1 /* $NetBSD: layer_vnops.c,v 1.72 2021/10/20 03:08:18 thorpej Exp $ */
2
3 /*
4 * Copyright (c) 1999 National Aeronautics & Space Administration
5 * All rights reserved.
6 *
7 * This software was written by William Studenmund of the
8 * Numerical Aerospace Simulation Facility, NASA Ames Research Center.
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. Neither the name of the National Aeronautics & Space Administration
19 * nor the names of its contributors may be used to endorse or promote
20 * products derived from this software without specific prior written
21 * permission.
22 *
23 * THIS SOFTWARE IS PROVIDED BY THE NATIONAL AERONAUTICS & SPACE ADMINISTRATION
24 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
25 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
26 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE ADMINISTRATION OR CONTRIB-
27 * UTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY,
28 * OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
29 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
30 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
31 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
32 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
33 * POSSIBILITY OF SUCH DAMAGE.
34 */
35
36 /*
37 * Copyright (c) 1992, 1993
38 * The Regents of the University of California. All rights reserved.
39 *
40 * This code is derived from software contributed to Berkeley by
41 * John Heidemann of the UCLA Ficus project.
42 *
43 * Redistribution and use in source and binary forms, with or without
44 * modification, are permitted provided that the following conditions
45 * are met:
46 * 1. Redistributions of source code must retain the above copyright
47 * notice, this list of conditions and the following disclaimer.
48 * 2. Redistributions in binary form must reproduce the above copyright
49 * notice, this list of conditions and the following disclaimer in the
50 * documentation and/or other materials provided with the distribution.
51 * 3. Neither the name of the University nor the names of its contributors
52 * may be used to endorse or promote products derived from this software
53 * without specific prior written permission.
54 *
55 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
56 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
57 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
58 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
59 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
60 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
61 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
62 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
63 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
64 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
65 * SUCH DAMAGE.
66 *
67 * @(#)null_vnops.c 8.6 (Berkeley) 5/27/95
68 *
69 * Ancestors:
70 * @(#)lofs_vnops.c 1.2 (Berkeley) 6/18/92
71 * Id: lofs_vnops.c,v 1.11 1992/05/30 10:05:43 jsp Exp jsp
72 * ...and...
73 * @(#)null_vnodeops.c 1.20 92/07/07 UCLA Ficus project
74 */
75
76 /*
77 * Generic layer vnode operations.
78 *
79 * The layer.h, layer_extern.h, layer_vfs.c, and layer_vnops.c files provide
80 * the core implementation of stacked file-systems.
81 *
82 * The layerfs duplicates a portion of the file system name space under
83 * a new name. In this respect, it is similar to the loopback file system.
84 * It differs from the loopback fs in two respects: it is implemented using
85 * a stackable layers technique, and it is "layerfs-nodes" stack above all
86 * lower-layer vnodes, not just over directory vnodes.
87 *
88 * OPERATION OF LAYERFS
89 *
90 * The layerfs is the minimum file system layer, bypassing all possible
91 * operations to the lower layer for processing there. The majority of its
92 * activity centers on the bypass routine, through which nearly all vnode
93 * operations pass.
94 *
95 * The bypass routine accepts arbitrary vnode operations for handling by
96 * the lower layer. It begins by examining vnode operation arguments and
97 * replacing any layered nodes by their lower-layer equivalents. It then
98 * invokes an operation on the lower layer. Finally, it replaces the
99 * layered nodes in the arguments and, if a vnode is returned by the
100 * operation, stacks a layered node on top of the returned vnode.
101 *
102 * The bypass routine in this file, layer_bypass(), is suitable for use
103 * by many different layered filesystems. It can be used by multiple
104 * filesystems simultaneously. Alternatively, a layered fs may provide
105 * its own bypass routine, in which case layer_bypass() should be used as
106 * a model. For instance, the main functionality provided by umapfs, the user
107 * identity mapping file system, is handled by a custom bypass routine.
108 *
109 * Typically a layered fs registers its selected bypass routine as the
110 * default vnode operation in its vnodeopv_entry_desc table. Additionally
111 * the filesystem must store the bypass entry point in the layerm_bypass
112 * field of struct layer_mount. All other layer routines in this file will
113 * use the layerm_bypass() routine.
