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) {
175 avc->hnext = (struct vcache *)NULL;
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;
710 afs_int32 target = anumber;
714 for (tq = VLRU.prev; tq != &VLRU && anumber > 0; tq = uq) {
717 if (tvc->f.states & CVFlushed) {
718 refpanic("CVFlushed on VLRU");
719 /* In the other path, this was 2 * afs_cacheStats */
720 } else if (!afsd_dynamic_vcaches && i++ > afs_maxvcount) {
721 refpanic("Exceeded pool of AFS vnodes(VLRU cycle?)");
722 } else if (QNext(uq) != tq) {
723 refpanic("VLRU inconsistent");
724 } else if (tvc->f.states & CVInit) {
729 if (osi_TryEvictVCache(tvc, &fv_slept, defersleep))
737 continue; /* start over - may have raced. */
740 if (anumber && !defersleep) {
748 if (!afsd_dynamic_vcaches && anumber == target) {
749 afs_warn("afs_ShakeLooseVCaches: warning none freed, using %d of %d\n",
750 afs_vcount, afs_maxvcount);
756 /* Alloc new vnode. */
758 static struct vcache *
759 afs_AllocVCache(void)
763 tvc = osi_NewVnode();
768 if (afsd_dynamic_vcaches && afs_maxvcount < afs_vcount) {
769 afs_maxvcount = afs_vcount;
770 /*printf("peak vnodes: %d\n", afs_maxvcount);*/
773 afs_stats_cmperf.vcacheXAllocs++; /* count in case we have a leak */
775 /* If we create a new inode, we either give it a new slot number,
776 * or if one's available, use a slot number from the slot free list
778 if (afs_freeSlotList != NULL) {
779 struct afs_slotlist *tmp;
781 tvc->diskSlot = afs_freeSlotList->slot;
782 tmp = afs_freeSlotList;
783 afs_freeSlotList = tmp->next;
784 afs_osi_Free(tmp, sizeof(struct afs_slotlist));
786 tvc->diskSlot = afs_nextVcacheSlot++;
792 /* Pre populate a newly allocated vcache. On platforms where the actual
793 * vnode is attached to the vcache, this function is called before attachment,
794 * therefore it cannot perform any actions on the vnode itself */
797 afs_PrePopulateVCache(struct vcache *avc, struct VenusFid *afid,
798 struct server *serverp) {
801 slot = avc->diskSlot;
803 osi_PrePopulateVCache(avc);
805 avc->diskSlot = slot;
806 QZero(&avc->metadirty);
808 AFS_RWLOCK_INIT(&avc->lock, "vcache lock");
811 avc->linkData = NULL;
814 avc->execsOrWriters = 0;
816 avc->f.states = CVInit;
817 avc->last_looker = 0;
819 avc->asynchrony = -1;
823 avc->f.truncPos = AFS_NOTRUNC; /* don't truncate until we need to */
824 hzero(avc->f.m.DataVersion); /* in case we copy it into flushDV */
826 avc->callback = serverp; /* to minimize chance that clear
829 #if defined(AFS_CACHE_BYPASS)
830 avc->cachingStates = 0;
831 avc->cachingTransitions = 0;
836 afs_FlushAllVCaches(void)
839 struct vcache *tvc, *nvc;
841 ObtainWriteLock(&afs_xvcache, 867);
844 for (i = 0; i < VCSIZE; i++) {
845 for (tvc = afs_vhashT[i]; tvc; tvc = nvc) {
849 if (afs_FlushVCache(tvc, &slept)) {
850 afs_warn("Failed to flush vcache 0x%lx\n", (unsigned long)(uintptrsz)tvc);
859 ReleaseWriteLock(&afs_xvcache);
863 * This routine is responsible for allocating a new cache entry
864 * from the free list. It formats the cache entry and inserts it
865 * into the appropriate hash tables. It must be called with
866 * afs_xvcache write-locked so as to prevent several processes from
867 * trying to create a new cache entry simultaneously.
869 * LOCK: afs_NewVCache afs_xvcache W
871 * \param afid The file id of the file whose cache entry is being created.
873 * \return The new vcache struct.
876 static_inline struct vcache *
877 afs_NewVCache_int(struct VenusFid *afid, struct server *serverp, int seq)
881 afs_int32 anumber = VCACHE_FREE;
883 AFS_STATCNT(afs_NewVCache);
885 afs_FlushReclaimedVcaches();
887 #if defined(AFS_LINUX22_ENV)
888 if(!afsd_dynamic_vcaches && afs_vcount >= afs_maxvcount) {
889 afs_ShakeLooseVCaches(anumber);
890 if (afs_vcount >= afs_maxvcount) {
891 afs_warn("afs_NewVCache - none freed\n");
895 tvc = afs_AllocVCache();
896 #else /* AFS_LINUX22_ENV */
897 /* pull out a free cache entry */
899 afs_ShakeLooseVCaches(anumber);
903 tvc = afs_AllocVCache();
905 tvc = freeVCList; /* take from free list */
906 freeVCList = tvc->nextfree;
907 tvc->nextfree = NULL;
908 afs_vcount++; /* balanced by FlushVCache */
909 } /* end of if (!freeVCList) */
911 #endif /* AFS_LINUX22_ENV */
913 #if defined(AFS_XBSD_ENV) || defined(AFS_DARWIN_ENV)
915 panic("afs_NewVCache(): free vcache with vnode attached");
918 /* Populate the vcache with as much as we can. */
919 afs_PrePopulateVCache(tvc, afid, serverp);
921 /* Thread the vcache onto the VLRU */
926 tvc->hnext = afs_vhashT[i];
928 QAdd(&afs_vhashTV[j], &tvc->vhashq);
930 if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) {
931 refpanic("NewVCache VLRU inconsistent");
933 QAdd(&VLRU, &tvc->vlruq); /* put in lruq */
934 if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) {
935 refpanic("NewVCache VLRU inconsistent2");
937 if (tvc->vlruq.next->prev != &(tvc->vlruq)) {
938 refpanic("NewVCache VLRU inconsistent3");
940 if (tvc->vlruq.prev->next != &(tvc->vlruq)) {
941 refpanic("NewVCache VLRU inconsistent4");
945 /* it should now be safe to drop the xvcache lock - so attach an inode
946 * to this vcache, where necessary */
947 osi_AttachVnode(tvc, seq);
949 /* Get a reference count to hold this vcache for the VLRUQ. Note that
950 * we have to do this after attaching the vnode, because the reference
951 * count may be held in the vnode itself */
953 #if defined(AFS_LINUX22_ENV)
954 /* Hold it for the LRU (should make count 2) */
956 #elif !(defined (AFS_DARWIN_ENV) || defined(AFS_XBSD_ENV))
957 VREFCOUNT_SET(tvc, 1); /* us */
960 #if defined (AFS_FBSD_ENV)
961 if (tvc->f.states & CVInit)
963 afs_PostPopulateVCache(tvc, afid, seq);
970 afs_NewVCache(struct VenusFid *afid, struct server *serverp)
972 return afs_NewVCache_int(afid, serverp, 0);
976 afs_NewBulkVCache(struct VenusFid *afid, struct server *serverp, int seq)
978 return afs_NewVCache_int(afid, serverp, seq);
984 * LOCK: afs_FlushActiveVcaches afs_xvcache N
986 * \param doflocks : Do we handle flocks?
989 afs_FlushActiveVcaches(afs_int32 doflocks)
995 afs_ucred_t *cred = NULL;
996 struct vrequest treq, ureq;
997 struct AFSVolSync tsync;
1000 AFS_STATCNT(afs_FlushActiveVcaches);
1001 ObtainReadLock(&afs_xvcache);
1002 for (i = 0; i < VCSIZE; i++) {
1003 for (tvc = afs_vhashT[i]; tvc; tvc = tvc->hnext) {
1004 if (tvc->f.states & CVInit) continue;
1005 #ifdef AFS_DARWIN80_ENV
1006 if (tvc->f.states & CDeadVnode &&
1007 (tvc->f.states & (CCore|CUnlinkedDel) ||
1008 tvc->flockCount)) panic("Dead vnode has core/unlinkedel/flock");
1010 if (doflocks && tvc->flockCount != 0) {
1011 struct rx_connection *rxconn;
1012 /* if this entry has an flock, send a keep-alive call out */
1014 ReleaseReadLock(&afs_xvcache);
1015 ObtainWriteLock(&tvc->lock, 51);
1017 afs_InitReq(&treq, afs_osi_credp);
1018 treq.flags |= O_NONBLOCK;
1020 tc = afs_Conn(&tvc->f.fid, &treq, SHARED_LOCK, &rxconn);
1022 XSTATS_START_TIME(AFS_STATS_FS_RPCIDX_EXTENDLOCK);
1025 RXAFS_ExtendLock(rxconn,
1026 (struct AFSFid *)&tvc->f.fid.Fid,
1032 } while (afs_Analyze
1033 (tc, rxconn, code, &tvc->f.fid, &treq,
1034 AFS_STATS_FS_RPCIDX_EXTENDLOCK, SHARED_LOCK, NULL));
1036 ReleaseWriteLock(&tvc->lock);
1037 #ifdef AFS_DARWIN80_ENV
1039 ObtainReadLock(&afs_xvcache);
1041 ObtainReadLock(&afs_xvcache);
1046 if ((tvc->f.states & CCore) || (tvc->f.states & CUnlinkedDel)) {
1048 * Don't let it evaporate in case someone else is in
1049 * this code. Also, drop the afs_xvcache lock while
1050 * getting vcache locks.
