xref: /freebsd/sys/kern/vfs_init.c (revision c17d43407fe04133a94055b0dbc7ea8965654a9f)
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  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)vfs_init.c	8.3 (Berkeley) 1/4/94
39  * $FreeBSD$
40  */
41 
42 
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/kernel.h>
46 #include <sys/mount.h>
47 #include <sys/sysctl.h>
48 #include <sys/vnode.h>
49 #include <sys/malloc.h>
50 
51 
52 MALLOC_DEFINE(M_VNODE, "vnodes", "Dynamically allocated vnodes");
53 
54 /*
55  * The highest defined VFS number.
56  */
57 int maxvfsconf = VFS_GENERIC + 1;
58 
59 /*
60  * Single-linked list of configured VFSes.
61  * New entries are added/deleted by vfs_register()/vfs_unregister()
62  */
63 struct vfsconf *vfsconf;
64 
65 /*
66  * vfs_init.c
67  *
68  * Allocate and fill in operations vectors.
69  *
70  * An undocumented feature of this approach to defining operations is that
71  * there can be multiple entries in vfs_opv_descs for the same operations
72  * vector. This allows third parties to extend the set of operations
73  * supported by another layer in a binary compatibile way. For example,
74  * assume that NFS needed to be modified to support Ficus. NFS has an entry
75  * (probably nfs_vnopdeop_decls) declaring all the operations NFS supports by
76  * default. Ficus could add another entry (ficus_nfs_vnodeop_decl_entensions)
77  * listing those new operations Ficus adds to NFS, all without modifying the
78  * NFS code. (Of couse, the OTW NFS protocol still needs to be munged, but
79  * that is a(whole)nother story.) This is a feature.
80  */
81 
82 /* Table of known vnodeop vectors (list of VFS vnode vectors) */
83 static const struct vnodeopv_desc **vnodeopv_descs;
84 static int vnodeopv_num;
85 
86 /* Table of known descs (list of vnode op handlers "vop_access_desc") */
87 static struct vnodeop_desc **vfs_op_descs;
88 /* Reference counts for vfs_op_descs */
89 static int *vfs_op_desc_refs;
90 /* Number of descriptions */
91 static int num_op_descs;
92 /* Number of entries in each description */
93 static int vfs_opv_numops;
94 
95 /*
96  * Recalculate the operations vector/description (those parts of it that can
97  * be recalculated, that is.)
98  * XXX It may be preferable to replace this function with an invariant check
99  * and a set of functions that should keep the table invariant.
100  */
101 static void
102 vfs_opv_recalc(void)
103 {
104 	int i, j;
105 	vop_t ***opv_desc_vector_p;
106 	vop_t **opv_desc_vector;
107 	struct vnodeopv_entry_desc *opve_descp;
108 	const struct vnodeopv_desc *opv;
109 
110 	if (vfs_op_descs == NULL)
111 		panic("vfs_opv_recalc called with null vfs_op_descs");
112 
113 	/*
114 	 * Run through and make sure all known descs have an offset
115 	 *
116 	 * vop_default_desc is hardwired at offset 1, and offset 0
117 	 * is a panic sanity check.
