Home | History | Annotate | Line # | Download | only in kern
vfs_cache.c revision 1.91
      1 /*	$NetBSD: vfs_cache.c,v 1.91 2012/11/05 17:27:39 dholland 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 1.91 2012/11/05 17:27:39 dholland Exp $");
     62 
     63 #include "opt_ddb.h"
     64 #include "opt_revcache.h"
     65 
     66 #include <sys/param.h>
     67 #include <sys/systm.h>
     68 #include <sys/time.h>
     69 #include <sys/mount.h>
     70 #include <sys/vnode.h>
     71 #include <sys/namei.h>
     72 #include <sys/errno.h>
     73 #include <sys/pool.h>
     74 #include <sys/mutex.h>
     75 #include <sys/atomic.h>
     76 #include <sys/kthread.h>
     77 #include <sys/kernel.h>
     78 #include <sys/cpu.h>
     79 #include <sys/evcnt.h>
     80 
     81 #define NAMECACHE_ENTER_REVERSE
     82 /*
     83  * Name caching works as follows:
     84  *
     85  * Names found by directory scans are retained in a cache
     86  * for future reference.  It is managed LRU, so frequently
     87  * used names will hang around.  Cache is indexed by hash value
     88  * obtained from (dvp, name) where dvp refers to the directory
     89  * containing name.
     90  *
     91  * For simplicity (and economy of storage), names longer than
     92  * a maximum length of NCHNAMLEN are not cached; they occur
     93  * infrequently in any case, and are almost never of interest.
     94  *
     95  * Upon reaching the last segment of a path, if the reference
     96  * is for DELETE, or NOCACHE is set (rewrite), and the
     97  * name is located in the cache, it will be dropped.
     98  * The entry is dropped also when it was not possible to lock
     99  * the cached vnode, either because vget() failed or the generation
    100  * number has changed while waiting for the lock.
    101  */
    102 
    103 /*
    104  * Per-cpu namecache data.
    105  */
    106 struct nchcpu {
    107 	kmutex_t	cpu_lock;
    108 	struct nchstats	cpu_stats;
    109 };
    110 
    111 /*
    112  * The type for the hash code. While the hash function generates a
    113  * u32, the hash code has historically been passed around as a u_long,
    114  * and the value is modified by xor'ing a uintptr_t, so it's not
    115  * entirely clear what the best type is. For now I'll leave it
    116  * unchanged as u_long.
    117  */
    118 
    119 typedef u_long nchash_t;
    120 
    121 /*
    122  * Structures associated with name cacheing.
    123  */
    124 
    125 static kmutex_t *namecache_lock __read_mostly;
    126 static pool_cache_t namecache_cache __read_mostly;
    127 static TAILQ_HEAD(, namecache) nclruhead __cacheline_aligned;
    128 
    129 static LIST_HEAD(nchashhead, namecache) *nchashtbl __read_mostly;
    130 static u_long	nchash __read_mostly;
    131 
    132 #define	NCHASH2(hash, dvp)	\
    133 	(((hash) ^ ((uintptr_t)(dvp) >> 3)) & nchash)
    134 
    135 static LIST_HEAD(ncvhashhead, namecache) *ncvhashtbl __read_mostly;
    136 static u_long	ncvhash __read_mostly;
    137 
    138 #define	NCVHASH(vp)		(((uintptr_t)(vp) >> 3) & ncvhash)
    139 
    140 /* Number of cache entries allocated. */
    141 static long	numcache __cacheline_aligned;
    142 
    143 /* Garbage collection queue and number of entries pending in it. */
    144 static void	*cache_gcqueue;
    145 static u_int	cache_gcpend;
    146 
    147 /* Cache effectiveness statistics. */
    148 struct nchstats	nchstats __cacheline_aligned;
    149 #define	COUNT(c,x)	(c.x++)
    150 
    151 static const int cache_lowat = 95;
    152 static const int cache_hiwat = 98;
    153 static const int cache_hottime = 5;	/* number of seconds */
    154 static int doingcache = 1;		/* 1 => enable the cache */
    155 
    156 static struct evcnt cache_ev_scan;
    157 static struct evcnt cache_ev_gc;
    158 static struct evcnt cache_ev_over;
    159 static struct evcnt cache_ev_under;
    160 static struct evcnt cache_ev_forced;
    161 
    162 static void cache_invalidate(struct namecache *);
    163 static struct namecache *cache_lookup_entry(
    164     const struct vnode *, const char *, size_t);
    165 static void cache_thread(void *);
    166 static void cache_invalidate(struct namecache *);
    167 static void cache_disassociate(struct namecache *);
    168 static void cache_reclaim(void);
    169 static int cache_ctor(void *, void *, int);
    170 static void cache_dtor(void *, void *);
    171 
    172 /*
    173  * Compute the hash for an entry.
    174  *
    175  * (This is for now a wrapper around namei_hash, whose interface is
    176  * for the time being slightly inconvenient.)
    177  */
    178 static nchash_t
    179 cache_hash(const char *name, size_t namelen)
    180 {
    181 	const char *endptr;
    182 
    183 	endptr = name + namelen;
    184 	return namei_hash(name, &endptr);
    185 }
    186 
    187 /*
    188  * Invalidate a cache entry and enqueue it for garbage collection.
