xref: /dragonfly/sys/vfs/devfs/devfs_core.c (revision 678e8cc6)
1 /*
2  * Copyright (c) 2009 The DragonFly Project.  All rights reserved.
3  *
4  * This code is derived from software contributed to The DragonFly Project
5  * by Alex Hornung <ahornung@gmail.com>
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  *
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in
15  *    the documentation and/or other materials provided with the
16  *    distribution.
17  * 3. Neither the name of The DragonFly Project nor the names of its
18  *    contributors may be used to endorse or promote products derived
19  *    from this software without specific, prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
25  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  */
34 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/kernel.h>
37 #include <sys/mount.h>
38 #include <sys/vnode.h>
39 #include <sys/types.h>
40 #include <sys/lock.h>
41 #include <sys/msgport.h>
42 #include <sys/sysctl.h>
43 #include <sys/ucred.h>
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/devfs.h>
47 #include <sys/devfs_rules.h>
48 #include <sys/udev.h>
49 
50 #include <sys/msgport2.h>
51 #include <sys/spinlock2.h>
52 #include <sys/mplock2.h>
53 #include <sys/sysref2.h>
54 
55 MALLOC_DEFINE(M_DEVFS, "devfs", "Device File System (devfs) allocations");
56 DEVFS_DECLARE_CLONE_BITMAP(ops_id);
57 /*
58  * SYSREF Integration - reference counting, allocation,
59  * sysid and syslink integration.
60  */
61 static void devfs_cdev_terminate(cdev_t dev);
62 static void devfs_cdev_lock(cdev_t dev);
63 static void devfs_cdev_unlock(cdev_t dev);
64 static struct sysref_class     cdev_sysref_class = {
65 	.name =         "cdev",
66 	.mtype =        M_DEVFS,
67 	.proto =        SYSREF_PROTO_DEV,
68 	.offset =       offsetof(struct cdev, si_sysref),
69 	.objsize =      sizeof(struct cdev),
70 	.nom_cache =	32,
71 	.flags =        0,
72 	.ops =  {
73 		.terminate = (sysref_terminate_func_t)devfs_cdev_terminate,
74 		.lock = (sysref_lock_func_t)devfs_cdev_lock,
75 		.unlock = (sysref_unlock_func_t)devfs_cdev_unlock
76 	}
77 };
78 
79 static struct objcache	*devfs_node_cache;
80 static struct objcache 	*devfs_msg_cache;
81 static struct objcache	*devfs_dev_cache;
82 
83 static struct objcache_malloc_args devfs_node_malloc_args = {
84 	sizeof(struct devfs_node), M_DEVFS };
85 struct objcache_malloc_args devfs_msg_malloc_args = {
86 	sizeof(struct devfs_msg), M_DEVFS };
87 struct objcache_malloc_args devfs_dev_malloc_args = {
88 	sizeof(struct cdev), M_DEVFS };
89 
90 static struct devfs_dev_head devfs_dev_list =
91 		TAILQ_HEAD_INITIALIZER(devfs_dev_list);
92 static struct devfs_mnt_head devfs_mnt_list =
93 		TAILQ_HEAD_INITIALIZER(devfs_mnt_list);
94 static struct devfs_chandler_head devfs_chandler_list =
95 		TAILQ_HEAD_INITIALIZER(devfs_chandler_list);
96 static struct devfs_alias_head devfs_alias_list =
97 		TAILQ_HEAD_INITIALIZER(devfs_alias_list);
98 static struct devfs_dev_ops_head devfs_dev_ops_list =
99 		TAILQ_HEAD_INITIALIZER(devfs_dev_ops_list);
100 
101 struct lock 		devfs_lock;
102 static struct lwkt_port devfs_dispose_port;
103 static struct lwkt_port devfs_msg_port;
104 static struct thread 	*td_core;
105 
106 static struct spinlock  ino_lock;
107 static ino_t 	d_ino;
108 static int	devfs_debug_enable;
109 static int	devfs_run;
110 
111 static ino_t devfs_fetch_ino(void);
112 static int devfs_create_all_dev_worker(struct devfs_node *);
113 static int devfs_create_dev_worker(cdev_t, uid_t, gid_t, int);
114 static int devfs_destroy_dev_worker(cdev_t);
115 static int devfs_destroy_related_worker(cdev_t);
116 static int devfs_destroy_dev_by_ops_worker(struct dev_ops *, int);
117 static int devfs_propagate_dev(cdev_t, int);
118 static int devfs_unlink_dev(cdev_t dev);
119 static void devfs_msg_exec(devfs_msg_t msg);
120 
121 static int devfs_chandler_add_worker(const char *, d_clone_t *);
122 static int devfs_chandler_del_worker(const char *);
123 
124 static void devfs_msg_autofree_reply(lwkt_port_t, lwkt_msg_t);
125 static void devfs_msg_core(void *);
126 
127 static int devfs_find_device_by_name_worker(devfs_msg_t);
128 static int devfs_find_device_by_udev_worker(devfs_msg_t);
129 
130 static int devfs_apply_reset_rules_caller(char *, int);
131 
132 static int devfs_scan_callback_worker(devfs_scan_t *, void *);
133 
134 static struct devfs_node *devfs_resolve_or_create_dir(struct devfs_node *,
135 		char *, size_t, int);
136 
137 static int devfs_make_alias_worker(struct devfs_alias *);
138 static int devfs_destroy_alias_worker(struct devfs_alias *);
139 static int devfs_alias_remove(cdev_t);
140 static int devfs_alias_reap(void);
141 static int devfs_alias_propagate(struct devfs_alias *, int);
142 static int devfs_alias_apply(struct devfs_node *, struct devfs_alias *);
143 static int devfs_alias_check_create(struct devfs_node *);
144 
145 static int devfs_clr_related_flag_worker(cdev_t, uint32_t);
146 static int devfs_destroy_related_without_flag_worker(cdev_t, uint32_t);
147 
148 static void *devfs_reaperp_callback(struct devfs_node *, void *);
149 static void *devfs_gc_dirs_callback(struct devfs_node *, void *);
150 static void *devfs_gc_links_callback(struct devfs_node *, struct devfs_node *);
151 static void *
152 devfs_inode_to_vnode_worker_callback(struct devfs_node *, ino_t *);
153 
154 /*
155  * devfs_debug() is a SYSCTL and TUNABLE controlled debug output function
156  * using kvprintf
157  */
158 int
159 devfs_debug(int level, char *fmt, ...)
160 {
161 	__va_list ap;
162 
163 	__va_start(ap, fmt);
164 	if (level <= devfs_debug_enable)
165 		kvprintf(fmt, ap);
166 	__va_end(ap);
167 
168 	return 0;
169 }
170 
171 /*
172  * devfs_allocp() Allocates a new devfs node with the specified
173  * parameters. The node is also automatically linked into the topology
174  * if a parent is specified. It also calls the rule and alias stuff to
175  * be applied on the new node
176  */
177 struct devfs_node *
178 devfs_allocp(devfs_nodetype devfsnodetype, char *name,
179 	     struct devfs_node *parent, struct mount *mp, cdev_t dev)
180 {
181 	struct devfs_node *node = NULL;
182 	size_t namlen = strlen(name);
183 
184 	node = objcache_get(devfs_node_cache, M_WAITOK);
185 	bzero(node, sizeof(*node));
186 
187 	atomic_add_long(&DEVFS_MNTDATA(mp)->leak_count, 1);
188 
189 	node->d_dev = NULL;
190 	node->nchildren = 1;
191 	node->mp = mp;
192 	node->d_dir.d_ino = devfs_fetch_ino();
193 
194 	/*
195 	 * Cookie jar for children. Leave 0 and 1 for '.' and '..' entries
196 	 * respectively.
197 	 */
198 	node->cookie_jar = 2;
199 
200 	/*
201 	 * Access Control members
202 	 */
203 	node->mode = DEVFS_DEFAULT_MODE;
204 	node->uid = DEVFS_DEFAULT_UID;
205 	node->gid = DEVFS_DEFAULT_GID;
206 
207 	switch (devfsnodetype) {
208 	case Proot:
209 		/*
210 		 * Ensure that we don't recycle the root vnode by marking it as
211 		 * linked into the topology.
212 		 */
213 		node->flags |= DEVFS_NODE_LINKED;
214 	case Pdir:
215 		TAILQ_INIT(DEVFS_DENODE_HEAD(node));
216 		node->d_dir.d_type = DT_DIR;
217 		node->nchildren = 2;
218 		break;
219 
220 	case Plink:
221 		node->d_dir.d_type = DT_LNK;
222 		break;
223 
224 	case Preg:
225 		node->d_dir.d_type = DT_REG;
226 		break;
227 
228 	case Pdev:
229 		if (dev != NULL) {
230 			node->d_dir.d_type = DT_CHR;
231 			node->d_dev = dev;
232 
233 			node->mode = dev->si_perms;
234 			node->uid = dev->si_uid;
235 			node->gid = dev->si_gid;
236 
237 			devfs_alias_check_create(node);
238 		}
239 		break;
240 
241 	default:
242 		panic("devfs_allocp: unknown node type");
243 	}
244 
245 	node->v_node = NULL;
246 	node->node_type = devfsnodetype;
247 
248 	/* Initialize the dirent structure of each devfs vnode */
249 	node->d_dir.d_namlen = namlen;
250 	node->d_dir.d_name = kmalloc(namlen+1, M_DEVFS, M_WAITOK);
251 	memcpy(node->d_dir.d_name, name, namlen);
252 	node->d_dir.d_name[namlen] = '\0';
253 
254 	/* Initialize the parent node element */
255 	node->parent = parent;
256 
257 	/* Initialize *time members */
258 	nanotime(&node->atime);
259 	node->mtime = node->ctime = node->atime;
260 
261 	/*
262 	 * Associate with parent as last step, clean out namecache
263 	 * reference.
264 	 */
265 	if ((parent != NULL) &&
266 	    ((parent->node_type == Proot) || (parent->node_type == Pdir))) {
267 		parent->nchildren++;
268 		node->cookie = parent->cookie_jar++;
269 		node->flags |= DEVFS_NODE_LINKED;
270 		TAILQ_INSERT_TAIL(DEVFS_DENODE_HEAD(parent), node, link);
271 
272 		/* This forces negative namecache lookups to clear */
273 		++mp->mnt_namecache_gen;
274 	}
275 
276 	/* Apply rules */
277 	devfs_rule_check_apply(node, NULL);
278 
279 	atomic_add_long(&DEVFS_MNTDATA(mp)->file_count, 1);
280 
281 	return node;
282 }
283 
284 /*
285  * devfs_allocv() allocates a new vnode based on a devfs node.
286  */
287 int
288 devfs_allocv(struct vnode **vpp, struct devfs_node *node)
289 {
290 	struct vnode *vp;
291 	int error = 0;
292 
293 	KKASSERT(node);
294 
295 try_again:
296 	while ((vp = node->v_node) != NULL) {
297 		error = vget(vp, LK_EXCLUSIVE);
298 		if (error == 0) {
299 			*vpp = vp;
300 			goto out;
301 		}
302 		if (error != ENOENT) {
303 			*vpp = NULL;
304 			goto out;
305 		}
306 	}
307 
308 	if ((error = getnewvnode(VT_DEVFS, node->mp, vpp, 0, 0)) != 0)
309 		goto out;
310 
311 	vp = *vpp;
312 
313 	if (node->v_node != NULL) {
314 		vp->v_type = VBAD;
315 		vx_put(vp);
316 		goto try_again;
317 	}
318 
319 	vp->v_data = node;
320 	node->v_node = vp;
321 
322 	switch (node->node_type) {
323 	case Proot:
324 		vsetflags(vp, VROOT);
325 		/* fall through */
326 	case Pdir:
327 		vp->v_type = VDIR;
328 		break;
329 
330 	case Plink:
331 		vp->v_type = VLNK;
332 		break;
333 
334 	case Preg:
335 		vp->v_type = VREG;
336 		break;
337 
338 	case Pdev:
339 		vp->v_type = VCHR;
340 		KKASSERT(node->d_dev);
341 
342 		vp->v_uminor = node->d_dev->si_uminor;
343 		vp->v_umajor = 0;
344 
345 		v_associate_rdev(vp, node->d_dev);
346 		vp->v_ops = &node->mp->mnt_vn_spec_ops;
347 		break;
348 
349 	default:
350 		panic("devfs_allocv: unknown node type");
351 	}
352 
353 out:
354 	return error;
355 }
356 
357 /*
358  * devfs_allocvp allocates both a devfs node (with the given settings) and a vnode
359  * based on the newly created devfs node.
