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
39 #include <afsconfig.h>
40 #include "afs/param.h"
42 #include "afs/sysincludes.h" /*Standard vendor system headers */
43 #include "afsincludes.h" /*AFS-based standard headers */
44 #include "afs/afs_stats.h"
45 #include "afs/afs_cbqueue.h"
46 #include "afs/afs_osidnlc.h"
48 afs_int32 afs_maxvcount = 0; /* max number of vcache entries */
49 afs_int32 afs_vcount = 0; /* number of vcache in use now */
57 #endif /* AFS_SGI64_ENV */
59 /* Exported variables */
60 afs_rwlock_t afs_xvcdirty; /*Lock: discon vcache dirty list mgmt */
61 afs_rwlock_t afs_xvcache; /*Lock: alloc new stat cache entries */
62 afs_rwlock_t afs_xvreclaim; /*Lock: entries reclaimed, not on free list */
63 afs_lock_t afs_xvcb; /*Lock: fids on which there are callbacks */
64 #if !defined(AFS_LINUX22_ENV)
65 static struct vcache *freeVCList; /*Free list for stat cache entries */
66 struct vcache *ReclaimedVCList; /*Reclaimed list for stat entries */
67 static struct vcache *Initial_freeVCList; /*Initial list for above */
69 struct afs_q VLRU; /*vcache LRU */
70 afs_int32 vcachegen = 0;
71 unsigned int afs_paniconwarn = 0;
72 struct vcache *afs_vhashT[VCSIZE];
73 struct afs_q afs_vhashTV[VCSIZE];
74 static struct afs_cbr *afs_cbrHashT[CBRSIZE];
75 afs_int32 afs_bulkStatsLost;
76 int afs_norefpanic = 0;
79 /* Disk backed vcache definitions
80 * Both protected by xvcache */
81 static int afs_nextVcacheSlot = 0;
82 static struct afs_slotlist *afs_freeSlotList = NULL;
84 /* Forward declarations */
85 static afs_int32 afs_QueueVCB(struct vcache *avc, int *slept);
89 * The PFlush algorithm makes use of the fact that Fid.Unique is not used in
90 * below hash algorithms. Change it if need be so that flushing algorithm
91 * doesn't move things from one hash chain to another.
93 /* Don't hash on the cell; our callback-breaking code sometimes fails to compute
94 * the cell correctly, and only scans one hash bucket. */
95 int VCHash(struct VenusFid *fid)
97 return opr_jhash_int2(fid->Fid.Volume, fid->Fid.Vnode, 0) &
98 opr_jhash_mask(VCSIZEBITS);
100 /* Hash only on volume to speed up volume callbacks. */
101 int VCHashV(struct VenusFid *fid)
103 return opr_jhash_int(fid->Fid.Volume, 0) & opr_jhash_mask(VCSIZEBITS);
107 * Generate an index into the hash table for a given Fid.
109 * \return The hash value.
112 afs_HashCBRFid(struct AFSFid *fid)
114 return (fid->Volume + fid->Vnode + fid->Unique) % CBRSIZE;
118 * Insert a CBR entry into the hash table.
119 * Must be called with afs_xvcb held.
124 afs_InsertHashCBR(struct afs_cbr *cbr)
126 int slot = afs_HashCBRFid(&cbr->fid);
128 cbr->hash_next = afs_cbrHashT[slot];
129 if (afs_cbrHashT[slot])
130 afs_cbrHashT[slot]->hash_pprev = &cbr->hash_next;
132 cbr->hash_pprev = &afs_cbrHashT[slot];
133 afs_cbrHashT[slot] = cbr;
138 * Flush the given vcache entry.
141 * afs_xvcache lock must be held for writing upon entry to
142 * prevent people from changing the vrefCount field, and to
143 * protect the lruq and hnext fields.
144 * LOCK: afs_FlushVCache afs_xvcache W
145 * REFCNT: vcache ref count must be zero on entry except for osf1
146 * RACE: lock is dropped and reobtained, permitting race in caller
148 * \param avc Pointer to vcache entry to flush.
149 * \param slept Pointer to int to set 1 if we sleep/drop locks, 0 if we don't.
153 afs_FlushVCache(struct vcache *avc, int *slept)
154 { /*afs_FlushVCache */
157 struct vcache **uvc, *wvc;
159 /* NOTE: We must have nothing drop afs_xvcache until we have removed all
160 * possible references to this vcache. This means all hash tables, queues,
164 AFS_STATCNT(afs_FlushVCache);
165 afs_Trace2(afs_iclSetp, CM_TRACE_FLUSHV, ICL_TYPE_POINTER, avc,
166 ICL_TYPE_INT32, avc->f.states);
168 code = osi_VM_FlushVCache(avc);
172 if (avc->f.states & CVFlushed) {
176 #if !defined(AFS_LINUX22_ENV)
177 if (avc->nextfree || !avc->vlruq.prev || !avc->vlruq.next) { /* qv afs.h */
178 refpanic("LRU vs. Free inconsistency");
181 avc->f.states |= CVFlushed;
182 /* pull the entry out of the lruq and put it on the free list */
183 QRemove(&avc->vlruq);
185 /* keep track of # of files that we bulk stat'd, but never used
186 * before they got recycled.
188 if (avc->f.states & CBulkStat)
191 /* remove entry from the hash chain */
192 i = VCHash(&avc->f.fid);
193 uvc = &afs_vhashT[i];
194 for (wvc = *uvc; wvc; uvc = &wvc->hnext, wvc = *uvc) {
202 /* remove entry from the volume hash table */
203 QRemove(&avc->vhashq);
205 #if defined(AFS_LINUX26_ENV)
207 struct pagewriter *pw, *store;
208 struct list_head tofree;
210 INIT_LIST_HEAD(&tofree);
211 spin_lock(&avc->pagewriter_lock);
212 list_for_each_entry_safe(pw, store, &avc->pagewriters, link) {
214 /* afs_osi_Free may sleep so we need to defer it */
215 list_add_tail(&pw->link, &tofree);
217 spin_unlock(&avc->pagewriter_lock);
218 list_for_each_entry_safe(pw, store, &tofree, link) {
220 afs_osi_Free(pw, sizeof(struct pagewriter));
225 if (avc->mvid.target_root)
226 osi_FreeSmallSpace(avc->mvid.target_root);
227 avc->mvid.target_root = NULL;
229 afs_osi_Free(avc->linkData, strlen(avc->linkData) + 1);
230 avc->linkData = NULL;
232 #if defined(AFS_XBSD_ENV) || defined(AFS_DARWIN_ENV)
233 /* OK, there are no internal vrefCounts, so there shouldn't
234 * be any more refs here. */
236 # ifdef AFS_DARWIN80_ENV
237 vnode_clearfsnode(AFSTOV(avc));
238 vnode_removefsref(AFSTOV(avc));
240 avc->v->v_data = NULL; /* remove from vnode */
242 AFSTOV(avc) = NULL; /* also drop the ptr to vnode */
246 #ifdef AFS_SUN511_ENV
251 #elif defined(AFS_SUN510_ENV)
252 /* As we use private vnodes, cleanup is up to us */
253 vn_reinit(AFSTOV(avc));
255 afs_FreeAllAxs(&(avc->Access));
256 afs_StaleVCacheFlags(avc, AFS_STALEVC_FILENAME, CUnique);
258 /* By this point, the vcache has been removed from all global structures
259 * via which someone could try to use the vcache. It is okay to drop
260 * afs_xvcache at this point (if *slept is set). */
262 if (afs_shuttingdown == AFS_RUNNING)
263 afs_QueueVCB(avc, slept);
266 * Next, keep track of which vnodes we've deleted for create's
267 * optimistic synchronization algorithm
270 if (avc->f.fid.Fid.Vnode & 1)
276 #if !defined(AFS_LINUX22_ENV)
277 /* put the entry in the free list */
278 avc->nextfree = freeVCList;
280 if (avc->vlruq.prev || avc->vlruq.next) {
281 refpanic("LRU vs. Free inconsistency");
283 avc->f.states |= CVFlushed;
285 /* This should put it back on the vnode free list since usecount is 1 */
287 if (VREFCOUNT_GT(avc,0)) {
288 AFS_RELE(AFSTOV(avc));
289 afs_stats_cmperf.vcacheXAllocs--;
291 if (afs_norefpanic) {
292 afs_warn("flush vc refcnt < 1");
295 osi_Panic("flush vc refcnt < 1");
297 #endif /* AFS_LINUX22_ENV */
302 } /*afs_FlushVCache */
306 * The core of the inactive vnode op for all but IRIX.
312 afs_InactiveVCache(struct vcache *avc, afs_ucred_t *acred)
314 AFS_STATCNT(afs_inactive);
315 if (avc->f.states & CDirty) {
316 /* we can't keep trying to push back dirty data forever. Give up. */
317 afs_InvalidateAllSegments(avc); /* turns off dirty bit */
319 avc->f.states &= ~CMAPPED;
320 avc->f.states &= ~CDirty; /* Turn it off */
321 if (avc->f.states & CUnlinked) {
322 if (CheckLock(&afs_xvcache) || CheckLock(&afs_xdcache)) {
323 avc->f.states |= CUnlinkedDel;
326 afs_remunlink(avc, 1); /* ignore any return code */
333 * Allocate a callback return structure from the
334 * free list and return it.
336 * Environment: The alloc and free routines are both called with the afs_xvcb lock
337 * held, so we don't have to worry about blocking in osi_Alloc.
339 * \return The allocated afs_cbr.
341 static struct afs_cbr *afs_cbrSpace = 0;
342 /* if alloc limit below changes, fix me! */
343 static struct afs_cbr *afs_cbrHeads[16];
350 while (!afs_cbrSpace) {
351 if (afs_stats_cmperf.CallBackAlloced >= sizeof(afs_cbrHeads)/sizeof(afs_cbrHeads[0])) {
352 /* don't allocate more than 16 * AFS_NCBRS for now */
354 afs_stats_cmperf.CallBackFlushes++;
357 tsp = afs_osi_Alloc(AFS_NCBRS * sizeof(struct afs_cbr));
358 osi_Assert(tsp != NULL);
359 for (i = 0; i < AFS_NCBRS - 1; i++) {
360 tsp[i].next = &tsp[i + 1];
362 tsp[AFS_NCBRS - 1].next = 0;
364 afs_cbrHeads[afs_stats_cmperf.CallBackAlloced] = tsp;
365 afs_stats_cmperf.CallBackAlloced++;
369 afs_cbrSpace = tsp->next;
374 * Free a callback return structure, removing it from all lists.
376 * Environment: the xvcb lock is held over these calls.
378 * \param asp The address of the structure to free.
383 afs_FreeCBR(struct afs_cbr *asp)
385 *(asp->pprev) = asp->next;
387 asp->next->pprev = asp->pprev;
389 *(asp->hash_pprev) = asp->hash_next;
391 asp->hash_next->hash_pprev = asp->hash_pprev;
393 asp->next = afs_cbrSpace;
399 FlushAllVCBs(int nconns, struct rx_connection **rxconns,
400 struct afs_conn **conns)
405 results = afs_osi_Alloc(nconns * sizeof (afs_int32));
406 osi_Assert(results != NULL);
409 multi_Rx(rxconns,nconns)
411 multi_RXAFS_GiveUpAllCallBacks();
412 results[multi_i] = multi_error;
417 * Freeing the CBR will unlink it from the server's CBR list
418 * do it here, not in the loop, because a dynamic CBR will call
419 * into the memory management routines.
421 for ( i = 0 ; i < nconns ; i++ ) {
422 if (results[i] == 0) {
423 /* Unchain all of them */
424 while (conns[i]->parent->srvr->server->cbrs)
425 afs_FreeCBR(conns[i]->parent->srvr->server->cbrs);
428 afs_osi_Free(results, nconns * sizeof(afs_int32));
432 * Flush all queued callbacks to all servers.
434 * Environment: holds xvcb lock over RPC to guard against race conditions
435 * when a new callback is granted for the same file later on.
437 * \return 0 for success.