1053 ReleaseReadLock(&afs_xvcache);
1054 #ifdef AFS_BOZONLOCK_ENV
1055 afs_BozonLock(&tvc->pvnLock, tvc);
1057 #if defined(AFS_SGI_ENV)
1059 * That's because if we come in via the CUnlinkedDel bit state path we'll be have 0 refcnt
1061 osi_Assert(VREFCOUNT_GT(tvc,0));
1062 AFS_RWLOCK((vnode_t *) tvc, VRWLOCK_WRITE);
1064 ObtainWriteLock(&tvc->lock, 52);
1065 if (tvc->f.states & CCore) {
1066 tvc->f.states &= ~CCore;
1067 /* XXXX Find better place-holder for cred XXXX */
1068 cred = (afs_ucred_t *)tvc->linkData;
1069 tvc->linkData = NULL; /* XXX */
1070 afs_InitReq(&ureq, cred);
1071 afs_Trace2(afs_iclSetp, CM_TRACE_ACTCCORE,
1072 ICL_TYPE_POINTER, tvc, ICL_TYPE_INT32,
1073 tvc->execsOrWriters);
1074 code = afs_StoreOnLastReference(tvc, &ureq);
1075 ReleaseWriteLock(&tvc->lock);
1076 #ifdef AFS_BOZONLOCK_ENV
1077 afs_BozonUnlock(&tvc->pvnLock, tvc);
1079 hzero(tvc->flushDV);
1082 if (code && code != VNOVNODE) {
1083 afs_StoreWarn(code, tvc->f.fid.Fid.Volume,
1084 /* /dev/console */ 1);
1086 } else if (tvc->f.states & CUnlinkedDel) {
1090 ReleaseWriteLock(&tvc->lock);
1091 #ifdef AFS_BOZONLOCK_ENV
1092 afs_BozonUnlock(&tvc->pvnLock, tvc);
1094 #if defined(AFS_SGI_ENV)
1095 AFS_RWUNLOCK((vnode_t *) tvc, VRWLOCK_WRITE);
1097 afs_remunlink(tvc, 0);
1098 #if defined(AFS_SGI_ENV)
1099 AFS_RWLOCK((vnode_t *) tvc, VRWLOCK_WRITE);
1102 /* lost (or won, perhaps) the race condition */
1103 ReleaseWriteLock(&tvc->lock);
1104 #ifdef AFS_BOZONLOCK_ENV
1105 afs_BozonUnlock(&tvc->pvnLock, tvc);
1108 #if defined(AFS_SGI_ENV)
1109 AFS_RWUNLOCK((vnode_t *) tvc, VRWLOCK_WRITE);
1111 #ifdef AFS_DARWIN80_ENV
1114 AFS_RELE(AFSTOV(tvc));
1115 /* Matches write code setting CCore flag */
1118 ObtainReadLock(&afs_xvcache);
1120 ObtainReadLock(&afs_xvcache);
1123 AFS_RELE(AFSTOV(tvc));
1124 /* Matches write code setting CCore flag */
1131 ReleaseReadLock(&afs_xvcache);
1137 * Make sure a cache entry is up-to-date status-wise.
1139 * NOTE: everywhere that calls this can potentially be sped up
1140 * by checking CStatd first, and avoiding doing the InitReq
1141 * if this is up-to-date.
1143 * Anymore, the only places that call this KNOW already that the
1144 * vcache is not up-to-date, so we don't screw around.
1146 * \param avc : Ptr to vcache entry to verify.
1152 * Make sure a cache entry is up-to-date status-wise.
1154 * NOTE: everywhere that calls this can potentially be sped up
1155 * by checking CStatd first, and avoiding doing the InitReq
1156 * if this is up-to-date.
1158 * Anymore, the only places that call this KNOW already that the
1159 * vcache is not up-to-date, so we don't screw around.
1161 * \param avc Pointer to vcache entry to verify.
1164 * \return 0 for success or other error codes.
1167 afs_VerifyVCache2(struct vcache *avc, struct vrequest *areq)
1171 AFS_STATCNT(afs_VerifyVCache);
1173 /* otherwise we must fetch the status info */
1175 ObtainWriteLock(&avc->lock, 53);
1176 if (avc->f.states & CStatd) {
1177 ReleaseWriteLock(&avc->lock);
1180 ObtainWriteLock(&afs_xcbhash, 461);
1181 avc->f.states &= ~(CStatd | CUnique);
1182 avc->callback = NULL;
1183 afs_DequeueCallback(avc);
1184 ReleaseWriteLock(&afs_xcbhash);
1185 ReleaseWriteLock(&avc->lock);
1187 /* since we've been called back, or the callback has expired,
1188 * it's possible that the contents of this directory, or this
1189 * file's name have changed, thus invalidating the dnlc contents.
1191 if ((avc->f.states & CForeign) || (avc->f.fid.Fid.Vnode & 1))
1192 osi_dnlc_purgedp(avc);
1194 osi_dnlc_purgevp(avc);
1196 /* fetch the status info */
1197 tvc = afs_GetVCache(&avc->f.fid, areq, NULL, avc);
1200 /* Put it back; caller has already incremented vrefCount */
1204 } /*afs_VerifyVCache */
1208 * Simple copy of stat info into cache.
1210 * Callers:as of 1992-04-29, only called by WriteVCache
1212 * \param avc Ptr to vcache entry involved.
1213 * \param astat Ptr to stat info to copy.
1217 afs_SimpleVStat(struct vcache *avc,
1218 struct AFSFetchStatus *astat, struct vrequest *areq)
1221 AFS_STATCNT(afs_SimpleVStat);
1223 #ifdef AFS_64BIT_CLIENT
1224 FillInt64(length, astat->Length_hi, astat->Length);
1225 #else /* AFS_64BIT_CLIENT */
1226 length = astat->Length;
1227 #endif /* AFS_64BIT_CLIENT */
1229 #if defined(AFS_SGI_ENV)
1230 if ((avc->execsOrWriters <= 0) && !afs_DirtyPages(avc)
1231 && !AFS_VN_MAPPED((vnode_t *) avc)) {
1232 osi_Assert((valusema(&avc->vc_rwlock) <= 0)
1233 && (OSI_GET_LOCKID() == avc->vc_rwlockid));
1234 if (length < avc->f.m.Length) {
1235 vnode_t *vp = (vnode_t *) avc;
1237 osi_Assert(WriteLocked(&avc->lock));
1238 ReleaseWriteLock(&avc->lock);
1240 PTOSSVP(vp, (off_t) length, (off_t) MAXLONG);
1242 ObtainWriteLock(&avc->lock, 67);
1247 if (!afs_DirtyPages(avc)) {
1248 /* if actively writing the file, don't fetch over this value */
1249 afs_Trace3(afs_iclSetp, CM_TRACE_SIMPLEVSTAT, ICL_TYPE_POINTER, avc,
1250 ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(avc->f.m.Length),
1251 ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(length));
1252 avc->f.m.Length = length;
1253 avc->f.m.Date = astat->ClientModTime;
1255 avc->f.m.Owner = astat->Owner;
1256 avc->f.m.Group = astat->Group;
1257 avc->f.m.Mode = astat->UnixModeBits;
1258 if (vType(avc) == VREG) {
1259 avc->f.m.Mode |= S_IFREG;
1260 } else if (vType(avc) == VDIR) {
1261 avc->f.m.Mode |= S_IFDIR;
1262 } else if (vType(avc) == VLNK) {
1263 avc->f.m.Mode |= S_IFLNK;
1264 if ((avc->f.m.Mode & 0111) == 0)
1267 if (avc->f.states & CForeign) {
1268 struct axscache *ac;
1269 avc->f.anyAccess = astat->AnonymousAccess;
1271 if ((astat->CallerAccess & ~astat->AnonymousAccess))
1273 * Caller has at least one bit not covered by anonymous, and
1274 * thus may have interesting rights.
1276 * HOWEVER, this is a really bad idea, because any access query
1277 * for bits which aren't covered by anonymous, on behalf of a user
1278 * who doesn't have any special rights, will result in an answer of
1279 * the form "I don't know, lets make a FetchStatus RPC and find out!"
1280 * It's an especially bad idea under Ultrix, since (due to the lack of
1281 * a proper access() call) it must perform several afs_access() calls
1282 * in order to create magic mode bits that vary according to who makes
1283 * the call. In other words, _every_ stat() generates a test for
1286 #endif /* badidea */
1287 if (avc->Access && (ac = afs_FindAxs(avc->Access, areq->uid)))
1288 ac->axess = astat->CallerAccess;
1289 else /* not found, add a new one if possible */
1290 afs_AddAxs(avc->Access, areq->uid, astat->CallerAccess);
1293 } /*afs_SimpleVStat */
1297 * Store the status info *only* back to the server for a
1300 * Environment: Must be called with a shared lock held on the vnode.
1302 * \param avc Ptr to the vcache entry.
1303 * \param astatus Ptr to the status info to store.
1304 * \param areq Ptr to the associated vrequest.
1306 * \return Operation status.
1310 afs_WriteVCache(struct vcache *avc,
1311 struct AFSStoreStatus *astatus,
1312 struct vrequest *areq)
1315 struct afs_conn *tc;
1316 struct AFSFetchStatus OutStatus;
1317 struct AFSVolSync tsync;
1318 struct rx_connection *rxconn;
1320 AFS_STATCNT(afs_WriteVCache);
1321 afs_Trace2(afs_iclSetp, CM_TRACE_WVCACHE, ICL_TYPE_POINTER, avc,
1322 ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(avc->f.m.Length));
1324 tc = afs_Conn(&avc->f.fid, areq, SHARED_LOCK, &rxconn);
1326 XSTATS_START_TIME(AFS_STATS_FS_RPCIDX_STORESTATUS);
1329 RXAFS_StoreStatus(rxconn, (struct AFSFid *)&avc->f.fid.Fid,
1330 astatus, &OutStatus, &tsync);
1335 } while (afs_Analyze
1336 (tc, rxconn, code, &avc->f.fid, areq, AFS_STATS_FS_RPCIDX_STORESTATUS,
1337 SHARED_LOCK, NULL));
1339 UpgradeSToWLock(&avc->lock, 20);
1341 /* success, do the changes locally */
1342 afs_SimpleVStat(avc, &OutStatus, areq);
1344 * Update the date, too. SimpleVStat didn't do this, since
1345 * it thought we were doing this after fetching new status
1346 * over a file being written.
