1 /*	$NetBSD: libdm-common.c,v 1.5 2009/12/05 11:42:24 haad Exp $	*/
2 
3 /*
4  * Copyright (C) 2001-2004 Sistina Software, Inc. All rights reserved.
5  * Copyright (C) 2004-2007 Red Hat, Inc. All rights reserved.
6  *
7  * This file is part of the device-mapper userspace tools.
8  *
9  * This copyrighted material is made available to anyone wishing to use,
10  * modify, copy, or redistribute it subject to the terms and conditions
11  * of the GNU Lesser General Public License v.2.1.
12  *
13  * You should have received a copy of the GNU Lesser General Public License
14  * along with this program; if not, write to the Free Software Foundation,
15  * Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
16  */
17 
18 #include "dmlib.h"
19 #include "libdm-targets.h"
20 #include "libdm-common.h"
21 #ifdef linux
22 #include "kdev_t.h"
23 #endif
24 #include "dm-ioctl.h"
25 
26 #include <stdarg.h>
27 #include <sys/param.h>
28 #include <sys/ioctl.h>
29 #include <fcntl.h>
30 #include <dirent.h>
31 
32 #ifdef UDEV_SYNC_SUPPORT
33 #  include <sys/types.h>
34 #  include <sys/ipc.h>
35 #  include <sys/sem.h>
36 #ifdef HAVE_UDEV_QUEUE_GET_UDEV_IS_ACTIVE
37 #  define LIBUDEV_I_KNOW_THE_API_IS_SUBJECT_TO_CHANGE
38 #  include <libudev.h>
39 #endif
40 #endif
41 
42 #ifdef linux
43 #  include <linux/fs.h>
44 #endif
45 
46 #ifdef HAVE_SELINUX
47 #  include <selinux/selinux.h>
48 #endif
49 
50 #if defined(__NetBSD__) || defined(__DragonFly__)
51 #include "libdm-netbsd.h"
52 #endif
53 
54 #define DEV_DIR "/dev/"
55 
56 static char _dm_dir[PATH_MAX] = DEV_DIR DM_DIR;
57 
58 static int _verbose = 0;
59 
60 #ifdef UDEV_SYNC_SUPPORT
61 static int _udev_running = -1;
62 static int _sync_with_udev = 1;
63 #endif
64 
65 /*
66  * Library users can provide their own logging
67  * function.
68  */
69 
70 static void _default_log_line(int level,
71 	    const char *file __attribute((unused)),
72 	    int line __attribute((unused)), int dm_errno,
73 	    const char *f, va_list ap)
74 {
75 	int use_stderr = level & _LOG_STDERR;
76 
77 	level &= ~_LOG_STDERR;
78 
79 	if (level > _LOG_WARN && !_verbose)
80 		return;
81 
82 	if (level < _LOG_WARN)
83 		vfprintf(stderr, f, ap);
84 	else
85 		vfprintf(use_stderr ? stderr : stdout, f, ap);
86 
87 	if (level < _LOG_WARN)
88 		fprintf(stderr, "\n");
89 	else
90 		fprintf(use_stderr ? stderr : stdout, "\n");
91 }
92 
93 static void _default_log_with_errno(int level,
94 	    const char *file __attribute((unused)),
95 	    int line __attribute((unused)), int dm_errno,
96 	    const char *f, ...)
97 {
98 	va_list ap;
99 
100 	va_start(ap, f);
101 	_default_log_line(level, file, line, dm_errno, f, ap);
102 	va_end(ap);
103 }
104 
105 static void _default_log(int level, const char *file,
106 			 int line, const char *f, ...)
107 {
108 	va_list ap;
109 
110 	va_start(ap, f);
111 	_default_log_line(level, file, line, 0, f, ap);
112 	va_end(ap);
113 }
114 
115 dm_log_fn dm_log = _default_log;
116 dm_log_with_errno_fn dm_log_with_errno = _default_log_with_errno;
117 
118 void dm_log_init(dm_log_fn fn)
119 {
120 	if (fn)
121 		dm_log = fn;
122 	else
123 		dm_log = _default_log;
124 
125 	dm_log_with_errno = _default_log_with_errno;
126 }
127 
128 int dm_log_is_non_default(void)
129 {
130 	return (dm_log == _default_log) ? 0 : 1;
131 }
132 
133 void dm_log_with_errno_init(dm_log_with_errno_fn fn)
134 {
135 	if (fn)
136 		dm_log_with_errno = fn;
137 	else
138 		dm_log_with_errno = _default_log_with_errno;
139 
140 	dm_log = _default_log;
141 }
142 
143 void dm_log_init_verbose(int level)
144 {
145 	_verbose = level;
146 }
147 
148 static void _build_dev_path(char *buffer, size_t len, const char *dev_name)
149 {
150 	/* If there's a /, assume caller knows what they're doing */
151 	if (strchr(dev_name, '/'))
152 		snprintf(buffer, len, "%s", dev_name);
153 	else
154 		snprintf(buffer, len, "%s/%s", _dm_dir, dev_name);
155 }
156 
157 int dm_get_library_version(char *version, size_t size)
158 {
159 	strncpy(version, DM_LIB_VERSION, size);
160 	return 1;
161 }
162 
163 struct dm_task *dm_task_create(int type)
164 {
165 	struct dm_task *dmt = dm_malloc(sizeof(*dmt));
166 
167 	if (!dmt) {
168 		log_error("dm_task_create: malloc(%" PRIsize_t ") failed",
169 			  sizeof(*dmt));
170 		return NULL;
171 	}
172 
173 	if (!dm_check_version()) {
174 		dm_free(dmt);
175 		return NULL;
176 	}
177 
178 	memset(dmt, 0, sizeof(*dmt));
179 
180 	dmt->type = type;
181 	dmt->minor = -1;
182 	dmt->major = -1;
183 	dmt->allow_default_major_fallback = 1;
184 	dmt->uid = DM_DEVICE_UID;
185 	dmt->gid = DM_DEVICE_GID;
186 	dmt->mode = DM_DEVICE_MODE;
187 	dmt->no_open_count = 0;
188 	dmt->read_ahead = DM_READ_AHEAD_AUTO;
189 	dmt->read_ahead_flags = 0;
190 	dmt->event_nr = 0;
191 	dmt->cookie_set = 0;
192 	dmt->query_inactive_table = 0;
193 
194 	return dmt;
195 }
196 
197 /*
198  * Find the name associated with a given device number by scanning _dm_dir.
