xref: /freebsd/sys/fs/devfs/devfs_devs.c (revision d8b878873e7aa8df1972cc6a642804b17eb61087) 
1 /*-
2  * Copyright (c) 2000,2004
3  *	Poul-Henning Kamp.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Neither the name of the University nor the names of its contributors
11  *    may be used to endorse or promote products derived from this software
12  *    without specific prior written permission.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  *
26  * From: FreeBSD: src/sys/miscfs/kernfs/kernfs_vfsops.c 1.36
27  *
28  * $FreeBSD$
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 
60 static MALLOC_DEFINE(M_DEVFS2, "DEVFS2", "DEVFS data 2");
61 static MALLOC_DEFINE(M_DEVFS3, "DEVFS3", "DEVFS data 3");
62 static MALLOC_DEFINE(M_CDEVP, "DEVFS1", "DEVFS cdev_priv storage");
63 
64 static SYSCTL_NODE(_vfs, OID_AUTO, devfs, CTLFLAG_RW, 0, "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
79 sysctl_devname(SYSCTL_HANDLER_ARGS)
80 {
81 	int error;
82 	dev_t ud;
83 	struct cdev_priv *cdp;
84 	struct cdev *dev;
85 
86 	error = SYSCTL_IN(req, &ud, sizeof (ud));
87 	if (error)
88 		return (error);
89 	if (ud == NODEV)
90 		return (EINVAL);
91 	dev = NULL;
92 	dev_lock();
93 	TAILQ_FOREACH(cdp, &cdevp_list, cdp_list)
94 		if (cdp->cdp_inode == ud) {
95 			dev = &cdp->cdp_c;
96 			dev_refl(dev);
97 			break;
98 		}
99 	dev_unlock();
100 	if (dev == NULL)
101 		return (ENOENT);
102 	error = SYSCTL_OUT(req, dev->si_name, strlen(dev->si_name) + 1);
103 	dev_rel(dev);
104 	return (error);
105 }
106 
107 SYSCTL_PROC(_kern, OID_AUTO, devname,
108     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_ANYBODY|CTLFLAG_MPSAFE,
109     NULL, 0, sysctl_devname, "", "devname(3) handler");
110 
111 SYSCTL_INT(_debug_sizeof, OID_AUTO, cdev, CTLFLAG_RD,
112     0, sizeof(struct cdev), "sizeof(struct cdev)");
113 
114 SYSCTL_INT(_debug_sizeof, OID_AUTO, cdev_priv, CTLFLAG_RD,
115     0, sizeof(struct cdev_priv), "sizeof(struct cdev_priv)");
116 
117 struct cdev *
118 devfs_alloc(void)
119 {
120 	struct cdev_priv *cdp;
121 	struct cdev *cdev;
122 	struct timespec ts;
123 
124 	cdp = malloc(sizeof *cdp, M_CDEVP, M_USE_RESERVE | M_ZERO | M_WAITOK);
125 
126 	cdp->cdp_dirents = &cdp->cdp_dirent0;
127 	cdp->cdp_dirent0 = NULL;
128 	cdp->cdp_maxdirent = 0;
129 
130 	cdev = &cdp->cdp_c;
131 
132 	cdev->si_name = cdev->__si_namebuf;
133 	LIST_INIT(&cdev->si_children);
134 	vfs_timestamp(&ts);
135 	cdev->si_atime = cdev->si_mtime = cdev->si_ctime = ts;
136 
137 	return (cdev);
138 }
139 
140 void
141 devfs_free(struct cdev *cdev)
142 {
143 	struct cdev_priv *cdp;
144 
145 	cdp = cdev2priv(cdev);
146 	if (cdev->si_cred != NULL)
147 		crfree(cdev->si_cred);
148 	if (cdp->cdp_inode > 0)
