xref: /illumos-gate/usr/src/cmd/devfsadm/disk_link.c (revision 406fc5100dac8d225a315a6def6be8d628f34e24)
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  * Copyright 2016 Toomas Soome <tsoome@me.com>
23  * Copyright 2012 Nexenta Systems, Inc.  All rights reserved.
24  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
25  * Use is subject to license terms.
26  */
27 
28 #include <devfsadm.h>
29 #include <stdio.h>
30 #include <strings.h>
31 #include <stdlib.h>
32 #include <limits.h>
33 #include <ctype.h>
34 #include <unistd.h>
35 #include <sys/int_fmtio.h>
36 #include <sys/stat.h>
37 #include <bsm/devalloc.h>
38 #include <sys/scsi/scsi_address.h>
39 #include <sys/libdevid.h>
40 #include <sys/lofi.h>
41 
42 #define	DISK_SUBPATH_MAX 100
43 #define	RM_STALE 0x01
44 #define	DISK_LINK_RE	"^r?dsk/c[0-9]+(t[0-9A-F]+)?d[0-9]+(((s|p))[0-9]+)?$"
45 #define	DISK_LINK_TO_UPPER(ch)\
46 	(((ch) >= 'a' && (ch) <= 'z') ? (ch - 'a' + 'A') : ch)
47 
48 #define	SLICE_SMI	"s7"
49 #define	SLICE_EFI	""
50 
51 #define	MN_SMI		"h"
52 #define	MN_EFI		"wd"
53 #define	ASCIIWWNSIZE	255
54 #if defined(__i386) || defined(__amd64)
55 /*
56  * The number of minor nodes per LUN is defined by the disk drivers.
57  * Currently it is set to 64. Refer CMLBUNIT_SHIFT (cmlb_impl.h)
58  */
59 #define	NUM_MINORS_PER_INSTANCE	64
60 #endif
61 
62 
63 extern int system_labeled;
64 
65 static int disk_callback_chan(di_minor_t minor, di_node_t node);
66 static int disk_callback_nchan(di_minor_t minor, di_node_t node);
67 static int disk_callback_wwn(di_minor_t minor, di_node_t node);
68 static int disk_callback_xvmd(di_minor_t minor, di_node_t node);
69 static int disk_callback_fabric(di_minor_t minor, di_node_t node);
70 static int disk_callback_sas(di_minor_t minor, di_node_t node);
71 static void disk_common(di_minor_t minor, di_node_t node, char *disk,
72 				int flags);
73 static char *diskctrl(di_node_t node, di_minor_t minor);
74 static int reserved_links_exist(di_node_t node, di_minor_t minor, int nflags);
75 static void disk_rm_lofi_all(char *file);
76 
77 
78 static devfsadm_create_t disk_cbt[] = {
79 	{ "disk", DDI_NT_BLOCK, NULL,
80 	    TYPE_EXACT, ILEVEL_0, disk_callback_nchan
81 	},
82 	{ "disk", DDI_NT_BLOCK_CHAN, NULL,
83 	    TYPE_EXACT, ILEVEL_0, disk_callback_chan
84 	},
85 	{ "disk", DDI_NT_BLOCK_FABRIC, NULL,
86 		TYPE_EXACT, ILEVEL_0, disk_callback_fabric
87 	},
88 	{ "disk", DDI_NT_BLOCK_WWN, NULL,
89 	    TYPE_EXACT, ILEVEL_0, disk_callback_wwn
90 	},
91 	{ "disk", DDI_NT_BLOCK_SAS, NULL,
92 	    TYPE_EXACT, ILEVEL_0, disk_callback_sas
93 	},
94 	{ "disk", DDI_NT_CD, NULL,
95 	    TYPE_EXACT, ILEVEL_0, disk_callback_nchan
96 	},
97 	{ "disk", DDI_NT_CD_CHAN, NULL,
98 	    TYPE_EXACT, ILEVEL_0, disk_callback_chan
99 	},
100 	{ "disk", DDI_NT_BLOCK_XVMD, NULL,
101 	    TYPE_EXACT, ILEVEL_0, disk_callback_xvmd
102 	},
103 	{ "disk", DDI_NT_CD_XVMD, NULL,
104 	    TYPE_EXACT, ILEVEL_0, disk_callback_xvmd
105 	},
106 };
107 
108 DEVFSADM_CREATE_INIT_V0(disk_cbt);
109 
110 /*
111  * HOT auto cleanup of disks is done for lofi devices only.