1348 avc->f.m.Date = OutStatus.ClientModTime;
1350 /* failure, set up to check with server next time */
1351 ObtainWriteLock(&afs_xcbhash, 462);
1352 afs_DequeueCallback(avc);
1353 avc->f.states &= ~(CStatd | CUnique); /* turn off stat valid flag */
1354 ReleaseWriteLock(&afs_xcbhash);
1355 if ((avc->f.states & CForeign) || (avc->f.fid.Fid.Vnode & 1))
1356 osi_dnlc_purgedp(avc); /* if it (could be) a directory */
1358 ConvertWToSLock(&avc->lock);
1361 } /*afs_WriteVCache */
1364 * Store status info only locally, set the proper disconnection flags
1365 * and add to dirty list.
1367 * \param avc The vcache to be written locally.
1368 * \param astatus Get attr fields from local store.
1369 * \param attrs This one is only of the vs_size.
1371 * \note Must be called with a shared lock on the vnode
1374 afs_WriteVCacheDiscon(struct vcache *avc,
1375 struct AFSStoreStatus *astatus,
1376 struct vattr *attrs)
1379 afs_int32 flags = 0;
1381 UpgradeSToWLock(&avc->lock, 700);
1383 if (!astatus->Mask) {
1389 /* Set attributes. */
1390 if (astatus->Mask & AFS_SETMODTIME) {
1391 avc->f.m.Date = astatus->ClientModTime;
1392 flags |= VDisconSetTime;
1395 if (astatus->Mask & AFS_SETOWNER) {
1396 /* printf("Not allowed yet. \n"); */
1397 /*avc->f.m.Owner = astatus->Owner;*/
1400 if (astatus->Mask & AFS_SETGROUP) {
1401 /* printf("Not allowed yet. \n"); */
1402 /*avc->f.m.Group = astatus->Group;*/
1405 if (astatus->Mask & AFS_SETMODE) {
1406 avc->f.m.Mode = astatus->UnixModeBits;
1408 #if 0 /* XXX: Leaving this out, so it doesn't mess up the file type flag.*/
1410 if (vType(avc) == VREG) {
1411 avc->f.m.Mode |= S_IFREG;
1412 } else if (vType(avc) == VDIR) {
1413 avc->f.m.Mode |= S_IFDIR;
1414 } else if (vType(avc) == VLNK) {
1415 avc->f.m.Mode |= S_IFLNK;
1416 if ((avc->f.m.Mode & 0111) == 0)
1420 flags |= VDisconSetMode;
1421 } /* if(astatus.Mask & AFS_SETMODE) */
1423 } /* if (!astatus->Mask) */
1425 if (attrs->va_size > 0) {
1426 /* XXX: Do I need more checks? */
1427 /* Truncation operation. */
1428 flags |= VDisconTrunc;
1432 afs_DisconAddDirty(avc, flags, 1);
1434 /* XXX: How about the rest of the fields? */
1436 ConvertWToSLock(&avc->lock);
1442 * Copy astat block into vcache info
1444 * \note This code may get dataversion and length out of sync if the file has
1445 * been modified. This is less than ideal. I haven't thought about it sufficiently
1446 * to be certain that it is adequate.
1448 * \note Environment: Must be called under a write lock
1450 * \param avc Ptr to vcache entry.
1451 * \param astat Ptr to stat block to copy in.
1452 * \param areq Ptr to associated request.
1455 afs_ProcessFS(struct vcache *avc,
1456 struct AFSFetchStatus *astat, struct vrequest *areq)
1459 AFS_STATCNT(afs_ProcessFS);
1461 #ifdef AFS_64BIT_CLIENT
1462 FillInt64(length, astat->Length_hi, astat->Length);
1463 #else /* AFS_64BIT_CLIENT */
1464 length = astat->Length;
1465 #endif /* AFS_64BIT_CLIENT */
1466 /* WARNING: afs_DoBulkStat uses the Length field to store a sequence
1467 * number for each bulk status request. Under no circumstances
1468 * should afs_DoBulkStat store a sequence number if the new
1469 * length will be ignored when afs_ProcessFS is called with
1470 * new stats. If you change the following conditional then you
1471 * also need to change the conditional in afs_DoBulkStat. */
1473 if ((avc->execsOrWriters <= 0) && !afs_DirtyPages(avc)
1474 && !AFS_VN_MAPPED((vnode_t *) avc)) {
1476 if ((avc->execsOrWriters <= 0) && !afs_DirtyPages(avc)) {
1478 /* if we're writing or mapping this file, don't fetch over these
1481 afs_Trace3(afs_iclSetp, CM_TRACE_PROCESSFS, ICL_TYPE_POINTER, avc,
1482 ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(avc->f.m.Length),
1483 ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(length));
1484 avc->f.m.Length = length;
1485 avc->f.m.Date = astat->ClientModTime;
1487 hset64(avc->f.m.DataVersion, astat->dataVersionHigh, astat->DataVersion);
1488 avc->f.m.Owner = astat->Owner;
1489 avc->f.m.Mode = astat->UnixModeBits;
1490 avc->f.m.Group = astat->Group;
1491 avc->f.m.LinkCount = astat->LinkCount;
1492 if (astat->FileType == File) {
1493 vSetType(avc, VREG);
1494 avc->f.m.Mode |= S_IFREG;
1495 } else if (astat->FileType == Directory) {
1496 vSetType(avc, VDIR);
1497 avc->f.m.Mode |= S_IFDIR;
1498 } else if (astat->FileType == SymbolicLink) {
1499 if (afs_fakestat_enable && (avc->f.m.Mode & 0111) == 0) {
1500 vSetType(avc, VDIR);
1501 avc->f.m.Mode |= S_IFDIR;
1503 vSetType(avc, VLNK);
1504 avc->f.m.Mode |= S_IFLNK;
1506 if ((avc->f.m.Mode & 0111) == 0) {
1510 avc->f.anyAccess = astat->AnonymousAccess;
1512 if ((astat->CallerAccess & ~astat->AnonymousAccess))
1514 * Caller has at least one bit not covered by anonymous, and
1515 * thus may have interesting rights.
1517 * HOWEVER, this is a really bad idea, because any access query
1518 * for bits which aren't covered by anonymous, on behalf of a user
1519 * who doesn't have any special rights, will result in an answer of
1520 * the form "I don't know, lets make a FetchStatus RPC and find out!"
1521 * It's an especially bad idea under Ultrix, since (due to the lack of
1522 * a proper access() call) it must perform several afs_access() calls
1523 * in order to create magic mode bits that vary according to who makes
1524 * the call. In other words, _every_ stat() generates a test for
1527 #endif /* badidea */
1529 struct axscache *ac;
1530 if (avc->Access && (ac = afs_FindAxs(avc->Access, areq->uid)))
1531 ac->axess = astat->CallerAccess;
1532 else /* not found, add a new one if possible */
1533 afs_AddAxs(avc->Access, areq->uid, astat->CallerAccess);
1535 } /*afs_ProcessFS */
1539 * Get fid from server.
1542 * \param areq Request to be passed on.
1543 * \param name Name of ?? to lookup.
1544 * \param OutStatus Fetch status.
1549 * \return Success status of operation.
1552 afs_RemoteLookup(struct VenusFid *afid, struct vrequest *areq,
1553 char *name, struct VenusFid *nfid,
1554 struct AFSFetchStatus *OutStatusp,
1555 struct AFSCallBack *CallBackp, struct server **serverp,
1556 struct AFSVolSync *tsyncp)
1559 struct afs_conn *tc;
1560 struct rx_connection *rxconn;
1561 struct AFSFetchStatus OutDirStatus;
1564 name = ""; /* XXX */
1566 tc = afs_Conn(afid, areq, SHARED_LOCK, &rxconn);
1569 *serverp = tc->parent->srvr->server;
1570 XSTATS_START_TIME(AFS_STATS_FS_RPCIDX_XLOOKUP);
1573 RXAFS_Lookup(rxconn, (struct AFSFid *)&afid->Fid, name,
1574 (struct AFSFid *)&nfid->Fid, OutStatusp,
1575 &OutDirStatus, CallBackp, tsyncp);
1580 } while (afs_Analyze
1581 (tc, rxconn, code, afid, areq, AFS_STATS_FS_RPCIDX_XLOOKUP, SHARED_LOCK,
1591 * Given a file id and a vrequest structure, fetch the status
1592 * information associated with the file.
1594 * \param afid File ID.
1595 * \param areq Ptr to associated vrequest structure, specifying the
1596 * user whose authentication tokens will be used.
1597 * \param avc Caller may already have a vcache for this file, which is
1600 * \note Environment:
1601 * The cache entry is returned with an increased vrefCount field.
1602 * The entry must be discarded by calling afs_PutVCache when you
1603 * are through using the pointer to the cache entry.
1605 * You should not hold any locks when calling this function, except
1606 * locks on other vcache entries. If you lock more than one vcache
1607 * entry simultaneously, you should lock them in this order:
1609 * 1. Lock all files first, then directories.
1610 * 2. Within a particular type, lock entries in Fid.Vnode order.
1612 * This locking hierarchy is convenient because it allows locking
1613 * of a parent dir cache entry, given a file (to check its access
1614 * control list). It also allows renames to be handled easily by
1615 * locking directories in a constant order.
1617 * \note NB. NewVCache -> FlushVCache presently (4/10/95) drops the xvcache lock.
