xref: /freebsd/sys/fs/devfs/devfs_devs.c (revision 48d6b52add36cc09e7fb1fbec44ab66c0742f320)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2000,2004
5  *	Poul-Henning Kamp.  All rights reserved.
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  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Neither the name of the University nor the names of its contributors
13  *    may be used to endorse or promote products derived from this software
14  *    without specific prior written permission.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  *
28  * From: FreeBSD: src/sys/miscfs/kernfs/kernfs_vfsops.c 1.36
29  */
30 
31 #include <sys/param.h>
32 #include <sys/systm.h>
33 #include <sys/conf.h>
34 #include <sys/dirent.h>
35 #include <sys/kernel.h>
36 #include <sys/limits.h>
37 #include <sys/lock.h>
38 #include <sys/malloc.h>
39 #include <sys/proc.h>
40 #include <sys/sx.h>
41 #include <sys/sysctl.h>
42 #include <sys/vnode.h>
43 
44 #include <sys/kdb.h>
45 
46 #include <fs/devfs/devfs.h>
47 #include <fs/devfs/devfs_int.h>
48 
49 #include <security/mac/mac_framework.h>
50 
51 /*
52  * The one true (but secret) list of active devices in the system.
53  * Locked by dev_lock()/devmtx
54  */
55 struct cdev_priv_list cdevp_list = TAILQ_HEAD_INITIALIZER(cdevp_list);
56 
57 struct unrhdr *devfs_inos;
58 
59 static MALLOC_DEFINE(M_DEVFS2, "DEVFS2", "DEVFS data 2");
60 static MALLOC_DEFINE(M_DEVFS3, "DEVFS3", "DEVFS data 3");
61 static MALLOC_DEFINE(M_CDEVP, "DEVFS1", "DEVFS cdev_priv storage");
62 
63 SYSCTL_NODE(_vfs, OID_AUTO, devfs, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
64     "DEVFS filesystem");
65 
66 static unsigned devfs_generation;
67 SYSCTL_UINT(_vfs_devfs, OID_AUTO, generation, CTLFLAG_RD,
68 	&devfs_generation, 0, "DEVFS generation number");
69 
70 unsigned devfs_rule_depth = 1;
71 SYSCTL_UINT(_vfs_devfs, OID_AUTO, rule_depth, CTLFLAG_RW,
72 	&devfs_rule_depth, 0, "Max depth of ruleset include");
73 
74 /*
75  * Helper sysctl for devname(3).  We're given a dev_t and return the
76  * name, if any, registered by the device driver.
77  */
78 static int
sysctl_devname(SYSCTL_HANDLER_ARGS)79 sysctl_devname(SYSCTL_HANDLER_ARGS)
80 {
81 	int error;
82 	dev_t ud;
83 #ifdef COMPAT_FREEBSD11
84 	uint32_t ud_compat;
85 #endif
86 	struct cdev_priv *cdp;
87 	struct cdev *dev;
88 
89 	if (req->newptr == NULL)
90 		return (EINVAL);
91 
92 #ifdef COMPAT_FREEBSD11
93 	if (req->newlen == sizeof(ud_compat)) {
94 		error = SYSCTL_IN(req, &ud_compat, sizeof(ud_compat));
95 		if (error == 0)
96 			ud = ud_compat == (uint32_t)NODEV ? NODEV : ud_compat;
97 	} else
98 #endif
99 		error = SYSCTL_IN(req, &ud, sizeof (ud));
100 	if (error)
101 		return (error);
102 	if (ud == NODEV)
103 		return (EINVAL);
104 	dev = NULL;
105 	dev_lock();
106 	TAILQ_FOREACH(cdp, &cdevp_list, cdp_list)
107 		if (cdp->cdp_inode == ud) {
108 			dev = &cdp->cdp_c;
109 			dev_refl(dev);
110 			break;
111 		}
