static void *
SlowCall(void * rock)
{
struct rx_connection *conn = rock;
u_long ntime;
u_long now;
long temp_rc;
#ifdef AFS_PTHREAD_ENV
pthread_mutex_lock(&slowCallLock);
#endif
slowCallCode = RXKST_PROCESSRUNNING;
#ifdef AFS_PTHREAD_ENV
pthread_cond_signal(&slowCallCV);
#else
LWP_NoYieldSignal(&slowCallCode);
#endif
slowCallCode = RXKST_Slow(conn, 1, &ntime);
if (!slowCallCode) {
now = FT_ApproxTime();
if ((ntime < now - maxSkew) || (ntime > now + maxSkew))
slowCallCode = RXKST_TIMESKEW;
}
temp_rc = slowCallCode;
#ifdef AFS_PTHREAD_ENV
pthread_cond_signal(&slowCallCV);
pthread_mutex_unlock(&slowCallLock);
#else
LWP_NoYieldSignal(&slowCallCode);
#endif
return (void *)(intptr_t)temp_rc;
}
static void
re_probe (ConnCacheEntry *e)
{
Listitem *item;
struct timeval tv;
assert (e->probe != NULL);
gettimeofday (&tv, NULL);
if (e->probe_le) {
listdel (connprobelist, e->probe_le);
e->probe_next = min(tv.tv_sec + (1 << e->ntries), e->probe_next);
} else
e->probe_next = tv.tv_sec + (1 << e->ntries);
if (e->ntries <= MAX_RETRIES)
++e->ntries;
for (item = listhead (connprobelist);
item;
item = listnext (connprobelist, item)) {
ConnCacheEntry *this = (ConnCacheEntry *)listdata (item);
if (e->probe_next < this->probe_next) {
e->probe_le = listaddbefore (connprobelist, item, e);
LWP_NoYieldSignal (connprobelist);
return;
}
}
e->probe_le = listaddtail (connprobelist, e);
LWP_NoYieldSignal (connprobelist);
}
static void
Consumer(void *foo)
{
LWP_NoYieldSignal ((char *)Producer);
while (1) {
LWP_WaitProcess((char *)Consumer);
printf("[consumer] eating %c\n", pcfoo);
LWP_NoYieldSignal ((char *)Producer);
}
}
/* release a lock, giving preference to new writers */
void
Lock_ReleaseW(struct Lock *lock)
{
if (lock->wait_states & EXCL_LOCKS) {
lock->wait_states &= ~EXCL_LOCKS;
LWP_NoYieldSignal(&lock->excl_locked);
}
else {
lock->wait_states &= ~READ_LOCK;
LWP_NoYieldSignal(&lock->readers_reading);
}
}
/*! \brief
* Called at initialization to figure out version of the dbase we really have.
*
* This routine is called after replaying the log; it reads the restored labels.
*/
static int
InitializeDB(register struct ubik_dbase *adbase)
{
register afs_int32 code;
code = (*adbase->getlabel) (adbase, 0, &adbase->version);
if (code) {
/* try setting the label to a new value */
adbase->version.epoch = 1; /* value for newly-initialized db */
adbase->version.counter = 1;
code = (*adbase->setlabel) (adbase, 0, &adbase->version);
if (code) {
/* failed, try to set it back */
adbase->version.epoch = 0;
adbase->version.counter = 0;
(*adbase->setlabel) (adbase, 0, &adbase->version);
}
#ifdef AFS_PTHREAD_ENV
assert(pthread_cond_broadcast(&adbase->version_cond) == 0);
#else
LWP_NoYieldSignal(&adbase->version);
#endif
}
return 0;
}
/*! \brief
* Called at initialization to figure out version of the dbase we really have.
*
* This routine is called after replaying the log; it reads the restored labels.
*/
static int
InitializeDB(struct ubik_dbase *adbase)
{
afs_int32 code;
code = (*adbase->getlabel) (adbase, 0, &adbase->version);
if (code) {
/* try setting the label to a new value */
UBIK_VERSION_LOCK;
adbase->version.epoch = 1; /* value for newly-initialized db */
adbase->version.counter = 1;
code = (*adbase->setlabel) (adbase, 0, &adbase->version);
if (code) {
/* failed, try to set it back */
adbase->version.epoch = 0;
adbase->version.counter = 0;
(*adbase->setlabel) (adbase, 0, &adbase->version);
}
#ifdef AFS_PTHREAD_ENV
CV_BROADCAST(&adbase->version_cond);
