Deleted Added
full compact
null_vnops.c (2142) null_vnops.c (2946)
1/*
2 * Copyright (c) 1992, 1993
3 * The Regents of the University of California. All rights reserved.
4 *
5 * This code is derived from software contributed to Berkeley by
6 * John Heidemann of the UCLA Ficus project.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. All advertising materials mentioning features or use of this software
17 * must display the following acknowledgement:
18 * This product includes software developed by the University of
19 * California, Berkeley and its contributors.
20 * 4. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
23 *
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * SUCH DAMAGE.
35 *
36 * @(#)null_vnops.c 8.1 (Berkeley) 6/10/93
37 *
1/*
2 * Copyright (c) 1992, 1993
3 * The Regents of the University of California. All rights reserved.
4 *
5 * This code is derived from software contributed to Berkeley by
6 * John Heidemann of the UCLA Ficus project.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. All advertising materials mentioning features or use of this software
17 * must display the following acknowledgement:
18 * This product includes software developed by the University of
19 * California, Berkeley and its contributors.
20 * 4. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
23 *
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * SUCH DAMAGE.
35 *
36 * @(#)null_vnops.c 8.1 (Berkeley) 6/10/93
37 *
38 * $Id: null_vnops.c,v 1.2 1994/08/02 07:45:06 davidg Exp $
38 * $Id: null_vnops.c,v 1.3 1994/08/20 03:48:51 davidg Exp $
39 */
40
41/*
42 * Null Layer
43 *
44 * (See mount_null(8) for more information.)
45 *
46 * The null layer duplicates a portion of the file system
47 * name space under a new name. In this respect, it is
48 * similar to the loopback file system. It differs from
49 * the loopback fs in two respects: it is implemented using
50 * a stackable layers techniques, and it's "null-node"s stack above
51 * all lower-layer vnodes, not just over directory vnodes.
52 *
53 * The null layer has two purposes. First, it serves as a demonstration
54 * of layering by proving a layer which does nothing. (It actually
55 * does everything the loopback file system does, which is slightly
56 * more than nothing.) Second, the null layer can serve as a prototype
57 * layer. Since it provides all necessary layer framework,
58 * new file system layers can be created very easily be starting
59 * with a null layer.
60 *
61 * The remainder of this man page examines the null layer as a basis
62 * for constructing new layers.
63 *
64 *
65 * INSTANTIATING NEW NULL LAYERS
66 *
67 * New null layers are created with mount_null(8).
68 * Mount_null(8) takes two arguments, the pathname
69 * of the lower vfs (target-pn) and the pathname where the null
70 * layer will appear in the namespace (alias-pn). After
71 * the null layer is put into place, the contents
72 * of target-pn subtree will be aliased under alias-pn.
73 *
74 *
75 * OPERATION OF A NULL LAYER
76 *
77 * The null layer is the minimum file system layer,
78 * simply bypassing all possible operations to the lower layer
79 * for processing there. The majority of its activity centers
80 * on the bypass routine, though which nearly all vnode operations
81 * pass.
82 *
83 * The bypass routine accepts arbitrary vnode operations for
84 * handling by the lower layer. It begins by examing vnode
85 * operation arguments and replacing any null-nodes by their
86 * lower-layer equivlants. It then invokes the operation
87 * on the lower layer. Finally, it replaces the null-nodes
88 * in the arguments and, if a vnode is return by the operation,
89 * stacks a null-node on top of the returned vnode.
90 *
91 * Although bypass handles most operations,
92 * vop_getattr, _inactive, _reclaim, and _print are not bypassed.
93 * Vop_getattr must change the fsid being returned.
94 * Vop_inactive and vop_reclaim are not bypassed so that
95 * they can handle freeing null-layer specific data.
96 * Vop_print is not bypassed to avoid excessive debugging
97 * information.
98 *
99 *
100 * INSTANTIATING VNODE STACKS
101 *
102 * Mounting associates the null layer with a lower layer,
103 * effect stacking two VFSes. Vnode stacks are instead
104 * created on demand as files are accessed.
105 *
106 * The initial mount creates a single vnode stack for the
107 * root of the new null layer. All other vnode stacks
108 * are created as a result of vnode operations on
109 * this or other null vnode stacks.
110 *
111 * New vnode stacks come into existance as a result of
112 * an operation which returns a vnode.