199  */
200 static char *_find_dm_name_of_device(dev_t st_rdev)
201 {
202 	const char *name;
203 	char path[PATH_MAX];
204 	struct dirent *dirent;
205 	DIR *d;
206 	struct stat buf;
207 	char *new_name = NULL;
208 
209 	if (!(d = opendir(_dm_dir))) {
210 		log_sys_error("opendir", _dm_dir);
211 		return NULL;
212 	}
213 
214 	while ((dirent = readdir(d))) {
215 		name = dirent->d_name;
216 
217 		if (!strcmp(name, ".") || !strcmp(name, ".."))
218 			continue;
219 
220 		if (dm_snprintf(path, sizeof(path), "%s/%s", _dm_dir,
221 				name) == -1) {
222 			log_error("Couldn't create path for %s", name);
223 			continue;
224 		}
225 
226 		if (stat(path, &buf))
227 			continue;
228 
229 		if (buf.st_rdev == st_rdev) {
230 			if (!(new_name = dm_strdup(name)))
231 				log_error("dm_task_set_name: strdup(%s) failed",
232 					  name);
233 			break;
234 		}
235 	}
236 
237 	if (closedir(d))
238 		log_sys_error("closedir", _dm_dir);
239 
240 	return new_name;
241 }
242 
243 int dm_task_set_name(struct dm_task *dmt, const char *name)
244 {
245 	char *pos;
246 	char *new_name = NULL;
247 	char path[PATH_MAX];
248 	struct stat st1, st2;
249 
250 	if (dmt->dev_name) {
251 		dm_free(dmt->dev_name);
252 		dmt->dev_name = NULL;
253 	}
254 
255 	/*
256 	 * Path supplied for existing device?
257 	 */
258 	if ((pos = strrchr(name, '/'))) {
259 		if (dmt->type == DM_DEVICE_CREATE) {
260 			log_error("Name \"%s\" invalid. It contains \"/\".", name);
261 			return 0;
262 		}
263 
264 		if (stat(name, &st1)) {
265 			log_error("Device %s not found", name);
266 			return 0;
267 		}
268 
269 		/*
270 		 * If supplied path points to same device as last component
271 		 * under /dev/mapper, use that name directly.  Otherwise call
272 		 * _find_dm_name_of_device() to scan _dm_dir for a match.
273 		 */
274 		if (dm_snprintf(path, sizeof(path), "%s/%s", _dm_dir,
275 				pos + 1) == -1) {
276 			log_error("Couldn't create path for %s", pos + 1);
277 			return 0;
278 		}
279 
280 		if (!stat(path, &st2) && (st1.st_rdev == st2.st_rdev))
281 			name = pos + 1;
282 		else if ((new_name = _find_dm_name_of_device(st1.st_rdev)))
283 			name = new_name;
284 		else {
285 			log_error("Device %s not found", name);
286 			return 0;
287 		}
288 	}
289 
290 	if (strlen(name) >= DM_NAME_LEN) {
291 		log_error("Name \"%s\" too long", name);
292 		if (new_name)
293 			dm_free(new_name);
294 		return 0;
295 	}
296 
297 	if (new_name)
298 		dmt->dev_name = new_name;
299 	else if (!(dmt->dev_name = dm_strdup(name))) {
300 		log_error("dm_task_set_name: strdup(%s) failed", name);
301 		return 0;
302 	}
303 
304 	return 1;
305 }
306 
307 int dm_task_set_uuid(struct dm_task *dmt, const char *uuid)
308 {
309 	if (dmt->uuid) {
310 		dm_free(dmt->uuid);
311 		dmt->uuid = NULL;
312 	}
313 
314 	if (!(dmt->uuid = dm_strdup(uuid))) {
315 		log_error("dm_task_set_uuid: strdup(%s) failed", uuid);
316 		return 0;
317 	}
318 
319 	return 1;
320 }
321 
322 int dm_task_set_major(struct dm_task *dmt, int major)
323 {
324 	dmt->major = major;
325 	dmt->allow_default_major_fallback = 0;
326 
327 	return 1;
328 }
329 
330 int dm_task_set_minor(struct dm_task *dmt, int minor)
331 {
332 	dmt->minor = minor;
333 
334 	return 1;
335 }
336 
337 int dm_task_set_major_minor(struct dm_task *dmt, int major, int minor,
338 			    int allow_default_major_fallback)
339 {
340 	dmt->major = major;
341 	dmt->minor = minor;
342 	dmt->allow_default_major_fallback = allow_default_major_fallback;
343 
344 	return 1;
345 }
346 
347 int dm_task_set_uid(struct dm_task *dmt, uid_t uid)
348 {
349 	dmt->uid = uid;
350 
351 	return 1;
352 }
353 
354 int dm_task_set_gid(struct dm_task *dmt, gid_t gid)
355 {
356 	dmt->gid = gid;
357 
358 	return 1;
359 }
360 
361 int dm_task_set_mode(struct dm_task *dmt, mode_t mode)
362 {
363 	dmt->mode = mode;
364 
365 	return 1;
366 }
367 
368 int dm_task_add_target(struct dm_task *dmt, uint64_t start, uint64_t size,
369 		       const char *ttype, const char *params)
370 {
371 	struct target *t = create_target(start, size, ttype, params);
372 
373 	if (!t)
374 		return 0;
375 
376 	if (!dmt->head)
377 		dmt->head = dmt->tail = t;
378 	else {
379 		dmt->tail->next = t;
380 		dmt->tail = t;
381 	}
382 
383 	return 1;
384 }
385 
