xref: /titanic_41/usr/src/uts/common/io/cmlb.c (revision fad1d7055df96950837a53ae555fcf0d85aa4d19)
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(diskaddr_t capacity, struct dk_geom *cl_g);
317 #endif
318 
319 static int cmlb_dkio_get_geometry(struct cmlb_lun *cl, caddr_t arg, int flag,
320     void *tg_cookie);
321 static int cmlb_dkio_set_geometry(struct cmlb_lun *cl, caddr_t arg, int flag);
322 static int cmlb_dkio_get_partition(struct cmlb_lun *cl, caddr_t arg, int flag,
323     void *tg_cookie);
324 static int cmlb_dkio_set_partition(struct cmlb_lun *cl, caddr_t arg, int flag);
325 static int cmlb_dkio_get_efi(struct cmlb_lun *cl, caddr_t arg, int flag,
326     void *tg_cookie);
327 static int cmlb_dkio_set_efi(struct cmlb_lun *cl, dev_t dev, caddr_t arg,
328     int flag, void *tg_cookie);
329 static int cmlb_dkio_get_vtoc(struct cmlb_lun *cl, caddr_t arg, int flag,
330     void *tg_cookie);
331 static int cmlb_dkio_get_extvtoc(struct cmlb_lun *cl, caddr_t arg, int flag,
332     void *tg_cookie);
333 static int cmlb_dkio_set_vtoc(struct cmlb_lun *cl, dev_t dev, caddr_t arg,
334     int flag, void *tg_cookie);
335 static int cmlb_dkio_set_extvtoc(struct cmlb_lun *cl, dev_t dev, caddr_t arg,
336     int flag, void *tg_cookie);
337 static int cmlb_dkio_get_mboot(struct cmlb_lun *cl, caddr_t arg, int flag,
338     void *tg_cookie);
339 static int cmlb_dkio_set_mboot(struct cmlb_lun *cl, caddr_t arg, int flag,
340     void *tg_cookie);
341 static int cmlb_dkio_partition(struct cmlb_lun *cl, caddr_t arg, int flag,
342     void *tg_cookie);
343 
344 #if defined(__i386) || defined(__amd64)
345 static int cmlb_dkio_set_ext_part(struct cmlb_lun *cl, caddr_t arg, int flag,
346     void *tg_cookie);
347 static int cmlb_validate_ext_part(struct cmlb_lun *cl, int part, int epart,
348     uint32_t start, uint32_t size);
349 static int cmlb_is_linux_swap(struct cmlb_lun *cl, uint32_t part_start,
350     void *tg_cookie);
351 static int cmlb_dkio_get_virtgeom(struct cmlb_lun *cl, caddr_t arg, int flag);
352 static int cmlb_dkio_get_phygeom(struct cmlb_lun *cl, caddr_t  arg, int flag);
353 static int cmlb_dkio_partinfo(struct cmlb_lun *cl, dev_t dev, caddr_t arg,
354     int flag);
355 static int cmlb_dkio_extpartinfo(struct cmlb_lun *cl, dev_t dev, caddr_t arg,
356     int flag);
357 #endif
358 
359 static void cmlb_dbg(uint_t comp, struct cmlb_lun *cl, const char *fmt, ...);
360 static void cmlb_v_log(dev_info_t *dev, char *label, uint_t level,
361     const char *fmt, va_list ap);
362 static void cmlb_log(dev_info_t *dev, char *label, uint_t level,
363     const char *fmt, ...);
364 
365 int
366 _init(void)
367 {
368 	mutex_init(&cmlb_log_mutex, NULL, MUTEX_DRIVER, NULL);
369 	return (mod_install(&modlinkage));
370 }
371 
372 int
373 _info(struct modinfo *modinfop)
374 {
375 	return (mod_info(&modlinkage, modinfop));
376 }
377 
378 int
379 _fini(void)
380 {
381 	int err;
382 
383 	if ((err = mod_remove(&modlinkage)) != 0) {
384 		return (err);
385 	}
386 
387 	mutex_destroy(&cmlb_log_mutex);
388 	return (err);
389 }
390 
391 /*
392  * cmlb_dbg is used for debugging to log additional info
393  * Level of output is controlled via cmlb_level_mask setting.
394  */
395 static void
396 cmlb_dbg(uint_t comp, struct cmlb_lun *cl, const char *fmt, ...)
397 {
398 	va_list		ap;
399 	dev_info_t	*dev;
400 	uint_t		level_mask = 0;
401 
402 	ASSERT(cl != NULL);
403 	dev = CMLB_DEVINFO(cl);
404 	ASSERT(dev != NULL);
405 	/*
406 	 * Filter messages based on the global component and level masks,
407 	 * also print if cl matches the value of cmlb_debug_cl, or if
408 	 * cmlb_debug_cl is set to NULL.
409 	 */
410 	if (comp & CMLB_TRACE)
411 		level_mask |= CMLB_LOGMASK_TRACE;
412 
413 	if (comp & CMLB_INFO)
414 		level_mask |= CMLB_LOGMASK_INFO;
415 
416 	if (comp & CMLB_ERROR)
417 		level_mask |= CMLB_LOGMASK_ERROR;
418 
419 	if ((cmlb_level_mask & level_mask) &&
420 	    ((cmlb_debug_cl == NULL) || (cmlb_debug_cl == cl))) {
421 		va_start(ap, fmt);
422 		cmlb_v_log(dev, CMLB_LABEL(cl), CE_CONT, fmt, ap);
423 		va_end(ap);
424 	}
425 }
426 
427 /*
428  * cmlb_log is basically a duplicate of scsi_log. It is redefined here
429  * so that this module does not depend on scsi module.
430  */
431 static void
432 cmlb_log(dev_info_t *dev, char *label, uint_t level, const char *fmt, ...)
433 {
434 	va_list		ap;
435 
436 	va_start(ap, fmt);
437 	cmlb_v_log(dev, label, level, fmt, ap);
438 	va_end(ap);
439 }
440 
441 static void
442 cmlb_v_log(dev_info_t *dev, char *label, uint_t level, const char *fmt,
443     va_list ap)
444 {
445 	static char 	name[256];
446 	int 		log_only = 0;
447 	int 		boot_only = 0;
448 	int 		console_only = 0;
449 
450 	mutex_enter(&cmlb_log_mutex);
451 
452 	if (dev) {
453 		if (level == CE_PANIC || level == CE_WARN ||
454 		    level == CE_NOTE) {
455 			(void) sprintf(name, "%s (%s%d):\n",
456 			    ddi_pathname(dev, cmlb_log_buffer),
457 			    label, ddi_get_instance(dev));
458 		} else {
459 			name[0] = '\0';
460 		}
461 	} else {
462 		(void) sprintf(name, "%s:", label);
463 	}
464 
465 	(void) vsprintf(cmlb_log_buffer, fmt, ap);
466 
467 	switch (cmlb_log_buffer[0]) {
468 	case '!':
469 		log_only = 1;
470 		break;
471 	case '?':
472 		boot_only = 1;
473 		break;
474 	case '^':
475 		console_only = 1;
476 		break;
477 	}
478 
479 	switch (level) {
480 	case CE_NOTE:
481 		level = CE_CONT;
482 		/* FALLTHROUGH */
483 	case CE_CONT:
484 	case CE_WARN:
485 	case CE_PANIC:
486 		if (boot_only) {
487 			cmn_err(level, "?%s\t%s", name, &cmlb_log_buffer[1]);
488 		} else if (console_only) {
489 			cmn_err(level, "^%s\t%s", name, &cmlb_log_buffer[1]);
490 		} else if (log_only) {
491 			cmn_err(level, "!%s\t%s", name, &cmlb_log_buffer[1]);
492 		} else {
493 			cmn_err(level, "%s\t%s", name, cmlb_log_buffer);
494 		}
495 		break;
496 	case CE_IGNORE:
497 		break;
498 	default:
499 		cmn_err(CE_CONT, "^DEBUG: %s\t%s", name, cmlb_log_buffer);
500 		break;
501 	}
502 	mutex_exit(&cmlb_log_mutex);
503 }
504 
505 
506 /*
507  * cmlb_alloc_handle:
508  *
509  *	Allocates a handle.
510  *
511  * Arguments:
512  *	cmlbhandlep	pointer to handle
513  *
514  * Notes:
515  *	Allocates a handle and stores the allocated handle in the area
516  *	pointed to by cmlbhandlep
517  *
518  * Context:
519  *	Kernel thread only (can sleep).
520  */
521 void
522 cmlb_alloc_handle(cmlb_handle_t *cmlbhandlep)
523 {
524 	struct cmlb_lun 	*cl;
525 
526 	cl = kmem_zalloc(sizeof (struct cmlb_lun), KM_SLEEP);
527 	ASSERT(cmlbhandlep != NULL);
528 
529 	cl->cl_state = CMLB_INITED;
530 	cl->cl_def_labeltype = CMLB_LABEL_UNDEF;
531 	mutex_init(CMLB_MUTEX(cl), NULL, MUTEX_DRIVER, NULL);
532 
533 	*cmlbhandlep = (cmlb_handle_t)(cl);
534 }
535 
536 /*
537  * cmlb_free_handle
538  *
539  *	Frees handle.
540  *
541  * Arguments:
542  *	cmlbhandlep	pointer to handle
543  */
544 void
545 cmlb_free_handle(cmlb_handle_t *cmlbhandlep)
546 {
547 	struct cmlb_lun 	*cl;
548 
549 	cl = (struct cmlb_lun *)*cmlbhandlep;
550 	if (cl != NULL) {
551 		mutex_destroy(CMLB_MUTEX(cl));
552 		kmem_free(cl, sizeof (struct cmlb_lun));
553 	}
554 
555 }
556 
557 /*
558  * cmlb_attach:
559  *
560  *	Attach handle to device, create minor nodes for device.
561  *
562  * Arguments:
563  * 	devi		pointer to device's dev_info structure.
564  * 	tgopsp		pointer to array of functions cmlb can use to callback
565  *			to target driver.
566  *
567  *	device_type	Peripheral device type as defined in
568  *			scsi/generic/inquiry.h
569  *
570  *	is_removable	whether or not device is removable.
571  *
572  *	is_hotpluggable	whether or not device is hotpluggable.
573  *
574  *	node_type	minor node type (as used by ddi_create_minor_node)
575  *
576  *	alter_behavior
577  *			bit flags:
578  *
579  *			CMLB_CREATE_ALTSLICE_VTOC_16_DTYPE_DIRECT: create
580  *			an alternate slice for the default label, if
581  *			device type is DTYPE_DIRECT an architectures default
582  *			label type is VTOC16.
583  *			Otherwise alternate slice will no be created.
584  *
585  *
586  *			CMLB_FAKE_GEOM_LABEL_IOCTLS_VTOC8: report a default
587  *			geometry and label for DKIOCGGEOM and DKIOCGVTOC
588  *			on architecture with VTOC8 label types.
589  *
590  * 			CMLB_OFF_BY_ONE: do the workaround for legacy off-by-
591  *                      one bug in obtaining capacity (in sd):
592  *			SCSI READ_CAPACITY command returns the LBA number of the
593  *			last logical block, but sd once treated this number as
594  *			disks' capacity on x86 platform. And LBAs are addressed
595  *			based 0. So the last block was lost on x86 platform.
596  *
597  *			Now, we remove this workaround. In order for present sd
598  *			driver to work with disks which are labeled/partitioned
599  *			via previous sd, we add workaround as follows:
600  *
601  *			1) Locate backup EFI label: cmlb searches the next to
602  *			   last
603  *			   block for backup EFI label. If fails, it will
604  *			   turn to the last block for backup EFI label;
605  *
606  *			2) Clear backup EFI label: cmlb first search the last
607  *			   block for backup EFI label, and will search the
608  *			   next to last block only if failed for the last
609  *			   block.
610  *
611  *			3) Calculate geometry:refer to cmlb_convert_geometry()
612  *			   If capacity increasing by 1 causes disks' capacity
613  *			   to cross over the limits in geometry calculation,
614  *			   geometry info will change. This will raise an issue:
615  *			   In case that primary VTOC label is destroyed, format
616  *			   commandline can restore it via backup VTOC labels.
617  *			   And format locates backup VTOC labels by use of
618  *			   geometry. So changing geometry will
619  *			   prevent format from finding backup VTOC labels. To
620  *			   eliminate this side effect for compatibility,
621  *			   sd uses (capacity -1) to calculate geometry;
622  *
623  *			4) 1TB disks: some important data structures use
624  *			   32-bit signed long/int (for example, daddr_t),
625  *			   so that sd doesn't support a disk with capacity
626  *			   larger than 1TB on 32-bit platform. However,
627  *			   for exactly 1TB disk, it was treated as (1T - 512)B
628  *			   in the past, and could have valid Solaris
629  *			   partitions. To workaround this, if an exactly 1TB
630  *			   disk has Solaris fdisk partition, it will be allowed
631  *			   to work with sd.
632  *
633  *
634  *
635  *			CMLB_FAKE_LABEL_ONE_PARTITION: create s0 and s2 covering
636  *			the entire disk, if there is no valid partition info.
637  *			If there is a valid Solaris partition, s0 and s2 will
638  *			only cover the entire Solaris partition.
639  *
640  *
641  *	cmlbhandle	cmlb handle associated with device
642  *
643  *	tg_cookie	cookie from target driver to be passed back to target
644  *			driver when we call back to it through tg_ops.
645  *
646  * Notes:
647  *	Assumes a default label based on capacity for non-removable devices.
648  *	If capacity > 1TB, EFI is assumed otherwise VTOC (default VTOC
649  *	for the architecture).
650  *
651  *	For removable devices, default label type is assumed to be VTOC
652  *	type. Create minor nodes based on a default label type.
653  *	Label on the media is not validated.
654  *	minor number consists of:
655  *		if _SUNOS_VTOC_8 is defined
656  *			lowest 3 bits is taken as partition number
657  *			the rest is instance number
658  *		if _SUNOS_VTOC_16 is defined
659  *			lowest 6 bits is taken as partition number
660  *			the rest is instance number
661  *
662  *
663  * Return values:
664  *	0 	Success
665  * 	ENXIO 	creating minor nodes failed.
666  *	EINVAL  invalid arg, unsupported tg_ops version
667  */
668 int
669 cmlb_attach(dev_info_t *devi, cmlb_tg_ops_t *tgopsp, int device_type,
670     boolean_t is_removable, boolean_t is_hotpluggable, char *node_type,
671     int alter_behavior, cmlb_handle_t cmlbhandle, void *tg_cookie)
672 {
673 
674 	struct cmlb_lun	*cl = (struct cmlb_lun *)cmlbhandle;
675 	diskaddr_t	cap;
676 	int		status;
677 
678 	ASSERT(VALID_BOOLEAN(is_removable));
679 	ASSERT(VALID_BOOLEAN(is_hotpluggable));
680 
681 	if (tgopsp->tg_version < TG_DK_OPS_VERSION_1)
682 		return (EINVAL);
683 
684 	mutex_enter(CMLB_MUTEX(cl));
685 
686 	CMLB_DEVINFO(cl) = devi;
687 	cl->cmlb_tg_ops = tgopsp;
688 	cl->cl_device_type = device_type;
689 	cl->cl_is_removable = is_removable;
690 	cl->cl_is_hotpluggable = is_hotpluggable;
691 	cl->cl_node_type = node_type;
692 	cl->cl_sys_blocksize = DEV_BSIZE;
693 	cl->cl_f_geometry_is_valid = B_FALSE;
694 	cl->cl_def_labeltype = CMLB_LABEL_VTOC;
695 	cl->cl_alter_behavior = alter_behavior;
696 	cl->cl_reserved = -1;
697 	cl->cl_msglog_flag |= CMLB_ALLOW_2TB_WARN;
698 #if defined(__i386) || defined(__amd64)
699 	cl->cl_logical_drive_count = 0;
700 #endif
701 
702 	if (!is_removable) {
703 		mutex_exit(CMLB_MUTEX(cl));
704 		status = DK_TG_GETCAP(cl, &cap, tg_cookie);
705 		mutex_enter(CMLB_MUTEX(cl));
706 		if (status == 0 && cap > CMLB_EXTVTOC_LIMIT) {
707 			/* set default EFI if > 2TB */
708 			cl->cl_def_labeltype = CMLB_LABEL_EFI;
709 		}
710 	}
711 
712 	/* create minor nodes based on default label type */
713 	cl->cl_last_labeltype = CMLB_LABEL_UNDEF;
714 	cl->cl_cur_labeltype = CMLB_LABEL_UNDEF;
715 
716 	if (cmlb_create_minor_nodes(cl) != 0) {
717 		mutex_exit(CMLB_MUTEX(cl));
718 		return (ENXIO);
719 	}
720 
721 	/* Define the dynamic properties for devinfo spapshots. */
722 	i_ddi_prop_dyn_driver_set(CMLB_DEVINFO(cl), cmlb_prop_dyn);
723 
724 	cl->cl_state = CMLB_ATTACHED;
725 
726 	mutex_exit(CMLB_MUTEX(cl));
727 	return (0);
728 }
729 
730 /*
731  * cmlb_detach:
732  *
733  * Invalidate in-core labeling data and remove all minor nodes for
734  * the device associate with handle.
735  *
736  * Arguments:
737  *	cmlbhandle	cmlb handle associated with device.
738  *
739  *	tg_cookie	cookie from target driver to be passed back to target
740  *			driver when we call back to it through tg_ops.
741  *
742  */
743 /*ARGSUSED1*/
744 void
745 cmlb_detach(cmlb_handle_t cmlbhandle, void *tg_cookie)
746 {
747 	struct cmlb_lun *cl = (struct cmlb_lun *)cmlbhandle;
748 
749 	mutex_enter(CMLB_MUTEX(cl));
750 	cl->cl_def_labeltype = CMLB_LABEL_UNDEF;
751 	cl->cl_f_geometry_is_valid = B_FALSE;
752 	ddi_remove_minor_node(CMLB_DEVINFO(cl), NULL);
753 	i_ddi_prop_dyn_driver_set(CMLB_DEVINFO(cl), NULL);
754 	cl->cl_state = CMLB_INITED;
755 	mutex_exit(CMLB_MUTEX(cl));
756 }
757 
758 /*
759  * cmlb_validate:
760  *
761  *	Validates label.
762  *
763  * Arguments
764  *	cmlbhandle	cmlb handle associated with device.
765  *
766  *	flags		operation flags. used for verbosity control
767  *
768  *	tg_cookie	cookie from target driver to be passed back to target
769  *			driver when we call back to it through tg_ops.
770  *
771  *
772  * Notes:
773  *	If new label type is different from the current, adjust minor nodes
774  *	accordingly.
775  *
776  * Return values:
777  *	0		success
778  *			Note: having fdisk but no solaris partition is assumed
779  *			success.
780  *
781  *	ENOMEM		memory allocation failed
782  *	EIO		i/o errors during read or get capacity
783  * 	EACCESS		reservation conflicts
784  * 	EINVAL		label was corrupt, or no default label was assumed
785  *	ENXIO		invalid handle
786  */
787 int
788 cmlb_validate(cmlb_handle_t cmlbhandle, int flags, void *tg_cookie)
789 {
790 	struct cmlb_lun *cl = (struct cmlb_lun *)cmlbhandle;
791 	int 		rval;
792 	int  		ret = 0;
793 
794 	/*
795 	 * Temp work-around checking cl for NULL since there is a bug
796 	 * in sd_detach calling this routine from taskq_dispatch
797 	 * inited function.
798 	 */
799 	if (cl == NULL)
800 		return (ENXIO);
801 
802 	mutex_enter(CMLB_MUTEX(cl));
803 	if (cl->cl_state < CMLB_ATTACHED) {
804 		mutex_exit(CMLB_MUTEX(cl));
805 		return (ENXIO);
806 	}
807 
808 	rval = cmlb_validate_geometry((struct cmlb_lun *)cmlbhandle, B_TRUE,
809 	    flags, tg_cookie);
810 
811 	if (rval == ENOTSUP) {
812 		if (cl->cl_f_geometry_is_valid) {
813 			cl->cl_cur_labeltype = CMLB_LABEL_EFI;
814 			ret = 0;
815 		} else {
816 			ret = EINVAL;
817 		}
818 	} else {
819 		ret = rval;
820 		if (ret == 0)
821 			cl->cl_cur_labeltype = CMLB_LABEL_VTOC;
822 	}
823 
824 	if (ret == 0)
825 		(void) cmlb_create_minor_nodes(cl);
826 
827 	mutex_exit(CMLB_MUTEX(cl));
828 	return (ret);
829 }
830 
831 /*
832  * cmlb_invalidate:
833  *	Invalidate in core label data
834  *
835  * Arguments:
836  *	cmlbhandle	cmlb handle associated with device.
837  *	tg_cookie	cookie from target driver to be passed back to target
838  *			driver when we call back to it through tg_ops.
839  */
840 /*ARGSUSED1*/
841 void
842 cmlb_invalidate(cmlb_handle_t cmlbhandle, void *tg_cookie)
843 {
844 	struct cmlb_lun *cl = (struct cmlb_lun *)cmlbhandle;
845 
846 	if (cl == NULL)
847 		return;
848 
849 	mutex_enter(CMLB_MUTEX(cl));
850 	cl->cl_f_geometry_is_valid = B_FALSE;
851 	mutex_exit(CMLB_MUTEX(cl));
852 }
853 
854 /*
855  * cmlb_is_valid
856  * 	Get status on whether the incore label/geom data is valid
857  *
858  * Arguments:
859  *	cmlbhandle      cmlb handle associated with device.
860  *
861  * Return values:
862  *	B_TRUE if incore label/geom data is valid.
863  *	B_FALSE otherwise.
864  *
865  */
866 
867 
868 boolean_t
869 cmlb_is_valid(cmlb_handle_t cmlbhandle)
870 {
871 	struct cmlb_lun *cl = (struct cmlb_lun *)cmlbhandle;
872 
873 	if (cmlbhandle == NULL)
874 		return (B_FALSE);
875 
876 	return (cl->cl_f_geometry_is_valid);
877 
878 }
879 
880 
881 
882 /*
883  * cmlb_close:
884  *
885  * Close the device, revert to a default label minor node for the device,
886  * if it is removable.
887  *
888  * Arguments:
889  *	cmlbhandle	cmlb handle associated with device.
890  *
891  *	tg_cookie	cookie from target driver to be passed back to target
892  *			driver when we call back to it through tg_ops.
893  * Return values:
894  *	0	Success
895  * 	ENXIO	Re-creating minor node failed.
896  */
897 /*ARGSUSED1*/
898 int
899 cmlb_close(cmlb_handle_t cmlbhandle, void *tg_cookie)
900 {
901 	struct cmlb_lun *cl = (struct cmlb_lun *)cmlbhandle;
902 
903 	mutex_enter(CMLB_MUTEX(cl));
904 	cl->cl_f_geometry_is_valid = B_FALSE;
905 
906 	/* revert to default minor node for this device */
907 	if (ISREMOVABLE(cl)) {
908 		cl->cl_cur_labeltype = CMLB_LABEL_UNDEF;
909 		(void) cmlb_create_minor_nodes(cl);
910 	}
911 
912 	mutex_exit(CMLB_MUTEX(cl));
913 	return (0);
914 }
915 
916 /*
917  * cmlb_get_devid_block:
918  *	 get the block number where device id is stored.
919  *
920  * Arguments:
921  *	cmlbhandle	cmlb handle associated with device.
922  *	devidblockp	pointer to block number.
923  *	tg_cookie	cookie from target driver to be passed back to target
924  *			driver when we call back to it through tg_ops.
925  *
926  * Notes:
927  *	It stores the block number of device id in the area pointed to
928  *	by devidblockp.
929  * 	with the block number of device id.
930  *
931  * Return values:
932  *	0	success
933  *	EINVAL 	device id does not apply to current label type.
934  */
935 /*ARGSUSED2*/
936 int
937 cmlb_get_devid_block(cmlb_handle_t cmlbhandle, diskaddr_t *devidblockp,
938     void *tg_cookie)
939 {
940 	daddr_t			spc, blk, head, cyl;
941 	struct cmlb_lun *cl = (struct cmlb_lun *)cmlbhandle;
942 
943 	mutex_enter(CMLB_MUTEX(cl));
944 	if (cl->cl_state < CMLB_ATTACHED) {
945 		mutex_exit(CMLB_MUTEX(cl));
946 		return (EINVAL);
947 	}
948 
949 	if ((!cl->cl_f_geometry_is_valid) ||
950 	    (cl->cl_solaris_size < DK_LABEL_LOC)) {
951 		mutex_exit(CMLB_MUTEX(cl));
952 		return (EINVAL);
953 	}
954 
955 	if (cl->cl_cur_labeltype == CMLB_LABEL_EFI) {
956 		if (cl->cl_reserved != -1) {
957 			blk = cl->cl_map[cl->cl_reserved].dkl_cylno;
958 		} else {
959 			mutex_exit(CMLB_MUTEX(cl));
960 			return (EINVAL);
961 		}
962 	} else {
963 		/* if the disk is unlabeled, don't write a devid to it */
964 		if (cl->cl_label_from_media != CMLB_LABEL_VTOC) {
965 			mutex_exit(CMLB_MUTEX(cl));
966 			return (EINVAL);
967 		}
968 
969 		/* this geometry doesn't allow us to write a devid */
970 		if (cl->cl_g.dkg_acyl < 2) {
971 			mutex_exit(CMLB_MUTEX(cl));
972 			return (EINVAL);
973 		}
974 
975 		/*
976 		 * Subtract 2 guarantees that the next to last cylinder
977 		 * is used
978 		 */
979 		cyl  = cl->cl_g.dkg_ncyl  + cl->cl_g.dkg_acyl - 2;
980 		spc  = cl->cl_g.dkg_nhead * cl->cl_g.dkg_nsect;
981 		head = cl->cl_g.dkg_nhead - 1;
982 		blk  = cl->cl_solaris_offset +
983 		    (cyl * (spc - cl->cl_g.dkg_apc)) +
984 		    (head * cl->cl_g.dkg_nsect) + 1;
985 	}
986 
987 	*devidblockp = blk;
988 	mutex_exit(CMLB_MUTEX(cl));
989 	return (0);
990 }
991 
992 /*
993  * cmlb_partinfo:
994  *	Get partition info for specified partition number.
995  *
996  * Arguments:
997  *	cmlbhandle	cmlb handle associated with device.
998  *	part		partition number
999  *	nblocksp	pointer to number of blocks
1000  *	startblockp	pointer to starting block
1001  *	partnamep	pointer to name of partition
1002  *	tagp		pointer to tag info
1003  *	tg_cookie	cookie from target driver to be passed back to target
1004  *			driver when we call back to it through tg_ops.
1005  *
1006  *
1007  * Notes:
1008  *	If in-core label is not valid, this functions tries to revalidate
1009  *	the label. If label is valid, it stores the total number of blocks
1010  *	in this partition in the area pointed to by nblocksp, starting
1011  *	block number in area pointed to by startblockp,  pointer to partition
1012  *	name in area pointed to by partnamep, and tag value in area
1013  *	pointed by tagp.
1014  *	For EFI labels, tag value will be set to 0.
1015  *
1016  *	For all nblocksp, startblockp and partnamep, tagp, a value of NULL
1017  *	indicates the corresponding info is not requested.
1018  *
1019  *
1020  * Return values:
1021  *	0	success
1022  *	EINVAL  no valid label or requested partition number is invalid.