113 * The bypass routine stacks a null-node above the new
114 * vnode before returning it to the caller.
115 *
116 * For example, imagine mounting a null layer with
117 * "mount_null /usr/include /dev/layer/null".
118 * Changing directory to /dev/layer/null will assign
119 * the root null-node (which was created when the null layer was mounted).
120 * Now consider opening "sys". A vop_lookup would be
121 * done on the root null-node. This operation would bypass through
122 * to the lower layer which would return a vnode representing
123 * the UFS "sys". Null_bypass then builds a null-node
124 * aliasing the UFS "sys" and returns this to the caller.
125 * Later operations on the null-node "sys" will repeat this
126 * process when constructing other vnode stacks.
127 *
128 *
129 * CREATING OTHER FILE SYSTEM LAYERS
130 *
131 * One of the easiest ways to construct new file system layers is to make
132 * a copy of the null layer, rename all files and variables, and
133 * then begin modifing the copy. Sed can be used to easily rename
134 * all variables.
135 *
136 * The umap layer is an example of a layer descended from the
137 * null layer.
138 *
139 *
140 * INVOKING OPERATIONS ON LOWER LAYERS
141 *
142 * There are two techniques to invoke operations on a lower layer
143 * when the operation cannot be completely bypassed. Each method
144 * is appropriate in different situations. In both cases,
145 * it is the responsibility of the aliasing layer to make
146 * the operation arguments "correct" for the lower layer
147 * by mapping an vnode arguments to the lower layer.
148 *
149 * The first approach is to call the aliasing layer's bypass routine.
150 * This method is most suitable when you wish to invoke the operation
151 * currently being hanldled on the lower layer. It has the advantage
152 * that the bypass routine already must do argument mapping.
153 * An example of this is null_getattrs in the null layer.
154 *
155 * A second approach is to directly invoked vnode operations on
156 * the lower layer with the VOP_OPERATIONNAME interface.
157 * The advantage of this method is that it is easy to invoke
158 * arbitrary operations on the lower layer. The disadvantage
159 * is that vnodes arguments must be manualy mapped.
160 *
161 */
162
163#include <sys/param.h>
164#include <sys/systm.h>
165#include <sys/proc.h>
166#include <sys/time.h>
167#include <sys/types.h>
168#include <sys/vnode.h>
169#include <sys/mount.h>
170#include <sys/namei.h>
171#include <sys/malloc.h>
172#include <sys/buf.h>
173#include <miscfs/nullfs/null.h>
174
175
176int null_bug_bypass = 0; /* for debugging: enables bypass printf'ing */
177
178/*
179 * This is the 10-Apr-92 bypass routine.
180 * This version has been optimized for speed, throwing away some
181 * safety checks. It should still always work, but it's not as
182 * robust to programmer errors.
183 * Define SAFETY to include some error checking code.
184 *
185 * In general, we map all vnodes going down and unmap them on the way back.
186 * As an exception to this, vnodes can be marked "unmapped" by setting
187 * the Nth bit in operation's vdesc_flags.
188 *
189 * Also, some BSD vnode operations have the side effect of vrele'ing
190 * their arguments. With stacking, the reference counts are held
191 * by the upper node, not the lower one, so we must handle these
192 * side-effects here. This is not of concern in Sun-derived systems
193 * since there are no such side-effects.
194 *
195 * This makes the following assumptions:
196 * - only one returned vpp
197 * - no INOUT vpp's (Sun's vop_open has one of these)
198 * - the vnode operation vector of the first vnode should be used
199 * to determine what implementation of the op should be invoked
200 * - all mapped vnodes are of our vnode-type (NEEDSWORK:
201 * problems on rmdir'ing mount points and renaming?)
202 */
203int
204null_bypass(ap)
205 struct vop_generic_args /* {
206 struct vnodeop_desc *a_desc;
207 <other random data follows, presumably>
208 } */ *ap;
209{
210 register struct vnode **this_vp_p;
211 int error;
212 struct vnode *old_vps[VDESC_MAX_VPS];
213 struct vnode **vps_p[VDESC_MAX_VPS];
214 struct vnode ***vppp;
215 struct vnodeop_desc *descp = ap->a_desc;
216 int reles, i;
217
218 if (null_bug_bypass)
219 printf ("null_bypass: %s\n", descp->vdesc_name);
220
221#ifdef SAFETY
222 /*
223 * We require at least one vp.
