xref: /freebsd/sys/fs/nullfs/null_vfsops.c (revision e40c72c823c6502906d4024933f0c119b5ca17cf)
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 
35 /*
36  * Null Layer
37  * (See null_vnops.c for a description of what this does.)
38  */
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/fcntl.h>
43 #include <sys/kernel.h>
44 #include <sys/lock.h>
45 #include <sys/malloc.h>
46 #include <sys/mount.h>
47 #include <sys/namei.h>
48 #include <sys/proc.h>
49 #include <sys/sysctl.h>
50 #include <sys/vnode.h>
51 #include <sys/jail.h>
52 
53 #include <fs/nullfs/null.h>
54 
55 static MALLOC_DEFINE(M_NULLFSMNT, "nullfs_mount", "NULLFS mount structure");
56 
57 static vfs_fhtovp_t	nullfs_fhtovp;
58 static vfs_mount_t	nullfs_mount;
59 static vfs_quotactl_t	nullfs_quotactl;
60 static vfs_root_t	nullfs_root;
61 static vfs_sync_t	nullfs_sync;
62 static vfs_statfs_t	nullfs_statfs;
63 static vfs_unmount_t	nullfs_unmount;
64 static vfs_vget_t	nullfs_vget;
65 static vfs_extattrctl_t	nullfs_extattrctl;
66 
67 SYSCTL_NODE(_vfs, OID_AUTO, nullfs, CTLFLAG_RW, 0, "nullfs");
68 
69 static bool null_cache_vnodes = true;
70 SYSCTL_BOOL(_vfs_nullfs, OID_AUTO, cache_vnodes, CTLFLAG_RWTUN,
71     &null_cache_vnodes, 0,
72     "cache free nullfs vnodes");
73 
74 /*
75  * Mount null layer
76  */
77 static int
78 nullfs_mount(struct mount *mp)
79 {
80 	struct vnode *lowerrootvp;
81 	struct vnode *nullm_rootvp;
82 	struct null_mount *xmp;
83 	struct null_node *nn;
84 	struct nameidata nd, *ndp;
85 	char *target;
86 	int error, len;
87 	bool isvnunlocked;
88 	static const char cache_opt_name[] = "cache";
89 	static const char nocache_opt_name[] = "nocache";
90 	static const char unixbypass_opt_name[] = "unixbypass";
91 	static const char nounixbypass_opt_name[] = "nounixbypass";
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 (null_is_nullfs_vnode(mp->mnt_vnodecovered) &&
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 	 * Do not allow to mount a vnode over itself.
156 	 */
157 	if (mp->mnt_vnodecovered == lowerrootvp) {
158 		vput(lowerrootvp);
159 		return (EDEADLK);
160 	}
161 
162 	/*
163 	 * Check multi null mount to avoid `lock against myself' panic.
164 	 */
165 	if (null_is_nullfs_vnode(mp->mnt_vnodecovered)) {
166 		nn = VTONULL(mp->mnt_vnodecovered);
167 		if (nn == NULL || lowerrootvp == nn->null_lowervp) {
168 			NULLFSDEBUG("nullfs_mount: multi null mount?\n");
169 			vput(lowerrootvp);
170 			return (EDEADLK);
171 		}
172 	}
173 
174 	/*
175 	 * Lower vnode must be the same type as the covered vnode - we
176 	 * don't allow mounting directories to files or vice versa.
177 	 */
178 	if ((lowerrootvp->v_type != VDIR && lowerrootvp->v_type != VREG &&
179 	    lowerrootvp->v_type != VSOCK) ||
180 	    lowerrootvp->v_type != mp->mnt_vnodecovered->v_type) {
181 		NULLFSDEBUG("nullfs_mount: target must be same type as fspath");
182 		vput(lowerrootvp);
183 		return (EINVAL);
184 	}
185 
186 	xmp = malloc(sizeof(struct null_mount), M_NULLFSMNT,
187 	    M_WAITOK | M_ZERO);
188 
189 	/*
190 	 * Save pointer to underlying FS and the reference to the
191 	 * lower root vnode.
192 	 */
193 	xmp->nullm_vfs = vfs_register_upper_from_vp(lowerrootvp, mp,
194 	    &xmp->upper_node);
195 	if (xmp->nullm_vfs == NULL) {
196 		vput(lowerrootvp);
197 		free(xmp, M_NULLFSMNT);
198 		return (ENOENT);
199 	}
200 	vref(lowerrootvp);
201 	xmp->nullm_lowerrootvp = lowerrootvp;
202 	mp->mnt_data = xmp;
203 
204 	/*
205 	 * Make sure the node alias worked.
