xref: /illumos-gate/usr/src/uts/common/io/cmlb.c (revision a1ec709f09b030b52805b585412bd7c449af04b7)
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 2009 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 /*
28  * This module provides support for labeling operations for target
29  * drivers.
30  */
31 
32 #include <sys/scsi/scsi.h>
33 #include <sys/sunddi.h>
34 #include <sys/dklabel.h>
35 #include <sys/dkio.h>
36 #include <sys/vtoc.h>
37 #include <sys/dktp/fdisk.h>
38 #include <sys/vtrace.h>
39 #include <sys/efi_partition.h>
40 #include <sys/cmlb.h>
41 #include <sys/cmlb_impl.h>
42 #if defined(__i386) || defined(__amd64)
43 #include <sys/fs/dv_node.h>
44 #endif
45 #include <sys/ddi_impldefs.h>
46 
47 /*
48  * Driver minor node structure and data table
49  */
50 struct driver_minor_data {
51 	char	*name;
52 	minor_t	minor;
53 	int	type;
54 };
55 
56 static struct driver_minor_data dk_minor_data[] = {
57 	{"a", 0, S_IFBLK},
58 	{"b", 1, S_IFBLK},
59 	{"c", 2, S_IFBLK},
60 	{"d", 3, S_IFBLK},
61 	{"e", 4, S_IFBLK},
62 	{"f", 5, S_IFBLK},
63 	{"g", 6, S_IFBLK},
64 	{"h", 7, S_IFBLK},
65 #if defined(_SUNOS_VTOC_16)
66 	{"i", 8, S_IFBLK},
67 	{"j", 9, S_IFBLK},
68 	{"k", 10, S_IFBLK},
69 	{"l", 11, S_IFBLK},
70 	{"m", 12, S_IFBLK},
71 	{"n", 13, S_IFBLK},
72 	{"o", 14, S_IFBLK},
73 	{"p", 15, S_IFBLK},
74 #endif			/* defined(_SUNOS_VTOC_16) */
75 #if defined(_FIRMWARE_NEEDS_FDISK)
76 	{"q", 16, S_IFBLK},
77 	{"r", 17, S_IFBLK},
78 	{"s", 18, S_IFBLK},
79 	{"t", 19, S_IFBLK},
80 	{"u", 20, S_IFBLK},
81 #endif			/* defined(_FIRMWARE_NEEDS_FDISK) */
82 	{"a,raw", 0, S_IFCHR},
83 	{"b,raw", 1, S_IFCHR},
84 	{"c,raw", 2, S_IFCHR},
85 	{"d,raw", 3, S_IFCHR},
86 	{"e,raw", 4, S_IFCHR},
87 	{"f,raw", 5, S_IFCHR},
88 	{"g,raw", 6, S_IFCHR},
89 	{"h,raw", 7, S_IFCHR},
90 #if defined(_SUNOS_VTOC_16)
91 	{"i,raw", 8, S_IFCHR},
92 	{"j,raw", 9, S_IFCHR},
93 	{"k,raw", 10, S_IFCHR},
94 	{"l,raw", 11, S_IFCHR},
95 	{"m,raw", 12, S_IFCHR},
96 	{"n,raw", 13, S_IFCHR},
97 	{"o,raw", 14, S_IFCHR},
98 	{"p,raw", 15, S_IFCHR},
99 #endif			/* defined(_SUNOS_VTOC_16) */
100 #if defined(_FIRMWARE_NEEDS_FDISK)
101 	{"q,raw", 16, S_IFCHR},
102 	{"r,raw", 17, S_IFCHR},
103 	{"s,raw", 18, S_IFCHR},
104 	{"t,raw", 19, S_IFCHR},
105 	{"u,raw", 20, S_IFCHR},
106 #endif			/* defined(_FIRMWARE_NEEDS_FDISK) */
107 	{0}
108 };
109 
110 #if defined(__i386) || defined(__amd64)
111 #if defined(_FIRMWARE_NEEDS_FDISK)
112 static struct driver_minor_data dk_ext_minor_data[] = {
113 	{"p5", 21, S_IFBLK},
114 	{"p6", 22, S_IFBLK},
115 	{"p7", 23, S_IFBLK},
116 	{"p8", 24, S_IFBLK},
117 	{"p9", 25, S_IFBLK},
118 	{"p10", 26, S_IFBLK},
119 	{"p11", 27, S_IFBLK},
120 	{"p12", 28, S_IFBLK},
121 	{"p13", 29, S_IFBLK},
122 	{"p14", 30, S_IFBLK},
123 	{"p15", 31, S_IFBLK},
124 	{"p16", 32, S_IFBLK},
125 	{"p17", 33, S_IFBLK},
126 	{"p18", 34, S_IFBLK},
127 	{"p19", 35, S_IFBLK},
128 	{"p20", 36, S_IFBLK},
129 	{"p21", 37, S_IFBLK},
130 	{"p22", 38, S_IFBLK},
131 	{"p23", 39, S_IFBLK},
132 	{"p24", 40, S_IFBLK},
133 	{"p25", 41, S_IFBLK},
134 	{"p26", 42, S_IFBLK},
135 	{"p27", 43, S_IFBLK},
136 	{"p28", 44, S_IFBLK},
137 	{"p29", 45, S_IFBLK},
138 	{"p30", 46, S_IFBLK},
139 	{"p31", 47, S_IFBLK},
140 	{"p32", 48, S_IFBLK},
141 	{"p33", 49, S_IFBLK},
142 	{"p34", 50, S_IFBLK},
143 	{"p35", 51, S_IFBLK},
144 	{"p36", 52, S_IFBLK},
145 	{"p5,raw", 21, S_IFCHR},
146 	{"p6,raw", 22, S_IFCHR},
147 	{"p7,raw", 23, S_IFCHR},
148 	{"p8,raw", 24, S_IFCHR},
149 	{"p9,raw", 25, S_IFCHR},
150 	{"p10,raw", 26, S_IFCHR},
151 	{"p11,raw", 27, S_IFCHR},
152 	{"p12,raw", 28, S_IFCHR},
153 	{"p13,raw", 29, S_IFCHR},
154 	{"p14,raw", 30, S_IFCHR},
155 	{"p15,raw", 31, S_IFCHR},
156 	{"p16,raw", 32, S_IFCHR},
157 	{"p17,raw", 33, S_IFCHR},
158 	{"p18,raw", 34, S_IFCHR},
159 	{"p19,raw", 35, S_IFCHR},
160 	{"p20,raw", 36, S_IFCHR},
161 	{"p21,raw", 37, S_IFCHR},
162 	{"p22,raw", 38, S_IFCHR},
163 	{"p23,raw", 39, S_IFCHR},
164 	{"p24,raw", 40, S_IFCHR},
165 	{"p25,raw", 41, S_IFCHR},
166 	{"p26,raw", 42, S_IFCHR},
167 	{"p27,raw", 43, S_IFCHR},
168 	{"p28,raw", 44, S_IFCHR},
169 	{"p29,raw", 45, S_IFCHR},
170 	{"p30,raw", 46, S_IFCHR},
171 	{"p31,raw", 47, S_IFCHR},
172 	{"p32,raw", 48, S_IFCHR},
173 	{"p33,raw", 49, S_IFCHR},
174 	{"p34,raw", 50, S_IFCHR},
175 	{"p35,raw", 51, S_IFCHR},
176 	{"p36,raw", 52, S_IFCHR},
177 	{0}
178 };
179 #endif			/* defined(_FIRMWARE_NEEDS_FDISK) */
180 #endif			/* if defined(__i386) || defined(__amd64) */
181 
182 static struct driver_minor_data dk_minor_data_efi[] = {
183 	{"a", 0, S_IFBLK},
184 	{"b", 1, S_IFBLK},
185 	{"c", 2, S_IFBLK},
186 	{"d", 3, S_IFBLK},
187 	{"e", 4, S_IFBLK},
188 	{"f", 5, S_IFBLK},
189 	{"g", 6, S_IFBLK},
190 	{"wd", 7, S_IFBLK},
191 #if defined(_SUNOS_VTOC_16)
192 	{"i", 8, S_IFBLK},
193 	{"j", 9, S_IFBLK},
194 	{"k", 10, S_IFBLK},
195 	{"l", 11, S_IFBLK},
196 	{"m", 12, S_IFBLK},
197 	{"n", 13, S_IFBLK},
198 	{"o", 14, S_IFBLK},
199 	{"p", 15, S_IFBLK},
200 #endif			/* defined(_SUNOS_VTOC_16) */
201 #if defined(_FIRMWARE_NEEDS_FDISK)
202 	{"q", 16, S_IFBLK},
203 	{"r", 17, S_IFBLK},
204 	{"s", 18, S_IFBLK},
205 	{"t", 19, S_IFBLK},
206 	{"u", 20, S_IFBLK},
207 #endif			/* defined(_FIRMWARE_NEEDS_FDISK) */
208 	{"a,raw", 0, S_IFCHR},
209 	{"b,raw", 1, S_IFCHR},
210 	{"c,raw", 2, S_IFCHR},
211 	{"d,raw", 3, S_IFCHR},
212 	{"e,raw", 4, S_IFCHR},
213 	{"f,raw", 5, S_IFCHR},
214 	{"g,raw", 6, S_IFCHR},
215 	{"wd,raw", 7, S_IFCHR},
216 #if defined(_SUNOS_VTOC_16)
217 	{"i,raw", 8, S_IFCHR},
218 	{"j,raw", 9, S_IFCHR},
219 	{"k,raw", 10, S_IFCHR},
220 	{"l,raw", 11, S_IFCHR},
221 	{"m,raw", 12, S_IFCHR},
222 	{"n,raw", 13, S_IFCHR},
223 	{"o,raw", 14, S_IFCHR},
224 	{"p,raw", 15, S_IFCHR},
225 #endif			/* defined(_SUNOS_VTOC_16) */
226 #if defined(_FIRMWARE_NEEDS_FDISK)
227 	{"q,raw", 16, S_IFCHR},
228 	{"r,raw", 17, S_IFCHR},
229 	{"s,raw", 18, S_IFCHR},
230 	{"t,raw", 19, S_IFCHR},
231 	{"u,raw", 20, S_IFCHR},
232 #endif			/* defined(_FIRMWARE_NEEDS_FDISK) */
233 	{0}
234 };
235 
236 /*
237  * Declare the dynamic properties implemented in prop_op(9E) implementation
238  * that we want to have show up in a di_init(3DEVINFO) device tree snapshot
239  * of drivers that call cmlb_attach().
240  */
241 static i_ddi_prop_dyn_t cmlb_prop_dyn[] = {
242 	{"Nblocks",		DDI_PROP_TYPE_INT64,	S_IFBLK},
243 	{"Size",		DDI_PROP_TYPE_INT64,	S_IFCHR},
244 	{"device-nblocks",	DDI_PROP_TYPE_INT64},
245 	{"device-blksize",	DDI_PROP_TYPE_INT},
246 	{NULL}
247 };
248 
249 /*
250  * External kernel interfaces
251  */
252 extern struct mod_ops mod_miscops;
253 
254 extern int ddi_create_internal_pathname(dev_info_t *dip, char *name,
255     int spec_type, minor_t minor_num);
256 
257 /*
258  * Global buffer and mutex for debug logging
259  */
260 static char	cmlb_log_buffer[1024];
261 static kmutex_t	cmlb_log_mutex;
262 
263 
264 struct cmlb_lun *cmlb_debug_cl = NULL;
265 uint_t cmlb_level_mask = 0x0;
266 
267 int cmlb_rot_delay = 4;	/* default rotational delay */
268 
269 static struct modlmisc modlmisc = {
270 	&mod_miscops,   /* Type of module */
271 	"Common Labeling module"
272 };
273 
274 static struct modlinkage modlinkage = {
275 	MODREV_1, (void *)&modlmisc, NULL
276 };
277 
278 /* Local function prototypes */
279 static dev_t cmlb_make_device(struct cmlb_lun *cl);
280 static int cmlb_validate_geometry(struct cmlb_lun *cl, boolean_t forcerevalid,
281     int flags, void *tg_cookie);
282 static void cmlb_resync_geom_caches(struct cmlb_lun *cl, diskaddr_t capacity,
283     void *tg_cookie);
284 static int cmlb_read_fdisk(struct cmlb_lun *cl, diskaddr_t capacity,
285     void *tg_cookie);
286 static void cmlb_swap_efi_gpt(efi_gpt_t *e);
287 static void cmlb_swap_efi_gpe(int nparts, efi_gpe_t *p);
288 static int cmlb_validate_efi(efi_gpt_t *labp);
289 static int cmlb_use_efi(struct cmlb_lun *cl, diskaddr_t capacity, int flags,
290     void *tg_cookie);
291 static void cmlb_build_default_label(struct cmlb_lun *cl, void *tg_cookie);
292 static int  cmlb_uselabel(struct cmlb_lun *cl,  struct dk_label *l, int flags);
293 #if defined(_SUNOS_VTOC_8)
294 static void cmlb_build_user_vtoc(struct cmlb_lun *cl, struct vtoc *user_vtoc);
295 #endif
296 static int cmlb_build_label_vtoc(struct cmlb_lun *cl, struct vtoc *user_vtoc);
297 static int cmlb_write_label(struct cmlb_lun *cl, void *tg_cookie);
298 static int cmlb_set_vtoc(struct cmlb_lun *cl, struct dk_label *dkl,
299     void *tg_cookie);
300 static void cmlb_clear_efi(struct cmlb_lun *cl, void *tg_cookie);
301 static void cmlb_clear_vtoc(struct cmlb_lun *cl, void *tg_cookie);
302 static void cmlb_setup_default_geometry(struct cmlb_lun *cl, void *tg_cookie);
303 static int cmlb_create_minor_nodes(struct cmlb_lun *cl);
304 static int cmlb_check_update_blockcount(struct cmlb_lun *cl, void *tg_cookie);
305 static boolean_t cmlb_check_efi_mbr(uchar_t *buf, boolean_t *is_mbr);
306 
307 #if defined(__i386) || defined(__amd64)
308 static int cmlb_update_fdisk_and_vtoc(struct cmlb_lun *cl, void *tg_cookie);
309 #endif
310 
311 #if defined(_FIRMWARE_NEEDS_FDISK)
312 static boolean_t  cmlb_has_max_chs_vals(struct ipart *fdp);
313 #endif
314 
315 #if defined(_SUNOS_VTOC_16)
316 static void cmlb_convert_geometry(struct cmlb_lun *cl, diskaddr_t capacity,
317     struct dk_geom *cl_g, void *tg_cookie);
318 #endif
319 
320 static int cmlb_dkio_get_geometry(struct cmlb_lun *cl, caddr_t arg, int flag,
321     void *tg_cookie);
322 static int cmlb_dkio_set_geometry(struct cmlb_lun *cl, caddr_t arg, int flag);
323 static int cmlb_dkio_get_partition(struct cmlb_lun *cl, caddr_t arg, int flag,
324     void *tg_cookie);
325 static int cmlb_dkio_set_partition(struct cmlb_lun *cl, caddr_t arg, int flag);
326 static int cmlb_dkio_get_efi(struct cmlb_lun *cl, caddr_t arg, int flag,
327     void *tg_cookie);
328 static int cmlb_dkio_set_efi(struct cmlb_lun *cl, dev_t dev, caddr_t arg,
329     int flag, void *tg_cookie);
330 static int cmlb_dkio_get_vtoc(struct cmlb_lun *cl, caddr_t arg, int flag,
331     void *tg_cookie);
332 static int cmlb_dkio_get_extvtoc(struct cmlb_lun *cl, caddr_t arg, int flag,
333     void *tg_cookie);
334 static int cmlb_dkio_set_vtoc(struct cmlb_lun *cl, dev_t dev, caddr_t arg,
335     int flag, void *tg_cookie);
336 static int cmlb_dkio_set_extvtoc(struct cmlb_lun *cl, dev_t dev, caddr_t arg,
337     int flag, void *tg_cookie);
338 static int cmlb_dkio_get_mboot(struct cmlb_lun *cl, caddr_t arg, int flag,
339     void *tg_cookie);
340 static int cmlb_dkio_set_mboot(struct cmlb_lun *cl, caddr_t arg, int flag,
341     void *tg_cookie);
342 static int cmlb_dkio_partition(struct cmlb_lun *cl, caddr_t arg, int flag,
343     void *tg_cookie);
344 
345 #if defined(__i386) || defined(__amd64)
346 static int cmlb_dkio_set_ext_part(struct cmlb_lun *cl, caddr_t arg, int flag,
347     void *tg_cookie);
348 static int cmlb_validate_ext_part(struct cmlb_lun *cl, int part, int epart,
349     uint32_t start, uint32_t size);
350 static int cmlb_is_linux_swap(struct cmlb_lun *cl, uint32_t part_start,
351     void *tg_cookie);
352 static int cmlb_dkio_get_virtgeom(struct cmlb_lun *cl, caddr_t arg, int flag);
353 static int cmlb_dkio_get_phygeom(struct cmlb_lun *cl, caddr_t  arg, int flag,
354     void *tg_cookie);
355 static int cmlb_dkio_partinfo(struct cmlb_lun *cl, dev_t dev, caddr_t arg,
356     int flag);
357 static int cmlb_dkio_extpartinfo(struct cmlb_lun *cl, dev_t dev, caddr_t arg,
358     int flag);
359 #endif
360 
361 static void cmlb_dbg(uint_t comp, struct cmlb_lun *cl, const char *fmt, ...);
362 static void cmlb_v_log(dev_info_t *dev, char *label, uint_t level,
363     const char *fmt, va_list ap);
364 static void cmlb_log(dev_info_t *dev, char *label, uint_t level,
365     const char *fmt, ...);
366 
367 int
368 _init(void)
369 {
370 	mutex_init(&cmlb_log_mutex, NULL, MUTEX_DRIVER, NULL);
371 	return (mod_install(&modlinkage));
372 }
373 
374 int
375 _info(struct modinfo *modinfop)
376 {
377 	return (mod_info(&modlinkage, modinfop));
378 }
379 
380 int
381 _fini(void)
382 {
383 	int err;
384 
385 	if ((err = mod_remove(&modlinkage)) != 0) {
386 		return (err);
387 	}
388 
389 	mutex_destroy(&cmlb_log_mutex);
390 	return (err);
391 }
392 
393 /*
394  * cmlb_dbg is used for debugging to log additional info
395  * Level of output is controlled via cmlb_level_mask setting.
396  */
397 static void
398 cmlb_dbg(uint_t comp, struct cmlb_lun *cl, const char *fmt, ...)
399 {
400 	va_list		ap;
401 	dev_info_t	*dev;
402 	uint_t		level_mask = 0;
403 
404 	ASSERT(cl != NULL);
405 	dev = CMLB_DEVINFO(cl);
406 	ASSERT(dev != NULL);
407 	/*
408 	 * Filter messages based on the global component and level masks,
409 	 * also print if cl matches the value of cmlb_debug_cl, or if
410 	 * cmlb_debug_cl is set to NULL.
411 	 */
412 	if (comp & CMLB_TRACE)
413 		level_mask |= CMLB_LOGMASK_TRACE;
414 
415 	if (comp & CMLB_INFO)
416 		level_mask |= CMLB_LOGMASK_INFO;
417 
418 	if (comp & CMLB_ERROR)
419 		level_mask |= CMLB_LOGMASK_ERROR;
420 
421 	if ((cmlb_level_mask & level_mask) &&
422 	    ((cmlb_debug_cl == NULL) || (cmlb_debug_cl == cl))) {
423 		va_start(ap, fmt);
424 		cmlb_v_log(dev, CMLB_LABEL(cl), CE_CONT, fmt, ap);
425 		va_end(ap);
426 	}
427 }
428 
429 /*
430  * cmlb_log is basically a duplicate of scsi_log. It is redefined here
431  * so that this module does not depend on scsi module.
432  */
433 static void
434 cmlb_log(dev_info_t *dev, char *label, uint_t level, const char *fmt, ...)
435 {
436 	va_list		ap;
437 
438 	va_start(ap, fmt);
439 	cmlb_v_log(dev, label, level, fmt, ap);
440 	va_end(ap);
441 }
442 
443 static void
444 cmlb_v_log(dev_info_t *dev, char *label, uint_t level, const char *fmt,
445     va_list ap)
446 {
447 	static char 	name[256];
448 	int 		log_only = 0;
449 	int 		boot_only = 0;
450 	int 		console_only = 0;
451 
452 	mutex_enter(&cmlb_log_mutex);
453 
454 	if (dev) {
455 		if (level == CE_PANIC || level == CE_WARN ||
456 		    level == CE_NOTE) {
457 			(void) sprintf(name, "%s (%s%d):\n",
458 			    ddi_pathname(dev, cmlb_log_buffer),
459 			    label, ddi_get_instance(dev));
460 		} else {
461 			name[0] = '\0';
462 		}
463 	} else {
464 		(void) sprintf(name, "%s:", label);
465 	}
466 
467 	(void) vsprintf(cmlb_log_buffer, fmt, ap);
468 
469 	switch (cmlb_log_buffer[0]) {
470 	case '!':
471 		log_only = 1;
472 		break;
473 	case '?':
474 		boot_only = 1;
475 		break;
476 	case '^':
477 		console_only = 1;
478 		break;
479 	}
480 
481 	switch (level) {
482 	case CE_NOTE:
483 		level = CE_CONT;
484 		/* FALLTHROUGH */
485 	case CE_CONT:
486 	case CE_WARN:
487 	case CE_PANIC:
488 		if (boot_only) {
489 			cmn_err(level, "?%s\t%s", name, &cmlb_log_buffer[1]);
490 		} else if (console_only) {
491 			cmn_err(level, "^%s\t%s", name, &cmlb_log_buffer[1]);
492 		} else if (log_only) {
493 			cmn_err(level, "!%s\t%s", name, &cmlb_log_buffer[1]);
494 		} else {
495 			cmn_err(level, "%s\t%s", name, cmlb_log_buffer);
496 		}
497 		break;
498 	case CE_IGNORE:
499 		break;
500 	default:
501 		cmn_err(CE_CONT, "^DEBUG: %s\t%s", name, cmlb_log_buffer);
502 		break;
503 	}
504 	mutex_exit(&cmlb_log_mutex);
505 }
506 
507 
508 /*
509  * cmlb_alloc_handle:
510  *
511  *	Allocates a handle.
512  *
513  * Arguments:
514  *	cmlbhandlep	pointer to handle
515  *
516  * Notes:
517  *	Allocates a handle and stores the allocated handle in the area
518  *	pointed to by cmlbhandlep
519  *
520  * Context:
521  *	Kernel thread only (can sleep).
522  */
523 void
524 cmlb_alloc_handle(cmlb_handle_t *cmlbhandlep)
525 {
526 	struct cmlb_lun 	*cl;
527 
528 	cl = kmem_zalloc(sizeof (struct cmlb_lun), KM_SLEEP);
529 	ASSERT(cmlbhandlep != NULL);
530 
531 	cl->cl_state = CMLB_INITED;
532 	cl->cl_def_labeltype = CMLB_LABEL_UNDEF;
533 	mutex_init(CMLB_MUTEX(cl), NULL, MUTEX_DRIVER, NULL);
534 
535 	*cmlbhandlep = (cmlb_handle_t)(cl);
536 }
537 
538 /*
539  * cmlb_free_handle
540  *
541  *	Frees handle.
542  *
543  * Arguments:
544  *	cmlbhandlep	pointer to handle
545  */
546 void
547 cmlb_free_handle(cmlb_handle_t *cmlbhandlep)
548 {
549 	struct cmlb_lun 	*cl;
550 
551 	cl = (struct cmlb_lun *)*cmlbhandlep;
552 	if (cl != NULL) {
553 		mutex_destroy(CMLB_MUTEX(cl));
554 		kmem_free(cl, sizeof (struct cmlb_lun));
555 	}
556 
557 }
558 
559 /*
560  * cmlb_attach:
561  *
562  *	Attach handle to device, create minor nodes for device.
563  *
564  * Arguments:
565  * 	devi		pointer to device's dev_info structure.
566  * 	tgopsp		pointer to array of functions cmlb can use to callback
567  *			to target driver.
568  *
569  *	device_type	Peripheral device type as defined in
570  *			scsi/generic/inquiry.h
571  *
572  *	is_removable	whether or not device is removable.
573  *
574  *	is_hotpluggable	whether or not device is hotpluggable.
575  *
576  *	node_type	minor node type (as used by ddi_create_minor_node)
577  *
578  *	alter_behavior
579  *			bit flags:
580  *
581  *			CMLB_CREATE_ALTSLICE_VTOC_16_DTYPE_DIRECT: create
582  *			an alternate slice for the default label, if
583  *			device type is DTYPE_DIRECT an architectures default
584  *			label type is VTOC16.
585  *			Otherwise alternate slice will no be created.
586  *
587  *
588  *			CMLB_FAKE_GEOM_LABEL_IOCTLS_VTOC8: report a default
589  *			geometry and label for DKIOCGGEOM and DKIOCGVTOC
590  *			on architecture with VTOC8 label types.
591  *
592  * 			CMLB_OFF_BY_ONE: do the workaround for legacy off-by-
593  *                      one bug in obtaining capacity (in sd):
594  *			SCSI READ_CAPACITY command returns the LBA number of the
595  *			last logical block, but sd once treated this number as
596  *			disks' capacity on x86 platform. And LBAs are addressed
597  *			based 0. So the last block was lost on x86 platform.
598  *
599  *			Now, we remove this workaround. In order for present sd
600  *			driver to work with disks which are labeled/partitioned
601  *			via previous sd, we add workaround as follows:
602  *
603  *			1) Locate backup EFI label: cmlb searches the next to
604  *			   last
605  *			   block for backup EFI label. If fails, it will
606  *			   turn to the last block for backup EFI label;
607  *
608  *			2) Clear backup EFI label: cmlb first search the last
609  *			   block for backup EFI label, and will search the
610  *			   next to last block only if failed for the last
611  *			   block.
612  *
613  *			3) Calculate geometry:refer to cmlb_convert_geometry()
614  *			   If capacity increasing by 1 causes disks' capacity
615  *			   to cross over the limits in geometry calculation,
616  *			   geometry info will change. This will raise an issue:
617  *			   In case that primary VTOC label is destroyed, format
618  *			   commandline can restore it via backup VTOC labels.
619  *			   And format locates backup VTOC labels by use of
620  *			   geometry. So changing geometry will
621  *			   prevent format from finding backup VTOC labels. To
622  *			   eliminate this side effect for compatibility,
623  *			   sd uses (capacity -1) to calculate geometry;
624  *
625  *			4) 1TB disks: some important data structures use
626  *			   32-bit signed long/int (for example, daddr_t),
627  *			   so that sd doesn't support a disk with capacity
628  *			   larger than 1TB on 32-bit platform. However,
629  *			   for exactly 1TB disk, it was treated as (1T - 512)B
630  *			   in the past, and could have valid Solaris
631  *			   partitions. To workaround this, if an exactly 1TB
632  *			   disk has Solaris fdisk partition, it will be allowed
633  *			   to work with sd.
634  *
635  *
636  *
637  *			CMLB_FAKE_LABEL_ONE_PARTITION: create s0 and s2 covering
638  *			the entire disk, if there is no valid partition info.
639  *			If there is a valid Solaris partition, s0 and s2 will
640  *			only cover the entire Solaris partition.
641  *
642  *
643  *	cmlbhandle	cmlb handle associated with device
644  *
645  *	tg_cookie	cookie from target driver to be passed back to target
646  *			driver when we call back to it through tg_ops.
647  *
648  * Notes:
649  *	Assumes a default label based on capacity for non-removable devices.
650  *	If capacity > 1TB, EFI is assumed otherwise VTOC (default VTOC
651  *	for the architecture).
652  *
653  *	For removable devices, default label type is assumed to be VTOC
654  *	type. Create minor nodes based on a default label type.
655  *	Label on the media is not validated.
656  *	minor number consists of:
657  *		if _SUNOS_VTOC_8 is defined
658  *			lowest 3 bits is taken as partition number
659  *			the rest is instance number
660  *		if _SUNOS_VTOC_16 is defined
661  *			lowest 6 bits is taken as partition number
662  *			the rest is instance number
663  *
664  *
665  * Return values:
666  *	0 	Success
667  * 	ENXIO 	creating minor nodes failed.
668  *	EINVAL  invalid arg, unsupported tg_ops version
669  */
670 int
671 cmlb_attach(dev_info_t *devi, cmlb_tg_ops_t *tgopsp, int device_type,
672     boolean_t is_removable, boolean_t is_hotpluggable, char *node_type,
673     int alter_behavior, cmlb_handle_t cmlbhandle, void *tg_cookie)
674 {
675 
676 	struct cmlb_lun	*cl = (struct cmlb_lun *)cmlbhandle;
677 	diskaddr_t	cap;
678 	int		status;
679 
680 	ASSERT(VALID_BOOLEAN(is_removable));
681 	ASSERT(VALID_BOOLEAN(is_hotpluggable));
682 
683 	if (tgopsp->tg_version < TG_DK_OPS_VERSION_1)
684 		return (EINVAL);
685 
686 	mutex_enter(CMLB_MUTEX(cl));
687 
688 	CMLB_DEVINFO(cl) = devi;
689 	cl->cmlb_tg_ops = tgopsp;
690 	cl->cl_device_type = device_type;
691 	cl->cl_is_removable = is_removable;
692 	cl->cl_is_hotpluggable = is_hotpluggable;
693 	cl->cl_node_type = node_type;
694 	cl->cl_sys_blocksize = DEV_BSIZE;
695 	cl->cl_f_geometry_is_valid = B_FALSE;
696 	cl->cl_def_labeltype = CMLB_LABEL_VTOC;
697 	cl->cl_alter_behavior = alter_behavior;
698 	cl->cl_reserved = -1;
699 	cl->cl_msglog_flag |= CMLB_ALLOW_2TB_WARN;
700 #if defined(__i386) || defined(__amd64)
701 	cl->cl_logical_drive_count = 0;
702 #endif
703 
704 	if (!is_removable) {
705 		mutex_exit(CMLB_MUTEX(cl));
706 		status = DK_TG_GETCAP(cl, &cap, tg_cookie);
707 		mutex_enter(CMLB_MUTEX(cl));
708 		if (status == 0 && cap > CMLB_EXTVTOC_LIMIT) {
709 			/* set default EFI if > 2TB */
710 			cl->cl_def_labeltype = CMLB_LABEL_EFI;
711 		}
712 	}
713 
714 	/* create minor nodes based on default label type */
715 	cl->cl_last_labeltype = CMLB_LABEL_UNDEF;
716 	cl->cl_cur_labeltype = CMLB_LABEL_UNDEF;
717 
718 	if (cmlb_create_minor_nodes(cl) != 0) {
719 		mutex_exit(CMLB_MUTEX(cl));
720 		return (ENXIO);
721 	}
722 
723 	/* Define the dynamic properties for devinfo spapshots. */
724 	i_ddi_prop_dyn_driver_set(CMLB_DEVINFO(cl), cmlb_prop_dyn);
725 
726 	cl->cl_state = CMLB_ATTACHED;
727 
728 	mutex_exit(CMLB_MUTEX(cl));
729 	return (0);
730 }
731 
732 /*
733  * cmlb_detach:
734  *
735  * Invalidate in-core labeling data and remove all minor nodes for
736  * the device associate with handle.
