From f43b9324340b6930d9ace6224cfdae4c2bcec84e Mon Sep 17 00:00:00 2001
From: alex <alex>
Date: Sat, 7 May 2005 19:01:31 +0000
Subject: [PATCH] b=6063
- to avoid possible lock collision during replay, we should replay all
request before any locks
---
lustre/include/linux/lustre_export.h | 3 +-
lustre/include/linux/lustre_idl.h | 2 +
lustre/include/linux/obd.h | 11 +-
lustre/include/linux/obd_class.h | 6 +-
lustre/ldlm/ldlm_lib.c | 360 ++++++++++++++++++++-------
lustre/ldlm/ldlm_lockd.c | 4 +-
lustre/ldlm/ldlm_request.c | 5 +
lustre/mds/handler.c | 8 -
lustre/mds/mds_fs.c | 10 +-
lustre/obdclass/genops.c | 10 +-
lustre/obdclass/obd_config.c | 5 +-
lustre/obdfilter/filter.c | 5 +-
lustre/ost/ost_handler.c | 9 -
lustre/ptlrpc/import.c | 2 +-
14 files changed, 316 insertions(+), 124 deletions(-)
diff --git a/lustre/include/linux/lustre_export.h b/lustre/include/linux/lustre_export.h
index e4a0bdbc68..ec36e7215c 100644
--- a/lustre/include/linux/lustre_export.h
+++ b/lustre/include/linux/lustre_export.h
@@ -84,7 +84,8 @@ struct obd_export {
/* ^ protects exp_outstanding_replies too */
unsigned long exp_flags;
int exp_failed:1,
- exp_replay_needed:1,
+ exp_req_replay_needed:1,
+ exp_lock_replay_needed:1,
exp_libclient:1, /* liblustre client? */
exp_sync:1;
union {
diff --git a/lustre/include/linux/lustre_idl.h b/lustre/include/linux/lustre_idl.h
index 8a0926c75c..1125ccfa94 100644
--- a/lustre/include/linux/lustre_idl.h
+++ b/lustre/include/linux/lustre_idl.h
@@ -174,6 +174,8 @@ struct lustre_msg {
#define MSG_LAST_REPLAY 1
#define MSG_RESENT 2
#define MSG_REPLAY 4
+#define MSG_REQ_REPLAY_DONE 8
+#define MSG_LOCK_REPLAY_DONE 16
static inline int lustre_msg_get_flags(struct lustre_msg *msg)
{
diff --git a/lustre/include/linux/obd.h b/lustre/include/linux/obd.h
index ac5d138fea..0b581eb2c6 100644
--- a/lustre/include/linux/obd.h
+++ b/lustre/include/linux/obd.h
@@ -648,16 +648,23 @@ struct obd_device {
spinlock_t obd_processing_task_lock;
__u64 obd_next_recovery_transno;
int obd_replayed_requests;
+ int obd_replayed_locks;
int obd_requests_queued_for_recovery;
wait_queue_head_t obd_next_transno_waitq;
struct list_head obd_uncommitted_replies;
spinlock_t obd_uncommitted_replies_lock;
struct timer_list obd_recovery_timer;
- struct list_head obd_recovery_queue;
- struct list_head obd_delayed_reply_queue;
time_t obd_recovery_start;
time_t obd_recovery_end;
+ atomic_t obd_req_replay_clients;
+ atomic_t obd_lock_replay_clients;
+
+ struct list_head obd_req_replay_queue;
+ struct list_head obd_lock_replay_queue;
+ struct list_head obd_final_req_queue;
+ int obd_recovery_stage;
+
union {
struct filter_obd filter;
struct mds_obd mds;
diff --git a/lustre/include/linux/obd_class.h b/lustre/include/linux/obd_class.h
index 9f41ed278d..2da67304ac 100644
--- a/lustre/include/linux/obd_class.h
+++ b/lustre/include/linux/obd_class.h
@@ -154,8 +154,10 @@ void class_put_type(struct obd_type *type);
int class_connect(struct lustre_handle *conn, struct obd_device *obd,
struct obd_uuid *cluuid);
int class_disconnect(struct obd_export *exp, unsigned long flags);
-void class_disconnect_exports(struct obd_device *obddev, unsigned long flags);
-void class_disconnect_stale_exports(struct obd_device *obddev, unsigned long flags);
+void class_disconnect_exports(struct obd_device *, unsigned long);
