xref: /freebsd/sys/fs/nullfs/null_vfsops.c (revision cc68614da8232d8baaca0ae0d0dd8f890f06623e)
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  * $FreeBSD$
38  */
39 
40 /*
41  * Null Layer
42  * (See null_vnops.c for a description of what this does.)
43  */
44 
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/fcntl.h>
48 #include <sys/kernel.h>
49 #include <sys/lock.h>
50 #include <sys/malloc.h>
51 #include <sys/mount.h>
52 #include <sys/namei.h>
53 #include <sys/proc.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 /*
72  * Mount null layer
73  */
74 static int
75 nullfs_mount(struct mount *mp)
76 {
77 	struct vnode *lowerrootvp;
78 	struct vnode *nullm_rootvp;
79 	struct null_mount *xmp;
80 	struct null_node *nn;
81 	struct nameidata nd, *ndp;
82 	char *target;
83 	int error, len;
84 	bool isvnunlocked;
85 
86 	NULLFSDEBUG("nullfs_mount(mp = %p)\n", (void *)mp);
87 
88 	if (mp->mnt_flag & MNT_ROOTFS)
89 		return (EOPNOTSUPP);
90 
91 	/*
92 	 * Update is a no-op
93 	 */
94 	if (mp->mnt_flag & MNT_UPDATE) {
95 		/*
96 		 * Only support update mounts for NFS export.
97 		 */
98 		if (vfs_flagopt(mp->mnt_optnew, "export", NULL, 0))
99 			return (0);
100 		else
101 			return (EOPNOTSUPP);
102 	}
103 
104 	/*
105 	 * Get argument
106 	 */
107 	error = vfs_getopt(mp->mnt_optnew, "from", (void **)&target, &len);
108 	if (error != 0)
109 		error = vfs_getopt(mp->mnt_optnew, "target", (void **)&target, &len);
110 	if (error || target[len - 1] != '\0')
111 		return (EINVAL);
112 
113 	/*
114 	 * Unlock lower node to avoid possible deadlock.
115 	 */
116 	if (mp->mnt_vnodecovered->v_op == &null_vnodeops &&
117 	    VOP_ISLOCKED(mp->mnt_vnodecovered) == LK_EXCLUSIVE) {
118 		VOP_UNLOCK(mp->mnt_vnodecovered);
119 		isvnunlocked = true;
120 	} else {
121 		isvnunlocked = false;
122 	}
123 
124 	/*
125 	 * Find lower node
126 	 */
127 	ndp = &nd;
128 	NDINIT(ndp, LOOKUP, FOLLOW|LOCKLEAF, UIO_SYSSPACE, target);
129 	error = namei(ndp);
130 
131 	/*
132 	 * Re-lock vnode.
133 	 * XXXKIB This is deadlock-prone as well.
134 	 */
135 	if (isvnunlocked)
136 		vn_lock(mp->mnt_vnodecovered, LK_EXCLUSIVE | LK_RETRY);
137 
138 	if (error)
139 		return (error);
140 	NDFREE(ndp, NDF_ONLY_PNBUF);
141 
142 	/*
143 	 * Sanity check on lower vnode
144 	 */
145 	lowerrootvp = ndp->ni_vp;
146 
147 	/*
148 	 * Check multi null mount to avoid `lock against myself' panic.
149 	 */
150 	if (mp->mnt_vnodecovered->v_op == &null_vnodeops) {
151 		nn = VTONULL(mp->mnt_vnodecovered);
152 		if (nn == NULL || lowerrootvp == nn->null_lowervp) {
153 			NULLFSDEBUG("nullfs_mount: multi null mount?\n");
154 			vput(lowerrootvp);
155 			return (EDEADLK);
156 		}
157 	}
158 
159 	xmp = (struct null_mount *) malloc(sizeof(struct null_mount),
160 	    M_NULLFSMNT, M_WAITOK | M_ZERO);
161 
162 	/*
163 	 * Save pointer to underlying FS and the reference to the
164 	 * lower root vnode.
165 	 */
166 	xmp->nullm_vfs = vfs_register_upper_from_vp(lowerrootvp, mp,
167 	    &xmp->upper_node);
168 	if (xmp->nullm_vfs == NULL) {
169 		vput(lowerrootvp);
170 		free(xmp, M_NULLFSMNT);
171 		return (ENOENT);
172 	}
173 	vref(lowerrootvp);
174 	xmp->nullm_lowerrootvp = lowerrootvp;
175 	mp->mnt_data = xmp;
176 
177 	/*
178 	 * Make sure the node alias worked.