114 *
115 * Although the bypass routine handles most operations outright, a number
116 * of operations are special cased and handled by the layerfs. For instance,
117 * layer_getattr() must change the fsid being returned. While layer_lock()
118 * and layer_unlock() must handle any locking for the current vnode as well
119 * as pass the lock request down. layer_inactive() and layer_reclaim() are
120 * not bypassed so that they can handle freeing layerfs-specific data. Also,
121 * certain vnode operations (create, mknod, remove, link, rename, mkdir,
122 * rmdir, and symlink) change the locking state within the operation. Ideally
123 * these operations should not change the lock state, but should be changed
124 * to let the caller of the function unlock them. Otherwise, all intermediate
125 * vnode layers (such as union, umapfs, etc) must catch these functions to do
126 * the necessary locking at their layer.
127 *
128 * INSTANTIATING VNODE STACKS
129 *
130 * Mounting associates "layerfs-nodes" stack and lower layer, in effect
131 * stacking two VFSes. The initial mount creates a single vnode stack for
132 * the root of the new layerfs. All other vnode stacks are created as a
133 * result of vnode operations on this or other layerfs vnode stacks.
134 *
135 * New vnode stacks come into existence as a result of an operation which
136 * returns a vnode. The bypass routine stacks a layerfs-node above the new
137 * vnode before returning it to the caller.
138 *
139 * For example, imagine mounting a null layer with:
140 *
141 * "mount_null /usr/include /dev/layer/null"
142 *
143 * Changing directory to /dev/layer/null will assign the root layerfs-node,
144 * which was created when the null layer was mounted). Now consider opening
145 * "sys". A layer_lookup() would be performed on the root layerfs-node.
146 * This operation would bypass through to the lower layer which would return
147 * a vnode representing the UFS "sys". Then, layer_bypass() builds a
148 * layerfs-node aliasing the UFS "sys" and returns this to the caller.
149 * Later operations on the layerfs-node "sys" will repeat this process when
150 * constructing other vnode stacks.
151 *
152 * INVOKING OPERATIONS ON LOWER LAYERS
153 *
154 * There are two techniques to invoke operations on a lower layer when the
155 * operation cannot be completely bypassed. Each method is appropriate in
156 * different situations. In both cases, it is the responsibility of the
157 * aliasing layer to make the operation arguments "correct" for the lower
158 * layer by mapping any vnode arguments to the lower layer.
159 *
160 * The first approach is to call the aliasing layer's bypass routine. This
161 * method is most suitable when you wish to invoke the operation currently
162 * being handled on the lower layer. It has the advantage that the bypass
163 * routine already must do argument mapping. An example of this is
164 * layer_getattr().
165 *
166 * A second approach is to directly invoke vnode operations on the lower
167 * layer with the VOP_OPERATIONNAME interface. The advantage of this method
168 * is that it is easy to invoke arbitrary operations on the lower layer.
169 * The disadvantage is that vnode's arguments must be manually mapped.
170 */
171
172 #include <sys/cdefs.h>
173 __KERNEL_RCSID(0, "$NetBSD: layer_vnops.c,v 1.72 2021/10/20 03:08:18 thorpej Exp $");
174
175 #include <sys/param.h>
176 #include <sys/systm.h>
177 #include <sys/proc.h>
178 #include <sys/time.h>
179 #include <sys/vnode.h>
180 #include <sys/mount.h>
181 #include <sys/namei.h>
182 #include <sys/kmem.h>
183 #include <sys/buf.h>
184 #include <sys/kauth.h>
185 #include <sys/fcntl.h>
186 #include <sys/fstrans.h>
187
188 #include <miscfs/genfs/layer.h>
189 #include <miscfs/genfs/layer_extern.h>
190 #include <miscfs/genfs/genfs.h>
191 #include <miscfs/specfs/specdev.h>
192
193 /*
194 * This is the 08-June-99 bypass routine, based on the 10-Apr-92 bypass
195 * routine by John Heidemann.
196 * The new element for this version is that the whole nullfs
197 * system gained the concept of locks on the lower node.
198 * The 10-Apr-92 version was optimized for speed, throwing away some
199 * safety checks. It should still always work, but it's not as
200 * robust to programmer errors.
201 *
202 * In general, we map all vnodes going down and unmap them on the way back.
203 *
204 * Also, some BSD vnode operations have the side effect of vrele'ing
205 * their arguments. With stacking, the reference counts are held
206 * by the upper node, not the lower one, so we must handle these
207 * side-effects here. This is not of concern in Sun-derived systems
208 * since there are no such side-effects.