    189  */
    190 static void
    191 cache_invalidate(struct namecache *ncp)
    192 {
    193 	void *head;
    194 
    195 	KASSERT(mutex_owned(&ncp->nc_lock));
    196 
    197 	if (ncp->nc_dvp != NULL) {
    198 		ncp->nc_vp = NULL;
    199 		ncp->nc_dvp = NULL;
    200 		do {
    201 			head = cache_gcqueue;
    202 			ncp->nc_gcqueue = head;
    203 		} while (atomic_cas_ptr(&cache_gcqueue, head, ncp) != head);
    204 		atomic_inc_uint(&cache_gcpend);
    205 	}
    206 }
    207 
    208 /*
    209  * Disassociate a namecache entry from any vnodes it is attached to,
    210  * and remove from the global LRU list.
    211  */
    212 static void
    213 cache_disassociate(struct namecache *ncp)
    214 {
    215 
    216 	KASSERT(mutex_owned(namecache_lock));
    217 	KASSERT(ncp->nc_dvp == NULL);
    218 
    219 	if (ncp->nc_lru.tqe_prev != NULL) {
    220 		TAILQ_REMOVE(&nclruhead, ncp, nc_lru);
    221 		ncp->nc_lru.tqe_prev = NULL;
    222 	}
    223 	if (ncp->nc_vhash.le_prev != NULL) {
    224 		LIST_REMOVE(ncp, nc_vhash);
    225 		ncp->nc_vhash.le_prev = NULL;
    226 	}
    227 	if (ncp->nc_vlist.le_prev != NULL) {
    228 		LIST_REMOVE(ncp, nc_vlist);
    229 		ncp->nc_vlist.le_prev = NULL;
    230 	}
    231 	if (ncp->nc_dvlist.le_prev != NULL) {
    232 		LIST_REMOVE(ncp, nc_dvlist);
    233 		ncp->nc_dvlist.le_prev = NULL;
    234 	}
    235 }
    236 
    237 /*
    238  * Lock all CPUs to prevent any cache lookup activity.  Conceptually,
    239  * this locks out all "readers".
    240  */
    241 static void
    242 cache_lock_cpus(void)
    243 {
    244 	CPU_INFO_ITERATOR cii;
    245 	struct cpu_info *ci;
    246 	struct nchcpu *cpup;
    247 	long *s, *d, *m;
    248 
    249 	for (CPU_INFO_FOREACH(cii, ci)) {
    250 		cpup = ci->ci_data.cpu_nch;
    251 		mutex_enter(&cpup->cpu_lock);
    252 
    253 		/* Collate statistics. */
    254 		d = (long *)&nchstats;
    255 		s = (long *)&cpup->cpu_stats;
    256 		m = s + sizeof(nchstats) / sizeof(long);
    257 		for (; s < m; s++, d++) {
    258 			*d += *s;
    259 			*s = 0;
    260 		}
    261 	}
    262 }
    263 
    264 /*
    265  * Release all CPU locks.
    266  */
    267 static void
    268 cache_unlock_cpus(void)
    269 {
    270 	CPU_INFO_ITERATOR cii;
    271 	struct cpu_info *ci;
    272 	struct nchcpu *cpup;
    273 
    274 	for (CPU_INFO_FOREACH(cii, ci)) {
    275 		cpup = ci->ci_data.cpu_nch;
    276 		mutex_exit(&cpup->cpu_lock);
    277 	}
    278 }
    279 
    280 /*
    281  * Find a single cache entry and return it locked.  'namecache_lock' or
    282  * at least one of the per-CPU locks must be held.
    283  */
    284 static struct namecache *
    285 cache_lookup_entry(const struct vnode *dvp, const char *name, size_t namelen)
    286 {
    287 	struct nchashhead *ncpp;
    288 	struct namecache *ncp;
    289 	nchash_t hash;
    290 
    291 	KASSERT(dvp != NULL);
    292 	hash = cache_hash(name, namelen);
    293 	ncpp = &nchashtbl[NCHASH2(hash, dvp)];
    294 
    295 	LIST_FOREACH(ncp, ncpp, nc_hash) {
    296 		if (ncp->nc_dvp != dvp ||
    297 		    ncp->nc_nlen != namelen ||
    298 		    memcmp(ncp->nc_name, name, (u_int)ncp->nc_nlen))
    299 		    	continue;
    300 	    	mutex_enter(&ncp->nc_lock);
    301 		if (__predict_true(ncp->nc_dvp == dvp)) {
    302 			ncp->nc_hittime = hardclock_ticks;
    303 			return ncp;
    304 		}
    305 		/* Raced: entry has been nullified. */
    306 		mutex_exit(&ncp->nc_lock);
    307 	}
    308 
    309 	return NULL;
    310 }
    311 
    312 /*
    313  * Look for a the name in the cache. We don't do this
    314  * if the segment name is long, simply so the cache can avoid
    315  * holding long names (which would either waste space, or
    316  * add greatly to the complexity).
    317  *
    318  * Lookup is called with DVP pointing to the directory to search,
    319  * and CNP providing the name of the entry being sought: cn_nameptr
    320  * is the name, cn_namelen is its length, and cn_flags is the flags
    321  * word from the namei operation.
    322  *
    323  * DVP must be locked.
    324  *
    325  * There are three possible non-error return states:
    326  *    1. Nothing was found in the cache. Nothing is known about
    327  *       the requested name.
    328  *    2. A negative entry was found in the cache, meaning that the
    329  *       requested name definitely does not exist.