+int class_disconnect_stale_exports(struct obd_device *,
+ int (*test_export)(struct obd_export *),
+ unsigned long);
/* generic operations shared by various OBD types */
int class_multi_setup(struct obd_device *obddev, uint32_t len, void *data);
diff --git a/lustre/ldlm/ldlm_lib.c b/lustre/ldlm/ldlm_lib.c
index a4ea476301..9244c92033 100644
--- a/lustre/ldlm/ldlm_lib.c
+++ b/lustre/ldlm/ldlm_lib.c
@@ -898,12 +898,8 @@ static void target_release_saved_req(struct ptlrpc_request *req)
static void target_finish_recovery(struct obd_device *obd)
{
- struct list_head *tmp, *n;
int rc;
- CWARN("%s: sending delayed replies to recovered clients\n",
- obd->obd_name);
-
ldlm_reprocess_all_ns(obd->obd_namespace);
/* when recovery finished, cleanup orphans on mds and ost */
@@ -916,26 +912,40 @@ static void target_finish_recovery(struct obd_device *obd)
CERROR("postrecov failed %d\n", rc);
}
+ obd->obd_recovery_end = LTIME_S(CURRENT_TIME);
+ return;
+}
- list_for_each_safe(tmp, n, &obd->obd_delayed_reply_queue) {
- struct ptlrpc_request *req;
+static void abort_req_replay_queue(struct obd_device *obd)
+{
+ struct ptlrpc_request *req;
+ struct list_head *tmp, *n;
+ int rc;
+
+ list_for_each_safe(tmp, n, &obd->obd_req_replay_queue) {
req = list_entry(tmp, struct ptlrpc_request, rq_list);
list_del(&req->rq_list);
- DEBUG_REQ(D_ERROR, req, "delayed:");
- ptlrpc_reply(req);
+ DEBUG_REQ(D_ERROR, req, "aborted:");
+ req->rq_status = -ENOTCONN;
+ req->rq_type = PTL_RPC_MSG_ERR;
+ rc = lustre_pack_reply(req, 0, NULL, NULL);
+ if (rc == 0) {
+ ptlrpc_reply(req);
+ } else {
+ DEBUG_REQ(D_ERROR, req,
+ "packing failed for abort-reply; skipping");
+ }
target_release_saved_req(req);
}
- obd->obd_recovery_end = LTIME_S(CURRENT_TIME);
- return;
}
-static void abort_recovery_queue(struct obd_device *obd)
+static void abort_lock_replay_queue(struct obd_device *obd)
{
struct ptlrpc_request *req;
struct list_head *tmp, *n;
int rc;
- list_for_each_safe(tmp, n, &obd->obd_recovery_queue) {
+ list_for_each_safe(tmp, n, &obd->obd_lock_replay_queue) {
req = list_entry(tmp, struct ptlrpc_request, rq_list);
list_del(&req->rq_list);
DEBUG_REQ(D_ERROR, req, "aborted:");
@@ -976,14 +986,19 @@ void target_cleanup_recovery(struct obd_device *obd)
target_cancel_recovery_timer(obd);
spin_unlock_bh(&obd->obd_processing_task_lock);
- list_for_each_safe(tmp, n, &obd->obd_delayed_reply_queue) {
+ list_for_each_safe(tmp, n, &obd->obd_req_replay_queue) {
req = list_entry(tmp, struct ptlrpc_request, rq_list);
list_del(&req->rq_list);
- LASSERT (req->rq_reply_state);
- lustre_free_reply_state(req->rq_reply_state);
+ LASSERT (req->rq_reply_state == 0);
target_release_saved_req(req);
}
- list_for_each_safe(tmp, n, &obd->obd_recovery_queue) {
+ list_for_each_safe(tmp, n, &obd->obd_lock_replay_queue) {
+ req = list_entry(tmp, struct ptlrpc_request, rq_list);
+ list_del(&req->rq_list);
+ LASSERT (req->rq_reply_state == 0);
+ target_release_saved_req(req);
+ }
+ list_for_each_safe(tmp, n, &obd->obd_final_req_queue) {
req = list_entry(tmp, struct ptlrpc_request, rq_list);
list_del(&req->rq_list);
LASSERT (req->rq_reply_state == 0);
@@ -991,6 +1006,7 @@ void target_cleanup_recovery(struct obd_device *obd)
}
}
+#if 0
static void target_abort_recovery(void *data)
{
struct obd_device *obd = data;
@@ -1006,11 +1022,11 @@ static void target_abort_recovery(void *data)
