#include <assert.h>
#include <afs/opr.h>
+#ifdef AFS_PTHREAD_ENV
+# include <opr/lock.h>
+#else
+# include <opr/lockstub.h>
+#endif
+
#include <lock.h>
#include <rx/xdr.h>
#include <rx/rx.h>
return code;
}
DBHOLD(ubik_dbase);
+ if (urecovery_AllBetter(ubik_dbase, 0) == 0) {
+ code = UNOQUORUM;
+ goto out;
+ }
urecovery_CheckTid(atid, 1);
code = udisk_begin(ubik_dbase, UBIK_WRITETRANS, &ubik_currentTrans);
if (!code && ubik_currentTrans) {
ubik_currentTrans->tid.epoch = atid->epoch;
ubik_currentTrans->tid.counter = atid->counter;
}
+ out:
DBRELE(ubik_dbase);
return code;
}
char tbuffer[256];
afs_int32 tlen;
afs_int32 length;
+ struct rx_peer *tpeer;
+ struct rx_connection *tconn;
+ afs_uint32 otherHost = 0;
+ char hoststr[16];
if ((code = ubik_CheckAuth(rxcall))) {
return code;
}
+
+ tconn = rx_ConnectionOf(rxcall);
+ tpeer = rx_PeerOf(tconn);
+ otherHost = ubikGetPrimaryInterfaceAddr(rx_HostOf(tpeer));
+ ViceLog(0, ("Ubik: Synchronize database: send (via GetFile) "
+ "to server %s begin\n",
+ afs_inet_ntoa_r(otherHost, hoststr)));
+
dbase = ubik_dbase;
DBHOLD(dbase);
code = (*dbase->stat) (dbase, file, &ubikstat);
if (code < 0) {
- DBRELE(dbase);
- return code;
+ ViceLog(0, ("database stat() error:%d\n", code));
+ goto failed;
}
length = ubikstat.size;
tlen = htonl(length);
code = rx_Write(rxcall, (char *)&tlen, sizeof(afs_int32));
if (code != sizeof(afs_int32)) {
- DBRELE(dbase);
- ubik_dprint("Rx-write length error=%d\n", code);
- return BULK_ERROR;
+ ViceLog(0, ("Rx-write length error=%d\n", code));
+ code = BULK_ERROR;
+ goto failed;
}
offset = 0;
while (length > 0) {
tlen = (length > sizeof(tbuffer) ? sizeof(tbuffer) : length);
code = (*dbase->read) (dbase, file, tbuffer, offset, tlen);
if (code != tlen) {
- DBRELE(dbase);
- ubik_dprint("read failed error=%d\n", code);
- return UIOERROR;
+ ViceLog(0, ("read failed error=%d\n", code));
+ code = UIOERROR;
+ goto failed;
}
code = rx_Write(rxcall, tbuffer, tlen);
if (code != tlen) {
- DBRELE(dbase);
- ubik_dprint("Rx-write length error=%d\n", code);
- return BULK_ERROR;
+ ViceLog(0, ("Rx-write data error=%d\n", code));
+ code = BULK_ERROR;
+ goto failed;
}
length -= tlen;
offset += tlen;
}
code = (*dbase->getlabel) (dbase, file, version); /* return the dbase, too */
+ if (code)
+ ViceLog(0, ("getlabel error=%d\n", code));
+
+ failed:
DBRELE(dbase);
+ if (code) {
+ ViceLog(0,
+ ("Ubik: Synchronize database: send (via GetFile) to "
+ "server %s failed (error = %d)\n",
+ afs_inet_ntoa_r(otherHost, hoststr), code));
+ } else {
+ ViceLog(0,
+ ("Ubik: Synchronize database: send (via GetFile) to "
+ "server %s complete, version: %d.%d\n",
+ afs_inet_ntoa_r(otherHost, hoststr), version->epoch, version->counter));
+ }
return code;
}
int tlen;
struct rx_peer *tpeer;
struct rx_connection *tconn;
+ afs_uint32 syncHost = 0;
afs_uint32 otherHost = 0;
char hoststr[16];
char pbuffer[1028];
/* send the file back to the requester */
dbase = ubik_dbase;
+ pbuffer[0] = '\0';
if ((code = ubik_CheckAuth(rxcall))) {
- DBHOLD(dbase);
- goto failed;
+ return code;
}
/* next, we do a sanity check to see if the guy sending us the database is
* 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();
+ syncHost = uvote_GetSyncSite();
tconn = rx_ConnectionOf(rxcall);
tpeer = rx_PeerOf(tconn);
otherHost = ubikGetPrimaryInterfaceAddr(rx_HostOf(tpeer));
- if (offset && offset != otherHost) {
+ if (syncHost && syncHost != otherHost) {
/* we *know* this is the wrong guy */
- code = USYNC;
- DBHOLD(dbase);
- goto failed;
+ char sync_hoststr[16];
+ ViceLog(0,
+ ("Ubik: Refusing synchronization with server %s since it is not the sync-site (%s).\n",
+ afs_inet_ntoa_r(otherHost, hoststr),
+ afs_inet_ntoa_r(syncHost, sync_hoststr)));
+ return USYNC;
}
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));
+ ViceLog(0, ("Ubik: Synchronize database: receive (via SendFile) from server %s begin\n",
+ afs_inet_ntoa_r(otherHost, hoststr)));
offset = 0;
