37
38#include "opt_kdtrace.h"
39#include "opt_ktrace.h"
40
41#include <sys/param.h>
42#include <sys/filedesc.h>
43#include <sys/fnv_hash.h>
44#include <sys/kernel.h>
45#include <sys/lock.h>
46#include <sys/malloc.h>
47#include <sys/mount.h>
48#include <sys/namei.h>
49#include <sys/proc.h>
50#include <sys/rwlock.h>
51#include <sys/sdt.h>
52#include <sys/syscallsubr.h>
53#include <sys/sysctl.h>
54#include <sys/sysproto.h>
55#include <sys/systm.h>
56#include <sys/vnode.h>
57#ifdef KTRACE
58#include <sys/ktrace.h>
59#endif
60
61#include <vm/uma.h>
62
63SDT_PROVIDER_DECLARE(vfs);
64SDT_PROBE_DEFINE3(vfs, namecache, enter, done, done, "struct vnode *", "char *",
65 "struct vnode *");
66SDT_PROBE_DEFINE2(vfs, namecache, enter_negative, done, done, "struct vnode *",
67 "char *");
68SDT_PROBE_DEFINE1(vfs, namecache, fullpath, entry, entry, "struct vnode *");
69SDT_PROBE_DEFINE3(vfs, namecache, fullpath, hit, hit, "struct vnode *",
70 "struct char *", "struct vnode *");
71SDT_PROBE_DEFINE1(vfs, namecache, fullpath, miss, miss, "struct vnode *");
72SDT_PROBE_DEFINE3(vfs, namecache, fullpath, return, return, "int",
73 "struct vnode *", "struct char *");
74SDT_PROBE_DEFINE3(vfs, namecache, lookup, hit, hit, "struct vnode *", "char *",
75 "struct vnode *");
76SDT_PROBE_DEFINE2(vfs, namecache, lookup, hit_negative, hit-negative,
77 "struct vnode *", "char *");
78SDT_PROBE_DEFINE2(vfs, namecache, lookup, miss, miss, "struct vnode *",
79 "char *");
80SDT_PROBE_DEFINE1(vfs, namecache, purge, done, done, "struct vnode *");
81SDT_PROBE_DEFINE1(vfs, namecache, purge_negative, done, done, "struct vnode *");
82SDT_PROBE_DEFINE1(vfs, namecache, purgevfs, done, done, "struct mount *");
83SDT_PROBE_DEFINE3(vfs, namecache, zap, done, done, "struct vnode *", "char *",
84 "struct vnode *");
85SDT_PROBE_DEFINE2(vfs, namecache, zap_negative, done, done, "struct vnode *",
86 "char *");
87
88/*
89 * This structure describes the elements in the cache of recent
90 * names looked up by namei.
91 */
92
93struct namecache {
94 LIST_ENTRY(namecache) nc_hash; /* hash chain */
95 LIST_ENTRY(namecache) nc_src; /* source vnode list */
96 TAILQ_ENTRY(namecache) nc_dst; /* destination vnode list */
97 struct vnode *nc_dvp; /* vnode of parent of name */
98 struct vnode *nc_vp; /* vnode the name refers to */
99 u_char nc_flag; /* flag bits */
100 u_char nc_nlen; /* length of name */
101 char nc_name[0]; /* segment name + nul */
102};
103
104/*
105 * Name caching works as follows:
106 *
107 * Names found by directory scans are retained in a cache
108 * for future reference. It is managed LRU, so frequently
109 * used names will hang around. Cache is indexed by hash value
110 * obtained from (vp, name) where vp refers to the directory
111 * containing name.
112 *
113 * If it is a "negative" entry, (i.e. for a name that is known NOT to
114 * exist) the vnode pointer will be NULL.
115 *
116 * Upon reaching the last segment of a path, if the reference
117 * is for DELETE, or NOCACHE is set (rewrite), and the
118 * name is located in the cache, it will be dropped.
119 */
120
121/*
122 * Structures associated with name cacheing.
123 */
124#define NCHHASH(hash) \
125 (&nchashtbl[(hash) & nchash])
126static LIST_HEAD(nchashhead, namecache) *nchashtbl; /* Hash Table */
127static TAILQ_HEAD(, namecache) ncneg; /* Hash Table */
128static u_long nchash; /* size of hash table */
129SYSCTL_ULONG(_debug, OID_AUTO, nchash, CTLFLAG_RD, &nchash, 0,
130 "Size of namecache hash table");
131static u_long ncnegfactor = 16; /* ratio of negative entries */
132SYSCTL_ULONG(_vfs, OID_AUTO, ncnegfactor, CTLFLAG_RW, &ncnegfactor, 0,
133 "Ratio of negative namecache entries");
134static u_long numneg; /* number of negative entries allocated */
135SYSCTL_ULONG(_debug, OID_AUTO, numneg, CTLFLAG_RD, &numneg, 0,
136 "Number of negative entries in namecache");
137static u_long numcache; /* number of cache entries allocated */
138SYSCTL_ULONG(_debug, OID_AUTO, numcache, CTLFLAG_RD, &numcache, 0,
139 "Number of namecache entries");
140static u_long numcachehv; /* number of cache entries with vnodes held */
141SYSCTL_ULONG(_debug, OID_AUTO, numcachehv, CTLFLAG_RD, &numcachehv, 0,
142 "Number of namecache entries with vnodes held");
143static u_int ncsizefactor = 2;
144SYSCTL_UINT(_vfs, OID_AUTO, ncsizefactor, CTLFLAG_RW, &ncsizefactor, 0,
145 "Size factor for namecache");
146
147struct nchstats nchstats; /* cache effectiveness statistics */
148
149static struct rwlock cache_lock;
150RW_SYSINIT(vfscache, &cache_lock, "Name Cache");
151
152#define CACHE_UPGRADE_LOCK() rw_try_upgrade(&cache_lock)
153#define CACHE_RLOCK() rw_rlock(&cache_lock)
154#define CACHE_RUNLOCK() rw_runlock(&cache_lock)
155#define CACHE_WLOCK() rw_wlock(&cache_lock)
156#define CACHE_WUNLOCK() rw_wunlock(&cache_lock)
157
158/*
159 * UMA zones for the VFS cache.
160 *
161 * The small cache is used for entries with short names, which are the
162 * most common. The large cache is used for entries which are too big to
163 * fit in the small cache.
164 */
165static uma_zone_t cache_zone_small;
166static uma_zone_t cache_zone_large;
167
168#define CACHE_PATH_CUTOFF 35
169#define CACHE_ZONE_SMALL (sizeof(struct namecache) + CACHE_PATH_CUTOFF \
170 + 1)
171#define CACHE_ZONE_LARGE (sizeof(struct namecache) + NAME_MAX + 1)
172
173#define cache_alloc(len) uma_zalloc(((len) <= CACHE_PATH_CUTOFF) ? \
174 cache_zone_small : cache_zone_large, M_WAITOK)
175#define cache_free(ncp) do { \
176 if (ncp != NULL) \
177 uma_zfree(((ncp)->nc_nlen <= CACHE_PATH_CUTOFF) ? \
178 cache_zone_small : cache_zone_large, (ncp)); \
179} while (0)
180
181static int doingcache = 1; /* 1 => enable the cache */
182SYSCTL_INT(_debug, OID_AUTO, vfscache, CTLFLAG_RW, &doingcache, 0,
183 "VFS namecache enabled");
184
185/* Export size information to userland */
186SYSCTL_INT(_debug_sizeof, OID_AUTO, namecache, CTLFLAG_RD, 0,
187 sizeof(struct namecache), "sizeof(struct namecache)");
188
189/*
190 * The new name cache statistics
191 */
192static SYSCTL_NODE(_vfs, OID_AUTO, cache, CTLFLAG_RW, 0,
193 "Name cache statistics");
194#define STATNODE(mode, name, var, descr) \
195 SYSCTL_ULONG(_vfs_cache, OID_AUTO, name, mode, var, 0, descr);
196STATNODE(CTLFLAG_RD, numneg, &numneg, "Number of negative cache entries");
197STATNODE(CTLFLAG_RD, numcache, &numcache, "Number of cache entries");
198static u_long numcalls; STATNODE(CTLFLAG_RD, numcalls, &numcalls,
199 "Number of cache lookups");
200static u_long dothits; STATNODE(CTLFLAG_RD, dothits, &dothits,
201 "Number of '.' hits");
202static u_long dotdothits; STATNODE(CTLFLAG_RD, dotdothits, &dotdothits,
203 "Number of '..' hits");
204static u_long numchecks; STATNODE(CTLFLAG_RD, numchecks, &numchecks,
205 "Number of checks in lookup");
206static u_long nummiss; STATNODE(CTLFLAG_RD, nummiss, &nummiss,
207 "Number of cache misses");
208static u_long nummisszap; STATNODE(CTLFLAG_RD, nummisszap, &nummisszap,
209 "Number of cache misses we do not want to cache");
210static u_long numposzaps; STATNODE(CTLFLAG_RD, numposzaps, &numposzaps,
211 "Number of cache hits (positive) we do not want to cache");
212static u_long numposhits; STATNODE(CTLFLAG_RD, numposhits, &numposhits,
213 "Number of cache hits (positive)");
214static u_long numnegzaps; STATNODE(CTLFLAG_RD, numnegzaps, &numnegzaps,
215 "Number of cache hits (negative) we do not want to cache");
216static u_long numneghits; STATNODE(CTLFLAG_RD, numneghits, &numneghits,
217 "Number of cache hits (negative)");
218static u_long numupgrades; STATNODE(CTLFLAG_RD, numupgrades, &numupgrades,
219 "Number of updates of the cache after lookup (write lock + retry)");
220
221SYSCTL_OPAQUE(_vfs_cache, OID_AUTO, nchstats, CTLFLAG_RD | CTLFLAG_MPSAFE,
222 &nchstats, sizeof(nchstats), "LU",
223 "VFS cache effectiveness statistics");
224
225
226
227static void cache_zap(struct namecache *ncp);
228static int vn_vptocnp_locked(struct vnode **vp, struct ucred *cred, char *buf,
229 u_int *buflen);
230static int vn_fullpath1(struct thread *td, struct vnode *vp, struct vnode *rdir,