360  */
361 int
362 devfs_allocvp(struct mount *mp, struct vnode **vpp, devfs_nodetype devfsnodetype,
363 		char *name, struct devfs_node *parent, cdev_t dev)
364 {
365 	struct devfs_node *node;
366 
367 	node = devfs_allocp(devfsnodetype, name, parent, mp, dev);
368 
369 	if (node != NULL)
370 		devfs_allocv(vpp, node);
371 	else
372 		*vpp = NULL;
373 
374 	return 0;
375 }
376 
377 /*
378  * Destroy the devfs_node.  The node must be unlinked from the topology.
379  *
380  * This function will also destroy any vnode association with the node
381  * and device.
382  *
383  * The cdev_t itself remains intact.
384  *
385  * The core lock is not necessarily held on call and must be temporarily
386  * released if it is to avoid a deadlock.
387  */
388 int
389 devfs_freep(struct devfs_node *node)
390 {
391 	struct vnode *vp;
392 	int relock;
393 
394 	KKASSERT(node);
395 	KKASSERT(((node->flags & DEVFS_NODE_LINKED) == 0) ||
396 		 (node->node_type == Proot));
397 
398 	/*
399 	 * Protect against double frees
400 	 */
401 	KKASSERT((node->flags & DEVFS_DESTROYED) == 0);
402 	node->flags |= DEVFS_DESTROYED;
403 
404 	/*
405 	 * Avoid deadlocks between devfs_lock and the vnode lock when
406 	 * disassociating the vnode (stress2 pty vs ls -la /dev/pts).
407 	 *
408 	 * This also prevents the vnode reclaim code from double-freeing
409 	 * the node.  The vget() is required to safely modified the vp
410 	 * and cycle the refs to terminate an inactive vp.
411 	 */
412 	if (lockstatus(&devfs_lock, curthread) == LK_EXCLUSIVE) {
413 		lockmgr(&devfs_lock, LK_RELEASE);
414 		relock = 1;
415 	} else {
416 		relock = 0;
417 	}
418 
419 	while ((vp = node->v_node) != NULL) {
420 		if (vget(vp, LK_EXCLUSIVE | LK_RETRY) != 0)
421 			break;
422 		v_release_rdev(vp);
423 		vp->v_data = NULL;
424 		node->v_node = NULL;
425 		cache_inval_vp(vp, CINV_DESTROY);
426 		vput(vp);
427 	}
428 
429 	/*
430 	 * Remaining cleanup
431 	 */
432 	atomic_subtract_long(&DEVFS_MNTDATA(node->mp)->leak_count, 1);
433 	if (node->symlink_name)	{
434 		kfree(node->symlink_name, M_DEVFS);
435 		node->symlink_name = NULL;
436 	}
437 
438 	/*
439 	 * Remove the node from the orphan list if it is still on it.
440 	 */
441 	if (node->flags & DEVFS_ORPHANED)
442 		devfs_tracer_del_orphan(node);
443 
444 	if (node->d_dir.d_name) {
445 		kfree(node->d_dir.d_name, M_DEVFS);
446 		node->d_dir.d_name = NULL;
447 	}
448 	atomic_subtract_long(&DEVFS_MNTDATA(node->mp)->file_count, 1);
449 	objcache_put(devfs_node_cache, node);
450 
451 	if (relock)
452 		lockmgr(&devfs_lock, LK_EXCLUSIVE);
453 
454 	return 0;
455 }
456 
457 /*
458  * Unlink the devfs node from the topology and add it to the orphan list.
459  * The node will later be destroyed by freep.
460  *
461  * Any vnode association, including the v_rdev and v_data, remains intact
462  * until the freep.
463  */
464 int
465 devfs_unlinkp(struct devfs_node *node)
466 {
467 	struct devfs_node *parent;
468 	KKASSERT(node);
469 
470 	/*
471 	 * Add the node to the orphan list, so it is referenced somewhere, to
472 	 * so we don't leak it.
473 	 */
474 	devfs_tracer_add_orphan(node);
475 
476 	parent = node->parent;
477 
478 	/*
479 	 * If the parent is known we can unlink the node out of the topology
480 	 */
481 	if (parent)	{
482 		TAILQ_REMOVE(DEVFS_DENODE_HEAD(parent), node, link);
483 		parent->nchildren--;
484 		node->flags &= ~DEVFS_NODE_LINKED;
485 	}
486 
487 	node->parent = NULL;
488 	return 0;
489 }
490 
491 void *
492 devfs_iterate_topology(struct devfs_node *node,
493 		devfs_iterate_callback_t *callback, void *arg1)
494 {
495 	struct devfs_node *node1, *node2;
496 	void *ret = NULL;
497 
498 	if ((node->node_type == Proot) || (node->node_type == Pdir)) {
499 		if (node->nchildren > 2) {
500 			TAILQ_FOREACH_MUTABLE(node1, DEVFS_DENODE_HEAD(node),
501 							link, node2) {
502 				if ((ret = devfs_iterate_topology(node1, callback, arg1)))
503 					return ret;
504 			}
505 		}
506 	}
507 
508 	ret = callback(node, arg1);
509 	return ret;
510 }
511 
512 /*
513  * devfs_reaperp() is a recursive function that iterates through all the
514  * topology, unlinking and freeing all devfs nodes.
515  */
516 static void *
517 devfs_reaperp_callback(struct devfs_node *node, void *unused)
518 {
519 	devfs_unlinkp(node);
520 	devfs_freep(node);
521 
522 	return NULL;
523 }
524 
525 static void *
526 devfs_gc_dirs_callback(struct devfs_node *node, void *unused)
527 {
528 	if (node->node_type == Pdir) {
529 		if ((node->nchildren == 2) &&
530 		    !(node->flags & DEVFS_USER_CREATED)) {
531 			devfs_unlinkp(node);
532 			devfs_freep(node);
533 		}
534 	}
535 
536 	return NULL;
537 }
538 
539 static void *
540 devfs_gc_links_callback(struct devfs_node *node, struct devfs_node *target)
541 {
542 	if ((node->node_type == Plink) && (node->link_target == target)) {
543 		devfs_unlinkp(node);
544 		devfs_freep(node);
545 	}
546 
547 	return NULL;
548 }
549 
550 /*
551  * devfs_gc() is devfs garbage collector. It takes care of unlinking and
552  * freeing a node, but also removes empty directories and links that link
553  * via devfs auto-link mechanism to the node being deleted.
554  */
555 int
556 devfs_gc(struct devfs_node *node)
557 {
558 	struct devfs_node *root_node = DEVFS_MNTDATA(node->mp)->root_node;
559 
560 	if (node->nlinks > 0)
561 		devfs_iterate_topology(root_node,
562 				(devfs_iterate_callback_t *)devfs_gc_links_callback, node);
563 
564 	devfs_unlinkp(node);
565 	devfs_iterate_topology(root_node,
566 			(devfs_iterate_callback_t *)devfs_gc_dirs_callback, NULL);
567 
568 	devfs_freep(node);
569 
570 	return 0;
571 }
572 
573 /*
574  * devfs_create_dev() is the asynchronous entry point for device creation.
575  * It just sends a message with the relevant details to the devfs core.
576  *
577  * This function will reference the passed device.  The reference is owned
578  * by devfs and represents all of the device's node associations.
579  */
580 int
581 devfs_create_dev(cdev_t dev, uid_t uid, gid_t gid, int perms)
582 {
583 	reference_dev(dev);
584 	devfs_msg_send_dev(DEVFS_DEVICE_CREATE, dev, uid, gid, perms);
585 
586 	return 0;
587 }
588 
589 /*
590  * devfs_destroy_dev() is the asynchronous entry point for device destruction.
591  * It just sends a message with the relevant details to the devfs core.
592  */
593 int
594 devfs_destroy_dev(cdev_t dev)
595 {
596 	devfs_msg_send_dev(DEVFS_DEVICE_DESTROY, dev, 0, 0, 0);
597 	return 0;
598 }
599 
600 /*
601  * devfs_mount_add() is the synchronous entry point for adding a new devfs
602  * mount.  It sends a synchronous message with the relevant details to the
603  * devfs core.
604  */
605 int
606 devfs_mount_add(struct devfs_mnt_data *mnt)
607 {
608 	devfs_msg_t msg;
609 
610 	msg = devfs_msg_get();
611 	msg->mdv_mnt = mnt;
612 	msg = devfs_msg_send_sync(DEVFS_MOUNT_ADD, msg);
613 	devfs_msg_put(msg);
614 
615 	return 0;
616 }
617 
618 /*
619  * devfs_mount_del() is the synchronous entry point for removing a devfs mount.
620  * It sends a synchronous message with the relevant details to the devfs core.
621  */
622 int
623 devfs_mount_del(struct devfs_mnt_data *mnt)
624 {
625 	devfs_msg_t msg;
626 
627 	msg = devfs_msg_get();
628 	msg->mdv_mnt = mnt;
629 	msg = devfs_msg_send_sync(DEVFS_MOUNT_DEL, msg);
630 	devfs_msg_put(msg);
631 
632 	return 0;
633 }
634 
635 /*
636  * devfs_destroy_related() is the synchronous entry point for device
637  * destruction by subname. It just sends a message with the relevant details to
638  * the devfs core.
639  */
640 int
641 devfs_destroy_related(cdev_t dev)
642 {
643 	devfs_msg_t msg;
644 
645 	msg = devfs_msg_get();
646 	msg->mdv_load = dev;
647 	msg = devfs_msg_send_sync(DEVFS_DESTROY_RELATED, msg);
648 	devfs_msg_put(msg);
649 	return 0;
650 }
651 
652 int
653 devfs_clr_related_flag(cdev_t dev, uint32_t flag)
654 {
655 	devfs_msg_t msg;
656 
657 	msg = devfs_msg_get();
658 	msg->mdv_flags.dev = dev;
659 	msg->mdv_flags.flag = flag;
660 	msg = devfs_msg_send_sync(DEVFS_CLR_RELATED_FLAG, msg);
661 	devfs_msg_put(msg);
662 
663 	return 0;
664 }
665 
666 int
667 devfs_destroy_related_without_flag(cdev_t dev, uint32_t flag)
668 {
669 	devfs_msg_t msg;
670 
671 	msg = devfs_msg_get();
672 	msg->mdv_flags.dev = dev;
673 	msg->mdv_flags.flag = flag;
674 	msg = devfs_msg_send_sync(DEVFS_DESTROY_RELATED_WO_FLAG, msg);
675 	devfs_msg_put(msg);
676 
677 	return 0;
678 }
679 
680 /*
681  * devfs_create_all_dev is the asynchronous entry point to trigger device
682  * node creation.  It just sends a message with the relevant details to
683  * the devfs core.
684  */
685 int
686 devfs_create_all_dev(struct devfs_node *root)
687 {
688 	devfs_msg_send_generic(DEVFS_CREATE_ALL_DEV, root);
689 	return 0;
690 }
691 
692 /*
693  * devfs_destroy_dev_by_ops is the asynchronous entry point to destroy all
694  * devices with a specific set of dev_ops and minor.  It just sends a
695  * message with the relevant details to the devfs core.
696  */
697 int
698 devfs_destroy_dev_by_ops(struct dev_ops *ops, int minor)
699 {
700 	devfs_msg_send_ops(DEVFS_DESTROY_DEV_BY_OPS, ops, minor);
701 	return 0;
702 }
703 
704 /*
705  * devfs_clone_handler_add is the synchronous entry point to add a new
706  * clone handler.  It just sends a message with the relevant details to
707  * the devfs core.