112  */
113 static devfsadm_remove_t disk_remove_cbt[] = {
114 	{ "disk", DISK_LINK_RE, RM_HOT | RM_POST | RM_ALWAYS,
115 		ILEVEL_0, disk_rm_lofi_all
116 	},
117 	{ "disk", DISK_LINK_RE, RM_POST,
118 		ILEVEL_0, devfsadm_rm_all
119 	}
120 };
121 
122 DEVFSADM_REMOVE_INIT_V0(disk_remove_cbt);
123 
124 static devlink_re_t disks_re_array[] = {
125 	{"^r?dsk/c([0-9]+)", 1},
126 	{"^cfg/c([0-9]+)$", 1},
127 	{"^scsi/.+/c([0-9]+)", 1},
128 	{NULL}
129 };
130 
131 static char *disk_mid = "disk_mid";
132 static char *modname = "disk_link";
133 
134 /*
135  * Check if link is from lofi by checking path from readlink().
136  */
137 static int
138 is_lofi_disk(char *file)
139 {
140 	char buf[PATH_MAX + 1];
141 	char filepath[PATH_MAX];
142 	char *ptr;
143 	ssize_t size;
144 
145 	size = snprintf(filepath, sizeof (filepath), "%s/dev/%s",
146 	    devfsadm_root_path(), file);
147 	if (size > sizeof (filepath))
148 		return (0);
149 
150 	size = readlink(filepath, buf, sizeof (buf) - 1);
151 	if (size == -1)
152 		return (0);
153 	buf[size] = '\0';
154 	ptr = strchr(buf, '@');
155 	if (ptr == NULL)
156 		return (0);
157 	ptr[1] = '\0';
158 	if (strcmp(buf, "../../devices/pseudo/lofi@") != 0)
159 		return (0);
160 	return (1);
161 }
162 
163 /*
164  * Wrapper around devfsadm_rm_link() for lofi devices.
165  */
166 static void disk_rm_lofi_all(char *file)
167 {
168 	if (is_lofi_disk(file))
169 		devfsadm_rm_link(file);
170 }
171 
172 int
173 minor_init()
174 {
175 	devfsadm_print(disk_mid,
176 	    "%s: minor_init(): Creating disks reserved ID cache\n",
177 	    modname);
178 	return (devfsadm_reserve_id_cache(disks_re_array, NULL));
179 }
180 
181 static int
182 disk_callback_chan(di_minor_t minor, di_node_t node)
183 {
184 	char *addr;
185 	char disk[23];
186 	char *driver;
187 	uint_t targ = 0;
188 	uint_t lun = 0;
189 
190 	driver = di_driver_name(node);
191 	if (strcmp(driver, LOFI_DRIVER_NAME) != 0) {
192 		addr = di_bus_addr(node);
193 		(void) sscanf(addr, "%X,%X", &targ, &lun);
194 	} else {
195 		targ = di_instance(node);
196 	}
197 
198 	(void) snprintf(disk, sizeof (disk), "t%dd%d", targ, lun);
199 	disk_common(minor, node, disk, 0);
200 	return (DEVFSADM_CONTINUE);
201 
202 }
203 
204 static int
205 disk_callback_nchan(di_minor_t minor, di_node_t node)
206 {
207 	char *addr;
208 	char disk[10];
209 	uint_t lun;
210 
211 	addr = di_bus_addr(node);
212 	(void) sscanf(addr, "%X", &lun);
213 	(void) sprintf(disk, "d%d", lun);
214 	disk_common(minor, node, disk, 0);
215 	return (DEVFSADM_CONTINUE);
216 
217 }
218 
219 static int
220 disk_callback_wwn(di_minor_t minor, di_node_t node)
221 {
222 	char disk[10];
223 	int lun;
224 	int targ;
225 	int *intp;
226 
227 	if (di_prop_lookup_ints(DDI_DEV_T_ANY, node, SCSI_ADDR_PROP_TARGET,
