xref: /freebsd/sys/fs/nullfs/null_subr.c (revision 5956d97f4b3204318ceb6aa9c77bd0bc6ea87a41)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1992, 1993
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_subr.c	8.7 (Berkeley) 5/14/95
35  *
36  * $FreeBSD$
37  */
38 
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/kernel.h>
42 #include <sys/lock.h>
43 #include <sys/rwlock.h>
44 #include <sys/malloc.h>
45 #include <sys/mount.h>
46 #include <sys/proc.h>
47 #include <sys/vnode.h>
48 
49 #include <fs/nullfs/null.h>
50 
51 /*
52  * Null layer cache:
53  * Each cache entry holds a reference to the lower vnode
54  * along with a pointer to the alias vnode.  When an
55  * entry is added the lower vnode is VREF'd.  When the
56  * alias is removed the lower vnode is vrele'd.
57  */
58 
59 #define	NULL_NHASH(vp) (&null_node_hashtbl[vfs_hash_index(vp) & null_hash_mask])
60 
61 static LIST_HEAD(null_node_hashhead, null_node) *null_node_hashtbl;
62 static struct rwlock null_hash_lock;
63 static u_long null_hash_mask;
64 
65 static MALLOC_DEFINE(M_NULLFSHASH, "nullfs_hash", "NULLFS hash table");
66 MALLOC_DEFINE(M_NULLFSNODE, "nullfs_node", "NULLFS vnode private part");
67 
68 static void null_hashins(struct mount *, struct null_node *);
69 
70 /*
71  * Initialise cache headers
72  */
73 int
74 nullfs_init(struct vfsconf *vfsp)
75 {
76 
77 	null_node_hashtbl = hashinit(desiredvnodes, M_NULLFSHASH,
78 	    &null_hash_mask);
79 	rw_init(&null_hash_lock, "nullhs");
80 	return (0);
81 }
82 
83 int
84 nullfs_uninit(struct vfsconf *vfsp)
85 {
86 
87 	rw_destroy(&null_hash_lock);
88 	hashdestroy(null_node_hashtbl, M_NULLFSHASH, null_hash_mask);
89 	return (0);
90 }
91 
92 /*
93  * Return a VREF'ed alias for lower vnode if already exists, else 0.
94  * Lower vnode should be locked on entry and will be left locked on exit.
95  */
96 static struct vnode *
97 null_hashget_locked(struct mount *mp, struct vnode *lowervp)
98 {
99 	struct null_node_hashhead *hd;
100 	struct null_node *a;
101 	struct vnode *vp;
102 
103 	ASSERT_VOP_LOCKED(lowervp, "null_hashget");
104 	rw_assert(&null_hash_lock, RA_LOCKED);
105 
106 	/*
107 	 * Find hash base, and then search the (two-way) linked
108 	 * list looking for a null_node structure which is referencing
109 	 * the lower vnode.  If found, the increment the null_node
110 	 * reference count (but NOT the lower vnode's VREF counter).
111 	 */
112 	hd = NULL_NHASH(lowervp);
113 	LIST_FOREACH(a, hd, null_hash) {
114 		if (a->null_lowervp == lowervp && NULLTOV(a)->v_mount == mp) {
115 			/*
116 			 * Since we have the lower node locked the nullfs
117 			 * node can not be in the process of recycling.  If
118 			 * it had been recycled before we grabed the lower
119 			 * lock it would not have been found on the hash.
120 			 */
121 			vp = NULLTOV(a);
122 			vref(vp);
123 			return (vp);
124 		}
125 	}
126 	return (NULLVP);
127 }
128 
129 struct vnode *
130 null_hashget(struct mount *mp, struct vnode *lowervp)
131 {
132 	struct null_node_hashhead *hd;
133 	struct vnode *vp;
134 
135 	hd = NULL_NHASH(lowervp);
136 	if (LIST_EMPTY(hd))
137 		return (NULLVP);
138 
139 	rw_rlock(&null_hash_lock);
140 	vp = null_hashget_locked(mp, lowervp);
141 	rw_runlock(&null_hash_lock);
142 
143 	return (vp);
144 }
145 
146 static void
147 null_hashins(struct mount *mp, struct null_node *xp)
148 {
149 	struct null_node_hashhead *hd;
150 #ifdef INVARIANTS
151 	struct null_node *oxp;
152 #endif
153 
154 	rw_assert(&null_hash_lock, RA_WLOCKED);
155 
156 	hd = NULL_NHASH(xp->null_lowervp);
157 #ifdef INVARIANTS
158 	LIST_FOREACH(oxp, hd, null_hash) {
159 		if (oxp->null_lowervp == xp->null_lowervp &&
160 		    NULLTOV(oxp)->v_mount == mp) {
161 			VNASSERT(0, NULLTOV(oxp),
162 			    ("vnode already in hash"));
163 		}
164 	}
165 #endif
166 	LIST_INSERT_HEAD(hd, xp, null_hash);
167 }
168 
169 static void
170 null_destroy_proto(struct vnode *vp, void *xp)
171 {
172 
173 	lockmgr(&vp->v_lock, LK_EXCLUSIVE, NULL);
174 	VI_LOCK(vp);
175 	vp->v_data = NULL;
176 	vp->v_vnlock = &vp->v_lock;
177 	vp->v_op = &dead_vnodeops;
178 	VI_UNLOCK(vp);
179 	vgone(vp);
180 	vput(vp);
181 	free(xp, M_NULLFSNODE);
182 }
183 
184 /*
185  * Make a new or get existing nullfs node.