1023  *
1024  */
1025 int
1026 cmlb_partinfo(cmlb_handle_t cmlbhandle, int part, diskaddr_t *nblocksp,
1027     diskaddr_t *startblockp, char **partnamep, uint16_t *tagp, void *tg_cookie)
1028 {
1029 
1030 	struct cmlb_lun *cl = (struct cmlb_lun *)cmlbhandle;
1031 	int rval;
1032 #if defined(__i386) || defined(__amd64)
1033 	int ext_part;
1034 #endif
1035 
1036 	ASSERT(cl != NULL);
1037 	mutex_enter(CMLB_MUTEX(cl));
1038 	if (cl->cl_state < CMLB_ATTACHED) {
1039 		mutex_exit(CMLB_MUTEX(cl));
1040 		return (EINVAL);
1041 	}
1042 
1043 	if (part  < 0 || part >= MAXPART) {
1044 		rval = EINVAL;
1045 	} else {
1046 		if (!cl->cl_f_geometry_is_valid)
1047 			(void) cmlb_validate_geometry((struct cmlb_lun *)cl,
1048 			    B_FALSE, 0, tg_cookie);
1049 
1050 #if defined(_SUNOS_VTOC_16)
1051 		if (((!cl->cl_f_geometry_is_valid) ||
1052 		    (part < NDKMAP && cl->cl_solaris_size == 0)) &&
1053 		    (part != P0_RAW_DISK)) {
1054 #else
1055 		if ((!cl->cl_f_geometry_is_valid) ||
1056 		    (part < NDKMAP && cl->cl_solaris_size == 0)) {
1057 #endif
1058 			rval = EINVAL;
1059 		} else {
1060 			if (startblockp != NULL)
1061 				*startblockp = (diskaddr_t)cl->cl_offset[part];
1062 
1063 			if (nblocksp != NULL)
1064 				*nblocksp = (diskaddr_t)
1065 				    cl->cl_map[part].dkl_nblk;
1066 
1067 			if (tagp != NULL)
1068 				if (cl->cl_cur_labeltype == CMLB_LABEL_EFI)
1069 					*tagp = V_UNASSIGNED;
1070 				else
1071 					*tagp = cl->cl_vtoc.v_part[part].p_tag;
1072 			rval = 0;
1073 		}
1074 
1075 		/* consistent with behavior of sd for getting minor name */
1076 		if (partnamep != NULL) {
1077 #if defined(__i386) || defined(__amd64)
1078 #if defined(_FIRMWARE_NEEDS_FDISK)
1079 		if (part > FDISK_P4) {
1080 			ext_part = part-FDISK_P4-1;
1081 			*partnamep = dk_ext_minor_data[ext_part].name;
1082 		} else
1083 #endif
1084 #endif
1085 			*partnamep = dk_minor_data[part].name;
1086 		}
1087 
1088 	}
1089 
1090 	mutex_exit(CMLB_MUTEX(cl));
1091 	return (rval);
1092 }
1093 
1094 /*
1095  * cmlb_efi_label_capacity:
1096  *	Get capacity stored in EFI disk label.
1097  *
1098  * Arguments:
1099  *	cmlbhandle	cmlb handle associated with device.
1100  *	capacity	pointer to capacity stored in EFI disk label.
1101  *	tg_cookie	cookie from target driver to be passed back to target
1102  *			driver when we call back to it through tg_ops.
1103  *
1104  *
1105  * Notes:
1106  *	If in-core label is not valid, this functions tries to revalidate
1107  *	the label. If label is valid and is an EFI label, it stores the capacity
1108  *      in disk label in the area pointed to by capacity.
1109  *
1110  *
1111  * Return values:
1112  *	0	success
1113  *	EINVAL  no valid EFI label or capacity is NULL.
1114  *
1115  */
1116 int
1117 cmlb_efi_label_capacity(cmlb_handle_t cmlbhandle, diskaddr_t *capacity,
1118     void *tg_cookie)
1119 {
1120 	struct cmlb_lun *cl = (struct cmlb_lun *)cmlbhandle;
1121 	int rval;
1122 
1123 	ASSERT(cl != NULL);
1124 	mutex_enter(CMLB_MUTEX(cl));
1125 	if (cl->cl_state < CMLB_ATTACHED) {
1126 		mutex_exit(CMLB_MUTEX(cl));
1127 		return (EINVAL);
1128 	}
1129 
1130 	if (!cl->cl_f_geometry_is_valid)
1131 		(void) cmlb_validate_geometry((struct cmlb_lun *)cl, B_FALSE,
1132 		    0, tg_cookie);
1133 
1134 	if ((!cl->cl_f_geometry_is_valid) || (capacity == NULL) ||
1135 	    (cl->cl_cur_labeltype != CMLB_LABEL_EFI)) {
1136 		rval = EINVAL;
1137 	} else {
1138 		*capacity = (diskaddr_t)cl->cl_map[WD_NODE].dkl_nblk;
1139 		rval = 0;
1140 	}
1141 
1142 	mutex_exit(CMLB_MUTEX(cl));
1143 	return (rval);
1144 }
1145 
1146 /* Caller should make sure Test Unit Ready succeeds before calling this. */
1147 /*ARGSUSED*/
1148 int
1149 cmlb_ioctl(cmlb_handle_t cmlbhandle, dev_t dev, int cmd, intptr_t arg,
1150     int flag, cred_t *cred_p, int *rval_p, void *tg_cookie)
1151 {
1152 
1153 	int err;
1154 	struct cmlb_lun *cl;
1155 
1156 	cl = (struct cmlb_lun *)cmlbhandle;
1157 
1158 	ASSERT(cl != NULL);
1159 
1160 	mutex_enter(CMLB_MUTEX(cl));
1161 	if (cl->cl_state < CMLB_ATTACHED) {
1162 		mutex_exit(CMLB_MUTEX(cl));
1163 		return (EIO);
1164 	}
1165 
1166 	switch (cmd) {
1167 		case DKIOCSEXTVTOC:
1168 		case DKIOCSGEOM:
1169 		case DKIOCSETEFI:
1170 		case DKIOCSMBOOT:
1171 #if defined(__i386) || defined(__amd64)
1172 		case DKIOCSETEXTPART:
1173 #endif
1174 			break;
1175 		case DKIOCSVTOC:
1176 #if defined(__i386) || defined(__amd64)
1177 		case DKIOCPARTINFO:
1178 #endif
1179 			if (cl->cl_blockcount > CMLB_OLDVTOC_LIMIT) {
1180 				mutex_exit(CMLB_MUTEX(cl));
1181 				return (EOVERFLOW);
1182 			}
1183 			break;
1184 		default:
1185 			(void) cmlb_validate_geometry(cl, 1, CMLB_SILENT,
1186 			    tg_cookie);
1187 
1188 			switch (cmd) {
1189 			case DKIOCGVTOC:
1190 			case DKIOCGAPART:
1191 			case DKIOCSAPART:
1192 
1193 				if (cl->cl_label_from_media == CMLB_LABEL_EFI) {
1194 					/* GPT label on disk */
1195 					mutex_exit(CMLB_MUTEX(cl));
1196 					return (ENOTSUP);
1197 				} else if
1198 				    (cl->cl_blockcount > CMLB_OLDVTOC_LIMIT) {
1199 					mutex_exit(CMLB_MUTEX(cl));
1200 					return (EOVERFLOW);
1201 				}
1202 				break;
1203 
1204 			case DKIOCGGEOM:
1205 				if (cl->cl_label_from_media == CMLB_LABEL_EFI) {
1206 					/* GPT label on disk */
1207 					mutex_exit(CMLB_MUTEX(cl));
1208 					return (ENOTSUP);
1209 				}
1210 				break;
1211 			default:
1212 				break;
1213 			}
1214 	}
1215 
1216 	mutex_exit(CMLB_MUTEX(cl));
1217 
1218 	switch (cmd) {
1219 	case DKIOCGGEOM:
1220 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCGGEOM\n");
1221 		err = cmlb_dkio_get_geometry(cl, (caddr_t)arg, flag, tg_cookie);
1222 		break;
1223 
1224 	case DKIOCSGEOM:
1225 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCSGEOM\n");
1226 		err = cmlb_dkio_set_geometry(cl, (caddr_t)arg, flag);
1227 		break;
1228 
1229 	case DKIOCGAPART:
1230 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCGAPART\n");
1231 		err = cmlb_dkio_get_partition(cl, (caddr_t)arg,
1232 		    flag, tg_cookie);
1233 		break;
1234 
1235 	case DKIOCSAPART:
1236 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCSAPART\n");
1237 		err = cmlb_dkio_set_partition(cl, (caddr_t)arg, flag);
1238 		break;
1239 
1240 	case DKIOCGVTOC:
1241 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCGVTOC\n");
1242 		err = cmlb_dkio_get_vtoc(cl, (caddr_t)arg, flag, tg_cookie);
1243 		break;
1244 
1245 	case DKIOCGEXTVTOC:
1246 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCGVTOC\n");
1247 		err = cmlb_dkio_get_extvtoc(cl, (caddr_t)arg, flag, tg_cookie);
1248 		break;
1249 
1250 	case DKIOCGETEFI:
1251 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCGETEFI\n");
1252 		err = cmlb_dkio_get_efi(cl, (caddr_t)arg, flag, tg_cookie);
1253 		break;
1254 
1255 	case DKIOCPARTITION:
1256 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCPARTITION\n");
1257 		err = cmlb_dkio_partition(cl, (caddr_t)arg, flag, tg_cookie);
1258 		break;
1259 
1260 	case DKIOCSVTOC:
1261 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCSVTOC\n");
1262 		err = cmlb_dkio_set_vtoc(cl, dev, (caddr_t)arg, flag,
1263 		    tg_cookie);
1264 		break;
1265 
1266 	case DKIOCSEXTVTOC:
1267 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCSVTOC\n");
1268 		err = cmlb_dkio_set_extvtoc(cl, dev, (caddr_t)arg, flag,
1269 		    tg_cookie);
1270 		break;
1271 
1272 	case DKIOCSETEFI:
1273 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCSETEFI\n");
1274 		err = cmlb_dkio_set_efi(cl, dev, (caddr_t)arg, flag, tg_cookie);
1275 		break;
1276 
1277 	case DKIOCGMBOOT:
1278 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCGMBOOT\n");
1279 		err = cmlb_dkio_get_mboot(cl, (caddr_t)arg, flag, tg_cookie);
1280 		break;
1281 
1282 	case DKIOCSMBOOT:
1283 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCSMBOOT\n");
1284 		err = cmlb_dkio_set_mboot(cl, (caddr_t)arg, flag, tg_cookie);
1285 		break;
1286 	case DKIOCG_PHYGEOM:
1287 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCG_PHYGEOM\n");
1288 #if defined(__i386) || defined(__amd64)
1289 		err = cmlb_dkio_get_phygeom(cl, (caddr_t)arg, flag);
1290 #else
1291 		err = ENOTTY;
1292 #endif
1293 		break;
1294 	case DKIOCG_VIRTGEOM:
1295 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCG_VIRTGEOM\n");
1296 #if defined(__i386) || defined(__amd64)
1297 		err = cmlb_dkio_get_virtgeom(cl, (caddr_t)arg, flag);
1298 #else
1299 		err = ENOTTY;
1300 #endif
1301 		break;
1302 	case DKIOCPARTINFO:
1303 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCPARTINFO");
1304 #if defined(__i386) || defined(__amd64)
1305 		err = cmlb_dkio_partinfo(cl, dev, (caddr_t)arg, flag);
1306 #else
1307 		err = ENOTTY;
1308 #endif
1309 		break;
1310 	case DKIOCEXTPARTINFO:
1311 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCPARTINFO");
1312 #if defined(__i386) || defined(__amd64)
1313 		err = cmlb_dkio_extpartinfo(cl, dev, (caddr_t)arg, flag);
1314 #else
1315 		err = ENOTTY;
1316 #endif
1317 		break;
1318 #if defined(__i386) || defined(__amd64)
1319 	case DKIOCSETEXTPART:
1320 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCSETEXTPART");
1321 		err = cmlb_dkio_set_ext_part(cl, (caddr_t)arg, flag, tg_cookie);
1322 		break;
1323 #endif
1324 	default:
1325 		err = ENOTTY;
1326 
1327 	}
1328 
1329 	/*
1330 	 * An ioctl that succeeds and changed ('set') size(9P) information
1331 	 * needs to invalidate the cached devinfo snapshot to avoid having
1332 	 * old information being returned in a snapshots.
1333 	 *
1334 	 * NB: When available, call ddi_change_minor_node() to clear
1335 	 * SSIZEVALID in specfs vnodes via spec_size_invalidate().
1336 	 */
1337 	if (err == 0) {
1338 		switch (cmd) {
1339 		case DKIOCSGEOM:
1340 		case DKIOCSAPART:
1341 		case DKIOCSVTOC:
1342 		case DKIOCSEXTVTOC:
1343 		case DKIOCSETEFI:
1344 			i_ddi_prop_dyn_cache_invalidate(CMLB_DEVINFO(cl),
1345 			    i_ddi_prop_dyn_driver_get(CMLB_DEVINFO(cl)));
1346 		}
1347 	}
1348 	return (err);
1349 }
1350 
1351 dev_t
1352 cmlb_make_device(struct cmlb_lun *cl)
1353 {
1354 	return (makedevice(ddi_driver_major(CMLB_DEVINFO(cl)),
1355 	    ddi_get_instance(CMLB_DEVINFO(cl)) << CMLBUNIT_SHIFT));
1356 }
1357 
1358 /*
1359  * Function: cmlb_check_update_blockcount
1360  *
1361  * Description: If current capacity value is invalid, obtains the
1362  *		current capacity from target driver.
1363  *
1364  * Return Code: 0	success
1365  *		EIO	failure
1366  */
1367 static int
1368 cmlb_check_update_blockcount(struct cmlb_lun *cl, void *tg_cookie)
1369 {
1370 	int status;
1371 	diskaddr_t capacity;
1372 	uint32_t lbasize;
1373 
1374 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
1375 
1376 	if (cl->cl_f_geometry_is_valid)
1377 		return (0);
1378 
1379 	mutex_exit(CMLB_MUTEX(cl));
1380 	status = DK_TG_GETCAP(cl, &capacity, tg_cookie);
1381 	if (status != 0) {
1382 		mutex_enter(CMLB_MUTEX(cl));
1383 		return (EIO);
1384 	}
1385 
1386 	status = DK_TG_GETBLOCKSIZE(cl, &lbasize, tg_cookie);
1387 	mutex_enter(CMLB_MUTEX(cl));
1388 	if (status != 0)
1389 		return (EIO);
1390 
1391 	if ((capacity != 0) && (lbasize != 0)) {
1392 		cl->cl_blockcount = capacity;
1393 		cl->cl_tgt_blocksize = lbasize;
1394 		if (!cl->cl_is_removable) {
1395 			cl->cl_sys_blocksize = lbasize;
1396 		}
1397 		return (0);
1398 	} else {
1399 		return (EIO);
1400 	}
1401 }
1402 
1403 static int
1404 cmlb_create_minor(dev_info_t *dip, char *name, int spec_type,
1405     minor_t minor_num, char *node_type, int flag, boolean_t internal)
1406 {
1407 	ASSERT(VALID_BOOLEAN(internal));
1408 
1409 	if (internal)
1410 		return (ddi_create_internal_pathname(dip,
1411 		    name, spec_type, minor_num));
1412 	else
1413 		return (ddi_create_minor_node(dip,
1414 		    name, spec_type, minor_num, node_type, flag));
1415 }
1416 
1417 /*
1418  *    Function: cmlb_create_minor_nodes
1419  *
1420  * Description: Create or adjust the minor device nodes for the instance.
1421  * 		Minor nodes are created based on default label type,
1422  *		current label type and last label type we created
1423  *		minor nodes based on.
1424  *
1425  *
1426  *   Arguments: cl - driver soft state (unit) structure
1427  *
1428  * Return Code: 0 success
1429  *		ENXIO	failure.
1430  *
1431  *     Context: Kernel thread context
1432  */
1433 static int
1434 cmlb_create_minor_nodes(struct cmlb_lun *cl)
1435 {
1436 	struct driver_minor_data	*dmdp;
1437 	int				instance;
1438 	char				name[48];
1439 	cmlb_label_t			newlabeltype;
1440 	boolean_t			internal;
1441 
1442 	ASSERT(cl != NULL);
1443 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
1444 
1445 	internal = VOID2BOOLEAN(
1446 	    (cl->cl_alter_behavior & (CMLB_INTERNAL_MINOR_NODES)) != 0);
1447 
1448 	/* check the most common case */
1449 	if (cl->cl_cur_labeltype != CMLB_LABEL_UNDEF &&
1450 	    cl->cl_last_labeltype == cl->cl_cur_labeltype) {
1451 		/* do nothing */
1452 		return (0);
1453 	}
1454 
1455 	if (cl->cl_def_labeltype == CMLB_LABEL_UNDEF) {
1456 		/* we should never get here */
1457 		return (ENXIO);
1458 	}
1459 
1460 	if (cl->cl_last_labeltype == CMLB_LABEL_UNDEF) {
1461 		/* first time during attach */
1462 		newlabeltype = cl->cl_def_labeltype;
1463 
1464 		instance = ddi_get_instance(CMLB_DEVINFO(cl));
1465 
1466 		/* Create all the minor nodes for this target. */
1467 		dmdp = (newlabeltype == CMLB_LABEL_EFI) ? dk_minor_data_efi :
1468 		    dk_minor_data;
1469 		while (dmdp->name != NULL) {
1470 
1471 			(void) sprintf(name, "%s", dmdp->name);
1472 
1473 			if (cmlb_create_minor(CMLB_DEVINFO(cl), name,
1474 			    dmdp->type,
1475 			    (instance << CMLBUNIT_SHIFT) | dmdp->minor,
1476 			    cl->cl_node_type, NULL, internal) == DDI_FAILURE) {
1477 				/*
1478 				 * Clean up any nodes that may have been
1479 				 * created, in case this fails in the middle
1480 				 * of the loop.
1481 				 */
1482 				ddi_remove_minor_node(CMLB_DEVINFO(cl), NULL);
1483 				return (ENXIO);
1484 			}
1485 			dmdp++;
1486 		}
1487 		cl->cl_last_labeltype = newlabeltype;
1488 		return (0);
1489 	}
1490 
1491 	/* Not first time  */
1492 	if (cl->cl_cur_labeltype == CMLB_LABEL_UNDEF) {
1493 		if (cl->cl_last_labeltype != cl->cl_def_labeltype) {
1494 			/* close time, revert to default. */
1495 			newlabeltype = cl->cl_def_labeltype;
1496 		} else {
1497 			/*
1498 			 * do nothing since the type for which we last created
1499 			 * nodes matches the default
1500 			 */
1501 			return (0);
1502 		}
1503 	} else {
1504 		if (cl->cl_cur_labeltype != cl->cl_last_labeltype) {
1505 			/* We are not closing, use current label type */
1506 			newlabeltype = cl->cl_cur_labeltype;
1507 		} else {
1508 			/*
1509 			 * do nothing since the type for which we last created
1510 			 * nodes matches the current label type
1511 			 */
1512 			return (0);
1513 		}
1514 	}
1515 
1516 	instance = ddi_get_instance(CMLB_DEVINFO(cl));
1517 
1518 	/*
1519 	 * Currently we only fix up the s7 node when we are switching
1520 	 * label types from or to EFI. This is consistent with
1521 	 * current behavior of sd.
1522 	 */
1523 	if (newlabeltype == CMLB_LABEL_EFI &&
1524 	    cl->cl_last_labeltype != CMLB_LABEL_EFI) {
1525 		/* from vtoc to EFI */
1526 		ddi_remove_minor_node(CMLB_DEVINFO(cl), "h");
1527 		ddi_remove_minor_node(CMLB_DEVINFO(cl), "h,raw");
1528 		(void) cmlb_create_minor(CMLB_DEVINFO(cl), "wd",
1529 		    S_IFBLK, (instance << CMLBUNIT_SHIFT) | WD_NODE,
1530 		    cl->cl_node_type, NULL, internal);
1531 		(void) cmlb_create_minor(CMLB_DEVINFO(cl), "wd,raw",
1532 		    S_IFCHR, (instance << CMLBUNIT_SHIFT) | WD_NODE,
1533 		    cl->cl_node_type, NULL, internal);
1534 	} else {
1535 		/* from efi to vtoc */
1536 		ddi_remove_minor_node(CMLB_DEVINFO(cl), "wd");
1537 		ddi_remove_minor_node(CMLB_DEVINFO(cl), "wd,raw");
1538 		(void) cmlb_create_minor(CMLB_DEVINFO(cl), "h",
1539 		    S_IFBLK, (instance << CMLBUNIT_SHIFT) | WD_NODE,
1540 		    cl->cl_node_type, NULL, internal);
1541 		(void) cmlb_create_minor(CMLB_DEVINFO(cl), "h,raw",
1542 		    S_IFCHR, (instance << CMLBUNIT_SHIFT) | WD_NODE,
1543 		    cl->cl_node_type, NULL, internal);
1544 	}
1545 
1546 	cl->cl_last_labeltype = newlabeltype;
1547 	return (0);
1548 }
1549 
1550 /*
1551  *    Function: cmlb_validate_geometry
1552  *
1553  * Description: Read the label from the disk (if present). Update the unit's
1554  *		geometry and vtoc information from the data in the label.
1555  *		Verify that the label is valid.
1556  *
1557  *   Arguments:
1558  *	cl		driver soft state (unit) structure
1559  *
1560  *	forcerevalid	force revalidation even if we are already valid.
1561  *	flags		operation flags from target driver. Used for verbosity
1562  *			control	at this time.
1563  *	tg_cookie	cookie from target driver to be passed back to target
1564  *			driver when we call back to it through tg_ops.
1565  *
1566  * Return Code: 0 - Successful completion
1567  *		EINVAL  - Invalid value in cl->cl_tgt_blocksize or
1568  *			  cl->cl_blockcount; or label on disk is corrupted
1569  *			  or unreadable.
1570  *		EACCES  - Reservation conflict at the device.
1571  *		ENOMEM  - Resource allocation error
1572  *		ENOTSUP - geometry not applicable
1573  *
1574  *     Context: Kernel thread only (can sleep).
1575  */
1576 static int
1577 cmlb_validate_geometry(struct cmlb_lun *cl, boolean_t forcerevalid, int flags,
1578     void *tg_cookie)
1579 {
1580 	int		label_error = 0;
1581 	diskaddr_t	capacity;
1582 	int		count;
1583 
1584 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
1585 	ASSERT(VALID_BOOLEAN(forcerevalid));
1586 
1587 	if ((cl->cl_f_geometry_is_valid) && (!forcerevalid)) {
1588 		if (cl->cl_cur_labeltype == CMLB_LABEL_EFI)
1589 			return (ENOTSUP);
1590 		return (0);
1591 	}
1592 
1593 	if (cmlb_check_update_blockcount(cl, tg_cookie) != 0)
1594 		return (EIO);
1595 
1596 	capacity = cl->cl_blockcount;
1597 
1598 #if defined(_SUNOS_VTOC_16)
1599 	/*
1600 	 * Set up the "whole disk" fdisk partition; this should always
1601 	 * exist, regardless of whether the disk contains an fdisk table
1602 	 * or vtoc.
1603 	 */
1604 	cl->cl_map[P0_RAW_DISK].dkl_cylno = 0;
1605 	cl->cl_offset[P0_RAW_DISK] = 0;
1606 	/*
1607 	 * note if capacity > int32_max(1TB) we are in 64bit environment
1608 	 * so no truncation happens
1609 	 */
1610 	cl->cl_map[P0_RAW_DISK].dkl_nblk  = capacity;
1611 #endif
1612 	/*
1613 	 * Refresh the logical and physical geometry caches.
1614 	 * (data from MODE SENSE format/rigid disk geometry pages,
1615 	 * and scsi_ifgetcap("geometry").
1616 	 */
1617 	cmlb_resync_geom_caches(cl, capacity, tg_cookie);
1618 
1619 	cl->cl_label_from_media = CMLB_LABEL_UNDEF;
1620 	label_error = cmlb_use_efi(cl, capacity, flags, tg_cookie);
1621 	if (label_error == 0) {
1622 
1623 		/* found a valid EFI label */
1624 		cmlb_dbg(CMLB_TRACE, cl,
1625 		    "cmlb_validate_geometry: found EFI label\n");
1626 		/*
1627 		 * solaris_size and geometry_is_valid are set in
1628 		 * cmlb_use_efi
1629 		 */
1630 		return (ENOTSUP);
1631 	}
1632 
1633 	/* NO EFI label found */
1634 
1635 	if (capacity > CMLB_EXTVTOC_LIMIT) {
1636 		if (label_error == ESRCH) {
1637 			/*
1638 			 * they've configured a LUN over 2TB, but used
1639 			 * format.dat to restrict format's view of the
1640 			 * capacity to be under 2TB in some earlier Solaris
1641 			 * release.
1642 			 */
1643 			/* i.e > 2TB with a VTOC < 2TB */
1644 			if (!(flags & CMLB_SILENT) &&
1645 			    (cl->cl_msglog_flag & CMLB_ALLOW_2TB_WARN)) {
1646 
1647 				cmlb_log(CMLB_DEVINFO(cl), CMLB_LABEL(cl),
1648 				    CE_NOTE, "!Disk (%s%d) is limited to 2 TB "
1649 				    "due to VTOC label. To use the full "
1650 				    "capacity of the disk, use format(1M) to "
1651 				    "relabel the disk with EFI/GPT label.\n",
1652 				    CMLB_LABEL(cl),
1653 				    ddi_get_instance(CMLB_DEVINFO(cl)));
1654 
1655 				cl->cl_msglog_flag &= ~CMLB_ALLOW_2TB_WARN;
1656 			}
1657 		} else {
1658 				return (ENOTSUP);
1659 		}
1660 	}
1661 
1662 	label_error = 0;
1663 
1664 	/*
1665 	 * at this point it is either labeled with a VTOC or it is
1666 	 * under 1TB (<= 1TB actually for off-by-1)
1667 	 */
1668 
1669 	/*
1670 	 * Only DIRECT ACCESS devices will have Scl labels.
1671 	 * CD's supposedly have a Scl label, too
1672 	 */
1673 	if (cl->cl_device_type == DTYPE_DIRECT || ISREMOVABLE(cl)) {
1674 		struct	dk_label *dkl;
1675 		offset_t label_addr;
1676 		int	rval;
1677 		size_t	buffer_size;
1678 
1679 		/*
1680 		 * Note: This will set up cl->cl_solaris_size and
1681 		 * cl->cl_solaris_offset.
1682 		 */
1683 		rval = cmlb_read_fdisk(cl, capacity, tg_cookie);
1684 		if ((rval != 0) && !ISCD(cl)) {
1685 			ASSERT(mutex_owned(CMLB_MUTEX(cl)));
1686 			return (rval);
1687 		}
1688 
1689 		if (cl->cl_solaris_size <= DK_LABEL_LOC) {
1690 			/*
1691 			 * Found fdisk table but no Solaris partition entry,
1692 			 * so don't call cmlb_uselabel() and don't create
1693 			 * a default label.
1694 			 */
1695 			label_error = 0;
1696 			cl->cl_f_geometry_is_valid = B_TRUE;
1697 			goto no_solaris_partition;
1698 		}
1699 
1700 		label_addr = (daddr_t)(cl->cl_solaris_offset + DK_LABEL_LOC);
1701 
1702 		buffer_size = cl->cl_sys_blocksize;
1703 
1704 		cmlb_dbg(CMLB_TRACE, cl, "cmlb_validate_geometry: "
1705 		    "label_addr: 0x%x allocation size: 0x%x\n",
1706 		    label_addr, buffer_size);
1707 
1708 		if ((dkl = kmem_zalloc(buffer_size, KM_NOSLEEP)) == NULL)
1709 			return (ENOMEM);
1710 
1711 		mutex_exit(CMLB_MUTEX(cl));
1712 		rval = DK_TG_READ(cl, dkl, label_addr, buffer_size, tg_cookie);
1713 		mutex_enter(CMLB_MUTEX(cl));
1714 
1715 		switch (rval) {
1716 		case 0:
1717 			/*
1718 			 * cmlb_uselabel will establish that the geometry
1719 			 * is valid.
1720 			 */
1721 			if (cmlb_uselabel(cl,
1722 			    (struct dk_label *)(uintptr_t)dkl, flags) !=
1723 			    CMLB_LABEL_IS_VALID) {
1724 				label_error = EINVAL;
1725 			} else
1726 				cl->cl_label_from_media = CMLB_LABEL_VTOC;
1727 			break;
1728 		case EACCES:
1729 			label_error = EACCES;
1730 			break;
1731 		default:
1732 			label_error = EINVAL;
1733 			break;
1734 		}
1735 
1736 		kmem_free(dkl, buffer_size);
1737 	}
1738 
1739 	/*
1740 	 * If a valid label was not found, AND if no reservation conflict
1741 	 * was detected, then go ahead and create a default label (4069506).
1742 	 *
1743 	 * Note: currently, for VTOC_8 devices, the default label is created
1744 	 * for removables and hotpluggables only.  For VTOC_16 devices, the
1745 	 * default label will be created for all devices.
1746 	 * (see cmlb_build_default_label)
1747 	 */
1748 #if defined(_SUNOS_VTOC_8)
1749 	if ((ISREMOVABLE(cl) || ISHOTPLUGGABLE(cl)) &&
1750 	    (label_error != EACCES)) {
1751 #elif defined(_SUNOS_VTOC_16)
1752 	if (label_error != EACCES) {
1753 #endif
1754 		if (!cl->cl_f_geometry_is_valid) {
1755 			cmlb_build_default_label(cl, tg_cookie);
1756 		}
1757 		label_error = 0;
1758 	}
1759 
1760 no_solaris_partition:
1761 
1762 #if defined(_SUNOS_VTOC_16)
1763 	/*
1764 	 * If we have valid geometry, set up the remaining fdisk partitions.
