vfs_cache.c revision 74384
1266077Sdes/*
2266077Sdes * Copyright (c) 1989, 1993, 1995
3266077Sdes *	The Regents of the University of California.  All rights reserved.
4266077Sdes *
5266077Sdes * This code is derived from software contributed to Berkeley by
6266077Sdes * Poul-Henning Kamp of the FreeBSD Project.
7266077Sdes *
8266077Sdes * Redistribution and use in source and binary forms, with or without
9266077Sdes * modification, are permitted provided that the following conditions
10266077Sdes * are met:
11266077Sdes * 1. Redistributions of source code must retain the above copyright
12266077Sdes *    notice, this list of conditions and the following disclaimer.
13266077Sdes * 2. Redistributions in binary form must reproduce the above copyright
14266077Sdes *    notice, this list of conditions and the following disclaimer in the
15266077Sdes *    documentation and/or other materials provided with the distribution.
16266077Sdes * 3. All advertising materials mentioning features or use of this software
17266077Sdes *    must display the following acknowledgement:
18266077Sdes *	This product includes software developed by the University of
19266077Sdes *	California, Berkeley and its contributors.
20266077Sdes * 4. Neither the name of the University nor the names of its contributors
21266077Sdes *    may be used to endorse or promote products derived from this software
22266077Sdes *    without specific prior written permission.
23266077Sdes *
24266077Sdes * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25266077Sdes * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26266077Sdes * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27266077Sdes * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28266077Sdes * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29266077Sdes * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30266077Sdes * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31266077Sdes * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32266077Sdes * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33266077Sdes * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34266077Sdes * SUCH DAMAGE.
35266077Sdes *
36266077Sdes *	@(#)vfs_cache.c	8.5 (Berkeley) 3/22/95
37266077Sdes * $FreeBSD: head/sys/kern/vfs_cache.c 74384 2001-03-17 09:31:06Z peter $
38266077Sdes */
39266077Sdes
40266077Sdes#include <sys/param.h>
41266077Sdes#include <sys/systm.h>
42266077Sdes#include <sys/kernel.h>
43266077Sdes#include <sys/sysctl.h>
44266077Sdes#include <sys/mount.h>
45266077Sdes#include <sys/vnode.h>
46266077Sdes#include <sys/namei.h>
47266077Sdes#include <sys/malloc.h>
48266077Sdes#include <sys/sysproto.h>
49266077Sdes#include <sys/proc.h>
50266077Sdes#include <sys/filedesc.h>
51266077Sdes#include <sys/fnv_hash.h>
52266077Sdes
53266077Sdes/*
54266077Sdes * This structure describes the elements in the cache of recent
55266077Sdes * names looked up by namei.
56266077Sdes */
57266077Sdes
58266077Sdesstruct	namecache {
59266077Sdes	LIST_ENTRY(namecache) nc_hash;	/* hash chain */
60266077Sdes	LIST_ENTRY(namecache) nc_src;	/* source vnode list */
61266077Sdes	TAILQ_ENTRY(namecache) nc_dst;	/* destination vnode list */
62266077Sdes	struct	vnode *nc_dvp;		/* vnode of parent of name */
63266077Sdes	struct	vnode *nc_vp;		/* vnode the name refers to */
64266077Sdes	u_char	nc_flag;		/* flag bits */
65266077Sdes	u_char	nc_nlen;		/* length of name */
66266077Sdes	char	nc_name[0];		/* segment name */
67266077Sdes};
68266077Sdes
69266077Sdes/*
70266077Sdes * Name caching works as follows:
71266077Sdes *
72266077Sdes * Names found by directory scans are retained in a cache
73266077Sdes * for future reference.  It is managed LRU, so frequently
74266077Sdes * used names will hang around.  Cache is indexed by hash value
75266077Sdes * obtained from (vp, name) where vp refers to the directory
76266077Sdes * containing name.
77266077Sdes *
78266077Sdes * If it is a "negative" entry, (i.e. for a name that is known NOT to
79266077Sdes * exist) the vnode pointer will be NULL.
80266077Sdes *
81266077Sdes * Upon reaching the last segment of a path, if the reference
82266077Sdes * is for DELETE, or NOCACHE is set (rewrite), and the
83266077Sdes * name is located in the cache, it will be dropped.
84266077Sdes */
85266077Sdes
86266077Sdes/*
87266077Sdes * Structures associated with name cacheing.
