summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
Diffstat
-rw-r--r--libglusterfs/src/mem-types.h1
-rw-r--r--xlators/mgmt/glusterd/src/glusterd-handler.c20
-rw-r--r--xlators/mgmt/glusterd/src/glusterd-mgmt-handler.c4
-rw-r--r--xlators/mgmt/glusterd/src/glusterd-mgmt.c319
-rw-r--r--xlators/mgmt/glusterd/src/glusterd-op-sm.c67
-rw-r--r--xlators/mgmt/glusterd/src/glusterd-op-sm.h3
-rw-r--r--xlators/mgmt/glusterd/src/glusterd-peer-utils.c13
-rw-r--r--xlators/mgmt/glusterd/src/glusterd-sm.h1
-rw-r--r--xlators/mgmt/glusterd/src/glusterd-syncop.c250
-rw-r--r--xlators/mgmt/glusterd/src/glusterd-syncop.h9
-rw-r--r--xlators/mgmt/glusterd/src/glusterd-utils.c19
-rw-r--r--xlators/mgmt/glusterd/src/glusterd-utils.h14
-rw-r--r--xlators/mgmt/glusterd/src/glusterd.h3
13 files changed, 404 insertions, 319 deletions
diff --git a/libglusterfs/src/mem-types.h b/libglusterfs/src/mem-types.h
index f4d3974f0b2..fc06d52239b 100644
--- a/libglusterfs/src/mem-types.h
+++ b/libglusterfs/src/mem-types.h
@@ -150,7 +150,6 @@ enum gf_common_mem_types_ {
gf_common_mt_nfs_exports = 131,
gf_common_mt_gf_brick_spec_t = 132,
gf_common_mt_gf_timer_entry_t = 133,
- gf_common_mt_list_head_t = 134,
gf_common_mt_end
};
#endif
diff --git a/xlators/mgmt/glusterd/src/glusterd-handler.c b/xlators/mgmt/glusterd/src/glusterd-handler.c
index 469c95c9890..ac69fc8712d 100644
--- a/xlators/mgmt/glusterd/src/glusterd-handler.c
+++ b/xlators/mgmt/glusterd/src/glusterd-handler.c
@@ -626,22 +626,6 @@ glusterd_op_txn_begin (rpcsvc_request_t *req, glusterd_op_t op, void *ctx,
gf_log (this->name, GF_LOG_DEBUG, "Acquired lock on localhost");
local_locking_done:
- txn_op_info.local_xaction_peers =
- GF_CALLOC (1, sizeof (struct cds_list_head),
- gf_common_mt_list_head_t);
- if (!txn_op_info.local_xaction_peers) {
- ret = -1;
- gf_log (this->name, GF_LOG_ERROR, "Out of memory");
- goto out;
- }
- CDS_INIT_LIST_HEAD (txn_op_info.local_xaction_peers);
-
- /* Maintain xaction_peers on per transaction basis */
- npeers = gd_build_local_xaction_peers_list
- (&priv->peers,
- txn_op_info.local_xaction_peers,
- op);
-
/* If no volname is given as a part of the command, locks will
* not be held, hence sending stage event. */
if (volname || (priv->op_version < GD_OP_VERSION_3_6_0))
@@ -898,8 +882,8 @@ __glusterd_handle_stage_op (rpcsvc_request_t *req)
"No transaction's opinfo set");
state.state = GD_OP_STATE_LOCKED;
- glusterd_txn_opinfo_init (&txn_op_info, &state,
- &op_req.op, req_ctx->dict, req);
+ glusterd_txn_opinfo_init (&txn_op_info, &state, &op_req.op,
+ req_ctx->dict, req);
ret = glusterd_set_txn_opinfo (txn_id, &txn_op_info);
if (ret) {
diff --git a/xlators/mgmt/glusterd/src/glusterd-mgmt-handler.c b/xlators/mgmt/glusterd/src/glusterd-mgmt-handler.c
index a200c8e1230..859690eee65 100644
--- a/xlators/mgmt/glusterd/src/glusterd-mgmt-handler.c
+++ b/xlators/mgmt/glusterd/src/glusterd-mgmt-handler.c
@@ -93,8 +93,8 @@ glusterd_op_state_machine_mgmt_v3_lock (rpcsvc_request_t *req,
GF_ASSERT (this);
GF_ASSERT (req);
- glusterd_txn_opinfo_init (&txn_op_info, NULL, &lock_req->op,
- ctx->dict, req);
+ glusterd_txn_opinfo_init (&txn_op_info, NULL, &lock_req->op, ctx->dict,
+ req);
ret = glusterd_set_txn_opinfo (&lock_req->txn_id, &txn_op_info);
if (ret) {
diff --git a/xlators/mgmt/glusterd/src/glusterd-mgmt.c b/xlators/mgmt/glusterd/src/glusterd-mgmt.c
index 2a34c78b35b..4e4dd047281 100644
--- a/xlators/mgmt/glusterd/src/glusterd-mgmt.c
+++ b/xlators/mgmt/glusterd/src/glusterd-mgmt.c
@@ -383,9 +383,9 @@ out:
int
glusterd_mgmt_v3_initiate_lockdown (glusterd_op_t op, dict_t *dict,
- char **op_errstr, int npeers,
+ char **op_errstr,
gf_boolean_t *is_acquired,
- struct cds_list_head *peers)
+ uint64_t txn_generation)
{
char *volname = NULL;
glusterd_peerinfo_t *peerinfo = NULL;
@@ -394,9 +394,13 @@ glusterd_mgmt_v3_initiate_lockdown (glusterd_op_t op, dict_t *dict,
struct syncargs args = {0};
uuid_t peer_uuid = {0};
xlator_t *this = NULL;
+ glusterd_conf_t *conf = NULL;
this = THIS;
GF_ASSERT (this);
+ conf = this->private;
+ GF_ASSERT (conf);
+
GF_ASSERT (dict);
GF_ASSERT (op_errstr);
GF_ASSERT (is_acquired);
@@ -411,20 +415,36 @@ glusterd_mgmt_v3_initiate_lockdown (glusterd_op_t op, dict_t *dict,
*is_acquired = _gf_true;
- if (!npeers) {
- ret = 0;
- goto out;
- }
-
/* Sending mgmt_v3 lock req to other nodes in the cluster */
gd_syncargs_init (&args, NULL);
synctask_barrier_init((&args));
peer_cnt = 0;
- list_for_each_local_xaction_peers (peerinfo, peers) {
+
+ rcu_read_lock ();
+ cds_list_for_each_entry_rcu (peerinfo, &conf->peers, uuid_list) {
+ /* Only send requests to peers who were available before the
+ * transaction started
+ */
+ if (peerinfo->generation > txn_generation)
+ continue;
+
+ if (!peerinfo->connected)
+ continue;
+ if (op != GD_OP_SYNC_VOLUME &&
+ peerinfo->state.state != GD_FRIEND_STATE_BEFRIENDED)
+ continue;
+
gd_mgmt_v3_lock (op, dict, peerinfo, &args,
MY_UUID, peer_uuid);
peer_cnt++;
}
+ rcu_read_unlock ();
+
+ if (0 == peer_cnt) {
+ ret = 0;
+ goto out;
+ }
+
gd_synctask_barrier_wait((&args), peer_cnt);
if (args.errstr)
@@ -633,8 +653,8 @@ out:
int
glusterd_mgmt_v3_pre_validate (glusterd_op_t op, dict_t *req_dict,
- char **op_errstr, int npeers,
- struct cds_list_head *peers)
+ char **op_errstr,
