LU-1287 mountconf: write failover nid config correctly

Once OST starts before MDT, MDT needs steal OSC config from client
config, but both target nids and failover nids are stored as the
same type of config LCFG_ADD_UUID. To handle this correctly, all
LCFG_ADD_UUID before LCFG_SETUP is treated as target nids, while
after as failover nids. NB this implies that records should be
written into config file in fixed order.

Signed-off-by: Lai Siyao <laisiyao@whamcloud.com>
Change-Id: I42d014911edad1bc53bb1a66def160ee0da3e670
Reviewed-on: http://review.whamcloud.com/3107
Reviewed-by: Sebastien Buisson <sebastien.buisson@bull.net>
Reviewed-by: Mike Pershin <mike.pershin@intel.com>
Tested-by: Hudson
Tested-by: Maloo <whamcloud.maloo@gmail.com>
Reviewed-by: Oleg Drokin <green@whamcloud.com>

diff --git a/lustre/mgs/mgs_llog.c b/lustre/mgs/mgs_llog.c
index d20b59e..911edb2 100644 (file)
--- a/lustre/mgs/mgs_llog.c
+++ b/lustre/mgs/mgs_llog.c
@@ -1106,6 +1106,24 @@ static int mgs_write_log_osc_to_lov(const struct lu_env *env,
 static int name_create_mdt_and_lov(char **logname, char **lovname,
                                   struct fs_db *fsdb, int i);
 
 static int name_create_mdt_and_lov(char **logname, char **lovname,
                                   struct fs_db *fsdb, int i);
 

+static int add_param(char *params, char *key, char *val)
+{
+       char *start = params + strlen(params);
+       char *end = params + sizeof(((struct mgs_target_info *)0)->mti_params);
+       int keylen = 0;
+
+       if (key != NULL)
+               keylen = strlen(key);
+       if (start + 1 + keylen + strlen(val) >= end) {
+               CERROR("params are too long: %s %s%s\n",
+                      params, key != NULL ? key : "", val);
+               return -EINVAL;
+       }
+
+       sprintf(start, " %s%s", key != NULL ? key : "", val);
+       return 0;
+}
+

 static int mgs_steal_llog_handler(const struct lu_env *env,
                                  struct llog_handle *llh,
                                  struct llog_rec_hdr *rec, void *data)
 static int mgs_steal_llog_handler(const struct lu_env *env,
                                  struct llog_handle *llh,
                                  struct llog_rec_hdr *rec, void *data)


@@ -1201,15 +1219,23 @@ static int mgs_steal_llog_handler(const struct lu_env *env,
         if (got_an_osc_or_mdc == 0 || last_step < 0)
                 RETURN(rc);
 
         if (got_an_osc_or_mdc == 0 || last_step < 0)
                 RETURN(rc);
 

-        if (lcfg->lcfg_command == LCFG_ADD_UUID) {
-                uint64_t nodenid;
-                nodenid = lcfg->lcfg_nid;
+       if (lcfg->lcfg_command == LCFG_ADD_UUID) {
+               uint64_t nodenid = lcfg->lcfg_nid;

 
 

-                tmti->mti_nids[tmti->mti_nid_count] = nodenid;
-                tmti->mti_nid_count++;
+               if (strlen(tmti->mti_uuid) == 0) {
+                       /* target uuid not set, this config record is before
+                        * LCFG_SETUP, this nid is one of target node nid.
+                        */
+                       tmti->mti_nids[tmti->mti_nid_count] = nodenid;
+                       tmti->mti_nid_count++;
+               } else {
+                       /* failover node nid */
+                       rc = add_param(tmti->mti_params, PARAM_FAILNODE,
+                                      libcfs_nid2str(nodenid));
+               }

 
 

-                RETURN(rc);
-        }
+               RETURN(rc);
+       }

 
         if (lcfg->lcfg_command == LCFG_SETUP) {
                 char *target;
 
         if (lcfg->lcfg_command == LCFG_SETUP) {
                 char *target;


@@ -1892,11 +1918,18 @@ static int mgs_write_log_osc_to_lov(const struct lu_env *env,
        rc = record_start_log(env, mgs, &llh, logname);
         if (rc)
                GOTO(out_free, rc);
        rc = record_start_log(env, mgs, &llh, logname);
         if (rc)
                GOTO(out_free, rc);

+

         /* FIXME these should be a single journal transaction */
        rc = record_marker(env, llh, fsdb, CM_START | flags, mti->mti_svname,
                            "add osc");
        if (rc)
                GOTO(out_end, rc);
         /* FIXME these should be a single journal transaction */
        rc = record_marker(env, llh, fsdb, CM_START | flags, mti->mti_svname,
                            "add osc");
        if (rc)
                GOTO(out_end, rc);

+
+       /* NB: don't change record order, because upon MDT steal OSC config
+        * from client, it treats all nids before LCFG_SETUP as target nids
+        * (multiple interfaces), while nids after as failover node nids.
+        * See mgs_steal_llog_handler() LCFG_ADD_UUID.
+        */

         for (i = 0; i < mti->mti_nid_count; i++) {
                 CDEBUG(D_MGS, "add nid %s\n", libcfs_nid2str(mti->mti_nids[i]));
                rc = record_add_uuid(env, llh, mti->mti_nids[i], nodeuuid);
         for (i = 0; i < mti->mti_nid_count; i++) {
                 CDEBUG(D_MGS, "add nid %s\n", libcfs_nid2str(mti->mti_nids[i]));
                rc = record_add_uuid(env, llh, mti->mti_nids[i], nodeuuid);