118 	 */
119 	vfs_opv_numops = 0;
120 	for (i = 0; i < num_op_descs; i++)
121 		if (vfs_opv_numops < (vfs_op_descs[i]->vdesc_offset + 1))
122 			vfs_opv_numops = vfs_op_descs[i]->vdesc_offset + 1;
123 	for (i = 0; i < num_op_descs; i++)
124 		if (vfs_op_descs[i]->vdesc_offset == 0)
125 			vfs_op_descs[i]->vdesc_offset = vfs_opv_numops++;
126 	/*
127 	 * Allocate and fill in the vectors
128 	 */
129 	for (i = 0; i < vnodeopv_num; i++) {
130 		opv = vnodeopv_descs[i];
131 		opv_desc_vector_p = opv->opv_desc_vector_p;
132 		if (*opv_desc_vector_p)
133 			FREE(*opv_desc_vector_p, M_VNODE);
134 		MALLOC(*opv_desc_vector_p, vop_t **,
135 			vfs_opv_numops * sizeof(vop_t *), M_VNODE,
136 			M_WAITOK | M_ZERO);
137 		if (*opv_desc_vector_p == NULL)
138 			panic("no memory for vop_t ** vector");
139 
140 		/* Fill in, with slot 0 being to return EOPNOTSUPP */
141 		opv_desc_vector = *opv_desc_vector_p;
142 		opv_desc_vector[0] = (vop_t *)vop_eopnotsupp;
143 		for (j = 0; opv->opv_desc_ops[j].opve_op; j++) {
144 			opve_descp = &(opv->opv_desc_ops[j]);
145 			opv_desc_vector[opve_descp->opve_op->vdesc_offset] =
146 				opve_descp->opve_impl;
147 		}
148 
149 		/* Replace unfilled routines with their default (slot 1). */
150 		opv_desc_vector = *(opv->opv_desc_vector_p);
151 		if (opv_desc_vector[1] == NULL)
152 			panic("vfs_opv_recalc: vector without a default.");
153 		for (j = 0; j < vfs_opv_numops; j++)
154 			if (opv_desc_vector[j] == NULL)
155 				opv_desc_vector[j] = opv_desc_vector[1];
156 	}
157 }
158 
159 /* Add a set of vnode operations (a description) to the table above. */
160 void
161 vfs_add_vnodeops(const void *data)
162 {
163 	const struct vnodeopv_desc *opv;
164 	const struct vnodeopv_desc **newopv;
165 	struct vnodeop_desc **newop;
166 	int *newref;
167 	vop_t **opv_desc_vector;
168 	struct vnodeop_desc *desc;
169 	int i, j;
170 
171 	opv = (const struct vnodeopv_desc *)data;
172 	MALLOC(newopv, const struct vnodeopv_desc **,
173 	       (vnodeopv_num + 1) * sizeof(*newopv), M_VNODE, M_WAITOK);
174 	if (newopv == NULL)
175 		panic("vfs_add_vnodeops: no memory");
176 	if (vnodeopv_descs) {
177 		bcopy(vnodeopv_descs, newopv, vnodeopv_num * sizeof(*newopv));
178 		FREE(vnodeopv_descs, M_VNODE);
179 	}
180 	newopv[vnodeopv_num] = opv;
181 	vnodeopv_descs = newopv;
182 	vnodeopv_num++;
183 
184 	/* See if we have turned up a new vnode op desc */
185 	opv_desc_vector = *(opv->opv_desc_vector_p);
186 	for (i = 0; (desc = opv->opv_desc_ops[i].opve_op); i++) {
187 		for (j = 0; j < num_op_descs; j++) {
188 			if (desc == vfs_op_descs[j]) {
189 				/* found it, increase reference count */
190 				vfs_op_desc_refs[j]++;
191 				break;
192 			}
193 		}
194 		if (j == num_op_descs) {
195 			/* not found, new entry */
196 			MALLOC(newop, struct vnodeop_desc **,
197 			       (num_op_descs + 1) * sizeof(*newop),
198 			       M_VNODE, M_WAITOK);
199 			if (newop == NULL)
200 				panic("vfs_add_vnodeops: no memory for desc");
201 			/* new reference count (for unload) */
202 			MALLOC(newref, int *,
203 				(num_op_descs + 1) * sizeof(*newref),
204 				M_VNODE, M_WAITOK);
205 			if (newref == NULL)
206 				panic("vfs_add_vnodeops: no memory for refs");
207 			if (vfs_op_descs) {
208 				bcopy(vfs_op_descs, newop,
