_buffer),
468 			    label, ddi_get_instance(dev));
469 		} else {
470 			name[0] = '\0';
471 		}
472 	} else {
473 		(void) sprintf(name, "%s:", label);
474 	}
475 
476 	(void) vsprintf(cmlb_log_buffer, fmt, ap);
477 
478 	switch (cmlb_log_buffer[0]) {
479 	case '!':
480 		log_only = 1;
481 		break;
482 	case '?':
483 		boot_only = 1;
484 		break;
485 	case '^':
486 		console_only = 1;
487 		break;
488 	}
489 
490 	switch (level) {
491 	case CE_NOTE:
492 		level = CE_CONT;
493 		/* FALLTHROUGH */
494 	case CE_CONT:
495 	case CE_WARN:
496 	case CE_PANIC:
497 		if (boot_only) {
498 			cmn_err(level, "?%s\t%s", name, &cmlb_log_buffer[1]);
499 		} else if (console_only) {
500 			cmn_err(level, "^%s\t%s", name, &cmlb_log_buffer[1]);
501 		} else if (log_only) {
502 			cmn_err(level, "!%s\t%s", name, &cmlb_log_buffer[1]);
503 		} else {
504 			cmn_err(level, "%s\t%s", name, cmlb_log_buffer);
505 		}
506 		break;
507 	case CE_IGNORE:
508 		break;
509 	default:
510 		cmn_err(CE_CONT, "^DEBUG: %s\t%s", name, cmlb_log_buffer);
511 		break;
512 	}
513 	mutex_exit(&cmlb_log_mutex);
514 }
515 
516 
517 /*
518  * cmlb_alloc_handle:
519  *
520  *	Allocates a handle.
521  *
522  * Arguments:
523  *	cmlbhandlep	pointer to handle
524  *
525  * Notes:
526  *	Allocates a handle and stores the allocated handle in the area
527  *	pointed to by cmlbhandlep
528  *
529  * Context:
530  *	Kernel thread only (can sleep).
531  */
532 void
533 cmlb_alloc_handle(cmlb_handle_t *cmlbhandlep)
534 {
535 	struct cmlb_lun 	*cl;
536 
537 	cl = kmem_zalloc(sizeof (struct cmlb_lun), KM_SLEEP);
538 	ASSERT(cmlbhandlep != NULL);
539 
540 	cl->cl_state = CMLB_INITED;
541 	cl->cl_def_labeltype = CMLB_LABEL_UNDEF;
542 	mutex_init(CMLB_MUTEX(cl), NULL, MUTEX_DRIVER, NULL);
543 
544 	*cmlbhandlep = (cmlb_handle_t)(cl);
545 }
546 
547 /*
548  * cmlb_free_handle
549  *
550  *	Frees handle.
551  *
552  * Arguments:
553  *	cmlbhandlep	pointer to handle
554  */
555 void
556 cmlb_free_handle(cmlb_handle_t *cmlbhandlep)
557 {
558 	struct cmlb_lun 	*cl;
559 
560 	cl = (struct cmlb_lun *)*cmlbhandlep;
561 	if (cl != NULL) {
562 		mutex_destroy(CMLB_MUTEX(cl));
563 		kmem_free(cl, sizeof (struct cmlb_lun));
564 	}
565 
566 }
567 
568 /*
569  * cmlb_attach:
570  *
571  *	Attach handle to device, create minor nodes for device.
572  *
573  * Arguments:
574  * 	devi		pointer to device's dev_info structure.
575  * 	tgopsp		pointer to array of functions cmlb can use to callback
576  *			to target driver.
577  *
578  *	device_type	Peripheral device type as defined in
579  *			scsi/generic/inquiry.h
580  *
581  *	is_removable	whether or not device is removable.
582  *
583  *	is_hotpluggable	whether or not device is hotpluggable.
584  *
585  *	node_type	minor node type (as used by ddi_create_minor_node)
586  *
587  *	alter_behavior
588  *			bit flags:
589  *
590  *			CMLB_CREATE_ALTSLICE_VTOC_16_DTYPE_DIRECT: create
591  *			an alternate slice for the default label, if
592  *			device type is DTYPE_DIRECT an architectures default
593  *			label type is VTOC16.
594  *			Otherwise alternate slice will no be created.
595  *
596  *
597  *			CMLB_FAKE_GEOM_LABEL_IOCTLS_VTOC8: report a default
598  *			geometry and label for DKIOCGGEOM and DKIOCGVTOC
599  *			on architecture with VTOC8 label types.
600  *
601  * 			CMLB_OFF_BY_ONE: do the workaround for legacy off-by-
602  *                      one bug in obtaining capacity (in sd):
603  *			SCSI READ_CAPACITY command returns the LBA number of the
604  *			last logical block, but sd once treated this number as
605  *			disks' capacity on x86 platform. And LBAs are addressed
606  *			based 0. So the last block was lost on x86 platform.
607  *
608  *			Now, we remove this workaround. In order for present sd
609  *			driver to work with disks which are labeled/partitioned
610  *			via previous sd, we add workaround as follows:
611  *
612  *			1) Locate backup EFI label: cmlb searches the next to
613  *			   last
614  *			   block for backup EFI label. If fails, it will
615  *			   turn to the last block for backup EFI label;
616  *
617  *			2) Clear backup EFI label: cmlb first search the last
618  *			   block for backup EFI label, and will search the
619  *			   next to last block only if failed for the last
620  *			   block.
621  *
622  *			3) Calculate geometry:refer to cmlb_convert_geometry()
623  *			   If capacity increasing by 1 causes disks' capacity
624  *			   to cross over the limits in geometry calculation,
625  *			   geometry info will change. This will raise an issue:
626  *			   In case that primary VTOC label is destroyed, format
627  *			   commandline can restore it via backup VTOC labels.
628  *			   And format locates backup VTOC labels by use of
629  *			   geometry. So changing geometry will
630  *			   prevent format from finding backup VTOC labels. To
631  *			   eliminate this side effect for compatibility,
632  *			   sd uses (capacity -1) to calculate geometry;
633  *
634  *			4) 1TB disks: some important data structures use
635  *			   32-bit signed long/int (for example, daddr_t),
636  *			   so that sd doesn't support a disk with capacity
637  *			   larger than 1TB on 32-bit platform. However,
638  *			   for exactly 1TB disk, it was treated as (1T - 512)B
639  *			   in the past, and could have valid Solaris
640  *			   partitions. To workaround this, if an exactly 1TB
641  *			   disk has Solaris fdisk partition, it will be allowed
642  *			   to work with sd.
643  *
644  *
645  *
646  *			CMLB_FAKE_LABEL_ONE_PARTITION: create s0 and s2 covering
647  *			the entire disk, if there is no valid partition info.
648  *			If there is a valid Solaris partition, s0 and s2 will
649  *			only cover the entire Solaris partition.
650  *
651  *			CMLB_CREATE_P0_MINOR_NODE: create p0 node covering
652  *			the entire disk. Used by lofi to ensure presence of
653  *			whole disk device node in case of LOFI_MAP_FILE ioctl.
654  *
655  *	cmlbhandle	cmlb handle associated with device
656  *
657  *	tg_cookie	cookie from target driver to be passed back to target
658  *			driver when we call back to it through tg_ops.
659  *
660  * Notes:
661  *	Assumes a default label based on capacity for non-removable devices.
662  *	If capacity > 1TB, EFI is assumed otherwise VTOC (default VTOC
663  *	for the architecture).
664  *
665  *	For removable devices, default label type is assumed to be VTOC
666  *	type. Create minor nodes based on a default label type.
667  *	Label on the media is not validated.
668  *	minor number consists of:
669  *		if _SUNOS_VTOC_8 is defined
670  *			lowest 3 bits is taken as partition number
671  *			the rest is instance number
672  *		if _SUNOS_VTOC_16 is defined
673  *			lowest 6 bits is taken as partition number
674  *			the rest is instance number
675  *
676  *
677  * Return values:
678  *	0 	Success
679  * 	ENXIO 	creating minor nodes failed.
680  *	EINVAL  invalid arg, unsupported tg_ops version
681  */
682 int
683 cmlb_attach(dev_info_t *devi, cmlb_tg_ops_t *tgopsp, int device_type,
684     boolean_t is_removable, boolean_t is_hotpluggable, char *node_type,
685     int alter_behavior, cmlb_handle_t cmlbhandle, void *tg_cookie)
686 {
687 
688 	struct cmlb_lun	*cl = (struct cmlb_lun *)cmlbhandle;
689 	diskaddr_t	cap;
690 	int		status;
691 
692 	ASSERT(VALID_BOOLEAN(is_removable));
693 	ASSERT(VALID_BOOLEAN(is_hotpluggable));
694 
695 	if (tgopsp->tg_version < TG_DK_OPS_VERSION_1)
696 		return (EINVAL);
697 
698 	mutex_enter(CMLB_MUTEX(cl));
699 
700 	CMLB_DEVINFO(cl) = devi;
701 	cl->cmlb_tg_ops = tgopsp;
702 	cl->cl_device_type = device_type;
703 	cl->cl_is_removable = is_removable;
704 	cl->cl_is_hotpluggable = is_hotpluggable;
705 	cl->cl_node_type = node_type;
706 	cl->cl_sys_blocksize = DEV_BSIZE;
707 	cl->cl_f_geometry_is_valid = B_FALSE;
708 	cl->cl_def_labeltype = CMLB_LABEL_VTOC;
709 	cl->cl_alter_behavior = alter_behavior;
710 	cl->cl_reserved = -1;
711 	cl->cl_msglog_flag |= CMLB_ALLOW_2TB_WARN;
712 #if defined(__i386) || defined(__amd64)
713 	cl->cl_logical_drive_count = 0;
714 #endif
715 
716 	if (!is_removable) {
717 		mutex_exit(CMLB_MUTEX(cl));
718 		status = DK_TG_GETCAP(cl, &cap, tg_cookie);
719 		mutex_enter(CMLB_MUTEX(cl));
720 		if (status == 0 && cap > CMLB_EXTVTOC_LIMIT) {
721 			/* set default EFI if > 2TB */
722 			cl->cl_def_labeltype = CMLB_LABEL_EFI;
723 		}
724 	}
725 
726 	/* create minor nodes based on default label type */
727 	cl->cl_last_labeltype = CMLB_LABEL_UNDEF;
728 	cl->cl_cur_labeltype = CMLB_LABEL_UNDEF;
729 
730 	if (cmlb_create_minor_nodes(cl) != 0) {
731 		mutex_exit(CMLB_MUTEX(cl));
732 		return (ENXIO);
733 	}
734 
735 	/* Define the dynamic properties for devinfo spapshots. */
736 	i_ddi_prop_dyn_driver_set(CMLB_DEVINFO(cl), cmlb_prop_dyn);
737 
738 	cl->cl_state = CMLB_ATTACHED;
739 
740 	mutex_exit(CMLB_MUTEX(cl));
741 	return (0);
742 }
743 
744 /*
745  * cmlb_detach:
746  *
747  * Invalidate in-core labeling data and remove all minor nodes for
748  * the device associate with handle.
749  *
750  * Arguments:
751  *	cmlbhandle	cmlb handle associated with device.
752  *
753  *	tg_cookie	cookie from target driver to be passed back to target
754  *			driver when we call back to it through tg_ops.
755  *
756  */
757 /*ARGSUSED1*/
758 void
759 cmlb_detach(cmlb_handle_t cmlbhandle, void *tg_cookie)
760 {
761 	struct cmlb_lun *cl = (struct cmlb_lun *)cmlbhandle;
762 
763 	mutex_enter(CMLB_MUTEX(cl));
764 	cl->cl_def_labeltype = CMLB_LABEL_UNDEF;
765 	cl->cl_f_geometry_is_valid = B_FALSE;
766 	ddi_remove_minor_node(CMLB_DEVINFO(cl), NULL);
767 	i_ddi_prop_dyn_driver_set(CMLB_DEVINFO(cl), NULL);
768 	cl->cl_state = CMLB_INITED;
769 	mutex_exit(CMLB_MUTEX(cl));
770 }
771 
772 /*
773  * cmlb_validate:
774  *
775  *	Validates label.
776  *
777  * Arguments
778  *	cmlbhandle	cmlb handle associated with device.
779  *
780  *	flags		operation flags. used for verbosity control
781  *
782  *	tg_cookie	cookie from target driver to be passed back to target
783  *			driver when we call back to it through tg_ops.
784  *
785  *
786  * Notes:
787  *	If new label type is different from the current, adjust minor nodes
788  *	accordingly.
789  *
790  * Return values:
791  *	0		success
792  *			Note: having fdisk but no solaris partition is assumed
793  *			success.
794  *
795  *	ENOMEM		memory allocation failed
796  *	EIO		i/o errors during read or get capacity
797  * 	EACCESS		reservation conflicts
798  * 	EINVAL		label was corrupt, or no default label was assumed
799  *	ENXIO		invalid handle
800  */
801 int
802 cmlb_validate(cmlb_handle_t cmlbhandle, int flags, void *tg_cookie)
803 {
804 	struct cmlb_lun *cl = (struct cmlb_lun *)cmlbhandle;
805 	int 		rval;
806 	int  		ret = 0;
807 
808 	/*
809 	 * Temp work-around checking cl for NULL since there is a bug
810 	 * in sd_detach calling this routine from taskq_dispatch
811 	 * inited function.
812 	 */
813 	if (cl == NULL)
814 		return (ENXIO);
815 
816 	mutex_enter(CMLB_MUTEX(cl));
817 	if (cl->cl_state < CMLB_ATTACHED) {
818 		mutex_exit(CMLB_MUTEX(cl));
819 		return (ENXIO);
820 	}
821 
822 	rval = cmlb_validate_geometry((struct cmlb_lun *)cmlbhandle, B_TRUE,
823 	    flags, tg_cookie);
824 
825 	if (rval == ENOTSUP) {
826 		if (cl->cl_f_geometry_is_valid) {
827 			cl->cl_cur_labeltype = CMLB_LABEL_EFI;
828 			ret = 0;
829 		} else {
830 			ret = EINVAL;
831 		}
832 	} else {
833 		ret = rval;
834 		if (ret == 0)
835 			cl->cl_cur_labeltype = CMLB_LABEL_VTOC;
836 	}
837 
838 	if (ret == 0)
839 		(void) cmlb_create_minor_nodes(cl);
840 
841 	mutex_exit(CMLB_MUTEX(cl));
842 	return (ret);
843 }
844 
845 /*
846  * cmlb_invalidate:
847  *	Invalidate in core label data
848  *
849  * Arguments:
850  *	cmlbhandle	cmlb handle associated with device.
851  *	tg_cookie	cookie from target driver to be passed back to target
852  *			driver when we call back to it through tg_ops.
853  */
854 /*ARGSUSED1*/
855 void
856 cmlb_invalidate(cmlb_handle_t cmlbhandle, void *tg_cookie)
857 {
858 	struct cmlb_lun *cl = (struct cmlb_lun *)cmlbhandle;
859 
860 	if (cl == NULL)
861 		return;
862 
863 	mutex_enter(CMLB_MUTEX(cl));
864 	cl->cl_f_geometry_is_valid = B_FALSE;
865 	mutex_exit(CMLB_MUTEX(cl));
866 }
867 
868 /*
869  * cmlb_is_valid
870  * 	Get status on whether the incore label/geom data is valid
871  *
872  * Arguments:
873  *	cmlbhandle      cmlb handle associated with device.
874  *
875  * Return values:
876  *	B_TRUE if incore label/geom data is valid.
877  *	B_FALSE otherwise.
878  *
879  */
880 
881 
882 boolean_t
883 cmlb_is_valid(cmlb_handle_t cmlbhandle)
884 {
885 	struct cmlb_lun *cl = (struct cmlb_lun *)cmlbhandle;
886 
887 	if (cmlbhandle == NULL)
888 		return (B_FALSE);
889 
890 	return (cl->cl_f_geometry_is_valid);
891 
892 }
893 
894 
895 
896 /*
897  * cmlb_close:
898  *
899  * Close the device, revert to a default label minor node for the device,
900  * if it is removable.
901  *
902  * Arguments:
903  *	cmlbhandle	cmlb handle associated with device.
904  *
905  *	tg_cookie	cookie from target driver to be passed back to target
906  *			driver when we call back to it through tg_ops.
907  * Return values:
908  *	0	Success
909  * 	ENXIO	Re-creating minor node failed.
910  */
911 /*ARGSUSED1*/
912 int
913 cmlb_close(cmlb_handle_t cmlbhandle, void *tg_cookie)
914 {
915 	struct cmlb_lun *cl = (struct cmlb_lun *)cmlbhandle;
916 
917 	mutex_enter(CMLB_MUTEX(cl));
918 	cl->cl_f_geometry_is_valid = B_FALSE;
919 
920 	/* revert to default minor node for this device */
921 	if (ISREMOVABLE(cl)) {
922 		cl->cl_cur_labeltype = CMLB_LABEL_UNDEF;
923 		(void) cmlb_create_minor_nodes(cl);
924 	}
925 
926 	mutex_exit(CMLB_MUTEX(cl));
927 	return (0);
928 }
929 
930 /*
931  * cmlb_get_devid_block:
932  *	 get the block number where device id is stored.
933  *
934  * Arguments:
935  *	cmlbhandle	cmlb handle associated with device.
936  *	devidblockp	pointer to block number.
937  *	tg_cookie	cookie from target driver to be passed back to target
938  *			driver when we call back to it through tg_ops.
939  *
940  * Notes:
941  *	It stores the block number of device id in the area pointed to
942  *	by devidblockp.
943  * 	with the block number of device id.
944  *
945  * Return values:
946  *	0	success
947  *	EINVAL 	device id does not apply to current label type.
948  */
949 /*ARGSUSED2*/
950 int
951 cmlb_get_devid_block(cmlb_handle_t cmlbhandle, diskaddr_t *devidblockp,
952     void *tg_cookie)
953 {
954 	daddr_t			spc, blk, head, cyl;
955 	struct cmlb_lun *cl = (struct cmlb_lun *)cmlbhandle;
956 
957 	mutex_enter(CMLB_MUTEX(cl));
958 	if (cl->cl_state < CMLB_ATTACHED) {
959 		mutex_exit(CMLB_MUTEX(cl));
960 		return (EINVAL);
961 	}
962 
963 	if ((!cl->cl_f_geometry_is_valid) ||
964 	    (cl->cl_solaris_size < DK_LABEL_LOC)) {
965 		mutex_exit(CMLB_MUTEX(cl));
966 		return (EINVAL);
967 	}
968 
969 	if (cl->cl_cur_labeltype == CMLB_LABEL_EFI) {
970 		if (cl->cl_reserved != -1) {
971 			blk = cl->cl_map[cl->cl_reserved].dkl_cylno;
972 		} else {
973 			mutex_exit(CMLB_MUTEX(cl));
974 			return (EINVAL);
975 		}
976 	} else {
977 		/* if the disk is unlabeled, don't write a devid to it */
978 		if (cl->cl_label_from_media != CMLB_LABEL_VTOC) {
979 			mutex_exit(CMLB_MUTEX(cl));
980 			return (EINVAL);
981 		}
982 
983 		/* this geometry doesn't allow us to write a devid */
984 		if (cl->cl_g.dkg_acyl < 2) {
985 			mutex_exit(CMLB_MUTEX(cl));
986 			return (EINVAL);
987 		}
988 
989 		/*
990 		 * Subtract 2 guarantees that the next to last cylinder
991 		 * is used
992 		 */
993 		cyl  = cl->cl_g.dkg_ncyl  + cl->cl_g.dkg_acyl - 2;
994 		spc  = cl->cl_g.dkg_nhead * cl->cl_g.dkg_nsect;
995 		head = cl->cl_g.dkg_nhead - 1;
996 		blk  = cl->cl_solaris_offset +
997 		    (cyl * (spc - cl->cl_g.dkg_apc)) +
998 		    (head * cl->cl_g.dkg_nsect) + 1;
999 	}
1000 
1001 	*devidblockp = blk;
1002 	mutex_exit(CMLB_MUTEX(cl));
1003 	return (0);
1004 }
1005 
1006 /*
1007  * cmlb_partinfo:
1008  *	Get partition info for specified partition number.
1009  *
1010  * Arguments:
1011  *	cmlbhandle	cmlb handle associated with device.
1012  *	part		partition number
1013  *	nblocksp	pointer to number of blocks
1014  *	startblockp	pointer to starting block
1015  *	partnamep	pointer to name of partition
1016  *	tagp		pointer to tag info
1017  *	tg_cookie	cookie from target driver to be passed back to target
1018  *			driver when we call back to it through tg_ops.
1019  *
1020  *
1021  * Notes:
1022  *	If in-core label is not valid, this functions tries to revalidate
1023  *	the label. If label is valid, it stores the total number of blocks
1024  *	in this partition in the area pointed to by nblocksp, starting
1025  *	block number in area pointed to by startblockp,  pointer to partition
1026  *	name in area pointed to by partnamep, and tag value in area
1027  *	pointed by tagp.
1028  *	For EFI labels, tag value will be set to 0.
1029  *
1030  *	For all nblocksp, startblockp and partnamep, tagp, a value of NULL
1031  *	indicates the corresponding info is not requested.
1032  *
1033  *
1034  * Return values:
1035  *	0	success
1036  *	EINVAL  no valid label or requested partition number is invalid.
1037  *
1038  */
1039 int
1040 cmlb_partinfo(cmlb_handle_t cmlbhandle, int part, diskaddr_t *nblocksp,
1041     diskaddr_t *startblockp, char **partnamep, uint16_t *tagp, void *tg_cookie)
1042 {
1043 
1044 	struct cmlb_lun *cl = (struct cmlb_lun *)cmlbhandle;
1045 	int rval;
1046 #if defined(__i386) || defined(__amd64)
1047 	int ext_part;
1048 #endif
1049 
1050 	ASSERT(cl != NULL);
1051 	mutex_enter(CMLB_MUTEX(cl));
1052 	if (cl->cl_state < CMLB_ATTACHED) {
1053 		mutex_exit(CMLB_MUTEX(cl));
1054 		return (EINVAL);
1055 	}
1056 
1057 	if (part  < 0 || part >= MAXPART) {
1058 		rval = EINVAL;
1059 	} else {
1060 		if (!cl->cl_f_geometry_is_valid)
1061 			(void) cmlb_validate_geometry((struct cmlb_lun *)cl,
1062 			    B_FALSE, 0, tg_cookie);
1063 
1064 		if (((!cl->cl_f_geometry_is_valid) ||
1065 		    (part < NDKMAP && cl->cl_solaris_size == 0)) &&
1066 		    (part != P0_RAW_DISK)) {
1067 			rval = EINVAL;
1068 		} else {
1069 			if (startblockp != NULL)
1070 				*startblockp = (diskaddr_t)cl->cl_offset[part];
1071 
1072 			if (nblocksp != NULL)
1073 				*nblocksp = (diskaddr_t)
1074 				    cl->cl_map[part].dkl_nblk;
1075 
1076 			if (tagp != NULL)
1077 				*tagp =
1078 				    ((cl->cl_cur_labeltype == CMLB_LABEL_EFI) ||
1079 				    (part >= NDKMAP)) ? V_UNASSIGNED :
1080 				    cl->cl_vtoc.v_part[part].p_tag;
1081 			rval = 0;
1082 		}
1083 
1084 		/* consistent with behavior of sd for getting minor name */
1085 		if (partnamep != NULL) {
1086 #if defined(__i386) || defined(__amd64)
1087 #if defined(_FIRMWARE_NEEDS_FDISK)
1088 		if (part > FDISK_P4) {
1089 			ext_part = part-FDISK_P4-1;
1090 			*partnamep = dk_ext_minor_data[ext_part].name;
1091 		} else
1092 #endif
1093 #endif
1094 			*partnamep = dk_minor_data[part].name;
1095 		}
1096 
1097 	}
1098 
1099 	mutex_exit(CMLB_MUTEX(cl));
1100 	return (rval);
1101 }
1102 
1103 /*
1104  * cmlb_efi_label_capacity:
1105  *	Get capacity stored in EFI disk label.
1106  *
1107  * Arguments:
1108  *	cmlbhandle	cmlb handle associated with device.
1109  *	capacity	pointer to capacity stored in EFI disk label.
1110  *	tg_cookie	cookie from target driver to be passed back to target
1111  *			driver when we call back to it through tg_ops.
1112  *
1113  *
1114  * Notes:
1115  *	If in-core label is not valid, this functions tries to revalidate
1116  *	the label. If label is valid and is an EFI label, it stores the capacity
1117  *      in disk label in the area pointed to by capacity.
1118  *
1119  *
1120  * Return values:
1121  *	0	success
1122  *	EINVAL  no valid EFI label or capacity is NULL.