1619 * \note Might have a vcache structure already, which must
1620 * already be held by the caller
1623 afs_GetVCache(struct VenusFid *afid, struct vrequest *areq,
1624 afs_int32 * cached, struct vcache *avc)
1627 afs_int32 code, newvcache = 0;
1632 AFS_STATCNT(afs_GetVCache);
1635 *cached = 0; /* Init just in case */
1637 #if defined(AFS_SGI_ENV) && !defined(AFS_SGI53_ENV)
1641 ObtainSharedLock(&afs_xvcache, 5);
1643 tvc = afs_FindVCache(afid, &retry, DO_STATS | DO_VLRU | IS_SLOCK);
1645 #if defined(AFS_SGI_ENV) && !defined(AFS_SGI53_ENV)
1646 ReleaseSharedLock(&afs_xvcache);
1647 spunlock_psema(tvc->v.v_lock, retry, &tvc->v.v_sync, PINOD);
1654 osi_Assert((tvc->f.states & CVInit) == 0);
1655 /* If we are in readdir, return the vnode even if not statd */
1656 if ((tvc->f.states & CStatd) || afs_InReadDir(tvc)) {
1657 ReleaseSharedLock(&afs_xvcache);
1661 UpgradeSToWLock(&afs_xvcache, 21);
1663 /* no cache entry, better grab one */
1664 tvc = afs_NewVCache(afid, NULL);
1667 ConvertWToSLock(&afs_xvcache);
1670 ReleaseSharedLock(&afs_xvcache);
1674 afs_stats_cmperf.vcacheMisses++;
1677 ReleaseSharedLock(&afs_xvcache);
1679 ObtainWriteLock(&tvc->lock, 54);
1681 if (tvc->f.states & CStatd) {
1682 ReleaseWriteLock(&tvc->lock);
1685 #ifdef AFS_DARWIN80_ENV
1686 /* Darwin 8.0 only has bufs in nfs, so we shouldn't have to worry about them.
1689 #if defined(AFS_DARWIN_ENV) || defined(AFS_FBSD_ENV)
1691 * XXX - I really don't like this. Should try to understand better.
1692 * It seems that sometimes, when we get called, we already hold the
1693 * lock on the vnode (e.g., from afs_getattr via afs_VerifyVCache).
1694 * We can't drop the vnode lock, because that could result in a race.
1695 * Sometimes, though, we get here and don't hold the vnode lock.
1696 * I hate code paths that sometimes hold locks and sometimes don't.
1697 * In any event, the dodge we use here is to check whether the vnode
1698 * is locked, and if it isn't, then we gain and drop it around the call
1699 * to vinvalbuf; otherwise, we leave it alone.
1702 struct vnode *vp = AFSTOV(tvc);
1705 #if defined(AFS_DARWIN_ENV)
1706 iheldthelock = VOP_ISLOCKED(vp);
1708 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, current_proc());
1709 /* this is messy. we can call fsync which will try to reobtain this */
1710 if (VTOAFS(vp) == tvc)
1711 ReleaseWriteLock(&tvc->lock);
1712 if (UBCINFOEXISTS(vp)) {
1713 vinvalbuf(vp, V_SAVE, &afs_osi_cred, current_proc(), PINOD, 0);
1715 if (VTOAFS(vp) == tvc)
1716 ObtainWriteLock(&tvc->lock, 954);
1718 VOP_UNLOCK(vp, LK_EXCLUSIVE, current_proc());
1719 #elif defined(AFS_FBSD80_ENV)
1720 iheldthelock = VOP_ISLOCKED(vp);
1721 if (!iheldthelock) {
1722 /* nosleep/sleep lock order reversal */
1723 int glocked = ISAFS_GLOCK();
1726 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1730 vinvalbuf(vp, V_SAVE, PINOD, 0); /* changed late in 8.0-CURRENT */
1733 #elif defined(AFS_FBSD60_ENV)
1734 iheldthelock = VOP_ISLOCKED(vp, curthread);
1736 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, curthread);
1738 vinvalbuf(vp, V_SAVE, curthread, PINOD, 0);
1741 VOP_UNLOCK(vp, LK_EXCLUSIVE, curthread);
1742 #elif defined(AFS_FBSD_ENV)
1743 iheldthelock = VOP_ISLOCKED(vp, curthread);
1745 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, curthread);
1746 vinvalbuf(vp, V_SAVE, osi_curcred(), curthread, PINOD, 0);
1748 VOP_UNLOCK(vp, LK_EXCLUSIVE, curthread);
1749 #elif defined(AFS_OBSD_ENV)
1750 iheldthelock = VOP_ISLOCKED(vp, curproc);
1752 VOP_LOCK(vp, LK_EXCLUSIVE | LK_RETRY, curproc);
1753 uvm_vnp_uncache(vp);
1755 VOP_UNLOCK(vp, 0, curproc);
1756 #elif defined(AFS_NBSD40_ENV)
1757 iheldthelock = VOP_ISLOCKED(vp);
1758 if (!iheldthelock) {
1759 VOP_LOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1761 uvm_vnp_uncache(vp);
1769 ObtainWriteLock(&afs_xcbhash, 464);
1770 tvc->f.states &= ~CUnique;
1772 afs_DequeueCallback(tvc);
1773 ReleaseWriteLock(&afs_xcbhash);
1775 /* It is always appropriate to throw away all the access rights? */
1776 afs_FreeAllAxs(&(tvc->Access));
1777 tvp = afs_GetVolume(afid, areq, READ_LOCK); /* copy useful per-volume info */
1779 if ((tvp->states & VForeign)) {
1781 tvc->f.states |= CForeign;
1782 if (newvcache && (tvp->rootVnode == afid->Fid.Vnode)
1783 && (tvp->rootUnique == afid->Fid.Unique)) {
1787 if (tvp->states & VRO)
1788 tvc->f.states |= CRO;
1789 if (tvp->states & VBackup)
1790 tvc->f.states |= CBackup;
1791 /* now copy ".." entry back out of volume structure, if necessary */
1792 if (tvc->mvstat == 2 && tvp->dotdot.Fid.Volume != 0) {
1794 tvc->mvid = (struct VenusFid *)
1795 osi_AllocSmallSpace(sizeof(struct VenusFid));
1796 *tvc->mvid = tvp->dotdot;
1798 afs_PutVolume(tvp, READ_LOCK);
1802 afs_RemoveVCB(afid);
1804 struct AFSFetchStatus OutStatus;
1806 if (afs_DynrootNewVnode(tvc, &OutStatus)) {
1807 afs_ProcessFS(tvc, &OutStatus, areq);
1808 tvc->f.states |= CStatd | CUnique;
1809 tvc->f.parent.vnode = OutStatus.ParentVnode;
1810 tvc->f.parent.unique = OutStatus.ParentUnique;
1814 if (AFS_IS_DISCONNECTED) {
1815 /* Nothing to do otherwise...*/
1817 /* printf("Network is down in afs_GetCache"); */
1819 code = afs_FetchStatus(tvc, afid, areq, &OutStatus);
1821 /* For the NFS translator's benefit, make sure
1822 * non-directory vnodes always have their parent FID set
1823 * correctly, even when created as a result of decoding an
1824 * NFS filehandle. It would be nice to also do this for
1825 * directories, but we can't because the fileserver fills
1826 * in the FID of the directory itself instead of that of
1829 if (!code && OutStatus.FileType != Directory &&
1830 !tvc->f.parent.vnode) {
1831 tvc->f.parent.vnode = OutStatus.ParentVnode;
1832 tvc->f.parent.unique = OutStatus.ParentUnique;
1833 /* XXX - SXW - It's conceivable we should mark ourselves
1834 * as dirty again here, incase we've been raced
1835 * out of the FetchStatus call.
1842 ReleaseWriteLock(&tvc->lock);
1848 ReleaseWriteLock(&tvc->lock);
1851 } /*afs_GetVCache */
1856 * Lookup a vcache by fid. Look inside the cache first, if not
1857 * there, lookup the file on the server, and then get it's fresh
1862 * \param cached Is element cached? If NULL, don't answer.
1866 * \return The found element or NULL.
