summaryrefslogtreecommitdiffstats
path: root/xlators/mgmt/glusterd/src/glusterd-brick-ops.c
diff options
context:
space:
mode:
Diffstat (limited to 'xlators/mgmt/glusterd/src/glusterd-brick-ops.c')
-rw-r--r--xlators/mgmt/glusterd/src/glusterd-brick-ops.c709
1 files changed, 530 insertions, 179 deletions
diff --git a/xlators/mgmt/glusterd/src/glusterd-brick-ops.c b/xlators/mgmt/glusterd/src/glusterd-brick-ops.c
index a14828e98..596503c21 100644
--- a/xlators/mgmt/glusterd/src/glusterd-brick-ops.c
+++ b/xlators/mgmt/glusterd/src/glusterd-brick-ops.c
@@ -335,7 +335,7 @@ out:
/* Handler functions */
int
-glusterd_handle_add_brick (rpcsvc_request_t *req)
+__glusterd_handle_add_brick (rpcsvc_request_t *req)
{
int32_t ret = -1;
gf_cli_req cli_req = {{0,}};
@@ -358,15 +358,16 @@ glusterd_handle_add_brick (rpcsvc_request_t *req)
GF_ASSERT (req);
- if (!xdr_to_generic (req->msg[0], &cli_req,
- (xdrproc_t)xdr_gf_cli_req)) {
+ ret = xdr_to_generic (req->msg[0], &cli_req,
+ (xdrproc_t)xdr_gf_cli_req);
+ if (ret < 0) {
//failed to decode msg;
req->rpc_err = GARBAGE_ARGS;
snprintf (err_str, sizeof (err_str), "Garbage args received");
goto out;
}
- gf_log ("glusterd", GF_LOG_INFO, "Received add brick req");
+ gf_log (this->name, GF_LOG_INFO, "Received add brick req");
if (cli_req.dict.dict_len) {
/* Unserialize the dictionary */
@@ -376,11 +377,11 @@ glusterd_handle_add_brick (rpcsvc_request_t *req)
cli_req.dict.dict_len,
&dict);
if (ret < 0) {
- gf_log ("glusterd", GF_LOG_ERROR,
+ gf_log (this->name, GF_LOG_ERROR,
"failed to "
"unserialize req-buffer to dictionary");
snprintf (err_str, sizeof (err_str), "Unable to decode "
- "the buffer");
+ "the command");
goto out;
}
}
@@ -388,45 +389,50 @@ glusterd_handle_add_brick (rpcsvc_request_t *req)
ret = dict_get_str (dict, "volname", &volname);
if (ret) {
- gf_log ("", GF_LOG_ERROR, "Unable to get volume name");
snprintf (err_str, sizeof (err_str), "Unable to get volume "
"name");
+ gf_log (this->name, GF_LOG_ERROR, "%s", err_str);
goto out;
}
if (!(ret = glusterd_check_volume_exists (volname))) {
ret = -1;
- snprintf(err_str, 2048, "Volume %s does not exist", volname);
- gf_log ("glusterd", GF_LOG_ERROR, "%s", err_str);
+ snprintf (err_str, sizeof (err_str), "Volume %s does not exist",
+ volname);
+ gf_log (this->name, GF_LOG_ERROR, "%s", err_str);
goto out;
}
ret = dict_get_int32 (dict, "count", &brick_count);
if (ret) {
- gf_log ("", GF_LOG_ERROR, "Unable to get count");
snprintf (err_str, sizeof (err_str), "Unable to get volume "
"brick count");
+ gf_log (this->name, GF_LOG_ERROR, "%s", err_str);
goto out;
}
ret = dict_get_int32 (dict, "replica-count", &replica_count);
if (!ret) {
- gf_log (THIS->name, GF_LOG_INFO, "replica-count is %d",
+ gf_log (this->name, GF_LOG_INFO, "replica-count is %d",
replica_count);
}
ret = dict_get_int32 (dict, "stripe-count", &stripe_count);
if (!ret) {
- gf_log (THIS->name, GF_LOG_INFO, "stripe-count is %d",
+ gf_log (this->name, GF_LOG_INFO, "stripe-count is %d",
stripe_count);
}
+ if (!dict_get (dict, "force")) {
+ gf_log (this->name, GF_LOG_ERROR, "Failed to get flag");
+ goto out;
+ }
ret = glusterd_volinfo_find (volname, &volinfo);
if (ret) {
snprintf (err_str, sizeof (err_str), "Unable to get volinfo "
"for volume name %s", volname);
- gf_log ("glusterd", GF_LOG_ERROR, "%s", err_str);
+ gf_log (this->name, GF_LOG_ERROR, "%s", err_str);
goto out;
}
@@ -442,10 +448,10 @@ glusterd_handle_add_brick (rpcsvc_request_t *req)
goto brick_val;
if ((brick_count % volinfo->dist_leaf_count) != 0) {
- snprintf(err_str, 2048, "Incorrect number of bricks"
- " supplied %d with count %d",
+ snprintf (err_str, sizeof (err_str), "Incorrect number "
+ "of bricks supplied %d with count %d",
brick_count, volinfo->dist_leaf_count);
- gf_log("glusterd", GF_LOG_ERROR, "%s", err_str);
+ gf_log (this->name, GF_LOG_ERROR, "%s", err_str);
ret = -1;
goto out;
}
@@ -461,7 +467,7 @@ glusterd_handle_add_brick (rpcsvc_request_t *req)
err_str,
sizeof (err_str));
if (ret == -1) {
- gf_log ("glusterd", GF_LOG_ERROR, "%s", err_str);
+ gf_log (this->name, GF_LOG_ERROR, "%s", err_str);
goto out;
}
@@ -471,7 +477,7 @@ glusterd_handle_add_brick (rpcsvc_request_t *req)
ret = dict_set_int32 (dict, "stripe-count", stripe_count);
if (ret) {
- gf_log (THIS->name, GF_LOG_ERROR,
+ gf_log (this->name, GF_LOG_ERROR,
"failed to set the stripe-count in dict");
goto out;
}
@@ -483,7 +489,7 @@ glusterd_handle_add_brick (rpcsvc_request_t *req)
&type, err_str,
sizeof (err_str));
if (ret == -1) {
- gf_log ("glusterd", GF_LOG_ERROR, "%s", err_str);