+ uint64_t txn_generation)
{
int32_t ret = -1;
int32_t peer_cnt = 0;
@@ -643,9 +663,13 @@ glusterd_mgmt_v3_pre_validate (glusterd_op_t op, dict_t *req_dict,
struct syncargs args = {0};
uuid_t peer_uuid = {0};
xlator_t *this = NULL;
+ glusterd_conf_t *conf = NULL;
this = THIS;
GF_ASSERT (this);
+ conf = this->private;
+ GF_ASSERT (conf);
+
GF_ASSERT (req_dict);
GF_ASSERT (op_errstr);
@@ -691,20 +715,36 @@ glusterd_mgmt_v3_pre_validate (glusterd_op_t op, dict_t *req_dict,
dict_unref (rsp_dict);
rsp_dict = NULL;
- if (!npeers) {
- ret = 0;
- goto out;
- }
-
/* Sending Pre Validation req to other nodes in the cluster */
gd_syncargs_init (&args, req_dict);
synctask_barrier_init((&args));
peer_cnt = 0;
- list_for_each_local_xaction_peers (peerinfo, peers) {
+
+ rcu_read_lock ();
+ cds_list_for_each_entry_rcu (peerinfo, &conf->peers, uuid_list) {
+ /* Only send requests to peers who were available before the
+ * transaction started
+ */
+ if (peerinfo->generation > txn_generation)
+ continue;
+
+ if (!peerinfo->connected)
+ continue;
+ if (op != GD_OP_SYNC_VOLUME &&
+ peerinfo->state.state != GD_FRIEND_STATE_BEFRIENDED)
+ continue;
+
gd_mgmt_v3_pre_validate_req (op, req_dict, peerinfo, &args,
MY_UUID, peer_uuid);
peer_cnt++;
}
+ rcu_read_unlock ();
+
+ if (0 == peer_cnt) {
+ ret = 0;
+ goto out;
+ }
+
gd_synctask_barrier_wait((&args), peer_cnt);
if (args.op_ret) {
@@ -865,9 +905,8 @@ out:
}
int
-glusterd_mgmt_v3_brick_op (glusterd_op_t op, dict_t *req_dict,
- char **op_errstr, int npeers,
- struct cds_list_head *peers)
+glusterd_mgmt_v3_brick_op (glusterd_op_t op, dict_t *req_dict, char **op_errstr,
+ uint64_t txn_generation)
{
int32_t ret = -1;
int32_t peer_cnt = 0;
@@ -876,9 +915,13 @@ glusterd_mgmt_v3_brick_op (glusterd_op_t op, dict_t *req_dict,
struct syncargs args = {0};
uuid_t peer_uuid = {0};
xlator_t *this = NULL;
+ glusterd_conf_t *conf = NULL;
this = THIS;
GF_ASSERT (this);
+ conf = this->private;
+ GF_ASSERT (conf);
+
GF_ASSERT (req_dict);
GF_ASSERT (op_errstr);
@@ -915,20 +958,36 @@ glusterd_mgmt_v3_brick_op (glusterd_op_t op, dict_t *req_dict,
dict_unref (rsp_dict);
rsp_dict = NULL;
- if (!npeers) {
- ret = 0;
- goto out;
- }
-
/* Sending brick op req to other nodes in the cluster */
gd_syncargs_init (&args, NULL);
synctask_barrier_init((&args));
peer_cnt = 0;
- list_for_each_local_xaction_peers (peerinfo, peers) {
+
+ rcu_read_lock ();
+ cds_list_for_each_entry_rcu (peerinfo, &conf->peers, uuid_list) {
+ /* Only send requests to peers who were available before the
+ * transaction started
+ */
+ if (peerinfo->generation > txn_generation)
+ continue;
+
+ if (!peerinfo->connected)
+ continue;
+ if (op != GD_OP_SYNC_VOLUME &&
+ peerinfo->state.state != GD_FRIEND_STATE_BEFRIENDED)
+ continue;
+
gd_mgmt_v3_brick_op_req (op, req_dict, peerinfo, &args,
MY_UUID, peer_uuid);
peer_cnt++;
}
+ rcu_read_unlock ();
+
+ if (0 == peer_cnt) {
+ ret = 0;
+ goto out;
+ }
+
gd_synctask_barrier_wait((&args), peer_cnt);
if (args.op_ret) {
@@ -1084,9 +1143,8 @@ out:
}
int
-glusterd_mgmt_v3_commit (glusterd_op_t op, dict_t *op_ctx,
- dict_t *req_dict, char **op_errstr,
- int npeers, struct cds_list_head *peers)
+glusterd_mgmt_v3_commit (glusterd_op_t op, dict_t *op_ctx, dict_t *req_dict,
+ char **op_errstr, uint64_t txn_generation)
{
int32_t ret = -1;
int32_t peer_cnt = 0;
@@ -1095,9 +1153,13 @@ glusterd_mgmt_v3_commit (glusterd_op_t op, dict_t *op_ctx,
struct syncargs args = {0};
uuid_t peer_uuid = {0};
xlator_t *this = NULL;
+ glusterd_conf_t *conf = NULL;
this = THIS;
GF_ASSERT (this);
+ conf = this->private;
+ GF_ASSERT (conf);
+
GF_ASSERT (op_ctx);
GF_ASSERT (req_dict);
GF_ASSERT (op_errstr);
@@ -1144,20 +1206,36 @@ glusterd_mgmt_v3_commit (glusterd_op_t op, dict_t *op_ctx,
dict_unref (rsp_dict);
rsp_dict = NULL;
- if (!npeers) {
- ret = 0;
- goto out;
- }
-
/* Sending commit req to other nodes in the cluster */
gd_syncargs_init (&args, op_ctx);
synctask_barrier_init((&args));
peer_cnt = 0;
- list_for_each_local_xaction_peers (peerinfo, peers) {
+
+ rcu_read_lock ();
+ cds_list_for_each_entry_rcu (peerinfo, &conf->peers, uuid_list) {
+ /* Only send requests to peers who were available before the
+ * transaction started
+ */
+ if (peerinfo->generation > txn_generation)
+ continue;
+
+ if (!peerinfo->connected)
+ continue;
+ if (op != GD_OP_SYNC_VOLUME &&
+ peerinfo->state.state != GD_FRIEND_STATE_BEFRIENDED)
+ continue;
+
gd_mgmt_v3_commit_req (op, req_dict, peerinfo, &args,
MY_UUID, peer_uuid);
peer_cnt++;
}
+ rcu_read_unlock ();
+
+ if (0 == peer_cnt) {
+ ret = 0;
+ goto out;
+ }
+
gd_synctask_barrier_wait((&args), peer_cnt);
if (args.op_ret) {
@@ -1282,8 +1360,8 @@ out:
int
glusterd_mgmt_v3_post_validate (glusterd_op_t op, int32_t op_ret, dict_t *dict,
- dict_t *req_dict, char **op_errstr, int npeers,
- struct cds_list_head *peers)
+ dict_t *req_dict, char **op_errstr,
+ uint64_t txn_generation)
{
int32_t ret = -1;
int32_t peer_cnt = 0;
@@ -1292,9 +1370,13 @@ glusterd_mgmt_v3_post_validate (glusterd_op_t op, int32_t op_ret, dict_t *dict,
struct syncargs args = {0};
uuid_t peer_uuid = {0};
xlator_t *this = NULL;
+ glusterd_conf_t *conf = NULL;
this = THIS;
GF_ASSERT (this);
+ conf = this->private;
+ GF_ASSERT (conf);
+
GF_ASSERT (dict);
GF_VALIDATE_OR_GOTO (this->name, req_dict, out);
