glusterd/snapshot: Change file-system uuid to file-system label

Problem: In XFS changing file-system UUID with xfs_admin is causing
too much delay with large file-system. The time taken by xfs_admin
tool to change UUID is directly proportional to the size of the file
system.

Cause: In XFS file-system UUID is stored in file-system superblock.
Therefore for chaning UUID all the superblock needs to be changed.

Fix: Instead of using file-system UUID use file-system label.

Change-Id: Ifb4c668fb29cfc1c89d9b221abc8d09dc09589ec
BUG: 1115107
Signed-off-by: Rajesh Joseph <rjoseph@redhat.com>
Reviewed-on: http://review.gluster.org/8215
Reviewed-by: Atin Mukherjee <amukherj@redhat.com>
Reviewed-by: Avra Sengupta <asengupt@redhat.com>
Reviewed-by: Vijay Bellur <vbellur@redhat.com>
Tested-by: Vijay Bellur <vbellur@redhat.com>

6 files changed, 129 insertions, 96 deletions

diff --git a/xlators/mgmt/glusterd/src/glusterd-handshake.c b/xlators/mgmt/glusterd/src/glusterd-handshake.c
index 0fd60806da2..7d582d443f4 100644
--- a/xlators/mgmt/glusterd/src/glusterd-handshake.c
+++ b/xlators/mgmt/glusterd/src/glusterd-handshake.c
@@ -454,16 +454,16 @@ glusterd_create_missed_snap (glusterd_missed_snap_info *missed_snapinfo,
         }
 
         /* After the snapshot both the origin brick (LVM brick) and
-         * the snapshot brick will have the same file-system UUID. This
+         * the snapshot brick will have the same file-system label. This
          * will cause lot of problems at mount time. Therefore we must
-         * generate a new UUID for the snapshot brick
+         * generate a new label for the snapshot brick
          */
-        ret = glusterd_update_fs_uuid (brickinfo);
+        ret = glusterd_update_fs_label (brickinfo);
         if (ret) {
                 gf_log (this->name, GF_LOG_ERROR, "Failed to update "
-                        "file-system uuid for %s brick", brickinfo->path);
-                /* Failing to update UUID should not cause snapshot failure.
-                 * Currently UUID is updated only for XFS and ext2/ext3/ext4
+                        "file-system label for %s brick", brickinfo->path);
+                /* Failing to update label should not cause snapshot failure.
+                 * Currently label is updated only for XFS and ext2/ext3/ext4
                  * file-system.
                  */
         }
diff --git a/xlators/mgmt/glusterd/src/glusterd-snapshot.c b/xlators/mgmt/glusterd/src/glusterd-snapshot.c
index c1ddc0849cc..7678d77b3d5 100644
--- a/xlators/mgmt/glusterd/src/glusterd-snapshot.c
+++ b/xlators/mgmt/glusterd/src/glusterd-snapshot.c
@@ -971,6 +971,17 @@ glusterd_snapshot_restore_prevalidate (dict_t *dict, char **op_errstr,
                                         "Failed to set %s", key);
                                 goto out;
                         }
+
+                        snprintf (key, sizeof (key),
+                                  "snap%d.brick%d.fs_type",
+                                  volcount, brick_count);
+                        ret = dict_set_str (rsp_dict, key,
+                                            brickinfo->fstype);
+                        if (ret) {
+                                gf_log (this->name, GF_LOG_ERROR,
+                                        "Failed to set %s", key);
+                                goto out;
+                        }
                 }
 
                 snprintf (key, sizeof (key), "snap%d.brick_count", volcount);
@@ -1466,6 +1477,21 @@ glusterd_snap_restore_use_rsp_dict (dict_t *dst, dict_t *src)
                                         "Failed to set %s", key);
                                 goto out;
                         }
+
+                        snprintf (key, sizeof (key),
+                                  "snap%d.brick%d.fs_type", i, j);
+                        ret = dict_get_str (src, key, &strvalue);
+                        if (ret) {
+                                gf_log (this->name, GF_LOG_ERROR,
+                                        "Failed to get %s", key);
+                                goto out;
+                        }
+                        ret = dict_set_dynstr_with_alloc (dst, key, strvalue);
+                        if (ret) {
+                                gf_log (this->name, GF_LOG_DEBUG,
+                                        "Failed to set %s", key);
+                                goto out;
+                        }
                 }
         }
 
