xref: /netbsd-src/sys/kern/vfs_cache.c (revision 200d779b75dbeafa7bc01fd0f60bc61185f6967b)
1 /*	$NetBSD: vfs_cache.c,v 1.107 2015/08/24 22:50:32 pooka Exp $	*/
2 
3 /*-
4  * Copyright (c) 2008 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
17  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
18  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
19  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
20  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
21  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
22  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
23  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
24  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
25  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
26  * POSSIBILITY OF SUCH DAMAGE.
27  */
28 
29 /*
30  * Copyright (c) 1989, 1993
31  *	The Regents of the University of California.  All rights reserved.
32  *
33  * Redistribution and use in source and binary forms, with or without
34  * modification, are permitted provided that the following conditions
35  * are met:
36  * 1. Redistributions of source code must retain the above copyright
37  *    notice, this list of conditions and the following disclaimer.
38  * 2. Redistributions in binary form must reproduce the above copyright
39  *    notice, this list of conditions and the following disclaimer in the
40  *    documentation and/or other materials provided with the distribution.
41  * 3. Neither the name of the University nor the names of its contributors
42  *    may be used to endorse or promote products derived from this software
43  *    without specific prior written permission.
44  *
45  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
46  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
47  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
48  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
49  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
50  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
51  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
52  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
53  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
54  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
55  * SUCH DAMAGE.
56  *
57  *	@(#)vfs_cache.c	8.3 (Berkeley) 8/22/94
58  */
59 
60 #include <sys/cdefs.h>
61 __KERNEL_RCSID(0, "$NetBSD: vfs_cache.c,v 1.107 2015/08/24 22:50:32 pooka Exp $");
62 
63 #ifdef _KERNEL_OPT
64 #include "opt_ddb.h"
65 #include "opt_revcache.h"
66 #endif
67 
68 #include <sys/param.h>
69 #include <sys/systm.h>
70 #include <sys/sysctl.h>
71 #include <sys/time.h>
72 #include <sys/mount.h>
73 #include <sys/vnode.h>
74 #include <sys/namei.h>
75 #include <sys/errno.h>
76 #include <sys/pool.h>
77 #include <sys/mutex.h>
78 #include <sys/atomic.h>
79 #include <sys/kthread.h>
80 #include <sys/kernel.h>
81 #include <sys/cpu.h>
82 #include <sys/evcnt.h>
83 
84 #define NAMECACHE_ENTER_REVERSE
85 /*
86  * Name caching works as follows:
87  *
88  * Names found by directory scans are retained in a cache
89  * for future reference.  It is managed LRU, so frequently
90  * used names will hang around.  Cache is indexed by hash value
91  * obtained from (dvp, name) where dvp refers to the directory
92  * containing name.
93  *
94  * For simplicity (and economy of storage), names longer than
95  * a maximum length of NCHNAMLEN are not cached; they occur
96  * infrequently in any case, and are almost never of interest.
97  *
98  * Upon reaching the last segment of a path, if the reference
99  * is for DELETE, or NOCACHE is set (rewrite), and the
100  * name is located in the cache, it will be dropped.
101  * The entry is dropped also when it was not possible to lock
102  * the cached vnode, either because vget() failed or the generation
103  * number has changed while waiting for the lock.
104  */
105 
106 /*
107  * The locking in this subsystem works as follows:
108  *
109  * When an entry is added to the cache, via cache_enter(),
110  * namecache_lock is taken to exclude other writers.  The new
111  * entry is added to the hash list in a way which permits
112  * concurrent lookups and invalidations in the cache done on
113  * other CPUs to continue in parallel.
114  *
115  * When a lookup is done in the cache, via cache_lookup() or
116  * cache_lookup_raw(), the per-cpu lock below is taken.  This
117  * protects calls to cache_lookup_entry() and cache_invalidate()
118  * against cache_reclaim() but allows lookups to continue in
119  * parallel with cache_enter().
120  *
121  * cache_revlookup() takes namecache_lock to exclude cache_enter()
122  * and cache_reclaim() since the list it operates on is not
123  * maintained to allow concurrent reads.
124  *
125  * When cache_reclaim() is called namecache_lock is held to hold
126  * off calls to cache_enter()/cache_revlookup() and each of the
127  * per-cpu locks is taken to hold off lookups.  Holding all these
128  * locks essentially idles the subsystem, ensuring there are no
129  * concurrent references to the cache entries being freed.
130  *
131  * 32 bit per-cpu statistic counters (struct nchstats_percpu) are
132  * incremented when the operations they count are performed while
133  * running on the corresponding CPU.  Frequently individual counters
134  * are incremented while holding a lock (either a per-cpu lock or
135  * namecache_lock) sufficient to preclude concurrent increments
136  * being done to the same counter, so non-atomic increments are
137  * done using the COUNT() macro.  Counters which are incremented
138  * when one of these locks is not held use the COUNT_UNL() macro
139  * instead.  COUNT_UNL() could be defined to do atomic increments
140  * but currently just does what COUNT() does, on the theory that
141  * it is unlikely the non-atomic increment will be interrupted
142  * by something on the same CPU that increments the same counter,
143  * but even if it does happen the consequences aren't serious.
144  *
145  * N.B.: Attempting to protect COUNT_UNL() increments by taking
146  * a per-cpu lock in the namecache_count_*() functions causes
147  * a deadlock.  Don't do that, use atomic increments instead if
148  * the imperfections here bug you.
149  *
150  * The 64 bit system-wide statistic counts (struct nchstats) are
151  * maintained by sampling the per-cpu counters periodically, adding
152  * in the deltas since the last samples and recording the current
153  * samples to use to compute the next delta.  The sampling is done
154  * as a side effect of cache_reclaim() which is run periodically,
155  * for its own purposes, often enough to avoid overflow of the 32
156  * bit counters.  While sampling in this fashion requires no locking
157  * it is never-the-less done only after all locks have been taken by
158  * cache_reclaim() to allow cache_stat_sysctl() to hold off
159  * cache_reclaim() with minimal locking.
160  *
161  * cache_stat_sysctl() takes its CPU's per-cpu lock to hold off
162  * cache_reclaim() so that it can copy the subsystem total stats
163  * without them being concurrently modified.  If CACHE_STATS_CURRENT
164  * is defined it also harvests the per-cpu increments into the total,
165  * which again requires cache_reclaim() to be held off.
