/*
* 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
*/
+#include <afsconfig.h>
#include <afs/param.h>
-#include <sys/types.h>
-#ifdef AFS_NT40_ENV
-#include <winsock2.h>
+
+#include <roken.h>
+
+#include <assert.h>
+
+#include <afs/opr.h>
+#ifdef AFS_PTHREAD_ENV
+# include <opr/lock.h>
#else
-#include <sys/file.h>
-#include <netinet/in.h>
+# include <opr/lockstub.h>
#endif
+
#include <lock.h>
#include <rx/xdr.h>
#include <rx/rx.h>
+#include <afs/afsutil.h>
#define UBIK_INTERNALS
#include "ubik.h"
#include "ubik_int.h"
-int (*ubik_CheckRXSecurityProc)();
-char *ubik_CheckRXSecurityRock;
-/* routines for handling requests remotely-submitted by the sync site. These are
- only write transactions (we don't propagate read trans), and there is at most one
- write transaction extant at any one time.
-*/
+static void printServerInfo(void);
+
+/*! \file
+ * routines for handling requests remotely-submitted by the sync site. These are
+ * only write transactions (we don't propagate read trans), and there is at most one
+ * write transaction extant at any one time.
+ */
struct ubik_trans *ubik_currentTrans = 0;
-ubik_CheckAuth(acall)
-register struct rx_call *acall; {
- register afs_int32 code;
- if (ubik_CheckRXSecurityProc) {
- code = (*ubik_CheckRXSecurityProc)(ubik_CheckRXSecurityRock, acall);
- return code;
- }
- else return 0;
-}
/* the rest of these guys handle remote execution of write
* transactions: this is the code executed on the other servers when a
* sync site is executing a write transaction.
*/
-SDISK_Begin(rxcall, atid)
- register struct rx_call *rxcall;
- struct ubik_tid *atid;
+afs_int32
+SDISK_Begin(struct rx_call *rxcall, struct ubik_tid *atid)
{
- register afs_int32 code;
+ afs_int32 code;
- if (code = ubik_CheckAuth(rxcall)) {
- return code;
+ if ((code = ubik_CheckAuth(rxcall))) {
+ return code;
}
DBHOLD(ubik_dbase);
- urecovery_CheckTid(atid);
- if (ubik_currentTrans) {
- /* If the thread is not waiting for lock - ok to end it */
- if (ubik_currentTrans->locktype != LOCKWAIT) {
- udisk_end(ubik_currentTrans);
- }
- ubik_currentTrans = (struct ubik_trans *) 0;
+ 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) {
/* label this trans with the right trans id */
ubik_currentTrans->tid.epoch = atid->epoch;
ubik_currentTrans->tid.counter = atid->counter;
}
+ out:
DBRELE(ubik_dbase);
- return code;}
+ return code;
+}
-SDISK_Commit(rxcall, atid)
- register struct rx_call *rxcall;
- struct ubik_tid *atid;
+afs_int32
+SDISK_Commit(struct rx_call *rxcall, struct ubik_tid *atid)
{
- register afs_int32 code;
- register struct ubik_dbase *dbase;
-
- if (code = ubik_CheckAuth(rxcall)) {
- return code;
- }
+ afs_int32 code;
+ if ((code = ubik_CheckAuth(rxcall))) {
+ return code;
+ }
+ ObtainWriteLock(&ubik_dbase->cache_lock);
+ DBHOLD(ubik_dbase);
if (!ubik_currentTrans) {
- return USYNC;
+ code = USYNC;
+ goto done;
}
/*
* sanity check to make sure only write trans appear here
*/
if (ubik_currentTrans->type != UBIK_WRITETRANS) {