+ gf_log (this->name, GF_LOG_ERROR, "%s", err_str);
goto out;
}
@@ -493,7 +499,7 @@ glusterd_handle_add_brick (rpcsvc_request_t *req)
ret = dict_set_int32 (dict, "replica-count", replica_count);
if (ret) {
- gf_log (THIS->name, GF_LOG_ERROR,
+ gf_log (this->name, GF_LOG_ERROR,
"failed to set the replica-count in dict");
goto out;
}
@@ -503,14 +509,14 @@ brick_val:
if (ret) {
snprintf (err_str, sizeof (err_str), "Unable to get volume "
"bricks");
- gf_log ("glusterd", GF_LOG_ERROR, "%s", err_str);
+ gf_log (this->name, GF_LOG_ERROR, "%s", err_str);
goto out;
}
if (type != volinfo->type) {
ret = dict_set_int32 (dict, "type", type);
if (ret)
- gf_log (THIS->name, GF_LOG_ERROR,
+ gf_log (this->name, GF_LOG_ERROR,
"failed to set the new type in dict");
}
@@ -526,8 +532,6 @@ out:
cli_rsp = &rsp;
glusterd_to_cli (req, cli_rsp, NULL, 0, NULL,
(xdrproc_t)xdr_gf_cli_rsp, dict);
- if (dict)
- dict_unref (dict);
ret = 0; //sent error to cli, prevent second reply
}
@@ -536,9 +540,80 @@ out:
return ret;
}
+int
+glusterd_handle_add_brick (rpcsvc_request_t *req)
+{
+ return glusterd_big_locked_handler (req, __glusterd_handle_add_brick);
+}
+
+static int
+subvol_matcher_init (int **subvols, int count)
+{
+ int ret = -1;
+
+ *subvols = GF_CALLOC (count, sizeof(int), gf_gld_mt_int);
+ if (*subvols)
+ ret = 0;
+
+ return ret;
+}
+
+static void
+subvol_matcher_update (int *subvols, glusterd_volinfo_t *volinfo,
+ glusterd_brickinfo_t *brickinfo)
+{
+ glusterd_brickinfo_t *tmp = NULL;
+ int32_t sub_volume = 0;
+ int pos = 0;
+
+ list_for_each_entry (tmp, &volinfo->bricks, brick_list) {
+
+ if (strcmp (tmp->hostname, brickinfo->hostname) ||
+ strcmp (tmp->path, brickinfo->path)) {
+ pos++;
+ continue;
+ }
+ gf_log (THIS->name, GF_LOG_DEBUG, LOGSTR_FOUND_BRICK,
+ brickinfo->hostname, brickinfo->path,
+ volinfo->volname);
+ sub_volume = (pos / volinfo->dist_leaf_count);
+ subvols[sub_volume]++;
+ break;
+ }
+
+}
+
+static int
+subvol_matcher_verify (int *subvols, glusterd_volinfo_t *volinfo, char *err_str,
+ size_t err_len, char *vol_type)
+{
+ int i = 0;
+ int ret = 0;
+
+ do {
+
+ if (subvols[i] % volinfo->dist_leaf_count == 0) {
+ continue;
+ } else {
+ ret = -1;
+ snprintf (err_str, err_len,
+ "Bricks not from same subvol for %s", vol_type);
+ gf_log (THIS->name, GF_LOG_ERROR, "%s", err_str);
+ break;
+ }
+ } while (++i < volinfo->subvol_count);
+
+ return ret;
+}
+
+static void
+subvol_matcher_destroy (int *subvols)
+{
+ GF_FREE (subvols);
+}
int
-glusterd_handle_remove_brick (rpcsvc_request_t *req)
+__glusterd_handle_remove_brick (rpcsvc_request_t *req)
{
int32_t ret = -1;
gf_cli_req cli_req = {{0,}};
@@ -550,10 +625,7 @@ glusterd_handle_remove_brick (rpcsvc_request_t *req)
int i = 1;
glusterd_volinfo_t *volinfo = NULL;
glusterd_brickinfo_t *brickinfo = NULL;
- int32_t pos = 0;
- int32_t sub_volume = 0;
- int32_t sub_volume_start = 0;
- int32_t sub_volume_end = 0;
+ int *subvols = NULL;
glusterd_brickinfo_t *tmp = NULL;
char err_str[2048] = {0};
gf_cli_rsp rsp = {0,};
@@ -565,18 +637,23 @@ glusterd_handle_remove_brick (rpcsvc_request_t *req)
int found = 0;
int diff_count = 0;
char *volname = 0;
+ xlator_t *this = NULL;
GF_ASSERT (req);
+ this = THIS;
+ GF_ASSERT (this);
- if (!xdr_to_generic (req->msg[0], &cli_req,
- (xdrproc_t)xdr_gf_cli_req)) {
+ ret = xdr_to_generic (req->msg[0], &cli_req,
+ (xdrproc_t)xdr_gf_cli_req);
+ if (ret < 0) {
//failed to decode msg;
req->rpc_err = GARBAGE_ARGS;
+ snprintf (err_str, sizeof (err_str), "Received garbage args");
goto out;
}
- gf_log ("glusterd", GF_LOG_INFO, "Received rem brick req");
+ gf_log (this->name, GF_LOG_INFO, "Received rem brick req");
if (cli_req.dict.dict_len) {
/* Unserialize the dictionary */
@@ -586,36 +663,42 @@ glusterd_handle_remove_brick (rpcsvc_request_t *req)
cli_req.dict.dict_len,
&dict);
if (ret < 0) {
- gf_log ("glusterd", GF_LOG_ERROR,
+ gf_log (this->name, GF_LOG_ERROR,
"failed to "
"unserialize req-buffer to dictionary");
+ snprintf (err_str, sizeof (err_str), "Unable to decode "
+ "the command");
goto out;
}
}
ret = dict_get_str (dict, "volname", &volname);
if (ret) {
- gf_log (THIS->name, GF_LOG_ERROR, "Unable to get volname");
+ snprintf (err_str, sizeof (err_str), "Unable to get volume "
+ "name");
+ gf_log (this->name, GF_LOG_ERROR, "%s", err_str);
goto out;
}
ret = dict_get_int32 (dict, "count", &count);
if (ret) {
- gf_log ("", GF_LOG_ERROR, "Unable to get count");
+ snprintf (err_str, sizeof (err_str), "Unable to get brick "
+ "count");
+ gf_log (this->name, GF_LOG_ERROR, "%s", err_str);
