xref: /netbsd-src/sys/miscfs/umapfs/umap_vnops.c (revision 73704c4ce4ee2a60eb617e693ce7e9f03902613e)
1 /*	$NetBSD: umap_vnops.c,v 1.23 2003/08/07 16:32:46 agc Exp $	*/
2 
3 /*
4  * Copyright (c) 1992, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software donated to Berkeley by
8  * the 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. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)umap_vnops.c	8.6 (Berkeley) 5/22/95
35  */
36 
37 /*
38  * Umap Layer
39  */
40 
41 #include <sys/cdefs.h>
42 __KERNEL_RCSID(0, "$NetBSD: umap_vnops.c,v 1.23 2003/08/07 16:32:46 agc Exp $");
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/time.h>
47 #include <sys/vnode.h>
48 #include <sys/mount.h>
49 #include <sys/namei.h>
50 #include <sys/malloc.h>
51 #include <sys/buf.h>
52 #include <miscfs/umapfs/umap.h>
53 #include <miscfs/genfs/genfs.h>
54 #include <miscfs/genfs/layer_extern.h>
55 
56 int	umap_lookup	__P((void *));
57 int	umap_getattr	__P((void *));
58 int	umap_print	__P((void *));
59 int	umap_rename	__P((void *));
60 
61 /*
62  * Global vfs data structures
63  */
64 /*
65  * XXX - strategy, bwrite are hand coded currently.  They should
66  * go away with a merged buffer/block cache.
67  *
68  */
69 int (**umap_vnodeop_p) __P((void *));
70 const struct vnodeopv_entry_desc umap_vnodeop_entries[] = {
71 	{ &vop_default_desc,	umap_bypass },
72 
73 	{ &vop_lookup_desc,	umap_lookup },
74 	{ &vop_getattr_desc,	umap_getattr },
75 	{ &vop_print_desc,	umap_print },
76 	{ &vop_rename_desc,	umap_rename },
77 
78 	{ &vop_lock_desc,	layer_lock },
79 	{ &vop_unlock_desc,	layer_unlock },
80 	{ &vop_islocked_desc,	layer_islocked },
81 	{ &vop_fsync_desc,	layer_fsync },
82 	{ &vop_inactive_desc,	layer_inactive },
83 	{ &vop_reclaim_desc,	layer_reclaim },
84 	{ &vop_open_desc,	layer_open },
85 	{ &vop_setattr_desc,	layer_setattr },
86 	{ &vop_access_desc,	layer_access },
87 
88 	{ &vop_strategy_desc,	layer_strategy },
89 	{ &vop_bwrite_desc,	layer_bwrite },
90 	{ &vop_bmap_desc,	layer_bmap },
91 	{ &vop_getpages_desc,	layer_getpages },
92 	{ &vop_putpages_desc,	layer_putpages },
93 
94 	{ NULL, NULL }
95 };
96 const struct vnodeopv_desc umapfs_vnodeop_opv_desc =
97 	{ &umap_vnodeop_p, umap_vnodeop_entries };
98 
99 /*
100  * This is the 08-June-1999 bypass routine.
101  * See layer_vnops.c:layer_bypass for more details.
102  */
103 int
104 umap_bypass(v)
105 	void *v;
106 {
107 	struct vop_generic_args /* {
108 		struct vnodeop_desc *a_desc;
109 		<other random data follows, presumably>
110 	} */ *ap = v;
111 	struct ucred **credpp = 0, *credp = 0;
112 	struct ucred *savecredp = 0, *savecompcredp = 0;
113 	struct ucred *compcredp = 0;
114 	struct vnode **this_vp_p;
115 	int error, error1;
116 	int (**our_vnodeop_p) __P((void *));
117 	struct vnode *old_vps[VDESC_MAX_VPS], *vp0;
118 	struct vnode **vps_p[VDESC_MAX_VPS];
119 	struct vnode ***vppp;
120 	struct vnodeop_desc *descp = ap->a_desc;
121 	int reles, i, flags;
122 	struct componentname **compnamepp = 0;
123 
124 #ifdef SAFETY
125 	/*
126 	 * We require at least one vp.