@@ -3725,7 +3751,8 @@ glusterd_snap_brick_create (glusterd_volinfo_t *snap_volinfo,
            But for now, mounting using runner apis.
         */
         ret = glusterd_mount_lvm_snapshot (brickinfo->device_path,
-                                           snap_brick_mount_path);
+                                           snap_brick_mount_path,
+                                           brickinfo->fstype);
         if (ret) {
                 gf_log (this->name, GF_LOG_ERROR,
                         "Failed to mount lvm snapshot.");
@@ -3978,7 +4005,7 @@ out:
         return ret;
 }
 
-/* This function will update the file-system UUID of the
+/* This function will update the file-system label of the
  * backend snapshot brick.
  *
  * @param brickinfo     brickinfo of the snap volume
@@ -3986,14 +4013,11 @@ out:
  * @return 0 on success and -1 on failure
  */
 int
-glusterd_update_fs_uuid (glusterd_brickinfo_t *brickinfo)
+glusterd_update_fs_label (glusterd_brickinfo_t *brickinfo)
 {
         int32_t         ret                     = -1;
         char            msg [PATH_MAX]          = "";
-        char            uuid_str [NAME_MAX]     = "";
-        char            template []             = "/tmp/xfsmountXXXXXX";
-        char           *mount_path              = NULL;
-        char           *cmd                     = NULL;
+        char            label [NAME_MAX]        = "";
         uuid_t          uuid                    = {0,};
         runner_t        runner                  = {0,};
         xlator_t       *this                    = NULL;
@@ -4005,82 +4029,39 @@ glusterd_update_fs_uuid (glusterd_brickinfo_t *brickinfo)
         /* Generate a new UUID */
         uuid_generate (uuid);
 
-        if (NULL == uuid_utoa_r (uuid, uuid_str)) {
-                gf_log (this->name, GF_LOG_ERROR, "Failed to convert "
-                        "uuid to string for %s brick", brickinfo->path);
-                goto out;
-        }
+        GLUSTERD_GET_UUID_NOHYPHEN (label, uuid);
 
         runinit (&runner);
-        snprintf (msg, sizeof (msg), "Changing filesystem uuid of %s brick to "
-                  "%s", brickinfo->path, uuid_str);
 
         /* Call the file-system specific tools to update the file-system
-         * UUID. Currently we are only supporting xfs and ext2/ext3/ext4
+         * label. Currently we are only supporting xfs and ext2/ext3/ext4
          * file-system.
          */
         if (0 == strcmp (brickinfo->fstype, "xfs")) {
-                /* TODO: xfs_admin tool is used to replace the file-system
-                 * UUID. As of now the tool is failing with error when trying
-                 * to replace the UUID. Therefore as a workaround we mount
-                 * the file-system to a temporary location and then unmount
-                 * it.
-                 * After this xfs_admin tool works fine.
-                 */
-                mount_path = mkdtemp (template);
-                if (NULL == mount_path) {
-                        gf_log (this->name, GF_LOG_ERROR, "Failed to create "
-                                "temporary folder. Error: %s",
-                                strerror (errno));
-                        runner_end (&runner);
-                        goto out;
-                }
-
-                /* First we mount the brick to a temporary path. Please
-                 * note that we are using "nouuid" option here because
-                 * both the origin volume brick and this brick will have
-                 * the same UUID. And XFS does not allow us to mount a
-                 * file-system which is sharing the UUID with any other
-                 * file-system on the system.
-                 */
-#ifdef GF_LINUX_HOST_OS
-                ret = mount (brickinfo->device_path, mount_path,
-                             brickinfo->fstype, 0, "nouuid");
-#else
-                ret = -1;
-#endif
-
-                if (ret) {
-                        gf_log (this->name, GF_LOG_ERROR, "Failed to mount "
-                                "%s at %s", brickinfo->device_path,
-                                mount_path);
-                        runner_end (&runner);
-                        goto out;
-                }
-
-                /* Now unmount the brick. */
-                ret = glusterd_umount (mount_path);
-                if (ret) {
-                        gf_log (this->name, GF_LOG_ERROR, "Failed to unmount "
-                                "%s from %s", brickinfo->device_path,
-                                mount_path);
-                        runner_end (&runner);
-                        goto out;
-                }
-
-                /* Now we are ready to run xfs_admin tool */
-                runner_add_args (&runner, "xfs_admin", "-U", uuid_str,
+                /* XFS label is of size 12. Therefore we should truncate the
+                 * label to 12 bytes*/
+                label [12] = '\0';
+                snprintf (msg, sizeof (msg), "Changing filesystem label of "
+                          "%s brick to %s", brickinfo->path, label);
+                /* Run the run xfs_admin tool to change the label
+                 * of the file-system */
+                runner_add_args (&runner, "xfs_admin", "-L", label,
                                  brickinfo->device_path, NULL);
         } else if (0 == strcmp (brickinfo->fstype, "ext4") ||
                    0 == strcmp (brickinfo->fstype, "ext3") ||
                    0 == strcmp (brickinfo->fstype, "ext2")) {
+                /* Ext2/Ext3/Ext4 label is of size 16. Therefore we should
+                 * truncate the label to 16 bytes*/
+                label [16] = '\0';
+                snprintf (msg, sizeof (msg), "Changing filesystem label of "
+                          "%s brick to %s", brickinfo->path, label);
                 /* For ext2/ext3/ext4 run tune2fs to change the
-                 * file-system UUID */
-                runner_add_args (&runner, "tune2fs", "-U", uuid_str,
+                 * file-system label */
+                runner_add_args (&runner, "tune2fs", "-L", label,
                                  brickinfo->device_path, NULL);
         } else {
                 gf_log (this->name, GF_LOG_WARNING, "Changing file-system "
-                        "UUID of %s file-system is not supported as of now",
+                        "label of %s file-system is not supported as of now",
                         brickinfo->fstype);
                 runner_end (&runner);
                 ret = -1;
@@ -4091,20 +4072,13 @@ glusterd_update_fs_uuid (glusterd_brickinfo_t *brickinfo)
         ret = runner_run (&runner);
         if (ret) {
                 gf_log (this->name, GF_LOG_ERROR, "Failed to change "
-                        "filesystem uuid of %s brick to %s",
-                        brickinfo->path, uuid_str);
+                        "filesystem label of %s brick to %s",
+                        brickinfo->path, label);
                 goto out;
         }
 