GF_ASSERT (op_errstr);
@@ -1335,20 +1417,36 @@ glusterd_mgmt_v3_post_validate (glusterd_op_t op, int32_t op_ret, dict_t *dict,
dict_unref (rsp_dict);
rsp_dict = NULL;
- if (!npeers) {
- ret = 0;
- goto out;
- }
-
/* Sending Post Validation req to other nodes in the cluster */
gd_syncargs_init (&args, req_dict);
synctask_barrier_init((&args));
peer_cnt = 0;
- list_for_each_local_xaction_peers (peerinfo, peers) {
+
+ rcu_read_lock ();
+ cds_list_for_each_entry_rcu (peerinfo, &conf->peers, uuid_list) {
+ /* Only send requests to peers who were available before the
+ * transaction started
+ */
+ if (peerinfo->generation > txn_generation)
+ continue;
+
+ if (!peerinfo->connected)
+ continue;
+ if (op != GD_OP_SYNC_VOLUME &&
+ peerinfo->state.state != GD_FRIEND_STATE_BEFRIENDED)
+ continue;
+
gd_mgmt_v3_post_validate_req (op, op_ret, req_dict, peerinfo,
&args, MY_UUID, peer_uuid);
peer_cnt++;
}
+ rcu_read_unlock ();
+
+ if (0 == peer_cnt) {
+ ret = 0;
+ goto out;
+ }
+
gd_synctask_barrier_wait((&args), peer_cnt);
if (args.op_ret) {
@@ -1468,11 +1566,10 @@ out:
}
int
-glusterd_mgmt_v3_release_peer_locks (glusterd_op_t op,
- dict_t *dict, int32_t op_ret,
- char **op_errstr, int npeers,
+glusterd_mgmt_v3_release_peer_locks (glusterd_op_t op, dict_t *dict,
+ int32_t op_ret, char **op_errstr,
gf_boolean_t is_acquired,
- struct cds_list_head *peers)
+ uint64_t txn_generation)
{
int32_t ret = -1;
int32_t peer_cnt = 0;
@@ -1480,9 +1577,13 @@ glusterd_mgmt_v3_release_peer_locks (glusterd_op_t op,
xlator_t *this = NULL;
glusterd_peerinfo_t *peerinfo = NULL;
struct syncargs args = {0};
+ glusterd_conf_t *conf = NULL;
this = THIS;
GF_ASSERT (this);
+ conf = this->private;
+ GF_ASSERT (conf);
+
GF_ASSERT (dict);
GF_ASSERT (op_errstr);
@@ -1491,20 +1592,36 @@ glusterd_mgmt_v3_release_peer_locks (glusterd_op_t op,
if (!is_acquired)
goto out;
- if (!npeers) {
- ret = 0;
- goto out;
- }
-
/* Sending mgmt_v3 unlock req to other nodes in the cluster */
gd_syncargs_init (&args, NULL);
synctask_barrier_init((&args));
peer_cnt = 0;
- list_for_each_local_xaction_peers (peerinfo, peers) {
+
+ rcu_read_lock ();
+ cds_list_for_each_entry_rcu (peerinfo, &conf->peers, uuid_list) {
+ /* Only send requests to peers who were available before the
+ * transaction started
+ */
+ if (peerinfo->generation > txn_generation)
+ continue;
+
+ if (!peerinfo->connected)
+ continue;
+ if (op != GD_OP_SYNC_VOLUME &&
+ peerinfo->state.state != GD_FRIEND_STATE_BEFRIENDED)
+ continue;
+
gd_mgmt_v3_unlock (op, dict, peerinfo, &args,
MY_UUID, peer_uuid);
peer_cnt++;
}
+ rcu_read_unlock ();
+
+ if (0 == peer_cnt) {
+ ret = 0;
+ goto out;
+ }
+
gd_synctask_barrier_wait((&args), peer_cnt);
if (args.op_ret) {
@@ -1530,7 +1647,6 @@ glusterd_mgmt_v3_initiate_all_phases (rpcsvc_request_t *req, glusterd_op_t op,
{
int32_t ret = -1;
int32_t op_ret = -1;
- int32_t npeers = 0;
dict_t *req_dict = NULL;
dict_t *tmp_dict = NULL;
glusterd_conf_t *conf = NULL;
@@ -1538,7 +1654,7 @@ glusterd_mgmt_v3_initiate_all_phases (rpcsvc_request_t *req, glusterd_op_t op,
xlator_t *this = NULL;
gf_boolean_t is_acquired = _gf_false;
uuid_t *originator_uuid = NULL;
- struct cds_list_head xaction_peers = {0,};
+ uint64_t txn_generation = 0;
this = THIS;
GF_ASSERT (this);
@@ -1547,14 +1663,14 @@ glusterd_mgmt_v3_initiate_all_phases (rpcsvc_request_t *req, glusterd_op_t op,
conf = this->private;
GF_ASSERT (conf);
- CDS_INIT_LIST_HEAD (&xaction_peers);
- npeers = gd_build_local_xaction_peers_list (&conf->peers,
- &xaction_peers, op);
- if (npeers == -1) {
- gf_log (this->name, GF_LOG_ERROR, "building local peers list "
- "failed");
- goto rsp;
- }
+ /* Save the peer list generation */
+ txn_generation = conf->generation;
+ cmm_smp_rmb ();
+ /* This read memory barrier makes sure that this assignment happens here
+ * only and is not reordered and optimized by either the compiler or the
+ * processor.
+ */
+
/* Save the MY_UUID as the originator_uuid. This originator_uuid
* will be used by is_origin_glusterd() to determine if a node
@@ -1594,8 +1710,7 @@ glusterd_mgmt_v3_initiate_all_phases (rpcsvc_request_t *req, glusterd_op_t op,
/* LOCKDOWN PHASE - Acquire mgmt_v3 locks */
ret = glusterd_mgmt_v3_initiate_lockdown (op, dict, &op_errstr,
- npeers, &is_acquired,
- &xaction_peers);
+ &is_acquired, txn_generation);
if (ret) {
gf_log (this->name, GF_LOG_ERROR, "mgmt_v3 lockdown failed.");
goto out;
@@ -1612,17 +1727,16 @@ glusterd_mgmt_v3_initiate_all_phases (rpcsvc_request_t *req, glusterd_op_t op,
}
/* PRE-COMMIT VALIDATE PHASE */
- ret = glusterd_mgmt_v3_pre_validate (op, req_dict,
- &op_errstr, npeers,
- &xaction_peers);
+ ret = glusterd_mgmt_v3_pre_validate (op, req_dict, &op_errstr,
+ txn_generation);
if (ret) {
gf_log (this->name, GF_LOG_ERROR, "Pre Validation Failed");
goto out;
}
/* COMMIT OP PHASE */
- ret = glusterd_mgmt_v3_commit (op, dict, req_dict,
- &op_errstr, npeers, &xaction_peers);
+ ret = glusterd_mgmt_v3_commit (op, dict, req_dict, &op_errstr,
+ txn_generation);
if (ret) {
gf_log (this->name, GF_LOG_ERROR, "Commit Op Failed");
goto out;
@@ -1633,9 +1747,8 @@ glusterd_mgmt_v3_initiate_all_phases (rpcsvc_request_t *req, glusterd_op_t op,
commands other than snapshot. So as of now, I am
sending 0 (op_ret as 0).