231 char *buf, char **retbuf, u_int buflen);
232
233static MALLOC_DEFINE(M_VFSCACHE, "vfscache", "VFS name cache entries");
234
235/*
236 * Flags in namecache.nc_flag
237 */
238#define NCF_WHITE 0x01
239#define NCF_ISDOTDOT 0x02
240
241#ifdef DIAGNOSTIC
242/*
243 * Grab an atomic snapshot of the name cache hash chain lengths
244 */
245static SYSCTL_NODE(_debug, OID_AUTO, hashstat, CTLFLAG_RW, NULL,
246 "hash table stats");
247
248static int
249sysctl_debug_hashstat_rawnchash(SYSCTL_HANDLER_ARGS)
250{
251 int error;
252 struct nchashhead *ncpp;
253 struct namecache *ncp;
254 int n_nchash;
255 int count;
256
257 n_nchash = nchash + 1; /* nchash is max index, not count */
258 if (!req->oldptr)
259 return SYSCTL_OUT(req, 0, n_nchash * sizeof(int));
260
261 /* Scan hash tables for applicable entries */
262 for (ncpp = nchashtbl; n_nchash > 0; n_nchash--, ncpp++) {
263 CACHE_RLOCK();
264 count = 0;
265 LIST_FOREACH(ncp, ncpp, nc_hash) {
266 count++;
267 }
268 CACHE_RUNLOCK();
269 error = SYSCTL_OUT(req, &count, sizeof(count));
270 if (error)
271 return (error);
272 }
273 return (0);
274}
275SYSCTL_PROC(_debug_hashstat, OID_AUTO, rawnchash, CTLTYPE_INT|CTLFLAG_RD|
276 CTLFLAG_MPSAFE, 0, 0, sysctl_debug_hashstat_rawnchash, "S,int",
277 "nchash chain lengths");
278
279static int
280sysctl_debug_hashstat_nchash(SYSCTL_HANDLER_ARGS)
281{
282 int error;
283 struct nchashhead *ncpp;
284 struct namecache *ncp;
285 int n_nchash;
286 int count, maxlength, used, pct;
287
288 if (!req->oldptr)
289 return SYSCTL_OUT(req, 0, 4 * sizeof(int));
290
291 n_nchash = nchash + 1; /* nchash is max index, not count */
292 used = 0;
293 maxlength = 0;
294
295 /* Scan hash tables for applicable entries */
296 for (ncpp = nchashtbl; n_nchash > 0; n_nchash--, ncpp++) {
297 count = 0;
298 CACHE_RLOCK();
299 LIST_FOREACH(ncp, ncpp, nc_hash) {
300 count++;
301 }
302 CACHE_RUNLOCK();
303 if (count)
304 used++;
305 if (maxlength < count)
306 maxlength = count;
307 }
308 n_nchash = nchash + 1;
309 pct = (used * 100 * 100) / n_nchash;
310 error = SYSCTL_OUT(req, &n_nchash, sizeof(n_nchash));
311 if (error)
312 return (error);
313 error = SYSCTL_OUT(req, &used, sizeof(used));
314 if (error)
315 return (error);
316 error = SYSCTL_OUT(req, &maxlength, sizeof(maxlength));
317 if (error)
318 return (error);
319 error = SYSCTL_OUT(req, &pct, sizeof(pct));
320 if (error)
321 return (error);
322 return (0);
323}
324SYSCTL_PROC(_debug_hashstat, OID_AUTO, nchash, CTLTYPE_INT|CTLFLAG_RD|
325 CTLFLAG_MPSAFE, 0, 0, sysctl_debug_hashstat_nchash, "I",
326 "nchash chain lengths");
327#endif
328
329/*
330 * cache_zap():
331 *
332 * Removes a namecache entry from cache, whether it contains an actual
333 * pointer to a vnode or if it is just a negative cache entry.
334 */
335static void
336cache_zap(ncp)
337 struct namecache *ncp;
338{
339 struct vnode *vp;
340
341 rw_assert(&cache_lock, RA_WLOCKED);
342 CTR2(KTR_VFS, "cache_zap(%p) vp %p", ncp, ncp->nc_vp);
343#ifdef KDTRACE_HOOKS
344 if (ncp->nc_vp != NULL) {
345 SDT_PROBE(vfs, namecache, zap, done, ncp->nc_dvp,
346 ncp->nc_name, ncp->nc_vp, 0, 0);
347 } else {
348 SDT_PROBE(vfs, namecache, zap_negative, done, ncp->nc_dvp,
349 ncp->nc_name, 0, 0, 0);
350 }
351#endif
352 vp = NULL;
353 LIST_REMOVE(ncp, nc_hash);
354 if (ncp->nc_flag & NCF_ISDOTDOT) {
355 if (ncp == ncp->nc_dvp->v_cache_dd)
356 ncp->nc_dvp->v_cache_dd = NULL;
357 } else {
358 LIST_REMOVE(ncp, nc_src);
359 if (LIST_EMPTY(&ncp->nc_dvp->v_cache_src)) {
360 vp = ncp->nc_dvp;
361 numcachehv--;
362 }
363 }
364 if (ncp->nc_vp) {
365 TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst, ncp, nc_dst);
366 if (ncp == ncp->nc_vp->v_cache_dd)
367 ncp->nc_vp->v_cache_dd = NULL;
368 } else {
369 TAILQ_REMOVE(&ncneg, ncp, nc_dst);
370 numneg--;
371 }
372 numcache--;
373 cache_free(ncp);
374 if (vp)
375 vdrop(vp);
376}
377
378/*
379 * Lookup an entry in the cache
380 *
381 * Lookup is called with dvp pointing to the directory to search,
382 * cnp pointing to the name of the entry being sought. If the lookup
383 * succeeds, the vnode is returned in *vpp, and a status of -1 is
384 * returned. If the lookup determines that the name does not exist
385 * (negative cacheing), a status of ENOENT is returned. If the lookup
386 * fails, a status of zero is returned. If the directory vnode is
387 * recycled out from under us due to a forced unmount, a status of
388 * ENOENT is returned.
389 *
390 * vpp is locked and ref'd on return. If we're looking up DOTDOT, dvp is
391 * unlocked. If we're looking up . an extra ref is taken, but the lock is
392 * not recursively acquired.
393 */
394
395int
396cache_lookup(dvp, vpp, cnp)
397 struct vnode *dvp;
398 struct vnode **vpp;
399 struct componentname *cnp;
400{
401 struct namecache *ncp;
402 uint32_t hash;
403 int error, ltype, wlocked;
404
405 if (!doingcache) {
406 cnp->cn_flags &= ~MAKEENTRY;
407 return (0);
408 }
409retry:
410 CACHE_RLOCK();
411 wlocked = 0;
412 numcalls++;
413 error = 0;
414
415retry_wlocked:
416 if (cnp->cn_nameptr[0] == '.') {
417 if (cnp->cn_namelen == 1) {
418 *vpp = dvp;
419 CTR2(KTR_VFS, "cache_lookup(%p, %s) found via .",
420 dvp, cnp->cn_nameptr);
421 dothits++;
422 SDT_PROBE(vfs, namecache, lookup, hit, dvp, ".",
423 *vpp, 0, 0);
424 goto success;
425 }
426 if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
427 dotdothits++;
428 if (dvp->v_cache_dd == NULL) {
429 SDT_PROBE(vfs, namecache, lookup, miss, dvp,
430 "..", NULL, 0, 0);
431 goto unlock;
432 }
433 if ((cnp->cn_flags & MAKEENTRY) == 0) {
434 if (!wlocked && !CACHE_UPGRADE_LOCK())
435 goto wlock;
436 if (dvp->v_cache_dd->nc_flag & NCF_ISDOTDOT)
437 cache_zap(dvp->v_cache_dd);
438 dvp->v_cache_dd = NULL;
439 CACHE_WUNLOCK();
440 return (0);
441 }
442 if (dvp->v_cache_dd->nc_flag & NCF_ISDOTDOT)
443 *vpp = dvp->v_cache_dd->nc_vp;
444 else
445 *vpp = dvp->v_cache_dd->nc_dvp;
446 /* Return failure if negative entry was found. */
447 if (*vpp == NULL) {
448 ncp = dvp->v_cache_dd;
449 goto negative_success;
450 }
451 CTR3(KTR_VFS, "cache_lookup(%p, %s) found %p via ..",
452 dvp, cnp->cn_nameptr, *vpp);
453 SDT_PROBE(vfs, namecache, lookup, hit, dvp, "..",
454 *vpp, 0, 0);
455 goto success;
456 }
457 }
458
459 hash = fnv_32_buf(cnp->cn_nameptr, cnp->cn_namelen, FNV1_32_INIT);
460 hash = fnv_32_buf(&dvp, sizeof(dvp), hash);
461 LIST_FOREACH(ncp, (NCHHASH(hash)), nc_hash) {
462 numchecks++;
463 if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen &&
464 !bcmp(ncp->nc_name, cnp->cn_nameptr, ncp->nc_nlen))
465 break;
466 }
467
468 /* We failed to find an entry */
469 if (ncp == NULL) {
470 SDT_PROBE(vfs, namecache, lookup, miss, dvp, cnp->cn_nameptr,
471 NULL, 0, 0);
472 if ((cnp->cn_flags & MAKEENTRY) == 0) {
473 nummisszap++;
474 } else {
475 nummiss++;
476 }
477 nchstats.ncs_miss++;
478 goto unlock;
479 }
480
481 /* We don't want to have an entry, so dump it */
482 if ((cnp->cn_flags & MAKEENTRY) == 0) {
483 numposzaps++;
484 nchstats.ncs_badhits++;
485 if (!wlocked && !CACHE_UPGRADE_LOCK())
486 goto wlock;
487 cache_zap(ncp);
488 CACHE_WUNLOCK();
489 return (0);
490 }
491
492 /* We found a "positive" match, return the vnode */
493 if (ncp->nc_vp) {
494 numposhits++;
495 nchstats.ncs_goodhits++;
496 *vpp = ncp->nc_vp;
497 CTR4(KTR_VFS, "cache_lookup(%p, %s) found %p via ncp %p",
498 dvp, cnp->cn_nameptr, *vpp, ncp);
499 SDT_PROBE(vfs, namecache, lookup, hit, dvp, ncp->nc_name,
500 *vpp, 0, 0);
501 goto success;
502 }
503
504negative_success:
505 /* We found a negative match, and want to create it, so purge */
506 if (cnp->cn_nameiop == CREATE) {
507 numnegzaps++;
508 nchstats.ncs_badhits++;
509 if (!wlocked && !CACHE_UPGRADE_LOCK())
510 goto wlock;
511 cache_zap(ncp);
512 CACHE_WUNLOCK();
513 return (0);
514 }
515
516 if (!wlocked && !CACHE_UPGRADE_LOCK())
517 goto wlock;
518 numneghits++;
519 /*
520 * We found a "negative" match, so we shift it to the end of
521 * the "negative" cache entries queue to satisfy LRU. Also,
522 * check to see if the entry is a whiteout; indicate this to
523 * the componentname, if so.