1765 	 * Note that dkl_cylno is not used for the fdisk map entries, so
1766 	 * we set it to an entirely bogus value.
1767 	 */
1768 	for (count = 0; count < FDISK_PARTS; count++) {
1769 		cl->cl_map[FDISK_P1 + count].dkl_cylno = UINT16_MAX;
1770 		cl->cl_map[FDISK_P1 + count].dkl_nblk =
1771 		    cl->cl_fmap[count].fmap_nblk;
1772 
1773 		cl->cl_offset[FDISK_P1 + count] =
1774 		    cl->cl_fmap[count].fmap_start;
1775 	}
1776 #endif
1777 
1778 	for (count = 0; count < NDKMAP; count++) {
1779 #if defined(_SUNOS_VTOC_8)
1780 		struct dk_map *lp  = &cl->cl_map[count];
1781 		cl->cl_offset[count] =
1782 		    cl->cl_g.dkg_nhead * cl->cl_g.dkg_nsect * lp->dkl_cylno;
1783 #elif defined(_SUNOS_VTOC_16)
1784 		struct dkl_partition *vp = &cl->cl_vtoc.v_part[count];
1785 
1786 		cl->cl_offset[count] = vp->p_start + cl->cl_solaris_offset;
1787 #else
1788 #error "No VTOC format defined."
1789 #endif
1790 	}
1791 
1792 	return (label_error);
1793 }
1794 
1795 #if defined(_SUNOS_VTOC_16)
1796 /*
1797  *    Function: cmlb_convert_geometry
1798  *
1799  * Description: Convert physical geometry into a dk_geom structure. In
1800  *		other words, make sure we don't wrap 16-bit values.
1801  *		e.g. converting from geom_cache to dk_geom
1802  *
1803  *     Context: Kernel thread only
1804  */
1805 static void
1806 cmlb_convert_geometry(diskaddr_t capacity, struct dk_geom *cl_g)
1807 {
1808 
1809 	/* Unlabeled SCSI floppy device */
1810 	if (capacity <= 0x1000) {
1811 		cl_g->dkg_nhead = 2;
1812 		cl_g->dkg_ncyl = 80;
1813 		cl_g->dkg_nsect = capacity / (cl_g->dkg_nhead * cl_g->dkg_ncyl);
1814 		return;
1815 	}
1816 
1817 	/*
1818 	 * For all devices we calculate cylinders using the heads and sectors
1819 	 * we assign based on capacity of the device.  The algorithm is
1820 	 * designed to be compatible with the way other operating systems
1821 	 * lay out fdisk tables for X86 and to insure that the cylinders never
1822 	 * exceed 65535 to prevent problems with X86 ioctls that report
1823 	 * geometry.
1824 	 * For some smaller disk sizes we report geometry that matches those
1825 	 * used by X86 BIOS usage. For larger disks, we use SPT that are
1826 	 * multiples of 63, since other OSes that are not limited to 16-bits
1827 	 * for cylinders stop at 63 SPT we make do by using multiples of 63 SPT.
1828 	 *
1829 	 * The following table (in order) illustrates some end result
1830 	 * calculations:
1831 	 *
1832 	 * Maximum number of blocks 		nhead	nsect
1833 	 *
1834 	 * 2097152 (1GB)			64	32
1835 	 * 16777216 (8GB)			128	32
1836 	 * 1052819775 (502.02GB)		255  	63
1837 	 * 2105639550 (0.98TB)			255	126
1838 	 * 3158459325 (1.47TB)			255  	189
1839 	 * 4211279100 (1.96TB)			255  	252
1840 	 * 5264098875 (2.45TB)			255  	315
1841 	 * ...
1842 	 */
1843 
1844 	if (capacity <= 0x200000) {
1845 		cl_g->dkg_nhead = 64;
1846 		cl_g->dkg_nsect = 32;
1847 	} else if (capacity <= 0x01000000) {
1848 		cl_g->dkg_nhead = 128;
1849 		cl_g->dkg_nsect = 32;
1850 	} else {
1851 		cl_g->dkg_nhead = 255;
1852 
1853 		/* make nsect be smallest multiple of 63 */
1854 		cl_g->dkg_nsect = ((capacity +
1855 		    (UINT16_MAX * 255 * 63) - 1) /
1856 		    (UINT16_MAX * 255 * 63)) * 63;
1857 
1858 		if (cl_g->dkg_nsect == 0)
1859 			cl_g->dkg_nsect = (UINT16_MAX / 63) * 63;
1860 	}
1861 
1862 }
1863 #endif
1864 
1865 /*
1866  *    Function: cmlb_resync_geom_caches
1867  *
1868  * Description: (Re)initialize both geometry caches: the virtual geometry
1869  *            information is extracted from the HBA (the "geometry"
1870  *            capability), and the physical geometry cache data is
1871  *            generated by issuing MODE SENSE commands.
1872  *
1873  *   Arguments:
1874  *	cl 		driver soft state (unit) structure
1875  *	capacity	disk capacity in #blocks
1876  *	tg_cookie	cookie from target driver to be passed back to target
1877  *			driver when we call back to it through tg_ops.
1878  *
1879  *     Context: Kernel thread only (can sleep).
1880  */
1881 static void
1882 cmlb_resync_geom_caches(struct cmlb_lun *cl, diskaddr_t capacity,
1883     void *tg_cookie)
1884 {
1885 	struct	cmlb_geom 	pgeom;
1886 	struct	cmlb_geom	lgeom;
1887 	struct 	cmlb_geom	*pgeomp = &pgeom;
1888 	unsigned short 		nhead;
1889 	unsigned short 		nsect;
1890 	int 			spc;
1891 	int			ret;
1892 
1893 	ASSERT(cl != NULL);
1894 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
1895 
1896 	/*
1897 	 * Ask the controller for its logical geometry.
1898 	 * Note: if the HBA does not support scsi_ifgetcap("geometry"),
1899 	 * then the lgeom cache will be invalid.
1900 	 */
1901 	mutex_exit(CMLB_MUTEX(cl));
1902 	bzero(&lgeom, sizeof (struct cmlb_geom));
1903 	ret = DK_TG_GETVIRTGEOM(cl, &lgeom, tg_cookie);
1904 	mutex_enter(CMLB_MUTEX(cl));
1905 
1906 	bcopy(&lgeom, &cl->cl_lgeom, sizeof (cl->cl_lgeom));
1907 
1908 	/*
1909 	 * Initialize the pgeom cache from lgeom, so that if MODE SENSE
1910 	 * doesn't work, DKIOCG_PHYSGEOM can return reasonable values.
1911 	 */
1912 	if (ret != 0 || cl->cl_lgeom.g_nsect == 0 ||
1913 	    cl->cl_lgeom.g_nhead == 0) {
1914 		/*
1915 		 * Note: Perhaps this needs to be more adaptive? The rationale
1916 		 * is that, if there's no HBA geometry from the HBA driver, any
1917 		 * guess is good, since this is the physical geometry. If MODE
1918 		 * SENSE fails this gives a max cylinder size for non-LBA access
1919 		 */
1920 		nhead = 255;
1921 		nsect = 63;
1922 	} else {
1923 		nhead = cl->cl_lgeom.g_nhead;
1924 		nsect = cl->cl_lgeom.g_nsect;
1925 	}
1926 
1927 	if (ISCD(cl)) {
1928 		pgeomp->g_nhead = 1;
1929 		pgeomp->g_nsect = nsect * nhead;
1930 	} else {
1931 		pgeomp->g_nhead = nhead;
1932 		pgeomp->g_nsect = nsect;
1933 	}
1934 
1935 	spc = pgeomp->g_nhead * pgeomp->g_nsect;
1936 	pgeomp->g_capacity = capacity;
1937 	if (spc == 0)
1938 		pgeomp->g_ncyl = 0;
1939 	else
1940 		pgeomp->g_ncyl = pgeomp->g_capacity / spc;
1941 	pgeomp->g_acyl = 0;
1942 
1943 	/*
1944 	 * Retrieve fresh geometry data from the hardware, stash it
1945 	 * here temporarily before we rebuild the incore label.
1946 	 *
1947 	 * We want to use the MODE SENSE commands to derive the
1948 	 * physical geometry of the device, but if either command
1949 	 * fails, the logical geometry is used as the fallback for
1950 	 * disk label geometry.
1951 	 */
1952 
1953 	mutex_exit(CMLB_MUTEX(cl));
1954 	(void) DK_TG_GETPHYGEOM(cl, pgeomp, tg_cookie);
1955 	mutex_enter(CMLB_MUTEX(cl));
1956 
1957 	/*
1958 	 * Now update the real copy while holding the mutex. This
1959 	 * way the global copy is never in an inconsistent state.
1960 	 */
1961 	bcopy(pgeomp, &cl->cl_pgeom,  sizeof (cl->cl_pgeom));
1962 
1963 	cmlb_dbg(CMLB_INFO, cl, "cmlb_resync_geom_caches: "
1964 	    "(cached from lgeom)\n");
1965 	cmlb_dbg(CMLB_INFO,  cl,
1966 	    "   ncyl: %ld; acyl: %d; nhead: %d; nsect: %d\n",
1967 	    cl->cl_pgeom.g_ncyl, cl->cl_pgeom.g_acyl,
1968 	    cl->cl_pgeom.g_nhead, cl->cl_pgeom.g_nsect);
1969 	cmlb_dbg(CMLB_INFO,  cl, "   lbasize: %d; capacity: %ld; "
1970 	    "intrlv: %d; rpm: %d\n", cl->cl_pgeom.g_secsize,
1971 	    cl->cl_pgeom.g_capacity, cl->cl_pgeom.g_intrlv,
1972 	    cl->cl_pgeom.g_rpm);
1973 }
1974 
1975 
1976 #if defined(__i386) || defined(__amd64)
1977 /*
1978  *    Function: cmlb_update_ext_minor_nodes
1979  *
1980  * Description: Routine to add/remove extended partition device nodes
1981  *
1982  *   Arguments:
1983  *	cl		driver soft state (unit) structure
1984  *	num_parts	Number of logical drives found on the LUN
1985  *
1986  * Should be called with the mutex held
1987  *
1988  * Return Code: 0 for success
1989  *
1990  *     Context: User and Kernel thread
1991  *
1992  */
1993 static int
1994 cmlb_update_ext_minor_nodes(struct cmlb_lun *cl, int num_parts)
1995 {
1996 	int				i, count;
1997 	char				name[48];
1998 	int				instance;
1999 	struct driver_minor_data	*demdp, *demdpr;
2000 	char				*devnm;
2001 	dev_info_t			*pdip;
2002 	boolean_t 			internal;
2003 
2004 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
2005 	ASSERT(cl->cl_update_ext_minor_nodes == 1);
2006 
2007 	internal = VOID2BOOLEAN(
2008 	    (cl->cl_alter_behavior & (CMLB_INTERNAL_MINOR_NODES)) != 0);
2009 	instance = ddi_get_instance(CMLB_DEVINFO(cl));
2010 	demdp = dk_ext_minor_data;
2011 	demdpr = &dk_ext_minor_data[MAX_EXT_PARTS];
2012 
2013 
2014 	if (cl->cl_logical_drive_count) {
2015 		for (i = 0; i < cl->cl_logical_drive_count; i++) {
2016 			(void) sprintf(name, "%s", demdp->name);
2017 			ddi_remove_minor_node(CMLB_DEVINFO(cl), name);
2018 			(void) sprintf(name, "%s", demdpr->name);
2019 			ddi_remove_minor_node(CMLB_DEVINFO(cl), name);
2020 			demdp++;
2021 			demdpr++;
2022 		}
2023 		/* There are existing device nodes. Remove them */
2024 		devnm = kmem_alloc(MAXNAMELEN + 1, KM_SLEEP);
2025 		(void) ddi_deviname(cl->cl_devi, devnm);
2026 		pdip = ddi_get_parent(cl->cl_devi);
2027 		(void) devfs_clean(pdip, devnm + 1, DV_CLEAN_FORCE);
2028 		kmem_free(devnm, MAXNAMELEN + 1);
2029 	}
2030 
2031 	demdp = dk_ext_minor_data;
2032 	demdpr = &dk_ext_minor_data[MAX_EXT_PARTS];
2033 
2034 	for (i = 0; i < num_parts; i++) {
2035 		(void) sprintf(name, "%s", demdp->name);
2036 		if (cmlb_create_minor(CMLB_DEVINFO(cl), name,
2037 		    demdp->type,
2038 		    (instance << CMLBUNIT_SHIFT) | demdp->minor,
2039 		    cl->cl_node_type, NULL, internal) == DDI_FAILURE) {
2040 			/*
2041 			 * Clean up any nodes that may have been
2042 			 * created, in case this fails in the middle
2043 			 * of the loop.
2044 			 */
2045 			ddi_remove_minor_node(CMLB_DEVINFO(cl), NULL);
2046 			cl->cl_logical_drive_count = 0;
2047 			return (ENXIO);
2048 		}
2049 		(void) sprintf(name, "%s", demdpr->name);
2050 		if (ddi_create_minor_node(CMLB_DEVINFO(cl), name,
2051 		    demdpr->type,
2052 		    (instance << CMLBUNIT_SHIFT) | demdpr->minor,
2053 		    cl->cl_node_type, NULL) == DDI_FAILURE) {
2054 			/*
2055 			 * Clean up any nodes that may have been
2056 			 * created, in case this fails in the middle
2057 			 * of the loop.
2058 			 */
2059 			ddi_remove_minor_node(CMLB_DEVINFO(cl), NULL);
2060 			cl->cl_logical_drive_count = 0;
2061 			return (ENXIO);
2062 		}
2063 		demdp++;
2064 		demdpr++;
2065 	}
2066 
2067 	/* Update the cl_map array for logical drives */
2068 	for (count = 0; count < MAX_EXT_PARTS; count++) {
2069 		cl->cl_map[FDISK_P4 + 1 + count].dkl_cylno = UINT32_MAX;
2070 		cl->cl_map[FDISK_P4 + 1 + count].dkl_nblk =
2071 		    cl->cl_fmap[FD_NUMPART + count].fmap_nblk;
2072 		cl->cl_offset[FDISK_P4 + 1 + count] =
2073 		    cl->cl_fmap[FD_NUMPART + count].fmap_start;
2074 	}
2075 
2076 	cl->cl_logical_drive_count = i;
2077 	cl->cl_update_ext_minor_nodes = 0;
2078 	return (0);
2079 }
2080 /*
2081  *    Function: cmlb_validate_ext_part
2082  *
2083  * Description: utility routine to validate an extended partition's
2084  *		metadata as found on disk
2085  *
2086  *   Arguments:
2087  *	cl		driver soft state (unit) structure
2088  *	part		partition number of the extended partition
2089  *	epart		partition number of the logical drive
2090  *	start		absolute sector number of the start of the logical
2091  *			drive being validated
2092  *	size		size of logical drive being validated
2093  *
2094  * Return Code: 0 for success
2095  *
2096  *     Context: User and Kernel thread
2097  *
2098  * Algorithm :
2099  * Error cases are :
2100  *	1. If start block is lesser than or equal to the end block
2101  *	2. If either start block or end block is beyond the bounadry
2102  *	   of the extended partition.
2103  *	3. start or end block overlap with existing partitions.
2104  *		To check this, first make sure that the start block doesnt
2105  *		overlap with existing partitions. Then, calculate the
2106  *		possible end block for the given start block that doesnt
2107  *		overlap with existing partitions. This can be calculated by
2108  *		first setting the possible end block to the end of the
2109  *		extended partition (optimistic) and then, checking if there
2110  *		is any other partition that lies after the start of the
2111  *		partition being validated. If so, set the possible end to
2112  *		one block less than the beginning of the next nearest partition
2113  *		If the actual end block is greater than the calculated end
2114  *		block, we have an overlap.
2115  *
2116  */
2117 static int
2118 cmlb_validate_ext_part(struct cmlb_lun *cl, int part, int epart, uint32_t start,
2119     uint32_t size)
2120 {
2121 	int i;
2122 	uint32_t end = start + size - 1;
2123 	uint32_t ext_start = cl->cl_fmap[part].fmap_start;
2124 	uint32_t ext_end = ext_start + cl->cl_fmap[part].fmap_nblk - 1;
2125 	uint32_t ts, te;
2126 	uint32_t poss_end = ext_end;
2127 
2128 	if (end <= start) {
2129 		return (1);
2130 	}
2131 
2132 	/*
2133 	 * Check if the logical drive boundaries are within that of the
2134 	 * extended partition.
2135 	 */
2136 	if (start <= ext_start || start > ext_end || end <= ext_start ||
2137 	    end > ext_end) {
2138 		return (1);
2139 	}
2140 
2141 	/*
2142 	 * epart will be equal to FD_NUMPART if it is the first logical drive.
2143 	 * There is no need to check for overlaps with other logical drives,
2144 	 * since it is the only logical drive that we have come across so far.
2145 	 */
2146 	if (epart == FD_NUMPART) {
2147 		return (0);
2148 	}
2149 
2150 	/* Check for overlaps with existing logical drives */
2151 	i = FD_NUMPART;
2152 	ts = cl->cl_fmap[FD_NUMPART].fmap_start;
2153 	te = ts + cl->cl_fmap[FD_NUMPART].fmap_nblk - 1;
2154 
2155 	while ((i < epart) && ts && te) {
2156 		if (start >= ts && start <= te) {
2157 			return (1);
2158 		}
2159 
2160 		if ((ts < poss_end) && (ts > start)) {
2161 			poss_end = ts - 1;
2162 		}
2163 
2164 		i++;
2165 		ts = cl->cl_fmap[i].fmap_start;
2166 		te = ts + cl->cl_fmap[i].fmap_nblk - 1;
2167 	}
2168 
2169 	if (end > poss_end) {
2170 		return (1);
2171 	}
2172 
2173 	return (0);
2174 }
2175 
2176 
2177 /*
2178  *    Function: cmlb_is_linux_swap
2179  *
2180  * Description: utility routine to verify if a partition is a linux swap
2181  *		partition or not.
2182  *
2183  *   Arguments:
2184  *	cl		driver soft state (unit) structure
2185  *	part_start	absolute sector number of the start of the partition
2186  *			being verified
2187  *	tg_cookie	cookie from target driver to be passed back to target
2188  *			driver when we call back to it through tg_ops.
2189  *
2190  * Return Code: 0 for success
2191  *
2192  *     Context: User and Kernel thread
2193  *
2194  * Notes:
2195  *	The linux swap magic "SWAP-SPACE" or "SWAPSPACE2" is found as the
2196  *	last 10 bytes of a disk block whose size is that of the linux page
2197  *	size. This disk block is found at the beginning of the swap partition.
2198  */
2199 static int
2200 cmlb_is_linux_swap(struct cmlb_lun *cl, uint32_t part_start, void *tg_cookie)
2201 {
2202 	int		i;
2203 	int		rval = -1;
2204 	uint32_t	seek_offset;
2205 	uint32_t	linux_pg_size;
2206 	char 		*buf, *linux_swap_magic;
2207 	int		sec_sz = cl->cl_sys_blocksize;
2208 	/* Known linux kernel page sizes */
2209 	uint32_t	linux_pg_size_arr[] = {4096, };
2210 
2211 	ASSERT(cl != NULL);
2212 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
2213 
2214 	if ((buf = kmem_zalloc(sec_sz, KM_NOSLEEP)) == NULL) {
2215 		return (ENOMEM);
2216 	}
2217 
2218 	linux_swap_magic = buf + sec_sz - 10;
2219 
2220 	for (i = 0; i < sizeof (linux_pg_size_arr)/sizeof (uint32_t); i++) {
2221 		linux_pg_size = linux_pg_size_arr[i];
2222 		seek_offset = linux_pg_size/sec_sz - 1;
2223 		seek_offset += part_start;
2224 
2225 		mutex_exit(CMLB_MUTEX(cl));
2226 		rval = DK_TG_READ(cl, buf, seek_offset, sec_sz, tg_cookie);
2227 		mutex_enter(CMLB_MUTEX(cl));
2228 
2229 		if (rval != 0) {
2230 			cmlb_dbg(CMLB_ERROR,  cl,
2231 			    "cmlb_is_linux_swap: disk read err\n");
2232 			rval = EIO;
2233 			break;
2234 		}
2235 
2236 		rval = -1;
2237 
2238 		if ((strncmp(linux_swap_magic, "SWAP-SPACE", 10) == 0) ||
2239 		    (strncmp(linux_swap_magic, "SWAPSPACE2", 10) == 0)) {
2240 			/* Found a linux swap */
2241 			rval = 0;
2242 			break;
2243 		}
2244 	}
2245 
2246 	kmem_free(buf, sec_sz);
2247 	return (rval);
2248 }
2249 #endif
2250 
2251 /*
2252  *    Function: cmlb_read_fdisk
2253  *
2254  * Description: utility routine to read the fdisk table.
2255  *
2256  *   Arguments:
2257  *	cl		driver soft state (unit) structure
2258  *	capacity	disk capacity in #blocks
2259  *	tg_cookie	cookie from target driver to be passed back to target
2260  *			driver when we call back to it through tg_ops.
2261  *
2262  * Return Code: 0 for success (includes not reading for no_fdisk_present case
2263  *		errnos from tg_rw if failed to read the first block.
2264  *
2265  *     Context: Kernel thread only (can sleep).
2266  */
2267 /*ARGSUSED*/
2268 static int
2269 cmlb_read_fdisk(struct cmlb_lun *cl, diskaddr_t capacity, void *tg_cookie)
2270 {
2271 #if defined(_NO_FDISK_PRESENT)
2272 
2273 	cl->cl_solaris_offset = 0;
2274 	cl->cl_solaris_size = capacity;
2275 	bzero(cl->cl_fmap, sizeof (struct fmap) * FD_NUMPART);
2276 	return (0);
2277 
2278 #elif defined(_FIRMWARE_NEEDS_FDISK)
2279 
2280 	struct ipart	*fdp;
2281 	struct mboot	*mbp;
2282 	struct ipart	fdisk[FD_NUMPART];
2283 	int		i, k;
2284 	char		sigbuf[2];
2285 	caddr_t		bufp;
2286 	int		uidx;
2287 	int 		rval;
2288 	int		lba = 0;
2289 	uint_t		solaris_offset;	/* offset to solaris part. */
2290 	daddr_t		solaris_size;	/* size of solaris partition */
2291 	uint32_t	blocksize;
2292 #if defined(__i386) || defined(__amd64)
2293 	struct ipart	eparts[2];
2294 	struct ipart	*efdp1 = &eparts[0];
2295 	struct ipart	*efdp2 = &eparts[1];
2296 	int		ext_part_exists = 0;
2297 	int		ld_count = 0;
2298 	int		is_linux_swap;
2299 #endif
2300 
2301 	ASSERT(cl != NULL);
2302 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
2303 
2304 	/*
2305 	 * Start off assuming no fdisk table
2306 	 */
2307 	solaris_offset = 0;
2308 	solaris_size   = capacity;
2309 
2310 	blocksize = cl->cl_tgt_blocksize;
2311 
2312 	bufp = kmem_zalloc(blocksize, KM_SLEEP);
2313 
2314 	mutex_exit(CMLB_MUTEX(cl));
2315 	rval = DK_TG_READ(cl, bufp, 0, blocksize, tg_cookie);
2316 	mutex_enter(CMLB_MUTEX(cl));
2317 
2318 	if (rval != 0) {
2319 		cmlb_dbg(CMLB_ERROR,  cl,
2320 		    "cmlb_read_fdisk: fdisk read err\n");
2321 		bzero(cl->cl_fmap, sizeof (struct fmap) * FD_NUMPART);
2322 		goto done;
2323 	}
2324 
2325 	mbp = (struct mboot *)bufp;
2326 
2327 	/*
2328 	 * The fdisk table does not begin on a 4-byte boundary within the
2329 	 * master boot record, so we copy it to an aligned structure to avoid
2330 	 * alignment exceptions on some processors.
2331 	 */
2332 	bcopy(&mbp->parts[0], fdisk, sizeof (fdisk));
2333 
2334 	/*
2335 	 * Check for lba support before verifying sig; sig might not be
2336 	 * there, say on a blank disk, but the max_chs mark may still
2337 	 * be present.
2338 	 *
2339 	 * Note: LBA support and BEFs are an x86-only concept but this
2340 	 * code should work OK on SPARC as well.
2341 	 */
2342 
2343 	/*
2344 	 * First, check for lba-access-ok on root node (or prom root node)
2345 	 * if present there, don't need to search fdisk table.
2346 	 */
2347 	if (ddi_getprop(DDI_DEV_T_ANY, ddi_root_node(), 0,
2348 	    "lba-access-ok", 0) != 0) {
2349 		/* All drives do LBA; don't search fdisk table */
2350 		lba = 1;
2351 	} else {
2352 		/* Okay, look for mark in fdisk table */
2353 		for (fdp = fdisk, i = 0; i < FD_NUMPART; i++, fdp++) {
2354 			/* accumulate "lba" value from all partitions */
2355 			lba = (lba || cmlb_has_max_chs_vals(fdp));
2356 		}
2357 	}
2358 
2359 	if (lba != 0) {
2360 		dev_t dev = cmlb_make_device(cl);
2361 
2362 		if (ddi_getprop(dev, CMLB_DEVINFO(cl), DDI_PROP_DONTPASS,
2363 		    "lba-access-ok", 0) == 0) {
2364 			/* not found; create it */
2365 			if (ddi_prop_create(dev, CMLB_DEVINFO(cl), 0,
2366 			    "lba-access-ok", (caddr_t)NULL, 0) !=
2367 			    DDI_PROP_SUCCESS) {
2368 				cmlb_dbg(CMLB_ERROR,  cl,
2369 				    "cmlb_read_fdisk: Can't create lba "
2370 				    "property for instance %d\n",
2371 				    ddi_get_instance(CMLB_DEVINFO(cl)));
2372 			}
2373 		}
2374 	}
2375 
2376 	bcopy(&mbp->signature, sigbuf, sizeof (sigbuf));
2377 
2378 	/*
2379 	 * Endian-independent signature check
2380 	 */
2381 	if (((sigbuf[1] & 0xFF) != ((MBB_MAGIC >> 8) & 0xFF)) ||
2382 	    (sigbuf[0] != (MBB_MAGIC & 0xFF))) {
2383 		cmlb_dbg(CMLB_ERROR,  cl,
2384 		    "cmlb_read_fdisk: no fdisk\n");
2385 		bzero(cl->cl_fmap, sizeof (struct fmap) * FD_NUMPART);
2386 		goto done;
2387 	}
2388 
2389 #ifdef CMLBDEBUG
2390 	if (cmlb_level_mask & CMLB_LOGMASK_INFO) {
2391 		fdp = fdisk;
2392 		cmlb_dbg(CMLB_INFO,  cl, "cmlb_read_fdisk:\n");
2393 		cmlb_dbg(CMLB_INFO,  cl, "         relsect    "
2394 		    "numsect         sysid       bootid\n");
2395 		for (i = 0; i < FD_NUMPART; i++, fdp++) {
2396 			cmlb_dbg(CMLB_INFO,  cl,
2397 			    "    %d:  %8d   %8d     0x%08x     0x%08x\n",
2398 			    i, fdp->relsect, fdp->numsect,
2399 			    fdp->systid, fdp->bootid);
2400 		}
2401 	}
2402 #endif
2403 
2404 	/*
2405 	 * Try to find the unix partition
2406 	 */
2407 	uidx = -1;
2408 	solaris_offset = 0;
2409 	solaris_size   = 0;
2410 
2411 	for (fdp = fdisk, i = 0; i < FD_NUMPART; i++, fdp++) {
2412 		uint32_t relsect;
2413 		uint32_t numsect;
2414 		uchar_t systid;
2415 #if defined(__i386) || defined(__amd64)
2416 		/*
2417 		 * Stores relative block offset from the beginning of the
2418 		 * Extended Partition.
2419 		 */
2420 		int	ext_relsect = 0;
2421 #endif
2422 
2423 		if (fdp->numsect == 0) {
2424 			cl->cl_fmap[i].fmap_start = 0;
2425 			cl->cl_fmap[i].fmap_nblk  = 0;
2426 			continue;
2427 		}
2428 
2429 		/*
2430 		 * Data in the fdisk table is little-endian.
