xref: /freebsd/sys/fs/nullfs/null_subr.c (revision f1fa1ba3d0f52d88fd30d242e58297fe85f66d2d)
1d167cf6fSWarner Losh /*-
251369649SPedro F. Giffuni  * SPDX-License-Identifier: BSD-3-Clause
351369649SPedro F. Giffuni  *
4df8bae1dSRodney W. Grimes  * Copyright (c) 1992, 1993
5df8bae1dSRodney W. Grimes  *	The Regents of the University of California.  All rights reserved.
6df8bae1dSRodney W. Grimes  *
7df8bae1dSRodney W. Grimes  * This code is derived from software donated to Berkeley by
8df8bae1dSRodney W. Grimes  * Jan-Simon Pendry.
9df8bae1dSRodney W. Grimes  *
10df8bae1dSRodney W. Grimes  * Redistribution and use in source and binary forms, with or without
11df8bae1dSRodney W. Grimes  * modification, are permitted provided that the following conditions
12df8bae1dSRodney W. Grimes  * are met:
13df8bae1dSRodney W. Grimes  * 1. Redistributions of source code must retain the above copyright
14df8bae1dSRodney W. Grimes  *    notice, this list of conditions and the following disclaimer.
15df8bae1dSRodney W. Grimes  * 2. Redistributions in binary form must reproduce the above copyright
16df8bae1dSRodney W. Grimes  *    notice, this list of conditions and the following disclaimer in the
17df8bae1dSRodney W. Grimes  *    documentation and/or other materials provided with the distribution.
18fbbd9655SWarner Losh  * 3. Neither the name of the University nor the names of its contributors
19df8bae1dSRodney W. Grimes  *    may be used to endorse or promote products derived from this software
20df8bae1dSRodney W. Grimes  *    without specific prior written permission.
21df8bae1dSRodney W. Grimes  *
22df8bae1dSRodney W. Grimes  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23df8bae1dSRodney W. Grimes  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24df8bae1dSRodney W. Grimes  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25df8bae1dSRodney W. Grimes  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26df8bae1dSRodney W. Grimes  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27df8bae1dSRodney W. Grimes  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28df8bae1dSRodney W. Grimes  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29df8bae1dSRodney W. Grimes  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30df8bae1dSRodney W. Grimes  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31df8bae1dSRodney W. Grimes  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32df8bae1dSRodney W. Grimes  * SUCH DAMAGE.
33df8bae1dSRodney W. Grimes  *
34996c772fSJohn Dyson  *	@(#)null_subr.c	8.7 (Berkeley) 5/14/95
35df8bae1dSRodney W. Grimes  *
36c3aac50fSPeter Wemm  * $FreeBSD$
37df8bae1dSRodney W. Grimes  */
38df8bae1dSRodney W. Grimes 
39df8bae1dSRodney W. Grimes #include <sys/param.h>
40df8bae1dSRodney W. Grimes #include <sys/systm.h>
418da80660SBoris Popov #include <sys/kernel.h>
42fb919e4dSMark Murray #include <sys/lock.h>
43cd29b292SMateusz Guzik #include <sys/rwlock.h>
44df8bae1dSRodney W. Grimes #include <sys/malloc.h>
45fb919e4dSMark Murray #include <sys/mount.h>
46fb919e4dSMark Murray #include <sys/proc.h>
47fb919e4dSMark Murray #include <sys/vnode.h>
48fb919e4dSMark Murray 
4999d300a1SRuslan Ermilov #include <fs/nullfs/null.h>
50df8bae1dSRodney W. Grimes 
51df8bae1dSRodney W. Grimes /*
52df8bae1dSRodney W. Grimes  * Null layer cache:
53df8bae1dSRodney W. Grimes  * Each cache entry holds a reference to the lower vnode
54df8bae1dSRodney W. Grimes  * along with a pointer to the alias vnode.  When an
55df8bae1dSRodney W. Grimes  * entry is added the lower vnode is VREF'd.  When the
56df8bae1dSRodney W. Grimes  * alias is removed the lower vnode is vrele'd.
