xref: /freebsd/sys/fs/devfs/devfs_devs.c (revision d72a078647de20e31ab09295e671f1b8b4ad89b2) 
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 "opt_mac.h"
32 
33 #include <sys/param.h>
34 #include <sys/systm.h>
35 #include <sys/conf.h>
36 #include <sys/dirent.h>
37 #include <sys/kernel.h>
38 #include <sys/limits.h>
39 #include <sys/lock.h>
40 #include <sys/mac.h>
41 #include <sys/malloc.h>
42 #include <sys/proc.h>
43 #include <sys/sx.h>
44 #include <sys/sysctl.h>
45 #include <sys/vnode.h>
46 
47 #include <sys/kdb.h>
48 
49 #include <fs/devfs/devfs.h>
50 #include <fs/devfs/devfs_int.h>
51 
52 /*
53  * The one true (but secret) list of active devices in the system.
54  * Locked by dev_lock()/devmtx
55  */
56 static TAILQ_HEAD(,cdev_priv) cdevp_list = TAILQ_HEAD_INITIALIZER(cdevp_list);
57 
58 struct unrhdr *devfs_inos;
59 
60 
61 static MALLOC_DEFINE(M_DEVFS2, "DEVFS2", "DEVFS data 2");
62 static MALLOC_DEFINE(M_DEVFS3, "DEVFS3", "DEVFS data 3");
63 static MALLOC_DEFINE(M_CDEVP, "DEVFS1", "DEVFS cdev_priv storage");
64 
65 static SYSCTL_NODE(_vfs, OID_AUTO, devfs, CTLFLAG_RW, 0, "DEVFS filesystem");
66 
67 static unsigned devfs_generation;
68 SYSCTL_UINT(_vfs_devfs, OID_AUTO, generation, CTLFLAG_RD,
69 	&devfs_generation, 0, "DEVFS generation number");
70 
71 unsigned devfs_rule_depth = 1;
72 SYSCTL_UINT(_vfs_devfs, OID_AUTO, rule_depth, CTLFLAG_RW,
73 	&devfs_rule_depth, 0, "Max depth of ruleset include");
74 
75 /*
76  * Helper sysctl for devname(3).  We're given a struct cdev * and return
77  * the name, if any, registered by the device driver.
78  */
79 static int
80 sysctl_devname(SYSCTL_HANDLER_ARGS)
81 {
82 	int error;
83 	dev_t ud;
84 	struct cdev_priv *cdp;
85 
86 	error = SYSCTL_IN(req, &ud, sizeof (ud));
87 	if (error)
88 		return (error);
89 	if (ud == NODEV)
90 		return(EINVAL);
91 /*
92 	ud ^ devfs_random();
93 */
94 	dev_lock();
95 	TAILQ_FOREACH(cdp, &cdevp_list, cdp_list)
96 		if (cdp->cdp_inode == ud)
97 			break;
98 	dev_unlock();
99 	if (cdp == NULL)
100 		return(ENOENT);
101 	return(SYSCTL_OUT(req, cdp->cdp_c.si_name, strlen(cdp->cdp_c.si_name) + 1));
102 	return (error);
103 }
104 
105 SYSCTL_PROC(_kern, OID_AUTO, devname, CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_ANYBODY,
106 	NULL, 0, sysctl_devname, "", "devname(3) handler");
107 
108 SYSCTL_INT(_debug_sizeof, OID_AUTO, cdev, CTLFLAG_RD,
109     0, sizeof(struct cdev), "sizeof(struct cdev)");
110 
111 SYSCTL_INT(_debug_sizeof, OID_AUTO, cdev_priv, CTLFLAG_RD,
112     0, sizeof(struct cdev_priv), "sizeof(struct cdev_priv)");
113 
114 struct cdev *
115 devfs_alloc(void)
116 {
117 	struct cdev_priv *cdp;
118 	struct cdev *cdev;
119 
120 	cdp = malloc(sizeof *cdp, M_CDEVP, M_USE_RESERVE | M_ZERO | M_WAITOK);