88266077Sdes */
89266077Sdes#define NCHHASH(dvp, hash) \
90266077Sdes	(&nchashtbl[((dvp)->v_id + (hash)) & nchash])
91266077Sdesstatic LIST_HEAD(nchashhead, namecache) *nchashtbl;	/* Hash Table */
92266077Sdesstatic TAILQ_HEAD(, namecache) ncneg;	/* Hash Table */
93266077Sdesstatic u_long	nchash;			/* size of hash table */
94266077SdesSYSCTL_ULONG(_debug, OID_AUTO, nchash, CTLFLAG_RD, &nchash, 0, "");
95266077Sdesstatic u_long	ncnegfactor = 16;	/* ratio of negative entries */
96266077SdesSYSCTL_ULONG(_debug, OID_AUTO, ncnegfactor, CTLFLAG_RW, &ncnegfactor, 0, "");
97266077Sdesstatic u_long	numneg;		/* number of cache entries allocated */
98266077SdesSYSCTL_ULONG(_debug, OID_AUTO, numneg, CTLFLAG_RD, &numneg, 0, "");
99266077Sdesstatic u_long	numcache;		/* number of cache entries allocated */
100266077SdesSYSCTL_ULONG(_debug, OID_AUTO, numcache, CTLFLAG_RD, &numcache, 0, "");
101266077Sdesstruct	nchstats nchstats;		/* cache effectiveness statistics */
102266077Sdes
103266077Sdesstatic int	doingcache = 1;		/* 1 => enable the cache */
104266077SdesSYSCTL_INT(_debug, OID_AUTO, vfscache, CTLFLAG_RW, &doingcache, 0, "");
105266077SdesSYSCTL_INT(_debug, OID_AUTO, vnsize, CTLFLAG_RD, 0, sizeof(struct vnode), "");
106266077SdesSYSCTL_INT(_debug, OID_AUTO, ncsize, CTLFLAG_RD, 0, sizeof(struct namecache), "");
107266077Sdes
108266077Sdes/*
109266077Sdes * The new name cache statistics
110266077Sdes */
111266077SdesSYSCTL_NODE(_vfs, OID_AUTO, cache, CTLFLAG_RW, 0, "Name cache statistics");
112266077Sdes#define STATNODE(mode, name, var) \
113266077Sdes	SYSCTL_ULONG(_vfs_cache, OID_AUTO, name, mode, var, 0, "");
114266077SdesSTATNODE(CTLFLAG_RD, numneg, &numneg);
115266077SdesSTATNODE(CTLFLAG_RD, numcache, &numcache);
116266077Sdesstatic u_long numcalls; STATNODE(CTLFLAG_RD, numcalls, &numcalls);
117266077Sdesstatic u_long dothits; STATNODE(CTLFLAG_RD, dothits, &dothits);
118266077Sdesstatic u_long dotdothits; STATNODE(CTLFLAG_RD, dotdothits, &dotdothits);
119266077Sdesstatic u_long numchecks; STATNODE(CTLFLAG_RD, numchecks, &numchecks);
120266077Sdesstatic u_long nummiss; STATNODE(CTLFLAG_RD, nummiss, &nummiss);
121266077Sdesstatic u_long nummisszap; STATNODE(CTLFLAG_RD, nummisszap, &nummisszap);
122266077Sdesstatic u_long numposzaps; STATNODE(CTLFLAG_RD, numposzaps, &numposzaps);
123266077Sdesstatic u_long numposhits; STATNODE(CTLFLAG_RD, numposhits, &numposhits);
124266077Sdesstatic u_long numnegzaps; STATNODE(CTLFLAG_RD, numnegzaps, &numnegzaps);
125266077Sdesstatic u_long numneghits; STATNODE(CTLFLAG_RD, numneghits, &numneghits);
126266077Sdes
127266077SdesSYSCTL_OPAQUE(_vfs_cache, OID_AUTO, nchstats, CTLFLAG_RD, &nchstats,
128266077Sdes        sizeof(nchstats), "LU", "VFS cache effectiveness statistics");
129266077Sdes
130266077Sdes
131266077Sdes
132266077Sdesstatic void cache_zap __P((struct namecache *ncp));
133266077Sdes
134266077Sdesstatic MALLOC_DEFINE(M_VFSCACHE, "vfscache", "VFS name cache entries");
135266077Sdes
136266077Sdes/*
137266077Sdes * Flags in namecache.nc_flag
138266077Sdes */
139266077Sdes#define NCF_WHITE	1
140266077Sdes/*
141266077Sdes * Delete an entry from its hash list and move it to the front
142266077Sdes * of the LRU list for immediate reuse.