#else
LWP_NoYieldSignal(&adbase->version);
#endif
UBIK_VERSION_UNLOCK;
}
return 0;
}
/* client receives a reply and bundles it off to the right sle */
static void HandleCurrentReply(RPC2_PacketBuffer *pb, struct CEntry *ce)
{
struct SL_Entry *sl;
say(1, RPC2_DebugLevel, "HandleCurrentReply()\n");
rpc2_Recvd.Replies++;
/* should this assert ?? XXXX */
if (BogusSl(ce, pb))
return;
pb = ShrinkPacket(pb);
/* convert rpc2 error value to system errno */
pb->Header.ReturnCode = RPC2_R2SError(pb->Header.ReturnCode);
rpc2_UpdateRTT(pb, ce);
rpc2_Recvd.GoodReplies++;
sl = ce->MySl;
sl->data = pb;
SetState(ce, C_THINK);
rpc2_IncrementSeqNumber(ce);
rpc2_DeactivateSle(sl, ARRIVED);
LWP_NoYieldSignal((char *)sl);
}
/* Special packet from socketlistener: never encrypted */
static void HandleSLPacket(RPC2_PacketBuffer *pb, struct CEntry *ce)
{
rpc2_ntohp(pb);
if (pb->Header.Opcode != RPC2_NAKED) {
BOGUS(pb, "HandleSLPacket: bogus opcode\n");
return;
}
if (!TestState(ce, CLIENT, (C_AWAITREPLY | C_AWAITINIT2))) {
BOGUS(pb, "HandleSLPacket: state != AWAIT\n");
return;
}
say(1, RPC2_DebugLevel, "HandleNak()\n");
rpc2_Recvd.Naks++;
if (BogusSl(ce, pb))
return;
rpc2_SetConnError(ce);
rpc2_DeactivateSle(ce->MySl, NAKED);
LWP_NoYieldSignal((char *)ce->MySl);
RPC2_FreeBuffer(&pb);
}
void rpc2_ExpireEvents()
{
int i;
struct SL_Entry *sl;
struct TM_Elem *t;
for (i = TM_Rescan(rpc2_TimerQueue); i > 0; i--) {
t = TM_GetExpired(rpc2_TimerQueue);
if (!t)
continue;
sl = (struct SL_Entry *)t->BackPointer;
rpc2_DeactivateSle(sl, TIMEOUT);
if (sl->Type == REPLY)
FreeHeld(sl);
else if (sl->Type == DELACK)
DelayedAck(sl);
else if (sl->Type == DELAYED_SEND)
rpc2_SendDelayedPacket(sl);
else if (sl->Type == DELAYED_RECV) {
RPC2_PacketBuffer *pb = rpc2_RecvDelayedPacket(sl);
DispatchPacket(pb);
} else
LWP_NoYieldSignal((char *)sl);
}
}
static void TermSignal() {
LogMsg(0,LogLevel,LogFile, "Term signal caught");
lobotomy = mtrue;
/* wake up the BrainSurgeon */
lwp_debug = 1;
LWP_NoYieldSignal(DoLobotomy);
return;
}
/*!
* \brief sync site
*
* routine called when a non-sync site server goes down; restarts recovery
* process to send missing server the new db when it comes back up.
*
* \note This routine should not do anything with variables used by non-sync site servers.
*/
int
urecovery_LostServer(void)
{
#if !defined(AFS_PTHREAD_ENV)
/* No corresponding LWP_WaitProcess found anywhere for this -- klm */
LWP_NoYieldSignal(&urecovery_state);
#endif
return 0;
}
static int
bnode_Check(struct bnode *abnode)
{
if (abnode->flags & BNODE_WAIT) {
abnode->flags &= ~BNODE_WAIT;
LWP_NoYieldSignal(abnode);
}
return 0;
}
/* release a lock, giving preference to new writers */
void
Afs_Lock_ReleaseW(struct Lock *lock)
{
if (lock->wait_states & EXCL_LOCKS) {
lock->wait_states &= ~EXCL_LOCKS;
#ifdef AFS_PTHREAD_ENV
opr_Verify(pthread_cond_broadcast(&lock->write_cv) == 0);
#else /* AFS_PTHREAD_ENV */
LWP_NoYieldSignal(&lock->excl_locked);
#endif /* AFS_PTHREAD_ENV */
} else {
lock->wait_states &= ~READ_LOCK;
#ifdef AFS_PTHREAD_ENV
opr_Verify(pthread_cond_broadcast(&lock->read_cv) == 0);
#else /* AFS_PTHREAD_ENV */
LWP_NoYieldSignal(&lock->readers_reading);
#endif /* AFS_PTHREAD_ENV */
}
}
/*!
* \brief sync site
*
* routine called when a non-sync site server goes down; restarts recovery
* process to send missing server the new db when it comes back up for
* non-sync site servers.
*
* \note This routine should not do anything with variables used by non-sync site servers.
