* Copyright (C) 2003 Cluster File Systems, Inc.
* Author: Andreas Dilger <adilger@clusterfs.com>
*
- * This file is part of Lustre, http://www.lustre.org.
+ * This file is part of the Lustre file system, http://www.lustre.org
+ * Lustre is a trademark of Cluster File Systems, Inc.
*
- * Lustre is free software; you can redistribute it and/or
- * modify it under the terms of version 2 of the GNU General Public
- * License as published by the Free Software Foundation.
+ * You may have signed or agreed to another license before downloading
+ * this software. If so, you are bound by the terms and conditions
+ * of that agreement, and the following does not apply to you. See the
+ * LICENSE file included with this distribution for more information.
*
- * Lustre is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
- * GNU General Public License for more details.
+ * If you did not agree to a different license, then this copy of Lustre
+ * is open source software; you can redistribute it and/or modify it
+ * under the terms of version 2 of the GNU General Public License as
+ * published by the Free Software Foundation.
*
- * You should have received a copy of the GNU General Public License
- * along with Lustre; if not, write to the Free Software
- * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ * In either case, Lustre is distributed in the hope that it will be
+ * useful, but WITHOUT ANY WARRANTY; without even the implied warranty
+ * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * license text for more details.
*
* OST<->MDS recovery logging thread.
*
#endif
#ifdef __KERNEL__
-#include <linux/fs.h>
+# include <libcfs/libcfs.h>
#else
-# include <portals/list.h>
+# include <libcfs/list.h>
# include <liblustre.h>
#endif
-#include <linux/kp30.h>
-#include <linux/obd_class.h>
-#include <linux/lustre_commit_confd.h>
-#include <linux/obd_support.h>
-#include <linux/obd_class.h>
-#include <linux/lustre_net.h>
-#include <portals/types.h>
-#include <portals/list.h>
-#include <linux/lustre_log.h>
+#include <obd_class.h>
+#include <lustre_commit_confd.h>
+#include <obd_support.h>
+#include <obd_class.h>
+#include <lustre_net.h>
+#include <lnet/types.h>
+#include <libcfs/list.h>
+#include <lustre_log.h>
#include "ptlrpc_internal.h"
#ifdef __KERNEL__
-static struct llog_commit_master lustre_lcm;
-static struct llog_commit_master *lcm = &lustre_lcm;
-
-/* Allocate new commit structs in case we do not have enough */
-static int llcd_alloc(void)
+/* Allocate new commit structs in case we do not have enough.
+ * Make the llcd size small enough that it fits into a single page when we
+ * are sending/receiving it. */
+static int llcd_alloc(struct llog_commit_master *lcm)
{
struct llog_canceld_ctxt *llcd;
- int offset = offsetof(struct llog_canceld_ctxt, llcd_cookies);
+ int llcd_size;