143266077Sdes */
144266077Sdesstatic void
145266077Sdescache_zap(ncp)
146266077Sdes	struct namecache *ncp;
147266077Sdes{
148266077Sdes	LIST_REMOVE(ncp, nc_hash);
149266077Sdes	LIST_REMOVE(ncp, nc_src);
150266077Sdes	if (LIST_EMPTY(&ncp->nc_dvp->v_cache_src))
151266077Sdes		vdrop(ncp->nc_dvp);
152266077Sdes	if (ncp->nc_vp) {
153266077Sdes		TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst, ncp, nc_dst);
154266077Sdes	} else {
155266077Sdes		TAILQ_REMOVE(&ncneg, ncp, nc_dst);
156266077Sdes		numneg--;
157266077Sdes	}
158266077Sdes	numcache--;
159266077Sdes	free(ncp, M_VFSCACHE);
160266077Sdes}
161266077Sdes
162266077Sdes/*
163266077Sdes * Lookup an entry in the cache
164266077Sdes *
165266077Sdes * We don't do this if the segment name is long, simply so the cache
166266077Sdes * can avoid holding long names (which would either waste space, or
167266077Sdes * add greatly to the complexity).
168266077Sdes *
169266077Sdes * Lookup is called with dvp pointing to the directory to search,
170266077Sdes * cnp pointing to the name of the entry being sought. If the lookup
171266077Sdes * succeeds, the vnode is returned in *vpp, and a status of -1 is
172266077Sdes * returned. If the lookup determines that the name does not exist
173266077Sdes * (negative cacheing), a status of ENOENT is returned. If the lookup
174266077Sdes * fails, a status of zero is returned.
175266077Sdes */
176266077Sdes
177266077Sdesint
178266077Sdescache_lookup(dvp, vpp, cnp)
179266077Sdes	struct vnode *dvp;
180266077Sdes	struct vnode **vpp;
181266077Sdes	struct componentname *cnp;
182266077Sdes{
183266077Sdes	struct namecache *ncp;
184266077Sdes	u_int32_t hash;
185266077Sdes
186266077Sdes	if (!doingcache) {
187266077Sdes		cnp->cn_flags &= ~MAKEENTRY;
188266077Sdes		return (0);
189266077Sdes	}
190266077Sdes
191266077Sdes	numcalls++;
192266077Sdes
193266077Sdes	if (cnp->cn_nameptr[0] == '.') {
194266077Sdes		if (cnp->cn_namelen == 1) {
195266077Sdes			*vpp = dvp;
196266077Sdes			dothits++;
197266077Sdes			return (-1);
198266077Sdes		}
199266077Sdes		if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
200266077Sdes			dotdothits++;
201266077Sdes			if (dvp->v_dd->v_id != dvp->v_ddid ||
202266077Sdes			    (cnp->cn_flags & MAKEENTRY) == 0) {
203266077Sdes				dvp->v_ddid = 0;
204266077Sdes				return (0);
205266077Sdes			}
206266077Sdes			*vpp = dvp->v_dd;
207266077Sdes			return (-1);
208266077Sdes		}
209266077Sdes	}
210266077Sdes
211266077Sdes	hash = fnv32_hashbuf(cnp->cn_nameptr, cnp->cn_namelen);
212266077Sdes	LIST_FOREACH(ncp, (NCHHASH(dvp, hash)), nc_hash) {
213266077Sdes		numchecks++;
214266077Sdes		if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen &&
215266077Sdes		    !bcmp(ncp->nc_name, cnp->cn_nameptr, ncp->nc_nlen))
216266077Sdes			break;
217266077Sdes	}
218266077Sdes
219266077Sdes	/* We failed to find an entry */
220266077Sdes	if (ncp == 0) {
221266077Sdes		if ((cnp->cn_flags & MAKEENTRY) == 0) {
222266077Sdes			nummisszap++;
223266077Sdes		} else {
224266077Sdes			nummiss++;
225266077Sdes		}
226266077Sdes		nchstats.ncs_miss++;
227266077Sdes		return (0);
228266077Sdes	}
229266077Sdes
230266077Sdes	/* We don't want to have an entry, so dump it */
231266077Sdes	if ((cnp->cn_flags & MAKEENTRY) == 0) {
232266077Sdes		numposzaps++;
233266077Sdes		nchstats.ncs_badhits++;
234266077Sdes		cache_zap(ncp);
235266077Sdes		return (0);
236266077Sdes	}
237266077Sdes
238266077Sdes	/* We found a "positive" match, return the vnode */
239266077Sdes        if (ncp->nc_vp) {
240266077Sdes		numposhits++;
241266077Sdes		nchstats.ncs_goodhits++;
242266077Sdes		*vpp = ncp->nc_vp;
243266077Sdes		return (-1);
244266077Sdes	}
245266077Sdes
246266077Sdes	/* We found a negative match, and want to create it, so purge */
247266077Sdes	if (cnp->cn_nameiop == CREATE) {
248266077Sdes		numnegzaps++;
249266077Sdes		nchstats.ncs_badhits++;
250266077Sdes		cache_zap(ncp);
251266077Sdes		return (0);
252266077Sdes	}
253266077Sdes
254266077Sdes	numneghits++;
255266077Sdes	/*
256266077Sdes	 * We found a "negative" match, ENOENT notifies client of this match.
