2 * Copyright 2000, International Business Machines Corporation and others.
5 * This software has been released under the terms of the IBM Public
6 * License. For details, see the LICENSE file in the top-level source
7 * directory or online at http://www.openafs.org/dl/license10.html
19 * afs_FlushActiveVcaches
22 * afs_WriteVCacheDiscon
40 #include <afsconfig.h>
41 #include "afs/param.h"
43 #include "afs/sysincludes.h" /*Standard vendor system headers */
44 #include "afsincludes.h" /*AFS-based standard headers */
45 #include "afs/afs_stats.h"
46 #include "afs/afs_cbqueue.h"
47 #include "afs/afs_osidnlc.h"
49 afs_int32 afs_maxvcount = 0; /* max number of vcache entries */
50 afs_int32 afs_vcount = 0; /* number of vcache in use now */
58 #endif /* AFS_SGI64_ENV */
60 /* Exported variables */
61 afs_rwlock_t afs_xvcdirty; /*Lock: discon vcache dirty list mgmt */
62 afs_rwlock_t afs_xvcache; /*Lock: alloc new stat cache entries */
63 afs_rwlock_t afs_xvreclaim; /*Lock: entries reclaimed, not on free list */
64 afs_lock_t afs_xvcb; /*Lock: fids on which there are callbacks */
65 #if !defined(AFS_LINUX22_ENV)
66 static struct vcache *freeVCList; /*Free list for stat cache entries */
67 struct vcache *ReclaimedVCList; /*Reclaimed list for stat entries */
68 static struct vcache *Initial_freeVCList; /*Initial list for above */
70 struct afs_q VLRU; /*vcache LRU */
71 afs_int32 vcachegen = 0;
72 unsigned int afs_paniconwarn = 0;
73 struct vcache *afs_vhashT[VCSIZE];
74 struct afs_q afs_vhashTV[VCSIZE];
75 static struct afs_cbr *afs_cbrHashT[CBRSIZE];
76 afs_int32 afs_bulkStatsLost;
77 int afs_norefpanic = 0;
80 /* Disk backed vcache definitions
81 * Both protected by xvcache */
82 static int afs_nextVcacheSlot = 0;
83 static struct afs_slotlist *afs_freeSlotList = NULL;
85 /* Forward declarations */
86 static afs_int32 afs_QueueVCB(struct vcache *avc, int *slept);
89 * Generate an index into the hash table for a given Fid.
91 * \return The hash value.
94 afs_HashCBRFid(struct AFSFid *fid)
96 return (fid->Volume + fid->Vnode + fid->Unique) % CBRSIZE;
100 * Insert a CBR entry into the hash table.
101 * Must be called with afs_xvcb held.
106 afs_InsertHashCBR(struct afs_cbr *cbr)
108 int slot = afs_HashCBRFid(&cbr->fid);
110 cbr->hash_next = afs_cbrHashT[slot];
111 if (afs_cbrHashT[slot])
112 afs_cbrHashT[slot]->hash_pprev = &cbr->hash_next;
114 cbr->hash_pprev = &afs_cbrHashT[slot];
115 afs_cbrHashT[slot] = cbr;
120 * Flush the given vcache entry.
123 * afs_xvcache lock must be held for writing upon entry to
124 * prevent people from changing the vrefCount field, and to
125 * protect the lruq and hnext fields.
126 * LOCK: afs_FlushVCache afs_xvcache W
127 * REFCNT: vcache ref count must be zero on entry except for osf1
128 * RACE: lock is dropped and reobtained, permitting race in caller
130 * \param avc Pointer to vcache entry to flush.
131 * \param slept Pointer to int to set 1 if we sleep/drop locks, 0 if we don't.
135 afs_FlushVCache(struct vcache *avc, int *slept)
136 { /*afs_FlushVCache */
139 struct vcache **uvc, *wvc;
142 AFS_STATCNT(afs_FlushVCache);
143 afs_Trace2(afs_iclSetp, CM_TRACE_FLUSHV, ICL_TYPE_POINTER, avc,
144 ICL_TYPE_INT32, avc->f.states);
146 code = osi_VM_FlushVCache(avc, slept);
150 if (avc->f.states & CVFlushed) {
154 #if !defined(AFS_LINUX22_ENV)
155 if (avc->nextfree || !avc->vlruq.prev || !avc->vlruq.next) { /* qv afs.h */
156 refpanic("LRU vs. Free inconsistency");
159 avc->f.states |= CVFlushed;
160 /* pull the entry out of the lruq and put it on the free list */
161 QRemove(&avc->vlruq);
163 /* keep track of # of files that we bulk stat'd, but never used
164 * before they got recycled.
166 if (avc->f.states & CBulkStat)
169 /* remove entry from the hash chain */
170 i = VCHash(&avc->f.fid);
171 uvc = &afs_vhashT[i];
172 for (wvc = *uvc; wvc; uvc = &wvc->hnext, wvc = *uvc) {
180 /* remove entry from the volume hash table */
181 QRemove(&avc->vhashq);
184 osi_FreeSmallSpace(avc->mvid);
185 avc->mvid = (struct VenusFid *)0;
187 afs_osi_Free(avc->linkData, strlen(avc->linkData) + 1);
188 avc->linkData = NULL;
190 #if defined(AFS_XBSD_ENV) || defined(AFS_DARWIN_ENV)
191 /* OK, there are no internal vrefCounts, so there shouldn't
192 * be any more refs here. */
194 #ifdef AFS_DARWIN80_ENV
195 vnode_clearfsnode(AFSTOV(avc));
196 vnode_removefsref(AFSTOV(avc));
198 avc->v->v_data = NULL; /* remove from vnode */
200 AFSTOV(avc) = NULL; /* also drop the ptr to vnode */
203 #ifdef AFS_SUN510_ENV
204 /* As we use private vnodes, cleanup is up to us */
205 vn_reinit(AFSTOV(avc));
207 afs_FreeAllAxs(&(avc->Access));
208 ObtainWriteLock(&afs_xcbhash, 460);
209 afs_DequeueCallback(avc); /* remove it from queued callbacks list */
210 avc->f.states &= ~(CStatd | CUnique);
211 ReleaseWriteLock(&afs_xcbhash);
212 if ((avc->f.states & CForeign) || (avc->f.fid.Fid.Vnode & 1))
213 osi_dnlc_purgedp(avc); /* if it (could be) a directory */
215 osi_dnlc_purgevp(avc);
217 if (!afs_shuttingdown)
218 afs_QueueVCB(avc, slept);
221 * Next, keep track of which vnodes we've deleted for create's
222 * optimistic synchronization algorithm
225 if (avc->f.fid.Fid.Vnode & 1)
231 #if !defined(AFS_LINUX22_ENV)
232 /* put the entry in the free list */
233 avc->nextfree = freeVCList;
235 if (avc->vlruq.prev || avc->vlruq.next) {
236 refpanic("LRU vs. Free inconsistency");
238 avc->f.states |= CVFlushed;
240 /* This should put it back on the vnode free list since usecount is 1 */
242 if (VREFCOUNT_GT(avc,0)) {
243 AFS_RELE(AFSTOV(avc));
244 afs_stats_cmperf.vcacheXAllocs--;
246 if (afs_norefpanic) {
247 afs_warn("flush vc refcnt < 1");
250 osi_Panic("flush vc refcnt < 1");
252 #endif /* AFS_LINUX22_ENV */
257 } /*afs_FlushVCache */
261 * The core of the inactive vnode op for all but IRIX.
267 afs_InactiveVCache(struct vcache *avc, afs_ucred_t *acred)
269 AFS_STATCNT(afs_inactive);
270 if (avc->f.states & CDirty) {
271 /* we can't keep trying to push back dirty data forever. Give up. */
272 afs_InvalidateAllSegments(avc); /* turns off dirty bit */
274 avc->f.states &= ~CMAPPED; /* mainly used by SunOS 4.0.x */
275 avc->f.states &= ~CDirty; /* Turn it off */
276 if (avc->f.states & CUnlinked) {
277 if (CheckLock(&afs_xvcache) || CheckLock(&afs_xdcache)) {
278 avc->f.states |= CUnlinkedDel;
281 afs_remunlink(avc, 1); /* ignore any return code */
288 * Allocate a callback return structure from the
289 * free list and return it.
291 * Environment: The alloc and free routines are both called with the afs_xvcb lock
292 * held, so we don't have to worry about blocking in osi_Alloc.
294 * \return The allocated afs_cbr.
296 static struct afs_cbr *afs_cbrSpace = 0;
297 /* if alloc limit below changes, fix me! */
298 static struct afs_cbr *afs_cbrHeads[16];
305 while (!afs_cbrSpace) {
306 if (afs_stats_cmperf.CallBackAlloced >= sizeof(afs_cbrHeads)/sizeof(afs_cbrHeads[0])) {
307 /* don't allocate more than 16 * AFS_NCBRS for now */
309 afs_stats_cmperf.CallBackFlushes++;
312 tsp = afs_osi_Alloc(AFS_NCBRS * sizeof(struct afs_cbr));
313 osi_Assert(tsp != NULL);
314 for (i = 0; i < AFS_NCBRS - 1; i++) {
315 tsp[i].next = &tsp[i + 1];
317 tsp[AFS_NCBRS - 1].next = 0;
319 afs_cbrHeads[afs_stats_cmperf.CallBackAlloced] = tsp;
320 afs_stats_cmperf.CallBackAlloced++;
324 afs_cbrSpace = tsp->next;
329 * Free a callback return structure, removing it from all lists.
331 * Environment: the xvcb lock is held over these calls.
333 * \param asp The address of the structure to free.
338 afs_FreeCBR(struct afs_cbr *asp)
340 *(asp->pprev) = asp->next;
342 asp->next->pprev = asp->pprev;
344 *(asp->hash_pprev) = asp->hash_next;
346 asp->hash_next->hash_pprev = asp->hash_pprev;
348 asp->next = afs_cbrSpace;
354 FlushAllVCBs(int nconns, struct rx_connection **rxconns,
355 struct afs_conn **conns)
360 results = afs_osi_Alloc(nconns * sizeof (afs_int32));
361 osi_Assert(results != NULL);
364 multi_Rx(rxconns,nconns)
366 multi_RXAFS_GiveUpAllCallBacks();
367 results[multi_i] = multi_error;
372 * Freeing the CBR will unlink it from the server's CBR list
373 * do it here, not in the loop, because a dynamic CBR will call
374 * into the memory management routines.
376 for ( i = 0 ; i < nconns ; i++ ) {
377 if (results[i] == 0) {
378 /* Unchain all of them */
379 while (conns[i]->parent->srvr->server->cbrs)
380 afs_FreeCBR(conns[i]->parent->srvr->server->cbrs);
383 afs_osi_Free(results, nconns * sizeof(afs_int32));
387 * Flush all queued callbacks to all servers.
389 * Environment: holds xvcb lock over RPC to guard against race conditions
390 * when a new callback is granted for the same file later on.
392 * \return 0 for success.
395 afs_FlushVCBs(afs_int32 lockit)
397 struct AFSFid *tfids;
398 struct AFSCallBack callBacks[1];
399 struct AFSCBFids fidArray;
400 struct AFSCBs cbArray;
402 struct afs_cbr *tcbrp;
406 struct vrequest treq;
408 int safety1, safety2, safety3;
411 if (AFS_IS_DISCONNECTED)
414 if ((code = afs_InitReq(&treq, afs_osi_credp)))
416 treq.flags |= O_NONBLOCK;
417 tfids = afs_osi_Alloc(sizeof(struct AFSFid) * AFS_MAXCBRSCALL);
418 osi_Assert(tfids != NULL);
421 ObtainWriteLock(&afs_xvcb, 273);
424 * First, attempt a multi across everything, all addresses
425 * for all servers we know of.
429 afs_LoopServers(AFS_LS_ALL, NULL, 0, FlushAllVCBs, NULL);
431 ObtainReadLock(&afs_xserver);
432 for (i = 0; i < NSERVERS; i++) {
433 for (safety1 = 0, tsp = afs_servers[i];
434 tsp && safety1 < afs_totalServers + 10;
435 tsp = tsp->next, safety1++) {
437 if (tsp->cbrs == (struct afs_cbr *)0)
440 /* otherwise, grab a block of AFS_MAXCBRSCALL from the list
441 * and make an RPC, over and over again.
