xref: /netbsd-src/sys/kern/vfs_init.c (revision e94a5d02693120d4ad9d909e488894e9fcf0eb76)
1 /*	$NetBSD: vfs_init.c,v 1.67 2024/12/07 02:27:38 riastradh Exp $	*/
2 
3 /*-
4  * Copyright (c) 1998, 2000, 2008 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility,
9  * NASA Ames Research Center.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
22  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
23  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
24  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30  * POSSIBILITY OF SUCH DAMAGE.
31  */
32 
33 /*
34  * Copyright (c) 1989, 1993
35  *	The Regents of the University of California.  All rights reserved.
36  *
37  * This code is derived from software contributed
38  * to Berkeley by John Heidemann of the UCLA Ficus project.
39  *
40  * Source: * @(#)i405_init.c 2.10 92/04/27 UCLA Ficus project
41  *
42  * Redistribution and use in source and binary forms, with or without
43  * modification, are permitted provided that the following conditions
44  * are met:
45  * 1. Redistributions of source code must retain the above copyright
46  *    notice, this list of conditions and the following disclaimer.
47  * 2. Redistributions in binary form must reproduce the above copyright
48  *    notice, this list of conditions and the following disclaimer in the
49  *    documentation and/or other materials provided with the distribution.
50  * 3. Neither the name of the University nor the names of its contributors
51  *    may be used to endorse or promote products derived from this software
52  *    without specific prior written permission.
53  *
54  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
64  * SUCH DAMAGE.
65  *
66  *	@(#)vfs_init.c	8.5 (Berkeley) 5/11/95
67  */
68 
69 #include <sys/cdefs.h>
70 __KERNEL_RCSID(0, "$NetBSD: vfs_init.c,v 1.67 2024/12/07 02:27:38 riastradh Exp $");
71 
72 #include <sys/param.h>
73 #include <sys/types.h>
74 
75 #include <sys/buf.h>
76 #include <sys/dirhash.h>
77 #include <sys/errno.h>
78 #include <sys/kauth.h>
79 #include <sys/kmem.h>
80 #include <sys/module.h>
81 #include <sys/mount.h>
82 #include <sys/namei.h>
83 #include <sys/sdt.h>
84 #include <sys/stat.h>
85 #include <sys/sysctl.h>
86 #include <sys/systm.h>
87 #include <sys/time.h>
88 #include <sys/ucred.h>
89 #include <sys/vnode.h>
90 #include <sys/vnode_impl.h>
91 
92 #include <miscfs/deadfs/deadfs.h>
93 #include <miscfs/fifofs/fifo.h>
94 #include <miscfs/specfs/specdev.h>
95 
96 /*
97  * Sigh, such primitive tools are these...
98  */
99 #if 0
100 #define DODEBUG(A) A
101 #else
102 #define DODEBUG(A) __nothing
103 #endif
104 
105 SDT_PROVIDER_DEFINE(vfs);
106 
107 /*
108  * These vnodeopv_descs are listed here because they are not
109  * associated with any particular file system, and thus cannot
110  * be initialized by vfs_attach().
111  */
112 const struct vnodeopv_desc * const vfs_special_vnodeopv_descs[] = {
113 	&dead_vnodeop_opv_desc,
114 	&fifo_vnodeop_opv_desc,
115 	&spec_vnodeop_opv_desc,
116 	NULL,
117 };
118 
119 struct vfs_list_head vfs_list =			/* vfs list */
120     LIST_HEAD_INITIALIZER(vfs_list);
121 
122 static kauth_listener_t mount_listener;
123 
124 /*
125  * This code doesn't work if the defn is **vnodop_defns with cc.
126  * The problem is because of the compiler sometimes putting in an
127  * extra level of indirection for arrays.  It's an interesting
128  * "feature" of C.
129  */
130 typedef int (*PFI)(void *);
131 
132 /*
133  * A miscellaneous routine.
