xref: /freebsd/sys/kern/vfs_init.c (revision 7f3dea244c40159a41ab22da77a434d7c5b5e85a)
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  * $Id: vfs_init.c,v 1.45 1999/03/07 16:06:41 dfr Exp $
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 #include <vm/vm_zone.h>
51 
52 
53 MALLOC_DEFINE(M_VNODE, "vnodes", "Dynamically allocated vnodes");
54 
55 /*
56  * XXX this bloat just exands the sysctl__vfs linker set a little so that
57  * we can attach sysctls for VFS modules without expanding the linker set.
58  * Currently (1998/09/06), only one VFS uses sysctls, so 2 extra linker
59  * set slots are more than sufficient.
60  */
61 static int mod_xx;
62 SYSCTL_INT(_vfs, OID_AUTO, mod0, CTLFLAG_RD, &mod_xx, 0, "");
63 SYSCTL_INT(_vfs, OID_AUTO, mod1, CTLFLAG_RD, &mod_xx, 0, "");
64 
65 /*
66  * Zone for namei
67  */
68 struct vm_zone *namei_zone;
69 
70 /*
71  * vfs_init.c
72  *
73  * Allocate and fill in operations vectors.
74  *
75  * An undocumented feature of this approach to defining operations is that
76  * there can be multiple entries in vfs_opv_descs for the same operations
77  * vector. This allows third parties to extend the set of operations
78  * supported by another layer in a binary compatibile way. For example,
79  * assume that NFS needed to be modified to support Ficus. NFS has an entry
80  * (probably nfs_vnopdeop_decls) declaring all the operations NFS supports by
81  * default. Ficus could add another entry (ficus_nfs_vnodeop_decl_entensions)
82  * listing those new operations Ficus adds to NFS, all without modifying the
83  * NFS code. (Of couse, the OTW NFS protocol still needs to be munged, but
84  * that is a(whole)nother story.) This is a feature.
85  */
86 
87 /* Table of known vnodeop vectors (list of VFS vnode vectors) */
88 static const struct vnodeopv_desc **vnodeopv_descs;
89 static int vnodeopv_num;
90 
91 /* Table of known descs (list of vnode op handlers "vop_access_desc") */
92 static struct vnodeop_desc **vfs_op_descs;
93 static int *vfs_op_desc_refs;			/* reference counts */
94 static int num_op_descs;
95 static int vfs_opv_numops;
96 
97 static void
98 vfs_opv_recalc(void)
99 {
100 	int i, j;
101 	vop_t ***opv_desc_vector_p;
102 	vop_t **opv_desc_vector;
103 	struct vnodeopv_entry_desc *opve_descp;
104 	const struct vnodeopv_desc *opv;
105 
106 	if (vfs_op_descs == NULL)
107 		panic("vfs_opv_recalc called with null vfs_op_descs");
108 
109 	/*
110 	 * Run through and make sure all known descs have an offset
111 	 *
112 	 * vop_default_desc is hardwired at offset 1, and offset 0
113 	 * is a panic sanity check.
