[binutils-gdb] gdb/Windows: Fix detach while running

[binutils-gdb] gdb/Windows: Fix detach while running

[Pedro Alves]

https://sourceware.org/git/gitweb.cgi?p=binutils-gdb.git;h=56f4dea207d690fff70a3ccb6fe10ebe1e4697c0

commit 56f4dea207d690fff70a3ccb6fe10ebe1e4697c0
Author: Pedro Alves <pedro@palves.net>
Date:   Wed Apr 17 18:57:01 2024 +0100

    gdb/Windows: Fix detach while running
    
    While testing a WIP Cygwin GDB that supports non-stop, I noticed that
    gdb.threads/attach-non-stop.exp exposes that this:
    
     (gdb) attach PID&
     ...
     (gdb) detach
    
    ... hangs.
    
    And it turns out that it hangs in all-stop as well.  This commits
    fixes that.
    
    After "attach &", the target is set running, we've called
    ContinueDebugEvent and the process_thread thread is waiting for
    WaitForDebugEvent events.  It is the equivalent of "attach; c&".
    
    In windows_nat_target::detach, the first thing we do is
    unconditionally call windows_continue (for ContinueDebugEvent), which
    blocks in do_synchronously, until the process_thread sees an event out
    of WaitForDebugEvent.  Unless the inferior happens to run into a
    breakpoint, etc., then this hangs indefinitely.
    
    If we've already called ContinueDebugEvent earlier, then we shouldn't
    be calling it again in ::detach.
    
    Still in windows_nat_target::detach, we have an interesting issue that
    ends up being the bulk of the patch -- only the process_thread thread
    can call DebugActiveProcessStop, but if it is blocked in
    WaitForDebugEvent, we need to somehow force it to break out of it.
    The only way to do that, is to force the inferior to do something that
    causes WaitForDebugEvent to return some event.
    
    This patch uses CreateRemoteThread to do it, which results in
    WaitForDebugEvent reporting CREATE_THREAD_DEBUG_EVENT.  We then
    terminate the injected thread before it has a chance to run any
    userspace code.
    
    Note that Win32 functions like DebugBreakProcess and
    GenerateConsoleCtrlEvent would also inject a new thread in the
    inferior.  I first used DebugBreakProcess, but that is actually more
    complicated to use, because we'd have to be sure to consume the
    breakpoint event before detaching, otherwise the inferior would likely
    die due a breakpoint exception being raised with no debugger around to
    intercept it.
    
    See the new break_out_process_thread method.
    
    So the fix has two parts:
    
     - Keep track of whether we've called ContinueDebugEvent and the
       process_thread thread is waiting for events, or whether
       WaitForDebugEvent already returned an event.
    
     - In windows_nat_target::detach, if the process_thread thread is
       waiting for events, unblock out of its WaitForDebugEvent, before
       proceeding with the actual detach.
    
    New test included.  Passes cleanly on GNU/Linux native and gdbserver,
    and also passes cleanly on Cygwin and MinGW, with the fix.  Before the
    fix, it would hang and fail with a timeout.
    
    Tested-By: Hannes Domani <ssbssa@yahoo.de>
    Reviewed-By: Tom Tromey <tom@tromey.com>
    Change-Id: Ifb91c58c08af1a9bcbafecedc93dfce001040905

Diff:
---
 gdb/testsuite/gdb.base/detach-while-running.c   |  27 +++++
 gdb/testsuite/gdb.base/detach-while-running.exp |  95 +++++++++++++++
 gdb/windows-nat.c                               | 153 ++++++++++++++++++++++--
 3 files changed, 267 insertions(+), 8 deletions(-)