209 					num_op_descs * sizeof(*newop));
210 				FREE(vfs_op_descs, M_VNODE);
211 			}
212 			if (vfs_op_desc_refs) {
213 				bcopy(vfs_op_desc_refs, newref,
214 					num_op_descs * sizeof(*newref));
215 				FREE(vfs_op_desc_refs, M_VNODE);
216 			}
217 			newop[num_op_descs] = desc;
218 			newref[num_op_descs] = 1;
219 			vfs_op_descs = newop;
220 			vfs_op_desc_refs = newref;
221 			num_op_descs++;
222 		}
223 	}
224 	vfs_opv_recalc();
225 }
226 
227 /* Remove a vnode type from the vnode description table above. */
228 void
229 vfs_rm_vnodeops(const void *data)
230 {
231 	const struct vnodeopv_desc *opv;
232 	const struct vnodeopv_desc **newopv;
233 	struct vnodeop_desc **newop;
234 	int *newref;
235 	vop_t **opv_desc_vector;
236 	struct vnodeop_desc *desc;
237 	int i, j, k;
238 
239 	opv = (const struct vnodeopv_desc *)data;
240 	/* Lower ref counts on descs in the table and release if zero */
241 	opv_desc_vector = *(opv->opv_desc_vector_p);
242 	for (i = 0; (desc = opv->opv_desc_ops[i].opve_op); i++) {
243 		for (j = 0; j < num_op_descs; j++) {
244 			if (desc == vfs_op_descs[j]) {
245 				/* found it, decrease reference count */
246 				vfs_op_desc_refs[j]--;
247 				break;
248 			}
249 		}
250 		for (j = 0; j < num_op_descs; j++) {
251 			if (vfs_op_desc_refs[j] > 0)
252 				continue;
253 			if (vfs_op_desc_refs[j] < 0)
254 				panic("vfs_remove_vnodeops: negative refcnt");
255 			MALLOC(newop, struct vnodeop_desc **,
256 			       (num_op_descs - 1) * sizeof(*newop),
257 			       M_VNODE, M_WAITOK);
258 			if (newop == NULL)
259 				panic("vfs_remove_vnodeops: no memory for desc");
260 			/* new reference count (for unload) */
261 			MALLOC(newref, int *,
262 				(num_op_descs - 1) * sizeof(*newref),
263 				M_VNODE, M_WAITOK);
264 			if (newref == NULL)
265 				panic("vfs_remove_vnodeops: no memory for refs");
266 			for (k = j; k < (num_op_descs - 1); k++) {
267 				vfs_op_descs[k] = vfs_op_descs[k + 1];
268 				vfs_op_desc_refs[k] = vfs_op_desc_refs[k + 1];
269 			}
270 			bcopy(vfs_op_descs, newop,
271 				(num_op_descs - 1) * sizeof(*newop));
272 			bcopy(vfs_op_desc_refs, newref,
273 				(num_op_descs - 1) * sizeof(*newref));
274 			FREE(vfs_op_descs, M_VNODE);
275 			FREE(vfs_op_desc_refs, M_VNODE);
276 			vfs_op_descs = newop;
277 			vfs_op_desc_refs = newref;
278 			num_op_descs--;
279 		}
280 	}
281 
282 	for (i = 0; i < vnodeopv_num; i++) {
283 		if (vnodeopv_descs[i] == opv) {
284 			for (j = i; j < (vnodeopv_num - 1); j++)
285 				vnodeopv_descs[j] = vnodeopv_descs[j + 1];
286 			break;
287 		}
288 	}
289 	if (i == vnodeopv_num)
290 		panic("vfs_remove_vnodeops: opv not found");
291 	MALLOC(newopv, const struct vnodeopv_desc **,
292 	       (vnodeopv_num - 1) * sizeof(*newopv), M_VNODE, M_WAITOK);
293 	if (newopv == NULL)
294 		panic("vfs_remove_vnodeops: no memory");
295 	bcopy(vnodeopv_descs, newopv, (vnodeopv_num - 1) * sizeof(*newopv));
296 	FREE(vnodeopv_descs, M_VNODE);
297 	vnodeopv_descs = newopv;
298 	vnodeopv_num--;
299 
300 	vfs_opv_recalc();
301 }
302 
303 /*
304  * Routines having to do with the management of the vnode table.
305  */
306 struct vattr va_null;
307 
308 /*
309  * Initialize the vnode structures and initialize each file system type.