737  *
738  * Arguments:
739  *	cmlbhandle	cmlb handle associated with device.
740  *
741  *	tg_cookie	cookie from target driver to be passed back to target
742  *			driver when we call back to it through tg_ops.
743  *
744  */
745 /*ARGSUSED1*/
746 void
747 cmlb_detach(cmlb_handle_t cmlbhandle, void *tg_cookie)
748 {
749 	struct cmlb_lun *cl = (struct cmlb_lun *)cmlbhandle;
750 
751 	mutex_enter(CMLB_MUTEX(cl));
752 	cl->cl_def_labeltype = CMLB_LABEL_UNDEF;
753 	cl->cl_f_geometry_is_valid = B_FALSE;
754 	ddi_remove_minor_node(CMLB_DEVINFO(cl), NULL);
755 	i_ddi_prop_dyn_driver_set(CMLB_DEVINFO(cl), NULL);
756 	cl->cl_state = CMLB_INITED;
757 	mutex_exit(CMLB_MUTEX(cl));
758 }
759 
760 /*
761  * cmlb_validate:
762  *
763  *	Validates label.
764  *
765  * Arguments
766  *	cmlbhandle	cmlb handle associated with device.
767  *
768  *	flags		operation flags. used for verbosity control
769  *
770  *	tg_cookie	cookie from target driver to be passed back to target
771  *			driver when we call back to it through tg_ops.
772  *
773  *
774  * Notes:
775  *	If new label type is different from the current, adjust minor nodes
776  *	accordingly.
777  *
778  * Return values:
779  *	0		success
780  *			Note: having fdisk but no solaris partition is assumed
781  *			success.
782  *
783  *	ENOMEM		memory allocation failed
784  *	EIO		i/o errors during read or get capacity
785  * 	EACCESS		reservation conflicts
786  * 	EINVAL		label was corrupt, or no default label was assumed
787  *	ENXIO		invalid handle
788  */
789 int
790 cmlb_validate(cmlb_handle_t cmlbhandle, int flags, void *tg_cookie)
791 {
792 	struct cmlb_lun *cl = (struct cmlb_lun *)cmlbhandle;
793 	int 		rval;
794 	int  		ret = 0;
795 
796 	/*
797 	 * Temp work-around checking cl for NULL since there is a bug
798 	 * in sd_detach calling this routine from taskq_dispatch
799 	 * inited function.
800 	 */
801 	if (cl == NULL)
802 		return (ENXIO);
803 
804 	mutex_enter(CMLB_MUTEX(cl));
805 	if (cl->cl_state < CMLB_ATTACHED) {
806 		mutex_exit(CMLB_MUTEX(cl));
807 		return (ENXIO);
808 	}
809 
810 	rval = cmlb_validate_geometry((struct cmlb_lun *)cmlbhandle, B_TRUE,
811 	    flags, tg_cookie);
812 
813 	if (rval == ENOTSUP) {
814 		if (cl->cl_f_geometry_is_valid) {
815 			cl->cl_cur_labeltype = CMLB_LABEL_EFI;
816 			ret = 0;
817 		} else {
818 			ret = EINVAL;
819 		}
820 	} else {
821 		ret = rval;
822 		if (ret == 0)
823 			cl->cl_cur_labeltype = CMLB_LABEL_VTOC;
824 	}
825 
826 	if (ret == 0)
827 		(void) cmlb_create_minor_nodes(cl);
828 
829 	mutex_exit(CMLB_MUTEX(cl));
830 	return (ret);
831 }
832 
833 /*
834  * cmlb_invalidate:
835  *	Invalidate in core label data
836  *
837  * Arguments:
838  *	cmlbhandle	cmlb handle associated with device.
839  *	tg_cookie	cookie from target driver to be passed back to target
840  *			driver when we call back to it through tg_ops.
841  */
842 /*ARGSUSED1*/
843 void
844 cmlb_invalidate(cmlb_handle_t cmlbhandle, void *tg_cookie)
845 {
846 	struct cmlb_lun *cl = (struct cmlb_lun *)cmlbhandle;
847 
848 	if (cl == NULL)
849 		return;
850 
851 	mutex_enter(CMLB_MUTEX(cl));
852 	cl->cl_f_geometry_is_valid = B_FALSE;
853 	mutex_exit(CMLB_MUTEX(cl));
854 }
855 
856 /*
857  * cmlb_is_valid
858  * 	Get status on whether the incore label/geom data is valid
859  *
860  * Arguments:
861  *	cmlbhandle      cmlb handle associated with device.
862  *
863  * Return values:
864  *	B_TRUE if incore label/geom data is valid.
865  *	B_FALSE otherwise.
866  *
867  */
868 
869 
870 boolean_t
871 cmlb_is_valid(cmlb_handle_t cmlbhandle)
872 {
873 	struct cmlb_lun *cl = (struct cmlb_lun *)cmlbhandle;
874 
875 	if (cmlbhandle == NULL)
876 		return (B_FALSE);
877 
878 	return (cl->cl_f_geometry_is_valid);
879 
880 }
881 
882 
883 
884 /*
885  * cmlb_close:
886  *
887  * Close the device, revert to a default label minor node for the device,
888  * if it is removable.
889  *
890  * Arguments:
891  *	cmlbhandle	cmlb handle associated with device.
892  *
893  *	tg_cookie	cookie from target driver to be passed back to target
894  *			driver when we call back to it through tg_ops.
895  * Return values:
896  *	0	Success
897  * 	ENXIO	Re-creating minor node failed.
898  */
899 /*ARGSUSED1*/
900 int
901 cmlb_close(cmlb_handle_t cmlbhandle, void *tg_cookie)
902 {
903 	struct cmlb_lun *cl = (struct cmlb_lun *)cmlbhandle;
904 
905 	mutex_enter(CMLB_MUTEX(cl));
906 	cl->cl_f_geometry_is_valid = B_FALSE;
907 
908 	/* revert to default minor node for this device */
909 	if (ISREMOVABLE(cl)) {
910 		cl->cl_cur_labeltype = CMLB_LABEL_UNDEF;
911 		(void) cmlb_create_minor_nodes(cl);
912 	}
913 
914 	mutex_exit(CMLB_MUTEX(cl));
915 	return (0);
916 }
917 
918 /*
919  * cmlb_get_devid_block:
920  *	 get the block number where device id is stored.
921  *
922  * Arguments:
923  *	cmlbhandle	cmlb handle associated with device.
924  *	devidblockp	pointer to block number.
925  *	tg_cookie	cookie from target driver to be passed back to target
926  *			driver when we call back to it through tg_ops.
927  *
928  * Notes:
929  *	It stores the block number of device id in the area pointed to
930  *	by devidblockp.
931  * 	with the block number of device id.
932  *
933  * Return values:
934  *	0	success
935  *	EINVAL 	device id does not apply to current label type.
936  */
937 /*ARGSUSED2*/
938 int
939 cmlb_get_devid_block(cmlb_handle_t cmlbhandle, diskaddr_t *devidblockp,
940     void *tg_cookie)
941 {
942 	daddr_t			spc, blk, head, cyl;
943 	struct cmlb_lun *cl = (struct cmlb_lun *)cmlbhandle;
944 
945 	mutex_enter(CMLB_MUTEX(cl));
946 	if (cl->cl_state < CMLB_ATTACHED) {
947 		mutex_exit(CMLB_MUTEX(cl));
948 		return (EINVAL);
949 	}
950 
951 	if ((!cl->cl_f_geometry_is_valid) ||
952 	    (cl->cl_solaris_size < DK_LABEL_LOC)) {
953 		mutex_exit(CMLB_MUTEX(cl));
954 		return (EINVAL);
955 	}
956 
957 	if (cl->cl_cur_labeltype == CMLB_LABEL_EFI) {
958 		if (cl->cl_reserved != -1) {
959 			blk = cl->cl_map[cl->cl_reserved].dkl_cylno;
960 		} else {
961 			mutex_exit(CMLB_MUTEX(cl));
962 			return (EINVAL);
963 		}
964 	} else {
965 		/* if the disk is unlabeled, don't write a devid to it */
966 		if (cl->cl_label_from_media != CMLB_LABEL_VTOC) {
967 			mutex_exit(CMLB_MUTEX(cl));
968 			return (EINVAL);
969 		}
970 
971 		/* this geometry doesn't allow us to write a devid */
972 		if (cl->cl_g.dkg_acyl < 2) {
973 			mutex_exit(CMLB_MUTEX(cl));
974 			return (EINVAL);
975 		}
976 
977 		/*
978 		 * Subtract 2 guarantees that the next to last cylinder
979 		 * is used
980 		 */
981 		cyl  = cl->cl_g.dkg_ncyl  + cl->cl_g.dkg_acyl - 2;
982 		spc  = cl->cl_g.dkg_nhead * cl->cl_g.dkg_nsect;
983 		head = cl->cl_g.dkg_nhead - 1;
984 		blk  = cl->cl_solaris_offset +
985 		    (cyl * (spc - cl->cl_g.dkg_apc)) +
986 		    (head * cl->cl_g.dkg_nsect) + 1;
987 	}
988 
989 	*devidblockp = blk;
990 	mutex_exit(CMLB_MUTEX(cl));
991 	return (0);
992 }
993 
994 /*
995  * cmlb_partinfo:
996  *	Get partition info for specified partition number.
997  *
998  * Arguments:
999  *	cmlbhandle	cmlb handle associated with device.
1000  *	part		partition number
1001  *	nblocksp	pointer to number of blocks
1002  *	startblockp	pointer to starting block
1003  *	partnamep	pointer to name of partition
1004  *	tagp		pointer to tag info
1005  *	tg_cookie	cookie from target driver to be passed back to target
1006  *			driver when we call back to it through tg_ops.
1007  *
1008  *
1009  * Notes:
1010  *	If in-core label is not valid, this functions tries to revalidate
1011  *	the label. If label is valid, it stores the total number of blocks
1012  *	in this partition in the area pointed to by nblocksp, starting
1013  *	block number in area pointed to by startblockp,  pointer to partition
1014  *	name in area pointed to by partnamep, and tag value in area
1015  *	pointed by tagp.
1016  *	For EFI labels, tag value will be set to 0.
1017  *
1018  *	For all nblocksp, startblockp and partnamep, tagp, a value of NULL
1019  *	indicates the corresponding info is not requested.
1020  *
1021  *
1022  * Return values:
1023  *	0	success
1024  *	EINVAL  no valid label or requested partition number is invalid.
1025  *
1026  */
1027 int
1028 cmlb_partinfo(cmlb_handle_t cmlbhandle, int part, diskaddr_t *nblocksp,
1029     diskaddr_t *startblockp, char **partnamep, uint16_t *tagp, void *tg_cookie)
1030 {
1031 
1032 	struct cmlb_lun *cl = (struct cmlb_lun *)cmlbhandle;
1033 	int rval;
1034 #if defined(__i386) || defined(__amd64)
1035 	int ext_part;
1036 #endif
1037 
1038 	ASSERT(cl != NULL);
1039 	mutex_enter(CMLB_MUTEX(cl));
1040 	if (cl->cl_state < CMLB_ATTACHED) {
1041 		mutex_exit(CMLB_MUTEX(cl));
1042 		return (EINVAL);
1043 	}
1044 
1045 	if (part  < 0 || part >= MAXPART) {
1046 		rval = EINVAL;
1047 	} else {
1048 		if (!cl->cl_f_geometry_is_valid)
1049 			(void) cmlb_validate_geometry((struct cmlb_lun *)cl,
1050 			    B_FALSE, 0, tg_cookie);
1051 
1052 #if defined(_SUNOS_VTOC_16)
1053 		if (((!cl->cl_f_geometry_is_valid) ||
1054 		    (part < NDKMAP && cl->cl_solaris_size == 0)) &&
1055 		    (part != P0_RAW_DISK)) {
1056 #else
1057 		if ((!cl->cl_f_geometry_is_valid) ||
1058 		    (part < NDKMAP && cl->cl_solaris_size == 0)) {
1059 #endif
1060 			rval = EINVAL;
1061 		} else {
1062 			if (startblockp != NULL)
1063 				*startblockp = (diskaddr_t)cl->cl_offset[part];
1064 
1065 			if (nblocksp != NULL)
1066 				*nblocksp = (diskaddr_t)
1067 				    cl->cl_map[part].dkl_nblk;
1068 
1069 			if (tagp != NULL)
1070 				if (cl->cl_cur_labeltype == CMLB_LABEL_EFI)
1071 					*tagp = V_UNASSIGNED;
1072 				else
1073 					*tagp = cl->cl_vtoc.v_part[part].p_tag;
1074 			rval = 0;
1075 		}
1076 
1077 		/* consistent with behavior of sd for getting minor name */
1078 		if (partnamep != NULL) {
1079 #if defined(__i386) || defined(__amd64)
1080 #if defined(_FIRMWARE_NEEDS_FDISK)
1081 		if (part > FDISK_P4) {
1082 			ext_part = part-FDISK_P4-1;
1083 			*partnamep = dk_ext_minor_data[ext_part].name;
1084 		} else
1085 #endif
1086 #endif
1087 			*partnamep = dk_minor_data[part].name;
1088 		}
1089 
1090 	}
1091 
1092 	mutex_exit(CMLB_MUTEX(cl));
1093 	return (rval);
1094 }
1095 
1096 /*
1097  * cmlb_efi_label_capacity:
1098  *	Get capacity stored in EFI disk label.
1099  *
1100  * Arguments:
1101  *	cmlbhandle	cmlb handle associated with device.
1102  *	capacity	pointer to capacity stored in EFI disk label.
1103  *	tg_cookie	cookie from target driver to be passed back to target
1104  *			driver when we call back to it through tg_ops.
1105  *
1106  *
1107  * Notes:
1108  *	If in-core label is not valid, this functions tries to revalidate
1109  *	the label. If label is valid and is an EFI label, it stores the capacity
1110  *      in disk label in the area pointed to by capacity.
1111  *
1112  *
1113  * Return values:
1114  *	0	success
1115  *	EINVAL  no valid EFI label or capacity is NULL.
1116  *
1117  */
1118 int
1119 cmlb_efi_label_capacity(cmlb_handle_t cmlbhandle, diskaddr_t *capacity,
1120     void *tg_cookie)
1121 {
1122 	struct cmlb_lun *cl = (struct cmlb_lun *)cmlbhandle;
1123 	int rval;
1124 
1125 	ASSERT(cl != NULL);
1126 	mutex_enter(CMLB_MUTEX(cl));
1127 	if (cl->cl_state < CMLB_ATTACHED) {
1128 		mutex_exit(CMLB_MUTEX(cl));
1129 		return (EINVAL);
1130 	}
1131 
1132 	if (!cl->cl_f_geometry_is_valid)
1133 		(void) cmlb_validate_geometry((struct cmlb_lun *)cl, B_FALSE,
1134 		    0, tg_cookie);
1135 
1136 	if ((!cl->cl_f_geometry_is_valid) || (capacity == NULL) ||
1137 	    (cl->cl_cur_labeltype != CMLB_LABEL_EFI)) {
1138 		rval = EINVAL;
1139 	} else {
1140 		*capacity = (diskaddr_t)cl->cl_map[WD_NODE].dkl_nblk;
1141 		rval = 0;
1142 	}
1143 
1144 	mutex_exit(CMLB_MUTEX(cl));
1145 	return (rval);
1146 }
1147 
1148 /* Caller should make sure Test Unit Ready succeeds before calling this. */
1149 /*ARGSUSED*/
1150 int
1151 cmlb_ioctl(cmlb_handle_t cmlbhandle, dev_t dev, int cmd, intptr_t arg,
1152     int flag, cred_t *cred_p, int *rval_p, void *tg_cookie)
1153 {
1154 
1155 	int err;
1156 	struct cmlb_lun *cl;
1157 
1158 	cl = (struct cmlb_lun *)cmlbhandle;
1159 
1160 	ASSERT(cl != NULL);
1161 
1162 	mutex_enter(CMLB_MUTEX(cl));
1163 	if (cl->cl_state < CMLB_ATTACHED) {
1164 		mutex_exit(CMLB_MUTEX(cl));
1165 		return (EIO);
1166 	}
1167 
1168 	switch (cmd) {
1169 		case DKIOCSEXTVTOC:
1170 		case DKIOCSGEOM:
1171 		case DKIOCSETEFI:
1172 		case DKIOCSMBOOT:
1173 #if defined(__i386) || defined(__amd64)
1174 		case DKIOCSETEXTPART:
1175 #endif
1176 			break;
1177 		case DKIOCSVTOC:
1178 #if defined(__i386) || defined(__amd64)
1179 		case DKIOCPARTINFO:
1180 #endif
1181 			if (cl->cl_blockcount > CMLB_OLDVTOC_LIMIT) {
1182 				mutex_exit(CMLB_MUTEX(cl));
1183 				return (EOVERFLOW);
1184 			}
1185 			break;
1186 		default:
1187 			(void) cmlb_validate_geometry(cl, 1, CMLB_SILENT,
1188 			    tg_cookie);
1189 
1190 			switch (cmd) {
1191 			case DKIOCGVTOC:
1192 			case DKIOCGAPART:
1193 			case DKIOCSAPART:
1194 
1195 				if (cl->cl_label_from_media == CMLB_LABEL_EFI) {
1196 					/* GPT label on disk */
1197 					mutex_exit(CMLB_MUTEX(cl));
1198 					return (ENOTSUP);
1199 				} else if
1200 				    (cl->cl_blockcount > CMLB_OLDVTOC_LIMIT) {
1201 					mutex_exit(CMLB_MUTEX(cl));
1202 					return (EOVERFLOW);
1203 				}
1204 				break;
1205 
1206 			case DKIOCGGEOM:
1207 				if (cl->cl_label_from_media == CMLB_LABEL_EFI) {
1208 					/* GPT label on disk */
1209 					mutex_exit(CMLB_MUTEX(cl));
1210 					return (ENOTSUP);
1211 				}
1212 				break;
1213 			default:
1214 				break;
1215 			}
1216 	}
1217 
1218 	mutex_exit(CMLB_MUTEX(cl));
1219 
1220 	switch (cmd) {
1221 	case DKIOCGGEOM:
1222 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCGGEOM\n");
1223 		err = cmlb_dkio_get_geometry(cl, (caddr_t)arg, flag, tg_cookie);
1224 		break;
1225 
1226 	case DKIOCSGEOM:
1227 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCSGEOM\n");
1228 		err = cmlb_dkio_set_geometry(cl, (caddr_t)arg, flag);
1229 		break;
1230 
1231 	case DKIOCGAPART:
1232 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCGAPART\n");
1233 		err = cmlb_dkio_get_partition(cl, (caddr_t)arg,
1234 		    flag, tg_cookie);
1235 		break;
1236 
1237 	case DKIOCSAPART:
1238 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCSAPART\n");
1239 		err = cmlb_dkio_set_partition(cl, (caddr_t)arg, flag);
1240 		break;
1241 
1242 	case DKIOCGVTOC:
1243 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCGVTOC\n");
1244 		err = cmlb_dkio_get_vtoc(cl, (caddr_t)arg, flag, tg_cookie);
1245 		break;
1246 
1247 	case DKIOCGEXTVTOC:
1248 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCGVTOC\n");
1249 		err = cmlb_dkio_get_extvtoc(cl, (caddr_t)arg, flag, tg_cookie);
1250 		break;
1251 
1252 	case DKIOCGETEFI:
1253 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCGETEFI\n");
1254 		err = cmlb_dkio_get_efi(cl, (caddr_t)arg, flag, tg_cookie);
1255 		break;
1256 
1257 	case DKIOCPARTITION:
1258 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCPARTITION\n");
1259 		err = cmlb_dkio_partition(cl, (caddr_t)arg, flag, tg_cookie);
1260 		break;
1261 
1262 	case DKIOCSVTOC:
1263 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCSVTOC\n");
1264 		err = cmlb_dkio_set_vtoc(cl, dev, (caddr_t)arg, flag,
1265 		    tg_cookie);
1266 		break;
1267 
1268 	case DKIOCSEXTVTOC:
1269 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCSVTOC\n");
1270 		err = cmlb_dkio_set_extvtoc(cl, dev, (caddr_t)arg, flag,
1271 		    tg_cookie);
1272 		break;
1273 
1274 	case DKIOCSETEFI:
1275 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCSETEFI\n");
1276 		err = cmlb_dkio_set_efi(cl, dev, (caddr_t)arg, flag, tg_cookie);
1277 		break;
1278 
1279 	case DKIOCGMBOOT:
1280 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCGMBOOT\n");
1281 		err = cmlb_dkio_get_mboot(cl, (caddr_t)arg, flag, tg_cookie);
1282 		break;
1283 
1284 	case DKIOCSMBOOT:
1285 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCSMBOOT\n");
1286 		err = cmlb_dkio_set_mboot(cl, (caddr_t)arg, flag, tg_cookie);
1287 		break;
1288 	case DKIOCG_PHYGEOM:
1289 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCG_PHYGEOM\n");
1290 #if defined(__i386) || defined(__amd64)
1291 		err = cmlb_dkio_get_phygeom(cl, (caddr_t)arg, flag, tg_cookie);
1292 #else
1293 		err = ENOTTY;
1294 #endif
1295 		break;
1296 	case DKIOCG_VIRTGEOM:
1297 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCG_VIRTGEOM\n");
1298 #if defined(__i386) || defined(__amd64)
1299 		err = cmlb_dkio_get_virtgeom(cl, (caddr_t)arg, flag);
1300 #else
1301 		err = ENOTTY;
1302 #endif
1303 		break;
1304 	case DKIOCPARTINFO:
1305 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCPARTINFO");
1306 #if defined(__i386) || defined(__amd64)
1307 		err = cmlb_dkio_partinfo(cl, dev, (caddr_t)arg, flag);
1308 #else
1309 		err = ENOTTY;
1310 #endif
1311 		break;
1312 	case DKIOCEXTPARTINFO:
1313 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCPARTINFO");
1314 #if defined(__i386) || defined(__amd64)
1315 		err = cmlb_dkio_extpartinfo(cl, dev, (caddr_t)arg, flag);
1316 #else
1317 		err = ENOTTY;
1318 #endif
1319 		break;
1320 #if defined(__i386) || defined(__amd64)
1321 	case DKIOCSETEXTPART:
1322 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCSETEXTPART");
1323 		err = cmlb_dkio_set_ext_part(cl, (caddr_t)arg, flag, tg_cookie);
1324 		break;
1325 #endif
1326 	default:
1327 		err = ENOTTY;
1328 
1329 	}
1330 
1331 	/*
1332 	 * An ioctl that succeeds and changed ('set') size(9P) information
1333 	 * needs to invalidate the cached devinfo snapshot to avoid having
1334 	 * old information being returned in a snapshots.
1335 	 *
1336 	 * NB: When available, call ddi_change_minor_node() to clear
1337 	 * SSIZEVALID in specfs vnodes via spec_size_invalidate().
1338 	 */
1339 	if (err == 0) {
1340 		switch (cmd) {
1341 		case DKIOCSGEOM:
1342 		case DKIOCSAPART:
1343 		case DKIOCSVTOC:
1344 		case DKIOCSEXTVTOC:
1345 		case DKIOCSETEFI:
1346 			i_ddi_prop_dyn_cache_invalidate(CMLB_DEVINFO(cl),
1347 			    i_ddi_prop_dyn_driver_get(CMLB_DEVINFO(cl)));
1348 		}
1349 	}
1350 	return (err);
1351 }
1352 
1353 dev_t
1354 cmlb_make_device(struct cmlb_lun *cl)
1355 {
1356 	return (makedevice(ddi_driver_major(CMLB_DEVINFO(cl)),
1357 	    ddi_get_instance(CMLB_DEVINFO(cl)) << CMLBUNIT_SHIFT));
1358 }
1359 
1360 /*
1361  * Function: cmlb_check_update_blockcount
1362  *
1363  * Description: If current capacity value is invalid, obtains the
1364  *		current capacity from target driver.
1365  *
1366  * Return Code: 0	success
1367  *		EIO	failure
1368  */
1369 static int
1370 cmlb_check_update_blockcount(struct cmlb_lun *cl, void *tg_cookie)
1371 {
1372 	int status;
1373 	diskaddr_t capacity;
1374 	uint32_t lbasize;
1375 
1376 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
1377 
1378 	if (cl->cl_f_geometry_is_valid)
1379 		return (0);
1380 
1381 	mutex_exit(CMLB_MUTEX(cl));
1382 	status = DK_TG_GETCAP(cl, &capacity, tg_cookie);
1383 	if (status != 0) {
1384 		mutex_enter(CMLB_MUTEX(cl));
1385 		return (EIO);
1386 	}
1387 
1388 	status = DK_TG_GETBLOCKSIZE(cl, &lbasize, tg_cookie);
1389 	mutex_enter(CMLB_MUTEX(cl));
1390 	if (status != 0)
1391 		return (EIO);
1392 
1393 	if ((capacity != 0) && (lbasize != 0)) {
1394 		cl->cl_blockcount = capacity;
1395 		cl->cl_tgt_blocksize = lbasize;
1396 		if (!cl->cl_is_removable) {
1397 			cl->cl_sys_blocksize = lbasize;
1398 		}
1399 		return (0);
1400 	} else {
1401 		return (EIO);
1402 	}
1403 }
1404 
1405 static int
1406 cmlb_create_minor(dev_info_t *dip, char *name, int spec_type,
1407     minor_t minor_num, char *node_type, int flag, boolean_t internal)
1408 {
1409 	ASSERT(VALID_BOOLEAN(internal));
1410 
1411 	if (internal)
1412 		return (ddi_create_internal_pathname(dip,
1413 		    name, spec_type, minor_num));
1414 	else
1415 		return (ddi_create_minor_node(dip,
1416 		    name, spec_type, minor_num, node_type, flag));
1417 }
1418 
1419 /*
1420  *    Function: cmlb_create_minor_nodes
1421  *
1422  * Description: Create or adjust the minor device nodes for the instance.
1423  * 		Minor nodes are created based on default label type,
1424  *		current label type and last label type we created
1425  *		minor nodes based on.
1426  *
1427  *
1428  *   Arguments: cl - driver soft state (unit) structure
1429  *
1430  * Return Code: 0 success
1431  *		ENXIO	failure.
1432  *
1433  *     Context: Kernel thread context
1434  */
1435 static int
1436 cmlb_create_minor_nodes(struct cmlb_lun *cl)
1437 {
1438 	struct driver_minor_data	*dmdp;
1439 	int				instance;
1440 	char				name[48];
1441 	cmlb_label_t			newlabeltype;
1442 	boolean_t			internal;
1443 
1444 	ASSERT(cl != NULL);
1445 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
1446 
1447 	internal = VOID2BOOLEAN(
1448 	    (cl->cl_alter_behavior & (CMLB_INTERNAL_MINOR_NODES)) != 0);
1449 
1450 	/* check the most common case */
1451 	if (cl->cl_cur_labeltype != CMLB_LABEL_UNDEF &&
1452 	    cl->cl_last_labeltype == cl->cl_cur_labeltype) {
1453 		/* do nothing */
1454 		return (0);
1455 	}
1456 
1457 	if (cl->cl_def_labeltype == CMLB_LABEL_UNDEF) {
1458 		/* we should never get here */
1459 		return (ENXIO);
1460 	}
1461 
1462 	if (cl->cl_last_labeltype == CMLB_LABEL_UNDEF) {
1463 		/* first time during attach */
1464 		newlabeltype = cl->cl_def_labeltype;
1465 
1466 		instance = ddi_get_instance(CMLB_DEVINFO(cl));
1467 
1468 		/* Create all the minor nodes for this target. */
1469 		dmdp = (newlabeltype == CMLB_LABEL_EFI) ? dk_minor_data_efi :
1470 		    dk_minor_data;
1471 		while (dmdp->name != NULL) {
1472 
1473 			(void) sprintf(name, "%s", dmdp->name);
1474 
1475 			if (cmlb_create_minor(CMLB_DEVINFO(cl), name,
1476 			    dmdp->type,
1477 			    (instance << CMLBUNIT_SHIFT) | dmdp->minor,
1478 			    cl->cl_node_type, NULL, internal) == DDI_FAILURE) {
1479 				/*
1480 				 * Clean up any nodes that may have been
1481 				 * created, in case this fails in the middle
1482 				 * of the loop.
1483 				 */
1484 				ddi_remove_minor_node(CMLB_DEVINFO(cl), NULL);
1485 				return (ENXIO);
1486 			}
1487 			dmdp++;
1488 		}
1489 		cl->cl_last_labeltype = newlabeltype;
1490 		return (0);
1491 	}
1492 
1493 	/* Not first time  */
1494 	if (cl->cl_cur_labeltype == CMLB_LABEL_UNDEF) {
1495 		if (cl->cl_last_labeltype != cl->cl_def_labeltype) {
1496 			/* close time, revert to default. */
1497 			newlabeltype = cl->cl_def_labeltype;
1498 		} else {
1499 			/*
1500 			 * do nothing since the type for which we last created
1501 			 * nodes matches the default
1502 			 */
1503 			return (0);
1504 		}
1505 	} else {
1506 		if (cl->cl_cur_labeltype != cl->cl_last_labeltype) {
1507 			/* We are not closing, use current label type */
1508 			newlabeltype = cl->cl_cur_labeltype;
1509 		} else {
1510 			/*
1511 			 * do nothing since the type for which we last created
1512 			 * nodes matches the current label type
1513 			 */
1514 			return (0);
1515 		}
1516 	}
1517 
1518 	instance = ddi_get_instance(CMLB_DEVINFO(cl));
1519 
1520 	/*
1521 	 * Currently we only fix up the s7 node when we are switching
1522 	 * label types from or to EFI. This is consistent with
1523 	 * current behavior of sd.