*/
- ret = glusterd_mgmt_v3_post_validate (op, 0, dict, req_dict,
- &op_errstr, npeers,
- &xaction_peers);
+ ret = glusterd_mgmt_v3_post_validate (op, 0, dict, req_dict, &op_errstr,
+ txn_generation);
if (ret) {
gf_log (this->name, GF_LOG_ERROR, "Post Validation Failed");
goto out;
@@ -1645,10 +1758,9 @@ glusterd_mgmt_v3_initiate_all_phases (rpcsvc_request_t *req, glusterd_op_t op,
out:
op_ret = ret;
/* UNLOCK PHASE FOR PEERS*/
- (void) glusterd_mgmt_v3_release_peer_locks (op, dict,
- op_ret, &op_errstr,
- npeers, is_acquired,
- &xaction_peers);
+ (void) glusterd_mgmt_v3_release_peer_locks (op, dict, op_ret,
+ &op_errstr, is_acquired,
+ txn_generation);
/* LOCAL VOLUME(S) UNLOCK */
if (is_acquired) {
@@ -1660,12 +1772,10 @@ out:
op_ret = ret;
}
}
-rsp:
+
/* SEND CLI RESPONSE */
glusterd_op_send_cli_response (op, op_ret, 0, req, dict, op_errstr);
- gd_cleanup_local_xaction_peers_list (&xaction_peers);
-
if (req_dict)
dict_unref (req_dict);
@@ -1748,7 +1858,6 @@ glusterd_mgmt_v3_initiate_snap_phases (rpcsvc_request_t *req, glusterd_op_t op,
{
int32_t ret = -1;
int32_t op_ret = -1;
- int32_t npeers = 0;
dict_t *req_dict = NULL;
dict_t *tmp_dict = NULL;
glusterd_conf_t *conf = NULL;
@@ -1758,7 +1867,7 @@ glusterd_mgmt_v3_initiate_snap_phases (rpcsvc_request_t *req, glusterd_op_t op,
uuid_t *originator_uuid = NULL;
gf_boolean_t success = _gf_false;
char *cli_errstr = NULL;
- struct cds_list_head xaction_peers = {0,};
+ uint64_t txn_generation = 0;
this = THIS;
GF_ASSERT (this);
@@ -1767,14 +1876,13 @@ glusterd_mgmt_v3_initiate_snap_phases (rpcsvc_request_t *req, glusterd_op_t op,
conf = this->private;
GF_ASSERT (conf);
- CDS_INIT_LIST_HEAD (&xaction_peers);
- npeers = gd_build_local_xaction_peers_list (&conf->peers,
- &xaction_peers, op);
- if (npeers == -1) {
- gf_log (this->name, GF_LOG_ERROR, "building local peers list "
- "failed");
- goto rsp;
- }
+ /* Save the peer list generation */
+ txn_generation = conf->generation;
+ cmm_smp_rmb ();
+ /* This read memory barrier makes sure that this assignment happens here
+ * only and is not reordered and optimized by either the compiler or the
+ * processor.
+ */
/* Save the MY_UUID as the originator_uuid. This originator_uuid
* will be used by is_origin_glusterd() to determine if a node
@@ -1814,8 +1922,7 @@ glusterd_mgmt_v3_initiate_snap_phases (rpcsvc_request_t *req, glusterd_op_t op,
/* LOCKDOWN PHASE - Acquire mgmt_v3 locks */
ret = glusterd_mgmt_v3_initiate_lockdown (op, dict, &op_errstr,
- npeers, &is_acquired,
- &xaction_peers);
+ &is_acquired, txn_generation);
if (ret) {
gf_log (this->name, GF_LOG_ERROR, "mgmt_v3 lockdown failed.");
goto out;
@@ -1832,8 +1939,8 @@ glusterd_mgmt_v3_initiate_snap_phases (rpcsvc_request_t *req, glusterd_op_t op,
}
/* PRE-COMMIT VALIDATE PHASE */
- ret = glusterd_mgmt_v3_pre_validate (op, req_dict,
- &op_errstr, npeers, &xaction_peers);
+ ret = glusterd_mgmt_v3_pre_validate (op, req_dict, &op_errstr,
+ txn_generation);
if (ret) {
gf_log (this->name, GF_LOG_ERROR, "Pre Validation Failed");
goto out;
@@ -1857,8 +1964,8 @@ glusterd_mgmt_v3_initiate_snap_phases (rpcsvc_request_t *req, glusterd_op_t op,
goto out;
}
- ret = glusterd_mgmt_v3_brick_op (op, req_dict,
- &op_errstr, npeers, &xaction_peers);
+ ret = glusterd_mgmt_v3_brick_op (op, req_dict, &op_errstr,
+ txn_generation);
if (ret) {
gf_log (this->name, GF_LOG_ERROR, "Brick Ops Failed");
goto unbarrier;
@@ -1888,8 +1995,8 @@ glusterd_mgmt_v3_initiate_snap_phases (rpcsvc_request_t *req, glusterd_op_t op,
goto unbarrier;
}
- ret = glusterd_mgmt_v3_commit (op, dict, req_dict,
- &op_errstr, npeers, &xaction_peers);
+ ret = glusterd_mgmt_v3_commit (op, dict, req_dict, &op_errstr,
+ txn_generation);
if (ret) {
gf_log (this->name, GF_LOG_ERROR, "Commit Op Failed");
/* If the main op fails, we should save the error string.
@@ -1914,8 +2021,8 @@ unbarrier:
goto out;
}
- ret = glusterd_mgmt_v3_brick_op (op, req_dict,
- &op_errstr, npeers, &xaction_peers);
+ ret = glusterd_mgmt_v3_brick_op (op, req_dict, &op_errstr,
+ txn_generation);
if (ret) {
gf_log (this->name, GF_LOG_ERROR, "Brick Ops Failed");
@@ -1943,18 +2050,16 @@ out:
/* POST-COMMIT VALIDATE PHASE */
ret = glusterd_mgmt_v3_post_validate (op, op_ret, dict, req_dict,
- &op_errstr, npeers,
- &xaction_peers);
+ &op_errstr, txn_generation);
if (ret) {
gf_log (this->name, GF_LOG_ERROR, "Post Validation Failed");
op_ret = -1;
}
/* UNLOCK PHASE FOR PEERS*/
- (void) glusterd_mgmt_v3_release_peer_locks (op, dict,
- op_ret, &op_errstr,
- npeers, is_acquired,
- &xaction_peers);
+ (void) glusterd_mgmt_v3_release_peer_locks (op, dict, op_ret,
+ &op_errstr, is_acquired,
+ txn_generation);
/* If the commit op (snapshot taking) failed, then the error is stored
in cli_errstr and unbarrier is called. Suppose, if unbarrier also
@@ -1978,12 +2083,10 @@ out:
op_ret = ret;
}
}
-rsp:
+
/* SEND CLI RESPONSE */
glusterd_op_send_cli_response (op, op_ret, 0, req, dict, op_errstr);
- gd_cleanup_local_xaction_peers_list (&xaction_peers);
-
if (req_dict)
dict_unref (req_dict);
diff --git a/xlators/mgmt/glusterd/src/glusterd-op-sm.c b/xlators/mgmt/glusterd/src/glusterd-op-sm.c
index d7694258301..5bfdb0bb43e 100644
--- a/xlators/mgmt/glusterd/src/glusterd-op-sm.c
+++ b/xlators/mgmt/glusterd/src/glusterd-op-sm.c
@@ -119,13 +119,16 @@ glusterd_txn_opinfo_dict_fini ()
void
glusterd_txn_opinfo_init (glusterd_op_info_t *opinfo,
- glusterd_op_sm_state_info_t *state,
- glusterd_op_t *op,
