xref: /freebsd/sys/fs/nullfs/null_subr.c (revision 74bf4e164ba5851606a27d4feff27717452583e5)
1 /*
2  * Copyright (c) 1992, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from software donated to Berkeley by
6  * Jan-Simon Pendry.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 4. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  *	@(#)null_subr.c	8.7 (Berkeley) 5/14/95
33  *
34  * $FreeBSD$
35  */
36 
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/kernel.h>
40 #include <sys/lock.h>
41 #include <sys/mutex.h>
42 #include <sys/malloc.h>
43 #include <sys/mount.h>
44 #include <sys/proc.h>
45 #include <sys/vnode.h>
46 
47 #include <fs/nullfs/null.h>
48 
49 #define LOG2_SIZEVNODE 8		/* log2(sizeof struct vnode) */
50 #define	NNULLNODECACHE 16
51 
52 /*
53  * Null layer cache:
54  * Each cache entry holds a reference to the lower vnode
55  * along with a pointer to the alias vnode.  When an
56  * entry is added the lower vnode is VREF'd.  When the
57  * alias is removed the lower vnode is vrele'd.
58  */
59 
60 #define	NULL_NHASH(vp) \
61 	(&null_node_hashtbl[(((uintptr_t)vp)>>LOG2_SIZEVNODE) & null_node_hash])
62 
63 static LIST_HEAD(null_node_hashhead, null_node) *null_node_hashtbl;
64 static u_long null_node_hash;
65 struct mtx null_hashmtx;
66 
67 static MALLOC_DEFINE(M_NULLFSHASH, "NULLFS hash", "NULLFS hash table");
68 MALLOC_DEFINE(M_NULLFSNODE, "NULLFS node", "NULLFS vnode private part");
69 
70 static struct vnode * null_hashget(struct mount *, struct vnode *);
71 static struct vnode * null_hashins(struct mount *, struct null_node *);
72 
73 /*
74  * Initialise cache headers
75  */
76 int
77 nullfs_init(vfsp)
78 	struct vfsconf *vfsp;
79 {
80 
81 	NULLFSDEBUG("nullfs_init\n");		/* printed during system boot */
82 	null_node_hashtbl = hashinit(NNULLNODECACHE, M_NULLFSHASH, &null_node_hash);
83 	mtx_init(&null_hashmtx, "nullhs", NULL, MTX_DEF);
84 	return (0);
85 }
86 
87 int
88 nullfs_uninit(vfsp)
89 	struct vfsconf *vfsp;
90 {
91 
92 	mtx_destroy(&null_hashmtx);
93 	free(null_node_hashtbl, M_NULLFSHASH);
94 	return (0);
95 }
96 
97 /*
98  * Return a VREF'ed alias for lower vnode if already exists, else 0.
99  * Lower vnode should be locked on entry and will be left locked on exit.
100  */
101 static struct vnode *
102 null_hashget(mp, lowervp)
103 	struct mount *mp;
104 	struct vnode *lowervp;
105 {
106 	struct thread *td = curthread;	/* XXX */
107 	struct null_node_hashhead *hd;
108 	struct null_node *a;
109 	struct vnode *vp;
110 
111 	/*
112 	 * Find hash base, and then search the (two-way) linked
113 	 * list looking for a null_node structure which is referencing
114 	 * the lower vnode.  If found, the increment the null_node
115 	 * reference count (but NOT the lower vnode's VREF counter).
116 	 */
117 	hd = NULL_NHASH(lowervp);
118 loop:
119 	mtx_lock(&null_hashmtx);
120 	LIST_FOREACH(a, hd, null_hash) {
121 		if (a->null_lowervp == lowervp && NULLTOV(a)->v_mount == mp) {
122 			vp = NULLTOV(a);
123 			mtx_lock(&vp->v_interlock);
124 			/*
125 			 * Don't block if nullfs vnode is being recycled.
126 			 * We already hold a lock on the lower vnode, thus
127 			 * waiting might deadlock against the thread
128 			 * recycling the nullfs vnode or another thread
129 			 * in vrele() waiting for the vnode lock.
