xref: /freebsd/sys/fs/nullfs/null_vfsops.c (revision 8c2dd68caa963f1900a8228b0732b04f5d530ffa)
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, curthread);
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_pin_from_vp(lowerrootvp);
167 	if (xmp->nullm_vfs == NULL) {
168 		vput(lowerrootvp);
169 		free(xmp, M_NULLFSMNT);
170 		return (ENOENT);
171 	}
172 	vref(lowerrootvp);
173 	xmp->nullm_lowerrootvp = lowerrootvp;
174 	mp->mnt_data = xmp;
175 
176 	/*
177 	 * Make sure the node alias worked.
178 	 */
179 	error = null_nodeget(mp, lowerrootvp, &nullm_rootvp);
180 	if (error != 0) {
181 		vfs_unpin(xmp->nullm_vfs);
182 		vrele(lowerrootvp);
183 		free(xmp, M_NULLFSMNT);
184 		return (error);
185 	}
186 
187 	if (NULLVPTOLOWERVP(nullm_rootvp)->v_mount->mnt_flag & MNT_LOCAL) {
188 		MNT_ILOCK(mp);
189 		mp->mnt_flag |= MNT_LOCAL;
190 		MNT_IUNLOCK(mp);
191 	}
192 
193 	xmp->nullm_flags |= NULLM_CACHE;
194 	if (vfs_getopt(mp->mnt_optnew, "nocache", NULL, NULL) == 0 ||
195 	    (xmp->nullm_vfs->mnt_kern_flag & MNTK_NULL_NOCACHE) != 0)
196 		xmp->nullm_flags &= ~NULLM_CACHE;
197 
198 	MNT_ILOCK(mp);
199 	if ((xmp->nullm_flags & NULLM_CACHE) != 0) {
200 		mp->mnt_kern_flag |= lowerrootvp->v_mount->mnt_kern_flag &
201 		    (MNTK_SHARED_WRITES | MNTK_LOOKUP_SHARED |
202 		    MNTK_EXTENDED_SHARED);
203 	}
204 	mp->mnt_kern_flag |= MNTK_LOOKUP_EXCL_DOTDOT | MNTK_NOMSYNC;
205 	mp->mnt_kern_flag |= lowerrootvp->v_mount->mnt_kern_flag &
206 	    (MNTK_USES_BCACHE | MNTK_NO_IOPF | MNTK_UNMAPPED_BUFS);
207 	MNT_IUNLOCK(mp);
208 	vfs_getnewfsid(mp);
209 	if ((xmp->nullm_flags & NULLM_CACHE) != 0) {
210 		MNT_ILOCK(xmp->nullm_vfs);
211 		TAILQ_INSERT_TAIL(&xmp->nullm_vfs->mnt_uppers, mp,
212 		    mnt_upper_link);
213 		MNT_IUNLOCK(xmp->nullm_vfs);
214 	}
215 
216 	vfs_mountedfrom(mp, target);
217 	vput(nullm_rootvp);
218 
219 	NULLFSDEBUG("nullfs_mount: lower %s, alias at %s\n",
220 		mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntonname);
221 	return (0);
222 }
223 
224 /*
225  * Free reference to null layer
226  */
227 static int
228 nullfs_unmount(mp, mntflags)
229 	struct mount *mp;
230 	int mntflags;
231 {
232 	struct null_mount *mntdata;
233 	struct mount *ump;
234 	int error, flags;
235 
236 	NULLFSDEBUG("nullfs_unmount: mp = %p\n", (void *)mp);
237 
238 	if (mntflags & MNT_FORCE)
239 		flags = FORCECLOSE;
240 	else
241 		flags = 0;
242 
243 	for (;;) {
244 		/* There is 1 extra root vnode reference (nullm_rootvp). */
245 		error = vflush(mp, 0, flags, curthread);
246 		if (error)
247 			return (error);
248 		MNT_ILOCK(mp);
249 		if (mp->mnt_nvnodelistsize == 0) {
250 			MNT_IUNLOCK(mp);
251 			break;
252 		}
253 		MNT_IUNLOCK(mp);
254 		if ((mntflags & MNT_FORCE) == 0)
255 			return (EBUSY);
256 	}
257 
258 	/*
259 	 * Finally, throw away the null_mount structure
260 	 */
261 	mntdata = mp->mnt_data;
262 	ump = mntdata->nullm_vfs;
263 	if ((mntdata->nullm_flags & NULLM_CACHE) != 0) {
264 		MNT_ILOCK(ump);
265 		while ((ump->mnt_kern_flag & MNTK_VGONE_UPPER) != 0) {
266 			ump->mnt_kern_flag |= MNTK_VGONE_WAITER;
267 			msleep(&ump->mnt_uppers, &ump->mnt_mtx, 0, "vgnupw", 0);
268 		}
269 		TAILQ_REMOVE(&ump->mnt_uppers, mp, mnt_upper_link);
270 		MNT_IUNLOCK(ump);
271 	}
272 	vfs_unpin(ump);
273 	vrele(mntdata->nullm_lowerrootvp);
274 	mp->mnt_data = NULL;
275 	free(mntdata, M_NULLFSMNT);
276 	return (0);
277 }
278 
279 static int
280 nullfs_root(mp, flags, vpp)
281 	struct mount *mp;
282 	int flags;
283 	struct vnode **vpp;
284 {
285 	struct vnode *vp;
286 	struct null_mount *mntdata;
287 	int error;
288 
289 	mntdata = MOUNTTONULLMOUNT(mp);
290 	NULLFSDEBUG("nullfs_root(mp = %p, vp = %p)\n", mp,
291 	    mntdata->nullm_lowerrootvp);
292 
293 	error = vget(mntdata->nullm_lowerrootvp, flags);
294 	if (error == 0) {
295 		error = null_nodeget(mp, mntdata->nullm_lowerrootvp, &vp);
296 		if (error == 0) {
297 			*vpp = vp;
298 		}
299 	}
300 	return (error);
301 }
302 
303 static int
304 nullfs_quotactl(mp, cmd, uid, arg, mp_busy)
305 	struct mount *mp;
306 	int cmd;
307 	uid_t uid;
308 	void *arg;
309 	bool *mp_busy;
310 {
311 	struct mount *lowermp;
312 	struct null_mount *mntdata;
313 	int error;
314 	bool unbusy;
315 
316 	mntdata = MOUNTTONULLMOUNT(mp);
317 	lowermp = atomic_load_ptr(&mntdata->nullm_vfs);
318 	KASSERT(*mp_busy == true, ("upper mount not busy"));
319 	/*
320 	 * See comment in sys_quotactl() for an explanation of why the
321 	 * lower mount needs to be busied by the caller of VFS_QUOTACTL()
322 	 * but may be unbusied by the implementation.  We must unbusy
323 	 * the upper mount for the same reason; otherwise a namei lookup
324 	 * issued by the VFS_QUOTACTL() implementation could traverse the
325 	 * upper mount and deadlock.