target_finish_recovery(obd);
ptlrpc_run_recovery_over_upcall(obd);
}
+#endif
static void target_recovery_expired(unsigned long castmeharder)
{
struct obd_device *obd = (struct obd_device *)castmeharder;
- CERROR("recovery timed out, aborting\n");
spin_lock_bh(&obd->obd_processing_task_lock);
if (obd->obd_recovering)
obd->obd_abort_recovery = 1;
@@ -1066,8 +1082,8 @@ static int check_for_next_transno(struct obd_device *obd)
__u64 next_transno, req_transno;
spin_lock_bh(&obd->obd_processing_task_lock);
- if (!list_empty(&obd->obd_recovery_queue)) {
- req = list_entry(obd->obd_recovery_queue.next,
+ if (!list_empty(&obd->obd_req_replay_queue)) {
+ req = list_entry(obd->obd_req_replay_queue.next,
struct ptlrpc_request, rq_list);
req_transno = req->rq_reqmsg->transno;
} else {
@@ -1076,7 +1092,7 @@ static int check_for_next_transno(struct obd_device *obd)
max = obd->obd_max_recoverable_clients;
connected = obd->obd_connected_clients;
- completed = max - obd->obd_recoverable_clients;
+ completed = max - atomic_read(&obd->obd_req_replay_clients);
queue_len = obd->obd_requests_queued_for_recovery;
next_transno = obd->obd_next_recovery_transno;
@@ -1086,7 +1102,7 @@ static int check_for_next_transno(struct obd_device *obd)
if (obd->obd_abort_recovery) {
CDEBUG(D_HA, "waking for aborted recovery\n");
wake_up = 1;
- } else if (max == completed) {
+ } else if (atomic_read(&obd->obd_req_replay_clients) == 0) {
CDEBUG(D_HA, "waking for completed recovery\n");
wake_up = 1;
} else if (req_transno == next_transno) {
@@ -1120,8 +1136,8 @@ target_next_replay_req(struct obd_device *obd)
spin_lock_bh(&obd->obd_processing_task_lock);
if (obd->obd_abort_recovery) {
req = NULL;
- } else if (!list_empty(&obd->obd_recovery_queue)) {
- req = list_entry(obd->obd_recovery_queue.next,
+ } else if (!list_empty(&obd->obd_req_replay_queue)) {
+ req = list_entry(obd->obd_req_replay_queue.next,
struct ptlrpc_request, rq_list);
list_del_init(&req->rq_list);
obd->obd_requests_queued_for_recovery--;
@@ -1132,11 +1148,90 @@ target_next_replay_req(struct obd_device *obd)
return req;
}
+static int check_for_next_lock(struct obd_device *obd)
+{
+ struct ptlrpc_request *req = NULL;
+ int wake_up = 0;
+
+ spin_lock_bh(&obd->obd_processing_task_lock);
+ if (!list_empty(&obd->obd_lock_replay_queue)) {
+ req = list_entry(obd->obd_lock_replay_queue.next,
+ struct ptlrpc_request, rq_list);
+ CDEBUG(D_HA, "waking for next lock\n");
+ wake_up = 1;
+ } else if (atomic_read(&obd->obd_lock_replay_clients) == 0) {
+ CDEBUG(D_HA, "waking for completed lock replay\n");
+ wake_up = 1;
+ } else if (obd->obd_abort_recovery) {
+ CDEBUG(D_HA, "waking for aborted recovery\n");
+ wake_up = 1;
+ }
+ spin_unlock_bh(&obd->obd_processing_task_lock);
+
+ return wake_up;
+}
+
+static struct ptlrpc_request *
+target_next_replay_lock(struct obd_device *obd)
+{
+ struct l_wait_info lwi = { 0 };
+ struct ptlrpc_request *req;
+
+ CDEBUG(D_HA, "Waiting for lock\n");
+ l_wait_event(obd->obd_next_transno_waitq,
+ check_for_next_lock(obd), &lwi);
+
+ spin_lock_bh(&obd->obd_processing_task_lock);
+ if (obd->obd_abort_recovery) {
+ req = NULL;
+ } else if (!list_empty(&obd->obd_lock_replay_queue)) {
+ req = list_entry(obd->obd_lock_replay_queue.next,
+ struct ptlrpc_request, rq_list);
+ list_del_init(&req->rq_list);