149 		free_unr(devfs_inos, cdp->cdp_inode);
150 	if (cdp->cdp_maxdirent > 0)
151 		free(cdp->cdp_dirents, M_DEVFS2);
152 	free(cdp, M_CDEVP);
153 }
154 
155 struct devfs_dirent *
156 devfs_find(struct devfs_dirent *dd, const char *name, int namelen)
157 {
158 	struct devfs_dirent *de;
159 
160 	TAILQ_FOREACH(de, &dd->de_dlist, de_list) {
161 		if (namelen != de->de_dirent->d_namlen)
162 			continue;
163 		if (bcmp(name, de->de_dirent->d_name, namelen) != 0)
164 			continue;
165 		break;
166 	}
167 	return (de);
168 }
169 
170 struct devfs_dirent *
171 devfs_newdirent(char *name, int namelen)
172 {
173 	int i;
174 	struct devfs_dirent *de;
175 	struct dirent d;
176 
177 	d.d_namlen = namelen;
178 	i = sizeof (*de) + GENERIC_DIRSIZ(&d);
179 	de = malloc(i, M_DEVFS3, M_WAITOK | M_ZERO);
180 	de->de_dirent = (struct dirent *)(de + 1);
181 	de->de_dirent->d_namlen = namelen;
182 	de->de_dirent->d_reclen = GENERIC_DIRSIZ(&d);
183 	bcopy(name, de->de_dirent->d_name, namelen);
184 	de->de_dirent->d_name[namelen] = '\0';
185 	vfs_timestamp(&de->de_ctime);
186 	de->de_mtime = de->de_atime = de->de_ctime;
187 	de->de_links = 1;
188 	de->de_holdcnt = 1;
189 #ifdef MAC
190 	mac_devfs_init(de);
191 #endif
192 	return (de);
193 }
194 
195 struct devfs_dirent *
196 devfs_vmkdir(struct devfs_mount *dmp, char *name, int namelen, struct devfs_dirent *dotdot, u_int inode)
197 {
198 	struct devfs_dirent *dd;
199 	struct devfs_dirent *de;
200 
201 	/* Create the new directory */
202 	dd = devfs_newdirent(name, namelen);
203 	TAILQ_INIT(&dd->de_dlist);
204 	dd->de_dirent->d_type = DT_DIR;
205 	dd->de_mode = 0555;
206 	dd->de_links = 2;
207 	dd->de_dir = dd;
208 	if (inode != 0)
209 		dd->de_inode = inode;
210 	else
211 		dd->de_inode = alloc_unr(devfs_inos);
212 
213 	/* Create the "." entry in the new directory */
214 	de = devfs_newdirent(".", 1);
215 	de->de_dirent->d_type = DT_DIR;
216 	de->de_flags |= DE_DOT;
217 	TAILQ_INSERT_TAIL(&dd->de_dlist, de, de_list);
218 	de->de_dir = dd;
219 
220 	/* Create the ".." entry in the new directory */
221 	de = devfs_newdirent("..", 2);
222 	de->de_dirent->d_type = DT_DIR;
223 	de->de_flags |= DE_DOTDOT;
224 	TAILQ_INSERT_TAIL(&dd->de_dlist, de, de_list);
225 	if (dotdot == NULL) {
226 		de->de_dir = dd;
227 	} else {
228 		de->de_dir = dotdot;
229 		TAILQ_INSERT_TAIL(&dotdot->de_dlist, dd, de_list);
230 		dotdot->de_links++;
231 	}
232 
233 #ifdef MAC
234 	mac_devfs_create_directory(dmp->dm_mount, name, namelen, dd);
235 #endif
236 	return (dd);
237 }
238 
239 void
240 devfs_dirent_free(struct devfs_dirent *de)
241 {
242 	free(de, M_DEVFS3);
243 }
244 
245 /*
246  * The caller needs to hold the dm for the duration of the call since
247  * dm->dm_lock may be temporary dropped.