443 tcount = 0; /* number found so far */
444 for (safety2 = 0; safety2 < afs_cacheStats; safety2++) {
445 if (tcount >= AFS_MAXCBRSCALL || !tsp->cbrs) {
446 struct rx_connection *rxconn;
447 /* if buffer is full, or we've queued all we're going
448 * to from this server, we should flush out the
451 fidArray.AFSCBFids_len = tcount;
452 fidArray.AFSCBFids_val = (struct AFSFid *)tfids;
453 cbArray.AFSCBs_len = 1;
454 cbArray.AFSCBs_val = callBacks;
455 memset(&callBacks[0], 0, sizeof(callBacks[0]));
456 callBacks[0].CallBackType = CB_EXCLUSIVE;
457 for (safety3 = 0; safety3 < AFS_MAXHOSTS * 2; safety3++) {
458 tc = afs_ConnByHost(tsp, tsp->cell->fsport,
459 tsp->cell->cellNum, &treq, 0,
460 SHARED_LOCK, 0, &rxconn);
463 (AFS_STATS_FS_RPCIDX_GIVEUPCALLBACKS);
466 RXAFS_GiveUpCallBacks(rxconn, &fidArray,
473 (tc, rxconn, code, 0, &treq,
474 AFS_STATS_FS_RPCIDX_GIVEUPCALLBACKS, SHARED_LOCK,
479 /* ignore return code, since callbacks may have
480 * been returned anyway, we shouldn't leave them
481 * around to be returned again.
483 * Next, see if we are done with this server, and if so,
484 * break to deal with the next one.
490 /* if to flush full buffer */
491 /* if we make it here, we have an entry at the head of cbrs,
492 * which we should copy to the file ID array and then free.
495 tfids[tcount++] = tcbrp->fid;
497 /* Freeing the CBR will unlink it from the server's CBR list */
499 } /* while loop for this one server */
500 if (safety2 > afs_cacheStats) {
501 afs_warn("possible internal error afs_flushVCBs (%d)\n",
504 } /* for loop for this hash chain */
505 } /* loop through all hash chains */
506 if (safety1 > afs_totalServers + 2) {
508 ("AFS internal error (afs_flushVCBs) (%d > %d), continuing...\n",
509 safety1, afs_totalServers + 2);
511 osi_Panic("afs_flushVCBS safety1");
514 ReleaseReadLock(&afs_xserver);
516 ReleaseWriteLock(&afs_xvcb);
517 afs_osi_Free(tfids, sizeof(struct AFSFid) * AFS_MAXCBRSCALL);
522 * Queue a callback on the given fid.
525 * Locks the xvcb lock.
526 * Called when the xvcache lock is already held.
527 * RACE: afs_xvcache may be dropped and reacquired
529 * \param avc vcache entry
530 * \param slep Set to 1 if we dropped afs_xvcache
531 * \return 1 if queued, 0 otherwise
535 afs_QueueVCB(struct vcache *avc, int *slept)
539 struct afs_cbr *tcbp;
542 AFS_STATCNT(afs_QueueVCB);
544 ObtainWriteLock(&afs_xvcb, 274);
546 /* we can't really give back callbacks on RO files, since the
547 * server only tracks them on a per-volume basis, and we don't
548 * know whether we still have some other files from the same
550 if (!((avc->f.states & CRO) == 0 && avc->callback)) {
554 /* The callback is really just a struct server ptr. */
555 tsp = (struct server *)(avc->callback);
558 /* If we don't have CBR space, AllocCBR may block or hit the net for
559 * clearing up CBRs. Hitting the net may involve a fileserver
560 * needing to contact us, so we must drop xvcache so we don't block
561 * those requests from going through. */
562 reacquire = *slept = 1;
563 ReleaseWriteLock(&afs_xvcache);
566 /* we now have a pointer to the server, so we just allocate
567 * a queue entry and queue it.
569 tcbp = afs_AllocCBR();
570 tcbp->fid = avc->f.fid.Fid;
572 tcbp->next = tsp->cbrs;
574 tsp->cbrs->pprev = &tcbp->next;
577 tcbp->pprev = &tsp->cbrs;
579 afs_InsertHashCBR(tcbp);
583 /* now release locks and return */
584 ReleaseWriteLock(&afs_xvcb);
587 /* make sure this is after dropping xvcb, for locking order */
588 ObtainWriteLock(&afs_xvcache, 279);
595 * Remove a queued callback for a given Fid.
598 * Locks xvcb and xserver locks.
599 * Typically called with xdcache, xvcache and/or individual vcache
602 * \param afid The fid we want cleansed of queued callbacks.
607 afs_RemoveVCB(struct VenusFid *afid)
610 struct afs_cbr *cbr, *ncbr;
612 AFS_STATCNT(afs_RemoveVCB);
613 ObtainWriteLock(&afs_xvcb, 275);
615 slot = afs_HashCBRFid(&afid->Fid);
616 ncbr = afs_cbrHashT[slot];
620 ncbr = cbr->hash_next;
622 if (afid->Fid.Volume == cbr->fid.Volume &&
623 afid->Fid.Vnode == cbr->fid.Vnode &&
624 afid->Fid.Unique == cbr->fid.Unique) {
629 ReleaseWriteLock(&afs_xvcb);
633 afs_FlushReclaimedVcaches(void)
635 #if !defined(AFS_LINUX22_ENV)
638 struct vcache *tmpReclaimedVCList = NULL;
640 ObtainWriteLock(&afs_xvreclaim, 76);
641 while (ReclaimedVCList) {
642 tvc = ReclaimedVCList; /* take from free list */
643 ReclaimedVCList = tvc->nextfree;
644 tvc->nextfree = NULL;
645 code = afs_FlushVCache(tvc, &fv_slept);
647 /* Ok, so, if we got code != 0, uh, wtf do we do? */
648 /* Probably, build a temporary list and then put all back when we
649 get to the end of the list */
650 /* This is actually really crappy, but we need to not leak these.
651 We probably need a way to be smarter about this. */
652 tvc->nextfree = tmpReclaimedVCList;
653 tmpReclaimedVCList = tvc;
654 /* printf("Reclaim list flush %lx failed: %d\n", (unsigned long) tvc, code); */
656 if (tvc->f.states & (CVInit
657 #ifdef AFS_DARWIN80_ENV
661 tvc->f.states &= ~(CVInit
662 #ifdef AFS_DARWIN80_ENV
666 afs_osi_Wakeup(&tvc->f.states);
669 if (tmpReclaimedVCList)
670 ReclaimedVCList = tmpReclaimedVCList;
672 ReleaseWriteLock(&afs_xvreclaim);
677 afs_PostPopulateVCache(struct vcache *avc, struct VenusFid *afid, int seq)
680 * The proper value for mvstat (for root fids) is setup by the caller.
683 if (afid->Fid.Vnode == 1 && afid->Fid.Unique == 1)
686 if (afs_globalVFS == 0)
687 osi_Panic("afs globalvfs");
689 osi_PostPopulateVCache(avc);
692 osi_dnlc_purgedp(avc); /* this may be overkill */
693 memset(&(avc->callsort), 0, sizeof(struct afs_q));
695 avc->f.states &=~ CVInit;
697 avc->f.states |= CBulkFetching;
698 avc->f.m.Length = seq;
700 afs_osi_Wakeup(&avc->f.states);
704 afs_ShakeLooseVCaches(afs_int32 anumber)
708 struct afs_q *tq, *uq;
709 int fv_slept, defersleep = 0;
711 afs_int32 target = anumber;
718 for (tq = VLRU.prev; tq != &VLRU && anumber > 0; tq = uq) {
721 if (tvc->f.states & CVFlushed) {
722 refpanic("CVFlushed on VLRU");
723 } else if (i++ > limit) {
724 afs_warn("afs_ShakeLooseVCaches: i %d limit %d afs_vcount %d afs_maxvcount %d\n",
725 (int)i, limit, (int)afs_vcount, (int)afs_maxvcount);
726 refpanic("Found too many AFS vnodes on VLRU (VLRU cycle?)");
727 } else if (QNext(uq) != tq) {
728 refpanic("VLRU inconsistent");
729 } else if (tvc->f.states & CVInit) {
734 if (osi_TryEvictVCache(tvc, &fv_slept, defersleep))
740 goto retry; /* start over - may have raced. */
743 if (anumber && !defersleep) {
750 if (!afsd_dynamic_vcaches && anumber == target) {
751 afs_warn("afs_ShakeLooseVCaches: warning none freed, using %d of %d\n",
752 afs_vcount, afs_maxvcount);
758 /* Alloc new vnode. */
760 static struct vcache *
761 afs_AllocVCache(void)
765 tvc = osi_NewVnode();
770 if (afsd_dynamic_vcaches && afs_maxvcount < afs_vcount) {
771 afs_maxvcount = afs_vcount;
772 /*printf("peak vnodes: %d\n", afs_maxvcount);*/
775 afs_stats_cmperf.vcacheXAllocs++; /* count in case we have a leak */
777 /* If we create a new inode, we either give it a new slot number,
778 * or if one's available, use a slot number from the slot free list
780 if (afs_freeSlotList != NULL) {
781 struct afs_slotlist *tmp;
783 tvc->diskSlot = afs_freeSlotList->slot;
784 tmp = afs_freeSlotList;
785 afs_freeSlotList = tmp->next;
786 afs_osi_Free(tmp, sizeof(struct afs_slotlist));
788 tvc->diskSlot = afs_nextVcacheSlot++;
794 /* Pre populate a newly allocated vcache. On platforms where the actual
795 * vnode is attached to the vcache, this function is called before attachment,
796 * therefore it cannot perform any actions on the vnode itself */
799 afs_PrePopulateVCache(struct vcache *avc, struct VenusFid *afid,
800 struct server *serverp) {
803 slot = avc->diskSlot;
805 osi_PrePopulateVCache(avc);
807 avc->diskSlot = slot;
808 QZero(&avc->metadirty);
810 AFS_RWLOCK_INIT(&avc->lock, "vcache lock");
813 avc->linkData = NULL;
816 avc->execsOrWriters = 0;
818 avc->f.states = CVInit;
819 avc->last_looker = 0;
821 avc->asynchrony = -1;
825 avc->f.truncPos = AFS_NOTRUNC; /* don't truncate until we need to */
826 hzero(avc->f.m.DataVersion); /* in case we copy it into flushDV */
828 avc->callback = serverp; /* to minimize chance that clear
831 #if defined(AFS_CACHE_BYPASS)
832 avc->cachingStates = 0;
833 avc->cachingTransitions = 0;
838 afs_FlushAllVCaches(void)
841 struct vcache *tvc, *nvc;
843 ObtainWriteLock(&afs_xvcache, 867);
846 for (i = 0; i < VCSIZE; i++) {
847 for (tvc = afs_vhashT[i]; tvc; tvc = nvc) {
851 if (afs_FlushVCache(tvc, &slept)) {
852 afs_warn("Failed to flush vcache 0x%lx\n", (unsigned long)(uintptrsz)tvc);
860 ReleaseWriteLock(&afs_xvcache);
864 * This routine is responsible for allocating a new cache entry
865 * from the free list. It formats the cache entry and inserts it
866 * into the appropriate hash tables. It must be called with
867 * afs_xvcache write-locked so as to prevent several processes from
868 * trying to create a new cache entry simultaneously.
870 * LOCK: afs_NewVCache afs_xvcache W
872 * \param afid The file id of the file whose cache entry is being created.
874 * \return The new vcache struct.