57df8bae1dSRodney W. Grimes  */
58df8bae1dSRodney W. Grimes 
59603f963eSKonstantin Belousov #define	NULL_NHASH(vp) (&null_node_hashtbl[vfs_hash_index(vp) & null_hash_mask])
608da80660SBoris Popov 
61e3975643SJake Burkholder static LIST_HEAD(null_node_hashhead, null_node) *null_node_hashtbl;
62cd29b292SMateusz Guzik static struct rwlock null_hash_lock;
63603f963eSKonstantin Belousov static u_long null_hash_mask;
648da80660SBoris Popov 
655bb84bc8SRobert Watson static MALLOC_DEFINE(M_NULLFSHASH, "nullfs_hash", "NULLFS hash table");
665bb84bc8SRobert Watson MALLOC_DEFINE(M_NULLFSNODE, "nullfs_node", "NULLFS vnode private part");
67df8bae1dSRodney W. Grimes 
6854939875STim J. Robbins static struct vnode * null_hashins(struct mount *, struct null_node *);
699b5e8b3aSBruce Evans 
70df8bae1dSRodney W. Grimes /*
71df8bae1dSRodney W. Grimes  * Initialise cache headers
72df8bae1dSRodney W. Grimes  */
7326f9a767SRodney W. Grimes int
74996c772fSJohn Dyson nullfs_init(vfsp)
75996c772fSJohn Dyson 	struct vfsconf *vfsp;
76df8bae1dSRodney W. Grimes {
77996c772fSJohn Dyson 
78603f963eSKonstantin Belousov 	null_node_hashtbl = hashinit(desiredvnodes, M_NULLFSHASH,
79603f963eSKonstantin Belousov 	    &null_hash_mask);
80cd29b292SMateusz Guzik 	rw_init(&null_hash_lock, "nullhs");
818da80660SBoris Popov 	return (0);
828da80660SBoris Popov }
838da80660SBoris Popov 
848da80660SBoris Popov int
858da80660SBoris Popov nullfs_uninit(vfsp)
868da80660SBoris Popov 	struct vfsconf *vfsp;
878da80660SBoris Popov {
888da80660SBoris Popov 
89cd29b292SMateusz Guzik 	rw_destroy(&null_hash_lock);
90603f963eSKonstantin Belousov 	hashdestroy(null_node_hashtbl, M_NULLFSHASH, null_hash_mask);
9126f9a767SRodney W. Grimes 	return (0);
92df8bae1dSRodney W. Grimes }
93df8bae1dSRodney W. Grimes 
94df8bae1dSRodney W. Grimes /*
95df8bae1dSRodney W. Grimes  * Return a VREF'ed alias for lower vnode if already exists, else 0.
964451405fSBoris Popov  * Lower vnode should be locked on entry and will be left locked on exit.
97df8bae1dSRodney W. Grimes  */
98d9e9650aSKonstantin Belousov struct vnode *
9954939875STim J. Robbins null_hashget(mp, lowervp)
10054939875STim J. Robbins 	struct mount *mp;
101df8bae1dSRodney W. Grimes 	struct vnode *lowervp;
102df8bae1dSRodney W. Grimes {
103996c772fSJohn Dyson 	struct null_node_hashhead *hd;
104df8bae1dSRodney W. Grimes 	struct null_node *a;
105df8bae1dSRodney W. Grimes 	struct vnode *vp;
1069c12e631SJeff Roberson 
1079c12e631SJeff Roberson 	ASSERT_VOP_LOCKED(lowervp, "null_hashget");
108df8bae1dSRodney W. Grimes 
109df8bae1dSRodney W. Grimes 	/*
110df8bae1dSRodney W. Grimes 	 * Find hash base, and then search the (two-way) linked
111df8bae1dSRodney W. Grimes 	 * list looking for a null_node structure which is referencing
112df8bae1dSRodney W. Grimes 	 * the lower vnode.  If found, the increment the null_node
113df8bae1dSRodney W. Grimes 	 * reference count (but NOT the lower vnode's VREF counter).
