xref: /titanic_44/usr/src/uts/common/io/cmlb.c (revision 5cd2e4b941a477f4673d418d1d6ab5984a910a72)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright 2012 DEY Storage Systems, Inc.  All rights reserved.
24  * Copyright 2010 Sun Microsystems, Inc.  All rights reserved.
25  * Use is subject to license terms.
26  */
27 
28 /*
29  * This module provides support for labeling operations for target
30  * drivers.
31  */
32 
33 #include <sys/scsi/scsi.h>
34 #include <sys/sunddi.h>
35 #include <sys/dklabel.h>
36 #include <sys/dkio.h>
37 #include <sys/vtoc.h>
38 #include <sys/dktp/fdisk.h>
39 #include <sys/vtrace.h>
40 #include <sys/efi_partition.h>
41 #include <sys/cmlb.h>
42 #include <sys/cmlb_impl.h>
43 #if defined(__i386) || defined(__amd64)
44 #include <sys/fs/dv_node.h>
45 #endif
46 #include <sys/ddi_impldefs.h>
47 
48 /*
49  * Driver minor node structure and data table
50  */
51 struct driver_minor_data {
52 	char	*name;
53 	minor_t	minor;
54 	int	type;
55 };
56 
57 static struct driver_minor_data dk_minor_data[] = {
58 	{"a", 0, S_IFBLK},
59 	{"b", 1, S_IFBLK},
60 	{"c", 2, S_IFBLK},
61 	{"d", 3, S_IFBLK},
62 	{"e", 4, S_IFBLK},
63 	{"f", 5, S_IFBLK},
64 	{"g", 6, S_IFBLK},
65 	{"h", 7, S_IFBLK},
66 #if defined(_SUNOS_VTOC_16)
67 	{"i", 8, S_IFBLK},
68 	{"j", 9, S_IFBLK},
69 	{"k", 10, S_IFBLK},
70 	{"l", 11, S_IFBLK},
71 	{"m", 12, S_IFBLK},
72 	{"n", 13, S_IFBLK},
73 	{"o", 14, S_IFBLK},
74 	{"p", 15, S_IFBLK},
75 #endif			/* defined(_SUNOS_VTOC_16) */
76 #if defined(_FIRMWARE_NEEDS_FDISK)
77 	{"q", 16, S_IFBLK},
78 	{"r", 17, S_IFBLK},
79 	{"s", 18, S_IFBLK},
80 	{"t", 19, S_IFBLK},
81 	{"u", 20, S_IFBLK},
82 #endif			/* defined(_FIRMWARE_NEEDS_FDISK) */
83 	{"a,raw", 0, S_IFCHR},
84 	{"b,raw", 1, S_IFCHR},
85 	{"c,raw", 2, S_IFCHR},
86 	{"d,raw", 3, S_IFCHR},
87 	{"e,raw", 4, S_IFCHR},
88 	{"f,raw", 5, S_IFCHR},
89 	{"g,raw", 6, S_IFCHR},
90 	{"h,raw", 7, S_IFCHR},
91 #if defined(_SUNOS_VTOC_16)
92 	{"i,raw", 8, S_IFCHR},
93 	{"j,raw", 9, S_IFCHR},
94 	{"k,raw", 10, S_IFCHR},
95 	{"l,raw", 11, S_IFCHR},
96 	{"m,raw", 12, S_IFCHR},
97 	{"n,raw", 13, S_IFCHR},
98 	{"o,raw", 14, S_IFCHR},
99 	{"p,raw", 15, S_IFCHR},
100 #endif			/* defined(_SUNOS_VTOC_16) */
101 #if defined(_FIRMWARE_NEEDS_FDISK)
102 	{"q,raw", 16, S_IFCHR},
103 	{"r,raw", 17, S_IFCHR},
104 	{"s,raw", 18, S_IFCHR},
105 	{"t,raw", 19, S_IFCHR},
106 	{"u,raw", 20, S_IFCHR},
107 #endif			/* defined(_FIRMWARE_NEEDS_FDISK) */
108 	{0}
109 };
110 
111 #if defined(__i386) || defined(__amd64)
112 #if defined(_FIRMWARE_NEEDS_FDISK)
113 static struct driver_minor_data dk_ext_minor_data[] = {
114 	{"p5", 21, S_IFBLK},
115 	{"p6", 22, S_IFBLK},
116 	{"p7", 23, S_IFBLK},
117 	{"p8", 24, S_IFBLK},
118 	{"p9", 25, S_IFBLK},
119 	{"p10", 26, S_IFBLK},
120 	{"p11", 27, S_IFBLK},
121 	{"p12", 28, S_IFBLK},
122 	{"p13", 29, S_IFBLK},
123 	{"p14", 30, S_IFBLK},
124 	{"p15", 31, S_IFBLK},
125 	{"p16", 32, S_IFBLK},
126 	{"p17", 33, S_IFBLK},
127 	{"p18", 34, S_IFBLK},
128 	{"p19", 35, S_IFBLK},
129 	{"p20", 36, S_IFBLK},
130 	{"p21", 37, S_IFBLK},
131 	{"p22", 38, S_IFBLK},
132 	{"p23", 39, S_IFBLK},
133 	{"p24", 40, S_IFBLK},
134 	{"p25", 41, S_IFBLK},
135 	{"p26", 42, S_IFBLK},
136 	{"p27", 43, S_IFBLK},
137 	{"p28", 44, S_IFBLK},
138 	{"p29", 45, S_IFBLK},
139 	{"p30", 46, S_IFBLK},
140 	{"p31", 47, S_IFBLK},
141 	{"p32", 48, S_IFBLK},
142 	{"p33", 49, S_IFBLK},
143 	{"p34", 50, S_IFBLK},
144 	{"p35", 51, S_IFBLK},
145 	{"p36", 52, S_IFBLK},
146 	{"p5,raw", 21, S_IFCHR},
147 	{"p6,raw", 22, S_IFCHR},
148 	{"p7,raw", 23, S_IFCHR},
149 	{"p8,raw", 24, S_IFCHR},
150 	{"p9,raw", 25, S_IFCHR},
151 	{"p10,raw", 26, S_IFCHR},
152 	{"p11,raw", 27, S_IFCHR},
153 	{"p12,raw", 28, S_IFCHR},
154 	{"p13,raw", 29, S_IFCHR},
155 	{"p14,raw", 30, S_IFCHR},
156 	{"p15,raw", 31, S_IFCHR},
157 	{"p16,raw", 32, S_IFCHR},
158 	{"p17,raw", 33, S_IFCHR},
159 	{"p18,raw", 34, S_IFCHR},
160 	{"p19,raw", 35, S_IFCHR},
161 	{"p20,raw", 36, S_IFCHR},
162 	{"p21,raw", 37, S_IFCHR},
163 	{"p22,raw", 38, S_IFCHR},
164 	{"p23,raw", 39, S_IFCHR},
165 	{"p24,raw", 40, S_IFCHR},
166 	{"p25,raw", 41, S_IFCHR},
167 	{"p26,raw", 42, S_IFCHR},
168 	{"p27,raw", 43, S_IFCHR},
169 	{"p28,raw", 44, S_IFCHR},
170 	{"p29,raw", 45, S_IFCHR},
171 	{"p30,raw", 46, S_IFCHR},
172 	{"p31,raw", 47, S_IFCHR},
173 	{"p32,raw", 48, S_IFCHR},
174 	{"p33,raw", 49, S_IFCHR},
175 	{"p34,raw", 50, S_IFCHR},
176 	{"p35,raw", 51, S_IFCHR},
177 	{"p36,raw", 52, S_IFCHR},
178 	{0}
179 };
180 #endif			/* defined(_FIRMWARE_NEEDS_FDISK) */
181 #endif			/* if defined(__i386) || defined(__amd64) */
182 
183 static struct driver_minor_data dk_minor_data_efi[] = {
184 	{"a", 0, S_IFBLK},
185 	{"b", 1, S_IFBLK},
186 	{"c", 2, S_IFBLK},
187 	{"d", 3, S_IFBLK},
188 	{"e", 4, S_IFBLK},
189 	{"f", 5, S_IFBLK},
190 	{"g", 6, S_IFBLK},
191 	{"wd", 7, S_IFBLK},
192 #if defined(_SUNOS_VTOC_16)
193 	{"i", 8, S_IFBLK},
194 	{"j", 9, S_IFBLK},
195 	{"k", 10, S_IFBLK},
196 	{"l", 11, S_IFBLK},
197 	{"m", 12, S_IFBLK},
198 	{"n", 13, S_IFBLK},
199 	{"o", 14, S_IFBLK},
200 	{"p", 15, S_IFBLK},
201 #endif			/* defined(_SUNOS_VTOC_16) */
202 #if defined(_FIRMWARE_NEEDS_FDISK)
203 	{"q", 16, S_IFBLK},
204 	{"r", 17, S_IFBLK},
205 	{"s", 18, S_IFBLK},
206 	{"t", 19, S_IFBLK},
207 	{"u", 20, S_IFBLK},
208 #endif			/* defined(_FIRMWARE_NEEDS_FDISK) */
209 	{"a,raw", 0, S_IFCHR},
210 	{"b,raw", 1, S_IFCHR},
211 	{"c,raw", 2, S_IFCHR},
212 	{"d,raw", 3, S_IFCHR},
213 	{"e,raw", 4, S_IFCHR},
214 	{"f,raw", 5, S_IFCHR},
215 	{"g,raw", 6, S_IFCHR},
216 	{"wd,raw", 7, S_IFCHR},
217 #if defined(_SUNOS_VTOC_16)
218 	{"i,raw", 8, S_IFCHR},
219 	{"j,raw", 9, S_IFCHR},
220 	{"k,raw", 10, S_IFCHR},
221 	{"l,raw", 11, S_IFCHR},
222 	{"m,raw", 12, S_IFCHR},
223 	{"n,raw", 13, S_IFCHR},
224 	{"o,raw", 14, S_IFCHR},
225 	{"p,raw", 15, S_IFCHR},
226 #endif			/* defined(_SUNOS_VTOC_16) */
227 #if defined(_FIRMWARE_NEEDS_FDISK)
228 	{"q,raw", 16, S_IFCHR},
229 	{"r,raw", 17, S_IFCHR},
230 	{"s,raw", 18, S_IFCHR},
231 	{"t,raw", 19, S_IFCHR},
232 	{"u,raw", 20, S_IFCHR},
233 #endif			/* defined(_FIRMWARE_NEEDS_FDISK) */
234 	{0}
235 };
236 
237 /*
238  * Declare the dynamic properties implemented in prop_op(9E) implementation
239  * that we want to have show up in a di_init(3DEVINFO) device tree snapshot
240  * of drivers that call cmlb_attach().
241  */
242 static i_ddi_prop_dyn_t cmlb_prop_dyn[] = {
243 	{"Nblocks",		DDI_PROP_TYPE_INT64,	S_IFBLK},
244 	{"Size",		DDI_PROP_TYPE_INT64,	S_IFCHR},
245 	{"device-nblocks",	DDI_PROP_TYPE_INT64},
246 	{"device-blksize",	DDI_PROP_TYPE_INT},
247 	{"device-solid-state",	DDI_PROP_TYPE_INT},
248 	{"device-rotational",	DDI_PROP_TYPE_INT},
249 	{NULL}
250 };
251 
252 /*
253  * This implies an upper limit of 8192 GPT partitions
254  * in one transfer for GUID Partition Entry Array.
255  */
256 len_t cmlb_tg_max_efi_xfer = 1024 * 1024;
257 
258 /*
259  * External kernel interfaces
260  */
261 extern struct mod_ops mod_miscops;
262 
263 extern int ddi_create_internal_pathname(dev_info_t *dip, char *name,
264     int spec_type, minor_t minor_num);
265 
266 /*
267  * Global buffer and mutex for debug logging
268  */
269 static char	cmlb_log_buffer[1024];
270 static kmutex_t	cmlb_log_mutex;
271 
272 
273 struct cmlb_lun *cmlb_debug_cl = NULL;
274 uint_t cmlb_level_mask = 0x0;
275 
276 int cmlb_rot_delay = 4;	/* default rotational delay */
277 
278 static struct modlmisc modlmisc = {
279 	&mod_miscops,   /* Type of module */
280 	"Common Labeling module"
281 };
282 
283 static struct modlinkage modlinkage = {
284 	MODREV_1, (void *)&modlmisc, NULL
285 };
286 
287 /* Local function prototypes */
288 static dev_t cmlb_make_device(struct cmlb_lun *cl);
289 static int cmlb_validate_geometry(struct cmlb_lun *cl, boolean_t forcerevalid,
290     int flags, void *tg_cookie);
291 static void cmlb_resync_geom_caches(struct cmlb_lun *cl, diskaddr_t capacity,
292     void *tg_cookie);
293 static int cmlb_read_fdisk(struct cmlb_lun *cl, diskaddr_t capacity,
294     void *tg_cookie);
295 static void cmlb_swap_efi_gpt(efi_gpt_t *e);
296 static void cmlb_swap_efi_gpe(int nparts, efi_gpe_t *p);
297 static int cmlb_validate_efi(efi_gpt_t *labp);
298 static int cmlb_use_efi(struct cmlb_lun *cl, diskaddr_t capacity, int flags,
299     void *tg_cookie);
300 static void cmlb_build_default_label(struct cmlb_lun *cl, void *tg_cookie);
301 static int  cmlb_uselabel(struct cmlb_lun *cl,  struct dk_label *l, int flags);
302 #if defined(_SUNOS_VTOC_8)
303 static void cmlb_build_user_vtoc(struct cmlb_lun *cl, struct vtoc *user_vtoc);
304 #endif
305 static int cmlb_build_label_vtoc(struct cmlb_lun *cl, struct vtoc *user_vtoc);
306 static int cmlb_write_label(struct cmlb_lun *cl, void *tg_cookie);
307 static int cmlb_set_vtoc(struct cmlb_lun *cl, struct dk_label *dkl,
308     void *tg_cookie);
309 static void cmlb_clear_efi(struct cmlb_lun *cl, void *tg_cookie);
310 static void cmlb_clear_vtoc(struct cmlb_lun *cl, void *tg_cookie);
311 static void cmlb_setup_default_geometry(struct cmlb_lun *cl, void *tg_cookie);
312 static int cmlb_create_minor_nodes(struct cmlb_lun *cl);
313 static int cmlb_check_update_blockcount(struct cmlb_lun *cl, void *tg_cookie);
314 static boolean_t cmlb_check_efi_mbr(uchar_t *buf, boolean_t *is_mbr);
315 
316 #if defined(__i386) || defined(__amd64)
317 static int cmlb_update_fdisk_and_vtoc(struct cmlb_lun *cl, void *tg_cookie);
318 #endif
319 
320 #if defined(_FIRMWARE_NEEDS_FDISK)
321 static boolean_t  cmlb_has_max_chs_vals(struct ipart *fdp);
322 #endif
323 
324 #if defined(_SUNOS_VTOC_16)
325 static void cmlb_convert_geometry(struct cmlb_lun *cl, diskaddr_t capacity,
326     struct dk_geom *cl_g, void *tg_cookie);
327 #endif
328 
329 static int cmlb_dkio_get_geometry(struct cmlb_lun *cl, caddr_t arg, int flag,
330     void *tg_cookie);
331 static int cmlb_dkio_set_geometry(struct cmlb_lun *cl, caddr_t arg, int flag);
332 static int cmlb_dkio_get_partition(struct cmlb_lun *cl, caddr_t arg, int flag,
333     void *tg_cookie);
334 static int cmlb_dkio_set_partition(struct cmlb_lun *cl, caddr_t arg, int flag);
335 static int cmlb_dkio_get_efi(struct cmlb_lun *cl, caddr_t arg, int flag,
336     void *tg_cookie);
337 static int cmlb_dkio_set_efi(struct cmlb_lun *cl, dev_t dev, caddr_t arg,
338     int flag, void *tg_cookie);
339 static int cmlb_dkio_get_vtoc(struct cmlb_lun *cl, caddr_t arg, int flag,
340     void *tg_cookie);
341 static int cmlb_dkio_get_extvtoc(struct cmlb_lun *cl, caddr_t arg, int flag,
342     void *tg_cookie);
343 static int cmlb_dkio_set_vtoc(struct cmlb_lun *cl, dev_t dev, caddr_t arg,
344     int flag, void *tg_cookie);
345 static int cmlb_dkio_set_extvtoc(struct cmlb_lun *cl, dev_t dev, caddr_t arg,
346     int flag, void *tg_cookie);
347 static int cmlb_dkio_get_mboot(struct cmlb_lun *cl, caddr_t arg, int flag,
348     void *tg_cookie);
349 static int cmlb_dkio_set_mboot(struct cmlb_lun *cl, caddr_t arg, int flag,
350     void *tg_cookie);
351 static int cmlb_dkio_partition(struct cmlb_lun *cl, caddr_t arg, int flag,
352     void *tg_cookie);
353 
354 #if defined(__i386) || defined(__amd64)
355 static int cmlb_dkio_set_ext_part(struct cmlb_lun *cl, caddr_t arg, int flag,
356     void *tg_cookie);
357 static int cmlb_validate_ext_part(struct cmlb_lun *cl, int part, int epart,
358     uint32_t start, uint32_t size);
359 static int cmlb_is_linux_swap(struct cmlb_lun *cl, uint32_t part_start,
360     void *tg_cookie);
361 static int cmlb_dkio_get_virtgeom(struct cmlb_lun *cl, caddr_t arg, int flag);
362 static int cmlb_dkio_get_phygeom(struct cmlb_lun *cl, caddr_t  arg, int flag,
363     void *tg_cookie);
364 static int cmlb_dkio_partinfo(struct cmlb_lun *cl, dev_t dev, caddr_t arg,
365     int flag);
366 static int cmlb_dkio_extpartinfo(struct cmlb_lun *cl, dev_t dev, caddr_t arg,
367     int flag);
368 #endif
369 
370 static void cmlb_dbg(uint_t comp, struct cmlb_lun *cl, const char *fmt, ...);
371 static void cmlb_v_log(dev_info_t *dev, const char *label, uint_t level,
372     const char *fmt, va_list ap);
373 static void cmlb_log(dev_info_t *dev, const char *label, uint_t level,
374     const char *fmt, ...);
375 
376 int
_init(void)377 _init(void)
378 {
379 	mutex_init(&cmlb_log_mutex, NULL, MUTEX_DRIVER, NULL);
380 	return (mod_install(&modlinkage));
381 }
382 
383 int
_info(struct modinfo * modinfop)384 _info(struct modinfo *modinfop)
385 {
386 	return (mod_info(&modlinkage, modinfop));
387 }
388 
389 int
_fini(void)390 _fini(void)
391 {
392 	int err;
393 
394 	if ((err = mod_remove(&modlinkage)) != 0) {
395 		return (err);
396 	}
397 
398 	mutex_destroy(&cmlb_log_mutex);
399 	return (err);
400 }
401 
402 /*
403  * cmlb_dbg is used for debugging to log additional info
404  * Level of output is controlled via cmlb_level_mask setting.
405  */
406 static void
cmlb_dbg(uint_t comp,struct cmlb_lun * cl,const char * fmt,...)407 cmlb_dbg(uint_t comp, struct cmlb_lun *cl, const char *fmt, ...)
408 {
409 	va_list		ap;
410 	dev_info_t	*dev;
411 	uint_t		level_mask = 0;
412 
413 	ASSERT(cl != NULL);
414 	dev = CMLB_DEVINFO(cl);
415 	ASSERT(dev != NULL);
416 	/*
417 	 * Filter messages based on the global component and level masks,
418 	 * also print if cl matches the value of cmlb_debug_cl, or if
419 	 * cmlb_debug_cl is set to NULL.
420 	 */
421 	if (comp & CMLB_TRACE)
422 		level_mask |= CMLB_LOGMASK_TRACE;
423 
424 	if (comp & CMLB_INFO)
425 		level_mask |= CMLB_LOGMASK_INFO;
426 
427 	if (comp & CMLB_ERROR)
428 		level_mask |= CMLB_LOGMASK_ERROR;
429 
430 	if ((cmlb_level_mask & level_mask) &&
431 	    ((cmlb_debug_cl == NULL) || (cmlb_debug_cl == cl))) {
432 		va_start(ap, fmt);
433 		cmlb_v_log(dev, CMLB_LABEL(cl), CE_CONT, fmt, ap);
434 		va_end(ap);
435 	}
436 }
437 
438 /*
439  * cmlb_log is basically a duplicate of scsi_log. It is redefined here
440  * so that this module does not depend on scsi module.
441  */
442 static void
cmlb_log(dev_info_t * dev,const char * label,uint_t level,const char * fmt,...)443 cmlb_log(dev_info_t *dev, const char *label, uint_t level, const char *fmt, ...)
444 {
445 	va_list		ap;
446 
447 	va_start(ap, fmt);
448 	cmlb_v_log(dev, label, level, fmt, ap);
449 	va_end(ap);
450 }
451 
452 static void
cmlb_v_log(dev_info_t * dev,const char * label,uint_t level,const char * fmt,va_list ap)453 cmlb_v_log(dev_info_t *dev, const char *label, uint_t level, const char *fmt,
454     va_list ap)
455 {
456 	static char 	name[256];
457 	int 		log_only = 0;
458 	int 		boot_only = 0;
459 	int 		console_only = 0;
460 
461 	mutex_enter(&cmlb_log_mutex);
462 
463 	if (dev) {
464 		if (level == CE_PANIC || level == CE_WARN ||
465 		    level == CE_NOTE) {
466 			(void) sprintf(name, "%s (%s%d):\n",
467 			    ddi_pathname(dev, cmlb_log_buffer),
468 			    label, ddi_get_instance(dev));
469 		} else {
470 			name[0] = '\0';
471 		}
472 	} else {
473 		(void) sprintf(name, "%s:", label);
474 	}
475 
476 	(void) vsprintf(cmlb_log_buffer, fmt, ap);
477 
478 	switch (cmlb_log_buffer[0]) {
479 	case '!':
480 		log_only = 1;
481 		break;
482 	case '?':
483 		boot_only = 1;
484 		break;
485 	case '^':
486 		console_only = 1;
487 		break;
488 	}
489 
490 	switch (level) {
491 	case CE_NOTE:
492 		level = CE_CONT;
493 		/* FALLTHROUGH */
494 	case CE_CONT:
495 	case CE_WARN:
496 	case CE_PANIC:
497 		if (boot_only) {
498 			cmn_err(level, "?%s\t%s", name, &cmlb_log_buffer[1]);
499 		} else if (console_only) {
500 			cmn_err(level, "^%s\t%s", name, &cmlb_log_buffer[1]);
501 		} else if (log_only) {
502 			cmn_err(level, "!%s\t%s", name, &cmlb_log_buffer[1]);
503 		} else {
504 			cmn_err(level, "%s\t%s", name, cmlb_log_buffer);
505 		}
506 		break;
507 	case CE_IGNORE:
508 		break;
509 	default:
510 		cmn_err(CE_CONT, "^DEBUG: %s\t%s", name, cmlb_log_buffer);
511 		break;
512 	}
513 	mutex_exit(&cmlb_log_mutex);
514 }
515 
516 
517 /*
518  * cmlb_alloc_handle:
519  *
520  *	Allocates a handle.
521  *
522  * Arguments:
523  *	cmlbhandlep	pointer to handle
524  *
525  * Notes:
526  *	Allocates a handle and stores the allocated handle in the area
527  *	pointed to by cmlbhandlep
528  *
529  * Context:
530  *	Kernel thread only (can sleep).
531  */
532 void
cmlb_alloc_handle(cmlb_handle_t * cmlbhandlep)533 cmlb_alloc_handle(cmlb_handle_t *cmlbhandlep)
534 {
535 	struct cmlb_lun 	*cl;
536 
537 	cl = kmem_zalloc(sizeof (struct cmlb_lun), KM_SLEEP);
538 	ASSERT(cmlbhandlep != NULL);
539 
540 	cl->cl_state = CMLB_INITED;
541 	cl->cl_def_labeltype = CMLB_LABEL_UNDEF;
542 	mutex_init(CMLB_MUTEX(cl), NULL, MUTEX_DRIVER, NULL);
543 
544 	*cmlbhandlep = (cmlb_handle_t)(cl);
545 }
546 
547 /*
548  * cmlb_free_handle
549  *
550  *	Frees handle.
551  *
552  * Arguments:
553  *	cmlbhandlep	pointer to handle
554  */
555 void
cmlb_free_handle(cmlb_handle_t * cmlbhandlep)556 cmlb_free_handle(cmlb_handle_t *cmlbhandlep)
557 {
558 	struct cmlb_lun 	*cl;
559 
560 	cl = (struct cmlb_lun *)*cmlbhandlep;
561 	if (cl != NULL) {
562 		mutex_destroy(CMLB_MUTEX(cl));
563 		kmem_free(cl, sizeof (struct cmlb_lun));
564 	}
565 
566 }
567 
568 /*
569  * cmlb_attach:
570  *
571  *	Attach handle to device, create minor nodes for device.
572  *
573  * Arguments:
574  * 	devi		pointer to device's dev_info structure.
575  * 	tgopsp		pointer to array of functions cmlb can use to callback
576  *			to target driver.
577  *
578  *	device_type	Peripheral device type as defined in
579  *			scsi/generic/inquiry.h
580  *
581  *	is_removable	whether or not device is removable.
582  *
583  *	is_hotpluggable	whether or not device is hotpluggable.
584  *
585  *	node_type	minor node type (as used by ddi_create_minor_node)
586  *
587  *	alter_behavior
588  *			bit flags:
589  *
590  *			CMLB_CREATE_ALTSLICE_VTOC_16_DTYPE_DIRECT: create
591  *			an alternate slice for the default label, if
592  *			device type is DTYPE_DIRECT an architectures default
593  *			label type is VTOC16.
594  *			Otherwise alternate slice will no be created.
595  *
596  *
597  *			CMLB_FAKE_GEOM_LABEL_IOCTLS_VTOC8: report a default
598  *			geometry and label for DKIOCGGEOM and DKIOCGVTOC
599  *			on architecture with VTOC8 label types.
600  *
601  * 			CMLB_OFF_BY_ONE: do the workaround for legacy off-by-
602  *                      one bug in obtaining capacity (in sd):
603  *			SCSI READ_CAPACITY command returns the LBA number of the
604  *			last logical block, but sd once treated this number as
605  *			disks' capacity on x86 platform. And LBAs are addressed
606  *			based 0. So the last block was lost on x86 platform.
607  *
608  *			Now, we remove this workaround. In order for present sd
609  *			driver to work with disks which are labeled/partitioned
610  *			via previous sd, we add workaround as follows:
611  *
612  *			1) Locate backup EFI label: cmlb searches the next to
613  *			   last
614  *			   block for backup EFI label. If fails, it will
615  *			   turn to the last block for backup EFI label;
616  *
617  *			2) Clear backup EFI label: cmlb first search the last
618  *			   block for backup EFI label, and will search the
619  *			   next to last block only if failed for the last
620  *			   block.
621  *
622  *			3) Calculate geometry:refer to cmlb_convert_geometry()
623  *			   If capacity increasing by 1 causes disks' capacity
624  *			   to cross over the limits in geometry calculation,
625  *			   geometry info will change. This will raise an issue:
626  *			   In case that primary VTOC label is destroyed, format
627  *			   commandline can restore it via backup VTOC labels.
628  *			   And format locates backup VTOC labels by use of
629  *			   geometry. So changing geometry will
630  *			   prevent format from finding backup VTOC labels. To
631  *			   eliminate this side effect for compatibility,
632  *			   sd uses (capacity -1) to calculate geometry;
633  *
634  *			4) 1TB disks: some important data structures use
635  *			   32-bit signed long/int (for example, daddr_t),
636  *			   so that sd doesn't support a disk with capacity
637  *			   larger than 1TB on 32-bit platform. However,
638  *			   for exactly 1TB disk, it was treated as (1T - 512)B
639  *			   in the past, and could have valid Solaris
640  *			   partitions. To workaround this, if an exactly 1TB
641  *			   disk has Solaris fdisk partition, it will be allowed
642  *			   to work with sd.
643  *
644  *
645  *
646  *			CMLB_FAKE_LABEL_ONE_PARTITION: create s0 and s2 covering
647  *			the entire disk, if there is no valid partition info.
648  *			If there is a valid Solaris partition, s0 and s2 will
649  *			only cover the entire Solaris partition.
650  *
651  *
652  *	cmlbhandle	cmlb handle associated with device
653  *
654  *	tg_cookie	cookie from target driver to be passed back to target
655  *			driver when we call back to it through tg_ops.
656  *
657  * Notes:
658  *	Assumes a default label based on capacity for non-removable devices.
659  *	If capacity > 1TB, EFI is assumed otherwise VTOC (default VTOC
660  *	for the architecture).
661  *
662  *	For removable devices, default label type is assumed to be VTOC
663  *	type. Create minor nodes based on a default label type.
664  *	Label on the media is not validated.
665  *	minor number consists of:
666  *		if _SUNOS_VTOC_8 is defined
667  *			lowest 3 bits is taken as partition number
668  *			the rest is instance number
669  *		if _SUNOS_VTOC_16 is defined
670  *			lowest 6 bits is taken as partition number
671  *			the rest is instance number
672  *
673  *
674  * Return values:
675  *	0 	Success
676  * 	ENXIO 	creating minor nodes failed.
677  *	EINVAL  invalid arg, unsupported tg_ops version
678  */
679 int
cmlb_attach(dev_info_t * devi,cmlb_tg_ops_t * tgopsp,int device_type,boolean_t is_removable,boolean_t is_hotpluggable,char * node_type,int alter_behavior,cmlb_handle_t cmlbhandle,void * tg_cookie)680 cmlb_attach(dev_info_t *devi, cmlb_tg_ops_t *tgopsp, int device_type,
681     boolean_t is_removable, boolean_t is_hotpluggable, char *node_type,
682     int alter_behavior, cmlb_handle_t cmlbhandle, void *tg_cookie)
683 {
684 
685 	struct cmlb_lun	*cl = (struct cmlb_lun *)cmlbhandle;
686 	diskaddr_t	cap;
687 	int		status;
688 
689 	ASSERT(VALID_BOOLEAN(is_removable));
690 	ASSERT(VALID_BOOLEAN(is_hotpluggable));
691 
692 	if (tgopsp->tg_version < TG_DK_OPS_VERSION_1)
693 		return (EINVAL);
694 
695 	mutex_enter(CMLB_MUTEX(cl));
696 
697 	CMLB_DEVINFO(cl) = devi;
698 	cl->cmlb_tg_ops = tgopsp;
699 	cl->cl_device_type = device_type;
700 	cl->cl_is_removable = is_removable;
701 	cl->cl_is_hotpluggable = is_hotpluggable;
702 	cl->cl_node_type = node_type;
703 	cl->cl_sys_blocksize = DEV_BSIZE;
704 	cl->cl_f_geometry_is_valid = B_FALSE;
705 	cl->cl_def_labeltype = CMLB_LABEL_VTOC;
706 	cl->cl_alter_behavior = alter_behavior;
707 	cl->cl_reserved = -1;
708 	cl->cl_msglog_flag |= CMLB_ALLOW_2TB_WARN;
709 #if defined(__i386) || defined(__amd64)
710 	cl->cl_logical_drive_count = 0;
711 #endif
712 
713 	if (!is_removable) {
714 		mutex_exit(CMLB_MUTEX(cl));
715 		status = DK_TG_GETCAP(cl, &cap, tg_cookie);
716 		mutex_enter(CMLB_MUTEX(cl));
717 		if (status == 0 && cap > CMLB_EXTVTOC_LIMIT) {
718 			/* set default EFI if > 2TB */
719 			cl->cl_def_labeltype = CMLB_LABEL_EFI;
720 		}
721 	}
722 
723 	/* create minor nodes based on default label type */
724 	cl->cl_last_labeltype = CMLB_LABEL_UNDEF;
725 	cl->cl_cur_labeltype = CMLB_LABEL_UNDEF;
726 
727 	if (cmlb_create_minor_nodes(cl) != 0) {
728 		mutex_exit(CMLB_MUTEX(cl));
729 		return (ENXIO);
730 	}
731 
732 	/* Define the dynamic properties for devinfo spapshots. */
733 	i_ddi_prop_dyn_driver_set(CMLB_DEVINFO(cl), cmlb_prop_dyn);
734 
735 	cl->cl_state = CMLB_ATTACHED;
736 
737 	mutex_exit(CMLB_MUTEX(cl));
738 	return (0);
739 }
740 
741 /*
742  * cmlb_detach:
743  *
744  * Invalidate in-core labeling data and remove all minor nodes for
745  * the device associate with handle.