224 */
225 if (descp->vdesc_vp_offsets == NULL ||
226 descp->vdesc_vp_offsets[0] == VDESC_NO_OFFSET)
227 panic ("null_bypass: no vp's in map.\n");
228#endif
229
230 /*
231 * Map the vnodes going in.
232 * Later, we'll invoke the operation based on
233 * the first mapped vnode's operation vector.
234 */
235 reles = descp->vdesc_flags;
236 for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) {
237 if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET)
238 break; /* bail out at end of list */
239 vps_p[i] = this_vp_p =
240 VOPARG_OFFSETTO(struct vnode**,descp->vdesc_vp_offsets[i],ap);
241 /*
242 * We're not guaranteed that any but the first vnode
243 * are of our type. Check for and don't map any
244 * that aren't. (We must always map first vp or vclean fails.)
245 */
246 if (i && (*this_vp_p)->v_op != null_vnodeop_p) {
247 old_vps[i] = NULL;
248 } else {
249 old_vps[i] = *this_vp_p;
250 *(vps_p[i]) = NULLVPTOLOWERVP(*this_vp_p);
251 /*
252 * XXX - Several operations have the side effect
253 * of vrele'ing their vp's. We must account for
254 * that. (This should go away in the future.)
255 */
256 if (reles & 1)
257 VREF(*this_vp_p);
258 }
259
260 }
261
262 /*
263 * Call the operation on the lower layer
264 * with the modified argument structure.
265 */
266 error = VCALL(*(vps_p[0]), descp->vdesc_offset, ap);
267
268 /*
269 * Maintain the illusion of call-by-value
270 * by restoring vnodes in the argument structure
271 * to their original value.
272 */
273 reles = descp->vdesc_flags;
274 for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) {
275 if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET)
276 break; /* bail out at end of list */
277 if (old_vps[i]) {
278 *(vps_p[i]) = old_vps[i];
279 if (reles & 1)
280 vrele(*(vps_p[i]));
281 }
282 }
283
284 /*
285 * Map the possible out-going vpp
286 * (Assumes that the lower layer always returns
287 * a VREF'ed vpp unless it gets an error.)
288 */
289 if (descp->vdesc_vpp_offset != VDESC_NO_OFFSET &&
290 !(descp->vdesc_flags & VDESC_NOMAP_VPP) &&
291 !error) {
292 /*
293 * XXX - even though some ops have vpp returned vp's,
294 * several ops actually vrele this before returning.
295 * We must avoid these ops.
296 * (This should go away when these ops are regularized.)
297 */
298 if (descp->vdesc_flags & VDESC_VPP_WILLRELE)
299 goto out;
300 vppp = VOPARG_OFFSETTO(struct vnode***,
301 descp->vdesc_vpp_offset,ap);
302 error = null_node_create(old_vps[0]->v_mount, **vppp, *vppp);
303 }
304
305 out:
306 return (error);
307}
308
309
310/*
311 * We handle getattr only to change the fsid.
312 */
313int
314null_getattr(ap)
315 struct vop_getattr_args /* {
316 struct vnode *a_vp;
317 struct vattr *a_vap;
318 struct ucred *a_cred;
319 struct proc *a_p;
320 } */ *ap;
321{
322 int error;
323 if (error = null_bypass(ap))
324 return (error);
325 /* Requires that arguments be restored. */
326 ap->a_vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0];
327 return (0);
328}
329
330
331int
332null_inactive(ap)
333 struct vop_inactive_args /* {
334 struct vnode *a_vp;
335 } */ *ap;
336{
337 /*
338 * Do nothing (and _don't_ bypass).
339 * Wait to vrele lowervp until reclaim,
340 * so that until then our null_node is in the
341 * cache and reusable.
342 *
343 * NEEDSWORK: Someday, consider inactive'ing
344 * the lowervp and then trying to reactivate it
345 * with capabilities (v_id)
346 * like they do in the name lookup cache code.
347 * That's too much work for now.
348 */
349 return (0);
350}
351
352int
353null_reclaim(ap)
354 struct vop_reclaim_args /* {
355 struct vnode *a_vp;
356 } */ *ap;
357{
358 struct vnode *vp = ap->a_vp;
359 struct null_node *xp = VTONULL(vp);
360 struct vnode *lowervp = xp->null_lowervp;
361
362 /*
363 * Note: in vop_reclaim, vp->v_op == dead_vnodeop_p,
364 * so we can't call VOPs on ourself.