*/
int
urecovery_LostServer(struct ubik_server *ts)
{
ubeacon_ReinitServer(ts);
#if !defined(AFS_PTHREAD_ENV)
/* No corresponding LWP_WaitProcess found anywhere for this -- klm */
LWP_NoYieldSignal(&urecovery_state);
#endif
return 0;
}
static void
Producer(void *foo)
{
while (1) {
LWP_WaitProcess((char *)Producer);
pcfoo++;
if (pcfoo > 'z')
pcfoo = 'a';
printf("[producer] creating %c\n", pcfoo);
LWP_NoYieldSignal ((char *)Consumer);
}
}
/* wake up readers waiting for this lock */
void
Afs_Lock_WakeupR(struct Lock *lock)
{
if (lock->wait_states & READ_LOCK) {
lock->wait_states &= ~READ_LOCK;
#ifdef AFS_PTHREAD_ENV
opr_Verify(pthread_cond_broadcast(&lock->read_cv) == 0);
#else /* AFS_PTHREAD_ENV */
LWP_NoYieldSignal(&lock->readers_reading);
#endif /* AFS_PTHREAD_ENV */
}
}
static struct ropa_ccpair *
add_client (struct ropa_cb *cb, struct ropa_client *c)
{
struct timeval tv;
struct ropa_ccpair cckey, *cc;
assert (cb && c);
cckey.client = c;
cckey.cb = cb;
cc = hashtabsearch (ht_ccpairs, &cckey);
if (cc) {
listdel (lru_ccpair, cc->li);
cc->li = listaddhead (lru_ccpair, cc);
return cc;
}
/* The reverse of these are in break_ccpair */
callback_ref (cb);
client_ref (c);
cc = listdeltail (lru_ccpair);
DIAGNOSTIC_CHECK_CCPAIR(cc);
cc->li = NULL;
if (ccpairs_inuse_p (cc))
break_ccpair (cc, TRUE);
/* XXX do it for real */
gettimeofday(&tv, NULL);
cc->expire = tv.tv_sec + 3600;
heap_insert (heap_ccpairs, cc, &cc->heap);
LWP_NoYieldSignal (heap_ccpairs);
cc->cb_li = listaddtail (cb->ccpairs, cc);
cc->client = c;
cc->cb = cb;
cc->li = listaddhead (lru_ccpair, cc);
hashtabadd (ht_ccpairs, cc);
mlog_log (MDEBROPA, "add_client: added %x to callback %x.%x.%x",
c->addr[0].addr_in, cb->fid.Volume, cb->fid.Vnode, cb->fid.Unique);
return cc;
}
/* Looks like server code */
static void HandleNewRequest(RPC2_PacketBuffer *pb, struct CEntry *ce)
{
struct SL_Entry *sl;
say(1, RPC2_DebugLevel, "HandleNewRequest()\n");
pb = ShrinkPacket(pb);
ce->TimeStampEcho = pb->Header.TimeStamp;
TVTOTS(&pb->Prefix.RecvStamp, ce->RequestTime);
say(15, RPC2_DebugLevel, "handlenewrequest TS %u RQ %u\n",
ce->TimeStampEcho, ce->RequestTime);
rpc2_Recvd.Requests++;
rpc2_Recvd.GoodRequests++;
sl = ce->MySl;
/* Free held packet and SL entry */
if (sl != NULL) {
rpc2_DeactivateSle(sl, 0);
FreeHeld(sl);
}
rpc2_IncrementSeqNumber(ce);
{ /* set up a timer to send a unsolicited ack response (actually an
RCP2_BUSY) within 100ms. */
struct timeval tv;
tv.tv_sec = 0;
tv.tv_usec = RPC2_DELACK_DELAY;
sl = rpc2_AllocSle(DELACK, ce);
rpc2_ActivateSle(sl, &tv);
}
/* Look for a waiting recipient */
sl = FindRecipient(pb);
if (sl != NULL) {
SetState(ce, S_PROCESS);
rpc2_DeactivateSle(sl, ARRIVED);
sl->data = pb;
LWP_NoYieldSignal((char *)sl);
} else {
/* hold for a future RPC2_GetRequest() */
rpc2_HoldPacket(pb);
SetState(ce, S_REQINQUEUE);
}
}
/* Keep alive for current request */
static void HandleBusy(RPC2_PacketBuffer *pb, struct CEntry *ce)
{
struct SL_Entry *sl;
say(1, RPC2_DebugLevel, "HandleBusy(%x)\n", ce->UniqueCID);
rpc2_Recvd.Busies++;
if (BogusSl(ce, pb))
return;
/* update rtt/bw measurements */
rpc2_UpdateRTT(pb, ce);
rpc2_Recvd.GoodBusies++;
sl = ce->MySl;
rpc2_DeactivateSle(sl, KEPTALIVE);
LWP_NoYieldSignal((char *)sl);
RPC2_FreeBuffer(&pb);
}
static void *
DoWorker(void * rock)
{
struct worker *w = rock;
long code;
code = (*w->proc) (w->index, w->rock);
#ifdef AFS_PTHREAD_ENV
pthread_mutex_lock(&workerLock);
#endif
w->exitCode = code;
#ifdef AFS_PTHREAD_ENV
pthread_mutex_unlock(&workerLock);
#endif
#ifdef AFS_PTHREAD_ENV
pthread_cond_signal(&workerCV);
#else
LWP_NoYieldSignal(&workers);
#endif
return (void *)(intptr_t)code;
}
static void HandleInit2(RPC2_PacketBuffer *pb, struct CEntry *ce)
{
struct SL_Entry *sl;
say(1, RPC2_DebugLevel, "HandleInit2()\n");
rpc2_Recvd.Requests++;
if (BogusSl(ce, pb))
return;
pb = ShrinkPacket(pb);
rpc2_UpdateRTT(pb, ce);
sl = ce->MySl;
sl->data = pb;
SetState(ce, C_AWAITINIT4);
rpc2_DeactivateSle(sl, ARRIVED);
LWP_NoYieldSignal((char *)sl);
}
void
doneWriting(afs_int32 error)
{
#ifndef AFS_PTHREAD_ENV
afs_int32 code = 0;
#endif
/* wait for the reader */
ObtainWriteLock(&dumpSyncPtr->ds_lock);
while (dumpSyncPtr->ds_readerStatus != DS_WAITING) {
LogDebug(4, "doneWriting: waiting for Reader\n");
dumpSyncPtr->ds_writerStatus = DS_WAITING;
ReleaseWriteLock(&dumpSyncPtr->ds_lock);
#ifdef AFS_PTHREAD_ENV
MUTEX_ENTER(&dumpSyncPtr->ds_writerStatus_mutex);
CV_WAIT(&dumpSyncPtr->ds_writerStatus_cond, &dumpSyncPtr->ds_writerStatus_mutex);
MUTEX_EXIT(&dumpSyncPtr->ds_writerStatus_mutex);
#else
LWP_WaitProcess(&dumpSyncPtr->ds_writerStatus);
#endif
ObtainWriteLock(&dumpSyncPtr->ds_lock);
}
LogDebug(4, "doneWriting: setting done\n");
/* signal that we are done */
if (error)
dumpSyncPtr->ds_writerStatus = DS_DONE_ERROR;
else
dumpSyncPtr->ds_writerStatus = DS_DONE;
dumpSyncPtr->ds_readerStatus = 0;
#ifdef AFS_PTHREAD_ENV
CV_BROADCAST(&dumpSyncPtr->ds_readerStatus_cond);
#else
code = LWP_NoYieldSignal(&dumpSyncPtr->ds_readerStatus);
if (code)
LogError(code, "doneWriting: Signal delivery failed\n");
#endif
ReleaseWriteLock(&dumpSyncPtr->ds_lock);
}
/* lineTest() : wait until a whole line has been entered before processing.
