xref: /freebsd/sys/fs/devfs/devfs_devs.c (revision 69f1cb3c91c3377cedc28a9fe37673bda10602cd) 
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 #ifdef MAC
183 	mac_init_devfsdirent(de);
184 #endif
185 	return (de);
186 }
187 
188 struct devfs_dirent *
189 devfs_vmkdir(struct devfs_mount *dmp, char *name, int namelen, struct devfs_dirent *dotdot, u_int inode)
190 {
191 	struct devfs_dirent *dd;
192 	struct devfs_dirent *de;
193 
194 	/* Create the new directory */
195 	dd = devfs_newdirent(name, namelen);
196 	TAILQ_INIT(&dd->de_dlist);
197 	dd->de_dirent->d_type = DT_DIR;
198 	dd->de_mode = 0555;
199 	dd->de_links = 2;
200 	dd->de_dir = dd;
201 	if (inode != 0)
202 		dd->de_inode = inode;
203 	else
204 		dd->de_inode = alloc_unr(devfs_inos);
205 
206 	/* Create the "." entry in the new directory */
207 	de = devfs_newdirent(".", 1);
208 	de->de_dirent->d_type = DT_DIR;
209 	de->de_flags |= DE_DOT;
210 	TAILQ_INSERT_TAIL(&dd->de_dlist, de, de_list);
211 	de->de_dir = dd;
212 
213 	/* Create the ".." entry in the new directory */
214 	de = devfs_newdirent("..", 2);
215 	de->de_dirent->d_type = DT_DIR;
216 	de->de_flags |= DE_DOTDOT;
217 	TAILQ_INSERT_TAIL(&dd->de_dlist, de, de_list);
218 	if (dotdot == NULL) {
219 		de->de_dir = dd;
220 	} else {
221 		de->de_dir = dotdot;
222 		TAILQ_INSERT_TAIL(&dotdot->de_dlist, dd, de_list);
223 		dotdot->de_links++;
224 	}
225 
226 #ifdef MAC
227 	mac_create_devfs_directory(dmp->dm_mount, name, namelen, dd);
228 #endif
229 	return (dd);
230 }
231 
232 void
233 devfs_delete(struct devfs_mount *dm, struct devfs_dirent *de)
234 {
235 
236 	if (de->de_symlink) {
237 		free(de->de_symlink, M_DEVFS);
238 		de->de_symlink = NULL;
239 	}
240 	if (de->de_vnode != NULL) {
241 		vhold(de->de_vnode);
242 		de->de_vnode->v_data = NULL;
243 		vgone(de->de_vnode);
244 		vdrop(de->de_vnode);
245 		de->de_vnode = NULL;
246 	}
247 #ifdef MAC
248 	mac_destroy_devfsdirent(de);
249 #endif
250 	if (de->de_inode > DEVFS_ROOTINO) {
251 		free_unr(devfs_inos, de->de_inode);
252 		de->de_inode = 0;
253 	}
254 	free(de, M_DEVFS3);
255 }
256 
257 /*
258  * Called on unmount.
259  * Recursively removes the entire tree
260  */
261 
262 static void
263 devfs_purge(struct devfs_mount *dm, struct devfs_dirent *dd)
264 {
265 	struct devfs_dirent *de;
266 
267 	sx_assert(&dm->dm_lock, SX_XLOCKED);
268 	for (;;) {
269 		de = TAILQ_FIRST(&dd->de_dlist);
270 		if (de == NULL)
271 			break;
272 		TAILQ_REMOVE(&dd->de_dlist, de, de_list);
273 		if (de->de_flags & (DE_DOT|DE_DOTDOT))
274 			devfs_delete(dm, de);
275 		else if (de->de_dirent->d_type == DT_DIR)
276 			devfs_purge(dm, de);
277 		else
278 			devfs_delete(dm, de);
279 	}
280 	devfs_delete(dm, dd);
281 }
282 
283 /*
284  * Each cdev_priv has an array of pointers to devfs_dirent which is indexed
285  * by the mount points dm_idx.
286  * This function extends the array when necessary, taking into account that
287  * the default array is 1 element and not malloc'ed.
