Whamcloud - gitweb
LU-9679 lustre: use LIST_HEAD() for local lists.
[fs/lustre-release.git] / lustre / osp / osp_sync.c
index 6aa00d2..cbe6481 100644 (file)
@@ -40,6 +40,7 @@
 #define DEBUG_SUBSYSTEM S_MDS
 
 #include <linux/kthread.h>
+#include <linux/delay.h>
 #include <lustre_log.h>
 #include <lustre_update.h>
 #include "osp_internal.h"
@@ -562,7 +563,7 @@ static int osp_sync_interpret(const struct lu_env *env,
        struct osp_device *d = req->rq_cb_data;
 
        if (jra->jra_magic != OSP_JOB_MAGIC) {
-               DEBUG_REQ(D_ERROR, req, "bad magic %u\n", jra->jra_magic);
+               DEBUG_REQ(D_ERROR, req, "bad magic %u", jra->jra_magic);
                LBUG();
        }
        LASSERT(d);
@@ -1015,7 +1016,8 @@ static void osp_sync_process_committed(const struct lu_env *env,
        struct llog_ctxt        *ctxt;
        struct llog_handle      *llh;
        int                     *arr;
-       struct list_head         list, *le;
+       LIST_HEAD(list);
+       struct list_head         *le;
        struct llog_logid        lgid;
        int                      rc, i, count = 0, done = 0;
 
@@ -1046,7 +1048,6 @@ static void osp_sync_process_committed(const struct lu_env *env,
        llh = ctxt->loc_handle;
        LASSERT(llh);
 
-       INIT_LIST_HEAD(&list);
        spin_lock(&d->opd_sync_lock);
        list_splice(&d->opd_sync_committed_there, &list);
        INIT_LIST_HEAD(&d->opd_sync_committed_there);
@@ -1152,8 +1153,6 @@ static int osp_sync_process_queues(const struct lu_env *env,
        struct osp_device       *d = data;
 
        do {
-               struct l_wait_info lwi = { 0 };
-
                if (!osp_sync_running(d)) {
                        CDEBUG(D_HA, "stop llog processing\n");
                        return LLOG_PROC_BREAK;
@@ -1178,12 +1177,14 @@ static int osp_sync_process_queues(const struct lu_env *env,
                        llh = NULL;
                        rec = NULL;
                }
-
-               l_wait_event(d->opd_sync_waitq,
-                            !osp_sync_running(d) ||
-                            osp_sync_can_process_new(d, rec) ||
-                            !list_empty(&d->opd_sync_committed_there),
-                            &lwi);
+               if (OBD_FAIL_PRECHECK(OBD_FAIL_CATALOG_FULL_CHECK) &&
+                           cfs_fail_val != 1)
+                       msleep(1 * MSEC_PER_SEC);
+
+               wait_event_idle(d->opd_sync_waitq,
+                               !osp_sync_running(d) ||
+                               osp_sync_can_process_new(d, rec) ||
+                               !list_empty(&d->opd_sync_committed_there));
        } while (1);
 }
 
@@ -1211,7 +1212,6 @@ static int osp_sync_thread(void *_arg)
 {
        struct osp_device       *d = _arg;
        struct ptlrpc_thread    *thread = &d->opd_sync_thread;
-       struct l_wait_info       lwi = { 0 };
        struct llog_ctxt        *ctxt;
        struct obd_device       *obd = d->opd_obd;
        struct llog_handle      *llh;
@@ -1279,12 +1279,11 @@ next:
                /* processing reaches catalog bottom */
                if (d->opd_sync_last_catalog_idx == size)
                        d->opd_sync_last_catalog_idx = LLOG_CAT_FIRST;
-               else if (wrapped)
-                       /* If catalog is wrapped we can`t predict last index of
-                        * processing because lgh_last_idx could be changed.
-                        * Starting form the next one */
-                       d->opd_sync_last_catalog_idx++;
-
+               /* If catalog is wrapped we can`t predict last index of
+                * processing because lgh_last_idx could be changed.
+                * Starting form the next one. Index would be increased
+                * at llog_process_thread
+                */
        } while (rc == 0 && (wrapped ||
                             d->opd_sync_last_catalog_idx == LLOG_CAT_FIRST));
 
