/* * Copyright 2000, International Business Machines Corporation and others. * All Rights Reserved. * * This software has been released under the terms of the IBM Public * License. For details, see the LICENSE file in the top-level source * directory or online at http://www.openafs.org/dl/license10.html */ /* * Implements: * afs_FlushVCache * afs_AllocCBR * afs_FreeCBR * afs_FlushVCBs * afs_QueueVCB * afs_RemoveVCB * afs_NewVCache * afs_FlushActiveVcaches * afs_VerifyVCache2 * afs_WriteVCache * afs_WriteVCacheDiscon * afs_SimpleVStat * afs_ProcessFS * TellALittleWhiteLie * afs_RemoteLookup * afs_GetVCache * afs_LookupVCache * afs_GetRootVCache * afs_UpdateStatus * afs_FetchStatus * afs_StuffVcache * afs_PutVCache * afs_FindVCache * afs_NFSFindVCache * afs_vcacheInit * shutdown_vcache * */ #include #include "afs/param.h" #include "afs/sysincludes.h" /*Standard vendor system headers */ #include "afsincludes.h" /*AFS-based standard headers */ #include "afs/afs_stats.h" #include "afs/afs_cbqueue.h" #include "afs/afs_osidnlc.h" afs_int32 afs_maxvcount = 0; /* max number of vcache entries */ afs_int32 afs_vcount = 0; /* number of vcache in use now */ #ifdef AFS_SGI_ENV int afsvnumbers = 0; #endif #ifdef AFS_SGI64_ENV char *makesname(); #endif /* AFS_SGI64_ENV */ /* Exported variables */ afs_rwlock_t afs_xvcdirty; /*Lock: discon vcache dirty list mgmt */ afs_rwlock_t afs_xvcache; /*Lock: alloc new stat cache entries */ afs_rwlock_t afs_xvreclaim; /*Lock: entries reclaimed, not on free list */ afs_lock_t afs_xvcb; /*Lock: fids on which there are callbacks */ #if !defined(AFS_LINUX22_ENV) static struct vcache *freeVCList; /*Free list for stat cache entries */ struct vcache *ReclaimedVCList; /*Reclaimed list for stat entries */ static struct vcache *Initial_freeVCList; /*Initial list for above */ #endif struct afs_q VLRU; /*vcache LRU */ afs_int32 vcachegen = 0; unsigned int afs_paniconwarn = 0; struct vcache *afs_vhashT[VCSIZE]; struct afs_q afs_vhashTV[VCSIZE]; static struct afs_cbr *afs_cbrHashT[CBRSIZE]; afs_int32 afs_bulkStatsLost; int afs_norefpanic = 0; /* Disk backed vcache definitions * Both protected by xvcache */ static int afs_nextVcacheSlot = 0; static struct afs_slotlist *afs_freeSlotList = NULL; /* Forward declarations */ static afs_int32 afs_QueueVCB(struct vcache *avc); /*! * Generate an index into the hash table for a given Fid. * \param fid * \return The hash value. */ static int afs_HashCBRFid(struct AFSFid *fid) { return (fid->Volume + fid->Vnode + fid->Unique) % CBRSIZE; } /*! * Insert a CBR entry into the hash table. * Must be called with afs_xvcb held. * \param cbr * \return */ static void afs_InsertHashCBR(struct afs_cbr *cbr) { int slot = afs_HashCBRFid(&cbr->fid); cbr->hash_next = afs_cbrHashT[slot]; if (afs_cbrHashT[slot]) afs_cbrHashT[slot]->hash_pprev = &cbr->hash_next; cbr->hash_pprev = &afs_cbrHashT[slot]; afs_cbrHashT[slot] = cbr; } /*! * * Flush the given vcache entry. * * Environment: * afs_xvcache lock must be held for writing upon entry to * prevent people from changing the vrefCount field, and to * protect the lruq and hnext fields. * LOCK: afs_FlushVCache afs_xvcache W * REFCNT: vcache ref count must be zero on entry except for osf1 * RACE: lock is dropped and reobtained, permitting race in caller * * \param avc Pointer to vcache entry to flush. * \param slept Pointer to int to set 1 if we sleep/drop locks, 0 if we don't. * */ int afs_FlushVCache(struct vcache *avc, int *slept) { /*afs_FlushVCache */ afs_int32 i, code; struct vcache **uvc, *wvc; *slept = 0; AFS_STATCNT(afs_FlushVCache); afs_Trace2(afs_iclSetp, CM_TRACE_FLUSHV, ICL_TYPE_POINTER, avc, ICL_TYPE_INT32, avc->f.states); code = osi_VM_FlushVCache(avc, slept); if (code) goto bad; if (avc->f.states & CVFlushed) { code = EBUSY; goto bad; } #if !defined(AFS_LINUX22_ENV) if (avc->nextfree || !avc->vlruq.prev || !avc->vlruq.next) { /* qv afs.h */ refpanic("LRU vs. Free inconsistency"); } #endif avc->f.states |= CVFlushed; /* pull the entry out of the lruq and put it on the free list */ QRemove(&avc->vlruq); /* keep track of # of files that we bulk stat'd, but never used * before they got recycled. */ if (avc->f.states & CBulkStat) afs_bulkStatsLost++; vcachegen++; /* remove entry from the hash chain */ i = VCHash(&avc->f.fid); uvc = &afs_vhashT[i]; for (wvc = *uvc; wvc; uvc = &wvc->hnext, wvc = *uvc) { if (avc == wvc) { *uvc = avc->hnext; avc->hnext = (struct vcache *)NULL; break; } } /* remove entry from the volume hash table */ QRemove(&avc->vhashq); if (avc->mvid) osi_FreeSmallSpace(avc->mvid); avc->mvid = (struct VenusFid *)0; if (avc->linkData) { afs_osi_Free(avc->linkData, strlen(avc->linkData) + 1); avc->linkData = NULL; } #if defined(AFS_XBSD_ENV) || defined(AFS_DARWIN_ENV) /* OK, there are no internal vrefCounts, so there shouldn't * be any more refs here. */ if (avc->v) { #ifdef AFS_DARWIN80_ENV vnode_clearfsnode(AFSTOV(avc)); vnode_removefsref(AFSTOV(avc)); #else avc->v->v_data = NULL; /* remove from vnode */ #endif AFSTOV(avc) = NULL; /* also drop the ptr to vnode */ } #endif #ifdef AFS_SUN510_ENV /* As we use private vnodes, cleanup is up to us */ vn_reinit(AFSTOV(avc)); #endif afs_FreeAllAxs(&(avc->Access)); afs_QueueVCB(avc); ObtainWriteLock(&afs_xcbhash, 460); afs_DequeueCallback(avc); /* remove it from queued callbacks list */ avc->f.states &= ~(CStatd | CUnique); ReleaseWriteLock(&afs_xcbhash); if ((avc->f.states & CForeign) || (avc->f.fid.Fid.Vnode & 1)) osi_dnlc_purgedp(avc); /* if it (could be) a directory */ else osi_dnlc_purgevp(avc); /* * Next, keep track of which vnodes we've deleted for create's * optimistic synchronization algorithm */ afs_allZaps++; if (avc->f.fid.Fid.Vnode & 1) afs_oddZaps++; else afs_evenZaps++; afs_vcount--; #if !defined(AFS_LINUX22_ENV) /* put the entry in the free list */ avc->nextfree = freeVCList; freeVCList = avc; if (avc->vlruq.prev || avc->vlruq.next) { refpanic("LRU vs. Free inconsistency"); } avc->f.states |= CVFlushed; #else /* This should put it back on the vnode free list since usecount is 1 */ vSetType(avc, VREG); if (VREFCOUNT_GT(avc,0)) { AFS_RELE(AFSTOV(avc)); afs_stats_cmperf.vcacheXAllocs--; } else { if (afs_norefpanic) { afs_warn("flush vc refcnt < 1"); afs_norefpanic++; } else osi_Panic("flush vc refcnt < 1"); } #endif /* AFS_LINUX22_ENV */ return 0; bad: return code; } /*afs_FlushVCache */ #ifndef AFS_SGI_ENV /*! * The core of the inactive vnode op for all but IRIX. * * \param avc * \param acred */ void afs_InactiveVCache(struct vcache *avc, afs_ucred_t *acred) { AFS_STATCNT(afs_inactive); if (avc->f.states & CDirty) { /* we can't keep trying to push back dirty data forever. Give up. */ afs_InvalidateAllSegments(avc); /* turns off dirty bit */ } avc->f.states &= ~CMAPPED; /* mainly used by SunOS 4.0.x */ avc->f.states &= ~CDirty; /* Turn it off */ if (avc->f.states & CUnlinked) { if (CheckLock(&afs_xvcache) || CheckLock(&afs_xdcache)) { avc->f.states |= CUnlinkedDel; return; } afs_remunlink(avc, 1); /* ignore any return code */ } } #endif /*! * Allocate a callback return structure from the * free list and return it. * * Environment: The alloc and free routines are both called with the afs_xvcb lock * held, so we don't have to worry about blocking in osi_Alloc. * * \return The allocated afs_cbr. */ static struct afs_cbr *afs_cbrSpace = 0; /* if alloc limit below changes, fix me! */ static struct afs_cbr *afs_cbrHeads[16]; struct afs_cbr * afs_AllocCBR(void) { register struct afs_cbr *tsp; int i; if (!afs_cbrSpace) { afs_osi_CancelWait(&AFS_WaitHandler); /* trigger FlushVCBs asap */ if (afs_stats_cmperf.CallBackAlloced >= sizeof(afs_cbrHeads)/sizeof(afs_cbrHeads[0])) { /* don't allocate more than 16 * AFS_NCBRS for now */ tsp = (struct afs_cbr *)osi_AllocSmallSpace(sizeof(*tsp)); tsp->dynalloc = 1; tsp->next = NULL; afs_stats_cmperf.CallBackFlushes++; } else { /* try allocating */ tsp = (struct afs_cbr *)afs_osi_Alloc(AFS_NCBRS * sizeof(struct afs_cbr)); for (i = 0; i < AFS_NCBRS - 1; i++) { tsp[i].next = &tsp[i + 1]; tsp[i].dynalloc = 0; } tsp[AFS_NCBRS - 1].next = 0; tsp[AFS_NCBRS - 1].dynalloc = 0; afs_cbrSpace = tsp->next; afs_cbrHeads[afs_stats_cmperf.CallBackAlloced] = tsp; afs_stats_cmperf.CallBackAlloced++; } } else { tsp = afs_cbrSpace; afs_cbrSpace = tsp->next; } return tsp; } /*! * Free a callback return structure, removing it from all lists. * * Environment: the xvcb lock is held over these calls. * * \param asp The address of the structure to free. * * \rerurn 0 */ int afs_FreeCBR(register struct afs_cbr *asp) { *(asp->pprev) = asp->next; if (asp->next) asp->next->pprev = asp->pprev; *(asp->hash_pprev) = asp->hash_next; if (asp->hash_next) asp->hash_next->hash_pprev = asp->hash_pprev; if (asp->dynalloc) { osi_FreeSmallSpace(asp); } else { asp->next = afs_cbrSpace; afs_cbrSpace = asp; } return 0; } /*! * Flush all queued callbacks to all servers. * * Environment: holds xvcb lock over RPC to guard against race conditions * when a new callback is granted for the same file later on. * * \return 0 for success. */ afs_int32 afs_FlushVCBs(afs_int32 lockit) { struct AFSFid *tfids; struct AFSCallBack callBacks[1]; struct AFSCBFids fidArray; struct AFSCBs cbArray; afs_int32 code; struct afs_cbr *tcbrp; int tcount; struct server *tsp; int i; struct vrequest treq; struct afs_conn *tc; int safety1, safety2, safety3; XSTATS_DECLS; if ((code = afs_InitReq(&treq, afs_osi_credp))) return code; treq.flags |= O_NONBLOCK; tfids = afs_osi_Alloc(sizeof(struct