* directory or online at http://www.openafs.org/dl/license10.html
*/
+#include <afsconfig.h>
#include <afs/param.h>
+#include <roken.h>
+
#include <afs/stds.h>
#include <windows.h>
afs_uint32 cryptall = 0;
afs_uint32 cm_anonvldb = 0;
+afs_uint32 rx_pmtu_discovery = 0;
void cm_PutConn(cm_conn_t *connp)
{
long code;
char addr[16]="unknown";
int forcing_new = 0;
+ int location_updated = 0;
+ char *format;
+ DWORD msgID;
osi_Log2(afsd_logp, "cm_Analyze connp 0x%p, code 0x%x",
connp, errorCode);
}
else if (errorCode == CM_ERROR_ALLDOWN) {
- osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_ALLDOWN.");
/* Servers marked DOWN will be restored by the background daemon
* thread as they become available. The volume status is
* updated as the server state changes.
*/
+ if (fidp) {
+ osi_Log2(afsd_logp, "cm_Analyze passed CM_ERROR_DOWN (FS cell %s vol 0x%x)",
+ cellp->name, fidp->volume);
+ msgID = MSG_ALL_SERVERS_DOWN;
+ format = "All servers are unreachable when accessing cell %s volume %d.";
+ LogEvent(EVENTLOG_WARNING_TYPE, msgID, cellp->name, fidp->volume);
+ } else {
+ osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_ALLDOWN (VL Server)");
+ }
}
else if (errorCode == CM_ERROR_ALLOFFLINE) {
- osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_ALLOFFLINE.");
/* Volume instances marked offline will be restored by the
* background daemon thread as they become available
*/
if (fidp) {
+ osi_Log2(afsd_logp, "cm_Analyze passed CM_ERROR_ALLOFFLINE (FS cell %s vol 0x%x)",
+ cellp->name, fidp->volume);
+ msgID = MSG_ALL_SERVERS_OFFLINE;
+ format = "All servers are offline when accessing cell %s volume %d.";
+ LogEvent(EVENTLOG_WARNING_TYPE, msgID, cellp->name, fidp->volume);
+
code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
CM_GETVOL_FLAG_NO_LRU_UPDATE,
&volp);
if (code == 0) {
- if (timeLeft > 7) {
+ /*
+ * Do not perform a cm_CheckOfflineVolume() if cm_Analyze()
+ * was called by cm_CheckOfflineVolumeState().
+ */
+ if (!(reqp->flags & CM_REQ_OFFLINE_VOL_CHK) && timeLeft > 7) {
thrd_Sleep(5000);
/* cm_CheckOfflineVolume() resets the serverRef state */
lock_ReleaseRead(&cm_volumeLock);
volp = NULL;
}
- }
+ } else {
+ osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_ALLOFFLINE (VL Server)");
+ }
}
else if (errorCode == CM_ERROR_ALLBUSY) {
/* Volumes that are busy cannot be determined to be non-busy
* without actually attempting to access them.