257266077Sdes	 * The nc_vpid field records whether this is a whiteout.
258266077Sdes	 */
259266077Sdes	TAILQ_REMOVE(&ncneg, ncp, nc_dst);
260266077Sdes	TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
261266077Sdes	nchstats.ncs_neghits++;
262266077Sdes	if (ncp->nc_flag & NCF_WHITE)
263266077Sdes		cnp->cn_flags |= ISWHITEOUT;
264266077Sdes	return (ENOENT);
265266077Sdes}
266266077Sdes
267266077Sdes/*
268266077Sdes * Add an entry to the cache.
269266077Sdes */
270266077Sdesvoid
271266077Sdescache_enter(dvp, vp, cnp)
272266077Sdes	struct vnode *dvp;
273266077Sdes	struct vnode *vp;
274266077Sdes	struct componentname *cnp;
275266077Sdes{
276266077Sdes	struct namecache *ncp;
277266077Sdes	struct nchashhead *ncpp;
278266077Sdes	u_int32_t hash;
279266077Sdes	int len;
280266077Sdes
281266077Sdes	if (!doingcache)
282266077Sdes		return;
283266077Sdes
284266077Sdes	if (cnp->cn_nameptr[0] == '.') {
285266077Sdes		if (cnp->cn_namelen == 1) {
286266077Sdes			return;
287266077Sdes		}
288266077Sdes		if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
289266077Sdes			if (vp) {
290266077Sdes				dvp->v_dd = vp;
291266077Sdes				dvp->v_ddid = vp->v_id;
292266077Sdes			} else {
293266077Sdes				dvp->v_dd = dvp;
294266077Sdes				dvp->v_ddid = 0;
295266077Sdes			}
296266077Sdes			return;
297266077Sdes		}
298266077Sdes	}
299266077Sdes
300266077Sdes	ncp = (struct namecache *)
301266077Sdes		malloc(sizeof *ncp + cnp->cn_namelen, M_VFSCACHE, M_WAITOK);
302266077Sdes	bzero((char *)ncp, sizeof *ncp);
303266077Sdes	numcache++;
304266077Sdes	if (!vp) {
305266077Sdes		numneg++;
306266077Sdes		ncp->nc_flag = cnp->cn_flags & ISWHITEOUT ? NCF_WHITE : 0;
307266077Sdes	} else if (vp->v_type == VDIR) {
308266077Sdes		vp->v_dd = dvp;
309266077Sdes		vp->v_ddid = dvp->v_id;
310266077Sdes	}
311266077Sdes
312266077Sdes	/*
313266077Sdes	 * Fill in cache info, if vp is NULL this is a "negative" cache entry.
314266077Sdes	 * For negative entries, we have to record whether it is a whiteout.
315266077Sdes	 * the whiteout flag is stored in the nc_vpid field which is
316266077Sdes	 * otherwise unused.