- OBD_ALLOC(llcd, PAGE_SIZE + offset);
+ /* payload of lustre_msg V2 is bigger */
+ llcd_size = 4096 - lustre_msg_size(LUSTRE_MSG_MAGIC_V2, 1, NULL);
+ OBD_ALLOC(llcd,
+ llcd_size + offsetof(struct llog_canceld_ctxt, llcd_cookies));
if (llcd == NULL)
return -ENOMEM;
+ llcd->llcd_size = llcd_size;
llcd->llcd_lcm = lcm;
spin_lock(&lcm->lcm_llcd_lock);
}
/* Get a free cookie struct from the list */
-struct llog_canceld_ctxt *llcd_grab(void)
+static struct llog_canceld_ctxt *llcd_grab(struct llog_commit_master *lcm)
{
struct llog_canceld_ctxt *llcd;
+repeat:
spin_lock(&lcm->lcm_llcd_lock);
if (list_empty(&lcm->lcm_llcd_free)) {
spin_unlock(&lcm->lcm_llcd_lock);
- if (llcd_alloc() < 0) {
+ if (llcd_alloc(lcm) < 0) {
CERROR("unable to allocate log commit data!\n");
return NULL;
}
- spin_lock(&lcm->lcm_llcd_lock);
+ /* check new llcd wasn't grabbed while lock dropped, b=7407 */
+ goto repeat;
}
llcd = list_entry(lcm->lcm_llcd_free.next, typeof(*llcd), llcd_list);
atomic_dec(&lcm->lcm_llcd_numfree);
spin_unlock(&lcm->lcm_llcd_lock);
- llcd->llcd_tries = 0;
llcd->llcd_cookiebytes = 0;
return llcd;
}
-EXPORT_SYMBOL(llcd_grab);
static void llcd_put(struct llog_canceld_ctxt *llcd)
{
- int offset = offsetof(struct llog_canceld_ctxt, llcd_cookies);
+ struct llog_commit_master *lcm = llcd->llcd_lcm;
+ llog_ctxt_put(llcd->llcd_ctxt);
if (atomic_read(&lcm->lcm_llcd_numfree) >= lcm->lcm_llcd_maxfree) {
- OBD_FREE(llcd, PAGE_SIZE + offset);
+ int llcd_size = llcd->llcd_size +
+ offsetof(struct llog_canceld_ctxt, llcd_cookies);
+ OBD_FREE(llcd, llcd_size);
} else {
spin_lock(&lcm->lcm_llcd_lock);
list_add(&llcd->llcd_list, &lcm->lcm_llcd_free);
}
/* Send some cookies to the appropriate target */
-void llcd_send(struct llog_canceld_ctxt *llcd)
+static void llcd_send(struct llog_canceld_ctxt *llcd)
+{
+ if (!(llcd->llcd_lcm->lcm_flags & LLOG_LCM_FL_EXIT)) {
+ spin_lock(&llcd->llcd_lcm->lcm_llcd_lock);
+ list_add_tail(&llcd->llcd_list,
+ &llcd->llcd_lcm->lcm_llcd_pending);
+ spin_unlock(&llcd->llcd_lcm->lcm_llcd_lock);
+ }
+ cfs_waitq_signal_nr(&llcd->llcd_lcm->lcm_waitq, 1);
+}
+
+/**
+ * Grab llcd and assign it to passed @ctxt. Also set up backward link
+ * and get ref on @ctxt.
+ */
+static struct llog_canceld_ctxt *ctxt_llcd_grab(struct llog_ctxt *ctxt)
{
- spin_lock(&llcd->llcd_lcm->lcm_llcd_lock);
- list_add_tail(&llcd->llcd_list, &llcd->llcd_lcm->lcm_llcd_pending);
- spin_unlock(&llcd->llcd_lcm->lcm_llcd_lock);
+ struct llog_canceld_ctxt *llcd;
+
+ LASSERT_SEM_LOCKED(&ctxt->loc_sem);
+ llcd = llcd_grab(ctxt->loc_lcm);
+ if (llcd == NULL)
+ return NULL;
+
+ llcd->llcd_ctxt = llog_ctxt_get(ctxt);
+ ctxt->loc_llcd = llcd;
- wake_up_nr(&llcd->llcd_lcm->lcm_waitq, 1);
+ CDEBUG(D_RPCTRACE,"grab llcd %p:%p\n", ctxt->loc_llcd, ctxt);
+ return llcd;
+}
+
+/**
+ * Put llcd in passed @ctxt. Set ->loc_llcd to NULL.