112 	dev_unlock();
113 	if (dev == NULL)
114 		return (ENOENT);
115 	error = SYSCTL_OUT(req, dev->si_name, strlen(dev->si_name) + 1);
116 	dev_rel(dev);
117 	return (error);
118 }
119 
120 SYSCTL_PROC(_kern, OID_AUTO, devname,
121     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_ANYBODY|CTLFLAG_MPSAFE,
122     NULL, 0, sysctl_devname, "", "devname(3) handler");
123 
124 SYSCTL_INT(_debug_sizeof, OID_AUTO, cdev, CTLFLAG_RD,
125     SYSCTL_NULL_INT_PTR, sizeof(struct cdev), "sizeof(struct cdev)");
126 
127 SYSCTL_INT(_debug_sizeof, OID_AUTO, cdev_priv, CTLFLAG_RD,
128     SYSCTL_NULL_INT_PTR, sizeof(struct cdev_priv), "sizeof(struct cdev_priv)");
129 
130 struct cdev *
devfs_alloc(int flags)131 devfs_alloc(int flags)
132 {
133 	struct cdev_priv *cdp;
134 	struct cdev *cdev;
135 	struct timespec ts;
136 
137 	cdp = malloc(sizeof *cdp, M_CDEVP, M_ZERO |
138 	    ((flags & MAKEDEV_NOWAIT) ? M_NOWAIT : M_WAITOK));
139 	if (cdp == NULL)
140 		return (NULL);
141 
142 	mtx_init(&cdp->cdp_threadlock, "devthrd", NULL, MTX_DEF);
143 
144 	cdp->cdp_dirents = &cdp->cdp_dirent0;
145 
146 	cdev = &cdp->cdp_c;
147 	LIST_INIT(&cdev->si_children);
148 	vfs_timestamp(&ts);
149 	cdev->si_atime = cdev->si_mtime = cdev->si_ctime = ts;
150 
151 	return (cdev);
152 }
153 
154 int
devfs_dev_exists(const char * name)155 devfs_dev_exists(const char *name)
156 {
157 	struct cdev_priv *cdp;
158 
159 	dev_lock_assert_locked();
160 
161 	TAILQ_FOREACH(cdp, &cdevp_list, cdp_list) {
162 		if ((cdp->cdp_flags & CDP_ACTIVE) == 0)
163 			continue;
164 		if (devfs_pathpath(cdp->cdp_c.si_name, name) != 0)
165 			return (1);
166 		if (devfs_pathpath(name, cdp->cdp_c.si_name) != 0)
167 			return (1);
168 	}
169 	if (devfs_dir_find(name) != 0)
170 		return (1);
171 
172 	return (0);
173 }
174 
175 void
devfs_free(struct cdev * cdev)176 devfs_free(struct cdev *cdev)
177 {
178 	struct cdev_priv *cdp;
179 
180 	cdp = cdev2priv(cdev);
181 	KASSERT((cdp->cdp_flags & (CDP_ACTIVE | CDP_ON_ACTIVE_LIST)) == 0,
182 	    ("%s: cdp %p (%s) still on active list",
183 	    __func__, cdp, cdev->si_name));
184 	if (cdev->si_cred != NULL)
185 		crfree(cdev->si_cred);
186 	devfs_free_cdp_inode(cdp->cdp_inode);
187 	if (cdp->cdp_maxdirent > 0)
188 		free(cdp->cdp_dirents, M_DEVFS2);
189 	mtx_destroy(&cdp->cdp_threadlock);
190 	free(cdp, M_CDEVP);
191 }
192 
193 struct devfs_dirent *
devfs_find(struct devfs_dirent * dd,const char * name,int namelen,int type)194 devfs_find(struct devfs_dirent *dd, const char *name, int namelen, int type)
195 {
196 	struct devfs_dirent *de;
197 
198 	TAILQ_FOREACH(de, &dd->de_dlist, de_list) {
199 		if (namelen != de->de_dirent->d_namlen)
200 			continue;
201 		if (type != 0 && type != de->de_dirent->d_type)
202 			continue;
203 
204 		/*
205 		 * The race with finding non-active name is not
206 		 * completely closed by the check, but it is similar
207 		 * to the devfs_allocv() in making it unlikely enough.
