Menu
▾
▴
lustre-cvs
|
From: Phil S. <ps...@us...> - 2003-04-20 15:43:30
|
Update of /cvsroot/lustre/lustre/ldlm
In directory sc8-pr-cvs1:/tmp/cvs-serv21268/ldlm
Modified Files:
Tag: b_devel
ldlm_lockd.c
Log Message:
b=1111
- add ptlrpc_ldlm_hooks_referenced(), so we can determine in
ldlm_cleanup if "force" is needed
- add a little bit more debug information to hunt connection leaks
Index: ldlm_lockd.c
===================================================================
RCS file: /cvsroot/lustre/lustre/ldlm/ldlm_lockd.c,v
retrieving revision 1.131.2.16
retrieving revision 1.131.2.17
diff -u -w -b -B -p -r1.131.2.16 -r1.131.2.17
--- ldlm_lockd.c 14 Apr 2003 20:51:56 -0000 1.131.2.16
+++ ldlm_lockd.c 20 Apr 2003 15:43:26 -0000 1.131.2.17
@@ -851,8 +851,13 @@ static int ldlm_cleanup(struct obd_devic
}
#ifdef __KERNEL__
- if (force)
+ if (force) {
ptlrpc_put_ldlm_hooks();
+ } else if (ptlrpc_ldlm_hooks_referenced()) {
+ CERROR("Some connections weren't cleaned up; run lconf with "
+ "--force to forcibly unload.\n");
+ RETURN(-EBUSY);
+ }
ptlrpc_stop_all_threads(ldlm->ldlm_cb_service);
ptlrpc_unregister_service(ldlm->ldlm_cb_service);
|
|
From: Mike S. <sh...@us...> - 2003-04-22 23:17:02
|
Update of /cvsroot/lustre/lustre/ldlm
In directory sc8-pr-cvs1:/tmp/cvs-serv2542
Modified Files:
Tag: b_devel
ldlm_lockd.c
Log Message:
b=1127: Run expired-lock recovery on a (new) DLM thread, rather than in the
timer context, because it will schedule when it sends ASTs, etc.
Index: ldlm_lockd.c
===================================================================
RCS file: /cvsroot/lustre/lustre/ldlm/ldlm_lockd.c,v
retrieving revision 1.131.2.18
retrieving revision 1.131.2.19
diff -u -w -b -B -p -r1.131.2.18 -r1.131.2.19
--- ldlm_lockd.c 22 Apr 2003 20:06:39 -0000 1.131.2.18
+++ ldlm_lockd.c 22 Apr 2003 23:16:58 -0000 1.131.2.19
@@ -28,6 +28,7 @@
# include <linux/module.h>
# include <linux/slab.h>
# include <linux/init.h>
+# include <linux/wait.h>
#else
# include <liblustre.h>
#endif
@@ -53,6 +53,86 @@ static spinlock_t waiting_locks_spinlock
static struct timer_list waiting_locks_timer;
static int ldlm_already_setup = 0;
+#ifdef __KERNEL__
+
+static struct expired_lock_thread {
+ wait_queue_head_t elt_waitq;
+ int elt_state;
+ struct list_head elt_expired_locks;
+ spinlock_t elt_lock;
+} expired_lock_thread;
+
+#define ELT_STOPPED 0
+#define ELT_READY 1
+#define ELT_TERMINATE 2
+
+static inline int have_expired_locks(void)
+{
+ int need_to_run;
+
+ spin_lock_bh(&expired_lock_thread.elt_lock);
+ need_to_run = !list_empty(&expired_lock_thread.elt_expired_locks);
+ spin_unlock_bh(&expired_lock_thread.elt_lock);
+
+ RETURN(need_to_run);
+}
+
+static int expired_lock_main(void *arg)
+{
+ struct list_head *expired = &expired_lock_thread.elt_expired_locks;
+ struct l_wait_info lwi = { 0 };
+ unsigned long flags;
+
+ ENTRY;
+ lock_kernel();
+ daemonize();
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0)
+ sigfillset(¤t->blocked);
+ recalc_sigpending();
+#else
+ spin_lock_irqsave(¤t->sigmask_lock, flags);
+ sigfillset(¤t->blocked);
+ recalc_sigpending(current);
+ spin_unlock_irqrestore(¤t->sigmask_lock, flags);
+#endif
+
+ strcpy(current->comm, "ldlm_elt");
+ unlock_kernel();
+
+ expired_lock_thread.elt_state = ELT_READY;
+ wake_up(&expired_lock_thread.elt_waitq);
+
+ while (1) {
+ l_wait_event(expired_lock_thread.elt_waitq,
+ have_expired_locks() ||
+ expired_lock_thread.elt_state == ELT_TERMINATE,
+ &lwi);
+
+ spin_lock_bh(&expired_lock_thread.elt_lock);
+ while (!list_empty(expired)) {
+ struct ldlm_lock *lock = list_entry(expired->next,
+ struct ldlm_lock,
+ l_pending_chain);
+ spin_unlock_bh(&expired_lock_thread.elt_lock);
+
+ ptlrpc_fail_export(lock->l_export);
+
+ spin_lock_bh(&expired_lock_thread.elt_lock);