2431 		 */
2432 		relsect = LE_32(fdp->relsect);
2433 		numsect = LE_32(fdp->numsect);
2434 
2435 		cl->cl_fmap[i].fmap_start = relsect;
2436 		cl->cl_fmap[i].fmap_nblk  = numsect;
2437 		cl->cl_fmap[i].fmap_systid = LE_8(fdp->systid);
2438 
2439 #if defined(__i386) || defined(__amd64)
2440 		/* Support only one extended partition per LUN */
2441 		if ((fdp->systid == EXTDOS || fdp->systid == FDISK_EXTLBA) &&
2442 		    (ext_part_exists == 0)) {
2443 			int j;
2444 			uint32_t logdrive_offset;
2445 			uint32_t ext_numsect;
2446 			uint32_t abs_secnum;
2447 
2448 			ext_part_exists = 1;
2449 
2450 			for (j = FD_NUMPART; j < FDISK_PARTS; j++) {
2451 				mutex_exit(CMLB_MUTEX(cl));
2452 				rval = DK_TG_READ(cl, bufp,
2453 				    (relsect + ext_relsect), blocksize,
2454 				    tg_cookie);
2455 				mutex_enter(CMLB_MUTEX(cl));
2456 
2457 				if (rval != 0) {
2458 					cmlb_dbg(CMLB_ERROR,  cl,
2459 					    "cmlb_read_fdisk: Extended "
2460 					    "partition read err\n");
2461 					goto done;
2462 				}
2463 				/*
2464 				 * The first ipart entry provides the offset
2465 				 * at which the logical drive starts off from
2466 				 * the beginning of the container partition
2467 				 * and the size of the logical drive.
2468 				 * The second ipart entry provides the offset
2469 				 * of the next container partition from the
2470 				 * beginning of the extended partition.
2471 				 */
2472 				bcopy(&bufp[FDISK_PART_TABLE_START], eparts,
2473 				    sizeof (eparts));
2474 				logdrive_offset = LE_32(efdp1->relsect);
2475 				ext_numsect = LE_32(efdp1->numsect);
2476 				systid = LE_8(efdp1->systid);
2477 				if (logdrive_offset <= 0 || ext_numsect <= 0)
2478 					break;
2479 				abs_secnum = relsect + ext_relsect +
2480 				    logdrive_offset;
2481 
2482 				/* Boundary condition and overlap checking */
2483 				if (cmlb_validate_ext_part(cl, i, j, abs_secnum,
2484 				    ext_numsect)) {
2485 					break;
2486 				}
2487 
2488 				if ((cl->cl_fmap[j].fmap_start != abs_secnum) ||
2489 				    (cl->cl_fmap[j].fmap_nblk != ext_numsect) ||
2490 				    (cl->cl_fmap[j].fmap_systid != systid)) {
2491 					/*
2492 					 * Indicates change from previous
2493 					 * partinfo. Need to recreate
2494 					 * logical device nodes.
2495 					 */
2496 					cl->cl_update_ext_minor_nodes = 1;
2497 				}
2498 				cl->cl_fmap[j].fmap_start = abs_secnum;
2499 				cl->cl_fmap[j].fmap_nblk  = ext_numsect;
2500 				cl->cl_fmap[j].fmap_systid = systid;
2501 				ld_count++;
2502 
2503 				is_linux_swap = 0;
2504 				if (efdp1->systid == SUNIXOS) {
2505 					if (cmlb_is_linux_swap(cl, abs_secnum,
2506 					    tg_cookie) == 0) {
2507 						is_linux_swap = 1;
2508 					}
2509 				}
2510 
2511 				if ((efdp1->systid == SUNIXOS) ||
2512 				    (efdp1->systid == SUNIXOS2)) {
2513 					if ((uidx == -1) && (!is_linux_swap)) {
2514 						uidx = 0;
2515 						solaris_offset = abs_secnum;
2516 						solaris_size = ext_numsect;
2517 					}
2518 				}
2519 
2520 				if ((ext_relsect = LE_32(efdp2->relsect)) == 0)
2521 					break;
2522 			}
2523 		}
2524 
2525 #endif
2526 
2527 		if (fdp->systid != SUNIXOS &&
2528 		    fdp->systid != SUNIXOS2 &&
2529 		    fdp->systid != EFI_PMBR) {
2530 			continue;
2531 		}
2532 
2533 		/*
2534 		 * use the last active solaris partition id found
2535 		 * (there should only be 1 active partition id)
2536 		 *
2537 		 * if there are no active solaris partition id
2538 		 * then use the first inactive solaris partition id
2539 		 */
2540 		if ((uidx == -1) || (fdp->bootid == ACTIVE)) {
2541 #if defined(__i386) || defined(__amd64)
2542 			is_linux_swap = 0;
2543 			if (fdp->systid == SUNIXOS) {
2544 				if (cmlb_is_linux_swap(cl, relsect,
2545 				    tg_cookie) == 0) {
2546 					is_linux_swap = 1;
2547 				}
2548 			}
2549 			if (!is_linux_swap) {
2550 #endif
2551 				uidx = i;
2552 				solaris_offset = relsect;
2553 				solaris_size   = numsect;
2554 #if defined(__i386) || defined(__amd64)
2555 			}
2556 #endif
2557 		}
2558 	}
2559 #if defined(__i386) || defined(__amd64)
2560 	if (ld_count < cl->cl_logical_drive_count) {
2561 		/*
2562 		 * Some/all logical drives were deleted. Clear out
2563 		 * the fmap entries correspoding to those deleted drives.
2564 		 */
2565 		for (k = ld_count + FD_NUMPART;
2566 		    k < cl->cl_logical_drive_count + FD_NUMPART; k++) {
2567 			cl->cl_fmap[k].fmap_start = 0;
2568 			cl->cl_fmap[k].fmap_nblk  = 0;
2569 			cl->cl_fmap[k].fmap_systid = 0;
2570 		}
2571 		cl->cl_update_ext_minor_nodes = 1;
2572 	}
2573 	if (cl->cl_update_ext_minor_nodes) {
2574 		rval = cmlb_update_ext_minor_nodes(cl, ld_count);
2575 		if (rval != 0) {
2576 			goto done;
2577 		}
2578 	}
2579 #endif
2580 	cmlb_dbg(CMLB_INFO,  cl, "fdisk 0x%x 0x%lx",
2581 	    cl->cl_solaris_offset, cl->cl_solaris_size);
2582 done:
2583 
2584 	/*
2585 	 * Clear the VTOC info, only if the Solaris partition entry
2586 	 * has moved, changed size, been deleted, or if the size of
2587 	 * the partition is too small to even fit the label sector.
2588 	 */
2589 	if ((cl->cl_solaris_offset != solaris_offset) ||
2590 	    (cl->cl_solaris_size != solaris_size) ||
2591 	    solaris_size <= DK_LABEL_LOC) {
2592 		cmlb_dbg(CMLB_INFO,  cl, "fdisk moved 0x%x 0x%lx",
2593 		    solaris_offset, solaris_size);
2594 		bzero(&cl->cl_g, sizeof (struct dk_geom));
2595 		bzero(&cl->cl_vtoc, sizeof (struct dk_vtoc));
2596 		bzero(&cl->cl_map, NDKMAP * (sizeof (struct dk_map)));
2597 		cl->cl_f_geometry_is_valid = B_FALSE;
2598 	}
2599 	cl->cl_solaris_offset = solaris_offset;
2600 	cl->cl_solaris_size = solaris_size;
2601 	kmem_free(bufp, blocksize);
2602 	return (rval);
2603 
2604 #else	/* #elif defined(_FIRMWARE_NEEDS_FDISK) */
2605 #error "fdisk table presence undetermined for this platform."
2606 #endif	/* #if defined(_NO_FDISK_PRESENT) */
2607 }
2608 
2609 static void
2610 cmlb_swap_efi_gpt(efi_gpt_t *e)
2611 {
2612 	_NOTE(ASSUMING_PROTECTED(*e))
2613 	e->efi_gpt_Signature = LE_64(e->efi_gpt_Signature);
2614 	e->efi_gpt_Revision = LE_32(e->efi_gpt_Revision);
2615 	e->efi_gpt_HeaderSize = LE_32(e->efi_gpt_HeaderSize);
2616 	e->efi_gpt_HeaderCRC32 = LE_32(e->efi_gpt_HeaderCRC32);
2617 	e->efi_gpt_MyLBA = LE_64(e->efi_gpt_MyLBA);
2618 	e->efi_gpt_AlternateLBA = LE_64(e->efi_gpt_AlternateLBA);
2619 	e->efi_gpt_FirstUsableLBA = LE_64(e->efi_gpt_FirstUsableLBA);
2620 	e->efi_gpt_LastUsableLBA = LE_64(e->efi_gpt_LastUsableLBA);
2621 	UUID_LE_CONVERT(e->efi_gpt_DiskGUID, e->efi_gpt_DiskGUID);
2622 	e->efi_gpt_PartitionEntryLBA = LE_64(e->efi_gpt_PartitionEntryLBA);
2623 	e->efi_gpt_NumberOfPartitionEntries =
2624 	    LE_32(e->efi_gpt_NumberOfPartitionEntries);
2625 	e->efi_gpt_SizeOfPartitionEntry =
2626 	    LE_32(e->efi_gpt_SizeOfPartitionEntry);
2627 	e->efi_gpt_PartitionEntryArrayCRC32 =
2628 	    LE_32(e->efi_gpt_PartitionEntryArrayCRC32);
2629 }
2630 
2631 static void
2632 cmlb_swap_efi_gpe(int nparts, efi_gpe_t *p)
2633 {
2634 	int i;
2635 
2636 	_NOTE(ASSUMING_PROTECTED(*p))
2637 	for (i = 0; i < nparts; i++) {
2638 		UUID_LE_CONVERT(p[i].efi_gpe_PartitionTypeGUID,
2639 		    p[i].efi_gpe_PartitionTypeGUID);
2640 		p[i].efi_gpe_StartingLBA = LE_64(p[i].efi_gpe_StartingLBA);
2641 		p[i].efi_gpe_EndingLBA = LE_64(p[i].efi_gpe_EndingLBA);
2642 		/* PartitionAttrs */
2643 	}
2644 }
2645 
2646 static int
2647 cmlb_validate_efi(efi_gpt_t *labp)
2648 {
2649 	if (labp->efi_gpt_Signature != EFI_SIGNATURE)
2650 		return (EINVAL);
2651 	/* at least 96 bytes in this version of the spec. */
2652 	if (sizeof (efi_gpt_t) - sizeof (labp->efi_gpt_Reserved2) >
2653 	    labp->efi_gpt_HeaderSize)
2654 		return (EINVAL);
2655 	/* this should be 128 bytes */
2656 	if (labp->efi_gpt_SizeOfPartitionEntry != sizeof (efi_gpe_t))
2657 		return (EINVAL);
2658 	return (0);
2659 }
2660 
2661 /*
2662  * This function returns B_FALSE if there is a valid MBR signature and no
2663  * partition table entries of type EFI_PMBR (0xEE). Otherwise it returns B_TRUE.
2664  *
2665  * The EFI spec (1.10 and later) requires having a Protective MBR (PMBR) to
2666  * recognize the disk as GPT partitioned. However, some other OS creates an MBR
2667  * where a PMBR entry is not the only one. Also, if the first block has been
2668  * corrupted, currently best attempt to allow data access would be to try to
2669  * check for GPT headers. Hence in case of more than one partition entry, but
2670  * at least one EFI_PMBR partition type or no valid magic number, the function
2671  * returns B_TRUE to continue with looking for GPT header.
2672  */
2673 
2674 static boolean_t
2675 cmlb_check_efi_mbr(uchar_t *buf, boolean_t *is_mbr)
2676 {
2677 	struct ipart	*fdp;
2678 	struct mboot	*mbp = (struct mboot *)buf;
2679 	struct ipart	fdisk[FD_NUMPART];
2680 	int		i;
2681 
2682 	if (is_mbr != NULL)
2683 		*is_mbr = B_TRUE;
2684 
2685 	if (LE_16(mbp->signature) != MBB_MAGIC) {
2686 		if (is_mbr != NULL)
2687 			*is_mbr = B_FALSE;
2688 		return (B_TRUE);
2689 	}
2690 
2691 	bcopy(&mbp->parts[0], fdisk, sizeof (fdisk));
2692 
2693 	for (fdp = fdisk, i = 0; i < FD_NUMPART; i++, fdp++) {
2694 		if (fdp->systid == EFI_PMBR)
2695 			return (B_TRUE);
2696 	}
2697 
2698 	return (B_FALSE);
2699 }
2700 
2701 static int
2702 cmlb_use_efi(struct cmlb_lun *cl, diskaddr_t capacity, int flags,
2703     void *tg_cookie)
2704 {
2705 	int		i;
2706 	int		rval = 0;
2707 	efi_gpe_t	*partitions;
2708 	uchar_t		*buf;
2709 	uint_t		lbasize;	/* is really how much to read */
2710 	diskaddr_t	cap = 0;
2711 	uint_t		nparts;
2712 	diskaddr_t	gpe_lba;
2713 	diskaddr_t	alternate_lba;
2714 	int		iofailed = 0;
2715 	struct uuid	uuid_type_reserved = EFI_RESERVED;
2716 #if defined(_FIRMWARE_NEEDS_FDISK)
2717 	boolean_t 	is_mbr;
2718 #endif
2719 
2720 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
2721 
2722 	lbasize = cl->cl_sys_blocksize;
2723 
2724 	cl->cl_reserved = -1;
2725 	mutex_exit(CMLB_MUTEX(cl));
2726 
2727 	buf = kmem_zalloc(EFI_MIN_ARRAY_SIZE, KM_SLEEP);
2728 
2729 	rval = DK_TG_READ(cl, buf,  0, lbasize, tg_cookie);
2730 	if (rval) {
2731 		iofailed = 1;
2732 		goto done_err;
2733 	}
2734 	if (((struct dk_label *)buf)->dkl_magic == DKL_MAGIC) {
2735 		/* not ours */
2736 		rval = ESRCH;
2737 		goto done_err;
2738 	}
2739 
2740 #if defined(_FIRMWARE_NEEDS_FDISK)
2741 	if (!cmlb_check_efi_mbr(buf, &is_mbr)) {
2742 		if (is_mbr)
2743 			rval = ESRCH;
2744 		else
2745 			rval = EINVAL;
2746 		goto done_err;
2747 	}
2748 #else
2749 	if (!cmlb_check_efi_mbr(buf, NULL)) {
2750 		rval = EINVAL;
2751 		goto done_err;
2752 	}
2753 
2754 #endif
2755 
2756 	rval = DK_TG_READ(cl, buf, 1, lbasize, tg_cookie);
2757 	if (rval) {
2758 		iofailed = 1;
2759 		goto done_err;
2760 	}
2761 	cmlb_swap_efi_gpt((efi_gpt_t *)buf);
2762 
2763 	if ((rval = cmlb_validate_efi((efi_gpt_t *)buf)) != 0) {
2764 		/*
2765 		 * Couldn't read the primary, try the backup.  Our
2766 		 * capacity at this point could be based on CHS, so
2767 		 * check what the device reports.
2768 		 */
2769 		rval = DK_TG_GETCAP(cl, &cap, tg_cookie);
2770 		if (rval) {
2771 			iofailed = 1;
2772 			goto done_err;
2773 		}
2774 
2775 		/*
2776 		 * CMLB_OFF_BY_ONE case, we check the next to last block first
2777 		 * for backup GPT header, otherwise check the last block.
2778 		 */
2779 
2780 		if ((rval = DK_TG_READ(cl, buf,
2781 		    cap - ((cl->cl_alter_behavior & CMLB_OFF_BY_ONE) ? 2 : 1),
2782 		    lbasize, tg_cookie))
2783 		    != 0) {
2784 			iofailed = 1;
2785 			goto done_err;
2786 		}
2787 		cmlb_swap_efi_gpt((efi_gpt_t *)buf);
2788 
2789 		if ((rval = cmlb_validate_efi((efi_gpt_t *)buf)) != 0) {
2790 
2791 			if (!(cl->cl_alter_behavior & CMLB_OFF_BY_ONE))
2792 				goto done_err;
2793 			if ((rval = DK_TG_READ(cl, buf, cap - 1, lbasize,
2794 			    tg_cookie)) != 0)
2795 				goto done_err;
2796 			cmlb_swap_efi_gpt((efi_gpt_t *)buf);
2797 			if ((rval = cmlb_validate_efi((efi_gpt_t *)buf)) != 0)
2798 				goto done_err;
2799 		}
2800 		if (!(flags & CMLB_SILENT))
2801 			cmlb_log(CMLB_DEVINFO(cl), CMLB_LABEL(cl), CE_WARN,
2802 			    "primary label corrupt; using backup\n");
2803 	}
2804 
2805 	nparts = ((efi_gpt_t *)buf)->efi_gpt_NumberOfPartitionEntries;
2806 	gpe_lba = ((efi_gpt_t *)buf)->efi_gpt_PartitionEntryLBA;
2807 	alternate_lba = ((efi_gpt_t *)buf)->efi_gpt_AlternateLBA;
2808 
2809 	rval = DK_TG_READ(cl, buf, gpe_lba, EFI_MIN_ARRAY_SIZE, tg_cookie);
2810 	if (rval) {
2811 		iofailed = 1;
2812 		goto done_err;
2813 	}
2814 	partitions = (efi_gpe_t *)buf;
2815 
2816 	if (nparts > MAXPART) {
2817 		nparts = MAXPART;
2818 	}
2819 	cmlb_swap_efi_gpe(nparts, partitions);
2820 
2821 	mutex_enter(CMLB_MUTEX(cl));
2822 
2823 	/* Fill in partition table. */
2824 	for (i = 0; i < nparts; i++) {
2825 		if (partitions->efi_gpe_StartingLBA != 0 ||
2826 		    partitions->efi_gpe_EndingLBA != 0) {
2827 			cl->cl_map[i].dkl_cylno =
2828 			    partitions->efi_gpe_StartingLBA;
2829 			cl->cl_map[i].dkl_nblk =
2830 			    partitions->efi_gpe_EndingLBA -
2831 			    partitions->efi_gpe_StartingLBA + 1;
2832 			cl->cl_offset[i] =
2833 			    partitions->efi_gpe_StartingLBA;
2834 		}
2835 
2836 		if (cl->cl_reserved == -1) {
2837 			if (bcmp(&partitions->efi_gpe_PartitionTypeGUID,
2838 			    &uuid_type_reserved, sizeof (struct uuid)) == 0) {
2839 				cl->cl_reserved = i;
2840 			}
2841 		}
2842 		if (i == WD_NODE) {
2843 			/*
2844 			 * minor number 7 corresponds to the whole disk
2845 			 * if the disk capacity is expanded after disk is
2846 			 * labeled, minor number 7 represents the capacity
2847 			 * indicated by the disk label.
2848 			 */
2849 			cl->cl_map[i].dkl_cylno = 0;
2850 			if (alternate_lba == 1) {
2851 				/*
2852 				 * We are using backup label. Since we can
2853 				 * find a valid label at the end of disk,
2854 				 * the disk capacity is not expanded.
2855 				 */
2856 				cl->cl_map[i].dkl_nblk = capacity;
2857 			} else {
2858 				cl->cl_map[i].dkl_nblk = alternate_lba + 1;
2859 			}
2860 			cl->cl_offset[i] = 0;
2861 		}
2862 		partitions++;
2863 	}
2864 	cl->cl_solaris_offset = 0;
2865 	cl->cl_solaris_size = capacity;
2866 	cl->cl_label_from_media = CMLB_LABEL_EFI;
2867 	cl->cl_f_geometry_is_valid = B_TRUE;
2868 
2869 	/* clear the vtoc label */
2870 	bzero(&cl->cl_vtoc, sizeof (struct dk_vtoc));
2871 
2872 	kmem_free(buf, EFI_MIN_ARRAY_SIZE);
2873 	return (0);
2874 
2875 done_err:
2876 	kmem_free(buf, EFI_MIN_ARRAY_SIZE);
2877 	mutex_enter(CMLB_MUTEX(cl));
2878 done_err1:
2879 	/*
2880 	 * if we didn't find something that could look like a VTOC
2881 	 * and the disk is over 1TB, we know there isn't a valid label.
2882 	 * Otherwise let cmlb_uselabel decide what to do.  We only
2883 	 * want to invalidate this if we're certain the label isn't
2884 	 * valid because cmlb_prop_op will now fail, which in turn
2885 	 * causes things like opens and stats on the partition to fail.
2886 	 */
2887 	if ((capacity > CMLB_EXTVTOC_LIMIT) && (rval != ESRCH) && !iofailed) {
2888 		cl->cl_f_geometry_is_valid = B_FALSE;
2889 	}
2890 	return (rval);
2891 }
2892 
2893 
2894 /*
2895  *    Function: cmlb_uselabel
2896  *
2897  * Description: Validate the disk label and update the relevant data (geometry,
2898  *		partition, vtoc, and capacity data) in the cmlb_lun struct.
2899  *		Marks the geometry of the unit as being valid.
2900  *
2901  *   Arguments: cl: unit struct.
2902  *		dk_label: disk label
2903  *
2904  * Return Code: CMLB_LABEL_IS_VALID: Label read from disk is OK; geometry,
2905  *		partition, vtoc, and capacity data are good.
2906  *
2907  *		CMLB_LABEL_IS_INVALID: Magic number or checksum error in the
2908  *		label; or computed capacity does not jibe with capacity
2909  *		reported from the READ CAPACITY command.
2910  *
2911  *     Context: Kernel thread only (can sleep).
2912  */
2913 static int
2914 cmlb_uselabel(struct cmlb_lun *cl, struct dk_label *labp, int flags)
2915 {
2916 	short		*sp;
2917 	short		sum;
2918 	short		count;
2919 	int		label_error = CMLB_LABEL_IS_VALID;
2920 	int		i;
2921 	diskaddr_t	label_capacity;
2922 	uint32_t	part_end;
2923 	diskaddr_t	track_capacity;
2924 #if defined(_SUNOS_VTOC_16)
2925 	struct	dkl_partition	*vpartp;
2926 #endif
2927 	ASSERT(cl != NULL);
2928 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
2929 
2930 	/* Validate the magic number of the label. */
2931 	if (labp->dkl_magic != DKL_MAGIC) {
2932 #if defined(__sparc)
2933 		if (!ISREMOVABLE(cl) && !ISHOTPLUGGABLE(cl)) {
2934 			if (!(flags & CMLB_SILENT))
2935 				cmlb_log(CMLB_DEVINFO(cl), CMLB_LABEL(cl),
2936 				    CE_WARN,
2937 				    "Corrupt label; wrong magic number\n");
2938 		}
2939 #endif
2940 		return (CMLB_LABEL_IS_INVALID);
2941 	}
2942 
2943 	/* Validate the checksum of the label. */
2944 	sp  = (short *)labp;
2945 	sum = 0;
2946 	count = sizeof (struct dk_label) / sizeof (short);
2947 	while (count--)	 {
2948 		sum ^= *sp++;
2949 	}
2950 
2951 	if (sum != 0) {
2952 #if defined(_SUNOS_VTOC_16)
2953 		if (!ISCD(cl)) {
2954 #elif defined(_SUNOS_VTOC_8)
2955 		if (!ISREMOVABLE(cl) && !ISHOTPLUGGABLE(cl)) {
2956 #endif
2957 			if (!(flags & CMLB_SILENT))
2958 				cmlb_log(CMLB_DEVINFO(cl), CMLB_LABEL(cl),
2959 				    CE_WARN,
2960 				    "Corrupt label - label checksum failed\n");
2961 		}
2962 		return (CMLB_LABEL_IS_INVALID);
2963 	}
2964 
2965 
2966 	/*
2967 	 * Fill in geometry structure with data from label.
2968 	 */
2969 	bzero(&cl->cl_g, sizeof (struct dk_geom));
2970 	cl->cl_g.dkg_ncyl   = labp->dkl_ncyl;
2971 	cl->cl_g.dkg_acyl   = labp->dkl_acyl;
2972 	cl->cl_g.dkg_bcyl   = 0;
2973 	cl->cl_g.dkg_nhead  = labp->dkl_nhead;
2974 	cl->cl_g.dkg_nsect  = labp->dkl_nsect;
2975 	cl->cl_g.dkg_intrlv = labp->dkl_intrlv;
2976 
2977 #if defined(_SUNOS_VTOC_8)
2978 	cl->cl_g.dkg_gap1   = labp->dkl_gap1;
2979 	cl->cl_g.dkg_gap2   = labp->dkl_gap2;
2980 	cl->cl_g.dkg_bhead  = labp->dkl_bhead;
2981 #endif
2982 #if defined(_SUNOS_VTOC_16)
2983 	cl->cl_dkg_skew = labp->dkl_skew;
2984 #endif
2985 
2986 #if defined(__i386) || defined(__amd64)
2987 	cl->cl_g.dkg_apc = labp->dkl_apc;
2988 #endif
2989 
2990 	/*
2991 	 * Currently we rely on the values in the label being accurate. If
2992 	 * dkl_rpm or dkl_pcly are zero in the label, use a default value.
2993 	 *
2994 	 * Note: In the future a MODE SENSE may be used to retrieve this data,
2995 	 * although this command is optional in SCSI-2.
2996 	 */
2997 	cl->cl_g.dkg_rpm  = (labp->dkl_rpm  != 0) ? labp->dkl_rpm  : 3600;
2998 	cl->cl_g.dkg_pcyl = (labp->dkl_pcyl != 0) ? labp->dkl_pcyl :
2999 	    (cl->cl_g.dkg_ncyl + cl->cl_g.dkg_acyl);
3000 
3001 	/*
3002 	 * The Read and Write reinstruct values may not be valid
3003 	 * for older disks.
3004 	 */
3005 	cl->cl_g.dkg_read_reinstruct  = labp->dkl_read_reinstruct;
3006 	cl->cl_g.dkg_write_reinstruct = labp->dkl_write_reinstruct;
3007 
3008 	/* Fill in partition table. */
3009 #if defined(_SUNOS_VTOC_8)
3010 	for (i = 0; i < NDKMAP; i++) {
3011 		cl->cl_map[i].dkl_cylno = labp->dkl_map[i].dkl_cylno;
3012 		cl->cl_map[i].dkl_nblk  = labp->dkl_map[i].dkl_nblk;
3013 	}
3014 #endif
3015 #if  defined(_SUNOS_VTOC_16)
3016 	vpartp		= labp->dkl_vtoc.v_part;
3017 	track_capacity	= labp->dkl_nhead * labp->dkl_nsect;
3018 
3019 	/* Prevent divide by zero */
3020 	if (track_capacity == 0) {
3021 		if (!(flags & CMLB_SILENT))
3022 			cmlb_log(CMLB_DEVINFO(cl), CMLB_LABEL(cl), CE_WARN,
3023 			    "Corrupt label - zero nhead or nsect value\n");
3024 
3025 		return (CMLB_LABEL_IS_INVALID);
3026 	}
3027 
3028 	for (i = 0; i < NDKMAP; i++, vpartp++) {
3029 		cl->cl_map[i].dkl_cylno = vpartp->p_start / track_capacity;
3030 		cl->cl_map[i].dkl_nblk  = vpartp->p_size;
3031 	}
3032 #endif
3033 
3034 	/* Fill in VTOC Structure. */
3035 	bcopy(&labp->dkl_vtoc, &cl->cl_vtoc, sizeof (struct dk_vtoc));
3036 #if defined(_SUNOS_VTOC_8)
3037 	/*
3038 	 * The 8-slice vtoc does not include the ascii label; save it into
3039 	 * the device's soft state structure here.
3040 	 */
3041 	bcopy(labp->dkl_asciilabel, cl->cl_asciilabel, LEN_DKL_ASCII);
3042 #endif
3043 
3044 	/* Now look for a valid capacity. */
3045 	track_capacity	= (cl->cl_g.dkg_nhead * cl->cl_g.dkg_nsect);
3046 	label_capacity	= (cl->cl_g.dkg_ncyl  * track_capacity);
3047 
3048 	if (cl->cl_g.dkg_acyl) {
3049 #if defined(__i386) || defined(__amd64)
3050 		/* we may have > 1 alts cylinder */
3051 		label_capacity += (track_capacity * cl->cl_g.dkg_acyl);
3052 #else
3053 		label_capacity += track_capacity;
3054 #endif
3055 	}
3056 
3057 	/*
3058 	 * Force check here to ensure the computed capacity is valid.
3059 	 * If capacity is zero, it indicates an invalid label and
3060 	 * we should abort updating the relevant data then.