- dict_t *op_ctx,
- rpcsvc_request_t *req)
+ glusterd_op_sm_state_info_t *state, glusterd_op_t *op,
+ dict_t *op_ctx, rpcsvc_request_t *req)
{
+ glusterd_conf_t *conf = NULL;
+
GF_ASSERT (opinfo);
+ conf = THIS->private;
+ GF_ASSERT (conf);
+
if (state)
opinfo->state = *state;
@@ -140,6 +143,9 @@ glusterd_txn_opinfo_init (glusterd_op_info_t *opinfo,
if (req)
opinfo->req = req;
+ opinfo->txn_generation = conf->generation;
+ cmm_smp_rmb ();
+
return;
}
@@ -314,9 +320,6 @@ glusterd_clear_txn_opinfo (uuid_t *txn_id)
dict_del(priv->glusterd_txn_opinfo, uuid_utoa (*txn_id));
- if (txn_op_info.local_xaction_peers)
- GF_FREE (txn_op_info.local_xaction_peers);
-
gf_log ("", GF_LOG_DEBUG,
"Successfully cleared opinfo for transaction ID : %s",
uuid_utoa (*txn_id));
@@ -2919,9 +2922,13 @@ glusterd_op_ac_send_lock (glusterd_op_sm_event_t *event, void *ctx)
priv = this->private;
GF_ASSERT (priv);
- list_for_each_local_xaction_peers (peerinfo,
- opinfo.local_xaction_peers) {
- GF_ASSERT (peerinfo);
+ rcu_read_lock ();
+ cds_list_for_each_entry_rcu (peerinfo, &priv->peers, uuid_list) {
+ /* Only send requests to peers who were available before the
+ * transaction started
+ */
+ if (peerinfo->generation > opinfo.txn_generation)
+ continue;
if (!peerinfo->connected || !peerinfo->mgmt)
continue;
@@ -2936,6 +2943,7 @@ glusterd_op_ac_send_lock (glusterd_op_sm_event_t *event, void *ctx)
if (proc->fn) {
ret = proc->fn (NULL, this, peerinfo);
if (ret) {
+ rcu_read_unlock ();
gf_log (this->name, GF_LOG_WARNING,
"Failed to send lock request "
"for operation 'Volume %s' to "
@@ -2958,6 +2966,7 @@ glusterd_op_ac_send_lock (glusterd_op_sm_event_t *event, void *ctx)
ret = dict_set_static_ptr (dict, "peerinfo",
peerinfo);
if (ret) {
+ rcu_read_unlock ();
gf_log (this->name, GF_LOG_ERROR,
"failed to set peerinfo");
dict_unref (dict);
@@ -2966,6 +2975,7 @@ glusterd_op_ac_send_lock (glusterd_op_sm_event_t *event, void *ctx)
ret = proc->fn (NULL, this, dict);
if (ret) {
+ rcu_read_unlock ();
gf_log (this->name, GF_LOG_WARNING,
"Failed to send mgmt_v3 lock "
"request for operation "
@@ -2981,6 +2991,7 @@ glusterd_op_ac_send_lock (glusterd_op_sm_event_t *event, void *ctx)
}
}
}
+ rcu_read_unlock ();
opinfo.pending_count = pending_count;
if (!opinfo.pending_count)
@@ -3009,9 +3020,13 @@ glusterd_op_ac_send_unlock (glusterd_op_sm_event_t *event, void *ctx)
priv = this->private;
GF_ASSERT (priv);
- list_for_each_local_xaction_peers (peerinfo,
- opinfo.local_xaction_peers) {
- GF_ASSERT (peerinfo);
+ rcu_read_lock ();
+ cds_list_for_each_entry_rcu (peerinfo, &priv->peers, uuid_list) {
+ /* Only send requests to peers who were available before the
+ * transaction started
+ */
+ if (peerinfo->generation > opinfo.txn_generation)
+ continue;
if (!peerinfo->connected || !peerinfo->mgmt ||
!peerinfo->locked)
@@ -3083,6 +3098,7 @@ glusterd_op_ac_send_unlock (glusterd_op_sm_event_t *event, void *ctx)
}
}
}
+ rcu_read_unlock ();
opinfo.pending_count = pending_count;
if (!opinfo.pending_count)
@@ -3562,9 +3578,13 @@ glusterd_op_ac_send_stage_op (glusterd_op_sm_event_t *event, void *ctx)
if (op == GD_OP_REPLACE_BRICK)
glusterd_rb_use_rsp_dict (NULL, rsp_dict);
- list_for_each_local_xaction_peers (peerinfo,
- opinfo.local_xaction_peers) {
- GF_ASSERT (peerinfo);
+ rcu_read_lock ();
+ cds_list_for_each_entry_rcu (peerinfo, &priv->peers, uuid_list) {
+ /* Only send requests to peers who were available before the
+ * transaction started
+ */
+ if (peerinfo->generation > opinfo.txn_generation)
+ continue;
if (!peerinfo->connected || !peerinfo->mgmt)
continue;
@@ -3577,6 +3597,7 @@ glusterd_op_ac_send_stage_op (glusterd_op_sm_event_t *event, void *ctx)
if (proc->fn) {
ret = dict_set_static_ptr (dict, "peerinfo", peerinfo);
if (ret) {
+ rcu_read_unlock ();
gf_log (this->name, GF_LOG_ERROR, "failed to "
"set peerinfo");
goto out;
@@ -3593,6 +3614,7 @@ glusterd_op_ac_send_stage_op (glusterd_op_sm_event_t *event, void *ctx)
pending_count++;
}
}
+ rcu_read_unlock ();
opinfo.pending_count = pending_count;
out:
@@ -4212,9 +4234,13 @@ glusterd_op_ac_send_commit_op (glusterd_op_sm_event_t *event, void *ctx)
goto out;
}
- list_for_each_local_xaction_peers (peerinfo,
- opinfo.local_xaction_peers) {
- GF_ASSERT (peerinfo);
+ rcu_read_lock ();
+ cds_list_for_each_entry_rcu (peerinfo, &priv->peers, uuid_list) {
+ /* Only send requests to peers who were available before the
+ * transaction started
+ */
+ if (peerinfo->generation > opinfo.txn_generation)
+ continue;
if (!peerinfo->connected || !peerinfo->mgmt)
continue;
@@ -4227,6 +4253,7 @@ glusterd_op_ac_send_commit_op (glusterd_op_sm_event_t *event, void *ctx)
if (proc->fn) {
ret = dict_set_static_ptr (dict, "peerinfo", peerinfo);
if (ret) {
+ rcu_read_unlock ();
gf_log (this->name, GF_LOG_ERROR,
"failed to set peerinfo");
goto out;
@@ -4242,6 +4269,7 @@ glusterd_op_ac_send_commit_op (glusterd_op_sm_event_t *event, void *ctx)
pending_count++;
}
}
+ rcu_read_unlock ();
opinfo.pending_count = pending_count;
gf_log (this->name, GF_LOG_DEBUG, "Sent commit op req for 'Volume %s' "
@@ -4528,7 +4556,6 @@ glusterd_op_txn_complete (uuid_t *txn_id)
glusterd_op_clear_op ();
glusterd_op_reset_ctx ();
glusterd_op_clear_errstr ();
- gd_cleanup_local_xaction_peers_list (opinfo.local_xaction_peers);
/* Based on the op-version, we release the cluster or mgmt_v3 lock */
if (priv->op_version < GD_OP_VERSION_3_6_0) {