317266077Sdes	 */
318266077Sdes	ncp->nc_vp = vp;
319266077Sdes	ncp->nc_dvp = dvp;
320266077Sdes	len = ncp->nc_nlen = cnp->cn_namelen;
321266077Sdes	hash = fnv32_hashbuf(cnp->cn_nameptr, len);
322266077Sdes	bcopy(cnp->cn_nameptr, ncp->nc_name, len);
323266077Sdes	ncpp = NCHHASH(dvp, hash);
324266077Sdes	LIST_INSERT_HEAD(ncpp, ncp, nc_hash);
325266077Sdes	if (LIST_EMPTY(&dvp->v_cache_src))
326266077Sdes		vhold(dvp);
327266077Sdes	LIST_INSERT_HEAD(&dvp->v_cache_src, ncp, nc_src);
328266077Sdes	if (vp) {
329266077Sdes		TAILQ_INSERT_HEAD(&vp->v_cache_dst, ncp, nc_dst);
330266077Sdes	} else {
331266077Sdes		TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
332266077Sdes	}
333266077Sdes	if (numneg * ncnegfactor > numcache) {
334266077Sdes		ncp = TAILQ_FIRST(&ncneg);
335266077Sdes		cache_zap(ncp);
336266077Sdes	}
337266077Sdes}
338266077Sdes
339266077Sdes/*
340266077Sdes * Name cache initialization, from vfs_init() when we are booting
341266077Sdes */
342266077Sdesstatic void
343266077Sdesnchinit(void *dummy __unused)
344266077Sdes{
345266077Sdes
346266077Sdes	TAILQ_INIT(&ncneg);
347266077Sdes	nchashtbl = hashinit(desiredvnodes * 2, M_VFSCACHE, &nchash);
348266077Sdes}
349266077SdesSYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nchinit, NULL)
350266077Sdes
351266077Sdes
352266077Sdes/*
353266077Sdes * Invalidate all entries to a particular vnode.
354266077Sdes *
355266077Sdes * Remove all entries in the namecache relating to this vnode and
356266077Sdes * change the v_id.  We take the v_id from a global counter, since
357266077Sdes * it becomes a handy sequence number in crash-dumps that way.
358266077Sdes * No valid vnode will ever have (v_id == 0).
359266077Sdes *
360266077Sdes * XXX: Only time and the size of v_id prevents this from failing:
361266077Sdes * XXX: In theory we should hunt down all (struct vnode*, v_id)
362266077Sdes * XXX: soft references and nuke them, at least on the global
363266077Sdes * XXX: v_id wraparound.  The period of resistance can be extended
364285206Sdes * XXX: by incrementing each vnodes v_id individually instead of
365285206Sdes * XXX: using the global v_id.
366266077Sdes */
367266077Sdes
368266077Sdesvoid
369266077Sdescache_purge(vp)
370266077Sdes	struct vnode *vp;
371266077Sdes{
372266077Sdes	static u_long nextid;
373266077Sdes
374266077Sdes	while (!LIST_EMPTY(&vp->v_cache_src))
375266077Sdes		cache_zap(LIST_FIRST(&vp->v_cache_src));
376266077Sdes	while (!TAILQ_EMPTY(&vp->v_cache_dst))
377266077Sdes		cache_zap(TAILQ_FIRST(&vp->v_cache_dst));
378266077Sdes
379266077Sdes	do
380266077Sdes		nextid++;
381266077Sdes	while (nextid == vp->v_id || !nextid);
382266077Sdes	vp->v_id = nextid;
383266077Sdes	vp->v_dd = vp;
384266077Sdes	vp->v_ddid = 0;
385266077Sdes}
386266077Sdes
387266077Sdes/*
388266077Sdes * Flush all entries referencing a particular filesystem.
389266077Sdes *
390266077Sdes * Since we need to check it anyway, we will flush all the invalid
391266077Sdes * entries at the same time.
392266077Sdes */
393266077Sdesvoid
394266077Sdescache_purgevfs(mp)
395266077Sdes	struct mount *mp;
396266077Sdes{
397266077Sdes	struct nchashhead *ncpp;
398266077Sdes	struct namecache *ncp, *nnp;
399266077Sdes
400266077Sdes	/* Scan hash tables for applicable entries */
401266077Sdes	for (ncpp = &nchashtbl[nchash]; ncpp >= nchashtbl; ncpp--) {
402266077Sdes		for (ncp = LIST_FIRST(ncpp); ncp != 0; ncp = nnp) {
403266077Sdes			nnp = LIST_NEXT(ncp, nc_hash);
404266077Sdes			if (ncp->nc_dvp->v_mount == mp) {
405266077Sdes				cache_zap(ncp);
406266077Sdes			}
407266077Sdes		}
408266077Sdes	}
409266077Sdes}
410266077Sdes
411266077Sdes/*
412266077Sdes * Perform canonical checks and cache lookup and pass on to filesystem
413266077Sdes * through the vop_cachedlookup only if needed.