206 	 */
207 	error = null_nodeget(mp, lowerrootvp, &nullm_rootvp);
208 	if (error != 0) {
209 		vfs_unregister_upper(xmp->nullm_vfs, &xmp->upper_node);
210 		vrele(lowerrootvp);
211 		free(xmp, M_NULLFSMNT);
212 		return (error);
213 	}
214 
215 	if (NULLVPTOLOWERVP(nullm_rootvp)->v_mount->mnt_flag & MNT_LOCAL) {
216 		MNT_ILOCK(mp);
217 		mp->mnt_flag |= MNT_LOCAL;
218 		MNT_IUNLOCK(mp);
219 	}
220 
221 	if (vfs_getopt(mp->mnt_optnew, cache_opt_name, NULL, NULL) == 0) {
222 		xmp->nullm_flags |= NULLM_CACHE;
223 	} else if (vfs_getopt(mp->mnt_optnew, nocache_opt_name, NULL,
224 	    NULL) == 0) {
225 		;
226 	} else if (null_cache_vnodes &&
227 	    (xmp->nullm_vfs->mnt_kern_flag & MNTK_NULL_NOCACHE) == 0) {
228 		xmp->nullm_flags |= NULLM_CACHE;
229 	}
230 
231 	if ((xmp->nullm_flags & NULLM_CACHE) != 0) {
232 		vfs_register_for_notification(xmp->nullm_vfs, mp,
233 		    &xmp->notify_node);
234 	}
235 
236 	if (vfs_getopt(mp->mnt_optnew, unixbypass_opt_name, NULL, NULL) == 0) {
237 		;
238 	} else if (vfs_getopt(mp->mnt_optnew, nounixbypass_opt_name, NULL,
239 	    NULL) == 0) {
240 		xmp->nullm_flags |= NULLM_NOUNPBYPASS;
241 	}
242 
243 	if (lowerrootvp == mp->mnt_vnodecovered) {
244 		vn_lock(lowerrootvp, LK_EXCLUSIVE | LK_RETRY | LK_CANRECURSE);
245 		lowerrootvp->v_vflag |= VV_CROSSLOCK;
246 		VOP_UNLOCK(lowerrootvp);
247 	}
248 
249 	MNT_ILOCK(mp);
250 	if ((xmp->nullm_flags & NULLM_CACHE) != 0) {
251 		mp->mnt_kern_flag |= lowerrootvp->v_mount->mnt_kern_flag &
252 		    (MNTK_SHARED_WRITES | MNTK_LOOKUP_SHARED |
253 		    MNTK_EXTENDED_SHARED);
254 	}
255 	mp->mnt_kern_flag |= MNTK_NOMSYNC | MNTK_UNLOCKED_INSMNTQUE;
256 	mp->mnt_kern_flag |= lowerrootvp->v_mount->mnt_kern_flag &
257 	    (MNTK_USES_BCACHE | MNTK_NO_IOPF | MNTK_UNMAPPED_BUFS);
258 	MNT_IUNLOCK(mp);
259 	vfs_getnewfsid(mp);
260 	vfs_mountedfrom(mp, target);
261 	vput(nullm_rootvp);
262 
263 	NULLFSDEBUG("nullfs_mount: lower %s, alias at %s\n",
264 		mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntonname);
265 	return (0);
266 }
267 
268 /*
269  * Free reference to null layer
270  */
271 static int
272 nullfs_unmount(struct mount *mp, int mntflags)
273 {
274 	struct null_mount *mntdata;
275 	int error, flags;
276 
277 	NULLFSDEBUG("nullfs_unmount: mp = %p\n", (void *)mp);
278 
279 	if (mntflags & MNT_FORCE)
280 		flags = FORCECLOSE;
281 	else
282 		flags = 0;
283 
284 	for (;;) {
285 		/* There is 1 extra root vnode reference (nullm_rootvp). */
286 		error = vflush(mp, 0, flags, curthread);
287 		if (error)
288 			return (error);
289 		MNT_ILOCK(mp);
290 		if (mp->mnt_nvnodelistsize == 0) {
291 			MNT_IUNLOCK(mp);
292 			break;
293 		}
294 		MNT_IUNLOCK(mp);
295 		if ((mntflags & MNT_FORCE) == 0)
296 			return (EBUSY);
297 	}
298 
299 	/*
300 	 * Finally, throw away the null_mount structure
301 	 */
302 	mntdata = mp->mnt_data;
303 	if ((mntdata->nullm_flags & NULLM_CACHE) != 0) {
304 		vfs_unregister_for_notification(mntdata->nullm_vfs,
305 		    &mntdata->notify_node);
306 	}
