xref: /titanic_50/usr/src/lib/libdevinfo/devinfo_devlink.c (revision 49f9b365248ee858ee91baa36eab27c5200f6dca)
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 2009 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #include "libdevinfo.h"
27 #include "devinfo_devlink.h"
28 #include "device_info.h"
29 
30 #undef	DEBUG
31 #ifndef	DEBUG
32 #define	NDEBUG 1
33 #else
34 #undef	NDEBUG
35 #endif
36 
37 #include <assert.h>
38 
39 static mutex_t update_mutex = DEFAULTMUTEX; /* Protects update record lock */
40 static mutex_t temp_file_mutex = DEFAULTMUTEX; /* for file creation tests */
41 
42 static const size_t elem_sizes[DB_TYPES] = {
43 	sizeof (struct db_node),
44 	sizeof (struct db_minor),
45 	sizeof (struct db_link),
46 	sizeof (char)
47 };
48 
49 /*
50  * List of directories/files skipped while physically walking /dev
51  * Paths are relative to "<root>/dev/"
52  */
53 static const char *skip_dirs[] = {"fd"};
54 static const char *skip_files[] = {
55 	"stdout",
56 	"stdin",
57 	"stderr"
58 };
59 
60 #define	N_SKIP_DIRS	(sizeof (skip_dirs) / sizeof (skip_dirs[0]))
61 #define	N_SKIP_FILES	(sizeof (skip_files) / sizeof (skip_files[0]))
62 
63 #define	DI_TEST_DB	ETCDEV "di_test_db"
64 
65 /*
66  *
67  * This file contains two sets of interfaces which operate on the reverse
68  * links database. One set (which includes di_devlink_open()/_close())
69  * allows link generators like devfsadm(1M) and ucblinks(1B) (writers) to
70  * populate the database with /devices -> /dev mappings. Another set
71  * of interfaces (which includes di_devlink_init()/_fini()) allows
72  * applications (readers) to lookup the database for /dev links corresponding
73  * to a given minor.
74  *
75  * Writers operate on a cached version of the database. The cache is created
76  * when di_devlink_open() is called. As links in /dev are created and removed,
77  * the cache is updated to keep it in synch with /dev. When the /dev updates
78  * are complete, the link generator calls di_devlink_close() which writes
79  * out the cache to the database.
80  *
81  * Applications which need to lookup the database, call di_devlink_init().
82  * di_devlink_init() checks the database file (if one exists). If the
83  * database is valid, it is mapped into the address space of the
84  * application. The database file consists of several segments. Each
85  * segment can be mapped in independently and is mapped on demand.
86  *
87  *		   Database Layout
88  *
89  *		---------------------
90  *		|	Magic #     |
91  *		| ----------------- |
92  *		|       Version	    |	HEADER
93  *		| ----------------- |
94  *		|        ...        |
95  *		---------------------
96  *		|		    |
97  *		|		    |	NODES
98  *		|	            |
99  *		|		    |
100  *		---------------------
101  *		|		    |
102  *		|		    |	MINORS
103  *		|	            |
104  *		|		    |
105  *		---------------------
106  *		|		    |
107  *		|		    |   LINKS
108  *		|	            |
109  *		|		    |
110  *		---------------------
111  *		|		    |
112  *		|		    |	STRINGS
113  *		|	            |
114  *		|		    |
115  *		---------------------
116  *
117  * Readers can lookup /dev links for a specific minor or
118  * lookup all /dev links. In the latter case, the node
119  * and minor segments are not mapped in and the reader
120  * walks through every link in the link segment.
121  *
122  */
123 di_devlink_handle_t
124 di_devlink_open(const char *root_dir, uint_t flags)
125 {
126 	int err;
127 	char path[PATH_MAX];
128 	struct di_devlink_handle *hdp;
129 	int retried = 0;
130 
131 retry:
132 	/*
133 	 * Allocate a read-write handle but open the DB in readonly
134 	 * mode. We do writes only to a temporary copy of the database.
135 	 */
136 	if ((hdp = handle_alloc(root_dir, OPEN_RDWR)) == NULL) {
137 		return (NULL);
138 	}
139 
140 	err = open_db(hdp, OPEN_RDONLY);
141 
142 	/*
143 	 * We don't want to unlink the db at this point - if we did we
144 	 * would be creating a window where consumers would take a slow
145 	 * code path (and those consumers might also trigger requests for
146 	 * db creation, which we are already in the process of doing).
147 	 * When we are done with our update, we use rename to install the
148 	 * latest version of the db file.
149 	 */
150 	get_db_path(hdp, DB_FILE, path, sizeof (path));
151 
152 	/*
153 	 * The flags argument is reserved for future use.
154 	 */
155 	if (flags != 0) {
156 		handle_free(&hdp); /* also closes the DB */
157 		errno = EINVAL;
158 		return (NULL);
159 	}
160 
161 	if (cache_alloc(hdp) != 0) {
162 		handle_free(&hdp);
163 		return (NULL);
164 	}
165 
166 	if (err) {
167 		/*
168 		 * Failed to open DB.
169 		 * The most likely cause is that DB file did not exist.
170 		 * Call di_devlink_close() to recreate the DB file and
171 		 * retry di_devlink_open().
172 		 */
173 		if (retried == 0) {
174 			(void) di_devlink_close(&hdp, 0);
175 			retried = 1;
176 			goto retry;
177 		}
178 
179 		/*
180 		 * DB cannot be opened, just return the
181 		 * handle. We will recreate the DB later.
182 		 */
183 		return (hdp);
184 	}
185 
186 	/* Read the database into the cache */
187 	CACHE(hdp)->update_count = DB_HDR(hdp)->update_count;
188 	(void) read_nodes(hdp, NULL, DB_HDR(hdp)->root_idx);
189 	(void) read_links(hdp, NULL, DB_HDR(hdp)->dngl_idx);
190 
191 	(void) close_db(hdp);
192 
193 	return (hdp);
194 }
195 
196 static void
197 get_db_path(
198 	struct di_devlink_handle *hdp,
199 	const char *fname,
200 	char *buf,
201 	size_t blen)
202 {
203 	char *dir = NULL;
204 
205 #ifdef	DEBUG
206 	if (dir = getenv(ALT_DB_DIR)) {
207 		(void) dprintf(DBG_INFO, "get_db_path: alternate db dir: %s\n",
208 		    dir);
209 	}
210 #endif
211 	if (dir == NULL) {
212 		dir = hdp->db_dir;
213 	}
214 
215 	(void) snprintf(buf, blen, "%s/%s", dir, fname);
216 }
217 
218 static int
219 open_db(struct di_devlink_handle *hdp, int flags)
220 {
221 	size_t sz;
222 	long page_sz;
223 	int fd, rv, flg;
224 	struct stat sbuf;
225 	uint32_t count[DB_TYPES] = {0};
226 	char path[PATH_MAX];
227 	void *cp;
228 
229 	assert(!DB_OPEN(hdp));
230 
231 #ifdef	DEBUG
232 	if (getenv(SKIP_DB)) {
233 		(void) dprintf(DBG_INFO, "open_db: skipping database\n");
234 		return (-1);
235 	}
236 #endif
237 	if ((page_sz = sysconf(_SC_PAGE_SIZE)) == -1) {
238 		return (-1);
239 	}
240 
241 	/*
242 	 * Use O_TRUNC flag for write access, so that the subsequent ftruncate()
243 	 * call will zero-fill the entire file
244 	 */
245 	if (IS_RDONLY(flags)) {
246 		flg = O_RDONLY;
247 		get_db_path(hdp, DB_FILE, path, sizeof (path));
248 	} else {
249 		flg = O_RDWR|O_CREAT|O_TRUNC;
250 		get_db_path(hdp, DB_TMP, path, sizeof (path));
251 	}
252 
253 	/*
254 	 * Avoid triggering /dev reconfigure for read when not present
255 	 */
256 	if (IS_RDONLY(flags) &&
257 	    (strncmp(path, "/dev/", 5) == 0) && !device_exists(path)) {
258 		return (-1);
259 	}
260 
261 	if ((fd = open(path, flg, DB_PERMS)) == -1) {
262 		return (-1);
263 	}
264 
265 	if (IS_RDONLY(flags)) {
266 		flg = PROT_READ;
267 		rv = fstat(fd, &sbuf);
268 		sz = sbuf.st_size;
269 	} else {
270 		flg = PROT_READ | PROT_WRITE;
271 		sz = size_db(hdp, page_sz, count);
272 		rv = ftruncate(fd, sz);
273 	}
274 
275 	if (rv == -1 || sz < HDR_LEN) {
276 		if (rv != -1)
277 			errno = EINVAL;
278 		(void) close(fd);
279 		return (-1);
280 	}
281 
282 	cp = mmap(0, HDR_LEN, flg, MAP_SHARED, fd, 0);
283 	if (cp == MAP_FAILED) {
284 		(void) close(fd);
285 		return (-1);
286 	}
287 	DB(hdp)->hdr = (struct db_hdr *)cp;
288 	DB(hdp)->db_fd = fd;
289 	DB(hdp)->flags = flags;
290 
291 	if (IS_RDONLY(flags)) {
292 		rv = invalid_db(hdp, sz, page_sz);
293 	} else {
294 		rv = init_hdr(hdp, page_sz, count);
295 	}
296 
297 	if (rv) {
298 		(void) dprintf(DBG_ERR, "open_db: invalid DB(%s)\n", path);
299 		(void) close_db(hdp);
300 		return (-1);
301 	} else {
302 		(void) dprintf(DBG_STEP, "open_db: DB(%s): opened\n", path);
303 		return (0);
304 	}
305 }
306 
307 /*
308  * A handle can be allocated for read-only or read-write access
309  */
310 static struct di_devlink_handle *
311 handle_alloc(const char *root_dir, uint_t flags)
312 {
313 	char dev_dir[PATH_MAX], path[PATH_MAX], db_dir[PATH_MAX];
314 	struct di_devlink_handle *hdp, proto = {0};
315 	int install = 0;
316 	int isroot = 0;
317 	struct stat sb;
318 	char can_path[PATH_MAX];
319 
320 	assert(flags == OPEN_RDWR || flags == OPEN_RDONLY);
321 
322 	dev_dir[0] = '\0';
323 	db_dir[0] = '\0';
324 
325 	/*
326 	 * NULL and the empty string are equivalent to "/"
327 	 */
328 	if (root_dir && root_dir[0] != '\0') {
329 
330 		if (root_dir[0] != '/') {
331 			errno = EINVAL;
332 			return (NULL);
333 		}
334 
335 #ifdef	DEBUG
336 		/*LINTED*/
337 		assert(sizeof (dev_dir) >= PATH_MAX);
338 #endif
339 		if ((realpath(root_dir, dev_dir) == NULL) ||
340 		    (realpath(root_dir, db_dir) == NULL)) {
341 			return (NULL);
342 		}
343 	} else {
344 		/*
345 		 * The dev dir is at /dev i.e. we are not doing a -r /altroot
346 		 */
347 		isroot = 1;
348 	}
349 
350 	if (strcmp(dev_dir, "/") == 0) {
351 		dev_dir[0] = 0;
352 		db_dir[0] = 0;
353 	} else {
354 		(void) strlcpy(db_dir, dev_dir, sizeof (db_dir));
355 	}
356 
357 	(void) strlcat(dev_dir, DEV, sizeof (dev_dir));
358 	(void) strlcat(db_dir, ETCDEV, sizeof (db_dir));
359 
360 	/*
361 	 * The following code is for install. Readers and writers need
362 	 * to be redirected to /tmp/etc/dev for the database file.
363 	 * Note that we test for readonly /etc by actually creating a
364 	 * file since statvfs is not a reliable method for determining
365 	 * readonly filesystems.
366 	 */
367 	install = 0;
368 	(void) snprintf(can_path, sizeof (can_path), "%s/%s", ETCDEV, DB_FILE);
369 	if (flags == OPEN_RDWR && isroot) {
370 		char di_test_db[PATH_MAX];
371 		int fd;
372 		(void) mutex_lock(&temp_file_mutex);
373 		(void) snprintf(di_test_db, sizeof (di_test_db), "%s.%d",
374 		    DI_TEST_DB, getpid());
375 		fd = open(di_test_db, O_CREAT|O_RDWR|O_EXCL, 0644);
376 		if (fd == -1 && errno == EROFS && stat(can_path, &sb) == -1)
377 			install = 1;
378 		if (fd != -1) {
379 			(void) close(fd);
380 			(void) unlink(di_test_db);
381 		}
382 		(void) mutex_unlock(&temp_file_mutex);
383 	} else if (isroot) {
384 		/*
385 		 * Readers can be non-privileged so we cannot test by creating
386 		 * a file in /etc/dev. Instead we check if the database
387 		 * file is missing in /etc/dev and is present in /tmp/etc/dev
388 		 * and is owned by root.
389 		 */
390 		char install_path[PATH_MAX];
391 
392 		(void) snprintf(install_path, sizeof (install_path),
393 		    "/tmp%s/%s", ETCDEV, DB_FILE);
394 		if (stat(can_path, &sb) == -1 && stat(install_path, &sb)
395 		    != -1 && sb.st_uid == 0) {
396 			install = 1;
397 		}
398 	}
399 
400 	/*
401 	 * Check if we are in install. If we are, the database will be in
402 	 * /tmp/etc/dev
403 	 */
404 	if (install)
405 		(void) snprintf(db_dir, sizeof (db_dir), "/tmp%s", ETCDEV);
406 
407 	proto.dev_dir = dev_dir;
408 	proto.db_dir = db_dir;
409 	proto.flags = flags;
410 	proto.lock_fd = -1;
411 
412 	/*
413 	 * Lock database if a read-write handle is being allocated.
414 	 * Locks are needed to protect against multiple writers.
415 	 * Readers don't need locks.
416 	 */
417 	if (HDL_RDWR(&proto)) {
418 		if (enter_db_lock(&proto, root_dir) != 1) {
419 			return (NULL);
420 		}
421 	}
422 
423 	DB(&proto)->db_fd = -1;
424 
425 	hdp = calloc(1, sizeof (struct di_devlink_handle));
426 	if (hdp == NULL) {
427 		goto error;
428 	}
429 
430 	*hdp = proto;
431 
432 	/*
433 	 * The handle hdp now contains a pointer to local storage
434 	 * in the dev_dir field (obtained from the proto handle).
435 	 * In the following line, a dynamically allocated version
436 	 * is substituted.
437 	 */
438 
439 	if ((hdp->dev_dir = strdup(proto.dev_dir)) == NULL) {
440 		free(hdp);
441 		goto error;
442 	}
443 
444 	if ((hdp->db_dir = strdup(proto.db_dir)) == NULL) {
445 		free(hdp->dev_dir);
446 		free(hdp);
447 		goto error;
448 	}
449 
450 	return (hdp);
451 
452 error:
453 	if (HDL_RDWR(&proto)) {
454 		/* Unlink DB file on error */
455 		get_db_path(&proto, DB_FILE, path, sizeof (path));
456 		(void) unlink(path);
457 		exit_db_lock(&proto);
458 	}
459 	return (NULL);
460 }
461 
462 
463 static int
464 cache_alloc(struct di_devlink_handle *hdp)
465 {
466 	size_t hash_sz = 0;
467 
468 	assert(HDL_RDWR(hdp));
469 
470 	if (DB_OPEN(hdp)) {
471 		hash_sz = DB_NUM(hdp, DB_LINK) / AVG_CHAIN_SIZE;
472 	}
473 	hash_sz = (hash_sz >= MIN_HASH_SIZE) ? hash_sz : MIN_HASH_SIZE;
474 
475 	CACHE(hdp)->hash = calloc(hash_sz, sizeof (cache_link_t *));
476 	if (CACHE(hdp)->hash == NULL) {
477 		return (-1);
478 	}
479 	CACHE(hdp)->hash_sz = hash_sz;
480 
481 	return (0);
482 }
483 
484 
485 static int
486 invalid_db(struct di_devlink_handle *hdp, size_t fsize, long page_sz)
487 {
488 	int i;
489 	char *cp;
490 	size_t sz;
491 
492 	if (DB_HDR(hdp)->magic != DB_MAGIC || DB_HDR(hdp)->vers != DB_VERSION) {
493 		return (1);
494 	}
495 
496 	if (DB_HDR(hdp)->page_sz == 0 || DB_HDR(hdp)->page_sz != page_sz) {
497 		return (1);
498 	}
499 
500 	sz = seg_size(hdp, DB_HEADER);
501 	for (i = 0; i < DB_TYPES; i++) {
502 		(void) dprintf(DBG_INFO, "N[%u] = %u\n", i, DB_NUM(hdp, i));
503 		/* There must be at least 1 element of each type */
504 		if (DB_NUM(hdp, i) < 1) {
505 			return (1);
506 		}
507 		sz += seg_size(hdp, i);
508 		assert(sz % page_sz == 0);
509 	}
510 
511 	if (sz != fsize) {
512 		return (1);
513 	}
514 
515 	if (!VALID_INDEX(hdp, DB_NODE, DB_HDR(hdp)->root_idx)) {
516 		return (1);
517 	}
518 
519 	if (!VALID_INDEX(hdp, DB_LINK, DB_HDR(hdp)->dngl_idx)) {
520 		return (1);
521 	}
522 
523 	if (DB_EMPTY(hdp)) {
524 		return (1);
525 	}
526 
527 	/*
528 	 * The last character in the string segment must be a NUL char.
