From 874f562aebeb86c9cc3c9b19d3c60e919489d905 Mon Sep 17 00:00:00 2001
From: alex <alex>
Date: Fri, 9 Jul 2004 07:15:17 +0000
Subject: [PATCH] b=3405
robert's fixes:
- add CONNECT_INITAL flag for the initial connect message from a client. When
server sees an old client reconnect with that flag, then it needs to set the
handles as if it was a new client.
- set the initial imp_conn_cnt from the server's conn_cnt. When client
connects with INITIAL set, the server sends back the old conn_cnt+1.
- timeout for inter-MDS requests is set to obd_timeout / 2
---
lustre/include/linux/lustre_idl.h | 1 +
lustre/include/linux/lustre_lib.h | 2 +-
lustre/ldlm/ldlm_lib.c | 32 +++++++++++++++++--------------
lustre/ptlrpc/client.c | 5 ++++-
lustre/ptlrpc/import.c | 15 +++++++++++----
5 files changed, 35 insertions(+), 20 deletions(-)
diff --git a/lustre/include/linux/lustre_idl.h b/lustre/include/linux/lustre_idl.h
index 4ab5ce813e..588882627f 100644
--- a/lustre/include/linux/lustre_idl.h
+++ b/lustre/include/linux/lustre_idl.h
@@ -212,6 +212,7 @@ static inline void lustre_msg_set_op_flags(struct lustre_msg *msg, int flags)
#define MSG_CONNECT_REPLAYABLE 0x4
//#define MSG_CONNECT_PEER 0x8
#define MSG_CONNECT_LIBCLIENT 0x10
+#define MSG_CONNECT_INITIAL 0x20
/*
* OST requests: OBDO & OBD request records
diff --git a/lustre/include/linux/lustre_lib.h b/lustre/include/linux/lustre_lib.h
index f354ecb6d5..b57146d228 100644
--- a/lustre/include/linux/lustre_lib.h
+++ b/lustre/include/linux/lustre_lib.h
@@ -79,7 +79,7 @@ int target_handle_connect(struct ptlrpc_request *req, svc_handler_t handler);
int target_handle_disconnect(struct ptlrpc_request *req);
void target_destroy_export(struct obd_export *exp);
int target_handle_reconnect(struct lustre_handle *conn, struct obd_export *exp,
- struct obd_uuid *cluuid);
+ struct obd_uuid *cluuid, int);
int target_handle_ping(struct ptlrpc_request *req);
void target_cancel_recovery_timer(struct obd_device *obd);
diff --git a/lustre/ldlm/ldlm_lib.c b/lustre/ldlm/ldlm_lib.c
index b6ae61c98f..d764484870 100644
--- a/lustre/ldlm/ldlm_lib.c
+++ b/lustre/ldlm/ldlm_lib.c
@@ -333,19 +333,16 @@ int client_disconnect_export(struct obd_export *exp, int failover)
* -------------------------------------------------------------------------- */
int target_handle_reconnect(struct lustre_handle *conn, struct obd_export *exp,
- struct obd_uuid *cluuid)
+ struct obd_uuid *cluuid, int initial_conn)
{
- if (exp->exp_connection) {
+ if (exp->exp_connection && !initial_conn) {
struct lustre_handle *hdl;
hdl = &exp->exp_imp_reverse->imp_remote_handle;
/* Might be a re-connect after a partition. */
-#warning "FIXME ASAP"
- memcpy(&hdl->cookie, &conn->cookie, sizeof(conn->cookie));
- if (1 || !memcmp(&conn->cookie, &hdl->cookie, sizeof conn->cookie)) {
+ if (!memcmp(&conn->cookie, &hdl->cookie, sizeof conn->cookie)) {
CERROR("%s reconnecting\n", cluuid->uuid);
conn->cookie = exp->exp_handle.h_cookie;
- /*RETURN(EALREADY);*/
- RETURN(0);
+ RETURN(EALREADY);
} else {
CERROR("%s reconnecting from %s, "
"handle mismatch (ours "LPX64", theirs "
@@ -377,6 +374,7 @@ int target_handle_connect(struct ptlrpc_request *req, svc_handler_t handler)
char *str, *tmp;
int rc = 0, abort_recovery;
unsigned long flags;
+ int initial_conn = 0;
ENTRY;
OBD_RACE(OBD_FAIL_TGT_CONN_RACE);
@@ -444,6 +442,9 @@ int target_handle_connect(struct ptlrpc_request *req, svc_handler_t handler)
if (obd_uuid_equals(&cluuid, &target->obd_uuid))
goto dont_check_exports;
+ if (lustre_msg_get_op_flags(req->rq_reqmsg) & MSG_CONNECT_INITIAL)
+ initial_conn = 1;
+
spin_lock(&target->obd_dev_lock);
list_for_each(p, &target->obd_exports) {
export = list_entry(p, struct obd_export, exp_obd_chain);
@@ -451,7 +452,8 @@ int target_handle_connect(struct ptlrpc_request *req, svc_handler_t handler)
spin_unlock(&target->obd_dev_lock);
LASSERT(export->exp_obd == target);
- rc = target_handle_reconnect(&conn, export, &cluuid);
