xref: /freebsd/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_ctldir.c (revision 59f5f100b774de8824fb2fc1a8a11a93bbc2dafd)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or https://opensource.org/licenses/CDDL-1.0.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright (c) 2012, 2015 by Delphix. All rights reserved.
24  * Copyright 2015, OmniTI Computer Consulting, Inc. All rights reserved.
25  */
26 
27 /*
28  * ZFS control directory (a.k.a. ".zfs")
29  *
30  * This directory provides a common location for all ZFS meta-objects.
31  * Currently, this is only the 'snapshot' directory, but this may expand in the
32  * future.  The elements are built using the GFS primitives, as the hierarchy
33  * does not actually exist on disk.
34  *
35  * For 'snapshot', we don't want to have all snapshots always mounted, because
36  * this would take up a huge amount of space in /etc/mnttab.  We have three
37  * types of objects:
38  *
39  * 	ctldir ------> snapshotdir -------> snapshot
40  *                                             |
41  *                                             |
42  *                                             V
43  *                                         mounted fs
44  *
45  * The 'snapshot' node contains just enough information to lookup '..' and act
46  * as a mountpoint for the snapshot.  Whenever we lookup a specific snapshot, we
47  * perform an automount of the underlying filesystem and return the
48  * corresponding vnode.
49  *
50  * All mounts are handled automatically by the kernel, but unmounts are
51  * (currently) handled from user land.  The main reason is that there is no
52  * reliable way to auto-unmount the filesystem when it's "no longer in use".
53  * When the user unmounts a filesystem, we call zfsctl_unmount(), which
54  * unmounts any snapshots within the snapshot directory.
55  *
56  * The '.zfs', '.zfs/snapshot', and all directories created under
57  * '.zfs/snapshot' (ie: '.zfs/snapshot/<snapname>') are all GFS nodes and
58  * share the same vfs_t as the head filesystem (what '.zfs' lives under).
59  *
60  * File systems mounted ontop of the GFS nodes '.zfs/snapshot/<snapname>'
61  * (ie: snapshots) are ZFS nodes and have their own unique vfs_t.
62  * However, vnodes within these mounted on file systems have their v_vfsp
63  * fields set to the head filesystem to make NFS happy (see
64  * zfsctl_snapdir_lookup()). We VFS_HOLD the head filesystem's vfs_t
65  * so that it cannot be freed until all snapshots have been unmounted.
66  */
67 
68 #include <sys/types.h>
69 #include <sys/param.h>
70 #include <sys/libkern.h>
71 #include <sys/dirent.h>
72 #include <sys/zfs_context.h>
73 #include <sys/zfs_ctldir.h>
74 #include <sys/zfs_ioctl.h>
75 #include <sys/zfs_vfsops.h>
76 #include <sys/namei.h>
77 #include <sys/stat.h>
78 #include <sys/dmu.h>
79 #include <sys/dsl_dataset.h>
80 #include <sys/dsl_destroy.h>
81 #include <sys/dsl_deleg.h>
82 #include <sys/mount.h>
83 #include <sys/zap.h>
84 #include <sys/sysproto.h>
85 
86 #include "zfs_namecheck.h"
87 
88 #include <sys/kernel.h>
89 #include <sys/ccompat.h>
90 
91 /* Common access mode for all virtual directories under the ctldir */
92 const uint16_t zfsctl_ctldir_mode = S_IRUSR | S_IXUSR | S_IRGRP | S_IXGRP |
93     S_IROTH | S_IXOTH;
94 
95 /*
96  * "Synthetic" filesystem implementation.
97  */
98 
99 /*
100  * Assert that A implies B.
101  */
102 #define	KASSERT_IMPLY(A, B, msg)	KASSERT(!(A) || (B), (msg));
103 
104 static MALLOC_DEFINE(M_SFSNODES, "sfs_nodes", "synthetic-fs nodes");
105 
106 typedef struct sfs_node {
107 	char		sn_name[ZFS_MAX_DATASET_NAME_LEN];
108 	uint64_t	sn_parent_id;
109 	uint64_t	sn_id;
110 } sfs_node_t;
111 
112 /*
113  * Check the parent's ID as well as the node's to account for a chance
114  * that IDs originating from different domains (snapshot IDs, artificial
115  * IDs, znode IDs) may clash.
116  */
117 static int
sfs_compare_ids(struct vnode * vp,void * arg)118 sfs_compare_ids(struct vnode *vp, void *arg)
119 {
120 	sfs_node_t *n1 = vp->v_data;
121 	sfs_node_t *n2 = arg;
122 	bool equal;
123 
124 	equal = n1->sn_id == n2->sn_id &&
125 	    n1->sn_parent_id == n2->sn_parent_id;
126 
127 	/* Zero means equality. */
128 	return (!equal);
129 }
130 
131 static int
sfs_vnode_get(const struct mount * mp,int flags,uint64_t parent_id,uint64_t id,struct vnode ** vpp)132 sfs_vnode_get(const struct mount *mp, int flags, uint64_t parent_id,
133     uint64_t id, struct vnode **vpp)
134 {
135 	sfs_node_t search;
136 	int err;
137 
138 	search.sn_id = id;
139 	search.sn_parent_id = parent_id;
140 	err = vfs_hash_get(mp, (uint32_t)id, flags, curthread, vpp,
141 	    sfs_compare_ids, &search);
142 	return (err);
143 }
144 
145 static int
sfs_vnode_insert(struct vnode * vp,int flags,uint64_t parent_id,uint64_t id,struct vnode ** vpp)146 sfs_vnode_insert(struct vnode *vp, int flags, uint64_t parent_id,
147     uint64_t id, struct vnode **vpp)
148 {
149 	int err;
150 
151 	KASSERT(vp->v_data != NULL, ("sfs_vnode_insert with NULL v_data"));
152 	err = vfs_hash_insert(vp, (uint32_t)id, flags, curthread, vpp,
153 	    sfs_compare_ids, vp->v_data);
154 	return (err);
155 }
156 
157 static void
sfs_vnode_remove(struct vnode * vp)158 sfs_vnode_remove(struct vnode *vp)
159 {
160 	vfs_hash_remove(vp);
161 }
162 
163 typedef void sfs_vnode_setup_fn(vnode_t *vp, void *arg);
164 
165 static int
sfs_vgetx(struct mount * mp,int flags,uint64_t parent_id,uint64_t id,const char * tag,struct vop_vector * vops,sfs_vnode_setup_fn setup,void * arg,struct vnode ** vpp)166 sfs_vgetx(struct mount *mp, int flags, uint64_t parent_id, uint64_t id,
167     const char *tag, struct vop_vector *vops,
168     sfs_vnode_setup_fn setup, void *arg,
169     struct vnode **vpp)
170 {
171 	struct vnode *vp;
172 	int error;
173 
174 	error = sfs_vnode_get(mp, flags, parent_id, id, vpp);
175 	if (error != 0 || *vpp != NULL) {
176 		KASSERT_IMPLY(error == 0, (*vpp)->v_data != NULL,
177 		    "sfs vnode with no data");
178 		return (error);
179 	}
180 
181 	/* Allocate a new vnode/inode. */
182 	error = getnewvnode(tag, mp, vops, &vp);
183 	if (error != 0) {
184 		*vpp = NULL;
185 		return (error);
186 	}
187 
188 	/*
189 	 * Exclusively lock the vnode vnode while it's being constructed.