114df8bae1dSRodney W. Grimes 	 */
115996c772fSJohn Dyson 	hd = NULL_NHASH(lowervp);
1161e0006e4SMateusz Guzik 	if (LIST_EMPTY(hd))
1171e0006e4SMateusz Guzik 		return (NULLVP);
118cd29b292SMateusz Guzik 	rw_rlock(&null_hash_lock);
119fc2ffbe6SPoul-Henning Kamp 	LIST_FOREACH(a, hd, null_hash) {
12054939875STim J. Robbins 		if (a->null_lowervp == lowervp && NULLTOV(a)->v_mount == mp) {
1219c12e631SJeff Roberson 			/*
1229c12e631SJeff Roberson 			 * Since we have the lower node locked the nullfs
1239c12e631SJeff Roberson 			 * node can not be in the process of recycling.  If
1249c12e631SJeff Roberson 			 * it had been recycled before we grabed the lower
1259c12e631SJeff Roberson 			 * lock it would not have been found on the hash.
1269c12e631SJeff Roberson 			 */
1274c65d593SJeff Roberson 			vp = NULLTOV(a);
1284c65d593SJeff Roberson 			vref(vp);
129cd29b292SMateusz Guzik 			rw_runlock(&null_hash_lock);
130df8bae1dSRodney W. Grimes 			return (vp);
131df8bae1dSRodney W. Grimes 		}
132df8bae1dSRodney W. Grimes 	}
133cd29b292SMateusz Guzik 	rw_runlock(&null_hash_lock);
1341cfdefbbSSemen Ustimenko 	return (NULLVP);
135df8bae1dSRodney W. Grimes }
136df8bae1dSRodney W. Grimes 
1371cfdefbbSSemen Ustimenko /*
1381cfdefbbSSemen Ustimenko  * Act like null_hashget, but add passed null_node to hash if no existing
1391cfdefbbSSemen Ustimenko  * node found.
1401cfdefbbSSemen Ustimenko  */
1411cfdefbbSSemen Ustimenko static struct vnode *
14254939875STim J. Robbins null_hashins(mp, xp)
14354939875STim J. Robbins 	struct mount *mp;
1441cfdefbbSSemen Ustimenko 	struct null_node *xp;
1451cfdefbbSSemen Ustimenko {
1461cfdefbbSSemen Ustimenko 	struct null_node_hashhead *hd;
1471cfdefbbSSemen Ustimenko 	struct null_node *oxp;
1481cfdefbbSSemen Ustimenko 	struct vnode *ovp;
1491cfdefbbSSemen Ustimenko 
1501cfdefbbSSemen Ustimenko 	hd = NULL_NHASH(xp->null_lowervp);
151cd29b292SMateusz Guzik 	rw_wlock(&null_hash_lock);
1521cfdefbbSSemen Ustimenko 	LIST_FOREACH(oxp, hd, null_hash) {
15354939875STim J. Robbins 		if (oxp->null_lowervp == xp->null_lowervp &&
15454939875STim J. Robbins 		    NULLTOV(oxp)->v_mount == mp) {
1559c12e631SJeff Roberson 			/*
1569c12e631SJeff Roberson 			 * See null_hashget for a description of this
1579c12e631SJeff Roberson 			 * operation.