326 	 */
327 	vfs_unbusy(mp);
328 	*mp_busy = false;
329 	unbusy = true;
330 	error = vfs_busy(lowermp, 0);
331 	if (error == 0)
332 		error = VFS_QUOTACTL(lowermp, cmd, uid, arg, &unbusy);
333 	if (unbusy)
334 		vfs_unbusy(lowermp);
335 
336 	return (error);
337 }
338 
339 static int
340 nullfs_statfs(mp, sbp)
341 	struct mount *mp;
342 	struct statfs *sbp;
343 {
344 	int error;
345 	struct statfs *mstat;
346 
347 	NULLFSDEBUG("nullfs_statfs(mp = %p, vp = %p->%p)\n", (void *)mp,
348 	    (void *)MOUNTTONULLMOUNT(mp)->nullm_rootvp,
349 	    (void *)NULLVPTOLOWERVP(MOUNTTONULLMOUNT(mp)->nullm_rootvp));
350 
351 	mstat = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK | M_ZERO);
352 
353 	error = VFS_STATFS(MOUNTTONULLMOUNT(mp)->nullm_vfs, mstat);
354 	if (error) {
355 		free(mstat, M_STATFS);
356 		return (error);
357 	}
358 
359 	/* now copy across the "interesting" information and fake the rest */
360 	sbp->f_type = mstat->f_type;
361 	sbp->f_flags = (sbp->f_flags & (MNT_RDONLY | MNT_NOEXEC | MNT_NOSUID |
362 	    MNT_UNION | MNT_NOSYMFOLLOW | MNT_AUTOMOUNTED)) |
363 	    (mstat->f_flags & ~(MNT_ROOTFS | MNT_AUTOMOUNTED));
364 	sbp->f_bsize = mstat->f_bsize;
365 	sbp->f_iosize = mstat->f_iosize;
366 	sbp->f_blocks = mstat->f_blocks;
367 	sbp->f_bfree = mstat->f_bfree;
368 	sbp->f_bavail = mstat->f_bavail;
369 	sbp->f_files = mstat->f_files;
370 	sbp->f_ffree = mstat->f_ffree;
371 
372 	free(mstat, M_STATFS);
373 	return (0);
374 }
375 
376 static int
377 nullfs_sync(mp, waitfor)
378 	struct mount *mp;
379 	int waitfor;
380 {
381 	/*
382 	 * XXX - Assumes no data cached at null layer.
383 	 */
384 	return (0);
385 }
386 
387 static int
388 nullfs_vget(mp, ino, flags, vpp)
389 	struct mount *mp;
390 	ino_t ino;
391 	int flags;
392 	struct vnode **vpp;
393 {
394 	int error;
395 
396 	KASSERT((flags & LK_TYPE_MASK) != 0,
397 	    ("nullfs_vget: no lock requested"));
398 
399 	error = VFS_VGET(MOUNTTONULLMOUNT(mp)->nullm_vfs, ino, flags, vpp);
400 	if (error != 0)
401 		return (error);
402 	return (null_nodeget(mp, *vpp, vpp));
403 }
404 
405 static int
406 nullfs_fhtovp(mp, fidp, flags, vpp)
407 	struct mount *mp;
408 	struct fid *fidp;
409 	int flags;
410 	struct vnode **vpp;
411 {
412 	int error;
413 
414 	error = VFS_FHTOVP(MOUNTTONULLMOUNT(mp)->nullm_vfs, fidp, flags,
415 	    vpp);
416 	if (error != 0)
417 		return (error);
418 	return (null_nodeget(mp, *vpp, vpp));
419 }
420 
421 static int
422 nullfs_extattrctl(mp, cmd, filename_vp, namespace, attrname)
423 	struct mount *mp;
424 	int cmd;
425 	struct vnode *filename_vp;
426 	int namespace;
427 	const char *attrname;
428 {
429 
430 	return (VFS_EXTATTRCTL(MOUNTTONULLMOUNT(mp)->nullm_vfs, cmd,
431 	    filename_vp, namespace, attrname));
432 }
433 
434 static void
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
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);
504