209 *
210 * New for the 08-June-99 version: we also handle operations which unlock
211 * the passed-in node (typically they vput the node).
212 *
213 * This makes the following assumptions:
214 * - only one returned vpp
215 * - no INOUT vpp's (Sun's vop_open has one of these)
216 * - the vnode operation vector of the first vnode should be used
217 * to determine what implementation of the op should be invoked
218 * - all mapped vnodes are of our vnode-type (NEEDSWORK:
219 * problems on rmdir'ing mount points and renaming?)
220 */
221 int
layer_bypass(void * v)222 layer_bypass(void *v)
223 {
224 struct vop_generic_args /* {
225 struct vnodeop_desc *a_desc;
226 <other random data follows, presumably>
227 } */ *ap = v;
228 int (**our_vnodeop_p)(void *);
229 struct vnode **this_vp_p;
230 int error;
231 struct vnode *old_vps[VDESC_MAX_VPS], *vp0;
232 struct vnode **vps_p[VDESC_MAX_VPS];
233 struct vnode ***vppp;
234 struct mount *mp;
235 struct vnodeop_desc *descp = ap->a_desc;
236 int reles, i, flags;
237
238 #ifdef DIAGNOSTIC
239 /*
240 * We require at least one vp.
241 */
242 if (descp->vdesc_vp_offsets == NULL ||
243 descp->vdesc_vp_offsets[0] == VDESC_NO_OFFSET)
244 panic("%s: no vp's in map.\n", __func__);
245 #endif
246
247 vps_p[0] =
248 VOPARG_OFFSETTO(struct vnode**, descp->vdesc_vp_offsets[0], ap);
249 vp0 = *vps_p[0];
250 mp = vp0->v_mount;
251 flags = MOUNTTOLAYERMOUNT(mp)->layerm_flags;
252 our_vnodeop_p = vp0->v_op;
253
254 if (flags & LAYERFS_MBYPASSDEBUG)
255 printf("%s: %s\n", __func__, descp->vdesc_name);
256
257 /*
258 * Map the vnodes going in.
259 * Later, we'll invoke the operation based on
260 * the first mapped vnode's operation vector.
261 */
262 reles = descp->vdesc_flags;
263 for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) {
264 if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET)
265 break; /* bail out at end of list */
266 vps_p[i] = this_vp_p =
267 VOPARG_OFFSETTO(struct vnode**, descp->vdesc_vp_offsets[i],
268 ap);
269 /*
270 * We're not guaranteed that any but the first vnode
271 * are of our type. Check for and don't map any
272 * that aren't. (We must always map first vp or vclean fails.)
273 */
274 if (i && (*this_vp_p == NULL ||
275 (*this_vp_p)->v_op != our_vnodeop_p)) {
276 old_vps[i] = NULL;
277 } else {
278 old_vps[i] = *this_vp_p;
279 *(vps_p[i]) = LAYERVPTOLOWERVP(*this_vp_p);
280 /*
281 * XXX - Several operations have the side effect
282 * of vrele'ing their vp's. We must account for
283 * that. (This should go away in the future.)
284 */
285 if (reles & VDESC_VP0_WILLRELE)
286 vref(*this_vp_p);
287 }
288 }
289
290 /*
291 * Call the operation on the lower layer
292 * with the modified argument structure.
293 */
294 error = VCALL(*vps_p[0], descp->vdesc_offset, ap);
295
296 /*
297 * Maintain the illusion of call-by-value
298 * by restoring vnodes in the argument structure
299 * to their original value.
300 */
301 reles = descp->vdesc_flags;
302 for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) {
303 if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET)
304 break; /* bail out at end of list */
305 if (old_vps[i]) {
306 *(vps_p[i]) = old_vps[i];
307 if (reles & VDESC_VP0_WILLRELE)
308 vrele(*(vps_p[i]));
309 }
310 }
311
312 /*
313 * Map the possible out-going vpp
314 * (Assumes that the lower layer always returns
315 * a VREF'ed vpp unless it gets an error.)
316 */
317 if (descp->vdesc_vpp_offset != VDESC_NO_OFFSET && !error) {
318 vppp = VOPARG_OFFSETTO(struct vnode***,
319 descp->vdesc_vpp_offset, ap);
320 /*
321 * Only vop_lookup, vop_create, vop_makedir, vop_mknod
322 * and vop_symlink return vpp's. vop_lookup doesn't call bypass
323 * as a lookup on "." would generate a locking error.