1123  *
1124  */
1125 int
1126 cmlb_efi_label_capacity(cmlb_handle_t cmlbhandle, diskaddr_t *capacity,
1127     void *tg_cookie)
1128 {
1129 	struct cmlb_lun *cl = (struct cmlb_lun *)cmlbhandle;
1130 	int rval;
1131 
1132 	ASSERT(cl != NULL);
1133 	mutex_enter(CMLB_MUTEX(cl));
1134 	if (cl->cl_state < CMLB_ATTACHED) {
1135 		mutex_exit(CMLB_MUTEX(cl));
1136 		return (EINVAL);
1137 	}
1138 
1139 	if (!cl->cl_f_geometry_is_valid)
1140 		(void) cmlb_validate_geometry((struct cmlb_lun *)cl, B_FALSE,
1141 		    0, tg_cookie);
1142 
1143 	if ((!cl->cl_f_geometry_is_valid) || (capacity == NULL) ||
1144 	    (cl->cl_cur_labeltype != CMLB_LABEL_EFI)) {
1145 		rval = EINVAL;
1146 	} else {
1147 		*capacity = (diskaddr_t)cl->cl_map[WD_NODE].dkl_nblk;
1148 		rval = 0;
1149 	}
1150 
1151 	mutex_exit(CMLB_MUTEX(cl));
1152 	return (rval);
1153 }
1154 
1155 /* Caller should make sure Test Unit Ready succeeds before calling this. */
1156 /*ARGSUSED*/
1157 int
1158 cmlb_ioctl(cmlb_handle_t cmlbhandle, dev_t dev, int cmd, intptr_t arg,
1159     int flag, cred_t *cred_p, int *rval_p, void *tg_cookie)
1160 {
1161 
1162 	int err;
1163 	struct cmlb_lun *cl;
1164 
1165 	cl = (struct cmlb_lun *)cmlbhandle;
1166 
1167 	ASSERT(cl != NULL);
1168 
1169 	mutex_enter(CMLB_MUTEX(cl));
1170 	if (cl->cl_state < CMLB_ATTACHED) {
1171 		mutex_exit(CMLB_MUTEX(cl));
1172 		return (EIO);
1173 	}
1174 
1175 	switch (cmd) {
1176 		case DKIOCSEXTVTOC:
1177 		case DKIOCSGEOM:
1178 		case DKIOCSETEFI:
1179 		case DKIOCSMBOOT:
1180 #if defined(__i386) || defined(__amd64)
1181 		case DKIOCSETEXTPART:
1182 #endif
1183 			break;
1184 		case DKIOCSVTOC:
1185 #if defined(__i386) || defined(__amd64)
1186 		case DKIOCPARTINFO:
1187 #endif
1188 			if (cl->cl_blockcount > CMLB_OLDVTOC_LIMIT) {
1189 				mutex_exit(CMLB_MUTEX(cl));
1190 				return (EOVERFLOW);
1191 			}
1192 			break;
1193 		default:
1194 			(void) cmlb_validate_geometry(cl, 1, CMLB_SILENT,
1195 			    tg_cookie);
1196 
1197 			switch (cmd) {
1198 			case DKIOCGVTOC:
1199 			case DKIOCGAPART:
1200 			case DKIOCSAPART:
1201 
1202 				if (cl->cl_label_from_media == CMLB_LABEL_EFI) {
1203 					/* GPT label on disk */
1204 					mutex_exit(CMLB_MUTEX(cl));
1205 					return (ENOTSUP);
1206 				} else if
1207 				    (cl->cl_blockcount > CMLB_OLDVTOC_LIMIT) {
1208 					mutex_exit(CMLB_MUTEX(cl));
1209 					return (EOVERFLOW);
1210 				}
1211 				break;
1212 
1213 			case DKIOCGGEOM:
1214 				if (cl->cl_label_from_media == CMLB_LABEL_EFI) {
1215 					/* GPT label on disk */
1216 					mutex_exit(CMLB_MUTEX(cl));
1217 					return (ENOTSUP);
1218 				}
1219 				break;
1220 			default:
1221 				break;
1222 			}
1223 	}
1224 
1225 	mutex_exit(CMLB_MUTEX(cl));
1226 
1227 	switch (cmd) {
1228 	case DKIOCGGEOM:
1229 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCGGEOM\n");
1230 		err = cmlb_dkio_get_geometry(cl, (caddr_t)arg, flag, tg_cookie);
1231 		break;
1232 
1233 	case DKIOCSGEOM:
1234 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCSGEOM\n");
1235 		err = cmlb_dkio_set_geometry(cl, (caddr_t)arg, flag);
1236 		break;
1237 
1238 	case DKIOCGAPART:
1239 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCGAPART\n");
1240 		err = cmlb_dkio_get_partition(cl, (caddr_t)arg,
1241 		    flag, tg_cookie);
1242 		break;
1243 
1244 	case DKIOCSAPART:
1245 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCSAPART\n");
1246 		err = cmlb_dkio_set_partition(cl, (caddr_t)arg, flag);
1247 		break;
1248 
1249 	case DKIOCGVTOC:
1250 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCGVTOC\n");
1251 		err = cmlb_dkio_get_vtoc(cl, (caddr_t)arg, flag, tg_cookie);
1252 		break;
1253 
1254 	case DKIOCGEXTVTOC:
1255 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCGVTOC\n");
1256 		err = cmlb_dkio_get_extvtoc(cl, (caddr_t)arg, flag, tg_cookie);
1257 		break;
1258 
1259 	case DKIOCGETEFI:
1260 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCGETEFI\n");
1261 		err = cmlb_dkio_get_efi(cl, (caddr_t)arg, flag, tg_cookie);
1262 		break;
1263 
1264 	case DKIOCPARTITION:
1265 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCPARTITION\n");
1266 		err = cmlb_dkio_partition(cl, (caddr_t)arg, flag, tg_cookie);
1267 		break;
1268 
1269 	case DKIOCSVTOC:
1270 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCSVTOC\n");
1271 		err = cmlb_dkio_set_vtoc(cl, dev, (caddr_t)arg, flag,
1272 		    tg_cookie);
1273 		break;
1274 
1275 	case DKIOCSEXTVTOC:
1276 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCSVTOC\n");
1277 		err = cmlb_dkio_set_extvtoc(cl, dev, (caddr_t)arg, flag,
1278 		    tg_cookie);
1279 		break;
1280 
1281 	case DKIOCSETEFI:
1282 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCSETEFI\n");
1283 		err = cmlb_dkio_set_efi(cl, dev, (caddr_t)arg, flag, tg_cookie);
1284 		break;
1285 
1286 	case DKIOCGMBOOT:
1287 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCGMBOOT\n");
1288 		err = cmlb_dkio_get_mboot(cl, (caddr_t)arg, flag, tg_cookie);
1289 		break;
1290 
1291 	case DKIOCSMBOOT:
1292 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCSMBOOT\n");
1293 		err = cmlb_dkio_set_mboot(cl, (caddr_t)arg, flag, tg_cookie);
1294 		break;
1295 	case DKIOCG_PHYGEOM:
1296 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCG_PHYGEOM\n");
1297 #if defined(__i386) || defined(__amd64)
1298 		err = cmlb_dkio_get_phygeom(cl, (caddr_t)arg, flag, tg_cookie);
1299 #else
1300 		err = ENOTTY;
1301 #endif
1302 		break;
1303 	case DKIOCG_VIRTGEOM:
1304 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCG_VIRTGEOM\n");
1305 #if defined(__i386) || defined(__amd64)
1306 		err = cmlb_dkio_get_virtgeom(cl, (caddr_t)arg, flag);
1307 #else
1308 		err = ENOTTY;
1309 #endif
1310 		break;
1311 	case DKIOCPARTINFO:
1312 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCPARTINFO");
1313 #if defined(__i386) || defined(__amd64)
1314 		err = cmlb_dkio_partinfo(cl, dev, (caddr_t)arg, flag);
1315 #else
1316 		err = ENOTTY;
1317 #endif
1318 		break;
1319 	case DKIOCEXTPARTINFO:
1320 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCPARTINFO");
1321 #if defined(__i386) || defined(__amd64)
1322 		err = cmlb_dkio_extpartinfo(cl, dev, (caddr_t)arg, flag);
1323 #else
1324 		err = ENOTTY;
1325 #endif
1326 		break;
1327 #if defined(__i386) || defined(__amd64)
1328 	case DKIOCSETEXTPART:
1329 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCSETEXTPART");
1330 		err = cmlb_dkio_set_ext_part(cl, (caddr_t)arg, flag, tg_cookie);
1331 		break;
1332 #endif
1333 	default:
1334 		err = ENOTTY;
1335 
1336 	}
1337 
1338 	/*
1339 	 * An ioctl that succeeds and changed ('set') size(9P) information
1340 	 * needs to invalidate the cached devinfo snapshot to avoid having
1341 	 * old information being returned in a snapshots.
1342 	 *
1343 	 * NB: When available, call ddi_change_minor_node() to clear
1344 	 * SSIZEVALID in specfs vnodes via spec_size_invalidate().
1345 	 */
1346 	if (err == 0) {
1347 		switch (cmd) {
1348 		case DKIOCSGEOM:
1349 		case DKIOCSAPART:
1350 		case DKIOCSVTOC:
1351 		case DKIOCSEXTVTOC:
1352 		case DKIOCSETEFI:
1353 			i_ddi_prop_dyn_cache_invalidate(CMLB_DEVINFO(cl),
1354 			    i_ddi_prop_dyn_driver_get(CMLB_DEVINFO(cl)));
1355 		}
1356 	}
1357 	return (err);
1358 }
1359 
1360 dev_t
1361 cmlb_make_device(struct cmlb_lun *cl)
1362 {
1363 	if (cl->cl_alter_behavior & CMLB_CREATE_P0_MINOR_NODE) {
1364 		return (makedevice(ddi_driver_major(CMLB_DEVINFO(cl)),
1365 		    ddi_get_instance(
1366 		    CMLB_DEVINFO(cl)) << CMLBUNIT_FORCE_P0_SHIFT));
1367 	} else {
1368 		return (makedevice(ddi_driver_major(CMLB_DEVINFO(cl)),
1369 		    ddi_get_instance(CMLB_DEVINFO(cl)) << CMLBUNIT_SHIFT));
1370 	}
1371 }
1372 
1373 /*
1374  * Function: cmlb_check_update_blockcount
1375  *
1376  * Description: If current capacity value is invalid, obtains the
1377  *		current capacity from target driver.
1378  *
1379  * Return Code: 0	success
1380  *		EIO	failure
1381  */
1382 static int
1383 cmlb_check_update_blockcount(struct cmlb_lun *cl, void *tg_cookie)
1384 {
1385 	int status;
1386 	diskaddr_t capacity;
1387 	uint32_t lbasize;
1388 
1389 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
1390 
1391 	if (cl->cl_f_geometry_is_valid)
1392 		return (0);
1393 
1394 	mutex_exit(CMLB_MUTEX(cl));
1395 	status = DK_TG_GETCAP(cl, &capacity, tg_cookie);
1396 	if (status != 0) {
1397 		mutex_enter(CMLB_MUTEX(cl));
1398 		return (EIO);
1399 	}
1400 
1401 	status = DK_TG_GETBLOCKSIZE(cl, &lbasize, tg_cookie);
1402 	mutex_enter(CMLB_MUTEX(cl));
1403 	if (status != 0)
1404 		return (EIO);
1405 
1406 	if ((capacity != 0) && (lbasize != 0)) {
1407 		cl->cl_blockcount = capacity;
1408 		cl->cl_tgt_blocksize = lbasize;
1409 		if (!cl->cl_is_removable) {
1410 			cl->cl_sys_blocksize = lbasize;
1411 		}
1412 		return (0);
1413 	} else {
1414 		return (EIO);
1415 	}
1416 }
1417 
1418 static int
1419 cmlb_create_minor(dev_info_t *dip, char *name, int spec_type,
1420     minor_t minor_num, char *node_type, int flag, boolean_t internal)
1421 {
1422 	ASSERT(VALID_BOOLEAN(internal));
1423 
1424 	if (internal)
1425 		return (ddi_create_internal_pathname(dip,
1426 		    name, spec_type, minor_num));
1427 	else
1428 		return (ddi_create_minor_node(dip,
1429 		    name, spec_type, minor_num, node_type, flag));
1430 }
1431 
1432 /*
1433  *    Function: cmlb_create_minor_nodes
1434  *
1435  * Description: Create or adjust the minor device nodes for the instance.
1436  * 		Minor nodes are created based on default label type,
1437  *		current label type and last label type we created
1438  *		minor nodes based on.
1439  *
1440  *
1441  *   Arguments: cl - driver soft state (unit) structure
1442  *
1443  * Return Code: 0 success
1444  *		ENXIO	failure.
1445  *
1446  *     Context: Kernel thread context
1447  */
1448 static int
1449 cmlb_create_minor_nodes(struct cmlb_lun *cl)
1450 {
1451 	struct driver_minor_data	*dmdp;
1452 	int				instance, shift;
1453 	char				name[48];
1454 	cmlb_label_t			newlabeltype;
1455 	boolean_t			internal;
1456 
1457 	ASSERT(cl != NULL);
1458 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
1459 
1460 	internal = VOID2BOOLEAN(
1461 	    (cl->cl_alter_behavior & (CMLB_INTERNAL_MINOR_NODES)) != 0);
1462 
1463 	if (cl->cl_alter_behavior & CMLB_CREATE_P0_MINOR_NODE)
1464 		shift = CMLBUNIT_FORCE_P0_SHIFT;
1465 	else
1466 		shift = CMLBUNIT_SHIFT;
1467 
1468 	/* check the most common case */
1469 	if (cl->cl_cur_labeltype != CMLB_LABEL_UNDEF &&
1470 	    cl->cl_last_labeltype == cl->cl_cur_labeltype) {
1471 		/* do nothing */
1472 		return (0);
1473 	}
1474 
1475 	if (cl->cl_def_labeltype == CMLB_LABEL_UNDEF) {
1476 		/* we should never get here */
1477 		return (ENXIO);
1478 	}
1479 
1480 	if (cl->cl_last_labeltype == CMLB_LABEL_UNDEF) {
1481 		/* first time during attach */
1482 		newlabeltype = cl->cl_def_labeltype;
1483 
1484 		instance = ddi_get_instance(CMLB_DEVINFO(cl));
1485 
1486 		/* Create all the minor nodes for this target. */
1487 		dmdp = (newlabeltype == CMLB_LABEL_EFI) ? dk_minor_data_efi :
1488 		    dk_minor_data;
1489 		while (dmdp->name != NULL) {
1490 
1491 			(void) sprintf(name, "%s", dmdp->name);
1492 
1493 			if (cmlb_create_minor(CMLB_DEVINFO(cl), name,
1494 			    dmdp->type,
1495 			    (instance << shift) | dmdp->minor,
1496 			    cl->cl_node_type, NULL, internal) == DDI_FAILURE) {
1497 				/*
1498 				 * Clean up any nodes that may have been
1499 				 * created, in case this fails in the middle
1500 				 * of the loop.
1501 				 */
1502 				ddi_remove_minor_node(CMLB_DEVINFO(cl), NULL);
1503 				return (ENXIO);
1504 			}
1505 			dmdp++;
1506 		}
1507 		cl->cl_last_labeltype = newlabeltype;
1508 #if defined(_SUNOS_VTOC_8)
1509 		/*
1510 		 * "emulate" p0 device for sparc, used by lofi
1511 		 */
1512 		if (cl->cl_alter_behavior & CMLB_CREATE_P0_MINOR_NODE) {
1513 			if (cmlb_create_minor(CMLB_DEVINFO(cl), "q", S_IFBLK,
1514 			    (instance << CMLBUNIT_FORCE_P0_SHIFT) | P0_RAW_DISK,
1515 			    cl->cl_node_type, NULL, internal) == DDI_FAILURE) {
1516 				ddi_remove_minor_node(CMLB_DEVINFO(cl), NULL);
1517 				return (ENXIO);
1518 			}
1519 
1520 			if (cmlb_create_minor(CMLB_DEVINFO(cl), "q,raw",
1521 			    S_IFCHR,
1522 			    (instance << CMLBUNIT_FORCE_P0_SHIFT) | P0_RAW_DISK,
1523 			    cl->cl_node_type, NULL, internal) == DDI_FAILURE) {
1524 				ddi_remove_minor_node(CMLB_DEVINFO(cl), NULL);
1525 				return (ENXIO);
1526 			}
1527 		}
1528 #endif	/* defined(_SUNOS_VTOC_8) */
1529 		return (0);
1530 	}
1531 
1532 	/* Not first time  */
1533 	if (cl->cl_cur_labeltype == CMLB_LABEL_UNDEF) {
1534 		if (cl->cl_last_labeltype != cl->cl_def_labeltype) {
1535 			/* close time, revert to default. */
1536 			newlabeltype = cl->cl_def_labeltype;
1537 		} else {
1538 			/*
1539 			 * do nothing since the type for which we last created
1540 			 * nodes matches the default
1541 			 */
1542 			return (0);
1543 		}
1544 	} else {
1545 		if (cl->cl_cur_labeltype != cl->cl_last_labeltype) {
1546 			/* We are not closing, use current label type */
1547 			newlabeltype = cl->cl_cur_labeltype;
1548 		} else {
1549 			/*
1550 			 * do nothing since the type for which we last created
1551 			 * nodes matches the current label type
1552 			 */
1553 			return (0);
1554 		}
1555 	}
1556 
1557 	instance = ddi_get_instance(CMLB_DEVINFO(cl));
1558 
1559 	/*
1560 	 * Currently we only fix up the s7 node when we are switching
1561 	 * label types from or to EFI. This is consistent with
1562 	 * current behavior of sd.
1563 	 */
1564 	if (newlabeltype == CMLB_LABEL_EFI &&
1565 	    cl->cl_last_labeltype != CMLB_LABEL_EFI) {
1566 		/* from vtoc to EFI */
1567 		ddi_remove_minor_node(CMLB_DEVINFO(cl), "h");
1568 		ddi_remove_minor_node(CMLB_DEVINFO(cl), "h,raw");
1569 		(void) cmlb_create_minor(CMLB_DEVINFO(cl), "wd",
1570 		    S_IFBLK, (instance << shift) | WD_NODE,
1571 		    cl->cl_node_type, NULL, internal);
1572 		(void) cmlb_create_minor(CMLB_DEVINFO(cl), "wd,raw",
1573 		    S_IFCHR, (instance << shift) | WD_NODE,
1574 		    cl->cl_node_type, NULL, internal);
1575 	} else {
1576 		/* from efi to vtoc */
1577 		ddi_remove_minor_node(CMLB_DEVINFO(cl), "wd");
1578 		ddi_remove_minor_node(CMLB_DEVINFO(cl), "wd,raw");
1579 		(void) cmlb_create_minor(CMLB_DEVINFO(cl), "h",
1580 		    S_IFBLK, (instance << shift) | WD_NODE,
1581 		    cl->cl_node_type, NULL, internal);
1582 		(void) cmlb_create_minor(CMLB_DEVINFO(cl), "h,raw",
1583 		    S_IFCHR, (instance << shift) | WD_NODE,
1584 		    cl->cl_node_type, NULL, internal);
1585 	}
1586 
1587 	cl->cl_last_labeltype = newlabeltype;
1588 	return (0);
1589 }
1590 
1591 /*
1592  *    Function: cmlb_validate_geometry
1593  *
1594  * Description: Read the label from the disk (if present). Update the unit's
1595  *		geometry and vtoc information from the data in the label.
1596  *		Verify that the label is valid.
1597  *
1598  *   Arguments:
1599  *	cl		driver soft state (unit) structure
1600  *
1601  *	forcerevalid	force revalidation even if we are already valid.
1602  *	flags		operation flags from target driver. Used for verbosity
1603  *			control	at this time.
1604  *	tg_cookie	cookie from target driver to be passed back to target
1605  *			driver when we call back to it through tg_ops.
1606  *
1607  * Return Code: 0 - Successful completion
1608  *		EINVAL  - Invalid value in cl->cl_tgt_blocksize or
1609  *			  cl->cl_blockcount; or label on disk is corrupted
1610  *			  or unreadable.
1611  *		EACCES  - Reservation conflict at the device.
1612  *		ENOMEM  - Resource allocation error
1613  *		ENOTSUP - geometry not applicable
1614  *
1615  *     Context: Kernel thread only (can sleep).
1616  */
1617 static int
1618 cmlb_validate_geometry(struct cmlb_lun *cl, boolean_t forcerevalid, int flags,
1619     void *tg_cookie)
1620 {
1621 	int		label_error = 0;
1622 	diskaddr_t	capacity;
1623 	int		count;
1624 
1625 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
1626 	ASSERT(VALID_BOOLEAN(forcerevalid));
1627 
1628 	if ((cl->cl_f_geometry_is_valid) && (!forcerevalid)) {
1629 		if (cl->cl_cur_labeltype == CMLB_LABEL_EFI)
1630 			return (ENOTSUP);
1631 		return (0);
1632 	}
1633 
1634 	if (cmlb_check_update_blockcount(cl, tg_cookie) != 0)
1635 		return (EIO);
1636 
1637 	capacity = cl->cl_blockcount;
1638 
1639 	/*
1640 	 * Set up the "whole disk" fdisk partition; this should always
1641 	 * exist, regardless of whether the disk contains an fdisk table
1642 	 * or vtoc.
1643 	 */
1644 	cl->cl_map[P0_RAW_DISK].dkl_cylno = 0;
1645 	cl->cl_offset[P0_RAW_DISK] = 0;
1646 	/*
1647 	 * note if capacity > int32_max(1TB) we are in 64bit environment
1648 	 * so no truncation happens
1649 	 */
1650 	cl->cl_map[P0_RAW_DISK].dkl_nblk  = capacity;
1651 
1652 	/*
1653 	 * Refresh the logical and physical geometry caches.
1654 	 * (data from MODE SENSE format/rigid disk geometry pages,
1655 	 * and scsi_ifgetcap("geometry").
1656 	 */
1657 	cmlb_resync_geom_caches(cl, capacity, tg_cookie);
1658 
1659 	cl->cl_label_from_media = CMLB_LABEL_UNDEF;
1660 	label_error = cmlb_use_efi(cl, capacity, flags, tg_cookie);
1661 	if (label_error == 0) {
1662 
1663 		/* found a valid EFI label */
1664 		cmlb_dbg(CMLB_TRACE, cl,
1665 		    "cmlb_validate_geometry: found EFI label\n");
1666 		/*
1667 		 * solaris_size and geometry_is_valid are set in
1668 		 * cmlb_use_efi
1669 		 */
1670 		return (ENOTSUP);
1671 	}
1672 
1673 	/* NO EFI label found */
1674 
1675 	if (capacity > CMLB_EXTVTOC_LIMIT) {
1676 		if (label_error == ESRCH) {
1677 			/*
1678 			 * they've configured a LUN over 2TB, but used
1679 			 * format.dat to restrict format's view of the
1680 			 * capacity to be under 2TB in some earlier Solaris
1681 			 * release.
1682 			 */
1683 			/* i.e > 2TB with a VTOC < 2TB */
1684 			if (!(flags & CMLB_SILENT) &&
1685 			    (cl->cl_msglog_flag & CMLB_ALLOW_2TB_WARN)) {
1686 
1687 				cmlb_log(CMLB_DEVINFO(cl), CMLB_LABEL(cl),
1688 				    CE_NOTE, "!Disk (%s%d) is limited to 2 TB "
1689 				    "due to VTOC label. To use the full "
1690 				    "capacity of the disk, use format(1M) to "
1691 				    "relabel the disk with EFI/GPT label.\n",
1692 				    CMLB_LABEL(cl),
1693 				    ddi_get_instance(CMLB_DEVINFO(cl)));
1694 
1695 				cl->cl_msglog_flag &= ~CMLB_ALLOW_2TB_WARN;
1696 			}
1697 		} else {
1698 				return (ENOTSUP);
1699 		}
1700 	}
1701 
1702 	label_error = 0;
1703 
1704 	/*
1705 	 * at this point it is either labeled with a VTOC or it is
1706 	 * under 1TB (<= 1TB actually for off-by-1)
1707 	 */
1708 
1709 	/*
1710 	 * Only DIRECT ACCESS devices will have Scl labels.
1711 	 * CD's supposedly have a Scl label, too
1712 	 */
1713 	if (cl->cl_device_type == DTYPE_DIRECT || ISREMOVABLE(cl)) {
1714 		struct	dk_label *dkl;
1715 		offset_t label_addr;
1716 		int	rval;
1717 		size_t	buffer_size;
1718 
1719 		/*
1720 		 * Note: This will set up cl->cl_solaris_size and
1721 		 * cl->cl_solaris_offset.
1722 		 */
1723 		rval = cmlb_read_fdisk(cl, capacity, tg_cookie);
1724 		if ((rval != 0) && !ISCD(cl)) {
1725 			ASSERT(mutex_owned(CMLB_MUTEX(cl)));
1726 			return (rval);
1727 		}
1728 
1729 		if (cl->cl_solaris_size <= DK_LABEL_LOC) {
1730 			/*
1731 			 * Found fdisk table but no Solaris partition entry,
1732 			 * so don't call cmlb_uselabel() and don't create
1733 			 * a default label.
1734 			 */
1735 			label_error = 0;
1736 			cl->cl_f_geometry_is_valid = B_TRUE;
1737 			goto no_solaris_partition;
1738 		}
1739 
1740 		label_addr = (daddr_t)(cl->cl_solaris_offset + DK_LABEL_LOC);
1741 
1742 		buffer_size = cl->cl_sys_blocksize;
1743 
1744 		cmlb_dbg(CMLB_TRACE, cl, "cmlb_validate_geometry: "
1745 		    "label_addr: 0x%x allocation size: 0x%x\n",
1746 		    label_addr, buffer_size);
1747 
1748 		if ((dkl = kmem_zalloc(buffer_size, KM_NOSLEEP)) == NULL)
1749 			return (ENOMEM);
1750 
1751 		mutex_exit(CMLB_MUTEX(cl));
1752 		rval = DK_TG_READ(cl, dkl, label_addr, buffer_size, tg_cookie);
1753 		mutex_enter(CMLB_MUTEX(cl));
1754 
1755 		switch (rval) {
1756 		case 0:
1757 			/*
1758 			 * cmlb_uselabel will establish that the geometry
1759 			 * is valid.
1760 			 */
1761 			if (cmlb_uselabel(cl,
1762 			    (struct dk_label *)(uintptr_t)dkl, flags) !=
1763 			    CMLB_LABEL_IS_VALID) {
1764 				label_error = EINVAL;
1765 			} else
1766 				cl->cl_label_from_media = CMLB_LABEL_VTOC;
1767 			break;
1768 		case EACCES:
1769 			label_error = EACCES;
1770 			break;
1771 		default:
1772 			label_error = EINVAL;
1773 			break;
1774 		}
1775 
1776 		kmem_free(dkl, buffer_size);
1777 	}
1778 
1779 	/*
1780 	 * If a valid label was not found, AND if no reservation conflict
1781 	 * was detected, then go ahead and create a default label (4069506).
1782 	 *
1783 	 * Note: currently, for VTOC_8 devices, the default label is created
1784 	 * for removables and hotpluggables only.  For VTOC_16 devices, the
1785 	 * default label will be created for all devices.
1786 	 * (see cmlb_build_default_label)
1787 	 */
1788 #if defined(_SUNOS_VTOC_8)
1789 	if ((ISREMOVABLE(cl) || ISHOTPLUGGABLE(cl)) &&
1790 	    (label_error != EACCES)) {
1791 #elif defined(_SUNOS_VTOC_16)
1792 	if (label_error != EACCES) {
1793 #endif
1794 		if (!cl->cl_f_geometry_is_valid) {
1795 			cmlb_build_default_label(cl, tg_cookie);
1796 		}
1797 		label_error = 0;
1798 	}
1799 
1800 no_solaris_partition:
1801 
1802 #if defined(_SUNOS_VTOC_16)
1803 	/*
1804 	 * If we have valid geometry, set up the remaining fdisk partitions.
1805 	 * Note that dkl_cylno is not used for the fdisk map entries, so
1806 	 * we set it to an entirely bogus value.
1807 	 */
1808 	for (count = 0; count < FDISK_PARTS; count++) {
1809 		cl->cl_map[FDISK_P1 + count].dkl_cylno = UINT16_MAX;
1810 		cl->cl_map[FDISK_P1 + count].dkl_nblk =
1811 		    cl->cl_fmap[count].fmap_nblk;
1812 
1813 		cl->cl_offset[FDISK_P1 + count] =
1814 		    cl->cl_fmap[count].fmap_start;
1815 	}
1816 #endif
1817 
1818 	for (count = 0; count < NDKMAP; count++) {
1819 #if defined(_SUNOS_VTOC_8)
1820 		struct dk_map *lp  = &cl->cl_map[count];
1821 		cl->cl_offset[count] =
1822 		    cl->cl_g.dkg_nhead * cl->cl_g.dkg_nsect * lp->dkl_cylno;
1823 #elif defined(_SUNOS_VTOC_16)
1824 		struct dkl_partition *vp = &cl->cl_vtoc.v_part[count];
1825 
1826 		cl->cl_offset[count] = vp->p_start + cl->cl_solaris_offset;
1827 #else
1828 #error "No VTOC format defined."
1829 #endif
1830 	}
1831 
1832 	return (label_error);
1833 }
1834 
1835 #if defined(_SUNOS_VTOC_16)
1836 /*
1837  *    Function: cmlb_convert_geometry
1838  *
1839  * Description: Convert physical geometry into a dk_geom structure. In
1840  *		other words, make sure we don't wrap 16-bit values.
1841  *		e.g. converting from geom_cache to dk_geom
1842  *
1843  *     Context: Kernel thread only
1844  */
1845 static void
1846 cmlb_convert_geometry(struct cmlb_lun *cl, diskaddr_t capacity,
1847     struct dk_geom *cl_g, void *tg_cookie)
1848 {
1849 
1850 	ASSERT(cl != NULL);
1851 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
1852 
1853 	/* Unlabeled SCSI floppy device */
1854 	if (capacity < 160) {
1855 		/* Less than 80K */
1856 		cl_g->dkg_nhead = 1;
1857 		cl_g->dkg_ncyl = capacity;
1858 		cl_g->dkg_nsect = 1;
1859 		return;
1860 	} else if (capacity <= 0x1000) {
1861 		cl_g->dkg_nhead = 2;
1862 		cl_g->dkg_ncyl = 80;
1863 		cl_g->dkg_nsect = capacity / (cl_g->dkg_nhead * cl_g->dkg_ncyl);
1864 		return;
1865 	}
1866 
1867 	/*
1868 	 * For all devices we calculate cylinders using the heads and sectors
1869 	 * we assign based on capacity of the device.  The algorithm is
1870 	 * designed to be compatible with the way other operating systems
1871 	 * lay out fdisk tables for X86 and to insure that the cylinders never
1872 	 * exceed 65535 to prevent problems with X86 ioctls that report
1873 	 * geometry.
1874 	 * For some smaller disk sizes we report geometry that matches those
1875 	 * used by X86 BIOS usage. For larger disks, we use SPT that are
1876 	 * multiples of 63, since other OSes that are not limited to 16-bits
1877 	 * for cylinders stop at 63 SPT we make do by using multiples of 63 SPT.
1878 	 *
1879 	 * The following table (in order) illustrates some end result
1880 	 * calculations:
1881 	 *
1882 	 * Maximum number of blocks 		nhead	nsect
1883 	 *
1884 	 * 2097152 (1GB)			64	32
1885 	 * 16777216 (8GB)			128	32
1886 	 * 1052819775 (502.02GB)		255  	63
1887 	 * 2105639550 (0.98TB)			255	126
1888 	 * 3158459325 (1.47TB)			255  	189
1889 	 * 4211279100 (1.96TB)			255  	252
1890 	 * 5264098875 (2.45TB)			255  	315
1891 	 * ...
1892 	 *
1893 	 * For Solid State Drive(SSD), it uses 4K page size inside and may be
1894 	 * double with every new generation. If the I/O is not aligned with
1895 	 * page size on SSDs, SSDs perform a lot slower.
1896 	 * By default, Solaris partition starts from cylinder 1. It will be
1897 	 * misaligned even with 4K if using heads(255) and SPT(63). To
1898 	 * workaround the problem, if the device is SSD, we use heads(224) and
1899 	 * SPT multiple of 56. Thus the default Solaris partition starts from
1900 	 * a position that aligns with 128K on a 512 bytes sector size SSD.
1901 	 */
1902 
1903 	if (capacity <= 0x200000) {
1904 		cl_g->dkg_nhead = 64;
1905 		cl_g->dkg_nsect = 32;
1906 	} else if (capacity <= 0x01000000) {
1907 		cl_g->dkg_nhead = 128;
1908 		cl_g->dkg_nsect = 32;
1909 	} else {
1910 		tg_attribute_t tgattribute;
1911 		int is_solid_state;
1912 		unsigned short nhead;
1913 		unsigned short nsect;
1914 
1915 		bzero(&tgattribute, sizeof (tg_attribute_t));
1916 
1917 		mutex_exit(CMLB_MUTEX(cl));
1918 		is_solid_state =
1919 		    (DK_TG_GETATTRIBUTE(cl, &tgattribute, tg_cookie) == 0) ?