1524 	 */
1525 	if (newlabeltype == CMLB_LABEL_EFI &&
1526 	    cl->cl_last_labeltype != CMLB_LABEL_EFI) {
1527 		/* from vtoc to EFI */
1528 		ddi_remove_minor_node(CMLB_DEVINFO(cl), "h");
1529 		ddi_remove_minor_node(CMLB_DEVINFO(cl), "h,raw");
1530 		(void) cmlb_create_minor(CMLB_DEVINFO(cl), "wd",
1531 		    S_IFBLK, (instance << CMLBUNIT_SHIFT) | WD_NODE,
1532 		    cl->cl_node_type, NULL, internal);
1533 		(void) cmlb_create_minor(CMLB_DEVINFO(cl), "wd,raw",
1534 		    S_IFCHR, (instance << CMLBUNIT_SHIFT) | WD_NODE,
1535 		    cl->cl_node_type, NULL, internal);
1536 	} else {
1537 		/* from efi to vtoc */
1538 		ddi_remove_minor_node(CMLB_DEVINFO(cl), "wd");
1539 		ddi_remove_minor_node(CMLB_DEVINFO(cl), "wd,raw");
1540 		(void) cmlb_create_minor(CMLB_DEVINFO(cl), "h",
1541 		    S_IFBLK, (instance << CMLBUNIT_SHIFT) | WD_NODE,
1542 		    cl->cl_node_type, NULL, internal);
1543 		(void) cmlb_create_minor(CMLB_DEVINFO(cl), "h,raw",
1544 		    S_IFCHR, (instance << CMLBUNIT_SHIFT) | WD_NODE,
1545 		    cl->cl_node_type, NULL, internal);
1546 	}
1547 
1548 	cl->cl_last_labeltype = newlabeltype;
1549 	return (0);
1550 }
1551 
1552 /*
1553  *    Function: cmlb_validate_geometry
1554  *
1555  * Description: Read the label from the disk (if present). Update the unit's
1556  *		geometry and vtoc information from the data in the label.
1557  *		Verify that the label is valid.
1558  *
1559  *   Arguments:
1560  *	cl		driver soft state (unit) structure
1561  *
1562  *	forcerevalid	force revalidation even if we are already valid.
1563  *	flags		operation flags from target driver. Used for verbosity
1564  *			control	at this time.
1565  *	tg_cookie	cookie from target driver to be passed back to target
1566  *			driver when we call back to it through tg_ops.
1567  *
1568  * Return Code: 0 - Successful completion
1569  *		EINVAL  - Invalid value in cl->cl_tgt_blocksize or
1570  *			  cl->cl_blockcount; or label on disk is corrupted
1571  *			  or unreadable.
1572  *		EACCES  - Reservation conflict at the device.
1573  *		ENOMEM  - Resource allocation error
1574  *		ENOTSUP - geometry not applicable
1575  *
1576  *     Context: Kernel thread only (can sleep).
1577  */
1578 static int
1579 cmlb_validate_geometry(struct cmlb_lun *cl, boolean_t forcerevalid, int flags,
1580     void *tg_cookie)
1581 {
1582 	int		label_error = 0;
1583 	diskaddr_t	capacity;
1584 	int		count;
1585 
1586 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
1587 	ASSERT(VALID_BOOLEAN(forcerevalid));
1588 
1589 	if ((cl->cl_f_geometry_is_valid) && (!forcerevalid)) {
1590 		if (cl->cl_cur_labeltype == CMLB_LABEL_EFI)
1591 			return (ENOTSUP);
1592 		return (0);
1593 	}
1594 
1595 	if (cmlb_check_update_blockcount(cl, tg_cookie) != 0)
1596 		return (EIO);
1597 
1598 	capacity = cl->cl_blockcount;
1599 
1600 #if defined(_SUNOS_VTOC_16)
1601 	/*
1602 	 * Set up the "whole disk" fdisk partition; this should always
1603 	 * exist, regardless of whether the disk contains an fdisk table
1604 	 * or vtoc.
1605 	 */
1606 	cl->cl_map[P0_RAW_DISK].dkl_cylno = 0;
1607 	cl->cl_offset[P0_RAW_DISK] = 0;
1608 	/*
1609 	 * note if capacity > int32_max(1TB) we are in 64bit environment
1610 	 * so no truncation happens
1611 	 */
1612 	cl->cl_map[P0_RAW_DISK].dkl_nblk  = capacity;
1613 #endif
1614 	/*
1615 	 * Refresh the logical and physical geometry caches.
1616 	 * (data from MODE SENSE format/rigid disk geometry pages,
1617 	 * and scsi_ifgetcap("geometry").
1618 	 */
1619 	cmlb_resync_geom_caches(cl, capacity, tg_cookie);
1620 
1621 	cl->cl_label_from_media = CMLB_LABEL_UNDEF;
1622 	label_error = cmlb_use_efi(cl, capacity, flags, tg_cookie);
1623 	if (label_error == 0) {
1624 
1625 		/* found a valid EFI label */
1626 		cmlb_dbg(CMLB_TRACE, cl,
1627 		    "cmlb_validate_geometry: found EFI label\n");
1628 		/*
1629 		 * solaris_size and geometry_is_valid are set in
1630 		 * cmlb_use_efi
1631 		 */
1632 		return (ENOTSUP);
1633 	}
1634 
1635 	/* NO EFI label found */
1636 
1637 	if (capacity > CMLB_EXTVTOC_LIMIT) {
1638 		if (label_error == ESRCH) {
1639 			/*
1640 			 * they've configured a LUN over 2TB, but used
1641 			 * format.dat to restrict format's view of the
1642 			 * capacity to be under 2TB in some earlier Solaris
1643 			 * release.
1644 			 */
1645 			/* i.e > 2TB with a VTOC < 2TB */
1646 			if (!(flags & CMLB_SILENT) &&
1647 			    (cl->cl_msglog_flag & CMLB_ALLOW_2TB_WARN)) {
1648 
1649 				cmlb_log(CMLB_DEVINFO(cl), CMLB_LABEL(cl),
1650 				    CE_NOTE, "!Disk (%s%d) is limited to 2 TB "
1651 				    "due to VTOC label. To use the full "
1652 				    "capacity of the disk, use format(1M) to "
1653 				    "relabel the disk with EFI/GPT label.\n",
1654 				    CMLB_LABEL(cl),
1655 				    ddi_get_instance(CMLB_DEVINFO(cl)));
1656 
1657 				cl->cl_msglog_flag &= ~CMLB_ALLOW_2TB_WARN;
1658 			}
1659 		} else {
1660 				return (ENOTSUP);
1661 		}
1662 	}
1663 
1664 	label_error = 0;
1665 
1666 	/*
1667 	 * at this point it is either labeled with a VTOC or it is
1668 	 * under 1TB (<= 1TB actually for off-by-1)
1669 	 */
1670 
1671 	/*
1672 	 * Only DIRECT ACCESS devices will have Scl labels.
1673 	 * CD's supposedly have a Scl label, too
1674 	 */
1675 	if (cl->cl_device_type == DTYPE_DIRECT || ISREMOVABLE(cl)) {
1676 		struct	dk_label *dkl;
1677 		offset_t label_addr;
1678 		int	rval;
1679 		size_t	buffer_size;
1680 
1681 		/*
1682 		 * Note: This will set up cl->cl_solaris_size and
1683 		 * cl->cl_solaris_offset.
1684 		 */
1685 		rval = cmlb_read_fdisk(cl, capacity, tg_cookie);
1686 		if ((rval != 0) && !ISCD(cl)) {
1687 			ASSERT(mutex_owned(CMLB_MUTEX(cl)));
1688 			return (rval);
1689 		}
1690 
1691 		if (cl->cl_solaris_size <= DK_LABEL_LOC) {
1692 			/*
1693 			 * Found fdisk table but no Solaris partition entry,
1694 			 * so don't call cmlb_uselabel() and don't create
1695 			 * a default label.
1696 			 */
1697 			label_error = 0;
1698 			cl->cl_f_geometry_is_valid = B_TRUE;
1699 			goto no_solaris_partition;
1700 		}
1701 
1702 		label_addr = (daddr_t)(cl->cl_solaris_offset + DK_LABEL_LOC);
1703 
1704 		buffer_size = cl->cl_sys_blocksize;
1705 
1706 		cmlb_dbg(CMLB_TRACE, cl, "cmlb_validate_geometry: "
1707 		    "label_addr: 0x%x allocation size: 0x%x\n",
1708 		    label_addr, buffer_size);
1709 
1710 		if ((dkl = kmem_zalloc(buffer_size, KM_NOSLEEP)) == NULL)
1711 			return (ENOMEM);
1712 
1713 		mutex_exit(CMLB_MUTEX(cl));
1714 		rval = DK_TG_READ(cl, dkl, label_addr, buffer_size, tg_cookie);
1715 		mutex_enter(CMLB_MUTEX(cl));
1716 
1717 		switch (rval) {
1718 		case 0:
1719 			/*
1720 			 * cmlb_uselabel will establish that the geometry
1721 			 * is valid.
1722 			 */
1723 			if (cmlb_uselabel(cl,
1724 			    (struct dk_label *)(uintptr_t)dkl, flags) !=
1725 			    CMLB_LABEL_IS_VALID) {
1726 				label_error = EINVAL;
1727 			} else
1728 				cl->cl_label_from_media = CMLB_LABEL_VTOC;
1729 			break;
1730 		case EACCES:
1731 			label_error = EACCES;
1732 			break;
1733 		default:
1734 			label_error = EINVAL;
1735 			break;
1736 		}
1737 
1738 		kmem_free(dkl, buffer_size);
1739 	}
1740 
1741 	/*
1742 	 * If a valid label was not found, AND if no reservation conflict
1743 	 * was detected, then go ahead and create a default label (4069506).
1744 	 *
1745 	 * Note: currently, for VTOC_8 devices, the default label is created
1746 	 * for removables and hotpluggables only.  For VTOC_16 devices, the
1747 	 * default label will be created for all devices.
1748 	 * (see cmlb_build_default_label)
1749 	 */
1750 #if defined(_SUNOS_VTOC_8)
1751 	if ((ISREMOVABLE(cl) || ISHOTPLUGGABLE(cl)) &&
1752 	    (label_error != EACCES)) {
1753 #elif defined(_SUNOS_VTOC_16)
1754 	if (label_error != EACCES) {
1755 #endif
1756 		if (!cl->cl_f_geometry_is_valid) {
1757 			cmlb_build_default_label(cl, tg_cookie);
1758 		}
1759 		label_error = 0;
1760 	}
1761 
1762 no_solaris_partition:
1763 
1764 #if defined(_SUNOS_VTOC_16)
1765 	/*
1766 	 * If we have valid geometry, set up the remaining fdisk partitions.
1767 	 * Note that dkl_cylno is not used for the fdisk map entries, so
1768 	 * we set it to an entirely bogus value.
1769 	 */
1770 	for (count = 0; count < FDISK_PARTS; count++) {
1771 		cl->cl_map[FDISK_P1 + count].dkl_cylno = UINT16_MAX;
1772 		cl->cl_map[FDISK_P1 + count].dkl_nblk =
1773 		    cl->cl_fmap[count].fmap_nblk;
1774 
1775 		cl->cl_offset[FDISK_P1 + count] =
1776 		    cl->cl_fmap[count].fmap_start;
1777 	}
1778 #endif
1779 
1780 	for (count = 0; count < NDKMAP; count++) {
1781 #if defined(_SUNOS_VTOC_8)
1782 		struct dk_map *lp  = &cl->cl_map[count];
1783 		cl->cl_offset[count] =
1784 		    cl->cl_g.dkg_nhead * cl->cl_g.dkg_nsect * lp->dkl_cylno;
1785 #elif defined(_SUNOS_VTOC_16)
1786 		struct dkl_partition *vp = &cl->cl_vtoc.v_part[count];
1787 
1788 		cl->cl_offset[count] = vp->p_start + cl->cl_solaris_offset;
1789 #else
1790 #error "No VTOC format defined."
1791 #endif
1792 	}
1793 
1794 	return (label_error);
1795 }
1796 
1797 #if defined(_SUNOS_VTOC_16)
1798 /*
1799  *    Function: cmlb_convert_geometry
1800  *
1801  * Description: Convert physical geometry into a dk_geom structure. In
1802  *		other words, make sure we don't wrap 16-bit values.
1803  *		e.g. converting from geom_cache to dk_geom
1804  *
1805  *     Context: Kernel thread only
1806  */
1807 static void
1808 cmlb_convert_geometry(struct cmlb_lun *cl, diskaddr_t capacity,
1809     struct dk_geom *cl_g, void *tg_cookie)
1810 {
1811 
1812 	ASSERT(cl != NULL);
1813 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
1814 
1815 	/* Unlabeled SCSI floppy device */
1816 	if (capacity <= 0x1000) {
1817 		cl_g->dkg_nhead = 2;
1818 		cl_g->dkg_ncyl = 80;
1819 		cl_g->dkg_nsect = capacity / (cl_g->dkg_nhead * cl_g->dkg_ncyl);
1820 		return;
1821 	}
1822 
1823 	/*
1824 	 * For all devices we calculate cylinders using the heads and sectors
1825 	 * we assign based on capacity of the device.  The algorithm is
1826 	 * designed to be compatible with the way other operating systems
1827 	 * lay out fdisk tables for X86 and to insure that the cylinders never
1828 	 * exceed 65535 to prevent problems with X86 ioctls that report
1829 	 * geometry.
1830 	 * For some smaller disk sizes we report geometry that matches those
1831 	 * used by X86 BIOS usage. For larger disks, we use SPT that are
1832 	 * multiples of 63, since other OSes that are not limited to 16-bits
1833 	 * for cylinders stop at 63 SPT we make do by using multiples of 63 SPT.
1834 	 *
1835 	 * The following table (in order) illustrates some end result
1836 	 * calculations:
1837 	 *
1838 	 * Maximum number of blocks 		nhead	nsect
1839 	 *
1840 	 * 2097152 (1GB)			64	32
1841 	 * 16777216 (8GB)			128	32
1842 	 * 1052819775 (502.02GB)		255  	63
1843 	 * 2105639550 (0.98TB)			255	126
1844 	 * 3158459325 (1.47TB)			255  	189
1845 	 * 4211279100 (1.96TB)			255  	252
1846 	 * 5264098875 (2.45TB)			255  	315
1847 	 * ...
1848 	 *
1849 	 * For Solid State Drive(SSD), it uses 4K page size inside and may be
1850 	 * double with every new generation. If the I/O is not aligned with
1851 	 * page size on SSDs, SSDs perform a lot slower.
1852 	 * By default, Solaris partition starts from cylinder 1. It will be
1853 	 * misaligned even with 4K if using heads(255) and SPT(63). To
1854 	 * workaround the problem, if the device is SSD, we use heads(224) and
1855 	 * SPT multiple of 56. Thus the default Solaris partition starts from
1856 	 * a position that aligns with 128K on a 512 bytes sector size SSD.
1857 	 */
1858 
1859 	if (capacity <= 0x200000) {
1860 		cl_g->dkg_nhead = 64;
1861 		cl_g->dkg_nsect = 32;
1862 	} else if (capacity <= 0x01000000) {
1863 		cl_g->dkg_nhead = 128;
1864 		cl_g->dkg_nsect = 32;
1865 	} else {
1866 		tg_attribute_t tgattribute;
1867 		int is_solid_state;
1868 		unsigned short nhead;
1869 		unsigned short nsect;
1870 
1871 		bzero(&tgattribute, sizeof (tg_attribute_t));
1872 
1873 		mutex_exit(CMLB_MUTEX(cl));
1874 		is_solid_state =
1875 		    (DK_TG_GETATTRIBUTE(cl, &tgattribute, tg_cookie) == 0) ?
1876 		    tgattribute.media_is_solid_state : FALSE;
1877 		mutex_enter(CMLB_MUTEX(cl));
1878 
1879 		if (is_solid_state) {
1880 			nhead = 224;
1881 			nsect = 56;
1882 		} else {
1883 			nhead = 255;
1884 			nsect = 63;
1885 		}
1886 
1887 		cl_g->dkg_nhead = nhead;
1888 
1889 		/* make nsect be smallest multiple of nhead */
1890 		cl_g->dkg_nsect = ((capacity +
1891 		    (UINT16_MAX * nhead * nsect) - 1) /
1892 		    (UINT16_MAX * nhead * nsect)) * nsect;
1893 
1894 		if (cl_g->dkg_nsect == 0)
1895 			cl_g->dkg_nsect = (UINT16_MAX / nsect) * nsect;
1896 	}
1897 
1898 }
1899 #endif
1900 
1901 /*
1902  *    Function: cmlb_resync_geom_caches
1903  *
1904  * Description: (Re)initialize both geometry caches: the virtual geometry
1905  *            information is extracted from the HBA (the "geometry"
1906  *            capability), and the physical geometry cache data is
1907  *            generated by issuing MODE SENSE commands.
1908  *
1909  *   Arguments:
1910  *	cl 		driver soft state (unit) structure
1911  *	capacity	disk capacity in #blocks
1912  *	tg_cookie	cookie from target driver to be passed back to target
1913  *			driver when we call back to it through tg_ops.
1914  *
1915  *     Context: Kernel thread only (can sleep).
1916  */
1917 static void
1918 cmlb_resync_geom_caches(struct cmlb_lun *cl, diskaddr_t capacity,
1919     void *tg_cookie)
1920 {
1921 	struct	cmlb_geom 	pgeom;
1922 	struct	cmlb_geom	lgeom;
1923 	struct 	cmlb_geom	*pgeomp = &pgeom;
1924 	unsigned short 		nhead;
1925 	unsigned short 		nsect;
1926 	int 			spc;
1927 	int			ret;
1928 
1929 	ASSERT(cl != NULL);
1930 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
1931 
1932 	/*
1933 	 * Ask the controller for its logical geometry.
1934 	 * Note: if the HBA does not support scsi_ifgetcap("geometry"),
1935 	 * then the lgeom cache will be invalid.
1936 	 */
1937 	mutex_exit(CMLB_MUTEX(cl));
1938 	bzero(&lgeom, sizeof (struct cmlb_geom));
1939 	ret = DK_TG_GETVIRTGEOM(cl, &lgeom, tg_cookie);
1940 	mutex_enter(CMLB_MUTEX(cl));
1941 
1942 	bcopy(&lgeom, &cl->cl_lgeom, sizeof (cl->cl_lgeom));
1943 
1944 	/*
1945 	 * Initialize the pgeom cache from lgeom, so that if MODE SENSE
1946 	 * doesn't work, DKIOCG_PHYSGEOM can return reasonable values.
1947 	 */
1948 	if (ret != 0 || cl->cl_lgeom.g_nsect == 0 ||
1949 	    cl->cl_lgeom.g_nhead == 0) {
1950 		/*
1951 		 * Note: Perhaps this needs to be more adaptive? The rationale
1952 		 * is that, if there's no HBA geometry from the HBA driver, any
1953 		 * guess is good, since this is the physical geometry. If MODE
1954 		 * SENSE fails this gives a max cylinder size for non-LBA access
1955 		 */
1956 		nhead = 255;
1957 		nsect = 63;
1958 	} else {
1959 		nhead = cl->cl_lgeom.g_nhead;
1960 		nsect = cl->cl_lgeom.g_nsect;
1961 	}
1962 
1963 	if (ISCD(cl)) {
1964 		pgeomp->g_nhead = 1;
1965 		pgeomp->g_nsect = nsect * nhead;
1966 	} else {
1967 		pgeomp->g_nhead = nhead;
1968 		pgeomp->g_nsect = nsect;
1969 	}
1970 
1971 	spc = pgeomp->g_nhead * pgeomp->g_nsect;
1972 	pgeomp->g_capacity = capacity;
1973 	if (spc == 0)
1974 		pgeomp->g_ncyl = 0;
1975 	else
1976 		pgeomp->g_ncyl = pgeomp->g_capacity / spc;
1977 	pgeomp->g_acyl = 0;
1978 
1979 	/*
1980 	 * Retrieve fresh geometry data from the hardware, stash it
1981 	 * here temporarily before we rebuild the incore label.
1982 	 *
1983 	 * We want to use the MODE SENSE commands to derive the
1984 	 * physical geometry of the device, but if either command
1985 	 * fails, the logical geometry is used as the fallback for
1986 	 * disk label geometry.
1987 	 */
1988 
1989 	mutex_exit(CMLB_MUTEX(cl));
1990 	(void) DK_TG_GETPHYGEOM(cl, pgeomp, tg_cookie);
1991 	mutex_enter(CMLB_MUTEX(cl));
1992 
1993 	/*
1994 	 * Now update the real copy while holding the mutex. This
1995 	 * way the global copy is never in an inconsistent state.
1996 	 */
1997 	bcopy(pgeomp, &cl->cl_pgeom,  sizeof (cl->cl_pgeom));
1998 
1999 	cmlb_dbg(CMLB_INFO, cl, "cmlb_resync_geom_caches: "
2000 	    "(cached from lgeom)\n");
2001 	cmlb_dbg(CMLB_INFO,  cl,
2002 	    "   ncyl: %ld; acyl: %d; nhead: %d; nsect: %d\n",
2003 	    cl->cl_pgeom.g_ncyl, cl->cl_pgeom.g_acyl,
2004 	    cl->cl_pgeom.g_nhead, cl->cl_pgeom.g_nsect);
2005 	cmlb_dbg(CMLB_INFO,  cl, "   lbasize: %d; capacity: %ld; "
2006 	    "intrlv: %d; rpm: %d\n", cl->cl_pgeom.g_secsize,
2007 	    cl->cl_pgeom.g_capacity, cl->cl_pgeom.g_intrlv,
2008 	    cl->cl_pgeom.g_rpm);
2009 }
2010 
2011 
2012 #if defined(__i386) || defined(__amd64)
2013 /*
2014  *    Function: cmlb_update_ext_minor_nodes
2015  *
2016  * Description: Routine to add/remove extended partition device nodes
2017  *
2018  *   Arguments:
2019  *	cl		driver soft state (unit) structure
2020  *	num_parts	Number of logical drives found on the LUN
2021  *
2022  * Should be called with the mutex held
2023  *
2024  * Return Code: 0 for success
2025  *
2026  *     Context: User and Kernel thread
2027  *
2028  */
2029 static int
2030 cmlb_update_ext_minor_nodes(struct cmlb_lun *cl, int num_parts)
2031 {
2032 	int				i, count;
2033 	char				name[48];
2034 	int				instance;
2035 	struct driver_minor_data	*demdp, *demdpr;
2036 	char				*devnm;
2037 	dev_info_t			*pdip;
2038 	boolean_t 			internal;
2039 
2040 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
2041 	ASSERT(cl->cl_update_ext_minor_nodes == 1);
2042 
2043 	internal = VOID2BOOLEAN(
2044 	    (cl->cl_alter_behavior & (CMLB_INTERNAL_MINOR_NODES)) != 0);
2045 	instance = ddi_get_instance(CMLB_DEVINFO(cl));
2046 	demdp = dk_ext_minor_data;
2047 	demdpr = &dk_ext_minor_data[MAX_EXT_PARTS];
2048 
2049 
2050 	if (cl->cl_logical_drive_count) {
2051 		for (i = 0; i < cl->cl_logical_drive_count; i++) {
2052 			(void) sprintf(name, "%s", demdp->name);
2053 			ddi_remove_minor_node(CMLB_DEVINFO(cl), name);
2054 			(void) sprintf(name, "%s", demdpr->name);
2055 			ddi_remove_minor_node(CMLB_DEVINFO(cl), name);
2056 			demdp++;
2057 			demdpr++;
2058 		}
2059 		/* There are existing device nodes. Remove them */
2060 		devnm = kmem_alloc(MAXNAMELEN + 1, KM_SLEEP);
2061 		(void) ddi_deviname(cl->cl_devi, devnm);
2062 		pdip = ddi_get_parent(cl->cl_devi);
2063 		(void) devfs_clean(pdip, devnm + 1, DV_CLEAN_FORCE);
2064 		kmem_free(devnm, MAXNAMELEN + 1);
2065 	}
2066 
2067 	demdp = dk_ext_minor_data;
2068 	demdpr = &dk_ext_minor_data[MAX_EXT_PARTS];
2069 
2070 	for (i = 0; i < num_parts; i++) {
2071 		(void) sprintf(name, "%s", demdp->name);
2072 		if (cmlb_create_minor(CMLB_DEVINFO(cl), name,
2073 		    demdp->type,
2074 		    (instance << CMLBUNIT_SHIFT) | demdp->minor,
2075 		    cl->cl_node_type, NULL, internal) == DDI_FAILURE) {
2076 			/*
2077 			 * Clean up any nodes that may have been
2078 			 * created, in case this fails in the middle
2079 			 * of the loop.
2080 			 */
2081 			ddi_remove_minor_node(CMLB_DEVINFO(cl), NULL);
2082 			cl->cl_logical_drive_count = 0;
2083 			return (ENXIO);
2084 		}
2085 		(void) sprintf(name, "%s", demdpr->name);
2086 		if (ddi_create_minor_node(CMLB_DEVINFO(cl), name,
2087 		    demdpr->type,
2088 		    (instance << CMLBUNIT_SHIFT) | demdpr->minor,
2089 		    cl->cl_node_type, NULL) == DDI_FAILURE) {
2090 			/*
2091 			 * Clean up any nodes that may have been
2092 			 * created, in case this fails in the middle
2093 			 * of the loop.
2094 			 */
2095 			ddi_remove_minor_node(CMLB_DEVINFO(cl), NULL);
2096 			cl->cl_logical_drive_count = 0;
2097 			return (ENXIO);
2098 		}
2099 		demdp++;
2100 		demdpr++;
2101 	}
2102 
2103 	/* Update the cl_map array for logical drives */
2104 	for (count = 0; count < MAX_EXT_PARTS; count++) {
2105 		cl->cl_map[FDISK_P4 + 1 + count].dkl_cylno = UINT32_MAX;
2106 		cl->cl_map[FDISK_P4 + 1 + count].dkl_nblk =
2107 		    cl->cl_fmap[FD_NUMPART + count].fmap_nblk;
2108 		cl->cl_offset[FDISK_P4 + 1 + count] =
2109 		    cl->cl_fmap[FD_NUMPART + count].fmap_start;
2110 	}
2111 
2112 	cl->cl_logical_drive_count = i;
2113 	cl->cl_update_ext_minor_nodes = 0;
2114 	return (0);
2115 }
2116 /*
2117  *    Function: cmlb_validate_ext_part
2118  *
2119  * Description: utility routine to validate an extended partition's
2120  *		metadata as found on disk
2121  *
2122  *   Arguments:
2123  *	cl		driver soft state (unit) structure
2124  *	part		partition number of the extended partition
2125  *	epart		partition number of the logical drive
2126  *	start		absolute sector number of the start of the logical
2127  *			drive being validated
2128  *	size		size of logical drive being validated
2129  *
2130  * Return Code: 0 for success
2131  *
2132  *     Context: User and Kernel thread
2133  *
2134  * Algorithm :
2135  * Error cases are :
2136  *	1. If start block is lesser than or equal to the end block
2137  *	2. If either start block or end block is beyond the bounadry
2138  *	   of the extended partition.
2139  *	3. start or end block overlap with existing partitions.
2140  *		To check this, first make sure that the start block doesnt
2141  *		overlap with existing partitions. Then, calculate the
2142  *		possible end block for the given start block that doesnt
2143  *		overlap with existing partitions. This can be calculated by
2144  *		first setting the possible end block to the end of the
2145  *		extended partition (optimistic) and then, checking if there
2146  *		is any other partition that lies after the start of the
2147  *		partition being validated. If so, set the possible end to
2148  *		one block less than the beginning of the next nearest partition
2149  *		If the actual end block is greater than the calculated end
2150  *		block, we have an overlap.
2151  *
2152  */
2153 static int
2154 cmlb_validate_ext_part(struct cmlb_lun *cl, int part, int epart, uint32_t start,
2155     uint32_t size)
2156 {
2157 	int i;
2158 	uint32_t end = start + size - 1;
2159 	uint32_t ext_start = cl->cl_fmap[part].fmap_start;
2160 	uint32_t ext_end = ext_start + cl->cl_fmap[part].fmap_nblk - 1;
2161 	uint32_t ts, te;
2162 	uint32_t poss_end = ext_end;
2163 
2164 	if (end <= start) {
2165 		return (1);
2166 	}
2167 
2168 	/*
2169 	 * Check if the logical drive boundaries are within that of the
2170 	 * extended partition.
2171 	 */
2172 	if (start <= ext_start || start > ext_end || end <= ext_start ||
2173 	    end > ext_end) {
2174 		return (1);
2175 	}
2176 
2177 	/*
2178 	 * epart will be equal to FD_NUMPART if it is the first logical drive.
2179 	 * There is no need to check for overlaps with other logical drives,
2180 	 * since it is the only logical drive that we have come across so far.
2181 	 */
2182 	if (epart == FD_NUMPART) {
2183 		return (0);
2184 	}
2185 
2186 	/* Check for overlaps with existing logical drives */
2187 	i = FD_NUMPART;
2188 	ts = cl->cl_fmap[FD_NUMPART].fmap_start;
2189 	te = ts + cl->cl_fmap[FD_NUMPART].fmap_nblk - 1;
2190 
2191 	while ((i < epart) && ts && te) {
2192 		if (start >= ts && start <= te) {
2193 			return (1);
2194 		}
2195 
2196 		if ((ts < poss_end) && (ts > start)) {
2197 			poss_end = ts - 1;
2198 		}
2199 
2200 		i++;
2201 		ts = cl->cl_fmap[i].fmap_start;
2202 		te = ts + cl->cl_fmap[i].fmap_nblk - 1;
2203 	}
2204 
2205 	if (end > poss_end) {
2206 		return (1);
2207 	}
2208 
2209 	return (0);
2210 }
2211 
2212 
2213 /*
2214  *    Function: cmlb_is_linux_swap
2215  *
2216  * Description: utility routine to verify if a partition is a linux swap
2217  *		partition or not.
2218  *
2219  *   Arguments:
2220  *	cl		driver soft state (unit) structure
2221  *	part_start	absolute sector number of the start of the partition
2222  *			being verified
2223  *	tg_cookie	cookie from target driver to be passed back to target
2224  *			driver when we call back to it through tg_ops.