    330  *    3. A positive entry was found in the cache, meaning that the
    331  *       requested name does exist and that we are providing the
    332  *       vnode.
    333  * In these cases the results are:
    334  *    1. 0 returned; VN is set to NULL.
    335  *    2. 1 returned; VN is set to NULL.
    336  *    3. 1 returned; VN is set to the vnode found.
    337  *
    338  * The additional result argument ISWHT is set to zero, unless a
    339  * negative entry is found that was entered as a whiteout, in which
    340  * case ISWHT is set to one.
    341  *
    342  * The ISWHT_RET argument pointer may be null. In this case an
    343  * assertion is made that the whiteout flag is not set. File systems
    344  * that do not support whiteouts can/should do this.
    345  *
    346  * Filesystems that do support whiteouts should add ISWHITEOUT to
    347  * cnp->cn_flags if ISWHT comes back nonzero.
    348  *
    349  * When a vnode is returned, it is locked, as per the vnode lookup
    350  * locking protocol.
    351  *
    352  * There is no way for this function to fail, in the sense of
    353  * generating an error that requires aborting the namei operation.
    354  *
    355  * (Prior to October 2012, this function returned an integer status,
    356  * and a vnode, and mucked with the flags word in CNP for whiteouts.
    357  * The integer status was -1 for "nothing found", ENOENT for "a
    358  * negative entry found", 0 for "a positive entry found", and possibly
    359  * other errors, and the value of VN might or might not have been set
    360  * depending on what error occurred.)
    361  */
    362 int
    363 cache_lookup(struct vnode *dvp, const char *name, size_t namelen,
    364 	     uint32_t nameiop, uint32_t cnflags,
    365 	     int *iswht_ret, struct vnode **vn_ret)
    366 {
    367 	struct namecache *ncp;
    368 	struct vnode *vp;
    369 	struct nchcpu *cpup;
    370 	int error;
    371 
    372 	/* Establish default result values */
    373 	if (iswht_ret != NULL) {
    374 		*iswht_ret = 0;
    375 	}
    376 	*vn_ret = NULL;
    377 
    378 	if (__predict_false(!doingcache)) {
    379 		return 0;
    380 	}
    381 
    382 	cpup = curcpu()->ci_data.cpu_nch;
    383 	mutex_enter(&cpup->cpu_lock);
    384 	if (__predict_false(namelen > NCHNAMLEN)) {
    385 		COUNT(cpup->cpu_stats, ncs_long);
    386 		mutex_exit(&cpup->cpu_lock);
    387 		/* found nothing */
    388 		return 0;
    389 	}
    390 	ncp = cache_lookup_entry(dvp, name, namelen);
    391 	if (__predict_false(ncp == NULL)) {
    392 		COUNT(cpup->cpu_stats, ncs_miss);
    393 		mutex_exit(&cpup->cpu_lock);
    394 		/* found nothing */
    395 		return 0;
    396 	}
    397 	if ((cnflags & MAKEENTRY) == 0) {
    398 		COUNT(cpup->cpu_stats, ncs_badhits);
    399 		/*
    400 		 * Last component and we are renaming or deleting,
    401 		 * the cache entry is invalid, or otherwise don't
    402 		 * want cache entry to exist.
    403 		 */
    404 		cache_invalidate(ncp);
    405 		mutex_exit(&ncp->nc_lock);
    406 		mutex_exit(&cpup->cpu_lock);
    407 		/* found nothing */
    408 		return 0;
    409 	}
    410 	if (ncp->nc_vp == NULL) {
    411 		if (iswht_ret != NULL) {
    412 			/*
    413 			 * Restore the ISWHITEOUT flag saved earlier.
    414 			 */
    415 			KASSERT((ncp->nc_flags & ~ISWHITEOUT) == 0);
    416 			*iswht_ret = (ncp->nc_flags & ISWHITEOUT) != 0;
    417 		} else {
    418 			KASSERT(ncp->nc_flags == 0);
    419 		}
    420 
    421 		if (__predict_true(nameiop != CREATE ||
    422 		    (cnflags & ISLASTCN) == 0)) {
    423 			COUNT(cpup->cpu_stats, ncs_neghits);
    424 			mutex_exit(&ncp->nc_lock);
    425 			mutex_exit(&cpup->cpu_lock);
    426 			/* found neg entry; vn is already null from above */
    427 			return 1;
    428 		} else {
    429 			COUNT(cpup->cpu_stats, ncs_badhits);
    430 			/*
    431 			 * Last component and we are renaming or
    432 			 * deleting, the cache entry is invalid,
    433 			 * or otherwise don't want cache entry to
    434 			 * exist.
    435 			 */
    436 			cache_invalidate(ncp);
    437 			mutex_exit(&ncp->nc_lock);
    438 			mutex_exit(&cpup->cpu_lock);
    439 			/* found nothing */
    440 			return 0;
    441 		}
    442 	}
    443 
    444 	vp = ncp->nc_vp;
    445 	if (vtryget(vp)) {
    446 		mutex_exit(&ncp->nc_lock);
    447 		mutex_exit(&cpup->cpu_lock);
    448 	} else {
    449 		mutex_enter(vp->v_interlock);
    450 		mutex_exit(&ncp->nc_lock);
    451 		mutex_exit(&cpup->cpu_lock);
    452 		error = vget(vp, LK_NOWAIT);
    453 		if (error) {
    454 			KASSERT(error == EBUSY);
    455 			/*
    456 			 * This vnode is being cleaned out.