130 			 */
131 			if ((vp->v_iflag & VI_XLOCK) != 0) {
132 				VI_UNLOCK(vp);
133 				continue;
134 			}
135 			mtx_unlock(&null_hashmtx);
136 			/*
137 			 * We need vget for the VXLOCK
138 			 * stuff, but we don't want to lock
139 			 * the lower node.
140 			 */
141 			if (vget(vp, LK_EXCLUSIVE | LK_THISLAYER | LK_INTERLOCK, td))
142 				goto loop;
143 
144 			return (vp);
145 		}
146 	}
147 	mtx_unlock(&null_hashmtx);
148 	return (NULLVP);
149 }
150 
151 /*
152  * Act like null_hashget, but add passed null_node to hash if no existing
153  * node found.
154  */
155 static struct vnode *
156 null_hashins(mp, xp)
157 	struct mount *mp;
158 	struct null_node *xp;
159 {
160 	struct thread *td = curthread;	/* XXX */
161 	struct null_node_hashhead *hd;
162 	struct null_node *oxp;
163 	struct vnode *ovp;
164 
165 	hd = NULL_NHASH(xp->null_lowervp);
166 loop:
167 	mtx_lock(&null_hashmtx);
168 	LIST_FOREACH(oxp, hd, null_hash) {
169 		if (oxp->null_lowervp == xp->null_lowervp &&
170 		    NULLTOV(oxp)->v_mount == mp) {
171 			ovp = NULLTOV(oxp);
172 			mtx_lock(&ovp->v_interlock);
173 			/*
174 			 * Don't block if nullfs vnode is being recycled.
175 			 * We already hold a lock on the lower vnode, thus
176 			 * waiting might deadlock against the thread
177 			 * recycling the nullfs vnode or another thread
178 			 * in vrele() waiting for the vnode lock.
179 			 */
180 			if ((ovp->v_iflag & VI_XLOCK) != 0) {
181 				VI_UNLOCK(ovp);
182 				continue;
183 			}
184 			mtx_unlock(&null_hashmtx);
185 			if (vget(ovp, LK_EXCLUSIVE | LK_THISLAYER | LK_INTERLOCK, td))
186 				goto loop;
187 
188 			return (ovp);
189 		}
190 	}
191 	LIST_INSERT_HEAD(hd, xp, null_hash);
192 	mtx_unlock(&null_hashmtx);
193 	return (NULLVP);
194 }
195 
196 /*
197  * Make a new or get existing nullfs node.
198  * Vp is the alias vnode, lowervp is the lower vnode.
199  *
200  * The lowervp assumed to be locked and having "spare" reference. This routine
201  * vrele lowervp if nullfs node was taken from hash. Otherwise it "transfers"
202  * the caller's "spare" reference to created nullfs vnode.
203  */
204 int
205 null_nodeget(mp, lowervp, vpp)
206 	struct mount *mp;
207 	struct vnode *lowervp;
208 	struct vnode **vpp;
209 {
210 	struct thread *td = curthread;	/* XXX */
211 	struct null_node *xp;
212 	struct vnode *vp;
213 	int error;
214 
215 	/* Lookup the hash firstly */
216 	*vpp = null_hashget(mp, lowervp);
217 	if (*vpp != NULL) {
218 		vrele(lowervp);
219 		return (0);
220 	}
221 
222 	/*
223 	 * We do not serialize vnode creation, instead we will check for
224 	 * duplicates later, when adding new vnode to hash.
225 	 *
226 	 * Note that duplicate can only appear in hash if the lowervp is
227 	 * locked LK_SHARED.
228 	 */
229 
230 	/*
231 	 * Do the MALLOC before the getnewvnode since doing so afterward
232 	 * might cause a bogus v_data pointer to get dereferenced
233 	 * elsewhere if MALLOC should block.
234 	 */
235 	MALLOC(xp, struct null_node *, sizeof(struct null_node),
236 	    M_NULLFSNODE, M_WAITOK);
237 
238 	error = getnewvnode("null", mp, null_vnodeop_p, &vp);
239 	if (error) {
240 		FREE(xp, M_NULLFSNODE);
241 		return (error);
242 	}
243 
244 	xp->null_vnode = vp;
245 	xp->null_lowervp = lowervp;
246 	xp->null_pending_locks = 0;
247 	xp->null_drain_wakeup = 0;
248 
249 	vp->v_type = lowervp->v_type;
250 	vp->v_data = xp;
251 
252 	/*
253 	 * From NetBSD:
254 	 * Now lock the new node. We rely on the fact that we were passed
255 	 * a locked vnode. If the lower node is exporting a struct lock
256 	 * (v_vnlock != NULL) then we just set the upper v_vnlock to the
257 	 * lower one, and both are now locked. If the lower node is exporting
258 	 * NULL, then we copy that up and manually lock the new vnode.
