xref: /freebsd-14.2/sys/fs/nullfs/null_vfsops.c (revision 0ee0770f)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1992, 1993, 1995
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software donated to Berkeley by
8  * Jan-Simon Pendry.
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  *	@(#)null_vfsops.c	8.2 (Berkeley) 1/21/94
35  *
36  * @(#)lofs_vfsops.c	1.2 (Berkeley) 6/18/92
37  */
38 
39 /*
40  * Null Layer
41  * (See null_vnops.c for a description of what this does.)
42  */
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/fcntl.h>
47 #include <sys/kernel.h>
48 #include <sys/lock.h>
49 #include <sys/malloc.h>
50 #include <sys/mount.h>
51 #include <sys/namei.h>
52 #include <sys/proc.h>
53 #include <sys/sysctl.h>
54 #include <sys/vnode.h>
55 #include <sys/jail.h>
56 
57 #include <fs/nullfs/null.h>
58 
59 static MALLOC_DEFINE(M_NULLFSMNT, "nullfs_mount", "NULLFS mount structure");
60 
61 static vfs_fhtovp_t	nullfs_fhtovp;
62 static vfs_mount_t	nullfs_mount;
63 static vfs_quotactl_t	nullfs_quotactl;
64 static vfs_root_t	nullfs_root;
65 static vfs_sync_t	nullfs_sync;
66 static vfs_statfs_t	nullfs_statfs;
67 static vfs_unmount_t	nullfs_unmount;
68 static vfs_vget_t	nullfs_vget;
69 static vfs_extattrctl_t	nullfs_extattrctl;
70 
71 SYSCTL_NODE(_vfs, OID_AUTO, nullfs, CTLFLAG_RW, 0, "nullfs");
72 
73 static bool null_cache_vnodes = true;
74 SYSCTL_BOOL(_vfs_nullfs, OID_AUTO, cache_vnodes, CTLFLAG_RWTUN,
75     &null_cache_vnodes, 0,
76     "cache free nullfs vnodes");
77 
78 /*
79  * Mount null layer
80  */
81 static int
nullfs_mount(struct mount * mp)82 nullfs_mount(struct mount *mp)
83 {
84 	struct vnode *lowerrootvp;
85 	struct vnode *nullm_rootvp;
86 	struct null_mount *xmp;
87 	struct null_node *nn;
88 	struct nameidata nd, *ndp;
89 	char *target;
90 	int error, len;
91 	bool isvnunlocked;
92 
93 	NULLFSDEBUG("nullfs_mount(mp = %p)\n", (void *)mp);
94 
95 	if (mp->mnt_flag & MNT_ROOTFS)
96 		return (EOPNOTSUPP);
97 
98 	/*
99 	 * Update is a no-op
100 	 */
101 	if (mp->mnt_flag & MNT_UPDATE) {
102 		/*
103 		 * Only support update mounts for NFS export.
104 		 */
105 		if (vfs_flagopt(mp->mnt_optnew, "export", NULL, 0))
106 			return (0);
107 		else
108 			return (EOPNOTSUPP);
109 	}
110 
111 	/*
112 	 * Get argument
113 	 */
114 	error = vfs_getopt(mp->mnt_optnew, "from", (void **)&target, &len);
115 	if (error != 0)
116 		error = vfs_getopt(mp->mnt_optnew, "target", (void **)&target, &len);
117 	if (error || target[len - 1] != '\0')
118 		return (EINVAL);
119 
120 	/*
121 	 * Unlock lower node to avoid possible deadlock.
122 	 */
123 	if (mp->mnt_vnodecovered->v_op == &null_vnodeops &&
124 	    VOP_ISLOCKED(mp->mnt_vnodecovered) == LK_EXCLUSIVE) {
125 		VOP_UNLOCK(mp->mnt_vnodecovered);
126 		isvnunlocked = true;
127 	} else {
128 		isvnunlocked = false;
129 	}
130 
131 	/*
132 	 * Find lower node
133 	 */
134 	ndp = &nd;
135 	NDINIT(ndp, LOOKUP, FOLLOW|LOCKLEAF, UIO_SYSSPACE, target);
136 	error = namei(ndp);
137 
138 	/*
139 	 * Re-lock vnode.