307 	if (mntdata->nullm_lowerrootvp == mp->mnt_vnodecovered) {
308 		vn_lock(mp->mnt_vnodecovered, LK_EXCLUSIVE | LK_RETRY | LK_CANRECURSE);
309 		mp->mnt_vnodecovered->v_vflag &= ~VV_CROSSLOCK;
310 		VOP_UNLOCK(mp->mnt_vnodecovered);
311 	}
312 	vfs_unregister_upper(mntdata->nullm_vfs, &mntdata->upper_node);
313 	vrele(mntdata->nullm_lowerrootvp);
314 	mp->mnt_data = NULL;
315 	free(mntdata, M_NULLFSMNT);
316 	return (0);
317 }
318 
319 static int
320 nullfs_root(struct mount *mp, int flags, struct vnode **vpp)
321 {
322 	struct vnode *vp;
323 	struct null_mount *mntdata;
324 	int error;
325 
326 	mntdata = MOUNTTONULLMOUNT(mp);
327 	NULLFSDEBUG("nullfs_root(mp = %p, vp = %p)\n", mp,
328 	    mntdata->nullm_lowerrootvp);
329 
330 	error = vget(mntdata->nullm_lowerrootvp, flags);
331 	if (error == 0) {
332 		error = null_nodeget(mp, mntdata->nullm_lowerrootvp, &vp);
333 		if (error == 0) {
334 			*vpp = vp;
335 		}
336 	}
337 	return (error);
338 }
339 
340 static int
341 nullfs_quotactl(struct mount *mp, int cmd, uid_t uid, void *arg, bool *mp_busy)
342 {
343 	struct mount *lowermp;
344 	struct null_mount *mntdata;
345 	int error;
346 	bool unbusy;
347 
348 	mntdata = MOUNTTONULLMOUNT(mp);
349 	lowermp = atomic_load_ptr(&mntdata->nullm_vfs);
350 	KASSERT(*mp_busy == true, ("upper mount not busy"));
351 	/*
352 	 * See comment in sys_quotactl() for an explanation of why the
353 	 * lower mount needs to be busied by the caller of VFS_QUOTACTL()
354 	 * but may be unbusied by the implementation.  We must unbusy
355 	 * the upper mount for the same reason; otherwise a namei lookup
356 	 * issued by the VFS_QUOTACTL() implementation could traverse the
357 	 * upper mount and deadlock.
358 	 */
359 	vfs_unbusy(mp);
360 	*mp_busy = false;
361 	unbusy = true;
362 	error = vfs_busy(lowermp, 0);
363 	if (error == 0)
364 		error = VFS_QUOTACTL(lowermp, cmd, uid, arg, &unbusy);
365 	if (unbusy)
366 		vfs_unbusy(lowermp);
367 
368 	return (error);
369 }
370 
371 static int
372 nullfs_statfs(struct mount *mp, struct statfs *sbp)
373 {
374 	int error;
375 	struct statfs *mstat;
376 
377 	NULLFSDEBUG("nullfs_statfs(mp = %p, vp = %p->%p)\n", (void *)mp,
378 	    (void *)MOUNTTONULLMOUNT(mp)->nullm_rootvp,
379 	    (void *)NULLVPTOLOWERVP(MOUNTTONULLMOUNT(mp)->nullm_rootvp));
380 
381 	mstat = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK | M_ZERO);
382 
383 	error = VFS_STATFS(MOUNTTONULLMOUNT(mp)->nullm_vfs, mstat);
384 	if (error) {
385 		free(mstat, M_STATFS);
386 		return (error);
387 	}
388 
389 	sbp->f_type = mstat->f_type;
390 	sbp->f_bsize = mstat->f_bsize;
391 	sbp->f_iosize = mstat->f_iosize;
392 	sbp->f_blocks = mstat->f_blocks;
393 	sbp->f_bfree = mstat->f_bfree;
394 	sbp->f_bavail = mstat->f_bavail;
395 	sbp->f_files = mstat->f_files;
396 	sbp->f_ffree = mstat->f_ffree;
397 
398 	free(mstat, M_STATFS);
399 	return (0);
400 }
401 
402 static int
403 nullfs_sync(struct mount *mp, int waitfor)
404 {
405 	/*
406 	 * XXX - Assumes no data cached at null layer.