414266077Sdes */
415266077Sdes
416266077Sdesint
417266077Sdesvfs_cache_lookup(ap)
418266077Sdes	struct vop_lookup_args /* {
419266077Sdes		struct vnode *a_dvp;
420266077Sdes		struct vnode **a_vpp;
421266077Sdes		struct componentname *a_cnp;
422266077Sdes	} */ *ap;
423266077Sdes{
424266077Sdes	struct vnode *dvp, *vp;
425266077Sdes	int lockparent;
426266077Sdes	int error;
427266077Sdes	struct vnode **vpp = ap->a_vpp;
428266077Sdes	struct componentname *cnp = ap->a_cnp;
429266077Sdes	struct ucred *cred = cnp->cn_cred;
430266077Sdes	int flags = cnp->cn_flags;
431266077Sdes	struct proc *p = cnp->cn_proc;
432266077Sdes	u_long vpid;	/* capability number of vnode */
433266077Sdes
434266077Sdes	*vpp = NULL;
435266077Sdes	dvp = ap->a_dvp;
436266077Sdes	lockparent = flags & LOCKPARENT;
437266077Sdes
438266077Sdes	if (dvp->v_type != VDIR)
439266077Sdes                return (ENOTDIR);
440266077Sdes
441266077Sdes	if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
442266077Sdes	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
443266077Sdes		return (EROFS);
444266077Sdes
445266077Sdes	error = VOP_ACCESS(dvp, VEXEC, cred, p);
446266077Sdes
447266077Sdes	if (error)
448266077Sdes		return (error);
449266077Sdes
450266077Sdes	error = cache_lookup(dvp, vpp, cnp);
451266077Sdes
452266077Sdes	if (!error)
453266077Sdes		return (VOP_CACHEDLOOKUP(dvp, vpp, cnp));
454266077Sdes
455266077Sdes	if (error == ENOENT)
456266077Sdes		return (error);
457266077Sdes
458266077Sdes	vp = *vpp;
459266077Sdes	vpid = vp->v_id;
460266077Sdes	cnp->cn_flags &= ~PDIRUNLOCK;
461266077Sdes	if (dvp == vp) {   /* lookup on "." */
462266077Sdes		VREF(vp);
463266077Sdes		error = 0;
464266077Sdes	} else if (flags & ISDOTDOT) {
465266077Sdes		VOP_UNLOCK(dvp, 0, p);
466266077Sdes		cnp->cn_flags |= PDIRUNLOCK;
467266077Sdes		error = vget(vp, LK_EXCLUSIVE, p);
468266077Sdes		if (!error && lockparent && (flags & ISLASTCN)) {
469266077Sdes			if ((error = vn_lock(dvp, LK_EXCLUSIVE, p)) == 0)
470266077Sdes				cnp->cn_flags &= ~PDIRUNLOCK;
471266077Sdes		}
472266077Sdes	} else {
473266077Sdes		error = vget(vp, LK_EXCLUSIVE, p);
474266077Sdes		if (!lockparent || error || !(flags & ISLASTCN)) {
475266077Sdes			VOP_UNLOCK(dvp, 0, p);
476266077Sdes			cnp->cn_flags |= PDIRUNLOCK;
477266077Sdes		}
478266077Sdes	}
479266077Sdes	/*
480266077Sdes	 * Check that the capability number did not change
481266077Sdes	 * while we were waiting for the lock.
482266077Sdes	 */
483266077Sdes	if (!error) {
484266077Sdes		if (vpid == vp->v_id)
485266077Sdes			return (0);
486266077Sdes		vput(vp);
487266077Sdes		if (lockparent && dvp != vp && (flags & ISLASTCN)) {
488266077Sdes			VOP_UNLOCK(dvp, 0, p);
489266077Sdes			cnp->cn_flags |= PDIRUNLOCK;
490266077Sdes		}
491266077Sdes	}
492266077Sdes	if (cnp->cn_flags & PDIRUNLOCK) {
493266077Sdes		error = vn_lock(dvp, LK_EXCLUSIVE, p);
494266077Sdes		if (error)
495266077Sdes			return (error);
496266077Sdes		cnp->cn_flags &= ~PDIRUNLOCK;
497266077Sdes	}
498266077Sdes	return (VOP_CACHEDLOOKUP(dvp, vpp, cnp));
499266077Sdes}
500266077Sdes
501266077Sdes
502266077Sdes#ifndef _SYS_SYSPROTO_H_
503266077Sdesstruct  __getcwd_args {
504266077Sdes	u_char	*buf;
505266077Sdes	u_int	buflen;
506266077Sdes};
507266077Sdes#endif
508266077Sdes