140 	 * XXXKIB This is deadlock-prone as well.
141 	 */
142 	if (isvnunlocked)
143 		vn_lock(mp->mnt_vnodecovered, LK_EXCLUSIVE | LK_RETRY);
144 
145 	if (error)
146 		return (error);
147 	NDFREE_PNBUF(ndp);
148 
149 	/*
150 	 * Sanity check on lower vnode
151 	 */
152 	lowerrootvp = ndp->ni_vp;
153 
154 	/*
155 	 * Check multi null mount to avoid `lock against myself' panic.
156 	 */
157 	if (mp->mnt_vnodecovered->v_op == &null_vnodeops) {
158 		nn = VTONULL(mp->mnt_vnodecovered);
159 		if (nn == NULL || lowerrootvp == nn->null_lowervp) {
160 			NULLFSDEBUG("nullfs_mount: multi null mount?\n");
161 			vput(lowerrootvp);
162 			return (EDEADLK);
163 		}
164 	}
165 
166 	/*
167 	 * Lower vnode must be the same type as the covered vnode - we
168 	 * don't allow mounting directories to files or vice versa.
169 	 */
170 	if ((lowerrootvp->v_type != VDIR && lowerrootvp->v_type != VREG) ||
171 	    lowerrootvp->v_type != mp->mnt_vnodecovered->v_type) {
172 		NULLFSDEBUG("nullfs_mount: target must be same type as fspath");
173 		vput(lowerrootvp);
174 		return (EINVAL);
175 	}
176 
177 	xmp = malloc(sizeof(struct null_mount), M_NULLFSMNT,
178 	    M_WAITOK | M_ZERO);
179 
180 	/*
181 	 * Save pointer to underlying FS and the reference to the
182 	 * lower root vnode.
183 	 */
184 	xmp->nullm_vfs = vfs_register_upper_from_vp(lowerrootvp, mp,
185 	    &xmp->upper_node);
186 	if (xmp->nullm_vfs == NULL) {
187 		vput(lowerrootvp);
188 		free(xmp, M_NULLFSMNT);
189 		return (ENOENT);
190 	}
191 	vref(lowerrootvp);
192 	xmp->nullm_lowerrootvp = lowerrootvp;
193 	mp->mnt_data = xmp;
194 
195 	/*
196 	 * Make sure the node alias worked.
197 	 */
198 	error = null_nodeget(mp, lowerrootvp, &nullm_rootvp);
199 	if (error != 0) {
200 		vfs_unregister_upper(xmp->nullm_vfs, &xmp->upper_node);
201 		vrele(lowerrootvp);
202 		free(xmp, M_NULLFSMNT);
203 		return (error);
204 	}
205 
206 	if (NULLVPTOLOWERVP(nullm_rootvp)->v_mount->mnt_flag & MNT_LOCAL) {
207 		MNT_ILOCK(mp);
208 		mp->mnt_flag |= MNT_LOCAL;
209 		MNT_IUNLOCK(mp);
210 	}
211 
212 	if (vfs_getopt(mp->mnt_optnew, "cache", NULL, NULL) == 0) {
213 		xmp->nullm_flags |= NULLM_CACHE;
214 	} else if (vfs_getopt(mp->mnt_optnew, "nocache", NULL, NULL) == 0) {
215 		;
216 	} else if (null_cache_vnodes &&
217 	    (xmp->nullm_vfs->mnt_kern_flag & MNTK_NULL_NOCACHE) == 0) {
218 		xmp->nullm_flags |= NULLM_CACHE;
219 	}
220 
221 	if ((xmp->nullm_flags & NULLM_CACHE) != 0) {
222 		vfs_register_for_notification(xmp->nullm_vfs, mp,
223 		    &xmp->notify_node);
224 	}
225 
226 	if (lowerrootvp == mp->mnt_vnodecovered) {
227 		vn_lock(lowerrootvp, LK_EXCLUSIVE | LK_RETRY | LK_CANRECURSE);
228 		lowerrootvp->v_vflag |= VV_CROSSLOCK;
229 		VOP_UNLOCK(lowerrootvp);
230 	}
231 
232 	MNT_ILOCK(mp);
233 	if ((xmp->nullm_flags & NULLM_CACHE) != 0) {
234 		mp->mnt_kern_flag |= lowerrootvp->v_mount->mnt_kern_flag &
235 		    (MNTK_SHARED_WRITES | MNTK_LOOKUP_SHARED |
236 		    MNTK_EXTENDED_SHARED);
237 	}
238 	mp->mnt_kern_flag |= MNTK_NOMSYNC | MNTK_UNLOCKED_INSMNTQUE;
239 	mp->mnt_kern_flag |= lowerrootvp->v_mount->mnt_kern_flag &
240 	    (MNTK_USES_BCACHE | MNTK_NO_IOPF | MNTK_UNMAPPED_BUFS);