121 
122 	cdp->cdp_dirents = &cdp->cdp_dirent0;
123 	cdp->cdp_dirent0 = NULL;
124 	cdp->cdp_maxdirent = 0;
125 
126 	cdev = &cdp->cdp_c;
127 	cdev->si_priv = cdp;
128 
129 	cdev->si_name = cdev->__si_namebuf;
130 	LIST_INIT(&cdev->si_children);
131 	return (cdev);
132 }
133 
134 void
135 devfs_free(struct cdev *cdev)
136 {
137 	struct cdev_priv *cdp;
138 
139 	cdp = cdev->si_priv;
140 	if (cdev->si_cred != NULL)
141 		crfree(cdev->si_cred);
142 	if (cdp->cdp_inode > 0)
143 		free_unr(devfs_inos, cdp->cdp_inode);
144 	if (cdp->cdp_maxdirent > 0)
145 		free(cdp->cdp_dirents, M_DEVFS2);
146 	free(cdp, M_CDEVP);
147 }
148 
149 struct devfs_dirent *
150 devfs_find(struct devfs_dirent *dd, const char *name, int namelen)
151 {
152 	struct devfs_dirent *de;
153 
154 	TAILQ_FOREACH(de, &dd->de_dlist, de_list) {
155 		if (namelen != de->de_dirent->d_namlen)
156 			continue;
157 		if (bcmp(name, de->de_dirent->d_name, namelen) != 0)
158 			continue;
159 		break;
160 	}
161 	return (de);
162 }
163 
164 struct devfs_dirent *
165 devfs_newdirent(char *name, int namelen)
166 {
167 	int i;
168 	struct devfs_dirent *de;
169 	struct dirent d;
170 
171 	d.d_namlen = namelen;
172 	i = sizeof (*de) + GENERIC_DIRSIZ(&d);
173 	de = malloc(i, M_DEVFS3, M_WAITOK | M_ZERO);
174 	de->de_dirent = (struct dirent *)(de + 1);
175 	de->de_dirent->d_namlen = namelen;
176 	de->de_dirent->d_reclen = GENERIC_DIRSIZ(&d);
177 	bcopy(name, de->de_dirent->d_name, namelen);
178 	de->de_dirent->d_name[namelen] = '\0';
179 	vfs_timestamp(&de->de_ctime);
180 	de->de_mtime = de->de_atime = de->de_ctime;
181 	de->de_links = 1;
182 	de->de_holdcnt = 1;
183 #ifdef MAC
184 	mac_init_devfsdirent(de);
185 #endif
186 	return (de);
187 }
188 
189 struct devfs_dirent *
190 devfs_vmkdir(struct devfs_mount *dmp, char *name, int namelen, struct devfs_dirent *dotdot, u_int inode)
191 {
192 	struct devfs_dirent *dd;
193 	struct devfs_dirent *de;
194 
195 	/* Create the new directory */
196 	dd = devfs_newdirent(name, namelen);
197 	TAILQ_INIT(&dd->de_dlist);
198 	dd->de_dirent->d_type = DT_DIR;
199 	dd->de_mode = 0555;
200 	dd->de_links = 2;
201 	dd->de_dir = dd;
202 	if (inode != 0)
203 		dd->de_inode = inode;
204 	else
205 		dd->de_inode = alloc_unr(devfs_inos);
206 
207 	/* Create the "." entry in the new directory */
208 	de = devfs_newdirent(".", 1);
209 	de->de_dirent->d_type = DT_DIR;
210 	de->de_flags |= DE_DOT;
211 	TAILQ_INSERT_TAIL(&dd->de_dlist, de, de_list);
212 	de->de_dir = dd;
213 
214 	/* Create the ".." entry in the new directory */
215 	de = devfs_newdirent("..", 2);
216 	de->de_dirent->d_type = DT_DIR;
217 	de->de_flags |= DE_DOTDOT;
218 	TAILQ_INSERT_TAIL(&dd->de_dlist, de, de_list);