529 	 */
530 	cp = get_string(hdp, DB_NUM(hdp, DB_STR) - 1);
531 	if (cp == NULL || *cp != '\0') {
532 		return (1);
533 	}
534 
535 	return (0);
536 }
537 
538 static int
539 read_nodes(struct di_devlink_handle *hdp, cache_node_t *pcnp, uint32_t nidx)
540 {
541 	char *path;
542 	cache_node_t *cnp;
543 	struct db_node *dnp;
544 	const char *fcn = "read_nodes";
545 
546 	assert(HDL_RDWR(hdp));
547 
548 	/*
549 	 * parent node should be NULL only for the root node
550 	 */
551 	if ((pcnp == NULL) ^ (nidx == DB_HDR(hdp)->root_idx)) {
552 		(void) dprintf(DBG_ERR, "%s: invalid parent or index(%u)\n",
553 		    fcn, nidx);
554 		SET_DB_ERR(hdp);
555 		return (-1);
556 	}
557 
558 	for (; dnp = get_node(hdp, nidx); nidx = dnp->sib) {
559 
560 		path = get_string(hdp, dnp->path);
561 
562 		/*
563 		 * Insert at head of list to recreate original order
564 		 */
565 		cnp = node_insert(hdp, pcnp, path, INSERT_HEAD);
566 		if (cnp == NULL) {
567 			SET_DB_ERR(hdp);
568 			break;
569 		}
570 
571 		assert(strcmp(path, "/") ^ (nidx == DB_HDR(hdp)->root_idx));
572 		assert(strcmp(path, "/") != 0 || dnp->sib == DB_NIL);
573 
574 		if (read_minors(hdp, cnp, dnp->minor) != 0 ||
575 		    read_nodes(hdp, cnp, dnp->child) != 0) {
576 			break;
577 		}
578 
579 		(void) dprintf(DBG_STEP, "%s: node[%u]: %s\n", fcn, nidx,
580 		    cnp->path);
581 	}
582 
583 	return (dnp ? -1 : 0);
584 }
585 
586 static int
587 read_minors(struct di_devlink_handle *hdp, cache_node_t *pcnp, uint32_t nidx)
588 {
589 	cache_minor_t *cmnp;
590 	struct db_minor *dmp;
591 	char *name, *nodetype;
592 	const char *fcn = "read_minors";
593 
594 	assert(HDL_RDWR(hdp));
595 
596 	if (pcnp == NULL) {
597 		(void) dprintf(DBG_ERR, "%s: minor[%u]: orphan minor\n", fcn,
598 		    nidx);
599 		SET_DB_ERR(hdp);
600 		return (-1);
601 	}
602 
603 	for (; dmp = get_minor(hdp, nidx); nidx = dmp->sib) {
604 
605 		name = get_string(hdp, dmp->name);
606 		nodetype = get_string(hdp, dmp->nodetype);
607 
608 		cmnp = minor_insert(hdp, pcnp, name, nodetype, NULL);
609 		if (cmnp == NULL) {
610 			SET_DB_ERR(hdp);
611 			break;
612 		}
613 
614 		(void) dprintf(DBG_STEP, "%s: minor[%u]: %s\n", fcn, nidx,
615 		    cmnp->name);
616 
617 		if (read_links(hdp, cmnp, dmp->link) != 0) {
618 			break;
619 		}
620 	}
621 
622 	return (dmp ? -1 : 0);
623 }
624 
625 /*
626  * If the link is dangling the corresponding minor will be absent.
627  */
628 static int
629 read_links(struct di_devlink_handle *hdp, cache_minor_t *pcmp, uint32_t nidx)
630 {
631 	cache_link_t *clp;
632 	struct db_link *dlp;
633 	char *path, *content;
634 
635 	assert(HDL_RDWR(hdp));
636 
637 	if (nidx != DB_NIL &&
638 	    ((pcmp == NULL) ^ (nidx == DB_HDR(hdp)->dngl_idx))) {
639 		(void) dprintf(DBG_ERR, "read_links: invalid minor or"
640 		    " index(%u)\n", nidx);
641 		SET_DB_ERR(hdp);
642 		return (-1);
643 	}
644 
645 	for (; dlp = get_link(hdp, nidx); nidx = dlp->sib) {
646 
647 		path = get_string(hdp, dlp->path);
648 		content = get_string(hdp, dlp->content);
649 
650 		clp = link_insert(hdp, pcmp, path, content, dlp->attr);
651 		if (clp == NULL) {
652 			SET_DB_ERR(hdp);
653 			break;
654 		}
655 
656 		(void) dprintf(DBG_STEP, "read_links: link[%u]: %s%s\n",
657 		    nidx, clp->path, pcmp == NULL ? "(DANGLING)" : "");
658 	}
659 
660 	return (dlp ? -1 : 0);
661 }
662 
663 int
664 di_devlink_close(di_devlink_handle_t *pp, int flag)
665 {
666 	int i, rv;
667 	char tmp[PATH_MAX];
668 	char file[PATH_MAX];
669 	uint32_t next[DB_TYPES] = {0};
670 	struct di_devlink_handle *hdp;
671 
672 	if (pp == NULL || *pp == NULL || !HDL_RDWR(*pp)) {
673 		errno = EINVAL;
674 		return (-1);
675 	}
676 
677 	hdp = *pp;
678 	*pp = NULL;
679 
680 	/*
681 	 * The caller encountered some error in their processing.
682 	 * so handle isn't valid. Discard it and return success.
683 	 */
684 	if (flag == DI_LINK_ERROR) {
685 		handle_free(&hdp);
686 		return (0);
687 	}
688 
689 	if (DB_ERR(hdp)) {
690 		handle_free(&hdp);
691 		errno = EINVAL;
692 		return (-1);
693 	}
694 
695 	/*
696 	 * Extract the DB path before the handle is freed.
697 	 */
698 	get_db_path(hdp, DB_FILE, file, sizeof (file));
699 	get_db_path(hdp, DB_TMP, tmp, sizeof (tmp));
700 
701 	/*
702 	 * update database with actual contents of /dev
703 	 */
704 	(void) dprintf(DBG_INFO, "di_devlink_close: update_count = %u\n",
705 	    CACHE(hdp)->update_count);
706 
707 	/*
708 	 * For performance reasons, synchronization of the database
709 	 * with /dev is turned off by default. However, applications
710 	 * with appropriate permissions can request a "sync" by
711 	 * calling di_devlink_update().
712 	 */
713 	if (CACHE(hdp)->update_count == 0) {
714 		CACHE(hdp)->update_count = 1;
715 		(void) dprintf(DBG_INFO,
716 		    "di_devlink_close: synchronizing DB\n");
717 		(void) synchronize_db(hdp);
718 	}
719 
720 	/*
721 	 * Resolve dangling links AFTER synchronizing DB with /dev as the
722 	 * synchronization process may create dangling links.
723 	 */
724 	resolve_dangling_links(hdp);
725 
726 	/*
727 	 * All changes to the cache are complete. Write out the cache
728 	 * to the database only if it is not empty.
729 	 */
730 	if (CACHE_EMPTY(hdp)) {
731 		(void) dprintf(DBG_INFO, "di_devlink_close: skipping write\n");
732 		(void) unlink(file);
733 		handle_free(&hdp);
734 		return (0);
735 	}
736 
737 	if (open_db(hdp, OPEN_RDWR) != 0) {
738 		handle_free(&hdp);
739 		return (-1);
740 	}
741 
742 	/*
743 	 * Keep track of array assignments. There is at least
744 	 * 1 element (the "NIL" element) per type.
745 	 */
746 	for (i = 0; i < DB_TYPES; i++) {
747 		next[i] = 1;
748 	}
749 
750 	(void) write_nodes(hdp, NULL, CACHE_ROOT(hdp), next);
751 	(void) write_links(hdp, NULL, CACHE(hdp)->dngl, next);
752 	DB_HDR(hdp)->update_count = CACHE(hdp)->update_count;
753 
754 	rv = close_db(hdp);
755 
756 	if (rv != 0 || DB_ERR(hdp) || rename(tmp, file) != 0) {
757 		(void) dprintf(DBG_ERR, "di_devlink_close: %s error: %s\n",
758 		    rv ? "close_db" : "DB or rename", strerror(errno));
759 		(void) unlink(tmp);
760 		(void) unlink(file);
761 		handle_free(&hdp);
762 		return (-1);
763 	}
764 
765 	handle_free(&hdp);
766 
767 	(void) dprintf(DBG_INFO, "di_devlink_close: wrote DB(%s)\n", file);
768 
769 	return (0);
770 }
771 
772 /*
773  * Inits the database header.
774  */
775 static int
776 init_hdr(struct di_devlink_handle *hdp, long page_sz, uint32_t *count)
777 {
778 	int i;
779 
780 	DB_HDR(hdp)->magic = DB_MAGIC;
781 	DB_HDR(hdp)->vers = DB_VERSION;
782 	DB_HDR(hdp)->root_idx = DB_NIL;
783 	DB_HDR(hdp)->dngl_idx = DB_NIL;
784 	DB_HDR(hdp)->page_sz = (uint32_t)page_sz;
785 
786 	for (i = 0; i < DB_TYPES; i++) {
787 		assert(count[i] >= 1);
788 		DB_NUM(hdp, i) = count[i];
789 	}
790 
791 	return (0);
792 }
793 
794 static int
795 write_nodes(
796 	struct di_devlink_handle *hdp,
797 	struct db_node *pdnp,
798 	cache_node_t *cnp,
799 	uint32_t *next)
800 {
801 	uint32_t idx;
802 	struct db_node *dnp;
803 	const char *fcn = "write_nodes";
804 
805 	assert(HDL_RDWR(hdp));
806 
807 	for (; cnp != NULL; cnp = cnp->sib) {
808 
809 		assert(cnp->path != NULL);
810 
811 		/* parent node should only be NULL for root node */
812 		if ((pdnp == NULL) ^ (cnp == CACHE_ROOT(hdp))) {
813 			(void) dprintf(DBG_ERR, "%s: invalid parent for: %s\n",
814 			    fcn, cnp->path);
815 			SET_DB_ERR(hdp);
816 			break;
817 		}
818 
819 		assert((strcmp(cnp->path, "/") != 0) ^
820 		    (cnp == CACHE_ROOT(hdp)));
821 
822 		idx = next[DB_NODE];
823 		if ((dnp = set_node(hdp, idx)) == NULL) {
824 			SET_DB_ERR(hdp);
825 			break;
826 		}
827 
828 		dnp->path = write_string(hdp, cnp->path, next);
829 		if (dnp->path == DB_NIL) {
830 			SET_DB_ERR(hdp);
831 			break;
832 		}
833 		/* commit write for this node */
834 		next[DB_NODE]++;
835 
836 		if (pdnp == NULL) {
837 			assert(DB_HDR(hdp)->root_idx == DB_NIL);
838 			DB_HDR(hdp)->root_idx = idx;
839 		} else {
840 			dnp->sib = pdnp->child;
841 			pdnp->child = idx;
842 		}
843 
844 		(void) dprintf(DBG_STEP, "%s: node[%u]: %s\n", fcn, idx,
845 		    cnp->path);
846 
847 		if (write_minors(hdp, dnp, cnp->minor, next) != 0 ||
848 		    write_nodes(hdp, dnp, cnp->child, next) != 0) {
849 			break;
850 		}
851 	}
852 
853 	return (cnp ? -1 : 0);
854 }
855 
856 static int
857 write_minors(
858 	struct di_devlink_handle *hdp,
859 	struct db_node *pdnp,
860 	cache_minor_t *cmnp,
861 	uint32_t *next)
862 {
863 	uint32_t idx;
864 	struct db_minor *dmp;
865 	const char *fcn = "write_minors";
866 
867 	assert(HDL_RDWR(hdp));
868 
869 	if (pdnp == NULL) {
870 		(void) dprintf(DBG_ERR, "%s: no node for minor: %s\n", fcn,
871 		    cmnp ? cmnp->name : "<NULL>");
872 		SET_DB_ERR(hdp);
873 		return (-1);
874 	}
875 
876 	for (; cmnp != NULL; cmnp = cmnp->sib) {
877 
878 		assert(cmnp->name != NULL);
879 
880 		idx = next[DB_MINOR];
881 		if ((dmp = set_minor(hdp, idx)) == NULL) {
882 			SET_DB_ERR(hdp);
883 			break;
884 		}
885 
886 		dmp->name = write_string(hdp, cmnp->name, next);
887 		dmp->nodetype = write_string(hdp, cmnp->nodetype, next);
888 		if (dmp->name == DB_NIL || dmp->nodetype == DB_NIL) {
889 			dmp->name = dmp->nodetype = DB_NIL;
890 			SET_DB_ERR(hdp);
891 			break;
892 		}
893 
894 		/* Commit writes to this minor */
895 		next[DB_MINOR]++;
896 
897 		dmp->sib = pdnp->minor;
898 		pdnp->minor = idx;
899 
900 		(void) dprintf(DBG_STEP, "%s: minor[%u]: %s\n", fcn, idx,
901 		    cmnp->name);
902 
903 		if (write_links(hdp, dmp, cmnp->link, next) != 0) {
904 			break;
905 		}
906 	}
907 
908 	return (cmnp ? -1 : 0);
909 }
910 
911 static int
912 write_links(
913 	struct di_devlink_handle *hdp,
914 	struct db_minor *pdmp,
915 	cache_link_t *clp,
916 	uint32_t *next)
917 {
918 	uint32_t idx;
919 	struct db_link *dlp;
920 	const char *fcn = "write_links";
921 
922 	assert(HDL_RDWR(hdp));
923 
924 	/* A NULL minor if and only if the links are dangling */
925 	if (clp != NULL && ((pdmp == NULL) ^ (clp == CACHE(hdp)->dngl))) {
926 		(void) dprintf(DBG_ERR, "%s: invalid minor for link\n", fcn);
927 		SET_DB_ERR(hdp);
928 		return (-1);
929 	}
930 
931 	for (; clp != NULL; clp = clp->sib) {
932 
933 		assert(clp->path != NULL);
934 
935 		if ((pdmp == NULL) ^ (clp->minor == NULL)) {
936 			(void) dprintf(DBG_ERR, "%s: invalid minor for link"
937 			    "(%s)\n", fcn, clp->path);
938 			SET_DB_ERR(hdp);
939 			break;
940 		}
941 
942 		idx = next[DB_LINK];
943 		if ((dlp = set_link(hdp, idx)) == NULL) {
944 			SET_DB_ERR(hdp);
945 			break;
946 		}
947 
948 		dlp->path = write_string(hdp, clp->path, next);
949 		dlp->content = write_string(hdp, clp->content, next);
950 		if (dlp->path == DB_NIL || dlp->content == DB_NIL) {
951 			dlp->path = dlp->content = DB_NIL;
952 			SET_DB_ERR(hdp);
953 			break;
954 		}
955 
956 		dlp->attr = clp->attr;
957 
958 		/* Commit writes to this link */
959 		next[DB_LINK]++;
960 
961 		if (pdmp != NULL) {
962 			dlp->sib = pdmp->link;
963 			pdmp->link = idx;
964 		} else {
965 			dlp->sib = DB_HDR(hdp)->dngl_idx;
966 			DB_HDR(hdp)->dngl_idx = idx;
967 		}
968 
969 		(void) dprintf(DBG_STEP, "%s: link[%u]: %s%s\n", fcn, idx,
970 		    clp->path, pdmp == NULL ? "(DANGLING)" : "");
971 	}
972 
973 	return (clp ? -1 : 0);
974 }
975 
976 
977 static uint32_t
978 write_string(struct di_devlink_handle *hdp, const char *str, uint32_t *next)
979 {
980 	char *dstr;
981 	uint32_t idx;
982 
983 	assert(HDL_RDWR(hdp));
984 
985 	if (str == NULL) {
986 		(void) dprintf(DBG_ERR, "write_string: NULL argument\n");
987 		return (DB_NIL);
988 	}
989 
990 	idx = next[DB_STR];
991 	if (!VALID_STR(hdp, idx, str)) {
992 		(void) dprintf(DBG_ERR, "write_string: invalid index[%u],"
993 		    " string(%s)\n", idx, str);
994 		return (DB_NIL);
995 	}
996 
997 	if ((dstr = set_string(hdp, idx)) == NULL) {
998 		return (DB_NIL);
999 	}
1000 
1001 	(void) strcpy(dstr, str);
1002 
1003 	next[DB_STR] += strlen(dstr) + 1;
1004 
1005 	return (idx);
1006 }
1007 
1008 static int
1009 close_db(struct di_devlink_handle *hdp)
1010 {
1011 	int i, rv = 0;
1012 	size_t sz;
1013 
1014 	if (!DB_OPEN(hdp)) {
1015 #ifdef	DEBUG
1016 		assert(DB(hdp)->db_fd == -1);
1017 		assert(DB(hdp)->flags == 0);
1018 		for (i = 0; i < DB_TYPES; i++) {
1019 			assert(DB_SEG(hdp, i) == NULL);
1020 			assert(DB_SEG_PROT(hdp, i) == 0);
1021 		}
1022 #endif
1023 		return (0);
1024 	}
1025 
1026 	/* Unmap header after unmapping all other mapped segments */
1027 	for (i = 0; i < DB_TYPES; i++) {
1028 		if (DB_SEG(hdp, i)) {
1029 			sz = seg_size(hdp, i);
1030 			if (DB_RDWR(hdp))
1031 				rv += msync(DB_SEG(hdp, i), sz, MS_SYNC);
1032 			(void) munmap(DB_SEG(hdp, i), sz);
1033 			DB_SEG(hdp, i) = NULL;
1034 			DB_SEG_PROT(hdp, i) = 0;
1035 		}
1036 	}
1037 
1038 	if (DB_RDWR(hdp))
1039 		rv += msync((caddr_t)DB_HDR(hdp), HDR_LEN, MS_SYNC);
1040 	(void) munmap((caddr_t)DB_HDR(hdp), HDR_LEN);
1041 	DB(hdp)->hdr = NULL;
1042 
1043 	(void) close(DB(hdp)->db_fd);
1044 	DB(hdp)->db_fd = -1;
1045 	DB(hdp)->flags = 0;
1046 
1047 	return (rv ? -1 : 0);
1048 }
1049 
1050 
1051 static void
1052 cache_free(struct di_devlink_handle *hdp)
1053 {
1054 	cache_link_t *clp;
1055 
1056 	subtree_free(hdp, &(CACHE_ROOT(hdp)));
1057 	assert(CACHE_LAST(hdp) == NULL);
1058 
1059 	/*
1060 	 * Don't bother removing links from hash table chains,
1061 	 * as we are freeing the hash table itself.