1920 		    tgattribute.media_is_solid_state : FALSE;
1921 		mutex_enter(CMLB_MUTEX(cl));
1922 
1923 		if (is_solid_state) {
1924 			nhead = 224;
1925 			nsect = 56;
1926 		} else {
1927 			nhead = 255;
1928 			nsect = 63;
1929 		}
1930 
1931 		cl_g->dkg_nhead = nhead;
1932 
1933 		/* make dkg_nsect be smallest multiple of nsect */
1934 		cl_g->dkg_nsect = ((capacity +
1935 		    (UINT16_MAX * nhead * nsect) - 1) /
1936 		    (UINT16_MAX * nhead * nsect)) * nsect;
1937 
1938 		if (cl_g->dkg_nsect == 0)
1939 			cl_g->dkg_nsect = (UINT16_MAX / nsect) * nsect;
1940 	}
1941 
1942 }
1943 #endif
1944 
1945 /*
1946  *    Function: cmlb_resync_geom_caches
1947  *
1948  * Description: (Re)initialize both geometry caches: the virtual geometry
1949  *            information is extracted from the HBA (the "geometry"
1950  *            capability), and the physical geometry cache data is
1951  *            generated by issuing MODE SENSE commands.
1952  *
1953  *   Arguments:
1954  *	cl 		driver soft state (unit) structure
1955  *	capacity	disk capacity in #blocks
1956  *	tg_cookie	cookie from target driver to be passed back to target
1957  *			driver when we call back to it through tg_ops.
1958  *
1959  *     Context: Kernel thread only (can sleep).
1960  */
1961 static void
1962 cmlb_resync_geom_caches(struct cmlb_lun *cl, diskaddr_t capacity,
1963     void *tg_cookie)
1964 {
1965 	struct	cmlb_geom 	pgeom;
1966 	struct	cmlb_geom	lgeom;
1967 	struct 	cmlb_geom	*pgeomp = &pgeom;
1968 	unsigned short 		nhead;
1969 	unsigned short 		nsect;
1970 	int 			spc;
1971 	int			ret;
1972 
1973 	ASSERT(cl != NULL);
1974 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
1975 
1976 	/*
1977 	 * Ask the controller for its logical geometry.
1978 	 * Note: if the HBA does not support scsi_ifgetcap("geometry"),
1979 	 * then the lgeom cache will be invalid.
1980 	 */
1981 	mutex_exit(CMLB_MUTEX(cl));
1982 	bzero(&lgeom, sizeof (struct cmlb_geom));
1983 	ret = DK_TG_GETVIRTGEOM(cl, &lgeom, tg_cookie);
1984 	mutex_enter(CMLB_MUTEX(cl));
1985 
1986 	bcopy(&lgeom, &cl->cl_lgeom, sizeof (cl->cl_lgeom));
1987 
1988 	/*
1989 	 * Initialize the pgeom cache from lgeom, so that if MODE SENSE
1990 	 * doesn't work, DKIOCG_PHYSGEOM can return reasonable values.
1991 	 */
1992 	if (ret != 0 || cl->cl_lgeom.g_nsect == 0 ||
1993 	    cl->cl_lgeom.g_nhead == 0) {
1994 		/*
1995 		 * Note: Perhaps this needs to be more adaptive? The rationale
1996 		 * is that, if there's no HBA geometry from the HBA driver, any
1997 		 * guess is good, since this is the physical geometry. If MODE
1998 		 * SENSE fails this gives a max cylinder size for non-LBA access
1999 		 */
2000 		nhead = 255;
2001 		nsect = 63;
2002 	} else {
2003 		nhead = cl->cl_lgeom.g_nhead;
2004 		nsect = cl->cl_lgeom.g_nsect;
2005 	}
2006 
2007 	if (ISCD(cl)) {
2008 		pgeomp->g_nhead = 1;
2009 		pgeomp->g_nsect = nsect * nhead;
2010 	} else {
2011 		pgeomp->g_nhead = nhead;
2012 		pgeomp->g_nsect = nsect;
2013 	}
2014 
2015 	spc = pgeomp->g_nhead * pgeomp->g_nsect;
2016 	pgeomp->g_capacity = capacity;
2017 	if (spc == 0)
2018 		pgeomp->g_ncyl = 0;
2019 	else
2020 		pgeomp->g_ncyl = pgeomp->g_capacity / spc;
2021 	pgeomp->g_acyl = 0;
2022 
2023 	/*
2024 	 * Retrieve fresh geometry data from the hardware, stash it
2025 	 * here temporarily before we rebuild the incore label.
2026 	 *
2027 	 * We want to use the MODE SENSE commands to derive the
2028 	 * physical geometry of the device, but if either command
2029 	 * fails, the logical geometry is used as the fallback for
2030 	 * disk label geometry.
2031 	 */
2032 
2033 	mutex_exit(CMLB_MUTEX(cl));
2034 	(void) DK_TG_GETPHYGEOM(cl, pgeomp, tg_cookie);
2035 	mutex_enter(CMLB_MUTEX(cl));
2036 
2037 	/*
2038 	 * Now update the real copy while holding the mutex. This
2039 	 * way the global copy is never in an inconsistent state.
2040 	 */
2041 	bcopy(pgeomp, &cl->cl_pgeom,  sizeof (cl->cl_pgeom));
2042 
2043 	cmlb_dbg(CMLB_INFO, cl, "cmlb_resync_geom_caches: "
2044 	    "(cached from lgeom)\n");
2045 	cmlb_dbg(CMLB_INFO,  cl,
2046 	    "   ncyl: %ld; acyl: %d; nhead: %d; nsect: %d\n",
2047 	    cl->cl_pgeom.g_ncyl, cl->cl_pgeom.g_acyl,
2048 	    cl->cl_pgeom.g_nhead, cl->cl_pgeom.g_nsect);
2049 	cmlb_dbg(CMLB_INFO,  cl, "   lbasize: %d; capacity: %ld; "
2050 	    "intrlv: %d; rpm: %d\n", cl->cl_pgeom.g_secsize,
2051 	    cl->cl_pgeom.g_capacity, cl->cl_pgeom.g_intrlv,
2052 	    cl->cl_pgeom.g_rpm);
2053 }
2054 
2055 
2056 #if defined(__i386) || defined(__amd64)
2057 /*
2058  *    Function: cmlb_update_ext_minor_nodes
2059  *
2060  * Description: Routine to add/remove extended partition device nodes
2061  *
2062  *   Arguments:
2063  *	cl		driver soft state (unit) structure
2064  *	num_parts	Number of logical drives found on the LUN
2065  *
2066  * Should be called with the mutex held
2067  *
2068  * Return Code: 0 for success
2069  *
2070  *     Context: User and Kernel thread
2071  *
2072  */
2073 static int
2074 cmlb_update_ext_minor_nodes(struct cmlb_lun *cl, int num_parts)
2075 {
2076 	int				i, count, shift;
2077 	char				name[48];
2078 	int				instance;
2079 	struct driver_minor_data	*demdp, *demdpr;
2080 	char				*devnm;
2081 	dev_info_t			*pdip;
2082 	boolean_t 			internal;
2083 
2084 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
2085 	ASSERT(cl->cl_update_ext_minor_nodes == 1);
2086 
2087 	internal = VOID2BOOLEAN(
2088 	    (cl->cl_alter_behavior & (CMLB_INTERNAL_MINOR_NODES)) != 0);
2089 	instance = ddi_get_instance(CMLB_DEVINFO(cl));
2090 	demdp = dk_ext_minor_data;
2091 	demdpr = &dk_ext_minor_data[MAX_EXT_PARTS];
2092 
2093 	if (cl->cl_alter_behavior & CMLB_CREATE_P0_MINOR_NODE)
2094 		shift = CMLBUNIT_FORCE_P0_SHIFT;
2095 	else
2096 		shift = CMLBUNIT_SHIFT;
2097 
2098 	if (cl->cl_logical_drive_count) {
2099 		for (i = 0; i < cl->cl_logical_drive_count; i++) {
2100 			(void) sprintf(name, "%s", demdp->name);
2101 			ddi_remove_minor_node(CMLB_DEVINFO(cl), name);
2102 			(void) sprintf(name, "%s", demdpr->name);
2103 			ddi_remove_minor_node(CMLB_DEVINFO(cl), name);
2104 			demdp++;
2105 			demdpr++;
2106 		}
2107 		/* There are existing device nodes. Remove them */
2108 		devnm = kmem_alloc(MAXNAMELEN + 1, KM_SLEEP);
2109 		(void) ddi_deviname(cl->cl_devi, devnm);
2110 		pdip = ddi_get_parent(cl->cl_devi);
2111 		(void) devfs_clean(pdip, devnm + 1, DV_CLEAN_FORCE);
2112 		kmem_free(devnm, MAXNAMELEN + 1);
2113 	}
2114 
2115 	demdp = dk_ext_minor_data;
2116 	demdpr = &dk_ext_minor_data[MAX_EXT_PARTS];
2117 
2118 	for (i = 0; i < num_parts; i++) {
2119 		(void) sprintf(name, "%s", demdp->name);
2120 		if (cmlb_create_minor(CMLB_DEVINFO(cl), name,
2121 		    demdp->type,
2122 		    (instance << shift) | demdp->minor,
2123 		    cl->cl_node_type, NULL, internal) == DDI_FAILURE) {
2124 			/*
2125 			 * Clean up any nodes that may have been
2126 			 * created, in case this fails in the middle
2127 			 * of the loop.
2128 			 */
2129 			ddi_remove_minor_node(CMLB_DEVINFO(cl), NULL);
2130 			cl->cl_logical_drive_count = 0;
2131 			return (ENXIO);
2132 		}
2133 		(void) sprintf(name, "%s", demdpr->name);
2134 		if (ddi_create_minor_node(CMLB_DEVINFO(cl), name,
2135 		    demdpr->type,
2136 		    (instance << shift) | demdpr->minor,
2137 		    cl->cl_node_type, NULL) == DDI_FAILURE) {
2138 			/*
2139 			 * Clean up any nodes that may have been
2140 			 * created, in case this fails in the middle
2141 			 * of the loop.
2142 			 */
2143 			ddi_remove_minor_node(CMLB_DEVINFO(cl), NULL);
2144 			cl->cl_logical_drive_count = 0;
2145 			return (ENXIO);
2146 		}
2147 		demdp++;
2148 		demdpr++;
2149 	}
2150 
2151 	/* Update the cl_map array for logical drives */
2152 	for (count = 0; count < MAX_EXT_PARTS; count++) {
2153 		cl->cl_map[FDISK_P4 + 1 + count].dkl_cylno = UINT32_MAX;
2154 		cl->cl_map[FDISK_P4 + 1 + count].dkl_nblk =
2155 		    cl->cl_fmap[FD_NUMPART + count].fmap_nblk;
2156 		cl->cl_offset[FDISK_P4 + 1 + count] =
2157 		    cl->cl_fmap[FD_NUMPART + count].fmap_start;
2158 	}
2159 
2160 	cl->cl_logical_drive_count = i;
2161 	cl->cl_update_ext_minor_nodes = 0;
2162 	return (0);
2163 }
2164 /*
2165  *    Function: cmlb_validate_ext_part
2166  *
2167  * Description: utility routine to validate an extended partition's
2168  *		metadata as found on disk
2169  *
2170  *   Arguments:
2171  *	cl		driver soft state (unit) structure
2172  *	part		partition number of the extended partition
2173  *	epart		partition number of the logical drive
2174  *	start		absolute sector number of the start of the logical
2175  *			drive being validated
2176  *	size		size of logical drive being validated
2177  *
2178  * Return Code: 0 for success
2179  *
2180  *     Context: User and Kernel thread
2181  *
2182  * Algorithm :
2183  * Error cases are :
2184  *	1. If start block is lesser than or equal to the end block
2185  *	2. If either start block or end block is beyond the bounadry
2186  *	   of the extended partition.
2187  *	3. start or end block overlap with existing partitions.
2188  *		To check this, first make sure that the start block doesnt
2189  *		overlap with existing partitions. Then, calculate the
2190  *		possible end block for the given start block that doesnt
2191  *		overlap with existing partitions. This can be calculated by
2192  *		first setting the possible end block to the end of the
2193  *		extended partition (optimistic) and then, checking if there
2194  *		is any other partition that lies after the start of the
2195  *		partition being validated. If so, set the possible end to
2196  *		one block less than the beginning of the next nearest partition
2197  *		If the actual end block is greater than the calculated end
2198  *		block, we have an overlap.
2199  *
2200  */
2201 static int
2202 cmlb_validate_ext_part(struct cmlb_lun *cl, int part, int epart, uint32_t start,
2203     uint32_t size)
2204 {
2205 	int i;
2206 	uint32_t end = start + size - 1;
2207 	uint32_t ext_start = cl->cl_fmap[part].fmap_start;
2208 	uint32_t ext_end = ext_start + cl->cl_fmap[part].fmap_nblk - 1;
2209 	uint32_t ts, te;
2210 	uint32_t poss_end = ext_end;
2211 
2212 	if (end <= start) {
2213 		return (1);
2214 	}
2215 
2216 	/*
2217 	 * Check if the logical drive boundaries are within that of the
2218 	 * extended partition.
2219 	 */
2220 	if (start <= ext_start || start > ext_end || end <= ext_start ||
2221 	    end > ext_end) {
2222 		return (1);
2223 	}
2224 
2225 	/*
2226 	 * epart will be equal to FD_NUMPART if it is the first logical drive.
2227 	 * There is no need to check for overlaps with other logical drives,
2228 	 * since it is the only logical drive that we have come across so far.
2229 	 */
2230 	if (epart == FD_NUMPART) {
2231 		return (0);
2232 	}
2233 
2234 	/* Check for overlaps with existing logical drives */
2235 	i = FD_NUMPART;
2236 	ts = cl->cl_fmap[FD_NUMPART].fmap_start;
2237 	te = ts + cl->cl_fmap[FD_NUMPART].fmap_nblk - 1;
2238 
2239 	while ((i < epart) && ts && te) {
2240 		if (start >= ts && start <= te) {
2241 			return (1);
2242 		}
2243 
2244 		if ((ts < poss_end) && (ts > start)) {
2245 			poss_end = ts - 1;
2246 		}
2247 
2248 		i++;
2249 		ts = cl->cl_fmap[i].fmap_start;
2250 		te = ts + cl->cl_fmap[i].fmap_nblk - 1;
2251 	}
2252 
2253 	if (end > poss_end) {
2254 		return (1);
2255 	}
2256 
2257 	return (0);
2258 }
2259 
2260 
2261 /*
2262  *    Function: cmlb_is_linux_swap
2263  *
2264  * Description: utility routine to verify if a partition is a linux swap
2265  *		partition or not.
2266  *
2267  *   Arguments:
2268  *	cl		driver soft state (unit) structure
2269  *	part_start	absolute sector number of the start of the partition
2270  *			being verified
2271  *	tg_cookie	cookie from target driver to be passed back to target
2272  *			driver when we call back to it through tg_ops.
2273  *
2274  * Return Code: 0 for success
2275  *
2276  *     Context: User and Kernel thread
2277  *
2278  * Notes:
2279  *	The linux swap magic "SWAP-SPACE" or "SWAPSPACE2" is found as the
2280  *	last 10 bytes of a disk block whose size is that of the linux page
2281  *	size. This disk block is found at the beginning of the swap partition.
2282  */
2283 static int
2284 cmlb_is_linux_swap(struct cmlb_lun *cl, uint32_t part_start, void *tg_cookie)
2285 {
2286 	int		i;
2287 	int		rval = -1;
2288 	uint32_t	seek_offset;
2289 	uint32_t	linux_pg_size;
2290 	char 		*buf, *linux_swap_magic;
2291 	int		sec_sz = cl->cl_sys_blocksize;
2292 	/* Known linux kernel page sizes */
2293 	uint32_t	linux_pg_size_arr[] = {4096, };
2294 
2295 	ASSERT(cl != NULL);
2296 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
2297 
2298 	if ((buf = kmem_zalloc(sec_sz, KM_NOSLEEP)) == NULL) {
2299 		return (ENOMEM);
2300 	}
2301 
2302 	/*
2303 	 * Check if there is a sane Solaris VTOC
2304 	 * If there is a valid vtoc, no need to lookup
2305 	 * for the linux swap signature.
2306 	 */
2307 	mutex_exit(CMLB_MUTEX(cl));
2308 	rval = DK_TG_READ(cl, buf, part_start + DK_LABEL_LOC,
2309 	    sec_sz, tg_cookie);
2310 	mutex_enter(CMLB_MUTEX(cl));
2311 	if (rval != 0) {
2312 		cmlb_dbg(CMLB_ERROR,  cl,
2313 		    "cmlb_is_linux_swap: disk vtoc read err\n");
2314 		rval = EIO;
2315 		goto done;
2316 	}
2317 
2318 	if ((((struct dk_label *)buf)->dkl_magic == DKL_MAGIC) &&
2319 	    (((struct dk_label *)buf)->dkl_vtoc.v_sanity == VTOC_SANE)) {
2320 		rval = -1;
2321 		goto done;
2322 	}
2323 
2324 
2325 	/* No valid vtoc, so check for linux swap signature */
2326 	linux_swap_magic = buf + sec_sz - 10;
2327 
2328 	for (i = 0; i < sizeof (linux_pg_size_arr)/sizeof (uint32_t); i++) {
2329 		linux_pg_size = linux_pg_size_arr[i];
2330 		seek_offset = linux_pg_size/sec_sz - 1;
2331 		seek_offset += part_start;
2332 
2333 		mutex_exit(CMLB_MUTEX(cl));
2334 		rval = DK_TG_READ(cl, buf, seek_offset, sec_sz, tg_cookie);
2335 		mutex_enter(CMLB_MUTEX(cl));
2336 
2337 		if (rval != 0) {
2338 			cmlb_dbg(CMLB_ERROR,  cl,
2339 			    "cmlb_is_linux_swap: disk read err\n");
2340 			rval = EIO;
2341 			break;
2342 		}
2343 
2344 		rval = -1;
2345 
2346 		if ((strncmp(linux_swap_magic, "SWAP-SPACE", 10) == 0) ||
2347 		    (strncmp(linux_swap_magic, "SWAPSPACE2", 10) == 0)) {
2348 			/* Found a linux swap */
2349 			rval = 0;
2350 			break;
2351 		}
2352 	}
2353 
2354 done:
2355 	kmem_free(buf, sec_sz);
2356 	return (rval);
2357 }
2358 #endif
2359 
2360 /*
2361  *    Function: cmlb_read_fdisk
2362  *
2363  * Description: utility routine to read the fdisk table.
2364  *
2365  *   Arguments:
2366  *	cl		driver soft state (unit) structure
2367  *	capacity	disk capacity in #blocks
2368  *	tg_cookie	cookie from target driver to be passed back to target
2369  *			driver when we call back to it through tg_ops.
2370  *
2371  * Return Code: 0 for success (includes not reading for no_fdisk_present case
2372  *		errnos from tg_rw if failed to read the first block.
2373  *
2374  *     Context: Kernel thread only (can sleep).
2375  */
2376 /*ARGSUSED*/
2377 static int
2378 cmlb_read_fdisk(struct cmlb_lun *cl, diskaddr_t capacity, void *tg_cookie)
2379 {
2380 #if defined(_NO_FDISK_PRESENT)
2381 
2382 	cl->cl_solaris_offset = 0;
2383 	cl->cl_solaris_size = capacity;
2384 	bzero(cl->cl_fmap, sizeof (struct fmap) * FD_NUMPART);
2385 	return (0);
2386 
2387 #elif defined(_FIRMWARE_NEEDS_FDISK)
2388 
2389 	struct ipart	*fdp;
2390 	struct mboot	*mbp;
2391 	struct ipart	fdisk[FD_NUMPART];
2392 	int		i, k;
2393 	char		sigbuf[2];
2394 	caddr_t		bufp;
2395 	int		uidx;
2396 	int 		rval;
2397 	int		lba = 0;
2398 	uint_t		solaris_offset;	/* offset to solaris part. */
2399 	daddr_t		solaris_size;	/* size of solaris partition */
2400 	uint32_t	blocksize;
2401 #if defined(__i386) || defined(__amd64)
2402 	struct ipart	eparts[2];
2403 	struct ipart	*efdp1 = &eparts[0];
2404 	struct ipart	*efdp2 = &eparts[1];
2405 	int		ext_part_exists = 0;
2406 	int		ld_count = 0;
2407 #endif
2408 
2409 	ASSERT(cl != NULL);
2410 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
2411 
2412 	/*
2413 	 * Start off assuming no fdisk table
2414 	 */
2415 	solaris_offset = 0;
2416 	solaris_size   = capacity;
2417 
2418 	blocksize = cl->cl_tgt_blocksize;
2419 
2420 	bufp = kmem_zalloc(blocksize, KM_SLEEP);
2421 
2422 	mutex_exit(CMLB_MUTEX(cl));
2423 	rval = DK_TG_READ(cl, bufp, 0, blocksize, tg_cookie);
2424 	mutex_enter(CMLB_MUTEX(cl));
2425 
2426 	if (rval != 0) {
2427 		cmlb_dbg(CMLB_ERROR,  cl,
2428 		    "cmlb_read_fdisk: fdisk read err\n");
2429 		bzero(cl->cl_fmap, sizeof (struct fmap) * FD_NUMPART);
2430 		goto done;
2431 	}
2432 
2433 	mbp = (struct mboot *)bufp;
2434 
2435 	/*
2436 	 * The fdisk table does not begin on a 4-byte boundary within the
2437 	 * master boot record, so we copy it to an aligned structure to avoid
2438 	 * alignment exceptions on some processors.
2439 	 */
2440 	bcopy(&mbp->parts[0], fdisk, sizeof (fdisk));
2441 
2442 	/*
2443 	 * Check for lba support before verifying sig; sig might not be
2444 	 * there, say on a blank disk, but the max_chs mark may still
2445 	 * be present.
2446 	 *
2447 	 * Note: LBA support and BEFs are an x86-only concept but this
2448 	 * code should work OK on SPARC as well.
2449 	 */
2450 
2451 	/*
2452 	 * First, check for lba-access-ok on root node (or prom root node)
2453 	 * if present there, don't need to search fdisk table.
2454 	 */
2455 	if (ddi_getprop(DDI_DEV_T_ANY, ddi_root_node(), 0,
2456 	    "lba-access-ok", 0) != 0) {
2457 		/* All drives do LBA; don't search fdisk table */
2458 		lba = 1;
2459 	} else {
2460 		/* Okay, look for mark in fdisk table */
2461 		for (fdp = fdisk, i = 0; i < FD_NUMPART; i++, fdp++) {
2462 			/* accumulate "lba" value from all partitions */
2463 			lba = (lba || cmlb_has_max_chs_vals(fdp));
2464 		}
2465 	}
2466 
2467 	if (lba != 0) {
2468 		dev_t dev = cmlb_make_device(cl);
2469 
2470 		if (ddi_getprop(dev, CMLB_DEVINFO(cl), DDI_PROP_DONTPASS,
2471 		    "lba-access-ok", 0) == 0) {
2472 			/* not found; create it */
2473 			if (ddi_prop_create(dev, CMLB_DEVINFO(cl), 0,
2474 			    "lba-access-ok", (caddr_t)NULL, 0) !=
2475 			    DDI_PROP_SUCCESS) {
2476 				cmlb_dbg(CMLB_ERROR,  cl,
2477 				    "cmlb_read_fdisk: Can't create lba "
2478 				    "property for instance %d\n",
2479 				    ddi_get_instance(CMLB_DEVINFO(cl)));
2480 			}
2481 		}
2482 	}
2483 
2484 	bcopy(&mbp->signature, sigbuf, sizeof (sigbuf));
2485 
2486 	/*
2487 	 * Endian-independent signature check
2488 	 */
2489 	if (((sigbuf[1] & 0xFF) != ((MBB_MAGIC >> 8) & 0xFF)) ||
2490 	    (sigbuf[0] != (MBB_MAGIC & 0xFF))) {
2491 		cmlb_dbg(CMLB_ERROR,  cl,
2492 		    "cmlb_read_fdisk: no fdisk\n");
2493 		bzero(cl->cl_fmap, sizeof (struct fmap) * FD_NUMPART);
2494 		goto done;
2495 	}
2496 
2497 #ifdef CMLBDEBUG
2498 	if (cmlb_level_mask & CMLB_LOGMASK_INFO) {
2499 		fdp = fdisk;
2500 		cmlb_dbg(CMLB_INFO,  cl, "cmlb_read_fdisk:\n");
2501 		cmlb_dbg(CMLB_INFO,  cl, "         relsect    "
2502 		    "numsect         sysid       bootid\n");
2503 		for (i = 0; i < FD_NUMPART; i++, fdp++) {
2504 			cmlb_dbg(CMLB_INFO,  cl,
2505 			    "    %d:  %8d   %8d     0x%08x     0x%08x\n",
2506 			    i, fdp->relsect, fdp->numsect,
2507 			    fdp->systid, fdp->bootid);
2508 		}
2509 	}
2510 #endif
2511 
2512 	/*
2513 	 * Try to find the unix partition
2514 	 */
2515 	uidx = -1;
2516 	solaris_offset = 0;
2517 	solaris_size   = 0;
2518 
2519 	for (fdp = fdisk, i = 0; i < FD_NUMPART; i++, fdp++) {
2520 		uint32_t relsect;
2521 		uint32_t numsect;
2522 		uchar_t systid;
2523 #if defined(__i386) || defined(__amd64)
2524 		/*
2525 		 * Stores relative block offset from the beginning of the
2526 		 * Extended Partition.
2527 		 */
2528 		int	ext_relsect = 0;
2529 #endif
2530 
2531 		if (fdp->numsect == 0) {
2532 			cl->cl_fmap[i].fmap_start = 0;
2533 			cl->cl_fmap[i].fmap_nblk  = 0;
2534 			continue;
2535 		}
2536 
2537 		/*
2538 		 * Data in the fdisk table is little-endian.
2539 		 */
2540 		relsect = LE_32(fdp->relsect);
2541 		numsect = LE_32(fdp->numsect);
2542 
2543 		cl->cl_fmap[i].fmap_start = relsect;
2544 		cl->cl_fmap[i].fmap_nblk  = numsect;
2545 		cl->cl_fmap[i].fmap_systid = LE_8(fdp->systid);
2546 
2547 #if defined(__i386) || defined(__amd64)
2548 		/* Support only one extended partition per LUN */
2549 		if ((fdp->systid == EXTDOS || fdp->systid == FDISK_EXTLBA) &&
2550 		    (ext_part_exists == 0)) {
2551 			int j;
2552 			uint32_t logdrive_offset;
2553 			uint32_t ext_numsect;
2554 			uint32_t abs_secnum;
2555 
2556 			ext_part_exists = 1;
2557 
2558 			for (j = FD_NUMPART; j < FDISK_PARTS; j++) {
2559 				mutex_exit(CMLB_MUTEX(cl));
2560 				rval = DK_TG_READ(cl, bufp,
2561 				    (relsect + ext_relsect), blocksize,
2562 				    tg_cookie);
2563 				mutex_enter(CMLB_MUTEX(cl));
2564 
2565 				if (rval != 0) {
2566 					cmlb_dbg(CMLB_ERROR,  cl,
2567 					    "cmlb_read_fdisk: Extended "
2568 					    "partition read err\n");
2569 					goto done;
2570 				}
2571 				/*
2572 				 * The first ipart entry provides the offset
2573 				 * at which the logical drive starts off from
2574 				 * the beginning of the container partition
2575 				 * and the size of the logical drive.
2576 				 * The second ipart entry provides the offset
2577 				 * of the next container partition from the
2578 				 * beginning of the extended partition.
2579 				 */
2580 				bcopy(&bufp[FDISK_PART_TABLE_START], eparts,
2581 				    sizeof (eparts));
2582 				logdrive_offset = LE_32(efdp1->relsect);
2583 				ext_numsect = LE_32(efdp1->numsect);
2584 				systid = LE_8(efdp1->systid);
2585 				if (logdrive_offset <= 0 || ext_numsect <= 0)
2586 					break;
2587 				abs_secnum = relsect + ext_relsect +
2588 				    logdrive_offset;
2589 
2590 				/* Boundary condition and overlap checking */
2591 				if (cmlb_validate_ext_part(cl, i, j, abs_secnum,
2592 				    ext_numsect)) {
2593 					break;
2594 				}
2595 
2596 				if ((cl->cl_fmap[j].fmap_start != abs_secnum) ||
2597 				    (cl->cl_fmap[j].fmap_nblk != ext_numsect) ||
2598 				    (cl->cl_fmap[j].fmap_systid != systid)) {
2599 					/*
2600 					 * Indicates change from previous
2601 					 * partinfo. Need to recreate
2602 					 * logical device nodes.