708  */
709 int
710 devfs_clone_handler_add(const char *name, d_clone_t *nhandler)
711 {
712 	devfs_msg_t msg;
713 
714 	msg = devfs_msg_get();
715 	msg->mdv_chandler.name = name;
716 	msg->mdv_chandler.nhandler = nhandler;
717 	msg = devfs_msg_send_sync(DEVFS_CHANDLER_ADD, msg);
718 	devfs_msg_put(msg);
719 	return 0;
720 }
721 
722 /*
723  * devfs_clone_handler_del is the synchronous entry point to remove a
724  * clone handler.  It just sends a message with the relevant details to
725  * the devfs core.
726  */
727 int
728 devfs_clone_handler_del(const char *name)
729 {
730 	devfs_msg_t msg;
731 
732 	msg = devfs_msg_get();
733 	msg->mdv_chandler.name = name;
734 	msg->mdv_chandler.nhandler = NULL;
735 	msg = devfs_msg_send_sync(DEVFS_CHANDLER_DEL, msg);
736 	devfs_msg_put(msg);
737 	return 0;
738 }
739 
740 /*
741  * devfs_find_device_by_name is the synchronous entry point to find a
742  * device given its name.  It sends a synchronous message with the
743  * relevant details to the devfs core and returns the answer.
744  */
745 cdev_t
746 devfs_find_device_by_name(const char *fmt, ...)
747 {
748 	cdev_t found = NULL;
749 	devfs_msg_t msg;
750 	char *target;
751 	__va_list ap;
752 
753 	if (fmt == NULL)
754 		return NULL;
755 
756 	__va_start(ap, fmt);
757 	kvasnrprintf(&target, PATH_MAX, 10, fmt, ap);
758 	__va_end(ap);
759 
760 	msg = devfs_msg_get();
761 	msg->mdv_name = target;
762 	msg = devfs_msg_send_sync(DEVFS_FIND_DEVICE_BY_NAME, msg);
763 	found = msg->mdv_cdev;
764 	devfs_msg_put(msg);
765 	kvasfree(&target);
766 
767 	return found;
768 }
769 
770 /*
771  * devfs_find_device_by_udev is the synchronous entry point to find a
772  * device given its udev number.  It sends a synchronous message with
773  * the relevant details to the devfs core and returns the answer.
774  */
775 cdev_t
776 devfs_find_device_by_udev(udev_t udev)
777 {
778 	cdev_t found = NULL;
779 	devfs_msg_t msg;
780 
781 	msg = devfs_msg_get();
782 	msg->mdv_udev = udev;
783 	msg = devfs_msg_send_sync(DEVFS_FIND_DEVICE_BY_UDEV, msg);
784 	found = msg->mdv_cdev;
785 	devfs_msg_put(msg);
786 
787 	devfs_debug(DEVFS_DEBUG_DEBUG,
788 		    "devfs_find_device_by_udev found? %s  -end:3-\n",
789 		    ((found) ? found->si_name:"NO"));
790 	return found;
791 }
792 
793 struct vnode *
794 devfs_inode_to_vnode(struct mount *mp, ino_t target)
795 {
796 	struct vnode *vp = NULL;
797 	devfs_msg_t msg;
798 
799 	if (mp == NULL)
800 		return NULL;
801 
802 	msg = devfs_msg_get();
803 	msg->mdv_ino.mp = mp;
804 	msg->mdv_ino.ino = target;
805 	msg = devfs_msg_send_sync(DEVFS_INODE_TO_VNODE, msg);
806 	vp = msg->mdv_ino.vp;
807 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
808 	devfs_msg_put(msg);
809 
810 	return vp;
811 }
812 
813 /*
814  * devfs_make_alias is the asynchronous entry point to register an alias
815  * for a device.  It just sends a message with the relevant details to the
816  * devfs core.
817  */
818 int
819 devfs_make_alias(const char *name, cdev_t dev_target)
820 {
821 	struct devfs_alias *alias;
822 	size_t len;
823 
824 	len = strlen(name);
825 
826 	alias = kmalloc(sizeof(struct devfs_alias), M_DEVFS, M_WAITOK);
827 	alias->name = kstrdup(name, M_DEVFS);
828 	alias->namlen = len;
829 	alias->dev_target = dev_target;
830 
831 	devfs_msg_send_generic(DEVFS_MAKE_ALIAS, alias);
832 	return 0;
833 }
834 
835 /*
836  * devfs_destroy_alias is the asynchronous entry point to deregister an alias
837  * for a device.  It just sends a message with the relevant details to the
838  * devfs core.
839  */
840 int
841 devfs_destroy_alias(const char *name, cdev_t dev_target)
842 {
843 	struct devfs_alias *alias;
844 	size_t len;
845 
846 	len = strlen(name);
847 
848 	alias = kmalloc(sizeof(struct devfs_alias), M_DEVFS, M_WAITOK);
849 	alias->name = kstrdup(name, M_DEVFS);
850 	alias->namlen = len;
851 	alias->dev_target = dev_target;
852 
853 	devfs_msg_send_generic(DEVFS_DESTROY_ALIAS, alias);
854 	return 0;
855 }
856 
857 /*
858  * devfs_apply_rules is the asynchronous entry point to trigger application
859  * of all rules.  It just sends a message with the relevant details to the
860  * devfs core.
861  */
862 int
863 devfs_apply_rules(char *mntto)
864 {
865 	char *new_name;
866 
867 	new_name = kstrdup(mntto, M_DEVFS);
868 	devfs_msg_send_name(DEVFS_APPLY_RULES, new_name);
869 
870 	return 0;
871 }
872 
873 /*
874  * devfs_reset_rules is the asynchronous entry point to trigger reset of all
875  * rules. It just sends a message with the relevant details to the devfs core.
876  */
877 int
878 devfs_reset_rules(char *mntto)
879 {
880 	char *new_name;
881 
882 	new_name = kstrdup(mntto, M_DEVFS);
883 	devfs_msg_send_name(DEVFS_RESET_RULES, new_name);
884 
885 	return 0;
886 }
887 
888 
889 /*
890  * devfs_scan_callback is the asynchronous entry point to call a callback
891  * on all cdevs.
892  * It just sends a message with the relevant details to the devfs core.
893  */
894 int
895 devfs_scan_callback(devfs_scan_t *callback, void *arg)
896 {
897 	devfs_msg_t msg;
898 
899 	KKASSERT(sizeof(callback) == sizeof(void *));
900 
901 	msg = devfs_msg_get();
902 	msg->mdv_load = callback;
903 	msg->mdv_load2 = arg;
904 	msg = devfs_msg_send_sync(DEVFS_SCAN_CALLBACK, msg);
905 	devfs_msg_put(msg);
906 
907 	return 0;
908 }
909 
910 
911 /*
912  * Acts as a message drain. Any message that is replied to here gets destroyed
913  * and the memory freed.
914  */
915 static void
916 devfs_msg_autofree_reply(lwkt_port_t port, lwkt_msg_t msg)
917 {
918 	devfs_msg_put((devfs_msg_t)msg);
919 }
920 
921 /*
922  * devfs_msg_get allocates a new devfs msg and returns it.
923  */
924 devfs_msg_t
925 devfs_msg_get(void)
926 {
927 	return objcache_get(devfs_msg_cache, M_WAITOK);
928 }
929 
930 /*
931  * devfs_msg_put deallocates a given devfs msg.
932  */
933 int
934 devfs_msg_put(devfs_msg_t msg)
935 {
936 	objcache_put(devfs_msg_cache, msg);
937 	return 0;
938 }
939 
940 /*
941  * devfs_msg_send is the generic asynchronous message sending facility
942  * for devfs. By default the reply port is the automatic disposal port.
943  *
944  * If the current thread is the devfs_msg_port thread we execute the
945  * operation synchronously.
946  */
947 void
948 devfs_msg_send(uint32_t cmd, devfs_msg_t devfs_msg)
949 {
950 	lwkt_port_t port = &devfs_msg_port;
951 
952 	lwkt_initmsg(&devfs_msg->hdr, &devfs_dispose_port, 0);
953 
954 	devfs_msg->hdr.u.ms_result = cmd;
955 
956 	if (port->mpu_td == curthread) {
957 		devfs_msg_exec(devfs_msg);
958 		lwkt_replymsg(&devfs_msg->hdr, 0);
959 	} else {
960 		lwkt_sendmsg(port, (lwkt_msg_t)devfs_msg);
961 	}
962 }
963 
964 /*
965  * devfs_msg_send_sync is the generic synchronous message sending
966  * facility for devfs. It initializes a local reply port and waits
967  * for the core's answer. This answer is then returned.
968  */
969 devfs_msg_t
970 devfs_msg_send_sync(uint32_t cmd, devfs_msg_t devfs_msg)
971 {
972 	struct lwkt_port rep_port;
973 	devfs_msg_t	msg_incoming;
974 	lwkt_port_t port = &devfs_msg_port;
975 
976 	lwkt_initport_thread(&rep_port, curthread);
977 	lwkt_initmsg(&devfs_msg->hdr, &rep_port, 0);
978 
979 	devfs_msg->hdr.u.ms_result = cmd;
980 
981 	lwkt_sendmsg(port, (lwkt_msg_t)devfs_msg);
982 	msg_incoming = lwkt_waitport(&rep_port, 0);
983 
984 	return msg_incoming;
985 }
986 
987 /*
988  * sends a message with a generic argument.
989  */
990 void
991 devfs_msg_send_generic(uint32_t cmd, void *load)
992 {
993 	devfs_msg_t devfs_msg = devfs_msg_get();
994 
995 	devfs_msg->mdv_load = load;
996 	devfs_msg_send(cmd, devfs_msg);
997 }
998 
999 /*
1000  * sends a message with a name argument.
1001  */
1002 void
1003 devfs_msg_send_name(uint32_t cmd, char *name)
1004 {
1005 	devfs_msg_t devfs_msg = devfs_msg_get();
1006 
1007 	devfs_msg->mdv_name = name;
1008 	devfs_msg_send(cmd, devfs_msg);
1009 }
1010 
1011 /*
1012  * sends a message with a mount argument.
1013  */
1014 void
1015 devfs_msg_send_mount(uint32_t cmd, struct devfs_mnt_data *mnt)
1016 {
1017 	devfs_msg_t devfs_msg = devfs_msg_get();
1018 
1019 	devfs_msg->mdv_mnt = mnt;
1020 	devfs_msg_send(cmd, devfs_msg);
1021 }
1022 
1023 /*
1024  * sends a message with an ops argument.
1025  */
1026 void
1027 devfs_msg_send_ops(uint32_t cmd, struct dev_ops *ops, int minor)
1028 {
1029 	devfs_msg_t devfs_msg = devfs_msg_get();
1030 
1031 	devfs_msg->mdv_ops.ops = ops;
1032 	devfs_msg->mdv_ops.minor = minor;
1033 	devfs_msg_send(cmd, devfs_msg);
1034 }
1035 
1036 /*
1037  * sends a message with a clone handler argument.
1038  */
1039 void
1040 devfs_msg_send_chandler(uint32_t cmd, char *name, d_clone_t handler)
1041 {
1042 	devfs_msg_t devfs_msg = devfs_msg_get();
1043 
1044 	devfs_msg->mdv_chandler.name = name;
1045 	devfs_msg->mdv_chandler.nhandler = handler;
1046 	devfs_msg_send(cmd, devfs_msg);
1047 }
1048 
1049 /*
1050  * sends a message with a device argument.
1051  */
1052 void
1053 devfs_msg_send_dev(uint32_t cmd, cdev_t dev, uid_t uid, gid_t gid, int perms)
1054 {
1055 	devfs_msg_t devfs_msg = devfs_msg_get();
1056 
1057 	devfs_msg->mdv_dev.dev = dev;
1058 	devfs_msg->mdv_dev.uid = uid;
1059 	devfs_msg->mdv_dev.gid = gid;
1060 	devfs_msg->mdv_dev.perms = perms;
1061 
1062 	devfs_msg_send(cmd, devfs_msg);
1063 }
1064 
1065 /*
1066  * sends a message with a link argument.
