xref: /freebsd/sys/kern/vfs_init.c (revision d2387d42b8da231a5b95cbc313825fb2aadf26f6)
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  */
40 
41 #include <sys/cdefs.h>
42 __FBSDID("$FreeBSD$");
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/kernel.h>
47 #include <sys/mount.h>
48 #include <sys/sysctl.h>
49 #include <sys/vnode.h>
50 #include <sys/malloc.h>
51 
52 
53 MALLOC_DEFINE(M_VNODE, "vnodes", "Dynamically allocated vnodes");
54 
55 /*
56  * The highest defined VFS number.
57  */
58 int maxvfsconf = VFS_GENERIC + 1;
59 
60 /*
61  * Single-linked list of configured VFSes.
62  * New entries are added/deleted by vfs_register()/vfs_unregister()
63  */
64 struct vfsconf *vfsconf;
65 
66 /*
67  * vfs_init.c
68  *
69  * Allocate and fill in operations vectors.
70  *
71  * An undocumented feature of this approach to defining operations is that
72  * there can be multiple entries in vfs_opv_descs for the same operations
73  * vector. This allows third parties to extend the set of operations
74  * supported by another layer in a binary compatibile way. For example,
75  * assume that NFS needed to be modified to support Ficus. NFS has an entry
76  * (probably nfs_vnopdeop_decls) declaring all the operations NFS supports by
77  * default. Ficus could add another entry (ficus_nfs_vnodeop_decl_entensions)
78  * listing those new operations Ficus adds to NFS, all without modifying the
79  * NFS code. (Of couse, the OTW NFS protocol still needs to be munged, but
80  * that is a(whole)nother story.) This is a feature.
81  */
82 
83 /* Table of known vnodeop vectors (list of VFS vnode vectors) */
84 static const struct vnodeopv_desc **vnodeopv_descs;
85 static int vnodeopv_num;
86 
87 /* Table of known descs (list of vnode op handlers "vop_access_desc") */
88 static struct vnodeop_desc **vfs_op_descs;
89 /* Reference counts for vfs_op_descs */
90 static int *vfs_op_desc_refs;
91 /* Number of descriptions */
92 static int num_op_descs;
93 /* Number of entries in each description */
94 static int vfs_opv_numops = 64;
95 
96 /* Allow this number to be tuned at boot */
97 TUNABLE_INT("vfs.opv_numops", &vfs_opv_numops);
98 SYSCTL_INT(_vfs, OID_AUTO, opv_numops, CTLFLAG_RDTUN, &vfs_opv_numops,
99 	0, "Maximum number of operations in vop_t vector");
100 
101 static int int_cmp(const void *a, const void *b);
102 
103 static int
104 int_cmp(const void *a, const void *b)
105 {
106 	return(*(const int *)a - *(const int *)b);
107 }
108 
109 /*
110  * Recalculate the operations vector/description (those parts of it that can
111  * be recalculated, that is.)
112  * Always allocate operations vector large enough to hold vfs_opv_numops
113  * entries. The vector is never freed or deallocated once it is initialized,
114  * so that vnodes might safely reference it through their v_op pointer without
115  * vector changing suddenly from under them.
116  */
117 static void
118 vfs_opv_recalc(void)
119 {
120 	int i, j, k;
121 	int *vfs_op_offsets;
122 	vop_t ***opv_desc_vector_p;
123 	vop_t **opv_desc_vector;
124 	struct vnodeopv_entry_desc *opve_descp;
125 	const struct vnodeopv_desc *opv;
126 
127 	if (vfs_op_descs == NULL)
128 		panic("vfs_opv_recalc called with null vfs_op_descs");
129 
130 	/*
131 	 * Allocate and initialize temporary array to store
132 	 * offsets. Sort it to put all uninitialized entries
133 	 * first and to make holes in existing offset sequence
134 	 * detectable.
135 	 */
136 	MALLOC(vfs_op_offsets, int *,
137 		num_op_descs * sizeof(int), M_TEMP, M_WAITOK);
138 	if (vfs_op_offsets == NULL)
139 		panic("vfs_opv_recalc: no memory");
140 	for (i = 0; i < num_op_descs; i++)
141 		vfs_op_offsets[i] = vfs_op_descs[i]->vdesc_offset;
142 	qsort(vfs_op_offsets, num_op_descs, sizeof(int), int_cmp);
143 
144 	/*
145 	 * Run through and make sure all known descs have an offset.