1869 afs_LookupVCache(struct VenusFid *afid, struct vrequest *areq,
1870 afs_int32 * cached, struct vcache *adp, char *aname)
1872 afs_int32 code, now, newvcache = 0;
1873 struct VenusFid nfid;
1876 struct AFSFetchStatus OutStatus;
1877 struct AFSCallBack CallBack;
1878 struct AFSVolSync tsync;
1879 struct server *serverp = 0;
1883 AFS_STATCNT(afs_GetVCache);
1885 *cached = 0; /* Init just in case */
1887 #if defined(AFS_SGI_ENV) && !defined(AFS_SGI53_ENV)
1891 ObtainReadLock(&afs_xvcache);
1892 tvc = afs_FindVCache(afid, &retry, DO_STATS /* no vlru */ );
1895 ReleaseReadLock(&afs_xvcache);
1897 #if defined(AFS_SGI_ENV) && !defined(AFS_SGI53_ENV)
1898 spunlock_psema(tvc->v.v_lock, retry, &tvc->v.v_sync, PINOD);
1902 ObtainReadLock(&tvc->lock);
1904 if (tvc->f.states & CStatd) {
1908 ReleaseReadLock(&tvc->lock);
1911 tvc->f.states &= ~CUnique;
1913 ReleaseReadLock(&tvc->lock);
1915 ObtainReadLock(&afs_xvcache);
1918 ReleaseReadLock(&afs_xvcache);
1920 /* lookup the file */
1923 origCBs = afs_allCBs; /* if anything changes, we don't have a cb */
1925 if (AFS_IS_DISCONNECTED) {
1926 /* printf("Network is down in afs_LookupVcache\n"); */
1930 afs_RemoteLookup(&adp->f.fid, areq, aname, &nfid, &OutStatus,
1931 &CallBack, &serverp, &tsync);
1933 #if defined(AFS_SGI_ENV) && !defined(AFS_SGI53_ENV)
1937 ObtainSharedLock(&afs_xvcache, 6);
1938 tvc = afs_FindVCache(&nfid, &retry, DO_VLRU | IS_SLOCK/* no xstats now */ );
1940 #if defined(AFS_SGI_ENV) && !defined(AFS_SGI53_ENV)
1941 ReleaseSharedLock(&afs_xvcache);
1942 spunlock_psema(tvc->v.v_lock, retry, &tvc->v.v_sync, PINOD);
1948 /* no cache entry, better grab one */
1949 UpgradeSToWLock(&afs_xvcache, 22);
1950 tvc = afs_NewVCache(&nfid, serverp);
1952 ConvertWToSLock(&afs_xvcache);
1955 ReleaseSharedLock(&afs_xvcache);
1960 ReleaseSharedLock(&afs_xvcache);
1961 ObtainWriteLock(&tvc->lock, 55);
1963 /* It is always appropriate to throw away all the access rights? */
1964 afs_FreeAllAxs(&(tvc->Access));
1965 tvp = afs_GetVolume(afid, areq, READ_LOCK); /* copy useful per-vol info */
1967 if ((tvp->states & VForeign)) {
1969 tvc->f.states |= CForeign;
1970 if (newvcache && (tvp->rootVnode == afid->Fid.Vnode)
1971 && (tvp->rootUnique == afid->Fid.Unique))
1974 if (tvp->states & VRO)
1975 tvc->f.states |= CRO;
1976 if (tvp->states & VBackup)
1977 tvc->f.states |= CBackup;
1978 /* now copy ".." entry back out of volume structure, if necessary */
1979 if (tvc->mvstat == 2 && tvp->dotdot.Fid.Volume != 0) {
1981 tvc->mvid = (struct VenusFid *)
1982 osi_AllocSmallSpace(sizeof(struct VenusFid));
1983 *tvc->mvid = tvp->dotdot;
1988 ObtainWriteLock(&afs_xcbhash, 465);
1989 afs_DequeueCallback(tvc);
1990 tvc->f.states &= ~(CStatd | CUnique);
1991 ReleaseWriteLock(&afs_xcbhash);
1992 if ((tvc->f.states & CForeign) || (tvc->f.fid.Fid.Vnode & 1))
1993 osi_dnlc_purgedp(tvc); /* if it (could be) a directory */
1995 afs_PutVolume(tvp, READ_LOCK);
1996 ReleaseWriteLock(&tvc->lock);
2001 ObtainWriteLock(&afs_xcbhash, 466);
2002 if (origCBs == afs_allCBs) {
2003 if (CallBack.ExpirationTime) {
2004 tvc->callback = serverp;
2005 tvc->cbExpires = CallBack.ExpirationTime + now;
2006 tvc->f.states |= CStatd | CUnique;
2007 tvc->f.states &= ~CBulkFetching;
2008 afs_QueueCallback(tvc, CBHash(CallBack.ExpirationTime), tvp);
2009 } else if (tvc->f.states & CRO) {
2010 /* adapt gives us an hour. */
2011 tvc->cbExpires = 3600 + osi_Time();
2012 /*XXX*/ tvc->f.states |= CStatd | CUnique;
2013 tvc->f.states &= ~CBulkFetching;
2014 afs_QueueCallback(tvc, CBHash(3600), tvp);
2016 tvc->callback = NULL;
2017 afs_DequeueCallback(tvc);
2018 tvc->f.states &= ~(CStatd | CUnique);
2019 if ((tvc->f.states & CForeign) || (tvc->f.fid.Fid.Vnode & 1))
2020 osi_dnlc_purgedp(tvc); /* if it (could be) a directory */
2023 afs_DequeueCallback(tvc);
2024 tvc->f.states &= ~CStatd;
2025 tvc->f.states &= ~CUnique;
2026 tvc->callback = NULL;
2027 if ((tvc->f.states & CForeign) || (tvc->f.fid.Fid.Vnode & 1))
2028 osi_dnlc_purgedp(tvc); /* if it (could be) a directory */
2030 ReleaseWriteLock(&afs_xcbhash);
2032 afs_PutVolume(tvp, READ_LOCK);
2033 afs_ProcessFS(tvc, &OutStatus, areq);
2035 ReleaseWriteLock(&tvc->lock);
2041 afs_GetRootVCache(struct VenusFid *afid, struct vrequest *areq,
2042 afs_int32 * cached, struct volume *tvolp)
2044 afs_int32 code = 0, i, newvcache = 0, haveStatus = 0;
2045 afs_int32 getNewFid = 0;
2047 struct VenusFid nfid;
2049 struct server *serverp = 0;
2050 struct AFSFetchStatus OutStatus;
2051 struct AFSCallBack CallBack;
2052 struct AFSVolSync tsync;
2054 #ifdef AFS_DARWIN80_ENV
2061 if (!tvolp->rootVnode || getNewFid) {
2062 struct VenusFid tfid;
2065 tfid.Fid.Vnode = 0; /* Means get rootfid of volume */
2066 origCBs = afs_allCBs; /* ignore InitCallBackState */
2068 afs_RemoteLookup(&tfid, areq, NULL, &nfid, &OutStatus, &CallBack,
2073 /* ReleaseReadLock(&tvolp->lock); */
2074 ObtainWriteLock(&tvolp->lock, 56);
2075 tvolp->rootVnode = afid->Fid.Vnode = nfid.Fid.Vnode;
2076 tvolp->rootUnique = afid->Fid.Unique = nfid.Fid.Unique;
2077 ReleaseWriteLock(&tvolp->lock);
2078 /* ObtainReadLock(&tvolp->lock);*/
2081 afid->Fid.Vnode = tvolp->rootVnode;
2082 afid->Fid.Unique = tvolp->rootUnique;
2086 ObtainSharedLock(&afs_xvcache, 7);
2088 for (tvc = afs_vhashT[i]; tvc; tvc = tvc->hnext) {
2089 if (!FidCmp(&(tvc->f.fid), afid)) {
2090 if (tvc->f.states & CVInit) {
2091 ReleaseSharedLock(&afs_xvcache);
2092 afs_osi_Sleep(&tvc->f.states);
2095 #ifdef AFS_DARWIN80_ENV
2096 if (tvc->f.states & CDeadVnode) {
2097 ReleaseSharedLock(&afs_xvcache);
2098 afs_osi_Sleep(&tvc->f.states);
2102 if (vnode_get(tvp)) /* this bumps ref count */
2104 if (vnode_ref(tvp)) {
2106 /* AFSTOV(tvc) may be NULL */
2116 if (!haveStatus && (!tvc || !(tvc->f.states & CStatd))) {
2117 /* Mount point no longer stat'd or unknown. FID may have changed. */
2119 ReleaseSharedLock(&afs_xvcache);
2120 #ifdef AFS_DARWIN80_ENV
2123 vnode_put(AFSTOV(tvc));
2124 vnode_rele(AFSTOV(tvc));
2133 UpgradeSToWLock(&afs_xvcache, 23);
2134 /* no cache entry, better grab one */
2135 tvc = afs_NewVCache(afid, NULL);
2138 ReleaseWriteLock(&afs_xvcache);
2142 afs_stats_cmperf.vcacheMisses++;
2146 afs_stats_cmperf.vcacheHits++;
2147 #if defined(AFS_DARWIN80_ENV)
2148 /* we already bumped the ref count in the for loop above */
2149 #else /* AFS_DARWIN80_ENV */
2152 UpgradeSToWLock(&afs_xvcache, 24);
2153 if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) {
2154 refpanic("GRVC VLRU inconsistent0");
2156 if (tvc->vlruq.next->prev != &(tvc->vlruq)) {
2157 refpanic("GRVC VLRU inconsistent1");
2159 if (tvc->vlruq.prev->next != &(tvc->vlruq)) {
2160 refpanic("GRVC VLRU inconsistent2");
2162 QRemove(&tvc->vlruq); /* move to lruq head */
2163 QAdd(&VLRU, &tvc->vlruq);
2164 if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) {
2165 refpanic("GRVC VLRU inconsistent3");
2167 if (tvc->vlruq.next->prev != &(tvc->vlruq)) {
2168 refpanic("GRVC VLRU inconsistent4");
2170 if (tvc->vlruq.prev->next != &(tvc->vlruq)) {
2171 refpanic("GRVC VLRU inconsistent5");
2176 ReleaseWriteLock(&afs_xvcache);
2178 if (tvc->f.states & CStatd) {
2182 ObtainReadLock(&tvc->lock);
2183 tvc->f.states &= ~CUnique;
2184 tvc->callback = NULL; /* redundant, perhaps */
2185 ReleaseReadLock(&tvc->lock);
2188 ObtainWriteLock(&tvc->lock, 57);
2190 /* It is always appropriate to throw away all the access rights? */
2191 afs_FreeAllAxs(&(tvc->Access));
2194 tvc->f.states |= CForeign;
2195 if (tvolp->states & VRO)
2196 tvc->f.states |= CRO;
2197 if (tvolp->states & VBackup)
2198 tvc->f.states |= CBackup;
2199 /* now copy ".." entry back out of volume structure, if necessary */
2200 if (newvcache && (tvolp->rootVnode == afid->Fid.Vnode)
2201 && (tvolp->rootUnique == afid->Fid.Unique)) {
2204 if (tvc->mvstat == 2 && tvolp->dotdot.Fid.Volume != 0) {
2206 tvc->mvid = (struct VenusFid *)
2207 osi_AllocSmallSpace(sizeof(struct VenusFid));
2208 *tvc->mvid = tvolp->dotdot;
2212 afs_RemoveVCB(afid);
2215 struct VenusFid tfid;
2218 tfid.Fid.Vnode = 0; /* Means get rootfid of volume */
2219 origCBs = afs_allCBs; /* ignore InitCallBackState */
2221 afs_RemoteLookup(&tfid, areq, NULL, &nfid, &OutStatus, &CallBack,
2226 ObtainWriteLock(&afs_xcbhash, 467);
2227 afs_DequeueCallback(tvc);
2228 tvc->callback = NULL;
2229 tvc->f.states &= ~(CStatd | CUnique);
2230 ReleaseWriteLock(&afs_xcbhash);
2231 if ((tvc->f.states & CForeign) || (tvc->f.fid.Fid.Vnode & 1))
2232 osi_dnlc_purgedp(tvc); /* if it (could be) a directory */
2233 ReleaseWriteLock(&tvc->lock);
2238 ObtainWriteLock(&afs_xcbhash, 468);
2239 if (origCBs == afs_allCBs) {
2240 tvc->f.states |= CTruth;
2241 tvc->callback = serverp;
2242 if (CallBack.ExpirationTime != 0) {
2243 tvc->cbExpires = CallBack.ExpirationTime + start;
2244 tvc->f.states |= CStatd;
2245 tvc->f.states &= ~CBulkFetching;
2246 afs_QueueCallback(tvc, CBHash(CallBack.ExpirationTime), tvolp);
2247 } else if (tvc->f.states & CRO) {
2248 /* adapt gives us an hour. */
2249 tvc->cbExpires = 3600 + osi_Time();
2250 /*XXX*/ tvc->f.states |= CStatd;
2251 tvc->f.states &= ~CBulkFetching;
2252 afs_QueueCallback(tvc, CBHash(3600), tvolp);
2255 afs_DequeueCallback(tvc);
2256 tvc->callback = NULL;
2257 tvc->f.states &= ~(CStatd | CUnique);
2258 if ((tvc->f.states & CForeign) || (tvc->f.fid.Fid.Vnode & 1))
2259 osi_dnlc_purgedp(tvc); /* if it (could be) a directory */
2261 ReleaseWriteLock(&afs_xcbhash);
2262 afs_ProcessFS(tvc, &OutStatus, areq);
2264 ReleaseWriteLock(&tvc->lock);
2270 * Update callback status and (sometimes) attributes of a vnode.