365 */
366 /* After this assignment, this node will not be re-used. */
367 xp->null_lowervp = NULL;
368 remque(xp);
369 FREE(vp->v_data, M_TEMP);
370 vp->v_data = NULL;
371 vrele (lowervp);
372 return (0);
373}
374
375
376int
377null_print(ap)
378 struct vop_print_args /* {
379 struct vnode *a_vp;
380 } */ *ap;
381{
382 register struct vnode *vp = ap->a_vp;
383 printf ("\ttag VT_NULLFS, vp=%x, lowervp=%x\n", vp, NULLVPTOLOWERVP(vp));
384 return (0);
385}
386
387
388/*
389 * XXX - vop_strategy must be hand coded because it has no
390 * vnode in its arguments.
391 * This goes away with a merged VM/buffer cache.
392 */
393int
394null_strategy(ap)
395 struct vop_strategy_args /* {
396 struct buf *a_bp;
397 } */ *ap;
398{
399 struct buf *bp = ap->a_bp;
400 int error;
401 struct vnode *savedvp;
402
403 savedvp = bp->b_vp;
404 bp->b_vp = NULLVPTOLOWERVP(bp->b_vp);
405
406 error = VOP_STRATEGY(bp);
407
408 bp->b_vp = savedvp;
409
410 return (error);
411}
412
413
414/*
415 * XXX - like vop_strategy, vop_bwrite must be hand coded because it has no
416 * vnode in its arguments.
417 * This goes away with a merged VM/buffer cache.
418 */
419int
420null_bwrite(ap)
421 struct vop_bwrite_args /* {
422 struct buf *a_bp;
423 } */ *ap;
424{
425 struct buf *bp = ap->a_bp;
426 int error;
427 struct vnode *savedvp;
428
429 savedvp = bp->b_vp;
430 bp->b_vp = NULLVPTOLOWERVP(bp->b_vp);
431
432 error = VOP_BWRITE(bp);
433
434 bp->b_vp = savedvp;
435
436 return (error);
437}
438
439/*
440 * Global vfs data structures
441 */
442int (**null_vnodeop_p)();
443struct vnodeopv_entry_desc null_vnodeop_entries[] = {
444 { &vop_default_desc, null_bypass },
445
446 { &vop_getattr_desc, null_getattr },
447 { &vop_inactive_desc, null_inactive },
448 { &vop_reclaim_desc, null_reclaim },
449 { &vop_print_desc, null_print },
450
451 { &vop_strategy_desc, null_strategy },
452 { &vop_bwrite_desc, null_bwrite },
453
454 { (struct vnodeop_desc*)NULL, (int(*)())NULL }
455};
456struct vnodeopv_desc null_vnodeop_opv_desc =
457 { &null_vnodeop_p, null_vnodeop_entries };
39 */
40
41/*
42 * Null Layer
43 *
44 * (See mount_null(8) for more information.)
45 *
46 * The null layer duplicates a portion of the file system
47 * name space under a new name. In this respect, it is
48 * similar to the loopback file system. It differs from
49 * the loopback fs in two respects: it is implemented using
50 * a stackable layers techniques, and it's "null-node"s stack above
51 * all lower-layer vnodes, not just over directory vnodes.
52 *
53 * The null layer has two purposes. First, it serves as a demonstration
54 * of layering by proving a layer which does nothing. (It actually
55 * does everything the loopback file system does, which is slightly
56 * more than nothing.) Second, the null layer can serve as a prototype
57 * layer. Since it provides all necessary layer framework,
58 * new file system layers can be created very easily be starting
59 * with a null layer.
60 *
61 * The remainder of this man page examines the null layer as a basis
62 * for constructing new layers.
63 *
64 *
65 * INSTANTIATING NEW NULL LAYERS
66 *
67 * New null layers are created with mount_null(8).
68 * Mount_null(8) takes two arguments, the pathname
69 * of the lower vfs (target-pn) and the pathname where the null
70 * layer will appear in the namespace (alias-pn). After
71 * the null layer is put into place, the contents
72 * of target-pn subtree will be aliased under alias-pn.
73 *
74 *
75 * OPERATION OF A NULL LAYER
76 *
77 * The null layer is the minimum file system layer,
78 * simply bypassing all possible operations to the lower layer
79 * for processing there. The majority of its activity centers
80 * on the bypass routine, though which nearly all vnode operations
81 * pass.