+ } else {
+ req = NULL;
+ }
+ spin_unlock_bh(&obd->obd_processing_task_lock);
+ return req;
+}
+
+static struct ptlrpc_request *
+target_next_final_ping(struct obd_device *obd)
+{
+ struct ptlrpc_request *req;
+
+ spin_lock_bh(&obd->obd_processing_task_lock);
+ if (!list_empty(&obd->obd_final_req_queue)) {
+ req = list_entry(obd->obd_final_req_queue.next,
+ struct ptlrpc_request, rq_list);
+ list_del_init(&req->rq_list);
+ } else {
+ req = NULL;
+ }
+ spin_unlock_bh(&obd->obd_processing_task_lock);
+ return req;
+}
+
+static int req_replay_done(struct obd_export *exp)
+{
+ if (exp->exp_req_replay_needed)
+ return 0;
+ return 1;
+}
+
+static int lock_replay_done(struct obd_export *exp)
+{
+ if (exp->exp_lock_replay_needed)
+ return 0;
+ return 1;
+}
+
static int target_recovery_thread(void *arg)
{
struct obd_device *obd = arg;
struct ptlrpc_request *req;
struct target_recovery_data *trd = &obd->obd_recovery_data;
+ char peer_str[PTL_NALFMT_SIZE];
unsigned long flags;
ENTRY;
@@ -1154,40 +1249,85 @@ static int target_recovery_thread(void *arg)
obd->obd_recovering = 1;
complete(&trd->trd_starting);
- while (obd->obd_recovering) {
+ /* The first stage: replay requests */
+ CWARN("1: request replay stage - %d clients\n",
+ atomic_read(&obd->obd_req_replay_clients));
+ while ((req = target_next_replay_req(obd))) {
LASSERT(trd->trd_processing_task == current->pid);
- req = target_next_replay_req(obd);
- if (req != NULL) {
- char peer_str[PTL_NALFMT_SIZE];
- DEBUG_REQ(D_HA, req, "processing t"LPD64" from %s: ",
- req->rq_reqmsg->transno,
- ptlrpc_peernid2str(&req->rq_peer, peer_str));
- (void)trd->trd_recovery_handler(req);
- obd->obd_replayed_requests++;
- reset_recovery_timer(obd);
- /* bug 1580: decide how to properly sync() in recovery*/
- //mds_fsync_super(mds->mds_sb);
- ptlrpc_free_clone(req);
- spin_lock_bh(&obd->obd_processing_task_lock);
- obd->obd_next_recovery_transno++;
- spin_unlock_bh(&obd->obd_processing_task_lock);
- } else {
- /* recovery is over */
- spin_lock_bh(&obd->obd_processing_task_lock);
- obd->obd_recovering = 0;
- target_cancel_recovery_timer(obd);
- if (obd->obd_abort_recovery) {
- obd->obd_abort_recovery = 0;
- spin_unlock_bh(&obd->obd_processing_task_lock);
- target_abort_recovery(obd);
- } else {
- LASSERT(obd->obd_recoverable_clients == 0);
- spin_unlock_bh(&obd->obd_processing_task_lock);
- target_finish_recovery(obd);
- }
- }
+ DEBUG_REQ(D_HA, req, "processing t"LPD64" from %s: ",
+ req->rq_reqmsg->transno,
+ ptlrpc_peernid2str(&req->rq_peer, peer_str));
+ (void)trd->trd_recovery_handler(req);
+ obd->obd_replayed_requests++;
+ reset_recovery_timer(obd);
+ /* bug 1580: decide how to properly sync() in recovery*/
+ //mds_fsync_super(mds->mds_sb);
+ ptlrpc_free_clone(req);
+ spin_lock_bh(&obd->obd_processing_task_lock);
+ obd->obd_next_recovery_transno++;
+ spin_unlock_bh(&obd->obd_processing_task_lock);
}
+ spin_lock_bh(&obd->obd_processing_task_lock);
+ target_cancel_recovery_timer(obd);
+ spin_unlock_bh(&obd->obd_processing_task_lock);
+
+ /* If some clients haven't replayed requests in time, evict them */
+ if (obd->obd_abort_recovery) {
+ int stale;
+ CERROR("req replay timed out, aborting ...\n");
+ obd->obd_abort_recovery = 0;
+ stale = class_disconnect_stale_exports(obd, req_replay_done, 0);
+ atomic_sub(stale, &obd->obd_lock_replay_clients);