2271 * Called after doing a fetch status RPC. Whilst disconnected, attributes
2272 * shouldn't be written to the vcache here.
2277 * \param Outsp Server status after rpc call.
2278 * \param acb Callback for this vnode.
2280 * \note The vcache must be write locked.
2283 afs_UpdateStatus(struct vcache *avc, struct VenusFid *afid,
2284 struct vrequest *areq, struct AFSFetchStatus *Outsp,
2285 struct AFSCallBack *acb, afs_uint32 start)
2287 struct volume *volp;
2290 /* Dont write status in vcache if resyncing after a disconnection. */
2291 afs_ProcessFS(avc, Outsp, areq);
2293 volp = afs_GetVolume(afid, areq, READ_LOCK);
2294 ObtainWriteLock(&afs_xcbhash, 469);
2295 avc->f.states |= CTruth;
2296 if (avc->callback /* check for race */ ) {
2297 if (acb->ExpirationTime != 0) {
2298 avc->cbExpires = acb->ExpirationTime + start;
2299 avc->f.states |= CStatd;
2300 avc->f.states &= ~CBulkFetching;
2301 afs_QueueCallback(avc, CBHash(acb->ExpirationTime), volp);
2302 } else if (avc->f.states & CRO) {
2303 /* ordinary callback on a read-only volume -- AFS 3.2 style */
2304 avc->cbExpires = 3600 + start;
2305 avc->f.states |= CStatd;
2306 avc->f.states &= ~CBulkFetching;
2307 afs_QueueCallback(avc, CBHash(3600), volp);
2309 afs_DequeueCallback(avc);
2310 avc->callback = NULL;
2311 avc->f.states &= ~(CStatd | CUnique);
2312 if ((avc->f.states & CForeign) || (avc->f.fid.Fid.Vnode & 1))
2313 osi_dnlc_purgedp(avc); /* if it (could be) a directory */
2316 afs_DequeueCallback(avc);
2317 avc->callback = NULL;
2318 avc->f.states &= ~(CStatd | CUnique);
2319 if ((avc->f.states & CForeign) || (avc->f.fid.Fid.Vnode & 1))
2320 osi_dnlc_purgedp(avc); /* if it (could be) a directory */
2322 ReleaseWriteLock(&afs_xcbhash);
2324 afs_PutVolume(volp, READ_LOCK);
2328 afs_BadFetchStatus(struct afs_conn *tc)
2330 int addr = ntohl(tc->parent->srvr->sa_ip);
2331 afs_warn("afs: Invalid AFSFetchStatus from server %u.%u.%u.%u\n",
2332 (addr >> 24) & 0xff, (addr >> 16) & 0xff, (addr >> 8) & 0xff,
2334 afs_warn("afs: This suggests the server may be sending bad data that "
2335 "can lead to availability issues or data corruption. The "
2336 "issue has been avoided for now, but it may not always be "
2337 "detectable. Please upgrade the server if possible.\n");
2341 * Check if a given AFSFetchStatus structure is sane.
2343 * @param[in] tc The server from which we received the status
2344 * @param[in] status The status we received
2346 * @return whether the given structure is valid or not
2347 * @retval 0 the structure is fine
2348 * @retval nonzero the structure looks like garbage; act as if we received
2349 * the returned error code from the server
2352 afs_CheckFetchStatus(struct afs_conn *tc, struct AFSFetchStatus *status)
2354 if (status->errorCode ||
2355 status->InterfaceVersion != 1 ||
2356 !(status->FileType > Invalid && status->FileType <= SymbolicLink) ||
2357 status->ParentVnode == 0 || status->ParentUnique == 0) {
2359 afs_warn("afs: FetchStatus ec %u iv %u ft %u pv %u pu %u\n",
2360 (unsigned)status->errorCode, (unsigned)status->InterfaceVersion,
2361 (unsigned)status->FileType, (unsigned)status->ParentVnode,
2362 (unsigned)status->ParentUnique);
2363 afs_BadFetchStatus(tc);
2371 * Must be called with avc write-locked
2372 * don't absolutely have to invalidate the hint unless the dv has
2373 * changed, but be sure to get it right else there will be consistency bugs.
2376 afs_FetchStatus(struct vcache * avc, struct VenusFid * afid,
2377 struct vrequest * areq, struct AFSFetchStatus * Outsp)
2380 afs_uint32 start = 0;
2381 struct afs_conn *tc;
2382 struct AFSCallBack CallBack;
2383 struct AFSVolSync tsync;
2384 struct rx_connection *rxconn;
2387 tc = afs_Conn(afid, areq, SHARED_LOCK, &rxconn);
2388 avc->dchint = NULL; /* invalidate hints */
2390 avc->callback = tc->parent->srvr->server;
2392 XSTATS_START_TIME(AFS_STATS_FS_RPCIDX_FETCHSTATUS);
2395 RXAFS_FetchStatus(rxconn, (struct AFSFid *)&afid->Fid, Outsp,
2402 code = afs_CheckFetchStatus(tc, Outsp);
2407 } while (afs_Analyze
2408 (tc, rxconn, code, afid, areq, AFS_STATS_FS_RPCIDX_FETCHSTATUS,
2409 SHARED_LOCK, NULL));
2412 afs_UpdateStatus(avc, afid, areq, Outsp, &CallBack, start);
2414 /* used to undo the local callback, but that's too extreme.
2415 * There are plenty of good reasons that fetchstatus might return
2416 * an error, such as EPERM. If we have the vnode cached, statd,
2417 * with callback, might as well keep track of the fact that we
2418 * don't have access...
2420 if (code == EPERM || code == EACCES) {
2421 struct axscache *ac;
2422 if (avc->Access && (ac = afs_FindAxs(avc->Access, areq->uid)))
2424 else /* not found, add a new one if possible */
2425 afs_AddAxs(avc->Access, areq->uid, 0);
2436 * Stuff some information into the vcache for the given file.
2439 * afid : File in question.
2440 * OutStatus : Fetch status on the file.
2441 * CallBack : Callback info.
2442 * tc : RPC connection involved.
2443 * areq : vrequest involved.
2446 * Nothing interesting.