+ */
+static void ctxt_llcd_put(struct llog_ctxt *ctxt)
+{
+ mutex_down(&ctxt->loc_sem);
+ if (ctxt->loc_llcd != NULL) {
+ CDEBUG(D_RPCTRACE,"put llcd %p:%p\n", ctxt->loc_llcd, ctxt);
+ llcd_put(ctxt->loc_llcd);
+ ctxt->loc_llcd = NULL;
+ }
+ ctxt->loc_imp = NULL;
+ mutex_up(&ctxt->loc_sem);
}
-EXPORT_SYMBOL(llcd_send);
/* deleted objects have a commit callback that cancels the MDS
- * log record for the deletion. The commit callback calls this
- * function
+ * log record for the deletion. The commit callback calls this
+ * function
*/
int llog_obd_repl_cancel(struct llog_ctxt *ctxt,
struct lov_stripe_md *lsm, int count,
LASSERT(ctxt);
- if (count == 0 || cookies == NULL) {
- down(&ctxt->loc_sem);
- if (ctxt->loc_llcd == NULL || !(flags & OBD_LLOG_FL_SENDNOW))
- GOTO(out, rc);
+ mutex_down(&ctxt->loc_sem);
+ llcd = ctxt->loc_llcd;
- llcd = ctxt->loc_llcd;
- GOTO(send_now, rc);
+ if (ctxt->loc_imp == NULL) {
+ CDEBUG(D_RPCTRACE, "no import for ctxt %p\n", ctxt);
+ GOTO(out, rc = 0);
}
- down(&ctxt->loc_sem);
- llcd = ctxt->loc_llcd;
- if (llcd == NULL) {
- llcd = llcd_grab();
+ if (count > 0 && cookies != NULL) {
if (llcd == NULL) {
- CERROR("couldn't get an llcd - dropped "LPX64":%x+%u\n",
- cookies->lgc_lgl.lgl_oid,
- cookies->lgc_lgl.lgl_ogen, cookies->lgc_index);
- GOTO(out, rc = -ENOMEM);
+ llcd = ctxt_llcd_grab(ctxt);
+ if (llcd == NULL) {
+ CERROR("couldn't get an llcd - dropped "LPX64
+ ":%x+%u\n",
+ cookies->lgc_lgl.lgl_oid,
+ cookies->lgc_lgl.lgl_ogen,
+ cookies->lgc_index);
+ GOTO(out, rc = -ENOMEM);
+ }
}
- llcd->llcd_import = ctxt->loc_imp;
- llcd->llcd_gen = ctxt->loc_gen;
- ctxt->loc_llcd = llcd;
- }
- memcpy((char *)llcd->llcd_cookies + llcd->llcd_cookiebytes, cookies,
- sizeof(*cookies));
- llcd->llcd_cookiebytes += sizeof(*cookies);
+ memcpy((char *)llcd->llcd_cookies + llcd->llcd_cookiebytes,
+ cookies, sizeof(*cookies));
+ llcd->llcd_cookiebytes += sizeof(*cookies);
+ } else {
+ if (llcd == NULL || !(flags & OBD_LLOG_FL_SENDNOW))
+ GOTO(out, rc);
+ }
-send_now:
- if ((PAGE_SIZE - llcd->llcd_cookiebytes < sizeof(*cookies) ||
- flags & OBD_LLOG_FL_SENDNOW)) {
- CDEBUG(D_HA, "send llcd: %p\n", llcd);
+ if ((llcd->llcd_size - llcd->llcd_cookiebytes) < sizeof(*cookies) ||
+ (flags & OBD_LLOG_FL_SENDNOW)) {
+ CDEBUG(D_RPCTRACE, "send llcd %p:%p\n", llcd, llcd->llcd_ctxt);
ctxt->loc_llcd = NULL;
llcd_send(llcd);
}
out:
- up(&ctxt->loc_sem);
+ mutex_up(&ctxt->loc_sem);
return rc;
}
EXPORT_SYMBOL(llog_obd_repl_cancel);
int rc = 0;
ENTRY;
- LASSERT(ctxt->loc_llcd);
-
if (exp && (ctxt->loc_imp == exp->exp_imp_reverse)) {
- CWARN("import will be destroyed, put llcd %p\n",
- ctxt->loc_llcd);
- llcd_put(ctxt->loc_llcd);
- ctxt->loc_llcd = NULL;
- up(&ctxt->loc_sem);
+ CDEBUG(D_RPCTRACE,"reverse import disconnect\n");
+ /*
+ * We put llcd because it is not going to sending list and
+ * thus, its refc will not be handled. We will handle it here.
+ */
+ ctxt_llcd_put(ctxt);
} else {
- up(&ctxt->loc_sem);
+ /*
+ * Sending cancel. This means that ctxt->loc_llcd wil be
+ * put on sending list in llog_obd_repl_cancel() and in
+ * this case recovery thread will take care of it refc.
