xref: /freebsd/sys/kern/vfs_init.c (revision f9218d3d4fd34f082473b3a021c6d4d109fb47cf)
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 = 64;
94 
95 /* Allow this number to be tuned at boot */
96 TUNABLE_INT("vfs.opv_numops", &vfs_opv_numops);
97 SYSCTL_INT(_vfs, OID_AUTO, opv_numops, CTLFLAG_RD, &vfs_opv_numops,
98 	0, "Maximum number of operations in vop_t vector");
99 
100 static int int_cmp(const void *a, const void *b);
101 
102 static int
103 int_cmp(const void *a, const void *b)
104 {
105 	return(*(const int *)a - *(const int *)b);
106 }
107 
108 /*
109  * Recalculate the operations vector/description (those parts of it that can
110  * be recalculated, that is.)
111  * Always allocate operations vector large enough to hold vfs_opv_numops
112  * entries. The vector is never freed or deallocated once it is initialized,
113  * so that vnodes might safely reference it through their v_op pointer without
114  * vector changing suddenly from under them.
115  */
116 static void
117 vfs_opv_recalc(void)
118 {
119 	int i, j, k;
120 	int *vfs_op_offsets;
121 	vop_t ***opv_desc_vector_p;
122 	vop_t **opv_desc_vector;
123 	struct vnodeopv_entry_desc *opve_descp;
124 	const struct vnodeopv_desc *opv;
125 
126 	if (vfs_op_descs == NULL)
127 		panic("vfs_opv_recalc called with null vfs_op_descs");
128 
129 	/*
130 	 * Allocate and initialize temporary array to store
131 	 * offsets. Sort it to put all uninitialized entries
132 	 * first and to make holes in existing offset sequence
133 	 * detectable.
134 	 */
135 	MALLOC(vfs_op_offsets, int *,
136 		num_op_descs * sizeof(int), M_TEMP, M_WAITOK);
137 	if (vfs_op_offsets == NULL)
138 		panic("vfs_opv_recalc: no memory");
139 	for (i = 0; i < num_op_descs; i++)
140 		vfs_op_offsets[i] = vfs_op_descs[i]->vdesc_offset;
141 	qsort(vfs_op_offsets, num_op_descs, sizeof(int), int_cmp);
142 
143 	/*
144 	 * Run through and make sure all known descs have an offset.
145 	 * Use vfs_op_offsets to locate holes in offset sequence and
146 	 * reuse them.
147 	 * vop_default_desc is hardwired at offset 1, and offset 0
148 	 * is a panic sanity check.
149 	 */
150 	j = 1; k = 1;
151 	for (i = 0; i < num_op_descs; i++) {
152 		if (vfs_op_descs[i]->vdesc_offset != 0)
153 			continue;
154 		/*
155 		 * Look at two adjacent entries vfs_op_offsets[j - 1] and
156 		 * vfs_op_offsets[j] and see if we can fit a new offset
157 		 * number in between. If not, look at the next pair until
158 		 * hole is found or the end of the vfs_op_offsets vector is
159 		 * reached. j has been initialized to 1 above so that
160 		 * referencing (j-1)-th element is safe and the loop will
161 		 * never execute if num_op_descs is 1. For each new value s
162 		 * of i the j loop pick up from where previous iteration has
163 		 * left off. When the last hole has been consumed or if no
164 		 * hole has been found, we will start allocating new numbers
165 		 * starting from the biggest already available offset + 1.
