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