877 static_inline struct vcache *
878 afs_NewVCache_int(struct VenusFid *afid, struct server *serverp, int seq)
882 afs_int32 anumber = VCACHE_FREE;
884 AFS_STATCNT(afs_NewVCache);
886 afs_FlushReclaimedVcaches();
888 #if defined(AFS_LINUX22_ENV)
889 if(!afsd_dynamic_vcaches && afs_vcount >= afs_maxvcount) {
890 afs_ShakeLooseVCaches(anumber);
891 if (afs_vcount >= afs_maxvcount) {
892 afs_warn("afs_NewVCache - none freed\n");
896 tvc = afs_AllocVCache();
897 #else /* AFS_LINUX22_ENV */
898 /* pull out a free cache entry */
900 afs_ShakeLooseVCaches(anumber);
904 tvc = afs_AllocVCache();
906 tvc = freeVCList; /* take from free list */
907 freeVCList = tvc->nextfree;
908 tvc->nextfree = NULL;
909 afs_vcount++; /* balanced by FlushVCache */
910 } /* end of if (!freeVCList) */
912 #endif /* AFS_LINUX22_ENV */
914 #if defined(AFS_XBSD_ENV) || defined(AFS_DARWIN_ENV)
916 panic("afs_NewVCache(): free vcache with vnode attached");
919 /* Populate the vcache with as much as we can. */
920 afs_PrePopulateVCache(tvc, afid, serverp);
922 /* Thread the vcache onto the VLRU */
927 tvc->hnext = afs_vhashT[i];
929 QAdd(&afs_vhashTV[j], &tvc->vhashq);
931 if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) {
932 refpanic("NewVCache VLRU inconsistent");
934 QAdd(&VLRU, &tvc->vlruq); /* put in lruq */
935 if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) {
936 refpanic("NewVCache VLRU inconsistent2");
938 if (tvc->vlruq.next->prev != &(tvc->vlruq)) {
939 refpanic("NewVCache VLRU inconsistent3");
941 if (tvc->vlruq.prev->next != &(tvc->vlruq)) {
942 refpanic("NewVCache VLRU inconsistent4");
946 /* it should now be safe to drop the xvcache lock - so attach an inode
947 * to this vcache, where necessary */
948 osi_AttachVnode(tvc, seq);
950 /* Get a reference count to hold this vcache for the VLRUQ. Note that
951 * we have to do this after attaching the vnode, because the reference
952 * count may be held in the vnode itself */
954 #if defined(AFS_LINUX22_ENV)
955 /* Hold it for the LRU (should make count 2) */
957 #elif !(defined (AFS_DARWIN_ENV) || defined(AFS_XBSD_ENV))
958 VREFCOUNT_SET(tvc, 1); /* us */
961 #if defined (AFS_FBSD_ENV)
962 if (tvc->f.states & CVInit)
964 afs_PostPopulateVCache(tvc, afid, seq);
971 afs_NewVCache(struct VenusFid *afid, struct server *serverp)
973 return afs_NewVCache_int(afid, serverp, 0);
977 afs_NewBulkVCache(struct VenusFid *afid, struct server *serverp, int seq)
979 return afs_NewVCache_int(afid, serverp, seq);
985 * LOCK: afs_FlushActiveVcaches afs_xvcache N
987 * \param doflocks : Do we handle flocks?
990 afs_FlushActiveVcaches(afs_int32 doflocks)
996 afs_ucred_t *cred = NULL;
997 struct vrequest treq, ureq;
998 struct AFSVolSync tsync;
1001 AFS_STATCNT(afs_FlushActiveVcaches);
1002 ObtainReadLock(&afs_xvcache);
1003 for (i = 0; i < VCSIZE; i++) {
1004 for (tvc = afs_vhashT[i]; tvc; tvc = tvc->hnext) {
1005 if (tvc->f.states & CVInit) continue;
1006 #ifdef AFS_DARWIN80_ENV
1007 if (tvc->f.states & CDeadVnode &&
1008 (tvc->f.states & (CCore|CUnlinkedDel) ||
1009 tvc->flockCount)) panic("Dead vnode has core/unlinkedel/flock");
1011 if (doflocks && tvc->flockCount != 0) {
1012 struct rx_connection *rxconn;
1013 /* if this entry has an flock, send a keep-alive call out */
1015 ReleaseReadLock(&afs_xvcache);
1016 ObtainWriteLock(&tvc->lock, 51);
1018 afs_InitReq(&treq, afs_osi_credp);
1019 treq.flags |= O_NONBLOCK;
1021 tc = afs_Conn(&tvc->f.fid, &treq, SHARED_LOCK, &rxconn);
1023 XSTATS_START_TIME(AFS_STATS_FS_RPCIDX_EXTENDLOCK);
1026 RXAFS_ExtendLock(rxconn,
1027 (struct AFSFid *)&tvc->f.fid.Fid,
1033 } while (afs_Analyze
1034 (tc, rxconn, code, &tvc->f.fid, &treq,
1035 AFS_STATS_FS_RPCIDX_EXTENDLOCK, SHARED_LOCK, NULL));
1037 ReleaseWriteLock(&tvc->lock);
1038 #ifdef AFS_DARWIN80_ENV
1040 ObtainReadLock(&afs_xvcache);
1042 ObtainReadLock(&afs_xvcache);
1047 if ((tvc->f.states & CCore) || (tvc->f.states & CUnlinkedDel)) {
1049 * Don't let it evaporate in case someone else is in
1050 * this code. Also, drop the afs_xvcache lock while
1051 * getting vcache locks.
1054 ReleaseReadLock(&afs_xvcache);
1055 #ifdef AFS_BOZONLOCK_ENV
1056 afs_BozonLock(&tvc->pvnLock, tvc);
1058 #if defined(AFS_SGI_ENV)
1060 * That's because if we come in via the CUnlinkedDel bit state path we'll be have 0 refcnt
1062 osi_Assert(VREFCOUNT_GT(tvc,0));
1063 AFS_RWLOCK((vnode_t *) tvc, VRWLOCK_WRITE);
1065 ObtainWriteLock(&tvc->lock, 52);
1066 if (tvc->f.states & CCore) {
1067 tvc->f.states &= ~CCore;
1068 /* XXXX Find better place-holder for cred XXXX */
1069 cred = (afs_ucred_t *)tvc->linkData;
1070 tvc->linkData = NULL; /* XXX */
1071 afs_InitReq(&ureq, cred);
1072 afs_Trace2(afs_iclSetp, CM_TRACE_ACTCCORE,
1073 ICL_TYPE_POINTER, tvc, ICL_TYPE_INT32,
1074 tvc->execsOrWriters);
1075 code = afs_StoreOnLastReference(tvc, &ureq);
1076 ReleaseWriteLock(&tvc->lock);
1077 #ifdef AFS_BOZONLOCK_ENV
1078 afs_BozonUnlock(&tvc->pvnLock, tvc);
1080 hzero(tvc->flushDV);
1083 if (code && code != VNOVNODE) {
1084 afs_StoreWarn(code, tvc->f.fid.Fid.Volume,
1085 /* /dev/console */ 1);
1087 } else if (tvc->f.states & CUnlinkedDel) {
1091 ReleaseWriteLock(&tvc->lock);
1092 #ifdef AFS_BOZONLOCK_ENV
1093 afs_BozonUnlock(&tvc->pvnLock, tvc);
1095 #if defined(AFS_SGI_ENV)
1096 AFS_RWUNLOCK((vnode_t *) tvc, VRWLOCK_WRITE);
1098 afs_remunlink(tvc, 0);
1099 #if defined(AFS_SGI_ENV)
1100 AFS_RWLOCK((vnode_t *) tvc, VRWLOCK_WRITE);
1103 /* lost (or won, perhaps) the race condition */
1104 ReleaseWriteLock(&tvc->lock);
1105 #ifdef AFS_BOZONLOCK_ENV
1106 afs_BozonUnlock(&tvc->pvnLock, tvc);
1109 #if defined(AFS_SGI_ENV)
1110 AFS_RWUNLOCK((vnode_t *) tvc, VRWLOCK_WRITE);
1112 #ifdef AFS_DARWIN80_ENV
1115 AFS_RELE(AFSTOV(tvc));
1116 /* Matches write code setting CCore flag */
1119 ObtainReadLock(&afs_xvcache);
1121 ObtainReadLock(&afs_xvcache);
1124 AFS_RELE(AFSTOV(tvc));
1125 /* Matches write code setting CCore flag */
1132 ReleaseReadLock(&afs_xvcache);
1138 * Make sure a cache entry is up-to-date status-wise.
1140 * NOTE: everywhere that calls this can potentially be sped up
1141 * by checking CStatd first, and avoiding doing the InitReq
1142 * if this is up-to-date.
1144 * Anymore, the only places that call this KNOW already that the
1145 * vcache is not up-to-date, so we don't screw around.
1147 * \param avc : Ptr to vcache entry to verify.
1153 * Make sure a cache entry is up-to-date status-wise.
1155 * NOTE: everywhere that calls this can potentially be sped up
1156 * by checking CStatd first, and avoiding doing the InitReq
1157 * if this is up-to-date.
1159 * Anymore, the only places that call this KNOW already that the
1160 * vcache is not up-to-date, so we don't screw around.
1162 * \param avc Pointer to vcache entry to verify.
1165 * \return 0 for success or other error codes.
1168 afs_VerifyVCache2(struct vcache *avc, struct vrequest *areq)
1172 AFS_STATCNT(afs_VerifyVCache);
1174 /* otherwise we must fetch the status info */
1176 ObtainWriteLock(&avc->lock, 53);
1177 if (avc->f.states & CStatd) {
1178 ReleaseWriteLock(&avc->lock);
1181 ObtainWriteLock(&afs_xcbhash, 461);
1182 avc->f.states &= ~(CStatd | CUnique);
1183 avc->callback = NULL;
1184 afs_DequeueCallback(avc);
1185 ReleaseWriteLock(&afs_xcbhash);
1186 ReleaseWriteLock(&avc->lock);
1188 /* since we've been called back, or the callback has expired,
1189 * it's possible that the contents of this directory, or this
1190 * file's name have changed, thus invalidating the dnlc contents.
1192 if ((avc->f.states & CForeign) || (avc->f.fid.Fid.Vnode & 1))
1193 osi_dnlc_purgedp(avc);
1195 osi_dnlc_purgevp(avc);
1197 /* fetch the status info */
1198 tvc = afs_GetVCache(&avc->f.fid, areq, NULL, avc);
1201 /* Put it back; caller has already incremented vrefCount */
1205 } /*afs_VerifyVCache */
1209 * Simple copy of stat info into cache.
1211 * Callers:as of 1992-04-29, only called by WriteVCache
1213 * \param avc Ptr to vcache entry involved.
1214 * \param astat Ptr to stat info to copy.
1218 afs_SimpleVStat(struct vcache *avc,
1219 struct AFSFetchStatus *astat, struct vrequest *areq)
1222 AFS_STATCNT(afs_SimpleVStat);
1224 #ifdef AFS_64BIT_CLIENT
1225 FillInt64(length, astat->Length_hi, astat->Length);
1226 #else /* AFS_64BIT_CLIENT */
1227 length = astat->Length;
1228 #endif /* AFS_64BIT_CLIENT */
1230 #if defined(AFS_SGI_ENV)
1231 if ((avc->execsOrWriters <= 0) && !afs_DirtyPages(avc)
1232 && !AFS_VN_MAPPED((vnode_t *) avc)) {
1233 osi_Assert((valusema(&avc->vc_rwlock) <= 0)
1234 && (OSI_GET_LOCKID() == avc->vc_rwlockid));
1235 if (length < avc->f.m.Length) {
1236 vnode_t *vp = (vnode_t *) avc;
1238 osi_Assert(WriteLocked(&avc->lock));
1239 ReleaseWriteLock(&avc->lock);
1241 PTOSSVP(vp, (off_t) length, (off_t) MAXLONG);
1243 ObtainWriteLock(&avc->lock, 67);
1248 if (!afs_DirtyPages(avc)) {
1249 /* if actively writing the file, don't fetch over this value */
1250 afs_Trace3(afs_iclSetp, CM_TRACE_SIMPLEVSTAT, ICL_TYPE_POINTER, avc,
1251 ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(avc->f.m.Length),
1252 ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(length));
1253 avc->f.m.Length = length;
1254 avc->f.m.Date = astat->ClientModTime;
1256 avc->f.m.Owner = astat->Owner;
1257 avc->f.m.Group = astat->Group;
1258 avc->f.m.Mode = astat->UnixModeBits;
1259 if (vType(avc) == VREG) {
1260 avc->f.m.Mode |= S_IFREG;
1261 } else if (vType(avc) == VDIR) {
1262 avc->f.m.Mode |= S_IFDIR;
1263 } else if (vType(avc) == VLNK) {
1264 avc->f.m.Mode |= S_IFLNK;
1265 if ((avc->f.m.Mode & 0111) == 0)
1268 if (avc->f.states & CForeign) {
1269 struct axscache *ac;
1270 avc->f.anyAccess = astat->AnonymousAccess;
1272 if ((astat->CallerAccess & ~astat->AnonymousAccess))
1274 * Caller has at least one bit not covered by anonymous, and
1275 * thus may have interesting rights.