190 	 */
191 	lockmgr(vp->v_vnlock, LK_EXCLUSIVE, NULL);
192 	error = insmntque(vp, mp);
193 	if (error != 0) {
194 		*vpp = NULL;
195 		return (error);
196 	}
197 
198 	setup(vp, arg);
199 
200 	error = sfs_vnode_insert(vp, flags, parent_id, id, vpp);
201 	if (error != 0 || *vpp != NULL) {
202 		KASSERT_IMPLY(error == 0, (*vpp)->v_data != NULL,
203 		    "sfs vnode with no data");
204 		return (error);
205 	}
206 
207 #if __FreeBSD_version >= 1400077
208 	vn_set_state(vp, VSTATE_CONSTRUCTED);
209 #endif
210 
211 	*vpp = vp;
212 	return (0);
213 }
214 
215 static void
sfs_print_node(sfs_node_t * node)216 sfs_print_node(sfs_node_t *node)
217 {
218 	printf("\tname = %s\n", node->sn_name);
219 	printf("\tparent_id = %ju\n", (uintmax_t)node->sn_parent_id);
220 	printf("\tid = %ju\n", (uintmax_t)node->sn_id);
221 }
222 
223 static sfs_node_t *
sfs_alloc_node(size_t size,const char * name,uint64_t parent_id,uint64_t id)224 sfs_alloc_node(size_t size, const char *name, uint64_t parent_id, uint64_t id)
225 {
226 	struct sfs_node *node;
227 
228 	KASSERT(strlen(name) < sizeof (node->sn_name),
229 	    ("sfs node name is too long"));
230 	KASSERT(size >= sizeof (*node), ("sfs node size is too small"));
231 	node = malloc(size, M_SFSNODES, M_WAITOK | M_ZERO);
232 	strlcpy(node->sn_name, name, sizeof (node->sn_name));
233 	node->sn_parent_id = parent_id;
234 	node->sn_id = id;
235 
236 	return (node);
237 }
238 
239 static void
sfs_destroy_node(sfs_node_t * node)240 sfs_destroy_node(sfs_node_t *node)
241 {
242 	free(node, M_SFSNODES);
243 }
244 
245 static void *
sfs_reclaim_vnode(vnode_t * vp)246 sfs_reclaim_vnode(vnode_t *vp)
247 {
248 	void *data;
249 
250 	sfs_vnode_remove(vp);
251 	data = vp->v_data;
252 	vp->v_data = NULL;
253 	return (data);
254 }
255 
256 static int
sfs_readdir_common(uint64_t parent_id,uint64_t id,struct vop_readdir_args * ap,zfs_uio_t * uio,off_t * offp)257 sfs_readdir_common(uint64_t parent_id, uint64_t id, struct vop_readdir_args *ap,
258     zfs_uio_t *uio, off_t *offp)
259 {
260 	struct dirent entry;
261 	int error;
262 
263 	/* Reset ncookies for subsequent use of vfs_read_dirent. */
264 	if (ap->a_ncookies != NULL)
265 		*ap->a_ncookies = 0;
266 
267 	if (zfs_uio_resid(uio) < sizeof (entry))
268 		return (SET_ERROR(EINVAL));
269 
270 	if (zfs_uio_offset(uio) < 0)
271 		return (SET_ERROR(EINVAL));
272 	if (zfs_uio_offset(uio) == 0) {
273 		entry.d_fileno = id;
274 		entry.d_type = DT_DIR;
275 		entry.d_name[0] = '.';
276 		entry.d_name[1] = '\0';
277 		entry.d_namlen = 1;
278 		entry.d_reclen = sizeof (entry);
279 		error = vfs_read_dirent(ap, &entry, zfs_uio_offset(uio));
280 		if (error != 0)
281 			return (SET_ERROR(error));
282 	}
283 
284 	if (zfs_uio_offset(uio) < sizeof (entry))
285 		return (SET_ERROR(EINVAL));
286 	if (zfs_uio_offset(uio) == sizeof (entry)) {
287 		entry.d_fileno = parent_id;
288 		entry.d_type = DT_DIR;
289 		entry.d_name[0] = '.';
290 		entry.d_name[1] = '.';
291 		entry.d_name[2] = '\0';
292 		entry.d_namlen = 2;
293 		entry.d_reclen = sizeof (entry);
294 		error = vfs_read_dirent(ap, &entry, zfs_uio_offset(uio));
295 		if (error != 0)
296 			return (SET_ERROR(error));
297 	}
298 
299 	if (offp != NULL)
300 		*offp = 2 * sizeof (entry);
301 	return (0);
302 }
303 
304 
305 /*
306  * .zfs inode namespace
307  *
308  * We need to generate unique inode numbers for all files and directories
309  * within the .zfs pseudo-filesystem.  We use the following scheme:
310  *
311  * 	ENTRY			ZFSCTL_INODE
312  * 	.zfs			1
313  * 	.zfs/snapshot		2
314  * 	.zfs/snapshot/<snap>	objectid(snap)
315  */
316 #define	ZFSCTL_INO_SNAP(id)	(id)
317 
318 static struct vop_vector zfsctl_ops_root;
319 static struct vop_vector zfsctl_ops_snapdir;
320 static struct vop_vector zfsctl_ops_snapshot;
321 
322 void
zfsctl_init(void)323 zfsctl_init(void)
324 {
325 }
326 
327 void
zfsctl_fini(void)328 zfsctl_fini(void)
329 {
330 }
331 
332 boolean_t
zfsctl_is_node(vnode_t * vp)333 zfsctl_is_node(vnode_t *vp)
334 {
335 	return (vn_matchops(vp, zfsctl_ops_root) ||
336 	    vn_matchops(vp, zfsctl_ops_snapdir) ||
337 	    vn_matchops(vp, zfsctl_ops_snapshot));
338 
339 }
340 
341 typedef struct zfsctl_root {
342 	sfs_node_t	node;
343 	sfs_node_t	*snapdir;
344 	timestruc_t	cmtime;
345 } zfsctl_root_t;
346 
347 
348 /*
349  * Create the '.zfs' directory.
350  */
351 void
zfsctl_create(zfsvfs_t * zfsvfs)352 zfsctl_create(zfsvfs_t *zfsvfs)
353 {
354 	zfsctl_root_t *dot_zfs;
355 	sfs_node_t *snapdir;
356 	vnode_t *rvp;
357 	uint64_t crtime[2];
358 
359 	ASSERT3P(zfsvfs->z_ctldir, ==, NULL);
360 
361 	snapdir = sfs_alloc_node(sizeof (*snapdir), "snapshot", ZFSCTL_INO_ROOT,
362 	    ZFSCTL_INO_SNAPDIR);
363 	dot_zfs = (zfsctl_root_t *)sfs_alloc_node(sizeof (*dot_zfs), ".zfs", 0,
364 	    ZFSCTL_INO_ROOT);
365 	dot_zfs->snapdir = snapdir;
366 
367 	VERIFY0(VFS_ROOT(zfsvfs->z_vfs, LK_EXCLUSIVE, &rvp));
368 	VERIFY0(sa_lookup(VTOZ(rvp)->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
369 	    &crtime, sizeof (crtime)));
370 	ZFS_TIME_DECODE(&dot_zfs->cmtime, crtime);
371 	vput(rvp);
372 
373 	zfsvfs->z_ctldir = dot_zfs;
374 }
375 
376 /*
377  * Destroy the '.zfs' directory.  Only called when the filesystem is unmounted.