208 		 */
209 		if (de->de_dirent->d_type == DT_CHR &&
210 		    (de->de_cdp->cdp_flags & CDP_ACTIVE) == 0)
211 			continue;
212 
213 		if (bcmp(name, de->de_dirent->d_name, namelen) != 0)
214 			continue;
215 		break;
216 	}
217 	KASSERT(de == NULL || (de->de_flags & DE_DOOMED) == 0,
218 	    ("devfs_find: returning a doomed entry"));
219 	return (de);
220 }
221 
222 struct devfs_dirent *
devfs_newdirent(char * name,int namelen)223 devfs_newdirent(char *name, int namelen)
224 {
225 	int i;
226 	struct devfs_dirent *de;
227 	struct dirent d;
228 
229 	d.d_namlen = namelen;
230 	i = sizeof(*de) + GENERIC_DIRSIZ(&d);
231 	de = malloc(i, M_DEVFS3, M_WAITOK | M_ZERO);
232 	de->de_dirent = (struct dirent *)(de + 1);
233 	de->de_dirent->d_namlen = namelen;
234 	de->de_dirent->d_reclen = GENERIC_DIRSIZ(&d);
235 	bcopy(name, de->de_dirent->d_name, namelen);
236 	dirent_terminate(de->de_dirent);
237 	vfs_timestamp(&de->de_ctime);
238 	de->de_mtime = de->de_atime = de->de_ctime;
239 	de->de_links = 1;
240 	de->de_holdcnt = 1;
241 #ifdef MAC
242 	mac_devfs_init(de);
243 #endif
244 	return (de);
245 }
246 
247 struct devfs_dirent *
devfs_parent_dirent(struct devfs_dirent * de)248 devfs_parent_dirent(struct devfs_dirent *de)
249 {
250 
251 	if (de->de_dirent->d_type != DT_DIR)
252 		return (de->de_dir);
253 
254 	if (de->de_flags & (DE_DOT | DE_DOTDOT))
255 		return (NULL);
256 
257 	de = TAILQ_FIRST(&de->de_dlist);	/* "." */
258 	if (de == NULL)
259 		return (NULL);
260 	de = TAILQ_NEXT(de, de_list);		/* ".." */
261 	if (de == NULL)
262 		return (NULL);
263 
264 	return (de->de_dir);
265 }
266 
267 struct devfs_dirent *
devfs_vmkdir(struct devfs_mount * dmp,char * name,int namelen,struct devfs_dirent * dotdot,u_int inode)268 devfs_vmkdir(struct devfs_mount *dmp, char *name, int namelen,
269     struct devfs_dirent *dotdot, u_int inode)
270 {
271 	struct devfs_dirent *dd;
272 	struct devfs_dirent *de;
273 
274 	/* Create the new directory */
275 	dd = devfs_newdirent(name, namelen);
276 	TAILQ_INIT(&dd->de_dlist);
277 	dd->de_dirent->d_type = DT_DIR;
278 	dd->de_mode = 0555;
279 	dd->de_links = 2;
280 	dd->de_dir = dd;
281 	if (inode != 0)
282 		dd->de_inode = inode;
283 	else
284 		dd->de_inode = alloc_unr(devfs_inos);
285 
286 	/*
287 	 * "." and ".." are always the two first entries in the
288 	 * de_dlist list.
289 	 *
290 	 * Create the "." entry in the new directory.