186  * Vp is the alias vnode, lowervp is the lower vnode.
187  *
188  * The lowervp assumed to be locked and having "spare" reference. This routine
189  * vrele lowervp if nullfs node was taken from hash. Otherwise it "transfers"
190  * the caller's "spare" reference to created nullfs vnode.
191  */
192 int
193 null_nodeget(struct mount *mp, struct vnode *lowervp, struct vnode **vpp)
194 {
195 	struct null_node *xp;
196 	struct vnode *vp;
197 	int error;
198 
199 	ASSERT_VOP_LOCKED(lowervp, "lowervp");
200 	VNPASS(lowervp->v_usecount > 0, lowervp);
201 
202 	/* Lookup the hash firstly. */
203 	*vpp = null_hashget(mp, lowervp);
204 	if (*vpp != NULL) {
205 		vrele(lowervp);
206 		return (0);
207 	}
208 
209 	/*
210 	 * We do not serialize vnode creation, instead we will check for
211 	 * duplicates later, when adding new vnode to hash.
212 	 * Note that duplicate can only appear in hash if the lowervp is
213 	 * locked LK_SHARED.
214 	 */
215 	xp = malloc(sizeof(struct null_node), M_NULLFSNODE, M_WAITOK);
216 
217 	error = getnewvnode("nullfs", mp, &null_vnodeops, &vp);
218 	if (error) {
219 		vput(lowervp);
220 		free(xp, M_NULLFSNODE);
221 		return (error);
222 	}
223 
224 	VNPASS(vp->v_object == NULL, vp);
225 	VNPASS((vn_irflag_read(vp) & VIRF_PGREAD) == 0, vp);
226 
227 	rw_wlock(&null_hash_lock);
228 	xp->null_vnode = vp;
229 	xp->null_lowervp = lowervp;
230 	xp->null_flags = 0;
231 	vp->v_type = lowervp->v_type;
232 	vp->v_data = xp;
233 	vp->v_vnlock = lowervp->v_vnlock;
234 	*vpp = null_hashget_locked(mp, lowervp);
235 	if (*vpp != NULL) {
236 		rw_wunlock(&null_hash_lock);
237 		vrele(lowervp);
238 		null_destroy_proto(vp, xp);
239 		return (0);
240 	}
241 
242 	/*
243 	 * We might miss the case where lower vnode sets VIRF_PGREAD
244 	 * some time after construction, which is typical case.
245 	 * null_open rechecks.
246 	 */
247 	if ((vn_irflag_read(lowervp) & VIRF_PGREAD) != 0) {
248 		MPASS(lowervp->v_object != NULL);
249 		vp->v_object = lowervp->v_object;
250 		vn_irflag_set(vp, VIRF_PGREAD);
251 	}
252 	if (lowervp == MOUNTTONULLMOUNT(mp)->nullm_lowerrootvp)
253 		vp->v_vflag |= VV_ROOT;
254 
255 	error = insmntque1(vp, mp);
256 	if (error != 0) {
257 		rw_wunlock(&null_hash_lock);
258 		vput(lowervp);
259 		vp->v_object = NULL;
260 		null_destroy_proto(vp, xp);
261 		return (error);
262 	}
263 
264 	null_hashins(mp, xp);
265 	rw_wunlock(&null_hash_lock);
266 	*vpp = vp;
267 
268 	return (0);
269 }
270 
271 /*
272  * Remove node from hash.
273  */
274 void
275 null_hashrem(struct null_node *xp)
276 {
277 
278 	rw_wlock(&null_hash_lock);
279 	LIST_REMOVE(xp, null_hash);
280 	rw_wunlock(&null_hash_lock);
281 }
282 
283 #ifdef DIAGNOSTIC
284 
285 struct vnode *
286 null_checkvp(struct vnode *vp, char *fil, int lno)
287 {
288 	struct null_node *a = VTONULL(vp);
289 
290 #ifdef notyet
291 	/*
292 	 * Can't do this check because vop_reclaim runs
293 	 * with a funny vop vector.
294 	 */
295 	if (vp->v_op != null_vnodeop_p) {
296 		printf ("null_checkvp: on non-null-node\n");
297 		panic("null_checkvp");
298 	}
299 #endif
300 	if (a->null_lowervp == NULLVP) {
301 		/* Should never happen */
302 		panic("null_checkvp %p", vp);
303 	}
304 	VI_LOCK_FLAGS(a->null_lowervp, MTX_DUPOK);
305 	if (a->null_lowervp->v_usecount < 1)
306 		panic ("null with unref'ed lowervp, vp %p lvp %p",
307 		    vp, a->null_lowervp);
308 	VI_UNLOCK(a->null_lowervp);
309 #ifdef notyet
310 	printf("null %x/%d -> %x/%d [%s, %d]\n",
311 	        NULLTOV(a), vrefcnt(NULLTOV(a)),
312 		a->null_lowervp, vrefcnt(a->null_lowervp),
313 		fil, lno);
314 #endif
315 	return (a->null_lowervp);
316 }
317 #endif
318