*/
- osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_ALLBUSY.");
-
if (fidp) { /* File Server query */
+ osi_Log2(afsd_logp, "cm_Analyze passed CM_ERROR_ALLBUSY (FS cell %s vol 0x%x)",
+ cellp->name, fidp->volume);
+ msgID = MSG_ALL_SERVERS_BUSY;
+ format = "All servers are busy when accessing cell %s volume %d.";
+ LogEvent(EVENTLOG_WARNING_TYPE, msgID, cellp->name, fidp->volume);
+
code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
CM_GETVOL_FLAG_NO_LRU_UPDATE,
&volp);
lock_ReleaseWrite(&cm_serverLock);
if (free_svr_list) {
cm_FreeServerList(serverspp, 0);
- serverspp = NULL;
serversp = NULL;
free_svr_list = 0;
}
volp = NULL;
}
} else { /* VL Server query */
+ osi_Log0(afsd_logp, "cm_Analyze passed CM_ERROR_ALLBUSY (VL Server).");
+
if (timeLeft > 7) {
thrd_Sleep(5000);
free_svr_list = 1;
}
}
+
+ switch ( errorCode ) {
+ case VBUSY:
+ msgID = MSG_SERVER_REPORTS_VBUSY;
+ format = "Server %s reported busy when accessing volume %d in cell %s.";
+ break;
+ case VRESTARTING:
+ msgID = MSG_SERVER_REPORTS_VRESTARTING;
+ format = "Server %s reported restarting when accessing volume %d in cell %s.";
+ break;
+ }
+
+ if (serverp && fidp) {
+ /* Log server being offline for this volume */
+ sprintf(addr, "%d.%d.%d.%d",
+ ((serverp->addr.sin_addr.s_addr & 0xff)),
+ ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
+ ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
+ ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
+
+ osi_Log3(afsd_logp, format, osi_LogSaveString(afsd_logp,addr), fidp->volume, cellp->name);
+ LogEvent(EVENTLOG_WARNING_TYPE, msgID, addr, fidp->volume, cellp->name);
+ }
+
lock_ObtainWrite(&cm_serverLock);
for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
if (tsrp->status == srv_deleted)
if (free_svr_list) {
cm_FreeServerList(serverspp, 0);
- serverspp = NULL;
serversp = NULL;
free_svr_list = 0;
}
else if (errorCode == VNOVOL || errorCode == VMOVED || errorCode == VOFFLINE ||
errorCode == VSALVAGE || errorCode == VNOSERVICE || errorCode == VIO)
{
- char *format;
- DWORD msgID;
-
/* In case of timeout */
reqp->volumeError = errorCode;
switch ( errorCode ) {
case VNOVOL:
msgID = MSG_SERVER_REPORTS_VNOVOL;
- format = "Server %s reported volume %d as not attached (does not exist).";
+ format = "Server %s reported volume %d in cell %s as not attached (may have been moved or deleted).";
break;
case VMOVED:
msgID = MSG_SERVER_REPORTS_VMOVED;
- format = "Server %s reported volume %d as moved.";
+ format = "Server %s reported volume %d in cell %s as moved.";
break;
case VOFFLINE:
msgID = MSG_SERVER_REPORTS_VOFFLINE;
- format = "Server %s reported volume %d as offline.";
+ format = "Server %s reported volume %d in cell %s as offline.";
break;
case VSALVAGE:
msgID = MSG_SERVER_REPORTS_VSALVAGE;
- format = "Server %s reported volume %d as needs salvage.";
+ format = "Server %s reported volume %d in cell %s as needs salvage.";
break;
case VNOSERVICE:
msgID = MSG_SERVER_REPORTS_VNOSERVICE;
- format = "Server %s reported volume %d as not in service.";
+ format = "Server %s reported volume %d in cell %s as not in service.";
break;
case VIO:
msgID = MSG_SERVER_REPORTS_VIO;
- format = "Server %s reported volume %d as temporarily unaccessible.";
+ format = "Server %s reported volume %d in cell %s as temporarily unaccessible.";
break;
}
- if (serverp && fidp) {
- /* Log server being offline for this volume */
- sprintf(addr, "%d.%d.%d.%d",
- ((serverp->addr.sin_addr.s_addr & 0xff)),
- ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
- ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
- ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
+ if (fidp) { /* File Server query */
+ if (serverp) {
+ /* Log server being unavailable for this volume */
+ sprintf(addr, "%d.%d.%d.%d",
+ ((serverp->addr.sin_addr.s_addr & 0xff)),
+ ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
+ ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
+ ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
+
+ osi_Log3(afsd_logp, format, osi_LogSaveString(afsd_logp,addr), fidp->volume, cellp->name);
+ LogEvent(EVENTLOG_WARNING_TYPE, msgID, addr, fidp->volume, cellp->name);
+ }
- osi_Log2(afsd_logp, format, osi_LogSaveString(afsd_logp,addr), fidp->volume);
- LogEvent(EVENTLOG_WARNING_TYPE, msgID, addr, fidp->volume);
- }
+ code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
+ CM_GETVOL_FLAG_NO_LRU_UPDATE,
+ &volp);
+ if (code == 0)
+ statep = cm_VolumeStateByID(volp, fidp->volume);
- /*
- * Mark server offline for this volume or delete the volume
- * from the server list if it was moved or is not present.
