xref: /freebsd/sys/kern/vfs_init.c (revision ff0ba87247820afbdfdc1b307c803f7923d0e4d3)
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/sx.h>
48 #include <sys/syscallsubr.h>
49 #include <sys/sysctl.h>
50 #include <sys/vnode.h>
51 #include <sys/malloc.h>
52 
53 static int	vfs_register(struct vfsconf *);
54 static int	vfs_unregister(struct vfsconf *);
55 
56 MALLOC_DEFINE(M_VNODE, "vnodes", "Dynamically allocated vnodes");
57 
58 /*
59  * The highest defined VFS number.
60  */
61 int maxvfsconf = VFS_GENERIC + 1;
62 
63 /*
64  * Single-linked list of configured VFSes.
65  * New entries are added/deleted by vfs_register()/vfs_unregister()
66  */
67 struct vfsconfhead vfsconf = TAILQ_HEAD_INITIALIZER(vfsconf);
68 struct sx vfsconf_sx;
69 SX_SYSINIT(vfsconf, &vfsconf_sx, "vfsconf");
70 
71 /*
72  * Loader.conf variable vfs.typenumhash enables setting vfc_typenum using a hash
73  * calculation on vfc_name, so that it doesn't change when file systems are
74  * loaded in a different order. This will avoid the NFS server file handles from
75  * changing for file systems that use vfc_typenum in their fsid.
76  */
77 static int	vfs_typenumhash = 1;
78 SYSCTL_INT(_vfs, OID_AUTO, typenumhash, CTLFLAG_RDTUN, &vfs_typenumhash, 0,
79     "Set vfc_typenum using a hash calculation on vfc_name, so that it does not"
80     "change when file systems are loaded in a different order.");
81 
82 /*
83  * A Zen vnode attribute structure.
84  *
85  * Initialized when the first filesystem registers by vfs_register().
86  */
87 struct vattr va_null;
88 
89 /*
90  * vfs_init.c
91  *
92  * Allocate and fill in operations vectors.
93  *
94  * An undocumented feature of this approach to defining operations is that
95  * there can be multiple entries in vfs_opv_descs for the same operations
96  * vector. This allows third parties to extend the set of operations
97  * supported by another layer in a binary compatibile way. For example,
98  * assume that NFS needed to be modified to support Ficus. NFS has an entry
99  * (probably nfs_vnopdeop_decls) declaring all the operations NFS supports by
100  * default. Ficus could add another entry (ficus_nfs_vnodeop_decl_entensions)
101  * listing those new operations Ficus adds to NFS, all without modifying the
102  * NFS code. (Of couse, the OTW NFS protocol still needs to be munged, but
103  * that is a(whole)nother story.) This is a feature.
104  */
105 
106 /*
107  * Routines having to do with the management of the vnode table.
108  */
109 
110 static struct vfsconf *
111 vfs_byname_locked(const char *name)
112 {
113 	struct vfsconf *vfsp;
114 
115 	sx_assert(&vfsconf_sx, SA_LOCKED);
116 	if (!strcmp(name, "ffs"))
117 		name = "ufs";
118 	TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
119 		if (!strcmp(name, vfsp->vfc_name))
120 			return (vfsp);
121 	}
122 	return (NULL);
123 }
124 
125 struct vfsconf *
126 vfs_byname(const char *name)
127 {
128 	struct vfsconf *vfsp;
129 
130 	vfsconf_slock();
131 	vfsp = vfs_byname_locked(name);
132 	vfsconf_sunlock();
133 	return (vfsp);
134 }
135 
136 struct vfsconf *
137 vfs_byname_kld(const char *fstype, struct thread *td, int *error)
138 {
139 	struct vfsconf *vfsp;
140 	int fileid, loaded;
141 
142 	vfsp = vfs_byname(fstype);
143 	if (vfsp != NULL)
144 		return (vfsp);
145 
146 	/* Try to load the respective module. */
147 	*error = kern_kldload(td, fstype, &fileid);
148 	loaded = (*error == 0);
149 	if (*error == EEXIST)
150 		*error = 0;
151 	if (*error)
152 		return (NULL);
153 
154 	/* Look up again to see if the VFS was loaded. */
155 	vfsp = vfs_byname(fstype);
156 	if (vfsp == NULL) {
157 		if (loaded)
158 			(void)kern_kldunload(td, fileid, LINKER_UNLOAD_FORCE);
159 		*error = ENODEV;
160 		return (NULL);
161 	}
162 	return (vfsp);
163 }
164 
165 
166 /* Register a new filesystem type in the global table */
167 static int
168 vfs_register(struct vfsconf *vfc)
169 {
170 	struct sysctl_oid *oidp;
171 	struct vfsops *vfsops;
172 	static int once;
173 	struct vfsconf *tvfc;
174 	uint32_t hashval;
175 	int secondpass;
176 
177 	if (!once) {
178 		vattr_null(&va_null);
179 		once = 1;
180 	}
181 
182 	if (vfc->vfc_version != VFS_VERSION) {
183 		printf("ERROR: filesystem %s, unsupported ABI version %x\n",
184 		    vfc->vfc_name, vfc->vfc_version);
185 		return (EINVAL);
186 	}
187 	vfsconf_lock();
188 	if (vfs_byname_locked(vfc->vfc_name) != NULL) {
189 		vfsconf_unlock();
190 		return (EEXIST);
191 	}
192 
193 	if (vfs_typenumhash != 0) {
194 		/*
195 		 * Calculate a hash on vfc_name to use for vfc_typenum. Unless
196 		 * all of 1<->255 are assigned, it is limited to 8bits since
197 		 * that is what ZFS uses from vfc_typenum and is also the
198 		 * preferred range for vfs_getnewfsid().