+ */
rc = llog_cancel(ctxt, NULL, 0, NULL, OBD_LLOG_FL_SENDNOW);
}
-
RETURN(rc);
}
EXPORT_SYMBOL(llog_obd_repl_sync);
+static inline void stop_log_commit(struct llog_commit_master *lcm,
+ struct llog_commit_daemon *lcd,
+ int rc)
+{
+ CERROR("error preparing commit: rc %d\n", rc);
+
+ spin_lock(&lcm->lcm_llcd_lock);
+ list_splice_init(&lcd->lcd_llcd_list, &lcm->lcm_llcd_resend);
+ spin_unlock(&lcm->lcm_llcd_lock);
+}
+
static int log_commit_thread(void *arg)
{
struct llog_commit_master *lcm = arg;
struct llog_commit_daemon *lcd;
struct llog_canceld_ctxt *llcd, *n;
- unsigned long flags;
+ struct obd_import *import = NULL;
ENTRY;
OBD_ALLOC(lcd, sizeof(*lcd));
if (lcd == NULL)
RETURN(-ENOMEM);
- lock_kernel();
- ptlrpc_daemonize(); /* thread never needs to do IO */
-
- SIGNAL_MASK_LOCK(current, flags);
- sigfillset(¤t->blocked);
- RECALC_SIGPENDING;
- SIGNAL_MASK_UNLOCK(current, flags);
-
spin_lock(&lcm->lcm_thread_lock);
- THREAD_NAME(current->comm, "ll_log_commit_%d",
- atomic_read(&lcm->lcm_thread_total));
+ THREAD_NAME(cfs_curproc_comm(), CFS_CURPROC_COMM_MAX - 1,
+ "ll_log_comt_%02d", atomic_read(&lcm->lcm_thread_total));
atomic_inc(&lcm->lcm_thread_total);
spin_unlock(&lcm->lcm_thread_lock);
- unlock_kernel();
- INIT_LIST_HEAD(&lcd->lcd_lcm_list);
- INIT_LIST_HEAD(&lcd->lcd_llcd_list);
+ ptlrpc_daemonize(cfs_curproc_comm()); /* thread never needs to do IO */
+
+ CFS_INIT_LIST_HEAD(&lcd->lcd_lcm_list);
+ CFS_INIT_LIST_HEAD(&lcd->lcd_llcd_list);
lcd->lcd_lcm = lcm;
- CDEBUG(D_HA, "%s started\n", current->comm);
+ CDEBUG(D_HA, "%s started\n", cfs_curproc_comm());
do {
struct ptlrpc_request *request;
- struct obd_import *import = NULL;
struct list_head *sending_list;
int rc = 0;
+ if (import)
+ class_import_put(import);
+ import = NULL;
+
/* If we do not have enough pages available, allocate some */
while (atomic_read(&lcm->lcm_llcd_numfree) <
lcm->lcm_llcd_minfree) {
- if (llcd_alloc() < 0)
+ if (llcd_alloc(lcm) < 0)
break;
}
sending_list = &lcm->lcm_llcd_pending;
resend:
+ if (import)
+ class_import_put(import);
+ import = NULL;
if (lcm->lcm_flags & LLOG_LCM_FL_EXIT) {
lcm->lcm_llcd_maxfree = 0;
lcm->lcm_llcd_minfree = 0;
if (atomic_read(&lcm->lcm_thread_numidle) <= 1 &&
atomic_read(&lcm->lcm_thread_total) < lcm->lcm_thread_max) {
- rc = llog_start_commit_thread();
+ rc = llog_start_commit_thread(lcm);
if (rc < 0)
CERROR("error starting thread: rc %d\n", rc);
}
llcd = list_entry(lcd->lcd_llcd_list.next,
typeof(*llcd), llcd_list);
LASSERT(llcd->llcd_lcm == lcm);
- import = llcd->llcd_import;
+ import = llcd->llcd_ctxt->loc_imp;
+ if (import)