*/
void
lineTest()
{
char ch;
int rc;
char line[256];
printf("Will print dots until enter a line\n");
/* start the dotwriter thread running */
waitingForAnswer = 1;
LWP_NoYieldSignal(&waitingForAnswer);
rc = LWP_GetLine(line, 256);
waitingForAnswer = 0;
printf("You entered : %s\n", line);
if (rc)
printf("linebuf was too small\n");
return;
}
static void HandleInit3(RPC2_PacketBuffer *pb, struct CEntry *ce)
{
struct SL_Entry *sl;
say(1, RPC2_DebugLevel, "HandleInit3()\n");
rpc2_Recvd.Requests++;
/* Am I expecting this packet? */
if (!TestState(ce, SERVER, S_AWAITINIT3)) {
if (ce->HeldPacket) {
/* My Init4 must have got lost; resend it */
ce->HeldPacket->Header.TimeStamp = htonl(pb->Header.TimeStamp);
rpc2_XmitPacket(ce->HeldPacket, ce->HostInfo->Addr, 1);
} else
say(1, RPC2_DebugLevel, "Bogus Init3\n");
/* Throw packet away anyway */
RPC2_FreeBuffer(&pb);
return;
}
/* Expected Init3 */
if (BogusSl(ce, pb))
return;
pb = ShrinkPacket(pb);
ce->TimeStampEcho = pb->Header.TimeStamp;
TVTOTS(&pb->Prefix.RecvStamp, ce->RequestTime);
say(15, RPC2_DebugLevel, "handleinit3 TS %u RQ %u\n", ce->TimeStampEcho,
ce->RequestTime);
sl = ce->MySl;
sl->data = pb;
SetState(ce, S_FINISHBIND);
rpc2_DeactivateSle(sl, ARRIVED);
LWP_NoYieldSignal((char *)sl);
}
/* interTest() : wait for key press and beep to remind user every five
* seconds.
*/
void
interTest()
{
char ch;
int rc;
printf("Will print dots until you hit a key!\n");
/* start the dotwriter thread running */
waitingForAnswer = 1;
LWP_NoYieldSignal(&waitingForAnswer);
do {
rc = LWP_GetResponseKey(5, &ch);
if (rc == 0)
printf("\a");
} while (rc == 0);
waitingForAnswer = 0; /* turn off dotwriter lwp */
printf("\nYou typed %c\n", ch);
return;
}
static void HandleInit4(RPC2_PacketBuffer *pb, struct CEntry *ce)
{
struct SL_Entry *sl;
say(1, RPC2_DebugLevel, "HandleInit4()\n");
rpc2_Recvd.Requests++;
if (BogusSl(ce, pb))
return;
pb = ShrinkPacket(pb);
rpc2_UpdateRTT(pb, ce);
sl = ce->MySl;
sl->data = pb;
/* really C_THINK, but we don't want this connection to get used
* until we're really sure all setup is complete. */
SetState(ce, C_AWAITREPLY);
rpc2_DeactivateSle(sl, ARRIVED);
LWP_NoYieldSignal((char *)sl);
}
int main(void)
{
long go;
GetRoot(); /* Also creates a child process for transcribing stdout */
GetParms();
MakeFiles(); /* in test directory */
InitRPC();
MakeWorkers();
GetConns();
GetVar(&go, "Say when: ");
DoBindings();
MakeClients();
/* wait for all clients to get ready */
while (ClientsReady < Clients) LWP_DispatchProcess();
LWP_NoYieldSignal((char *)&ClientsReady);
LWP_WaitProcess((char *)main); /* infinite wait */
return 0; /* make compiler happy */
}
void
main(int ac, char **av)
{
int delay = 0;
int iters = 0;
int inter = 0;
int line = 0;
int i;
PROCESS dotpid;
int rc;
for (i = 1; i < ac; i++) {
if (!strcmp("-delay", av[i])) {
if (++i >= ac) {
printf("Missing delay time for -delay option.\n");
}
delay = atoi(av[i]);
if (delay < 0) {
printf("Delay must be at least 0 seconds.\n");
Usage();
}
} else if (!strcmp("-iters", av[i])) {
if (++i >= ac) {
printf("Missing iteration count for -iters option.\n");
}
iters = atoi(av[i]);
if (iters < 0) {
printf("Number of iterations must be at least 0.\n");
Usage();
}
} else if (!strcmp("-nobuf", av[i])) {
rc = setvbuf(stdin, NULL, _IONBF, 0);
if (rc < 0) {
perror("Setting -nobuf for stdin");
}
} else if (!strcmp("-inter", av[i])) {
inter = 1;
} else if (!strcmp("-line", av[i])) {
line = 1;
} else
Usage();
}
IOMGR_Initialize();
LWP_CreateProcess(DotWriter, 32000, LWP_NORMAL_PRIORITY, (char *)0,
"DotWriter", &dotpid);
if (inter) {
interTest();
exit(1);
}
if (line) {
lineTest();
exit(1);
}
if (delay == 0) {
delay = -1; /* Means wait indefinitely. */
}
for (; iters >= 0; iters--) {
waitingForAnswer = 1;
LWP_NoYieldSignal(&waitingForAnswer);
rc = LWP_WaitForKeystroke(delay);
waitingForAnswer = 0;
if (rc) {
printf("\n'%c'\n", getchar());
printf("Flushing remaining input.\n");
while (LWP_WaitForKeystroke(0)) {
printf("'%c'\n", getchar());
}
} else {
printf("\nNo data available on this iteration.\n");
}
}
}
/*!