166  *
167  * The per-cpu data (a lock and the per-cpu stats structures)
168  * are defined next.
169  */
170 struct nchstats_percpu _NAMEI_CACHE_STATS(uint32_t);
171 
172 struct nchcpu {
173 	kmutex_t		cpu_lock;
174 	struct nchstats_percpu	cpu_stats;
175 	/* XXX maybe __cacheline_aligned would improve this? */
176 	struct nchstats_percpu	cpu_stats_last;	/* from last sample */
177 };
178 
179 /*
180  * The type for the hash code. While the hash function generates a
181  * u32, the hash code has historically been passed around as a u_long,
182  * and the value is modified by xor'ing a uintptr_t, so it's not
183  * entirely clear what the best type is. For now I'll leave it
184  * unchanged as u_long.
185  */
186 
187 typedef u_long nchash_t;
188 
189 /*
190  * Structures associated with name cacheing.
191  */
192 
193 static kmutex_t *namecache_lock __read_mostly;
194 static pool_cache_t namecache_cache __read_mostly;
195 static TAILQ_HEAD(, namecache) nclruhead __cacheline_aligned;
196 
197 static LIST_HEAD(nchashhead, namecache) *nchashtbl __read_mostly;
198 static u_long	nchash __read_mostly;
199 
200 #define	NCHASH2(hash, dvp)	\
201 	(((hash) ^ ((uintptr_t)(dvp) >> 3)) & nchash)
202 
203 static LIST_HEAD(ncvhashhead, namecache) *ncvhashtbl __read_mostly;
204 static u_long	ncvhash __read_mostly;
205 
206 #define	NCVHASH(vp)		(((uintptr_t)(vp) >> 3) & ncvhash)
207 
208 /* Number of cache entries allocated. */
209 static long	numcache __cacheline_aligned;
210 
211 /* Garbage collection queue and number of entries pending in it. */
212 static void	*cache_gcqueue;
213 static u_int	cache_gcpend;
214 
215 /* Cache effectiveness statistics.  This holds total from per-cpu stats */
216 struct nchstats	nchstats __cacheline_aligned;
217 
218 /*
219  * Macros to count an event, update the central stats with per-cpu
220  * values and add current per-cpu increments to the subsystem total
221  * last collected by cache_reclaim().
222  */
223 #define	CACHE_STATS_CURRENT	/* nothing */
224 
225 #define	COUNT(cpup, f)	((cpup)->cpu_stats.f++)
226 
227 #define	UPDATE(cpup, f) do { \
228 	struct nchcpu *Xcpup = (cpup); \
229 	uint32_t Xcnt = (volatile uint32_t) Xcpup->cpu_stats.f; \
230 	nchstats.f += Xcnt - Xcpup->cpu_stats_last.f; \
231 	Xcpup->cpu_stats_last.f = Xcnt; \
232 } while (/* CONSTCOND */ 0)
233 
234 #define	ADD(stats, cpup, f) do { \
235 	struct nchcpu *Xcpup = (cpup); \
236 	stats.f += Xcpup->cpu_stats.f - Xcpup->cpu_stats_last.f; \
237 } while (/* CONSTCOND */ 0)
238 
239 /* Do unlocked stats the same way. Use a different name to allow mind changes */
240 #define	COUNT_UNL(cpup, f)	COUNT((cpup), f)
241 
242 static const int cache_lowat = 95;
243 static const int cache_hiwat = 98;
244 static const int cache_hottime = 5;	/* number of seconds */
245 static int doingcache = 1;		/* 1 => enable the cache */
246 
247 static struct evcnt cache_ev_scan;
248 static struct evcnt cache_ev_gc;
249 static struct evcnt cache_ev_over;
250 static struct evcnt cache_ev_under;
251 static struct evcnt cache_ev_forced;
252 
253 static void cache_invalidate(struct namecache *);
254 static struct namecache *cache_lookup_entry(
255     const struct vnode *, const char *, size_t);
256 static void cache_thread(void *);
257 static void cache_invalidate(struct namecache *);
258 static void cache_disassociate(struct namecache *);
259 static void cache_reclaim(void);
260 static int cache_ctor(void *, void *, int);
261 static void cache_dtor(void *, void *);
262 
263 static struct sysctllog *sysctllog;
264 static void sysctl_cache_stat_setup(void);
265 
266 /*
267  * Compute the hash for an entry.
268  *
269  * (This is for now a wrapper around namei_hash, whose interface is
270  * for the time being slightly inconvenient.)
271  */
272 static nchash_t
273 cache_hash(const char *name, size_t namelen)
274 {
275 	const char *endptr;
276 
277 	endptr = name + namelen;
278 	return namei_hash(name, &endptr);
279 }
280 
281 /*
282  * Invalidate a cache entry and enqueue it for garbage collection.
283  * The caller needs to hold namecache_lock or a per-cpu lock to hold
284  * off cache_reclaim().
285  */
286 static void
287 cache_invalidate(struct namecache *ncp)
288 {
289 	void *head;
290 
291 	KASSERT(mutex_owned(&ncp->nc_lock));
292 
293 	if (ncp->nc_dvp != NULL) {
294 		ncp->nc_vp = NULL;
295 		ncp->nc_dvp = NULL;
296 		do {
297 			head = cache_gcqueue;
298 			ncp->nc_gcqueue = head;
299 		} while (atomic_cas_ptr(&cache_gcqueue, head, ncp) != head);
300 		atomic_inc_uint(&cache_gcpend);
301 	}
302 }
303 
304 /*
305  * Disassociate a namecache entry from any vnodes it is attached to,
306  * and remove from the global LRU list.