2225  *
2226  * Return Code: 0 for success
2227  *
2228  *     Context: User and Kernel thread
2229  *
2230  * Notes:
2231  *	The linux swap magic "SWAP-SPACE" or "SWAPSPACE2" is found as the
2232  *	last 10 bytes of a disk block whose size is that of the linux page
2233  *	size. This disk block is found at the beginning of the swap partition.
2234  */
2235 static int
2236 cmlb_is_linux_swap(struct cmlb_lun *cl, uint32_t part_start, void *tg_cookie)
2237 {
2238 	int		i;
2239 	int		rval = -1;
2240 	uint32_t	seek_offset;
2241 	uint32_t	linux_pg_size;
2242 	char 		*buf, *linux_swap_magic;
2243 	int		sec_sz = cl->cl_sys_blocksize;
2244 	/* Known linux kernel page sizes */
2245 	uint32_t	linux_pg_size_arr[] = {4096, };
2246 
2247 	ASSERT(cl != NULL);
2248 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
2249 
2250 	if ((buf = kmem_zalloc(sec_sz, KM_NOSLEEP)) == NULL) {
2251 		return (ENOMEM);
2252 	}
2253 
2254 	linux_swap_magic = buf + sec_sz - 10;
2255 
2256 	for (i = 0; i < sizeof (linux_pg_size_arr)/sizeof (uint32_t); i++) {
2257 		linux_pg_size = linux_pg_size_arr[i];
2258 		seek_offset = linux_pg_size/sec_sz - 1;
2259 		seek_offset += part_start;
2260 
2261 		mutex_exit(CMLB_MUTEX(cl));
2262 		rval = DK_TG_READ(cl, buf, seek_offset, sec_sz, tg_cookie);
2263 		mutex_enter(CMLB_MUTEX(cl));
2264 
2265 		if (rval != 0) {
2266 			cmlb_dbg(CMLB_ERROR,  cl,
2267 			    "cmlb_is_linux_swap: disk read err\n");
2268 			rval = EIO;
2269 			break;
2270 		}
2271 
2272 		rval = -1;
2273 
2274 		if ((strncmp(linux_swap_magic, "SWAP-SPACE", 10) == 0) ||
2275 		    (strncmp(linux_swap_magic, "SWAPSPACE2", 10) == 0)) {
2276 			/* Found a linux swap */
2277 			rval = 0;
2278 			break;
2279 		}
2280 	}
2281 
2282 	kmem_free(buf, sec_sz);
2283 	return (rval);
2284 }
2285 #endif
2286 
2287 /*
2288  *    Function: cmlb_read_fdisk
2289  *
2290  * Description: utility routine to read the fdisk table.
2291  *
2292  *   Arguments:
2293  *	cl		driver soft state (unit) structure
2294  *	capacity	disk capacity in #blocks
2295  *	tg_cookie	cookie from target driver to be passed back to target
2296  *			driver when we call back to it through tg_ops.
2297  *
2298  * Return Code: 0 for success (includes not reading for no_fdisk_present case
2299  *		errnos from tg_rw if failed to read the first block.
2300  *
2301  *     Context: Kernel thread only (can sleep).
2302  */
2303 /*ARGSUSED*/
2304 static int
2305 cmlb_read_fdisk(struct cmlb_lun *cl, diskaddr_t capacity, void *tg_cookie)
2306 {
2307 #if defined(_NO_FDISK_PRESENT)
2308 
2309 	cl->cl_solaris_offset = 0;
2310 	cl->cl_solaris_size = capacity;
2311 	bzero(cl->cl_fmap, sizeof (struct fmap) * FD_NUMPART);
2312 	return (0);
2313 
2314 #elif defined(_FIRMWARE_NEEDS_FDISK)
2315 
2316 	struct ipart	*fdp;
2317 	struct mboot	*mbp;
2318 	struct ipart	fdisk[FD_NUMPART];
2319 	int		i, k;
2320 	char		sigbuf[2];
2321 	caddr_t		bufp;
2322 	int		uidx;
2323 	int 		rval;
2324 	int		lba = 0;
2325 	uint_t		solaris_offset;	/* offset to solaris part. */
2326 	daddr_t		solaris_size;	/* size of solaris partition */
2327 	uint32_t	blocksize;
2328 #if defined(__i386) || defined(__amd64)
2329 	struct ipart	eparts[2];
2330 	struct ipart	*efdp1 = &eparts[0];
2331 	struct ipart	*efdp2 = &eparts[1];
2332 	int		ext_part_exists = 0;
2333 	int		ld_count = 0;
2334 	int		is_linux_swap;
2335 #endif
2336 
2337 	ASSERT(cl != NULL);
2338 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
2339 
2340 	/*
2341 	 * Start off assuming no fdisk table
2342 	 */
2343 	solaris_offset = 0;
2344 	solaris_size   = capacity;
2345 
2346 	blocksize = cl->cl_tgt_blocksize;
2347 
2348 	bufp = kmem_zalloc(blocksize, KM_SLEEP);
2349 
2350 	mutex_exit(CMLB_MUTEX(cl));
2351 	rval = DK_TG_READ(cl, bufp, 0, blocksize, tg_cookie);
2352 	mutex_enter(CMLB_MUTEX(cl));
2353 
2354 	if (rval != 0) {
2355 		cmlb_dbg(CMLB_ERROR,  cl,
2356 		    "cmlb_read_fdisk: fdisk read err\n");
2357 		bzero(cl->cl_fmap, sizeof (struct fmap) * FD_NUMPART);
2358 		goto done;
2359 	}
2360 
2361 	mbp = (struct mboot *)bufp;
2362 
2363 	/*
2364 	 * The fdisk table does not begin on a 4-byte boundary within the
2365 	 * master boot record, so we copy it to an aligned structure to avoid
2366 	 * alignment exceptions on some processors.
2367 	 */
2368 	bcopy(&mbp->parts[0], fdisk, sizeof (fdisk));
2369 
2370 	/*
2371 	 * Check for lba support before verifying sig; sig might not be
2372 	 * there, say on a blank disk, but the max_chs mark may still
2373 	 * be present.
2374 	 *
2375 	 * Note: LBA support and BEFs are an x86-only concept but this
2376 	 * code should work OK on SPARC as well.
2377 	 */
2378 
2379 	/*
2380 	 * First, check for lba-access-ok on root node (or prom root node)
2381 	 * if present there, don't need to search fdisk table.
2382 	 */
2383 	if (ddi_getprop(DDI_DEV_T_ANY, ddi_root_node(), 0,
2384 	    "lba-access-ok", 0) != 0) {
2385 		/* All drives do LBA; don't search fdisk table */
2386 		lba = 1;
2387 	} else {
2388 		/* Okay, look for mark in fdisk table */
2389 		for (fdp = fdisk, i = 0; i < FD_NUMPART; i++, fdp++) {
2390 			/* accumulate "lba" value from all partitions */
2391 			lba = (lba || cmlb_has_max_chs_vals(fdp));
2392 		}
2393 	}
2394 
2395 	if (lba != 0) {
2396 		dev_t dev = cmlb_make_device(cl);
2397 
2398 		if (ddi_getprop(dev, CMLB_DEVINFO(cl), DDI_PROP_DONTPASS,
2399 		    "lba-access-ok", 0) == 0) {
2400 			/* not found; create it */
2401 			if (ddi_prop_create(dev, CMLB_DEVINFO(cl), 0,
2402 			    "lba-access-ok", (caddr_t)NULL, 0) !=
2403 			    DDI_PROP_SUCCESS) {
2404 				cmlb_dbg(CMLB_ERROR,  cl,
2405 				    "cmlb_read_fdisk: Can't create lba "
2406 				    "property for instance %d\n",
2407 				    ddi_get_instance(CMLB_DEVINFO(cl)));
2408 			}
2409 		}
2410 	}
2411 
2412 	bcopy(&mbp->signature, sigbuf, sizeof (sigbuf));
2413 
2414 	/*
2415 	 * Endian-independent signature check
2416 	 */
2417 	if (((sigbuf[1] & 0xFF) != ((MBB_MAGIC >> 8) & 0xFF)) ||
2418 	    (sigbuf[0] != (MBB_MAGIC & 0xFF))) {
2419 		cmlb_dbg(CMLB_ERROR,  cl,
2420 		    "cmlb_read_fdisk: no fdisk\n");
2421 		bzero(cl->cl_fmap, sizeof (struct fmap) * FD_NUMPART);
2422 		goto done;
2423 	}
2424 
2425 #ifdef CMLBDEBUG
2426 	if (cmlb_level_mask & CMLB_LOGMASK_INFO) {
2427 		fdp = fdisk;
2428 		cmlb_dbg(CMLB_INFO,  cl, "cmlb_read_fdisk:\n");
2429 		cmlb_dbg(CMLB_INFO,  cl, "         relsect    "
2430 		    "numsect         sysid       bootid\n");
2431 		for (i = 0; i < FD_NUMPART; i++, fdp++) {
2432 			cmlb_dbg(CMLB_INFO,  cl,
2433 			    "    %d:  %8d   %8d     0x%08x     0x%08x\n",
2434 			    i, fdp->relsect, fdp->numsect,
2435 			    fdp->systid, fdp->bootid);
2436 		}
2437 	}
2438 #endif
2439 
2440 	/*
2441 	 * Try to find the unix partition
2442 	 */
2443 	uidx = -1;
2444 	solaris_offset = 0;
2445 	solaris_size   = 0;
2446 
2447 	for (fdp = fdisk, i = 0; i < FD_NUMPART; i++, fdp++) {
2448 		uint32_t relsect;
2449 		uint32_t numsect;
2450 		uchar_t systid;
2451 #if defined(__i386) || defined(__amd64)
2452 		/*
2453 		 * Stores relative block offset from the beginning of the
2454 		 * Extended Partition.
2455 		 */
2456 		int	ext_relsect = 0;
2457 #endif
2458 
2459 		if (fdp->numsect == 0) {
2460 			cl->cl_fmap[i].fmap_start = 0;
2461 			cl->cl_fmap[i].fmap_nblk  = 0;
2462 			continue;
2463 		}
2464 
2465 		/*
2466 		 * Data in the fdisk table is little-endian.
2467 		 */
2468 		relsect = LE_32(fdp->relsect);
2469 		numsect = LE_32(fdp->numsect);
2470 
2471 		cl->cl_fmap[i].fmap_start = relsect;
2472 		cl->cl_fmap[i].fmap_nblk  = numsect;
2473 		cl->cl_fmap[i].fmap_systid = LE_8(fdp->systid);
2474 
2475 #if defined(__i386) || defined(__amd64)
2476 		/* Support only one extended partition per LUN */
2477 		if ((fdp->systid == EXTDOS || fdp->systid == FDISK_EXTLBA) &&
2478 		    (ext_part_exists == 0)) {
2479 			int j;
2480 			uint32_t logdrive_offset;
2481 			uint32_t ext_numsect;
2482 			uint32_t abs_secnum;
2483 
2484 			ext_part_exists = 1;
2485 
2486 			for (j = FD_NUMPART; j < FDISK_PARTS; j++) {
2487 				mutex_exit(CMLB_MUTEX(cl));
2488 				rval = DK_TG_READ(cl, bufp,
2489 				    (relsect + ext_relsect), blocksize,
2490 				    tg_cookie);
2491 				mutex_enter(CMLB_MUTEX(cl));
2492 
2493 				if (rval != 0) {
2494 					cmlb_dbg(CMLB_ERROR,  cl,
2495 					    "cmlb_read_fdisk: Extended "
2496 					    "partition read err\n");
2497 					goto done;
2498 				}
2499 				/*
2500 				 * The first ipart entry provides the offset
2501 				 * at which the logical drive starts off from
2502 				 * the beginning of the container partition
2503 				 * and the size of the logical drive.
2504 				 * The second ipart entry provides the offset
2505 				 * of the next container partition from the
2506 				 * beginning of the extended partition.
2507 				 */
2508 				bcopy(&bufp[FDISK_PART_TABLE_START], eparts,
2509 				    sizeof (eparts));
2510 				logdrive_offset = LE_32(efdp1->relsect);
2511 				ext_numsect = LE_32(efdp1->numsect);
2512 				systid = LE_8(efdp1->systid);
2513 				if (logdrive_offset <= 0 || ext_numsect <= 0)
2514 					break;
2515 				abs_secnum = relsect + ext_relsect +
2516 				    logdrive_offset;
2517 
2518 				/* Boundary condition and overlap checking */
2519 				if (cmlb_validate_ext_part(cl, i, j, abs_secnum,
2520 				    ext_numsect)) {
2521 					break;
2522 				}
2523 
2524 				if ((cl->cl_fmap[j].fmap_start != abs_secnum) ||
2525 				    (cl->cl_fmap[j].fmap_nblk != ext_numsect) ||
2526 				    (cl->cl_fmap[j].fmap_systid != systid)) {
2527 					/*
2528 					 * Indicates change from previous
2529 					 * partinfo. Need to recreate
2530 					 * logical device nodes.
2531 					 */
2532 					cl->cl_update_ext_minor_nodes = 1;
2533 				}
2534 				cl->cl_fmap[j].fmap_start = abs_secnum;
2535 				cl->cl_fmap[j].fmap_nblk  = ext_numsect;
2536 				cl->cl_fmap[j].fmap_systid = systid;
2537 				ld_count++;
2538 
2539 				is_linux_swap = 0;
2540 				if (efdp1->systid == SUNIXOS) {
2541 					if (cmlb_is_linux_swap(cl, abs_secnum,
2542 					    tg_cookie) == 0) {
2543 						is_linux_swap = 1;
2544 					}
2545 				}
2546 
2547 				if ((efdp1->systid == SUNIXOS) ||
2548 				    (efdp1->systid == SUNIXOS2)) {
2549 					if ((uidx == -1) && (!is_linux_swap)) {
2550 						uidx = 0;
2551 						solaris_offset = abs_secnum;
2552 						solaris_size = ext_numsect;
2553 					}
2554 				}
2555 
2556 				if ((ext_relsect = LE_32(efdp2->relsect)) == 0)
2557 					break;
2558 			}
2559 		}
2560 
2561 #endif
2562 
2563 		if (fdp->systid != SUNIXOS &&
2564 		    fdp->systid != SUNIXOS2 &&
2565 		    fdp->systid != EFI_PMBR) {
2566 			continue;
2567 		}
2568 
2569 		/*
2570 		 * use the last active solaris partition id found
2571 		 * (there should only be 1 active partition id)
2572 		 *
2573 		 * if there are no active solaris partition id
2574 		 * then use the first inactive solaris partition id
2575 		 */
2576 		if ((uidx == -1) || (fdp->bootid == ACTIVE)) {
2577 #if defined(__i386) || defined(__amd64)
2578 			is_linux_swap = 0;
2579 			if (fdp->systid == SUNIXOS) {
2580 				if (cmlb_is_linux_swap(cl, relsect,
2581 				    tg_cookie) == 0) {
2582 					is_linux_swap = 1;
2583 				}
2584 			}
2585 			if (!is_linux_swap) {
2586 #endif
2587 				uidx = i;
2588 				solaris_offset = relsect;
2589 				solaris_size   = numsect;
2590 #if defined(__i386) || defined(__amd64)
2591 			}
2592 #endif
2593 		}
2594 	}
2595 #if defined(__i386) || defined(__amd64)
2596 	if (ld_count < cl->cl_logical_drive_count) {
2597 		/*
2598 		 * Some/all logical drives were deleted. Clear out
2599 		 * the fmap entries correspoding to those deleted drives.
2600 		 */
2601 		for (k = ld_count + FD_NUMPART;
2602 		    k < cl->cl_logical_drive_count + FD_NUMPART; k++) {
2603 			cl->cl_fmap[k].fmap_start = 0;
2604 			cl->cl_fmap[k].fmap_nblk  = 0;
2605 			cl->cl_fmap[k].fmap_systid = 0;
2606 		}
2607 		cl->cl_update_ext_minor_nodes = 1;
2608 	}
2609 	if (cl->cl_update_ext_minor_nodes) {
2610 		rval = cmlb_update_ext_minor_nodes(cl, ld_count);
2611 		if (rval != 0) {
2612 			goto done;
2613 		}
2614 	}
2615 #endif
2616 	cmlb_dbg(CMLB_INFO,  cl, "fdisk 0x%x 0x%lx",
2617 	    cl->cl_solaris_offset, cl->cl_solaris_size);
2618 done:
2619 
2620 	/*
2621 	 * Clear the VTOC info, only if the Solaris partition entry
2622 	 * has moved, changed size, been deleted, or if the size of
2623 	 * the partition is too small to even fit the label sector.
2624 	 */
2625 	if ((cl->cl_solaris_offset != solaris_offset) ||
2626 	    (cl->cl_solaris_size != solaris_size) ||
2627 	    solaris_size <= DK_LABEL_LOC) {
2628 		cmlb_dbg(CMLB_INFO,  cl, "fdisk moved 0x%x 0x%lx",
2629 		    solaris_offset, solaris_size);
2630 		bzero(&cl->cl_g, sizeof (struct dk_geom));
2631 		bzero(&cl->cl_vtoc, sizeof (struct dk_vtoc));
2632 		bzero(&cl->cl_map, NDKMAP * (sizeof (struct dk_map)));
2633 		cl->cl_f_geometry_is_valid = B_FALSE;
2634 	}
2635 	cl->cl_solaris_offset = solaris_offset;
2636 	cl->cl_solaris_size = solaris_size;
2637 	kmem_free(bufp, blocksize);
2638 	return (rval);
2639 
2640 #else	/* #elif defined(_FIRMWARE_NEEDS_FDISK) */
2641 #error "fdisk table presence undetermined for this platform."
2642 #endif	/* #if defined(_NO_FDISK_PRESENT) */
2643 }
2644 
2645 static void
2646 cmlb_swap_efi_gpt(efi_gpt_t *e)
2647 {
2648 	_NOTE(ASSUMING_PROTECTED(*e))
2649 	e->efi_gpt_Signature = LE_64(e->efi_gpt_Signature);
2650 	e->efi_gpt_Revision = LE_32(e->efi_gpt_Revision);
2651 	e->efi_gpt_HeaderSize = LE_32(e->efi_gpt_HeaderSize);
2652 	e->efi_gpt_HeaderCRC32 = LE_32(e->efi_gpt_HeaderCRC32);
2653 	e->efi_gpt_MyLBA = LE_64(e->efi_gpt_MyLBA);
2654 	e->efi_gpt_AlternateLBA = LE_64(e->efi_gpt_AlternateLBA);
2655 	e->efi_gpt_FirstUsableLBA = LE_64(e->efi_gpt_FirstUsableLBA);
2656 	e->efi_gpt_LastUsableLBA = LE_64(e->efi_gpt_LastUsableLBA);
2657 	UUID_LE_CONVERT(e->efi_gpt_DiskGUID, e->efi_gpt_DiskGUID);
2658 	e->efi_gpt_PartitionEntryLBA = LE_64(e->efi_gpt_PartitionEntryLBA);
2659 	e->efi_gpt_NumberOfPartitionEntries =
2660 	    LE_32(e->efi_gpt_NumberOfPartitionEntries);
2661 	e->efi_gpt_SizeOfPartitionEntry =
2662 	    LE_32(e->efi_gpt_SizeOfPartitionEntry);
2663 	e->efi_gpt_PartitionEntryArrayCRC32 =
2664 	    LE_32(e->efi_gpt_PartitionEntryArrayCRC32);
2665 }
2666 
2667 static void
2668 cmlb_swap_efi_gpe(int nparts, efi_gpe_t *p)
2669 {
2670 	int i;
2671 
2672 	_NOTE(ASSUMING_PROTECTED(*p))
2673 	for (i = 0; i < nparts; i++) {
2674 		UUID_LE_CONVERT(p[i].efi_gpe_PartitionTypeGUID,
2675 		    p[i].efi_gpe_PartitionTypeGUID);
2676 		p[i].efi_gpe_StartingLBA = LE_64(p[i].efi_gpe_StartingLBA);
2677 		p[i].efi_gpe_EndingLBA = LE_64(p[i].efi_gpe_EndingLBA);
2678 		/* PartitionAttrs */
2679 	}
2680 }
2681 
2682 static int
2683 cmlb_validate_efi(efi_gpt_t *labp)
2684 {
2685 	if (labp->efi_gpt_Signature != EFI_SIGNATURE)
2686 		return (EINVAL);
2687 	/* at least 96 bytes in this version of the spec. */
2688 	if (sizeof (efi_gpt_t) - sizeof (labp->efi_gpt_Reserved2) >
2689 	    labp->efi_gpt_HeaderSize)
2690 		return (EINVAL);
2691 	/* this should be 128 bytes */
2692 	if (labp->efi_gpt_SizeOfPartitionEntry != sizeof (efi_gpe_t))
2693 		return (EINVAL);
2694 	return (0);
2695 }
2696 
2697 /*
2698  * This function returns B_FALSE if there is a valid MBR signature and no
2699  * partition table entries of type EFI_PMBR (0xEE). Otherwise it returns B_TRUE.
2700  *
2701  * The EFI spec (1.10 and later) requires having a Protective MBR (PMBR) to
2702  * recognize the disk as GPT partitioned. However, some other OS creates an MBR
2703  * where a PMBR entry is not the only one. Also, if the first block has been
2704  * corrupted, currently best attempt to allow data access would be to try to
2705  * check for GPT headers. Hence in case of more than one partition entry, but
2706  * at least one EFI_PMBR partition type or no valid magic number, the function
2707  * returns B_TRUE to continue with looking for GPT header.
2708  */
2709 
2710 static boolean_t
2711 cmlb_check_efi_mbr(uchar_t *buf, boolean_t *is_mbr)
2712 {
2713 	struct ipart	*fdp;
2714 	struct mboot	*mbp = (struct mboot *)buf;
2715 	struct ipart	fdisk[FD_NUMPART];
2716 	int		i;
2717 
2718 	if (is_mbr != NULL)
2719 		*is_mbr = B_TRUE;
2720 
2721 	if (LE_16(mbp->signature) != MBB_MAGIC) {
2722 		if (is_mbr != NULL)
2723 			*is_mbr = B_FALSE;
2724 		return (B_TRUE);
2725 	}
2726 
2727 	bcopy(&mbp->parts[0], fdisk, sizeof (fdisk));
2728 
2729 	for (fdp = fdisk, i = 0; i < FD_NUMPART; i++, fdp++) {
2730 		if (fdp->systid == EFI_PMBR)
2731 			return (B_TRUE);
2732 	}
2733 
2734 	return (B_FALSE);
2735 }
2736 
2737 static int
2738 cmlb_use_efi(struct cmlb_lun *cl, diskaddr_t capacity, int flags,
2739     void *tg_cookie)
2740 {
2741 	int		i;
2742 	int		rval = 0;
2743 	efi_gpe_t	*partitions;
2744 	uchar_t		*buf;
2745 	uint_t		lbasize;	/* is really how much to read */
2746 	diskaddr_t	cap = 0;
2747 	uint_t		nparts;
2748 	diskaddr_t	gpe_lba;
2749 	diskaddr_t	alternate_lba;
2750 	int		iofailed = 0;
2751 	struct uuid	uuid_type_reserved = EFI_RESERVED;
2752 #if defined(_FIRMWARE_NEEDS_FDISK)
2753 	boolean_t 	is_mbr;
2754 #endif
2755 
2756 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
2757 
2758 	lbasize = cl->cl_sys_blocksize;
2759 
2760 	cl->cl_reserved = -1;
2761 	mutex_exit(CMLB_MUTEX(cl));
2762 
2763 	buf = kmem_zalloc(EFI_MIN_ARRAY_SIZE, KM_SLEEP);
2764 
2765 	rval = DK_TG_READ(cl, buf,  0, lbasize, tg_cookie);
2766 	if (rval) {
2767 		iofailed = 1;
2768 		goto done_err;
2769 	}
2770 	if (((struct dk_label *)buf)->dkl_magic == DKL_MAGIC) {
2771 		/* not ours */
2772 		rval = ESRCH;
2773 		goto done_err;
2774 	}
2775 
2776 #if defined(_FIRMWARE_NEEDS_FDISK)
2777 	if (!cmlb_check_efi_mbr(buf, &is_mbr)) {
2778 		if (is_mbr)
2779 			rval = ESRCH;
2780 		else
2781 			rval = EINVAL;
2782 		goto done_err;
2783 	}
2784 #else
2785 	if (!cmlb_check_efi_mbr(buf, NULL)) {
2786 		rval = EINVAL;
2787 		goto done_err;
2788 	}
2789 
2790 #endif
2791 
2792 	rval = DK_TG_READ(cl, buf, 1, lbasize, tg_cookie);
2793 	if (rval) {
2794 		iofailed = 1;
2795 		goto done_err;
2796 	}
2797 	cmlb_swap_efi_gpt((efi_gpt_t *)buf);
2798 
2799 	if ((rval = cmlb_validate_efi((efi_gpt_t *)buf)) != 0) {
2800 		/*
2801 		 * Couldn't read the primary, try the backup.  Our
2802 		 * capacity at this point could be based on CHS, so
2803 		 * check what the device reports.
2804 		 */
2805 		rval = DK_TG_GETCAP(cl, &cap, tg_cookie);
2806 		if (rval) {
2807 			iofailed = 1;
2808 			goto done_err;
2809 		}
2810 
2811 		/*
2812 		 * CMLB_OFF_BY_ONE case, we check the next to last block first
2813 		 * for backup GPT header, otherwise check the last block.
2814 		 */
2815 
2816 		if ((rval = DK_TG_READ(cl, buf,
2817 		    cap - ((cl->cl_alter_behavior & CMLB_OFF_BY_ONE) ? 2 : 1),
2818 		    lbasize, tg_cookie))
2819 		    != 0) {
2820 			iofailed = 1;
2821 			goto done_err;
2822 		}
2823 		cmlb_swap_efi_gpt((efi_gpt_t *)buf);
2824 
2825 		if ((rval = cmlb_validate_efi((efi_gpt_t *)buf)) != 0) {
2826 
2827 			if (!(cl->cl_alter_behavior & CMLB_OFF_BY_ONE))
2828 				goto done_err;
2829 			if ((rval = DK_TG_READ(cl, buf, cap - 1, lbasize,
2830 			    tg_cookie)) != 0)
2831 				goto done_err;
2832 			cmlb_swap_efi_gpt((efi_gpt_t *)buf);
2833 			if ((rval = cmlb_validate_efi((efi_gpt_t *)buf)) != 0)
2834 				goto done_err;
2835 		}
2836 		if (!(flags & CMLB_SILENT))
2837 			cmlb_log(CMLB_DEVINFO(cl), CMLB_LABEL(cl), CE_WARN,
2838 			    "primary label corrupt; using backup\n");
2839 	}
2840 
2841 	nparts = ((efi_gpt_t *)buf)->efi_gpt_NumberOfPartitionEntries;
2842 	gpe_lba = ((efi_gpt_t *)buf)->efi_gpt_PartitionEntryLBA;
2843 	alternate_lba = ((efi_gpt_t *)buf)->efi_gpt_AlternateLBA;
2844 
2845 	rval = DK_TG_READ(cl, buf, gpe_lba, EFI_MIN_ARRAY_SIZE, tg_cookie);
2846 	if (rval) {
2847 		iofailed = 1;
2848 		goto done_err;
2849 	}
2850 	partitions = (efi_gpe_t *)buf;
2851 
2852 	if (nparts > MAXPART) {
2853 		nparts = MAXPART;
2854 	}
2855 	cmlb_swap_efi_gpe(nparts, partitions);
2856 
2857 	mutex_enter(CMLB_MUTEX(cl));
2858 
2859 	/* Fill in partition table. */
2860 	for (i = 0; i < nparts; i++) {
2861 		if (partitions->efi_gpe_StartingLBA != 0 ||
2862 		    partitions->efi_gpe_EndingLBA != 0) {
2863 			cl->cl_map[i].dkl_cylno =
2864 			    partitions->efi_gpe_StartingLBA;
2865 			cl->cl_map[i].dkl_nblk =
2866 			    partitions->efi_gpe_EndingLBA -
2867 			    partitions->efi_gpe_StartingLBA + 1;
2868 			cl->cl_offset[i] =
2869 			    partitions->efi_gpe_StartingLBA;
2870 		}
2871 
2872 		if (cl->cl_reserved == -1) {
2873 			if (bcmp(&partitions->efi_gpe_PartitionTypeGUID,
2874 			    &uuid_type_reserved, sizeof (struct uuid)) == 0) {
2875 				cl->cl_reserved = i;
2876 			}
2877 		}
2878 		if (i == WD_NODE) {
2879 			/*
2880 			 * minor number 7 corresponds to the whole disk
2881 			 * if the disk capacity is expanded after disk is
2882 			 * labeled, minor number 7 represents the capacity
2883 			 * indicated by the disk label.
2884 			 */
2885 			cl->cl_map[i].dkl_cylno = 0;
2886 			if (alternate_lba == 1) {
2887 				/*
2888 				 * We are using backup label. Since we can
2889 				 * find a valid label at the end of disk,
2890 				 * the disk capacity is not expanded.
2891 				 */
2892 				cl->cl_map[i].dkl_nblk = capacity;
2893 			} else {
2894 				cl->cl_map[i].dkl_nblk = alternate_lba + 1;
2895 			}
2896 			cl->cl_offset[i] = 0;
2897 		}
2898 		partitions++;
2899 	}
2900 	cl->cl_solaris_offset = 0;
2901 	cl->cl_solaris_size = capacity;
2902 	cl->cl_label_from_media = CMLB_LABEL_EFI;
2903 	cl->cl_f_geometry_is_valid = B_TRUE;
2904 
2905 	/* clear the vtoc label */
2906 	bzero(&cl->cl_vtoc, sizeof (struct dk_vtoc));
2907 
2908 	kmem_free(buf, EFI_MIN_ARRAY_SIZE);
2909 	return (0);
2910 
2911 done_err:
2912 	kmem_free(buf, EFI_MIN_ARRAY_SIZE);
2913 	mutex_enter(CMLB_MUTEX(cl));
2914 done_err1:
2915 	/*
2916 	 * if we didn't find something that could look like a VTOC
2917 	 * and the disk is over 1TB, we know there isn't a valid label.
2918 	 * Otherwise let cmlb_uselabel decide what to do.  We only
2919 	 * want to invalidate this if we're certain the label isn't
2920 	 * valid because cmlb_prop_op will now fail, which in turn
2921 	 * causes things like opens and stats on the partition to fail.