1589c12e631SJeff Roberson 			 */
1591cfdefbbSSemen Ustimenko 			ovp = NULLTOV(oxp);
1604c65d593SJeff Roberson 			vref(ovp);
161cd29b292SMateusz Guzik 			rw_wunlock(&null_hash_lock);
1621cfdefbbSSemen Ustimenko 			return (ovp);
1631cfdefbbSSemen Ustimenko 		}
1641cfdefbbSSemen Ustimenko 	}
1651cfdefbbSSemen Ustimenko 	LIST_INSERT_HEAD(hd, xp, null_hash);
166cd29b292SMateusz Guzik 	rw_wunlock(&null_hash_lock);
1671cfdefbbSSemen Ustimenko 	return (NULLVP);
1681cfdefbbSSemen Ustimenko }
169df8bae1dSRodney W. Grimes 
17061b9d89fSTor Egge static void
17167e3d54fSKonstantin Belousov null_destroy_proto(struct vnode *vp, void *xp)
17267e3d54fSKonstantin Belousov {
17367e3d54fSKonstantin Belousov 
17466f02f4bSKonstantin Belousov 	lockmgr(&vp->v_lock, LK_EXCLUSIVE, NULL);
17567e3d54fSKonstantin Belousov 	VI_LOCK(vp);
17667e3d54fSKonstantin Belousov 	vp->v_data = NULL;
17767e3d54fSKonstantin Belousov 	vp->v_vnlock = &vp->v_lock;
17867e3d54fSKonstantin Belousov 	vp->v_op = &dead_vnodeops;
17967e3d54fSKonstantin Belousov 	VI_UNLOCK(vp);
18067e3d54fSKonstantin Belousov 	vgone(vp);
18167e3d54fSKonstantin Belousov 	vput(vp);
18267e3d54fSKonstantin Belousov 	free(xp, M_NULLFSNODE);
18367e3d54fSKonstantin Belousov }
18467e3d54fSKonstantin Belousov 
18567e3d54fSKonstantin Belousov static void
18661b9d89fSTor Egge null_insmntque_dtr(struct vnode *vp, void *xp)
18761b9d89fSTor Egge {
188dd0f9532SKonstantin Belousov 
189dd0f9532SKonstantin Belousov 	vput(((struct null_node *)xp)->null_lowervp);
19067e3d54fSKonstantin Belousov 	null_destroy_proto(vp, xp);
19161b9d89fSTor Egge }
19261b9d89fSTor Egge 
193df8bae1dSRodney W. Grimes /*
1941cfdefbbSSemen Ustimenko  * Make a new or get existing nullfs node.
1951cfdefbbSSemen Ustimenko  * Vp is the alias vnode, lowervp is the lower vnode.
1961cfdefbbSSemen Ustimenko  *
1971cfdefbbSSemen Ustimenko  * The lowervp assumed to be locked and having "spare" reference. This routine
1981cfdefbbSSemen Ustimenko  * vrele lowervp if nullfs node was taken from hash. Otherwise it "transfers"
1991cfdefbbSSemen Ustimenko  * the caller's "spare" reference to created nullfs vnode.
200df8bae1dSRodney W. Grimes  */
2011cfdefbbSSemen Ustimenko int
2021cfdefbbSSemen Ustimenko null_nodeget(mp, lowervp, vpp)
203df8bae1dSRodney W. Grimes 	struct mount *mp;
204df8bae1dSRodney W. Grimes 	struct vnode *lowervp;
205df8bae1dSRodney W. Grimes 	struct vnode **vpp;
206df8bae1dSRodney W. Grimes {
207df8bae1dSRodney W. Grimes 	struct null_node *xp;
2081cfdefbbSSemen Ustimenko 	struct vnode *vp;
209df8bae1dSRodney W. Grimes 	int error;
210df8bae1dSRodney W. Grimes 
211d9e9650aSKonstantin Belousov 	ASSERT_VOP_LOCKED(lowervp, "lowervp");
212*f1fa1ba3SMateusz Guzik 	VNPASS(lowervp->v_usecount > 0, lowervp);
21348a1e3f6SKonstantin Belousov 
214d9e9650aSKonstantin Belousov 	/* Lookup the hash firstly. */
21554939875STim J. Robbins 	*vpp = null_hashget(mp, lowervp);
2161cfdefbbSSemen Ustimenko 	if (*vpp != NULL) {
2171cfdefbbSSemen Ustimenko 		vrele(lowervp);
2181cfdefbbSSemen Ustimenko 		return (0);
2191cfdefbbSSemen Ustimenko 	}
2201cfdefbbSSemen Ustimenko 
2211cfdefbbSSemen Ustimenko 	/*
222d9e9650aSKonstantin Belousov 	 * The insmntque1() call below requires the exclusive lock on
223d9e9650aSKonstantin Belousov 	 * the nullfs vnode.  Upgrade the lock now if hash failed to
224d9e9650aSKonstantin Belousov 	 * provide ready to use vnode.