307  */
308 static void
309 cache_disassociate(struct namecache *ncp)
310 {
311 
312 	KASSERT(mutex_owned(namecache_lock));
313 	KASSERT(ncp->nc_dvp == NULL);
314 
315 	if (ncp->nc_lru.tqe_prev != NULL) {
316 		TAILQ_REMOVE(&nclruhead, ncp, nc_lru);
317 		ncp->nc_lru.tqe_prev = NULL;
318 	}
319 	if (ncp->nc_vhash.le_prev != NULL) {
320 		LIST_REMOVE(ncp, nc_vhash);
321 		ncp->nc_vhash.le_prev = NULL;
322 	}
323 	if (ncp->nc_vlist.le_prev != NULL) {
324 		LIST_REMOVE(ncp, nc_vlist);
325 		ncp->nc_vlist.le_prev = NULL;
326 	}
327 	if (ncp->nc_dvlist.le_prev != NULL) {
328 		LIST_REMOVE(ncp, nc_dvlist);
329 		ncp->nc_dvlist.le_prev = NULL;
330 	}
331 }
332 
333 /*
334  * Lock all CPUs to prevent any cache lookup activity.  Conceptually,
335  * this locks out all "readers".
336  */
337 static void
338 cache_lock_cpus(void)
339 {
340 	CPU_INFO_ITERATOR cii;
341 	struct cpu_info *ci;
342 	struct nchcpu *cpup;
343 
344 	/*
345 	 * Lock out all CPUs first, then harvest per-cpu stats.  This
346 	 * is probably not quite as cache-efficient as doing the lock
347 	 * and harvest at the same time, but allows cache_stat_sysctl()
348 	 * to make do with a per-cpu lock.
349 	 */
350 	for (CPU_INFO_FOREACH(cii, ci)) {
351 		cpup = ci->ci_data.cpu_nch;
352 		mutex_enter(&cpup->cpu_lock);
353 	}
354 	for (CPU_INFO_FOREACH(cii, ci)) {
355 		cpup = ci->ci_data.cpu_nch;
356 		UPDATE(cpup, ncs_goodhits);
357 		UPDATE(cpup, ncs_neghits);
358 		UPDATE(cpup, ncs_badhits);
359 		UPDATE(cpup, ncs_falsehits);
360 		UPDATE(cpup, ncs_miss);
361 		UPDATE(cpup, ncs_long);
362 		UPDATE(cpup, ncs_pass2);
363 		UPDATE(cpup, ncs_2passes);
364 		UPDATE(cpup, ncs_revhits);
365 		UPDATE(cpup, ncs_revmiss);
366 	}
367 }
368 
369 /*
370  * Release all CPU locks.
371  */
372 static void
373 cache_unlock_cpus(void)
374 {
375 	CPU_INFO_ITERATOR cii;
376 	struct cpu_info *ci;
377 	struct nchcpu *cpup;
378 
379 	for (CPU_INFO_FOREACH(cii, ci)) {
380 		cpup = ci->ci_data.cpu_nch;
381 		mutex_exit(&cpup->cpu_lock);
382 	}
383 }
384 
385 /*
386  * Find a single cache entry and return it locked.
387  * The caller needs to hold namecache_lock or a per-cpu lock to hold
388  * off cache_reclaim().
389  */
390 static struct namecache *
391 cache_lookup_entry(const struct vnode *dvp, const char *name, size_t namelen)
392 {
393 	struct nchashhead *ncpp;
394 	struct namecache *ncp;
395 	nchash_t hash;
396 
397 	KASSERT(dvp != NULL);
398 	hash = cache_hash(name, namelen);
399 	ncpp = &nchashtbl[NCHASH2(hash, dvp)];
400 
401 	LIST_FOREACH(ncp, ncpp, nc_hash) {
402 		membar_datadep_consumer();	/* for Alpha... */
403 		if (ncp->nc_dvp != dvp ||
404 		    ncp->nc_nlen != namelen ||
405 		    memcmp(ncp->nc_name, name, (u_int)ncp->nc_nlen))
406 		    	continue;
407 	    	mutex_enter(&ncp->nc_lock);
408 		if (__predict_true(ncp->nc_dvp == dvp)) {
409 			ncp->nc_hittime = hardclock_ticks;
410 			return ncp;
411 		}
412 		/* Raced: entry has been nullified. */
413 		mutex_exit(&ncp->nc_lock);
414 	}
415 
416 	return NULL;
417 }
418 
419 /*
420  * Look for a the name in the cache. We don't do this
421  * if the segment name is long, simply so the cache can avoid
422  * holding long names (which would either waste space, or
423  * add greatly to the complexity).
424  *
425  * Lookup is called with DVP pointing to the directory to search,
426  * and CNP providing the name of the entry being sought: cn_nameptr
427  * is the name, cn_namelen is its length, and cn_flags is the flags
428  * word from the namei operation.
429  *
430  * DVP must be locked.
431  *
432  * There are three possible non-error return states:
433  *    1. Nothing was found in the cache. Nothing is known about
434  *       the requested name.
435  *    2. A negative entry was found in the cache, meaning that the
436  *       requested name definitely does not exist.
437  *    3. A positive entry was found in the cache, meaning that the
438  *       requested name does exist and that we are providing the
439  *       vnode.
440  * In these cases the results are:
441  *    1. 0 returned; VN is set to NULL.
442  *    2. 1 returned; VN is set to NULL.
443  *    3. 1 returned; VN is set to the vnode found.
444  *
445  * The additional result argument ISWHT is set to zero, unless a
446  * negative entry is found that was entered as a whiteout, in which
447  * case ISWHT is set to one.
448  *
449  * The ISWHT_RET argument pointer may be null. In this case an
450  * assertion is made that the whiteout flag is not set. File systems
451  * that do not support whiteouts can/should do this.
452  *
453  * Filesystems that do support whiteouts should add ISWHITEOUT to
454  * cnp->cn_flags if ISWHT comes back nonzero.
455  *
456  * When a vnode is returned, it is locked, as per the vnode lookup
457  * locking protocol.
458  *
459  * There is no way for this function to fail, in the sense of
460  * generating an error that requires aborting the namei operation.
461  *
462  * (Prior to October 2012, this function returned an integer status,
463  * and a vnode, and mucked with the flags word in CNP for whiteouts.
464  * The integer status was -1 for "nothing found", ENOENT for "a
465  * negative entry found", 0 for "a positive entry found", and possibly
466  * other errors, and the value of VN might or might not have been set
467  * depending on what error occurred.)