AFSFid) * AFS_MAXCBRSCALL); if (lockit) ObtainWriteLock(&afs_xvcb, 273); ObtainReadLock(&afs_xserver); for (i = 0; i < NSERVERS; i++) { for (safety1 = 0, tsp = afs_servers[i]; tsp && safety1 < afs_totalServers + 10; tsp = tsp->next, safety1++) { /* don't have any */ if (tsp->cbrs == (struct afs_cbr *)0) continue; /* otherwise, grab a block of AFS_MAXCBRSCALL from the list * and make an RPC, over and over again. */ tcount = 0; /* number found so far */ for (safety2 = 0; safety2 < afs_cacheStats; safety2++) { if (tcount >= AFS_MAXCBRSCALL || !tsp->cbrs) { /* if buffer is full, or we've queued all we're going * to from this server, we should flush out the * callbacks. */ fidArray.AFSCBFids_len = tcount; fidArray.AFSCBFids_val = (struct AFSFid *)tfids; cbArray.AFSCBs_len = 1; cbArray.AFSCBs_val = callBacks; memset(&callBacks[0], 0, sizeof(callBacks[0])); callBacks[0].CallBackType = CB_EXCLUSIVE; for (safety3 = 0; safety3 < AFS_MAXHOSTS * 2; safety3++) { tc = afs_ConnByHost(tsp, tsp->cell->fsport, tsp->cell->cellNum, &treq, 0, SHARED_LOCK); if (tc) { XSTATS_START_TIME (AFS_STATS_FS_RPCIDX_GIVEUPCALLBACKS); RX_AFS_GUNLOCK(); code = RXAFS_GiveUpCallBacks(tc->id, &fidArray, &cbArray); RX_AFS_GLOCK(); XSTATS_END_TIME; } else code = -1; if (!afs_Analyze (tc, code, 0, &treq, AFS_STATS_FS_RPCIDX_GIVEUPCALLBACKS, SHARED_LOCK, tsp->cell)) { break; } } /* ignore return code, since callbacks may have * been returned anyway, we shouldn't leave them * around to be returned again. * * Next, see if we are done with this server, and if so, * break to deal with the next one. */ if (!tsp->cbrs) break; tcount = 0; } /* if to flush full buffer */ /* if we make it here, we have an entry at the head of cbrs, * which we should copy to the file ID array and then free. */ tcbrp = tsp->cbrs; tfids[tcount++] = tcbrp->fid; /* Freeing the CBR will unlink it from the server's CBR list */ afs_FreeCBR(tcbrp); } /* while loop for this one server */ if (safety2 > afs_cacheStats) { afs_warn("possible internal error afs_flushVCBs (%d)\n", safety2); } } /* for loop for this hash chain */ } /* loop through all hash chains */ if (safety1 > afs_totalServers + 2) { afs_warn ("AFS internal error (afs_flushVCBs) (%d > %d), continuing...\n", safety1, afs_totalServers + 2); if (afs_paniconwarn) osi_Panic("afs_flushVCBS safety1"); } ReleaseReadLock(&afs_xserver); if (lockit) ReleaseWriteLock(&afs_xvcb); afs_osi_Free(tfids, sizeof(struct AFSFid) * AFS_MAXCBRSCALL); return 0; } /*! * Queue a callback on the given fid. * * Environment: * Locks the xvcb lock. * Called when the xvcache lock is already held. * * \param avc vcache entry * \return 1 if queued, 0 otherwise */ static afs_int32 afs_QueueVCB(struct vcache *avc) { int queued = 0; struct server *tsp; struct afs_cbr *tcbp; AFS_STATCNT(afs_QueueVCB); ObtainWriteLock(&afs_xvcb, 274); /* we can't really give back callbacks on RO files, since the * server only tracks them on a per-volume basis, and we don't * know whether we still have some other files from the same * volume. */ if (!((avc->f.states & CRO) == 0 && avc->callback)) { goto done; } /* The callback is really just a struct server ptr. */ tsp = (struct server *)(avc->callback); /* we now have a pointer to the server, so we just allocate * a queue entry and queue it. */ tcbp = afs_AllocCBR(); tcbp->fid = avc->f.fid.Fid; tcbp->next = tsp->cbrs; if (tsp->cbrs) tsp->cbrs->pprev = &tcbp->next; tsp->cbrs = tcbp; tcbp->pprev = &tsp->cbrs; afs_InsertHashCBR(tcbp); queued = 1; done: /* now release locks and return */ ReleaseWriteLock(&afs_xvcb); return queued; } /*! * Remove a queued callback for a given Fid. * * Environment: * Locks xvcb and xserver locks. * Typically called with xdcache, xvcache and/or individual vcache * entries locked. * * \param afid The fid we want cleansed of queued callbacks. * */ void afs_RemoveVCB(struct VenusFid *afid) { int slot; struct afs_cbr *cbr, *ncbr; AFS_STATCNT(afs_RemoveVCB); ObtainWriteLock(&afs_xvcb, 275); slot = afs_HashCBRFid(&afid->Fid); ncbr = afs_cbrHashT[slot]; while (ncbr) { cbr = ncbr; ncbr = cbr->hash_next; if (afid->Fid.Volume == cbr->fid.Volume && afid->Fid.Vnode == cbr->fid.Vnode && afid->Fid.Unique == cbr->fid.Unique) { afs_FreeCBR(cbr); } } ReleaseWriteLock(&afs_xvcb); } void afs_FlushReclaimedVcaches(void) { #if !defined(AFS_LINUX22_ENV) struct vcache *tvc; int code, fv_slept; struct vcache *tmpReclaimedVCList = NULL; ObtainWriteLock(&afs_xvreclaim, 76); while (ReclaimedVCList) { tvc = ReclaimedVCList; /* take from free list */ ReclaimedVCList = tvc->nextfree; tvc->nextfree = NULL; code = afs_FlushVCache(tvc, &fv_slept); if (code) { /* Ok, so, if we got code != 0, uh, wtf do we do? */ /* Probably, build a temporary list and then put all back when we get to the end of the list */ /* This is actually really crappy, but we need to not leak these. We probably need a way to be smarter about this. */ tvc->nextfree = tmpReclaimedVCList; tmpReclaimedVCList = tvc; /* printf("Reclaim list flush %lx failed: %d\n", (unsigned long) tvc, code); */ } if (tvc->f.states & (CVInit #ifdef AFS_DARWIN80_ENV | CDeadVnode #endif )) { tvc->f.states &= ~(CVInit #ifdef AFS_DARWIN80_ENV | CDeadVnode #endif ); afs_osi_Wakeup(&tvc->f.states); } } if (tmpReclaimedVCList) ReclaimedVCList = tmpReclaimedVCList; ReleaseWriteLock(&afs_xvreclaim); #endif } void afs_PostPopulateVCache(struct vcache *avc, struct VenusFid *afid, int seq) { /* * The proper value for mvstat (for root fids) is setup by the caller. */ avc->mvstat = 0; if (afid->Fid.Vnode == 1 && afid->Fid.Unique == 1) avc->mvstat = 2; if (afs_globalVFS == 0) osi_Panic("afs globalvfs"); osi_PostPopulateVCache(avc); avc->dchint = NULL; osi_dnlc_purgedp(avc); /* this may be overkill */ memset(&(avc->callsort), 0, sizeof(struct afs_q)); avc->slocks = NULL; avc->f.states &=~ CVInit; if (seq) { avc->f.states |= CBulkFetching; avc->f.m.Length = seq; } afs_osi_Wakeup(&avc->f.states); } int afs_ShakeLooseVCaches(afs_int32 anumber) { afs_int32 i, loop; struct vcache *tvc; struct afs_q *tq, *uq; int fv_slept; afs_int32 target = anumber; i = 0; loop = 0; for (tq = VLRU.prev; tq != &VLRU && anumber > 0; tq = uq) { tvc = QTOV(tq); uq = QPrev(tq); if (tvc->f.states & CVFlushed) { refpanic("CVFlushed on VLRU"); /* In the other path, this was 2 * afs_cacheStats */ } else if (!afsd_dynamic_vcaches && i++ > afs_maxvcount) { refpanic("Exceeded pool of AFS vnodes(VLRU cycle?)"); } else if (QNext(uq) != tq) { refpanic("VLRU inconsistent"); } else if (tvc->f.states & CVInit) { continue; } fv_slept = 0; if (osi_TryEvictVCache(tvc, &fv_slept)) anumber--; if (fv_slept) { if (loop++ > 100) break; uq = VLRU.prev; i = 0; continue; /* start over - may have raced. */ } if (tq == uq) break; } if (!afsd_dynamic_vcaches && anumber == target) { afs_warn("afs_ShakeLooseVCaches: warning none freed, using %d of %d\n", afs_vcount, afs_maxvcount); } return 0; } /* Alloc new vnode. */ static struct vcache * afs_AllocVCache(void) { struct vcache *tvc; tvc = osi_NewVnode(); afs_vcount++; /* track the peak */ if (afsd_dynamic_vcaches && afs_maxvcount < afs_vcount) { afs_maxvcount = afs_vcount; /*printf("peak vnodes: %d\n", afs_maxvcount);*/ } afs_stats_cmperf.vcacheXAllocs++; /* count in case we have a leak */ /* If we create a new inode, we either give it a new slot number, * or if one's available, use a slot number from the slot free list */ if (afs_freeSlotList != NULL) { struct afs_slotlist *tmp; tvc->diskSlot = afs_freeSlotList->slot; tmp = afs_freeSlotList; afs_freeSlotList = tmp->next; afs_osi_Free(tmp, sizeof(struct afs_slotlist)); } else { tvc->diskSlot = afs_nextVcacheSlot++; } return tvc; } /* Pre populate a newly allocated vcache. On platforms where the actual * vnode is attached to the vcache, this function is called before attachment, * therefore it cannot perform any actions on the vnode itself */ static void afs_PrePopulateVCache(struct vcache *avc, struct VenusFid *afid, struct server *serverp) { afs_uint32 slot; slot = avc->diskSlot; osi_PrePopulateVCache(avc); avc->diskSlot = slot; QZero(&avc->metadirty); AFS_RWLOCK_INIT(&avc->lock, "vcache lock"); avc->mvid = NULL; avc->linkData = NULL; avc->cbExpires = 0; avc->opens = 0; avc->execsOrWriters = 0; avc->flockCount = 0; avc->f.states = CVInit; avc->last_looker = 0; avc->f.fid = *afid; avc->asynchrony = -1; avc->vc_error = 0; hzero(avc->mapDV); avc->f.truncPos = AFS_NOTRUNC; /* don't truncate until we need to */ hzero(avc->f.m.DataVersion); /* in case we copy it into flushDV */ avc->Access = NULL; avc->callback = serverp; /* to minimize chance that clear * request is lost */ #if defined(AFS_CACHE_BYPASS) avc->cachingStates = 0; avc->cachingTransitions = 0; #endif } /*! * This routine is responsible for allocating a new cache entry * from the free list. It formats the cache entry and inserts it * into the appropriate hash tables. It must be called with * afs_xvcache write-locked so as to prevent several processes from * trying to create a new cache entry simultaneously. * * LOCK: afs_NewVCache afs_xvcache W * * \param afid The file id of the file whose cache entry is being created. * * \return The new vcache struct. */ static_inline struct vcache * afs_NewVCache_int(struct VenusFid *afid, struct server *serverp, int seq) { struct vcache *tvc; afs_int32 i, j; afs_int32 anumber = VCACHE_FREE; AFS_STATCNT(afs_NewVCache); afs_FlushReclaimedVcaches(); #if defined(AFS_LINUX22_ENV) if(!afsd_dynamic_vcaches) { afs_ShakeLooseVCaches(anumber); if (afs_vcount >= afs_maxvcount) { afs_warn("afs_NewVCache - none freed\n"); return NULL; } } tvc = afs_AllocVCache(); #else /* AFS_LINUX22_ENV */ /* pull out a free cache entry */ if (!freeVCList) { afs_ShakeLooseVCaches(anumber); } if (!freeVCList) { tvc = afs_AllocVCache(); } else { tvc = freeVCList; /* take from free list */ freeVCList = tvc->nextfree; tvc->nextfree = NULL; } /* end of if (!freeVCList) */ #endif /* AFS_LINUX22_ENV */ #if defined(AFS_XBSD_ENV) || defined(AFS_DARWIN_ENV) if (tvc->v) panic("afs_NewVCache(): free vcache with vnode attached"); #endif /* Populate the vcache with as much as we can. */ afs_PrePopulateVCache(tvc, afid, serverp); /* Thread the vcache onto the VLRU */ i = VCHash(afid); j = VCHashV(afid); tvc->hnext = afs_vhashT[i]; afs_vhashT[i] = tvc; QAdd(&afs_vhashTV[j], &tvc->vhashq); if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) { refpanic("NewVCache VLRU inconsistent"); } QAdd(&VLRU, &tvc->vlruq); /* put in lruq */ if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) { refpanic("NewVCache VLRU inconsistent2"); } if (tvc->vlruq.next->prev != &(tvc->vlruq)) { refpanic("NewVCache VLRU inconsistent3"); } if (tvc->vlruq.prev->next != &(tvc->vlruq)) { refpanic("NewVCache VLRU inconsistent4"); } vcachegen++; /* it should now be safe to drop the xvcache lock - so attach an inode * to this vcache, where necessary */ osi_AttachVnode(tvc, seq); /* Get a reference count to hold this vcache for the VLRUQ. Note that * we have to do this after attaching the vnode, because the reference * count may be held in the vnode itself */ #if defined(AFS_LINUX22_ENV) /* Hold it for the LRU (should make count 2) */ AFS_FAST_HOLD(tvc); #elif !(defined (AFS_DARWIN_ENV) || defined(AFS_XBSD_ENV)) VREFCOUNT_SET(tvc, 1); /* us */ #endif #if defined (AFS_FBSD_ENV) if (tvc->f.states & CVInit) #endif afs_PostPopulateVCache(tvc, afid, seq); return tvc; } /*afs_NewVCache */ struct vcache * afs_NewVCache(struct VenusFid *afid, struct server *serverp) { return afs_NewVCache_int(afid, serverp, 0); } struct vcache * afs_NewBulkVCache(struct VenusFid *afid, struct server *serverp, int seq) { return afs_NewVCache_int(afid, serverp, seq); } /*! * ??? * * LOCK: afs_FlushActiveVcaches afs_xvcache N * * \param doflocks : Do we handle flocks? */ void afs_FlushActiveVcaches(register afs_int32 doflocks) { register struct vcache *tvc; register int i; register struct afs_conn *tc; register afs_int32 code; afs_ucred_t *cred = NULL; struct vrequest treq, ureq; struct AFSVolSync tsync; int didCore; XSTATS_DECLS; AFS_STATCNT(afs_FlushActiveVcaches); ObtainReadLock(&afs_xvcache); for (i = 0; i < VCSIZE; i++) { for (tvc = afs_vhashT[i]; tvc; tvc = tvc->hnext) { if (tvc->f.states & CVInit) continue; #ifdef AFS_DARWIN80_ENV if (tvc->f.states & CDeadVnode && (tvc->f.states & (CCore|CUnlinkedDel) || tvc->flockCount)) panic("Dead vnode has core/unlinkedel/flock"); #endif if (doflocks && tvc->flockCount != 0) { /* if this entry has an flock, send a keep-alive call out */ osi_vnhold(tvc, 0); ReleaseReadLock(&afs_xvcache); ObtainWriteLock(&tvc->lock, 51); do { afs_InitReq(&treq, afs_osi_credp); treq.flags |= O_NONBLOCK; tc = afs_Conn(&tvc->f.fid, &treq, SHARED_LOCK); if (tc) { XSTATS_START_TIME(AFS_STATS_FS_RPCIDX_EXTENDLOCK); RX_AFS_GUNLOCK(); code = RXAFS_ExtendLock(tc->id, (struct AFSFid *)&tvc->f.fid.Fid, &tsync); RX_AFS_GLOCK(); XSTATS_END_TIME; } else code = -1; } while (afs_Analyze (tc, code, &tvc->f.fid, &treq, AFS_STATS_FS_RPCIDX_EXTENDLOCK, SHARED_LOCK, NULL)); ReleaseWriteLock(&tvc->lock); #ifdef AFS_DARWIN80_ENV AFS_FAST_RELE(tvc); ObtainReadLock(&afs_xvcache); #else ObtainReadLock(&afs_xvcache); AFS_FAST_RELE(tvc); #endif } didCore = 0; if ((tvc->f.states & CCore) || (tvc->f.states & CUnlinkedDel)) { /* * Don't let it evaporate in case someone else is in * this code. Also, drop the afs_xvcache lock while * getting vcache locks. */ osi_vnhold(tvc, 0); ReleaseReadLock(&afs_xvcache); #ifdef AFS_BOZONLOCK_ENV afs_BozonLock(&tvc->pvnLock, tvc); #endif #if defined(AFS_SGI_ENV) /* * That's because if we come in via the CUnlinkedDel bit state path we'll be have 0 refcnt */ osi_Assert(VREFCOUNT_GT(tvc,0)); AFS_RWLOCK((vnode_t *) tvc, VRWLOCK_WRITE); #endif ObtainWriteLock(&tvc->lock, 52); if (tvc->f.states & CCore) { tvc->f.states &= ~CCore; /* XXXX Find better place-holder for cred XXXX */ cred = (afs_ucred_t *)tvc->linkData; tvc->linkData = NULL; /* XXX */ afs_InitReq(&ureq, cred); afs_Trace2(afs_iclSetp, CM_TRACE_ACTCCORE, ICL_TYPE_POINTER, tvc, ICL_TYPE_INT32, tvc->execsOrWriters); code = afs_StoreOnLastReference(tvc, &ureq); ReleaseWriteLock(&tvc->lock); #ifdef AFS_BOZONLOCK_ENV afs_BozonUnlock(&tvc->pvnLock, tvc); #endif hzero(tvc->flushDV); osi_FlushText(tvc); didCore = 1; if (code && code != VNOVNODE) { afs_StoreWarn(code, tvc->f.fid.Fid.Volume, /* /dev/console */ 1); } } else if (tvc->f.states & CUnlinkedDel) { /* * Ignore errors */ ReleaseWriteLock(&tvc->lock); #ifdef AFS_BOZONLOCK_ENV afs_BozonUnlock(&tvc->pvnLock, tvc); #endif #if defined(AFS_SGI_ENV) AFS_RWUNLOCK((vnode_t *) tvc, VRWLOCK_WRITE); #endif afs_remunlink(tvc, 0); #if defined(AFS_SGI_ENV) AFS_RWLOCK((vnode_t *) tvc, VRWLOCK_WRITE); #endif } else { /* lost (or won, perhaps) the race condition */ ReleaseWriteLock(&tvc->lock); #ifdef AFS_BOZONLOCK_ENV afs_BozonUnlock(&tvc->pvnLock, tvc); #endif } #if defined(AFS_SGI_ENV) AFS_RWUNLOCK((vnode_t *) tvc, VRWLOCK_WRITE); #endif #ifdef AFS_DARWIN80_ENV AFS_FAST_RELE(tvc); if (didCore) { AFS_RELE(AFSTOV(tvc)); /* Matches write code setting CCore flag */ crfree(cred); } ObtainReadLock(&afs_xvcache); #else ObtainReadLock(&afs_xvcache); AFS_FAST_RELE(tvc); if (didCore) { AFS_RELE(AFSTOV(tvc)); /* Matches write code setting CCore flag */ crfree(cred); } #endif } } } ReleaseReadLock(&afs_xvcache); } /*! * Make sure a cache entry is up-to-date status-wise. * * NOTE: everywhere that calls this can potentially be sped up * by checking CStatd first, and avoiding doing the InitReq * if this is up-to-date. * * Anymore, the only places that call this KNOW already that the * vcache is not up-to-date, so we don't screw around. * * \param avc : Ptr to vcache entry to verify. * \param areq : ??? */ /*! * * Make sure a cache entry is up-to-date status-wise. * * NOTE: everywhere that calls this can potentially be sped up * by checking CStatd first, and avoiding doing the InitReq * if this is up-to-date. * * Anymore, the only places that call this KNOW already that the * vcache is not up-to-date, so we don't screw around. * * \param avc Pointer to vcache entry to verify. * \param areq * * \return 0 for success or other error codes. */ int afs_VerifyVCache2(struct vcache *avc, struct vrequest *areq) { register struct vcache *tvc; AFS_STATCNT(afs_VerifyVCache); /* otherwise we must fetch the status info */ ObtainWriteLock(&avc->lock, 53); if (avc->f.states & CStatd) { ReleaseWriteLock(&avc->lock); return 0; } ObtainWriteLock(&afs_xcbhash, 461); avc->f.states &= ~(CStatd | CUnique); avc->callback = NULL; afs_DequeueCallback(avc); ReleaseWriteLock(&afs_xcbhash); ReleaseWriteLock(&avc->lock); /* since we've been called back, or the callback has expired, * it's possible that the contents of this directory, or this * file's name have changed, thus invalidating the dnlc contents. */ if ((avc->f.states & CForeign) || (avc->f.fid.Fid.Vnode & 1)) osi_dnlc_purgedp(avc); else osi_dnlc_purgevp(avc); /* fetch the status info */ tvc = afs_GetVCache(&avc->f.fid, areq, NULL, avc); if (!tvc) return ENOENT; /* Put it back; caller has already incremented vrefCount */ afs_PutVCache(tvc); return 0; } /*afs_VerifyVCache */ /*! * Simple copy of stat info into cache. * * Callers:as of 1992-04-29, only called by WriteVCache * * \param avc Ptr to vcache entry involved. * \param astat Ptr to stat info to copy. * */ static void afs_SimpleVStat(register struct vcache *avc, register struct AFSFetchStatus *astat, struct vrequest *areq) { afs_size_t length; AFS_STATCNT(afs_SimpleVStat); #ifdef AFS_64BIT_CLIENT FillInt64(length, astat->Length_hi, astat->Length); #else /* AFS_64BIT_CLIENT */ length = astat->Length; #endif /* AFS_64BIT_CLIENT */ #if defined(AFS_SGI_ENV) if ((avc->execsOrWriters <= 0) && !afs_DirtyPages(avc) && !AFS_VN_MAPPED((vnode_t *) avc)) { osi_Assert((valusema(&avc->vc_rwlock) <= 0) && (OSI_GET_LOCKID() == avc->vc_rwlockid)); if (length < avc->f.m.Length) { vnode_t *vp = (vnode_t *) avc; osi_Assert(WriteLocked(&avc->lock)); ReleaseWriteLock(&avc->lock); AFS_GUNLOCK(); PTOSSVP(vp, (off_t) length, (off_t) MAXLONG); AFS_GLOCK(); ObtainWriteLock(&avc->lock, 67); } } #endif if (!afs_DirtyPages(avc)) { /* if actively writing the file, don't fetch over this value */ afs_Trace3(afs_iclSetp, CM_TRACE_SIMPLEVSTAT, ICL_TYPE_POINTER, avc, ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(avc->f.m.Length), ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(length)); avc->f.m.Length = length; avc->f.m.Date = astat->ClientModTime; } avc->f.m.Owner = astat->Owner; avc->f.m.Group = astat->Group; avc->f.m.Mode = astat->UnixModeBits; if (vType(avc) == VREG) { avc->f.m.Mode |= S_IFREG; } else if (vType(avc) == VDIR) { avc->f.m.Mode |= S_IFDIR; } else if (vType(avc) == VLNK) { avc->f.m.Mode |= S_IFLNK; if ((avc->f.m.Mode & 0111) == 0) avc->mvstat = 1; } if (avc->f.states & CForeign) { struct axscache *ac; avc->f.anyAccess = astat->AnonymousAccess; #ifdef badidea if ((astat->CallerAccess & ~astat->AnonymousAccess)) /* USED TO SAY : * Caller has at least one bit not covered by anonymous, and * thus