3061 	 */
3062 	if (label_capacity == 0) {
3063 		if (!(flags & CMLB_SILENT))
3064 			cmlb_log(CMLB_DEVINFO(cl), CMLB_LABEL(cl), CE_WARN,
3065 			    "Corrupt label - no valid capacity could be "
3066 			    "retrieved\n");
3067 
3068 		return (CMLB_LABEL_IS_INVALID);
3069 	}
3070 
3071 	/* Mark the geometry as valid. */
3072 	cl->cl_f_geometry_is_valid = B_TRUE;
3073 
3074 	/*
3075 	 * if we got invalidated when mutex exit and entered again,
3076 	 * if blockcount different than when we came in, need to
3077 	 * retry from beginning of cmlb_validate_geometry.
3078 	 * revisit this on next phase of utilizing this for
3079 	 * sd.
3080 	 */
3081 
3082 	if (label_capacity <= cl->cl_blockcount) {
3083 #if defined(_SUNOS_VTOC_8)
3084 		/*
3085 		 * We can't let this happen on drives that are subdivided
3086 		 * into logical disks (i.e., that have an fdisk table).
3087 		 * The cl_blockcount field should always hold the full media
3088 		 * size in sectors, period.  This code would overwrite
3089 		 * cl_blockcount with the size of the Solaris fdisk partition.
3090 		 */
3091 		cmlb_dbg(CMLB_ERROR,  cl,
3092 		    "cmlb_uselabel: Label %d blocks; Drive %d blocks\n",
3093 		    label_capacity, cl->cl_blockcount);
3094 		cl->cl_solaris_size = label_capacity;
3095 
3096 #endif	/* defined(_SUNOS_VTOC_8) */
3097 		goto done;
3098 	}
3099 
3100 	if (ISCD(cl)) {
3101 		/* For CDROMs, we trust that the data in the label is OK. */
3102 #if defined(_SUNOS_VTOC_8)
3103 		for (i = 0; i < NDKMAP; i++) {
3104 			part_end = labp->dkl_nhead * labp->dkl_nsect *
3105 			    labp->dkl_map[i].dkl_cylno +
3106 			    labp->dkl_map[i].dkl_nblk  - 1;
3107 
3108 			if ((labp->dkl_map[i].dkl_nblk) &&
3109 			    (part_end > cl->cl_blockcount)) {
3110 				cl->cl_f_geometry_is_valid = B_FALSE;
3111 				break;
3112 			}
3113 		}
3114 #endif
3115 #if defined(_SUNOS_VTOC_16)
3116 		vpartp = &(labp->dkl_vtoc.v_part[0]);
3117 		for (i = 0; i < NDKMAP; i++, vpartp++) {
3118 			part_end = vpartp->p_start + vpartp->p_size;
3119 			if ((vpartp->p_size > 0) &&
3120 			    (part_end > cl->cl_blockcount)) {
3121 				cl->cl_f_geometry_is_valid = B_FALSE;
3122 				break;
3123 			}
3124 		}
3125 #endif
3126 	} else {
3127 		/* label_capacity > cl->cl_blockcount */
3128 		if (!(flags & CMLB_SILENT)) {
3129 			cmlb_log(CMLB_DEVINFO(cl), CMLB_LABEL(cl), CE_WARN,
3130 			    "Corrupt label - bad geometry\n");
3131 			cmlb_log(CMLB_DEVINFO(cl), CMLB_LABEL(cl), CE_CONT,
3132 			    "Label says %llu blocks; Drive says %llu blocks\n",
3133 			    label_capacity, cl->cl_blockcount);
3134 		}
3135 		cl->cl_f_geometry_is_valid = B_FALSE;
3136 		label_error = CMLB_LABEL_IS_INVALID;
3137 	}
3138 
3139 done:
3140 
3141 	cmlb_dbg(CMLB_INFO,  cl, "cmlb_uselabel: (label geometry)\n");
3142 	cmlb_dbg(CMLB_INFO,  cl,
3143 	    "   ncyl: %d; acyl: %d; nhead: %d; nsect: %d\n",
3144 	    cl->cl_g.dkg_ncyl,  cl->cl_g.dkg_acyl,
3145 	    cl->cl_g.dkg_nhead, cl->cl_g.dkg_nsect);
3146 
3147 	cmlb_dbg(CMLB_INFO,  cl,
3148 	    "   label_capacity: %d; intrlv: %d; rpm: %d\n",
3149 	    cl->cl_blockcount, cl->cl_g.dkg_intrlv, cl->cl_g.dkg_rpm);
3150 	cmlb_dbg(CMLB_INFO,  cl, "   wrt_reinstr: %d; rd_reinstr: %d\n",
3151 	    cl->cl_g.dkg_write_reinstruct, cl->cl_g.dkg_read_reinstruct);
3152 
3153 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
3154 
3155 	return (label_error);
3156 }
3157 
3158 
3159 /*
3160  *    Function: cmlb_build_default_label
3161  *
3162  * Description: Generate a default label for those devices that do not have
3163  *		one, e.g., new media, removable cartridges, etc..
3164  *
3165  *     Context: Kernel thread only
3166  */
3167 /*ARGSUSED*/
3168 static void
3169 cmlb_build_default_label(struct cmlb_lun *cl, void *tg_cookie)
3170 {
3171 #if defined(_SUNOS_VTOC_16)
3172 	uint_t	phys_spc;
3173 	uint_t	disksize;
3174 	struct  dk_geom cl_g;
3175 	diskaddr_t capacity;
3176 #endif
3177 
3178 	ASSERT(cl != NULL);
3179 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
3180 
3181 #if defined(_SUNOS_VTOC_8)
3182 	/*
3183 	 * Note: This is a legacy check for non-removable devices on VTOC_8
3184 	 * only. This may be a valid check for VTOC_16 as well.
3185 	 * Once we understand why there is this difference between SPARC and
3186 	 * x86 platform, we could remove this legacy check.
3187 	 */
3188 	if (!ISREMOVABLE(cl) && !ISHOTPLUGGABLE(cl)) {
3189 		return;
3190 	}
3191 #endif
3192 
3193 	bzero(&cl->cl_g, sizeof (struct dk_geom));
3194 	bzero(&cl->cl_vtoc, sizeof (struct dk_vtoc));
3195 	bzero(&cl->cl_map, NDKMAP * (sizeof (struct dk_map)));
3196 
3197 #if defined(_SUNOS_VTOC_8)
3198 
3199 	/*
3200 	 * It's a REMOVABLE media, therefore no label (on sparc, anyway).
3201 	 * But it is still necessary to set up various geometry information,
3202 	 * and we are doing this here.
3203 	 */
3204 
3205 	/*
3206 	 * For the rpm, we use the minimum for the disk.  For the head, cyl,
3207 	 * and number of sector per track, if the capacity <= 1GB, head = 64,
3208 	 * sect = 32.  else head = 255, sect 63 Note: the capacity should be
3209 	 * equal to C*H*S values.  This will cause some truncation of size due
3210 	 * to round off errors. For CD-ROMs, this truncation can have adverse
3211 	 * side effects, so returning ncyl and nhead as 1. The nsect will
3212 	 * overflow for most of CD-ROMs as nsect is of type ushort. (4190569)
3213 	 */
3214 	cl->cl_solaris_size = cl->cl_blockcount;
3215 	if (ISCD(cl)) {
3216 		tg_attribute_t tgattribute;
3217 		int is_writable;
3218 		/*
3219 		 * Preserve the old behavior for non-writable
3220 		 * medias. Since dkg_nsect is a ushort, it
3221 		 * will lose bits as cdroms have more than
3222 		 * 65536 sectors. So if we recalculate
3223 		 * capacity, it will become much shorter.
3224 		 * But the dkg_* information is not
3225 		 * used for CDROMs so it is OK. But for
3226 		 * Writable CDs we need this information
3227 		 * to be valid (for newfs say). So we
3228 		 * make nsect and nhead > 1 that way
3229 		 * nsect can still stay within ushort limit
3230 		 * without losing any bits.
3231 		 */
3232 
3233 		bzero(&tgattribute, sizeof (tg_attribute_t));
3234 
3235 		mutex_exit(CMLB_MUTEX(cl));
3236 		is_writable =
3237 		    (DK_TG_GETATTRIBUTE(cl, &tgattribute, tg_cookie) == 0) ?
3238 		    tgattribute.media_is_writable : 1;
3239 		mutex_enter(CMLB_MUTEX(cl));
3240 
3241 		if (is_writable) {
3242 			cl->cl_g.dkg_nhead = 64;
3243 			cl->cl_g.dkg_nsect = 32;
3244 			cl->cl_g.dkg_ncyl = cl->cl_blockcount / (64 * 32);
3245 			cl->cl_solaris_size = (diskaddr_t)cl->cl_g.dkg_ncyl *
3246 			    cl->cl_g.dkg_nhead * cl->cl_g.dkg_nsect;
3247 		} else {
3248 			cl->cl_g.dkg_ncyl  = 1;
3249 			cl->cl_g.dkg_nhead = 1;
3250 			cl->cl_g.dkg_nsect = cl->cl_blockcount;
3251 		}
3252 	} else {
3253 		if (cl->cl_blockcount <= 0x1000) {
3254 			/* unlabeled SCSI floppy device */
3255 			cl->cl_g.dkg_nhead = 2;
3256 			cl->cl_g.dkg_ncyl = 80;
3257 			cl->cl_g.dkg_nsect = cl->cl_blockcount / (2 * 80);
3258 		} else if (cl->cl_blockcount <= 0x200000) {
3259 			cl->cl_g.dkg_nhead = 64;
3260 			cl->cl_g.dkg_nsect = 32;
3261 			cl->cl_g.dkg_ncyl  = cl->cl_blockcount / (64 * 32);
3262 		} else {
3263 			cl->cl_g.dkg_nhead = 255;
3264 
3265 			cl->cl_g.dkg_nsect = ((cl->cl_blockcount +
3266 			    (UINT16_MAX * 255 * 63) - 1) /
3267 			    (UINT16_MAX * 255 * 63)) * 63;
3268 
3269 			if (cl->cl_g.dkg_nsect == 0)
3270 				cl->cl_g.dkg_nsect = (UINT16_MAX / 63) * 63;
3271 
3272 			cl->cl_g.dkg_ncyl = cl->cl_blockcount /
3273 			    (255 * cl->cl_g.dkg_nsect);
3274 		}
3275 
3276 		cl->cl_solaris_size =
3277 		    (diskaddr_t)cl->cl_g.dkg_ncyl * cl->cl_g.dkg_nhead *
3278 		    cl->cl_g.dkg_nsect;
3279 
3280 	}
3281 
3282 	cl->cl_g.dkg_acyl	= 0;
3283 	cl->cl_g.dkg_bcyl	= 0;
3284 	cl->cl_g.dkg_rpm	= 200;
3285 	cl->cl_asciilabel[0]	= '\0';
3286 	cl->cl_g.dkg_pcyl	= cl->cl_g.dkg_ncyl;
3287 
3288 	cl->cl_map[0].dkl_cylno = 0;
3289 	cl->cl_map[0].dkl_nblk  = cl->cl_solaris_size;
3290 
3291 	cl->cl_map[2].dkl_cylno = 0;
3292 	cl->cl_map[2].dkl_nblk  = cl->cl_solaris_size;
3293 
3294 #elif defined(_SUNOS_VTOC_16)
3295 
3296 	if (cl->cl_solaris_size == 0) {
3297 		/*
3298 		 * Got fdisk table but no solaris entry therefore
3299 		 * don't create a default label
3300 		 */
3301 		cl->cl_f_geometry_is_valid = B_TRUE;
3302 		return;
3303 	}
3304 
3305 	/*
3306 	 * For CDs we continue to use the physical geometry to calculate
3307 	 * number of cylinders. All other devices must convert the
3308 	 * physical geometry (cmlb_geom) to values that will fit
3309 	 * in a dk_geom structure.
3310 	 */
3311 	if (ISCD(cl)) {
3312 		phys_spc = cl->cl_pgeom.g_nhead * cl->cl_pgeom.g_nsect;
3313 	} else {
3314 		/* Convert physical geometry to disk geometry */
3315 		bzero(&cl_g, sizeof (struct dk_geom));
3316 
3317 		/*
3318 		 * Refer to comments related to off-by-1 at the
3319 		 * header of this file.
3320 		 * Before calculating geometry, capacity should be
3321 		 * decreased by 1.
3322 		 */
3323 
3324 		if (cl->cl_alter_behavior & CMLB_OFF_BY_ONE)
3325 			capacity = cl->cl_blockcount - 1;
3326 		else
3327 			capacity = cl->cl_blockcount;
3328 
3329 
3330 		cmlb_convert_geometry(capacity, &cl_g);
3331 		bcopy(&cl_g, &cl->cl_g, sizeof (cl->cl_g));
3332 		phys_spc = cl->cl_g.dkg_nhead * cl->cl_g.dkg_nsect;
3333 	}
3334 
3335 	if (phys_spc == 0)
3336 		return;
3337 	cl->cl_g.dkg_pcyl = cl->cl_solaris_size / phys_spc;
3338 	if (cl->cl_alter_behavior & CMLB_FAKE_LABEL_ONE_PARTITION) {
3339 		/* disable devid */
3340 		cl->cl_g.dkg_ncyl = cl->cl_g.dkg_pcyl;
3341 		disksize = cl->cl_solaris_size;
3342 	} else {
3343 		cl->cl_g.dkg_acyl = DK_ACYL;
3344 		cl->cl_g.dkg_ncyl = cl->cl_g.dkg_pcyl - DK_ACYL;
3345 		disksize = cl->cl_g.dkg_ncyl * phys_spc;
3346 	}
3347 
3348 	if (ISCD(cl)) {
3349 		/*
3350 		 * CD's don't use the "heads * sectors * cyls"-type of
3351 		 * geometry, but instead use the entire capacity of the media.
3352 		 */
3353 		disksize = cl->cl_solaris_size;
3354 		cl->cl_g.dkg_nhead = 1;
3355 		cl->cl_g.dkg_nsect = 1;
3356 		cl->cl_g.dkg_rpm =
3357 		    (cl->cl_pgeom.g_rpm == 0) ? 200 : cl->cl_pgeom.g_rpm;
3358 
3359 		cl->cl_vtoc.v_part[0].p_start = 0;
3360 		cl->cl_vtoc.v_part[0].p_size  = disksize;
3361 		cl->cl_vtoc.v_part[0].p_tag   = V_BACKUP;
3362 		cl->cl_vtoc.v_part[0].p_flag  = V_UNMNT;
3363 
3364 		cl->cl_map[0].dkl_cylno = 0;
3365 		cl->cl_map[0].dkl_nblk  = disksize;
3366 		cl->cl_offset[0] = 0;
3367 
3368 	} else {
3369 		/*
3370 		 * Hard disks and removable media cartridges
3371 		 */
3372 		cl->cl_g.dkg_rpm =
3373 		    (cl->cl_pgeom.g_rpm == 0) ? 3600: cl->cl_pgeom.g_rpm;
3374 		cl->cl_vtoc.v_sectorsz = cl->cl_sys_blocksize;
3375 
3376 		/* Add boot slice */
3377 		cl->cl_vtoc.v_part[8].p_start = 0;
3378 		cl->cl_vtoc.v_part[8].p_size  = phys_spc;
3379 		cl->cl_vtoc.v_part[8].p_tag   = V_BOOT;
3380 		cl->cl_vtoc.v_part[8].p_flag  = V_UNMNT;
3381 
3382 		cl->cl_map[8].dkl_cylno = 0;
3383 		cl->cl_map[8].dkl_nblk  = phys_spc;
3384 		cl->cl_offset[8] = 0;
3385 
3386 		if ((cl->cl_alter_behavior &
3387 		    CMLB_CREATE_ALTSLICE_VTOC_16_DTYPE_DIRECT) &&
3388 		    cl->cl_device_type == DTYPE_DIRECT) {
3389 			cl->cl_vtoc.v_part[9].p_start = phys_spc;
3390 			cl->cl_vtoc.v_part[9].p_size  = 2 * phys_spc;
3391 			cl->cl_vtoc.v_part[9].p_tag   = V_ALTSCTR;
3392 			cl->cl_vtoc.v_part[9].p_flag  = 0;
3393 
3394 			cl->cl_map[9].dkl_cylno = 1;
3395 			cl->cl_map[9].dkl_nblk  = 2 * phys_spc;
3396 			cl->cl_offset[9] = phys_spc;
3397 		}
3398 	}
3399 
3400 	cl->cl_g.dkg_apc = 0;
3401 
3402 	/* Add backup slice */
3403 	cl->cl_vtoc.v_part[2].p_start = 0;
3404 	cl->cl_vtoc.v_part[2].p_size  = disksize;
3405 	cl->cl_vtoc.v_part[2].p_tag   = V_BACKUP;
3406 	cl->cl_vtoc.v_part[2].p_flag  = V_UNMNT;
3407 
3408 	cl->cl_map[2].dkl_cylno = 0;
3409 	cl->cl_map[2].dkl_nblk  = disksize;
3410 	cl->cl_offset[2] = 0;
3411 
3412 	/*
3413 	 * single slice (s0) covering the entire disk
3414 	 */
3415 	if (cl->cl_alter_behavior & CMLB_FAKE_LABEL_ONE_PARTITION) {
3416 		cl->cl_vtoc.v_part[0].p_start = 0;
3417 		cl->cl_vtoc.v_part[0].p_tag   = V_UNASSIGNED;
3418 		cl->cl_vtoc.v_part[0].p_flag  = 0;
3419 		cl->cl_vtoc.v_part[0].p_size  = disksize;
3420 		cl->cl_map[0].dkl_cylno = 0;
3421 		cl->cl_map[0].dkl_nblk  = disksize;
3422 		cl->cl_offset[0] = 0;
3423 	}
3424 
3425 	(void) sprintf(cl->cl_vtoc.v_asciilabel, "DEFAULT cyl %d alt %d"
3426 	    " hd %d sec %d", cl->cl_g.dkg_ncyl, cl->cl_g.dkg_acyl,
3427 	    cl->cl_g.dkg_nhead, cl->cl_g.dkg_nsect);
3428 
3429 #else
3430 #error "No VTOC format defined."
3431 #endif
3432 
3433 	cl->cl_g.dkg_read_reinstruct  = 0;
3434 	cl->cl_g.dkg_write_reinstruct = 0;
3435 
3436 	cl->cl_g.dkg_intrlv = 1;
3437 
3438 	cl->cl_vtoc.v_sanity  = VTOC_SANE;
3439 	cl->cl_vtoc.v_nparts = V_NUMPAR;
3440 	cl->cl_vtoc.v_version = V_VERSION;
3441 
3442 	cl->cl_f_geometry_is_valid = B_TRUE;
3443 	cl->cl_label_from_media = CMLB_LABEL_UNDEF;
3444 
3445 	cmlb_dbg(CMLB_INFO,  cl,
3446 	    "cmlb_build_default_label: Default label created: "
3447 	    "cyl: %d\tacyl: %d\tnhead: %d\tnsect: %d\tcap: %d\n",
3448 	    cl->cl_g.dkg_ncyl, cl->cl_g.dkg_acyl, cl->cl_g.dkg_nhead,
3449 	    cl->cl_g.dkg_nsect, cl->cl_blockcount);
3450 }
3451 
3452 
3453 #if defined(_FIRMWARE_NEEDS_FDISK)
3454 /*
3455  * Max CHS values, as they are encoded into bytes, for 1022/254/63
3456  */
3457 #define	LBA_MAX_SECT	(63 | ((1022 & 0x300) >> 2))
3458 #define	LBA_MAX_CYL	(1022 & 0xFF)
3459 #define	LBA_MAX_HEAD	(254)
3460 
3461 
3462 /*
3463  *    Function: cmlb_has_max_chs_vals
3464  *
3465  * Description: Return B_TRUE if Cylinder-Head-Sector values are all at maximum.
3466  *
3467  *   Arguments: fdp - ptr to CHS info
3468  *
3469  * Return Code: True or false
3470  *
3471  *     Context: Any.
3472  */
3473 static boolean_t
3474 cmlb_has_max_chs_vals(struct ipart *fdp)
3475 {
3476 	return ((fdp->begcyl  == LBA_MAX_CYL)	&&
3477 	    (fdp->beghead == LBA_MAX_HEAD)	&&
3478 	    (fdp->begsect == LBA_MAX_SECT)	&&
3479 	    (fdp->endcyl  == LBA_MAX_CYL)	&&
3480 	    (fdp->endhead == LBA_MAX_HEAD)	&&
3481 	    (fdp->endsect == LBA_MAX_SECT));
3482 }
3483 #endif
3484 
3485 /*
3486  *    Function: cmlb_dkio_get_geometry
3487  *
3488  * Description: This routine is the driver entry point for handling user
3489  *		requests to get the device geometry (DKIOCGGEOM).
3490  *
3491  *   Arguments:
3492  *	arg		pointer to user provided dk_geom structure specifying
3493  *			the controller's notion of the current geometry.
3494  *
3495  *	flag 		this argument is a pass through to ddi_copyxxx()
3496  *			directly from the mode argument of ioctl().
3497  *
3498  *	tg_cookie	cookie from target driver to be passed back to target
3499  *			driver when we call back to it through tg_ops.
3500  *
3501  * Return Code: 0
3502  *		EFAULT
3503  *		ENXIO
3504  *		EIO
3505  */
3506 static int
3507 cmlb_dkio_get_geometry(struct cmlb_lun *cl, caddr_t arg, int flag,
3508     void *tg_cookie)
3509 {
3510 	struct dk_geom	*tmp_geom = NULL;
3511 	int		rval = 0;
3512 
3513 	/*
3514 	 * cmlb_validate_geometry does not spin a disk up
3515 	 * if it was spcl down. We need to make sure it
3516 	 * is ready.
3517 	 */
3518 	mutex_enter(CMLB_MUTEX(cl));
3519 	rval = cmlb_validate_geometry(cl, B_TRUE, 0, tg_cookie);
3520 #if defined(_SUNOS_VTOC_8)
3521 	if (rval == EINVAL &&
3522 	    cl->cl_alter_behavior & CMLB_FAKE_GEOM_LABEL_IOCTLS_VTOC8) {
3523 		/*
3524 		 * This is to return a default label geometry even when we
3525 		 * do not really assume a default label for the device.
3526 		 * dad driver utilizes this.
3527 		 */
3528 		if (cl->cl_blockcount <= CMLB_OLDVTOC_LIMIT) {
3529 			cmlb_setup_default_geometry(cl, tg_cookie);
3530 			rval = 0;
3531 		}
3532 	}
3533 #endif
3534 	if (rval) {
3535 		mutex_exit(CMLB_MUTEX(cl));
3536 		return (rval);
3537 	}
3538 
3539 #if defined(__i386) || defined(__amd64)
3540 	if (cl->cl_solaris_size == 0) {
3541 		mutex_exit(CMLB_MUTEX(cl));
3542 		return (EIO);
3543 	}
3544 #endif
3545 
3546 	/*
3547 	 * Make a local copy of the soft state geometry to avoid some potential
3548 	 * race conditions associated with holding the mutex and updating the
3549 	 * write_reinstruct value
3550 	 */
3551 	tmp_geom = kmem_zalloc(sizeof (struct dk_geom), KM_SLEEP);
3552 	bcopy(&cl->cl_g, tmp_geom, sizeof (struct dk_geom));
3553 
3554 	if (tmp_geom->dkg_write_reinstruct == 0) {
3555 		tmp_geom->dkg_write_reinstruct =
3556 		    (int)((int)(tmp_geom->dkg_nsect * tmp_geom->dkg_rpm *
3557 		    cmlb_rot_delay) / (int)60000);
3558 	}
3559 	mutex_exit(CMLB_MUTEX(cl));
3560 
3561 	rval = ddi_copyout(tmp_geom, (void *)arg, sizeof (struct dk_geom),
3562 	    flag);
3563 	if (rval != 0) {
3564 		rval = EFAULT;
3565 	}
3566 
3567 	kmem_free(tmp_geom, sizeof (struct dk_geom));
3568 	return (rval);
3569 
3570 }
3571 
3572 
3573 /*
3574  *    Function: cmlb_dkio_set_geometry
3575  *
3576  * Description: This routine is the driver entry point for handling user
3577  *		requests to set the device geometry (DKIOCSGEOM). The actual
3578  *		device geometry is not updated, just the driver "notion" of it.
3579  *
3580  *   Arguments:
3581  *	arg		pointer to user provided dk_geom structure used to set
3582  *			the controller's notion of the current geometry.
3583  *
3584  *	flag 		this argument is a pass through to ddi_copyxxx()
3585  *			directly from the mode argument of ioctl().
3586  *
3587  *	tg_cookie	cookie from target driver to be passed back to target
3588  *			driver when we call back to it through tg_ops.
3589  *
3590  * Return Code: 0
3591  *		EFAULT
3592  *		ENXIO
3593  *		EIO
3594  */
3595 static int
3596 cmlb_dkio_set_geometry(struct cmlb_lun *cl, caddr_t arg, int flag)
3597 {
3598 	struct dk_geom	*tmp_geom;
3599 	struct dk_map	*lp;
3600 	int		rval = 0;
3601 	int		i;
3602 
3603 
3604 #if defined(__i386) || defined(__amd64)
3605 	if (cl->cl_solaris_size == 0) {
3606 		return (EIO);
3607 	}
3608 #endif
3609 	/*
3610 	 * We need to copy the user specified geometry into local
3611 	 * storage and then update the softstate. We don't want to hold
3612 	 * the mutex and copyin directly from the user to the soft state
3613 	 */
3614 	tmp_geom = (struct dk_geom *)
3615 	    kmem_zalloc(sizeof (struct dk_geom), KM_SLEEP);
3616 	rval = ddi_copyin(arg, tmp_geom, sizeof (struct dk_geom), flag);
3617 	if (rval != 0) {
3618 		kmem_free(tmp_geom, sizeof (struct dk_geom));
3619 		return (EFAULT);
3620 	}
3621 
3622 	mutex_enter(CMLB_MUTEX(cl));
3623 	bcopy(tmp_geom, &cl->cl_g, sizeof (struct dk_geom));
3624 	for (i = 0; i < NDKMAP; i++) {
3625 		lp  = &cl->cl_map[i];
3626 		cl->cl_offset[i] =
3627 		    cl->cl_g.dkg_nhead * cl->cl_g.dkg_nsect * lp->dkl_cylno;
3628 #if defined(__i386) || defined(__amd64)
3629 		cl->cl_offset[i] += cl->cl_solaris_offset;
3630 #endif
3631 	}
3632 	cl->cl_f_geometry_is_valid = B_FALSE;
3633 	mutex_exit(CMLB_MUTEX(cl));
3634 	kmem_free(tmp_geom, sizeof (struct dk_geom));
3635 
3636 	return (rval);
3637 }
3638 
3639 /*
3640  *    Function: cmlb_dkio_get_partition
3641  *
3642  * Description: This routine is the driver entry point for handling user
3643  *		requests to get the partition table (DKIOCGAPART).
3644  *
3645  *   Arguments:
3646  *	arg		pointer to user provided dk_allmap structure specifying
3647  *			the controller's notion of the current partition table.
3648  *
3649  *	flag		this argument is a pass through to ddi_copyxxx()
3650  *			directly from the mode argument of ioctl().
3651  *
3652  *	tg_cookie	cookie from target driver to be passed back to target
3653  *			driver when we call back to it through tg_ops.
3654  *
3655  * Return Code: 0
3656  *		EFAULT
3657  *		ENXIO
3658  *		EIO
3659  */
3660 static int
3661 cmlb_dkio_get_partition(struct cmlb_lun *cl, caddr_t arg, int flag,
3662     void *tg_cookie)
3663 {
3664 	int		rval = 0;
3665 	int		size;
3666 
3667 	/*
3668 	 * Make sure the geometry is valid before getting the partition
3669 	 * information.