         ret = 0;
 out:
-        if (NULL != mount_path) {
-                if (rmdir (mount_path)) {
-                        gf_log (this->name, GF_LOG_ERROR, "Failed to remove "
-                                "temporary folder %s. Error: %s",
-                                mount_path, strerror (errno));
-                }
-        }
         return ret;
 }
 
@@ -4148,16 +4122,16 @@ glusterd_take_brick_snapshot (dict_t *dict, glusterd_volinfo_t *snap_vol,
         }
 
         /* After the snapshot both the origin brick (LVM brick) and
-         * the snapshot brick will have the same file-system UUID. This
+         * the snapshot brick will have the same file-system label. This
          * will cause lot of problems at mount time. Therefore we must
-         * generate a new UUID for the snapshot brick
+         * generate a new label for the snapshot brick
          */
-        ret = glusterd_update_fs_uuid (brickinfo);
+        ret = glusterd_update_fs_label (brickinfo);
         if (ret) {
                 gf_log (this->name, GF_LOG_ERROR, "Failed to update "
-                        "file-system uuid for %s brick", brickinfo->path);
-                /* Failing to update UUID should not cause snapshot failure.
-                 * Currently UUID is updated only for XFS and ext2/ext3/ext4
+                        "file-system label for %s brick", brickinfo->path);
+                /* Failing to update label should not cause snapshot failure.
+                 * Currently label is updated only for XFS and ext2/ext3/ext4
                  * file-system.
                  */
         }
diff --git a/xlators/mgmt/glusterd/src/glusterd-store.c b/xlators/mgmt/glusterd/src/glusterd-store.c
index e03ec7af9f5..c31d8a8ad71 100644
--- a/xlators/mgmt/glusterd/src/glusterd-store.c
+++ b/xlators/mgmt/glusterd/src/glusterd-store.c
@@ -325,6 +325,17 @@ gd_store_brick_snap_details_write (int fd, glusterd_brickinfo_t *brickinfo)
                         goto out;
         }
 