2603 					 */
2604 					cl->cl_update_ext_minor_nodes = 1;
2605 				}
2606 				cl->cl_fmap[j].fmap_start = abs_secnum;
2607 				cl->cl_fmap[j].fmap_nblk  = ext_numsect;
2608 				cl->cl_fmap[j].fmap_systid = systid;
2609 				ld_count++;
2610 
2611 				if ((efdp1->systid == SUNIXOS &&
2612 				    (cmlb_is_linux_swap(cl, abs_secnum,
2613 				    tg_cookie) != 0)) ||
2614 				    efdp1->systid == SUNIXOS2) {
2615 					if (uidx == -1) {
2616 						uidx = 0;
2617 						solaris_offset = abs_secnum;
2618 						solaris_size = ext_numsect;
2619 					}
2620 				}
2621 
2622 				if ((ext_relsect = LE_32(efdp2->relsect)) == 0)
2623 					break;
2624 			}
2625 		}
2626 
2627 #endif
2628 
2629 		if (fdp->systid != SUNIXOS &&
2630 		    fdp->systid != SUNIXOS2 &&
2631 		    fdp->systid != EFI_PMBR) {
2632 			continue;
2633 		}
2634 
2635 		/*
2636 		 * use the last active solaris partition id found
2637 		 * (there should only be 1 active partition id)
2638 		 *
2639 		 * if there are no active solaris partition id
2640 		 * then use the first inactive solaris partition id
2641 		 */
2642 		if ((uidx == -1) || (fdp->bootid == ACTIVE)) {
2643 #if defined(__i386) || defined(__amd64)
2644 			if (fdp->systid != SUNIXOS ||
2645 			    (fdp->systid == SUNIXOS &&
2646 			    (cmlb_is_linux_swap(cl, relsect,
2647 			    tg_cookie) != 0))) {
2648 #endif
2649 				uidx = i;
2650 				solaris_offset = relsect;
2651 				solaris_size   = numsect;
2652 #if defined(__i386) || defined(__amd64)
2653 			}
2654 #endif
2655 		}
2656 	}
2657 #if defined(__i386) || defined(__amd64)
2658 	if (ld_count < cl->cl_logical_drive_count) {
2659 		/*
2660 		 * Some/all logical drives were deleted. Clear out
2661 		 * the fmap entries correspoding to those deleted drives.
2662 		 */
2663 		for (k = ld_count + FD_NUMPART;
2664 		    k < cl->cl_logical_drive_count + FD_NUMPART; k++) {
2665 			cl->cl_fmap[k].fmap_start = 0;
2666 			cl->cl_fmap[k].fmap_nblk  = 0;
2667 			cl->cl_fmap[k].fmap_systid = 0;
2668 		}
2669 		cl->cl_update_ext_minor_nodes = 1;
2670 	}
2671 	if (cl->cl_update_ext_minor_nodes) {
2672 		rval = cmlb_update_ext_minor_nodes(cl, ld_count);
2673 		if (rval != 0) {
2674 			goto done;
2675 		}
2676 	}
2677 #endif
2678 	cmlb_dbg(CMLB_INFO,  cl, "fdisk 0x%x 0x%lx",
2679 	    cl->cl_solaris_offset, cl->cl_solaris_size);
2680 done:
2681 
2682 	/*
2683 	 * Clear the VTOC info, only if the Solaris partition entry
2684 	 * has moved, changed size, been deleted, or if the size of
2685 	 * the partition is too small to even fit the label sector.
2686 	 */
2687 	if ((cl->cl_solaris_offset != solaris_offset) ||
2688 	    (cl->cl_solaris_size != solaris_size) ||
2689 	    solaris_size <= DK_LABEL_LOC) {
2690 		cmlb_dbg(CMLB_INFO,  cl, "fdisk moved 0x%x 0x%lx",
2691 		    solaris_offset, solaris_size);
2692 		bzero(&cl->cl_g, sizeof (struct dk_geom));
2693 		bzero(&cl->cl_vtoc, sizeof (struct dk_vtoc));
2694 		bzero(&cl->cl_map, NDKMAP * (sizeof (struct dk_map)));
2695 		cl->cl_f_geometry_is_valid = B_FALSE;
2696 	}
2697 	cl->cl_solaris_offset = solaris_offset;
2698 	cl->cl_solaris_size = solaris_size;
2699 	kmem_free(bufp, blocksize);
2700 	return (rval);
2701 
2702 #else	/* #elif defined(_FIRMWARE_NEEDS_FDISK) */
2703 #error "fdisk table presence undetermined for this platform."
2704 #endif	/* #if defined(_NO_FDISK_PRESENT) */
2705 }
2706 
2707 static void
2708 cmlb_swap_efi_gpt(efi_gpt_t *e)
2709 {
2710 	_NOTE(ASSUMING_PROTECTED(*e))
2711 	e->efi_gpt_Signature = LE_64(e->efi_gpt_Signature);
2712 	e->efi_gpt_Revision = LE_32(e->efi_gpt_Revision);
2713 	e->efi_gpt_HeaderSize = LE_32(e->efi_gpt_HeaderSize);
2714 	e->efi_gpt_HeaderCRC32 = LE_32(e->efi_gpt_HeaderCRC32);
2715 	e->efi_gpt_MyLBA = LE_64(e->efi_gpt_MyLBA);
2716 	e->efi_gpt_AlternateLBA = LE_64(e->efi_gpt_AlternateLBA);
2717 	e->efi_gpt_FirstUsableLBA = LE_64(e->efi_gpt_FirstUsableLBA);
2718 	e->efi_gpt_LastUsableLBA = LE_64(e->efi_gpt_LastUsableLBA);
2719 	UUID_LE_CONVERT(e->efi_gpt_DiskGUID, e->efi_gpt_DiskGUID);
2720 	e->efi_gpt_PartitionEntryLBA = LE_64(e->efi_gpt_PartitionEntryLBA);
2721 	e->efi_gpt_NumberOfPartitionEntries =
2722 	    LE_32(e->efi_gpt_NumberOfPartitionEntries);
2723 	e->efi_gpt_SizeOfPartitionEntry =
2724 	    LE_32(e->efi_gpt_SizeOfPartitionEntry);
2725 	e->efi_gpt_PartitionEntryArrayCRC32 =
2726 	    LE_32(e->efi_gpt_PartitionEntryArrayCRC32);
2727 }
2728 
2729 static void
2730 cmlb_swap_efi_gpe(int nparts, efi_gpe_t *p)
2731 {
2732 	int i;
2733 
2734 	_NOTE(ASSUMING_PROTECTED(*p))
2735 	for (i = 0; i < nparts; i++) {
2736 		UUID_LE_CONVERT(p[i].efi_gpe_PartitionTypeGUID,
2737 		    p[i].efi_gpe_PartitionTypeGUID);
2738 		p[i].efi_gpe_StartingLBA = LE_64(p[i].efi_gpe_StartingLBA);
2739 		p[i].efi_gpe_EndingLBA = LE_64(p[i].efi_gpe_EndingLBA);
2740 		/* PartitionAttrs */
2741 	}
2742 }
2743 
2744 static int
2745 cmlb_validate_efi(efi_gpt_t *labp)
2746 {
2747 	if (labp->efi_gpt_Signature != EFI_SIGNATURE)
2748 		return (EINVAL);
2749 	/* at least 96 bytes in this version of the spec. */
2750 	if (sizeof (efi_gpt_t) - sizeof (labp->efi_gpt_Reserved2) >
2751 	    labp->efi_gpt_HeaderSize)
2752 		return (EINVAL);
2753 	/* this should be 128 bytes */
2754 	if (labp->efi_gpt_SizeOfPartitionEntry != sizeof (efi_gpe_t))
2755 		return (EINVAL);
2756 	return (0);
2757 }
2758 
2759 /*
2760  * This function returns B_FALSE if there is a valid MBR signature and no
2761  * partition table entries of type EFI_PMBR (0xEE). Otherwise it returns B_TRUE.
2762  *
2763  * The EFI spec (1.10 and later) requires having a Protective MBR (PMBR) to
2764  * recognize the disk as GPT partitioned. However, some other OS creates an MBR
2765  * where a PMBR entry is not the only one. Also, if the first block has been
2766  * corrupted, currently best attempt to allow data access would be to try to
2767  * check for GPT headers. Hence in case of more than one partition entry, but
2768  * at least one EFI_PMBR partition type or no valid magic number, the function
2769  * returns B_TRUE to continue with looking for GPT header.
2770  */
2771 
2772 static boolean_t
2773 cmlb_check_efi_mbr(uchar_t *buf, boolean_t *is_mbr)
2774 {
2775 	struct ipart	*fdp;
2776 	struct mboot	*mbp = (struct mboot *)buf;
2777 	struct ipart	fdisk[FD_NUMPART];
2778 	int		i;
2779 
2780 	if (is_mbr != NULL)
2781 		*is_mbr = B_TRUE;
2782 
2783 	if (LE_16(mbp->signature) != MBB_MAGIC) {
2784 		if (is_mbr != NULL)
2785 			*is_mbr = B_FALSE;
2786 		return (B_TRUE);
2787 	}
2788 
2789 	bcopy(&mbp->parts[0], fdisk, sizeof (fdisk));
2790 
2791 	for (fdp = fdisk, i = 0; i < FD_NUMPART; i++, fdp++) {
2792 		if (fdp->systid == EFI_PMBR)
2793 			return (B_TRUE);
2794 	}
2795 
2796 	return (B_FALSE);
2797 }
2798 
2799 static int
2800 cmlb_use_efi(struct cmlb_lun *cl, diskaddr_t capacity, int flags,
2801     void *tg_cookie)
2802 {
2803 	int		i;
2804 	int		rval = 0;
2805 	efi_gpe_t	*partitions;
2806 	uchar_t		*buf;
2807 	uint_t		lbasize;	/* is really how much to read */
2808 	diskaddr_t	cap = 0;
2809 	uint_t		nparts;
2810 	diskaddr_t	gpe_lba;
2811 	diskaddr_t	alternate_lba;
2812 	int		iofailed = 0;
2813 	struct uuid	uuid_type_reserved = EFI_RESERVED;
2814 #if defined(_FIRMWARE_NEEDS_FDISK)
2815 	boolean_t 	is_mbr;
2816 #endif
2817 
2818 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
2819 
2820 	lbasize = cl->cl_sys_blocksize;
2821 
2822 	cl->cl_reserved = -1;
2823 	mutex_exit(CMLB_MUTEX(cl));
2824 
2825 	buf = kmem_zalloc(EFI_MIN_ARRAY_SIZE, KM_SLEEP);
2826 
2827 	rval = DK_TG_READ(cl, buf,  0, lbasize, tg_cookie);
2828 	if (rval) {
2829 		iofailed = 1;
2830 		goto done_err;
2831 	}
2832 	if (((struct dk_label *)buf)->dkl_magic == DKL_MAGIC) {
2833 		/* not ours */
2834 		rval = ESRCH;
2835 		goto done_err;
2836 	}
2837 
2838 #if defined(_FIRMWARE_NEEDS_FDISK)
2839 	if (!cmlb_check_efi_mbr(buf, &is_mbr)) {
2840 		if (is_mbr)
2841 			rval = ESRCH;
2842 		else
2843 			rval = EINVAL;
2844 		goto done_err;
2845 	}
2846 #else
2847 	if (!cmlb_check_efi_mbr(buf, NULL)) {
2848 		rval = EINVAL;
2849 		goto done_err;
2850 	}
2851 
2852 #endif
2853 
2854 	rval = DK_TG_READ(cl, buf, 1, lbasize, tg_cookie);
2855 	if (rval) {
2856 		iofailed = 1;
2857 		goto done_err;
2858 	}
2859 	cmlb_swap_efi_gpt((efi_gpt_t *)buf);
2860 
2861 	if ((rval = cmlb_validate_efi((efi_gpt_t *)buf)) != 0) {
2862 		/*
2863 		 * Couldn't read the primary, try the backup.  Our
2864 		 * capacity at this point could be based on CHS, so
2865 		 * check what the device reports.
2866 		 */
2867 		rval = DK_TG_GETCAP(cl, &cap, tg_cookie);
2868 		if (rval) {
2869 			iofailed = 1;
2870 			goto done_err;
2871 		}
2872 
2873 		/*
2874 		 * CMLB_OFF_BY_ONE case, we check the next to last block first
2875 		 * for backup GPT header, otherwise check the last block.
2876 		 */
2877 
2878 		if ((rval = DK_TG_READ(cl, buf,
2879 		    cap - ((cl->cl_alter_behavior & CMLB_OFF_BY_ONE) ? 2 : 1),
2880 		    lbasize, tg_cookie))
2881 		    != 0) {
2882 			iofailed = 1;
2883 			goto done_err;
2884 		}
2885 		cmlb_swap_efi_gpt((efi_gpt_t *)buf);
2886 
2887 		if ((rval = cmlb_validate_efi((efi_gpt_t *)buf)) != 0) {
2888 
2889 			if (!(cl->cl_alter_behavior & CMLB_OFF_BY_ONE))
2890 				goto done_err;
2891 			if ((rval = DK_TG_READ(cl, buf, cap - 1, lbasize,
2892 			    tg_cookie)) != 0)
2893 				goto done_err;
2894 			cmlb_swap_efi_gpt((efi_gpt_t *)buf);
2895 			if ((rval = cmlb_validate_efi((efi_gpt_t *)buf)) != 0)
2896 				goto done_err;
2897 		}
2898 		if (!(flags & CMLB_SILENT))
2899 			cmlb_log(CMLB_DEVINFO(cl), CMLB_LABEL(cl), CE_WARN,
2900 			    "primary label corrupt; using backup\n");
2901 	}
2902 
2903 	nparts = ((efi_gpt_t *)buf)->efi_gpt_NumberOfPartitionEntries;
2904 	gpe_lba = ((efi_gpt_t *)buf)->efi_gpt_PartitionEntryLBA;
2905 	alternate_lba = ((efi_gpt_t *)buf)->efi_gpt_AlternateLBA;
2906 
2907 	rval = DK_TG_READ(cl, buf, gpe_lba, EFI_MIN_ARRAY_SIZE, tg_cookie);
2908 	if (rval) {
2909 		iofailed = 1;
2910 		goto done_err;
2911 	}
2912 	partitions = (efi_gpe_t *)buf;
2913 
2914 	if (nparts > MAXPART) {
2915 		nparts = MAXPART;
2916 	}
2917 	cmlb_swap_efi_gpe(nparts, partitions);
2918 
2919 	mutex_enter(CMLB_MUTEX(cl));
2920 
2921 	/* Fill in partition table. */
2922 	for (i = 0; i < nparts; i++) {
2923 		if (partitions->efi_gpe_StartingLBA != 0 ||
2924 		    partitions->efi_gpe_EndingLBA != 0) {
2925 			cl->cl_map[i].dkl_cylno =
2926 			    partitions->efi_gpe_StartingLBA;
2927 			cl->cl_map[i].dkl_nblk =
2928 			    partitions->efi_gpe_EndingLBA -
2929 			    partitions->efi_gpe_StartingLBA + 1;
2930 			cl->cl_offset[i] =
2931 			    partitions->efi_gpe_StartingLBA;
2932 		}
2933 
2934 		if (cl->cl_reserved == -1) {
2935 			if (bcmp(&partitions->efi_gpe_PartitionTypeGUID,
2936 			    &uuid_type_reserved, sizeof (struct uuid)) == 0) {
2937 				cl->cl_reserved = i;
2938 			}
2939 		}
2940 		if (i == WD_NODE) {
2941 			/*
2942 			 * minor number 7 corresponds to the whole disk
2943 			 * if the disk capacity is expanded after disk is
2944 			 * labeled, minor number 7 represents the capacity
2945 			 * indicated by the disk label.
2946 			 */
2947 			cl->cl_map[i].dkl_cylno = 0;
2948 			if (alternate_lba == 1) {
2949 				/*
2950 				 * We are using backup label. Since we can
2951 				 * find a valid label at the end of disk,
2952 				 * the disk capacity is not expanded.
2953 				 */
2954 				cl->cl_map[i].dkl_nblk = capacity;
2955 			} else {
2956 				cl->cl_map[i].dkl_nblk = alternate_lba + 1;
2957 			}
2958 			cl->cl_offset[i] = 0;
2959 		}
2960 		partitions++;
2961 	}
2962 	cl->cl_solaris_offset = 0;
2963 	cl->cl_solaris_size = capacity;
2964 	cl->cl_label_from_media = CMLB_LABEL_EFI;
2965 	cl->cl_f_geometry_is_valid = B_TRUE;
2966 
2967 	/* clear the vtoc label */
2968 	bzero(&cl->cl_vtoc, sizeof (struct dk_vtoc));
2969 
2970 	kmem_free(buf, EFI_MIN_ARRAY_SIZE);
2971 	return (0);
2972 
2973 done_err:
2974 	kmem_free(buf, EFI_MIN_ARRAY_SIZE);
2975 	mutex_enter(CMLB_MUTEX(cl));
2976 done_err1:
2977 	/*
2978 	 * if we didn't find something that could look like a VTOC
2979 	 * and the disk is over 1TB, we know there isn't a valid label.
2980 	 * Otherwise let cmlb_uselabel decide what to do.  We only
2981 	 * want to invalidate this if we're certain the label isn't
2982 	 * valid because cmlb_prop_op will now fail, which in turn
2983 	 * causes things like opens and stats on the partition to fail.
2984 	 */
2985 	if ((capacity > CMLB_EXTVTOC_LIMIT) && (rval != ESRCH) && !iofailed) {
2986 		cl->cl_f_geometry_is_valid = B_FALSE;
2987 	}
2988 	return (rval);
2989 }
2990 
2991 
2992 /*
2993  *    Function: cmlb_uselabel
2994  *
2995  * Description: Validate the disk label and update the relevant data (geometry,
2996  *		partition, vtoc, and capacity data) in the cmlb_lun struct.
2997  *		Marks the geometry of the unit as being valid.
2998  *
2999  *   Arguments: cl: unit struct.
3000  *		dk_label: disk label
3001  *
3002  * Return Code: CMLB_LABEL_IS_VALID: Label read from disk is OK; geometry,
3003  *		partition, vtoc, and capacity data are good.
3004  *
3005  *		CMLB_LABEL_IS_INVALID: Magic number or checksum error in the
3006  *		label; or computed capacity does not jibe with capacity
3007  *		reported from the READ CAPACITY command.
3008  *
3009  *     Context: Kernel thread only (can sleep).
3010  */
3011 static int
3012 cmlb_uselabel(struct cmlb_lun *cl, struct dk_label *labp, int flags)
3013 {
3014 	short		*sp;
3015 	short		sum;
3016 	short		count;
3017 	int		label_error = CMLB_LABEL_IS_VALID;
3018 	int		i;
3019 	diskaddr_t	label_capacity;
3020 	uint32_t	part_end;
3021 	diskaddr_t	track_capacity;
3022 #if defined(_SUNOS_VTOC_16)
3023 	struct	dkl_partition	*vpartp;
3024 #endif
3025 	ASSERT(cl != NULL);
3026 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
3027 
3028 	/* Validate the magic number of the label. */
3029 	if (labp->dkl_magic != DKL_MAGIC) {
3030 #if defined(__sparc)
3031 		if (!ISREMOVABLE(cl) && !ISHOTPLUGGABLE(cl)) {
3032 			if (!(flags & CMLB_SILENT))
3033 				cmlb_log(CMLB_DEVINFO(cl), CMLB_LABEL(cl),
3034 				    CE_WARN,
3035 				    "Corrupt label; wrong magic number\n");
3036 		}
3037 #endif
3038 		return (CMLB_LABEL_IS_INVALID);
3039 	}
3040 
3041 	/* Validate the checksum of the label. */
3042 	sp  = (short *)labp;
3043 	sum = 0;
3044 	count = sizeof (struct dk_label) / sizeof (short);
3045 	while (count--)	 {
3046 		sum ^= *sp++;
3047 	}
3048 
3049 	if (sum != 0) {
3050 #if defined(_SUNOS_VTOC_16)
3051 		if (!ISCD(cl)) {
3052 #elif defined(_SUNOS_VTOC_8)
3053 		if (!ISREMOVABLE(cl) && !ISHOTPLUGGABLE(cl)) {
3054 #endif
3055 			if (!(flags & CMLB_SILENT))
3056 				cmlb_log(CMLB_DEVINFO(cl), CMLB_LABEL(cl),
3057 				    CE_WARN,
3058 				    "Corrupt label - label checksum failed\n");
3059 		}
3060 		return (CMLB_LABEL_IS_INVALID);
3061 	}
3062 
3063 
3064 	/*
3065 	 * Fill in geometry structure with data from label.
3066 	 */
3067 	bzero(&cl->cl_g, sizeof (struct dk_geom));
3068 	cl->cl_g.dkg_ncyl   = labp->dkl_ncyl;
3069 	cl->cl_g.dkg_acyl   = labp->dkl_acyl;
3070 	cl->cl_g.dkg_bcyl   = 0;
3071 	cl->cl_g.dkg_nhead  = labp->dkl_nhead;
3072 	cl->cl_g.dkg_nsect  = labp->dkl_nsect;
3073 	cl->cl_g.dkg_intrlv = labp->dkl_intrlv;
3074 
3075 #if defined(_SUNOS_VTOC_8)
3076 	cl->cl_g.dkg_gap1   = labp->dkl_gap1;
3077 	cl->cl_g.dkg_gap2   = labp->dkl_gap2;
3078 	cl->cl_g.dkg_bhead  = labp->dkl_bhead;
3079 #endif
3080 #if defined(_SUNOS_VTOC_16)
3081 	cl->cl_dkg_skew = labp->dkl_skew;
3082 #endif
3083 
3084 #if defined(__i386) || defined(__amd64)
3085 	cl->cl_g.dkg_apc = labp->dkl_apc;
3086 #endif
3087 
3088 	/*
3089 	 * Currently we rely on the values in the label being accurate. If
3090 	 * dkl_rpm or dkl_pcly are zero in the label, use a default value.
3091 	 *
3092 	 * Note: In the future a MODE SENSE may be used to retrieve this data,
3093 	 * although this command is optional in SCSI-2.
3094 	 */
3095 	cl->cl_g.dkg_rpm  = (labp->dkl_rpm  != 0) ? labp->dkl_rpm  : 3600;
3096 	cl->cl_g.dkg_pcyl = (labp->dkl_pcyl != 0) ? labp->dkl_pcyl :
3097 	    (cl->cl_g.dkg_ncyl + cl->cl_g.dkg_acyl);
3098 
3099 	/*
3100 	 * The Read and Write reinstruct values may not be valid
3101 	 * for older disks.
3102 	 */
3103 	cl->cl_g.dkg_read_reinstruct  = labp->dkl_read_reinstruct;
3104 	cl->cl_g.dkg_write_reinstruct = labp->dkl_write_reinstruct;
3105 
3106 	/* Fill in partition table. */
3107 #if defined(_SUNOS_VTOC_8)
3108 	for (i = 0; i < NDKMAP; i++) {
3109 		cl->cl_map[i].dkl_cylno = labp->dkl_map[i].dkl_cylno;
3110 		cl->cl_map[i].dkl_nblk  = labp->dkl_map[i].dkl_nblk;
3111 	}
3112 #endif
3113 #if  defined(_SUNOS_VTOC_16)
3114 	vpartp		= labp->dkl_vtoc.v_part;
3115 	track_capacity	= labp->dkl_nhead * labp->dkl_nsect;
3116 
3117 	/* Prevent divide by zero */
3118 	if (track_capacity == 0) {
3119 		if (!(flags & CMLB_SILENT))
3120 			cmlb_log(CMLB_DEVINFO(cl), CMLB_LABEL(cl), CE_WARN,
3121 			    "Corrupt label - zero nhead or nsect value\n");
3122 
3123 		return (CMLB_LABEL_IS_INVALID);
3124 	}
3125 
3126 	for (i = 0; i < NDKMAP; i++, vpartp++) {
3127 		cl->cl_map[i].dkl_cylno = vpartp->p_start / track_capacity;
3128 		cl->cl_map[i].dkl_nblk  = vpartp->p_size;
3129 	}
3130 #endif
3131 
3132 	/* Fill in VTOC Structure. */
3133 	bcopy(&labp->dkl_vtoc, &cl->cl_vtoc, sizeof (struct dk_vtoc));
3134 #if defined(_SUNOS_VTOC_8)
3135 	/*
3136 	 * The 8-slice vtoc does not include the ascii label; save it into
3137 	 * the device's soft state structure here.
3138 	 */
3139 	bcopy(labp->dkl_asciilabel, cl->cl_asciilabel, LEN_DKL_ASCII);
3140 #endif
3141 
3142 	/* Now look for a valid capacity. */
3143 	track_capacity	= (cl->cl_g.dkg_nhead * cl->cl_g.dkg_nsect);
3144 	label_capacity	= (cl->cl_g.dkg_ncyl  * track_capacity);
3145 
3146 	if (cl->cl_g.dkg_acyl) {
3147 #if defined(__i386) || defined(__amd64)
3148 		/* we may have > 1 alts cylinder */
3149 		label_capacity += (track_capacity * cl->cl_g.dkg_acyl);
3150 #else
3151 		label_capacity += track_capacity;
3152 #endif
3153 	}
3154 
3155 	/*
3156 	 * Force check here to ensure the computed capacity is valid.
3157 	 * If capacity is zero, it indicates an invalid label and
3158 	 * we should abort updating the relevant data then.
3159 	 */
3160 	if (label_capacity == 0) {
3161 		if (!(flags & CMLB_SILENT))
3162 			cmlb_log(CMLB_DEVINFO(cl), CMLB_LABEL(cl), CE_WARN,
3163 			    "Corrupt label - no valid capacity could be "
3164 			    "retrieved\n");
3165 
3166 		return (CMLB_LABEL_IS_INVALID);
3167 	}
3168 
3169 	/* Mark the geometry as valid. */
3170 	cl->cl_f_geometry_is_valid = B_TRUE;
3171 
3172 	/*
3173 	 * if we got invalidated when mutex exit and entered again,
3174 	 * if blockcount different than when we came in, need to
3175 	 * retry from beginning of cmlb_validate_geometry.
3176 	 * revisit this on next phase of utilizing this for
3177 	 * sd.
3178 	 */
3179 
3180 	if (label_capacity <= cl->cl_blockcount) {
3181 #if defined(_SUNOS_VTOC_8)
3182 		/*
3183 		 * We can't let this happen on drives that are subdivided
3184 		 * into logical disks (i.e., that have an fdisk table).
3185 		 * The cl_blockcount field should always hold the full media
3186 		 * size in sectors, period.  This code would overwrite
3187 		 * cl_blockcount with the size of the Solaris fdisk partition.
3188 		 */
3189 		cmlb_dbg(CMLB_ERROR,  cl,
3190 		    "cmlb_uselabel: Label %d blocks; Drive %d blocks\n",
3191 		    label_capacity, cl->cl_blockcount);
3192 		cl->cl_solaris_size = label_capacity;
3193 
3194 #endif	/* defined(_SUNOS_VTOC_8) */
3195 		goto done;
3196 	}
3197 
3198 	if (ISCD(cl)) {
3199 		/* For CDROMs, we trust that the data in the label is OK. */
3200 #if defined(_SUNOS_VTOC_8)
3201 		for (i = 0; i < NDKMAP; i++) {
3202 			part_end = labp->dkl_nhead * labp->dkl_nsect *
3203 			    labp->dkl_map[i].dkl_cylno +
3204 			    labp->dkl_map[i].dkl_nblk  - 1;
3205 
3206 			if ((labp->dkl_map[i].dkl_nblk) &&
3207 			    (part_end > cl->cl_blockcount)) {
3208 				cl->cl_f_geometry_is_valid = B_FALSE;
3209 				break;
3210 			}
3211 		}
3212 #endif
3213 #if defined(_SUNOS_VTOC_16)
3214 		vpartp = &(labp->dkl_vtoc.v_part[0]);
3215 		for (i = 0; i < NDKMAP; i++, vpartp++) {
3216 			part_end = vpartp->p_start + vpartp->p_size;
3217 			if ((vpartp->p_size > 0) &&
3218 			    (part_end > cl->cl_blockcount)) {
3219 				cl->cl_f_geometry_is_valid = B_FALSE;
3220 				break;
3221 			}
3222 		}
3223 #endif
3224 	} else {
3225 		/* label_capacity > cl->cl_blockcount */
3226 		if (!(flags & CMLB_SILENT)) {
3227 			cmlb_log(CMLB_DEVINFO(cl), CMLB_LABEL(cl), CE_WARN,
3228 			    "Corrupt label - bad geometry\n");
3229 			cmlb_log(CMLB_DEVINFO(cl), CMLB_LABEL(cl), CE_CONT,
3230 			    "Label says %llu blocks; Drive says %llu blocks\n",
3231 			    label_capacity, cl->cl_blockcount);
3232 		}
3233 		cl->cl_f_geometry_is_valid = B_FALSE;
3234 		label_error = CMLB_LABEL_IS_INVALID;
3235 	}
3236 
3237 done:
3238 
3239 	cmlb_dbg(CMLB_INFO,  cl, "cmlb_uselabel: (label geometry)\n");
3240 	cmlb_dbg(CMLB_INFO,  cl,
3241 	    "   ncyl: %d; acyl: %d; nhead: %d; nsect: %d\n",
3242 	    cl->cl_g.dkg_ncyl,  cl->cl_g.dkg_acyl,
3243 	    cl->cl_g.dkg_nhead, cl->cl_g.dkg_nsect);
3244 
3245 	cmlb_dbg(CMLB_INFO,  cl,
3246 	    "   label_capacity: %d; intrlv: %d; rpm: %d\n",
3247 	    cl->cl_blockcount, cl->cl_g.dkg_intrlv, cl->cl_g.dkg_rpm);
3248 	cmlb_dbg(CMLB_INFO,  cl, "   wrt_reinstr: %d; rd_reinstr: %d\n",
3249 	    cl->cl_g.dkg_write_reinstruct, cl->cl_g.dkg_read_reinstruct);
3250 
3251 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
3252 
3253 	return (label_error);
3254 }
3255 
3256 
3257 /*
3258  *    Function: cmlb_build_default_label
3259  *
3260  * Description: Generate a default label for those devices that do not have
3261  *		one, e.g., new media, removable cartridges, etc..