1062 	 */
1063 	while (CACHE(hdp)->dngl != NULL) {
1064 		clp = CACHE(hdp)->dngl;
1065 		CACHE(hdp)->dngl = clp->sib;
1066 		assert(clp->minor == NULL);
1067 		link_free(&clp);
1068 	}
1069 
1070 	assert((CACHE(hdp)->hash == NULL) ^ (CACHE(hdp)->hash_sz != 0));
1071 
1072 	free(CACHE(hdp)->hash);
1073 	CACHE(hdp)->hash = NULL;
1074 	CACHE(hdp)->hash_sz = 0;
1075 }
1076 
1077 static void
1078 handle_free(struct di_devlink_handle **pp)
1079 {
1080 	struct di_devlink_handle *hdp = *pp;
1081 
1082 	*pp = NULL;
1083 
1084 	if (hdp == NULL)
1085 		return;
1086 
1087 	(void) close_db(hdp);
1088 	cache_free(hdp);
1089 
1090 	if (HDL_RDWR(hdp))
1091 		exit_db_lock(hdp);
1092 	assert(hdp->lock_fd == -1);
1093 
1094 	free(hdp->dev_dir);
1095 	free(hdp->db_dir);
1096 	free(hdp);
1097 }
1098 
1099 /*
1100  * Frees the tree rooted at a node. Siblings of the subtree root
1101  * have to be handled by the caller.
1102  */
1103 static void
1104 subtree_free(struct di_devlink_handle *hdp, cache_node_t **pp)
1105 {
1106 	cache_node_t *np;
1107 	cache_link_t *clp;
1108 	cache_minor_t *cmnp;
1109 
1110 	if (pp == NULL || *pp == NULL)
1111 		return;
1112 
1113 	while ((*pp)->child != NULL) {
1114 		np = (*pp)->child;
1115 		(*pp)->child = np->sib;
1116 		subtree_free(hdp, &np);
1117 	}
1118 
1119 	while ((*pp)->minor != NULL) {
1120 		cmnp = (*pp)->minor;
1121 		(*pp)->minor = cmnp->sib;
1122 
1123 		while (cmnp->link != NULL) {
1124 			clp = cmnp->link;
1125 			cmnp->link = clp->sib;
1126 			rm_link_from_hash(hdp, clp);
1127 			link_free(&clp);
1128 		}
1129 		minor_free(hdp, &cmnp);
1130 	}
1131 
1132 	node_free(pp);
1133 }
1134 
1135 static void
1136 rm_link_from_hash(struct di_devlink_handle *hdp, cache_link_t *clp)
1137 {
1138 	int hval;
1139 	cache_link_t **pp;
1140 
1141 	if (clp == NULL)
1142 		return;
1143 
1144 	if (clp->path == NULL)
1145 		return;
1146 
1147 	hval = hashfn(hdp, clp->path);
1148 	pp = &(CACHE_HASH(hdp, hval));
1149 	for (; *pp != NULL; pp = &(*pp)->hash) {
1150 		if (*pp == clp) {
1151 			*pp = clp->hash;
1152 			clp->hash = NULL;
1153 			return;
1154 		}
1155 	}
1156 
1157 	dprintf(DBG_ERR, "rm_link_from_hash: link(%s) not found\n", clp->path);
1158 }
1159 
1160 static cache_link_t *
1161 link_hash(di_devlink_handle_t hdp, const char *link, uint_t flags)
1162 {
1163 	int hval;
1164 	cache_link_t **pp, *clp;
1165 
1166 	if (link == NULL)
1167 		return (NULL);
1168 
1169 	hval = hashfn(hdp, link);
1170 	pp = &(CACHE_HASH(hdp, hval));
1171 	for (; (clp = *pp) != NULL; pp = &clp->hash) {
1172 		if (strcmp(clp->path, link) == 0) {
1173 			break;
1174 		}
1175 	}
1176 
1177 	if (clp == NULL)
1178 		return (NULL);
1179 
1180 	if ((flags & UNLINK_FROM_HASH) == UNLINK_FROM_HASH) {
1181 		*pp = clp->hash;
1182 		clp->hash = NULL;
1183 	}
1184 
1185 	return (clp);
1186 }
1187 
1188 static cache_minor_t *
1189 link2minor(struct di_devlink_handle *hdp, cache_link_t *clp)
1190 {
1191 	cache_link_t *plp;
1192 	const char *minor_path;
1193 	char *cp, buf[PATH_MAX], link[PATH_MAX];
1194 	char abspath[PATH_MAX];
1195 	struct stat st;
1196 
1197 	if (TYPE_PRI(attr2type(clp->attr))) {
1198 		/*
1199 		 * For primary link, content should point to a /devices node.
1200 		 */
1201 		if (!is_minor_node(clp->content, &minor_path)) {
1202 			return (NULL);
1203 		}
1204 
1205 		return (lookup_minor(hdp, minor_path, NULL,
1206 		    TYPE_CACHE|CREATE_FLAG));
1207 
1208 	}
1209 
1210 	/*
1211 	 * If secondary, the primary link is derived from the secondary
1212 	 * link contents. Secondary link contents can have two formats:
1213 	 *	audio -> /dev/sound/0
1214 	 *	fb0 -> fbs/afb0
1215 	 */
1216 
1217 	buf[0] = '\0';
1218 	if (strncmp(clp->content, DEV"/", strlen(DEV"/")) == 0) {
1219 		cp = &clp->content[strlen(DEV"/")];
1220 	} else if (clp->content[0] != '/') {
1221 		if ((cp = strrchr(clp->path, '/')) != NULL) {
1222 			char savechar = *(cp + 1);
1223 			*(cp + 1) = '\0';
1224 			(void) snprintf(buf, sizeof (buf), "%s", clp->path);
1225 			*(cp + 1) = savechar;
1226 		}
1227 		(void) strlcat(buf, clp->content, sizeof (buf));
1228 		cp = buf;
1229 	} else {
1230 		goto follow_link;
1231 	}
1232 
1233 	/*
1234 	 * Lookup the primary link if possible and find its minor.
1235 	 */
1236 	if ((plp = link_hash(hdp, cp, 0)) != NULL && plp->minor != NULL) {
1237 		return (plp->minor);
1238 	}
1239 
1240 	/* realpath() used only as a last resort because it is expensive */
1241 follow_link:
1242 	(void) snprintf(link, sizeof (link), "%s/%s", hdp->dev_dir, clp->path);
1243 
1244 #ifdef	DEBUG
1245 	/*LINTED*/
1246 	assert(sizeof (buf) >= PATH_MAX);
1247 #endif
1248 
1249 	/*
1250 	 * A realpath attempt to lookup a dangling link can invoke implicit
1251 	 * reconfig so verify there's an actual device behind the link first.
1252 	 */
1253 	if (lstat(link, &st) == -1)
1254 		return (NULL);
1255 	if (S_ISLNK(st.st_mode)) {
1256 		if (s_readlink(link, buf, sizeof (buf)) < 0)
1257 			return (NULL);
1258 		if (buf[0] != '/') {
1259 			char *p;
1260 			size_t n = sizeof (abspath);
1261 			if (strlcpy(abspath, link, n) >= n)
1262 				return (NULL);
1263 			p = strrchr(abspath, '/') + 1;
1264 			*p = 0;
1265 			n = sizeof (abspath) - strlen(p);
1266 			if (strlcpy(p, buf, n) >= n)
1267 				return (NULL);
1268 		} else {
1269 			if (strlcpy(abspath, buf, sizeof (abspath)) >=
1270 			    sizeof (abspath))
1271 				return (NULL);
1272 		}
1273 		if (!device_exists(abspath))
1274 			return (NULL);
1275 	}
1276 
1277 	if (s_realpath(link, buf) == NULL || !is_minor_node(buf, &minor_path)) {
1278 		return (NULL);
1279 	}
1280 	return (lookup_minor(hdp, minor_path, NULL, TYPE_CACHE|CREATE_FLAG));
1281 }
1282 
1283 
1284 static void
1285 resolve_dangling_links(struct di_devlink_handle *hdp)
1286 {
1287 	cache_minor_t *cmnp;
1288 	cache_link_t *clp, **pp;
1289 
1290 	for (pp = &(CACHE(hdp)->dngl); *pp != NULL; ) {
1291 		clp = *pp;
1292 		if ((cmnp = link2minor(hdp, clp)) != NULL) {
1293 			*pp = clp->sib;
1294 			clp->sib = cmnp->link;
1295 			cmnp->link = clp;
1296 			assert(clp->minor == NULL);
1297 			clp->minor = cmnp;
1298 		} else {
1299 			dprintf(DBG_INFO, "resolve_dangling_links: link(%s):"
1300 			    " unresolved\n", clp->path);
1301 			pp = &clp->sib;
1302 		}
1303 	}
1304 }
1305 
1306 
1307 /*
1308  * The elements are assumed to be detached from the cache tree.
1309  */
1310 static void
1311 node_free(cache_node_t **pp)
1312 {
1313 	cache_node_t *cnp = *pp;
1314 
1315 	*pp = NULL;
1316 
1317 	if (cnp == NULL)
1318 		return;
1319 
1320 	free(cnp->path);
1321 	free(cnp);
1322 }
1323 
1324 static void
1325 minor_free(struct di_devlink_handle *hdp, cache_minor_t **pp)
1326 {
1327 	cache_minor_t *cmnp = *pp;
1328 
1329 	*pp = NULL;
1330 
1331 	if (cmnp == NULL)
1332 		return;
1333 
1334 	if (CACHE_LAST(hdp) == cmnp) {
1335 		dprintf(DBG_STEP, "minor_free: last_minor(%s)\n", cmnp->name);
1336 		CACHE_LAST(hdp) = NULL;
1337 	}
1338 
1339 	free(cmnp->name);
1340 	free(cmnp->nodetype);
1341 	free(cmnp);
1342 }
1343 
1344 static void
1345 link_free(cache_link_t **pp)
1346 {
1347 	cache_link_t *clp = *pp;
1348 
1349 	*pp = NULL;
1350 
1351 	if (clp == NULL)
1352 		return;
1353 
1354 	free(clp->path);
1355 	free(clp->content);
1356 	free(clp);
1357 }
1358 
1359 /*
1360  * Returns the ':' preceding the minor name
1361  */
1362 static char *
1363 minor_colon(const char *path)
1364 {
1365 	char *cp;
1366 
1367 	if ((cp = strrchr(path, '/')) == NULL) {
1368 		return (NULL);
1369 	}
1370 
1371 	return (strchr(cp, ':'));
1372 }
1373 
1374 static void *
1375 lookup_minor(
1376 	struct di_devlink_handle *hdp,
1377 	const char *minor_path,
1378 	const char *nodetype,
1379 	const int flags)
1380 {
1381 	void *vp;
1382 	char *colon;
1383 	char pdup[PATH_MAX];
1384 	const char *fcn = "lookup_minor";
1385 
1386 	if (minor_path == NULL) {
1387 		errno = EINVAL;
1388 		return (NULL);
1389 	}
1390 
1391 	(void) snprintf(pdup, sizeof (pdup), "%s", minor_path);
1392 
1393 	if ((colon = minor_colon(pdup)) == NULL) {
1394 		(void) dprintf(DBG_ERR, "%s: invalid minor path(%s)\n", fcn,
1395 		    minor_path);
1396 		errno = EINVAL;
1397 		return (NULL);
1398 	}
1399 	*colon = '\0';
1400 
1401 	if ((vp = get_last_minor(hdp, pdup, colon + 1, flags)) != NULL) {
1402 		return (vp);
1403 	}
1404 
1405 	if ((vp = lookup_node(hdp, pdup, flags)) == NULL) {
1406 		(void) dprintf(DBG_ERR, "%s: node(%s) not found\n", fcn, pdup);
1407 		return (NULL);
1408 	}
1409 	*colon = ':';
1410 
1411 	if (LOOKUP_CACHE(flags)) {
1412 		cache_minor_t **pp;
1413 
1414 		pp = &((cache_node_t *)vp)->minor;
1415 		for (; *pp != NULL; pp = &(*pp)->sib) {
1416 			if (strcmp((*pp)->name, colon + 1) == 0)
1417 				break;
1418 		}
1419 
1420 		if (*pp == NULL && CREATE_ELEM(flags)) {
1421 			*pp = minor_insert(hdp, vp, colon + 1, nodetype, pp);
1422 		}
1423 		set_last_minor(hdp, *pp, flags);
1424 
1425 		return (*pp);
1426 	} else {
1427 		char *cp;
1428 		uint32_t nidx;
1429 		struct db_minor *dmp;
1430 
1431 		nidx = (((struct db_node *)vp)->minor);
1432 		for (; dmp = get_minor(hdp, nidx); nidx = dmp->sib) {
1433 			cp = get_string(hdp, dmp->name);
1434 			if (cp && strcmp(cp, colon + 1) == 0)
1435 				break;
1436 		}
1437 		return (dmp);
1438 	}
1439 }
1440 
1441 static void *
1442 lookup_node(struct di_devlink_handle *hdp, char *path, const int flags)
1443 {
1444 	struct tnode tnd = {NULL};
1445 
1446 	if (tnd.node = get_last_node(hdp, path, flags))
1447 		return (tnd.node);
1448 
1449 	tnd.handle = hdp;
1450 	tnd.flags = flags;
1451 
1452 	if (walk_tree(path, &tnd, visit_node) != 0)
1453 		return (NULL);
1454 
1455 	return (tnd.node);
1456 }
1457 
1458 /*
1459  * last_minor is used for nodes of TYPE_CACHE only.