goto out;
}
ret = glusterd_volinfo_find (volname, &volinfo);
if (ret) {
- snprintf (err_str, 2048, "Volume %s does not exist",
- volname);
- gf_log ("", GF_LOG_ERROR, "%s", err_str);
+ snprintf (err_str, sizeof (err_str),"Volume %s does not exist",
+ volname);
+ gf_log (this->name, GF_LOG_ERROR, "%s", err_str);
goto out;
}
ret = dict_get_int32 (dict, "replica-count", &replica_count);
if (!ret) {
- gf_log (THIS->name, GF_LOG_INFO,
+ gf_log (this->name, GF_LOG_INFO,
"request to change replica-count to %d", replica_count);
ret = gd_rmbr_validate_replica_count (volinfo, replica_count,
count, err_str,
@@ -632,7 +715,7 @@ glusterd_handle_remove_brick (rpcsvc_request_t *req)
ret = dict_set_int32 (dict, "replica-count",
replica_count);
if (ret) {
- gf_log (THIS->name, GF_LOG_WARNING,
+ gf_log (this->name, GF_LOG_WARNING,
"failed to set the replica_count "
"in dict");
goto out;
@@ -651,12 +734,12 @@ glusterd_handle_remove_brick (rpcsvc_request_t *req)
strcpy (vol_type, "distribute");
}
- /* Do not allow remove-brick if the volume is plain stripe */
+ /* Do not allow remove-brick if the volume is a stripe volume*/
if ((volinfo->type == GF_CLUSTER_TYPE_STRIPE) &&
(volinfo->brick_count == volinfo->stripe_count)) {
- snprintf (err_str, 2048,
- "Removing brick from a plain stripe is not allowed");
- gf_log ("glusterd", GF_LOG_ERROR, "%s", err_str);
+ snprintf (err_str, sizeof (err_str),
+ "Removing brick from a stripe volume is not allowed");
+ gf_log (this->name, GF_LOG_ERROR, "%s", err_str);
ret = -1;
goto out;
}
@@ -664,11 +747,11 @@ glusterd_handle_remove_brick (rpcsvc_request_t *req)
if (!replica_count &&
(volinfo->type == GF_CLUSTER_TYPE_STRIPE_REPLICATE) &&
(volinfo->brick_count == volinfo->dist_leaf_count)) {
- snprintf (err_str, 2048,
+ snprintf (err_str, sizeof(err_str),
"Removing bricks from stripe-replicate"
" configuration is not allowed without reducing "
"replica or stripe count explicitly.");
- gf_log (THIS->name, GF_LOG_ERROR, "%s", err_str);
+ gf_log (this->name, GF_LOG_ERROR, "%s", err_str);
ret = -1;
goto out;
}
@@ -676,11 +759,11 @@ glusterd_handle_remove_brick (rpcsvc_request_t *req)
if (!replica_count &&
(volinfo->type == GF_CLUSTER_TYPE_REPLICATE) &&
(volinfo->brick_count == volinfo->dist_leaf_count)) {
- snprintf (err_str, 2048,
+ snprintf (err_str, sizeof (err_str),
"Removing bricks from replicate configuration "
"is not allowed without reducing replica count "
"explicitly.");
- gf_log (THIS->name, GF_LOG_ERROR, "%s", err_str);
+ gf_log (this->name, GF_LOG_ERROR, "%s", err_str);
ret = -1;
goto out;
}
@@ -690,10 +773,10 @@ glusterd_handle_remove_brick (rpcsvc_request_t *req)
if (!replica_count && (volinfo->type != GF_CLUSTER_TYPE_NONE)) {
if (volinfo->dist_leaf_count &&
(count % volinfo->dist_leaf_count)) {
- snprintf (err_str, 2048, "Remove brick incorrect"
- " brick count of %d for %s %d",
+ snprintf (err_str, sizeof (err_str), "Remove brick "
+ "incorrect brick count of %d for %s %d",
count, vol_type, volinfo->dist_leaf_count);
- gf_log ("", GF_LOG_ERROR, "%s", err_str);
+ gf_log (this->name, GF_LOG_ERROR, "%s", err_str);
ret = -1;
goto out;
}
@@ -707,22 +790,32 @@ glusterd_handle_remove_brick (rpcsvc_request_t *req)
}
strcpy (brick_list, " ");
+
+ if ((volinfo->type != GF_CLUSTER_TYPE_NONE) &&
+ (volinfo->subvol_count > 1)) {
+ ret = subvol_matcher_init (&subvols, volinfo->subvol_count);
+ if (ret)
+ goto out;
+ }
+
while ( i <= count) {
- snprintf (key, 256, "brick%d", i);
+ snprintf (key, sizeof (key), "brick%d", i);
ret = dict_get_str (dict, key, &brick);
if (ret) {
- gf_log ("", GF_LOG_ERROR, "Unable to get %s", key);
+ snprintf (err_str, sizeof (err_str), "Unable to get %s",
+ key);
+ gf_log (this->name, GF_LOG_ERROR, "%s", err_str);
goto out;
}
- gf_log ("", GF_LOG_DEBUG, "Remove brick count %d brick: %s",
- i, brick);
+ gf_log (this->name, GF_LOG_DEBUG, "Remove brick count %d brick:"
+ " %s", i, brick);
ret = glusterd_volume_brickinfo_get_by_brick(brick, volinfo,
&brickinfo);
if (ret) {
- snprintf(err_str, 2048,"Incorrect brick %s for volume"
- " %s", brick, volname);
- gf_log ("", GF_LOG_ERROR, "%s", err_str);
+ snprintf (err_str, sizeof (err_str), "Incorrect brick "
+ "%s for volume %s", brick, volname);
+ gf_log (this->name, GF_LOG_ERROR, "%s", err_str);
goto out;
}
strcat(brick_list, brick);
@@ -743,7 +836,7 @@ glusterd_handle_remove_brick (rpcsvc_request_t *req)
found = 0;
list_for_each_entry (tmp, &volinfo->bricks, brick_list) {
tmp_brick_idx++;
- gf_log (THIS->name, GF_LOG_TRACE,
+ gf_log (this->name, GF_LOG_TRACE,