- */
- if (!serversp && fidp) {
- code = cm_GetServerList(fidp, userp, reqp, &serverspp);
- if (code == 0) {
- serversp = *serverspp;
- free_svr_list = 1;
+ if ((errorCode == VMOVED || errorCode == VNOVOL || errorCode == VOFFLINE) &&
+ !(reqp->flags & CM_REQ_VOLUME_UPDATED))
+ {
+ code = cm_ForceUpdateVolume(fidp, userp, reqp);
+ if (code == 0)
+ location_updated = 1;
+
+ /* Even if the update fails, there might still be another replica */
+
+ reqp->flags |= CM_REQ_VOLUME_UPDATED;
+ osi_Log3(afsd_logp, "cm_Analyze called cm_ForceUpdateVolume cell %u vol %u code 0x%x",
+ fidp->cell, fidp->volume, code);
+ }
+
+ if (statep) {
+ cm_UpdateVolumeStatus(volp, statep->ID);
+ osi_Log3(afsd_logp, "cm_Analyze NewVolState cell %u vol %u state %u",
+ fidp->cell, fidp->volume, statep->state);
+ }
+
+ if (volp) {
+ lock_ObtainRead(&cm_volumeLock);
+ cm_PutVolume(volp);
+ lock_ReleaseRead(&cm_volumeLock);
+ volp = NULL;
+ }
+
+ /*
+ * Mark server offline for this volume or delete the volume
+ * from the server list if it was moved or is not present.
+ */
+ if (!serversp || location_updated) {
+ code = cm_GetServerList(fidp, userp, reqp, &serverspp);
+ if (code == 0) {
+ serversp = *serverspp;
+ free_svr_list = 1;
+ }
}
}
((tsrp->server->addr.sin_addr.s_addr & 0xff)),
((tsrp->server->addr.sin_addr.s_addr & 0xff00)>> 8),
((tsrp->server->addr.sin_addr.s_addr & 0xff0000)>> 16),
- ((tsrp->server->addr.sin_addr.s_addr & 0xff000000)>> 24));
+ ((tsrp->server->addr.sin_addr.s_addr & 0xff000000)>> 24));
- if (tsrp->server == serverp) {
+ if (cm_ServerEqual(tsrp->server, serverp)) {
/* REDIRECT */
if (errorCode == VMOVED || errorCode == VNOVOL) {
- osi_Log2(afsd_logp, "volume %d not present on server %s",
+ osi_Log2(afsd_logp, "volume %d not present on server %s",
fidp->volume, osi_LogSaveString(afsd_logp,addr));
tsrp->status = srv_deleted;
if (fidp)
cm_RemoveVolumeFromServer(serverp, fidp->volume);
} else {
- osi_Log2(afsd_logp, "volume %d instance on server %s marked offline",
+ osi_Log2(afsd_logp, "volume %d instance on server %s marked offline",
fidp->volume, osi_LogSaveString(afsd_logp,addr));
tsrp->status = srv_offline;
}
/* break; */
} else {
- osi_Log3(afsd_logp, "volume %d exists on server %s with status %u",
+ osi_Log3(afsd_logp, "volume %d exists on server %s with status %u",
fidp->volume, osi_LogSaveString(afsd_logp,addr), tsrp->status);
}
- }
+ }
lock_ReleaseWrite(&cm_serverLock);
/* Free the server list before cm_ForceUpdateVolume is called */
if (free_svr_list) {