378  * The nodes must not have any associated vnodes by now as they should be
379  * vflush-ed.
380  */
381 void
zfsctl_destroy(zfsvfs_t * zfsvfs)382 zfsctl_destroy(zfsvfs_t *zfsvfs)
383 {
384 	sfs_destroy_node(zfsvfs->z_ctldir->snapdir);
385 	sfs_destroy_node((sfs_node_t *)zfsvfs->z_ctldir);
386 	zfsvfs->z_ctldir = NULL;
387 }
388 
389 static int
zfsctl_fs_root_vnode(struct mount * mp,void * arg __unused,int flags,struct vnode ** vpp)390 zfsctl_fs_root_vnode(struct mount *mp, void *arg __unused, int flags,
391     struct vnode **vpp)
392 {
393 	return (VFS_ROOT(mp, flags, vpp));
394 }
395 
396 static void
zfsctl_common_vnode_setup(vnode_t * vp,void * arg)397 zfsctl_common_vnode_setup(vnode_t *vp, void *arg)
398 {
399 	ASSERT_VOP_ELOCKED(vp, __func__);
400 
401 	/* We support shared locking. */
402 	VN_LOCK_ASHARE(vp);
403 	vp->v_type = VDIR;
404 	vp->v_data = arg;
405 }
406 
407 static int
zfsctl_root_vnode(struct mount * mp,void * arg __unused,int flags,struct vnode ** vpp)408 zfsctl_root_vnode(struct mount *mp, void *arg __unused, int flags,
409     struct vnode **vpp)
410 {
411 	void *node;
412 	int err;
413 
414 	node = ((zfsvfs_t *)mp->mnt_data)->z_ctldir;
415 	err = sfs_vgetx(mp, flags, 0, ZFSCTL_INO_ROOT, "zfs", &zfsctl_ops_root,
416 	    zfsctl_common_vnode_setup, node, vpp);
417 	return (err);
418 }
419 
420 static int
zfsctl_snapdir_vnode(struct mount * mp,void * arg __unused,int flags,struct vnode ** vpp)421 zfsctl_snapdir_vnode(struct mount *mp, void *arg __unused, int flags,
422     struct vnode **vpp)
423 {
424 	void *node;
425 	int err;
426 
427 	node = ((zfsvfs_t *)mp->mnt_data)->z_ctldir->snapdir;
428 	err = sfs_vgetx(mp, flags, ZFSCTL_INO_ROOT, ZFSCTL_INO_SNAPDIR, "zfs",
429 	    &zfsctl_ops_snapdir, zfsctl_common_vnode_setup, node, vpp);
430 	return (err);
431 }
432 
433 /*
434  * Given a root znode, retrieve the associated .zfs directory.
435  * Add a hold to the vnode and return it.
436  */
437 int
zfsctl_root(zfsvfs_t * zfsvfs,int flags,vnode_t ** vpp)438 zfsctl_root(zfsvfs_t *zfsvfs, int flags, vnode_t **vpp)
439 {
440 	int error;
441 
442 	error = zfsctl_root_vnode(zfsvfs->z_vfs, NULL, flags, vpp);
443 	return (error);
444 }
445 
446 /*
447  * Common open routine.  Disallow any write access.
448  */
449 static int
zfsctl_common_open(struct vop_open_args * ap)450 zfsctl_common_open(struct vop_open_args *ap)
451 {
452 	int flags = ap->a_mode;
453 
454 	if (flags & FWRITE)
455 		return (SET_ERROR(EACCES));
456 
457 	return (0);
458 }
459 
460 /*
461  * Common close routine.  Nothing to do here.
462  */
463 static int
zfsctl_common_close(struct vop_close_args * ap)464 zfsctl_common_close(struct vop_close_args *ap)
465 {
466 	(void) ap;
467 	return (0);
468 }
469 
470 /*
471  * Common access routine.  Disallow writes.
472  */
473 static int
zfsctl_common_access(struct vop_access_args * ap)474 zfsctl_common_access(struct vop_access_args *ap)
475 {
476 	accmode_t accmode = ap->a_accmode;
477 
478 	if (accmode & VWRITE)
479 		return (SET_ERROR(EACCES));
480 	return (0);
481 }
482 
483 /*
484  * Common getattr function.  Fill in basic information.
485  */
486 static void
zfsctl_common_getattr(vnode_t * vp,vattr_t * vap)487 zfsctl_common_getattr(vnode_t *vp, vattr_t *vap)
488 {
489 	timestruc_t	now;
490 	sfs_node_t *node;
491 
492 	node = vp->v_data;
493 
494 	vap->va_uid = 0;
495 	vap->va_gid = 0;
496 	vap->va_rdev = 0;
497 	/*
498 	 * We are a purely virtual object, so we have no
499 	 * blocksize or allocated blocks.
500 	 */
501 	vap->va_blksize = 0;
502 	vap->va_nblocks = 0;
503 	vap->va_gen = 0;
504 	vn_fsid(vp, vap);
505 	vap->va_mode = zfsctl_ctldir_mode;
506 	vap->va_type = VDIR;
507 	/*
508 	 * We live in the now (for atime).
509 	 */
510 	gethrestime(&now);
511 	vap->va_atime = now;
512 	/* FreeBSD: Reset chflags(2) flags. */
513 	vap->va_flags = 0;
514 
515 	vap->va_nodeid = node->sn_id;
516 
517 	/* At least '.' and '..'. */
518 	vap->va_nlink = 2;
519 }
520 
521 #ifndef _OPENSOLARIS_SYS_VNODE_H_
522 struct vop_fid_args {
523 	struct vnode *a_vp;
524 	struct fid *a_fid;
525 };
526 #endif
527 
528 static int
zfsctl_common_fid(struct vop_fid_args * ap)529 zfsctl_common_fid(struct vop_fid_args *ap)
530 {
531 	vnode_t		*vp = ap->a_vp;
532 	fid_t		*fidp = (void *)ap->a_fid;
533 	sfs_node_t	*node = vp->v_data;
534 	uint64_t	object = node->sn_id;
535 	zfid_short_t	*zfid;
536 	int		i;
537 
538 	zfid = (zfid_short_t *)fidp;
539 	zfid->zf_len = SHORT_FID_LEN;
540 
541 	for (i = 0; i < sizeof (zfid->zf_object); i++)
542 		zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
543 
544 	/* .zfs nodes always have a generation number of 0 */
545 	for (i = 0; i < sizeof (zfid->zf_gen); i++)
546 		zfid->zf_gen[i] = 0;
547 
548 	return (0);
549 }
550 
551 #ifndef _SYS_SYSPROTO_H_
552 struct vop_reclaim_args {
553 	struct vnode *a_vp;
554 	struct thread *a_td;
555 };
556 #endif
557 
558 static int
zfsctl_common_reclaim(struct vop_reclaim_args * ap)559 zfsctl_common_reclaim(struct vop_reclaim_args *ap)
560 {
561 	vnode_t *vp = ap->a_vp;
562 
563 	(void) sfs_reclaim_vnode(vp);
564 	return (0);
565 }
566 
567 #ifndef _SYS_SYSPROTO_H_
568 struct vop_print_args {
569 	struct vnode *a_vp;
570 };
571 #endif
572 
573 static int
zfsctl_common_print(struct vop_print_args * ap)574 zfsctl_common_print(struct vop_print_args *ap)
575 {
576 	sfs_print_node(ap->a_vp->v_data);
577 	return (0);
578 }
579 
580 #ifndef _SYS_SYSPROTO_H_
581 struct vop_getattr_args {
582 	struct vnode *a_vp;
583 	struct vattr *a_vap;
584 	struct ucred *a_cred;
585 };
586 #endif
587 
588 /*
589  * Get root directory attributes.