1067  */
1068 void
1069 devfs_msg_send_link(uint32_t cmd, char *name, char *target, struct mount *mp)
1070 {
1071 	devfs_msg_t devfs_msg = devfs_msg_get();
1072 
1073 	devfs_msg->mdv_link.name = name;
1074 	devfs_msg->mdv_link.target = target;
1075 	devfs_msg->mdv_link.mp = mp;
1076 	devfs_msg_send(cmd, devfs_msg);
1077 }
1078 
1079 /*
1080  * devfs_msg_core is the main devfs thread. It handles all incoming messages
1081  * and calls the relevant worker functions. By using messages it's assured
1082  * that events occur in the correct order.
1083  */
1084 static void
1085 devfs_msg_core(void *arg)
1086 {
1087 	devfs_msg_t msg;
1088 
1089 	lwkt_initport_thread(&devfs_msg_port, curthread);
1090 
1091 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
1092 	devfs_run = 1;
1093 	wakeup(td_core);
1094 	lockmgr(&devfs_lock, LK_RELEASE);
1095 
1096 	get_mplock();	/* mpsafe yet? */
1097 
1098 	while (devfs_run) {
1099 		msg = (devfs_msg_t)lwkt_waitport(&devfs_msg_port, 0);
1100 		devfs_debug(DEVFS_DEBUG_DEBUG,
1101 				"devfs_msg_core, new msg: %x\n",
1102 				(unsigned int)msg->hdr.u.ms_result);
1103 		devfs_msg_exec(msg);
1104 		lwkt_replymsg(&msg->hdr, 0);
1105 	}
1106 
1107 	rel_mplock();
1108 	wakeup(td_core);
1109 
1110 	lwkt_exit();
1111 }
1112 
1113 static void
1114 devfs_msg_exec(devfs_msg_t msg)
1115 {
1116 	struct devfs_mnt_data *mnt;
1117 	struct devfs_node *node;
1118 	cdev_t	dev;
1119 
1120 	/*
1121 	 * Acquire the devfs lock to ensure safety of all called functions
1122 	 */
1123 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
1124 
1125 	switch (msg->hdr.u.ms_result) {
1126 	case DEVFS_DEVICE_CREATE:
1127 		dev = msg->mdv_dev.dev;
1128 		devfs_create_dev_worker(dev,
1129 					msg->mdv_dev.uid,
1130 					msg->mdv_dev.gid,
1131 					msg->mdv_dev.perms);
1132 		break;
1133 	case DEVFS_DEVICE_DESTROY:
1134 		dev = msg->mdv_dev.dev;
1135 		devfs_destroy_dev_worker(dev);
1136 		break;
1137 	case DEVFS_DESTROY_RELATED:
1138 		devfs_destroy_related_worker(msg->mdv_load);
1139 		break;
1140 	case DEVFS_DESTROY_DEV_BY_OPS:
1141 		devfs_destroy_dev_by_ops_worker(msg->mdv_ops.ops,
1142 						msg->mdv_ops.minor);
1143 		break;
1144 	case DEVFS_CREATE_ALL_DEV:
1145 		node = (struct devfs_node *)msg->mdv_load;
1146 		devfs_create_all_dev_worker(node);
1147 		break;
1148 	case DEVFS_MOUNT_ADD:
1149 		mnt = msg->mdv_mnt;
1150 		TAILQ_INSERT_TAIL(&devfs_mnt_list, mnt, link);
1151 		devfs_create_all_dev_worker(mnt->root_node);
1152 		break;
1153 	case DEVFS_MOUNT_DEL:
1154 		mnt = msg->mdv_mnt;
1155 		TAILQ_REMOVE(&devfs_mnt_list, mnt, link);
1156 		devfs_iterate_topology(mnt->root_node, devfs_reaperp_callback,
1157 				       NULL);
1158 		if (mnt->leak_count) {
1159 			devfs_debug(DEVFS_DEBUG_SHOW,
1160 				    "Leaked %ld devfs_node elements!\n",
1161 				    mnt->leak_count);
1162 		}
1163 		break;
1164 	case DEVFS_CHANDLER_ADD:
1165 		devfs_chandler_add_worker(msg->mdv_chandler.name,
1166 				msg->mdv_chandler.nhandler);
1167 		break;
1168 	case DEVFS_CHANDLER_DEL:
1169 		devfs_chandler_del_worker(msg->mdv_chandler.name);
1170 		break;
1171 	case DEVFS_FIND_DEVICE_BY_NAME:
1172 		devfs_find_device_by_name_worker(msg);
1173 		break;
1174 	case DEVFS_FIND_DEVICE_BY_UDEV:
1175 		devfs_find_device_by_udev_worker(msg);
1176 		break;
1177 	case DEVFS_MAKE_ALIAS:
1178 		devfs_make_alias_worker((struct devfs_alias *)msg->mdv_load);
1179 		break;
1180 	case DEVFS_DESTROY_ALIAS:
1181 		devfs_destroy_alias_worker((struct devfs_alias *)msg->mdv_load);
1182 		break;
1183 	case DEVFS_APPLY_RULES:
1184 		devfs_apply_reset_rules_caller(msg->mdv_name, 1);
1185 		break;
1186 	case DEVFS_RESET_RULES:
1187 		devfs_apply_reset_rules_caller(msg->mdv_name, 0);
1188 		break;
1189 	case DEVFS_SCAN_CALLBACK:
1190 		devfs_scan_callback_worker((devfs_scan_t *)msg->mdv_load,
1191 			msg->mdv_load2);
1192 		break;
1193 	case DEVFS_CLR_RELATED_FLAG:
1194 		devfs_clr_related_flag_worker(msg->mdv_flags.dev,
1195 				msg->mdv_flags.flag);
1196 		break;
1197 	case DEVFS_DESTROY_RELATED_WO_FLAG:
1198 		devfs_destroy_related_without_flag_worker(msg->mdv_flags.dev,
1199 				msg->mdv_flags.flag);
1200 		break;
1201 	case DEVFS_INODE_TO_VNODE:
1202 		msg->mdv_ino.vp = devfs_iterate_topology(
1203 			DEVFS_MNTDATA(msg->mdv_ino.mp)->root_node,
1204 			(devfs_iterate_callback_t *)devfs_inode_to_vnode_worker_callback,
1205 			&msg->mdv_ino.ino);
1206 		break;
1207 	case DEVFS_TERMINATE_CORE:
1208 		devfs_run = 0;
1209 		break;
1210 	case DEVFS_SYNC:
1211 		break;
1212 	default:
1213 		devfs_debug(DEVFS_DEBUG_WARNING,
1214 			    "devfs_msg_core: unknown message "
1215 			    "received at core\n");
1216 		break;
1217 	}
1218 	lockmgr(&devfs_lock, LK_RELEASE);
1219 }
1220 
1221 /*
1222  * Worker function to insert a new dev into the dev list and initialize its
1223  * permissions. It also calls devfs_propagate_dev which in turn propagates
1224  * the change to all mount points.
1225  *
1226  * The passed dev is already referenced.  This reference is eaten by this
1227  * function and represents the dev's linkage into devfs_dev_list.
1228  */
1229 static int
1230 devfs_create_dev_worker(cdev_t dev, uid_t uid, gid_t gid, int perms)
1231 {
1232 	KKASSERT(dev);
1233 
1234 	dev->si_uid = uid;
1235 	dev->si_gid = gid;
1236 	dev->si_perms = perms;
1237 
1238 	devfs_link_dev(dev);
1239 	devfs_propagate_dev(dev, 1);
1240 
1241 	udev_event_attach(dev, NULL, 0);
1242 
1243 	return 0;
1244 }
1245 
1246 /*
1247  * Worker function to delete a dev from the dev list and free the cdev.
1248  * It also calls devfs_propagate_dev which in turn propagates the change
1249  * to all mount points.
1250  */
1251 static int
1252 devfs_destroy_dev_worker(cdev_t dev)
1253 {
1254 	int error;
1255 
1256 	KKASSERT(dev);
1257 	KKASSERT((lockstatus(&devfs_lock, curthread)) == LK_EXCLUSIVE);
1258 
1259 	error = devfs_unlink_dev(dev);
1260 	devfs_propagate_dev(dev, 0);
1261 
1262 	udev_event_detach(dev, NULL, 0);
1263 
1264 	if (error == 0)
1265 		release_dev(dev);	/* link ref */
1266 	release_dev(dev);
1267 	release_dev(dev);
1268 
1269 	return 0;
1270 }
1271 
1272 /*
1273  * Worker function to destroy all devices with a certain basename.
1274  * Calls devfs_destroy_dev_worker for the actual destruction.
1275  */
1276 static int
1277 devfs_destroy_related_worker(cdev_t needle)
1278 {
1279 	cdev_t dev;
1280 
1281 restart:
1282 	devfs_debug(DEVFS_DEBUG_DEBUG, "related worker: %s\n",
1283 	    needle->si_name);
1284 	TAILQ_FOREACH(dev, &devfs_dev_list, link) {
1285 		if (dev->si_parent == needle) {
1286 			devfs_destroy_related_worker(dev);
1287 			devfs_destroy_dev_worker(dev);
1288 			goto restart;
1289 		}
1290 	}
1291 	return 0;
1292 }
1293 
1294 static int
1295 devfs_clr_related_flag_worker(cdev_t needle, uint32_t flag)
1296 {
1297 	cdev_t dev, dev1;
1298 
1299 	TAILQ_FOREACH_MUTABLE(dev, &devfs_dev_list, link, dev1) {
1300 		if (dev->si_parent == needle) {
1301 			devfs_clr_related_flag_worker(dev, flag);
1302 			dev->si_flags &= ~flag;
1303 		}
1304 	}
1305 
1306 	return 0;
1307 }
1308 
1309 static int
1310 devfs_destroy_related_without_flag_worker(cdev_t needle, uint32_t flag)
1311 {
1312 	cdev_t dev;
1313 
1314 restart:
1315 	devfs_debug(DEVFS_DEBUG_DEBUG, "related_wo_flag: %s\n",
1316 	    needle->si_name);
1317 
1318 	TAILQ_FOREACH(dev, &devfs_dev_list, link) {
1319 		if (dev->si_parent == needle) {
1320 			devfs_destroy_related_without_flag_worker(dev, flag);
1321 			if (!(dev->si_flags & flag)) {
1322 				devfs_destroy_dev_worker(dev);
1323 				devfs_debug(DEVFS_DEBUG_DEBUG,
1324 				    "related_wo_flag: %s restart\n", dev->si_name);
1325 				goto restart;
1326 			}
1327 		}
1328 	}
1329 
1330 	return 0;
1331 }
1332 
1333 /*
1334  * Worker function that creates all device nodes on top of a devfs
1335  * root node.
1336  */
1337 static int
1338 devfs_create_all_dev_worker(struct devfs_node *root)
1339 {
1340 	cdev_t dev;
1341 
1342 	KKASSERT(root);
1343 
1344 	TAILQ_FOREACH(dev, &devfs_dev_list, link) {
1345 		devfs_create_device_node(root, dev, NULL, NULL);
1346 	}
1347 
1348 	return 0;
1349 }
1350 
1351 /*
1352  * Worker function that destroys all devices that match a specific
1353  * dev_ops and/or minor. If minor is less than 0, it is not matched
1354  * against. It also propagates all changes.
1355  */
1356 static int
1357 devfs_destroy_dev_by_ops_worker(struct dev_ops *ops, int minor)
1358 {
1359 	cdev_t dev, dev1;
1360 
1361 	KKASSERT(ops);
1362 
1363 	TAILQ_FOREACH_MUTABLE(dev, &devfs_dev_list, link, dev1) {
1364 		if (dev->si_ops != ops)
1365 			continue;
1366 		if ((minor < 0) || (dev->si_uminor == minor)) {
1367 			devfs_destroy_dev_worker(dev);
1368 		}
1369 	}
1370 
1371 	return 0;
1372 }
1373 
1374 /*
1375  * Worker function that registers a new clone handler in devfs.
