argv[0] = obd_lustre_upcall;
argv[1] = "RECOVERY_OVER";
- argv[2] = obd->obd_uuid.uuid;
+ argv[2] = (char *)obd->obd_uuid.uuid;
argv[3] = NULL;
envp[0] = "HOME=/";
argv[0], argv[1], argv[2], rc);
} else {
- CERROR("Invoked upcall %s %s %s\n",
- argv[0], argv[1], argv[2]);
+ CWARN("Invoked upcall %s %s %s\n",
+ argv[0], argv[1], argv[2]);
}
}
argv[0] = obd_lustre_upcall;
argv[1] = "FAILED_IMPORT";
- argv[2] = imp->imp_target_uuid.uuid;
+ argv[2] = (char *)imp->imp_target_uuid.uuid;
argv[3] = imp->imp_obd->obd_name;
- argv[4] = imp->imp_connection->c_remote_uuid.uuid;
- argv[5] = imp->imp_obd->obd_uuid.uuid;
+ argv[4] = (char *)imp->imp_connection->c_remote_uuid.uuid;
+ argv[5] = (char *)imp->imp_obd->obd_uuid.uuid;
argv[6] = NULL;
envp[0] = "HOME=/";
rc = USERMODEHELPER(argv[0], argv, envp);
if (rc < 0) {
CERROR("Error invoking recovery upcall %s %s %s %s %s: %d; "
- "check /proc/sys/lustre/lustre_upcall\n",
+ "check /proc/sys/lustre/upcall\n",
argv[0], argv[1], argv[2], argv[3], argv[4],rc);
} else {
- CERROR("Invoked upcall %s %s %s %s %s\n",
- argv[0], argv[1], argv[2], argv[3], argv[4]);
+ CWARN("Invoked upcall %s %s %s %s %s\n",
+ argv[0], argv[1], argv[2], argv[3], argv[4]);
}
#else
if (imp->imp_state == LUSTRE_IMP_CLOSED) {
#endif
}
-int ptlrpc_replay_next(struct obd_import *imp)
+/* This might block waiting for the upcall to start, so it should
+ * not be called from a thread that shouldn't block. (Like ptlrpcd) */
+void ptlrpc_initiate_recovery(struct obd_import *imp)
+{
+ ENTRY;
+
+ LASSERT (obd_lustre_upcall != NULL);
+
+ if (strcmp(obd_lustre_upcall, "DEFAULT") == 0) {
+ CDEBUG(D_HA, "%s: starting recovery without upcall\n",
+ imp->imp_target_uuid.uuid);
+ ptlrpc_connect_import(imp, NULL);
+ }
+ else if (strcmp(obd_lustre_upcall, "NONE") == 0) {
+ CDEBUG(D_HA, "%s: recovery disabled\n",
+ imp->imp_target_uuid.uuid);
+ }
+ else {
+ CDEBUG(D_HA, "%s: calling upcall to start recovery\n",
+ imp->imp_target_uuid.uuid);
+ ptlrpc_run_failed_import_upcall(imp);
+ }
+
+ EXIT;
+}
+
+int ptlrpc_replay_next(struct obd_import *imp, int *inflight)
{
int rc = 0;
struct list_head *tmp, *pos;
- struct ptlrpc_request *req;
+ struct ptlrpc_request *req = NULL;
unsigned long flags;
__u64 last_transno;
- int sent_req = 0;
ENTRY;
+ *inflight = 0;
+
/* It might have committed some after we last spoke, so make sure we
* get rid of them now.
