xref: /freebsd/sys/kern/vfs_init.c (revision 95d45410b5100e07f6f98450bcd841a8945d4726)
1 /*-
2  * Copyright (c) 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from software contributed
6  * to Berkeley by John Heidemann of the UCLA Ficus project.
7  *
8  * Source: * @(#)i405_init.c 2.10 92/04/27 UCLA Ficus project
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 4. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)vfs_init.c	8.3 (Berkeley) 1/4/94
35  */
36 
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/fnv_hash.h>
43 #include <sys/kernel.h>
44 #include <sys/linker.h>
45 #include <sys/mount.h>
46 #include <sys/proc.h>
47 #include <sys/syscallsubr.h>
48 #include <sys/sysctl.h>
49 #include <sys/vnode.h>
50 #include <sys/malloc.h>
51 
52 static int	vfs_register(struct vfsconf *);
53 static int	vfs_unregister(struct vfsconf *);
54 
55 MALLOC_DEFINE(M_VNODE, "vnodes", "Dynamically allocated vnodes");
56 
57 /*
58  * The highest defined VFS number.
59  */
60 int maxvfsconf = VFS_GENERIC + 1;
61 
62 /*
63  * Single-linked list of configured VFSes.
64  * New entries are added/deleted by vfs_register()/vfs_unregister()
65  */
66 struct vfsconfhead vfsconf = TAILQ_HEAD_INITIALIZER(vfsconf);
67 
68 /*
69  * Loader.conf variable vfs.typenumhash enables setting vfc_typenum using a hash
70  * calculation on vfc_name, so that it doesn't change when file systems are
71  * loaded in a different order. This will avoid the NFS server file handles from
72  * changing for file systems that use vfc_typenum in their fsid.
73  */
74 static int	vfs_typenumhash = 1;
75 SYSCTL_INT(_vfs, OID_AUTO, typenumhash, CTLFLAG_RDTUN, &vfs_typenumhash, 0,
76     "Set vfc_typenum using a hash calculation on vfc_name, so that it does not"
77     "change when file systems are loaded in a different order.");
78 
79 /*
80  * A Zen vnode attribute structure.
81  *
82  * Initialized when the first filesystem registers by vfs_register().
83  */
84 struct vattr va_null;
85 
86 /*
87  * vfs_init.c
88  *
89  * Allocate and fill in operations vectors.
90  *
91  * An undocumented feature of this approach to defining operations is that
92  * there can be multiple entries in vfs_opv_descs for the same operations
93  * vector. This allows third parties to extend the set of operations
94  * supported by another layer in a binary compatibile way. For example,
95  * assume that NFS needed to be modified to support Ficus. NFS has an entry
96  * (probably nfs_vnopdeop_decls) declaring all the operations NFS supports by
97  * default. Ficus could add another entry (ficus_nfs_vnodeop_decl_entensions)
98  * listing those new operations Ficus adds to NFS, all without modifying the
99  * NFS code. (Of couse, the OTW NFS protocol still needs to be munged, but
100  * that is a(whole)nother story.) This is a feature.
101  */
102 
103 /*
104  * Routines having to do with the management of the vnode table.