1376  */
1377 static int
1378 devfs_chandler_add_worker(const char *name, d_clone_t *nhandler)
1379 {
1380 	struct devfs_clone_handler *chandler = NULL;
1381 	u_char len = strlen(name);
1382 
1383 	if (len == 0)
1384 		return 1;
1385 
1386 	TAILQ_FOREACH(chandler, &devfs_chandler_list, link) {
1387 		if (chandler->namlen != len)
1388 			continue;
1389 
1390 		if (!memcmp(chandler->name, name, len)) {
1391 			/* Clonable basename already exists */
1392 			return 1;
1393 		}
1394 	}
1395 
1396 	chandler = kmalloc(sizeof(*chandler), M_DEVFS, M_WAITOK | M_ZERO);
1397 	chandler->name = kstrdup(name, M_DEVFS);
1398 	chandler->namlen = len;
1399 	chandler->nhandler = nhandler;
1400 
1401 	TAILQ_INSERT_TAIL(&devfs_chandler_list, chandler, link);
1402 	return 0;
1403 }
1404 
1405 /*
1406  * Worker function that removes a given clone handler from the
1407  * clone handler list.
1408  */
1409 static int
1410 devfs_chandler_del_worker(const char *name)
1411 {
1412 	struct devfs_clone_handler *chandler, *chandler2;
1413 	u_char len = strlen(name);
1414 
1415 	if (len == 0)
1416 		return 1;
1417 
1418 	TAILQ_FOREACH_MUTABLE(chandler, &devfs_chandler_list, link, chandler2) {
1419 		if (chandler->namlen != len)
1420 			continue;
1421 		if (memcmp(chandler->name, name, len))
1422 			continue;
1423 
1424 		TAILQ_REMOVE(&devfs_chandler_list, chandler, link);
1425 		kfree(chandler->name, M_DEVFS);
1426 		kfree(chandler, M_DEVFS);
1427 		break;
1428 	}
1429 
1430 	return 0;
1431 }
1432 
1433 /*
1434  * Worker function that finds a given device name and changes
1435  * the message received accordingly so that when replied to,
1436  * the answer is returned to the caller.
1437  */
1438 static int
1439 devfs_find_device_by_name_worker(devfs_msg_t devfs_msg)
1440 {
1441 	struct devfs_alias *alias;
1442 	cdev_t dev;
1443 	cdev_t found = NULL;
1444 
1445 	TAILQ_FOREACH(dev, &devfs_dev_list, link) {
1446 		if (strcmp(devfs_msg->mdv_name, dev->si_name) == 0) {
1447 			found = dev;
1448 			break;
1449 		}
1450 	}
1451 	if (found == NULL) {
1452 		TAILQ_FOREACH(alias, &devfs_alias_list, link) {
1453 			if (strcmp(devfs_msg->mdv_name, alias->name) == 0) {
1454 				found = alias->dev_target;
1455 				break;
1456 			}
1457 		}
1458 	}
1459 	devfs_msg->mdv_cdev = found;
1460 
1461 	return 0;
1462 }
1463 
1464 /*
1465  * Worker function that finds a given device udev and changes
1466  * the message received accordingly so that when replied to,
1467  * the answer is returned to the caller.
1468  */
1469 static int
1470 devfs_find_device_by_udev_worker(devfs_msg_t devfs_msg)
1471 {
1472 	cdev_t dev, dev1;
1473 	cdev_t found = NULL;
1474 
1475 	TAILQ_FOREACH_MUTABLE(dev, &devfs_dev_list, link, dev1) {
1476 		if (((udev_t)dev->si_inode) == devfs_msg->mdv_udev) {
1477 			found = dev;
1478 			break;
1479 		}
1480 	}
1481 	devfs_msg->mdv_cdev = found;
1482 
1483 	return 0;
1484 }
1485 
1486 /*
1487  * Worker function that inserts a given alias into the
1488  * alias list, and propagates the alias to all mount
1489  * points.
1490  */
1491 static int
1492 devfs_make_alias_worker(struct devfs_alias *alias)
1493 {
1494 	struct devfs_alias *alias2;
1495 	size_t len = strlen(alias->name);
1496 	int found = 0;
1497 
1498 	TAILQ_FOREACH(alias2, &devfs_alias_list, link) {
1499 		if (len != alias2->namlen)
1500 			continue;
1501 
1502 		if (!memcmp(alias->name, alias2->name, len)) {
1503 			found = 1;
1504 			break;
1505 		}
1506 	}
1507 
1508 	if (!found) {
1509 		/*
1510 		 * The alias doesn't exist yet, so we add it to the alias list
1511 		 */
1512 		TAILQ_INSERT_TAIL(&devfs_alias_list, alias, link);
1513 		devfs_alias_propagate(alias, 0);
1514 		udev_event_attach(alias->dev_target, alias->name, 1);
1515 	} else {
1516 		devfs_debug(DEVFS_DEBUG_WARNING,
1517 			    "Warning: duplicate devfs_make_alias for %s\n",
1518 			    alias->name);
1519 		kfree(alias->name, M_DEVFS);
1520 		kfree(alias, M_DEVFS);
1521 	}
1522 
1523 	return 0;
1524 }
1525 
1526 /*
1527  * Worker function that delete a given alias from the
1528  * alias list, and propagates the removal to all mount
1529  * points.
1530  */
1531 static int
1532 devfs_destroy_alias_worker(struct devfs_alias *alias)
1533 {
1534 	struct devfs_alias *alias2;
1535 	int found = 0;
1536 
1537 	TAILQ_FOREACH(alias2, &devfs_alias_list, link) {
1538 		if (alias->dev_target != alias2->dev_target)
1539 			continue;
1540 
1541 		if (devfs_WildCmp(alias->name, alias2->name) == 0) {
1542 			found = 1;
1543 			break;
1544 		}
1545 	}
1546 
1547 	if (!found) {
1548 		devfs_debug(DEVFS_DEBUG_WARNING,
1549 		    "Warning: devfs_destroy_alias for inexistant alias: %s\n",
1550 		    alias->name);
1551 		kfree(alias->name, M_DEVFS);
1552 		kfree(alias, M_DEVFS);
1553 	} else {
1554 		/*
1555 		 * The alias exists, so we delete it from the alias list
1556 		 */
1557 		TAILQ_REMOVE(&devfs_alias_list, alias2, link);
1558 		devfs_alias_propagate(alias2, 1);
1559 		udev_event_detach(alias2->dev_target, alias2->name, 1);
1560 		kfree(alias->name, M_DEVFS);
1561 		kfree(alias, M_DEVFS);
1562 		kfree(alias2->name, M_DEVFS);
1563 		kfree(alias2, M_DEVFS);
1564 	}
1565 
1566 	return 0;
1567 }
1568 
1569 /*
1570  * Function that removes and frees all aliases.
1571  */
1572 static int
1573 devfs_alias_reap(void)
1574 {
1575 	struct devfs_alias *alias, *alias2;
1576 
1577 	TAILQ_FOREACH_MUTABLE(alias, &devfs_alias_list, link, alias2) {
1578 		TAILQ_REMOVE(&devfs_alias_list, alias, link);
1579 		kfree(alias->name, M_DEVFS);
1580 		kfree(alias, M_DEVFS);
1581 	}
1582 	return 0;
1583 }
1584 
1585 /*
1586  * Function that removes an alias matching a specific cdev and frees
1587  * it accordingly.
1588  */
1589 static int
1590 devfs_alias_remove(cdev_t dev)
1591 {
1592 	struct devfs_alias *alias, *alias2;
1593 
1594 	TAILQ_FOREACH_MUTABLE(alias, &devfs_alias_list, link, alias2) {
1595 		if (alias->dev_target == dev) {
1596 			TAILQ_REMOVE(&devfs_alias_list, alias, link);
1597 			udev_event_detach(alias->dev_target, alias->name, 1);
1598 			kfree(alias->name, M_DEVFS);
1599 			kfree(alias, M_DEVFS);
1600 		}
1601 	}
1602 	return 0;
1603 }
1604 
1605 /*
1606  * This function propagates an alias addition or removal to
1607  * all mount points.
1608  */
1609 static int
1610 devfs_alias_propagate(struct devfs_alias *alias, int remove)
1611 {
1612 	struct devfs_mnt_data *mnt;
1613 
1614 	TAILQ_FOREACH(mnt, &devfs_mnt_list, link) {
1615 		if (remove) {
1616 			devfs_destroy_node(mnt->root_node, alias->name);
1617 		} else {
1618 			devfs_alias_apply(mnt->root_node, alias);
1619 		}
1620 	}
1621 	return 0;
1622 }
1623 
1624 /*
1625  * This function is a recursive function iterating through
1626  * all device nodes in the topology and, if applicable,
1627  * creating the relevant alias for a device node.
1628  */
1629 static int
1630 devfs_alias_apply(struct devfs_node *node, struct devfs_alias *alias)
1631 {
1632 	struct devfs_node *node1, *node2;
1633 
1634 	KKASSERT(alias != NULL);
1635 
1636 	if ((node->node_type == Proot) || (node->node_type == Pdir)) {
1637 		if (node->nchildren > 2) {
1638 			TAILQ_FOREACH_MUTABLE(node1, DEVFS_DENODE_HEAD(node), link, node2) {
1639 				devfs_alias_apply(node1, alias);
1640 			}
1641 		}
1642 	} else {
1643 		if (node->d_dev == alias->dev_target)
1644 			devfs_alias_create(alias->name, node, 0);
1645 	}
1646 	return 0;
1647 }
1648 
1649 /*
1650  * This function checks if any alias possibly is applicable
1651  * to the given node. If so, the alias is created.
1652  */
1653 static int
1654 devfs_alias_check_create(struct devfs_node *node)
1655 {
1656 	struct devfs_alias *alias;
1657 
1658 	TAILQ_FOREACH(alias, &devfs_alias_list, link) {
1659 		if (node->d_dev == alias->dev_target)
1660 			devfs_alias_create(alias->name, node, 0);
1661 	}
1662 	return 0;
1663 }
1664 
1665 /*
1666  * This function creates an alias with a given name
1667  * linking to a given devfs node. It also increments
1668  * the link count on the target node.
1669  */
1670 int
1671 devfs_alias_create(char *name_orig, struct devfs_node *target, int rule_based)
1672 {
1673 	struct mount *mp = target->mp;
1674 	struct devfs_node *parent = DEVFS_MNTDATA(mp)->root_node;
1675 	struct devfs_node *linknode;
1676 	char *create_path = NULL;
1677 	char *name;
1678 	char *name_buf;
1679 	int result = 0;
1680 
1681 	KKASSERT((lockstatus(&devfs_lock, curthread)) == LK_EXCLUSIVE);
1682 
1683 	name_buf = kmalloc(PATH_MAX, M_TEMP, M_WAITOK);
1684 	devfs_resolve_name_path(name_orig, name_buf, &create_path, &name);
1685 
1686 	if (create_path)
1687 		parent = devfs_resolve_or_create_path(parent, create_path, 1);
1688 
1689 
1690 	if (devfs_find_device_node_by_name(parent, name)) {
1691 		devfs_debug(DEVFS_DEBUG_WARNING,
1692 			    "Node already exists: %s "
1693 			    "(devfs_make_alias_worker)!\n",
1694 			    name);
1695 		result = 1;
1696 		goto done;
1697 	}
1698 
1699 	linknode = devfs_allocp(Plink, name, parent, mp, NULL);
1700 	if (linknode == NULL) {
1701 		result = 1;
1702 		goto done;
1703 	}
1704 
1705 	linknode->link_target = target;
1706 	target->nlinks++;
1707 
1708 	if (rule_based)
1709 		linknode->flags |= DEVFS_RULE_CREATED;
1710 
1711 done:
1712 	kfree(name_buf, M_TEMP);
1713 	return (result);
1714 }
1715 
1716 /*
1717  * This function is called by the core and handles mount point
1718  * strings. It either calls the relevant worker (devfs_apply_
1719  * reset_rules_worker) on all mountpoints or only a specific
1720  * one.