* \brief Main interaction loop for the recovery manager
*
* The recovery light-weight process only runs when you're the
* synchronization site. It performs the following tasks, if and only
* if the prerequisite tasks have been performed successfully (it
* keeps track of which ones have been performed in its bit map,
* \p urecovery_state).
*
* First, it is responsible for probing that all servers are up. This
* is the only operation that must be performed even if this is not
* yet the sync site, since otherwise this site may not notice that
* enough other machines are running to even elect this guy to be the
* sync site.
*
* After that, the recovery process does nothing until the beacon and
* voting modules manage to get this site elected sync site.
*
* After becoming sync site, recovery first attempts to find the best
* database available in the network (it must do this in order to
* ensure finding the latest committed data). After finding the right
* database, it must fetch this dbase to the sync site.
*
* After fetching the dbase, it relabels it with a new version number,
* to ensure that everyone recognizes this dbase as the most recent
* dbase.
*
* One the dbase has been relabelled, this machine can start handling
* requests. However, the recovery module still has one more task:
* propagating the dbase out to everyone who is up in the network.
*/
void *
urecovery_Interact(void *dummy)
{
afs_int32 code, tcode;
struct ubik_server *bestServer = NULL;
struct ubik_server *ts;
int dbok, doingRPC, now;
afs_int32 lastProbeTime;
/* if we're the sync site, the best db version we've found yet */
static struct ubik_version bestDBVersion;
struct ubik_version tversion;
struct timeval tv;
int length, tlen, offset, file, nbytes;
struct rx_call *rxcall;
char tbuffer[1024];
struct ubik_stat ubikstat;
struct in_addr inAddr;
char hoststr[16];
char pbuffer[1028];
int fd = -1;
afs_int32 pass;
afs_pthread_setname_self("recovery");
/* otherwise, begin interaction */
urecovery_state = 0;
lastProbeTime = 0;
while (1) {
/* Run through this loop every 4 seconds */
tv.tv_sec = 4;
tv.tv_usec = 0;
#ifdef AFS_PTHREAD_ENV
select(0, 0, 0, 0, &tv);
#else
IOMGR_Select(0, 0, 0, 0, &tv);
#endif
ubik_dprint("recovery running in state %x\n", urecovery_state);
/* Every 30 seconds, check all the down servers and mark them
* as up if they respond. When a server comes up or found to
* not be current, then re-find the the best database and
* propogate it.
*/
if ((now = FT_ApproxTime()) > 30 + lastProbeTime) {
for (ts = ubik_servers, doingRPC = 0; ts; ts = ts->next) {
UBIK_BEACON_LOCK;
if (!ts->up) {
UBIK_BEACON_UNLOCK;
doingRPC = 1;
code = DoProbe(ts);
if (code == 0) {
UBIK_BEACON_LOCK;
ts->up = 1;
UBIK_BEACON_UNLOCK;
DBHOLD(ubik_dbase);
urecovery_state &= ~UBIK_RECFOUNDDB;
DBRELE(ubik_dbase);
}
} else {
UBIK_BEACON_UNLOCK;
DBHOLD(ubik_dbase);
if (!ts->currentDB)
urecovery_state &= ~UBIK_RECFOUNDDB;
DBRELE(ubik_dbase);
}
}
if (doingRPC)
now = FT_ApproxTime();
lastProbeTime = now;
}
/* Mark whether we are the sync site */
DBHOLD(ubik_dbase);
if (!ubeacon_AmSyncSite()) {
urecovery_state &= ~UBIK_RECSYNCSITE;
DBRELE(ubik_dbase);
continue; /* nothing to do */
}
urecovery_state |= UBIK_RECSYNCSITE;
/* If a server has just come up or if we have not found the
* most current database, then go find the most current db.
*/
if (!(urecovery_state & UBIK_RECFOUNDDB)) {
DBRELE(ubik_dbase);
bestServer = (struct ubik_server *)0;
bestDBVersion.epoch = 0;
bestDBVersion.counter = 0;
for (ts = ubik_servers; ts; ts = ts->next) {
UBIK_BEACON_LOCK;
if (!ts->up) {
UBIK_BEACON_UNLOCK;
continue; /* don't bother with these guys */
}
UBIK_BEACON_UNLOCK;
if (ts->isClone)
continue;
UBIK_ADDR_LOCK;
code = DISK_GetVersion(ts->disk_rxcid, &ts->version);
UBIK_ADDR_UNLOCK;
if (code == 0) {
/* perhaps this is the best version */
if (vcmp(ts->version, bestDBVersion) > 0) {
/* new best version */
bestDBVersion = ts->version;
bestServer = ts;
}
}
}
/* take into consideration our version. Remember if we,
* the sync site, have the best version. Also note that
* we may need to send the best version out.