468  */
469 int
470 cache_lookup(struct vnode *dvp, const char *name, size_t namelen,
471 	     uint32_t nameiop, uint32_t cnflags,
472 	     int *iswht_ret, struct vnode **vn_ret)
473 {
474 	struct namecache *ncp;
475 	struct vnode *vp;
476 	struct nchcpu *cpup;
477 	int error, ret_value;
478 
479 
480 	/* Establish default result values */
481 	if (iswht_ret != NULL) {
482 		*iswht_ret = 0;
483 	}
484 	*vn_ret = NULL;
485 
486 	if (__predict_false(!doingcache)) {
487 		return 0;
488 	}
489 
490 	cpup = curcpu()->ci_data.cpu_nch;
491 	mutex_enter(&cpup->cpu_lock);
492 	if (__predict_false(namelen > NCHNAMLEN)) {
493 		COUNT(cpup, ncs_long);
494 		mutex_exit(&cpup->cpu_lock);
495 		/* found nothing */
496 		return 0;
497 	}
498 
499 	ncp = cache_lookup_entry(dvp, name, namelen);
500 	if (__predict_false(ncp == NULL)) {
501 		COUNT(cpup, ncs_miss);
502 		mutex_exit(&cpup->cpu_lock);
503 		/* found nothing */
504 		return 0;
505 	}
506 	if ((cnflags & MAKEENTRY) == 0) {
507 		COUNT(cpup, ncs_badhits);
508 		/*
509 		 * Last component and we are renaming or deleting,
510 		 * the cache entry is invalid, or otherwise don't
511 		 * want cache entry to exist.
512 		 */
513 		cache_invalidate(ncp);
514 		mutex_exit(&ncp->nc_lock);
515 		mutex_exit(&cpup->cpu_lock);
516 		/* found nothing */
517 		return 0;
518 	}
519 	if (ncp->nc_vp == NULL) {
520 		if (iswht_ret != NULL) {
521 			/*
522 			 * Restore the ISWHITEOUT flag saved earlier.
523 			 */
524 			KASSERT((ncp->nc_flags & ~ISWHITEOUT) == 0);
525 			*iswht_ret = (ncp->nc_flags & ISWHITEOUT) != 0;
526 		} else {
527 			KASSERT(ncp->nc_flags == 0);
528 		}
529 
530 		if (__predict_true(nameiop != CREATE ||
531 		    (cnflags & ISLASTCN) == 0)) {
532 			COUNT(cpup, ncs_neghits);
533 			/* found neg entry; vn is already null from above */
534 			ret_value = 1;
535 		} else {
536 			COUNT(cpup, ncs_badhits);
537 			/*
538 			 * Last component and we are renaming or
539 			 * deleting, the cache entry is invalid,
540 			 * or otherwise don't want cache entry to
541 			 * exist.
542 			 */
543 			cache_invalidate(ncp);
544 			/* found nothing */
545 			ret_value = 0;
546 		}
547 		mutex_exit(&ncp->nc_lock);
548 		mutex_exit(&cpup->cpu_lock);
549 		return ret_value;
550 	}
551 
552 	vp = ncp->nc_vp;
553 	mutex_enter(vp->v_interlock);
554 	mutex_exit(&ncp->nc_lock);
555 	mutex_exit(&cpup->cpu_lock);
556 
557 	/*
558 	 * Unlocked except for the vnode interlock.  Call vget().
559 	 */
560 	error = vget(vp, LK_NOWAIT, false /* !wait */);
561 	if (error) {
562 		KASSERT(error == EBUSY);
563 		/*
564 		 * This vnode is being cleaned out.
565 		 * XXX badhits?
566 		 */
567 		COUNT_UNL(cpup, ncs_falsehits);
568 		/* found nothing */
569 		return 0;
570 	}
571 
572 	COUNT_UNL(cpup, ncs_goodhits);
573 	/* found it */
574 	*vn_ret = vp;
575 	return 1;
576 }
577 
578 
579 /*
580  * Cut-'n-pasted version of the above without the nameiop argument.
581  */
582 int
583 cache_lookup_raw(struct vnode *dvp, const char *name, size_t namelen,
584 		 uint32_t cnflags,
585 		 int *iswht_ret, struct vnode **vn_ret)
586 {
587 	struct namecache *ncp;
588 	struct vnode *vp;
589 	struct nchcpu *cpup;
590 	int error;
591 
592 	/* Establish default results. */
593 	if (iswht_ret != NULL) {
594 		*iswht_ret = 0;
595 	}
596 	*vn_ret = NULL;
597 
598 	if (__predict_false(!doingcache)) {
599 		/* found nothing */
600 		return 0;
601 	}
602 
603 	cpup = curcpu()->ci_data.cpu_nch;
604 	mutex_enter(&cpup->cpu_lock);
605 	if (__predict_false(namelen > NCHNAMLEN)) {
606 		COUNT(cpup, ncs_long);
607 		mutex_exit(&cpup->cpu_lock);
608 		/* found nothing */
609 		return 0;
610 	}
611 	ncp = cache_lookup_entry(dvp, name, namelen);
612 	if (__predict_false(ncp == NULL)) {
613 		COUNT(cpup, ncs_miss);
614 		mutex_exit(&cpup->cpu_lock);
615 		/* found nothing */
616 		return 0;
617 	}
618 	vp = ncp->nc_vp;
619 	if (vp == NULL) {
620 		/*
621 		 * Restore the ISWHITEOUT flag saved earlier.
622 		 */
623 		if (iswht_ret != NULL) {
624 			KASSERT((ncp->nc_flags & ~ISWHITEOUT) == 0);
625 			/*cnp->cn_flags |= ncp->nc_flags;*/
626 			*iswht_ret = (ncp->nc_flags & ISWHITEOUT) != 0;
627 		}
628 		COUNT(cpup, ncs_neghits);
629 		mutex_exit(&ncp->nc_lock);
630 		mutex_exit(&cpup->cpu_lock);
631 		/* found negative entry; vn is already null from above */
632 		return 1;
633 	}
634 	mutex_enter(vp->v_interlock);
635 	mutex_exit(&ncp->nc_lock);
636 	mutex_exit(&cpup->cpu_lock);
637 
638 	/*
639 	 * Unlocked except for the vnode interlock.  Call vget().
640 	 */
641 	error = vget(vp, LK_NOWAIT, false /* !wait */);
642 	if (error) {
643 		KASSERT(error == EBUSY);
644 		/*
645 		 * This vnode is being cleaned out.
646 		 * XXX badhits?