+ }
+ spin_unlock_bh(&expired_lock_thread.elt_lock);
+
+ if (expired_lock_thread.elt_state == ELT_TERMINATE)
+ break;
+ }
+
+ expired_lock_thread.elt_state = ELT_STOPPED;
+ wake_up(&expired_lock_thread.elt_waitq);
+ RETURN(0);
+}
+
+#endif /* __KERNEL__ */
+
static void waiting_locks_callback(unsigned long unused)
{
struct ldlm_lock *lock;
@@ -77,9 +157,11 @@ static void waiting_locks_callback(unsig
LDLM_DEBUG(lock, "timer expired");
/* ptlrpc_fail_export must be called with this lock released */
- spin_unlock_bh(&waiting_locks_spinlock);
- ptlrpc_fail_export(lock->l_export);
- spin_lock_bh(&waiting_locks_spinlock);
+ spin_lock_bh(&expired_lock_thread.elt_lock);
+ list_del(&lock->l_pending_chain);
+ list_add(&lock->l_pending_chain,
+ &expired_lock_thread.elt_expired_locks);
+ spin_unlock_bh(&expired_lock_thread.elt_lock);
}
spin_unlock_bh(&waiting_locks_spinlock);
@@ -109,6 +191,8 @@ static int ldlm_add_waiting_lock(struct
}
list_add_tail(&lock->l_pending_chain, &waiting_locks_list); /* FIFO */
spin_unlock_bh(&waiting_locks_spinlock);
+ /* We drop this ref when we get removed from the list. */
+ class_export_get(lock->l_export);
RETURN(1);
}
@@ -146,6 +230,8 @@ int ldlm_del_waiting_lock(struct ldlm_lo
}
list_del_init(&lock->l_pending_chain);
spin_unlock_bh(&waiting_locks_spinlock);
+ /* We got this ref when we were added to the list. */
+ class_export_put(lock->l_export);
RETURN(1);
}
@@ -816,6 +902,19 @@ static int ldlm_setup(struct obd_device
}
}
+ rc = kernel_thread(expired_lock_main, NULL, CLONE_VM | CLONE_FS);
+ if (rc < 0) {
+ CERROR("Cannot start ldlm expired-lock thread: %d\n", rc);
+ GOTO(out_thread, rc);
+ }
+
+ INIT_LIST_HEAD(&expired_lock_thread.elt_expired_locks);
+ spin_lock_init(&expired_lock_thread.elt_lock);
+ expired_lock_thread.elt_state = ELT_STOPPED;
+ init_waitqueue_head(&expired_lock_thread.elt_waitq);
+
+ wait_event(expired_lock_thread.elt_waitq,
+ expired_lock_thread.elt_state == ELT_READY);
#endif
INIT_LIST_HEAD(&waiting_locks_list);
spin_lock_init(&waiting_locks_spinlock);
@@ -865,6 +964,11 @@ static int ldlm_cleanup(struct obd_devic
ptlrpc_stop_all_threads(ldlm->ldlm_cancel_service);
ptlrpc_unregister_service(ldlm->ldlm_cancel_service);
ldlm_proc_cleanup(obddev);
+
+ expired_lock_thread.elt_state = ELT_TERMINATE;
+ wake_up(&expired_lock_thread.elt_waitq);
+ wait_event(expired_lock_thread.elt_waitq,
+ expired_lock_thread.elt_state == ELT_STOPPED);
inter_module_unregister("ldlm_namespace_cleanup");
inter_module_unregister("ldlm_cli_cancel_unused");
|
|
From: Phil S. <ps...@us...> - 2003-04-23 03:13:38
|
Update of /cvsroot/lustre/lustre/ldlm
In directory sc8-pr-cvs1:/tmp/cvs-serv20701
Modified Files:
Tag: b_devel
ldlm_lockd.c
Log Message:
if we receive an invalid export, print the sending NID
Index: ldlm_lockd.c
===================================================================
RCS file: /cvsroot/lustre/lustre/ldlm/ldlm_lockd.c,v
retrieving revision 1.131.2.19
retrieving revision 1.131.2.20
diff -u -w -b -B -p -r1.131.2.19 -r1.131.2.20
--- ldlm_lockd.c 22 Apr 2003 23:16:58 -0000 1.131.2.19
+++ ldlm_lockd.c 23 Apr 2003 03:13:31 -0000 1.131.2.20
@@ -681,9 +681,10 @@ static int ldlm_callback_handler(struct
if (req->rq_export == NULL) {
struct ldlm_request *dlm_req;
- CERROR("operation %d with bad export (ptl req %d/rep %d)\n",
- req->rq_reqmsg->opc, req->rq_request_portal,
- req->rq_reply_portal);
+ CERROR("operation %d from nid 0x%x with bad export "
+ "(ptl req %d/rep %d)\n", req->rq_reqmsg->opc,
+ req->rq_connection->c_peer.peer_nid,
+ req->rq_request_portal, req->rq_reply_portal);
CERROR("--> export addr: "LPX64", cookie: "LPX64"\n",
req->rq_reqmsg->handle.addr,
req->rq_reqmsg->handle.cookie);
|
|
From: Phil S. <ps...@us...> - 2003-04-23 03:14:27
|
