xref: /freebsd/sys/kern/vfs_init.c (revision 5e53a4f90f82c4345f277dd87cc9292f26e04a29)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed
8  * to Berkeley by John Heidemann of the UCLA Ficus project.
9  *
10  * Source: * @(#)i405_init.c 2.10 92/04/27 UCLA Ficus project
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)vfs_init.c	8.3 (Berkeley) 1/4/94
37  */
38 
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41 
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/fnv_hash.h>
45 #include <sys/kernel.h>
46 #include <sys/linker.h>
47 #include <sys/mount.h>
48 #include <sys/proc.h>
49 #include <sys/sx.h>
50 #include <sys/syscallsubr.h>
51 #include <sys/sysctl.h>
52 #include <sys/vnode.h>
53 #include <sys/malloc.h>
54 
55 static int	vfs_register(struct vfsconf *);
56 static int	vfs_unregister(struct vfsconf *);
57 
58 MALLOC_DEFINE(M_VNODE, "vnodes", "Dynamically allocated vnodes");
59 
60 /*
61  * The highest defined VFS number.
62  */
63 int maxvfsconf = VFS_GENERIC + 1;
64 
65 /*
66  * Single-linked list of configured VFSes.
67  * New entries are added/deleted by vfs_register()/vfs_unregister()
68  */
69 struct vfsconfhead vfsconf = TAILQ_HEAD_INITIALIZER(vfsconf);
70 struct sx vfsconf_sx;
71 SX_SYSINIT(vfsconf, &vfsconf_sx, "vfsconf");
72 
73 /*
74  * Loader.conf variable vfs.typenumhash enables setting vfc_typenum using a hash
75  * calculation on vfc_name, so that it doesn't change when file systems are
76  * loaded in a different order. This will avoid the NFS server file handles from
77  * changing for file systems that use vfc_typenum in their fsid.
78  */
79 static int	vfs_typenumhash = 1;
80 SYSCTL_INT(_vfs, OID_AUTO, typenumhash, CTLFLAG_RDTUN, &vfs_typenumhash, 0,
81     "Set vfc_typenum using a hash calculation on vfc_name, so that it does not"
82     "change when file systems are loaded in a different order.");
83 
84 /*
85  * A Zen vnode attribute structure.
86  *
87  * Initialized when the first filesystem registers by vfs_register().
88  */
89 struct vattr va_null;
90 
91 /*
92  * vfs_init.c
93  *
94  * Allocate and fill in operations vectors.
95  *
96  * An undocumented feature of this approach to defining operations is that
97  * there can be multiple entries in vfs_opv_descs for the same operations
98  * vector. This allows third parties to extend the set of operations
99  * supported by another layer in a binary compatibile way. For example,
100  * assume that NFS needed to be modified to support Ficus. NFS has an entry
101  * (probably nfs_vnopdeop_decls) declaring all the operations NFS supports by
102  * default. Ficus could add another entry (ficus_nfs_vnodeop_decl_entensions)
103  * listing those new operations Ficus adds to NFS, all without modifying the
104  * NFS code. (Of couse, the OTW NFS protocol still needs to be munged, but
105  * that is a(whole)nother story.) This is a feature.
106  */
107 
108 /*
109  * Routines having to do with the management of the vnode table.