1721  */
1722 static int
1723 devfs_apply_reset_rules_caller(char *mountto, int apply)
1724 {
1725 	struct devfs_mnt_data *mnt;
1726 
1727 	if (mountto[0] == '*') {
1728 		TAILQ_FOREACH(mnt, &devfs_mnt_list, link) {
1729 			devfs_iterate_topology(mnt->root_node,
1730 					(apply)?(devfs_rule_check_apply):(devfs_rule_reset_node),
1731 					NULL);
1732 		}
1733 	} else {
1734 		TAILQ_FOREACH(mnt, &devfs_mnt_list, link) {
1735 			if (!strcmp(mnt->mp->mnt_stat.f_mntonname, mountto)) {
1736 				devfs_iterate_topology(mnt->root_node,
1737 					(apply)?(devfs_rule_check_apply):(devfs_rule_reset_node),
1738 					NULL);
1739 				break;
1740 			}
1741 		}
1742 	}
1743 
1744 	kfree(mountto, M_DEVFS);
1745 	return 0;
1746 }
1747 
1748 /*
1749  * This function calls a given callback function for
1750  * every dev node in the devfs dev list.
1751  */
1752 static int
1753 devfs_scan_callback_worker(devfs_scan_t *callback, void *arg)
1754 {
1755 	cdev_t dev, dev1;
1756 
1757 	TAILQ_FOREACH_MUTABLE(dev, &devfs_dev_list, link, dev1) {
1758 		callback(dev, arg);
1759 	}
1760 
1761 	return 0;
1762 }
1763 
1764 /*
1765  * This function tries to resolve a given directory, or if not
1766  * found and creation requested, creates the given directory.
1767  */
1768 static struct devfs_node *
1769 devfs_resolve_or_create_dir(struct devfs_node *parent, char *dir_name,
1770 			    size_t name_len, int create)
1771 {
1772 	struct devfs_node *node, *found = NULL;
1773 
1774 	TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(parent), link) {
1775 		if (name_len != node->d_dir.d_namlen)
1776 			continue;
1777 
1778 		if (!memcmp(dir_name, node->d_dir.d_name, name_len)) {
1779 			found = node;
1780 			break;
1781 		}
1782 	}
1783 
1784 	if ((found == NULL) && (create)) {
1785 		found = devfs_allocp(Pdir, dir_name, parent, parent->mp, NULL);
1786 	}
1787 
1788 	return found;
1789 }
1790 
1791 /*
1792  * This function tries to resolve a complete path. If creation is requested,
1793  * if a given part of the path cannot be resolved (because it doesn't exist),
1794  * it is created.
1795  */
1796 struct devfs_node *
1797 devfs_resolve_or_create_path(struct devfs_node *parent, char *path, int create)
1798 {
1799 	struct devfs_node *node = parent;
1800 	char *buf;
1801 	size_t idx = 0;
1802 
1803 	if (path == NULL)
1804 		return parent;
1805 
1806 	buf = kmalloc(PATH_MAX, M_TEMP, M_WAITOK);
1807 
1808 	while (*path && idx < PATH_MAX - 1) {
1809 		if (*path != '/') {
1810 			buf[idx++] = *path;
1811 		} else {
1812 			buf[idx] = '\0';
1813 			node = devfs_resolve_or_create_dir(node, buf, idx, create);
1814 			if (node == NULL) {
1815 				kfree(buf, M_TEMP);
1816 				return NULL;
1817 			}
1818 			idx = 0;
1819 		}
1820 		++path;
1821 	}
1822 	buf[idx] = '\0';
1823 	node = devfs_resolve_or_create_dir(node, buf, idx, create);
1824 	kfree (buf, M_TEMP);
1825 	return (node);
1826 }
1827 
1828 /*
1829  * Takes a full path and strips it into a directory path and a name.
1830  * For a/b/c/foo, it returns foo in namep and a/b/c in pathp. It
1831  * requires a working buffer with enough size to keep the whole
1832  * fullpath.
1833  */
1834 int
1835 devfs_resolve_name_path(char *fullpath, char *buf, char **pathp, char **namep)
1836 {
1837 	char *name = NULL;
1838 	char *path = NULL;
1839 	size_t len = strlen(fullpath) + 1;
1840 	int i;
1841 
1842 	KKASSERT((fullpath != NULL) && (buf != NULL));
1843 	KKASSERT((pathp != NULL) && (namep != NULL));
1844 
1845 	memcpy(buf, fullpath, len);
1846 
1847 	for (i = len-1; i>= 0; i--) {
1848 		if (buf[i] == '/') {
1849 			buf[i] = '\0';
1850 			name = &(buf[i+1]);
1851 			path = buf;
1852 			break;
1853 		}
1854 	}
1855 
1856 	*pathp = path;
1857 
1858 	if (name) {
1859 		*namep = name;
1860 	} else {
1861 		*namep = buf;
1862 	}
1863 
1864 	return 0;
1865 }
1866 
1867 /*
1868  * This function creates a new devfs node for a given device.  It can
1869  * handle a complete path as device name, and accordingly creates
1870  * the path and the final device node.
1871  *
1872  * The reference count on the passed dev remains unchanged.
1873  */
1874 struct devfs_node *
1875 devfs_create_device_node(struct devfs_node *root, cdev_t dev,
1876 			 char *dev_name, char *path_fmt, ...)
1877 {
1878 	struct devfs_node *parent, *node = NULL;
1879 	char *path = NULL;
1880 	char *name;
1881 	char *name_buf;
1882 	__va_list ap;
1883 	int i, found;
1884 	char *create_path = NULL;
1885 	char *names = "pqrsPQRS";
1886 
1887 	name_buf = kmalloc(PATH_MAX, M_TEMP, M_WAITOK);
1888 
1889 	if (path_fmt != NULL) {
1890 		__va_start(ap, path_fmt);
1891 		kvasnrprintf(&path, PATH_MAX, 10, path_fmt, ap);
1892 		__va_end(ap);
1893 	}
1894 
1895 	parent = devfs_resolve_or_create_path(root, path, 1);
1896 	KKASSERT(parent);
1897 
1898 	devfs_resolve_name_path(
1899 			((dev_name == NULL) && (dev))?(dev->si_name):(dev_name),
1900 			name_buf, &create_path, &name);
1901 
1902 	if (create_path)
1903 		parent = devfs_resolve_or_create_path(parent, create_path, 1);
1904 
1905 
1906 	if (devfs_find_device_node_by_name(parent, name)) {
1907 		devfs_debug(DEVFS_DEBUG_WARNING, "devfs_create_device_node: "
1908 			"DEVICE %s ALREADY EXISTS!!! Ignoring creation request.\n", name);
1909 		goto out;
1910 	}
1911 
1912 	node = devfs_allocp(Pdev, name, parent, parent->mp, dev);
1913 	nanotime(&parent->mtime);
1914 
1915 	/*
1916 	 * Ugly unix98 pty magic, to hide pty master (ptm) devices and their
1917 	 * directory
1918 	 */
1919 	if ((dev) && (strlen(dev->si_name) >= 4) &&
1920 			(!memcmp(dev->si_name, "ptm/", 4))) {
1921 		node->parent->flags |= DEVFS_HIDDEN;
1922 		node->flags |= DEVFS_HIDDEN;
1923 	}
1924 
1925 	/*
1926 	 * Ugly pty magic, to tag pty devices as such and hide them if needed.
1927 	 */
1928 	if ((strlen(name) >= 3) && (!memcmp(name, "pty", 3)))
1929 		node->flags |= (DEVFS_PTY | DEVFS_INVISIBLE);
1930 
1931 	if ((strlen(name) >= 3) && (!memcmp(name, "tty", 3))) {
1932 		found = 0;
1933 		for (i = 0; i < strlen(names); i++) {
1934 			if (name[3] == names[i]) {
1935 				found = 1;
1936 				break;
1937 			}
1938 		}
1939 		if (found)
1940 			node->flags |= (DEVFS_PTY | DEVFS_INVISIBLE);
1941 	}
1942 
1943 out:
1944 	kfree(name_buf, M_TEMP);
1945 	kvasfree(&path);
1946 	return node;
1947 }
1948 
1949 /*
1950  * This function finds a given device node in the topology with a given
1951  * cdev.
1952  */
1953 void *
1954 devfs_find_device_node_callback(struct devfs_node *node, cdev_t target)
1955 {
1956 	if ((node->node_type == Pdev) && (node->d_dev == target)) {
1957 		return node;
1958 	}
1959 
1960 	return NULL;
1961 }
1962 
1963 /*
1964  * This function finds a device node in the given parent directory by its
1965  * name and returns it.
1966  */
1967 struct devfs_node *
1968 devfs_find_device_node_by_name(struct devfs_node *parent, char *target)
1969 {
1970 	struct devfs_node *node, *found = NULL;
1971 	size_t len = strlen(target);
1972 
1973 	TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(parent), link) {
1974 		if (len != node->d_dir.d_namlen)
1975 			continue;
1976 
1977 		if (!memcmp(node->d_dir.d_name, target, len)) {
1978 			found = node;
1979 			break;
1980 		}
1981 	}
1982 
1983 	return found;
1984 }
1985 
1986 static void *
1987 devfs_inode_to_vnode_worker_callback(struct devfs_node *node, ino_t *inop)
1988 {
1989 	struct vnode *vp = NULL;
1990 	ino_t target = *inop;
1991 
1992 	if (node->d_dir.d_ino == target) {
1993 		if (node->v_node) {
1994 			vp = node->v_node;
1995 			vget(vp, LK_EXCLUSIVE | LK_RETRY);
1996 			vn_unlock(vp);
1997 		} else {
1998 			devfs_allocv(&vp, node);
1999 			vn_unlock(vp);
2000 		}
2001 	}
2002 
2003 	return vp;
2004 }
2005 
2006 /*
2007  * This function takes a cdev and removes its devfs node in the
2008  * given topology.  The cdev remains intact.
2009  */
2010 int
2011 devfs_destroy_device_node(struct devfs_node *root, cdev_t target)
2012 {
2013 	KKASSERT(target != NULL);
2014 	return devfs_destroy_node(root, target->si_name);
2015 }
2016 
2017 /*
2018  * This function takes a path to a devfs node, resolves it and
2019  * removes the devfs node from the given topology.
2020  */
2021 int
2022 devfs_destroy_node(struct devfs_node *root, char *target)
2023 {
2024 	struct devfs_node *node, *parent;
2025 	char *name;
2026 	char *name_buf;
2027 	char *create_path = NULL;
2028 
2029 	KKASSERT(target);
2030 
2031 	name_buf = kmalloc(PATH_MAX, M_TEMP, M_WAITOK);
2032 	ksnprintf(name_buf, PATH_MAX, "%s", target);
2033 
2034 	devfs_resolve_name_path(target, name_buf, &create_path, &name);
2035 
2036 	if (create_path)
2037 		parent = devfs_resolve_or_create_path(root, create_path, 0);
2038 	else
2039 		parent = root;
2040 
2041 	if (parent == NULL) {
2042 		kfree(name_buf, M_TEMP);
2043 		return 1;
2044 	}
2045 
2046 	node = devfs_find_device_node_by_name(parent, name);
2047 
2048 	if (node) {
2049 		nanotime(&node->parent->mtime);
2050 		devfs_gc(node);
2051 	}
2052 
2053 	kfree(name_buf, M_TEMP);
2054 
2055 	return 0;
2056 }
2057 
2058 /*
2059  * Just set perms and ownership for given node.
2060  */
2061 int
2062 devfs_set_perms(struct devfs_node *node, uid_t uid, gid_t gid,
2063 		u_short mode, u_long flags)
2064 {
2065 	node->mode = mode;
2066 	node->uid = uid;
2067 	node->gid = gid;
2068 
2069 	return 0;
2070 }
2071 
2072 /*
2073  * Propagates a device attach/detach to all mount
2074  * points. Also takes care of automatic alias removal
2075  * for a deleted cdev.