Update of /cvsroot/lustre/lustre/ldlm
In directory sc8-pr-cvs1:/tmp/cvs-serv20898
Modified Files:
Tag: b_devel
ldlm_lockd.c
Log Message:
oops, NID is LPU64
Index: ldlm_lockd.c
===================================================================
RCS file: /cvsroot/lustre/lustre/ldlm/ldlm_lockd.c,v
retrieving revision 1.131.2.20
retrieving revision 1.131.2.21
diff -u -w -b -B -p -r1.131.2.20 -r1.131.2.21
--- ldlm_lockd.c 23 Apr 2003 03:13:31 -0000 1.131.2.20
+++ ldlm_lockd.c 23 Apr 2003 03:14:17 -0000 1.131.2.21
@@ -681,7 +681,7 @@ static int ldlm_callback_handler(struct
if (req->rq_export == NULL) {
struct ldlm_request *dlm_req;
- CERROR("operation %d from nid 0x%x with bad export "
+ CERROR("operation %d from nid "LPU64" with bad export "
"(ptl req %d/rep %d)\n", req->rq_reqmsg->opc,
req->rq_connection->c_peer.peer_nid,
req->rq_request_portal, req->rq_reply_portal);
|
|
From: Mike S. <sh...@us...> - 2003-04-23 16:23:30
|
Update of /cvsroot/lustre/lustre/ldlm
In directory sc8-pr-cvs1:/tmp/cvs-serv30306
Modified Files:
Tag: b_devel
ldlm_lockd.c
Log Message:
Important: initializing the expired_lock_thread structures _before_ starting
the thread.
Index: ldlm_lockd.c
===================================================================
RCS file: /cvsroot/lustre/lustre/ldlm/ldlm_lockd.c,v
retrieving revision 1.131.2.21
retrieving revision 1.131.2.22
diff -u -w -b -B -p -r1.131.2.21 -r1.131.2.22
--- ldlm_lockd.c 23 Apr 2003 03:14:17 -0000 1.131.2.21
+++ ldlm_lockd.c 23 Apr 2003 16:23:26 -0000 1.131.2.22
@@ -42,6 +42,10 @@ extern struct list_head ldlm_namespace_l
extern int (*mds_reint_p)(int offset, struct ptlrpc_request *req);
extern int (*mds_getattr_name_p)(int offset, struct ptlrpc_request *req);
+static int ldlm_already_setup = 0;
+
+#ifdef __KERNEL__
+
inline unsigned long round_timeout(unsigned long timeout)
{
return ((timeout / HZ) + 1) * HZ;
@@ -51,9 +55,6 @@ inline unsigned long round_timeout(unsig
static struct list_head waiting_locks_list;
static spinlock_t waiting_locks_spinlock;
static struct timer_list waiting_locks_timer;
-static int ldlm_already_setup = 0;
-
-#ifdef __KERNEL__
static struct expired_lock_thread {
wait_queue_head_t elt_waitq;
@@ -131,8 +132,6 @@ static int expired_lock_main(void *arg)
RETURN(0);
}
-#endif /* __KERNEL__ */
-
static void waiting_locks_callback(unsigned long unused)
{
struct ldlm_lock *lock;
@@ -235,6 +234,20 @@ int ldlm_del_waiting_lock(struct ldlm_lo
RETURN(1);
}
+#else /* !__KERNEL__ */
+
+static int ldlm_add_waiting_lock(struct ldlm_lock *lock)
+{
+ RETURN(1);
+}
+
+int ldlm_del_waiting_lock(struct ldlm_lock *lock)
+{
+ RETURN(0);
+}
+
+#endif /* __KERNEL__ */
+
static inline void ldlm_failed_ast(struct ldlm_lock *lock, int rc,
char *ast_type)
{
@@ -903,16 +916,16 @@ static int ldlm_setup(struct obd_device
}
}
+ INIT_LIST_HEAD(&expired_lock_thread.elt_expired_locks);
+ spin_lock_init(&expired_lock_thread.elt_lock);
+ expired_lock_thread.elt_state = ELT_STOPPED;
+ init_waitqueue_head(&expired_lock_thread.elt_waitq);
+
rc = kernel_thread(expired_lock_main, NULL, CLONE_VM | CLONE_FS);
if (rc < 0) {
CERROR("Cannot start ldlm expired-lock thread: %d\n", rc);
GOTO(out_thread, rc);
}
-
- INIT_LIST_HEAD(&expired_lock_thread.elt_expired_locks);
- spin_lock_init(&expired_lock_thread.elt_lock);
- expired_lock_thread.elt_state = ELT_STOPPED;
- init_waitqueue_head(&expired_lock_thread.elt_waitq);
wait_event(expired_lock_thread.elt_waitq,
expired_lock_thread.elt_state == ELT_READY);
|
|
From: Phil S. <ps...@us...> - 2003-04-23 22:12:12
|
Update of /cvsroot/lustre/lustre/ldlm
In directory sc8-pr-cvs1:/tmp/cvs-serv5760
Modified Files:
Tag: b_devel
ldlm_lockd.c
Log Message:
Make "ldlm_del_waiting_lock" a no-op on the client-side, because there
is no waiting locks list.