may have interesting rights. * * HOWEVER, this is a really bad idea, because any access query * for bits which aren't covered by anonymous, on behalf of a user * who doesn't have any special rights, will result in an answer of * the form "I don't know, lets make a FetchStatus RPC and find out!" * It's an especially bad idea under Ultrix, since (due to the lack of * a proper access() call) it must perform several afs_access() calls * in order to create magic mode bits that vary according to who makes * the call. In other words, _every_ stat() generates a test for * writeability... */ #endif /* badidea */ if (avc->Access && (ac = afs_FindAxs(avc->Access, areq->uid))) ac->axess = astat->CallerAccess; else /* not found, add a new one if possible */ afs_AddAxs(avc->Access, areq->uid, astat->CallerAccess); } } /*afs_SimpleVStat */ /*! * Store the status info *only* back to the server for a * fid/vrequest. * * Environment: Must be called with a shared lock held on the vnode. * * \param avc Ptr to the vcache entry. * \param astatus Ptr to the status info to store. * \param areq Ptr to the associated vrequest. * * \return Operation status. */ int afs_WriteVCache(register struct vcache *avc, register struct AFSStoreStatus *astatus, struct vrequest *areq) { afs_int32 code; struct afs_conn *tc; struct AFSFetchStatus OutStatus; struct AFSVolSync tsync; XSTATS_DECLS; AFS_STATCNT(afs_WriteVCache); afs_Trace2(afs_iclSetp, CM_TRACE_WVCACHE, ICL_TYPE_POINTER, avc, ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(avc->f.m.Length)); do { tc = afs_Conn(&avc->f.fid, areq, SHARED_LOCK); if (tc) { XSTATS_START_TIME(AFS_STATS_FS_RPCIDX_STORESTATUS); RX_AFS_GUNLOCK(); code = RXAFS_StoreStatus(tc->id, (struct AFSFid *)&avc->f.fid.Fid, astatus, &OutStatus, &tsync); RX_AFS_GLOCK(); XSTATS_END_TIME; } else code = -1; } while (afs_Analyze (tc, code, &avc->f.fid, areq, AFS_STATS_FS_RPCIDX_STORESTATUS, SHARED_LOCK, NULL)); UpgradeSToWLock(&avc->lock, 20); if (code == 0) { /* success, do the changes locally */ afs_SimpleVStat(avc, &OutStatus, areq); /* * Update the date, too. SimpleVStat didn't do this, since * it thought we were doing this after fetching new status * over a file being written. */ avc->f.m.Date = OutStatus.ClientModTime; } else { /* failure, set up to check with server next time */ ObtainWriteLock(&afs_xcbhash, 462); afs_DequeueCallback(avc); avc->f.states &= ~(CStatd | CUnique); /* turn off stat valid flag */ ReleaseWriteLock(&afs_xcbhash); if ((avc->f.states & CForeign) || (avc->f.fid.Fid.Vnode & 1)) osi_dnlc_purgedp(avc); /* if it (could be) a directory */ } ConvertWToSLock(&avc->lock); return code; } /*afs_WriteVCache */ /*! * Store status info only locally, set the proper disconnection flags * and add to dirty list. * * \param avc The vcache to be written locally. * \param astatus Get attr fields from local store. * \param attrs This one is only of the vs_size. * * \note Must be called with a shared lock on the vnode */ int afs_WriteVCacheDiscon(register struct vcache *avc, register struct AFSStoreStatus *astatus, struct vattr *attrs) { afs_int32 code = 0; afs_int32 flags = 0; UpgradeSToWLock(&avc->lock, 700); if (!astatus->Mask) { return code; } else { /* Set attributes. */ if (astatus->Mask & AFS_SETMODTIME) { avc->f.m.Date = astatus->ClientModTime; flags |= VDisconSetTime; } if (astatus->Mask & AFS_SETOWNER) { /* printf("Not allowed yet. \n"); */ /*avc->f.m.Owner = astatus->Owner;*/ } if (astatus->Mask & AFS_SETGROUP) { /* printf("Not allowed yet. \n"); */ /*avc->f.m.Group = astatus->Group;*/ } if (astatus->Mask & AFS_SETMODE) { avc->f.m.Mode = astatus->UnixModeBits; #if 0 /* XXX: Leaving this out, so it doesn't mess up the file type flag.*/ if (vType(avc) == VREG) { avc->f.m.Mode |= S_IFREG; } else if (vType(avc) == VDIR) { avc->f.m.Mode |= S_IFDIR; } else if (vType(avc) == VLNK) { avc->f.m.Mode |= S_IFLNK; if ((avc->f.m.Mode & 0111) == 0) avc->mvstat = 1; } #endif flags |= VDisconSetMode; } /* if(astatus.Mask & AFS_SETMODE) */ } /* if (!astatus->Mask) */ if (attrs->va_size > 0) { /* XXX: Do I need more checks? */ /* Truncation operation. */ flags |= VDisconTrunc; } if (flags) afs_DisconAddDirty(avc, flags, 1); /* XXX: How about the rest of the fields? */ ConvertWToSLock(&avc->lock); return code; } /*! * Copy astat block into vcache info * * \note This code may get dataversion and length out of sync if the file has * been modified. This is less than ideal. I haven't thought about it sufficiently * to be certain that it is adequate. * * \note Environment: Must be called under a write lock * * \param avc Ptr to vcache entry. * \param astat Ptr to stat block to copy in. * \param areq Ptr to associated request. */ void afs_ProcessFS(register struct vcache *avc, register struct AFSFetchStatus *astat, struct vrequest *areq) { afs_size_t length; AFS_STATCNT(afs_ProcessFS); #ifdef AFS_64BIT_CLIENT FillInt64(length, astat->Length_hi, astat->Length); #else /* AFS_64BIT_CLIENT */ length = astat->Length; #endif /* AFS_64BIT_CLIENT */ /* WARNING: afs_DoBulkStat uses the Length field to store a sequence * number for each bulk status request. Under no circumstances * should afs_DoBulkStat store a sequence number if the new * length will be ignored when afs_ProcessFS is called with * new stats. If you change the following conditional then you * also need to change the conditional in afs_DoBulkStat. */ #ifdef AFS_SGI_ENV if ((avc->execsOrWriters <= 0) && !afs_DirtyPages(avc) && !AFS_VN_MAPPED((vnode_t *) avc)) { #else if ((avc->execsOrWriters <= 0) && !afs_DirtyPages(avc)) { #endif /* if we're writing or mapping this file, don't fetch over these * values. */ afs_Trace3(afs_iclSetp, CM_TRACE_PROCESSFS, ICL_TYPE_POINTER, avc, ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(avc->f.m.Length), ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(length)); avc->f.m.Length = length; avc->f.m.Date = astat->ClientModTime; } hset64(avc->f.m.DataVersion, astat->dataVersionHigh, astat->DataVersion); avc->f.m.Owner = astat->Owner; avc->f.m.Mode = astat->UnixModeBits; avc->f.m.Group = astat->Group; avc->f.m.LinkCount = astat->LinkCount; if (astat->FileType == File) { vSetType(avc, VREG); avc->f.m.Mode |= S_IFREG; } else if (astat->FileType == Directory) { vSetType(avc, VDIR); avc->f.m.Mode |= S_IFDIR; } else if (astat->FileType == SymbolicLink) { if (afs_fakestat_enable && (avc->f.m.Mode & 0111) == 0) { vSetType(avc, VDIR); avc->f.m.Mode |= S_IFDIR; } else { vSetType(avc, VLNK); avc->f.m.Mode |= S_IFLNK; } if ((avc->f.m.Mode & 0111) == 0) { avc->mvstat = 1; } } avc->f.anyAccess = astat->AnonymousAccess; #ifdef badidea if ((astat->CallerAccess & ~astat->AnonymousAccess)) /* USED TO SAY : * Caller has at least one bit not covered by anonymous, and * thus may have interesting rights. * * HOWEVER, this is a really bad idea, because any access query * for bits which aren't covered by anonymous, on behalf of a user * who doesn't have any special rights, will result in an answer of * the form "I don't know, lets make a FetchStatus RPC and find out!" * It's an especially bad idea under Ultrix, since (due to the lack of * a proper access() call) it must perform several afs_access() calls * in order to create magic mode bits that vary according to who makes * the call. In other words, _every_ stat() generates a test for * writeability... */ #endif /* badidea */ { struct axscache *ac; if (avc->Access && (ac = afs_FindAxs(avc->Access, areq->uid))) ac->axess = astat->CallerAccess; else /* not found, add a new one if possible */ afs_AddAxs(avc->Access, areq->uid, astat->CallerAccess); } } /*afs_ProcessFS */ /*! * Get fid from server. * * \param afid * \param areq Request to be passed on. * \param name Name of ?? to lookup. * \param OutStatus Fetch status. * \param CallBackp * \param serverp * \param tsyncp * * \return Success status of operation. */ int afs_RemoteLookup(register struct VenusFid *afid, struct vrequest *areq, char *name, struct VenusFid *nfid, struct AFSFetchStatus *OutStatusp, struct AFSCallBack *CallBackp, struct server **serverp, struct AFSVolSync *tsyncp) { afs_int32 code; afs_uint32 start; register struct afs_conn *tc; struct AFSFetchStatus OutDirStatus; XSTATS_DECLS; if (!name) name = ""; /* XXX */ do { tc = afs_Conn(afid, areq, SHARED_LOCK); if (tc) { if (serverp) *serverp = tc->srvr->server; start = osi_Time(); XSTATS_START_TIME(AFS_STATS_FS_RPCIDX_XLOOKUP); RX_AFS_GUNLOCK(); code = RXAFS_Lookup(tc->id, (struct AFSFid *)&afid->Fid, name, (struct AFSFid *)&nfid->Fid, OutStatusp, &OutDirStatus, CallBackp, tsyncp); RX_AFS_GLOCK(); XSTATS_END_TIME; } else code = -1; } while (afs_Analyze (tc, code, afid, areq, AFS_STATS_FS_RPCIDX_XLOOKUP, SHARED_LOCK, NULL)); return code; } /*! * afs_GetVCache * * Given a file id and a vrequest structure, fetch the status * information associated with the file. * * \param afid File ID. * \param areq Ptr to associated vrequest structure, specifying the * user whose authentication tokens will be used. * \param avc Caller may already have a vcache for this file, which is * already held. * * \note Environment: * The cache entry is returned with an increased vrefCount field. * The entry must be discarded by calling afs_PutVCache when you * are through using the pointer to the cache entry. * * You should not hold any locks when calling this function, except * locks on other vcache entries. If you lock more than one vcache * entry simultaneously, you should lock them in this order: * * 1. Lock all files