+ rc = target_handle_reconnect(&conn, export, &cluuid,
+ initial_conn);
break;
}
export = NULL;
@@ -459,15 +461,16 @@ int target_handle_connect(struct ptlrpc_request *req, svc_handler_t handler)
/* If we found an export, we already unlocked. */
if (!export) {
spin_unlock(&target->obd_dev_lock);
- } else if (req->rq_reqmsg->conn_cnt == 1) {
+ } else if (req->rq_reqmsg->conn_cnt == 1 && !initial_conn) {
CERROR("%s reconnected with 1 conn_cnt; cookies not random?\n",
cluuid.uuid);
-#warning "FIXME ASAP"
- /*GOTO(out, rc = -EALREADY);*/
+ GOTO(out, rc = -EALREADY);
}
/* Tell the client if we're in recovery. */
/* If this is the first client, start the recovery timer */
+ CWARN("%s: connection from %s %s\n", target->obd_name, cluuid.uuid,
+ target->obd_recovering ? "(recovering)" : "");
if (target->obd_recovering) {
lustre_msg_add_op_flags(req->rq_repmsg, MSG_CONNECT_RECOVERING);
target_start_recovery_timer(target, handler);
@@ -516,14 +519,15 @@ int target_handle_connect(struct ptlrpc_request *req, svc_handler_t handler)
LASSERT(export != NULL);
spin_lock_irqsave(&export->exp_lock, flags);
-#warning "FIXME ASAP"
- if (0 && export->exp_conn_cnt >= req->rq_reqmsg->conn_cnt) {
+ if (initial_conn) {
+ req->rq_repmsg->conn_cnt = export->exp_conn_cnt + 1;
+ } else if (export->exp_conn_cnt >= req->rq_reqmsg->conn_cnt) {
CERROR("%s: already connected at a higher conn_cnt: %d > %d\n",
cluuid.uuid, export->exp_conn_cnt,
req->rq_reqmsg->conn_cnt);
spin_unlock_irqrestore(&export->exp_lock, flags);
GOTO(out, rc = -EALREADY);
- }
+ }
export->exp_conn_cnt = req->rq_reqmsg->conn_cnt;
spin_unlock_irqrestore(&export->exp_lock, flags);
diff --git a/lustre/ptlrpc/client.c b/lustre/ptlrpc/client.c
index 452f3edd43..5a4f9ae9cf 100644
--- a/lustre/ptlrpc/client.c
+++ b/lustre/ptlrpc/client.c
@@ -203,7 +203,10 @@ struct ptlrpc_request *ptlrpc_prep_req(struct obd_import *imp, int opcode,
RETURN(NULL);
}
- request->rq_timeout = obd_timeout;
+ if (imp->imp_server_timeout)
+ request->rq_timeout = obd_timeout / 2;
+ else
+ request->rq_timeout = obd_timeout;
request->rq_send_state = LUSTRE_IMP_FULL;
request->rq_type = PTL_RPC_MSG_REQUEST;
request->rq_import = class_import_get(imp);
diff --git a/lustre/ptlrpc/import.c b/lustre/ptlrpc/import.c
index e2106b480b..86f3adf64c 100644
--- a/lustre/ptlrpc/import.c
+++ b/lustre/ptlrpc/import.c
@@ -255,14 +255,13 @@ int ptlrpc_connect_import(struct obd_import *imp, char * new_uuid)
IMPORT_SET_STATE_NOLOCK(imp, LUSTRE_IMP_CONNECTING);
- imp->imp_conn_cnt++;
imp->imp_resend_replay = 0;
if (imp->imp_remote_handle.cookie == 0) {
initial_connect = 1;
} else {
committed_before_reconnect = imp->imp_peer_committed_transno;;
-
+ imp->imp_conn_cnt++;
}
@@ -322,8 +321,11 @@ int ptlrpc_connect_import(struct obd_import *imp, char * new_uuid)
aa->pcaa_peer_committed = committed_before_reconnect;
aa->pcaa_initial_connect = initial_connect;
- if (aa->pcaa_initial_connect)
- imp->imp_replayable = 1;
+ if (aa->pcaa_initial_connect) {
+ lustre_msg_add_op_flags(request->rq_reqmsg,
+ MSG_CONNECT_INITIAL);
+ imp->imp_replayable = 1;
+ }
ptlrpcd_add_req(request);
rc = 0;
@@ -366,6 +368,10 @@ static int ptlrpc_connect_interpret(struct ptlrpc_request *request,
} else {
imp->imp_replayable = 0;
}
+ LASSERTF(imp->imp_conn_cnt < request->rq_repmsg->conn_cnt,
+ "imp conn_cnt %d req conn_cnt %d",
+ imp->imp_conn_cnt, request->rq_repmsg->conn_cnt);
+ imp->imp_conn_cnt = request->rq_repmsg->conn_cnt;
imp->imp_remote_handle = request->rq_repmsg->handle;
IMPORT_SET_STATE(imp, LUSTRE_IMP_FULL);
GOTO(finish, rc = 0);
@@ -412,6 +418,7 @@ static int ptlrpc_connect_interpret(struct ptlrpc_request *request,
imp->imp_last_replay_transno = 0;
IMPORT_SET_STATE(imp, LUSTRE_IMP_REPLAY);
} else {
+ CWARN("oops! we get evicted from %s\n", imp->imp_target_uuid.uuid);
imp->imp_remote_handle = request->rq_repmsg->handle;
IMPORT_SET_STATE(imp, LUSTRE_IMP_EVICTED);
}
--
GitLab