/* Forward declarations */
static afs_int32 afs_QueueVCB(struct vcache *avc, int *slept);
+
+/*
+ * The PFlush algorithm makes use of the fact that Fid.Unique is not used in
+ * below hash algorithms. Change it if need be so that flushing algorithm
+ * doesn't move things from one hash chain to another.
+ */
+/* Don't hash on the cell; our callback-breaking code sometimes fails to compute
+ * the cell correctly, and only scans one hash bucket. */
+int VCHash(struct VenusFid *fid)
+{
+ return opr_jhash_int2(fid->Fid.Volume, fid->Fid.Vnode, 0) &
+ opr_jhash_mask(VCSIZEBITS);
+}
+/* Hash only on volume to speed up volume callbacks. */
+int VCHashV(struct VenusFid *fid)
+{
+ return opr_jhash_int(fid->Fid.Vnode, 0) & opr_jhash_mask(VCSIZEBITS);
+}
+
/*!
* Generate an index into the hash table for a given Fid.
* \param fid
afs_int32 i, code;
struct vcache **uvc, *wvc;
+ /* NOTE: We must have nothing drop afs_xvcache until we have removed all
+ * possible references to this vcache. This means all hash tables, queues,
+ * DNLC, etc. */
+
*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);
+ code = osi_VM_FlushVCache(avc);
if (code)
goto bad;
for (wvc = *uvc; wvc; uvc = &wvc->hnext, wvc = *uvc) {
if (avc == wvc) {
*uvc = avc->hnext;
- avc->hnext = (struct vcache *)NULL;
+ avc->hnext = 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 defined(AFS_LINUX26_ENV)
+ {
+ struct pagewriter *pw, *store;
+ struct list_head tofree;
+
+ INIT_LIST_HEAD(&tofree);
+ spin_lock(&avc->pagewriter_lock);
+ list_for_each_entry_safe(pw, store, &avc->pagewriters, link) {
+ list_del(&pw->link);
+ /* afs_osi_Free may sleep so we need to defer it */
+ list_add_tail(&pw->link, &tofree);
+ }
+ spin_unlock(&avc->pagewriter_lock);
+ list_for_each_entry_safe(pw, store, &tofree, link) {
+ list_del(&pw->link);
+ afs_osi_Free(pw, sizeof(struct pagewriter));
+ }
+ }
+#endif
+
+ if (avc->mvid.target_root)
+ osi_FreeSmallSpace(avc->mvid.target_root);
+ avc->mvid.target_root = NULL;
if (avc->linkData) {
afs_osi_Free(avc->linkData, strlen(avc->linkData) + 1);
avc->linkData = NULL;
vn_reinit(AFSTOV(avc));
#endif
afs_FreeAllAxs(&(avc->Access));
- if (!afs_shuttingdown)
- afs_QueueVCB(avc, slept);
ObtainWriteLock(&afs_xcbhash, 460);
afs_DequeueCallback(avc); /* remove it from queued callbacks list */
avc->f.states &= ~(CStatd | CUnique);
else
osi_dnlc_purgevp(avc);
+ /* By this point, the vcache has been removed from all global structures
+ * via which someone could try to use the vcache. It is okay to drop
+ * afs_xvcache at this point (if *slept is set). */
+
+ if (afs_shuttingdown == AFS_RUNNING)
+ afs_QueueVCB(avc, slept);
+
/*
* Next, keep track of which vnodes we've deleted for create's
* optimistic synchronization algorithm
int tcount;
struct server *tsp;
int i;
- struct vrequest treq;
+ struct vrequest *treq = NULL;
struct afs_conn *tc;
int safety1, safety2, safety3;
XSTATS_DECLS;
if (AFS_IS_DISCONNECTED)
return ENETDOWN;
- if ((code = afs_InitReq(&treq, afs_osi_credp)))
+ if ((code = afs_CreateReq(&treq, afs_osi_credp)))
return code;
- treq.flags |= O_NONBLOCK;
+ treq->flags |= O_NONBLOCK;
tfids = afs_osi_Alloc(sizeof(struct AFSFid) * AFS_MAXCBRSCALL);
osi_Assert(tfids != NULL);
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,
+ tsp->cell->cellNum, treq, 0,
SHARED_LOCK, 0, &rxconn);
if (tc) {
XSTATS_START_TIME
} else
code = -1;
if (!afs_Analyze
- (tc, rxconn, code, 0, &treq,
+ (tc, rxconn, code, 0, treq,
AFS_STATS_FS_RPCIDX_GIVEUPCALLBACKS, SHARED_LOCK,
tsp->cell)) {
break;
if (lockit)
ReleaseWriteLock(&afs_xvcb);
afs_osi_Free(tfids, sizeof(struct AFSFid) * AFS_MAXCBRSCALL);
+ afs_DestroyReq(treq);
return 0;
}
/*
* The proper value for mvstat (for root fids) is setup by the caller.