3262  *
3263  *     Context: Kernel thread only
3264  */
3265 /*ARGSUSED*/
3266 static void
3267 cmlb_build_default_label(struct cmlb_lun *cl, void *tg_cookie)
3268 {
3269 #if defined(_SUNOS_VTOC_16)
3270 	uint_t	phys_spc;
3271 	uint_t	disksize;
3272 	struct  dk_geom cl_g;
3273 	diskaddr_t capacity;
3274 #endif
3275 
3276 	ASSERT(cl != NULL);
3277 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
3278 
3279 #if defined(_SUNOS_VTOC_8)
3280 	/*
3281 	 * Note: This is a legacy check for non-removable devices on VTOC_8
3282 	 * only. This may be a valid check for VTOC_16 as well.
3283 	 * Once we understand why there is this difference between SPARC and
3284 	 * x86 platform, we could remove this legacy check.
3285 	 */
3286 	if (!ISREMOVABLE(cl) && !ISHOTPLUGGABLE(cl)) {
3287 		return;
3288 	}
3289 #endif
3290 
3291 	bzero(&cl->cl_g, sizeof (struct dk_geom));
3292 	bzero(&cl->cl_vtoc, sizeof (struct dk_vtoc));
3293 	bzero(&cl->cl_map, NDKMAP * (sizeof (struct dk_map)));
3294 
3295 #if defined(_SUNOS_VTOC_8)
3296 
3297 	/*
3298 	 * It's a REMOVABLE media, therefore no label (on sparc, anyway).
3299 	 * But it is still necessary to set up various geometry information,
3300 	 * and we are doing this here.
3301 	 */
3302 
3303 	/*
3304 	 * For the rpm, we use the minimum for the disk.  For the head, cyl,
3305 	 * and number of sector per track, if the capacity <= 1GB, head = 64,
3306 	 * sect = 32.  else head = 255, sect 63 Note: the capacity should be
3307 	 * equal to C*H*S values.  This will cause some truncation of size due
3308 	 * to round off errors. For CD-ROMs, this truncation can have adverse
3309 	 * side effects, so returning ncyl and nhead as 1. The nsect will
3310 	 * overflow for most of CD-ROMs as nsect is of type ushort. (4190569)
3311 	 */
3312 	cl->cl_solaris_size = cl->cl_blockcount;
3313 	if (ISCD(cl)) {
3314 		tg_attribute_t tgattribute;
3315 		int is_writable;
3316 		/*
3317 		 * Preserve the old behavior for non-writable
3318 		 * medias. Since dkg_nsect is a ushort, it
3319 		 * will lose bits as cdroms have more than
3320 		 * 65536 sectors. So if we recalculate
3321 		 * capacity, it will become much shorter.
3322 		 * But the dkg_* information is not
3323 		 * used for CDROMs so it is OK. But for
3324 		 * Writable CDs we need this information
3325 		 * to be valid (for newfs say). So we
3326 		 * make nsect and nhead > 1 that way
3327 		 * nsect can still stay within ushort limit
3328 		 * without losing any bits.
3329 		 */
3330 
3331 		bzero(&tgattribute, sizeof (tg_attribute_t));
3332 
3333 		mutex_exit(CMLB_MUTEX(cl));
3334 		is_writable =
3335 		    (DK_TG_GETATTRIBUTE(cl, &tgattribute, tg_cookie) == 0) ?
3336 		    tgattribute.media_is_writable : 1;
3337 		mutex_enter(CMLB_MUTEX(cl));
3338 
3339 		if (is_writable) {
3340 			cl->cl_g.dkg_nhead = 64;
3341 			cl->cl_g.dkg_nsect = 32;
3342 			cl->cl_g.dkg_ncyl = cl->cl_blockcount / (64 * 32);
3343 			cl->cl_solaris_size = (diskaddr_t)cl->cl_g.dkg_ncyl *
3344 			    cl->cl_g.dkg_nhead * cl->cl_g.dkg_nsect;
3345 		} else {
3346 			cl->cl_g.dkg_ncyl  = 1;
3347 			cl->cl_g.dkg_nhead = 1;
3348 			cl->cl_g.dkg_nsect = cl->cl_blockcount;
3349 		}
3350 	} else {
3351 		if (cl->cl_blockcount < 160) {
3352 			/* Less than 80K */
3353 			cl->cl_g.dkg_nhead = 1;
3354 			cl->cl_g.dkg_ncyl = cl->cl_blockcount;
3355 			cl->cl_g.dkg_nsect = 1;
3356 		} else if (cl->cl_blockcount <= 0x1000) {
3357 			/* unlabeled SCSI floppy device */
3358 			cl->cl_g.dkg_nhead = 2;
3359 			cl->cl_g.dkg_ncyl = 80;
3360 			cl->cl_g.dkg_nsect = cl->cl_blockcount / (2 * 80);
3361 		} else if (cl->cl_blockcount <= 0x200000) {
3362 			cl->cl_g.dkg_nhead = 64;
3363 			cl->cl_g.dkg_nsect = 32;
3364 			cl->cl_g.dkg_ncyl  = cl->cl_blockcount / (64 * 32);
3365 		} else {
3366 			cl->cl_g.dkg_nhead = 255;
3367 
3368 			cl->cl_g.dkg_nsect = ((cl->cl_blockcount +
3369 			    (UINT16_MAX * 255 * 63) - 1) /
3370 			    (UINT16_MAX * 255 * 63)) * 63;
3371 
3372 			if (cl->cl_g.dkg_nsect == 0)
3373 				cl->cl_g.dkg_nsect = (UINT16_MAX / 63) * 63;
3374 
3375 			cl->cl_g.dkg_ncyl = cl->cl_blockcount /
3376 			    (255 * cl->cl_g.dkg_nsect);
3377 		}
3378 
3379 		cl->cl_solaris_size =
3380 		    (diskaddr_t)cl->cl_g.dkg_ncyl * cl->cl_g.dkg_nhead *
3381 		    cl->cl_g.dkg_nsect;
3382 
3383 	}
3384 
3385 	cl->cl_g.dkg_acyl	= 0;
3386 	cl->cl_g.dkg_bcyl	= 0;
3387 	cl->cl_g.dkg_rpm	= 200;
3388 	cl->cl_asciilabel[0]	= '\0';
3389 	cl->cl_g.dkg_pcyl	= cl->cl_g.dkg_ncyl;
3390 
3391 	cl->cl_map[0].dkl_cylno = 0;
3392 	cl->cl_map[0].dkl_nblk  = cl->cl_solaris_size;
3393 
3394 	cl->cl_map[2].dkl_cylno = 0;
3395 	cl->cl_map[2].dkl_nblk  = cl->cl_solaris_size;
3396 
3397 #elif defined(_SUNOS_VTOC_16)
3398 
3399 	if (cl->cl_solaris_size == 0) {
3400 		/*
3401 		 * Got fdisk table but no solaris entry therefore
3402 		 * don't create a default label
3403 		 */
3404 		cl->cl_f_geometry_is_valid = B_TRUE;
3405 		return;
3406 	}
3407 
3408 	/*
3409 	 * For CDs we continue to use the physical geometry to calculate
3410 	 * number of cylinders. All other devices must convert the
3411 	 * physical geometry (cmlb_geom) to values that will fit
3412 	 * in a dk_geom structure.
3413 	 */
3414 	if (ISCD(cl)) {
3415 		phys_spc = cl->cl_pgeom.g_nhead * cl->cl_pgeom.g_nsect;
3416 	} else {
3417 		/* Convert physical geometry to disk geometry */
3418 		bzero(&cl_g, sizeof (struct dk_geom));
3419 
3420 		/*
3421 		 * Refer to comments related to off-by-1 at the
3422 		 * header of this file.
3423 		 * Before calculating geometry, capacity should be
3424 		 * decreased by 1.
3425 		 */
3426 
3427 		if (cl->cl_alter_behavior & CMLB_OFF_BY_ONE)
3428 			capacity = cl->cl_blockcount - 1;
3429 		else
3430 			capacity = cl->cl_blockcount;
3431 
3432 
3433 		cmlb_convert_geometry(cl, capacity, &cl_g, tg_cookie);
3434 		bcopy(&cl_g, &cl->cl_g, sizeof (cl->cl_g));
3435 		phys_spc = cl->cl_g.dkg_nhead * cl->cl_g.dkg_nsect;
3436 	}
3437 
3438 	if (phys_spc == 0)
3439 		return;
3440 	cl->cl_g.dkg_pcyl = cl->cl_solaris_size / phys_spc;
3441 	if (cl->cl_alter_behavior & CMLB_FAKE_LABEL_ONE_PARTITION) {
3442 		/* disable devid */
3443 		cl->cl_g.dkg_ncyl = cl->cl_g.dkg_pcyl;
3444 		disksize = cl->cl_solaris_size;
3445 	} else {
3446 		cl->cl_g.dkg_acyl = DK_ACYL;
3447 		cl->cl_g.dkg_ncyl = cl->cl_g.dkg_pcyl - DK_ACYL;
3448 		disksize = cl->cl_g.dkg_ncyl * phys_spc;
3449 	}
3450 
3451 	if (ISCD(cl)) {
3452 		/*
3453 		 * CD's don't use the "heads * sectors * cyls"-type of
3454 		 * geometry, but instead use the entire capacity of the media.
3455 		 */
3456 		disksize = cl->cl_solaris_size;
3457 		cl->cl_g.dkg_nhead = 1;
3458 		cl->cl_g.dkg_nsect = 1;
3459 		cl->cl_g.dkg_rpm =
3460 		    (cl->cl_pgeom.g_rpm == 0) ? 200 : cl->cl_pgeom.g_rpm;
3461 
3462 		cl->cl_vtoc.v_part[0].p_start = 0;
3463 		cl->cl_vtoc.v_part[0].p_size  = disksize;
3464 		cl->cl_vtoc.v_part[0].p_tag   = V_BACKUP;
3465 		cl->cl_vtoc.v_part[0].p_flag  = V_UNMNT;
3466 
3467 		cl->cl_map[0].dkl_cylno = 0;
3468 		cl->cl_map[0].dkl_nblk  = disksize;
3469 		cl->cl_offset[0] = 0;
3470 
3471 	} else {
3472 		/*
3473 		 * Hard disks and removable media cartridges
3474 		 */
3475 		cl->cl_g.dkg_rpm =
3476 		    (cl->cl_pgeom.g_rpm == 0) ? 3600: cl->cl_pgeom.g_rpm;
3477 		cl->cl_vtoc.v_sectorsz = cl->cl_sys_blocksize;
3478 
3479 		/* Add boot slice */
3480 		cl->cl_vtoc.v_part[8].p_start = 0;
3481 		cl->cl_vtoc.v_part[8].p_size  = phys_spc;
3482 		cl->cl_vtoc.v_part[8].p_tag   = V_BOOT;
3483 		cl->cl_vtoc.v_part[8].p_flag  = V_UNMNT;
3484 
3485 		cl->cl_map[8].dkl_cylno = 0;
3486 		cl->cl_map[8].dkl_nblk  = phys_spc;
3487 		cl->cl_offset[8] = 0;
3488 
3489 		if ((cl->cl_alter_behavior &
3490 		    CMLB_CREATE_ALTSLICE_VTOC_16_DTYPE_DIRECT) &&
3491 		    cl->cl_device_type == DTYPE_DIRECT) {
3492 			cl->cl_vtoc.v_part[9].p_start = phys_spc;
3493 			cl->cl_vtoc.v_part[9].p_size  = 2 * phys_spc;
3494 			cl->cl_vtoc.v_part[9].p_tag   = V_ALTSCTR;
3495 			cl->cl_vtoc.v_part[9].p_flag  = 0;
3496 
3497 			cl->cl_map[9].dkl_cylno = 1;
3498 			cl->cl_map[9].dkl_nblk  = 2 * phys_spc;
3499 			cl->cl_offset[9] = phys_spc;
3500 		}
3501 	}
3502 
3503 	cl->cl_g.dkg_apc = 0;
3504 
3505 	/* Add backup slice */
3506 	cl->cl_vtoc.v_part[2].p_start = 0;
3507 	cl->cl_vtoc.v_part[2].p_size  = disksize;
3508 	cl->cl_vtoc.v_part[2].p_tag   = V_BACKUP;
3509 	cl->cl_vtoc.v_part[2].p_flag  = V_UNMNT;
3510 
3511 	cl->cl_map[2].dkl_cylno = 0;
3512 	cl->cl_map[2].dkl_nblk  = disksize;
3513 	cl->cl_offset[2] = 0;
3514 
3515 	/*
3516 	 * single slice (s0) covering the entire disk
3517 	 */
3518 	if (cl->cl_alter_behavior & CMLB_FAKE_LABEL_ONE_PARTITION) {
3519 		cl->cl_vtoc.v_part[0].p_start = 0;
3520 		cl->cl_vtoc.v_part[0].p_tag   = V_UNASSIGNED;
3521 		cl->cl_vtoc.v_part[0].p_flag  = 0;
3522 		cl->cl_vtoc.v_part[0].p_size  = disksize;
3523 		cl->cl_map[0].dkl_cylno = 0;
3524 		cl->cl_map[0].dkl_nblk  = disksize;
3525 		cl->cl_offset[0] = 0;
3526 	}
3527 
3528 	(void) sprintf(cl->cl_vtoc.v_asciilabel, "DEFAULT cyl %d alt %d"
3529 	    " hd %d sec %d", cl->cl_g.dkg_ncyl, cl->cl_g.dkg_acyl,
3530 	    cl->cl_g.dkg_nhead, cl->cl_g.dkg_nsect);
3531 
3532 #else
3533 #error "No VTOC format defined."
3534 #endif
3535 
3536 	cl->cl_g.dkg_read_reinstruct  = 0;
3537 	cl->cl_g.dkg_write_reinstruct = 0;
3538 
3539 	cl->cl_g.dkg_intrlv = 1;
3540 
3541 	cl->cl_vtoc.v_sanity  = VTOC_SANE;
3542 	cl->cl_vtoc.v_nparts = V_NUMPAR;
3543 	cl->cl_vtoc.v_version = V_VERSION;
3544 
3545 	cl->cl_f_geometry_is_valid = B_TRUE;
3546 	cl->cl_label_from_media = CMLB_LABEL_UNDEF;
3547 
3548 	cmlb_dbg(CMLB_INFO,  cl,
3549 	    "cmlb_build_default_label: Default label created: "
3550 	    "cyl: %d\tacyl: %d\tnhead: %d\tnsect: %d\tcap: %d\n",
3551 	    cl->cl_g.dkg_ncyl, cl->cl_g.dkg_acyl, cl->cl_g.dkg_nhead,
3552 	    cl->cl_g.dkg_nsect, cl->cl_blockcount);
3553 }
3554 
3555 
3556 #if defined(_FIRMWARE_NEEDS_FDISK)
3557 /*
3558  * Max CHS values, as they are encoded into bytes, for 1022/254/63
3559  */
3560 #define	LBA_MAX_SECT	(63 | ((1022 & 0x300) >> 2))
3561 #define	LBA_MAX_CYL	(1022 & 0xFF)
3562 #define	LBA_MAX_HEAD	(254)
3563 
3564 
3565 /*
3566  *    Function: cmlb_has_max_chs_vals
3567  *
3568  * Description: Return B_TRUE if Cylinder-Head-Sector values are all at maximum.
3569  *
3570  *   Arguments: fdp - ptr to CHS info
3571  *
3572  * Return Code: True or false
3573  *
3574  *     Context: Any.
3575  */
3576 static boolean_t
3577 cmlb_has_max_chs_vals(struct ipart *fdp)
3578 {
3579 	return ((fdp->begcyl  == LBA_MAX_CYL)	&&
3580 	    (fdp->beghead == LBA_MAX_HEAD)	&&
3581 	    (fdp->begsect == LBA_MAX_SECT)	&&
3582 	    (fdp->endcyl  == LBA_MAX_CYL)	&&
3583 	    (fdp->endhead == LBA_MAX_HEAD)	&&
3584 	    (fdp->endsect == LBA_MAX_SECT));
3585 }
3586 #endif
3587 
3588 /*
3589  *    Function: cmlb_dkio_get_geometry
3590  *
3591  * Description: This routine is the driver entry point for handling user
3592  *		requests to get the device geometry (DKIOCGGEOM).
3593  *
3594  *   Arguments:
3595  *	arg		pointer to user provided dk_geom structure specifying
3596  *			the controller's notion of the current geometry.
3597  *
3598  *	flag 		this argument is a pass through to ddi_copyxxx()
3599  *			directly from the mode argument of ioctl().
3600  *
3601  *	tg_cookie	cookie from target driver to be passed back to target
3602  *			driver when we call back to it through tg_ops.
3603  *
3604  * Return Code: 0
3605  *		EFAULT
3606  *		ENXIO
3607  *		EIO
3608  */
3609 static int
3610 cmlb_dkio_get_geometry(struct cmlb_lun *cl, caddr_t arg, int flag,
3611     void *tg_cookie)
3612 {
3613 	struct dk_geom	*tmp_geom = NULL;
3614 	int		rval = 0;
3615 
3616 	/*
3617 	 * cmlb_validate_geometry does not spin a disk up
3618 	 * if it was spcl down. We need to make sure it
3619 	 * is ready.
3620 	 */
3621 	mutex_enter(CMLB_MUTEX(cl));
3622 	rval = cmlb_validate_geometry(cl, B_TRUE, 0, tg_cookie);
3623 #if defined(_SUNOS_VTOC_8)
3624 	if (rval == EINVAL &&
3625 	    cl->cl_alter_behavior & CMLB_FAKE_GEOM_LABEL_IOCTLS_VTOC8) {
3626 		/*
3627 		 * This is to return a default label geometry even when we
3628 		 * do not really assume a default label for the device.
3629 		 * dad driver utilizes this.
3630 		 */
3631 		if (cl->cl_blockcount <= CMLB_OLDVTOC_LIMIT) {
3632 			cmlb_setup_default_geometry(cl, tg_cookie);
3633 			rval = 0;
3634 		}
3635 	}
3636 #endif
3637 	if (rval) {
3638 		mutex_exit(CMLB_MUTEX(cl));
3639 		return (rval);
3640 	}
3641 
3642 #if defined(__i386) || defined(__amd64)
3643 	if (cl->cl_solaris_size == 0) {
3644 		mutex_exit(CMLB_MUTEX(cl));
3645 		return (EIO);
3646 	}
3647 #endif
3648 
3649 	/*
3650 	 * Make a local copy of the soft state geometry to avoid some potential
3651 	 * race conditions associated with holding the mutex and updating the
3652 	 * write_reinstruct value
3653 	 */
3654 	tmp_geom = kmem_zalloc(sizeof (struct dk_geom), KM_SLEEP);
3655 	bcopy(&cl->cl_g, tmp_geom, sizeof (struct dk_geom));
3656 
3657 	if (tmp_geom->dkg_write_reinstruct == 0) {
3658 		tmp_geom->dkg_write_reinstruct =
3659 		    (int)((int)(tmp_geom->dkg_nsect * tmp_geom->dkg_rpm *
3660 		    cmlb_rot_delay) / (int)60000);
3661 	}
3662 	mutex_exit(CMLB_MUTEX(cl));
3663 
3664 	rval = ddi_copyout(tmp_geom, (void *)arg, sizeof (struct dk_geom),
3665 	    flag);
3666 	if (rval != 0) {
3667 		rval = EFAULT;
3668 	}
3669 
3670 	kmem_free(tmp_geom, sizeof (struct dk_geom));
3671 	return (rval);
3672 
3673 }
3674 
3675 
3676 /*
3677  *    Function: cmlb_dkio_set_geometry
3678  *
3679  * Description: This routine is the driver entry point for handling user
3680  *		requests to set the device geometry (DKIOCSGEOM). The actual
3681  *		device geometry is not updated, just the driver "notion" of it.
3682  *
3683  *   Arguments:
3684  *	arg		pointer to user provided dk_geom structure used to set
3685  *			the controller's notion of the current geometry.
3686  *
3687  *	flag 		this argument is a pass through to ddi_copyxxx()
3688  *			directly from the mode argument of ioctl().
3689  *
3690  *	tg_cookie	cookie from target driver to be passed back to target
3691  *			driver when we call back to it through tg_ops.
3692  *
3693  * Return Code: 0
3694  *		EFAULT
3695  *		ENXIO
3696  *		EIO
3697  */
3698 static int
3699 cmlb_dkio_set_geometry(struct cmlb_lun *cl, caddr_t arg, int flag)
3700 {
3701 	struct dk_geom	*tmp_geom;
3702 	struct dk_map	*lp;
3703 	int		rval = 0;
3704 	int		i;
3705 
3706 
3707 #if defined(__i386) || defined(__amd64)
3708 	if (cl->cl_solaris_size == 0) {
3709 		return (EIO);
3710 	}
3711 #endif
3712 	/*
3713 	 * We need to copy the user specified geometry into local
3714 	 * storage and then update the softstate. We don't want to hold
3715 	 * the mutex and copyin directly from the user to the soft state
3716 	 */
3717 	tmp_geom = (struct dk_geom *)
3718 	    kmem_zalloc(sizeof (struct dk_geom), KM_SLEEP);
3719 	rval = ddi_copyin(arg, tmp_geom, sizeof (struct dk_geom), flag);
3720 	if (rval != 0) {
3721 		kmem_free(tmp_geom, sizeof (struct dk_geom));
3722 		return (EFAULT);
3723 	}
3724 
3725 	mutex_enter(CMLB_MUTEX(cl));
3726 	bcopy(tmp_geom, &cl->cl_g, sizeof (struct dk_geom));
3727 	for (i = 0; i < NDKMAP; i++) {
3728 		lp  = &cl->cl_map[i];
3729 		cl->cl_offset[i] =
3730 		    cl->cl_g.dkg_nhead * cl->cl_g.dkg_nsect * lp->dkl_cylno;
3731 #if defined(__i386) || defined(__amd64)
3732 		cl->cl_offset[i] += cl->cl_solaris_offset;
3733 #endif
3734 	}
3735 	cl->cl_f_geometry_is_valid = B_FALSE;
3736 	mutex_exit(CMLB_MUTEX(cl));
3737 	kmem_free(tmp_geom, sizeof (struct dk_geom));
3738 
3739 	return (rval);
3740 }
3741 
3742 /*
3743  *    Function: cmlb_dkio_get_partition
3744  *
3745  * Description: This routine is the driver entry point for handling user
3746  *		requests to get the partition table (DKIOCGAPART).
3747  *
3748  *   Arguments:
3749  *	arg		pointer to user provided dk_allmap structure specifying
3750  *			the controller's notion of the current partition table.
3751  *
3752  *	flag		this argument is a pass through to ddi_copyxxx()
3753  *			directly from the mode argument of ioctl().
3754  *
3755  *	tg_cookie	cookie from target driver to be passed back to target
3756  *			driver when we call back to it through tg_ops.
3757  *
3758  * Return Code: 0
3759  *		EFAULT
3760  *		ENXIO
3761  *		EIO
3762  */
3763 static int
3764 cmlb_dkio_get_partition(struct cmlb_lun *cl, caddr_t arg, int flag,
3765     void *tg_cookie)
3766 {
3767 	int		rval = 0;
3768 	int		size;
3769 
3770 	/*
3771 	 * Make sure the geometry is valid before getting the partition
3772 	 * information.
3773 	 */
3774 	mutex_enter(CMLB_MUTEX(cl));
3775 	if ((rval = cmlb_validate_geometry(cl, B_TRUE, 0, tg_cookie)) != 0) {
3776 		mutex_exit(CMLB_MUTEX(cl));
3777 		return (rval);
3778 	}
3779 	mutex_exit(CMLB_MUTEX(cl));
3780 
3781 #if defined(__i386) || defined(__amd64)
3782 	if (cl->cl_solaris_size == 0) {
3783 		return (EIO);
3784 	}
3785 #endif
3786 
3787 #ifdef _MULTI_DATAMODEL
3788 	switch (ddi_model_convert_from(flag & FMODELS)) {
3789 	case DDI_MODEL_ILP32: {
3790 		struct dk_map32 dk_map32[NDKMAP];
3791 		int		i;
3792 
3793 		for (i = 0; i < NDKMAP; i++) {
3794 			dk_map32[i].dkl_cylno = cl->cl_map[i].dkl_cylno;
3795 			dk_map32[i].dkl_nblk  = cl->cl_map[i].dkl_nblk;
3796 		}
3797 		size = NDKMAP * sizeof (struct dk_map32);
3798 		rval = ddi_copyout(dk_map32, (void *)arg, size, flag);
3799 		if (rval != 0) {
3800 			rval = EFAULT;
3801 		}
3802 		break;
3803 	}
3804 	case DDI_MODEL_NONE:
3805 		size = NDKMAP * sizeof (struct dk_map);
3806 		rval = ddi_copyout(cl->cl_map, (void *)arg, size, flag);
3807 		if (rval != 0) {
3808 			rval = EFAULT;
3809 		}
3810 		break;
3811 	}
3812 #else /* ! _MULTI_DATAMODEL */
3813 	size = NDKMAP * sizeof (struct dk_map);
3814 	rval = ddi_copyout(cl->cl_map, (void *)arg, size, flag);
3815 	if (rval != 0) {
3816 		rval = EFAULT;
3817 	}
3818 #endif /* _MULTI_DATAMODEL */
3819 	return (rval);
3820 }
3821 
3822 /*
3823  *    Function: cmlb_dkio_set_partition
3824  *
3825  * Description: This routine is the driver entry point for handling user
3826  *		requests to set the partition table (DKIOCSAPART). The actual
3827  *		device partition is not updated.
3828  *
3829  *   Arguments:
3830  *		arg  - pointer to user provided dk_allmap structure used to set
3831  *			the controller's notion of the partition table.
3832  *		flag - this argument is a pass through to ddi_copyxxx()
3833  *		       directly from the mode argument of ioctl().
3834  *
3835  * Return Code: 0
3836  *		EINVAL
3837  *		EFAULT
3838  *		ENXIO
3839  *		EIO
3840  */
3841 static int
3842 cmlb_dkio_set_partition(struct cmlb_lun *cl, caddr_t arg, int flag)
3843 {
3844 	struct dk_map	dk_map[NDKMAP];
3845 	struct dk_map	*lp;
3846 	int		rval = 0;
3847 	int		size;
3848 	int		i;
3849 #if defined(_SUNOS_VTOC_16)
3850 	struct dkl_partition	*vp;
3851 #endif
3852 
3853 	/*
3854 	 * Set the map for all logical partitions.  We lock
3855 	 * the priority just to make sure an interrupt doesn't
3856 	 * come in while the map is half updated.
