LU-57 Allow OSTs to be created with no primary node

Add a --servicenode parameter for mkfs.lustre to treat all service
nodes equally.

Bugzilla: 19064
Change-Id: I7851cb7781cdaf59d2cf18690d2da35121432cc1
Signed-off-by: Bobi Jam <bobijam@whamcloud.com>

diff --git a/lustre/doc/mkfs.lustre.8 b/lustre/doc/mkfs.lustre.8
index d79a4cc..80ed09a 100644 (file)
--- a/lustre/doc/mkfs.lustre.8
+++ b/lustre/doc/mkfs.lustre.8
@@ -47,6 +47,11 @@ Only print what would be done; does not affect the disk
 .TP
 .BI \--failnode= nid,...  
 Set the NID(s) of a failover partner. This option can be repeated as desired.
+Cannot be used with --servicenode.
+.TP
+.BI \--servicenode= nid,....
+Set the NID(s) of all service partner. This option treats all nodes as equal
+service nodes. Cannot be used with --failnode.
 .TP
 .BI \--fsname= filesystem_name  
 The Lustre filesystem this service will be part of.  The maximum

diff --git a/lustre/include/lustre_disk.h b/lustre/include/lustre_disk.h
index 7446b9f..99a5e7f 100644 (file)
--- a/lustre/include/lustre_disk.h
+++ b/lustre/include/lustre_disk.h
@@ -68,6 +68,7 @@
 #define LDD_F_WRITECONF     0x0100 /* regenerate all logs for this fs */
 #define LDD_F_UPGRADE14     0x0200 /* COMPAT_14 */
 #define LDD_F_PARAM         0x0400 /* process as lctl conf_param */
+#define LDD_F_NO_PRIMNODE   0x1000 /* all nodes are specified as servicenodes */
 
 enum ldd_mount_type {
         LDD_MT_EXT3 = 0,
@@ -104,7 +105,7 @@ struct lustre_disk_data {
 
         __u32      ldd_config_ver;      /* config rewrite count - not used */
         __u32      ldd_flags;           /* LDD_SV_TYPE */
-        __u32      ldd_svindex;         /* server index (0001), must match 
+        __u32      ldd_svindex;         /* server index (0001), must match
                                            svname */
         __u32      ldd_mount_type;      /* target fs type LDD_MT_* */
         char       ldd_fsname[64];      /* filesystem this server is part of */
@@ -128,7 +129,7 @@ static inline int server_make_name(__u32 flags, __u16 index, char *fs,
 {
         if (flags & (LDD_F_SV_TYPE_MDT | LDD_F_SV_TYPE_OST)) {
                 sprintf(name, "%.8s-%s%04x", fs,
-                        (flags & LDD_F_SV_TYPE_MDT) ? "MDT" : "OST",  
+                        (flags & LDD_F_SV_TYPE_MDT) ? "MDT" : "OST",
                         index);
         } else if (flags & LDD_F_SV_TYPE_MGS) {
                 sprintf(name, "MGS");
@@ -145,7 +146,7 @@ int server_name2index(char *svname, __u32 *idx, char **endptr);
 
 /****************** mount command *********************/
 
-/* The lmd is only used internally by Lustre; mount simply passes 
+/* The lmd is only used internally by Lustre; mount simply passes
    everything as string options */
 
 #define LMD_MAGIC    0xbdacbd03
@@ -160,7 +161,7 @@ struct lustre_mount_data {
         int        lmd_recovery_time_hard;
         char      *lmd_dev;           /* device name */
         char      *lmd_profile;       /* client only */
-        char      *lmd_opts;          /* lustre mount options (as opposed to 
+        char      *lmd_opts;          /* lustre mount options (as opposed to
                                          _device_ mount options) */
         __u32     *lmd_exclude;       /* array of OSTs to ignore */
 };
@@ -174,7 +175,7 @@ struct lustre_mount_data {
                                         existing MGS services */
 #define LMD_FLG_WRITECONF    0x0040  /* Rewrite config log */
 
-#define lmd_is_client(x) ((x)->lmd_flags & LMD_FLG_CLIENT) 
+#define lmd_is_client(x) ((x)->lmd_flags & LMD_FLG_CLIENT)
 
 
 /****************** last_rcvd file *********************/
@@ -272,7 +273,7 @@ static inline void check_lcd(char *obd_name, int index,
                 lcd->lcd_uuid[length - 1] = '\0';
 