- return UBADTYPE;
+ code = UBADTYPE;
+ goto done;
}
- dbase = ubik_currentTrans->dbase;
- DBHOLD(dbase);
- urecovery_CheckTid(atid);
+ urecovery_CheckTid(atid, 0);
if (!ubik_currentTrans) {
- DBRELE(dbase);
- return USYNC;
+ code = USYNC;
+ goto done;
}
code = udisk_commit(ubik_currentTrans);
if (code == 0) {
/* sync site should now match */
- ubik_dbVersion = ubik_dbase->version;
+ uvote_set_dbVersion(ubik_dbase->version);
}
- DBRELE(dbase);
+done:
+ DBRELE(ubik_dbase);
+ ReleaseWriteLock(&ubik_dbase->cache_lock);
return code;
}
-SDISK_ReleaseLocks(rxcall, atid)
- register struct rx_call *rxcall;
- struct ubik_tid *atid;
+afs_int32
+SDISK_ReleaseLocks(struct rx_call *rxcall, struct ubik_tid *atid)
{
- register struct ubik_dbase *dbase;
- register afs_int32 code;
+ afs_int32 code;
- if (code = ubik_CheckAuth(rxcall)) {
- return code;
+ if ((code = ubik_CheckAuth(rxcall))) {
+ return code;
}
+ DBHOLD(ubik_dbase);
+
if (!ubik_currentTrans) {
- return USYNC;
+ code = USYNC;
+ goto done;
}
/* sanity check to make sure only write trans appear here */
if (ubik_currentTrans->type != UBIK_WRITETRANS) {
- return UBADTYPE;
+ code = UBADTYPE;
+ goto done;
}
- dbase = ubik_currentTrans->dbase;
- DBHOLD(dbase);
- urecovery_CheckTid(atid);
+ urecovery_CheckTid(atid, 0);
if (!ubik_currentTrans) {
- DBRELE(dbase);
- return USYNC;
+ code = USYNC;
+ goto done;
}
/* If the thread is not waiting for lock - ok to end it */
if (ubik_currentTrans->locktype != LOCKWAIT) {
- udisk_end(ubik_currentTrans);
+ udisk_end(ubik_currentTrans);
}
- ubik_currentTrans = (struct ubik_trans *) 0;
- DBRELE(dbase);
- return 0;
+ ubik_currentTrans = (struct ubik_trans *)0;
+done:
+ DBRELE(ubik_dbase);
+ return code;
}
-SDISK_Abort(rxcall, atid)
- register struct rx_call *rxcall;
- struct ubik_tid *atid;
+afs_int32
+SDISK_Abort(struct rx_call *rxcall, struct ubik_tid *atid)
{
- register afs_int32 code;
- register struct ubik_dbase *dbase;
-
- if (code = ubik_CheckAuth(rxcall)) {
- return code;
- }
+ afs_int32 code;
+ if ((code = ubik_CheckAuth(rxcall))) {
+ return code;
+ }
+ DBHOLD(ubik_dbase);
if (!ubik_currentTrans) {
- return USYNC;
+ code = USYNC;
+ goto done;
}
/* sanity check to make sure only write trans appear here */
if (ubik_currentTrans->type != UBIK_WRITETRANS) {
- return UBADTYPE;
+ code = UBADTYPE;
+ goto done;
}
- dbase = ubik_currentTrans->dbase;
- DBHOLD(dbase);
- urecovery_CheckTid(atid);
+ urecovery_CheckTid(atid, 0);
if (!ubik_currentTrans) {
- DBRELE(dbase);
- return USYNC;
+ code = USYNC;
+ goto done;
}
code = udisk_abort(ubik_currentTrans);
/* If the thread is not waiting for lock - ok to end it */
if (ubik_currentTrans->locktype != LOCKWAIT) {
- udisk_end(ubik_currentTrans);
+ udisk_end(ubik_currentTrans);
}
- ubik_currentTrans = (struct ubik_trans *) 0;
- DBRELE(dbase);
+ ubik_currentTrans = (struct ubik_trans *)0;
+done:
+ DBRELE(ubik_dbase);
return code;
}
-SDISK_Lock(rxcall, atid, afile, apos, alen, atype)
- register struct rx_call *rxcall;