746  *
747  * Arguments:
748  *	cmlbhandle	cmlb handle associated with device.
749  *
750  *	tg_cookie	cookie from target driver to be passed back to target
751  *			driver when we call back to it through tg_ops.
752  *
753  */
754 /*ARGSUSED1*/
755 void
cmlb_detach(cmlb_handle_t cmlbhandle,void * tg_cookie)756 cmlb_detach(cmlb_handle_t cmlbhandle, void *tg_cookie)
757 {
758 	struct cmlb_lun *cl = (struct cmlb_lun *)cmlbhandle;
759 
760 	mutex_enter(CMLB_MUTEX(cl));
761 	cl->cl_def_labeltype = CMLB_LABEL_UNDEF;
762 	cl->cl_f_geometry_is_valid = B_FALSE;
763 	ddi_remove_minor_node(CMLB_DEVINFO(cl), NULL);
764 	i_ddi_prop_dyn_driver_set(CMLB_DEVINFO(cl), NULL);
765 	cl->cl_state = CMLB_INITED;
766 	mutex_exit(CMLB_MUTEX(cl));
767 }
768 
769 /*
770  * cmlb_validate:
771  *
772  *	Validates label.
773  *
774  * Arguments
775  *	cmlbhandle	cmlb handle associated with device.
776  *
777  *	flags		operation flags. used for verbosity control
778  *
779  *	tg_cookie	cookie from target driver to be passed back to target
780  *			driver when we call back to it through tg_ops.
781  *
782  *
783  * Notes:
784  *	If new label type is different from the current, adjust minor nodes
785  *	accordingly.
786  *
787  * Return values:
788  *	0		success
789  *			Note: having fdisk but no solaris partition is assumed
790  *			success.
791  *
792  *	ENOMEM		memory allocation failed
793  *	EIO		i/o errors during read or get capacity
794  * 	EACCESS		reservation conflicts
795  * 	EINVAL		label was corrupt, or no default label was assumed
796  *	ENXIO		invalid handle
797  */
798 int
cmlb_validate(cmlb_handle_t cmlbhandle,int flags,void * tg_cookie)799 cmlb_validate(cmlb_handle_t cmlbhandle, int flags, void *tg_cookie)
800 {
801 	struct cmlb_lun *cl = (struct cmlb_lun *)cmlbhandle;
802 	int 		rval;
803 	int  		ret = 0;
804 
805 	/*
806 	 * Temp work-around checking cl for NULL since there is a bug
807 	 * in sd_detach calling this routine from taskq_dispatch
808 	 * inited function.
809 	 */
810 	if (cl == NULL)
811 		return (ENXIO);
812 
813 	mutex_enter(CMLB_MUTEX(cl));
814 	if (cl->cl_state < CMLB_ATTACHED) {
815 		mutex_exit(CMLB_MUTEX(cl));
816 		return (ENXIO);
817 	}
818 
819 	rval = cmlb_validate_geometry((struct cmlb_lun *)cmlbhandle, B_TRUE,
820 	    flags, tg_cookie);
821 
822 	if (rval == ENOTSUP) {
823 		if (cl->cl_f_geometry_is_valid) {
824 			cl->cl_cur_labeltype = CMLB_LABEL_EFI;
825 			ret = 0;
826 		} else {
827 			ret = EINVAL;
828 		}
829 	} else {
830 		ret = rval;
831 		if (ret == 0)
832 			cl->cl_cur_labeltype = CMLB_LABEL_VTOC;
833 	}
834 
835 	if (ret == 0)
836 		(void) cmlb_create_minor_nodes(cl);
837 
838 	mutex_exit(CMLB_MUTEX(cl));
839 	return (ret);
840 }
841 
842 /*
843  * cmlb_invalidate:
844  *	Invalidate in core label data
845  *
846  * Arguments:
847  *	cmlbhandle	cmlb handle associated with device.
848  *	tg_cookie	cookie from target driver to be passed back to target
849  *			driver when we call back to it through tg_ops.
850  */
851 /*ARGSUSED1*/
852 void
cmlb_invalidate(cmlb_handle_t cmlbhandle,void * tg_cookie)853 cmlb_invalidate(cmlb_handle_t cmlbhandle, void *tg_cookie)
854 {
855 	struct cmlb_lun *cl = (struct cmlb_lun *)cmlbhandle;
856 
857 	if (cl == NULL)
858 		return;
859 
860 	mutex_enter(CMLB_MUTEX(cl));
861 	cl->cl_f_geometry_is_valid = B_FALSE;
862 	mutex_exit(CMLB_MUTEX(cl));
863 }
864 
865 /*
866  * cmlb_is_valid
867  * 	Get status on whether the incore label/geom data is valid
868  *
869  * Arguments:
870  *	cmlbhandle      cmlb handle associated with device.
871  *
872  * Return values:
873  *	B_TRUE if incore label/geom data is valid.
874  *	B_FALSE otherwise.
875  *
876  */
877 
878 
879 boolean_t
cmlb_is_valid(cmlb_handle_t cmlbhandle)880 cmlb_is_valid(cmlb_handle_t cmlbhandle)
881 {
882 	struct cmlb_lun *cl = (struct cmlb_lun *)cmlbhandle;
883 
884 	if (cmlbhandle == NULL)
885 		return (B_FALSE);
886 
887 	return (cl->cl_f_geometry_is_valid);
888 
889 }
890 
891 
892 
893 /*
894  * cmlb_close:
895  *
896  * Close the device, revert to a default label minor node for the device,
897  * if it is removable.
898  *
899  * Arguments:
900  *	cmlbhandle	cmlb handle associated with device.
901  *
902  *	tg_cookie	cookie from target driver to be passed back to target
903  *			driver when we call back to it through tg_ops.
904  * Return values:
905  *	0	Success
906  * 	ENXIO	Re-creating minor node failed.
907  */
908 /*ARGSUSED1*/
909 int
cmlb_close(cmlb_handle_t cmlbhandle,void * tg_cookie)910 cmlb_close(cmlb_handle_t cmlbhandle, void *tg_cookie)
911 {
912 	struct cmlb_lun *cl = (struct cmlb_lun *)cmlbhandle;
913 
914 	mutex_enter(CMLB_MUTEX(cl));
915 	cl->cl_f_geometry_is_valid = B_FALSE;
916 
917 	/* revert to default minor node for this device */
918 	if (ISREMOVABLE(cl)) {
919 		cl->cl_cur_labeltype = CMLB_LABEL_UNDEF;
920 		(void) cmlb_create_minor_nodes(cl);
921 	}
922 
923 	mutex_exit(CMLB_MUTEX(cl));
924 	return (0);
925 }
926 
927 /*
928  * cmlb_get_devid_block:
929  *	 get the block number where device id is stored.
930  *
931  * Arguments:
932  *	cmlbhandle	cmlb handle associated with device.
933  *	devidblockp	pointer to block number.
934  *	tg_cookie	cookie from target driver to be passed back to target
935  *			driver when we call back to it through tg_ops.
936  *
937  * Notes:
938  *	It stores the block number of device id in the area pointed to
939  *	by devidblockp.
940  * 	with the block number of device id.
941  *
942  * Return values:
943  *	0	success
944  *	EINVAL 	device id does not apply to current label type.
945  */
946 /*ARGSUSED2*/
947 int
cmlb_get_devid_block(cmlb_handle_t cmlbhandle,diskaddr_t * devidblockp,void * tg_cookie)948 cmlb_get_devid_block(cmlb_handle_t cmlbhandle, diskaddr_t *devidblockp,
949     void *tg_cookie)
950 {
951 	daddr_t			spc, blk, head, cyl;
952 	struct cmlb_lun *cl = (struct cmlb_lun *)cmlbhandle;
953 
954 	mutex_enter(CMLB_MUTEX(cl));
955 	if (cl->cl_state < CMLB_ATTACHED) {
956 		mutex_exit(CMLB_MUTEX(cl));
957 		return (EINVAL);
958 	}
959 
960 	if ((!cl->cl_f_geometry_is_valid) ||
961 	    (cl->cl_solaris_size < DK_LABEL_LOC)) {
962 		mutex_exit(CMLB_MUTEX(cl));
963 		return (EINVAL);
964 	}
965 
966 	if (cl->cl_cur_labeltype == CMLB_LABEL_EFI) {
967 		if (cl->cl_reserved != -1) {
968 			blk = cl->cl_map[cl->cl_reserved].dkl_cylno;
969 		} else {
970 			mutex_exit(CMLB_MUTEX(cl));
971 			return (EINVAL);
972 		}
973 	} else {
974 		/* if the disk is unlabeled, don't write a devid to it */
975 		if (cl->cl_label_from_media != CMLB_LABEL_VTOC) {
976 			mutex_exit(CMLB_MUTEX(cl));
977 			return (EINVAL);
978 		}
979 
980 		/* this geometry doesn't allow us to write a devid */
981 		if (cl->cl_g.dkg_acyl < 2) {
982 			mutex_exit(CMLB_MUTEX(cl));
983 			return (EINVAL);
984 		}
985 
986 		/*
987 		 * Subtract 2 guarantees that the next to last cylinder
988 		 * is used
989 		 */
990 		cyl  = cl->cl_g.dkg_ncyl  + cl->cl_g.dkg_acyl - 2;
991 		spc  = cl->cl_g.dkg_nhead * cl->cl_g.dkg_nsect;
992 		head = cl->cl_g.dkg_nhead - 1;
993 		blk  = cl->cl_solaris_offset +
994 		    (cyl * (spc - cl->cl_g.dkg_apc)) +
995 		    (head * cl->cl_g.dkg_nsect) + 1;
996 	}
997 
998 	*devidblockp = blk;
999 	mutex_exit(CMLB_MUTEX(cl));
1000 	return (0);
1001 }
1002 
1003 /*
1004  * cmlb_partinfo:
1005  *	Get partition info for specified partition number.
1006  *
1007  * Arguments:
1008  *	cmlbhandle	cmlb handle associated with device.
1009  *	part		partition number
1010  *	nblocksp	pointer to number of blocks
1011  *	startblockp	pointer to starting block
1012  *	partnamep	pointer to name of partition
1013  *	tagp		pointer to tag info
1014  *	tg_cookie	cookie from target driver to be passed back to target
1015  *			driver when we call back to it through tg_ops.
1016  *
1017  *
1018  * Notes:
1019  *	If in-core label is not valid, this functions tries to revalidate
1020  *	the label. If label is valid, it stores the total number of blocks
1021  *	in this partition in the area pointed to by nblocksp, starting
1022  *	block number in area pointed to by startblockp,  pointer to partition
1023  *	name in area pointed to by partnamep, and tag value in area
1024  *	pointed by tagp.
1025  *	For EFI labels, tag value will be set to 0.
1026  *
1027  *	For all nblocksp, startblockp and partnamep, tagp, a value of NULL
1028  *	indicates the corresponding info is not requested.
1029  *
1030  *
1031  * Return values:
1032  *	0	success
1033  *	EINVAL  no valid label or requested partition number is invalid.
1034  *
1035  */
1036 int
cmlb_partinfo(cmlb_handle_t cmlbhandle,int part,diskaddr_t * nblocksp,diskaddr_t * startblockp,char ** partnamep,uint16_t * tagp,void * tg_cookie)1037 cmlb_partinfo(cmlb_handle_t cmlbhandle, int part, diskaddr_t *nblocksp,
1038     diskaddr_t *startblockp, char **partnamep, uint16_t *tagp, void *tg_cookie)
1039 {
1040 
1041 	struct cmlb_lun *cl = (struct cmlb_lun *)cmlbhandle;
1042 	int rval;
1043 #if defined(__i386) || defined(__amd64)
1044 	int ext_part;
1045 #endif
1046 
1047 	ASSERT(cl != NULL);
1048 	mutex_enter(CMLB_MUTEX(cl));
1049 	if (cl->cl_state < CMLB_ATTACHED) {
1050 		mutex_exit(CMLB_MUTEX(cl));
1051 		return (EINVAL);
1052 	}
1053 
1054 	if (part  < 0 || part >= MAXPART) {
1055 		rval = EINVAL;
1056 	} else {
1057 		if (!cl->cl_f_geometry_is_valid)
1058 			(void) cmlb_validate_geometry((struct cmlb_lun *)cl,
1059 			    B_FALSE, 0, tg_cookie);
1060 
1061 #if defined(_SUNOS_VTOC_16)
1062 		if (((!cl->cl_f_geometry_is_valid) ||
1063 		    (part < NDKMAP && cl->cl_solaris_size == 0)) &&
1064 		    (part != P0_RAW_DISK)) {
1065 #else
1066 		if ((!cl->cl_f_geometry_is_valid) ||
1067 		    (part < NDKMAP && cl->cl_solaris_size == 0)) {
1068 #endif
1069 			rval = EINVAL;
1070 		} else {
1071 			if (startblockp != NULL)
1072 				*startblockp = (diskaddr_t)cl->cl_offset[part];
1073 
1074 			if (nblocksp != NULL)
1075 				*nblocksp = (diskaddr_t)
1076 				    cl->cl_map[part].dkl_nblk;
1077 
1078 			if (tagp != NULL)
1079 				*tagp =
1080 				    ((cl->cl_cur_labeltype == CMLB_LABEL_EFI) ||
1081 				    (part >= NDKMAP)) ? V_UNASSIGNED :
1082 				    cl->cl_vtoc.v_part[part].p_tag;
1083 			rval = 0;
1084 		}
1085 
1086 		/* consistent with behavior of sd for getting minor name */
1087 		if (partnamep != NULL) {
1088 #if defined(__i386) || defined(__amd64)
1089 #if defined(_FIRMWARE_NEEDS_FDISK)
1090 		if (part > FDISK_P4) {
1091 			ext_part = part-FDISK_P4-1;
1092 			*partnamep = dk_ext_minor_data[ext_part].name;
1093 		} else
1094 #endif
1095 #endif
1096 			*partnamep = dk_minor_data[part].name;
1097 		}
1098 
1099 	}
1100 
1101 	mutex_exit(CMLB_MUTEX(cl));
1102 	return (rval);
1103 }
1104 
1105 /*
1106  * cmlb_efi_label_capacity:
1107  *	Get capacity stored in EFI disk label.
1108  *
1109  * Arguments:
1110  *	cmlbhandle	cmlb handle associated with device.
1111  *	capacity	pointer to capacity stored in EFI disk label.
1112  *	tg_cookie	cookie from target driver to be passed back to target
1113  *			driver when we call back to it through tg_ops.
1114  *
1115  *
1116  * Notes:
1117  *	If in-core label is not valid, this functions tries to revalidate
1118  *	the label. If label is valid and is an EFI label, it stores the capacity
1119  *      in disk label in the area pointed to by capacity.
1120  *
1121  *
1122  * Return values:
1123  *	0	success
1124  *	EINVAL  no valid EFI label or capacity is NULL.
1125  *
1126  */
1127 int
1128 cmlb_efi_label_capacity(cmlb_handle_t cmlbhandle, diskaddr_t *capacity,
1129     void *tg_cookie)
1130 {
1131 	struct cmlb_lun *cl = (struct cmlb_lun *)cmlbhandle;
1132 	int rval;
1133 
1134 	ASSERT(cl != NULL);
1135 	mutex_enter(CMLB_MUTEX(cl));
1136 	if (cl->cl_state < CMLB_ATTACHED) {
1137 		mutex_exit(CMLB_MUTEX(cl));
1138 		return (EINVAL);
1139 	}
1140 
1141 	if (!cl->cl_f_geometry_is_valid)
1142 		(void) cmlb_validate_geometry((struct cmlb_lun *)cl, B_FALSE,
1143 		    0, tg_cookie);
1144 
1145 	if ((!cl->cl_f_geometry_is_valid) || (capacity == NULL) ||
1146 	    (cl->cl_cur_labeltype != CMLB_LABEL_EFI)) {
1147 		rval = EINVAL;
1148 	} else {
1149 		*capacity = (diskaddr_t)cl->cl_map[WD_NODE].dkl_nblk;
1150 		rval = 0;
1151 	}
1152 
1153 	mutex_exit(CMLB_MUTEX(cl));
1154 	return (rval);
1155 }
1156 
1157 /* Caller should make sure Test Unit Ready succeeds before calling this. */
1158 /*ARGSUSED*/
1159 int
1160 cmlb_ioctl(cmlb_handle_t cmlbhandle, dev_t dev, int cmd, intptr_t arg,
1161     int flag, cred_t *cred_p, int *rval_p, void *tg_cookie)
1162 {
1163 
1164 	int err;
1165 	struct cmlb_lun *cl;
1166 
1167 	cl = (struct cmlb_lun *)cmlbhandle;
1168 
1169 	ASSERT(cl != NULL);
1170 
1171 	mutex_enter(CMLB_MUTEX(cl));
1172 	if (cl->cl_state < CMLB_ATTACHED) {
1173 		mutex_exit(CMLB_MUTEX(cl));
1174 		return (EIO);
1175 	}
1176 
1177 	switch (cmd) {
1178 		case DKIOCSEXTVTOC:
1179 		case DKIOCSGEOM:
1180 		case DKIOCSETEFI:
1181 		case DKIOCSMBOOT:
1182 #if defined(__i386) || defined(__amd64)
1183 		case DKIOCSETEXTPART:
1184 #endif
1185 			break;
1186 		case DKIOCSVTOC:
1187 #if defined(__i386) || defined(__amd64)
1188 		case DKIOCPARTINFO:
1189 #endif
1190 			if (cl->cl_blockcount > CMLB_OLDVTOC_LIMIT) {
1191 				mutex_exit(CMLB_MUTEX(cl));
1192 				return (EOVERFLOW);
1193 			}
1194 			break;
1195 		default:
1196 			(void) cmlb_validate_geometry(cl, 1, CMLB_SILENT,
1197 			    tg_cookie);
1198 
1199 			switch (cmd) {
1200 			case DKIOCGVTOC:
1201 			case DKIOCGAPART:
1202 			case DKIOCSAPART:
1203 
1204 				if (cl->cl_label_from_media == CMLB_LABEL_EFI) {
1205 					/* GPT label on disk */
1206 					mutex_exit(CMLB_MUTEX(cl));
1207 					return (ENOTSUP);
1208 				} else if
1209 				    (cl->cl_blockcount > CMLB_OLDVTOC_LIMIT) {
1210 					mutex_exit(CMLB_MUTEX(cl));
1211 					return (EOVERFLOW);
1212 				}
1213 				break;
1214 
1215 			case DKIOCGGEOM:
1216 				if (cl->cl_label_from_media == CMLB_LABEL_EFI) {
1217 					/* GPT label on disk */
1218 					mutex_exit(CMLB_MUTEX(cl));
1219 					return (ENOTSUP);
1220 				}
1221 				break;
1222 			default:
1223 				break;
1224 			}
1225 	}
1226 
1227 	mutex_exit(CMLB_MUTEX(cl));
1228 
1229 	switch (cmd) {
1230 	case DKIOCGGEOM:
1231 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCGGEOM\n");
1232 		err = cmlb_dkio_get_geometry(cl, (caddr_t)arg, flag, tg_cookie);
1233 		break;
1234 
1235 	case DKIOCSGEOM:
1236 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCSGEOM\n");
1237 		err = cmlb_dkio_set_geometry(cl, (caddr_t)arg, flag);
1238 		break;
1239 
1240 	case DKIOCGAPART:
1241 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCGAPART\n");
1242 		err = cmlb_dkio_get_partition(cl, (caddr_t)arg,
1243 		    flag, tg_cookie);
1244 		break;
1245 
1246 	case DKIOCSAPART:
1247 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCSAPART\n");
1248 		err = cmlb_dkio_set_partition(cl, (caddr_t)arg, flag);
1249 		break;
1250 
1251 	case DKIOCGVTOC:
1252 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCGVTOC\n");
1253 		err = cmlb_dkio_get_vtoc(cl, (caddr_t)arg, flag, tg_cookie);
1254 		break;
1255 
1256 	case DKIOCGEXTVTOC:
1257 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCGVTOC\n");
1258 		err = cmlb_dkio_get_extvtoc(cl, (caddr_t)arg, flag, tg_cookie);
1259 		break;
1260 
1261 	case DKIOCGETEFI:
1262 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCGETEFI\n");
1263 		err = cmlb_dkio_get_efi(cl, (caddr_t)arg, flag, tg_cookie);
1264 		break;
1265 
1266 	case DKIOCPARTITION:
1267 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCPARTITION\n");
1268 		err = cmlb_dkio_partition(cl, (caddr_t)arg, flag, tg_cookie);
1269 		break;
1270 
1271 	case DKIOCSVTOC:
1272 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCSVTOC\n");
1273 		err = cmlb_dkio_set_vtoc(cl, dev, (caddr_t)arg, flag,
1274 		    tg_cookie);
1275 		break;
1276 
1277 	case DKIOCSEXTVTOC:
1278 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCSVTOC\n");
1279 		err = cmlb_dkio_set_extvtoc(cl, dev, (caddr_t)arg, flag,
1280 		    tg_cookie);
1281 		break;
1282 
1283 	case DKIOCSETEFI:
1284 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCSETEFI\n");
1285 		err = cmlb_dkio_set_efi(cl, dev, (caddr_t)arg, flag, tg_cookie);
1286 		break;
1287 
1288 	case DKIOCGMBOOT:
1289 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCGMBOOT\n");
1290 		err = cmlb_dkio_get_mboot(cl, (caddr_t)arg, flag, tg_cookie);
1291 		break;
1292 
1293 	case DKIOCSMBOOT:
1294 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCSMBOOT\n");
1295 		err = cmlb_dkio_set_mboot(cl, (caddr_t)arg, flag, tg_cookie);
1296 		break;
1297 	case DKIOCG_PHYGEOM:
1298 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCG_PHYGEOM\n");
1299 #if defined(__i386) || defined(__amd64)
1300 		err = cmlb_dkio_get_phygeom(cl, (caddr_t)arg, flag, tg_cookie);
1301 #else
1302 		err = ENOTTY;
1303 #endif
1304 		break;
1305 	case DKIOCG_VIRTGEOM:
1306 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCG_VIRTGEOM\n");
1307 #if defined(__i386) || defined(__amd64)
1308 		err = cmlb_dkio_get_virtgeom(cl, (caddr_t)arg, flag);
1309 #else
1310 		err = ENOTTY;
1311 #endif
1312 		break;
1313 	case DKIOCPARTINFO:
1314 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCPARTINFO");
1315 #if defined(__i386) || defined(__amd64)
1316 		err = cmlb_dkio_partinfo(cl, dev, (caddr_t)arg, flag);
1317 #else
1318 		err = ENOTTY;
1319 #endif
1320 		break;
1321 	case DKIOCEXTPARTINFO:
1322 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCPARTINFO");
1323 #if defined(__i386) || defined(__amd64)
1324 		err = cmlb_dkio_extpartinfo(cl, dev, (caddr_t)arg, flag);
1325 #else
1326 		err = ENOTTY;
1327 #endif
1328 		break;
1329 #if defined(__i386) || defined(__amd64)
1330 	case DKIOCSETEXTPART:
1331 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCSETEXTPART");
1332 		err = cmlb_dkio_set_ext_part(cl, (caddr_t)arg, flag, tg_cookie);
1333 		break;
1334 #endif
1335 	default:
1336 		err = ENOTTY;
1337 
1338 	}
1339 
1340 	/*
1341 	 * An ioctl that succeeds and changed ('set') size(9P) information
1342 	 * needs to invalidate the cached devinfo snapshot to avoid having
1343 	 * old information being returned in a snapshots.
1344 	 *
1345 	 * NB: When available, call ddi_change_minor_node() to clear
1346 	 * SSIZEVALID in specfs vnodes via spec_size_invalidate().
1347 	 */
1348 	if (err == 0) {
1349 		switch (cmd) {
1350 		case DKIOCSGEOM:
1351 		case DKIOCSAPART:
1352 		case DKIOCSVTOC:
1353 		case DKIOCSEXTVTOC:
1354 		case DKIOCSETEFI:
1355 			i_ddi_prop_dyn_cache_invalidate(CMLB_DEVINFO(cl),
1356 			    i_ddi_prop_dyn_driver_get(CMLB_DEVINFO(cl)));
1357 		}
1358 	}
1359 	return (err);
1360 }
1361 
1362 dev_t
1363 cmlb_make_device(struct cmlb_lun *cl)
1364 {
1365 	return (makedevice(ddi_driver_major(CMLB_DEVINFO(cl)),
1366 	    ddi_get_instance(CMLB_DEVINFO(cl)) << CMLBUNIT_SHIFT));
1367 }
1368 
1369 /*
1370  * Function: cmlb_check_update_blockcount
1371  *
1372  * Description: If current capacity value is invalid, obtains the
1373  *		current capacity from target driver.
1374  *
1375  * Return Code: 0	success
1376  *		EIO	failure
1377  */
1378 static int
1379 cmlb_check_update_blockcount(struct cmlb_lun *cl, void *tg_cookie)
1380 {
1381 	int status;
1382 	diskaddr_t capacity;
1383 	uint32_t lbasize;
1384 
1385 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
1386 
1387 	if (cl->cl_f_geometry_is_valid)
1388 		return (0);
1389 
1390 	mutex_exit(CMLB_MUTEX(cl));
1391 	status = DK_TG_GETCAP(cl, &capacity, tg_cookie);
1392 	if (status != 0) {
1393 		mutex_enter(CMLB_MUTEX(cl));
1394 		return (EIO);
1395 	}
1396 
1397 	status = DK_TG_GETBLOCKSIZE(cl, &lbasize, tg_cookie);
1398 	mutex_enter(CMLB_MUTEX(cl));
1399 	if (status != 0)
1400 		return (EIO);
1401 
1402 	if ((capacity != 0) && (lbasize != 0)) {
1403 		cl->cl_blockcount = capacity;
1404 		cl->cl_tgt_blocksize = lbasize;
1405 		if (!cl->cl_is_removable) {
1406 			cl->cl_sys_blocksize = lbasize;
1407 		}
1408 		return (0);
1409 	} else {
1410 		return (EIO);
1411 	}
1412 }
1413 
1414 static int
1415 cmlb_create_minor(dev_info_t *dip, char *name, int spec_type,
1416     minor_t minor_num, char *node_type, int flag, boolean_t internal)
1417 {
1418 	ASSERT(VALID_BOOLEAN(internal));
1419 
1420 	if (internal)
1421 		return (ddi_create_internal_pathname(dip,
1422 		    name, spec_type, minor_num));
1423 	else
1424 		return (ddi_create_minor_node(dip,
1425 		    name, spec_type, minor_num, node_type, flag));
1426 }
1427 
1428 /*
1429  *    Function: cmlb_create_minor_nodes
1430  *
1431  * Description: Create or adjust the minor device nodes for the instance.
1432  * 		Minor nodes are created based on default label type,
1433  *		current label type and last label type we created
1434  *		minor nodes based on.
1435  *
1436  *
1437  *   Arguments: cl - driver soft state (unit) structure
1438  *
1439  * Return Code: 0 success
1440  *		ENXIO	failure.
1441  *
1442  *     Context: Kernel thread context
1443  */
1444 static int
1445 cmlb_create_minor_nodes(struct cmlb_lun *cl)
1446 {
1447 	struct driver_minor_data	*dmdp;
1448 	int				instance;
1449 	char				name[48];
1450 	cmlb_label_t			newlabeltype;
1451 	boolean_t			internal;
1452 
1453 	ASSERT(cl != NULL);
1454 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
1455 
1456 	internal = VOID2BOOLEAN(
1457 	    (cl->cl_alter_behavior & (CMLB_INTERNAL_MINOR_NODES)) != 0);
1458 
1459 	/* check the most common case */
1460 	if (cl->cl_cur_labeltype != CMLB_LABEL_UNDEF &&
1461 	    cl->cl_last_labeltype == cl->cl_cur_labeltype) {
1462 		/* do nothing */
1463 		return (0);
1464 	}
1465 
1466 	if (cl->cl_def_labeltype == CMLB_LABEL_UNDEF) {
1467 		/* we should never get here */
1468 		return (ENXIO);
1469 	}
1470 
1471 	if (cl->cl_last_labeltype == CMLB_LABEL_UNDEF) {
1472 		/* first time during attach */
1473 		newlabeltype = cl->cl_def_labeltype;
1474 
1475 		instance = ddi_get_instance(CMLB_DEVINFO(cl));
1476 
1477 		/* Create all the minor nodes for this target. */
1478 		dmdp = (newlabeltype == CMLB_LABEL_EFI) ? dk_minor_data_efi :
1479 		    dk_minor_data;
1480 		while (dmdp->name != NULL) {
1481 
1482 			(void) sprintf(name, "%s", dmdp->name);
1483 
1484 			if (cmlb_create_minor(CMLB_DEVINFO(cl), name,
1485 			    dmdp->type,
1486 			    (instance << CMLBUNIT_SHIFT) | dmdp->minor,
1487 			    cl->cl_node_type, NULL, internal) == DDI_FAILURE) {
1488 				/*
1489 				 * Clean up any nodes that may have been
1490 				 * created, in case this fails in the middle
1491 				 * of the loop.
1492 				 */
1493 				ddi_remove_minor_node(CMLB_DEVINFO(cl), NULL);
1494 				return (ENXIO);
1495 			}
1496 			dmdp++;
1497 		}
1498 		cl->cl_last_labeltype = newlabeltype;
1499 		return (0);
1500 	}
1501 
1502 	/* Not first time  */
1503 	if (cl->cl_cur_labeltype == CMLB_LABEL_UNDEF) {
1504 		if (cl->cl_last_labeltype != cl->cl_def_labeltype) {
1505 			/* close time, revert to default. */
1506 			newlabeltype = cl->cl_def_labeltype;
1507 		} else {
1508 			/*
1509 			 * do nothing since the type for which we last created
1510 			 * nodes matches the default
1511 			 */
1512 			return (0);
1513 		}
1514 	} else {
1515 		if (cl->cl_cur_labeltype != cl->cl_last_labeltype) {
1516 			/* We are not closing, use current label type */
1517 			newlabeltype = cl->cl_cur_labeltype;
1518 		} else {
1519 			/*
1520 			 * do nothing since the type for which we last created
1521 			 * nodes matches the current label type
1522 			 */
1523 			return (0);
1524 		}
1525 	}
1526 
1527 	instance = ddi_get_instance(CMLB_DEVINFO(cl));
1528 
1529 	/*
1530 	 * Currently we only fix up the s7 node when we are switching
1531 	 * label types from or to EFI. This is consistent with
1532 	 * current behavior of sd.