524 */
525 TAILQ_REMOVE(&ncneg, ncp, nc_dst);
526 TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
527 nchstats.ncs_neghits++;
528 if (ncp->nc_flag & NCF_WHITE)
529 cnp->cn_flags |= ISWHITEOUT;
530 SDT_PROBE(vfs, namecache, lookup, hit_negative, dvp, ncp->nc_name,
531 0, 0, 0);
532 CACHE_WUNLOCK();
533 return (ENOENT);
534
535wlock:
536 /*
537 * We need to update the cache after our lookup, so upgrade to
538 * a write lock and retry the operation.
539 */
540 CACHE_RUNLOCK();
541 CACHE_WLOCK();
542 numupgrades++;
543 wlocked = 1;
544 goto retry_wlocked;
545
546success:
547 /*
548 * On success we return a locked and ref'd vnode as per the lookup
549 * protocol.
550 */
551 if (dvp == *vpp) { /* lookup on "." */
552 VREF(*vpp);
553 if (wlocked)
554 CACHE_WUNLOCK();
555 else
556 CACHE_RUNLOCK();
557 /*
558 * When we lookup "." we still can be asked to lock it
559 * differently...
560 */
561 ltype = cnp->cn_lkflags & LK_TYPE_MASK;
562 if (ltype != VOP_ISLOCKED(*vpp)) {
563 if (ltype == LK_EXCLUSIVE) {
564 vn_lock(*vpp, LK_UPGRADE | LK_RETRY);
565 if ((*vpp)->v_iflag & VI_DOOMED) {
566 /* forced unmount */
567 vrele(*vpp);
568 *vpp = NULL;
569 return (ENOENT);
570 }
571 } else
572 vn_lock(*vpp, LK_DOWNGRADE | LK_RETRY);
573 }
574 return (-1);
575 }
576 ltype = 0; /* silence gcc warning */
577 if (cnp->cn_flags & ISDOTDOT) {
578 ltype = VOP_ISLOCKED(dvp);
579 VOP_UNLOCK(dvp, 0);
580 }
581 VI_LOCK(*vpp);
582 if (wlocked)
583 CACHE_WUNLOCK();
584 else
585 CACHE_RUNLOCK();
586 error = vget(*vpp, cnp->cn_lkflags | LK_INTERLOCK, cnp->cn_thread);
587 if (cnp->cn_flags & ISDOTDOT) {
588 vn_lock(dvp, ltype | LK_RETRY);
589 if (dvp->v_iflag & VI_DOOMED) {
590 if (error == 0)
591 vput(*vpp);
592 *vpp = NULL;
593 return (ENOENT);
594 }
595 }
596 if (error) {
597 *vpp = NULL;
598 goto retry;
599 }
600 if ((cnp->cn_flags & ISLASTCN) &&
601 (cnp->cn_lkflags & LK_TYPE_MASK) == LK_EXCLUSIVE) {
602 ASSERT_VOP_ELOCKED(*vpp, "cache_lookup");
603 }
604 return (-1);
605
606unlock:
607 if (wlocked)
608 CACHE_WUNLOCK();
609 else
610 CACHE_RUNLOCK();
611 return (0);
612}
613
614/*
615 * Add an entry to the cache.
616 */
617void
618cache_enter(dvp, vp, cnp)
619 struct vnode *dvp;
620 struct vnode *vp;
621 struct componentname *cnp;
622{
623 struct namecache *ncp, *n2;
624 struct nchashhead *ncpp;
625 uint32_t hash;
626 int flag;
627 int hold;
628 int zap;
629 int len;
630
631 CTR3(KTR_VFS, "cache_enter(%p, %p, %s)", dvp, vp, cnp->cn_nameptr);
632 VNASSERT(vp == NULL || (vp->v_iflag & VI_DOOMED) == 0, vp,
633 ("cache_enter: Adding a doomed vnode"));
634 VNASSERT(dvp == NULL || (dvp->v_iflag & VI_DOOMED) == 0, dvp,
635 ("cache_enter: Doomed vnode used as src"));
636
637 if (!doingcache)
638 return;
639
640 /*
641 * Avoid blowout in namecache entries.
642 */
643 if (numcache >= desiredvnodes * ncsizefactor)
644 return;
645
646 flag = 0;
647 if (cnp->cn_nameptr[0] == '.') {
648 if (cnp->cn_namelen == 1)
649 return;
650 if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
651 CACHE_WLOCK();
652 /*
653 * If dotdot entry already exists, just retarget it
654 * to new parent vnode, otherwise continue with new
655 * namecache entry allocation.
656 */
657 if ((ncp = dvp->v_cache_dd) != NULL &&
658 ncp->nc_flag & NCF_ISDOTDOT) {
659 KASSERT(ncp->nc_dvp == dvp,
660 ("wrong isdotdot parent"));
661 if (ncp->nc_vp != NULL)
662 TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst,
663 ncp, nc_dst);
664 else
665 TAILQ_REMOVE(&ncneg, ncp, nc_dst);
666 if (vp != NULL)
667 TAILQ_INSERT_HEAD(&vp->v_cache_dst,
668 ncp, nc_dst);
669 else
670 TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
671 ncp->nc_vp = vp;
672 CACHE_WUNLOCK();
673 return;
674 }
675 dvp->v_cache_dd = NULL;
676 SDT_PROBE(vfs, namecache, enter, done, dvp, "..", vp,
677 0, 0);
678 CACHE_WUNLOCK();
679 flag = NCF_ISDOTDOT;
680 }
681 }
682
683 hold = 0;
684 zap = 0;
685
686 /*
687 * Calculate the hash key and setup as much of the new
688 * namecache entry as possible before acquiring the lock.
689 */
690 ncp = cache_alloc(cnp->cn_namelen);
691 ncp->nc_vp = vp;
692 ncp->nc_dvp = dvp;
693 ncp->nc_flag = flag;
694 len = ncp->nc_nlen = cnp->cn_namelen;
695 hash = fnv_32_buf(cnp->cn_nameptr, len, FNV1_32_INIT);
696 strlcpy(ncp->nc_name, cnp->cn_nameptr, len + 1);
697 hash = fnv_32_buf(&dvp, sizeof(dvp), hash);
698 CACHE_WLOCK();
699
700 /*
701 * See if this vnode or negative entry is already in the cache
702 * with this name. This can happen with concurrent lookups of
703 * the same path name.
704 */
705 ncpp = NCHHASH(hash);
706 LIST_FOREACH(n2, ncpp, nc_hash) {
707 if (n2->nc_dvp == dvp &&
708 n2->nc_nlen == cnp->cn_namelen &&
709 !bcmp(n2->nc_name, cnp->cn_nameptr, n2->nc_nlen)) {
710 CACHE_WUNLOCK();
711 cache_free(ncp);
712 return;
713 }
714 }
715
716 if (flag == NCF_ISDOTDOT) {
717 /*
718 * See if we are trying to add .. entry, but some other lookup
719 * has populated v_cache_dd pointer already.
720 */
721 if (dvp->v_cache_dd != NULL) {
722 CACHE_WUNLOCK();
723 cache_free(ncp);
724 return;
725 }
726 KASSERT(vp == NULL || vp->v_type == VDIR,
727 ("wrong vnode type %p", vp));
728 dvp->v_cache_dd = ncp;
729 }
730
731 numcache++;
732 if (!vp) {
733 numneg++;
734 if (cnp->cn_flags & ISWHITEOUT)
735 ncp->nc_flag |= NCF_WHITE;
736 } else if (vp->v_type == VDIR) {
737 if (flag != NCF_ISDOTDOT) {
738 if ((n2 = vp->v_cache_dd) != NULL &&
739 (n2->nc_flag & NCF_ISDOTDOT) != 0)
740 cache_zap(n2);
741 vp->v_cache_dd = ncp;
742 }
743 } else {
744 vp->v_cache_dd = NULL;
745 }
746
747 /*
748 * Insert the new namecache entry into the appropriate chain
749 * within the cache entries table.
750 */
751 LIST_INSERT_HEAD(ncpp, ncp, nc_hash);
752 if (flag != NCF_ISDOTDOT) {
753 if (LIST_EMPTY(&dvp->v_cache_src)) {
754 hold = 1;
755 numcachehv++;
756 }
757 LIST_INSERT_HEAD(&dvp->v_cache_src, ncp, nc_src);
758 }
759
760 /*
761 * If the entry is "negative", we place it into the
762 * "negative" cache queue, otherwise, we place it into the
763 * destination vnode's cache entries queue.
