2 * Copyright (c) 2000,2004
3 * Poul-Henning Kamp. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
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.
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
26 * From: FreeBSD: src/sys/miscfs/kernfs/kernfs_vfsops.c 1.36
31 #include <sys/param.h>
32 #include <sys/systm.h>
34 #include <sys/dirent.h>
35 #include <sys/kernel.h>
36 #include <sys/limits.h>
38 #include <sys/malloc.h>
41 #include <sys/sysctl.h>
42 #include <sys/vnode.h>
46 #include <fs/devfs/devfs.h>
47 #include <fs/devfs/devfs_int.h>
49 #include <security/mac/mac_framework.h>
52 * The one true (but secret) list of active devices in the system.
53 * Locked by dev_lock()/devmtx
55 struct cdev_priv_list cdevp_list = TAILQ_HEAD_INITIALIZER(cdevp_list);
57 struct unrhdr *devfs_inos;
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");
64 SYSCTL_NODE(_vfs, OID_AUTO, devfs, CTLFLAG_RW, 0, "DEVFS filesystem");
66 static unsigned devfs_generation;
67 SYSCTL_UINT(_vfs_devfs, OID_AUTO, generation, CTLFLAG_RD,
68 &devfs_generation, 0, "DEVFS generation number");
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");
75 * Helper sysctl for devname(3). We're given a dev_t and return the
76 * name, if any, registered by the device driver.
79 sysctl_devname(SYSCTL_HANDLER_ARGS)
83 struct cdev_priv *cdp;
86 error = SYSCTL_IN(req, &ud, sizeof (ud));
93 TAILQ_FOREACH(cdp, &cdevp_list, cdp_list)
94 if (cdp->cdp_inode == ud) {
102 error = SYSCTL_OUT(req, dev->si_name, strlen(dev->si_name) + 1);
107 SYSCTL_PROC(_kern, OID_AUTO, devname,
108 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_ANYBODY|CTLFLAG_MPSAFE,
109 NULL, 0, sysctl_devname, "", "devname(3) handler");
111 SYSCTL_INT(_debug_sizeof, OID_AUTO, cdev, CTLFLAG_RD,
112 SYSCTL_NULL_INT_PTR, sizeof(struct cdev), "sizeof(struct cdev)");
114 SYSCTL_INT(_debug_sizeof, OID_AUTO, cdev_priv, CTLFLAG_RD,
115 SYSCTL_NULL_INT_PTR, sizeof(struct cdev_priv), "sizeof(struct cdev_priv)");
118 devfs_alloc(int flags)
120 struct cdev_priv *cdp;
124 cdp = malloc(sizeof *cdp, M_CDEVP, M_ZERO |
125 ((flags & MAKEDEV_NOWAIT) ? M_NOWAIT : M_WAITOK));
129 cdp->cdp_dirents = &cdp->cdp_dirent0;
130 cdp->cdp_dirent0 = NULL;
131 cdp->cdp_maxdirent = 0;
136 LIST_INIT(&cdev->si_children);
138 cdev->si_atime = cdev->si_mtime = cdev->si_ctime = ts;
139 cdev->si_cred = NULL;
145 devfs_dev_exists(const char *name)
147 struct cdev_priv *cdp;
149 mtx_assert(&devmtx, MA_OWNED);
151 TAILQ_FOREACH(cdp, &cdevp_list, cdp_list) {
152 if ((cdp->cdp_flags & CDP_ACTIVE) == 0)
154 if (devfs_pathpath(cdp->cdp_c.si_name, name) != 0)
156 if (devfs_pathpath(name, cdp->cdp_c.si_name) != 0)
159 if (devfs_dir_find(name) != 0)
166 devfs_free(struct cdev *cdev)
168 struct cdev_priv *cdp;
170 cdp = cdev2priv(cdev);
171 if (cdev->si_cred != NULL)
172 crfree(cdev->si_cred);
173 devfs_free_cdp_inode(cdp->cdp_inode);
174 if (cdp->cdp_maxdirent > 0)
175 free(cdp->cdp_dirents, M_DEVFS2);
179 struct devfs_dirent *
180 devfs_find(struct devfs_dirent *dd, const char *name, int namelen, int type)
182 struct devfs_dirent *de;
184 TAILQ_FOREACH(de, &dd->de_dlist, de_list) {
185 if (namelen != de->de_dirent->d_namlen)
187 if (type != 0 && type != de->de_dirent->d_type)
191 * The race with finding non-active name is not
192 * completely closed by the check, but it is similar
193 * to the devfs_allocv() in making it unlikely enough.