127 	 */
128 	if (descp->vdesc_vp_offsets == NULL ||
129 	    descp->vdesc_vp_offsets[0] == VDESC_NO_OFFSET)
130 		panic ("umap_bypass: no vp's in map.\n");
131 #endif
132 	vps_p[0] = VOPARG_OFFSETTO(struct vnode**,descp->vdesc_vp_offsets[0],
133 				ap);
134 	vp0 = *vps_p[0];
135 	flags = MOUNTTOUMAPMOUNT(vp0->v_mount)->umapm_flags;
136 	our_vnodeop_p = vp0->v_op;
137 
138 	if (flags & LAYERFS_MBYPASSDEBUG)
139 		printf("umap_bypass: %s\n", descp->vdesc_name);
140 
141 	/*
142 	 * Map the vnodes going in.
143 	 * Later, we'll invoke the operation based on
144 	 * the first mapped vnode's operation vector.
145 	 */
146 	reles = descp->vdesc_flags;
147 	for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) {
148 		if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET)
149 			break;   /* bail out at end of list */
150 		vps_p[i] = this_vp_p =
151 			VOPARG_OFFSETTO(struct vnode**, descp->vdesc_vp_offsets[i], ap);
152 
153 		/*
154 		 * We're not guaranteed that any but the first vnode
155 		 * are of our type.  Check for and don't map any
156 		 * that aren't.  (Must map first vp or vclean fails.)
157 		 */
158 
159 		if (i && ((*this_vp_p)==NULL ||
160 		    (*this_vp_p)->v_op != our_vnodeop_p)) {
161 			old_vps[i] = NULL;
162 		} else {
163 			old_vps[i] = *this_vp_p;
164 			*(vps_p[i]) = UMAPVPTOLOWERVP(*this_vp_p);
165 			if (reles & 1)
166 				VREF(*this_vp_p);
167 		}
168 
169 	}
170 
171 	/*
172 	 * Fix the credentials.  (That's the purpose of this layer.)
173 	 */
174 
175 	if (descp->vdesc_cred_offset != VDESC_NO_OFFSET) {
176 
177 		credpp = VOPARG_OFFSETTO(struct ucred**,
178 		    descp->vdesc_cred_offset, ap);
179 
180 		/* Save old values */
181 
182 		savecredp = *credpp;
183 		if (savecredp != NOCRED)
184 			*credpp = crdup(savecredp);
185 		credp = *credpp;
186 
187 		if ((flags & LAYERFS_MBYPASSDEBUG) && credp->cr_uid != 0)
188 			printf("umap_bypass: user was %d, group %d\n",
189 			    credp->cr_uid, credp->cr_gid);
190 
191 		/* Map all ids in the credential structure. */
192 
193 		umap_mapids(vp0->v_mount, credp);
194 
195 		if ((flags & LAYERFS_MBYPASSDEBUG) && credp->cr_uid != 0)
196 			printf("umap_bypass: user now %d, group %d\n",
197 			    credp->cr_uid, credp->cr_gid);
198 	}
199 
200 	/* BSD often keeps a credential in the componentname structure
201 	 * for speed.  If there is one, it better get mapped, too.
202 	 */
203 
204 	if (descp->vdesc_componentname_offset != VDESC_NO_OFFSET) {
205 
206 		compnamepp = VOPARG_OFFSETTO(struct componentname**,
207 		    descp->vdesc_componentname_offset, ap);
208 
209 		savecompcredp = (*compnamepp)->cn_cred;
210 		if (savecompcredp != NOCRED)
211 			(*compnamepp)->cn_cred = crdup(savecompcredp);
212 		compcredp = (*compnamepp)->cn_cred;
213 
214 		if ((flags & LAYERFS_MBYPASSDEBUG) && compcredp->cr_uid != 0)
215 			printf("umap_bypass: component credit user was %d, group %d\n",
216 			    compcredp->cr_uid, compcredp->cr_gid);
217 
218 		/* Map all ids in the credential structure. */
219 
220 		umap_mapids(vp0->v_mount, compcredp);
221 
222 		if ((flags & LAYERFS_MBYPASSDEBUG) && compcredp->cr_uid != 0)
223 			printf("umap_bypass: component credit user now %d, group %d\n",
224 			    compcredp->cr_uid, compcredp->cr_gid);
225 	}
226 
227 	/*
228 	 * Call the operation on the lower layer
229 	 * with the modified argument structure.