Index: ldlm_lockd.c
===================================================================
RCS file: /cvsroot/lustre/lustre/ldlm/ldlm_lockd.c,v
retrieving revision 1.131.2.22
retrieving revision 1.131.2.23
diff -u -w -b -B -p -r1.131.2.22 -r1.131.2.23
--- ldlm_lockd.c 23 Apr 2003 16:23:26 -0000 1.131.2.22
+++ ldlm_lockd.c 23 Apr 2003 22:12:07 -0000 1.131.2.23
@@ -203,8 +203,12 @@ static int ldlm_add_waiting_lock(struct
int ldlm_del_waiting_lock(struct ldlm_lock *lock)
{
struct list_head *list_next;
-
ENTRY;
+
+ if (lock->l_export == NULL) {
+ /* We don't have a "waiting locks list" on clients. */
+ RETURN(0);
+ }
spin_lock_bh(&waiting_locks_spinlock);
|
|
From: Robert R. <rr...@us...> - 2003-04-24 01:38:21
|
Update of /cvsroot/lustre/lustre/ldlm
In directory sc8-pr-cvs1:/tmp/cvs-serv13228/ldlm
Modified Files:
Tag: b_devel
ldlm_lockd.c
Log Message:
b=1075
* added --failover option to lconf and lctl cleanup
* failover flag added to obd_disconnect and obd_cleanup
* if failover ==1, then MDS and OST will not update the client
state in last_rcvd.
* Also changes to lconf to add the --group support to MDS.
Index: ldlm_lockd.c
===================================================================
RCS file: /cvsroot/lustre/lustre/ldlm/ldlm_lockd.c,v
retrieving revision 1.131.2.23
retrieving revision 1.131.2.24
diff -u -w -b -B -p -r1.131.2.23 -r1.131.2.24
--- ldlm_lockd.c 23 Apr 2003 22:12:07 -0000 1.131.2.23
+++ ldlm_lockd.c 24 Apr 2003 01:37:48 -0000 1.131.2.24
@@ -957,7 +957,7 @@ static int ldlm_setup(struct obd_device
return rc;
}
-static int ldlm_cleanup(struct obd_device *obddev, int force)
+static int ldlm_cleanup(struct obd_device *obddev, int force, int failover)
{
struct ldlm_obd *ldlm = &obddev->u.ldlm;
ENTRY;
|
|
From: Mei <me...@us...> - 2003-04-26 09:25:32
|
Update of /cvsroot/lustre/lustre/ldlm
In directory sc8-pr-cvs1:/tmp/cvs-serv27150
Modified Files:
Tag: b_devel
ldlm_lockd.c
Log Message:
fix compilation for liblustre
Index: ldlm_lockd.c
===================================================================
RCS file: /cvsroot/lustre/lustre/ldlm/ldlm_lockd.c,v
retrieving revision 1.131.2.24
retrieving revision 1.131.2.25
diff -u -w -b -B -p -r1.131.2.24 -r1.131.2.25
--- ldlm_lockd.c 24 Apr 2003 01:37:48 -0000 1.131.2.24
+++ ldlm_lockd.c 26 Apr 2003 09:25:22 -0000 1.131.2.25
@@ -933,12 +933,13 @@ static int ldlm_setup(struct obd_device
wait_event(expired_lock_thread.elt_waitq,
expired_lock_thread.elt_state == ELT_READY);
-#endif
+
INIT_LIST_HEAD(&waiting_locks_list);
spin_lock_init(&waiting_locks_spinlock);
waiting_locks_timer.function = waiting_locks_callback;
waiting_locks_timer.data = 0;
init_timer(&waiting_locks_timer);
+#endif
ldlm_already_setup = 1;
|
|
From: Phil S. <ps...@us...> - 2003-05-05 21:35:18
|
Update of /cvsroot/lustre/lustre/ldlm
In directory sc8-pr-cvs1:/tmp/cvs-serv10244
Modified Files:
Tag: b_devel
ldlm_lockd.c
Log Message:
Print rq_reqpmsg->status instead of rq_status, to get useful error code
Index: ldlm_lockd.c
===================================================================
RCS file: /cvsroot/lustre/lustre/ldlm/ldlm_lockd.c,v
retrieving revision 1.131.2.26
retrieving revision 1.131.2.27
diff -u -w -b -B -p -r1.131.2.26 -r1.131.2.27
--- ldlm_lockd.c 27 Apr 2003 17:21:48 -0000 1.131.2.26
+++ ldlm_lockd.c 5 May 2003 21:35:14 -0000 1.131.2.27
@@ -328,7 +328,7 @@ int ldlm_server_blocking_ast(struct ldlm
ldlm_failed_ast(lock, rc, "blocking");
} else if (rc) {
CERROR("client returned %d from blocking AST for lock %p\n",
- req->rq_status, lock);
+ req->rq_repmsg->status, lock);
LDLM_DEBUG(lock, "client returned error %d from blocking AST",
req->rq_status);
ldlm_lock_cancel(lock);
|
|
From: Robert R. <rr...@us...> - 2003-05-13 00:46:12
|
Update of /cvsroot/lustre/lustre/ldlm
In directory sc8-pr-cvs1:/tmp/cvs-serv25471/ldlm
Modified Files:
Tag: b_devel
ldlm_lockd.c
Log Message:
- add ptlrpc_dump_connections() for debugging ldlm cleanup problems
- cleanup the import's export ptlrpc_import_connect if the connect fails.
this prevents an obd_refcount clean on the client during a failed mount.