1277 * HOWEVER, this is a really bad idea, because any access query
1278 * for bits which aren't covered by anonymous, on behalf of a user
1279 * who doesn't have any special rights, will result in an answer of
1280 * the form "I don't know, lets make a FetchStatus RPC and find out!"
1281 * It's an especially bad idea under Ultrix, since (due to the lack of
1282 * a proper access() call) it must perform several afs_access() calls
1283 * in order to create magic mode bits that vary according to who makes
1284 * the call. In other words, _every_ stat() generates a test for
1287 #endif /* badidea */
1288 if (avc->Access && (ac = afs_FindAxs(avc->Access, areq->uid)))
1289 ac->axess = astat->CallerAccess;
1290 else /* not found, add a new one if possible */
1291 afs_AddAxs(avc->Access, areq->uid, astat->CallerAccess);
1294 } /*afs_SimpleVStat */
1298 * Store the status info *only* back to the server for a
1301 * Environment: Must be called with a shared lock held on the vnode.
1303 * \param avc Ptr to the vcache entry.
1304 * \param astatus Ptr to the status info to store.
1305 * \param areq Ptr to the associated vrequest.
1307 * \return Operation status.
1311 afs_WriteVCache(struct vcache *avc,
1312 struct AFSStoreStatus *astatus,
1313 struct vrequest *areq)
1316 struct afs_conn *tc;
1317 struct AFSFetchStatus OutStatus;
1318 struct AFSVolSync tsync;
1319 struct rx_connection *rxconn;
1321 AFS_STATCNT(afs_WriteVCache);
1322 afs_Trace2(afs_iclSetp, CM_TRACE_WVCACHE, ICL_TYPE_POINTER, avc,
1323 ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(avc->f.m.Length));
1325 tc = afs_Conn(&avc->f.fid, areq, SHARED_LOCK, &rxconn);
1327 XSTATS_START_TIME(AFS_STATS_FS_RPCIDX_STORESTATUS);
1330 RXAFS_StoreStatus(rxconn, (struct AFSFid *)&avc->f.fid.Fid,
1331 astatus, &OutStatus, &tsync);
1336 } while (afs_Analyze
1337 (tc, rxconn, code, &avc->f.fid, areq, AFS_STATS_FS_RPCIDX_STORESTATUS,
1338 SHARED_LOCK, NULL));
1340 UpgradeSToWLock(&avc->lock, 20);
1342 /* success, do the changes locally */
1343 afs_SimpleVStat(avc, &OutStatus, areq);
1345 * Update the date, too. SimpleVStat didn't do this, since
1346 * it thought we were doing this after fetching new status
1347 * over a file being written.
1349 avc->f.m.Date = OutStatus.ClientModTime;
1351 /* failure, set up to check with server next time */
1352 ObtainWriteLock(&afs_xcbhash, 462);
1353 afs_DequeueCallback(avc);
1354 avc->f.states &= ~(CStatd | CUnique); /* turn off stat valid flag */
1355 ReleaseWriteLock(&afs_xcbhash);
1356 if ((avc->f.states & CForeign) || (avc->f.fid.Fid.Vnode & 1))
1357 osi_dnlc_purgedp(avc); /* if it (could be) a directory */
1359 ConvertWToSLock(&avc->lock);
1362 } /*afs_WriteVCache */
1365 * Store status info only locally, set the proper disconnection flags
1366 * and add to dirty list.
1368 * \param avc The vcache to be written locally.
1369 * \param astatus Get attr fields from local store.
1370 * \param attrs This one is only of the vs_size.
1372 * \note Must be called with a shared lock on the vnode
1375 afs_WriteVCacheDiscon(struct vcache *avc,
1376 struct AFSStoreStatus *astatus,
1377 struct vattr *attrs)
1380 afs_int32 flags = 0;
1382 UpgradeSToWLock(&avc->lock, 700);
1384 if (!astatus->Mask) {
1390 /* Set attributes. */
1391 if (astatus->Mask & AFS_SETMODTIME) {
1392 avc->f.m.Date = astatus->ClientModTime;
1393 flags |= VDisconSetTime;
1396 if (astatus->Mask & AFS_SETOWNER) {
1397 /* printf("Not allowed yet. \n"); */
1398 /*avc->f.m.Owner = astatus->Owner;*/
1401 if (astatus->Mask & AFS_SETGROUP) {
1402 /* printf("Not allowed yet. \n"); */
1403 /*avc->f.m.Group = astatus->Group;*/
1406 if (astatus->Mask & AFS_SETMODE) {
1407 avc->f.m.Mode = astatus->UnixModeBits;
1409 #if 0 /* XXX: Leaving this out, so it doesn't mess up the file type flag.*/
1411 if (vType(avc) == VREG) {
1412 avc->f.m.Mode |= S_IFREG;
1413 } else if (vType(avc) == VDIR) {
1414 avc->f.m.Mode |= S_IFDIR;
1415 } else if (vType(avc) == VLNK) {
1416 avc->f.m.Mode |= S_IFLNK;
1417 if ((avc->f.m.Mode & 0111) == 0)
1421 flags |= VDisconSetMode;
1422 } /* if(astatus.Mask & AFS_SETMODE) */
1424 } /* if (!astatus->Mask) */
1426 if (attrs->va_size > 0) {
1427 /* XXX: Do I need more checks? */
1428 /* Truncation operation. */
1429 flags |= VDisconTrunc;
1433 afs_DisconAddDirty(avc, flags, 1);
1435 /* XXX: How about the rest of the fields? */
1437 ConvertWToSLock(&avc->lock);
1443 * Copy astat block into vcache info
1445 * \note This code may get dataversion and length out of sync if the file has
1446 * been modified. This is less than ideal. I haven't thought about it sufficiently
1447 * to be certain that it is adequate.
1449 * \note Environment: Must be called under a write lock
1451 * \param avc Ptr to vcache entry.
1452 * \param astat Ptr to stat block to copy in.
1453 * \param areq Ptr to associated request.
1456 afs_ProcessFS(struct vcache *avc,
1457 struct AFSFetchStatus *astat, struct vrequest *areq)
1460 AFS_STATCNT(afs_ProcessFS);
1462 #ifdef AFS_64BIT_CLIENT
1463 FillInt64(length, astat->Length_hi, astat->Length);
1464 #else /* AFS_64BIT_CLIENT */
1465 length = astat->Length;
1466 #endif /* AFS_64BIT_CLIENT */
1467 /* WARNING: afs_DoBulkStat uses the Length field to store a sequence
1468 * number for each bulk status request. Under no circumstances
1469 * should afs_DoBulkStat store a sequence number if the new
1470 * length will be ignored when afs_ProcessFS is called with
1471 * new stats. If you change the following conditional then you
1472 * also need to change the conditional in afs_DoBulkStat. */
1474 if ((avc->execsOrWriters <= 0) && !afs_DirtyPages(avc)
1475 && !AFS_VN_MAPPED((vnode_t *) avc)) {
1477 if ((avc->execsOrWriters <= 0) && !afs_DirtyPages(avc)) {
1479 /* if we're writing or mapping this file, don't fetch over these
1482 afs_Trace3(afs_iclSetp, CM_TRACE_PROCESSFS, ICL_TYPE_POINTER, avc,
1483 ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(avc->f.m.Length),
1484 ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(length));
1485 avc->f.m.Length = length;
1486 avc->f.m.Date = astat->ClientModTime;
1488 hset64(avc->f.m.DataVersion, astat->dataVersionHigh, astat->DataVersion);
1489 avc->f.m.Owner = astat->Owner;
1490 avc->f.m.Mode = astat->UnixModeBits;
1491 avc->f.m.Group = astat->Group;
1492 avc->f.m.LinkCount = astat->LinkCount;
1493 if (astat->FileType == File) {
1494 vSetType(avc, VREG);
1495 avc->f.m.Mode |= S_IFREG;
1496 } else if (astat->FileType == Directory) {
1497 vSetType(avc, VDIR);
1498 avc->f.m.Mode |= S_IFDIR;
1499 } else if (astat->FileType == SymbolicLink) {
1500 if (afs_fakestat_enable && (avc->f.m.Mode & 0111) == 0) {
1501 vSetType(avc, VDIR);
1502 avc->f.m.Mode |= S_IFDIR;
1504 vSetType(avc, VLNK);
1505 avc->f.m.Mode |= S_IFLNK;
1507 if ((avc->f.m.Mode & 0111) == 0) {
1511 avc->f.anyAccess = astat->AnonymousAccess;
1513 if ((astat->CallerAccess & ~astat->AnonymousAccess))
1515 * Caller has at least one bit not covered by anonymous, and
1516 * thus may have interesting rights.
1518 * HOWEVER, this is a really bad idea, because any access query
1519 * for bits which aren't covered by anonymous, on behalf of a user
1520 * who doesn't have any special rights, will result in an answer of
1521 * the form "I don't know, lets make a FetchStatus RPC and find out!"
1522 * It's an especially bad idea under Ultrix, since (due to the lack of
1523 * a proper access() call) it must perform several afs_access() calls
1524 * in order to create magic mode bits that vary according to who makes
1525 * the call. In other words, _every_ stat() generates a test for
1528 #endif /* badidea */
1530 struct axscache *ac;
1531 if (avc->Access && (ac = afs_FindAxs(avc->Access, areq->uid)))
1532 ac->axess = astat->CallerAccess;
1533 else /* not found, add a new one if possible */
1534 afs_AddAxs(avc->Access, areq->uid, astat->CallerAccess);
1536 } /*afs_ProcessFS */
1540 * Get fid from server.
1543 * \param areq Request to be passed on.
1544 * \param name Name of ?? to lookup.
1545 * \param OutStatus Fetch status.
1550 * \return Success status of operation.
1553 afs_RemoteLookup(struct VenusFid *afid, struct vrequest *areq,
1554 char *name, struct VenusFid *nfid,
1555 struct AFSFetchStatus *OutStatusp,
1556 struct AFSCallBack *CallBackp, struct server **serverp,
1557 struct AFSVolSync *tsyncp)
1560 struct afs_conn *tc;
1561 struct rx_connection *rxconn;
1562 struct AFSFetchStatus OutDirStatus;
1565 name = ""; /* XXX */
1567 tc = afs_Conn(afid, areq, SHARED_LOCK, &rxconn);
1570 *serverp = tc->parent->srvr->server;
1571 XSTATS_START_TIME(AFS_STATS_FS_RPCIDX_XLOOKUP);
1574 RXAFS_Lookup(rxconn, (struct AFSFid *)&afid->Fid, name,
1575 (struct AFSFid *)&nfid->Fid, OutStatusp,
1576 &OutDirStatus, CallBackp, tsyncp);
1581 } while (afs_Analyze
1582 (tc, rxconn, code, afid, areq, AFS_STATS_FS_RPCIDX_XLOOKUP, SHARED_LOCK,
1592 * Given a file id and a vrequest structure, fetch the status
1593 * information associated with the file.
1595 * \param afid File ID.
1596 * \param areq Ptr to associated vrequest structure, specifying the
1597 * user whose authentication tokens will be used.
1598 * \param avc Caller may already have a vcache for this file, which is
1601 * \note Environment:
1602 * The cache entry is returned with an increased vrefCount field.
1603 * The entry must be discarded by calling afs_PutVCache when you
1604 * are through using the pointer to the cache entry.
1606 * You should not hold any locks when calling this function, except
1607 * locks on other vcache entries. If you lock more than one vcache
1608 * entry simultaneously, you should lock them in this order:
1610 * 1. Lock all files first, then directories.
1611 * 2. Within a particular type, lock entries in Fid.Vnode order.
1613 * This locking hierarchy is convenient because it allows locking
1614 * of a parent dir cache entry, given a file (to check its access
1615 * control list). It also allows renames to be handled easily by
1616 * locking directories in a constant order.
1618 * \note NB. NewVCache -> FlushVCache presently (4/10/95) drops the xvcache lock.