*/
DBHOLD(ubik_dbase);
if (vcmp(ubik_dbase->version, bestDBVersion) >= 0) {
bestDBVersion = ubik_dbase->version;
bestServer = (struct ubik_server *)0;
urecovery_state |= UBIK_RECHAVEDB;
} else {
/* Clear the flag only when we know we have to retrieve
* the db. Because urecovery_AllBetter() looks at it.
*/
urecovery_state &= ~UBIK_RECHAVEDB;
}
urecovery_state |= UBIK_RECFOUNDDB;
urecovery_state &= ~UBIK_RECSENTDB;
}
if (!(urecovery_state & UBIK_RECFOUNDDB)) {
DBRELE(ubik_dbase);
continue; /* not ready */
}
/* If we, the sync site, do not have the best db version, then
* go and get it from the server that does.
*/
if ((urecovery_state & UBIK_RECHAVEDB) || !bestServer) {
urecovery_state |= UBIK_RECHAVEDB;
} else {
/* we don't have the best version; we should fetch it. */
urecovery_AbortAll(ubik_dbase);
/* Rx code to do the Bulk fetch */
file = 0;
offset = 0;
UBIK_ADDR_LOCK;
rxcall = rx_NewCall(bestServer->disk_rxcid);
ubik_print("Ubik: Synchronize database with server %s\n",
afs_inet_ntoa_r(bestServer->addr[0], hoststr));
UBIK_ADDR_UNLOCK;
code = StartDISK_GetFile(rxcall, file);
if (code) {
ubik_dprint("StartDiskGetFile failed=%d\n", code);
goto FetchEndCall;
}
nbytes = rx_Read(rxcall, (char *)&length, sizeof(afs_int32));
length = ntohl(length);
if (nbytes != sizeof(afs_int32)) {
ubik_dprint("Rx-read length error=%d\n", code = BULK_ERROR);
code = EIO;
goto FetchEndCall;
}
/* give invalid label during file transit */
UBIK_VERSION_LOCK;
tversion.epoch = 0;
code = (*ubik_dbase->setlabel) (ubik_dbase, file, &tversion);
UBIK_VERSION_UNLOCK;
if (code) {
ubik_dprint("setlabel io error=%d\n", code);
goto FetchEndCall;
}
snprintf(pbuffer, sizeof(pbuffer), "%s.DB%s%d.TMP",
ubik_dbase->pathName, (file<0)?"SYS":"",
(file<0)?-file:file);
fd = open(pbuffer, O_CREAT | O_RDWR | O_TRUNC, 0600);
if (fd < 0) {
code = errno;
goto FetchEndCall;
}
code = lseek(fd, HDRSIZE, 0);
if (code != HDRSIZE) {
close(fd);
goto FetchEndCall;
}
pass = 0;
while (length > 0) {
tlen = (length > sizeof(tbuffer) ? sizeof(tbuffer) : length);
#ifndef AFS_PTHREAD_ENV
if (pass % 4 == 0)
IOMGR_Poll();
#endif
nbytes = rx_Read(rxcall, tbuffer, tlen);
if (nbytes != tlen) {
ubik_dprint("Rx-read bulk error=%d\n", code = BULK_ERROR);
code = EIO;
close(fd);
goto FetchEndCall;
}
nbytes = write(fd, tbuffer, tlen);
pass++;
if (nbytes != tlen) {
code = UIOERROR;
close(fd);
goto FetchEndCall;
}
offset += tlen;
length -= tlen;
}
code = close(fd);
if (code)
goto FetchEndCall;
code = EndDISK_GetFile(rxcall, &tversion);
FetchEndCall:
tcode = rx_EndCall(rxcall, code);
if (!code)
code = tcode;
if (!code) {
/* we got a new file, set up its header */
urecovery_state |= UBIK_RECHAVEDB;
UBIK_VERSION_LOCK;
memcpy(&ubik_dbase->version, &tversion,
sizeof(struct ubik_version));
snprintf(tbuffer, sizeof(tbuffer), "%s.DB%s%d",
ubik_dbase->pathName, (file<0)?"SYS":"",
(file<0)?-file:file);
#ifdef AFS_NT40_ENV
snprintf(pbuffer, sizeof(pbuffer), "%s.DB%s%d.OLD",
ubik_dbase->pathName, (file<0)?"SYS":"",
(file<0)?-file:file);
code = unlink(pbuffer);
if (!code)
code = rename(tbuffer, pbuffer);
snprintf(pbuffer, sizeof(pbuffer), "%s.DB%s%d.TMP",
ubik_dbase->pathName, (file<0)?"SYS":"",
(file<0)?-file:file);
#endif
if (!code)
code = rename(pbuffer, tbuffer);
if (!code) {
(*ubik_dbase->open) (ubik_dbase, file);
/* after data is good, sync disk with correct label */
code =
(*ubik_dbase->setlabel) (ubik_dbase, 0,
&ubik_dbase->version);
}
UBIK_VERSION_UNLOCK;
#ifdef AFS_NT40_ENV
snprintf(pbuffer, sizeof(pbuffer), "%s.DB%s%d.OLD",
ubik_dbase->pathName, (file<0)?"SYS":"",
(file<0)?-file:file);
unlink(pbuffer);
#endif
}
if (code) {
unlink(pbuffer);
/*
* We will effectively invalidate the old data forever now.
* Unclear if we *should* but we do.