509266077Sdesstatic int disablecwd;
510266077SdesSYSCTL_INT(_debug, OID_AUTO, disablecwd, CTLFLAG_RW, &disablecwd, 0, "");
511266077Sdes
512266077Sdesstatic u_long numcwdcalls; STATNODE(CTLFLAG_RD, numcwdcalls, &numcwdcalls);
513266077Sdesstatic u_long numcwdfail1; STATNODE(CTLFLAG_RD, numcwdfail1, &numcwdfail1);
514266077Sdesstatic u_long numcwdfail2; STATNODE(CTLFLAG_RD, numcwdfail2, &numcwdfail2);
515266077Sdesstatic u_long numcwdfail3; STATNODE(CTLFLAG_RD, numcwdfail3, &numcwdfail3);
516266077Sdesstatic u_long numcwdfail4; STATNODE(CTLFLAG_RD, numcwdfail4, &numcwdfail4);
517266077Sdesstatic u_long numcwdfound; STATNODE(CTLFLAG_RD, numcwdfound, &numcwdfound);
518266077Sdesint
519266077Sdes__getcwd(p, uap)
520266077Sdes	struct proc *p;
521266077Sdes	struct __getcwd_args *uap;
522266077Sdes{
523266077Sdes	char *bp, *buf;
524266077Sdes	int error, i, slash_prefixed;
525266077Sdes	struct filedesc *fdp;
526266077Sdes	struct namecache *ncp;
527266077Sdes	struct vnode *vp;
528266077Sdes
529266077Sdes	numcwdcalls++;
530266077Sdes	if (disablecwd)
531266077Sdes		return (ENODEV);
532266077Sdes	if (uap->buflen < 2)
533266077Sdes		return (EINVAL);
534266077Sdes	if (uap->buflen > MAXPATHLEN)
535266077Sdes		uap->buflen = MAXPATHLEN;
536266077Sdes	buf = bp = malloc(uap->buflen, M_TEMP, M_WAITOK);
537266077Sdes	bp += uap->buflen - 1;
538266077Sdes	*bp = '\0';
539266077Sdes	fdp = p->p_fd;
540266077Sdes	slash_prefixed = 0;
541266077Sdes	for (vp = fdp->fd_cdir; vp != fdp->fd_rdir && vp != rootvnode;) {
542266077Sdes		if (vp->v_flag & VROOT) {
543266077Sdes			if (vp->v_mount == NULL)	/* forced unmount */
544266077Sdes				return (EBADF);
545266077Sdes			vp = vp->v_mount->mnt_vnodecovered;
546266077Sdes			continue;
547266077Sdes		}
548266077Sdes		if (vp->v_dd->v_id != vp->v_ddid) {
549266077Sdes			numcwdfail1++;
550266077Sdes			free(buf, M_TEMP);
551266077Sdes			return (ENOTDIR);
552266077Sdes		}
553266077Sdes		ncp = TAILQ_FIRST(&vp->v_cache_dst);
554266077Sdes		if (!ncp) {
555266077Sdes			numcwdfail2++;
556266077Sdes			free(buf, M_TEMP);
557266077Sdes			return (ENOENT);
558266077Sdes		}
559266077Sdes		if (ncp->nc_dvp != vp->v_dd) {
560266077Sdes			numcwdfail3++;
561266077Sdes			free(buf, M_TEMP);
562266077Sdes			return (EBADF);
563266077Sdes		}
564266077Sdes		for (i = ncp->nc_nlen - 1; i >= 0; i--) {
565266077Sdes			if (bp == buf) {
566266077Sdes				numcwdfail4++;
567266077Sdes				free(buf, M_TEMP);
568266077Sdes				return (ENOMEM);
569266077Sdes			}
570266077Sdes			*--bp = ncp->nc_name[i];
571266077Sdes		}
572266077Sdes		if (bp == buf) {
573266077Sdes			numcwdfail4++;
574266077Sdes			free(buf, M_TEMP);
575266077Sdes			return (ENOMEM);
576266077Sdes		}
577266077Sdes		*--bp = '/';
578266077Sdes		slash_prefixed = 1;
579266077Sdes		vp = vp->v_dd;
580266077Sdes	}
581266077Sdes	if (!slash_prefixed) {
582266077Sdes		if (bp == buf) {
583266077Sdes			numcwdfail4++;
584266077Sdes			free(buf, M_TEMP);
585266077Sdes			return (ENOMEM);
586266077Sdes		}
587266077Sdes		*--bp = '/';
588266077Sdes	}
589266077Sdes	numcwdfound++;
590266077Sdes	error = copyout(bp, uap->buf, strlen(bp) + 1);
591266077Sdes	free(buf, M_TEMP);
592266077Sdes	return (error);
593266077Sdes}
594266077Sdes
595266077Sdes/*
596266077Sdes * Thus begins the fullpath magic.
