LU-7638 recovery: do not abort update recovery.

[fs/lustre-release.git] / lustre / ofd / ofd_obd.c

diff --git a/lustre/ofd/ofd_obd.c b/lustre/ofd/ofd_obd.c
index 24848f6..a1ee602 100644 (file)
--- a/lustre/ofd/ofd_obd.c
+++ b/lustre/ofd/ofd_obd.c
@@ -23,7 +23,7 @@
  * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  *
- * Copyright (c) 2012, 2014 Intel Corporation.
+ * Copyright (c) 2012, 2015, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
@@ -297,13 +297,17 @@ static int ofd_obd_reconnect(const struct lu_env *env, struct obd_export *exp,
        if (exp == NULL || obd == NULL || cluuid == NULL)
                RETURN(-EINVAL);
 
+       rc = nodemap_add_member(*(lnet_nid_t *)client_nid, exp);
+       if (rc != 0 && rc != -EEXIST)
+               RETURN(rc);
+
        ofd = ofd_dev(obd->obd_lu_dev);
 
        rc = ofd_parse_connect_data(env, exp, data, false);
        if (rc == 0)
                ofd_export_stats_init(ofd, exp, client_nid);
-
-       nodemap_add_member(*(lnet_nid_t *)client_nid, exp);
+       else
+               nodemap_del_member(exp);
 
        RETURN(rc);
 }
@@ -333,7 +337,6 @@ static int ofd_obd_connect(const struct lu_env *env, struct obd_export **_exp,
        struct ofd_device       *ofd;
        struct lustre_handle     conn = { 0 };
        int                      rc;
-       lnet_nid_t              *client_nid;
        ENTRY;
 
        if (_exp == NULL || obd == NULL || cluuid == NULL)
@@ -348,15 +351,19 @@ static int ofd_obd_connect(const struct lu_env *env, struct obd_export **_exp,
        exp = class_conn2export(&conn);
        LASSERT(exp != NULL);
 
+       if (localdata != NULL) {
+               rc = nodemap_add_member(*(lnet_nid_t *)localdata, exp);
+               if (rc != 0 && rc != -EEXIST)
+                       GOTO(out, rc);
+       } else {
+               CDEBUG(D_HA, "%s: cannot find nodemap for client %s: "
+                      "nid is null\n", obd->obd_name, cluuid->uuid);
+       }
+
        rc = ofd_parse_connect_data(env, exp, data, true);
        if (rc)
                GOTO(out, rc);
 
-       if (localdata != NULL) {
-               client_nid = localdata;
-               nodemap_add_member(*client_nid, exp);
-       }
-
        if (obd->obd_replayable) {
                struct tg_export_data *ted = &exp->exp_target_data;
 
@@ -373,8 +380,8 @@ static int ofd_obd_connect(const struct lu_env *env, struct obd_export **_exp,
 
 out:
        if (rc != 0) {
-               nodemap_del_member(exp);
                class_disconnect(exp);
+               nodemap_del_member(exp);
                *_exp = NULL;
        } else {
                *_exp = exp;
@@ -407,7 +414,6 @@ int ofd_obd_disconnect(struct obd_export *exp)
        if (!(exp->exp_flags & OBD_OPT_FORCE))
                ofd_grant_sanity_check(ofd_obd(ofd), __FUNCTION__);
 
-       nodemap_del_member(exp);
        rc = server_disconnect_export(exp);
 
        ofd_grant_discard(exp);
@@ -423,6 +429,7 @@ int ofd_obd_disconnect(struct obd_export *exp)
                lu_env_fini(&env);
        }
 out:
+       nodemap_del_member(exp);
        class_export_put(exp);
        RETURN(rc);
 }
@@ -528,17 +535,20 @@ static int ofd_destroy_export(struct obd_export *exp)
 int ofd_postrecov(const struct lu_env *env, struct ofd_device *ofd)
 {
        struct lu_device *ldev = &ofd->ofd_dt_dev.dd_lu_dev;
-       struct lfsck_start_param lsp;
        int rc;
 