+ class_import_get(import);
}
list_for_each_entry_safe(llcd, n, sending_list, llcd_list) {
LASSERT(llcd->llcd_lcm == lcm);
- if (import == llcd->llcd_import)
+ if (import == llcd->llcd_ctxt->loc_imp)
list_move_tail(&llcd->llcd_list,
&lcd->lcd_llcd_list);
}
list_for_each_entry_safe(llcd, n, &lcm->lcm_llcd_resend,
llcd_list) {
LASSERT(llcd->llcd_lcm == lcm);
- if (import == llcd->llcd_import)
+ if (import == llcd->llcd_ctxt->loc_imp)
list_move_tail(&llcd->llcd_list,
&lcd->lcd_llcd_list);
}
/* We are the only one manipulating our local list - no lock */
list_for_each_entry_safe(llcd,n, &lcd->lcd_llcd_list,llcd_list){
- char *bufs[1] = {(char *)llcd->llcd_cookies};
- struct obd_device *obd = import->imp_obd;
- struct llog_ctxt *ctxt;
+ char *bufs[2] = { NULL, (char *)llcd->llcd_cookies };
list_del(&llcd->llcd_list);
if (llcd->llcd_cookiebytes == 0) {
- CDEBUG(D_HA, "just put empty llcd %p\n", llcd);
+ CDEBUG(D_RPCTRACE, "put empty llcd %p:%p\n",
+ llcd, llcd->llcd_ctxt);
llcd_put(llcd);
continue;
}
- /* check whether the cookies are new. if new then send, otherwise
- * just put llcd */
- ctxt = llog_get_context(obd, llcd->llcd_cookies[0].lgc_subsys + 1);
- LASSERT(ctxt != NULL);
- down(&ctxt->loc_sem);
- if (log_gen_lt(llcd->llcd_gen, ctxt->loc_gen)) {
- up(&ctxt->loc_sem);
- CDEBUG(D_HA, "just put stale llcd %p\n", llcd);
+
+ mutex_down(&llcd->llcd_ctxt->loc_sem);
+ if (llcd->llcd_ctxt->loc_imp == NULL) {
+ mutex_up(&llcd->llcd_ctxt->loc_sem);
+ CWARN("import will be destroyed, put "
+ "llcd %p:%p\n", llcd, llcd->llcd_ctxt);
llcd_put(llcd);
continue;
}
- up(&ctxt->loc_sem);
+ mutex_up(&llcd->llcd_ctxt->loc_sem);
- request = ptlrpc_prep_req(import, OBD_LOG_CANCEL, 1,
- &llcd->llcd_cookiebytes,
- bufs);
+ if (!import || (import == LP_POISON) ||
+ (import->imp_client == LP_POISON)) {
+ CERROR("No import %p (llcd=%p, ctxt=%p)\n",
+ import, llcd, llcd->llcd_ctxt);
+ llcd_put(llcd);
+ continue;
+ }
+
+ OBD_FAIL_TIMEOUT(OBD_FAIL_PTLRPC_DELAY_RECOV, 10);
+
+ request = ptlrpc_request_alloc(import, &RQF_LOG_CANCEL);
if (request == NULL) {
rc = -ENOMEM;
- CERROR("error preparing commit: rc %d\n", rc);
+ stop_log_commit(lcm, lcd, rc);
+ break;
+ }
- spin_lock(&lcm->lcm_llcd_lock);
- list_splice(&lcd->lcd_llcd_list,
- &lcm->lcm_llcd_resend);
- INIT_LIST_HEAD(&lcd->lcd_llcd_list);
- spin_unlock(&lcm->lcm_llcd_lock);
+ req_capsule_set_size(&request->rq_pill, &RMF_LOGCOOKIES,
+ RCL_CLIENT,llcd->llcd_cookiebytes);
+
+ rc = ptlrpc_request_bufs_pack(request,
+ LUSTRE_LOG_VERSION,
+ OBD_LOG_CANCEL, bufs,
+ NULL);
+ if (rc) {
+ ptlrpc_request_free(request);
+ stop_log_commit(lcm, lcd, rc);