134  * A generic "default" routine that just returns an error.
135  */
136 /*ARGSUSED*/
137 int
138 vn_default_error(void *v)
139 {
140 
141 	return SET_ERROR(EOPNOTSUPP);
142 }
143 
144 static struct sysctllog *vfs_sysctllog;
145 
146 /*
147  * Top level filesystem related information gathering.
148  */
149 static void
150 sysctl_vfs_setup(void)
151 {
152 
153 	sysctl_createv(&vfs_sysctllog, 0, NULL, NULL,
154 	    CTLFLAG_PERMANENT,
155 	    CTLTYPE_NODE, "generic",
156 	    SYSCTL_DESCR("Non-specific vfs related information"),
157 	    NULL, 0, NULL, 0,
158 	    CTL_VFS, VFS_GENERIC, CTL_EOL);
159 	sysctl_createv(&vfs_sysctllog, 0, NULL, NULL,
160 	    CTLFLAG_PERMANENT,
161 	    CTLTYPE_STRING, "fstypes",
162 	    SYSCTL_DESCR("List of file systems present"),
163 	    sysctl_vfs_generic_fstypes, 0, NULL, 0,
164 	    CTL_VFS, VFS_GENERIC, CTL_CREATE, CTL_EOL);
165 	sysctl_createv(&vfs_sysctllog, 0, NULL, NULL,
166 	    CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
167 	    CTLTYPE_INT, "magiclinks",
168 	    SYSCTL_DESCR("Whether \"magic\" symlinks are expanded"),
169 	    NULL, 0, &vfs_magiclinks, 0,
170 	    CTL_VFS, VFS_GENERIC, VFS_MAGICLINKS, CTL_EOL);
171 	sysctl_createv(&vfs_sysctllog, 0, NULL, NULL,
172 	    CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
173 	    CTLTYPE_INT, "timestamp_precision",
174 	    SYSCTL_DESCR("File timestamp precision"),
175 	    NULL, 0, &vfs_timestamp_precision, 0,
176 	    CTL_VFS, VFS_GENERIC, VFS_TIMESTAMP_PRECISION,
177 	    CTL_EOL);
178 }
179 
180 /*
181  * vfs_init.c
182  *
183  * Allocate and fill in operations vectors.
184  *
185  * An undocumented feature of this approach to defining operations is that
186  * there can be multiple entries in vfs_opv_descs for the same operations
187  * vector. This allows third parties to extend the set of operations
188  * supported by another layer in a binary compatibile way. For example,
189  * assume that NFS needed to be modified to support Ficus. NFS has an entry
190  * (probably nfs_vnopdeop_decls) declaring all the operations NFS supports by
191  * default. Ficus could add another entry (ficus_nfs_vnodeop_decl_entensions)
192  * listing those new operations Ficus adds to NFS, all without modifying the
193  * NFS code. (Of couse, the OTW NFS protocol still needs to be munged, but
194  * that is a(whole)nother story.) This is a feature.
195  */
196 
197 /*
198  * Init the vector, if it needs it.
199  * Also handle backwards compatibility.
200  */
201 static void
202 vfs_opv_init_explicit(const struct vnodeopv_desc *vfs_opv_desc)
203 {
204 	int (**opv_desc_vector)(void *);
205 	const struct vnodeopv_entry_desc *opve_descp;
206 
207 	opv_desc_vector = *(vfs_opv_desc->opv_desc_vector_p);
208 
209 	for (opve_descp = vfs_opv_desc->opv_desc_ops;
210 	     opve_descp->opve_op;
211 	     opve_descp++) {
212 		/*
213 		 * Sanity check:  is this operation listed
214 		 * in the list of operations?  We check this
215 		 * by seeing if its offset is zero.  Since
216 		 * the default routine should always be listed
217 		 * first, it should be the only one with a zero
218 		 * offset.  Any other operation with a zero
219 		 * offset is probably not listed in
220 		 * vfs_op_descs, and so is probably an error.