3857 	 */
3858 	_NOTE(DATA_READABLE_WITHOUT_LOCK(cmlb_lun::cl_solaris_size))
3859 	mutex_enter(CMLB_MUTEX(cl));
3860 
3861 	if (cl->cl_blockcount > CMLB_OLDVTOC_LIMIT) {
3862 		mutex_exit(CMLB_MUTEX(cl));
3863 		return (ENOTSUP);
3864 	}
3865 	mutex_exit(CMLB_MUTEX(cl));
3866 	if (cl->cl_solaris_size == 0) {
3867 		return (EIO);
3868 	}
3869 
3870 #ifdef _MULTI_DATAMODEL
3871 	switch (ddi_model_convert_from(flag & FMODELS)) {
3872 	case DDI_MODEL_ILP32: {
3873 		struct dk_map32 dk_map32[NDKMAP];
3874 
3875 		size = NDKMAP * sizeof (struct dk_map32);
3876 		rval = ddi_copyin((void *)arg, dk_map32, size, flag);
3877 		if (rval != 0) {
3878 			return (EFAULT);
3879 		}
3880 		for (i = 0; i < NDKMAP; i++) {
3881 			dk_map[i].dkl_cylno = dk_map32[i].dkl_cylno;
3882 			dk_map[i].dkl_nblk  = dk_map32[i].dkl_nblk;
3883 		}
3884 		break;
3885 	}
3886 	case DDI_MODEL_NONE:
3887 		size = NDKMAP * sizeof (struct dk_map);
3888 		rval = ddi_copyin((void *)arg, dk_map, size, flag);
3889 		if (rval != 0) {
3890 			return (EFAULT);
3891 		}
3892 		break;
3893 	}
3894 #else /* ! _MULTI_DATAMODEL */
3895 	size = NDKMAP * sizeof (struct dk_map);
3896 	rval = ddi_copyin((void *)arg, dk_map, size, flag);
3897 	if (rval != 0) {
3898 		return (EFAULT);
3899 	}
3900 #endif /* _MULTI_DATAMODEL */
3901 
3902 	mutex_enter(CMLB_MUTEX(cl));
3903 	/* Note: The size used in this bcopy is set based upon the data model */
3904 	bcopy(dk_map, cl->cl_map, size);
3905 #if defined(_SUNOS_VTOC_16)
3906 	vp = (struct dkl_partition *)&(cl->cl_vtoc);
3907 #endif	/* defined(_SUNOS_VTOC_16) */
3908 	for (i = 0; i < NDKMAP; i++) {
3909 		lp  = &cl->cl_map[i];
3910 		cl->cl_offset[i] =
3911 		    cl->cl_g.dkg_nhead * cl->cl_g.dkg_nsect * lp->dkl_cylno;
3912 #if defined(_SUNOS_VTOC_16)
3913 		vp->p_start = cl->cl_offset[i];
3914 		vp->p_size = lp->dkl_nblk;
3915 		vp++;
3916 #endif	/* defined(_SUNOS_VTOC_16) */
3917 #if defined(__i386) || defined(__amd64)
3918 		cl->cl_offset[i] += cl->cl_solaris_offset;
3919 #endif
3920 	}
3921 	mutex_exit(CMLB_MUTEX(cl));
3922 	return (rval);
3923 }
3924 
3925 
3926 /*
3927  *    Function: cmlb_dkio_get_vtoc
3928  *
3929  * Description: This routine is the driver entry point for handling user
3930  *		requests to get the current volume table of contents
3931  *		(DKIOCGVTOC).
3932  *
3933  *   Arguments:
3934  *	arg		pointer to user provided vtoc structure specifying
3935  *			the current vtoc.
3936  *
3937  *	flag		this argument is a pass through to ddi_copyxxx()
3938  *			directly from the mode argument of ioctl().
3939  *
3940  *	tg_cookie	cookie from target driver to be passed back to target
3941  *			driver when we call back to it through tg_ops.
3942  *
3943  * Return Code: 0
3944  *		EFAULT
3945  *		ENXIO
3946  *		EIO
3947  */
3948 static int
3949 cmlb_dkio_get_vtoc(struct cmlb_lun *cl, caddr_t arg, int flag, void *tg_cookie)
3950 {
3951 #if defined(_SUNOS_VTOC_8)
3952 	struct vtoc	user_vtoc;
3953 #endif	/* defined(_SUNOS_VTOC_8) */
3954 	int		rval = 0;
3955 
3956 	mutex_enter(CMLB_MUTEX(cl));
3957 	if (cl->cl_blockcount > CMLB_OLDVTOC_LIMIT) {
3958 		mutex_exit(CMLB_MUTEX(cl));
3959 		return (EOVERFLOW);
3960 	}
3961 
3962 	rval = cmlb_validate_geometry(cl, B_TRUE, 0, tg_cookie);
3963 
3964 #if defined(_SUNOS_VTOC_8)
3965 	if (rval == EINVAL &&
3966 	    (cl->cl_alter_behavior & CMLB_FAKE_GEOM_LABEL_IOCTLS_VTOC8)) {
3967 		/*
3968 		 * This is to return a default label even when we do not
3969 		 * really assume a default label for the device.
3970 		 * dad driver utilizes this.
3971 		 */
3972 		if (cl->cl_blockcount <= CMLB_OLDVTOC_LIMIT) {
3973 			cmlb_setup_default_geometry(cl, tg_cookie);
3974 			rval = 0;
3975 		}
3976 	}
3977 #endif
3978 	if (rval) {
3979 		mutex_exit(CMLB_MUTEX(cl));
3980 		return (rval);
3981 	}
3982 
3983 #if defined(_SUNOS_VTOC_8)
3984 	cmlb_build_user_vtoc(cl, &user_vtoc);
3985 	mutex_exit(CMLB_MUTEX(cl));
3986 
3987 #ifdef _MULTI_DATAMODEL
3988 	switch (ddi_model_convert_from(flag & FMODELS)) {
3989 	case DDI_MODEL_ILP32: {
3990 		struct vtoc32 user_vtoc32;
3991 
3992 		vtoctovtoc32(user_vtoc, user_vtoc32);
3993 		if (ddi_copyout(&user_vtoc32, (void *)arg,
3994 		    sizeof (struct vtoc32), flag)) {
3995 			return (EFAULT);
3996 		}
3997 		break;
3998 	}
3999 
4000 	case DDI_MODEL_NONE:
4001 		if (ddi_copyout(&user_vtoc, (void *)arg,
4002 		    sizeof (struct vtoc), flag)) {
4003 			return (EFAULT);
4004 		}
4005 		break;
4006 	}
4007 #else /* ! _MULTI_DATAMODEL */
4008 	if (ddi_copyout(&user_vtoc, (void *)arg, sizeof (struct vtoc), flag)) {
4009 		return (EFAULT);
4010 	}
4011 #endif /* _MULTI_DATAMODEL */
4012 
4013 #elif defined(_SUNOS_VTOC_16)
4014 	mutex_exit(CMLB_MUTEX(cl));
4015 
4016 #ifdef _MULTI_DATAMODEL
4017 	/*
4018 	 * The cl_vtoc structure is a "struct dk_vtoc"  which is always
4019 	 * 32-bit to maintain compatibility with existing on-disk
4020 	 * structures.  Thus, we need to convert the structure when copying
4021 	 * it out to a datamodel-dependent "struct vtoc" in a 64-bit
4022 	 * program.  If the target is a 32-bit program, then no conversion
4023 	 * is necessary.
4024 	 */
4025 	/* LINTED: logical expression always true: op "||" */
4026 	ASSERT(sizeof (cl->cl_vtoc) == sizeof (struct vtoc32));
4027 	switch (ddi_model_convert_from(flag & FMODELS)) {
4028 	case DDI_MODEL_ILP32:
4029 		if (ddi_copyout(&(cl->cl_vtoc), (void *)arg,
4030 		    sizeof (cl->cl_vtoc), flag)) {
4031 			return (EFAULT);
4032 		}
4033 		break;
4034 
4035 	case DDI_MODEL_NONE: {
4036 		struct vtoc user_vtoc;
4037 
4038 		vtoc32tovtoc(cl->cl_vtoc, user_vtoc);
4039 		if (ddi_copyout(&user_vtoc, (void *)arg,
4040 		    sizeof (struct vtoc), flag)) {
4041 			return (EFAULT);
4042 		}
4043 		break;
4044 	}
4045 	}
4046 #else /* ! _MULTI_DATAMODEL */
4047 	if (ddi_copyout(&(cl->cl_vtoc), (void *)arg, sizeof (cl->cl_vtoc),
4048 	    flag)) {
4049 		return (EFAULT);
4050 	}
4051 #endif /* _MULTI_DATAMODEL */
4052 #else
4053 #error "No VTOC format defined."
4054 #endif
4055 
4056 	return (rval);
4057 }
4058 
4059 
4060 /*
4061  *    Function: cmlb_dkio_get_extvtoc
4062  */
4063 static int
4064 cmlb_dkio_get_extvtoc(struct cmlb_lun *cl, caddr_t arg, int flag,
4065     void *tg_cookie)
4066 {
4067 	struct extvtoc	ext_vtoc;
4068 #if defined(_SUNOS_VTOC_8)
4069 	struct vtoc	user_vtoc;
4070 #endif	/* defined(_SUNOS_VTOC_8) */
4071 	int		rval = 0;
4072 
4073 	bzero(&ext_vtoc, sizeof (struct extvtoc));
4074 	mutex_enter(CMLB_MUTEX(cl));
4075 	rval = cmlb_validate_geometry(cl, B_TRUE, 0, tg_cookie);
4076 
4077 #if defined(_SUNOS_VTOC_8)
4078 	if (rval == EINVAL &&
4079 	    (cl->cl_alter_behavior & CMLB_FAKE_GEOM_LABEL_IOCTLS_VTOC8)) {
4080 		/*
4081 		 * This is to return a default label even when we do not
4082 		 * really assume a default label for the device.
4083 		 * dad driver utilizes this.
4084 		 */
4085 		if (cl->cl_blockcount <= CMLB_OLDVTOC_LIMIT) {
4086 			cmlb_setup_default_geometry(cl, tg_cookie);
4087 			rval = 0;
4088 		}
4089 	}
4090 #endif
4091 	if (rval) {
4092 		mutex_exit(CMLB_MUTEX(cl));
4093 		return (rval);
4094 	}
4095 
4096 #if defined(_SUNOS_VTOC_8)
4097 	cmlb_build_user_vtoc(cl, &user_vtoc);
4098 	mutex_exit(CMLB_MUTEX(cl));
4099 
4100 	/*
4101 	 * Checking callers data model does not make much sense here
4102 	 * since extvtoc will always be equivalent to 64bit vtoc.
4103 	 * What is important is whether the kernel is in 32 or 64 bit
4104 	 */
4105 
4106 #ifdef _LP64
4107 		if (ddi_copyout(&user_vtoc, (void *)arg,
4108 		    sizeof (struct extvtoc), flag)) {
4109 			return (EFAULT);
4110 		}
4111 #else
4112 		vtoc32tovtoc(user_vtoc, ext_vtoc);
4113 		if (ddi_copyout(&ext_vtoc, (void *)arg,
4114 		    sizeof (struct extvtoc), flag)) {
4115 			return (EFAULT);
4116 		}
4117 #endif
4118 
4119 #elif defined(_SUNOS_VTOC_16)
4120 	/*
4121 	 * The cl_vtoc structure is a "struct dk_vtoc"  which is always
4122 	 * 32-bit to maintain compatibility with existing on-disk
4123 	 * structures.  Thus, we need to convert the structure when copying
4124 	 * it out to extvtoc
4125 	 */
4126 	vtoc32tovtoc(cl->cl_vtoc, ext_vtoc);
4127 	mutex_exit(CMLB_MUTEX(cl));
4128 
4129 	if (ddi_copyout(&ext_vtoc, (void *)arg, sizeof (struct extvtoc), flag))
4130 		return (EFAULT);
4131 #else
4132 #error "No VTOC format defined."
4133 #endif
4134 
4135 	return (rval);
4136 }
4137 
4138 /*
4139  * This routine implements the DKIOCGETEFI ioctl. This ioctl is currently
4140  * used to read the GPT Partition Table Header (primary/backup), the GUID
4141  * partition Entry Array (primary/backup), and the MBR.
4142  */
4143 static int
4144 cmlb_dkio_get_efi(struct cmlb_lun *cl, caddr_t arg, int flag, void *tg_cookie)
4145 {
4146 	dk_efi_t	user_efi;
4147 	int		rval = 0;
4148 	void		*buffer;
4149 	diskaddr_t	tgt_lba;
4150 
4151 	if (ddi_copyin(arg, &user_efi, sizeof (dk_efi_t), flag))
4152 		return (EFAULT);
4153 
4154 	user_efi.dki_data = (void *)(uintptr_t)user_efi.dki_data_64;
4155 
4156 	if (user_efi.dki_length == 0 ||
4157 	    user_efi.dki_length > cmlb_tg_max_efi_xfer)
4158 		return (EINVAL);
4159 
4160 	tgt_lba = user_efi.dki_lba;
4161 
4162 	mutex_enter(CMLB_MUTEX(cl));
4163 	if ((cmlb_check_update_blockcount(cl, tg_cookie) != 0) ||
4164 	    (cl->cl_tgt_blocksize == 0) ||
4165 	    (user_efi.dki_length % cl->cl_sys_blocksize)) {
4166 		mutex_exit(CMLB_MUTEX(cl));
4167 		return (EINVAL);
4168 	}
4169 	if (cl->cl_tgt_blocksize != cl->cl_sys_blocksize)
4170 		tgt_lba = tgt_lba * cl->cl_tgt_blocksize /
4171 		    cl->cl_sys_blocksize;
4172 	mutex_exit(CMLB_MUTEX(cl));
4173 
4174 	buffer = kmem_alloc(user_efi.dki_length, KM_SLEEP);
4175 	rval = DK_TG_READ(cl, buffer, tgt_lba, user_efi.dki_length, tg_cookie);
4176 	if (rval == 0 && ddi_copyout(buffer, user_efi.dki_data,
4177 	    user_efi.dki_length, flag) != 0)
4178 		rval = EFAULT;
4179 
4180 	kmem_free(buffer, user_efi.dki_length);
4181 	return (rval);
4182 }
4183 
4184 #if defined(_SUNOS_VTOC_8)
4185 /*
4186  *    Function: cmlb_build_user_vtoc
4187  *
4188  * Description: This routine populates a pass by reference variable with the
4189  *		current volume table of contents.
4190  *
4191  *   Arguments: cl - driver soft state (unit) structure
4192  *		user_vtoc - pointer to vtoc structure to be populated
4193  */
4194 static void
4195 cmlb_build_user_vtoc(struct cmlb_lun *cl, struct vtoc *user_vtoc)
4196 {
4197 	struct dk_map2		*lpart;
4198 	struct dk_map		*lmap;
4199 	struct partition	*vpart;
4200 	uint32_t		nblks;
4201 	int			i;
4202 
4203 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
4204 
4205 	/*
4206 	 * Return vtoc structure fields in the provided VTOC area, addressed
4207 	 * by *vtoc.
4208 	 */
4209 	bzero(user_vtoc, sizeof (struct vtoc));
4210 	user_vtoc->v_bootinfo[0] = cl->cl_vtoc.v_bootinfo[0];
4211 	user_vtoc->v_bootinfo[1] = cl->cl_vtoc.v_bootinfo[1];
4212 	user_vtoc->v_bootinfo[2] = cl->cl_vtoc.v_bootinfo[2];
4213 	user_vtoc->v_sanity	= VTOC_SANE;
4214 	user_vtoc->v_version	= cl->cl_vtoc.v_version;
4215 	bcopy(cl->cl_vtoc.v_volume, user_vtoc->v_volume, LEN_DKL_VVOL);
4216 	user_vtoc->v_sectorsz = cl->cl_sys_blocksize;
4217 	user_vtoc->v_nparts = cl->cl_vtoc.v_nparts;
4218 
4219 	for (i = 0; i < 10; i++)
4220 		user_vtoc->v_reserved[i] = cl->cl_vtoc.v_reserved[i];
4221 
4222 	/*
4223 	 * Convert partitioning information.
4224 	 *
4225 	 * Note the conversion from starting cylinder number
4226 	 * to starting sector number.
4227 	 */
4228 	lmap = cl->cl_map;
4229 	lpart = (struct dk_map2 *)cl->cl_vtoc.v_part;
4230 	vpart = user_vtoc->v_part;
4231 
4232 	nblks = cl->cl_g.dkg_nsect * cl->cl_g.dkg_nhead;
4233 
4234 	for (i = 0; i < V_NUMPAR; i++) {
4235 		vpart->p_tag	= lpart->p_tag;
4236 		vpart->p_flag	= lpart->p_flag;
4237 		vpart->p_start	= lmap->dkl_cylno * nblks;
4238 		vpart->p_size	= lmap->dkl_nblk;
4239 		lmap++;
4240 		lpart++;
4241 		vpart++;
4242 
4243 		/* (4364927) */
4244 		user_vtoc->timestamp[i] = (time_t)cl->cl_vtoc.v_timestamp[i];
4245 	}
4246 
4247 	bcopy(cl->cl_asciilabel, user_vtoc->v_asciilabel, LEN_DKL_ASCII);
4248 }
4249 #endif
4250 
4251 static int
4252 cmlb_dkio_partition(struct cmlb_lun *cl, caddr_t arg, int flag,
4253     void *tg_cookie)
4254 {
4255 	struct partition64	p64;
4256 	int			rval = 0;
4257 	uint_t			nparts;
4258 	efi_gpe_t		*partitions;
4259 	efi_gpt_t		*buffer;
4260 	diskaddr_t		gpe_lba;
4261 	int			n_gpe_per_blk = 0;
4262 
4263 	if (ddi_copyin((const void *)arg, &p64,
4264 	    sizeof (struct partition64), flag)) {
4265 		return (EFAULT);
4266 	}
4267 
4268 	buffer = kmem_alloc(cl->cl_sys_blocksize, KM_SLEEP);
4269 	rval = DK_TG_READ(cl, buffer, 1, cl->cl_sys_blocksize, tg_cookie);
4270 	if (rval != 0)
4271 		goto done_error;
4272 
4273 	cmlb_swap_efi_gpt(buffer);
4274 
4275 	if ((rval = cmlb_validate_efi(buffer)) != 0)
4276 		goto done_error;
4277 
4278 	nparts = buffer->efi_gpt_NumberOfPartitionEntries;
4279 	gpe_lba = buffer->efi_gpt_PartitionEntryLBA;
4280 	if (p64.p_partno >= nparts) {
4281 		/* couldn't find it */
4282 		rval = ESRCH;
4283 		goto done_error;
4284 	}
4285 	/*
4286 	 * Read the block that contains the requested GPE.
4287 	 */
4288 	n_gpe_per_blk = cl->cl_sys_blocksize / sizeof (efi_gpe_t);
4289 	gpe_lba += p64.p_partno / n_gpe_per_blk;
4290 	rval = DK_TG_READ(cl, buffer, gpe_lba, cl->cl_sys_blocksize, tg_cookie);
4291 
4292 	if (rval) {
4293 		goto done_error;
4294 	}
4295 	partitions = (efi_gpe_t *)buffer;
4296 	partitions += p64.p_partno % n_gpe_per_blk;
4297 
4298 	/* Byte swap only the requested GPE */
4299 	cmlb_swap_efi_gpe(1, partitions);
4300 
4301 	bcopy(&partitions->efi_gpe_PartitionTypeGUID, &p64.p_type,
4302 	    sizeof (struct uuid));
4303 	p64.p_start = partitions->efi_gpe_StartingLBA;
4304 	p64.p_size = partitions->efi_gpe_EndingLBA -
4305 	    p64.p_start + 1;
4306 
4307 	if (ddi_copyout(&p64, (void *)arg, sizeof (struct partition64), flag))
4308 		rval = EFAULT;
4309 
4310 done_error:
4311 	kmem_free(buffer, cl->cl_sys_blocksize);
4312 	return (rval);
4313 }
4314 
4315 
4316 /*
4317  *    Function: cmlb_dkio_set_vtoc
4318  *
4319  * Description: This routine is the driver entry point for handling user
4320  *		requests to set the current volume table of contents
4321  *		(DKIOCSVTOC).
4322  *
4323  *   Arguments:
4324  *	dev		the device number
4325  *	arg		pointer to user provided vtoc structure used to set the
4326  *			current vtoc.
4327  *
4328  *	flag		this argument is a pass through to ddi_copyxxx()
4329  *			directly from the mode argument of ioctl().
4330  *
4331  *	tg_cookie	cookie from target driver to be passed back to target
4332  *			driver when we call back to it through tg_ops.
4333  *
4334  * Return Code: 0
4335  *		EFAULT
4336  *		ENXIO
4337  *		EINVAL
4338  *		ENOTSUP
4339  */
4340 static int
4341 cmlb_dkio_set_vtoc(struct cmlb_lun *cl, dev_t dev, caddr_t arg, int flag,
4342     void *tg_cookie)
4343 {
4344 	struct vtoc	user_vtoc;
4345 	int		shift, rval = 0;
4346 	boolean_t	internal;
4347 
4348 	internal = VOID2BOOLEAN(
4349 	    (cl->cl_alter_behavior & (CMLB_INTERNAL_MINOR_NODES)) != 0);
4350 
4351 	if (cl->cl_alter_behavior & CMLB_CREATE_P0_MINOR_NODE)
4352 		shift = CMLBUNIT_FORCE_P0_SHIFT;
4353 	else
4354 		shift = CMLBUNIT_SHIFT;
4355 
4356 #ifdef _MULTI_DATAMODEL
4357 	switch (ddi_model_convert_from(flag & FMODELS)) {
4358 	case DDI_MODEL_ILP32: {
4359 		struct vtoc32 user_vtoc32;
4360 
4361 		if (ddi_copyin((const void *)arg, &user_vtoc32,
4362 		    sizeof (struct vtoc32), flag)) {
4363 			return (EFAULT);
4364 		}
4365 		vtoc32tovtoc(user_vtoc32, user_vtoc);
4366 		break;
4367 	}
4368 
4369 	case DDI_MODEL_NONE:
4370 		if (ddi_copyin((const void *)arg, &user_vtoc,
4371 		    sizeof (struct vtoc), flag)) {
4372 			return (EFAULT);
4373 		}
4374 		break;
4375 	}
4376 #else /* ! _MULTI_DATAMODEL */
4377 	if (ddi_copyin((const void *)arg, &user_vtoc,
4378 	    sizeof (struct vtoc), flag)) {
4379 		return (EFAULT);
4380 	}
4381 #endif /* _MULTI_DATAMODEL */
4382 
4383 	mutex_enter(CMLB_MUTEX(cl));
4384 
4385 	if (cl->cl_blockcount > CMLB_OLDVTOC_LIMIT) {
4386 		mutex_exit(CMLB_MUTEX(cl));
4387 		return (EOVERFLOW);
4388 	}
4389 
4390 #if defined(__i386) || defined(__amd64)
4391 	if (cl->cl_tgt_blocksize != cl->cl_sys_blocksize) {
4392 		mutex_exit(CMLB_MUTEX(cl));
4393 		return (EINVAL);
4394 	}
4395 #endif
4396 
4397 	if (cl->cl_g.dkg_ncyl == 0) {
4398 		mutex_exit(CMLB_MUTEX(cl));
4399 		return (EINVAL);
4400 	}
4401 
4402 	mutex_exit(CMLB_MUTEX(cl));
4403 	cmlb_clear_efi(cl, tg_cookie);
4404 	ddi_remove_minor_node(CMLB_DEVINFO(cl), "wd");
4405 	ddi_remove_minor_node(CMLB_DEVINFO(cl), "wd,raw");
4406 
4407 	/*
4408 	 * cmlb_dkio_set_vtoc creates duplicate minor nodes when
4409 	 * relabeling an SMI disk. To avoid that we remove them
4410 	 * before creating.
4411 	 * It should be OK to remove a non-existed minor node.
4412 	 */
4413 	ddi_remove_minor_node(CMLB_DEVINFO(cl), "h");
4414 	ddi_remove_minor_node(CMLB_DEVINFO(cl), "h,raw");
4415 
4416 	(void) cmlb_create_minor(CMLB_DEVINFO(cl), "h",
4417 	    S_IFBLK, (CMLBUNIT(dev, shift) << shift) | WD_NODE,
4418 	    cl->cl_node_type, NULL, internal);
4419 	(void) cmlb_create_minor(CMLB_DEVINFO(cl), "h,raw",
4420 	    S_IFCHR, (CMLBUNIT(dev, shift) << shift) | WD_NODE,
4421 	    cl->cl_node_type, NULL, internal);
4422 	mutex_enter(CMLB_MUTEX(cl));
4423 
4424 	if ((rval = cmlb_build_label_vtoc(cl, &user_vtoc)) == 0) {
4425 		if ((rval = cmlb_write_label(cl, tg_cookie)) == 0) {
4426 			if (cmlb_validate_geometry(cl,
4427 			    B_TRUE, 0, tg_cookie) != 0) {
4428 				cmlb_dbg(CMLB_ERROR, cl,
4429 				    "cmlb_dkio_set_vtoc: "
4430 				    "Failed validate geometry\n");
4431 			}
4432 			cl->cl_msglog_flag |= CMLB_ALLOW_2TB_WARN;
4433 		}
4434 	}
4435 	mutex_exit(CMLB_MUTEX(cl));
4436 	return (rval);
4437 }
4438 
4439 /*
4440  *    Function: cmlb_dkio_set_extvtoc
4441  */
4442 static int
4443 cmlb_dkio_set_extvtoc(struct cmlb_lun *cl, dev_t dev, caddr_t arg, int flag,
4444     void *tg_cookie)
4445 {
4446 	int		shift, rval = 0;
4447 	struct vtoc	user_vtoc;
4448 	boolean_t	internal;
4449 
4450 	if (cl->cl_alter_behavior & CMLB_CREATE_P0_MINOR_NODE)
4451 		shift = CMLBUNIT_FORCE_P0_SHIFT;
4452 	else
4453 		shift = CMLBUNIT_SHIFT;
4454 
4455 	/*
4456 	 * Checking callers data model does not make much sense here
4457 	 * since extvtoc will always be equivalent to 64bit vtoc.
4458 	 * What is important is whether the kernel is in 32 or 64 bit
4459 	 */
4460 
4461 #ifdef _LP64
4462 	if (ddi_copyin((const void *)arg, &user_vtoc,
4463 		    sizeof (struct extvtoc), flag)) {
4464 			return (EFAULT);
4465 	}
4466 #else
4467 	struct	extvtoc	user_extvtoc;
4468 	if (ddi_copyin((const void *)arg, &user_extvtoc,
4469 		    sizeof (struct extvtoc), flag)) {
4470 			return (EFAULT);
4471 	}
4472 
4473 	vtoctovtoc32(user_extvtoc, user_vtoc);
4474 #endif
4475 
4476 	internal = VOID2BOOLEAN(
4477 	    (cl->cl_alter_behavior & (CMLB_INTERNAL_MINOR_NODES)) != 0);
4478 	mutex_enter(CMLB_MUTEX(cl));
4479 #if defined(__i386) || defined(__amd64)
4480 	if (cl->cl_tgt_blocksize != cl->cl_sys_blocksize) {
4481 		mutex_exit(CMLB_MUTEX(cl));
4482 		return (EINVAL);
4483 	}
4484 #endif
4485 
4486 	if (cl->cl_g.dkg_ncyl == 0) {
4487 		mutex_exit(CMLB_MUTEX(cl));
4488 		return (EINVAL);
4489 	}
4490 
4491 	mutex_exit(CMLB_MUTEX(cl));
4492 	cmlb_clear_efi(cl, tg_cookie);
4493 	ddi_remove_minor_node(CMLB_DEVINFO(cl), "wd");
4494 	ddi_remove_minor_node(CMLB_DEVINFO(cl), "wd,raw");
4495 	/*
4496 	 * cmlb_dkio_set_extvtoc creates duplicate minor nodes when
4497 	 * relabeling an SMI disk. To avoid that we remove them
4498 	 * before creating.
4499 	 * It should be OK to remove a non-existed minor node.
4500 	 */
4501 	ddi_remove_minor_node(CMLB_DEVINFO(cl), "h");
4502 	ddi_remove_minor_node(CMLB_DEVINFO(cl), "h,raw");
4503 
4504 	(void) cmlb_create_minor(CMLB_DEVINFO(cl), "h",
4505 	    S_IFBLK, (CMLBUNIT(dev, shift) << shift) | WD_NODE,
4506 	    cl->cl_node_type, NULL, internal);
4507 	(void) cmlb_create_minor(CMLB_DEVINFO(cl), "h,raw",
4508 	    S_IFCHR, (CMLBUNIT(dev, shift) << shift) | WD_NODE,
4509 	    cl->cl_node_type, NULL, internal);
4510 
4511 	mutex_enter(CMLB_MUTEX(cl));
4512 
4513 	if ((rval = cmlb_build_label_vtoc(cl, &user_vtoc)) == 0) {
4514 		if ((rval = cmlb_write_label(cl, tg_cookie)) == 0) {
4515 			if (cmlb_validate_geometry(cl,
4516 			    B_TRUE, 0, tg_cookie) != 0) {
4517 				cmlb_dbg(CMLB_ERROR, cl,
4518 				    "cmlb_dkio_set_vtoc: "
4519 				    "Failed validate geometry\n");
4520 			}
4521 		}
4522 	}
4523 	mutex_exit(CMLB_MUTEX(cl));
4524 	return (rval);
4525 }
4526 
4527 /*
4528  *    Function: cmlb_build_label_vtoc
4529  *
4530  * Description: This routine updates the driver soft state current volume table
4531  *		of contents based on a user specified vtoc.