166 		 */
167 		for (; j < num_op_descs; j++) {
168 			if (vfs_op_offsets[j - 1] < k && vfs_op_offsets[j] > k)
169 				break;
170 			k = vfs_op_offsets[j] + 1;
171 		}
172 		vfs_op_descs[i]->vdesc_offset = k++;
173 	}
174 	FREE(vfs_op_offsets, M_TEMP);
175 
176 	/* Panic if new vops will cause vector overflow */
177 	if (k > vfs_opv_numops)
178 		panic("VFS: Ran out of vop_t vector entries. %d entries required, only %d available.\n", k, vfs_opv_numops);
179 
180 	/*
181 	 * Allocate and fill in the vectors
182 	 */
183 	for (i = 0; i < vnodeopv_num; i++) {
184 		opv = vnodeopv_descs[i];
185 		opv_desc_vector_p = opv->opv_desc_vector_p;
186 		if (*opv_desc_vector_p == NULL)
187 			MALLOC(*opv_desc_vector_p, vop_t **,
188 				vfs_opv_numops * sizeof(vop_t *), M_VNODE,
189 				M_WAITOK | M_ZERO);
190 
191 		/* Fill in, with slot 0 being to return EOPNOTSUPP */
192 		opv_desc_vector = *opv_desc_vector_p;
193 		opv_desc_vector[0] = (vop_t *)vop_eopnotsupp;
194 		for (j = 0; opv->opv_desc_ops[j].opve_op; j++) {
195 			opve_descp = &(opv->opv_desc_ops[j]);
196 			opv_desc_vector[opve_descp->opve_op->vdesc_offset] =
197 				opve_descp->opve_impl;
198 		}
199 
200 		/* Replace unfilled routines with their default (slot 1). */
201 		opv_desc_vector = *(opv->opv_desc_vector_p);
202 		if (opv_desc_vector[1] == NULL)
203 			panic("vfs_opv_recalc: vector without a default.");
204 		for (j = 0; j < vfs_opv_numops; j++)
205 			if (opv_desc_vector[j] == NULL)
206 				opv_desc_vector[j] = opv_desc_vector[1];
207 	}
208 }
209 
210 /* Add a set of vnode operations (a description) to the table above. */
211 void
212 vfs_add_vnodeops(const void *data)
213 {
214 	const struct vnodeopv_desc *opv;
215 	const struct vnodeopv_desc **newopv;
216 	struct vnodeop_desc **newop;
217 	int *newref;
218 	vop_t **opv_desc_vector;
219 	struct vnodeop_desc *desc;
220 	int i, j;
221 
222 	opv = (const struct vnodeopv_desc *)data;
223 	MALLOC(newopv, const struct vnodeopv_desc **,
224 	       (vnodeopv_num + 1) * sizeof(*newopv), M_VNODE, M_WAITOK);
225 	if (vnodeopv_descs) {
226 		bcopy(vnodeopv_descs, newopv, vnodeopv_num * sizeof(*newopv));
227 		FREE(vnodeopv_descs, M_VNODE);
228 	}
229 	newopv[vnodeopv_num] = opv;
230 	vnodeopv_descs = newopv;
231 	vnodeopv_num++;
232 
233 	/* See if we have turned up a new vnode op desc */
234 	opv_desc_vector = *(opv->opv_desc_vector_p);
235 	for (i = 0; (desc = opv->opv_desc_ops[i].opve_op); i++) {
236 		for (j = 0; j < num_op_descs; j++) {
237 			if (desc == vfs_op_descs[j]) {
238 				/* found it, increase reference count */
239 				vfs_op_desc_refs[j]++;
240 				break;
241 			}
242 		}
243 		if (j == num_op_descs) {
244 			/* not found, new entry */
245 			MALLOC(newop, struct vnodeop_desc **,
246 			       (num_op_descs + 1) * sizeof(*newop),
247 			       M_VNODE, M_WAITOK);
248 			/* new reference count (for unload) */
249 			MALLOC(newref, int *,
250 				(num_op_descs + 1) * sizeof(*newref),
251 				M_VNODE, M_WAITOK);
252 			if (vfs_op_descs) {
253 				bcopy(vfs_op_descs, newop,
254 					num_op_descs * sizeof(*newop));
255 				FREE(vfs_op_descs, M_VNODE);
256 			}
257 			if (vfs_op_desc_refs) {
258 				bcopy(vfs_op_desc_refs, newref,