440 afs_FlushVCBs(afs_int32 lockit)
442 struct AFSFid *tfids;
443 struct AFSCallBack callBacks[1];
444 struct AFSCBFids fidArray;
445 struct AFSCBs cbArray;
447 struct afs_cbr *tcbrp;
451 struct vrequest *treq = NULL;
453 int safety1, safety2, safety3;
456 if (AFS_IS_DISCONNECTED)
459 if ((code = afs_CreateReq(&treq, afs_osi_credp)))
461 treq->flags |= O_NONBLOCK;
462 tfids = afs_osi_Alloc(sizeof(struct AFSFid) * AFS_MAXCBRSCALL);
463 osi_Assert(tfids != NULL);
466 ObtainWriteLock(&afs_xvcb, 273);
469 * First, attempt a multi across everything, all addresses
470 * for all servers we know of.
474 afs_LoopServers(AFS_LS_ALL, NULL, 0, FlushAllVCBs, NULL);
476 ObtainReadLock(&afs_xserver);
477 for (i = 0; i < NSERVERS; i++) {
478 for (safety1 = 0, tsp = afs_servers[i];
479 tsp && safety1 < afs_totalServers + 10;
480 tsp = tsp->next, safety1++) {
482 if (tsp->cbrs == (struct afs_cbr *)0)
485 /* otherwise, grab a block of AFS_MAXCBRSCALL from the list
486 * and make an RPC, over and over again.
488 tcount = 0; /* number found so far */
489 for (safety2 = 0; safety2 < afs_cacheStats; safety2++) {
490 if (tcount >= AFS_MAXCBRSCALL || !tsp->cbrs) {
491 struct rx_connection *rxconn;
492 /* if buffer is full, or we've queued all we're going
493 * to from this server, we should flush out the
496 fidArray.AFSCBFids_len = tcount;
497 fidArray.AFSCBFids_val = (struct AFSFid *)tfids;
498 cbArray.AFSCBs_len = 1;
499 cbArray.AFSCBs_val = callBacks;
500 memset(&callBacks[0], 0, sizeof(callBacks[0]));
501 callBacks[0].CallBackType = CB_EXCLUSIVE;
502 for (safety3 = 0; safety3 < AFS_MAXHOSTS * 2; safety3++) {
503 tc = afs_ConnByHost(tsp, tsp->cell->fsport,
504 tsp->cell->cellNum, treq, 0,
505 SHARED_LOCK, 0, &rxconn);
508 (AFS_STATS_FS_RPCIDX_GIVEUPCALLBACKS);
511 RXAFS_GiveUpCallBacks(rxconn, &fidArray,
518 (tc, rxconn, code, 0, treq,
519 AFS_STATS_FS_RPCIDX_GIVEUPCALLBACKS, SHARED_LOCK,
524 /* ignore return code, since callbacks may have
525 * been returned anyway, we shouldn't leave them
526 * around to be returned again.
528 * Next, see if we are done with this server, and if so,
529 * break to deal with the next one.
535 /* if to flush full buffer */
536 /* if we make it here, we have an entry at the head of cbrs,
537 * which we should copy to the file ID array and then free.
540 tfids[tcount++] = tcbrp->fid;
542 /* Freeing the CBR will unlink it from the server's CBR list */
544 } /* while loop for this one server */
545 if (safety2 > afs_cacheStats) {
546 afs_warn("possible internal error afs_flushVCBs (%d)\n",
549 } /* for loop for this hash chain */
550 } /* loop through all hash chains */
551 if (safety1 > afs_totalServers + 2) {
553 ("AFS internal error (afs_flushVCBs) (%d > %d), continuing...\n",
554 safety1, afs_totalServers + 2);
556 osi_Panic("afs_flushVCBS safety1");
559 ReleaseReadLock(&afs_xserver);
561 ReleaseWriteLock(&afs_xvcb);
562 afs_osi_Free(tfids, sizeof(struct AFSFid) * AFS_MAXCBRSCALL);
563 afs_DestroyReq(treq);
568 * Queue a callback on the given fid.
571 * Locks the xvcb lock.
572 * Called when the xvcache lock is already held.
573 * RACE: afs_xvcache may be dropped and reacquired
575 * \param avc vcache entry
576 * \param slep Set to 1 if we dropped afs_xvcache
577 * \return 1 if queued, 0 otherwise
581 afs_QueueVCB(struct vcache *avc, int *slept)
585 struct afs_cbr *tcbp;
588 AFS_STATCNT(afs_QueueVCB);
590 ObtainWriteLock(&afs_xvcb, 274);
592 /* we can't really give back callbacks on RO files, since the
593 * server only tracks them on a per-volume basis, and we don't
594 * know whether we still have some other files from the same
596 if (!((avc->f.states & CRO) == 0 && avc->callback)) {
600 /* The callback is really just a struct server ptr. */
601 tsp = (struct server *)(avc->callback);
604 /* If we don't have CBR space, AllocCBR may block or hit the net for
605 * clearing up CBRs. Hitting the net may involve a fileserver
606 * needing to contact us, so we must drop xvcache so we don't block
607 * those requests from going through. */
608 reacquire = *slept = 1;
609 ReleaseWriteLock(&afs_xvcache);
612 /* we now have a pointer to the server, so we just allocate
613 * a queue entry and queue it.
615 tcbp = afs_AllocCBR();
616 tcbp->fid = avc->f.fid.Fid;
618 tcbp->next = tsp->cbrs;
620 tsp->cbrs->pprev = &tcbp->next;
623 tcbp->pprev = &tsp->cbrs;
625 afs_InsertHashCBR(tcbp);
629 /* now release locks and return */
630 ReleaseWriteLock(&afs_xvcb);
633 /* make sure this is after dropping xvcb, for locking order */
634 ObtainWriteLock(&afs_xvcache, 279);
641 * Remove a queued callback for a given Fid.
644 * Locks xvcb and xserver locks.
645 * Typically called with xdcache, xvcache and/or individual vcache
648 * \param afid The fid we want cleansed of queued callbacks.
653 afs_RemoveVCB(struct VenusFid *afid)
656 struct afs_cbr *cbr, *ncbr;
658 AFS_STATCNT(afs_RemoveVCB);
659 ObtainWriteLock(&afs_xvcb, 275);
661 slot = afs_HashCBRFid(&afid->Fid);
662 ncbr = afs_cbrHashT[slot];
666 ncbr = cbr->hash_next;
668 if (afid->Fid.Volume == cbr->fid.Volume &&
669 afid->Fid.Vnode == cbr->fid.Vnode &&
670 afid->Fid.Unique == cbr->fid.Unique) {
675 ReleaseWriteLock(&afs_xvcb);
679 afs_FlushReclaimedVcaches(void)
681 #if !defined(AFS_LINUX22_ENV)
684 struct vcache *tmpReclaimedVCList = NULL;
686 ObtainWriteLock(&afs_xvreclaim, 76);
687 while (ReclaimedVCList) {
688 tvc = ReclaimedVCList; /* take from free list */
689 ReclaimedVCList = tvc->nextfree;
690 tvc->nextfree = NULL;
691 code = afs_FlushVCache(tvc, &fv_slept);
693 /* Ok, so, if we got code != 0, uh, wtf do we do? */
694 /* Probably, build a temporary list and then put all back when we
695 get to the end of the list */
696 /* This is actually really crappy, but we need to not leak these.
697 We probably need a way to be smarter about this. */
698 tvc->nextfree = tmpReclaimedVCList;
699 tmpReclaimedVCList = tvc;
700 /* printf("Reclaim list flush %lx failed: %d\n", (unsigned long) tvc, code); */
702 if (tvc->f.states & (CVInit
703 # ifdef AFS_DARWIN80_ENV
707 tvc->f.states &= ~(CVInit
708 # ifdef AFS_DARWIN80_ENV
712 afs_osi_Wakeup(&tvc->f.states);
715 if (tmpReclaimedVCList)
716 ReclaimedVCList = tmpReclaimedVCList;
718 ReleaseWriteLock(&afs_xvreclaim);
723 afs_PostPopulateVCache(struct vcache *avc, struct VenusFid *afid, int seq)
726 * The proper value for mvstat (for root fids) is setup by the caller.
728 avc->mvstat = AFS_MVSTAT_FILE;
729 if (afid->Fid.Vnode == 1 && afid->Fid.Unique == 1)
730 avc->mvstat = AFS_MVSTAT_ROOT;
732 if (afs_globalVFS == 0)
733 osi_Panic("afs globalvfs");
735 osi_PostPopulateVCache(avc);
738 osi_dnlc_purgedp(avc); /* this may be overkill */
739 memset(&(avc->callsort), 0, sizeof(struct afs_q));
741 avc->f.states &=~ CVInit;
743 avc->f.states |= CBulkFetching;
744 avc->f.m.Length = seq;
746 afs_osi_Wakeup(&avc->f.states);
750 * afs_VCacheStressed() is intended to determine if the stat cache looks
751 * stressed / full-ish. Due to the different strategies of allocating vcaches
752 * on different platforms, the definition of "stressed" varies, and is somewhat
753 * arbitrary. We just try to make a reasonable guess here.
755 * Returns 1 if the stat cache looks stressed, and 0 otherwise.
757 #ifdef AFS_LINUX26_ENV
759 afs_VCacheStressed(void)
761 if (afsd_dynamic_vcaches) {
763 * For dynamic vcaches, the number of vcaches in use can vary wildly.
764 * Consider us stressed if we're significantly above the configured
765 * threshold. VCACHE_DYNAMIC_STRESSED is the arbitrary point at which
766 * we're considered "significantly" over the threshold.
768 if (afs_vcount > afs_cacheStats + VCACHE_DYNAMIC_STRESSED) {
775 * For non-dynamic vcaches, we should never go above the configured
776 * limit, and ShakeLooseVCaches should try to get us to VCACHE_FREE
777 * under the limit. So if we're closer then VCACHE_FREE/2, then we're
778 * very close to the limit, so consider us stressed.
780 if (afs_vcount > afs_cacheStats || afs_cacheStats - afs_vcount < VCACHE_FREE/2) {
786 #else /* AFS_LINUX26_ENV */
788 afs_VCacheStressed(void)
790 /* If we don't have any vcaches in the free list, then consider the stat
792 if (freeVCList != NULL) {
797 #endif /* AFS_LINUX26_ENV */
800 afs_ShakeLooseVCaches(afs_int32 anumber)
802 /* Try not to run for more than about 3 seconds */
803 static const int DEADLINE = 3;
808 struct afs_q *tq, *uq;
809 int fv_slept, defersleep = 0;
811 afs_uint32 start = osi_Time();
818 for (tq = VLRU.prev; tq != &VLRU && anumber > 0; tq = uq) {
821 if (tvc->f.states & CVFlushed) {
822 refpanic("CVFlushed on VLRU");
823 } else if (i++ > limit) {
824 afs_warn("afs_ShakeLooseVCaches: i %d limit %d afs_vcount %d afs_maxvcount %d\n",
825 (int)i, limit, (int)afs_vcount, (int)afs_maxvcount);
826 refpanic("Found too many AFS vnodes on VLRU (VLRU cycle?)");
827 } else if (QNext(uq) != tq) {
828 refpanic("VLRU inconsistent");
829 } else if (tvc->f.states & CVInit) {
834 evicted = osi_TryEvictVCache(tvc, &fv_slept, defersleep);
841 afs_uint32 now = osi_Time();
846 if (now - start >= DEADLINE) {
847 static afs_uint32 last_warned;
848 /* Warn about this at most every VCACHE_STRESS_LOGINTERVAL secs */
849 if (now < last_warned ||
850 now - last_warned > VCACHE_STRESS_LOGINTERVAL) {
852 afs_warn("afs: Warning: it took us a long time (around "
853 "%d seconds) to try to trim our stat cache "
854 "down to a reasonable size. This may indicate "
855 "someone is accessing an excessive number of "
856 "files, or something is wrong with the AFS "
859 afs_warn("afs: Consider raising the afsd -stat parameter "
860 "(current setting: %d, current vcount: %d), or "
861 "figure out what is accessing so many files.\n",
862 afs_cacheStats, afs_vcount);
869 * This vcache was busy and we slept while trying to evict it.