291 	 */
292 	de = devfs_newdirent(".", 1);
293 	de->de_dirent->d_type = DT_DIR;
294 	de->de_flags |= DE_DOT;
295 	TAILQ_INSERT_TAIL(&dd->de_dlist, de, de_list);
296 	de->de_dir = dd;
297 
298 	/* Create the ".." entry in the new directory. */
299 	de = devfs_newdirent("..", 2);
300 	de->de_dirent->d_type = DT_DIR;
301 	de->de_flags |= DE_DOTDOT;
302 	TAILQ_INSERT_TAIL(&dd->de_dlist, de, de_list);
303 	if (dotdot == NULL) {
304 		de->de_dir = dd;
305 	} else {
306 		de->de_dir = dotdot;
307 		sx_assert(&dmp->dm_lock, SX_XLOCKED);
308 		TAILQ_INSERT_TAIL(&dotdot->de_dlist, dd, de_list);
309 		dotdot->de_links++;
310 		devfs_rules_apply(dmp, dd);
311 	}
312 
313 #ifdef MAC
314 	mac_devfs_create_directory(dmp->dm_mount, name, namelen, dd);
315 #endif
316 	return (dd);
317 }
318 
319 void
devfs_dirent_free(struct devfs_dirent * de)320 devfs_dirent_free(struct devfs_dirent *de)
321 {
322 	struct vnode *vp;
323 
324 	vp = de->de_vnode;
325 	mtx_lock(&devfs_de_interlock);
326 	if (vp != NULL && vp->v_data == de)
327 		vp->v_data = NULL;
328 	mtx_unlock(&devfs_de_interlock);
329 	free(de, M_DEVFS3);
330 }
331 
332 /*
333  * Removes a directory if it is empty. Also empty parent directories are
334  * removed recursively.
335  */
336 static void
devfs_rmdir_empty(struct devfs_mount * dm,struct devfs_dirent * de)337 devfs_rmdir_empty(struct devfs_mount *dm, struct devfs_dirent *de)
338 {
339 	struct devfs_dirent *dd, *de_dot, *de_dotdot;
340 
341 	sx_assert(&dm->dm_lock, SX_XLOCKED);
342 
343 	for (;;) {
344 		KASSERT(de->de_dirent->d_type == DT_DIR,
345 		    ("devfs_rmdir_empty: de is not a directory"));
346 
347 		if ((de->de_flags & DE_DOOMED) != 0 || de == dm->dm_rootdir)
348 			return;
349 
350 		de_dot = TAILQ_FIRST(&de->de_dlist);
351 		KASSERT(de_dot != NULL, ("devfs_rmdir_empty: . missing"));
352 		de_dotdot = TAILQ_NEXT(de_dot, de_list);
353 		KASSERT(de_dotdot != NULL, ("devfs_rmdir_empty: .. missing"));
354 		/* Return if the directory is not empty. */
355 		if (TAILQ_NEXT(de_dotdot, de_list) != NULL)
356 			return;
357 
358 		dd = devfs_parent_dirent(de);
359 		KASSERT(dd != NULL, ("devfs_rmdir_empty: NULL dd"));
360 		TAILQ_REMOVE(&de->de_dlist, de_dot, de_list);
361 		TAILQ_REMOVE(&de->de_dlist, de_dotdot, de_list);
362 		TAILQ_REMOVE(&dd->de_dlist, de, de_list);
363 		DEVFS_DE_HOLD(dd);
364 		devfs_delete(dm, de, DEVFS_DEL_NORECURSE);
365 		devfs_delete(dm, de_dot, DEVFS_DEL_NORECURSE);
366 		devfs_delete(dm, de_dotdot, DEVFS_DEL_NORECURSE);
367 		if (DEVFS_DE_DROP(dd)) {
368 			devfs_dirent_free(dd);
369 			return;
370 		}
371 
372 		de = dd;
373 	}
374 }
375 
376 /*
377  * The caller needs to hold the dm for the duration of the call since
378  * dm->dm_lock may be temporary dropped.