195 if (de->de_dirent->d_type == DT_CHR &&
196 (de->de_cdp->cdp_flags & CDP_ACTIVE) == 0)
199 if (bcmp(name, de->de_dirent->d_name, namelen) != 0)
203 KASSERT(de == NULL || (de->de_flags & DE_DOOMED) == 0,
204 ("devfs_find: returning a doomed entry"));
208 struct devfs_dirent *
209 devfs_newdirent(char *name, int namelen)
212 struct devfs_dirent *de;
215 d.d_namlen = namelen;
216 i = sizeof (*de) + GENERIC_DIRSIZ(&d);
217 de = malloc(i, M_DEVFS3, M_WAITOK | M_ZERO);
218 de->de_dirent = (struct dirent *)(de + 1);
219 de->de_dirent->d_namlen = namelen;
220 de->de_dirent->d_reclen = GENERIC_DIRSIZ(&d);
221 bcopy(name, de->de_dirent->d_name, namelen);
222 de->de_dirent->d_name[namelen] = '\0';
223 vfs_timestamp(&de->de_ctime);
224 de->de_mtime = de->de_atime = de->de_ctime;
233 struct devfs_dirent *
234 devfs_parent_dirent(struct devfs_dirent *de)
237 if (de->de_dirent->d_type != DT_DIR)
240 if (de->de_flags & (DE_DOT | DE_DOTDOT))
243 de = TAILQ_FIRST(&de->de_dlist); /* "." */
246 de = TAILQ_NEXT(de, de_list); /* ".." */
253 struct devfs_dirent *
254 devfs_vmkdir(struct devfs_mount *dmp, char *name, int namelen, struct devfs_dirent *dotdot, u_int inode)
256 struct devfs_dirent *dd;
257 struct devfs_dirent *de;
259 /* Create the new directory */
260 dd = devfs_newdirent(name, namelen);
261 TAILQ_INIT(&dd->de_dlist);
262 dd->de_dirent->d_type = DT_DIR;
267 dd->de_inode = inode;
269 dd->de_inode = alloc_unr(devfs_inos);
272 * "." and ".." are always the two first entries in the
275 * Create the "." entry in the new directory.
277 de = devfs_newdirent(".", 1);
278 de->de_dirent->d_type = DT_DIR;
279 de->de_flags |= DE_DOT;
280 TAILQ_INSERT_TAIL(&dd->de_dlist, de, de_list);
283 /* Create the ".." entry in the new directory. */
284 de = devfs_newdirent("..", 2);
285 de->de_dirent->d_type = DT_DIR;
286 de->de_flags |= DE_DOTDOT;
287 TAILQ_INSERT_TAIL(&dd->de_dlist, de, de_list);
288 if (dotdot == NULL) {
292 sx_assert(&dmp->dm_lock, SX_XLOCKED);
293 TAILQ_INSERT_TAIL(&dotdot->de_dlist, dd, de_list);
295 devfs_rules_apply(dmp, dd);
299 mac_devfs_create_directory(dmp->dm_mount, name, namelen, dd);
305 devfs_dirent_free(struct devfs_dirent *de)
311 * Removes a directory if it is empty. Also empty parent directories are
312 * removed recursively.