+        if (strlen (brickinfo->fstype) > 0) {
+                snprintf (value, sizeof (value), "%s", brickinfo->fstype);
+                ret = gf_store_save_value (fd,
+                                GLUSTERD_STORE_KEY_BRICK_FSTYPE, value);
+                if (ret) {
+                        gf_log (this->name, GF_LOG_ERROR, "Failed to save "
+                                "brick fs type of brick %s", brickinfo->path);
+                        goto out;
+                }
+        }
+
         memset (value, 0, sizeof (value));
         snprintf (value, sizeof(value), "%d", brickinfo->snap_status);
         ret = gf_store_save_value (fd, GLUSTERD_STORE_KEY_BRICK_SNAP_STATUS,
@@ -2272,6 +2283,10 @@ glusterd_store_retrieve_bricks (glusterd_volinfo_t *volinfo)
                         } else if (!strncmp (key, GLUSTERD_STORE_KEY_BRICK_SNAP_STATUS,
                                              strlen (GLUSTERD_STORE_KEY_BRICK_SNAP_STATUS))) {
                                 gf_string2int (value, &brickinfo->snap_status);
+                        } else if (!strncmp (key, GLUSTERD_STORE_KEY_BRICK_FSTYPE,
+                                             strlen (GLUSTERD_STORE_KEY_BRICK_FSTYPE))) {
+                                strncpy (brickinfo->fstype, value,
+                                         sizeof (brickinfo->fstype));
                         } else if (!strncmp (key,
                                     GLUSTERD_STORE_KEY_BRICK_VGNAME,
                                     strlen (GLUSTERD_STORE_KEY_BRICK_VGNAME))) {
@@ -3086,7 +3101,8 @@ out:
  * at the brick_mount_path
  */
 int32_t
-glusterd_mount_brick_paths (char *brick_mount_path, char *device_path)
+glusterd_mount_brick_paths (char *brick_mount_path, char *device_path,
+                            const char *fstype)
 {
         int32_t                  ret                 = -1;
         runner_t                 runner              = {0, };
@@ -3134,7 +3150,8 @@ glusterd_mount_brick_paths (char *brick_mount_path, char *device_path)
                         "Activating %s successful", device_path);
 
         /* Mount the snapshot */
-        ret = glusterd_mount_lvm_snapshot (device_path, brick_mount_path);
+        ret = glusterd_mount_lvm_snapshot (device_path, brick_mount_path,
+                                           fstype);
         if (ret) {
                 gf_log (this->name, GF_LOG_ERROR,
                         "Failed to mount lvm snapshot.");
@@ -3208,7 +3225,7 @@ glusterd_recreate_vol_brick_mounts (xlator_t  *this,
                 /* Check if brick_mount_path is already mounted.
                  * If not, mount the device_path at the brick_mount_path */
                 ret = glusterd_mount_brick_paths (brick_mount_path,
-                                                  brickinfo->device_path);
+                                brickinfo->device_path, brickinfo->fstype);
                 if (ret) {
                         gf_log (this->name, GF_LOG_ERROR,
                                 "Failed to mount brick_mount_path");
diff --git a/xlators/mgmt/glusterd/src/glusterd-store.h b/xlators/mgmt/glusterd/src/glusterd-store.h
index 4c0f0d42321..89cf24de789 100644
--- a/xlators/mgmt/glusterd/src/glusterd-store.h
+++ b/xlators/mgmt/glusterd/src/glusterd-store.h
@@ -82,6 +82,7 @@ typedef enum glusterd_store_ver_ac_{
 #define GLUSTERD_STORE_KEY_BRICK_DEVICE_PATH    "device_path"
 #define GLUSTERD_STORE_KEY_BRICK_MOUNT_DIR      "mount_dir"
 #define GLUSTERD_STORE_KEY_BRICK_SNAP_STATUS    "snap-status"
+#define GLUSTERD_STORE_KEY_BRICK_FSTYPE         "fs-type"
 #define GLUSTERD_STORE_KEY_BRICK_ID             "brick-id"
 
 #define GLUSTERD_STORE_KEY_PEER_UUID            "uuid"
diff --git a/xlators/mgmt/glusterd/src/glusterd-utils.c b/xlators/mgmt/glusterd/src/glusterd-utils.c
index 6b973a3638d..02b2ba923e5 100644
--- a/xlators/mgmt/glusterd/src/glusterd-utils.c
+++ b/xlators/mgmt/glusterd/src/glusterd-utils.c
@@ -612,6 +612,7 @@ glusterd_brickinfo_dup (glusterd_brickinfo_t *brickinfo,
         strcpy (dup_brickinfo->hostname, brickinfo->hostname);
         strcpy (dup_brickinfo->path, brickinfo->path);
         strcpy (dup_brickinfo->device_path, brickinfo->device_path);
+        strcpy (dup_brickinfo->fstype, brickinfo->fstype);
         ret = gf_canonicalize_path (dup_brickinfo->path);
         if (ret) {
                 gf_log (THIS->name, GF_LOG_ERROR, "Failed to canonicalize "
@@ -710,6 +711,13 @@ glusterd_snap_volinfo_restore (dict_t *dict, dict_t *rsp_dict,
                         strncpy (new_brickinfo->device_path, value,
                                  sizeof(new_brickinfo->device_path));
 