230 	 */
231 	error = VCALL(*(vps_p[0]), descp->vdesc_offset, ap);
232 
233 	/*
234 	 * Maintain the illusion of call-by-value
235 	 * by restoring vnodes in the argument structure
236 	 * to their original value.
237 	 */
238 	reles = descp->vdesc_flags;
239 	for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) {
240 		if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET)
241 			break;   /* bail out at end of list */
242 		if (old_vps[i]) {
243 			*(vps_p[i]) = old_vps[i];
244 			if (reles & VDESC_VP0_WILLUNLOCK)
245 				LAYERFS_UPPERUNLOCK(*(vps_p[i]), 0, error1);
246 			if (reles & VDESC_VP0_WILLRELE)
247 				vrele(*(vps_p[i]));
248 		};
249 	};
250 
251 	/*
252 	 * Map the possible out-going vpp
253 	 * (Assumes that the lower layer always returns
254 	 * a VREF'ed vpp unless it gets an error.)
255 	 */
256 	if (descp->vdesc_vpp_offset != VDESC_NO_OFFSET &&
257 	    !(descp->vdesc_flags & VDESC_NOMAP_VPP) &&
258 	    !error) {
259 		if (descp->vdesc_flags & VDESC_VPP_WILLRELE)
260 			goto out;
261 		vppp = VOPARG_OFFSETTO(struct vnode***,
262 				 descp->vdesc_vpp_offset, ap);
263 		error = layer_node_create(old_vps[0]->v_mount, **vppp, *vppp);
264 	};
265 
266  out:
267 	/*
268 	 * Free duplicate cred structure and restore old one.
269 	 */
270 	if (descp->vdesc_cred_offset != VDESC_NO_OFFSET) {
271 		if ((flags & LAYERFS_MBYPASSDEBUG) && credp &&
272 					credp->cr_uid != 0)
273 			printf("umap_bypass: returning-user was %d\n",
274 			    credp->cr_uid);
275 
276 		if (savecredp != NOCRED) {
277 			crfree(credp);
278 			*credpp = savecredp;
279 			if ((flags & LAYERFS_MBYPASSDEBUG) && credpp &&
280 					(*credpp)->cr_uid != 0)
281 			 	printf("umap_bypass: returning-user now %d\n\n",
282 				    savecredp->cr_uid);
283 		}
284 	}
285 
286 	if (descp->vdesc_componentname_offset != VDESC_NO_OFFSET) {
287 		if ((flags & LAYERFS_MBYPASSDEBUG) && compcredp &&
288 					compcredp->cr_uid != 0)
289 			printf("umap_bypass: returning-component-user was %d\n",
290 			    compcredp->cr_uid);
291 
292 		if (savecompcredp != NOCRED) {
293 			crfree(compcredp);
294 			(*compnamepp)->cn_cred = savecompcredp;
295 			if ((flags & LAYERFS_MBYPASSDEBUG) && savecompcredp &&
296 					savecompcredp->cr_uid != 0)
297 			 	printf("umap_bypass: returning-component-user now %d\n",
298 				    savecompcredp->cr_uid);
299 		}
300 	}
301 
302 	return (error);
303 }
304 
305 /*
306  * This is based on the 08-June-1999 bypass routine.
307  * See layer_vnops.c:layer_bypass for more details.
308  */
309 int
310 umap_lookup(v)
311 	void *v;
312 {
313 	struct vop_lookup_args /* {
314 		struct vnodeop_desc *a_desc;
315 		struct vnode * a_dvp;
316 		struct vnode ** a_vpp;
317 		struct componentname * a_cnp;
318 	} */ *ap = v;
319 	struct componentname *cnp = ap->a_cnp;
320 	struct ucred *savecompcredp = NULL;
321 	struct ucred *compcredp = NULL;
322 	struct vnode *dvp, *vp, *ldvp;
323 	struct mount *mp;
324 	int error;
325 	int i, flags, cnf = cnp->cn_flags;
326 
327 	dvp = ap->a_dvp;
328 	mp = dvp->v_mount;
329 
330 	if ((cnf & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
331 		(cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
332 		return (EROFS);
333 
334 	flags = MOUNTTOUMAPMOUNT(mp)->umapm_flags;
335 	ldvp = UMAPVPTOLOWERVP(dvp);
336 
337 	if (flags & LAYERFS_MBYPASSDEBUG)
338 		printf("umap_lookup\n");
339 
340 	/*
341 	 * Fix the credentials.  (That's the purpose of this layer.)