590  */
591 static int
zfsctl_root_getattr(struct vop_getattr_args * ap)592 zfsctl_root_getattr(struct vop_getattr_args *ap)
593 {
594 	struct vnode *vp = ap->a_vp;
595 	struct vattr *vap = ap->a_vap;
596 	zfsctl_root_t *node = vp->v_data;
597 
598 	zfsctl_common_getattr(vp, vap);
599 	vap->va_ctime = node->cmtime;
600 	vap->va_mtime = vap->va_ctime;
601 	vap->va_birthtime = vap->va_ctime;
602 	vap->va_nlink += 1; /* snapdir */
603 	vap->va_size = vap->va_nlink;
604 	return (0);
605 }
606 
607 /*
608  * When we lookup "." we still can be asked to lock it
609  * differently, can't we?
610  */
611 static int
zfsctl_relock_dot(vnode_t * dvp,int ltype)612 zfsctl_relock_dot(vnode_t *dvp, int ltype)
613 {
614 	vref(dvp);
615 	if (ltype != VOP_ISLOCKED(dvp)) {
616 		if (ltype == LK_EXCLUSIVE)
617 			vn_lock(dvp, LK_UPGRADE | LK_RETRY);
618 		else /* if (ltype == LK_SHARED) */
619 			vn_lock(dvp, LK_DOWNGRADE | LK_RETRY);
620 
621 		/* Relock for the "." case may left us with reclaimed vnode. */
622 		if (VN_IS_DOOMED(dvp)) {
623 			vrele(dvp);
624 			return (SET_ERROR(ENOENT));
625 		}
626 	}
627 	return (0);
628 }
629 
630 /*
631  * Special case the handling of "..".
632  */
633 static int
zfsctl_root_lookup(struct vop_lookup_args * ap)634 zfsctl_root_lookup(struct vop_lookup_args *ap)
635 {
636 	struct componentname *cnp = ap->a_cnp;
637 	vnode_t *dvp = ap->a_dvp;
638 	vnode_t **vpp = ap->a_vpp;
639 	int flags = ap->a_cnp->cn_flags;
640 	int lkflags = ap->a_cnp->cn_lkflags;
641 	int nameiop = ap->a_cnp->cn_nameiop;
642 	int err;
643 
644 	ASSERT3S(dvp->v_type, ==, VDIR);
645 
646 	if ((flags & ISLASTCN) != 0 && nameiop != LOOKUP)
647 		return (SET_ERROR(ENOTSUP));
648 
649 	if (cnp->cn_namelen == 1 && *cnp->cn_nameptr == '.') {
650 		err = zfsctl_relock_dot(dvp, lkflags & LK_TYPE_MASK);
651 		if (err == 0)
652 			*vpp = dvp;
653 	} else if ((flags & ISDOTDOT) != 0) {
654 		err = vn_vget_ino_gen(dvp, zfsctl_fs_root_vnode, NULL,
655 		    lkflags, vpp);
656 	} else if (strncmp(cnp->cn_nameptr, "snapshot", cnp->cn_namelen) == 0) {
657 		err = zfsctl_snapdir_vnode(dvp->v_mount, NULL, lkflags, vpp);
658 	} else {
659 		err = SET_ERROR(ENOENT);
660 	}
661 	if (err != 0)
662 		*vpp = NULL;
663 	return (err);
664 }
665 
666 static int
zfsctl_root_readdir(struct vop_readdir_args * ap)667 zfsctl_root_readdir(struct vop_readdir_args *ap)
668 {
669 	struct dirent entry;
670 	vnode_t *vp = ap->a_vp;
671 	zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
672 	zfsctl_root_t *node = vp->v_data;
673 	zfs_uio_t uio;
674 	int *eofp = ap->a_eofflag;
675 	off_t dots_offset;
676 	int error;
677 
678 	zfs_uio_init(&uio, ap->a_uio);
679 
680 	ASSERT3S(vp->v_type, ==, VDIR);
681 
682 	/*
683 	 * FIXME: this routine only ever emits 3 entries and does not tolerate
684 	 * being called with a buffer too small to handle all of them.
685 	 *
686 	 * The check below facilitates the idiom of repeating calls until the
687 	 * count to return is 0.
688 	 */
689 	if (zfs_uio_offset(&uio) == 3 * sizeof (entry)) {
690 		return (0);
691 	}
692 
693 	error = sfs_readdir_common(zfsvfs->z_root, ZFSCTL_INO_ROOT, ap, &uio,
694 	    &dots_offset);
695 	if (error != 0) {
696 		if (error == ENAMETOOLONG) /* ran out of destination space */
697 			error = 0;
698 		return (error);
699 	}
700 	if (zfs_uio_offset(&uio) != dots_offset)
701 		return (SET_ERROR(EINVAL));
702 
703 	_Static_assert(sizeof (node->snapdir->sn_name) <= sizeof (entry.d_name),
704 	    "node->snapdir->sn_name too big for entry.d_name");
705 	entry.d_fileno = node->snapdir->sn_id;
706 	entry.d_type = DT_DIR;
707 	strcpy(entry.d_name, node->snapdir->sn_name);
708 	entry.d_namlen = strlen(entry.d_name);
709 	entry.d_reclen = sizeof (entry);
710 	error = vfs_read_dirent(ap, &entry, zfs_uio_offset(&uio));
711 	if (error != 0) {
712 		if (error == ENAMETOOLONG)
713 			error = 0;
714 		return (SET_ERROR(error));
715 	}
716 	if (eofp != NULL)
717 		*eofp = 1;
718 	return (0);
719 }
720 
721 static int
zfsctl_root_vptocnp(struct vop_vptocnp_args * ap)722 zfsctl_root_vptocnp(struct vop_vptocnp_args *ap)
723 {
724 	static const char dotzfs_name[4] = ".zfs";
725 	vnode_t *dvp;
726 	int error;
727 
728 	if (*ap->a_buflen < sizeof (dotzfs_name))
729 		return (SET_ERROR(ENOMEM));
730 
731 	error = vn_vget_ino_gen(ap->a_vp, zfsctl_fs_root_vnode, NULL,
732 	    LK_SHARED, &dvp);
733 	if (error != 0)
734 		return (SET_ERROR(error));
735 
736 	VOP_UNLOCK(dvp);
737 	*ap->a_vpp = dvp;
738 	*ap->a_buflen -= sizeof (dotzfs_name);
739 	memcpy(ap->a_buf + *ap->a_buflen, dotzfs_name, sizeof (dotzfs_name));
740 	return (0);
741 }
742 
743 static int
zfsctl_common_pathconf(struct vop_pathconf_args * ap)744 zfsctl_common_pathconf(struct vop_pathconf_args *ap)
745 {
746 	/*
747 	 * We care about ACL variables so that user land utilities like ls
748 	 * can display them correctly.  Since the ctldir's st_dev is set to be
749 	 * the same as the parent dataset, we must support all variables that
750 	 * it supports.