2076  */
2077 static int
2078 devfs_propagate_dev(cdev_t dev, int attach)
2079 {
2080 	struct devfs_mnt_data *mnt;
2081 
2082 	TAILQ_FOREACH(mnt, &devfs_mnt_list, link) {
2083 		if (attach) {
2084 			/* Device is being attached */
2085 			devfs_create_device_node(mnt->root_node, dev,
2086 						 NULL, NULL );
2087 		} else {
2088 			/* Device is being detached */
2089 			devfs_alias_remove(dev);
2090 			devfs_destroy_device_node(mnt->root_node, dev);
2091 		}
2092 	}
2093 	return 0;
2094 }
2095 
2096 /*
2097  * devfs_clone either returns a basename from a complete name by
2098  * returning the length of the name without trailing digits, or,
2099  * if clone != 0, calls the device's clone handler to get a new
2100  * device, which in turn is returned in devp.
2101  */
2102 cdev_t
2103 devfs_clone(cdev_t dev, const char *name, size_t len, int mode,
2104 		struct ucred *cred)
2105 {
2106 	int error;
2107 	struct devfs_clone_handler *chandler;
2108 	struct dev_clone_args ap;
2109 
2110 	TAILQ_FOREACH(chandler, &devfs_chandler_list, link) {
2111 		if (chandler->namlen != len)
2112 			continue;
2113 		if ((!memcmp(chandler->name, name, len)) && (chandler->nhandler)) {
2114 			lockmgr(&devfs_lock, LK_RELEASE);
2115 			devfs_config();
2116 			lockmgr(&devfs_lock, LK_EXCLUSIVE);
2117 
2118 			ap.a_head.a_dev = dev;
2119 			ap.a_dev = NULL;
2120 			ap.a_name = name;
2121 			ap.a_namelen = len;
2122 			ap.a_mode = mode;
2123 			ap.a_cred = cred;
2124 			error = (chandler->nhandler)(&ap);
2125 			if (error)
2126 				continue;
2127 
2128 			return ap.a_dev;
2129 		}
2130 	}
2131 
2132 	return NULL;
2133 }
2134 
2135 
2136 /*
2137  * Registers a new orphan in the orphan list.
2138  */
2139 void
2140 devfs_tracer_add_orphan(struct devfs_node *node)
2141 {
2142 	struct devfs_orphan *orphan;
2143 
2144 	KKASSERT(node);
2145 	orphan = kmalloc(sizeof(struct devfs_orphan), M_DEVFS, M_WAITOK);
2146 	orphan->node = node;
2147 
2148 	KKASSERT((node->flags & DEVFS_ORPHANED) == 0);
2149 	node->flags |= DEVFS_ORPHANED;
2150 	TAILQ_INSERT_TAIL(DEVFS_ORPHANLIST(node->mp), orphan, link);
2151 }
2152 
2153 /*
2154  * Removes an orphan from the orphan list.
2155  */
2156 void
2157 devfs_tracer_del_orphan(struct devfs_node *node)
2158 {
2159 	struct devfs_orphan *orphan;
2160 
2161 	KKASSERT(node);
2162 
2163 	TAILQ_FOREACH(orphan, DEVFS_ORPHANLIST(node->mp), link)	{
2164 		if (orphan->node == node) {
2165 			node->flags &= ~DEVFS_ORPHANED;
2166 			TAILQ_REMOVE(DEVFS_ORPHANLIST(node->mp), orphan, link);
2167 			kfree(orphan, M_DEVFS);
2168 			break;
2169 		}
2170 	}
2171 }
2172 
2173 /*
2174  * Counts the orphans in the orphan list, and if cleanup
2175  * is specified, also frees the orphan and removes it from
2176  * the list.
2177  */
2178 size_t
2179 devfs_tracer_orphan_count(struct mount *mp, int cleanup)
2180 {
2181 	struct devfs_orphan *orphan, *orphan2;
2182 	size_t count = 0;
2183 
2184 	TAILQ_FOREACH_MUTABLE(orphan, DEVFS_ORPHANLIST(mp), link, orphan2)	{
2185 		count++;
2186 		/*
2187 		 * If we are instructed to clean up, we do so.
2188 		 */
2189 		if (cleanup) {
2190 			TAILQ_REMOVE(DEVFS_ORPHANLIST(mp), orphan, link);
2191 			orphan->node->flags &= ~DEVFS_ORPHANED;
2192 			devfs_freep(orphan->node);
2193 			kfree(orphan, M_DEVFS);
2194 		}
2195 	}
2196 
2197 	return count;
2198 }
2199 
2200 /*
2201  * Fetch an ino_t from the global d_ino by increasing it
2202  * while spinlocked.
2203  */
2204 static ino_t
2205 devfs_fetch_ino(void)
2206 {
2207 	ino_t	ret;
2208 
2209 	spin_lock(&ino_lock);
2210 	ret = d_ino++;
2211 	spin_unlock(&ino_lock);
2212 
2213 	return ret;
2214 }
2215 
2216 /*
2217  * Allocates a new cdev and initializes it's most basic
2218  * fields.
2219  */
2220 cdev_t
2221 devfs_new_cdev(struct dev_ops *ops, int minor, struct dev_ops *bops)
2222 {
2223 	cdev_t dev = sysref_alloc(&cdev_sysref_class);
2224 
2225 	sysref_activate(&dev->si_sysref);
2226 	reference_dev(dev);
2227 	bzero(dev, offsetof(struct cdev, si_sysref));
2228 
2229 	dev->si_uid = 0;
2230 	dev->si_gid = 0;
2231 	dev->si_perms = 0;
2232 	dev->si_drv1 = NULL;
2233 	dev->si_drv2 = NULL;
2234 	dev->si_lastread = 0;		/* time_second */
2235 	dev->si_lastwrite = 0;		/* time_second */
2236 
2237 	dev->si_dict = NULL;
2238 	dev->si_parent = NULL;
2239 	dev->si_ops = ops;
2240 	dev->si_flags = 0;
2241 	dev->si_umajor = 0;
2242 	dev->si_uminor = minor;
2243 	dev->si_bops = bops;
2244 
2245 	/*
2246 	 * Since the disk subsystem is in the way, we need to
2247 	 * propagate the D_CANFREE from bops (and ops) to
2248 	 * si_flags.
2249 	 */
2250 	if (bops && (bops->head.flags & D_CANFREE)) {
2251 		dev->si_flags |= SI_CANFREE;
2252 	} else if (ops->head.flags & D_CANFREE) {
2253 		dev->si_flags |= SI_CANFREE;
2254 	}
2255 
2256 	/* If there is a backing device, we reference its ops */
2257 	dev->si_inode = makeudev(
2258 		    devfs_reference_ops((bops)?(bops):(ops)),
2259 		    minor );
2260 
2261 	return dev;
2262 }
2263 
2264 static void
2265 devfs_cdev_terminate(cdev_t dev)
2266 {
2267 	int locked = 0;
2268 
2269 	/* Check if it is locked already. if not, we acquire the devfs lock */
2270 	if (!(lockstatus(&devfs_lock, curthread)) == LK_EXCLUSIVE) {
2271 		lockmgr(&devfs_lock, LK_EXCLUSIVE);
2272 		locked = 1;
2273 	}
2274 
2275 	/*
2276 	 * Make sure the node isn't linked anymore. Otherwise we've screwed
2277 	 * up somewhere, since normal devs are unlinked on the call to
2278 	 * destroy_dev and only-cdevs that have not been used for cloning
2279 	 * are not linked in the first place. only-cdevs used for cloning
2280 	 * will be linked in, too, and should only be destroyed via
2281 	 * destroy_dev, not destroy_only_dev, so we catch that problem, too.
2282 	 */
2283 	KKASSERT((dev->si_flags & SI_DEVFS_LINKED) == 0);
2284 
2285 	/* If we acquired the lock, we also get rid of it */
2286 	if (locked)
2287 		lockmgr(&devfs_lock, LK_RELEASE);
2288 
2289 	/* If there is a backing device, we release the backing device's ops */
2290 	devfs_release_ops((dev->si_bops)?(dev->si_bops):(dev->si_ops));
2291 
2292 	/* Finally destroy the device */
2293 	sysref_put(&dev->si_sysref);
2294 }
2295 
2296 /*
2297  * Dummies for now (individual locks for MPSAFE)
2298  */
2299 static void
2300 devfs_cdev_lock(cdev_t dev)
2301 {
2302 }
2303 
2304 static void
2305 devfs_cdev_unlock(cdev_t dev)
2306 {
2307 }
2308 
2309 static int
2310 devfs_detached_filter_eof(struct knote *kn, long hint)
2311 {
2312 	kn->kn_flags |= (EV_EOF | EV_NODATA);
2313 	return (1);
2314 }
2315 
2316 static void
2317 devfs_detached_filter_detach(struct knote *kn)
2318 {
2319 	cdev_t dev = (cdev_t)kn->kn_hook;
2320 
2321 	knote_remove(&dev->si_kqinfo.ki_note, kn);
2322 }
2323 
2324 static struct filterops devfs_detached_filterops =
2325 	{ FILTEROP_ISFD, NULL,
2326 	  devfs_detached_filter_detach,
2327 	  devfs_detached_filter_eof };
2328 
2329 /*
2330  * Delegates knote filter handling responsibility to devfs
2331  *
2332  * Any device that implements kqfilter event handling and could be detached
2333  * or shut down out from under the kevent subsystem must allow devfs to
2334  * assume responsibility for any knotes it may hold.
2335  */
2336 void
2337 devfs_assume_knotes(cdev_t dev, struct kqinfo *kqi)
2338 {
2339 	/*
2340 	 * Let kern/kern_event.c do the heavy lifting.
2341 	 */
2342 	knote_assume_knotes(kqi, &dev->si_kqinfo,
2343 			    &devfs_detached_filterops, (void *)dev);
2344 
2345 	/*
2346 	 * These should probably be activated individually, but doing so
2347 	 * would require refactoring kq's public in-kernel interface.
2348 	 */
2349 	KNOTE(&dev->si_kqinfo.ki_note, 0);
2350 }
2351 
2352 /*
2353  * Links a given cdev into the dev list.
2354  */
2355 int
2356 devfs_link_dev(cdev_t dev)
2357 {
2358 	KKASSERT((dev->si_flags & SI_DEVFS_LINKED) == 0);
2359 	dev->si_flags |= SI_DEVFS_LINKED;
2360 	TAILQ_INSERT_TAIL(&devfs_dev_list, dev, link);
2361 
2362 	return 0;
2363 }
2364 
2365 /*
2366  * Removes a given cdev from the dev list.  The caller is responsible for
2367  * releasing the reference on the device associated with the linkage.
2368  *
2369  * Returns EALREADY if the dev has already been unlinked.