break;
}
- request->rq_replen = lustre_msg_size(0, NULL);
+ /* XXX FIXME bug 249, 5515 */
+ request->rq_request_portal = LDLM_CANCEL_REQUEST_PORTAL;
+ request->rq_reply_portal = LDLM_CANCEL_REPLY_PORTAL;
+
+ ptlrpc_request_set_replen(request);
+ mutex_down(&llcd->llcd_ctxt->loc_sem);
+ if (llcd->llcd_ctxt->loc_imp == NULL) {
+ mutex_up(&llcd->llcd_ctxt->loc_sem);
+ CWARN("import will be destroyed, put "
+ "llcd %p:%p\n", llcd, llcd->llcd_ctxt);
+ llcd_put(llcd);
+ ptlrpc_req_finished(request);
+ continue;
+ }
+ mutex_up(&llcd->llcd_ctxt->loc_sem);
rc = ptlrpc_queue_wait(request);
ptlrpc_req_finished(request);
continue;
}
-#if 0 /* FIXME just put llcd, not send it again */
- spin_lock(&lcm->lcm_llcd_lock);
- list_splice(&lcd->lcd_llcd_list, &lcm->lcm_llcd_resend);
- if (++llcd->llcd_tries < 5) {
- CERROR("commit %p failed on attempt %d: rc %d\n",
- llcd, llcd->llcd_tries, rc);
-
- list_add_tail(&llcd->llcd_list,
- &lcm->lcm_llcd_resend);
- spin_unlock(&lcm->lcm_llcd_lock);
- } else {
- spin_unlock(&lcm->lcm_llcd_lock);
-#endif
- CERROR("commit %p dropped %d cookies: rc %d\n",
- llcd, (int)(llcd->llcd_cookiebytes /
- sizeof(*llcd->llcd_cookies)),
- rc);
- llcd_put(llcd);
-// }
- break;
+ CERROR("commit %p:%p drop %d cookies: rc %d\n",
+ llcd, llcd->llcd_ctxt,
+ (int)(llcd->llcd_cookiebytes /
+ sizeof(*llcd->llcd_cookies)), rc);
+ llcd_put(llcd);
}
if (rc == 0) {
}
} while(1);
+ if (import)
+ class_import_put(import);
+
/* If we are force exiting, just drop all of the cookies. */
if (lcm->lcm_flags & LLOG_LCM_FL_EXIT_FORCE) {
spin_lock(&lcm->lcm_llcd_lock);
- list_splice(&lcm->lcm_llcd_pending, &lcd->lcd_llcd_list);
- list_splice(&lcm->lcm_llcd_resend, &lcd->lcd_llcd_list);
- list_splice(&lcm->lcm_llcd_free, &lcd->lcd_llcd_list);
+ list_splice_init(&lcm->lcm_llcd_pending, &lcd->lcd_llcd_list);
+ list_splice_init(&lcm->lcm_llcd_resend, &lcd->lcd_llcd_list);
+ list_splice_init(&lcm->lcm_llcd_free, &lcd->lcd_llcd_list);
spin_unlock(&lcm->lcm_llcd_lock);
list_for_each_entry_safe(llcd, n, &lcd->lcd_llcd_list,llcd_list)
spin_unlock(&lcm->lcm_thread_lock);
OBD_FREE(lcd, sizeof(*lcd));
+ CDEBUG(D_HA, "%s exiting\n", cfs_curproc_comm());
+
spin_lock(&lcm->lcm_thread_lock);
atomic_dec(&lcm->lcm_thread_total);
spin_unlock(&lcm->lcm_thread_lock);
- wake_up(&lcm->lcm_waitq);
+ cfs_waitq_signal(&lcm->lcm_waitq);
- CDEBUG(D_HA, "%s exiting\n", current->comm);
return 0;
}
-int llog_start_commit_thread(void)
+int llog_start_commit_thread(struct llog_commit_master *lcm)
{
int rc;
ENTRY;
if (atomic_read(&lcm->lcm_thread_total) >= lcm->lcm_thread_max)
RETURN(0);
- rc = kernel_thread(log_commit_thread, lcm, CLONE_VM | CLONE_FILES);
+ rc = cfs_kernel_thread(log_commit_thread, lcm, CLONE_VM | CLONE_FILES);
if (rc < 0) {
CERROR("error starting thread #%d: %d\n",
atomic_read(&lcm->lcm_thread_total), rc);
EXPORT_SYMBOL(llog_start_commit_thread);
static struct llog_process_args {
- struct semaphore llpa_sem;
+ struct semaphore llpa_sem;
struct llog_ctxt *llpa_ctxt;
void *llpa_cb;
void *llpa_arg;
} llpa;
-int llog_init_commit_master(void)
+
+int llog_init_commit_master(struct llog_commit_master *lcm)
{
- INIT_LIST_HEAD(&lcm->lcm_thread_busy);
- INIT_LIST_HEAD(&lcm->lcm_thread_idle);
+ CFS_INIT_LIST_HEAD(&lcm->lcm_thread_busy);
+ CFS_INIT_LIST_HEAD(&lcm->lcm_thread_idle);
spin_lock_init(&lcm->lcm_thread_lock);
atomic_set(&lcm->lcm_thread_numidle, 0);
- init_waitqueue_head(&lcm->lcm_waitq);
- INIT_LIST_HEAD(&lcm->lcm_llcd_pending);
- INIT_LIST_HEAD(&lcm->lcm_llcd_resend);
- INIT_LIST_HEAD(&lcm->lcm_llcd_free);
+ cfs_waitq_init(&lcm->lcm_waitq);
+ CFS_INIT_LIST_HEAD(&lcm->lcm_llcd_pending);
+ CFS_INIT_LIST_HEAD(&lcm->lcm_llcd_resend);
+ CFS_INIT_LIST_HEAD(&lcm->lcm_llcd_free);
spin_lock_init(&lcm->lcm_llcd_lock);
atomic_set(&lcm->lcm_llcd_numfree, 0);
lcm->lcm_llcd_minfree = 0;
sema_init(&llpa.llpa_sem, 1);
return 0;
}
+EXPORT_SYMBOL(llog_init_commit_master);
-int llog_cleanup_commit_master(int force)
+int llog_cleanup_commit_master(struct llog_commit_master *lcm,
+ int force)
{
lcm->lcm_flags |= LLOG_LCM_FL_EXIT;
if (force)
lcm->lcm_flags |= LLOG_LCM_FL_EXIT_FORCE;
- wake_up(&lcm->lcm_waitq);
+ cfs_waitq_signal(&lcm->lcm_waitq);
wait_event_interruptible(lcm->lcm_waitq,
atomic_read(&lcm->lcm_thread_total) == 0);
return 0;
}
-
+EXPORT_SYMBOL(llog_cleanup_commit_master);
static int log_process_thread(void *args)
{
void *cb = data->llpa_cb;
struct llog_logid logid = *(struct llog_logid *)(data->llpa_arg);
struct llog_handle *llh = NULL;
- unsigned long flags;
int rc;
ENTRY;
-
- up(&data->llpa_sem);
- lock_kernel();
- ptlrpc_daemonize(); /* thread never needs to do IO */
-
- SIGNAL_MASK_LOCK(current, flags);
- sigfillset(¤t->blocked);
- RECALC_SIGPENDING;
- SIGNAL_MASK_UNLOCK(current, flags);
- unlock_kernel();
-
+
+ mutex_up(&data->llpa_sem);
+ ptlrpc_daemonize("llog_process"); /* thread does IO to log files */
+
rc = llog_create(ctxt, &llh, &logid, NULL);
if (rc) {
CERROR("llog_create failed %d\n", rc);
- RETURN(rc);
+ GOTO(out, rc);
}
rc = llog_init_handle(llh, LLOG_F_IS_CAT, NULL);
if (rc) {
CERROR("llog_init_handle failed %d\n", rc);
- GOTO(out, rc);
+ GOTO(release_llh, rc);
}
-
+
if (cb) {
rc = llog_cat_process(llh, (llog_cb_t)cb, NULL);