764 */
765 if (vp) {
766 TAILQ_INSERT_HEAD(&vp->v_cache_dst, ncp, nc_dst);
767 SDT_PROBE(vfs, namecache, enter, done, dvp, ncp->nc_name, vp,
768 0, 0);
769 } else {
770 TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
771 SDT_PROBE(vfs, namecache, enter_negative, done, dvp,
772 ncp->nc_name, 0, 0, 0);
773 }
774 if (numneg * ncnegfactor > numcache) {
775 ncp = TAILQ_FIRST(&ncneg);
776 zap = 1;
777 }
778 if (hold)
779 vhold(dvp);
780 if (zap)
781 cache_zap(ncp);
782 CACHE_WUNLOCK();
783}
784
785/*
786 * Name cache initialization, from vfs_init() when we are booting
787 */
788static void
789nchinit(void *dummy __unused)
790{
791
792 TAILQ_INIT(&ncneg);
793
794 cache_zone_small = uma_zcreate("S VFS Cache", CACHE_ZONE_SMALL, NULL,
795 NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
796 cache_zone_large = uma_zcreate("L VFS Cache", CACHE_ZONE_LARGE, NULL,
797 NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
798
799 nchashtbl = hashinit(desiredvnodes * 2, M_VFSCACHE, &nchash);
800}
801SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nchinit, NULL);
802
803
804/*
805 * Invalidate all entries to a particular vnode.
806 */
807void
808cache_purge(vp)
809 struct vnode *vp;
810{
811
812 CTR1(KTR_VFS, "cache_purge(%p)", vp);
813 SDT_PROBE(vfs, namecache, purge, done, vp, 0, 0, 0, 0);
814 CACHE_WLOCK();
815 while (!LIST_EMPTY(&vp->v_cache_src))
816 cache_zap(LIST_FIRST(&vp->v_cache_src));
817 while (!TAILQ_EMPTY(&vp->v_cache_dst))
818 cache_zap(TAILQ_FIRST(&vp->v_cache_dst));
819 if (vp->v_cache_dd != NULL) {
820 KASSERT(vp->v_cache_dd->nc_flag & NCF_ISDOTDOT,
821 ("lost dotdot link"));
822 cache_zap(vp->v_cache_dd);
823 }
824 KASSERT(vp->v_cache_dd == NULL, ("incomplete purge"));
825 CACHE_WUNLOCK();
826}
827
828/*
829 * Invalidate all negative entries for a particular directory vnode.
830 */
831void
832cache_purge_negative(vp)
833 struct vnode *vp;
834{
835 struct namecache *cp, *ncp;
836
837 CTR1(KTR_VFS, "cache_purge_negative(%p)", vp);
838 SDT_PROBE(vfs, namecache, purge_negative, done, vp, 0, 0, 0, 0);
839 CACHE_WLOCK();
840 LIST_FOREACH_SAFE(cp, &vp->v_cache_src, nc_src, ncp) {
841 if (cp->nc_vp == NULL)
842 cache_zap(cp);
843 }
844 CACHE_WUNLOCK();
845}
846
847/*
848 * Flush all entries referencing a particular filesystem.
849 */
850void
851cache_purgevfs(mp)
852 struct mount *mp;
853{
854 struct nchashhead *ncpp;
855 struct namecache *ncp, *nnp;
856
857 /* Scan hash tables for applicable entries */
858 SDT_PROBE(vfs, namecache, purgevfs, done, mp, 0, 0, 0, 0);
859 CACHE_WLOCK();
860 for (ncpp = &nchashtbl[nchash]; ncpp >= nchashtbl; ncpp--) {
861 LIST_FOREACH_SAFE(ncp, ncpp, nc_hash, nnp) {
862 if (ncp->nc_dvp->v_mount == mp)
863 cache_zap(ncp);
864 }
865 }
866 CACHE_WUNLOCK();
867}
868
869/*
870 * Perform canonical checks and cache lookup and pass on to filesystem
871 * through the vop_cachedlookup only if needed.
872 */
873
874int
875vfs_cache_lookup(ap)
876 struct vop_lookup_args /* {
877 struct vnode *a_dvp;
878 struct vnode **a_vpp;
879 struct componentname *a_cnp;
880 } */ *ap;
881{
882 struct vnode *dvp;
883 int error;
884 struct vnode **vpp = ap->a_vpp;
885 struct componentname *cnp = ap->a_cnp;
886 struct ucred *cred = cnp->cn_cred;
887 int flags = cnp->cn_flags;
888 struct thread *td = cnp->cn_thread;
889
890 *vpp = NULL;
891 dvp = ap->a_dvp;
892
893 if (dvp->v_type != VDIR)
894 return (ENOTDIR);
895
896 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
897 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
898 return (EROFS);
899
900 error = VOP_ACCESS(dvp, VEXEC, cred, td);
901 if (error)
902 return (error);
903
904 error = cache_lookup(dvp, vpp, cnp);
905 if (error == 0)
906 return (VOP_CACHEDLOOKUP(dvp, vpp, cnp));
907 if (error == -1)
908 return (0);
909 return (error);
910}
911
912
913#ifndef _SYS_SYSPROTO_H_
914struct __getcwd_args {
915 u_char *buf;
916 u_int buflen;
917};
918#endif
919
920/*
921 * XXX All of these sysctls would probably be more productive dead.
922 */
923static int disablecwd;
924SYSCTL_INT(_debug, OID_AUTO, disablecwd, CTLFLAG_RW, &disablecwd, 0,
925 "Disable the getcwd syscall");
926
927/* Implementation of the getcwd syscall. */
928int
929sys___getcwd(td, uap)
930 struct thread *td;
931 struct __getcwd_args *uap;
932{
933
934 return (kern___getcwd(td, uap->buf, UIO_USERSPACE, uap->buflen));
935}
936
937int
938kern___getcwd(struct thread *td, u_char *buf, enum uio_seg bufseg, u_int buflen)
939{
940 char *bp, *tmpbuf;
941 struct filedesc *fdp;
942 struct vnode *cdir, *rdir;
943 int error, vfslocked;
944
945 if (disablecwd)
946 return (ENODEV);
947 if (buflen < 2)
948 return (EINVAL);
949 if (buflen > MAXPATHLEN)
950 buflen = MAXPATHLEN;
951
952 tmpbuf = malloc(buflen, M_TEMP, M_WAITOK);
953 fdp = td->td_proc->p_fd;
954 FILEDESC_SLOCK(fdp);
955 cdir = fdp->fd_cdir;
956 VREF(cdir);
957 rdir = fdp->fd_rdir;
958 VREF(rdir);
959 FILEDESC_SUNLOCK(fdp);
960 error = vn_fullpath1(td, cdir, rdir, tmpbuf, &bp, buflen);
961 vfslocked = VFS_LOCK_GIANT(rdir->v_mount);
962 vrele(rdir);
963 VFS_UNLOCK_GIANT(vfslocked);
964 vfslocked = VFS_LOCK_GIANT(cdir->v_mount);
965 vrele(cdir);
966 VFS_UNLOCK_GIANT(vfslocked);
967
968 if (!error) {
969 if (bufseg == UIO_SYSSPACE)
970 bcopy(bp, buf, strlen(bp) + 1);
971 else
972 error = copyout(bp, buf, strlen(bp) + 1);
973#ifdef KTRACE
974 if (KTRPOINT(curthread, KTR_NAMEI))
975 ktrnamei(bp);
976#endif
977 }
978 free(tmpbuf, M_TEMP);
979 return (error);
980}
981
982/*
983 * Thus begins the fullpath magic.
984 */
985
986#undef STATNODE
987#define STATNODE(name, descr) \
988 static u_int name; \
989 SYSCTL_UINT(_vfs_cache, OID_AUTO, name, CTLFLAG_RD, &name, 0, descr)
990
991static int disablefullpath;
992SYSCTL_INT(_debug, OID_AUTO, disablefullpath, CTLFLAG_RW, &disablefullpath, 0,
993 "Disable the vn_fullpath function");
994
995/* These count for kern___getcwd(), too. */
996STATNODE(numfullpathcalls, "Number of fullpath search calls");
997STATNODE(numfullpathfail1, "Number of fullpath search errors (ENOTDIR)");
998STATNODE(numfullpathfail2,
999 "Number of fullpath search errors (VOP_VPTOCNP failures)");
1000STATNODE(numfullpathfail4, "Number of fullpath search errors (ENOMEM)");
1001STATNODE(numfullpathfound, "Number of successful fullpath calls");
1002
1003/*
1004 * Retrieve the full filesystem path that correspond to a vnode from the name
1005 * cache (if available)
1006 */
1007int
1008vn_fullpath(struct thread *td, struct vnode *vn, char **retbuf, char **freebuf)
1009{
1010 char *buf;
1011 struct filedesc *fdp;
1012 struct vnode *rdir;
1013 int error, vfslocked;
1014
1015 if (disablefullpath)
1016 return (ENODEV);
1017 if (vn == NULL)
1018 return (EINVAL);
1019
1020 buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1021 fdp = td->td_proc->p_fd;
1022 FILEDESC_SLOCK(fdp);
1023 rdir = fdp->fd_rdir;
1024 VREF(rdir);
1025 FILEDESC_SUNLOCK(fdp);
1026 error = vn_fullpath1(td, vn, rdir, buf, retbuf, MAXPATHLEN);
1027 vfslocked = VFS_LOCK_GIANT(rdir->v_mount);
1028 vrele(rdir);
1029 VFS_UNLOCK_GIANT(vfslocked);
1030
1031 if (!error)
1032 *freebuf = buf;
1033 else
1034 free(buf, M_TEMP);
1035 return (error);
1036}
1037
1038/*
1039 * This function is similar to vn_fullpath, but it attempts to lookup the
1040 * pathname relative to the global root mount point. This is required for the
1041 * auditing sub-system, as audited pathnames must be absolute, relative to the
1042 * global root mount point.