379  */
380 void
devfs_delete(struct devfs_mount * dm,struct devfs_dirent * de,int flags)381 devfs_delete(struct devfs_mount *dm, struct devfs_dirent *de, int flags)
382 {
383 	struct devfs_dirent *dd;
384 	struct vnode *vp;
385 
386 	KASSERT((de->de_flags & DE_DOOMED) == 0,
387 		("devfs_delete doomed dirent"));
388 	de->de_flags |= DE_DOOMED;
389 
390 	if ((flags & DEVFS_DEL_NORECURSE) == 0) {
391 		dd = devfs_parent_dirent(de);
392 		if (dd != NULL)
393 			DEVFS_DE_HOLD(dd);
394 		if (de->de_flags & DE_USER) {
395 			KASSERT(dd != NULL, ("devfs_delete: NULL dd"));
396 			devfs_dir_unref_de(dm, dd);
397 		}
398 	} else
399 		dd = NULL;
400 
401 	mtx_lock(&devfs_de_interlock);
402 	vp = de->de_vnode;
403 	if (vp != NULL) {
404 		vhold(vp);
405 		mtx_unlock(&devfs_de_interlock);
406 		sx_unlock(&dm->dm_lock);
407 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
408 		vgone(vp);
409 		VOP_UNLOCK(vp);
410 		vdrop(vp);
411 		sx_xlock(&dm->dm_lock);
412 	} else
413 		mtx_unlock(&devfs_de_interlock);
414 	if (de->de_symlink) {
415 		free(de->de_symlink, M_DEVFS);
416 		de->de_symlink = NULL;
417 	}
418 #ifdef MAC
419 	mac_devfs_destroy(de);
420 #endif
421 	if (de->de_inode > DEVFS_ROOTINO) {
422 		devfs_free_cdp_inode(de->de_inode);
423 		de->de_inode = 0;
424 	}
425 	if (DEVFS_DE_DROP(de))
426 		devfs_dirent_free(de);
427 
428 	if (dd != NULL) {
429 		if (DEVFS_DE_DROP(dd))
430 			devfs_dirent_free(dd);
431 		else
432 			devfs_rmdir_empty(dm, dd);
433 	}
434 }
435 
436 /*
437  * Called on unmount.
438  * Recursively removes the entire tree.
439  * The caller needs to hold the dm for the duration of the call.
440  */
441 
442 static void
devfs_purge(struct devfs_mount * dm,struct devfs_dirent * dd)443 devfs_purge(struct devfs_mount *dm, struct devfs_dirent *dd)
444 {
445 	struct devfs_dirent *de;
446 
447 	sx_assert(&dm->dm_lock, SX_XLOCKED);
448 
449 	DEVFS_DE_HOLD(dd);
450 	for (;;) {
451 		/*
452 		 * Use TAILQ_LAST() to remove "." and ".." last.
453 		 * We might need ".." to resolve a path in
454 		 * devfs_dir_unref_de().
455 		 */
456 		de = TAILQ_LAST(&dd->de_dlist, devfs_dlist_head);
457 		if (de == NULL)
458 			break;
459 		TAILQ_REMOVE(&dd->de_dlist, de, de_list);
460 		if (de->de_flags & DE_USER)
461 			devfs_dir_unref_de(dm, dd);
462 		if (de->de_flags & (DE_DOT | DE_DOTDOT))
463 			devfs_delete(dm, de, DEVFS_DEL_NORECURSE);
464 		else if (de->de_dirent->d_type == DT_DIR)
465 			devfs_purge(dm, de);
466 		else
467 			devfs_delete(dm, de, DEVFS_DEL_NORECURSE);
468 	}
469 	if (DEVFS_DE_DROP(dd))
470 		devfs_dirent_free(dd);
471 	else if ((dd->de_flags & DE_DOOMED) == 0)
472 		devfs_delete(dm, dd, DEVFS_DEL_NORECURSE);
473 }
474 
475 /*
476  * Each cdev_priv has an array of pointers to devfs_dirent which is indexed
477  * by the mount points dm_idx.
478  * This function extends the array when necessary, taking into account that
479  * the default array is 1 element and not malloc'ed.