1533 	 */
1534 	if (newlabeltype == CMLB_LABEL_EFI &&
1535 	    cl->cl_last_labeltype != CMLB_LABEL_EFI) {
1536 		/* from vtoc to EFI */
1537 		ddi_remove_minor_node(CMLB_DEVINFO(cl), "h");
1538 		ddi_remove_minor_node(CMLB_DEVINFO(cl), "h,raw");
1539 		(void) cmlb_create_minor(CMLB_DEVINFO(cl), "wd",
1540 		    S_IFBLK, (instance << CMLBUNIT_SHIFT) | WD_NODE,
1541 		    cl->cl_node_type, NULL, internal);
1542 		(void) cmlb_create_minor(CMLB_DEVINFO(cl), "wd,raw",
1543 		    S_IFCHR, (instance << CMLBUNIT_SHIFT) | WD_NODE,
1544 		    cl->cl_node_type, NULL, internal);
1545 	} else {
1546 		/* from efi to vtoc */
1547 		ddi_remove_minor_node(CMLB_DEVINFO(cl), "wd");
1548 		ddi_remove_minor_node(CMLB_DEVINFO(cl), "wd,raw");
1549 		(void) cmlb_create_minor(CMLB_DEVINFO(cl), "h",
1550 		    S_IFBLK, (instance << CMLBUNIT_SHIFT) | WD_NODE,
1551 		    cl->cl_node_type, NULL, internal);
1552 		(void) cmlb_create_minor(CMLB_DEVINFO(cl), "h,raw",
1553 		    S_IFCHR, (instance << CMLBUNIT_SHIFT) | WD_NODE,
1554 		    cl->cl_node_type, NULL, internal);
1555 	}
1556 
1557 	cl->cl_last_labeltype = newlabeltype;
1558 	return (0);
1559 }
1560 
1561 /*
1562  *    Function: cmlb_validate_geometry
1563  *
1564  * Description: Read the label from the disk (if present). Update the unit's
1565  *		geometry and vtoc information from the data in the label.
1566  *		Verify that the label is valid.
1567  *
1568  *   Arguments:
1569  *	cl		driver soft state (unit) structure
1570  *
1571  *	forcerevalid	force revalidation even if we are already valid.
1572  *	flags		operation flags from target driver. Used for verbosity
1573  *			control	at this time.
1574  *	tg_cookie	cookie from target driver to be passed back to target
1575  *			driver when we call back to it through tg_ops.
1576  *
1577  * Return Code: 0 - Successful completion
1578  *		EINVAL  - Invalid value in cl->cl_tgt_blocksize or
1579  *			  cl->cl_blockcount; or label on disk is corrupted
1580  *			  or unreadable.
1581  *		EACCES  - Reservation conflict at the device.
1582  *		ENOMEM  - Resource allocation error
1583  *		ENOTSUP - geometry not applicable
1584  *
1585  *     Context: Kernel thread only (can sleep).
1586  */
1587 static int
1588 cmlb_validate_geometry(struct cmlb_lun *cl, boolean_t forcerevalid, int flags,
1589     void *tg_cookie)
1590 {
1591 	int		label_error = 0;
1592 	diskaddr_t	capacity;
1593 	int		count;
1594 
1595 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
1596 	ASSERT(VALID_BOOLEAN(forcerevalid));
1597 
1598 	if ((cl->cl_f_geometry_is_valid) && (!forcerevalid)) {
1599 		if (cl->cl_cur_labeltype == CMLB_LABEL_EFI)
1600 			return (ENOTSUP);
1601 		return (0);
1602 	}
1603 
1604 	if (cmlb_check_update_blockcount(cl, tg_cookie) != 0)
1605 		return (EIO);
1606 
1607 	capacity = cl->cl_blockcount;
1608 
1609 #if defined(_SUNOS_VTOC_16)
1610 	/*
1611 	 * Set up the "whole disk" fdisk partition; this should always
1612 	 * exist, regardless of whether the disk contains an fdisk table
1613 	 * or vtoc.
1614 	 */
1615 	cl->cl_map[P0_RAW_DISK].dkl_cylno = 0;
1616 	cl->cl_offset[P0_RAW_DISK] = 0;
1617 	/*
1618 	 * note if capacity > int32_max(1TB) we are in 64bit environment
1619 	 * so no truncation happens
1620 	 */
1621 	cl->cl_map[P0_RAW_DISK].dkl_nblk  = capacity;
1622 #endif
1623 	/*
1624 	 * Refresh the logical and physical geometry caches.
1625 	 * (data from MODE SENSE format/rigid disk geometry pages,
1626 	 * and scsi_ifgetcap("geometry").
1627 	 */
1628 	cmlb_resync_geom_caches(cl, capacity, tg_cookie);
1629 
1630 	cl->cl_label_from_media = CMLB_LABEL_UNDEF;
1631 	label_error = cmlb_use_efi(cl, capacity, flags, tg_cookie);
1632 	if (label_error == 0) {
1633 
1634 		/* found a valid EFI label */
1635 		cmlb_dbg(CMLB_TRACE, cl,
1636 		    "cmlb_validate_geometry: found EFI label\n");
1637 		/*
1638 		 * solaris_size and geometry_is_valid are set in
1639 		 * cmlb_use_efi
1640 		 */
1641 		return (ENOTSUP);
1642 	}
1643 
1644 	/* NO EFI label found */
1645 
1646 	if (capacity > CMLB_EXTVTOC_LIMIT) {
1647 		if (label_error == ESRCH) {
1648 			/*
1649 			 * they've configured a LUN over 2TB, but used
1650 			 * format.dat to restrict format's view of the
1651 			 * capacity to be under 2TB in some earlier Solaris
1652 			 * release.
1653 			 */
1654 			/* i.e > 2TB with a VTOC < 2TB */
1655 			if (!(flags & CMLB_SILENT) &&
1656 			    (cl->cl_msglog_flag & CMLB_ALLOW_2TB_WARN)) {
1657 
1658 				cmlb_log(CMLB_DEVINFO(cl), CMLB_LABEL(cl),
1659 				    CE_NOTE, "!Disk (%s%d) is limited to 2 TB "
1660 				    "due to VTOC label. To use the full "
1661 				    "capacity of the disk, use format(1M) to "
1662 				    "relabel the disk with EFI/GPT label.\n",
1663 				    CMLB_LABEL(cl),
1664 				    ddi_get_instance(CMLB_DEVINFO(cl)));
1665 
1666 				cl->cl_msglog_flag &= ~CMLB_ALLOW_2TB_WARN;
1667 			}
1668 		} else {
1669 				return (ENOTSUP);
1670 		}
1671 	}
1672 
1673 	label_error = 0;
1674 
1675 	/*
1676 	 * at this point it is either labeled with a VTOC or it is
1677 	 * under 1TB (<= 1TB actually for off-by-1)
1678 	 */
1679 
1680 	/*
1681 	 * Only DIRECT ACCESS devices will have Scl labels.
1682 	 * CD's supposedly have a Scl label, too
1683 	 */
1684 	if (cl->cl_device_type == DTYPE_DIRECT || ISREMOVABLE(cl)) {
1685 		struct	dk_label *dkl;
1686 		offset_t label_addr;
1687 		int	rval;
1688 		size_t	buffer_size;
1689 
1690 		/*
1691 		 * Note: This will set up cl->cl_solaris_size and
1692 		 * cl->cl_solaris_offset.
1693 		 */
1694 		rval = cmlb_read_fdisk(cl, capacity, tg_cookie);
1695 		if ((rval != 0) && !ISCD(cl)) {
1696 			ASSERT(mutex_owned(CMLB_MUTEX(cl)));
1697 			return (rval);
1698 		}
1699 
1700 		if (cl->cl_solaris_size <= DK_LABEL_LOC) {
1701 			/*
1702 			 * Found fdisk table but no Solaris partition entry,
1703 			 * so don't call cmlb_uselabel() and don't create
1704 			 * a default label.
1705 			 */
1706 			label_error = 0;
1707 			cl->cl_f_geometry_is_valid = B_TRUE;
1708 			goto no_solaris_partition;
1709 		}
1710 
1711 		label_addr = (daddr_t)(cl->cl_solaris_offset + DK_LABEL_LOC);
1712 
1713 		buffer_size = cl->cl_sys_blocksize;
1714 
1715 		cmlb_dbg(CMLB_TRACE, cl, "cmlb_validate_geometry: "
1716 		    "label_addr: 0x%x allocation size: 0x%x\n",
1717 		    label_addr, buffer_size);
1718 
1719 		if ((dkl = kmem_zalloc(buffer_size, KM_NOSLEEP)) == NULL)
1720 			return (ENOMEM);
1721 
1722 		mutex_exit(CMLB_MUTEX(cl));
1723 		rval = DK_TG_READ(cl, dkl, label_addr, buffer_size, tg_cookie);
1724 		mutex_enter(CMLB_MUTEX(cl));
1725 
1726 		switch (rval) {
1727 		case 0:
1728 			/*
1729 			 * cmlb_uselabel will establish that the geometry
1730 			 * is valid.
1731 			 */
1732 			if (cmlb_uselabel(cl,
1733 			    (struct dk_label *)(uintptr_t)dkl, flags) !=
1734 			    CMLB_LABEL_IS_VALID) {
1735 				label_error = EINVAL;
1736 			} else
1737 				cl->cl_label_from_media = CMLB_LABEL_VTOC;
1738 			break;
1739 		case EACCES:
1740 			label_error = EACCES;
1741 			break;
1742 		default:
1743 			label_error = EINVAL;
1744 			break;
1745 		}
1746 
1747 		kmem_free(dkl, buffer_size);
1748 	}
1749 
1750 	/*
1751 	 * If a valid label was not found, AND if no reservation conflict
1752 	 * was detected, then go ahead and create a default label (4069506).
1753 	 *
1754 	 * Note: currently, for VTOC_8 devices, the default label is created
1755 	 * for removables and hotpluggables only.  For VTOC_16 devices, the
1756 	 * default label will be created for all devices.
1757 	 * (see cmlb_build_default_label)
1758 	 */
1759 #if defined(_SUNOS_VTOC_8)
1760 	if ((ISREMOVABLE(cl) || ISHOTPLUGGABLE(cl)) &&
1761 	    (label_error != EACCES)) {
1762 #elif defined(_SUNOS_VTOC_16)
1763 	if (label_error != EACCES) {
1764 #endif
1765 		if (!cl->cl_f_geometry_is_valid) {
1766 			cmlb_build_default_label(cl, tg_cookie);
1767 		}
1768 		label_error = 0;
1769 	}
1770 
1771 no_solaris_partition:
1772 
1773 #if defined(_SUNOS_VTOC_16)
1774 	/*
1775 	 * If we have valid geometry, set up the remaining fdisk partitions.
1776 	 * Note that dkl_cylno is not used for the fdisk map entries, so
1777 	 * we set it to an entirely bogus value.
1778 	 */
1779 	for (count = 0; count < FDISK_PARTS; count++) {
1780 		cl->cl_map[FDISK_P1 + count].dkl_cylno = UINT16_MAX;
1781 		cl->cl_map[FDISK_P1 + count].dkl_nblk =
1782 		    cl->cl_fmap[count].fmap_nblk;
1783 
1784 		cl->cl_offset[FDISK_P1 + count] =
1785 		    cl->cl_fmap[count].fmap_start;
1786 	}
1787 #endif
1788 
1789 	for (count = 0; count < NDKMAP; count++) {
1790 #if defined(_SUNOS_VTOC_8)
1791 		struct dk_map *lp  = &cl->cl_map[count];
1792 		cl->cl_offset[count] =
1793 		    cl->cl_g.dkg_nhead * cl->cl_g.dkg_nsect * lp->dkl_cylno;
1794 #elif defined(_SUNOS_VTOC_16)
1795 		struct dkl_partition *vp = &cl->cl_vtoc.v_part[count];
1796 
1797 		cl->cl_offset[count] = vp->p_start + cl->cl_solaris_offset;
1798 #else
1799 #error "No VTOC format defined."
1800 #endif
1801 	}
1802 
1803 	return (label_error);
1804 }
1805 
1806 #if defined(_SUNOS_VTOC_16)
1807 /*
1808  *    Function: cmlb_convert_geometry
1809  *
1810  * Description: Convert physical geometry into a dk_geom structure. In
1811  *		other words, make sure we don't wrap 16-bit values.
1812  *		e.g. converting from geom_cache to dk_geom
1813  *
1814  *     Context: Kernel thread only
1815  */
1816 static void
1817 cmlb_convert_geometry(struct cmlb_lun *cl, diskaddr_t capacity,
1818     struct dk_geom *cl_g, void *tg_cookie)
1819 {
1820 
1821 	ASSERT(cl != NULL);
1822 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
1823 
1824 	/* Unlabeled SCSI floppy device */
1825 	if (capacity < 160) {
1826 		/* Less than 80K */
1827 		cl_g->dkg_nhead = 1;
1828 		cl_g->dkg_ncyl = capacity;
1829 		cl_g->dkg_nsect = 1;
1830 		return;
1831 	} else if (capacity <= 0x1000) {
1832 		cl_g->dkg_nhead = 2;
1833 		cl_g->dkg_ncyl = 80;
1834 		cl_g->dkg_nsect = capacity / (cl_g->dkg_nhead * cl_g->dkg_ncyl);
1835 		return;
1836 	}
1837 
1838 	/*
1839 	 * For all devices we calculate cylinders using the heads and sectors
1840 	 * we assign based on capacity of the device.  The algorithm is
1841 	 * designed to be compatible with the way other operating systems
1842 	 * lay out fdisk tables for X86 and to insure that the cylinders never
1843 	 * exceed 65535 to prevent problems with X86 ioctls that report
1844 	 * geometry.
1845 	 * For some smaller disk sizes we report geometry that matches those
1846 	 * used by X86 BIOS usage. For larger disks, we use SPT that are
1847 	 * multiples of 63, since other OSes that are not limited to 16-bits
1848 	 * for cylinders stop at 63 SPT we make do by using multiples of 63 SPT.
1849 	 *
1850 	 * The following table (in order) illustrates some end result
1851 	 * calculations:
1852 	 *
1853 	 * Maximum number of blocks 		nhead	nsect
1854 	 *
1855 	 * 2097152 (1GB)			64	32
1856 	 * 16777216 (8GB)			128	32
1857 	 * 1052819775 (502.02GB)		255  	63
1858 	 * 2105639550 (0.98TB)			255	126
1859 	 * 3158459325 (1.47TB)			255  	189
1860 	 * 4211279100 (1.96TB)			255  	252
1861 	 * 5264098875 (2.45TB)			255  	315
1862 	 * ...
1863 	 *
1864 	 * For Solid State Drive(SSD), it uses 4K page size inside and may be
1865 	 * double with every new generation. If the I/O is not aligned with
1866 	 * page size on SSDs, SSDs perform a lot slower.
1867 	 * By default, Solaris partition starts from cylinder 1. It will be
1868 	 * misaligned even with 4K if using heads(255) and SPT(63). To
1869 	 * workaround the problem, if the device is SSD, we use heads(224) and
1870 	 * SPT multiple of 56. Thus the default Solaris partition starts from
1871 	 * a position that aligns with 128K on a 512 bytes sector size SSD.
1872 	 */
1873 
1874 	if (capacity <= 0x200000) {
1875 		cl_g->dkg_nhead = 64;
1876 		cl_g->dkg_nsect = 32;
1877 	} else if (capacity <= 0x01000000) {
1878 		cl_g->dkg_nhead = 128;
1879 		cl_g->dkg_nsect = 32;
1880 	} else {
1881 		tg_attribute_t tgattribute;
1882 		int is_solid_state;
1883 		unsigned short nhead;
1884 		unsigned short nsect;
1885 
1886 		bzero(&tgattribute, sizeof (tg_attribute_t));
1887 
1888 		mutex_exit(CMLB_MUTEX(cl));
1889 		is_solid_state =
1890 		    (DK_TG_GETATTRIBUTE(cl, &tgattribute, tg_cookie) == 0) ?
1891 		    tgattribute.media_is_solid_state : FALSE;
1892 		mutex_enter(CMLB_MUTEX(cl));
1893 
1894 		if (is_solid_state) {
1895 			nhead = 224;
1896 			nsect = 56;
1897 		} else {
1898 			nhead = 255;
1899 			nsect = 63;
1900 		}
1901 
1902 		cl_g->dkg_nhead = nhead;
1903 
1904 		/* make dkg_nsect be smallest multiple of nsect */
1905 		cl_g->dkg_nsect = ((capacity +
1906 		    (UINT16_MAX * nhead * nsect) - 1) /
1907 		    (UINT16_MAX * nhead * nsect)) * nsect;
1908 
1909 		if (cl_g->dkg_nsect == 0)
1910 			cl_g->dkg_nsect = (UINT16_MAX / nsect) * nsect;
1911 	}
1912 
1913 }
1914 #endif
1915 
1916 /*
1917  *    Function: cmlb_resync_geom_caches
1918  *
1919  * Description: (Re)initialize both geometry caches: the virtual geometry
1920  *            information is extracted from the HBA (the "geometry"
1921  *            capability), and the physical geometry cache data is
1922  *            generated by issuing MODE SENSE commands.
1923  *
1924  *   Arguments:
1925  *	cl 		driver soft state (unit) structure
1926  *	capacity	disk capacity in #blocks
1927  *	tg_cookie	cookie from target driver to be passed back to target
1928  *			driver when we call back to it through tg_ops.
1929  *
1930  *     Context: Kernel thread only (can sleep).
1931  */
1932 static void
1933 cmlb_resync_geom_caches(struct cmlb_lun *cl, diskaddr_t capacity,
1934     void *tg_cookie)
1935 {
1936 	struct	cmlb_geom 	pgeom;
1937 	struct	cmlb_geom	lgeom;
1938 	struct 	cmlb_geom	*pgeomp = &pgeom;
1939 	unsigned short 		nhead;
1940 	unsigned short 		nsect;
1941 	int 			spc;
1942 	int			ret;
1943 
1944 	ASSERT(cl != NULL);
1945 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
1946 
1947 	/*
1948 	 * Ask the controller for its logical geometry.
1949 	 * Note: if the HBA does not support scsi_ifgetcap("geometry"),
1950 	 * then the lgeom cache will be invalid.
1951 	 */
1952 	mutex_exit(CMLB_MUTEX(cl));
1953 	bzero(&lgeom, sizeof (struct cmlb_geom));
1954 	ret = DK_TG_GETVIRTGEOM(cl, &lgeom, tg_cookie);
1955 	mutex_enter(CMLB_MUTEX(cl));
1956 
1957 	bcopy(&lgeom, &cl->cl_lgeom, sizeof (cl->cl_lgeom));
1958 
1959 	/*
1960 	 * Initialize the pgeom cache from lgeom, so that if MODE SENSE
1961 	 * doesn't work, DKIOCG_PHYSGEOM can return reasonable values.
1962 	 */
1963 	if (ret != 0 || cl->cl_lgeom.g_nsect == 0 ||
1964 	    cl->cl_lgeom.g_nhead == 0) {
1965 		/*
1966 		 * Note: Perhaps this needs to be more adaptive? The rationale
1967 		 * is that, if there's no HBA geometry from the HBA driver, any
1968 		 * guess is good, since this is the physical geometry. If MODE
1969 		 * SENSE fails this gives a max cylinder size for non-LBA access
1970 		 */
1971 		nhead = 255;
1972 		nsect = 63;
1973 	} else {
1974 		nhead = cl->cl_lgeom.g_nhead;
1975 		nsect = cl->cl_lgeom.g_nsect;
1976 	}
1977 
1978 	if (ISCD(cl)) {
1979 		pgeomp->g_nhead = 1;
1980 		pgeomp->g_nsect = nsect * nhead;
1981 	} else {
1982 		pgeomp->g_nhead = nhead;
1983 		pgeomp->g_nsect = nsect;
1984 	}
1985 
1986 	spc = pgeomp->g_nhead * pgeomp->g_nsect;
1987 	pgeomp->g_capacity = capacity;
1988 	if (spc == 0)
1989 		pgeomp->g_ncyl = 0;
1990 	else
1991 		pgeomp->g_ncyl = pgeomp->g_capacity / spc;
1992 	pgeomp->g_acyl = 0;
1993 
1994 	/*
1995 	 * Retrieve fresh geometry data from the hardware, stash it
1996 	 * here temporarily before we rebuild the incore label.
1997 	 *
1998 	 * We want to use the MODE SENSE commands to derive the
1999 	 * physical geometry of the device, but if either command
2000 	 * fails, the logical geometry is used as the fallback for
2001 	 * disk label geometry.
2002 	 */
2003 
2004 	mutex_exit(CMLB_MUTEX(cl));
2005 	(void) DK_TG_GETPHYGEOM(cl, pgeomp, tg_cookie);
2006 	mutex_enter(CMLB_MUTEX(cl));
2007 
2008 	/*
2009 	 * Now update the real copy while holding the mutex. This
2010 	 * way the global copy is never in an inconsistent state.
2011 	 */
2012 	bcopy(pgeomp, &cl->cl_pgeom,  sizeof (cl->cl_pgeom));
2013 
2014 	cmlb_dbg(CMLB_INFO, cl, "cmlb_resync_geom_caches: "
2015 	    "(cached from lgeom)\n");
2016 	cmlb_dbg(CMLB_INFO,  cl,
2017 	    "   ncyl: %ld; acyl: %d; nhead: %d; nsect: %d\n",
2018 	    cl->cl_pgeom.g_ncyl, cl->cl_pgeom.g_acyl,
2019 	    cl->cl_pgeom.g_nhead, cl->cl_pgeom.g_nsect);
2020 	cmlb_dbg(CMLB_INFO,  cl, "   lbasize: %d; capacity: %ld; "
2021 	    "intrlv: %d; rpm: %d\n", cl->cl_pgeom.g_secsize,
2022 	    cl->cl_pgeom.g_capacity, cl->cl_pgeom.g_intrlv,
2023 	    cl->cl_pgeom.g_rpm);
2024 }
2025 
2026 
2027 #if defined(__i386) || defined(__amd64)
2028 /*
2029  *    Function: cmlb_update_ext_minor_nodes
2030  *
2031  * Description: Routine to add/remove extended partition device nodes
2032  *
2033  *   Arguments:
2034  *	cl		driver soft state (unit) structure
2035  *	num_parts	Number of logical drives found on the LUN
2036  *
2037  * Should be called with the mutex held
2038  *
2039  * Return Code: 0 for success
2040  *
2041  *     Context: User and Kernel thread
2042  *
2043  */
2044 static int
2045 cmlb_update_ext_minor_nodes(struct cmlb_lun *cl, int num_parts)
2046 {
2047 	int				i, count;
2048 	char				name[48];
2049 	int				instance;
2050 	struct driver_minor_data	*demdp, *demdpr;
2051 	char				*devnm;
2052 	dev_info_t			*pdip;
2053 	boolean_t 			internal;
2054 
2055 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
2056 	ASSERT(cl->cl_update_ext_minor_nodes == 1);
2057 
2058 	internal = VOID2BOOLEAN(
2059 	    (cl->cl_alter_behavior & (CMLB_INTERNAL_MINOR_NODES)) != 0);
2060 	instance = ddi_get_instance(CMLB_DEVINFO(cl));
2061 	demdp = dk_ext_minor_data;
2062 	demdpr = &dk_ext_minor_data[MAX_EXT_PARTS];
2063 
2064 
2065 	if (cl->cl_logical_drive_count) {
2066 		for (i = 0; i < cl->cl_logical_drive_count; i++) {
2067 			(void) sprintf(name, "%s", demdp->name);
2068 			ddi_remove_minor_node(CMLB_DEVINFO(cl), name);
2069 			(void) sprintf(name, "%s", demdpr->name);
2070 			ddi_remove_minor_node(CMLB_DEVINFO(cl), name);
2071 			demdp++;
2072 			demdpr++;
2073 		}
2074 		/* There are existing device nodes. Remove them */
2075 		devnm = kmem_alloc(MAXNAMELEN + 1, KM_SLEEP);
2076 		(void) ddi_deviname(cl->cl_devi, devnm);
2077 		pdip = ddi_get_parent(cl->cl_devi);
2078 		(void) devfs_clean(pdip, devnm + 1, DV_CLEAN_FORCE);
2079 		kmem_free(devnm, MAXNAMELEN + 1);
2080 	}
2081 
2082 	demdp = dk_ext_minor_data;
2083 	demdpr = &dk_ext_minor_data[MAX_EXT_PARTS];
2084 
2085 	for (i = 0; i < num_parts; i++) {
2086 		(void) sprintf(name, "%s", demdp->name);
2087 		if (cmlb_create_minor(CMLB_DEVINFO(cl), name,
2088 		    demdp->type,
2089 		    (instance << CMLBUNIT_SHIFT) | demdp->minor,
2090 		    cl->cl_node_type, NULL, internal) == DDI_FAILURE) {
2091 			/*
2092 			 * Clean up any nodes that may have been
2093 			 * created, in case this fails in the middle
2094 			 * of the loop.
2095 			 */
2096 			ddi_remove_minor_node(CMLB_DEVINFO(cl), NULL);
2097 			cl->cl_logical_drive_count = 0;
2098 			return (ENXIO);
2099 		}
2100 		(void) sprintf(name, "%s", demdpr->name);
2101 		if (ddi_create_minor_node(CMLB_DEVINFO(cl), name,
2102 		    demdpr->type,
2103 		    (instance << CMLBUNIT_SHIFT) | demdpr->minor,
2104 		    cl->cl_node_type, NULL) == DDI_FAILURE) {
2105 			/*
2106 			 * Clean up any nodes that may have been
2107 			 * created, in case this fails in the middle
2108 			 * of the loop.
2109 			 */
2110 			ddi_remove_minor_node(CMLB_DEVINFO(cl), NULL);
2111 			cl->cl_logical_drive_count = 0;
2112 			return (ENXIO);
2113 		}
2114 		demdp++;
2115 		demdpr++;
2116 	}
2117 
2118 	/* Update the cl_map array for logical drives */
2119 	for (count = 0; count < MAX_EXT_PARTS; count++) {
2120 		cl->cl_map[FDISK_P4 + 1 + count].dkl_cylno = UINT32_MAX;
2121 		cl->cl_map[FDISK_P4 + 1 + count].dkl_nblk =
2122 		    cl->cl_fmap[FD_NUMPART + count].fmap_nblk;
2123 		cl->cl_offset[FDISK_P4 + 1 + count] =
2124 		    cl->cl_fmap[FD_NUMPART + count].fmap_start;
2125 	}
2126 
2127 	cl->cl_logical_drive_count = i;
2128 	cl->cl_update_ext_minor_nodes = 0;
2129 	return (0);
2130 }
2131 /*
2132  *    Function: cmlb_validate_ext_part
2133  *
2134  * Description: utility routine to validate an extended partition's
2135  *		metadata as found on disk
2136  *
2137  *   Arguments:
2138  *	cl		driver soft state (unit) structure
2139  *	part		partition number of the extended partition
2140  *	epart		partition number of the logical drive
2141  *	start		absolute sector number of the start of the logical
2142  *			drive being validated
2143  *	size		size of logical drive being validated
2144  *
2145  * Return Code: 0 for success
2146  *
2147  *     Context: User and Kernel thread
2148  *
2149  * Algorithm :
2150  * Error cases are :
2151  *	1. If start block is lesser than or equal to the end block
2152  *	2. If either start block or end block is beyond the bounadry
2153  *	   of the extended partition.
2154  *	3. start or end block overlap with existing partitions.
2155  *		To check this, first make sure that the start block doesnt
2156  *		overlap with existing partitions. Then, calculate the
2157  *		possible end block for the given start block that doesnt
2158  *		overlap with existing partitions. This can be calculated by
2159  *		first setting the possible end block to the end of the
2160  *		extended partition (optimistic) and then, checking if there
2161  *		is any other partition that lies after the start of the
2162  *		partition being validated. If so, set the possible end to
2163  *		one block less than the beginning of the next nearest partition
2164  *		If the actual end block is greater than the calculated end
2165  *		block, we have an overlap.
2166  *
2167  */
2168 static int
2169 cmlb_validate_ext_part(struct cmlb_lun *cl, int part, int epart, uint32_t start,
2170     uint32_t size)
2171 {
2172 	int i;
2173 	uint32_t end = start + size - 1;
2174 	uint32_t ext_start = cl->cl_fmap[part].fmap_start;
2175 	uint32_t ext_end = ext_start + cl->cl_fmap[part].fmap_nblk - 1;
2176 	uint32_t ts, te;
2177 	uint32_t poss_end = ext_end;
2178 
2179 	if (end <= start) {
2180 		return (1);
2181 	}
2182 
2183 	/*
2184 	 * Check if the logical drive boundaries are within that of the
2185 	 * extended partition.
2186 	 */
2187 	if (start <= ext_start || start > ext_end || end <= ext_start ||
2188 	    end > ext_end) {
2189 		return (1);
2190 	}
2191 
2192 	/*
2193 	 * epart will be equal to FD_NUMPART if it is the first logical drive.
2194 	 * There is no need to check for overlaps with other logical drives,
2195 	 * since it is the only logical drive that we have come across so far.
2196 	 */
2197 	if (epart == FD_NUMPART) {
2198 		return (0);
2199 	}
2200 
2201 	/* Check for overlaps with existing logical drives */
2202 	i = FD_NUMPART;
2203 	ts = cl->cl_fmap[FD_NUMPART].fmap_start;
2204 	te = ts + cl->cl_fmap[FD_NUMPART].fmap_nblk - 1;
2205 
2206 	while ((i < epart) && ts && te) {
2207 		if (start >= ts && start <= te) {
2208 			return (1);
2209 		}
2210 
2211 		if ((ts < poss_end) && (ts > start)) {
2212 			poss_end = ts - 1;
2213 		}
2214 
2215 		i++;
2216 		ts = cl->cl_fmap[i].fmap_start;
2217 		te = ts + cl->cl_fmap[i].fmap_nblk - 1;
2218 	}
2219 
2220 	if (end > poss_end) {
2221 		return (1);
2222 	}
2223 
2224 	return (0);
2225 }
2226 
2227 
2228 /*
2229  *    Function: cmlb_is_linux_swap
2230  *
2231  * Description: utility routine to verify if a partition is a linux swap
2232  *		partition or not.
2233  *
2234  *   Arguments:
2235  *	cl		driver soft state (unit) structure
2236  *	part_start	absolute sector number of the start of the partition
2237  *			being verified
2238  *	tg_cookie	cookie from target driver to be passed back to target
2239  *			driver when we call back to it through tg_ops.
2240  *
2241  * Return Code: 0 for success
2242  *
2243  *     Context: User and Kernel thread
2244  *
2245  * Notes:
2246  *	The linux swap magic "SWAP-SPACE" or "SWAPSPACE2" is found as the
2247  *	last 10 bytes of a disk block whose size is that of the linux page
2248  *	size. This disk block is found at the beginning of the swap partition.