first, then directories. * 2. Within a particular type, lock entries in Fid.Vnode order. * * This locking hierarchy is convenient because it allows locking * of a parent dir cache entry, given a file (to check its access * control list). It also allows renames to be handled easily by * locking directories in a constant order. * * \note NB. NewVCache -> FlushVCache presently (4/10/95) drops the xvcache lock. * * \note Might have a vcache structure already, which must * already be held by the caller */ struct vcache * afs_GetVCache(register struct VenusFid *afid, struct vrequest *areq, afs_int32 * cached, struct vcache *avc) { afs_int32 code, newvcache = 0; register struct vcache *tvc; struct volume *tvp; afs_int32 retry; AFS_STATCNT(afs_GetVCache); if (cached) *cached = 0; /* Init just in case */ #if defined(AFS_SGI_ENV) && !defined(AFS_SGI53_ENV) loop: #endif ObtainSharedLock(&afs_xvcache, 5); tvc = afs_FindVCache(afid, &retry, DO_STATS | DO_VLRU | IS_SLOCK); if (tvc && retry) { #if defined(AFS_SGI_ENV) && !defined(AFS_SGI53_ENV) ReleaseSharedLock(&afs_xvcache); spunlock_psema(tvc->v.v_lock, retry, &tvc->v.v_sync, PINOD); goto loop; #endif } if (tvc) { if (cached) *cached = 1; osi_Assert((tvc->f.states & CVInit) == 0); /* If we are in readdir, return the vnode even if not statd */ if ((tvc->f.states & CStatd) || afs_InReadDir(tvc)) { ReleaseSharedLock(&afs_xvcache); return tvc; } } else { UpgradeSToWLock(&afs_xvcache, 21); /* no cache entry, better grab one */ tvc = afs_NewVCache(afid, NULL); newvcache = 1; ConvertWToSLock(&afs_xvcache); if (tvc == NULL) { ReleaseSharedLock(&afs_xvcache); return NULL; } afs_stats_cmperf.vcacheMisses++; } ReleaseSharedLock(&afs_xvcache); ObtainWriteLock(&tvc->lock, 54); if (tvc->f.states & CStatd) { ReleaseWriteLock(&tvc->lock); return tvc; } #ifdef AFS_DARWIN80_ENV /* Darwin 8.0 only has bufs in nfs, so we shouldn't have to worry about them. What about ubc? */ #else #if defined(AFS_DARWIN_ENV) || defined(AFS_FBSD_ENV) /* * XXX - I really don't like this. Should try to understand better. * It seems that sometimes, when we get called, we already hold the * lock on the vnode (e.g., from afs_getattr via afs_VerifyVCache). * We can't drop the vnode lock, because that could result in a race. * Sometimes, though, we get here and don't hold the vnode lock. * I hate code paths that sometimes hold locks and sometimes don't. * In any event, the dodge we use here is to check whether the vnode * is locked, and if it isn't, then we gain and drop it around the call * to vinvalbuf; otherwise, we leave it alone. */ { struct vnode *vp = AFSTOV(tvc); int iheldthelock; #if defined(AFS_DARWIN_ENV) iheldthelock = VOP_ISLOCKED(vp); if (!iheldthelock) vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, current_proc()); /* this is messy. we can call fsync which will try to reobtain this */ if (VTOAFS(vp) == tvc) ReleaseWriteLock(&tvc->lock); if (UBCINFOEXISTS(vp)) { vinvalbuf(vp, V_SAVE, &afs_osi_cred, current_proc(), PINOD, 0); } if (VTOAFS(vp) == tvc) ObtainWriteLock(&tvc->lock, 954); if (!iheldthelock) VOP_UNLOCK(vp, LK_EXCLUSIVE, current_proc()); #elif defined(AFS_FBSD80_ENV) iheldthelock = VOP_ISLOCKED(vp); if (!iheldthelock) { /* nosleep/sleep lock order reversal */ int glocked = ISAFS_GLOCK(); if (glocked) AFS_GUNLOCK(); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); if (glocked) AFS_GLOCK(); } vinvalbuf(vp, V_SAVE, PINOD, 0); /* changed late in 8.0-CURRENT */ if (!iheldthelock) VOP_UNLOCK(vp, 0); #elif defined(AFS_FBSD60_ENV) iheldthelock = VOP_ISLOCKED(vp, curthread); if (!iheldthelock) vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, curthread); vinvalbuf(vp, V_SAVE, curthread, PINOD, 0); if (!iheldthelock) VOP_UNLOCK(vp, LK_EXCLUSIVE, curthread); #elif defined(AFS_FBSD_ENV) iheldthelock = VOP_ISLOCKED(vp, curthread); if (!iheldthelock) vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, curthread); vinvalbuf(vp, V_SAVE, osi_curcred(), curthread, PINOD, 0); if (!iheldthelock) VOP_UNLOCK(vp, LK_EXCLUSIVE, curthread); #elif defined(AFS_OBSD_ENV) iheldthelock = VOP_ISLOCKED(vp, curproc); if (!iheldthelock) VOP_LOCK(vp, LK_EXCLUSIVE | LK_RETRY, curproc); uvm_vnp_uncache(vp); if (!iheldthelock) VOP_UNLOCK(vp, 0, curproc); #elif defined(AFS_NBSD40_ENV) iheldthelock = VOP_ISLOCKED(vp); if (!iheldthelock) { VOP_LOCK(vp, LK_EXCLUSIVE | LK_RETRY); } uvm_vnp_uncache(vp); if (!iheldthelock) VOP_UNLOCK(vp, 0); #endif } #endif #endif ObtainWriteLock(&afs_xcbhash, 464); tvc->f.states &= ~CUnique; tvc->callback = 0; afs_DequeueCallback(tvc); ReleaseWriteLock(&afs_xcbhash); /* It is always appropriate to throw away all the access rights? */ afs_FreeAllAxs(&(tvc->Access)); tvp = afs_GetVolume(afid, areq, READ_LOCK); /* copy useful per-volume info */ if (tvp) { if ((tvp->states & VForeign)) { if (newvcache) tvc->f.states |= CForeign; if (newvcache && (tvp->rootVnode == afid->Fid.Vnode) && (tvp->rootUnique == afid->Fid.Unique)) { tvc->mvstat = 2; } } if (tvp->states & VRO) tvc->f.states |= CRO; if (tvp->states & VBackup) tvc->f.states |= CBackup; /* now copy ".." entry back out of volume structure, if necessary */ if (tvc->mvstat == 2 && tvp->dotdot.Fid.Volume != 0) { if (!tvc->mvid) tvc->mvid = (struct VenusFid *) osi_AllocSmallSpace(sizeof(struct VenusFid)); *tvc->mvid = tvp->dotdot; } afs_PutVolume(tvp, READ_LOCK); } /* stat the file */ afs_RemoveVCB(afid); { struct AFSFetchStatus OutStatus; if (afs_DynrootNewVnode(tvc, &OutStatus)) { afs_ProcessFS(tvc, &OutStatus, areq); tvc->f.states |= CStatd | CUnique; tvc->f.parent.vnode = OutStatus.ParentVnode; tvc->f.parent.unique = OutStatus.ParentUnique; code = 0; } else { if (AFS_IS_DISCONNECTED) { /* Nothing to do otherwise...*/ code = ENETDOWN; /* printf("Network is down in afs_GetCache"); */ } else code = afs_FetchStatus(tvc, afid, areq, &OutStatus); /* For the NFS translator's benefit, make sure * non-directory vnodes always have their parent FID set * correctly, even when created as a result of decoding an * NFS filehandle. It would be nice to also do this for * directories, but we can't because the fileserver fills * in the FID of the directory itself instead of that of * its parent. */ if (!code && OutStatus.FileType != Directory && !tvc->f.parent.vnode) { tvc->f.parent.vnode = OutStatus.ParentVnode; tvc->f.parent.unique = OutStatus.ParentUnique; /* XXX - SXW - It's conceivable we should mark ourselves * as dirty again here, incase we've been raced * out of the FetchStatus call. */ } } } if (code) { ReleaseWriteLock(&tvc->lock); afs_PutVCache(tvc); return NULL; } ReleaseWriteLock(&tvc->lock); return tvc; } /*afs_GetVCache */ /*! * Lookup a vcache by fid. Look inside the cache first, if not * there, lookup the file on the server, and then get it's fresh * cache entry. * * \param afid * \param areq * \param cached Is element cached? If NULL, don't answer. * \param adp * \param aname * * \return The found element or NULL. */ struct vcache * afs_LookupVCache(struct VenusFid *afid, struct vrequest *areq, afs_int32 * cached, struct vcache *adp, char *aname) { afs_int32 code, now, newvcache = 0; struct VenusFid nfid; register struct vcache *tvc; struct volume *tvp; struct AFSFetchStatus OutStatus; struct AFSCallBack CallBack; struct AFSVolSync tsync; struct server *serverp = 0; afs_int32 origCBs; afs_int32 retry; AFS_STATCNT(afs_GetVCache); if (cached) *cached = 0; /* Init just in case */ #if defined(AFS_SGI_ENV) && !defined(AFS_SGI53_ENV) loop1: #endif ObtainReadLock(&afs_xvcache); tvc = afs_FindVCache(afid, &retry, DO_STATS /* no vlru */ ); if (tvc) { ReleaseReadLock(&afs_xvcache); if (retry) { #if defined(AFS_SGI_ENV) && !defined(AFS_SGI53_ENV) spunlock_psema(tvc->v.v_lock, retry, &tvc->v.v_sync, PINOD); goto loop1; #endif } ObtainReadLock(&tvc->lock); if (tvc->f.states & CStatd) { if (cached) { *cached = 1; } ReleaseReadLock(&tvc->lock); return tvc; } tvc->f.states &= ~CUnique; ReleaseReadLock(&tvc->lock); afs_PutVCache(tvc); ObtainReadLock(&afs_xvcache); } /* if (tvc) */ ReleaseReadLock(&afs_xvcache); /* lookup the file */ nfid = *afid; now = osi_Time(); origCBs = afs_allCBs; /* if anything changes, we don't have a cb */ if (AFS_IS_DISCONNECTED) { /* printf("Network is down in afs_LookupVcache\n"); */ code = ENETDOWN; } else code = afs_RemoteLookup(&adp->f.fid, areq, aname, &nfid, &OutStatus, &CallBack, &serverp, &tsync); #if defined(AFS_SGI_ENV) && !defined(AFS_SGI53_ENV) loop2: #endif ObtainSharedLock(&afs_xvcache, 6); tvc = afs_FindVCache(&nfid, &retry, DO_VLRU | IS_SLOCK/* no xstats now */ ); if (tvc && retry) { #if defined(AFS_SGI_ENV) && !defined(AFS_SGI53_ENV) ReleaseSharedLock(&afs_xvcache); spunlock_psema(tvc->v.v_lock, retry, &tvc->v.v_sync, PINOD); goto loop2; #endif } if (!tvc) { /* no cache entry, better grab one */ UpgradeSToWLock(&afs_xvcache, 22); tvc = afs_NewVCache(&nfid, serverp); newvcache = 1; ConvertWToSLock(&afs_xvcache); if (!tvc) { ReleaseSharedLock(&afs_xvcache); return NULL; } } ReleaseSharedLock(&afs_xvcache); ObtainWriteLock(&tvc->lock, 55); /* It is always appropriate to throw away all the access rights? */ afs_FreeAllAxs(&(tvc->Access)); tvp = afs_GetVolume(afid, areq, READ_LOCK); /* copy useful per-vol info */ if (tvp) { if ((tvp->states & VForeign)) { if (newvcache) tvc->f.states |= CForeign; if (newvcache && (tvp->rootVnode == afid->Fid.Vnode) && (tvp->rootUnique == afid->Fid.Unique)) tvc->mvstat = 2; } if (tvp->states & VRO) tvc->f.states |= CRO; if (tvp->states & VBackup) tvc->f.states |= CBackup; /* now