324 * So all the calls which get us here have a unlocked vpp. :-)
325 */
326 error = layer_node_create(mp, **vppp, *vppp);
327 if (error) {
328 vrele(**vppp);
329 **vppp = NULL;
330 }
331 }
332 return error;
333 }
334
335 /*
336 * We have to carry on the locking protocol on the layer vnodes
337 * as we progress through the tree. We also have to enforce read-only
338 * if this layer is mounted read-only.
339 */
340 int
layer_lookup(void * v)341 layer_lookup(void *v)
342 {
343 struct vop_lookup_v2_args /* {
344 struct vnodeop_desc *a_desc;
345 struct vnode * a_dvp;
346 struct vnode ** a_vpp;
347 struct componentname * a_cnp;
348 } */ *ap = v;
349 struct componentname *cnp = ap->a_cnp;
350 struct vnode *dvp, *lvp, *ldvp;
351 int error, flags = cnp->cn_flags;
352
353 dvp = ap->a_dvp;
354
355 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
356 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
357 *ap->a_vpp = NULL;
358 return EROFS;
359 }
360
361 ldvp = LAYERVPTOLOWERVP(dvp);
362 ap->a_dvp = ldvp;
363 error = VCALL(ldvp, ap->a_desc->vdesc_offset, ap);
364 lvp = *ap->a_vpp;
365 *ap->a_vpp = NULL;
366
367 if (error == EJUSTRETURN && (flags & ISLASTCN) &&
368 (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
369 (cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME))
370 error = EROFS;
371
372 /*
373 * We must do the same locking and unlocking at this layer as
374 * is done in the layers below us.
375 */
376 if (ldvp == lvp) {
377 /*
378 * Got the same object back, because we looked up ".",
379 * or ".." in the root node of a mount point.
380 * So we make another reference to dvp and return it.
381 */
382 vref(dvp);
383 *ap->a_vpp = dvp;
384 vrele(lvp);
385 } else if (lvp != NULL) {
386 /* Note: dvp and ldvp are both locked. */
387 KASSERT(error != ENOLCK);
388 error = layer_node_create(dvp->v_mount, lvp, ap->a_vpp);
389 if (error) {
390 vrele(lvp);
391 }
392 }
393 return error;
394 }
395
396 /*
397 * Setattr call. Disallow write attempts if the layer is mounted read-only.
398 */
399 int
layer_setattr(void * v)400 layer_setattr(void *v)
401 {
402 struct vop_setattr_args /* {
403 struct vnodeop_desc *a_desc;
404 struct vnode *a_vp;
405 struct vattr *a_vap;
406 kauth_cred_t a_cred;
407 struct lwp *a_l;
408 } */ *ap = v;
409 struct vnode *vp = ap->a_vp;
410 struct vattr *vap = ap->a_vap;
411
412 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
413 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
414 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
415 (vp->v_mount->mnt_flag & MNT_RDONLY))
416 return EROFS;
417 if (vap->va_size != VNOVAL) {
418 switch (vp->v_type) {
419 case VDIR:
420 return EISDIR;
421 case VCHR:
422 case VBLK:
423 case VSOCK:
424 case VFIFO:
425 return 0;
426 case VREG:
427 case VLNK:
428 default:
429 /*
430 * Disallow write attempts if the filesystem is
431 * mounted read-only.
432 */
433 if (vp->v_mount->mnt_flag & MNT_RDONLY)
434 return EROFS;
435 }
436 }
437 return LAYERFS_DO_BYPASS(vp, ap);
438 }
439
440 /*
441 * We handle getattr only to change the fsid.