179 	 */
180 	error = null_nodeget(mp, lowerrootvp, &nullm_rootvp);
181 	if (error != 0) {
182 		vfs_unregister_upper(xmp->nullm_vfs, &xmp->upper_node);
183 		vrele(lowerrootvp);
184 		free(xmp, M_NULLFSMNT);
185 		return (error);
186 	}
187 
188 	if (NULLVPTOLOWERVP(nullm_rootvp)->v_mount->mnt_flag & MNT_LOCAL) {
189 		MNT_ILOCK(mp);
190 		mp->mnt_flag |= MNT_LOCAL;
191 		MNT_IUNLOCK(mp);
192 	}
193 
194 	xmp->nullm_flags |= NULLM_CACHE;
195 	if (vfs_getopt(mp->mnt_optnew, "nocache", NULL, NULL) == 0 ||
196 	    (xmp->nullm_vfs->mnt_kern_flag & MNTK_NULL_NOCACHE) != 0)
197 		xmp->nullm_flags &= ~NULLM_CACHE;
198 
199 	if ((xmp->nullm_flags & NULLM_CACHE) != 0) {
200 		vfs_register_for_notification(xmp->nullm_vfs, mp,
201 		    &xmp->notify_node);
202 	}
203 
204 	MNT_ILOCK(mp);
205 	if ((xmp->nullm_flags & NULLM_CACHE) != 0) {
206 		mp->mnt_kern_flag |= lowerrootvp->v_mount->mnt_kern_flag &
207 		    (MNTK_SHARED_WRITES | MNTK_LOOKUP_SHARED |
208 		    MNTK_EXTENDED_SHARED);
209 	}
210 	mp->mnt_kern_flag |= MNTK_LOOKUP_EXCL_DOTDOT | MNTK_NOMSYNC;
211 	mp->mnt_kern_flag |= lowerrootvp->v_mount->mnt_kern_flag &
212 	    (MNTK_USES_BCACHE | MNTK_NO_IOPF | MNTK_UNMAPPED_BUFS);
213 	MNT_IUNLOCK(mp);
214 	vfs_getnewfsid(mp);
215 	vfs_mountedfrom(mp, target);
216 	vput(nullm_rootvp);
217 
218 	NULLFSDEBUG("nullfs_mount: lower %s, alias at %s\n",
219 		mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntonname);
220 	return (0);
221 }
222 
223 /*
224  * Free reference to null layer
225  */
226 static int
227 nullfs_unmount(mp, mntflags)
228 	struct mount *mp;
229 	int mntflags;
230 {
231 	struct null_mount *mntdata;
232 	int error, flags;
233 
234 	NULLFSDEBUG("nullfs_unmount: mp = %p\n", (void *)mp);
235 
236 	if (mntflags & MNT_FORCE)
237 		flags = FORCECLOSE;
238 	else
239 		flags = 0;
240 
241 	for (;;) {
242 		/* There is 1 extra root vnode reference (nullm_rootvp). */
243 		error = vflush(mp, 0, flags, curthread);
244 		if (error)
245 			return (error);
246 		MNT_ILOCK(mp);
247 		if (mp->mnt_nvnodelistsize == 0) {
248 			MNT_IUNLOCK(mp);
249 			break;
250 		}
251 		MNT_IUNLOCK(mp);
252 		if ((mntflags & MNT_FORCE) == 0)
253 			return (EBUSY);
254 	}
255 
256 	/*
257 	 * Finally, throw away the null_mount structure
258 	 */
259 	mntdata = mp->mnt_data;
260 	if ((mntdata->nullm_flags & NULLM_CACHE) != 0) {
261 		vfs_unregister_for_notification(mntdata->nullm_vfs,
262 		    &mntdata->notify_node);
263 	}
264 	vfs_unregister_upper(mntdata->nullm_vfs, &mntdata->upper_node);
265 	vrele(mntdata->nullm_lowerrootvp);
266 	mp->mnt_data = NULL;
267 	free(mntdata, M_NULLFSMNT);
268 	return (0);
269 }
270 
271 static int
272 nullfs_root(mp, flags, vpp)
273 	struct mount *mp;
274 	int flags;
275 	struct vnode **vpp;
276 {
277 	struct vnode *vp;
278 	struct null_mount *mntdata;