+                snprintf (key, sizeof (key), "snap%d.brick%d.fs_type",
+                          volcount, brick_count);
+                ret = dict_get_str (dict, key, &value);
+                if (!ret)
+                        strncpy (new_brickinfo->fstype, value,
+                                 sizeof(new_brickinfo->fstype));
+
                 /* If the brick is not of this peer, or snapshot is missed *
                  * for the brick do not replace the xattr for it */
                 if ((!uuid_compare (brickinfo->uuid, MY_UUID)) &&
@@ -2363,6 +2371,15 @@ gd_add_brick_snap_details_to_dict (dict_t *dict, char *prefix,
                 goto out;
         }
 
+        snprintf (key, sizeof (key), "%s.fs_type", prefix);
+        ret = dict_set_str (dict, key, brickinfo->fstype);
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "Failed to set fstype for %s:%s",
+                         brickinfo->hostname, brickinfo->path);
+                goto out;
+        }
+
         memset (key, 0, sizeof (key));
         snprintf (key, sizeof (key), "%s.mount_dir", prefix);
         ret = dict_set_str (dict, key, brickinfo->mount_dir);
@@ -3695,6 +3712,7 @@ gd_import_new_brick_snap_details (dict_t *dict, char *prefix,
         glusterd_conf_t *conf        = NULL;
         char             key[512]    = {0,};
         char            *snap_device = NULL;
+        char            *fs_type     = NULL;
         char            *mount_dir   = NULL;
 
         this = THIS;
@@ -3727,6 +3745,14 @@ gd_import_new_brick_snap_details (dict_t *dict, char *prefix,
         }
         strcpy (brickinfo->device_path, snap_device);
 
+        snprintf (key, sizeof (key), "%s.fs_type", prefix);
+        ret = dict_get_str (dict, key, &fs_type);
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR, "%s missing in payload", key);
+                goto out;
+        }
+        strcpy (brickinfo->fstype, fs_type);
+
         memset (key, 0, sizeof (key));
         snprintf (key, sizeof (key), "%s.mount_dir", prefix);
         ret = dict_get_str (dict, key, &mount_dir);
@@ -12206,7 +12232,8 @@ glusterd_compare_volume_name(struct list_head *list1, struct list_head *list2)
 }
 
 int32_t
-glusterd_mount_lvm_snapshot (char *device_path, char *brick_mount_path)
+glusterd_mount_lvm_snapshot (char *device_path, char *brick_mount_path,
+                             const char *fstype)
 {
         char               msg[NAME_MAX] = "";
         int32_t            ret           = -1;
@@ -12222,7 +12249,20 @@ glusterd_mount_lvm_snapshot (char *device_path, char *brick_mount_path)
         runinit (&runner);
         snprintf (msg, sizeof (msg), "mount %s %s",
                   device_path, brick_mount_path);
-        runner_add_args (&runner, "mount", device_path, brick_mount_path, NULL);
+
+        /* XFS file-system does not allow to mount file-system with duplicate
+         * UUID. File-system UUID of snapshot and its origin volume is same.
+         * Therefore to mount such a snapshot in XFS we need to pass nouuid
+         * option
+         */
+        if (!strcmp (fstype, "xfs")) {
+                runner_add_args (&runner, "mount", "-o", "nouuid",
+                                 device_path, brick_mount_path, NULL);
+        } else {
+                runner_add_args (&runner, "mount", device_path,
+                                 brick_mount_path, NULL);
+        }
+
         runner_log (&runner, this->name, GF_LOG_DEBUG, msg);
         ret = runner_run (&runner);
         if (ret) {
diff --git a/xlators/mgmt/glusterd/src/glusterd-utils.h b/xlators/mgmt/glusterd/src/glusterd-utils.h
index 3edb0c55db4..8d3af0689fd 100644
--- a/xlators/mgmt/glusterd/src/glusterd-utils.h
+++ b/xlators/mgmt/glusterd/src/glusterd-utils.h
@@ -757,7 +757,8 @@ gd_restore_snap_volume (dict_t *dict, dict_t *rsp_dict,
                         int32_t volcount);
 
 int32_t
-glusterd_mount_lvm_snapshot (char *device_path, char *brick_mount_path);
+glusterd_mount_lvm_snapshot (char *device_path, char *brick_mount_path,
+                             const char *fstype);
 
 int32_t
 glusterd_umount (const char *path);
@@ -926,6 +927,6 @@ glusterd_update_fstype (char *orig_brick_path,
                         char *fstype, size_t fslen);
 
 int
-glusterd_update_fs_uuid (glusterd_brickinfo_t *brickinfo);
+glusterd_update_fs_label (glusterd_brickinfo_t *brickinfo);
 
 #endif