228 	    &intp) <= 0) {
229 		return (DEVFSADM_CONTINUE);
230 	}
231 	targ = *intp;
232 	if (di_prop_lookup_ints(DDI_DEV_T_ANY, node, SCSI_ADDR_PROP_LUN,
233 	    &intp) <= 0) {
234 		lun = 0;
235 	} else {
236 		lun = *intp;
237 	}
238 	(void) sprintf(disk, "t%dd%d", targ, lun);
239 
240 	disk_common(minor, node, disk, RM_STALE);
241 
242 	return (DEVFSADM_CONTINUE);
243 }
244 
245 static int
246 disk_callback_fabric(di_minor_t minor, di_node_t node)
247 {
248 	char disk[DISK_SUBPATH_MAX];
249 	int lun;
250 	int count;
251 	int *intp;
252 	uchar_t *str;
253 	uchar_t *wwn;
254 	uchar_t ascii_wwn[ASCIIWWNSIZE];
255 
256 	if (di_prop_lookup_strings(DDI_DEV_T_ANY, node,
257 	    "client-guid", (char **)&wwn) > 0) {
258 		if (strlcpy((char *)ascii_wwn, (char *)wwn,
259 		    sizeof (ascii_wwn)) >= sizeof (ascii_wwn)) {
260 			devfsadm_errprint("SUNW_disk_link: GUID too long:%d",
261 			    strlen((char *)wwn));
262 			return (DEVFSADM_CONTINUE);
263 		}
264 		lun = 0;
265 	} else if (di_prop_lookup_bytes(DDI_DEV_T_ANY, node,
266 	    "port-wwn", &wwn) > 0) {
267 		if (di_prop_lookup_ints(DDI_DEV_T_ANY, node,
268 		    SCSI_ADDR_PROP_LUN, &intp) > 0) {
269 			lun = *intp;
270 		} else {
271 			lun = 0;
272 		}
273 
274 		for (count = 0, str = ascii_wwn; count < 8; count++, str += 2) {
275 			(void) sprintf((caddr_t)str, "%02x", wwn[count]);
276 		}
277 		*str = '\0';
278 	} else {
279 		return (DEVFSADM_CONTINUE);
280 	}
281 
282 	for (str = ascii_wwn; *str != '\0'; str++) {
283 		*str = DISK_LINK_TO_UPPER(*str);
284 	}
285 
286 	(void) snprintf(disk, DISK_SUBPATH_MAX, "t%sd%d", ascii_wwn, lun);
287 
288 	disk_common(minor, node, disk, RM_STALE);
289 
290 	return (DEVFSADM_CONTINUE);
291 }
292 
293 static int
294 disk_callback_sas(di_minor_t minor, di_node_t node)
295 {
296 	char disk[DISK_SUBPATH_MAX];
297 	int lun64_found = 0;
298 	scsi_lun64_t lun64, sl;
299 	scsi_lun_t lun;
300 	int64_t *lun64p;
301 	uint64_t wwn;
302 	int *intp;
303 	char *tgt_port;
304 	uchar_t addr_method;
305 
306 	/* Get lun property */
307 	if (di_prop_lookup_int64(DDI_DEV_T_ANY, node,
308 	    SCSI_ADDR_PROP_LUN64, &lun64p) > 0) {
309 		if (*lun64p != SCSI_LUN64_ILLEGAL) {
310 			lun64_found = 1;
311 			lun64 = (uint64_t)*lun64p;
312 		}
313 	}
314 	if ((!lun64_found) && (di_prop_lookup_ints(DDI_DEV_T_ANY, node,
315 	    SCSI_ADDR_PROP_LUN, &intp) > 0)) {
316 		lun64 = (uint64_t)*intp;
317 	}
318 
319 	lun = scsi_lun64_to_lun(lun64);
320 
321 	addr_method = (lun.sl_lun1_msb & SCSI_LUN_AM_MASK);
322 
323 	if (di_prop_lookup_strings(DDI_DEV_T_ANY, node,
324 	    SCSI_ADDR_PROP_TARGET_PORT, &tgt_port) > 0) {
325 		(void) scsi_wwnstr_to_wwn(tgt_port, &wwn);
326 		if ((addr_method == SCSI_LUN_AM_PDEV) &&
