[Lldb-commits] [PATCH] Fix TestProcesslaunch regression caused by D7372

[Pavel Labath]

Hi ovyalov,

After closing all the leaked file descriptors to the inferior tty, the following problem occured:
- when stdin, stdout and stderr are redirected, there are no slave descriptors open (which is good)
- lldb has a reader thread, which attempts to read from the master end of the tty
- this thread receives an EOF
- in response, it closes it's master end
- as this is the last open file descriptor for the master end, this deletes the tty and sends
  SIGHUP to the inferior (this is bad)

I fix this problem by making sure the master end remains open for the duration of the inferior
process by storing a copy of the file descriptor in ProcessMonitor. I create a copy to avoid
ownership issues with the reading thread.

http://reviews.llvm.org/D7440

Files:
  source/Plugins/Process/Linux/ProcessMonitor.cpp
  source/Plugins/Process/Linux/ProcessMonitor.h
  source/Plugins/Process/POSIX/ProcessPOSIX.cpp

Index: source/Plugins/Process/Linux/ProcessMonitor.cpp
===================================================================
--- source/Plugins/Process/Linux/ProcessMonitor.cpp
+++ source/Plugins/Process/Linux/ProcessMonitor.cpp
@@ -2347,11 +2347,10 @@
     StopOpThread();
     sem_destroy(&m_operation_pending);
     sem_destroy(&m_operation_done);
-
-    // Note: ProcessPOSIX passes the m_terminal_fd file descriptor to
-    // Process::SetSTDIOFileDescriptor, which in turn transfers ownership of
-    // the descriptor to a ConnectionFileDescriptor object.  Consequently
-    // even though still has the file descriptor, we shouldn't close it here.
+    if (m_terminal_fd >= 0) {
+        close(m_terminal_fd);
+        m_terminal_fd = -1;
+    }
 }
 
 void
Index: source/Plugins/Process/Linux/ProcessMonitor.h
===================================================================
--- source/Plugins/Process/Linux/ProcessMonitor.h
+++ source/Plugins/Process/Linux/ProcessMonitor.h
@@ -84,7 +84,8 @@
     /// Reads from this file descriptor yield both the standard output and
     /// standard error of this debugee.  Even if stderr and stdout were
     /// redirected on launch it may still happen that data is available on this
-    /// descriptor (if the inferior process opens /dev/tty, for example).
+    /// descriptor (if the inferior process opens /dev/tty, for example). This descriptor is
+    /// closed after a call to StopMonitor().
     ///
     /// If this monitor was attached to an existing process this method returns
     /// -1.
Index: source/Plugins/Process/POSIX/ProcessPOSIX.cpp
===================================================================
--- source/Plugins/Process/POSIX/ProcessPOSIX.cpp
+++ source/Plugins/Process/POSIX/ProcessPOSIX.cpp
@@ -252,7 +252,16 @@
     if (!error.Success())
         return error;
 
-    SetSTDIOFileDescriptor(m_monitor->GetTerminalFD());
+    int terminal = m_monitor->GetTerminalFD();
+    if (terminal >= 0) {
+        // The reader thread will close the file descriptor when done, so we pass it a copy.
+        int stdio = fcntl(terminal, F_DUPFD_CLOEXEC, 0);
+        if (stdio == -1) {
+            error.SetErrorToErrno();
+            return error;
+        }
+        SetSTDIOFileDescriptor(stdio);
+    }
 
     SetID(m_monitor->GetPID());
     return error;

EMAIL PREFERENCES
  http://reviews.llvm.org/settings/panel/emailpreferences/
-------------- next part --------------
A non-text attachment was scrubbed...
Name: D7440.19409.patch
Type: text/x-patch
Size: 2333 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/lldb-commits/attachments/20150205/c1ec7e52/attachment.bin>