259 					num_op_descs * sizeof(*newref));
260 				FREE(vfs_op_desc_refs, M_VNODE);
261 			}
262 			newop[num_op_descs] = desc;
263 			newref[num_op_descs] = 1;
264 			vfs_op_descs = newop;
265 			vfs_op_desc_refs = newref;
266 			num_op_descs++;
267 		}
268 	}
269 	vfs_opv_recalc();
270 }
271 
272 /* Remove a vnode type from the vnode description table above. */
273 void
274 vfs_rm_vnodeops(const void *data)
275 {
276 	const struct vnodeopv_desc *opv;
277 	const struct vnodeopv_desc **newopv;
278 	struct vnodeop_desc **newop;
279 	int *newref;
280 	vop_t **opv_desc_vector;
281 	struct vnodeop_desc *desc;
282 	int i, j, k;
283 
284 	opv = (const struct vnodeopv_desc *)data;
285 	/* Lower ref counts on descs in the table and release if zero */
286 	for (i = 0; (desc = opv->opv_desc_ops[i].opve_op); i++) {
287 		for (j = 0; j < num_op_descs; j++) {
288 			if (desc == vfs_op_descs[j]) {
289 				/* found it, decrease reference count */
290 				vfs_op_desc_refs[j]--;
291 				break;
292 			}
293 		}
294 		for (j = 0; j < num_op_descs; j++) {
295 			if (vfs_op_desc_refs[j] > 0)
296 				continue;
297 			if (vfs_op_desc_refs[j] < 0)
298 				panic("vfs_remove_vnodeops: negative refcnt");
299 			/* Entry is going away - replace it with defaultop */
300 			for (k = 0; k < vnodeopv_num; k++) {
301 				opv_desc_vector =
302 					*(vnodeopv_descs[k]->opv_desc_vector_p);
303 				if (opv_desc_vector != NULL)
304 					opv_desc_vector[desc->vdesc_offset] =
305 						opv_desc_vector[1];
306 			}
307 			MALLOC(newop, struct vnodeop_desc **,
308 			       (num_op_descs - 1) * sizeof(*newop),
309 			       M_VNODE, M_WAITOK);
310 			/* new reference count (for unload) */
311 			MALLOC(newref, int *,
312 				(num_op_descs - 1) * sizeof(*newref),
313 				M_VNODE, M_WAITOK);
314 			for (k = j; k < (num_op_descs - 1); k++) {
315 				vfs_op_descs[k] = vfs_op_descs[k + 1];
316 				vfs_op_desc_refs[k] = vfs_op_desc_refs[k + 1];
317 			}
318 			bcopy(vfs_op_descs, newop,
319 				(num_op_descs - 1) * sizeof(*newop));
320 			bcopy(vfs_op_desc_refs, newref,
321 				(num_op_descs - 1) * sizeof(*newref));
322 			FREE(vfs_op_descs, M_VNODE);
323 			FREE(vfs_op_desc_refs, M_VNODE);
324 			vfs_op_descs = newop;
325 			vfs_op_desc_refs = newref;
326 			num_op_descs--;
327 		}
328 	}
329 
330 	for (i = 0; i < vnodeopv_num; i++) {
331 		if (vnodeopv_descs[i] == opv) {
332 			for (j = i; j < (vnodeopv_num - 1); j++)
333 				vnodeopv_descs[j] = vnodeopv_descs[j + 1];
334 			break;
335 		}
336 	}
337 	if (i == vnodeopv_num)
338 		panic("vfs_remove_vnodeops: opv not found");
339 	opv_desc_vector = *(opv->opv_desc_vector_p);
340 	if (opv_desc_vector != NULL)
341 		FREE(opv_desc_vector, M_VNODE);
342 	MALLOC(newopv, const struct vnodeopv_desc **,
343 	       (vnodeopv_num - 1) * sizeof(*newopv), M_VNODE, M_WAITOK);
344 	bcopy(vnodeopv_descs, newopv, (vnodeopv_num - 1) * sizeof(*newopv));
345 	FREE(vnodeopv_descs, M_VNODE);
346 	vnodeopv_descs = newopv;
347 	vnodeopv_num--;
348 
349 	vfs_opv_recalc();
350 }
351 
352 /*
353  * Routines having to do with the management of the vnode table.
354  */
355 struct vattr va_null;
356 
357 /*
358  * Initialize the vnode structures and initialize each filesystem type.