386 int dm_set_selinux_context(const char *path, mode_t mode)
387 {
388 #ifdef HAVE_SELINUX
389 	security_context_t scontext;
390 
391 	if (is_selinux_enabled() <= 0)
392 		return 1;
393 
394 	if (matchpathcon(path, mode, &scontext) < 0) {
395 		log_error("%s: matchpathcon %07o failed: %s", path, mode,
396 			  strerror(errno));
397 		return 0;
398 	}
399 
400 	log_debug("Setting SELinux context for %s to %s.", path, scontext);
401 
402 	if ((lsetfilecon(path, scontext) < 0) && (errno != ENOTSUP)) {
403 		log_sys_error("lsetfilecon", path);
404 		freecon(scontext);
405 		return 0;
406 	}
407 
408 	freecon(scontext);
409 #endif
410 	return 1;
411 }
412 
413 static int _add_dev_node(const char *dev_name, uint32_t major, uint32_t minor,
414 			 uid_t uid, gid_t gid, mode_t mode, int check_udev)
415 {
416 	char path[PATH_MAX];
417 	struct stat info;
418 	dev_t dev = MKDEV(major, minor);
419 	mode_t old_mask;
420 
421 	#ifdef __NetBSD__
422 	char rpath[PATH_MAX];
423 	uint32_t raw_major;
424 	dev_t rdev;
425 	char raw_devname[DM_NAME_LEN+1]; /* r + other device name */
426 
427 	nbsd_get_dm_major(&raw_major,DM_CHAR_MAJOR);
428 	rdev = MKDEV(raw_major,minor);
429 
430 	snprintf(raw_devname,sizeof(raw_devname),"r%s",dev_name);
431 
432 	_build_dev_path(rpath, sizeof(rpath), raw_devname);
433 
434 	if (stat(rpath, &info) >= 0) {
435 		if (!S_ISCHR(info.st_mode)) {
436 			log_error("A non-raw device file at '%s' "
437 			    "is already present", rpath);
438 			return 0;
439 		}
440 
441 		/* If right inode already exists we don't touch uid etc. */
442 		if (info.st_rdev == rdev)
443 			return 1;
444 
445 		if (unlink(rpath) < 0) {
446 			log_error("Unable to unlink device node for '%s'",
447 			    raw_devname);
448 			return 0;
449 		}
450 	}
451 
452 	old_mask = umask(0);
453 
454 	if (mknod(rpath, S_IFCHR | mode, rdev) < 0) {
455 		log_error("Unable to make device node for '%s'", raw_devname);
456 		return 0;
457 	}
458 #elif defined(__DragonFly__)
459 	_build_dev_path(path, sizeof(path), dev_name);
460 	log_error("Skipping all sanity checks here");
461 	return 1;
462 #endif
463 
464 	_build_dev_path(path, sizeof(path), dev_name);
465 
466 	if (stat(path, &info) >= 0) {
467 		if (!S_ISBLK(info.st_mode)) {
468 			log_error("A non-block device file at '%s' "
469 				  "is already present", path);
470 			return 0;
471 		}
472 
473 		/* If right inode already exists we don't touch uid etc. */
474 		if (info.st_rdev == dev)
475 			return 1;
476 
477 		if (unlink(path) < 0) {
478 			log_error("Unable to unlink device node for '%s'",
479 				  dev_name);
480 			return 0;
481 		}
482 	} else if (dm_udev_get_sync_support() && check_udev)
483 		log_warn("%s not set up by udev: Falling back to direct "
484 			 "node creation.", path);
485 
486 	old_mask = umask(0);
487 	if (mknod(path, S_IFBLK | mode, dev) < 0) {
488 		umask(old_mask);
489 		log_error("Unable to make device node for '%s'", dev_name);
490 		return 0;
491 	}
492 	umask(old_mask);
493 
494 	if (chown(path, uid, gid) < 0) {
495 		log_sys_error("chown", path);
496 		return 0;
497 	}
498 
499 	log_debug("Created %s", path);
500 
501 	if (!dm_set_selinux_context(path, S_IFBLK))
502 		return 0;
503 
504 	return 1;
505 }
506 
507 static int _rm_dev_node(const char *dev_name, int check_udev)
508 {
509 	char path[PATH_MAX];
510 	struct stat info;
511 
512 	#ifdef __NetBSD__
513 	char rpath[PATH_MAX];
514 	char raw_devname[DM_NAME_LEN+1]; /* r + other device name */
515 
516 	snprintf(raw_devname,sizeof(raw_devname),"r%s",dev_name);
517 
518 	_build_dev_path(rpath, sizeof(rpath), raw_devname);
519 
520 	if (stat(rpath, &info) < 0)
521 		return 1;
522 
523 	if (unlink(rpath) < 0) {
524 		log_error("Unable to unlink device node for '%s'", raw_devname);
525 		return 0;
526 	}
527 
528 	log_debug("Removed %s", rpath);
529 #endif
530 
531 	_build_dev_path(path, sizeof(path), dev_name);
532 
533 	if (stat(path, &info) < 0)
534 		return 1;
535 	else if (dm_udev_get_sync_support() && check_udev)
536 		log_warn("Node %s was not removed by udev. "
537 			 "Falling back to direct node removal.", path);
538 
539 	if (unlink(path) < 0) {
540 		log_error("Unable to unlink device node for '%s'", dev_name);
541 		return 0;
542 	}
543 
544 	log_debug("Removed %s", path);
545 
546 	return 1;