870 * Move this busy vcache to the head of the VLRU so vcaches
871 * following this busy vcache can be evicted during the retry.
873 QRemove(&tvc->vlruq);
874 QAdd(&VLRU, &tvc->vlruq);
876 goto retry; /* start over - may have raced. */
879 if (anumber && !defersleep) {
887 if (afs_VCacheStressed()) {
889 * If it looks like we have too many vcaches, right after
890 * ShakeLooseVCaches has tried to trim down the number of vcaches, then
891 * maybe -stat should be increased. Log a warning, so if this is
892 * causing problems the user has a chance at noticing.
894 static afs_uint32 last_warned;
895 afs_uint32 now = osi_Time();
897 /* Warn about this at most once every VCACHE_STRESS_LOGINTERVAL secs */
898 if (now - last_warned > VCACHE_STRESS_LOGINTERVAL) {
900 afs_warn("afs: Warning: We are having trouble keeping the AFS stat "
901 "cache trimmed down under the configured limit (current "
902 "-stat setting: %d, current vcache usage: %d).\n",
903 afs_cacheStats, afs_vcount);
904 afs_warn("afs: If AFS access seems slow, consider raising the "
905 "-stat setting for afsd.\n");
912 /* Alloc new vnode. */
914 static struct vcache *
915 afs_AllocVCache(void)
919 tvc = osi_NewVnode();
927 if (afsd_dynamic_vcaches && afs_maxvcount < afs_vcount) {
928 afs_maxvcount = afs_vcount;
929 /*printf("peak vnodes: %d\n", afs_maxvcount);*/
932 afs_stats_cmperf.vcacheXAllocs++; /* count in case we have a leak */
934 /* If we create a new inode, we either give it a new slot number,
935 * or if one's available, use a slot number from the slot free list
937 if (afs_freeSlotList != NULL) {
938 struct afs_slotlist *tmp;
940 tvc->diskSlot = afs_freeSlotList->slot;
941 tmp = afs_freeSlotList;
942 afs_freeSlotList = tmp->next;
943 afs_osi_Free(tmp, sizeof(struct afs_slotlist));
945 tvc->diskSlot = afs_nextVcacheSlot++;
951 /* Pre populate a newly allocated vcache. On platforms where the actual
952 * vnode is attached to the vcache, this function is called before attachment,
953 * therefore it cannot perform any actions on the vnode itself */
956 afs_PrePopulateVCache(struct vcache *avc, struct VenusFid *afid,
957 struct server *serverp) {
961 slot = avc->diskSlot;
963 osi_PrePopulateVCache(avc);
965 avc->diskSlot = slot;
966 QZero(&avc->metadirty);
968 AFS_RWLOCK_INIT(&avc->lock, "vcache lock");
970 memset(&avc->mvid, 0, sizeof(avc->mvid));
971 avc->linkData = NULL;
974 avc->execsOrWriters = 0;
976 avc->f.states = CVInit;
977 avc->last_looker = 0;
979 avc->asynchrony = -1;
983 avc->f.truncPos = AFS_NOTRUNC; /* don't truncate until we need to */
985 afs_SetDataVersion(avc, &zero); /* in case we copy it into flushDV */
987 avc->callback = serverp; /* to minimize chance that clear
990 #if defined(AFS_CACHE_BYPASS)
991 avc->cachingStates = 0;
992 avc->cachingTransitions = 0;
997 afs_FlushAllVCaches(void)
1000 struct vcache *tvc, *nvc;
1002 ObtainWriteLock(&afs_xvcache, 867);
1005 for (i = 0; i < VCSIZE; i++) {
1006 for (tvc = afs_vhashT[i]; tvc; tvc = nvc) {
1010 if (afs_FlushVCache(tvc, &slept)) {
1011 afs_warn("Failed to flush vcache 0x%lx\n", (unsigned long)(uintptrsz)tvc);
1019 ReleaseWriteLock(&afs_xvcache);
1023 * This routine is responsible for allocating a new cache entry
1024 * from the free list. It formats the cache entry and inserts it
1025 * into the appropriate hash tables. It must be called with
1026 * afs_xvcache write-locked so as to prevent several processes from
1027 * trying to create a new cache entry simultaneously.
1029 * LOCK: afs_NewVCache afs_xvcache W
1031 * \param afid The file id of the file whose cache entry is being created.
1033 * \return The new vcache struct.
1036 static_inline struct vcache *
1037 afs_NewVCache_int(struct VenusFid *afid, struct server *serverp, int seq)
1041 afs_int32 anumber = VCACHE_FREE;
1043 AFS_STATCNT(afs_NewVCache);
1045 afs_FlushReclaimedVcaches();
1047 #if defined(AFS_LINUX22_ENV)
1048 if(!afsd_dynamic_vcaches && afs_vcount >= afs_maxvcount) {
1049 afs_ShakeLooseVCaches(anumber);
1050 if (afs_vcount >= afs_maxvcount) {
1051 afs_warn("afs_NewVCache - none freed\n");
1055 tvc = afs_AllocVCache();
1059 #else /* AFS_LINUX22_ENV */
1060 /* pull out a free cache entry */
1062 afs_ShakeLooseVCaches(anumber);
1066 tvc = afs_AllocVCache();
1071 tvc = freeVCList; /* take from free list */
1072 freeVCList = tvc->nextfree;
1073 tvc->nextfree = NULL;
1074 afs_vcount++; /* balanced by FlushVCache */
1075 } /* end of if (!freeVCList) */
1077 #endif /* AFS_LINUX22_ENV */
1079 #if defined(AFS_XBSD_ENV) || defined(AFS_DARWIN_ENV)
1081 panic("afs_NewVCache(): free vcache with vnode attached");
1084 /* Populate the vcache with as much as we can. */
1085 afs_PrePopulateVCache(tvc, afid, serverp);
1087 /* Thread the vcache onto the VLRU */
1092 tvc->hnext = afs_vhashT[i];
1093 afs_vhashT[i] = tvc;
1094 QAdd(&afs_vhashTV[j], &tvc->vhashq);
1096 if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) {
1097 refpanic("NewVCache VLRU inconsistent");
1099 QAdd(&VLRU, &tvc->vlruq); /* put in lruq */
1100 if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) {
1101 refpanic("NewVCache VLRU inconsistent2");
1103 if (tvc->vlruq.next->prev != &(tvc->vlruq)) {
1104 refpanic("NewVCache VLRU inconsistent3");
1106 if (tvc->vlruq.prev->next != &(tvc->vlruq)) {
1107 refpanic("NewVCache VLRU inconsistent4");
1111 /* it should now be safe to drop the xvcache lock - so attach an inode
1112 * to this vcache, where necessary */
1113 osi_AttachVnode(tvc, seq);
1115 /* Get a reference count to hold this vcache for the VLRUQ. Note that
1116 * we have to do this after attaching the vnode, because the reference
1117 * count may be held in the vnode itself */
1119 #if defined(AFS_LINUX22_ENV)
1120 /* Hold it for the LRU (should make count 2) */
1121 osi_Assert(osi_vnhold(tvc) == 0);
1122 #elif !(defined (AFS_DARWIN_ENV) || defined(AFS_XBSD_ENV))
1123 VREFCOUNT_SET(tvc, 1); /* us */
1126 #if defined (AFS_FBSD_ENV)
1127 if (tvc->f.states & CVInit)
1129 afs_PostPopulateVCache(tvc, afid, seq);
1132 } /*afs_NewVCache */
1136 afs_NewVCache(struct VenusFid *afid, struct server *serverp)
1138 return afs_NewVCache_int(afid, serverp, 0);
1142 afs_NewBulkVCache(struct VenusFid *afid, struct server *serverp, int seq)
1144 return afs_NewVCache_int(afid, serverp, seq);
1150 * LOCK: afs_FlushActiveVcaches afs_xvcache N
1152 * \param doflocks : Do we handle flocks?
1155 afs_FlushActiveVcaches(afs_int32 doflocks)
1159 struct afs_conn *tc;
1161 afs_ucred_t *cred = NULL;
1162 struct vrequest *treq = NULL;
1163 struct AFSVolSync tsync;
1166 AFS_STATCNT(afs_FlushActiveVcaches);
1168 code = afs_CreateReq(&treq, afs_osi_credp);
1170 afs_warn("unable to alloc treq\n");
1174 ObtainReadLock(&afs_xvcache);
1175 for (i = 0; i < VCSIZE; i++) {
1176 for (tvc = afs_vhashT[i]; tvc; tvc = tvc->hnext) {
1177 if (tvc->f.states & CVInit) continue;
1178 #ifdef AFS_DARWIN80_ENV
1179 if (tvc->f.states & CDeadVnode &&
1180 (tvc->f.states & (CCore|CUnlinkedDel) ||
1181 tvc->flockCount)) panic("Dead vnode has core/unlinkedel/flock");
1183 if (doflocks && tvc->flockCount != 0) {
1184 struct rx_connection *rxconn;
1185 if (osi_vnhold(tvc) != 0) {
1188 /* if this entry has an flock, send a keep-alive call out */
1189 ReleaseReadLock(&afs_xvcache);
1190 ObtainWriteLock(&tvc->lock, 51);
1192 code = afs_InitReq(treq, afs_osi_credp);
1195 break; /* shutting down: do not try to extend the lock */
1197 treq->flags |= O_NONBLOCK;
1199 tc = afs_Conn(&tvc->f.fid, treq, SHARED_LOCK, &rxconn);
1201 XSTATS_START_TIME(AFS_STATS_FS_RPCIDX_EXTENDLOCK);
1204 RXAFS_ExtendLock(rxconn,
1205 (struct AFSFid *)&tvc->f.fid.Fid,
1211 } while (afs_Analyze
1212 (tc, rxconn, code, &tvc->f.fid, treq,
1213 AFS_STATS_FS_RPCIDX_EXTENDLOCK, SHARED_LOCK, NULL));
1215 ReleaseWriteLock(&tvc->lock);
1216 #ifdef AFS_DARWIN80_ENV
1218 ObtainReadLock(&afs_xvcache);
1220 ObtainReadLock(&afs_xvcache);
1225 if ((tvc->f.states & CCore) || (tvc->f.states & CUnlinkedDel)) {
1227 * Don't let it evaporate in case someone else is in
1228 * this code. Also, drop the afs_xvcache lock while
1229 * getting vcache locks.
1231 if (osi_vnhold(tvc) != 0) {
1234 ReleaseReadLock(&afs_xvcache);
1235 #if defined(AFS_SGI_ENV)
1237 * That's because if we come in via the CUnlinkedDel bit state path we'll be have 0 refcnt
1239 osi_Assert(VREFCOUNT_GT(tvc,0));
1240 AFS_RWLOCK((vnode_t *) tvc, VRWLOCK_WRITE);
1242 ObtainWriteLock(&tvc->lock, 52);
1243 if (tvc->f.states & CCore) {
1244 tvc->f.states &= ~CCore;
1245 /* XXXX Find better place-holder for cred XXXX */
1246 cred = (afs_ucred_t *)tvc->linkData;
1247 tvc->linkData = NULL; /* XXX */
1248 code = afs_InitReq(treq, cred);
1249 afs_Trace2(afs_iclSetp, CM_TRACE_ACTCCORE,
1250 ICL_TYPE_POINTER, tvc, ICL_TYPE_INT32,
1251 tvc->execsOrWriters);
1252 if (!code) { /* avoid store when shutting down */
1253 code = afs_StoreOnLastReference(tvc, treq);
1255 ReleaseWriteLock(&tvc->lock);
1256 hzero(tvc->flushDV);
1259 if (code && code != VNOVNODE) {
1260 afs_StoreWarn(code, tvc->f.fid.Fid.Volume,
1261 /* /dev/console */ 1);
1263 } else if (tvc->f.states & CUnlinkedDel) {
1267 ReleaseWriteLock(&tvc->lock);
1268 #if defined(AFS_SGI_ENV)
1269 AFS_RWUNLOCK((vnode_t *) tvc, VRWLOCK_WRITE);
1271 afs_remunlink(tvc, 0);
1272 #if defined(AFS_SGI_ENV)
1273 AFS_RWLOCK((vnode_t *) tvc, VRWLOCK_WRITE);
1276 /* lost (or won, perhaps) the race condition */
1277 ReleaseWriteLock(&tvc->lock);
1279 #if defined(AFS_SGI_ENV)
1280 AFS_RWUNLOCK((vnode_t *) tvc, VRWLOCK_WRITE);
1282 #ifdef AFS_DARWIN80_ENV
1285 AFS_RELE(AFSTOV(tvc));
1286 /* Matches write code setting CCore flag */
1289 ObtainReadLock(&afs_xvcache);
1291 ObtainReadLock(&afs_xvcache);
1294 AFS_RELE(AFSTOV(tvc));
1295 /* Matches write code setting CCore flag */
1302 ReleaseReadLock(&afs_xvcache);
1303 afs_DestroyReq(treq);
1309 * Make sure a cache entry is up-to-date status-wise.