1460  */
1461 static void *
1462 get_last_node(struct di_devlink_handle *hdp, const char *path, int flags)
1463 {
1464 	cache_node_t *cnp;
1465 
1466 #ifdef	DEBUG
1467 	if (getenv(SKIP_LAST_CACHE)) {
1468 		(void) dprintf(DBG_INFO, "get_last_node: SKIPPING \"last\" "
1469 		    "node cache\n");
1470 		return (NULL);
1471 	}
1472 #endif
1473 
1474 	if (!LOOKUP_CACHE(flags) || CACHE_LAST(hdp) == NULL ||
1475 	    CACHE_LAST(hdp)->node == NULL) {
1476 		return (NULL);
1477 	}
1478 
1479 	cnp = CACHE_LAST(hdp)->node;
1480 	if (strcmp(cnp->path, path) == 0) {
1481 		return (cnp);
1482 	}
1483 
1484 	cnp = cnp->sib;
1485 	if (cnp && strcmp(cnp->path, path) == 0) {
1486 		return (cnp);
1487 	}
1488 
1489 	return (NULL);
1490 }
1491 
1492 static void *
1493 get_last_minor(
1494 	struct di_devlink_handle *hdp,
1495 	const char *devfs_path,
1496 	const char *minor_name,
1497 	int flags)
1498 {
1499 	cache_minor_t *cmnp;
1500 
1501 #ifdef	DEBUG
1502 	if (getenv(SKIP_LAST_CACHE)) {
1503 		(void) dprintf(DBG_INFO, "get_last_minor: SKIPPING \"last\" "
1504 		    "minor cache\n");
1505 		return (NULL);
1506 	}
1507 #endif
1508 
1509 	if (!LOOKUP_CACHE(flags) || CACHE_LAST(hdp) == NULL) {
1510 		return (NULL);
1511 	}
1512 
1513 	cmnp = CACHE_LAST(hdp);
1514 	if (strcmp(cmnp->name, minor_name) == 0 && cmnp->node &&
1515 	    strcmp(cmnp->node->path, devfs_path) == 0) {
1516 		return (cmnp);
1517 	}
1518 
1519 	cmnp = cmnp->sib;
1520 	if (cmnp && strcmp(cmnp->name, minor_name) == 0 && cmnp->node &&
1521 	    strcmp(cmnp->node->path, devfs_path) == 0) {
1522 		set_last_minor(hdp, cmnp, TYPE_CACHE);
1523 		return (cmnp);
1524 	}
1525 
1526 	return (NULL);
1527 }
1528 
1529 static void
1530 set_last_minor(struct di_devlink_handle *hdp, cache_minor_t *cmnp, int flags)
1531 {
1532 #ifdef	DEBUG
1533 	if (getenv(SKIP_LAST_CACHE)) {
1534 		(void) dprintf(DBG_INFO, "set_last_minor: SKIPPING \"last\" "
1535 		    "minor cache\n");
1536 		return;
1537 	}
1538 #endif
1539 
1540 	if (LOOKUP_CACHE(flags) && cmnp) {
1541 		CACHE_LAST(hdp) = cmnp;
1542 	}
1543 }
1544 
1545 
1546 /*
1547  * Returns 0 if normal return or -1 otherwise.
1548  */
1549 static int
1550 walk_tree(
1551 	char *cur,
1552 	void *arg,
1553 	int (*node_callback)(const char *path, void *arg))
1554 {
1555 	char *slash, buf[PATH_MAX];
1556 
1557 	if (cur == NULL || cur[0] != '/' || strlen(cur) > sizeof (buf) - 1) {
1558 		errno = EINVAL;
1559 		return (-1);
1560 	}
1561 
1562 	(void) strcpy(buf, "/");
1563 
1564 	for (;;) {
1565 
1566 		if (node_callback(buf, arg) != DI_WALK_CONTINUE)
1567 			break;
1568 
1569 		while (*cur == '/')
1570 			cur++;
1571 
1572 		if (*cur == '\0')
1573 			break;
1574 
1575 		/*
1576 		 * There is a next component(s). Append a "/" separator for all
1577 		 * but the first (root) component.
1578 		 */
1579 		if (buf[1] != '\0') {
1580 			(void) strlcat(buf, "/", sizeof (buf));
1581 		}
1582 
1583 		if (slash = strchr(cur, '/')) {
1584 			*slash = '\0';
1585 			(void) strlcat(buf, cur, sizeof (buf));
1586 			*slash = '/';
1587 			cur = slash;
1588 		} else {
1589 			(void) strlcat(buf, cur, sizeof (buf));
1590 			cur += strlen(cur);
1591 		}
1592 
1593 	}
1594 
1595 	return (0);
1596 }
1597 
1598 
1599 static int
1600 visit_node(const char *path, void *arg)
1601 {
1602 	struct tnode *tnp = arg;
1603 
1604 	if (LOOKUP_CACHE(tnp->flags)) {
1605 
1606 		cache_node_t *cnp = tnp->node;
1607 
1608 		cnp = (cnp) ? cnp->child : CACHE_ROOT(tnp->handle);
1609 
1610 		for (; cnp != NULL; cnp = cnp->sib) {
1611 			if (strcmp(cnp->path, path) == 0)
1612 				break;
1613 		}
1614 		if (cnp == NULL && CREATE_ELEM(tnp->flags)) {
1615 			cnp = node_insert(tnp->handle, tnp->node, path,
1616 			    INSERT_TAIL);
1617 		}
1618 		tnp->node = cnp;
1619 	} else {
1620 		char *cp;
1621 		struct db_node *dnp = tnp->node;
1622 
1623 		dnp = (dnp) ? get_node(tnp->handle, dnp->child)
1624 		    : get_node(tnp->handle, DB_HDR(tnp->handle)->root_idx);
1625 
1626 		for (; dnp != NULL; dnp = get_node(tnp->handle, dnp->sib)) {
1627 			cp = get_string(tnp->handle, dnp->path);
1628 			if (cp && strcmp(cp, path) == 0) {
1629 				break;
1630 			}
1631 		}
1632 		tnp->node = dnp;
1633 	}
1634 
1635 	/*
1636 	 * Terminate walk if node is not found for a path component.
1637 	 */
1638 	return (tnp->node ? DI_WALK_CONTINUE : DI_WALK_TERMINATE);
1639 }
1640 
1641 static void
1642 minor_delete(di_devlink_handle_t hdp, cache_minor_t *cmnp)
1643 {
1644 	cache_link_t **lpp;
1645 	cache_minor_t **mpp;
1646 	const char *fcn = "minor_delete";
1647 
1648 	(void) dprintf(DBG_STEP, "%s: removing minor: %s\n", fcn, cmnp->name);
1649 
1650 	/* detach minor from node */
1651 	if (cmnp->node != NULL) {
1652 		mpp = &cmnp->node->minor;
1653 		for (; *mpp != NULL; mpp = &(*mpp)->sib) {
1654 			if (*mpp == cmnp)
1655 				break;
1656 		}
1657 
1658 		if (*mpp == NULL) {
1659 			(void) dprintf(DBG_ERR, "%s: dangling minor: %s\n",
1660 			    fcn, cmnp->name);
1661 		} else {
1662 			*mpp = cmnp->sib;
1663 		}
1664 	} else {
1665 		(void) dprintf(DBG_ERR, "%s: orphan minor(%s)\n", fcn,
1666 		    cmnp->name);
1667 	}
1668 
1669 	delete_unused_nodes(hdp, cmnp->node);
1670 
1671 	cmnp->node = NULL;
1672 	cmnp->sib = NULL;
1673 
1674 	/* Move all remaining links to dangling list */
1675 	for (lpp = &cmnp->link; *lpp != NULL; lpp = &(*lpp)->sib) {
1676 		(*lpp)->minor = NULL;
1677 	}
1678 	*lpp = CACHE(hdp)->dngl;
1679 	CACHE(hdp)->dngl = cmnp->link;
1680 	cmnp->link = NULL;
1681 
1682 	minor_free(hdp, &cmnp);
1683 }
1684 
1685 static void
1686 delete_unused_nodes(di_devlink_handle_t hdp, cache_node_t *cnp)
1687 {
1688 	cache_node_t **npp;
1689 	const char *fcn = "delete_unused_nodes";
1690 
1691 	if (cnp == NULL)
1692 		return;
1693 
1694 	if (cnp->minor != NULL || cnp->child != NULL)
1695 		return;
1696 
1697 	(void) dprintf(DBG_INFO, "%s: removing unused node: %s\n", fcn,
1698 	    cnp->path);
1699 
1700 	/* Unlink node from tree */
1701 	if (cnp->parent != NULL) {
1702 		npp = &cnp->parent->child;
1703 		for (; *npp != NULL; npp = &(*npp)->sib) {
1704 			if (*npp == cnp)
1705 				break;
1706 		}
1707 
1708 		if (*npp == NULL) {
1709 			(void) dprintf(DBG_ERR, "%s: dangling node: %s\n", fcn,
1710 			    cnp->path);
1711 		} else {
1712 			*npp = cnp->sib;
1713 		}
1714 	} else if (cnp == CACHE_ROOT(hdp)) {
1715 		CACHE_ROOT(hdp) = NULL;
1716 	} else {
1717 		(void) dprintf(DBG_ERR, "%s: orphan node (%s)\n", fcn,
1718 		    cnp->path);
1719 	}
1720 
1721 	delete_unused_nodes(hdp, cnp->parent);
1722 
1723 	cnp->parent = cnp->sib = NULL;
1724 
1725 	node_free(&cnp);
1726 }
1727 
1728 static int
1729 rm_link(di_devlink_handle_t hdp, const char *link)
1730 {
1731 	cache_link_t *clp;
1732 	const char *fcn = "rm_link";
1733 
1734 	if (hdp == NULL || DB_ERR(hdp) || link == NULL || link[0] == '/' ||
1735 	    (!HDL_RDWR(hdp) && !HDL_RDONLY(hdp))) {
1736 		dprintf(DBG_ERR, "%s: %s: invalid args\n",
1737 		    fcn, link ? link : "<NULL>");
1738 		errno = EINVAL;
1739 		return (-1);
1740 	}
1741 
1742 	dprintf(DBG_STEP, "%s: link(%s)\n", fcn, link);
1743 
1744 	if ((clp = link_hash(hdp, link, UNLINK_FROM_HASH)) == NULL) {
1745 		return (0);
1746 	}
1747 
1748 	link_delete(hdp, clp);
1749 
1750 	return (0);
1751 }
1752 
1753 int
1754 di_devlink_rm_link(di_devlink_handle_t hdp, const char *link)
1755 {
1756 	if (hdp == NULL || !HDL_RDWR(hdp)) {
1757 		errno = EINVAL;
1758 		return (-1);
1759 	}
1760 
1761 	return (rm_link(hdp, link));
1762 }
1763 
1764 static void
1765 link_delete(di_devlink_handle_t hdp, cache_link_t *clp)
1766 {
1767 	cache_link_t **pp;
1768 	const char *fcn = "link_delete";
1769 
1770 	(void) dprintf(DBG_STEP, "%s: removing link: %s\n", fcn, clp->path);
1771 
1772 	if (clp->minor == NULL)
1773 		pp = &(CACHE(hdp)->dngl);
1774 	else
1775 		pp = &clp->minor->link;
1776 
1777 	for (; *pp != NULL; pp = &(*pp)->sib) {
1778 		if (*pp == clp)
1779 			break;
1780 	}
1781 
1782 	if (*pp == NULL) {
1783 		(void) dprintf(DBG_ERR, "%s: link(%s) not on list\n",
1784 		    fcn, clp->path);
1785 	} else {
1786 		*pp = clp->sib;
1787 	}
1788 
1789 	delete_unused_minor(hdp, clp->minor);
1790 
1791 	clp->minor = NULL;
1792 
1793 	link_free(&clp);
1794 }
1795 
1796 static void
1797 delete_unused_minor(di_devlink_handle_t hdp, cache_minor_t *cmnp)
1798 {
1799 	if (cmnp == NULL)
1800 		return;
1801 
1802 	if (cmnp->link != NULL)
1803 		return;
1804 
1805 	dprintf(DBG_STEP, "delete_unused_minor: removing minor(%s)\n",
1806 	    cmnp->name);
1807 
1808 	minor_delete(hdp, cmnp);
1809 }
1810 
1811 int
1812 di_devlink_add_link(
1813 	di_devlink_handle_t hdp,
1814 	const char *link,
1815 	const char *content,
1816 	int flags)
1817 {
1818 	return (add_link(hdp, link, content, flags) != NULL ? 0 : -1);
1819 }
1820 
1821 static cache_link_t *
1822 add_link(
1823 	struct di_devlink_handle *hdp,
1824 	const char *link,
1825 	const char *content,
1826 	int flags)
1827 {
1828 	uint32_t attr;
1829 	cache_link_t *clp;
1830 	cache_minor_t *cmnp;
1831 	const char *fcn = "add_link";
1832 
1833 	if (hdp == NULL || DB_ERR(hdp) || link == NULL ||
1834 	    link[0] == '/' || content == NULL || !link_flag(flags) ||
1835 	    (!HDL_RDWR(hdp) && !HDL_RDONLY(hdp))) {
1836 		dprintf(DBG_ERR, "%s: %s: invalid args\n",
1837 		    fcn, link ? link : "<NULL>");
1838 		errno = EINVAL;
1839 		return (NULL);
1840 	}
1841 
1842 	if ((clp = link_hash(hdp, link, 0)) != NULL) {
1843 		if (link_cmp(clp, content, LINK_TYPE(flags)) != 0) {
1844 			(void) rm_link(hdp, link);
1845 		} else {
1846 			return (clp);
1847 		}
1848 	}
1849 
1850 	if (TYPE_PRI(flags)) {
1851 		const char *minor_path = NULL;
1852 
1853 		if (!is_minor_node(content, &minor_path)) {
1854 			(void) dprintf(DBG_ERR, "%s: invalid content(%s)"
1855 			    " for primary link\n", fcn, content);
1856 			errno = EINVAL;
1857 			return (NULL);
1858 		}
1859 		if ((cmnp = lookup_minor(hdp, minor_path, NULL,
1860 		    TYPE_CACHE|CREATE_FLAG)) == NULL) {
1861 			return (NULL);
1862 		}
1863 		attr = A_PRIMARY;
1864 	} else {
1865 		/*
1866 		 * Defer resolving a secondary link to a minor until the
1867 		 * database is closed. This ensures that the primary link
1868 		 * (required for a successful resolve) has also been created.
1869 		 */
1870 		cmnp = NULL;
1871 		attr = A_SECONDARY;
1872 	}
1873 
1874 	return (link_insert(hdp, cmnp, link, content, attr));
1875 }
1876 
1877 /*
1878  * Returns 0 on match or 1 otherwise.
1879  */
1880 static int
1881 link_cmp(cache_link_t *clp, const char *content, int type)
1882 {
1883 	if (strcmp(clp->content, content) != 0)
1884 		return (1);
1885 
1886 	if (attr2type(clp->attr) != type)
1887 		return (1);
1888 
1889 	return (0);
1890 }
1891 
1892 int
1893 di_devlink_update(di_devlink_handle_t hdp)
1894 {
1895 	if (hdp == NULL || !HDL_RDWR(hdp) || DB_ERR(hdp)) {
1896 		errno = EINVAL;
1897 		return (-1);
1898 	}
1899 
1900 	/*
1901 	 * Reset the counter to schedule a synchronization with /dev on the next
1902 	 * di_devlink_close().
1903 	 */
1904 	CACHE(hdp)->update_count = 0;
1905 
1906 	return (0);
1907 }
1908 
1909 static int
1910 synchronize_db(di_devlink_handle_t hdp)
1911 {
1912 	int hval;
1913 	cache_link_t *clp;
1914 	char pdup[PATH_MAX];
1915 	recurse_t rec = {NULL};
1916 	const char *fcn = "synchronize_db";
1917 
1918 	rec.data = NULL;
1919 	rec.fcn = cache_dev_link;
1920 
1921 	/*
1922 	 * Walk through $ROOT/dev, reading every link and marking the
1923 	 * corresponding cached version as valid(adding new links as needed).
1924 	 * Then walk through the cache and remove all unmarked links.
1925 	 */
1926 	if (recurse_dev(hdp, &rec) != 0) {
1927 		return (-1);
1928 	}
1929 
1930 	for (hval = 0; hval < CACHE(hdp)->hash_sz; hval++) {
1931 		for (clp = CACHE_HASH(hdp, hval); clp != NULL; ) {
1932 			if (GET_VALID_ATTR(clp->attr)) {
1933 				CLR_VALID_ATTR(clp->attr);
1934 				clp = clp->hash;
1935 				continue;
1936 			}
1937 
1938 			/*
1939 			 * The link is stale, so remove it. Since the link
1940 			 * will be destroyed, use a copy of the link path to
1941 			 * invoke the remove function.