310  */
311 /* ARGSUSED*/
312 static void
313 vfsinit(void *dummy)
314 {
315 
316 	vattr_null(&va_null);
317 }
318 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vfsinit, NULL)
319 
320 /* Register a new file system type in the global table */
321 int
322 vfs_register(struct vfsconf *vfc)
323 {
324 	struct sysctl_oid *oidp;
325 	struct vfsconf *vfsp;
326 
327 	vfsp = NULL;
328 	if (vfsconf)
329 		for (vfsp = vfsconf; vfsp->vfc_next; vfsp = vfsp->vfc_next)
330 			if (strcmp(vfc->vfc_name, vfsp->vfc_name) == 0)
331 				return EEXIST;
332 
333 	vfc->vfc_typenum = maxvfsconf++;
334 	if (vfsp)
335 		vfsp->vfc_next = vfc;
336 	else
337 		vfsconf = vfc;
338 	vfc->vfc_next = NULL;
339 
340 	/*
341 	 * If this filesystem has a sysctl node under vfs
342 	 * (i.e. vfs.xxfs), then change the oid number of that node to
343 	 * match the filesystem's type number.  This allows user code
344 	 * which uses the type number to read sysctl variables defined
345 	 * by the filesystem to continue working. Since the oids are
346 	 * in a sorted list, we need to make sure the order is
347 	 * preserved by re-registering the oid after modifying its
348 	 * number.
349 	 */
350 	SLIST_FOREACH(oidp, &sysctl__vfs_children, oid_link)
351 		if (strcmp(oidp->oid_name, vfc->vfc_name) == 0) {
352 			sysctl_unregister_oid(oidp);
353 			oidp->oid_number = vfc->vfc_typenum;
354 			sysctl_register_oid(oidp);
355 		}
356 
357 	/*
358 	 * Call init function for this VFS...
359 	 */
360 	(*(vfc->vfc_vfsops->vfs_init))(vfc);
361 
362 	return 0;
363 }
364 
365 
366 /* Remove registration of a file system type */
367 int
368 vfs_unregister(struct vfsconf *vfc)
369 {
370 	struct vfsconf *vfsp, *prev_vfsp;
371 	int error, i, maxtypenum;
372 
373 	i = vfc->vfc_typenum;
374 
375 	prev_vfsp = NULL;
376 	for (vfsp = vfsconf; vfsp;
377 			prev_vfsp = vfsp, vfsp = vfsp->vfc_next) {
378 		if (!strcmp(vfc->vfc_name, vfsp->vfc_name))
379 			break;
380 	}
381 	if (vfsp == NULL)
382 		return EINVAL;
383 	if (vfsp->vfc_refcount)
384 		return EBUSY;
385 	if (vfc->vfc_vfsops->vfs_uninit != NULL) {
386 		error = (*vfc->vfc_vfsops->vfs_uninit)(vfsp);
387 		if (error)
388 			return (error);
389 	}
390 	if (prev_vfsp)
391 		prev_vfsp->vfc_next = vfsp->vfc_next;
392 	else
393 		vfsconf = vfsp->vfc_next;
394 	maxtypenum = VFS_GENERIC;
395 	for (vfsp = vfsconf; vfsp != NULL; vfsp = vfsp->vfc_next)
396 		if (maxtypenum < vfsp->vfc_typenum)
397 			maxtypenum = vfsp->vfc_typenum;
398 	maxvfsconf = maxtypenum + 1;
399 	return 0;
400 }
401 
402 /*
403  * Standard kernel module handling code for file system modules.
404  * Referenced from VFS_SET().
405  */
406 int
407 vfs_modevent(module_t mod, int type, void *data)
408 {
409 	struct vfsconf *vfc;
410 	int error = 0;
411 
412 	vfc = (struct vfsconf *)data;
413 
414 	switch (type) {
415 	case MOD_LOAD:
416 		if (vfc)
417 			error = vfs_register(vfc);
418 		break;
419 
420 	case MOD_UNLOAD:
421 		if (vfc)
422 			error = vfs_unregister(vfc);
423 		break;
424 	default:	/* including MOD_SHUTDOWN */
425 		break;
426 	}
427 	return (error);
428 }
429