1311 * NOTE: everywhere that calls this can potentially be sped up
1312 * by checking CStatd first, and avoiding doing the InitReq
1313 * if this is up-to-date.
1315 * Anymore, the only places that call this KNOW already that the
1316 * vcache is not up-to-date, so we don't screw around.
1318 * \param avc : Ptr to vcache entry to verify.
1324 * Make sure a cache entry is up-to-date status-wise.
1326 * NOTE: everywhere that calls this can potentially be sped up
1327 * by checking CStatd first, and avoiding doing the InitReq
1328 * if this is up-to-date.
1330 * Anymore, the only places that call this KNOW already that the
1331 * vcache is not up-to-date, so we don't screw around.
1333 * \param avc Pointer to vcache entry to verify.
1336 * \return 0 for success or other error codes.
1339 afs_VerifyVCache2(struct vcache *avc, struct vrequest *areq)
1343 AFS_STATCNT(afs_VerifyVCache);
1345 /* otherwise we must fetch the status info */
1347 ObtainWriteLock(&avc->lock, 53);
1348 if (avc->f.states & CStatd) {
1349 ReleaseWriteLock(&avc->lock);
1352 afs_StaleVCacheFlags(avc, AFS_STALEVC_FILENAME | AFS_STALEVC_CLEARCB,
1354 ReleaseWriteLock(&avc->lock);
1356 /* fetch the status info */
1357 tvc = afs_GetVCache(&avc->f.fid, areq);
1360 /* Put it back; caller has already incremented vrefCount */
1364 } /*afs_VerifyVCache */
1368 * Simple copy of stat info into cache.
1370 * Callers:as of 1992-04-29, only called by WriteVCache
1372 * \param avc Ptr to vcache entry involved.
1373 * \param astat Ptr to stat info to copy.
1377 afs_SimpleVStat(struct vcache *avc,
1378 struct AFSFetchStatus *astat, struct vrequest *areq)
1381 AFS_STATCNT(afs_SimpleVStat);
1383 #ifdef AFS_64BIT_CLIENT
1384 FillInt64(length, astat->Length_hi, astat->Length);
1385 #else /* AFS_64BIT_CLIENT */
1386 length = astat->Length;
1387 #endif /* AFS_64BIT_CLIENT */
1389 #if defined(AFS_SGI_ENV)
1390 if ((avc->execsOrWriters <= 0) && !afs_DirtyPages(avc)
1391 && !AFS_VN_MAPPED((vnode_t *) avc)) {
1392 osi_Assert((valusema(&avc->vc_rwlock) <= 0)
1393 && (OSI_GET_LOCKID() == avc->vc_rwlockid));
1394 if (length < avc->f.m.Length) {
1395 vnode_t *vp = (vnode_t *) avc;
1397 osi_Assert(WriteLocked(&avc->lock));
1398 ReleaseWriteLock(&avc->lock);
1400 PTOSSVP(vp, (off_t) length, (off_t) MAXLONG);
1402 ObtainWriteLock(&avc->lock, 67);
1407 if (!afs_DirtyPages(avc)) {
1408 /* if actively writing the file, don't fetch over this value */
1409 afs_Trace3(afs_iclSetp, CM_TRACE_SIMPLEVSTAT, ICL_TYPE_POINTER, avc,
1410 ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(avc->f.m.Length),
1411 ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(length));
1412 avc->f.m.Length = length;
1413 avc->f.m.Date = astat->ClientModTime;
1415 avc->f.m.Owner = astat->Owner;
1416 avc->f.m.Group = astat->Group;
1417 avc->f.m.Mode = astat->UnixModeBits;
1418 if (vType(avc) == VREG) {
1419 avc->f.m.Mode |= S_IFREG;
1420 } else if (vType(avc) == VDIR) {
1421 avc->f.m.Mode |= S_IFDIR;
1422 } else if (vType(avc) == VLNK) {
1423 avc->f.m.Mode |= S_IFLNK;
1424 if ((avc->f.m.Mode & 0111) == 0)
1425 avc->mvstat = AFS_MVSTAT_MTPT;
1427 if (avc->f.states & CForeign) {
1428 struct axscache *ac;
1429 avc->f.anyAccess = astat->AnonymousAccess;
1431 if ((astat->CallerAccess & ~astat->AnonymousAccess))
1433 * Caller has at least one bit not covered by anonymous, and
1434 * thus may have interesting rights.
1436 * HOWEVER, this is a really bad idea, because any access query
1437 * for bits which aren't covered by anonymous, on behalf of a user
1438 * who doesn't have any special rights, will result in an answer of
1439 * the form "I don't know, lets make a FetchStatus RPC and find out!"
1440 * It's an especially bad idea under Ultrix, since (due to the lack of
1441 * a proper access() call) it must perform several afs_access() calls
1442 * in order to create magic mode bits that vary according to who makes
1443 * the call. In other words, _every_ stat() generates a test for
1446 #endif /* badidea */
1447 if (avc->Access && (ac = afs_FindAxs(avc->Access, areq->uid)))
1448 ac->axess = astat->CallerAccess;
1449 else /* not found, add a new one if possible */
1450 afs_AddAxs(avc->Access, areq->uid, astat->CallerAccess);
1453 } /*afs_SimpleVStat */
1457 * Store the status info *only* back to the server for a
1460 * Environment: Must be called with a shared lock held on the vnode.
1462 * \param avc Ptr to the vcache entry.
1463 * \param astatus Ptr to the status info to store.
1464 * \param areq Ptr to the associated vrequest.
1466 * \return Operation status.
1470 afs_WriteVCache(struct vcache *avc,
1471 struct AFSStoreStatus *astatus,
1472 struct vrequest *areq)
1475 struct afs_conn *tc;
1476 struct AFSFetchStatus OutStatus;
1477 struct AFSVolSync tsync;
1478 struct rx_connection *rxconn;
1480 AFS_STATCNT(afs_WriteVCache);
1481 afs_Trace2(afs_iclSetp, CM_TRACE_WVCACHE, ICL_TYPE_POINTER, avc,
1482 ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(avc->f.m.Length));
1484 tc = afs_Conn(&avc->f.fid, areq, SHARED_LOCK, &rxconn);
1486 XSTATS_START_TIME(AFS_STATS_FS_RPCIDX_STORESTATUS);
1489 RXAFS_StoreStatus(rxconn, (struct AFSFid *)&avc->f.fid.Fid,
1490 astatus, &OutStatus, &tsync);
1495 } while (afs_Analyze
1496 (tc, rxconn, code, &avc->f.fid, areq, AFS_STATS_FS_RPCIDX_STORESTATUS,
1497 SHARED_LOCK, NULL));
1499 UpgradeSToWLock(&avc->lock, 20);
1501 /* success, do the changes locally */
1502 afs_SimpleVStat(avc, &OutStatus, areq);
1504 * Update the date, too. SimpleVStat didn't do this, since
1505 * it thought we were doing this after fetching new status
1506 * over a file being written.
1508 avc->f.m.Date = OutStatus.ClientModTime;
1510 /* failure, set up to check with server next time */
1511 afs_StaleVCacheFlags(avc, 0, CUnique);
1513 ConvertWToSLock(&avc->lock);
1516 } /*afs_WriteVCache */
1519 * Store status info only locally, set the proper disconnection flags
1520 * and add to dirty list.
1522 * \param avc The vcache to be written locally.
1523 * \param astatus Get attr fields from local store.
1524 * \param attrs This one is only of the vs_size.
1526 * \note Must be called with a shared lock on the vnode
1529 afs_WriteVCacheDiscon(struct vcache *avc,
1530 struct AFSStoreStatus *astatus,
1531 struct vattr *attrs)
1534 afs_int32 flags = 0;
1536 UpgradeSToWLock(&avc->lock, 700);
1538 if (!astatus->Mask) {
1544 /* Set attributes. */
1545 if (astatus->Mask & AFS_SETMODTIME) {
1546 avc->f.m.Date = astatus->ClientModTime;
1547 flags |= VDisconSetTime;
1550 if (astatus->Mask & AFS_SETOWNER) {
1551 /* printf("Not allowed yet. \n"); */
1552 /*avc->f.m.Owner = astatus->Owner;*/
1555 if (astatus->Mask & AFS_SETGROUP) {
1556 /* printf("Not allowed yet. \n"); */
1557 /*avc->f.m.Group = astatus->Group;*/
1560 if (astatus->Mask & AFS_SETMODE) {
1561 avc->f.m.Mode = astatus->UnixModeBits;
1563 flags |= VDisconSetMode;
1564 } /* if(astatus.Mask & AFS_SETMODE) */
1566 } /* if (!astatus->Mask) */
1568 if (attrs->va_size > 0) {
1569 /* XXX: Do I need more checks? */
1570 /* Truncation operation. */
1571 flags |= VDisconTrunc;
1575 afs_DisconAddDirty(avc, flags, 1);
1577 /* XXX: How about the rest of the fields? */
1579 ConvertWToSLock(&avc->lock);
1585 * Copy astat block into vcache info
1587 * \note This code may get dataversion and length out of sync if the file has
1588 * been modified. This is less than ideal. I haven't thought about it sufficiently
1589 * to be certain that it is adequate.
1591 * \note Environment: Must be called under a write lock
1593 * \param avc Ptr to vcache entry.
1594 * \param astat Ptr to stat block to copy in.
1595 * \param areq Ptr to associated request.
1598 afs_ProcessFS(struct vcache *avc,
1599 struct AFSFetchStatus *astat, struct vrequest *areq)
1603 AFS_STATCNT(afs_ProcessFS);
1605 #ifdef AFS_64BIT_CLIENT
1606 FillInt64(length, astat->Length_hi, astat->Length);
1607 #else /* AFS_64BIT_CLIENT */
1608 length = astat->Length;
1609 #endif /* AFS_64BIT_CLIENT */
1610 /* WARNING: afs_DoBulkStat uses the Length field to store a sequence
1611 * number for each bulk status request. Under no circumstances
1612 * should afs_DoBulkStat store a sequence number if the new
1613 * length will be ignored when afs_ProcessFS is called with
1614 * new stats. If you change the following conditional then you
1615 * also need to change the conditional in afs_DoBulkStat. */
1617 if ((avc->execsOrWriters <= 0) && !afs_DirtyPages(avc)
1618 && !AFS_VN_MAPPED((vnode_t *) avc)) {
1620 if ((avc->execsOrWriters <= 0) && !afs_DirtyPages(avc)) {
1622 /* if we're writing or mapping this file, don't fetch over these
1625 afs_Trace3(afs_iclSetp, CM_TRACE_PROCESSFS, ICL_TYPE_POINTER, avc,
1626 ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(avc->f.m.Length),
1627 ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(length));
1628 avc->f.m.Length = length;
1629 avc->f.m.Date = astat->ClientModTime;
1631 hset64(newDV, astat->dataVersionHigh, astat->DataVersion);
1632 afs_SetDataVersion(avc, &newDV);
1633 avc->f.m.Owner = astat->Owner;
1634 avc->f.m.Mode = astat->UnixModeBits;
1635 avc->f.m.Group = astat->Group;
1636 avc->f.m.LinkCount = astat->LinkCount;
1637 if (astat->FileType == File) {
1638 vSetType(avc, VREG);
1639 avc->f.m.Mode |= S_IFREG;
1640 } else if (astat->FileType == Directory) {
1641 vSetType(avc, VDIR);
1642 avc->f.m.Mode |= S_IFDIR;
1643 } else if (astat->FileType == SymbolicLink) {
1644 if (afs_fakestat_enable && (avc->f.m.Mode & 0111) == 0) {
1645 vSetType(avc, VDIR);
1646 avc->f.m.Mode |= S_IFDIR;
1648 vSetType(avc, VLNK);
1649 avc->f.m.Mode |= S_IFLNK;
1651 if ((avc->f.m.Mode & 0111) == 0) {
1652 avc->mvstat = AFS_MVSTAT_MTPT;
1655 avc->f.anyAccess = astat->AnonymousAccess;
1657 if ((astat->CallerAccess & ~astat->AnonymousAccess))
1659 * Caller has at least one bit not covered by anonymous, and
1660 * thus may have interesting rights.
