xref: /freebsd/sys/kern/vfs_init.c (revision 197186496661d8f63639b5b27b33851f727d4da7)
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;
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 	if (*error)
134 		return (NULL);
135 
136 	/* Look up again to see if the VFS was loaded. */
137 	vfsp = vfs_byname(fstype);
138 	if (vfsp == NULL) {
139 		(void)kern_kldunload(td, fileid, LINKER_UNLOAD_FORCE);
140 		*error = ENODEV;
141 		return (NULL);
142 	}
143 	return (vfsp);
144 }
145 
146 
147 /* Register a new filesystem type in the global table */
148 static int
149 vfs_register(struct vfsconf *vfc)
150 {
151 	struct sysctl_oid *oidp;
152 	struct vfsops *vfsops;
153 	static int once;
154 	struct vfsconf *tvfc;
155 	uint32_t hashval;
156 	int secondpass;
157 
158 	if (!once) {
159 		vattr_null(&va_null);
160 		once = 1;
161 	}
162 
163 	if (vfc->vfc_version != VFS_VERSION) {
164 		printf("ERROR: filesystem %s, unsupported ABI version %x\n",
165 		    vfc->vfc_name, vfc->vfc_version);
166 		return (EINVAL);
167 	}
168 	if (vfs_byname(vfc->vfc_name) != NULL)
169 		return (EEXIST);
170 
171 	if (vfs_typenumhash != 0) {
172 		/*
173 		 * Calculate a hash on vfc_name to use for vfc_typenum. Unless
174 		 * all of 1<->255 are assigned, it is limited to 8bits since
175 		 * that is what ZFS uses from vfc_typenum and is also the
176 		 * preferred range for vfs_getnewfsid().
177 		 */
178 		hashval = fnv_32_str(vfc->vfc_name, FNV1_32_INIT);
179 		hashval &= 0xff;
180 		secondpass = 0;
181 		do {
182 			/* Look for and fix any collision. */
183 			TAILQ_FOREACH(tvfc, &vfsconf, vfc_list) {
184 				if (hashval == tvfc->vfc_typenum) {
185 					if (hashval == 255 && secondpass == 0) {
186 						hashval = 1;
187 						secondpass = 1;
188 					} else
189 						hashval++;
190 					break;
191 				}
192 			}
193 		} while (tvfc != NULL);
194 		vfc->vfc_typenum = hashval;
195 		if (vfc->vfc_typenum >= maxvfsconf)
196 			maxvfsconf = vfc->vfc_typenum + 1;
197 	} else
198 		vfc->vfc_typenum = maxvfsconf++;
199 	TAILQ_INSERT_TAIL(&vfsconf, vfc, vfc_list);
200 
201 	/*
202 	 * If this filesystem has a sysctl node under vfs
203 	 * (i.e. vfs.xxfs), then change the oid number of that node to
204 	 * match the filesystem's type number.  This allows user code
205 	 * which uses the type number to read sysctl variables defined
206 	 * by the filesystem to continue working. Since the oids are
207 	 * in a sorted list, we need to make sure the order is
208 	 * preserved by re-registering the oid after modifying its
209 	 * number.
210 	 */
211 	sysctl_lock();
212 	SLIST_FOREACH(oidp, &sysctl__vfs_children, oid_link)
213 		if (strcmp(oidp->oid_name, vfc->vfc_name) == 0) {
214 			sysctl_unregister_oid(oidp);
215 			oidp->oid_number = vfc->vfc_typenum;
216 			sysctl_register_oid(oidp);
217 			break;
218 		}
219 	sysctl_unlock();
220 
221 	/*
222 	 * Initialise unused ``struct vfsops'' fields, to use
223 	 * the vfs_std*() functions.  Note, we need the mount
224 	 * and unmount operations, at the least.  The check
225 	 * for vfsops available is just a debugging aid.
226 	 */
227 	KASSERT(vfc->vfc_vfsops != NULL,
228 	    ("Filesystem %s has no vfsops", vfc->vfc_name));
229 	/*
230 	 * Check the mount and unmount operations.