547 }
548 
549 static int _rename_dev_node(const char *old_name, const char *new_name,
550 			    int check_udev)
551 {
552 	char oldpath[PATH_MAX];
553 	char newpath[PATH_MAX];
554 	struct stat info;
555 
556 #ifdef __NetBSD__
557 	char rpath[PATH_MAX];
558 	char nrpath[PATH_MAX];
559 	char raw_devname[DM_NAME_LEN+1]; /* r + other device name */
560 	char nraw_devname[DM_NAME_LEN+1]; /* r + other device name */
561 
562 	snprintf(nraw_devname,sizeof(raw_devname),"r%s",new_name);
563 	snprintf(raw_devname,sizeof(raw_devname),"r%s",old_name);
564 
565 	_build_dev_path(nrpath, sizeof(nrpath), nraw_devname);
566 	_build_dev_path(rpath, sizeof(rpath), raw_devname);
567 
568 	if (stat(nrpath, &info) == 0) {
569 		if (S_ISBLK(info.st_mode)) {
570 			log_error("A block device file at '%s' "
571 			    "is present where raw device should be.", newpath);
572 			return 0;
573 		}
574 
575 		if (unlink(nrpath) < 0) {
576 			log_error("Unable to unlink device node for '%s'",
577 			    nraw_devname);
578 			return 0;
579 		}
580 	}
581 
582 	if (rename(rpath, nrpath) < 0) {
583 		log_error("Unable to rename device node from '%s' to '%s'",
584 		    raw_devname, nraw_devname);
585 		return 0;
586 	}
587 
588 	log_debug("Renamed %s to %s", rpath, nrpath);
589 
590 #endif
591 
592 	_build_dev_path(oldpath, sizeof(oldpath), old_name);
593 	_build_dev_path(newpath, sizeof(newpath), new_name);
594 
595 	if (stat(newpath, &info) == 0) {
596 		if (!S_ISBLK(info.st_mode)) {
597 			log_error("A non-block device file at '%s' "
598 				  "is already present", newpath);
599 			return 0;
600 		}
601 		else if (dm_udev_get_sync_support() && check_udev) {
602 			if (stat(oldpath, &info) < 0 &&
603 				 errno == ENOENT)
604 				/* assume udev already deleted this */
605 				return 1;
606 			else {
607 				log_warn("The node %s should have been renamed to %s "
608 					 "by udev but old node is still present. "
609 					 "Falling back to direct old node removal.",
610 					 oldpath, newpath);
611 				return _rm_dev_node(old_name, 0);
612 			}
613 		}
614 
615 		if (unlink(newpath) < 0) {
616 			if (errno == EPERM) {
617 				/* devfs, entry has already been renamed */
618 				return 1;
619 			}
620 			log_error("Unable to unlink device node for '%s'",
621 				  new_name);
622 			return 0;
623 		}
624 	}
625 	else if (dm_udev_get_sync_support() && check_udev)
626 		log_warn("The node %s should have been renamed to %s "
627 			 "by udev but new node is not present. "
628 			 "Falling back to direct node rename.",
629 			 oldpath, newpath);
630 
631 	if (rename(oldpath, newpath) < 0) {
632 		log_error("Unable to rename device node from '%s' to '%s'",
633 			  old_name, new_name);
634 		return 0;
635 	}
636 
637 	log_debug("Renamed %s to %s", oldpath, newpath);
638 
639 	return 1;
640 }
641 
642 #ifdef linux
643 static int _open_dev_node(const char *dev_name)
644 {
645 	int fd = -1;
646 	char path[PATH_MAX];
647 
648 	_build_dev_path(path, sizeof(path), dev_name);
649 
650 	if ((fd = open(path, O_RDONLY, 0)) < 0)
651 		log_sys_error("open", path);
652 
653 	return fd;
654 }
655 
656 int get_dev_node_read_ahead(const char *dev_name, uint32_t *read_ahead)
657 {
658 	int r = 1;
659 	int fd;
660 	long read_ahead_long;
661 
662 	if (!*dev_name) {
663 		log_error("Empty device name passed to BLKRAGET");
664 		return 0;
665 	}
666 
667 	if ((fd = _open_dev_node(dev_name)) < 0)
668 		return_0;
669 
670 	if (ioctl(fd, BLKRAGET, &read_ahead_long)) {
671 		log_sys_error("BLKRAGET", dev_name);
672 		*read_ahead = 0;
673 		r = 0;
674 	}  else {
675 		*read_ahead = (uint32_t) read_ahead_long;
676 		log_debug("%s: read ahead is %" PRIu32, dev_name, *read_ahead);
677 	}
678 
679 	if (close(fd))
680 		stack;
681 
682 	return r;
683 }
684 
685 static int _set_read_ahead(const char *dev_name, uint32_t read_ahead)
686 {
687 	int r = 1;
688 	int fd;
689 	long read_ahead_long = (long) read_ahead;
690 
691 	if (!*dev_name) {
692 		log_error("Empty device name passed to BLKRAGET");
693 		return 0;
694 	}
695 
696 	if ((fd = _open_dev_node(dev_name)) < 0)
697 		return_0;
698 
699 	log_debug("%s: Setting read ahead to %" PRIu32, dev_name, read_ahead);
700 
701 	if (ioctl(fd, BLKRASET, read_ahead_long)) {
702 		log_sys_error("BLKRASET", dev_name);
703 		r = 0;