315 devfs_rmdir_empty(struct devfs_mount *dm, struct devfs_dirent *de)
317 struct devfs_dirent *dd, *de_dot, *de_dotdot;
319 sx_assert(&dm->dm_lock, SX_XLOCKED);
322 KASSERT(de->de_dirent->d_type == DT_DIR,
323 ("devfs_rmdir_empty: de is not a directory"));
325 if ((de->de_flags & DE_DOOMED) != 0 || de == dm->dm_rootdir)
328 de_dot = TAILQ_FIRST(&de->de_dlist);
329 KASSERT(de_dot != NULL, ("devfs_rmdir_empty: . missing"));
330 de_dotdot = TAILQ_NEXT(de_dot, de_list);
331 KASSERT(de_dotdot != NULL, ("devfs_rmdir_empty: .. missing"));
332 /* Return if the directory is not empty. */
333 if (TAILQ_NEXT(de_dotdot, de_list) != NULL)
336 dd = devfs_parent_dirent(de);
337 KASSERT(dd != NULL, ("devfs_rmdir_empty: NULL dd"));
338 TAILQ_REMOVE(&de->de_dlist, de_dot, de_list);
339 TAILQ_REMOVE(&de->de_dlist, de_dotdot, de_list);
340 TAILQ_REMOVE(&dd->de_dlist, de, de_list);
342 devfs_delete(dm, de, DEVFS_DEL_NORECURSE);
343 devfs_delete(dm, de_dot, DEVFS_DEL_NORECURSE);
344 devfs_delete(dm, de_dotdot, DEVFS_DEL_NORECURSE);
345 if (DEVFS_DE_DROP(dd)) {
346 devfs_dirent_free(dd);
355 * The caller needs to hold the dm for the duration of the call since
356 * dm->dm_lock may be temporary dropped.
359 devfs_delete(struct devfs_mount *dm, struct devfs_dirent *de, int flags)
361 struct devfs_dirent *dd;
364 KASSERT((de->de_flags & DE_DOOMED) == 0,
365 ("devfs_delete doomed dirent"));
366 de->de_flags |= DE_DOOMED;
368 if ((flags & DEVFS_DEL_NORECURSE) == 0) {
369 dd = devfs_parent_dirent(de);
372 if (de->de_flags & DE_USER) {
373 KASSERT(dd != NULL, ("devfs_delete: NULL dd"));
374 devfs_dir_unref_de(dm, dd);
379 mtx_lock(&devfs_de_interlock);
383 mtx_unlock(&devfs_de_interlock);
385 sx_unlock(&dm->dm_lock);
386 if ((flags & DEVFS_DEL_VNLOCKED) == 0)
387 vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY);
391 if ((flags & DEVFS_DEL_VNLOCKED) == 0)
394 sx_xlock(&dm->dm_lock);
396 mtx_unlock(&devfs_de_interlock);
397 if (de->de_symlink) {
398 free(de->de_symlink, M_DEVFS);
399 de->de_symlink = NULL;
402 mac_devfs_destroy(de);
404 if (de->de_inode > DEVFS_ROOTINO) {
405 devfs_free_cdp_inode(de->de_inode);
408 if (DEVFS_DE_DROP(de))
409 devfs_dirent_free(de);
412 if (DEVFS_DE_DROP(dd))
413 devfs_dirent_free(dd);
415 devfs_rmdir_empty(dm, dd);
421 * Recursively removes the entire tree.
422 * The caller needs to hold the dm for the duration of the call.
426 devfs_purge(struct devfs_mount *dm, struct devfs_dirent *dd)
428 struct devfs_dirent *de;
430 sx_assert(&dm->dm_lock, SX_XLOCKED);
435 * Use TAILQ_LAST() to remove "." and ".." last.
436 * We might need ".." to resolve a path in
437 * devfs_dir_unref_de().