2249  */
2250 static int
2251 cmlb_is_linux_swap(struct cmlb_lun *cl, uint32_t part_start, void *tg_cookie)
2252 {
2253 	int		i;
2254 	int		rval = -1;
2255 	uint32_t	seek_offset;
2256 	uint32_t	linux_pg_size;
2257 	char 		*buf, *linux_swap_magic;
2258 	int		sec_sz = cl->cl_sys_blocksize;
2259 	/* Known linux kernel page sizes */
2260 	uint32_t	linux_pg_size_arr[] = {4096, };
2261 
2262 	ASSERT(cl != NULL);
2263 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
2264 
2265 	if ((buf = kmem_zalloc(sec_sz, KM_NOSLEEP)) == NULL) {
2266 		return (ENOMEM);
2267 	}
2268 
2269 	/*
2270 	 * Check if there is a sane Solaris VTOC
2271 	 * If there is a valid vtoc, no need to lookup
2272 	 * for the linux swap signature.
2273 	 */
2274 	mutex_exit(CMLB_MUTEX(cl));
2275 	rval = DK_TG_READ(cl, buf, part_start + DK_LABEL_LOC,
2276 	    sec_sz, tg_cookie);
2277 	mutex_enter(CMLB_MUTEX(cl));
2278 	if (rval != 0) {
2279 		cmlb_dbg(CMLB_ERROR,  cl,
2280 		    "cmlb_is_linux_swap: disk vtoc read err\n");
2281 		rval = EIO;
2282 		goto done;
2283 	}
2284 
2285 	if ((((struct dk_label *)buf)->dkl_magic == DKL_MAGIC) &&
2286 	    (((struct dk_label *)buf)->dkl_vtoc.v_sanity == VTOC_SANE)) {
2287 		rval = -1;
2288 		goto done;
2289 	}
2290 
2291 
2292 	/* No valid vtoc, so check for linux swap signature */
2293 	linux_swap_magic = buf + sec_sz - 10;
2294 
2295 	for (i = 0; i < sizeof (linux_pg_size_arr)/sizeof (uint32_t); i++) {
2296 		linux_pg_size = linux_pg_size_arr[i];
2297 		seek_offset = linux_pg_size/sec_sz - 1;
2298 		seek_offset += part_start;
2299 
2300 		mutex_exit(CMLB_MUTEX(cl));
2301 		rval = DK_TG_READ(cl, buf, seek_offset, sec_sz, tg_cookie);
2302 		mutex_enter(CMLB_MUTEX(cl));
2303 
2304 		if (rval != 0) {
2305 			cmlb_dbg(CMLB_ERROR,  cl,
2306 			    "cmlb_is_linux_swap: disk read err\n");
2307 			rval = EIO;
2308 			break;
2309 		}
2310 
2311 		rval = -1;
2312 
2313 		if ((strncmp(linux_swap_magic, "SWAP-SPACE", 10) == 0) ||
2314 		    (strncmp(linux_swap_magic, "SWAPSPACE2", 10) == 0)) {
2315 			/* Found a linux swap */
2316 			rval = 0;
2317 			break;
2318 		}
2319 	}
2320 
2321 done:
2322 	kmem_free(buf, sec_sz);
2323 	return (rval);
2324 }
2325 #endif
2326 
2327 /*
2328  *    Function: cmlb_read_fdisk
2329  *
2330  * Description: utility routine to read the fdisk table.
2331  *
2332  *   Arguments:
2333  *	cl		driver soft state (unit) structure
2334  *	capacity	disk capacity in #blocks
2335  *	tg_cookie	cookie from target driver to be passed back to target
2336  *			driver when we call back to it through tg_ops.
2337  *
2338  * Return Code: 0 for success (includes not reading for no_fdisk_present case
2339  *		errnos from tg_rw if failed to read the first block.
2340  *
2341  *     Context: Kernel thread only (can sleep).
2342  */
2343 /*ARGSUSED*/
2344 static int
2345 cmlb_read_fdisk(struct cmlb_lun *cl, diskaddr_t capacity, void *tg_cookie)
2346 {
2347 #if defined(_NO_FDISK_PRESENT)
2348 
2349 	cl->cl_solaris_offset = 0;
2350 	cl->cl_solaris_size = capacity;
2351 	bzero(cl->cl_fmap, sizeof (struct fmap) * FD_NUMPART);
2352 	return (0);
2353 
2354 #elif defined(_FIRMWARE_NEEDS_FDISK)
2355 
2356 	struct ipart	*fdp;
2357 	struct mboot	*mbp;
2358 	struct ipart	fdisk[FD_NUMPART];
2359 	int		i, k;
2360 	char		sigbuf[2];
2361 	caddr_t		bufp;
2362 	int		uidx;
2363 	int 		rval;
2364 	int		lba = 0;
2365 	uint_t		solaris_offset;	/* offset to solaris part. */
2366 	daddr_t		solaris_size;	/* size of solaris partition */
2367 	uint32_t	blocksize;
2368 #if defined(__i386) || defined(__amd64)
2369 	struct ipart	eparts[2];
2370 	struct ipart	*efdp1 = &eparts[0];
2371 	struct ipart	*efdp2 = &eparts[1];
2372 	int		ext_part_exists = 0;
2373 	int		ld_count = 0;
2374 #endif
2375 
2376 	ASSERT(cl != NULL);
2377 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
2378 
2379 	/*
2380 	 * Start off assuming no fdisk table
2381 	 */
2382 	solaris_offset = 0;
2383 	solaris_size   = capacity;
2384 
2385 	blocksize = cl->cl_tgt_blocksize;
2386 
2387 	bufp = kmem_zalloc(blocksize, KM_SLEEP);
2388 
2389 	mutex_exit(CMLB_MUTEX(cl));
2390 	rval = DK_TG_READ(cl, bufp, 0, blocksize, tg_cookie);
2391 	mutex_enter(CMLB_MUTEX(cl));
2392 
2393 	if (rval != 0) {
2394 		cmlb_dbg(CMLB_ERROR,  cl,
2395 		    "cmlb_read_fdisk: fdisk read err\n");
2396 		bzero(cl->cl_fmap, sizeof (struct fmap) * FD_NUMPART);
2397 		goto done;
2398 	}
2399 
2400 	mbp = (struct mboot *)bufp;
2401 
2402 	/*
2403 	 * The fdisk table does not begin on a 4-byte boundary within the
2404 	 * master boot record, so we copy it to an aligned structure to avoid
2405 	 * alignment exceptions on some processors.
2406 	 */
2407 	bcopy(&mbp->parts[0], fdisk, sizeof (fdisk));
2408 
2409 	/*
2410 	 * Check for lba support before verifying sig; sig might not be
2411 	 * there, say on a blank disk, but the max_chs mark may still
2412 	 * be present.
2413 	 *
2414 	 * Note: LBA support and BEFs are an x86-only concept but this
2415 	 * code should work OK on SPARC as well.
2416 	 */
2417 
2418 	/*
2419 	 * First, check for lba-access-ok on root node (or prom root node)
2420 	 * if present there, don't need to search fdisk table.
2421 	 */
2422 	if (ddi_getprop(DDI_DEV_T_ANY, ddi_root_node(), 0,
2423 	    "lba-access-ok", 0) != 0) {
2424 		/* All drives do LBA; don't search fdisk table */
2425 		lba = 1;
2426 	} else {
2427 		/* Okay, look for mark in fdisk table */
2428 		for (fdp = fdisk, i = 0; i < FD_NUMPART; i++, fdp++) {
2429 			/* accumulate "lba" value from all partitions */
2430 			lba = (lba || cmlb_has_max_chs_vals(fdp));
2431 		}
2432 	}
2433 
2434 	if (lba != 0) {
2435 		dev_t dev = cmlb_make_device(cl);
2436 
2437 		if (ddi_getprop(dev, CMLB_DEVINFO(cl), DDI_PROP_DONTPASS,
2438 		    "lba-access-ok", 0) == 0) {
2439 			/* not found; create it */
2440 			if (ddi_prop_create(dev, CMLB_DEVINFO(cl), 0,
2441 			    "lba-access-ok", (caddr_t)NULL, 0) !=
2442 			    DDI_PROP_SUCCESS) {
2443 				cmlb_dbg(CMLB_ERROR,  cl,
2444 				    "cmlb_read_fdisk: Can't create lba "
2445 				    "property for instance %d\n",
2446 				    ddi_get_instance(CMLB_DEVINFO(cl)));
2447 			}
2448 		}
2449 	}
2450 
2451 	bcopy(&mbp->signature, sigbuf, sizeof (sigbuf));
2452 
2453 	/*
2454 	 * Endian-independent signature check
2455 	 */
2456 	if (((sigbuf[1] & 0xFF) != ((MBB_MAGIC >> 8) & 0xFF)) ||
2457 	    (sigbuf[0] != (MBB_MAGIC & 0xFF))) {
2458 		cmlb_dbg(CMLB_ERROR,  cl,
2459 		    "cmlb_read_fdisk: no fdisk\n");
2460 		bzero(cl->cl_fmap, sizeof (struct fmap) * FD_NUMPART);
2461 		goto done;
2462 	}
2463 
2464 #ifdef CMLBDEBUG
2465 	if (cmlb_level_mask & CMLB_LOGMASK_INFO) {
2466 		fdp = fdisk;
2467 		cmlb_dbg(CMLB_INFO,  cl, "cmlb_read_fdisk:\n");
2468 		cmlb_dbg(CMLB_INFO,  cl, "         relsect    "
2469 		    "numsect         sysid       bootid\n");
2470 		for (i = 0; i < FD_NUMPART; i++, fdp++) {
2471 			cmlb_dbg(CMLB_INFO,  cl,
2472 			    "    %d:  %8d   %8d     0x%08x     0x%08x\n",
2473 			    i, fdp->relsect, fdp->numsect,
2474 			    fdp->systid, fdp->bootid);
2475 		}
2476 	}
2477 #endif
2478 
2479 	/*
2480 	 * Try to find the unix partition
2481 	 */
2482 	uidx = -1;
2483 	solaris_offset = 0;
2484 	solaris_size   = 0;
2485 
2486 	for (fdp = fdisk, i = 0; i < FD_NUMPART; i++, fdp++) {
2487 		uint32_t relsect;
2488 		uint32_t numsect;
2489 		uchar_t systid;
2490 #if defined(__i386) || defined(__amd64)
2491 		/*
2492 		 * Stores relative block offset from the beginning of the
2493 		 * Extended Partition.
2494 		 */
2495 		int	ext_relsect = 0;
2496 #endif
2497 
2498 		if (fdp->numsect == 0) {
2499 			cl->cl_fmap[i].fmap_start = 0;
2500 			cl->cl_fmap[i].fmap_nblk  = 0;
2501 			continue;
2502 		}
2503 
2504 		/*
2505 		 * Data in the fdisk table is little-endian.
2506 		 */
2507 		relsect = LE_32(fdp->relsect);
2508 		numsect = LE_32(fdp->numsect);
2509 
2510 		cl->cl_fmap[i].fmap_start = relsect;
2511 		cl->cl_fmap[i].fmap_nblk  = numsect;
2512 		cl->cl_fmap[i].fmap_systid = LE_8(fdp->systid);
2513 
2514 #if defined(__i386) || defined(__amd64)
2515 		/* Support only one extended partition per LUN */
2516 		if ((fdp->systid == EXTDOS || fdp->systid == FDISK_EXTLBA) &&
2517 		    (ext_part_exists == 0)) {
2518 			int j;
2519 			uint32_t logdrive_offset;
2520 			uint32_t ext_numsect;
2521 			uint32_t abs_secnum;
2522 
2523 			ext_part_exists = 1;
2524 
2525 			for (j = FD_NUMPART; j < FDISK_PARTS; j++) {
2526 				mutex_exit(CMLB_MUTEX(cl));
2527 				rval = DK_TG_READ(cl, bufp,
2528 				    (relsect + ext_relsect), blocksize,
2529 				    tg_cookie);
2530 				mutex_enter(CMLB_MUTEX(cl));
2531 
2532 				if (rval != 0) {
2533 					cmlb_dbg(CMLB_ERROR,  cl,
2534 					    "cmlb_read_fdisk: Extended "
2535 					    "partition read err\n");
2536 					goto done;
2537 				}
2538 				/*
2539 				 * The first ipart entry provides the offset
2540 				 * at which the logical drive starts off from
2541 				 * the beginning of the container partition
2542 				 * and the size of the logical drive.
2543 				 * The second ipart entry provides the offset
2544 				 * of the next container partition from the
2545 				 * beginning of the extended partition.
2546 				 */
2547 				bcopy(&bufp[FDISK_PART_TABLE_START], eparts,
2548 				    sizeof (eparts));
2549 				logdrive_offset = LE_32(efdp1->relsect);
2550 				ext_numsect = LE_32(efdp1->numsect);
2551 				systid = LE_8(efdp1->systid);
2552 				if (logdrive_offset <= 0 || ext_numsect <= 0)
2553 					break;
2554 				abs_secnum = relsect + ext_relsect +
2555 				    logdrive_offset;
2556 
2557 				/* Boundary condition and overlap checking */
2558 				if (cmlb_validate_ext_part(cl, i, j, abs_secnum,
2559 				    ext_numsect)) {
2560 					break;
2561 				}
2562 
2563 				if ((cl->cl_fmap[j].fmap_start != abs_secnum) ||
2564 				    (cl->cl_fmap[j].fmap_nblk != ext_numsect) ||
2565 				    (cl->cl_fmap[j].fmap_systid != systid)) {
2566 					/*
2567 					 * Indicates change from previous
2568 					 * partinfo. Need to recreate
2569 					 * logical device nodes.
2570 					 */
2571 					cl->cl_update_ext_minor_nodes = 1;
2572 				}
2573 				cl->cl_fmap[j].fmap_start = abs_secnum;
2574 				cl->cl_fmap[j].fmap_nblk  = ext_numsect;
2575 				cl->cl_fmap[j].fmap_systid = systid;
2576 				ld_count++;
2577 
2578 				if ((efdp1->systid == SUNIXOS &&
2579 				    (cmlb_is_linux_swap(cl, abs_secnum,
2580 				    tg_cookie) != 0)) ||
2581 				    efdp1->systid == SUNIXOS2) {
2582 					if (uidx == -1) {
2583 						uidx = 0;
2584 						solaris_offset = abs_secnum;
2585 						solaris_size = ext_numsect;
2586 					}
2587 				}
2588 
2589 				if ((ext_relsect = LE_32(efdp2->relsect)) == 0)
2590 					break;
2591 			}
2592 		}
2593 
2594 #endif
2595 
2596 		if (fdp->systid != SUNIXOS &&
2597 		    fdp->systid != SUNIXOS2 &&
2598 		    fdp->systid != EFI_PMBR) {
2599 			continue;
2600 		}
2601 
2602 		/*
2603 		 * use the last active solaris partition id found
2604 		 * (there should only be 1 active partition id)
2605 		 *
2606 		 * if there are no active solaris partition id
2607 		 * then use the first inactive solaris partition id
2608 		 */
2609 		if ((uidx == -1) || (fdp->bootid == ACTIVE)) {
2610 #if defined(__i386) || defined(__amd64)
2611 			if (fdp->systid != SUNIXOS ||
2612 			    (fdp->systid == SUNIXOS &&
2613 			    (cmlb_is_linux_swap(cl, relsect,
2614 			    tg_cookie) != 0))) {
2615 #endif
2616 				uidx = i;
2617 				solaris_offset = relsect;
2618 				solaris_size   = numsect;
2619 #if defined(__i386) || defined(__amd64)
2620 			}
2621 #endif
2622 		}
2623 	}
2624 #if defined(__i386) || defined(__amd64)
2625 	if (ld_count < cl->cl_logical_drive_count) {
2626 		/*
2627 		 * Some/all logical drives were deleted. Clear out
2628 		 * the fmap entries correspoding to those deleted drives.
2629 		 */
2630 		for (k = ld_count + FD_NUMPART;
2631 		    k < cl->cl_logical_drive_count + FD_NUMPART; k++) {
2632 			cl->cl_fmap[k].fmap_start = 0;
2633 			cl->cl_fmap[k].fmap_nblk  = 0;
2634 			cl->cl_fmap[k].fmap_systid = 0;
2635 		}
2636 		cl->cl_update_ext_minor_nodes = 1;
2637 	}
2638 	if (cl->cl_update_ext_minor_nodes) {
2639 		rval = cmlb_update_ext_minor_nodes(cl, ld_count);
2640 		if (rval != 0) {
2641 			goto done;
2642 		}
2643 	}
2644 #endif
2645 	cmlb_dbg(CMLB_INFO,  cl, "fdisk 0x%x 0x%lx",
2646 	    cl->cl_solaris_offset, cl->cl_solaris_size);
2647 done:
2648 
2649 	/*
2650 	 * Clear the VTOC info, only if the Solaris partition entry
2651 	 * has moved, changed size, been deleted, or if the size of
2652 	 * the partition is too small to even fit the label sector.
2653 	 */
2654 	if ((cl->cl_solaris_offset != solaris_offset) ||
2655 	    (cl->cl_solaris_size != solaris_size) ||
2656 	    solaris_size <= DK_LABEL_LOC) {
2657 		cmlb_dbg(CMLB_INFO,  cl, "fdisk moved 0x%x 0x%lx",
2658 		    solaris_offset, solaris_size);
2659 		bzero(&cl->cl_g, sizeof (struct dk_geom));
2660 		bzero(&cl->cl_vtoc, sizeof (struct dk_vtoc));
2661 		bzero(&cl->cl_map, NDKMAP * (sizeof (struct dk_map)));
2662 		cl->cl_f_geometry_is_valid = B_FALSE;
2663 	}
2664 	cl->cl_solaris_offset = solaris_offset;
2665 	cl->cl_solaris_size = solaris_size;
2666 	kmem_free(bufp, blocksize);
2667 	return (rval);
2668 
2669 #else	/* #elif defined(_FIRMWARE_NEEDS_FDISK) */
2670 #error "fdisk table presence undetermined for this platform."
2671 #endif	/* #if defined(_NO_FDISK_PRESENT) */
2672 }
2673 
2674 static void
2675 cmlb_swap_efi_gpt(efi_gpt_t *e)
2676 {
2677 	_NOTE(ASSUMING_PROTECTED(*e))
2678 	e->efi_gpt_Signature = LE_64(e->efi_gpt_Signature);
2679 	e->efi_gpt_Revision = LE_32(e->efi_gpt_Revision);
2680 	e->efi_gpt_HeaderSize = LE_32(e->efi_gpt_HeaderSize);
2681 	e->efi_gpt_HeaderCRC32 = LE_32(e->efi_gpt_HeaderCRC32);
2682 	e->efi_gpt_MyLBA = LE_64(e->efi_gpt_MyLBA);
2683 	e->efi_gpt_AlternateLBA = LE_64(e->efi_gpt_AlternateLBA);
2684 	e->efi_gpt_FirstUsableLBA = LE_64(e->efi_gpt_FirstUsableLBA);
2685 	e->efi_gpt_LastUsableLBA = LE_64(e->efi_gpt_LastUsableLBA);
2686 	UUID_LE_CONVERT(e->efi_gpt_DiskGUID, e->efi_gpt_DiskGUID);
2687 	e->efi_gpt_PartitionEntryLBA = LE_64(e->efi_gpt_PartitionEntryLBA);
2688 	e->efi_gpt_NumberOfPartitionEntries =
2689 	    LE_32(e->efi_gpt_NumberOfPartitionEntries);
2690 	e->efi_gpt_SizeOfPartitionEntry =
2691 	    LE_32(e->efi_gpt_SizeOfPartitionEntry);
2692 	e->efi_gpt_PartitionEntryArrayCRC32 =
2693 	    LE_32(e->efi_gpt_PartitionEntryArrayCRC32);
2694 }
2695 
2696 static void
2697 cmlb_swap_efi_gpe(int nparts, efi_gpe_t *p)
2698 {
2699 	int i;
2700 
2701 	_NOTE(ASSUMING_PROTECTED(*p))
2702 	for (i = 0; i < nparts; i++) {
2703 		UUID_LE_CONVERT(p[i].efi_gpe_PartitionTypeGUID,
2704 		    p[i].efi_gpe_PartitionTypeGUID);
2705 		p[i].efi_gpe_StartingLBA = LE_64(p[i].efi_gpe_StartingLBA);
2706 		p[i].efi_gpe_EndingLBA = LE_64(p[i].efi_gpe_EndingLBA);
2707 		/* PartitionAttrs */
2708 	}
2709 }
2710 
2711 static int
2712 cmlb_validate_efi(efi_gpt_t *labp)
2713 {
2714 	if (labp->efi_gpt_Signature != EFI_SIGNATURE)
2715 		return (EINVAL);
2716 	/* at least 96 bytes in this version of the spec. */
2717 	if (sizeof (efi_gpt_t) - sizeof (labp->efi_gpt_Reserved2) >
2718 	    labp->efi_gpt_HeaderSize)
2719 		return (EINVAL);
2720 	/* this should be 128 bytes */
2721 	if (labp->efi_gpt_SizeOfPartitionEntry != sizeof (efi_gpe_t))
2722 		return (EINVAL);
2723 	return (0);
2724 }
2725 
2726 /*
2727  * This function returns B_FALSE if there is a valid MBR signature and no
2728  * partition table entries of type EFI_PMBR (0xEE). Otherwise it returns B_TRUE.
2729  *
2730  * The EFI spec (1.10 and later) requires having a Protective MBR (PMBR) to
2731  * recognize the disk as GPT partitioned. However, some other OS creates an MBR
2732  * where a PMBR entry is not the only one. Also, if the first block has been
2733  * corrupted, currently best attempt to allow data access would be to try to
2734  * check for GPT headers. Hence in case of more than one partition entry, but
2735  * at least one EFI_PMBR partition type or no valid magic number, the function
2736  * returns B_TRUE to continue with looking for GPT header.
2737  */
2738 
2739 static boolean_t
2740 cmlb_check_efi_mbr(uchar_t *buf, boolean_t *is_mbr)
2741 {
2742 	struct ipart	*fdp;
2743 	struct mboot	*mbp = (struct mboot *)buf;
2744 	struct ipart	fdisk[FD_NUMPART];
2745 	int		i;
2746 
2747 	if (is_mbr != NULL)
2748 		*is_mbr = B_TRUE;
2749 
2750 	if (LE_16(mbp->signature) != MBB_MAGIC) {
2751 		if (is_mbr != NULL)
2752 			*is_mbr = B_FALSE;
2753 		return (B_TRUE);
2754 	}
2755 
2756 	bcopy(&mbp->parts[0], fdisk, sizeof (fdisk));
2757 
2758 	for (fdp = fdisk, i = 0; i < FD_NUMPART; i++, fdp++) {
2759 		if (fdp->systid == EFI_PMBR)
2760 			return (B_TRUE);
2761 	}
2762 
2763 	return (B_FALSE);
2764 }
2765 
2766 static int
2767 cmlb_use_efi(struct cmlb_lun *cl, diskaddr_t capacity, int flags,
2768     void *tg_cookie)
2769 {
2770 	int		i;
2771 	int		rval = 0;
2772 	efi_gpe_t	*partitions;
2773 	uchar_t		*buf;
2774 	uint_t		lbasize;	/* is really how much to read */
2775 	diskaddr_t	cap = 0;
2776 	uint_t		nparts;
2777 	diskaddr_t	gpe_lba;
2778 	diskaddr_t	alternate_lba;
2779 	int		iofailed = 0;
2780 	struct uuid	uuid_type_reserved = EFI_RESERVED;
2781 #if defined(_FIRMWARE_NEEDS_FDISK)
2782 	boolean_t 	is_mbr;
2783 #endif
2784 
2785 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
2786 
2787 	lbasize = cl->cl_sys_blocksize;
2788 
2789 	cl->cl_reserved = -1;
2790 	mutex_exit(CMLB_MUTEX(cl));
2791 
2792 	buf = kmem_zalloc(EFI_MIN_ARRAY_SIZE, KM_SLEEP);
2793 
2794 	rval = DK_TG_READ(cl, buf,  0, lbasize, tg_cookie);
2795 	if (rval) {
2796 		iofailed = 1;
2797 		goto done_err;
2798 	}
2799 	if (((struct dk_label *)buf)->dkl_magic == DKL_MAGIC) {
2800 		/* not ours */
2801 		rval = ESRCH;
2802 		goto done_err;
2803 	}
2804 
2805 #if defined(_FIRMWARE_NEEDS_FDISK)
2806 	if (!cmlb_check_efi_mbr(buf, &is_mbr)) {
2807 		if (is_mbr)
2808 			rval = ESRCH;
2809 		else
2810 			rval = EINVAL;
2811 		goto done_err;
2812 	}
2813 #else
2814 	if (!cmlb_check_efi_mbr(buf, NULL)) {
2815 		rval = EINVAL;
2816 		goto done_err;
2817 	}
2818 
2819 #endif
2820 
2821 	rval = DK_TG_READ(cl, buf, 1, lbasize, tg_cookie);
2822 	if (rval) {
2823 		iofailed = 1;
2824 		goto done_err;
2825 	}
2826 	cmlb_swap_efi_gpt((efi_gpt_t *)buf);
2827 
2828 	if ((rval = cmlb_validate_efi((efi_gpt_t *)buf)) != 0) {
2829 		/*
2830 		 * Couldn't read the primary, try the backup.  Our
2831 		 * capacity at this point could be based on CHS, so
2832 		 * check what the device reports.
2833 		 */
2834 		rval = DK_TG_GETCAP(cl, &cap, tg_cookie);
2835 		if (rval) {
2836 			iofailed = 1;
2837 			goto done_err;
2838 		}
2839 
2840 		/*
2841 		 * CMLB_OFF_BY_ONE case, we check the next to last block first
2842 		 * for backup GPT header, otherwise check the last block.
2843 		 */
2844 
2845 		if ((rval = DK_TG_READ(cl, buf,
2846 		    cap - ((cl->cl_alter_behavior & CMLB_OFF_BY_ONE) ? 2 : 1),
2847 		    lbasize, tg_cookie))
2848 		    != 0) {
2849 			iofailed = 1;
2850 			goto done_err;
2851 		}
2852 		cmlb_swap_efi_gpt((efi_gpt_t *)buf);
2853 
2854 		if ((rval = cmlb_validate_efi((efi_gpt_t *)buf)) != 0) {
2855 
2856 			if (!(cl->cl_alter_behavior & CMLB_OFF_BY_ONE))
2857 				goto done_err;
2858 			if ((rval = DK_TG_READ(cl, buf, cap - 1, lbasize,
2859 			    tg_cookie)) != 0)
2860 				goto done_err;
2861 			cmlb_swap_efi_gpt((efi_gpt_t *)buf);
2862 			if ((rval = cmlb_validate_efi((efi_gpt_t *)buf)) != 0)
2863 				goto done_err;
2864 		}
2865 		if (!(flags & CMLB_SILENT))
2866 			cmlb_log(CMLB_DEVINFO(cl), CMLB_LABEL(cl), CE_WARN,
2867 			    "primary label corrupt; using backup\n");
2868 	}
2869 
2870 	nparts = ((efi_gpt_t *)buf)->efi_gpt_NumberOfPartitionEntries;
2871 	gpe_lba = ((efi_gpt_t *)buf)->efi_gpt_PartitionEntryLBA;
2872 	alternate_lba = ((efi_gpt_t *)buf)->efi_gpt_AlternateLBA;
2873 
2874 	rval = DK_TG_READ(cl, buf, gpe_lba, EFI_MIN_ARRAY_SIZE, tg_cookie);
2875 	if (rval) {
2876 		iofailed = 1;
2877 		goto done_err;
2878 	}
2879 	partitions = (efi_gpe_t *)buf;
2880 
2881 	if (nparts > MAXPART) {
2882 		nparts = MAXPART;
2883 	}
2884 	cmlb_swap_efi_gpe(nparts, partitions);
2885 
2886 	mutex_enter(CMLB_MUTEX(cl));
2887 
2888 	/* Fill in partition table. */
2889 	for (i = 0; i < nparts; i++) {
2890 		if (partitions->efi_gpe_StartingLBA != 0 ||
2891 		    partitions->efi_gpe_EndingLBA != 0) {
2892 			cl->cl_map[i].dkl_cylno =
2893 			    partitions->efi_gpe_StartingLBA;
2894 			cl->cl_map[i].dkl_nblk =
2895 			    partitions->efi_gpe_EndingLBA -
2896 			    partitions->efi_gpe_StartingLBA + 1;
2897 			cl->cl_offset[i] =
2898 			    partitions->efi_gpe_StartingLBA;
2899 		}
2900 
2901 		if (cl->cl_reserved == -1) {
2902 			if (bcmp(&partitions->efi_gpe_PartitionTypeGUID,
2903 			    &uuid_type_reserved, sizeof (struct uuid)) == 0) {
2904 				cl->cl_reserved = i;
2905 			}
2906 		}
2907 		if (i == WD_NODE) {
2908 			/*
2909 			 * minor number 7 corresponds to the whole disk
2910 			 * if the disk capacity is expanded after disk is
2911 			 * labeled, minor number 7 represents the capacity
2912 			 * indicated by the disk label.
2913 			 */
2914 			cl->cl_map[i].dkl_cylno = 0;
2915 			if (alternate_lba == 1) {
2916 				/*
2917 				 * We are using backup label. Since we can
2918 				 * find a valid label at the end of disk,
2919 				 * the disk capacity is not expanded.
2920 				 */
2921 				cl->cl_map[i].dkl_nblk = capacity;
2922 			} else {
2923 				cl->cl_map[i].dkl_nblk = alternate_lba + 1;
2924 			}
2925 			cl->cl_offset[i] = 0;
2926 		}
2927 		partitions++;
2928 	}
2929 	cl->cl_solaris_offset = 0;
2930 	cl->cl_solaris_size = capacity;
2931 	cl->cl_label_from_media = CMLB_LABEL_EFI;
2932 	cl->cl_f_geometry_is_valid = B_TRUE;
2933 
2934 	/* clear the vtoc label */
2935 	bzero(&cl->cl_vtoc, sizeof (struct dk_vtoc));
2936 
2937 	kmem_free(buf, EFI_MIN_ARRAY_SIZE);
2938 	return (0);
2939 
2940 done_err:
2941 	kmem_free(buf, EFI_MIN_ARRAY_SIZE);
2942 	mutex_enter(CMLB_MUTEX(cl));
2943 done_err1:
2944 	/*
2945 	 * if we didn't find something that could look like a VTOC
2946 	 * and the disk is over 1TB, we know there isn't a valid label.
2947 	 * Otherwise let cmlb_uselabel decide what to do.  We only
2948 	 * want to invalidate this if we're certain the label isn't
2949 	 * valid because cmlb_prop_op will now fail, which in turn
2950 	 * causes things like opens and stats on the partition to fail.
2951 	 */
2952 	if ((capacity > CMLB_EXTVTOC_LIMIT) && (rval != ESRCH) && !iofailed) {
2953 		cl->cl_f_geometry_is_valid = B_FALSE;
2954 	}
2955 	return (rval);
2956 }
2957 
2958 
2959 /*
2960  *    Function: cmlb_uselabel
2961  *
2962  * Description: Validate the disk label and update the relevant data (geometry,
2963  *		partition, vtoc, and capacity data) in the cmlb_lun struct.
2964  *		Marks the geometry of the unit as being valid.
2965  *
2966  *   Arguments: cl: unit struct.