*/
- avc->mvstat = 0;
+ avc->mvstat = AFS_MVSTAT_FILE;
if (afid->Fid.Vnode == 1 && afid->Fid.Unique == 1)
- avc->mvstat = 2;
+ avc->mvstat = AFS_MVSTAT_ROOT;
if (afs_globalVFS == 0)
osi_Panic("afs globalvfs");
afs_ShakeLooseVCaches(afs_int32 anumber)
{
afs_int32 i, loop;
+ int evicted;
struct vcache *tvc;
struct afs_q *tq, *uq;
int fv_slept, defersleep = 0;
+ int limit;
afs_int32 target = anumber;
- i = 0;
loop = 0;
+
+ retry:
+ i = 0;
+ limit = afs_vcount;
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 (i++ > limit) {
+ afs_warn("afs_ShakeLooseVCaches: i %d limit %d afs_vcount %d afs_maxvcount %d\n",
+ (int)i, limit, (int)afs_vcount, (int)afs_maxvcount);
+ refpanic("Found too many AFS vnodes on VLRU (VLRU cycle?)");
} else if (QNext(uq) != tq) {
refpanic("VLRU inconsistent");
} else if (tvc->f.states & CVInit) {
}
fv_slept = 0;
- if (osi_TryEvictVCache(tvc, &fv_slept, defersleep))
+ evicted = osi_TryEvictVCache(tvc, &fv_slept, defersleep);
+ if (evicted) {
anumber--;
+ }
if (fv_slept) {
if (loop++ > 100)
break;
- uq = VLRU.prev;
- i = 0;
- continue; /* start over - may have raced. */
+ if (!evicted) {
+ /*
+ * This vcache was busy and we slept while trying to evict it.
+ * Move this busy vcache to the head of the VLRU so vcaches
+ * following this busy vcache can be evicted during the retry.
+ */
+ QRemove(&tvc->vlruq);
+ QAdd(&VLRU, &tvc->vlruq);
+ }
+ goto retry; /* start over - may have raced. */
}
- if (tq == uq) {
+ if (uq == &VLRU) {
if (anumber && !defersleep) {
defersleep = 1;
- tq = VLRU.prev;
- continue;
+ goto retry;
}
break;
}
AFS_RWLOCK_INIT(&avc->lock, "vcache lock");
- avc->mvid = NULL;
+ memset(&avc->mvid, 0, sizeof(avc->mvid));
avc->linkData = NULL;
avc->cbExpires = 0;
avc->opens = 0;
if (slept) {
goto retry;
}
- tvc = nvc;
}
}
struct afs_conn *tc;
afs_int32 code;
afs_ucred_t *cred = NULL;
- struct vrequest treq, ureq;
+ struct vrequest *treq = NULL;
struct AFSVolSync tsync;
int didCore;
XSTATS_DECLS;
AFS_STATCNT(afs_FlushActiveVcaches);
+
+ code = afs_CreateReq(&treq, afs_osi_credp);
+ if (code) {
+ afs_warn("unable to alloc treq\n");
+ return;
+ }
+
ObtainReadLock(&afs_xvcache);
for (i = 0; i < VCSIZE; i++) {
for (tvc = afs_vhashT[i]; tvc; tvc = tvc->hnext) {
ReleaseReadLock(&afs_xvcache);
ObtainWriteLock(&tvc->lock, 51);
do {
- afs_InitReq(&treq, afs_osi_credp);
- treq.flags |= O_NONBLOCK;
+ code = afs_InitReq(treq, afs_osi_credp);
+ if (code) {
+ code = -1;
+ break; /* shutting down: do not try to extend the lock */
+ }
+ treq->flags |= O_NONBLOCK;
- tc = afs_Conn(&tvc->f.fid, &treq, SHARED_LOCK, &rxconn);
+ tc = afs_Conn(&tvc->f.fid, treq, SHARED_LOCK, &rxconn);
if (tc) {
XSTATS_START_TIME(AFS_STATS_FS_RPCIDX_EXTENDLOCK);
RX_AFS_GUNLOCK();
} else
code = -1;
} while (afs_Analyze
- (tc, rxconn, code, &tvc->f.fid, &treq,
+ (tc, rxconn, code, &tvc->f.fid, treq,
AFS_STATS_FS_RPCIDX_EXTENDLOCK, SHARED_LOCK, NULL));
ReleaseWriteLock(&tvc->lock);
*/
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
/* XXXX Find better place-holder for cred XXXX */
cred = (afs_ucred_t *)tvc->linkData;
tvc->linkData = NULL; /* XXX */
- afs_InitReq(&ureq, cred);
+ code = afs_InitReq(treq, cred);
afs_Trace2(afs_iclSetp, CM_TRACE_ACTCCORE,