2922 	 */
2923 	if ((capacity > CMLB_EXTVTOC_LIMIT) && (rval != ESRCH) && !iofailed) {
2924 		cl->cl_f_geometry_is_valid = B_FALSE;
2925 	}
2926 	return (rval);
2927 }
2928 
2929 
2930 /*
2931  *    Function: cmlb_uselabel
2932  *
2933  * Description: Validate the disk label and update the relevant data (geometry,
2934  *		partition, vtoc, and capacity data) in the cmlb_lun struct.
2935  *		Marks the geometry of the unit as being valid.
2936  *
2937  *   Arguments: cl: unit struct.
2938  *		dk_label: disk label
2939  *
2940  * Return Code: CMLB_LABEL_IS_VALID: Label read from disk is OK; geometry,
2941  *		partition, vtoc, and capacity data are good.
2942  *
2943  *		CMLB_LABEL_IS_INVALID: Magic number or checksum error in the
2944  *		label; or computed capacity does not jibe with capacity
2945  *		reported from the READ CAPACITY command.
2946  *
2947  *     Context: Kernel thread only (can sleep).
2948  */
2949 static int
2950 cmlb_uselabel(struct cmlb_lun *cl, struct dk_label *labp, int flags)
2951 {
2952 	short		*sp;
2953 	short		sum;
2954 	short		count;
2955 	int		label_error = CMLB_LABEL_IS_VALID;
2956 	int		i;
2957 	diskaddr_t	label_capacity;
2958 	uint32_t	part_end;
2959 	diskaddr_t	track_capacity;
2960 #if defined(_SUNOS_VTOC_16)
2961 	struct	dkl_partition	*vpartp;
2962 #endif
2963 	ASSERT(cl != NULL);
2964 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
2965 
2966 	/* Validate the magic number of the label. */
2967 	if (labp->dkl_magic != DKL_MAGIC) {
2968 #if defined(__sparc)
2969 		if (!ISREMOVABLE(cl) && !ISHOTPLUGGABLE(cl)) {
2970 			if (!(flags & CMLB_SILENT))
2971 				cmlb_log(CMLB_DEVINFO(cl), CMLB_LABEL(cl),
2972 				    CE_WARN,
2973 				    "Corrupt label; wrong magic number\n");
2974 		}
2975 #endif
2976 		return (CMLB_LABEL_IS_INVALID);
2977 	}
2978 
2979 	/* Validate the checksum of the label. */
2980 	sp  = (short *)labp;
2981 	sum = 0;
2982 	count = sizeof (struct dk_label) / sizeof (short);
2983 	while (count--)	 {
2984 		sum ^= *sp++;
2985 	}
2986 
2987 	if (sum != 0) {
2988 #if defined(_SUNOS_VTOC_16)
2989 		if (!ISCD(cl)) {
2990 #elif defined(_SUNOS_VTOC_8)
2991 		if (!ISREMOVABLE(cl) && !ISHOTPLUGGABLE(cl)) {
2992 #endif
2993 			if (!(flags & CMLB_SILENT))
2994 				cmlb_log(CMLB_DEVINFO(cl), CMLB_LABEL(cl),
2995 				    CE_WARN,
2996 				    "Corrupt label - label checksum failed\n");
2997 		}
2998 		return (CMLB_LABEL_IS_INVALID);
2999 	}
3000 
3001 
3002 	/*
3003 	 * Fill in geometry structure with data from label.
3004 	 */
3005 	bzero(&cl->cl_g, sizeof (struct dk_geom));
3006 	cl->cl_g.dkg_ncyl   = labp->dkl_ncyl;
3007 	cl->cl_g.dkg_acyl   = labp->dkl_acyl;
3008 	cl->cl_g.dkg_bcyl   = 0;
3009 	cl->cl_g.dkg_nhead  = labp->dkl_nhead;
3010 	cl->cl_g.dkg_nsect  = labp->dkl_nsect;
3011 	cl->cl_g.dkg_intrlv = labp->dkl_intrlv;
3012 
3013 #if defined(_SUNOS_VTOC_8)
3014 	cl->cl_g.dkg_gap1   = labp->dkl_gap1;
3015 	cl->cl_g.dkg_gap2   = labp->dkl_gap2;
3016 	cl->cl_g.dkg_bhead  = labp->dkl_bhead;
3017 #endif
3018 #if defined(_SUNOS_VTOC_16)
3019 	cl->cl_dkg_skew = labp->dkl_skew;
3020 #endif
3021 
3022 #if defined(__i386) || defined(__amd64)
3023 	cl->cl_g.dkg_apc = labp->dkl_apc;
3024 #endif
3025 
3026 	/*
3027 	 * Currently we rely on the values in the label being accurate. If
3028 	 * dkl_rpm or dkl_pcly are zero in the label, use a default value.
3029 	 *
3030 	 * Note: In the future a MODE SENSE may be used to retrieve this data,
3031 	 * although this command is optional in SCSI-2.
3032 	 */
3033 	cl->cl_g.dkg_rpm  = (labp->dkl_rpm  != 0) ? labp->dkl_rpm  : 3600;
3034 	cl->cl_g.dkg_pcyl = (labp->dkl_pcyl != 0) ? labp->dkl_pcyl :
3035 	    (cl->cl_g.dkg_ncyl + cl->cl_g.dkg_acyl);
3036 
3037 	/*
3038 	 * The Read and Write reinstruct values may not be valid
3039 	 * for older disks.
3040 	 */
3041 	cl->cl_g.dkg_read_reinstruct  = labp->dkl_read_reinstruct;
3042 	cl->cl_g.dkg_write_reinstruct = labp->dkl_write_reinstruct;
3043 
3044 	/* Fill in partition table. */
3045 #if defined(_SUNOS_VTOC_8)
3046 	for (i = 0; i < NDKMAP; i++) {
3047 		cl->cl_map[i].dkl_cylno = labp->dkl_map[i].dkl_cylno;
3048 		cl->cl_map[i].dkl_nblk  = labp->dkl_map[i].dkl_nblk;
3049 	}
3050 #endif
3051 #if  defined(_SUNOS_VTOC_16)
3052 	vpartp		= labp->dkl_vtoc.v_part;
3053 	track_capacity	= labp->dkl_nhead * labp->dkl_nsect;
3054 
3055 	/* Prevent divide by zero */
3056 	if (track_capacity == 0) {
3057 		if (!(flags & CMLB_SILENT))
3058 			cmlb_log(CMLB_DEVINFO(cl), CMLB_LABEL(cl), CE_WARN,
3059 			    "Corrupt label - zero nhead or nsect value\n");
3060 
3061 		return (CMLB_LABEL_IS_INVALID);
3062 	}
3063 
3064 	for (i = 0; i < NDKMAP; i++, vpartp++) {
3065 		cl->cl_map[i].dkl_cylno = vpartp->p_start / track_capacity;
3066 		cl->cl_map[i].dkl_nblk  = vpartp->p_size;
3067 	}
3068 #endif
3069 
3070 	/* Fill in VTOC Structure. */
3071 	bcopy(&labp->dkl_vtoc, &cl->cl_vtoc, sizeof (struct dk_vtoc));
3072 #if defined(_SUNOS_VTOC_8)
3073 	/*
3074 	 * The 8-slice vtoc does not include the ascii label; save it into
3075 	 * the device's soft state structure here.
3076 	 */
3077 	bcopy(labp->dkl_asciilabel, cl->cl_asciilabel, LEN_DKL_ASCII);
3078 #endif
3079 
3080 	/* Now look for a valid capacity. */
3081 	track_capacity	= (cl->cl_g.dkg_nhead * cl->cl_g.dkg_nsect);
3082 	label_capacity	= (cl->cl_g.dkg_ncyl  * track_capacity);
3083 
3084 	if (cl->cl_g.dkg_acyl) {
3085 #if defined(__i386) || defined(__amd64)
3086 		/* we may have > 1 alts cylinder */
3087 		label_capacity += (track_capacity * cl->cl_g.dkg_acyl);
3088 #else
3089 		label_capacity += track_capacity;
3090 #endif
3091 	}
3092 
3093 	/*
3094 	 * Force check here to ensure the computed capacity is valid.
3095 	 * If capacity is zero, it indicates an invalid label and
3096 	 * we should abort updating the relevant data then.
3097 	 */
3098 	if (label_capacity == 0) {
3099 		if (!(flags & CMLB_SILENT))
3100 			cmlb_log(CMLB_DEVINFO(cl), CMLB_LABEL(cl), CE_WARN,
3101 			    "Corrupt label - no valid capacity could be "
3102 			    "retrieved\n");
3103 
3104 		return (CMLB_LABEL_IS_INVALID);
3105 	}
3106 
3107 	/* Mark the geometry as valid. */
3108 	cl->cl_f_geometry_is_valid = B_TRUE;
3109 
3110 	/*
3111 	 * if we got invalidated when mutex exit and entered again,
3112 	 * if blockcount different than when we came in, need to
3113 	 * retry from beginning of cmlb_validate_geometry.
3114 	 * revisit this on next phase of utilizing this for
3115 	 * sd.
3116 	 */
3117 
3118 	if (label_capacity <= cl->cl_blockcount) {
3119 #if defined(_SUNOS_VTOC_8)
3120 		/*
3121 		 * We can't let this happen on drives that are subdivided
3122 		 * into logical disks (i.e., that have an fdisk table).
3123 		 * The cl_blockcount field should always hold the full media
3124 		 * size in sectors, period.  This code would overwrite
3125 		 * cl_blockcount with the size of the Solaris fdisk partition.
3126 		 */
3127 		cmlb_dbg(CMLB_ERROR,  cl,
3128 		    "cmlb_uselabel: Label %d blocks; Drive %d blocks\n",
3129 		    label_capacity, cl->cl_blockcount);
3130 		cl->cl_solaris_size = label_capacity;
3131 
3132 #endif	/* defined(_SUNOS_VTOC_8) */
3133 		goto done;
3134 	}
3135 
3136 	if (ISCD(cl)) {
3137 		/* For CDROMs, we trust that the data in the label is OK. */
3138 #if defined(_SUNOS_VTOC_8)
3139 		for (i = 0; i < NDKMAP; i++) {
3140 			part_end = labp->dkl_nhead * labp->dkl_nsect *
3141 			    labp->dkl_map[i].dkl_cylno +
3142 			    labp->dkl_map[i].dkl_nblk  - 1;
3143 
3144 			if ((labp->dkl_map[i].dkl_nblk) &&
3145 			    (part_end > cl->cl_blockcount)) {
3146 				cl->cl_f_geometry_is_valid = B_FALSE;
3147 				break;
3148 			}
3149 		}
3150 #endif
3151 #if defined(_SUNOS_VTOC_16)
3152 		vpartp = &(labp->dkl_vtoc.v_part[0]);
3153 		for (i = 0; i < NDKMAP; i++, vpartp++) {
3154 			part_end = vpartp->p_start + vpartp->p_size;
3155 			if ((vpartp->p_size > 0) &&
3156 			    (part_end > cl->cl_blockcount)) {
3157 				cl->cl_f_geometry_is_valid = B_FALSE;
3158 				break;
3159 			}
3160 		}
3161 #endif
3162 	} else {
3163 		/* label_capacity > cl->cl_blockcount */
3164 		if (!(flags & CMLB_SILENT)) {
3165 			cmlb_log(CMLB_DEVINFO(cl), CMLB_LABEL(cl), CE_WARN,
3166 			    "Corrupt label - bad geometry\n");
3167 			cmlb_log(CMLB_DEVINFO(cl), CMLB_LABEL(cl), CE_CONT,
3168 			    "Label says %llu blocks; Drive says %llu blocks\n",
3169 			    label_capacity, cl->cl_blockcount);
3170 		}
3171 		cl->cl_f_geometry_is_valid = B_FALSE;
3172 		label_error = CMLB_LABEL_IS_INVALID;
3173 	}
3174 
3175 done:
3176 
3177 	cmlb_dbg(CMLB_INFO,  cl, "cmlb_uselabel: (label geometry)\n");
3178 	cmlb_dbg(CMLB_INFO,  cl,
3179 	    "   ncyl: %d; acyl: %d; nhead: %d; nsect: %d\n",
3180 	    cl->cl_g.dkg_ncyl,  cl->cl_g.dkg_acyl,
3181 	    cl->cl_g.dkg_nhead, cl->cl_g.dkg_nsect);
3182 
3183 	cmlb_dbg(CMLB_INFO,  cl,
3184 	    "   label_capacity: %d; intrlv: %d; rpm: %d\n",
3185 	    cl->cl_blockcount, cl->cl_g.dkg_intrlv, cl->cl_g.dkg_rpm);
3186 	cmlb_dbg(CMLB_INFO,  cl, "   wrt_reinstr: %d; rd_reinstr: %d\n",
3187 	    cl->cl_g.dkg_write_reinstruct, cl->cl_g.dkg_read_reinstruct);
3188 
3189 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
3190 
3191 	return (label_error);
3192 }
3193 
3194 
3195 /*
3196  *    Function: cmlb_build_default_label
3197  *
3198  * Description: Generate a default label for those devices that do not have
3199  *		one, e.g., new media, removable cartridges, etc..
3200  *
3201  *     Context: Kernel thread only
3202  */
3203 /*ARGSUSED*/
3204 static void
3205 cmlb_build_default_label(struct cmlb_lun *cl, void *tg_cookie)
3206 {
3207 #if defined(_SUNOS_VTOC_16)
3208 	uint_t	phys_spc;
3209 	uint_t	disksize;
3210 	struct  dk_geom cl_g;
3211 	diskaddr_t capacity;
3212 #endif
3213 
3214 	ASSERT(cl != NULL);
3215 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
3216 
3217 #if defined(_SUNOS_VTOC_8)
3218 	/*
3219 	 * Note: This is a legacy check for non-removable devices on VTOC_8
3220 	 * only. This may be a valid check for VTOC_16 as well.
3221 	 * Once we understand why there is this difference between SPARC and
3222 	 * x86 platform, we could remove this legacy check.
3223 	 */
3224 	if (!ISREMOVABLE(cl) && !ISHOTPLUGGABLE(cl)) {
3225 		return;
3226 	}
3227 #endif
3228 
3229 	bzero(&cl->cl_g, sizeof (struct dk_geom));
3230 	bzero(&cl->cl_vtoc, sizeof (struct dk_vtoc));
3231 	bzero(&cl->cl_map, NDKMAP * (sizeof (struct dk_map)));
3232 
3233 #if defined(_SUNOS_VTOC_8)
3234 
3235 	/*
3236 	 * It's a REMOVABLE media, therefore no label (on sparc, anyway).
3237 	 * But it is still necessary to set up various geometry information,
3238 	 * and we are doing this here.
3239 	 */
3240 
3241 	/*
3242 	 * For the rpm, we use the minimum for the disk.  For the head, cyl,
3243 	 * and number of sector per track, if the capacity <= 1GB, head = 64,
3244 	 * sect = 32.  else head = 255, sect 63 Note: the capacity should be
3245 	 * equal to C*H*S values.  This will cause some truncation of size due
3246 	 * to round off errors. For CD-ROMs, this truncation can have adverse
3247 	 * side effects, so returning ncyl and nhead as 1. The nsect will
3248 	 * overflow for most of CD-ROMs as nsect is of type ushort. (4190569)
3249 	 */
3250 	cl->cl_solaris_size = cl->cl_blockcount;
3251 	if (ISCD(cl)) {
3252 		tg_attribute_t tgattribute;
3253 		int is_writable;
3254 		/*
3255 		 * Preserve the old behavior for non-writable
3256 		 * medias. Since dkg_nsect is a ushort, it
3257 		 * will lose bits as cdroms have more than
3258 		 * 65536 sectors. So if we recalculate
3259 		 * capacity, it will become much shorter.
3260 		 * But the dkg_* information is not
3261 		 * used for CDROMs so it is OK. But for
3262 		 * Writable CDs we need this information
3263 		 * to be valid (for newfs say). So we
3264 		 * make nsect and nhead > 1 that way
3265 		 * nsect can still stay within ushort limit
3266 		 * without losing any bits.
3267 		 */
3268 
3269 		bzero(&tgattribute, sizeof (tg_attribute_t));
3270 
3271 		mutex_exit(CMLB_MUTEX(cl));
3272 		is_writable =
3273 		    (DK_TG_GETATTRIBUTE(cl, &tgattribute, tg_cookie) == 0) ?
3274 		    tgattribute.media_is_writable : 1;
3275 		mutex_enter(CMLB_MUTEX(cl));
3276 
3277 		if (is_writable) {
3278 			cl->cl_g.dkg_nhead = 64;
3279 			cl->cl_g.dkg_nsect = 32;
3280 			cl->cl_g.dkg_ncyl = cl->cl_blockcount / (64 * 32);
3281 			cl->cl_solaris_size = (diskaddr_t)cl->cl_g.dkg_ncyl *
3282 			    cl->cl_g.dkg_nhead * cl->cl_g.dkg_nsect;
3283 		} else {
3284 			cl->cl_g.dkg_ncyl  = 1;
3285 			cl->cl_g.dkg_nhead = 1;
3286 			cl->cl_g.dkg_nsect = cl->cl_blockcount;
3287 		}
3288 	} else {
3289 		if (cl->cl_blockcount <= 0x1000) {
3290 			/* unlabeled SCSI floppy device */
3291 			cl->cl_g.dkg_nhead = 2;
3292 			cl->cl_g.dkg_ncyl = 80;
3293 			cl->cl_g.dkg_nsect = cl->cl_blockcount / (2 * 80);
3294 		} else if (cl->cl_blockcount <= 0x200000) {
3295 			cl->cl_g.dkg_nhead = 64;
3296 			cl->cl_g.dkg_nsect = 32;
3297 			cl->cl_g.dkg_ncyl  = cl->cl_blockcount / (64 * 32);
3298 		} else {
3299 			cl->cl_g.dkg_nhead = 255;
3300 
3301 			cl->cl_g.dkg_nsect = ((cl->cl_blockcount +
3302 			    (UINT16_MAX * 255 * 63) - 1) /
3303 			    (UINT16_MAX * 255 * 63)) * 63;
3304 
3305 			if (cl->cl_g.dkg_nsect == 0)
3306 				cl->cl_g.dkg_nsect = (UINT16_MAX / 63) * 63;
3307 
3308 			cl->cl_g.dkg_ncyl = cl->cl_blockcount /
3309 			    (255 * cl->cl_g.dkg_nsect);
3310 		}
3311 
3312 		cl->cl_solaris_size =
3313 		    (diskaddr_t)cl->cl_g.dkg_ncyl * cl->cl_g.dkg_nhead *
3314 		    cl->cl_g.dkg_nsect;
3315 
3316 	}
3317 
3318 	cl->cl_g.dkg_acyl	= 0;
3319 	cl->cl_g.dkg_bcyl	= 0;
3320 	cl->cl_g.dkg_rpm	= 200;
3321 	cl->cl_asciilabel[0]	= '\0';
3322 	cl->cl_g.dkg_pcyl	= cl->cl_g.dkg_ncyl;
3323 
3324 	cl->cl_map[0].dkl_cylno = 0;
3325 	cl->cl_map[0].dkl_nblk  = cl->cl_solaris_size;
3326 
3327 	cl->cl_map[2].dkl_cylno = 0;
3328 	cl->cl_map[2].dkl_nblk  = cl->cl_solaris_size;
3329 
3330 #elif defined(_SUNOS_VTOC_16)
3331 
3332 	if (cl->cl_solaris_size == 0) {
3333 		/*
3334 		 * Got fdisk table but no solaris entry therefore
3335 		 * don't create a default label
3336 		 */
3337 		cl->cl_f_geometry_is_valid = B_TRUE;
3338 		return;
3339 	}
3340 
3341 	/*
3342 	 * For CDs we continue to use the physical geometry to calculate
3343 	 * number of cylinders. All other devices must convert the
3344 	 * physical geometry (cmlb_geom) to values that will fit
3345 	 * in a dk_geom structure.
3346 	 */
3347 	if (ISCD(cl)) {
3348 		phys_spc = cl->cl_pgeom.g_nhead * cl->cl_pgeom.g_nsect;
3349 	} else {
3350 		/* Convert physical geometry to disk geometry */
3351 		bzero(&cl_g, sizeof (struct dk_geom));
3352 
3353 		/*
3354 		 * Refer to comments related to off-by-1 at the
3355 		 * header of this file.
3356 		 * Before calculating geometry, capacity should be
3357 		 * decreased by 1.
3358 		 */
3359 
3360 		if (cl->cl_alter_behavior & CMLB_OFF_BY_ONE)
3361 			capacity = cl->cl_blockcount - 1;
3362 		else
3363 			capacity = cl->cl_blockcount;
3364 
3365 
3366 		cmlb_convert_geometry(cl, capacity, &cl_g, tg_cookie);
3367 		bcopy(&cl_g, &cl->cl_g, sizeof (cl->cl_g));
3368 		phys_spc = cl->cl_g.dkg_nhead * cl->cl_g.dkg_nsect;
3369 	}
3370 
3371 	if (phys_spc == 0)
3372 		return;
3373 	cl->cl_g.dkg_pcyl = cl->cl_solaris_size / phys_spc;
3374 	if (cl->cl_alter_behavior & CMLB_FAKE_LABEL_ONE_PARTITION) {
3375 		/* disable devid */
3376 		cl->cl_g.dkg_ncyl = cl->cl_g.dkg_pcyl;
3377 		disksize = cl->cl_solaris_size;
3378 	} else {
3379 		cl->cl_g.dkg_acyl = DK_ACYL;
3380 		cl->cl_g.dkg_ncyl = cl->cl_g.dkg_pcyl - DK_ACYL;
3381 		disksize = cl->cl_g.dkg_ncyl * phys_spc;
3382 	}
3383 
3384 	if (ISCD(cl)) {
3385 		/*
3386 		 * CD's don't use the "heads * sectors * cyls"-type of
3387 		 * geometry, but instead use the entire capacity of the media.
3388 		 */
3389 		disksize = cl->cl_solaris_size;
3390 		cl->cl_g.dkg_nhead = 1;
3391 		cl->cl_g.dkg_nsect = 1;
3392 		cl->cl_g.dkg_rpm =
3393 		    (cl->cl_pgeom.g_rpm == 0) ? 200 : cl->cl_pgeom.g_rpm;
3394 
3395 		cl->cl_vtoc.v_part[0].p_start = 0;
3396 		cl->cl_vtoc.v_part[0].p_size  = disksize;
3397 		cl->cl_vtoc.v_part[0].p_tag   = V_BACKUP;
3398 		cl->cl_vtoc.v_part[0].p_flag  = V_UNMNT;
3399 
3400 		cl->cl_map[0].dkl_cylno = 0;
3401 		cl->cl_map[0].dkl_nblk  = disksize;
3402 		cl->cl_offset[0] = 0;
3403 
3404 	} else {
3405 		/*
3406 		 * Hard disks and removable media cartridges
3407 		 */
3408 		cl->cl_g.dkg_rpm =
3409 		    (cl->cl_pgeom.g_rpm == 0) ? 3600: cl->cl_pgeom.g_rpm;
3410 		cl->cl_vtoc.v_sectorsz = cl->cl_sys_blocksize;
3411 
3412 		/* Add boot slice */
3413 		cl->cl_vtoc.v_part[8].p_start = 0;
3414 		cl->cl_vtoc.v_part[8].p_size  = phys_spc;
3415 		cl->cl_vtoc.v_part[8].p_tag   = V_BOOT;
3416 		cl->cl_vtoc.v_part[8].p_flag  = V_UNMNT;
3417 
3418 		cl->cl_map[8].dkl_cylno = 0;
3419 		cl->cl_map[8].dkl_nblk  = phys_spc;
3420 		cl->cl_offset[8] = 0;
3421 
3422 		if ((cl->cl_alter_behavior &
3423 		    CMLB_CREATE_ALTSLICE_VTOC_16_DTYPE_DIRECT) &&
3424 		    cl->cl_device_type == DTYPE_DIRECT) {
3425 			cl->cl_vtoc.v_part[9].p_start = phys_spc;
3426 			cl->cl_vtoc.v_part[9].p_size  = 2 * phys_spc;
3427 			cl->cl_vtoc.v_part[9].p_tag   = V_ALTSCTR;
3428 			cl->cl_vtoc.v_part[9].p_flag  = 0;
3429 
3430 			cl->cl_map[9].dkl_cylno = 1;
3431 			cl->cl_map[9].dkl_nblk  = 2 * phys_spc;
3432 			cl->cl_offset[9] = phys_spc;
3433 		}
3434 	}
3435 
3436 	cl->cl_g.dkg_apc = 0;
3437 
3438 	/* Add backup slice */
3439 	cl->cl_vtoc.v_part[2].p_start = 0;
3440 	cl->cl_vtoc.v_part[2].p_size  = disksize;
3441 	cl->cl_vtoc.v_part[2].p_tag   = V_BACKUP;
3442 	cl->cl_vtoc.v_part[2].p_flag  = V_UNMNT;
3443 
3444 	cl->cl_map[2].dkl_cylno = 0;
3445 	cl->cl_map[2].dkl_nblk  = disksize;
3446 	cl->cl_offset[2] = 0;
3447 
3448 	/*
3449 	 * single slice (s0) covering the entire disk
3450 	 */
3451 	if (cl->cl_alter_behavior & CMLB_FAKE_LABEL_ONE_PARTITION) {
3452 		cl->cl_vtoc.v_part[0].p_start = 0;
3453 		cl->cl_vtoc.v_part[0].p_tag   = V_UNASSIGNED;
3454 		cl->cl_vtoc.v_part[0].p_flag  = 0;
3455 		cl->cl_vtoc.v_part[0].p_size  = disksize;
3456 		cl->cl_map[0].dkl_cylno = 0;
3457 		cl->cl_map[0].dkl_nblk  = disksize;
3458 		cl->cl_offset[0] = 0;
3459 	}
3460 
3461 	(void) sprintf(cl->cl_vtoc.v_asciilabel, "DEFAULT cyl %d alt %d"
3462 	    " hd %d sec %d", cl->cl_g.dkg_ncyl, cl->cl_g.dkg_acyl,
3463 	    cl->cl_g.dkg_nhead, cl->cl_g.dkg_nsect);
3464 
3465 #else
3466 #error "No VTOC format defined."
3467 #endif
3468 
3469 	cl->cl_g.dkg_read_reinstruct  = 0;
3470 	cl->cl_g.dkg_write_reinstruct = 0;
3471 
3472 	cl->cl_g.dkg_intrlv = 1;
3473 
3474 	cl->cl_vtoc.v_sanity  = VTOC_SANE;
3475 	cl->cl_vtoc.v_nparts = V_NUMPAR;
3476 	cl->cl_vtoc.v_version = V_VERSION;
3477 
3478 	cl->cl_f_geometry_is_valid = B_TRUE;
3479 	cl->cl_label_from_media = CMLB_LABEL_UNDEF;
3480 
3481 	cmlb_dbg(CMLB_INFO,  cl,
3482 	    "cmlb_build_default_label: Default label created: "
3483 	    "cyl: %d\tacyl: %d\tnhead: %d\tnsect: %d\tcap: %d\n",
3484 	    cl->cl_g.dkg_ncyl, cl->cl_g.dkg_acyl, cl->cl_g.dkg_nhead,
3485 	    cl->cl_g.dkg_nsect, cl->cl_blockcount);
3486 }
3487 
3488 
3489 #if defined(_FIRMWARE_NEEDS_FDISK)
3490 /*
3491  * Max CHS values, as they are encoded into bytes, for 1022/254/63
3492  */
3493 #define	LBA_MAX_SECT	(63 | ((1022 & 0x300) >> 2))
3494 #define	LBA_MAX_CYL	(1022 & 0xFF)
3495 #define	LBA_MAX_HEAD	(254)
3496 
3497 
3498 /*
3499  *    Function: cmlb_has_max_chs_vals
3500  *
3501  * Description: Return B_TRUE if Cylinder-Head-Sector values are all at maximum.
3502  *
3503  *   Arguments: fdp - ptr to CHS info
3504  *
3505  * Return Code: True or false
3506  *
3507  *     Context: Any.
3508  */
3509 static boolean_t
3510 cmlb_has_max_chs_vals(struct ipart *fdp)
3511 {
3512 	return ((fdp->begcyl  == LBA_MAX_CYL)	&&
3513 	    (fdp->beghead == LBA_MAX_HEAD)	&&
3514 	    (fdp->begsect == LBA_MAX_SECT)	&&
3515 	    (fdp->endcyl  == LBA_MAX_CYL)	&&
3516 	    (fdp->endhead == LBA_MAX_HEAD)	&&
3517 	    (fdp->endsect == LBA_MAX_SECT));
3518 }
3519 #endif
3520 
3521 /*
3522  *    Function: cmlb_dkio_get_geometry
3523  *
3524  * Description: This routine is the driver entry point for handling user
3525  *		requests to get the device geometry (DKIOCGGEOM).
3526  *
3527  *   Arguments:
3528  *	arg		pointer to user provided dk_geom structure specifying
3529  *			the controller's notion of the current geometry.
3530  *
3531  *	flag 		this argument is a pass through to ddi_copyxxx()
3532  *			directly from the mode argument of ioctl().
3533  *
3534  *	tg_cookie	cookie from target driver to be passed back to target
3535  *			driver when we call back to it through tg_ops.
3536  *
3537  * Return Code: 0
3538  *		EFAULT
3539  *		ENXIO
3540  *		EIO
3541  */
3542 static int
3543 cmlb_dkio_get_geometry(struct cmlb_lun *cl, caddr_t arg, int flag,
3544     void *tg_cookie)
3545 {
3546 	struct dk_geom	*tmp_geom = NULL;
3547 	int		rval = 0;
3548 
3549 	/*
3550 	 * cmlb_validate_geometry does not spin a disk up
3551 	 * if it was spcl down. We need to make sure it
3552 	 * is ready.
3553 	 */
3554 	mutex_enter(CMLB_MUTEX(cl));
3555 	rval = cmlb_validate_geometry(cl, B_TRUE, 0, tg_cookie);
3556 #if defined(_SUNOS_VTOC_8)
3557 	if (rval == EINVAL &&
3558 	    cl->cl_alter_behavior & CMLB_FAKE_GEOM_LABEL_IOCTLS_VTOC8) {
3559 		/*
3560 		 * This is to return a default label geometry even when we
3561 		 * do not really assume a default label for the device.
3562 		 * dad driver utilizes this.