1942 			 */
1943 			(void) snprintf(pdup, sizeof (pdup), "%s", clp->path);
1944 			clp = clp->hash;
1945 			(void) dprintf(DBG_STEP, "%s: removing invalid link:"
1946 			    " %s\n", fcn, pdup);
1947 			(void) di_devlink_rm_link(hdp, pdup);
1948 		}
1949 	}
1950 
1951 	(void) dprintf(DBG_STEP, "%s: update completed\n", fcn);
1952 
1953 	return (0);
1954 }
1955 
1956 static di_devlink_handle_t
1957 di_devlink_init_impl(const char *root, const char *name, uint_t flags)
1958 {
1959 	int	err = 0;
1960 
1961 	if ((flags != 0 && flags != DI_MAKE_LINK) ||
1962 	    (flags == 0 && name != NULL)) {
1963 		errno = EINVAL;
1964 		return (NULL);
1965 	}
1966 
1967 	if ((flags == DI_MAKE_LINK) &&
1968 	    (err = devlink_create(root, name, DCA_DEVLINK_CACHE))) {
1969 		errno = err;
1970 		return (NULL);
1971 	}
1972 
1973 	(void) dprintf(DBG_INFO, "devlink_init_impl: success\n");
1974 
1975 	return (devlink_snapshot(root));
1976 }
1977 
1978 di_devlink_handle_t
1979 di_devlink_init(const char *name, uint_t flags)
1980 {
1981 	return (di_devlink_init_impl("/", name, flags));
1982 }
1983 
1984 di_devlink_handle_t
1985 di_devlink_init_root(const char *root, const char *name, uint_t flags)
1986 {
1987 	return (di_devlink_init_impl(root, name, flags));
1988 }
1989 
1990 static di_devlink_handle_t
1991 devlink_snapshot(const char *root_dir)
1992 {
1993 	struct di_devlink_handle *hdp;
1994 	int		err;
1995 	static int	retried = 0;
1996 
1997 	if ((hdp = handle_alloc(root_dir, OPEN_RDONLY)) == NULL) {
1998 		return (NULL);
1999 	}
2000 
2001 	/*
2002 	 * We don't need to lock.  If a consumer wants the very latest db
2003 	 * then he must perform a di_devlink_init with the DI_MAKE_LINK
2004 	 * flag to force a sync with devfsadm first.  Otherwise, the
2005 	 * current database file is opened and mmaped on demand: the rename
2006 	 * associated with a db update does not change the contents
2007 	 * of files already opened.
2008 	 */
2009 again:	err = open_db(hdp, OPEN_RDONLY);
2010 
2011 	/*
2012 	 * If we failed to open DB the most likely cause is that DB file did
2013 	 * not exist. If we have not done a retry, signal devfsadmd to
2014 	 * recreate the DB file and retry. If we fail to open the DB after
2015 	 * retry, we will walk /dev in di_devlink_walk.
2016 	 */
2017 	if (err && (retried == 0)) {
2018 		retried++;
2019 		(void) devlink_create(root_dir, NULL, DCA_DEVLINK_SYNC);
2020 		goto again;
2021 	}
2022 	return (hdp);
2023 }
2024 
2025 int
2026 di_devlink_fini(di_devlink_handle_t *pp)
2027 {
2028 	if (pp == NULL || *pp == NULL || !HDL_RDONLY(*pp)) {
2029 		errno = EINVAL;
2030 		return (-1);
2031 	}
2032 
2033 	/* Freeing the handle also closes the DB */
2034 	handle_free(pp);
2035 
2036 	return (0);
2037 }
2038 
2039 int
2040 di_devlink_walk(
2041 	di_devlink_handle_t hdp,
2042 	const char *re,
2043 	const char *minor_path,
2044 	uint_t flags,
2045 	void *arg,
2046 	int (*devlink_callback)(di_devlink_t, void *))
2047 {
2048 	int rv;
2049 	regex_t reg;
2050 	link_desc_t linkd = {NULL};
2051 
2052 	if (hdp == NULL || !HDL_RDONLY(hdp)) {
2053 		errno = EINVAL;
2054 		return (-1);
2055 	}
2056 
2057 	linkd.minor_path = minor_path;
2058 	linkd.flags = flags;
2059 	linkd.arg = arg;
2060 	linkd.fcn = devlink_callback;
2061 
2062 	if (re) {
2063 		if (regcomp(&reg, re, REG_EXTENDED) != 0)
2064 			return (-1);
2065 		linkd.regp = &reg;
2066 	}
2067 
2068 	if (check_args(&linkd)) {
2069 		errno = EINVAL;
2070 		rv = -1;
2071 		goto out;
2072 	}
2073 
2074 	if (DB_OPEN(hdp)) {
2075 		rv = walk_db(hdp, &linkd);
2076 	} else {
2077 		rv = walk_dev(hdp, &linkd);
2078 	}
2079 
2080 out:
2081 	if (re) {
2082 		regfree(&reg);
2083 	}
2084 
2085 	return (rv ? -1 : 0);
2086 }
2087 
2088 static int
2089 link_flag(uint_t flags)
2090 {
2091 	if (flags != 0 && flags != DI_PRIMARY_LINK &&
2092 	    flags != DI_SECONDARY_LINK) {
2093 		return (0);
2094 	}
2095 
2096 	return (1);
2097 }
2098 
2099 /*
2100  * Currently allowed flags are:
2101  *	DI_PRIMARY_LINK
2102  *	DI_SECONDARY_LINK
2103  */
2104 static int
2105 check_args(link_desc_t *linkp)
2106 {
2107 	if (linkp->fcn == NULL)
2108 		return (-1);
2109 
2110 	if (!link_flag(linkp->flags)) {
2111 		return (-1);
2112 	}
2113 
2114 	/*
2115 	 * Minor path can be NULL. In that case, all links will be
2116 	 * selected.
2117 	 */
2118 	if (linkp->minor_path) {
2119 		if (linkp->minor_path[0] != '/' ||
2120 		    minor_colon(linkp->minor_path) == NULL) {
2121 			return (-1);
2122 		}
2123 	}
2124 
2125 	return (0);
2126 }
2127 
2128 
2129 /*
2130  * Walk all links in database if no minor path is specified.
2131  */
2132 static int
2133 walk_db(struct di_devlink_handle *hdp, link_desc_t *linkp)
2134 {
2135 	assert(DB_OPEN(hdp));
2136 
2137 	if (linkp->minor_path == NULL) {
2138 		return (walk_all_links(hdp, linkp));
2139 	} else {
2140 		return (walk_matching_links(hdp, linkp));
2141 	}
2142 }
2143 
2144 static int
2145 cache_dev(struct di_devlink_handle *hdp)
2146 {
2147 	size_t sz;
2148 	recurse_t rec = {NULL};
2149 
2150 	assert(hdp);
2151 	assert(HDL_RDONLY(hdp));
2152 
2153 	if (hdp == NULL || !HDL_RDONLY(hdp)) {
2154 		dprintf(DBG_ERR, "cache_dev: invalid arg\n");
2155 		return (-1);
2156 	}
2157 
2158 	sz = MIN_HASH_SIZE;
2159 
2160 	CACHE(hdp)->hash = calloc(sz, sizeof (cache_link_t *));
2161 	if (CACHE(hdp)->hash == NULL) {
2162 		return (-1);
2163 	}
2164 	CACHE(hdp)->hash_sz = sz;
2165 
2166 	rec.data = NULL;
2167 	rec.fcn = cache_dev_link;
2168 
2169 	return (recurse_dev(hdp, &rec));
2170 }
2171 
2172 static int
2173 walk_dev(struct di_devlink_handle *hdp, link_desc_t *linkp)
2174 {
2175 	assert(hdp && linkp);
2176 	assert(!DB_OPEN(hdp));
2177 	assert(HDL_RDONLY(hdp));
2178 
2179 	if (hdp == NULL || !HDL_RDONLY(hdp) || DB_OPEN(hdp)) {
2180 		dprintf(DBG_ERR, "walk_dev: invalid args\n");
2181 		return (-1);
2182 	}
2183 
2184 	if (CACHE_EMPTY(hdp) && cache_dev(hdp) != 0) {
2185 		dprintf(DBG_ERR, "walk_dev: /dev caching failed\n");
2186 		return (-1);
2187 	}
2188 
2189 	if (linkp->minor_path)
2190 		walk_cache_minor(hdp, linkp->minor_path, linkp);
2191 	else
2192 		walk_all_cache(hdp, linkp);
2193 
2194 	return (linkp->retval);
2195 }
2196 
2197 /* ARGSUSED */
2198 static int
2199 cache_dev_link(struct di_devlink_handle *hdp, void *data, const char *link)
2200 {
2201 	int flags;
2202 	cache_link_t *clp;
2203 	char content[PATH_MAX];
2204 
2205 	assert(HDL_RDWR(hdp) || HDL_RDONLY(hdp));
2206 
2207 	if (s_readlink(link, content, sizeof (content)) < 0) {
2208 		return (DI_WALK_CONTINUE);
2209 	}
2210 
2211 	if (is_minor_node(content, NULL)) {
2212 		flags = DI_PRIMARY_LINK;
2213 	} else {
2214 		flags = DI_SECONDARY_LINK;
2215 	}
2216 
2217 	assert(strncmp(link, hdp->dev_dir, strlen(hdp->dev_dir)) == 0);
2218 
2219 	/*
2220 	 * Store only the part after <root-dir>/dev/
2221 	 */
2222 	link += strlen(hdp->dev_dir) + 1;
2223 
2224 	if ((clp = add_link(hdp, link, content, flags)) != NULL) {
2225 		SET_VALID_ATTR(clp->attr);
2226 	}
2227 
2228 	return (DI_WALK_CONTINUE);
2229 }
2230 
2231 
2232 static int
2233 walk_all_links(struct di_devlink_handle *hdp, link_desc_t *linkp)
2234 {
2235 	struct db_link *dlp;
2236 	uint32_t nidx, eidx;
2237 
2238 	assert(DB_NUM(hdp, DB_LINK) >= 1);
2239 
2240 	eidx = DB_NUM(hdp, DB_LINK);
2241 
2242 	/* Skip the "NIL" (index == 0) link. */
2243 	for (nidx = 1; nidx < eidx; nidx++) {
2244 		/*
2245 		 * Declare this local to the block with zero
2246 		 * initializer so that it gets rezeroed
2247 		 * for each iteration.
2248 		 */
2249 		struct di_devlink vlink = {NULL};
2250 
2251 		if ((dlp = get_link(hdp, nidx)) == NULL)
2252 			continue;
2253 
2254 		vlink.rel_path = get_string(hdp, dlp->path);
2255 		vlink.content = get_string(hdp, dlp->content);
2256 		vlink.type = attr2type(dlp->attr);
2257 
2258 		if (visit_link(hdp, linkp, &vlink) != DI_WALK_CONTINUE) {
2259 			break;
2260 		}
2261 	}
2262 
2263 	return (linkp->retval);
2264 }
2265 
2266 static int
2267 walk_matching_links(struct di_devlink_handle *hdp, link_desc_t *linkp)
2268 {
2269 	uint32_t nidx;
2270 	struct db_link *dlp;
2271 	struct db_minor *dmp;
2272 
2273 	assert(linkp->minor_path != NULL);
2274 
2275 	dmp = lookup_minor(hdp, linkp->minor_path, NULL, TYPE_DB);
2276 
2277 	/*
2278 	 * If a minor matching the path exists, walk that minor's devlinks list.
2279 	 * Then walk the dangling devlinks list. Non-matching devlinks will be
2280 	 * filtered out in visit_link.
2281 	 */
2282 	for (;;) {
2283 		nidx = dmp ? dmp->link : DB_HDR(hdp)->dngl_idx;
2284 		for (; dlp = get_link(hdp, nidx); nidx = dlp->sib) {
2285 			struct di_devlink vlink = {NULL};
2286 
2287 			vlink.rel_path = get_string(hdp, dlp->path);
2288 			vlink.content = get_string(hdp, dlp->content);
2289 			vlink.type = attr2type(dlp->attr);
2290 
2291 			if (visit_link(hdp, linkp, &vlink) != DI_WALK_CONTINUE)
2292 				goto out;
2293 		}
2294 		if (dmp == NULL) {
2295 			break;
2296 		} else {
2297 			dmp = NULL;
2298 		}
2299 	}
2300 
2301 out:
2302 	return (linkp->retval);
2303 }
2304 
2305 static int
2306 visit_link(
2307 	struct di_devlink_handle *hdp,
2308 	link_desc_t *linkp,
2309 	struct di_devlink *vlp)
2310 {
2311 	struct stat sbuf;
2312 	const char *minor_path = NULL;
2313 	char abs_path[PATH_MAX], cont[PATH_MAX];
2314 
2315 	/*
2316 	 * It is legal for the link's content and type to be unknown.
2317 	 * but one of absolute or relative path must be set.
2318 	 */
2319 	if (vlp->rel_path == NULL && vlp->abs_path == NULL) {
2320 		(void) dprintf(DBG_ERR, "visit_link: invalid arguments\n");
2321 		return (DI_WALK_CONTINUE);
2322 	}
2323 
2324 	if (vlp->rel_path == NULL) {
2325 		vlp->rel_path = (char *)rel_path(hdp, vlp->abs_path);
2326 		if (vlp->rel_path == NULL || vlp->rel_path[0] == '\0')
2327 			return (DI_WALK_CONTINUE);
2328 	}
2329 
2330 	if (linkp->regp) {
2331 		if (regexec(linkp->regp, vlp->rel_path, 0, NULL, 0) != 0)
2332 			return (DI_WALK_CONTINUE);
2333 	}
2334 
2335 	if (vlp->abs_path == NULL) {
2336 		assert(vlp->rel_path[0] != '/');
2337 		(void) snprintf(abs_path, sizeof (abs_path), "%s/%s",
2338 		    hdp->dev_dir, vlp->rel_path);
2339 		vlp->abs_path = abs_path;
2340 	}
2341 
2342 	if (vlp->content == NULL) {
2343 		if (s_readlink(vlp->abs_path, cont, sizeof (cont)) < 0) {
2344 			return (DI_WALK_CONTINUE);
2345 		}
2346 		vlp->content = cont;
2347 	}
2348 
2349 
2350 	if (vlp->type == 0) {
2351 		if (is_minor_node(vlp->content, &minor_path)) {
2352 			vlp->type = DI_PRIMARY_LINK;
2353 		} else {
2354 			vlp->type = DI_SECONDARY_LINK;
2355 		}
2356 	}
2357 
2358 	/*
2359 	 * Filter based on minor path
2360 	 */
2361 	if (linkp->minor_path) {
2362 		char tmp[PATH_MAX];
2363 
2364 		/*
2365 		 * derive minor path
2366 		 */
2367 		if (vlp->type == DI_SECONDARY_LINK) {
2368 
2369 #ifdef	DEBUG
2370 			/*LINTED*/
2371 			assert(sizeof (tmp) >= PATH_MAX);
2372 #endif
2373 			if (s_realpath(vlp->abs_path, tmp) == NULL)
2374 				return (DI_WALK_CONTINUE);
2375 
2376 			if (!is_minor_node(tmp, &minor_path))
2377 				return (DI_WALK_CONTINUE);
2378 
2379 		} else if (minor_path == NULL) {
2380 			if (!is_minor_node(vlp->content, &minor_path))
2381 				return (DI_WALK_CONTINUE);
2382 		}
2383 
2384 		assert(minor_path != NULL);
2385 
2386 		if (strcmp(linkp->minor_path, minor_path) != 0)
2387 			return (DI_WALK_CONTINUE);
2388 	}
2389 
2390 	/*
2391 	 * Filter based on link type
2392 	 */
2393 	if (!TYPE_NONE(linkp->flags) && LINK_TYPE(linkp->flags) != vlp->type) {
2394 		return (DI_WALK_CONTINUE);
2395 	}
2396 
2397 	if (lstat(vlp->abs_path, &sbuf) < 0) {
2398 		dprintf(DBG_ERR, "visit_link: %s: lstat failed: %s\n",
2399 		    vlp->abs_path, strerror(errno));
2400 		return (DI_WALK_CONTINUE);
2401 	}
2402 
2403 	return (linkp->fcn(vlp, linkp->arg));
2404 }
2405 
2406 static int
2407 devlink_valid(di_devlink_t devlink)
2408 {
2409 	if (devlink == NULL || devlink->rel_path == NULL ||
2410 	    devlink->abs_path == NULL || devlink->content == NULL ||
2411 	    TYPE_NONE(devlink->type)) {
2412 		return (0);
2413 	}
2414 
2415 	return (1);
2416 }
2417 
2418 const char *
2419 di_devlink_path(di_devlink_t devlink)
2420 {
2421 	if (!devlink_valid(devlink)) {
2422 		errno = EINVAL;
2423 		return (NULL);
2424 	}
2425 
2426 	return (devlink->abs_path);
2427 }
2428 
2429 const char *
2430 di_devlink_content(di_devlink_t devlink)
2431 {
2432 	if (!devlink_valid(devlink)) {
2433 		errno = EINVAL;
2434 		return (NULL);
2435 	}
2436 
2437 	return (devlink->content);
2438 }
2439 
2440 int
2441 di_devlink_type(di_devlink_t devlink)
2442 {
2443 	if (!devlink_valid(devlink)) {
2444 		errno = EINVAL;
2445 		return (-1);
2446 	}
2447 
2448 	return (devlink->type);
2449 }
2450 
2451 di_devlink_t
2452 di_devlink_dup(di_devlink_t devlink)
2453 {
2454 	struct di_devlink *duplink;
2455 
2456 	if (!devlink_valid(devlink)) {
2457 		errno = EINVAL;
2458 		return (NULL);
2459 	}
2460 
2461 	if ((duplink = calloc(1, sizeof (struct di_devlink))) == NULL) {
2462 		return (NULL);
2463 	}
2464 
2465 	duplink->rel_path = strdup(devlink->rel_path);
2466 	duplink->abs_path = strdup(devlink->abs_path);
2467 	duplink->content  = strdup(devlink->content);
2468 	duplink->type	  = devlink->type;
2469 
2470 	if (!devlink_valid(duplink)) {
2471 		(void) di_devlink_free(duplink);
2472 		errno = ENOMEM;
2473 		return (NULL);
2474 	}
2475 
2476 	return (duplink);
2477 }
2478 
2479 int
2480 di_devlink_free(di_devlink_t devlink)
2481 {
2482 	if (devlink == NULL) {
2483 		errno = EINVAL;
2484 		return (-1);
2485 	}
2486 
2487 	free(devlink->rel_path);
2488 	free(devlink->abs_path);
2489 	free(devlink->content);
2490 	free(devlink);
2491 
2492 	return (0);
2493 }
2494 
2495 /*
2496  * Obtain path relative to dev_dir
2497  */
2498 static const char *
2499 rel_path(struct di_devlink_handle *hdp, const char *path)
2500 {
2501 	const size_t len = strlen(hdp->dev_dir);
2502 
2503 	if (strncmp(path, hdp->dev_dir, len) != 0)
2504 		return (NULL);
2505 
2506 	if (path[len] == '\0')
2507 		return (&path[len]);
2508 
2509 	if (path[len] != '/')
2510 		return (NULL);
2511 
2512 	return (&path[len+1]);
2513 }
2514 
2515 static int
2516 recurse_dev(struct di_devlink_handle *hdp, recurse_t *rp)
2517 {
2518 	int ret = 0;
2519 
2520 	(void) do_recurse(hdp->dev_dir, hdp, rp, &ret);
2521 
2522 	return (ret);
2523 }
2524 
2525 static int
2526 do_recurse(
2527 	const char *dir,
2528 	struct di_devlink_handle *hdp,
2529 	recurse_t *rp,
2530 	int *retp)
2531 {
2532 	size_t len;
2533 	const char *rel;
2534 	struct stat sbuf;
2535 	char cur[PATH_MAX], *cp;
2536 	int i, rv = DI_WALK_CONTINUE;
2537 	finddevhdl_t handle;
2538 	char *d_name;
2539 
2540 
2541 	if ((rel = rel_path(hdp, dir)) == NULL)
2542 		return (DI_WALK_CONTINUE);
2543 
2544 	/*
2545 	 * Skip directories we are not interested in.