cm_FreeServerList(serverspp, 0);
- serverspp = NULL;
serversp = NULL;
free_svr_list = 0;
}
- if (fidp) { /* File Server query */
- code = cm_FindVolumeByID(cellp, fidp->volume, userp, reqp,
- CM_GETVOL_FLAG_NO_LRU_UPDATE,
- &volp);
- if (code == 0)
- statep = cm_VolumeStateByID(volp, fidp->volume);
-
- if ((errorCode == VMOVED || errorCode == VNOVOL) &&
- !(reqp->flags & CM_REQ_VOLUME_UPDATED))
- {
- code = cm_ForceUpdateVolume(fidp, userp, reqp);
- if (code)
- timeLeft = 0; /* prevent a retry on failure */
- else
- reqp->flags |= CM_REQ_VOLUME_UPDATED;
- osi_Log3(afsd_logp, "cm_Analyze called cm_ForceUpdateVolume cell %u vol %u code 0x%x",
- fidp->cell, fidp->volume, code);
- }
-
- if (statep) {
- cm_UpdateVolumeStatus(volp, statep->ID);
- osi_Log3(afsd_logp, "cm_Analyze NewVolState cell %u vol %u state %u",
- fidp->cell, fidp->volume, statep->state);
- }
-
- if (volp) {
- lock_ObtainRead(&cm_volumeLock);
- cm_PutVolume(volp);
- lock_ReleaseRead(&cm_volumeLock);
- volp = NULL;
- }
- }
-
if ( timeLeft > 2 )
retry = 1;
} else if ( errorCode == VNOVNODE ) {
osi_LogSaveString(afsd_logp,addr));
reqp->tokenIdleErrorServp = serverp;
reqp->idleError++;
+ }
- if (timeLeft > 2) {
- if (!fidp) { /* vldb */
- retry = 1;
- } else { /* file */
- cm_volume_t *volp = cm_GetVolumeByFID(fidp);
- if (volp) {
- if (fidp->volume == cm_GetROVolumeID(volp))
- retry = 1;
- cm_PutVolume(volp);
- }
+ if (timeLeft > 2) {
+ if (!fidp) { /* vldb */
+ retry = 1;
+ } else { /* file */
+ cm_volume_t *volp = cm_GetVolumeByFID(fidp);
+ if (volp) {
+ if (fidp->volume == cm_GetROVolumeID(volp))
+ retry = 1;
+ cm_PutVolume(volp);
}
}
}
}
+ else if (errorCode == RX_MSGSIZE) {
+ /*
+ * RPC failed because a transmitted rx packet
+ * may have grown larger than the path mtu.
+ * Force a retry and the Rx library will try
+ * with a smaller mtu size.
+ */
+
+ if (serverp)
+ sprintf(addr, "%d.%d.%d.%d",
+ ((serverp->addr.sin_addr.s_addr & 0xff)),
+ ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
+ ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
+ ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
+
+ LogEvent(EVENTLOG_WARNING_TYPE, MSG_RX_MSGSIZE_EXCEEDED, addr);
+ osi_Log1(afsd_logp, "cm_Analyze: Path MTU may have been exceeded addr[%s]",
+ osi_LogSaveString(afsd_logp,addr));
+
+ retry = 1;
+ }
+ else if (errorCode == CM_RX_RETRY_BUSY_CALL) {
+ /*
+ * RPC failed because the selected call channel
+ * is currently busy on the server. Unconditionally
+ * retry the request so an alternate call channel can be used.