- struct ubik_tid *atid;
- afs_int32 afile, apos, alen, atype; /* apos and alen are not used */
+/* apos and alen are not used */
+afs_int32
+SDISK_Lock(struct rx_call *rxcall, struct ubik_tid *atid,
+ afs_int32 afile, afs_int32 apos, afs_int32 alen, afs_int32 atype)
{
- register afs_int32 code;
- register struct ubik_dbase *dbase;
+ afs_int32 code;
struct ubik_trans *ubik_thisTrans;
- if (code = ubik_CheckAuth(rxcall)) {
- return code;
+ if ((code = ubik_CheckAuth(rxcall))) {
+ return code;
}
+ DBHOLD(ubik_dbase);
if (!ubik_currentTrans) {
- return USYNC;
+ code = USYNC;
+ goto done;
}
/* sanity check to make sure only write trans appear here */
if (ubik_currentTrans->type != UBIK_WRITETRANS) {
- return UBADTYPE;
+ code = UBADTYPE;
+ goto done;
}
if (alen != 1) {
- return UBADLOCK;
+ code = UBADLOCK;
+ goto done;
}
- dbase = ubik_currentTrans->dbase;
- DBHOLD(dbase);
- urecovery_CheckTid(atid);
+ urecovery_CheckTid(atid, 0);
if (!ubik_currentTrans) {
- DBRELE(dbase);
- return USYNC;
+ code = USYNC;
+ goto done;
}
ubik_thisTrans = ubik_currentTrans;
* case, end the transaction here.
*/
if (!code && (ubik_currentTrans != ubik_thisTrans)) {
- udisk_end(ubik_thisTrans);
- code = USYNC;
+ udisk_end(ubik_thisTrans);
+ code = USYNC;
}
-
- DBRELE(dbase);
+done:
+ DBRELE(ubik_dbase);
return code;
}
-/* Write a vector of data */
-SDISK_WriteV(rxcall, atid, io_vector, io_buffer)
- register struct rx_call *rxcall;
- struct ubik_tid *atid;
- iovec_wrt *io_vector;
- iovec_buf *io_buffer;
+/*!
+ * \brief Write a vector of data
+ */
+afs_int32
+SDISK_WriteV(struct rx_call *rxcall, struct ubik_tid *atid,
+ iovec_wrt *io_vector, iovec_buf *io_buffer)
{
afs_int32 code, i, offset;
- struct ubik_dbase *dbase;
struct ubik_iovec *iovec;
- char *iobuf;
+ char *iobuf;
- if (code = ubik_CheckAuth(rxcall)) {
- return code;
+ if ((code = ubik_CheckAuth(rxcall))) {
+ return code;
}
+ DBHOLD(ubik_dbase);
if (!ubik_currentTrans) {
- return USYNC;
+ code = USYNC;
+ goto done;
}
/* sanity check to make sure only write trans appear here */
if (ubik_currentTrans->type != UBIK_WRITETRANS) {
- return UBADTYPE;
+ code = UBADTYPE;
+ goto done;
}
- dbase = ubik_currentTrans->dbase;
- DBHOLD(dbase);
- urecovery_CheckTid(atid);
+ urecovery_CheckTid(atid, 0);
if (!ubik_currentTrans) {
- DBRELE(dbase);
- return USYNC;
+ code = USYNC;
+ goto done;
}
iovec = (struct ubik_iovec *)io_vector->iovec_wrt_val;
iobuf = (char *)io_buffer->iovec_buf_val;
- for (i=0, offset=0; i<io_vector->iovec_wrt_len; i++) {
- /* Sanity check for going off end of buffer */
- if ((offset + iovec[i].length) > io_buffer->iovec_buf_len) {
- code = UINTERNAL;
- } else {
- code = udisk_write(ubik_currentTrans, iovec[i].file, &iobuf[offset],
- iovec[i].position, iovec[i].length);
- }
- if (code) break;
+ for (i = 0, offset = 0; i < io_vector->iovec_wrt_len; i++) {
+ /* Sanity check for going off end of buffer */
+ if ((offset + iovec[i].length) > io_buffer->iovec_buf_len) {
+ code = UINTERNAL;