442 */
443 int
layer_getattr(void * v)444 layer_getattr(void *v)
445 {
446 struct vop_getattr_args /* {
447 struct vnode *a_vp;
448 struct vattr *a_vap;
449 kauth_cred_t a_cred;
450 struct lwp *a_l;
451 } */ *ap = v;
452 struct vnode *vp = ap->a_vp;
453 int error;
454
455 error = LAYERFS_DO_BYPASS(vp, ap);
456 if (error) {
457 return error;
458 }
459 /* Requires that arguments be restored. */
460 ap->a_vap->va_fsid = vp->v_mount->mnt_stat.f_fsidx.__fsid_val[0];
461 return 0;
462 }
463
464 int
layer_access(void * v)465 layer_access(void *v)
466 {
467 struct vop_access_args /* {
468 struct vnode *a_vp;
469 accmode_t a_accmode;
470 kauth_cred_t a_cred;
471 struct lwp *a_l;
472 } */ *ap = v;
473 struct vnode *vp = ap->a_vp;
474 accmode_t accmode = ap->a_accmode;
475
476 /*
477 * Disallow write attempts on read-only layers;
478 * unless the file is a socket, fifo, or a block or
479 * character device resident on the file system.
480 */
481 if (accmode & VWRITE) {
482 switch (vp->v_type) {
483 case VDIR:
484 case VLNK:
485 case VREG:
486 if (vp->v_mount->mnt_flag & MNT_RDONLY)
487 return EROFS;
488 break;
489 default:
490 break;
491 }
492 }
493 return LAYERFS_DO_BYPASS(vp, ap);
494 }
495
496 /*
497 * We must handle open to be able to catch MNT_NODEV and friends
498 * and increment the lower v_writecount.
499 */
500 int
layer_open(void * v)501 layer_open(void *v)
502 {
503 struct vop_open_args /* {
504 const struct vnodeop_desc *a_desc;
505 struct vnode *a_vp;
506 int a_mode;
507 kauth_cred_t a_cred;
508 } */ *ap = v;
509 struct vnode *vp = ap->a_vp;
510 struct vnode *lvp = LAYERVPTOLOWERVP(vp);
511 int error;
512
513 if (((lvp->v_type == VBLK) || (lvp->v_type == VCHR)) &&
514 (vp->v_mount->mnt_flag & MNT_NODEV))
515 return ENXIO;
516
517 error = LAYERFS_DO_BYPASS(vp, ap);
518 if (error == 0 && (ap->a_mode & FWRITE)) {
519 mutex_enter(lvp->v_interlock);
520 lvp->v_writecount++;
521 mutex_exit(lvp->v_interlock);
522 }
523 return error;
524 }
525
526 /*
527 * We must handle close to decrement the lower v_writecount.
528 */
529 int
layer_close(void * v)530 layer_close(void *v)
531 {
532 struct vop_close_args /* {
533 const struct vnodeop_desc *a_desc;
534 struct vnode *a_vp;
535 int a_fflag;
536 kauth_cred_t a_cred;
537 } */ *ap = v;
538 struct vnode *vp = ap->a_vp;
539 struct vnode *lvp = LAYERVPTOLOWERVP(vp);
540
541 if ((ap->a_fflag & FWRITE)) {
542 mutex_enter(lvp->v_interlock);
543 KASSERT(lvp->v_writecount > 0);
544 lvp->v_writecount--;
545 mutex_exit(lvp->v_interlock);
546 }
547 return LAYERFS_DO_BYPASS(vp, ap);
548 }
549
550 /*
551 * If vinvalbuf is calling us, it's a "shallow fsync" -- don't bother
552 * syncing the underlying vnodes, since they'll be fsync'ed when
553 * reclaimed; otherwise, pass it through to the underlying layer.
554 *
555 * XXX Do we still need to worry about shallow fsync?
556 */
557 int
layer_fsync(void * v)558 layer_fsync(void *v)
559 {
560 struct vop_fsync_args /* {
561 struct vnode *a_vp;
562 kauth_cred_t a_cred;
563 int a_flags;
564 off_t offlo;
565 off_t offhi;
566 struct lwp *a_l;
567 } */ *ap = v;
568 int error;
569
570 if (ap->a_flags & FSYNC_RECLAIM) {
571 return 0;
572 }
573 if (ap->a_vp->v_type == VBLK || ap->a_vp->v_type == VCHR) {
574 error = spec_fsync(v);
575 if (error)
576 return error;
577 }
578 return LAYERFS_DO_BYPASS(ap->a_vp, ap);
579 }
580
581 int
layer_inactive(void * v)582 layer_inactive(void *v)
583 {
584 struct vop_inactive_v2_args /* {
585 struct vnode *a_vp;
586 bool *a_recycle;
587 } */ *ap = v;
588 struct vnode *vp = ap->a_vp;
589
590 /*
591 * If we did a remove, don't cache the node.