3670 	 */
3671 	mutex_enter(CMLB_MUTEX(cl));
3672 	if ((rval = cmlb_validate_geometry(cl, B_TRUE, 0, tg_cookie)) != 0) {
3673 		mutex_exit(CMLB_MUTEX(cl));
3674 		return (rval);
3675 	}
3676 	mutex_exit(CMLB_MUTEX(cl));
3677 
3678 #if defined(__i386) || defined(__amd64)
3679 	if (cl->cl_solaris_size == 0) {
3680 		return (EIO);
3681 	}
3682 #endif
3683 
3684 #ifdef _MULTI_DATAMODEL
3685 	switch (ddi_model_convert_from(flag & FMODELS)) {
3686 	case DDI_MODEL_ILP32: {
3687 		struct dk_map32 dk_map32[NDKMAP];
3688 		int		i;
3689 
3690 		for (i = 0; i < NDKMAP; i++) {
3691 			dk_map32[i].dkl_cylno = cl->cl_map[i].dkl_cylno;
3692 			dk_map32[i].dkl_nblk  = cl->cl_map[i].dkl_nblk;
3693 		}
3694 		size = NDKMAP * sizeof (struct dk_map32);
3695 		rval = ddi_copyout(dk_map32, (void *)arg, size, flag);
3696 		if (rval != 0) {
3697 			rval = EFAULT;
3698 		}
3699 		break;
3700 	}
3701 	case DDI_MODEL_NONE:
3702 		size = NDKMAP * sizeof (struct dk_map);
3703 		rval = ddi_copyout(cl->cl_map, (void *)arg, size, flag);
3704 		if (rval != 0) {
3705 			rval = EFAULT;
3706 		}
3707 		break;
3708 	}
3709 #else /* ! _MULTI_DATAMODEL */
3710 	size = NDKMAP * sizeof (struct dk_map);
3711 	rval = ddi_copyout(cl->cl_map, (void *)arg, size, flag);
3712 	if (rval != 0) {
3713 		rval = EFAULT;
3714 	}
3715 #endif /* _MULTI_DATAMODEL */
3716 	return (rval);
3717 }
3718 
3719 /*
3720  *    Function: cmlb_dkio_set_partition
3721  *
3722  * Description: This routine is the driver entry point for handling user
3723  *		requests to set the partition table (DKIOCSAPART). The actual
3724  *		device partition is not updated.
3725  *
3726  *   Arguments:
3727  *		arg  - pointer to user provided dk_allmap structure used to set
3728  *			the controller's notion of the partition table.
3729  *		flag - this argument is a pass through to ddi_copyxxx()
3730  *		       directly from the mode argument of ioctl().
3731  *
3732  * Return Code: 0
3733  *		EINVAL
3734  *		EFAULT
3735  *		ENXIO
3736  *		EIO
3737  */
3738 static int
3739 cmlb_dkio_set_partition(struct cmlb_lun *cl, caddr_t arg, int flag)
3740 {
3741 	struct dk_map	dk_map[NDKMAP];
3742 	struct dk_map	*lp;
3743 	int		rval = 0;
3744 	int		size;
3745 	int		i;
3746 #if defined(_SUNOS_VTOC_16)
3747 	struct dkl_partition	*vp;
3748 #endif
3749 
3750 	/*
3751 	 * Set the map for all logical partitions.  We lock
3752 	 * the priority just to make sure an interrupt doesn't
3753 	 * come in while the map is half updated.
3754 	 */
3755 	_NOTE(DATA_READABLE_WITHOUT_LOCK(cmlb_lun::cl_solaris_size))
3756 	mutex_enter(CMLB_MUTEX(cl));
3757 
3758 	if (cl->cl_blockcount > CMLB_OLDVTOC_LIMIT) {
3759 		mutex_exit(CMLB_MUTEX(cl));
3760 		return (ENOTSUP);
3761 	}
3762 	mutex_exit(CMLB_MUTEX(cl));
3763 	if (cl->cl_solaris_size == 0) {
3764 		return (EIO);
3765 	}
3766 
3767 #ifdef _MULTI_DATAMODEL
3768 	switch (ddi_model_convert_from(flag & FMODELS)) {
3769 	case DDI_MODEL_ILP32: {
3770 		struct dk_map32 dk_map32[NDKMAP];
3771 
3772 		size = NDKMAP * sizeof (struct dk_map32);
3773 		rval = ddi_copyin((void *)arg, dk_map32, size, flag);
3774 		if (rval != 0) {
3775 			return (EFAULT);
3776 		}
3777 		for (i = 0; i < NDKMAP; i++) {
3778 			dk_map[i].dkl_cylno = dk_map32[i].dkl_cylno;
3779 			dk_map[i].dkl_nblk  = dk_map32[i].dkl_nblk;
3780 		}
3781 		break;
3782 	}
3783 	case DDI_MODEL_NONE:
3784 		size = NDKMAP * sizeof (struct dk_map);
3785 		rval = ddi_copyin((void *)arg, dk_map, size, flag);
3786 		if (rval != 0) {
3787 			return (EFAULT);
3788 		}
3789 		break;
3790 	}
3791 #else /* ! _MULTI_DATAMODEL */
3792 	size = NDKMAP * sizeof (struct dk_map);
3793 	rval = ddi_copyin((void *)arg, dk_map, size, flag);
3794 	if (rval != 0) {
3795 		return (EFAULT);
3796 	}
3797 #endif /* _MULTI_DATAMODEL */
3798 
3799 	mutex_enter(CMLB_MUTEX(cl));
3800 	/* Note: The size used in this bcopy is set based upon the data model */
3801 	bcopy(dk_map, cl->cl_map, size);
3802 #if defined(_SUNOS_VTOC_16)
3803 	vp = (struct dkl_partition *)&(cl->cl_vtoc);
3804 #endif	/* defined(_SUNOS_VTOC_16) */
3805 	for (i = 0; i < NDKMAP; i++) {
3806 		lp  = &cl->cl_map[i];
3807 		cl->cl_offset[i] =
3808 		    cl->cl_g.dkg_nhead * cl->cl_g.dkg_nsect * lp->dkl_cylno;
3809 #if defined(_SUNOS_VTOC_16)
3810 		vp->p_start = cl->cl_offset[i];
3811 		vp->p_size = lp->dkl_nblk;
3812 		vp++;
3813 #endif	/* defined(_SUNOS_VTOC_16) */
3814 #if defined(__i386) || defined(__amd64)
3815 		cl->cl_offset[i] += cl->cl_solaris_offset;
3816 #endif
3817 	}
3818 	mutex_exit(CMLB_MUTEX(cl));
3819 	return (rval);
3820 }
3821 
3822 
3823 /*
3824  *    Function: cmlb_dkio_get_vtoc
3825  *
3826  * Description: This routine is the driver entry point for handling user
3827  *		requests to get the current volume table of contents
3828  *		(DKIOCGVTOC).
3829  *
3830  *   Arguments:
3831  *	arg		pointer to user provided vtoc structure specifying
3832  *			the current vtoc.
3833  *
3834  *	flag		this argument is a pass through to ddi_copyxxx()
3835  *			directly from the mode argument of ioctl().
3836  *
3837  *	tg_cookie	cookie from target driver to be passed back to target
3838  *			driver when we call back to it through tg_ops.
3839  *
3840  * Return Code: 0
3841  *		EFAULT
3842  *		ENXIO
3843  *		EIO
3844  */
3845 static int
3846 cmlb_dkio_get_vtoc(struct cmlb_lun *cl, caddr_t arg, int flag, void *tg_cookie)
3847 {
3848 #if defined(_SUNOS_VTOC_8)
3849 	struct vtoc	user_vtoc;
3850 #endif	/* defined(_SUNOS_VTOC_8) */
3851 	int		rval = 0;
3852 
3853 	mutex_enter(CMLB_MUTEX(cl));
3854 	if (cl->cl_blockcount > CMLB_OLDVTOC_LIMIT) {
3855 		mutex_exit(CMLB_MUTEX(cl));
3856 		return (EOVERFLOW);
3857 	}
3858 
3859 	rval = cmlb_validate_geometry(cl, B_TRUE, 0, tg_cookie);
3860 
3861 #if defined(_SUNOS_VTOC_8)
3862 	if (rval == EINVAL &&
3863 	    (cl->cl_alter_behavior & CMLB_FAKE_GEOM_LABEL_IOCTLS_VTOC8)) {
3864 		/*
3865 		 * This is to return a default label even when we do not
3866 		 * really assume a default label for the device.
3867 		 * dad driver utilizes this.
3868 		 */
3869 		if (cl->cl_blockcount <= CMLB_OLDVTOC_LIMIT) {
3870 			cmlb_setup_default_geometry(cl, tg_cookie);
3871 			rval = 0;
3872 		}
3873 	}
3874 #endif
3875 	if (rval) {
3876 		mutex_exit(CMLB_MUTEX(cl));
3877 		return (rval);
3878 	}
3879 
3880 #if defined(_SUNOS_VTOC_8)
3881 	cmlb_build_user_vtoc(cl, &user_vtoc);
3882 	mutex_exit(CMLB_MUTEX(cl));
3883 
3884 #ifdef _MULTI_DATAMODEL
3885 	switch (ddi_model_convert_from(flag & FMODELS)) {
3886 	case DDI_MODEL_ILP32: {
3887 		struct vtoc32 user_vtoc32;
3888 
3889 		vtoctovtoc32(user_vtoc, user_vtoc32);
3890 		if (ddi_copyout(&user_vtoc32, (void *)arg,
3891 		    sizeof (struct vtoc32), flag)) {
3892 			return (EFAULT);
3893 		}
3894 		break;
3895 	}
3896 
3897 	case DDI_MODEL_NONE:
3898 		if (ddi_copyout(&user_vtoc, (void *)arg,
3899 		    sizeof (struct vtoc), flag)) {
3900 			return (EFAULT);
3901 		}
3902 		break;
3903 	}
3904 #else /* ! _MULTI_DATAMODEL */
3905 	if (ddi_copyout(&user_vtoc, (void *)arg, sizeof (struct vtoc), flag)) {
3906 		return (EFAULT);
3907 	}
3908 #endif /* _MULTI_DATAMODEL */
3909 
3910 #elif defined(_SUNOS_VTOC_16)
3911 	mutex_exit(CMLB_MUTEX(cl));
3912 
3913 #ifdef _MULTI_DATAMODEL
3914 	/*
3915 	 * The cl_vtoc structure is a "struct dk_vtoc"  which is always
3916 	 * 32-bit to maintain compatibility with existing on-disk
3917 	 * structures.  Thus, we need to convert the structure when copying
3918 	 * it out to a datamodel-dependent "struct vtoc" in a 64-bit
3919 	 * program.  If the target is a 32-bit program, then no conversion
3920 	 * is necessary.
3921 	 */
3922 	/* LINTED: logical expression always true: op "||" */
3923 	ASSERT(sizeof (cl->cl_vtoc) == sizeof (struct vtoc32));
3924 	switch (ddi_model_convert_from(flag & FMODELS)) {
3925 	case DDI_MODEL_ILP32:
3926 		if (ddi_copyout(&(cl->cl_vtoc), (void *)arg,
3927 		    sizeof (cl->cl_vtoc), flag)) {
3928 			return (EFAULT);
3929 		}
3930 		break;
3931 
3932 	case DDI_MODEL_NONE: {
3933 		struct vtoc user_vtoc;
3934 
3935 		vtoc32tovtoc(cl->cl_vtoc, user_vtoc);
3936 		if (ddi_copyout(&user_vtoc, (void *)arg,
3937 		    sizeof (struct vtoc), flag)) {
3938 			return (EFAULT);
3939 		}
3940 		break;
3941 	}
3942 	}
3943 #else /* ! _MULTI_DATAMODEL */
3944 	if (ddi_copyout(&(cl->cl_vtoc), (void *)arg, sizeof (cl->cl_vtoc),
3945 	    flag)) {
3946 		return (EFAULT);
3947 	}
3948 #endif /* _MULTI_DATAMODEL */
3949 #else
3950 #error "No VTOC format defined."
3951 #endif
3952 
3953 	return (rval);
3954 }
3955 
3956 
3957 /*
3958  *    Function: cmlb_dkio_get_extvtoc
3959  */
3960 static int
3961 cmlb_dkio_get_extvtoc(struct cmlb_lun *cl, caddr_t arg, int flag,
3962     void *tg_cookie)
3963 {
3964 	struct extvtoc	ext_vtoc;
3965 #if defined(_SUNOS_VTOC_8)
3966 	struct vtoc	user_vtoc;
3967 #endif	/* defined(_SUNOS_VTOC_8) */
3968 	int		rval = 0;
3969 
3970 	bzero(&ext_vtoc, sizeof (struct extvtoc));
3971 	mutex_enter(CMLB_MUTEX(cl));
3972 	rval = cmlb_validate_geometry(cl, B_TRUE, 0, tg_cookie);
3973 
3974 #if defined(_SUNOS_VTOC_8)
3975 	if (rval == EINVAL &&
3976 	    (cl->cl_alter_behavior & CMLB_FAKE_GEOM_LABEL_IOCTLS_VTOC8)) {
3977 		/*
3978 		 * This is to return a default label even when we do not
3979 		 * really assume a default label for the device.
3980 		 * dad driver utilizes this.
3981 		 */
3982 		if (cl->cl_blockcount <= CMLB_OLDVTOC_LIMIT) {
3983 			cmlb_setup_default_geometry(cl, tg_cookie);
3984 			rval = 0;
3985 		}
3986 	}
3987 #endif
3988 	if (rval) {
3989 		mutex_exit(CMLB_MUTEX(cl));
3990 		return (rval);
3991 	}
3992 
3993 #if defined(_SUNOS_VTOC_8)
3994 	cmlb_build_user_vtoc(cl, &user_vtoc);
3995 	mutex_exit(CMLB_MUTEX(cl));
3996 
3997 	/*
3998 	 * Checking callers data model does not make much sense here
3999 	 * since extvtoc will always be equivalent to 64bit vtoc.
4000 	 * What is important is whether the kernel is in 32 or 64 bit
4001 	 */
4002 
4003 #ifdef _LP64
4004 		if (ddi_copyout(&user_vtoc, (void *)arg,
4005 		    sizeof (struct extvtoc), flag)) {
4006 			return (EFAULT);
4007 		}
4008 #else
4009 		vtoc32tovtoc(user_vtoc, ext_vtoc);
4010 		if (ddi_copyout(&ext_vtoc, (void *)arg,
4011 		    sizeof (struct extvtoc), flag)) {
4012 			return (EFAULT);
4013 		}
4014 #endif
4015 
4016 #elif defined(_SUNOS_VTOC_16)
4017 	/*
4018 	 * The cl_vtoc structure is a "struct dk_vtoc"  which is always
4019 	 * 32-bit to maintain compatibility with existing on-disk
4020 	 * structures.  Thus, we need to convert the structure when copying
4021 	 * it out to extvtoc
4022 	 */
4023 	vtoc32tovtoc(cl->cl_vtoc, ext_vtoc);
4024 	mutex_exit(CMLB_MUTEX(cl));
4025 
4026 	if (ddi_copyout(&ext_vtoc, (void *)arg, sizeof (struct extvtoc), flag))
4027 		return (EFAULT);
4028 #else
4029 #error "No VTOC format defined."
4030 #endif
4031 
4032 	return (rval);
4033 }
4034 static int
4035 cmlb_dkio_get_efi(struct cmlb_lun *cl, caddr_t arg, int flag, void *tg_cookie)
4036 {
4037 	dk_efi_t	user_efi;
4038 	int		rval = 0;
4039 	void		*buffer;
4040 	diskaddr_t	tgt_lba;
4041 
4042 	if (ddi_copyin(arg, &user_efi, sizeof (dk_efi_t), flag))
4043 		return (EFAULT);
4044 
4045 	user_efi.dki_data = (void *)(uintptr_t)user_efi.dki_data_64;
4046 
4047 	tgt_lba = user_efi.dki_lba;
4048 
4049 	mutex_enter(CMLB_MUTEX(cl));
4050 	if ((cmlb_check_update_blockcount(cl, tg_cookie) != 0) ||
4051 	    (cl->cl_tgt_blocksize == 0)) {
4052 		mutex_exit(CMLB_MUTEX(cl));
4053 		return (EINVAL);
4054 	}
4055 	if (cl->cl_tgt_blocksize != cl->cl_sys_blocksize)
4056 		tgt_lba = tgt_lba * cl->cl_tgt_blocksize /
4057 		    cl->cl_sys_blocksize;
4058 	mutex_exit(CMLB_MUTEX(cl));
4059 
4060 	buffer = kmem_alloc(user_efi.dki_length, KM_SLEEP);
4061 	rval = DK_TG_READ(cl, buffer, tgt_lba, user_efi.dki_length, tg_cookie);
4062 	if (rval == 0 && ddi_copyout(buffer, user_efi.dki_data,
4063 	    user_efi.dki_length, flag) != 0)
4064 		rval = EFAULT;
4065 
4066 	kmem_free(buffer, user_efi.dki_length);
4067 	return (rval);
4068 }
4069 
4070 #if defined(_SUNOS_VTOC_8)
4071 /*
4072  *    Function: cmlb_build_user_vtoc
4073  *
4074  * Description: This routine populates a pass by reference variable with the
4075  *		current volume table of contents.
4076  *
4077  *   Arguments: cl - driver soft state (unit) structure
4078  *		user_vtoc - pointer to vtoc structure to be populated
4079  */
4080 static void
4081 cmlb_build_user_vtoc(struct cmlb_lun *cl, struct vtoc *user_vtoc)
4082 {
4083 	struct dk_map2		*lpart;
4084 	struct dk_map		*lmap;
4085 	struct partition	*vpart;
4086 	uint32_t		nblks;
4087 	int			i;
4088 
4089 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
4090 
4091 	/*
4092 	 * Return vtoc structure fields in the provided VTOC area, addressed
4093 	 * by *vtoc.
4094 	 */
4095 	bzero(user_vtoc, sizeof (struct vtoc));
4096 	user_vtoc->v_bootinfo[0] = cl->cl_vtoc.v_bootinfo[0];
4097 	user_vtoc->v_bootinfo[1] = cl->cl_vtoc.v_bootinfo[1];
4098 	user_vtoc->v_bootinfo[2] = cl->cl_vtoc.v_bootinfo[2];
4099 	user_vtoc->v_sanity	= VTOC_SANE;
4100 	user_vtoc->v_version	= cl->cl_vtoc.v_version;
4101 	bcopy(cl->cl_vtoc.v_volume, user_vtoc->v_volume, LEN_DKL_VVOL);
4102 	user_vtoc->v_sectorsz = cl->cl_sys_blocksize;
4103 	user_vtoc->v_nparts = cl->cl_vtoc.v_nparts;
4104 
4105 	for (i = 0; i < 10; i++)
4106 		user_vtoc->v_reserved[i] = cl->cl_vtoc.v_reserved[i];
4107 
4108 	/*
4109 	 * Convert partitioning information.
4110 	 *
4111 	 * Note the conversion from starting cylinder number
4112 	 * to starting sector number.
4113 	 */
4114 	lmap = cl->cl_map;
4115 	lpart = (struct dk_map2 *)cl->cl_vtoc.v_part;
4116 	vpart = user_vtoc->v_part;
4117 
4118 	nblks = cl->cl_g.dkg_nsect * cl->cl_g.dkg_nhead;
4119 
4120 	for (i = 0; i < V_NUMPAR; i++) {
4121 		vpart->p_tag	= lpart->p_tag;
4122 		vpart->p_flag	= lpart->p_flag;
4123 		vpart->p_start	= lmap->dkl_cylno * nblks;
4124 		vpart->p_size	= lmap->dkl_nblk;
4125 		lmap++;
4126 		lpart++;
4127 		vpart++;
4128 
4129 		/* (4364927) */
4130 		user_vtoc->timestamp[i] = (time_t)cl->cl_vtoc.v_timestamp[i];
4131 	}
4132 
4133 	bcopy(cl->cl_asciilabel, user_vtoc->v_asciilabel, LEN_DKL_ASCII);
4134 }
4135 #endif
4136 
4137 static int
4138 cmlb_dkio_partition(struct cmlb_lun *cl, caddr_t arg, int flag,
4139     void *tg_cookie)
4140 {
4141 	struct partition64	p64;
4142 	int			rval = 0;
4143 	uint_t			nparts;
4144 	efi_gpe_t		*partitions;
4145 	efi_gpt_t		*buffer;
4146 	diskaddr_t		gpe_lba;
4147 
4148 	if (ddi_copyin((const void *)arg, &p64,
4149 	    sizeof (struct partition64), flag)) {
4150 		return (EFAULT);
4151 	}
4152 
4153 	buffer = kmem_alloc(EFI_MIN_ARRAY_SIZE, KM_SLEEP);
4154 	rval = DK_TG_READ(cl, buffer, 1, cl->cl_sys_blocksize, tg_cookie);
4155 	if (rval != 0)
4156 		goto done_error;
4157 
4158 	cmlb_swap_efi_gpt(buffer);
4159 
4160 	if ((rval = cmlb_validate_efi(buffer)) != 0)
4161 		goto done_error;
4162 
4163 	nparts = buffer->efi_gpt_NumberOfPartitionEntries;
4164 	gpe_lba = buffer->efi_gpt_PartitionEntryLBA;
4165 	if (p64.p_partno > nparts) {
4166 		/* couldn't find it */
4167 		rval = ESRCH;
4168 		goto done_error;
4169 	}
4170 	/*
4171 	 * if we're dealing with a partition that's out of the normal
4172 	 * 16K block, adjust accordingly
4173 	 */
4174 	gpe_lba += p64.p_partno / sizeof (efi_gpe_t);
4175 	rval = DK_TG_READ(cl, buffer, gpe_lba, EFI_MIN_ARRAY_SIZE, tg_cookie);
4176 
4177 	if (rval) {
4178 		goto done_error;
4179 	}
4180 	partitions = (efi_gpe_t *)buffer;
4181 
4182 	cmlb_swap_efi_gpe(nparts, partitions);
4183 
4184 	partitions += p64.p_partno;
4185 	bcopy(&partitions->efi_gpe_PartitionTypeGUID, &p64.p_type,
4186 	    sizeof (struct uuid));
4187 	p64.p_start = partitions->efi_gpe_StartingLBA;
4188 	p64.p_size = partitions->efi_gpe_EndingLBA -
4189 	    p64.p_start + 1;
4190 
4191 	if (ddi_copyout(&p64, (void *)arg, sizeof (struct partition64), flag))
4192 		rval = EFAULT;
4193 
4194 done_error:
4195 	kmem_free(buffer, EFI_MIN_ARRAY_SIZE);
4196 	return (rval);
4197 }
4198 
4199 
4200 /*
4201  *    Function: cmlb_dkio_set_vtoc
4202  *
4203  * Description: This routine is the driver entry point for handling user
4204  *		requests to set the current volume table of contents
4205  *		(DKIOCSVTOC).
4206  *
4207  *   Arguments:
4208  *	dev		the device number
4209  *	arg		pointer to user provided vtoc structure used to set the
4210  *			current vtoc.
4211  *
4212  *	flag		this argument is a pass through to ddi_copyxxx()
4213  *			directly from the mode argument of ioctl().
4214  *
4215  *	tg_cookie	cookie from target driver to be passed back to target
4216  *			driver when we call back to it through tg_ops.
4217  *
4218  * Return Code: 0
4219  *		EFAULT
4220  *		ENXIO
4221  *		EINVAL
4222  *		ENOTSUP
4223  */
4224 static int
4225 cmlb_dkio_set_vtoc(struct cmlb_lun *cl, dev_t dev, caddr_t arg, int flag,
4226     void *tg_cookie)
4227 {
4228 	struct vtoc	user_vtoc;
4229 	int		rval = 0;
4230 	boolean_t	internal;
4231 
4232 	internal = VOID2BOOLEAN(
4233 	    (cl->cl_alter_behavior & (CMLB_INTERNAL_MINOR_NODES)) != 0);
4234 
4235 #ifdef _MULTI_DATAMODEL
4236 	switch (ddi_model_convert_from(flag & FMODELS)) {
4237 	case DDI_MODEL_ILP32: {
4238 		struct vtoc32 user_vtoc32;
4239 
4240 		if (ddi_copyin((const void *)arg, &user_vtoc32,
4241 		    sizeof (struct vtoc32), flag)) {
4242 			return (EFAULT);
4243 		}
4244 		vtoc32tovtoc(user_vtoc32, user_vtoc);
4245 		break;
4246 	}
4247 
4248 	case DDI_MODEL_NONE:
4249 		if (ddi_copyin((const void *)arg, &user_vtoc,
4250 		    sizeof (struct vtoc), flag)) {
4251 			return (EFAULT);
4252 		}
4253 		break;
4254 	}
4255 #else /* ! _MULTI_DATAMODEL */
4256 	if (ddi_copyin((const void *)arg, &user_vtoc,
4257 	    sizeof (struct vtoc), flag)) {
4258 		return (EFAULT);
4259 	}
4260 #endif /* _MULTI_DATAMODEL */
4261 
4262 	mutex_enter(CMLB_MUTEX(cl));
4263 
4264 	if (cl->cl_blockcount > CMLB_OLDVTOC_LIMIT) {
4265 		mutex_exit(CMLB_MUTEX(cl));
4266 		return (EOVERFLOW);
4267 	}
4268 
4269 #if defined(__i386) || defined(__amd64)
4270 	if (cl->cl_tgt_blocksize != cl->cl_sys_blocksize) {
4271 		mutex_exit(CMLB_MUTEX(cl));
4272 		return (EINVAL);
4273 	}
4274 #endif
4275 
4276 	if (cl->cl_g.dkg_ncyl == 0) {
4277 		mutex_exit(CMLB_MUTEX(cl));
4278 		return (EINVAL);
4279 	}
4280 
4281 	mutex_exit(CMLB_MUTEX(cl));
4282 	cmlb_clear_efi(cl, tg_cookie);
4283 	ddi_remove_minor_node(CMLB_DEVINFO(cl), "wd");
4284 	ddi_remove_minor_node(CMLB_DEVINFO(cl), "wd,raw");
4285 
4286 	/*
4287 	 * cmlb_dkio_set_vtoc creates duplicate minor nodes when
4288 	 * relabeling an SMI disk. To avoid that we remove them
4289 	 * before creating.
4290 	 * It should be OK to remove a non-existed minor node.
4291 	 */
4292 	ddi_remove_minor_node(CMLB_DEVINFO(cl), "h");
4293 	ddi_remove_minor_node(CMLB_DEVINFO(cl), "h,raw");
4294 
4295 	(void) cmlb_create_minor(CMLB_DEVINFO(cl), "h",
4296 	    S_IFBLK, (CMLBUNIT(dev) << CMLBUNIT_SHIFT) | WD_NODE,
4297 	    cl->cl_node_type, NULL, internal);
4298 	(void) cmlb_create_minor(CMLB_DEVINFO(cl), "h,raw",
4299 	    S_IFCHR, (CMLBUNIT(dev) << CMLBUNIT_SHIFT) | WD_NODE,
4300 	    cl->cl_node_type, NULL, internal);
4301 	mutex_enter(CMLB_MUTEX(cl));
4302 
4303 	if ((rval = cmlb_build_label_vtoc(cl, &user_vtoc)) == 0) {
4304 		if ((rval = cmlb_write_label(cl, tg_cookie)) == 0) {
4305 			if (cmlb_validate_geometry(cl,
4306 			    B_TRUE, 0, tg_cookie) != 0) {
4307 				cmlb_dbg(CMLB_ERROR, cl,
4308 				    "cmlb_dkio_set_vtoc: "
4309 				    "Failed validate geometry\n");
4310 			}
4311 			cl->cl_msglog_flag |= CMLB_ALLOW_2TB_WARN;
4312 		}
4313 	}
4314 	mutex_exit(CMLB_MUTEX(cl));
4315 	return (rval);
4316 }
4317 
4318 /*
4319  *    Function: cmlb_dkio_set_extvtoc
4320  */
4321 static int
4322 cmlb_dkio_set_extvtoc(struct cmlb_lun *cl, dev_t dev, caddr_t arg, int flag,
4323     void *tg_cookie)
4324 {
4325 	int		rval = 0;
4326 	struct vtoc	user_vtoc;
4327 	boolean_t	internal;
4328 
4329 
4330 	/*
4331 	 * Checking callers data model does not make much sense here
4332 	 * since extvtoc will always be equivalent to 64bit vtoc.