+ } else {
+ code =
+ udisk_write(ubik_currentTrans, iovec[i].file, &iobuf[offset],
+ iovec[i].position, iovec[i].length);
+ }
+ if (code)
+ break;
- offset += iovec[i].length;
+ offset += iovec[i].length;
}
-
- DBRELE(dbase);
+done:
+ DBRELE(ubik_dbase);
return code;
}
-SDISK_Write(rxcall, atid, afile, apos, adata)
- register struct rx_call *rxcall;
- struct ubik_tid *atid;
- afs_int32 afile, apos;
- register bulkdata *adata;
+afs_int32
+SDISK_Write(struct rx_call *rxcall, struct ubik_tid *atid,
+ afs_int32 afile, afs_int32 apos, bulkdata *adata)
{
- register afs_int32 code;
- register struct ubik_dbase *dbase;
+ afs_int32 code;
- if (code = ubik_CheckAuth(rxcall)) {
- return code;
+ if ((code = ubik_CheckAuth(rxcall))) {
+ return code;
}
+ DBHOLD(ubik_dbase);
if (!ubik_currentTrans) {
- return USYNC;
+ code = USYNC;
+ goto done;
}
/* sanity check to make sure only write trans appear here */
if (ubik_currentTrans->type != UBIK_WRITETRANS) {
- return UBADTYPE;
+ code = UBADTYPE;
+ goto done;
}
- dbase = ubik_currentTrans->dbase;
- DBHOLD(dbase);
- urecovery_CheckTid(atid);
+ urecovery_CheckTid(atid, 0);
if (!ubik_currentTrans) {
- DBRELE(dbase);
- return USYNC;
+ code = USYNC;
+ goto done;
}
- code = udisk_write(ubik_currentTrans, afile, adata->bulkdata_val, apos, adata->bulkdata_len);
- DBRELE(dbase);
+ code =
+ udisk_write(ubik_currentTrans, afile, adata->bulkdata_val, apos,
+ adata->bulkdata_len);
+done:
+ DBRELE(ubik_dbase);
return code;
}
-SDISK_Truncate(rxcall, atid, afile, alen)
- register struct rx_call *rxcall;
- struct ubik_tid *atid;
- afs_int32 afile;
- afs_int32 alen;
+afs_int32
+SDISK_Truncate(struct rx_call *rxcall, struct ubik_tid *atid,
+ afs_int32 afile, afs_int32 alen)
{
- register afs_int32 code;
- register struct ubik_dbase *dbase;
+ afs_int32 code;
- if (code = ubik_CheckAuth(rxcall)) {
- return code;
+ if ((code = ubik_CheckAuth(rxcall))) {
+ return code;
}
- if (!ubik_currentTrans) {
- return USYNC;
+ DBHOLD(ubik_dbase);
+ if (!ubik_currentTrans) {
+ code = USYNC;
+ goto done;
}
/* sanity check to make sure only write trans appear here */
if (ubik_currentTrans->type != UBIK_WRITETRANS) {
- return UBADTYPE;
+ code = UBADTYPE;
+ goto done;
}
- dbase = ubik_currentTrans->dbase;
- DBHOLD(dbase);
- urecovery_CheckTid(atid);
+ urecovery_CheckTid(atid, 0);
if (!ubik_currentTrans) {
- DBRELE(dbase);
- return USYNC;
+ code = USYNC;
+ goto done;
}
code = udisk_truncate(ubik_currentTrans, afile, alen);
- DBRELE(dbase);
+done:
+ DBRELE(ubik_dbase);
return code;
}
-SDISK_GetVersion(rxcall, aversion)
- register struct rx_call *rxcall;
- register struct ubik_version *aversion;
+afs_int32
+SDISK_GetVersion(struct rx_call *rxcall,
+ struct ubik_version *aversion)
{
- register afs_int32 code;
+ afs_int32 code;
- if (code = ubik_CheckAuth(rxcall)) {
- return code;
+ if ((code = ubik_CheckAuth(rxcall))) {
+ return code;
}
/*
* we are not the sync site any more, all write transactions would
* fail with UNOQUORUM anyway.