146 	 * Use vfs_op_offsets to locate holes in offset sequence and
147 	 * reuse them.
148 	 * vop_default_desc is hardwired at offset 1, and offset 0
149 	 * is a panic sanity check.
150 	 */
151 	j = 1; k = 1;
152 	for (i = 0; i < num_op_descs; i++) {
153 		if (vfs_op_descs[i]->vdesc_offset != 0)
154 			continue;
155 		/*
156 		 * Look at two adjacent entries vfs_op_offsets[j - 1] and
157 		 * vfs_op_offsets[j] and see if we can fit a new offset
158 		 * number in between. If not, look at the next pair until
159 		 * hole is found or the end of the vfs_op_offsets vector is
160 		 * reached. j has been initialized to 1 above so that
161 		 * referencing (j-1)-th element is safe and the loop will
162 		 * never execute if num_op_descs is 1. For each new value s
163 		 * of i the j loop pick up from where previous iteration has
164 		 * left off. When the last hole has been consumed or if no
165 		 * hole has been found, we will start allocating new numbers
166 		 * starting from the biggest already available offset + 1.
167 		 */
168 		for (; j < num_op_descs; j++) {
169 			if (vfs_op_offsets[j - 1] < k && vfs_op_offsets[j] > k)
170 				break;
171 			k = vfs_op_offsets[j] + 1;
172 		}
173 		vfs_op_descs[i]->vdesc_offset = k++;
174 	}
175 	FREE(vfs_op_offsets, M_TEMP);
176 
177 	/* Panic if new vops will cause vector overflow */
178 	if (k > vfs_opv_numops)
179 		panic("VFS: Ran out of vop_t vector entries. %d entries required, only %d available.\n", k, vfs_opv_numops);
180 
181 	/*
182 	 * Allocate and fill in the vectors
183 	 */
184 	for (i = 0; i < vnodeopv_num; i++) {
185 		opv = vnodeopv_descs[i];
186 		opv_desc_vector_p = opv->opv_desc_vector_p;
187 		if (*opv_desc_vector_p == NULL)
188 			MALLOC(*opv_desc_vector_p, vop_t **,
189 				vfs_opv_numops * sizeof(vop_t *), M_VNODE,
190 				M_WAITOK | M_ZERO);
191 
192 		/* Fill in, with slot 0 being to return EOPNOTSUPP */
193 		opv_desc_vector = *opv_desc_vector_p;
194 		opv_desc_vector[0] = (vop_t *)vop_eopnotsupp;
195 		for (j = 0; opv->opv_desc_ops[j].opve_op; j++) {
196 			opve_descp = &(opv->opv_desc_ops[j]);
197 			opv_desc_vector[opve_descp->opve_op->vdesc_offset] =
198 				opve_descp->opve_impl;
199 		}
200 
201 		/* Replace unfilled routines with their default (slot 1). */
202 		opv_desc_vector = *(opv->opv_desc_vector_p);
203 		if (opv_desc_vector[1] == NULL)
204 			panic("vfs_opv_recalc: vector without a default.");
205 		for (j = 0; j < vfs_opv_numops; j++)
206 			if (opv_desc_vector[j] == NULL)
207 				opv_desc_vector[j] = opv_desc_vector[1];
208 	}
209 }
210 
211 /* Add a set of vnode operations (a description) to the table above. */
212 void
213 vfs_add_vnodeops(const void *data)
214 {
215 	const struct vnodeopv_desc *opv;
216 	const struct vnodeopv_desc **newopv;
217 	struct vnodeop_desc **newop;
218 	int *newref;
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 	for (i = 0; (desc = opv->opv_desc_ops[i].opve_op); i++) {
235 		for (j = 0; j < num_op_descs; j++) {
236 			if (desc == vfs_op_descs[j]) {
237 				/* found it, increase reference count */
238 				vfs_op_desc_refs[j]++;
239 				break;
240 			}
241 		}
242 		if (j == num_op_descs) {
243 			/* not found, new entry */
244 			MALLOC(newop, struct vnodeop_desc **,
245 			       (num_op_descs + 1) * sizeof(*newop),
246 			       M_VNODE, M_WAITOK);
247 			/* new reference count (for unload) */
248 			MALLOC(newref, int *,