- if (rc)
+ if (rc != LLOG_PROC_BREAK)
CERROR("llog_cat_process failed %d\n", rc);
- } else
+ } else {
CWARN("no callback function for recovery\n");
+ }
- CDEBUG(D_HA, "send to llcd :%p forcibly\n", ctxt->loc_llcd);
+ CDEBUG(D_HA, "send llcd %p:%p forcibly after recovery\n",
+ ctxt->loc_llcd, ctxt);
llog_sync(ctxt, NULL);
-out:
+
+release_llh:
rc = llog_cat_put(llh);
if (rc)
CERROR("llog_cat_put failed %d\n", rc);
-
+out:
+ llog_ctxt_put(ctxt);
RETURN(rc);
}
-static int llog_recovery_generic(struct llog_ctxt *ctxt,
- void *handle,
- void *arg)
+
+static int llog_recovery_generic(struct llog_ctxt *ctxt, void *handle,void *arg)
{
+ struct obd_device *obd = ctxt->loc_obd;
int rc;
ENTRY;
- down(&llpa.llpa_sem);
- llpa.llpa_ctxt = ctxt;
+ if (obd->obd_stopping)
+ RETURN(-ENODEV);
+
+ mutex_down(&llpa.llpa_sem);
llpa.llpa_cb = handle;
llpa.llpa_arg = arg;
-
- rc = kernel_thread(log_process_thread, &llpa, CLONE_VM | CLONE_FILES);
- if (rc < 0)
+ llpa.llpa_ctxt = llog_ctxt_get(ctxt);
+ if (!llpa.llpa_ctxt) {
+ up(&llpa.llpa_sem);
+ RETURN(-ENODEV);
+ }
+ rc = cfs_kernel_thread(log_process_thread, &llpa, CLONE_VM | CLONE_FILES);
+ if (rc < 0) {
+ llog_ctxt_put(ctxt);
CERROR("error starting log_process_thread: %d\n", rc);
- else {
+ } else {
CDEBUG(D_HA, "log_process_thread: %d\n", rc);
rc = 0;
}
RETURN(rc);
}
+
int llog_repl_connect(struct llog_ctxt *ctxt, int count,
- struct llog_logid *logid, struct llog_ctxt_gen *gen)
+ struct llog_logid *logid, struct llog_gen *gen,
+ struct obd_uuid *uuid)
{
struct llog_canceld_ctxt *llcd;
int rc;
ENTRY;
-
- down(&ctxt->loc_sem);
- ctxt->loc_gen = *gen;
- llcd = ctxt->loc_llcd;
- if (llcd) {
- CDEBUG(D_HA, "put current llcd when new connection arrives\n");
- llcd_put(llcd);
+
+ /* send back llcd before recovery from llog */
+ if (ctxt->loc_llcd != NULL) {
+ CWARN("llcd %p:%p not empty\n", ctxt->loc_llcd, ctxt);
+ llog_sync(ctxt, NULL);
}
- llcd = llcd_grab();
+
+ mutex_down(&ctxt->loc_sem);
+ ctxt->loc_gen = *gen;
+ llcd = ctxt_llcd_grab(ctxt);
if (llcd == NULL) {
CERROR("couldn't get an llcd\n");
+ mutex_up(&ctxt->loc_sem);
RETURN(-ENOMEM);
}
- llcd->llcd_import = ctxt->loc_imp;
- llcd->llcd_gen = ctxt->loc_gen;
- ctxt->loc_llcd = llcd;
- up(&ctxt->loc_sem);
+ mutex_up(&ctxt->loc_sem);
- rc = llog_recovery_generic(ctxt, ctxt->llog_proc_cb, logid);
- if (rc != 0)
+ rc = llog_recovery_generic(ctxt, ctxt->llog_proc_cb, logid);
+ if (rc != 0) {
+ ctxt_llcd_put(ctxt);
CERROR("error recovery process: %d\n", rc);
-
+ }
RETURN(rc);
}
EXPORT_SYMBOL(llog_repl_connect);
git://git.whamcloud.com / fs / lustre-release.git / blobdiff