*/
last_transno = imp->imp_last_replay_transno;
spin_unlock_irqrestore(&imp->imp_lock, flags);
- CDEBUG(D_HA, "import %p from %s has committed "LPD64"\n",
- imp, imp->imp_target_uuid.uuid, imp->imp_peer_committed_transno);
+ CDEBUG(D_HA, "import %p from %s committed "LPU64" last "LPU64"\n",
+ imp, imp->imp_target_uuid.uuid, imp->imp_peer_committed_transno,
+ last_transno);
+
/* Do I need to hold a lock across this iteration? We shouldn't be
* racing with any additions to the list, because we're in recovery
* and are therefore not processing additional requests to add. Calls
*/
list_for_each_safe(tmp, pos, &imp->imp_replay_list) {
req = list_entry(tmp, struct ptlrpc_request, rq_replay_list);
+
+ /* If need to resend, stop on the matching one first. It's
+ possible though it's already been committed, so in that case
+ we'll just continue with replay */
+ if (imp->imp_resend_replay &&
+ req->rq_transno == last_transno) {
+ lustre_msg_add_flags(req->rq_reqmsg, MSG_RESENT);
+ break;
+ }
+
if (req->rq_transno > last_transno) {
- /* remove from list so ptlrpcd can send the
- req, it should be reinserted after it is
- sent and replied. Perhaps better solution
- would be to add req->rq_replay_list so the
- req can be saved for replay and still go
- through the normal send thread. */
- rc = ptlrpc_replay_req(req);
- if (rc) {
- CERROR("recovery replay error %d for req "LPD64"\n",
- rc, req->rq_xid);
- RETURN(rc);
- }
- sent_req = 1;
+ imp->imp_last_replay_transno = req->rq_transno;
break;
}
+ req = NULL;
}
- RETURN(sent_req);
+ imp->imp_resend_replay = 0;
+
+ if (req != NULL) {
+ rc = ptlrpc_replay_req(req);
+ if (rc) {
+ CERROR("recovery replay error %d for req "
+ LPD64"\n", rc, req->rq_xid);
+ RETURN(rc);
+ }
+ imp->imp_reqs_replayed++;
+ *inflight = 1;
+ }
+ RETURN(rc);
}
int ptlrpc_resend(struct obd_import *imp)
spin_unlock_irqrestore(&imp->imp_lock, flags);
}
-inline void ptlrpc_invalidate_import_state(struct obd_import *imp)
-{
- struct obd_device *obd = imp->imp_obd;
- struct ldlm_namespace *ns = obd->obd_namespace;
-
- ptlrpc_abort_inflight(imp);
-
- obd_invalidate_import(obd, imp);
-
- ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
-}
-
-void ptlrpc_handle_failed_import(struct obd_import *imp)
-{
- ENTRY;
-
- if (!imp->imp_replayable) {
- CDEBUG(D_HA,
- "import %s@%s for %s not replayable, deactivating\n",
- imp->imp_target_uuid.uuid,
- imp->imp_connection->c_remote_uuid.uuid,
- imp->imp_obd->obd_name);
- ptlrpc_set_import_active(imp, 0);
- }
-
- ptlrpc_run_failed_import_upcall(imp);
- EXIT;
-}
-
void ptlrpc_request_handle_notconn(struct ptlrpc_request *failed_req)
{
int rc;
imp->imp_target_uuid.uuid,
imp->imp_connection->c_remote_uuid.uuid);
- ptlrpc_set_import_discon(imp);
+ if (ptlrpc_set_import_discon(imp)) {
+ if (!imp->imp_replayable) {
+ CDEBUG(D_HA, "import %s@%s for %s not replayable, "
+ "auto-deactivating\n",
+ imp->imp_target_uuid.uuid,
+ imp->imp_connection->c_remote_uuid.uuid,
+ imp->imp_obd->obd_name);
+ ptlrpc_deactivate_import(imp);
+ }
+
+ /* to control recovery via lctl {disable|enable}_recovery */
+ if (imp->imp_deactive == 0)
+ rc = ptlrpc_connect_import(imp, NULL);
+ }
- rc = ptlrpc_connect_import(imp, NULL);
-
/* Wait for recovery to complete and resend. If evicted, then
this request will be errored out later.*/
spin_lock_irqsave(&failed_req->rq_lock, flags);
- if (!failed_req->rq_no_resend)
- failed_req->rq_resend = 1;
+ failed_req->rq_resend = 1;
spin_unlock_irqrestore(&failed_req->rq_lock, flags);
-
+
EXIT;
}
+/*
+ * This should only be called by the ioctl interface, currently
+ * with the lctl deactivate and activate commands.