2967  *		dk_label: disk label
2968  *
2969  * Return Code: CMLB_LABEL_IS_VALID: Label read from disk is OK; geometry,
2970  *		partition, vtoc, and capacity data are good.
2971  *
2972  *		CMLB_LABEL_IS_INVALID: Magic number or checksum error in the
2973  *		label; or computed capacity does not jibe with capacity
2974  *		reported from the READ CAPACITY command.
2975  *
2976  *     Context: Kernel thread only (can sleep).
2977  */
2978 static int
2979 cmlb_uselabel(struct cmlb_lun *cl, struct dk_label *labp, int flags)
2980 {
2981 	short		*sp;
2982 	short		sum;
2983 	short		count;
2984 	int		label_error = CMLB_LABEL_IS_VALID;
2985 	int		i;
2986 	diskaddr_t	label_capacity;
2987 	uint32_t	part_end;
2988 	diskaddr_t	track_capacity;
2989 #if defined(_SUNOS_VTOC_16)
2990 	struct	dkl_partition	*vpartp;
2991 #endif
2992 	ASSERT(cl != NULL);
2993 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
2994 
2995 	/* Validate the magic number of the label. */
2996 	if (labp->dkl_magic != DKL_MAGIC) {
2997 #if defined(__sparc)
2998 		if (!ISREMOVABLE(cl) && !ISHOTPLUGGABLE(cl)) {
2999 			if (!(flags & CMLB_SILENT))
3000 				cmlb_log(CMLB_DEVINFO(cl), CMLB_LABEL(cl),
3001 				    CE_WARN,
3002 				    "Corrupt label; wrong magic number\n");
3003 		}
3004 #endif
3005 		return (CMLB_LABEL_IS_INVALID);
3006 	}
3007 
3008 	/* Validate the checksum of the label. */
3009 	sp  = (short *)labp;
3010 	sum = 0;
3011 	count = sizeof (struct dk_label) / sizeof (short);
3012 	while (count--)	 {
3013 		sum ^= *sp++;
3014 	}
3015 
3016 	if (sum != 0) {
3017 #if defined(_SUNOS_VTOC_16)
3018 		if (!ISCD(cl)) {
3019 #elif defined(_SUNOS_VTOC_8)
3020 		if (!ISREMOVABLE(cl) && !ISHOTPLUGGABLE(cl)) {
3021 #endif
3022 			if (!(flags & CMLB_SILENT))
3023 				cmlb_log(CMLB_DEVINFO(cl), CMLB_LABEL(cl),
3024 				    CE_WARN,
3025 				    "Corrupt label - label checksum failed\n");
3026 		}
3027 		return (CMLB_LABEL_IS_INVALID);
3028 	}
3029 
3030 
3031 	/*
3032 	 * Fill in geometry structure with data from label.
3033 	 */
3034 	bzero(&cl->cl_g, sizeof (struct dk_geom));
3035 	cl->cl_g.dkg_ncyl   = labp->dkl_ncyl;
3036 	cl->cl_g.dkg_acyl   = labp->dkl_acyl;
3037 	cl->cl_g.dkg_bcyl   = 0;
3038 	cl->cl_g.dkg_nhead  = labp->dkl_nhead;
3039 	cl->cl_g.dkg_nsect  = labp->dkl_nsect;
3040 	cl->cl_g.dkg_intrlv = labp->dkl_intrlv;
3041 
3042 #if defined(_SUNOS_VTOC_8)
3043 	cl->cl_g.dkg_gap1   = labp->dkl_gap1;
3044 	cl->cl_g.dkg_gap2   = labp->dkl_gap2;
3045 	cl->cl_g.dkg_bhead  = labp->dkl_bhead;
3046 #endif
3047 #if defined(_SUNOS_VTOC_16)
3048 	cl->cl_dkg_skew = labp->dkl_skew;
3049 #endif
3050 
3051 #if defined(__i386) || defined(__amd64)
3052 	cl->cl_g.dkg_apc = labp->dkl_apc;
3053 #endif
3054 
3055 	/*
3056 	 * Currently we rely on the values in the label being accurate. If
3057 	 * dkl_rpm or dkl_pcly are zero in the label, use a default value.
3058 	 *
3059 	 * Note: In the future a MODE SENSE may be used to retrieve this data,
3060 	 * although this command is optional in SCSI-2.
3061 	 */
3062 	cl->cl_g.dkg_rpm  = (labp->dkl_rpm  != 0) ? labp->dkl_rpm  : 3600;
3063 	cl->cl_g.dkg_pcyl = (labp->dkl_pcyl != 0) ? labp->dkl_pcyl :
3064 	    (cl->cl_g.dkg_ncyl + cl->cl_g.dkg_acyl);
3065 
3066 	/*
3067 	 * The Read and Write reinstruct values may not be valid
3068 	 * for older disks.
3069 	 */
3070 	cl->cl_g.dkg_read_reinstruct  = labp->dkl_read_reinstruct;
3071 	cl->cl_g.dkg_write_reinstruct = labp->dkl_write_reinstruct;
3072 
3073 	/* Fill in partition table. */
3074 #if defined(_SUNOS_VTOC_8)
3075 	for (i = 0; i < NDKMAP; i++) {
3076 		cl->cl_map[i].dkl_cylno = labp->dkl_map[i].dkl_cylno;
3077 		cl->cl_map[i].dkl_nblk  = labp->dkl_map[i].dkl_nblk;
3078 	}
3079 #endif
3080 #if  defined(_SUNOS_VTOC_16)
3081 	vpartp		= labp->dkl_vtoc.v_part;
3082 	track_capacity	= labp->dkl_nhead * labp->dkl_nsect;
3083 
3084 	/* Prevent divide by zero */
3085 	if (track_capacity == 0) {
3086 		if (!(flags & CMLB_SILENT))
3087 			cmlb_log(CMLB_DEVINFO(cl), CMLB_LABEL(cl), CE_WARN,
3088 			    "Corrupt label - zero nhead or nsect value\n");
3089 
3090 		return (CMLB_LABEL_IS_INVALID);
3091 	}
3092 
3093 	for (i = 0; i < NDKMAP; i++, vpartp++) {
3094 		cl->cl_map[i].dkl_cylno = vpartp->p_start / track_capacity;
3095 		cl->cl_map[i].dkl_nblk  = vpartp->p_size;
3096 	}
3097 #endif
3098 
3099 	/* Fill in VTOC Structure. */
3100 	bcopy(&labp->dkl_vtoc, &cl->cl_vtoc, sizeof (struct dk_vtoc));
3101 #if defined(_SUNOS_VTOC_8)
3102 	/*
3103 	 * The 8-slice vtoc does not include the ascii label; save it into
3104 	 * the device's soft state structure here.
3105 	 */
3106 	bcopy(labp->dkl_asciilabel, cl->cl_asciilabel, LEN_DKL_ASCII);
3107 #endif
3108 
3109 	/* Now look for a valid capacity. */
3110 	track_capacity	= (cl->cl_g.dkg_nhead * cl->cl_g.dkg_nsect);
3111 	label_capacity	= (cl->cl_g.dkg_ncyl  * track_capacity);
3112 
3113 	if (cl->cl_g.dkg_acyl) {
3114 #if defined(__i386) || defined(__amd64)
3115 		/* we may have > 1 alts cylinder */
3116 		label_capacity += (track_capacity * cl->cl_g.dkg_acyl);
3117 #else
3118 		label_capacity += track_capacity;
3119 #endif
3120 	}
3121 
3122 	/*
3123 	 * Force check here to ensure the computed capacity is valid.
3124 	 * If capacity is zero, it indicates an invalid label and
3125 	 * we should abort updating the relevant data then.
3126 	 */
3127 	if (label_capacity == 0) {
3128 		if (!(flags & CMLB_SILENT))
3129 			cmlb_log(CMLB_DEVINFO(cl), CMLB_LABEL(cl), CE_WARN,
3130 			    "Corrupt label - no valid capacity could be "
3131 			    "retrieved\n");
3132 
3133 		return (CMLB_LABEL_IS_INVALID);
3134 	}
3135 
3136 	/* Mark the geometry as valid. */
3137 	cl->cl_f_geometry_is_valid = B_TRUE;
3138 
3139 	/*
3140 	 * if we got invalidated when mutex exit and entered again,
3141 	 * if blockcount different than when we came in, need to
3142 	 * retry from beginning of cmlb_validate_geometry.
3143 	 * revisit this on next phase of utilizing this for
3144 	 * sd.
3145 	 */
3146 
3147 	if (label_capacity <= cl->cl_blockcount) {
3148 #if defined(_SUNOS_VTOC_8)
3149 		/*
3150 		 * We can't let this happen on drives that are subdivided
3151 		 * into logical disks (i.e., that have an fdisk table).
3152 		 * The cl_blockcount field should always hold the full media
3153 		 * size in sectors, period.  This code would overwrite
3154 		 * cl_blockcount with the size of the Solaris fdisk partition.
3155 		 */
3156 		cmlb_dbg(CMLB_ERROR,  cl,
3157 		    "cmlb_uselabel: Label %d blocks; Drive %d blocks\n",
3158 		    label_capacity, cl->cl_blockcount);
3159 		cl->cl_solaris_size = label_capacity;
3160 
3161 #endif	/* defined(_SUNOS_VTOC_8) */
3162 		goto done;
3163 	}
3164 
3165 	if (ISCD(cl)) {
3166 		/* For CDROMs, we trust that the data in the label is OK. */
3167 #if defined(_SUNOS_VTOC_8)
3168 		for (i = 0; i < NDKMAP; i++) {
3169 			part_end = labp->dkl_nhead * labp->dkl_nsect *
3170 			    labp->dkl_map[i].dkl_cylno +
3171 			    labp->dkl_map[i].dkl_nblk  - 1;
3172 
3173 			if ((labp->dkl_map[i].dkl_nblk) &&
3174 			    (part_end > cl->cl_blockcount)) {
3175 				cl->cl_f_geometry_is_valid = B_FALSE;
3176 				break;
3177 			}
3178 		}
3179 #endif
3180 #if defined(_SUNOS_VTOC_16)
3181 		vpartp = &(labp->dkl_vtoc.v_part[0]);
3182 		for (i = 0; i < NDKMAP; i++, vpartp++) {
3183 			part_end = vpartp->p_start + vpartp->p_size;
3184 			if ((vpartp->p_size > 0) &&
3185 			    (part_end > cl->cl_blockcount)) {
3186 				cl->cl_f_geometry_is_valid = B_FALSE;
3187 				break;
3188 			}
3189 		}
3190 #endif
3191 	} else {
3192 		/* label_capacity > cl->cl_blockcount */
3193 		if (!(flags & CMLB_SILENT)) {
3194 			cmlb_log(CMLB_DEVINFO(cl), CMLB_LABEL(cl), CE_WARN,
3195 			    "Corrupt label - bad geometry\n");
3196 			cmlb_log(CMLB_DEVINFO(cl), CMLB_LABEL(cl), CE_CONT,
3197 			    "Label says %llu blocks; Drive says %llu blocks\n",
3198 			    label_capacity, cl->cl_blockcount);
3199 		}
3200 		cl->cl_f_geometry_is_valid = B_FALSE;
3201 		label_error = CMLB_LABEL_IS_INVALID;
3202 	}
3203 
3204 done:
3205 
3206 	cmlb_dbg(CMLB_INFO,  cl, "cmlb_uselabel: (label geometry)\n");
3207 	cmlb_dbg(CMLB_INFO,  cl,
3208 	    "   ncyl: %d; acyl: %d; nhead: %d; nsect: %d\n",
3209 	    cl->cl_g.dkg_ncyl,  cl->cl_g.dkg_acyl,
3210 	    cl->cl_g.dkg_nhead, cl->cl_g.dkg_nsect);
3211 
3212 	cmlb_dbg(CMLB_INFO,  cl,
3213 	    "   label_capacity: %d; intrlv: %d; rpm: %d\n",
3214 	    cl->cl_blockcount, cl->cl_g.dkg_intrlv, cl->cl_g.dkg_rpm);
3215 	cmlb_dbg(CMLB_INFO,  cl, "   wrt_reinstr: %d; rd_reinstr: %d\n",
3216 	    cl->cl_g.dkg_write_reinstruct, cl->cl_g.dkg_read_reinstruct);
3217 
3218 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
3219 
3220 	return (label_error);
3221 }
3222 
3223 
3224 /*
3225  *    Function: cmlb_build_default_label
3226  *
3227  * Description: Generate a default label for those devices that do not have
3228  *		one, e.g., new media, removable cartridges, etc..
3229  *
3230  *     Context: Kernel thread only
3231  */
3232 /*ARGSUSED*/
3233 static void
3234 cmlb_build_default_label(struct cmlb_lun *cl, void *tg_cookie)
3235 {
3236 #if defined(_SUNOS_VTOC_16)
3237 	uint_t	phys_spc;
3238 	uint_t	disksize;
3239 	struct  dk_geom cl_g;
3240 	diskaddr_t capacity;
3241 #endif
3242 
3243 	ASSERT(cl != NULL);
3244 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
3245 
3246 #if defined(_SUNOS_VTOC_8)
3247 	/*
3248 	 * Note: This is a legacy check for non-removable devices on VTOC_8
3249 	 * only. This may be a valid check for VTOC_16 as well.
3250 	 * Once we understand why there is this difference between SPARC and
3251 	 * x86 platform, we could remove this legacy check.
3252 	 */
3253 	if (!ISREMOVABLE(cl) && !ISHOTPLUGGABLE(cl)) {
3254 		return;
3255 	}
3256 #endif
3257 
3258 	bzero(&cl->cl_g, sizeof (struct dk_geom));
3259 	bzero(&cl->cl_vtoc, sizeof (struct dk_vtoc));
3260 	bzero(&cl->cl_map, NDKMAP * (sizeof (struct dk_map)));
3261 
3262 #if defined(_SUNOS_VTOC_8)
3263 
3264 	/*
3265 	 * It's a REMOVABLE media, therefore no label (on sparc, anyway).
3266 	 * But it is still necessary to set up various geometry information,
3267 	 * and we are doing this here.
3268 	 */
3269 
3270 	/*
3271 	 * For the rpm, we use the minimum for the disk.  For the head, cyl,
3272 	 * and number of sector per track, if the capacity <= 1GB, head = 64,
3273 	 * sect = 32.  else head = 255, sect 63 Note: the capacity should be
3274 	 * equal to C*H*S values.  This will cause some truncation of size due
3275 	 * to round off errors. For CD-ROMs, this truncation can have adverse
3276 	 * side effects, so returning ncyl and nhead as 1. The nsect will
3277 	 * overflow for most of CD-ROMs as nsect is of type ushort. (4190569)
3278 	 */
3279 	cl->cl_solaris_size = cl->cl_blockcount;
3280 	if (ISCD(cl)) {
3281 		tg_attribute_t tgattribute;
3282 		int is_writable;
3283 		/*
3284 		 * Preserve the old behavior for non-writable
3285 		 * medias. Since dkg_nsect is a ushort, it
3286 		 * will lose bits as cdroms have more than
3287 		 * 65536 sectors. So if we recalculate
3288 		 * capacity, it will become much shorter.
3289 		 * But the dkg_* information is not
3290 		 * used for CDROMs so it is OK. But for
3291 		 * Writable CDs we need this information
3292 		 * to be valid (for newfs say). So we
3293 		 * make nsect and nhead > 1 that way
3294 		 * nsect can still stay within ushort limit
3295 		 * without losing any bits.
3296 		 */
3297 
3298 		bzero(&tgattribute, sizeof (tg_attribute_t));
3299 
3300 		mutex_exit(CMLB_MUTEX(cl));
3301 		is_writable =
3302 		    (DK_TG_GETATTRIBUTE(cl, &tgattribute, tg_cookie) == 0) ?
3303 		    tgattribute.media_is_writable : 1;
3304 		mutex_enter(CMLB_MUTEX(cl));
3305 
3306 		if (is_writable) {
3307 			cl->cl_g.dkg_nhead = 64;
3308 			cl->cl_g.dkg_nsect = 32;
3309 			cl->cl_g.dkg_ncyl = cl->cl_blockcount / (64 * 32);
3310 			cl->cl_solaris_size = (diskaddr_t)cl->cl_g.dkg_ncyl *
3311 			    cl->cl_g.dkg_nhead * cl->cl_g.dkg_nsect;
3312 		} else {
3313 			cl->cl_g.dkg_ncyl  = 1;
3314 			cl->cl_g.dkg_nhead = 1;
3315 			cl->cl_g.dkg_nsect = cl->cl_blockcount;
3316 		}
3317 	} else {
3318 		if (cl->cl_blockcount < 160) {
3319 			/* Less than 80K */
3320 			cl->cl_g.dkg_nhead = 1;
3321 			cl->cl_g.dkg_ncyl = cl->cl_blockcount;
3322 			cl->cl_g.dkg_nsect = 1;
3323 		} else if (cl->cl_blockcount <= 0x1000) {
3324 			/* unlabeled SCSI floppy device */
3325 			cl->cl_g.dkg_nhead = 2;
3326 			cl->cl_g.dkg_ncyl = 80;
3327 			cl->cl_g.dkg_nsect = cl->cl_blockcount / (2 * 80);
3328 		} else if (cl->cl_blockcount <= 0x200000) {
3329 			cl->cl_g.dkg_nhead = 64;
3330 			cl->cl_g.dkg_nsect = 32;
3331 			cl->cl_g.dkg_ncyl  = cl->cl_blockcount / (64 * 32);
3332 		} else {
3333 			cl->cl_g.dkg_nhead = 255;
3334 
3335 			cl->cl_g.dkg_nsect = ((cl->cl_blockcount +
3336 			    (UINT16_MAX * 255 * 63) - 1) /
3337 			    (UINT16_MAX * 255 * 63)) * 63;
3338 
3339 			if (cl->cl_g.dkg_nsect == 0)
3340 				cl->cl_g.dkg_nsect = (UINT16_MAX / 63) * 63;
3341 
3342 			cl->cl_g.dkg_ncyl = cl->cl_blockcount /
3343 			    (255 * cl->cl_g.dkg_nsect);
3344 		}
3345 
3346 		cl->cl_solaris_size =
3347 		    (diskaddr_t)cl->cl_g.dkg_ncyl * cl->cl_g.dkg_nhead *
3348 		    cl->cl_g.dkg_nsect;
3349 
3350 	}
3351 
3352 	cl->cl_g.dkg_acyl	= 0;
3353 	cl->cl_g.dkg_bcyl	= 0;
3354 	cl->cl_g.dkg_rpm	= 200;
3355 	cl->cl_asciilabel[0]	= '\0';
3356 	cl->cl_g.dkg_pcyl	= cl->cl_g.dkg_ncyl;
3357 
3358 	cl->cl_map[0].dkl_cylno = 0;
3359 	cl->cl_map[0].dkl_nblk  = cl->cl_solaris_size;
3360 
3361 	cl->cl_map[2].dkl_cylno = 0;
3362 	cl->cl_map[2].dkl_nblk  = cl->cl_solaris_size;
3363 
3364 #elif defined(_SUNOS_VTOC_16)
3365 
3366 	if (cl->cl_solaris_size == 0) {
3367 		/*
3368 		 * Got fdisk table but no solaris entry therefore
3369 		 * don't create a default label
3370 		 */
3371 		cl->cl_f_geometry_is_valid = B_TRUE;
3372 		return;
3373 	}
3374 
3375 	/*
3376 	 * For CDs we continue to use the physical geometry to calculate
3377 	 * number of cylinders. All other devices must convert the
3378 	 * physical geometry (cmlb_geom) to values that will fit
3379 	 * in a dk_geom structure.
3380 	 */
3381 	if (ISCD(cl)) {
3382 		phys_spc = cl->cl_pgeom.g_nhead * cl->cl_pgeom.g_nsect;
3383 	} else {
3384 		/* Convert physical geometry to disk geometry */
3385 		bzero(&cl_g, sizeof (struct dk_geom));
3386 
3387 		/*
3388 		 * Refer to comments related to off-by-1 at the
3389 		 * header of this file.
3390 		 * Before calculating geometry, capacity should be
3391 		 * decreased by 1.
3392 		 */
3393 
3394 		if (cl->cl_alter_behavior & CMLB_OFF_BY_ONE)
3395 			capacity = cl->cl_blockcount - 1;
3396 		else
3397 			capacity = cl->cl_blockcount;
3398 
3399 
3400 		cmlb_convert_geometry(cl, capacity, &cl_g, tg_cookie);
3401 		bcopy(&cl_g, &cl->cl_g, sizeof (cl->cl_g));
3402 		phys_spc = cl->cl_g.dkg_nhead * cl->cl_g.dkg_nsect;
3403 	}
3404 
3405 	if (phys_spc == 0)
3406 		return;
3407 	cl->cl_g.dkg_pcyl = cl->cl_solaris_size / phys_spc;
3408 	if (cl->cl_alter_behavior & CMLB_FAKE_LABEL_ONE_PARTITION) {
3409 		/* disable devid */
3410 		cl->cl_g.dkg_ncyl = cl->cl_g.dkg_pcyl;
3411 		disksize = cl->cl_solaris_size;
3412 	} else {
3413 		cl->cl_g.dkg_acyl = DK_ACYL;
3414 		cl->cl_g.dkg_ncyl = cl->cl_g.dkg_pcyl - DK_ACYL;
3415 		disksize = cl->cl_g.dkg_ncyl * phys_spc;
3416 	}
3417 
3418 	if (ISCD(cl)) {
3419 		/*
3420 		 * CD's don't use the "heads * sectors * cyls"-type of
3421 		 * geometry, but instead use the entire capacity of the media.
3422 		 */
3423 		disksize = cl->cl_solaris_size;
3424 		cl->cl_g.dkg_nhead = 1;
3425 		cl->cl_g.dkg_nsect = 1;
3426 		cl->cl_g.dkg_rpm =
3427 		    (cl->cl_pgeom.g_rpm == 0) ? 200 : cl->cl_pgeom.g_rpm;
3428 
3429 		cl->cl_vtoc.v_part[0].p_start = 0;
3430 		cl->cl_vtoc.v_part[0].p_size  = disksize;
3431 		cl->cl_vtoc.v_part[0].p_tag   = V_BACKUP;
3432 		cl->cl_vtoc.v_part[0].p_flag  = V_UNMNT;
3433 
3434 		cl->cl_map[0].dkl_cylno = 0;
3435 		cl->cl_map[0].dkl_nblk  = disksize;
3436 		cl->cl_offset[0] = 0;
3437 
3438 	} else {
3439 		/*
3440 		 * Hard disks and removable media cartridges
3441 		 */
3442 		cl->cl_g.dkg_rpm =
3443 		    (cl->cl_pgeom.g_rpm == 0) ? 3600: cl->cl_pgeom.g_rpm;
3444 		cl->cl_vtoc.v_sectorsz = cl->cl_sys_blocksize;
3445 
3446 		/* Add boot slice */
3447 		cl->cl_vtoc.v_part[8].p_start = 0;
3448 		cl->cl_vtoc.v_part[8].p_size  = phys_spc;
3449 		cl->cl_vtoc.v_part[8].p_tag   = V_BOOT;
3450 		cl->cl_vtoc.v_part[8].p_flag  = V_UNMNT;
3451 
3452 		cl->cl_map[8].dkl_cylno = 0;
3453 		cl->cl_map[8].dkl_nblk  = phys_spc;
3454 		cl->cl_offset[8] = 0;
3455 
3456 		if ((cl->cl_alter_behavior &
3457 		    CMLB_CREATE_ALTSLICE_VTOC_16_DTYPE_DIRECT) &&
3458 		    cl->cl_device_type == DTYPE_DIRECT) {
3459 			cl->cl_vtoc.v_part[9].p_start = phys_spc;
3460 			cl->cl_vtoc.v_part[9].p_size  = 2 * phys_spc;
3461 			cl->cl_vtoc.v_part[9].p_tag   = V_ALTSCTR;
3462 			cl->cl_vtoc.v_part[9].p_flag  = 0;
3463 
3464 			cl->cl_map[9].dkl_cylno = 1;
3465 			cl->cl_map[9].dkl_nblk  = 2 * phys_spc;
3466 			cl->cl_offset[9] = phys_spc;
3467 		}
3468 	}
3469 
3470 	cl->cl_g.dkg_apc = 0;
3471 
3472 	/* Add backup slice */
3473 	cl->cl_vtoc.v_part[2].p_start = 0;
3474 	cl->cl_vtoc.v_part[2].p_size  = disksize;
3475 	cl->cl_vtoc.v_part[2].p_tag   = V_BACKUP;
3476 	cl->cl_vtoc.v_part[2].p_flag  = V_UNMNT;
3477 
3478 	cl->cl_map[2].dkl_cylno = 0;
3479 	cl->cl_map[2].dkl_nblk  = disksize;
3480 	cl->cl_offset[2] = 0;
3481 
3482 	/*
3483 	 * single slice (s0) covering the entire disk
3484 	 */
3485 	if (cl->cl_alter_behavior & CMLB_FAKE_LABEL_ONE_PARTITION) {
3486 		cl->cl_vtoc.v_part[0].p_start = 0;
3487 		cl->cl_vtoc.v_part[0].p_tag   = V_UNASSIGNED;
3488 		cl->cl_vtoc.v_part[0].p_flag  = 0;
3489 		cl->cl_vtoc.v_part[0].p_size  = disksize;
3490 		cl->cl_map[0].dkl_cylno = 0;
3491 		cl->cl_map[0].dkl_nblk  = disksize;
3492 		cl->cl_offset[0] = 0;
3493 	}
3494 
3495 	(void) sprintf(cl->cl_vtoc.v_asciilabel, "DEFAULT cyl %d alt %d"
3496 	    " hd %d sec %d", cl->cl_g.dkg_ncyl, cl->cl_g.dkg_acyl,
3497 	    cl->cl_g.dkg_nhead, cl->cl_g.dkg_nsect);
3498 
3499 #else
3500 #error "No VTOC format defined."
3501 #endif
3502 
3503 	cl->cl_g.dkg_read_reinstruct  = 0;
3504 	cl->cl_g.dkg_write_reinstruct = 0;
3505 
3506 	cl->cl_g.dkg_intrlv = 1;
3507 
3508 	cl->cl_vtoc.v_sanity  = VTOC_SANE;
3509 	cl->cl_vtoc.v_nparts = V_NUMPAR;
3510 	cl->cl_vtoc.v_version = V_VERSION;
3511 
3512 	cl->cl_f_geometry_is_valid = B_TRUE;
3513 	cl->cl_label_from_media = CMLB_LABEL_UNDEF;
3514 
3515 	cmlb_dbg(CMLB_INFO,  cl,
3516 	    "cmlb_build_default_label: Default label created: "
3517 	    "cyl: %d\tacyl: %d\tnhead: %d\tnsect: %d\tcap: %d\n",
3518 	    cl->cl_g.dkg_ncyl, cl->cl_g.dkg_acyl, cl->cl_g.dkg_nhead,
3519 	    cl->cl_g.dkg_nsect, cl->cl_blockcount);
3520 }
3521 
3522 
3523 #if defined(_FIRMWARE_NEEDS_FDISK)
3524 /*
3525  * Max CHS values, as they are encoded into bytes, for 1022/254/63
3526  */
3527 #define	LBA_MAX_SECT	(63 | ((1022 & 0x300) >> 2))
3528 #define	LBA_MAX_CYL	(1022 & 0xFF)
3529 #define	LBA_MAX_HEAD	(254)
3530 
3531 
3532 /*
3533  *    Function: cmlb_has_max_chs_vals
3534  *
3535  * Description: Return B_TRUE if Cylinder-Head-Sector values are all at maximum.
3536  *
3537  *   Arguments: fdp - ptr to CHS info
3538  *
3539  * Return Code: True or false
3540  *
3541  *     Context: Any.
3542  */
3543 static boolean_t
3544 cmlb_has_max_chs_vals(struct ipart *fdp)
3545 {
3546 	return ((fdp->begcyl  == LBA_MAX_CYL)	&&
3547 	    (fdp->beghead == LBA_MAX_HEAD)	&&
3548 	    (fdp->begsect == LBA_MAX_SECT)	&&
3549 	    (fdp->endcyl  == LBA_MAX_CYL)	&&
3550 	    (fdp->endhead == LBA_MAX_HEAD)	&&
3551 	    (fdp->endsect == LBA_MAX_SECT));
3552 }
3553 #endif
3554 
3555 /*
3556  *    Function: cmlb_dkio_get_geometry
3557  *
3558  * Description: This routine is the driver entry point for handling user
3559  *		requests to get the device geometry (DKIOCGGEOM).
3560  *
3561  *   Arguments:
3562  *	arg		pointer to user provided dk_geom structure specifying
3563  *			the controller's notion of the current geometry.
3564  *
3565  *	flag 		this argument is a pass through to ddi_copyxxx()
3566  *			directly from the mode argument of ioctl().
3567  *
3568  *	tg_cookie	cookie from target driver to be passed back to target
3569  *			driver when we call back to it through tg_ops.
3570  *
3571  * Return Code: 0
3572  *		EFAULT
3573  *		ENXIO
3574  *		EIO
3575  */
3576 static int
3577 cmlb_dkio_get_geometry(struct cmlb_lun *cl, caddr_t arg, int flag,
3578     void *tg_cookie)
3579 {
3580 	struct dk_geom	*tmp_geom = NULL;
3581 	int		rval = 0;
3582 
3583 	/*
3584 	 * cmlb_validate_geometry does not spin a disk up
3585 	 * if it was spcl down. We need to make sure it
3586 	 * is ready.
3587 	 */
3588 	mutex_enter(CMLB_MUTEX(cl));
3589 	rval = cmlb_validate_geometry(cl, B_TRUE, 0, tg_cookie);
3590 #if defined(_SUNOS_VTOC_8)
3591 	if (rval == EINVAL &&
3592 	    cl->cl_alter_behavior & CMLB_FAKE_GEOM_LABEL_IOCTLS_VTOC8) {
3593 		/*
3594 		 * This is to return a default label geometry even when we
3595 		 * do not really assume a default label for the device.
3596 		 * dad driver utilizes this.
3597 		 */
3598 		if (cl->cl_blockcount <= CMLB_OLDVTOC_LIMIT) {
3599 			cmlb_setup_default_geometry(cl, tg_cookie);
3600 			rval = 0;
3601 		}
3602 	}
3603 #endif
3604 	if (rval) {
3605 		mutex_exit(CMLB_MUTEX(cl));
3606 		return (rval);
3607 	}
3608 
3609 #if defined(__i386) || defined(__amd64)
3610 	if (cl->cl_solaris_size == 0) {
3611 		mutex_exit(CMLB_MUTEX(cl));
3612 		return (EIO);
3613 	}
3614 #endif
3615 
3616 	/*
3617 	 * Make a local copy of the soft state geometry to avoid some potential
3618 	 * race conditions associated with holding the mutex and updating the
3619 	 * write_reinstruct value
3620 	 */
3621 	tmp_geom = kmem_zalloc(sizeof (struct dk_geom), KM_SLEEP);
3622 	bcopy(&cl->cl_g, tmp_geom, sizeof (struct dk_geom));
3623 
3624 	if (tmp_geom->dkg_write_reinstruct == 0) {
3625 		tmp_geom->dkg_write_reinstruct =
3626 		    (int)((int)(tmp_geom->dkg_nsect * tmp_geom->dkg_rpm *
3627 		    cmlb_rot_delay) / (int)60000);
3628 	}
3629 	mutex_exit(CMLB_MUTEX(cl));
3630 
3631 	rval = ddi_copyout(tmp_geom, (void *)arg, sizeof (struct dk_geom),
3632 	    flag);
3633 	if (rval != 0) {
3634 		rval = EFAULT;
3635 	}
3636 
3637 	kmem_free(tmp_geom, sizeof (struct dk_geom));
3638 	return (rval);
3639 
3640 }
3641 
3642 
3643 /*
3644  *    Function: cmlb_dkio_set_geometry
3645  *
3646  * Description: This routine is the driver entry point for handling user
3647  *		requests to set the device geometry (DKIOCSGEOM). The actual
3648  *		device geometry is not updated, just the driver "notion" of it.