1043 */
1044int
1045vn_fullpath_global(struct thread *td, struct vnode *vn,
1046 char **retbuf, char **freebuf)
1047{
1048 char *buf;
1049 int error;
1050
1051 if (disablefullpath)
1052 return (ENODEV);
1053 if (vn == NULL)
1054 return (EINVAL);
1055 buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1056 error = vn_fullpath1(td, vn, rootvnode, buf, retbuf, MAXPATHLEN);
1057 if (!error)
1058 *freebuf = buf;
1059 else
1060 free(buf, M_TEMP);
1061 return (error);
1062}
1063
1064int
1065vn_vptocnp(struct vnode **vp, struct ucred *cred, char *buf, u_int *buflen)
1066{
1067 int error;
1068
1069 CACHE_RLOCK();
1070 error = vn_vptocnp_locked(vp, cred, buf, buflen);
| 37
38#include "opt_kdtrace.h"
39#include "opt_ktrace.h"
40
41#include <sys/param.h>
42#include <sys/filedesc.h>
43#include <sys/fnv_hash.h>
44#include <sys/kernel.h>
45#include <sys/lock.h>
46#include <sys/malloc.h>
47#include <sys/mount.h>
48#include <sys/namei.h>
49#include <sys/proc.h>
50#include <sys/rwlock.h>
51#include <sys/sdt.h>
52#include <sys/syscallsubr.h>
53#include <sys/sysctl.h>
54#include <sys/sysproto.h>
55#include <sys/systm.h>
56#include <sys/vnode.h>
57#ifdef KTRACE
58#include <sys/ktrace.h>
59#endif
60
61#include <vm/uma.h>
62
63SDT_PROVIDER_DECLARE(vfs);
64SDT_PROBE_DEFINE3(vfs, namecache, enter, done, done, "struct vnode *", "char *",
65 "struct vnode *");
66SDT_PROBE_DEFINE2(vfs, namecache, enter_negative, done, done, "struct vnode *",
67 "char *");
68SDT_PROBE_DEFINE1(vfs, namecache, fullpath, entry, entry, "struct vnode *");
69SDT_PROBE_DEFINE3(vfs, namecache, fullpath, hit, hit, "struct vnode *",
70 "struct char *", "struct vnode *");
71SDT_PROBE_DEFINE1(vfs, namecache, fullpath, miss, miss, "struct vnode *");
72SDT_PROBE_DEFINE3(vfs, namecache, fullpath, return, return, "int",
73 "struct vnode *", "struct char *");
74SDT_PROBE_DEFINE3(vfs, namecache, lookup, hit, hit, "struct vnode *", "char *",
75 "struct vnode *");
76SDT_PROBE_DEFINE2(vfs, namecache, lookup, hit_negative, hit-negative,
77 "struct vnode *", "char *");
78SDT_PROBE_DEFINE2(vfs, namecache, lookup, miss, miss, "struct vnode *",
79 "char *");
80SDT_PROBE_DEFINE1(vfs, namecache, purge, done, done, "struct vnode *");
81SDT_PROBE_DEFINE1(vfs, namecache, purge_negative, done, done, "struct vnode *");
82SDT_PROBE_DEFINE1(vfs, namecache, purgevfs, done, done, "struct mount *");
83SDT_PROBE_DEFINE3(vfs, namecache, zap, done, done, "struct vnode *", "char *",
84 "struct vnode *");
85SDT_PROBE_DEFINE2(vfs, namecache, zap_negative, done, done, "struct vnode *",
86 "char *");
87
88/*
89 * This structure describes the elements in the cache of recent
90 * names looked up by namei.
91 */
92
93struct namecache {
94 LIST_ENTRY(namecache) nc_hash; /* hash chain */
95 LIST_ENTRY(namecache) nc_src; /* source vnode list */
96 TAILQ_ENTRY(namecache) nc_dst; /* destination vnode list */
97 struct vnode *nc_dvp; /* vnode of parent of name */
98 struct vnode *nc_vp; /* vnode the name refers to */
99 u_char nc_flag; /* flag bits */
100 u_char nc_nlen; /* length of name */
101 char nc_name[0]; /* segment name + nul */
102};
103
104/*
105 * Name caching works as follows:
106 *
107 * Names found by directory scans are retained in a cache
108 * for future reference. It is managed LRU, so frequently
109 * used names will hang around. Cache is indexed by hash value
110 * obtained from (vp, name) where vp refers to the directory
111 * containing name.
112 *
113 * If it is a "negative" entry, (i.e. for a name that is known NOT to
114 * exist) the vnode pointer will be NULL.
115 *
116 * Upon reaching the last segment of a path, if the reference
117 * is for DELETE, or NOCACHE is set (rewrite), and the
118 * name is located in the cache, it will be dropped.
119 */
120
121/*
122 * Structures associated with name cacheing.
123 */
124#define NCHHASH(hash) \
125 (&nchashtbl[(hash) & nchash])
126static LIST_HEAD(nchashhead, namecache) *nchashtbl; /* Hash Table */
127static TAILQ_HEAD(, namecache) ncneg; /* Hash Table */
128static u_long nchash; /* size of hash table */
129SYSCTL_ULONG(_debug, OID_AUTO, nchash, CTLFLAG_RD, &nchash, 0,
130 "Size of namecache hash table");
131static u_long ncnegfactor = 16; /* ratio of negative entries */
132SYSCTL_ULONG(_vfs, OID_AUTO, ncnegfactor, CTLFLAG_RW, &ncnegfactor, 0,
133 "Ratio of negative namecache entries");
134static u_long numneg; /* number of negative entries allocated */
135SYSCTL_ULONG(_debug, OID_AUTO, numneg, CTLFLAG_RD, &numneg, 0,
136 "Number of negative entries in namecache");
137static u_long numcache; /* number of cache entries allocated */
138SYSCTL_ULONG(_debug, OID_AUTO, numcache, CTLFLAG_RD, &numcache, 0,
139 "Number of namecache entries");
140static u_long numcachehv; /* number of cache entries with vnodes held */
141SYSCTL_ULONG(_debug, OID_AUTO, numcachehv, CTLFLAG_RD, &numcachehv, 0,
142 "Number of namecache entries with vnodes held");
143static u_int ncsizefactor = 2;
144SYSCTL_UINT(_vfs, OID_AUTO, ncsizefactor, CTLFLAG_RW, &ncsizefactor, 0,
145 "Size factor for namecache");
146
147struct nchstats nchstats; /* cache effectiveness statistics */
148
149static struct rwlock cache_lock;
150RW_SYSINIT(vfscache, &cache_lock, "Name Cache");
151
152#define CACHE_UPGRADE_LOCK() rw_try_upgrade(&cache_lock)
153#define CACHE_RLOCK() rw_rlock(&cache_lock)
154#define CACHE_RUNLOCK() rw_runlock(&cache_lock)
155#define CACHE_WLOCK() rw_wlock(&cache_lock)
156#define CACHE_WUNLOCK() rw_wunlock(&cache_lock)
157
158/*
159 * UMA zones for the VFS cache.
160 *
161 * The small cache is used for entries with short names, which are the
162 * most common. The large cache is used for entries which are too big to
163 * fit in the small cache.
164 */
165static uma_zone_t cache_zone_small;
166static uma_zone_t cache_zone_large;
167
168#define CACHE_PATH_CUTOFF 35
169#define CACHE_ZONE_SMALL (sizeof(struct namecache) + CACHE_PATH_CUTOFF \
170 + 1)
171#define CACHE_ZONE_LARGE (sizeof(struct namecache) + NAME_MAX + 1)
172
173#define cache_alloc(len) uma_zalloc(((len) <= CACHE_PATH_CUTOFF) ? \
174 cache_zone_small : cache_zone_large, M_WAITOK)
175#define cache_free(ncp) do { \
176 if (ncp != NULL) \
177 uma_zfree(((ncp)->nc_nlen <= CACHE_PATH_CUTOFF) ? \
178 cache_zone_small : cache_zone_large, (ncp)); \
179} while (0)
180
181static int doingcache = 1; /* 1 => enable the cache */
182SYSCTL_INT(_debug, OID_AUTO, vfscache, CTLFLAG_RW, &doingcache, 0,
183 "VFS namecache enabled");
184
185/* Export size information to userland */
186SYSCTL_INT(_debug_sizeof, OID_AUTO, namecache, CTLFLAG_RD, 0,
187 sizeof(struct namecache), "sizeof(struct namecache)");
188
189/*
190 * The new name cache statistics
191 */
192static SYSCTL_NODE(_vfs, OID_AUTO, cache, CTLFLAG_RW, 0,
193 "Name cache statistics");
194#define STATNODE(mode, name, var, descr) \
195 SYSCTL_ULONG(_vfs_cache, OID_AUTO, name, mode, var, 0, descr);
196STATNODE(CTLFLAG_RD, numneg, &numneg, "Number of negative cache entries");
197STATNODE(CTLFLAG_RD, numcache, &numcache, "Number of cache entries");
198static u_long numcalls; STATNODE(CTLFLAG_RD, numcalls, &numcalls,
199 "Number of cache lookups");
200static u_long dothits; STATNODE(CTLFLAG_RD, dothits, &dothits,
201 "Number of '.' hits");
202static u_long dotdothits; STATNODE(CTLFLAG_RD, dotdothits, &dotdothits,
203 "Number of '..' hits");
204static u_long numchecks; STATNODE(CTLFLAG_RD, numchecks, &numchecks,
205 "Number of checks in lookup");
206static u_long nummiss; STATNODE(CTLFLAG_RD, nummiss, &nummiss,
207 "Number of cache misses");
208static u_long nummisszap; STATNODE(CTLFLAG_RD, nummisszap, &nummisszap,
209 "Number of cache misses we do not want to cache");
210static u_long numposzaps; STATNODE(CTLFLAG_RD, numposzaps, &numposzaps,
211 "Number of cache hits (positive) we do not want to cache");
212static u_long numposhits; STATNODE(CTLFLAG_RD, numposhits, &numposhits,
213 "Number of cache hits (positive)");
214static u_long numnegzaps; STATNODE(CTLFLAG_RD, numnegzaps, &numnegzaps,
215 "Number of cache hits (negative) we do not want to cache");
216static u_long numneghits; STATNODE(CTLFLAG_RD, numneghits, &numneghits,
217 "Number of cache hits (negative)");
218static u_long numupgrades; STATNODE(CTLFLAG_RD, numupgrades, &numupgrades,
219 "Number of updates of the cache after lookup (write lock + retry)");
220
221SYSCTL_OPAQUE(_vfs_cache, OID_AUTO, nchstats, CTLFLAG_RD | CTLFLAG_MPSAFE,