248  */
249 void
250 devfs_delete(struct devfs_mount *dm, struct devfs_dirent *de, int vp_locked)
251 {
252 	struct vnode *vp;
253 
254 	KASSERT((de->de_flags & DE_DOOMED) == 0,
255 		("devfs_delete doomed dirent"));
256 	de->de_flags |= DE_DOOMED;
257 	mtx_lock(&devfs_de_interlock);
258 	vp = de->de_vnode;
259 	if (vp != NULL) {
260 		VI_LOCK(vp);
261 		mtx_unlock(&devfs_de_interlock);
262 		vholdl(vp);
263 		sx_unlock(&dm->dm_lock);
264 		if (!vp_locked)
265 			vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY);
266 		else
267 			VI_UNLOCK(vp);
268 		vgone(vp);
269 		if (!vp_locked)
270 			VOP_UNLOCK(vp, 0);
271 		vdrop(vp);
272 		sx_xlock(&dm->dm_lock);
273 	} else
274 		mtx_unlock(&devfs_de_interlock);
275 	if (de->de_symlink) {
276 		free(de->de_symlink, M_DEVFS);
277 		de->de_symlink = NULL;
278 	}
279 #ifdef MAC
280 	mac_devfs_destroy(de);
281 #endif
282 	if (de->de_inode > DEVFS_ROOTINO) {
283 		free_unr(devfs_inos, de->de_inode);
284 		de->de_inode = 0;
285 	}
286 	if (DEVFS_DE_DROP(de))
287 		devfs_dirent_free(de);
288 }
289 
290 /*
291  * Called on unmount.
292  * Recursively removes the entire tree.
293  * The caller needs to hold the dm for the duration of the call.
294  */
295 
296 static void
297 devfs_purge(struct devfs_mount *dm, struct devfs_dirent *dd)
298 {
299 	struct devfs_dirent *de;
300 
301 	sx_assert(&dm->dm_lock, SX_XLOCKED);
302 	for (;;) {
303 		de = TAILQ_FIRST(&dd->de_dlist);
304 		if (de == NULL)
305 			break;
306 		TAILQ_REMOVE(&dd->de_dlist, de, de_list);
307 		if (de->de_flags & (DE_DOT|DE_DOTDOT))
308 			devfs_delete(dm, de, 0);
309 		else if (de->de_dirent->d_type == DT_DIR)
310 			devfs_purge(dm, de);
311 		else
312 			devfs_delete(dm, de, 0);
313 	}
314 	devfs_delete(dm, dd, 0);
315 }
316 
317 /*
318  * Each cdev_priv has an array of pointers to devfs_dirent which is indexed
319  * by the mount points dm_idx.
320  * This function extends the array when necessary, taking into account that
321  * the default array is 1 element and not malloc'ed.
322  */
323 static void
324 devfs_metoo(struct cdev_priv *cdp, struct devfs_mount *dm)
325 {
326 	struct devfs_dirent **dep;
327 	int siz;
328 
329 	siz = (dm->dm_idx + 1) * sizeof *dep;
330 	dep = malloc(siz, M_DEVFS2, M_WAITOK | M_ZERO);
331 	dev_lock();
332 	if (dm->dm_idx <= cdp->cdp_maxdirent) {
333 		/* We got raced */
334 		dev_unlock();
335 		free(dep, M_DEVFS2);
336 		return;
337 	}
338 	memcpy(dep, cdp->cdp_dirents, (cdp->cdp_maxdirent + 1) * sizeof *dep);
339 	if (cdp->cdp_maxdirent > 0)
340 		free(cdp->cdp_dirents, M_DEVFS2);
341 	cdp->cdp_dirents = dep;
342 	/*
343 	 * XXX: if malloc told us how much we actually got this could
344 	 * XXX: be optimized.
345 	 */
346 	cdp->cdp_maxdirent = dm->dm_idx;
347 	dev_unlock();
348 }
349 
350 /*
351  * The caller needs to hold the dm for the duration of the call.