1662 * HOWEVER, this is a really bad idea, because any access query
1663 * for bits which aren't covered by anonymous, on behalf of a user
1664 * who doesn't have any special rights, will result in an answer of
1665 * the form "I don't know, lets make a FetchStatus RPC and find out!"
1666 * It's an especially bad idea under Ultrix, since (due to the lack of
1667 * a proper access() call) it must perform several afs_access() calls
1668 * in order to create magic mode bits that vary according to who makes
1669 * the call. In other words, _every_ stat() generates a test for
1672 #endif /* badidea */
1674 struct axscache *ac;
1675 if (avc->Access && (ac = afs_FindAxs(avc->Access, areq->uid)))
1676 ac->axess = astat->CallerAccess;
1677 else /* not found, add a new one if possible */
1678 afs_AddAxs(avc->Access, areq->uid, astat->CallerAccess);
1680 } /*afs_ProcessFS */
1684 * Get fid from server.
1687 * \param areq Request to be passed on.
1688 * \param name Name of ?? to lookup.
1689 * \param OutStatus Fetch status.
1694 * \return Success status of operation.
1697 afs_RemoteLookup(struct VenusFid *afid, struct vrequest *areq,
1698 char *name, struct VenusFid *nfid,
1699 struct AFSFetchStatus *OutStatusp,
1700 struct AFSCallBack *CallBackp, struct server **serverp,
1701 struct AFSVolSync *tsyncp)
1704 struct afs_conn *tc;
1705 struct rx_connection *rxconn;
1706 struct AFSFetchStatus OutDirStatus;
1709 name = ""; /* XXX */
1711 tc = afs_Conn(afid, areq, SHARED_LOCK, &rxconn);
1714 *serverp = tc->parent->srvr->server;
1715 XSTATS_START_TIME(AFS_STATS_FS_RPCIDX_XLOOKUP);
1718 RXAFS_Lookup(rxconn, (struct AFSFid *)&afid->Fid, name,
1719 (struct AFSFid *)&nfid->Fid, OutStatusp,
1720 &OutDirStatus, CallBackp, tsyncp);
1725 } while (afs_Analyze
1726 (tc, rxconn, code, afid, areq, AFS_STATS_FS_RPCIDX_XLOOKUP, SHARED_LOCK,
1736 * Given a file id and a vrequest structure, fetch the status
1737 * information associated with the file.
1739 * \param afid File ID.
1740 * \param areq Ptr to associated vrequest structure, specifying the
1741 * user whose authentication tokens will be used.
1743 * \note Environment:
1744 * The cache entry is returned with an increased vrefCount field.
1745 * The entry must be discarded by calling afs_PutVCache when you
1746 * are through using the pointer to the cache entry.
1748 * You should not hold any locks when calling this function, except
1749 * locks on other vcache entries. If you lock more than one vcache
1750 * entry simultaneously, you should lock them in this order:
1752 * 1. Lock all files first, then directories.
1753 * 2. Within a particular type, lock entries in Fid.Vnode order.
1755 * This locking hierarchy is convenient because it allows locking
1756 * of a parent dir cache entry, given a file (to check its access
1757 * control list). It also allows renames to be handled easily by
1758 * locking directories in a constant order.
1760 * \note NB. NewVCache -> FlushVCache presently (4/10/95) drops the xvcache lock.
1763 afs_GetVCache(struct VenusFid *afid, struct vrequest *areq)
1766 afs_int32 code, newvcache = 0;
1771 AFS_STATCNT(afs_GetVCache);
1773 #if defined(AFS_SGI_ENV) && !defined(AFS_SGI53_ENV)
1777 ObtainSharedLock(&afs_xvcache, 5);
1779 tvc = afs_FindVCache(afid, &retry, DO_STATS | DO_VLRU | IS_SLOCK);
1781 #if defined(AFS_SGI_ENV) && !defined(AFS_SGI53_ENV)
1782 ReleaseSharedLock(&afs_xvcache);
1783 spunlock_psema(tvc->v.v_lock, retry, &tvc->v.v_sync, PINOD);
1788 osi_Assert((tvc->f.states & CVInit) == 0);
1789 /* If we are in readdir, return the vnode even if not statd */
1790 if ((tvc->f.states & CStatd) || afs_InReadDir(tvc)) {
1791 ReleaseSharedLock(&afs_xvcache);
1795 UpgradeSToWLock(&afs_xvcache, 21);
1797 /* no cache entry, better grab one */
1798 tvc = afs_NewVCache(afid, NULL);
1801 ConvertWToSLock(&afs_xvcache);
1804 ReleaseSharedLock(&afs_xvcache);
1808 afs_stats_cmperf.vcacheMisses++;
1811 ReleaseSharedLock(&afs_xvcache);
1813 ObtainWriteLock(&tvc->lock, 54);
1815 if (tvc->f.states & CStatd) {
1816 ReleaseWriteLock(&tvc->lock);
1819 #ifdef AFS_DARWIN80_ENV
1820 /* Darwin 8.0 only has bufs in nfs, so we shouldn't have to worry about them.
1823 # if defined(AFS_DARWIN_ENV) || defined(AFS_FBSD_ENV)
1825 * XXX - I really don't like this. Should try to understand better.
1826 * It seems that sometimes, when we get called, we already hold the
1827 * lock on the vnode (e.g., from afs_getattr via afs_VerifyVCache).
1828 * We can't drop the vnode lock, because that could result in a race.
1829 * Sometimes, though, we get here and don't hold the vnode lock.
1830 * I hate code paths that sometimes hold locks and sometimes don't.
1831 * In any event, the dodge we use here is to check whether the vnode
1832 * is locked, and if it isn't, then we gain and drop it around the call
1833 * to vinvalbuf; otherwise, we leave it alone.
1836 struct vnode *vp = AFSTOV(tvc);
1839 # if defined(AFS_DARWIN_ENV)
1840 iheldthelock = VOP_ISLOCKED(vp);
1842 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, current_proc());
1843 /* this is messy. we can call fsync which will try to reobtain this */
1844 if (VTOAFS(vp) == tvc)
1845 ReleaseWriteLock(&tvc->lock);
1846 if (UBCINFOEXISTS(vp)) {
1847 vinvalbuf(vp, V_SAVE, &afs_osi_cred, current_proc(), PINOD, 0);
1849 if (VTOAFS(vp) == tvc)
1850 ObtainWriteLock(&tvc->lock, 954);
1852 VOP_UNLOCK(vp, LK_EXCLUSIVE, current_proc());
1853 # elif defined(AFS_FBSD_ENV)
1855 iheldthelock = VOP_ISLOCKED(vp);
1856 if (!iheldthelock) {
1857 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1859 vinvalbuf(vp, V_SAVE, PINOD, 0); /* changed late in 8.0-CURRENT */
1863 # elif defined(AFS_OBSD_ENV)
1864 iheldthelock = VOP_ISLOCKED(vp, curproc);
1866 VOP_LOCK(vp, LK_EXCLUSIVE | LK_RETRY, curproc);
1867 uvm_vnp_uncache(vp);
1869 VOP_UNLOCK(vp, 0, curproc);
1870 # elif defined(AFS_NBSD40_ENV)
1871 iheldthelock = VOP_ISLOCKED(vp);
1872 if (!iheldthelock) {
1873 VOP_LOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1875 uvm_vnp_uncache(vp);
1883 afs_StaleVCacheFlags(tvc, AFS_STALEVC_NODNLC | AFS_STALEVC_CLEARCB,
1886 /* It is always appropriate to throw away all the access rights? */
1887 afs_FreeAllAxs(&(tvc->Access));
1888 tvp = afs_GetVolume(afid, areq, READ_LOCK); /* copy useful per-volume info */
1890 if ((tvp->states & VForeign)) {
1892 tvc->f.states |= CForeign;
1893 if (newvcache && (tvp->rootVnode == afid->Fid.Vnode)
1894 && (tvp->rootUnique == afid->Fid.Unique)) {
1895 tvc->mvstat = AFS_MVSTAT_ROOT;
1898 if (tvp->states & VRO)
1899 tvc->f.states |= CRO;
1900 if (tvp->states & VBackup)
1901 tvc->f.states |= CBackup;
1902 /* now copy ".." entry back out of volume structure, if necessary */
1903 if (tvc->mvstat == AFS_MVSTAT_ROOT && tvp->dotdot.Fid.Volume != 0) {
1904 if (!tvc->mvid.parent)
1905 tvc->mvid.parent = (struct VenusFid *)
1906 osi_AllocSmallSpace(sizeof(struct VenusFid));
1907 *tvc->mvid.parent = tvp->dotdot;
1909 afs_PutVolume(tvp, READ_LOCK);
1913 afs_RemoveVCB(afid);
1915 struct AFSFetchStatus OutStatus;
1917 if (afs_DynrootNewVnode(tvc, &OutStatus)) {
1918 afs_ProcessFS(tvc, &OutStatus, areq);
1919 tvc->f.states |= CStatd | CUnique;
1920 tvc->f.parent.vnode = OutStatus.ParentVnode;
1921 tvc->f.parent.unique = OutStatus.ParentUnique;
1925 if (AFS_IS_DISCONNECTED) {
1926 /* Nothing to do otherwise...*/
1928 /* printf("Network is down in afs_GetCache"); */
1930 code = afs_FetchStatus(tvc, afid, areq, &OutStatus);
1932 /* For the NFS translator's benefit, make sure
1933 * non-directory vnodes always have their parent FID set
1934 * correctly, even when created as a result of decoding an
1935 * NFS filehandle. It would be nice to also do this for
1936 * directories, but we can't because the fileserver fills
1937 * in the FID of the directory itself instead of that of
1940 if (!code && OutStatus.FileType != Directory &&
1941 !tvc->f.parent.vnode) {
1942 tvc->f.parent.vnode = OutStatus.ParentVnode;
1943 tvc->f.parent.unique = OutStatus.ParentUnique;
1944 /* XXX - SXW - It's conceivable we should mark ourselves
1945 * as dirty again here, incase we've been raced
1946 * out of the FetchStatus call.
1953 ReleaseWriteLock(&tvc->lock);
1959 ReleaseWriteLock(&tvc->lock);
1962 } /*afs_GetVCache */
1967 * Lookup a vcache by fid. Look inside the cache first, if not
1968 * there, lookup the file on the server, and then get it's fresh
1976 * \return The found element or NULL.