249 				(num_op_descs + 1) * sizeof(*newref),
250 				M_VNODE, M_WAITOK);
251 			if (vfs_op_descs) {
252 				bcopy(vfs_op_descs, newop,
253 					num_op_descs * sizeof(*newop));
254 				FREE(vfs_op_descs, M_VNODE);
255 			}
256 			if (vfs_op_desc_refs) {
257 				bcopy(vfs_op_desc_refs, newref,
258 					num_op_descs * sizeof(*newref));
259 				FREE(vfs_op_desc_refs, M_VNODE);
260 			}
261 			newop[num_op_descs] = desc;
262 			newref[num_op_descs] = 1;
263 			vfs_op_descs = newop;
264 			vfs_op_desc_refs = newref;
265 			num_op_descs++;
266 		}
267 	}
268 	vfs_opv_recalc();
269 }
270 
271 /* Remove a vnode type from the vnode description table above. */
272 void
273 vfs_rm_vnodeops(const void *data)
274 {
275 	const struct vnodeopv_desc *opv;
276 	const struct vnodeopv_desc **newopv;
277 	struct vnodeop_desc **newop;
278 	int *newref;
279 	vop_t **opv_desc_vector;
280 	struct vnodeop_desc *desc;
281 	int i, j, k;
282 
283 	opv = (const struct vnodeopv_desc *)data;
284 	/* Lower ref counts on descs in the table and release if zero */
285 	for (i = 0; (desc = opv->opv_desc_ops[i].opve_op); i++) {
286 		for (j = 0; j < num_op_descs; j++) {
287 			if (desc == vfs_op_descs[j]) {
288 				/* found it, decrease reference count */
289 				vfs_op_desc_refs[j]--;
290 				break;
291 			}
292 		}
293 		for (j = 0; j < num_op_descs; j++) {
294 			if (vfs_op_desc_refs[j] > 0)
295 				continue;
296 			if (vfs_op_desc_refs[j] < 0)
297 				panic("vfs_remove_vnodeops: negative refcnt");
298 			/* Entry is going away - replace it with defaultop */
299 			for (k = 0; k < vnodeopv_num; k++) {
300 				opv_desc_vector =
301 					*(vnodeopv_descs[k]->opv_desc_vector_p);
302 				if (opv_desc_vector != NULL)
303 					opv_desc_vector[desc->vdesc_offset] =
304 						opv_desc_vector[1];
305 			}
306 			MALLOC(newop, struct vnodeop_desc **,
307 			       (num_op_descs - 1) * sizeof(*newop),
308 			       M_VNODE, M_WAITOK);
309 			/* new reference count (for unload) */
310 			MALLOC(newref, int *,
311 				(num_op_descs - 1) * sizeof(*newref),
312 				M_VNODE, M_WAITOK);
313 			for (k = j; k < (num_op_descs - 1); k++) {
314 				vfs_op_descs[k] = vfs_op_descs[k + 1];
315 				vfs_op_desc_refs[k] = vfs_op_desc_refs[k + 1];
316 			}
317 			bcopy(vfs_op_descs, newop,
318 				(num_op_descs - 1) * sizeof(*newop));
319 			bcopy(vfs_op_desc_refs, newref,
320 				(num_op_descs - 1) * sizeof(*newref));
321 			FREE(vfs_op_descs, M_VNODE);
322 			FREE(vfs_op_desc_refs, M_VNODE);
323 			vfs_op_descs = newop;
324 			vfs_op_desc_refs = newref;
325 			num_op_descs--;
326 		}
327 	}
328 
329 	for (i = 0; i < vnodeopv_num; i++) {
330 		if (vnodeopv_descs[i] == opv) {
331 			for (j = i; j < (vnodeopv_num - 1); j++)
332 				vnodeopv_descs[j] = vnodeopv_descs[j + 1];
333 			break;
334 		}
335 	}
336 	if (i == vnodeopv_num)
337 		panic("vfs_remove_vnodeops: opv not found");
338 	opv_desc_vector = *(opv->opv_desc_vector_p);
339 	if (opv_desc_vector != NULL)
340 		FREE(opv_desc_vector, M_VNODE);
341 	MALLOC(newopv, const struct vnodeopv_desc **,
342 	       (vnodeopv_num - 1) * sizeof(*newopv), M_VNODE, M_WAITOK);
343 	bcopy(vnodeopv_descs, newopv, (vnodeopv_num - 1) * sizeof(*newopv));
344 	FREE(vnodeopv_descs, M_VNODE);
345 	vnodeopv_descs = newopv;
346 	vnodeopv_num--;
347 
348 	vfs_opv_recalc();
349 }
350 
351 /*
352  * Routines having to do with the management of the vnode table.