279 	int error;
280 
281 	mntdata = MOUNTTONULLMOUNT(mp);
282 	NULLFSDEBUG("nullfs_root(mp = %p, vp = %p)\n", mp,
283 	    mntdata->nullm_lowerrootvp);
284 
285 	error = vget(mntdata->nullm_lowerrootvp, flags);
286 	if (error == 0) {
287 		error = null_nodeget(mp, mntdata->nullm_lowerrootvp, &vp);
288 		if (error == 0) {
289 			*vpp = vp;
290 		}
291 	}
292 	return (error);
293 }
294 
295 static int
296 nullfs_quotactl(mp, cmd, uid, arg, mp_busy)
297 	struct mount *mp;
298 	int cmd;
299 	uid_t uid;
300 	void *arg;
301 	bool *mp_busy;
302 {
303 	struct mount *lowermp;
304 	struct null_mount *mntdata;
305 	int error;
306 	bool unbusy;
307 
308 	mntdata = MOUNTTONULLMOUNT(mp);
309 	lowermp = atomic_load_ptr(&mntdata->nullm_vfs);
310 	KASSERT(*mp_busy == true, ("upper mount not busy"));
311 	/*
312 	 * See comment in sys_quotactl() for an explanation of why the
313 	 * lower mount needs to be busied by the caller of VFS_QUOTACTL()
314 	 * but may be unbusied by the implementation.  We must unbusy
315 	 * the upper mount for the same reason; otherwise a namei lookup
316 	 * issued by the VFS_QUOTACTL() implementation could traverse the
317 	 * upper mount and deadlock.
318 	 */
319 	vfs_unbusy(mp);
320 	*mp_busy = false;
321 	unbusy = true;
322 	error = vfs_busy(lowermp, 0);
323 	if (error == 0)
324 		error = VFS_QUOTACTL(lowermp, cmd, uid, arg, &unbusy);
325 	if (unbusy)
326 		vfs_unbusy(lowermp);
327 
328 	return (error);
329 }
330 
331 static int
332 nullfs_statfs(mp, sbp)
333 	struct mount *mp;
334 	struct statfs *sbp;
335 {
336 	int error;
337 	struct statfs *mstat;
338 
339 	NULLFSDEBUG("nullfs_statfs(mp = %p, vp = %p->%p)\n", (void *)mp,
340 	    (void *)MOUNTTONULLMOUNT(mp)->nullm_rootvp,
341 	    (void *)NULLVPTOLOWERVP(MOUNTTONULLMOUNT(mp)->nullm_rootvp));
342 
343 	mstat = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK | M_ZERO);
344 
345 	error = VFS_STATFS(MOUNTTONULLMOUNT(mp)->nullm_vfs, mstat);
346 	if (error) {
347 		free(mstat, M_STATFS);
348 		return (error);
349 	}
350 
351 	/* now copy across the "interesting" information and fake the rest */
352 	sbp->f_type = mstat->f_type;
353 	sbp->f_flags = (sbp->f_flags & (MNT_RDONLY | MNT_NOEXEC | MNT_NOSUID |
354 	    MNT_UNION | MNT_NOSYMFOLLOW | MNT_AUTOMOUNTED)) |
355 	    (mstat->f_flags & ~(MNT_ROOTFS | MNT_AUTOMOUNTED));
356 	sbp->f_bsize = mstat->f_bsize;
357 	sbp->f_iosize = mstat->f_iosize;
358 	sbp->f_blocks = mstat->f_blocks;
359 	sbp->f_bfree = mstat->f_bfree;
360 	sbp->f_bavail = mstat->f_bavail;
361 	sbp->f_files = mstat->f_files;
362 	sbp->f_ffree = mstat->f_ffree;
363 
364 	free(mstat, M_STATFS);
365 	return (0);
366 }
367 
368 static int
369 nullfs_sync(mp, waitfor)
370 	struct mount *mp;
371 	int waitfor;
372 {
373 	/*
374 	 * XXX - Assumes no data cached at null layer.