*/
UBIK_VERSION_LOCK;
ubik_dbase->version.epoch = 0;
ubik_dbase->version.counter = 0;
UBIK_VERSION_UNLOCK;
ubik_print("Ubik: Synchronize database failed (error = %d)\n",
code);
} else {
ubik_print("Ubik: Synchronize database completed\n");
urecovery_state |= UBIK_RECHAVEDB;
}
udisk_Invalidate(ubik_dbase, 0); /* data has changed */
#ifdef AFS_PTHREAD_ENV
CV_BROADCAST(&ubik_dbase->version_cond);
#else
LWP_NoYieldSignal(&ubik_dbase->version);
#endif
}
if (!(urecovery_state & UBIK_RECHAVEDB)) {
DBRELE(ubik_dbase);
continue; /* not ready */
}
/* If the database was newly initialized, then when we establish quorum, write
* a new label. This allows urecovery_AllBetter() to allow access for reads.
* Setting it to 2 also allows another site to come along with a newer
* database and overwrite this one.
*/
if (ubik_dbase->version.epoch == 1) {
urecovery_AbortAll(ubik_dbase);
UBIK_VERSION_LOCK;
version_globals.ubik_epochTime = 2;
ubik_dbase->version.epoch = version_globals.ubik_epochTime;
ubik_dbase->version.counter = 1;
code =
(*ubik_dbase->setlabel) (ubik_dbase, 0, &ubik_dbase->version);
UBIK_VERSION_UNLOCK;
udisk_Invalidate(ubik_dbase, 0); /* data may have changed */
#ifdef AFS_PTHREAD_ENV
CV_BROADCAST(&ubik_dbase->version_cond);
#else
LWP_NoYieldSignal(&ubik_dbase->version);
#endif
}
/* Check the other sites and send the database to them if they
* do not have the current db.
*/
if (!(urecovery_state & UBIK_RECSENTDB)) {
/* now propagate out new version to everyone else */
dbok = 1; /* start off assuming they all worked */
/*
* Check if a write transaction is in progress. We can't send the
* db when a write is in progress here because the db would be
* obsolete as soon as it goes there. Also, ops after the begin
* trans would reach the recepient and wouldn't find a transaction
* pending there. Frankly, I don't think it's possible to get past
* the write-lock above if there is a write transaction in progress,
* but then, it won't hurt to check, will it?
*/
if (ubik_dbase->flags & DBWRITING) {
struct timeval tv;
int safety = 0;
long cur_usec = 50000;
while ((ubik_dbase->flags & DBWRITING) && (safety < 500)) {
DBRELE(ubik_dbase);
/* sleep for a little while */
tv.tv_sec = 0;
tv.tv_usec = cur_usec;
#ifdef AFS_PTHREAD_ENV
select(0, 0, 0, 0, &tv);
#else
IOMGR_Select(0, 0, 0, 0, &tv);
#endif
cur_usec += 10000;
safety++;
DBHOLD(ubik_dbase);
}
}
for (ts = ubik_servers; ts; ts = ts->next) {
UBIK_ADDR_LOCK;
inAddr.s_addr = ts->addr[0];
UBIK_ADDR_UNLOCK;
UBIK_BEACON_LOCK;
if (!ts->up) {
UBIK_BEACON_UNLOCK;
ubik_dprint("recovery cannot send version to %s\n",
afs_inet_ntoa_r(inAddr.s_addr, hoststr));
dbok = 0;
continue;
}
UBIK_BEACON_UNLOCK;
ubik_dprint("recovery sending version to %s\n",
afs_inet_ntoa_r(inAddr.s_addr, hoststr));
if (vcmp(ts->version, ubik_dbase->version) != 0) {
ubik_dprint("recovery stating local database\n");
/* Rx code to do the Bulk Store */
code = (*ubik_dbase->stat) (ubik_dbase, 0, &ubikstat);
if (!code) {
length = ubikstat.size;
file = offset = 0;
UBIK_ADDR_LOCK;
rxcall = rx_NewCall(ts->disk_rxcid);
UBIK_ADDR_UNLOCK;
code =
StartDISK_SendFile(rxcall, file, length,
&ubik_dbase->version);
if (code) {
ubik_dprint("StartDiskSendFile failed=%d\n",
code);
goto StoreEndCall;
}
while (length > 0) {
tlen =
(length >
sizeof(tbuffer) ? sizeof(tbuffer) : length);
nbytes =
(*ubik_dbase->read) (ubik_dbase, file,
tbuffer, offset, tlen);
if (nbytes != tlen) {
ubik_dprint("Local disk read error=%d\n",
code = UIOERROR);
goto StoreEndCall;
}
nbytes = rx_Write(rxcall, tbuffer, tlen);
if (nbytes != tlen) {
ubik_dprint("Rx-write bulk error=%d\n", code =
BULK_ERROR);
goto StoreEndCall;
}
offset += tlen;
length -= tlen;
}
code = EndDISK_SendFile(rxcall);
StoreEndCall:
code = rx_EndCall(rxcall, code);
}
if (code == 0) {
/* we set a new file, process its header */
ts->version = ubik_dbase->version;
ts->currentDB = 1;
} else
dbok = 0;
} else {
/* mark file up to date */
ts->currentDB = 1;
}
}
if (dbok)
urecovery_state |= UBIK_RECSENTDB;
}
DBRELE(ubik_dbase);
}
return NULL;
}
afs_int32
SDISK_SendFile(struct rx_call *rxcall, afs_int32 file,
afs_int32 length, struct ubik_version *avers)
{
afs_int32 code;
struct ubik_dbase *dbase = NULL;
char tbuffer[1024];
afs_int32 offset;
struct ubik_version tversion;
int tlen;
struct rx_peer *tpeer;
struct rx_connection *tconn;
afs_uint32 otherHost = 0;
char hoststr[16];
char pbuffer[1028];
int fd = -1;
afs_int32 epoch = 0;
afs_int32 pass;
/* send the file back to the requester */
dbase = ubik_dbase;
if ((code = ubik_CheckAuth(rxcall))) {
DBHOLD(dbase);
goto failed;
}
/* next, we do a sanity check to see if the guy sending us the database is
* the guy we think is the sync site. It turns out that we might not have
* decided yet that someone's the sync site, but they could have enough
* votes from others to be sync site anyway, and could send us the database
* in advance of getting our votes. This is fine, what we're really trying
* to check is that some authenticated bogon isn't sending a random database
* into another configuration. This could happen on a bad configuration
* screwup. Thus, we only object if we're sure we know who the sync site
* is, and it ain't the guy talking to us.