225d9e9650aSKonstantin Belousov 	 */
226d9e9650aSKonstantin Belousov 	if (VOP_ISLOCKED(lowervp) != LK_EXCLUSIVE) {
227d9e9650aSKonstantin Belousov 		vn_lock(lowervp, LK_UPGRADE | LK_RETRY);
228abd80ddbSMateusz Guzik 		if (VN_IS_DOOMED(lowervp)) {
229d9e9650aSKonstantin Belousov 			vput(lowervp);
230d9e9650aSKonstantin Belousov 			return (ENOENT);
231d9e9650aSKonstantin Belousov 		}
232d9e9650aSKonstantin Belousov 	}
233d9e9650aSKonstantin Belousov 
234d9e9650aSKonstantin Belousov 	/*
2351cfdefbbSSemen Ustimenko 	 * We do not serialize vnode creation, instead we will check for
2361cfdefbbSSemen Ustimenko 	 * duplicates later, when adding new vnode to hash.
2371cfdefbbSSemen Ustimenko 	 * Note that duplicate can only appear in hash if the lowervp is
2381cfdefbbSSemen Ustimenko 	 * locked LK_SHARED.
2392f9bae59SDavid Greenman 	 */
240d9e9650aSKonstantin Belousov 	xp = malloc(sizeof(struct null_node), M_NULLFSNODE, M_WAITOK);
2412f9bae59SDavid Greenman 
242e583d999SEdward Tomasz Napierala 	error = getnewvnode("nullfs", mp, &null_vnodeops, &vp);
2432f9bae59SDavid Greenman 	if (error) {
244dd0f9532SKonstantin Belousov 		vput(lowervp);
2451ede983cSDag-Erling Smørgrav 		free(xp, M_NULLFSNODE);
246df8bae1dSRodney W. Grimes 		return (error);
2472f9bae59SDavid Greenman 	}
248df8bae1dSRodney W. Grimes 
249df8bae1dSRodney W. Grimes 	xp->null_vnode = vp;
250df8bae1dSRodney W. Grimes 	xp->null_lowervp = lowervp;
2510fc6daa7SKonstantin Belousov 	xp->null_flags = 0;
25208720e34SSemen Ustimenko 	vp->v_type = lowervp->v_type;
25308720e34SSemen Ustimenko 	vp->v_data = xp;
2544451405fSBoris Popov 	vp->v_vnlock = lowervp->v_vnlock;
25561b9d89fSTor Egge 	error = insmntque1(vp, mp, null_insmntque_dtr, xp);
25661b9d89fSTor Egge 	if (error != 0)
25761b9d89fSTor Egge 		return (error);
258dc1d2cc6SKonstantin Belousov 	if (lowervp == MOUNTTONULLMOUNT(mp)->nullm_lowerrootvp)
259dc1d2cc6SKonstantin Belousov 		vp->v_vflag |= VV_ROOT;
260dc1d2cc6SKonstantin Belousov 
261df8bae1dSRodney W. Grimes 	/*
2621cfdefbbSSemen Ustimenko 	 * Atomically insert our new node into the hash or vget existing
2631cfdefbbSSemen Ustimenko 	 * if someone else has beaten us to it.
264df8bae1dSRodney W. Grimes 	 */
26554939875STim J. Robbins 	*vpp = null_hashins(mp, xp);
2661cfdefbbSSemen Ustimenko 	if (*vpp != NULL) {
2674451405fSBoris Popov 		vrele(lowervp);
26867e3d54fSKonstantin Belousov 		null_destroy_proto(vp, xp);
269df8bae1dSRodney W. Grimes 		return (0);
270df8bae1dSRodney W. Grimes 	}
2711cfdefbbSSemen Ustimenko 	*vpp = vp;
2721cfdefbbSSemen Ustimenko 
2731cfdefbbSSemen Ustimenko 	return (0);
2741cfdefbbSSemen Ustimenko }
2751cfdefbbSSemen Ustimenko 
2761cfdefbbSSemen Ustimenko /*
2771cfdefbbSSemen Ustimenko  * Remove node from hash.