219 	if (dotdot == NULL) {
220 		de->de_dir = dd;
221 	} else {
222 		de->de_dir = dotdot;
223 		TAILQ_INSERT_TAIL(&dotdot->de_dlist, dd, de_list);
224 		dotdot->de_links++;
225 	}
226 
227 #ifdef MAC
228 	mac_create_devfs_directory(dmp->dm_mount, name, namelen, dd);
229 #endif
230 	return (dd);
231 }
232 
233 void
234 devfs_dirent_free(struct devfs_dirent *de)
235 {
236 	free(de, M_DEVFS3);
237 }
238 
239 void
240 devfs_delete(struct devfs_mount *dm, struct devfs_dirent *de)
241 {
242 
243 	KASSERT((de->de_flags & DE_DOOMED) == 0,
244 		("devfs_delete doomed dirent"));
245 	de->de_flags |= DE_DOOMED;
246 	if (de->de_symlink) {
247 		free(de->de_symlink, M_DEVFS);
248 		de->de_symlink = NULL;
249 	}
250 	if (de->de_vnode != NULL) {
251 		vhold(de->de_vnode);
252 		de->de_vnode->v_data = NULL;
253 		vgone(de->de_vnode);
254 		vdrop(de->de_vnode);
255 		de->de_vnode = NULL;
256 	}
257 #ifdef MAC
258 	mac_destroy_devfsdirent(de);
259 #endif
260 	if (de->de_inode > DEVFS_ROOTINO) {
261 		free_unr(devfs_inos, de->de_inode);
262 		de->de_inode = 0;
263 	}
264 	if (DEVFS_DE_DROP(de))
265 		devfs_dirent_free(de);
266 }
267 
268 /*
269  * Called on unmount.
270  * Recursively removes the entire tree
271  */
272 
273 static void
274 devfs_purge(struct devfs_mount *dm, struct devfs_dirent *dd)
275 {
276 	struct devfs_dirent *de;
277 
278 	sx_assert(&dm->dm_lock, SX_XLOCKED);
279 	for (;;) {
280 		de = TAILQ_FIRST(&dd->de_dlist);
281 		if (de == NULL)
282 			break;
283 		TAILQ_REMOVE(&dd->de_dlist, de, de_list);
284 		if (de->de_flags & (DE_DOT|DE_DOTDOT))
285 			devfs_delete(dm, de);
286 		else if (de->de_dirent->d_type == DT_DIR)
287 			devfs_purge(dm, de);
288 		else
289 			devfs_delete(dm, de);
290 	}
291 	devfs_delete(dm, dd);
292 }
293 
294 /*
295  * Each cdev_priv has an array of pointers to devfs_dirent which is indexed
296  * by the mount points dm_idx.
297  * This function extends the array when necessary, taking into account that
298  * the default array is 1 element and not malloc'ed.
299  */
300 static void
301 devfs_metoo(struct cdev_priv *cdp, struct devfs_mount *dm)
302 {
303 	struct devfs_dirent **dep;
304 	int siz;
305 
306 	siz = (dm->dm_idx + 1) * sizeof *dep;
307 	dep = malloc(siz, M_DEVFS2, M_WAITOK | M_ZERO);
308 	dev_lock();
309 	if (dm->dm_idx <= cdp->cdp_maxdirent) {
310 		/* We got raced */
311 		dev_unlock();
312 		free(dep, M_DEVFS2);
313 		return;
314 	}
315 	memcpy(dep, cdp->cdp_dirents, (cdp->cdp_maxdirent + 1) * sizeof *dep);
316 	if (cdp->cdp_maxdirent > 0)
317 		free(cdp->cdp_dirents, M_DEVFS2);
318 	cdp->cdp_dirents = dep;
319 	/*
320 	 * XXX: if malloc told us how much we actually got this could
321 	 * XXX: be optimized.