    457 			 * XXX badhits?
    458 			 */
    459 			COUNT(cpup->cpu_stats, ncs_falsehits);
    460 			/* found nothing */
    461 			return 0;
    462 		}
    463 	}
    464 
    465 #ifdef DEBUG
    466 	/*
    467 	 * since we released nb->nb_lock,
    468 	 * we can't use this pointer any more.
    469 	 */
    470 	ncp = NULL;
    471 #endif /* DEBUG */
    472 
    473 	if (vp == dvp) {	/* lookup on "." */
    474 		error = 0;
    475 	} else if (cnflags & ISDOTDOT) {
    476 		VOP_UNLOCK(dvp);
    477 		error = vn_lock(vp, LK_EXCLUSIVE);
    478 		vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
    479 	} else {
    480 		error = vn_lock(vp, LK_EXCLUSIVE);
    481 	}
    482 
    483 	/*
    484 	 * Check that the lock succeeded.
    485 	 */
    486 	if (error) {
    487 		/* We don't have the right lock, but this is only for stats. */
    488 		COUNT(cpup->cpu_stats, ncs_badhits);
    489 
    490 		vrele(vp);
    491 		/* found nothing */
    492 		return 0;
    493 	}
    494 
    495 	/* We don't have the right lock, but this is only for stats. */
    496 	COUNT(cpup->cpu_stats, ncs_goodhits);
    497 
    498 	/* found it */
    499 	*vn_ret = vp;
    500 	return 1;
    501 }
    502 
    503 int
    504 cache_lookup_raw(struct vnode *dvp, const char *name, size_t namelen,
    505 		 uint32_t cnflags,
    506 		 int *iswht_ret, struct vnode **vn_ret)
    507 {
    508 	struct namecache *ncp;
    509 	struct vnode *vp;
    510 	struct nchcpu *cpup;
    511 	int error;
    512 
    513 	/* Establish default results. */
    514 	if (iswht_ret != NULL) {
    515 		*iswht_ret = 0;
    516 	}
    517 	*vn_ret = NULL;
    518 
    519 	if (__predict_false(!doingcache)) {
    520 		/* found nothing */
    521 		return 0;
    522 	}
    523 
    524 	cpup = curcpu()->ci_data.cpu_nch;
    525 	mutex_enter(&cpup->cpu_lock);
    526 	if (__predict_false(namelen > NCHNAMLEN)) {
    527 		COUNT(cpup->cpu_stats, ncs_long);
    528 		mutex_exit(&cpup->cpu_lock);
    529 		/* found nothing */
    530 		return 0;
    531 	}
    532 	ncp = cache_lookup_entry(dvp, name, namelen);
    533 	if (__predict_false(ncp == NULL)) {
    534 		COUNT(cpup->cpu_stats, ncs_miss);
    535 		mutex_exit(&cpup->cpu_lock);
    536 		/* found nothing */
    537 		return 0;
    538 	}
    539 	vp = ncp->nc_vp;
    540 	if (vp == NULL) {
    541 		/*
    542 		 * Restore the ISWHITEOUT flag saved earlier.
    543 		 */
    544 		if (iswht_ret != NULL) {
    545 			KASSERT((ncp->nc_flags & ~ISWHITEOUT) == 0);
    546 			/*cnp->cn_flags |= ncp->nc_flags;*/
    547 			*iswht_ret = (ncp->nc_flags & ISWHITEOUT) != 0;
    548 		}
    549 		COUNT(cpup->cpu_stats, ncs_neghits);
    550 		mutex_exit(&ncp->nc_lock);
    551 		mutex_exit(&cpup->cpu_lock);
    552 		/* found negative entry; vn is already null from above */
    553 		return 1;
    554 	}
    555 	if (vtryget(vp)) {
    556 		mutex_exit(&ncp->nc_lock);
    557 		mutex_exit(&cpup->cpu_lock);
    558 	} else {
    559 		mutex_enter(vp->v_interlock);
    560 		mutex_exit(&ncp->nc_lock);
    561 		mutex_exit(&cpup->cpu_lock);
    562 		error = vget(vp, LK_NOWAIT);
    563 		if (error) {
    564 			KASSERT(error == EBUSY);
    565 			/*
    566 			 * This vnode is being cleaned out.
    567 			 * XXX badhits?
    568 			 */
    569 			COUNT(cpup->cpu_stats, ncs_falsehits);
    570 			/* found nothing */
    571 			return 0;
    572 		}
    573 	}
    574 
    575 	/* Unlocked, but only for stats. */
    576 	COUNT(cpup->cpu_stats, ncs_goodhits); /* XXX can be "badhits" */
    577 
    578 	/* found it */
    579 	*vn_ret = vp;
    580 	return 1;
    581 }
    582 
    583 /*
    584  * Scan cache looking for name of directory entry pointing at vp.
    585  *
    586  * If the lookup succeeds the vnode is referenced and stored in dvpp.
    587  *
    588  * If bufp is non-NULL, also place the name in the buffer which starts
    589  * at bufp, immediately before *bpp, and move bpp backwards to point
    590  * at the start of it.  (Yes, this is a little baroque, but it's done
    591  * this way to cater to the whims of getcwd).
    592  *
    593  * Returns 0 on success, -1 on cache miss, positive errno on failure.