copy ".." entry back out of volume structure, if necessary */ if (tvc->mvstat == 2 && tvp->dotdot.Fid.Volume != 0) { if (!tvc->mvid) tvc->mvid = (struct VenusFid *) osi_AllocSmallSpace(sizeof(struct VenusFid)); *tvc->mvid = tvp->dotdot; } } if (code) { ObtainWriteLock(&afs_xcbhash, 465); afs_DequeueCallback(tvc); tvc->f.states &= ~(CStatd | CUnique); ReleaseWriteLock(&afs_xcbhash); if ((tvc->f.states & CForeign) || (tvc->f.fid.Fid.Vnode & 1)) osi_dnlc_purgedp(tvc); /* if it (could be) a directory */ if (tvp) afs_PutVolume(tvp, READ_LOCK); ReleaseWriteLock(&tvc->lock); afs_PutVCache(tvc); return NULL; } ObtainWriteLock(&afs_xcbhash, 466); if (origCBs == afs_allCBs) { if (CallBack.ExpirationTime) { tvc->callback = serverp; tvc->cbExpires = CallBack.ExpirationTime + now; tvc->f.states |= CStatd | CUnique; tvc->f.states &= ~CBulkFetching; afs_QueueCallback(tvc, CBHash(CallBack.ExpirationTime), tvp); } else if (tvc->f.states & CRO) { /* adapt gives us an hour. */ tvc->cbExpires = 3600 + osi_Time(); /*XXX*/ tvc->f.states |= CStatd | CUnique; tvc->f.states &= ~CBulkFetching; afs_QueueCallback(tvc, CBHash(3600), tvp); } else { tvc->callback = NULL; afs_DequeueCallback(tvc); tvc->f.states &= ~(CStatd | CUnique); if ((tvc->f.states & CForeign) || (tvc->f.fid.Fid.Vnode & 1)) osi_dnlc_purgedp(tvc); /* if it (could be) a directory */ } } else { afs_DequeueCallback(tvc); tvc->f.states &= ~CStatd; tvc->f.states &= ~CUnique; tvc->callback = NULL; if ((tvc->f.states & CForeign) || (tvc->f.fid.Fid.Vnode & 1)) osi_dnlc_purgedp(tvc); /* if it (could be) a directory */ } ReleaseWriteLock(&afs_xcbhash); if (tvp) afs_PutVolume(tvp, READ_LOCK); afs_ProcessFS(tvc, &OutStatus, areq); ReleaseWriteLock(&tvc->lock); return tvc; } struct vcache * afs_GetRootVCache(struct VenusFid *afid, struct vrequest *areq, afs_int32 * cached, struct volume *tvolp) { afs_int32 code = 0, i, newvcache = 0, haveStatus = 0; afs_int32 getNewFid = 0; afs_uint32 start; struct VenusFid nfid; register struct vcache *tvc; struct server *serverp = 0; struct AFSFetchStatus OutStatus; struct AFSCallBack CallBack; struct AFSVolSync tsync; int origCBs = 0; #ifdef AFS_DARWIN80_ENV vnode_t tvp; #endif start = osi_Time(); newmtpt: if (!tvolp->rootVnode || getNewFid) { struct VenusFid tfid; tfid = *afid; tfid.Fid.Vnode = 0; /* Means get rootfid of volume */ origCBs = afs_allCBs; /* ignore InitCallBackState */ code = afs_RemoteLookup(&tfid, areq, NULL, &nfid, &OutStatus, &CallBack, &serverp, &tsync); if (code) { return NULL; } /* ReleaseReadLock(&tvolp->lock); */ ObtainWriteLock(&tvolp->lock, 56); tvolp->rootVnode = afid->Fid.Vnode = nfid.Fid.Vnode; tvolp->rootUnique = afid->Fid.Unique = nfid.Fid.Unique; ReleaseWriteLock(&tvolp->lock); /* ObtainReadLock(&tvolp->lock);*/ haveStatus = 1; } else { afid->Fid.Vnode = tvolp->rootVnode; afid->Fid.Unique = tvolp->rootUnique; } rootvc_loop: ObtainSharedLock(&afs_xvcache, 7); i = VCHash(afid); for (tvc = afs_vhashT[i]; tvc; tvc = tvc->hnext) { if (!FidCmp(&(tvc->f.fid), afid)) { if (tvc->f.states & CVInit) { ReleaseSharedLock(&afs_xvcache); afs_osi_Sleep(&tvc->f.states); goto rootvc_loop; } #ifdef AFS_DARWIN80_ENV if (tvc->f.states & CDeadVnode) { if (!(tvc->f.states & CBulkFetching)) { ReleaseSharedLock(&afs_xvcache); afs_osi_Sleep(&tvc->f.states); goto rootvc_loop; } } tvp = AFSTOV(tvc); if (vnode_get(tvp)) /* this bumps ref count */ continue; if (vnode_ref(tvp)) { AFS_GUNLOCK(); /* AFSTOV(tvc) may be NULL */ vnode_put(tvp); AFS_GLOCK(); continue; } if (tvc->f.states & (CBulkFetching|CDeadVnode)) { AFS_GUNLOCK(); vnode_recycle(AFSTOV(tvc)); AFS_GLOCK(); } #endif break; } } if (!haveStatus && (!tvc || !(tvc->f.states & CStatd))) { /* Mount point no longer stat'd or unknown. FID may have changed. */ getNewFid = 1; ReleaseSharedLock(&afs_xvcache); #ifdef AFS_DARWIN80_ENV if (tvc) { AFS_GUNLOCK(); vnode_put(AFSTOV(tvc)); vnode_rele(AFSTOV(tvc)); AFS_GLOCK(); } #endif tvc = NULL; goto newmtpt; } if (!tvc) { UpgradeSToWLock(&afs_xvcache, 23); /* no cache entry, better grab one */ tvc = afs_NewVCache(afid, NULL); if (!tvc) { ReleaseWriteLock(&afs_xvcache); return NULL; } newvcache = 1; afs_stats_cmperf.vcacheMisses++; } else { if (cached) *cached = 1; afs_stats_cmperf.vcacheHits++; #if defined(AFS_DARWIN80_ENV) /* we already bumped the ref count in the for loop above */ #else /* AFS_DARWIN80_ENV */ osi_vnhold(tvc, 0); #endif UpgradeSToWLock(&afs_xvcache, 24); if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) { refpanic("GRVC VLRU inconsistent0"); } if (tvc->vlruq.next->prev != &(tvc->vlruq)) { refpanic("GRVC VLRU inconsistent1"); } if (tvc->vlruq.prev->next != &(tvc->vlruq)) { refpanic("GRVC VLRU inconsistent2"); } QRemove(&tvc->vlruq); /* move to lruq head */ QAdd(&VLRU, &tvc->vlruq); if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) { refpanic("GRVC VLRU inconsistent3"); } if (tvc->vlruq.next->prev != &(tvc->vlruq)) { refpanic("GRVC VLRU inconsistent4"); } if (tvc->vlruq.prev->next != &(tvc->vlruq)) { refpanic("GRVC VLRU inconsistent5"); } vcachegen++; } ReleaseWriteLock(&afs_xvcache); if (tvc->f.states & CStatd) { return tvc; } else { ObtainReadLock(&tvc->lock); tvc->f.states &= ~CUnique; tvc->callback = NULL; /* redundant, perhaps */ ReleaseReadLock(&tvc->lock); } ObtainWriteLock(&tvc->lock, 57); /* It is always appropriate to throw away all the access rights? */ afs_FreeAllAxs(&(tvc->Access)); if (newvcache) tvc->f.states |= CForeign; if (tvolp->states & VRO) tvc->f.states |= CRO; if (tvolp->states & VBackup) tvc->f.states |= CBackup; /* now copy ".." entry back out of volume structure, if necessary */ if (newvcache && (tvolp->rootVnode == afid->Fid.Vnode) && (tvolp->rootUnique == afid->Fid.Unique)) { tvc->mvstat = 2; } if (tvc->mvstat == 2 && tvolp->dotdot.Fid.Volume != 0) { if (!tvc->mvid) tvc->mvid = (struct VenusFid *) osi_AllocSmallSpace(sizeof(struct VenusFid)); *tvc->mvid = tvolp->dotdot; } /* stat the file */ afs_RemoveVCB(afid); if (!haveStatus) { struct VenusFid tfid; tfid = *afid; tfid.Fid.Vnode = 0; /* Means get rootfid of volume */ origCBs = afs_allCBs; /* ignore InitCallBackState */ code = afs_RemoteLookup(&tfid, areq, NULL, &nfid, &OutStatus, &CallBack, &serverp, &tsync); } if (code) { ObtainWriteLock(&afs_xcbhash, 467); afs_DequeueCallback(tvc); tvc->callback = NULL; tvc->f.states &= ~(CStatd | CUnique); ReleaseWriteLock(&afs_xcbhash); if ((tvc->f.states & CForeign) || (tvc->f.fid.Fid.Vnode & 1)) osi_dnlc_purgedp(tvc); /* if it (could be) a directory */ ReleaseWriteLock(&tvc->lock); afs_PutVCache(tvc); return NULL; } ObtainWriteLock(&afs_xcbhash, 468); if (origCBs == afs_allCBs) { tvc->f.states |= CTruth; tvc->callback = serverp; if (CallBack.ExpirationTime != 0) { tvc->cbExpires = CallBack.ExpirationTime + start; tvc->f.states |= CStatd; tvc->f.states &= ~CBulkFetching; afs_QueueCallback(tvc, CBHash(CallBack.ExpirationTime), tvolp); } else if (tvc->f.states & CRO) { /* adapt gives us an hour. */ tvc->cbExpires = 3600 + osi_Time(); /*XXX*/ tvc->f.states |= CStatd; tvc->f.states &= ~CBulkFetching; afs_QueueCallback(tvc, CBHash(3600), tvolp); } } else { afs_DequeueCallback(tvc); tvc->callback = NULL; tvc->f.states &= ~(CStatd | CUnique); if ((tvc->f.states & CForeign) || (tvc->f.fid.Fid.Vnode & 1)) osi_dnlc_purgedp(tvc); /* if it (could be) a directory */ } ReleaseWriteLock(&afs_xcbhash); afs_ProcessFS(tvc, &OutStatus, areq); ReleaseWriteLock(&tvc->lock); return tvc; } /*! * Update callback status and (sometimes) attributes of a vnode. * Called after doing a fetch status RPC. Whilst disconnected, attributes * shouldn't be written to the vcache here. * * \param avc * \param afid * \param areq * \param Outsp Server status after rpc call. * \param acb Callback for this vnode. * * \note The vcache must be write locked. */ void afs_UpdateStatus(struct vcache *avc, struct VenusFid *afid, struct vrequest *areq, struct AFSFetchStatus *Outsp, struct AFSCallBack *acb, afs_uint32 start) { struct volume *volp; if (!AFS_IN_SYNC) /* Dont write status in vcache if resyncing after a disconnection. */ afs_ProcessFS(avc, Outsp, areq); volp = afs_GetVolume(afid, areq, READ_LOCK); ObtainWriteLock(&afs_xcbhash, 469); avc->f.states |= CTruth; if (avc->callback /* check for race */ ) { if (acb->ExpirationTime != 0) { avc->cbExpires = acb->ExpirationTime + start; avc->f.states |= CStatd; avc->f.states &= ~CBulkFetching; afs_QueueCallback(avc, CBHash(acb->ExpirationTime), volp); } else if (avc->f.states & CRO) { /* ordinary callback on a read-only volume -- AFS 3.2 style */ avc->cbExpires = 3600 + start; avc->f.states |= CStatd; avc->f.states &= ~CBulkFetching; afs_QueueCallback(avc, CBHash(3600), volp); } else { afs_DequeueCallback(avc); avc->callback = NULL; avc->f.states &= ~(CStatd | CUnique); if ((avc->f.states & CForeign) || (avc->f.fid.Fid.Vnode & 1)) osi_dnlc_purgedp(avc); /* if it (could be) a directory */ } } else { afs_DequeueCallback(avc); avc->callback = NULL; avc->f.states &= ~(CStatd | CUnique); if ((avc->f.states & CForeign) || (avc->f.fid.Fid.Vnode & 1)) osi_dnlc_purgedp(avc); /* if it (could be) a directory */ } ReleaseWriteLock(&afs_xcbhash); if (volp) afs_PutVolume(volp, READ_LOCK); } /*! * Must be called with avc write-locked * don't absolutely have to invalidate the hint unless the dv has * changed, but be sure to get it right else there will be consistency bugs. */ afs_int32 afs_FetchStatus(struct vcache * avc, struct VenusFid * afid, struct vrequest * areq, struct AFSFetchStatus * Outsp) { int code; afs_uint32 