Hi Vu,
Don't you need to protect the list in ~LogClient()? And in
NotifyClientAboutLostInvocations(), does it need to protect before
'read' in the 'for' loop? Otherwise it's ack from me.
Thanks
Minh
On 6/1/20 2:15 pm, Vu Minh Nguyen wrote:
log agent did not protect the resource `unacked_invocations_ list` from
accessing by multiple threads, so caused segmentation fault.
This patch introduces a mutex in order to synchronize the access to that
common resource.
---
src/log/agent/lga_client.h | 12 +++++++++++-
1 file changed, 11 insertions(+), 1 deletion(-)
diff --git a/src/log/agent/lga_client.h b/src/log/agent/lga_client.h
index f5fa6faa4..c999d148e 100644
--- a/src/log/agent/lga_client.h
+++ b/src/log/agent/lga_client.h
@@ -174,11 +174,15 @@ class LogClient {
// get acknowledgement from it.
void KeepTrack(SaInvocationT inv, uint32_t ack_flags) {
if (ack_flags != SA_LOG_RECORD_WRITE_ACK) return;
+ base::Lock scope_lock{mutex_unacked_list_};
unacked_invocations_.push_back(inv);
}
// Got an acknowledgment, so remove from the track list.
- void RemoveTrack(SaInvocationT inv) { unacked_invocations_.remove(inv); }
+ void RemoveTrack(SaInvocationT inv) {
+ base::Lock scope_lock{mutex_unacked_list_};
+ unacked_invocations_.remove(inv);
+ }
void NotifyClientAboutLostInvocations() {
for (const auto& i : unacked_invocations_) {
@@ -196,6 +200,8 @@ class LogClient {
SendMsgToMbx(msg, MDS_SEND_PRIORITY_HIGH);
}
+
+ base::Lock scope_lock{mutex_unacked_list_};
unacked_invocations_.clear();
}
@@ -290,6 +296,10 @@ class LogClient {
// If cluster goes to headless, log agent will inform to log client with
// SA_AIS_ERR_TRY_AGAIN code for these invocations.
std::list<SaInvocationT> unacked_invocations_{};
+
+ // To protect the `unacked_invocations_` list.
+ base::Mutex mutex_unacked_list_{};
+
// LOG handle (derived from hdl-mngr)
SaLogHandleT handle_;
_______________________________________________
Opensaf-devel mailing list
Opensaf-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/opensaf-devel