3563 		 */
3564 		if (cl->cl_blockcount <= CMLB_OLDVTOC_LIMIT) {
3565 			cmlb_setup_default_geometry(cl, tg_cookie);
3566 			rval = 0;
3567 		}
3568 	}
3569 #endif
3570 	if (rval) {
3571 		mutex_exit(CMLB_MUTEX(cl));
3572 		return (rval);
3573 	}
3574 
3575 #if defined(__i386) || defined(__amd64)
3576 	if (cl->cl_solaris_size == 0) {
3577 		mutex_exit(CMLB_MUTEX(cl));
3578 		return (EIO);
3579 	}
3580 #endif
3581 
3582 	/*
3583 	 * Make a local copy of the soft state geometry to avoid some potential
3584 	 * race conditions associated with holding the mutex and updating the
3585 	 * write_reinstruct value
3586 	 */
3587 	tmp_geom = kmem_zalloc(sizeof (struct dk_geom), KM_SLEEP);
3588 	bcopy(&cl->cl_g, tmp_geom, sizeof (struct dk_geom));
3589 
3590 	if (tmp_geom->dkg_write_reinstruct == 0) {
3591 		tmp_geom->dkg_write_reinstruct =
3592 		    (int)((int)(tmp_geom->dkg_nsect * tmp_geom->dkg_rpm *
3593 		    cmlb_rot_delay) / (int)60000);
3594 	}
3595 	mutex_exit(CMLB_MUTEX(cl));
3596 
3597 	rval = ddi_copyout(tmp_geom, (void *)arg, sizeof (struct dk_geom),
3598 	    flag);
3599 	if (rval != 0) {
3600 		rval = EFAULT;
3601 	}
3602 
3603 	kmem_free(tmp_geom, sizeof (struct dk_geom));
3604 	return (rval);
3605 
3606 }
3607 
3608 
3609 /*
3610  *    Function: cmlb_dkio_set_geometry
3611  *
3612  * Description: This routine is the driver entry point for handling user
3613  *		requests to set the device geometry (DKIOCSGEOM). The actual
3614  *		device geometry is not updated, just the driver "notion" of it.
3615  *
3616  *   Arguments:
3617  *	arg		pointer to user provided dk_geom structure used to set
3618  *			the controller's notion of the current geometry.
3619  *
3620  *	flag 		this argument is a pass through to ddi_copyxxx()
3621  *			directly from the mode argument of ioctl().
3622  *
3623  *	tg_cookie	cookie from target driver to be passed back to target
3624  *			driver when we call back to it through tg_ops.
3625  *
3626  * Return Code: 0
3627  *		EFAULT
3628  *		ENXIO
3629  *		EIO
3630  */
3631 static int
3632 cmlb_dkio_set_geometry(struct cmlb_lun *cl, caddr_t arg, int flag)
3633 {
3634 	struct dk_geom	*tmp_geom;
3635 	struct dk_map	*lp;
3636 	int		rval = 0;
3637 	int		i;
3638 
3639 
3640 #if defined(__i386) || defined(__amd64)
3641 	if (cl->cl_solaris_size == 0) {
3642 		return (EIO);
3643 	}
3644 #endif
3645 	/*
3646 	 * We need to copy the user specified geometry into local
3647 	 * storage and then update the softstate. We don't want to hold
3648 	 * the mutex and copyin directly from the user to the soft state
3649 	 */
3650 	tmp_geom = (struct dk_geom *)
3651 	    kmem_zalloc(sizeof (struct dk_geom), KM_SLEEP);
3652 	rval = ddi_copyin(arg, tmp_geom, sizeof (struct dk_geom), flag);
3653 	if (rval != 0) {
3654 		kmem_free(tmp_geom, sizeof (struct dk_geom));
3655 		return (EFAULT);
3656 	}
3657 
3658 	mutex_enter(CMLB_MUTEX(cl));
3659 	bcopy(tmp_geom, &cl->cl_g, sizeof (struct dk_geom));
3660 	for (i = 0; i < NDKMAP; i++) {
3661 		lp  = &cl->cl_map[i];
3662 		cl->cl_offset[i] =
3663 		    cl->cl_g.dkg_nhead * cl->cl_g.dkg_nsect * lp->dkl_cylno;
3664 #if defined(__i386) || defined(__amd64)
3665 		cl->cl_offset[i] += cl->cl_solaris_offset;
3666 #endif
3667 	}
3668 	cl->cl_f_geometry_is_valid = B_FALSE;
3669 	mutex_exit(CMLB_MUTEX(cl));
3670 	kmem_free(tmp_geom, sizeof (struct dk_geom));
3671 
3672 	return (rval);
3673 }
3674 
3675 /*
3676  *    Function: cmlb_dkio_get_partition
3677  *
3678  * Description: This routine is the driver entry point for handling user
3679  *		requests to get the partition table (DKIOCGAPART).
3680  *
3681  *   Arguments:
3682  *	arg		pointer to user provided dk_allmap structure specifying
3683  *			the controller's notion of the current partition table.
3684  *
3685  *	flag		this argument is a pass through to ddi_copyxxx()
3686  *			directly from the mode argument of ioctl().
3687  *
3688  *	tg_cookie	cookie from target driver to be passed back to target
3689  *			driver when we call back to it through tg_ops.
3690  *
3691  * Return Code: 0
3692  *		EFAULT
3693  *		ENXIO
3694  *		EIO
3695  */
3696 static int
3697 cmlb_dkio_get_partition(struct cmlb_lun *cl, caddr_t arg, int flag,
3698     void *tg_cookie)
3699 {
3700 	int		rval = 0;
3701 	int		size;
3702 
3703 	/*
3704 	 * Make sure the geometry is valid before getting the partition
3705 	 * information.
3706 	 */
3707 	mutex_enter(CMLB_MUTEX(cl));
3708 	if ((rval = cmlb_validate_geometry(cl, B_TRUE, 0, tg_cookie)) != 0) {
3709 		mutex_exit(CMLB_MUTEX(cl));
3710 		return (rval);
3711 	}
3712 	mutex_exit(CMLB_MUTEX(cl));
3713 
3714 #if defined(__i386) || defined(__amd64)
3715 	if (cl->cl_solaris_size == 0) {
3716 		return (EIO);
3717 	}
3718 #endif
3719 
3720 #ifdef _MULTI_DATAMODEL
3721 	switch (ddi_model_convert_from(flag & FMODELS)) {
3722 	case DDI_MODEL_ILP32: {
3723 		struct dk_map32 dk_map32[NDKMAP];
3724 		int		i;
3725 
3726 		for (i = 0; i < NDKMAP; i++) {
3727 			dk_map32[i].dkl_cylno = cl->cl_map[i].dkl_cylno;
3728 			dk_map32[i].dkl_nblk  = cl->cl_map[i].dkl_nblk;
3729 		}
3730 		size = NDKMAP * sizeof (struct dk_map32);
3731 		rval = ddi_copyout(dk_map32, (void *)arg, size, flag);
3732 		if (rval != 0) {
3733 			rval = EFAULT;
3734 		}
3735 		break;
3736 	}
3737 	case DDI_MODEL_NONE:
3738 		size = NDKMAP * sizeof (struct dk_map);
3739 		rval = ddi_copyout(cl->cl_map, (void *)arg, size, flag);
3740 		if (rval != 0) {
3741 			rval = EFAULT;
3742 		}
3743 		break;
3744 	}
3745 #else /* ! _MULTI_DATAMODEL */
3746 	size = NDKMAP * sizeof (struct dk_map);
3747 	rval = ddi_copyout(cl->cl_map, (void *)arg, size, flag);
3748 	if (rval != 0) {
3749 		rval = EFAULT;
3750 	}
3751 #endif /* _MULTI_DATAMODEL */
3752 	return (rval);
3753 }
3754 
3755 /*
3756  *    Function: cmlb_dkio_set_partition
3757  *
3758  * Description: This routine is the driver entry point for handling user
3759  *		requests to set the partition table (DKIOCSAPART). The actual
3760  *		device partition is not updated.
3761  *
3762  *   Arguments:
3763  *		arg  - pointer to user provided dk_allmap structure used to set
3764  *			the controller's notion of the partition table.
3765  *		flag - this argument is a pass through to ddi_copyxxx()
3766  *		       directly from the mode argument of ioctl().
3767  *
3768  * Return Code: 0
3769  *		EINVAL
3770  *		EFAULT
3771  *		ENXIO
3772  *		EIO
3773  */
3774 static int
3775 cmlb_dkio_set_partition(struct cmlb_lun *cl, caddr_t arg, int flag)
3776 {
3777 	struct dk_map	dk_map[NDKMAP];
3778 	struct dk_map	*lp;
3779 	int		rval = 0;
3780 	int		size;
3781 	int		i;
3782 #if defined(_SUNOS_VTOC_16)
3783 	struct dkl_partition	*vp;
3784 #endif
3785 
3786 	/*
3787 	 * Set the map for all logical partitions.  We lock
3788 	 * the priority just to make sure an interrupt doesn't
3789 	 * come in while the map is half updated.
3790 	 */
3791 	_NOTE(DATA_READABLE_WITHOUT_LOCK(cmlb_lun::cl_solaris_size))
3792 	mutex_enter(CMLB_MUTEX(cl));
3793 
3794 	if (cl->cl_blockcount > CMLB_OLDVTOC_LIMIT) {
3795 		mutex_exit(CMLB_MUTEX(cl));
3796 		return (ENOTSUP);
3797 	}
3798 	mutex_exit(CMLB_MUTEX(cl));
3799 	if (cl->cl_solaris_size == 0) {
3800 		return (EIO);
3801 	}
3802 
3803 #ifdef _MULTI_DATAMODEL
3804 	switch (ddi_model_convert_from(flag & FMODELS)) {
3805 	case DDI_MODEL_ILP32: {
3806 		struct dk_map32 dk_map32[NDKMAP];
3807 
3808 		size = NDKMAP * sizeof (struct dk_map32);
3809 		rval = ddi_copyin((void *)arg, dk_map32, size, flag);
3810 		if (rval != 0) {
3811 			return (EFAULT);
3812 		}
3813 		for (i = 0; i < NDKMAP; i++) {
3814 			dk_map[i].dkl_cylno = dk_map32[i].dkl_cylno;
3815 			dk_map[i].dkl_nblk  = dk_map32[i].dkl_nblk;
3816 		}
3817 		break;
3818 	}
3819 	case DDI_MODEL_NONE:
3820 		size = NDKMAP * sizeof (struct dk_map);
3821 		rval = ddi_copyin((void *)arg, dk_map, size, flag);
3822 		if (rval != 0) {
3823 			return (EFAULT);
3824 		}
3825 		break;
3826 	}
3827 #else /* ! _MULTI_DATAMODEL */
3828 	size = NDKMAP * sizeof (struct dk_map);
3829 	rval = ddi_copyin((void *)arg, dk_map, size, flag);
3830 	if (rval != 0) {
3831 		return (EFAULT);
3832 	}
3833 #endif /* _MULTI_DATAMODEL */
3834 
3835 	mutex_enter(CMLB_MUTEX(cl));
3836 	/* Note: The size used in this bcopy is set based upon the data model */
3837 	bcopy(dk_map, cl->cl_map, size);
3838 #if defined(_SUNOS_VTOC_16)
3839 	vp = (struct dkl_partition *)&(cl->cl_vtoc);
3840 #endif	/* defined(_SUNOS_VTOC_16) */
3841 	for (i = 0; i < NDKMAP; i++) {
3842 		lp  = &cl->cl_map[i];
3843 		cl->cl_offset[i] =
3844 		    cl->cl_g.dkg_nhead * cl->cl_g.dkg_nsect * lp->dkl_cylno;
3845 #if defined(_SUNOS_VTOC_16)
3846 		vp->p_start = cl->cl_offset[i];
3847 		vp->p_size = lp->dkl_nblk;
3848 		vp++;
3849 #endif	/* defined(_SUNOS_VTOC_16) */
3850 #if defined(__i386) || defined(__amd64)
3851 		cl->cl_offset[i] += cl->cl_solaris_offset;
3852 #endif
3853 	}
3854 	mutex_exit(CMLB_MUTEX(cl));
3855 	return (rval);
3856 }
3857 
3858 
3859 /*
3860  *    Function: cmlb_dkio_get_vtoc
3861  *
3862  * Description: This routine is the driver entry point for handling user
3863  *		requests to get the current volume table of contents
3864  *		(DKIOCGVTOC).
3865  *
3866  *   Arguments:
3867  *	arg		pointer to user provided vtoc structure specifying
3868  *			the current vtoc.
3869  *
3870  *	flag		this argument is a pass through to ddi_copyxxx()
3871  *			directly from the mode argument of ioctl().
3872  *
3873  *	tg_cookie	cookie from target driver to be passed back to target
3874  *			driver when we call back to it through tg_ops.
3875  *
3876  * Return Code: 0
3877  *		EFAULT
3878  *		ENXIO
3879  *		EIO
3880  */
3881 static int
3882 cmlb_dkio_get_vtoc(struct cmlb_lun *cl, caddr_t arg, int flag, void *tg_cookie)
3883 {
3884 #if defined(_SUNOS_VTOC_8)
3885 	struct vtoc	user_vtoc;
3886 #endif	/* defined(_SUNOS_VTOC_8) */
3887 	int		rval = 0;
3888 
3889 	mutex_enter(CMLB_MUTEX(cl));
3890 	if (cl->cl_blockcount > CMLB_OLDVTOC_LIMIT) {
3891 		mutex_exit(CMLB_MUTEX(cl));
3892 		return (EOVERFLOW);
3893 	}
3894 
3895 	rval = cmlb_validate_geometry(cl, B_TRUE, 0, tg_cookie);
3896 
3897 #if defined(_SUNOS_VTOC_8)
3898 	if (rval == EINVAL &&
3899 	    (cl->cl_alter_behavior & CMLB_FAKE_GEOM_LABEL_IOCTLS_VTOC8)) {
3900 		/*
3901 		 * This is to return a default label even when we do not
3902 		 * really assume a default label for the device.
3903 		 * dad driver utilizes this.
3904 		 */
3905 		if (cl->cl_blockcount <= CMLB_OLDVTOC_LIMIT) {
3906 			cmlb_setup_default_geometry(cl, tg_cookie);
3907 			rval = 0;
3908 		}
3909 	}
3910 #endif
3911 	if (rval) {
3912 		mutex_exit(CMLB_MUTEX(cl));
3913 		return (rval);
3914 	}
3915 
3916 #if defined(_SUNOS_VTOC_8)
3917 	cmlb_build_user_vtoc(cl, &user_vtoc);
3918 	mutex_exit(CMLB_MUTEX(cl));
3919 
3920 #ifdef _MULTI_DATAMODEL
3921 	switch (ddi_model_convert_from(flag & FMODELS)) {
3922 	case DDI_MODEL_ILP32: {
3923 		struct vtoc32 user_vtoc32;
3924 
3925 		vtoctovtoc32(user_vtoc, user_vtoc32);
3926 		if (ddi_copyout(&user_vtoc32, (void *)arg,
3927 		    sizeof (struct vtoc32), flag)) {
3928 			return (EFAULT);
3929 		}
3930 		break;
3931 	}
3932 
3933 	case DDI_MODEL_NONE:
3934 		if (ddi_copyout(&user_vtoc, (void *)arg,
3935 		    sizeof (struct vtoc), flag)) {
3936 			return (EFAULT);
3937 		}
3938 		break;
3939 	}
3940 #else /* ! _MULTI_DATAMODEL */
3941 	if (ddi_copyout(&user_vtoc, (void *)arg, sizeof (struct vtoc), flag)) {
3942 		return (EFAULT);
3943 	}
3944 #endif /* _MULTI_DATAMODEL */
3945 
3946 #elif defined(_SUNOS_VTOC_16)
3947 	mutex_exit(CMLB_MUTEX(cl));
3948 
3949 #ifdef _MULTI_DATAMODEL
3950 	/*
3951 	 * The cl_vtoc structure is a "struct dk_vtoc"  which is always
3952 	 * 32-bit to maintain compatibility with existing on-disk
3953 	 * structures.  Thus, we need to convert the structure when copying
3954 	 * it out to a datamodel-dependent "struct vtoc" in a 64-bit
3955 	 * program.  If the target is a 32-bit program, then no conversion
3956 	 * is necessary.
3957 	 */
3958 	/* LINTED: logical expression always true: op "||" */
3959 	ASSERT(sizeof (cl->cl_vtoc) == sizeof (struct vtoc32));
3960 	switch (ddi_model_convert_from(flag & FMODELS)) {
3961 	case DDI_MODEL_ILP32:
3962 		if (ddi_copyout(&(cl->cl_vtoc), (void *)arg,
3963 		    sizeof (cl->cl_vtoc), flag)) {
3964 			return (EFAULT);
3965 		}
3966 		break;
3967 
3968 	case DDI_MODEL_NONE: {
3969 		struct vtoc user_vtoc;
3970 
3971 		vtoc32tovtoc(cl->cl_vtoc, user_vtoc);
3972 		if (ddi_copyout(&user_vtoc, (void *)arg,
3973 		    sizeof (struct vtoc), flag)) {
3974 			return (EFAULT);
3975 		}
3976 		break;
3977 	}
3978 	}
3979 #else /* ! _MULTI_DATAMODEL */
3980 	if (ddi_copyout(&(cl->cl_vtoc), (void *)arg, sizeof (cl->cl_vtoc),
3981 	    flag)) {
3982 		return (EFAULT);
3983 	}
3984 #endif /* _MULTI_DATAMODEL */
3985 #else
3986 #error "No VTOC format defined."
3987 #endif
3988 
3989 	return (rval);
3990 }
3991 
3992 
3993 /*
3994  *    Function: cmlb_dkio_get_extvtoc
3995  */
3996 static int
3997 cmlb_dkio_get_extvtoc(struct cmlb_lun *cl, caddr_t arg, int flag,
3998     void *tg_cookie)
3999 {
4000 	struct extvtoc	ext_vtoc;
4001 #if defined(_SUNOS_VTOC_8)
4002 	struct vtoc	user_vtoc;
4003 #endif	/* defined(_SUNOS_VTOC_8) */
4004 	int		rval = 0;
4005 
4006 	bzero(&ext_vtoc, sizeof (struct extvtoc));
4007 	mutex_enter(CMLB_MUTEX(cl));
4008 	rval = cmlb_validate_geometry(cl, B_TRUE, 0, tg_cookie);
4009 
4010 #if defined(_SUNOS_VTOC_8)
4011 	if (rval == EINVAL &&
4012 	    (cl->cl_alter_behavior & CMLB_FAKE_GEOM_LABEL_IOCTLS_VTOC8)) {
4013 		/*
4014 		 * This is to return a default label even when we do not
4015 		 * really assume a default label for the device.
4016 		 * dad driver utilizes this.
4017 		 */
4018 		if (cl->cl_blockcount <= CMLB_OLDVTOC_LIMIT) {
4019 			cmlb_setup_default_geometry(cl, tg_cookie);
4020 			rval = 0;
4021 		}
4022 	}
4023 #endif
4024 	if (rval) {
4025 		mutex_exit(CMLB_MUTEX(cl));
4026 		return (rval);
4027 	}
4028 
4029 #if defined(_SUNOS_VTOC_8)
4030 	cmlb_build_user_vtoc(cl, &user_vtoc);
4031 	mutex_exit(CMLB_MUTEX(cl));
4032 
4033 	/*
4034 	 * Checking callers data model does not make much sense here
4035 	 * since extvtoc will always be equivalent to 64bit vtoc.
4036 	 * What is important is whether the kernel is in 32 or 64 bit
4037 	 */
4038 
4039 #ifdef _LP64
4040 		if (ddi_copyout(&user_vtoc, (void *)arg,
4041 		    sizeof (struct extvtoc), flag)) {
4042 			return (EFAULT);
4043 		}
4044 #else
4045 		vtoc32tovtoc(user_vtoc, ext_vtoc);
4046 		if (ddi_copyout(&ext_vtoc, (void *)arg,
4047 		    sizeof (struct extvtoc), flag)) {
4048 			return (EFAULT);
4049 		}
4050 #endif
4051 
4052 #elif defined(_SUNOS_VTOC_16)
4053 	/*
4054 	 * The cl_vtoc structure is a "struct dk_vtoc"  which is always
4055 	 * 32-bit to maintain compatibility with existing on-disk
4056 	 * structures.  Thus, we need to convert the structure when copying
4057 	 * it out to extvtoc
4058 	 */
4059 	vtoc32tovtoc(cl->cl_vtoc, ext_vtoc);
4060 	mutex_exit(CMLB_MUTEX(cl));
4061 
4062 	if (ddi_copyout(&ext_vtoc, (void *)arg, sizeof (struct extvtoc), flag))
4063 		return (EFAULT);
4064 #else
4065 #error "No VTOC format defined."
4066 #endif
4067 
4068 	return (rval);
4069 }
4070 static int
4071 cmlb_dkio_get_efi(struct cmlb_lun *cl, caddr_t arg, int flag, void *tg_cookie)
4072 {
4073 	dk_efi_t	user_efi;
4074 	int		rval = 0;
4075 	void		*buffer;
4076 	diskaddr_t	tgt_lba;
4077 
4078 	if (ddi_copyin(arg, &user_efi, sizeof (dk_efi_t), flag))
4079 		return (EFAULT);
4080 
4081 	user_efi.dki_data = (void *)(uintptr_t)user_efi.dki_data_64;
4082 
4083 	tgt_lba = user_efi.dki_lba;
4084 
4085 	mutex_enter(CMLB_MUTEX(cl));
4086 	if ((cmlb_check_update_blockcount(cl, tg_cookie) != 0) ||
4087 	    (cl->cl_tgt_blocksize == 0)) {
4088 		mutex_exit(CMLB_MUTEX(cl));
4089 		return (EINVAL);
4090 	}
4091 	if (cl->cl_tgt_blocksize != cl->cl_sys_blocksize)
4092 		tgt_lba = tgt_lba * cl->cl_tgt_blocksize /
4093 		    cl->cl_sys_blocksize;
4094 	mutex_exit(CMLB_MUTEX(cl));
4095 
4096 	buffer = kmem_alloc(user_efi.dki_length, KM_SLEEP);
4097 	rval = DK_TG_READ(cl, buffer, tgt_lba, user_efi.dki_length, tg_cookie);
4098 	if (rval == 0 && ddi_copyout(buffer, user_efi.dki_data,
4099 	    user_efi.dki_length, flag) != 0)
4100 		rval = EFAULT;
4101 
4102 	kmem_free(buffer, user_efi.dki_length);
4103 	return (rval);
4104 }
4105 
4106 #if defined(_SUNOS_VTOC_8)
4107 /*
4108  *    Function: cmlb_build_user_vtoc
4109  *
4110  * Description: This routine populates a pass by reference variable with the
4111  *		current volume table of contents.
4112  *
4113  *   Arguments: cl - driver soft state (unit) structure
4114  *		user_vtoc - pointer to vtoc structure to be populated
4115  */
4116 static void
4117 cmlb_build_user_vtoc(struct cmlb_lun *cl, struct vtoc *user_vtoc)
4118 {
4119 	struct dk_map2		*lpart;
4120 	struct dk_map		*lmap;
4121 	struct partition	*vpart;
4122 	uint32_t		nblks;
4123 	int			i;
4124 
4125 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
4126 
4127 	/*
4128 	 * Return vtoc structure fields in the provided VTOC area, addressed
4129 	 * by *vtoc.
4130 	 */
4131 	bzero(user_vtoc, sizeof (struct vtoc));
4132 	user_vtoc->v_bootinfo[0] = cl->cl_vtoc.v_bootinfo[0];
4133 	user_vtoc->v_bootinfo[1] = cl->cl_vtoc.v_bootinfo[1];
4134 	user_vtoc->v_bootinfo[2] = cl->cl_vtoc.v_bootinfo[2];
4135 	user_vtoc->v_sanity	= VTOC_SANE;
4136 	user_vtoc->v_version	= cl->cl_vtoc.v_version;
4137 	bcopy(cl->cl_vtoc.v_volume, user_vtoc->v_volume, LEN_DKL_VVOL);
4138 	user_vtoc->v_sectorsz = cl->cl_sys_blocksize;
4139 	user_vtoc->v_nparts = cl->cl_vtoc.v_nparts;
4140 
4141 	for (i = 0; i < 10; i++)
4142 		user_vtoc->v_reserved[i] = cl->cl_vtoc.v_reserved[i];
4143 
4144 	/*
4145 	 * Convert partitioning information.
4146 	 *
4147 	 * Note the conversion from starting cylinder number
4148 	 * to starting sector number.
4149 	 */
4150 	lmap = cl->cl_map;
4151 	lpart = (struct dk_map2 *)cl->cl_vtoc.v_part;
4152 	vpart = user_vtoc->v_part;
4153 
4154 	nblks = cl->cl_g.dkg_nsect * cl->cl_g.dkg_nhead;
4155 
4156 	for (i = 0; i < V_NUMPAR; i++) {
4157 		vpart->p_tag	= lpart->p_tag;
4158 		vpart->p_flag	= lpart->p_flag;
4159 		vpart->p_start	= lmap->dkl_cylno * nblks;
4160 		vpart->p_size	= lmap->dkl_nblk;
4161 		lmap++;
4162 		lpart++;
4163 		vpart++;
4164 
4165 		/* (4364927) */
4166 		user_vtoc->timestamp[i] = (time_t)cl->cl_vtoc.v_timestamp[i];
4167 	}
4168 
4169 	bcopy(cl->cl_asciilabel, user_vtoc->v_asciilabel, LEN_DKL_ASCII);
4170 }
4171 #endif
4172 
4173 static int
4174 cmlb_dkio_partition(struct cmlb_lun *cl, caddr_t arg, int flag,
4175     void *tg_cookie)
4176 {
4177 	struct partition64	p64;
4178 	int			rval = 0;
4179 	uint_t			nparts;
4180 	efi_gpe_t		*partitions;
4181 	efi_gpt_t		*buffer;
4182 	diskaddr_t		gpe_lba;
4183 
4184 	if (ddi_copyin((const void *)arg, &p64,
4185 	    sizeof (struct partition64), flag)) {
4186 		return (EFAULT);
4187 	}
4188 
4189 	buffer = kmem_alloc(EFI_MIN_ARRAY_SIZE, KM_SLEEP);
4190 	rval = DK_TG_READ(cl, buffer, 1, cl->cl_sys_blocksize, tg_cookie);
4191 	if (rval != 0)
4192 		goto done_error;
4193 
4194 	cmlb_swap_efi_gpt(buffer);
4195 
4196 	if ((rval = cmlb_validate_efi(buffer)) != 0)
4197 		goto done_error;
4198 
4199 	nparts = buffer->efi_gpt_NumberOfPartitionEntries;
4200 	gpe_lba = buffer->efi_gpt_PartitionEntryLBA;
4201 	if (p64.p_partno > nparts) {
4202 		/* couldn't find it */
4203 		rval = ESRCH;
4204 		goto done_error;
4205 	}
4206 	/*
4207 	 * if we're dealing with a partition that's out of the normal
4208 	 * 16K block, adjust accordingly
4209 	 */
4210 	gpe_lba += p64.p_partno / sizeof (efi_gpe_t);
4211 	rval = DK_TG_READ(cl, buffer, gpe_lba, EFI_MIN_ARRAY_SIZE, tg_cookie);
4212 
4213 	if (rval) {
4214 		goto done_error;
4215 	}
4216 	partitions = (efi_gpe_t *)buffer;
4217 
4218 	cmlb_swap_efi_gpe(nparts, partitions);
4219 
4220 	partitions += p64.p_partno;
4221 	bcopy(&partitions->efi_gpe_PartitionTypeGUID, &p64.p_type,
4222 	    sizeof (struct uuid));
4223 	p64.p_start = partitions->efi_gpe_StartingLBA;
4224 	p64.p_size = partitions->efi_gpe_EndingLBA -
4225 	    p64.p_start + 1;
4226 
4227 	if (ddi_copyout(&p64, (void *)arg, sizeof (struct partition64), flag))
4228 		rval = EFAULT;
4229 
4230 done_error:
4231 	kmem_free(buffer, EFI_MIN_ARRAY_SIZE);
4232 	return (rval);
4233 }
4234 
4235 
4236 /*
4237  *    Function: cmlb_dkio_set_vtoc
4238  *
4239  * Description: This routine is the driver entry point for handling user
4240  *		requests to set the current volume table of contents
4241  *		(DKIOCSVTOC).
4242  *
4243  *   Arguments:
4244  *	dev		the device number
4245  *	arg		pointer to user provided vtoc structure used to set the
4246  *			current vtoc.
4247  *
4248  *	flag		this argument is a pass through to ddi_copyxxx()
4249  *			directly from the mode argument of ioctl().
4250  *
4251  *	tg_cookie	cookie from target driver to be passed back to target
4252  *			driver when we call back to it through tg_ops.
4253  *
4254  * Return Code: 0
4255  *		EFAULT
4256  *		ENXIO
4257  *		EINVAL
4258  *		ENOTSUP
4259  */
4260 static int
4261 cmlb_dkio_set_vtoc(struct cmlb_lun *cl, dev_t dev, caddr_t arg, int flag,
4262     void *tg_cookie)
4263 {
4264 	struct vtoc	user_vtoc;
4265 	int		rval = 0;
4266 	boolean_t	internal;
4267 
4268 	internal = VOID2BOOLEAN(
4269 	    (cl->cl_alter_behavior & (CMLB_INTERNAL_MINOR_NODES)) != 0);
4270 
4271 #ifdef _MULTI_DATAMODEL
4272 	switch (ddi_model_convert_from(flag & FMODELS)) {
4273 	case DDI_MODEL_ILP32: {
4274 		struct vtoc32 user_vtoc32;
4275 
4276 		if (ddi_copyin((const void *)arg, &user_vtoc32,
4277 		    sizeof (struct vtoc32), flag)) {
4278 			return (EFAULT);
4279 		}
4280 		vtoc32tovtoc(user_vtoc32, user_vtoc);
4281 		break;
4282 	}
4283 
4284 	case DDI_MODEL_NONE:
4285 		if (ddi_copyin((const void *)arg, &user_vtoc,
4286 		    sizeof (struct vtoc), flag)) {
4287 			return (EFAULT);
4288 		}
4289 		break;
4290 	}
4291 #else /* ! _MULTI_DATAMODEL */
4292 	if (ddi_copyin((const void *)arg, &user_vtoc,
4293 	    sizeof (struct vtoc), flag)) {
4294 		return (EFAULT);
4295 	}
4296 #endif /* _MULTI_DATAMODEL */
4297 
4298 	mutex_enter(CMLB_MUTEX(cl));
4299 
4300 	if (cl->cl_blockcount > CMLB_OLDVTOC_LIMIT) {
4301 		mutex_exit(CMLB_MUTEX(cl));
4302 		return (EOVERFLOW);
4303 	}
4304 
4305 #if defined(__i386) || defined(__amd64)
4306 	if (cl->cl_tgt_blocksize != cl->cl_sys_blocksize) {
4307 		mutex_exit(CMLB_MUTEX(cl));
4308 		return (EINVAL);
4309 	}
4310 #endif
4311 
4312 	if (cl->cl_g.dkg_ncyl == 0) {
4313 		mutex_exit(CMLB_MUTEX(cl));
4314 		return (EINVAL);
4315 	}
4316 
4317 	mutex_exit(CMLB_MUTEX(cl));
4318 	cmlb_clear_efi(cl, tg_cookie);
4319 	ddi_remove_minor_node(CMLB_DEVINFO(cl), "wd");
4320 	ddi_remove_minor_node(CMLB_DEVINFO(cl), "wd,raw");
4321 
4322 	/*
4323 	 * cmlb_dkio_set_vtoc creates duplicate minor nodes when
4324 	 * relabeling an SMI disk. To avoid that we remove them
4325 	 * before creating.