342 	 *
343 	 * BSD often keeps a credential in the componentname structure
344 	 * for speed.  If there is one, it better get mapped, too.
345 	 */
346 
347 	if ((savecompcredp = cnp->cn_cred)) {
348 		compcredp = crdup(savecompcredp);
349 		cnp->cn_cred = compcredp;
350 
351 		if ((flags & LAYERFS_MBYPASSDEBUG) && compcredp->cr_uid != 0)
352 			printf("umap_lookup: component credit user was %d, group %d\n",
353 			    compcredp->cr_uid, compcredp->cr_gid);
354 
355 		/* Map all ids in the credential structure. */
356 		umap_mapids(mp, compcredp);
357 	}
358 
359 	if ((flags & LAYERFS_MBYPASSDEBUG) && compcredp->cr_uid != 0)
360 		printf("umap_lookup: component credit user now %d, group %d\n",
361 		    compcredp->cr_uid, compcredp->cr_gid);
362 
363 	ap->a_dvp = ldvp;
364 	error = VCALL(ldvp, ap->a_desc->vdesc_offset, ap);
365 	vp = *ap->a_vpp;
366 
367 	if (error == EJUSTRETURN && (cnf & ISLASTCN) &&
368 	    (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
369 	    (cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME))
370 		error = EROFS;
371 
372 	/* Do locking fixup as appropriate. See layer_lookup() for info */
373 	if ((cnp->cn_flags & PDIRUNLOCK)) {
374 		LAYERFS_UPPERUNLOCK(dvp, 0, i);
375 	}
376 	if (ldvp == vp) {
377 		*ap->a_vpp = dvp;
378 		VREF(dvp);
379 		vrele(vp);
380 	} else if (vp != NULL) {
381 		error = layer_node_create(mp, vp, ap->a_vpp);
382 	}
383 
384 	/*
385 	 * Free duplicate cred structure and restore old one.
386 	 */
387 	if ((flags & LAYERFS_MBYPASSDEBUG) && compcredp &&
388 					compcredp->cr_uid != 0)
389 		printf("umap_lookup: returning-component-user was %d\n",
390 			    compcredp->cr_uid);
391 
392 	if (savecompcredp != NOCRED) {
393 		crfree(compcredp);
394 		cnp->cn_cred = savecompcredp;
395 		if ((flags & LAYERFS_MBYPASSDEBUG) && savecompcredp &&
396 				savecompcredp->cr_uid != 0)
397 		 	printf("umap_lookup: returning-component-user now %d\n",
398 			    savecompcredp->cr_uid);
399 	}
400 
401 	return (error);
402 }
403 
404 /*
405  *  We handle getattr to change the fsid.
406  */
407 int
408 umap_getattr(v)
409 	void *v;
410 {
411 	struct vop_getattr_args /* {
412 		struct vnode *a_vp;
413 		struct vattr *a_vap;
414 		struct ucred *a_cred;
415 		struct proc *a_p;
416 	} */ *ap = v;
417 	uid_t uid;
418 	gid_t gid;
419 	int error, tmpid, nentries, gnentries, flags;
420 	u_long (*mapdata)[2];
421 	u_long (*gmapdata)[2];
422 	struct vnode **vp1p;
423 	const struct vnodeop_desc *descp = ap->a_desc;
424 
425 	if ((error = umap_bypass(ap)) != 0)
426 		return (error);
427 	/* Requires that arguments be restored. */
428 	ap->a_vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0];
429 
430 	flags = MOUNTTOUMAPMOUNT(ap->a_vp->v_mount)->umapm_flags;
431 	/*
432 	 * Umap needs to map the uid and gid returned by a stat
433 	 * into the proper values for this site.  This involves
434 	 * finding the returned uid in the mapping information,
435 	 * translating it into the uid on the other end,
436 	 * and filling in the proper field in the vattr
437 	 * structure pointed to by ap->a_vap.  The group
438 	 * is easier, since currently all groups will be
439 	 * translate to the NULLGROUP.