    594  */
    595 int
    596 cache_revlookup(struct vnode *vp, struct vnode **dvpp, char **bpp, char *bufp)
    597 {
    598 	struct namecache *ncp;
    599 	struct vnode *dvp;
    600 	struct ncvhashhead *nvcpp;
    601 	char *bp;
    602 	int error, nlen;
    603 
    604 	if (!doingcache)
    605 		goto out;
    606 
    607 	nvcpp = &ncvhashtbl[NCVHASH(vp)];
    608 
    609 	mutex_enter(namecache_lock);
    610 	LIST_FOREACH(ncp, nvcpp, nc_vhash) {
    611 		mutex_enter(&ncp->nc_lock);
    612 		if (ncp->nc_vp == vp &&
    613 		    (dvp = ncp->nc_dvp) != NULL &&
    614 		    dvp != vp) { 		/* avoid pesky . entries.. */
    615 
    616 #ifdef DIAGNOSTIC
    617 			if (ncp->nc_nlen == 1 &&
    618 			    ncp->nc_name[0] == '.')
    619 				panic("cache_revlookup: found entry for .");
    620 
    621 			if (ncp->nc_nlen == 2 &&
    622 			    ncp->nc_name[0] == '.' &&
    623 			    ncp->nc_name[1] == '.')
    624 				panic("cache_revlookup: found entry for ..");
    625 #endif
    626 			COUNT(nchstats, ncs_revhits);
    627 			nlen = ncp->nc_nlen;
    628 
    629 			if (bufp) {
    630 				bp = *bpp;
    631 				bp -= nlen;
    632 				if (bp <= bufp) {
    633 					*dvpp = NULL;
    634 					mutex_exit(&ncp->nc_lock);
    635 					mutex_exit(namecache_lock);
    636 					return (ERANGE);
    637 				}
    638 				memcpy(bp, ncp->nc_name, nlen);
    639 				*bpp = bp;
    640 			}
    641 
    642 			if (vtryget(dvp)) {
    643 				mutex_exit(&ncp->nc_lock);
    644 				mutex_exit(namecache_lock);
    645 			} else {
    646 				mutex_enter(dvp->v_interlock);
    647 				mutex_exit(&ncp->nc_lock);
    648 				mutex_exit(namecache_lock);
    649 				error = vget(dvp, LK_NOWAIT);
    650 				if (error) {
    651 					KASSERT(error == EBUSY);
    652 					if (bufp)
    653 						(*bpp) += nlen;
    654 					*dvpp = NULL;
    655 					return -1;
    656 				}
    657 			}
    658 			*dvpp = dvp;
    659 			return (0);
    660 		}
    661 		mutex_exit(&ncp->nc_lock);
    662 	}
    663 	COUNT(nchstats, ncs_revmiss);
    664 	mutex_exit(namecache_lock);
    665  out:
    666 	*dvpp = NULL;
    667 	return (-1);
    668 }
    669 
    670 /*
    671  * Add an entry to the cache
    672  */
    673 void
    674 cache_enter(struct vnode *dvp, struct vnode *vp,
    675 	    const char *name, size_t namelen, uint32_t cnflags)
    676 {
    677 	struct namecache *ncp;
    678 	struct namecache *oncp;
    679 	struct nchashhead *ncpp;
    680 	struct ncvhashhead *nvcpp;
    681 	nchash_t hash;
    682 
    683 	/* First, check whether we can/should add a cache entry. */
    684 	if ((cnflags & MAKEENTRY) == 0 ||
    685 	    __predict_false(namelen > NCHNAMLEN || !doingcache)) {
    686 		return;
    687 	}
    688 
    689 	if (numcache > desiredvnodes) {
    690 		mutex_enter(namecache_lock);
    691 		cache_ev_forced.ev_count++;
    692 		cache_reclaim();
    693 		mutex_exit(namecache_lock);
    694 	}
    695 
    696 	ncp = pool_cache_get(namecache_cache, PR_WAITOK);
    697 	mutex_enter(namecache_lock);
    698 	numcache++;
    699 
    700 	/*
    701 	 * Concurrent lookups in the same directory may race for a
    702 	 * cache entry.  if there's a duplicated entry, free it.
    703 	 */
    704 	oncp = cache_lookup_entry(dvp, name, namelen);
    705 	if (oncp) {
    706 		cache_invalidate(oncp);
    707 		mutex_exit(&oncp->nc_lock);
    708 	}
    709 
    710 	/* Grab the vnode we just found. */
    711 	mutex_enter(&ncp->nc_lock);
    712 	ncp->nc_vp = vp;
    713 	ncp->nc_flags = 0;
    714 	ncp->nc_hittime = 0;
    715 	ncp->nc_gcqueue = NULL;
    716 	if (vp == NULL) {
    717 		/*
    718 		 * For negative hits, save the ISWHITEOUT flag so we can
    719 		 * restore it later when the cache entry is used again.