359  */
360 /* ARGSUSED*/
361 static void
362 vfsinit(void *dummy)
363 {
364 
365 	vattr_null(&va_null);
366 }
367 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vfsinit, NULL)
368 
369 /* Register a new filesystem type in the global table */
370 int
371 vfs_register(struct vfsconf *vfc)
372 {
373 	struct sysctl_oid *oidp;
374 	struct vfsconf *vfsp;
375 
376 	vfsp = NULL;
377 	if (vfsconf)
378 		for (vfsp = vfsconf; vfsp->vfc_next; vfsp = vfsp->vfc_next)
379 			if (strcmp(vfc->vfc_name, vfsp->vfc_name) == 0)
380 				return EEXIST;
381 
382 	vfc->vfc_typenum = maxvfsconf++;
383 	if (vfsp)
384 		vfsp->vfc_next = vfc;
385 	else
386 		vfsconf = vfc;
387 	vfc->vfc_next = NULL;
388 
389 	/*
390 	 * If this filesystem has a sysctl node under vfs
391 	 * (i.e. vfs.xxfs), then change the oid number of that node to
392 	 * match the filesystem's type number.  This allows user code
393 	 * which uses the type number to read sysctl variables defined
394 	 * by the filesystem to continue working. Since the oids are
395 	 * in a sorted list, we need to make sure the order is
396 	 * preserved by re-registering the oid after modifying its
397 	 * number.
398 	 */
399 	SLIST_FOREACH(oidp, &sysctl__vfs_children, oid_link)
400 		if (strcmp(oidp->oid_name, vfc->vfc_name) == 0) {
401 			sysctl_unregister_oid(oidp);
402 			oidp->oid_number = vfc->vfc_typenum;
403 			sysctl_register_oid(oidp);
404 		}
405 
406 	/*
407 	 * Call init function for this VFS...
408 	 */
409 	(*(vfc->vfc_vfsops->vfs_init))(vfc);
410 
411 	return 0;
412 }
413 
414 
415 /* Remove registration of a filesystem type */
416 int
417 vfs_unregister(struct vfsconf *vfc)
418 {
419 	struct vfsconf *vfsp, *prev_vfsp;
420 	int error, i, maxtypenum;
421 
422 	i = vfc->vfc_typenum;
423 
424 	prev_vfsp = NULL;
425 	for (vfsp = vfsconf; vfsp;
426 			prev_vfsp = vfsp, vfsp = vfsp->vfc_next) {
427 		if (!strcmp(vfc->vfc_name, vfsp->vfc_name))
428 			break;
429 	}
430 	if (vfsp == NULL)
431 		return EINVAL;
432 	if (vfsp->vfc_refcount)
433 		return EBUSY;
434 	if (vfc->vfc_vfsops->vfs_uninit != NULL) {
435 		error = (*vfc->vfc_vfsops->vfs_uninit)(vfsp);
436 		if (error)
437 			return (error);
438 	}
439 	if (prev_vfsp)
440 		prev_vfsp->vfc_next = vfsp->vfc_next;
441 	else
442 		vfsconf = vfsp->vfc_next;
443 	maxtypenum = VFS_GENERIC;
444 	for (vfsp = vfsconf; vfsp != NULL; vfsp = vfsp->vfc_next)
445 		if (maxtypenum < vfsp->vfc_typenum)
446 			maxtypenum = vfsp->vfc_typenum;
447 	maxvfsconf = maxtypenum + 1;
448 	return 0;
449 }
450 
451 /*
452  * Standard kernel module handling code for filesystem modules.
453  * Referenced from VFS_SET().
454  */
455 int
456 vfs_modevent(module_t mod, int type, void *data)
457 {
458 	struct vfsconf *vfc;
459 	int error = 0;
460 
461 	vfc = (struct vfsconf *)data;
462 
463 	switch (type) {
464 	case MOD_LOAD:
465 		if (vfc)
466 			error = vfs_register(vfc);
467 		break;
468 
469 	case MOD_UNLOAD:
470 		if (vfc)
471 			error = vfs_unregister(vfc);
472 		break;
473 	default:	/* including MOD_SHUTDOWN */
474 		break;
475 	}
476 	return (error);
477 }
478