xref: /freebsd/sys/kern/vfs_init.c (revision 3bdf775801b218aa5a89564839405b122f4b233e)
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 TUNABLE_INT("vfs.typenumhash", &vfs_typenumhash);
76 SYSCTL_INT(_vfs, OID_AUTO, typenumhash, CTLFLAG_RDTUN, &vfs_typenumhash, 0,
77     "Set vfc_typenum using a hash calculation on vfc_name, so that it does not"
78     "change when file systems are loaded in a different order.");
79 
80 /*
81  * A Zen vnode attribute structure.
82  *
83  * Initialized when the first filesystem registers by vfs_register().
84  */
85 struct vattr va_null;
86 
87 /*
88  * vfs_init.c
89  *
90  * Allocate and fill in operations vectors.
91  *
92  * An undocumented feature of this approach to defining operations is that
93  * there can be multiple entries in vfs_opv_descs for the same operations
94  * vector. This allows third parties to extend the set of operations
95  * supported by another layer in a binary compatibile way. For example,
96  * assume that NFS needed to be modified to support Ficus. NFS has an entry
97  * (probably nfs_vnopdeop_decls) declaring all the operations NFS supports by
98  * default. Ficus could add another entry (ficus_nfs_vnodeop_decl_entensions)
99  * listing those new operations Ficus adds to NFS, all without modifying the
100  * NFS code. (Of couse, the OTW NFS protocol still needs to be munged, but
101  * that is a(whole)nother story.) This is a feature.
102  */
103 
104 /*
105  * Routines having to do with the management of the vnode table.
106  */
107 
108 struct vfsconf *
109 vfs_byname(const char *name)
110 {
111 	struct vfsconf *vfsp;
112 
113 	if (!strcmp(name, "ffs"))
114 		name = "ufs";
115 	TAILQ_FOREACH(vfsp, &vfsconf, vfc_list)
116 		if (!strcmp(name, vfsp->vfc_name))
117 			return (vfsp);
118 	return (NULL);
119 }
120 
121 struct vfsconf *
122 vfs_byname_kld(const char *fstype, struct thread *td, int *error)
123 {
124 	struct vfsconf *vfsp;
125 	int fileid, loaded;
126 
127 	vfsp = vfs_byname(fstype);
128 	if (vfsp != NULL)
129 		return (vfsp);
130 
131 	/* Try to load the respective module. */
132 	*error = kern_kldload(td, fstype, &fileid);
133 	loaded = (*error == 0);
134 	if (*error == EEXIST)
135 		*error = 0;
136 	if (*error)
137 		return (NULL);
138 
139 	/* Look up again to see if the VFS was loaded. */
140 	vfsp = vfs_byname(fstype);
141 	if (vfsp == NULL) {
142 		if (loaded)
143 			(void)kern_kldunload(td, fileid, LINKER_UNLOAD_FORCE);
144 		*error = ENODEV;
145 		return (NULL);
146 	}
147 	return (vfsp);
148 }
149 
150 
151 /* Register a new filesystem type in the global table */
152 static int
153 vfs_register(struct vfsconf *vfc)
154 {
155 	struct sysctl_oid *oidp;
156 	struct vfsops *vfsops;
157 	static int once;
158 	struct vfsconf *tvfc;
159 	uint32_t hashval;
160 	int secondpass;
161 
162 	if (!once) {
163 		vattr_null(&va_null);
164 		once = 1;
165 	}
166 
167 	if (vfc->vfc_version != VFS_VERSION) {
168 		printf("ERROR: filesystem %s, unsupported ABI version %x\n",
169 		    vfc->vfc_name, vfc->vfc_version);
170 		return (EINVAL);
171 	}
172 	if (vfs_byname(vfc->vfc_name) != NULL)
173 		return (EEXIST);
174 
175 	if (vfs_typenumhash != 0) {
176 		/*
177 		 * Calculate a hash on vfc_name to use for vfc_typenum. Unless
178 		 * all of 1<->255 are assigned, it is limited to 8bits since
179 		 * that is what ZFS uses from vfc_typenum and is also the
180 		 * preferred range for vfs_getnewfsid().