114 	 */
115 	vfs_opv_numops = 0;
116 	for (i = 0; i < num_op_descs; i++)
117 		if (vfs_opv_numops < (vfs_op_descs[i]->vdesc_offset + 1))
118 			vfs_opv_numops = vfs_op_descs[i]->vdesc_offset + 1;
119 	for (i = 0; i < num_op_descs; i++)
120 		if (vfs_op_descs[i]->vdesc_offset == 0)
121 			vfs_op_descs[i]->vdesc_offset = vfs_opv_numops++;
122 	/*
123 	 * Allocate and fill in the vectors
124 	 */
125 	for (i = 0; i < vnodeopv_num; i++) {
126 		opv = vnodeopv_descs[i];
127 		opv_desc_vector_p = opv->opv_desc_vector_p;
128 		if (*opv_desc_vector_p)
129 			FREE(*opv_desc_vector_p, M_VNODE);
130 		MALLOC(*opv_desc_vector_p, vop_t **,
131 		       vfs_opv_numops * sizeof(vop_t *), M_VNODE, M_WAITOK);
132 		if (*opv_desc_vector_p == NULL)
133 			panic("no memory for vop_t ** vector");
134 		bzero(*opv_desc_vector_p, vfs_opv_numops * sizeof(vop_t *));
135 
136 		/* Fill in, with slot 0 being panic */
137 		opv_desc_vector = *opv_desc_vector_p;
138 		opv_desc_vector[0] = (vop_t *)vop_panic;
139 		for (j = 0; opv->opv_desc_ops[j].opve_op; j++) {
140 			opve_descp = &(opv->opv_desc_ops[j]);
141 			opv_desc_vector[opve_descp->opve_op->vdesc_offset] =
142 				opve_descp->opve_impl;
143 		}
144 
145 		/* Replace unfilled routines with their default (slot 1). */
146 		opv_desc_vector = *(opv->opv_desc_vector_p);
147 		if (opv_desc_vector[1] == NULL)
148 			panic("vfs_opv_recalc: vector without a default.");
149 		for (j = 0; j < vfs_opv_numops; j++)
150 			if (opv_desc_vector[j] == NULL)
151 				opv_desc_vector[j] = opv_desc_vector[1];
152 	}
153 }
154 
155 void
156 vfs_add_vnodeops(const void *data)
157 {
158 	const struct vnodeopv_desc *opv;
159 	const struct vnodeopv_desc **newopv;
160 	struct vnodeop_desc **newop;
161 	int *newref;
162 	vop_t **opv_desc_vector;
163 	struct vnodeop_desc *desc;
164 	int i, j;
165 
166 	opv = (const struct vnodeopv_desc *)data;
167 	MALLOC(newopv, const struct vnodeopv_desc **,
168 	       (vnodeopv_num + 1) * sizeof(*newopv), M_VNODE, M_WAITOK);
169 	if (newopv == NULL)
170 		panic("vfs_add_vnodeops: no memory");
171 	if (vnodeopv_descs) {
172 		bcopy(vnodeopv_descs, newopv, vnodeopv_num * sizeof(*newopv));
173 		FREE(vnodeopv_descs, M_VNODE);
174 	}
175 	newopv[vnodeopv_num] = opv;
176 	vnodeopv_descs = newopv;
177 	vnodeopv_num++;
178 
179 	/* See if we have turned up a new vnode op desc */
180 	opv_desc_vector = *(opv->opv_desc_vector_p);
181 	for (i = 0; (desc = opv->opv_desc_ops[i].opve_op); i++) {
182 		for (j = 0; j < num_op_descs; j++) {
183 			if (desc == vfs_op_descs[j]) {
184 				/* found it, increase reference count */
185 				vfs_op_desc_refs[j]++;
186 				break;
187 			}
188 		}
189 		if (j == num_op_descs) {
190 			/* not found, new entry */
191 			MALLOC(newop, struct vnodeop_desc **,
192 			       (num_op_descs + 1) * sizeof(*newop),
193 			       M_VNODE, M_WAITOK);
194 			if (newop == NULL)
195 				panic("vfs_add_vnodeops: no memory for desc");
196 			/* new reference count (for unload) */
197 			MALLOC(newref, int *,
198 				(num_op_descs + 1) * sizeof(*newref),
199 				M_VNODE, M_WAITOK);
200 			if (newref == NULL)
201 				panic("vfs_add_vnodeops: no memory for refs");
202 			if (vfs_op_descs) {
203 				bcopy(vfs_op_descs, newop,
204 					num_op_descs * sizeof(*newop));
205 				FREE(vfs_op_descs, M_VNODE);
206 			}
207 			if (vfs_op_desc_refs) {