751 	 */
752 	switch (ap->a_name) {
753 	case _PC_LINK_MAX:
754 		*ap->a_retval = MIN(LONG_MAX, ZFS_LINK_MAX);
755 		return (0);
756 
757 	case _PC_FILESIZEBITS:
758 		*ap->a_retval = 64;
759 		return (0);
760 
761 	case _PC_MIN_HOLE_SIZE:
762 		*ap->a_retval = (int)SPA_MINBLOCKSIZE;
763 		return (0);
764 
765 	case _PC_ACL_EXTENDED:
766 		*ap->a_retval = 0;
767 		return (0);
768 
769 	case _PC_ACL_NFS4:
770 		*ap->a_retval = 1;
771 		return (0);
772 
773 	case _PC_ACL_PATH_MAX:
774 		*ap->a_retval = ACL_MAX_ENTRIES;
775 		return (0);
776 
777 	case _PC_NAME_MAX:
778 		*ap->a_retval = NAME_MAX;
779 		return (0);
780 
781 	default:
782 		return (vop_stdpathconf(ap));
783 	}
784 }
785 
786 /*
787  * Returns a trivial ACL
788  */
789 static int
zfsctl_common_getacl(struct vop_getacl_args * ap)790 zfsctl_common_getacl(struct vop_getacl_args *ap)
791 {
792 	int i;
793 
794 	if (ap->a_type != ACL_TYPE_NFS4)
795 		return (EINVAL);
796 
797 	acl_nfs4_sync_acl_from_mode(ap->a_aclp, zfsctl_ctldir_mode, 0);
798 	/*
799 	 * acl_nfs4_sync_acl_from_mode assumes that the owner can always modify
800 	 * attributes.  That is not the case for the ctldir, so we must clear
801 	 * those bits.  We also must clear ACL_READ_NAMED_ATTRS, because xattrs
802 	 * aren't supported by the ctldir.
803 	 */
804 	for (i = 0; i < ap->a_aclp->acl_cnt; i++) {
805 		struct acl_entry *entry;
806 		entry = &(ap->a_aclp->acl_entry[i]);
807 		entry->ae_perm &= ~(ACL_WRITE_ACL | ACL_WRITE_OWNER |
808 		    ACL_WRITE_ATTRIBUTES | ACL_WRITE_NAMED_ATTRS |
809 		    ACL_READ_NAMED_ATTRS);
810 	}
811 
812 	return (0);
813 }
814 
815 static struct vop_vector zfsctl_ops_root = {
816 	.vop_default =	&default_vnodeops,
817 	.vop_fplookup_vexec = VOP_EAGAIN,
818 	.vop_fplookup_symlink = VOP_EAGAIN,
819 	.vop_open =	zfsctl_common_open,
820 	.vop_close =	zfsctl_common_close,
821 	.vop_ioctl =	VOP_EINVAL,
822 	.vop_getattr =	zfsctl_root_getattr,
823 	.vop_access =	zfsctl_common_access,
824 	.vop_readdir =	zfsctl_root_readdir,
825 	.vop_lookup =	zfsctl_root_lookup,
826 	.vop_inactive =	VOP_NULL,
827 	.vop_reclaim =	zfsctl_common_reclaim,
828 	.vop_fid =	zfsctl_common_fid,
829 	.vop_print =	zfsctl_common_print,
830 	.vop_vptocnp =	zfsctl_root_vptocnp,
831 	.vop_pathconf =	zfsctl_common_pathconf,
832 	.vop_getacl =	zfsctl_common_getacl,
833 #if __FreeBSD_version >= 1400043
834 	.vop_add_writecount =	vop_stdadd_writecount_nomsync,
835 #endif
836 };
837 VFS_VOP_VECTOR_REGISTER(zfsctl_ops_root);
838 
839 static int
zfsctl_snapshot_zname(vnode_t * vp,const char * name,int len,char * zname)840 zfsctl_snapshot_zname(vnode_t *vp, const char *name, int len, char *zname)
841 {
842 	objset_t *os = ((zfsvfs_t *)((vp)->v_vfsp->vfs_data))->z_os;
843 
844 	dmu_objset_name(os, zname);
845 	if (strlen(zname) + 1 + strlen(name) >= len)
846 		return (SET_ERROR(ENAMETOOLONG));
847 	(void) strcat(zname, "@");
848 	(void) strcat(zname, name);
849 	return (0);
850 }
851 
852 static int
zfsctl_snapshot_lookup(vnode_t * vp,const char * name,uint64_t * id)853 zfsctl_snapshot_lookup(vnode_t *vp, const char *name, uint64_t *id)
854 {
855 	objset_t *os = ((zfsvfs_t *)((vp)->v_vfsp->vfs_data))->z_os;
856 	int err;
857 
858 	err = dsl_dataset_snap_lookup(dmu_objset_ds(os), name, id);
859 	return (err);
860 }
861 
862 /*
863  * Given a vnode get a root vnode of a filesystem mounted on top of
864  * the vnode, if any.  The root vnode is referenced and locked.
865  * If no filesystem is mounted then the orinal vnode remains referenced
866  * and locked.  If any error happens the orinal vnode is unlocked and
867  * released.
868  */
869 static int
zfsctl_mounted_here(vnode_t ** vpp,int flags)870 zfsctl_mounted_here(vnode_t **vpp, int flags)
871 {
872 	struct mount *mp;
873 	int err;
874 
875 	ASSERT_VOP_LOCKED(*vpp, __func__);
876 	ASSERT3S((*vpp)->v_type, ==, VDIR);
877 
878 	if ((mp = (*vpp)->v_mountedhere) != NULL) {
879 		err = vfs_busy(mp, 0);
880 		KASSERT(err == 0, ("vfs_busy(mp, 0) failed with %d", err));
881 		KASSERT(vrefcnt(*vpp) > 1, ("unreferenced mountpoint"));
882 		vput(*vpp);
883 		err = VFS_ROOT(mp, flags, vpp);
884 		vfs_unbusy(mp);
885 		return (err);
886 	}
887 	return (EJUSTRETURN);
888 }
889 
890 typedef struct {
891 	const char *snap_name;
892 	uint64_t    snap_id;
893 } snapshot_setup_arg_t;
894 
895 static void
zfsctl_snapshot_vnode_setup(vnode_t * vp,void * arg)896 zfsctl_snapshot_vnode_setup(vnode_t *vp, void *arg)
897 {
898 	snapshot_setup_arg_t *ssa = arg;
899 	sfs_node_t *node;
900 
901 	ASSERT_VOP_ELOCKED(vp, __func__);
902 
903 	node = sfs_alloc_node(sizeof (sfs_node_t),
904 	    ssa->snap_name, ZFSCTL_INO_SNAPDIR, ssa->snap_id);
905 	zfsctl_common_vnode_setup(vp, node);
906 
907 	/* We have to support recursive locking. */
908 	VN_LOCK_AREC(vp);
909 }
910 
911 /*
912  * Lookup entry point for the 'snapshot' directory.  Try to open the
913  * snapshot if it exist, creating the pseudo filesystem vnode as necessary.