440 	 */
441 
442 	/* Find entry in map */
443 
444 	uid = ap->a_vap->va_uid;
445 	gid = ap->a_vap->va_gid;
446 	if ((flags & LAYERFS_MBYPASSDEBUG))
447 		printf("umap_getattr: mapped uid = %d, mapped gid = %d\n", uid,
448 		    gid);
449 
450 	vp1p = VOPARG_OFFSETTO(struct vnode**, descp->vdesc_vp_offsets[0], ap);
451 	nentries =  MOUNTTOUMAPMOUNT((*vp1p)->v_mount)->info_nentries;
452 	mapdata =  (MOUNTTOUMAPMOUNT((*vp1p)->v_mount)->info_mapdata);
453 	gnentries =  MOUNTTOUMAPMOUNT((*vp1p)->v_mount)->info_gnentries;
454 	gmapdata =  (MOUNTTOUMAPMOUNT((*vp1p)->v_mount)->info_gmapdata);
455 
456 	/* Reverse map the uid for the vnode.  Since it's a reverse
457 		map, we can't use umap_mapids() to do it. */
458 
459 	tmpid = umap_reverse_findid(uid, mapdata, nentries);
460 
461 	if (tmpid != -1) {
462 		ap->a_vap->va_uid = (uid_t) tmpid;
463 		if ((flags & LAYERFS_MBYPASSDEBUG))
464 			printf("umap_getattr: original uid = %d\n", uid);
465 	} else
466 		ap->a_vap->va_uid = (uid_t) NOBODY;
467 
468 	/* Reverse map the gid for the vnode. */
469 
470 	tmpid = umap_reverse_findid(gid, gmapdata, gnentries);
471 
472 	if (tmpid != -1) {
473 		ap->a_vap->va_gid = (gid_t) tmpid;
474 		if ((flags & LAYERFS_MBYPASSDEBUG))
475 			printf("umap_getattr: original gid = %d\n", gid);
476 	} else
477 		ap->a_vap->va_gid = (gid_t) NULLGROUP;
478 
479 	return (0);
480 }
481 
482 int
483 umap_print(v)
484 	void *v;
485 {
486 	struct vop_print_args /* {
487 		struct vnode *a_vp;
488 	} */ *ap = v;
489 	struct vnode *vp = ap->a_vp;
490 	printf("\ttag VT_UMAPFS, vp=%p, lowervp=%p\n", vp,
491 	    UMAPVPTOLOWERVP(vp));
492 	return (0);
493 }
494 
495 int
496 umap_rename(v)
497 	void *v;
498 {
499 	struct vop_rename_args  /* {
500 		struct vnode *a_fdvp;
501 		struct vnode *a_fvp;
502 		struct componentname *a_fcnp;
503 		struct vnode *a_tdvp;
504 		struct vnode *a_tvp;
505 		struct componentname *a_tcnp;
506 	} */ *ap = v;
507 	int error, flags;
508 	struct componentname *compnamep;
509 	struct ucred *compcredp, *savecompcredp;
510 	struct vnode *vp;
511 
512 	/*
513 	 * Rename is irregular, having two componentname structures.
514 	 * We need to map the cre in the second structure,
515 	 * and then bypass takes care of the rest.
516 	 */
517 
518 	vp = ap->a_fdvp;
519 	flags = MOUNTTOUMAPMOUNT(vp->v_mount)->umapm_flags;
520 	compnamep = ap->a_tcnp;
521 	compcredp = compnamep->cn_cred;
522 
523 	savecompcredp = compcredp;
524 	compcredp = compnamep->cn_cred = crdup(savecompcredp);
525 
526 	if ((flags & LAYERFS_MBYPASSDEBUG) && compcredp->cr_uid != 0)
527 		printf("umap_rename: rename component credit user was %d, group %d\n",
528 		    compcredp->cr_uid, compcredp->cr_gid);
529 
530 	/* Map all ids in the credential structure. */
531 
532 	umap_mapids(vp->v_mount, compcredp);
533 
534 	if ((flags & LAYERFS_MBYPASSDEBUG) && compcredp->cr_uid != 0)
535 		printf("umap_rename: rename component credit user now %d, group %d\n",
536 		    compcredp->cr_uid, compcredp->cr_gid);
537 
538 	error = umap_bypass(ap);
539 
540 	/* Restore the additional mapped componentname cred structure. */
541 
542 	crfree(compcredp);
543 	compnamep->cn_cred = savecompcredp;
544 
545 	return error;
546 }
547