110  */
111 
112 static struct vfsconf *
113 vfs_byname_locked(const char *name)
114 {
115 	struct vfsconf *vfsp;
116 
117 	sx_assert(&vfsconf_sx, SA_LOCKED);
118 	if (!strcmp(name, "ffs"))
119 		name = "ufs";
120 	TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
121 		if (!strcmp(name, vfsp->vfc_name))
122 			return (vfsp);
123 	}
124 	return (NULL);
125 }
126 
127 struct vfsconf *
128 vfs_byname(const char *name)
129 {
130 	struct vfsconf *vfsp;
131 
132 	vfsconf_slock();
133 	vfsp = vfs_byname_locked(name);
134 	vfsconf_sunlock();
135 	return (vfsp);
136 }
137 
138 struct vfsconf *
139 vfs_byname_kld(const char *fstype, struct thread *td, int *error)
140 {
141 	struct vfsconf *vfsp;
142 	int fileid, loaded;
143 
144 	vfsp = vfs_byname(fstype);
145 	if (vfsp != NULL)
146 		return (vfsp);
147 
148 	/* Try to load the respective module. */
149 	*error = kern_kldload(td, fstype, &fileid);
150 	loaded = (*error == 0);
151 	if (*error == EEXIST)
152 		*error = 0;
153 	if (*error)
154 		return (NULL);
155 
156 	/* Look up again to see if the VFS was loaded. */
157 	vfsp = vfs_byname(fstype);
158 	if (vfsp == NULL) {
159 		if (loaded)
160 			(void)kern_kldunload(td, fileid, LINKER_UNLOAD_FORCE);
161 		*error = ENODEV;
162 		return (NULL);
163 	}
164 	return (vfsp);
165 }
166 
167 
168 /* Register a new filesystem type in the global table */
169 static int
170 vfs_register(struct vfsconf *vfc)
171 {
172 	struct sysctl_oid *oidp;
173 	struct vfsops *vfsops;
174 	static int once;
175 	struct vfsconf *tvfc;
176 	uint32_t hashval;
177 	int secondpass;
178 
179 	if (!once) {
180 		vattr_null(&va_null);
181 		once = 1;
182 	}
183 
184 	if (vfc->vfc_version != VFS_VERSION) {
185 		printf("ERROR: filesystem %s, unsupported ABI version %x\n",
186 		    vfc->vfc_name, vfc->vfc_version);
187 		return (EINVAL);
188 	}
189 	vfsconf_lock();
190 	if (vfs_byname_locked(vfc->vfc_name) != NULL) {
191 		vfsconf_unlock();
192 		return (EEXIST);
193 	}
194 
195 	if (vfs_typenumhash != 0) {
196 		/*
197 		 * Calculate a hash on vfc_name to use for vfc_typenum. Unless
198 		 * all of 1<->255 are assigned, it is limited to 8bits since
199 		 * that is what ZFS uses from vfc_typenum and is also the
200 		 * preferred range for vfs_getnewfsid().
201 		 */
202 		hashval = fnv_32_str(vfc->vfc_name, FNV1_32_INIT);
203 		hashval &= 0xff;
204 		secondpass = 0;
205 		do {
206 			/* Look for and fix any collision. */
207 			TAILQ_FOREACH(tvfc, &vfsconf, vfc_list) {
208 				if (hashval == tvfc->vfc_typenum) {
209 					if (hashval == 255 && secondpass == 0) {
210 						hashval = 1;
211 						secondpass = 1;
212 					} else
213 						hashval++;
214 					break;
215 				}
216 			}
217 		} while (tvfc != NULL);
218 		vfc->vfc_typenum = hashval;
219 		if (vfc->vfc_typenum >= maxvfsconf)
220 			maxvfsconf = vfc->vfc_typenum + 1;
221 	} else
222 		vfc->vfc_typenum = maxvfsconf++;
223 	TAILQ_INSERT_TAIL(&vfsconf, vfc, vfc_list);
224 
225 	/*
226 	 * Initialise unused ``struct vfsops'' fields, to use
227 	 * the vfs_std*() functions.  Note, we need the mount
228 	 * and unmount operations, at the least.  The check
229 	 * for vfsops available is just a debugging aid.
230 	 */
231 	KASSERT(vfc->vfc_vfsops != NULL,
232 	    ("Filesystem %s has no vfsops", vfc->vfc_name));
233 	/*
234 	 * Check the mount and unmount operations.
235 	 */
236 	vfsops = vfc->vfc_vfsops;
237 	KASSERT(vfsops->vfs_mount != NULL,
238 	    ("Filesystem %s has no mount op", vfc->vfc_name));
239 	KASSERT(vfsops->vfs_unmount != NULL,
240 	    ("Filesystem %s has no unmount op", vfc->vfc_name));
241 
242 	if (vfsops->vfs_root == NULL)
243 		/* return file system's root vnode */
244 		vfsops->vfs_root =	vfs_stdroot;
245 	if (vfsops->vfs_quotactl == NULL)
246 		/* quota control */
247 		vfsops->vfs_quotactl =	vfs_stdquotactl;
248 	if (vfsops->vfs_statfs == NULL)
249 		/* return file system's status */
250 		vfsops->vfs_statfs =	vfs_stdstatfs;
251 	if (vfsops->vfs_sync == NULL)
252 		/*
253 		 * flush unwritten data (nosync)
254 		 * file systems can use vfs_stdsync
255 		 * explicitly by setting it in the
256 		 * vfsop vector.