480  */
481 static void
devfs_metoo(struct cdev_priv * cdp,struct devfs_mount * dm)482 devfs_metoo(struct cdev_priv *cdp, struct devfs_mount *dm)
483 {
484 	struct devfs_dirent **dep, **olddep;
485 	int siz;
486 
487 	siz = (dm->dm_idx + 1) * sizeof *dep;
488 	dep = malloc(siz, M_DEVFS2, M_WAITOK | M_ZERO);
489 	dev_lock();
490 	if (dm->dm_idx <= cdp->cdp_maxdirent) {
491 		/* We got raced */
492 		dev_unlock();
493 		free(dep, M_DEVFS2);
494 		return;
495 	}
496 	memcpy(dep, cdp->cdp_dirents, (cdp->cdp_maxdirent + 1) * sizeof *dep);
497 	olddep = cdp->cdp_maxdirent > 0 ? cdp->cdp_dirents : NULL;
498 	cdp->cdp_dirents = dep;
499 	/*
500 	 * XXX: if malloc told us how much we actually got this could
501 	 * XXX: be optimized.
502 	 */
503 	cdp->cdp_maxdirent = dm->dm_idx;
504 	dev_unlock();
505 	free(olddep, M_DEVFS2);
506 }
507 
508 /*
509  * The caller needs to hold the dm for the duration of the call.
510  */
511 static int
devfs_populate_loop(struct devfs_mount * dm,int cleanup)512 devfs_populate_loop(struct devfs_mount *dm, int cleanup)
513 {
514 	struct cdev_priv *cdp;
515 	struct devfs_dirent *de;
516 	struct devfs_dirent *dd, *dt;
517 	struct cdev *pdev;
518 	int de_flags, depth, j;
519 	char *q, *s;
520 
521 	sx_assert(&dm->dm_lock, SX_XLOCKED);
522 	dev_lock();
523 	TAILQ_FOREACH(cdp, &cdevp_list, cdp_list) {
524 		KASSERT(cdp->cdp_dirents != NULL, ("NULL cdp_dirents"));
525 		KASSERT((cdp->cdp_flags & CDP_ON_ACTIVE_LIST) != 0,
526 		    ("%s: cdp %p (%s) should not be on active list",
527 		    __func__, cdp, cdp->cdp_c.si_name));
528 
529 		/*
530 		 * If we are unmounting, or the device has been destroyed,
531 		 * clean up our dirent.
532 		 */
533 		if ((cleanup || !(cdp->cdp_flags & CDP_ACTIVE)) &&
534 		    dm->dm_idx <= cdp->cdp_maxdirent &&
535 		    cdp->cdp_dirents[dm->dm_idx] != NULL) {
536 			de = cdp->cdp_dirents[dm->dm_idx];
537 			cdp->cdp_dirents[dm->dm_idx] = NULL;
538 			KASSERT(cdp == de->de_cdp,
539 			    ("%s %d %s %p %p", __func__, __LINE__,
540 			    cdp->cdp_c.si_name, cdp, de->de_cdp));
541 			KASSERT(de->de_dir != NULL, ("Null de->de_dir"));
542 			dev_unlock();
543 
544 			TAILQ_REMOVE(&de->de_dir->de_dlist, de, de_list);
545 			de->de_cdp = NULL;
546 			de->de_inode = 0;
547 			devfs_delete(dm, de, 0);
548 			dev_lock();
549 			cdp->cdp_inuse--;
550 			dev_unlock();
551 			return (1);
552 		}
553 		/*
554 	 	 * GC any lingering devices
555 		 */
556 		if (!(cdp->cdp_flags & CDP_ACTIVE)) {
557 			if (cdp->cdp_inuse > 0)
558 				continue;
559 			cdp->cdp_flags &= ~CDP_ON_ACTIVE_LIST;
560 			TAILQ_REMOVE(&cdevp_list, cdp, cdp_list);
561 			dev_unlock();
562 			dev_rel(&cdp->cdp_c);
563 			return (1);
564 		}
565 		/*
566 		 * Don't create any new dirents if we are unmounting
567 		 */
568 		if (cleanup)
569 			continue;
570 		KASSERT((cdp->cdp_flags & CDP_ACTIVE), ("Bogons, I tell ya'!"));
571 
572 		if (dm->dm_idx <= cdp->cdp_maxdirent &&