+ abort_req_replay_queue(obd);
+ }
+
+ /* The second stage: replay locks */
+ CWARN("2: lock replay stage - %d clients\n",
+ atomic_read(&obd->obd_lock_replay_clients));
+ while ((req = target_next_replay_lock(obd))) {
+ LASSERT(trd->trd_processing_task == current->pid);
+ DEBUG_REQ(D_HA, req, "processing lock from %s: ",
+ ptlrpc_peernid2str(&req->rq_peer, peer_str));
+ (void)trd->trd_recovery_handler(req);
+ reset_recovery_timer(obd);
+ ptlrpc_free_clone(req);
+ obd->obd_replayed_locks++;
+ }
+
+ spin_lock_bh(&obd->obd_processing_task_lock);
+ target_cancel_recovery_timer(obd);
+ spin_unlock_bh(&obd->obd_processing_task_lock);
+
+ /* If some clients haven't replayed requests in time, evict them */
+ if (obd->obd_abort_recovery) {
+ int stale;
+ CERROR("lock replay timed out, aborting ...\n");
+ obd->obd_abort_recovery = 0;
+ stale = class_disconnect_stale_exports(obd, lock_replay_done, 0);
+ abort_lock_replay_queue(obd);
+ }
+
+ /* We drop recoverying flag to forward all new requests
+ * to regular mds_handle() since now */
+ spin_lock_bh(&obd->obd_processing_task_lock);
+ obd->obd_recovering = 0;
+ spin_unlock_bh(&obd->obd_processing_task_lock);
+
+ /* The third stage: reply on final pings */
+ CWARN("3: final stage - process recovery completion pings\n");
+ while ((req = target_next_final_ping(obd))) {
+ LASSERT(trd->trd_processing_task == current->pid);
+ DEBUG_REQ(D_HA, req, "processing final ping from %s: ",
+ ptlrpc_peernid2str(&req->rq_peer, peer_str));
+ (void)trd->trd_recovery_handler(req);
+ ptlrpc_free_clone(req);
+ }
+
+ CWARN("4: recovery completed - %d/%d reqs/locks replayed\n",
+ obd->obd_replayed_requests, obd->obd_replayed_locks);
+ target_finish_recovery(obd);
+
trd->trd_processing_task = 0;
complete(&trd->trd_finishing);
return 0;
@@ -1227,6 +1367,43 @@ void target_stop_recovery_thread(struct obd_device *obd)
}
#endif
+int target_process_req_flags(struct obd_device *obd, struct ptlrpc_request *req)
+{
+ struct obd_export *exp = req->rq_export;
+ LASSERT(exp != NULL);
+ if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_REQ_REPLAY_DONE) {
+ /* client declares he's ready to replay locks */
+ spin_lock_bh(&obd->obd_processing_task_lock);
+ if (exp->exp_req_replay_needed) {
+ LASSERT(atomic_read(&obd->obd_req_replay_clients) > 0);
+ exp->exp_req_replay_needed = 0;
+ atomic_dec(&obd->obd_req_replay_clients);
+ if (atomic_read(&obd->obd_req_replay_clients) == 0) {
+ CDEBUG(D_HA, "all clients have replayed reqs\n");
+ wake_up(&obd->obd_next_transno_waitq);
+ }
+ }
+ spin_unlock_bh(&obd->obd_processing_task_lock);
+ }
+ if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_LOCK_REPLAY_DONE) {
+ /* client declares he's ready to complete recovery
+ * so, we put the request on th final queue */
+ spin_lock_bh(&obd->obd_processing_task_lock);
+ if (exp->exp_lock_replay_needed) {
+ LASSERT(atomic_read(&obd->obd_lock_replay_clients) > 0);
+ exp->exp_lock_replay_needed = 0;
+ atomic_dec(&obd->obd_lock_replay_clients);
+ if (atomic_read(&obd->obd_lock_replay_clients) == 0) {
+ CDEBUG(D_HA, "all clients have replayed locks\n");
+ wake_up(&obd->obd_next_transno_waitq);
+ }
+ }
+ spin_unlock_bh(&obd->obd_processing_task_lock);
+ }
+
+ return 0;
+}
+
int target_queue_recovery_request(struct ptlrpc_request *req,
struct obd_device *obd)
{