327 		    (lun.sl_lun2_msb == 0) && (lun.sl_lun2_lsb == 0) &&
328 		    (lun.sl_lun3_msb == 0) && (lun.sl_lun3_lsb == 0) &&
329 		    (lun.sl_lun4_msb == 0) && (lun.sl_lun4_lsb == 0)) {
330 			(void) snprintf(disk, DISK_SUBPATH_MAX,
331 			    "t%"PRIX64"d%"PRId64, wwn, lun64);
332 		} else if ((addr_method == SCSI_LUN_AM_FLAT) &&
333 		    (lun.sl_lun2_msb == 0) && (lun.sl_lun2_lsb == 0) &&
334 		    (lun.sl_lun3_msb == 0) && (lun.sl_lun3_lsb == 0) &&
335 		    (lun.sl_lun4_msb == 0) && (lun.sl_lun4_lsb == 0)) {
336 			sl = (lun.sl_lun1_msb << 8) | lun.sl_lun1_lsb;
337 			(void) snprintf(disk, DISK_SUBPATH_MAX,
338 			    "t%"PRIX64"d%"PRIX16, wwn, sl);
339 		} else {
340 			(void) snprintf(disk, DISK_SUBPATH_MAX,
341 			    "t%"PRIX64"d%"PRIX64, wwn, lun64);
342 		}
343 	} else if (di_prop_lookup_ints(DDI_DEV_T_ANY, node,
344 	    SCSI_ADDR_PROP_SATA_PHY, &intp) > 0) {
345 		/* Use phy format naming, for SATA devices without wwn */
346 		if ((addr_method == SCSI_LUN_AM_PDEV) &&
347 		    (lun.sl_lun2_msb == 0) && (lun.sl_lun2_lsb == 0) &&
348 		    (lun.sl_lun3_msb == 0) && (lun.sl_lun3_lsb == 0) &&
349 		    (lun.sl_lun4_msb == 0) && (lun.sl_lun4_lsb == 0)) {
350 			(void) snprintf(disk, DISK_SUBPATH_MAX,
351 			    "t%dd%"PRId64, *intp, lun64);
352 		} else if ((addr_method == SCSI_LUN_AM_FLAT) &&
353 		    (lun.sl_lun2_msb == 0) && (lun.sl_lun2_lsb == 0) &&
354 		    (lun.sl_lun3_msb == 0) && (lun.sl_lun3_lsb == 0) &&
355 		    (lun.sl_lun4_msb == 0) && (lun.sl_lun4_lsb == 0)) {
356 			sl = (lun.sl_lun1_msb << 8) | lun.sl_lun1_lsb;
357 			(void) snprintf(disk, DISK_SUBPATH_MAX,
358 			    "t%dd%"PRIX16, *intp, sl);
359 		} else {
360 			(void) snprintf(disk, DISK_SUBPATH_MAX,
361 			    "t%dd%"PRIX64, *intp, lun64);
362 		}
363 	} else {
364 		return (DEVFSADM_CONTINUE);
365 	}
366 
367 	disk_common(minor, node, disk, RM_STALE);
368 
369 	return (DEVFSADM_CONTINUE);
370 }
371 
372 /*
373  * xVM virtual block device
374  *
375  * Xen passes device number in next format:
376  *
377  *    1 << 28 | disk << 8 | partition      xvd, disks or partitions 16 onwards
378  *  202 <<  8 | disk << 4 | partition      xvd, disks and partitions up to 15
379  *    8 <<  8 | disk << 4 | partition      sd, disks and partitions up to 15
380  *    3 <<  8 | disk << 6 | partition      hd, disks 0..1, partitions 0..63
381  *   22 <<  8 | (disk-2) << 6 | partition  hd, disks 2..3, partitions 0..63
382  *    2 << 28 onwards                      reserved for future use
383  *   other values less than 1 << 28        deprecated / reserved
384  *
385  * The corresponding Solaris /dev/dsk name can be:
386  *
387  *          c0tYdXsN
388  *
389  * where Y,X >= 0.
390  *
391  * For PV guests using the legacy naming (0, 1, 2, ...)
