LU-12616 obclass: fix MDS start/stop race

[fs/lustre-release.git] / lustre / mdd / mdd_orphans.c

diff --git a/lustre/mdd/mdd_orphans.c b/lustre/mdd/mdd_orphans.c
index 7b99a93..de1500b 100644 (file)
--- a/lustre/mdd/mdd_orphans.c
+++ b/lustre/mdd/mdd_orphans.c
@@ -45,7 +45,7 @@
 #include <lustre_fid.h>
 #include "mdd_internal.h"
 
-static const char mdd_orphan_index_name[] = "PENDING";
+static const char mdd_orphan_index_name[] = MDT_ORPHAN_DIR;
 static const char dotdot[] = "..";
 
 enum {
@@ -164,7 +164,7 @@ int mdd_orphan_insert(const struct lu_env *env, struct mdd_object *obj,
        LASSERT(mdd_write_locked(env, obj) != 0);
        LASSERT(!(obj->mod_flags & ORPHAN_OBJ));
 
-       dt_write_lock(env, mdd->mdd_orphans, MOR_TGT_ORPHAN);
+       dt_write_lock(env, mdd->mdd_orphans, DT_TGT_ORPHAN);
 
        rc = mdd_orphan_insert_obj(env, mdd, obj, th);
        if (rc)
@@ -210,13 +210,13 @@ int mdd_orphan_declare_delete(const struct lu_env *env, struct mdd_object *obj,
        if (rc)
                return rc;
 
+       if (!mdd_object_exists(obj))
+               return -ENOENT;
+
         rc = mdo_declare_ref_del(env, obj, th);
         if (rc)
                 return rc;
 
-       if (!lu_object_exists(&obj->mod_obj.mo_lu))
-               return -ENOENT;
-
         if (S_ISDIR(mdd_object_type(obj))) {
                 rc = mdo_declare_ref_del(env, obj, th);
                 if (rc)
@@ -244,7 +244,7 @@ int mdd_orphan_delete(const struct lu_env *env, struct mdd_object *obj,
        struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
        struct dt_object *dor = mdd->mdd_orphans;
        struct dt_key *key;
-       int rc;
+       int rc = 0;
 
        ENTRY;
 
@@ -255,7 +255,10 @@ int mdd_orphan_delete(const struct lu_env *env, struct mdd_object *obj,
        LASSERT(dor);
 
        key = mdd_orphan_key_fill(env, mdo2fid(obj));
-       dt_write_lock(env, mdd->mdd_orphans, MOR_TGT_ORPHAN);
+       dt_write_lock(env, mdd->mdd_orphans, DT_TGT_ORPHAN);
+
+       if (OBD_FAIL_CHECK(OBD_FAIL_MDS_ORPHAN_DELETE))
+               goto ref_del;
 
        rc = dt_delete(env, mdd->mdd_orphans, key, th);
        if (rc == -ENOENT) {
@@ -263,6 +266,7 @@ int mdd_orphan_delete(const struct lu_env *env, struct mdd_object *obj,
                rc = dt_delete(env, mdd->mdd_orphans, key, th);
        }
 
+ref_del:
        if (!rc) {
                /* lov objects will be destroyed by caller */
                mdo_ref_del(env, obj, th);
@@ -286,6 +290,7 @@ static int mdd_orphan_destroy(const struct lu_env *env, struct mdd_object *obj,
 {
        struct thandle *th = NULL;
        struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
+       bool orphan_exists = true;
        int rc = 0;
        ENTRY;
 
@@ -298,23 +303,31 @@ static int mdd_orphan_destroy(const struct lu_env *env, struct mdd_object *obj,
                RETURN(rc);
        }
 
+       mdd_write_lock(env, obj, DT_TGT_CHILD);
        rc = mdd_orphan_declare_delete(env, obj, th);
-       if (rc)
-               GOTO(stop, rc);
-
-       rc = mdo_declare_destroy(env, obj, th);
-       if (rc)
-               GOTO(stop, rc);
+       if (rc == -ENOENT)
+               orphan_exists = false;
+       else if (rc)
+               GOTO(unlock, rc);
+
+       if (orphan_exists) {
+               rc = mdo_declare_destroy(env, obj, th);
+               if (rc)
+                       GOTO(unlock, rc);
+       }
 
        rc = mdd_trans_start(env, mdd, th);
        if (rc)
-               GOTO(stop, rc);
+               GOTO(unlock, rc);
 
-       mdd_write_lock(env, obj, MOR_TGT_CHILD);
        if (likely(obj->mod_count == 0)) {
-               dt_write_lock(env, mdd->mdd_orphans, MOR_TGT_ORPHAN);
+               dt_write_lock(env, mdd->mdd_orphans, DT_TGT_ORPHAN);
                rc = dt_delete(env, mdd->mdd_orphans, key, th);
-               if (rc == 0) {
+               if (rc) {
+                       CERROR("%s: could not delete orphan "DFID": rc = %d\n",
+                              mdd2obd_dev(mdd)->obd_name, PFID(mdo2fid(obj)),
+                              rc);
+               } else if (orphan_exists) {
                        mdo_ref_del(env, obj, th);
                        if (S_ISDIR(mdd_object_type(obj))) {
                                mdo_ref_del(env, obj, th);
@@ -322,15 +335,15 @@ static int mdd_orphan_destroy(const struct lu_env *env, struct mdd_object *obj,
                        }
                        rc = mdo_destroy(env, obj, th);
                } else {
-                       CERROR("%s: could not delete orphan "DFID": rc = %d\n",
-                              mdd2obd_dev(mdd)->obd_name, PFID(mdo2fid(obj)),
-                              rc);
+                       CWARN("%s: orphan %s "DFID" doesn't exist\n",
+                             mdd2obd_dev(mdd)->obd_name, (char *)key,
+                             PFID(mdo2fid(obj)));
                }
                dt_write_unlock(env, mdd->mdd_orphans);
        }
+unlock:
        mdd_write_unlock(env, obj);
 
-stop:
        rc = mdd_trans_stop(env, mdd, 0, th);
 
        RETURN(rc);
@@ -366,7 +379,7 @@ static int mdd_orphan_key_test_and_delete(const struct lu_env *env,
                        CERROR("%s: error unlinking orphan "DFID": rc = %d\n",
                               mdd2obd_dev(mdd)->obd_name, PFID(lf), rc);
         } else {
-                mdd_write_lock(env, mdo, MOR_TGT_CHILD);
+               mdd_write_lock(env, mdo, DT_TGT_CHILD);
                 if (likely(mdo->mod_count > 0)) {
                         CDEBUG(D_HA, "Found orphan "DFID" count %d, skip it\n",
                                PFID(lf), mdo->mod_count);