914  * Perform a mount of the associated dataset on top of the vnode.
915  * There are four possibilities:
916  * - the snapshot node and vnode do not exist
917  * - the snapshot vnode is covered by the mounted snapshot
918  * - the snapshot vnode is not covered yet, the mount operation is in progress
919  * - the snapshot vnode is not covered, because the snapshot has been unmounted
920  * The last two states are transient and should be relatively short-lived.
921  */
922 static int
zfsctl_snapdir_lookup(struct vop_lookup_args * ap)923 zfsctl_snapdir_lookup(struct vop_lookup_args *ap)
924 {
925 	vnode_t *dvp = ap->a_dvp;
926 	vnode_t **vpp = ap->a_vpp;
927 	struct componentname *cnp = ap->a_cnp;
928 	char name[NAME_MAX + 1];
929 	char fullname[ZFS_MAX_DATASET_NAME_LEN];
930 	char *mountpoint;
931 	size_t mountpoint_len;
932 	zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data;
933 	uint64_t snap_id;
934 	int nameiop = cnp->cn_nameiop;
935 	int lkflags = cnp->cn_lkflags;
936 	int flags = cnp->cn_flags;
937 	int err;
938 
939 	ASSERT3S(dvp->v_type, ==, VDIR);
940 
941 	if ((flags & ISLASTCN) != 0 && nameiop != LOOKUP)
942 		return (SET_ERROR(ENOTSUP));
943 
944 	if (cnp->cn_namelen == 1 && *cnp->cn_nameptr == '.') {
945 		err = zfsctl_relock_dot(dvp, lkflags & LK_TYPE_MASK);
946 		if (err == 0)
947 			*vpp = dvp;
948 		return (err);
949 	}
950 	if (flags & ISDOTDOT) {
951 		err = vn_vget_ino_gen(dvp, zfsctl_root_vnode, NULL, lkflags,
952 		    vpp);
953 		return (err);
954 	}
955 
956 	if (cnp->cn_namelen >= sizeof (name))
957 		return (SET_ERROR(ENAMETOOLONG));
958 
959 	strlcpy(name, ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen + 1);
960 	err = zfsctl_snapshot_lookup(dvp, name, &snap_id);
961 	if (err != 0)
962 		return (SET_ERROR(ENOENT));
963 
964 	for (;;) {
965 		snapshot_setup_arg_t ssa;
966 
967 		ssa.snap_name = name;
968 		ssa.snap_id = snap_id;
969 		err = sfs_vgetx(dvp->v_mount, LK_SHARED, ZFSCTL_INO_SNAPDIR,
970 		    snap_id, "zfs", &zfsctl_ops_snapshot,
971 		    zfsctl_snapshot_vnode_setup, &ssa, vpp);
972 		if (err != 0)
973 			return (err);
974 
975 		/* Check if a new vnode has just been created. */
976 		if (VOP_ISLOCKED(*vpp) == LK_EXCLUSIVE)
977 			break;
978 
979 		/*
980 		 * Check if a snapshot is already mounted on top of the vnode.
981 		 */
982 		err = zfsctl_mounted_here(vpp, lkflags);
983 		if (err != EJUSTRETURN)
984 			return (err);
985 
986 		/*
987 		 * If the vnode is not covered, then either the mount operation
988 		 * is in progress or the snapshot has already been unmounted
989 		 * but the vnode hasn't been inactivated and reclaimed yet.
990 		 * We can try to re-use the vnode in the latter case.
991 		 */
992 		VI_LOCK(*vpp);
993 		if (((*vpp)->v_iflag & VI_MOUNT) == 0) {
994 			VI_UNLOCK(*vpp);
995 			/*
996 			 * Upgrade to exclusive lock in order to:
997 			 * - avoid race conditions
998 			 * - satisfy the contract of mount_snapshot()
999 			 */
1000 			err = VOP_LOCK(*vpp, LK_TRYUPGRADE);
1001 			if (err == 0)
1002 				break;
1003 		} else {
1004 			VI_UNLOCK(*vpp);
1005 		}
1006 
1007 		/*
1008 		 * In this state we can loop on uncontested locks and starve
1009 		 * the thread doing the lengthy, non-trivial mount operation.
1010 		 * So, yield to prevent that from happening.
1011 		 */
1012 		vput(*vpp);
1013 		kern_yield(PRI_USER);
1014 	}
1015 
1016 	VERIFY0(zfsctl_snapshot_zname(dvp, name, sizeof (fullname), fullname));
1017 
1018 	mountpoint_len = strlen(dvp->v_vfsp->mnt_stat.f_mntonname) +
1019 	    strlen("/" ZFS_CTLDIR_NAME "/snapshot/") + strlen(name) + 1;
1020 	mountpoint = kmem_alloc(mountpoint_len, KM_SLEEP);
1021 	(void) snprintf(mountpoint, mountpoint_len,
1022 	    "%s/" ZFS_CTLDIR_NAME "/snapshot/%s",
1023 	    dvp->v_vfsp->mnt_stat.f_mntonname, name);
1024 
1025 	err = mount_snapshot(curthread, vpp, "zfs", mountpoint, fullname, 0,
1026 	    dvp->v_vfsp);
1027 	kmem_free(mountpoint, mountpoint_len);
1028 	if (err == 0) {
1029 		/*
1030 		 * Fix up the root vnode mounted on .zfs/snapshot/<snapname>.
1031 		 *
1032 		 * This is where we lie about our v_vfsp in order to
1033 		 * make .zfs/snapshot/<snapname> accessible over NFS
1034 		 * without requiring manual mounts of <snapname>.