288  */
289 static void
290 devfs_metoo(struct cdev_priv *cdp, struct devfs_mount *dm)
291 {
292 	struct devfs_dirent **dep;
293 	int siz;
294 
295 	siz = (dm->dm_idx + 1) * sizeof *dep;
296 	dep = malloc(siz, M_DEVFS2, M_WAITOK | M_ZERO);
297 	dev_lock();
298 	if (dm->dm_idx <= cdp->cdp_maxdirent) {
299 		/* We got raced */
300 		dev_unlock();
301 		free(dep, M_DEVFS2);
302 		return;
303 	}
304 	memcpy(dep, cdp->cdp_dirents, (cdp->cdp_maxdirent + 1) * sizeof *dep);
305 	if (cdp->cdp_maxdirent > 0)
306 		free(cdp->cdp_dirents, M_DEVFS2);
307 	cdp->cdp_dirents = dep;
308 	/*
309 	 * XXX: if malloc told us how much we actually got this could
310 	 * XXX: be optimized.
311 	 */
312 	cdp->cdp_maxdirent = dm->dm_idx;
313 	dev_unlock();
314 }
315 
316 static int
317 devfs_populate_loop(struct devfs_mount *dm, int cleanup)
318 {
319 	struct cdev_priv *cdp;
320 	struct devfs_dirent *de;
321 	struct devfs_dirent *dd;
322 	struct cdev *pdev;
323 	int j;
324 	char *q, *s;
325 
326 	sx_assert(&dm->dm_lock, SX_XLOCKED);
327 	dev_lock();
328 	TAILQ_FOREACH(cdp, &cdevp_list, cdp_list) {
329 
330 		KASSERT(cdp->cdp_dirents != NULL, ("NULL cdp_dirents"));
331 
332 		/*
333 		 * If we are unmounting, or the device has been destroyed,
334 		 * clean up our dirent.
335 		 */
336 		if ((cleanup || !(cdp->cdp_flags & CDP_ACTIVE)) &&
337 		    dm->dm_idx <= cdp->cdp_maxdirent &&
338 		    cdp->cdp_dirents[dm->dm_idx] != NULL) {
339 			de = cdp->cdp_dirents[dm->dm_idx];
340 			cdp->cdp_dirents[dm->dm_idx] = NULL;
341 			cdp->cdp_inuse--;
342 			KASSERT(cdp == de->de_cdp,
343 			    ("%s %d %s %p %p", __func__, __LINE__,
344 			    cdp->cdp_c.si_name, cdp, de->de_cdp));
345 			KASSERT(de->de_dir != NULL, ("Null de->de_dir"));
346 			dev_unlock();
347 
348 			TAILQ_REMOVE(&de->de_dir->de_dlist, de, de_list);
349 			de->de_cdp = NULL;
350 			de->de_inode = 0;
351 			devfs_delete(dm, de);
352 			return (1);
353 		}
354 		/*
355 	 	 * GC any lingering devices
356 		 */
357 		if (!(cdp->cdp_flags & CDP_ACTIVE)) {
358 			if (cdp->cdp_inuse > 0)
359 				continue;
360 			TAILQ_REMOVE(&cdevp_list, cdp, cdp_list);
361 			dev_unlock();
362 			dev_rel(&cdp->cdp_c);
363 			return (1);
364 		}
365 		/*
366 		 * Don't create any new dirents if we are unmounting
367 		 */
368 		if (cleanup)
369 			continue;
370 		KASSERT((cdp->cdp_flags & CDP_ACTIVE), ("Bogons, I tell ya'!"));
371 
372 		if (dm->dm_idx <= cdp->cdp_maxdirent &&
373 		    cdp->cdp_dirents[dm->dm_idx] != NULL) {
374 			de = cdp->cdp_dirents[dm->dm_idx];
375 			KASSERT(cdp == de->de_cdp, ("inconsistent cdp"));
376 			continue;
377 		}
378 
379 
380 		cdp->cdp_inuse++;
381 		dev_unlock();
382 
383 		if (dm->dm_idx > cdp->cdp_maxdirent)
384 		        devfs_metoo(cdp, dm);
385 
386 		dd = dm->dm_rootdir;
387 		s = cdp->cdp_c.si_name;
388 		for (;;) {
389 			for (q = s; *q != '/' && *q != '\0'; q++)
390 				continue;