1979 afs_LookupVCache(struct VenusFid *afid, struct vrequest *areq,
1980 struct vcache *adp, char *aname)
1982 afs_int32 code, now, newvcache = 0;
1983 struct VenusFid nfid;
1986 struct AFSFetchStatus OutStatus;
1987 struct AFSCallBack CallBack;
1988 struct AFSVolSync tsync;
1989 struct server *serverp = 0;
1993 AFS_STATCNT(afs_GetVCache);
1995 #if defined(AFS_SGI_ENV) && !defined(AFS_SGI53_ENV)
1999 ObtainReadLock(&afs_xvcache);
2000 tvc = afs_FindVCache(afid, &retry, DO_STATS /* no vlru */ );
2003 ReleaseReadLock(&afs_xvcache);
2005 #if defined(AFS_SGI_ENV) && !defined(AFS_SGI53_ENV)
2006 spunlock_psema(tvc->v.v_lock, retry, &tvc->v.v_sync, PINOD);
2010 ObtainReadLock(&tvc->lock);
2012 if (tvc->f.states & CStatd) {
2013 ReleaseReadLock(&tvc->lock);
2016 tvc->f.states &= ~CUnique;
2018 ReleaseReadLock(&tvc->lock);
2020 ObtainReadLock(&afs_xvcache);
2023 ReleaseReadLock(&afs_xvcache);
2025 /* lookup the file */
2028 origCBs = afs_allCBs; /* if anything changes, we don't have a cb */
2030 if (AFS_IS_DISCONNECTED) {
2031 /* printf("Network is down in afs_LookupVcache\n"); */
2035 afs_RemoteLookup(&adp->f.fid, areq, aname, &nfid, &OutStatus,
2036 &CallBack, &serverp, &tsync);
2038 #if defined(AFS_SGI_ENV) && !defined(AFS_SGI53_ENV)
2042 ObtainSharedLock(&afs_xvcache, 6);
2043 tvc = afs_FindVCache(&nfid, &retry, DO_VLRU | IS_SLOCK/* no xstats now */ );
2045 #if defined(AFS_SGI_ENV) && !defined(AFS_SGI53_ENV)
2046 ReleaseSharedLock(&afs_xvcache);
2047 spunlock_psema(tvc->v.v_lock, retry, &tvc->v.v_sync, PINOD);
2053 /* no cache entry, better grab one */
2054 UpgradeSToWLock(&afs_xvcache, 22);
2055 tvc = afs_NewVCache(&nfid, serverp);
2057 ConvertWToSLock(&afs_xvcache);
2060 ReleaseSharedLock(&afs_xvcache);
2065 ReleaseSharedLock(&afs_xvcache);
2066 ObtainWriteLock(&tvc->lock, 55);
2068 /* It is always appropriate to throw away all the access rights? */
2069 afs_FreeAllAxs(&(tvc->Access));
2070 tvp = afs_GetVolume(afid, areq, READ_LOCK); /* copy useful per-vol info */
2072 if ((tvp->states & VForeign)) {
2074 tvc->f.states |= CForeign;
2075 if (newvcache && (tvp->rootVnode == afid->Fid.Vnode)
2076 && (tvp->rootUnique == afid->Fid.Unique))
2077 tvc->mvstat = AFS_MVSTAT_ROOT;
2079 if (tvp->states & VRO)
2080 tvc->f.states |= CRO;
2081 if (tvp->states & VBackup)
2082 tvc->f.states |= CBackup;
2083 /* now copy ".." entry back out of volume structure, if necessary */
2084 if (tvc->mvstat == AFS_MVSTAT_ROOT && tvp->dotdot.Fid.Volume != 0) {
2085 if (!tvc->mvid.parent)
2086 tvc->mvid.parent = (struct VenusFid *)
2087 osi_AllocSmallSpace(sizeof(struct VenusFid));
2088 *tvc->mvid.parent = tvp->dotdot;
2093 afs_StaleVCacheFlags(tvc, 0, CUnique);
2095 afs_PutVolume(tvp, READ_LOCK);
2096 ReleaseWriteLock(&tvc->lock);
2101 ObtainWriteLock(&afs_xcbhash, 466);
2102 if (origCBs == afs_allCBs) {
2103 if (CallBack.ExpirationTime) {
2104 tvc->callback = serverp;
2105 tvc->cbExpires = CallBack.ExpirationTime + now;
2106 tvc->f.states |= CStatd | CUnique;
2107 tvc->f.states &= ~CBulkFetching;
2108 afs_QueueCallback(tvc, CBHash(CallBack.ExpirationTime), tvp);
2109 } else if (tvc->f.states & CRO) {
2110 /* adapt gives us an hour. */
2111 tvc->cbExpires = 3600 + osi_Time();
2112 /*XXX*/ tvc->f.states |= CStatd | CUnique;
2113 tvc->f.states &= ~CBulkFetching;
2114 afs_QueueCallback(tvc, CBHash(3600), tvp);
2116 afs_StaleVCacheFlags(tvc,
2117 AFS_STALEVC_CBLOCKED | AFS_STALEVC_CLEARCB,
2121 afs_StaleVCacheFlags(tvc,
2122 AFS_STALEVC_CBLOCKED | AFS_STALEVC_CLEARCB,
2125 ReleaseWriteLock(&afs_xcbhash);
2127 afs_PutVolume(tvp, READ_LOCK);
2128 afs_ProcessFS(tvc, &OutStatus, areq);
2130 ReleaseWriteLock(&tvc->lock);
2136 afs_GetRootVCache(struct VenusFid *afid, struct vrequest *areq,
2137 struct volume *tvolp)
2139 afs_int32 code = 0, i, newvcache = 0, haveStatus = 0;
2140 afs_int32 getNewFid = 0;
2142 struct VenusFid nfid;
2144 struct server *serverp = 0;
2145 struct AFSFetchStatus OutStatus;
2146 struct AFSCallBack CallBack;
2147 struct AFSVolSync tsync;
2149 #ifdef AFS_DARWIN80_ENV
2156 if (!tvolp->rootVnode || getNewFid) {
2157 struct VenusFid tfid;
2160 tfid.Fid.Vnode = 0; /* Means get rootfid of volume */
2161 origCBs = afs_allCBs; /* ignore InitCallBackState */
2163 afs_RemoteLookup(&tfid, areq, NULL, &nfid, &OutStatus, &CallBack,
2168 /* ReleaseReadLock(&tvolp->lock); */
2169 ObtainWriteLock(&tvolp->lock, 56);
2170 tvolp->rootVnode = afid->Fid.Vnode = nfid.Fid.Vnode;
2171 tvolp->rootUnique = afid->Fid.Unique = nfid.Fid.Unique;
2172 ReleaseWriteLock(&tvolp->lock);
2173 /* ObtainReadLock(&tvolp->lock);*/
2176 afid->Fid.Vnode = tvolp->rootVnode;
2177 afid->Fid.Unique = tvolp->rootUnique;
2181 ObtainSharedLock(&afs_xvcache, 7);
2183 for (tvc = afs_vhashT[i]; tvc; tvc = tvc->hnext) {
2184 if (!FidCmp(&(tvc->f.fid), afid)) {
2185 if (tvc->f.states & CVInit) {
2186 ReleaseSharedLock(&afs_xvcache);
2187 afs_osi_Sleep(&tvc->f.states);
2190 #ifdef AFS_DARWIN80_ENV
2191 if (tvc->f.states & CDeadVnode) {
2192 ReleaseSharedLock(&afs_xvcache);
2193 afs_osi_Sleep(&tvc->f.states);
2197 if (vnode_get(tvp)) /* this bumps ref count */
2199 if (vnode_ref(tvp)) {
2201 /* AFSTOV(tvc) may be NULL */
2207 if (osi_vnhold(tvc) != 0) {
2215 if (!haveStatus && (!tvc || !(tvc->f.states & CStatd))) {
2216 /* Mount point no longer stat'd or unknown. FID may have changed. */
2218 #ifdef AFS_DARWIN80_ENV
2219 ReleaseSharedLock(&afs_xvcache);
2222 vnode_put(AFSTOV(tvc));
2223 vnode_rele(AFSTOV(tvc));
2230 ReleaseSharedLock(&afs_xvcache);
2237 UpgradeSToWLock(&afs_xvcache, 23);
2238 /* no cache entry, better grab one */
2239 tvc = afs_NewVCache(afid, NULL);
2242 ReleaseWriteLock(&afs_xvcache);
2246 afs_stats_cmperf.vcacheMisses++;
2248 afs_stats_cmperf.vcacheHits++;
2249 UpgradeSToWLock(&afs_xvcache, 24);
2250 if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) {
2251 refpanic("GRVC VLRU inconsistent0");
2253 if (tvc->vlruq.next->prev != &(tvc->vlruq)) {
2254 refpanic("GRVC VLRU inconsistent1");
2256 if (tvc->vlruq.prev->next != &(tvc->vlruq)) {
2257 refpanic("GRVC VLRU inconsistent2");
2259 QRemove(&tvc->vlruq); /* move to lruq head */
2260 QAdd(&VLRU, &tvc->vlruq);
2261 if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) {
2262 refpanic("GRVC VLRU inconsistent3");
2264 if (tvc->vlruq.next->prev != &(tvc->vlruq)) {
2265 refpanic("GRVC VLRU inconsistent4");
2267 if (tvc->vlruq.prev->next != &(tvc->vlruq)) {
2268 refpanic("GRVC VLRU inconsistent5");
2273 ReleaseWriteLock(&afs_xvcache);
2275 if (tvc->f.states & CStatd) {
2279 ObtainReadLock(&tvc->lock);
2280 tvc->f.states &= ~CUnique;
2281 tvc->callback = NULL; /* redundant, perhaps */
2282 ReleaseReadLock(&tvc->lock);
2285 ObtainWriteLock(&tvc->lock, 57);
2287 /* It is always appropriate to throw away all the access rights? */
2288 afs_FreeAllAxs(&(tvc->Access));
2291 tvc->f.states |= CForeign;
2292 if (tvolp->states & VRO)
2293 tvc->f.states |= CRO;
2294 if (tvolp->states & VBackup)
2295 tvc->f.states |= CBackup;
2296 /* now copy ".." entry back out of volume structure, if necessary */
2297 if (newvcache && (tvolp->rootVnode == afid->Fid.Vnode)
2298 && (tvolp->rootUnique == afid->Fid.Unique)) {
2299 tvc->mvstat = AFS_MVSTAT_ROOT;
2301 if (tvc->mvstat == AFS_MVSTAT_ROOT && tvolp->dotdot.Fid.Volume != 0) {
2302 if (!tvc->mvid.parent)
2303 tvc->mvid.parent = (struct VenusFid *)
2304 osi_AllocSmallSpace(sizeof(struct VenusFid));
2305 *tvc->mvid.parent = tvolp->dotdot;
2309 afs_RemoveVCB(afid);
2312 struct VenusFid tfid;
2315 tfid.Fid.Vnode = 0; /* Means get rootfid of volume */
2316 origCBs = afs_allCBs; /* ignore InitCallBackState */
2318 afs_RemoteLookup(&tfid, areq, NULL, &nfid, &OutStatus, &CallBack,
2323 afs_StaleVCacheFlags(tvc, AFS_STALEVC_CLEARCB, CUnique);
2324 ReleaseWriteLock(&tvc->lock);
2329 ObtainWriteLock(&afs_xcbhash, 468);
2330 if (origCBs == afs_allCBs) {
2331 tvc->f.states |= CTruth;
2332 tvc->callback = serverp;
2333 if (CallBack.ExpirationTime != 0) {
2334 tvc->cbExpires = CallBack.ExpirationTime + start;
2335 tvc->f.states |= CStatd;
2336 tvc->f.states &= ~CBulkFetching;
2337 afs_QueueCallback(tvc, CBHash(CallBack.ExpirationTime), tvolp);
2338 } else if (tvc->f.states & CRO) {
2339 /* adapt gives us an hour. */
2340 tvc->cbExpires = 3600 + osi_Time();
2341 /*XXX*/ tvc->f.states |= CStatd;
2342 tvc->f.states &= ~CBulkFetching;
2343 afs_QueueCallback(tvc, CBHash(3600), tvolp);
2346 afs_StaleVCacheFlags(tvc, AFS_STALEVC_CBLOCKED | AFS_STALEVC_CLEARCB,
2349 ReleaseWriteLock(&afs_xcbhash);
2350 afs_ProcessFS(tvc, &OutStatus, areq);
2352 ReleaseWriteLock(&tvc->lock);
2358 * Update callback status and (sometimes) attributes of a vnode.