322 	 */
323 	cdp->cdp_maxdirent = dm->dm_idx;
324 	dev_unlock();
325 }
326 
327 static int
328 devfs_populate_loop(struct devfs_mount *dm, int cleanup)
329 {
330 	struct cdev_priv *cdp;
331 	struct devfs_dirent *de;
332 	struct devfs_dirent *dd;
333 	struct cdev *pdev;
334 	int j;
335 	char *q, *s;
336 
337 	sx_assert(&dm->dm_lock, SX_XLOCKED);
338 	dev_lock();
339 	TAILQ_FOREACH(cdp, &cdevp_list, cdp_list) {
340 
341 		KASSERT(cdp->cdp_dirents != NULL, ("NULL cdp_dirents"));
342 
343 		/*
344 		 * If we are unmounting, or the device has been destroyed,
345 		 * clean up our dirent.
346 		 */
347 		if ((cleanup || !(cdp->cdp_flags & CDP_ACTIVE)) &&
348 		    dm->dm_idx <= cdp->cdp_maxdirent &&
349 		    cdp->cdp_dirents[dm->dm_idx] != NULL) {
350 			de = cdp->cdp_dirents[dm->dm_idx];
351 			cdp->cdp_dirents[dm->dm_idx] = NULL;
352 			cdp->cdp_inuse--;
353 			KASSERT(cdp == de->de_cdp,
354 			    ("%s %d %s %p %p", __func__, __LINE__,
355 			    cdp->cdp_c.si_name, cdp, de->de_cdp));
356 			KASSERT(de->de_dir != NULL, ("Null de->de_dir"));
357 			dev_unlock();
358 
359 			TAILQ_REMOVE(&de->de_dir->de_dlist, de, de_list);
360 			de->de_cdp = NULL;
361 			de->de_inode = 0;
362 			devfs_delete(dm, de);
363 			return (1);
364 		}
365 		/*
366 	 	 * GC any lingering devices
367 		 */
368 		if (!(cdp->cdp_flags & CDP_ACTIVE)) {
369 			if (cdp->cdp_inuse > 0)
370 				continue;
371 			TAILQ_REMOVE(&cdevp_list, cdp, cdp_list);
372 			dev_unlock();
373 			dev_rel(&cdp->cdp_c);
374 			return (1);
375 		}
376 		/*
377 		 * Don't create any new dirents if we are unmounting
378 		 */
379 		if (cleanup)
380 			continue;
381 		KASSERT((cdp->cdp_flags & CDP_ACTIVE), ("Bogons, I tell ya'!"));
382 
383 		if (dm->dm_idx <= cdp->cdp_maxdirent &&
384 		    cdp->cdp_dirents[dm->dm_idx] != NULL) {
385 			de = cdp->cdp_dirents[dm->dm_idx];
386 			KASSERT(cdp == de->de_cdp, ("inconsistent cdp"));
387 			continue;
388 		}
389 
390 
391 		cdp->cdp_inuse++;
392 		dev_unlock();
393 
394 		if (dm->dm_idx > cdp->cdp_maxdirent)
395 		        devfs_metoo(cdp, dm);
396 
397 		dd = dm->dm_rootdir;
398 		s = cdp->cdp_c.si_name;
399 		for (;;) {
400 			for (q = s; *q != '/' && *q != '\0'; q++)
401 				continue;
402 			if (*q != '/')
403 				break;
404 			de = devfs_find(dd, s, q - s);
405 			if (de == NULL)
406 				de = devfs_vmkdir(dm, s, q - s, dd, 0);
407 			s = q + 1;
408 			dd = de;
409 		}
410 
411 		de = devfs_newdirent(s, q - s);
412 		if (cdp->cdp_c.si_flags & SI_ALIAS) {
413 			de->de_uid = 0;
414 			de->de_gid = 0;
415 			de->de_mode = 0755;