352  */
353 static int
354 devfs_populate_loop(struct devfs_mount *dm, int cleanup)
355 {
356 	struct cdev_priv *cdp;
357 	struct devfs_dirent *de;
358 	struct devfs_dirent *dd;
359 	struct cdev *pdev;
360 	int j;
361 	char *q, *s;
362 
363 	sx_assert(&dm->dm_lock, SX_XLOCKED);
364 	dev_lock();
365 	TAILQ_FOREACH(cdp, &cdevp_list, cdp_list) {
366 
367 		KASSERT(cdp->cdp_dirents != NULL, ("NULL cdp_dirents"));
368 
369 		/*
370 		 * If we are unmounting, or the device has been destroyed,
371 		 * clean up our dirent.
372 		 */
373 		if ((cleanup || !(cdp->cdp_flags & CDP_ACTIVE)) &&
374 		    dm->dm_idx <= cdp->cdp_maxdirent &&
375 		    cdp->cdp_dirents[dm->dm_idx] != NULL) {
376 			de = cdp->cdp_dirents[dm->dm_idx];
377 			cdp->cdp_dirents[dm->dm_idx] = NULL;
378 			KASSERT(cdp == de->de_cdp,
379 			    ("%s %d %s %p %p", __func__, __LINE__,
380 			    cdp->cdp_c.si_name, cdp, de->de_cdp));
381 			KASSERT(de->de_dir != NULL, ("Null de->de_dir"));
382 			dev_unlock();
383 
384 			TAILQ_REMOVE(&de->de_dir->de_dlist, de, de_list);
385 			de->de_cdp = NULL;
386 			de->de_inode = 0;
387 			devfs_delete(dm, de, 0);
388 			dev_lock();
389 			cdp->cdp_inuse--;
390 			dev_unlock();
391 			return (1);
392 		}
393 		/*
394 	 	 * GC any lingering devices
395 		 */
396 		if (!(cdp->cdp_flags & CDP_ACTIVE)) {
397 			if (cdp->cdp_inuse > 0)
398 				continue;
399 			TAILQ_REMOVE(&cdevp_list, cdp, cdp_list);
400 			dev_unlock();
401 			dev_rel(&cdp->cdp_c);
402 			return (1);
403 		}
404 		/*
405 		 * Don't create any new dirents if we are unmounting
406 		 */
407 		if (cleanup)
408 			continue;
409 		KASSERT((cdp->cdp_flags & CDP_ACTIVE), ("Bogons, I tell ya'!"));
410 
411 		if (dm->dm_idx <= cdp->cdp_maxdirent &&
412 		    cdp->cdp_dirents[dm->dm_idx] != NULL) {
413 			de = cdp->cdp_dirents[dm->dm_idx];
414 			KASSERT(cdp == de->de_cdp, ("inconsistent cdp"));
415 			continue;
416 		}
417 
418 
419 		cdp->cdp_inuse++;
420 		dev_unlock();
421 
422 		if (dm->dm_idx > cdp->cdp_maxdirent)
423 		        devfs_metoo(cdp, dm);
424 
425 		dd = dm->dm_rootdir;
426 		s = cdp->cdp_c.si_name;
427 		for (;;) {
428 			for (q = s; *q != '/' && *q != '\0'; q++)
429 				continue;
430 			if (*q != '/')
431 				break;
432 			de = devfs_find(dd, s, q - s);
433 			if (de == NULL)
434 				de = devfs_vmkdir(dm, s, q - s, dd, 0);
435 			s = q + 1;
436 			dd = de;
437 		}
438 
439 		de = devfs_newdirent(s, q - s);
440 		if (cdp->cdp_c.si_flags & SI_ALIAS) {
441 			de->de_uid = 0;
442 			de->de_gid = 0;
443 			de->de_mode = 0755;
444 			de->de_dirent->d_type = DT_LNK;
445 			pdev = cdp->cdp_c.si_parent;
446 			j = strlen(pdev->si_name) + 1;