3649  *
3650  *   Arguments:
3651  *	arg		pointer to user provided dk_geom structure used to set
3652  *			the controller's notion of the current geometry.
3653  *
3654  *	flag 		this argument is a pass through to ddi_copyxxx()
3655  *			directly from the mode argument of ioctl().
3656  *
3657  *	tg_cookie	cookie from target driver to be passed back to target
3658  *			driver when we call back to it through tg_ops.
3659  *
3660  * Return Code: 0
3661  *		EFAULT
3662  *		ENXIO
3663  *		EIO
3664  */
3665 static int
3666 cmlb_dkio_set_geometry(struct cmlb_lun *cl, caddr_t arg, int flag)
3667 {
3668 	struct dk_geom	*tmp_geom;
3669 	struct dk_map	*lp;
3670 	int		rval = 0;
3671 	int		i;
3672 
3673 
3674 #if defined(__i386) || defined(__amd64)
3675 	if (cl->cl_solaris_size == 0) {
3676 		return (EIO);
3677 	}
3678 #endif
3679 	/*
3680 	 * We need to copy the user specified geometry into local
3681 	 * storage and then update the softstate. We don't want to hold
3682 	 * the mutex and copyin directly from the user to the soft state
3683 	 */
3684 	tmp_geom = (struct dk_geom *)
3685 	    kmem_zalloc(sizeof (struct dk_geom), KM_SLEEP);
3686 	rval = ddi_copyin(arg, tmp_geom, sizeof (struct dk_geom), flag);
3687 	if (rval != 0) {
3688 		kmem_free(tmp_geom, sizeof (struct dk_geom));
3689 		return (EFAULT);
3690 	}
3691 
3692 	mutex_enter(CMLB_MUTEX(cl));
3693 	bcopy(tmp_geom, &cl->cl_g, sizeof (struct dk_geom));
3694 	for (i = 0; i < NDKMAP; i++) {
3695 		lp  = &cl->cl_map[i];
3696 		cl->cl_offset[i] =
3697 		    cl->cl_g.dkg_nhead * cl->cl_g.dkg_nsect * lp->dkl_cylno;
3698 #if defined(__i386) || defined(__amd64)
3699 		cl->cl_offset[i] += cl->cl_solaris_offset;
3700 #endif
3701 	}
3702 	cl->cl_f_geometry_is_valid = B_FALSE;
3703 	mutex_exit(CMLB_MUTEX(cl));
3704 	kmem_free(tmp_geom, sizeof (struct dk_geom));
3705 
3706 	return (rval);
3707 }
3708 
3709 /*
3710  *    Function: cmlb_dkio_get_partition
3711  *
3712  * Description: This routine is the driver entry point for handling user
3713  *		requests to get the partition table (DKIOCGAPART).
3714  *
3715  *   Arguments:
3716  *	arg		pointer to user provided dk_allmap structure specifying
3717  *			the controller's notion of the current partition table.
3718  *
3719  *	flag		this argument is a pass through to ddi_copyxxx()
3720  *			directly from the mode argument of ioctl().
3721  *
3722  *	tg_cookie	cookie from target driver to be passed back to target
3723  *			driver when we call back to it through tg_ops.
3724  *
3725  * Return Code: 0
3726  *		EFAULT
3727  *		ENXIO
3728  *		EIO
3729  */
3730 static int
3731 cmlb_dkio_get_partition(struct cmlb_lun *cl, caddr_t arg, int flag,
3732     void *tg_cookie)
3733 {
3734 	int		rval = 0;
3735 	int		size;
3736 
3737 	/*
3738 	 * Make sure the geometry is valid before getting the partition
3739 	 * information.
3740 	 */
3741 	mutex_enter(CMLB_MUTEX(cl));
3742 	if ((rval = cmlb_validate_geometry(cl, B_TRUE, 0, tg_cookie)) != 0) {
3743 		mutex_exit(CMLB_MUTEX(cl));
3744 		return (rval);
3745 	}
3746 	mutex_exit(CMLB_MUTEX(cl));
3747 
3748 #if defined(__i386) || defined(__amd64)
3749 	if (cl->cl_solaris_size == 0) {
3750 		return (EIO);
3751 	}
3752 #endif
3753 
3754 #ifdef _MULTI_DATAMODEL
3755 	switch (ddi_model_convert_from(flag & FMODELS)) {
3756 	case DDI_MODEL_ILP32: {
3757 		struct dk_map32 dk_map32[NDKMAP];
3758 		int		i;
3759 
3760 		for (i = 0; i < NDKMAP; i++) {
3761 			dk_map32[i].dkl_cylno = cl->cl_map[i].dkl_cylno;
3762 			dk_map32[i].dkl_nblk  = cl->cl_map[i].dkl_nblk;
3763 		}
3764 		size = NDKMAP * sizeof (struct dk_map32);
3765 		rval = ddi_copyout(dk_map32, (void *)arg, size, flag);
3766 		if (rval != 0) {
3767 			rval = EFAULT;
3768 		}
3769 		break;
3770 	}
3771 	case DDI_MODEL_NONE:
3772 		size = NDKMAP * sizeof (struct dk_map);
3773 		rval = ddi_copyout(cl->cl_map, (void *)arg, size, flag);
3774 		if (rval != 0) {
3775 			rval = EFAULT;
3776 		}
3777 		break;
3778 	}
3779 #else /* ! _MULTI_DATAMODEL */
3780 	size = NDKMAP * sizeof (struct dk_map);
3781 	rval = ddi_copyout(cl->cl_map, (void *)arg, size, flag);
3782 	if (rval != 0) {
3783 		rval = EFAULT;
3784 	}
3785 #endif /* _MULTI_DATAMODEL */
3786 	return (rval);
3787 }
3788 
3789 /*
3790  *    Function: cmlb_dkio_set_partition
3791  *
3792  * Description: This routine is the driver entry point for handling user
3793  *		requests to set the partition table (DKIOCSAPART). The actual
3794  *		device partition is not updated.
3795  *
3796  *   Arguments:
3797  *		arg  - pointer to user provided dk_allmap structure used to set
3798  *			the controller's notion of the partition table.
3799  *		flag - this argument is a pass through to ddi_copyxxx()
3800  *		       directly from the mode argument of ioctl().
3801  *
3802  * Return Code: 0
3803  *		EINVAL
3804  *		EFAULT
3805  *		ENXIO
3806  *		EIO
3807  */
3808 static int
3809 cmlb_dkio_set_partition(struct cmlb_lun *cl, caddr_t arg, int flag)
3810 {
3811 	struct dk_map	dk_map[NDKMAP];
3812 	struct dk_map	*lp;
3813 	int		rval = 0;
3814 	int		size;
3815 	int		i;
3816 #if defined(_SUNOS_VTOC_16)
3817 	struct dkl_partition	*vp;
3818 #endif
3819 
3820 	/*
3821 	 * Set the map for all logical partitions.  We lock
3822 	 * the priority just to make sure an interrupt doesn't
3823 	 * come in while the map is half updated.
3824 	 */
3825 	_NOTE(DATA_READABLE_WITHOUT_LOCK(cmlb_lun::cl_solaris_size))
3826 	mutex_enter(CMLB_MUTEX(cl));
3827 
3828 	if (cl->cl_blockcount > CMLB_OLDVTOC_LIMIT) {
3829 		mutex_exit(CMLB_MUTEX(cl));
3830 		return (ENOTSUP);
3831 	}
3832 	mutex_exit(CMLB_MUTEX(cl));
3833 	if (cl->cl_solaris_size == 0) {
3834 		return (EIO);
3835 	}
3836 
3837 #ifdef _MULTI_DATAMODEL
3838 	switch (ddi_model_convert_from(flag & FMODELS)) {
3839 	case DDI_MODEL_ILP32: {
3840 		struct dk_map32 dk_map32[NDKMAP];
3841 
3842 		size = NDKMAP * sizeof (struct dk_map32);
3843 		rval = ddi_copyin((void *)arg, dk_map32, size, flag);
3844 		if (rval != 0) {
3845 			return (EFAULT);
3846 		}
3847 		for (i = 0; i < NDKMAP; i++) {
3848 			dk_map[i].dkl_cylno = dk_map32[i].dkl_cylno;
3849 			dk_map[i].dkl_nblk  = dk_map32[i].dkl_nblk;
3850 		}
3851 		break;
3852 	}
3853 	case DDI_MODEL_NONE:
3854 		size = NDKMAP * sizeof (struct dk_map);
3855 		rval = ddi_copyin((void *)arg, dk_map, size, flag);
3856 		if (rval != 0) {
3857 			return (EFAULT);
3858 		}
3859 		break;
3860 	}
3861 #else /* ! _MULTI_DATAMODEL */
3862 	size = NDKMAP * sizeof (struct dk_map);
3863 	rval = ddi_copyin((void *)arg, dk_map, size, flag);
3864 	if (rval != 0) {
3865 		return (EFAULT);
3866 	}
3867 #endif /* _MULTI_DATAMODEL */
3868 
3869 	mutex_enter(CMLB_MUTEX(cl));
3870 	/* Note: The size used in this bcopy is set based upon the data model */
3871 	bcopy(dk_map, cl->cl_map, size);
3872 #if defined(_SUNOS_VTOC_16)
3873 	vp = (struct dkl_partition *)&(cl->cl_vtoc);
3874 #endif	/* defined(_SUNOS_VTOC_16) */
3875 	for (i = 0; i < NDKMAP; i++) {
3876 		lp  = &cl->cl_map[i];
3877 		cl->cl_offset[i] =
3878 		    cl->cl_g.dkg_nhead * cl->cl_g.dkg_nsect * lp->dkl_cylno;
3879 #if defined(_SUNOS_VTOC_16)
3880 		vp->p_start = cl->cl_offset[i];
3881 		vp->p_size = lp->dkl_nblk;
3882 		vp++;
3883 #endif	/* defined(_SUNOS_VTOC_16) */
3884 #if defined(__i386) || defined(__amd64)
3885 		cl->cl_offset[i] += cl->cl_solaris_offset;
3886 #endif
3887 	}
3888 	mutex_exit(CMLB_MUTEX(cl));
3889 	return (rval);
3890 }
3891 
3892 
3893 /*
3894  *    Function: cmlb_dkio_get_vtoc
3895  *
3896  * Description: This routine is the driver entry point for handling user
3897  *		requests to get the current volume table of contents
3898  *		(DKIOCGVTOC).
3899  *
3900  *   Arguments:
3901  *	arg		pointer to user provided vtoc structure specifying
3902  *			the current vtoc.
3903  *
3904  *	flag		this argument is a pass through to ddi_copyxxx()
3905  *			directly from the mode argument of ioctl().
3906  *
3907  *	tg_cookie	cookie from target driver to be passed back to target
3908  *			driver when we call back to it through tg_ops.
3909  *
3910  * Return Code: 0
3911  *		EFAULT
3912  *		ENXIO
3913  *		EIO
3914  */
3915 static int
3916 cmlb_dkio_get_vtoc(struct cmlb_lun *cl, caddr_t arg, int flag, void *tg_cookie)
3917 {
3918 #if defined(_SUNOS_VTOC_8)
3919 	struct vtoc	user_vtoc;
3920 #endif	/* defined(_SUNOS_VTOC_8) */
3921 	int		rval = 0;
3922 
3923 	mutex_enter(CMLB_MUTEX(cl));
3924 	if (cl->cl_blockcount > CMLB_OLDVTOC_LIMIT) {
3925 		mutex_exit(CMLB_MUTEX(cl));
3926 		return (EOVERFLOW);
3927 	}
3928 
3929 	rval = cmlb_validate_geometry(cl, B_TRUE, 0, tg_cookie);
3930 
3931 #if defined(_SUNOS_VTOC_8)
3932 	if (rval == EINVAL &&
3933 	    (cl->cl_alter_behavior & CMLB_FAKE_GEOM_LABEL_IOCTLS_VTOC8)) {
3934 		/*
3935 		 * This is to return a default label even when we do not
3936 		 * really assume a default label for the device.
3937 		 * dad driver utilizes this.
3938 		 */
3939 		if (cl->cl_blockcount <= CMLB_OLDVTOC_LIMIT) {
3940 			cmlb_setup_default_geometry(cl, tg_cookie);
3941 			rval = 0;
3942 		}
3943 	}
3944 #endif
3945 	if (rval) {
3946 		mutex_exit(CMLB_MUTEX(cl));
3947 		return (rval);
3948 	}
3949 
3950 #if defined(_SUNOS_VTOC_8)
3951 	cmlb_build_user_vtoc(cl, &user_vtoc);
3952 	mutex_exit(CMLB_MUTEX(cl));
3953 
3954 #ifdef _MULTI_DATAMODEL
3955 	switch (ddi_model_convert_from(flag & FMODELS)) {
3956 	case DDI_MODEL_ILP32: {
3957 		struct vtoc32 user_vtoc32;
3958 
3959 		vtoctovtoc32(user_vtoc, user_vtoc32);
3960 		if (ddi_copyout(&user_vtoc32, (void *)arg,
3961 		    sizeof (struct vtoc32), flag)) {
3962 			return (EFAULT);
3963 		}
3964 		break;
3965 	}
3966 
3967 	case DDI_MODEL_NONE:
3968 		if (ddi_copyout(&user_vtoc, (void *)arg,
3969 		    sizeof (struct vtoc), flag)) {
3970 			return (EFAULT);
3971 		}
3972 		break;
3973 	}
3974 #else /* ! _MULTI_DATAMODEL */
3975 	if (ddi_copyout(&user_vtoc, (void *)arg, sizeof (struct vtoc), flag)) {
3976 		return (EFAULT);
3977 	}
3978 #endif /* _MULTI_DATAMODEL */
3979 
3980 #elif defined(_SUNOS_VTOC_16)
3981 	mutex_exit(CMLB_MUTEX(cl));
3982 
3983 #ifdef _MULTI_DATAMODEL
3984 	/*
3985 	 * The cl_vtoc structure is a "struct dk_vtoc"  which is always
3986 	 * 32-bit to maintain compatibility with existing on-disk
3987 	 * structures.  Thus, we need to convert the structure when copying
3988 	 * it out to a datamodel-dependent "struct vtoc" in a 64-bit
3989 	 * program.  If the target is a 32-bit program, then no conversion
3990 	 * is necessary.
3991 	 */
3992 	/* LINTED: logical expression always true: op "||" */
3993 	ASSERT(sizeof (cl->cl_vtoc) == sizeof (struct vtoc32));
3994 	switch (ddi_model_convert_from(flag & FMODELS)) {
3995 	case DDI_MODEL_ILP32:
3996 		if (ddi_copyout(&(cl->cl_vtoc), (void *)arg,
3997 		    sizeof (cl->cl_vtoc), flag)) {
3998 			return (EFAULT);
3999 		}
4000 		break;
4001 
4002 	case DDI_MODEL_NONE: {
4003 		struct vtoc user_vtoc;
4004 
4005 		vtoc32tovtoc(cl->cl_vtoc, user_vtoc);
4006 		if (ddi_copyout(&user_vtoc, (void *)arg,
4007 		    sizeof (struct vtoc), flag)) {
4008 			return (EFAULT);
4009 		}
4010 		break;
4011 	}
4012 	}
4013 #else /* ! _MULTI_DATAMODEL */
4014 	if (ddi_copyout(&(cl->cl_vtoc), (void *)arg, sizeof (cl->cl_vtoc),
4015 	    flag)) {
4016 		return (EFAULT);
4017 	}
4018 #endif /* _MULTI_DATAMODEL */
4019 #else
4020 #error "No VTOC format defined."
4021 #endif
4022 
4023 	return (rval);
4024 }
4025 
4026 
4027 /*
4028  *    Function: cmlb_dkio_get_extvtoc
4029  */
4030 static int
4031 cmlb_dkio_get_extvtoc(struct cmlb_lun *cl, caddr_t arg, int flag,
4032     void *tg_cookie)
4033 {
4034 	struct extvtoc	ext_vtoc;
4035 #if defined(_SUNOS_VTOC_8)
4036 	struct vtoc	user_vtoc;
4037 #endif	/* defined(_SUNOS_VTOC_8) */
4038 	int		rval = 0;
4039 
4040 	bzero(&ext_vtoc, sizeof (struct extvtoc));
4041 	mutex_enter(CMLB_MUTEX(cl));
4042 	rval = cmlb_validate_geometry(cl, B_TRUE, 0, tg_cookie);
4043 
4044 #if defined(_SUNOS_VTOC_8)
4045 	if (rval == EINVAL &&
4046 	    (cl->cl_alter_behavior & CMLB_FAKE_GEOM_LABEL_IOCTLS_VTOC8)) {
4047 		/*
4048 		 * This is to return a default label even when we do not
4049 		 * really assume a default label for the device.
4050 		 * dad driver utilizes this.
4051 		 */
4052 		if (cl->cl_blockcount <= CMLB_OLDVTOC_LIMIT) {
4053 			cmlb_setup_default_geometry(cl, tg_cookie);
4054 			rval = 0;
4055 		}
4056 	}
4057 #endif
4058 	if (rval) {
4059 		mutex_exit(CMLB_MUTEX(cl));
4060 		return (rval);
4061 	}
4062 
4063 #if defined(_SUNOS_VTOC_8)
4064 	cmlb_build_user_vtoc(cl, &user_vtoc);
4065 	mutex_exit(CMLB_MUTEX(cl));
4066 
4067 	/*
4068 	 * Checking callers data model does not make much sense here
4069 	 * since extvtoc will always be equivalent to 64bit vtoc.
4070 	 * What is important is whether the kernel is in 32 or 64 bit
4071 	 */
4072 
4073 #ifdef _LP64
4074 		if (ddi_copyout(&user_vtoc, (void *)arg,
4075 		    sizeof (struct extvtoc), flag)) {
4076 			return (EFAULT);
4077 		}
4078 #else
4079 		vtoc32tovtoc(user_vtoc, ext_vtoc);
4080 		if (ddi_copyout(&ext_vtoc, (void *)arg,
4081 		    sizeof (struct extvtoc), flag)) {
4082 			return (EFAULT);
4083 		}
4084 #endif
4085 
4086 #elif defined(_SUNOS_VTOC_16)
4087 	/*
4088 	 * The cl_vtoc structure is a "struct dk_vtoc"  which is always
4089 	 * 32-bit to maintain compatibility with existing on-disk
4090 	 * structures.  Thus, we need to convert the structure when copying
4091 	 * it out to extvtoc
4092 	 */
4093 	vtoc32tovtoc(cl->cl_vtoc, ext_vtoc);
4094 	mutex_exit(CMLB_MUTEX(cl));
4095 
4096 	if (ddi_copyout(&ext_vtoc, (void *)arg, sizeof (struct extvtoc), flag))
4097 		return (EFAULT);
4098 #else
4099 #error "No VTOC format defined."
4100 #endif
4101 
4102 	return (rval);
4103 }
4104 
4105 /*
4106  * This routine implements the DKIOCGETEFI ioctl. This ioctl is currently
4107  * used to read the GPT Partition Table Header (primary/backup), the GUID
4108  * partition Entry Array (primary/backup), and the MBR.
4109  */
4110 static int
4111 cmlb_dkio_get_efi(struct cmlb_lun *cl, caddr_t arg, int flag, void *tg_cookie)
4112 {
4113 	dk_efi_t	user_efi;
4114 	int		rval = 0;
4115 	void		*buffer;
4116 	diskaddr_t	tgt_lba;
4117 
4118 	if (ddi_copyin(arg, &user_efi, sizeof (dk_efi_t), flag))
4119 		return (EFAULT);
4120 
4121 	user_efi.dki_data = (void *)(uintptr_t)user_efi.dki_data_64;
4122 
4123 	if (user_efi.dki_length == 0 ||
4124 	    user_efi.dki_length > cmlb_tg_max_efi_xfer)
4125 		return (EINVAL);
4126 
4127 	tgt_lba = user_efi.dki_lba;
4128 
4129 	mutex_enter(CMLB_MUTEX(cl));
4130 	if ((cmlb_check_update_blockcount(cl, tg_cookie) != 0) ||
4131 	    (cl->cl_tgt_blocksize == 0) ||
4132 	    (user_efi.dki_length % cl->cl_sys_blocksize)) {
4133 		mutex_exit(CMLB_MUTEX(cl));
4134 		return (EINVAL);
4135 	}
4136 	if (cl->cl_tgt_blocksize != cl->cl_sys_blocksize)
4137 		tgt_lba = tgt_lba * cl->cl_tgt_blocksize /
4138 		    cl->cl_sys_blocksize;
4139 	mutex_exit(CMLB_MUTEX(cl));
4140 
4141 	buffer = kmem_alloc(user_efi.dki_length, KM_SLEEP);
4142 	rval = DK_TG_READ(cl, buffer, tgt_lba, user_efi.dki_length, tg_cookie);
4143 	if (rval == 0 && ddi_copyout(buffer, user_efi.dki_data,
4144 	    user_efi.dki_length, flag) != 0)
4145 		rval = EFAULT;
4146 
4147 	kmem_free(buffer, user_efi.dki_length);
4148 	return (rval);
4149 }
4150 
4151 #if defined(_SUNOS_VTOC_8)
4152 /*
4153  *    Function: cmlb_build_user_vtoc
4154  *
4155  * Description: This routine populates a pass by reference variable with the
4156  *		current volume table of contents.
4157  *
4158  *   Arguments: cl - driver soft state (unit) structure
4159  *		user_vtoc - pointer to vtoc structure to be populated
4160  */
4161 static void
4162 cmlb_build_user_vtoc(struct cmlb_lun *cl, struct vtoc *user_vtoc)
4163 {
4164 	struct dk_map2		*lpart;
4165 	struct dk_map		*lmap;
4166 	struct partition	*vpart;
4167 	uint32_t		nblks;
4168 	int			i;
4169 
4170 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
4171 
4172 	/*
4173 	 * Return vtoc structure fields in the provided VTOC area, addressed
4174 	 * by *vtoc.
4175 	 */
4176 	bzero(user_vtoc, sizeof (struct vtoc));
4177 	user_vtoc->v_bootinfo[0] = cl->cl_vtoc.v_bootinfo[0];
4178 	user_vtoc->v_bootinfo[1] = cl->cl_vtoc.v_bootinfo[1];
4179 	user_vtoc->v_bootinfo[2] = cl->cl_vtoc.v_bootinfo[2];
4180 	user_vtoc->v_sanity	= VTOC_SANE;
4181 	user_vtoc->v_version	= cl->cl_vtoc.v_version;
4182 	bcopy(cl->cl_vtoc.v_volume, user_vtoc->v_volume, LEN_DKL_VVOL);
4183 	user_vtoc->v_sectorsz = cl->cl_sys_blocksize;
4184 	user_vtoc->v_nparts = cl->cl_vtoc.v_nparts;
4185 
4186 	for (i = 0; i < 10; i++)
4187 		user_vtoc->v_reserved[i] = cl->cl_vtoc.v_reserved[i];
4188 
4189 	/*
4190 	 * Convert partitioning information.
4191 	 *
4192 	 * Note the conversion from starting cylinder number
4193 	 * to starting sector number.
4194 	 */
4195 	lmap = cl->cl_map;
4196 	lpart = (struct dk_map2 *)cl->cl_vtoc.v_part;
4197 	vpart = user_vtoc->v_part;
4198 
4199 	nblks = cl->cl_g.dkg_nsect * cl->cl_g.dkg_nhead;
4200 
4201 	for (i = 0; i < V_NUMPAR; i++) {
4202 		vpart->p_tag	= lpart->p_tag;
4203 		vpart->p_flag	= lpart->p_flag;
4204 		vpart->p_start	= lmap->dkl_cylno * nblks;
4205 		vpart->p_size	= lmap->dkl_nblk;
4206 		lmap++;
4207 		lpart++;
4208 		vpart++;
4209 
4210 		/* (4364927) */
4211 		user_vtoc->timestamp[i] = (time_t)cl->cl_vtoc.v_timestamp[i];
4212 	}
4213 
4214 	bcopy(cl->cl_asciilabel, user_vtoc->v_asciilabel, LEN_DKL_ASCII);
4215 }
4216 #endif
4217 
4218 static int
4219 cmlb_dkio_partition(struct cmlb_lun *cl, caddr_t arg, int flag,
4220     void *tg_cookie)
4221 {
4222 	struct partition64	p64;
4223 	int			rval = 0;
4224 	uint_t			nparts;
4225 	efi_gpe_t		*partitions;
4226 	efi_gpt_t		*buffer;
4227 	diskaddr_t		gpe_lba;
4228 	int			n_gpe_per_blk = 0;
4229 
4230 	if (ddi_copyin((const void *)arg, &p64,
4231 	    sizeof (struct partition64), flag)) {
4232 		return (EFAULT);
4233 	}
4234 
4235 	buffer = kmem_alloc(cl->cl_sys_blocksize, KM_SLEEP);
4236 	rval = DK_TG_READ(cl, buffer, 1, cl->cl_sys_blocksize, tg_cookie);
4237 	if (rval != 0)
4238 		goto done_error;
4239 
4240 	cmlb_swap_efi_gpt(buffer);
4241 
4242 	if ((rval = cmlb_validate_efi(buffer)) != 0)
4243 		goto done_error;
4244 
4245 	nparts = buffer->efi_gpt_NumberOfPartitionEntries;
4246 	gpe_lba = buffer->efi_gpt_PartitionEntryLBA;
4247 	if (p64.p_partno >= nparts) {
4248 		/* couldn't find it */
4249 		rval = ESRCH;
4250 		goto done_error;
4251 	}
4252 	/*
4253 	 * Read the block that contains the requested GPE.
4254 	 */
4255 	n_gpe_per_blk = cl->cl_sys_blocksize / sizeof (efi_gpe_t);
4256 	gpe_lba += p64.p_partno / n_gpe_per_blk;
4257 	rval = DK_TG_READ(cl, buffer, gpe_lba, cl->cl_sys_blocksize, tg_cookie);
4258 
4259 	if (rval) {
4260 		goto done_error;
4261 	}
4262 	partitions = (efi_gpe_t *)buffer;
4263 	partitions += p64.p_partno % n_gpe_per_blk;
4264 
4265 	/* Byte swap only the requested GPE */
4266 	cmlb_swap_efi_gpe(1, partitions);
4267 
4268 	bcopy(&partitions->efi_gpe_PartitionTypeGUID, &p64.p_type,
4269 	    sizeof (struct uuid));
4270 	p64.p_start = partitions->efi_gpe_StartingLBA;
4271 	p64.p_size = partitions->efi_gpe_EndingLBA -
4272 	    p64.p_start + 1;
4273 
4274 	if (ddi_copyout(&p64, (void *)arg, sizeof (struct partition64), flag))
4275 		rval = EFAULT;
4276 
4277 done_error:
4278 	kmem_free(buffer, cl->cl_sys_blocksize);
4279 	return (rval);
4280 }
4281 
4282 
4283 /*
4284  *    Function: cmlb_dkio_set_vtoc
4285  *
4286  * Description: This routine is the driver entry point for handling user
4287  *		requests to set the current volume table of contents
4288  *		(DKIOCSVTOC).
4289  *
4290  *   Arguments:
4291  *	dev		the device number
4292  *	arg		pointer to user provided vtoc structure used to set the
4293  *			current vtoc.
4294  *
4295  *	flag		this argument is a pass through to ddi_copyxxx()
4296  *			directly from the mode argument of ioctl().
4297  *
4298  *	tg_cookie	cookie from target driver to be passed back to target
4299  *			driver when we call back to it through tg_ops.
4300  *
4301  * Return Code: 0
4302  *		EFAULT
4303  *		ENXIO
4304  *		EINVAL
4305  *		ENOTSUP
4306  */
4307 static int
4308 cmlb_dkio_set_vtoc(struct cmlb_lun *cl, dev_t dev, caddr_t arg, int flag,
4309     void *tg_cookie)
4310 {
4311 	struct vtoc	user_vtoc;
4312 	int		rval = 0;
4313 	boolean_t	internal;
4314 
4315 	internal = VOID2BOOLEAN(
4316 	    (cl->cl_alter_behavior & (CMLB_INTERNAL_MINOR_NODES)) != 0);
4317 
4318 #ifdef _MULTI_DATAMODEL
4319 	switch (ddi_model_convert_from(flag & FMODELS)) {
4320 	case DDI_MODEL_ILP32: {
4321 		struct vtoc32 user_vtoc32;
4322 
4323 		if (ddi_copyin((const void *)arg, &user_vtoc32,
4324 		    sizeof (struct vtoc32), flag)) {
4325 			return (EFAULT);
4326 		}
4327 		vtoc32tovtoc(user_vtoc32, user_vtoc);
4328 		break;
4329 	}
4330 
4331 	case DDI_MODEL_NONE:
4332 		if (ddi_copyin((const void *)arg, &user_vtoc,
4333 		    sizeof (struct vtoc), flag)) {
4334 			return (EFAULT);
4335 		}
4336 		break;
4337 	}
4338 #else /* ! _MULTI_DATAMODEL */
4339 	if (ddi_copyin((const void *)arg, &user_vtoc,
4340 	    sizeof (struct vtoc), flag)) {
4341 		return (EFAULT);
4342 	}
4343 #endif /* _MULTI_DATAMODEL */
4344 
4345 	mutex_enter(CMLB_MUTEX(cl));
4346 
4347 	if (cl->cl_blockcount > CMLB_OLDVTOC_LIMIT) {
4348 		mutex_exit(CMLB_MUTEX(cl));
4349 		return (EOVERFLOW);
4350 	}
4351 
4352 #if defined(__i386) || defined(__amd64)
4353 	if (cl->cl_tgt_blocksize != cl->cl_sys_blocksize) {
4354 		mutex_exit(CMLB_MUTEX(cl));
4355 		return (EINVAL);
4356 	}
4357 #endif
4358 
4359 	if (cl->cl_g.dkg_ncyl == 0) {
4360 		mutex_exit(CMLB_MUTEX(cl));
4361 		return (EINVAL);
4362 	}
4363 
4364 	mutex_exit(CMLB_MUTEX(cl));
4365 	cmlb_clear_efi(cl, tg_cookie);
4366 	ddi_remove_minor_node(CMLB_DEVINFO(cl), "wd");
4367 	ddi_remove_minor_node(CMLB_DEVINFO(cl), "wd,raw");
4368 
4369 	/*
4370 	 * cmlb_dkio_set_vtoc creates duplicate minor nodes when
4371 	 * relabeling an SMI disk. To avoid that we remove them
4372 	 * before creating.
4373 	 * It should be OK to remove a non-existed minor node.