647 		 */
648 		COUNT_UNL(cpup, ncs_falsehits);
649 		/* found nothing */
650 		return 0;
651 	}
652 
653 	COUNT_UNL(cpup, ncs_goodhits); /* XXX can be "badhits" */
654 	/* found it */
655 	*vn_ret = vp;
656 	return 1;
657 }
658 
659 /*
660  * Scan cache looking for name of directory entry pointing at vp.
661  *
662  * If the lookup succeeds the vnode is referenced and stored in dvpp.
663  *
664  * If bufp is non-NULL, also place the name in the buffer which starts
665  * at bufp, immediately before *bpp, and move bpp backwards to point
666  * at the start of it.  (Yes, this is a little baroque, but it's done
667  * this way to cater to the whims of getcwd).
668  *
669  * Returns 0 on success, -1 on cache miss, positive errno on failure.
670  */
671 int
672 cache_revlookup(struct vnode *vp, struct vnode **dvpp, char **bpp, char *bufp)
673 {
674 	struct namecache *ncp;
675 	struct vnode *dvp;
676 	struct ncvhashhead *nvcpp;
677 	struct nchcpu *cpup;
678 	char *bp;
679 	int error, nlen;
680 
681 	if (!doingcache)
682 		goto out;
683 
684 	nvcpp = &ncvhashtbl[NCVHASH(vp)];
685 
686 	/*
687 	 * We increment counters in the local CPU's per-cpu stats.
688 	 * We don't take the per-cpu lock, however, since this function
689 	 * is the only place these counters are incremented so no one
690 	 * will be racing with us to increment them.
691 	 */
692 	cpup = curcpu()->ci_data.cpu_nch;
693 	mutex_enter(namecache_lock);
694 	LIST_FOREACH(ncp, nvcpp, nc_vhash) {
695 		mutex_enter(&ncp->nc_lock);
696 		if (ncp->nc_vp == vp &&
697 		    (dvp = ncp->nc_dvp) != NULL &&
698 		    dvp != vp) { 		/* avoid pesky . entries.. */
699 
700 #ifdef DIAGNOSTIC
701 			if (ncp->nc_nlen == 1 &&
702 			    ncp->nc_name[0] == '.')
703 				panic("cache_revlookup: found entry for .");
704 
705 			if (ncp->nc_nlen == 2 &&
706 			    ncp->nc_name[0] == '.' &&
707 			    ncp->nc_name[1] == '.')
708 				panic("cache_revlookup: found entry for ..");
709 #endif
710 			COUNT(cpup, ncs_revhits);
711 			nlen = ncp->nc_nlen;
712 
713 			if (bufp) {
714 				bp = *bpp;
715 				bp -= nlen;
716 				if (bp <= bufp) {
717 					*dvpp = NULL;
718 					mutex_exit(&ncp->nc_lock);
719 					mutex_exit(namecache_lock);
720 					return (ERANGE);
721 				}
722 				memcpy(bp, ncp->nc_name, nlen);
723 				*bpp = bp;
724 			}
725 
726 			mutex_enter(dvp->v_interlock);
727 			mutex_exit(&ncp->nc_lock);
728 			mutex_exit(namecache_lock);
729 			error = vget(dvp, LK_NOWAIT, false /* !wait */);
730 			if (error) {
731 				KASSERT(error == EBUSY);
732 				if (bufp)
733 					(*bpp) += nlen;
734 				*dvpp = NULL;
735 				return -1;
736 			}
737 			*dvpp = dvp;
738 			return (0);
739 		}
740 		mutex_exit(&ncp->nc_lock);
741 	}
742 	COUNT(cpup, ncs_revmiss);
743 	mutex_exit(namecache_lock);
744  out:
745 	*dvpp = NULL;
746 	return (-1);
747 }
748 
749 /*
750  * Add an entry to the cache
751  */
752 void
753 cache_enter(struct vnode *dvp, struct vnode *vp,
754 	    const char *name, size_t namelen, uint32_t cnflags)
755 {
756 	struct namecache *ncp;
757 	struct namecache *oncp;
758 	struct nchashhead *ncpp;
759 	struct ncvhashhead *nvcpp;
760 	nchash_t hash;
761 
762 	/* First, check whether we can/should add a cache entry. */
763 	if ((cnflags & MAKEENTRY) == 0 ||
764 	    __predict_false(namelen > NCHNAMLEN || !doingcache)) {
765 		return;
766 	}
767 
768 	if (numcache > desiredvnodes) {
769 		mutex_enter(namecache_lock);
770 		cache_ev_forced.ev_count++;
771 		cache_reclaim();
772 		mutex_exit(namecache_lock);
773 	}
774 
775 	ncp = pool_cache_get(namecache_cache, PR_WAITOK);
776 	mutex_enter(namecache_lock);
777 	numcache++;
778 
779 	/*
780 	 * Concurrent lookups in the same directory may race for a
781 	 * cache entry.  if there's a duplicated entry, free it.
782 	 */
783 	oncp = cache_lookup_entry(dvp, name, namelen);
784 	if (oncp) {
785 		cache_invalidate(oncp);
786 		mutex_exit(&oncp->nc_lock);
787 	}
788 
789 	/* Grab the vnode we just found. */
790 	mutex_enter(&ncp->nc_lock);
791 	ncp->nc_vp = vp;
792 	ncp->nc_flags = 0;
793 	ncp->nc_hittime = 0;
794 	ncp->nc_gcqueue = NULL;
795 	if (vp == NULL) {
796 		/*
797 		 * For negative hits, save the ISWHITEOUT flag so we can
798 		 * restore it later when the cache entry is used again.
799 		 */
800 		ncp->nc_flags = cnflags & ISWHITEOUT;
801 	}
802 
803 	/* Fill in cache info. */
804 	ncp->nc_dvp = dvp;
805 	LIST_INSERT_HEAD(&dvp->v_dnclist, ncp, nc_dvlist);
806 	if (vp)
807 		LIST_INSERT_HEAD(&vp->v_nclist, ncp, nc_vlist);
808 	else {
809 		ncp->nc_vlist.le_prev = NULL;
810 		ncp->nc_vlist.le_next = NULL;
811 	}
812 	KASSERT(namelen <= NCHNAMLEN);
813 	ncp->nc_nlen = namelen;
814 	memcpy(ncp->nc_name, name, (unsigned)ncp->nc_nlen);
815 	TAILQ_INSERT_TAIL(&nclruhead, ncp, nc_lru);
816 	hash = cache_hash(name, namelen);
817 	ncpp = &nchashtbl[NCHASH2(hash, dvp)];
818 
819 	/*
820 	 * Flush updates before making visible in table.  No need for a
821 	 * memory barrier on the other side: to see modifications the
822 	 * list must be followed, meaning a dependent pointer load.