375 	 */
376 	return (0);
377 }
378 
379 static int
380 nullfs_vget(mp, ino, flags, vpp)
381 	struct mount *mp;
382 	ino_t ino;
383 	int flags;
384 	struct vnode **vpp;
385 {
386 	int error;
387 
388 	KASSERT((flags & LK_TYPE_MASK) != 0,
389 	    ("nullfs_vget: no lock requested"));
390 
391 	error = VFS_VGET(MOUNTTONULLMOUNT(mp)->nullm_vfs, ino, flags, vpp);
392 	if (error != 0)
393 		return (error);
394 	return (null_nodeget(mp, *vpp, vpp));
395 }
396 
397 static int
398 nullfs_fhtovp(mp, fidp, flags, vpp)
399 	struct mount *mp;
400 	struct fid *fidp;
401 	int flags;
402 	struct vnode **vpp;
403 {
404 	int error;
405 
406 	error = VFS_FHTOVP(MOUNTTONULLMOUNT(mp)->nullm_vfs, fidp, flags,
407 	    vpp);
408 	if (error != 0)
409 		return (error);
410 	return (null_nodeget(mp, *vpp, vpp));
411 }
412 
413 static int
414 nullfs_extattrctl(mp, cmd, filename_vp, namespace, attrname)
415 	struct mount *mp;
416 	int cmd;
417 	struct vnode *filename_vp;
418 	int namespace;
419 	const char *attrname;
420 {
421 
422 	return (VFS_EXTATTRCTL(MOUNTTONULLMOUNT(mp)->nullm_vfs, cmd,
423 	    filename_vp, namespace, attrname));
424 }
425 
426 static void
427 nullfs_reclaim_lowervp(struct mount *mp, struct vnode *lowervp)
428 {
429 	struct vnode *vp;
430 
431 	vp = null_hashget(mp, lowervp);
432 	if (vp == NULL)
433 		return;
434 	VTONULL(vp)->null_flags |= NULLV_NOUNLOCK;
435 	vgone(vp);
436 	vput(vp);
437 }
438 
439 static void
440 nullfs_unlink_lowervp(struct mount *mp, struct vnode *lowervp)
441 {
442 	struct vnode *vp;
443 	struct null_node *xp;
444 
445 	vp = null_hashget(mp, lowervp);
446 	if (vp == NULL)
447 		return;
448 	xp = VTONULL(vp);
449 	xp->null_flags |= NULLV_DROP | NULLV_NOUNLOCK;
450 	vhold(vp);
451 	vunref(vp);
452 
453 	if (vp->v_usecount == 0) {
454 		/*
455 		 * If vunref() dropped the last use reference on the
456 		 * nullfs vnode, it must be reclaimed, and its lock
457 		 * was split from the lower vnode lock.  Need to do
458 		 * extra unlock before allowing the final vdrop() to
459 		 * free the vnode.
460 		 */
461 		KASSERT(VN_IS_DOOMED(vp),
462 		    ("not reclaimed nullfs vnode %p", vp));
463 		VOP_UNLOCK(vp);
464 	} else {
465 		/*
466 		 * Otherwise, the nullfs vnode still shares the lock
467 		 * with the lower vnode, and must not be unlocked.
468 		 * Also clear the NULLV_NOUNLOCK, the flag is not
469 		 * relevant for future reclamations.
470 		 */
471 		ASSERT_VOP_ELOCKED(vp, "unlink_lowervp");
472 		KASSERT(!VN_IS_DOOMED(vp),
473 		    ("reclaimed nullfs vnode %p", vp));
474 		xp->null_flags &= ~NULLV_NOUNLOCK;
475 	}
476 	vdrop(vp);
477 }
478 
479 static struct vfsops null_vfsops = {
480 	.vfs_extattrctl =	nullfs_extattrctl,
481 	.vfs_fhtovp =		nullfs_fhtovp,
482 	.vfs_init =		nullfs_init,
483 	.vfs_mount =		nullfs_mount,
484 	.vfs_quotactl =		nullfs_quotactl,
485 	.vfs_root =		nullfs_root,
486 	.vfs_statfs =		nullfs_statfs,
487 	.vfs_sync =		nullfs_sync,
488 	.vfs_uninit =		nullfs_uninit,
489 	.vfs_unmount =		nullfs_unmount,
490 	.vfs_vget =		nullfs_vget,
491 	.vfs_reclaim_lowervp =	nullfs_reclaim_lowervp,
492 	.vfs_unlink_lowervp =	nullfs_unlink_lowervp,
493 };
494 
495 VFS_SET(null_vfsops, nullfs, VFCF_LOOPBACK | VFCF_JAIL);
496