597266077Sdes */
598266077Sdes
599266077Sdes#undef STATNODE
600266077Sdes#define STATNODE(name)							\
601266077Sdes	static u_int name;						\
602266077Sdes	SYSCTL_UINT(_vfs_cache, OID_AUTO, name, CTLFLAG_RD, &name, 0, "")
603266077Sdes
604266077Sdesstatic int disablefullpath;
605266077SdesSYSCTL_INT(_debug, OID_AUTO, disablefullpath, CTLFLAG_RW,
606266077Sdes    &disablefullpath, 0, "");
607266077Sdes
608266077SdesSTATNODE(numfullpathcalls);
609266077SdesSTATNODE(numfullpathfail1);
610266077SdesSTATNODE(numfullpathfail2);
611266077SdesSTATNODE(numfullpathfail3);
612266077SdesSTATNODE(numfullpathfail4);
613266077SdesSTATNODE(numfullpathfound);
614266077Sdes
615266077Sdesint
616266077Sdestextvp_fullpath(struct proc *p, char **retbuf, char **retfreebuf) {
617266077Sdes	char *bp, *buf;
618266077Sdes	int i, slash_prefixed;
619266077Sdes	struct filedesc *fdp;
620266077Sdes	struct namecache *ncp;
621266077Sdes	struct vnode *vp, *textvp;
622266077Sdes
623266077Sdes	numfullpathcalls++;
624266077Sdes	if (disablefullpath)
625266077Sdes		return (ENODEV);
626266077Sdes	textvp = p->p_textvp;
627266077Sdes	if (textvp == NULL)
628266077Sdes		return (EINVAL);
629266077Sdes	buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
630266077Sdes	bp = buf + MAXPATHLEN - 1;
631266077Sdes	*bp = '\0';
632266077Sdes	fdp = p->p_fd;
633266077Sdes	slash_prefixed = 0;
634266077Sdes	for (vp = textvp; vp != fdp->fd_rdir && vp != rootvnode;) {
635266077Sdes		if (vp->v_flag & VROOT) {
636266077Sdes			if (vp->v_mount == NULL) {	/* forced unmount */
637266077Sdes				free(buf, M_TEMP);
638266077Sdes				return (EBADF);
639266077Sdes			}
640266077Sdes			vp = vp->v_mount->mnt_vnodecovered;
641266077Sdes			continue;
642266077Sdes		}
643266077Sdes		if (vp != textvp && vp->v_dd->v_id != vp->v_ddid) {
644266077Sdes			numfullpathfail1++;
645266077Sdes			free(buf, M_TEMP);
646266077Sdes			return (ENOTDIR);
647266077Sdes		}
648266077Sdes		ncp = TAILQ_FIRST(&vp->v_cache_dst);
649266077Sdes		if (!ncp) {
650266077Sdes			numfullpathfail2++;
651266077Sdes			free(buf, M_TEMP);
652266077Sdes			return (ENOENT);
653266077Sdes		}
654266077Sdes		if (vp != textvp && ncp->nc_dvp != vp->v_dd) {
655266077Sdes			numfullpathfail3++;
656266077Sdes			free(buf, M_TEMP);
657266077Sdes			return (EBADF);
658266077Sdes		}
659266077Sdes		for (i = ncp->nc_nlen - 1; i >= 0; i--) {
660266077Sdes			if (bp == buf) {
661266077Sdes				numfullpathfail4++;
662266077Sdes				free(buf, M_TEMP);
663266077Sdes				return (ENOMEM);
664266077Sdes			}
665266077Sdes			*--bp = ncp->nc_name[i];
666266077Sdes		}
667266077Sdes		if (bp == buf) {
668266077Sdes			numfullpathfail4++;
669266077Sdes			free(buf, M_TEMP);
670266077Sdes			return (ENOMEM);
671266077Sdes		}
672266077Sdes		*--bp = '/';
673266077Sdes		slash_prefixed = 1;
674266077Sdes		vp = ncp->nc_dvp;
675266077Sdes	}
676266077Sdes	if (!slash_prefixed) {
677266077Sdes		if (bp == buf) {
678266077Sdes			numfullpathfail4++;
679266077Sdes			free(buf, M_TEMP);
680266077Sdes			return (ENOMEM);
681266077Sdes		}
682266077Sdes		*--bp = '/';
683266077Sdes	}
684266077Sdes	numfullpathfound++;
685266077Sdes	*retbuf = bp;
686266077Sdes	*retfreebuf = buf;
687266077Sdes	return (0);
688266077Sdes}
689266077Sdes