4326 	 * It should be OK to remove a non-existed minor node.
4327 	 */
4328 	ddi_remove_minor_node(CMLB_DEVINFO(cl), "h");
4329 	ddi_remove_minor_node(CMLB_DEVINFO(cl), "h,raw");
4330 
4331 	(void) cmlb_create_minor(CMLB_DEVINFO(cl), "h",
4332 	    S_IFBLK, (CMLBUNIT(dev) << CMLBUNIT_SHIFT) | WD_NODE,
4333 	    cl->cl_node_type, NULL, internal);
4334 	(void) cmlb_create_minor(CMLB_DEVINFO(cl), "h,raw",
4335 	    S_IFCHR, (CMLBUNIT(dev) << CMLBUNIT_SHIFT) | WD_NODE,
4336 	    cl->cl_node_type, NULL, internal);
4337 	mutex_enter(CMLB_MUTEX(cl));
4338 
4339 	if ((rval = cmlb_build_label_vtoc(cl, &user_vtoc)) == 0) {
4340 		if ((rval = cmlb_write_label(cl, tg_cookie)) == 0) {
4341 			if (cmlb_validate_geometry(cl,
4342 			    B_TRUE, 0, tg_cookie) != 0) {
4343 				cmlb_dbg(CMLB_ERROR, cl,
4344 				    "cmlb_dkio_set_vtoc: "
4345 				    "Failed validate geometry\n");
4346 			}
4347 			cl->cl_msglog_flag |= CMLB_ALLOW_2TB_WARN;
4348 		}
4349 	}
4350 	mutex_exit(CMLB_MUTEX(cl));
4351 	return (rval);
4352 }
4353 
4354 /*
4355  *    Function: cmlb_dkio_set_extvtoc
4356  */
4357 static int
4358 cmlb_dkio_set_extvtoc(struct cmlb_lun *cl, dev_t dev, caddr_t arg, int flag,
4359     void *tg_cookie)
4360 {
4361 	int		rval = 0;
4362 	struct vtoc	user_vtoc;
4363 	boolean_t	internal;
4364 
4365 
4366 	/*
4367 	 * Checking callers data model does not make much sense here
4368 	 * since extvtoc will always be equivalent to 64bit vtoc.
4369 	 * What is important is whether the kernel is in 32 or 64 bit
4370 	 */
4371 
4372 #ifdef _LP64
4373 	if (ddi_copyin((const void *)arg, &user_vtoc,
4374 		    sizeof (struct extvtoc), flag)) {
4375 			return (EFAULT);
4376 	}
4377 #else
4378 	struct	extvtoc	user_extvtoc;
4379 	if (ddi_copyin((const void *)arg, &user_extvtoc,
4380 		    sizeof (struct extvtoc), flag)) {
4381 			return (EFAULT);
4382 	}
4383 
4384 	vtoctovtoc32(user_extvtoc, user_vtoc);
4385 #endif
4386 
4387 	internal = VOID2BOOLEAN(
4388 	    (cl->cl_alter_behavior & (CMLB_INTERNAL_MINOR_NODES)) != 0);
4389 	mutex_enter(CMLB_MUTEX(cl));
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 	 * cmlb_dkio_set_extvtoc creates duplicate minor nodes when
4408 	 * relabeling an SMI disk. To avoid that we remove them
4409 	 * before creating.
4410 	 * It should be OK to remove a non-existed minor node.
4411 	 */
4412 	ddi_remove_minor_node(CMLB_DEVINFO(cl), "h");
4413 	ddi_remove_minor_node(CMLB_DEVINFO(cl), "h,raw");
4414 
4415 	(void) cmlb_create_minor(CMLB_DEVINFO(cl), "h",
4416 	    S_IFBLK, (CMLBUNIT(dev) << CMLBUNIT_SHIFT) | WD_NODE,
4417 	    cl->cl_node_type, NULL, internal);
4418 	(void) cmlb_create_minor(CMLB_DEVINFO(cl), "h,raw",
4419 	    S_IFCHR, (CMLBUNIT(dev) << CMLBUNIT_SHIFT) | WD_NODE,
4420 	    cl->cl_node_type, NULL, internal);
4421 
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 		}
4433 	}
4434 	mutex_exit(CMLB_MUTEX(cl));
4435 	return (rval);
4436 }
4437 
4438 /*
4439  *    Function: cmlb_build_label_vtoc
4440  *
4441  * Description: This routine updates the driver soft state current volume table
4442  *		of contents based on a user specified vtoc.
4443  *
4444  *   Arguments: cl - driver soft state (unit) structure
4445  *		user_vtoc - pointer to vtoc structure specifying vtoc to be used
4446  *			    to update the driver soft state.
4447  *
4448  * Return Code: 0
4449  *		EINVAL
4450  */
4451 static int
4452 cmlb_build_label_vtoc(struct cmlb_lun *cl, struct vtoc *user_vtoc)
4453 {
4454 	struct dk_map		*lmap;
4455 	struct partition	*vpart;
4456 	uint_t			nblks;
4457 #if defined(_SUNOS_VTOC_8)
4458 	int			ncyl;
4459 	struct dk_map2		*lpart;
4460 #endif	/* defined(_SUNOS_VTOC_8) */
4461 	int			i;
4462 
4463 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
4464 
4465 	/* Sanity-check the vtoc */
4466 	if (user_vtoc->v_sanity != VTOC_SANE ||
4467 	    user_vtoc->v_sectorsz != cl->cl_sys_blocksize ||
4468 	    user_vtoc->v_nparts != V_NUMPAR) {
4469 		cmlb_dbg(CMLB_INFO,  cl,
4470 		    "cmlb_build_label_vtoc: vtoc not valid\n");
4471 		return (EINVAL);
4472 	}
4473 
4474 	nblks = cl->cl_g.dkg_nsect * cl->cl_g.dkg_nhead;
4475 	if (nblks == 0) {
4476 		cmlb_dbg(CMLB_INFO,  cl,
4477 		    "cmlb_build_label_vtoc: geom nblks is 0\n");
4478 		return (EINVAL);
4479 	}
4480 
4481 #if defined(_SUNOS_VTOC_8)
4482 	vpart = user_vtoc->v_part;
4483 	for (i = 0; i < V_NUMPAR; i++) {
4484 		if (((unsigned)vpart->p_start % nblks) != 0) {
4485 			cmlb_dbg(CMLB_INFO,  cl,
4486 			    "cmlb_build_label_vtoc: p_start not multiply of"
4487 			    "nblks part %d p_start %d nblks %d\n", i,
4488 			    vpart->p_start, nblks);
4489 			return (EINVAL);
4490 		}
4491 		ncyl = (unsigned)vpart->p_start / nblks;
4492 		ncyl += (unsigned)vpart->p_size / nblks;
4493 		if (((unsigned)vpart->p_size % nblks) != 0) {
4494 			ncyl++;
4495 		}
4496 		if (ncyl > (int)cl->cl_g.dkg_ncyl) {
4497 			cmlb_dbg(CMLB_INFO,  cl,
4498 			    "cmlb_build_label_vtoc: ncyl %d  > dkg_ncyl %d"
4499 			    "p_size %ld p_start %ld nblks %d  part number %d"
4500 			    "tag %d\n",
4501 			    ncyl, cl->cl_g.dkg_ncyl, vpart->p_size,
4502 			    vpart->p_start, nblks,
4503 			    i, vpart->p_tag);
4504 
4505 			return (EINVAL);
4506 		}
4507 		vpart++;
4508 	}
4509 #endif	/* defined(_SUNOS_VTOC_8) */
4510 
4511 	/* Put appropriate vtoc structure fields into the disk label */
4512 #if defined(_SUNOS_VTOC_16)
4513 	/*
4514 	 * The vtoc is always a 32bit data structure to maintain the
4515 	 * on-disk format. Convert "in place" instead of doing bcopy.
4516 	 */
4517 	vtoctovtoc32((*user_vtoc), (*((struct vtoc32 *)&(cl->cl_vtoc))));
4518 
4519 	/*
4520 	 * in the 16-slice vtoc, starting sectors are expressed in
4521 	 * numbers *relative* to the start of the Solaris fdisk partition.
4522 	 */
4523 	lmap = cl->cl_map;
4524 	vpart = user_vtoc->v_part;
4525 
4526 	for (i = 0; i < (int)user_vtoc->v_nparts; i++, lmap++, vpart++) {
4527 		lmap->dkl_cylno = (unsigned)vpart->p_start / nblks;
4528 		lmap->dkl_nblk = (unsigned)vpart->p_size;
4529 	}
4530 
4531 #elif defined(_SUNOS_VTOC_8)
4532 
4533 	cl->cl_vtoc.v_bootinfo[0] = (uint32_t)user_vtoc->v_bootinfo[0];
4534 	cl->cl_vtoc.v_bootinfo[1] = (uint32_t)user_vtoc->v_bootinfo[1];
4535 	cl->cl_vtoc.v_bootinfo[2] = (uint32_t)user_vtoc->v_bootinfo[2];
4536 
4537 	cl->cl_vtoc.v_sanity = (uint32_t)user_vtoc->v_sanity;
4538 	cl->cl_vtoc.v_version = (uint32_t)user_vtoc->v_version;
4539 
4540 	bcopy(user_vtoc->v_volume, cl->cl_vtoc.v_volume, LEN_DKL_VVOL);
4541 
4542 	cl->cl_vtoc.v_nparts = user_vtoc->v_nparts;
4543 
4544 	for (i = 0; i < 10; i++)
4545 		cl->cl_vtoc.v_reserved[i] =  user_vtoc->v_reserved[i];
4546 
4547 	/*
4548 	 * Note the conversion from starting sector number
4549 	 * to starting cylinder number.
4550 	 * Return error if division results in a remainder.
4551 	 */
4552 	lmap = cl->cl_map;
4553 	lpart = cl->cl_vtoc.v_part;
4554 	vpart = user_vtoc->v_part;
4555 
4556 	for (i = 0; i < (int)user_vtoc->v_nparts; i++) {
4557 		lpart->p_tag  = vpart->p_tag;
4558 		lpart->p_flag = vpart->p_flag;
4559 		lmap->dkl_cylno = (unsigned)vpart->p_start / nblks;
4560 		lmap->dkl_nblk = (unsigned)vpart->p_size;
4561 
4562 		lmap++;
4563 		lpart++;
4564 		vpart++;
4565 
4566 		/* (4387723) */
4567 #ifdef _LP64
4568 		if (user_vtoc->timestamp[i] > TIME32_MAX) {
4569 			cl->cl_vtoc.v_timestamp[i] = TIME32_MAX;
4570 		} else {
4571 			cl->cl_vtoc.v_timestamp[i] = user_vtoc->timestamp[i];
4572 		}
4573 #else
4574 		cl->cl_vtoc.v_timestamp[i] = user_vtoc->timestamp[i];
4575 #endif
4576 	}
4577 
4578 	bcopy(user_vtoc->v_asciilabel, cl->cl_asciilabel, LEN_DKL_ASCII);
4579 #else
4580 #error "No VTOC format defined."
4581 #endif
4582 	return (0);
4583 }
4584 
4585 /*
4586  *    Function: cmlb_clear_efi
4587  *
4588  * Description: This routine clears all EFI labels.
4589  *
4590  *   Arguments:
4591  *	cl		 driver soft state (unit) structure
4592  *
4593  *	tg_cookie	cookie from target driver to be passed back to target
4594  *			driver when we call back to it through tg_ops.
4595  * Return Code: void
4596  */
4597 static void
4598 cmlb_clear_efi(struct cmlb_lun *cl, void *tg_cookie)
4599 {
4600 	efi_gpt_t	*gpt;
4601 	diskaddr_t	cap;
4602 	int		rval;
4603 
4604 	ASSERT(!mutex_owned(CMLB_MUTEX(cl)));
4605 
4606 	mutex_enter(CMLB_MUTEX(cl));
4607 	cl->cl_reserved = -1;
4608 	mutex_exit(CMLB_MUTEX(cl));
4609 
4610 	gpt = kmem_alloc(cl->cl_sys_blocksize, KM_SLEEP);
4611 
4612 	if (DK_TG_READ(cl, gpt, 1, cl->cl_sys_blocksize, tg_cookie) != 0) {
4613 		goto done;
4614 	}
4615 
4616 	cmlb_swap_efi_gpt(gpt);
4617 	rval = cmlb_validate_efi(gpt);
4618 	if (rval == 0) {
4619 		/* clear primary */
4620 		bzero(gpt, sizeof (efi_gpt_t));
4621 		if (rval = DK_TG_WRITE(cl, gpt, 1, cl->cl_sys_blocksize,
4622 		    tg_cookie)) {
4623 			cmlb_dbg(CMLB_INFO,  cl,
4624 			    "cmlb_clear_efi: clear primary label failed\n");
4625 		}
4626 	}
4627 	/* the backup */
4628 	rval = DK_TG_GETCAP(cl, &cap, tg_cookie);
4629 	if (rval) {
4630 		goto done;
4631 	}
4632 
4633 	if ((rval = DK_TG_READ(cl, gpt, cap - 1, cl->cl_sys_blocksize,
4634 	    tg_cookie)) != 0) {
4635 		goto done;
4636 	}
4637 	cmlb_swap_efi_gpt(gpt);
4638 	rval = cmlb_validate_efi(gpt);
4639 	if (rval == 0) {
4640 		/* clear backup */
4641 		cmlb_dbg(CMLB_TRACE,  cl,
4642 		    "cmlb_clear_efi clear backup@%lu\n", cap - 1);
4643 		bzero(gpt, sizeof (efi_gpt_t));
4644 		if ((rval = DK_TG_WRITE(cl,  gpt, cap - 1, cl->cl_sys_blocksize,
4645 		    tg_cookie))) {
4646 			cmlb_dbg(CMLB_INFO,  cl,
4647 			    "cmlb_clear_efi: clear backup label failed\n");
4648 		}
4649 	} else {
4650 		/*
4651 		 * Refer to comments related to off-by-1 at the
4652 		 * header of this file
4653 		 */
4654 		if ((rval = DK_TG_READ(cl, gpt, cap - 2,
4655 		    cl->cl_sys_blocksize, tg_cookie)) != 0) {
4656 			goto done;
4657 		}
4658 		cmlb_swap_efi_gpt(gpt);
4659 		rval = cmlb_validate_efi(gpt);
4660 		if (rval == 0) {
4661 			/* clear legacy backup EFI label */
4662 			cmlb_dbg(CMLB_TRACE,  cl,
4663 			    "cmlb_clear_efi clear legacy backup@%lu\n",
4664 			    cap - 2);
4665 			bzero(gpt, sizeof (efi_gpt_t));
4666 			if ((rval = DK_TG_WRITE(cl,  gpt, cap - 2,
4667 			    cl->cl_sys_blocksize, tg_cookie))) {
4668 				cmlb_dbg(CMLB_INFO,  cl,
4669 				"cmlb_clear_efi: clear legacy backup label "
4670 				"failed\n");
4671 			}
4672 		}
4673 	}
4674 
4675 done:
4676 	kmem_free(gpt, cl->cl_sys_blocksize);
4677 }
4678 
4679 /*
4680  *    Function: cmlb_set_vtoc
4681  *
4682  * Description: This routine writes data to the appropriate positions
4683  *
4684  *   Arguments:
4685  *	cl		driver soft state (unit) structure
4686  *
4687  *	dkl		the data to be written
4688  *
4689  *	tg_cookie	cookie from target driver to be passed back to target
4690  *			driver when we call back to it through tg_ops.
4691  *
4692  * Return: void
4693  */
4694 static int
4695 cmlb_set_vtoc(struct cmlb_lun *cl, struct dk_label *dkl, void *tg_cookie)
4696 {
4697 	uint_t	label_addr;
4698 	int	sec;
4699 	diskaddr_t	blk;
4700 	int	head;
4701 	int	cyl;
4702 	int	rval;
4703 
4704 #if defined(__i386) || defined(__amd64)
4705 	label_addr = cl->cl_solaris_offset + DK_LABEL_LOC;
4706 #else
4707 	/* Write the primary label at block 0 of the solaris partition. */
4708 	label_addr = 0;
4709 #endif
4710 
4711 	rval = DK_TG_WRITE(cl, dkl, label_addr, cl->cl_sys_blocksize,
4712 	    tg_cookie);
4713 
4714 	if (rval != 0) {
4715 		return (rval);
4716 	}
4717 
4718 	/*
4719 	 * Calculate where the backup labels go.  They are always on
4720 	 * the last alternate cylinder, but some older drives put them
4721 	 * on head 2 instead of the last head.	They are always on the
4722 	 * first 5 odd sectors of the appropriate track.
4723 	 *
4724 	 * We have no choice at this point, but to believe that the
4725 	 * disk label is valid.	 Use the geometry of the disk
4726 	 * as described in the label.
4727 	 */
4728 	cyl  = dkl->dkl_ncyl  + dkl->dkl_acyl - 1;
4729 	head = dkl->dkl_nhead - 1;
4730 
4731 	/*
4732 	 * Write and verify the backup labels. Make sure we don't try to
4733 	 * write past the last cylinder.
4734 	 */
4735 	for (sec = 1; ((sec < 5 * 2 + 1) && (sec < dkl->dkl_nsect)); sec += 2) {
4736 		blk = (diskaddr_t)(
4737 		    (cyl * ((dkl->dkl_nhead * dkl->dkl_nsect) - dkl->dkl_apc)) +
4738 		    (head * dkl->dkl_nsect) + sec);
4739 #if defined(__i386) || defined(__amd64)
4740 		blk += cl->cl_solaris_offset;
4741 #endif
4742 		rval = DK_TG_WRITE(cl, dkl, blk, cl->cl_sys_blocksize,
4743 		    tg_cookie);
4744 		cmlb_dbg(CMLB_INFO,  cl,
4745 		"cmlb_set_vtoc: wrote backup label %llx\n", blk);
4746 		if (rval != 0) {
4747 			goto exit;
4748 		}
4749 	}
4750 exit:
4751 	return (rval);
4752 }
4753 
4754 /*
4755  *    Function: cmlb_clear_vtoc
4756  *
4757  * Description: This routine clears out the VTOC labels.
4758  *
4759  *   Arguments:
4760  *	cl		driver soft state (unit) structure
4761  *
4762  *	tg_cookie	cookie from target driver to be passed back to target
4763  *			driver when we call back to it through tg_ops.
4764  *
4765  * Return: void
4766  */
4767 static void
4768 cmlb_clear_vtoc(struct cmlb_lun *cl, void *tg_cookie)
4769 {
4770 	struct dk_label		*dkl;
4771 
4772 	mutex_exit(CMLB_MUTEX(cl));
4773 	dkl = kmem_zalloc(cl->cl_sys_blocksize, KM_SLEEP);
4774 	mutex_enter(CMLB_MUTEX(cl));
4775 	/*
4776 	 * cmlb_set_vtoc uses these fields in order to figure out
4777 	 * where to overwrite the backup labels
4778 	 */
4779 	dkl->dkl_apc    = cl->cl_g.dkg_apc;
4780 	dkl->dkl_ncyl   = cl->cl_g.dkg_ncyl;
4781 	dkl->dkl_acyl   = cl->cl_g.dkg_acyl;
4782 	dkl->dkl_nhead  = cl->cl_g.dkg_nhead;
4783 	dkl->dkl_nsect  = cl->cl_g.dkg_nsect;
4784 	mutex_exit(CMLB_MUTEX(cl));
4785 	(void) cmlb_set_vtoc(cl, dkl, tg_cookie);
4786 	kmem_free(dkl, cl->cl_sys_blocksize);
4787 
4788 	mutex_enter(CMLB_MUTEX(cl));
4789 }
4790 
4791 /*
4792  *    Function: cmlb_write_label
4793  *
4794  * Description: This routine will validate and write the driver soft state vtoc
4795  *		contents to the device.
4796  *
4797  *   Arguments:
4798  *	cl		cmlb handle
4799  *
4800  *	tg_cookie	cookie from target driver to be passed back to target
4801  *			driver when we call back to it through tg_ops.
4802  *
4803  *
4804  * Return Code: the code returned by cmlb_send_scsi_cmd()
4805  *		0
4806  *		EINVAL
4807  *		ENXIO
4808  *		ENOMEM
4809  */
4810 static int
4811 cmlb_write_label(struct cmlb_lun *cl, void *tg_cookie)
4812 {
4813 	struct dk_label	*dkl;
4814 	short		sum;
4815 	short		*sp;
4816 	int		i;
4817 	int		rval;
4818 
4819 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
4820 	mutex_exit(CMLB_MUTEX(cl));
4821 	dkl = kmem_zalloc(cl->cl_sys_blocksize, KM_SLEEP);
4822 	mutex_enter(CMLB_MUTEX(cl));
4823 
4824 	bcopy(&cl->cl_vtoc, &dkl->dkl_vtoc, sizeof (struct dk_vtoc));
4825 	dkl->dkl_rpm	= cl->cl_g.dkg_rpm;
4826 	dkl->dkl_pcyl	= cl->cl_g.dkg_pcyl;
4827 	dkl->dkl_apc	= cl->cl_g.dkg_apc;
4828 	dkl->dkl_intrlv = cl->cl_g.dkg_intrlv;
4829 	dkl->dkl_ncyl	= cl->cl_g.dkg_ncyl;
4830 	dkl->dkl_acyl	= cl->cl_g.dkg_acyl;
4831 	dkl->dkl_nhead	= cl->cl_g.dkg_nhead;
4832 	dkl->dkl_nsect	= cl->cl_g.dkg_nsect;
4833 
4834 #if defined(_SUNOS_VTOC_8)
4835 	dkl->dkl_obs1	= cl->cl_g.dkg_obs1;
4836 	dkl->dkl_obs2	= cl->cl_g.dkg_obs2;
4837 	dkl->dkl_obs3	= cl->cl_g.dkg_obs3;
4838 	for (i = 0; i < NDKMAP; i++) {
4839 		dkl->dkl_map[i].dkl_cylno = cl->cl_map[i].dkl_cylno;
4840 		dkl->dkl_map[i].dkl_nblk  = cl->cl_map[i].dkl_nblk;
4841 	}
4842 	bcopy(cl->cl_asciilabel, dkl->dkl_asciilabel, LEN_DKL_ASCII);
4843 #elif defined(_SUNOS_VTOC_16)
4844 	dkl->dkl_skew	= cl->cl_dkg_skew;
4845 #else
4846 #error "No VTOC format defined."
4847 #endif
4848 
4849 	dkl->dkl_magic			= DKL_MAGIC;
4850 	dkl->dkl_write_reinstruct	= cl->cl_g.dkg_write_reinstruct;
4851 	dkl->dkl_read_reinstruct	= cl->cl_g.dkg_read_reinstruct;
4852 
4853 	/* Construct checksum for the new disk label */
4854 	sum = 0;
4855 	sp = (short *)dkl;
4856 	i = sizeof (struct dk_label) / sizeof (short);
4857 	while (i--) {
4858 		sum ^= *sp++;
4859 	}
4860 	dkl->dkl_cksum = sum;
4861 
4862 	mutex_exit(CMLB_MUTEX(cl));
4863 
4864 	rval = cmlb_set_vtoc(cl, dkl, tg_cookie);
4865 exit:
4866 	kmem_free(dkl, cl->cl_sys_blocksize);
4867 	mutex_enter(CMLB_MUTEX(cl));
4868 	return (rval);
4869 }
4870 
4871 static int
4872 cmlb_dkio_set_efi(struct cmlb_lun *cl, dev_t dev, caddr_t arg, int flag,
4873     void *tg_cookie)
4874 {
4875 	dk_efi_t	user_efi;
4876 	int		rval = 0;
4877 	void		*buffer;
4878 	diskaddr_t	tgt_lba;
4879 	boolean_t	internal;
4880 
4881 	if (ddi_copyin(arg, &user_efi, sizeof (dk_efi_t), flag))
4882 		return (EFAULT);
4883 
4884 	internal = VOID2BOOLEAN(
4885 	    (cl->cl_alter_behavior & (CMLB_INTERNAL_MINOR_NODES)) != 0);
4886 
4887 	user_efi.dki_data = (void *)(uintptr_t)user_efi.dki_data_64;
4888 
4889 	buffer = kmem_alloc(user_efi.dki_length, KM_SLEEP);
4890 	if (ddi_copyin(user_efi.dki_data, buffer, user_efi.dki_length, flag)) {
4891 		rval = EFAULT;
4892 	} else {
4893 		/*
4894 		 * let's clear the vtoc labels and clear the softstate
4895 		 * vtoc.
4896 		 */
4897 		mutex_enter(CMLB_MUTEX(cl));
4898 		if (cl->cl_vtoc.v_sanity == VTOC_SANE) {
4899 			cmlb_dbg(CMLB_TRACE,  cl,
4900 			    "cmlb_dkio_set_efi: CLEAR VTOC\n");
4901 			if (cl->cl_label_from_media == CMLB_LABEL_VTOC)
4902 				cmlb_clear_vtoc(cl, tg_cookie);
4903 			bzero(&cl->cl_vtoc, sizeof (struct dk_vtoc));
4904 			mutex_exit(CMLB_MUTEX(cl));
4905 			ddi_remove_minor_node(CMLB_DEVINFO(cl), "h");
4906 			ddi_remove_minor_node(CMLB_DEVINFO(cl), "h,raw");
4907 			(void) cmlb_create_minor(CMLB_DEVINFO(cl), "wd",
4908 			    S_IFBLK,
4909 			    (CMLBUNIT(dev) << CMLBUNIT_SHIFT) | WD_NODE,
4910 			    cl->cl_node_type, NULL, internal);
4911 			(void) cmlb_create_minor(CMLB_DEVINFO(cl), "wd,raw",
4912 			    S_IFCHR,
4913 			    (CMLBUNIT(dev) << CMLBUNIT_SHIFT) | WD_NODE,
4914 			    cl->cl_node_type, NULL, internal);
4915 		} else
4916 			mutex_exit(CMLB_MUTEX(cl));
4917 
4918 		tgt_lba = user_efi.dki_lba;
4919 
4920 		mutex_enter(CMLB_MUTEX(cl));
4921 		if ((cmlb_check_update_blockcount(cl, tg_cookie) != 0) ||
4922 		    (cl->cl_tgt_blocksize == 0)) {
4923 			kmem_free(buffer, user_efi.dki_length);
4924 			mutex_exit(CMLB_MUTEX(cl));
4925 			return (EINVAL);
4926 		}
4927 		if (cl->cl_tgt_blocksize != cl->cl_sys_blocksize)
4928 			tgt_lba = tgt_lba *
4929 			    cl->cl_tgt_blocksize / cl->cl_sys_blocksize;
4930 
4931 		mutex_exit(CMLB_MUTEX(cl));
4932 		rval = DK_TG_WRITE(cl, buffer, tgt_lba, user_efi.dki_length,
4933 		    tg_cookie);
4934 
4935 		if (rval == 0) {
4936 			mutex_enter(CMLB_MUTEX(cl));
4937 			cl->cl_f_geometry_is_valid = B_FALSE;
4938 			mutex_exit(CMLB_MUTEX(cl));
4939 		}
4940 	}
4941 	kmem_free(buffer, user_efi.dki_length);
4942 	return (rval);
4943 }
4944 
4945 /*
4946  *    Function: cmlb_dkio_get_mboot
4947  *
4948  * Description: This routine is the driver entry point for handling user
4949  *		requests to get the current device mboot (DKIOCGMBOOT)
4950  *
4951  *   Arguments:
4952  *	arg		pointer to user provided mboot structure specifying
4953  *			the current mboot.
4954  *
4955  *	flag		this argument is a pass through to ddi_copyxxx()
4956  *			directly from the mode argument of ioctl().
4957  *
4958  *	tg_cookie	cookie from target driver to be passed back to target
4959  *			driver when we call back to it through tg_ops.
4960  *
4961  * Return Code: 0
4962  *		EINVAL
4963  *		EFAULT
4964  *		ENXIO
4965  */
4966 static int
4967 cmlb_dkio_get_mboot(struct cmlb_lun *cl, caddr_t arg, int flag, void *tg_cookie)
4968 {
4969 	struct mboot	*mboot;
4970 	int		rval;
4971 	size_t		buffer_size;
4972 
4973 
4974 #if defined(_SUNOS_VTOC_8)
4975 	if ((!ISREMOVABLE(cl) && !ISHOTPLUGGABLE(cl)) || (arg == NULL)) {
4976 #elif defined(_SUNOS_VTOC_16)
4977 	if (arg == NULL) {
4978 #endif
4979 		return (EINVAL);
4980 	}
4981 
4982 	/*
4983 	 * Read the mboot block, located at absolute block 0 on the target.
4984 	 */
4985 	buffer_size = cl->cl_sys_blocksize;
4986 
4987 	cmlb_dbg(CMLB_TRACE,  cl,
4988 	    "cmlb_dkio_get_mboot: allocation size: 0x%x\n", buffer_size);
4989 
4990 	mboot = kmem_zalloc(buffer_size, KM_SLEEP);
4991 	if ((rval = DK_TG_READ(cl, mboot, 0, buffer_size, tg_cookie)) == 0) {
4992 		if (ddi_copyout(mboot, (void *)arg,
4993 		    sizeof (struct mboot), flag) != 0) {
4994 			rval = EFAULT;
4995 		}
4996 	}
4997 	kmem_free(mboot, buffer_size);
4998 	return (rval);
4999 }
5000 
5001 
5002 /*
5003  *    Function: cmlb_dkio_set_mboot
5004  *
5005  * Description: This routine is the driver entry point for handling user
5006  *		requests to validate and set the device master boot
5007  *		(DKIOCSMBOOT).