"validate brick %s:%s (%d %d %d)",
tmp->hostname, tmp->path, tmp_brick_idx,
brick_index, volinfo->replica_count);
@@ -761,8 +854,9 @@ glusterd_handle_remove_brick (rpcsvc_request_t *req)
if (found)
continue;
- snprintf(err_str, 2048,"Bricks are from same subvol");
- gf_log (THIS->name, GF_LOG_INFO,
+ snprintf (err_str, sizeof (err_str), "Bricks are from "
+ "same subvol");
+ gf_log (this->name, GF_LOG_INFO,
"failed to validate brick %s:%s (%d %d %d)",
tmp->hostname, tmp->path, tmp_brick_idx,
brick_index, volinfo->replica_count);
@@ -771,36 +865,18 @@ glusterd_handle_remove_brick (rpcsvc_request_t *req)
goto out;
}
- pos = 0;
- list_for_each_entry (tmp, &volinfo->bricks, brick_list) {
-
- if (strcmp (tmp->hostname,brickinfo->hostname) ||
- strcmp (tmp->path, brickinfo->path)) {
- pos++;
- continue;
- }
+ /* Find which subvolume the brick belongs to */
+ subvol_matcher_update (subvols, volinfo, brickinfo);
+ }
- gf_log ("", GF_LOG_INFO, "Found brick");
- if (!sub_volume && (volinfo->dist_leaf_count > 1)) {
- sub_volume = (pos / volinfo->dist_leaf_count) + 1;
- sub_volume_start = (volinfo->dist_leaf_count *
- (sub_volume - 1));
- sub_volume_end = (volinfo->dist_leaf_count *
- sub_volume) - 1;
- } else {
- if (pos < sub_volume_start ||
- pos >sub_volume_end) {
- ret = -1;
- snprintf(err_str, 2048,"Bricks not from"
- " same subvol for %s",
- vol_type);
- gf_log ("", GF_LOG_ERROR,
- "%s", err_str);
- goto out;
- }
- }
- break;
- }
+ /* Check if the bricks belong to the same subvolumes.*/
+ if ((volinfo->type != GF_CLUSTER_TYPE_NONE) &&
+ (volinfo->subvol_count > 1)) {
+ ret = subvol_matcher_verify (subvols, volinfo,
+ err_str, sizeof(err_str),
+ vol_type);
+ if (ret)
+ goto out;
}
ret = glusterd_op_begin_synctask (req, GD_OP_REMOVE_BRICK, dict);
@@ -810,24 +886,125 @@ out:
rsp.op_ret = -1;
rsp.op_errno = 0;
if (err_str[0] == '\0')
- snprintf (err_str, sizeof (err_str), "Operation failed");
- gf_log ("", GF_LOG_ERROR, "%s", err_str);
+ snprintf (err_str, sizeof (err_str),
+ "Operation failed");
+ gf_log (this->name, GF_LOG_ERROR, "%s", err_str);
rsp.op_errstr = err_str;
cli_rsp = &rsp;
glusterd_to_cli (req, cli_rsp, NULL, 0, NULL,
(xdrproc_t)xdr_gf_cli_rsp, dict);
- if (dict)
- dict_unref (dict);
ret = 0; //sent error to cli, prevent second reply
}
+
GF_FREE (brick_list);
+ subvol_matcher_destroy (subvols);
free (cli_req.dict.dict_val); //its malloced by xdr
return ret;
}
+int
+glusterd_handle_remove_brick (rpcsvc_request_t *req)
+{
+ return glusterd_big_locked_handler (req,
+ __glusterd_handle_remove_brick);
+}
+
+static int
+_glusterd_restart_gsync_session (dict_t *this, char *key,
+ data_t *value, void *data)
+{
+ char *slave = NULL;
+ char *slave_buf = NULL;
+ char *path_list = NULL;
+ char *slave_vol = NULL;
+ char *slave_ip = NULL;
+ char *conf_path = NULL;
+ char **errmsg = NULL;
+ int ret = -1;
+ glusterd_gsync_status_temp_t *param = NULL;
+ gf_boolean_t is_running = _gf_false;
+
+ param = (glusterd_gsync_status_temp_t *)data;
+
+ GF_ASSERT (param);
+ GF_ASSERT (param->volinfo);
+
+ slave = strchr(value->data, ':');
+ if (slave) {
+ slave++;
+ slave_buf = gf_strdup (slave);
+ if (!slave_buf) {
+ gf_log ("", GF_LOG_ERROR,
+ "Failed to gf_strdup");
+ ret = -1;
+ goto out;
+ }
+ }
+ else
+ return 0;
+
+ ret = dict_set_dynstr (param->rsp_dict, "slave", slave_buf);
+ if (ret) {
+ gf_log ("", GF_LOG_ERROR,
+ "Unable to store slave");
+ if (slave_buf)
+ GF_FREE(slave_buf);
+ goto out;
+ }
+
+ ret = glusterd_get_slave_details_confpath (param->volinfo,
+ param->rsp_dict,
+ &slave_ip, &slave_vol,
+ &conf_path, errmsg);
+ if (ret) {
+ if (*errmsg)
+ gf_log ("", GF_LOG_ERROR, "%s", *errmsg);
+ else
+ gf_log ("", GF_LOG_ERROR,
+ "Unable to fetch slave or confpath details.");
+ goto out;
+ }
+
+ /* In cases that gsyncd is not running, we will not invoke it
+ * because of add-brick. */
+ ret = glusterd_check_gsync_running_local (param->volinfo->volname,
+ slave, conf_path,
+ &is_running);
+ if (ret) {
+ gf_log ("", GF_LOG_ERROR, "gsync running validation failed.");
+ goto out;
+ }
+ if (_gf_false == is_running) {
+ gf_log ("", GF_LOG_DEBUG, "gsync session for %s and %s is"
+ " not running on this node. Hence not restarting.",
+ param->volinfo->volname, slave);
+ ret = 0;
+ goto out;
+ }
+
+ ret = glusterd_get_local_brickpaths (param->volinfo, &path_list);
+ if (!path_list) {
+ gf_log ("", GF_LOG_DEBUG, "This node not being part of"
+ " volume should not be running gsyncd. Hence"
+ " no gsyncd process to restart.");
+ ret = 0;
+ goto out;
+ }