592 */
593 *ap->a_recycle = ((VTOLAYER(vp)->layer_flags & LAYERFS_REMOVED) != 0);
594
595 /*
596 * Do nothing (and _don't_ bypass).
597 * Wait to vrele lowervp until reclaim,
598 * so that until then our layer_node is in the
599 * cache and reusable.
600 *
601 * NEEDSWORK: Someday, consider inactive'ing
602 * the lowervp and then trying to reactivate it
603 * with capabilities (v_id)
604 * like they do in the name lookup cache code.
605 * That's too much work for now.
606 */
607
608 return 0;
609 }
610
611 int
layer_remove(void * v)612 layer_remove(void *v)
613 {
614 struct vop_remove_v3_args /* {
615 struct vnode *a_dvp;
616 struct vnode *a_vp;
617 struct componentname *a_cnp;
618 nlink_t ctx_vp_new_nlink;
619 } */ *ap = v;
620 struct vnode *vp = ap->a_vp;
621 int error;
622
623 vref(vp);
624 error = LAYERFS_DO_BYPASS(vp, ap);
625 if (error == 0) {
626 VTOLAYER(vp)->layer_flags |= LAYERFS_REMOVED;
627 }
628 vrele(vp);
629
630 return error;
631 }
632
633 int
layer_rename(void * v)634 layer_rename(void *v)
635 {
636 struct vop_rename_args /* {
637 struct vnode *a_fdvp;
638 struct vnode *a_fvp;
639 struct componentname *a_fcnp;
640 struct vnode *a_tdvp;
641 struct vnode *a_tvp;
642 struct componentname *a_tcnp;
643 } */ *ap = v;
644 struct vnode *fdvp = ap->a_fdvp, *tvp;
645 int error;
646
647 tvp = ap->a_tvp;
648 if (tvp) {
649 if (tvp->v_mount != fdvp->v_mount)
650 tvp = NULL;
651 else
652 vref(tvp);
653 }
654 error = LAYERFS_DO_BYPASS(fdvp, ap);
655 if (tvp) {
656 if (error == 0)
657 VTOLAYER(tvp)->layer_flags |= LAYERFS_REMOVED;
658 vrele(tvp);
659 }
660 return error;
661 }
662
663 int
layer_rmdir(void * v)664 layer_rmdir(void *v)
665 {
666 struct vop_rmdir_v2_args /* {
667 struct vnode *a_dvp;
668 struct vnode *a_vp;
669 struct componentname *a_cnp;
670 } */ *ap = v;
671 int error;
672 struct vnode *vp = ap->a_vp;
673
674 vref(vp);
675 error = LAYERFS_DO_BYPASS(vp, ap);
676 if (error == 0) {
677 VTOLAYER(vp)->layer_flags |= LAYERFS_REMOVED;
678 }
679 vrele(vp);
680
681 return error;
682 }
683
684 int
layer_revoke(void * v)685 layer_revoke(void *v)
686 {
687 struct vop_revoke_args /* {
688 struct vnode *a_vp;
689 int a_flags;
690 } */ *ap = v;
691 struct vnode *vp = ap->a_vp;
692 struct vnode *lvp = LAYERVPTOLOWERVP(vp);
693 int error;
694
695 /*
696 * We will most likely end up in vclean which uses the usecount
697 * to determine if a vnode is active. Take an extra reference on
698 * the lower vnode so it will always close and inactivate.
699 */
700 vref(lvp);
701 error = LAYERFS_DO_BYPASS(vp, ap);
702 vrele(lvp);
703
704 return error;
705 }
706
707 int
layer_reclaim(void * v)708 layer_reclaim(void *v)
709 {
710 struct vop_reclaim_v2_args /* {
711 struct vnode *a_vp;
712 struct lwp *a_l;
713 } */ *ap = v;
714 struct vnode *vp = ap->a_vp;
715 struct layer_mount *lmp = MOUNTTOLAYERMOUNT(vp->v_mount);
716 struct layer_node *xp = VTOLAYER(vp);
717 struct vnode *lowervp = xp->layer_lowervp;
718
719 VOP_UNLOCK(vp);
720
721 /*
722 * Note: in vop_reclaim, the node's struct lock has been
723 * decomissioned, so we have to be careful about calling
724 * VOP's on ourself. We must be careful as VXLOCK is set.