221 		 *
222 		 * A panic here means the layer programmer
223 		 * has committed the all-too common bug
224 		 * of adding a new operation to the layer's
225 		 * list of vnode operations but
226 		 * not adding the operation to the system-wide
227 		 * list of supported operations.
228 		 */
229 		if (opve_descp->opve_op->vdesc_offset == 0 &&
230 		    opve_descp->opve_op->vdesc_offset != VOFFSET(vop_default))
231 		{
232 			printf("operation %s not listed in %s.\n",
233 			    opve_descp->opve_op->vdesc_name, "vfs_op_descs");
234 			panic("vfs_opv_init: bad operation");
235 		}
236 
237 		/*
238 		 * Fill in this entry.
239 		 */
240 		opv_desc_vector[opve_descp->opve_op->vdesc_offset] =
241 		    opve_descp->opve_impl;
242 	}
243 }
244 
245 static void
246 vfs_opv_init_default(const struct vnodeopv_desc *vfs_opv_desc)
247 {
248 	int j;
249 	int (**opv_desc_vector)(void *);
250 
251 	opv_desc_vector = *(vfs_opv_desc->opv_desc_vector_p);
252 
253 	/*
254 	 * Force every operations vector to have a default routine.
255 	 */
256 	if (opv_desc_vector[VOFFSET(vop_default)] == NULL)
257 		panic("vfs_opv_init: operation vector without vop_default");
258 
259 	for (j = 0; j < VNODE_OPS_COUNT; j++) {
260 		if (opv_desc_vector[j] == NULL) {
261 			opv_desc_vector[j] =
262 			    opv_desc_vector[VOFFSET(vop_default)];
263 		}
264 	}
265 }
266 
267 void
268 vfs_opv_init(const struct vnodeopv_desc * const *vopvdpp)
269 {
270 	int (**opv_desc_vector)(void *);
271 	int i;
272 
273 	/*
274 	 * Allocate the vectors.
275 	 */
276 	for (i = 0; vopvdpp[i] != NULL; i++) {
277 		opv_desc_vector =
278 		    kmem_alloc(VNODE_OPS_COUNT * sizeof(PFI), KM_SLEEP);
279 		memset(opv_desc_vector, 0, VNODE_OPS_COUNT * sizeof(PFI));
280 		*(vopvdpp[i]->opv_desc_vector_p) = opv_desc_vector;
281 		DODEBUG(printf("vector at %p allocated\n",
282 		    opv_desc_vector_p));
283 	}
284 
285 	/*
286 	 * ...and fill them in.
287 	 */
288 	for (i = 0; vopvdpp[i] != NULL; i++)
289 		vfs_opv_init_explicit(vopvdpp[i]);
290 
291 	/*
292 	 * Finally, go back and replace unfilled routines
293 	 * with their default.
294 	 */
295 	for (i = 0; vopvdpp[i] != NULL; i++)
296 		vfs_opv_init_default(vopvdpp[i]);
297 }
298 
299 void
300 vfs_opv_free(const struct vnodeopv_desc * const *vopvdpp)
301 {
302 	int i;
303 
304 	/*
305 	 * Free the vectors allocated in vfs_opv_init().
306 	 */
307 	for (i = 0; vopvdpp[i] != NULL; i++) {
308 		kmem_free(*(vopvdpp[i]->opv_desc_vector_p),
309 		    VNODE_OPS_COUNT * sizeof(PFI));
310 		*(vopvdpp[i]->opv_desc_vector_p) = NULL;
311 	}
312 }
313 
314 #ifdef DEBUG
315 static void
316 vfs_op_check(void)
317 {
318 	int i;
319 
320 	DODEBUG(printf("Vnode_interface_init.\n"));
321 
322 	/*
323 	 * Check offset of each op.