823 	 * The below is LIST_INSERT_HEAD() inlined, with the memory
824 	 * barrier included in the correct place.
825 	 */
826 	if ((ncp->nc_hash.le_next = ncpp->lh_first) != NULL)
827 		ncpp->lh_first->nc_hash.le_prev = &ncp->nc_hash.le_next;
828 	ncp->nc_hash.le_prev = &ncpp->lh_first;
829 	membar_producer();
830 	ncpp->lh_first = ncp;
831 
832 	ncp->nc_vhash.le_prev = NULL;
833 	ncp->nc_vhash.le_next = NULL;
834 
835 	/*
836 	 * Create reverse-cache entries (used in getcwd) for directories.
837 	 * (and in linux procfs exe node)
838 	 */
839 	if (vp != NULL &&
840 	    vp != dvp &&
841 #ifndef NAMECACHE_ENTER_REVERSE
842 	    vp->v_type == VDIR &&
843 #endif
844 	    (ncp->nc_nlen > 2 ||
845 	    (ncp->nc_nlen > 1 && ncp->nc_name[1] != '.') ||
846 	    (/* ncp->nc_nlen > 0 && */ ncp->nc_name[0] != '.'))) {
847 		nvcpp = &ncvhashtbl[NCVHASH(vp)];
848 		LIST_INSERT_HEAD(nvcpp, ncp, nc_vhash);
849 	}
850 	mutex_exit(&ncp->nc_lock);
851 	mutex_exit(namecache_lock);
852 }
853 
854 /*
855  * Name cache initialization, from vfs_init() when we are booting
856  */
857 void
858 nchinit(void)
859 {
860 	int error;
861 
862 	TAILQ_INIT(&nclruhead);
863 	namecache_cache = pool_cache_init(sizeof(struct namecache),
864 	    coherency_unit, 0, 0, "ncache", NULL, IPL_NONE, cache_ctor,
865 	    cache_dtor, NULL);
866 	KASSERT(namecache_cache != NULL);
867 
868 	namecache_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
869 
870 	nchashtbl = hashinit(desiredvnodes, HASH_LIST, true, &nchash);
871 	ncvhashtbl =
872 #ifdef NAMECACHE_ENTER_REVERSE
873 	    hashinit(desiredvnodes, HASH_LIST, true, &ncvhash);
874 #else
875 	    hashinit(desiredvnodes/8, HASH_LIST, true, &ncvhash);
876 #endif
877 
878 	error = kthread_create(PRI_VM, KTHREAD_MPSAFE, NULL, cache_thread,
879 	    NULL, NULL, "cachegc");
880 	if (error != 0)
881 		panic("nchinit %d", error);
882 
883 	evcnt_attach_dynamic(&cache_ev_scan, EVCNT_TYPE_MISC, NULL,
884 	   "namecache", "entries scanned");
885 	evcnt_attach_dynamic(&cache_ev_gc, EVCNT_TYPE_MISC, NULL,
886 	   "namecache", "entries collected");
887 	evcnt_attach_dynamic(&cache_ev_over, EVCNT_TYPE_MISC, NULL,
888 	   "namecache", "over scan target");
889 	evcnt_attach_dynamic(&cache_ev_under, EVCNT_TYPE_MISC, NULL,
890 	   "namecache", "under scan target");
891 	evcnt_attach_dynamic(&cache_ev_forced, EVCNT_TYPE_MISC, NULL,
892 	   "namecache", "forced reclaims");
893 
894 	sysctl_cache_stat_setup();
895 }
896 
897 static int
898 cache_ctor(void *arg, void *obj, int flag)
899 {
900 	struct namecache *ncp;
901 
902 	ncp = obj;
903 	mutex_init(&ncp->nc_lock, MUTEX_DEFAULT, IPL_NONE);
904 
905 	return 0;
906 }
907 
908 static void
909 cache_dtor(void *arg, void *obj)
910 {
911 	struct namecache *ncp;
912 
913 	ncp = obj;
914 	mutex_destroy(&ncp->nc_lock);
915 }
916 
917 /*
918  * Called once for each CPU in the system as attached.
919  */
920 void
921 cache_cpu_init(struct cpu_info *ci)
922 {
923 	struct nchcpu *cpup;
924 	size_t sz;
925 
926 	sz = roundup2(sizeof(*cpup), coherency_unit) + coherency_unit;
927 	cpup = kmem_zalloc(sz, KM_SLEEP);
928 	cpup = (void *)roundup2((uintptr_t)cpup, coherency_unit);
929 	mutex_init(&cpup->cpu_lock, MUTEX_DEFAULT, IPL_NONE);
930 	ci->ci_data.cpu_nch = cpup;
931 }
932 
933 /*
934  * Name cache reinitialization, for when the maximum number of vnodes increases.