447 			de->de_symlink = malloc(j, M_DEVFS, M_WAITOK);
448 			bcopy(pdev->si_name, de->de_symlink, j);
449 		} else {
450 			de->de_uid = cdp->cdp_c.si_uid;
451 			de->de_gid = cdp->cdp_c.si_gid;
452 			de->de_mode = cdp->cdp_c.si_mode;
453 			de->de_dirent->d_type = DT_CHR;
454 		}
455 		de->de_inode = cdp->cdp_inode;
456 		de->de_cdp = cdp;
457 #ifdef MAC
458 		mac_devfs_create_device(cdp->cdp_c.si_cred, dm->dm_mount,
459 		    &cdp->cdp_c, de);
460 #endif
461 		de->de_dir = dd;
462 		TAILQ_INSERT_TAIL(&dd->de_dlist, de, de_list);
463 		devfs_rules_apply(dm, de);
464 		dev_lock();
465 		/* XXX: could check that cdp is still active here */
466 		KASSERT(cdp->cdp_dirents[dm->dm_idx] == NULL,
467 		    ("%s %d\n", __func__, __LINE__));
468 		cdp->cdp_dirents[dm->dm_idx] = de;
469 		KASSERT(de->de_cdp != (void *)0xdeadc0de,
470 		    ("%s %d\n", __func__, __LINE__));
471 		dev_unlock();
472 		return (1);
473 	}
474 	dev_unlock();
475 	return (0);
476 }
477 
478 /*
479  * The caller needs to hold the dm for the duration of the call.
480  */
481 void
482 devfs_populate(struct devfs_mount *dm)
483 {
484 
485 	sx_assert(&dm->dm_lock, SX_XLOCKED);
486 	if (dm->dm_generation == devfs_generation)
487 		return;
488 	while (devfs_populate_loop(dm, 0))
489 		continue;
490 	dm->dm_generation = devfs_generation;
491 }
492 
493 /*
494  * The caller needs to hold the dm for the duration of the call.
495  */
496 void
497 devfs_cleanup(struct devfs_mount *dm)
498 {
499 
500 	sx_assert(&dm->dm_lock, SX_XLOCKED);
501 	while (devfs_populate_loop(dm, 1))
502 		continue;
503 	devfs_purge(dm, dm->dm_rootdir);
504 }
505 
506 /*
507  * devfs_create() and devfs_destroy() are called from kern_conf.c and
508  * in both cases the devlock() mutex is held, so no further locking
509  * is necesary and no sleeping allowed.
510  */
511 
512 void
513 devfs_create(struct cdev *dev)
514 {
515 	struct cdev_priv *cdp;
516 
517 	mtx_assert(&devmtx, MA_OWNED);
518 	cdp = cdev2priv(dev);
519 	cdp->cdp_flags |= CDP_ACTIVE;
520 	cdp->cdp_inode = alloc_unrl(devfs_inos);
521 	dev_refl(dev);
522 	TAILQ_INSERT_TAIL(&cdevp_list, cdp, cdp_list);
523 	devfs_generation++;
524 }
525 
526 void
527 devfs_destroy(struct cdev *dev)
528 {
529 	struct cdev_priv *cdp;
530 
531 	mtx_assert(&devmtx, MA_OWNED);
532 	cdp = cdev2priv(dev);
533 	cdp->cdp_flags &= ~CDP_ACTIVE;
534 	devfs_generation++;
535 }
536 
537 static void
538 devfs_devs_init(void *junk __unused)
539 {
540 
541 	devfs_inos = new_unrhdr(DEVFS_ROOTINO + 1, INT_MAX, &devmtx);
542 }
543 
544 SYSINIT(devfs_devs, SI_SUB_DEVFS, SI_ORDER_FIRST, devfs_devs_init, NULL);
545