+ */
int ptlrpc_set_import_active(struct obd_import *imp, int active)
{
struct obd_device *obd = imp->imp_obd;
- unsigned long flags;
+ int rc = 0;
LASSERT(obd);
/* When deactivating, mark import invalid, and abort in-flight
* requests. */
if (!active) {
- spin_lock_irqsave(&imp->imp_lock, flags);
- /* This is a bit of a hack, but invalidating replayable
- * imports makes a temporary reconnect failure into a much more
- * ugly -- and hard to remedy -- situation. */
- if (!imp->imp_replayable) {
- CDEBUG(D_HA, "setting import %s INVALID\n",
- imp->imp_target_uuid.uuid);
- imp->imp_invalid = 1;
- }
- imp->imp_generation++;
- spin_unlock_irqrestore(&imp->imp_lock, flags);
- ptlrpc_invalidate_import_state(imp);
+ ptlrpc_invalidate_import(imp, 0);
+ imp->imp_deactive = 1;
}
- /* When activating, mark import valid */
+ /* When activating, mark import valid, and attempt recovery */
if (active) {
+ imp->imp_deactive = 0;
CDEBUG(D_HA, "setting import %s VALID\n",
imp->imp_target_uuid.uuid);
- spin_lock_irqsave(&imp->imp_lock, flags);
- imp->imp_invalid = 0;
- spin_unlock_irqrestore(&imp->imp_lock, flags);
+ rc = ptlrpc_recover_import(imp, NULL);
}
- if (obd->obd_observer)
- RETURN(obd_notify(obd->obd_observer, obd, active));
-
- RETURN(0);
+ RETURN(rc);
}
int ptlrpc_recover_import(struct obd_import *imp, char *new_uuid)
{
int rc;
ENTRY;
-
+
/* force import to be disconnected. */
ptlrpc_set_import_discon(imp);
-
+
rc = ptlrpc_recover_import_no_retry(imp, new_uuid);
RETURN(rc);
return in_recovery;
}
+int ptlrpc_import_control_recovery(struct obd_import *imp, int disable)
+{
+ unsigned long flags;
+
+ /* with imp_deactivate == 1 pinger won't initiate re-connect */
+ spin_lock_irqsave(&imp->imp_lock, flags);
+ if (disable)
+ imp->imp_deactive = 1;
+ else
+ imp->imp_deactive = 0;
+ if (imp->imp_state == LUSTRE_IMP_DISCON) {
+ imp->imp_force_verify = 1;
+ ptlrpc_pinger_wake_up();
+ }
+ spin_unlock_irqrestore(&imp->imp_lock, flags);
+ RETURN(0);
+}
+
static int ptlrpc_recover_import_no_retry(struct obd_import *imp,
char *new_uuid)
{
spin_lock_irqsave(&imp->imp_lock, flags);
if (imp->imp_state != LUSTRE_IMP_DISCON) {
in_recovery = 1;
- }
+ }
spin_unlock_irqrestore(&imp->imp_lock, flags);
if (in_recovery == 1)
RETURN(-EALREADY);
-
rc = ptlrpc_connect_import(imp, new_uuid);
if (rc)
RETURN(rc);
- CDEBUG(D_ERROR, "%s: recovery started, waiting\n",
- imp->imp_client->cli_name);
+ CDEBUG(D_HA, "%s: recovery started, waiting\n",
+ imp->imp_target_uuid.uuid);
lwi = LWI_TIMEOUT(MAX(obd_timeout * HZ, 1), NULL, NULL);
- rc = l_wait_event(imp->imp_recovery_waitq,
+ rc = l_wait_event(imp->imp_recovery_waitq,
!ptlrpc_import_in_recovery(imp), &lwi);
- CDEBUG(D_ERROR, "%s: recovery finished\n",
- imp->imp_client->cli_name);
+ CDEBUG(D_HA, "%s: recovery finished\n",
+ imp->imp_target_uuid.uuid);
RETURN(rc);
-
}
void ptlrpc_fail_export(struct obd_export *exp)
CDEBUG(D_HA, "disconnecting export %p/%s\n",
exp, exp->exp_client_uuid.uuid);
+ if (obd_dump_on_timeout)
+ portals_debug_dumplog();
+
/* Most callers into obd_disconnect are removing their own reference
* (request, for example) in addition to the one from the hash table.
* We don't have such a reference here, so make one. */
git://git.whamcloud.com / fs / lustre-release.git / blobdiff