241 	MNT_IUNLOCK(mp);
242 	vfs_getnewfsid(mp);
243 	vfs_mountedfrom(mp, target);
244 	vput(nullm_rootvp);
245 
246 	NULLFSDEBUG("nullfs_mount: lower %s, alias at %s\n",
247 		mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntonname);
248 	return (0);
249 }
250 
251 /*
252  * Free reference to null layer
253  */
254 static int
nullfs_unmount(struct mount * mp,int mntflags)255 nullfs_unmount(struct mount *mp, int mntflags)
256 {
257 	struct null_mount *mntdata;
258 	int error, flags;
259 
260 	NULLFSDEBUG("nullfs_unmount: mp = %p\n", (void *)mp);
261 
262 	if (mntflags & MNT_FORCE)
263 		flags = FORCECLOSE;
264 	else
265 		flags = 0;
266 
267 	for (;;) {
268 		/* There is 1 extra root vnode reference (nullm_rootvp). */
269 		error = vflush(mp, 0, flags, curthread);
270 		if (error)
271 			return (error);
272 		MNT_ILOCK(mp);
273 		if (mp->mnt_nvnodelistsize == 0) {
274 			MNT_IUNLOCK(mp);
275 			break;
276 		}
277 		MNT_IUNLOCK(mp);
278 		if ((mntflags & MNT_FORCE) == 0)
279 			return (EBUSY);
280 	}
281 
282 	/*
283 	 * Finally, throw away the null_mount structure
284 	 */
285 	mntdata = mp->mnt_data;
286 	if ((mntdata->nullm_flags & NULLM_CACHE) != 0) {
287 		vfs_unregister_for_notification(mntdata->nullm_vfs,
288 		    &mntdata->notify_node);
289 	}
290 	if (mntdata->nullm_lowerrootvp == mp->mnt_vnodecovered) {
291 		vn_lock(mp->mnt_vnodecovered, LK_EXCLUSIVE | LK_RETRY | LK_CANRECURSE);
292 		mp->mnt_vnodecovered->v_vflag &= ~VV_CROSSLOCK;
293 		VOP_UNLOCK(mp->mnt_vnodecovered);
294 	}
295 	vfs_unregister_upper(mntdata->nullm_vfs, &mntdata->upper_node);
296 	vrele(mntdata->nullm_lowerrootvp);
297 	mp->mnt_data = NULL;
298 	free(mntdata, M_NULLFSMNT);
299 	return (0);
300 }
301 
302 static int
nullfs_root(struct mount * mp,int flags,struct vnode ** vpp)303 nullfs_root(struct mount *mp, int flags, struct vnode **vpp)
304 {
305 	struct vnode *vp;
306 	struct null_mount *mntdata;
307 	int error;
308 
309 	mntdata = MOUNTTONULLMOUNT(mp);
310 	NULLFSDEBUG("nullfs_root(mp = %p, vp = %p)\n", mp,
311 	    mntdata->nullm_lowerrootvp);
312 
313 	error = vget(mntdata->nullm_lowerrootvp, flags);
314 	if (error == 0) {
315 		error = null_nodeget(mp, mntdata->nullm_lowerrootvp, &vp);
316 		if (error == 0) {
317 			*vpp = vp;
318 		}
319 	}
320 	return (error);
321 }
322 
323 static int
nullfs_quotactl(struct mount * mp,int cmd,uid_t uid,void * arg,bool * mp_busy)324 nullfs_quotactl(struct mount *mp, int cmd, uid_t uid, void *arg, bool *mp_busy)
325 {
326 	struct mount *lowermp;
327 	struct null_mount *mntdata;
328 	int error;
329 	bool unbusy;
330 
331 	mntdata = MOUNTTONULLMOUNT(mp);
332 	lowermp = atomic_load_ptr(&mntdata->nullm_vfs);
333 	KASSERT(*mp_busy == true, ("upper mount not busy"));
334 	/*
335 	 * See comment in sys_quotactl() for an explanation of why the
336 	 * lower mount needs to be busied by the caller of VFS_QUOTACTL()
337 	 * but may be unbusied by the implementation.  We must unbusy
338 	 * the upper mount for the same reason; otherwise a namei lookup
339 	 * issued by the VFS_QUOTACTL() implementation could traverse the
340 	 * upper mount and deadlock.