@@ -1234,6 +1411,39 @@ int target_queue_recovery_request(struct ptlrpc_request *req,
int inserted = 0;
__u64 transno = req->rq_reqmsg->transno;
+ if (obd->obd_recovery_data.trd_processing_task == current->pid) {
+ /* Processing the queue right now, don't re-add. */
+ return 1;
+ }
+
+ target_process_req_flags(obd, req);
+
+ if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_LOCK_REPLAY_DONE) {
+ /* client declares he's ready to complete recovery
+ * so, we put the request on th final queue */
+ req = ptlrpc_clone_req(req);
+ if (req == NULL)
+ return -ENOMEM;
+ DEBUG_REQ(D_HA, req, "queue final req");
+ spin_lock_bh(&obd->obd_processing_task_lock);
+ list_add_tail(&req->rq_list, &obd->obd_final_req_queue);
+ spin_unlock_bh(&obd->obd_processing_task_lock);
+ return 0;
+ }
+ if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_REQ_REPLAY_DONE) {
+ /* client declares he's ready to replay locks */
+ req = ptlrpc_clone_req(req);
+ if (req == NULL)
+ return -ENOMEM;
+ DEBUG_REQ(D_HA, req, "queue lock replay req");
+ spin_lock_bh(&obd->obd_processing_task_lock);
+ list_add_tail(&req->rq_list, &obd->obd_lock_replay_queue);
+ spin_unlock_bh(&obd->obd_processing_task_lock);
+ wake_up(&obd->obd_next_transno_waitq);
+ return 0;
+ }
+
+
/* CAVEAT EMPTOR: The incoming request message has been swabbed
* (i.e. buflens etc are in my own byte order), but type-dependent
* buffers (eg mds_body, ost_body etc) have NOT been swabbed. */
@@ -1256,8 +1466,7 @@ int target_queue_recovery_request(struct ptlrpc_request *req,
* handled will pass through here and be processed immediately.
*/
spin_lock_bh(&obd->obd_processing_task_lock);
- if (obd->obd_recovery_data.trd_processing_task == current->pid ||
- transno < obd->obd_next_recovery_transno) {
+ if (transno < obd->obd_next_recovery_transno) {
/* Processing the queue right now, don't re-add. */
LASSERT(list_empty(&req->rq_list));
spin_unlock_bh(&obd->obd_processing_task_lock);
@@ -1280,7 +1489,7 @@ int target_queue_recovery_request(struct ptlrpc_request *req,
spin_lock_bh(&obd->obd_processing_task_lock);
/* XXX O(n^2) */
- list_for_each(tmp, &obd->obd_recovery_queue) {
+ list_for_each(tmp, &obd->obd_req_replay_queue) {
struct ptlrpc_request *reqiter =
list_entry(tmp, struct ptlrpc_request, rq_list);
@@ -1292,7 +1501,7 @@ int target_queue_recovery_request(struct ptlrpc_request *req,
}
if (!inserted)
- list_add_tail(&req->rq_list, &obd->obd_recovery_queue);
+ list_add_tail(&req->rq_list, &obd->obd_req_replay_queue);
obd->obd_requests_queued_for_recovery++;
wake_up(&obd->obd_next_transno_waitq);
@@ -1305,41 +1514,6 @@ struct obd_device * target_req2obd(struct ptlrpc_request *req)
return req->rq_export->exp_obd;
}
-int target_queue_final_reply(struct ptlrpc_request *req, int rc)
-{
- struct obd_device *obd = target_req2obd(req);
-
- LASSERT ((rc == 0) == (req->rq_reply_state != NULL));
-
- if (rc) {
- /* Just like ptlrpc_error, but without the sending. */
- rc = lustre_pack_reply(req, 0, NULL, NULL);
- LASSERT(rc == 0); /* XXX handle this */
- req->rq_type = PTL_RPC_MSG_ERR;
- }
-
- LASSERT (!req->rq_reply_state->rs_difficult);
- LASSERT(list_empty(&req->rq_list));
-
- req = ptlrpc_clone_req(req);
-
- spin_lock_bh(&obd->obd_processing_task_lock);
-
- list_add(&req->rq_list, &obd->obd_delayed_reply_queue);
-
- /* only count the first "replay over" request from each