1620 * \note Might have a vcache structure already, which must
1621 * already be held by the caller
1624 afs_GetVCache(struct VenusFid *afid, struct vrequest *areq,
1625 afs_int32 * cached, struct vcache *avc)
1628 afs_int32 code, newvcache = 0;
1633 AFS_STATCNT(afs_GetVCache);
1636 *cached = 0; /* Init just in case */
1638 #if defined(AFS_SGI_ENV) && !defined(AFS_SGI53_ENV)
1642 ObtainSharedLock(&afs_xvcache, 5);
1644 tvc = afs_FindVCache(afid, &retry, DO_STATS | DO_VLRU | IS_SLOCK);
1646 #if defined(AFS_SGI_ENV) && !defined(AFS_SGI53_ENV)
1647 ReleaseSharedLock(&afs_xvcache);
1648 spunlock_psema(tvc->v.v_lock, retry, &tvc->v.v_sync, PINOD);
1655 osi_Assert((tvc->f.states & CVInit) == 0);
1656 /* If we are in readdir, return the vnode even if not statd */
1657 if ((tvc->f.states & CStatd) || afs_InReadDir(tvc)) {
1658 ReleaseSharedLock(&afs_xvcache);
1662 UpgradeSToWLock(&afs_xvcache, 21);
1664 /* no cache entry, better grab one */
1665 tvc = afs_NewVCache(afid, NULL);
1668 ConvertWToSLock(&afs_xvcache);
1671 ReleaseSharedLock(&afs_xvcache);
1675 afs_stats_cmperf.vcacheMisses++;
1678 ReleaseSharedLock(&afs_xvcache);
1680 ObtainWriteLock(&tvc->lock, 54);
1682 if (tvc->f.states & CStatd) {
1683 ReleaseWriteLock(&tvc->lock);
1686 #ifdef AFS_DARWIN80_ENV
1687 /* Darwin 8.0 only has bufs in nfs, so we shouldn't have to worry about them.
1690 #if defined(AFS_DARWIN_ENV) || defined(AFS_FBSD_ENV)
1692 * XXX - I really don't like this. Should try to understand better.
1693 * It seems that sometimes, when we get called, we already hold the
1694 * lock on the vnode (e.g., from afs_getattr via afs_VerifyVCache).
1695 * We can't drop the vnode lock, because that could result in a race.
1696 * Sometimes, though, we get here and don't hold the vnode lock.
1697 * I hate code paths that sometimes hold locks and sometimes don't.
1698 * In any event, the dodge we use here is to check whether the vnode
1699 * is locked, and if it isn't, then we gain and drop it around the call
1700 * to vinvalbuf; otherwise, we leave it alone.
1703 struct vnode *vp = AFSTOV(tvc);
1706 #if defined(AFS_DARWIN_ENV)
1707 iheldthelock = VOP_ISLOCKED(vp);
1709 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, current_proc());
1710 /* this is messy. we can call fsync which will try to reobtain this */
1711 if (VTOAFS(vp) == tvc)
1712 ReleaseWriteLock(&tvc->lock);
1713 if (UBCINFOEXISTS(vp)) {
1714 vinvalbuf(vp, V_SAVE, &afs_osi_cred, current_proc(), PINOD, 0);
1716 if (VTOAFS(vp) == tvc)
1717 ObtainWriteLock(&tvc->lock, 954);
1719 VOP_UNLOCK(vp, LK_EXCLUSIVE, current_proc());
1720 #elif defined(AFS_FBSD80_ENV)
1721 iheldthelock = VOP_ISLOCKED(vp);
1722 if (!iheldthelock) {
1723 /* nosleep/sleep lock order reversal */
1724 int glocked = ISAFS_GLOCK();
1727 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1731 vinvalbuf(vp, V_SAVE, PINOD, 0); /* changed late in 8.0-CURRENT */
1734 #elif defined(AFS_FBSD60_ENV)
1735 iheldthelock = VOP_ISLOCKED(vp, curthread);
1737 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, curthread);
1739 vinvalbuf(vp, V_SAVE, curthread, PINOD, 0);
1742 VOP_UNLOCK(vp, LK_EXCLUSIVE, curthread);
1743 #elif defined(AFS_FBSD_ENV)
1744 iheldthelock = VOP_ISLOCKED(vp, curthread);
1746 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, curthread);
1747 vinvalbuf(vp, V_SAVE, osi_curcred(), curthread, PINOD, 0);
1749 VOP_UNLOCK(vp, LK_EXCLUSIVE, curthread);
1750 #elif defined(AFS_OBSD_ENV)
1751 iheldthelock = VOP_ISLOCKED(vp, curproc);
1753 VOP_LOCK(vp, LK_EXCLUSIVE | LK_RETRY, curproc);
1754 uvm_vnp_uncache(vp);
1756 VOP_UNLOCK(vp, 0, curproc);
1757 #elif defined(AFS_NBSD40_ENV)
1758 iheldthelock = VOP_ISLOCKED(vp);
1759 if (!iheldthelock) {
1760 VOP_LOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1762 uvm_vnp_uncache(vp);
1770 ObtainWriteLock(&afs_xcbhash, 464);
1771 tvc->f.states &= ~CUnique;
1773 afs_DequeueCallback(tvc);
1774 ReleaseWriteLock(&afs_xcbhash);
1776 /* It is always appropriate to throw away all the access rights? */
1777 afs_FreeAllAxs(&(tvc->Access));
1778 tvp = afs_GetVolume(afid, areq, READ_LOCK); /* copy useful per-volume info */
1780 if ((tvp->states & VForeign)) {
1782 tvc->f.states |= CForeign;
1783 if (newvcache && (tvp->rootVnode == afid->Fid.Vnode)
1784 && (tvp->rootUnique == afid->Fid.Unique)) {
1788 if (tvp->states & VRO)
1789 tvc->f.states |= CRO;
1790 if (tvp->states & VBackup)
1791 tvc->f.states |= CBackup;
1792 /* now copy ".." entry back out of volume structure, if necessary */
1793 if (tvc->mvstat == 2 && tvp->dotdot.Fid.Volume != 0) {
1795 tvc->mvid = (struct VenusFid *)
1796 osi_AllocSmallSpace(sizeof(struct VenusFid));
1797 *tvc->mvid = tvp->dotdot;
1799 afs_PutVolume(tvp, READ_LOCK);
1803 afs_RemoveVCB(afid);
1805 struct AFSFetchStatus OutStatus;
1807 if (afs_DynrootNewVnode(tvc, &OutStatus)) {
1808 afs_ProcessFS(tvc, &OutStatus, areq);
1809 tvc->f.states |= CStatd | CUnique;
1810 tvc->f.parent.vnode = OutStatus.ParentVnode;
1811 tvc->f.parent.unique = OutStatus.ParentUnique;
1815 if (AFS_IS_DISCONNECTED) {
1816 /* Nothing to do otherwise...*/
1818 /* printf("Network is down in afs_GetCache"); */
1820 code = afs_FetchStatus(tvc, afid, areq, &OutStatus);
1822 /* For the NFS translator's benefit, make sure
1823 * non-directory vnodes always have their parent FID set
1824 * correctly, even when created as a result of decoding an
1825 * NFS filehandle. It would be nice to also do this for
1826 * directories, but we can't because the fileserver fills
1827 * in the FID of the directory itself instead of that of
1830 if (!code && OutStatus.FileType != Directory &&
1831 !tvc->f.parent.vnode) {
1832 tvc->f.parent.vnode = OutStatus.ParentVnode;
1833 tvc->f.parent.unique = OutStatus.ParentUnique;
1834 /* XXX - SXW - It's conceivable we should mark ourselves
1835 * as dirty again here, incase we've been raced
1836 * out of the FetchStatus call.
1843 ReleaseWriteLock(&tvc->lock);
1849 ReleaseWriteLock(&tvc->lock);
1852 } /*afs_GetVCache */
1857 * Lookup a vcache by fid. Look inside the cache first, if not
1858 * there, lookup the file on the server, and then get it's fresh
1863 * \param cached Is element cached? If NULL, don't answer.
1867 * \return The found element or NULL.
1870 afs_LookupVCache(struct VenusFid *afid, struct vrequest *areq,
1871 afs_int32 * cached, struct vcache *adp, char *aname)
1873 afs_int32 code, now, newvcache = 0;
1874 struct VenusFid nfid;
1877 struct AFSFetchStatus OutStatus;
1878 struct AFSCallBack CallBack;
1879 struct AFSVolSync tsync;
1880 struct server *serverp = 0;
1884 AFS_STATCNT(afs_GetVCache);
1886 *cached = 0; /* Init just in case */
1888 #if defined(AFS_SGI_ENV) && !defined(AFS_SGI53_ENV)
1892 ObtainReadLock(&afs_xvcache);
1893 tvc = afs_FindVCache(afid, &retry, DO_STATS /* no vlru */ );
1896 ReleaseReadLock(&afs_xvcache);
1898 #if defined(AFS_SGI_ENV) && !defined(AFS_SGI53_ENV)
1899 spunlock_psema(tvc->v.v_lock, retry, &tvc->v.v_sync, PINOD);
1903 ObtainReadLock(&tvc->lock);
1905 if (tvc->f.states & CStatd) {
1909 ReleaseReadLock(&tvc->lock);
1912 tvc->f.states &= ~CUnique;
1914 ReleaseReadLock(&tvc->lock);
1916 ObtainReadLock(&afs_xvcache);
1919 ReleaseReadLock(&afs_xvcache);
1921 /* lookup the file */
1924 origCBs = afs_allCBs; /* if anything changes, we don't have a cb */
1926 if (AFS_IS_DISCONNECTED) {
1927 /* printf("Network is down in afs_LookupVcache\n"); */
1931 afs_RemoteLookup(&adp->f.fid, areq, aname, &nfid, &OutStatus,
1932 &CallBack, &serverp, &tsync);
1934 #if defined(AFS_SGI_ENV) && !defined(AFS_SGI53_ENV)
1938 ObtainSharedLock(&afs_xvcache, 6);
1939 tvc = afs_FindVCache(&nfid, &retry, DO_VLRU | IS_SLOCK/* no xstats now */ );
1941 #if defined(AFS_SGI_ENV) && !defined(AFS_SGI53_ENV)
1942 ReleaseSharedLock(&afs_xvcache);
1943 spunlock_psema(tvc->v.v_lock, retry, &tvc->v.v_sync, PINOD);
1949 /* no cache entry, better grab one */
1950 UpgradeSToWLock(&afs_xvcache, 22);
1951 tvc = afs_NewVCache(&nfid, serverp);
1953 ConvertWToSLock(&afs_xvcache);
1956 ReleaseSharedLock(&afs_xvcache);
1961 ReleaseSharedLock(&afs_xvcache);
1962 ObtainWriteLock(&tvc->lock, 55);
1964 /* It is always appropriate to throw away all the access rights? */
1965 afs_FreeAllAxs(&(tvc->Access));
1966 tvp = afs_GetVolume(afid, areq, READ_LOCK); /* copy useful per-vol info */
1968 if ((tvp->states & VForeign)) {
1970 tvc->f.states |= CForeign;
1971 if (newvcache && (tvp->rootVnode == afid->Fid.Vnode)
1972 && (tvp->rootUnique == afid->Fid.Unique))
1975 if (tvp->states & VRO)
1976 tvc->f.states |= CRO;
1977 if (tvp->states & VBackup)
1978 tvc->f.states |= CBackup;
1979 /* now copy ".." entry back out of volume structure, if necessary */
1980 if (tvc->mvstat == 2 && tvp->dotdot.Fid.Volume != 0) {
1982 tvc->mvid = (struct VenusFid *)
1983 osi_AllocSmallSpace(sizeof(struct VenusFid));
1984 *tvc->mvid = tvp->dotdot;
1989 ObtainWriteLock(&afs_xcbhash, 465);
1990 afs_DequeueCallback(tvc);
1991 tvc->f.states &= ~(CStatd | CUnique);
1992 ReleaseWriteLock(&afs_xcbhash);
1993 if ((tvc->f.states & CForeign) || (tvc->f.fid.Fid.Vnode & 1))
1994 osi_dnlc_purgedp(tvc); /* if it (could be) a directory */
1996 afs_PutVolume(tvp, READ_LOCK);
1997 ReleaseWriteLock(&tvc->lock);
2002 ObtainWriteLock(&afs_xcbhash, 466);
2003 if (origCBs == afs_allCBs) {
2004 if (CallBack.ExpirationTime) {
2005 tvc->callback = serverp;
2006 tvc->cbExpires = CallBack.ExpirationTime + now;
2007 tvc->f.states |= CStatd | CUnique;
2008 tvc->f.states &= ~CBulkFetching;
2009 afs_QueueCallback(tvc, CBHash(CallBack.ExpirationTime), tvp);
2010 } else if (tvc->f.states & CRO) {
2011 /* adapt gives us an hour. */