935  */
936 void
937 nchreinit(void)
938 {
939 	struct namecache *ncp;
940 	struct nchashhead *oldhash1, *hash1;
941 	struct ncvhashhead *oldhash2, *hash2;
942 	u_long i, oldmask1, oldmask2, mask1, mask2;
943 
944 	hash1 = hashinit(desiredvnodes, HASH_LIST, true, &mask1);
945 	hash2 =
946 #ifdef NAMECACHE_ENTER_REVERSE
947 	    hashinit(desiredvnodes, HASH_LIST, true, &mask2);
948 #else
949 	    hashinit(desiredvnodes/8, HASH_LIST, true, &mask2);
950 #endif
951 	mutex_enter(namecache_lock);
952 	cache_lock_cpus();
953 	oldhash1 = nchashtbl;
954 	oldmask1 = nchash;
955 	nchashtbl = hash1;
956 	nchash = mask1;
957 	oldhash2 = ncvhashtbl;
958 	oldmask2 = ncvhash;
959 	ncvhashtbl = hash2;
960 	ncvhash = mask2;
961 	for (i = 0; i <= oldmask1; i++) {
962 		while ((ncp = LIST_FIRST(&oldhash1[i])) != NULL) {
963 			LIST_REMOVE(ncp, nc_hash);
964 			ncp->nc_hash.le_prev = NULL;
965 		}
966 	}
967 	for (i = 0; i <= oldmask2; i++) {
968 		while ((ncp = LIST_FIRST(&oldhash2[i])) != NULL) {
969 			LIST_REMOVE(ncp, nc_vhash);
970 			ncp->nc_vhash.le_prev = NULL;
971 		}
972 	}
973 	cache_unlock_cpus();
974 	mutex_exit(namecache_lock);
975 	hashdone(oldhash1, HASH_LIST, oldmask1);
976 	hashdone(oldhash2, HASH_LIST, oldmask2);
977 }
978 
979 /*
980  * Cache flush, a particular vnode; called when a vnode is renamed to
981  * hide entries that would now be invalid
982  */
983 void
984 cache_purge1(struct vnode *vp, const char *name, size_t namelen, int flags)
985 {
986 	struct namecache *ncp, *ncnext;
987 
988 	mutex_enter(namecache_lock);
989 	if (flags & PURGE_PARENTS) {
990 		for (ncp = LIST_FIRST(&vp->v_nclist); ncp != NULL;
991 		    ncp = ncnext) {
992 			ncnext = LIST_NEXT(ncp, nc_vlist);
993 			mutex_enter(&ncp->nc_lock);
994 			cache_invalidate(ncp);
995 			mutex_exit(&ncp->nc_lock);
996 			cache_disassociate(ncp);
997 		}
998 	}
999 	if (flags & PURGE_CHILDREN) {
1000 		for (ncp = LIST_FIRST(&vp->v_dnclist); ncp != NULL;
1001 		    ncp = ncnext) {
1002 			ncnext = LIST_NEXT(ncp, nc_dvlist);
1003 			mutex_enter(&ncp->nc_lock);
1004 			cache_invalidate(ncp);
1005 			mutex_exit(&ncp->nc_lock);
1006 			cache_disassociate(ncp);
1007 		}
1008 	}
1009 	if (name != NULL) {
1010 		ncp = cache_lookup_entry(vp, name, namelen);
1011 		if (ncp) {
1012 			cache_invalidate(ncp);
1013 			mutex_exit(&ncp->nc_lock);
1014 			cache_disassociate(ncp);
1015 		}
1016 	}
1017 	mutex_exit(namecache_lock);
1018 }
1019 
1020 /*
1021  * Cache flush, a whole filesystem; called when filesys is umounted to
1022  * remove entries that would now be invalid.
1023  */
1024 void
1025 cache_purgevfs(struct mount *mp)
1026 {
1027 	struct namecache *ncp, *nxtcp;
1028 
1029 	mutex_enter(namecache_lock);
1030 	for (ncp = TAILQ_FIRST(&nclruhead); ncp != NULL; ncp = nxtcp) {
1031 		nxtcp = TAILQ_NEXT(ncp, nc_lru);
1032 		mutex_enter(&ncp->nc_lock);
1033 		if (ncp->nc_dvp != NULL && ncp->nc_dvp->v_mount == mp) {
1034 			/* Free the resources we had. */
1035 			cache_invalidate(ncp);
1036 			cache_disassociate(ncp);
1037 		}
1038 		mutex_exit(&ncp->nc_lock);
1039 	}
1040 	cache_reclaim();
1041 	mutex_exit(namecache_lock);
1042 }
1043 
1044 /*
1045  * Scan global list invalidating entries until we meet a preset target.
1046  * Prefer to invalidate entries that have not scored a hit within
1047  * cache_hottime seconds.  We sort the LRU list only for this routine's
1048  * benefit.
1049  */
1050 static void
1051 cache_prune(int incache, int target)
1052 {
1053 	struct namecache *ncp, *nxtcp, *sentinel;
1054 	int items, recent, tryharder;
1055 
1056 	KASSERT(mutex_owned(namecache_lock));
1057 
1058 	items = 0;
1059 	tryharder = 0;
1060 	recent = hardclock_ticks - hz * cache_hottime;
1061 	sentinel = NULL;
1062 	for (ncp = TAILQ_FIRST(&nclruhead); ncp != NULL; ncp = nxtcp) {
1063 		if (incache <= target)
1064 			break;
1065 		items++;
1066 		nxtcp = TAILQ_NEXT(ncp, nc_lru);
1067 		if (ncp == sentinel) {
1068 			/*
1069 			 * If we looped back on ourself, then ignore
1070 			 * recent entries and purge whatever we find.
1071 			 */
1072 			tryharder = 1;
1073 		}
1074 		if (ncp->nc_dvp == NULL)
1075 			continue;
1076 		if (!tryharder && (ncp->nc_hittime - recent) > 0) {
1077 			if (sentinel == NULL)
1078 				sentinel = ncp;
1079 			TAILQ_REMOVE(&nclruhead, ncp, nc_lru);
1080 			TAILQ_INSERT_TAIL(&nclruhead, ncp, nc_lru);
1081 			continue;
1082 		}
1083 		mutex_enter(&ncp->nc_lock);
1084 		if (ncp->nc_dvp != NULL) {
1085 			cache_invalidate(ncp);
1086 			cache_disassociate(ncp);
1087 			incache--;
1088 		}
1089 		mutex_exit(&ncp->nc_lock);
1090 	}
1091 	cache_ev_scan.ev_count += items;
1092 }
1093 
1094 /*
1095  * Collect dead cache entries from all CPUs and garbage collect.
1096  */
1097 static void
1098 cache_reclaim(void)
1099 {
1100 	struct namecache *ncp, *next;
1101 	int items;
1102 
1103 	KASSERT(mutex_owned(namecache_lock));
1104 
1105 	/*
1106 	 * If the number of extant entries not awaiting garbage collection
1107 	 * exceeds the high water mark, then reclaim stale entries until we
1108 	 * reach our low water mark.