+ */
+ if (serverp)
+ sprintf(addr, "%d.%d.%d.%d",
+ ((serverp->addr.sin_addr.s_addr & 0xff)),
+ ((serverp->addr.sin_addr.s_addr & 0xff00)>> 8),
+ ((serverp->addr.sin_addr.s_addr & 0xff0000)>> 16),
+ ((serverp->addr.sin_addr.s_addr & 0xff000000)>> 24));
+
+ LogEvent(EVENTLOG_WARNING_TYPE, MSG_RX_BUSY_CALL_CHANNEL, addr);
+ osi_Log1(afsd_logp, "cm_Analyze: Retry RPC due to busy call channel addr[%s]",
+ osi_LogSaveString(afsd_logp,addr));
+ retry = 1;
+ }
else if (errorCode >= -64 && errorCode < 0) {
/* mark server as down */
if (serverp)
}
/* If not allowed to retry, don't */
- if (!forcing_new && (reqp->flags & CM_REQ_NORETRY))
+ if (!forcing_new && (reqp->flags & CM_REQ_NORETRY) &&
+ (errorCode != RX_MSGSIZE && errorCode != CM_RX_RETRY_BUSY_CALL))
retry = 0;
else if (retry && dead_session)
retry = 0;
- out:
/* drop this on the way out */
if (connp)
cm_PutConn(connp);
lock_ReleaseRead(&cm_serverLock);
if (firstError == 0) {
- if (allDown)
+ if (allDown) {
firstError = (reqp->tokenError ? reqp->tokenError :
(reqp->idleError ? RX_CALL_TIMEOUT : CM_ERROR_ALLDOWN));
- else if (allBusy)
+ /*
+ * if we experienced either a token error or and idle dead time error
+ * and now all of the servers are down, we have either tried them
+ * all or lost connectivity. Clear the error we are returning so
+ * we will not return it indefinitely if the request is retried.
+ */
+ reqp->idleError = reqp->tokenError = 0;
+ } else if (allBusy) {
firstError = CM_ERROR_ALLBUSY;
- else if (allOffline || (someBusy && someOffline))
+ } else if (allOffline || (someBusy && someOffline)) {
firstError = CM_ERROR_ALLOFFLINE;
- else {
+ } else {
osi_Log0(afsd_logp, "cm_ConnByMServers returning impossible error TIMEDOUT");
firstError = CM_ERROR_TIMEDOUT;
}
cm_PutServer(tcp->serverp);
cm_ReleaseUser(userp);
*lcpp = tcp->nextp;
+ rx_SetConnSecondsUntilNatPing(tcp->rxconnp, 0);
rx_DestroyConnection(tcp->rxconnp);
lock_FinalizeMutex(&tcp->mx);
free(tcp);
rx_SetConnDeadTime(tcp->rxconnp, ConnDeadtimeout);
rx_SetConnHardDeadTime(tcp->rxconnp, HardDeadtimeout);
rx_SetConnIdleDeadTime(tcp->rxconnp, IdleDeadtimeout);
- if (NatPingInterval)
+
+ /*
+ * Register the error to be returned on an idle dead timeout
+ */
+ rx_SetServerConnIdleDeadErr(tcp->rxconnp, RX_CALL_DEAD);
+
+ /*
+ * Let the Rx library know that we can auto-retry if an
+ * RX_MSGSIZE error is returned.
+ */
+ if (rx_pmtu_discovery)
+ rx_SetMsgsizeRetryErr(tcp->rxconnp, RX_MSGSIZE);
+
+ /*
+ * Attempt to limit NAT pings to the anonymous file server connections.
+ * Only file servers implement client callbacks and we only need one ping
+ * to be sent to each server.
+ */
+ if (NatPingInterval && serverp->type == CM_SERVER_FILE && secIndex == 0)
rx_SetConnSecondsUntilNatPing(tcp->rxconnp, NatPingInterval);
+
tcp->ucgen = ucellp->gen;
if (secObjp)
rxs_Release(secObjp); /* Decrement the initial refCount */
else
osi_Log0(afsd_logp, "cm_ConnByServer replace connection due to crypt change");
tcp->flags &= ~CM_CONN_FLAG_FORCE_NEW;
+ rx_SetConnSecondsUntilNatPing(tcp->rxconnp, 0);
rx_DestroyConnection(tcp->rxconnp);
cm_NewRXConnection(tcp, ucellp, serverp);
}
git://git.openafs.org / openafs.git / blobdiff