diff --git a/xlators/mgmt/glusterd/src/glusterd-op-sm.h b/xlators/mgmt/glusterd/src/glusterd-op-sm.h
index 69bfd4c92a5..f6eaa372f35 100644
--- a/xlators/mgmt/glusterd/src/glusterd-op-sm.h
+++ b/xlators/mgmt/glusterd/src/glusterd-op-sm.h
@@ -103,8 +103,7 @@ struct glusterd_op_info_ {
int32_t op_errno;
char *op_errstr;
struct cds_list_head pending_bricks;
- struct cds_list_head *local_xaction_peers;
-
+ uint64_t txn_generation;
};
typedef struct glusterd_op_info_ glusterd_op_info_t;
diff --git a/xlators/mgmt/glusterd/src/glusterd-peer-utils.c b/xlators/mgmt/glusterd/src/glusterd-peer-utils.c
index 5f098839398..f3241e918f7 100644
--- a/xlators/mgmt/glusterd/src/glusterd-peer-utils.c
+++ b/xlators/mgmt/glusterd/src/glusterd-peer-utils.c
@@ -258,8 +258,15 @@ glusterd_peerinfo_t *
glusterd_peerinfo_new (glusterd_friend_sm_state_t state, uuid_t *uuid,
const char *hostname, int port)
{
- glusterd_peerinfo_t *new_peer = NULL;
- int ret = -1;
+ glusterd_peerinfo_t *new_peer = NULL;
+ int ret = -1;
+ xlator_t *this = NULL;
+ glusterd_conf_t *conf = NULL;
+
+ this = THIS;
+ GF_ASSERT (this);
+ conf = this->private;
+ GF_ASSERT (conf);
new_peer = GF_CALLOC (1, sizeof (*new_peer), gf_gld_mt_peerinfo_t);
if (!new_peer)
@@ -297,6 +304,8 @@ glusterd_peerinfo_new (glusterd_friend_sm_state_t state, uuid_t *uuid,
new_peer->port = port;
pthread_mutex_init (&new_peer->delete_lock, NULL);
+
+ new_peer->generation = uatomic_add_return (&conf->generation, 1);
out:
if (ret && new_peer) {
glusterd_peerinfo_cleanup (new_peer);
diff --git a/xlators/mgmt/glusterd/src/glusterd-sm.h b/xlators/mgmt/glusterd/src/glusterd-sm.h
index be137802a64..dceaa3e46b9 100644
--- a/xlators/mgmt/glusterd/src/glusterd-sm.h
+++ b/xlators/mgmt/glusterd/src/glusterd-sm.h
@@ -104,6 +104,7 @@ struct glusterd_peerinfo_ {
/* Members required for proper cleanup using RCU */
gd_rcu_head rcu_head;
pthread_mutex_t delete_lock;
+ uint64_t generation;
};
typedef struct glusterd_peerinfo_ glusterd_peerinfo_t;
diff --git a/xlators/mgmt/glusterd/src/glusterd-syncop.c b/xlators/mgmt/glusterd/src/glusterd-syncop.c
index a00438e6d96..c5066b015a3 100644
--- a/xlators/mgmt/glusterd/src/glusterd-syncop.c
+++ b/xlators/mgmt/glusterd/src/glusterd-syncop.c
@@ -1067,89 +1067,36 @@ out:
int
-gd_build_peers_list (struct cds_list_head *peers,
- struct cds_list_head *xact_peers, glusterd_op_t op)
+gd_lock_op_phase (glusterd_conf_t *conf, glusterd_op_t op, dict_t *op_ctx,
+ char **op_errstr, uuid_t txn_id,
+ glusterd_op_info_t *txn_opinfo)
{
- glusterd_peerinfo_t *peerinfo = NULL;
- int npeers = 0;
+ int ret = -1;
+ int peer_cnt = 0;
+ uuid_t peer_uuid = {0};
+ xlator_t *this = NULL;
+ glusterd_peerinfo_t *peerinfo = NULL;
+ struct syncargs args = {0};
- GF_ASSERT (peers);
- GF_ASSERT (xact_peers);
+ this = THIS;
+ synctask_barrier_init((&args));
+ peer_cnt = 0;
rcu_read_lock ();
- cds_list_for_each_entry_rcu (peerinfo, peers, uuid_list) {
- if (!peerinfo->connected)
+ cds_list_for_each_entry_rcu (peerinfo, &conf->peers, uuid_list) {
+ /* Only send requests to peers who were available before the
+ * transaction started
+ */
+ if (peerinfo->generation > txn_opinfo->txn_generation)
continue;
- if (op != GD_OP_SYNC_VOLUME &&
- peerinfo->state.state != GD_FRIEND_STATE_BEFRIENDED)
- continue;
-
- cds_list_add_tail (&peerinfo->op_peers_list, xact_peers);
- npeers++;
- }
- rcu_read_unlock ();
-
- return npeers;
-}
-int
-gd_build_local_xaction_peers_list (struct cds_list_head *peers,
- struct cds_list_head *xact_peers,
- glusterd_op_t op)
-{
- glusterd_peerinfo_t *peerinfo = NULL;
- glusterd_local_peers_t *local_peers = NULL;
- int npeers = 0;
-
- GF_ASSERT (peers);
- GF_ASSERT (xact_peers);
-
- rcu_read_lock ();
- cds_list_for_each_entry_rcu (peerinfo, peers, uuid_list) {
if (!peerinfo->connected)
continue;
if (op != GD_OP_SYNC_VOLUME &&
peerinfo->state.state != GD_FRIEND_STATE_BEFRIENDED)
continue;
- local_peers = GF_CALLOC (1, sizeof (*local_peers),
- gf_gld_mt_local_peers_t);
- if (!local_peers) {
- npeers = -1;
- goto unlock;
- }
- CDS_INIT_LIST_HEAD (&local_peers->op_peers_list);
- local_peers->peerinfo = peerinfo;
- cds_list_add_tail (&local_peers->op_peers_list, xact_peers);
- npeers++;
- }
-unlock:
- rcu_read_unlock ();
-
- return npeers;
-}
-int
-gd_lock_op_phase (glusterd_conf_t *conf, glusterd_op_t op, dict_t *op_ctx,
- char **op_errstr, int npeers, uuid_t txn_id,
- struct cds_list_head *peers)
-{
- int ret = -1;
- int peer_cnt = 0;
- uuid_t peer_uuid = {0};
- xlator_t *this = NULL;
- glusterd_peerinfo_t *peerinfo = NULL;
- struct syncargs args = {0};
-
- if (!npeers) {
- ret = 0;
- goto out;
- }
-
- this = THIS;
- synctask_barrier_init((&args));
- peer_cnt = 0;
- list_for_each_local_xaction_peers (peerinfo, peers) {
if (conf->op_version < GD_OP_VERSION_3_6_0) {
/* Reset lock status */
peerinfo->locked = _gf_false;
@@ -1160,6 +1107,13 @@ gd_lock_op_phase (glusterd_conf_t *conf, glusterd_op_t op, dict_t *op_ctx,
MY_UUID, peer_uuid, txn_id);
peer_cnt++;
}
+ rcu_read_unlock ();
+
+ if (0 == peer_cnt) {
+ ret = 0;
+ goto out;
+ }
+
gd_synctask_barrier_wait((&args), peer_cnt);
if (args.op_ret) {
@@ -1187,15 +1141,15 @@ out:
}
int
-gd_stage_op_phase (struct cds_list_head *peers, glusterd_op_t op,
- dict_t *op_ctx, dict_t *req_dict, char **op_errstr,
- int npeers)
+gd_stage_op_phase (glusterd_op_t op, dict_t *op_ctx, dict_t *req_dict,
+ char **op_errstr, glusterd_op_info_t *txn_opinfo)
{
int ret = -1;
int peer_cnt = 0;
dict_t *rsp_dict = NULL;
char *hostname = NULL;