5008  *
5009  *   Arguments:
5010  *	arg		pointer to user provided mboot structure used to set the
5011  *			master boot.
5012  *
5013  *	flag		this argument is a pass through to ddi_copyxxx()
5014  *			directly from the mode argument of ioctl().
5015  *
5016  *	tg_cookie	cookie from target driver to be passed back to target
5017  *			driver when we call back to it through tg_ops.
5018  *
5019  * Return Code: 0
5020  *		EINVAL
5021  *		EFAULT
5022  *		ENXIO
5023  */
5024 static int
5025 cmlb_dkio_set_mboot(struct cmlb_lun *cl, caddr_t arg, int flag, void *tg_cookie)
5026 {
5027 	struct mboot	*mboot = NULL;
5028 	int		rval;
5029 	ushort_t	magic;
5030 
5031 
5032 	ASSERT(!mutex_owned(CMLB_MUTEX(cl)));
5033 
5034 #if defined(_SUNOS_VTOC_8)
5035 	if (!ISREMOVABLE(cl) && !ISHOTPLUGGABLE(cl)) {
5036 		return (EINVAL);
5037 	}
5038 #endif
5039 
5040 	if (arg == NULL) {
5041 		return (EINVAL);
5042 	}
5043 
5044 	mboot = kmem_zalloc(cl->cl_sys_blocksize, KM_SLEEP);
5045 
5046 	if (ddi_copyin((const void *)arg, mboot,
5047 	    cl->cl_sys_blocksize, flag) != 0) {
5048 		kmem_free(mboot, cl->cl_sys_blocksize);
5049 		return (EFAULT);
5050 	}
5051 
5052 	/* Is this really a master boot record? */
5053 	magic = LE_16(mboot->signature);
5054 	if (magic != MBB_MAGIC) {
5055 		kmem_free(mboot, cl->cl_sys_blocksize);
5056 		return (EINVAL);
5057 	}
5058 
5059 	rval = DK_TG_WRITE(cl, mboot, 0, cl->cl_sys_blocksize, tg_cookie);
5060 
5061 	mutex_enter(CMLB_MUTEX(cl));
5062 #if defined(__i386) || defined(__amd64)
5063 	if (rval == 0) {
5064 		/*
5065 		 * mboot has been written successfully.
5066 		 * update the fdisk and vtoc tables in memory
5067 		 */
5068 		rval = cmlb_update_fdisk_and_vtoc(cl, tg_cookie);
5069 		if ((!cl->cl_f_geometry_is_valid) || (rval != 0)) {
5070 			mutex_exit(CMLB_MUTEX(cl));
5071 			kmem_free(mboot, cl->cl_sys_blocksize);
5072 			return (rval);
5073 		}
5074 	}
5075 
5076 #ifdef __lock_lint
5077 	cmlb_setup_default_geometry(cl, tg_cookie);
5078 #endif
5079 
5080 #else
5081 	if (rval == 0) {
5082 		/*
5083 		 * mboot has been written successfully.
5084 		 * set up the default geometry and VTOC
5085 		 */
5086 		if (cl->cl_blockcount <= CMLB_EXTVTOC_LIMIT)
5087 			cmlb_setup_default_geometry(cl, tg_cookie);
5088 	}
5089 #endif
5090 	cl->cl_msglog_flag |= CMLB_ALLOW_2TB_WARN;
5091 	mutex_exit(CMLB_MUTEX(cl));
5092 	kmem_free(mboot, cl->cl_sys_blocksize);
5093 	return (rval);
5094 }
5095 
5096 
5097 #if defined(__i386) || defined(__amd64)
5098 /*ARGSUSED*/
5099 static int
5100 cmlb_dkio_set_ext_part(struct cmlb_lun *cl, caddr_t arg, int flag,
5101     void *tg_cookie)
5102 {
5103 	int fdisk_rval;
5104 	diskaddr_t capacity;
5105 
5106 	ASSERT(!mutex_owned(CMLB_MUTEX(cl)));
5107 
5108 	mutex_enter(CMLB_MUTEX(cl));
5109 	capacity = cl->cl_blockcount;
5110 	fdisk_rval = cmlb_read_fdisk(cl, capacity, tg_cookie);
5111 	if (fdisk_rval != 0) {
5112 		mutex_exit(CMLB_MUTEX(cl));
5113 		return (fdisk_rval);
5114 	}
5115 
5116 	mutex_exit(CMLB_MUTEX(cl));
5117 	return (fdisk_rval);
5118 }
5119 #endif
5120 
5121 /*
5122  *    Function: cmlb_setup_default_geometry
5123  *
5124  * Description: This local utility routine sets the default geometry as part of
5125  *		setting the device mboot.
5126  *
5127  *   Arguments:
5128  *	cl		driver soft state (unit) structure
5129  *
5130  *	tg_cookie	cookie from target driver to be passed back to target
5131  *			driver when we call back to it through tg_ops.
5132  *
5133  *
5134  * Note: This may be redundant with cmlb_build_default_label.
5135  */
5136 static void
5137 cmlb_setup_default_geometry(struct cmlb_lun *cl, void *tg_cookie)
5138 {
5139 	struct cmlb_geom	pgeom;
5140 	struct cmlb_geom	*pgeomp = &pgeom;
5141 	int			ret;
5142 	int			geom_base_cap = 1;
5143 
5144 
5145 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
5146 
5147 	/* zero out the soft state geometry and partition table. */
5148 	bzero(&cl->cl_g, sizeof (struct dk_geom));
5149 	bzero(&cl->cl_vtoc, sizeof (struct dk_vtoc));
5150 	bzero(cl->cl_map, NDKMAP * (sizeof (struct dk_map)));
5151 
5152 	/*
5153 	 * For the rpm, we use the minimum for the disk.
5154 	 * For the head, cyl and number of sector per track,
5155 	 * if the capacity <= 1GB, head = 64, sect = 32.
5156 	 * else head = 255, sect 63
5157 	 * Note: the capacity should be equal to C*H*S values.
5158 	 * This will cause some truncation of size due to
5159 	 * round off errors. For CD-ROMs, this truncation can
5160 	 * have adverse side effects, so returning ncyl and
5161 	 * nhead as 1. The nsect will overflow for most of
5162 	 * CD-ROMs as nsect is of type ushort.
5163 	 */
5164 	if (cl->cl_alter_behavior & CMLB_FAKE_GEOM_LABEL_IOCTLS_VTOC8) {
5165 		/*
5166 		 * newfs currently can not handle 255 ntracks for SPARC
5167 		 * so get the geometry from target driver instead of coming up
5168 		 * with one based on capacity.
5169 		 */
5170 		mutex_exit(CMLB_MUTEX(cl));
5171 		ret = DK_TG_GETPHYGEOM(cl, pgeomp, tg_cookie);
5172 		mutex_enter(CMLB_MUTEX(cl));
5173 
5174 		if (ret == 0) {
5175 			geom_base_cap = 0;
5176 		} else {
5177 			cmlb_dbg(CMLB_ERROR,  cl,
5178 			    "cmlb_setup_default_geometry: "
5179 			    "tg_getphygeom failed %d\n", ret);
5180 
5181 			/* do default setting, geometry based on capacity */
5182 		}
5183 	}
5184 
5185 	if (geom_base_cap) {
5186 		if (ISCD(cl)) {
5187 			cl->cl_g.dkg_ncyl = 1;
5188 			cl->cl_g.dkg_nhead = 1;
5189 			cl->cl_g.dkg_nsect = cl->cl_blockcount;
5190 		} else if (cl->cl_blockcount <= 0x1000) {
5191 			/* Needed for unlabeled SCSI floppies. */
5192 			cl->cl_g.dkg_nhead = 2;
5193 			cl->cl_g.dkg_ncyl = 80;
5194 			cl->cl_g.dkg_pcyl = 80;
5195 			cl->cl_g.dkg_nsect = cl->cl_blockcount / (2 * 80);
5196 		} else if (cl->cl_blockcount <= 0x200000) {
5197 			cl->cl_g.dkg_nhead = 64;
5198 			cl->cl_g.dkg_nsect = 32;
5199 			cl->cl_g.dkg_ncyl = cl->cl_blockcount / (64 * 32);
5200 		} else {
5201 			cl->cl_g.dkg_nhead = 255;
5202 
5203 			cl->cl_g.dkg_nsect = ((cl->cl_blockcount +
5204 			    (UINT16_MAX * 255 * 63) - 1) /
5205 			    (UINT16_MAX * 255 * 63)) * 63;
5206 
5207 			if (cl->cl_g.dkg_nsect == 0)
5208 				cl->cl_g.dkg_nsect = (UINT16_MAX / 63) * 63;
5209 
5210 			cl->cl_g.dkg_ncyl = cl->cl_blockcount /
5211 			    (255 * cl->cl_g.dkg_nsect);
5212 		}
5213 
5214 		cl->cl_g.dkg_acyl = 0;
5215 		cl->cl_g.dkg_bcyl = 0;
5216 		cl->cl_g.dkg_intrlv = 1;
5217 		cl->cl_g.dkg_rpm = 200;
5218 		if (cl->cl_g.dkg_pcyl == 0)
5219 			cl->cl_g.dkg_pcyl = cl->cl_g.dkg_ncyl +
5220 			    cl->cl_g.dkg_acyl;
5221 	} else {
5222 		cl->cl_g.dkg_ncyl = (short)pgeomp->g_ncyl;
5223 		cl->cl_g.dkg_acyl = pgeomp->g_acyl;
5224 		cl->cl_g.dkg_nhead = pgeomp->g_nhead;
5225 		cl->cl_g.dkg_nsect = pgeomp->g_nsect;
5226 		cl->cl_g.dkg_intrlv = pgeomp->g_intrlv;
5227 		cl->cl_g.dkg_rpm = pgeomp->g_rpm;
5228 		cl->cl_g.dkg_pcyl = cl->cl_g.dkg_ncyl + cl->cl_g.dkg_acyl;
5229 	}
5230 
5231 	cl->cl_g.dkg_read_reinstruct = 0;
5232 	cl->cl_g.dkg_write_reinstruct = 0;
5233 	cl->cl_solaris_size = cl->cl_g.dkg_ncyl *
5234 	    cl->cl_g.dkg_nhead * cl->cl_g.dkg_nsect;
5235 
5236 	cl->cl_map['a'-'a'].dkl_cylno = 0;
5237 	cl->cl_map['a'-'a'].dkl_nblk = cl->cl_solaris_size;
5238 
5239 	cl->cl_map['c'-'a'].dkl_cylno = 0;
5240 	cl->cl_map['c'-'a'].dkl_nblk = cl->cl_solaris_size;
5241 
5242 	cl->cl_vtoc.v_part[2].p_tag   = V_BACKUP;
5243 	cl->cl_vtoc.v_part[2].p_flag  = V_UNMNT;
5244 	cl->cl_vtoc.v_nparts = V_NUMPAR;
5245 	cl->cl_vtoc.v_version = V_VERSION;
5246 	(void) sprintf((char *)cl->cl_asciilabel, "DEFAULT cyl %d alt %d"
5247 	    " hd %d sec %d", cl->cl_g.dkg_ncyl, cl->cl_g.dkg_acyl,
5248 	    cl->cl_g.dkg_nhead, cl->cl_g.dkg_nsect);
5249 
5250 	cl->cl_f_geometry_is_valid = B_FALSE;
5251 }
5252 
5253 
5254 #if defined(__i386) || defined(__amd64)
5255 /*
5256  *    Function: cmlb_update_fdisk_and_vtoc
5257  *
5258  * Description: This local utility routine updates the device fdisk and vtoc
5259  *		as part of setting the device mboot.
5260  *
5261  *   Arguments:
5262  *	cl		driver soft state (unit) structure
5263  *
5264  *	tg_cookie	cookie from target driver to be passed back to target
5265  *			driver when we call back to it through tg_ops.
5266  *
5267  *
5268  * Return Code: 0 for success or errno-type return code.
5269  *
5270  *    Note:x86: This looks like a duplicate of cmlb_validate_geometry(), but
5271  *		these did exist separately in x86 sd.c.
5272  */
5273 static int
5274 cmlb_update_fdisk_and_vtoc(struct cmlb_lun *cl, void *tg_cookie)
5275 {
5276 	int		count;
5277 	int		label_rc = 0;
5278 	int		fdisk_rval;
5279 	diskaddr_t	capacity;
5280 
5281 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
5282 
5283 	if (cmlb_check_update_blockcount(cl, tg_cookie) != 0)
5284 		return (EINVAL);
5285 
5286 #if defined(_SUNOS_VTOC_16)
5287 	/*
5288 	 * Set up the "whole disk" fdisk partition; this should always
5289 	 * exist, regardless of whether the disk contains an fdisk table
5290 	 * or vtoc.
5291 	 */
5292 	cl->cl_map[P0_RAW_DISK].dkl_cylno = 0;
5293 	cl->cl_map[P0_RAW_DISK].dkl_nblk = cl->cl_blockcount;
5294 #endif	/* defined(_SUNOS_VTOC_16) */
5295 
5296 	/*
5297 	 * copy the lbasize and capacity so that if they're
5298 	 * reset while we're not holding the CMLB_MUTEX(cl), we will
5299 	 * continue to use valid values after the CMLB_MUTEX(cl) is
5300 	 * reacquired.
5301 	 */
5302 	capacity = cl->cl_blockcount;
5303 
5304 	/*
5305 	 * refresh the logical and physical geometry caches.
5306 	 * (data from mode sense format/rigid disk geometry pages,
5307 	 * and scsi_ifgetcap("geometry").
5308 	 */
5309 	cmlb_resync_geom_caches(cl, capacity, tg_cookie);
5310 
5311 	/*
5312 	 * Only DIRECT ACCESS devices will have Scl labels.
5313 	 * CD's supposedly have a Scl label, too
5314 	 */
5315 	if (cl->cl_device_type == DTYPE_DIRECT || ISREMOVABLE(cl)) {
5316 		fdisk_rval = cmlb_read_fdisk(cl, capacity, tg_cookie);
5317 		if (fdisk_rval != 0) {
5318 			ASSERT(mutex_owned(CMLB_MUTEX(cl)));
5319 			return (fdisk_rval);
5320 		}
5321 
5322 		if (cl->cl_solaris_size <= DK_LABEL_LOC) {
5323 			/*
5324 			 * Found fdisk table but no Solaris partition entry,
5325 			 * so don't call cmlb_uselabel() and don't create
5326 			 * a default label.
5327 			 */
5328 			label_rc = 0;
5329 			cl->cl_f_geometry_is_valid = B_TRUE;
5330 			goto no_solaris_partition;
5331 		}
5332 	} else if (capacity < 0) {
5333 		ASSERT(mutex_owned(CMLB_MUTEX(cl)));
5334 		return (EINVAL);
5335 	}
5336 
5337 	/*
5338 	 * For Removable media We reach here if we have found a
5339 	 * SOLARIS PARTITION.
5340 	 * If cl_f_geometry_is_valid is B_FALSE it indicates that the SOLARIS
5341 	 * PARTITION has changed from the previous one, hence we will setup a
5342 	 * default VTOC in this case.
5343 	 */
5344 	if (!cl->cl_f_geometry_is_valid) {
5345 		/* if we get here it is writable */
5346 		/* we are called from SMBOOT, and after a write of fdisk */
5347 		cmlb_build_default_label(cl, tg_cookie);
5348 		label_rc = 0;
5349 	}
5350 
5351 no_solaris_partition:
5352 
5353 #if defined(_SUNOS_VTOC_16)
5354 	/*
5355 	 * If we have valid geometry, set up the remaining fdisk partitions.
5356 	 * Note that dkl_cylno is not used for the fdisk map entries, so
5357 	 * we set it to an entirely bogus value.
5358 	 */
5359 	for (count = 0; count < FDISK_PARTS; count++) {
5360 		cl->cl_map[FDISK_P1 + count].dkl_cylno = UINT32_MAX;
5361 		cl->cl_map[FDISK_P1 + count].dkl_nblk =
5362 		    cl->cl_fmap[count].fmap_nblk;
5363 		cl->cl_offset[FDISK_P1 + count] =
5364 		    cl->cl_fmap[count].fmap_start;
5365 	}
5366 #endif
5367 
5368 	for (count = 0; count < NDKMAP; count++) {
5369 #if defined(_SUNOS_VTOC_8)
5370 		struct dk_map *lp  = &cl->cl_map[count];
5371 		cl->cl_offset[count] =
5372 		    cl->cl_g.dkg_nhead * cl->cl_g.dkg_nsect * lp->dkl_cylno;
5373 #elif defined(_SUNOS_VTOC_16)
5374 		struct dkl_partition *vp = &cl->cl_vtoc.v_part[count];
5375 		cl->cl_offset[count] = vp->p_start + cl->cl_solaris_offset;
5376 #else
5377 #error "No VTOC format defined."
5378 #endif
5379 	}
5380 
5381 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
5382 	return (label_rc);
5383 }
5384 #endif
5385 
5386 #if defined(__i386) || defined(__amd64)
5387 static int
5388 cmlb_dkio_get_virtgeom(struct cmlb_lun *cl, caddr_t arg, int flag)
5389 {
5390 	int err = 0;
5391 
5392 	/* Return the driver's notion of the media's logical geometry */
5393 	struct dk_geom	disk_geom;
5394 	struct dk_geom	*dkgp = &disk_geom;
5395 
5396 	mutex_enter(CMLB_MUTEX(cl));
5397 	/*
5398 	 * If there is no HBA geometry available, or
5399 	 * if the HBA returned us something that doesn't
5400 	 * really fit into an Int 13/function 8 geometry
5401 	 * result, just fail the ioctl.  See PSARC 1998/313.
5402 	 */
5403 	if (cl->cl_lgeom.g_nhead == 0 ||
5404 	    cl->cl_lgeom.g_nsect == 0 ||
5405 	    cl->cl_lgeom.g_ncyl > 1024) {
5406 		mutex_exit(CMLB_MUTEX(cl));
5407 		err = EINVAL;
5408 	} else {
5409 		dkgp->dkg_ncyl	= cl->cl_lgeom.g_ncyl;
5410 		dkgp->dkg_acyl	= cl->cl_lgeom.g_acyl;
5411 		dkgp->dkg_pcyl	= dkgp->dkg_ncyl + dkgp->dkg_acyl;
5412 		dkgp->dkg_nhead	= cl->cl_lgeom.g_nhead;
5413 		dkgp->dkg_nsect	= cl->cl_lgeom.g_nsect;
5414 
5415 		mutex_exit(CMLB_MUTEX(cl));
5416 		if (ddi_copyout(dkgp, (void *)arg,
5417 		    sizeof (struct dk_geom), flag)) {
5418 			err = EFAULT;
5419 		} else {
5420 			err = 0;
5421 		}
5422 	}
5423 	return (err);
5424 }
5425 #endif
5426 
5427 #if defined(__i386) || defined(__amd64)
5428 static int
5429 cmlb_dkio_get_phygeom(struct cmlb_lun *cl, caddr_t  arg, int flag,
5430     void *tg_cookie)
5431 {
5432 	int err = 0;
5433 	diskaddr_t capacity;
5434 
5435 
5436 	/* Return the driver's notion of the media physical geometry */
5437 	struct dk_geom	disk_geom;
5438 	struct dk_geom	*dkgp = &disk_geom;
5439 
5440 	mutex_enter(CMLB_MUTEX(cl));
5441 
5442 	if (cl->cl_g.dkg_nhead != 0 &&
5443 	    cl->cl_g.dkg_nsect != 0) {
5444 		/*
5445 		 * We succeeded in getting a geometry, but
5446 		 * right now it is being reported as just the
5447 		 * Solaris fdisk partition, just like for
5448 		 * DKIOCGGEOM. We need to change that to be
5449 		 * correct for the entire disk now.
5450 		 */
5451 		bcopy(&cl->cl_g, dkgp, sizeof (*dkgp));
5452 		dkgp->dkg_acyl = 0;
5453 		dkgp->dkg_ncyl = cl->cl_blockcount /
5454 		    (dkgp->dkg_nhead * dkgp->dkg_nsect);
5455 	} else {
5456 		bzero(dkgp, sizeof (struct dk_geom));
5457 		/*
5458 		 * This disk does not have a Solaris VTOC
5459 		 * so we must present a physical geometry
5460 		 * that will remain consistent regardless
5461 		 * of how the disk is used. This will ensure
5462 		 * that the geometry does not change regardless
5463 		 * of the fdisk partition type (ie. EFI, FAT32,
5464 		 * Solaris, etc).
5465 		 */
5466 		if (ISCD(cl)) {
5467 			dkgp->dkg_nhead = cl->cl_pgeom.g_nhead;
5468 			dkgp->dkg_nsect = cl->cl_pgeom.g_nsect;
5469 			dkgp->dkg_ncyl = cl->cl_pgeom.g_ncyl;
5470 			dkgp->dkg_acyl = cl->cl_pgeom.g_acyl;
5471 		} else {
5472 			/*
5473 			 * Invalid cl_blockcount can generate invalid
5474 			 * dk_geom and may result in division by zero
5475 			 * system failure. Should make sure blockcount
5476 			 * is valid before using it here.
5477 			 */
5478 			if (cl->cl_blockcount == 0) {
5479 				mutex_exit(CMLB_MUTEX(cl));
5480 				err = EIO;
5481 				return (err);
5482 			}
5483 			/*
5484 			 * Refer to comments related to off-by-1 at the
5485 			 * header of this file
5486 			 */
5487 			if (cl->cl_alter_behavior & CMLB_OFF_BY_ONE)
5488 				capacity = cl->cl_blockcount - 1;
5489 			else
5490 				capacity = cl->cl_blockcount;
5491 
5492 			cmlb_convert_geometry(cl, capacity, dkgp, tg_cookie);
5493 			dkgp->dkg_acyl = 0;
5494 			dkgp->dkg_ncyl = capacity /
5495 			    (dkgp->dkg_nhead * dkgp->dkg_nsect);
5496 		}
5497 	}
5498 	dkgp->dkg_pcyl = dkgp->dkg_ncyl + dkgp->dkg_acyl;
5499 
5500 	mutex_exit(CMLB_MUTEX(cl));
5501 	if (ddi_copyout(dkgp, (void *)arg, sizeof (struct dk_geom), flag))
5502 		err = EFAULT;
5503 
5504 	return (err);
5505 }
5506 #endif
5507 
5508 #if defined(__i386) || defined(__amd64)
5509 static int
5510 cmlb_dkio_partinfo(struct cmlb_lun *cl, dev_t dev, caddr_t  arg, int flag)
5511 {
5512 	int err = 0;
5513 
5514 	/*
5515 	 * Return parameters describing the selected disk slice.
5516 	 * Note: this ioctl is for the intel platform only
5517 	 */
5518 	int part;
5519 
5520 	part = CMLBPART(dev);
5521 
5522 	mutex_enter(CMLB_MUTEX(cl));
5523 	/* don't check cl_solaris_size for pN */
5524 	if (part < P0_RAW_DISK && cl->cl_solaris_size == 0) {
5525 		err = EIO;
5526 		mutex_exit(CMLB_MUTEX(cl));
5527 	} else {
5528 		struct part_info p;
5529 
5530 		p.p_start = (daddr_t)cl->cl_offset[part];
5531 		p.p_length = (int)cl->cl_map[part].dkl_nblk;
5532 		mutex_exit(CMLB_MUTEX(cl));
5533 #ifdef _MULTI_DATAMODEL
5534 		switch (ddi_model_convert_from(flag & FMODELS)) {
5535 		case DDI_MODEL_ILP32:
5536 		{
5537 			struct part_info32 p32;
5538 
5539 			p32.p_start = (daddr32_t)p.p_start;
5540 			p32.p_length = p.p_length;
5541 			if (ddi_copyout(&p32, (void *)arg,
5542 			    sizeof (p32), flag))
5543 				err = EFAULT;
5544 			break;
5545 		}
5546 
5547 		case DDI_MODEL_NONE:
5548 		{
5549 			if (ddi_copyout(&p, (void *)arg, sizeof (p),
5550 			    flag))
5551 				err = EFAULT;
5552 			break;
5553 		}
5554 		}
5555 #else /* ! _MULTI_DATAMODEL */
5556 		if (ddi_copyout(&p, (void *)arg, sizeof (p), flag))
5557 			err = EFAULT;
5558 #endif /* _MULTI_DATAMODEL */
5559 	}
5560 	return (err);
5561 }
5562 static int
5563 cmlb_dkio_extpartinfo(struct cmlb_lun *cl, dev_t dev, caddr_t  arg, int flag)
5564 {
5565 	int err = 0;
5566 
5567 	/*
5568 	 * Return parameters describing the selected disk slice.
5569 	 * Note: this ioctl is for the intel platform only
5570 	 */
5571 	int part;
5572 
5573 	part = CMLBPART(dev);
5574 
5575 	mutex_enter(CMLB_MUTEX(cl));
5576 	/* don't check cl_solaris_size for pN */
5577 	if (part < P0_RAW_DISK && cl->cl_solaris_size == 0) {
5578 		err = EIO;
5579 		mutex_exit(CMLB_MUTEX(cl));
5580 	} else {
5581 		struct extpart_info p;
5582 
5583 		p.p_start = (diskaddr_t)cl->cl_offset[part];
5584 		p.p_length = (diskaddr_t)cl->cl_map[part].dkl_nblk;
5585 		mutex_exit(CMLB_MUTEX(cl));
5586 		if (ddi_copyout(&p, (void *)arg, sizeof (p), flag))
5587 			err = EFAULT;
5588 	}
5589 	return (err);
5590 }
5591 #endif
5592 
5593 int
5594 cmlb_prop_op(cmlb_handle_t cmlbhandle,
5595     dev_t dev, dev_info_t *dip, ddi_prop_op_t prop_op, int mod_flags,
5596     char *name, caddr_t valuep, int *lengthp, int part, void *tg_cookie)
5597 {
5598 	struct cmlb_lun	*cl;
5599 	diskaddr_t	capacity;
5600 	uint32_t	lbasize;
5601 	enum		dp { DP_NBLOCKS, DP_BLKSIZE } dp;
5602 	int		callers_length;
5603 	caddr_t		buffer;
5604 	uint64_t	nblocks64;
5605 	uint_t		dblk;
5606 
5607 	/* Always fallback to ddi_prop_op... */
5608 	cl = (struct cmlb_lun *)cmlbhandle;
5609 	if (cl == NULL) {
5610 fallback:	return (ddi_prop_op(dev, dip, prop_op, mod_flags,
5611 		    name, valuep, lengthp));
5612 	}
5613 
5614 	/* Pick up capacity and blocksize information. */
5615 	capacity = cl->cl_blockcount;
5616 	if (capacity == 0)
5617 		goto fallback;
5618 	lbasize = cl->cl_tgt_blocksize;
5619 	if (lbasize == 0)
5620 		lbasize = DEV_BSIZE;	/* 0 -> DEV_BSIZE units */
5621 
5622 	/* Check for dynamic property of whole device. */
5623 	if (dev == DDI_DEV_T_ANY) {
5624 		/* Fallback to ddi_prop_op if we don't understand.  */
5625 		if (strcmp(name, "device-nblocks") == 0)
5626 			dp = DP_NBLOCKS;
5627 		else if (strcmp(name, "device-blksize") == 0)
5628 			dp = DP_BLKSIZE;
5629 		else
5630 			goto fallback;
5631 
5632 		/* get callers length, establish length of our dynamic prop */
5633 		callers_length = *lengthp;
5634 		if (dp == DP_NBLOCKS)
5635 			*lengthp = sizeof (uint64_t);
5636 		else if (dp == DP_BLKSIZE)
5637 			*lengthp = sizeof (uint32_t);
5638 
5639 		/* service request for the length of the property */
5640 		if (prop_op == PROP_LEN)
5641 			return (DDI_PROP_SUCCESS);
5642 
5643 		switch (prop_op) {
5644 		case PROP_LEN_AND_VAL_ALLOC:
5645 			if ((buffer = kmem_alloc(*lengthp,
5646 			    (mod_flags & DDI_PROP_CANSLEEP) ?
5647 			    KM_SLEEP : KM_NOSLEEP)) == NULL)
5648 				return (DDI_PROP_NO_MEMORY);
5649 			*(caddr_t *)valuep = buffer;	/* set callers buf */
5650 			break;
5651 
5652 		case PROP_LEN_AND_VAL_BUF:
5653 			/* the length of the prop and the request must match */
5654 			if (callers_length != *lengthp)
5655 				return (DDI_PROP_INVAL_ARG);
5656 			buffer = valuep;		/* get callers buf */
5657 			break;
5658 
5659 		default:
5660 			return (DDI_PROP_INVAL_ARG);
5661 		}
5662 
5663 		/* transfer the value into the buffer */
5664 		if (dp == DP_NBLOCKS)
5665 			*((uint64_t *)buffer) = capacity;
5666 		else if (dp == DP_BLKSIZE)
5667 			*((uint32_t *)buffer) = lbasize;
5668 
5669 		return (DDI_PROP_SUCCESS);
5670 	}
5671 
5672 	/*
5673 	 * Support dynamic size oriented properties of partition. Requests
5674 	 * issued under conditions where size is valid are passed to
5675 	 * ddi_prop_op_nblocks with the size information, otherwise the
5676 	 * request is passed to ddi_prop_op. Size depends on valid geometry.
5677 	 */
5678 	if (!cmlb_is_valid(cmlbhandle))
5679 		goto fallback;
5680 
5681 	/* Get partition nblocks value. */
5682 	(void) cmlb_partinfo(cmlbhandle, part,
5683 	    (diskaddr_t *)&nblocks64, NULL, NULL, NULL, tg_cookie);
5684 
5685 	/*
5686 	 * Assume partition information is in sys_blocksize units, compute
5687 	 * divisor for size(9P) property representation.
5688 	 */
5689 	dblk = lbasize / cl->cl_sys_blocksize;
5690 
5691 	/* Now let ddi_prop_op_nblocks_blksize() handle the request. */
5692 	return (ddi_prop_op_nblocks_blksize(dev, dip, prop_op, mod_flags,
5693 	    name, valuep, lengthp, nblocks64 / dblk, lbasize));
5694 }
5695