2546 	 */
2547 	for (i = 0; i < N_SKIP_DIRS; i++) {
2548 		if (strcmp(rel, skip_dirs[i]) == 0) {
2549 			(void) dprintf(DBG_STEP, "do_recurse: skipping %s\n",
2550 			    dir);
2551 			return (DI_WALK_CONTINUE);
2552 		}
2553 	}
2554 
2555 	(void) dprintf(DBG_STEP, "do_recurse: dir = %s\n", dir);
2556 
2557 	if (finddev_readdir(dir, &handle) != 0)
2558 		return (DI_WALK_CONTINUE);
2559 
2560 	(void) snprintf(cur, sizeof (cur), "%s/", dir);
2561 	len = strlen(cur);
2562 	cp = cur + len;
2563 	len = sizeof (cur) - len;
2564 
2565 	for (;;) {
2566 		if ((d_name = (char *)finddev_next(handle)) == NULL)
2567 			break;
2568 
2569 		if (strlcpy(cp, d_name, len) >= len)
2570 			break;
2571 
2572 		/*
2573 		 * Skip files we are not interested in.
2574 		 */
2575 		for (i = 0; i < N_SKIP_FILES; i++) {
2576 
2577 			rel = rel_path(hdp, cur);
2578 			if (rel == NULL || strcmp(rel, skip_files[i]) == 0) {
2579 				(void) dprintf(DBG_STEP,
2580 				    "do_recurse: skipping %s\n", cur);
2581 				goto next_entry;
2582 			}
2583 		}
2584 
2585 		if (lstat(cur, &sbuf) == 0) {
2586 			if (S_ISDIR(sbuf.st_mode)) {
2587 				rv = do_recurse(cur, hdp, rp, retp);
2588 			} else if (S_ISLNK(sbuf.st_mode)) {
2589 				rv = rp->fcn(hdp, rp->data, cur);
2590 			} else {
2591 				(void) dprintf(DBG_STEP,
2592 				    "do_recurse: Skipping entry: %s\n", cur);
2593 			}
2594 		} else {
2595 			(void) dprintf(DBG_ERR, "do_recurse: cur(%s): lstat"
2596 			    " failed: %s\n", cur, strerror(errno));
2597 		}
2598 
2599 next_entry:
2600 		*cp = '\0';
2601 
2602 		if (rv != DI_WALK_CONTINUE)
2603 			break;
2604 	}
2605 
2606 	finddev_close(handle);
2607 
2608 	return (rv);
2609 }
2610 
2611 
2612 static int
2613 check_attr(uint32_t attr)
2614 {
2615 	switch (attr & A_LINK_TYPES) {
2616 		case A_PRIMARY:
2617 		case A_SECONDARY:
2618 			return (1);
2619 		default:
2620 			dprintf(DBG_ERR, "check_attr: incorrect attr(%u)\n",
2621 			    attr);
2622 			return (0);
2623 	}
2624 }
2625 
2626 static int
2627 attr2type(uint32_t attr)
2628 {
2629 	switch (attr & A_LINK_TYPES) {
2630 		case A_PRIMARY:
2631 			return (DI_PRIMARY_LINK);
2632 		case A_SECONDARY:
2633 			return (DI_SECONDARY_LINK);
2634 		default:
2635 			dprintf(DBG_ERR, "attr2type: incorrect attr(%u)\n",
2636 			    attr);
2637 			return (0);
2638 	}
2639 }
2640 
2641 /* Allocate new node and link it in */
2642 static cache_node_t *
2643 node_insert(
2644 	struct di_devlink_handle *hdp,
2645 	cache_node_t *pcnp,
2646 	const char *path,
2647 	int insert)
2648 {
2649 	cache_node_t *cnp;
2650 
2651 	if (path == NULL) {
2652 		errno = EINVAL;
2653 		SET_DB_ERR(hdp);
2654 		return (NULL);
2655 	}
2656 
2657 	if ((cnp = calloc(1, sizeof (cache_node_t))) == NULL) {
2658 		SET_DB_ERR(hdp);
2659 		return (NULL);
2660 	}
2661 
2662 	if ((cnp->path = strdup(path)) == NULL) {
2663 		SET_DB_ERR(hdp);
2664 		free(cnp);
2665 		return (NULL);
2666 	}
2667 
2668 	cnp->parent = pcnp;
2669 
2670 	if (pcnp == NULL) {
2671 		assert(strcmp(path, "/") == 0);
2672 		assert(CACHE(hdp)->root == NULL);
2673 		CACHE(hdp)->root = cnp;
2674 	} else if (insert == INSERT_HEAD) {
2675 		cnp->sib = pcnp->child;
2676 		pcnp->child = cnp;
2677 	} else if (CACHE_LAST(hdp) && CACHE_LAST(hdp)->node &&
2678 	    CACHE_LAST(hdp)->node->parent == pcnp &&
2679 	    CACHE_LAST(hdp)->node->sib == NULL) {
2680 
2681 		CACHE_LAST(hdp)->node->sib = cnp;
2682 
2683 	} else {
2684 		cache_node_t **pp;
2685 
2686 		for (pp = &pcnp->child; *pp != NULL; pp = &(*pp)->sib)
2687 			;
2688 		*pp = cnp;
2689 	}
2690 
2691 	return (cnp);
2692 }
2693 
2694 /*
2695  * Allocate a new minor and link it in either at the tail or head
2696  * of the minor list depending on the value of "prev".
2697  */
2698 static cache_minor_t *
2699 minor_insert(
2700 	struct di_devlink_handle *hdp,
2701 	cache_node_t *pcnp,
2702 	const char *name,
2703 	const char *nodetype,
2704 	cache_minor_t **prev)
2705 {
2706 	cache_minor_t *cmnp;
2707 
2708 	if (pcnp == NULL || name == NULL) {
2709 		errno = EINVAL;
2710 		SET_DB_ERR(hdp);
2711 		return (NULL);
2712 	}
2713 
2714 	/*
2715 	 * Some pseudo drivers don't specify nodetype. Assume pseudo if
2716 	 * nodetype is not specified.
2717 	 */
2718 	if (nodetype == NULL)
2719 		nodetype = DDI_PSEUDO;
2720 
2721 	if ((cmnp = calloc(1, sizeof (cache_minor_t))) == NULL) {
2722 		SET_DB_ERR(hdp);
2723 		return (NULL);
2724 	}
2725 
2726 	cmnp->name = strdup(name);
2727 	cmnp->nodetype = strdup(nodetype);
2728 	if (cmnp->name == NULL || cmnp->nodetype == NULL) {
2729 		SET_DB_ERR(hdp);
2730 		free(cmnp->name);
2731 		free(cmnp->nodetype);
2732 		free(cmnp);
2733 		return (NULL);
2734 	}
2735 
2736 	cmnp->node = pcnp;
2737 
2738 	/* Add to node's minor list */
2739 	if (prev == NULL) {
2740 		cmnp->sib = pcnp->minor;
2741 		pcnp->minor = cmnp;
2742 	} else {
2743 		assert(*prev == NULL);
2744 		*prev = cmnp;
2745 	}
2746 
2747 	return (cmnp);
2748 }
2749 
2750 static cache_link_t *
2751 link_insert(
2752 	struct di_devlink_handle *hdp,
2753 	cache_minor_t *cmnp,
2754 	const char *path,
2755 	const char *content,
2756 	uint32_t attr)
2757 {
2758 	cache_link_t *clp;
2759 
2760 	if (path == NULL || content == NULL || !check_attr(attr)) {
2761 		errno = EINVAL;
2762 		SET_DB_ERR(hdp);
2763 		return (NULL);
2764 	}
2765 
2766 	if ((clp = calloc(1, sizeof (cache_link_t))) == NULL) {
2767 		SET_DB_ERR(hdp);
2768 		return (NULL);
2769 	}
2770 
2771 	clp->path = strdup(path);
2772 	clp->content = strdup(content);
2773 	if (clp->path == NULL || clp->content == NULL) {
2774 		SET_DB_ERR(hdp);
2775 		link_free(&clp);
2776 		return (NULL);
2777 	}
2778 
2779 	clp->attr = attr;
2780 	hash_insert(hdp, clp);
2781 	clp->minor = cmnp;
2782 
2783 	/* Add to minor's link list */
2784 	if (cmnp != NULL) {
2785 		clp->sib = cmnp->link;
2786 		cmnp->link = clp;
2787 	} else {
2788 		clp->sib = CACHE(hdp)->dngl;
2789 		CACHE(hdp)->dngl = clp;
2790 	}
2791 
2792 	return (clp);
2793 }
2794 
2795 static void
2796 hash_insert(struct di_devlink_handle *hdp, cache_link_t *clp)
2797 {
2798 	uint_t hval;
2799 
2800 	hval = hashfn(hdp, clp->path);
2801 	clp->hash = CACHE_HASH(hdp, hval);
2802 	CACHE_HASH(hdp, hval) = clp;
2803 }
2804 
2805 
2806 static struct db_node *
2807 get_node(struct di_devlink_handle *hdp, uint32_t idx)
2808 {
2809 	return (map_seg(hdp, idx, PROT_READ, DB_NODE));
2810 }
2811 
2812 static struct db_node *
2813 set_node(struct di_devlink_handle *hdp, uint32_t idx)
2814 {
2815 	return (map_seg(hdp, idx, PROT_READ | PROT_WRITE, DB_NODE));
2816 }
2817 
2818 static struct db_minor *
2819 get_minor(struct di_devlink_handle *hdp, uint32_t idx)
2820 {
2821 	return (map_seg(hdp, idx, PROT_READ, DB_MINOR));
2822 }
2823 
2824 static struct db_minor *
2825 set_minor(struct di_devlink_handle *hdp, uint32_t idx)
2826 {
2827 	return (map_seg(hdp, idx, PROT_READ | PROT_WRITE, DB_MINOR));
2828 }
2829 
2830 static struct db_link *
2831 get_link(struct di_devlink_handle *hdp, uint32_t idx)
2832 {
2833 	return (map_seg(hdp, idx, PROT_READ, DB_LINK));
2834 }
2835 
2836 static struct db_link *
2837 set_link(struct di_devlink_handle *hdp, uint32_t idx)
2838 {
2839 	return (map_seg(hdp, idx, PROT_READ | PROT_WRITE, DB_LINK));
2840 }
2841 
2842 static char *
2843 get_string(struct di_devlink_handle *hdp, uint32_t idx)
2844 {
2845 	return (map_seg(hdp, idx, PROT_READ, DB_STR));
2846 }
2847 
2848 static char *
2849 set_string(struct di_devlink_handle *hdp, uint32_t idx)
2850 {
2851 	return (map_seg(hdp, idx, PROT_READ | PROT_WRITE, DB_STR));
2852 }
2853 
2854 
2855 /*
2856  * Returns the element corresponding to idx. If the portion of file involved
2857  * is not yet mapped, does an mmap() as well. Existing mappings are not changed.
2858  */
2859 static void *
2860 map_seg(
2861 	struct di_devlink_handle *hdp,
2862 	uint32_t idx,
2863 	int prot,
2864 	db_seg_t seg)
2865 {
2866 	int s;
2867 	off_t off;
2868 	size_t slen;
2869 	caddr_t addr;
2870 
2871 	if (idx == DB_NIL) {
2872 		return (NULL);
2873 	}
2874 
2875 	if (!VALID_INDEX(hdp, seg, idx)) {
2876 		(void) dprintf(DBG_ERR, "map_seg: seg(%d): invalid idx(%u)\n",
2877 		    seg, idx);
2878 		return (NULL);
2879 	}
2880 
2881 	/*
2882 	 * If the seg is already mapped in, use it if the access type is
2883 	 * valid.
2884 	 */
2885 	if (DB_SEG(hdp, seg) != NULL) {
2886 		if (DB_SEG_PROT(hdp, seg) != prot) {
2887 			(void) dprintf(DBG_ERR, "map_seg: illegal access: "
2888 			    "seg[%d]: idx=%u, seg_prot=%d, access=%d\n",
2889 			    seg, idx, DB_SEG_PROT(hdp, seg), prot);
2890 			return (NULL);
2891 		}
2892 		return (DB_SEG(hdp, seg) + idx * elem_sizes[seg]);
2893 	}
2894 
2895 	/*
2896 	 * Segment is not mapped. Mmap() the segment.
2897 	 */
2898 	off = seg_size(hdp, DB_HEADER);
2899 	for (s = 0; s < seg; s++) {
2900 		off += seg_size(hdp, s);
2901 	}
2902 	slen = seg_size(hdp, seg);
2903 
2904 	addr = mmap(0, slen, prot, MAP_SHARED, DB(hdp)->db_fd, off);
2905 	if (addr == MAP_FAILED) {
2906 		(void) dprintf(DBG_ERR, "map_seg: seg[%d]: mmap failed: %s\n",
2907 		    seg, strerror(errno));
2908 		(void) dprintf(DBG_ERR, "map_seg: args: len=%lu, prot=%d,"
2909 		    " fd=%d, off=%ld\n", (ulong_t)slen, prot, DB(hdp)->db_fd,
2910 		    off);
2911 		return (NULL);
2912 	}
2913 
2914 	DB_SEG(hdp, seg) = addr;
2915 	DB_SEG_PROT(hdp, seg) = prot;
2916 
2917 	(void) dprintf(DBG_STEP, "map_seg: seg[%d]: len=%lu, prot=%d, fd=%d, "
2918 	    "off=%ld, seg_base=%p\n", seg, (ulong_t)slen, prot, DB(hdp)->db_fd,
2919 	    off, (void *)addr);
2920 
2921 	return (DB_SEG(hdp, seg) + idx * elem_sizes[seg]);
2922 }
2923 
2924 /*
2925  * Computes the size of a segment rounded up to the nearest page boundary.