2359 * Called after doing a fetch status RPC. Whilst disconnected, attributes
2360 * shouldn't be written to the vcache here.
2365 * \param Outsp Server status after rpc call.
2366 * \param acb Callback for this vnode.
2368 * \note The vcache must be write locked.
2371 afs_UpdateStatus(struct vcache *avc, struct VenusFid *afid,
2372 struct vrequest *areq, struct AFSFetchStatus *Outsp,
2373 struct AFSCallBack *acb, afs_uint32 start)
2375 struct volume *volp;
2378 /* Dont write status in vcache if resyncing after a disconnection. */
2379 afs_ProcessFS(avc, Outsp, areq);
2381 volp = afs_GetVolume(afid, areq, READ_LOCK);
2382 ObtainWriteLock(&afs_xcbhash, 469);
2383 avc->f.states |= CTruth;
2384 if (avc->callback /* check for race */ ) {
2385 if (acb->ExpirationTime != 0) {
2386 avc->cbExpires = acb->ExpirationTime + start;
2387 avc->f.states |= CStatd;
2388 avc->f.states &= ~CBulkFetching;
2389 afs_QueueCallback(avc, CBHash(acb->ExpirationTime), volp);
2390 } else if (avc->f.states & CRO) {
2391 /* ordinary callback on a read-only volume -- AFS 3.2 style */
2392 avc->cbExpires = 3600 + start;
2393 avc->f.states |= CStatd;
2394 avc->f.states &= ~CBulkFetching;
2395 afs_QueueCallback(avc, CBHash(3600), volp);
2397 afs_StaleVCacheFlags(avc,
2398 AFS_STALEVC_CBLOCKED | AFS_STALEVC_CLEARCB,
2402 afs_StaleVCacheFlags(avc, AFS_STALEVC_CBLOCKED | AFS_STALEVC_CLEARCB,
2405 ReleaseWriteLock(&afs_xcbhash);
2407 afs_PutVolume(volp, READ_LOCK);
2411 afs_BadFetchStatus(struct afs_conn *tc)
2413 int addr = ntohl(tc->parent->srvr->sa_ip);
2414 afs_warn("afs: Invalid AFSFetchStatus from server %u.%u.%u.%u\n",
2415 (addr >> 24) & 0xff, (addr >> 16) & 0xff, (addr >> 8) & 0xff,
2417 afs_warn("afs: This suggests the server may be sending bad data that "
2418 "can lead to availability issues or data corruption. The "
2419 "issue has been avoided for now, but it may not always be "
2420 "detectable. Please upgrade the server if possible.\n");
2424 * Check if a given AFSFetchStatus structure is sane.
2426 * @param[in] tc The server from which we received the status
2427 * @param[in] status The status we received
2429 * @return whether the given structure is valid or not
2430 * @retval 0 the structure is fine
2431 * @retval nonzero the structure looks like garbage; act as if we received
2432 * the returned error code from the server
2435 afs_CheckFetchStatus(struct afs_conn *tc, struct AFSFetchStatus *status)
2437 if (status->errorCode ||
2438 status->InterfaceVersion != 1 ||
2439 !(status->FileType > Invalid && status->FileType <= SymbolicLink) ||
2440 status->ParentVnode == 0 || status->ParentUnique == 0) {
2442 afs_warn("afs: FetchStatus ec %u iv %u ft %u pv %u pu %u\n",
2443 (unsigned)status->errorCode, (unsigned)status->InterfaceVersion,
2444 (unsigned)status->FileType, (unsigned)status->ParentVnode,
2445 (unsigned)status->ParentUnique);
2446 afs_BadFetchStatus(tc);
2454 * Must be called with avc write-locked
2455 * don't absolutely have to invalidate the hint unless the dv has
2456 * changed, but be sure to get it right else there will be consistency bugs.
2459 afs_FetchStatus(struct vcache * avc, struct VenusFid * afid,
2460 struct vrequest * areq, struct AFSFetchStatus * Outsp)
2463 afs_uint32 start = 0;
2464 struct afs_conn *tc;
2465 struct AFSCallBack CallBack;
2466 struct AFSVolSync tsync;
2467 struct rx_connection *rxconn;
2470 tc = afs_Conn(afid, areq, SHARED_LOCK, &rxconn);
2471 avc->dchint = NULL; /* invalidate hints */
2473 avc->callback = tc->parent->srvr->server;
2475 XSTATS_START_TIME(AFS_STATS_FS_RPCIDX_FETCHSTATUS);
2478 RXAFS_FetchStatus(rxconn, (struct AFSFid *)&afid->Fid, Outsp,
2485 code = afs_CheckFetchStatus(tc, Outsp);
2490 } while (afs_Analyze
2491 (tc, rxconn, code, afid, areq, AFS_STATS_FS_RPCIDX_FETCHSTATUS,
2492 SHARED_LOCK, NULL));
2495 afs_UpdateStatus(avc, afid, areq, Outsp, &CallBack, start);
2497 /* used to undo the local callback, but that's too extreme.
2498 * There are plenty of good reasons that fetchstatus might return
2499 * an error, such as EPERM. If we have the vnode cached, statd,
2500 * with callback, might as well keep track of the fact that we
2501 * don't have access...
2503 if (code == EPERM || code == EACCES) {
2504 struct axscache *ac;
2505 if (avc->Access && (ac = afs_FindAxs(avc->Access, areq->uid)))
2507 else /* not found, add a new one if possible */
2508 afs_AddAxs(avc->Access, areq->uid, 0);
2515 * Decrements the reference count on a cache entry.
2517 * \param avc Pointer to the cache entry to decrement.
2519 * \note Environment: Nothing interesting.
2522 afs_PutVCache(struct vcache *avc)
2524 AFS_STATCNT(afs_PutVCache);
2525 #ifdef AFS_DARWIN80_ENV
2526 vnode_put(AFSTOV(avc));
2530 * Can we use a read lock here?
2532 ObtainReadLock(&afs_xvcache);
2534 ReleaseReadLock(&afs_xvcache);
2536 } /*afs_PutVCache */
2540 * Reset a vcache entry, so local contents are ignored, and the
2541 * server will be reconsulted next time the vcache is used
2543 * \param avc Pointer to the cache entry to reset
2545 * \param skipdnlc skip the dnlc purge for this vnode
2547 * \note avc must be write locked on entry
2549 * \note The caller should purge the dnlc when skipdnlc is set.
2552 afs_ResetVCache(struct vcache *avc, afs_ucred_t *acred, afs_int32 skipdnlc)
2554 afs_stalevc_flags_t flags = 0;
2556 flags |= AFS_STALEVC_NODNLC;
2559 afs_StaleVCacheFlags(avc, flags, CDirty); /* next reference will re-stat */
2560 /* now find the disk cache entries */
2561 afs_TryToSmush(avc, acred, 1);
2562 if (avc->linkData && !(avc->f.states & CCore)) {
2563 afs_osi_Free(avc->linkData, strlen(avc->linkData) + 1);
2564 avc->linkData = NULL;
2569 * Sleepa when searching for a vcache. Releases all the pending locks,
2570 * sleeps then obtains the previously released locks.
2572 * \param vcache Enter sleep state.
2573 * \param flag Determines what locks to use.
2578 findvc_sleep(struct vcache *avc, int flag)
2580 if (flag & IS_SLOCK) {
2581 ReleaseSharedLock(&afs_xvcache);
2583 if (flag & IS_WLOCK) {
2584 ReleaseWriteLock(&afs_xvcache);
2586 ReleaseReadLock(&afs_xvcache);
2589 afs_osi_Sleep(&avc->f.states);
2590 if (flag & IS_SLOCK) {
2591 ObtainSharedLock(&afs_xvcache, 341);
2593 if (flag & IS_WLOCK) {
2594 ObtainWriteLock(&afs_xvcache, 343);
2596 ObtainReadLock(&afs_xvcache);
2602 * Add a reference on an existing vcache entry.
2604 * \param tvc Pointer to the vcache.
2606 * \note Environment: Must be called with at least one reference from
2607 * elsewhere on the vcache, even if that reference will be dropped.
2608 * The global lock is required.
2610 * \return 0 on success, -1 on failure.
2614 afs_RefVCache(struct vcache *tvc)
2616 #ifdef AFS_DARWIN80_ENV
2620 /* AFS_STATCNT(afs_RefVCache); */
2622 #ifdef AFS_DARWIN80_ENV
2626 if (vnode_ref(tvp)) {
2628 /* AFSTOV(tvc) may be NULL */
2634 if (osi_vnhold(tvc) != 0) {
2639 } /*afs_RefVCache */
2642 * Find a vcache entry given a fid.
2644 * \param afid Pointer to the fid whose cache entry we desire.
2645 * \param retry (SGI-specific) tell the caller to drop the lock on xvcache,
2646 * unlock the vnode, and try again.
2647 * \param flag Bit 1 to specify whether to compute hit statistics. Not
2648 * set if FindVCache is called as part of internal bookkeeping.
2650 * \note Environment: Must be called with the afs_xvcache lock at least held at
2651 * the read level. In order to do the VLRU adjustment, the xvcache lock
2652 * must be shared-- we upgrade it here.
2656 afs_FindVCache(struct VenusFid *afid, afs_int32 * retry, afs_int32 flag)
2661 #ifdef AFS_DARWIN80_ENV
2662 struct vcache *deadvc = NULL, *livevc = NULL;
2666 AFS_STATCNT(afs_FindVCache);
2670 for (tvc = afs_vhashT[i]; tvc; tvc = tvc->hnext) {
2671 if (FidMatches(afid, tvc)) {
2672 if (tvc->f.states & CVInit) {
2673 findvc_sleep(tvc, flag);
2676 #ifdef AFS_DARWIN80_ENV
2677 if (tvc->f.states & CDeadVnode) {
2678 findvc_sleep(tvc, flag);
2686 /* should I have a read lock on the vnode here? */
2688 #if defined(AFS_DARWIN80_ENV)
2692 if (tvp && vnode_ref(tvp)) {
2694 /* AFSTOV(tvc) may be NULL */
2703 #elif defined(AFS_DARWIN_ENV)
2704 tvc->f.states |= CUBCinit;
2706 if (UBCINFOMISSING(AFSTOV(tvc)) ||
2707 UBCINFORECLAIMED(AFSTOV(tvc))) {
2708 ubc_info_init(AFSTOV(tvc));
2711 tvc->f.states &= ~CUBCinit;
2713 if (osi_vnhold(tvc) != 0) {
2720 * only move to front of vlru if we have proper vcache locking)
2722 if (flag & DO_VLRU) {
2723 if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) {
2724 refpanic("FindVC VLRU inconsistent1");
2726 if (tvc->vlruq.next->prev != &(tvc->vlruq)) {
2727 refpanic("FindVC VLRU inconsistent1");
2729 if (tvc->vlruq.prev->next != &(tvc->vlruq)) {
2730 refpanic("FindVC VLRU inconsistent2");
2732 UpgradeSToWLock(&afs_xvcache, 26);
2733 QRemove(&tvc->vlruq);
2734 QAdd(&VLRU, &tvc->vlruq);
2735 ConvertWToSLock(&afs_xvcache);
2736 if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) {
2737 refpanic("FindVC VLRU inconsistent1");
2739 if (tvc->vlruq.next->prev != &(tvc->vlruq)) {
2740 refpanic("FindVC VLRU inconsistent2");
2742 if (tvc->vlruq.prev->next != &(tvc->vlruq)) {
2743 refpanic("FindVC VLRU inconsistent3");
2749 if (flag & DO_STATS) {
2751 afs_stats_cmperf.vcacheHits++;
2753 afs_stats_cmperf.vcacheMisses++;
2754 if (afs_IsPrimaryCellNum(afid->Cell))
2755 afs_stats_cmperf.vlocalAccesses++;
2757 afs_stats_cmperf.vremoteAccesses++;
2760 } /*afs_FindVCache */
2763 * Find a vcache entry given a fid. Does a wildcard match on what we
2764 * have for the fid. If more than one entry, don't return anything.