- export */
- if (req->rq_export->exp_replay_needed) {
- --obd->obd_recoverable_clients;
- req->rq_export->exp_replay_needed = 0;
- CWARN("%s: %d recoverable clients remain\n",
- obd->obd_name, obd->obd_recoverable_clients);
- }
- wake_up(&obd->obd_next_transno_waitq);
- spin_unlock_bh(&obd->obd_processing_task_lock);
- return 1;
-}
-
int
target_send_reply_msg (struct ptlrpc_request *req, int rc, int fail_id)
{
diff --git a/lustre/ldlm/ldlm_lockd.c b/lustre/ldlm/ldlm_lockd.c
index 0634cb7b04..952112beb6 100644
--- a/lustre/ldlm/ldlm_lockd.c
+++ b/lustre/ldlm/ldlm_lockd.c
@@ -399,6 +399,9 @@ int ldlm_server_blocking_ast(struct ldlm_lock *lock,
RETURN(0);
}
+ LASSERTF(lock->l_export->exp_obd->obd_recovering == 0,
+ "BUG 6063: lock collide during recovery");
+
LASSERT(lock);
l_lock(&lock->l_resource->lr_namespace->ns_lock);
@@ -1676,4 +1679,3 @@ EXPORT_SYMBOL(target_send_reply);
EXPORT_SYMBOL(target_queue_recovery_request);
EXPORT_SYMBOL(target_handle_ping);
EXPORT_SYMBOL(target_handle_disconnect);
-EXPORT_SYMBOL(target_queue_final_reply);
diff --git a/lustre/ldlm/ldlm_request.c b/lustre/ldlm/ldlm_request.c
index 90c988a73f..012b31eb7c 100644
--- a/lustre/ldlm/ldlm_request.c
+++ b/lustre/ldlm/ldlm_request.c
@@ -1036,6 +1036,11 @@ static int replay_one_lock(struct obd_import *imp, struct ldlm_lock *lock)
size[1] = lock->l_lvb_len;
}
req->rq_replen = lustre_msg_size(buffers, size);
+ /* notify the server we've replayed all requests.
+ * also, we mark the request to be put on a dedicated
+ * queue to be processed after all request replayes.
+ * bug 6063 */
+ lustre_msg_set_flags(req->rq_reqmsg, MSG_REQ_REPLAY_DONE);
LDLM_DEBUG(lock, "replaying lock:");
diff --git a/lustre/mds/handler.c b/lustre/mds/handler.c
index c6e48cf2fb..ecfc6a208e 100644
--- a/lustre/mds/handler.c
+++ b/lustre/mds/handler.c
@@ -2823,14 +2823,6 @@ int mds_handle(struct ptlrpc_request *req)
}
out:
- if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_LAST_REPLAY) {
- if (obd && obd->obd_recovering) {
- DEBUG_REQ(D_HA, req, "LAST_REPLAY, queuing reply");
- return target_queue_final_reply(req, rc);
- }
- /* Lost a race with recovery; let the error path DTRT. */
- rc = req->rq_status = -ENOTCONN;
- }
target_send_reply(req, rc, fail);
return 0;
diff --git a/lustre/mds/mds_fs.c b/lustre/mds/mds_fs.c
index 391654a4a4..b8bb10007c 100644
--- a/lustre/mds/mds_fs.c
+++ b/lustre/mds/mds_fs.c
@@ -380,9 +380,17 @@ static int mds_read_last_rcvd(struct obd_device *obd, struct file *file)
spin_lock_init(&med->med_open_lock);
mcd = NULL;
- exp->exp_replay_needed = 1;
+ exp->exp_req_replay_needed = 1;
obd->obd_recoverable_clients++;
obd->obd_max_recoverable_clients++;
+
+ /* track clients to separate req replay
+ * from lock replay. bug 6063 */
+ atomic_inc(&obd->obd_req_replay_clients);
+ exp->exp_req_replay_needed = 1;
+ atomic_inc(&obd->obd_lock_replay_clients);
+ exp->exp_lock_replay_needed = 1;
+
class_export_put(exp);
CDEBUG(D_OTHER, "client at idx %d has last_transno = "LPU64"\n",
diff --git a/lustre/obdclass/genops.c b/lustre/obdclass/genops.c
index 36ae1e8d3b..596bc0b6d4 100644
--- a/lustre/obdclass/genops.c
+++ b/lustre/obdclass/genops.c
@@ -738,7 +738,9 @@ void class_disconnect_exports(struct obd_device *obd, unsigned long flags)
/* Remove exports that have not completed recovery.