1109 	 */
1110 	items = numcache - cache_gcpend;
1111 	if (items > (uint64_t)desiredvnodes * cache_hiwat / 100) {
1112 		cache_prune(items, (int)((uint64_t)desiredvnodes *
1113 		    cache_lowat / 100));
1114 		cache_ev_over.ev_count++;
1115 	} else
1116 		cache_ev_under.ev_count++;
1117 
1118 	/*
1119 	 * Stop forward lookup activity on all CPUs and garbage collect dead
1120 	 * entries.
1121 	 */
1122 	cache_lock_cpus();
1123 	ncp = cache_gcqueue;
1124 	cache_gcqueue = NULL;
1125 	items = cache_gcpend;
1126 	cache_gcpend = 0;
1127 	while (ncp != NULL) {
1128 		next = ncp->nc_gcqueue;
1129 		cache_disassociate(ncp);
1130 		KASSERT(ncp->nc_dvp == NULL);
1131 		if (ncp->nc_hash.le_prev != NULL) {
1132 			LIST_REMOVE(ncp, nc_hash);
1133 			ncp->nc_hash.le_prev = NULL;
1134 		}
1135 		pool_cache_put(namecache_cache, ncp);
1136 		ncp = next;
1137 	}
1138 	cache_unlock_cpus();
1139 	numcache -= items;
1140 	cache_ev_gc.ev_count += items;
1141 }
1142 
1143 /*
1144  * Cache maintainence thread, awakening once per second to:
1145  *
1146  * => keep number of entries below the high water mark
1147  * => sort pseudo-LRU list
1148  * => garbage collect dead entries
1149  */
1150 static void
1151 cache_thread(void *arg)
1152 {
1153 
1154 	mutex_enter(namecache_lock);
1155 	for (;;) {
1156 		cache_reclaim();
1157 		kpause("cachegc", false, hz, namecache_lock);
1158 	}
1159 }
1160 
1161 #ifdef DDB
1162 void
1163 namecache_print(struct vnode *vp, void (*pr)(const char *, ...))
1164 {
1165 	struct vnode *dvp = NULL;
1166 	struct namecache *ncp;
1167 
1168 	TAILQ_FOREACH(ncp, &nclruhead, nc_lru) {
1169 		if (ncp->nc_vp == vp && ncp->nc_dvp != NULL) {
1170 			(*pr)("name %.*s\n", ncp->nc_nlen, ncp->nc_name);
1171 			dvp = ncp->nc_dvp;
1172 		}
1173 	}
1174 	if (dvp == NULL) {
1175 		(*pr)("name not found\n");
1176 		return;
1177 	}
1178 	vp = dvp;
1179 	TAILQ_FOREACH(ncp, &nclruhead, nc_lru) {
1180 		if (ncp->nc_vp == vp) {
1181 			(*pr)("parent %.*s\n", ncp->nc_nlen, ncp->nc_name);
1182 		}
1183 	}
1184 }
1185 #endif
1186 
1187 void
1188 namecache_count_pass2(void)
1189 {
1190 	struct nchcpu *cpup = curcpu()->ci_data.cpu_nch;
1191 
1192 	COUNT_UNL(cpup, ncs_pass2);
1193 }
1194 
1195 void
1196 namecache_count_2passes(void)
1197 {
1198 	struct nchcpu *cpup = curcpu()->ci_data.cpu_nch;
1199 
1200 	COUNT_UNL(cpup, ncs_2passes);
1201 }
1202 
1203 /*
1204  * Fetch the current values of the stats.  We return the most
1205  * recent values harvested into nchstats by cache_reclaim(), which
1206  * will be less than a second old.
1207  */
1208 static int
1209 cache_stat_sysctl(SYSCTLFN_ARGS)
1210 {
1211 	struct nchstats stats;
1212 	struct nchcpu *my_cpup;
1213 #ifdef CACHE_STATS_CURRENT
1214 	CPU_INFO_ITERATOR cii;
1215 	struct cpu_info *ci;
1216 #endif	/* CACHE_STATS_CURRENT */
1217 
1218 	if (oldp == NULL) {
1219 		*oldlenp = sizeof(stats);
1220 		return 0;
1221 	}
1222 
1223 	if (*oldlenp < sizeof(stats)) {
1224 		*oldlenp = 0;
1225 		return 0;
1226 	}
1227 
1228 	/*
1229 	 * Take this CPU's per-cpu lock to hold off cache_reclaim()
1230 	 * from doing a stats update while doing minimal damage to
1231 	 * concurrent operations.
1232 	 */
1233 	sysctl_unlock();
1234 	my_cpup = curcpu()->ci_data.cpu_nch;
1235 	mutex_enter(&my_cpup->cpu_lock);
1236 	stats = nchstats;
1237 #ifdef CACHE_STATS_CURRENT
1238 	for (CPU_INFO_FOREACH(cii, ci)) {
1239 		struct nchcpu *cpup = ci->ci_data.cpu_nch;
1240 
1241 		ADD(stats, cpup, ncs_goodhits);
1242 		ADD(stats, cpup, ncs_neghits);
1243 		ADD(stats, cpup, ncs_badhits);
1244 		ADD(stats, cpup, ncs_falsehits);
1245 		ADD(stats, cpup, ncs_miss);
1246 		ADD(stats, cpup, ncs_long);
1247 		ADD(stats, cpup, ncs_pass2);
1248 		ADD(stats, cpup, ncs_2passes);
1249 		ADD(stats, cpup, ncs_revhits);
1250 		ADD(stats, cpup, ncs_revmiss);
1251 	}
1252 #endif	/* CACHE_STATS_CURRENT */
1253 	mutex_exit(&my_cpup->cpu_lock);
1254 	sysctl_relock();
1255 
1256 	*oldlenp = sizeof(stats);
1257 	return sysctl_copyout(l, &stats, oldp, sizeof(stats));
1258 }
1259 
1260 static void
1261 sysctl_cache_stat_setup(void)
1262 {
1263 
1264 	KASSERT(sysctllog == NULL);
1265 	sysctl_createv(&sysctllog, 0, NULL, NULL,
1266 		       CTLFLAG_PERMANENT,
1267 		       CTLTYPE_STRUCT, "namecache_stats",
1268 		       SYSCTL_DESCR("namecache statistics"),
1269 		       cache_stat_sysctl, 0, NULL, 0,
1270 		       CTL_VFS, CTL_CREATE, CTL_EOL);
1271 }
1272