341 	 */
342 	vfs_unbusy(mp);
343 	*mp_busy = false;
344 	unbusy = true;
345 	error = vfs_busy(lowermp, 0);
346 	if (error == 0)
347 		error = VFS_QUOTACTL(lowermp, cmd, uid, arg, &unbusy);
348 	if (unbusy)
349 		vfs_unbusy(lowermp);
350 
351 	return (error);
352 }
353 
354 static int
nullfs_statfs(struct mount * mp,struct statfs * sbp)355 nullfs_statfs(struct mount *mp, struct statfs *sbp)
356 {
357 	int error;
358 	struct statfs *mstat;
359 
360 	NULLFSDEBUG("nullfs_statfs(mp = %p, vp = %p->%p)\n", (void *)mp,
361 	    (void *)MOUNTTONULLMOUNT(mp)->nullm_rootvp,
362 	    (void *)NULLVPTOLOWERVP(MOUNTTONULLMOUNT(mp)->nullm_rootvp));
363 
364 	mstat = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK | M_ZERO);
365 
366 	error = VFS_STATFS(MOUNTTONULLMOUNT(mp)->nullm_vfs, mstat);
367 	if (error) {
368 		free(mstat, M_STATFS);
369 		return (error);
370 	}
371 
372 	/* now copy across the "interesting" information and fake the rest */
373 	sbp->f_type = mstat->f_type;
374 	sbp->f_flags &= MNT_RDONLY | MNT_NOEXEC | MNT_NOSUID | MNT_UNION |
375 	    MNT_NOSYMFOLLOW | MNT_AUTOMOUNTED | MNT_EXPORTED | MNT_IGNORE;
376 	mstat->f_flags &= ~(MNT_ROOTFS | MNT_AUTOMOUNTED | MNT_EXPORTED);
377 	sbp->f_flags |= mstat->f_flags;
378 	sbp->f_bsize = mstat->f_bsize;
379 	sbp->f_iosize = mstat->f_iosize;
380 	sbp->f_blocks = mstat->f_blocks;
381 	sbp->f_bfree = mstat->f_bfree;
382 	sbp->f_bavail = mstat->f_bavail;
383 	sbp->f_files = mstat->f_files;
384 	sbp->f_ffree = mstat->f_ffree;
385 
386 	free(mstat, M_STATFS);
387 	return (0);
388 }
389 
390 static int
nullfs_sync(struct mount * mp,int waitfor)391 nullfs_sync(struct mount *mp, int waitfor)
392 {
393 	/*
394 	 * XXX - Assumes no data cached at null layer.