725 */
726 if (vp == lmp->layerm_rootvp) {
727 /*
728 * Oops! We no longer have a root node. Most likely reason is
729 * that someone forcably unmunted the underlying fs.
730 *
731 * Now getting the root vnode will fail. We're dead. :-(
732 */
733 lmp->layerm_rootvp = NULL;
734 }
735
736 mutex_enter(vp->v_interlock);
737 KASSERT(vp->v_interlock == lowervp->v_interlock);
738 lowervp->v_writecount -= vp->v_writecount;
739 mutex_exit(vp->v_interlock);
740
741 /* After this assignment, this node will not be re-used. */
742 xp->layer_lowervp = NULL;
743 kmem_free(vp->v_data, lmp->layerm_size);
744 vp->v_data = NULL;
745 vrele(lowervp);
746
747 return 0;
748 }
749
750 /*
751 * We just feed the returned vnode up to the caller - there's no need
752 * to build a layer node on top of the node on which we're going to do
753 * i/o. :-)
754 */
755 int
layer_bmap(void * v)756 layer_bmap(void *v)
757 {
758 struct vop_bmap_args /* {
759 struct vnode *a_vp;
760 daddr_t a_bn;
761 struct vnode **a_vpp;
762 daddr_t *a_bnp;
763 int *a_runp;
764 } */ *ap = v;
765 struct vnode *vp;
766
767 vp = LAYERVPTOLOWERVP(ap->a_vp);
768 ap->a_vp = vp;
769
770 return VCALL(vp, ap->a_desc->vdesc_offset, ap);
771 }
772
773 int
layer_print(void * v)774 layer_print(void *v)
775 {
776 struct vop_print_args /* {
777 struct vnode *a_vp;
778 } */ *ap = v;
779 struct vnode *vp = ap->a_vp;
780 printf ("\ttag VT_LAYERFS, vp=%p, lowervp=%p\n", vp, LAYERVPTOLOWERVP(vp));
781 return 0;
782 }
783
784 int
layer_getpages(void * v)785 layer_getpages(void *v)
786 {
787 struct vop_getpages_args /* {
788 struct vnode *a_vp;
789 voff_t a_offset;
790 struct vm_page **a_m;
791 int *a_count;
792 int a_centeridx;
793 vm_prot_t a_access_type;
794 int a_advice;
795 int a_flags;
796 } */ *ap = v;
797 struct vnode *vp = ap->a_vp;
798 struct mount *mp = vp->v_mount;
799 int error;
800 krw_t op;
801
802 KASSERT(rw_lock_held(vp->v_uobj.vmobjlock));
803
804 if (ap->a_flags & PGO_LOCKED) {
805 return EBUSY;
806 }
807 ap->a_vp = LAYERVPTOLOWERVP(vp);
808 KASSERT(vp->v_uobj.vmobjlock == ap->a_vp->v_uobj.vmobjlock);
809
810 /* Just pass the request on to the underlying layer. */
811 op = rw_lock_op(vp->v_uobj.vmobjlock);
812 rw_exit(vp->v_uobj.vmobjlock);
813 fstrans_start(mp);
814 rw_enter(vp->v_uobj.vmobjlock, op);
815 if (mp == vp->v_mount) {
816 /* Will release the lock. */
817 error = VCALL(ap->a_vp, VOFFSET(vop_getpages), ap);
818 } else {
819 rw_exit(vp->v_uobj.vmobjlock);
820 error = ENOENT;
821 }
822 fstrans_done(mp);
823
824 return error;
825 }
826
827 int
layer_putpages(void * v)828 layer_putpages(void *v)
829 {
830 struct vop_putpages_args /* {
831 struct vnode *a_vp;
832 voff_t a_offlo;
833 voff_t a_offhi;
834 int a_flags;
835 } */ *ap = v;
836 struct vnode *vp = ap->a_vp;
837
838 KASSERT(rw_write_held(vp->v_uobj.vmobjlock));
839
840 ap->a_vp = LAYERVPTOLOWERVP(vp);
841 KASSERT(vp->v_uobj.vmobjlock == ap->a_vp->v_uobj.vmobjlock);
842
843 if (ap->a_flags & PGO_RECLAIM) {
844 rw_exit(vp->v_uobj.vmobjlock);
845 return 0;
846 }
847
848 /* Just pass the request on to the underlying layer. */
849 return VCALL(ap->a_vp, VOFFSET(vop_putpages), ap);
850 }
851