208 				bcopy(vfs_op_desc_refs, newref,
209 					num_op_descs * sizeof(*newref));
210 				FREE(vfs_op_desc_refs, M_VNODE);
211 			}
212 			newop[num_op_descs] = desc;
213 			newref[num_op_descs] = 1;
214 			vfs_op_descs = newop;
215 			vfs_op_desc_refs = newref;
216 			num_op_descs++;
217 		}
218 	}
219 	vfs_opv_recalc();
220 }
221 
222 void
223 vfs_rm_vnodeops(const void *data)
224 {
225 	const struct vnodeopv_desc *opv;
226 	const struct vnodeopv_desc **newopv;
227 	struct vnodeop_desc **newop;
228 	int *newref;
229 	vop_t **opv_desc_vector;
230 	struct vnodeop_desc *desc;
231 	int i, j, k;
232 
233 	opv = (const struct vnodeopv_desc *)data;
234 	/* Lower ref counts on descs in the table and release if zero */
235 	opv_desc_vector = *(opv->opv_desc_vector_p);
236 	for (i = 0; (desc = opv->opv_desc_ops[i].opve_op); i++) {
237 		for (j = 0; j < num_op_descs; j++) {
238 			if (desc == vfs_op_descs[j]) {
239 				/* found it, decrease reference count */
240 				vfs_op_desc_refs[j]--;
241 				break;
242 			}
243 		}
244 		for (j = 0; j < num_op_descs; j++) {
245 			if (vfs_op_desc_refs[j] > 0)
246 				continue;
247 			if (vfs_op_desc_refs[j] < 0)
248 				panic("vfs_remove_vnodeops: negative refcnt");
249 			MALLOC(newop, struct vnodeop_desc **,
250 			       (num_op_descs - 1) * sizeof(*newop),
251 			       M_VNODE, M_WAITOK);
252 			if (newop == NULL)
253 				panic("vfs_remove_vnodeops: no memory for desc");
254 			/* new reference count (for unload) */
255 			MALLOC(newref, int *,
256 				(num_op_descs - 1) * sizeof(*newref),
257 				M_VNODE, M_WAITOK);
258 			if (newref == NULL)
259 				panic("vfs_remove_vnodeops: no memory for refs");
260 			for (k = j; k < (num_op_descs - 1); k++) {
261 				vfs_op_descs[k] = vfs_op_descs[k + 1];
262 				vfs_op_desc_refs[k] = vfs_op_desc_refs[k + 1];
263 			}
264 			bcopy(vfs_op_descs, newop,
265 				(num_op_descs - 1) * sizeof(*newop));
266 			bcopy(vfs_op_desc_refs, newref,
267 				(num_op_descs - 1) * sizeof(*newref));
268 			FREE(vfs_op_descs, M_VNODE);
269 			FREE(vfs_op_desc_refs, M_VNODE);
270 			vfs_op_descs = newop;
271 			vfs_op_desc_refs = newref;
272 			num_op_descs--;
273 		}
274 	}
275 
276 	for (i = 0; i < vnodeopv_num; i++) {
277 		if (vnodeopv_descs[i] == opv) {
278 			for (j = i; j < (vnodeopv_num - 1); j++)
279 				vnodeopv_descs[j] = vnodeopv_descs[j + 1];
280 			break;
281 		}
282 	}
283 	if (i == vnodeopv_num)
284 		panic("vfs_remove_vnodeops: opv not found");
285 	MALLOC(newopv, const struct vnodeopv_desc **,
286 	       (vnodeopv_num - 1) * sizeof(*newopv), M_VNODE, M_WAITOK);
287 	if (newopv == NULL)
288 		panic("vfs_remove_vnodeops: no memory");
289 	bcopy(vnodeopv_descs, newopv, (vnodeopv_num - 1) * sizeof(*newopv));
290 	FREE(vnodeopv_descs, M_VNODE);
291 	vnodeopv_descs = newopv;
292 	vnodeopv_num--;
293 
294 	vfs_opv_recalc();
295 }
296 
297 /*
298  * Routines having to do with the management of the vnode table.
299  */
300 struct vattr va_null;
301 
302 /*
303  * Initialize the vnode structures and initialize each file system type.
304  */
305 /* ARGSUSED*/
306 static void
307 vfsinit(void *dummy)
308 {
309 
310 	namei_zone = zinit("NAMEI", MAXPATHLEN, 0, 0, 2);
311 
312 	/*
313 	 * Initialize the vnode table
314 	 */
315 	vntblinit();
316 	/*
317 	 * Initialize the vnode name cache
318 	 */
319 	nchinit();
320 	/*
321 	 * Initialize each file system type.