353  */
354 struct vattr va_null;
355 
356 /*
357  * Initialize the vnode structures and initialize each filesystem type.
358  */
359 /* ARGSUSED*/
360 static void
361 vfsinit(void *dummy)
362 {
363 
364 	vattr_null(&va_null);
365 }
366 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vfsinit, NULL)
367 
368 /* Register a new filesystem type in the global table */
369 int
370 vfs_register(struct vfsconf *vfc)
371 {
372 	struct sysctl_oid *oidp;
373 	struct vfsconf *vfsp;
374 
375 	struct vfsops *vfsops;
376 
377 	vfsp = NULL;
378 	if (vfsconf)
379 		for (vfsp = vfsconf; vfsp->vfc_next; vfsp = vfsp->vfc_next)
380 			if (strcmp(vfc->vfc_name, vfsp->vfc_name) == 0)
381 				return EEXIST;
382 
383 	vfc->vfc_typenum = maxvfsconf++;
384 	if (vfsp)
385 		vfsp->vfc_next = vfc;
386 	else
387 		vfsconf = vfc;
388 	vfc->vfc_next = NULL;
389 
390 	/*
391 	 * If this filesystem has a sysctl node under vfs
392 	 * (i.e. vfs.xxfs), then change the oid number of that node to
393 	 * match the filesystem's type number.  This allows user code
394 	 * which uses the type number to read sysctl variables defined
395 	 * by the filesystem to continue working. Since the oids are
396 	 * in a sorted list, we need to make sure the order is
397 	 * preserved by re-registering the oid after modifying its
398 	 * number.
399 	 */
400 	SLIST_FOREACH(oidp, &sysctl__vfs_children, oid_link)
401 		if (strcmp(oidp->oid_name, vfc->vfc_name) == 0) {
402 			sysctl_unregister_oid(oidp);
403 			oidp->oid_number = vfc->vfc_typenum;
404 			sysctl_register_oid(oidp);
405 		}
406 
407 	/*
408 	 * Initialise unused ``struct vfsops'' fields, to use
409 	 * the vfs_std*() functions.  Note, we need the mount
410 	 * and unmount operations, at the least.  The check
411 	 * for vfsops available is just a debugging aid.
412 	 */
413 	KASSERT(vfc->vfc_vfsops != NULL,
414 	    ("Filesystem %s has no vfsops", vfc->vfc_name));
415 	/*
416 	 * Check the mount and unmount operations.
417 	 */
418 	vfsops = vfc->vfc_vfsops;
419 	KASSERT(vfsops->vfs_mount != NULL || vfsops->vfs_nmount != NULL,
420 	    ("Filesystem %s has no (n)mount op", vfc->vfc_name));
421 	KASSERT(vfsops->vfs_unmount != NULL,
422 	    ("Filesystem %s has no unmount op", vfc->vfc_name));
423 
424 	if (vfsops->vfs_start == NULL)
425 		/* make a file system operational */
426 		vfsops->vfs_start =	vfs_stdstart;
427 	if (vfsops->vfs_root == NULL)
428 		/* return file system's root vnode */
429 		vfsops->vfs_root =	vfs_stdroot;
430 	if (vfsops->vfs_quotactl == NULL)
431 		/* quota control */
432 		vfsops->vfs_quotactl =	vfs_stdquotactl;
433 	if (vfsops->vfs_statfs == NULL)
434 		/* return file system's status */
435 		vfsops->vfs_statfs =	vfs_stdstatfs;
436 	if (vfsops->vfs_sync == NULL)
437 		/*
438 		 * flush unwritten data (nosync)
439 		 * file systems can use vfs_stdsync
440 		 * explicitly by setting it in the
441 		 * vfsop vector.