1035 		 */
1036 		ASSERT3P(VTOZ(*vpp)->z_zfsvfs, !=, zfsvfs);
1037 		VTOZ(*vpp)->z_zfsvfs->z_parent = zfsvfs;
1038 
1039 		/* Clear the root flag (set via VFS_ROOT) as well. */
1040 		(*vpp)->v_vflag &= ~VV_ROOT;
1041 	}
1042 
1043 	if (err != 0)
1044 		*vpp = NULL;
1045 	return (err);
1046 }
1047 
1048 static int
zfsctl_snapdir_readdir(struct vop_readdir_args * ap)1049 zfsctl_snapdir_readdir(struct vop_readdir_args *ap)
1050 {
1051 	char snapname[ZFS_MAX_DATASET_NAME_LEN];
1052 	struct dirent entry;
1053 	vnode_t *vp = ap->a_vp;
1054 	zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
1055 	zfs_uio_t uio;
1056 	int *eofp = ap->a_eofflag;
1057 	off_t dots_offset;
1058 	int error;
1059 
1060 	zfs_uio_init(&uio, ap->a_uio);
1061 
1062 	ASSERT3S(vp->v_type, ==, VDIR);
1063 
1064 	error = sfs_readdir_common(ZFSCTL_INO_ROOT, ZFSCTL_INO_SNAPDIR, ap,
1065 	    &uio, &dots_offset);
1066 	if (error != 0) {
1067 		if (error == ENAMETOOLONG) /* ran out of destination space */
1068 			error = 0;
1069 		return (error);
1070 	}
1071 
1072 	if ((error = zfs_enter(zfsvfs, FTAG)) != 0)
1073 		return (error);
1074 	for (;;) {
1075 		uint64_t cookie;
1076 		uint64_t id;
1077 
1078 		cookie = zfs_uio_offset(&uio) - dots_offset;
1079 
1080 		dsl_pool_config_enter(dmu_objset_pool(zfsvfs->z_os), FTAG);
1081 		error = dmu_snapshot_list_next(zfsvfs->z_os, sizeof (snapname),
1082 		    snapname, &id, &cookie, NULL);
1083 		dsl_pool_config_exit(dmu_objset_pool(zfsvfs->z_os), FTAG);
1084 		if (error != 0) {
1085 			if (error == ENOENT) {
1086 				if (eofp != NULL)
1087 					*eofp = 1;
1088 				error = 0;
1089 			}
1090 			zfs_exit(zfsvfs, FTAG);
1091 			return (error);
1092 		}
1093 
1094 		entry.d_fileno = id;
1095 		entry.d_type = DT_DIR;
1096 		strcpy(entry.d_name, snapname);
1097 		entry.d_namlen = strlen(entry.d_name);
1098 		entry.d_reclen = sizeof (entry);
1099 		error = vfs_read_dirent(ap, &entry, zfs_uio_offset(&uio));
1100 		if (error != 0) {
1101 			if (error == ENAMETOOLONG)
1102 				error = 0;
1103 			zfs_exit(zfsvfs, FTAG);
1104 			return (SET_ERROR(error));
1105 		}
1106 		zfs_uio_setoffset(&uio, cookie + dots_offset);
1107 	}
1108 	__builtin_unreachable();
1109 }
1110 
1111 static int
zfsctl_snapdir_getattr(struct vop_getattr_args * ap)1112 zfsctl_snapdir_getattr(struct vop_getattr_args *ap)
1113 {
1114 	vnode_t *vp = ap->a_vp;
1115 	vattr_t *vap = ap->a_vap;
1116 	zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
1117 	dsl_dataset_t *ds;
1118 	uint64_t snap_count;
1119 	int err;
1120 
1121 	if ((err = zfs_enter(zfsvfs, FTAG)) != 0)
1122 		return (err);
1123 	ds = dmu_objset_ds(zfsvfs->z_os);
1124 	zfsctl_common_getattr(vp, vap);
1125 	vap->va_ctime = dmu_objset_snap_cmtime(zfsvfs->z_os);
1126 	vap->va_mtime = vap->va_ctime;
1127 	vap->va_birthtime = vap->va_ctime;
1128 	if (dsl_dataset_phys(ds)->ds_snapnames_zapobj != 0) {
1129 		err = zap_count(dmu_objset_pool(ds->ds_objset)->dp_meta_objset,
1130 		    dsl_dataset_phys(ds)->ds_snapnames_zapobj, &snap_count);
1131 		if (err != 0) {
1132 			zfs_exit(zfsvfs, FTAG);
1133 			return (err);
1134 		}
1135 		vap->va_nlink += snap_count;
1136 	}
1137 	vap->va_size = vap->va_nlink;
1138 
1139 	zfs_exit(zfsvfs, FTAG);
1140 	return (0);
1141 }
1142 
1143 static struct vop_vector zfsctl_ops_snapdir = {
1144 	.vop_default =	&default_vnodeops,
1145 	.vop_fplookup_vexec = VOP_EAGAIN,
1146 	.vop_fplookup_symlink = VOP_EAGAIN,
1147 	.vop_open =	zfsctl_common_open,
1148 	.vop_close =	zfsctl_common_close,
1149 	.vop_getattr =	zfsctl_snapdir_getattr,
1150 	.vop_access =	zfsctl_common_access,
1151 	.vop_readdir =	zfsctl_snapdir_readdir,
1152 	.vop_lookup =	zfsctl_snapdir_lookup,
1153 	.vop_reclaim =	zfsctl_common_reclaim,
1154 	.vop_fid =	zfsctl_common_fid,
1155 	.vop_print =	zfsctl_common_print,
1156 	.vop_pathconf =	zfsctl_common_pathconf,
1157 	.vop_getacl =	zfsctl_common_getacl,
1158 #if __FreeBSD_version >= 1400043
1159 	.vop_add_writecount =	vop_stdadd_writecount_nomsync,
1160 #endif
1161 };
1162 VFS_VOP_VECTOR_REGISTER(zfsctl_ops_snapdir);
1163 
1164 
1165 static int
zfsctl_snapshot_inactive(struct vop_inactive_args * ap)1166 zfsctl_snapshot_inactive(struct vop_inactive_args *ap)
1167 {
1168 	vnode_t *vp = ap->a_vp;
1169 
1170 	vrecycle(vp);
1171 	return (0);
1172 }
1173 
1174 static int
zfsctl_snapshot_reclaim(struct vop_reclaim_args * ap)1175 zfsctl_snapshot_reclaim(struct vop_reclaim_args *ap)
1176 {
1177 	vnode_t *vp = ap->a_vp;
1178 	void *data = vp->v_data;
1179 
1180 	sfs_reclaim_vnode(vp);
1181 	sfs_destroy_node(data);
1182 	return (0);
1183 }
1184 
1185 static int
zfsctl_snapshot_vptocnp(struct vop_vptocnp_args * ap)1186 zfsctl_snapshot_vptocnp(struct vop_vptocnp_args *ap)
1187 {
1188 	struct mount *mp;
1189 	vnode_t *dvp;
1190 	vnode_t *vp;
1191 	sfs_node_t *node;
1192 	size_t len;
1193 	int locked;
1194 	int error;
1195 
1196 	vp = ap->a_vp;
1197 	node = vp->v_data;
1198 	len = strlen(node->sn_name);
1199 	if (*ap->a_buflen < len)
1200 		return (SET_ERROR(ENOMEM));
1201 
1202 	/*
1203 	 * Prevent unmounting of the snapshot while the vnode lock
1204 	 * is not held.  That is not strictly required, but allows
1205 	 * us to assert that an uncovered snapshot vnode is never
1206 	 * "leaked".
1207 	 */
1208 	mp = vp->v_mountedhere;
1209 	if (mp == NULL)
1210 		return (SET_ERROR(ENOENT));
1211 	error = vfs_busy(mp, 0);
1212 	KASSERT(error == 0, ("vfs_busy(mp, 0) failed with %d", error));
1213 
1214 	/*
1215 	 * We can vput the vnode as we can now depend on the reference owned
1216 	 * by the busied mp.  But we also need to hold the vnode, because
1217 	 * the reference may go after vfs_unbusy() which has to be called
1218 	 * before we can lock the vnode again.