    720 		 */
    721 		ncp->nc_flags = cnflags & ISWHITEOUT;
    722 	}
    723 
    724 	/* Fill in cache info. */
    725 	ncp->nc_dvp = dvp;
    726 	LIST_INSERT_HEAD(&dvp->v_dnclist, ncp, nc_dvlist);
    727 	if (vp)
    728 		LIST_INSERT_HEAD(&vp->v_nclist, ncp, nc_vlist);
    729 	else {
    730 		ncp->nc_vlist.le_prev = NULL;
    731 		ncp->nc_vlist.le_next = NULL;
    732 	}
    733 	KASSERT(namelen <= NCHNAMLEN);
    734 	ncp->nc_nlen = namelen;
    735 	memcpy(ncp->nc_name, name, (unsigned)ncp->nc_nlen);
    736 	TAILQ_INSERT_TAIL(&nclruhead, ncp, nc_lru);
    737 	hash = cache_hash(name, namelen);
    738 	ncpp = &nchashtbl[NCHASH2(hash, dvp)];
    739 
    740 	/*
    741 	 * Flush updates before making visible in table.  No need for a
    742 	 * memory barrier on the other side: to see modifications the
    743 	 * list must be followed, meaning a dependent pointer load.
    744 	 * The below is LIST_INSERT_HEAD() inlined, with the memory
    745 	 * barrier included in the correct place.
    746 	 */
    747 	if ((ncp->nc_hash.le_next = ncpp->lh_first) != NULL)
    748 		ncpp->lh_first->nc_hash.le_prev = &ncp->nc_hash.le_next;
    749 	ncp->nc_hash.le_prev = &ncpp->lh_first;
    750 	membar_producer();
    751 	ncpp->lh_first = ncp;
    752 
    753 	ncp->nc_vhash.le_prev = NULL;
    754 	ncp->nc_vhash.le_next = NULL;
    755 
    756 	/*
    757 	 * Create reverse-cache entries (used in getcwd) for directories.
    758 	 * (and in linux procfs exe node)
    759 	 */
    760 	if (vp != NULL &&
    761 	    vp != dvp &&
    762 #ifndef NAMECACHE_ENTER_REVERSE
    763 	    vp->v_type == VDIR &&
    764 #endif
    765 	    (ncp->nc_nlen > 2 ||
    766 	    (ncp->nc_nlen > 1 && ncp->nc_name[1] != '.') ||
    767 	    (/* ncp->nc_nlen > 0 && */ ncp->nc_name[0] != '.'))) {
    768 		nvcpp = &ncvhashtbl[NCVHASH(vp)];
    769 		LIST_INSERT_HEAD(nvcpp, ncp, nc_vhash);
    770 	}
    771 	mutex_exit(&ncp->nc_lock);
    772 	mutex_exit(namecache_lock);
    773 }
    774 
    775 /*
    776  * Name cache initialization, from vfs_init() when we are booting
    777  */
    778 void
    779 nchinit(void)
    780 {
    781 	int error;
    782 
    783 	TAILQ_INIT(&nclruhead);
    784 	namecache_cache = pool_cache_init(sizeof(struct namecache),
    785 	    coherency_unit, 0, 0, "ncache", NULL, IPL_NONE, cache_ctor,
    786 	    cache_dtor, NULL);
    787 	KASSERT(namecache_cache != NULL);
    788 
    789 	namecache_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
    790 
    791 	nchashtbl = hashinit(desiredvnodes, HASH_LIST, true, &nchash);
    792 	ncvhashtbl =
    793 #ifdef NAMECACHE_ENTER_REVERSE
    794 	    hashinit(desiredvnodes, HASH_LIST, true, &ncvhash);
    795 #else
    796 	    hashinit(desiredvnodes/8, HASH_LIST, true, &ncvhash);
    797 #endif
    798 
    799 	error = kthread_create(PRI_VM, KTHREAD_MPSAFE, NULL, cache_thread,
    800 	    NULL, NULL, "cachegc");
    801 	if (error != 0)
    802 		panic("nchinit %d", error);
    803 
    804 	evcnt_attach_dynamic(&cache_ev_scan, EVCNT_TYPE_MISC, NULL,
    805 	   "namecache", "entries scanned");
    806 	evcnt_attach_dynamic(&cache_ev_gc, EVCNT_TYPE_MISC, NULL,
    807 	   "namecache", "entries collected");
    808 	evcnt_attach_dynamic(&cache_ev_over, EVCNT_TYPE_MISC, NULL,
    809 	   "namecache", "over scan target");
    810 	evcnt_attach_dynamic(&cache_ev_under, EVCNT_TYPE_MISC, NULL,
    811 	   "namecache", "under scan target");
    812 	evcnt_attach_dynamic(&cache_ev_forced, EVCNT_TYPE_MISC, NULL,
    813 	   "namecache", "forced reclaims");
    814 }
    815 
    816 static int
    817 cache_ctor(void *arg, void *obj, int flag)
    818 {
    819 	struct namecache *ncp;
    820 
    821 	ncp = obj;
    822 	mutex_init(&ncp->nc_lock, MUTEX_DEFAULT, IPL_NONE);
    823 
    824 	return 0;
    825 }
    826 
    827 static void
    828 cache_dtor(void *arg, void *obj)
    829 {
    830 	struct namecache *ncp;
    831 
    832 	ncp = obj;
    833 	mutex_destroy(&ncp->nc_lock);
    834 }
    835 
    836 /*
    837  * Called once for each CPU in the system as attached.
    838  */
    839 void
    840 cache_cpu_init(struct cpu_info *ci)
    841 {
    842 	struct nchcpu *cpup;
    843 	size_t sz;
    844 
    845 	sz = roundup2(sizeof(*cpup), coherency_unit) + coherency_unit;
    846 	cpup = kmem_zalloc(sz, KM_SLEEP);
    847 	cpup = (void *)roundup2((uintptr_t)cpup, coherency_unit);
    848 	mutex_init(&cpup->cpu_lock, MUTEX_DEFAULT, IPL_NONE);
    849 	ci->ci_data.cpu_nch = cpup;
    850 }
    851 
    852 /*
    853  * Name cache reinitialization, for when the maximum number of vnodes increases.