ICL_TYPE_POINTER, tvc, ICL_TYPE_INT32,
tvc->execsOrWriters);
- code = afs_StoreOnLastReference(tvc, &ureq);
+ if (!code) { /* avoid store when shutting down */
+ code = afs_StoreOnLastReference(tvc, treq);
+ }
ReleaseWriteLock(&tvc->lock);
-#ifdef AFS_BOZONLOCK_ENV
- afs_BozonUnlock(&tvc->pvnLock, tvc);
-#endif
hzero(tvc->flushDV);
osi_FlushText(tvc);
didCore = 1;
* 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
} 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);
}
}
ReleaseReadLock(&afs_xvcache);
+ afs_DestroyReq(treq);
}
/* fetch the status info */
tvc = afs_GetVCache(&avc->f.fid, areq, NULL, avc);
if (!tvc)
- return ENOENT;
+ return EIO;
/* Put it back; caller has already incremented vrefCount */
afs_PutVCache(tvc);
return 0;
} else if (vType(avc) == VLNK) {
avc->f.m.Mode |= S_IFLNK;
if ((avc->f.m.Mode & 0111) == 0)
- avc->mvstat = 1;
+ avc->mvstat = AFS_MVSTAT_MTPT;
}
if (avc->f.states & CForeign) {
struct axscache *ac;
} else if (vType(avc) == VLNK) {
avc->f.m.Mode |= S_IFLNK;
if ((avc->f.m.Mode & 0111) == 0)
- avc->mvstat = 1;
+ avc->mvstat = AFS_MVSTAT_MTPT;
}
#endif
flags |= VDisconSetMode;
struct AFSFetchStatus *astat, struct vrequest *areq)
{
afs_size_t length;
-#ifdef AFS_DARWIN80_ENV
- int fixup = 0;
-#endif
AFS_STATCNT(afs_ProcessFS);
#ifdef AFS_64BIT_CLIENT
avc->f.m.Group = astat->Group;
avc->f.m.LinkCount = astat->LinkCount;
if (astat->FileType == File) {
-#ifdef AFS_DARWIN80_ENV
- if (avc->f.m.Type != VREG)
- fixup = 1;
-#endif
vSetType(avc, VREG);
avc->f.m.Mode |= S_IFREG;
} else if (astat->FileType == Directory) {
-#ifdef AFS_DARWIN80_ENV
- if (avc->f.m.Type != VDIR)
- fixup = 1;
-#endif
vSetType(avc, VDIR);
avc->f.m.Mode |= S_IFDIR;
} else if (astat->FileType == SymbolicLink) {
if (afs_fakestat_enable && (avc->f.m.Mode & 0111) == 0) {
-#ifdef AFS_DARWIN80_ENV
- if (avc->f.m.Type != VDIR)
- fixup = 1;
-#endif
vSetType(avc, VDIR);
avc->f.m.Mode |= S_IFDIR;
} else {
-#ifdef AFS_DARWIN80_ENV
- if (avc->f.m.Type != VLNK)
- fixup = 1;
-#endif
vSetType(avc, VLNK);
avc->f.m.Mode |= S_IFLNK;
}
if ((avc->f.m.Mode & 0111) == 0) {
- avc->mvstat = 1;
+ avc->mvstat = AFS_MVSTAT_MTPT;
}
}
-#ifdef AFS_DARWIN80_ENV
- if (fixup)
- printf("found mistyped vnode!\n");
-#endif
avc->f.anyAccess = astat->AnonymousAccess;
#ifdef badidea
if ((astat->CallerAccess & ~astat->AnonymousAccess))
tvc->f.states |= CForeign;
if (newvcache && (tvp->rootVnode == afid->Fid.Vnode)
&& (tvp->rootUnique == afid->Fid.Unique)) {
- tvc->mvstat = 2;
+ tvc->mvstat = AFS_MVSTAT_ROOT;
}
}
if (tvp->states & VRO)
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 *)
+ if (tvc->mvstat == AFS_MVSTAT_ROOT && tvp->dotdot.Fid.Volume != 0) {
+ if (!tvc->mvid.parent)
+ tvc->mvid.parent = (struct VenusFid *)
osi_AllocSmallSpace(sizeof(struct VenusFid));
- *tvc->mvid = tvp->dotdot;
+ *tvc->mvid.parent = tvp->dotdot;
}
afs_PutVolume(tvp, READ_LOCK);
}
tvc->f.states |= CForeign;
if (newvcache && (tvp->rootVnode == afid->Fid.Vnode)
&& (tvp->rootUnique == afid->Fid.Unique))
- tvc->mvstat = 2;
+ tvc->mvstat = AFS_MVSTAT_ROOT;
}
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 *)
+ if (tvc->mvstat == AFS_MVSTAT_ROOT && tvp->dotdot.Fid.Volume != 0) {