395 	 */
396 	return (0);
397 }
398 
399 static int
nullfs_vget(struct mount * mp,ino_t ino,int flags,struct vnode ** vpp)400 nullfs_vget(struct mount *mp, ino_t ino, int flags, struct vnode **vpp)
401 {
402 	int error;
403 
404 	KASSERT((flags & LK_TYPE_MASK) != 0,
405 	    ("nullfs_vget: no lock requested"));
406 
407 	error = VFS_VGET(MOUNTTONULLMOUNT(mp)->nullm_vfs, ino, flags, vpp);
408 	if (error != 0)
409 		return (error);
410 	return (null_nodeget(mp, *vpp, vpp));
411 }
412 
413 static int
nullfs_fhtovp(struct mount * mp,struct fid * fidp,int flags,struct vnode ** vpp)414 nullfs_fhtovp(struct mount *mp, struct fid *fidp, int flags, struct vnode **vpp)
415 {
416 	int error;
417 
418 	error = VFS_FHTOVP(MOUNTTONULLMOUNT(mp)->nullm_vfs, fidp, flags,
419 	    vpp);
420 	if (error != 0)
421 		return (error);
422 	return (null_nodeget(mp, *vpp, vpp));
423 }
424 
425 static int
nullfs_extattrctl(struct mount * mp,int cmd,struct vnode * filename_vp,int namespace,const char * attrname)426 nullfs_extattrctl(struct mount *mp, int cmd, struct vnode *filename_vp,
427     int namespace, const char *attrname)
428 {
429 
430 	return (VFS_EXTATTRCTL(MOUNTTONULLMOUNT(mp)->nullm_vfs, cmd,
431 	    filename_vp, namespace, attrname));
432 }
433 
434 static void
nullfs_reclaim_lowervp(struct mount * mp,struct vnode * lowervp)435 nullfs_reclaim_lowervp(struct mount *mp, struct vnode *lowervp)
436 {
437 	struct vnode *vp;
438 
439 	vp = null_hashget(mp, lowervp);
440 	if (vp == NULL)
441 		return;
442 	VTONULL(vp)->null_flags |= NULLV_NOUNLOCK;
443 	vgone(vp);
444 	vput(vp);
445 }
446 
447 static void
nullfs_unlink_lowervp(struct mount * mp,struct vnode * lowervp)448 nullfs_unlink_lowervp(struct mount *mp, struct vnode *lowervp)
449 {
450 	struct vnode *vp;
451 	struct null_node *xp;
452 
453 	vp = null_hashget(mp, lowervp);
454 	if (vp == NULL)
455 		return;
456 	xp = VTONULL(vp);
457 	xp->null_flags |= NULLV_DROP | NULLV_NOUNLOCK;
458 	vhold(vp);
459 	vunref(vp);
460 
461 	if (vp->v_usecount == 0) {
462 		/*
463 		 * If vunref() dropped the last use reference on the
464 		 * nullfs vnode, it must be reclaimed, and its lock
465 		 * was split from the lower vnode lock.  Need to do
466 		 * extra unlock before allowing the final vdrop() to
467 		 * free the vnode.
468 		 */
469 		KASSERT(VN_IS_DOOMED(vp),
470 		    ("not reclaimed nullfs vnode %p", vp));
471 		VOP_UNLOCK(vp);
472 	} else {
473 		/*
474 		 * Otherwise, the nullfs vnode still shares the lock
475 		 * with the lower vnode, and must not be unlocked.
476 		 * Also clear the NULLV_NOUNLOCK, the flag is not
477 		 * relevant for future reclamations.
478 		 */
479 		ASSERT_VOP_ELOCKED(vp, "unlink_lowervp");
480 		KASSERT(!VN_IS_DOOMED(vp),
481 		    ("reclaimed nullfs vnode %p", vp));
482 		xp->null_flags &= ~NULLV_NOUNLOCK;
483 	}
484 	vdrop(vp);
485 }
486 
487 static struct vfsops null_vfsops = {
488 	.vfs_extattrctl =	nullfs_extattrctl,
489 	.vfs_fhtovp =		nullfs_fhtovp,
490 	.vfs_init =		nullfs_init,
491 	.vfs_mount =		nullfs_mount,
492 	.vfs_quotactl =		nullfs_quotactl,
493 	.vfs_root =		nullfs_root,
494 	.vfs_statfs =		nullfs_statfs,
495 	.vfs_sync =		nullfs_sync,
496 	.vfs_uninit =		nullfs_uninit,
497 	.vfs_unmount =		nullfs_unmount,
498 	.vfs_vget =		nullfs_vget,
499 	.vfs_reclaim_lowervp =	nullfs_reclaim_lowervp,
500 	.vfs_unlink_lowervp =	nullfs_unlink_lowervp,
501 };
502 
503 VFS_SET(null_vfsops, nullfs, VFCF_LOOPBACK | VFCF_JAIL | VFCF_FILEMOUNT);
504