2766 * \param avcp Fill in pointer if we found one and only one.
2767 * \param afid Pointer to the fid whose cache entry we desire.
2768 * \param retry (SGI-specific) tell the caller to drop the lock on xvcache,
2769 * unlock the vnode, and try again.
2770 * \param flags bit 1 to specify whether to compute hit statistics. Not
2771 * set if FindVCache is called as part of internal bookkeeping.
2773 * \note Environment: Must be called with the afs_xvcache lock at least held at
2774 * the read level. In order to do the VLRU adjustment, the xvcache lock
2775 * must be shared-- we upgrade it here.
2777 * \return Number of matches found.
2780 int afs_duplicate_nfs_fids = 0;
2783 afs_NFSFindVCache(struct vcache **avcp, struct VenusFid *afid)
2787 afs_int32 count = 0;
2788 struct vcache *found_tvc = NULL;
2789 #ifdef AFS_DARWIN80_ENV
2793 AFS_STATCNT(afs_FindVCache);
2797 ObtainSharedLock(&afs_xvcache, 331);
2800 for (tvc = afs_vhashT[i]; tvc; tvc = tvc->hnext) {
2801 /* Match only on what we have.... */
2802 if (((tvc->f.fid.Fid.Vnode & 0xffff) == afid->Fid.Vnode)
2803 && (tvc->f.fid.Fid.Volume == afid->Fid.Volume)
2804 && ((tvc->f.fid.Fid.Unique & 0xffffff) == afid->Fid.Unique)
2805 && (tvc->f.fid.Cell == afid->Cell)) {
2806 if (tvc->f.states & CVInit) {
2807 ReleaseSharedLock(&afs_xvcache);
2808 afs_osi_Sleep(&tvc->f.states);
2811 #ifdef AFS_DARWIN80_ENV
2812 if (tvc->f.states & CDeadVnode) {
2813 ReleaseSharedLock(&afs_xvcache);
2814 afs_osi_Sleep(&tvc->f.states);
2818 if (vnode_get(tvp)) {
2819 /* This vnode no longer exists. */
2822 if (vnode_ref(tvp)) {
2823 /* This vnode no longer exists. */
2825 /* AFSTOV(tvc) may be NULL */
2831 if (osi_vnhold(tvc) != 0) {
2834 #endif /* AFS_DARWIN80_ENV */
2838 afs_duplicate_nfs_fids++;
2839 #ifndef AFS_DARWIN80_ENV
2841 AFS_FAST_RELE(found_tvc);
2843 ReleaseSharedLock(&afs_xvcache);
2844 #ifdef AFS_DARWIN80_ENV
2845 /* Drop our reference counts. */
2846 vnode_put(AFSTOV(tvc));
2847 vnode_put(AFSTOV(found_tvc));
2856 /* should I have a read lock on the vnode here? */
2859 * We obtained the xvcache lock above.
2861 if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) {
2862 refpanic("FindVC VLRU inconsistent1");
2864 if (tvc->vlruq.next->prev != &(tvc->vlruq)) {
2865 refpanic("FindVC VLRU inconsistent1");
2867 if (tvc->vlruq.prev->next != &(tvc->vlruq)) {
2868 refpanic("FindVC VLRU inconsistent2");
2870 UpgradeSToWLock(&afs_xvcache, 568);
2871 QRemove(&tvc->vlruq);
2872 QAdd(&VLRU, &tvc->vlruq);
2873 ConvertWToSLock(&afs_xvcache);
2874 if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) {
2875 refpanic("FindVC VLRU inconsistent1");
2877 if (tvc->vlruq.next->prev != &(tvc->vlruq)) {
2878 refpanic("FindVC VLRU inconsistent2");
2880 if (tvc->vlruq.prev->next != &(tvc->vlruq)) {
2881 refpanic("FindVC VLRU inconsistent3");
2887 afs_stats_cmperf.vcacheHits++;
2889 afs_stats_cmperf.vcacheMisses++;
2890 if (afs_IsPrimaryCellNum(afid->Cell))
2891 afs_stats_cmperf.vlocalAccesses++;
2893 afs_stats_cmperf.vremoteAccesses++;
2895 *avcp = tvc; /* May be null */
2897 ReleaseSharedLock(&afs_xvcache);
2898 return (tvc ? 1 : 0);
2900 } /*afs_NFSFindVCache */
2906 * Initialize vcache related variables
2911 afs_vcacheInit(int astatSize)
2913 #if !defined(AFS_LINUX22_ENV)
2917 if (!afs_maxvcount) {
2918 afs_maxvcount = astatSize; /* no particular limit on linux? */
2920 #if !defined(AFS_LINUX22_ENV)
2924 AFS_RWLOCK_INIT(&afs_xvcache, "afs_xvcache");
2925 LOCK_INIT(&afs_xvcb, "afs_xvcb");
2927 #if !defined(AFS_LINUX22_ENV)
2928 /* Allocate and thread the struct vcache entries */
2929 tvp = afs_osi_Alloc(astatSize * sizeof(struct vcache));
2930 osi_Assert(tvp != NULL);
2931 memset(tvp, 0, sizeof(struct vcache) * astatSize);
2933 Initial_freeVCList = tvp;
2934 freeVCList = &(tvp[0]);
2935 for (i = 0; i < astatSize - 1; i++) {
2936 tvp[i].nextfree = &(tvp[i + 1]);
2938 tvp[astatSize - 1].nextfree = NULL;
2939 # ifdef KERNEL_HAVE_PIN
2940 pin((char *)tvp, astatSize * sizeof(struct vcache)); /* XXX */
2944 #if defined(AFS_SGI_ENV)
2945 for (i = 0; i < astatSize; i++) {
2946 char name[METER_NAMSZ];
2947 struct vcache *tvc = &tvp[i];
2949 tvc->v.v_number = ++afsvnumbers;
2950 tvc->vc_rwlockid = OSI_NO_LOCKID;
2951 initnsema(&tvc->vc_rwlock, 1,
2952 makesname(name, "vrw", tvc->v.v_number));
2953 # ifndef AFS_SGI53_ENV
2954 initnsema(&tvc->v.v_sync, 0, makesname(name, "vsy", tvc->v.v_number));
2956 # ifndef AFS_SGI62_ENV
2957 initnlock(&tvc->v.v_lock, makesname(name, "vlk", tvc->v.v_number));
2958 # endif /* AFS_SGI62_ENV */
2962 for(i = 0; i < VCSIZE; ++i)
2963 QInit(&afs_vhashTV[i]);
2970 shutdown_vcache(void)
2973 struct afs_cbr *tsp;
2975 * XXX We may potentially miss some of the vcaches because if when
2976 * there are no free vcache entries and all the vcache entries are active
2977 * ones then we allocate an additional one - admittedly we almost never
2982 struct afs_q *tq, *uq = NULL;
2984 for (tq = VLRU.prev; tq != &VLRU; tq = uq) {
2987 if (tvc->mvid.target_root) {
2988 osi_FreeSmallSpace(tvc->mvid.target_root);
2989 tvc->mvid.target_root = NULL;
2992 aix_gnode_rele(AFSTOV(tvc));
2994 if (tvc->linkData) {
2995 afs_osi_Free(tvc->linkData, strlen(tvc->linkData) + 1);
3000 * Also free the remaining ones in the Cache
3002 for (i = 0; i < VCSIZE; i++) {
3003 for (tvc = afs_vhashT[i]; tvc; tvc = tvc->hnext) {
3004 if (tvc->mvid.target_root) {
3005 osi_FreeSmallSpace(tvc->mvid.target_root);
3006 tvc->mvid.target_root = NULL;
3010 afs_osi_Free(tvc->v.v_gnode, sizeof(struct gnode));
3011 # ifdef AFS_AIX32_ENV
3014 vms_delete(tvc->segid);
3016 tvc->segid = tvc->vmh = NULL;
3017 if (VREFCOUNT_GT(tvc,0))
3018 osi_Panic("flushVcache: vm race");
3026 #if defined(AFS_SUN5_ENV)
3032 if (tvc->linkData) {
3033 afs_osi_Free(tvc->linkData, strlen(tvc->linkData) + 1);
3038 afs_FreeAllAxs(&(tvc->Access));
3044 * Free any leftover callback queue
3046 for (i = 0; i < afs_stats_cmperf.CallBackAlloced; i++) {
3047 tsp = afs_cbrHeads[i];
3048 afs_cbrHeads[i] = 0;
3049 afs_osi_Free((char *)tsp, AFS_NCBRS * sizeof(struct afs_cbr));
3053 #if !defined(AFS_LINUX22_ENV)
3054 afs_osi_Free(Initial_freeVCList, afs_cacheStats * sizeof(struct vcache));
3056 # ifdef KERNEL_HAVE_PIN
3057 unpin(Initial_freeVCList, afs_cacheStats * sizeof(struct vcache));
3060 freeVCList = Initial_freeVCList = 0;
3063 AFS_RWLOCK_INIT(&afs_xvcache, "afs_xvcache");
3064 LOCK_INIT(&afs_xvcb, "afs_xvcb");
3066 for(i = 0; i < VCSIZE; ++i)
3067 QInit(&afs_vhashTV[i]);
3071 afs_DisconGiveUpCallbacks(void)
3077 ObtainWriteLock(&afs_xvcache, 1002); /* XXX - should be a unique number */
3080 /* Somehow, walk the set of vcaches, with each one coming out as tvc */
3081 for (i = 0; i < VCSIZE; i++) {
3082 for (tvc = afs_vhashT[i]; tvc; tvc = tvc->hnext) {
3084 if (afs_QueueVCB(tvc, &slept)) {
3085 tvc->callback = NULL;
3094 ReleaseWriteLock(&afs_xvcache);
3101 * Clear the Statd flag from all vcaches
3103 * This function removes the Statd flag from all vcaches. It's used by
3104 * disconnected mode to tidy up during reconnection
3108 afs_ClearAllStatdFlag(void)
3113 ObtainWriteLock(&afs_xvcache, 715);
3115 for (i = 0; i < VCSIZE; i++) {
3116 for (tvc = afs_vhashT[i]; tvc; tvc = tvc->hnext) {
3117 afs_StaleVCacheFlags(tvc, AFS_STALEVC_NODNLC | AFS_STALEVC_NOCB,
3121 ReleaseWriteLock(&afs_xvcache);
3125 * Mark a vcache as stale; our metadata for the relevant file may be out of
3128 * @post Any subsequent access to this vcache will cause us to fetch the
3129 * metadata for this vcache again.
3132 afs_StaleVCacheFlags(struct vcache *avc, afs_stalevc_flags_t flags,
3136 int do_filename = 0;
3138 int lock_cbhash = 1;
3140 if ((flags & AFS_STALEVC_NODNLC)) {
3143 if ((flags & AFS_STALEVC_FILENAME)) {
3146 if ((flags & AFS_STALEVC_CBLOCKED)) {
3149 if ((flags & AFS_STALEVC_NOCB)) {
3155 ObtainWriteLock(&afs_xcbhash, 486);
3158 afs_DequeueCallback(avc);
3162 avc->f.states &= ~cflags;
3165 ReleaseWriteLock(&afs_xcbhash);
3168 if ((flags & AFS_STALEVC_SKIP_DNLC_FOR_INIT_FLUSHED) &&
3169 (avc->f.states & (CVInit | CVFlushed))) {
3173 if (flags & AFS_STALEVC_CLEARCB) {
3174 avc->callback = NULL;
3178 if ((avc->f.fid.Fid.Vnode & 1) ||
3179 AFSTOV(avc) == NULL || vType(avc) == VDIR ||
3180 (avc->f.states & CForeign)) {
3181 /* This vcache is (or could be) a directory. */
3182 osi_dnlc_purgedp(avc);
3184 } else if (do_filename) {
3185 osi_dnlc_purgevp(avc);
3191 afs_SetDataVersion(struct vcache *avc, afs_hyper_t *avers)
3193 hset(avc->f.m.DataVersion, *avers);