222 &nchstats, sizeof(nchstats), "LU",
223 "VFS cache effectiveness statistics");
224
225
226
227static void cache_zap(struct namecache *ncp);
228static int vn_vptocnp_locked(struct vnode **vp, struct ucred *cred, char *buf,
229 u_int *buflen);
230static int vn_fullpath1(struct thread *td, struct vnode *vp, struct vnode *rdir,
231 char *buf, char **retbuf, u_int buflen);
232
233static MALLOC_DEFINE(M_VFSCACHE, "vfscache", "VFS name cache entries");
234
235/*
236 * Flags in namecache.nc_flag
237 */
238#define NCF_WHITE 0x01
239#define NCF_ISDOTDOT 0x02
240
241#ifdef DIAGNOSTIC
242/*
243 * Grab an atomic snapshot of the name cache hash chain lengths
244 */
245static SYSCTL_NODE(_debug, OID_AUTO, hashstat, CTLFLAG_RW, NULL,
246 "hash table stats");
247
248static int
249sysctl_debug_hashstat_rawnchash(SYSCTL_HANDLER_ARGS)
250{
251 int error;
252 struct nchashhead *ncpp;
253 struct namecache *ncp;
254 int n_nchash;
255 int count;
256
257 n_nchash = nchash + 1; /* nchash is max index, not count */
258 if (!req->oldptr)
259 return SYSCTL_OUT(req, 0, n_nchash * sizeof(int));
260
261 /* Scan hash tables for applicable entries */
262 for (ncpp = nchashtbl; n_nchash > 0; n_nchash--, ncpp++) {
263 CACHE_RLOCK();
264 count = 0;
265 LIST_FOREACH(ncp, ncpp, nc_hash) {
266 count++;
267 }
268 CACHE_RUNLOCK();
269 error = SYSCTL_OUT(req, &count, sizeof(count));
270 if (error)
271 return (error);
272 }
273 return (0);
274}
275SYSCTL_PROC(_debug_hashstat, OID_AUTO, rawnchash, CTLTYPE_INT|CTLFLAG_RD|
276 CTLFLAG_MPSAFE, 0, 0, sysctl_debug_hashstat_rawnchash, "S,int",
277 "nchash chain lengths");
278
279static int
280sysctl_debug_hashstat_nchash(SYSCTL_HANDLER_ARGS)
281{
282 int error;
283 struct nchashhead *ncpp;
284 struct namecache *ncp;
285 int n_nchash;
286 int count, maxlength, used, pct;
287
288 if (!req->oldptr)
289 return SYSCTL_OUT(req, 0, 4 * sizeof(int));
290
291 n_nchash = nchash + 1; /* nchash is max index, not count */
292 used = 0;
293 maxlength = 0;
294
295 /* Scan hash tables for applicable entries */
296 for (ncpp = nchashtbl; n_nchash > 0; n_nchash--, ncpp++) {
297 count = 0;
298 CACHE_RLOCK();
299 LIST_FOREACH(ncp, ncpp, nc_hash) {
300 count++;
301 }
302 CACHE_RUNLOCK();
303 if (count)
304 used++;
305 if (maxlength < count)
306 maxlength = count;
307 }
308 n_nchash = nchash + 1;
309 pct = (used * 100 * 100) / n_nchash;
310 error = SYSCTL_OUT(req, &n_nchash, sizeof(n_nchash));
311 if (error)
312 return (error);
313 error = SYSCTL_OUT(req, &used, sizeof(used));
314 if (error)
315 return (error);
316 error = SYSCTL_OUT(req, &maxlength, sizeof(maxlength));
317 if (error)
318 return (error);
319 error = SYSCTL_OUT(req, &pct, sizeof(pct));
320 if (error)
321 return (error);
322 return (0);
323}
324SYSCTL_PROC(_debug_hashstat, OID_AUTO, nchash, CTLTYPE_INT|CTLFLAG_RD|
325 CTLFLAG_MPSAFE, 0, 0, sysctl_debug_hashstat_nchash, "I",
326 "nchash chain lengths");
327#endif
328
329/*
330 * cache_zap():
331 *
332 * Removes a namecache entry from cache, whether it contains an actual
333 * pointer to a vnode or if it is just a negative cache entry.
334 */
335static void
336cache_zap(ncp)
337 struct namecache *ncp;
338{
339 struct vnode *vp;
340
341 rw_assert(&cache_lock, RA_WLOCKED);
342 CTR2(KTR_VFS, "cache_zap(%p) vp %p", ncp, ncp->nc_vp);
343#ifdef KDTRACE_HOOKS
344 if (ncp->nc_vp != NULL) {
345 SDT_PROBE(vfs, namecache, zap, done, ncp->nc_dvp,
346 ncp->nc_name, ncp->nc_vp, 0, 0);
347 } else {
348 SDT_PROBE(vfs, namecache, zap_negative, done, ncp->nc_dvp,
349 ncp->nc_name, 0, 0, 0);
350 }
351#endif
352 vp = NULL;
353 LIST_REMOVE(ncp, nc_hash);
354 if (ncp->nc_flag & NCF_ISDOTDOT) {
355 if (ncp == ncp->nc_dvp->v_cache_dd)
356 ncp->nc_dvp->v_cache_dd = NULL;
357 } else {
358 LIST_REMOVE(ncp, nc_src);
359 if (LIST_EMPTY(&ncp->nc_dvp->v_cache_src)) {
360 vp = ncp->nc_dvp;
361 numcachehv--;
362 }
363 }
364 if (ncp->nc_vp) {
365 TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst, ncp, nc_dst);
366 if (ncp == ncp->nc_vp->v_cache_dd)
367 ncp->nc_vp->v_cache_dd = NULL;
368 } else {
369 TAILQ_REMOVE(&ncneg, ncp, nc_dst);
370 numneg--;
371 }
372 numcache--;
373 cache_free(ncp);
374 if (vp)
375 vdrop(vp);
376}
377
378/*
379 * Lookup an entry in the cache
380 *
381 * Lookup is called with dvp pointing to the directory to search,
382 * cnp pointing to the name of the entry being sought. If the lookup
383 * succeeds, the vnode is returned in *vpp, and a status of -1 is
384 * returned. If the lookup determines that the name does not exist
385 * (negative cacheing), a status of ENOENT is returned. If the lookup
386 * fails, a status of zero is returned. If the directory vnode is
387 * recycled out from under us due to a forced unmount, a status of
388 * ENOENT is returned.
389 *
390 * vpp is locked and ref'd on return. If we're looking up DOTDOT, dvp is
391 * unlocked. If we're looking up . an extra ref is taken, but the lock is
392 * not recursively acquired.
393 */
394
395int
396cache_lookup(dvp, vpp, cnp)
397 struct vnode *dvp;
398 struct vnode **vpp;
399 struct componentname *cnp;
400{
401 struct namecache *ncp;
402 uint32_t hash;
403 int error, ltype, wlocked;
404
405 if (!doingcache) {
406 cnp->cn_flags &= ~MAKEENTRY;
407 return (0);
408 }
409retry:
410 CACHE_RLOCK();
411 wlocked = 0;
412 numcalls++;
413 error = 0;
414
415retry_wlocked:
416 if (cnp->cn_nameptr[0] == '.') {
417 if (cnp->cn_namelen == 1) {
418 *vpp = dvp;
419 CTR2(KTR_VFS, "cache_lookup(%p, %s) found via .",
420 dvp, cnp->cn_nameptr);
421 dothits++;
422 SDT_PROBE(vfs, namecache, lookup, hit, dvp, ".",
423 *vpp, 0, 0);
424 goto success;
425 }
426 if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
427 dotdothits++;
428 if (dvp->v_cache_dd == NULL) {
429 SDT_PROBE(vfs, namecache, lookup, miss, dvp,
430 "..", NULL, 0, 0);
431 goto unlock;
432 }
433 if ((cnp->cn_flags & MAKEENTRY) == 0) {
434 if (!wlocked && !CACHE_UPGRADE_LOCK())
435 goto wlock;
436 if (dvp->v_cache_dd->nc_flag & NCF_ISDOTDOT)
437 cache_zap(dvp->v_cache_dd);
438 dvp->v_cache_dd = NULL;
439 CACHE_WUNLOCK();
440 return (0);
441 }
442 if (dvp->v_cache_dd->nc_flag & NCF_ISDOTDOT)
443 *vpp = dvp->v_cache_dd->nc_vp;
444 else
445 *vpp = dvp->v_cache_dd->nc_dvp;
446 /* Return failure if negative entry was found. */
447 if (*vpp == NULL) {
448 ncp = dvp->v_cache_dd;
449 goto negative_success;
450 }
451 CTR3(KTR_VFS, "cache_lookup(%p, %s) found %p via ..",
452 dvp, cnp->cn_nameptr, *vpp);
453 SDT_PROBE(vfs, namecache, lookup, hit, dvp, "..",
454 *vpp, 0, 0);
455 goto success;
456 }
457 }
458
459 hash = fnv_32_buf(cnp->cn_nameptr, cnp->cn_namelen, FNV1_32_INIT);
460 hash = fnv_32_buf(&dvp, sizeof(dvp), hash);
461 LIST_FOREACH(ncp, (NCHHASH(hash)), nc_hash) {
462 numchecks++;
463 if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen &&
464 !bcmp(ncp->nc_name, cnp->cn_nameptr, ncp->nc_nlen))
465 break;
466 }
467
468 /* We failed to find an entry */
469 if (ncp == NULL) {
470 SDT_PROBE(vfs, namecache, lookup, miss, dvp, cnp->cn_nameptr,
471 NULL, 0, 0);
472 if ((cnp->cn_flags & MAKEENTRY) == 0) {
473 nummisszap++;
474 } else {
475 nummiss++;
476 }
477 nchstats.ncs_miss++;
478 goto unlock;
479 }
480
481 /* We don't want to have an entry, so dump it */
482 if ((cnp->cn_flags & MAKEENTRY) == 0) {
483 numposzaps++;
484 nchstats.ncs_badhits++;
485 if (!wlocked && !CACHE_UPGRADE_LOCK())
486 goto wlock;
487 cache_zap(ncp);
488 CACHE_WUNLOCK();
489 return (0);
490 }
491
492 /* We found a "positive" match, return the vnode */
493 if (ncp->nc_vp) {
494 numposhits++;
495 nchstats.ncs_goodhits++;
496 *vpp = ncp->nc_vp;
497 CTR4(KTR_VFS, "cache_lookup(%p, %s) found %p via ncp %p",
498 dvp, cnp->cn_nameptr, *vpp, ncp);
499 SDT_PROBE(vfs, namecache, lookup, hit, dvp, ncp->nc_name,
500 *vpp, 0, 0);
501 goto success;
502 }
503
504negative_success:
505 /* We found a negative match, and want to create it, so purge */
506 if (cnp->cn_nameiop == CREATE) {
507 numnegzaps++;
508 nchstats.ncs_badhits++;
509 if (!wlocked && !CACHE_UPGRADE_LOCK())
510 goto wlock;
511 cache_zap(ncp);
512 CACHE_WUNLOCK();
513 return (0);
514 }
515
516 if (!wlocked && !CACHE_UPGRADE_LOCK())
517 goto wlock;
518 numneghits++;
519 /*
520 * We found a "negative" match, so we shift it to the end of
521 * the "negative" cache entries queue to satisfy LRU. Also,
522 * check to see if the entry is a whiteout; indicate this to
523 * the componentname, if so.