1219 	 */
1220 	locked = VOP_ISLOCKED(vp);
1221 	enum vgetstate vs = vget_prep(vp);
1222 	vput(vp);
1223 
1224 	/* Look up .zfs/snapshot, our parent. */
1225 	error = zfsctl_snapdir_vnode(vp->v_mount, NULL, LK_SHARED, &dvp);
1226 	if (error == 0) {
1227 		VOP_UNLOCK(dvp);
1228 		*ap->a_vpp = dvp;
1229 		*ap->a_buflen -= len;
1230 		memcpy(ap->a_buf + *ap->a_buflen, node->sn_name, len);
1231 	}
1232 	vfs_unbusy(mp);
1233 	vget_finish(vp, locked | LK_RETRY, vs);
1234 	return (error);
1235 }
1236 
1237 /*
1238  * These VP's should never see the light of day.  They should always
1239  * be covered.
1240  */
1241 static struct vop_vector zfsctl_ops_snapshot = {
1242 	.vop_default =		NULL, /* ensure very restricted access */
1243 	.vop_fplookup_vexec =	VOP_EAGAIN,
1244 	.vop_fplookup_symlink = VOP_EAGAIN,
1245 	.vop_open =		zfsctl_common_open,
1246 	.vop_close =		zfsctl_common_close,
1247 	.vop_inactive =		zfsctl_snapshot_inactive,
1248 	.vop_need_inactive =	vop_stdneed_inactive,
1249 	.vop_reclaim =		zfsctl_snapshot_reclaim,
1250 	.vop_vptocnp =		zfsctl_snapshot_vptocnp,
1251 	.vop_lock1 =		vop_stdlock,
1252 	.vop_unlock =		vop_stdunlock,
1253 	.vop_islocked =		vop_stdislocked,
1254 	.vop_advlockpurge =	vop_stdadvlockpurge, /* called by vgone */
1255 	.vop_print =		zfsctl_common_print,
1256 #if __FreeBSD_version >= 1400043
1257 	.vop_add_writecount =	vop_stdadd_writecount_nomsync,
1258 #endif
1259 };
1260 VFS_VOP_VECTOR_REGISTER(zfsctl_ops_snapshot);
1261 
1262 int
zfsctl_lookup_objset(vfs_t * vfsp,uint64_t objsetid,zfsvfs_t ** zfsvfsp)1263 zfsctl_lookup_objset(vfs_t *vfsp, uint64_t objsetid, zfsvfs_t **zfsvfsp)
1264 {
1265 	zfsvfs_t *zfsvfs __unused = vfsp->vfs_data;
1266 	vnode_t *vp;
1267 	int error;
1268 
1269 	ASSERT3P(zfsvfs->z_ctldir, !=, NULL);
1270 	*zfsvfsp = NULL;
1271 	error = sfs_vnode_get(vfsp, LK_EXCLUSIVE,
1272 	    ZFSCTL_INO_SNAPDIR, objsetid, &vp);
1273 	if (error == 0 && vp != NULL) {
1274 		/*
1275 		 * XXX Probably need to at least reference, if not busy, the mp.
1276 		 */
1277 		if (vp->v_mountedhere != NULL)
1278 			*zfsvfsp = vp->v_mountedhere->mnt_data;
1279 		vput(vp);
1280 	}
1281 	if (*zfsvfsp == NULL)
1282 		return (SET_ERROR(EINVAL));
1283 	return (0);
1284 }
1285 
1286 /*
1287  * Unmount any snapshots for the given filesystem.  This is called from
1288  * zfs_umount() - if we have a ctldir, then go through and unmount all the
1289  * snapshots.
1290  */
1291 int
zfsctl_umount_snapshots(vfs_t * vfsp,int fflags,cred_t * cr)1292 zfsctl_umount_snapshots(vfs_t *vfsp, int fflags, cred_t *cr)
1293 {
1294 	char snapname[ZFS_MAX_DATASET_NAME_LEN];
1295 	zfsvfs_t *zfsvfs = vfsp->vfs_data;
1296 	struct mount *mp;
1297 	vnode_t *vp;
1298 	uint64_t cookie;
1299 	int error;
1300 
1301 	ASSERT3P(zfsvfs->z_ctldir, !=, NULL);
1302 
1303 	cookie = 0;
1304 	for (;;) {
1305 		uint64_t id;
1306 
1307 		dsl_pool_config_enter(dmu_objset_pool(zfsvfs->z_os), FTAG);
1308 		error = dmu_snapshot_list_next(zfsvfs->z_os, sizeof (snapname),
1309 		    snapname, &id, &cookie, NULL);
1310 		dsl_pool_config_exit(dmu_objset_pool(zfsvfs->z_os), FTAG);
1311 		if (error != 0) {
1312 			if (error == ENOENT)
1313 				error = 0;
1314 			break;
1315 		}
1316 
1317 		for (;;) {
1318 			error = sfs_vnode_get(vfsp, LK_EXCLUSIVE,
1319 			    ZFSCTL_INO_SNAPDIR, id, &vp);
1320 			if (error != 0 || vp == NULL)
1321 				break;
1322 
1323 			mp = vp->v_mountedhere;
1324 
1325 			/*
1326 			 * v_mountedhere being NULL means that the
1327 			 * (uncovered) vnode is in a transient state
1328 			 * (mounting or unmounting), so loop until it
1329 			 * settles down.
1330 			 */
1331 			if (mp != NULL)
1332 				break;
1333 			vput(vp);
1334 		}
1335 		if (error != 0)
1336 			break;
1337 		if (vp == NULL)
1338 			continue;	/* no mountpoint, nothing to do */
1339 
1340 		/*
1341 		 * The mount-point vnode is kept locked to avoid spurious EBUSY
1342 		 * from a concurrent umount.
1343 		 * The vnode lock must have recursive locking enabled.
1344 		 */
1345 		vfs_ref(mp);
1346 		error = dounmount(mp, fflags, curthread);
1347 		KASSERT_IMPLY(error == 0, vrefcnt(vp) == 1,
1348 		    ("extra references after unmount"));
1349 		vput(vp);
1350 		if (error != 0)
1351 			break;
1352 	}
1353 	KASSERT_IMPLY((fflags & MS_FORCE) != 0, error == 0,
1354 	    ("force unmounting failed"));
1355 	return (error);
1356 }
1357 
1358 int
zfsctl_snapshot_unmount(const char * snapname,int flags __unused)1359 zfsctl_snapshot_unmount(const char *snapname, int flags __unused)
1360 {
1361 	vfs_t *vfsp = NULL;
1362 	zfsvfs_t *zfsvfs = NULL;
1363 
1364 	if (strchr(snapname, '@') == NULL)
1365 		return (0);
1366 
1367 	int err = getzfsvfs(snapname, &zfsvfs);
1368 	if (err != 0) {
1369 		ASSERT3P(zfsvfs, ==, NULL);
1370 		return (0);
1371 	}
1372 	vfsp = zfsvfs->z_vfs;
1373 
1374 	ASSERT(!dsl_pool_config_held(dmu_objset_pool(zfsvfs->z_os)));
1375 
1376 	vfs_ref(vfsp);
1377 	vfs_unbusy(vfsp);
1378 	return (dounmount(vfsp, MS_FORCE, curthread));
1379 }
1380