704 	}
705 
706 	if (close(fd))
707 		stack;
708 
709 	return r;
710 }
711 
712 static int _set_dev_node_read_ahead(const char *dev_name, uint32_t read_ahead,
713 				    uint32_t read_ahead_flags)
714 {
715 	uint32_t current_read_ahead;
716 
717 	if (read_ahead == DM_READ_AHEAD_AUTO)
718 		return 1;
719 
720 	if (read_ahead == DM_READ_AHEAD_NONE)
721 		read_ahead = 0;
722 
723 	if (read_ahead_flags & DM_READ_AHEAD_MINIMUM_FLAG) {
724 		if (!get_dev_node_read_ahead(dev_name, &current_read_ahead))
725 			return_0;
726 
727 		if (current_read_ahead > read_ahead) {
728 			log_debug("%s: retaining kernel read ahead of %" PRIu32
729 				  " (requested %" PRIu32 ")",
730 				  dev_name, current_read_ahead, read_ahead);
731 			return 1;
732 		}
733 	}
734 
735 	return _set_read_ahead(dev_name, read_ahead);
736 }
737 
738 #else
739 
740 int get_dev_node_read_ahead(const char *dev_name, uint32_t *read_ahead)
741 {
742 	*read_ahead = 0;
743 
744 	return 1;
745 }
746 
747 static int _set_dev_node_read_ahead(const char *dev_name, uint32_t read_ahead,
748 				    uint32_t read_ahead_flags)
749 {
750 	return 1;
751 }
752 #endif
753 
754 typedef enum {
755 	NODE_ADD,
756 	NODE_DEL,
757 	NODE_RENAME,
758 	NODE_READ_AHEAD
759 } node_op_t;
760 
761 static int _do_node_op(node_op_t type, const char *dev_name, uint32_t major,
762 		       uint32_t minor, uid_t uid, gid_t gid, mode_t mode,
763 		       const char *old_name, uint32_t read_ahead,
764 		       uint32_t read_ahead_flags, int check_udev)
765 {
766 	switch (type) {
767 	case NODE_ADD:
768 		return _add_dev_node(dev_name, major, minor, uid, gid,
769 				     mode, check_udev);
770 	case NODE_DEL:
771 		return _rm_dev_node(dev_name, check_udev);
772 	case NODE_RENAME:
773 		return _rename_dev_node(old_name, dev_name, check_udev);
774 	case NODE_READ_AHEAD:
775 		return _set_dev_node_read_ahead(dev_name, read_ahead,
776 						read_ahead_flags);
777 	}
778 
779 	return 1;
780 }
781 
782 static DM_LIST_INIT(_node_ops);
783 
784 struct node_op_parms {
785 	struct dm_list list;
786 	node_op_t type;
787 	char *dev_name;
788 	uint32_t major;
789 	uint32_t minor;
790 	uid_t uid;
791 	gid_t gid;
792 	mode_t mode;
793 	uint32_t read_ahead;
794 	uint32_t read_ahead_flags;
795 	char *old_name;
796 	int check_udev;
797 	char names[0];
798 };
799 
800 static void _store_str(char **pos, char **ptr, const char *str)
801 {
802 	strcpy(*pos, str);
803 	*ptr = *pos;
804 	*pos += strlen(*ptr) + 1;
805 }
806 
807 static int _stack_node_op(node_op_t type, const char *dev_name, uint32_t major,
808 			  uint32_t minor, uid_t uid, gid_t gid, mode_t mode,
809 			  const char *old_name, uint32_t read_ahead,
810 			  uint32_t read_ahead_flags, int check_udev)
811 {
812 	struct node_op_parms *nop;
813 	struct dm_list *noph, *nopht;
814 	size_t len = strlen(dev_name) + strlen(old_name) + 2;
815 	char *pos;
816 
817 	/*
818 	 * Ignore any outstanding operations on the node if deleting it
819 	 */
820 	if (type == NODE_DEL) {
821 		dm_list_iterate_safe(noph, nopht, &_node_ops) {
822 			nop = dm_list_item(noph, struct node_op_parms);
823 			if (!strcmp(dev_name, nop->dev_name)) {
824 				dm_list_del(&nop->list);
825 				dm_free(nop);
826 			}
827 		}
828 	}
829 
830 	if (!(nop = dm_malloc(sizeof(*nop) + len))) {
831 		log_error("Insufficient memory to stack mknod operation");
832 		return 0;
833 	}
834 
835 	pos = nop->names;
836 	nop->type = type;
837 	nop->major = major;
838 	nop->minor = minor;
839 	nop->uid = uid;
840 	nop->gid = gid;
841 	nop->mode = mode;
842 	nop->read_ahead = read_ahead;
843 	nop->read_ahead_flags = read_ahead_flags;
844 	nop->check_udev = check_udev;
845 
846 	_store_str(&pos, &nop->dev_name, dev_name);
847 	_store_str(&pos, &nop->old_name, old_name);
848 
849 	dm_list_add(&_node_ops, &nop->list);
850 
851 	return 1;
852 }
853 
854 static void _pop_node_ops(void)
855 {
856 	struct dm_list *noph, *nopht;
857 	struct node_op_parms *nop;
858 
859 	dm_list_iterate_safe(noph, nopht, &_node_ops) {
860 		nop = dm_list_item(noph, struct node_op_parms);
861 		_do_node_op(nop->type, nop->dev_name, nop->major, nop->minor,
862 			    nop->uid, nop->gid, nop->mode, nop->old_name,
863 			    nop->read_ahead, nop->read_ahead_flags,
864 			    nop->check_udev);