Index: ldlm_lockd.c
===================================================================
RCS file: /cvsroot/lustre/lustre/ldlm/ldlm_lockd.c,v
retrieving revision 1.131.2.27
retrieving revision 1.131.2.28
diff -u -w -b -B -p -r1.131.2.27 -r1.131.2.28
--- ldlm_lockd.c 5 May 2003 21:35:14 -0000 1.131.2.27
+++ ldlm_lockd.c 13 May 2003 00:45:59 -0000 1.131.2.28
@@ -972,6 +972,7 @@ static int ldlm_cleanup(struct obd_devic
} else if (ptlrpc_ldlm_hooks_referenced()) {
CERROR("Some connections weren't cleaned up; run lconf with "
"--force to forcibly unload.\n");
+ ptlrpc_dump_connections();
RETURN(-EBUSY);
}
|
|
From: Hariharan T. <th...@us...> - 2003-05-16 05:56:15
|
Update of /cvsroot/lustre/lustre/ldlm
In directory sc8-pr-cvs1:/tmp/cvs-serv12069/ldlm
Modified Files:
Tag: b_devel
ldlm_lockd.c
Log Message:
Added lprocfs counters for obd_ops stats, obdfilter and ptlrpc services. Bugzilla#1107.
Index: ldlm_lockd.c
===================================================================
RCS file: /cvsroot/lustre/lustre/ldlm/ldlm_lockd.c,v
retrieving revision 1.131.2.28
retrieving revision 1.131.2.29
diff -u -w -b -B -p -r1.131.2.28 -r1.131.2.29
--- ldlm_lockd.c 13 May 2003 00:45:59 -0000 1.131.2.28
+++ ldlm_lockd.c 16 May 2003 05:55:40 -0000 1.131.2.29
@@ -876,7 +876,7 @@ static int ldlm_setup(struct obd_device
ptlrpc_init_svc(LDLM_NEVENTS, LDLM_NBUFS, LDLM_BUFSIZE,
LDLM_MAXREQSIZE, LDLM_CB_REQUEST_PORTAL,
LDLM_CB_REPLY_PORTAL,
- ldlm_callback_handler, "ldlm_cbd");
+ ldlm_callback_handler, "ldlm_cbd", obddev);
if (!ldlm->ldlm_cb_service) {
CERROR("failed to start service\n");
@@ -887,7 +887,7 @@ static int ldlm_setup(struct obd_device
ptlrpc_init_svc(LDLM_NEVENTS, LDLM_NBUFS, LDLM_BUFSIZE,
LDLM_MAXREQSIZE, LDLM_CANCEL_REQUEST_PORTAL,
LDLM_CANCEL_REPLY_PORTAL,
- ldlm_cancel_handler, "ldlm_canceld");
+ ldlm_cancel_handler, "ldlm_canceld", obddev);
if (!ldlm->ldlm_cancel_service) {
CERROR("failed to start service\n");
|
|
From: Mike S. <sh...@us...> - 2003-05-16 14:49:50
|
Update of /cvsroot/lustre/lustre/ldlm
In directory sc8-pr-cvs1:/tmp/cvs-serv28479
Modified Files:
Tag: b_devel
ldlm_lockd.c
Log Message:
b=1289: need to wake the expired-locks thread when we add an expired lock to
its work list.
Index: ldlm_lockd.c
===================================================================
RCS file: /cvsroot/lustre/lustre/ldlm/ldlm_lockd.c,v
retrieving revision 1.131.2.29
retrieving revision 1.131.2.30
diff -u -w -b -B -p -r1.131.2.29 -r1.131.2.30
--- ldlm_lockd.c 16 May 2003 05:55:40 -0000 1.131.2.29
+++ ldlm_lockd.c 16 May 2003 14:49:46 -0000 1.131.2.30
@@ -155,12 +155,12 @@ static void waiting_locks_callback(unsig
LDLM_DEBUG(lock, "timer expired");
- /* ptlrpc_fail_export must be called with this lock released */
spin_lock_bh(&expired_lock_thread.elt_lock);
list_del(&lock->l_pending_chain);
list_add(&lock->l_pending_chain,
&expired_lock_thread.elt_expired_locks);
spin_unlock_bh(&expired_lock_thread.elt_lock);
+ wake_up(&expired_lock_thread.elt_waitq);
}
spin_unlock_bh(&waiting_locks_spinlock);
|
|
From: Phil S. <ps...@us...> - 2003-05-19 06:29:25
|
Update of /cvsroot/lustre/lustre/ldlm
In directory sc8-pr-cvs1:/tmp/cvs-serv30166/ldlm
Modified Files:
Tag: b_devel
ldlm_lockd.c
Log Message:
merge b_ad (cleaned up a bit) into b_devel; entirely 2.5 compat fixes
Index: ldlm_lockd.c
===================================================================
RCS file: /cvsroot/lustre/lustre/ldlm/ldlm_lockd.c,v
retrieving revision 1.131.2.30
retrieving revision 1.131.2.31
diff -u -w -b -B -p -r1.131.2.30 -r1.131.2.31
--- ldlm_lockd.c 16 May 2003 14:49:46 -0000 1.131.2.30
+++ ldlm_lockd.c 19 May 2003 06:29:21 -0000 1.131.2.31
@@ -86,19 +86,13 @@ static int expired_lock_main(void *arg)
ENTRY;
lock_kernel();
- daemonize();