2926  */
2927 static size_t
2928 seg_size(struct di_devlink_handle *hdp, int seg)
2929 {
2930 	size_t sz;
2931 
2932 	assert(DB_HDR(hdp)->page_sz);
2933 
2934 	if (seg == DB_HEADER) {
2935 		sz = HDR_LEN;
2936 	} else {
2937 		assert(DB_NUM(hdp, seg) >= 1);
2938 		sz = DB_NUM(hdp, seg) * elem_sizes[seg];
2939 	}
2940 
2941 	sz = (sz / DB_HDR(hdp)->page_sz) + 1;
2942 
2943 	sz *= DB_HDR(hdp)->page_sz;
2944 
2945 	return (sz);
2946 }
2947 
2948 static size_t
2949 size_db(struct di_devlink_handle *hdp, long page_sz, uint32_t *count)
2950 {
2951 	int i;
2952 	size_t sz;
2953 	cache_link_t *clp;
2954 
2955 	assert(page_sz > 0);
2956 
2957 	/* Take "NIL" element into account */
2958 	for (i = 0; i < DB_TYPES; i++) {
2959 		count[i] = 1;
2960 	}
2961 
2962 	count_node(CACHE(hdp)->root, count);
2963 
2964 	for (clp = CACHE(hdp)->dngl; clp != NULL; clp = clp->sib) {
2965 		count_link(clp, count);
2966 	}
2967 
2968 	sz = ((HDR_LEN / page_sz) + 1) * page_sz;
2969 	for (i = 0; i < DB_TYPES; i++) {
2970 		assert(count[i] >= 1);
2971 		sz += (((count[i] * elem_sizes[i]) / page_sz) + 1) * page_sz;
2972 		(void) dprintf(DBG_INFO, "N[%u]=%u\n", i, count[i]);
2973 	}
2974 	(void) dprintf(DBG_INFO, "DB size=%lu\n", (ulong_t)sz);
2975 
2976 	return (sz);
2977 }
2978 
2979 
2980 static void
2981 count_node(cache_node_t *cnp, uint32_t *count)
2982 {
2983 	cache_minor_t *cmnp;
2984 
2985 	if (cnp == NULL)
2986 		return;
2987 
2988 	count[DB_NODE]++;
2989 	count_string(cnp->path, count);
2990 
2991 	for (cmnp = cnp->minor; cmnp != NULL; cmnp = cmnp->sib) {
2992 		count_minor(cmnp, count);
2993 	}
2994 
2995 	for (cnp = cnp->child; cnp != NULL; cnp = cnp->sib) {
2996 		count_node(cnp, count);
2997 	}
2998 
2999 }
3000 
3001 static void
3002 count_minor(cache_minor_t *cmnp, uint32_t *count)
3003 {
3004 	cache_link_t *clp;
3005 
3006 	if (cmnp == NULL)
3007 		return;
3008 
3009 	count[DB_MINOR]++;
3010 	count_string(cmnp->name, count);
3011 	count_string(cmnp->nodetype, count);
3012 
3013 	for (clp = cmnp->link; clp != NULL; clp = clp->sib) {
3014 		count_link(clp, count);
3015 	}
3016 }
3017 
3018 static void
3019 count_link(cache_link_t *clp, uint32_t *count)
3020 {
3021 	if (clp == NULL)
3022 		return;
3023 
3024 	count[DB_LINK]++;
3025 	count_string(clp->path, count);
3026 	count_string(clp->content, count);
3027 }
3028 
3029 
3030 static void
3031 count_string(const char *str, uint32_t *count)
3032 {
3033 	if (str == NULL) {
3034 		(void) dprintf(DBG_ERR, "count_string: NULL argument\n");
3035 		return;
3036 	}
3037 
3038 	count[DB_STR] += strlen(str) + 1;
3039 }
3040 
3041 static uint_t
3042 hashfn(struct di_devlink_handle *hdp, const char *str)
3043 {
3044 	const char *cp;
3045 	ulong_t hval = 0;
3046 
3047 	if (str == NULL) {
3048 		return (0);
3049 	}
3050 
3051 	assert(CACHE(hdp)->hash_sz >= MIN_HASH_SIZE);
3052 
3053 	for (cp = str; *cp != '\0'; cp++) {
3054 		hval += *cp;
3055 	}
3056 
3057 	return (hval % CACHE(hdp)->hash_sz);
3058 }
3059 
3060 /*
3061  * enter_db_lock()
3062  *
3063  * If the handle is IS_RDWR then we lock as writer to "update" database,
3064  * if IS_RDONLY then we lock as reader to "snapshot" database. The
3065  * implementation uses advisory file locking.
3066  *
3067  * This function returns:
3068  *   == 1	success and grabbed the lock file, we can open the DB.
3069  *   == 0	success but did not lock the lock file,	reader must walk
3070  *		the /dev directory.
3071  *   == -1	failure.
3072  */
3073 static int
3074 enter_db_lock(struct di_devlink_handle *hdp, const char *root_dir)
3075 {
3076 	int		fd;
3077 	struct flock	lock;
3078 	char		lockfile[PATH_MAX];
3079 	int		rv;
3080 	int		writer = HDL_RDWR(hdp);
3081 	static int	did_sync = 0;
3082 	int		eintrs;
3083 
3084 	assert(hdp->lock_fd < 0);
3085 
3086 	get_db_path(hdp, DB_LOCK, lockfile, sizeof (lockfile));
3087 
3088 	dprintf(DBG_LCK, "enter_db_lock: %s BEGIN\n",
3089 	    writer ? "update" : "snapshot");
3090 
3091 	/* Record locks are per-process. Protect against multiple threads. */
3092 	(void) mutex_lock(&update_mutex);
3093 
3094 again:	if ((fd = open(lockfile,
3095 	    (writer ? (O_RDWR|O_CREAT) : O_RDONLY), DB_LOCK_PERMS)) < 0) {
3096 		/*
3097 		 * Typically the lock file and the database go hand in hand.
3098 		 * If we find that the lock file does not exist (for some
3099 		 * unknown reason) and we are the reader then we return
3100 		 * success (after triggering devfsadm to create the file and
3101 		 * a retry) so that we can still provide service via slow
3102 		 * /dev walk.  If we get a failure as a writer we want the
3103 		 * error to manifests itself.
3104 		 */
3105 		if ((errno == ENOENT) && !writer) {
3106 			/* If reader, signal once to get files created */
3107 			if (did_sync == 0) {
3108 				did_sync = 1;
3109 				dprintf(DBG_LCK, "enter_db_lock: %s OSYNC\n",
3110 				    writer ? "update" : "snapshot");
3111 
3112 				/* signal to get files created */
3113 				(void) devlink_create(root_dir, NULL,
3114 				    DCA_DEVLINK_SYNC);
3115 				goto again;
3116 			}
3117 			dprintf(DBG_LCK, "enter_db_lock: %s OPENFAILD %s: "
3118 			    "WALK\n", writer ? "update" : "snapshot",
3119 			    strerror(errno));
3120 			(void) mutex_unlock(&update_mutex);
3121 			return (0);		/* success, but not locked */
3122 		} else {
3123 			dprintf(DBG_LCK, "enter_db_lock: %s OPENFAILD %s\n",
3124 			    writer ? "update" : "snapshot", strerror(errno));
3125 			(void) mutex_unlock(&update_mutex);
3126 			return (-1);		/* failed */
3127 		}
3128 	}
3129 
3130 	lock.l_type = writer ? F_WRLCK : F_RDLCK;
3131 	lock.l_whence = SEEK_SET;
3132 	lock.l_start = 0;
3133 	lock.l_len = 0;
3134 
3135 	/* Enter the lock. */
3136 	for (eintrs = 0; eintrs < MAX_LOCK_RETRY; eintrs++) {
3137 		rv = fcntl(fd, F_SETLKW, &lock);
3138 		if ((rv != -1) || (errno != EINTR))
3139 			break;
3140 	}
3141 
3142 	if (rv != -1) {
3143 		hdp->lock_fd = fd;
3144 		dprintf(DBG_LCK, "enter_db_lock: %s LOCKED\n",
3145 		    writer ? "update" : "snapshot");
3146 		return (1);		/* success, locked */
3147 	}
3148 
3149 	(void) close(fd);
3150 	dprintf(DBG_ERR, "enter_db_lock: %s FAILED: %s: WALK\n",
3151 	    writer ? "update" : "snapshot", strerror(errno));
3152 	(void) mutex_unlock(&update_mutex);
3153 	return (-1);
3154 }
3155 
3156 /*
3157  * Close and re-open lock file every time so that it is recreated if deleted.
3158  */
3159 static void
3160 exit_db_lock(struct di_devlink_handle *hdp)
3161 {
3162 	struct flock	unlock;
3163 	int		writer = HDL_RDWR(hdp);
3164 
3165 	if (hdp->lock_fd < 0) {
3166 		return;
3167 	}
3168 
3169 	unlock.l_type = F_UNLCK;
3170 	unlock.l_whence = SEEK_SET;
3171 	unlock.l_start = 0;
3172 	unlock.l_len = 0;
3173 
3174 	dprintf(DBG_LCK, "exit_db_lock : %s UNLOCKED\n",
3175 	    writer ? "update" : "snapshot");
3176 	if (fcntl(hdp->lock_fd, F_SETLK, &unlock) == -1) {
3177 		dprintf(DBG_ERR, "exit_db_lock : %s failed: %s\n",
3178 		    writer ? "update" : "snapshot", strerror(errno));
3179 	}
3180 
3181 	(void) close(hdp->lock_fd);
3182 
3183 	hdp->lock_fd = -1;
3184 
3185 	(void) mutex_unlock(&update_mutex);
3186 }
3187 
3188 /*
3189  * returns 1 if contents is a minor node in /devices.
3190  * If mn_root is not NULL, mn_root is set to:
3191  *	if contents is a /dev node, mn_root = contents
3192  *			OR
3193  *	if contents is a /devices node, mn_root set to the '/'
3194  *	following /devices.
3195  */
3196 int
3197 is_minor_node(const char *contents, const char **mn_root)
3198 {
3199 	char *ptr, *prefix;
3200 
3201 	prefix = "../devices/";
3202 
3203 	if ((ptr = strstr(contents, prefix)) != NULL) {
3204 
3205 		/* mn_root should point to the / following /devices */
3206 		if (mn_root != NULL) {
3207 			*mn_root = ptr += strlen(prefix) - 1;
3208 		}
3209 		return (1);
3210 	}
3211 
3212 	prefix = "/devices/";
3213 
3214 	if (strncmp(contents, prefix, strlen(prefix)) == 0) {
3215 
3216 		/* mn_root should point to the / following /devices/ */
3217 		if (mn_root != NULL) {
3218 			*mn_root = contents + strlen(prefix) - 1;
3219 		}
3220 		return (1);
3221 	}
3222 
3223 	if (mn_root != NULL) {
3224 		*mn_root = contents;
3225 	}
3226 	return (0);
3227 }
3228 
3229 static int
3230 s_readlink(const char *link, char *buf, size_t blen)
3231 {
3232 	int rv;
3233 
3234 	if ((rv = readlink(link, buf, blen)) == -1)
3235 		goto bad;
3236 
3237 	if (rv >= blen && buf[blen - 1] != '\0') {
3238 		errno = ENAMETOOLONG;
3239 		goto bad;
3240 	} else if (rv < blen) {
3241 		buf[rv] = '\0';
3242 	}
3243 
3244 	return (0);
3245 bad:
3246 	dprintf(DBG_ERR, "s_readlink: %s: failed: %s\n",
3247 	    link, strerror(errno));
3248 	return (-1);
3249 }
3250 
3251 /*
3252  * Synchronous link creation interface routines
3253  * The scope of the operation is determined by the "name" arg.
3254  * "name" can be NULL, a driver name or a devfs pathname (without /devices)
3255  *
3256  *	"name"				creates
3257  *	======				=======
3258  *
3259  *	NULL		=>		All devlinks in system
3260  *	<driver>	=>		devlinks for named driver
3261  *	/pci@1		=>		devlinks for subtree rooted at pci@1
3262  *	/pseudo/foo@0:X	=>		devlinks for minor X
3263  *
3264  * devlink_create() returns 0 on success or an errno value on failure
3265  */
3266 
3267 #define	MAX_DAEMON_ATTEMPTS 2
3268 
3269 static int
3270 devlink_create(const char *root, const char *name, int dca_devlink_flag)
3271 {
3272 	int i;
3273 	int install;
3274 	struct dca_off dca;
3275 
3276 	assert(root);
3277 
3278 	/*
3279 	 * Convert name into arg for door_call
3280 	 */
3281 	if (dca_init(name, &dca, dca_devlink_flag) != 0)
3282 		return (EINVAL);
3283 
3284 	/*
3285 	 * Attempt to use the daemon first
3286 	 */
3287 	i = 0;
3288 	do {
3289 		install = daemon_call(root, &dca);
3290 
3291 		dprintf(DBG_INFO, "daemon_call() retval=%d\n", dca.dca_error);
3292 
3293 		/*
3294 		 * Retry only if door server isn't running
3295 		 */
3296 		if (dca.dca_error != ENOENT && dca.dca_error != EBADF) {
3297 			return (dca.dca_error);
3298 		}
3299 
3300 		dca.dca_error = 0;
3301 
3302 		/*
3303 		 * To improve performance defer this check until the first
3304 		 * failure. Safe to defer as door server checks perms.
3305 		 */
3306 		if (geteuid() != 0)
3307 			return (EPERM);
3308 	/*
3309 	 * Daemon may not be running. Try to start it.
3310 	 */
3311 	} while ((++i < MAX_DAEMON_ATTEMPTS) &&
3312 	    start_daemon(root, install) == 0);
3313 
3314 	dprintf(DBG_INFO, "devlink_create: can't start daemon\n");
3315 
3316 	assert(dca.dca_error == 0);
3317 
3318 	/*
3319 	 * If the daemon cannot be started execute the devfsadm command.
3320 	 */
3321 	exec_cmd(root, &dca);
3322 
3323 	return (dca.dca_error);
3324 }
3325 
3326 /*
3327  * The "name" member of "struct dca" contains data in the following order
3328  *	root'\0'minor'\0'driver'\0'
3329  * The root component is always present at offset 0 in the "name" field.
3330  * The driver and minor are optional. If present they have a non-zero
3331  * offset in the "name" member.
3332  */
3333 static int
3334 dca_init(const char *name, struct dca_off *dcp, int dca_flags)
3335 {
3336 	char *cp;
3337 
3338 	dcp->dca_root = 0;
3339 	dcp->dca_minor = 0;
3340 	dcp->dca_driver = 0;
3341 	dcp->dca_error = 0;
3342 	dcp->dca_flags = dca_flags;
3343 	dcp->dca_name[0] = '\0';
3344 
3345 	name = name ? name : "/";
3346 
3347 	/*
3348 	 *  Check if name is a driver name
3349 	 */
3350 	if (*name != '/') {
3351 		(void) snprintf(dcp->dca_name, sizeof (dcp->dca_name),
3352 		    "/ %s", name);
3353 		dcp->dca_root = 0;
3354 		*(dcp->dca_name + 1) = '\0';
3355 		dcp->dca_driver = 2;
3356 		return (0);
3357 	}
3358 
3359 	(void) snprintf(dcp->dca_name, sizeof (dcp->dca_name), "%s", name);
3360 
3361 	/*
3362 	 * "/devices" not allowed in devfs pathname
3363 	 */
3364 	if (is_minor_node(name, NULL))
3365 		return (-1);
3366 
3367 	dcp->dca_root = 0;
3368 	if (cp = strrchr(dcp->dca_name, ':')) {
3369 		*cp++ = '\0';
3370 		dcp->dca_minor = cp - dcp->dca_name;
3371 	}
3372 
3373 	return (0);
3374 }
3375 
3376 
3377 #define	DAEMON_STARTUP_TIME	1 /* 1 second. This may need to be adjusted */
3378 #define	DEVNAME_CHECK_FILE	"/etc/devname_check_RDONLY"
3379 
3380 static int
3381 daemon_call(const char *root, struct dca_off *dcp)
3382 {
3383 	door_arg_t	arg;
3384 	int		fd, door_error;
3385 	sigset_t	oset, nset;
3386 	char		synch_door[PATH_MAX];
3387 	struct stat	sb;
3388 	char		*prefix;
3389 	int		rofd;
3390 	int		rdonly;
3391 	int		install = 0;
3392 
3393 	/*
3394 	 * If root is readonly, there are two possibilities:
3395 	 *	- we are in some sort of install scenario
3396 	 *	- we are early in boot
3397 	 * If the latter we don't want daemon_call()  to succeed.