407 	 */
408 	return (0);
409 }
410 
411 static int
412 nullfs_vget(struct mount *mp, ino_t ino, int flags, struct vnode **vpp)
413 {
414 	int error;
415 
416 	KASSERT((flags & LK_TYPE_MASK) != 0,
417 	    ("nullfs_vget: no lock requested"));
418 
419 	error = VFS_VGET(MOUNTTONULLMOUNT(mp)->nullm_vfs, ino, flags, vpp);
420 	if (error != 0)
421 		return (error);
422 	return (null_nodeget(mp, *vpp, vpp));
423 }
424 
425 static int
426 nullfs_fhtovp(struct mount *mp, struct fid *fidp, int flags, struct vnode **vpp)
427 {
428 	int error;
429 
430 	error = VFS_FHTOVP(MOUNTTONULLMOUNT(mp)->nullm_vfs, fidp, flags,
431 	    vpp);
432 	if (error != 0)
433 		return (error);
434 	return (null_nodeget(mp, *vpp, vpp));
435 }
436 
437 static int
438 nullfs_extattrctl(struct mount *mp, int cmd, struct vnode *filename_vp,
439     int namespace, const char *attrname)
440 {
441 
442 	return (VFS_EXTATTRCTL(MOUNTTONULLMOUNT(mp)->nullm_vfs, cmd,
443 	    filename_vp, namespace, attrname));
444 }
445 
446 static void
447 nullfs_reclaim_lowervp(struct mount *mp, struct vnode *lowervp)
448 {
449 	struct vnode *vp;
450 
451 	vp = null_hashget(mp, lowervp);
452 	if (vp == NULL)
453 		return;
454 	VTONULL(vp)->null_flags |= NULLV_NOUNLOCK;
455 	vgone(vp);
456 	vput(vp);
457 }
458 
459 static void
460 nullfs_unlink_lowervp(struct mount *mp, struct vnode *lowervp)
461 {
462 	struct vnode *vp;
463 	struct null_node *xp;
464 
465 	vp = null_hashget(mp, lowervp);
466 	if (vp == NULL)
467 		return;
468 	xp = VTONULL(vp);
469 	xp->null_flags |= NULLV_DROP | NULLV_NOUNLOCK;
470 	vhold(vp);
471 	vunref(vp);
472 
473 	if (VN_IS_DOOMED(vp)) {
474 		/*
475 		 * If the vnode is doomed, its lock was split from the lower
476 		 * vnode lock.  Therefore we need to do an extra unlock before
477 		 * allowing the final vdrop() to free the vnode.
478 		 */
479 		VOP_UNLOCK(vp);
480 	} else {
481 		/*
482 		 * Otherwise, the nullfs vnode still shares the lock
483 		 * with the lower vnode, and must not be unlocked.
484 		 * Also clear the NULLV_NOUNLOCK, the flag is not
485 		 * relevant for future reclamations.
486 		 */
487 		ASSERT_VOP_ELOCKED(vp, "unlink_lowervp");
488 		xp->null_flags &= ~NULLV_NOUNLOCK;
489 	}
490 	vdrop(vp);
491 }
492 
493 static struct vfsops null_vfsops = {
494 	.vfs_extattrctl =	nullfs_extattrctl,
495 	.vfs_fhtovp =		nullfs_fhtovp,
496 	.vfs_init =		nullfs_init,
497 	.vfs_mount =		nullfs_mount,
498 	.vfs_quotactl =		nullfs_quotactl,
499 	.vfs_root =		nullfs_root,
500 	.vfs_statfs =		nullfs_statfs,
501 	.vfs_sync =		nullfs_sync,
502 	.vfs_uninit =		nullfs_uninit,
503 	.vfs_unmount =		nullfs_unmount,
504 	.vfs_vget =		nullfs_vget,
505 	.vfs_reclaim_lowervp =	nullfs_reclaim_lowervp,
506 	.vfs_unlink_lowervp =	nullfs_unlink_lowervp,
507 };
508 
509 VFS_SET(null_vfsops, nullfs, VFCF_LOOPBACK | VFCF_JAIL | VFCF_FILEMOUNT);
510