865 		dm_list_del(&nop->list);
866 		dm_free(nop);
867 	}
868 }
869 
870 int add_dev_node(const char *dev_name, uint32_t major, uint32_t minor,
871 		 uid_t uid, gid_t gid, mode_t mode, int check_udev)
872 {
873 	log_debug("%s: Stacking NODE_ADD (%" PRIu32 ",%" PRIu32 ") %u:%u 0%o",
874 		  dev_name, major, minor, uid, gid, mode);
875 
876 	return _stack_node_op(NODE_ADD, dev_name, major, minor, uid,
877 			      gid, mode, "", 0, 0, check_udev);
878 }
879 
880 int rename_dev_node(const char *old_name, const char *new_name, int check_udev)
881 {
882 	log_debug("%s: Stacking NODE_RENAME to %s", old_name, new_name);
883 
884 	return _stack_node_op(NODE_RENAME, new_name, 0, 0, 0,
885 			      0, 0, old_name, 0, 0, check_udev);
886 }
887 
888 int rm_dev_node(const char *dev_name, int check_udev)
889 {
890 	log_debug("%s: Stacking NODE_DEL (replaces other stacked ops)", dev_name);
891 
892 	return _stack_node_op(NODE_DEL, dev_name, 0, 0, 0,
893 			      0, 0, "", 0, 0, check_udev);
894 }
895 
896 int set_dev_node_read_ahead(const char *dev_name, uint32_t read_ahead,
897 			    uint32_t read_ahead_flags)
898 {
899 	if (read_ahead == DM_READ_AHEAD_AUTO)
900 		return 1;
901 
902 	log_debug("%s: Stacking NODE_READ_AHEAD %" PRIu32 " (flags=%" PRIu32
903 		  ")", dev_name, read_ahead, read_ahead_flags);
904 
905 	return _stack_node_op(NODE_READ_AHEAD, dev_name, 0, 0, 0, 0,
906                               0, "", read_ahead, read_ahead_flags, 0);
907 }
908 
909 void update_devs(void)
910 {
911 	_pop_node_ops();
912 }
913 
914 int dm_set_dev_dir(const char *dev_dir)
915 {
916 	size_t len;
917 	const char *slash;
918 	if (*dev_dir != '/') {
919 		log_debug("Invalid dev_dir value, %s: "
920 			  "not an absolute name.", dev_dir);
921 		return 0;
922 	}
923 
924 	len = strlen(dev_dir);
925 	slash = dev_dir[len-1] == '/' ? "" : "/";
926 
927 	if (snprintf(_dm_dir, sizeof _dm_dir, "%s%s%s", dev_dir, slash, DM_DIR)
928 	    >= sizeof _dm_dir) {
929 		log_debug("Invalid dev_dir value, %s: name too long.", dev_dir);
930 		return 0;
931 	}
932 
933 	return 1;
934 }
935 
936 const char *dm_dir(void)
937 {
938 	return _dm_dir;
939 }
940 
941 int dm_mknodes(const char *name)
942 {
943 	struct dm_task *dmt;
944 	int r = 0;
945 
946 	if (!(dmt = dm_task_create(DM_DEVICE_MKNODES)))
947 		return 0;
948 
949 	if (name && !dm_task_set_name(dmt, name))
950 		goto out;
951 
952 	if (!dm_task_no_open_count(dmt))
953 		goto out;
954 
955 	r = dm_task_run(dmt);
956 
957 out:
958 	dm_task_destroy(dmt);
959 	return r;
960 }
961 
962 int dm_driver_version(char *version, size_t size)
963 {
964 	struct dm_task *dmt;
965 	int r = 0;
966 
967 	if (!(dmt = dm_task_create(DM_DEVICE_VERSION)))
968 		return 0;
969 
970 	if (!dm_task_run(dmt))
971 		log_error("Failed to get driver version");
972 
973 	if (!dm_task_get_driver_version(dmt, version, size))
974 		goto out;
975 
976 	r = 1;
977 
978 out:
979 	dm_task_destroy(dmt);
980 	return r;
981 }
982 
983 #ifndef UDEV_SYNC_SUPPORT
984 void dm_udev_set_sync_support(int sync_with_udev)
985 {
986 }
987 
988 int dm_udev_get_sync_support(void)
989 {
990 	return 0;
991 }
992 
993 int dm_task_set_cookie(struct dm_task *dmt, uint32_t *cookie, uint16_t flags)
994 {
995 	if (dm_cookie_supported())
996 		dmt->event_nr = flags << DM_UDEV_FLAGS_SHIFT;
997 	*cookie = 0;
998 
999 	return 1;
1000 }
1001 
1002 int dm_udev_complete(uint32_t cookie)
1003 {
1004 	return 1;
1005 }
1006 
1007 int dm_udev_wait(uint32_t cookie)
1008 {
1009 	return 1;
1010 }
1011 
1012 #else		/* UDEV_SYNC_SUPPORT */
1013 
1014 
1015 static int _check_udev_is_running(void)
1016 {
1017 
1018 #  ifndef HAVE_UDEV_QUEUE_GET_UDEV_IS_ACTIVE
1019 
1020 	log_debug("Could not get udev state because libudev library "
1021 		  "was not found and it was not compiled in. "
1022 		  "Assuming udev is not running.");
1023 	return 0;
1024 
1025 #  else	/* HAVE_UDEV_QUEUE_GET_UDEV_IS_ACTIVE */
1026 
1027 	struct udev *udev;
1028 	struct udev_queue *udev_queue;
1029 	int r;
1030 
1031 	if (!(udev = udev_new()))
1032 		goto_bad;
1033 
1034 	if (!(udev_queue = udev_queue_new(udev))) {
1035 		udev_unref(udev);
1036 		goto_bad;
1037 	}
1038 
1039 	if (!(r = udev_queue_get_udev_is_active(udev_queue)))
1040 		log_debug("Udev is not running. "