439 de = TAILQ_LAST(&dd->de_dlist, devfs_dlist_head);
442 TAILQ_REMOVE(&dd->de_dlist, de, de_list);
443 if (de->de_flags & DE_USER)
444 devfs_dir_unref_de(dm, dd);
445 if (de->de_flags & (DE_DOT | DE_DOTDOT))
446 devfs_delete(dm, de, DEVFS_DEL_NORECURSE);
447 else if (de->de_dirent->d_type == DT_DIR)
450 devfs_delete(dm, de, DEVFS_DEL_NORECURSE);
452 if (DEVFS_DE_DROP(dd))
453 devfs_dirent_free(dd);
454 else if ((dd->de_flags & DE_DOOMED) == 0)
455 devfs_delete(dm, dd, DEVFS_DEL_NORECURSE);
459 * Each cdev_priv has an array of pointers to devfs_dirent which is indexed
460 * by the mount points dm_idx.
461 * This function extends the array when necessary, taking into account that
462 * the default array is 1 element and not malloc'ed.
465 devfs_metoo(struct cdev_priv *cdp, struct devfs_mount *dm)
467 struct devfs_dirent **dep;
470 siz = (dm->dm_idx + 1) * sizeof *dep;
471 dep = malloc(siz, M_DEVFS2, M_WAITOK | M_ZERO);
473 if (dm->dm_idx <= cdp->cdp_maxdirent) {
479 memcpy(dep, cdp->cdp_dirents, (cdp->cdp_maxdirent + 1) * sizeof *dep);
480 if (cdp->cdp_maxdirent > 0)
481 free(cdp->cdp_dirents, M_DEVFS2);
482 cdp->cdp_dirents = dep;
484 * XXX: if malloc told us how much we actually got this could
487 cdp->cdp_maxdirent = dm->dm_idx;
492 * The caller needs to hold the dm for the duration of the call.
495 devfs_populate_loop(struct devfs_mount *dm, int cleanup)
497 struct cdev_priv *cdp;
498 struct devfs_dirent *de;
499 struct devfs_dirent *dd;
504 sx_assert(&dm->dm_lock, SX_XLOCKED);
506 TAILQ_FOREACH(cdp, &cdevp_list, cdp_list) {
508 KASSERT(cdp->cdp_dirents != NULL, ("NULL cdp_dirents"));
511 * If we are unmounting, or the device has been destroyed,
512 * clean up our dirent.
514 if ((cleanup || !(cdp->cdp_flags & CDP_ACTIVE)) &&
515 dm->dm_idx <= cdp->cdp_maxdirent &&
516 cdp->cdp_dirents[dm->dm_idx] != NULL) {
517 de = cdp->cdp_dirents[dm->dm_idx];
518 cdp->cdp_dirents[dm->dm_idx] = NULL;
519 KASSERT(cdp == de->de_cdp,
520 ("%s %d %s %p %p", __func__, __LINE__,
521 cdp->cdp_c.si_name, cdp, de->de_cdp));
522 KASSERT(de->de_dir != NULL, ("Null de->de_dir"));
525 TAILQ_REMOVE(&de->de_dir->de_dlist, de, de_list);
528 devfs_delete(dm, de, 0);
535 * GC any lingering devices
537 if (!(cdp->cdp_flags & CDP_ACTIVE)) {
538 if (cdp->cdp_inuse > 0)
540 TAILQ_REMOVE(&cdevp_list, cdp, cdp_list);
542 dev_rel(&cdp->cdp_c);
546 * Don't create any new dirents if we are unmounting
550 KASSERT((cdp->cdp_flags & CDP_ACTIVE), ("Bogons, I tell ya'!"));
552 if (dm->dm_idx <= cdp->cdp_maxdirent &&
553 cdp->cdp_dirents[dm->dm_idx] != NULL) {
554 de = cdp->cdp_dirents[dm->dm_idx];
555 KASSERT(cdp == de->de_cdp, ("inconsistent cdp"));
563 if (dm->dm_idx > cdp->cdp_maxdirent)
564 devfs_metoo(cdp, dm);
567 s = cdp->cdp_c.si_name;
569 for (q = s; *q != '/' && *q != '\0'; q++)