392  * the Solaris disk names created will be c0d[0..767]sN
393  */
394 
395 #define	HD_BASE		(3 << 8)
396 #define	XEN_EXT_SHIFT	(28)
397 
398 /*
399  * Return: Number of parsed and written parameters
400  */
401 static int
402 decode_xen_device(uint_t device, uint_t *disk, uint_t *plun)
403 {
404 	uint_t dsk, lun = 0;
405 	int ret = 1;
406 
407 	if ((device >> XEN_EXT_SHIFT) > 1)
408 		return (0);
409 
410 	if (device < HD_BASE) {
411 		/* legacy device address */
412 		dsk = device;
413 		goto end;
414 	}
415 
416 	ret = 2;
417 	if (device & (1 << XEN_EXT_SHIFT)) {
418 		/* extended */
419 		dsk = device & (~0xff);
420 		lun = device & 0xff;
421 		goto end;
422 	}
423 
424 	switch (device >> 8) {
425 	case 202:				/* xvd */
426 		dsk = (device >> 4) & 0xf;
427 		lun =  device & 0xf;
428 		break;
429 	case 8:					/* sd */
430 		dsk = device & (~0xf);
431 		lun = device & 0xf;
432 		break;
433 	case 3:					/* hd, disk 0..1 */
434 		dsk = device & (~0x3f);
435 		lun = device & 0x3f;
436 		break;
437 	case 22:				/* hd, disk 2..3 */
438 		dsk = device & (~0x3f);
439 		lun = device & 0x3f;
440 		break;
441 	default:
442 		return (0);
443 	}
444 end:
445 	*disk = dsk;
446 	*plun = lun;
447 	return (ret);
448 }
449 
450 static int
451 disk_callback_xvmd(di_minor_t minor, di_node_t node)
452 {
453 	char *addr;
454 	char disk[16];
455 	uint_t targ;
456 	uint_t dsk, lun;
457 	int res;
458 
459 	addr = di_bus_addr(node);
460 	targ = strtol(addr, (char **)NULL, 10);
461 
462 	res = decode_xen_device(targ, &dsk, &lun);
463 
464 	/* HVM device names are generated using the standard generator */
465 
466 	if (res == 1)
467 		(void) snprintf(disk, sizeof (disk),  "d%d", dsk);
468 	else if (res == 2)
469 		(void) snprintf(disk, sizeof (disk), "t%dd%d", dsk, lun);
470 	else {
471 		devfsadm_errprint("%s: invalid disk device number (%s)\n",
472 		    modname, addr);
473 		return (DEVFSADM_CONTINUE);
474 	}
475 	disk_common(minor, node, disk, 0);
476 	return (DEVFSADM_CONTINUE);
477 
478 }
479 
480 /*
481  * This function is called for every disk minor node.
482  * Calls enumerate to assign a logical controller number, and
483  * then devfsadm_mklink to make the link.
484  */
485 static void
486 disk_common(di_minor_t minor, di_node_t node, char *disk, int flags)
487 {
488 	char l_path[PATH_MAX + 1];
489 	char sec_path[PATH_MAX + 1];
490 	char stale_re[DISK_SUBPATH_MAX];
491 	char *dir;
492 	char slice[4];
493 	char *mn;
494 	char *ctrl;
495 	char *nt = NULL;
496 	int *int_prop;
497 	int  nflags = 0;
498 #if defined(__i386) || defined(__amd64)
499 	char mn_copy[4];
500 	char *part;
501 	int part_num;
502 #endif
503 
504 	mn = di_minor_name(minor);
505 	if (strstr(mn, ",raw")) {
506 		dir = "rdsk";
507 #if defined(__i386) || defined(__amd64)
508 		(void) strncpy(mn_copy, mn, 4);
509 		part = strtok(mn_copy, ",");
510 #endif
511 	} else {
512 		dir = "dsk";
513 #if defined(__i386) || defined(__amd64)
514 		part = mn;
515 #endif
516 	}
517 
518 #if defined(__i386) || defined(__amd64)
519 	/*
520 	 * The following is a table describing the allocation of
521 	 * minor numbers, minor names and /dev/dsk names for partitions
522 	 * and slices on x86 systems.
523 	 *
524 	 *	Minor Number	Minor Name	/dev/dsk name
525 	 *	---------------------------------------------
526 	 *	0 to 15		"a" to "p"	s0 to s15
527 	 *	16		"q"		p0
528 	 *	17 to 20	"r" to "u"	p1 to p4
529 	 *	21 to 52	"p5" to "p36"	p5 to p36
530 	 *
531 	 */
532 	part_num = atoi(part + 1);
533 
534 	if ((mn[0] == 'p') && (part_num >= 5)) {
535 		/* logical drive */
536 		(void) snprintf(slice, 4, "%s", part);
537 	} else {
538 #endif
539 	if (mn[0] < 'q') {
540 		(void) sprintf(slice, "s%d", mn[0] - 'a');
541 	} else if (strncmp(mn, MN_EFI, 2) != 0) {
542 		(void) sprintf(slice, "p%d", mn[0] - 'q');
543 	} else {
544 		/* For EFI label */
545 		(void) sprintf(slice, SLICE_EFI);
546 	}
547 #if defined(__i386) || defined(__amd64)
548 	}
549 #endif
550 
551 	nflags = 0;
552 	if (system_labeled) {
553 		nt = di_minor_nodetype(minor);
554 		if ((nt != NULL) &&
555 		    ((strcmp(nt, DDI_NT_CD) == 0) ||
556 		    (strcmp(nt, DDI_NT_CD_CHAN) == 0) ||
557 		    (strcmp(nt, DDI_NT_BLOCK_CHAN) == 0))) {
558 			nflags = DA_ADD|DA_CD;
559 		}
560 	}
561 
562 	if (reserved_links_exist(node, minor, nflags) == DEVFSADM_SUCCESS) {
563 		devfsadm_print(disk_mid, "Reserved link exists. Not "
564 		    "creating links for slice %s\n", slice);
565 		return;
566 	}
567 
568 	if (NULL == (ctrl = diskctrl(node, minor)))
569 		return;
570 
571 	(void) strcpy(l_path, dir);
572 	(void) strcat(l_path, "/c");
573 	(void) strcat(l_path, ctrl);
574 	(void) strcat(l_path, disk);
575 
576 	/*
577 	 * If switching between SMI and EFI label or vice versa
578 	 * cleanup the previous label's devlinks.
579 	 */
580 	if (*mn == *(MN_SMI) || (strncmp(mn, MN_EFI, 2) == 0)) {
581 		char *s, tpath[PATH_MAX + 1];
582 		struct stat sb;
583 
584 		s = l_path + strlen(l_path);
585 		(void) strcat(l_path, (*mn == *(MN_SMI))
586 		    ? SLICE_EFI : SLICE_SMI);
587 		/*
588 		 * Attempt the remove only if the stale link exists
589 		 */
590 		(void) snprintf(tpath, sizeof (tpath), "%s/dev/%s",
591 		    devfsadm_root_path(), l_path);
592 		if (lstat(tpath, &sb) != -1)
593 			devfsadm_rm_all(l_path);
594 		*s = '\0';
595 	}
596 	(void) strcat(l_path, slice);
597 
598 	(void) devfsadm_mklink(l_path, node, minor, nflags);
599 
600 	/* secondary links for removable and hotpluggable devices */
601 	if (di_prop_lookup_ints(DDI_DEV_T_ANY, node, "removable-media",
602 	    &int_prop) >= 0) {
603 		(void) strcpy(sec_path, "removable-media/");
604 		(void) strcat(sec_path, l_path);
605 		(void) devfsadm_secondary_link(sec_path, l_path, 0);
606 	}
607 	if (di_prop_lookup_ints(DDI_DEV_T_ANY, node, "hotpluggable",
608 	    &int_prop) >= 0) {
609 		(void) strcpy(sec_path, "hotpluggable/");
610 		(void) strcat(sec_path, l_path);
611 		(void) devfsadm_secondary_link(sec_path, l_path, 0);
612 	}
613 
614 	if ((flags & RM_STALE) == RM_STALE) {
615 		(void) strcpy(stale_re, "^");
616 		(void) strcat(stale_re, dir);
617 		(void) strcat(stale_re, "/c");
618 		(void) strcat(stale_re, ctrl);
619 		(void) strcat(stale_re, "t[0-9A-F]+d[0-9]+(s[0-9]+)?$");
620 		/*
621 		 * optimizations are made inside of devfsadm_rm_stale_links
622 		 * instead of before calling the function, as it always
623 		 * needs to add the valid link to the cache.
624 		 */
625 		devfsadm_rm_stale_links(stale_re, l_path, node, minor);
626 	}
627 
628 	free(ctrl);
629 }
630 
631 
632 /* index of enumeration rule applicable to this module */
633 #define	RULE_INDEX	0
634 
635 static char *
636 diskctrl(di_node_t node, di_minor_t minor)
637 {
638 	char path[PATH_MAX + 1];
639 	char *devfspath;
640 	char *buf, *mn;
641 
642 	devfsadm_enumerate_t rules[3] = {
643 	    {"^r?dsk$/^c([0-9]+)", 1, MATCH_PARENT},
644 	    {"^cfg$/^c([0-9]+)$", 1, MATCH_ADDR},
645 	    {"^scsi$/^.+$/^c([0-9]+)", 1, MATCH_PARENT}
646 	};
647 
648 	mn = di_minor_name(minor);
649 
650 	if ((devfspath = di_devfs_path(node)) == NULL) {
651 		return (NULL);
652 	}
653 	(void) strcpy(path, devfspath);
654 	(void) strcat(path, ":");
655 	(void) strcat(path, mn);
656 	di_devfs_path_free(devfspath);
657 
658 	/*
659 	 * Use controller component of disk path
660 	 */
661 	if (disk_enumerate_int(path, RULE_INDEX, &buf, rules, 3) ==
662 	    DEVFSADM_MULTIPLE) {
663 
664 		/*
665 		 * We failed because there are multiple logical controller
666 		 * numbers for a single physical controller.  If we use node
667 		 * name also in the match it should fix this and only find one
668 		 * logical controller. (See 4045879).
669 		 * NOTE: Rules for controllers are not changed, as there is
670 		 * no unique controller number for them in this case.
671 		 *
672 		 * MATCH_UNCACHED flag is private to the "disks" and "sgen"
673 		 * modules. NOT to be used by other modules.
674 		 */
675 
676 		rules[0].flags = MATCH_NODE | MATCH_UNCACHED; /* disks */
677 		rules[2].flags = MATCH_NODE | MATCH_UNCACHED; /* generic scsi */
678 		if (devfsadm_enumerate_int(path, RULE_INDEX, &buf, rules, 3)) {
679 			return (NULL);
680 		}
681 	}
682 
683 	return (buf);
684 }
685 
686 typedef struct dvlist {
687 	char *dv_link;
688 	struct dvlist *dv_next;
689 } dvlist_t;
690 
691 static void
692 free_dvlist(dvlist_t **pp)
693 {
694 	dvlist_t *entry;
695 
696 	while (*pp) {
697 		entry = *pp;
698 		*pp = entry->dv_next;
699 		assert(entry->dv_link);
700 		free(entry->dv_link);
701 		free(entry);
702 	}
703 }
704 static int
705 dvlink_cb(di_devlink_t devlink, void *arg)
706 {
707 	char *path;
708 	char *can_path;
709 	dvlist_t **pp = (dvlist_t **)arg;
710 	dvlist_t *entry = NULL;
711 
712 	entry = calloc(1, sizeof (dvlist_t));
713 	if (entry == NULL) {
714 		devfsadm_errprint("%s: calloc failed\n", modname);
715 		goto error;
716 	}
717 
718 	path = (char *)di_devlink_path(devlink);
719 	assert(path);
720 	if (path == NULL) {
721 		devfsadm_errprint("%s: di_devlink_path() returned NULL\n",
722 		    modname);
723 		goto error;
724 	}
725 
726 	devfsadm_print(disk_mid, "%s: found link %s in reverse link cache\n",
727 	    modname, path);
728 
729 	/*
730 	 * Return linkname in canonical form i.e. without the
731 	 * "/dev/" prefix
732 	 */
733 	can_path = strstr(path, "/dev/");
734 	if (can_path == NULL) {
735 		devfsadm_errprint("%s: devlink path %s has no /dev/\n",
736 		    modname, path);
737 		goto error;
738 	}
739 
740 	entry->dv_link = s_strdup(can_path + strlen("/dev/"));
741 	entry->dv_next = *pp;
742 	*pp = entry;
743 
744 	return (DI_WALK_CONTINUE);
745 
746 error:
747 	free(entry);
748 	free_dvlist(pp);
749 	*pp = NULL;
750 	return (DI_WALK_TERMINATE);
751 }
752 
753 /*
754  * Returns success only if all goes well. If there is no matching reserved link
755  * or if there is an error, we assume no match. It is better to err on the side
756  * of caution by creating extra links than to miss out creating a required link.
757  */
758 static int
759 reserved_links_exist(di_node_t node, di_minor_t minor, int nflags)
760 {
761 	di_devlink_handle_t dvlink_cache = devfsadm_devlink_cache();
762 	char phys_path[PATH_MAX];
763 	char *minor_path;
764 	dvlist_t *head;
765 	dvlist_t *entry;
766 	char *s;
767 	char l[PATH_MAX];
768 	int switch_link = 0;
769 	char *mn = di_minor_name(minor);
770 
771 	if (dvlink_cache == NULL || mn == NULL) {
772 		devfsadm_errprint("%s: No minor or devlink cache\n", modname);
773 		return (DEVFSADM_FAILURE);
774 	}
775 
776 	if (!devfsadm_have_reserved()) {
777 		devfsadm_print(disk_mid, "%s: No reserved links\n", modname);
778 		return (DEVFSADM_FAILURE);
779 	}
780 
781 	minor_path = di_devfs_minor_path(minor);
782 	if (minor_path == NULL) {
783 		devfsadm_errprint("%s: di_devfs_minor_path failed\n", modname);
784 		return (DEVFSADM_FAILURE);
785 	}
786 
787 	(void) strlcpy(phys_path, minor_path, sizeof (phys_path));
788 
789 	di_devfs_path_free(minor_path);
790 
791 	head = NULL;
792 	(void) di_devlink_cache_walk(dvlink_cache, DISK_LINK_RE, phys_path,
793 	    DI_PRIMARY_LINK, &head, dvlink_cb);
794 
795 	/*
796 	 * We may be switching between EFI label and SMI label in which case
797 	 * we only have minors of the other type.