1041 			  "Not using udev synchronisation code.");
1042 
1043 	udev_queue_unref(udev_queue);
1044 	udev_unref(udev);
1045 
1046 	return r;
1047 
1048 bad:
1049 	log_error("Could not get udev state. Assuming udev is not running.");
1050 	return 0;
1051 
1052 #  endif	/* HAVE_UDEV_QUEUE_GET_UDEV_IS_ACTIVE */
1053 
1054 }
1055 
1056 void dm_udev_set_sync_support(int sync_with_udev)
1057 {
1058 	if (_udev_running < 0)
1059 		_udev_running = _check_udev_is_running();
1060 
1061 	_sync_with_udev = sync_with_udev;
1062 }
1063 
1064 int dm_udev_get_sync_support(void)
1065 {
1066 	if (_udev_running < 0)
1067 		_udev_running = _check_udev_is_running();
1068 
1069 	return dm_cookie_supported() && _udev_running && _sync_with_udev;
1070 }
1071 
1072 static int _get_cookie_sem(uint32_t cookie, int *semid)
1073 {
1074 	if (cookie >> 16 != DM_COOKIE_MAGIC) {
1075 		log_error("Could not continue to access notification "
1076 			  "semaphore identified by cookie value %"
1077 			  PRIu32 " (0x%x). Incorrect cookie prefix.",
1078 			  cookie, cookie);
1079 		return 0;
1080 	}
1081 
1082 	if ((*semid = semget((key_t) cookie, 1, 0)) >= 0)
1083 		return 1;
1084 
1085 	switch (errno) {
1086 		case ENOENT:
1087 			log_error("Could not find notification "
1088 				  "semaphore identified by cookie "
1089 				  "value %" PRIu32 " (0x%x)",
1090 				  cookie, cookie);
1091 			break;
1092 		case EACCES:
1093 			log_error("No permission to access "
1094 				  "notificaton semaphore identified "
1095 				  "by cookie value %" PRIu32 " (0x%x)",
1096 				  cookie, cookie);
1097 			break;
1098 		default:
1099 			log_error("Failed to access notification "
1100 				   "semaphore identified by cookie "
1101 				   "value %" PRIu32 " (0x%x): %s",
1102 				  cookie, cookie, strerror(errno));
1103 			break;
1104 	}
1105 
1106 	return 0;
1107 }
1108 
1109 static int _udev_notify_sem_inc(uint32_t cookie, int semid)
1110 {
1111 	struct sembuf sb = {0, 1, 0};
1112 
1113 	if (semop(semid, &sb, 1) < 0) {
1114 		log_error("semid %d: semop failed for cookie 0x%" PRIx32 ": %s",
1115 			  semid, cookie, strerror(errno));
1116 		return 0;
1117 	}
1118 
1119 	log_debug("Udev cookie 0x%" PRIx32 " (semid %d) incremented",
1120 		  cookie, semid);
1121 
1122 	return 1;
1123 }
1124 
1125 static int _udev_notify_sem_dec(uint32_t cookie, int semid)
1126 {
1127 	struct sembuf sb = {0, -1, IPC_NOWAIT};
1128 
1129 	if (semop(semid, &sb, 1) < 0) {
1130 		switch (errno) {
1131 			case EAGAIN:
1132 				log_error("semid %d: semop failed for cookie "
1133 					  "0x%" PRIx32 ": "
1134 					  "incorrect semaphore state",
1135 					  semid, cookie);
1136 				break;
1137 			default:
1138 				log_error("semid %d: semop failed for cookie "
1139 					  "0x%" PRIx32 ": %s",
1140 					  semid, cookie, strerror(errno));
1141 				break;
1142 		}
1143 		return 0;
1144 	}
1145 
1146 	log_debug("Udev cookie 0x%" PRIx32 " (semid %d) decremented",
1147 		  cookie, semid);
1148 
1149 	return 1;
1150 }
1151 
1152 static int _udev_notify_sem_destroy(uint32_t cookie, int semid)
1153 {
1154 	if (semctl(semid, 0, IPC_RMID, 0) < 0) {
1155 		log_error("Could not cleanup notification semaphore "
1156 			  "identified by cookie value %" PRIu32 " (0x%x): %s",
1157 			  cookie, cookie, strerror(errno));
1158 		return 0;
1159 	}
1160 
1161 	log_debug("Udev cookie 0x%" PRIx32 " (semid %d) destroyed", cookie,
1162 		  semid);
1163 
1164 	return 1;
1165 }
1166 
1167 static int _udev_notify_sem_create(uint32_t *cookie, int *semid)
1168 {
1169 	int fd;
1170 	int gen_semid;
1171 	uint16_t base_cookie;
1172 	uint32_t gen_cookie;
1173 
1174 	if ((fd = open("/dev/urandom", O_RDONLY)) < 0) {
1175 		log_error("Failed to open /dev/urandom "
1176 			  "to create random cookie value");
1177 		*cookie = 0;
1178 		return 0;
1179 	}
1180 
1181 	/* Generate random cookie value. Be sure it is unique and non-zero. */
1182 	do {
1183 		/* FIXME Handle non-error returns from read(). Move _io() into libdm? */
1184 		if (read(fd, &base_cookie, sizeof(base_cookie)) != sizeof(base_cookie)) {
1185 			log_error("Failed to initialize notification cookie");
1186 			goto bad;
1187 		}
1188 
1189 		gen_cookie = DM_COOKIE_MAGIC << 16 | base_cookie;
1190 
1191 		if (base_cookie && (gen_semid = semget((key_t) gen_cookie,
