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