4374 	 */
4375 	ddi_remove_minor_node(CMLB_DEVINFO(cl), "h");
4376 	ddi_remove_minor_node(CMLB_DEVINFO(cl), "h,raw");
4377 
4378 	(void) cmlb_create_minor(CMLB_DEVINFO(cl), "h",
4379 	    S_IFBLK, (CMLBUNIT(dev) << CMLBUNIT_SHIFT) | WD_NODE,
4380 	    cl->cl_node_type, NULL, internal);
4381 	(void) cmlb_create_minor(CMLB_DEVINFO(cl), "h,raw",
4382 	    S_IFCHR, (CMLBUNIT(dev) << CMLBUNIT_SHIFT) | WD_NODE,
4383 	    cl->cl_node_type, NULL, internal);
4384 	mutex_enter(CMLB_MUTEX(cl));
4385 
4386 	if ((rval = cmlb_build_label_vtoc(cl, &user_vtoc)) == 0) {
4387 		if ((rval = cmlb_write_label(cl, tg_cookie)) == 0) {
4388 			if (cmlb_validate_geometry(cl,
4389 			    B_TRUE, 0, tg_cookie) != 0) {
4390 				cmlb_dbg(CMLB_ERROR, cl,
4391 				    "cmlb_dkio_set_vtoc: "
4392 				    "Failed validate geometry\n");
4393 			}
4394 			cl->cl_msglog_flag |= CMLB_ALLOW_2TB_WARN;
4395 		}
4396 	}
4397 	mutex_exit(CMLB_MUTEX(cl));
4398 	return (rval);
4399 }
4400 
4401 /*
4402  *    Function: cmlb_dkio_set_extvtoc
4403  */
4404 static int
4405 cmlb_dkio_set_extvtoc(struct cmlb_lun *cl, dev_t dev, caddr_t arg, int flag,
4406     void *tg_cookie)
4407 {
4408 	int		rval = 0;
4409 	struct vtoc	user_vtoc;
4410 	boolean_t	internal;
4411 
4412 
4413 	/*
4414 	 * Checking callers data model does not make much sense here
4415 	 * since extvtoc will always be equivalent to 64bit vtoc.
4416 	 * What is important is whether the kernel is in 32 or 64 bit
4417 	 */
4418 
4419 #ifdef _LP64
4420 	if (ddi_copyin((const void *)arg, &user_vtoc,
4421 		    sizeof (struct extvtoc), flag)) {
4422 			return (EFAULT);
4423 	}
4424 #else
4425 	struct	extvtoc	user_extvtoc;
4426 	if (ddi_copyin((const void *)arg, &user_extvtoc,
4427 		    sizeof (struct extvtoc), flag)) {
4428 			return (EFAULT);
4429 	}
4430 
4431 	vtoctovtoc32(user_extvtoc, user_vtoc);
4432 #endif
4433 
4434 	internal = VOID2BOOLEAN(
4435 	    (cl->cl_alter_behavior & (CMLB_INTERNAL_MINOR_NODES)) != 0);
4436 	mutex_enter(CMLB_MUTEX(cl));
4437 #if defined(__i386) || defined(__amd64)
4438 	if (cl->cl_tgt_blocksize != cl->cl_sys_blocksize) {
4439 		mutex_exit(CMLB_MUTEX(cl));
4440 		return (EINVAL);
4441 	}
4442 #endif
4443 
4444 	if (cl->cl_g.dkg_ncyl == 0) {
4445 		mutex_exit(CMLB_MUTEX(cl));
4446 		return (EINVAL);
4447 	}
4448 
4449 	mutex_exit(CMLB_MUTEX(cl));
4450 	cmlb_clear_efi(cl, tg_cookie);
4451 	ddi_remove_minor_node(CMLB_DEVINFO(cl), "wd");
4452 	ddi_remove_minor_node(CMLB_DEVINFO(cl), "wd,raw");
4453 	/*
4454 	 * cmlb_dkio_set_extvtoc creates duplicate minor nodes when
4455 	 * relabeling an SMI disk. To avoid that we remove them
4456 	 * before creating.
4457 	 * It should be OK to remove a non-existed minor node.
4458 	 */
4459 	ddi_remove_minor_node(CMLB_DEVINFO(cl), "h");
4460 	ddi_remove_minor_node(CMLB_DEVINFO(cl), "h,raw");
4461 
4462 	(void) cmlb_create_minor(CMLB_DEVINFO(cl), "h",
4463 	    S_IFBLK, (CMLBUNIT(dev) << CMLBUNIT_SHIFT) | WD_NODE,
4464 	    cl->cl_node_type, NULL, internal);
4465 	(void) cmlb_create_minor(CMLB_DEVINFO(cl), "h,raw",
4466 	    S_IFCHR, (CMLBUNIT(dev) << CMLBUNIT_SHIFT) | WD_NODE,
4467 	    cl->cl_node_type, NULL, internal);
4468 
4469 	mutex_enter(CMLB_MUTEX(cl));
4470 
4471 	if ((rval = cmlb_build_label_vtoc(cl, &user_vtoc)) == 0) {
4472 		if ((rval = cmlb_write_label(cl, tg_cookie)) == 0) {
4473 			if (cmlb_validate_geometry(cl,
4474 			    B_TRUE, 0, tg_cookie) != 0) {
4475 				cmlb_dbg(CMLB_ERROR, cl,
4476 				    "cmlb_dkio_set_vtoc: "
4477 				    "Failed validate geometry\n");
4478 			}
4479 		}
4480 	}
4481 	mutex_exit(CMLB_MUTEX(cl));
4482 	return (rval);
4483 }
4484 
4485 /*
4486  *    Function: cmlb_build_label_vtoc
4487  *
4488  * Description: This routine updates the driver soft state current volume table
4489  *		of contents based on a user specified vtoc.
4490  *
4491  *   Arguments: cl - driver soft state (unit) structure
4492  *		user_vtoc - pointer to vtoc structure specifying vtoc to be used
4493  *			    to update the driver soft state.
4494  *
4495  * Return Code: 0
4496  *		EINVAL
4497  */
4498 static int
4499 cmlb_build_label_vtoc(struct cmlb_lun *cl, struct vtoc *user_vtoc)
4500 {
4501 	struct dk_map		*lmap;
4502 	struct partition	*vpart;
4503 	uint_t			nblks;
4504 #if defined(_SUNOS_VTOC_8)
4505 	int			ncyl;
4506 	struct dk_map2		*lpart;
4507 #endif	/* defined(_SUNOS_VTOC_8) */
4508 	int			i;
4509 
4510 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
4511 
4512 	/* Sanity-check the vtoc */
4513 	if (user_vtoc->v_sanity != VTOC_SANE ||
4514 	    user_vtoc->v_sectorsz != cl->cl_sys_blocksize ||
4515 	    user_vtoc->v_nparts != V_NUMPAR) {
4516 		cmlb_dbg(CMLB_INFO,  cl,
4517 		    "cmlb_build_label_vtoc: vtoc not valid\n");
4518 		return (EINVAL);
4519 	}
4520 
4521 	nblks = cl->cl_g.dkg_nsect * cl->cl_g.dkg_nhead;
4522 	if (nblks == 0) {
4523 		cmlb_dbg(CMLB_INFO,  cl,
4524 		    "cmlb_build_label_vtoc: geom nblks is 0\n");
4525 		return (EINVAL);
4526 	}
4527 
4528 #if defined(_SUNOS_VTOC_8)
4529 	vpart = user_vtoc->v_part;
4530 	for (i = 0; i < V_NUMPAR; i++) {
4531 		if (((unsigned)vpart->p_start % nblks) != 0) {
4532 			cmlb_dbg(CMLB_INFO,  cl,
4533 			    "cmlb_build_label_vtoc: p_start not multiply of"
4534 			    "nblks part %d p_start %d nblks %d\n", i,
4535 			    vpart->p_start, nblks);
4536 			return (EINVAL);
4537 		}
4538 		ncyl = (unsigned)vpart->p_start / nblks;
4539 		ncyl += (unsigned)vpart->p_size / nblks;
4540 		if (((unsigned)vpart->p_size % nblks) != 0) {
4541 			ncyl++;
4542 		}
4543 		if (ncyl > (int)cl->cl_g.dkg_ncyl) {
4544 			cmlb_dbg(CMLB_INFO,  cl,
4545 			    "cmlb_build_label_vtoc: ncyl %d  > dkg_ncyl %d"
4546 			    "p_size %ld p_start %ld nblks %d  part number %d"
4547 			    "tag %d\n",
4548 			    ncyl, cl->cl_g.dkg_ncyl, vpart->p_size,
4549 			    vpart->p_start, nblks,
4550 			    i, vpart->p_tag);
4551 
4552 			return (EINVAL);
4553 		}
4554 		vpart++;
4555 	}
4556 #endif	/* defined(_SUNOS_VTOC_8) */
4557 
4558 	/* Put appropriate vtoc structure fields into the disk label */
4559 #if defined(_SUNOS_VTOC_16)
4560 	/*
4561 	 * The vtoc is always a 32bit data structure to maintain the
4562 	 * on-disk format. Convert "in place" instead of doing bcopy.
4563 	 */
4564 	vtoctovtoc32((*user_vtoc), (*((struct vtoc32 *)&(cl->cl_vtoc))));
4565 
4566 	/*
4567 	 * in the 16-slice vtoc, starting sectors are expressed in
4568 	 * numbers *relative* to the start of the Solaris fdisk partition.
4569 	 */
4570 	lmap = cl->cl_map;
4571 	vpart = user_vtoc->v_part;
4572 
4573 	for (i = 0; i < (int)user_vtoc->v_nparts; i++, lmap++, vpart++) {
4574 		lmap->dkl_cylno = (unsigned)vpart->p_start / nblks;
4575 		lmap->dkl_nblk = (unsigned)vpart->p_size;
4576 	}
4577 
4578 #elif defined(_SUNOS_VTOC_8)
4579 
4580 	cl->cl_vtoc.v_bootinfo[0] = (uint32_t)user_vtoc->v_bootinfo[0];
4581 	cl->cl_vtoc.v_bootinfo[1] = (uint32_t)user_vtoc->v_bootinfo[1];
4582 	cl->cl_vtoc.v_bootinfo[2] = (uint32_t)user_vtoc->v_bootinfo[2];
4583 
4584 	cl->cl_vtoc.v_sanity = (uint32_t)user_vtoc->v_sanity;
4585 	cl->cl_vtoc.v_version = (uint32_t)user_vtoc->v_version;
4586 
4587 	bcopy(user_vtoc->v_volume, cl->cl_vtoc.v_volume, LEN_DKL_VVOL);
4588 
4589 	cl->cl_vtoc.v_nparts = user_vtoc->v_nparts;
4590 
4591 	for (i = 0; i < 10; i++)
4592 		cl->cl_vtoc.v_reserved[i] =  user_vtoc->v_reserved[i];
4593 
4594 	/*
4595 	 * Note the conversion from starting sector number
4596 	 * to starting cylinder number.
4597 	 * Return error if division results in a remainder.
4598 	 */
4599 	lmap = cl->cl_map;
4600 	lpart = cl->cl_vtoc.v_part;
4601 	vpart = user_vtoc->v_part;
4602 
4603 	for (i = 0; i < (int)user_vtoc->v_nparts; i++) {
4604 		lpart->p_tag  = vpart->p_tag;
4605 		lpart->p_flag = vpart->p_flag;
4606 		lmap->dkl_cylno = (unsigned)vpart->p_start / nblks;
4607 		lmap->dkl_nblk = (unsigned)vpart->p_size;
4608 
4609 		lmap++;
4610 		lpart++;
4611 		vpart++;
4612 
4613 		/* (4387723) */
4614 #ifdef _LP64
4615 		if (user_vtoc->timestamp[i] > TIME32_MAX) {
4616 			cl->cl_vtoc.v_timestamp[i] = TIME32_MAX;
4617 		} else {
4618 			cl->cl_vtoc.v_timestamp[i] = user_vtoc->timestamp[i];
4619 		}
4620 #else
4621 		cl->cl_vtoc.v_timestamp[i] = user_vtoc->timestamp[i];
4622 #endif
4623 	}
4624 
4625 	bcopy(user_vtoc->v_asciilabel, cl->cl_asciilabel, LEN_DKL_ASCII);
4626 #else
4627 #error "No VTOC format defined."
4628 #endif
4629 	return (0);
4630 }
4631 
4632 /*
4633  *    Function: cmlb_clear_efi
4634  *
4635  * Description: This routine clears all EFI labels.
4636  *
4637  *   Arguments:
4638  *	cl		 driver soft state (unit) structure
4639  *
4640  *	tg_cookie	cookie from target driver to be passed back to target
4641  *			driver when we call back to it through tg_ops.
4642  * Return Code: void
4643  */
4644 static void
4645 cmlb_clear_efi(struct cmlb_lun *cl, void *tg_cookie)
4646 {
4647 	efi_gpt_t	*gpt;
4648 	diskaddr_t	cap;
4649 	int		rval;
4650 
4651 	ASSERT(!mutex_owned(CMLB_MUTEX(cl)));
4652 
4653 	mutex_enter(CMLB_MUTEX(cl));
4654 	cl->cl_reserved = -1;
4655 	mutex_exit(CMLB_MUTEX(cl));
4656 
4657 	gpt = kmem_alloc(cl->cl_sys_blocksize, KM_SLEEP);
4658 
4659 	if (DK_TG_READ(cl, gpt, 1, cl->cl_sys_blocksize, tg_cookie) != 0) {
4660 		goto done;
4661 	}
4662 
4663 	cmlb_swap_efi_gpt(gpt);
4664 	rval = cmlb_validate_efi(gpt);
4665 	if (rval == 0) {
4666 		/* clear primary */
4667 		bzero(gpt, sizeof (efi_gpt_t));
4668 		if (rval = DK_TG_WRITE(cl, gpt, 1, cl->cl_sys_blocksize,
4669 		    tg_cookie)) {
4670 			cmlb_dbg(CMLB_INFO,  cl,
4671 			    "cmlb_clear_efi: clear primary label failed\n");
4672 		}
4673 	}
4674 	/* the backup */
4675 	rval = DK_TG_GETCAP(cl, &cap, tg_cookie);
4676 	if (rval) {
4677 		goto done;
4678 	}
4679 
4680 	if ((rval = DK_TG_READ(cl, gpt, cap - 1, cl->cl_sys_blocksize,
4681 	    tg_cookie)) != 0) {
4682 		goto done;
4683 	}
4684 	cmlb_swap_efi_gpt(gpt);
4685 	rval = cmlb_validate_efi(gpt);
4686 	if (rval == 0) {
4687 		/* clear backup */
4688 		cmlb_dbg(CMLB_TRACE,  cl,
4689 		    "cmlb_clear_efi clear backup@%lu\n", cap - 1);
4690 		bzero(gpt, sizeof (efi_gpt_t));
4691 		if ((rval = DK_TG_WRITE(cl,  gpt, cap - 1, cl->cl_sys_blocksize,
4692 		    tg_cookie))) {
4693 			cmlb_dbg(CMLB_INFO,  cl,
4694 			    "cmlb_clear_efi: clear backup label failed\n");
4695 		}
4696 	} else {
4697 		/*
4698 		 * Refer to comments related to off-by-1 at the
4699 		 * header of this file
4700 		 */
4701 		if ((rval = DK_TG_READ(cl, gpt, cap - 2,
4702 		    cl->cl_sys_blocksize, tg_cookie)) != 0) {
4703 			goto done;
4704 		}
4705 		cmlb_swap_efi_gpt(gpt);
4706 		rval = cmlb_validate_efi(gpt);
4707 		if (rval == 0) {
4708 			/* clear legacy backup EFI label */
4709 			cmlb_dbg(CMLB_TRACE,  cl,
4710 			    "cmlb_clear_efi clear legacy backup@%lu\n",
4711 			    cap - 2);
4712 			bzero(gpt, sizeof (efi_gpt_t));
4713 			if ((rval = DK_TG_WRITE(cl,  gpt, cap - 2,
4714 			    cl->cl_sys_blocksize, tg_cookie))) {
4715 				cmlb_dbg(CMLB_INFO,  cl,
4716 				"cmlb_clear_efi: clear legacy backup label "
4717 				"failed\n");
4718 			}
4719 		}
4720 	}
4721 
4722 done:
4723 	kmem_free(gpt, cl->cl_sys_blocksize);
4724 }
4725 
4726 /*
4727  *    Function: cmlb_set_vtoc
4728  *
4729  * Description: This routine writes data to the appropriate positions
4730  *
4731  *   Arguments:
4732  *	cl		driver soft state (unit) structure
4733  *
4734  *	dkl		the data to be written
4735  *
4736  *	tg_cookie	cookie from target driver to be passed back to target
4737  *			driver when we call back to it through tg_ops.
4738  *
4739  * Return: void
4740  */
4741 static int
4742 cmlb_set_vtoc(struct cmlb_lun *cl, struct dk_label *dkl, void *tg_cookie)
4743 {
4744 	uint_t	label_addr;
4745 	int	sec;
4746 	diskaddr_t	blk;
4747 	int	head;
4748 	int	cyl;
4749 	int	rval;
4750 
4751 #if defined(__i386) || defined(__amd64)
4752 	label_addr = cl->cl_solaris_offset + DK_LABEL_LOC;
4753 #else
4754 	/* Write the primary label at block 0 of the solaris partition. */
4755 	label_addr = 0;
4756 #endif
4757 
4758 	rval = DK_TG_WRITE(cl, dkl, label_addr, cl->cl_sys_blocksize,
4759 	    tg_cookie);
4760 
4761 	if (rval != 0) {
4762 		return (rval);
4763 	}
4764 
4765 	/*
4766 	 * Calculate where the backup labels go.  They are always on
4767 	 * the last alternate cylinder, but some older drives put them
4768 	 * on head 2 instead of the last head.	They are always on the
4769 	 * first 5 odd sectors of the appropriate track.
4770 	 *
4771 	 * We have no choice at this point, but to believe that the
4772 	 * disk label is valid.	 Use the geometry of the disk
4773 	 * as described in the label.
4774 	 */
4775 	cyl  = dkl->dkl_ncyl  + dkl->dkl_acyl - 1;
4776 	head = dkl->dkl_nhead - 1;
4777 
4778 	/*
4779 	 * Write and verify the backup labels. Make sure we don't try to
4780 	 * write past the last cylinder.
4781 	 */
4782 	for (sec = 1; ((sec < 5 * 2 + 1) && (sec < dkl->dkl_nsect)); sec += 2) {
4783 		blk = (diskaddr_t)(
4784 		    (cyl * ((dkl->dkl_nhead * dkl->dkl_nsect) - dkl->dkl_apc)) +
4785 		    (head * dkl->dkl_nsect) + sec);
4786 #if defined(__i386) || defined(__amd64)
4787 		blk += cl->cl_solaris_offset;
4788 #endif
4789 		rval = DK_TG_WRITE(cl, dkl, blk, cl->cl_sys_blocksize,
4790 		    tg_cookie);
4791 		cmlb_dbg(CMLB_INFO,  cl,
4792 		"cmlb_set_vtoc: wrote backup label %llx\n", blk);
4793 		if (rval != 0) {
4794 			goto exit;
4795 		}
4796 	}
4797 exit:
4798 	return (rval);
4799 }
4800 
4801 /*
4802  *    Function: cmlb_clear_vtoc
4803  *
4804  * Description: This routine clears out the VTOC labels.
4805  *
4806  *   Arguments:
4807  *	cl		driver soft state (unit) structure
4808  *
4809  *	tg_cookie	cookie from target driver to be passed back to target
4810  *			driver when we call back to it through tg_ops.
4811  *
4812  * Return: void
4813  */
4814 static void
4815 cmlb_clear_vtoc(struct cmlb_lun *cl, void *tg_cookie)
4816 {
4817 	struct dk_label		*dkl;
4818 
4819 	mutex_exit(CMLB_MUTEX(cl));
4820 	dkl = kmem_zalloc(cl->cl_sys_blocksize, KM_SLEEP);
4821 	mutex_enter(CMLB_MUTEX(cl));
4822 	/*
4823 	 * cmlb_set_vtoc uses these fields in order to figure out
4824 	 * where to overwrite the backup labels
4825 	 */
4826 	dkl->dkl_apc    = cl->cl_g.dkg_apc;
4827 	dkl->dkl_ncyl   = cl->cl_g.dkg_ncyl;
4828 	dkl->dkl_acyl   = cl->cl_g.dkg_acyl;
4829 	dkl->dkl_nhead  = cl->cl_g.dkg_nhead;
4830 	dkl->dkl_nsect  = cl->cl_g.dkg_nsect;
4831 	mutex_exit(CMLB_MUTEX(cl));
4832 	(void) cmlb_set_vtoc(cl, dkl, tg_cookie);
4833 	kmem_free(dkl, cl->cl_sys_blocksize);
4834 
4835 	mutex_enter(CMLB_MUTEX(cl));
4836 }
4837 
4838 /*
4839  *    Function: cmlb_write_label
4840  *
4841  * Description: This routine will validate and write the driver soft state vtoc
4842  *		contents to the device.
4843  *
4844  *   Arguments:
4845  *	cl		cmlb handle
4846  *
4847  *	tg_cookie	cookie from target driver to be passed back to target
4848  *			driver when we call back to it through tg_ops.
4849  *
4850  *
4851  * Return Code: the code returned by cmlb_send_scsi_cmd()
4852  *		0
4853  *		EINVAL
4854  *		ENXIO
4855  *		ENOMEM
4856  */
4857 static int
4858 cmlb_write_label(struct cmlb_lun *cl, void *tg_cookie)
4859 {
4860 	struct dk_label	*dkl;
4861 	short		sum;
4862 	short		*sp;
4863 	int		i;
4864 	int		rval;
4865 
4866 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
4867 	mutex_exit(CMLB_MUTEX(cl));
4868 	dkl = kmem_zalloc(cl->cl_sys_blocksize, KM_SLEEP);
4869 	mutex_enter(CMLB_MUTEX(cl));
4870 
4871 	bcopy(&cl->cl_vtoc, &dkl->dkl_vtoc, sizeof (struct dk_vtoc));
4872 	dkl->dkl_rpm	= cl->cl_g.dkg_rpm;
4873 	dkl->dkl_pcyl	= cl->cl_g.dkg_pcyl;
4874 	dkl->dkl_apc	= cl->cl_g.dkg_apc;
4875 	dkl->dkl_intrlv = cl->cl_g.dkg_intrlv;
4876 	dkl->dkl_ncyl	= cl->cl_g.dkg_ncyl;
4877 	dkl->dkl_acyl	= cl->cl_g.dkg_acyl;
4878 	dkl->dkl_nhead	= cl->cl_g.dkg_nhead;
4879 	dkl->dkl_nsect	= cl->cl_g.dkg_nsect;
4880 
4881 #if defined(_SUNOS_VTOC_8)
4882 	dkl->dkl_obs1	= cl->cl_g.dkg_obs1;
4883 	dkl->dkl_obs2	= cl->cl_g.dkg_obs2;
4884 	dkl->dkl_obs3	= cl->cl_g.dkg_obs3;
4885 	for (i = 0; i < NDKMAP; i++) {
4886 		dkl->dkl_map[i].dkl_cylno = cl->cl_map[i].dkl_cylno;
4887 		dkl->dkl_map[i].dkl_nblk  = cl->cl_map[i].dkl_nblk;
4888 	}
4889 	bcopy(cl->cl_asciilabel, dkl->dkl_asciilabel, LEN_DKL_ASCII);
4890 #elif defined(_SUNOS_VTOC_16)
4891 	dkl->dkl_skew	= cl->cl_dkg_skew;
4892 #else
4893 #error "No VTOC format defined."
4894 #endif
4895 
4896 	dkl->dkl_magic			= DKL_MAGIC;
4897 	dkl->dkl_write_reinstruct	= cl->cl_g.dkg_write_reinstruct;
4898 	dkl->dkl_read_reinstruct	= cl->cl_g.dkg_read_reinstruct;
4899 
4900 	/* Construct checksum for the new disk label */
4901 	sum = 0;
4902 	sp = (short *)dkl;
4903 	i = sizeof (struct dk_label) / sizeof (short);
4904 	while (i--) {
4905 		sum ^= *sp++;
4906 	}
4907 	dkl->dkl_cksum = sum;
4908 
4909 	mutex_exit(CMLB_MUTEX(cl));
4910 
4911 	rval = cmlb_set_vtoc(cl, dkl, tg_cookie);
4912 exit:
4913 	kmem_free(dkl, cl->cl_sys_blocksize);
4914 	mutex_enter(CMLB_MUTEX(cl));
4915 	return (rval);
4916 }
4917 
4918 /*
4919  * This routine implements the DKIOCSETEFI ioctl. This ioctl is currently
4920  * used to write (or clear) the GPT Partition Table header (primary/backup)
4921  * and GUID partition Entry Array (primary/backup). It is also used to write
4922  * the Protective MBR.
4923  */
4924 static int
4925 cmlb_dkio_set_efi(struct cmlb_lun *cl, dev_t dev, caddr_t arg, int flag,
4926     void *tg_cookie)
4927 {
4928 	dk_efi_t	user_efi;
4929 	int		rval = 0;
4930 	void		*buffer;
4931 	diskaddr_t	tgt_lba;
4932 	boolean_t	internal;
4933 
4934 	if (ddi_copyin(arg, &user_efi, sizeof (dk_efi_t), flag))
4935 		return (EFAULT);
4936 
4937 	internal = VOID2BOOLEAN(
4938 	    (cl->cl_alter_behavior & (CMLB_INTERNAL_MINOR_NODES)) != 0);
4939 
4940 	user_efi.dki_data = (void *)(uintptr_t)user_efi.dki_data_64;
4941 
4942 	if (user_efi.dki_length == 0 ||
4943 	    user_efi.dki_length > cmlb_tg_max_efi_xfer)
4944 		return (EINVAL);
4945 
4946 	tgt_lba = user_efi.dki_lba;
4947 
4948 	mutex_enter(CMLB_MUTEX(cl));
4949 	if ((cmlb_check_update_blockcount(cl, tg_cookie) != 0) ||
4950 	    (cl->cl_tgt_blocksize == 0) ||
4951 	    (user_efi.dki_length % cl->cl_sys_blocksize)) {
4952 		mutex_exit(CMLB_MUTEX(cl));
4953 		return (EINVAL);
4954 	}
4955 	if (cl->cl_tgt_blocksize != cl->cl_sys_blocksize)
4956 		tgt_lba = tgt_lba *
4957 		    cl->cl_tgt_blocksize / cl->cl_sys_blocksize;
4958 	mutex_exit(CMLB_MUTEX(cl));
4959 
4960 	buffer = kmem_alloc(user_efi.dki_length, KM_SLEEP);
4961 	if (ddi_copyin(user_efi.dki_data, buffer, user_efi.dki_length, flag)) {
4962 		rval = EFAULT;
4963 	} else {
4964 		/*
4965 		 * let's clear the vtoc labels and clear the softstate
4966 		 * vtoc.
4967 		 */
4968 		mutex_enter(CMLB_MUTEX(cl));
4969 		if (cl->cl_vtoc.v_sanity == VTOC_SANE) {
4970 			cmlb_dbg(CMLB_TRACE,  cl,
4971 			    "cmlb_dkio_set_efi: CLEAR VTOC\n");
4972 			if (cl->cl_label_from_media == CMLB_LABEL_VTOC)
4973 				cmlb_clear_vtoc(cl, tg_cookie);
4974 			bzero(&cl->cl_vtoc, sizeof (struct dk_vtoc));
4975 			mutex_exit(CMLB_MUTEX(cl));
4976 			ddi_remove_minor_node(CMLB_DEVINFO(cl), "h");
4977 			ddi_remove_minor_node(CMLB_DEVINFO(cl), "h,raw");
4978 			(void) cmlb_create_minor(CMLB_DEVINFO(cl), "wd",
4979 			    S_IFBLK,
4980 			    (CMLBUNIT(dev) << CMLBUNIT_SHIFT) | WD_NODE,
4981 			    cl->cl_node_type, NULL, internal);
4982 			(void) cmlb_create_minor(CMLB_DEVINFO(cl), "wd,raw",
4983 			    S_IFCHR,
4984 			    (CMLBUNIT(dev) << CMLBUNIT_SHIFT) | WD_NODE,
4985 			    cl->cl_node_type, NULL, internal);
4986 		} else
4987 			mutex_exit(CMLB_MUTEX(cl));
4988 
4989 		rval = DK_TG_WRITE(cl, buffer, tgt_lba, user_efi.dki_length,
4990 		    tg_cookie);
4991 
4992 		if (rval == 0) {
4993 			mutex_enter(CMLB_MUTEX(cl));
4994 			cl->cl_f_geometry_is_valid = B_FALSE;
4995 			mutex_exit(CMLB_MUTEX(cl));
4996 		}
4997 	}
4998 	kmem_free(buffer, user_efi.dki_length);
4999 	return (rval);
5000 }
5001 
5002 /*
5003  *    Function: cmlb_dkio_get_mboot
5004  *
5005  * Description: This routine is the driver entry point for handling user
5006  *		requests to get the current device mboot (DKIOCGMBOOT)
5007  *
5008  *   Arguments:
5009  *	arg		pointer to user provided mboot structure specifying
5010  *			the current mboot.
5011  *
5012  *	flag		this argument is a pass through to ddi_copyxxx()
5013  *			directly from the mode argument of ioctl().
5014  *
5015  *	tg_cookie	cookie from target driver to be passed back to target
5016  *			driver when we call back to it through tg_ops.
5017  *
5018  * Return Code: 0
5019  *		EINVAL
5020  *		EFAULT
5021  *		ENXIO
5022  */
5023 static int
5024 cmlb_dkio_get_mboot(struct cmlb_lun *cl, caddr_t arg, int flag, void *tg_cookie)
5025 {
5026 	struct mboot	*mboot;
5027 	int		rval;
5028 	size_t		buffer_size;
5029 
5030 
5031 #if defined(_SUNOS_VTOC_8)
5032 	if ((!ISREMOVABLE(cl) && !ISHOTPLUGGABLE(cl)) || (arg == NULL)) {
5033 #elif defined(_SUNOS_VTOC_16)
5034 	if (arg == NULL) {
5035 #endif
5036 		return (EINVAL);
5037 	}
5038 
5039 	/*
5040 	 * Read the mboot block, located at absolute block 0 on the target.
5041 	 */
5042 	buffer_size = cl->cl_sys_blocksize;
5043 
5044 	cmlb_dbg(CMLB_TRACE,  cl,
5045 	    "cmlb_dkio_get_mboot: allocation size: 0x%x\n", buffer_size);
5046 
5047 	mboot = kmem_zalloc(buffer_size, KM_SLEEP);
5048 	if ((rval = DK_TG_READ(cl, mboot, 0, buffer_size, tg_cookie)) == 0) {
5049 		if (ddi_copyout(mboot, (void *)arg,
5050 		    sizeof (struct mboot), flag) != 0) {
5051 			rval = EFAULT;
5052 		}
5053 	}
5054 	kmem_free(mboot, buffer_size);
5055 	return (rval);
5056 }
5057 
5058 
5059 /*
5060  *    Function: cmlb_dkio_set_mboot
5061  *
5062  * Description: This routine is the driver entry point for handling user
5063  *		requests to validate and set the device master boot
5064  *		(DKIOCSMBOOT).
5065  *
5066  *   Arguments:
5067  *	arg		pointer to user provided mboot structure used to set the
5068  *			master boot.
5069  *
5070  *	flag		this argument is a pass through to ddi_copyxxx()
5071  *			directly from the mode argument of ioctl().
5072  *
5073  *	tg_cookie	cookie from target driver to be passed back to target
5074  *			driver when we call back to it through tg_ops.
5075  *
5076  * Return Code: 0
5077  *		EINVAL
5078  *		EFAULT
5079  *		ENXIO
5080  */
5081 static int
5082 cmlb_dkio_set_mboot(struct cmlb_lun *cl, caddr_t arg, int flag, void *tg_cookie)
5083 {
5084 	struct mboot	*mboot = NULL;
5085 	int		rval;
5086 	ushort_t	magic;
5087 
5088 
5089 	ASSERT(!mutex_owned(CMLB_MUTEX(cl)));
5090 
5091 #if defined(_SUNOS_VTOC_8)
5092 	if (!ISREMOVABLE(cl) && !ISHOTPLUGGABLE(cl)) {
5093 		return (EINVAL);
5094 	}
5095 #endif
5096 
5097 	if (arg == NULL) {
5098 		return (EINVAL);
5099 	}
5100 
5101 	mboot = kmem_zalloc(cl->cl_sys_blocksize, KM_SLEEP);
5102 
5103 	if (ddi_copyin((const void *)arg, mboot,
5104 	    cl->cl_sys_blocksize, flag) != 0) {
5105 		kmem_free(mboot, cl->cl_sys_blocksize);
5106 		return (EFAULT);
5107 	}
5108 
5109 	/* Is this really a master boot record? */
5110 	magic = LE_16(mboot->signature);
5111 	if (magic != MBB_MAGIC) {
5112 		kmem_free(mboot, cl->cl_sys_blocksize);
5113 		return (EINVAL);
5114 	}
5115 
5116 	rval = DK_TG_WRITE(cl, mboot, 0, cl->cl_sys_blocksize, tg_cookie);
5117 
5118 	mutex_enter(CMLB_MUTEX(cl));
5119 #if defined(__i386) || defined(__amd64)
5120 	if (rval == 0) {
5121 		/*
5122 		 * mboot has been written successfully.
5123 		 * update the fdisk and vtoc tables in memory
5124 		 */
5125 		rval = cmlb_update_fdisk_and_vtoc(cl, tg_cookie);
5126 		if ((!cl->cl_f_geometry_is_valid) || (rval != 0)) {
5127 			mutex_exit(CMLB_MUTEX(cl));
5128 			kmem_free(mboot, cl->cl_sys_blocksize);
5129 			return (rval);
5130 		}
5131 	}
5132 
5133 #ifdef __lock_lint
5134 	cmlb_setup_default_geometry(cl, tg_cookie);
5135 #endif
5136 
5137 #else
5138 	if (rval == 0) {
5139 		/*
5140 		 * mboot has been written successfully.
5141 		 * set up the default geometry and VTOC
5142 		 */
5143 		if (cl->cl_blockcount <= CMLB_EXTVTOC_LIMIT)
5144 			cmlb_setup_default_geometry(cl, tg_cookie);
5145 	}
5146 #endif
5147 	cl->cl_msglog_flag |= CMLB_ALLOW_2TB_WARN;
5148 	mutex_exit(CMLB_MUTEX(cl));
5149 	kmem_free(mboot, cl->cl_sys_blocksize);
5150 	return (rval);
5151 }
5152 
5153 
5154 #if defined(__i386) || defined(__amd64)
5155 /*ARGSUSED*/
5156 static int
5157 cmlb_dkio_set_ext_part(struct cmlb_lun *cl, caddr_t arg, int flag,
5158     void *tg_cookie)
5159 {
5160 	int fdisk_rval;
5161 	diskaddr_t capacity;
5162 
5163 	ASSERT(!mutex_owned(CMLB_MUTEX(cl)));
5164 
5165 	mutex_enter(CMLB_MUTEX(cl));
5166 	capacity = cl->cl_blockcount;
5167 	fdisk_rval = cmlb_read_fdisk(cl, capacity, tg_cookie);
5168 	if (fdisk_rval != 0) {
5169 		mutex_exit(CMLB_MUTEX(cl));
5170 		return (fdisk_rval);
5171 	}
5172 
5173 	mutex_exit(CMLB_MUTEX(cl));
5174 	return (fdisk_rval);
5175 }
5176 #endif
5177 
5178 /*
5179  *    Function: cmlb_setup_default_geometry
5180  *
5181  * Description: This local utility routine sets the default geometry as part of
5182  *		setting the device mboot.
5183  *
5184  *   Arguments:
5185  *	cl		driver soft state (unit) structure
5186  *
5187  *	tg_cookie	cookie from target driver to be passed back to target
5188  *			driver when we call back to it through tg_ops.
5189  *
5190  *
5191  * Note: This may be redundant with cmlb_build_default_label.
5192  */
5193 static void
5194 cmlb_setup_default_geometry(struct cmlb_lun *cl, void *tg_cookie)
5195 {
5196 	struct cmlb_geom	pgeom;
5197 	struct cmlb_geom	*pgeomp = &pgeom;
5198 	int			ret;
5199 	int			geom_base_cap = 1;
5200 
5201 
5202 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
5203 
5204 	/* zero out the soft state geometry and partition table. */
5205 	bzero(&cl->cl_g, sizeof (struct dk_geom));
5206 	bzero(&cl->cl_vtoc, sizeof (struct dk_vtoc));
5207 	bzero(cl->cl_map, NDKMAP * (sizeof (struct dk_map)));
5208 
5209 	/*
5210 	 * For the rpm, we use the minimum for the disk.
5211 	 * For the head, cyl and number of sector per track,
5212 	 * if the capacity <= 1GB, head = 64, sect = 32.
5213 	 * else head = 255, sect 63
5214 	 * Note: the capacity should be equal to C*H*S values.
5215 	 * This will cause some truncation of size due to
5216 	 * round off errors. For CD-ROMs, this truncation can
5217 	 * have adverse side effects, so returning ncyl and
5218 	 * nhead as 1. The nsect will overflow for most of
5219 	 * CD-ROMs as nsect is of type ushort.
5220 	 */
5221 	if (cl->cl_alter_behavior & CMLB_FAKE_GEOM_LABEL_IOCTLS_VTOC8) {
5222 		/*
5223 		 * newfs currently can not handle 255 ntracks for SPARC
5224 		 * so get the geometry from target driver instead of coming up
5225 		 * with one based on capacity.
5226 		 */
5227 		mutex_exit(CMLB_MUTEX(cl));
5228 		ret = DK_TG_GETPHYGEOM(cl, pgeomp, tg_cookie);
5229 		mutex_enter(CMLB_MUTEX(cl));
5230 
5231 		if (ret == 0) {
5232 			geom_base_cap = 0;
5233 		} else {
5234 			cmlb_dbg(CMLB_ERROR,  cl,
5235 			    "cmlb_setup_default_geometry: "
5236 			    "tg_getphygeom failed %d\n", ret);
5237 
5238 			/* do default setting, geometry based on capacity */
5239 		}
5240 	}
5241 
5242 	if (geom_base_cap) {
5243 		if (ISCD(cl)) {
5244 			cl->cl_g.dkg_ncyl = 1;
5245 			cl->cl_g.dkg_nhead = 1;
5246 			cl->cl_g.dkg_nsect = cl->cl_blockcount;
5247 		} else if (cl->cl_blockcount < 160) {
5248 			/* Less than 80K */
5249 			cl->cl_g.dkg_nhead = 1;
5250 			cl->cl_g.dkg_ncyl = cl->cl_blockcount;
5251 			cl->cl_g.dkg_nsect = 1;
5252 		} else if (cl->cl_blockcount <= 0x1000) {
5253 			/* Needed for unlabeled SCSI floppies. */
5254 			cl->cl_g.dkg_nhead = 2;
5255 			cl->cl_g.dkg_ncyl = 80;
5256 			cl->cl_g.dkg_pcyl = 80;
5257 			cl->cl_g.dkg_nsect = cl->cl_blockcount / (2 * 80);
5258 		} else if (cl->cl_blockcount <= 0x200000) {
5259 			cl->cl_g.dkg_nhead = 64;
5260 			cl->cl_g.dkg_nsect = 32;
5261 			cl->cl_g.dkg_ncyl = cl->cl_blockcount / (64 * 32);
5262 		} else {
5263 			cl->cl_g.dkg_nhead = 255;
5264 
5265 			cl->cl_g.dkg_nsect = ((cl->cl_blockcount +
5266 			    (UINT16_MAX * 255 * 63) - 1) /
5267 			    (UINT16_MAX * 255 * 63)) * 63;
5268 
5269 			if (cl->cl_g.dkg_nsect == 0)
5270 				cl->cl_g.dkg_nsect = (UINT16_MAX / 63) * 63;
5271 
5272 			cl->cl_g.dkg_ncyl = cl->cl_blockcount /
5273 			    (255 * cl->cl_g.dkg_nsect);
5274 		}
5275 
5276 		cl->cl_g.dkg_acyl = 0;
5277 		cl->cl_g.dkg_bcyl = 0;
5278 		cl->cl_g.dkg_intrlv = 1;
5279 		cl->cl_g.dkg_rpm = 200;
5280 		if (cl->cl_g.dkg_pcyl == 0)
5281 			cl->cl_g.dkg_pcyl = cl->cl_g.dkg_ncyl +
5282 			    cl->cl_g.dkg_acyl;
5283 	} else {
5284 		cl->cl_g.dkg_ncyl = (short)pgeomp->g_ncyl;
5285 		cl->cl_g.dkg_acyl = pgeomp->g_acyl;
5286 		cl->cl_g.dkg_nhead = pgeomp->g_nhead;
5287 		cl->cl_g.dkg_nsect = pgeomp->g_nsect;
5288 		cl->cl_g.dkg_intrlv = pgeomp->g_intrlv;
5289 		cl->cl_g.dkg_rpm = pgeomp->g_rpm;
5290 		cl->cl_g.dkg_pcyl = cl->cl_g.dkg_ncyl + cl->cl_g.dkg_acyl;
5291 	}
5292 
5293 	cl->cl_g.dkg_read_reinstruct = 0;
5294 	cl->cl_g.dkg_write_reinstruct = 0;
5295 	cl->cl_solaris_size = cl->cl_g.dkg_ncyl *
5296 	    cl->cl_g.dkg_nhead * cl->cl_g.dkg_nsect;
5297 
5298 	cl->cl_map['a'-'a'].dkl_cylno = 0;
5299 	cl->cl_map['a'-'a'].dkl_nblk = cl->cl_solaris_size;
5300 
5301 	cl->cl_map['c'-'a'].dkl_cylno = 0;
5302 	cl->cl_map['c'-'a'].dkl_nblk = cl->cl_solaris_size;
5303 
5304 	cl->cl_vtoc.v_part[2].p_tag   = V_BACKUP;
5305 	cl->cl_vtoc.v_part[2].p_flag  = V_UNMNT;
5306 	cl->cl_vtoc.v_nparts = V_NUMPAR;
5307 	cl->cl_vtoc.v_version = V_VERSION;
5308 	(void) sprintf((char *)cl->cl_asciilabel, "DEFAULT cyl %d alt %d"
5309 	    " hd %d sec %d", cl->cl_g.dkg_ncyl, cl->cl_g.dkg_acyl,
5310 	    cl->cl_g.dkg_nhead, cl->cl_g.dkg_nsect);
5311 
5312 	cl->cl_f_geometry_is_valid = B_FALSE;
5313 }
5314 
5315 
5316 #if defined(__i386) || defined(__amd64)
5317 /*
5318  *    Function: cmlb_update_fdisk_and_vtoc
5319  *
5320  * Description: This local utility routine updates the device fdisk and vtoc
5321  *		as part of setting the device mboot.
5322  *
5323  *   Arguments:
5324  *	cl		driver soft state (unit) structure
5325  *
5326  *	tg_cookie	cookie from target driver to be passed back to target
5327  *			driver when we call back to it through tg_ops.
5328  *
5329  *
5330  * Return Code: 0 for success or errno-type return code.
5331  *
5332  *    Note:x86: This looks like a duplicate of cmlb_validate_geometry(), but
5333  *		these did exist separately in x86 sd.c.
5334  */
5335 static int
5336 cmlb_update_fdisk_and_vtoc(struct cmlb_lun *cl, void *tg_cookie)
5337 {
5338 	int		count;
5339 	int		label_rc = 0;
5340 	int		fdisk_rval;
5341 	diskaddr_t	capacity;
5342 
5343 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
5344 
5345 	if (cmlb_check_update_blockcount(cl, tg_cookie) != 0)
5346 		return (EINVAL);
5347 
5348 #if defined(_SUNOS_VTOC_16)
5349 	/*
5350 	 * Set up the "whole disk" fdisk partition; this should always
5351 	 * exist, regardless of whether the disk contains an fdisk table
5352 	 * or vtoc.
5353 	 */
5354 	cl->cl_map[P0_RAW_DISK].dkl_cylno = 0;
5355 	cl->cl_map[P0_RAW_DISK].dkl_nblk = cl->cl_blockcount;
5356 #endif	/* defined(_SUNOS_VTOC_16) */
5357 
5358 	/*
5359 	 * copy the lbasize and capacity so that if they're
5360 	 * reset while we're not holding the CMLB_MUTEX(cl), we will
5361 	 * continue to use valid values after the CMLB_MUTEX(cl) is
5362 	 * reacquired.
5363 	 */
5364 	capacity = cl->cl_blockcount;
5365 
5366 	/*
5367 	 * refresh the logical and physical geometry caches.
5368 	 * (data from mode sense format/rigid disk geometry pages,
5369 	 * and scsi_ifgetcap("geometry").
5370 	 */
5371 	cmlb_resync_geom_caches(cl, capacity, tg_cookie);
5372 
5373 	/*
5374 	 * Only DIRECT ACCESS devices will have Scl labels.
5375 	 * CD's supposedly have a Scl label, too
5376 	 */
5377 	if (cl->cl_device_type == DTYPE_DIRECT || ISREMOVABLE(cl)) {
5378 		fdisk_rval = cmlb_read_fdisk(cl, capacity, tg_cookie);
5379 		if (fdisk_rval != 0) {
5380 			ASSERT(mutex_owned(CMLB_MUTEX(cl)));
5381 			return (fdisk_rval);
5382 		}
5383 
5384 		if (cl->cl_solaris_size <= DK_LABEL_LOC) {
5385 			/*
5386 			 * Found fdisk table but no Solaris partition entry,
5387 			 * so don't call cmlb_uselabel() and don't create
5388 			 * a default label.
5389 			 */
5390 			label_rc = 0;
5391 			cl->cl_f_geometry_is_valid = B_TRUE;
5392 			goto no_solaris_partition;
5393 		}
5394 	} else if (capacity < 0) {
5395 		ASSERT(mutex_owned(CMLB_MUTEX(cl)));
5396 		return (EINVAL);
5397 	}
5398 
5399 	/*
5400 	 * For Removable media We reach here if we have found a
5401 	 * SOLARIS PARTITION.
5402 	 * If cl_f_geometry_is_valid is B_FALSE it indicates that the SOLARIS
5403 	 * PARTITION has changed from the previous one, hence we will setup a
5404 	 * default VTOC in this case.
5405 	 */
5406 	if (!cl->cl_f_geometry_is_valid) {
5407 		/* if we get here it is writable */
5408 		/* we are called from SMBOOT, and after a write of fdisk */
5409 		cmlb_build_default_label(cl, tg_cookie);
5410 		label_rc = 0;
5411 	}
5412 
5413 no_solaris_partition:
5414 
5415 #if defined(_SUNOS_VTOC_16)
5416 	/*
5417 	 * If we have valid geometry, set up the remaining fdisk partitions.
5418 	 * Note that dkl_cylno is not used for the fdisk map entries, so
5419 	 * we set it to an entirely bogus value.
5420 	 */
5421 	for (count = 0; count < FDISK_PARTS; count++) {
5422 		cl->cl_map[FDISK_P1 + count].dkl_cylno = UINT32_MAX;
5423 		cl->cl_map[FDISK_P1 + count].dkl_nblk =
5424 		    cl->cl_fmap[count].fmap_nblk;
5425 		cl->cl_offset[FDISK_P1 + count] =
5426 		    cl->cl_fmap[count].fmap_start;
5427 	}
5428 #endif
5429 
5430 	for (count = 0; count < NDKMAP; count++) {
5431 #if defined(_SUNOS_VTOC_8)
5432 		struct dk_map *lp  = &cl->cl_map[count];
5433 		cl->cl_offset[count] =
5434 		    cl->cl_g.dkg_nhead * cl->cl_g.dkg_nsect * lp->dkl_cylno;
5435 #elif defined(_SUNOS_VTOC_16)
5436 		struct dkl_partition *vp = &cl->cl_vtoc.v_part[count];
5437 		cl->cl_offset[count] = vp->p_start + cl->cl_solaris_offset;
5438 #else
5439 #error "No VTOC format defined."
5440 #endif
5441 	}
5442 
5443 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
5444 	return (label_rc);
5445 }
5446 #endif
5447 
5448 #if defined(__i386) || defined(__amd64)
5449 static int
5450 cmlb_dkio_get_virtgeom(struct cmlb_lun *cl, caddr_t arg, int flag)
5451 {
5452 	int err = 0;
5453 
5454 	/* Return the driver's notion of the media's logical geometry */
5455 	struct dk_geom	disk_geom;
5456 	struct dk_geom	*dkgp = &disk_geom;
5457 
5458 	mutex_enter(CMLB_MUTEX(cl));
5459 	/*
5460 	 * If there is no HBA geometry available, or
5461 	 * if the HBA returned us something that doesn't
5462 	 * really fit into an Int 13/function 8 geometry
5463 	 * result, just fail the ioctl.  See PSARC 1998/313.
5464 	 */
5465 	if (cl->cl_lgeom.g_nhead == 0 ||
5466 	    cl->cl_lgeom.g_nsect == 0 ||
5467 	    cl->cl_lgeom.g_ncyl > 1024) {
5468 		mutex_exit(CMLB_MUTEX(cl));
5469 		err = EINVAL;
5470 	} else {
5471 		dkgp->dkg_ncyl	= cl->cl_lgeom.g_ncyl;
5472 		dkgp->dkg_acyl	= cl->cl_lgeom.g_acyl;
5473 		dkgp->dkg_pcyl	= dkgp->dkg_ncyl + dkgp->dkg_acyl;
5474 		dkgp->dkg_nhead	= cl->cl_lgeom.g_nhead;
5475 		dkgp->dkg_nsect	= cl->cl_lgeom.g_nsect;
5476 
5477 		mutex_exit(CMLB_MUTEX(cl));
5478 		if (ddi_copyout(dkgp, (void *)arg,
5479 		    sizeof (struct dk_geom), flag)) {
5480 			err = EFAULT;
5481 		} else {
5482 			err = 0;
5483 		}
5484 	}
5485 	return (err);
5486 }
5487 #endif
5488 
5489 #if defined(__i386) || defined(__amd64)
5490 static int
5491 cmlb_dkio_get_phygeom(struct cmlb_lun *cl, caddr_t  arg, int flag,
5492     void *tg_cookie)
5493 {
5494 	int err = 0;
5495 	diskaddr_t capacity;
5496 
5497 
5498 	/* Return the driver's notion of the media physical geometry */
5499 	struct dk_geom	disk_geom;
5500 	struct dk_geom	*dkgp = &disk_geom;
5501 
5502 	mutex_enter(CMLB_MUTEX(cl));
5503 
5504 	if (cl->cl_g.dkg_nhead != 0 &&
5505 	    cl->cl_g.dkg_nsect != 0) {
5506 		/*
5507 		 * We succeeded in getting a geometry, but
5508 		 * right now it is being reported as just the
5509 		 * Solaris fdisk partition, just like for
5510 		 * DKIOCGGEOM. We need to change that to be
5511 		 * correct for the entire disk now.
5512 		 */
5513 		bcopy(&cl->cl_g, dkgp, sizeof (*dkgp));
5514 		dkgp->dkg_acyl = 0;
5515 		dkgp->dkg_ncyl = cl->cl_blockcount /
5516 		    (dkgp->dkg_nhead * dkgp->dkg_nsect);
5517 	} else {
5518 		bzero(dkgp, sizeof (struct dk_geom));
5519 		/*
5520 		 * This disk does not have a Solaris VTOC
5521 		 * so we must present a physical geometry
5522 		 * that will remain consistent regardless
5523 		 * of how the disk is used. This will ensure
5524 		 * that the geometry does not change regardless
5525 		 * of the fdisk partition type (ie. EFI, FAT32,
5526 		 * Solaris, etc).
5527 		 */
5528 		if (ISCD(cl)) {
5529 			dkgp->dkg_nhead = cl->cl_pgeom.g_nhead;
5530 			dkgp->dkg_nsect = cl->cl_pgeom.g_nsect;
5531 			dkgp->dkg_ncyl = cl->cl_pgeom.g_ncyl;
5532 			dkgp->dkg_acyl = cl->cl_pgeom.g_acyl;
5533 		} else {
5534 			/*
5535 			 * Invalid cl_blockcount can generate invalid
5536 			 * dk_geom and may result in division by zero
5537 			 * system failure. Should make sure blockcount
5538 			 * is valid before using it here.
5539 			 */
5540 			if (cl->cl_blockcount == 0) {
5541 				mutex_exit(CMLB_MUTEX(cl));
5542 				err = EIO;
5543 				return (err);
5544 			}
5545 			/*
5546 			 * Refer to comments related to off-by-1 at the
5547 			 * header of this file
5548 			 */
5549 			if (cl->cl_alter_behavior & CMLB_OFF_BY_ONE)
5550 				capacity = cl->cl_blockcount - 1;
5551 			else
5552 				capacity = cl->cl_blockcount;
5553 
5554 			cmlb_convert_geometry(cl, capacity, dkgp, tg_cookie);
5555 			dkgp->dkg_acyl = 0;
5556 			dkgp->dkg_ncyl = capacity /
5557 			    (dkgp->dkg_nhead * dkgp->dkg_nsect);
5558 		}
5559 	}
5560 	dkgp->dkg_pcyl = dkgp->dkg_ncyl + dkgp->dkg_acyl;
5561 
5562 	mutex_exit(CMLB_MUTEX(cl));
5563 	if (ddi_copyout(dkgp, (void *)arg, sizeof (struct dk_geom), flag))
5564 		err = EFAULT;
5565 
5566 	return (err);
5567 }
5568 #endif
5569 
5570 #if defined(__i386) || defined(__amd64)
5571 static int
5572 cmlb_dkio_partinfo(struct cmlb_lun *cl, dev_t dev, caddr_t  arg, int flag)
5573 {
5574 	int err = 0;
5575 
5576 	/*
5577 	 * Return parameters describing the selected disk slice.
5578 	 * Note: this ioctl is for the intel platform only
5579 	 */
5580 	int part;
5581 
5582 	part = CMLBPART(dev);
5583 
5584 	mutex_enter(CMLB_MUTEX(cl));
5585 	/* don't check cl_solaris_size for pN */
5586 	if (part < P0_RAW_DISK && cl->cl_solaris_size == 0) {
5587 		err = EIO;
5588 		mutex_exit(CMLB_MUTEX(cl));
5589 	} else {
5590 		struct part_info p;
5591 
5592 		p.p_start = (daddr_t)cl->cl_offset[part];
5593 		p.p_length = (int)cl->cl_map[part].dkl_nblk;
5594 		mutex_exit(CMLB_MUTEX(cl));
5595 #ifdef _MULTI_DATAMODEL
5596 		switch (ddi_model_convert_from(flag & FMODELS)) {
5597 		case DDI_MODEL_ILP32:
5598 		{
5599 			struct part_info32 p32;
5600 
5601 			p32.p_start = (daddr32_t)p.p_start;
5602 			p32.p_length = p.p_length;
5603 			if (ddi_copyout(&p32, (void *)arg,
5604 			    sizeof (p32), flag))
5605 				err = EFAULT;
5606 			break;
5607 		}
5608 
5609 		case DDI_MODEL_NONE:
5610 		{
5611 			if (ddi_copyout(&p, (void *)arg, sizeof (p),
5612 			    flag))
5613 				err = EFAULT;
5614 			break;
5615 		}
5616 		}
5617 #else /* ! _MULTI_DATAMODEL */
5618 		if (ddi_copyout(&p, (void *)arg, sizeof (p), flag))
5619 			err = EFAULT;
5620 #endif /* _MULTI_DATAMODEL */
5621 	}
5622 	return (err);
5623 }
5624 static int
5625 cmlb_dkio_extpartinfo(struct cmlb_lun *cl, dev_t dev, caddr_t  arg, int flag)
5626 {
5627 	int err = 0;
5628 
5629 	/*
5630 	 * Return parameters describing the selected disk slice.
5631 	 * Note: this ioctl is for the intel platform only
5632 	 */
5633 	int part;
5634 
5635 	part = CMLBPART(dev);
5636 
5637 	mutex_enter(CMLB_MUTEX(cl));
5638 	/* don't check cl_solaris_size for pN */
5639 	if (part < P0_RAW_DISK && cl->cl_solaris_size == 0) {
5640 		err = EIO;
5641 		mutex_exit(CMLB_MUTEX(cl));
5642 	} else {
5643 		struct extpart_info p;
5644 
5645 		p.p_start = (diskaddr_t)cl->cl_offset[part];
5646 		p.p_length = (diskaddr_t)cl->cl_map[part].dkl_nblk;
5647 		mutex_exit(CMLB_MUTEX(cl));
5648 		if (ddi_copyout(&p, (void *)arg, sizeof (p), flag))
5649 			err = EFAULT;
5650 	}
5651 	return (err);
5652 }
5653 #endif
5654 
5655 int
5656 cmlb_prop_op(cmlb_handle_t cmlbhandle,
5657     dev_t dev, dev_info_t *dip, ddi_prop_op_t prop_op, int mod_flags,
5658     char *name, caddr_t valuep, int *lengthp, int part, void *tg_cookie)
5659 {
5660 	struct cmlb_lun	*cl;
5661 	diskaddr_t	capacity;
5662 	uint32_t	lbasize;
5663 	enum		dp { DP_NBLOCKS, DP_BLKSIZE, DP_SSD, DP_ROT } dp;
5664 	int		callers_length;
5665 	caddr_t		buffer;
5666 	uint64_t	nblocks64;
5667 	uint_t		dblk;
5668 	tg_attribute_t	tgattr;
5669 
5670 	/* Always fallback to ddi_prop_op... */
5671 	cl = (struct cmlb_lun *)cmlbhandle;
5672 	if (cl == NULL) {
5673 fallback:	return (ddi_prop_op(dev, dip, prop_op, mod_flags,
5674 		    name, valuep, lengthp));
5675 	}
5676 
5677 	/* Pick up capacity and blocksize information. */
5678 	capacity = cl->cl_blockcount;
5679 	if (capacity == 0)
5680 		goto fallback;
5681 	lbasize = cl->cl_tgt_blocksize;
5682 	if (lbasize == 0)
5683 		lbasize = DEV_BSIZE;	/* 0 -> DEV_BSIZE units */
5684 
5685 	/* Check for dynamic property of whole device. */
5686 	if (dev == DDI_DEV_T_ANY) {
5687 		/* Fallback to ddi_prop_op if we don't understand.  */
5688 		if (strcmp(name, "device-nblocks") == 0)
5689 			dp = DP_NBLOCKS;
5690 		else if (strcmp(name, "device-blksize") == 0)
5691 			dp = DP_BLKSIZE;
5692 		else if (strcmp(name, "device-solid-state") == 0)
5693 			dp = DP_SSD;
5694 		else if (strcmp(name, "device-rotational") == 0)
5695 			dp = DP_ROT;
5696 		else
5697 			goto fallback;
5698 
5699 		/* get callers length, establish length of our dynamic prop */
5700 		callers_length = *lengthp;
5701 		if (dp == DP_NBLOCKS)
5702 			*lengthp = sizeof (uint64_t);
5703 		else if ((dp == DP_BLKSIZE) || (dp == DP_SSD))
5704 			*lengthp = sizeof (uint32_t);
5705 
5706 		/* service request for the length of the property */
5707 		if (prop_op == PROP_LEN)
5708 			return (DDI_PROP_SUCCESS);
5709 
5710 		switch (prop_op) {
5711 		case PROP_LEN_AND_VAL_ALLOC:
5712 			if ((buffer = kmem_alloc(*lengthp,
5713 			    (mod_flags & DDI_PROP_CANSLEEP) ?
5714 			    KM_SLEEP : KM_NOSLEEP)) == NULL)
5715 				return (DDI_PROP_NO_MEMORY);
5716 			*(caddr_t *)valuep = buffer;	/* set callers buf */
5717 			break;
5718 
5719 		case PROP_LEN_AND_VAL_BUF:
5720 			/* the length of the prop and the request must match */
5721 			if (callers_length != *lengthp)
5722 				return (DDI_PROP_INVAL_ARG);
5723 			buffer = valuep;		/* get callers buf */
5724 			break;
5725 
5726 		default:
5727 			return (DDI_PROP_INVAL_ARG);
5728 		}
5729 
5730 		/* transfer the value into the buffer */
5731 		switch (dp) {
5732 		case DP_NBLOCKS:
5733 			*((uint64_t *)buffer) = capacity;
5734 			break;
5735 		case DP_BLKSIZE:
5736 			*((uint32_t *)buffer) = lbasize;
5737 			break;
5738 		case DP_SSD:
5739 			if (DK_TG_GETATTRIBUTE(cl, &tgattr, tg_cookie) != 0)
5740 				tgattr.media_is_solid_state = B_FALSE;
5741 			*((uint32_t *)buffer) =
5742 			    tgattr.media_is_solid_state ? 1 : 0;
5743 			break;
5744 		case DP_ROT:
5745 			if (DK_TG_GETATTRIBUTE(cl, &tgattr, tg_cookie) != 0)
5746 				tgattr.media_is_rotational = B_TRUE;
5747 			*((uint32_t *)buffer) =
5748 			    tgattr.media_is_rotational ? 1 : 0;
5749 			break;
5750 		}
5751 		return (DDI_PROP_SUCCESS);
5752 	}
5753 
5754 	/*
5755 	 * Support dynamic size oriented properties of partition. Requests
5756 	 * issued under conditions where size is valid are passed to
5757 	 * ddi_prop_op_nblocks with the size information, otherwise the
5758 	 * request is passed to ddi_prop_op. Size depends on valid geometry.
5759 	 */
5760 	if (!cmlb_is_valid(cmlbhandle))
5761 		goto fallback;
5762 
5763 	/* Get partition nblocks value. */
5764 	(void) cmlb_partinfo(cmlbhandle, part,
5765 	    (diskaddr_t *)&nblocks64, NULL, NULL, NULL, tg_cookie);
5766 
5767 	/*
5768 	 * Assume partition information is in sys_blocksize units, compute
5769 	 * divisor for size(9P) property representation.
5770 	 */
5771 	dblk = lbasize / cl->cl_sys_blocksize;
5772 
5773 	/* Now let ddi_prop_op_nblocks_blksize() handle the request. */
5774 	return (ddi_prop_op_nblocks_blksize(dev, dip, prop_op, mod_flags,
5775 	    name, valuep, lengthp, nblocks64 / dblk, lbasize));
5776 }
5777