    854  */
    855 void
    856 nchreinit(void)
    857 {
    858 	struct namecache *ncp;
    859 	struct nchashhead *oldhash1, *hash1;
    860 	struct ncvhashhead *oldhash2, *hash2;
    861 	u_long i, oldmask1, oldmask2, mask1, mask2;
    862 
    863 	hash1 = hashinit(desiredvnodes, HASH_LIST, true, &mask1);
    864 	hash2 =
    865 #ifdef NAMECACHE_ENTER_REVERSE
    866 	    hashinit(desiredvnodes, HASH_LIST, true, &mask2);
    867 #else
    868 	    hashinit(desiredvnodes/8, HASH_LIST, true, &mask2);
    869 #endif
    870 	mutex_enter(namecache_lock);
    871 	cache_lock_cpus();
    872 	oldhash1 = nchashtbl;
    873 	oldmask1 = nchash;
    874 	nchashtbl = hash1;
    875 	nchash = mask1;
    876 	oldhash2 = ncvhashtbl;
    877 	oldmask2 = ncvhash;
    878 	ncvhashtbl = hash2;
    879 	ncvhash = mask2;
    880 	for (i = 0; i <= oldmask1; i++) {
    881 		while ((ncp = LIST_FIRST(&oldhash1[i])) != NULL) {
    882 			LIST_REMOVE(ncp, nc_hash);
    883 			ncp->nc_hash.le_prev = NULL;
    884 		}
    885 	}
    886 	for (i = 0; i <= oldmask2; i++) {
    887 		while ((ncp = LIST_FIRST(&oldhash2[i])) != NULL) {
    888 			LIST_REMOVE(ncp, nc_vhash);
    889 			ncp->nc_vhash.le_prev = NULL;
    890 		}
    891 	}
    892 	cache_unlock_cpus();
    893 	mutex_exit(namecache_lock);
    894 	hashdone(oldhash1, HASH_LIST, oldmask1);
    895 	hashdone(oldhash2, HASH_LIST, oldmask2);
    896 }
    897 
    898 /*
    899  * Cache flush, a particular vnode; called when a vnode is renamed to
    900  * hide entries that would now be invalid
    901  */
    902 void
    903 cache_purge1(struct vnode *vp, const char *name, size_t namelen, int flags)
    904 {
    905 	struct namecache *ncp, *ncnext;
    906 
    907 	mutex_enter(namecache_lock);
    908 	if (flags & PURGE_PARENTS) {
    909 		for (ncp = LIST_FIRST(&vp->v_nclist); ncp != NULL;
    910 		    ncp = ncnext) {
    911 			ncnext = LIST_NEXT(ncp, nc_vlist);
    912 			mutex_enter(&ncp->nc_lock);
    913 			cache_invalidate(ncp);
    914 			mutex_exit(&ncp->nc_lock);
    915 			cache_disassociate(ncp);
    916 		}
    917 	}
    918 	if (flags & PURGE_CHILDREN) {
    919 		for (ncp = LIST_FIRST(&vp->v_dnclist); ncp != NULL;
    920 		    ncp = ncnext) {
    921 			ncnext = LIST_NEXT(ncp, nc_dvlist);
    922 			mutex_enter(&ncp->nc_lock);
    923 			cache_invalidate(ncp);
    924 			mutex_exit(&ncp->nc_lock);
    925 			cache_disassociate(ncp);
    926 		}
    927 	}
    928 	if (name != NULL) {
    929 		ncp = cache_lookup_entry(vp, name, namelen);
    930 		if (ncp) {
    931 			cache_invalidate(ncp);
    932 			mutex_exit(&ncp->nc_lock);
    933 			cache_disassociate(ncp);
    934 		}
    935 	}
    936 	mutex_exit(namecache_lock);
    937 }
    938 
    939 /*
    940  * Cache flush, a whole filesystem; called when filesys is umounted to
    941  * remove entries that would now be invalid.
    942  */
    943 void
    944 cache_purgevfs(struct mount *mp)
    945 {
    946 	struct namecache *ncp, *nxtcp;
    947 
    948 	mutex_enter(namecache_lock);
    949 	for (ncp = TAILQ_FIRST(&nclruhead); ncp != NULL; ncp = nxtcp) {
    950 		nxtcp = TAILQ_NEXT(ncp, nc_lru);
    951 		mutex_enter(&ncp->nc_lock);
    952 		if (ncp->nc_dvp != NULL && ncp->nc_dvp->v_mount == mp) {
    953 			/* Free the resources we had. */
    954 			cache_invalidate(ncp);
    955 			cache_disassociate(ncp);
    956 		}
    957 		mutex_exit(&ncp->nc_lock);
    958 	}
    959 	cache_reclaim();
    960 	mutex_exit(namecache_lock);
    961 }
    962 
    963 /*
    964  * Scan global list invalidating entries until we meet a preset target.
    965  * Prefer to invalidate entries that have not scored a hit within
    966  * cache_hottime seconds.  We sort the LRU list only for this routine's
    967  * benefit.