573 		    cdp->cdp_dirents[dm->dm_idx] != NULL) {
574 			de = cdp->cdp_dirents[dm->dm_idx];
575 			KASSERT(cdp == de->de_cdp, ("inconsistent cdp"));
576 			continue;
577 		}
578 
579 		cdp->cdp_inuse++;
580 		dev_unlock();
581 
582 		if (dm->dm_idx > cdp->cdp_maxdirent)
583 		        devfs_metoo(cdp, dm);
584 
585 		dd = dm->dm_rootdir;
586 		s = cdp->cdp_c.si_name;
587 		for (;;) {
588 			for (q = s; *q != '/' && *q != '\0'; q++)
589 				continue;
590 			if (*q != '/')
591 				break;
592 			de = devfs_find(dd, s, q - s, 0);
593 			if (de == NULL)
594 				de = devfs_vmkdir(dm, s, q - s, dd, 0);
595 			else if (de->de_dirent->d_type == DT_LNK) {
596 				de = devfs_find(dd, s, q - s, DT_DIR);
597 				if (de == NULL)
598 					de = devfs_vmkdir(dm, s, q - s, dd, 0);
599 				de->de_flags |= DE_COVERED;
600 			}
601 			s = q + 1;
602 			dd = de;
603 			KASSERT(dd->de_dirent->d_type == DT_DIR &&
604 			    (dd->de_flags & (DE_DOT | DE_DOTDOT)) == 0,
605 			    ("%s: invalid directory (si_name=%s)",
606 			    __func__, cdp->cdp_c.si_name));
607 		}
608 		de_flags = 0;
609 		de = devfs_find(dd, s, q - s, DT_LNK);
610 		if (de != NULL)
611 			de_flags |= DE_COVERED;
612 
613 		de = devfs_newdirent(s, q - s);
614 		if (cdp->cdp_c.si_flags & SI_ALIAS) {
615 			de->de_uid = 0;
616 			de->de_gid = 0;
617 			de->de_mode = 0755;
618 			de->de_dirent->d_type = DT_LNK;
619 			pdev = cdp->cdp_c.si_parent;
620 			dt = dd;
621 			depth = 0;
622 			while (dt != dm->dm_rootdir &&
623 			    (dt = devfs_parent_dirent(dt)) != NULL)
624 				depth++;
625 			j = depth * 3 + strlen(pdev->si_name) + 1;
626 			de->de_symlink = malloc(j, M_DEVFS, M_WAITOK);
627 			de->de_symlink[0] = 0;
628 			while (depth-- > 0)
629 				strcat(de->de_symlink, "../");
630 			strcat(de->de_symlink, pdev->si_name);
631 		} else {
632 			de->de_uid = cdp->cdp_c.si_uid;
633 			de->de_gid = cdp->cdp_c.si_gid;
634 			de->de_mode = cdp->cdp_c.si_mode;
635 			de->de_dirent->d_type = DT_CHR;
636 		}
637 		de->de_flags |= de_flags;
638 		de->de_inode = cdp->cdp_inode;
639 		de->de_cdp = cdp;
640 #ifdef MAC
641 		mac_devfs_create_device(cdp->cdp_c.si_cred, dm->dm_mount,
642 		    &cdp->cdp_c, de);
643 #endif
644 		de->de_dir = dd;
645 		TAILQ_INSERT_TAIL(&dd->de_dlist, de, de_list);
646 		devfs_rules_apply(dm, de);
647 		dev_lock();
648 		/* XXX: could check that cdp is still active here */
649 		KASSERT(cdp->cdp_dirents[dm->dm_idx] == NULL,
650 		    ("%s %d\n", __func__, __LINE__));
651 		cdp->cdp_dirents[dm->dm_idx] = de;
652 		KASSERT(de->de_cdp != (void *)0xdeadc0de,
653 		    ("%s %d\n", __func__, __LINE__));
654 		dev_unlock();
655 		return (1);
656 	}
657 	dev_unlock();
658 	return (0);
659 }
660 
661 int
devfs_populate_needed(struct devfs_mount * dm)662 devfs_populate_needed(struct devfs_mount *dm)
663 {
664 
665 	return (dm->dm_generation != devfs_generation);
666 }
667 
668 /*
669  * The caller needs to hold the dm for the duration of the call.