2370  */
2371 static int
2372 devfs_unlink_dev(cdev_t dev)
2373 {
2374 	if ((dev->si_flags & SI_DEVFS_LINKED)) {
2375 		TAILQ_REMOVE(&devfs_dev_list, dev, link);
2376 		dev->si_flags &= ~SI_DEVFS_LINKED;
2377 		return (0);
2378 	}
2379 	return (EALREADY);
2380 }
2381 
2382 int
2383 devfs_node_is_accessible(struct devfs_node *node)
2384 {
2385 	if ((node) && (!(node->flags & DEVFS_HIDDEN)))
2386 		return 1;
2387 	else
2388 		return 0;
2389 }
2390 
2391 int
2392 devfs_reference_ops(struct dev_ops *ops)
2393 {
2394 	int unit;
2395 	struct devfs_dev_ops *found = NULL;
2396 	struct devfs_dev_ops *devops;
2397 
2398 	TAILQ_FOREACH(devops, &devfs_dev_ops_list, link) {
2399 		if (devops->ops == ops) {
2400 			found = devops;
2401 			break;
2402 		}
2403 	}
2404 
2405 	if (!found) {
2406 		found = kmalloc(sizeof(struct devfs_dev_ops), M_DEVFS, M_WAITOK);
2407 		found->ops = ops;
2408 		found->ref_count = 0;
2409 		TAILQ_INSERT_TAIL(&devfs_dev_ops_list, found, link);
2410 	}
2411 
2412 	KKASSERT(found);
2413 
2414 	if (found->ref_count == 0) {
2415 		found->id = devfs_clone_bitmap_get(&DEVFS_CLONE_BITMAP(ops_id), 255);
2416 		if (found->id == -1) {
2417 			/* Ran out of unique ids */
2418 			devfs_debug(DEVFS_DEBUG_WARNING,
2419 					"devfs_reference_ops: WARNING: ran out of unique ids\n");
2420 		}
2421 	}
2422 	unit = found->id;
2423 	++found->ref_count;
2424 
2425 	return unit;
2426 }
2427 
2428 void
2429 devfs_release_ops(struct dev_ops *ops)
2430 {
2431 	struct devfs_dev_ops *found = NULL;
2432 	struct devfs_dev_ops *devops;
2433 
2434 	TAILQ_FOREACH(devops, &devfs_dev_ops_list, link) {
2435 		if (devops->ops == ops) {
2436 			found = devops;
2437 			break;
2438 		}
2439 	}
2440 
2441 	KKASSERT(found);
2442 
2443 	--found->ref_count;
2444 
2445 	if (found->ref_count == 0) {
2446 		TAILQ_REMOVE(&devfs_dev_ops_list, found, link);
2447 		devfs_clone_bitmap_put(&DEVFS_CLONE_BITMAP(ops_id), found->id);
2448 		kfree(found, M_DEVFS);
2449 	}
2450 }
2451 
2452 /*
2453  * Wait for asynchronous messages to complete in the devfs helper
2454  * thread, then return.  Do nothing if the helper thread is dead
2455  * or we are being indirectly called from the helper thread itself.
2456  */
2457 void
2458 devfs_config(void)
2459 {
2460 	devfs_msg_t msg;
2461 
2462 	if (devfs_run && curthread != td_core) {
2463 		msg = devfs_msg_get();
2464 		msg = devfs_msg_send_sync(DEVFS_SYNC, msg);
2465 		devfs_msg_put(msg);
2466 	}
2467 }
2468 
2469 /*
2470  * Called on init of devfs; creates the objcaches and
2471  * spawns off the devfs core thread. Also initializes
2472  * locks.
2473  */
2474 static void
2475 devfs_init(void)
2476 {
2477 	devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_init() called\n");
2478 	/* Create objcaches for nodes, msgs and devs */
2479 	devfs_node_cache = objcache_create("devfs-node-cache", 0, 0,
2480 					   NULL, NULL, NULL,
2481 					   objcache_malloc_alloc,
2482 					   objcache_malloc_free,
2483 					   &devfs_node_malloc_args );
2484 
2485 	devfs_msg_cache = objcache_create("devfs-msg-cache", 0, 0,
2486 					  NULL, NULL, NULL,
2487 					  objcache_malloc_alloc,
2488 					  objcache_malloc_free,
2489 					  &devfs_msg_malloc_args );
2490 
2491 	devfs_dev_cache = objcache_create("devfs-dev-cache", 0, 0,
2492 					  NULL, NULL, NULL,
2493 					  objcache_malloc_alloc,
2494 					  objcache_malloc_free,
2495 					  &devfs_dev_malloc_args );
2496 
2497 	devfs_clone_bitmap_init(&DEVFS_CLONE_BITMAP(ops_id));
2498 
2499 	/* Initialize the reply-only port which acts as a message drain */
2500 	lwkt_initport_replyonly(&devfs_dispose_port, devfs_msg_autofree_reply);
2501 
2502 	/* Initialize *THE* devfs lock */
2503 	lockinit(&devfs_lock, "devfs_core lock", 0, 0);
2504 
2505 	lockmgr(&devfs_lock, LK_EXCLUSIVE);
2506 	lwkt_create(devfs_msg_core, /*args*/NULL, &td_core, NULL,
2507 		    0, -1, "devfs_msg_core");
2508 	while (devfs_run == 0)
2509 		lksleep(td_core, &devfs_lock, 0, "devfsc", 0);
2510 	lockmgr(&devfs_lock, LK_RELEASE);
2511 
2512 	devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_init finished\n");
2513 }
2514 
2515 /*
2516  * Called on unload of devfs; takes care of destroying the core
2517  * and the objcaches. Also removes aliases that are no longer needed.
2518  */
2519 static void
2520 devfs_uninit(void)
2521 {
2522 	devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_uninit() called\n");
2523 
2524 	devfs_msg_send(DEVFS_TERMINATE_CORE, NULL);
2525 	while (devfs_run)
2526 		tsleep(td_core, 0, "devfsc", hz*10);
2527 	tsleep(td_core, 0, "devfsc", hz);
2528 
2529 	devfs_clone_bitmap_uninit(&DEVFS_CLONE_BITMAP(ops_id));
2530 
2531 	/* Destroy the objcaches */
2532 	objcache_destroy(devfs_msg_cache);
2533 	objcache_destroy(devfs_node_cache);
2534 	objcache_destroy(devfs_dev_cache);
2535 
2536 	devfs_alias_reap();
2537 }
2538 
2539 /*
2540  * This is a sysctl handler to assist userland devname(3) to
2541  * find the device name for a given udev.
2542  */
2543 static int
2544 devfs_sysctl_devname_helper(SYSCTL_HANDLER_ARGS)
2545 {
2546 	udev_t 	udev;
2547 	cdev_t	found;
2548 	int		error;
2549 
2550 
2551 	if ((error = SYSCTL_IN(req, &udev, sizeof(udev_t))))
2552 		return (error);
2553 
2554 	devfs_debug(DEVFS_DEBUG_DEBUG, "devfs sysctl, received udev: %d\n", udev);
2555 
2556 	if (udev == NOUDEV)
2557 		return(EINVAL);
2558 
2559 	if ((found = devfs_find_device_by_udev(udev)) == NULL)
2560 		return(ENOENT);
2561 
2562 	return(SYSCTL_OUT(req, found->si_name, strlen(found->si_name) + 1));
2563 }
2564 
2565 
2566 SYSCTL_PROC(_kern, OID_AUTO, devname, CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_ANYBODY,
2567 			NULL, 0, devfs_sysctl_devname_helper, "", "helper for devname(3)");
2568 
2569 SYSCTL_NODE(_vfs, OID_AUTO, devfs, CTLFLAG_RW, 0, "devfs");
2570 TUNABLE_INT("vfs.devfs.debug", &devfs_debug_enable);
2571 SYSCTL_INT(_vfs_devfs, OID_AUTO, debug, CTLFLAG_RW, &devfs_debug_enable,
2572 		0, "Enable DevFS debugging");
2573 
2574 SYSINIT(vfs_devfs_register, SI_SUB_PRE_DRIVERS, SI_ORDER_FIRST,
2575 		devfs_init, NULL);
2576 SYSUNINIT(vfs_devfs_register, SI_SUB_PRE_DRIVERS, SI_ORDER_ANY,
2577 		devfs_uninit, NULL);
2578 
2579 /*
2580  * WildCmp() - compare wild string to sane string
2581  *
2582  *	Returns 0 on success, -1 on failure.
2583  */
2584 static int
2585 wildCmp(const char **mary, int d, const char *w, const char *s)
2586 {
2587     int i;
2588 
2589     /*
2590      * skip fixed portion
2591      */
2592     for (;;) {
2593 	switch(*w) {
2594 	case '*':
2595 	    /*
2596 	     * optimize terminator
2597 	     */
2598 	    if (w[1] == 0)
2599 		return(0);
2600 	    if (w[1] != '?' && w[1] != '*') {
2601 		/*
2602 		 * optimize * followed by non-wild
2603 		 */
2604 		for (i = 0; s + i < mary[d]; ++i) {
2605 		    if (s[i] == w[1] && wildCmp(mary, d + 1, w + 1, s + i) == 0)
2606 			return(0);
2607 		}
2608 	    } else {
2609 		/*
2610 		 * less-optimal
2611 		 */
2612 		for (i = 0; s + i < mary[d]; ++i) {
2613 		    if (wildCmp(mary, d + 1, w + 1, s + i) == 0)
2614 			return(0);
2615 		}
2616 	    }
2617 	    mary[d] = s;
2618 	    return(-1);
2619 	case '?':
2620 	    if (*s == 0)
2621 		return(-1);
2622 	    ++w;
2623 	    ++s;
2624 	    break;
2625 	default:
2626 	    if (*w != *s)
2627 		return(-1);
2628 	    if (*w == 0)	/* terminator */
2629 		return(0);
2630 	    ++w;
2631 	    ++s;
2632 	    break;
2633 	}
2634     }
2635     /* not reached */
2636     return(-1);
2637 }
2638 
2639 
2640 /*
2641  * WildCaseCmp() - compare wild string to sane string, case insensitive
2642  *
2643  *	Returns 0 on success, -1 on failure.
2644  */
2645 static int
2646 wildCaseCmp(const char **mary, int d, const char *w, const char *s)
2647 {
2648     int i;
2649 
2650     /*
2651      * skip fixed portion
2652      */
2653     for (;;) {
2654 	switch(*w) {
2655 	case '*':
2656 	    /*
2657 	     * optimize terminator
2658 	     */
2659 	    if (w[1] == 0)
2660 		return(0);
2661 	    if (w[1] != '?' && w[1] != '*') {
2662 		/*
2663 		 * optimize * followed by non-wild
2664 		 */
2665 		for (i = 0; s + i < mary[d]; ++i) {
2666 		    if (s[i] == w[1] && wildCaseCmp(mary, d + 1, w + 1, s + i) == 0)
2667 			return(0);
2668 		}
2669 	    } else {
2670 		/*
2671 		 * less-optimal
2672 		 */
2673 		for (i = 0; s + i < mary[d]; ++i) {
2674 		    if (wildCaseCmp(mary, d + 1, w + 1, s + i) == 0)
2675 			return(0);
2676 		}
2677 	    }
2678 	    mary[d] = s;
2679 	    return(-1);
2680 	case '?':
2681 	    if (*s == 0)
2682 		return(-1);
2683 	    ++w;
2684 	    ++s;
2685 	    break;
2686 	default:
2687 	    if (*w != *s) {
2688 #define tolower(x)	((x >= 'A' && x <= 'Z')?(x+('a'-'A')):(x))
2689 		if (tolower(*w) != tolower(*s))
2690 		    return(-1);
2691 	    }
2692 	    if (*w == 0)	/* terminator */
2693 		return(0);
2694 	    ++w;
2695 	    ++s;
2696 	    break;
2697 	}
2698     }
2699     /* not reached */
2700     return(-1);
2701 }
2702 
2703 int
2704 devfs_WildCmp(const char *w, const char *s)
2705 {
2706     int i;
2707     int c;
2708     int slen = strlen(s);
2709     const char **mary;
2710 
2711     for (i = c = 0; w[i]; ++i) {
2712 	if (w[i] == '*')
2713 	    ++c;
2714     }
2715     mary = kmalloc(sizeof(char *) * (c + 1), M_DEVFS, M_WAITOK);
2716     for (i = 0; i < c; ++i)
2717 	mary[i] = s + slen;
2718     i = wildCmp(mary, 0, w, s);
2719     kfree(mary, M_DEVFS);
2720     return(i);
2721 }
2722 
2723 int
2724 devfs_WildCaseCmp(const char *w, const char *s)
2725 {
2726     int i;
2727     int c;
2728     int slen = strlen(s);
2729     const char **mary;
2730 
2731     for (i = c = 0; w[i]; ++i) {
2732 	if (w[i] == '*')
2733 	    ++c;
2734     }
2735     mary = kmalloc(sizeof(char *) * (c + 1), M_DEVFS, M_WAITOK);
2736     for (i = 0; i < c; ++i)
2737 	mary[i] = s + slen;
2738     i = wildCaseCmp(mary, 0, w, s);
2739     kfree(mary, M_DEVFS);
2740     return(i);
2741 }
2742 
2743