1192 				    1, 0600 | IPC_CREAT | IPC_EXCL)) < 0) {
1193 			switch (errno) {
1194 				case EEXIST:
1195 					/* if the semaphore key exists, we
1196 					 * simply generate another random one */
1197 					base_cookie = 0;
1198 					break;
1199 				case ENOMEM:
1200 					log_error("Not enough memory to create "
1201 						  "notification semaphore");
1202 					goto bad;
1203 				case ENOSPC:
1204 					log_error("Limit for the maximum number "
1205 						  "of semaphores reached. You can "
1206 						  "check and set the limits in "
1207 						  "/proc/sys/kernel/sem.");
1208 					goto bad;
1209 				default:
1210 					log_error("Failed to create notification "
1211 						  "semaphore: %s", strerror(errno));
1212 					goto bad;
1213 			}
1214 		}
1215 	} while (!base_cookie);
1216 
1217 	log_debug("Udev cookie 0x%" PRIx32 " (semid %d) created",
1218 		  gen_cookie, gen_semid);
1219 
1220 	if (semctl(gen_semid, 0, SETVAL, 1) < 0) {
1221 		log_error("semid %d: semctl failed: %s", gen_semid, strerror(errno));
1222 		/* We have to destroy just created semaphore
1223 		 * so it won't stay in the system. */
1224 		(void) _udev_notify_sem_destroy(gen_cookie, gen_semid);
1225 		goto bad;
1226 	}
1227 
1228 	log_debug("Udev cookie 0x%" PRIx32 " (semid %d) incremented",
1229 		  gen_cookie, gen_semid);
1230 
1231 	if (close(fd))
1232 		stack;
1233 
1234 	*semid = gen_semid;
1235 	*cookie = gen_cookie;
1236 
1237 	return 1;
1238 
1239 bad:
1240 	if (close(fd))
1241 		stack;
1242 
1243 	*cookie = 0;
1244 
1245 	return 0;
1246 }
1247 
1248 int dm_task_set_cookie(struct dm_task *dmt, uint32_t *cookie, uint16_t flags)
1249 {
1250 	int semid;
1251 
1252 	if (dm_cookie_supported())
1253 		dmt->event_nr = flags << DM_UDEV_FLAGS_SHIFT;
1254 
1255 	if (!dm_udev_get_sync_support()) {
1256 		*cookie = 0;
1257 		return 1;
1258 	}
1259 
1260 	if (*cookie) {
1261 		if (!_get_cookie_sem(*cookie, &semid))
1262 			goto_bad;
1263 	} else if (!_udev_notify_sem_create(cookie, &semid))
1264 		goto_bad;
1265 
1266 	if (!_udev_notify_sem_inc(*cookie, semid)) {
1267 		log_error("Could not set notification semaphore "
1268 			  "identified by cookie value %" PRIu32 " (0x%x)",
1269 			  *cookie, *cookie);
1270 		goto bad;
1271 	}
1272 
1273 	dmt->event_nr |= ~DM_UDEV_FLAGS_MASK & *cookie;
1274 	dmt->cookie_set = 1;
1275 
1276 	log_debug("Udev cookie 0x%" PRIx32 " (semid %d) assigned to dm_task "
1277 		  "with flags 0x%" PRIx16, *cookie, semid, flags);
1278 
1279 	return 1;
1280 
1281 bad:
1282 	dmt->event_nr = 0;
1283 	return 0;
1284 }
1285 
1286 int dm_udev_complete(uint32_t cookie)
1287 {
1288 	int semid;
1289 
1290 	if (!cookie || !dm_udev_get_sync_support())
1291 		return 1;
1292 
1293 	if (!_get_cookie_sem(cookie, &semid))
1294 		return_0;
1295 
1296 	if (!_udev_notify_sem_dec(cookie, semid)) {
1297 		log_error("Could not signal waiting process using notification "
1298 			  "semaphore identified by cookie value %" PRIu32 " (0x%x)",
1299 			  cookie, cookie);
1300 		return 0;
1301 	}
1302 
1303 	return 1;
1304 }
1305 
1306 int dm_udev_wait(uint32_t cookie)
1307 {
1308 	int semid;
1309 	struct sembuf sb = {0, 0, 0};
1310 
1311 	if (!cookie || !dm_udev_get_sync_support())
1312 		return 1;
1313 
1314 	if (!_get_cookie_sem(cookie, &semid))
1315 		return_0;
1316 
1317 	if (!_udev_notify_sem_dec(cookie, semid)) {
1318 		log_error("Failed to set a proper state for notification "
1319 			  "semaphore identified by cookie value %" PRIu32 " (0x%x) "
1320 			  "to initialize waiting for incoming notifications.",
1321 			  cookie, cookie);
1322 		(void) _udev_notify_sem_destroy(cookie, semid);
1323 		return 0;
1324 	}
1325 
1326 	log_debug("Udev cookie 0x%" PRIx32 " (semid %d): Waiting for zero",
1327 		  cookie, semid);
1328 
1329 repeat_wait:
1330 	if (semop(semid, &sb, 1) < 0) {
1331 		if (errno == EINTR)
1332 			goto repeat_wait;
1333 		else if (errno == EIDRM)
1334 			return 1;
1335 
1336 		log_error("Could not set wait state for notification semaphore "
1337 			  "identified by cookie value %" PRIu32 " (0x%x): %s",
1338 			  cookie, cookie, strerror(errno));
1339 		(void) _udev_notify_sem_destroy(cookie, semid);
1340 		return 0;
1341 	}
1342 
1343 	return _udev_notify_sem_destroy(cookie, semid);
1344 }
1345 
1346 #endif		/* UDEV_SYNC_SUPPORT */
1347