670  */
671 void
devfs_populate(struct devfs_mount * dm)672 devfs_populate(struct devfs_mount *dm)
673 {
674 	unsigned gen;
675 
676 	sx_assert(&dm->dm_lock, SX_XLOCKED);
677 	if (!devfs_populate_needed(dm))
678 		return;
679 	gen = devfs_generation;
680 	while (devfs_populate_loop(dm, 0))
681 		continue;
682 	dm->dm_generation = gen;
683 }
684 
685 /*
686  * The caller needs to hold the dm for the duration of the call.
687  */
688 void
devfs_cleanup(struct devfs_mount * dm)689 devfs_cleanup(struct devfs_mount *dm)
690 {
691 
692 	sx_assert(&dm->dm_lock, SX_XLOCKED);
693 	while (devfs_populate_loop(dm, 1))
694 		continue;
695 	devfs_purge(dm, dm->dm_rootdir);
696 }
697 
698 /*
699  * devfs_create() and devfs_destroy() are called from kern_conf.c and
700  * in both cases the devlock() mutex is held, so no further locking
701  * is necessary and no sleeping allowed.
702  */
703 
704 void
devfs_create(struct cdev * dev)705 devfs_create(struct cdev *dev)
706 {
707 	struct cdev_priv *cdp;
708 
709 	dev_lock_assert_locked();
710 	cdp = cdev2priv(dev);
711 	KASSERT((cdp->cdp_flags & CDP_ON_ACTIVE_LIST) == 0,
712 	    ("%s: cdp %p (%s) already on active list",
713 	    __func__, cdp, dev->si_name));
714 	cdp->cdp_flags |= (CDP_ACTIVE | CDP_ON_ACTIVE_LIST);
715 	cdp->cdp_inode = alloc_unrl(devfs_inos);
716 	dev_refl(dev);
717 	TAILQ_INSERT_TAIL(&cdevp_list, cdp, cdp_list);
718 	devfs_generation++;
719 }
720 
721 void
devfs_destroy(struct cdev * dev)722 devfs_destroy(struct cdev *dev)
723 {
724 	struct cdev_priv *cdp;
725 
726 	dev_lock_assert_locked();
727 	cdp = cdev2priv(dev);
728 	cdp->cdp_flags &= ~CDP_ACTIVE;
729 	devfs_generation++;
730 }
731 
732 ino_t
devfs_alloc_cdp_inode(void)733 devfs_alloc_cdp_inode(void)
734 {
735 
736 	return (alloc_unr(devfs_inos));
737 }
738 
739 void
devfs_free_cdp_inode(ino_t ino)740 devfs_free_cdp_inode(ino_t ino)
741 {
742 
743 	if (ino > 0)
744 		free_unr(devfs_inos, ino);
745 }
746 
747 static void
devfs_devs_init(void * junk __unused)748 devfs_devs_init(void *junk __unused)
749 {
750 
751 	devfs_inos = new_unrhdr(DEVFS_ROOTINO + 1, INT_MAX, &devmtx);
752 }
753 
754 SYSINIT(devfs_devs, SI_SUB_DEVFS, SI_ORDER_FIRST, devfs_devs_init, NULL);
755