+
+ ret = glusterd_check_restart_gsync_session (param->volinfo, slave,
+ param->rsp_dict, path_list,
+ conf_path, 0);
+ if (ret)
+ gf_log ("", GF_LOG_ERROR,
+ "Unable to restart gsync session.");
+
+out:
+ gf_log ("", GF_LOG_DEBUG, "Returning %d.", ret);
+ return ret;
+}
/* op-sm */
@@ -835,17 +1012,21 @@ int
glusterd_op_perform_add_bricks (glusterd_volinfo_t *volinfo, int32_t count,
char *bricks, dict_t *dict)
{
- glusterd_brickinfo_t *brickinfo = NULL;
- char *brick = NULL;
- int32_t i = 1;
- char *brick_list = NULL;
- char *free_ptr1 = NULL;
- char *free_ptr2 = NULL;
- char *saveptr = NULL;
- int32_t ret = -1;
- int32_t stripe_count = 0;
- int32_t replica_count = 0;
- int32_t type = 0;
+ char *brick = NULL;
+ int32_t i = 1;
+ char *brick_list = NULL;
+ char *free_ptr1 = NULL;
+ char *free_ptr2 = NULL;
+ char *saveptr = NULL;
+ int32_t ret = -1;
+ int32_t stripe_count = 0;
+ int32_t replica_count = 0;
+ int32_t type = 0;
+ glusterd_brickinfo_t *brickinfo = NULL;
+ glusterd_gsync_status_temp_t param = {0, };
+ gf_boolean_t restart_needed = 0;
+ char msg[1024] __attribute__((unused)) = {0, };
+ int caps = 0;
GF_ASSERT (volinfo);
@@ -903,13 +1084,15 @@ glusterd_op_perform_add_bricks (glusterd_volinfo_t *volinfo, int32_t count,
if (stripe_count) {
volinfo->stripe_count = stripe_count;
}
- volinfo->dist_leaf_count = (volinfo->stripe_count *
- volinfo->replica_count);
+ volinfo->dist_leaf_count = glusterd_get_dist_leaf_count (volinfo);
/* backward compatibility */
volinfo->sub_count = ((volinfo->dist_leaf_count == 1) ? 0:
volinfo->dist_leaf_count);
+ volinfo->subvol_count = (volinfo->brick_count /
+ volinfo->dist_leaf_count);
+
ret = glusterd_create_volfiles_and_notify_services (volinfo);
if (ret)
goto out;
@@ -924,13 +1107,40 @@ glusterd_op_perform_add_bricks (glusterd_volinfo_t *volinfo, int32_t count,
if (count)
brick = strtok_r (brick_list+1, " \n", &saveptr);
+#ifdef HAVE_BD_XLATOR
+ if (brickinfo->vg[0])
+ caps = CAPS_BD | CAPS_THIN |
+ CAPS_OFFLOAD_COPY | CAPS_OFFLOAD_SNAPSHOT;
+#endif
while (i <= count) {
-
ret = glusterd_volume_brickinfo_get_by_brick (brick, volinfo,
&brickinfo);
if (ret)
goto out;
+#ifdef HAVE_BD_XLATOR
+ /* Check for VG/thin pool if its BD volume */
+ if (brickinfo->vg[0]) {
+ ret = glusterd_is_valid_vg (brickinfo, 0, msg);
+ if (ret) {
+ gf_log (THIS->name, GF_LOG_CRITICAL, "%s", msg);
+ goto out;
+ }
+ /* if anyone of the brick does not have thin support,
+ disable it for entire volume */
+ caps &= brickinfo->caps;
+ } else
+ caps = 0;
+#endif
+
+ if (uuid_is_null (brickinfo->uuid)) {
+ ret = glusterd_resolve_brick (brickinfo);
+ if (ret) {
+ gf_log ("", GF_LOG_ERROR, FMTSTR_RESOLVE_BRICK,
+ brickinfo->hostname, brickinfo->path);
+ goto out;
+ }
+ }
ret = glusterd_brick_start (volinfo, brickinfo,
_gf_true);
@@ -938,8 +1148,27 @@ glusterd_op_perform_add_bricks (glusterd_volinfo_t *volinfo, int32_t count,
goto out;
i++;
brick = strtok_r (NULL, " \n", &saveptr);
+
+ /* Check if the brick is added in this node, and set
+ * the restart_needed flag. */
+ if ((!uuid_compare (brickinfo->uuid, MY_UUID)) &&
+ !restart_needed) {
+ restart_needed = 1;
+ gf_log ("", GF_LOG_DEBUG,
+ "Restart gsyncd session, if it's already "
+ "running.");
+ }
}
+ /* If the restart_needed flag is set, restart gsyncd sessions for that
+ * particular master with all the slaves. */
+ if (restart_needed) {
+ param.rsp_dict = dict;
+ param.volinfo = volinfo;
+ dict_foreach (volinfo->gsync_slaves,
+ _glusterd_restart_gsync_session, &param);
+ }
+ volinfo->caps = caps;
out:
GF_FREE (free_ptr1);
GF_FREE (free_ptr2);
@@ -1016,6 +1245,7 @@ glusterd_op_stage_add_brick (dict_t *dict, char **op_errstr)
gf_boolean_t brick_alloc = _gf_false;
char *all_bricks = NULL;
char *str_ret = NULL;
+ gf_boolean_t is_force = _gf_false;
priv = THIS->private;
if (!priv)
@@ -1035,15 +1265,6 @@ glusterd_op_stage_add_brick (dict_t *dict, char **op_errstr)
goto out;
}
- if (volinfo->backend == GD_VOL_BK_BD) {
- snprintf (msg, sizeof (msg), "Add brick is not supported for "
- "Block backend volume %s.", volname);
- gf_log (THIS->name, GF_LOG_ERROR, "%s", msg);
- *op_errstr = gf_strdup (msg);
- ret = -1;
- goto out;
- }
-
ret = glusterd_validate_volume_id (dict, volinfo);
if (ret)
goto out;
@@ -1078,6 +1299,8 @@ glusterd_op_stage_add_brick (dict_t *dict, char **op_errstr)
goto out;
}
+ is_force = dict_get_str_boolean (dict, "force", _gf_false);
+
if (bricks) {
brick_list = gf_strdup (bricks);
all_bricks = gf_strdup (bricks);
@@ -1119,10 +1342,21 @@ glusterd_op_stage_add_brick (dict_t *dict, char **op_errstr)