324 	 */
325 	for (i = 0; vfs_op_descs[i]; i++) {
326 		if (vfs_op_descs[i]->vdesc_offset != i)
327 			panic("vfs_op_check: vfs_op_desc[] offset mismatch");
328 	}
329 
330 	if (i != VNODE_OPS_COUNT) {
331 		panic("vfs_op_check: vnode ops count mismatch (%d != %d)",
332 		    i, VNODE_OPS_COUNT);
333 	}
334 
335 	DODEBUG(printf ("vfs_opv_numops=%d\n", VNODE_OPS_COUNT));
336 }
337 #endif /* DEBUG */
338 
339 /*
340  * Common routine to check if an unprivileged mount is allowed.
341  *
342  * We export just this part (i.e., without the access control) so that if a
343  * secmodel wants to implement finer grained user mounts it can do so without
344  * copying too much code. More elaborate policies (i.e., specific users allowed
345  * to also create devices and/or introduce set-id binaries, or export
346  * file-systems) will require a different implementation.
347  *
348  * This routine is intended to be called from listener context, and as such
349  * does not take credentials as an argument.
350  */
351 int
352 usermount_common_policy(struct mount *mp, u_long flags)
353 {
354 
355 	/* No exporting if unprivileged. */
356 	if (flags & MNT_EXPORTED)
357 		return SET_ERROR(EPERM);
358 
359 	/* Must have 'nosuid' and 'nodev'. */
360 	if ((flags & MNT_NODEV) == 0 || (flags & MNT_NOSUID) == 0)
361 		return SET_ERROR(EPERM);
362 
363 	/* Retain 'noexec'. */
364 	if ((mp->mnt_flag & MNT_NOEXEC) && (flags & MNT_NOEXEC) == 0)
365 		return SET_ERROR(EPERM);
366 
367 	return 0;
368 }
369 
370 static int
371 mount_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie,
372     void *arg0, void *arg1, void *arg2, void *arg3)
373 {
374 	int result;
375 	enum kauth_system_req req;
376 
377 	result = KAUTH_RESULT_DEFER;
378 	req = (enum kauth_system_req)(uintptr_t)(uintptr_t)arg0;
379 
380 	if (action != KAUTH_SYSTEM_MOUNT)
381 		return result;
382 
383 	if (req == KAUTH_REQ_SYSTEM_MOUNT_GET)
384 		result = KAUTH_RESULT_ALLOW;
385 	else if (req == KAUTH_REQ_SYSTEM_MOUNT_DEVICE) {
386 		vnode_t *devvp = arg2;
387 		accmode_t accmode = (accmode_t)(unsigned long)arg3;
388 		int error;
389 
390 		error = VOP_ACCESS(devvp, accmode, cred);
391 		if (!error)
392 			result = KAUTH_RESULT_ALLOW;
393 	}
394 
395 	return result;
396 }
397 
398 /*
399  * Initialize the vnode structures and initialize each file system type.
400  */
401 void
402 vfsinit(void)
403 {
404 
405 	/*
406 	 * Attach sysctl nodes
407 	 */
408 	sysctl_vfs_setup();
409 
410 	/*
411 	 * Initialize the vnode table
412 	 */
413 	vntblinit();
414 
415 	/*
416 	 * Initialize the vnode name cache
417 	 */
418 	nchinit();
419 
420 #ifdef DEBUG
421 	/*
422 	 * Check the list of vnode operations.
423 	 */
424 	vfs_op_check();
425 #endif
426 
427 	/*
428 	 * Initialize the special vnode operations.
429 	 */
430 	vfs_opv_init(vfs_special_vnodeopv_descs);
431 
432 	/*
433 	 * Initialise generic dirhash.
434 	 */
435 	dirhash_init();
436 
437 	/*
438 	 * Initialise VFS hooks.