524 */
525 TAILQ_REMOVE(&ncneg, ncp, nc_dst);
526 TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
527 nchstats.ncs_neghits++;
528 if (ncp->nc_flag & NCF_WHITE)
529 cnp->cn_flags |= ISWHITEOUT;
530 SDT_PROBE(vfs, namecache, lookup, hit_negative, dvp, ncp->nc_name,
531 0, 0, 0);
532 CACHE_WUNLOCK();
533 return (ENOENT);
534
535wlock:
536 /*
537 * We need to update the cache after our lookup, so upgrade to
538 * a write lock and retry the operation.
539 */
540 CACHE_RUNLOCK();
541 CACHE_WLOCK();
542 numupgrades++;
543 wlocked = 1;
544 goto retry_wlocked;
545
546success:
547 /*
548 * On success we return a locked and ref'd vnode as per the lookup
549 * protocol.
550 */
551 if (dvp == *vpp) { /* lookup on "." */
552 VREF(*vpp);
553 if (wlocked)
554 CACHE_WUNLOCK();
555 else
556 CACHE_RUNLOCK();
557 /*
558 * When we lookup "." we still can be asked to lock it
559 * differently...
560 */
561 ltype = cnp->cn_lkflags & LK_TYPE_MASK;
562 if (ltype != VOP_ISLOCKED(*vpp)) {
563 if (ltype == LK_EXCLUSIVE) {
564 vn_lock(*vpp, LK_UPGRADE | LK_RETRY);
565 if ((*vpp)->v_iflag & VI_DOOMED) {
566 /* forced unmount */
567 vrele(*vpp);
568 *vpp = NULL;
569 return (ENOENT);
570 }
571 } else
572 vn_lock(*vpp, LK_DOWNGRADE | LK_RETRY);
573 }
574 return (-1);
575 }
576 ltype = 0; /* silence gcc warning */
577 if (cnp->cn_flags & ISDOTDOT) {
578 ltype = VOP_ISLOCKED(dvp);
579 VOP_UNLOCK(dvp, 0);
580 }
581 VI_LOCK(*vpp);
582 if (wlocked)
583 CACHE_WUNLOCK();
584 else
585 CACHE_RUNLOCK();
586 error = vget(*vpp, cnp->cn_lkflags | LK_INTERLOCK, cnp->cn_thread);
587 if (cnp->cn_flags & ISDOTDOT) {
588 vn_lock(dvp, ltype | LK_RETRY);
589 if (dvp->v_iflag & VI_DOOMED) {
590 if (error == 0)
591 vput(*vpp);
592 *vpp = NULL;
593 return (ENOENT);
594 }
595 }
596 if (error) {
597 *vpp = NULL;
598 goto retry;
599 }
600 if ((cnp->cn_flags & ISLASTCN) &&
601 (cnp->cn_lkflags & LK_TYPE_MASK) == LK_EXCLUSIVE) {
602 ASSERT_VOP_ELOCKED(*vpp, "cache_lookup");
603 }
604 return (-1);
605
606unlock:
607 if (wlocked)
608 CACHE_WUNLOCK();
609 else
610 CACHE_RUNLOCK();
611 return (0);
612}
613
614/*
615 * Add an entry to the cache.
616 */
617void
618cache_enter(dvp, vp, cnp)
619 struct vnode *dvp;
620 struct vnode *vp;
621 struct componentname *cnp;
622{
623 struct namecache *ncp, *n2;
624 struct nchashhead *ncpp;
625 uint32_t hash;
626 int flag;
627 int hold;
628 int zap;
629 int len;
630
631 CTR3(KTR_VFS, "cache_enter(%p, %p, %s)", dvp, vp, cnp->cn_nameptr);
632 VNASSERT(vp == NULL || (vp->v_iflag & VI_DOOMED) == 0, vp,
633 ("cache_enter: Adding a doomed vnode"));
634 VNASSERT(dvp == NULL || (dvp->v_iflag & VI_DOOMED) == 0, dvp,
635 ("cache_enter: Doomed vnode used as src"));
636
637 if (!doingcache)
638 return;
639
640 /*
641 * Avoid blowout in namecache entries.
642 */
643 if (numcache >= desiredvnodes * ncsizefactor)
644 return;
645
646 flag = 0;
647 if (cnp->cn_nameptr[0] == '.') {
648 if (cnp->cn_namelen == 1)
649 return;
650 if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
651 CACHE_WLOCK();
652 /*
653 * If dotdot entry already exists, just retarget it
654 * to new parent vnode, otherwise continue with new
655 * namecache entry allocation.
656 */
657 if ((ncp = dvp->v_cache_dd) != NULL &&
658 ncp->nc_flag & NCF_ISDOTDOT) {
659 KASSERT(ncp->nc_dvp == dvp,
660 ("wrong isdotdot parent"));
661 if (ncp->nc_vp != NULL)
662 TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst,
663 ncp, nc_dst);
664 else
665 TAILQ_REMOVE(&ncneg, ncp, nc_dst);
666 if (vp != NULL)
667 TAILQ_INSERT_HEAD(&vp->v_cache_dst,
668 ncp, nc_dst);
669 else
670 TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
671 ncp->nc_vp = vp;
672 CACHE_WUNLOCK();
673 return;
674 }
675 dvp->v_cache_dd = NULL;
676 SDT_PROBE(vfs, namecache, enter, done, dvp, "..", vp,
677 0, 0);
678 CACHE_WUNLOCK();
679 flag = NCF_ISDOTDOT;
680 }
681 }
682
683 hold = 0;
684 zap = 0;
685
686 /*
687 * Calculate the hash key and setup as much of the new
688 * namecache entry as possible before acquiring the lock.
689 */
690 ncp = cache_alloc(cnp->cn_namelen);
691 ncp->nc_vp = vp;
692 ncp->nc_dvp = dvp;
693 ncp->nc_flag = flag;
694 len = ncp->nc_nlen = cnp->cn_namelen;
695 hash = fnv_32_buf(cnp->cn_nameptr, len, FNV1_32_INIT);
696 strlcpy(ncp->nc_name, cnp->cn_nameptr, len + 1);
697 hash = fnv_32_buf(&dvp, sizeof(dvp), hash);
698 CACHE_WLOCK();
699
700 /*
701 * See if this vnode or negative entry is already in the cache
702 * with this name. This can happen with concurrent lookups of
703 * the same path name.
704 */
705 ncpp = NCHHASH(hash);
706 LIST_FOREACH(n2, ncpp, nc_hash) {
707 if (n2->nc_dvp == dvp &&
708 n2->nc_nlen == cnp->cn_namelen &&
709 !bcmp(n2->nc_name, cnp->cn_nameptr, n2->nc_nlen)) {
710 CACHE_WUNLOCK();
711 cache_free(ncp);
712 return;
713 }
714 }
715
716 if (flag == NCF_ISDOTDOT) {
717 /*
718 * See if we are trying to add .. entry, but some other lookup
719 * has populated v_cache_dd pointer already.
720 */
721 if (dvp->v_cache_dd != NULL) {
722 CACHE_WUNLOCK();
723 cache_free(ncp);
724 return;
725 }
726 KASSERT(vp == NULL || vp->v_type == VDIR,
727 ("wrong vnode type %p", vp));
728 dvp->v_cache_dd = ncp;
729 }
730
731 numcache++;
732 if (!vp) {
733 numneg++;
734 if (cnp->cn_flags & ISWHITEOUT)
735 ncp->nc_flag |= NCF_WHITE;
736 } else if (vp->v_type == VDIR) {
737 if (flag != NCF_ISDOTDOT) {
738 if ((n2 = vp->v_cache_dd) != NULL &&
739 (n2->nc_flag & NCF_ISDOTDOT) != 0)
740 cache_zap(n2);
741 vp->v_cache_dd = ncp;
742 }
743 } else {
744 vp->v_cache_dd = NULL;
745 }
746
747 /*
748 * Insert the new namecache entry into the appropriate chain
749 * within the cache entries table.
750 */
751 LIST_INSERT_HEAD(ncpp, ncp, nc_hash);
752 if (flag != NCF_ISDOTDOT) {
753 if (LIST_EMPTY(&dvp->v_cache_src)) {
754 hold = 1;
755 numcachehv++;
756 }
757 LIST_INSERT_HEAD(&dvp->v_cache_src, ncp, nc_src);
758 }
759
760 /*
761 * If the entry is "negative", we place it into the
762 * "negative" cache queue, otherwise, we place it into the
763 * destination vnode's cache entries queue.