}
if (!uuid_compare (brickinfo->uuid, MY_UUID)) {
- ret = glusterd_brick_create_path (brickinfo->hostname,
- brickinfo->path,
+#ifdef HAVE_BD_XLATOR
+ if (brickinfo->vg[0]) {
+ ret = glusterd_is_valid_vg (brickinfo, 1, msg);
+ if (ret) {
+ gf_log (THIS->name, GF_LOG_ERROR, "%s",
+ msg);
+ *op_errstr = gf_strdup (msg);
+ goto out;
+ }
+ }
+#endif
+
+ ret = glusterd_validate_and_create_brickpath (brickinfo,
volinfo->volume_id,
- op_errstr);
+ op_errstr, is_force);
if (ret)
goto out;
}
@@ -1157,17 +1391,22 @@ glusterd_op_stage_remove_brick (dict_t *dict, char **op_errstr)
char msg[2048] = {0,};
int32_t flag = 0;
gf1_op_commands cmd = GF_OP_CMD_NONE;
+ char *task_id_str = NULL;
+ xlator_t *this = NULL;
+
+ this = THIS;
+ GF_ASSERT (this);
ret = dict_get_str (dict, "volname", &volname);
if (ret) {
- gf_log ("", GF_LOG_ERROR, "Unable to get volume name");
+ gf_log (this->name, GF_LOG_ERROR, "Unable to get volume name");
goto out;
}
ret = glusterd_volinfo_find (volname, &volinfo);
if (ret) {
- gf_log ("", GF_LOG_ERROR, "Volume %s does not exist", volname);
+ gf_log (this->name, GF_LOG_ERROR, "Volume %s does not exist", volname);
goto out;
}
@@ -1179,7 +1418,7 @@ glusterd_op_stage_remove_brick (dict_t *dict, char **op_errstr)
snprintf (msg, sizeof (msg), "Replace brick is in progress on "
"volume %s. Please retry after replace-brick "
"operation is committed or aborted", volname);
- gf_log (THIS->name, GF_LOG_ERROR, "%s", msg);
+ gf_log (this->name, GF_LOG_ERROR, "%s", msg);
*op_errstr = gf_strdup (msg);
ret = -1;
goto out;
@@ -1187,7 +1426,7 @@ glusterd_op_stage_remove_brick (dict_t *dict, char **op_errstr)
ret = dict_get_int32 (dict, "command", &flag);
if (ret) {
- gf_log ("", GF_LOG_ERROR, "Unable to get brick count");
+ gf_log (this->name, GF_LOG_ERROR, "Unable to get brick count");
goto out;
}
cmd = flag;
@@ -1204,21 +1443,60 @@ glusterd_op_stage_remove_brick (dict_t *dict, char **op_errstr)
case GF_OP_CMD_START:
{
+ if ((volinfo->type == GF_CLUSTER_TYPE_REPLICATE) &&
+ dict_get (dict, "replica-count")) {
+ snprintf (msg, sizeof(msg), "Migration of data is not "
+ "needed when reducing replica count. Use the"
+ " 'force' option");
+ errstr = gf_strdup (msg);
+ gf_log (this->name, GF_LOG_ERROR, "%s", errstr);
+ goto out;
+ }
+
if (GLUSTERD_STATUS_STARTED != volinfo->status) {
- snprintf (msg, sizeof (msg), "Volume %s needs to be started "
- "before remove-brick (you can use 'force' or "
- "'commit' to override this behavior)",
+ snprintf (msg, sizeof (msg), "Volume %s needs to be "
+ "started before remove-brick (you can use "
+ "'force' or 'commit' to override this "
+ "behavior)", volinfo->volname);
+ errstr = gf_strdup (msg);
+ gf_log (this->name, GF_LOG_ERROR, "%s", errstr);
+ goto out;
+ }
+ if (!gd_is_remove_brick_committed (volinfo)) {
+ snprintf (msg, sizeof (msg), "An earlier remove-brick "
+ "task exists for volume %s. Either commit it"
+ " or stop it before starting a new task.",
volinfo->volname);
errstr = gf_strdup (msg);
- gf_log (THIS->name, GF_LOG_ERROR, "%s", errstr);
+ gf_log (this->name, GF_LOG_ERROR, "Earlier remove-brick"
+ " task exists for volume %s.",
+ volinfo->volname);
goto out;
}
if (glusterd_is_defrag_on(volinfo)) {
- errstr = gf_strdup("Rebalance is in progress. Please retry"
- " after completion");
- gf_log ("glusterd", GF_LOG_ERROR, "%s", errstr);
+ errstr = gf_strdup("Rebalance is in progress. Please "
+ "retry after completion");
+ gf_log (this->name, GF_LOG_ERROR, "%s", errstr);
goto out;
}
+
+ if (is_origin_glusterd (dict)) {
+ ret = glusterd_generate_and_set_task_id
+ (dict, GF_REMOVE_BRICK_TID_KEY);
+ if (ret) {
+ gf_log (this->name, GF_LOG_ERROR,
+ "Failed to generate task-id");
+ goto out;
+ }
+ } else {
+ ret = dict_get_str (dict, GF_REMOVE_BRICK_TID_KEY,
+ &task_id_str);
+ if (ret) {
+ gf_log (this->name, GF_LOG_WARNING,
+ "Missing remove-brick-id");
+ ret = 0;
+ }
+ }
break;
}
@@ -1240,7 +1518,7 @@ glusterd_op_stage_remove_brick (dict_t *dict, char **op_errstr)
ret = dict_get_int32 (dict, "count", &brick_count);
if (ret) {
- gf_log ("", GF_LOG_ERROR, "Unable to get brick count");
+ gf_log (this->name, GF_LOG_ERROR, "Unable to get brick count");
goto out;
}
@@ -1253,7 +1531,7 @@ glusterd_op_stage_remove_brick (dict_t *dict, char **op_errstr)
}
out:
- gf_log ("", GF_LOG_DEBUG, "Returning %d", ret);
+ gf_log (this->name, GF_LOG_DEBUG, "Returning %d", ret);
if (ret && errstr) {
if (op_errstr)
*op_errstr = errstr;
@@ -1389,15 +1667,6 @@ glusterd_op_add_brick (dict_t *dict, char **op_errstr)