2449 afs_StuffVcache(struct VenusFid *afid,
2450 struct AFSFetchStatus *OutStatus,
2451 struct AFSCallBack *CallBack, struct afs_conn *tc,
2452 struct vrequest *areq)
2454 afs_int32 code, i, newvcache = 0;
2456 struct AFSVolSync tsync;
2458 struct axscache *ac;
2461 AFS_STATCNT(afs_StuffVcache);
2462 #ifdef IFS_VCACHECOUNT
2467 ObtainSharedLock(&afs_xvcache, 8);
2469 tvc = afs_FindVCache(afid, &retry, DO_VLRU| IS_SLOCK /* no stats */ );
2471 #if defined(AFS_SGI_ENV) && !defined(AFS_SGI53_ENV)
2472 ReleaseSharedLock(&afs_xvcache);
2473 spunlock_psema(tvc->v.v_lock, retry, &tvc->v.v_sync, PINOD);
2479 /* no cache entry, better grab one */
2480 UpgradeSToWLock(&afs_xvcache, 25);
2481 tvc = afs_NewVCache(afid, NULL);
2483 ConvertWToSLock(&afs_xvcache);
2486 ReleaseSharedLock(&afs_xvcache);
2491 ReleaseSharedLock(&afs_xvcache);
2492 ObtainWriteLock(&tvc->lock, 58);
2494 tvc->f.states &= ~CStatd;
2495 if ((tvc->f.states & CForeign) || (tvc->f.fid.Fid.Vnode & 1))
2496 osi_dnlc_purgedp(tvc); /* if it (could be) a directory */
2498 /* Is it always appropriate to throw away all the access rights? */
2499 afs_FreeAllAxs(&(tvc->Access));
2501 /*Copy useful per-volume info */
2502 tvp = afs_GetVolume(afid, areq, READ_LOCK);
2504 if (newvcache && (tvp->states & VForeign))
2505 tvc->f.states |= CForeign;
2506 if (tvp->states & VRO)
2507 tvc->f.states |= CRO;
2508 if (tvp->states & VBackup)
2509 tvc->f.states |= CBackup;
2511 * Now, copy ".." entry back out of volume structure, if
2514 if (tvc->mvstat == 2 && tvp->dotdot.Fid.Volume != 0) {
2516 tvc->mvid = (struct VenusFid *)
2517 osi_AllocSmallSpace(sizeof(struct VenusFid));
2518 *tvc->mvid = tvp->dotdot;
2521 /* store the stat on the file */
2522 afs_RemoveVCB(afid);
2523 afs_ProcessFS(tvc, OutStatus, areq);
2524 tvc->callback = tc->srvr->server;
2526 /* we use osi_Time twice below. Ideally, we would use the time at which
2527 * the FetchStatus call began, instead, but we don't have it here. So we
2528 * make do with "now". In the CRO case, it doesn't really matter. In
2529 * the other case, we hope that the difference between "now" and when the
2530 * call actually began execution on the server won't be larger than the
2531 * padding which the server keeps. Subtract 1 second anyway, to be on
2532 * the safe side. Can't subtract more because we don't know how big
2533 * ExpirationTime is. Possible consistency problems may arise if the call
2534 * timeout period becomes longer than the server's expiration padding. */
2535 ObtainWriteLock(&afs_xcbhash, 470);
2536 if (CallBack->ExpirationTime != 0) {
2537 tvc->cbExpires = CallBack->ExpirationTime + osi_Time() - 1;
2538 tvc->f.states |= CStatd;
2539 tvc->f.states &= ~CBulkFetching;
2540 afs_QueueCallback(tvc, CBHash(CallBack->ExpirationTime), tvp);
2541 } else if (tvc->f.states & CRO) {
2542 /* old-fashioned AFS 3.2 style */
2543 tvc->cbExpires = 3600 + osi_Time();
2544 /*XXX*/ tvc->f.states |= CStatd;
2545 tvc->f.states &= ~CBulkFetching;
2546 afs_QueueCallback(tvc, CBHash(3600), tvp);
2548 afs_DequeueCallback(tvc);
2549 tvc->callback = NULL;
2550 tvc->f.states &= ~(CStatd | CUnique);
2551 if ((tvc->f.states & CForeign) || (tvc->f.fid.Fid.Vnode & 1))
2552 osi_dnlc_purgedp(tvc); /* if it (could be) a directory */
2554 ReleaseWriteLock(&afs_xcbhash);
2556 afs_PutVolume(tvp, READ_LOCK);
2558 /* look in per-pag cache */
2559 if (tvc->Access && (ac = afs_FindAxs(tvc->Access, areq->uid)))
2560 ac->axess = OutStatus->CallerAccess; /* substitute pags */
2561 else /* not found, add a new one if possible */
2562 afs_AddAxs(tvc->Access, areq->uid, OutStatus->CallerAccess);
2564 ReleaseWriteLock(&tvc->lock);
2565 afs_Trace4(afs_iclSetp, CM_TRACE_STUFFVCACHE, ICL_TYPE_POINTER, tvc,
2566 ICL_TYPE_POINTER, tvc->callback, ICL_TYPE_INT32,
2567 tvc->cbExpires, ICL_TYPE_INT32, tvc->cbExpires - osi_Time());
2569 * Release ref count... hope this guy stays around...
2572 } /*afs_StuffVcache */
2576 * Decrements the reference count on a cache entry.
2578 * \param avc Pointer to the cache entry to decrement.
2580 * \note Environment: Nothing interesting.
2583 afs_PutVCache(struct vcache *avc)
2585 AFS_STATCNT(afs_PutVCache);
2586 #ifdef AFS_DARWIN80_ENV
2587 vnode_put(AFSTOV(avc));
2591 * Can we use a read lock here?
2593 ObtainReadLock(&afs_xvcache);
2595 ReleaseReadLock(&afs_xvcache);
2597 } /*afs_PutVCache */
2601 * Reset a vcache entry, so local contents are ignored, and the
2602 * server will be reconsulted next time the vcache is used
2604 * \param avc Pointer to the cache entry to reset
2607 * \note avc must be write locked on entry
2610 afs_ResetVCache(struct vcache *avc, afs_ucred_t *acred)
2612 ObtainWriteLock(&afs_xcbhash, 456);
2613 afs_DequeueCallback(avc);
2614 avc->f.states &= ~(CStatd | CDirty); /* next reference will re-stat */
2615 ReleaseWriteLock(&afs_xcbhash);
2616 /* now find the disk cache entries */
2617 afs_TryToSmush(avc, acred, 1);
2618 osi_dnlc_purgedp(avc);
2619 if (avc->linkData && !(avc->f.states & CCore)) {
2620 afs_osi_Free(avc->linkData, strlen(avc->linkData) + 1);
2621 avc->linkData = NULL;
2626 * Sleepa when searching for a vcache. Releases all the pending locks,
2627 * sleeps then obtains the previously released locks.
2629 * \param vcache Enter sleep state.
2630 * \param flag Determines what locks to use.
2635 findvc_sleep(struct vcache *avc, int flag)
2637 if (flag & IS_SLOCK) {
2638 ReleaseSharedLock(&afs_xvcache);
2640 if (flag & IS_WLOCK) {
2641 ReleaseWriteLock(&afs_xvcache);
2643 ReleaseReadLock(&afs_xvcache);
2646 afs_osi_Sleep(&avc->f.states);
2647 if (flag & IS_SLOCK) {
2648 ObtainSharedLock(&afs_xvcache, 341);
2650 if (flag & IS_WLOCK) {
2651 ObtainWriteLock(&afs_xvcache, 343);
2653 ObtainReadLock(&afs_xvcache);
2659 * Add a reference on an existing vcache entry.
2661 * \param tvc Pointer to the vcache.
2663 * \note Environment: Must be called with at least one reference from
2664 * elsewhere on the vcache, even if that reference will be dropped.
2665 * The global lock is required.
2667 * \return 0 on success, -1 on failure.
2671 afs_RefVCache(struct vcache *tvc)
2673 #ifdef AFS_DARWIN80_ENV
2677 /* AFS_STATCNT(afs_RefVCache); */
2679 #ifdef AFS_DARWIN80_ENV
2683 if (vnode_ref(tvp)) {
2685 /* AFSTOV(tvc) may be NULL */
2694 } /*afs_RefVCache */
2697 * Find a vcache entry given a fid.
2699 * \param afid Pointer to the fid whose cache entry we desire.
2700 * \param retry (SGI-specific) tell the caller to drop the lock on xvcache,
2701 * unlock the vnode, and try again.
2702 * \param flag Bit 1 to specify whether to compute hit statistics. Not
2703 * set if FindVCache is called as part of internal bookkeeping.
2705 * \note Environment: Must be called with the afs_xvcache lock at least held at
2706 * the read level. In order to do the VLRU adjustment, the xvcache lock
2707 * must be shared-- we upgrade it here.
2711 afs_FindVCache(struct VenusFid *afid, afs_int32 * retry, afs_int32 flag)
2716 #ifdef AFS_DARWIN80_ENV
2717 struct vcache *deadvc = NULL, *livevc = NULL;
2721 AFS_STATCNT(afs_FindVCache);
2725 for (tvc = afs_vhashT[i]; tvc; tvc = tvc->hnext) {
2726 if (FidMatches(afid, tvc)) {
2727 if (tvc->f.states & CVInit) {
2728 findvc_sleep(tvc, flag);
2731 #ifdef AFS_DARWIN80_ENV
2732 if (tvc->f.states & CDeadVnode) {
2733 findvc_sleep(tvc, flag);
2741 /* should I have a read lock on the vnode here? */
2745 #if defined(AFS_DARWIN80_ENV)
2749 if (tvp && vnode_ref(tvp)) {
2751 /* AFSTOV(tvc) may be NULL */
2760 #elif defined(AFS_DARWIN_ENV)
2761 tvc->f.states |= CUBCinit;
2763 if (UBCINFOMISSING(AFSTOV(tvc)) ||
2764 UBCINFORECLAIMED(AFSTOV(tvc))) {
2765 ubc_info_init(AFSTOV(tvc));
2768 tvc->f.states &= ~CUBCinit;
2770 osi_vnhold(tvc, retry); /* already held, above */
2771 if (retry && *retry)
2775 * only move to front of vlru if we have proper vcache locking)
2777 if (flag & DO_VLRU) {
2778 if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) {
2779 refpanic("FindVC VLRU inconsistent1");
2781 if (tvc->vlruq.next->prev != &(tvc->vlruq)) {
2782 refpanic("FindVC VLRU inconsistent1");
2784 if (tvc->vlruq.prev->next != &(tvc->vlruq)) {
2785 refpanic("FindVC VLRU inconsistent2");
2787 UpgradeSToWLock(&afs_xvcache, 26);
2788 QRemove(&tvc->vlruq);
2789 QAdd(&VLRU, &tvc->vlruq);
2790 ConvertWToSLock(&afs_xvcache);
2791 if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) {
2792 refpanic("FindVC VLRU inconsistent1");
2794 if (tvc->vlruq.next->prev != &(tvc->vlruq)) {
2795 refpanic("FindVC VLRU inconsistent2");
2797 if (tvc->vlruq.prev->next != &(tvc->vlruq)) {
2798 refpanic("FindVC VLRU inconsistent3");
2804 if (flag & DO_STATS) {
2806 afs_stats_cmperf.vcacheHits++;
2808 afs_stats_cmperf.vcacheMisses++;
2809 if (afs_IsPrimaryCellNum(afid->Cell))
2810 afs_stats_cmperf.vlocalAccesses++;
2812 afs_stats_cmperf.vremoteAccesses++;
2815 } /*afs_FindVCache */
2818 * Find a vcache entry given a fid. Does a wildcard match on what we
2819 * have for the fid. If more than one entry, don't return anything.
2821 * \param avcp Fill in pointer if we found one and only one.
2822 * \param afid Pointer to the fid whose cache entry we desire.
2823 * \param retry (SGI-specific) tell the caller to drop the lock on xvcache,
2824 * unlock the vnode, and try again.
2825 * \param flags bit 1 to specify whether to compute hit statistics. Not
2826 * set if FindVCache is called as part of internal bookkeeping.
2828 * \note Environment: Must be called with the afs_xvcache lock at least held at
2829 * the read level. In order to do the VLRU adjustment, the xvcache lock
2830 * must be shared-- we upgrade it here.
