LU-8044 mgs: Only add OSP for registered MDT

[fs/lustre-release.git] / lustre / mgs / mgs_llog.c

diff --git a/lustre/mgs/mgs_llog.c b/lustre/mgs/mgs_llog.c
index 70a83dc..f9cb376 100644 (file)
--- a/lustre/mgs/mgs_llog.c
+++ b/lustre/mgs/mgs_llog.c
@@ -2293,34 +2293,34 @@ static int mgs_write_log_mdt(const struct lu_env *env,
         #14 L mount_option 0:  1:client  2:lov1  3:MDC_uml1_mdsA_MNT_client
         */
 
-                /* copy client info about lov/lmv */
-               mgi->mgi_comp.comp_mti = mti;
-               mgi->mgi_comp.comp_fsdb = fsdb;
+       /* copy client info about lov/lmv */
+       mgi->mgi_comp.comp_mti = mti;
+       mgi->mgi_comp.comp_fsdb = fsdb;
 
-               rc = mgs_steal_llog_for_mdt_from_client(env, mgs, cliname,
-                                                       &mgi->mgi_comp);
-               if (rc)
-                       GOTO(out_free, rc);
-               rc = mgs_write_log_mdc_to_lmv(env, mgs, fsdb, mti, cliname,
-                                              fsdb->fsdb_clilmv);
-               if (rc)
-                       GOTO(out_free, rc);
+       rc = mgs_steal_llog_for_mdt_from_client(env, mgs, cliname,
+                                               &mgi->mgi_comp);
+       if (rc)
+               GOTO(out_free, rc);
+       rc = mgs_write_log_mdc_to_lmv(env, mgs, fsdb, mti, cliname,
+                                     fsdb->fsdb_clilmv);
+       if (rc)
+               GOTO(out_free, rc);
 
-                /* add mountopts */
-               rc = record_start_log(env, mgs, &llh, cliname);
-               if (rc)
-                       GOTO(out_free, rc);
+       /* add mountopts */
+       rc = record_start_log(env, mgs, &llh, cliname);
+       if (rc)
+               GOTO(out_free, rc);
 
-               rc = record_marker(env, llh, fsdb, CM_START, cliname,
-                                   "mount opts");
-               if (rc)
-                       GOTO(out_end, rc);
-               rc = record_mount_opt(env, llh, cliname, fsdb->fsdb_clilov,
-                                      fsdb->fsdb_clilmv);
-               if (rc)
-                       GOTO(out_end, rc);
-               rc = record_marker(env, llh, fsdb, CM_END, cliname,
-                                   "mount opts");
+       rc = record_marker(env, llh, fsdb, CM_START, cliname,
+                          "mount opts");
+       if (rc)
+               GOTO(out_end, rc);
+       rc = record_mount_opt(env, llh, cliname, fsdb->fsdb_clilov,
+                             fsdb->fsdb_clilmv);
+       if (rc)
+               GOTO(out_end, rc);
+       rc = record_marker(env, llh, fsdb, CM_END, cliname,
+                          "mount opts");
 
        if (rc)
                GOTO(out_end, rc);
@@ -2335,8 +2335,34 @@ static int mgs_write_log_mdt(const struct lu_env *env,
                        if (rc)
                                GOTO(out_end, rc);
 
-                       rc = mgs_write_log_osp_to_mdt(env, mgs, fsdb, mti,
-                                                     i, logname);
+                       /* NB: If the log for the MDT is empty, it means
+                        * the MDT is only added to the index
+                        * map, and not being process yet, i.e. this
+                        * is an unregistered MDT, see mgs_write_log_target().
+                        * so we should skip it. Otherwise
+                        *
+                        * 1. MGS get register request for MDT1 and MDT2.
+                        *
+                        * 2. Then both MDT1 and MDT2 are added into
+                        * fsdb_mdt_index_map. (see mgs_set_index()).
+                        *
+                        * 3. Then MDT1 get the lock of fsdb_mutex, then
+                        * generate the config log, here, it will regard MDT2
+                        * as an existent MDT, and generate "add osp" for
+                        * lustre-MDT0001-osp-MDT0002. Note: at the moment
+                        * MDT0002 config log is still empty, so it will
+                        * add "add osp" even before "lov setup", which
+                        * will definitly cause trouble.
+                        *
+                        * 4. MDT1 registeration finished, fsdb_mutex is
+                        * released, then MDT2 get in, then in above
+                        * mgs_steal_llog_for_mdt_from_client(), it will
+                        * add another osp log for lustre-MDT0001-osp-MDT0002,
+                        * which will cause another trouble.*/
+                       if (!mgs_log_is_empty(env, mgs, logname))
+                               rc = mgs_write_log_osp_to_mdt(env, mgs, fsdb,
+                                                             mti, i, logname);
+
                        name_destroy(&logname);
                        if (rc)
                                GOTO(out_end, rc);
@@ -3745,6 +3771,9 @@ int mgs_write_log_target(const struct lu_env *env, struct mgs_device *mgs,
                rc = 0;
        }
 
+       OBD_FAIL_TIMEOUT(OBD_FAIL_MGS_WRITE_TARGET_DELAY, cfs_fail_val > 0 ?
+                        cfs_fail_val : 10);
+
        mutex_lock(&fsdb->fsdb_mutex);
 
         if (mti->mti_flags &