416 			de->de_dirent->d_type = DT_LNK;
417 			pdev = cdp->cdp_c.si_parent;
418 			j = strlen(pdev->si_name) + 1;
419 			de->de_symlink = malloc(j, M_DEVFS, M_WAITOK);
420 			bcopy(pdev->si_name, de->de_symlink, j);
421 		} else {
422 			de->de_uid = cdp->cdp_c.si_uid;
423 			de->de_gid = cdp->cdp_c.si_gid;
424 			de->de_mode = cdp->cdp_c.si_mode;
425 			de->de_dirent->d_type = DT_CHR;
426 		}
427 		de->de_inode = cdp->cdp_inode;
428 		de->de_cdp = cdp;
429 #ifdef MAC
430 		mac_create_devfs_device(cdp->cdp_c.si_cred, dm->dm_mount,
431 		    &cdp->cdp_c, de);
432 #endif
433 		de->de_dir = dd;
434 		TAILQ_INSERT_TAIL(&dd->de_dlist, de, de_list);
435 		devfs_rules_apply(dm, de);
436 		dev_lock();
437 		/* XXX: could check that cdp is still active here */
438 		KASSERT(cdp->cdp_dirents[dm->dm_idx] == NULL,
439 		    ("%s %d\n", __func__, __LINE__));
440 		cdp->cdp_dirents[dm->dm_idx] = de;
441 		KASSERT(de->de_cdp != (void *)0xdeadc0de,
442 		    ("%s %d\n", __func__, __LINE__));
443 		dev_unlock();
444 		return (1);
445 	}
446 	dev_unlock();
447 	return (0);
448 }
449 
450 void
451 devfs_populate(struct devfs_mount *dm)
452 {
453 
454 	sx_assert(&dm->dm_lock, SX_XLOCKED);
455 	if (dm->dm_generation == devfs_generation)
456 		return;
457 	while (devfs_populate_loop(dm, 0))
458 		continue;
459 	dm->dm_generation = devfs_generation;
460 }
461 
462 void
463 devfs_cleanup(struct devfs_mount *dm)
464 {
465 
466 	sx_assert(&dm->dm_lock, SX_XLOCKED);
467 	while (devfs_populate_loop(dm, 1))
468 		continue;
469 	devfs_purge(dm, dm->dm_rootdir);
470 }
471 
472 /*
473  * devfs_create() and devfs_destroy() are called from kern_conf.c and
474  * in both cases the devlock() mutex is held, so no further locking
475  * is necesary and no sleeping allowed.
476  */
477 
478 void
479 devfs_create(struct cdev *dev)
480 {
481 	struct cdev_priv *cdp;
482 
483 	mtx_assert(&devmtx, MA_OWNED);
484 	cdp = dev->si_priv;
485 	cdp->cdp_flags |= CDP_ACTIVE;
486 	cdp->cdp_inode = alloc_unrl(devfs_inos);
487 	dev_refl(dev);
488 	TAILQ_INSERT_TAIL(&cdevp_list, cdp, cdp_list);
489 	devfs_generation++;
490 }
491 
492 void
493 devfs_destroy(struct cdev *dev)
494 {
495 	struct cdev_priv *cdp;
496 
497 	mtx_assert(&devmtx, MA_OWNED);
498 	cdp = dev->si_priv;
499 	cdp->cdp_flags &= ~CDP_ACTIVE;
500 	devfs_generation++;
501 }
502 
503 static void
504 devfs_devs_init(void *junk __unused)
505 {
506 
507 	devfs_inos = new_unrhdr(DEVFS_ROOTINO + 1, INT_MAX, &devmtx);
508 }
509 
510 SYSINIT(devfs_devs, SI_SUB_DEVFS, SI_ORDER_FIRST, devfs_devs_init, NULL);
511