82 *
83 * The bypass routine accepts arbitrary vnode operations for
84 * handling by the lower layer. It begins by examing vnode
85 * operation arguments and replacing any null-nodes by their
86 * lower-layer equivlants. It then invokes the operation
87 * on the lower layer. Finally, it replaces the null-nodes
88 * in the arguments and, if a vnode is return by the operation,
89 * stacks a null-node on top of the returned vnode.
90 *
91 * Although bypass handles most operations,
92 * vop_getattr, _inactive, _reclaim, and _print are not bypassed.
93 * Vop_getattr must change the fsid being returned.
94 * Vop_inactive and vop_reclaim are not bypassed so that
95 * they can handle freeing null-layer specific data.
96 * Vop_print is not bypassed to avoid excessive debugging
97 * information.
98 *
99 *
100 * INSTANTIATING VNODE STACKS
101 *
102 * Mounting associates the null layer with a lower layer,
103 * effect stacking two VFSes. Vnode stacks are instead
104 * created on demand as files are accessed.
105 *
106 * The initial mount creates a single vnode stack for the
107 * root of the new null layer. All other vnode stacks
108 * are created as a result of vnode operations on
109 * this or other null vnode stacks.
110 *
111 * New vnode stacks come into existance as a result of
112 * an operation which returns a vnode.
113 * The bypass routine stacks a null-node above the new
114 * vnode before returning it to the caller.
115 *
116 * For example, imagine mounting a null layer with
117 * "mount_null /usr/include /dev/layer/null".
118 * Changing directory to /dev/layer/null will assign
119 * the root null-node (which was created when the null layer was mounted).
120 * Now consider opening "sys". A vop_lookup would be
121 * done on the root null-node. This operation would bypass through
122 * to the lower layer which would return a vnode representing
123 * the UFS "sys". Null_bypass then builds a null-node
124 * aliasing the UFS "sys" and returns this to the caller.
125 * Later operations on the null-node "sys" will repeat this
126 * process when constructing other vnode stacks.
127 *
128 *
129 * CREATING OTHER FILE SYSTEM LAYERS
130 *
131 * One of the easiest ways to construct new file system layers is to make
132 * a copy of the null layer, rename all files and variables, and
133 * then begin modifing the copy. Sed can be used to easily rename
134 * all variables.
135 *
136 * The umap layer is an example of a layer descended from the
137 * null layer.
138 *
139 *
140 * INVOKING OPERATIONS ON LOWER LAYERS
141 *
142 * There are two techniques to invoke operations on a lower layer
143 * when the operation cannot be completely bypassed. Each method
144 * is appropriate in different situations. In both cases,
145 * it is the responsibility of the aliasing layer to make
146 * the operation arguments "correct" for the lower layer
147 * by mapping an vnode arguments to the lower layer.
148 *
149 * The first approach is to call the aliasing layer's bypass routine.
150 * This method is most suitable when you wish to invoke the operation
151 * currently being hanldled on the lower layer. It has the advantage
152 * that the bypass routine already must do argument mapping.
153 * An example of this is null_getattrs in the null layer.
154 *
155 * A second approach is to directly invoked vnode operations on
156 * the lower layer with the VOP_OPERATIONNAME interface.
157 * The advantage of this method is that it is easy to invoke
158 * arbitrary operations on the lower layer. The disadvantage
159 * is that vnodes arguments must be manualy mapped.
160 *
161 */
162
163#include <sys/param.h>
164#include <sys/systm.h>
165#include <sys/proc.h>
166#include <sys/time.h>
167#include <sys/types.h>
168#include <sys/vnode.h>
169#include <sys/mount.h>
170#include <sys/namei.h>
171#include <sys/malloc.h>
172#include <sys/buf.h>
173#include <miscfs/nullfs/null.h>
174
175
176int null_bug_bypass = 0; /* for debugging: enables bypass printf'ing */
177
178/*
179 * This is the 10-Apr-92 bypass routine.
180 * This version has been optimized for speed, throwing away some
181 * safety checks. It should still always work, but it's not as
182 * robust to programmer errors.
183 * Define SAFETY to include some error checking code.
184 *
185 * In general, we map all vnodes going down and unmap them on the way back.
186 * As an exception to this, vnodes can be marked "unmapped" by setting
187 * the Nth bit in operation's vdesc_flags.