xlator_t *this = NULL;
+ glusterd_conf_t *conf = NULL;
glusterd_peerinfo_t *peerinfo = NULL;
uuid_t tmp_uuid = {0};
char *errstr = NULL;
@@ -1204,6 +1158,8 @@ gd_stage_op_phase (struct cds_list_head *peers, glusterd_op_t op,
this = THIS;
GF_ASSERT (this);
+ conf = this->private;
+ GF_ASSERT (conf);
rsp_dict = dict_new ();
if (!rsp_dict)
@@ -1252,21 +1208,36 @@ stage_done:
goto out;
}
- if (!npeers) {
- ret = 0;
- goto out;
- }
-
gd_syncargs_init (&args, aggr_dict);
synctask_barrier_init((&args));
peer_cnt = 0;
- list_for_each_local_xaction_peers (peerinfo, peers) {
+ rcu_read_lock ();
+ cds_list_for_each_entry_rcu (peerinfo, &conf->peers, uuid_list) {
+ /* Only send requests to peers who were available before the
+ * transaction started
+ */
+ if (peerinfo->generation > txn_opinfo->txn_generation)
+ continue;
+
+ if (!peerinfo->connected)
+ continue;
+ if (op != GD_OP_SYNC_VOLUME &&
+ peerinfo->state.state != GD_FRIEND_STATE_BEFRIENDED)
+ continue;
+
ret = gd_syncop_mgmt_stage_op (peerinfo, &args,
MY_UUID, tmp_uuid,
op, req_dict, op_ctx);
peer_cnt++;
}
+ rcu_read_unlock ();
+
+ if (0 == peer_cnt) {
+ ret = 0;
+ goto out;
+ }
+
gf_log (this->name, GF_LOG_DEBUG, "Sent stage op req for 'Volume %s' "
"to %d peers", gd_op_list[op], peer_cnt);
@@ -1295,9 +1266,8 @@ out:
}
int
-gd_commit_op_phase (struct cds_list_head *peers, glusterd_op_t op,
- dict_t *op_ctx, dict_t *req_dict, char **op_errstr,
- int npeers)
+gd_commit_op_phase (glusterd_op_t op, dict_t *op_ctx, dict_t *req_dict,
+ char **op_errstr, glusterd_op_info_t *txn_opinfo)
{
dict_t *rsp_dict = NULL;
int peer_cnt = -1;
@@ -1305,12 +1275,17 @@ gd_commit_op_phase (struct cds_list_head *peers, glusterd_op_t op,
char *hostname = NULL;
glusterd_peerinfo_t *peerinfo = NULL;
xlator_t *this = NULL;
+ glusterd_conf_t *conf = NULL;
uuid_t tmp_uuid = {0};
char *errstr = NULL;
struct syncargs args = {0};
int type = GF_QUOTA_OPTION_TYPE_NONE;
this = THIS;
+ GF_ASSERT (this);
+ conf = this->private;
+ GF_ASSERT (conf);
+
rsp_dict = dict_new ();
if (!rsp_dict) {
ret = -1;
@@ -1359,21 +1334,36 @@ commit_done:
goto out;
}
- if (!npeers) {
- ret = 0;
- goto out;
- }
-
gd_syncargs_init (&args, op_ctx);
synctask_barrier_init((&args));
peer_cnt = 0;
- list_for_each_local_xaction_peers (peerinfo, peers) {
+ rcu_read_lock ();
+ cds_list_for_each_entry_rcu (peerinfo, &conf->peers, uuid_list) {
+ /* Only send requests to peers who were available before the
+ * transaction started
+ */
+ if (peerinfo->generation > txn_opinfo->txn_generation)
+ continue;
+
+ if (!peerinfo->connected)
+ continue;
+ if (op != GD_OP_SYNC_VOLUME &&
+ peerinfo->state.state != GD_FRIEND_STATE_BEFRIENDED)
+ continue;
+
ret = gd_syncop_mgmt_commit_op (peerinfo, &args,
MY_UUID, tmp_uuid,
op, req_dict, op_ctx);
peer_cnt++;
}
+ rcu_read_unlock ();
+
+ if (0 == peer_cnt) {
+ ret = 0;
+ goto out;
+ }
+
gd_synctask_barrier_wait((&args), peer_cnt);
ret = args.op_ret;
if (args.errstr)
@@ -1399,8 +1389,8 @@ out:
int
gd_unlock_op_phase (glusterd_conf_t *conf, glusterd_op_t op, int *op_ret,
rpcsvc_request_t *req, dict_t *op_ctx, char *op_errstr,
- int npeers, char *volname, gf_boolean_t is_acquired,
- uuid_t txn_id, struct cds_list_head *peers)
+ char *volname, gf_boolean_t is_acquired, uuid_t txn_id,
+ glusterd_op_info_t *txn_opinfo)
{
glusterd_peerinfo_t *peerinfo = NULL;
uuid_t tmp_uuid = {0};
@@ -1412,11 +1402,6 @@ gd_unlock_op_phase (glusterd_conf_t *conf, glusterd_op_t op, int *op_ret,
this = THIS;
GF_ASSERT (this);
- if (!npeers) {
- ret = 0;
- goto out;
- }
-
/* If the lock has not been held during this
* transaction, do not send unlock requests */
if (!is_acquired) {
@@ -1428,7 +1413,21 @@ gd_unlock_op_phase (glusterd_conf_t *conf, glusterd_op_t op, int *op_ret,
peer_cnt = 0;
if (conf->op_version < GD_OP_VERSION_3_6_0) {
- list_for_each_local_xaction_peers (peerinfo, peers) {
+ rcu_read_lock ();
+ cds_list_for_each_entry_rcu (peerinfo, &conf->peers,
+ uuid_list) {
+ /* Only send requests to peers who were available before
+ * the transaction started
+ */
+ if (peerinfo->generation > txn_opinfo->txn_generation)
+ continue;
+
+ if (!peerinfo->connected)
+ continue;
+ if (op != GD_OP_SYNC_VOLUME &&
+ peerinfo->state.state != GD_FRIEND_STATE_BEFRIENDED)
+ continue;
+
/* Only unlock peers that were locked */
if (peerinfo->locked) {
gd_syncop_mgmt_unlock (peerinfo, &args,
@@ -1436,16 +1435,39 @@ gd_unlock_op_phase (glusterd_conf_t *conf, glusterd_op_t op, int *op_ret,
peer_cnt++;
}
}
+ rcu_read_unlock ();
} else {
if (volname) {
- list_for_each_local_xaction_peers (peerinfo, peers) {
+ rcu_read_lock ();
+ cds_list_for_each_entry_rcu (peerinfo, &conf->peers,
+ uuid_list) {
+ /* Only send requests to peers who were
+ * available before the transaction started
+ */
+ if (peerinfo->generation >
+ txn_opinfo->txn_generation)
+ continue;
+
+ if (!peerinfo->connected)
+ continue;
+ if (op != GD_OP_SYNC_VOLUME &&
+ peerinfo->state.state != GD_FRIEND_STATE_BEFRIENDED)
+ continue;
+
gd_syncop_mgmt_v3_unlock (op_ctx, peerinfo,
&args, MY_UUID,
tmp_uuid, txn_id);
peer_cnt++;
}
+ rcu_read_unlock ();
}
}
+
+ if (0 == peer_cnt) {
+ ret = 0;
+ goto out;
+ }
+
gd_synctask_barrier_wait((&args), peer_cnt);
ret = args.op_ret;
@@ -1585,7 +1607,6 @@ gd_sync_task_begin (dict_t *op_ctx, rpcsvc_request_t * req)
{
int ret = -1;
int op_ret = -1;
- int npeers = 0;
dict_t *req_dict = NULL;
glusterd_conf_t *conf = NULL;
glusterd_op_t op = 0;