*/
offset = uvote_GetSyncSite();
tconn = rx_ConnectionOf(rxcall);
tpeer = rx_PeerOf(tconn);
otherHost = ubikGetPrimaryInterfaceAddr(rx_HostOf(tpeer));
if (offset && offset != otherHost) {
/* we *know* this is the wrong guy */
code = USYNC;
DBHOLD(dbase);
goto failed;
}
DBHOLD(dbase);
/* abort any active trans that may scribble over the database */
urecovery_AbortAll(dbase);
ubik_print("Ubik: Synchronize database with server %s\n",
afs_inet_ntoa_r(otherHost, hoststr));
offset = 0;
UBIK_VERSION_LOCK;
epoch = tversion.epoch = 0; /* start off by labelling in-transit db as invalid */
(*dbase->setlabel) (dbase, file, &tversion); /* setlabel does sync */
snprintf(pbuffer, sizeof(pbuffer), "%s.DB%s%d.TMP",
ubik_dbase->pathName, (file<0)?"SYS":"",
(file<0)?-file:file);
fd = open(pbuffer, O_CREAT | O_RDWR | O_TRUNC, 0600);
if (fd < 0) {
code = errno;
goto failed_locked;
}
code = lseek(fd, HDRSIZE, 0);
if (code != HDRSIZE) {
close(fd);
goto failed_locked;
}
pass = 0;
memcpy(&ubik_dbase->version, &tversion, sizeof(struct ubik_version));
UBIK_VERSION_UNLOCK;
while (length > 0) {
tlen = (length > sizeof(tbuffer) ? sizeof(tbuffer) : length);
#if !defined(AFS_PTHREAD_ENV)
if (pass % 4 == 0)
IOMGR_Poll();
#endif
code = rx_Read(rxcall, tbuffer, tlen);
if (code != tlen) {
ubik_dprint("Rx-read length error=%d\n", code);
code = BULK_ERROR;
close(fd);
goto failed;
}
code = write(fd, tbuffer, tlen);
pass++;
if (code != tlen) {
ubik_dprint("write failed error=%d\n", code);
code = UIOERROR;
close(fd);
goto failed;
}
offset += tlen;
length -= tlen;
}
code = close(fd);
if (code)
goto failed;
/* sync data first, then write label and resync (resync done by setlabel call).
* This way, good label is only on good database. */
snprintf(tbuffer, sizeof(tbuffer), "%s.DB%s%d",
ubik_dbase->pathName, (file<0)?"SYS":"", (file<0)?-file:file);
#ifdef AFS_NT40_ENV
snprintf(pbuffer, sizeof(pbuffer), "%s.DB%s%d.OLD",
ubik_dbase->pathName, (file<0)?"SYS":"", (file<0)?-file:file);
code = unlink(pbuffer);
if (!code)
code = rename(tbuffer, pbuffer);
snprintf(pbuffer, sizeof(pbuffer), "%s.DB%s%d.TMP",
ubik_dbase->pathName, (file<0)?"SYS":"", (file<0)?-file:file);
#endif
if (!code)
code = rename(pbuffer, tbuffer);
UBIK_VERSION_LOCK;
if (!code) {
(*ubik_dbase->open) (ubik_dbase, file);
code = (*ubik_dbase->setlabel) (dbase, file, avers);
}
#ifdef AFS_NT40_ENV
snprintf(pbuffer, sizeof(pbuffer), "%s.DB%s%d.OLD",
ubik_dbase->pathName, (file<0)?"SYS":"", (file<0)?-file:file);
unlink(pbuffer);
#endif
memcpy(&ubik_dbase->version, avers, sizeof(struct ubik_version));
udisk_Invalidate(dbase, file); /* new dbase, flush disk buffers */
#ifdef AFS_PTHREAD_ENV
assert(pthread_cond_broadcast(&dbase->version_cond) == 0);
#else
LWP_NoYieldSignal(&dbase->version);
#endif
failed_locked:
UBIK_VERSION_UNLOCK;
failed:
if (code) {
unlink(pbuffer);
/* Failed to sync. Allow reads again for now. */
if (dbase != NULL) {
UBIK_VERSION_LOCK;
tversion.epoch = epoch;
(*dbase->setlabel) (dbase, file, &tversion);
UBIK_VERSION_UNLOCK;
}
ubik_print
("Ubik: Synchronize database with server %s failed (error = %d)\n",
afs_inet_ntoa_r(otherHost, hoststr), code);
} else {
ubik_print("Ubik: Synchronize database completed\n");
}
DBRELE(dbase);
return code;
}