                 LCONSOLE_ERROR("the client UUID (%s) on %s for exports"
-                               "stored in last_rcvd(index = %d) is bad!\n", 
+                               "stored in last_rcvd(index = %d) is bad!\n",
                                lcd->lcd_uuid, obd_name, index);
         }
 }
@@ -331,9 +332,9 @@ void lustre_register_kill_super_cb(void (*cfs)(struct super_block *sb));
 
 
 int lustre_common_put_super(struct super_block *sb);
-int lustre_process_log(struct super_block *sb, char *logname, 
+int lustre_process_log(struct super_block *sb, char *logname,
                      struct config_llog_instance *cfg);
-int lustre_end_log(struct super_block *sb, char *logname, 
+int lustre_end_log(struct super_block *sb, char *logname,
                        struct config_llog_instance *cfg);
 struct lustre_mount_info *server_find_mount_locked(char *name);
 struct lustre_mount_info *server_get_mount(char *name);

diff --git a/lustre/mgs/mgs_handler.c b/lustre/mgs/mgs_handler.c
index d9a9a3a..775b2ce 100644 (file)
--- a/lustre/mgs/mgs_handler.c
+++ b/lustre/mgs/mgs_handler.c
@@ -422,7 +422,8 @@ static int mgs_handle_target_reg(struct ptlrpc_request *req)
                         /* Nothing wrong, or fatal error */
                         GOTO(out_nolock, rc);
         } else {
-                if ((rc = mgs_check_failover_reg(mti)))
+                if (!(mti->mti_flags & LDD_F_NO_PRIMNODE)
+                    && (rc = mgs_check_failover_reg(mti)))
                         GOTO(out_nolock, rc);
         }
 

diff --git a/lustre/mgs/mgs_llog.c b/lustre/mgs/mgs_llog.c
index c57e28a..29e1d35 100644 (file)
--- a/lustre/mgs/mgs_llog.c
+++ b/lustre/mgs/mgs_llog.c
@@ -1008,7 +1008,6 @@ static int mgs_write_log_failnids(struct obd_device *obd,
         char *ptr = mti->mti_params;
         lnet_nid_t nid;
         int rc = 0;
-
         /*
         #03 L add_uuid  nid=uml1@tcp(0x20000c0a80201) nal=90 0:  1:uml1_UUID
         #04 L add_uuid  nid=1@elan(0x1000000000001)   nal=90 0:  1:uml1_UUID

diff --git a/lustre/obdclass/obd_mount.c b/lustre/obdclass/obd_mount.c
index 3654456..177a1bc 100644 (file)
--- a/lustre/obdclass/obd_mount.c
+++ b/lustre/obdclass/obd_mount.c
@@ -923,6 +923,8 @@ static int server_sb2mti(struct super_block *sb, struct mgs_target_info *mti)
         struct lustre_sb_info   *lsi = s2lsi(sb);
         struct lustre_disk_data *ldd = lsi->lsi_ldd;
         lnet_process_id_t        id;
+        lnet_nid_t               nid;
+        char                    *ptr;
         int i = 0;
         ENTRY;
 
@@ -938,6 +940,19 @@ static int server_sb2mti(struct super_block *sb, struct mgs_target_info *mti)
         while (LNetGetId(i++, &id) != -ENOENT) {
                 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
                         continue;
+                if ((ldd->ldd_flags & LDD_F_NO_PRIMNODE) != 0) {
+                        /* server use --servicenode param, only allow specified
+                         * nids be registered */
+                        ptr = ldd->ldd_params;
+                        while (class_find_param(ptr, PARAM_FAILNODE, &ptr) ==0){
+                                while (class_parse_nid(ptr, &nid, &ptr) == 0) {
+                                        if (nid == id.nid)
+                                                goto allowed;
+                                }
+                        }
+                        continue;
+                }
+allowed:
                 mti->mti_nids[mti->mti_nid_count] = id.nid;
                 mti->mti_nid_count++;
                 if (mti->mti_nid_count >= MTI_NIDS_MAX) {

diff --git a/lustre/tests/conf-sanity.sh b/lustre/tests/conf-sanity.sh
index 3d2c816..1c97e58 100644 (file)
--- a/lustre/tests/conf-sanity.sh
+++ b/lustre/tests/conf-sanity.sh
@@ -2288,16 +2288,25 @@ test_56() {
 }
 run_test 56 "check big indexes"
 
-test_57() { # bug 22656
+test_57a() { # bug 22656
        local NID=$(do_facet ost1 "$LCTL get_param nis" | tail -1 | awk '{print $1}')
        writeconf
        do_facet ost1 "$TUNEFS --failnode=$NID `ostdevname 1`" || error "tunefs failed"
        start_mgsmds
-       stop_mds
        start_ost && error "OST registration from failnode should fail"
        reformat
 }
-run_test 57 "initial registration from failnode should fail (should return errs)"
+run_test 57a "initial registration from failnode should fail (should return errs)"
+
+test_57b() {
+       local NID=$(do_facet ost1 "$LCTL get_param nis" | tail -1 | awk '{print $1}')
+       writeconf
+       do_facet ost1 "$TUNEFS --servicenode=$NID `ostdevname 1`" || error "tunefs failed"
+       start_mgsmds
+       start_ost || error "OST registration from servicenode should not fail"
+       reformat
+}
+run_test 57b "initial registration from servicenode should not fail"
 