4532  *
4533  *   Arguments: cl - driver soft state (unit) structure
4534  *		user_vtoc - pointer to vtoc structure specifying vtoc to be used
4535  *			    to update the driver soft state.
4536  *
4537  * Return Code: 0
4538  *		EINVAL
4539  */
4540 static int
4541 cmlb_build_label_vtoc(struct cmlb_lun *cl, struct vtoc *user_vtoc)
4542 {
4543 	struct dk_map		*lmap;
4544 	struct partition	*vpart;
4545 	uint_t			nblks;
4546 #if defined(_SUNOS_VTOC_8)
4547 	int			ncyl;
4548 	struct dk_map2		*lpart;
4549 #endif	/* defined(_SUNOS_VTOC_8) */
4550 	int			i;
4551 
4552 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
4553 
4554 	/* Sanity-check the vtoc */
4555 	if (user_vtoc->v_sanity != VTOC_SANE ||
4556 	    user_vtoc->v_sectorsz != cl->cl_sys_blocksize ||
4557 	    user_vtoc->v_nparts != V_NUMPAR) {
4558 		cmlb_dbg(CMLB_INFO,  cl,
4559 		    "cmlb_build_label_vtoc: vtoc not valid\n");
4560 		return (EINVAL);
4561 	}
4562 
4563 	nblks = cl->cl_g.dkg_nsect * cl->cl_g.dkg_nhead;
4564 	if (nblks == 0) {
4565 		cmlb_dbg(CMLB_INFO,  cl,
4566 		    "cmlb_build_label_vtoc: geom nblks is 0\n");
4567 		return (EINVAL);
4568 	}
4569 
4570 #if defined(_SUNOS_VTOC_8)
4571 	vpart = user_vtoc->v_part;
4572 	for (i = 0; i < V_NUMPAR; i++) {
4573 		if (((unsigned)vpart->p_start % nblks) != 0) {
4574 			cmlb_dbg(CMLB_INFO,  cl,
4575 			    "cmlb_build_label_vtoc: p_start not multiply of"
4576 			    "nblks part %d p_start %d nblks %d\n", i,
4577 			    vpart->p_start, nblks);
4578 			return (EINVAL);
4579 		}
4580 		ncyl = (unsigned)vpart->p_start / nblks;
4581 		ncyl += (unsigned)vpart->p_size / nblks;
4582 		if (((unsigned)vpart->p_size % nblks) != 0) {
4583 			ncyl++;
4584 		}
4585 		if (ncyl > (int)cl->cl_g.dkg_ncyl) {
4586 			cmlb_dbg(CMLB_INFO,  cl,
4587 			    "cmlb_build_label_vtoc: ncyl %d  > dkg_ncyl %d"
4588 			    "p_size %ld p_start %ld nblks %d  part number %d"
4589 			    "tag %d\n",
4590 			    ncyl, cl->cl_g.dkg_ncyl, vpart->p_size,
4591 			    vpart->p_start, nblks,
4592 			    i, vpart->p_tag);
4593 
4594 			return (EINVAL);
4595 		}
4596 		vpart++;
4597 	}
4598 #endif	/* defined(_SUNOS_VTOC_8) */
4599 
4600 	/* Put appropriate vtoc structure fields into the disk label */
4601 #if defined(_SUNOS_VTOC_16)
4602 	/*
4603 	 * The vtoc is always a 32bit data structure to maintain the
4604 	 * on-disk format. Convert "in place" instead of doing bcopy.
4605 	 */
4606 	vtoctovtoc32((*user_vtoc), (*((struct vtoc32 *)&(cl->cl_vtoc))));
4607 
4608 	/*
4609 	 * in the 16-slice vtoc, starting sectors are expressed in
4610 	 * numbers *relative* to the start of the Solaris fdisk partition.
4611 	 */
4612 	lmap = cl->cl_map;
4613 	vpart = user_vtoc->v_part;
4614 
4615 	for (i = 0; i < (int)user_vtoc->v_nparts; i++, lmap++, vpart++) {
4616 		lmap->dkl_cylno = (unsigned)vpart->p_start / nblks;
4617 		lmap->dkl_nblk = (unsigned)vpart->p_size;
4618 	}
4619 
4620 #elif defined(_SUNOS_VTOC_8)
4621 
4622 	cl->cl_vtoc.v_bootinfo[0] = (uint32_t)user_vtoc->v_bootinfo[0];
4623 	cl->cl_vtoc.v_bootinfo[1] = (uint32_t)user_vtoc->v_bootinfo[1];
4624 	cl->cl_vtoc.v_bootinfo[2] = (uint32_t)user_vtoc->v_bootinfo[2];
4625 
4626 	cl->cl_vtoc.v_sanity = (uint32_t)user_vtoc->v_sanity;
4627 	cl->cl_vtoc.v_version = (uint32_t)user_vtoc->v_version;
4628 
4629 	bcopy(user_vtoc->v_volume, cl->cl_vtoc.v_volume, LEN_DKL_VVOL);
4630 
4631 	cl->cl_vtoc.v_nparts = user_vtoc->v_nparts;
4632 
4633 	for (i = 0; i < 10; i++)
4634 		cl->cl_vtoc.v_reserved[i] =  user_vtoc->v_reserved[i];
4635 
4636 	/*
4637 	 * Note the conversion from starting sector number
4638 	 * to starting cylinder number.
4639 	 * Return error if division results in a remainder.
4640 	 */
4641 	lmap = cl->cl_map;
4642 	lpart = cl->cl_vtoc.v_part;
4643 	vpart = user_vtoc->v_part;
4644 
4645 	for (i = 0; i < (int)user_vtoc->v_nparts; i++) {
4646 		lpart->p_tag  = vpart->p_tag;
4647 		lpart->p_flag = vpart->p_flag;
4648 		lmap->dkl_cylno = (unsigned)vpart->p_start / nblks;
4649 		lmap->dkl_nblk = (unsigned)vpart->p_size;
4650 
4651 		lmap++;
4652 		lpart++;
4653 		vpart++;
4654 
4655 		/* (4387723) */
4656 #ifdef _LP64
4657 		if (user_vtoc->timestamp[i] > TIME32_MAX) {
4658 			cl->cl_vtoc.v_timestamp[i] = TIME32_MAX;
4659 		} else {
4660 			cl->cl_vtoc.v_timestamp[i] = user_vtoc->timestamp[i];
4661 		}
4662 #else
4663 		cl->cl_vtoc.v_timestamp[i] = user_vtoc->timestamp[i];
4664 #endif
4665 	}
4666 
4667 	bcopy(user_vtoc->v_asciilabel, cl->cl_asciilabel, LEN_DKL_ASCII);
4668 #else
4669 #error "No VTOC format defined."
4670 #endif
4671 	return (0);
4672 }
4673 
4674 /*
4675  *    Function: cmlb_clear_efi
4676  *
4677  * Description: This routine clears all EFI labels.
4678  *
4679  *   Arguments:
4680  *	cl		 driver soft state (unit) structure
4681  *
4682  *	tg_cookie	cookie from target driver to be passed back to target
4683  *			driver when we call back to it through tg_ops.
4684  * Return Code: void
4685  */
4686 static void
4687 cmlb_clear_efi(struct cmlb_lun *cl, void *tg_cookie)
4688 {
4689 	efi_gpt_t	*gpt;
4690 	diskaddr_t	cap;
4691 	int		rval;
4692 
4693 	ASSERT(!mutex_owned(CMLB_MUTEX(cl)));
4694 
4695 	mutex_enter(CMLB_MUTEX(cl));
4696 	cl->cl_reserved = -1;
4697 	mutex_exit(CMLB_MUTEX(cl));
4698 
4699 	gpt = kmem_alloc(cl->cl_sys_blocksize, KM_SLEEP);
4700 
4701 	if (DK_TG_READ(cl, gpt, 1, cl->cl_sys_blocksize, tg_cookie) != 0) {
4702 		goto done;
4703 	}
4704 
4705 	cmlb_swap_efi_gpt(gpt);
4706 	rval = cmlb_validate_efi(gpt);
4707 	if (rval == 0) {
4708 		/* clear primary */
4709 		bzero(gpt, sizeof (efi_gpt_t));
4710 		if (rval = DK_TG_WRITE(cl, gpt, 1, cl->cl_sys_blocksize,
4711 		    tg_cookie)) {
4712 			cmlb_dbg(CMLB_INFO,  cl,
4713 			    "cmlb_clear_efi: clear primary label failed\n");
4714 		}
4715 	}
4716 	/* the backup */
4717 	rval = DK_TG_GETCAP(cl, &cap, tg_cookie);
4718 	if (rval) {
4719 		goto done;
4720 	}
4721 
4722 	if ((rval = DK_TG_READ(cl, gpt, cap - 1, cl->cl_sys_blocksize,
4723 	    tg_cookie)) != 0) {
4724 		goto done;
4725 	}
4726 	cmlb_swap_efi_gpt(gpt);
4727 	rval = cmlb_validate_efi(gpt);
4728 	if (rval == 0) {
4729 		/* clear backup */
4730 		cmlb_dbg(CMLB_TRACE,  cl,
4731 		    "cmlb_clear_efi clear backup@%lu\n", cap - 1);
4732 		bzero(gpt, sizeof (efi_gpt_t));
4733 		if ((rval = DK_TG_WRITE(cl,  gpt, cap - 1, cl->cl_sys_blocksize,
4734 		    tg_cookie))) {
4735 			cmlb_dbg(CMLB_INFO,  cl,
4736 			    "cmlb_clear_efi: clear backup label failed\n");
4737 		}
4738 	} else {
4739 		/*
4740 		 * Refer to comments related to off-by-1 at the
4741 		 * header of this file
4742 		 */
4743 		if ((rval = DK_TG_READ(cl, gpt, cap - 2,
4744 		    cl->cl_sys_blocksize, tg_cookie)) != 0) {
4745 			goto done;
4746 		}
4747 		cmlb_swap_efi_gpt(gpt);
4748 		rval = cmlb_validate_efi(gpt);
4749 		if (rval == 0) {
4750 			/* clear legacy backup EFI label */
4751 			cmlb_dbg(CMLB_TRACE,  cl,
4752 			    "cmlb_clear_efi clear legacy backup@%lu\n",
4753 			    cap - 2);
4754 			bzero(gpt, sizeof (efi_gpt_t));
4755 			if ((rval = DK_TG_WRITE(cl,  gpt, cap - 2,
4756 			    cl->cl_sys_blocksize, tg_cookie))) {
4757 				cmlb_dbg(CMLB_INFO,  cl,
4758 				"cmlb_clear_efi: clear legacy backup label "
4759 				"failed\n");
4760 			}
4761 		}
4762 	}
4763 
4764 done:
4765 	kmem_free(gpt, cl->cl_sys_blocksize);
4766 }
4767 
4768 /*
4769  *    Function: cmlb_set_vtoc
4770  *
4771  * Description: This routine writes data to the appropriate positions
4772  *
4773  *   Arguments:
4774  *	cl		driver soft state (unit) structure
4775  *
4776  *	dkl		the data to be written
4777  *
4778  *	tg_cookie	cookie from target driver to be passed back to target
4779  *			driver when we call back to it through tg_ops.
4780  *
4781  * Return: void
4782  */
4783 static int
4784 cmlb_set_vtoc(struct cmlb_lun *cl, struct dk_label *dkl, void *tg_cookie)
4785 {
4786 	uint_t	label_addr;
4787 	int	sec;
4788 	diskaddr_t	blk;
4789 	int	head;
4790 	int	cyl;
4791 	int	rval;
4792 
4793 #if defined(__i386) || defined(__amd64)
4794 	label_addr = cl->cl_solaris_offset + DK_LABEL_LOC;
4795 #else
4796 	/* Write the primary label at block 0 of the solaris partition. */
4797 	label_addr = 0;
4798 #endif
4799 
4800 	rval = DK_TG_WRITE(cl, dkl, label_addr, cl->cl_sys_blocksize,
4801 	    tg_cookie);
4802 
4803 	if (rval != 0) {
4804 		return (rval);
4805 	}
4806 
4807 	/*
4808 	 * Calculate where the backup labels go.  They are always on
4809 	 * the last alternate cylinder, but some older drives put them
4810 	 * on head 2 instead of the last head.	They are always on the
4811 	 * first 5 odd sectors of the appropriate track.
4812 	 *
4813 	 * We have no choice at this point, but to believe that the
4814 	 * disk label is valid.	 Use the geometry of the disk
4815 	 * as described in the label.
4816 	 */
4817 	cyl  = dkl->dkl_ncyl  + dkl->dkl_acyl - 1;
4818 	head = dkl->dkl_nhead - 1;
4819 
4820 	/*
4821 	 * Write and verify the backup labels. Make sure we don't try to
4822 	 * write past the last cylinder.
4823 	 */
4824 	for (sec = 1; ((sec < 5 * 2 + 1) && (sec < dkl->dkl_nsect)); sec += 2) {
4825 		blk = (diskaddr_t)(
4826 		    (cyl * ((dkl->dkl_nhead * dkl->dkl_nsect) - dkl->dkl_apc)) +
4827 		    (head * dkl->dkl_nsect) + sec);
4828 #if defined(__i386) || defined(__amd64)
4829 		blk += cl->cl_solaris_offset;
4830 #endif
4831 		rval = DK_TG_WRITE(cl, dkl, blk, cl->cl_sys_blocksize,
4832 		    tg_cookie);
4833 		cmlb_dbg(CMLB_INFO,  cl,
4834 		"cmlb_set_vtoc: wrote backup label %llx\n", blk);
4835 		if (rval != 0) {
4836 			goto exit;
4837 		}
4838 	}
4839 exit:
4840 	return (rval);
4841 }
4842 
4843 /*
4844  *    Function: cmlb_clear_vtoc
4845  *
4846  * Description: This routine clears out the VTOC labels.
4847  *
4848  *   Arguments:
4849  *	cl		driver soft state (unit) structure
4850  *
4851  *	tg_cookie	cookie from target driver to be passed back to target
4852  *			driver when we call back to it through tg_ops.
4853  *
4854  * Return: void
4855  */
4856 static void
4857 cmlb_clear_vtoc(struct cmlb_lun *cl, void *tg_cookie)
4858 {
4859 	struct dk_label		*dkl;
4860 
4861 	mutex_exit(CMLB_MUTEX(cl));
4862 	dkl = kmem_zalloc(cl->cl_sys_blocksize, KM_SLEEP);
4863 	mutex_enter(CMLB_MUTEX(cl));
4864 	/*
4865 	 * cmlb_set_vtoc uses these fields in order to figure out
4866 	 * where to overwrite the backup labels
4867 	 */
4868 	dkl->dkl_apc    = cl->cl_g.dkg_apc;
4869 	dkl->dkl_ncyl   = cl->cl_g.dkg_ncyl;
4870 	dkl->dkl_acyl   = cl->cl_g.dkg_acyl;
4871 	dkl->dkl_nhead  = cl->cl_g.dkg_nhead;
4872 	dkl->dkl_nsect  = cl->cl_g.dkg_nsect;
4873 	mutex_exit(CMLB_MUTEX(cl));
4874 	(void) cmlb_set_vtoc(cl, dkl, tg_cookie);
4875 	kmem_free(dkl, cl->cl_sys_blocksize);
4876 
4877 	mutex_enter(CMLB_MUTEX(cl));
4878 }
4879 
4880 /*
4881  *    Function: cmlb_write_label
4882  *
4883  * Description: This routine will validate and write the driver soft state vtoc
4884  *		contents to the device.
4885  *
4886  *   Arguments:
4887  *	cl		cmlb handle
4888  *
4889  *	tg_cookie	cookie from target driver to be passed back to target
4890  *			driver when we call back to it through tg_ops.
4891  *
4892  *
4893  * Return Code: the code returned by cmlb_send_scsi_cmd()
4894  *		0
4895  *		EINVAL
4896  *		ENXIO
4897  *		ENOMEM
4898  */
4899 static int
4900 cmlb_write_label(struct cmlb_lun *cl, void *tg_cookie)
4901 {
4902 	struct dk_label	*dkl;
4903 	short		sum;
4904 	short		*sp;
4905 	int		i;
4906 	int		rval;
4907 
4908 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
4909 	mutex_exit(CMLB_MUTEX(cl));
4910 	dkl = kmem_zalloc(cl->cl_sys_blocksize, KM_SLEEP);
4911 	mutex_enter(CMLB_MUTEX(cl));
4912 
4913 	bcopy(&cl->cl_vtoc, &dkl->dkl_vtoc, sizeof (struct dk_vtoc));
4914 	dkl->dkl_rpm	= cl->cl_g.dkg_rpm;
4915 	dkl->dkl_pcyl	= cl->cl_g.dkg_pcyl;
4916 	dkl->dkl_apc	= cl->cl_g.dkg_apc;
4917 	dkl->dkl_intrlv = cl->cl_g.dkg_intrlv;
4918 	dkl->dkl_ncyl	= cl->cl_g.dkg_ncyl;
4919 	dkl->dkl_acyl	= cl->cl_g.dkg_acyl;
4920 	dkl->dkl_nhead	= cl->cl_g.dkg_nhead;
4921 	dkl->dkl_nsect	= cl->cl_g.dkg_nsect;
4922 
4923 #if defined(_SUNOS_VTOC_8)
4924 	dkl->dkl_obs1	= cl->cl_g.dkg_obs1;
4925 	dkl->dkl_obs2	= cl->cl_g.dkg_obs2;
4926 	dkl->dkl_obs3	= cl->cl_g.dkg_obs3;
4927 	for (i = 0; i < NDKMAP; i++) {
4928 		dkl->dkl_map[i].dkl_cylno = cl->cl_map[i].dkl_cylno;
4929 		dkl->dkl_map[i].dkl_nblk  = cl->cl_map[i].dkl_nblk;
4930 	}
4931 	bcopy(cl->cl_asciilabel, dkl->dkl_asciilabel, LEN_DKL_ASCII);
4932 #elif defined(_SUNOS_VTOC_16)
4933 	dkl->dkl_skew	= cl->cl_dkg_skew;
4934 #else
4935 #error "No VTOC format defined."
4936 #endif
4937 
4938 	dkl->dkl_magic			= DKL_MAGIC;
4939 	dkl->dkl_write_reinstruct	= cl->cl_g.dkg_write_reinstruct;
4940 	dkl->dkl_read_reinstruct	= cl->cl_g.dkg_read_reinstruct;
4941 
4942 	/* Construct checksum for the new disk label */
4943 	sum = 0;
4944 	sp = (short *)dkl;
4945 	i = sizeof (struct dk_label) / sizeof (short);
4946 	while (i--) {
4947 		sum ^= *sp++;
4948 	}
4949 	dkl->dkl_cksum = sum;
4950 
4951 	mutex_exit(CMLB_MUTEX(cl));
4952 
4953 	rval = cmlb_set_vtoc(cl, dkl, tg_cookie);
4954 exit:
4955 	kmem_free(dkl, cl->cl_sys_blocksize);
4956 	mutex_enter(CMLB_MUTEX(cl));
4957 	return (rval);
4958 }
4959 
4960 /*
4961  * This routine implements the DKIOCSETEFI ioctl. This ioctl is currently
4962  * used to write (or clear) the GPT Partition Table header (primary/backup)
4963  * and GUID partition Entry Array (primary/backup). It is also used to write
4964  * the Protective MBR.
4965  */
4966 static int
4967 cmlb_dkio_set_efi(struct cmlb_lun *cl, dev_t dev, caddr_t arg, int flag,
4968     void *tg_cookie)
4969 {
4970 	dk_efi_t	user_efi;
4971 	int		shift, rval = 0;
4972 	void		*buffer;
4973 	diskaddr_t	tgt_lba;
4974 	boolean_t	internal;
4975 
4976 	if (ddi_copyin(arg, &user_efi, sizeof (dk_efi_t), flag))
4977 		return (EFAULT);
4978 
4979 	internal = VOID2BOOLEAN(
4980 	    (cl->cl_alter_behavior & (CMLB_INTERNAL_MINOR_NODES)) != 0);
4981 
4982 	if (cl->cl_alter_behavior & CMLB_CREATE_P0_MINOR_NODE)
4983 		shift = CMLBUNIT_FORCE_P0_SHIFT;
4984 	else
4985 		shift = CMLBUNIT_SHIFT;
4986 
4987 	user_efi.dki_data = (void *)(uintptr_t)user_efi.dki_data_64;
4988 
4989 	if (user_efi.dki_length == 0 ||
4990 	    user_efi.dki_length > cmlb_tg_max_efi_xfer)
4991 		return (EINVAL);
4992 
4993 	tgt_lba = user_efi.dki_lba;
4994 
4995 	mutex_enter(CMLB_MUTEX(cl));
4996 	if ((cmlb_check_update_blockcount(cl, tg_cookie) != 0) ||
4997 	    (cl->cl_tgt_blocksize == 0) ||
4998 	    (user_efi.dki_length % cl->cl_sys_blocksize)) {
4999 		mutex_exit(CMLB_MUTEX(cl));
5000 		return (EINVAL);
5001 	}
5002 	if (cl->cl_tgt_blocksize != cl->cl_sys_blocksize)
5003 		tgt_lba = tgt_lba *
5004 		    cl->cl_tgt_blocksize / cl->cl_sys_blocksize;
5005 	mutex_exit(CMLB_MUTEX(cl));
5006 
5007 	buffer = kmem_alloc(user_efi.dki_length, KM_SLEEP);
5008 	if (ddi_copyin(user_efi.dki_data, buffer, user_efi.dki_length, flag)) {
5009 		rval = EFAULT;
5010 	} else {
5011 		/*
5012 		 * let's clear the vtoc labels and clear the softstate
5013 		 * vtoc.
5014 		 */
5015 		mutex_enter(CMLB_MUTEX(cl));
5016 		if (cl->cl_vtoc.v_sanity == VTOC_SANE) {
5017 			cmlb_dbg(CMLB_TRACE,  cl,
5018 			    "cmlb_dkio_set_efi: CLEAR VTOC\n");
5019 			if (cl->cl_label_from_media == CMLB_LABEL_VTOC)
5020 				cmlb_clear_vtoc(cl, tg_cookie);
5021 			bzero(&cl->cl_vtoc, sizeof (struct dk_vtoc));
5022 			mutex_exit(CMLB_MUTEX(cl));
5023 			ddi_remove_minor_node(CMLB_DEVINFO(cl), "h");
5024 			ddi_remove_minor_node(CMLB_DEVINFO(cl), "h,raw");
5025 			(void) cmlb_create_minor(CMLB_DEVINFO(cl), "wd",
5026 			    S_IFBLK,
5027 			    (CMLBUNIT(dev, shift) << shift) | WD_NODE,
5028 			    cl->cl_node_type, NULL, internal);
5029 			(void) cmlb_create_minor(CMLB_DEVINFO(cl), "wd,raw",
5030 			    S_IFCHR,
5031 			    (CMLBUNIT(dev, shift) << shift) | WD_NODE,
5032 			    cl->cl_node_type, NULL, internal);
5033 		} else
5034 			mutex_exit(CMLB_MUTEX(cl));
5035 
5036 		rval = DK_TG_WRITE(cl, buffer, tgt_lba, user_efi.dki_length,
5037 		    tg_cookie);
5038 
5039 		if (rval == 0) {
5040 			mutex_enter(CMLB_MUTEX(cl));
5041 			cl->cl_f_geometry_is_valid = B_FALSE;
5042 			mutex_exit(CMLB_MUTEX(cl));
5043 		}
5044 	}
5045 	kmem_free(buffer, user_efi.dki_length);
5046 	return (rval);
5047 }
5048 
5049 /*
5050  *    Function: cmlb_dkio_get_mboot
5051  *
5052  * Description: This routine is the driver entry point for handling user
5053  *		requests to get the current device mboot (DKIOCGMBOOT)
5054  *
5055  *   Arguments:
5056  *	arg		pointer to user provided mboot structure specifying
5057  *			the current mboot.
5058  *
5059  *	flag		this argument is a pass through to ddi_copyxxx()
5060  *			directly from the mode argument of ioctl().
5061  *
5062  *	tg_cookie	cookie from target driver to be passed back to target
5063  *			driver when we call back to it through tg_ops.
5064  *
5065  * Return Code: 0
5066  *		EINVAL
5067  *		EFAULT
5068  *		ENXIO
5069  */
5070 static int
5071 cmlb_dkio_get_mboot(struct cmlb_lun *cl, caddr_t arg, int flag, void *tg_cookie)
5072 {
5073 	struct mboot	*mboot;
5074 	int		rval;
5075 	size_t		buffer_size;
5076 
5077 
5078 #if defined(_SUNOS_VTOC_8)
5079 	if ((!ISREMOVABLE(cl) && !ISHOTPLUGGABLE(cl)) || (arg == NULL)) {
5080 #elif defined(_SUNOS_VTOC_16)
5081 	if (arg == NULL) {
5082 #endif
5083 		return (EINVAL);
5084 	}
5085 
5086 	/*
5087 	 * Read the mboot block, located at absolute block 0 on the target.
5088 	 */
5089 	buffer_size = cl->cl_sys_blocksize;
5090 
5091 	cmlb_dbg(CMLB_TRACE,  cl,
5092 	    "cmlb_dkio_get_mboot: allocation size: 0x%x\n", buffer_size);
5093 
5094 	mboot = kmem_zalloc(buffer_size, KM_SLEEP);
5095 	if ((rval = DK_TG_READ(cl, mboot, 0, buffer_size, tg_cookie)) == 0) {
5096 		if (ddi_copyout(mboot, (void *)arg,
5097 		    sizeof (struct mboot), flag) != 0) {
5098 			rval = EFAULT;
5099 		}
5100 	}
5101 	kmem_free(mboot, buffer_size);
5102 	return (rval);
5103 }
5104 
5105 
5106 /*
5107  *    Function: cmlb_dkio_set_mboot
5108  *
5109  * Description: This routine is the driver entry point for handling user
5110  *		requests to validate and set the device master boot
5111  *		(DKIOCSMBOOT).
5112  *
5113  *   Arguments:
5114  *	arg		pointer to user provided mboot structure used to set the
5115  *			master boot.
5116  *
5117  *	flag		this argument is a pass through to ddi_copyxxx()
5118  *			directly from the mode argument of ioctl().
5119  *
5120  *	tg_cookie	cookie from target driver to be passed back to target
5121  *			driver when we call back to it through tg_ops.
5122  *
5123  * Return Code: 0
5124  *		EINVAL
5125  *		EFAULT
5126  *		ENXIO
5127  */
5128 static int
5129 cmlb_dkio_set_mboot(struct cmlb_lun *cl, caddr_t arg, int flag, void *tg_cookie)
5130 {
5131 	struct mboot	*mboot = NULL;
5132 	int		rval;
5133 	ushort_t	magic;
5134 
5135 
5136 	ASSERT(!mutex_owned(CMLB_MUTEX(cl)));
5137 
5138 #if defined(_SUNOS_VTOC_8)
5139 	if (!ISREMOVABLE(cl) && !ISHOTPLUGGABLE(cl)) {
5140 		return (EINVAL);
5141 	}
5142 #endif
5143 
5144 	if (arg == NULL) {
5145 		return (EINVAL);
5146 	}
5147 
5148 	mboot = kmem_zalloc(cl->cl_sys_blocksize, KM_SLEEP);
5149 
5150 	if (ddi_copyin((const void *)arg, mboot,
5151 	    cl->cl_sys_blocksize, flag) != 0) {
5152 		kmem_free(mboot, cl->cl_sys_blocksize);
5153 		return (EFAULT);
5154 	}
5155 
5156 	/* Is this really a master boot record? */
5157 	magic = LE_16(mboot->signature);
5158 	if (magic != MBB_MAGIC) {
5159 		kmem_free(mboot, cl->cl_sys_blocksize);
5160 		return (EINVAL);
5161 	}
5162 
5163 	rval = DK_TG_WRITE(cl, mboot, 0, cl->cl_sys_blocksize, tg_cookie);
5164 
5165 	mutex_enter(CMLB_MUTEX(cl));
5166 #if defined(__i386) || defined(__amd64)
5167 	if (rval == 0) {
5168 		/*
5169 		 * mboot has been written successfully.