@@ -1322,11 +1321,11 @@ next:
        while (atomic_read(&d->opd_sync_rpcs_in_progress) > 0) {
                osp_sync_process_committed(&env, d);
 
-               lwi = LWI_TIMEOUT(cfs_time_seconds(5), NULL, NULL);
-               rc = l_wait_event(d->opd_sync_waitq,
-                               atomic_read(&d->opd_sync_rpcs_in_progress) == 0,
-                                 &lwi);
-               if (rc == -ETIMEDOUT)
+               rc = wait_event_idle_timeout(
+                       d->opd_sync_waitq,
+                       atomic_read(&d->opd_sync_rpcs_in_progress) == 0,
+                       cfs_time_seconds(5));
+               if (rc == 0)
                        count++;
                LASSERTF(count < 10, "%s: %d %d %sempty\n",
                         d->opd_obd->obd_name,
@@ -1424,13 +1423,30 @@ static int osp_sync_llog_init(const struct lu_env *env, struct osp_device *d)
        LASSERT(ctxt);
 
        if (likely(logid_id(&osi->osi_cid.lci_logid) != 0)) {
-               rc = llog_open(env, ctxt, &lgh, &osi->osi_cid.lci_logid, NULL,
-                              LLOG_OPEN_EXISTS);
-               /* re-create llog if it is missing */
-               if (rc == -ENOENT)
+               struct lu_fid fid_temp;
+
+               if (CFS_FAIL_CHECK(OBD_FAIL_OSP_INVALID_LOGID)) {
+                       memset(&osi->osi_cid, 0, sizeof(osi->osi_cid));
+                       logid_set_id(&osi->osi_cid.lci_logid, cfs_fail_val);
+               }
+
+               logid_to_fid(&osi->osi_cid.lci_logid, &fid_temp);
+               if (fid_is_sane(&fid_temp)) {
+                       rc = llog_open(env, ctxt, &lgh, &osi->osi_cid.lci_logid,
+                                      NULL, LLOG_OPEN_EXISTS);
+
+                       /* re-create llog if it is missing */
+                       if (rc == -ENOENT)
+                               logid_set_id(&osi->osi_cid.lci_logid, 0);
+                       else if (rc < 0)
+                               GOTO(out_cleanup, rc);
+               } else {
+                       CERROR("%s: the catid "DFID" for init llog %d is bad\n",
+                              obd->obd_name, PFID(&fid_temp), d->opd_index);
+
+                       /* it will be recreated later */
                        logid_set_id(&osi->osi_cid.lci_logid, 0);
-               else if (rc < 0)
-                       GOTO(out_cleanup, rc);
+               }
        }
 
        if (unlikely(logid_id(&osi->osi_cid.lci_logid) == 0)) {
@@ -1511,7 +1527,6 @@ static void osp_sync_llog_fini(const struct lu_env *env, struct osp_device *d)
  */
 int osp_sync_init(const struct lu_env *env, struct osp_device *d)
 {
-       struct l_wait_info       lwi = { 0 };
        struct task_struct      *task;
        int                      rc;
 
@@ -1552,8 +1567,8 @@ int osp_sync_init(const struct lu_env *env, struct osp_device *d)
                GOTO(err_llog, rc);
        }
 
-       l_wait_event(d->opd_sync_thread.t_ctl_waitq,
-                    osp_sync_running(d) || osp_sync_stopped(d), &lwi);
+       wait_event_idle(d->opd_sync_thread.t_ctl_waitq,
+                       osp_sync_running(d) || osp_sync_stopped(d));
 
        RETURN(0);
 err_llog: