xref: /illumos-gate/usr/src/uts/common/io/cmlb.c (revision 0803869a9bc0aff1c81287ac6eae15943cbba545)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright 2012 DEY Storage Systems, Inc.  All rights reserved.
24  * Copyright 2010 Sun Microsystems, Inc.  All rights reserved.
25  * Use is subject to license terms.
26  * Copyright 2016 Toomas Soome <tsoome@me.com>
27  */
28 
29 /*
30  * This module provides support for labeling operations for target
31  * drivers.
32  */
33 
34 #include <sys/scsi/scsi.h>
35 #include <sys/sunddi.h>
36 #include <sys/dklabel.h>
37 #include <sys/dkio.h>
38 #include <sys/vtoc.h>
39 #include <sys/dktp/fdisk.h>
40 #include <sys/vtrace.h>
41 #include <sys/efi_partition.h>
42 #include <sys/cmlb.h>
43 #include <sys/cmlb_impl.h>
44 #if defined(__i386) || defined(__amd64)
45 #include <sys/fs/dv_node.h>
46 #endif
47 #include <sys/ddi_impldefs.h>
48 
49 /*
50  * Driver minor node structure and data table
51  */
52 struct driver_minor_data {
53 	char	*name;
54 	minor_t	minor;
55 	int	type;
56 };
57 
58 static struct driver_minor_data dk_minor_data[] = {
59 	{"a", 0, S_IFBLK},
60 	{"b", 1, S_IFBLK},
61 	{"c", 2, S_IFBLK},
62 	{"d", 3, S_IFBLK},
63 	{"e", 4, S_IFBLK},
64 	{"f", 5, S_IFBLK},
65 	{"g", 6, S_IFBLK},
66 	{"h", 7, S_IFBLK},
67 #if defined(_SUNOS_VTOC_16)
68 	{"i", 8, S_IFBLK},
69 	{"j", 9, S_IFBLK},
70 	{"k", 10, S_IFBLK},
71 	{"l", 11, S_IFBLK},
72 	{"m", 12, S_IFBLK},
73 	{"n", 13, S_IFBLK},
74 	{"o", 14, S_IFBLK},
75 	{"p", 15, S_IFBLK},
76 #endif			/* defined(_SUNOS_VTOC_16) */
77 #if defined(_FIRMWARE_NEEDS_FDISK)
78 	{"q", 16, S_IFBLK},
79 	{"r", 17, S_IFBLK},
80 	{"s", 18, S_IFBLK},
81 	{"t", 19, S_IFBLK},
82 	{"u", 20, S_IFBLK},
83 #endif			/* defined(_FIRMWARE_NEEDS_FDISK) */
84 	{"a,raw", 0, S_IFCHR},
85 	{"b,raw", 1, S_IFCHR},
86 	{"c,raw", 2, S_IFCHR},
87 	{"d,raw", 3, S_IFCHR},
88 	{"e,raw", 4, S_IFCHR},
89 	{"f,raw", 5, S_IFCHR},
90 	{"g,raw", 6, S_IFCHR},
91 	{"h,raw", 7, S_IFCHR},
92 #if defined(_SUNOS_VTOC_16)
93 	{"i,raw", 8, S_IFCHR},
94 	{"j,raw", 9, S_IFCHR},
95 	{"k,raw", 10, S_IFCHR},
96 	{"l,raw", 11, S_IFCHR},
97 	{"m,raw", 12, S_IFCHR},
98 	{"n,raw", 13, S_IFCHR},
99 	{"o,raw", 14, S_IFCHR},
100 	{"p,raw", 15, S_IFCHR},
101 #endif			/* defined(_SUNOS_VTOC_16) */
102 #if defined(_FIRMWARE_NEEDS_FDISK)
103 	{"q,raw", 16, S_IFCHR},
104 	{"r,raw", 17, S_IFCHR},
105 	{"s,raw", 18, S_IFCHR},
106 	{"t,raw", 19, S_IFCHR},
107 	{"u,raw", 20, S_IFCHR},
108 #endif			/* defined(_FIRMWARE_NEEDS_FDISK) */
109 	{0}
110 };
111 
112 #if defined(__i386) || defined(__amd64)
113 #if defined(_FIRMWARE_NEEDS_FDISK)
114 static struct driver_minor_data dk_ext_minor_data[] = {
115 	{"p5", 21, S_IFBLK},
116 	{"p6", 22, S_IFBLK},
117 	{"p7", 23, S_IFBLK},
118 	{"p8", 24, S_IFBLK},
119 	{"p9", 25, S_IFBLK},
120 	{"p10", 26, S_IFBLK},
121 	{"p11", 27, S_IFBLK},
122 	{"p12", 28, S_IFBLK},
123 	{"p13", 29, S_IFBLK},
124 	{"p14", 30, S_IFBLK},
125 	{"p15", 31, S_IFBLK},
126 	{"p16", 32, S_IFBLK},
127 	{"p17", 33, S_IFBLK},
128 	{"p18", 34, S_IFBLK},
129 	{"p19", 35, S_IFBLK},
130 	{"p20", 36, S_IFBLK},
131 	{"p21", 37, S_IFBLK},
132 	{"p22", 38, S_IFBLK},
133 	{"p23", 39, S_IFBLK},
134 	{"p24", 40, S_IFBLK},
135 	{"p25", 41, S_IFBLK},
136 	{"p26", 42, S_IFBLK},
137 	{"p27", 43, S_IFBLK},
138 	{"p28", 44, S_IFBLK},
139 	{"p29", 45, S_IFBLK},
140 	{"p30", 46, S_IFBLK},
141 	{"p31", 47, S_IFBLK},
142 	{"p32", 48, S_IFBLK},
143 	{"p33", 49, S_IFBLK},
144 	{"p34", 50, S_IFBLK},
145 	{"p35", 51, S_IFBLK},
146 	{"p36", 52, S_IFBLK},
147 	{"p5,raw", 21, S_IFCHR},
148 	{"p6,raw", 22, S_IFCHR},
149 	{"p7,raw", 23, S_IFCHR},
150 	{"p8,raw", 24, S_IFCHR},
151 	{"p9,raw", 25, S_IFCHR},
152 	{"p10,raw", 26, S_IFCHR},
153 	{"p11,raw", 27, S_IFCHR},
154 	{"p12,raw", 28, S_IFCHR},
155 	{"p13,raw", 29, S_IFCHR},
156 	{"p14,raw", 30, S_IFCHR},
157 	{"p15,raw", 31, S_IFCHR},
158 	{"p16,raw", 32, S_IFCHR},
159 	{"p17,raw", 33, S_IFCHR},
160 	{"p18,raw", 34, S_IFCHR},
161 	{"p19,raw", 35, S_IFCHR},
162 	{"p20,raw", 36, S_IFCHR},
163 	{"p21,raw", 37, S_IFCHR},
164 	{"p22,raw", 38, S_IFCHR},
165 	{"p23,raw", 39, S_IFCHR},
166 	{"p24,raw", 40, S_IFCHR},
167 	{"p25,raw", 41, S_IFCHR},
168 	{"p26,raw", 42, S_IFCHR},
169 	{"p27,raw", 43, S_IFCHR},
170 	{"p28,raw", 44, S_IFCHR},
171 	{"p29,raw", 45, S_IFCHR},
172 	{"p30,raw", 46, S_IFCHR},
173 	{"p31,raw", 47, S_IFCHR},
174 	{"p32,raw", 48, S_IFCHR},
175 	{"p33,raw", 49, S_IFCHR},
176 	{"p34,raw", 50, S_IFCHR},
177 	{"p35,raw", 51, S_IFCHR},
178 	{"p36,raw", 52, S_IFCHR},
179 	{0}
180 };
181 #endif			/* defined(_FIRMWARE_NEEDS_FDISK) */
182 #endif			/* if defined(__i386) || defined(__amd64) */
183 
184 static struct driver_minor_data dk_minor_data_efi[] = {
185 	{"a", 0, S_IFBLK},
186 	{"b", 1, S_IFBLK},
187 	{"c", 2, S_IFBLK},
188 	{"d", 3, S_IFBLK},
189 	{"e", 4, S_IFBLK},
190 	{"f", 5, S_IFBLK},
191 	{"g", 6, S_IFBLK},
192 	{"wd", 7, S_IFBLK},
193 #if defined(_SUNOS_VTOC_16)
194 	{"i", 8, S_IFBLK},
195 	{"j", 9, S_IFBLK},
196 	{"k", 10, S_IFBLK},
197 	{"l", 11, S_IFBLK},
198 	{"m", 12, S_IFBLK},
199 	{"n", 13, S_IFBLK},
200 	{"o", 14, S_IFBLK},
201 	{"p", 15, S_IFBLK},
202 #endif			/* defined(_SUNOS_VTOC_16) */
203 #if defined(_FIRMWARE_NEEDS_FDISK)
204 	{"q", 16, S_IFBLK},
205 	{"r", 17, S_IFBLK},
206 	{"s", 18, S_IFBLK},
207 	{"t", 19, S_IFBLK},
208 	{"u", 20, S_IFBLK},
209 #endif			/* defined(_FIRMWARE_NEEDS_FDISK) */
210 	{"a,raw", 0, S_IFCHR},
211 	{"b,raw", 1, S_IFCHR},
212 	{"c,raw", 2, S_IFCHR},
213 	{"d,raw", 3, S_IFCHR},
214 	{"e,raw", 4, S_IFCHR},
215 	{"f,raw", 5, S_IFCHR},
216 	{"g,raw", 6, S_IFCHR},
217 	{"wd,raw", 7, S_IFCHR},
218 #if defined(_SUNOS_VTOC_16)
219 	{"i,raw", 8, S_IFCHR},
220 	{"j,raw", 9, S_IFCHR},
221 	{"k,raw", 10, S_IFCHR},
222 	{"l,raw", 11, S_IFCHR},
223 	{"m,raw", 12, S_IFCHR},
224 	{"n,raw", 13, S_IFCHR},
225 	{"o,raw", 14, S_IFCHR},
226 	{"p,raw", 15, S_IFCHR},
227 #endif			/* defined(_SUNOS_VTOC_16) */
228 #if defined(_FIRMWARE_NEEDS_FDISK)
229 	{"q,raw", 16, S_IFCHR},
230 	{"r,raw", 17, S_IFCHR},
231 	{"s,raw", 18, S_IFCHR},
232 	{"t,raw", 19, S_IFCHR},
233 	{"u,raw", 20, S_IFCHR},
234 #endif			/* defined(_FIRMWARE_NEEDS_FDISK) */
235 	{0}
236 };
237 
238 /*
239  * Declare the dynamic properties implemented in prop_op(9E) implementation
240  * that we want to have show up in a di_init(3DEVINFO) device tree snapshot
241  * of drivers that call cmlb_attach().
242  */
243 static i_ddi_prop_dyn_t cmlb_prop_dyn[] = {
244 	{"Nblocks",		DDI_PROP_TYPE_INT64,	S_IFBLK},
245 	{"Size",		DDI_PROP_TYPE_INT64,	S_IFCHR},
246 	{"device-nblocks",	DDI_PROP_TYPE_INT64},
247 	{"device-blksize",	DDI_PROP_TYPE_INT},
248 	{"device-solid-state",	DDI_PROP_TYPE_INT},
249 	{"device-rotational",	DDI_PROP_TYPE_INT},
250 	{NULL}
251 };
252 
253 /*
254  * This implies an upper limit of 8192 GPT partitions
255  * in one transfer for GUID Partition Entry Array.
256  */
257 len_t cmlb_tg_max_efi_xfer = 1024 * 1024;
258 
259 /*
260  * External kernel interfaces
261  */
262 extern struct mod_ops mod_miscops;
263 
264 extern int ddi_create_internal_pathname(dev_info_t *dip, char *name,
265     int spec_type, minor_t minor_num);
266 
267 /*
268  * Global buffer and mutex for debug logging
269  */
270 static char	cmlb_log_buffer[1024];
271 static kmutex_t	cmlb_log_mutex;
272 
273 
274 struct cmlb_lun *cmlb_debug_cl = NULL;
275 uint_t cmlb_level_mask = 0x0;
276 
277 int cmlb_rot_delay = 4;	/* default rotational delay */
278 
279 static struct modlmisc modlmisc = {
280 	&mod_miscops,   /* Type of module */
281 	"Common Labeling module"
282 };
283 
284 static struct modlinkage modlinkage = {
285 	MODREV_1, (void *)&modlmisc, NULL
286 };
287 
288 /* Local function prototypes */
289 static dev_t cmlb_make_device(struct cmlb_lun *cl);
290 static int cmlb_validate_geometry(struct cmlb_lun *cl, boolean_t forcerevalid,
291     int flags, void *tg_cookie);
292 static void cmlb_resync_geom_caches(struct cmlb_lun *cl, diskaddr_t capacity,
293     void *tg_cookie);
294 static int cmlb_read_fdisk(struct cmlb_lun *cl, diskaddr_t capacity,
295     void *tg_cookie);
296 static void cmlb_swap_efi_gpt(efi_gpt_t *e);
297 static void cmlb_swap_efi_gpe(int nparts, efi_gpe_t *p);
298 static int cmlb_validate_efi(efi_gpt_t *labp);
299 static int cmlb_use_efi(struct cmlb_lun *cl, diskaddr_t capacity, int flags,
300     void *tg_cookie);
301 static void cmlb_build_default_label(struct cmlb_lun *cl, void *tg_cookie);
302 static int  cmlb_uselabel(struct cmlb_lun *cl,  struct dk_label *l, int flags);
303 #if defined(_SUNOS_VTOC_8)
304 static void cmlb_build_user_vtoc(struct cmlb_lun *cl, struct vtoc *user_vtoc);
305 #endif
306 static int cmlb_build_label_vtoc(struct cmlb_lun *cl, struct vtoc *user_vtoc);
307 static int cmlb_write_label(struct cmlb_lun *cl, void *tg_cookie);
308 static int cmlb_set_vtoc(struct cmlb_lun *cl, struct dk_label *dkl,
309     void *tg_cookie);
310 static void cmlb_clear_efi(struct cmlb_lun *cl, void *tg_cookie);
311 static void cmlb_clear_vtoc(struct cmlb_lun *cl, void *tg_cookie);
312 static void cmlb_setup_default_geometry(struct cmlb_lun *cl, void *tg_cookie);
313 static int cmlb_create_minor_nodes(struct cmlb_lun *cl);
314 static int cmlb_check_update_blockcount(struct cmlb_lun *cl, void *tg_cookie);
315 static boolean_t cmlb_check_efi_mbr(uchar_t *buf, boolean_t *is_mbr);
316 
317 #if defined(__i386) || defined(__amd64)
318 static int cmlb_update_fdisk_and_vtoc(struct cmlb_lun *cl, void *tg_cookie);
319 #endif
320 
321 #if defined(_FIRMWARE_NEEDS_FDISK)
322 static boolean_t  cmlb_has_max_chs_vals(struct ipart *fdp);
323 #endif
324 
325 #if defined(_SUNOS_VTOC_16)
326 static void cmlb_convert_geometry(struct cmlb_lun *cl, diskaddr_t capacity,
327     struct dk_geom *cl_g, void *tg_cookie);
328 #endif
329 
330 static int cmlb_dkio_get_geometry(struct cmlb_lun *cl, caddr_t arg, int flag,
331     void *tg_cookie);
332 static int cmlb_dkio_set_geometry(struct cmlb_lun *cl, caddr_t arg, int flag);
333 static int cmlb_dkio_get_partition(struct cmlb_lun *cl, caddr_t arg, int flag,
334     void *tg_cookie);
335 static int cmlb_dkio_set_partition(struct cmlb_lun *cl, caddr_t arg, int flag);
336 static int cmlb_dkio_get_efi(struct cmlb_lun *cl, caddr_t arg, int flag,
337     void *tg_cookie);
338 static int cmlb_dkio_set_efi(struct cmlb_lun *cl, dev_t dev, caddr_t arg,
339     int flag, void *tg_cookie);
340 static int cmlb_dkio_get_vtoc(struct cmlb_lun *cl, caddr_t arg, int flag,
341     void *tg_cookie);
342 static int cmlb_dkio_get_extvtoc(struct cmlb_lun *cl, caddr_t arg, int flag,
343     void *tg_cookie);
344 static int cmlb_dkio_set_vtoc(struct cmlb_lun *cl, dev_t dev, caddr_t arg,
345     int flag, void *tg_cookie);
346 static int cmlb_dkio_set_extvtoc(struct cmlb_lun *cl, dev_t dev, caddr_t arg,
347     int flag, void *tg_cookie);
348 static int cmlb_dkio_get_mboot(struct cmlb_lun *cl, caddr_t arg, int flag,
349     void *tg_cookie);
350 static int cmlb_dkio_set_mboot(struct cmlb_lun *cl, caddr_t arg, int flag,
351     void *tg_cookie);
352 static int cmlb_dkio_partition(struct cmlb_lun *cl, caddr_t arg, int flag,
353     void *tg_cookie);
354 
355 #if defined(__i386) || defined(__amd64)
356 static int cmlb_dkio_set_ext_part(struct cmlb_lun *cl, caddr_t arg, int flag,
357     void *tg_cookie);
358 static int cmlb_validate_ext_part(struct cmlb_lun *cl, int part, int epart,
359     uint32_t start, uint32_t size);
360 static int cmlb_is_linux_swap(struct cmlb_lun *cl, uint32_t part_start,
361     void *tg_cookie);
362 static int cmlb_dkio_get_virtgeom(struct cmlb_lun *cl, caddr_t arg, int flag);
363 static int cmlb_dkio_get_phygeom(struct cmlb_lun *cl, caddr_t  arg, int flag,
364     void *tg_cookie);
365 static int cmlb_dkio_partinfo(struct cmlb_lun *cl, dev_t dev, caddr_t arg,
366     int flag);
367 static int cmlb_dkio_extpartinfo(struct cmlb_lun *cl, dev_t dev, caddr_t arg,
368     int flag);
369 #endif
370 
371 static void cmlb_dbg(uint_t comp, struct cmlb_lun *cl, const char *fmt, ...);
372 static void cmlb_v_log(dev_info_t *dev, const char *label, uint_t level,
373     const char *fmt, va_list ap);
374 static void cmlb_log(dev_info_t *dev, const char *label, uint_t level,
375     const char *fmt, ...);
376 
377 int
378 _init(void)
379 {
380 	mutex_init(&cmlb_log_mutex, NULL, MUTEX_DRIVER, NULL);
381 	return (mod_install(&modlinkage));
382 }
383 
384 int
385 _info(struct modinfo *modinfop)
386 {
387 	return (mod_info(&modlinkage, modinfop));
388 }
389 
390 int
391 _fini(void)
392 {
393 	int err;
394 
395 	if ((err = mod_remove(&modlinkage)) != 0) {
396 		return (err);
397 	}
398 
399 	mutex_destroy(&cmlb_log_mutex);
400 	return (err);
401 }
402 
403 /*
404  * cmlb_dbg is used for debugging to log additional info
405  * Level of output is controlled via cmlb_level_mask setting.
406  */
407 static void
408 cmlb_dbg(uint_t comp, struct cmlb_lun *cl, const char *fmt, ...)
409 {
410 	va_list		ap;
411 	dev_info_t	*dev;
412 	uint_t		level_mask = 0;
413 
414 	ASSERT(cl != NULL);
415 	dev = CMLB_DEVINFO(cl);
416 	ASSERT(dev != NULL);
417 	/*
418 	 * Filter messages based on the global component and level masks,
419 	 * also print if cl matches the value of cmlb_debug_cl, or if
420 	 * cmlb_debug_cl is set to NULL.
421 	 */
422 	if (comp & CMLB_TRACE)
423 		level_mask |= CMLB_LOGMASK_TRACE;
424 
425 	if (comp & CMLB_INFO)
426 		level_mask |= CMLB_LOGMASK_INFO;
427 
428 	if (comp & CMLB_ERROR)
429 		level_mask |= CMLB_LOGMASK_ERROR;
430 
431 	if ((cmlb_level_mask & level_mask) &&
432 	    ((cmlb_debug_cl == NULL) || (cmlb_debug_cl == cl))) {
433 		va_start(ap, fmt);
434 		cmlb_v_log(dev, CMLB_LABEL(cl), CE_CONT, fmt, ap);
435 		va_end(ap);
436 	}
437 }
438 
439 /*
440  * cmlb_log is basically a duplicate of scsi_log. It is redefined here
441  * so that this module does not depend on scsi module.
442  */
443 static void
444 cmlb_log(dev_info_t *dev, const char *label, uint_t level, const char *fmt, ...)
445 {
446 	va_list		ap;
447 
448 	va_start(ap, fmt);
449 	cmlb_v_log(dev, label, level, fmt, ap);
450 	va_end(ap);
451 }
452 
453 static void
454 cmlb_v_log(dev_info_t *dev, const char *label, uint_t level, const char *fmt,
455     va_list ap)
456 {
457 	static char 	name[256];
458 	int 		log_only = 0;
459 	int 		boot_only = 0;
460 	int 		console_only = 0;
461 
462 	mutex_enter(&cmlb_log_mutex);
463 
464 	if (dev) {
465 		if (level == CE_PANIC || level == CE_WARN ||
466 		    level == CE_NOTE) {
467 			(void) sprintf(name, "%s (%s%d):\n",
468 			    ddi_pathname(dev, cmlb_log_buffer),
469 			    label, ddi_get_instance(dev));
470 		} else {
471 			name[0] = '\0';
472 		}
473 	} else {
474 		(void) sprintf(name, "%s:", label);
475 	}
476 
477 	(void) vsprintf(cmlb_log_buffer, fmt, ap);
478 
479 	switch (cmlb_log_buffer[0]) {
480 	case '!':
481 		log_only = 1;
482 		break;
483 	case '?':
484 		boot_only = 1;
485 		break;
486 	case '^':
487 		console_only = 1;
488 		break;
489 	}
490 
491 	switch (level) {
492 	case CE_NOTE:
493 		level = CE_CONT;
494 		/* FALLTHROUGH */
495 	case CE_CONT:
496 	case CE_WARN:
497 	case CE_PANIC:
498 		if (boot_only) {
499 			cmn_err(level, "?%s\t%s", name, &cmlb_log_buffer[1]);
500 		} else if (console_only) {
501 			cmn_err(level, "^%s\t%s", name, &cmlb_log_buffer[1]);
502 		} else if (log_only) {
503 			cmn_err(level, "!%s\t%s", name, &cmlb_log_buffer[1]);
504 		} else {
505 			cmn_err(level, "%s\t%s", name, cmlb_log_buffer);
506 		}
507 		break;
508 	case CE_IGNORE:
509 		break;
510 	default:
511 		cmn_err(CE_CONT, "^DEBUG: %s\t%s", name, cmlb_log_buffer);
512 		break;
513 	}
514 	mutex_exit(&cmlb_log_mutex);
515 }
516 
517 
518 /*
519  * cmlb_alloc_handle:
520  *
521  *	Allocates a handle.
522  *
523  * Arguments:
524  *	cmlbhandlep	pointer to handle
525  *
526  * Notes:
527  *	Allocates a handle and stores the allocated handle in the area
528  *	pointed to by cmlbhandlep
529  *
530  * Context:
531  *	Kernel thread only (can sleep).
532  */
533 void
534 cmlb_alloc_handle(cmlb_handle_t *cmlbhandlep)
535 {
536 	struct cmlb_lun 	*cl;
537 
538 	cl = kmem_zalloc(sizeof (struct cmlb_lun), KM_SLEEP);
539 	ASSERT(cmlbhandlep != NULL);
540 
541 	cl->cl_state = CMLB_INITED;
542 	cl->cl_def_labeltype = CMLB_LABEL_UNDEF;
543 	mutex_init(CMLB_MUTEX(cl), NULL, MUTEX_DRIVER, NULL);
544 
545 	*cmlbhandlep = (cmlb_handle_t)(cl);
546 }
547 
548 /*
549  * cmlb_free_handle
550  *
551  *	Frees handle.
552  *
553  * Arguments:
554  *	cmlbhandlep	pointer to handle
555  */
556 void
557 cmlb_free_handle(cmlb_handle_t *cmlbhandlep)
558 {
559 	struct cmlb_lun 	*cl;
560 
561 	cl = (struct cmlb_lun *)*cmlbhandlep;
562 	if (cl != NULL) {
563 		mutex_destroy(CMLB_MUTEX(cl));
564 		kmem_free(cl, sizeof (struct cmlb_lun));
565 	}
566 
567 }
568 
569 /*
570  * cmlb_attach:
571  *
572  *	Attach handle to device, create minor nodes for device.
573  *
574  * Arguments:
575  * 	devi		pointer to device's dev_info structure.
576  * 	tgopsp		pointer to array of functions cmlb can use to callback
577  *			to target driver.
578  *
579  *	device_type	Peripheral device type as defined in
580  *			scsi/generic/inquiry.h
581  *
582  *	is_removable	whether or not device is removable.
583  *
584  *	is_hotpluggable	whether or not device is hotpluggable.
585  *
586  *	node_type	minor node type (as used by ddi_create_minor_node)
587  *
588  *	alter_behavior
589  *			bit flags:
590  *
591  *			CMLB_CREATE_ALTSLICE_VTOC_16_DTYPE_DIRECT: create
592  *			an alternate slice for the default label, if
593  *			device type is DTYPE_DIRECT an architectures default
594  *			label type is VTOC16.
595  *			Otherwise alternate slice will no be created.
596  *
597  *
598  *			CMLB_FAKE_GEOM_LABEL_IOCTLS_VTOC8: report a default
599  *			geometry and label for DKIOCGGEOM and DKIOCGVTOC
600  *			on architecture with VTOC8 label types.
601  *
602  * 			CMLB_OFF_BY_ONE: do the workaround for legacy off-by-
603  *                      one bug in obtaining capacity (in sd):
604  *			SCSI READ_CAPACITY command returns the LBA number of the
605  *			last logical block, but sd once treated this number as
606  *			disks' capacity on x86 platform. And LBAs are addressed
607  *			based 0. So the last block was lost on x86 platform.
608  *
609  *			Now, we remove this workaround. In order for present sd
610  *			driver to work with disks which are labeled/partitioned
611  *			via previous sd, we add workaround as follows:
612  *
613  *			1) Locate backup EFI label: cmlb searches the next to
614  *			   last
615  *			   block for backup EFI label. If fails, it will
616  *			   turn to the last block for backup EFI label;
617  *
618  *			2) Clear backup EFI label: cmlb first search the last
619  *			   block for backup EFI label, and will search the
620  *			   next to last block only if failed for the last
621  *			   block.
622  *
623  *			3) Calculate geometry:refer to cmlb_convert_geometry()
624  *			   If capacity increasing by 1 causes disks' capacity
625  *			   to cross over the limits in geometry calculation,
626  *			   geometry info will change. This will raise an issue:
627  *			   In case that primary VTOC label is destroyed, format
628  *			   commandline can restore it via backup VTOC labels.
629  *			   And format locates backup VTOC labels by use of
630  *			   geometry. So changing geometry will
631  *			   prevent format from finding backup VTOC labels. To
632  *			   eliminate this side effect for compatibility,
633  *			   sd uses (capacity -1) to calculate geometry;
634  *
635  *			4) 1TB disks: some important data structures use
636  *			   32-bit signed long/int (for example, daddr_t),
637  *			   so that sd doesn't support a disk with capacity
638  *			   larger than 1TB on 32-bit platform. However,
639  *			   for exactly 1TB disk, it was treated as (1T - 512)B
640  *			   in the past, and could have valid Solaris
641  *			   partitions. To workaround this, if an exactly 1TB
642  *			   disk has Solaris fdisk partition, it will be allowed
643  *			   to work with sd.
644  *
645  *
646  *
647  *			CMLB_FAKE_LABEL_ONE_PARTITION: create s0 and s2 covering
648  *			the entire disk, if there is no valid partition info.
649  *			If there is a valid Solaris partition, s0 and s2 will
650  *			only cover the entire Solaris partition.
651  *
652  *			CMLB_CREATE_P0_MINOR_NODE: create p0 node covering
653  *			the entire disk. Used by lofi to ensure presence of
654  *			whole disk device node in case of LOFI_MAP_FILE ioctl.
655  *
656  *	cmlbhandle	cmlb handle associated with device
657  *
658  *	tg_cookie	cookie from target driver to be passed back to target
659  *			driver when we call back to it through tg_ops.
660  *
661  * Notes:
662  *	Assumes a default label based on capacity for non-removable devices.
663  *	If capacity > 1TB, EFI is assumed otherwise VTOC (default VTOC
664  *	for the architecture).
665  *
666  *	For removable devices, default label type is assumed to be VTOC
667  *	type. Create minor nodes based on a default label type.
668  *	Label on the media is not validated.
669  *	minor number consists of:
670  *		if _SUNOS_VTOC_8 is defined
671  *			lowest 3 bits is taken as partition number
672  *			the rest is instance number
673  *		if _SUNOS_VTOC_16 is defined
674  *			lowest 6 bits is taken as partition number
675  *			the rest is instance number
676  *
677  *
678  * Return values:
679  *	0 	Success
680  * 	ENXIO 	creating minor nodes failed.
681  *	EINVAL  invalid arg, unsupported tg_ops version
682  */
683 int
684 cmlb_attach(dev_info_t *devi, cmlb_tg_ops_t *tgopsp, int device_type,
685     boolean_t is_removable, boolean_t is_hotpluggable, char *node_type,
686     int alter_behavior, cmlb_handle_t cmlbhandle, void *tg_cookie)
687 {
688 
689 	struct cmlb_lun	*cl = (struct cmlb_lun *)cmlbhandle;
690 	diskaddr_t	cap;
691 	int		status;
692 
693 	ASSERT(VALID_BOOLEAN(is_removable));
694 	ASSERT(VALID_BOOLEAN(is_hotpluggable));
695 
696 	if (tgopsp->tg_version < TG_DK_OPS_VERSION_1)
697 		return (EINVAL);
698 
699 	mutex_enter(CMLB_MUTEX(cl));
700 
701 	CMLB_DEVINFO(cl) = devi;
702 	cl->cmlb_tg_ops = tgopsp;
703 	cl->cl_device_type = device_type;
704 	cl->cl_is_removable = is_removable;
705 	cl->cl_is_hotpluggable = is_hotpluggable;
706 	cl->cl_node_type = node_type;
707 	cl->cl_sys_blocksize = DEV_BSIZE;
708 	cl->cl_f_geometry_is_valid = B_FALSE;
709 	cl->cl_def_labeltype = CMLB_LABEL_VTOC;
710 	cl->cl_alter_behavior = alter_behavior;
711 	cl->cl_reserved = -1;
712 	cl->cl_msglog_flag |= CMLB_ALLOW_2TB_WARN;
713 #if defined(__i386) || defined(__amd64)
714 	cl->cl_logical_drive_count = 0;
715 #endif
716 
717 	if (!is_removable) {
718 		mutex_exit(CMLB_MUTEX(cl));
719 		status = DK_TG_GETCAP(cl, &cap, tg_cookie);
720 		mutex_enter(CMLB_MUTEX(cl));
721 		if (status == 0 && cap > CMLB_EXTVTOC_LIMIT) {
722 			/* set default EFI if > 2TB */
723 			cl->cl_def_labeltype = CMLB_LABEL_EFI;
724 		}
725 	}
726 
727 	/* create minor nodes based on default label type */
728 	cl->cl_last_labeltype = CMLB_LABEL_UNDEF;
729 	cl->cl_cur_labeltype = CMLB_LABEL_UNDEF;
730 
731 	if (cmlb_create_minor_nodes(cl) != 0) {
732 		mutex_exit(CMLB_MUTEX(cl));
733 		return (ENXIO);
734 	}
735 
736 	/* Define the dynamic properties for devinfo spapshots. */
737 	i_ddi_prop_dyn_driver_set(CMLB_DEVINFO(cl), cmlb_prop_dyn);
738 
739 	cl->cl_state = CMLB_ATTACHED;
740 
741 	mutex_exit(CMLB_MUTEX(cl));
742 	return (0);
743 }
744 
745 /*
746  * cmlb_detach:
747  *
748  * Invalidate in-core labeling data and remove all minor nodes for
749  * the device associate with handle.
750  *
751  * Arguments:
752  *	cmlbhandle	cmlb handle associated with device.
753  *
754  *	tg_cookie	cookie from target driver to be passed back to target
755  *			driver when we call back to it through tg_ops.
756  *
757  */
758 /*ARGSUSED1*/
759 void
760 cmlb_detach(cmlb_handle_t cmlbhandle, void *tg_cookie)
761 {
762 	struct cmlb_lun *cl = (struct cmlb_lun *)cmlbhandle;
763 
764 	mutex_enter(CMLB_MUTEX(cl));
765 	cl->cl_def_labeltype = CMLB_LABEL_UNDEF;
766 	cl->cl_f_geometry_is_valid = B_FALSE;
767 	ddi_remove_minor_node(CMLB_DEVINFO(cl), NULL);
768 	i_ddi_prop_dyn_driver_set(CMLB_DEVINFO(cl), NULL);
769 	cl->cl_state = CMLB_INITED;
770 	mutex_exit(CMLB_MUTEX(cl));
771 }
772 
773 /*
774  * cmlb_validate:
775  *
776  *	Validates label.
777  *
778  * Arguments
779  *	cmlbhandle	cmlb handle associated with device.
780  *
781  *	flags		operation flags. used for verbosity control
782  *
783  *	tg_cookie	cookie from target driver to be passed back to target
784  *			driver when we call back to it through tg_ops.
785  *
786  *
787  * Notes:
788  *	If new label type is different from the current, adjust minor nodes
789  *	accordingly.
790  *
791  * Return values:
792  *	0		success
793  *			Note: having fdisk but no solaris partition is assumed
794  *			success.
795  *
796  *	ENOMEM		memory allocation failed
797  *	EIO		i/o errors during read or get capacity
798  * 	EACCESS		reservation conflicts
799  * 	EINVAL		label was corrupt, or no default label was assumed
800  *	ENXIO		invalid handle
801  */
802 int
803 cmlb_validate(cmlb_handle_t cmlbhandle, int flags, void *tg_cookie)
804 {
805 	struct cmlb_lun *cl = (struct cmlb_lun *)cmlbhandle;
806 	int 		rval;
807 	int  		ret = 0;
808 
809 	/*
810 	 * Temp work-around checking cl for NULL since there is a bug
811 	 * in sd_detach calling this routine from taskq_dispatch
812 	 * inited function.
813 	 */
814 	if (cl == NULL)
815 		return (ENXIO);
816 
817 	mutex_enter(CMLB_MUTEX(cl));
818 	if (cl->cl_state < CMLB_ATTACHED) {
819 		mutex_exit(CMLB_MUTEX(cl));
820 		return (ENXIO);
821 	}
822 
823 	rval = cmlb_validate_geometry((struct cmlb_lun *)cmlbhandle, B_TRUE,
824 	    flags, tg_cookie);
825 
826 	if (rval == ENOTSUP) {
827 		if (cl->cl_f_geometry_is_valid) {
828 			cl->cl_cur_labeltype = CMLB_LABEL_EFI;
829 			ret = 0;
830 		} else {
831 			ret = EINVAL;
832 		}
833 	} else {
834 		ret = rval;
835 		if (ret == 0)
836 			cl->cl_cur_labeltype = CMLB_LABEL_VTOC;
837 	}
838 
839 	if (ret == 0)
840 		(void) cmlb_create_minor_nodes(cl);
841 
842 	mutex_exit(CMLB_MUTEX(cl));
843 	return (ret);
844 }
845 
846 /*
847  * cmlb_invalidate:
848  *	Invalidate in core label data
849  *
850  * Arguments:
851  *	cmlbhandle	cmlb handle associated with device.
852  *	tg_cookie	cookie from target driver to be passed back to target
853  *			driver when we call back to it through tg_ops.
854  */
855 /*ARGSUSED1*/
856 void
857 cmlb_invalidate(cmlb_handle_t cmlbhandle, void *tg_cookie)
858 {
859 	struct cmlb_lun *cl = (struct cmlb_lun *)cmlbhandle;
860 
861 	if (cl == NULL)
862 		return;
863 
864 	mutex_enter(CMLB_MUTEX(cl));
865 	cl->cl_f_geometry_is_valid = B_FALSE;
866 	mutex_exit(CMLB_MUTEX(cl));
867 }
868 
869 /*
870  * cmlb_is_valid
871  * 	Get status on whether the incore label/geom data is valid
872  *
873  * Arguments:
874  *	cmlbhandle      cmlb handle associated with device.
875  *
876  * Return values:
877  *	B_TRUE if incore label/geom data is valid.
878  *	B_FALSE otherwise.
879  *
880  */
881 
882 
883 boolean_t
884 cmlb_is_valid(cmlb_handle_t cmlbhandle)
885 {
886 	struct cmlb_lun *cl = (struct cmlb_lun *)cmlbhandle;
887 
888 	if (cmlbhandle == NULL)
889 		return (B_FALSE);
890 
891 	return (cl->cl_f_geometry_is_valid);
892 
893 }
894 
895 
896 
897 /*
898  * cmlb_close:
899  *
900  * Close the device, revert to a default label minor node for the device,
901  * if it is removable.
902  *
903  * Arguments:
904  *	cmlbhandle	cmlb handle associated with device.
905  *
906  *	tg_cookie	cookie from target driver to be passed back to target
907  *			driver when we call back to it through tg_ops.
908  * Return values:
909  *	0	Success
910  * 	ENXIO	Re-creating minor node failed.
911  */
912 /*ARGSUSED1*/
913 int
914 cmlb_close(cmlb_handle_t cmlbhandle, void *tg_cookie)
915 {
916 	struct cmlb_lun *cl = (struct cmlb_lun *)cmlbhandle;
917 
918 	mutex_enter(CMLB_MUTEX(cl));
919 	cl->cl_f_geometry_is_valid = B_FALSE;
920 
921 	/* revert to default minor node for this device */
922 	if (ISREMOVABLE(cl)) {
923 		cl->cl_cur_labeltype = CMLB_LABEL_UNDEF;
924 		(void) cmlb_create_minor_nodes(cl);
925 	}
926 
927 	mutex_exit(CMLB_MUTEX(cl));
928 	return (0);
929 }
930 
931 /*
932  * cmlb_get_devid_block:
933  *	 get the block number where device id is stored.
934  *
935  * Arguments:
936  *	cmlbhandle	cmlb handle associated with device.
937  *	devidblockp	pointer to block number.
938  *	tg_cookie	cookie from target driver to be passed back to target
939  *			driver when we call back to it through tg_ops.
940  *
941  * Notes:
942  *	It stores the block number of device id in the area pointed to
943  *	by devidblockp.
944  * 	with the block number of device id.
945  *
946  * Return values:
947  *	0	success
948  *	EINVAL 	device id does not apply to current label type.
949  */
950 /*ARGSUSED2*/
951 int
952 cmlb_get_devid_block(cmlb_handle_t cmlbhandle, diskaddr_t *devidblockp,
953     void *tg_cookie)
954 {
955 	daddr_t			spc, blk, head, cyl;
956 	struct cmlb_lun *cl = (struct cmlb_lun *)cmlbhandle;
957 
958 	mutex_enter(CMLB_MUTEX(cl));
959 	if (cl->cl_state < CMLB_ATTACHED) {
960 		mutex_exit(CMLB_MUTEX(cl));
961 		return (EINVAL);
962 	}
963 
964 	if ((!cl->cl_f_geometry_is_valid) ||
965 	    (cl->cl_solaris_size < DK_LABEL_LOC)) {
966 		mutex_exit(CMLB_MUTEX(cl));
967 		return (EINVAL);
968 	}
969 
970 	if (cl->cl_cur_labeltype == CMLB_LABEL_EFI) {
971 		if (cl->cl_reserved != -1) {
972 			blk = cl->cl_map[cl->cl_reserved].dkl_cylno;
973 		} else {
974 			mutex_exit(CMLB_MUTEX(cl));
975 			return (EINVAL);
976 		}
977 	} else {
978 		/* if the disk is unlabeled, don't write a devid to it */
979 		if (cl->cl_label_from_media != CMLB_LABEL_VTOC) {
980 			mutex_exit(CMLB_MUTEX(cl));
981 			return (EINVAL);
982 		}
983 
984 		/* this geometry doesn't allow us to write a devid */
985 		if (cl->cl_g.dkg_acyl < 2) {
986 			mutex_exit(CMLB_MUTEX(cl));
987 			return (EINVAL);
988 		}
989 
990 		/*
991 		 * Subtract 2 guarantees that the next to last cylinder
992 		 * is used
993 		 */
994 		cyl  = cl->cl_g.dkg_ncyl  + cl->cl_g.dkg_acyl - 2;
995 		spc  = cl->cl_g.dkg_nhead * cl->cl_g.dkg_nsect;
996 		head = cl->cl_g.dkg_nhead - 1;
997 		blk  = cl->cl_solaris_offset +
998 		    (cyl * (spc - cl->cl_g.dkg_apc)) +
999 		    (head * cl->cl_g.dkg_nsect) + 1;
1000 	}
1001 
1002 	*devidblockp = blk;
1003 	mutex_exit(CMLB_MUTEX(cl));
1004 	return (0);
1005 }
1006 
1007 /*
1008  * cmlb_partinfo:
1009  *	Get partition info for specified partition number.
1010  *
1011  * Arguments:
1012  *	cmlbhandle	cmlb handle associated with device.
1013  *	part		partition number
1014  *	nblocksp	pointer to number of blocks
1015  *	startblockp	pointer to starting block
1016  *	partnamep	pointer to name of partition
1017  *	tagp		pointer to tag info
1018  *	tg_cookie	cookie from target driver to be passed back to target
1019  *			driver when we call back to it through tg_ops.
1020  *
1021  *
1022  * Notes:
1023  *	If in-core label is not valid, this functions tries to revalidate
1024  *	the label. If label is valid, it stores the total number of blocks
1025  *	in this partition in the area pointed to by nblocksp, starting
1026  *	block number in area pointed to by startblockp,  pointer to partition
1027  *	name in area pointed to by partnamep, and tag value in area
1028  *	pointed by tagp.
1029  *	For EFI labels, tag value will be set to 0.
1030  *
1031  *	For all nblocksp, startblockp and partnamep, tagp, a value of NULL
1032  *	indicates the corresponding info is not requested.
1033  *
1034  *
1035  * Return values:
1036  *	0	success
1037  *	EINVAL  no valid label or requested partition number is invalid.
1038  *
1039  */
1040 int
1041 cmlb_partinfo(cmlb_handle_t cmlbhandle, int part, diskaddr_t *nblocksp,
1042     diskaddr_t *startblockp, char **partnamep, uint16_t *tagp, void *tg_cookie)
1043 {
1044 
1045 	struct cmlb_lun *cl = (struct cmlb_lun *)cmlbhandle;
1046 	int rval;
1047 #if defined(__i386) || defined(__amd64)
1048 	int ext_part;
1049 #endif
1050 
1051 	ASSERT(cl != NULL);
1052 	mutex_enter(CMLB_MUTEX(cl));
1053 	if (cl->cl_state < CMLB_ATTACHED) {
1054 		mutex_exit(CMLB_MUTEX(cl));
1055 		return (EINVAL);
1056 	}
1057 
1058 	if (part  < 0 || part >= MAXPART) {
1059 		rval = EINVAL;
1060 	} else {
1061 		if (!cl->cl_f_geometry_is_valid)
1062 			(void) cmlb_validate_geometry((struct cmlb_lun *)cl,
1063 			    B_FALSE, 0, tg_cookie);
1064 
1065 		if (((!cl->cl_f_geometry_is_valid) ||
1066 		    (part < NDKMAP && cl->cl_solaris_size == 0)) &&
1067 		    (part != P0_RAW_DISK)) {
1068 			rval = EINVAL;
1069 		} else {
1070 			if (startblockp != NULL)
1071 				*startblockp = (diskaddr_t)cl->cl_offset[part];
1072 
1073 			if (nblocksp != NULL)
1074 				*nblocksp = (diskaddr_t)
1075 				    cl->cl_map[part].dkl_nblk;
1076 
1077 			if (tagp != NULL)
1078 				*tagp =
1079 				    ((cl->cl_cur_labeltype == CMLB_LABEL_EFI) ||
1080 				    (part >= NDKMAP)) ? V_UNASSIGNED :
1081 				    cl->cl_vtoc.v_part[part].p_tag;
1082 			rval = 0;
1083 		}
1084 
1085 		/* consistent with behavior of sd for getting minor name */
1086 		if (partnamep != NULL) {
1087 #if defined(__i386) || defined(__amd64)
1088 #if defined(_FIRMWARE_NEEDS_FDISK)
1089 		if (part > FDISK_P4) {
1090 			ext_part = part-FDISK_P4-1;
1091 			*partnamep = dk_ext_minor_data[ext_part].name;
1092 		} else
1093 #endif
1094 #endif
1095 			*partnamep = dk_minor_data[part].name;
1096 		}
1097 
1098 	}
1099 
1100 	mutex_exit(CMLB_MUTEX(cl));
1101 	return (rval);
1102 }
1103 
1104 /*
1105  * cmlb_efi_label_capacity:
1106  *	Get capacity stored in EFI disk label.
1107  *
1108  * Arguments:
1109  *	cmlbhandle	cmlb handle associated with device.
1110  *	capacity	pointer to capacity stored in EFI disk label.
1111  *	tg_cookie	cookie from target driver to be passed back to target
1112  *			driver when we call back to it through tg_ops.
1113  *
1114  *
1115  * Notes:
1116  *	If in-core label is not valid, this functions tries to revalidate
1117  *	the label. If label is valid and is an EFI label, it stores the capacity
1118  *      in disk label in the area pointed to by capacity.
1119  *
1120  *
1121  * Return values:
1122  *	0	success
1123  *	EINVAL  no valid EFI label or capacity is NULL.
1124  *
1125  */
1126 int
1127 cmlb_efi_label_capacity(cmlb_handle_t cmlbhandle, diskaddr_t *capacity,
1128     void *tg_cookie)
1129 {
1130 	struct cmlb_lun *cl = (struct cmlb_lun *)cmlbhandle;
1131 	int rval;
1132 
1133 	ASSERT(cl != NULL);
1134 	mutex_enter(CMLB_MUTEX(cl));
1135 	if (cl->cl_state < CMLB_ATTACHED) {
1136 		mutex_exit(CMLB_MUTEX(cl));
1137 		return (EINVAL);
1138 	}
1139 
1140 	if (!cl->cl_f_geometry_is_valid)
1141 		(void) cmlb_validate_geometry((struct cmlb_lun *)cl, B_FALSE,
1142 		    0, tg_cookie);
1143 
1144 	if ((!cl->cl_f_geometry_is_valid) || (capacity == NULL) ||
1145 	    (cl->cl_cur_labeltype != CMLB_LABEL_EFI)) {
1146 		rval = EINVAL;
1147 	} else {
1148 		*capacity = (diskaddr_t)cl->cl_map[WD_NODE].dkl_nblk;
1149 		rval = 0;
1150 	}
1151 
1152 	mutex_exit(CMLB_MUTEX(cl));
1153 	return (rval);
1154 }
1155 
1156 /* Caller should make sure Test Unit Ready succeeds before calling this. */
1157 /*ARGSUSED*/
1158 int
1159 cmlb_ioctl(cmlb_handle_t cmlbhandle, dev_t dev, int cmd, intptr_t arg,
1160     int flag, cred_t *cred_p, int *rval_p, void *tg_cookie)
1161 {
1162 
1163 	int err;
1164 	struct cmlb_lun *cl;
1165 
1166 	cl = (struct cmlb_lun *)cmlbhandle;
1167 
1168 	ASSERT(cl != NULL);
1169 
1170 	mutex_enter(CMLB_MUTEX(cl));
1171 	if (cl->cl_state < CMLB_ATTACHED) {
1172 		mutex_exit(CMLB_MUTEX(cl));
1173 		return (EIO);
1174 	}
1175 
1176 	switch (cmd) {
1177 		case DKIOCSEXTVTOC:
1178 		case DKIOCSGEOM:
1179 		case DKIOCSETEFI:
1180 		case DKIOCSMBOOT:
1181 #if defined(__i386) || defined(__amd64)
1182 		case DKIOCSETEXTPART:
1183 #endif
1184 			break;
1185 		case DKIOCSVTOC:
1186 #if defined(__i386) || defined(__amd64)
1187 		case DKIOCPARTINFO:
1188 #endif
1189 			if (cl->cl_blockcount > CMLB_OLDVTOC_LIMIT) {
1190 				mutex_exit(CMLB_MUTEX(cl));
1191 				return (EOVERFLOW);
1192 			}
1193 			break;
1194 		default:
1195 			(void) cmlb_validate_geometry(cl, 1, CMLB_SILENT,
1196 			    tg_cookie);
1197 
1198 			switch (cmd) {
1199 			case DKIOCGVTOC:
1200 			case DKIOCGAPART:
1201 			case DKIOCSAPART:
1202 
1203 				if (cl->cl_label_from_media == CMLB_LABEL_EFI) {
1204 					/* GPT label on disk */
1205 					mutex_exit(CMLB_MUTEX(cl));
1206 					return (ENOTSUP);
1207 				} else if
1208 				    (cl->cl_blockcount > CMLB_OLDVTOC_LIMIT) {
1209 					mutex_exit(CMLB_MUTEX(cl));
1210 					return (EOVERFLOW);
1211 				}
1212 				break;
1213 
1214 			case DKIOCGGEOM:
1215 				if (cl->cl_label_from_media == CMLB_LABEL_EFI) {
1216 					/* GPT label on disk */
1217 					mutex_exit(CMLB_MUTEX(cl));
1218 					return (ENOTSUP);
1219 				}
1220 				break;
1221 			default:
1222 				break;
1223 			}
1224 	}
1225 
1226 	mutex_exit(CMLB_MUTEX(cl));
1227 
1228 	switch (cmd) {
1229 	case DKIOCGGEOM:
1230 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCGGEOM\n");
1231 		err = cmlb_dkio_get_geometry(cl, (caddr_t)arg, flag, tg_cookie);
1232 		break;
1233 
1234 	case DKIOCSGEOM:
1235 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCSGEOM\n");
1236 		err = cmlb_dkio_set_geometry(cl, (caddr_t)arg, flag);
1237 		break;
1238 
1239 	case DKIOCGAPART:
1240 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCGAPART\n");
1241 		err = cmlb_dkio_get_partition(cl, (caddr_t)arg,
1242 		    flag, tg_cookie);
1243 		break;
1244 
1245 	case DKIOCSAPART:
1246 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCSAPART\n");
1247 		err = cmlb_dkio_set_partition(cl, (caddr_t)arg, flag);
1248 		break;
1249 
1250 	case DKIOCGVTOC:
1251 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCGVTOC\n");
1252 		err = cmlb_dkio_get_vtoc(cl, (caddr_t)arg, flag, tg_cookie);
1253 		break;
1254 
1255 	case DKIOCGEXTVTOC:
1256 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCGVTOC\n");
1257 		err = cmlb_dkio_get_extvtoc(cl, (caddr_t)arg, flag, tg_cookie);
1258 		break;
1259 
1260 	case DKIOCGETEFI:
1261 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCGETEFI\n");
1262 		err = cmlb_dkio_get_efi(cl, (caddr_t)arg, flag, tg_cookie);
1263 		break;
1264 
1265 	case DKIOCPARTITION:
1266 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCPARTITION\n");
1267 		err = cmlb_dkio_partition(cl, (caddr_t)arg, flag, tg_cookie);
1268 		break;
1269 
1270 	case DKIOCSVTOC:
1271 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCSVTOC\n");
1272 		err = cmlb_dkio_set_vtoc(cl, dev, (caddr_t)arg, flag,
1273 		    tg_cookie);
1274 		break;
1275 
1276 	case DKIOCSEXTVTOC:
1277 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCSVTOC\n");
1278 		err = cmlb_dkio_set_extvtoc(cl, dev, (caddr_t)arg, flag,
1279 		    tg_cookie);
1280 		break;
1281 
1282 	case DKIOCSETEFI:
1283 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCSETEFI\n");
1284 		err = cmlb_dkio_set_efi(cl, dev, (caddr_t)arg, flag, tg_cookie);
1285 		break;
1286 
1287 	case DKIOCGMBOOT:
1288 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCGMBOOT\n");
1289 		err = cmlb_dkio_get_mboot(cl, (caddr_t)arg, flag, tg_cookie);
1290 		break;
1291 
1292 	case DKIOCSMBOOT:
1293 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCSMBOOT\n");
1294 		err = cmlb_dkio_set_mboot(cl, (caddr_t)arg, flag, tg_cookie);
1295 		break;
1296 	case DKIOCG_PHYGEOM:
1297 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCG_PHYGEOM\n");
1298 #if defined(__i386) || defined(__amd64)
1299 		err = cmlb_dkio_get_phygeom(cl, (caddr_t)arg, flag, tg_cookie);
1300 #else
1301 		err = ENOTTY;
1302 #endif
1303 		break;
1304 	case DKIOCG_VIRTGEOM:
1305 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCG_VIRTGEOM\n");
1306 #if defined(__i386) || defined(__amd64)
1307 		err = cmlb_dkio_get_virtgeom(cl, (caddr_t)arg, flag);
1308 #else
1309 		err = ENOTTY;
1310 #endif
1311 		break;
1312 	case DKIOCPARTINFO:
1313 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCPARTINFO");
1314 #if defined(__i386) || defined(__amd64)
1315 		err = cmlb_dkio_partinfo(cl, dev, (caddr_t)arg, flag);
1316 #else
1317 		err = ENOTTY;
1318 #endif
1319 		break;
1320 	case DKIOCEXTPARTINFO:
1321 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCPARTINFO");
1322 #if defined(__i386) || defined(__amd64)
1323 		err = cmlb_dkio_extpartinfo(cl, dev, (caddr_t)arg, flag);
1324 #else
1325 		err = ENOTTY;
1326 #endif
1327 		break;
1328 #if defined(__i386) || defined(__amd64)
1329 	case DKIOCSETEXTPART:
1330 		cmlb_dbg(CMLB_TRACE, cl, "DKIOCSETEXTPART");
1331 		err = cmlb_dkio_set_ext_part(cl, (caddr_t)arg, flag, tg_cookie);
1332 		break;
1333 #endif
1334 	default:
1335 		err = ENOTTY;
1336 
1337 	}
1338 
1339 	/*
1340 	 * An ioctl that succeeds and changed ('set') size(9P) information
1341 	 * needs to invalidate the cached devinfo snapshot to avoid having
1342 	 * old information being returned in a snapshots.
1343 	 *
1344 	 * NB: When available, call ddi_change_minor_node() to clear
1345 	 * SSIZEVALID in specfs vnodes via spec_size_invalidate().
1346 	 */
1347 	if (err == 0) {
1348 		switch (cmd) {
1349 		case DKIOCSGEOM:
1350 		case DKIOCSAPART:
1351 		case DKIOCSVTOC:
1352 		case DKIOCSEXTVTOC:
1353 		case DKIOCSETEFI:
1354 			i_ddi_prop_dyn_cache_invalidate(CMLB_DEVINFO(cl),
1355 			    i_ddi_prop_dyn_driver_get(CMLB_DEVINFO(cl)));
1356 		}
1357 	}
1358 	return (err);
1359 }
1360 
1361 dev_t
1362 cmlb_make_device(struct cmlb_lun *cl)
1363 {
1364 	if (cl->cl_alter_behavior & CMLB_CREATE_P0_MINOR_NODE) {
1365 		return (makedevice(ddi_driver_major(CMLB_DEVINFO(cl)),
1366 		    ddi_get_instance(
1367 		    CMLB_DEVINFO(cl)) << CMLBUNIT_FORCE_P0_SHIFT));
1368 	} else {
1369 		return (makedevice(ddi_driver_major(CMLB_DEVINFO(cl)),
1370 		    ddi_get_instance(CMLB_DEVINFO(cl)) << CMLBUNIT_SHIFT));
1371 	}
1372 }
1373 
1374 /*
1375  * Function: cmlb_check_update_blockcount
1376  *
1377  * Description: If current capacity value is invalid, obtains the
1378  *		current capacity from target driver.
1379  *
1380  * Return Code: 0	success
1381  *		EIO	failure
1382  */
1383 static int
1384 cmlb_check_update_blockcount(struct cmlb_lun *cl, void *tg_cookie)
1385 {
1386 	int status;
1387 	diskaddr_t capacity;
1388 	uint32_t lbasize;
1389 
1390 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
1391 
1392 	if (cl->cl_f_geometry_is_valid)
1393 		return (0);
1394 
1395 	mutex_exit(CMLB_MUTEX(cl));
1396 	status = DK_TG_GETCAP(cl, &capacity, tg_cookie);
1397 	if (status != 0) {
1398 		mutex_enter(CMLB_MUTEX(cl));
1399 		return (EIO);
1400 	}
1401 
1402 	status = DK_TG_GETBLOCKSIZE(cl, &lbasize, tg_cookie);
1403 	mutex_enter(CMLB_MUTEX(cl));
1404 	if (status != 0)
1405 		return (EIO);
1406 
1407 	if ((capacity != 0) && (lbasize != 0)) {
1408 		cl->cl_blockcount = capacity;
1409 		cl->cl_tgt_blocksize = lbasize;
1410 		if (!cl->cl_is_removable) {
1411 			cl->cl_sys_blocksize = lbasize;
1412 		}
1413 		return (0);
1414 	} else {
1415 		return (EIO);
1416 	}
1417 }
1418 
1419 static int
1420 cmlb_create_minor(dev_info_t *dip, char *name, int spec_type,
1421     minor_t minor_num, char *node_type, int flag, boolean_t internal)
1422 {
1423 	ASSERT(VALID_BOOLEAN(internal));
1424 
1425 	if (internal)
1426 		return (ddi_create_internal_pathname(dip,
1427 		    name, spec_type, minor_num));
1428 	else
1429 		return (ddi_create_minor_node(dip,
1430 		    name, spec_type, minor_num, node_type, flag));
1431 }
1432 
1433 /*
1434  *    Function: cmlb_create_minor_nodes
1435  *
1436  * Description: Create or adjust the minor device nodes for the instance.
1437  * 		Minor nodes are created based on default label type,
1438  *		current label type and last label type we created
1439  *		minor nodes based on.
1440  *
1441  *
1442  *   Arguments: cl - driver soft state (unit) structure
1443  *
1444  * Return Code: 0 success
1445  *		ENXIO	failure.
1446  *
1447  *     Context: Kernel thread context
1448  */
1449 static int
1450 cmlb_create_minor_nodes(struct cmlb_lun *cl)
1451 {
1452 	struct driver_minor_data	*dmdp;
1453 	int				instance, shift;
1454 	char				name[48];
1455 	cmlb_label_t			newlabeltype;
1456 	boolean_t			internal;
1457 
1458 	ASSERT(cl != NULL);
1459 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
1460 
1461 	internal = VOID2BOOLEAN(
1462 	    (cl->cl_alter_behavior & (CMLB_INTERNAL_MINOR_NODES)) != 0);
1463 
1464 	if (cl->cl_alter_behavior & CMLB_CREATE_P0_MINOR_NODE)
1465 		shift = CMLBUNIT_FORCE_P0_SHIFT;
1466 	else
1467 		shift = CMLBUNIT_SHIFT;
1468 
1469 	/* check the most common case */
1470 	if (cl->cl_cur_labeltype != CMLB_LABEL_UNDEF &&
1471 	    cl->cl_last_labeltype == cl->cl_cur_labeltype) {
1472 		/* do nothing */
1473 		return (0);
1474 	}
1475 
1476 	if (cl->cl_def_labeltype == CMLB_LABEL_UNDEF) {
1477 		/* we should never get here */
1478 		return (ENXIO);
1479 	}
1480 
1481 	if (cl->cl_last_labeltype == CMLB_LABEL_UNDEF) {
1482 		/* first time during attach */
1483 		newlabeltype = cl->cl_def_labeltype;
1484 
1485 		instance = ddi_get_instance(CMLB_DEVINFO(cl));
1486 
1487 		/* Create all the minor nodes for this target. */
1488 		dmdp = (newlabeltype == CMLB_LABEL_EFI) ? dk_minor_data_efi :
1489 		    dk_minor_data;
1490 		while (dmdp->name != NULL) {
1491 
1492 			(void) sprintf(name, "%s", dmdp->name);
1493 
1494 			if (cmlb_create_minor(CMLB_DEVINFO(cl), name,
1495 			    dmdp->type,
1496 			    (instance << shift) | dmdp->minor,
1497 			    cl->cl_node_type, NULL, internal) == DDI_FAILURE) {
1498 				/*
1499 				 * Clean up any nodes that may have been
1500 				 * created, in case this fails in the middle
1501 				 * of the loop.
1502 				 */
1503 				ddi_remove_minor_node(CMLB_DEVINFO(cl), NULL);
1504 				return (ENXIO);
1505 			}
1506 			dmdp++;
1507 		}
1508 		cl->cl_last_labeltype = newlabeltype;
1509 #if defined(_SUNOS_VTOC_8)
1510 		/*
1511 		 * "emulate" p0 device for sparc, used by lofi
1512 		 */
1513 		if (cl->cl_alter_behavior & CMLB_CREATE_P0_MINOR_NODE) {
1514 			if (cmlb_create_minor(CMLB_DEVINFO(cl), "q", S_IFBLK,
1515 			    (instance << CMLBUNIT_FORCE_P0_SHIFT) | P0_RAW_DISK,
1516 			    cl->cl_node_type, NULL, internal) == DDI_FAILURE) {
1517 				ddi_remove_minor_node(CMLB_DEVINFO(cl), NULL);
1518 				return (ENXIO);
1519 			}
1520 
1521 			if (cmlb_create_minor(CMLB_DEVINFO(cl), "q,raw",
1522 			    S_IFCHR,
1523 			    (instance << CMLBUNIT_FORCE_P0_SHIFT) | P0_RAW_DISK,
1524 			    cl->cl_node_type, NULL, internal) == DDI_FAILURE) {
1525 				ddi_remove_minor_node(CMLB_DEVINFO(cl), NULL);
1526 				return (ENXIO);
1527 			}
1528 		}
1529 #endif	/* defined(_SUNOS_VTOC_8) */
1530 		return (0);
1531 	}
1532 
1533 	/* Not first time  */
1534 	if (cl->cl_cur_labeltype == CMLB_LABEL_UNDEF) {
1535 		if (cl->cl_last_labeltype != cl->cl_def_labeltype) {
1536 			/* close time, revert to default. */
1537 			newlabeltype = cl->cl_def_labeltype;
1538 		} else {
1539 			/*
1540 			 * do nothing since the type for which we last created
1541 			 * nodes matches the default
1542 			 */
1543 			return (0);
1544 		}
1545 	} else {
1546 		if (cl->cl_cur_labeltype != cl->cl_last_labeltype) {
1547 			/* We are not closing, use current label type */
1548 			newlabeltype = cl->cl_cur_labeltype;
1549 		} else {
1550 			/*
1551 			 * do nothing since the type for which we last created
1552 			 * nodes matches the current label type
1553 			 */
1554 			return (0);
1555 		}
1556 	}
1557 
1558 	instance = ddi_get_instance(CMLB_DEVINFO(cl));
1559 
1560 	/*
1561 	 * Currently we only fix up the s7 node when we are switching
1562 	 * label types from or to EFI. This is consistent with
1563 	 * current behavior of sd.
1564 	 */
1565 	if (newlabeltype == CMLB_LABEL_EFI &&
1566 	    cl->cl_last_labeltype != CMLB_LABEL_EFI) {
1567 		/* from vtoc to EFI */
1568 		ddi_remove_minor_node(CMLB_DEVINFO(cl), "h");
1569 		ddi_remove_minor_node(CMLB_DEVINFO(cl), "h,raw");
1570 		(void) cmlb_create_minor(CMLB_DEVINFO(cl), "wd",
1571 		    S_IFBLK, (instance << shift) | WD_NODE,
1572 		    cl->cl_node_type, NULL, internal);
1573 		(void) cmlb_create_minor(CMLB_DEVINFO(cl), "wd,raw",
1574 		    S_IFCHR, (instance << shift) | WD_NODE,
1575 		    cl->cl_node_type, NULL, internal);
1576 	} else {
1577 		/* from efi to vtoc */
1578 		ddi_remove_minor_node(CMLB_DEVINFO(cl), "wd");
1579 		ddi_remove_minor_node(CMLB_DEVINFO(cl), "wd,raw");
1580 		(void) cmlb_create_minor(CMLB_DEVINFO(cl), "h",
1581 		    S_IFBLK, (instance << shift) | WD_NODE,
1582 		    cl->cl_node_type, NULL, internal);
1583 		(void) cmlb_create_minor(CMLB_DEVINFO(cl), "h,raw",
1584 		    S_IFCHR, (instance << shift) | WD_NODE,
1585 		    cl->cl_node_type, NULL, internal);
1586 	}
1587 
1588 	cl->cl_last_labeltype = newlabeltype;
1589 	return (0);
1590 }
1591 
1592 /*
1593  *    Function: cmlb_validate_geometry
1594  *
1595  * Description: Read the label from the disk (if present). Update the unit's
1596  *		geometry and vtoc information from the data in the label.
1597  *		Verify that the label is valid.
1598  *
1599  *   Arguments:
1600  *	cl		driver soft state (unit) structure
1601  *
1602  *	forcerevalid	force revalidation even if we are already valid.
1603  *	flags		operation flags from target driver. Used for verbosity
1604  *			control	at this time.
1605  *	tg_cookie	cookie from target driver to be passed back to target
1606  *			driver when we call back to it through tg_ops.
1607  *
1608  * Return Code: 0 - Successful completion
1609  *		EINVAL  - Invalid value in cl->cl_tgt_blocksize or
1610  *			  cl->cl_blockcount; or label on disk is corrupted
1611  *			  or unreadable.
1612  *		EACCES  - Reservation conflict at the device.
1613  *		ENOMEM  - Resource allocation error
1614  *		ENOTSUP - geometry not applicable
1615  *
1616  *     Context: Kernel thread only (can sleep).
1617  */
1618 static int
1619 cmlb_validate_geometry(struct cmlb_lun *cl, boolean_t forcerevalid, int flags,
1620     void *tg_cookie)
1621 {
1622 	int		label_error = 0;
1623 	diskaddr_t	capacity;
1624 	int		count;
1625 
1626 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
1627 	ASSERT(VALID_BOOLEAN(forcerevalid));
1628 
1629 	if ((cl->cl_f_geometry_is_valid) && (!forcerevalid)) {
1630 		if (cl->cl_cur_labeltype == CMLB_LABEL_EFI)
1631 			return (ENOTSUP);
1632 		return (0);
1633 	}
1634 
1635 	if (cmlb_check_update_blockcount(cl, tg_cookie) != 0)
1636 		return (EIO);
1637 
1638 	capacity = cl->cl_blockcount;
1639 
1640 	/*
1641 	 * Set up the "whole disk" fdisk partition; this should always
1642 	 * exist, regardless of whether the disk contains an fdisk table
1643 	 * or vtoc.
1644 	 */
1645 	cl->cl_map[P0_RAW_DISK].dkl_cylno = 0;
1646 	cl->cl_offset[P0_RAW_DISK] = 0;
1647 	/*
1648 	 * note if capacity > int32_max(1TB) we are in 64bit environment
1649 	 * so no truncation happens
1650 	 */
1651 	cl->cl_map[P0_RAW_DISK].dkl_nblk  = capacity;
1652 
1653 	/*
1654 	 * Refresh the logical and physical geometry caches.
1655 	 * (data from MODE SENSE format/rigid disk geometry pages,
1656 	 * and scsi_ifgetcap("geometry").
1657 	 */
1658 	cmlb_resync_geom_caches(cl, capacity, tg_cookie);
1659 
1660 	cl->cl_label_from_media = CMLB_LABEL_UNDEF;
1661 	label_error = cmlb_use_efi(cl, capacity, flags, tg_cookie);
1662 	if (label_error == 0) {
1663 
1664 		/* found a valid EFI label */
1665 		cmlb_dbg(CMLB_TRACE, cl,
1666 		    "cmlb_validate_geometry: found EFI label\n");
1667 		/*
1668 		 * solaris_size and geometry_is_valid are set in
1669 		 * cmlb_use_efi
1670 		 */
1671 		return (ENOTSUP);
1672 	}
1673 
1674 	/* NO EFI label found */
1675 
1676 	if (capacity > CMLB_EXTVTOC_LIMIT) {
1677 		if (label_error == ESRCH) {
1678 			/*
1679 			 * they've configured a LUN over 2TB, but used
1680 			 * format.dat to restrict format's view of the
1681 			 * capacity to be under 2TB in some earlier Solaris
1682 			 * release.
1683 			 */
1684 			/* i.e > 2TB with a VTOC < 2TB */
1685 			if (!(flags & CMLB_SILENT) &&
1686 			    (cl->cl_msglog_flag & CMLB_ALLOW_2TB_WARN)) {
1687 
1688 				cmlb_log(CMLB_DEVINFO(cl), CMLB_LABEL(cl),
1689 				    CE_NOTE, "!Disk (%s%d) is limited to 2 TB "
1690 				    "due to VTOC label. To use the full "
1691 				    "capacity of the disk, use format(1M) to "
1692 				    "relabel the disk with EFI/GPT label.\n",
1693 				    CMLB_LABEL(cl),
1694 				    ddi_get_instance(CMLB_DEVINFO(cl)));
1695 
1696 				cl->cl_msglog_flag &= ~CMLB_ALLOW_2TB_WARN;
1697 			}
1698 		} else {
1699 				return (ENOTSUP);
1700 		}
1701 	}
1702 
1703 	label_error = 0;
1704 
1705 	/*
1706 	 * at this point it is either labeled with a VTOC or it is
1707 	 * under 1TB (<= 1TB actually for off-by-1)
1708 	 */
1709 
1710 	/*
1711 	 * Only DIRECT ACCESS devices will have Scl labels.
1712 	 * CD's supposedly have a Scl label, too
1713 	 */
1714 	if (cl->cl_device_type == DTYPE_DIRECT || ISREMOVABLE(cl)) {
1715 		struct	dk_label *dkl;
1716 		offset_t label_addr;
1717 		int	rval;
1718 		size_t	buffer_size;
1719 
1720 		/*
1721 		 * Note: This will set up cl->cl_solaris_size and
1722 		 * cl->cl_solaris_offset.
1723 		 */
1724 		rval = cmlb_read_fdisk(cl, capacity, tg_cookie);
1725 		if ((rval != 0) && !ISCD(cl)) {
1726 			ASSERT(mutex_owned(CMLB_MUTEX(cl)));
1727 			return (rval);
1728 		}
1729 
1730 		if (cl->cl_solaris_size <= DK_LABEL_LOC) {
1731 			/*
1732 			 * Found fdisk table but no Solaris partition entry,
1733 			 * so don't call cmlb_uselabel() and don't create
1734 			 * a default label.
1735 			 */
1736 			label_error = 0;
1737 			cl->cl_f_geometry_is_valid = B_TRUE;
1738 			goto no_solaris_partition;
1739 		}
1740 
1741 		label_addr = (daddr_t)(cl->cl_solaris_offset + DK_LABEL_LOC);
1742 
1743 		buffer_size = cl->cl_sys_blocksize;
1744 
1745 		cmlb_dbg(CMLB_TRACE, cl, "cmlb_validate_geometry: "
1746 		    "label_addr: 0x%x allocation size: 0x%x\n",
1747 		    label_addr, buffer_size);
1748 
1749 		if ((dkl = kmem_zalloc(buffer_size, KM_NOSLEEP)) == NULL)
1750 			return (ENOMEM);
1751 
1752 		mutex_exit(CMLB_MUTEX(cl));
1753 		rval = DK_TG_READ(cl, dkl, label_addr, buffer_size, tg_cookie);
1754 		mutex_enter(CMLB_MUTEX(cl));
1755 
1756 		switch (rval) {
1757 		case 0:
1758 			/*
1759 			 * cmlb_uselabel will establish that the geometry
1760 			 * is valid.
1761 			 */
1762 			if (cmlb_uselabel(cl,
1763 			    (struct dk_label *)(uintptr_t)dkl, flags) !=
1764 			    CMLB_LABEL_IS_VALID) {
1765 				label_error = EINVAL;
1766 			} else
1767 				cl->cl_label_from_media = CMLB_LABEL_VTOC;
1768 			break;
1769 		case EACCES:
1770 			label_error = EACCES;
1771 			break;
1772 		default:
1773 			label_error = EINVAL;
1774 			break;
1775 		}
1776 
1777 		kmem_free(dkl, buffer_size);
1778 	}
1779 
1780 	/*
1781 	 * If a valid label was not found, AND if no reservation conflict
1782 	 * was detected, then go ahead and create a default label (4069506).
1783 	 *
1784 	 * Note: currently, for VTOC_8 devices, the default label is created
1785 	 * for removables and hotpluggables only.  For VTOC_16 devices, the
1786 	 * default label will be created for all devices.
1787 	 * (see cmlb_build_default_label)
1788 	 */
1789 #if defined(_SUNOS_VTOC_8)
1790 	if ((ISREMOVABLE(cl) || ISHOTPLUGGABLE(cl)) &&
1791 	    (label_error != EACCES)) {
1792 #elif defined(_SUNOS_VTOC_16)
1793 	if (label_error != EACCES) {
1794 #endif
1795 		if (!cl->cl_f_geometry_is_valid) {
1796 			cmlb_build_default_label(cl, tg_cookie);
1797 		}
1798 		label_error = 0;
1799 	}
1800 
1801 no_solaris_partition:
1802 
1803 #if defined(_SUNOS_VTOC_16)
1804 	/*
1805 	 * If we have valid geometry, set up the remaining fdisk partitions.
1806 	 * Note that dkl_cylno is not used for the fdisk map entries, so
1807 	 * we set it to an entirely bogus value.
1808 	 */
1809 	for (count = 0; count < FDISK_PARTS; count++) {
1810 		cl->cl_map[FDISK_P1 + count].dkl_cylno = UINT16_MAX;
1811 		cl->cl_map[FDISK_P1 + count].dkl_nblk =
1812 		    cl->cl_fmap[count].fmap_nblk;
1813 
1814 		cl->cl_offset[FDISK_P1 + count] =
1815 		    cl->cl_fmap[count].fmap_start;
1816 	}
1817 #endif
1818 
1819 	for (count = 0; count < NDKMAP; count++) {
1820 #if defined(_SUNOS_VTOC_8)
1821 		struct dk_map *lp  = &cl->cl_map[count];
1822 		cl->cl_offset[count] =
1823 		    cl->cl_g.dkg_nhead * cl->cl_g.dkg_nsect * lp->dkl_cylno;
1824 #elif defined(_SUNOS_VTOC_16)
1825 		struct dkl_partition *vp = &cl->cl_vtoc.v_part[count];
1826 
1827 		cl->cl_offset[count] = vp->p_start + cl->cl_solaris_offset;
1828 #else
1829 #error "No VTOC format defined."
1830 #endif
1831 	}
1832 
1833 	return (label_error);
1834 }
1835 
1836 #if defined(_SUNOS_VTOC_16)
1837 /*
1838  *    Function: cmlb_convert_geometry
1839  *
1840  * Description: Convert physical geometry into a dk_geom structure. In
1841  *		other words, make sure we don't wrap 16-bit values.
1842  *		e.g. converting from geom_cache to dk_geom
1843  *
1844  *     Context: Kernel thread only
1845  */
1846 static void
1847 cmlb_convert_geometry(struct cmlb_lun *cl, diskaddr_t capacity,
1848     struct dk_geom *cl_g, void *tg_cookie)
1849 {
1850 
1851 	ASSERT(cl != NULL);
1852 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
1853 
1854 	/* Unlabeled SCSI floppy device */
1855 	if (capacity < 160) {
1856 		/* Less than 80K */
1857 		cl_g->dkg_nhead = 1;
1858 		cl_g->dkg_ncyl = capacity;
1859 		cl_g->dkg_nsect = 1;
1860 		return;
1861 	} else if (capacity <= 0x1000) {
1862 		cl_g->dkg_nhead = 2;
1863 		cl_g->dkg_ncyl = 80;
1864 		cl_g->dkg_nsect = capacity / (cl_g->dkg_nhead * cl_g->dkg_ncyl);
1865 		return;
1866 	}
1867 
1868 	/*
1869 	 * For all devices we calculate cylinders using the heads and sectors
1870 	 * we assign based on capacity of the device.  The algorithm is
1871 	 * designed to be compatible with the way other operating systems
1872 	 * lay out fdisk tables for X86 and to insure that the cylinders never
1873 	 * exceed 65535 to prevent problems with X86 ioctls that report
1874 	 * geometry.
1875 	 * For some smaller disk sizes we report geometry that matches those
1876 	 * used by X86 BIOS usage. For larger disks, we use SPT that are
1877 	 * multiples of 63, since other OSes that are not limited to 16-bits
1878 	 * for cylinders stop at 63 SPT we make do by using multiples of 63 SPT.
1879 	 *
1880 	 * The following table (in order) illustrates some end result
1881 	 * calculations:
1882 	 *
1883 	 * Maximum number of blocks 		nhead	nsect
1884 	 *
1885 	 * 2097152 (1GB)			64	32
1886 	 * 16777216 (8GB)			128	32
1887 	 * 1052819775 (502.02GB)		255  	63
1888 	 * 2105639550 (0.98TB)			255	126
1889 	 * 3158459325 (1.47TB)			255  	189
1890 	 * 4211279100 (1.96TB)			255  	252
1891 	 * 5264098875 (2.45TB)			255  	315
1892 	 * ...
1893 	 *
1894 	 * For Solid State Drive(SSD), it uses 4K page size inside and may be
1895 	 * double with every new generation. If the I/O is not aligned with
1896 	 * page size on SSDs, SSDs perform a lot slower.
1897 	 * By default, Solaris partition starts from cylinder 1. It will be
1898 	 * misaligned even with 4K if using heads(255) and SPT(63). To
1899 	 * workaround the problem, if the device is SSD, we use heads(224) and
1900 	 * SPT multiple of 56. Thus the default Solaris partition starts from
1901 	 * a position that aligns with 128K on a 512 bytes sector size SSD.
1902 	 */
1903 
1904 	if (capacity <= 0x200000) {
1905 		cl_g->dkg_nhead = 64;
1906 		cl_g->dkg_nsect = 32;
1907 	} else if (capacity <= 0x01000000) {
1908 		cl_g->dkg_nhead = 128;
1909 		cl_g->dkg_nsect = 32;
1910 	} else {
1911 		tg_attribute_t tgattribute;
1912 		int is_solid_state;
1913 		unsigned short nhead;
1914 		unsigned short nsect;
1915 
1916 		bzero(&tgattribute, sizeof (tg_attribute_t));
1917 
1918 		mutex_exit(CMLB_MUTEX(cl));
1919 		is_solid_state =
1920 		    (DK_TG_GETATTRIBUTE(cl, &tgattribute, tg_cookie) == 0) ?
1921 		    tgattribute.media_is_solid_state : FALSE;
1922 		mutex_enter(CMLB_MUTEX(cl));
1923 
1924 		if (is_solid_state) {
1925 			nhead = 224;
1926 			nsect = 56;
1927 		} else {
1928 			nhead = 255;
1929 			nsect = 63;
1930 		}
1931 
1932 		cl_g->dkg_nhead = nhead;
1933 
1934 		/* make dkg_nsect be smallest multiple of nsect */
1935 		cl_g->dkg_nsect = ((capacity +
1936 		    (UINT16_MAX * nhead * nsect) - 1) /
1937 		    (UINT16_MAX * nhead * nsect)) * nsect;
1938 
1939 		if (cl_g->dkg_nsect == 0)
1940 			cl_g->dkg_nsect = (UINT16_MAX / nsect) * nsect;
1941 	}
1942 
1943 }
1944 #endif
1945 
1946 /*
1947  *    Function: cmlb_resync_geom_caches
1948  *
1949  * Description: (Re)initialize both geometry caches: the virtual geometry
1950  *            information is extracted from the HBA (the "geometry"
1951  *            capability), and the physical geometry cache data is
1952  *            generated by issuing MODE SENSE commands.
1953  *
1954  *   Arguments:
1955  *	cl 		driver soft state (unit) structure
1956  *	capacity	disk capacity in #blocks
1957  *	tg_cookie	cookie from target driver to be passed back to target
1958  *			driver when we call back to it through tg_ops.
1959  *
1960  *     Context: Kernel thread only (can sleep).
1961  */
1962 static void
1963 cmlb_resync_geom_caches(struct cmlb_lun *cl, diskaddr_t capacity,
1964     void *tg_cookie)
1965 {
1966 	struct	cmlb_geom 	pgeom;
1967 	struct	cmlb_geom	lgeom;
1968 	struct 	cmlb_geom	*pgeomp = &pgeom;
1969 	unsigned short 		nhead;
1970 	unsigned short 		nsect;
1971 	int 			spc;
1972 	int			ret;
1973 
1974 	ASSERT(cl != NULL);
1975 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
1976 
1977 	/*
1978 	 * Ask the controller for its logical geometry.
1979 	 * Note: if the HBA does not support scsi_ifgetcap("geometry"),
1980 	 * then the lgeom cache will be invalid.
1981 	 */
1982 	mutex_exit(CMLB_MUTEX(cl));
1983 	bzero(&lgeom, sizeof (struct cmlb_geom));
1984 	ret = DK_TG_GETVIRTGEOM(cl, &lgeom, tg_cookie);
1985 	mutex_enter(CMLB_MUTEX(cl));
1986 
1987 	bcopy(&lgeom, &cl->cl_lgeom, sizeof (cl->cl_lgeom));
1988 
1989 	/*
1990 	 * Initialize the pgeom cache from lgeom, so that if MODE SENSE
1991 	 * doesn't work, DKIOCG_PHYSGEOM can return reasonable values.
1992 	 */
1993 	if (ret != 0 || cl->cl_lgeom.g_nsect == 0 ||
1994 	    cl->cl_lgeom.g_nhead == 0) {
1995 		/*
1996 		 * Note: Perhaps this needs to be more adaptive? The rationale
1997 		 * is that, if there's no HBA geometry from the HBA driver, any
1998 		 * guess is good, since this is the physical geometry. If MODE
1999 		 * SENSE fails this gives a max cylinder size for non-LBA access
2000 		 */
2001 		nhead = 255;
2002 		nsect = 63;
2003 	} else {
2004 		nhead = cl->cl_lgeom.g_nhead;
2005 		nsect = cl->cl_lgeom.g_nsect;
2006 	}
2007 
2008 	if (ISCD(cl)) {
2009 		pgeomp->g_nhead = 1;
2010 		pgeomp->g_nsect = nsect * nhead;
2011 	} else {
2012 		pgeomp->g_nhead = nhead;
2013 		pgeomp->g_nsect = nsect;
2014 	}
2015 
2016 	spc = pgeomp->g_nhead * pgeomp->g_nsect;
2017 	pgeomp->g_capacity = capacity;
2018 	if (spc == 0)
2019 		pgeomp->g_ncyl = 0;
2020 	else
2021 		pgeomp->g_ncyl = pgeomp->g_capacity / spc;
2022 	pgeomp->g_acyl = 0;
2023 
2024 	/*
2025 	 * Retrieve fresh geometry data from the hardware, stash it
2026 	 * here temporarily before we rebuild the incore label.
2027 	 *
2028 	 * We want to use the MODE SENSE commands to derive the
2029 	 * physical geometry of the device, but if either command
2030 	 * fails, the logical geometry is used as the fallback for
2031 	 * disk label geometry.
2032 	 */
2033 
2034 	mutex_exit(CMLB_MUTEX(cl));
2035 	(void) DK_TG_GETPHYGEOM(cl, pgeomp, tg_cookie);
2036 	mutex_enter(CMLB_MUTEX(cl));
2037 
2038 	/*
2039 	 * Now update the real copy while holding the mutex. This
2040 	 * way the global copy is never in an inconsistent state.
2041 	 */
2042 	bcopy(pgeomp, &cl->cl_pgeom,  sizeof (cl->cl_pgeom));
2043 
2044 	cmlb_dbg(CMLB_INFO, cl, "cmlb_resync_geom_caches: "
2045 	    "(cached from lgeom)\n");
2046 	cmlb_dbg(CMLB_INFO,  cl,
2047 	    "   ncyl: %ld; acyl: %d; nhead: %d; nsect: %d\n",
2048 	    cl->cl_pgeom.g_ncyl, cl->cl_pgeom.g_acyl,
2049 	    cl->cl_pgeom.g_nhead, cl->cl_pgeom.g_nsect);
2050 	cmlb_dbg(CMLB_INFO,  cl, "   lbasize: %d; capacity: %ld; "
2051 	    "intrlv: %d; rpm: %d\n", cl->cl_pgeom.g_secsize,
2052 	    cl->cl_pgeom.g_capacity, cl->cl_pgeom.g_intrlv,
2053 	    cl->cl_pgeom.g_rpm);
2054 }
2055 
2056 
2057 #if defined(__i386) || defined(__amd64)
2058 /*
2059  *    Function: cmlb_update_ext_minor_nodes
2060  *
2061  * Description: Routine to add/remove extended partition device nodes
2062  *
2063  *   Arguments:
2064  *	cl		driver soft state (unit) structure
2065  *	num_parts	Number of logical drives found on the LUN
2066  *
2067  * Should be called with the mutex held
2068  *
2069  * Return Code: 0 for success
2070  *
2071  *     Context: User and Kernel thread
2072  *
2073  */
2074 static int
2075 cmlb_update_ext_minor_nodes(struct cmlb_lun *cl, int num_parts)
2076 {
2077 	int				i, count, shift;
2078 	char				name[48];
2079 	int				instance;
2080 	struct driver_minor_data	*demdp, *demdpr;
2081 	char				*devnm;
2082 	dev_info_t			*pdip;
2083 	boolean_t 			internal;
2084 
2085 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
2086 	ASSERT(cl->cl_update_ext_minor_nodes == 1);
2087 
2088 	internal = VOID2BOOLEAN(
2089 	    (cl->cl_alter_behavior & (CMLB_INTERNAL_MINOR_NODES)) != 0);
2090 	instance = ddi_get_instance(CMLB_DEVINFO(cl));
2091 	demdp = dk_ext_minor_data;
2092 	demdpr = &dk_ext_minor_data[MAX_EXT_PARTS];
2093 
2094 	if (cl->cl_alter_behavior & CMLB_CREATE_P0_MINOR_NODE)
2095 		shift = CMLBUNIT_FORCE_P0_SHIFT;
2096 	else
2097 		shift = CMLBUNIT_SHIFT;
2098 
2099 	if (cl->cl_logical_drive_count) {
2100 		for (i = 0; i < cl->cl_logical_drive_count; i++) {
2101 			(void) sprintf(name, "%s", demdp->name);
2102 			ddi_remove_minor_node(CMLB_DEVINFO(cl), name);
2103 			(void) sprintf(name, "%s", demdpr->name);
2104 			ddi_remove_minor_node(CMLB_DEVINFO(cl), name);
2105 			demdp++;
2106 			demdpr++;
2107 		}
2108 		/* There are existing device nodes. Remove them */
2109 		devnm = kmem_alloc(MAXNAMELEN + 1, KM_SLEEP);
2110 		(void) ddi_deviname(cl->cl_devi, devnm);
2111 		pdip = ddi_get_parent(cl->cl_devi);
2112 		(void) devfs_clean(pdip, devnm + 1, DV_CLEAN_FORCE);
2113 		kmem_free(devnm, MAXNAMELEN + 1);
2114 	}
2115 
2116 	demdp = dk_ext_minor_data;
2117 	demdpr = &dk_ext_minor_data[MAX_EXT_PARTS];
2118 
2119 	for (i = 0; i < num_parts; i++) {
2120 		(void) sprintf(name, "%s", demdp->name);
2121 		if (cmlb_create_minor(CMLB_DEVINFO(cl), name,
2122 		    demdp->type,
2123 		    (instance << shift) | demdp->minor,
2124 		    cl->cl_node_type, NULL, internal) == DDI_FAILURE) {
2125 			/*
2126 			 * Clean up any nodes that may have been
2127 			 * created, in case this fails in the middle
2128 			 * of the loop.
2129 			 */
2130 			ddi_remove_minor_node(CMLB_DEVINFO(cl), NULL);
2131 			cl->cl_logical_drive_count = 0;
2132 			return (ENXIO);
2133 		}
2134 		(void) sprintf(name, "%s", demdpr->name);
2135 		if (ddi_create_minor_node(CMLB_DEVINFO(cl), name,
2136 		    demdpr->type,
2137 		    (instance << shift) | demdpr->minor,
2138 		    cl->cl_node_type, NULL) == DDI_FAILURE) {
2139 			/*
2140 			 * Clean up any nodes that may have been
2141 			 * created, in case this fails in the middle
2142 			 * of the loop.
2143 			 */
2144 			ddi_remove_minor_node(CMLB_DEVINFO(cl), NULL);
2145 			cl->cl_logical_drive_count = 0;
2146 			return (ENXIO);
2147 		}
2148 		demdp++;
2149 		demdpr++;
2150 	}
2151 
2152 	/* Update the cl_map array for logical drives */
2153 	for (count = 0; count < MAX_EXT_PARTS; count++) {
2154 		cl->cl_map[FDISK_P4 + 1 + count].dkl_cylno = UINT32_MAX;
2155 		cl->cl_map[FDISK_P4 + 1 + count].dkl_nblk =
2156 		    cl->cl_fmap[FD_NUMPART + count].fmap_nblk;
2157 		cl->cl_offset[FDISK_P4 + 1 + count] =
2158 		    cl->cl_fmap[FD_NUMPART + count].fmap_start;
2159 	}
2160 
2161 	cl->cl_logical_drive_count = i;
2162 	cl->cl_update_ext_minor_nodes = 0;
2163 	return (0);
2164 }
2165 /*
2166  *    Function: cmlb_validate_ext_part
2167  *
2168  * Description: utility routine to validate an extended partition's
2169  *		metadata as found on disk
2170  *
2171  *   Arguments:
2172  *	cl		driver soft state (unit) structure
2173  *	part		partition number of the extended partition
2174  *	epart		partition number of the logical drive
2175  *	start		absolute sector number of the start of the logical
2176  *			drive being validated
2177  *	size		size of logical drive being validated
2178  *
2179  * Return Code: 0 for success
2180  *
2181  *     Context: User and Kernel thread
2182  *
2183  * Algorithm :
2184  * Error cases are :
2185  *	1. If start block is lesser than or equal to the end block
2186  *	2. If either start block or end block is beyond the bounadry
2187  *	   of the extended partition.
2188  *	3. start or end block overlap with existing partitions.
2189  *		To check this, first make sure that the start block doesnt
2190  *		overlap with existing partitions. Then, calculate the
2191  *		possible end block for the given start block that doesnt
2192  *		overlap with existing partitions. This can be calculated by
2193  *		first setting the possible end block to the end of the
2194  *		extended partition (optimistic) and then, checking if there
2195  *		is any other partition that lies after the start of the
2196  *		partition being validated. If so, set the possible end to
2197  *		one block less than the beginning of the next nearest partition
2198  *		If the actual end block is greater than the calculated end
2199  *		block, we have an overlap.
2200  *
2201  */
2202 static int
2203 cmlb_validate_ext_part(struct cmlb_lun *cl, int part, int epart, uint32_t start,
2204     uint32_t size)
2205 {
2206 	int i;
2207 	uint32_t end = start + size - 1;
2208 	uint32_t ext_start = cl->cl_fmap[part].fmap_start;
2209 	uint32_t ext_end = ext_start + cl->cl_fmap[part].fmap_nblk - 1;
2210 	uint32_t ts, te;
2211 	uint32_t poss_end = ext_end;
2212 
2213 	if (end <= start) {
2214 		return (1);
2215 	}
2216 
2217 	/*
2218 	 * Check if the logical drive boundaries are within that of the
2219 	 * extended partition.
2220 	 */
2221 	if (start <= ext_start || start > ext_end || end <= ext_start ||
2222 	    end > ext_end) {
2223 		return (1);
2224 	}
2225 
2226 	/*
2227 	 * epart will be equal to FD_NUMPART if it is the first logical drive.
2228 	 * There is no need to check for overlaps with other logical drives,
2229 	 * since it is the only logical drive that we have come across so far.
2230 	 */
2231 	if (epart == FD_NUMPART) {
2232 		return (0);
2233 	}
2234 
2235 	/* Check for overlaps with existing logical drives */
2236 	i = FD_NUMPART;
2237 	ts = cl->cl_fmap[FD_NUMPART].fmap_start;
2238 	te = ts + cl->cl_fmap[FD_NUMPART].fmap_nblk - 1;
2239 
2240 	while ((i < epart) && ts && te) {
2241 		if (start >= ts && start <= te) {
2242 			return (1);
2243 		}
2244 
2245 		if ((ts < poss_end) && (ts > start)) {
2246 			poss_end = ts - 1;
2247 		}
2248 
2249 		i++;
2250 		ts = cl->cl_fmap[i].fmap_start;
2251 		te = ts + cl->cl_fmap[i].fmap_nblk - 1;
2252 	}
2253 
2254 	if (end > poss_end) {
2255 		return (1);
2256 	}
2257 
2258 	return (0);
2259 }
2260 
2261 
2262 /*
2263  *    Function: cmlb_is_linux_swap
2264  *
2265  * Description: utility routine to verify if a partition is a linux swap
2266  *		partition or not.
2267  *
2268  *   Arguments:
2269  *	cl		driver soft state (unit) structure
2270  *	part_start	absolute sector number of the start of the partition
2271  *			being verified
2272  *	tg_cookie	cookie from target driver to be passed back to target
2273  *			driver when we call back to it through tg_ops.
2274  *
2275  * Return Code: 0 for success
2276  *
2277  *     Context: User and Kernel thread
2278  *
2279  * Notes:
2280  *	The linux swap magic "SWAP-SPACE" or "SWAPSPACE2" is found as the
2281  *	last 10 bytes of a disk block whose size is that of the linux page
2282  *	size. This disk block is found at the beginning of the swap partition.
2283  */
2284 static int
2285 cmlb_is_linux_swap(struct cmlb_lun *cl, uint32_t part_start, void *tg_cookie)
2286 {
2287 	int		i;
2288 	int		rval = -1;
2289 	uint32_t	seek_offset;
2290 	uint32_t	linux_pg_size;
2291 	char 		*buf, *linux_swap_magic;
2292 	int		sec_sz = cl->cl_sys_blocksize;
2293 	/* Known linux kernel page sizes */
2294 	uint32_t	linux_pg_size_arr[] = {4096, };
2295 
2296 	ASSERT(cl != NULL);
2297 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
2298 
2299 	if ((buf = kmem_zalloc(sec_sz, KM_NOSLEEP)) == NULL) {
2300 		return (ENOMEM);
2301 	}
2302 
2303 	/*
2304 	 * Check if there is a sane Solaris VTOC
2305 	 * If there is a valid vtoc, no need to lookup
2306 	 * for the linux swap signature.
2307 	 */
2308 	mutex_exit(CMLB_MUTEX(cl));
2309 	rval = DK_TG_READ(cl, buf, part_start + DK_LABEL_LOC,
2310 	    sec_sz, tg_cookie);
2311 	mutex_enter(CMLB_MUTEX(cl));
2312 	if (rval != 0) {
2313 		cmlb_dbg(CMLB_ERROR,  cl,
2314 		    "cmlb_is_linux_swap: disk vtoc read err\n");
2315 		rval = EIO;
2316 		goto done;
2317 	}
2318 
2319 	if ((((struct dk_label *)buf)->dkl_magic == DKL_MAGIC) &&
2320 	    (((struct dk_label *)buf)->dkl_vtoc.v_sanity == VTOC_SANE)) {
2321 		rval = -1;
2322 		goto done;
2323 	}
2324 
2325 
2326 	/* No valid vtoc, so check for linux swap signature */
2327 	linux_swap_magic = buf + sec_sz - 10;
2328 
2329 	for (i = 0; i < sizeof (linux_pg_size_arr)/sizeof (uint32_t); i++) {
2330 		linux_pg_size = linux_pg_size_arr[i];
2331 		seek_offset = linux_pg_size/sec_sz - 1;
2332 		seek_offset += part_start;
2333 
2334 		mutex_exit(CMLB_MUTEX(cl));
2335 		rval = DK_TG_READ(cl, buf, seek_offset, sec_sz, tg_cookie);
2336 		mutex_enter(CMLB_MUTEX(cl));
2337 
2338 		if (rval != 0) {
2339 			cmlb_dbg(CMLB_ERROR,  cl,
2340 			    "cmlb_is_linux_swap: disk read err\n");
2341 			rval = EIO;
2342 			break;
2343 		}
2344 
2345 		rval = -1;
2346 
2347 		if ((strncmp(linux_swap_magic, "SWAP-SPACE", 10) == 0) ||
2348 		    (strncmp(linux_swap_magic, "SWAPSPACE2", 10) == 0)) {
2349 			/* Found a linux swap */
2350 			rval = 0;
2351 			break;
2352 		}
2353 	}
2354 
2355 done:
2356 	kmem_free(buf, sec_sz);
2357 	return (rval);
2358 }
2359 #endif
2360 
2361 /*
2362  *    Function: cmlb_read_fdisk
2363  *
2364  * Description: utility routine to read the fdisk table.
2365  *
2366  *   Arguments:
2367  *	cl		driver soft state (unit) structure
2368  *	capacity	disk capacity in #blocks
2369  *	tg_cookie	cookie from target driver to be passed back to target
2370  *			driver when we call back to it through tg_ops.
2371  *
2372  * Return Code: 0 for success (includes not reading for no_fdisk_present case
2373  *		errnos from tg_rw if failed to read the first block.
2374  *
2375  *     Context: Kernel thread only (can sleep).
2376  */
2377 /*ARGSUSED*/
2378 static int
2379 cmlb_read_fdisk(struct cmlb_lun *cl, diskaddr_t capacity, void *tg_cookie)
2380 {
2381 #if defined(_NO_FDISK_PRESENT)
2382 
2383 	cl->cl_solaris_offset = 0;
2384 	cl->cl_solaris_size = capacity;
2385 	bzero(cl->cl_fmap, sizeof (struct fmap) * FD_NUMPART);
2386 	return (0);
2387 
2388 #elif defined(_FIRMWARE_NEEDS_FDISK)
2389 
2390 	struct ipart	*fdp;
2391 	struct mboot	*mbp;
2392 	struct ipart	fdisk[FD_NUMPART];
2393 	int		i, k;
2394 	char		sigbuf[2];
2395 	caddr_t		bufp;
2396 	int		uidx;
2397 	int 		rval;
2398 	int		lba = 0;
2399 	uint_t		solaris_offset;	/* offset to solaris part. */
2400 	daddr_t		solaris_size;	/* size of solaris partition */
2401 	uint32_t	blocksize;
2402 #if defined(__i386) || defined(__amd64)
2403 	struct ipart	eparts[2];
2404 	struct ipart	*efdp1 = &eparts[0];
2405 	struct ipart	*efdp2 = &eparts[1];
2406 	int		ext_part_exists = 0;
2407 	int		ld_count = 0;
2408 #endif
2409 
2410 	ASSERT(cl != NULL);
2411 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
2412 
2413 	/*
2414 	 * Start off assuming no fdisk table
2415 	 */
2416 	solaris_offset = 0;
2417 	solaris_size   = capacity;
2418 
2419 	blocksize = cl->cl_tgt_blocksize;
2420 
2421 	bufp = kmem_zalloc(blocksize, KM_SLEEP);
2422 
2423 	mutex_exit(CMLB_MUTEX(cl));
2424 	rval = DK_TG_READ(cl, bufp, 0, blocksize, tg_cookie);
2425 	mutex_enter(CMLB_MUTEX(cl));
2426 
2427 	if (rval != 0) {
2428 		cmlb_dbg(CMLB_ERROR,  cl,
2429 		    "cmlb_read_fdisk: fdisk read err\n");
2430 		bzero(cl->cl_fmap, sizeof (struct fmap) * FD_NUMPART);
2431 		goto done;
2432 	}
2433 
2434 	mbp = (struct mboot *)bufp;
2435 
2436 	/*
2437 	 * The fdisk table does not begin on a 4-byte boundary within the
2438 	 * master boot record, so we copy it to an aligned structure to avoid
2439 	 * alignment exceptions on some processors.
2440 	 */
2441 	bcopy(&mbp->parts[0], fdisk, sizeof (fdisk));
2442 
2443 	/*
2444 	 * Check for lba support before verifying sig; sig might not be
2445 	 * there, say on a blank disk, but the max_chs mark may still
2446 	 * be present.
2447 	 *
2448 	 * Note: LBA support and BEFs are an x86-only concept but this
2449 	 * code should work OK on SPARC as well.
2450 	 */
2451 
2452 	/*
2453 	 * First, check for lba-access-ok on root node (or prom root node)
2454 	 * if present there, don't need to search fdisk table.
2455 	 */
2456 	if (ddi_getprop(DDI_DEV_T_ANY, ddi_root_node(), 0,
2457 	    "lba-access-ok", 0) != 0) {
2458 		/* All drives do LBA; don't search fdisk table */
2459 		lba = 1;
2460 	} else {
2461 		/* Okay, look for mark in fdisk table */
2462 		for (fdp = fdisk, i = 0; i < FD_NUMPART; i++, fdp++) {
2463 			/* accumulate "lba" value from all partitions */
2464 			lba = (lba || cmlb_has_max_chs_vals(fdp));
2465 		}
2466 	}
2467 
2468 	if (lba != 0) {
2469 		dev_t dev = cmlb_make_device(cl);
2470 
2471 		if (ddi_getprop(dev, CMLB_DEVINFO(cl), DDI_PROP_DONTPASS,
2472 		    "lba-access-ok", 0) == 0) {
2473 			/* not found; create it */
2474 			if (ddi_prop_create(dev, CMLB_DEVINFO(cl), 0,
2475 			    "lba-access-ok", (caddr_t)NULL, 0) !=
2476 			    DDI_PROP_SUCCESS) {
2477 				cmlb_dbg(CMLB_ERROR,  cl,
2478 				    "cmlb_read_fdisk: Can't create lba "
2479 				    "property for instance %d\n",
2480 				    ddi_get_instance(CMLB_DEVINFO(cl)));
2481 			}
2482 		}
2483 	}
2484 
2485 	bcopy(&mbp->signature, sigbuf, sizeof (sigbuf));
2486 
2487 	/*
2488 	 * Endian-independent signature check
2489 	 */
2490 	if (((sigbuf[1] & 0xFF) != ((MBB_MAGIC >> 8) & 0xFF)) ||
2491 	    (sigbuf[0] != (MBB_MAGIC & 0xFF))) {
2492 		cmlb_dbg(CMLB_ERROR,  cl,
2493 		    "cmlb_read_fdisk: no fdisk\n");
2494 		bzero(cl->cl_fmap, sizeof (struct fmap) * FD_NUMPART);
2495 		goto done;
2496 	}
2497 
2498 #ifdef CMLBDEBUG
2499 	if (cmlb_level_mask & CMLB_LOGMASK_INFO) {
2500 		fdp = fdisk;
2501 		cmlb_dbg(CMLB_INFO,  cl, "cmlb_read_fdisk:\n");
2502 		cmlb_dbg(CMLB_INFO,  cl, "         relsect    "
2503 		    "numsect         sysid       bootid\n");
2504 		for (i = 0; i < FD_NUMPART; i++, fdp++) {
2505 			cmlb_dbg(CMLB_INFO,  cl,
2506 			    "    %d:  %8d   %8d     0x%08x     0x%08x\n",
2507 			    i, fdp->relsect, fdp->numsect,
2508 			    fdp->systid, fdp->bootid);
2509 		}
2510 	}
2511 #endif
2512 
2513 	/*
2514 	 * Try to find the unix partition
2515 	 */
2516 	uidx = -1;
2517 	solaris_offset = 0;
2518 	solaris_size   = 0;
2519 
2520 	for (fdp = fdisk, i = 0; i < FD_NUMPART; i++, fdp++) {
2521 		uint32_t relsect;
2522 		uint32_t numsect;
2523 		uchar_t systid;
2524 #if defined(__i386) || defined(__amd64)
2525 		/*
2526 		 * Stores relative block offset from the beginning of the
2527 		 * Extended Partition.
2528 		 */
2529 		int	ext_relsect = 0;
2530 #endif
2531 
2532 		if (fdp->numsect == 0) {
2533 			cl->cl_fmap[i].fmap_start = 0;
2534 			cl->cl_fmap[i].fmap_nblk  = 0;
2535 			continue;
2536 		}
2537 
2538 		/*
2539 		 * Data in the fdisk table is little-endian.
2540 		 */
2541 		relsect = LE_32(fdp->relsect);
2542 		numsect = LE_32(fdp->numsect);
2543 
2544 		cl->cl_fmap[i].fmap_start = relsect;
2545 		cl->cl_fmap[i].fmap_nblk  = numsect;
2546 		cl->cl_fmap[i].fmap_systid = LE_8(fdp->systid);
2547 
2548 #if defined(__i386) || defined(__amd64)
2549 		/* Support only one extended partition per LUN */
2550 		if ((fdp->systid == EXTDOS || fdp->systid == FDISK_EXTLBA) &&
2551 		    (ext_part_exists == 0)) {
2552 			int j;
2553 			uint32_t logdrive_offset;
2554 			uint32_t ext_numsect;
2555 			uint32_t abs_secnum;
2556 
2557 			ext_part_exists = 1;
2558 
2559 			for (j = FD_NUMPART; j < FDISK_PARTS; j++) {
2560 				mutex_exit(CMLB_MUTEX(cl));
2561 				rval = DK_TG_READ(cl, bufp,
2562 				    (relsect + ext_relsect), blocksize,
2563 				    tg_cookie);
2564 				mutex_enter(CMLB_MUTEX(cl));
2565 
2566 				if (rval != 0) {
2567 					cmlb_dbg(CMLB_ERROR,  cl,
2568 					    "cmlb_read_fdisk: Extended "
2569 					    "partition read err\n");
2570 					goto done;
2571 				}
2572 				/*
2573 				 * The first ipart entry provides the offset
2574 				 * at which the logical drive starts off from
2575 				 * the beginning of the container partition
2576 				 * and the size of the logical drive.
2577 				 * The second ipart entry provides the offset
2578 				 * of the next container partition from the
2579 				 * beginning of the extended partition.
2580 				 */
2581 				bcopy(&bufp[FDISK_PART_TABLE_START], eparts,
2582 				    sizeof (eparts));
2583 				logdrive_offset = LE_32(efdp1->relsect);
2584 				ext_numsect = LE_32(efdp1->numsect);
2585 				systid = LE_8(efdp1->systid);
2586 				if (logdrive_offset <= 0 || ext_numsect <= 0)
2587 					break;
2588 				abs_secnum = relsect + ext_relsect +
2589 				    logdrive_offset;
2590 
2591 				/* Boundary condition and overlap checking */
2592 				if (cmlb_validate_ext_part(cl, i, j, abs_secnum,
2593 				    ext_numsect)) {
2594 					break;
2595 				}
2596 
2597 				if ((cl->cl_fmap[j].fmap_start != abs_secnum) ||
2598 				    (cl->cl_fmap[j].fmap_nblk != ext_numsect) ||
2599 				    (cl->cl_fmap[j].fmap_systid != systid)) {
2600 					/*
2601 					 * Indicates change from previous
2602 					 * partinfo. Need to recreate
2603 					 * logical device nodes.
2604 					 */
2605 					cl->cl_update_ext_minor_nodes = 1;
2606 				}
2607 				cl->cl_fmap[j].fmap_start = abs_secnum;
2608 				cl->cl_fmap[j].fmap_nblk  = ext_numsect;
2609 				cl->cl_fmap[j].fmap_systid = systid;
2610 				ld_count++;
2611 
2612 				if ((efdp1->systid == SUNIXOS &&
2613 				    (cmlb_is_linux_swap(cl, abs_secnum,
2614 				    tg_cookie) != 0)) ||
2615 				    efdp1->systid == SUNIXOS2) {
2616 					if (uidx == -1) {
2617 						uidx = 0;
2618 						solaris_offset = abs_secnum;
2619 						solaris_size = ext_numsect;
2620 					}
2621 				}
2622 
2623 				if ((ext_relsect = LE_32(efdp2->relsect)) == 0)
2624 					break;
2625 			}
2626 		}
2627 
2628 #endif
2629 
2630 		if (fdp->systid != SUNIXOS &&
2631 		    fdp->systid != SUNIXOS2 &&
2632 		    fdp->systid != EFI_PMBR) {
2633 			continue;
2634 		}
2635 
2636 		/*
2637 		 * use the last active solaris partition id found
2638 		 * (there should only be 1 active partition id)
2639 		 *
2640 		 * if there are no active solaris partition id
2641 		 * then use the first inactive solaris partition id
2642 		 */
2643 		if ((uidx == -1) || (fdp->bootid == ACTIVE)) {
2644 #if defined(__i386) || defined(__amd64)
2645 			if (fdp->systid != SUNIXOS ||
2646 			    (fdp->systid == SUNIXOS &&
2647 			    (cmlb_is_linux_swap(cl, relsect,
2648 			    tg_cookie) != 0))) {
2649 #endif
2650 				uidx = i;
2651 				solaris_offset = relsect;
2652 				solaris_size   = numsect;
2653 #if defined(__i386) || defined(__amd64)
2654 			}
2655 #endif
2656 		}
2657 	}
2658 #if defined(__i386) || defined(__amd64)
2659 	if (ld_count < cl->cl_logical_drive_count) {
2660 		/*
2661 		 * Some/all logical drives were deleted. Clear out
2662 		 * the fmap entries correspoding to those deleted drives.
2663 		 */
2664 		for (k = ld_count + FD_NUMPART;
2665 		    k < cl->cl_logical_drive_count + FD_NUMPART; k++) {
2666 			cl->cl_fmap[k].fmap_start = 0;
2667 			cl->cl_fmap[k].fmap_nblk  = 0;
2668 			cl->cl_fmap[k].fmap_systid = 0;
2669 		}
2670 		cl->cl_update_ext_minor_nodes = 1;
2671 	}
2672 	if (cl->cl_update_ext_minor_nodes) {
2673 		rval = cmlb_update_ext_minor_nodes(cl, ld_count);
2674 		if (rval != 0) {
2675 			goto done;
2676 		}
2677 	}
2678 #endif
2679 	cmlb_dbg(CMLB_INFO,  cl, "fdisk 0x%x 0x%lx",
2680 	    cl->cl_solaris_offset, cl->cl_solaris_size);
2681 done:
2682 
2683 	/*
2684 	 * Clear the VTOC info, only if the Solaris partition entry
2685 	 * has moved, changed size, been deleted, or if the size of
2686 	 * the partition is too small to even fit the label sector.
2687 	 */
2688 	if ((cl->cl_solaris_offset != solaris_offset) ||
2689 	    (cl->cl_solaris_size != solaris_size) ||
2690 	    solaris_size <= DK_LABEL_LOC) {
2691 		cmlb_dbg(CMLB_INFO,  cl, "fdisk moved 0x%x 0x%lx",
2692 		    solaris_offset, solaris_size);
2693 		bzero(&cl->cl_g, sizeof (struct dk_geom));
2694 		bzero(&cl->cl_vtoc, sizeof (struct dk_vtoc));
2695 		bzero(&cl->cl_map, NDKMAP * (sizeof (struct dk_map)));
2696 		cl->cl_f_geometry_is_valid = B_FALSE;
2697 	}
2698 	cl->cl_solaris_offset = solaris_offset;
2699 	cl->cl_solaris_size = solaris_size;
2700 	kmem_free(bufp, blocksize);
2701 	return (rval);
2702 
2703 #else	/* #elif defined(_FIRMWARE_NEEDS_FDISK) */
2704 #error "fdisk table presence undetermined for this platform."
2705 #endif	/* #if defined(_NO_FDISK_PRESENT) */
2706 }
2707 
2708 static void
2709 cmlb_swap_efi_gpt(efi_gpt_t *e)
2710 {
2711 	_NOTE(ASSUMING_PROTECTED(*e))
2712 	e->efi_gpt_Signature = LE_64(e->efi_gpt_Signature);
2713 	e->efi_gpt_Revision = LE_32(e->efi_gpt_Revision);
2714 	e->efi_gpt_HeaderSize = LE_32(e->efi_gpt_HeaderSize);
2715 	e->efi_gpt_HeaderCRC32 = LE_32(e->efi_gpt_HeaderCRC32);
2716 	e->efi_gpt_MyLBA = LE_64(e->efi_gpt_MyLBA);
2717 	e->efi_gpt_AlternateLBA = LE_64(e->efi_gpt_AlternateLBA);
2718 	e->efi_gpt_FirstUsableLBA = LE_64(e->efi_gpt_FirstUsableLBA);
2719 	e->efi_gpt_LastUsableLBA = LE_64(e->efi_gpt_LastUsableLBA);
2720 	UUID_LE_CONVERT(e->efi_gpt_DiskGUID, e->efi_gpt_DiskGUID);
2721 	e->efi_gpt_PartitionEntryLBA = LE_64(e->efi_gpt_PartitionEntryLBA);
2722 	e->efi_gpt_NumberOfPartitionEntries =
2723 	    LE_32(e->efi_gpt_NumberOfPartitionEntries);
2724 	e->efi_gpt_SizeOfPartitionEntry =
2725 	    LE_32(e->efi_gpt_SizeOfPartitionEntry);
2726 	e->efi_gpt_PartitionEntryArrayCRC32 =
2727 	    LE_32(e->efi_gpt_PartitionEntryArrayCRC32);
2728 }
2729 
2730 static void
2731 cmlb_swap_efi_gpe(int nparts, efi_gpe_t *p)
2732 {
2733 	int i;
2734 
2735 	_NOTE(ASSUMING_PROTECTED(*p))
2736 	for (i = 0; i < nparts; i++) {
2737 		UUID_LE_CONVERT(p[i].efi_gpe_PartitionTypeGUID,
2738 		    p[i].efi_gpe_PartitionTypeGUID);
2739 		p[i].efi_gpe_StartingLBA = LE_64(p[i].efi_gpe_StartingLBA);
2740 		p[i].efi_gpe_EndingLBA = LE_64(p[i].efi_gpe_EndingLBA);
2741 		/* PartitionAttrs */
2742 	}
2743 }
2744 
2745 static int
2746 cmlb_validate_efi(efi_gpt_t *labp)
2747 {
2748 	if (labp->efi_gpt_Signature != EFI_SIGNATURE)
2749 		return (EINVAL);
2750 	/* at least 96 bytes in this version of the spec. */
2751 	if (sizeof (efi_gpt_t) - sizeof (labp->efi_gpt_Reserved2) >
2752 	    labp->efi_gpt_HeaderSize)
2753 		return (EINVAL);
2754 	/* this should be 128 bytes */
2755 	if (labp->efi_gpt_SizeOfPartitionEntry != sizeof (efi_gpe_t))
2756 		return (EINVAL);
2757 	return (0);
2758 }
2759 
2760 /*
2761  * This function returns B_FALSE if there is a valid MBR signature and no
2762  * partition table entries of type EFI_PMBR (0xEE). Otherwise it returns B_TRUE.
2763  *
2764  * The EFI spec (1.10 and later) requires having a Protective MBR (PMBR) to
2765  * recognize the disk as GPT partitioned. However, some other OS creates an MBR
2766  * where a PMBR entry is not the only one. Also, if the first block has been
2767  * corrupted, currently best attempt to allow data access would be to try to
2768  * check for GPT headers. Hence in case of more than one partition entry, but
2769  * at least one EFI_PMBR partition type or no valid magic number, the function
2770  * returns B_TRUE to continue with looking for GPT header.
2771  */
2772 
2773 static boolean_t
2774 cmlb_check_efi_mbr(uchar_t *buf, boolean_t *is_mbr)
2775 {
2776 	struct ipart	*fdp;
2777 	struct mboot	*mbp = (struct mboot *)buf;
2778 	struct ipart	fdisk[FD_NUMPART];
2779 	int		i;
2780 
2781 	if (is_mbr != NULL)
2782 		*is_mbr = B_TRUE;
2783 
2784 	if (LE_16(mbp->signature) != MBB_MAGIC) {
2785 		if (is_mbr != NULL)
2786 			*is_mbr = B_FALSE;
2787 		return (B_TRUE);
2788 	}
2789 
2790 	bcopy(&mbp->parts[0], fdisk, sizeof (fdisk));
2791 
2792 	for (fdp = fdisk, i = 0; i < FD_NUMPART; i++, fdp++) {
2793 		if (fdp->systid == EFI_PMBR)
2794 			return (B_TRUE);
2795 	}
2796 
2797 	return (B_FALSE);
2798 }
2799 
2800 static int
2801 cmlb_use_efi(struct cmlb_lun *cl, diskaddr_t capacity, int flags,
2802     void *tg_cookie)
2803 {
2804 	int		i;
2805 	int		rval = 0;
2806 	efi_gpe_t	*partitions;
2807 	uchar_t		*buf;
2808 	uint_t		lbasize;	/* is really how much to read */
2809 	diskaddr_t	cap = 0;
2810 	uint_t		nparts;
2811 	diskaddr_t	gpe_lba;
2812 	diskaddr_t	alternate_lba;
2813 	int		iofailed = 0;
2814 	struct uuid	uuid_type_reserved = EFI_RESERVED;
2815 #if defined(_FIRMWARE_NEEDS_FDISK)
2816 	boolean_t 	is_mbr;
2817 #endif
2818 
2819 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
2820 
2821 	lbasize = cl->cl_sys_blocksize;
2822 
2823 	cl->cl_reserved = -1;
2824 	mutex_exit(CMLB_MUTEX(cl));
2825 
2826 	buf = kmem_zalloc(EFI_MIN_ARRAY_SIZE, KM_SLEEP);
2827 
2828 	rval = DK_TG_READ(cl, buf,  0, lbasize, tg_cookie);
2829 	if (rval) {
2830 		iofailed = 1;
2831 		goto done_err;
2832 	}
2833 	if (((struct dk_label *)buf)->dkl_magic == DKL_MAGIC) {
2834 		/* not ours */
2835 		rval = ESRCH;
2836 		goto done_err;
2837 	}
2838 
2839 #if defined(_FIRMWARE_NEEDS_FDISK)
2840 	if (!cmlb_check_efi_mbr(buf, &is_mbr)) {
2841 		if (is_mbr)
2842 			rval = ESRCH;
2843 		else
2844 			rval = EINVAL;
2845 		goto done_err;
2846 	}
2847 #else
2848 	if (!cmlb_check_efi_mbr(buf, NULL)) {
2849 		rval = EINVAL;
2850 		goto done_err;
2851 	}
2852 
2853 #endif
2854 
2855 	rval = DK_TG_READ(cl, buf, 1, lbasize, tg_cookie);
2856 	if (rval) {
2857 		iofailed = 1;
2858 		goto done_err;
2859 	}
2860 	cmlb_swap_efi_gpt((efi_gpt_t *)buf);
2861 
2862 	if ((rval = cmlb_validate_efi((efi_gpt_t *)buf)) != 0) {
2863 		/*
2864 		 * Couldn't read the primary, try the backup.  Our
2865 		 * capacity at this point could be based on CHS, so
2866 		 * check what the device reports.
2867 		 */
2868 		rval = DK_TG_GETCAP(cl, &cap, tg_cookie);
2869 		if (rval) {
2870 			iofailed = 1;
2871 			goto done_err;
2872 		}
2873 
2874 		/*
2875 		 * CMLB_OFF_BY_ONE case, we check the next to last block first
2876 		 * for backup GPT header, otherwise check the last block.
2877 		 */
2878 
2879 		if ((rval = DK_TG_READ(cl, buf,
2880 		    cap - ((cl->cl_alter_behavior & CMLB_OFF_BY_ONE) ? 2 : 1),
2881 		    lbasize, tg_cookie))
2882 		    != 0) {
2883 			iofailed = 1;
2884 			goto done_err;
2885 		}
2886 		cmlb_swap_efi_gpt((efi_gpt_t *)buf);
2887 
2888 		if ((rval = cmlb_validate_efi((efi_gpt_t *)buf)) != 0) {
2889 
2890 			if (!(cl->cl_alter_behavior & CMLB_OFF_BY_ONE))
2891 				goto done_err;
2892 			if ((rval = DK_TG_READ(cl, buf, cap - 1, lbasize,
2893 			    tg_cookie)) != 0)
2894 				goto done_err;
2895 			cmlb_swap_efi_gpt((efi_gpt_t *)buf);
2896 			if ((rval = cmlb_validate_efi((efi_gpt_t *)buf)) != 0)
2897 				goto done_err;
2898 		}
2899 		if (!(flags & CMLB_SILENT))
2900 			cmlb_log(CMLB_DEVINFO(cl), CMLB_LABEL(cl), CE_WARN,
2901 			    "primary label corrupt; using backup\n");
2902 	}
2903 
2904 	nparts = ((efi_gpt_t *)buf)->efi_gpt_NumberOfPartitionEntries;
2905 	gpe_lba = ((efi_gpt_t *)buf)->efi_gpt_PartitionEntryLBA;
2906 	alternate_lba = ((efi_gpt_t *)buf)->efi_gpt_AlternateLBA;
2907 
2908 	rval = DK_TG_READ(cl, buf, gpe_lba, EFI_MIN_ARRAY_SIZE, tg_cookie);
2909 	if (rval) {
2910 		iofailed = 1;
2911 		goto done_err;
2912 	}
2913 	partitions = (efi_gpe_t *)buf;
2914 
2915 	if (nparts > MAXPART) {
2916 		nparts = MAXPART;
2917 	}
2918 	cmlb_swap_efi_gpe(nparts, partitions);
2919 
2920 	mutex_enter(CMLB_MUTEX(cl));
2921 
2922 	/* Fill in partition table. */
2923 	for (i = 0; i < nparts; i++) {
2924 		if (partitions->efi_gpe_StartingLBA != 0 ||
2925 		    partitions->efi_gpe_EndingLBA != 0) {
2926 			cl->cl_map[i].dkl_cylno =
2927 			    partitions->efi_gpe_StartingLBA;
2928 			cl->cl_map[i].dkl_nblk =
2929 			    partitions->efi_gpe_EndingLBA -
2930 			    partitions->efi_gpe_StartingLBA + 1;
2931 			cl->cl_offset[i] =
2932 			    partitions->efi_gpe_StartingLBA;
2933 		}
2934 
2935 		if (cl->cl_reserved == -1) {
2936 			if (bcmp(&partitions->efi_gpe_PartitionTypeGUID,
2937 			    &uuid_type_reserved, sizeof (struct uuid)) == 0) {
2938 				cl->cl_reserved = i;
2939 			}
2940 		}
2941 		if (i == WD_NODE) {
2942 			/*
2943 			 * minor number 7 corresponds to the whole disk
2944 			 * if the disk capacity is expanded after disk is
2945 			 * labeled, minor number 7 represents the capacity
2946 			 * indicated by the disk label.
2947 			 */
2948 			cl->cl_map[i].dkl_cylno = 0;
2949 			if (alternate_lba == 1) {
2950 				/*
2951 				 * We are using backup label. Since we can
2952 				 * find a valid label at the end of disk,
2953 				 * the disk capacity is not expanded.
2954 				 */
2955 				cl->cl_map[i].dkl_nblk = capacity;
2956 			} else {
2957 				cl->cl_map[i].dkl_nblk = alternate_lba + 1;
2958 			}
2959 			cl->cl_offset[i] = 0;
2960 		}
2961 		partitions++;
2962 	}
2963 	cl->cl_solaris_offset = 0;
2964 	cl->cl_solaris_size = capacity;
2965 	cl->cl_label_from_media = CMLB_LABEL_EFI;
2966 	cl->cl_f_geometry_is_valid = B_TRUE;
2967 
2968 	/* clear the vtoc label */
2969 	bzero(&cl->cl_vtoc, sizeof (struct dk_vtoc));
2970 
2971 	kmem_free(buf, EFI_MIN_ARRAY_SIZE);
2972 	return (0);
2973 
2974 done_err:
2975 	kmem_free(buf, EFI_MIN_ARRAY_SIZE);
2976 	mutex_enter(CMLB_MUTEX(cl));
2977 done_err1:
2978 	/*
2979 	 * if we didn't find something that could look like a VTOC
2980 	 * and the disk is over 1TB, we know there isn't a valid label.
2981 	 * Otherwise let cmlb_uselabel decide what to do.  We only
2982 	 * want to invalidate this if we're certain the label isn't
2983 	 * valid because cmlb_prop_op will now fail, which in turn
2984 	 * causes things like opens and stats on the partition to fail.
2985 	 */
2986 	if ((capacity > CMLB_EXTVTOC_LIMIT) && (rval != ESRCH) && !iofailed) {
2987 		cl->cl_f_geometry_is_valid = B_FALSE;
2988 	}
2989 	return (rval);
2990 }
2991 
2992 
2993 /*
2994  *    Function: cmlb_uselabel
2995  *
2996  * Description: Validate the disk label and update the relevant data (geometry,
2997  *		partition, vtoc, and capacity data) in the cmlb_lun struct.
2998  *		Marks the geometry of the unit as being valid.
2999  *
3000  *   Arguments: cl: unit struct.
3001  *		dk_label: disk label
3002  *
3003  * Return Code: CMLB_LABEL_IS_VALID: Label read from disk is OK; geometry,
3004  *		partition, vtoc, and capacity data are good.
3005  *
3006  *		CMLB_LABEL_IS_INVALID: Magic number or checksum error in the
3007  *		label; or computed capacity does not jibe with capacity
3008  *		reported from the READ CAPACITY command.
3009  *
3010  *     Context: Kernel thread only (can sleep).
3011  */
3012 static int
3013 cmlb_uselabel(struct cmlb_lun *cl, struct dk_label *labp, int flags)
3014 {
3015 	short		*sp;
3016 	short		sum;
3017 	short		count;
3018 	int		label_error = CMLB_LABEL_IS_VALID;
3019 	int		i;
3020 	diskaddr_t	label_capacity;
3021 	uint32_t	part_end;
3022 	diskaddr_t	track_capacity;
3023 #if defined(_SUNOS_VTOC_16)
3024 	struct	dkl_partition	*vpartp;
3025 #endif
3026 	ASSERT(cl != NULL);
3027 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
3028 
3029 	/* Validate the magic number of the label. */
3030 	if (labp->dkl_magic != DKL_MAGIC) {
3031 #if defined(__sparc)
3032 		if (!ISREMOVABLE(cl) && !ISHOTPLUGGABLE(cl)) {
3033 			if (!(flags & CMLB_SILENT))
3034 				cmlb_log(CMLB_DEVINFO(cl), CMLB_LABEL(cl),
3035 				    CE_WARN,
3036 				    "Corrupt label; wrong magic number\n");
3037 		}
3038 #endif
3039 		return (CMLB_LABEL_IS_INVALID);
3040 	}
3041 
3042 	/* Validate the checksum of the label. */
3043 	sp  = (short *)labp;
3044 	sum = 0;
3045 	count = sizeof (struct dk_label) / sizeof (short);
3046 	while (count--)	 {
3047 		sum ^= *sp++;
3048 	}
3049 
3050 	if (sum != 0) {
3051 #if defined(_SUNOS_VTOC_16)
3052 		if (!ISCD(cl)) {
3053 #elif defined(_SUNOS_VTOC_8)
3054 		if (!ISREMOVABLE(cl) && !ISHOTPLUGGABLE(cl)) {
3055 #endif
3056 			if (!(flags & CMLB_SILENT))
3057 				cmlb_log(CMLB_DEVINFO(cl), CMLB_LABEL(cl),
3058 				    CE_WARN,
3059 				    "Corrupt label - label checksum failed\n");
3060 		}
3061 		return (CMLB_LABEL_IS_INVALID);
3062 	}
3063 
3064 
3065 	/*
3066 	 * Fill in geometry structure with data from label.
3067 	 */
3068 	bzero(&cl->cl_g, sizeof (struct dk_geom));
3069 	cl->cl_g.dkg_ncyl   = labp->dkl_ncyl;
3070 	cl->cl_g.dkg_acyl   = labp->dkl_acyl;
3071 	cl->cl_g.dkg_bcyl   = 0;
3072 	cl->cl_g.dkg_nhead  = labp->dkl_nhead;
3073 	cl->cl_g.dkg_nsect  = labp->dkl_nsect;
3074 	cl->cl_g.dkg_intrlv = labp->dkl_intrlv;
3075 
3076 #if defined(_SUNOS_VTOC_8)
3077 	cl->cl_g.dkg_gap1   = labp->dkl_gap1;
3078 	cl->cl_g.dkg_gap2   = labp->dkl_gap2;
3079 	cl->cl_g.dkg_bhead  = labp->dkl_bhead;
3080 #endif
3081 #if defined(_SUNOS_VTOC_16)
3082 	cl->cl_dkg_skew = labp->dkl_skew;
3083 #endif
3084 
3085 #if defined(__i386) || defined(__amd64)
3086 	cl->cl_g.dkg_apc = labp->dkl_apc;
3087 #endif
3088 
3089 	/*
3090 	 * Currently we rely on the values in the label being accurate. If
3091 	 * dkl_rpm or dkl_pcly are zero in the label, use a default value.
3092 	 *
3093 	 * Note: In the future a MODE SENSE may be used to retrieve this data,
3094 	 * although this command is optional in SCSI-2.
3095 	 */
3096 	cl->cl_g.dkg_rpm  = (labp->dkl_rpm  != 0) ? labp->dkl_rpm  : 3600;
3097 	cl->cl_g.dkg_pcyl = (labp->dkl_pcyl != 0) ? labp->dkl_pcyl :
3098 	    (cl->cl_g.dkg_ncyl + cl->cl_g.dkg_acyl);
3099 
3100 	/*
3101 	 * The Read and Write reinstruct values may not be valid
3102 	 * for older disks.
3103 	 */
3104 	cl->cl_g.dkg_read_reinstruct  = labp->dkl_read_reinstruct;
3105 	cl->cl_g.dkg_write_reinstruct = labp->dkl_write_reinstruct;
3106 
3107 	/* Fill in partition table. */
3108 #if defined(_SUNOS_VTOC_8)
3109 	for (i = 0; i < NDKMAP; i++) {
3110 		cl->cl_map[i].dkl_cylno = labp->dkl_map[i].dkl_cylno;
3111 		cl->cl_map[i].dkl_nblk  = labp->dkl_map[i].dkl_nblk;
3112 	}
3113 #endif
3114 #if  defined(_SUNOS_VTOC_16)
3115 	vpartp		= labp->dkl_vtoc.v_part;
3116 	track_capacity	= labp->dkl_nhead * labp->dkl_nsect;
3117 
3118 	/* Prevent divide by zero */
3119 	if (track_capacity == 0) {
3120 		if (!(flags & CMLB_SILENT))
3121 			cmlb_log(CMLB_DEVINFO(cl), CMLB_LABEL(cl), CE_WARN,
3122 			    "Corrupt label - zero nhead or nsect value\n");
3123 
3124 		return (CMLB_LABEL_IS_INVALID);
3125 	}
3126 
3127 	for (i = 0; i < NDKMAP; i++, vpartp++) {
3128 		cl->cl_map[i].dkl_cylno = vpartp->p_start / track_capacity;
3129 		cl->cl_map[i].dkl_nblk  = vpartp->p_size;
3130 	}
3131 #endif
3132 
3133 	/* Fill in VTOC Structure. */
3134 	bcopy(&labp->dkl_vtoc, &cl->cl_vtoc, sizeof (struct dk_vtoc));
3135 #if defined(_SUNOS_VTOC_8)
3136 	/*
3137 	 * The 8-slice vtoc does not include the ascii label; save it into
3138 	 * the device's soft state structure here.
3139 	 */
3140 	bcopy(labp->dkl_asciilabel, cl->cl_asciilabel, LEN_DKL_ASCII);
3141 #endif
3142 
3143 	/* Now look for a valid capacity. */
3144 	track_capacity	= (cl->cl_g.dkg_nhead * cl->cl_g.dkg_nsect);
3145 	label_capacity	= (cl->cl_g.dkg_ncyl  * track_capacity);
3146 
3147 	if (cl->cl_g.dkg_acyl) {
3148 #if defined(__i386) || defined(__amd64)
3149 		/* we may have > 1 alts cylinder */
3150 		label_capacity += (track_capacity * cl->cl_g.dkg_acyl);
3151 #else
3152 		label_capacity += track_capacity;
3153 #endif
3154 	}
3155 
3156 	/*
3157 	 * Force check here to ensure the computed capacity is valid.
3158 	 * If capacity is zero, it indicates an invalid label and
3159 	 * we should abort updating the relevant data then.
3160 	 */
3161 	if (label_capacity == 0) {
3162 		if (!(flags & CMLB_SILENT))
3163 			cmlb_log(CMLB_DEVINFO(cl), CMLB_LABEL(cl), CE_WARN,
3164 			    "Corrupt label - no valid capacity could be "
3165 			    "retrieved\n");
3166 
3167 		return (CMLB_LABEL_IS_INVALID);
3168 	}
3169 
3170 	/* Mark the geometry as valid. */
3171 	cl->cl_f_geometry_is_valid = B_TRUE;
3172 
3173 	/*
3174 	 * if we got invalidated when mutex exit and entered again,
3175 	 * if blockcount different than when we came in, need to
3176 	 * retry from beginning of cmlb_validate_geometry.
3177 	 * revisit this on next phase of utilizing this for
3178 	 * sd.
3179 	 */
3180 
3181 	if (label_capacity <= cl->cl_blockcount) {
3182 #if defined(_SUNOS_VTOC_8)
3183 		/*
3184 		 * We can't let this happen on drives that are subdivided
3185 		 * into logical disks (i.e., that have an fdisk table).
3186 		 * The cl_blockcount field should always hold the full media
3187 		 * size in sectors, period.  This code would overwrite
3188 		 * cl_blockcount with the size of the Solaris fdisk partition.
3189 		 */
3190 		cmlb_dbg(CMLB_ERROR,  cl,
3191 		    "cmlb_uselabel: Label %d blocks; Drive %d blocks\n",
3192 		    label_capacity, cl->cl_blockcount);
3193 		cl->cl_solaris_size = label_capacity;
3194 
3195 #endif	/* defined(_SUNOS_VTOC_8) */
3196 		goto done;
3197 	}
3198 
3199 	if (ISCD(cl)) {
3200 		/* For CDROMs, we trust that the data in the label is OK. */
3201 #if defined(_SUNOS_VTOC_8)
3202 		for (i = 0; i < NDKMAP; i++) {
3203 			part_end = labp->dkl_nhead * labp->dkl_nsect *
3204 			    labp->dkl_map[i].dkl_cylno +
3205 			    labp->dkl_map[i].dkl_nblk  - 1;
3206 
3207 			if ((labp->dkl_map[i].dkl_nblk) &&
3208 			    (part_end > cl->cl_blockcount)) {
3209 				cl->cl_f_geometry_is_valid = B_FALSE;
3210 				break;
3211 			}
3212 		}
3213 #endif
3214 #if defined(_SUNOS_VTOC_16)
3215 		vpartp = &(labp->dkl_vtoc.v_part[0]);
3216 		for (i = 0; i < NDKMAP; i++, vpartp++) {
3217 			part_end = vpartp->p_start + vpartp->p_size;
3218 			if ((vpartp->p_size > 0) &&
3219 			    (part_end > cl->cl_blockcount)) {
3220 				cl->cl_f_geometry_is_valid = B_FALSE;
3221 				break;
3222 			}
3223 		}
3224 #endif
3225 	} else {
3226 		/* label_capacity > cl->cl_blockcount */
3227 		if (!(flags & CMLB_SILENT)) {
3228 			cmlb_log(CMLB_DEVINFO(cl), CMLB_LABEL(cl), CE_WARN,
3229 			    "Corrupt label - bad geometry\n");
3230 			cmlb_log(CMLB_DEVINFO(cl), CMLB_LABEL(cl), CE_CONT,
3231 			    "Label says %llu blocks; Drive says %llu blocks\n",
3232 			    label_capacity, cl->cl_blockcount);
3233 		}
3234 		cl->cl_f_geometry_is_valid = B_FALSE;
3235 		label_error = CMLB_LABEL_IS_INVALID;
3236 	}
3237 
3238 done:
3239 
3240 	cmlb_dbg(CMLB_INFO,  cl, "cmlb_uselabel: (label geometry)\n");
3241 	cmlb_dbg(CMLB_INFO,  cl,
3242 	    "   ncyl: %d; acyl: %d; nhead: %d; nsect: %d\n",
3243 	    cl->cl_g.dkg_ncyl,  cl->cl_g.dkg_acyl,
3244 	    cl->cl_g.dkg_nhead, cl->cl_g.dkg_nsect);
3245 
3246 	cmlb_dbg(CMLB_INFO,  cl,
3247 	    "   label_capacity: %d; intrlv: %d; rpm: %d\n",
3248 	    cl->cl_blockcount, cl->cl_g.dkg_intrlv, cl->cl_g.dkg_rpm);
3249 	cmlb_dbg(CMLB_INFO,  cl, "   wrt_reinstr: %d; rd_reinstr: %d\n",
3250 	    cl->cl_g.dkg_write_reinstruct, cl->cl_g.dkg_read_reinstruct);
3251 
3252 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
3253 
3254 	return (label_error);
3255 }
3256 
3257 
3258 /*
3259  *    Function: cmlb_build_default_label
3260  *
3261  * Description: Generate a default label for those devices that do not have
3262  *		one, e.g., new media, removable cartridges, etc..
3263  *
3264  *     Context: Kernel thread only
3265  */
3266 /*ARGSUSED*/
3267 static void
3268 cmlb_build_default_label(struct cmlb_lun *cl, void *tg_cookie)
3269 {
3270 #if defined(_SUNOS_VTOC_16)
3271 	uint_t	phys_spc;
3272 	uint_t	disksize;
3273 	struct  dk_geom cl_g;
3274 	diskaddr_t capacity;
3275 #endif
3276 
3277 	ASSERT(cl != NULL);
3278 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
3279 
3280 #if defined(_SUNOS_VTOC_8)
3281 	/*
3282 	 * Note: This is a legacy check for non-removable devices on VTOC_8
3283 	 * only. This may be a valid check for VTOC_16 as well.
3284 	 * Once we understand why there is this difference between SPARC and
3285 	 * x86 platform, we could remove this legacy check.
3286 	 */
3287 	if (!ISREMOVABLE(cl) && !ISHOTPLUGGABLE(cl)) {
3288 		return;
3289 	}
3290 #endif
3291 
3292 	bzero(&cl->cl_g, sizeof (struct dk_geom));
3293 	bzero(&cl->cl_vtoc, sizeof (struct dk_vtoc));
3294 	bzero(&cl->cl_map, NDKMAP * (sizeof (struct dk_map)));
3295 
3296 #if defined(_SUNOS_VTOC_8)
3297 
3298 	/*
3299 	 * It's a REMOVABLE media, therefore no label (on sparc, anyway).
3300 	 * But it is still necessary to set up various geometry information,
3301 	 * and we are doing this here.
3302 	 */
3303 
3304 	/*
3305 	 * For the rpm, we use the minimum for the disk.  For the head, cyl,
3306 	 * and number of sector per track, if the capacity <= 1GB, head = 64,
3307 	 * sect = 32.  else head = 255, sect 63 Note: the capacity should be
3308 	 * equal to C*H*S values.  This will cause some truncation of size due
3309 	 * to round off errors. For CD-ROMs, this truncation can have adverse
3310 	 * side effects, so returning ncyl and nhead as 1. The nsect will
3311 	 * overflow for most of CD-ROMs as nsect is of type ushort. (4190569)
3312 	 */
3313 	cl->cl_solaris_size = cl->cl_blockcount;
3314 	if (ISCD(cl)) {
3315 		tg_attribute_t tgattribute;
3316 		int is_writable;
3317 		/*
3318 		 * Preserve the old behavior for non-writable
3319 		 * medias. Since dkg_nsect is a ushort, it
3320 		 * will lose bits as cdroms have more than
3321 		 * 65536 sectors. So if we recalculate
3322 		 * capacity, it will become much shorter.
3323 		 * But the dkg_* information is not
3324 		 * used for CDROMs so it is OK. But for
3325 		 * Writable CDs we need this information
3326 		 * to be valid (for newfs say). So we
3327 		 * make nsect and nhead > 1 that way
3328 		 * nsect can still stay within ushort limit
3329 		 * without losing any bits.
3330 		 */
3331 
3332 		bzero(&tgattribute, sizeof (tg_attribute_t));
3333 
3334 		mutex_exit(CMLB_MUTEX(cl));
3335 		is_writable =
3336 		    (DK_TG_GETATTRIBUTE(cl, &tgattribute, tg_cookie) == 0) ?
3337 		    tgattribute.media_is_writable : 1;
3338 		mutex_enter(CMLB_MUTEX(cl));
3339 
3340 		if (is_writable) {
3341 			cl->cl_g.dkg_nhead = 64;
3342 			cl->cl_g.dkg_nsect = 32;
3343 			cl->cl_g.dkg_ncyl = cl->cl_blockcount / (64 * 32);
3344 			cl->cl_solaris_size = (diskaddr_t)cl->cl_g.dkg_ncyl *
3345 			    cl->cl_g.dkg_nhead * cl->cl_g.dkg_nsect;
3346 		} else {
3347 			cl->cl_g.dkg_ncyl  = 1;
3348 			cl->cl_g.dkg_nhead = 1;
3349 			cl->cl_g.dkg_nsect = cl->cl_blockcount;
3350 		}
3351 	} else {
3352 		if (cl->cl_blockcount < 160) {
3353 			/* Less than 80K */
3354 			cl->cl_g.dkg_nhead = 1;
3355 			cl->cl_g.dkg_ncyl = cl->cl_blockcount;
3356 			cl->cl_g.dkg_nsect = 1;
3357 		} else if (cl->cl_blockcount <= 0x1000) {
3358 			/* unlabeled SCSI floppy device */
3359 			cl->cl_g.dkg_nhead = 2;
3360 			cl->cl_g.dkg_ncyl = 80;
3361 			cl->cl_g.dkg_nsect = cl->cl_blockcount / (2 * 80);
3362 		} else if (cl->cl_blockcount <= 0x200000) {
3363 			cl->cl_g.dkg_nhead = 64;
3364 			cl->cl_g.dkg_nsect = 32;
3365 			cl->cl_g.dkg_ncyl  = cl->cl_blockcount / (64 * 32);
3366 		} else {
3367 			cl->cl_g.dkg_nhead = 255;
3368 
3369 			cl->cl_g.dkg_nsect = ((cl->cl_blockcount +
3370 			    (UINT16_MAX * 255 * 63) - 1) /
3371 			    (UINT16_MAX * 255 * 63)) * 63;
3372 
3373 			if (cl->cl_g.dkg_nsect == 0)
3374 				cl->cl_g.dkg_nsect = (UINT16_MAX / 63) * 63;
3375 
3376 			cl->cl_g.dkg_ncyl = cl->cl_blockcount /
3377 			    (255 * cl->cl_g.dkg_nsect);
3378 		}
3379 
3380 		cl->cl_solaris_size =
3381 		    (diskaddr_t)cl->cl_g.dkg_ncyl * cl->cl_g.dkg_nhead *
3382 		    cl->cl_g.dkg_nsect;
3383 
3384 	}
3385 
3386 	cl->cl_g.dkg_acyl	= 0;
3387 	cl->cl_g.dkg_bcyl	= 0;
3388 	cl->cl_g.dkg_rpm	= 200;
3389 	cl->cl_asciilabel[0]	= '\0';
3390 	cl->cl_g.dkg_pcyl	= cl->cl_g.dkg_ncyl;
3391 
3392 	cl->cl_map[0].dkl_cylno = 0;
3393 	cl->cl_map[0].dkl_nblk  = cl->cl_solaris_size;
3394 
3395 	cl->cl_map[2].dkl_cylno = 0;
3396 	cl->cl_map[2].dkl_nblk  = cl->cl_solaris_size;
3397 
3398 #elif defined(_SUNOS_VTOC_16)
3399 
3400 	if (cl->cl_solaris_size == 0) {
3401 		/*
3402 		 * Got fdisk table but no solaris entry therefore
3403 		 * don't create a default label
3404 		 */
3405 		cl->cl_f_geometry_is_valid = B_TRUE;
3406 		return;
3407 	}
3408 
3409 	/*
3410 	 * For CDs we continue to use the physical geometry to calculate
3411 	 * number of cylinders. All other devices must convert the
3412 	 * physical geometry (cmlb_geom) to values that will fit
3413 	 * in a dk_geom structure.
3414 	 */
3415 	if (ISCD(cl)) {
3416 		phys_spc = cl->cl_pgeom.g_nhead * cl->cl_pgeom.g_nsect;
3417 	} else {
3418 		/* Convert physical geometry to disk geometry */
3419 		bzero(&cl_g, sizeof (struct dk_geom));
3420 
3421 		/*
3422 		 * Refer to comments related to off-by-1 at the
3423 		 * header of this file.
3424 		 * Before calculating geometry, capacity should be
3425 		 * decreased by 1.
3426 		 */
3427 
3428 		if (cl->cl_alter_behavior & CMLB_OFF_BY_ONE)
3429 			capacity = cl->cl_blockcount - 1;
3430 		else
3431 			capacity = cl->cl_blockcount;
3432 
3433 
3434 		cmlb_convert_geometry(cl, capacity, &cl_g, tg_cookie);
3435 		bcopy(&cl_g, &cl->cl_g, sizeof (cl->cl_g));
3436 		phys_spc = cl->cl_g.dkg_nhead * cl->cl_g.dkg_nsect;
3437 	}
3438 
3439 	if (phys_spc == 0)
3440 		return;
3441 	cl->cl_g.dkg_pcyl = cl->cl_solaris_size / phys_spc;
3442 	if (cl->cl_alter_behavior & CMLB_FAKE_LABEL_ONE_PARTITION) {
3443 		/* disable devid */
3444 		cl->cl_g.dkg_ncyl = cl->cl_g.dkg_pcyl;
3445 		disksize = cl->cl_solaris_size;
3446 	} else {
3447 		cl->cl_g.dkg_acyl = DK_ACYL;
3448 		cl->cl_g.dkg_ncyl = cl->cl_g.dkg_pcyl - DK_ACYL;
3449 		disksize = cl->cl_g.dkg_ncyl * phys_spc;
3450 	}
3451 
3452 	if (ISCD(cl)) {
3453 		/*
3454 		 * CD's don't use the "heads * sectors * cyls"-type of
3455 		 * geometry, but instead use the entire capacity of the media.
3456 		 */
3457 		disksize = cl->cl_solaris_size;
3458 		cl->cl_g.dkg_nhead = 1;
3459 		cl->cl_g.dkg_nsect = 1;
3460 		cl->cl_g.dkg_rpm =
3461 		    (cl->cl_pgeom.g_rpm == 0) ? 200 : cl->cl_pgeom.g_rpm;
3462 
3463 		cl->cl_vtoc.v_part[0].p_start = 0;
3464 		cl->cl_vtoc.v_part[0].p_size  = disksize;
3465 		cl->cl_vtoc.v_part[0].p_tag   = V_BACKUP;
3466 		cl->cl_vtoc.v_part[0].p_flag  = V_UNMNT;
3467 
3468 		cl->cl_map[0].dkl_cylno = 0;
3469 		cl->cl_map[0].dkl_nblk  = disksize;
3470 		cl->cl_offset[0] = 0;
3471 
3472 	} else {
3473 		/*
3474 		 * Hard disks and removable media cartridges
3475 		 */
3476 		cl->cl_g.dkg_rpm =
3477 		    (cl->cl_pgeom.g_rpm == 0) ? 3600: cl->cl_pgeom.g_rpm;
3478 		cl->cl_vtoc.v_sectorsz = cl->cl_sys_blocksize;
3479 
3480 		/* Add boot slice */
3481 		cl->cl_vtoc.v_part[8].p_start = 0;
3482 		cl->cl_vtoc.v_part[8].p_size  = phys_spc;
3483 		cl->cl_vtoc.v_part[8].p_tag   = V_BOOT;
3484 		cl->cl_vtoc.v_part[8].p_flag  = V_UNMNT;
3485 
3486 		cl->cl_map[8].dkl_cylno = 0;
3487 		cl->cl_map[8].dkl_nblk  = phys_spc;
3488 		cl->cl_offset[8] = 0;
3489 
3490 		if ((cl->cl_alter_behavior &
3491 		    CMLB_CREATE_ALTSLICE_VTOC_16_DTYPE_DIRECT) &&
3492 		    cl->cl_device_type == DTYPE_DIRECT) {
3493 			cl->cl_vtoc.v_part[9].p_start = phys_spc;
3494 			cl->cl_vtoc.v_part[9].p_size  = 2 * phys_spc;
3495 			cl->cl_vtoc.v_part[9].p_tag   = V_ALTSCTR;
3496 			cl->cl_vtoc.v_part[9].p_flag  = 0;
3497 
3498 			cl->cl_map[9].dkl_cylno = 1;
3499 			cl->cl_map[9].dkl_nblk  = 2 * phys_spc;
3500 			cl->cl_offset[9] = phys_spc;
3501 		}
3502 	}
3503 
3504 	cl->cl_g.dkg_apc = 0;
3505 
3506 	/* Add backup slice */
3507 	cl->cl_vtoc.v_part[2].p_start = 0;
3508 	cl->cl_vtoc.v_part[2].p_size  = disksize;
3509 	cl->cl_vtoc.v_part[2].p_tag   = V_BACKUP;
3510 	cl->cl_vtoc.v_part[2].p_flag  = V_UNMNT;
3511 
3512 	cl->cl_map[2].dkl_cylno = 0;
3513 	cl->cl_map[2].dkl_nblk  = disksize;
3514 	cl->cl_offset[2] = 0;
3515 
3516 	/*
3517 	 * single slice (s0) covering the entire disk
3518 	 */
3519 	if (cl->cl_alter_behavior & CMLB_FAKE_LABEL_ONE_PARTITION) {
3520 		cl->cl_vtoc.v_part[0].p_start = 0;
3521 		cl->cl_vtoc.v_part[0].p_tag   = V_UNASSIGNED;
3522 		cl->cl_vtoc.v_part[0].p_flag  = 0;
3523 		cl->cl_vtoc.v_part[0].p_size  = disksize;
3524 		cl->cl_map[0].dkl_cylno = 0;
3525 		cl->cl_map[0].dkl_nblk  = disksize;
3526 		cl->cl_offset[0] = 0;
3527 	}
3528 
3529 	(void) sprintf(cl->cl_vtoc.v_asciilabel, "DEFAULT cyl %d alt %d"
3530 	    " hd %d sec %d", cl->cl_g.dkg_ncyl, cl->cl_g.dkg_acyl,
3531 	    cl->cl_g.dkg_nhead, cl->cl_g.dkg_nsect);
3532 
3533 #else
3534 #error "No VTOC format defined."
3535 #endif
3536 
3537 	cl->cl_g.dkg_read_reinstruct  = 0;
3538 	cl->cl_g.dkg_write_reinstruct = 0;
3539 
3540 	cl->cl_g.dkg_intrlv = 1;
3541 
3542 	cl->cl_vtoc.v_sanity  = VTOC_SANE;
3543 	cl->cl_vtoc.v_nparts = V_NUMPAR;
3544 	cl->cl_vtoc.v_version = V_VERSION;
3545 
3546 	cl->cl_f_geometry_is_valid = B_TRUE;
3547 	cl->cl_label_from_media = CMLB_LABEL_UNDEF;
3548 
3549 	cmlb_dbg(CMLB_INFO,  cl,
3550 	    "cmlb_build_default_label: Default label created: "
3551 	    "cyl: %d\tacyl: %d\tnhead: %d\tnsect: %d\tcap: %d\n",
3552 	    cl->cl_g.dkg_ncyl, cl->cl_g.dkg_acyl, cl->cl_g.dkg_nhead,
3553 	    cl->cl_g.dkg_nsect, cl->cl_blockcount);
3554 }
3555 
3556 
3557 #if defined(_FIRMWARE_NEEDS_FDISK)
3558 /*
3559  * Max CHS values, as they are encoded into bytes, for 1022/254/63
3560  */
3561 #define	LBA_MAX_SECT	(63 | ((1022 & 0x300) >> 2))
3562 #define	LBA_MAX_CYL	(1022 & 0xFF)
3563 #define	LBA_MAX_HEAD	(254)
3564 
3565 
3566 /*
3567  *    Function: cmlb_has_max_chs_vals
3568  *
3569  * Description: Return B_TRUE if Cylinder-Head-Sector values are all at maximum.
3570  *
3571  *   Arguments: fdp - ptr to CHS info
3572  *
3573  * Return Code: True or false
3574  *
3575  *     Context: Any.
3576  */
3577 static boolean_t
3578 cmlb_has_max_chs_vals(struct ipart *fdp)
3579 {
3580 	return ((fdp->begcyl  == LBA_MAX_CYL)	&&
3581 	    (fdp->beghead == LBA_MAX_HEAD)	&&
3582 	    (fdp->begsect == LBA_MAX_SECT)	&&
3583 	    (fdp->endcyl  == LBA_MAX_CYL)	&&
3584 	    (fdp->endhead == LBA_MAX_HEAD)	&&
3585 	    (fdp->endsect == LBA_MAX_SECT));
3586 }
3587 #endif
3588 
3589 /*
3590  *    Function: cmlb_dkio_get_geometry
3591  *
3592  * Description: This routine is the driver entry point for handling user
3593  *		requests to get the device geometry (DKIOCGGEOM).
3594  *
3595  *   Arguments:
3596  *	arg		pointer to user provided dk_geom structure specifying
3597  *			the controller's notion of the current geometry.
3598  *
3599  *	flag 		this argument is a pass through to ddi_copyxxx()
3600  *			directly from the mode argument of ioctl().
3601  *
3602  *	tg_cookie	cookie from target driver to be passed back to target
3603  *			driver when we call back to it through tg_ops.
3604  *
3605  * Return Code: 0
3606  *		EFAULT
3607  *		ENXIO
3608  *		EIO
3609  */
3610 static int
3611 cmlb_dkio_get_geometry(struct cmlb_lun *cl, caddr_t arg, int flag,
3612     void *tg_cookie)
3613 {
3614 	struct dk_geom	*tmp_geom = NULL;
3615 	int		rval = 0;
3616 
3617 	/*
3618 	 * cmlb_validate_geometry does not spin a disk up
3619 	 * if it was spcl down. We need to make sure it
3620 	 * is ready.
3621 	 */
3622 	mutex_enter(CMLB_MUTEX(cl));
3623 	rval = cmlb_validate_geometry(cl, B_TRUE, 0, tg_cookie);
3624 #if defined(_SUNOS_VTOC_8)
3625 	if (rval == EINVAL &&
3626 	    cl->cl_alter_behavior & CMLB_FAKE_GEOM_LABEL_IOCTLS_VTOC8) {
3627 		/*
3628 		 * This is to return a default label geometry even when we
3629 		 * do not really assume a default label for the device.
3630 		 * dad driver utilizes this.
3631 		 */
3632 		if (cl->cl_blockcount <= CMLB_OLDVTOC_LIMIT) {
3633 			cmlb_setup_default_geometry(cl, tg_cookie);
3634 			rval = 0;
3635 		}
3636 	}
3637 #endif
3638 	if (rval) {
3639 		mutex_exit(CMLB_MUTEX(cl));
3640 		return (rval);
3641 	}
3642 
3643 #if defined(__i386) || defined(__amd64)
3644 	if (cl->cl_solaris_size == 0) {
3645 		mutex_exit(CMLB_MUTEX(cl));
3646 		return (EIO);
3647 	}
3648 #endif
3649 
3650 	/*
3651 	 * Make a local copy of the soft state geometry to avoid some potential
3652 	 * race conditions associated with holding the mutex and updating the
3653 	 * write_reinstruct value
3654 	 */
3655 	tmp_geom = kmem_zalloc(sizeof (struct dk_geom), KM_SLEEP);
3656 	bcopy(&cl->cl_g, tmp_geom, sizeof (struct dk_geom));
3657 
3658 	if (tmp_geom->dkg_write_reinstruct == 0) {
3659 		tmp_geom->dkg_write_reinstruct =
3660 		    (int)((int)(tmp_geom->dkg_nsect * tmp_geom->dkg_rpm *
3661 		    cmlb_rot_delay) / (int)60000);
3662 	}
3663 	mutex_exit(CMLB_MUTEX(cl));
3664 
3665 	rval = ddi_copyout(tmp_geom, (void *)arg, sizeof (struct dk_geom),
3666 	    flag);
3667 	if (rval != 0) {
3668 		rval = EFAULT;
3669 	}
3670 
3671 	kmem_free(tmp_geom, sizeof (struct dk_geom));
3672 	return (rval);
3673 
3674 }
3675 
3676 
3677 /*
3678  *    Function: cmlb_dkio_set_geometry
3679  *
3680  * Description: This routine is the driver entry point for handling user
3681  *		requests to set the device geometry (DKIOCSGEOM). The actual
3682  *		device geometry is not updated, just the driver "notion" of it.
3683  *
3684  *   Arguments:
3685  *	arg		pointer to user provided dk_geom structure used to set
3686  *			the controller's notion of the current geometry.
3687  *
3688  *	flag 		this argument is a pass through to ddi_copyxxx()
3689  *			directly from the mode argument of ioctl().
3690  *
3691  *	tg_cookie	cookie from target driver to be passed back to target
3692  *			driver when we call back to it through tg_ops.
3693  *
3694  * Return Code: 0
3695  *		EFAULT
3696  *		ENXIO
3697  *		EIO
3698  */
3699 static int
3700 cmlb_dkio_set_geometry(struct cmlb_lun *cl, caddr_t arg, int flag)
3701 {
3702 	struct dk_geom	*tmp_geom;
3703 	struct dk_map	*lp;
3704 	int		rval = 0;
3705 	int		i;
3706 
3707 
3708 #if defined(__i386) || defined(__amd64)
3709 	if (cl->cl_solaris_size == 0) {
3710 		return (EIO);
3711 	}
3712 #endif
3713 	/*
3714 	 * We need to copy the user specified geometry into local
3715 	 * storage and then update the softstate. We don't want to hold
3716 	 * the mutex and copyin directly from the user to the soft state
3717 	 */
3718 	tmp_geom = (struct dk_geom *)
3719 	    kmem_zalloc(sizeof (struct dk_geom), KM_SLEEP);
3720 	rval = ddi_copyin(arg, tmp_geom, sizeof (struct dk_geom), flag);
3721 	if (rval != 0) {
3722 		kmem_free(tmp_geom, sizeof (struct dk_geom));
3723 		return (EFAULT);
3724 	}
3725 
3726 	mutex_enter(CMLB_MUTEX(cl));
3727 	bcopy(tmp_geom, &cl->cl_g, sizeof (struct dk_geom));
3728 	for (i = 0; i < NDKMAP; i++) {
3729 		lp  = &cl->cl_map[i];
3730 		cl->cl_offset[i] =
3731 		    cl->cl_g.dkg_nhead * cl->cl_g.dkg_nsect * lp->dkl_cylno;
3732 #if defined(__i386) || defined(__amd64)
3733 		cl->cl_offset[i] += cl->cl_solaris_offset;
3734 #endif
3735 	}
3736 	cl->cl_f_geometry_is_valid = B_FALSE;
3737 	mutex_exit(CMLB_MUTEX(cl));
3738 	kmem_free(tmp_geom, sizeof (struct dk_geom));
3739 
3740 	return (rval);
3741 }
3742 
3743 /*
3744  *    Function: cmlb_dkio_get_partition
3745  *
3746  * Description: This routine is the driver entry point for handling user
3747  *		requests to get the partition table (DKIOCGAPART).
3748  *
3749  *   Arguments:
3750  *	arg		pointer to user provided dk_allmap structure specifying
3751  *			the controller's notion of the current partition table.
3752  *
3753  *	flag		this argument is a pass through to ddi_copyxxx()
3754  *			directly from the mode argument of ioctl().
3755  *
3756  *	tg_cookie	cookie from target driver to be passed back to target
3757  *			driver when we call back to it through tg_ops.
3758  *
3759  * Return Code: 0
3760  *		EFAULT
3761  *		ENXIO
3762  *		EIO
3763  */
3764 static int
3765 cmlb_dkio_get_partition(struct cmlb_lun *cl, caddr_t arg, int flag,
3766     void *tg_cookie)
3767 {
3768 	int		rval = 0;
3769 	int		size;
3770 
3771 	/*
3772 	 * Make sure the geometry is valid before getting the partition
3773 	 * information.
3774 	 */
3775 	mutex_enter(CMLB_MUTEX(cl));
3776 	if ((rval = cmlb_validate_geometry(cl, B_TRUE, 0, tg_cookie)) != 0) {
3777 		mutex_exit(CMLB_MUTEX(cl));
3778 		return (rval);
3779 	}
3780 	mutex_exit(CMLB_MUTEX(cl));
3781 
3782 #if defined(__i386) || defined(__amd64)
3783 	if (cl->cl_solaris_size == 0) {
3784 		return (EIO);
3785 	}
3786 #endif
3787 
3788 #ifdef _MULTI_DATAMODEL
3789 	switch (ddi_model_convert_from(flag & FMODELS)) {
3790 	case DDI_MODEL_ILP32: {
3791 		struct dk_map32 dk_map32[NDKMAP];
3792 		int		i;
3793 
3794 		for (i = 0; i < NDKMAP; i++) {
3795 			dk_map32[i].dkl_cylno = cl->cl_map[i].dkl_cylno;
3796 			dk_map32[i].dkl_nblk  = cl->cl_map[i].dkl_nblk;
3797 		}
3798 		size = NDKMAP * sizeof (struct dk_map32);
3799 		rval = ddi_copyout(dk_map32, (void *)arg, size, flag);
3800 		if (rval != 0) {
3801 			rval = EFAULT;
3802 		}
3803 		break;
3804 	}
3805 	case DDI_MODEL_NONE:
3806 		size = NDKMAP * sizeof (struct dk_map);
3807 		rval = ddi_copyout(cl->cl_map, (void *)arg, size, flag);
3808 		if (rval != 0) {
3809 			rval = EFAULT;
3810 		}
3811 		break;
3812 	}
3813 #else /* ! _MULTI_DATAMODEL */
3814 	size = NDKMAP * sizeof (struct dk_map);
3815 	rval = ddi_copyout(cl->cl_map, (void *)arg, size, flag);
3816 	if (rval != 0) {
3817 		rval = EFAULT;
3818 	}
3819 #endif /* _MULTI_DATAMODEL */
3820 	return (rval);
3821 }
3822 
3823 /*
3824  *    Function: cmlb_dkio_set_partition
3825  *
3826  * Description: This routine is the driver entry point for handling user
3827  *		requests to set the partition table (DKIOCSAPART). The actual
3828  *		device partition is not updated.
3829  *
3830  *   Arguments:
3831  *		arg  - pointer to user provided dk_allmap structure used to set
3832  *			the controller's notion of the partition table.
3833  *		flag - this argument is a pass through to ddi_copyxxx()
3834  *		       directly from the mode argument of ioctl().
3835  *
3836  * Return Code: 0
3837  *		EINVAL
3838  *		EFAULT
3839  *		ENXIO
3840  *		EIO
3841  */
3842 static int
3843 cmlb_dkio_set_partition(struct cmlb_lun *cl, caddr_t arg, int flag)
3844 {
3845 	struct dk_map	dk_map[NDKMAP];
3846 	struct dk_map	*lp;
3847 	int		rval = 0;
3848 	int		size;
3849 	int		i;
3850 #if defined(_SUNOS_VTOC_16)
3851 	struct dkl_partition	*vp;
3852 #endif
3853 
3854 	/*
3855 	 * Set the map for all logical partitions.  We lock
3856 	 * the priority just to make sure an interrupt doesn't
3857 	 * come in while the map is half updated.
3858 	 */
3859 	_NOTE(DATA_READABLE_WITHOUT_LOCK(cmlb_lun::cl_solaris_size))
3860 	mutex_enter(CMLB_MUTEX(cl));
3861 
3862 	if (cl->cl_blockcount > CMLB_OLDVTOC_LIMIT) {
3863 		mutex_exit(CMLB_MUTEX(cl));
3864 		return (ENOTSUP);
3865 	}
3866 	mutex_exit(CMLB_MUTEX(cl));
3867 	if (cl->cl_solaris_size == 0) {
3868 		return (EIO);
3869 	}
3870 
3871 #ifdef _MULTI_DATAMODEL
3872 	switch (ddi_model_convert_from(flag & FMODELS)) {
3873 	case DDI_MODEL_ILP32: {
3874 		struct dk_map32 dk_map32[NDKMAP];
3875 
3876 		size = NDKMAP * sizeof (struct dk_map32);
3877 		rval = ddi_copyin((void *)arg, dk_map32, size, flag);
3878 		if (rval != 0) {
3879 			return (EFAULT);
3880 		}
3881 		for (i = 0; i < NDKMAP; i++) {
3882 			dk_map[i].dkl_cylno = dk_map32[i].dkl_cylno;
3883 			dk_map[i].dkl_nblk  = dk_map32[i].dkl_nblk;
3884 		}
3885 		break;
3886 	}
3887 	case DDI_MODEL_NONE:
3888 		size = NDKMAP * sizeof (struct dk_map);
3889 		rval = ddi_copyin((void *)arg, dk_map, size, flag);
3890 		if (rval != 0) {
3891 			return (EFAULT);
3892 		}
3893 		break;
3894 	}
3895 #else /* ! _MULTI_DATAMODEL */
3896 	size = NDKMAP * sizeof (struct dk_map);
3897 	rval = ddi_copyin((void *)arg, dk_map, size, flag);
3898 	if (rval != 0) {
3899 		return (EFAULT);
3900 	}
3901 #endif /* _MULTI_DATAMODEL */
3902 
3903 	mutex_enter(CMLB_MUTEX(cl));
3904 	/* Note: The size used in this bcopy is set based upon the data model */
3905 	bcopy(dk_map, cl->cl_map, size);
3906 #if defined(_SUNOS_VTOC_16)
3907 	vp = (struct dkl_partition *)&(cl->cl_vtoc);
3908 #endif	/* defined(_SUNOS_VTOC_16) */
3909 	for (i = 0; i < NDKMAP; i++) {
3910 		lp  = &cl->cl_map[i];
3911 		cl->cl_offset[i] =
3912 		    cl->cl_g.dkg_nhead * cl->cl_g.dkg_nsect * lp->dkl_cylno;
3913 #if defined(_SUNOS_VTOC_16)
3914 		vp->p_start = cl->cl_offset[i];
3915 		vp->p_size = lp->dkl_nblk;
3916 		vp++;
3917 #endif	/* defined(_SUNOS_VTOC_16) */
3918 #if defined(__i386) || defined(__amd64)
3919 		cl->cl_offset[i] += cl->cl_solaris_offset;
3920 #endif
3921 	}
3922 	mutex_exit(CMLB_MUTEX(cl));
3923 	return (rval);
3924 }
3925 
3926 
3927 /*
3928  *    Function: cmlb_dkio_get_vtoc
3929  *
3930  * Description: This routine is the driver entry point for handling user
3931  *		requests to get the current volume table of contents
3932  *		(DKIOCGVTOC).
3933  *
3934  *   Arguments:
3935  *	arg		pointer to user provided vtoc structure specifying
3936  *			the current vtoc.
3937  *
3938  *	flag		this argument is a pass through to ddi_copyxxx()
3939  *			directly from the mode argument of ioctl().
3940  *
3941  *	tg_cookie	cookie from target driver to be passed back to target
3942  *			driver when we call back to it through tg_ops.
3943  *
3944  * Return Code: 0
3945  *		EFAULT
3946  *		ENXIO
3947  *		EIO
3948  */
3949 static int
3950 cmlb_dkio_get_vtoc(struct cmlb_lun *cl, caddr_t arg, int flag, void *tg_cookie)
3951 {
3952 #if defined(_SUNOS_VTOC_8)
3953 	struct vtoc	user_vtoc;
3954 #endif	/* defined(_SUNOS_VTOC_8) */
3955 	int		rval = 0;
3956 
3957 	mutex_enter(CMLB_MUTEX(cl));
3958 	if (cl->cl_blockcount > CMLB_OLDVTOC_LIMIT) {
3959 		mutex_exit(CMLB_MUTEX(cl));
3960 		return (EOVERFLOW);
3961 	}
3962 
3963 	rval = cmlb_validate_geometry(cl, B_TRUE, 0, tg_cookie);
3964 
3965 #if defined(_SUNOS_VTOC_8)
3966 	if (rval == EINVAL &&
3967 	    (cl->cl_alter_behavior & CMLB_FAKE_GEOM_LABEL_IOCTLS_VTOC8)) {
3968 		/*
3969 		 * This is to return a default label even when we do not
3970 		 * really assume a default label for the device.
3971 		 * dad driver utilizes this.
3972 		 */
3973 		if (cl->cl_blockcount <= CMLB_OLDVTOC_LIMIT) {
3974 			cmlb_setup_default_geometry(cl, tg_cookie);
3975 			rval = 0;
3976 		}
3977 	}
3978 #endif
3979 	if (rval) {
3980 		mutex_exit(CMLB_MUTEX(cl));
3981 		return (rval);
3982 	}
3983 
3984 #if defined(_SUNOS_VTOC_8)
3985 	cmlb_build_user_vtoc(cl, &user_vtoc);
3986 	mutex_exit(CMLB_MUTEX(cl));
3987 
3988 #ifdef _MULTI_DATAMODEL
3989 	switch (ddi_model_convert_from(flag & FMODELS)) {
3990 	case DDI_MODEL_ILP32: {
3991 		struct vtoc32 user_vtoc32;
3992 
3993 		vtoctovtoc32(user_vtoc, user_vtoc32);
3994 		if (ddi_copyout(&user_vtoc32, (void *)arg,
3995 		    sizeof (struct vtoc32), flag)) {
3996 			return (EFAULT);
3997 		}
3998 		break;
3999 	}
4000 
4001 	case DDI_MODEL_NONE:
4002 		if (ddi_copyout(&user_vtoc, (void *)arg,
4003 		    sizeof (struct vtoc), flag)) {
4004 			return (EFAULT);
4005 		}
4006 		break;
4007 	}
4008 #else /* ! _MULTI_DATAMODEL */
4009 	if (ddi_copyout(&user_vtoc, (void *)arg, sizeof (struct vtoc), flag)) {
4010 		return (EFAULT);
4011 	}
4012 #endif /* _MULTI_DATAMODEL */
4013 
4014 #elif defined(_SUNOS_VTOC_16)
4015 	mutex_exit(CMLB_MUTEX(cl));
4016 
4017 #ifdef _MULTI_DATAMODEL
4018 	/*
4019 	 * The cl_vtoc structure is a "struct dk_vtoc"  which is always
4020 	 * 32-bit to maintain compatibility with existing on-disk
4021 	 * structures.  Thus, we need to convert the structure when copying
4022 	 * it out to a datamodel-dependent "struct vtoc" in a 64-bit
4023 	 * program.  If the target is a 32-bit program, then no conversion
4024 	 * is necessary.
4025 	 */
4026 	/* LINTED: logical expression always true: op "||" */
4027 	ASSERT(sizeof (cl->cl_vtoc) == sizeof (struct vtoc32));
4028 	switch (ddi_model_convert_from(flag & FMODELS)) {
4029 	case DDI_MODEL_ILP32:
4030 		if (ddi_copyout(&(cl->cl_vtoc), (void *)arg,
4031 		    sizeof (cl->cl_vtoc), flag)) {
4032 			return (EFAULT);
4033 		}
4034 		break;
4035 
4036 	case DDI_MODEL_NONE: {
4037 		struct vtoc user_vtoc;
4038 
4039 		vtoc32tovtoc(cl->cl_vtoc, user_vtoc);
4040 		if (ddi_copyout(&user_vtoc, (void *)arg,
4041 		    sizeof (struct vtoc), flag)) {
4042 			return (EFAULT);
4043 		}
4044 		break;
4045 	}
4046 	}
4047 #else /* ! _MULTI_DATAMODEL */
4048 	if (ddi_copyout(&(cl->cl_vtoc), (void *)arg, sizeof (cl->cl_vtoc),
4049 	    flag)) {
4050 		return (EFAULT);
4051 	}
4052 #endif /* _MULTI_DATAMODEL */
4053 #else
4054 #error "No VTOC format defined."
4055 #endif
4056 
4057 	return (rval);
4058 }
4059 
4060 
4061 /*
4062  *    Function: cmlb_dkio_get_extvtoc
4063  */
4064 static int
4065 cmlb_dkio_get_extvtoc(struct cmlb_lun *cl, caddr_t arg, int flag,
4066     void *tg_cookie)
4067 {
4068 	struct extvtoc	ext_vtoc;
4069 #if defined(_SUNOS_VTOC_8)
4070 	struct vtoc	user_vtoc;
4071 #endif	/* defined(_SUNOS_VTOC_8) */
4072 	int		rval = 0;
4073 
4074 	bzero(&ext_vtoc, sizeof (struct extvtoc));
4075 	mutex_enter(CMLB_MUTEX(cl));
4076 	rval = cmlb_validate_geometry(cl, B_TRUE, 0, tg_cookie);
4077 
4078 #if defined(_SUNOS_VTOC_8)
4079 	if (rval == EINVAL &&
4080 	    (cl->cl_alter_behavior & CMLB_FAKE_GEOM_LABEL_IOCTLS_VTOC8)) {
4081 		/*
4082 		 * This is to return a default label even when we do not
4083 		 * really assume a default label for the device.
4084 		 * dad driver utilizes this.
4085 		 */
4086 		if (cl->cl_blockcount <= CMLB_OLDVTOC_LIMIT) {
4087 			cmlb_setup_default_geometry(cl, tg_cookie);
4088 			rval = 0;
4089 		}
4090 	}
4091 #endif
4092 	if (rval) {
4093 		mutex_exit(CMLB_MUTEX(cl));
4094 		return (rval);
4095 	}
4096 
4097 #if defined(_SUNOS_VTOC_8)
4098 	cmlb_build_user_vtoc(cl, &user_vtoc);
4099 	mutex_exit(CMLB_MUTEX(cl));
4100 
4101 	/*
4102 	 * Checking callers data model does not make much sense here
4103 	 * since extvtoc will always be equivalent to 64bit vtoc.
4104 	 * What is important is whether the kernel is in 32 or 64 bit
4105 	 */
4106 
4107 #ifdef _LP64
4108 		if (ddi_copyout(&user_vtoc, (void *)arg,
4109 		    sizeof (struct extvtoc), flag)) {
4110 			return (EFAULT);
4111 		}
4112 #else
4113 		vtoc32tovtoc(user_vtoc, ext_vtoc);
4114 		if (ddi_copyout(&ext_vtoc, (void *)arg,
4115 		    sizeof (struct extvtoc), flag)) {
4116 			return (EFAULT);
4117 		}
4118 #endif
4119 
4120 #elif defined(_SUNOS_VTOC_16)
4121 	/*
4122 	 * The cl_vtoc structure is a "struct dk_vtoc"  which is always
4123 	 * 32-bit to maintain compatibility with existing on-disk
4124 	 * structures.  Thus, we need to convert the structure when copying
4125 	 * it out to extvtoc
4126 	 */
4127 	vtoc32tovtoc(cl->cl_vtoc, ext_vtoc);
4128 	mutex_exit(CMLB_MUTEX(cl));
4129 
4130 	if (ddi_copyout(&ext_vtoc, (void *)arg, sizeof (struct extvtoc), flag))
4131 		return (EFAULT);
4132 #else
4133 #error "No VTOC format defined."
4134 #endif
4135 
4136 	return (rval);
4137 }
4138 
4139 /*
4140  * This routine implements the DKIOCGETEFI ioctl. This ioctl is currently
4141  * used to read the GPT Partition Table Header (primary/backup), the GUID
4142  * partition Entry Array (primary/backup), and the MBR.
4143  */
4144 static int
4145 cmlb_dkio_get_efi(struct cmlb_lun *cl, caddr_t arg, int flag, void *tg_cookie)
4146 {
4147 	dk_efi_t	user_efi;
4148 	int		rval = 0;
4149 	void		*buffer;
4150 	diskaddr_t	tgt_lba;
4151 
4152 	if (ddi_copyin(arg, &user_efi, sizeof (dk_efi_t), flag))
4153 		return (EFAULT);
4154 
4155 	user_efi.dki_data = (void *)(uintptr_t)user_efi.dki_data_64;
4156 
4157 	if (user_efi.dki_length == 0 ||
4158 	    user_efi.dki_length > cmlb_tg_max_efi_xfer)
4159 		return (EINVAL);
4160 
4161 	tgt_lba = user_efi.dki_lba;
4162 
4163 	mutex_enter(CMLB_MUTEX(cl));
4164 	if ((cmlb_check_update_blockcount(cl, tg_cookie) != 0) ||
4165 	    (cl->cl_tgt_blocksize == 0) ||
4166 	    (user_efi.dki_length % cl->cl_sys_blocksize)) {
4167 		mutex_exit(CMLB_MUTEX(cl));
4168 		return (EINVAL);
4169 	}
4170 	if (cl->cl_tgt_blocksize != cl->cl_sys_blocksize)
4171 		tgt_lba = tgt_lba * cl->cl_tgt_blocksize /
4172 		    cl->cl_sys_blocksize;
4173 	mutex_exit(CMLB_MUTEX(cl));
4174 
4175 	buffer = kmem_alloc(user_efi.dki_length, KM_SLEEP);
4176 	rval = DK_TG_READ(cl, buffer, tgt_lba, user_efi.dki_length, tg_cookie);
4177 	if (rval == 0 && ddi_copyout(buffer, user_efi.dki_data,
4178 	    user_efi.dki_length, flag) != 0)
4179 		rval = EFAULT;
4180 
4181 	kmem_free(buffer, user_efi.dki_length);
4182 	return (rval);
4183 }
4184 
4185 #if defined(_SUNOS_VTOC_8)
4186 /*
4187  *    Function: cmlb_build_user_vtoc
4188  *
4189  * Description: This routine populates a pass by reference variable with the
4190  *		current volume table of contents.
4191  *
4192  *   Arguments: cl - driver soft state (unit) structure
4193  *		user_vtoc - pointer to vtoc structure to be populated
4194  */
4195 static void
4196 cmlb_build_user_vtoc(struct cmlb_lun *cl, struct vtoc *user_vtoc)
4197 {
4198 	struct dk_map2		*lpart;
4199 	struct dk_map		*lmap;
4200 	struct partition	*vpart;
4201 	uint32_t		nblks;
4202 	int			i;
4203 
4204 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
4205 
4206 	/*
4207 	 * Return vtoc structure fields in the provided VTOC area, addressed
4208 	 * by *vtoc.
4209 	 */
4210 	bzero(user_vtoc, sizeof (struct vtoc));
4211 	user_vtoc->v_bootinfo[0] = cl->cl_vtoc.v_bootinfo[0];
4212 	user_vtoc->v_bootinfo[1] = cl->cl_vtoc.v_bootinfo[1];
4213 	user_vtoc->v_bootinfo[2] = cl->cl_vtoc.v_bootinfo[2];
4214 	user_vtoc->v_sanity	= VTOC_SANE;
4215 	user_vtoc->v_version	= cl->cl_vtoc.v_version;
4216 	bcopy(cl->cl_vtoc.v_volume, user_vtoc->v_volume, LEN_DKL_VVOL);
4217 	user_vtoc->v_sectorsz = cl->cl_sys_blocksize;
4218 	user_vtoc->v_nparts = cl->cl_vtoc.v_nparts;
4219 
4220 	for (i = 0; i < 10; i++)
4221 		user_vtoc->v_reserved[i] = cl->cl_vtoc.v_reserved[i];
4222 
4223 	/*
4224 	 * Convert partitioning information.
4225 	 *
4226 	 * Note the conversion from starting cylinder number
4227 	 * to starting sector number.
4228 	 */
4229 	lmap = cl->cl_map;
4230 	lpart = (struct dk_map2 *)cl->cl_vtoc.v_part;
4231 	vpart = user_vtoc->v_part;
4232 
4233 	nblks = cl->cl_g.dkg_nsect * cl->cl_g.dkg_nhead;
4234 
4235 	for (i = 0; i < V_NUMPAR; i++) {
4236 		vpart->p_tag	= lpart->p_tag;
4237 		vpart->p_flag	= lpart->p_flag;
4238 		vpart->p_start	= lmap->dkl_cylno * nblks;
4239 		vpart->p_size	= lmap->dkl_nblk;
4240 		lmap++;
4241 		lpart++;
4242 		vpart++;
4243 
4244 		/* (4364927) */
4245 		user_vtoc->timestamp[i] = (time_t)cl->cl_vtoc.v_timestamp[i];
4246 	}
4247 
4248 	bcopy(cl->cl_asciilabel, user_vtoc->v_asciilabel, LEN_DKL_ASCII);
4249 }
4250 #endif
4251 
4252 static int
4253 cmlb_dkio_partition(struct cmlb_lun *cl, caddr_t arg, int flag,
4254     void *tg_cookie)
4255 {
4256 	struct partition64	p64;
4257 	int			rval = 0;
4258 	uint_t			nparts;
4259 	efi_gpe_t		*partitions;
4260 	efi_gpt_t		*buffer;
4261 	diskaddr_t		gpe_lba;
4262 	int			n_gpe_per_blk = 0;
4263 
4264 	if (ddi_copyin((const void *)arg, &p64,
4265 	    sizeof (struct partition64), flag)) {
4266 		return (EFAULT);
4267 	}
4268 
4269 	buffer = kmem_alloc(cl->cl_sys_blocksize, KM_SLEEP);
4270 	rval = DK_TG_READ(cl, buffer, 1, cl->cl_sys_blocksize, tg_cookie);
4271 	if (rval != 0)
4272 		goto done_error;
4273 
4274 	cmlb_swap_efi_gpt(buffer);
4275 
4276 	if ((rval = cmlb_validate_efi(buffer)) != 0)
4277 		goto done_error;
4278 
4279 	nparts = buffer->efi_gpt_NumberOfPartitionEntries;
4280 	gpe_lba = buffer->efi_gpt_PartitionEntryLBA;
4281 	if (p64.p_partno >= nparts) {
4282 		/* couldn't find it */
4283 		rval = ESRCH;
4284 		goto done_error;
4285 	}
4286 	/*
4287 	 * Read the block that contains the requested GPE.
4288 	 */
4289 	n_gpe_per_blk = cl->cl_sys_blocksize / sizeof (efi_gpe_t);
4290 	gpe_lba += p64.p_partno / n_gpe_per_blk;
4291 	rval = DK_TG_READ(cl, buffer, gpe_lba, cl->cl_sys_blocksize, tg_cookie);
4292 
4293 	if (rval) {
4294 		goto done_error;
4295 	}
4296 	partitions = (efi_gpe_t *)buffer;
4297 	partitions += p64.p_partno % n_gpe_per_blk;
4298 
4299 	/* Byte swap only the requested GPE */
4300 	cmlb_swap_efi_gpe(1, partitions);
4301 
4302 	bcopy(&partitions->efi_gpe_PartitionTypeGUID, &p64.p_type,
4303 	    sizeof (struct uuid));
4304 	p64.p_start = partitions->efi_gpe_StartingLBA;
4305 	p64.p_size = partitions->efi_gpe_EndingLBA -
4306 	    p64.p_start + 1;
4307 
4308 	if (ddi_copyout(&p64, (void *)arg, sizeof (struct partition64), flag))
4309 		rval = EFAULT;
4310 
4311 done_error:
4312 	kmem_free(buffer, cl->cl_sys_blocksize);
4313 	return (rval);
4314 }
4315 
4316 
4317 /*
4318  *    Function: cmlb_dkio_set_vtoc
4319  *
4320  * Description: This routine is the driver entry point for handling user
4321  *		requests to set the current volume table of contents
4322  *		(DKIOCSVTOC).
4323  *
4324  *   Arguments:
4325  *	dev		the device number
4326  *	arg		pointer to user provided vtoc structure used to set the
4327  *			current vtoc.
4328  *
4329  *	flag		this argument is a pass through to ddi_copyxxx()
4330  *			directly from the mode argument of ioctl().
4331  *
4332  *	tg_cookie	cookie from target driver to be passed back to target
4333  *			driver when we call back to it through tg_ops.
4334  *
4335  * Return Code: 0
4336  *		EFAULT
4337  *		ENXIO
4338  *		EINVAL
4339  *		ENOTSUP
4340  */
4341 static int
4342 cmlb_dkio_set_vtoc(struct cmlb_lun *cl, dev_t dev, caddr_t arg, int flag,
4343     void *tg_cookie)
4344 {
4345 	struct vtoc	user_vtoc;
4346 	int		shift, rval = 0;
4347 	boolean_t	internal;
4348 
4349 	internal = VOID2BOOLEAN(
4350 	    (cl->cl_alter_behavior & (CMLB_INTERNAL_MINOR_NODES)) != 0);
4351 
4352 	if (cl->cl_alter_behavior & CMLB_CREATE_P0_MINOR_NODE)
4353 		shift = CMLBUNIT_FORCE_P0_SHIFT;
4354 	else
4355 		shift = CMLBUNIT_SHIFT;
4356 
4357 #ifdef _MULTI_DATAMODEL
4358 	switch (ddi_model_convert_from(flag & FMODELS)) {
4359 	case DDI_MODEL_ILP32: {
4360 		struct vtoc32 user_vtoc32;
4361 
4362 		if (ddi_copyin((const void *)arg, &user_vtoc32,
4363 		    sizeof (struct vtoc32), flag)) {
4364 			return (EFAULT);
4365 		}
4366 		vtoc32tovtoc(user_vtoc32, user_vtoc);
4367 		break;
4368 	}
4369 
4370 	case DDI_MODEL_NONE:
4371 		if (ddi_copyin((const void *)arg, &user_vtoc,
4372 		    sizeof (struct vtoc), flag)) {
4373 			return (EFAULT);
4374 		}
4375 		break;
4376 	}
4377 #else /* ! _MULTI_DATAMODEL */
4378 	if (ddi_copyin((const void *)arg, &user_vtoc,
4379 	    sizeof (struct vtoc), flag)) {
4380 		return (EFAULT);
4381 	}
4382 #endif /* _MULTI_DATAMODEL */
4383 
4384 	mutex_enter(CMLB_MUTEX(cl));
4385 
4386 	if (cl->cl_blockcount > CMLB_OLDVTOC_LIMIT) {
4387 		mutex_exit(CMLB_MUTEX(cl));
4388 		return (EOVERFLOW);
4389 	}
4390 
4391 #if defined(__i386) || defined(__amd64)
4392 	if (cl->cl_tgt_blocksize != cl->cl_sys_blocksize) {
4393 		mutex_exit(CMLB_MUTEX(cl));
4394 		return (EINVAL);
4395 	}
4396 #endif
4397 
4398 	if (cl->cl_g.dkg_ncyl == 0) {
4399 		mutex_exit(CMLB_MUTEX(cl));
4400 		return (EINVAL);
4401 	}
4402 
4403 	mutex_exit(CMLB_MUTEX(cl));
4404 	cmlb_clear_efi(cl, tg_cookie);
4405 	ddi_remove_minor_node(CMLB_DEVINFO(cl), "wd");
4406 	ddi_remove_minor_node(CMLB_DEVINFO(cl), "wd,raw");
4407 
4408 	/*
4409 	 * cmlb_dkio_set_vtoc creates duplicate minor nodes when
4410 	 * relabeling an SMI disk. To avoid that we remove them
4411 	 * before creating.
4412 	 * It should be OK to remove a non-existed minor node.
4413 	 */
4414 	ddi_remove_minor_node(CMLB_DEVINFO(cl), "h");
4415 	ddi_remove_minor_node(CMLB_DEVINFO(cl), "h,raw");
4416 
4417 	(void) cmlb_create_minor(CMLB_DEVINFO(cl), "h",
4418 	    S_IFBLK, (CMLBUNIT(dev, shift) << shift) | WD_NODE,
4419 	    cl->cl_node_type, NULL, internal);
4420 	(void) cmlb_create_minor(CMLB_DEVINFO(cl), "h,raw",
4421 	    S_IFCHR, (CMLBUNIT(dev, shift) << shift) | WD_NODE,
4422 	    cl->cl_node_type, NULL, internal);
4423 	mutex_enter(CMLB_MUTEX(cl));
4424 
4425 	if ((rval = cmlb_build_label_vtoc(cl, &user_vtoc)) == 0) {
4426 		if ((rval = cmlb_write_label(cl, tg_cookie)) == 0) {
4427 			if (cmlb_validate_geometry(cl,
4428 			    B_TRUE, 0, tg_cookie) != 0) {
4429 				cmlb_dbg(CMLB_ERROR, cl,
4430 				    "cmlb_dkio_set_vtoc: "
4431 				    "Failed validate geometry\n");
4432 			}
4433 			cl->cl_msglog_flag |= CMLB_ALLOW_2TB_WARN;
4434 		}
4435 	}
4436 	mutex_exit(CMLB_MUTEX(cl));
4437 	return (rval);
4438 }
4439 
4440 /*
4441  *    Function: cmlb_dkio_set_extvtoc
4442  */
4443 static int
4444 cmlb_dkio_set_extvtoc(struct cmlb_lun *cl, dev_t dev, caddr_t arg, int flag,
4445     void *tg_cookie)
4446 {
4447 	int		shift, rval = 0;
4448 	struct vtoc	user_vtoc;
4449 	boolean_t	internal;
4450 
4451 	if (cl->cl_alter_behavior & CMLB_CREATE_P0_MINOR_NODE)
4452 		shift = CMLBUNIT_FORCE_P0_SHIFT;
4453 	else
4454 		shift = CMLBUNIT_SHIFT;
4455 
4456 	/*
4457 	 * Checking callers data model does not make much sense here
4458 	 * since extvtoc will always be equivalent to 64bit vtoc.
4459 	 * What is important is whether the kernel is in 32 or 64 bit
4460 	 */
4461 
4462 #ifdef _LP64
4463 	if (ddi_copyin((const void *)arg, &user_vtoc,
4464 		    sizeof (struct extvtoc), flag)) {
4465 			return (EFAULT);
4466 	}
4467 #else
4468 	struct	extvtoc	user_extvtoc;
4469 	if (ddi_copyin((const void *)arg, &user_extvtoc,
4470 		    sizeof (struct extvtoc), flag)) {
4471 			return (EFAULT);
4472 	}
4473 
4474 	vtoctovtoc32(user_extvtoc, user_vtoc);
4475 #endif
4476 
4477 	internal = VOID2BOOLEAN(
4478 	    (cl->cl_alter_behavior & (CMLB_INTERNAL_MINOR_NODES)) != 0);
4479 	mutex_enter(CMLB_MUTEX(cl));
4480 #if defined(__i386) || defined(__amd64)
4481 	if (cl->cl_tgt_blocksize != cl->cl_sys_blocksize) {
4482 		mutex_exit(CMLB_MUTEX(cl));
4483 		return (EINVAL);
4484 	}
4485 #endif
4486 
4487 	if (cl->cl_g.dkg_ncyl == 0) {
4488 		mutex_exit(CMLB_MUTEX(cl));
4489 		return (EINVAL);
4490 	}
4491 
4492 	mutex_exit(CMLB_MUTEX(cl));
4493 	cmlb_clear_efi(cl, tg_cookie);
4494 	ddi_remove_minor_node(CMLB_DEVINFO(cl), "wd");
4495 	ddi_remove_minor_node(CMLB_DEVINFO(cl), "wd,raw");
4496 	/*
4497 	 * cmlb_dkio_set_extvtoc creates duplicate minor nodes when
4498 	 * relabeling an SMI disk. To avoid that we remove them
4499 	 * before creating.
4500 	 * It should be OK to remove a non-existed minor node.
4501 	 */
4502 	ddi_remove_minor_node(CMLB_DEVINFO(cl), "h");
4503 	ddi_remove_minor_node(CMLB_DEVINFO(cl), "h,raw");
4504 
4505 	(void) cmlb_create_minor(CMLB_DEVINFO(cl), "h",
4506 	    S_IFBLK, (CMLBUNIT(dev, shift) << shift) | WD_NODE,
4507 	    cl->cl_node_type, NULL, internal);
4508 	(void) cmlb_create_minor(CMLB_DEVINFO(cl), "h,raw",
4509 	    S_IFCHR, (CMLBUNIT(dev, shift) << shift) | WD_NODE,
4510 	    cl->cl_node_type, NULL, internal);
4511 
4512 	mutex_enter(CMLB_MUTEX(cl));
4513 
4514 	if ((rval = cmlb_build_label_vtoc(cl, &user_vtoc)) == 0) {
4515 		if ((rval = cmlb_write_label(cl, tg_cookie)) == 0) {
4516 			if (cmlb_validate_geometry(cl,
4517 			    B_TRUE, 0, tg_cookie) != 0) {
4518 				cmlb_dbg(CMLB_ERROR, cl,
4519 				    "cmlb_dkio_set_vtoc: "
4520 				    "Failed validate geometry\n");
4521 			}
4522 		}
4523 	}
4524 	mutex_exit(CMLB_MUTEX(cl));
4525 	return (rval);
4526 }
4527 
4528 /*
4529  *    Function: cmlb_build_label_vtoc
4530  *
4531  * Description: This routine updates the driver soft state current volume table
4532  *		of contents based on a user specified vtoc.
4533  *
4534  *   Arguments: cl - driver soft state (unit) structure
4535  *		user_vtoc - pointer to vtoc structure specifying vtoc to be used
4536  *			    to update the driver soft state.
4537  *
4538  * Return Code: 0
4539  *		EINVAL
4540  */
4541 static int
4542 cmlb_build_label_vtoc(struct cmlb_lun *cl, struct vtoc *user_vtoc)
4543 {
4544 	struct dk_map		*lmap;
4545 	struct partition	*vpart;
4546 	uint_t			nblks;
4547 #if defined(_SUNOS_VTOC_8)
4548 	int			ncyl;
4549 	struct dk_map2		*lpart;
4550 #endif	/* defined(_SUNOS_VTOC_8) */
4551 	int			i;
4552 
4553 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
4554 
4555 	/* Sanity-check the vtoc */
4556 	if (user_vtoc->v_sanity != VTOC_SANE ||
4557 	    user_vtoc->v_sectorsz != cl->cl_sys_blocksize ||
4558 	    user_vtoc->v_nparts != V_NUMPAR) {
4559 		cmlb_dbg(CMLB_INFO,  cl,
4560 		    "cmlb_build_label_vtoc: vtoc not valid\n");
4561 		return (EINVAL);
4562 	}
4563 
4564 	nblks = cl->cl_g.dkg_nsect * cl->cl_g.dkg_nhead;
4565 	if (nblks == 0) {
4566 		cmlb_dbg(CMLB_INFO,  cl,
4567 		    "cmlb_build_label_vtoc: geom nblks is 0\n");
4568 		return (EINVAL);
4569 	}
4570 
4571 #if defined(_SUNOS_VTOC_8)
4572 	vpart = user_vtoc->v_part;
4573 	for (i = 0; i < V_NUMPAR; i++) {
4574 		if (((unsigned)vpart->p_start % nblks) != 0) {
4575 			cmlb_dbg(CMLB_INFO,  cl,
4576 			    "cmlb_build_label_vtoc: p_start not multiply of"
4577 			    "nblks part %d p_start %d nblks %d\n", i,
4578 			    vpart->p_start, nblks);
4579 			return (EINVAL);
4580 		}
4581 		ncyl = (unsigned)vpart->p_start / nblks;
4582 		ncyl += (unsigned)vpart->p_size / nblks;
4583 		if (((unsigned)vpart->p_size % nblks) != 0) {
4584 			ncyl++;
4585 		}
4586 		if (ncyl > (int)cl->cl_g.dkg_ncyl) {
4587 			cmlb_dbg(CMLB_INFO,  cl,
4588 			    "cmlb_build_label_vtoc: ncyl %d  > dkg_ncyl %d"
4589 			    "p_size %ld p_start %ld nblks %d  part number %d"
4590 			    "tag %d\n",
4591 			    ncyl, cl->cl_g.dkg_ncyl, vpart->p_size,
4592 			    vpart->p_start, nblks,
4593 			    i, vpart->p_tag);
4594 
4595 			return (EINVAL);
4596 		}
4597 		vpart++;
4598 	}
4599 #endif	/* defined(_SUNOS_VTOC_8) */
4600 
4601 	/* Put appropriate vtoc structure fields into the disk label */
4602 #if defined(_SUNOS_VTOC_16)
4603 	/*
4604 	 * The vtoc is always a 32bit data structure to maintain the
4605 	 * on-disk format. Convert "in place" instead of doing bcopy.
4606 	 */
4607 	vtoctovtoc32((*user_vtoc), (*((struct vtoc32 *)&(cl->cl_vtoc))));
4608 
4609 	/*
4610 	 * in the 16-slice vtoc, starting sectors are expressed in
4611 	 * numbers *relative* to the start of the Solaris fdisk partition.
4612 	 */
4613 	lmap = cl->cl_map;
4614 	vpart = user_vtoc->v_part;
4615 
4616 	for (i = 0; i < (int)user_vtoc->v_nparts; i++, lmap++, vpart++) {
4617 		lmap->dkl_cylno = (unsigned)vpart->p_start / nblks;
4618 		lmap->dkl_nblk = (unsigned)vpart->p_size;
4619 	}
4620 
4621 #elif defined(_SUNOS_VTOC_8)
4622 
4623 	cl->cl_vtoc.v_bootinfo[0] = (uint32_t)user_vtoc->v_bootinfo[0];
4624 	cl->cl_vtoc.v_bootinfo[1] = (uint32_t)user_vtoc->v_bootinfo[1];
4625 	cl->cl_vtoc.v_bootinfo[2] = (uint32_t)user_vtoc->v_bootinfo[2];
4626 
4627 	cl->cl_vtoc.v_sanity = (uint32_t)user_vtoc->v_sanity;
4628 	cl->cl_vtoc.v_version = (uint32_t)user_vtoc->v_version;
4629 
4630 	bcopy(user_vtoc->v_volume, cl->cl_vtoc.v_volume, LEN_DKL_VVOL);
4631 
4632 	cl->cl_vtoc.v_nparts = user_vtoc->v_nparts;
4633 
4634 	for (i = 0; i < 10; i++)
4635 		cl->cl_vtoc.v_reserved[i] =  user_vtoc->v_reserved[i];
4636 
4637 	/*
4638 	 * Note the conversion from starting sector number
4639 	 * to starting cylinder number.
4640 	 * Return error if division results in a remainder.
4641 	 */
4642 	lmap = cl->cl_map;
4643 	lpart = cl->cl_vtoc.v_part;
4644 	vpart = user_vtoc->v_part;
4645 
4646 	for (i = 0; i < (int)user_vtoc->v_nparts; i++) {
4647 		lpart->p_tag  = vpart->p_tag;
4648 		lpart->p_flag = vpart->p_flag;
4649 		lmap->dkl_cylno = (unsigned)vpart->p_start / nblks;
4650 		lmap->dkl_nblk = (unsigned)vpart->p_size;
4651 
4652 		lmap++;
4653 		lpart++;
4654 		vpart++;
4655 
4656 		/* (4387723) */
4657 #ifdef _LP64
4658 		if (user_vtoc->timestamp[i] > TIME32_MAX) {
4659 			cl->cl_vtoc.v_timestamp[i] = TIME32_MAX;
4660 		} else {
4661 			cl->cl_vtoc.v_timestamp[i] = user_vtoc->timestamp[i];
4662 		}
4663 #else
4664 		cl->cl_vtoc.v_timestamp[i] = user_vtoc->timestamp[i];
4665 #endif
4666 	}
4667 
4668 	bcopy(user_vtoc->v_asciilabel, cl->cl_asciilabel, LEN_DKL_ASCII);
4669 #else
4670 #error "No VTOC format defined."
4671 #endif
4672 	return (0);
4673 }
4674 
4675 /*
4676  *    Function: cmlb_clear_efi
4677  *
4678  * Description: This routine clears all EFI labels.
4679  *
4680  *   Arguments:
4681  *	cl		 driver soft state (unit) structure
4682  *
4683  *	tg_cookie	cookie from target driver to be passed back to target
4684  *			driver when we call back to it through tg_ops.
4685  * Return Code: void
4686  */
4687 static void
4688 cmlb_clear_efi(struct cmlb_lun *cl, void *tg_cookie)
4689 {
4690 	efi_gpt_t	*gpt;
4691 	diskaddr_t	cap;
4692 	int		rval;
4693 
4694 	ASSERT(!mutex_owned(CMLB_MUTEX(cl)));
4695 
4696 	mutex_enter(CMLB_MUTEX(cl));
4697 	cl->cl_reserved = -1;
4698 	mutex_exit(CMLB_MUTEX(cl));
4699 
4700 	gpt = kmem_alloc(cl->cl_sys_blocksize, KM_SLEEP);
4701 
4702 	if (DK_TG_READ(cl, gpt, 1, cl->cl_sys_blocksize, tg_cookie) != 0) {
4703 		goto done;
4704 	}
4705 
4706 	cmlb_swap_efi_gpt(gpt);
4707 	rval = cmlb_validate_efi(gpt);
4708 	if (rval == 0) {
4709 		/* clear primary */
4710 		bzero(gpt, sizeof (efi_gpt_t));
4711 		if (rval = DK_TG_WRITE(cl, gpt, 1, cl->cl_sys_blocksize,
4712 		    tg_cookie)) {
4713 			cmlb_dbg(CMLB_INFO,  cl,
4714 			    "cmlb_clear_efi: clear primary label failed\n");
4715 		}
4716 	}
4717 	/* the backup */
4718 	rval = DK_TG_GETCAP(cl, &cap, tg_cookie);
4719 	if (rval) {
4720 		goto done;
4721 	}
4722 
4723 	if ((rval = DK_TG_READ(cl, gpt, cap - 1, cl->cl_sys_blocksize,
4724 	    tg_cookie)) != 0) {
4725 		goto done;
4726 	}
4727 	cmlb_swap_efi_gpt(gpt);
4728 	rval = cmlb_validate_efi(gpt);
4729 	if (rval == 0) {
4730 		/* clear backup */
4731 		cmlb_dbg(CMLB_TRACE,  cl,
4732 		    "cmlb_clear_efi clear backup@%lu\n", cap - 1);
4733 		bzero(gpt, sizeof (efi_gpt_t));
4734 		if ((rval = DK_TG_WRITE(cl,  gpt, cap - 1, cl->cl_sys_blocksize,
4735 		    tg_cookie))) {
4736 			cmlb_dbg(CMLB_INFO,  cl,
4737 			    "cmlb_clear_efi: clear backup label failed\n");
4738 		}
4739 	} else {
4740 		/*
4741 		 * Refer to comments related to off-by-1 at the
4742 		 * header of this file
4743 		 */
4744 		if ((rval = DK_TG_READ(cl, gpt, cap - 2,
4745 		    cl->cl_sys_blocksize, tg_cookie)) != 0) {
4746 			goto done;
4747 		}
4748 		cmlb_swap_efi_gpt(gpt);
4749 		rval = cmlb_validate_efi(gpt);
4750 		if (rval == 0) {
4751 			/* clear legacy backup EFI label */
4752 			cmlb_dbg(CMLB_TRACE,  cl,
4753 			    "cmlb_clear_efi clear legacy backup@%lu\n",
4754 			    cap - 2);
4755 			bzero(gpt, sizeof (efi_gpt_t));
4756 			if ((rval = DK_TG_WRITE(cl,  gpt, cap - 2,
4757 			    cl->cl_sys_blocksize, tg_cookie))) {
4758 				cmlb_dbg(CMLB_INFO,  cl,
4759 				"cmlb_clear_efi: clear legacy backup label "
4760 				"failed\n");
4761 			}
4762 		}
4763 	}
4764 
4765 done:
4766 	kmem_free(gpt, cl->cl_sys_blocksize);
4767 }
4768 
4769 /*
4770  *    Function: cmlb_set_vtoc
4771  *
4772  * Description: This routine writes data to the appropriate positions
4773  *
4774  *   Arguments:
4775  *	cl		driver soft state (unit) structure
4776  *
4777  *	dkl		the data to be written
4778  *
4779  *	tg_cookie	cookie from target driver to be passed back to target
4780  *			driver when we call back to it through tg_ops.
4781  *
4782  * Return: void
4783  */
4784 static int
4785 cmlb_set_vtoc(struct cmlb_lun *cl, struct dk_label *dkl, void *tg_cookie)
4786 {
4787 	uint_t	label_addr;
4788 	int	sec;
4789 	diskaddr_t	blk;
4790 	int	head;
4791 	int	cyl;
4792 	int	rval;
4793 
4794 #if defined(__i386) || defined(__amd64)
4795 	label_addr = cl->cl_solaris_offset + DK_LABEL_LOC;
4796 #else
4797 	/* Write the primary label at block 0 of the solaris partition. */
4798 	label_addr = 0;
4799 #endif
4800 
4801 	rval = DK_TG_WRITE(cl, dkl, label_addr, cl->cl_sys_blocksize,
4802 	    tg_cookie);
4803 
4804 	if (rval != 0) {
4805 		return (rval);
4806 	}
4807 
4808 	/*
4809 	 * Calculate where the backup labels go.  They are always on
4810 	 * the last alternate cylinder, but some older drives put them
4811 	 * on head 2 instead of the last head.	They are always on the
4812 	 * first 5 odd sectors of the appropriate track.
4813 	 *
4814 	 * We have no choice at this point, but to believe that the
4815 	 * disk label is valid.	 Use the geometry of the disk
4816 	 * as described in the label.
4817 	 */
4818 	cyl  = dkl->dkl_ncyl  + dkl->dkl_acyl - 1;
4819 	head = dkl->dkl_nhead - 1;
4820 
4821 	/*
4822 	 * Write and verify the backup labels. Make sure we don't try to
4823 	 * write past the last cylinder.
4824 	 */
4825 	for (sec = 1; ((sec < 5 * 2 + 1) && (sec < dkl->dkl_nsect)); sec += 2) {
4826 		blk = (diskaddr_t)(
4827 		    (cyl * ((dkl->dkl_nhead * dkl->dkl_nsect) - dkl->dkl_apc)) +
4828 		    (head * dkl->dkl_nsect) + sec);
4829 #if defined(__i386) || defined(__amd64)
4830 		blk += cl->cl_solaris_offset;
4831 #endif
4832 		rval = DK_TG_WRITE(cl, dkl, blk, cl->cl_sys_blocksize,
4833 		    tg_cookie);
4834 		cmlb_dbg(CMLB_INFO,  cl,
4835 		"cmlb_set_vtoc: wrote backup label %llx\n", blk);
4836 		if (rval != 0) {
4837 			goto exit;
4838 		}
4839 	}
4840 exit:
4841 	return (rval);
4842 }
4843 
4844 /*
4845  *    Function: cmlb_clear_vtoc
4846  *
4847  * Description: This routine clears out the VTOC labels.
4848  *
4849  *   Arguments:
4850  *	cl		driver soft state (unit) structure
4851  *
4852  *	tg_cookie	cookie from target driver to be passed back to target
4853  *			driver when we call back to it through tg_ops.
4854  *
4855  * Return: void
4856  */
4857 static void
4858 cmlb_clear_vtoc(struct cmlb_lun *cl, void *tg_cookie)
4859 {
4860 	struct dk_label		*dkl;
4861 
4862 	mutex_exit(CMLB_MUTEX(cl));
4863 	dkl = kmem_zalloc(cl->cl_sys_blocksize, KM_SLEEP);
4864 	mutex_enter(CMLB_MUTEX(cl));
4865 	/*
4866 	 * cmlb_set_vtoc uses these fields in order to figure out
4867 	 * where to overwrite the backup labels
4868 	 */
4869 	dkl->dkl_apc    = cl->cl_g.dkg_apc;
4870 	dkl->dkl_ncyl   = cl->cl_g.dkg_ncyl;
4871 	dkl->dkl_acyl   = cl->cl_g.dkg_acyl;
4872 	dkl->dkl_nhead  = cl->cl_g.dkg_nhead;
4873 	dkl->dkl_nsect  = cl->cl_g.dkg_nsect;
4874 	mutex_exit(CMLB_MUTEX(cl));
4875 	(void) cmlb_set_vtoc(cl, dkl, tg_cookie);
4876 	kmem_free(dkl, cl->cl_sys_blocksize);
4877 
4878 	mutex_enter(CMLB_MUTEX(cl));
4879 }
4880 
4881 /*
4882  *    Function: cmlb_write_label
4883  *
4884  * Description: This routine will validate and write the driver soft state vtoc
4885  *		contents to the device.
4886  *
4887  *   Arguments:
4888  *	cl		cmlb handle
4889  *
4890  *	tg_cookie	cookie from target driver to be passed back to target
4891  *			driver when we call back to it through tg_ops.
4892  *
4893  *
4894  * Return Code: the code returned by cmlb_send_scsi_cmd()
4895  *		0
4896  *		EINVAL
4897  *		ENXIO
4898  *		ENOMEM
4899  */
4900 static int
4901 cmlb_write_label(struct cmlb_lun *cl, void *tg_cookie)
4902 {
4903 	struct dk_label	*dkl;
4904 	short		sum;
4905 	short		*sp;
4906 	int		i;
4907 	int		rval;
4908 
4909 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
4910 	mutex_exit(CMLB_MUTEX(cl));
4911 	dkl = kmem_zalloc(cl->cl_sys_blocksize, KM_SLEEP);
4912 	mutex_enter(CMLB_MUTEX(cl));
4913 
4914 	bcopy(&cl->cl_vtoc, &dkl->dkl_vtoc, sizeof (struct dk_vtoc));
4915 	dkl->dkl_rpm	= cl->cl_g.dkg_rpm;
4916 	dkl->dkl_pcyl	= cl->cl_g.dkg_pcyl;
4917 	dkl->dkl_apc	= cl->cl_g.dkg_apc;
4918 	dkl->dkl_intrlv = cl->cl_g.dkg_intrlv;
4919 	dkl->dkl_ncyl	= cl->cl_g.dkg_ncyl;
4920 	dkl->dkl_acyl	= cl->cl_g.dkg_acyl;
4921 	dkl->dkl_nhead	= cl->cl_g.dkg_nhead;
4922 	dkl->dkl_nsect	= cl->cl_g.dkg_nsect;
4923 
4924 #if defined(_SUNOS_VTOC_8)
4925 	dkl->dkl_obs1	= cl->cl_g.dkg_obs1;
4926 	dkl->dkl_obs2	= cl->cl_g.dkg_obs2;
4927 	dkl->dkl_obs3	= cl->cl_g.dkg_obs3;
4928 	for (i = 0; i < NDKMAP; i++) {
4929 		dkl->dkl_map[i].dkl_cylno = cl->cl_map[i].dkl_cylno;
4930 		dkl->dkl_map[i].dkl_nblk  = cl->cl_map[i].dkl_nblk;
4931 	}
4932 	bcopy(cl->cl_asciilabel, dkl->dkl_asciilabel, LEN_DKL_ASCII);
4933 #elif defined(_SUNOS_VTOC_16)
4934 	dkl->dkl_skew	= cl->cl_dkg_skew;
4935 #else
4936 #error "No VTOC format defined."
4937 #endif
4938 
4939 	dkl->dkl_magic			= DKL_MAGIC;
4940 	dkl->dkl_write_reinstruct	= cl->cl_g.dkg_write_reinstruct;
4941 	dkl->dkl_read_reinstruct	= cl->cl_g.dkg_read_reinstruct;
4942 
4943 	/* Construct checksum for the new disk label */
4944 	sum = 0;
4945 	sp = (short *)dkl;
4946 	i = sizeof (struct dk_label) / sizeof (short);
4947 	while (i--) {
4948 		sum ^= *sp++;
4949 	}
4950 	dkl->dkl_cksum = sum;
4951 
4952 	mutex_exit(CMLB_MUTEX(cl));
4953 
4954 	rval = cmlb_set_vtoc(cl, dkl, tg_cookie);
4955 exit:
4956 	kmem_free(dkl, cl->cl_sys_blocksize);
4957 	mutex_enter(CMLB_MUTEX(cl));
4958 	return (rval);
4959 }
4960 
4961 /*
4962  * This routine implements the DKIOCSETEFI ioctl. This ioctl is currently
4963  * used to write (or clear) the GPT Partition Table header (primary/backup)
4964  * and GUID partition Entry Array (primary/backup). It is also used to write
4965  * the Protective MBR.
4966  */
4967 static int
4968 cmlb_dkio_set_efi(struct cmlb_lun *cl, dev_t dev, caddr_t arg, int flag,
4969     void *tg_cookie)
4970 {
4971 	dk_efi_t	user_efi;
4972 	int		shift, rval = 0;
4973 	void		*buffer;
4974 	diskaddr_t	tgt_lba;
4975 	boolean_t	internal;
4976 
4977 	if (ddi_copyin(arg, &user_efi, sizeof (dk_efi_t), flag))
4978 		return (EFAULT);
4979 
4980 	internal = VOID2BOOLEAN(
4981 	    (cl->cl_alter_behavior & (CMLB_INTERNAL_MINOR_NODES)) != 0);
4982 
4983 	if (cl->cl_alter_behavior & CMLB_CREATE_P0_MINOR_NODE)
4984 		shift = CMLBUNIT_FORCE_P0_SHIFT;
4985 	else
4986 		shift = CMLBUNIT_SHIFT;
4987 
4988 	user_efi.dki_data = (void *)(uintptr_t)user_efi.dki_data_64;
4989 
4990 	if (user_efi.dki_length == 0 ||
4991 	    user_efi.dki_length > cmlb_tg_max_efi_xfer)
4992 		return (EINVAL);
4993 
4994 	tgt_lba = user_efi.dki_lba;
4995 
4996 	mutex_enter(CMLB_MUTEX(cl));
4997 	if ((cmlb_check_update_blockcount(cl, tg_cookie) != 0) ||
4998 	    (cl->cl_tgt_blocksize == 0) ||
4999 	    (user_efi.dki_length % cl->cl_sys_blocksize)) {
5000 		mutex_exit(CMLB_MUTEX(cl));
5001 		return (EINVAL);
5002 	}
5003 	if (cl->cl_tgt_blocksize != cl->cl_sys_blocksize)
5004 		tgt_lba = tgt_lba *
5005 		    cl->cl_tgt_blocksize / cl->cl_sys_blocksize;
5006 	mutex_exit(CMLB_MUTEX(cl));
5007 
5008 	buffer = kmem_alloc(user_efi.dki_length, KM_SLEEP);
5009 	if (ddi_copyin(user_efi.dki_data, buffer, user_efi.dki_length, flag)) {
5010 		rval = EFAULT;
5011 	} else {
5012 		/*
5013 		 * let's clear the vtoc labels and clear the softstate
5014 		 * vtoc.
5015 		 */
5016 		mutex_enter(CMLB_MUTEX(cl));
5017 		if (cl->cl_vtoc.v_sanity == VTOC_SANE) {
5018 			cmlb_dbg(CMLB_TRACE,  cl,
5019 			    "cmlb_dkio_set_efi: CLEAR VTOC\n");
5020 			if (cl->cl_label_from_media == CMLB_LABEL_VTOC)
5021 				cmlb_clear_vtoc(cl, tg_cookie);
5022 			bzero(&cl->cl_vtoc, sizeof (struct dk_vtoc));
5023 			mutex_exit(CMLB_MUTEX(cl));
5024 			ddi_remove_minor_node(CMLB_DEVINFO(cl), "h");
5025 			ddi_remove_minor_node(CMLB_DEVINFO(cl), "h,raw");
5026 			(void) cmlb_create_minor(CMLB_DEVINFO(cl), "wd",
5027 			    S_IFBLK,
5028 			    (CMLBUNIT(dev, shift) << shift) | WD_NODE,
5029 			    cl->cl_node_type, NULL, internal);
5030 			(void) cmlb_create_minor(CMLB_DEVINFO(cl), "wd,raw",
5031 			    S_IFCHR,
5032 			    (CMLBUNIT(dev, shift) << shift) | WD_NODE,
5033 			    cl->cl_node_type, NULL, internal);
5034 		} else
5035 			mutex_exit(CMLB_MUTEX(cl));
5036 
5037 		rval = DK_TG_WRITE(cl, buffer, tgt_lba, user_efi.dki_length,
5038 		    tg_cookie);
5039 
5040 		if (rval == 0) {
5041 			mutex_enter(CMLB_MUTEX(cl));
5042 			cl->cl_f_geometry_is_valid = B_FALSE;
5043 			mutex_exit(CMLB_MUTEX(cl));
5044 		}
5045 	}
5046 	kmem_free(buffer, user_efi.dki_length);
5047 	return (rval);
5048 }
5049 
5050 /*
5051  *    Function: cmlb_dkio_get_mboot
5052  *
5053  * Description: This routine is the driver entry point for handling user
5054  *		requests to get the current device mboot (DKIOCGMBOOT)
5055  *
5056  *   Arguments:
5057  *	arg		pointer to user provided mboot structure specifying
5058  *			the current mboot.
5059  *
5060  *	flag		this argument is a pass through to ddi_copyxxx()
5061  *			directly from the mode argument of ioctl().
5062  *
5063  *	tg_cookie	cookie from target driver to be passed back to target
5064  *			driver when we call back to it through tg_ops.
5065  *
5066  * Return Code: 0
5067  *		EINVAL
5068  *		EFAULT
5069  *		ENXIO
5070  */
5071 static int
5072 cmlb_dkio_get_mboot(struct cmlb_lun *cl, caddr_t arg, int flag, void *tg_cookie)
5073 {
5074 	struct mboot	*mboot;
5075 	int		rval;
5076 	size_t		buffer_size;
5077 
5078 
5079 #if defined(_SUNOS_VTOC_8)
5080 	if ((!ISREMOVABLE(cl) && !ISHOTPLUGGABLE(cl)) || (arg == NULL)) {
5081 #elif defined(_SUNOS_VTOC_16)
5082 	if (arg == NULL) {
5083 #endif
5084 		return (EINVAL);
5085 	}
5086 
5087 	/*
5088 	 * Read the mboot block, located at absolute block 0 on the target.
5089 	 */
5090 	buffer_size = cl->cl_sys_blocksize;
5091 
5092 	cmlb_dbg(CMLB_TRACE,  cl,
5093 	    "cmlb_dkio_get_mboot: allocation size: 0x%x\n", buffer_size);
5094 
5095 	mboot = kmem_zalloc(buffer_size, KM_SLEEP);
5096 	if ((rval = DK_TG_READ(cl, mboot, 0, buffer_size, tg_cookie)) == 0) {
5097 		if (ddi_copyout(mboot, (void *)arg,
5098 		    sizeof (struct mboot), flag) != 0) {
5099 			rval = EFAULT;
5100 		}
5101 	}
5102 	kmem_free(mboot, buffer_size);
5103 	return (rval);
5104 }
5105 
5106 
5107 /*
5108  *    Function: cmlb_dkio_set_mboot
5109  *
5110  * Description: This routine is the driver entry point for handling user
5111  *		requests to validate and set the device master boot
5112  *		(DKIOCSMBOOT).
5113  *
5114  *   Arguments:
5115  *	arg		pointer to user provided mboot structure used to set the
5116  *			master boot.
5117  *
5118  *	flag		this argument is a pass through to ddi_copyxxx()
5119  *			directly from the mode argument of ioctl().