105  */
106 
107 struct vfsconf *
108 vfs_byname(const char *name)
109 {
110 	struct vfsconf *vfsp;
111 
112 	if (!strcmp(name, "ffs"))
113 		name = "ufs";
114 	TAILQ_FOREACH(vfsp, &vfsconf, vfc_list)
115 		if (!strcmp(name, vfsp->vfc_name))
116 			return (vfsp);
117 	return (NULL);
118 }
119 
120 struct vfsconf *
121 vfs_byname_kld(const char *fstype, struct thread *td, int *error)
122 {
123 	struct vfsconf *vfsp;
124 	int fileid, loaded;
125 
126 	vfsp = vfs_byname(fstype);
127 	if (vfsp != NULL)
128 		return (vfsp);
129 
130 	/* Try to load the respective module. */
131 	*error = kern_kldload(td, fstype, &fileid);
132 	loaded = (*error == 0);
133 	if (*error == EEXIST)
134 		*error = 0;
135 	if (*error)
136 		return (NULL);
137 
138 	/* Look up again to see if the VFS was loaded. */
139 	vfsp = vfs_byname(fstype);
140 	if (vfsp == NULL) {
141 		if (loaded)
142 			(void)kern_kldunload(td, fileid, LINKER_UNLOAD_FORCE);
143 		*error = ENODEV;
144 		return (NULL);
145 	}
146 	return (vfsp);
147 }
148 
149 
150 /* Register a new filesystem type in the global table */
151 static int
152 vfs_register(struct vfsconf *vfc)
153 {
154 	struct sysctl_oid *oidp;
155 	struct vfsops *vfsops;
156 	static int once;
157 	struct vfsconf *tvfc;
158 	uint32_t hashval;
159 	int secondpass;
160 
161 	if (!once) {
162 		vattr_null(&va_null);
163 		once = 1;
164 	}
165 
166 	if (vfc->vfc_version != VFS_VERSION) {
167 		printf("ERROR: filesystem %s, unsupported ABI version %x\n",
168 		    vfc->vfc_name, vfc->vfc_version);
169 		return (EINVAL);
170 	}
171 	if (vfs_byname(vfc->vfc_name) != NULL)
172 		return (EEXIST);
173 
174 	if (vfs_typenumhash != 0) {
175 		/*
176 		 * Calculate a hash on vfc_name to use for vfc_typenum. Unless
177 		 * all of 1<->255 are assigned, it is limited to 8bits since
178 		 * that is what ZFS uses from vfc_typenum and is also the
179 		 * preferred range for vfs_getnewfsid().
180 		 */
181 		hashval = fnv_32_str(vfc->vfc_name, FNV1_32_INIT);
182 		hashval &= 0xff;
183 		secondpass = 0;
184 		do {
185 			/* Look for and fix any collision. */
186 			TAILQ_FOREACH(tvfc, &vfsconf, vfc_list) {
187 				if (hashval == tvfc->vfc_typenum) {
188 					if (hashval == 255 && secondpass == 0) {
189 						hashval = 1;
190 						secondpass = 1;
191 					} else
192 						hashval++;
193 					break;
194 				}
195 			}
196 		} while (tvfc != NULL);
197 		vfc->vfc_typenum = hashval;
198 		if (vfc->vfc_typenum >= maxvfsconf)
199 			maxvfsconf = vfc->vfc_typenum + 1;
200 	} else
201 		vfc->vfc_typenum = maxvfsconf++;
202 	TAILQ_INSERT_TAIL(&vfsconf, vfc, vfc_list);
203 
204 	/*
205 	 * If this filesystem has a sysctl node under vfs
206 	 * (i.e. vfs.xxfs), then change the oid number of that node to
207 	 * match the filesystem's type number.  This allows user code
208 	 * which uses the type number to read sysctl variables defined
209 	 * by the filesystem to continue working. Since the oids are
210 	 * in a sorted list, we need to make sure the order is
211 	 * preserved by re-registering the oid after modifying its
212 	 * number.
213 	 */
214 	sysctl_lock();
215 	SLIST_FOREACH(oidp, SYSCTL_CHILDREN(&sysctl___vfs), oid_link)
216 		if (strcmp(oidp->oid_name, vfc->vfc_name) == 0) {
217 			sysctl_unregister_oid(oidp);
218 			oidp->oid_number = vfc->vfc_typenum;
219 			sysctl_register_oid(oidp);
220 			break;
221 		}
222 	sysctl_unlock();
223 
224 	/*
225 	 * Initialise unused ``struct vfsops'' fields, to use
226 	 * the vfs_std*() functions.  Note, we need the mount
227 	 * and unmount operations, at the least.  The check
228 	 * for vfsops available is just a debugging aid.
229 	 */
230 	KASSERT(vfc->vfc_vfsops != NULL,
231 	    ("Filesystem %s has no vfsops", vfc->vfc_name));
232 	/*
233 	 * Check the mount and unmount operations.