+ kportal_daemonize("ldlm_elt");
-#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0)
+ SIGNAL_MASK_LOCK(current, flags);
sigfillset(¤t->blocked);
- recalc_sigpending();
-#else
- spin_lock_irqsave(¤t->sigmask_lock, flags);
- sigfillset(¤t->blocked);
- recalc_sigpending(current);
- spin_unlock_irqrestore(¤t->sigmask_lock, flags);
-#endif
+ RECALC_SIGPENDING;
+ SIGNAL_MASK_UNLOCK(current, flags);
- strcpy(current->comm, "ldlm_elt");
unlock_kernel();
expired_lock_thread.elt_state = ELT_READY;
|
|
From: Eric B. <ee...@us...> - 2003-05-23 18:41:42
|
Update of /cvsroot/lustre/lustre/ldlm
In directory sc8-pr-cvs1:/tmp/cvs-serv19138/ldlm
Modified Files:
Tag: b_devel
ldlm_lockd.c
Log Message:
* Removed redundant parameter to ptlrpc_{error,reply}
* Reply messages set msg->opc to match the request. Client checks this
for sanity. Makes tcpdump tractible and should help in other debug
situations.
* Made mismatched req/rep transnos on replay a protocol error rather than
LASSERT (it LBUGs for the time being).
Index: ldlm_lockd.c
===================================================================
RCS file: /cvsroot/lustre/lustre/ldlm/ldlm_lockd.c,v
retrieving revision 1.131.2.32
retrieving revision 1.131.2.33
diff -u -w -b -B -p -r1.131.2.32 -r1.131.2.33
--- ldlm_lockd.c 22 May 2003 18:50:46 -0000 1.131.2.32
+++ ldlm_lockd.c 23 May 2003 18:41:08 -0000 1.131.2.33
@@ -566,7 +566,7 @@ int ldlm_handle_cancel(struct ptlrpc_req
req->rq_status = 0;
}
- if (ptlrpc_reply(req->rq_svc, req) != 0)
+ if (ptlrpc_reply(req) != 0)
LBUG();
if (lock) {
@@ -674,7 +674,7 @@ static int ldlm_callback_reply(struct pt
&req->rq_repmsg);
if (rc)
return rc;
- return ptlrpc_reply(req->rq_svc, req);
+ return ptlrpc_reply(req);
}
static int ldlm_callback_handler(struct ptlrpc_request *req)
|
|
From: Mike S. <sh...@us...> - 2003-05-29 13:58:58
|
Update of /cvsroot/lustre/lustre/ldlm
In directory sc8-pr-cvs1:/tmp/cvs-serv32644
Modified Files:
Tag: b_devel
ldlm_lockd.c
Log Message:
Phil noticed that I'd been stupid, and accidentally removed some needed return
statements in my last checkin. Better now!
Index: ldlm_lockd.c
===================================================================
RCS file: /cvsroot/lustre/lustre/ldlm/ldlm_lockd.c,v
retrieving revision 1.131.2.37
retrieving revision 1.131.2.38
diff -u -w -b -B -p -r1.131.2.37 -r1.131.2.38
--- ldlm_lockd.c 28 May 2003 21:17:24 -0000 1.131.2.37
+++ ldlm_lockd.c 29 May 2003 13:37:56 -0000 1.131.2.38
@@ -186,6 +186,7 @@ static int ldlm_add_waiting_lock(struct
spin_unlock_bh(&waiting_locks_spinlock);
/* We drop this ref when we get removed from the list. */
class_export_get(lock->l_export);
+ return 0;
}
/*
@@ -200,6 +201,7 @@ int ldlm_del_waiting_lock(struct ldlm_lo
if (lock->l_export == NULL) {
/* We don't have a "waiting locks list" on clients. */
LDLM_DEBUG(lock, "client lock: no-op");
+ return 0;
}
spin_lock_bh(&waiting_locks_spinlock);
@@ -207,7 +209,7 @@ int ldlm_del_waiting_lock(struct ldlm_lo
if (list_empty(&lock->l_pending_chain)) {
spin_unlock_bh(&waiting_locks_spinlock);
LDLM_DEBUG(lock, "wasn't waiting");
- RETURN(0);
+ return 0;
}
list_next = lock->l_pending_chain.next;
@@ -229,6 +231,7 @@ int ldlm_del_waiting_lock(struct ldlm_lo
/* We got this ref when we were added to the list. */
class_export_put(lock->l_export);
LDLM_DEBUG(lock, "removed");
+ return 1;
}
#else /* !__KERNEL__ */
|
|
From: Mike S. <sh...@us...> - 2003-05-29 16:43:09
|
Update of /cvsroot/lustre/lustre/ldlm
In directory sc8-pr-cvs1:/tmp/cvs-serv10993
Modified Files:
Tag: b_devel
ldlm_lockd.c
Log Message:
returning 0, meant 1; probably harmless, but hey
Index: ldlm_lockd.c
===================================================================
RCS file: /cvsroot/lustre/lustre/ldlm/ldlm_lockd.c,v
retrieving revision 1.131.2.38
retrieving revision 1.131.2.39
diff -u -w -b -B -p -r1.131.2.38 -r1.131.2.39
--- ldlm_lockd.c 29 May 2003 13:37:56 -0000 1.131.2.38
+++ ldlm_lockd.c 29 May 2003 16:43:04 -0000 1.131.2.39
@@ -186,7 +186,7 @@ static int ldlm_add_waiting_lock(struct
spin_unlock_bh(&waiting_locks_spinlock);
/* We drop this ref when we get removed from the list. */
class_export_get(lock->l_export);
- return 0;
+ return 1;
}
/*
|
|
From: Phil S. <ps...@us...> - 2003-06-02 04:02:18
|
Update of /cvsroot/lustre/lustre/ldlm
In directory sc8-pr-cvs1:/tmp/cvs-serv14395/ldlm
Modified Files:
Tag: b_devel
ldlm_lockd.c
Log Message:
- print the error code if portals_do_debug_dumplog fails
- demote "mmap readpage - check locks" message to a CDEBUG
- don't try to set mtime on OSTs if there's no LSM (file not yet opened)
- print which client (nid) returned an error from a blocking AST
Index: ldlm_lockd.c
===================================================================
RCS file: /cvsroot/lustre/lustre/ldlm/ldlm_lockd.c,v
retrieving revision 1.131.2.39
retrieving revision 1.131.2.40
diff -u -w -b -B -p -r1.131.2.39 -r1.131.2.40
--- ldlm_lockd.c 29 May 2003 16:43:04 -0000 1.131.2.39
+++ ldlm_lockd.c 2 Jun 2003 04:02:14 -0000 1.131.2.40
@@ -323,7 +323,8 @@ int ldlm_server_blocking_ast(struct ldlm
ldlm_del_waiting_lock(lock);
ldlm_failed_ast(lock, rc, "blocking");
} else if (rc) {
- CERROR("client returned %d from blocking AST for lock %p\n",
+ CERROR("client (nid "LPU64") returned %d from blocking AST for "
+ "lock %p\n", req->rq_connection->c_peer.peer_nid,
req->rq_repmsg->status, lock);
LDLM_DEBUG(lock, "client returned error %d from blocking AST",
req->rq_status);
|
|
From: Phil S. <ps...@us...> - 2003-06-05 15:44:10
|
Update of /cvsroot/lustre/lustre/ldlm
In directory sc8-pr-cvs1:/tmp/cvs-serv32031
Modified Files:
Tag: b_devel
ldlm_lockd.c
Log Message:
reduce noise in ldlm_server_blocking_ast: two of these error codes are
common, acceptable races or other conditions. Only CERROR if we get
something different.
Index: ldlm_lockd.c
===================================================================
RCS file: /cvsroot/lustre/lustre/ldlm/ldlm_lockd.c,v
retrieving revision 1.131.2.41
retrieving revision 1.131.2.42
diff -u -w -b -B -p -r1.131.2.41 -r1.131.2.42
--- ldlm_lockd.c 4 Jun 2003 03:56:46 -0000 1.131.2.41
+++ ldlm_lockd.c 5 Jun 2003 15:44:04 -0000 1.131.2.42
@@ -318,8 +318,21 @@ int ldlm_server_blocking_ast(struct ldlm
ldlm_del_waiting_lock(lock);
ldlm_failed_ast(lock, rc, "blocking");
} else if (rc) {
- CERROR("client (nid "LPU64") returned %d from blocking AST for "
- "lock %p\n", req->rq_connection->c_peer.peer_nid,
+ if (rc == -EINVAL)
+ CDEBUG(D_DLMTRACE, "client (nid "LPU64") returned %d "
+ "from blocking AST for lock %p--normal race\n",
+ req->rq_connection->c_peer.peer_nid,
+ req->rq_repmsg->status, lock);
+ else if (rc == -ENOTCONN)
+ CDEBUG(D_DLMTRACE, "client (nid "LPU64") returned %d "
+ "from blocking AST for lock %p--this client was "
+ "probably rebooted while it held a lock, nothing"
+ " serious\n",req->rq_connection->c_peer.peer_nid,
+ req->rq_repmsg->status, lock);
+ else
+ CDEBUG(D_ERROR, "client (nid "LPU64") returned %d "
+ "from blocking AST for lock %p\n",
+ req->rq_connection->c_peer.peer_nid,
req->rq_repmsg->status, lock);
LDLM_DEBUG(lock, "client returned error %d from blocking AST",
req->rq_status);
|
|
From: Phil S. <ps...@us...> - 2003-06-09 03:56:18
|
Update of /cvsroot/lustre/lustre/ldlm
In directory sc8-pr-cvs1:/tmp/cvs-serv24063/ldlm
Modified Files:
Tag: b_devel
ldlm_lockd.c
Log Message:
It is very common, if a node is rebooted with a lock held, to receive
messages to the ldlm_callback_handler with an invalid export; quiet
these down to D_RPCTRACE messages instead of CERRORs.