    968  */
    969 static void
    970 cache_prune(int incache, int target)
    971 {
    972 	struct namecache *ncp, *nxtcp, *sentinel;
    973 	int items, recent, tryharder;
    974 
    975 	KASSERT(mutex_owned(namecache_lock));
    976 
    977 	items = 0;
    978 	tryharder = 0;
    979 	recent = hardclock_ticks - hz * cache_hottime;
    980 	sentinel = NULL;
    981 	for (ncp = TAILQ_FIRST(&nclruhead); ncp != NULL; ncp = nxtcp) {
    982 		if (incache <= target)
    983 			break;
    984 		items++;
    985 		nxtcp = TAILQ_NEXT(ncp, nc_lru);
    986 		if (ncp->nc_dvp == NULL)
    987 			continue;
    988 		if (ncp == sentinel) {
    989 			/*
    990 			 * If we looped back on ourself, then ignore
    991 			 * recent entries and purge whatever we find.
    992 			 */
    993 			tryharder = 1;
    994 		}
    995 		if (!tryharder && (ncp->nc_hittime - recent) > 0) {
    996 			if (sentinel == NULL)
    997 				sentinel = ncp;
    998 			TAILQ_REMOVE(&nclruhead, ncp, nc_lru);
    999 			TAILQ_INSERT_TAIL(&nclruhead, ncp, nc_lru);
   1000 			continue;
   1001 		}
   1002 		mutex_enter(&ncp->nc_lock);
   1003 		if (ncp->nc_dvp != NULL) {
   1004 			cache_invalidate(ncp);
   1005 			cache_disassociate(ncp);
   1006 			incache--;
   1007 		}
   1008 		mutex_exit(&ncp->nc_lock);
   1009 	}
   1010 	cache_ev_scan.ev_count += items;
   1011 }
   1012 
   1013 /*
   1014  * Collect dead cache entries from all CPUs and garbage collect.
   1015  */
   1016 static void
   1017 cache_reclaim(void)
   1018 {
   1019 	struct namecache *ncp, *next;
   1020 	int items;
   1021 
   1022 	KASSERT(mutex_owned(namecache_lock));
   1023 
   1024 	/*
   1025 	 * If the number of extant entries not awaiting garbage collection
   1026 	 * exceeds the high water mark, then reclaim stale entries until we
   1027 	 * reach our low water mark.
   1028 	 */
   1029 	items = numcache - cache_gcpend;
   1030 	if (items > (uint64_t)desiredvnodes * cache_hiwat / 100) {
   1031 		cache_prune(items, (int)((uint64_t)desiredvnodes *
   1032 		    cache_lowat / 100));
   1033 		cache_ev_over.ev_count++;
   1034 	} else
   1035 		cache_ev_under.ev_count++;
   1036 
   1037 	/*
   1038 	 * Stop forward lookup activity on all CPUs and garbage collect dead
   1039 	 * entries.
   1040 	 */
   1041 	cache_lock_cpus();
   1042 	ncp = cache_gcqueue;
   1043 	cache_gcqueue = NULL;
   1044 	items = cache_gcpend;
   1045 	cache_gcpend = 0;
   1046 	while (ncp != NULL) {
   1047 		next = ncp->nc_gcqueue;
   1048 		cache_disassociate(ncp);
   1049 		KASSERT(ncp->nc_dvp == NULL);
   1050 		if (ncp->nc_hash.le_prev != NULL) {
   1051 			LIST_REMOVE(ncp, nc_hash);
   1052 			ncp->nc_hash.le_prev = NULL;
   1053 		}
   1054 		pool_cache_put(namecache_cache, ncp);
   1055 		ncp = next;
   1056 	}
   1057 	cache_unlock_cpus();
   1058 	numcache -= items;
   1059 	cache_ev_gc.ev_count += items;
   1060 }
   1061 
   1062 /*
   1063  * Cache maintainence thread, awakening once per second to:
   1064  *
   1065  * => keep number of entries below the high water mark
   1066  * => sort pseudo-LRU list
   1067  * => garbage collect dead entries
   1068  */
   1069 static void
   1070 cache_thread(void *arg)
   1071 {
   1072 
   1073 	mutex_enter(namecache_lock);
   1074 	for (;;) {
   1075 		cache_reclaim();
   1076 		kpause("cachegc", false, hz, namecache_lock);
   1077 	}
   1078 }
   1079 
   1080 #ifdef DDB
   1081 void
   1082 namecache_print(struct vnode *vp, void (*pr)(const char *, ...))
   1083 {
   1084 	struct vnode *dvp = NULL;
   1085 	struct namecache *ncp;
   1086 
   1087 	TAILQ_FOREACH(ncp, &nclruhead, nc_lru) {
   1088 		if (ncp->nc_vp == vp && ncp->nc_dvp != NULL) {
   1089 			(*pr)("name %.*s\n", ncp->nc_nlen, ncp->nc_name);
   1090 			dvp = ncp->nc_dvp;
   1091 		}
   1092 	}
   1093 	if (dvp == NULL) {
   1094 		(*pr)("name not found\n");
   1095 		return;
   1096 	}
   1097 	vp = dvp;
   1098 	TAILQ_FOREACH(ncp, &nclruhead, nc_lru) {
   1099 		if (ncp->nc_vp == vp) {
   1100 			(*pr)("parent %.*s\n", ncp->nc_nlen, ncp->nc_name);
   1101 		}
   1102 	}
   1103 }
   1104 #endif
   1105