234 	 */
235 	vfsops = vfc->vfc_vfsops;
236 	KASSERT(vfsops->vfs_mount != NULL,
237 	    ("Filesystem %s has no mount op", vfc->vfc_name));
238 	KASSERT(vfsops->vfs_unmount != NULL,
239 	    ("Filesystem %s has no unmount op", vfc->vfc_name));
240 
241 	if (vfsops->vfs_root == NULL)
242 		/* return file system's root vnode */
243 		vfsops->vfs_root =	vfs_stdroot;
244 	if (vfsops->vfs_quotactl == NULL)
245 		/* quota control */
246 		vfsops->vfs_quotactl =	vfs_stdquotactl;
247 	if (vfsops->vfs_statfs == NULL)
248 		/* return file system's status */
249 		vfsops->vfs_statfs =	vfs_stdstatfs;
250 	if (vfsops->vfs_sync == NULL)
251 		/*
252 		 * flush unwritten data (nosync)
253 		 * file systems can use vfs_stdsync
254 		 * explicitly by setting it in the
255 		 * vfsop vector.
256 		 */
257 		vfsops->vfs_sync =	vfs_stdnosync;
258 	if (vfsops->vfs_vget == NULL)
259 		/* convert an inode number to a vnode */
260 		vfsops->vfs_vget =	vfs_stdvget;
261 	if (vfsops->vfs_fhtovp == NULL)
262 		/* turn an NFS file handle into a vnode */
263 		vfsops->vfs_fhtovp =	vfs_stdfhtovp;
264 	if (vfsops->vfs_checkexp == NULL)
265 		/* check if file system is exported */
266 		vfsops->vfs_checkexp =	vfs_stdcheckexp;
267 	if (vfsops->vfs_init == NULL)
268 		/* file system specific initialisation */
269 		vfsops->vfs_init =	vfs_stdinit;
270 	if (vfsops->vfs_uninit == NULL)
271 		/* file system specific uninitialisation */
272 		vfsops->vfs_uninit =	vfs_stduninit;
273 	if (vfsops->vfs_extattrctl == NULL)
274 		/* extended attribute control */
275 		vfsops->vfs_extattrctl = vfs_stdextattrctl;
276 	if (vfsops->vfs_sysctl == NULL)
277 		vfsops->vfs_sysctl = vfs_stdsysctl;
278 
279 	/*
280 	 * Call init function for this VFS...
281 	 */
282 	(*(vfc->vfc_vfsops->vfs_init))(vfc);
283 
284 	return 0;
285 }
286 
287 
288 /* Remove registration of a filesystem type */
289 static int
290 vfs_unregister(struct vfsconf *vfc)
291 {
292 	struct vfsconf *vfsp;
293 	int error, i, maxtypenum;
294 
295 	i = vfc->vfc_typenum;
296 
297 	vfsp = vfs_byname(vfc->vfc_name);
298 	if (vfsp == NULL)
299 		return EINVAL;
300 	if (vfsp->vfc_refcount)
301 		return EBUSY;
302 	if (vfc->vfc_vfsops->vfs_uninit != NULL) {
303 		error = (*vfc->vfc_vfsops->vfs_uninit)(vfsp);
304 		if (error)
305 			return (error);
306 	}
307 	TAILQ_REMOVE(&vfsconf, vfsp, vfc_list);
308 	maxtypenum = VFS_GENERIC;
309 	TAILQ_FOREACH(vfsp, &vfsconf, vfc_list)
310 		if (maxtypenum < vfsp->vfc_typenum)
311 			maxtypenum = vfsp->vfc_typenum;
312 	maxvfsconf = maxtypenum + 1;
313 	return 0;
314 }
315 
316 /*
317  * Standard kernel module handling code for filesystem modules.
318  * Referenced from VFS_SET().
319  */
320 int
321 vfs_modevent(module_t mod, int type, void *data)
322 {
323 	struct vfsconf *vfc;
324 	int error = 0;
325 
326 	vfc = (struct vfsconf *)data;
327 
328 	switch (type) {
329 	case MOD_LOAD:
330 		if (vfc)
331 			error = vfs_register(vfc);
332 		break;
333 
334 	case MOD_UNLOAD:
335 		if (vfc)
336 			error = vfs_unregister(vfc);
337 		break;
338 	default:
339 		error = EOPNOTSUPP;
340 		break;
341 	}
342 	return (error);
343 }
344