391 			if (*q != '/')
392 				break;
393 			de = devfs_find(dd, s, q - s);
394 			if (de == NULL)
395 				de = devfs_vmkdir(dm, s, q - s, dd, 0);
396 			s = q + 1;
397 			dd = de;
398 		}
399 
400 		de = devfs_newdirent(s, q - s);
401 		if (cdp->cdp_c.si_flags & SI_ALIAS) {
402 			de->de_uid = 0;
403 			de->de_gid = 0;
404 			de->de_mode = 0755;
405 			de->de_dirent->d_type = DT_LNK;
406 			pdev = cdp->cdp_c.si_parent;
407 			j = strlen(pdev->si_name) + 1;
408 			de->de_symlink = malloc(j, M_DEVFS, M_WAITOK);
409 			bcopy(pdev->si_name, de->de_symlink, j);
410 		} else {
411 			de->de_uid = cdp->cdp_c.si_uid;
412 			de->de_gid = cdp->cdp_c.si_gid;
413 			de->de_mode = cdp->cdp_c.si_mode;
414 			de->de_dirent->d_type = DT_CHR;
415 		}
416 		de->de_inode = cdp->cdp_inode;
417 		de->de_cdp = cdp;
418 #ifdef MAC
419 		mac_create_devfs_device(cdp->cdp_c.si_cred, dm->dm_mount,
420 		    &cdp->cdp_c, de);
421 #endif
422 		de->de_dir = dd;
423 		TAILQ_INSERT_TAIL(&dd->de_dlist, de, de_list);
424 		devfs_rules_apply(dm, de);
425 		dev_lock();
426 		/* XXX: could check that cdp is still active here */
427 		KASSERT(cdp->cdp_dirents[dm->dm_idx] == NULL,
428 		    ("%s %d\n", __func__, __LINE__));
429 		cdp->cdp_dirents[dm->dm_idx] = de;
430 		KASSERT(de->de_cdp != (void *)0xdeadc0de,
431 		    ("%s %d\n", __func__, __LINE__));
432 		dev_unlock();
433 		return (1);
434 	}
435 	dev_unlock();
436 	return (0);
437 }
438 
439 void
440 devfs_populate(struct devfs_mount *dm)
441 {
442 
443 	sx_assert(&dm->dm_lock, SX_XLOCKED);
444 	if (dm->dm_generation == devfs_generation)
445 		return;
446 	while (devfs_populate_loop(dm, 0))
447 		continue;
448 	dm->dm_generation = devfs_generation;
449 }
450 
451 void
452 devfs_cleanup(struct devfs_mount *dm)
453 {
454 
455 	sx_assert(&dm->dm_lock, SX_XLOCKED);
456 	while (devfs_populate_loop(dm, 1))
457 		continue;
458 	devfs_purge(dm, dm->dm_rootdir);
459 }
460 
461 /*
462  * devfs_create() and devfs_destroy() are called from kern_conf.c and
463  * in both cases the devlock() mutex is held, so no further locking
464  * is necesary and no sleeping allowed.
465  */
466 
467 void
468 devfs_create(struct cdev *dev)
469 {
470 	struct cdev_priv *cdp;
471 
472 	mtx_assert(&devmtx, MA_OWNED);
473 	cdp = dev->si_priv;
474 	cdp->cdp_flags |= CDP_ACTIVE;
475 	cdp->cdp_inode = alloc_unrl(devfs_inos);
476 	dev_refl(dev);
477 	TAILQ_INSERT_TAIL(&cdevp_list, cdp, cdp_list);
478 	devfs_generation++;
479 }
480 
481 void
482 devfs_destroy(struct cdev *dev)
483 {
484 	struct cdev_priv *cdp;
485 
486 	mtx_assert(&devmtx, MA_OWNED);
487 	cdp = dev->si_priv;
488 	cdp->cdp_flags &= ~CDP_ACTIVE;
489 	devfs_generation++;
490 }
491 
492 static void
493 devfs_devs_init(void *junk __unused)
494 {
495 
496 	devfs_inos = new_unrhdr(DEVFS_ROOTINO + 1, INT_MAX, &devmtx);
497 }
498 
499 SYSINIT(devfs_devs, SI_SUB_DEVFS, SI_ORDER_FIRST, devfs_devs_init, NULL);
500