*/
+ DBHOLD(ubik_dbase);
if (ubeacon_AmSyncSite()) {
- return UDEADLOCK;
+ DBRELE(ubik_dbase);
+ return UDEADLOCK;
}
- DBHOLD(ubik_dbase);
code = (*ubik_dbase->getlabel) (ubik_dbase, 0, aversion);
DBRELE(ubik_dbase);
if (code) {
return 0;
}
-SDISK_GetFile(rxcall, file, version)
- register struct rx_call *rxcall;
- register afs_int32 file;
- struct ubik_version *version;
+afs_int32
+SDISK_GetFile(struct rx_call *rxcall, afs_int32 file,
+ struct ubik_version *version)
{
- register afs_int32 code;
- register struct ubik_dbase *dbase;
- register afs_int32 offset;
+ afs_int32 code;
+ struct ubik_dbase *dbase;
+ afs_int32 offset;
struct ubik_stat ubikstat;
char tbuffer[256];
afs_int32 tlen;
afs_int32 length;
-
- if (code = ubik_CheckAuth(rxcall)) {
- return code;
- }
-/* temporarily disabled because it causes problems for migration tool. Hey, it's just
- * a sanity check, anyway.
- if (ubeacon_AmSyncSite()) {
- return UDEADLOCK;
+
+ if ((code = ubik_CheckAuth(rxcall))) {
+ return code;
}
-*/
dbase = ubik_dbase;
DBHOLD(dbase);
code = (*dbase->stat) (dbase, file, &ubikstat);
}
length = ubikstat.size;
tlen = htonl(length);
- code = rx_Write(rxcall, &tlen, sizeof(afs_int32));
+ code = rx_Write(rxcall, (char *)&tlen, sizeof(afs_int32));
if (code != sizeof(afs_int32)) {
DBRELE(dbase);
+ ViceLog(0, ("Rx-write length error=%d\n", code));
return BULK_ERROR;
}
offset = 0;
while (length > 0) {
tlen = (length > sizeof(tbuffer) ? sizeof(tbuffer) : length);
- code = (*dbase->read)(dbase, file, tbuffer, offset, tlen);
+ code = (*dbase->read) (dbase, file, tbuffer, offset, tlen);
if (code != tlen) {
DBRELE(dbase);
+ ViceLog(0, ("read failed error=%d\n", code));
return UIOERROR;
}
code = rx_Write(rxcall, tbuffer, tlen);
if (code != tlen) {
DBRELE(dbase);
+ ViceLog(0, ("Rx-write length error=%d\n", code));
return BULK_ERROR;
}
length -= tlen;
offset += tlen;
}
- code = (*dbase->getlabel)(dbase, file, version); /* return the dbase, too */
+ code = (*dbase->getlabel) (dbase, file, version); /* return the dbase, too */
DBRELE(dbase);
return code;
}
-SDISK_SendFile(rxcall, file, length, avers)
- register struct rx_call *rxcall;
- afs_int32 file;
- afs_int32 length;
- struct ubik_version *avers;
+afs_int32
+SDISK_SendFile(struct rx_call *rxcall, afs_int32 file,
+ afs_int32 length, struct ubik_version *avers)
{
- register afs_int32 code;
- register struct ubik_dbase *dbase;
- char tbuffer[256];
- afs_int32 offset, t;
+ afs_int32 code;
+ struct ubik_dbase *dbase = NULL;
+ char tbuffer[1024];
+ afs_int32 offset;
struct ubik_version tversion;
- register int tlen;
+ int tlen;
struct rx_peer *tpeer;
struct rx_connection *tconn;
- afs_uint32 otherHost;
+ 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 */