764 */
765 if (vp) {
766 TAILQ_INSERT_HEAD(&vp->v_cache_dst, ncp, nc_dst);
767 SDT_PROBE(vfs, namecache, enter, done, dvp, ncp->nc_name, vp,
768 0, 0);
769 } else {
770 TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
771 SDT_PROBE(vfs, namecache, enter_negative, done, dvp,
772 ncp->nc_name, 0, 0, 0);
773 }
774 if (numneg * ncnegfactor > numcache) {
775 ncp = TAILQ_FIRST(&ncneg);
776 zap = 1;
777 }
778 if (hold)
779 vhold(dvp);
780 if (zap)
781 cache_zap(ncp);
782 CACHE_WUNLOCK();
783}
784
785/*
786 * Name cache initialization, from vfs_init() when we are booting
787 */
788static void
789nchinit(void *dummy __unused)
790{
791
792 TAILQ_INIT(&ncneg);
793
794 cache_zone_small = uma_zcreate("S VFS Cache", CACHE_ZONE_SMALL, NULL,
795 NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
796 cache_zone_large = uma_zcreate("L VFS Cache", CACHE_ZONE_LARGE, NULL,
797 NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
798
799 nchashtbl = hashinit(desiredvnodes * 2, M_VFSCACHE, &nchash);
800}
801SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nchinit, NULL);
802
803
804/*
805 * Invalidate all entries to a particular vnode.
806 */
807void
808cache_purge(vp)
809 struct vnode *vp;
810{
811
812 CTR1(KTR_VFS, "cache_purge(%p)", vp);
813 SDT_PROBE(vfs, namecache, purge, done, vp, 0, 0, 0, 0);
814 CACHE_WLOCK();
815 while (!LIST_EMPTY(&vp->v_cache_src))
816 cache_zap(LIST_FIRST(&vp->v_cache_src));
817 while (!TAILQ_EMPTY(&vp->v_cache_dst))
818 cache_zap(TAILQ_FIRST(&vp->v_cache_dst));
819 if (vp->v_cache_dd != NULL) {
820 KASSERT(vp->v_cache_dd->nc_flag & NCF_ISDOTDOT,
821 ("lost dotdot link"));
822 cache_zap(vp->v_cache_dd);
823 }
824 KASSERT(vp->v_cache_dd == NULL, ("incomplete purge"));
825 CACHE_WUNLOCK();
826}
827
828/*
829 * Invalidate all negative entries for a particular directory vnode.
830 */
831void
832cache_purge_negative(vp)
833 struct vnode *vp;
834{
835 struct namecache *cp, *ncp;
836
837 CTR1(KTR_VFS, "cache_purge_negative(%p)", vp);
838 SDT_PROBE(vfs, namecache, purge_negative, done, vp, 0, 0, 0, 0);
839 CACHE_WLOCK();
840 LIST_FOREACH_SAFE(cp, &vp->v_cache_src, nc_src, ncp) {
841 if (cp->nc_vp == NULL)
842 cache_zap(cp);
843 }
844 CACHE_WUNLOCK();
845}
846
847/*
848 * Flush all entries referencing a particular filesystem.
849 */
850void
851cache_purgevfs(mp)
852 struct mount *mp;
853{
854 struct nchashhead *ncpp;
855 struct namecache *ncp, *nnp;
856
857 /* Scan hash tables for applicable entries */
858 SDT_PROBE(vfs, namecache, purgevfs, done, mp, 0, 0, 0, 0);
859 CACHE_WLOCK();
860 for (ncpp = &nchashtbl[nchash]; ncpp >= nchashtbl; ncpp--) {
861 LIST_FOREACH_SAFE(ncp, ncpp, nc_hash, nnp) {
862 if (ncp->nc_dvp->v_mount == mp)
863 cache_zap(ncp);
864 }
865 }
866 CACHE_WUNLOCK();
867}
868
869/*
870 * Perform canonical checks and cache lookup and pass on to filesystem
871 * through the vop_cachedlookup only if needed.
872 */
873
874int
875vfs_cache_lookup(ap)
876 struct vop_lookup_args /* {
877 struct vnode *a_dvp;
878 struct vnode **a_vpp;
879 struct componentname *a_cnp;
880 } */ *ap;
881{
882 struct vnode *dvp;
883 int error;
884 struct vnode **vpp = ap->a_vpp;
885 struct componentname *cnp = ap->a_cnp;
886 struct ucred *cred = cnp->cn_cred;
887 int flags = cnp->cn_flags;
888 struct thread *td = cnp->cn_thread;
889
890 *vpp = NULL;
891 dvp = ap->a_dvp;
892
893 if (dvp->v_type != VDIR)
894 return (ENOTDIR);
895
896 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
897 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
898 return (EROFS);
899
900 error = VOP_ACCESS(dvp, VEXEC, cred, td);
901 if (error)
902 return (error);
903
904 error = cache_lookup(dvp, vpp, cnp);
905 if (error == 0)
906 return (VOP_CACHEDLOOKUP(dvp, vpp, cnp));
907 if (error == -1)
908 return (0);
909 return (error);
910}
911
912
913#ifndef _SYS_SYSPROTO_H_
914struct __getcwd_args {
915 u_char *buf;
916 u_int buflen;
917};
918#endif
919
920/*
921 * XXX All of these sysctls would probably be more productive dead.
922 */
923static int disablecwd;
924SYSCTL_INT(_debug, OID_AUTO, disablecwd, CTLFLAG_RW, &disablecwd, 0,
925 "Disable the getcwd syscall");
926
927/* Implementation of the getcwd syscall. */
928int
929sys___getcwd(td, uap)
930 struct thread *td;
931 struct __getcwd_args *uap;
932{
933
934 return (kern___getcwd(td, uap->buf, UIO_USERSPACE, uap->buflen));
935}
936
937int
938kern___getcwd(struct thread *td, u_char *buf, enum uio_seg bufseg, u_int buflen)
939{
940 char *bp, *tmpbuf;
941 struct filedesc *fdp;
942 struct vnode *cdir, *rdir;
943 int error, vfslocked;
944
945 if (disablecwd)
946 return (ENODEV);
947 if (buflen < 2)
948 return (EINVAL);
949 if (buflen > MAXPATHLEN)
950 buflen = MAXPATHLEN;
951
952 tmpbuf = malloc(buflen, M_TEMP, M_WAITOK);
953 fdp = td->td_proc->p_fd;
954 FILEDESC_SLOCK(fdp);
955 cdir = fdp->fd_cdir;
956 VREF(cdir);
957 rdir = fdp->fd_rdir;
958 VREF(rdir);
959 FILEDESC_SUNLOCK(fdp);
960 error = vn_fullpath1(td, cdir, rdir, tmpbuf, &bp, buflen);
961 vfslocked = VFS_LOCK_GIANT(rdir->v_mount);
962 vrele(rdir);
963 VFS_UNLOCK_GIANT(vfslocked);
964 vfslocked = VFS_LOCK_GIANT(cdir->v_mount);
965 vrele(cdir);
966 VFS_UNLOCK_GIANT(vfslocked);
967
968 if (!error) {
969 if (bufseg == UIO_SYSSPACE)
970 bcopy(bp, buf, strlen(bp) + 1);
971 else
972 error = copyout(bp, buf, strlen(bp) + 1);
973#ifdef KTRACE
974 if (KTRPOINT(curthread, KTR_NAMEI))
975 ktrnamei(bp);
976#endif
977 }
978 free(tmpbuf, M_TEMP);
979 return (error);
980}
981
982/*
983 * Thus begins the fullpath magic.
984 */
985
986#undef STATNODE
987#define STATNODE(name, descr) \
988 static u_int name; \
989 SYSCTL_UINT(_vfs_cache, OID_AUTO, name, CTLFLAG_RD, &name, 0, descr)
990
991static int disablefullpath;
992SYSCTL_INT(_debug, OID_AUTO, disablefullpath, CTLFLAG_RW, &disablefullpath, 0,
993 "Disable the vn_fullpath function");
994
995/* These count for kern___getcwd(), too. */
996STATNODE(numfullpathcalls, "Number of fullpath search calls");
997STATNODE(numfullpathfail1, "Number of fullpath search errors (ENOTDIR)");
998STATNODE(numfullpathfail2,
999 "Number of fullpath search errors (VOP_VPTOCNP failures)");
1000STATNODE(numfullpathfail4, "Number of fullpath search errors (ENOMEM)");
1001STATNODE(numfullpathfound, "Number of successful fullpath calls");
1002
1003/*
1004 * Retrieve the full filesystem path that correspond to a vnode from the name
1005 * cache (if available)
1006 */
1007int
1008vn_fullpath(struct thread *td, struct vnode *vn, char **retbuf, char **freebuf)
1009{
1010 char *buf;
1011 struct filedesc *fdp;
1012 struct vnode *rdir;
1013 int error, vfslocked;
1014
1015 if (disablefullpath)
1016 return (ENODEV);
1017 if (vn == NULL)
1018 return (EINVAL);
1019
1020 buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1021 fdp = td->td_proc->p_fd;
1022 FILEDESC_SLOCK(fdp);
1023 rdir = fdp->fd_rdir;
1024 VREF(rdir);
1025 FILEDESC_SUNLOCK(fdp);
1026 error = vn_fullpath1(td, vn, rdir, buf, retbuf, MAXPATHLEN);
1027 vfslocked = VFS_LOCK_GIANT(rdir->v_mount);
1028 vrele(rdir);
1029 VFS_UNLOCK_GIANT(vfslocked);
1030
1031 if (!error)
1032 *freebuf = buf;
1033 else
1034 free(buf, M_TEMP);
1035 return (error);
1036}
1037
1038/*
1039 * This function is similar to vn_fullpath, but it attempts to lookup the
1040 * pathname relative to the global root mount point. This is required for the
1041 * auditing sub-system, as audited pathnames must be absolute, relative to the
1042 * global root mount point.
1043 */
1044int
1045vn_fullpath_global(struct thread *td, struct vnode *vn,
1046 char **retbuf, char **freebuf)
1047{
1048 char *buf;
1049 int error;
1050
1051 if (disablefullpath)
1052 return (ENODEV);
1053 if (vn == NULL)
1054 return (EINVAL);
1055 buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1056 error = vn_fullpath1(td, vn, rootvnode, buf, retbuf, MAXPATHLEN);
1057 if (!error)
1058 *freebuf = buf;
1059 else
1060 free(buf, M_TEMP);
1061 return (error);
1062}
1063
1064int
1065vn_vptocnp(struct vnode **vp, struct ucred *cred, char *buf, u_int *buflen)
1066{
1067 int error;
1068
1069 CACHE_RLOCK();
1070 error = vn_vptocnp_locked(vp, cred, buf, buflen);
|