199 		 */
200 		hashval = fnv_32_str(vfc->vfc_name, FNV1_32_INIT);
201 		hashval &= 0xff;
202 		secondpass = 0;
203 		do {
204 			/* Look for and fix any collision. */
205 			TAILQ_FOREACH(tvfc, &vfsconf, vfc_list) {
206 				if (hashval == tvfc->vfc_typenum) {
207 					if (hashval == 255 && secondpass == 0) {
208 						hashval = 1;
209 						secondpass = 1;
210 					} else
211 						hashval++;
212 					break;
213 				}
214 			}
215 		} while (tvfc != NULL);
216 		vfc->vfc_typenum = hashval;
217 		if (vfc->vfc_typenum >= maxvfsconf)
218 			maxvfsconf = vfc->vfc_typenum + 1;
219 	} else
220 		vfc->vfc_typenum = maxvfsconf++;
221 	TAILQ_INSERT_TAIL(&vfsconf, vfc, vfc_list);
222 
223 	/*
224 	 * Initialise unused ``struct vfsops'' fields, to use
225 	 * the vfs_std*() functions.  Note, we need the mount
226 	 * and unmount operations, at the least.  The check
227 	 * for vfsops available is just a debugging aid.
228 	 */
229 	KASSERT(vfc->vfc_vfsops != NULL,
230 	    ("Filesystem %s has no vfsops", vfc->vfc_name));
231 	/*
232 	 * Check the mount and unmount operations.
233 	 */
234 	vfsops = vfc->vfc_vfsops;
235 	KASSERT(vfsops->vfs_mount != NULL,
236 	    ("Filesystem %s has no mount op", vfc->vfc_name));
237 	KASSERT(vfsops->vfs_unmount != NULL,
238 	    ("Filesystem %s has no unmount op", vfc->vfc_name));
239 
240 	if (vfsops->vfs_root == NULL)
241 		/* return file system's root vnode */
242 		vfsops->vfs_root =	vfs_stdroot;
243 	if (vfsops->vfs_quotactl == NULL)
244 		/* quota control */
245 		vfsops->vfs_quotactl =	vfs_stdquotactl;
246 	if (vfsops->vfs_statfs == NULL)
247 		/* return file system's status */
248 		vfsops->vfs_statfs =	vfs_stdstatfs;
249 	if (vfsops->vfs_sync == NULL)
250 		/*
251 		 * flush unwritten data (nosync)
252 		 * file systems can use vfs_stdsync
253 		 * explicitly by setting it in the
254 		 * vfsop vector.
255 		 */
256 		vfsops->vfs_sync =	vfs_stdnosync;
257 	if (vfsops->vfs_vget == NULL)
258 		/* convert an inode number to a vnode */
259 		vfsops->vfs_vget =	vfs_stdvget;
260 	if (vfsops->vfs_fhtovp == NULL)
261 		/* turn an NFS file handle into a vnode */
262 		vfsops->vfs_fhtovp =	vfs_stdfhtovp;
263 	if (vfsops->vfs_checkexp == NULL)
264 		/* check if file system is exported */
265 		vfsops->vfs_checkexp =	vfs_stdcheckexp;
266 	if (vfsops->vfs_init == NULL)
267 		/* file system specific initialisation */
268 		vfsops->vfs_init =	vfs_stdinit;
269 	if (vfsops->vfs_uninit == NULL)
270 		/* file system specific uninitialisation */
271 		vfsops->vfs_uninit =	vfs_stduninit;
272 	if (vfsops->vfs_extattrctl == NULL)
273 		/* extended attribute control */
274 		vfsops->vfs_extattrctl = vfs_stdextattrctl;
275 	if (vfsops->vfs_sysctl == NULL)
276 		vfsops->vfs_sysctl = vfs_stdsysctl;
277 
278 	/*
279 	 * Call init function for this VFS...
280 	 */
281 	(*(vfc->vfc_vfsops->vfs_init))(vfc);
282 	vfsconf_unlock();
283 
284 	/*
285 	 * If this filesystem has a sysctl node under vfs
286 	 * (i.e. vfs.xxfs), then change the oid number of that node to
287 	 * match the filesystem's type number.  This allows user code
288 	 * which uses the type number to read sysctl variables defined
289 	 * by the filesystem to continue working. Since the oids are
290 	 * in a sorted list, we need to make sure the order is
291 	 * preserved by re-registering the oid after modifying its
292 	 * number.
293 	 */
294 	sysctl_xlock();
295 	SLIST_FOREACH(oidp, SYSCTL_CHILDREN(&sysctl___vfs), oid_link) {
296 		if (strcmp(oidp->oid_name, vfc->vfc_name) == 0) {
297 			sysctl_unregister_oid(oidp);
298 			oidp->oid_number = vfc->vfc_typenum;
299 			sysctl_register_oid(oidp);
300 			break;
301 		}
302 	}
303 	sysctl_xunlock();
304 
305 	return (0);
306 }
307 
308 
309 /* Remove registration of a filesystem type */
310 static int
311 vfs_unregister(struct vfsconf *vfc)
312 {
313 	struct vfsconf *vfsp;
314 	int error, i, maxtypenum;
315 
316 	i = vfc->vfc_typenum;
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