-
- if (code = ubik_CheckAuth(rxcall)) {
- goto failed;
+
+ dbase = ubik_dbase;
+ pbuffer[0] = '\0';
+
+ if ((code = ubik_CheckAuth(rxcall))) {
+ return code;
}
/* next, we do a sanity check to see if the guy sending us the database is
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;
- goto failed;
+ if (offset && offset != otherHost) {
+ /* we *know* this is the wrong guy */
+ 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(offset, sync_hoststr)));
+ return USYNC;
}
- dbase = ubik_dbase;
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(otherHost));
+ ViceLog(0, ("Ubik: Synchronize database via DISK_SendFile from server %s\n",
+ afs_inet_ntoa_r(otherHost, hoststr)));
offset = 0;
- (*dbase->truncate) (dbase, file, 0); /* truncate first */
- tversion.epoch = 0; /* start off by labelling in-transit db as invalid */
- tversion.counter = 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) {
- DBRELE(dbase);
+ ViceLog(0, ("Rx-read length error=%d\n", code));
code = BULK_ERROR;
+ close(fd);
goto failed;
}
- code = (*dbase->write)(dbase, file, tbuffer, offset, tlen);
+ code = write(fd, tbuffer, tlen);
+ pass++;
if (code != tlen) {
- DBRELE(dbase);
+ ViceLog(0, ("write failed tlen=%d, error=%d\n", tlen, 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. */
- (*ubik_dbase->sync)(dbase, file);
- code = (*ubik_dbase->setlabel)(dbase, file, avers);
- bcopy(avers, &ubik_dbase->version, sizeof(struct ubik_version));
- udisk_Invalidate(dbase, file); /* new dbase, flush disk buffers */
+ * 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
+ opr_Assert(pthread_cond_broadcast(&dbase->version_cond) == 0);
+#else
LWP_NoYieldSignal(&dbase->version);
- DBRELE(dbase);
+#endif
+
+failed_locked:
+ UBIK_VERSION_UNLOCK;
+
failed:
if (code) {
- ubik_print("Ubik: Synchronize database with server %s failed (error = %d)\n",
- afs_inet_ntoa(otherHost), code);
+ if (pbuffer[0] != '\0')
+ 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;
+ }
+ ViceLog(0, ("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");
+ uvote_set_dbVersion(*avers);
+ ViceLog(0, ("Ubik: Synchronize database completed\n"));
}
+ DBRELE(dbase);
return code;
}
-SDISK_Probe(rxcall)
- register struct rx_call *rxcall;
+afs_int32
+SDISK_Probe(struct rx_call *rxcall)
{
return 0;
}
-/*
-* Update remote machines addresses in my server list
-* Send back my addresses to caller of this RPC
-* Returns zero on success, else 1.
-*/
-SDISK_UpdateInterfaceAddr(rxcall, inAddr, outAddr)
-register struct rx_call *rxcall;
-UbikInterfaceAddr *inAddr, *outAddr;
+/*!
+ * \brief Update remote machines addresses in my server list
+ *
+ * Send back my addresses to caller of this RPC
+ * \return zero on success, else 1.