+ if (!tvc->mvid.parent)
+ tvc->mvid.parent = (struct VenusFid *)
osi_AllocSmallSpace(sizeof(struct VenusFid));
- *tvc->mvid = tvp->dotdot;
+ *tvc->mvid.parent = tvp->dotdot;
}
}
/* 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;
+ tvc->mvstat = AFS_MVSTAT_ROOT;
}
- if (tvc->mvstat == 2 && tvolp->dotdot.Fid.Volume != 0) {
- if (!tvc->mvid)
- tvc->mvid = (struct VenusFid *)
+ if (tvc->mvstat == AFS_MVSTAT_ROOT && tvolp->dotdot.Fid.Volume != 0) {
+ if (!tvc->mvid.parent)
+ tvc->mvid.parent = (struct VenusFid *)
osi_AllocSmallSpace(sizeof(struct VenusFid));
- *tvc->mvid = tvolp->dotdot;
+ *tvc->mvid.parent = tvolp->dotdot;
}
/* stat the file */
afs_PutVolume(volp, READ_LOCK);
}
+void
+afs_BadFetchStatus(struct afs_conn *tc)
+{
+ int addr = ntohl(tc->parent->srvr->sa_ip);
+ afs_warn("afs: Invalid AFSFetchStatus from server %u.%u.%u.%u\n",
+ (addr >> 24) & 0xff, (addr >> 16) & 0xff, (addr >> 8) & 0xff,
+ (addr) & 0xff);
+ afs_warn("afs: This suggests the server may be sending bad data that "
+ "can lead to availability issues or data corruption. The "
+ "issue has been avoided for now, but it may not always be "
+ "detectable. Please upgrade the server if possible.\n");
+}
+
/**
* Check if a given AFSFetchStatus structure is sane.
*
!(status->FileType > Invalid && status->FileType <= SymbolicLink) ||
status->ParentVnode == 0 || status->ParentUnique == 0) {
+ afs_warn("afs: FetchStatus ec %u iv %u ft %u pv %u pu %u\n",
+ (unsigned)status->errorCode, (unsigned)status->InterfaceVersion,
+ (unsigned)status->FileType, (unsigned)status->ParentVnode,
+ (unsigned)status->ParentUnique);
+ afs_BadFetchStatus(tc);
+
return VBUSY;
}
return 0;
* 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 *)
+ if (tvc->mvstat == AFS_MVSTAT_ROOT && tvp->dotdot.Fid.Volume != 0) {
+ if (!tvc->mvid.parent)
+ tvc->mvid.parent = (struct VenusFid *)
osi_AllocSmallSpace(sizeof(struct VenusFid));
- *tvc->mvid = tvp->dotdot;
+ *tvc->mvid.parent = tvp->dotdot;
}
}
/* store the stat on the file */
*
* \param avc Pointer to the cache entry to reset
* \param acred
+ * \param skipdnlc skip the dnlc purge for this vnode
*
* \note avc must be write locked on entry
+ *
+ * \note The caller should purge the dnlc when skipdnlc is set.
*/
void
-afs_ResetVCache(struct vcache *avc, afs_ucred_t *acred)
+afs_ResetVCache(struct vcache *avc, afs_ucred_t *acred, afs_int32 skipdnlc)
{
ObtainWriteLock(&afs_xcbhash, 456);
afs_DequeueCallback(avc);
ReleaseWriteLock(&afs_xcbhash);
/* now find the disk cache entries */
afs_TryToSmush(avc, acred, 1);
- osi_dnlc_purgedp(avc);
+ if (!skipdnlc) {
+ osi_dnlc_purgedp(avc);
+ }
if (avc->linkData && !(avc->f.states & CCore)) {
afs_osi_Free(avc->linkData, strlen(avc->linkData) + 1);
avc->linkData = NULL;
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;
+ if (tvc->mvid.target_root) {
+ osi_FreeSmallSpace(tvc->mvid.target_root);
+ tvc->mvid.target_root = NULL;
}
#ifdef AFS_AIX_ENV
aix_gnode_rele(AFSTOV(tvc));
*/
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;
+ if (tvc->mvid.target_root) {
+ osi_FreeSmallSpace(tvc->mvid.target_root);
+ tvc->mvid.target_root = NULL;
}
#ifdef AFS_AIX_ENV
if (tvc->v.v_gnode)