4333 	 * What is important is whether the kernel is in 32 or 64 bit
4334 	 */
4335 
4336 #ifdef _LP64
4337 	if (ddi_copyin((const void *)arg, &user_vtoc,
4338 		    sizeof (struct extvtoc), flag)) {
4339 			return (EFAULT);
4340 	}
4341 #else
4342 	struct	extvtoc	user_extvtoc;
4343 	if (ddi_copyin((const void *)arg, &user_extvtoc,
4344 		    sizeof (struct extvtoc), flag)) {
4345 			return (EFAULT);
4346 	}
4347 
4348 	vtoctovtoc32(user_extvtoc, user_vtoc);
4349 #endif
4350 
4351 	internal = VOID2BOOLEAN(
4352 	    (cl->cl_alter_behavior & (CMLB_INTERNAL_MINOR_NODES)) != 0);
4353 	mutex_enter(CMLB_MUTEX(cl));
4354 #if defined(__i386) || defined(__amd64)
4355 	if (cl->cl_tgt_blocksize != cl->cl_sys_blocksize) {
4356 		mutex_exit(CMLB_MUTEX(cl));
4357 		return (EINVAL);
4358 	}
4359 #endif
4360 
4361 	if (cl->cl_g.dkg_ncyl == 0) {
4362 		mutex_exit(CMLB_MUTEX(cl));
4363 		return (EINVAL);
4364 	}
4365 
4366 	mutex_exit(CMLB_MUTEX(cl));
4367 	cmlb_clear_efi(cl, tg_cookie);
4368 	ddi_remove_minor_node(CMLB_DEVINFO(cl), "wd");
4369 	ddi_remove_minor_node(CMLB_DEVINFO(cl), "wd,raw");
4370 	(void) cmlb_create_minor(CMLB_DEVINFO(cl), "h",
4371 	    S_IFBLK, (CMLBUNIT(dev) << CMLBUNIT_SHIFT) | WD_NODE,
4372 	    cl->cl_node_type, NULL, internal);
4373 	(void) cmlb_create_minor(CMLB_DEVINFO(cl), "h,raw",
4374 	    S_IFCHR, (CMLBUNIT(dev) << CMLBUNIT_SHIFT) | WD_NODE,
4375 	    cl->cl_node_type, NULL, internal);
4376 
4377 	mutex_enter(CMLB_MUTEX(cl));
4378 
4379 	if ((rval = cmlb_build_label_vtoc(cl, &user_vtoc)) == 0) {
4380 		if ((rval = cmlb_write_label(cl, tg_cookie)) == 0) {
4381 			if (cmlb_validate_geometry(cl,
4382 			    B_TRUE, 0, tg_cookie) != 0) {
4383 				cmlb_dbg(CMLB_ERROR, cl,
4384 				    "cmlb_dkio_set_vtoc: "
4385 				    "Failed validate geometry\n");
4386 			}
4387 		}
4388 	}
4389 	mutex_exit(CMLB_MUTEX(cl));
4390 	return (rval);
4391 }
4392 
4393 /*
4394  *    Function: cmlb_build_label_vtoc
4395  *
4396  * Description: This routine updates the driver soft state current volume table
4397  *		of contents based on a user specified vtoc.
4398  *
4399  *   Arguments: cl - driver soft state (unit) structure
4400  *		user_vtoc - pointer to vtoc structure specifying vtoc to be used
4401  *			    to update the driver soft state.
4402  *
4403  * Return Code: 0
4404  *		EINVAL
4405  */
4406 static int
4407 cmlb_build_label_vtoc(struct cmlb_lun *cl, struct vtoc *user_vtoc)
4408 {
4409 	struct dk_map		*lmap;
4410 	struct partition	*vpart;
4411 	uint_t			nblks;
4412 #if defined(_SUNOS_VTOC_8)
4413 	int			ncyl;
4414 	struct dk_map2		*lpart;
4415 #endif	/* defined(_SUNOS_VTOC_8) */
4416 	int			i;
4417 
4418 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
4419 
4420 	/* Sanity-check the vtoc */
4421 	if (user_vtoc->v_sanity != VTOC_SANE ||
4422 	    user_vtoc->v_sectorsz != cl->cl_sys_blocksize ||
4423 	    user_vtoc->v_nparts != V_NUMPAR) {
4424 		cmlb_dbg(CMLB_INFO,  cl,
4425 		    "cmlb_build_label_vtoc: vtoc not valid\n");
4426 		return (EINVAL);
4427 	}
4428 
4429 	nblks = cl->cl_g.dkg_nsect * cl->cl_g.dkg_nhead;
4430 	if (nblks == 0) {
4431 		cmlb_dbg(CMLB_INFO,  cl,
4432 		    "cmlb_build_label_vtoc: geom nblks is 0\n");
4433 		return (EINVAL);
4434 	}
4435 
4436 #if defined(_SUNOS_VTOC_8)
4437 	vpart = user_vtoc->v_part;
4438 	for (i = 0; i < V_NUMPAR; i++) {
4439 		if (((unsigned)vpart->p_start % nblks) != 0) {
4440 			cmlb_dbg(CMLB_INFO,  cl,
4441 			    "cmlb_build_label_vtoc: p_start not multiply of"
4442 			    "nblks part %d p_start %d nblks %d\n", i,
4443 			    vpart->p_start, nblks);
4444 			return (EINVAL);
4445 		}
4446 		ncyl = (unsigned)vpart->p_start / nblks;
4447 		ncyl += (unsigned)vpart->p_size / nblks;
4448 		if (((unsigned)vpart->p_size % nblks) != 0) {
4449 			ncyl++;
4450 		}
4451 		if (ncyl > (int)cl->cl_g.dkg_ncyl) {
4452 			cmlb_dbg(CMLB_INFO,  cl,
4453 			    "cmlb_build_label_vtoc: ncyl %d  > dkg_ncyl %d"
4454 			    "p_size %ld p_start %ld nblks %d  part number %d"
4455 			    "tag %d\n",
4456 			    ncyl, cl->cl_g.dkg_ncyl, vpart->p_size,
4457 			    vpart->p_start, nblks,
4458 			    i, vpart->p_tag);
4459 
4460 			return (EINVAL);
4461 		}
4462 		vpart++;
4463 	}
4464 #endif	/* defined(_SUNOS_VTOC_8) */
4465 
4466 	/* Put appropriate vtoc structure fields into the disk label */
4467 #if defined(_SUNOS_VTOC_16)
4468 	/*
4469 	 * The vtoc is always a 32bit data structure to maintain the
4470 	 * on-disk format. Convert "in place" instead of doing bcopy.
4471 	 */
4472 	vtoctovtoc32((*user_vtoc), (*((struct vtoc32 *)&(cl->cl_vtoc))));
4473 
4474 	/*
4475 	 * in the 16-slice vtoc, starting sectors are expressed in
4476 	 * numbers *relative* to the start of the Solaris fdisk partition.
4477 	 */
4478 	lmap = cl->cl_map;
4479 	vpart = user_vtoc->v_part;
4480 
4481 	for (i = 0; i < (int)user_vtoc->v_nparts; i++, lmap++, vpart++) {
4482 		lmap->dkl_cylno = (unsigned)vpart->p_start / nblks;
4483 		lmap->dkl_nblk = (unsigned)vpart->p_size;
4484 	}
4485 
4486 #elif defined(_SUNOS_VTOC_8)
4487 
4488 	cl->cl_vtoc.v_bootinfo[0] = (uint32_t)user_vtoc->v_bootinfo[0];
4489 	cl->cl_vtoc.v_bootinfo[1] = (uint32_t)user_vtoc->v_bootinfo[1];
4490 	cl->cl_vtoc.v_bootinfo[2] = (uint32_t)user_vtoc->v_bootinfo[2];
4491 
4492 	cl->cl_vtoc.v_sanity = (uint32_t)user_vtoc->v_sanity;
4493 	cl->cl_vtoc.v_version = (uint32_t)user_vtoc->v_version;
4494 
4495 	bcopy(user_vtoc->v_volume, cl->cl_vtoc.v_volume, LEN_DKL_VVOL);
4496 
4497 	cl->cl_vtoc.v_nparts = user_vtoc->v_nparts;
4498 
4499 	for (i = 0; i < 10; i++)
4500 		cl->cl_vtoc.v_reserved[i] =  user_vtoc->v_reserved[i];
4501 
4502 	/*
4503 	 * Note the conversion from starting sector number
4504 	 * to starting cylinder number.
4505 	 * Return error if division results in a remainder.
4506 	 */
4507 	lmap = cl->cl_map;
4508 	lpart = cl->cl_vtoc.v_part;
4509 	vpart = user_vtoc->v_part;
4510 
4511 	for (i = 0; i < (int)user_vtoc->v_nparts; i++) {
4512 		lpart->p_tag  = vpart->p_tag;
4513 		lpart->p_flag = vpart->p_flag;
4514 		lmap->dkl_cylno = (unsigned)vpart->p_start / nblks;
4515 		lmap->dkl_nblk = (unsigned)vpart->p_size;
4516 
4517 		lmap++;
4518 		lpart++;
4519 		vpart++;
4520 
4521 		/* (4387723) */
4522 #ifdef _LP64
4523 		if (user_vtoc->timestamp[i] > TIME32_MAX) {
4524 			cl->cl_vtoc.v_timestamp[i] = TIME32_MAX;
4525 		} else {
4526 			cl->cl_vtoc.v_timestamp[i] = user_vtoc->timestamp[i];
4527 		}
4528 #else
4529 		cl->cl_vtoc.v_timestamp[i] = user_vtoc->timestamp[i];
4530 #endif
4531 	}
4532 
4533 	bcopy(user_vtoc->v_asciilabel, cl->cl_asciilabel, LEN_DKL_ASCII);
4534 #else
4535 #error "No VTOC format defined."
4536 #endif
4537 	return (0);
4538 }
4539 
4540 /*
4541  *    Function: cmlb_clear_efi
4542  *
4543  * Description: This routine clears all EFI labels.
4544  *
4545  *   Arguments:
4546  *	cl		 driver soft state (unit) structure
4547  *
4548  *	tg_cookie	cookie from target driver to be passed back to target
4549  *			driver when we call back to it through tg_ops.
4550  * Return Code: void
4551  */
4552 static void
4553 cmlb_clear_efi(struct cmlb_lun *cl, void *tg_cookie)
4554 {
4555 	efi_gpt_t	*gpt;
4556 	diskaddr_t	cap;
4557 	int		rval;
4558 
4559 	ASSERT(!mutex_owned(CMLB_MUTEX(cl)));
4560 
4561 	mutex_enter(CMLB_MUTEX(cl));
4562 	cl->cl_reserved = -1;
4563 	mutex_exit(CMLB_MUTEX(cl));
4564 
4565 	gpt = kmem_alloc(cl->cl_sys_blocksize, KM_SLEEP);
4566 
4567 	if (DK_TG_READ(cl, gpt, 1, cl->cl_sys_blocksize, tg_cookie) != 0) {
4568 		goto done;
4569 	}
4570 
4571 	cmlb_swap_efi_gpt(gpt);
4572 	rval = cmlb_validate_efi(gpt);
4573 	if (rval == 0) {
4574 		/* clear primary */
4575 		bzero(gpt, sizeof (efi_gpt_t));
4576 		if (rval = DK_TG_WRITE(cl, gpt, 1, cl->cl_sys_blocksize,
4577 		    tg_cookie)) {
4578 			cmlb_dbg(CMLB_INFO,  cl,
4579 			    "cmlb_clear_efi: clear primary label failed\n");
4580 		}
4581 	}
4582 	/* the backup */
4583 	rval = DK_TG_GETCAP(cl, &cap, tg_cookie);
4584 	if (rval) {
4585 		goto done;
4586 	}
4587 
4588 	if ((rval = DK_TG_READ(cl, gpt, cap - 1, cl->cl_sys_blocksize,
4589 	    tg_cookie)) != 0) {
4590 		goto done;
4591 	}
4592 	cmlb_swap_efi_gpt(gpt);
4593 	rval = cmlb_validate_efi(gpt);
4594 	if (rval == 0) {
4595 		/* clear backup */
4596 		cmlb_dbg(CMLB_TRACE,  cl,
4597 		    "cmlb_clear_efi clear backup@%lu\n", cap - 1);
4598 		bzero(gpt, sizeof (efi_gpt_t));
4599 		if ((rval = DK_TG_WRITE(cl,  gpt, cap - 1, cl->cl_sys_blocksize,
4600 		    tg_cookie))) {
4601 			cmlb_dbg(CMLB_INFO,  cl,
4602 			    "cmlb_clear_efi: clear backup label failed\n");
4603 		}
4604 	} else {
4605 		/*
4606 		 * Refer to comments related to off-by-1 at the
4607 		 * header of this file
4608 		 */
4609 		if ((rval = DK_TG_READ(cl, gpt, cap - 2,
4610 		    cl->cl_sys_blocksize, tg_cookie)) != 0) {
4611 			goto done;
4612 		}
4613 		cmlb_swap_efi_gpt(gpt);
4614 		rval = cmlb_validate_efi(gpt);
4615 		if (rval == 0) {
4616 			/* clear legacy backup EFI label */
4617 			cmlb_dbg(CMLB_TRACE,  cl,
4618 			    "cmlb_clear_efi clear legacy backup@%lu\n",
4619 			    cap - 2);
4620 			bzero(gpt, sizeof (efi_gpt_t));
4621 			if ((rval = DK_TG_WRITE(cl,  gpt, cap - 2,
4622 			    cl->cl_sys_blocksize, tg_cookie))) {
4623 				cmlb_dbg(CMLB_INFO,  cl,
4624 				"cmlb_clear_efi: clear legacy backup label "
4625 				"failed\n");
4626 			}
4627 		}
4628 	}
4629 
4630 done:
4631 	kmem_free(gpt, cl->cl_sys_blocksize);
4632 }
4633 
4634 /*
4635  *    Function: cmlb_set_vtoc
4636  *
4637  * Description: This routine writes data to the appropriate positions
4638  *
4639  *   Arguments:
4640  *	cl		driver soft state (unit) structure
4641  *
4642  *	dkl		the data to be written
4643  *
4644  *	tg_cookie	cookie from target driver to be passed back to target
4645  *			driver when we call back to it through tg_ops.
4646  *
4647  * Return: void
4648  */
4649 static int
4650 cmlb_set_vtoc(struct cmlb_lun *cl, struct dk_label *dkl, void *tg_cookie)
4651 {
4652 	uint_t	label_addr;
4653 	int	sec;
4654 	diskaddr_t	blk;
4655 	int	head;
4656 	int	cyl;
4657 	int	rval;
4658 
4659 #if defined(__i386) || defined(__amd64)
4660 	label_addr = cl->cl_solaris_offset + DK_LABEL_LOC;
4661 #else
4662 	/* Write the primary label at block 0 of the solaris partition. */
4663 	label_addr = 0;
4664 #endif
4665 
4666 	rval = DK_TG_WRITE(cl, dkl, label_addr, cl->cl_sys_blocksize,
4667 	    tg_cookie);
4668 
4669 	if (rval != 0) {
4670 		return (rval);
4671 	}
4672 
4673 	/*
4674 	 * Calculate where the backup labels go.  They are always on
4675 	 * the last alternate cylinder, but some older drives put them
4676 	 * on head 2 instead of the last head.	They are always on the
4677 	 * first 5 odd sectors of the appropriate track.
4678 	 *
4679 	 * We have no choice at this point, but to believe that the
4680 	 * disk label is valid.	 Use the geometry of the disk
4681 	 * as described in the label.
4682 	 */
4683 	cyl  = dkl->dkl_ncyl  + dkl->dkl_acyl - 1;
4684 	head = dkl->dkl_nhead - 1;
4685 
4686 	/*
4687 	 * Write and verify the backup labels. Make sure we don't try to
4688 	 * write past the last cylinder.
4689 	 */
4690 	for (sec = 1; ((sec < 5 * 2 + 1) && (sec < dkl->dkl_nsect)); sec += 2) {
4691 		blk = (diskaddr_t)(
4692 		    (cyl * ((dkl->dkl_nhead * dkl->dkl_nsect) - dkl->dkl_apc)) +
4693 		    (head * dkl->dkl_nsect) + sec);
4694 #if defined(__i386) || defined(__amd64)
4695 		blk += cl->cl_solaris_offset;
4696 #endif
4697 		rval = DK_TG_WRITE(cl, dkl, blk, cl->cl_sys_blocksize,
4698 		    tg_cookie);
4699 		cmlb_dbg(CMLB_INFO,  cl,
4700 		"cmlb_set_vtoc: wrote backup label %llx\n", blk);
4701 		if (rval != 0) {
4702 			goto exit;
4703 		}
4704 	}
4705 exit:
4706 	return (rval);
4707 }
4708 
4709 /*
4710  *    Function: cmlb_clear_vtoc
4711  *
4712  * Description: This routine clears out the VTOC labels.
4713  *
4714  *   Arguments:
4715  *	cl		driver soft state (unit) structure
4716  *
4717  *	tg_cookie	cookie from target driver to be passed back to target
4718  *			driver when we call back to it through tg_ops.
4719  *
4720  * Return: void
4721  */
4722 static void
4723 cmlb_clear_vtoc(struct cmlb_lun *cl, void *tg_cookie)
4724 {
4725 	struct dk_label		*dkl;
4726 
4727 	mutex_exit(CMLB_MUTEX(cl));
4728 	dkl = kmem_zalloc(cl->cl_sys_blocksize, KM_SLEEP);
4729 	mutex_enter(CMLB_MUTEX(cl));
4730 	/*
4731 	 * cmlb_set_vtoc uses these fields in order to figure out
4732 	 * where to overwrite the backup labels
4733 	 */
4734 	dkl->dkl_apc    = cl->cl_g.dkg_apc;
4735 	dkl->dkl_ncyl   = cl->cl_g.dkg_ncyl;
4736 	dkl->dkl_acyl   = cl->cl_g.dkg_acyl;
4737 	dkl->dkl_nhead  = cl->cl_g.dkg_nhead;
4738 	dkl->dkl_nsect  = cl->cl_g.dkg_nsect;
4739 	mutex_exit(CMLB_MUTEX(cl));
4740 	(void) cmlb_set_vtoc(cl, dkl, tg_cookie);
4741 	kmem_free(dkl, cl->cl_sys_blocksize);
4742 
4743 	mutex_enter(CMLB_MUTEX(cl));
4744 }
4745 
4746 /*
4747  *    Function: cmlb_write_label
4748  *
4749  * Description: This routine will validate and write the driver soft state vtoc
4750  *		contents to the device.
4751  *
4752  *   Arguments:
4753  *	cl		cmlb handle
4754  *
4755  *	tg_cookie	cookie from target driver to be passed back to target
4756  *			driver when we call back to it through tg_ops.
4757  *
4758  *
4759  * Return Code: the code returned by cmlb_send_scsi_cmd()
4760  *		0
4761  *		EINVAL
4762  *		ENXIO
4763  *		ENOMEM
4764  */
4765 static int
4766 cmlb_write_label(struct cmlb_lun *cl, void *tg_cookie)
4767 {
4768 	struct dk_label	*dkl;
4769 	short		sum;
4770 	short		*sp;
4771 	int		i;
4772 	int		rval;
4773 
4774 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
4775 	mutex_exit(CMLB_MUTEX(cl));
4776 	dkl = kmem_zalloc(cl->cl_sys_blocksize, KM_SLEEP);
4777 	mutex_enter(CMLB_MUTEX(cl));
4778 
4779 	bcopy(&cl->cl_vtoc, &dkl->dkl_vtoc, sizeof (struct dk_vtoc));
4780 	dkl->dkl_rpm	= cl->cl_g.dkg_rpm;
4781 	dkl->dkl_pcyl	= cl->cl_g.dkg_pcyl;
4782 	dkl->dkl_apc	= cl->cl_g.dkg_apc;
4783 	dkl->dkl_intrlv = cl->cl_g.dkg_intrlv;
4784 	dkl->dkl_ncyl	= cl->cl_g.dkg_ncyl;
4785 	dkl->dkl_acyl	= cl->cl_g.dkg_acyl;
4786 	dkl->dkl_nhead	= cl->cl_g.dkg_nhead;
4787 	dkl->dkl_nsect	= cl->cl_g.dkg_nsect;
4788 
4789 #if defined(_SUNOS_VTOC_8)
4790 	dkl->dkl_obs1	= cl->cl_g.dkg_obs1;
4791 	dkl->dkl_obs2	= cl->cl_g.dkg_obs2;
4792 	dkl->dkl_obs3	= cl->cl_g.dkg_obs3;
4793 	for (i = 0; i < NDKMAP; i++) {
4794 		dkl->dkl_map[i].dkl_cylno = cl->cl_map[i].dkl_cylno;
4795 		dkl->dkl_map[i].dkl_nblk  = cl->cl_map[i].dkl_nblk;
4796 	}
4797 	bcopy(cl->cl_asciilabel, dkl->dkl_asciilabel, LEN_DKL_ASCII);
4798 #elif defined(_SUNOS_VTOC_16)
4799 	dkl->dkl_skew	= cl->cl_dkg_skew;
4800 #else
4801 #error "No VTOC format defined."
4802 #endif
4803 
4804 	dkl->dkl_magic			= DKL_MAGIC;
4805 	dkl->dkl_write_reinstruct	= cl->cl_g.dkg_write_reinstruct;
4806 	dkl->dkl_read_reinstruct	= cl->cl_g.dkg_read_reinstruct;
4807 
4808 	/* Construct checksum for the new disk label */
4809 	sum = 0;
4810 	sp = (short *)dkl;
4811 	i = sizeof (struct dk_label) / sizeof (short);
4812 	while (i--) {
4813 		sum ^= *sp++;
4814 	}
4815 	dkl->dkl_cksum = sum;
4816 
4817 	mutex_exit(CMLB_MUTEX(cl));
4818 
4819 	rval = cmlb_set_vtoc(cl, dkl, tg_cookie);
4820 exit:
4821 	kmem_free(dkl, cl->cl_sys_blocksize);
4822 	mutex_enter(CMLB_MUTEX(cl));
4823 	return (rval);
4824 }
4825 
4826 static int
4827 cmlb_dkio_set_efi(struct cmlb_lun *cl, dev_t dev, caddr_t arg, int flag,
4828     void *tg_cookie)
4829 {
4830 	dk_efi_t	user_efi;
4831 	int		rval = 0;
4832 	void		*buffer;
4833 	diskaddr_t	tgt_lba;
4834 	boolean_t	internal;
4835 
4836 	if (ddi_copyin(arg, &user_efi, sizeof (dk_efi_t), flag))
4837 		return (EFAULT);
4838 
4839 	internal = VOID2BOOLEAN(
4840 	    (cl->cl_alter_behavior & (CMLB_INTERNAL_MINOR_NODES)) != 0);
4841 
4842 	user_efi.dki_data = (void *)(uintptr_t)user_efi.dki_data_64;
4843 
4844 	buffer = kmem_alloc(user_efi.dki_length, KM_SLEEP);
4845 	if (ddi_copyin(user_efi.dki_data, buffer, user_efi.dki_length, flag)) {
4846 		rval = EFAULT;
4847 	} else {
4848 		/*
4849 		 * let's clear the vtoc labels and clear the softstate
4850 		 * vtoc.
4851 		 */
4852 		mutex_enter(CMLB_MUTEX(cl));
4853 		if (cl->cl_vtoc.v_sanity == VTOC_SANE) {
4854 			cmlb_dbg(CMLB_TRACE,  cl,
4855 			    "cmlb_dkio_set_efi: CLEAR VTOC\n");
4856 			if (cl->cl_label_from_media == CMLB_LABEL_VTOC)
4857 				cmlb_clear_vtoc(cl, tg_cookie);
4858 			bzero(&cl->cl_vtoc, sizeof (struct dk_vtoc));
4859 			mutex_exit(CMLB_MUTEX(cl));
4860 			ddi_remove_minor_node(CMLB_DEVINFO(cl), "h");
4861 			ddi_remove_minor_node(CMLB_DEVINFO(cl), "h,raw");
4862 			(void) cmlb_create_minor(CMLB_DEVINFO(cl), "wd",
4863 			    S_IFBLK,
4864 			    (CMLBUNIT(dev) << CMLBUNIT_SHIFT) | WD_NODE,
4865 			    cl->cl_node_type, NULL, internal);
4866 			(void) cmlb_create_minor(CMLB_DEVINFO(cl), "wd,raw",
4867 			    S_IFCHR,
4868 			    (CMLBUNIT(dev) << CMLBUNIT_SHIFT) | WD_NODE,
4869 			    cl->cl_node_type, NULL, internal);
4870 		} else
4871 			mutex_exit(CMLB_MUTEX(cl));
4872 
4873 		tgt_lba = user_efi.dki_lba;
4874 
4875 		mutex_enter(CMLB_MUTEX(cl));
4876 		if ((cmlb_check_update_blockcount(cl, tg_cookie) != 0) ||
4877 		    (cl->cl_tgt_blocksize == 0)) {
4878 			kmem_free(buffer, user_efi.dki_length);
4879 			mutex_exit(CMLB_MUTEX(cl));
4880 			return (EINVAL);
4881 		}
4882 		if (cl->cl_tgt_blocksize != cl->cl_sys_blocksize)
4883 			tgt_lba = tgt_lba *
4884 			    cl->cl_tgt_blocksize / cl->cl_sys_blocksize;
4885 
4886 		mutex_exit(CMLB_MUTEX(cl));
4887 		rval = DK_TG_WRITE(cl, buffer, tgt_lba, user_efi.dki_length,
4888 		    tg_cookie);
4889 
4890 		if (rval == 0) {
4891 			mutex_enter(CMLB_MUTEX(cl));
4892 			cl->cl_f_geometry_is_valid = B_FALSE;
4893 			mutex_exit(CMLB_MUTEX(cl));
4894 		}
4895 	}
4896 	kmem_free(buffer, user_efi.dki_length);
4897 	return (rval);
4898 }
4899 
4900 /*
4901  *    Function: cmlb_dkio_get_mboot
4902  *
4903  * Description: This routine is the driver entry point for handling user
4904  *		requests to get the current device mboot (DKIOCGMBOOT)
4905  *
4906  *   Arguments:
4907  *	arg		pointer to user provided mboot structure specifying
4908  *			the current mboot.
4909  *
4910  *	flag		this argument is a pass through to ddi_copyxxx()
4911  *			directly from the mode argument of ioctl().
4912  *
4913  *	tg_cookie	cookie from target driver to be passed back to target
4914  *			driver when we call back to it through tg_ops.
4915  *
4916  * Return Code: 0
4917  *		EINVAL
4918  *		EFAULT
4919  *		ENXIO
4920  */
4921 static int
4922 cmlb_dkio_get_mboot(struct cmlb_lun *cl, caddr_t arg, int flag, void *tg_cookie)
4923 {
4924 	struct mboot	*mboot;
4925 	int		rval;
4926 	size_t		buffer_size;
4927 
4928 
4929 #if defined(_SUNOS_VTOC_8)
4930 	if ((!ISREMOVABLE(cl) && !ISHOTPLUGGABLE(cl)) || (arg == NULL)) {
4931 #elif defined(_SUNOS_VTOC_16)
4932 	if (arg == NULL) {
4933 #endif
4934 		return (EINVAL);
4935 	}
4936 
4937 	/*
4938 	 * Read the mboot block, located at absolute block 0 on the target.
4939 	 */
4940 	buffer_size = cl->cl_sys_blocksize;
4941 
4942 	cmlb_dbg(CMLB_TRACE,  cl,
4943 	    "cmlb_dkio_get_mboot: allocation size: 0x%x\n", buffer_size);
4944 
4945 	mboot = kmem_zalloc(buffer_size, KM_SLEEP);
4946 	if ((rval = DK_TG_READ(cl, mboot, 0, buffer_size, tg_cookie)) == 0) {
4947 		if (ddi_copyout(mboot, (void *)arg,
4948 		    sizeof (struct mboot), flag) != 0) {
4949 			rval = EFAULT;
4950 		}
4951 	}
4952 	kmem_free(mboot, buffer_size);
4953 	return (rval);
4954 }
4955 
4956 
4957 /*
4958  *    Function: cmlb_dkio_set_mboot
4959  *
4960  * Description: This routine is the driver entry point for handling user
4961  *		requests to validate and set the device master boot
4962  *		(DKIOCSMBOOT).
4963  *
4964  *   Arguments:
4965  *	arg		pointer to user provided mboot structure used to set the
4966  *			master boot.
4967  *
4968  *	flag		this argument is a pass through to ddi_copyxxx()
4969  *			directly from the mode argument of ioctl().
4970  *
4971  *	tg_cookie	cookie from target driver to be passed back to target
4972  *			driver when we call back to it through tg_ops.
4973  *
4974  * Return Code: 0
4975  *		EINVAL
4976  *		EFAULT
4977  *		ENXIO
4978  */
4979 static int
4980 cmlb_dkio_set_mboot(struct cmlb_lun *cl, caddr_t arg, int flag, void *tg_cookie)
4981 {
4982 	struct mboot	*mboot = NULL;
4983 	int		rval;
4984 	ushort_t	magic;
4985 
4986 
4987 	ASSERT(!mutex_owned(CMLB_MUTEX(cl)));
4988 
4989 #if defined(_SUNOS_VTOC_8)
4990 	if (!ISREMOVABLE(cl) && !ISHOTPLUGGABLE(cl)) {
4991 		return (EINVAL);
4992 	}
4993 #endif
4994 
4995 	if (arg == NULL) {
4996 		return (EINVAL);
4997 	}
4998 
4999 	mboot = kmem_zalloc(cl->cl_sys_blocksize, KM_SLEEP);
5000 
5001 	if (ddi_copyin((const void *)arg, mboot,
5002 	    cl->cl_sys_blocksize, flag) != 0) {
5003 		kmem_free(mboot, cl->cl_sys_blocksize);
5004 		return (EFAULT);
5005 	}
5006 
5007 	/* Is this really a master boot record? */
5008 	magic = LE_16(mboot->signature);
5009 	if (magic != MBB_MAGIC) {
5010 		kmem_free(mboot, cl->cl_sys_blocksize);
5011 		return (EINVAL);
5012 	}
5013 
5014 	rval = DK_TG_WRITE(cl, mboot, 0, cl->cl_sys_blocksize, tg_cookie);
5015 
5016 	mutex_enter(CMLB_MUTEX(cl));
5017 #if defined(__i386) || defined(__amd64)
5018 	if (rval == 0) {
5019 		/*
5020 		 * mboot has been written successfully.