181 		 */
182 		hashval = fnv_32_str(vfc->vfc_name, FNV1_32_INIT);
183 		hashval &= 0xff;
184 		secondpass = 0;
185 		do {
186 			/* Look for and fix any collision. */
187 			TAILQ_FOREACH(tvfc, &vfsconf, vfc_list) {
188 				if (hashval == tvfc->vfc_typenum) {
189 					if (hashval == 255 && secondpass == 0) {
190 						hashval = 1;
191 						secondpass = 1;
192 					} else
193 						hashval++;
194 					break;
195 				}
196 			}
197 		} while (tvfc != NULL);
198 		vfc->vfc_typenum = hashval;
199 		if (vfc->vfc_typenum >= maxvfsconf)
200 			maxvfsconf = vfc->vfc_typenum + 1;
201 	} else
202 		vfc->vfc_typenum = maxvfsconf++;
203 	TAILQ_INSERT_TAIL(&vfsconf, vfc, vfc_list);
204 
205 	/*
206 	 * If this filesystem has a sysctl node under vfs
207 	 * (i.e. vfs.xxfs), then change the oid number of that node to
208 	 * match the filesystem's type number.  This allows user code
209 	 * which uses the type number to read sysctl variables defined
210 	 * by the filesystem to continue working. Since the oids are
211 	 * in a sorted list, we need to make sure the order is
212 	 * preserved by re-registering the oid after modifying its
213 	 * number.
214 	 */
215 	sysctl_lock();
216 	SLIST_FOREACH(oidp, &sysctl__vfs_children, oid_link)
217 		if (strcmp(oidp->oid_name, vfc->vfc_name) == 0) {
218 			sysctl_unregister_oid(oidp);
219 			oidp->oid_number = vfc->vfc_typenum;
220 			sysctl_register_oid(oidp);
221 			break;
222 		}
223 	sysctl_unlock();
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 
285 	return 0;
286 }
287 
288 
289 /* Remove registration of a filesystem type */
290 static int
291 vfs_unregister(struct vfsconf *vfc)
292 {
293 	struct vfsconf *vfsp;
294 	int error, i, maxtypenum;
295 
296 	i = vfc->vfc_typenum;
297 
298 	vfsp = vfs_byname(vfc->vfc_name);
299 	if (vfsp == NULL)
300 		return EINVAL;
301 	if (vfsp->vfc_refcount)
302 		return EBUSY;
303 	if (vfc->vfc_vfsops->vfs_uninit != NULL) {
304 		error = (*vfc->vfc_vfsops->vfs_uninit)(vfsp);
305 		if (error)
306 			return (error);
307 	}
308 	TAILQ_REMOVE(&vfsconf, vfsp, vfc_list);
309 	maxtypenum = VFS_GENERIC;
310 	TAILQ_FOREACH(vfsp, &vfsconf, vfc_list)
311 		if (maxtypenum < vfsp->vfc_typenum)
312 			maxtypenum = vfsp->vfc_typenum;
313 	maxvfsconf = maxtypenum + 1;
314 	return 0;
315 }
316 
317 /*
318  * Standard kernel module handling code for filesystem modules.
319  * Referenced from VFS_SET().
320  */
321 int
322 vfs_modevent(module_t mod, int type, void *data)
323 {
324 	struct vfsconf *vfc;
325 	int error = 0;
326 
327 	vfc = (struct vfsconf *)data;
328 
329 	switch (type) {
330 	case MOD_LOAD:
331 		if (vfc)
332 			error = vfs_register(vfc);
333 		break;
334 
335 	case MOD_UNLOAD:
336 		if (vfc)
337 			error = vfs_unregister(vfc);
338 		break;
339 	default:
340 		error = EOPNOTSUPP;
341 		break;
342 	}
343 	return (error);
344 }
345