257 		 */
258 		vfsops->vfs_sync =	vfs_stdnosync;
259 	if (vfsops->vfs_vget == NULL)
260 		/* convert an inode number to a vnode */
261 		vfsops->vfs_vget =	vfs_stdvget;
262 	if (vfsops->vfs_fhtovp == NULL)
263 		/* turn an NFS file handle into a vnode */
264 		vfsops->vfs_fhtovp =	vfs_stdfhtovp;
265 	if (vfsops->vfs_checkexp == NULL)
266 		/* check if file system is exported */
267 		vfsops->vfs_checkexp =	vfs_stdcheckexp;
268 	if (vfsops->vfs_init == NULL)
269 		/* file system specific initialisation */
270 		vfsops->vfs_init =	vfs_stdinit;
271 	if (vfsops->vfs_uninit == NULL)
272 		/* file system specific uninitialisation */
273 		vfsops->vfs_uninit =	vfs_stduninit;
274 	if (vfsops->vfs_extattrctl == NULL)
275 		/* extended attribute control */
276 		vfsops->vfs_extattrctl = vfs_stdextattrctl;
277 	if (vfsops->vfs_sysctl == NULL)
278 		vfsops->vfs_sysctl = vfs_stdsysctl;
279 
280 	/*
281 	 * Call init function for this VFS...
282 	 */
283 	(*(vfc->vfc_vfsops->vfs_init))(vfc);
284 	vfsconf_unlock();
285 
286 	/*
287 	 * If this filesystem has a sysctl node under vfs
288 	 * (i.e. vfs.xxfs), then change the oid number of that node to
289 	 * match the filesystem's type number.  This allows user code
290 	 * which uses the type number to read sysctl variables defined
291 	 * by the filesystem to continue working. Since the oids are
292 	 * in a sorted list, we need to make sure the order is
293 	 * preserved by re-registering the oid after modifying its
294 	 * number.
295 	 */
296 	sysctl_wlock();
297 	SLIST_FOREACH(oidp, SYSCTL_CHILDREN(&sysctl___vfs), oid_link) {
298 		if (strcmp(oidp->oid_name, vfc->vfc_name) == 0) {
299 			sysctl_unregister_oid(oidp);
300 			oidp->oid_number = vfc->vfc_typenum;
301 			sysctl_register_oid(oidp);
302 			break;
303 		}
304 	}
305 	sysctl_wunlock();
306 
307 	return (0);
308 }
309 
310 
311 /* Remove registration of a filesystem type */
312 static int
313 vfs_unregister(struct vfsconf *vfc)
314 {
315 	struct vfsconf *vfsp;
316 	int error, maxtypenum;
317 
318 	vfsconf_lock();
319 	vfsp = vfs_byname_locked(vfc->vfc_name);
320 	if (vfsp == NULL) {
321 		vfsconf_unlock();
322 		return (EINVAL);
323 	}
324 	if (vfsp->vfc_refcount != 0) {
325 		vfsconf_unlock();
326 		return (EBUSY);
327 	}
328 	if (vfc->vfc_vfsops->vfs_uninit != NULL) {
329 		error = (*vfc->vfc_vfsops->vfs_uninit)(vfsp);
330 		if (error != 0) {
331 			vfsconf_unlock();
332 			return (error);
333 		}
334 	}
335 	TAILQ_REMOVE(&vfsconf, vfsp, vfc_list);
336 	maxtypenum = VFS_GENERIC;
337 	TAILQ_FOREACH(vfsp, &vfsconf, vfc_list)
338 		if (maxtypenum < vfsp->vfc_typenum)
339 			maxtypenum = vfsp->vfc_typenum;
340 	maxvfsconf = maxtypenum + 1;
341 	vfsconf_unlock();
342 	return (0);
343 }
344 
345 /*
346  * Standard kernel module handling code for filesystem modules.
347  * Referenced from VFS_SET().
348  */
349 int
350 vfs_modevent(module_t mod, int type, void *data)
351 {
352 	struct vfsconf *vfc;
353 	int error = 0;
354 
355 	vfc = (struct vfsconf *)data;
356 
357 	switch (type) {
358 	case MOD_LOAD:
359 		if (vfc)
360 			error = vfs_register(vfc);
361 		break;
362 
363 	case MOD_UNLOAD:
364 		if (vfc)
365 			error = vfs_unregister(vfc);
366 		break;
367 	default:
368 		error = EOPNOTSUPP;
369 		break;
370 	}
371 	return (error);
372 }
373