5170 		 * update the fdisk and vtoc tables in memory
5171 		 */
5172 		rval = cmlb_update_fdisk_and_vtoc(cl, tg_cookie);
5173 		if ((!cl->cl_f_geometry_is_valid) || (rval != 0)) {
5174 			mutex_exit(CMLB_MUTEX(cl));
5175 			kmem_free(mboot, cl->cl_sys_blocksize);
5176 			return (rval);
5177 		}
5178 	}
5179 
5180 #ifdef __lock_lint
5181 	cmlb_setup_default_geometry(cl, tg_cookie);
5182 #endif
5183 
5184 #else
5185 	if (rval == 0) {
5186 		/*
5187 		 * mboot has been written successfully.
5188 		 * set up the default geometry and VTOC
5189 		 */
5190 		if (cl->cl_blockcount <= CMLB_EXTVTOC_LIMIT)
5191 			cmlb_setup_default_geometry(cl, tg_cookie);
5192 	}
5193 #endif
5194 	cl->cl_msglog_flag |= CMLB_ALLOW_2TB_WARN;
5195 	mutex_exit(CMLB_MUTEX(cl));
5196 	kmem_free(mboot, cl->cl_sys_blocksize);
5197 	return (rval);
5198 }
5199 
5200 
5201 #if defined(__i386) || defined(__amd64)
5202 /*ARGSUSED*/
5203 static int
5204 cmlb_dkio_set_ext_part(struct cmlb_lun *cl, caddr_t arg, int flag,
5205     void *tg_cookie)
5206 {
5207 	int fdisk_rval;
5208 	diskaddr_t capacity;
5209 
5210 	ASSERT(!mutex_owned(CMLB_MUTEX(cl)));
5211 
5212 	mutex_enter(CMLB_MUTEX(cl));
5213 	capacity = cl->cl_blockcount;
5214 	fdisk_rval = cmlb_read_fdisk(cl, capacity, tg_cookie);
5215 	if (fdisk_rval != 0) {
5216 		mutex_exit(CMLB_MUTEX(cl));
5217 		return (fdisk_rval);
5218 	}
5219 
5220 	mutex_exit(CMLB_MUTEX(cl));
5221 	return (fdisk_rval);
5222 }
5223 #endif
5224 
5225 /*
5226  *    Function: cmlb_setup_default_geometry
5227  *
5228  * Description: This local utility routine sets the default geometry as part of
5229  *		setting the device mboot.
5230  *
5231  *   Arguments:
5232  *	cl		driver soft state (unit) structure
5233  *
5234  *	tg_cookie	cookie from target driver to be passed back to target
5235  *			driver when we call back to it through tg_ops.
5236  *
5237  *
5238  * Note: This may be redundant with cmlb_build_default_label.
5239  */
5240 static void
5241 cmlb_setup_default_geometry(struct cmlb_lun *cl, void *tg_cookie)
5242 {
5243 	struct cmlb_geom	pgeom;
5244 	struct cmlb_geom	*pgeomp = &pgeom;
5245 	int			ret;
5246 	int			geom_base_cap = 1;
5247 
5248 
5249 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
5250 
5251 	/* zero out the soft state geometry and partition table. */
5252 	bzero(&cl->cl_g, sizeof (struct dk_geom));
5253 	bzero(&cl->cl_vtoc, sizeof (struct dk_vtoc));
5254 	bzero(cl->cl_map, NDKMAP * (sizeof (struct dk_map)));
5255 
5256 	/*
5257 	 * For the rpm, we use the minimum for the disk.
5258 	 * For the head, cyl and number of sector per track,
5259 	 * if the capacity <= 1GB, head = 64, sect = 32.
5260 	 * else head = 255, sect 63
5261 	 * Note: the capacity should be equal to C*H*S values.
5262 	 * This will cause some truncation of size due to
5263 	 * round off errors. For CD-ROMs, this truncation can
5264 	 * have adverse side effects, so returning ncyl and
5265 	 * nhead as 1. The nsect will overflow for most of
5266 	 * CD-ROMs as nsect is of type ushort.
5267 	 */
5268 	if (cl->cl_alter_behavior & CMLB_FAKE_GEOM_LABEL_IOCTLS_VTOC8) {
5269 		/*
5270 		 * newfs currently can not handle 255 ntracks for SPARC
5271 		 * so get the geometry from target driver instead of coming up
5272 		 * with one based on capacity.
5273 		 */
5274 		mutex_exit(CMLB_MUTEX(cl));
5275 		ret = DK_TG_GETPHYGEOM(cl, pgeomp, tg_cookie);
5276 		mutex_enter(CMLB_MUTEX(cl));
5277 
5278 		if (ret == 0) {
5279 			geom_base_cap = 0;
5280 		} else {
5281 			cmlb_dbg(CMLB_ERROR,  cl,
5282 			    "cmlb_setup_default_geometry: "
5283 			    "tg_getphygeom failed %d\n", ret);
5284 
5285 			/* do default setting, geometry based on capacity */
5286 		}
5287 	}
5288 
5289 	if (geom_base_cap) {
5290 		if (ISCD(cl)) {
5291 			cl->cl_g.dkg_ncyl = 1;
5292 			cl->cl_g.dkg_nhead = 1;
5293 			cl->cl_g.dkg_nsect = cl->cl_blockcount;
5294 		} else if (cl->cl_blockcount < 160) {
5295 			/* Less than 80K */
5296 			cl->cl_g.dkg_nhead = 1;
5297 			cl->cl_g.dkg_ncyl = cl->cl_blockcount;
5298 			cl->cl_g.dkg_nsect = 1;
5299 		} else if (cl->cl_blockcount <= 0x1000) {
5300 			/* Needed for unlabeled SCSI floppies. */
5301 			cl->cl_g.dkg_nhead = 2;
5302 			cl->cl_g.dkg_ncyl = 80;
5303 			cl->cl_g.dkg_pcyl = 80;
5304 			cl->cl_g.dkg_nsect = cl->cl_blockcount / (2 * 80);
5305 		} else if (cl->cl_blockcount <= 0x200000) {
5306 			cl->cl_g.dkg_nhead = 64;
5307 			cl->cl_g.dkg_nsect = 32;
5308 			cl->cl_g.dkg_ncyl = cl->cl_blockcount / (64 * 32);
5309 		} else {
5310 			cl->cl_g.dkg_nhead = 255;
5311 
5312 			cl->cl_g.dkg_nsect = ((cl->cl_blockcount +
5313 			    (UINT16_MAX * 255 * 63) - 1) /
5314 			    (UINT16_MAX * 255 * 63)) * 63;
5315 
5316 			if (cl->cl_g.dkg_nsect == 0)
5317 				cl->cl_g.dkg_nsect = (UINT16_MAX / 63) * 63;
5318 
5319 			cl->cl_g.dkg_ncyl = cl->cl_blockcount /
5320 			    (255 * cl->cl_g.dkg_nsect);
5321 		}
5322 
5323 		cl->cl_g.dkg_acyl = 0;
5324 		cl->cl_g.dkg_bcyl = 0;
5325 		cl->cl_g.dkg_intrlv = 1;
5326 		cl->cl_g.dkg_rpm = 200;
5327 		if (cl->cl_g.dkg_pcyl == 0)
5328 			cl->cl_g.dkg_pcyl = cl->cl_g.dkg_ncyl +
5329 			    cl->cl_g.dkg_acyl;
5330 	} else {
5331 		cl->cl_g.dkg_ncyl = (short)pgeomp->g_ncyl;
5332 		cl->cl_g.dkg_acyl = pgeomp->g_acyl;
5333 		cl->cl_g.dkg_nhead = pgeomp->g_nhead;
5334 		cl->cl_g.dkg_nsect = pgeomp->g_nsect;
5335 		cl->cl_g.dkg_intrlv = pgeomp->g_intrlv;
5336 		cl->cl_g.dkg_rpm = pgeomp->g_rpm;
5337 		cl->cl_g.dkg_pcyl = cl->cl_g.dkg_ncyl + cl->cl_g.dkg_acyl;
5338 	}
5339 
5340 	cl->cl_g.dkg_read_reinstruct = 0;
5341 	cl->cl_g.dkg_write_reinstruct = 0;
5342 	cl->cl_solaris_size = cl->cl_g.dkg_ncyl *
5343 	    cl->cl_g.dkg_nhead * cl->cl_g.dkg_nsect;
5344 
5345 	cl->cl_map['a'-'a'].dkl_cylno = 0;
5346 	cl->cl_map['a'-'a'].dkl_nblk = cl->cl_solaris_size;
5347 
5348 	cl->cl_map['c'-'a'].dkl_cylno = 0;
5349 	cl->cl_map['c'-'a'].dkl_nblk = cl->cl_solaris_size;
5350 
5351 	cl->cl_vtoc.v_part[2].p_tag   = V_BACKUP;
5352 	cl->cl_vtoc.v_part[2].p_flag  = V_UNMNT;
5353 	cl->cl_vtoc.v_nparts = V_NUMPAR;
5354 	cl->cl_vtoc.v_version = V_VERSION;
5355 	(void) sprintf((char *)cl->cl_asciilabel, "DEFAULT cyl %d alt %d"
5356 	    " hd %d sec %d", cl->cl_g.dkg_ncyl, cl->cl_g.dkg_acyl,
5357 	    cl->cl_g.dkg_nhead, cl->cl_g.dkg_nsect);
5358 
5359 	cl->cl_f_geometry_is_valid = B_FALSE;
5360 }
5361 
5362 
5363 #if defined(__i386) || defined(__amd64)
5364 /*
5365  *    Function: cmlb_update_fdisk_and_vtoc
5366  *
5367  * Description: This local utility routine updates the device fdisk and vtoc
5368  *		as part of setting the device mboot.
5369  *
5370  *   Arguments:
5371  *	cl		driver soft state (unit) structure
5372  *
5373  *	tg_cookie	cookie from target driver to be passed back to target
5374  *			driver when we call back to it through tg_ops.
5375  *
5376  *
5377  * Return Code: 0 for success or errno-type return code.
5378  *
5379  *    Note:x86: This looks like a duplicate of cmlb_validate_geometry(), but
5380  *		these did exist separately in x86 sd.c.
5381  */
5382 static int
5383 cmlb_update_fdisk_and_vtoc(struct cmlb_lun *cl, void *tg_cookie)
5384 {
5385 	int		count;
5386 	int		label_rc = 0;
5387 	int		fdisk_rval;
5388 	diskaddr_t	capacity;
5389 
5390 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
5391 
5392 	if (cmlb_check_update_blockcount(cl, tg_cookie) != 0)
5393 		return (EINVAL);
5394 
5395 	/*
5396 	 * Set up the "whole disk" fdisk partition; this should always
5397 	 * exist, regardless of whether the disk contains an fdisk table
5398 	 * or vtoc.
5399 	 */
5400 	cl->cl_map[P0_RAW_DISK].dkl_cylno = 0;
5401 	cl->cl_map[P0_RAW_DISK].dkl_nblk = cl->cl_blockcount;
5402 
5403 	/*
5404 	 * copy the lbasize and capacity so that if they're
5405 	 * reset while we're not holding the CMLB_MUTEX(cl), we will
5406 	 * continue to use valid values after the CMLB_MUTEX(cl) is
5407 	 * reacquired.
5408 	 */
5409 	capacity = cl->cl_blockcount;
5410 
5411 	/*
5412 	 * refresh the logical and physical geometry caches.
5413 	 * (data from mode sense format/rigid disk geometry pages,
5414 	 * and scsi_ifgetcap("geometry").
5415 	 */
5416 	cmlb_resync_geom_caches(cl, capacity, tg_cookie);
5417 
5418 	/*
5419 	 * Only DIRECT ACCESS devices will have Scl labels.
5420 	 * CD's supposedly have a Scl label, too
5421 	 */
5422 	if (cl->cl_device_type == DTYPE_DIRECT || ISREMOVABLE(cl)) {
5423 		fdisk_rval = cmlb_read_fdisk(cl, capacity, tg_cookie);
5424 		if (fdisk_rval != 0) {
5425 			ASSERT(mutex_owned(CMLB_MUTEX(cl)));
5426 			return (fdisk_rval);
5427 		}
5428 
5429 		if (cl->cl_solaris_size <= DK_LABEL_LOC) {
5430 			/*
5431 			 * Found fdisk table but no Solaris partition entry,
5432 			 * so don't call cmlb_uselabel() and don't create
5433 			 * a default label.
5434 			 */
5435 			label_rc = 0;
5436 			cl->cl_f_geometry_is_valid = B_TRUE;
5437 			goto no_solaris_partition;
5438 		}
5439 	} else if (capacity < 0) {
5440 		ASSERT(mutex_owned(CMLB_MUTEX(cl)));
5441 		return (EINVAL);
5442 	}
5443 
5444 	/*
5445 	 * For Removable media We reach here if we have found a
5446 	 * SOLARIS PARTITION.
5447 	 * If cl_f_geometry_is_valid is B_FALSE it indicates that the SOLARIS
5448 	 * PARTITION has changed from the previous one, hence we will setup a
5449 	 * default VTOC in this case.
5450 	 */
5451 	if (!cl->cl_f_geometry_is_valid) {
5452 		/* if we get here it is writable */
5453 		/* we are called from SMBOOT, and after a write of fdisk */
5454 		cmlb_build_default_label(cl, tg_cookie);
5455 		label_rc = 0;
5456 	}
5457 
5458 no_solaris_partition:
5459 
5460 #if defined(_SUNOS_VTOC_16)
5461 	/*
5462 	 * If we have valid geometry, set up the remaining fdisk partitions.
5463 	 * Note that dkl_cylno is not used for the fdisk map entries, so
5464 	 * we set it to an entirely bogus value.
5465 	 */
5466 	for (count = 0; count < FDISK_PARTS; count++) {
5467 		cl->cl_map[FDISK_P1 + count].dkl_cylno = UINT32_MAX;
5468 		cl->cl_map[FDISK_P1 + count].dkl_nblk =
5469 		    cl->cl_fmap[count].fmap_nblk;
5470 		cl->cl_offset[FDISK_P1 + count] =
5471 		    cl->cl_fmap[count].fmap_start;
5472 	}
5473 #endif
5474 
5475 	for (count = 0; count < NDKMAP; count++) {
5476 #if defined(_SUNOS_VTOC_8)
5477 		struct dk_map *lp  = &cl->cl_map[count];
5478 		cl->cl_offset[count] =
5479 		    cl->cl_g.dkg_nhead * cl->cl_g.dkg_nsect * lp->dkl_cylno;
5480 #elif defined(_SUNOS_VTOC_16)
5481 		struct dkl_partition *vp = &cl->cl_vtoc.v_part[count];
5482 		cl->cl_offset[count] = vp->p_start + cl->cl_solaris_offset;
5483 #else
5484 #error "No VTOC format defined."
5485 #endif
5486 	}
5487 
5488 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
5489 	return (label_rc);
5490 }
5491 #endif
5492 
5493 #if defined(__i386) || defined(__amd64)
5494 static int
5495 cmlb_dkio_get_virtgeom(struct cmlb_lun *cl, caddr_t arg, int flag)
5496 {
5497 	int err = 0;
5498 
5499 	/* Return the driver's notion of the media's logical geometry */
5500 	struct dk_geom	disk_geom;
5501 	struct dk_geom	*dkgp = &disk_geom;
5502 
5503 	mutex_enter(CMLB_MUTEX(cl));
5504 	/*
5505 	 * If there is no HBA geometry available, or
5506 	 * if the HBA returned us something that doesn't
5507 	 * really fit into an Int 13/function 8 geometry
5508 	 * result, just fail the ioctl.  See PSARC 1998/313.
5509 	 */
5510 	if (cl->cl_lgeom.g_nhead == 0 ||
5511 	    cl->cl_lgeom.g_nsect == 0 ||
5512 	    cl->cl_lgeom.g_ncyl > 1024) {
5513 		mutex_exit(CMLB_MUTEX(cl));
5514 		err = EINVAL;
5515 	} else {
5516 		dkgp->dkg_ncyl	= cl->cl_lgeom.g_ncyl;
5517 		dkgp->dkg_acyl	= cl->cl_lgeom.g_acyl;
5518 		dkgp->dkg_pcyl	= dkgp->dkg_ncyl + dkgp->dkg_acyl;
5519 		dkgp->dkg_nhead	= cl->cl_lgeom.g_nhead;
5520 		dkgp->dkg_nsect	= cl->cl_lgeom.g_nsect;
5521 
5522 		mutex_exit(CMLB_MUTEX(cl));
5523 		if (ddi_copyout(dkgp, (void *)arg,
5524 		    sizeof (struct dk_geom), flag)) {
5525 			err = EFAULT;
5526 		} else {
5527 			err = 0;
5528 		}
5529 	}
5530 	return (err);
5531 }
5532 #endif
5533 
5534 #if defined(__i386) || defined(__amd64)
5535 static int
5536 cmlb_dkio_get_phygeom(struct cmlb_lun *cl, caddr_t  arg, int flag,
5537     void *tg_cookie)
5538 {
5539 	int err = 0;
5540 	diskaddr_t capacity;
5541 
5542 
5543 	/* Return the driver's notion of the media physical geometry */
5544 	struct dk_geom	disk_geom;
5545 	struct dk_geom	*dkgp = &disk_geom;
5546 
5547 	mutex_enter(CMLB_MUTEX(cl));
5548 
5549 	if (cl->cl_g.dkg_nhead != 0 &&
5550 	    cl->cl_g.dkg_nsect != 0) {
5551 		/*
5552 		 * We succeeded in getting a geometry, but
5553 		 * right now it is being reported as just the
5554 		 * Solaris fdisk partition, just like for
5555 		 * DKIOCGGEOM. We need to change that to be
5556 		 * correct for the entire disk now.
5557 		 */
5558 		bcopy(&cl->cl_g, dkgp, sizeof (*dkgp));
5559 		dkgp->dkg_acyl = 0;
5560 		dkgp->dkg_ncyl = cl->cl_blockcount /
5561 		    (dkgp->dkg_nhead * dkgp->dkg_nsect);
5562 	} else {
5563 		bzero(dkgp, sizeof (struct dk_geom));
5564 		/*
5565 		 * This disk does not have a Solaris VTOC
5566 		 * so we must present a physical geometry
5567 		 * that will remain consistent regardless
5568 		 * of how the disk is used. This will ensure
5569 		 * that the geometry does not change regardless
5570 		 * of the fdisk partition type (ie. EFI, FAT32,
5571 		 * Solaris, etc).
5572 		 */
5573 		if (ISCD(cl)) {
5574 			dkgp->dkg_nhead = cl->cl_pgeom.g_nhead;
5575 			dkgp->dkg_nsect = cl->cl_pgeom.g_nsect;
5576 			dkgp->dkg_ncyl = cl->cl_pgeom.g_ncyl;
5577 			dkgp->dkg_acyl = cl->cl_pgeom.g_acyl;
5578 		} else {
5579 			/*
5580 			 * Invalid cl_blockcount can generate invalid
5581 			 * dk_geom and may result in division by zero
5582 			 * system failure. Should make sure blockcount
5583 			 * is valid before using it here.
5584 			 */
5585 			if (cl->cl_blockcount == 0) {
5586 				mutex_exit(CMLB_MUTEX(cl));
5587 				err = EIO;
5588 				return (err);
5589 			}
5590 			/*
5591 			 * Refer to comments related to off-by-1 at the
5592 			 * header of this file
5593 			 */
5594 			if (cl->cl_alter_behavior & CMLB_OFF_BY_ONE)
5595 				capacity = cl->cl_blockcount - 1;
5596 			else
5597 				capacity = cl->cl_blockcount;
5598 
5599 			cmlb_convert_geometry(cl, capacity, dkgp, tg_cookie);
5600 			dkgp->dkg_acyl = 0;
5601 			dkgp->dkg_ncyl = capacity /
5602 			    (dkgp->dkg_nhead * dkgp->dkg_nsect);
5603 		}
5604 	}
5605 	dkgp->dkg_pcyl = dkgp->dkg_ncyl + dkgp->dkg_acyl;
5606 
5607 	mutex_exit(CMLB_MUTEX(cl));
5608 	if (ddi_copyout(dkgp, (void *)arg, sizeof (struct dk_geom), flag))
5609 		err = EFAULT;
5610 
5611 	return (err);
5612 }
5613 #endif
5614 
5615 #if defined(__i386) || defined(__amd64)
5616 static int
5617 cmlb_dkio_partinfo(struct cmlb_lun *cl, dev_t dev, caddr_t  arg, int flag)
5618 {
5619 	int err = 0;
5620 
5621 	/*
5622 	 * Return parameters describing the selected disk slice.
5623 	 * Note: this ioctl is for the intel platform only
5624 	 */
5625 	int part;
5626 
5627 	if (cl->cl_alter_behavior & CMLB_CREATE_P0_MINOR_NODE)
5628 		part = getminor(dev) & ((1 << CMLBUNIT_FORCE_P0_SHIFT) - 1);
5629 	else
5630 		part = CMLBPART(dev);
5631 
5632 	mutex_enter(CMLB_MUTEX(cl));
5633 	/* don't check cl_solaris_size for pN */
5634 	if (part < P0_RAW_DISK && cl->cl_solaris_size == 0) {
5635 		err = EIO;
5636 		mutex_exit(CMLB_MUTEX(cl));
5637 	} else {
5638 		struct part_info p;
5639 
5640 		p.p_start = (daddr_t)cl->cl_offset[part];
5641 		p.p_length = (int)cl->cl_map[part].dkl_nblk;
5642 		mutex_exit(CMLB_MUTEX(cl));
5643 #ifdef _MULTI_DATAMODEL
5644 		switch (ddi_model_convert_from(flag & FMODELS)) {
5645 		case DDI_MODEL_ILP32:
5646 		{
5647 			struct part_info32 p32;
5648 
5649 			p32.p_start = (daddr32_t)p.p_start;
5650 			p32.p_length = p.p_length;
5651 			if (ddi_copyout(&p32, (void *)arg,
5652 			    sizeof (p32), flag))
5653 				err = EFAULT;
5654 			break;
5655 		}
5656 
5657 		case DDI_MODEL_NONE:
5658 		{
5659 			if (ddi_copyout(&p, (void *)arg, sizeof (p),
5660 			    flag))
5661 				err = EFAULT;
5662 			break;
5663 		}
5664 		}
5665 #else /* ! _MULTI_DATAMODEL */
5666 		if (ddi_copyout(&p, (void *)arg, sizeof (p), flag))
5667 			err = EFAULT;
5668 #endif /* _MULTI_DATAMODEL */
5669 	}
5670 	return (err);
5671 }
5672 static int
5673 cmlb_dkio_extpartinfo(struct cmlb_lun *cl, dev_t dev, caddr_t  arg, int flag)
5674 {
5675 	int err = 0;
5676 
5677 	/*
5678 	 * Return parameters describing the selected disk slice.
5679 	 * Note: this ioctl is for the intel platform only
5680 	 */
5681 	int part;
5682 
5683 	if (cl->cl_alter_behavior & CMLB_CREATE_P0_MINOR_NODE)
5684 		part = getminor(dev) & ((1 << CMLBUNIT_FORCE_P0_SHIFT) - 1);
5685 	else
5686 		part = CMLBPART(dev);
5687 
5688 	mutex_enter(CMLB_MUTEX(cl));
5689 	/* don't check cl_solaris_size for pN */
5690 	if (part < P0_RAW_DISK && cl->cl_solaris_size == 0) {
5691 		err = EIO;
5692 		mutex_exit(CMLB_MUTEX(cl));
5693 	} else {
5694 		struct extpart_info p;
5695 
5696 		p.p_start = (diskaddr_t)cl->cl_offset[part];
5697 		p.p_length = (diskaddr_t)cl->cl_map[part].dkl_nblk;
5698 		mutex_exit(CMLB_MUTEX(cl));
5699 		if (ddi_copyout(&p, (void *)arg, sizeof (p), flag))
5700 			err = EFAULT;
5701 	}
5702 	return (err);
5703 }
5704 #endif
5705 
5706 int
5707 cmlb_prop_op(cmlb_handle_t cmlbhandle,
5708     dev_t dev, dev_info_t *dip, ddi_prop_op_t prop_op, int mod_flags,
5709     char *name, caddr_t valuep, int *lengthp, int part, void *tg_cookie)
5710 {
5711 	struct cmlb_lun	*cl;
5712 	diskaddr_t	capacity;
5713 	uint32_t	lbasize;
5714 	enum		dp { DP_NBLOCKS, DP_BLKSIZE, DP_SSD } dp;
5715 	int		callers_length;
5716 	caddr_t		buffer;
5717 	uint64_t	nblocks64;
5718 	uint_t		dblk;
5719 	tg_attribute_t	tgattr;
5720 
5721 	/* Always fallback to ddi_prop_op... */
5722 	cl = (struct cmlb_lun *)cmlbhandle;
5723 	if (cl == NULL) {
5724 fallback:	return (ddi_prop_op(dev, dip, prop_op, mod_flags,
5725 		    name, valuep, lengthp));
5726 	}
5727 
5728 	/* Pick up capacity and blocksize information. */
5729 	capacity = cl->cl_blockcount;
5730 	if (capacity == 0)
5731 		goto fallback;
5732 	lbasize = cl->cl_tgt_blocksize;
5733 	if (lbasize == 0)
5734 		lbasize = DEV_BSIZE;	/* 0 -> DEV_BSIZE units */
5735 
5736 	/* Check for dynamic property of whole device. */
5737 	if (dev == DDI_DEV_T_ANY) {
5738 		/* Fallback to ddi_prop_op if we don't understand.  */
5739 		if (strcmp(name, "device-nblocks") == 0)
5740 			dp = DP_NBLOCKS;
5741 		else if (strcmp(name, "device-blksize") == 0)
5742 			dp = DP_BLKSIZE;
5743 		else if (strcmp(name, "device-solid-state") == 0)
5744 			dp = DP_SSD;
5745 		else
5746 			goto fallback;
5747 
5748 		/* get callers length, establish length of our dynamic prop */
5749 		callers_length = *lengthp;
5750 		if (dp == DP_NBLOCKS)
5751 			*lengthp = sizeof (uint64_t);
5752 		else if ((dp == DP_BLKSIZE) || (dp == DP_SSD))
5753 			*lengthp = sizeof (uint32_t);
5754 
5755 		/* service request for the length of the property */
5756 		if (prop_op == PROP_LEN)
5757 			return (DDI_PROP_SUCCESS);
5758 
5759 		switch (prop_op) {
5760 		case PROP_LEN_AND_VAL_ALLOC:
5761 			if ((buffer = kmem_alloc(*lengthp,
5762 			    (mod_flags & DDI_PROP_CANSLEEP) ?
5763 			    KM_SLEEP : KM_NOSLEEP)) == NULL)
5764 				return (DDI_PROP_NO_MEMORY);
5765 			*(caddr_t *)valuep = buffer;	/* set callers buf */
5766 			break;
5767 
5768 		case PROP_LEN_AND_VAL_BUF:
5769 			/* the length of the prop and the request must match */
5770 			if (callers_length != *lengthp)
5771 				return (DDI_PROP_INVAL_ARG);
5772 			buffer = valuep;		/* get callers buf */
5773 			break;
5774 
5775 		default:
5776 			return (DDI_PROP_INVAL_ARG);
5777 		}
5778 
5779 		/* transfer the value into the buffer */
5780 		switch (dp) {
5781 		case DP_NBLOCKS:
5782 			*((uint64_t *)buffer) = capacity;
5783 			break;
5784 		case DP_BLKSIZE:
5785 			*((uint32_t *)buffer) = lbasize;
5786 			break;
5787 		case DP_SSD:
5788 			if (DK_TG_GETATTRIBUTE(cl, &tgattr, tg_cookie) != 0)
5789 				tgattr.media_is_solid_state = B_FALSE;
5790 			*((uint32_t *)buffer) =
5791 			    tgattr.media_is_solid_state ? 1 : 0;
5792 			break;
5793 		}
5794 		return (DDI_PROP_SUCCESS);
5795 	}
5796 
5797 	/*
5798 	 * Support dynamic size oriented properties of partition. Requests
5799 	 * issued under conditions where size is valid are passed to
5800 	 * ddi_prop_op_nblocks with the size information, otherwise the
5801 	 * request is passed to ddi_prop_op. Size depends on valid geometry.
5802 	 */
5803 	if (!cmlb_is_valid(cmlbhandle))
5804 		goto fallback;
5805 
5806 	/* Get partition nblocks value. */
5807 	(void) cmlb_partinfo(cmlbhandle, part,
5808 	    (diskaddr_t *)&nblocks64, NULL, NULL, NULL, tg_cookie);
5809 
5810 	/*
5811 	 * Assume partition information is in sys_blocksize units, compute
5812 	 * divisor for size(9P) property representation.
5813 	 */
5814 	dblk = lbasize / cl->cl_sys_blocksize;
5815 
5816 	/* Now let ddi_prop_op_nblocks_blksize() handle the request. */
5817 	return (ddi_prop_op_nblocks_blksize(dev, dip, prop_op, mod_flags,
5818 	    name, valuep, lengthp, nblocks64 / dblk, lbasize));
5819 }
5820