        CDEBUG(D_HA, "%s: recovery is over\n", ofd_name(ofd));
 
-       lsp.lsp_start = NULL;
-       lsp.lsp_index_valid = 0;
-       rc = lfsck_start(env, ofd->ofd_osd, &lsp);
-       if (rc != 0 && rc != -EALREADY)
-               CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
-                     ofd_name(ofd), rc);
+       if (!ofd->ofd_skip_lfsck) {
+               struct lfsck_start_param lsp;
+
+               lsp.lsp_start = NULL;
+               lsp.lsp_index_valid = 0;
+               rc = lfsck_start(env, ofd->ofd_osd, &lsp);
+               if (rc != 0 && rc != -EALREADY)
+                       CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
+                             ofd_name(ofd), rc);
+       }
 
        return ldev->ld_ops->ldo_recovery_complete(env, ldev);
 }
@@ -722,7 +732,7 @@ int ofd_statfs_internal(const struct lu_env *env, struct ofd_device *ofd,
                unstable = ofd->ofd_osfs_inflight - unstable;
                ofd->ofd_osfs_unstable = 0;
                if (unstable) {
-                       /* some writes completed while we were running statfs
+                       /* some writes committed while we were running statfs
                         * w/o the ofd_osfs_lock. Those ones got added to
                         * the cached statfs data that we are about to crunch.
                         * Take them into account in the new statfs data */
@@ -1072,6 +1082,7 @@ static int ofd_echo_create(const struct lu_env *env, struct obd_export *exp,
        u64                      seq = ostid_seq(&oa->o_oi);
        struct ofd_seq          *oseq;
        int                      rc = 0, diff = 1;
+       long                     granted;
        u64                      next_id;
        int                      count;
 
@@ -1099,8 +1110,10 @@ static int ofd_echo_create(const struct lu_env *env, struct obd_export *exp,
        }
 
        mutex_lock(&oseq->os_create_lock);
-       rc = ofd_grant_create(env, ofd_obd(ofd)->obd_self_export, &diff);
-       if (rc < 0) {
+       granted = ofd_grant_create(env, ofd_obd(ofd)->obd_self_export, &diff);
+       if (granted < 0) {
+               rc = granted;
+               granted = 0;
                CDEBUG(D_HA, "%s: failed to acquire grant space for "
                       "precreate (%d): rc = %d\n", ofd_name(ofd), diff, rc);
                diff = 0;
@@ -1120,7 +1133,7 @@ static int ofd_echo_create(const struct lu_env *env, struct obd_export *exp,
                rc = 0;
        }
 
-       ofd_grant_commit(env, ofd_obd(ofd)->obd_self_export, rc);
+       ofd_grant_commit(ofd_obd(ofd)->obd_self_export, granted, rc);
 out:
        mutex_unlock(&oseq->os_create_lock);
        ofd_seq_put(env, oseq);
@@ -1288,7 +1301,7 @@ static int ofd_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
        switch (cmd) {
        case OBD_IOC_ABORT_RECOVERY:
                CERROR("%s: aborting recovery\n", obd->obd_name);
-               obd->obd_force_abort_recovery = 1;
+               obd->obd_abort_recovery = 1;
                target_stop_recovery_thread(obd);
                break;
        case OBD_IOC_SYNC:
@@ -1349,24 +1362,14 @@ static int ofd_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
  * special actions, it just invokes target_recovery_cleanup().
  *
  * \param[in] obd      OBD device of OFD
- * \param[in] stage    cleanup stage
  *
  * \retval             0
  */
-static int ofd_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
+static int ofd_precleanup(struct obd_device *obd)
 {
-       int rc = 0;
-
        ENTRY;
-
-       switch(stage) {
-       case OBD_CLEANUP_EARLY:
-               break;
-       case OBD_CLEANUP_EXPORTS:
-               target_cleanup_recovery(obd);
-               break;
-       }
-       RETURN(rc);
+       target_cleanup_recovery(obd);
+       RETURN(0);
 }
 
 /**