3398 	 * else we want to use /tmp/etc/dev
3399 	 *
3400 	 * Both of these requrements are fulfilled if we check for
3401 	 * for a root owned door file in /tmp/etc/dev. If we are
3402 	 * early in boot, the door file won't exist, so this call
3403 	 * will fail.
3404 	 *
3405 	 * If we are in install, the door file will be present.
3406 	 *
3407 	 * If root is read-only, try only once, since libdevinfo
3408 	 * isn't capable of starting devfsadmd correctly in that
3409 	 * situation.
3410 	 *
3411 	 * Don't use statvfs() to check for readonly roots since it
3412 	 * doesn't always report the truth.
3413 	 */
3414 	rofd = -1;
3415 	rdonly = 0;
3416 	if ((rofd = open(DEVNAME_CHECK_FILE, O_WRONLY|O_CREAT|O_TRUNC, 0644))
3417 	    == -1 && errno == EROFS) {
3418 		rdonly = 1;
3419 		prefix = "/tmp";
3420 	} else {
3421 		if (rofd != -1) {
3422 			(void) close(rofd);
3423 			(void) unlink(DEVNAME_CHECK_FILE);
3424 		}
3425 		prefix = (char *)root;
3426 	}
3427 
3428 	if (rdonly && stat(DEVNAME_CHECK_FILE, &sb) != -1)
3429 		install = 1;
3430 
3431 	(void) snprintf(synch_door, sizeof (synch_door),
3432 	    "%s/etc/dev/%s", prefix, DEVFSADM_SYNCH_DOOR);
3433 
3434 	/*
3435 	 * Return ENOTSUP to prevent retries if root is readonly
3436 	 */
3437 	if (stat(synch_door, &sb) == -1 || sb.st_uid != 0) {
3438 		if (rdonly)
3439 			dcp->dca_error = ENOTSUP;
3440 		else
3441 			dcp->dca_error = ENOENT;
3442 		dprintf(DBG_ERR, "stat failed: %s: no file or not root owned\n",
3443 		    synch_door);
3444 		return (install);
3445 	}
3446 
3447 	if ((fd = open(synch_door, O_RDONLY)) == -1) {
3448 		dcp->dca_error = errno;
3449 		dprintf(DBG_ERR, "open of %s failed: %s\n",
3450 		    synch_door, strerror(errno));
3451 		return (install);
3452 	}
3453 
3454 	arg.data_ptr = (char *)dcp;
3455 	arg.data_size = sizeof (*dcp);
3456 	arg.desc_ptr = NULL;
3457 	arg.desc_num = 0;
3458 	arg.rbuf = (char *)dcp;
3459 	arg.rsize = sizeof (*dcp);
3460 
3461 	/*
3462 	 * Block signals to this thread until door call
3463 	 * completes.
3464 	 */
3465 	(void) sigfillset(&nset);
3466 	(void) sigemptyset(&oset);
3467 	(void) sigprocmask(SIG_SETMASK, &nset, &oset);
3468 	if (door_call(fd, &arg)) {
3469 		door_error = 1;
3470 		dcp->dca_error = errno;
3471 	}
3472 	(void) sigprocmask(SIG_SETMASK, &oset, NULL);
3473 
3474 	(void) close(fd);
3475 
3476 	if (door_error)
3477 		return (install);
3478 
3479 	assert(arg.data_ptr);
3480 
3481 	/*LINTED*/
3482 	dcp->dca_error = ((struct dca_off *)arg.data_ptr)->dca_error;
3483 
3484 	/*
3485 	 * The doors interface may return data in a different buffer
3486 	 * If that happens, deallocate buffer via munmap()
3487 	 */
3488 	if (arg.rbuf != (char *)dcp)
3489 		(void) munmap(arg.rbuf, arg.rsize);
3490 
3491 	return (install);
3492 }
3493 
3494 #define	DEVFSADM_PATH	"/usr/sbin/devfsadm"
3495 #define	DEVFSADM	"devfsadm"
3496 
3497 #define	DEVFSADMD_PATH	"/usr/lib/devfsadm/devfsadmd"
3498 #define	DEVFSADM_DAEMON	"devfsadmd"
3499 
3500 static int
3501 start_daemon(const char *root, int install)
3502 {
3503 	int rv, i = 0;
3504 	char *argv[20];
3505 
3506 	argv[i++] = DEVFSADM_DAEMON;
3507 	if (install) {
3508 		argv[i++] = "-a";
3509 		argv[i++] = "/tmp";
3510 		argv[i++] = "-p";
3511 		argv[i++] = "/tmp/root/etc/path_to_inst";
3512 	} else if (strcmp(root, "/")) {
3513 		argv[i++] = "-r";
3514 		argv[i++] = (char *)root;
3515 	}
3516 	argv[i++] = NULL;
3517 
3518 	rv = do_exec(DEVFSADMD_PATH, argv);
3519 
3520 	(void) sleep(DAEMON_STARTUP_TIME);
3521 
3522 	return (rv);
3523 }
3524 
3525 static void
3526 exec_cmd(const char *root, struct dca_off *dcp)
3527 {
3528 	int i;
3529 	char *argv[20];
3530 
3531 	i = 0;
3532 	argv[i++] = DEVFSADM;
3533 
3534 	/*
3535 	 * Load drivers only if -i is specified
3536 	 */
3537 	if (dcp->dca_driver) {
3538 		argv[i++] = "-i";
3539 		argv[i++] = &dcp->dca_name[dcp->dca_driver];
3540 	} else {
3541 		argv[i++] = "-n";
3542 	}
3543 
3544 	if (root != NULL && strcmp(root, "/") != 0) {
3545 		argv[i++] = "-r";
3546 		argv[i++] = (char *)root;
3547 	}
3548 
3549 	argv[i] = NULL;
3550 
3551 	if (do_exec(DEVFSADM_PATH, argv))
3552 		dcp->dca_error = errno;
3553 }
3554 
3555 static int
3556 do_exec(const char *path, char *const argv[])
3557 {
3558 	int i;
3559 	pid_t cpid;
3560 
3561 #ifdef	DEBUG
3562 	dprintf(DBG_INFO, "Executing %s\n\tArgument list:", path);
3563 	for (i = 0; argv[i] != NULL; i++) {
3564 		dprintf(DBG_INFO, " %s", argv[i]);
3565 	}
3566 	dprintf(DBG_INFO, "\n");
3567 #endif
3568 
3569 	if ((cpid = fork1()) == -1) {
3570 		dprintf(DBG_ERR, "fork1 failed: %s\n", strerror(errno));
3571 		return (-1);
3572 	}
3573 
3574 	if (cpid == 0) { /* child process */
3575 		int fd;
3576 
3577 		if ((fd = open("/dev/null", O_RDWR)) >= 0) {
3578 			(void) dup2(fd, fileno(stdout));
3579 			(void) dup2(fd, fileno(stderr));
3580 			(void) close(fd);
3581 
3582 			(void) execv(path, argv);
3583 		} else {
3584 			dprintf(DBG_ERR, "open of /dev/null failed: %s\n",
3585 			    strerror(errno));
3586 		}
3587 
3588 		_exit(-1);
3589 	}
3590 
3591 	/* Parent process */
3592 	if (waitpid(cpid, &i, 0) == cpid) {
3593 		if (WIFEXITED(i)) {
3594 			if (WEXITSTATUS(i) == 0) {
3595 				dprintf(DBG_STEP,
3596 				    "do_exec: child exited normally\n");
3597 				return (0);
3598 			} else
3599 				errno = EINVAL;
3600 		} else {
3601 			/*
3602 			 * The child was interrupted by a signal
3603 			 */
3604 			errno = EINTR;
3605 		}
3606 		dprintf(DBG_ERR, "child terminated abnormally: %s\n",
3607 		    strerror(errno));
3608 	} else {
3609 		dprintf(DBG_ERR, "waitpid failed: %s\n", strerror(errno));
3610 	}
3611 
3612 	return (-1);
3613 }
3614 
3615 static int
3616 walk_cache_links(di_devlink_handle_t hdp, cache_link_t *clp, link_desc_t *linkp)
3617 {
3618 	int i;
3619 
3620 	assert(HDL_RDWR(hdp) || HDL_RDONLY(hdp));
3621 
3622 	dprintf(DBG_INFO, "walk_cache_links: initial link: %s\n",
3623 	    clp ? clp->path : "<NULL>");
3624 
3625 	/*
3626 	 * First search the links under the specified minor. On the
3627 	 * 2nd pass, search the dangling list - secondary links may
3628 	 * exist on this list since they are not resolved during the
3629 	 * /dev walk.
3630 	 */
3631 	for (i = 0; i < 2; i++) {
3632 		for (; clp != NULL; clp = clp->sib) {
3633 			struct di_devlink vlink = {NULL};
3634 
3635 			assert(clp->path[0] != '/');
3636 
3637 			vlink.rel_path = clp->path;
3638 			vlink.content = clp->content;
3639 			vlink.type = attr2type(clp->attr);
3640 
3641 			if (visit_link(hdp, linkp, &vlink)
3642 			    != DI_WALK_CONTINUE) {
3643 				dprintf(DBG_INFO, "walk_cache_links: "
3644 				    "terminating at link: %s\n", clp->path);
3645 				goto out;
3646 			}
3647 		}
3648 
3649 		clp = CACHE(hdp)->dngl;
3650 	}
3651 
3652 out:
3653 
3654 	/* If i < 2, we terminated the walk prematurely */
3655 	return (i < 2 ? DI_WALK_TERMINATE : DI_WALK_CONTINUE);
3656 }
3657 
3658 static void
3659 walk_all_cache(di_devlink_handle_t hdp, link_desc_t *linkp)
3660 {
3661 	int i;
3662 	cache_link_t *clp;
3663 
3664 	dprintf(DBG_INFO, "walk_all_cache: entered\n");
3665 
3666 	for (i = 0; i < CACHE(hdp)->hash_sz; i++) {
3667 		clp = CACHE_HASH(hdp, i);
3668 		for (; clp; clp = clp->hash) {
3669 			struct di_devlink vlink = {NULL};
3670 
3671 			assert(clp->path[0] != '/');
3672 
3673 			vlink.rel_path = clp->path;
3674 			vlink.content = clp->content;
3675 			vlink.type = attr2type(clp->attr);
3676 			if (visit_link(hdp, linkp, &vlink) !=
3677 			    DI_WALK_CONTINUE) {
3678 				dprintf(DBG_INFO, "walk_all_cache: terminating "
3679 				    "walk at link: %s\n", clp->path);
3680 				return;
3681 			}
3682 		}
3683 	}
3684 }
3685 
3686 static void
3687 walk_cache_minor(di_devlink_handle_t hdp, const char *mpath, link_desc_t *linkp)
3688 {
3689 	cache_minor_t *cmnp;
3690 
3691 	assert(mpath);
3692 
3693 	if ((cmnp = lookup_minor(hdp, mpath, NULL, TYPE_CACHE)) != NULL) {
3694 		(void) walk_cache_links(hdp, cmnp->link, linkp);
3695 	} else {
3696 		dprintf(DBG_ERR, "lookup minor failed: %s\n", mpath);
3697 	}
3698 }
3699 
3700 static void
3701 walk_cache_node(di_devlink_handle_t hdp, const char *path, link_desc_t *linkp)
3702 {
3703 	cache_minor_t *cmnp;
3704 	cache_node_t *cnp;
3705 
3706 	assert(path);
3707 
3708 	if ((cnp = lookup_node(hdp, (char *)path, TYPE_CACHE)) == NULL) {
3709 		dprintf(DBG_ERR, "lookup node failed: %s\n", path);
3710 		return;
3711 	}
3712 
3713 	for (cmnp = cnp->minor; cmnp != NULL; cmnp = cmnp->sib) {
3714 		if (walk_cache_links(hdp, cmnp->link, linkp)
3715 		    == DI_WALK_TERMINATE)
3716 			break;
3717 	}
3718 }
3719 
3720 /*
3721  * Private function
3722  *
3723  * Walk cached links corresponding to the given path.
3724  *
3725  * path		path to a node or minor node.
3726  *
3727  * flags	specifies the type of devlinks to be selected.
3728  *		If DI_PRIMARY_LINK is used, only primary links are selected.
3729  *		If DI_SECONDARY_LINK is specified, only secondary links
3730  *		are selected.
3731  *		If neither flag is specified, all devlinks are selected.
3732  *
3733  * re		An extended regular expression in regex(5) format which
3734  *		selects the /dev links to be returned. The regular
3735  *		expression should use link pathnames relative to
3736  *		/dev. i.e. without the leading "/dev/" prefix.
3737  *		A NULL value matches all devlinks.
3738  */
3739 int
3740 di_devlink_cache_walk(di_devlink_handle_t hdp,
3741 	const char *re,
3742 	const char *path,
3743 	uint_t flags,
3744 	void *arg,
3745 	int (*devlink_callback)(di_devlink_t, void *))
3746 {
3747 	regex_t reg;
3748 	link_desc_t linkd = {NULL};
3749 
3750 	if (hdp == NULL || path == NULL || !link_flag(flags) ||
3751 	    !HDL_RDWR(hdp) || devlink_callback == NULL) {
3752 		errno = EINVAL;
3753 		return (-1);
3754 	}
3755 
3756 	linkd.flags = flags;
3757 	linkd.arg = arg;
3758 	linkd.fcn = devlink_callback;
3759 
3760 	if (re) {
3761 		if (regcomp(&reg, re, REG_EXTENDED) != 0)
3762 			return (-1);
3763 		linkd.regp = &reg;
3764 	}
3765 
3766 	if (minor_colon(path) == NULL) {
3767 		walk_cache_node(hdp, path, &linkd);
3768 	} else {
3769 		walk_cache_minor(hdp, path, &linkd);
3770 	}
3771 
3772 	if (re)
3773 		regfree(&reg);
3774 
3775 	return (0);
3776 }
3777 
3778 #define	DEBUG_ENV_VAR	"_DEVLINK_DEBUG"
3779 static int _devlink_debug = -1;
3780 
3781 /*
3782  * debug level is initialized to -1.
3783  * On first call into this routine, debug level is set.
3784  * If debug level is zero, debugging msgs are disabled.
3785  */
3786 static void
3787 debug_print(debug_level_t msglevel, const char *fmt, va_list ap)
3788 {
3789 	char	*cp;
3790 	int	save;
3791 
3792 	/*
3793 	 * We shouldn't be here if debug is disabled
3794 	 */
3795 	assert(_devlink_debug != 0);
3796 
3797 	/*
3798 	 * Set debug level on first call into this routine
3799 	 */
3800 	if (_devlink_debug < 0) {
3801 		if ((cp = getenv(DEBUG_ENV_VAR)) == NULL) {
3802 			_devlink_debug = 0;
3803 			return;
3804 		}
3805 
3806 		save = errno;
3807 		errno = 0;
3808 		_devlink_debug = strtol(cp, NULL, 10);
3809 		if (errno != 0 || _devlink_debug < 0)  {
3810 			_devlink_debug = 0;
3811 			errno = save;
3812 			return;
3813 		}
3814 		errno = save;
3815 
3816 		if (!_devlink_debug)
3817 			return;
3818 	}
3819 
3820 	/* debug msgs are enabled */
3821 	assert(_devlink_debug > 0);
3822 
3823 	if (_devlink_debug < msglevel)
3824 		return;
3825 	if ((_devlink_debug == DBG_LCK) && (msglevel != _devlink_debug))
3826 		return;
3827 
3828 	/* Print a distinctive label for error msgs */
3829 	if (msglevel == DBG_ERR) {
3830 		(void) fprintf(stderr, "[ERROR]: ");
3831 	}
3832 
3833 	(void) vfprintf(stderr, fmt, ap);
3834 	(void) fflush(stderr);
3835 }
3836 
3837 /* ARGSUSED */
3838 /* PRINTFLIKE2 */
3839 void
3840 dprintf(debug_level_t msglevel, const char *fmt, ...)
3841 {
3842 	va_list ap;
3843 
3844 	assert(msglevel > 0);
3845 	if (!_devlink_debug)
3846 		return;
3847 
3848 	va_start(ap, fmt);
3849 	debug_print(msglevel, fmt, ap);
3850 	va_end(ap);
3851 }
3852