goto out;
}
- /* Need to reset the defrag/rebalance status accordingly */
- switch (volinfo->defrag_status) {
- case GF_DEFRAG_STATUS_FAILED:
- case GF_DEFRAG_STATUS_COMPLETE:
- volinfo->defrag_status = 0;
- default:
- break;
- }
-
ret = glusterd_store_volinfo (volinfo, GLUSTERD_VOLINFO_VER_AC_INCREMENT);
if (ret)
goto out;
@@ -1412,42 +1681,74 @@ out:
int
glusterd_op_remove_brick (dict_t *dict, char **op_errstr)
{
- int ret = -1;
- char *volname = NULL;
- glusterd_volinfo_t *volinfo = NULL;
- char *brick = NULL;
- int32_t count = 0;
- int32_t i = 1;
- char key[256] = {0,};
- int32_t flag = 0;
- char err_str[4096] = {0,};
- int need_rebalance = 0;
- int force = 0;
- gf1_op_commands cmd = 0;
- int32_t replica_count = 0;
- glusterd_brickinfo_t *brickinfo = NULL;
- glusterd_brickinfo_t *tmp = NULL;
+ int ret = -1;
+ char *volname = NULL;
+ glusterd_volinfo_t *volinfo = NULL;
+ char *brick = NULL;
+ int32_t count = 0;
+ int32_t i = 1;
+ char key[256] = {0,};
+ int32_t flag = 0;
+ char err_str[4096] = {0,};
+ int need_rebalance = 0;
+ int force = 0;
+ gf1_op_commands cmd = 0;
+ int32_t replica_count = 0;
+ glusterd_brickinfo_t *brickinfo = NULL;
+ glusterd_brickinfo_t *tmp = NULL;
+ char *task_id_str = NULL;
+ xlator_t *this = NULL;
+ dict_t *bricks_dict = NULL;
+ char *brick_tmpstr = NULL;
+
+ this = THIS;
+ GF_ASSERT (this);
ret = dict_get_str (dict, "volname", &volname);
if (ret) {
- gf_log ("", GF_LOG_ERROR, "Unable to get volume name");
+ gf_log (this->name, GF_LOG_ERROR, "Unable to get volume name");
goto out;
}
ret = glusterd_volinfo_find (volname, &volinfo);
if (ret) {
- gf_log ("", GF_LOG_ERROR, "Unable to allocate memory");
+ gf_log (this->name, GF_LOG_ERROR, "Unable to allocate memory");
goto out;
}
ret = dict_get_int32 (dict, "command", &flag);
if (ret) {
- gf_log ("", GF_LOG_ERROR, "Unable to get brick count");
+ gf_log (this->name, GF_LOG_ERROR, "Unable to get command");
goto out;
}
cmd = flag;
+ /* Set task-id, if available, in ctx dict for operations other than
+ * start
+ */
+ if (is_origin_glusterd (dict) && (cmd != GF_OP_CMD_START)) {
+ if (!uuid_is_null (volinfo->rebal.rebalance_id)) {
+ ret = glusterd_copy_uuid_to_dict
+ (volinfo->rebal.rebalance_id, dict,
+ GF_REMOVE_BRICK_TID_KEY);
+ if (ret) {
+ gf_log (this->name, GF_LOG_ERROR,
+ "Failed to set remove-brick-id");
+ goto out;
+ }
+ }
+ }
+
+ /* Clear task-id, rebal.op and stored bricks on commmitting/stopping
+ * remove-brick */
+ if ((cmd != GF_OP_CMD_START) || (cmd != GF_OP_CMD_STATUS)) {
+ uuid_clear (volinfo->rebal.rebalance_id);
+ volinfo->rebal.op = GD_OP_NONE;
+ dict_unref (volinfo->rebal.dict);
+ volinfo->rebal.dict = NULL;
+ }
+
ret = -1;
switch (cmd) {
case GF_OP_CMD_NONE:
@@ -1468,7 +1769,7 @@ glusterd_op_remove_brick (dict_t *dict, char **op_errstr)
}
ret = glusterd_create_volfiles_and_notify_services (volinfo);
if (ret) {
- gf_log (THIS->name, GF_LOG_WARNING,
+ gf_log (this->name, GF_LOG_WARNING,
"failed to create volfiles");
goto out;
}
@@ -1476,7 +1777,7 @@ glusterd_op_remove_brick (dict_t *dict, char **op_errstr)
ret = glusterd_store_volinfo (volinfo,
GLUSTERD_VOLINFO_VER_AC_INCREMENT);
if (ret) {
- gf_log (THIS->name, GF_LOG_WARNING,
+ gf_log (this->name, GF_LOG_WARNING,
"failed to store volinfo");
goto out;
}
@@ -1486,6 +1787,15 @@ glusterd_op_remove_brick (dict_t *dict, char **op_errstr)
}
case GF_OP_CMD_START:
+ ret = dict_get_str (dict, GF_REMOVE_BRICK_TID_KEY, &task_id_str);
+ if (ret) {
+ gf_log (this->name, GF_LOG_DEBUG,
+ "Missing remove-brick-id");
+ ret = 0;
+ } else {
+ uuid_parse (task_id_str, volinfo->rebal.rebalance_id) ;
+ volinfo->rebal.op = GD_OP_REMOVE_BRICK;
+ }
force = 0;
break;
@@ -1496,13 +1806,14 @@ glusterd_op_remove_brick (dict_t *dict, char **op_errstr)
case GF_OP_CMD_COMMIT_FORCE:
if (volinfo->decommission_in_progress) {
- if (volinfo->defrag) {
- LOCK (&volinfo->defrag->lock);
+ if (volinfo->rebal.defrag) {
+ LOCK (&volinfo->rebal.defrag->lock);
/* Fake 'rebalance-complete' so the graph change
happens right away */
- volinfo->defrag_status = GF_DEFRAG_STATUS_COMPLETE;
+ volinfo->rebal.defrag_status =
+ GF_DEFRAG_STATUS_COMPLETE;
- UNLOCK (&volinfo->defrag->lock);
+ UNLOCK (&volinfo->rebal.defrag->lock);
}
/* Graph change happens in rebalance _cbk function,
no need to do anything here */
@@ -1520,15 +1831,44 @@ glusterd_op_remove_brick (dict_t *dict, char **op_errstr)
goto out;
}
-
+ /* Save the list of bricks for later usage. Right now this is required
+ * for displaying the task parameters with task status in volume status.