@@ -1596,7 +1617,6 @@ gd_sync_task_begin (dict_t *op_ctx, rpcsvc_request_t * req)
xlator_t *this = NULL;
gf_boolean_t is_acquired = _gf_false;
uuid_t *txn_id = NULL;
- struct cds_list_head xaction_peers = {0,};
glusterd_op_info_t txn_opinfo = {{0},};
this = THIS;
@@ -1604,8 +1624,6 @@ gd_sync_task_begin (dict_t *op_ctx, rpcsvc_request_t * req)
conf = this->private;
GF_ASSERT (conf);
- CDS_INIT_LIST_HEAD (&xaction_peers);
-
ret = dict_get_int32 (op_ctx, GD_SYNC_OPCODE_KEY, &tmp_op);
if (ret) {
gf_log (this->name, GF_LOG_ERROR, "Failed to get volume "
@@ -1686,20 +1704,11 @@ gd_sync_task_begin (dict_t *op_ctx, rpcsvc_request_t * req)
local_locking_done:
- /* Maintain xaction_peers on per transaction basis */
- npeers = gd_build_local_xaction_peers_list (&conf->peers,
- &xaction_peers, op);
- if (npeers == -1) {
- gf_log (this->name, GF_LOG_ERROR, "building local peers list "
- "failed");
- goto out;
- }
-
/* If no volname is given as a part of the command, locks will
* not be held */
if (volname || (conf->op_version < GD_OP_VERSION_3_6_0)) {
- ret = gd_lock_op_phase (conf, op, op_ctx, &op_errstr,
- npeers, *txn_id, &xaction_peers);
+ ret = gd_lock_op_phase (conf, op, op_ctx, &op_errstr, *txn_id,
+ &txn_opinfo);
if (ret) {
gf_log (this->name, GF_LOG_ERROR,
"Locking Peers Failed.");
@@ -1716,8 +1725,7 @@ local_locking_done:
goto out;
}
- ret = gd_stage_op_phase (&xaction_peers, op, op_ctx, req_dict,
- &op_errstr, npeers);
+ ret = gd_stage_op_phase (op, op_ctx, req_dict, &op_errstr, &txn_opinfo);
if (ret)
goto out;
@@ -1725,8 +1733,8 @@ local_locking_done:
if (ret)
goto out;
- ret = gd_commit_op_phase (&xaction_peers, op, op_ctx, req_dict,
- &op_errstr, npeers);
+ ret = gd_commit_op_phase (op, op_ctx, req_dict, &op_errstr,
+ &txn_opinfo);
if (ret)
goto out;
@@ -1734,11 +1742,9 @@ local_locking_done:
out:
op_ret = ret;
if (txn_id) {
- (void) gd_unlock_op_phase (conf, op, &op_ret, req,
- op_ctx, op_errstr,
- npeers, volname,
- is_acquired, *txn_id,
- &xaction_peers);
+ (void) gd_unlock_op_phase (conf, op, &op_ret, req, op_ctx,
+ op_errstr, volname, is_acquired,
+ *txn_id, &txn_opinfo);
/* Clearing the transaction opinfo */
ret = glusterd_clear_txn_opinfo (txn_id);
@@ -1751,8 +1757,6 @@ out:
glusterd_op_send_cli_response (op, op_ret, 0, req, op_ctx, op_errstr);
- gd_cleanup_local_xaction_peers_list (&xaction_peers);
-
if (volname)
GF_FREE (volname);
diff --git a/xlators/mgmt/glusterd/src/glusterd-syncop.h b/xlators/mgmt/glusterd/src/glusterd-syncop.h
index d86a5ba2131..87a3c76f9fb 100644
--- a/xlators/mgmt/glusterd/src/glusterd-syncop.h
+++ b/xlators/mgmt/glusterd/src/glusterd-syncop.h
@@ -59,15 +59,6 @@ void
gd_synctask_barrier_wait (struct syncargs *args, int count);
int
-gd_build_peers_list (struct cds_list_head *peers,
- struct cds_list_head *xact_peers, glusterd_op_t op);
-
-int
-gd_build_local_xaction_peers_list (struct cds_list_head *peers,
- struct cds_list_head *xact_peers,
- glusterd_op_t op);
-
-int
gd_brick_op_phase (glusterd_op_t op, dict_t *op_ctx, dict_t *req_dict,
char **op_errstr);
diff --git a/xlators/mgmt/glusterd/src/glusterd-utils.c b/xlators/mgmt/glusterd/src/glusterd-utils.c
index aa27ebb18d3..afc3faaefb5 100644
--- a/xlators/mgmt/glusterd/src/glusterd-utils.c
+++ b/xlators/mgmt/glusterd/src/glusterd-utils.c
@@ -9557,22 +9557,3 @@ glusterd_list_add_order (struct cds_list_head *new, struct cds_list_head *head,
cds_list_add_rcu (new, rcu_dereference (pos->prev));
}
-
-void
-gd_cleanup_local_xaction_peers_list (struct cds_list_head *xact_peers)
-{
- glusterd_local_peers_t *local_peers = NULL;
- glusterd_local_peers_t *tmp = NULL;
-
- GF_ASSERT (xact_peers);
-
- if (cds_list_empty (xact_peers))
- return;
-
- cds_list_for_each_entry_safe (local_peers, tmp, xact_peers,
- op_peers_list) {
- GF_FREE (local_peers);
- /* local_peers->peerinfo need not be freed because it does not
- * ownership of peerinfo, but merely refer it */
- }
-}
diff --git a/xlators/mgmt/glusterd/src/glusterd-utils.h b/xlators/mgmt/glusterd/src/glusterd-utils.h
index 80c7c19d508..d2dbddec3f1 100644
--- a/xlators/mgmt/glusterd/src/glusterd-utils.h
+++ b/xlators/mgmt/glusterd/src/glusterd-utils.h
@@ -37,16 +37,6 @@
volinfo->volname, brickid);\
} while (0)
-#define list_for_each_local_xaction_peers(xact_peer, xact_peers_head) \
- glusterd_local_peers_t *pos = NULL; \
- for (pos = cds_list_entry ((xact_peers_head)->next, \
- glusterd_local_peers_t, op_peers_list), \
- xact_peer = pos->peerinfo; \
- &pos->op_peers_list != (xact_peers_head); \
- pos = cds_list_entry(pos->op_peers_list.next, \
- glusterd_local_peers_t, op_peers_list), \
- xact_peer = pos->peerinfo)
-
struct glusterd_lock_ {
uuid_t owner;
time_t timestamp;
@@ -670,8 +660,4 @@ void
glusterd_list_add_order (struct cds_list_head *new, struct cds_list_head *head,
int (*compare)(struct cds_list_head *,
struct cds_list_head *));
-
-void
-gd_cleanup_local_xaction_peers_list (struct cds_list_head *peers);
-
#endif
diff --git a/xlators/mgmt/glusterd/src/glusterd.h b/xlators/mgmt/glusterd/src/glusterd.h
index 480cfc66269..ff63cce2234 100644
--- a/xlators/mgmt/glusterd/src/glusterd.h
+++ b/xlators/mgmt/glusterd/src/glusterd.h
@@ -171,7 +171,8 @@ typedef struct {
char *snap_bricks_directory;
gf_store_handle_t *missed_snaps_list_shandle;
struct cds_list_head missed_snaps_list;
- int ping_timeout;
+ int ping_timeout;
+ uint64_t generation;
} glusterd_conf_t;
generated by cgit (git 2.34.1) at 2025-11-04 13:34:32 +0000