start = 0; register struct afs_conn *tc; struct AFSCallBack CallBack; struct AFSVolSync tsync; XSTATS_DECLS; do { tc = afs_Conn(afid, areq, SHARED_LOCK); avc->dchint = NULL; /* invalidate hints */ if (tc) { avc->callback = tc->srvr->server; start = osi_Time(); XSTATS_START_TIME(AFS_STATS_FS_RPCIDX_FETCHSTATUS); RX_AFS_GUNLOCK(); code = RXAFS_FetchStatus(tc->id, (struct AFSFid *)&afid->Fid, Outsp, &CallBack, &tsync); RX_AFS_GLOCK(); XSTATS_END_TIME; } else code = -1; } while (afs_Analyze (tc, code, afid, areq, AFS_STATS_FS_RPCIDX_FETCHSTATUS, SHARED_LOCK, NULL)); if (!code) { afs_UpdateStatus(avc, afid, areq, Outsp, &CallBack, start); } else { /* used to undo the local callback, but that's too extreme. * There are plenty of good reasons that fetchstatus might return * an error, such as EPERM. If we have the vnode cached, statd, * with callback, might as well keep track of the fact that we * don't have access... */ if (code == EPERM || code == EACCES) { struct axscache *ac; if (avc->Access && (ac = afs_FindAxs(avc->Access, areq->uid))) ac->axess = 0; else /* not found, add a new one if possible */ afs_AddAxs(avc->Access, areq->uid, 0); } } return code; } #if 0 /* * afs_StuffVcache * * Description: * Stuff some information into the vcache for the given file. * * Parameters: * afid : File in question. * OutStatus : Fetch status on the file. * CallBack : Callback info. * tc : RPC connection involved. * areq : vrequest involved. * * Environment: * Nothing interesting. */ void afs_StuffVcache(register struct VenusFid *afid, struct AFSFetchStatus *OutStatus, struct AFSCallBack *CallBack, register struct afs_conn *tc, struct vrequest *areq) { register afs_int32 code, i, newvcache = 0; register struct vcache *tvc; struct AFSVolSync tsync; struct volume *tvp; struct axscache *ac; afs_int32 retry; AFS_STATCNT(afs_StuffVcache); #ifdef IFS_VCACHECOUNT ifs_gvcachecall++; #endif loop: ObtainSharedLock(&afs_xvcache, 8); tvc = afs_FindVCache(afid, &retry, DO_VLRU| IS_SLOCK /* no stats */ ); if (tvc && retry) { #if defined(AFS_SGI_ENV) && !defined(AFS_SGI53_ENV) ReleaseSharedLock(&afs_xvcache); spunlock_psema(tvc->v.v_lock, retry, &tvc->v.v_sync, PINOD); goto loop; #endif } if (!tvc) { /* no cache entry, better grab one */ UpgradeSToWLock(&afs_xvcache, 25); tvc = afs_NewVCache(afid, NULL); newvcache = 1; ConvertWToSLock(&afs_xvcache); if (!tvc) { ReleaseSharedLock(&afs_xvcache); return NULL; } } ReleaseSharedLock(&afs_xvcache); ObtainWriteLock(&tvc->lock, 58); tvc->f.states &= ~CStatd; if ((tvc->f.states & CForeign) || (tvc->f.fid.Fid.Vnode & 1)) osi_dnlc_purgedp(tvc); /* if it (could be) a directory */ /* Is it always appropriate to throw away all the access rights? */ afs_FreeAllAxs(&(tvc->Access)); /*Copy useful per-volume info */ tvp = afs_GetVolume(afid, areq, READ_LOCK); if (tvp) { if (newvcache && (tvp->states & VForeign)) tvc->f.states |= CForeign; if (tvp->states & VRO) tvc->f.states |= CRO; if (tvp->states & VBackup) tvc->f.states |= CBackup; /* * Now, copy ".." entry back out of volume structure, if * necessary */ if (tvc->mvstat == 2 && tvp->dotdot.Fid.Volume != 0) { if (!tvc->mvid) tvc->mvid = (struct VenusFid *) osi_AllocSmallSpace(sizeof(struct VenusFid)); *tvc->mvid = tvp->dotdot; } } /* store the stat on the file */ afs_RemoveVCB(afid); afs_ProcessFS(tvc, OutStatus, areq); tvc->callback = tc->srvr->server; /* we use osi_Time twice below. Ideally, we would use the time at which * the FetchStatus call began, instead, but we don't have it here. So we * make do with "now". In the CRO case, it doesn't really matter. In * the other case, we hope that the difference between "now" and when the * call actually began execution on the server won't be larger than the * padding which the server keeps. Subtract 1 second anyway, to be on * the safe side. Can't subtract more because we don't know how big * ExpirationTime is. Possible consistency problems may arise if the call * timeout period becomes longer than the server's expiration padding. */ ObtainWriteLock(&afs_xcbhash, 470); if (CallBack->ExpirationTime != 0) { tvc->cbExpires = CallBack->ExpirationTime + osi_Time() - 1; tvc->f.states |= CStatd; tvc->f.states &= ~CBulkFetching; afs_QueueCallback(tvc, CBHash(CallBack->ExpirationTime), tvp); } else if (tvc->f.states & CRO) { /* old-fashioned AFS 3.2 style */ tvc->cbExpires = 3600 + osi_Time(); /*XXX*/ tvc->f.states |= CStatd; tvc->f.states &= ~CBulkFetching; afs_QueueCallback(tvc, CBHash(3600), tvp); } else { afs_DequeueCallback(tvc); tvc->callback = NULL; tvc->f.states &= ~(CStatd | CUnique); if ((tvc->f.states & CForeign) || (tvc->f.fid.Fid.Vnode & 1)) osi_dnlc_purgedp(tvc); /* if it (could be) a directory */ } ReleaseWriteLock(&afs_xcbhash); if (tvp) afs_PutVolume(tvp, READ_LOCK); /* look in per-pag cache */ if (tvc->Access && (ac = afs_FindAxs(tvc->Access, areq->uid))) ac->axess = OutStatus->CallerAccess; /* substitute pags */ else /* not found, add a new one if possible */ afs_AddAxs(tvc->Access, areq->uid, OutStatus->CallerAccess); ReleaseWriteLock(&tvc->lock); afs_Trace4(afs_iclSetp, CM_TRACE_STUFFVCACHE, ICL_TYPE_POINTER, tvc, ICL_TYPE_POINTER, tvc->callback, ICL_TYPE_INT32, tvc->cbExpires, ICL_TYPE_INT32, tvc->cbExpires - osi_Time()); /* * Release ref count... hope this guy stays around... */ afs_PutVCache(tvc); } /*afs_StuffVcache */ #endif /*! * Decrements the reference count on a cache entry. * * \param avc Pointer to the cache entry to decrement. * * \note Environment: Nothing interesting. */ void afs_PutVCache(register struct vcache *avc) { AFS_STATCNT(afs_PutVCache); #ifdef AFS_DARWIN80_ENV vnode_put(AFSTOV(avc)); AFS_FAST_RELE(avc); #else /* * Can we use a read lock here? */ ObtainReadLock(&afs_xvcache); AFS_FAST_RELE(avc); ReleaseReadLock(&afs_xvcache); #endif } /*afs_PutVCache */ /*! * Reset a vcache entry, so local contents are ignored, and the * server will be reconsulted next time the vcache is used * * \param avc Pointer to the cache entry to reset * \param acred * * \note avc must be write locked on entry */ void afs_ResetVCache(struct vcache *avc, afs_ucred_t *acred) { ObtainWriteLock(&afs_xcbhash, 456); afs_DequeueCallback(avc); avc->f.states &= ~(CStatd | CDirty); /* next reference will re-stat */ ReleaseWriteLock(&afs_xcbhash); /* now find the disk cache entries */ afs_TryToSmush(avc, acred, 1); osi_dnlc_purgedp(avc); if (avc->linkData && !(avc->f.states & CCore)) { afs_osi_Free(avc->linkData, strlen(avc->linkData) + 1); avc->linkData = NULL; } } /*! * Sleepa when searching for a vcache. Releases all the pending locks, * sleeps then obtains the previously released locks. * * \param vcache Enter sleep state. * \param flag Determines what locks to use. * * \return */ static void findvc_sleep(struct vcache *avc, int flag) { int fstates = avc->f.states; if (flag & IS_SLOCK) { ReleaseSharedLock(&afs_xvcache); } else { if (flag & IS_WLOCK) { ReleaseWriteLock(&afs_xvcache); } else { ReleaseReadLock(&afs_xvcache); } } if (flag & FIND_CDEAD) { ObtainWriteLock(&afs_xvcache, 342); afs_FlushReclaimedVcaches(); if (fstates == avc->f.states) { ReleaseWriteLock(&afs_xvcache); afs_osi_Sleep(&avc->f.states); } else ReleaseWriteLock(&afs_xvcache); } else afs_osi_Sleep(&avc->f.states); if (flag & IS_SLOCK) { ObtainSharedLock(&afs_xvcache, 341); } else { if (flag & IS_WLOCK) { ObtainWriteLock(&afs_xvcache, 343); } else { ObtainReadLock(&afs_xvcache); } } } /*! * Add a reference on an existing vcache entry. * * \param tvc Pointer to the vcache. * * \note Environment: Must be called with at least one reference from * elsewhere on the vcache, even if that reference will be dropped. * The global lock is required. * * \return 0 on success, -1 on failure. */ int afs_RefVCache(struct vcache *tvc) { #ifdef AFS_DARWIN80_ENV vnode_t tvp; #endif /* AFS_STATCNT(afs_RefVCache); */ #ifdef AFS_DARWIN80_ENV tvp = AFSTOV(tvc); if (vnode_get(tvp)) return -1; if (vnode_ref(tvp)) { AFS_GUNLOCK(); /* AFSTOV(tvc) may be NULL */ vnode_put(tvp); AFS_GLOCK(); return -1; } #else osi_vnhold(tvc, 0); #endif return 0; } /*afs_RefVCache */ /*! * Find a vcache entry given a fid. * * \param afid Pointer to the fid whose cache entry we desire. * \param retry (SGI-specific) tell the caller to drop the lock on xvcache, * unlock the vnode, and try again. * \param flag Bit 1 to specify whether to compute hit statistics. Not * set if FindVCache is called as part of internal bookkeeping. * * \note Environment: Must be called with the afs_xvcache lock at least held at * the read level. In order to do the VLRU adjustment, the xvcache lock * must be shared-- we upgrade it here. */ struct vcache * afs_FindVCache(struct VenusFid *afid, afs_int32 * retry, afs_int32 flag) { register struct vcache *tvc; afs_int32 i; #ifdef AFS_DARWIN80_ENV vnode_t tvp; #endif AFS_STATCNT(afs_FindVCache); findloop: i = VCHash(afid); for (tvc = afs_vhashT[i]; tvc; tvc = tvc->hnext) { if (FidMatches(afid, tvc)) { if (tvc->f.states & CVInit) { findvc_sleep(tvc, flag); goto findloop; } #ifdef AFS_DARWIN80_ENV if (tvc->f.states & CDeadVnode) { if (!