442 		 */
443 		vfsops->vfs_sync =	vfs_stdnosync;
444 	if (vfsops->vfs_vget == NULL)
445 		/* convert an inode number to a vnode */
446 		vfsops->vfs_vget =	vfs_stdvget;
447 	if (vfsops->vfs_fhtovp == NULL)
448 		/* turn an NFS file handle into a vnode */
449 		vfsops->vfs_fhtovp =	vfs_stdfhtovp;
450 	if (vfsops->vfs_checkexp == NULL)
451 		/* check if file system is exported */
452 		vfsops->vfs_checkexp =	vfs_stdcheckexp;
453 	if (vfsops->vfs_vptofh == NULL)
454 		/* turn a vnode into an NFS file handle */
455 		vfsops->vfs_vptofh =	vfs_stdvptofh;
456 	if (vfsops->vfs_init == NULL)
457 		/* file system specific initialisation */
458 		vfsops->vfs_init =	vfs_stdinit;
459 	if (vfsops->vfs_uninit == NULL)
460 		/* file system specific uninitialisation */
461 		vfsops->vfs_uninit =	vfs_stduninit;
462 	if (vfsops->vfs_extattrctl == NULL)
463 		/* extended attribute control */
464 		vfsops->vfs_extattrctl = vfs_stdextattrctl;
465 
466 	/*
467 	 * Call init function for this VFS...
468 	 */
469 	(*(vfc->vfc_vfsops->vfs_init))(vfc);
470 
471 	return 0;
472 }
473 
474 
475 /* Remove registration of a filesystem type */
476 int
477 vfs_unregister(struct vfsconf *vfc)
478 {
479 	struct vfsconf *vfsp, *prev_vfsp;
480 	int error, i, maxtypenum;
481 
482 	i = vfc->vfc_typenum;
483 
484 	prev_vfsp = NULL;
485 	for (vfsp = vfsconf; vfsp;
486 			prev_vfsp = vfsp, vfsp = vfsp->vfc_next) {
487 		if (!strcmp(vfc->vfc_name, vfsp->vfc_name))
488 			break;
489 	}
490 	if (vfsp == NULL)
491 		return EINVAL;
492 	if (vfsp->vfc_refcount)
493 		return EBUSY;
494 	if (vfc->vfc_vfsops->vfs_uninit != NULL) {
495 		error = (*vfc->vfc_vfsops->vfs_uninit)(vfsp);
496 		if (error)
497 			return (error);
498 	}
499 	if (prev_vfsp)
500 		prev_vfsp->vfc_next = vfsp->vfc_next;
501 	else
502 		vfsconf = vfsp->vfc_next;
503 	maxtypenum = VFS_GENERIC;
504 	for (vfsp = vfsconf; vfsp != NULL; vfsp = vfsp->vfc_next)
505 		if (maxtypenum < vfsp->vfc_typenum)
506 			maxtypenum = vfsp->vfc_typenum;
507 	maxvfsconf = maxtypenum + 1;
508 	return 0;
509 }
510 
511 /*
512  * Standard kernel module handling code for filesystem modules.
513  * Referenced from VFS_SET().
514  */
515 int
516 vfs_modevent(module_t mod, int type, void *data)
517 {
518 	struct vfsconf *vfc;
519 	int error = 0;
520 
521 	vfc = (struct vfsconf *)data;
522 
523 	switch (type) {
524 	case MOD_LOAD:
525 		if (vfc)
526 			error = vfs_register(vfc);
527 		break;
528 
529 	case MOD_UNLOAD:
530 		if (vfc)
531 			error = vfs_unregister(vfc);
532 		break;
533 	default:	/* including MOD_SHUTDOWN */
534 		break;
535 	}
536 	return (error);
537 }
538