5120  *
5121  *	tg_cookie	cookie from target driver to be passed back to target
5122  *			driver when we call back to it through tg_ops.
5123  *
5124  * Return Code: 0
5125  *		EINVAL
5126  *		EFAULT
5127  *		ENXIO
5128  */
5129 static int
5130 cmlb_dkio_set_mboot(struct cmlb_lun *cl, caddr_t arg, int flag, void *tg_cookie)
5131 {
5132 	struct mboot	*mboot = NULL;
5133 	int		rval;
5134 	ushort_t	magic;
5135 
5136 
5137 	ASSERT(!mutex_owned(CMLB_MUTEX(cl)));
5138 
5139 #if defined(_SUNOS_VTOC_8)
5140 	if (!ISREMOVABLE(cl) && !ISHOTPLUGGABLE(cl)) {
5141 		return (EINVAL);
5142 	}
5143 #endif
5144 
5145 	if (arg == NULL) {
5146 		return (EINVAL);
5147 	}
5148 
5149 	mboot = kmem_zalloc(cl->cl_sys_blocksize, KM_SLEEP);
5150 
5151 	if (ddi_copyin((const void *)arg, mboot,
5152 	    cl->cl_sys_blocksize, flag) != 0) {
5153 		kmem_free(mboot, cl->cl_sys_blocksize);
5154 		return (EFAULT);
5155 	}
5156 
5157 	/* Is this really a master boot record? */
5158 	magic = LE_16(mboot->signature);
5159 	if (magic != MBB_MAGIC) {
5160 		kmem_free(mboot, cl->cl_sys_blocksize);
5161 		return (EINVAL);
5162 	}
5163 
5164 	rval = DK_TG_WRITE(cl, mboot, 0, cl->cl_sys_blocksize, tg_cookie);
5165 
5166 	mutex_enter(CMLB_MUTEX(cl));
5167 #if defined(__i386) || defined(__amd64)
5168 	if (rval == 0) {
5169 		/*
5170 		 * mboot has been written successfully.
5171 		 * update the fdisk and vtoc tables in memory
5172 		 */
5173 		rval = cmlb_update_fdisk_and_vtoc(cl, tg_cookie);
5174 		if ((!cl->cl_f_geometry_is_valid) || (rval != 0)) {
5175 			mutex_exit(CMLB_MUTEX(cl));
5176 			kmem_free(mboot, cl->cl_sys_blocksize);
5177 			return (rval);
5178 		}
5179 	}
5180 
5181 #ifdef __lock_lint
5182 	cmlb_setup_default_geometry(cl, tg_cookie);
5183 #endif
5184 
5185 #else
5186 	if (rval == 0) {
5187 		/*
5188 		 * mboot has been written successfully.
5189 		 * set up the default geometry and VTOC
5190 		 */
5191 		if (cl->cl_blockcount <= CMLB_EXTVTOC_LIMIT)
5192 			cmlb_setup_default_geometry(cl, tg_cookie);
5193 	}
5194 #endif
5195 	cl->cl_msglog_flag |= CMLB_ALLOW_2TB_WARN;
5196 	mutex_exit(CMLB_MUTEX(cl));
5197 	kmem_free(mboot, cl->cl_sys_blocksize);
5198 	return (rval);
5199 }
5200 
5201 
5202 #if defined(__i386) || defined(__amd64)
5203 /*ARGSUSED*/
5204 static int
5205 cmlb_dkio_set_ext_part(struct cmlb_lun *cl, caddr_t arg, int flag,
5206     void *tg_cookie)
5207 {
5208 	int fdisk_rval;
5209 	diskaddr_t capacity;
5210 
5211 	ASSERT(!mutex_owned(CMLB_MUTEX(cl)));
5212 
5213 	mutex_enter(CMLB_MUTEX(cl));
5214 	capacity = cl->cl_blockcount;
5215 	fdisk_rval = cmlb_read_fdisk(cl, capacity, tg_cookie);
5216 	if (fdisk_rval != 0) {
5217 		mutex_exit(CMLB_MUTEX(cl));
5218 		return (fdisk_rval);
5219 	}
5220 
5221 	mutex_exit(CMLB_MUTEX(cl));
5222 	return (fdisk_rval);
5223 }
5224 #endif
5225 
5226 /*
5227  *    Function: cmlb_setup_default_geometry
5228  *
5229  * Description: This local utility routine sets the default geometry as part of
5230  *		setting the device mboot.
5231  *
5232  *   Arguments:
5233  *	cl		driver soft state (unit) structure
5234  *
5235  *	tg_cookie	cookie from target driver to be passed back to target
5236  *			driver when we call back to it through tg_ops.
5237  *
5238  *
5239  * Note: This may be redundant with cmlb_build_default_label.
5240  */
5241 static void
5242 cmlb_setup_default_geometry(struct cmlb_lun *cl, void *tg_cookie)
5243 {
5244 	struct cmlb_geom	pgeom;
5245 	struct cmlb_geom	*pgeomp = &pgeom;
5246 	int			ret;
5247 	int			geom_base_cap = 1;
5248 
5249 
5250 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
5251 
5252 	/* zero out the soft state geometry and partition table. */
5253 	bzero(&cl->cl_g, sizeof (struct dk_geom));
5254 	bzero(&cl->cl_vtoc, sizeof (struct dk_vtoc));
5255 	bzero(cl->cl_map, NDKMAP * (sizeof (struct dk_map)));
5256 
5257 	/*
5258 	 * For the rpm, we use the minimum for the disk.
5259 	 * For the head, cyl and number of sector per track,
5260 	 * if the capacity <= 1GB, head = 64, sect = 32.
5261 	 * else head = 255, sect 63
5262 	 * Note: the capacity should be equal to C*H*S values.
5263 	 * This will cause some truncation of size due to
5264 	 * round off errors. For CD-ROMs, this truncation can
5265 	 * have adverse side effects, so returning ncyl and
5266 	 * nhead as 1. The nsect will overflow for most of
5267 	 * CD-ROMs as nsect is of type ushort.
5268 	 */
5269 	if (cl->cl_alter_behavior & CMLB_FAKE_GEOM_LABEL_IOCTLS_VTOC8) {
5270 		/*
5271 		 * newfs currently can not handle 255 ntracks for SPARC
5272 		 * so get the geometry from target driver instead of coming up
5273 		 * with one based on capacity.
5274 		 */
5275 		mutex_exit(CMLB_MUTEX(cl));
5276 		ret = DK_TG_GETPHYGEOM(cl, pgeomp, tg_cookie);
5277 		mutex_enter(CMLB_MUTEX(cl));
5278 
5279 		if (ret == 0) {
5280 			geom_base_cap = 0;
5281 		} else {
5282 			cmlb_dbg(CMLB_ERROR,  cl,
5283 			    "cmlb_setup_default_geometry: "
5284 			    "tg_getphygeom failed %d\n", ret);
5285 
5286 			/* do default setting, geometry based on capacity */
5287 		}
5288 	}
5289 
5290 	if (geom_base_cap) {
5291 		if (ISCD(cl)) {
5292 			cl->cl_g.dkg_ncyl = 1;
5293 			cl->cl_g.dkg_nhead = 1;
5294 			cl->cl_g.dkg_nsect = cl->cl_blockcount;
5295 		} else if (cl->cl_blockcount < 160) {
5296 			/* Less than 80K */
5297 			cl->cl_g.dkg_nhead = 1;
5298 			cl->cl_g.dkg_ncyl = cl->cl_blockcount;
5299 			cl->cl_g.dkg_nsect = 1;
5300 		} else if (cl->cl_blockcount <= 0x1000) {
5301 			/* Needed for unlabeled SCSI floppies. */
5302 			cl->cl_g.dkg_nhead = 2;
5303 			cl->cl_g.dkg_ncyl = 80;
5304 			cl->cl_g.dkg_pcyl = 80;
5305 			cl->cl_g.dkg_nsect = cl->cl_blockcount / (2 * 80);
5306 		} else if (cl->cl_blockcount <= 0x200000) {
5307 			cl->cl_g.dkg_nhead = 64;
5308 			cl->cl_g.dkg_nsect = 32;
5309 			cl->cl_g.dkg_ncyl = cl->cl_blockcount / (64 * 32);
5310 		} else {
5311 			cl->cl_g.dkg_nhead = 255;
5312 
5313 			cl->cl_g.dkg_nsect = ((cl->cl_blockcount +
5314 			    (UINT16_MAX * 255 * 63) - 1) /
5315 			    (UINT16_MAX * 255 * 63)) * 63;
5316 
5317 			if (cl->cl_g.dkg_nsect == 0)
5318 				cl->cl_g.dkg_nsect = (UINT16_MAX / 63) * 63;
5319 
5320 			cl->cl_g.dkg_ncyl = cl->cl_blockcount /
5321 			    (255 * cl->cl_g.dkg_nsect);
5322 		}
5323 
5324 		cl->cl_g.dkg_acyl = 0;
5325 		cl->cl_g.dkg_bcyl = 0;
5326 		cl->cl_g.dkg_intrlv = 1;
5327 		cl->cl_g.dkg_rpm = 200;
5328 		if (cl->cl_g.dkg_pcyl == 0)
5329 			cl->cl_g.dkg_pcyl = cl->cl_g.dkg_ncyl +
5330 			    cl->cl_g.dkg_acyl;
5331 	} else {
5332 		cl->cl_g.dkg_ncyl = (short)pgeomp->g_ncyl;
5333 		cl->cl_g.dkg_acyl = pgeomp->g_acyl;
5334 		cl->cl_g.dkg_nhead = pgeomp->g_nhead;
5335 		cl->cl_g.dkg_nsect = pgeomp->g_nsect;
5336 		cl->cl_g.dkg_intrlv = pgeomp->g_intrlv;
5337 		cl->cl_g.dkg_rpm = pgeomp->g_rpm;
5338 		cl->cl_g.dkg_pcyl = cl->cl_g.dkg_ncyl + cl->cl_g.dkg_acyl;
5339 	}
5340 
5341 	cl->cl_g.dkg_read_reinstruct = 0;
5342 	cl->cl_g.dkg_write_reinstruct = 0;
5343 	cl->cl_solaris_size = cl->cl_g.dkg_ncyl *
5344 	    cl->cl_g.dkg_nhead * cl->cl_g.dkg_nsect;
5345 
5346 	cl->cl_map['a'-'a'].dkl_cylno = 0;
5347 	cl->cl_map['a'-'a'].dkl_nblk = cl->cl_solaris_size;
5348 
5349 	cl->cl_map['c'-'a'].dkl_cylno = 0;
5350 	cl->cl_map['c'-'a'].dkl_nblk = cl->cl_solaris_size;
5351 
5352 	cl->cl_vtoc.v_part[2].p_tag   = V_BACKUP;
5353 	cl->cl_vtoc.v_part[2].p_flag  = V_UNMNT;
5354 	cl->cl_vtoc.v_nparts = V_NUMPAR;
5355 	cl->cl_vtoc.v_version = V_VERSION;
5356 	(void) sprintf((char *)cl->cl_asciilabel, "DEFAULT cyl %d alt %d"
5357 	    " hd %d sec %d", cl->cl_g.dkg_ncyl, cl->cl_g.dkg_acyl,
5358 	    cl->cl_g.dkg_nhead, cl->cl_g.dkg_nsect);
5359 
5360 	cl->cl_f_geometry_is_valid = B_FALSE;
5361 }
5362 
5363 
5364 #if defined(__i386) || defined(__amd64)
5365 /*
5366  *    Function: cmlb_update_fdisk_and_vtoc
5367  *
5368  * Description: This local utility routine updates the device fdisk and vtoc
5369  *		as part of setting the device mboot.
5370  *
5371  *   Arguments:
5372  *	cl		driver soft state (unit) structure
5373  *
5374  *	tg_cookie	cookie from target driver to be passed back to target
5375  *			driver when we call back to it through tg_ops.
5376  *
5377  *
5378  * Return Code: 0 for success or errno-type return code.
5379  *
5380  *    Note:x86: This looks like a duplicate of cmlb_validate_geometry(), but
5381  *		these did exist separately in x86 sd.c.
5382  */
5383 static int
5384 cmlb_update_fdisk_and_vtoc(struct cmlb_lun *cl, void *tg_cookie)
5385 {
5386 	int		count;
5387 	int		label_rc = 0;
5388 	int		fdisk_rval;
5389 	diskaddr_t	capacity;
5390 
5391 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
5392 
5393 	if (cmlb_check_update_blockcount(cl, tg_cookie) != 0)
5394 		return (EINVAL);
5395 
5396 	/*
5397 	 * Set up the "whole disk" fdisk partition; this should always
5398 	 * exist, regardless of whether the disk contains an fdisk table
5399 	 * or vtoc.
5400 	 */
5401 	cl->cl_map[P0_RAW_DISK].dkl_cylno = 0;
5402 	cl->cl_map[P0_RAW_DISK].dkl_nblk = cl->cl_blockcount;
5403 
5404 	/*
5405 	 * copy the lbasize and capacity so that if they're
5406 	 * reset while we're not holding the CMLB_MUTEX(cl), we will
5407 	 * continue to use valid values after the CMLB_MUTEX(cl) is
5408 	 * reacquired.
5409 	 */
5410 	capacity = cl->cl_blockcount;
5411 
5412 	/*
5413 	 * refresh the logical and physical geometry caches.
5414 	 * (data from mode sense format/rigid disk geometry pages,
5415 	 * and scsi_ifgetcap("geometry").
5416 	 */
5417 	cmlb_resync_geom_caches(cl, capacity, tg_cookie);
5418 
5419 	/*
5420 	 * Only DIRECT ACCESS devices will have Scl labels.
5421 	 * CD's supposedly have a Scl label, too
5422 	 */
5423 	if (cl->cl_device_type == DTYPE_DIRECT || ISREMOVABLE(cl)) {
5424 		fdisk_rval = cmlb_read_fdisk(cl, capacity, tg_cookie);
5425 		if (fdisk_rval != 0) {
5426 			ASSERT(mutex_owned(CMLB_MUTEX(cl)));
5427 			return (fdisk_rval);
5428 		}
5429 
5430 		if (cl->cl_solaris_size <= DK_LABEL_LOC) {
5431 			/*
5432 			 * Found fdisk table but no Solaris partition entry,
5433 			 * so don't call cmlb_uselabel() and don't create
5434 			 * a default label.
5435 			 */
5436 			label_rc = 0;
5437 			cl->cl_f_geometry_is_valid = B_TRUE;
5438 			goto no_solaris_partition;
5439 		}
5440 	} else if (capacity < 0) {
5441 		ASSERT(mutex_owned(CMLB_MUTEX(cl)));
5442 		return (EINVAL);
5443 	}
5444 
5445 	/*
5446 	 * For Removable media We reach here if we have found a
5447 	 * SOLARIS PARTITION.
5448 	 * If cl_f_geometry_is_valid is B_FALSE it indicates that the SOLARIS
5449 	 * PARTITION has changed from the previous one, hence we will setup a
5450 	 * default VTOC in this case.
5451 	 */
5452 	if (!cl->cl_f_geometry_is_valid) {
5453 		/* if we get here it is writable */
5454 		/* we are called from SMBOOT, and after a write of fdisk */
5455 		cmlb_build_default_label(cl, tg_cookie);
5456 		label_rc = 0;
5457 	}
5458 
5459 no_solaris_partition:
5460 
5461 #if defined(_SUNOS_VTOC_16)
5462 	/*
5463 	 * If we have valid geometry, set up the remaining fdisk partitions.
5464 	 * Note that dkl_cylno is not used for the fdisk map entries, so
5465 	 * we set it to an entirely bogus value.
5466 	 */
5467 	for (count = 0; count < FDISK_PARTS; count++) {
5468 		cl->cl_map[FDISK_P1 + count].dkl_cylno = UINT32_MAX;
5469 		cl->cl_map[FDISK_P1 + count].dkl_nblk =
5470 		    cl->cl_fmap[count].fmap_nblk;
5471 		cl->cl_offset[FDISK_P1 + count] =
5472 		    cl->cl_fmap[count].fmap_start;
5473 	}
5474 #endif
5475 
5476 	for (count = 0; count < NDKMAP; count++) {
5477 #if defined(_SUNOS_VTOC_8)
5478 		struct dk_map *lp  = &cl->cl_map[count];
5479 		cl->cl_offset[count] =
5480 		    cl->cl_g.dkg_nhead * cl->cl_g.dkg_nsect * lp->dkl_cylno;
5481 #elif defined(_SUNOS_VTOC_16)
5482 		struct dkl_partition *vp = &cl->cl_vtoc.v_part[count];
5483 		cl->cl_offset[count] = vp->p_start + cl->cl_solaris_offset;
5484 #else
5485 #error "No VTOC format defined."
5486 #endif
5487 	}
5488 
5489 	ASSERT(mutex_owned(CMLB_MUTEX(cl)));
5490 	return (label_rc);
5491 }
5492 #endif
5493 
5494 #if defined(__i386) || defined(__amd64)
5495 static int
5496 cmlb_dkio_get_virtgeom(struct cmlb_lun *cl, caddr_t arg, int flag)
5497 {
5498 	int err = 0;
5499 
5500 	/* Return the driver's notion of the media's logical geometry */
5501 	struct dk_geom	disk_geom;
5502 	struct dk_geom	*dkgp = &disk_geom;
5503 
5504 	mutex_enter(CMLB_MUTEX(cl));
5505 	/*
5506 	 * If there is no HBA geometry available, or
5507 	 * if the HBA returned us something that doesn't
5508 	 * really fit into an Int 13/function 8 geometry
5509 	 * result, just fail the ioctl.  See PSARC 1998/313.
5510 	 */
5511 	if (cl->cl_lgeom.g_nhead == 0 ||
5512 	    cl->cl_lgeom.g_nsect == 0 ||
5513 	    cl->cl_lgeom.g_ncyl > 1024) {
5514 		mutex_exit(CMLB_MUTEX(cl));
5515 		err = EINVAL;
5516 	} else {
5517 		dkgp->dkg_ncyl	= cl->cl_lgeom.g_ncyl;
5518 		dkgp->dkg_acyl	= cl->cl_lgeom.g_acyl;
5519 		dkgp->dkg_pcyl	= dkgp->dkg_ncyl + dkgp->dkg_acyl;
5520 		dkgp->dkg_nhead	= cl->cl_lgeom.g_nhead;
5521 		dkgp->dkg_nsect	= cl->cl_lgeom.g_nsect;
5522 
5523 		mutex_exit(CMLB_MUTEX(cl));
5524 		if (ddi_copyout(dkgp, (void *)arg,
5525 		    sizeof (struct dk_geom), flag)) {
5526 			err = EFAULT;
5527 		} else {
5528 			err = 0;
5529 		}
5530 	}
5531 	return (err);
5532 }
5533 #endif
5534 
5535 #if defined(__i386) || defined(__amd64)
5536 static int
5537 cmlb_dkio_get_phygeom(struct cmlb_lun *cl, caddr_t  arg, int flag,
5538     void *tg_cookie)
5539 {
5540 	int err = 0;
5541 	diskaddr_t capacity;
5542 
5543 
5544 	/* Return the driver's notion of the media physical geometry */
5545 	struct dk_geom	disk_geom;
5546 	struct dk_geom	*dkgp = &disk_geom;
5547 
5548 	mutex_enter(CMLB_MUTEX(cl));
5549 
5550 	if (cl->cl_g.dkg_nhead != 0 &&
5551 	    cl->cl_g.dkg_nsect != 0) {
5552 		/*
5553 		 * We succeeded in getting a geometry, but
5554 		 * right now it is being reported as just the
5555 		 * Solaris fdisk partition, just like for
5556 		 * DKIOCGGEOM. We need to change that to be
5557 		 * correct for the entire disk now.
5558 		 */
5559 		bcopy(&cl->cl_g, dkgp, sizeof (*dkgp));
5560 		dkgp->dkg_acyl = 0;
5561 		dkgp->dkg_ncyl = cl->cl_blockcount /
5562 		    (dkgp->dkg_nhead * dkgp->dkg_nsect);
5563 	} else {
5564 		bzero(dkgp, sizeof (struct dk_geom));
5565 		/*
5566 		 * This disk does not have a Solaris VTOC
5567 		 * so we must present a physical geometry
5568 		 * that will remain consistent regardless
5569 		 * of how the disk is used. This will ensure
5570 		 * that the geometry does not change regardless
5571 		 * of the fdisk partition type (ie. EFI, FAT32,
5572 		 * Solaris, etc).
5573 		 */
5574 		if (ISCD(cl)) {
5575 			dkgp->dkg_nhead = cl->cl_pgeom.g_nhead;
5576 			dkgp->dkg_nsect = cl->cl_pgeom.g_nsect;
5577 			dkgp->dkg_ncyl = cl->cl_pgeom.g_ncyl;
5578 			dkgp->dkg_acyl = cl->cl_pgeom.g_acyl;
5579 		} else {
5580 			/*
5581 			 * Invalid cl_blockcount can generate invalid
5582 			 * dk_geom and may result in division by zero
5583 			 * system failure. Should make sure blockcount
5584 			 * is valid before using it here.
5585 			 */
5586 			if (cl->cl_blockcount == 0) {
5587 				mutex_exit(CMLB_MUTEX(cl));
5588 				err = EIO;
5589 				return (err);
5590 			}
5591 			/*
5592 			 * Refer to comments related to off-by-1 at the
5593 			 * header of this file
5594 			 */
5595 			if (cl->cl_alter_behavior & CMLB_OFF_BY_ONE)
5596 				capacity = cl->cl_blockcount - 1;
5597 			else
5598 				capacity = cl->cl_blockcount;
5599 
5600 			cmlb_convert_geometry(cl, capacity, dkgp, tg_cookie);
5601 			dkgp->dkg_acyl = 0;
5602 			dkgp->dkg_ncyl = capacity /
5603 			    (dkgp->dkg_nhead * dkgp->dkg_nsect);
5604 		}
5605 	}
5606 	dkgp->dkg_pcyl = dkgp->dkg_ncyl + dkgp->dkg_acyl;
5607 
5608 	mutex_exit(CMLB_MUTEX(cl));
5609 	if (ddi_copyout(dkgp, (void *)arg, sizeof (struct dk_geom), flag))
5610 		err = EFAULT;
5611 
5612 	return (err);
5613 }
5614 #endif
5615 
5616 #if defined(__i386) || defined(__amd64)
5617 static int
5618 cmlb_dkio_partinfo(struct cmlb_lun *cl, dev_t dev, caddr_t  arg, int flag)
5619 {
5620 	int err = 0;
5621 
5622 	/*
5623 	 * Return parameters describing the selected disk slice.
5624 	 * Note: this ioctl is for the intel platform only
5625 	 */
5626 	int part;
5627 
5628 	if (cl->cl_alter_behavior & CMLB_CREATE_P0_MINOR_NODE)
5629 		part = getminor(dev) & ((1 << CMLBUNIT_FORCE_P0_SHIFT) - 1);
5630 	else
5631 		part = CMLBPART(dev);
5632 
5633 	mutex_enter(CMLB_MUTEX(cl));
5634 	/* don't check cl_solaris_size for pN */
5635 	if (part < P0_RAW_DISK && cl->cl_solaris_size == 0) {
5636 		err = EIO;
5637 		mutex_exit(CMLB_MUTEX(cl));
5638 	} else {
5639 		struct part_info p;
5640 
5641 		p.p_start = (daddr_t)cl->cl_offset[part];
5642 		p.p_length = (int)cl->cl_map[part].dkl_nblk;
5643 		mutex_exit(CMLB_MUTEX(cl));
5644 #ifdef _MULTI_DATAMODEL
5645 		switch (ddi_model_convert_from(flag & FMODELS)) {
5646 		case DDI_MODEL_ILP32:
5647 		{
5648 			struct part_info32 p32;
5649 
5650 			p32.p_start = (daddr32_t)p.p_start;
5651 			p32.p_length = p.p_length;
5652 			if (ddi_copyout(&p32, (void *)arg,
5653 			    sizeof (p32), flag))
5654 				err = EFAULT;
5655 			break;
5656 		}
5657 
5658 		case DDI_MODEL_NONE:
5659 		{
5660 			if (ddi_copyout(&p, (void *)arg, sizeof (p),
5661 			    flag))
5662 				err = EFAULT;
5663 			break;
5664 		}
5665 		}
5666 #else /* ! _MULTI_DATAMODEL */
5667 		if (ddi_copyout(&p, (void *)arg, sizeof (p), flag))
5668 			err = EFAULT;
5669 #endif /* _MULTI_DATAMODEL */
5670 	}
5671 	return (err);
5672 }
5673 static int
5674 cmlb_dkio_extpartinfo(struct cmlb_lun *cl, dev_t dev, caddr_t  arg, int flag)
5675 {
5676 	int err = 0;
5677 
5678 	/*
5679 	 * Return parameters describing the selected disk slice.
5680 	 * Note: this ioctl is for the intel platform only
5681 	 */
5682 	int part;
5683 
5684 	if (cl->cl_alter_behavior & CMLB_CREATE_P0_MINOR_NODE)
5685 		part = getminor(dev) & ((1 << CMLBUNIT_FORCE_P0_SHIFT) - 1);
5686 	else
5687 		part = CMLBPART(dev);
5688 
5689 	mutex_enter(CMLB_MUTEX(cl));
5690 	/* don't check cl_solaris_size for pN */
5691 	if (part < P0_RAW_DISK && cl->cl_solaris_size == 0) {
5692 		err = EIO;
5693 		mutex_exit(CMLB_MUTEX(cl));
5694 	} else {
5695 		struct extpart_info p;
5696 
5697 		p.p_start = (diskaddr_t)cl->cl_offset[part];
5698 		p.p_length = (diskaddr_t)cl->cl_map[part].dkl_nblk;
5699 		mutex_exit(CMLB_MUTEX(cl));
5700 		if (ddi_copyout(&p, (void *)arg, sizeof (p), flag))
5701 			err = EFAULT;
5702 	}
5703 	return (err);
5704 }
5705 #endif
5706 
5707 int
5708 cmlb_prop_op(cmlb_handle_t cmlbhandle,
5709     dev_t dev, dev_info_t *dip, ddi_prop_op_t prop_op, int mod_flags,
5710     char *name, caddr_t valuep, int *lengthp, int part, void *tg_cookie)
5711 {
5712 	struct cmlb_lun	*cl;
5713 	diskaddr_t	capacity;
5714 	uint32_t	lbasize;
5715 	enum		dp { DP_NBLOCKS, DP_BLKSIZE, DP_SSD, DP_ROT } dp;
5716 	int		callers_length;
5717 	caddr_t		buffer;
5718 	uint64_t	nblocks64;
5719 	uint_t		dblk;
5720 	tg_attribute_t	tgattr;
5721 
5722 	/* Always fallback to ddi_prop_op... */
5723 	cl = (struct cmlb_lun *)cmlbhandle;
5724 	if (cl == NULL) {
5725 fallback:	return (ddi_prop_op(dev, dip, prop_op, mod_flags,
5726 		    name, valuep, lengthp));
5727 	}
5728 
5729 	/* Pick up capacity and blocksize information. */
5730 	capacity = cl->cl_blockcount;
5731 	if (capacity == 0)
5732 		goto fallback;
5733 	lbasize = cl->cl_tgt_blocksize;
5734 	if (lbasize == 0)
5735 		lbasize = DEV_BSIZE;	/* 0 -> DEV_BSIZE units */
5736 
5737 	/* Check for dynamic property of whole device. */
5738 	if (dev == DDI_DEV_T_ANY) {
5739 		/* Fallback to ddi_prop_op if we don't understand.  */
5740 		if (strcmp(name, "device-nblocks") == 0)
5741 			dp = DP_NBLOCKS;
5742 		else if (strcmp(name, "device-blksize") == 0)
5743 			dp = DP_BLKSIZE;
5744 		else if (strcmp(name, "device-solid-state") == 0)
5745 			dp = DP_SSD;
5746 		else if (strcmp(name, "device-rotational") == 0)
5747 			dp = DP_ROT;
5748 		else
5749 			goto fallback;
5750 
5751 		/* get callers length, establish length of our dynamic prop */
5752 		callers_length = *lengthp;
5753 		if (dp == DP_NBLOCKS)
5754 			*lengthp = sizeof (uint64_t);
5755 		else if ((dp == DP_BLKSIZE) || (dp == DP_SSD))
5756 			*lengthp = sizeof (uint32_t);
5757 
5758 		/* service request for the length of the property */
5759 		if (prop_op == PROP_LEN)
5760 			return (DDI_PROP_SUCCESS);
5761 
5762 		switch (prop_op) {
5763 		case PROP_LEN_AND_VAL_ALLOC:
5764 			if ((buffer = kmem_alloc(*lengthp,
5765 			    (mod_flags & DDI_PROP_CANSLEEP) ?
5766 			    KM_SLEEP : KM_NOSLEEP)) == NULL)
5767 				return (DDI_PROP_NO_MEMORY);
5768 			*(caddr_t *)valuep = buffer;	/* set callers buf */
5769 			break;
5770 
5771 		case PROP_LEN_AND_VAL_BUF:
5772 			/* the length of the prop and the request must match */
5773 			if (callers_length != *lengthp)
5774 				return (DDI_PROP_INVAL_ARG);
5775 			buffer = valuep;		/* get callers buf */
5776 			break;
5777 
5778 		default:
5779 			return (DDI_PROP_INVAL_ARG);
5780 		}
5781 
5782 		/* transfer the value into the buffer */
5783 		switch (dp) {
5784 		case DP_NBLOCKS:
5785 			*((uint64_t *)buffer) = capacity;
5786 			break;
5787 		case DP_BLKSIZE:
5788 			*((uint32_t *)buffer) = lbasize;
5789 			break;
5790 		case DP_SSD:
5791 			if (DK_TG_GETATTRIBUTE(cl, &tgattr, tg_cookie) != 0)
5792 				tgattr.media_is_solid_state = B_FALSE;
5793 			*((uint32_t *)buffer) =
5794 			    tgattr.media_is_solid_state ? 1 : 0;
5795 			break;
5796 		case DP_ROT:
5797 			if (DK_TG_GETATTRIBUTE(cl, &tgattr, tg_cookie) != 0)
5798 				tgattr.media_is_rotational = B_TRUE;
5799 			*((uint32_t *)buffer) =
5800 			    tgattr.media_is_rotational ? 1 : 0;
5801 			break;
5802 		}
5803 		return (DDI_PROP_SUCCESS);
5804 	}
5805 
5806 	/*
5807 	 * Support dynamic size oriented properties of partition. Requests
5808 	 * issued under conditions where size is valid are passed to
5809 	 * ddi_prop_op_nblocks with the size information, otherwise the
5810 	 * request is passed to ddi_prop_op. Size depends on valid geometry.
5811 	 */
5812 	if (!cmlb_is_valid(cmlbhandle))
5813 		goto fallback;
5814 
5815 	/* Get partition nblocks value. */
5816 	(void) cmlb_partinfo(cmlbhandle, part,
5817 	    (diskaddr_t *)&nblocks64, NULL, NULL, NULL, tg_cookie);
5818 
5819 	/*
5820 	 * Assume partition information is in sys_blocksize units, compute
5821 	 * divisor for size(9P) property representation.
5822 	 */
5823 	dblk = lbasize / cl->cl_sys_blocksize;
5824 
5825 	/* Now let ddi_prop_op_nblocks_blksize() handle the request. */
5826 	return (ddi_prop_op_nblocks_blksize(dev, dip, prop_op, mod_flags,
5827 	    name, valuep, lengthp, nblocks64 / dblk, lbasize));
5828 }
5829