+ */
+afs_int32
+SDISK_UpdateInterfaceAddr(struct rx_call *rxcall,
+ UbikInterfaceAddr *inAddr,
+ UbikInterfaceAddr *outAddr)
{
struct ubik_server *ts, *tmp;
- afs_uint32 remoteAddr; /* in net byte order */
- int i, j, found=0, probableMatch=0;
+ afs_uint32 remoteAddr; /* in net byte order */
+ int i, j, found = 0, probableMatch = 0;
+ char hoststr[16];
+ UBIK_ADDR_LOCK;
/* copy the output parameters */
- for ( i=0; i < UBIK_MAX_INTERFACE_ADDR; i++)
+ for (i = 0; i < UBIK_MAX_INTERFACE_ADDR; i++)
outAddr->hostAddr[i] = ntohl(ubik_host[i]);
remoteAddr = htonl(inAddr->hostAddr[0]);
- for(ts = ubik_servers; ts; ts=ts->next)
- if ( ts->addr[0] == remoteAddr ) /* both in net byte order */
- {
+ for (ts = ubik_servers; ts; ts = ts->next)
+ if (ts->addr[0] == remoteAddr) { /* both in net byte order */
probableMatch = 1;
break;
}
- if ( probableMatch )
- {
+ if (probableMatch) {
/* verify that all addresses in the incoming RPC are
- ** not part of other server entries in my CellServDB
- */
- for ( i=0; !found && (i<UBIK_MAX_INTERFACE_ADDR)
- && inAddr->hostAddr[i]; i++)
- {
+ ** not part of other server entries in my CellServDB
+ */
+ for (i = 0; !found && (i < UBIK_MAX_INTERFACE_ADDR)
+ && inAddr->hostAddr[i]; i++) {
remoteAddr = htonl(inAddr->hostAddr[i]);
- for(tmp = ubik_servers; (!found && tmp); tmp=tmp->next)
- {
- if ( ts == tmp ) /* this is my server */
- continue;
- for ( j=0; (j<UBIK_MAX_INTERFACE_ADDR) && tmp->addr[j]; j++)
- if ( remoteAddr == tmp->addr[j] )
- {
+ for (tmp = ubik_servers; (!found && tmp); tmp = tmp->next) {
+ if (ts == tmp) /* this is my server */
+ continue;
+ for (j = 0; (j < UBIK_MAX_INTERFACE_ADDR) && tmp->addr[j];
+ j++)
+ if (remoteAddr == tmp->addr[j]) {
found = 1;
break;
}
}
}
- } /* if (probableMatch) */
-
+ }
+
+ /* if (probableMatch) */
/* inconsistent addresses in CellServDB */
- if ( !probableMatch || found )
- {
- ubik_print("Inconsistent Cell Info from server: ");
- for ( i=0; i < UBIK_MAX_INTERFACE_ADDR && inAddr->hostAddr[i]; i++)
- ubik_print("%s ", afs_inet_ntoa(htonl(inAddr->hostAddr[i])));
- ubik_print("\n");
+ if (!probableMatch || found) {
+ ViceLog(0, ("Inconsistent Cell Info from server:\n"));
+ for (i = 0; i < UBIK_MAX_INTERFACE_ADDR && inAddr->hostAddr[i]; i++)
+ ViceLog(0, ("... %s\n", afs_inet_ntoa_r(htonl(inAddr->hostAddr[i]), hoststr)));
+ fflush(stdout);
+ fflush(stderr);
printServerInfo();
- return UBADHOST;
+ UBIK_ADDR_UNLOCK;
+ return UBADHOST;
}
/* update our data structures */
- for ( i=1; i < UBIK_MAX_INTERFACE_ADDR; i++)
- ts->addr[i] = htonl(inAddr->hostAddr[i]);
-
- ubik_print("ubik: A Remote Server has addresses: ");
- for ( i=0; i < UBIK_MAX_INTERFACE_ADDR && ts->addr[i]; i++)
- ubik_print("%s ", afs_inet_ntoa(ts->addr[i]));
- ubik_print("\n");
+ for (i = 1; i < UBIK_MAX_INTERFACE_ADDR; i++)
+ ts->addr[i] = htonl(inAddr->hostAddr[i]);
+