 test_58() { # bug 22658
         [ "$FSTYPE" != "ldiskfs" ] && skip "not supported for $FSTYPE" && return

diff --git a/lustre/utils/mkfs_lustre.c b/lustre/utils/mkfs_lustre.c
index f6b0c83..f01eec6 100644 (file)
--- a/lustre/utils/mkfs_lustre.c
+++ b/lustre/utils/mkfs_lustre.c
@@ -122,6 +122,9 @@ void usage(FILE *out)
                 "\t\t\trequired for all targets other than the mgs node\n"
                 "\t\t--fsname=<filesystem_name> : default is 'lustre'\n"
                 "\t\t--failnode=<nid>[,<...>] : NID(s) of a failover partner\n"
+                "\t\t\tcannot be used with --servicenode\n"
+                "\t\t--servicenode=<nid>[,<...>] : NID(s) of all service partners\n"
+                "\t\t\ttreat all nodes as equal service node, cannot be used with --failnode\n"
                 "\t\t--param <key>=<value> : set a permanent parameter\n"
                 "\t\t\te.g. --param sys.timeout=40\n"
                 "\t\t\t     --param lov.stripesize=2M\n"
@@ -740,7 +743,7 @@ void print_ldd(char *str, struct lustre_disk_data *ldd)
         printf("Lustre FS:  %s\n", ldd->ldd_fsname);
         printf("Mount type: %s\n", MT_STR(ldd));
         printf("Flags:      %#x\n", ldd->ldd_flags);
-        printf("              (%s%s%s%s%s%s%s%s)\n",
+        printf("              (%s%s%s%s%s%s%s%s%s)\n",
                IS_MDT(ldd) ? "MDT ":"",
                IS_OST(ldd) ? "OST ":"",
                IS_MGS(ldd) ? "MGS ":"",
@@ -748,6 +751,7 @@ void print_ldd(char *str, struct lustre_disk_data *ldd)
                ldd->ldd_flags & LDD_F_VIRGIN     ? "first_time ":"",
                ldd->ldd_flags & LDD_F_UPDATE     ? "update ":"",
                ldd->ldd_flags & LDD_F_WRITECONF  ? "writeconf ":"",
+               ldd->ldd_flags & LDD_F_NO_PRIMNODE? "no_primnode ":"",
                ldd->ldd_flags & LDD_F_UPGRADE14  ? "upgrade1.4 ":"");
         printf("Persistent mount opts: %s\n", ldd->ldd_mount_opts);
         printf("Parameters:%s\n", ldd->ldd_params);
@@ -1187,15 +1191,17 @@ int parse_opts(int argc, char *const argv[], struct mkfs_opts *mop,
                 {"print", 0, 0, 'n'},
                 {"quiet", 0, 0, 'q'},
                 {"reformat", 0, 0, 'r'},
+                {"servicenode", 1, 0, 's'},
                 {"verbose", 0, 0, 'v'},
                 {"writeconf", 0, 0, 'w'},
                 {0, 0, 0, 0}
         };
-        char *optstring = "b:c:C:d:ef:Ghi:k:L:m:MnNo:Op:Pqru:vw";
+        char *optstring = "b:c:C:d:ef:Ghi:k:L:m:MnNo:Op:Pqrsu:vw";
         int opt;
         int rc, longidx;
         int upcall = 0;
         const size_t prefix_len = sizeof(PARAM_MDT_UPCALL) - 1;
+        int failnode_set = 0, servicenode_set = 0;
 
         while ((opt = getopt_long(argc, argv, optstring, long_opt, &longidx)) !=
                EOF) {
@@ -1237,8 +1243,19 @@ int parse_opts(int argc, char *const argv[], struct mkfs_opts *mop,
                         /* Must update the mgs logs */
                         mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
                         break;
-                case 'f': {
-                        char *nids = convert_hostnames(optarg);
+                case 'f':
+                case 's': {
+                        char *nids;
+
+                        if ((opt == 'f' && servicenode_set)
+                            || (opt == 's' && failnode_set)) {
+                                fprintf(stderr, "%s: %s cannot use with --%s\n",
+                                        progname, long_opt[longidx].name,
+                                        opt == 'f' ? "servicenode" : "failnode");
+                                return 1;
+                        }
+
+                        nids = convert_hostnames(optarg);
                         if (!nids)
                                 return 1;
                         rc = add_param(mop->mo_ldd.ldd_params, PARAM_FAILNODE,
@@ -1248,6 +1265,12 @@ int parse_opts(int argc, char *const argv[], struct mkfs_opts *mop,
                                 return rc;
                         /* Must update the mgs logs */
                         mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
+                        if (opt == 'f') {
+                                failnode_set = 1;
+                        } else {
+                                mop->mo_ldd.ldd_flags |= LDD_F_NO_PRIMNODE;
+                                servicenode_set = 1;
+                        }
                         failover = 1;
                         break;
                 }