2832 * \return Number of matches found.
2835 int afs_duplicate_nfs_fids = 0;
2838 afs_NFSFindVCache(struct vcache **avcp, struct VenusFid *afid)
2842 afs_int32 count = 0;
2843 struct vcache *found_tvc = NULL;
2844 #ifdef AFS_DARWIN80_ENV
2848 AFS_STATCNT(afs_FindVCache);
2852 ObtainSharedLock(&afs_xvcache, 331);
2855 for (tvc = afs_vhashT[i]; tvc; tvc = tvc->hnext) {
2856 /* Match only on what we have.... */
2857 if (((tvc->f.fid.Fid.Vnode & 0xffff) == afid->Fid.Vnode)
2858 && (tvc->f.fid.Fid.Volume == afid->Fid.Volume)
2859 && ((tvc->f.fid.Fid.Unique & 0xffffff) == afid->Fid.Unique)
2860 && (tvc->f.fid.Cell == afid->Cell)) {
2861 if (tvc->f.states & CVInit) {
2862 ReleaseSharedLock(&afs_xvcache);
2863 afs_osi_Sleep(&tvc->f.states);
2866 #ifdef AFS_DARWIN80_ENV
2867 if (tvc->f.states & CDeadVnode) {
2868 ReleaseSharedLock(&afs_xvcache);
2869 afs_osi_Sleep(&tvc->f.states);
2873 if (vnode_get(tvp)) {
2874 /* This vnode no longer exists. */
2877 if (vnode_ref(tvp)) {
2878 /* This vnode no longer exists. */
2880 /* AFSTOV(tvc) may be NULL */
2885 #endif /* AFS_DARWIN80_ENV */
2889 afs_duplicate_nfs_fids++;
2890 ReleaseSharedLock(&afs_xvcache);
2891 #ifdef AFS_DARWIN80_ENV
2892 /* Drop our reference counts. */
2893 vnode_put(AFSTOV(tvc));
2894 vnode_put(AFSTOV(found_tvc));
2903 /* should I have a read lock on the vnode here? */
2905 #ifndef AFS_DARWIN80_ENV
2906 #if defined(AFS_SGI_ENV) && !defined(AFS_SGI53_ENV)
2907 afs_int32 retry = 0;
2908 osi_vnhold(tvc, &retry);
2911 found_tvc = (struct vcache *)0;
2912 ReleaseSharedLock(&afs_xvcache);
2913 spunlock_psema(tvc->v.v_lock, retry, &tvc->v.v_sync, PINOD);
2917 osi_vnhold(tvc, (int *)0); /* already held, above */
2921 * We obtained the xvcache lock above.
2923 if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) {
2924 refpanic("FindVC VLRU inconsistent1");
2926 if (tvc->vlruq.next->prev != &(tvc->vlruq)) {
2927 refpanic("FindVC VLRU inconsistent1");
2929 if (tvc->vlruq.prev->next != &(tvc->vlruq)) {
2930 refpanic("FindVC VLRU inconsistent2");
2932 UpgradeSToWLock(&afs_xvcache, 568);
2933 QRemove(&tvc->vlruq);
2934 QAdd(&VLRU, &tvc->vlruq);
2935 ConvertWToSLock(&afs_xvcache);
2936 if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) {
2937 refpanic("FindVC VLRU inconsistent1");
2939 if (tvc->vlruq.next->prev != &(tvc->vlruq)) {
2940 refpanic("FindVC VLRU inconsistent2");
2942 if (tvc->vlruq.prev->next != &(tvc->vlruq)) {
2943 refpanic("FindVC VLRU inconsistent3");
2949 afs_stats_cmperf.vcacheHits++;
2951 afs_stats_cmperf.vcacheMisses++;
2952 if (afs_IsPrimaryCellNum(afid->Cell))
2953 afs_stats_cmperf.vlocalAccesses++;
2955 afs_stats_cmperf.vremoteAccesses++;
2957 *avcp = tvc; /* May be null */
2959 ReleaseSharedLock(&afs_xvcache);
2960 return (tvc ? 1 : 0);
2962 } /*afs_NFSFindVCache */
2968 * Initialize vcache related variables
2973 afs_vcacheInit(int astatSize)
2975 #if !defined(AFS_LINUX22_ENV)
2979 if (!afs_maxvcount) {
2980 afs_maxvcount = astatSize; /* no particular limit on linux? */
2982 #if !defined(AFS_LINUX22_ENV)
2986 AFS_RWLOCK_INIT(&afs_xvcache, "afs_xvcache");
2987 LOCK_INIT(&afs_xvcb, "afs_xvcb");
2989 #if !defined(AFS_LINUX22_ENV)
2990 /* Allocate and thread the struct vcache entries */
2991 tvp = afs_osi_Alloc(astatSize * sizeof(struct vcache));
2992 osi_Assert(tvp != NULL);
2993 memset(tvp, 0, sizeof(struct vcache) * astatSize);
2995 Initial_freeVCList = tvp;
2996 freeVCList = &(tvp[0]);
2997 for (i = 0; i < astatSize - 1; i++) {
2998 tvp[i].nextfree = &(tvp[i + 1]);
3000 tvp[astatSize - 1].nextfree = NULL;
3001 # ifdef KERNEL_HAVE_PIN
3002 pin((char *)tvp, astatSize * sizeof(struct vcache)); /* XXX */
3006 #if defined(AFS_SGI_ENV)
3007 for (i = 0; i < astatSize; i++) {
3008 char name[METER_NAMSZ];
3009 struct vcache *tvc = &tvp[i];
3011 tvc->v.v_number = ++afsvnumbers;
3012 tvc->vc_rwlockid = OSI_NO_LOCKID;
3013 initnsema(&tvc->vc_rwlock, 1,
3014 makesname(name, "vrw", tvc->v.v_number));
3015 #ifndef AFS_SGI53_ENV
3016 initnsema(&tvc->v.v_sync, 0, makesname(name, "vsy", tvc->v.v_number));
3018 #ifndef AFS_SGI62_ENV
3019 initnlock(&tvc->v.v_lock, makesname(name, "vlk", tvc->v.v_number));
3020 #endif /* AFS_SGI62_ENV */
3024 for(i = 0; i < VCSIZE; ++i)
3025 QInit(&afs_vhashTV[i]);
3032 shutdown_vcache(void)
3035 struct afs_cbr *tsp;
3037 * XXX We may potentially miss some of the vcaches because if when
3038 * there are no free vcache entries and all the vcache entries are active
3039 * ones then we allocate an additional one - admittedly we almost never
3044 struct afs_q *tq, *uq = NULL;
3046 for (tq = VLRU.prev; tq != &VLRU; tq = uq) {
3050 osi_FreeSmallSpace(tvc->mvid);
3051 tvc->mvid = (struct VenusFid *)0;
3054 aix_gnode_rele(AFSTOV(tvc));
3056 if (tvc->linkData) {
3057 afs_osi_Free(tvc->linkData, strlen(tvc->linkData) + 1);
3062 * Also free the remaining ones in the Cache
3064 for (i = 0; i < VCSIZE; i++) {
3065 for (tvc = afs_vhashT[i]; tvc; tvc = tvc->hnext) {
3067 osi_FreeSmallSpace(tvc->mvid);
3068 tvc->mvid = (struct VenusFid *)0;
3072 afs_osi_Free(tvc->v.v_gnode, sizeof(struct gnode));
3073 #ifdef AFS_AIX32_ENV
3076 vms_delete(tvc->segid);
3078 tvc->segid = tvc->vmh = NULL;
3079 if (VREFCOUNT_GT(tvc,0))
3080 osi_Panic("flushVcache: vm race");
3088 #if defined(AFS_SUN5_ENV)
3094 if (tvc->linkData) {
3095 afs_osi_Free(tvc->linkData, strlen(tvc->linkData) + 1);
3100 afs_FreeAllAxs(&(tvc->Access));
3106 * Free any leftover callback queue
3108 for (i = 0; i < afs_stats_cmperf.CallBackAlloced; i++) {
3109 tsp = afs_cbrHeads[i];
3110 afs_cbrHeads[i] = 0;
3111 afs_osi_Free((char *)tsp, AFS_NCBRS * sizeof(struct afs_cbr));
3115 #if !defined(AFS_LINUX22_ENV)
3116 afs_osi_Free(Initial_freeVCList, afs_cacheStats * sizeof(struct vcache));
3118 # ifdef KERNEL_HAVE_PIN
3119 unpin(Initial_freeVCList, afs_cacheStats * sizeof(struct vcache));
3122 freeVCList = Initial_freeVCList = 0;
3125 AFS_RWLOCK_INIT(&afs_xvcache, "afs_xvcache");
3126 LOCK_INIT(&afs_xvcb, "afs_xvcb");
3128 for(i = 0; i < VCSIZE; ++i)
3129 QInit(&afs_vhashTV[i]);
3133 afs_DisconGiveUpCallbacks(void)
3139 ObtainWriteLock(&afs_xvcache, 1002); /* XXX - should be a unique number */
3142 /* Somehow, walk the set of vcaches, with each one coming out as tvc */
3143 for (i = 0; i < VCSIZE; i++) {
3144 for (tvc = afs_vhashT[i]; tvc; tvc = tvc->hnext) {
3146 if (afs_QueueVCB(tvc, &slept)) {
3147 tvc->callback = NULL;
3156 ReleaseWriteLock(&afs_xvcache);
3163 * Clear the Statd flag from all vcaches
3165 * This function removes the Statd flag from all vcaches. It's used by
3166 * disconnected mode to tidy up during reconnection
3170 afs_ClearAllStatdFlag(void)
3175 ObtainWriteLock(&afs_xvcache, 715);
3177 for (i = 0; i < VCSIZE; i++) {
3178 for (tvc = afs_vhashT[i]; tvc; tvc = tvc->hnext) {
3179 tvc->f.states &= ~(CStatd|CUnique);
3182 ReleaseWriteLock(&afs_xvcache);