188 *
189 * Also, some BSD vnode operations have the side effect of vrele'ing
190 * their arguments. With stacking, the reference counts are held
191 * by the upper node, not the lower one, so we must handle these
192 * side-effects here. This is not of concern in Sun-derived systems
193 * since there are no such side-effects.
194 *
195 * This makes the following assumptions:
196 * - only one returned vpp
197 * - no INOUT vpp's (Sun's vop_open has one of these)
198 * - the vnode operation vector of the first vnode should be used
199 * to determine what implementation of the op should be invoked
200 * - all mapped vnodes are of our vnode-type (NEEDSWORK:
201 * problems on rmdir'ing mount points and renaming?)
202 */
203int
204null_bypass(ap)
205 struct vop_generic_args /* {
206 struct vnodeop_desc *a_desc;
207 <other random data follows, presumably>
208 } */ *ap;
209{
210 register struct vnode **this_vp_p;
211 int error;
212 struct vnode *old_vps[VDESC_MAX_VPS];
213 struct vnode **vps_p[VDESC_MAX_VPS];
214 struct vnode ***vppp;
215 struct vnodeop_desc *descp = ap->a_desc;
216 int reles, i;
217
218 if (null_bug_bypass)
219 printf ("null_bypass: %s\n", descp->vdesc_name);
220
221#ifdef SAFETY
222 /*
223 * We require at least one vp.
224 */
225 if (descp->vdesc_vp_offsets == NULL ||
226 descp->vdesc_vp_offsets[0] == VDESC_NO_OFFSET)
227 panic ("null_bypass: no vp's in map.\n");
228#endif
229
230 /*
231 * Map the vnodes going in.
232 * Later, we'll invoke the operation based on
233 * the first mapped vnode's operation vector.
234 */
235 reles = descp->vdesc_flags;
236 for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) {
237 if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET)
238 break; /* bail out at end of list */
239 vps_p[i] = this_vp_p =
240 VOPARG_OFFSETTO(struct vnode**,descp->vdesc_vp_offsets[i],ap);
241 /*
242 * We're not guaranteed that any but the first vnode
243 * are of our type. Check for and don't map any
244 * that aren't. (We must always map first vp or vclean fails.)
245 */
246 if (i && (*this_vp_p)->v_op != null_vnodeop_p) {
247 old_vps[i] = NULL;
248 } else {
249 old_vps[i] = *this_vp_p;
250 *(vps_p[i]) = NULLVPTOLOWERVP(*this_vp_p);
251 /*
252 * XXX - Several operations have the side effect
253 * of vrele'ing their vp's. We must account for
254 * that. (This should go away in the future.)
255 */
256 if (reles & 1)
257 VREF(*this_vp_p);
258 }
259
260 }
261
262 /*
263 * Call the operation on the lower layer
264 * with the modified argument structure.
265 */
266 error = VCALL(*(vps_p[0]), descp->vdesc_offset, ap);
267
268 /*
269 * Maintain the illusion of call-by-value
270 * by restoring vnodes in the argument structure
271 * to their original value.
272 */
273 reles = descp->vdesc_flags;
274 for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) {
275 if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET)
276 break; /* bail out at end of list */
277 if (old_vps[i]) {
278 *(vps_p[i]) = old_vps[i];
279 if (reles & 1)
280 vrele(*(vps_p[i]));
281 }
282 }
283
284 /*
285 * Map the possible out-going vpp
286 * (Assumes that the lower layer always returns
287 * a VREF'ed vpp unless it gets an error.)
288 */
289 if (descp->vdesc_vpp_offset != VDESC_NO_OFFSET &&
290 !(descp->vdesc_flags & VDESC_NOMAP_VPP) &&
291 !error) {
292 /*
293 * XXX - even though some ops have vpp returned vp's,
294 * several ops actually vrele this before returning.
295 * We must avoid these ops.
296 * (This should go away when these ops are regularized.)
297 */
298 if (descp->vdesc_flags & VDESC_VPP_WILLRELE)
299 goto out;
300 vppp = VOPARG_OFFSETTO(struct vnode***,
301 descp->vdesc_vpp_offset,ap);
302 error = null_node_create(old_vps[0]->v_mount, **vppp, *vppp);
303 }
304
305 out:
306 return (error);
307}
308
309
310/*
311 * We handle getattr only to change the fsid.