+ ViceLog(0, ("ubik: A Remote Server has addresses:\n"));
+ for (i = 0; i < UBIK_MAX_INTERFACE_ADDR && ts->addr[i]; i++)
+ 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;
}
-printServerInfo()
+static void
+printServerInfo(void)
{
struct ubik_server *ts;
- int i,j=1;
+ int i, j = 1;
+ char hoststr[16];
- ubik_print("Local CellServDB:");
- for ( ts=ubik_servers; ts; ts= ts->next, j++)
- {
- ubik_print("Server %d: ", j);
- for ( i=0; (i<UBIK_MAX_INTERFACE_ADDR) && ts->addr[i]; i++)
- ubik_print("%s ", afs_inet_ntoa(ts->addr[i]));
+ ViceLog(0, ("Local CellServDB:\n"));
+ for (ts = ubik_servers; ts; ts = ts->next, j++) {
+ ViceLog(0, (" Server %d:\n", j));
+ for (i = 0; (i < UBIK_MAX_INTERFACE_ADDR) && ts->addr[i]; i++)
+ ViceLog(0, (" ... %s\n", afs_inet_ntoa_r(ts->addr[i], hoststr)));
}
- ubik_print("\n");
}
-SDISK_SetVersion(rxcall, atid, oldversionp, newversionp)
- struct rx_call *rxcall;
- struct ubik_tid *atid;
- struct ubik_version *oldversionp;
- struct ubik_version *newversionp;
+afs_int32
+SDISK_SetVersion(struct rx_call *rxcall, struct ubik_tid *atid,
+ struct ubik_version *oldversionp,
+ struct ubik_version *newversionp)
{
- afs_int32 code=0;
- struct ubik_dbase *dbase;
-
- if (code = ubik_CheckAuth(rxcall)) {
- return(code);
- }
-
- if (!ubik_currentTrans) {
- return USYNC;
- }
- /* sanity check to make sure only write trans appear here */
- if (ubik_currentTrans->type != UBIK_WRITETRANS) {
- return UBADTYPE;
- }
-
- /* Should not get this for the sync site */
- if (ubeacon_AmSyncSite()) {
- return UDEADLOCK;
- }
-
- dbase = ubik_currentTrans->dbase;
- DBHOLD(dbase);
- urecovery_CheckTid(atid);
- if (!ubik_currentTrans) {
- DBRELE(dbase);
- return USYNC;
- }
-
- /* Set the label if its version matches the sync-site's */
- if ((oldversionp->epoch == ubik_dbVersion.epoch) &&
- (oldversionp->counter == ubik_dbVersion.counter)) {
- code = (*dbase->setlabel) (ubik_dbase, 0, newversionp);
- if (!code) {
- ubik_dbase->version = *newversionp;
- ubik_dbVersion = *newversionp;
- }
- } else {
- code = USYNC;
- }
-
- DBRELE(dbase);
- return code;
+ afs_int32 code = 0;
+
+ if ((code = ubik_CheckAuth(rxcall))) {
+ return (code);
+ }
+ DBHOLD(ubik_dbase);
+ if (!ubik_currentTrans) {
+ code = USYNC;
+ goto done;
+ }
+ /* sanity check to make sure only write trans appear here */
+ if (ubik_currentTrans->type != UBIK_WRITETRANS) {
+ code = UBADTYPE;
+ goto done;
+ }
+
+ /* Should not get this for the sync site */
+ if (ubeacon_AmSyncSite()) {
+ code = UDEADLOCK;
+ goto done;
+ }
+
+ urecovery_CheckTid(atid, 0);
+ if (!ubik_currentTrans) {
+ code = USYNC;
+ goto done;
+ }
+
+ /* 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) {
+ ubik_dbase->version = *newversionp;
+ uvote_set_dbVersion(*newversionp);
+ }
+ UBIK_VERSION_UNLOCK;
+ } else {
+ code = USYNC;
+ }
+done:
+ DBRELE(ubik_dbase);
+ return code;
}