798 	 */
799 	if (head == NULL && (*mn == *(MN_SMI) ||
800 	    (strncmp(mn, MN_EFI, 2) == 0))) {
801 		devfsadm_print(disk_mid, "%s: No links for minor %s in /dev. "
802 		    "Trying another label\n", modname, mn);
803 		s = strrchr(phys_path, ':');
804 		if (s == NULL) {
805 			devfsadm_errprint("%s: invalid minor path: %s\n",
806 			    modname, phys_path);
807 			return (DEVFSADM_FAILURE);
808 		}
809 		(void) snprintf(s+1, sizeof (phys_path) - (s + 1 - phys_path),
810 		    "%s%s", *mn == *(MN_SMI) ? MN_EFI : MN_SMI,
811 		    strstr(s, ",raw") ? ",raw" : "");
812 		(void) di_devlink_cache_walk(dvlink_cache, DISK_LINK_RE,
813 		    phys_path, DI_PRIMARY_LINK, &head, dvlink_cb);
814 	}
815 
816 	if (head == NULL) {
817 		devfsadm_print(disk_mid, "%s: minor %s has no links in /dev\n",
818 		    modname, phys_path);
819 		/* no links on disk */
820 		return (DEVFSADM_FAILURE);
821 	}
822 
823 	/*
824 	 * It suffices to use 1 link to this minor, since
825 	 * we are matching with reserved IDs on the basis of
826 	 * the controller number which will be the same for
827 	 * all links to this minor.
828 	 */
829 	if (!devfsadm_is_reserved(disks_re_array, head->dv_link)) {
830 		/* not reserved links */
831 		devfsadm_print(disk_mid, "%s: devlink %s and its minor "
832 		    "are NOT reserved\n", modname, head->dv_link);
833 		free_dvlist(&head);
834 		return (DEVFSADM_FAILURE);
835 	}
836 
837 	devfsadm_print(disk_mid, "%s: devlink %s and its minor are on "
838 	    "reserved list\n", modname, head->dv_link);
839 
840 	/*
841 	 * Switch between SMI and EFI labels if required
842 	 */
843 	switch_link = 0;
844 	if (*mn == *(MN_SMI) || (strncmp(mn, MN_EFI, 2) == 0)) {
845 		for (entry = head; entry; entry = entry->dv_next) {
846 			s = strrchr(entry->dv_link, '/');
847 			assert(s);
848 			if (s == NULL) {
849 				devfsadm_errprint("%s: disk link %s has no "
850 				    "directory\n", modname, entry->dv_link);
851 				continue;
852 			}
853 			if (*mn == *(MN_SMI) && strchr(s, 's') == NULL) {
854 				(void) snprintf(l, sizeof (l), "%s%s",
855 				    entry->dv_link, SLICE_SMI);
856 				switch_link = 1;
857 				devfsadm_print(disk_mid, "%s: switching "
858 				    "reserved link from EFI to SMI label. "
859 				    "New link is %s\n", modname, l);
860 			} else if (strncmp(mn, MN_EFI, 2) == 0 &&
861 			    (s = strchr(s, 's'))) {
862 				*s = '\0';
863 				(void) snprintf(l, sizeof (l), "%s",
864 				    entry->dv_link);
865 				*s = 's';
866 				switch_link = 1;
867 				devfsadm_print(disk_mid, "%s: switching "
868 				    "reserved link from SMI to EFI label. "
869 				    "New link is %s\n", modname, l);
870 			}
871 			if (switch_link) {
872 				devfsadm_print(disk_mid, "%s: switching "
873 				    "link: deleting %s and creating %s\n",
874 				    modname, entry->dv_link, l);
875 				devfsadm_rm_link(entry->dv_link);
876 				(void) devfsadm_mklink(l, node, minor, nflags);
877 			}
878 		}
879 	}
880 	free_dvlist(&head);
881 
882 	/*
883 	 * return SUCCESS to indicate that new links to this minor should not
884 	 * be created so that only compatibility links to this minor remain.
885 	 */
886 	return (DEVFSADM_SUCCESS);
887 }
888