439 	 */
440 	vfs_hooks_init();
441 
442 	mount_listener = kauth_listen_scope(KAUTH_SCOPE_SYSTEM,
443 	    mount_listener_cb, NULL);
444 
445 	/*
446 	 * Establish each file system which was statically
447 	 * included in the kernel.
448 	 */
449 	module_init_class(MODULE_CLASS_VFS);
450 
451 	/*
452 	 * Initialize EVFILT_FS for kqueue.
453 	 */
454 	vfs_evfilt_fs_init();
455 }
456 
457 /*
458  * Drop a reference to a file system type.
459  */
460 void
461 vfs_delref(struct vfsops *vfs)
462 {
463 
464 	mutex_enter(&vfs_list_lock);
465 	vfs->vfs_refcount--;
466 	mutex_exit(&vfs_list_lock);
467 }
468 
469 /*
470  * Establish a file system and initialize it.
471  */
472 int
473 vfs_attach(struct vfsops *vfs)
474 {
475 	struct vfsops *v;
476 	int error = 0;
477 
478 	mutex_enter(&vfs_list_lock);
479 
480 	/*
481 	 * Make sure this file system doesn't already exist.
482 	 */
483 	LIST_FOREACH(v, &vfs_list, vfs_list) {
484 		if (strcmp(vfs->vfs_name, v->vfs_name) == 0) {
485 			error = SET_ERROR(EEXIST);
486 			goto out;
487 		}
488 	}
489 
490 	/*
491 	 * Initialize the vnode operations for this file system.
492 	 */
493 	vfs_opv_init(vfs->vfs_opv_descs);
494 
495 	/*
496 	 * Now initialize the file system itself.
497 	 */
498 	(*vfs->vfs_init)();
499 
500 	/*
501 	 * ...and link it into the kernel's list.
502 	 */
503 	LIST_INSERT_HEAD(&vfs_list, vfs, vfs_list);
504 
505 	/*
506 	 * Sanity: make sure the reference count is 0.
507 	 */
508 	vfs->vfs_refcount = 0;
509 out:
510 	mutex_exit(&vfs_list_lock);
511 	return error;
512 }
513 
514 /*
515  * Remove a file system from the kernel.
516  */
517 int
518 vfs_detach(struct vfsops *vfs)
519 {
520 	struct vfsops *v;
521 	int error = 0;
522 
523 	mutex_enter(&vfs_list_lock);
524 
525 	/*
526 	 * Make sure no one is using the filesystem.
527 	 */
528 	if (vfs->vfs_refcount != 0) {
529 		error = SET_ERROR(EBUSY);
530 		goto out;
531 	}
532 
533 	/*
534 	 * ...and remove it from the kernel's list.
535 	 */
536 	LIST_FOREACH(v, &vfs_list, vfs_list) {
537 		if (v == vfs) {
538 			LIST_REMOVE(v, vfs_list);
539 			break;
540 		}
541 	}
542 
543 	if (v == NULL) {
544 		error = SET_ERROR(ESRCH);
545 		goto out;
546 	}
547 
548 	/*
549 	 * Now run the file system-specific cleanups.
550 	 */
551 	(*vfs->vfs_done)();
552 
553 	/*
554 	 * Free the vnode operations vector.
555 	 */
556 	vfs_opv_free(vfs->vfs_opv_descs);
557 out:
558 	mutex_exit(&vfs_list_lock);
559 	return error;
560 }
561 
562 void
563 vfs_reinit(void)
564 {
565 	struct vfsops *vfs;
566 
567 	mutex_enter(&vfs_list_lock);
568 	LIST_FOREACH(vfs, &vfs_list, vfs_list) {
569 		if (vfs->vfs_reinit) {
570 			vfs->vfs_refcount++;
571 			mutex_exit(&vfs_list_lock);
572 			(*vfs->vfs_reinit)();
573 			mutex_enter(&vfs_list_lock);
574 			vfs->vfs_refcount--;
575 		}
576 	}
577 	mutex_exit(&vfs_list_lock);
578 }
579