(flag & FIND_CDEAD)) { findvc_sleep(tvc, flag); goto findloop; } } tvp = AFSTOV(tvc); if (vnode_get(tvp)) continue; if (vnode_ref(tvp)) { AFS_GUNLOCK(); /* AFSTOV(tvc) may be NULL */ vnode_put(tvp); AFS_GLOCK(); continue; } if (tvc->f.states & (CBulkFetching|CDeadVnode)) { AFS_GUNLOCK(); vnode_recycle(AFSTOV(tvc)); AFS_GLOCK(); } #endif break; } } /* should I have a read lock on the vnode here? */ if (tvc) { if (retry) *retry = 0; #if !defined(AFS_DARWIN80_ENV) osi_vnhold(tvc, retry); /* already held, above */ if (retry && *retry) return 0; #endif #if defined(AFS_DARWIN_ENV) && !defined(AFS_DARWIN80_ENV) tvc->f.states |= CUBCinit; AFS_GUNLOCK(); if (UBCINFOMISSING(AFSTOV(tvc)) || UBCINFORECLAIMED(AFSTOV(tvc))) { ubc_info_init(AFSTOV(tvc)); } AFS_GLOCK(); tvc->f.states &= ~CUBCinit; #endif /* * only move to front of vlru if we have proper vcache locking) */ if (flag & DO_VLRU) { if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) { refpanic("FindVC VLRU inconsistent1"); } if (tvc->vlruq.next->prev != &(tvc->vlruq)) { refpanic("FindVC VLRU inconsistent1"); } if (tvc->vlruq.prev->next != &(tvc->vlruq)) { refpanic("FindVC VLRU inconsistent2"); } UpgradeSToWLock(&afs_xvcache, 26); QRemove(&tvc->vlruq); QAdd(&VLRU, &tvc->vlruq); ConvertWToSLock(&afs_xvcache); if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) { refpanic("FindVC VLRU inconsistent1"); } if (tvc->vlruq.next->prev != &(tvc->vlruq)) { refpanic("FindVC VLRU inconsistent2"); } if (tvc->vlruq.prev->next != &(tvc->vlruq)) { refpanic("FindVC VLRU inconsistent3"); } } vcachegen++; } if (flag & DO_STATS) { if (tvc) afs_stats_cmperf.vcacheHits++; else afs_stats_cmperf.vcacheMisses++; if (afs_IsPrimaryCellNum(afid->Cell)) afs_stats_cmperf.vlocalAccesses++; else afs_stats_cmperf.vremoteAccesses++; } return tvc; } /*afs_FindVCache */ /*! * Find a vcache entry given a fid. Does a wildcard match on what we * have for the fid. If more than one entry, don't return anything. * * \param avcp Fill in pointer if we found one and only one. * \param afid Pointer to the fid whose cache entry we desire. * \param retry (SGI-specific) tell the caller to drop the lock on xvcache, * unlock the vnode, and try again. * \param flags bit 1 to specify whether to compute hit statistics. Not * set if FindVCache is called as part of internal bookkeeping. * * \note Environment: Must be called with the afs_xvcache lock at least held at * the read level. In order to do the VLRU adjustment, the xvcache lock * must be shared-- we upgrade it here. * * \return Number of matches found. */ int afs_duplicate_nfs_fids = 0; afs_int32 afs_NFSFindVCache(struct vcache **avcp, struct VenusFid *afid) { register struct vcache *tvc; afs_int32 i; afs_int32 count = 0; struct vcache *found_tvc = NULL; #ifdef AFS_DARWIN80_ENV vnode_t tvp; #endif AFS_STATCNT(afs_FindVCache); loop: ObtainSharedLock(&afs_xvcache, 331); i = VCHash(afid); for (tvc = afs_vhashT[i]; tvc; tvc = tvc->hnext) { /* Match only on what we have.... */ if (((tvc->f.fid.Fid.Vnode & 0xffff) == afid->Fid.Vnode) && (tvc->f.fid.Fid.Volume == afid->Fid.Volume) && ((tvc->f.fid.Fid.Unique & 0xffffff) == afid->Fid.Unique) && (tvc->f.fid.Cell == afid->Cell)) { if (tvc->f.states & CVInit) { ReleaseSharedLock(&afs_xvcache); afs_osi_Sleep(&tvc->f.states); goto loop; } #ifdef AFS_DARWIN80_ENV if (tvc->f.states & CDeadVnode) { if (!(tvc->f.states & CBulkFetching)) { ReleaseSharedLock(&afs_xvcache); afs_osi_Sleep(&tvc->f.states); goto loop; } } tvp = AFSTOV(tvc); if (vnode_get(tvp)) { /* This vnode no longer exists. */ continue; } if (vnode_ref(tvp)) { /* This vnode no longer exists. */ AFS_GUNLOCK(); /* AFSTOV(tvc) may be NULL */ vnode_put(tvp); AFS_GLOCK(); continue; } if (tvc->f.states & (CBulkFetching|CDeadVnode)) { AFS_GUNLOCK(); vnode_recycle(AFSTOV(tvc)); AFS_GLOCK(); } #endif /* AFS_DARWIN80_ENV */ count++; if (found_tvc) { /* Duplicates */ afs_duplicate_nfs_fids++; ReleaseSharedLock(&afs_xvcache); #ifdef AFS_DARWIN80_ENV /* Drop our reference counts. */ vnode_put(AFSTOV(tvc)); vnode_put(AFSTOV(found_tvc)); #endif return count; } found_tvc = tvc; } } tvc = found_tvc; /* should I have a read lock on the vnode here? */ if (tvc) { #if defined(AFS_SGI_ENV) && !defined(AFS_SGI53_ENV) afs_int32 retry = 0; osi_vnhold(tvc, &retry); if (retry) { count = 0; found_tvc = (struct vcache *)0; ReleaseSharedLock(&afs_xvcache); spunlock_psema(tvc->v.v_lock, retry, &tvc->v.v_sync, PINOD); goto loop; } #else osi_vnhold(tvc, (int *)0); /* already held, above */ #endif /* * We obtained the xvcache lock above. */ if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) { refpanic("FindVC VLRU inconsistent1"); } if (tvc->vlruq.next->prev != &(tvc->vlruq)) { refpanic("FindVC VLRU inconsistent1"); } if (tvc->vlruq.prev->next != &(tvc->vlruq)) { refpanic("FindVC VLRU inconsistent2"); } UpgradeSToWLock(&afs_xvcache, 568); QRemove(&tvc->vlruq); QAdd(&VLRU, &tvc->vlruq); ConvertWToSLock(&afs_xvcache); if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) { refpanic("FindVC VLRU inconsistent1"); } if (tvc->vlruq.next->prev != &(tvc->vlruq)) { refpanic("FindVC VLRU inconsistent2"); } if (tvc->vlruq.prev->next != &(tvc->vlruq)) { refpanic("FindVC VLRU inconsistent3"); } } vcachegen++; if (tvc) afs_stats_cmperf.vcacheHits++; else afs_stats_cmperf.vcacheMisses++; if (afs_IsPrimaryCellNum(afid->Cell)) afs_stats_cmperf.vlocalAccesses++; else afs_stats_cmperf.vremoteAccesses++; *avcp = tvc; /* May be null */ ReleaseSharedLock(&afs_xvcache); return (tvc ? 1 : 0); } /*afs_NFSFindVCache */ /*! * Initialize vcache related variables * * \param astatSize */ void afs_vcacheInit(int astatSize) { #if !defined(AFS_LINUX22_ENV) register struct vcache *tvp; #endif int i; if (!afs_maxvcount) { afs_maxvcount = astatSize; /* no particular limit on linux? */ } #if !defined(AFS_LINUX22_ENV) freeVCList = NULL; #endif AFS_RWLOCK_INIT(&afs_xvcache, "afs_xvcache"); LOCK_INIT(&afs_xvcb, "afs_xvcb"); #if !defined(AFS_LINUX22_ENV) /* Allocate and thread the struct vcache entries */ tvp = (struct vcache *)afs_osi_Alloc(astatSize * sizeof(struct vcache)); memset(tvp, 0, sizeof(struct vcache) * astatSize); Initial_freeVCList = tvp; freeVCList = &(tvp[0]); for (i = 0; i < astatSize - 1; i++) { tvp[i].nextfree = &(tvp[i + 1]); } tvp[astatSize - 1].nextfree = NULL; # ifdef KERNEL_HAVE_PIN pin((char *)tvp, astatSize * sizeof(struct vcache)); /* XXX */ # endif #endif #if defined(AFS_SGI_ENV) for (i = 0; i < astatSize; i++) { char name[METER_NAMSZ]; struct vcache *tvc = &tvp[i]; tvc->v.v_number = ++afsvnumbers; tvc->vc_rwlockid = OSI_NO_LOCKID; initnsema(&tvc->vc_rwlock, 1, makesname(name, "vrw", tvc->v.v_number)); #ifndef AFS_SGI53_ENV initnsema(&tvc->v.v_sync, 0, makesname(name, "vsy", tvc->v.v_number)); #endif #ifndef AFS_SGI62_ENV initnlock(&tvc->v.v_lock, makesname(name, "vlk", tvc->v.v_number)); #endif /* AFS_SGI62_ENV */ } #endif QInit(&VLRU); for(i = 0; i < VCSIZE; ++i) QInit(&afs_vhashTV[i]); } /*! * Shutdown vcache. */ void shutdown_vcache(void) { int i; struct afs_cbr *tsp; /* * XXX We may potentially miss some of the vcaches because if when * there are no free vcache entries and all the vcache entries are active * ones then we allocate an additional one - admittedly we almost never * had that occur. */ { register struct afs_q *tq, *uq = NULL; register struct vcache *tvc; for (tq = VLRU.prev; tq != &VLRU; tq = uq) { tvc = QTOV(tq); uq = QPrev(tq); if (tvc->mvid) { osi_FreeSmallSpace(tvc->mvid); tvc->mvid = (struct VenusFid *)0; } #ifdef AFS_AIX_ENV aix_gnode_rele(AFSTOV(tvc)); #endif if (tvc->linkData) { afs_osi_Free(tvc->linkData, strlen(tvc->linkData) + 1); tvc->linkData = 0; } } /* * Also free the remaining ones in the Cache */ for (i = 0; i < VCSIZE; i++) { for (tvc = afs_vhashT[i]; tvc; tvc = tvc->hnext) { if (tvc->mvid) { osi_FreeSmallSpace(tvc->mvid); tvc->mvid = (struct VenusFid *)0; } #ifdef AFS_AIX_ENV if (tvc->v.v_gnode) afs_osi_Free(tvc->v.v_gnode, sizeof(struct gnode)); #ifdef AFS_AIX32_ENV if (tvc->segid) { AFS_GUNLOCK(); vms_delete(tvc->segid); AFS_GLOCK(); tvc->segid = tvc->vmh = NULL; if (VREFCOUNT_GT(tvc,0)) osi_Panic("flushVcache: vm race"); } if (tvc->credp) { crfree(tvc->credp); tvc->credp = NULL; } #endif #endif #if defined(AFS_SUN5_ENV) if (tvc->credp) { crfree(tvc->credp); tvc->credp = NULL; } #endif if (tvc->linkData) { afs_osi_Free(tvc->linkData, strlen(tvc->linkData) + 1); tvc->linkData = 0; } if (tvc->Access) afs_FreeAllAxs(&(tvc->Access)); } afs_vhashT[i] = 0; } } /* * Free any leftover callback queue */ for (i = 0; i < afs_stats_cmperf.CallBackAlloced; i++) { tsp = afs_cbrHeads[i]; afs_cbrHeads[i] = 0; afs_osi_Free((char *)tsp, AFS_NCBRS * sizeof(struct afs_cbr)); } afs_cbrSpace = 0; #if !defined(AFS_LINUX22_ENV) afs_osi_Free(Initial_freeVCList, afs_cacheStats * sizeof(struct vcache)); # ifdef KERNEL_HAVE_PIN unpin(Initial_freeVCList, afs_cacheStats * sizeof(struct vcache)); # endif freeVCList = Initial_freeVCList = 0; #endif AFS_RWLOCK_INIT(&afs_xvcache, "afs_xvcache"); LOCK_INIT(&afs_xvcb, "afs_xvcb"); QInit(&VLRU); for(i = 0; i < VCSIZE; ++i) QInit(&afs_vhashTV[i]); } void afs_DisconGiveUpCallbacks(void) { int i; struct vcache *tvc; int nq=0; ObtainWriteLock(&afs_xvcache, 1002); /* XXX - should be a unique number */ /* Somehow, walk the set of vcaches, with each one coming out as tvc */ for (i = 0; i < VCSIZE; i++) { for (tvc = afs_vhashT[i]; tvc; tvc = tvc->hnext) { if (afs_QueueVCB(tvc)) { tvc->callback = NULL; nq++; } } } ReleaseWriteLock(&afs_xvcache); afs_FlushVCBs(1); } /*! * * Clear the Statd flag from all vcaches * * This function removes the Statd flag from all vcaches. It's used by * disconnected mode to tidy up during reconnection * */ void afs_ClearAllStatdFlag(void) { int i; struct vcache *tvc; ObtainWriteLock(&afs_xvcache, 715); for (i = 0; i < VCSIZE; i++) { for (tvc = afs_vhashT[i]; tvc; tvc = tvc->hnext) { tvc->f.states &= ~(CStatd|CUnique); } } ReleaseWriteLock(&afs_xvcache); }