5021 		 * update the fdisk and vtoc tables in memory
5022 		 */
5023 		rval = cmlb_update_fdisk_and_vtoc(cl, tg_cookie);
5024 		if ((!cl->cl_f_geometry_is_valid) || (rval != 0)) {
5025 			mutex_exit(CMLB_MUTEX(cl));
5026 			kmem_free(mboot, cl->cl_sys_blocksize);
5027 			return (rval);
5028 		}
5029 	}
5030 
5031 #ifdef __lock_lint
5032 	cmlb_setup_default_geometry(cl, tg_cookie);
5033 #endif
5034 
5035 #else
5036 	if (rval == 0) {
5037 		/*
5038 		 * mboot has been written successfully.
5039 		 * set up the default geometry and VTOC
5040 		 */
5041 		if (cl->cl_blockcount <= CMLB_EXTVTOC_LIMIT)
5042 			cmlb_setup_default_geometry(cl, tg_cookie);
5043 	}
5044 #endif
5045 	cl->cl_msglog_flag |= CMLB_ALLOW_2TB_WARN;
5046 	mutex_exit(CMLB_MUTEX(cl));
5047 	kmem_free(mboot, cl->cl_sys_blocksize);
5048 	return (rval);
5049 }
5050 
5051 
5052 #if defined(__i386) || defined(__amd64)
5053 /*ARGSUSED*/
5054 static int
5055 cmlb_dkio_set_ext_part(struct cmlb_lun *cl, caddr_t arg, int flag,
5056     void *tg_cookie)
5057 {
5058 	int fdisk_rval;
5059 	diskaddr_t capacity;
5060 
5061 	ASSERT(!mutex_owned(CMLB_MUTEX(cl)));
5062 
5063 	mutex_enter(CMLB_MUTEX(cl));
5064 	capacity = cl->cl_blockcount;
5065 	fdisk_rval = cmlb_read_fdisk(cl, capacity, tg_cookie);
5066 	if (fdisk_rval != 0) {
5067 		mutex_exit(CMLB_MUTEX(cl));
5068 		return (fdisk_rval);
5069 	}
5070 
5071 	mutex_exit(CMLB_MUTEX(cl));
5072 	return (fdisk_rval);
5073 }
5074 #endif
5075 
5076 /*
5077  *    Function: cmlb_setup_default_geometry
5078  *
5079  * Description: This local utility routine sets the default geometry as part of
5080  *		setting the device mboot.
5081  *
5082  *   Arguments:
5083  *	cl		driver soft state (unit) structure
5084  *
5085  *	tg_cookie	cookie from target driver to be passed back to target
5086  *			driver when we call back to it through tg_ops.
5087  *
5088  *
5089  * Note: This may be redundant with cmlb_build_default_label.
5090  */
5091 static void
5092 cmlb_setup_default_geometry(struct cmlb_lun *cl, void *tg_cookie)
5093 {
5094 	struct cmlb_geom	pgeom;
5095 	struct cmlb_geom	*pgeomp = &pgeom;
5096 	int			ret;
5097 	int			geom_base_cap = 1;
5098 
5099 
5100 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
5101 
5102 	/* zero out the soft state geometry and partition table. */
5103 	bzero(&cl->cl_g, sizeof (struct dk_geom));
5104 	bzero(&cl->cl_vtoc, sizeof (struct dk_vtoc));
5105 	bzero(cl->cl_map, NDKMAP * (sizeof (struct dk_map)));
5106 
5107 	/*
5108 	 * For the rpm, we use the minimum for the disk.
5109 	 * For the head, cyl and number of sector per track,
5110 	 * if the capacity <= 1GB, head = 64, sect = 32.
5111 	 * else head = 255, sect 63
5112 	 * Note: the capacity should be equal to C*H*S values.
5113 	 * This will cause some truncation of size due to
5114 	 * round off errors. For CD-ROMs, this truncation can
5115 	 * have adverse side effects, so returning ncyl and
5116 	 * nhead as 1. The nsect will overflow for most of
5117 	 * CD-ROMs as nsect is of type ushort.
5118 	 */
5119 	if (cl->cl_alter_behavior & CMLB_FAKE_GEOM_LABEL_IOCTLS_VTOC8) {
5120 		/*
5121 		 * newfs currently can not handle 255 ntracks for SPARC
5122 		 * so get the geometry from target driver instead of coming up
5123 		 * with one based on capacity.
5124 		 */
5125 		mutex_exit(CMLB_MUTEX(cl));
5126 		ret = DK_TG_GETPHYGEOM(cl, pgeomp, tg_cookie);
5127 		mutex_enter(CMLB_MUTEX(cl));
5128 
5129 		if (ret == 0) {
5130 			geom_base_cap = 0;
5131 		} else {
5132 			cmlb_dbg(CMLB_ERROR,  cl,
5133 			    "cmlb_setup_default_geometry: "
5134 			    "tg_getphygeom failed %d\n", ret);
5135 
5136 			/* do default setting, geometry based on capacity */
5137 		}
5138 	}
5139 
5140 	if (geom_base_cap) {
5141 		if (ISCD(cl)) {
5142 			cl->cl_g.dkg_ncyl = 1;
5143 			cl->cl_g.dkg_nhead = 1;
5144 			cl->cl_g.dkg_nsect = cl->cl_blockcount;
5145 		} else if (cl->cl_blockcount <= 0x1000) {
5146 			/* Needed for unlabeled SCSI floppies. */
5147 			cl->cl_g.dkg_nhead = 2;
5148 			cl->cl_g.dkg_ncyl = 80;
5149 			cl->cl_g.dkg_pcyl = 80;
5150 			cl->cl_g.dkg_nsect = cl->cl_blockcount / (2 * 80);
5151 		} else if (cl->cl_blockcount <= 0x200000) {
5152 			cl->cl_g.dkg_nhead = 64;
5153 			cl->cl_g.dkg_nsect = 32;
5154 			cl->cl_g.dkg_ncyl = cl->cl_blockcount / (64 * 32);
5155 		} else {
5156 			cl->cl_g.dkg_nhead = 255;
5157 
5158 			cl->cl_g.dkg_nsect = ((cl->cl_blockcount +
5159 			    (UINT16_MAX * 255 * 63) - 1) /
5160 			    (UINT16_MAX * 255 * 63)) * 63;
5161 
5162 			if (cl->cl_g.dkg_nsect == 0)
5163 				cl->cl_g.dkg_nsect = (UINT16_MAX / 63) * 63;
5164 
5165 			cl->cl_g.dkg_ncyl = cl->cl_blockcount /
5166 			    (255 * cl->cl_g.dkg_nsect);
5167 		}
5168 
5169 		cl->cl_g.dkg_acyl = 0;
5170 		cl->cl_g.dkg_bcyl = 0;
5171 		cl->cl_g.dkg_intrlv = 1;
5172 		cl->cl_g.dkg_rpm = 200;
5173 		if (cl->cl_g.dkg_pcyl == 0)
5174 			cl->cl_g.dkg_pcyl = cl->cl_g.dkg_ncyl +
5175 			    cl->cl_g.dkg_acyl;
5176 	} else {
5177 		cl->cl_g.dkg_ncyl = (short)pgeomp->g_ncyl;
5178 		cl->cl_g.dkg_acyl = pgeomp->g_acyl;
5179 		cl->cl_g.dkg_nhead = pgeomp->g_nhead;
5180 		cl->cl_g.dkg_nsect = pgeomp->g_nsect;
5181 		cl->cl_g.dkg_intrlv = pgeomp->g_intrlv;
5182 		cl->cl_g.dkg_rpm = pgeomp->g_rpm;
5183 		cl->cl_g.dkg_pcyl = cl->cl_g.dkg_ncyl + cl->cl_g.dkg_acyl;
5184 	}
5185 
5186 	cl->cl_g.dkg_read_reinstruct = 0;
5187 	cl->cl_g.dkg_write_reinstruct = 0;
5188 	cl->cl_solaris_size = cl->cl_g.dkg_ncyl *
5189 	    cl->cl_g.dkg_nhead * cl->cl_g.dkg_nsect;
5190 
5191 	cl->cl_map['a'-'a'].dkl_cylno = 0;
5192 	cl->cl_map['a'-'a'].dkl_nblk = cl->cl_solaris_size;
5193 
5194 	cl->cl_map['c'-'a'].dkl_cylno = 0;
5195 	cl->cl_map['c'-'a'].dkl_nblk = cl->cl_solaris_size;
5196 
5197 	cl->cl_vtoc.v_part[2].p_tag   = V_BACKUP;
5198 	cl->cl_vtoc.v_part[2].p_flag  = V_UNMNT;
5199 	cl->cl_vtoc.v_nparts = V_NUMPAR;
5200 	cl->cl_vtoc.v_version = V_VERSION;
5201 	(void) sprintf((char *)cl->cl_asciilabel, "DEFAULT cyl %d alt %d"
5202 	    " hd %d sec %d", cl->cl_g.dkg_ncyl, cl->cl_g.dkg_acyl,
5203 	    cl->cl_g.dkg_nhead, cl->cl_g.dkg_nsect);
5204 
5205 	cl->cl_f_geometry_is_valid = B_FALSE;
5206 }
5207 
5208 
5209 #if defined(__i386) || defined(__amd64)
5210 /*
5211  *    Function: cmlb_update_fdisk_and_vtoc
5212  *
5213  * Description: This local utility routine updates the device fdisk and vtoc
5214  *		as part of setting the device mboot.
5215  *
5216  *   Arguments:
5217  *	cl		driver soft state (unit) structure
5218  *
5219  *	tg_cookie	cookie from target driver to be passed back to target
5220  *			driver when we call back to it through tg_ops.
5221  *
5222  *
5223  * Return Code: 0 for success or errno-type return code.
5224  *
5225  *    Note:x86: This looks like a duplicate of cmlb_validate_geometry(), but
5226  *		these did exist separately in x86 sd.c.
5227  */
5228 static int
5229 cmlb_update_fdisk_and_vtoc(struct cmlb_lun *cl, void *tg_cookie)
5230 {
5231 	int		count;
5232 	int		label_rc = 0;
5233 	int		fdisk_rval;
5234 	diskaddr_t	capacity;
5235 
5236 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
5237 
5238 	if (cmlb_check_update_blockcount(cl, tg_cookie) != 0)
5239 		return (EINVAL);
5240 
5241 #if defined(_SUNOS_VTOC_16)
5242 	/*
5243 	 * Set up the "whole disk" fdisk partition; this should always
5244 	 * exist, regardless of whether the disk contains an fdisk table
5245 	 * or vtoc.
5246 	 */
5247 	cl->cl_map[P0_RAW_DISK].dkl_cylno = 0;
5248 	cl->cl_map[P0_RAW_DISK].dkl_nblk = cl->cl_blockcount;
5249 #endif	/* defined(_SUNOS_VTOC_16) */
5250 
5251 	/*
5252 	 * copy the lbasize and capacity so that if they're
5253 	 * reset while we're not holding the CMLB_MUTEX(cl), we will
5254 	 * continue to use valid values after the CMLB_MUTEX(cl) is
5255 	 * reacquired.
5256 	 */
5257 	capacity = cl->cl_blockcount;
5258 
5259 	/*
5260 	 * refresh the logical and physical geometry caches.
5261 	 * (data from mode sense format/rigid disk geometry pages,
5262 	 * and scsi_ifgetcap("geometry").
5263 	 */
5264 	cmlb_resync_geom_caches(cl, capacity, tg_cookie);
5265 
5266 	/*
5267 	 * Only DIRECT ACCESS devices will have Scl labels.
5268 	 * CD's supposedly have a Scl label, too
5269 	 */
5270 	if (cl->cl_device_type == DTYPE_DIRECT || ISREMOVABLE(cl)) {
5271 		fdisk_rval = cmlb_read_fdisk(cl, capacity, tg_cookie);
5272 		if (fdisk_rval != 0) {
5273 			ASSERT(mutex_owned(CMLB_MUTEX(cl)));
5274 			return (fdisk_rval);
5275 		}
5276 
5277 		if (cl->cl_solaris_size <= DK_LABEL_LOC) {
5278 			/*
5279 			 * Found fdisk table but no Solaris partition entry,
5280 			 * so don't call cmlb_uselabel() and don't create
5281 			 * a default label.
5282 			 */
5283 			label_rc = 0;
5284 			cl->cl_f_geometry_is_valid = B_TRUE;
5285 			goto no_solaris_partition;
5286 		}
5287 	} else if (capacity < 0) {
5288 		ASSERT(mutex_owned(CMLB_MUTEX(cl)));
5289 		return (EINVAL);
5290 	}
5291 
5292 	/*
5293 	 * For Removable media We reach here if we have found a
5294 	 * SOLARIS PARTITION.
5295 	 * If cl_f_geometry_is_valid is B_FALSE it indicates that the SOLARIS
5296 	 * PARTITION has changed from the previous one, hence we will setup a
5297 	 * default VTOC in this case.
5298 	 */
5299 	if (!cl->cl_f_geometry_is_valid) {
5300 		/* if we get here it is writable */
5301 		/* we are called from SMBOOT, and after a write of fdisk */
5302 		cmlb_build_default_label(cl, tg_cookie);
5303 		label_rc = 0;
5304 	}
5305 
5306 no_solaris_partition:
5307 
5308 #if defined(_SUNOS_VTOC_16)
5309 	/*
5310 	 * If we have valid geometry, set up the remaining fdisk partitions.
5311 	 * Note that dkl_cylno is not used for the fdisk map entries, so
5312 	 * we set it to an entirely bogus value.
5313 	 */
5314 	for (count = 0; count < FDISK_PARTS; count++) {
5315 		cl->cl_map[FDISK_P1 + count].dkl_cylno = UINT32_MAX;
5316 		cl->cl_map[FDISK_P1 + count].dkl_nblk =
5317 		    cl->cl_fmap[count].fmap_nblk;
5318 		cl->cl_offset[FDISK_P1 + count] =
5319 		    cl->cl_fmap[count].fmap_start;
5320 	}
5321 #endif
5322 
5323 	for (count = 0; count < NDKMAP; count++) {
5324 #if defined(_SUNOS_VTOC_8)
5325 		struct dk_map *lp  = &cl->cl_map[count];
5326 		cl->cl_offset[count] =
5327 		    cl->cl_g.dkg_nhead * cl->cl_g.dkg_nsect * lp->dkl_cylno;
5328 #elif defined(_SUNOS_VTOC_16)
5329 		struct dkl_partition *vp = &cl->cl_vtoc.v_part[count];
5330 		cl->cl_offset[count] = vp->p_start + cl->cl_solaris_offset;
5331 #else
5332 #error "No VTOC format defined."
5333 #endif
5334 	}
5335 
5336 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
5337 	return (label_rc);
5338 }
5339 #endif
5340 
5341 #if defined(__i386) || defined(__amd64)
5342 static int
5343 cmlb_dkio_get_virtgeom(struct cmlb_lun *cl, caddr_t arg, int flag)
5344 {
5345 	int err = 0;
5346 
5347 	/* Return the driver's notion of the media's logical geometry */
5348 	struct dk_geom	disk_geom;
5349 	struct dk_geom	*dkgp = &disk_geom;
5350 
5351 	mutex_enter(CMLB_MUTEX(cl));
5352 	/*
5353 	 * If there is no HBA geometry available, or
5354 	 * if the HBA returned us something that doesn't
5355 	 * really fit into an Int 13/function 8 geometry
5356 	 * result, just fail the ioctl.  See PSARC 1998/313.
5357 	 */
5358 	if (cl->cl_lgeom.g_nhead == 0 ||
5359 	    cl->cl_lgeom.g_nsect == 0 ||
5360 	    cl->cl_lgeom.g_ncyl > 1024) {
5361 		mutex_exit(CMLB_MUTEX(cl));
5362 		err = EINVAL;
5363 	} else {
5364 		dkgp->dkg_ncyl	= cl->cl_lgeom.g_ncyl;
5365 		dkgp->dkg_acyl	= cl->cl_lgeom.g_acyl;
5366 		dkgp->dkg_pcyl	= dkgp->dkg_ncyl + dkgp->dkg_acyl;
5367 		dkgp->dkg_nhead	= cl->cl_lgeom.g_nhead;
5368 		dkgp->dkg_nsect	= cl->cl_lgeom.g_nsect;
5369 
5370 		mutex_exit(CMLB_MUTEX(cl));
5371 		if (ddi_copyout(dkgp, (void *)arg,
5372 		    sizeof (struct dk_geom), flag)) {
5373 			err = EFAULT;
5374 		} else {
5375 			err = 0;
5376 		}
5377 	}
5378 	return (err);
5379 }
5380 #endif
5381 
5382 #if defined(__i386) || defined(__amd64)
5383 static int
5384 cmlb_dkio_get_phygeom(struct cmlb_lun *cl, caddr_t  arg, int flag)
5385 {
5386 	int err = 0;
5387 	diskaddr_t capacity;
5388 
5389 
5390 	/* Return the driver's notion of the media physical geometry */
5391 	struct dk_geom	disk_geom;
5392 	struct dk_geom	*dkgp = &disk_geom;
5393 
5394 	mutex_enter(CMLB_MUTEX(cl));
5395 
5396 	if (cl->cl_g.dkg_nhead != 0 &&
5397 	    cl->cl_g.dkg_nsect != 0) {
5398 		/*
5399 		 * We succeeded in getting a geometry, but
5400 		 * right now it is being reported as just the
5401 		 * Solaris fdisk partition, just like for
5402 		 * DKIOCGGEOM. We need to change that to be
5403 		 * correct for the entire disk now.
5404 		 */
5405 		bcopy(&cl->cl_g, dkgp, sizeof (*dkgp));
5406 		dkgp->dkg_acyl = 0;
5407 		dkgp->dkg_ncyl = cl->cl_blockcount /
5408 		    (dkgp->dkg_nhead * dkgp->dkg_nsect);
5409 	} else {
5410 		bzero(dkgp, sizeof (struct dk_geom));
5411 		/*
5412 		 * This disk does not have a Solaris VTOC
5413 		 * so we must present a physical geometry
5414 		 * that will remain consistent regardless
5415 		 * of how the disk is used. This will ensure
5416 		 * that the geometry does not change regardless
5417 		 * of the fdisk partition type (ie. EFI, FAT32,
5418 		 * Solaris, etc).
5419 		 */
5420 		if (ISCD(cl)) {
5421 			dkgp->dkg_nhead = cl->cl_pgeom.g_nhead;
5422 			dkgp->dkg_nsect = cl->cl_pgeom.g_nsect;
5423 			dkgp->dkg_ncyl = cl->cl_pgeom.g_ncyl;
5424 			dkgp->dkg_acyl = cl->cl_pgeom.g_acyl;
5425 		} else {
5426 			/*
5427 			 * Invalid cl_blockcount can generate invalid
5428 			 * dk_geom and may result in division by zero
5429 			 * system failure. Should make sure blockcount
5430 			 * is valid before using it here.
5431 			 */
5432 			if (cl->cl_blockcount == 0) {
5433 				mutex_exit(CMLB_MUTEX(cl));
5434 				err = EIO;
5435 				return (err);
5436 			}
5437 			/*
5438 			 * Refer to comments related to off-by-1 at the
5439 			 * header of this file
5440 			 */
5441 			if (cl->cl_alter_behavior & CMLB_OFF_BY_ONE)
5442 				capacity = cl->cl_blockcount - 1;
5443 			else
5444 				capacity = cl->cl_blockcount;
5445 
5446 			cmlb_convert_geometry(capacity, dkgp);
5447 			dkgp->dkg_acyl = 0;
5448 			dkgp->dkg_ncyl = capacity /
5449 			    (dkgp->dkg_nhead * dkgp->dkg_nsect);
5450 		}
5451 	}
5452 	dkgp->dkg_pcyl = dkgp->dkg_ncyl + dkgp->dkg_acyl;
5453 
5454 	mutex_exit(CMLB_MUTEX(cl));
5455 	if (ddi_copyout(dkgp, (void *)arg, sizeof (struct dk_geom), flag))
5456 		err = EFAULT;
5457 
5458 	return (err);
5459 }
5460 #endif
5461 
5462 #if defined(__i386) || defined(__amd64)
5463 static int
5464 cmlb_dkio_partinfo(struct cmlb_lun *cl, dev_t dev, caddr_t  arg, int flag)
5465 {
5466 	int err = 0;
5467 
5468 	/*
5469 	 * Return parameters describing the selected disk slice.
5470 	 * Note: this ioctl is for the intel platform only
5471 	 */
5472 	int part;
5473 
5474 	part = CMLBPART(dev);
5475 
5476 	mutex_enter(CMLB_MUTEX(cl));
5477 	/* don't check cl_solaris_size for pN */
5478 	if (part < P0_RAW_DISK && cl->cl_solaris_size == 0) {
5479 		err = EIO;
5480 		mutex_exit(CMLB_MUTEX(cl));
5481 	} else {
5482 		struct part_info p;
5483 
5484 		p.p_start = (daddr_t)cl->cl_offset[part];
5485 		p.p_length = (int)cl->cl_map[part].dkl_nblk;
5486 		mutex_exit(CMLB_MUTEX(cl));
5487 #ifdef _MULTI_DATAMODEL
5488 		switch (ddi_model_convert_from(flag & FMODELS)) {
5489 		case DDI_MODEL_ILP32:
5490 		{
5491 			struct part_info32 p32;
5492 
5493 			p32.p_start = (daddr32_t)p.p_start;
5494 			p32.p_length = p.p_length;
5495 			if (ddi_copyout(&p32, (void *)arg,
5496 			    sizeof (p32), flag))
5497 				err = EFAULT;
5498 			break;
5499 		}
5500 
5501 		case DDI_MODEL_NONE:
5502 		{
5503 			if (ddi_copyout(&p, (void *)arg, sizeof (p),
5504 			    flag))
5505 				err = EFAULT;
5506 			break;
5507 		}
5508 		}
5509 #else /* ! _MULTI_DATAMODEL */
5510 		if (ddi_copyout(&p, (void *)arg, sizeof (p), flag))
5511 			err = EFAULT;
5512 #endif /* _MULTI_DATAMODEL */
5513 	}
5514 	return (err);
5515 }
5516 static int
5517 cmlb_dkio_extpartinfo(struct cmlb_lun *cl, dev_t dev, caddr_t  arg, int flag)
5518 {
5519 	int err = 0;
5520 
5521 	/*
5522 	 * Return parameters describing the selected disk slice.
5523 	 * Note: this ioctl is for the intel platform only
5524 	 */
5525 	int part;
5526 
5527 	part = CMLBPART(dev);
5528 
5529 	mutex_enter(CMLB_MUTEX(cl));
5530 	/* don't check cl_solaris_size for pN */
5531 	if (part < P0_RAW_DISK && cl->cl_solaris_size == 0) {
5532 		err = EIO;
5533 		mutex_exit(CMLB_MUTEX(cl));
5534 	} else {
5535 		struct extpart_info p;
5536 
5537 		p.p_start = (diskaddr_t)cl->cl_offset[part];
5538 		p.p_length = (diskaddr_t)cl->cl_map[part].dkl_nblk;
5539 		mutex_exit(CMLB_MUTEX(cl));
5540 		if (ddi_copyout(&p, (void *)arg, sizeof (p), flag))
5541 			err = EFAULT;
5542 	}
5543 	return (err);
5544 }
5545 #endif
5546 
5547 int
5548 cmlb_prop_op(cmlb_handle_t cmlbhandle,
5549     dev_t dev, dev_info_t *dip, ddi_prop_op_t prop_op, int mod_flags,
5550     char *name, caddr_t valuep, int *lengthp, int part, void *tg_cookie)
5551 {
5552 	struct cmlb_lun	*cl;
5553 	diskaddr_t	capacity;
5554 	uint32_t	lbasize;
5555 	enum		dp { DP_NBLOCKS, DP_BLKSIZE } dp;
5556 	int		callers_length;
5557 	caddr_t		buffer;
5558 	uint64_t	nblocks64;
5559 	uint_t		dblk;
5560 
5561 	/* Always fallback to ddi_prop_op... */
5562 	cl = (struct cmlb_lun *)cmlbhandle;
5563 	if (cl == NULL) {
5564 fallback:	return (ddi_prop_op(dev, dip, prop_op, mod_flags,
5565 		    name, valuep, lengthp));
5566 	}
5567 
5568 	/* Pick up capacity and blocksize information. */
5569 	capacity = cl->cl_blockcount;
5570 	if (capacity == 0)
5571 		goto fallback;
5572 	lbasize = cl->cl_tgt_blocksize;
5573 	if (lbasize == 0)
5574 		lbasize = DEV_BSIZE;	/* 0 -> DEV_BSIZE units */
5575 
5576 	/* Check for dynamic property of whole device. */
5577 	if (dev == DDI_DEV_T_ANY) {
5578 		/* Fallback to ddi_prop_op if we don't understand.  */
5579 		if (strcmp(name, "device-nblocks") == 0)
5580 			dp = DP_NBLOCKS;
5581 		else if (strcmp(name, "device-blksize") == 0)
5582 			dp = DP_BLKSIZE;
5583 		else
5584 			goto fallback;
5585 
5586 		/* get callers length, establish length of our dynamic prop */
5587 		callers_length = *lengthp;
5588 		if (dp == DP_NBLOCKS)
5589 			*lengthp = sizeof (uint64_t);
5590 		else if (dp == DP_BLKSIZE)
5591 			*lengthp = sizeof (uint32_t);
5592 
5593 		/* service request for the length of the property */
5594 		if (prop_op == PROP_LEN)
5595 			return (DDI_PROP_SUCCESS);
5596 
5597 		switch (prop_op) {
5598 		case PROP_LEN_AND_VAL_ALLOC:
5599 			if ((buffer = kmem_alloc(*lengthp,
5600 			    (mod_flags & DDI_PROP_CANSLEEP) ?
5601 			    KM_SLEEP : KM_NOSLEEP)) == NULL)
5602 				return (DDI_PROP_NO_MEMORY);
5603 			*(caddr_t *)valuep = buffer;	/* set callers buf */
5604 			break;
5605 
5606 		case PROP_LEN_AND_VAL_BUF:
5607 			/* the length of the prop and the request must match */
5608 			if (callers_length != *lengthp)
5609 				return (DDI_PROP_INVAL_ARG);
5610 			buffer = valuep;		/* get callers buf */
5611 			break;
5612 
5613 		default:
5614 			return (DDI_PROP_INVAL_ARG);
5615 		}
5616 
5617 		/* transfer the value into the buffer */
5618 		if (dp == DP_NBLOCKS)
5619 			*((uint64_t *)buffer) = capacity;
5620 		else if (dp == DP_BLKSIZE)
5621 			*((uint32_t *)buffer) = lbasize;
5622 
5623 		return (DDI_PROP_SUCCESS);
5624 	}
5625 
5626 	/*
5627 	 * Support dynamic size oriented properties of partition. Requests
5628 	 * issued under conditions where size is valid are passed to
5629 	 * ddi_prop_op_nblocks with the size information, otherwise the
5630 	 * request is passed to ddi_prop_op. Size depends on valid geometry.
5631 	 */
5632 	if (!cmlb_is_valid(cmlbhandle))
5633 		goto fallback;
5634 
5635 	/* Get partition nblocks value. */
5636 	(void) cmlb_partinfo(cmlbhandle, part,
5637 	    (diskaddr_t *)&nblocks64, NULL, NULL, NULL, tg_cookie);
5638 
5639 	/*
5640 	 * Assume partition information is in sys_blocksize units, compute
5641 	 * divisor for size(9P) property representation.
5642 	 */
5643 	dblk = lbasize / cl->cl_sys_blocksize;
5644 
5645 	/* Now let ddi_prop_op_nblocks_blksize() handle the request. */
5646 	return (ddi_prop_op_nblocks_blksize(dev, dip, prop_op, mod_flags,
5647 	    name, valuep, lengthp, nblocks64 / dblk, lbasize));
5648 }
5649