UBIK_VERSION_LOCK;
fd = open(pbuffer, O_CREAT | O_RDWR | O_TRUNC, 0600);
if (fd < 0) {
code = errno;
+ ViceLog(0, ("Open error=%d\n", code));
goto failed_locked;
}
code = lseek(fd, HDRSIZE, 0);
if (code != HDRSIZE) {
+ ViceLog(0, ("lseek error=%d\n", code));
close(fd);
goto failed_locked;
}
#endif
code = rx_Read(rxcall, tbuffer, tlen);
if (code != tlen) {
- ubik_dprint("Rx-read length error=%d\n", code);
+ ViceLog(0, ("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);
+ ViceLog(0, ("write failed tlen=%d, error=%d\n", tlen, code));
code = UIOERROR;
close(fd);
goto failed;
length -= tlen;
}
code = close(fd);
- if (code)
+ if (code) {
+ ViceLog(0, ("close failed error=%d\n", 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. */
#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);
+ if (pbuffer[0] != '\0')
+ unlink(pbuffer);
+
/* Failed to sync. Allow reads again for now. */
if (dbase != NULL) {
UBIK_VERSION_LOCK;
(*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);
+ ViceLog(0,
+ ("Ubik: Synchronize database: receive (via SendFile) from "
+ "server %s failed (error = %d)\n",
+ afs_inet_ntoa_r(otherHost, hoststr), code));
} else {
- ubik_print("Ubik: Synchronize database completed\n");
+ uvote_set_dbVersion(*avers);
+ ViceLog(0,
+ ("Ubik: Synchronize database: receive (via SendFile) from "
+ "server %s complete, version: %d.%d\n",
+ afs_inet_ntoa_r(otherHost, hoststr), avers->epoch, avers->counter));
}
DBRELE(dbase);
return code;
/* if (probableMatch) */
/* inconsistent addresses in CellServDB */
if (!probableMatch || found) {
- ubik_print("Inconsistent Cell Info from server: ");
+ ViceLog(0, ("Inconsistent Cell Info from server:\n"));
for (i = 0; i < UBIK_MAX_INTERFACE_ADDR && inAddr->hostAddr[i]; i++)
- ubik_print("%s ", afs_inet_ntoa_r(htonl(inAddr->hostAddr[i]), hoststr));
- ubik_print("\n");
+ ViceLog(0, ("... %s\n", afs_inet_ntoa_r(htonl(inAddr->hostAddr[i]), hoststr)));
fflush(stdout);
fflush(stderr);
printServerInfo();
for (i = 1; i < UBIK_MAX_INTERFACE_ADDR; i++)
ts->addr[i] = htonl(inAddr->hostAddr[i]);
- ubik_print("ubik: A Remote Server has addresses: ");
+ ViceLog(0, ("ubik: A Remote Server has addresses:\n"));
for (i = 0; i < UBIK_MAX_INTERFACE_ADDR && ts->addr[i]; i++)
- ubik_print("%s ", afs_inet_ntoa_r(ts->addr[i], hoststr));
- ubik_print("\n");
+ ViceLog(0, ("... %s\n", afs_inet_ntoa_r(ts->addr[i], hoststr)));
UBIK_ADDR_UNLOCK;
+
+ /*
+ * The most likely cause of a DISK_UpdateInterfaceAddr RPC
+ * is because the server was restarted. Reset its state
+ * so that no DISK_Begin RPCs will be issued until the
+ * known database version is current.
+ */
+ UBIK_BEACON_LOCK;
+ ts->beaconSinceDown = 0;
+ ts->currentDB = 0;
+ urecovery_LostServer(ts);
+ UBIK_BEACON_UNLOCK;
return 0;
}
int i, j = 1;
char hoststr[16];
- ubik_print("Local CellServDB:");
+ ViceLog(0, ("Local CellServDB:\n"));
for (ts = ubik_servers; ts; ts = ts->next, j++) {
- ubik_print("Server %d: ", j);
+ ViceLog(0, (" Server %d:\n", j));
for (i = 0; (i < UBIK_MAX_INTERFACE_ADDR) && ts->addr[i]; i++)
- ubik_print("%s ", afs_inet_ntoa_r(ts->addr[i], hoststr));
+ ViceLog(0, (" ... %s\n", afs_inet_ntoa_r(ts->addr[i], hoststr)));
}
- ubik_print("\n");
}
afs_int32
goto done;
}
- /* Set the label if its version matches the sync-site's */
- if (uvote_eq_dbVersion(*oldversionp)) {
+ /* Set the label if our version matches the sync-site's. Also set the label
+ * if our on-disk version matches the old version, and our view of the
+ * sync-site's version matches the new version. This suggests that
+ * ubik_dbVersion was updated while the sync-site was setting the new
+ * version, and it already told us via VOTE_Beacon. */
+ if (uvote_eq_dbVersion(*oldversionp)
+ || (uvote_eq_dbVersion(*newversionp)
+ && vcmp(ubik_dbase->version, *oldversionp) == 0)) {
UBIK_VERSION_LOCK;
code = (*ubik_dbase->setlabel) (ubik_dbase, 0, newversionp);
if (!code) {
git://git.openafs.org / openafs.git / blobdiff