2781cfdefbbSSemen Ustimenko  */
27908720e34SSemen Ustimenko void
28008720e34SSemen Ustimenko null_hashrem(xp)
28108720e34SSemen Ustimenko 	struct null_node *xp;
28208720e34SSemen Ustimenko {
28308720e34SSemen Ustimenko 
284cd29b292SMateusz Guzik 	rw_wlock(&null_hash_lock);
28508720e34SSemen Ustimenko 	LIST_REMOVE(xp, null_hash);
286cd29b292SMateusz Guzik 	rw_wunlock(&null_hash_lock);
28708720e34SSemen Ustimenko }
28808720e34SSemen Ustimenko 
289a0f40f54SBruce Evans #ifdef DIAGNOSTIC
2901bf978ceSKATO Takenori 
291df8bae1dSRodney W. Grimes struct vnode *
292df8bae1dSRodney W. Grimes null_checkvp(vp, fil, lno)
293df8bae1dSRodney W. Grimes 	struct vnode *vp;
294df8bae1dSRodney W. Grimes 	char *fil;
295df8bae1dSRodney W. Grimes 	int lno;
296df8bae1dSRodney W. Grimes {
297df8bae1dSRodney W. Grimes 	struct null_node *a = VTONULL(vp);
298b9131889SKonstantin Belousov 
299df8bae1dSRodney W. Grimes #ifdef notyet
300df8bae1dSRodney W. Grimes 	/*
301df8bae1dSRodney W. Grimes 	 * Can't do this check because vop_reclaim runs
302df8bae1dSRodney W. Grimes 	 * with a funny vop vector.
303df8bae1dSRodney W. Grimes 	 */
304df8bae1dSRodney W. Grimes 	if (vp->v_op != null_vnodeop_p) {
305df8bae1dSRodney W. Grimes 		printf ("null_checkvp: on non-null-node\n");
306df8bae1dSRodney W. Grimes 		panic("null_checkvp");
307b9131889SKonstantin Belousov 	}
308df8bae1dSRodney W. Grimes #endif
309c5e17d9eSKATO Takenori 	if (a->null_lowervp == NULLVP) {
310df8bae1dSRodney W. Grimes 		/* Should never happen */
3114d2310ddSKonstantin Belousov 		panic("null_checkvp %p", vp);
312df8bae1dSRodney W. Grimes 	}
313b9131889SKonstantin Belousov 	VI_LOCK_FLAGS(a->null_lowervp, MTX_DUPOK);
3144d2310ddSKonstantin Belousov 	if (a->null_lowervp->v_usecount < 1)
3154d2310ddSKonstantin Belousov 		panic ("null with unref'ed lowervp, vp %p lvp %p",
3164d2310ddSKonstantin Belousov 		    vp, a->null_lowervp);
317b9131889SKonstantin Belousov 	VI_UNLOCK(a->null_lowervp);
318df8bae1dSRodney W. Grimes #ifdef notyet
319df8bae1dSRodney W. Grimes 	printf("null %x/%d -> %x/%d [%s, %d]\n",
3204d93c0beSJeff Roberson 	        NULLTOV(a), vrefcnt(NULLTOV(a)),
3214d93c0beSJeff Roberson 		a->null_lowervp, vrefcnt(a->null_lowervp),
322df8bae1dSRodney W. Grimes 		fil, lno);
323df8bae1dSRodney W. Grimes #endif
324b9131889SKonstantin Belousov 	return (a->null_lowervp);
325df8bae1dSRodney W. Grimes }
326df8bae1dSRodney W. Grimes #endif
327