322 	 * Vfs type numbers must be distinct from VFS_GENERIC (and VFS_VFSCONF).
323 	 */
324 	vattr_null(&va_null);
325 	maxvfsconf = VFS_GENERIC + 1;
326 }
327 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vfsinit, NULL)
328 
329 int
330 vfs_register(struct vfsconf *vfc)
331 {
332 	struct sysctl_oid *oidp;
333 	struct vfsconf *vfsp;
334 
335 	vfsp = NULL;
336 	if (vfsconf)
337 		for (vfsp = vfsconf; vfsp->vfc_next; vfsp = vfsp->vfc_next)
338 			if (strcmp(vfc->vfc_name, vfsp->vfc_name) == 0)
339 				return EEXIST;
340 
341 	vfc->vfc_typenum = maxvfsconf++;
342 	if (vfsp)
343 		vfsp->vfc_next = vfc;
344 	else
345 		vfsconf = vfc;
346 	vfc->vfc_next = NULL;
347 
348 	/*
349 	 * If this filesystem has a sysctl node under vfs
350 	 * (i.e. vfs.xxfs), then change the oid number of that node to
351 	 * match the filesystem's type number.  This allows user code
352 	 * which uses the type number to read sysctl variables defined
353 	 * by the filesystem to continue working. Since the oids are
354 	 * in a sorted list, we need to make sure the order is
355 	 * preserved by re-registering the oid after modifying its
356 	 * number.
357 	 */
358 	for (oidp = SLIST_FIRST(&sysctl__vfs_children); oidp;
359 	     oidp = SLIST_NEXT(oidp, oid_link))
360 		if (strcmp(oidp->oid_name, vfc->vfc_name) == 0) {
361 			sysctl_unregister_oid(oidp);
362 			oidp->oid_number = vfc->vfc_typenum;
363 			sysctl_register_oid(oidp);
364 		}
365 
366 	/*
367 	 * Call init function for this VFS...
368 	 */
369 	(*(vfc->vfc_vfsops->vfs_init))(vfc);
370 
371 	return 0;
372 }
373 
374 
375 int
376 vfs_unregister(struct vfsconf *vfc)
377 {
378 	struct vfsconf *vfsp, *prev_vfsp;
379 	int error, i, maxtypenum;
380 
381 	i = vfc->vfc_typenum;
382 
383 	prev_vfsp = NULL;
384 	for (vfsp = vfsconf; vfsp;
385 			prev_vfsp = vfsp, vfsp = vfsp->vfc_next) {
386 		if (!strcmp(vfc->vfc_name, vfsp->vfc_name))
387 			break;
388 	}
389 	if (vfsp == NULL)
390 		return EINVAL;
391 	if (vfsp->vfc_refcount)
392 		return EBUSY;
393 	if (vfc->vfc_vfsops->vfs_uninit != NULL) {
394 		error = (*vfc->vfc_vfsops->vfs_uninit)(vfsp);
395 		if (error)
396 			return (error);
397 	}
398 	if (prev_vfsp)
399 		prev_vfsp->vfc_next = vfsp->vfc_next;
400 	else
401 		vfsconf = vfsp->vfc_next;
402 	maxtypenum = VFS_GENERIC;
403 	for (vfsp = vfsconf; vfsp != NULL; vfsp = vfsp->vfc_next)
404 		if (maxtypenum < vfsp->vfc_typenum)
405 			maxtypenum = vfsp->vfc_typenum;
406 	maxvfsconf = maxtypenum + 1;
407 	return 0;
408 }
409 
410 int
411 vfs_modevent(module_t mod, int type, void *data)
412 {
413 	struct vfsconf *vfc;
414 	int error = 0;
415 
416 	vfc = (struct vfsconf *)data;
417 
418 	switch (type) {
419 	case MOD_LOAD:
420 		if (vfc)
421 			error = vfs_register(vfc);
422 		break;
423 
424 	case MOD_UNLOAD:
425 		if (vfc)
426 			error = vfs_unregister(vfc);
427 		break;
428 	default:	/* including MOD_SHUTDOWN */
429 		break;
430 	}
431 	return (error);
432 }
433