231 	 */
232 	vfsops = vfc->vfc_vfsops;
233 	KASSERT(vfsops->vfs_mount != NULL,
234 	    ("Filesystem %s has no mount op", vfc->vfc_name));
235 	KASSERT(vfsops->vfs_unmount != NULL,
236 	    ("Filesystem %s has no unmount op", vfc->vfc_name));
237 
238 	if (vfsops->vfs_root == NULL)
239 		/* return file system's root vnode */
240 		vfsops->vfs_root =	vfs_stdroot;
241 	if (vfsops->vfs_quotactl == NULL)
242 		/* quota control */
243 		vfsops->vfs_quotactl =	vfs_stdquotactl;
244 	if (vfsops->vfs_statfs == NULL)
245 		/* return file system's status */
246 		vfsops->vfs_statfs =	vfs_stdstatfs;
247 	if (vfsops->vfs_sync == NULL)
248 		/*
249 		 * flush unwritten data (nosync)
250 		 * file systems can use vfs_stdsync
251 		 * explicitly by setting it in the
252 		 * vfsop vector.
253 		 */
254 		vfsops->vfs_sync =	vfs_stdnosync;
255 	if (vfsops->vfs_vget == NULL)
256 		/* convert an inode number to a vnode */
257 		vfsops->vfs_vget =	vfs_stdvget;
258 	if (vfsops->vfs_fhtovp == NULL)
259 		/* turn an NFS file handle into a vnode */
260 		vfsops->vfs_fhtovp =	vfs_stdfhtovp;
261 	if (vfsops->vfs_checkexp == NULL)
262 		/* check if file system is exported */
263 		vfsops->vfs_checkexp =	vfs_stdcheckexp;
264 	if (vfsops->vfs_init == NULL)
265 		/* file system specific initialisation */
266 		vfsops->vfs_init =	vfs_stdinit;
267 	if (vfsops->vfs_uninit == NULL)
268 		/* file system specific uninitialisation */
269 		vfsops->vfs_uninit =	vfs_stduninit;
270 	if (vfsops->vfs_extattrctl == NULL)
271 		/* extended attribute control */
272 		vfsops->vfs_extattrctl = vfs_stdextattrctl;
273 	if (vfsops->vfs_sysctl == NULL)
274 		vfsops->vfs_sysctl = vfs_stdsysctl;
275 
276 	/*
277 	 * Call init function for this VFS...
278 	 */
279 	(*(vfc->vfc_vfsops->vfs_init))(vfc);
280 
281 	return 0;
282 }
283 
284 
285 /* Remove registration of a filesystem type */
286 static int
287 vfs_unregister(struct vfsconf *vfc)
288 {
289 	struct vfsconf *vfsp;
290 	int error, i, maxtypenum;
291 
292 	i = vfc->vfc_typenum;
293 
294 	vfsp = vfs_byname(vfc->vfc_name);
295 	if (vfsp == NULL)
296 		return EINVAL;
297 	if (vfsp->vfc_refcount)
298 		return EBUSY;
299 	if (vfc->vfc_vfsops->vfs_uninit != NULL) {
300 		error = (*vfc->vfc_vfsops->vfs_uninit)(vfsp);
301 		if (error)
302 			return (error);
303 	}
304 	TAILQ_REMOVE(&vfsconf, vfsp, vfc_list);
305 	maxtypenum = VFS_GENERIC;
306 	TAILQ_FOREACH(vfsp, &vfsconf, vfc_list)
307 		if (maxtypenum < vfsp->vfc_typenum)
308 			maxtypenum = vfsp->vfc_typenum;
309 	maxvfsconf = maxtypenum + 1;
310 	return 0;
311 }
312 
313 /*
314  * Standard kernel module handling code for filesystem modules.
315  * Referenced from VFS_SET().
316  */
317 int
318 vfs_modevent(module_t mod, int type, void *data)
319 {
320 	struct vfsconf *vfc;
321 	int error = 0;
322 
323 	vfc = (struct vfsconf *)data;
324 
325 	switch (type) {
326 	case MOD_LOAD:
327 		if (vfc)
328 			error = vfs_register(vfc);
329 		break;
330 
331 	case MOD_UNLOAD:
332 		if (vfc)
333 			error = vfs_unregister(vfc);
334 		break;
335 	default:
336 		error = EOPNOTSUPP;
337 		break;
338 	}
339 	return (error);
340 }
341