2012 tvc->cbExpires = 3600 + osi_Time();
2013 /*XXX*/ tvc->f.states |= CStatd | CUnique;
2014 tvc->f.states &= ~CBulkFetching;
2015 afs_QueueCallback(tvc, CBHash(3600), tvp);
2017 tvc->callback = NULL;
2018 afs_DequeueCallback(tvc);
2019 tvc->f.states &= ~(CStatd | CUnique);
2020 if ((tvc->f.states & CForeign) || (tvc->f.fid.Fid.Vnode & 1))
2021 osi_dnlc_purgedp(tvc); /* if it (could be) a directory */
2024 afs_DequeueCallback(tvc);
2025 tvc->f.states &= ~CStatd;
2026 tvc->f.states &= ~CUnique;
2027 tvc->callback = NULL;
2028 if ((tvc->f.states & CForeign) || (tvc->f.fid.Fid.Vnode & 1))
2029 osi_dnlc_purgedp(tvc); /* if it (could be) a directory */
2031 ReleaseWriteLock(&afs_xcbhash);
2033 afs_PutVolume(tvp, READ_LOCK);
2034 afs_ProcessFS(tvc, &OutStatus, areq);
2036 ReleaseWriteLock(&tvc->lock);
2042 afs_GetRootVCache(struct VenusFid *afid, struct vrequest *areq,
2043 afs_int32 * cached, struct volume *tvolp)
2045 afs_int32 code = 0, i, newvcache = 0, haveStatus = 0;
2046 afs_int32 getNewFid = 0;
2048 struct VenusFid nfid;
2050 struct server *serverp = 0;
2051 struct AFSFetchStatus OutStatus;
2052 struct AFSCallBack CallBack;
2053 struct AFSVolSync tsync;
2055 #ifdef AFS_DARWIN80_ENV
2062 if (!tvolp->rootVnode || getNewFid) {
2063 struct VenusFid tfid;
2066 tfid.Fid.Vnode = 0; /* Means get rootfid of volume */
2067 origCBs = afs_allCBs; /* ignore InitCallBackState */
2069 afs_RemoteLookup(&tfid, areq, NULL, &nfid, &OutStatus, &CallBack,
2074 /* ReleaseReadLock(&tvolp->lock); */
2075 ObtainWriteLock(&tvolp->lock, 56);
2076 tvolp->rootVnode = afid->Fid.Vnode = nfid.Fid.Vnode;
2077 tvolp->rootUnique = afid->Fid.Unique = nfid.Fid.Unique;
2078 ReleaseWriteLock(&tvolp->lock);
2079 /* ObtainReadLock(&tvolp->lock);*/
2082 afid->Fid.Vnode = tvolp->rootVnode;
2083 afid->Fid.Unique = tvolp->rootUnique;
2087 ObtainSharedLock(&afs_xvcache, 7);
2089 for (tvc = afs_vhashT[i]; tvc; tvc = tvc->hnext) {
2090 if (!FidCmp(&(tvc->f.fid), afid)) {
2091 if (tvc->f.states & CVInit) {
2092 ReleaseSharedLock(&afs_xvcache);
2093 afs_osi_Sleep(&tvc->f.states);
2096 #ifdef AFS_DARWIN80_ENV
2097 if (tvc->f.states & CDeadVnode) {
2098 ReleaseSharedLock(&afs_xvcache);
2099 afs_osi_Sleep(&tvc->f.states);
2103 if (vnode_get(tvp)) /* this bumps ref count */
2105 if (vnode_ref(tvp)) {
2107 /* AFSTOV(tvc) may be NULL */
2117 if (!haveStatus && (!tvc || !(tvc->f.states & CStatd))) {
2118 /* Mount point no longer stat'd or unknown. FID may have changed. */
2120 ReleaseSharedLock(&afs_xvcache);
2121 #ifdef AFS_DARWIN80_ENV
2124 vnode_put(AFSTOV(tvc));
2125 vnode_rele(AFSTOV(tvc));
2134 UpgradeSToWLock(&afs_xvcache, 23);
2135 /* no cache entry, better grab one */
2136 tvc = afs_NewVCache(afid, NULL);
2139 ReleaseWriteLock(&afs_xvcache);
2143 afs_stats_cmperf.vcacheMisses++;
2147 afs_stats_cmperf.vcacheHits++;
2148 #if defined(AFS_DARWIN80_ENV)
2149 /* we already bumped the ref count in the for loop above */
2150 #else /* AFS_DARWIN80_ENV */
2153 UpgradeSToWLock(&afs_xvcache, 24);
2154 if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) {
2155 refpanic("GRVC VLRU inconsistent0");
2157 if (tvc->vlruq.next->prev != &(tvc->vlruq)) {
2158 refpanic("GRVC VLRU inconsistent1");
2160 if (tvc->vlruq.prev->next != &(tvc->vlruq)) {
2161 refpanic("GRVC VLRU inconsistent2");
2163 QRemove(&tvc->vlruq); /* move to lruq head */
2164 QAdd(&VLRU, &tvc->vlruq);
2165 if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) {
2166 refpanic("GRVC VLRU inconsistent3");
2168 if (tvc->vlruq.next->prev != &(tvc->vlruq)) {
2169 refpanic("GRVC VLRU inconsistent4");
2171 if (tvc->vlruq.prev->next != &(tvc->vlruq)) {
2172 refpanic("GRVC VLRU inconsistent5");
2177 ReleaseWriteLock(&afs_xvcache);
2179 if (tvc->f.states & CStatd) {
2183 ObtainReadLock(&tvc->lock);
2184 tvc->f.states &= ~CUnique;
2185 tvc->callback = NULL; /* redundant, perhaps */
2186 ReleaseReadLock(&tvc->lock);
2189 ObtainWriteLock(&tvc->lock, 57);
2191 /* It is always appropriate to throw away all the access rights? */
2192 afs_FreeAllAxs(&(tvc->Access));
2195 tvc->f.states |= CForeign;
2196 if (tvolp->states & VRO)
2197 tvc->f.states |= CRO;
2198 if (tvolp->states & VBackup)
2199 tvc->f.states |= CBackup;
2200 /* now copy ".." entry back out of volume structure, if necessary */
2201 if (newvcache && (tvolp->rootVnode == afid->Fid.Vnode)
2202 && (tvolp->rootUnique == afid->Fid.Unique)) {
2205 if (tvc->mvstat == 2 && tvolp->dotdot.Fid.Volume != 0) {
2207 tvc->mvid = (struct VenusFid *)
2208 osi_AllocSmallSpace(sizeof(struct VenusFid));
2209 *tvc->mvid = tvolp->dotdot;
2213 afs_RemoveVCB(afid);
2216 struct VenusFid tfid;
2219 tfid.Fid.Vnode = 0; /* Means get rootfid of volume */
2220 origCBs = afs_allCBs; /* ignore InitCallBackState */
2222 afs_RemoteLookup(&tfid, areq, NULL, &nfid, &OutStatus, &CallBack,
2227 ObtainWriteLock(&afs_xcbhash, 467);
2228 afs_DequeueCallback(tvc);
2229 tvc->callback = NULL;
2230 tvc->f.states &= ~(CStatd | CUnique);
2231 ReleaseWriteLock(&afs_xcbhash);
2232 if ((tvc->f.states & CForeign) || (tvc->f.fid.Fid.Vnode & 1))
2233 osi_dnlc_purgedp(tvc); /* if it (could be) a directory */
2234 ReleaseWriteLock(&tvc->lock);
2239 ObtainWriteLock(&afs_xcbhash, 468);
2240 if (origCBs == afs_allCBs) {
2241 tvc->f.states |= CTruth;
2242 tvc->callback = serverp;
2243 if (CallBack.ExpirationTime != 0) {
2244 tvc->cbExpires = CallBack.ExpirationTime + start;
2245 tvc->f.states |= CStatd;
2246 tvc->f.states &= ~CBulkFetching;
2247 afs_QueueCallback(tvc, CBHash(CallBack.ExpirationTime), tvolp);
2248 } else if (tvc->f.states & CRO) {
2249 /* adapt gives us an hour. */
2250 tvc->cbExpires = 3600 + osi_Time();
2251 /*XXX*/ tvc->f.states |= CStatd;
2252 tvc->f.states &= ~CBulkFetching;
2253 afs_QueueCallback(tvc, CBHash(3600), tvolp);
2256 afs_DequeueCallback(tvc);
2257 tvc->callback = NULL;
2258 tvc->f.states &= ~(CStatd | CUnique);
2259 if ((tvc->f.states & CForeign) || (tvc->f.fid.Fid.Vnode & 1))
2260 osi_dnlc_purgedp(tvc); /* if it (could be) a directory */
2262 ReleaseWriteLock(&afs_xcbhash);
2263 afs_ProcessFS(tvc, &OutStatus, areq);
2265 ReleaseWriteLock(&tvc->lock);
2271 * Update callback status and (sometimes) attributes of a vnode.
2272 * Called after doing a fetch status RPC. Whilst disconnected, attributes
2273 * shouldn't be written to the vcache here.
2278 * \param Outsp Server status after rpc call.
2279 * \param acb Callback for this vnode.
2281 * \note The vcache must be write locked.
2284 afs_UpdateStatus(struct vcache *avc, struct VenusFid *afid,
2285 struct vrequest *areq, struct AFSFetchStatus *Outsp,
2286 struct AFSCallBack *acb, afs_uint32 start)
2288 struct volume *volp;
2291 /* Dont write status in vcache if resyncing after a disconnection. */
2292 afs_ProcessFS(avc, Outsp, areq);
2294 volp = afs_GetVolume(afid, areq, READ_LOCK);
2295 ObtainWriteLock(&afs_xcbhash, 469);
2296 avc->f.states |= CTruth;
2297 if (avc->callback /* check for race */ ) {
2298 if (acb->ExpirationTime != 0) {
2299 avc->cbExpires = acb->ExpirationTime + start;
2300 avc->f.states |= CStatd;
2301 avc->f.states &= ~CBulkFetching;
2302 afs_QueueCallback(avc, CBHash(acb->ExpirationTime), volp);
2303 } else if (avc->f.states & CRO) {
2304 /* ordinary callback on a read-only volume -- AFS 3.2 style */
2305 avc->cbExpires = 3600 + start;
2306 avc->f.states |= CStatd;
2307 avc->f.states &= ~CBulkFetching;
2308 afs_QueueCallback(avc, CBHash(3600), volp);
2310 afs_DequeueCallback(avc);
2311 avc->callback = NULL;
2312 avc->f.states &= ~(CStatd | CUnique);
2313 if ((avc->f.states & CForeign) || (avc->f.fid.Fid.Vnode & 1))
2314 osi_dnlc_purgedp(avc); /* if it (could be) a directory */
2317 afs_DequeueCallback(avc);
2318 avc->callback = NULL;
2319 avc->f.states &= ~(CStatd | CUnique);
2320 if ((avc->f.states & CForeign) || (avc->f.fid.Fid.Vnode & 1))
2321 osi_dnlc_purgedp(avc); /* if it (could be) a directory */
2323 ReleaseWriteLock(&afs_xcbhash);
2325 afs_PutVolume(volp, READ_LOCK);
2329 afs_BadFetchStatus(struct afs_conn *tc)
2331 int addr = ntohl(tc->parent->srvr->sa_ip);
2332 afs_warn("afs: Invalid AFSFetchStatus from server %u.%u.%u.%u\n",
2333 (addr >> 24) & 0xff, (addr >> 16) & 0xff, (addr >> 8) & 0xff,
2335 afs_warn("afs: This suggests the server may be sending bad data that "
2336 "can lead to availability issues or data corruption. The "
2337 "issue has been avoided for now, but it may not always be "
2338 "detectable. Please upgrade the server if possible.\n");
2342 * Check if a given AFSFetchStatus structure is sane.
2344 * @param[in] tc The server from which we received the status
2345 * @param[in] status The status we received
2347 * @return whether the given structure is valid or not
2348 * @retval 0 the structure is fine
2349 * @retval nonzero the structure looks like garbage; act as if we received
2350 * the returned error code from the server
2353 afs_CheckFetchStatus(struct afs_conn *tc, struct AFSFetchStatus *status)
2355 if (status->errorCode ||
2356 status->InterfaceVersion != 1 ||
2357 !(status->FileType > Invalid && status->FileType <= SymbolicLink) ||
2358 status->ParentVnode == 0 || status->ParentUnique == 0) {
2360 afs_warn("afs: FetchStatus ec %u iv %u ft %u pv %u pu %u\n",
2361 (unsigned)status->errorCode, (unsigned)status->InterfaceVersion,
2362 (unsigned)status->FileType, (unsigned)status->ParentVnode,
2363 (unsigned)status->ParentUnique);
2364 afs_BadFetchStatus(tc);
2372 * Must be called with avc write-locked
2373 * don't absolutely have to invalidate the hint unless the dv has
2374 * changed, but be sure to get it right else there will be consistency bugs.
2377 afs_FetchStatus(struct vcache * avc, struct VenusFid * afid,
2378 struct vrequest * areq, struct AFSFetchStatus * Outsp)
2381 afs_uint32 start = 0;
2382 struct afs_conn *tc;
2383 struct AFSCallBack CallBack;
2384 struct AFSVolSync tsync;
2385 struct rx_connection *rxconn;
2388 tc = afs_Conn(afid, areq, SHARED_LOCK, &rxconn);
2389 avc->dchint = NULL; /* invalidate hints */
2391 avc->callback = tc->parent->srvr->server;
2393 XSTATS_START_TIME(AFS_STATS_FS_RPCIDX_FETCHSTATUS);
2396 RXAFS_FetchStatus(rxconn, (struct AFSFid *)&afid->Fid, Outsp,
2403 code = afs_CheckFetchStatus(tc, Outsp);
2408 } while (afs_Analyze
2409 (tc, rxconn, code, afid, areq, AFS_STATS_FS_RPCIDX_FETCHSTATUS,
2410 SHARED_LOCK, NULL));
2413 afs_UpdateStatus(avc, afid, areq, Outsp, &CallBack, start);
2415 /* used to undo the local callback, but that's too extreme.
2416 * There are plenty of good reasons that fetchstatus might return
2417 * an error, such as EPERM. If we have the vnode cached, statd,
2418 * with callback, might as well keep track of the fact that we
2419 * don't have access...
2421 if (code == EPERM || code == EACCES) {
2422 struct axscache *ac;
2423 if (avc->Access && (ac = afs_FindAxs(avc->Access, areq->uid)))
2425 else /* not found, add a new one if possible */
2426 afs_AddAxs(avc->Access, areq->uid, 0);
2437 * Stuff some information into the vcache for the given file.
2440 * afid : File in question.
2441 * OutStatus : Fetch status on the file.
2442 * CallBack : Callback info.
2443 * tc : RPC connection involved.
2444 * areq : vrequest involved.
2447 * Nothing interesting.
2450 afs_StuffVcache(struct VenusFid *afid,
2451 struct AFSFetchStatus *OutStatus,
2452 struct AFSCallBack *CallBack, struct afs_conn *tc,
2453 struct vrequest *areq)
2455 afs_int32 code, i, newvcache = 0;
2457 struct AFSVolSync tsync;
2459 struct axscache *ac;
2462 AFS_STATCNT(afs_StuffVcache);
2463 #ifdef IFS_VCACHECOUNT
2468 ObtainSharedLock(&afs_xvcache, 8);
2470 tvc = afs_FindVCache(afid, &retry, DO_VLRU| IS_SLOCK /* no stats */ );
2472 #if defined(AFS_SGI_ENV) && !defined(AFS_SGI53_ENV)
2473 ReleaseSharedLock(&afs_xvcache);
2474 spunlock_psema(tvc->v.v_lock, retry, &tvc->v.v_sync, PINOD);
2480 /* no cache entry, better grab one */
2481 UpgradeSToWLock(&afs_xvcache, 25);
2482 tvc = afs_NewVCache(afid, NULL);
2484 ConvertWToSLock(&afs_xvcache);
2487 ReleaseSharedLock(&afs_xvcache);
2492 ReleaseSharedLock(&afs_xvcache);
2493 ObtainWriteLock(&tvc->lock, 58);
2495 tvc->f.states &= ~CStatd;
2496 if ((tvc->f.states & CForeign) || (tvc->f.fid.Fid.Vnode & 1))
2497 osi_dnlc_purgedp(tvc); /* if it (could be) a directory */
2499 /* Is it always appropriate to throw away all the access rights? */
2500 afs_FreeAllAxs(&(tvc->Access));
2502 /*Copy useful per-volume info */
2503 tvp = afs_GetVolume(afid, areq, READ_LOCK);
2505 if (newvcache && (tvp->states & VForeign))
2506 tvc->f.states |= CForeign;
2507 if (tvp->states & VRO)
2508 tvc->f.states |= CRO;
2509 if (tvp->states & VBackup)
2510 tvc->f.states |= CBackup;
2512 * Now, copy ".." entry back out of volume structure, if
2515 if (tvc->mvstat == 2 && tvp->dotdot.Fid.Volume != 0) {
2517 tvc->mvid = (struct VenusFid *)
2518 osi_AllocSmallSpace(sizeof(struct VenusFid));
2519 *tvc->mvid = tvp->dotdot;
2522 /* store the stat on the file */
2523 afs_RemoveVCB(afid);
2524 afs_ProcessFS(tvc, OutStatus, areq);
2525 tvc->callback = tc->srvr->server;
2527 /* we use osi_Time twice below. Ideally, we would use the time at which
2528 * the FetchStatus call began, instead, but we don't have it here. So we
2529 * make do with "now". In the CRO case, it doesn't really matter. In
2530 * the other case, we hope that the difference between "now" and when the
2531 * call actually began execution on the server won't be larger than the
2532 * padding which the server keeps. Subtract 1 second anyway, to be on
2533 * the safe side. Can't subtract more because we don't know how big
2534 * ExpirationTime is. Possible consistency problems may arise if the call
2535 * timeout period becomes longer than the server's expiration padding. */
2536 ObtainWriteLock(&afs_xcbhash, 470);
2537 if (CallBack->ExpirationTime != 0) {
2538 tvc->cbExpires = CallBack->ExpirationTime + osi_Time() - 1;
2539 tvc->f.states |= CStatd;
2540 tvc->f.states &= ~CBulkFetching;
2541 afs_QueueCallback(tvc, CBHash(CallBack->ExpirationTime), tvp);
2542 } else if (tvc->f.states & CRO) {
2543 /* old-fashioned AFS 3.2 style */
2544 tvc->cbExpires = 3600 + osi_Time();
2545 /*XXX*/ tvc->f.states |= CStatd;
2546 tvc->f.states &= ~CBulkFetching;
2547 afs_QueueCallback(tvc, CBHash(3600), tvp);
2549 afs_DequeueCallback(tvc);
2550 tvc->callback = NULL;
2551 tvc->f.states &= ~(CStatd | CUnique);
2552 if ((tvc->f.states & CForeign) || (tvc->f.fid.Fid.Vnode & 1))
2553 osi_dnlc_purgedp(tvc); /* if it (could be) a directory */
2555 ReleaseWriteLock(&afs_xcbhash);
2557 afs_PutVolume(tvp, READ_LOCK);
2559 /* look in per-pag cache */
2560 if (tvc->Access && (ac = afs_FindAxs(tvc->Access, areq->uid)))
2561 ac->axess = OutStatus->CallerAccess; /* substitute pags */
2562 else /* not found, add a new one if possible */
2563 afs_AddAxs(tvc->Access, areq->uid, OutStatus->CallerAccess);
2565 ReleaseWriteLock(&tvc->lock);
2566 afs_Trace4(afs_iclSetp, CM_TRACE_STUFFVCACHE, ICL_TYPE_POINTER, tvc,
2567 ICL_TYPE_POINTER, tvc->callback, ICL_TYPE_INT32,
2568 tvc->cbExpires, ICL_TYPE_INT32, tvc->cbExpires - osi_Time());
2570 * Release ref count... hope this guy stays around...
2573 } /*afs_StuffVcache */
2577 * Decrements the reference count on a cache entry.
2579 * \param avc Pointer to the cache entry to decrement.
2581 * \note Environment: Nothing interesting.
2584 afs_PutVCache(struct vcache *avc)
2586 AFS_STATCNT(afs_PutVCache);
2587 #ifdef AFS_DARWIN80_ENV
2588 vnode_put(AFSTOV(avc));
2592 * Can we use a read lock here?
2594 ObtainReadLock(&afs_xvcache);
2596 ReleaseReadLock(&afs_xvcache);
2598 } /*afs_PutVCache */
2602 * Reset a vcache entry, so local contents are ignored, and the
2603 * server will be reconsulted next time the vcache is used
2605 * \param avc Pointer to the cache entry to reset
2607 * \param skipdnlc skip the dnlc purge for this vnode
2609 * \note avc must be write locked on entry
2611 * \note The caller should purge the dnlc when skipdnlc is set.
2614 afs_ResetVCache(struct vcache *avc, afs_ucred_t *acred, afs_int32 skipdnlc)
2616 ObtainWriteLock(&afs_xcbhash, 456);
2617 afs_DequeueCallback(avc);
2618 avc->f.states &= ~(CStatd | CDirty); /* next reference will re-stat */
2619 ReleaseWriteLock(&afs_xcbhash);
2620 /* now find the disk cache entries */
2621 afs_TryToSmush(avc, acred, 1);
2623 osi_dnlc_purgedp(avc);
2625 if (avc->linkData && !(avc->f.states & CCore)) {
2626 afs_osi_Free(avc->linkData, strlen(avc->linkData) + 1);
2627 avc->linkData = NULL;
2632 * Sleepa when searching for a vcache. Releases all the pending locks,
2633 * sleeps then obtains the previously released locks.
2635 * \param vcache Enter sleep state.
2636 * \param flag Determines what locks to use.
2641 findvc_sleep(struct vcache *avc, int flag)
2643 if (flag & IS_SLOCK) {
2644 ReleaseSharedLock(&afs_xvcache);
2646 if (flag & IS_WLOCK) {
2647 ReleaseWriteLock(&afs_xvcache);
2649 ReleaseReadLock(&afs_xvcache);
2652 afs_osi_Sleep(&avc->f.states);
2653 if (flag & IS_SLOCK) {
2654 ObtainSharedLock(&afs_xvcache, 341);
2656 if (flag & IS_WLOCK) {
2657 ObtainWriteLock(&afs_xvcache, 343);
2659 ObtainReadLock(&afs_xvcache);
2665 * Add a reference on an existing vcache entry.
2667 * \param tvc Pointer to the vcache.
2669 * \note Environment: Must be called with at least one reference from
2670 * elsewhere on the vcache, even if that reference will be dropped.
2671 * The global lock is required.
2673 * \return 0 on success, -1 on failure.
2677 afs_RefVCache(struct vcache *tvc)
2679 #ifdef AFS_DARWIN80_ENV
2683 /* AFS_STATCNT(afs_RefVCache); */
2685 #ifdef AFS_DARWIN80_ENV
2689 if (vnode_ref(tvp)) {
2691 /* AFSTOV(tvc) may be NULL */
2700 } /*afs_RefVCache */
2703 * Find a vcache entry given a fid.
2705 * \param afid Pointer to the fid whose cache entry we desire.
2706 * \param retry (SGI-specific) tell the caller to drop the lock on xvcache,
2707 * unlock the vnode, and try again.
2708 * \param flag Bit 1 to specify whether to compute hit statistics. Not
2709 * set if FindVCache is called as part of internal bookkeeping.
2711 * \note Environment: Must be called with the afs_xvcache lock at least held at
2712 * the read level. In order to do the VLRU adjustment, the xvcache lock
2713 * must be shared-- we upgrade it here.
2717 afs_FindVCache(struct VenusFid *afid, afs_int32 * retry, afs_int32 flag)
2722 #ifdef AFS_DARWIN80_ENV
2723 struct vcache *deadvc = NULL, *livevc = NULL;
2727 AFS_STATCNT(afs_FindVCache);
2731 for (tvc = afs_vhashT[i]; tvc; tvc = tvc->hnext) {
2732 if (FidMatches(afid, tvc)) {
2733 if (tvc->f.states & CVInit) {
2734 findvc_sleep(tvc, flag);
2737 #ifdef AFS_DARWIN80_ENV
2738 if (tvc->f.states & CDeadVnode) {
2739 findvc_sleep(tvc, flag);
2747 /* should I have a read lock on the vnode here? */
2751 #if defined(AFS_DARWIN80_ENV)
2755 if (tvp && vnode_ref(tvp)) {
2757 /* AFSTOV(tvc) may be NULL */
2766 #elif defined(AFS_DARWIN_ENV)
2767 tvc->f.states |= CUBCinit;
2769 if (UBCINFOMISSING(AFSTOV(tvc)) ||
2770 UBCINFORECLAIMED(AFSTOV(tvc))) {
2771 ubc_info_init(AFSTOV(tvc));
2774 tvc->f.states &= ~CUBCinit;
2776 osi_vnhold(tvc, retry); /* already held, above */
2777 if (retry && *retry)
2781 * only move to front of vlru if we have proper vcache locking)
2783 if (flag & DO_VLRU) {
2784 if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) {
2785 refpanic("FindVC VLRU inconsistent1");
2787 if (tvc->vlruq.next->prev != &(tvc->vlruq)) {
2788 refpanic("FindVC VLRU inconsistent1");
2790 if (tvc->vlruq.prev->next != &(tvc->vlruq)) {
2791 refpanic("FindVC VLRU inconsistent2");
2793 UpgradeSToWLock(&afs_xvcache, 26);
2794 QRemove(&tvc->vlruq);
2795 QAdd(&VLRU, &tvc->vlruq);
2796 ConvertWToSLock(&afs_xvcache);
2797 if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) {
2798 refpanic("FindVC VLRU inconsistent1");
2800 if (tvc->vlruq.next->prev != &(tvc->vlruq)) {
2801 refpanic("FindVC VLRU inconsistent2");
2803 if (tvc->vlruq.prev->next != &(tvc->vlruq)) {
2804 refpanic("FindVC VLRU inconsistent3");
2810 if (flag & DO_STATS) {
2812 afs_stats_cmperf.vcacheHits++;
2814 afs_stats_cmperf.vcacheMisses++;
2815 if (afs_IsPrimaryCellNum(afid->Cell))
2816 afs_stats_cmperf.vlocalAccesses++;
2818 afs_stats_cmperf.vremoteAccesses++;
2821 } /*afs_FindVCache */
2824 * Find a vcache entry given a fid. Does a wildcard match on what we
2825 * have for the fid. If more than one entry, don't return anything.
2827 * \param avcp Fill in pointer if we found one and only one.
2828 * \param afid Pointer to the fid whose cache entry we desire.
2829 * \param retry (SGI-specific) tell the caller to drop the lock on xvcache,
2830 * unlock the vnode, and try again.
2831 * \param flags bit 1 to specify whether to compute hit statistics. Not
2832 * set if FindVCache is called as part of internal bookkeeping.
2834 * \note Environment: Must be called with the afs_xvcache lock at least held at
2835 * the read level. In order to do the VLRU adjustment, the xvcache lock
2836 * must be shared-- we upgrade it here.
2838 * \return Number of matches found.
2841 int afs_duplicate_nfs_fids = 0;
2844 afs_NFSFindVCache(struct vcache **avcp, struct VenusFid *afid)
2848 afs_int32 count = 0;
2849 struct vcache *found_tvc = NULL;
2850 #ifdef AFS_DARWIN80_ENV
2854 AFS_STATCNT(afs_FindVCache);
2858 ObtainSharedLock(&afs_xvcache, 331);
2861 for (tvc = afs_vhashT[i]; tvc; tvc = tvc->hnext) {
2862 /* Match only on what we have.... */
2863 if (((tvc->f.fid.Fid.Vnode & 0xffff) == afid->Fid.Vnode)
2864 && (tvc->f.fid.Fid.Volume == afid->Fid.Volume)
2865 && ((tvc->f.fid.Fid.Unique & 0xffffff) == afid->Fid.Unique)
2866 && (tvc->f.fid.Cell == afid->Cell)) {
2867 if (tvc->f.states & CVInit) {
2868 ReleaseSharedLock(&afs_xvcache);
2869 afs_osi_Sleep(&tvc->f.states);
2872 #ifdef AFS_DARWIN80_ENV
2873 if (tvc->f.states & CDeadVnode) {
2874 ReleaseSharedLock(&afs_xvcache);
2875 afs_osi_Sleep(&tvc->f.states);
2879 if (vnode_get(tvp)) {
2880 /* This vnode no longer exists. */
2883 if (vnode_ref(tvp)) {
2884 /* This vnode no longer exists. */
2886 /* AFSTOV(tvc) may be NULL */
2891 #endif /* AFS_DARWIN80_ENV */
2895 afs_duplicate_nfs_fids++;
2896 ReleaseSharedLock(&afs_xvcache);
2897 #ifdef AFS_DARWIN80_ENV
2898 /* Drop our reference counts. */
2899 vnode_put(AFSTOV(tvc));
2900 vnode_put(AFSTOV(found_tvc));
2909 /* should I have a read lock on the vnode here? */
2911 #ifndef AFS_DARWIN80_ENV
2912 #if defined(AFS_SGI_ENV) && !defined(AFS_SGI53_ENV)
2913 afs_int32 retry = 0;
2914 osi_vnhold(tvc, &retry);
2917 found_tvc = (struct vcache *)0;
2918 ReleaseSharedLock(&afs_xvcache);
2919 spunlock_psema(tvc->v.v_lock, retry, &tvc->v.v_sync, PINOD);
2923 osi_vnhold(tvc, (int *)0); /* already held, above */
2927 * We obtained the xvcache lock above.
2929 if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) {
2930 refpanic("FindVC VLRU inconsistent1");
2932 if (tvc->vlruq.next->prev != &(tvc->vlruq)) {
2933 refpanic("FindVC VLRU inconsistent1");
2935 if (tvc->vlruq.prev->next != &(tvc->vlruq)) {
2936 refpanic("FindVC VLRU inconsistent2");
2938 UpgradeSToWLock(&afs_xvcache, 568);
2939 QRemove(&tvc->vlruq);
2940 QAdd(&VLRU, &tvc->vlruq);
2941 ConvertWToSLock(&afs_xvcache);
2942 if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) {
2943 refpanic("FindVC VLRU inconsistent1");
2945 if (tvc->vlruq.next->prev != &(tvc->vlruq)) {
2946 refpanic("FindVC VLRU inconsistent2");
2948 if (tvc->vlruq.prev->next != &(tvc->vlruq)) {
2949 refpanic("FindVC VLRU inconsistent3");
2955 afs_stats_cmperf.vcacheHits++;
2957 afs_stats_cmperf.vcacheMisses++;
2958 if (afs_IsPrimaryCellNum(afid->Cell))
2959 afs_stats_cmperf.vlocalAccesses++;
2961 afs_stats_cmperf.vremoteAccesses++;
2963 *avcp = tvc; /* May be null */
2965 ReleaseSharedLock(&afs_xvcache);
2966 return (tvc ? 1 : 0);
2968 } /*afs_NFSFindVCache */
2974 * Initialize vcache related variables
2979 afs_vcacheInit(int astatSize)
2981 #if !defined(AFS_LINUX22_ENV)
2985 if (!afs_maxvcount) {
2986 afs_maxvcount = astatSize; /* no particular limit on linux? */
2988 #if !defined(AFS_LINUX22_ENV)
2992 AFS_RWLOCK_INIT(&afs_xvcache, "afs_xvcache");
2993 LOCK_INIT(&afs_xvcb, "afs_xvcb");
2995 #if !defined(AFS_LINUX22_ENV)
2996 /* Allocate and thread the struct vcache entries */
2997 tvp = afs_osi_Alloc(astatSize * sizeof(struct vcache));
2998 osi_Assert(tvp != NULL);
2999 memset(tvp, 0, sizeof(struct vcache) * astatSize);
3001 Initial_freeVCList = tvp;
3002 freeVCList = &(tvp[0]);
3003 for (i = 0; i < astatSize - 1; i++) {
3004 tvp[i].nextfree = &(tvp[i + 1]);
3006 tvp[astatSize - 1].nextfree = NULL;
3007 # ifdef KERNEL_HAVE_PIN
3008 pin((char *)tvp, astatSize * sizeof(struct vcache)); /* XXX */
3012 #if defined(AFS_SGI_ENV)
3013 for (i = 0; i < astatSize; i++) {
3014 char name[METER_NAMSZ];
3015 struct vcache *tvc = &tvp[i];
3017 tvc->v.v_number = ++afsvnumbers;
3018 tvc->vc_rwlockid = OSI_NO_LOCKID;
3019 initnsema(&tvc->vc_rwlock, 1,
3020 makesname(name, "vrw", tvc->v.v_number));
3021 #ifndef AFS_SGI53_ENV
3022 initnsema(&tvc->v.v_sync, 0, makesname(name, "vsy", tvc->v.v_number));
3024 #ifndef AFS_SGI62_ENV
3025 initnlock(&tvc->v.v_lock, makesname(name, "vlk", tvc->v.v_number));
3026 #endif /* AFS_SGI62_ENV */
3030 for(i = 0; i < VCSIZE; ++i)
3031 QInit(&afs_vhashTV[i]);
3038 shutdown_vcache(void)
3041 struct afs_cbr *tsp;
3043 * XXX We may potentially miss some of the vcaches because if when
3044 * there are no free vcache entries and all the vcache entries are active
3045 * ones then we allocate an additional one - admittedly we almost never
3050 struct afs_q *tq, *uq = NULL;
3052 for (tq = VLRU.prev; tq != &VLRU; tq = uq) {
3056 osi_FreeSmallSpace(tvc->mvid);
3057 tvc->mvid = (struct VenusFid *)0;
3060 aix_gnode_rele(AFSTOV(tvc));
3062 if (tvc->linkData) {
3063 afs_osi_Free(tvc->linkData, strlen(tvc->linkData) + 1);
3068 * Also free the remaining ones in the Cache
3070 for (i = 0; i < VCSIZE; i++) {
3071 for (tvc = afs_vhashT[i]; tvc; tvc = tvc->hnext) {
3073 osi_FreeSmallSpace(tvc->mvid);
3074 tvc->mvid = (struct VenusFid *)0;
3078 afs_osi_Free(tvc->v.v_gnode, sizeof(struct gnode));
3079 #ifdef AFS_AIX32_ENV
3082 vms_delete(tvc->segid);
3084 tvc->segid = tvc->vmh = NULL;
3085 if (VREFCOUNT_GT(tvc,0))
3086 osi_Panic("flushVcache: vm race");
3094 #if defined(AFS_SUN5_ENV)
3100 if (tvc->linkData) {
3101 afs_osi_Free(tvc->linkData, strlen(tvc->linkData) + 1);
3106 afs_FreeAllAxs(&(tvc->Access));
3112 * Free any leftover callback queue
3114 for (i = 0; i < afs_stats_cmperf.CallBackAlloced; i++) {
3115 tsp = afs_cbrHeads[i];
3116 afs_cbrHeads[i] = 0;
3117 afs_osi_Free((char *)tsp, AFS_NCBRS * sizeof(struct afs_cbr));
3121 #if !defined(AFS_LINUX22_ENV)
3122 afs_osi_Free(Initial_freeVCList, afs_cacheStats * sizeof(struct vcache));
3124 # ifdef KERNEL_HAVE_PIN
3125 unpin(Initial_freeVCList, afs_cacheStats * sizeof(struct vcache));
3128 freeVCList = Initial_freeVCList = 0;
3131 AFS_RWLOCK_INIT(&afs_xvcache, "afs_xvcache");
3132 LOCK_INIT(&afs_xvcb, "afs_xvcb");
3134 for(i = 0; i < VCSIZE; ++i)
3135 QInit(&afs_vhashTV[i]);
3139 afs_DisconGiveUpCallbacks(void)
3145 ObtainWriteLock(&afs_xvcache, 1002); /* XXX - should be a unique number */
3148 /* Somehow, walk the set of vcaches, with each one coming out as tvc */
3149 for (i = 0; i < VCSIZE; i++) {
3150 for (tvc = afs_vhashT[i]; tvc; tvc = tvc->hnext) {
3152 if (afs_QueueVCB(tvc, &slept)) {
3153 tvc->callback = NULL;
3162 ReleaseWriteLock(&afs_xvcache);
3169 * Clear the Statd flag from all vcaches
3171 * This function removes the Statd flag from all vcaches. It's used by
3172 * disconnected mode to tidy up during reconnection
3176 afs_ClearAllStatdFlag(void)
3181 ObtainWriteLock(&afs_xvcache, 715);
3183 for (i = 0; i < VCSIZE; i++) {
3184 for (tvc = afs_vhashT[i]; tvc; tvc = tvc->hnext) {
3185 tvc->f.states &= ~(CStatd|CUnique);
3188 ReleaseWriteLock(&afs_xvcache);