*/
-void class_disconnect_stale_exports(struct obd_device *obd, unsigned long flags)
+int class_disconnect_stale_exports(struct obd_device *obd,
+ int (*test_export)(struct obd_export *),
+ unsigned long flags)
{
struct list_head work_list;
struct list_head *pos, *n;
@@ -750,10 +752,12 @@ void class_disconnect_stale_exports(struct obd_device *obd, unsigned long flags)
spin_lock(&obd->obd_dev_lock);
list_for_each_safe(pos, n, &obd->obd_exports) {
exp = list_entry(pos, struct obd_export, exp_obd_chain);
- if (exp->exp_replay_needed) {
+ if (!test_export(exp)) {
list_del(&exp->exp_obd_chain);
list_add(&exp->exp_obd_chain, &work_list);
cnt++;
+ CDEBUG(D_ERROR, "%s: disconnect stale client %s\n",
+ obd->obd_name, exp->exp_client_uuid.uuid);
}
}
spin_unlock(&obd->obd_dev_lock);
@@ -761,7 +765,7 @@ void class_disconnect_stale_exports(struct obd_device *obd, unsigned long flags)
CDEBUG(D_ERROR, "%s: disconnecting %d stale clients\n",
obd->obd_name, cnt);
class_disconnect_export_list(&work_list, flags);
- EXIT;
+ RETURN(cnt);
}
diff --git a/lustre/obdclass/obd_config.c b/lustre/obdclass/obd_config.c
index 414fde4f5c..c27c6ae5d6 100644
--- a/lustre/obdclass/obd_config.c
+++ b/lustre/obdclass/obd_config.c
@@ -116,8 +116,9 @@ static int class_attach(struct lustre_cfg *lcfg)
init_timer(&obd->obd_recovery_timer);
spin_lock_init(&obd->obd_processing_task_lock);
init_waitqueue_head(&obd->obd_next_transno_waitq);
- INIT_LIST_HEAD(&obd->obd_recovery_queue);
- INIT_LIST_HEAD(&obd->obd_delayed_reply_queue);
+ INIT_LIST_HEAD(&obd->obd_req_replay_queue);
+ INIT_LIST_HEAD(&obd->obd_lock_replay_queue);
+ INIT_LIST_HEAD(&obd->obd_final_req_queue);
spin_lock_init(&obd->obd_uncommitted_replies_lock);
INIT_LIST_HEAD(&obd->obd_uncommitted_replies);
diff --git a/lustre/obdfilter/filter.c b/lustre/obdfilter/filter.c
index c5dbac89ec..43a778d50d 100644
--- a/lustre/obdfilter/filter.c
+++ b/lustre/obdfilter/filter.c
@@ -499,7 +499,10 @@ static int filter_init_server_data(struct obd_device *obd, struct file * filp)
spin_lock_init(&fed->fed_lock);
fcd = NULL;
- exp->exp_replay_needed = 1;
+ exp->exp_req_replay_needed = 1;
+ exp->exp_lock_replay_needed = 1;
+ atomic_inc(&obd->obd_req_replay_clients);
+ atomic_inc(&obd->obd_lock_replay_clients);
obd->obd_recoverable_clients++;
obd->obd_max_recoverable_clients++;
class_export_put(exp);
diff --git a/lustre/ost/ost_handler.c b/lustre/ost/ost_handler.c
index 1090944866..cf1c6de90d 100644
--- a/lustre/ost/ost_handler.c
+++ b/lustre/ost/ost_handler.c
@@ -1228,15 +1228,6 @@ int ost_handle(struct ptlrpc_request *req)
}
out_check_req:
- if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_LAST_REPLAY) {
- if (obd && obd->obd_recovering) {
- DEBUG_REQ(D_HA, req, "LAST_REPLAY, queuing reply");
- rc = target_queue_final_reply(req, rc);
- GOTO(out_free_oti, rc);
- }
- /* Lost a race with recovery; let the error path DTRT. */
- rc = req->rq_status = -ENOTCONN;
- }
if (!rc)
oti_to_request(oti, req);
diff --git a/lustre/ptlrpc/import.c b/lustre/ptlrpc/import.c
index 7cf82f11dd..52f3587f3f 100644
--- a/lustre/ptlrpc/import.c
+++ b/lustre/ptlrpc/import.c
@@ -589,7 +589,7 @@ static int signal_completed_replay(struct obd_import *imp)
req->rq_replen = lustre_msg_size(0, NULL);
req->rq_send_state = LUSTRE_IMP_REPLAY_WAIT;
- req->rq_reqmsg->flags |= MSG_LAST_REPLAY;
+ req->rq_reqmsg->flags |= MSG_LOCK_REPLAY_DONE | MSG_REQ_REPLAY_DONE;
req->rq_timeout *= 3;
req->rq_interpret_reply = completed_replay_interpret;
--
GitLab