312 */
313int
314null_getattr(ap)
315 struct vop_getattr_args /* {
316 struct vnode *a_vp;
317 struct vattr *a_vap;
318 struct ucred *a_cred;
319 struct proc *a_p;
320 } */ *ap;
321{
322 int error;
323 if (error = null_bypass(ap))
324 return (error);
325 /* Requires that arguments be restored. */
326 ap->a_vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0];
327 return (0);
328}
329
330
331int
332null_inactive(ap)
333 struct vop_inactive_args /* {
334 struct vnode *a_vp;
335 } */ *ap;
336{
337 /*
338 * Do nothing (and _don't_ bypass).
339 * Wait to vrele lowervp until reclaim,
340 * so that until then our null_node is in the
341 * cache and reusable.
342 *
343 * NEEDSWORK: Someday, consider inactive'ing
344 * the lowervp and then trying to reactivate it
345 * with capabilities (v_id)
346 * like they do in the name lookup cache code.
347 * That's too much work for now.
348 */
349 return (0);
350}
351
352int
353null_reclaim(ap)
354 struct vop_reclaim_args /* {
355 struct vnode *a_vp;
356 } */ *ap;
357{
358 struct vnode *vp = ap->a_vp;
359 struct null_node *xp = VTONULL(vp);
360 struct vnode *lowervp = xp->null_lowervp;
361
362 /*
363 * Note: in vop_reclaim, vp->v_op == dead_vnodeop_p,
364 * so we can't call VOPs on ourself.
365 */
366 /* After this assignment, this node will not be re-used. */
367 xp->null_lowervp = NULL;
368 remque(xp);
369 FREE(vp->v_data, M_TEMP);
370 vp->v_data = NULL;
371 vrele (lowervp);
372 return (0);
373}
374
375
376int
377null_print(ap)
378 struct vop_print_args /* {
379 struct vnode *a_vp;
380 } */ *ap;
381{
382 register struct vnode *vp = ap->a_vp;
383 printf ("\ttag VT_NULLFS, vp=%x, lowervp=%x\n", vp, NULLVPTOLOWERVP(vp));
384 return (0);
385}
386
387
388/*
389 * XXX - vop_strategy must be hand coded because it has no
390 * vnode in its arguments.
391 * This goes away with a merged VM/buffer cache.
392 */
393int
394null_strategy(ap)
395 struct vop_strategy_args /* {
396 struct buf *a_bp;
397 } */ *ap;
398{
399 struct buf *bp = ap->a_bp;
400 int error;
401 struct vnode *savedvp;
402
403 savedvp = bp->b_vp;
404 bp->b_vp = NULLVPTOLOWERVP(bp->b_vp);
405
406 error = VOP_STRATEGY(bp);
407
408 bp->b_vp = savedvp;
409
410 return (error);
411}
412
413
414/*
415 * XXX - like vop_strategy, vop_bwrite must be hand coded because it has no
416 * vnode in its arguments.
417 * This goes away with a merged VM/buffer cache.
418 */
419int
420null_bwrite(ap)
421 struct vop_bwrite_args /* {
422 struct buf *a_bp;
423 } */ *ap;
424{
425 struct buf *bp = ap->a_bp;
426 int error;
427 struct vnode *savedvp;
428
429 savedvp = bp->b_vp;
430 bp->b_vp = NULLVPTOLOWERVP(bp->b_vp);
431
432 error = VOP_BWRITE(bp);
433
434 bp->b_vp = savedvp;
435
436 return (error);
437}
438
439/*
440 * Global vfs data structures
441 */
442int (**null_vnodeop_p)();
443struct vnodeopv_entry_desc null_vnodeop_entries[] = {
444 { &vop_default_desc, null_bypass },
445
446 { &vop_getattr_desc, null_getattr },
447 { &vop_inactive_desc, null_inactive },
448 { &vop_reclaim_desc, null_reclaim },
449 { &vop_print_desc, null_print },
450
451 { &vop_strategy_desc, null_strategy },
452 { &vop_bwrite_desc, null_bwrite },
453
454 { (struct vnodeop_desc*)NULL, (int(*)())NULL }
455};
456struct vnodeopv_desc null_vnodeop_opv_desc =
457 { &null_vnodeop_p, null_vnodeop_entries };
458
459VNODEOP_SET(null_vnodeop_opv_desc);