259 	 */
260 
261 	vp->v_vnlock = lowervp->v_vnlock;
262 	error = VOP_LOCK(vp, LK_EXCLUSIVE | LK_THISLAYER, td);
263 	if (error)
264 		panic("null_nodeget: can't lock new vnode\n");
265 
266 	/*
267 	 * Atomically insert our new node into the hash or vget existing
268 	 * if someone else has beaten us to it.
269 	 */
270 	*vpp = null_hashins(mp, xp);
271 	if (*vpp != NULL) {
272 		vrele(lowervp);
273 		VOP_UNLOCK(vp, LK_THISLAYER, td);
274 		vp->v_vnlock = NULL;
275 		xp->null_lowervp = NULL;
276 		vrele(vp);
277 		return (0);
278 	}
279 
280 	/*
281 	 * XXX We take extra vref just to workaround UFS's XXX:
282 	 * UFS can vrele() vnode in VOP_CLOSE() in some cases. Luckily, this
283 	 * can only happen if v_usecount == 1. To workaround, we just don't
284 	 * let v_usecount be 1, it will be 2 or more.
285 	 */
286 	VREF(lowervp);
287 
288 	*vpp = vp;
289 
290 	return (0);
291 }
292 
293 /*
294  * Remove node from hash.
295  */
296 void
297 null_hashrem(xp)
298 	struct null_node *xp;
299 {
300 
301 	mtx_lock(&null_hashmtx);
302 	LIST_REMOVE(xp, null_hash);
303 	mtx_unlock(&null_hashmtx);
304 }
305 
306 #ifdef DIAGNOSTIC
307 
308 #ifdef KDB
309 #define	null_checkvp_barrier	1
310 #else
311 #define	null_checkvp_barrier	0
312 #endif
313 
314 struct vnode *
315 null_checkvp(vp, fil, lno)
316 	struct vnode *vp;
317 	char *fil;
318 	int lno;
319 {
320 	struct null_node *a = VTONULL(vp);
321 #ifdef notyet
322 	/*
323 	 * Can't do this check because vop_reclaim runs
324 	 * with a funny vop vector.
325 	 */
326 	if (vp->v_op != null_vnodeop_p) {
327 		printf ("null_checkvp: on non-null-node\n");
328 		while (null_checkvp_barrier) /*WAIT*/ ;
329 		panic("null_checkvp");
330 	};
331 #endif
332 	if (a->null_lowervp == NULLVP) {
333 		/* Should never happen */
334 		int i; u_long *p;
335 		printf("vp = %p, ZERO ptr\n", (void *)vp);
336 		for (p = (u_long *) a, i = 0; i < 8; i++)
337 			printf(" %lx", p[i]);
338 		printf("\n");
339 		/* wait for debugger */
340 		while (null_checkvp_barrier) /*WAIT*/ ;
341 		panic("null_checkvp");
342 	}
343 	if (vrefcnt(a->null_lowervp) < 1) {
344 		int i; u_long *p;
345 		printf("vp = %p, unref'ed lowervp\n", (void *)vp);
346 		for (p = (u_long *) a, i = 0; i < 8; i++)
347 			printf(" %lx", p[i]);
348 		printf("\n");
349 		/* wait for debugger */
350 		while (null_checkvp_barrier) /*WAIT*/ ;
351 		panic ("null with unref'ed lowervp");
352 	};
353 #ifdef notyet
354 	printf("null %x/%d -> %x/%d [%s, %d]\n",
355 	        NULLTOV(a), vrefcnt(NULLTOV(a)),
356 		a->null_lowervp, vrefcnt(a->null_lowervp),
357 		fil, lno);
358 #endif
359 	return a->null_lowervp;
360 }
361 #endif
362