Index: ldlm_lockd.c
===================================================================
RCS file: /cvsroot/lustre/lustre/ldlm/ldlm_lockd.c,v
retrieving revision 1.131.2.42
retrieving revision 1.131.2.43
diff -u -w -b -B -p -r1.131.2.42 -r1.131.2.43
--- ldlm_lockd.c 5 Jun 2003 15:44:04 -0000 1.131.2.42
+++ ldlm_lockd.c 9 Jun 2003 03:56:14 -0000 1.131.2.43
@@ -703,16 +703,17 @@ static int ldlm_callback_handler(struct
if (req->rq_export == NULL) {
struct ldlm_request *dlm_req;
- CERROR("operation %d from nid "LPU64" with bad export "
- "(ptl req %d/rep %d)\n", req->rq_reqmsg->opc,
- req->rq_connection->c_peer.peer_nid,
+ CDEBUG(D_RPCTRACE, "operation %d from nid "LPU64" with bad "
+ "export cookie "LPX64" (ptl req %d/rep %d); this is "
+ "normal if this node rebooted with a lock held\n",
+ req->rq_reqmsg->opc, req->rq_connection->c_peer.peer_nid,
+ req->rq_reqmsg->handle.cookie,
req->rq_request_portal, req->rq_reply_portal);
- CERROR("--> export cookie: "LPX64"\n",
- req->rq_reqmsg->handle.cookie);
+
dlm_req = lustre_swab_reqbuf(req, 0, sizeof (*dlm_req),
lustre_swab_ldlm_request);
if (dlm_req != NULL)
- CERROR("--> lock cookie: "LPX64"\n",
+ CDEBUG(D_RPCTRACE, "--> lock cookie: "LPX64"\n",
dlm_req->lock_handle1.cookie);
ldlm_callback_reply(req, -ENOTCONN);
|
|
From: Mike S. <sh...@us...> - 2003-07-13 09:33:19
|
Update of /cvsroot/lustre/lustre/ldlm
In directory sc8-pr-cvs1:/tmp/cvs-serv32044/ldlm
Modified Files:
Tag: b_devel
ldlm_lockd.c
Log Message:
Whine to console if we take > obd_timeout to service a request, or grant a lock.
Index: ldlm_lockd.c
===================================================================
RCS file: /cvsroot/lustre/lustre/ldlm/ldlm_lockd.c,v
retrieving revision 1.131.2.46
retrieving revision 1.131.2.47
diff -u -w -b -B -p -r1.131.2.46 -r1.131.2.47
--- ldlm_lockd.c 5 Jul 2003 22:46:20 -0000 1.131.2.46
+++ ldlm_lockd.c 13 Jul 2003 09:33:10 -0000 1.131.2.47
@@ -347,10 +347,19 @@ int ldlm_server_blocking_ast(struct ldlm
RETURN(rc);
}
+/* XXX copied from ptlrpc/service.c */
+static long timeval_sub(struct timeval *large, struct timeval *small)
+{
+ return (large->tv_sec - small->tv_sec) * 1000000 +
+ (large->tv_usec - small->tv_usec);
+}
+
int ldlm_server_completion_ast(struct ldlm_lock *lock, int flags, void *data)
{
struct ldlm_request *body;
struct ptlrpc_request *req;
+ struct timeval granted_time;
+ long total_enqueue_wait;
int rc = 0, size = sizeof(*body);
ENTRY;
@@ -359,6 +368,12 @@ int ldlm_server_completion_ast(struct ld
RETURN(-EINVAL);
}
+ do_gettimeofday(&granted_time);
+ total_enqueue_wait = timeval_sub(&granted_time, &lock->l_enqueued_time);
+
+ if (total_enqueue_wait / 1000000 > obd_timeout)
+ LDLM_ERROR(lock, "enqueue wait took %ldus", total_enqueue_wait);
+
req = ptlrpc_prep_req(lock->l_export->exp_ldlm_data.led_import,
LDLM_CP_CALLBACK, 1, &size, NULL);
if (!req)
@@ -370,7 +385,8 @@ int ldlm_server_completion_ast(struct ld
body->lock_flags = flags;
ldlm_lock2desc(lock, &body->lock_desc);
- LDLM_DEBUG(lock, "server preparing completion AST");
+ LDLM_DEBUG(lock, "server preparing completion AST (after %ldus wait)",
+ total_enqueue_wait);
req->rq_replen = lustre_msg_size(0, NULL);
req->rq_level = LUSTRE_CONN_RECOVER;
@@ -447,6 +463,7 @@ int ldlm_handle_enqueue(struct ptlrpc_re
if (!lock)
GOTO(out, err = -ENOMEM);
+ do_gettimeofday(&lock->l_enqueued_time);
memcpy(&lock->l_remote_handle, &dlm_req->lock_handle1,
sizeof(lock->l_remote_handle));
LDLM_DEBUG(lock, "server-side enqueue handler, new lock created");
|
Thanks for helping keep SourceForge clean.
X