GPF in orchagent. ACL counters related code

[pavel-shirshov]

Reading symbols from /usr/bin/orchagent...Reading symbols from /usr/lib/debug/.build-id/ef/36e12d55ea2b4540a35abab394a233817622dc.debug...bdone.
done.
t[New LWP 108]
[New LWP 97]
[New LWP 107]
[New LWP 105]
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".
Core was generated by `orchagent -m 90:b1:1c:f4:a8:51'.
Program terminated with signal SIGSEGV, Segmentation fault.
#0 0x000000000046a139 in pair (this=0x7f1b1fffee50)
at /usr/include/c++/4.9/bits/stl_pair.h:127
127 /usr/include/c++/4.9/bits/stl_pair.h: No such file or directory.
(gdb) bt
#0 0x000000000046a139 in pair (this=0x7f1b1fffee50)
at /usr/include/c++/4.9/bits/stl_pair.h:127
#1 AclOrch::collectCountersThread (pAclOrch=0x2387690) at aclorch.cpp:1315
#2 0x00007f1b25a7f970 in ?? () from /usr/lib/x86_64-linux-gnu/libstdc++.so.6
#3 0x00007f1b26825064 in start_thread (arg=0x7f1b1ffff700) at pthread_create.c:309
#4 0x00007f1b251ef62d in clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:111

[lguohan]

@qiluo-msft , can you take a look at this issue?

[pavel-shirshov]

[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".
Core was generated by `orchagent -m 90:b1:1c:f4:a8:51'.
Program terminated with signal SIGSEGV, Segmentation fault.
#0 0x000000000046a139 in pair (this=0x7f1b1fffee50) at /usr/include/c++/4.9/bits/stl_pair.h:127
127 constexpr pair(const pair&) = default;
(gdb) bt
#0 0x000000000046a139 in pair (this=0x7f1b1fffee50) at /usr/include/c++/4.9/bits/stl_pair.h:127
#1 AclOrch::collectCountersThread (pAclOrch=0x2387690) at aclorch.cpp:1315
#2 0x00007f1b25a7f970 in ?? () from /usr/lib/x86_64-linux-gnu/libstdc++.so.6
#3 0x00007f1b26825064 in start_thread (arg=0x7f1b1ffff700) at pthread_create.c:309
#4 0x00007f1b251ef62d in clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:111

[oleksandrivantsiv]

Is there any steps to reproduce? Do you observe this crash during some test?

[oleksandrivantsiv]

Can you attach or share logs with me?

[oleksandrivantsiv]

Problem is in incorrect usage of thread class in AclOrch class. Thread is initialized and started in a middle of "AclOrch" class object construction. After thread starts it reads AclOrch object members. Due to race condition in some cases thread reads AclOrch object members before they are initialized.

I was able to simulate race condition by adding additional member with long term creation between thread and other members of AclOrch class. This allowed me achieve 100% reproducibility rate. Patch that allows to reproduce the issue is the following:

diff --git a/orchagent/aclorch.cpp b/orchagent/aclorch.cpp
index 371fdde..845cc70 100644
--- a/orchagent/aclorch.cpp
+++ b/orchagent/aclorch.cpp
@@ -791,9 +791,18 @@ bool AclRange::remove()
     return true;
 }

+Test::Test(void *ptr) :
+        m_ptr(ptr)
+{
+    SWSS_LOG_ERROR("Test class before sleep");
+    sleep(1);
+    SWSS_LOG_ERROR("Test class after sleep");
+}
+
 AclOrch::AclOrch(DBConnector *db, vector<string> tableNames, PortsOrch *portOrch, MirrorOrch *mirrorOrch) :
         Orch(db, tableNames),
         thread(AclOrch::collectCountersThread, this),
+        Test(this),
         m_portOrch(portOrch),
         m_mirrorOrch(mirrorOrch)
 {
diff --git a/orchagent/aclorch.h b/orchagent/aclorch.h
index a005a7e..e646b09 100644
--- a/orchagent/aclorch.h
+++ b/orchagent/aclorch.h
@@ -1,6 +1,7 @@
 #ifndef SWSS_ACLORCH_H
 #define SWSS_ACLORCH_H

+#include <unistd.h>
 #include <iostream>
 #include <sstream>
 #include <thread>
@@ -228,7 +229,15 @@ inline void split(string str, Iterable& out, char delim = ' ')
     }
 }

-class AclOrch : public Orch, public Observer, public thread
+class Test {
+public:
+    Test(void *ptr);
+
+private:
+    void *m_ptr;
+};
+
+class AclOrch : public Orch, public Observer, public thread, public Test
 {
 public:
     AclOrch(DBConnector *db, vector<string> tableNames, PortsOrch *portOrch, MirrorOrch *mirrorOrch);

[stcheng]

#203 and #205 fix this issue