+ */
+ bricks_dict = dict_new ();
+ if (!bricks_dict) {
+ ret = -1;
+ goto out;
+ }
+ ret = dict_set_int32 (bricks_dict, "count", count);
+ if (ret) {
+ gf_log (this->name, GF_LOG_ERROR,
+ "Failed to save remove-brick count");
+ goto out;
+ }
while ( i <= count) {
snprintf (key, 256, "brick%d", i);
ret = dict_get_str (dict, key, &brick);
if (ret) {
- gf_log ("", GF_LOG_ERROR, "Unable to get %s", key);
+ gf_log (this->name, GF_LOG_ERROR, "Unable to get %s",
+ key);
goto out;
}
+ brick_tmpstr = gf_strdup (brick);
+ if (!brick_tmpstr) {
+ ret = -1;
+ gf_log (this->name, GF_LOG_ERROR,
+ "Failed to duplicate brick name");
+ goto out;
+ }
+ ret = dict_set_dynstr (bricks_dict, key, brick_tmpstr);
+ if (ret) {
+ gf_log (this->name, GF_LOG_ERROR,
+ "Failed to add brick to dict");
+ goto out;
+ }
+ brick_tmpstr = NULL;
+
ret = glusterd_op_perform_remove_brick (volinfo, brick, force,
&need_rebalance);
if (ret)
@@ -1537,13 +1877,16 @@ glusterd_op_remove_brick (dict_t *dict, char **op_errstr)
}
ret = dict_get_int32 (dict, "replica-count", &replica_count);
if (!ret) {
- gf_log (THIS->name, GF_LOG_INFO,
+ gf_log (this->name, GF_LOG_INFO,
"changing replica count %d to %d on volume %s",
volinfo->replica_count, replica_count,
volinfo->volname);
volinfo->replica_count = replica_count;
- volinfo->dist_leaf_count = (volinfo->stripe_count *
- replica_count);
+ volinfo->sub_count = replica_count;
+ volinfo->dist_leaf_count = glusterd_get_dist_leaf_count (volinfo);
+ volinfo->subvol_count = (volinfo->brick_count /
+ volinfo->dist_leaf_count);
+
if (replica_count == 1) {
if (volinfo->type == GF_CLUSTER_TYPE_REPLICATE) {
volinfo->type = GF_CLUSTER_TYPE_NONE;
@@ -1556,37 +1899,41 @@ glusterd_op_remove_brick (dict_t *dict, char **op_errstr)
}
}
}
+ volinfo->rebal.dict = bricks_dict;
+ bricks_dict = NULL;
ret = glusterd_create_volfiles_and_notify_services (volinfo);
if (ret) {
- gf_log (THIS->name, GF_LOG_WARNING, "failed to create volfiles");
+ gf_log (this->name, GF_LOG_WARNING, "failed to create volfiles");
goto out;
}
ret = glusterd_store_volinfo (volinfo, GLUSTERD_VOLINFO_VER_AC_INCREMENT);
if (ret) {
- gf_log (THIS->name, GF_LOG_WARNING, "failed to store volinfo");
+ gf_log (this->name, GF_LOG_WARNING, "failed to store volinfo");
goto out;
}
/* Need to reset the defrag/rebalance status accordingly */
- switch (volinfo->defrag_status) {
+ switch (volinfo->rebal.defrag_status) {
case GF_DEFRAG_STATUS_FAILED:
case GF_DEFRAG_STATUS_COMPLETE:
- volinfo->defrag_status = 0;
+ volinfo->rebal.defrag_status = 0;
default:
break;
}
if (!force && need_rebalance) {
/* perform the rebalance operations */
- ret = glusterd_handle_defrag_start (volinfo, err_str, 4096,
- GF_DEFRAG_CMD_START_FORCE,
- glusterd_remove_brick_migrate_cbk);
+ ret = glusterd_handle_defrag_start
+ (volinfo, err_str, sizeof (err_str),
+ GF_DEFRAG_CMD_START_FORCE,
+ glusterd_remove_brick_migrate_cbk, GD_OP_REMOVE_BRICK);
+
if (!ret)
volinfo->decommission_in_progress = 1;
if (ret) {
- gf_log (THIS->name, GF_LOG_ERROR,
+ gf_log (this->name, GF_LOG_ERROR,
"failed to start the rebalance");
}
} else {
@@ -1598,5 +1945,9 @@ out:
if (ret && err_str[0] && op_errstr)
*op_errstr = gf_strdup (err_str);
+ GF_FREE (brick_tmpstr);
+ if (bricks_dict)
+ dict_unref (bricks_dict);
+
return ret;
}
generated by cgit (git 2.34.1) at 2026-01-05 20:54:25 +0000