573 de = devfs_find(dd, s, q - s, 0);
575 de = devfs_vmkdir(dm, s, q - s, dd, 0);
576 else if (de->de_dirent->d_type == DT_LNK) {
577 de = devfs_find(dd, s, q - s, DT_DIR);
579 de = devfs_vmkdir(dm, s, q - s, dd, 0);
580 de->de_flags |= DE_COVERED;
584 KASSERT(dd->de_dirent->d_type == DT_DIR &&
585 (dd->de_flags & (DE_DOT | DE_DOTDOT)) == 0,
586 ("%s: invalid directory (si_name=%s)",
587 __func__, cdp->cdp_c.si_name));
591 de = devfs_find(dd, s, q - s, DT_LNK);
593 de_flags |= DE_COVERED;
595 de = devfs_newdirent(s, q - s);
596 if (cdp->cdp_c.si_flags & SI_ALIAS) {
600 de->de_dirent->d_type = DT_LNK;
601 pdev = cdp->cdp_c.si_parent;
602 j = strlen(pdev->si_name) + 1;
603 de->de_symlink = malloc(j, M_DEVFS, M_WAITOK);
604 bcopy(pdev->si_name, de->de_symlink, j);
606 de->de_uid = cdp->cdp_c.si_uid;
607 de->de_gid = cdp->cdp_c.si_gid;
608 de->de_mode = cdp->cdp_c.si_mode;
609 de->de_dirent->d_type = DT_CHR;
611 de->de_flags |= de_flags;
612 de->de_inode = cdp->cdp_inode;
615 mac_devfs_create_device(cdp->cdp_c.si_cred, dm->dm_mount,
619 TAILQ_INSERT_TAIL(&dd->de_dlist, de, de_list);
620 devfs_rules_apply(dm, de);
622 /* XXX: could check that cdp is still active here */
623 KASSERT(cdp->cdp_dirents[dm->dm_idx] == NULL,
624 ("%s %d\n", __func__, __LINE__));
625 cdp->cdp_dirents[dm->dm_idx] = de;
626 KASSERT(de->de_cdp != (void *)0xdeadc0de,
627 ("%s %d\n", __func__, __LINE__));
636 * The caller needs to hold the dm for the duration of the call.
639 devfs_populate(struct devfs_mount *dm)
643 sx_assert(&dm->dm_lock, SX_XLOCKED);
644 gen = devfs_generation;
645 if (dm->dm_generation == gen)
647 while (devfs_populate_loop(dm, 0))
649 dm->dm_generation = gen;
653 * The caller needs to hold the dm for the duration of the call.
656 devfs_cleanup(struct devfs_mount *dm)
659 sx_assert(&dm->dm_lock, SX_XLOCKED);
660 while (devfs_populate_loop(dm, 1))
662 devfs_purge(dm, dm->dm_rootdir);
666 * devfs_create() and devfs_destroy() are called from kern_conf.c and
667 * in both cases the devlock() mutex is held, so no further locking
668 * is necessary and no sleeping allowed.
672 devfs_create(struct cdev *dev)
674 struct cdev_priv *cdp;
676 mtx_assert(&devmtx, MA_OWNED);
677 cdp = cdev2priv(dev);
678 cdp->cdp_flags |= CDP_ACTIVE;
679 cdp->cdp_inode = alloc_unrl(devfs_inos);
681 TAILQ_INSERT_TAIL(&cdevp_list, cdp, cdp_list);
686 devfs_destroy(struct cdev *dev)
688 struct cdev_priv *cdp;
690 mtx_assert(&devmtx, MA_OWNED);
691 cdp = cdev2priv(dev);
692 cdp->cdp_flags &= ~CDP_ACTIVE;
697 devfs_alloc_cdp_inode(void)
700 return (alloc_unr(devfs_inos));
704 devfs_free_cdp_inode(ino_t ino)
708 free_unr(devfs_inos, ino);
712 devfs_devs_init(void *junk __unused)
715 devfs_inos = new_unrhdr(DEVFS_ROOTINO + 1, INT_MAX, &devmtx);
718 SYSINIT(devfs_devs, SI_SUB_DEVFS, SI_ORDER_FIRST, devfs_devs_init, NULL);