diff --git a/gdb/testsuite/gdb.base/detach-while-running.c b/gdb/testsuite/gdb.base/detach-while-running.c
new file mode 100644
index 00000000000..9c038f4af18
--- /dev/null
+++ b/gdb/testsuite/gdb.base/detach-while-running.c
@@ -0,0 +1,27 @@
+/* This testcase is part of GDB, the GNU debugger.
+
+   Copyright 2024 Free Software Foundation, Inc.
+
+   This program is free software; you can redistribute it and/or modify
+   it under the terms of the GNU General Public License as published by
+   the Free Software Foundation; either version 3 of the License, or
+   (at your option) any later version.
+
+   This program is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+   GNU General Public License for more details.
+
+   You should have received a copy of the GNU General Public License
+   along with this program.  If not, see <http://www.gnu.org/licenses/>.  */
+
+#include <unistd.h>
+
+int
+main (int argc, char **argv)
+{
+  for (int i = 0; i < 30; i++)
+    sleep (1);
+
+  return 0;
+}
diff --git a/gdb/testsuite/gdb.base/detach-while-running.exp b/gdb/testsuite/gdb.base/detach-while-running.exp
new file mode 100644
index 00000000000..4ce6c7045ea
--- /dev/null
+++ b/gdb/testsuite/gdb.base/detach-while-running.exp
@@ -0,0 +1,95 @@
+# Copyright 2024 Free Software Foundation, Inc.
+
+# This program is free software; you can redistribute it and/or modify
+# it under the terms of the GNU General Public License as published by
+# the Free Software Foundation; either version 3 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with this program.  If not, see <http://www.gnu.org/licenses/>.
+
+# Test detaching while the inferior is running.  Basically:
+#
+#  (gdb) attach PID
+#  (gdb) c&
+#  (gdb) detach
+
+require can_spawn_for_attach
+
+standard_testfile
+
+if {[build_executable "failed to prepare" $testfile $srcfile {debug}] == -1} {
+    return
+}
+
+# The test proper.  See description above.
+
+proc test {} {
+    global binfile gdb_prompt
+
+    # This test requires executing commands while the target is
+    # running, which, when testing with the remote target, requires
+    # non-stop remote protocol.  Until that variant of the RSP is the
+    # default, force target non-stop mode on.
+    set is_remote \
+	[expr {[target_info exists gdb_protocol] \
+		   && ([target_info gdb_protocol] == "remote" \
+			   || [target_info gdb_protocol] == "extended-remote")}]
+
+    save_vars { ::GDBFLAGS } {
+	if {$is_remote} {
+	    append ::GDBFLAGS " -ex \"maint set target-non-stop on\""
+	}
+	clean_restart ${binfile}
+    }
+
+    set test_spawn_id [spawn_wait_for_attach $binfile]
+    set testpid [spawn_id_get_pid $test_spawn_id]
+
+    set any "\[^\r\n\]*"
+
+    # Iterate more than once so that we test re-attaching after
+    # detaching, in case GDB incorrectly detaches and the process
+    # crashes after the detach.
+    set n_iters 2
+    for {set iter 1} {$iter <= $n_iters} {incr iter} {
+	with_test_prefix "iter=$iter" {
+	    set attached 0
+
+	    gdb_test_multiple "attach $testpid" "attach" {
+		-re "Attaching to program:${any}process $testpid\r\n.*$gdb_prompt " {
+		    pass $gdb_test_name
+		    set attached 1
+		}
+	    }
+
+	    if {!$attached} {
+		break
+	    }
+
+	    gdb_test_multiple "continue &" "" {
+		-re "Continuing\.\r\n$::gdb_prompt " {
+		    pass $gdb_test_name
+		}
+	    }
+
+	    gdb_test "detach" "Detaching from.*"
+
+	    # Sleep a bit before reattaching to let the detached
+	    # process crash and exit if e.g., GDB managed to leave
+	    # breakpoint traps behind.
+	    if {$iter != $n_iters} {
+		sleep 1
+	    }
+	}
+    }
+
+    kill_wait_spawned_process $test_spawn_id
+}
+
+test
diff --git a/gdb/windows-nat.c b/gdb/windows-nat.c
index 6fc41231023..a53b6a6e053 100644
--- a/gdb/windows-nat.c
+++ b/gdb/windows-nat.c
@@ -356,6 +356,13 @@ private:
      needed.  */
   void wait_for_debug_event_main_thread (DEBUG_EVENT *event);
 
+  /* Force the process_thread thread to return from WaitForDebugEvent.
+     PROCESS_ALIVE is set to false if the inferior process exits while
+     we're trying to break out the process_thread thread.  This can
+     happen because this is called while all threads are running free,
+     while we're trying to detach.  */
+  void break_out_process_thread (bool &process_alive);
+
   /* Queue used to send requests to process_thread.  This is
      implicitly locked.  */
   std::queue<gdb::function_view<bool ()>> m_queue;
@@ -378,6 +385,12 @@ private:
 
   /* True if currently in async mode.  */
   bool m_is_async = false;
+
+  /* True if we last called ContinueDebugEvent and the process_thread
+     thread is now waiting for events.  False if WaitForDebugEvent
+     already returned an event, and we need to ContinueDebugEvent
+     again to restart the inferior.  */
+  bool m_continued = false;
 };
 
 static void
@@ -497,6 +510,8 @@ windows_nat_target::wait_for_debug_event_main_thread (DEBUG_EVENT *event)
 	wait_for_debug_event (event, INFINITE);
       return false;
     });
+
+  m_continued = false;
 }
 
 /* See nat/windows-nat.h.  */
@@ -1351,6 +1366,8 @@ windows_nat_target::windows_continue (DWORD continue_status, int id,
 				" - ContinueDebugEvent failed"),
 			      *err);
 
+  m_continued = !last_call;
+
   return TRUE;
 }
 
@@ -2071,20 +2088,140 @@ windows_nat_target::attach (const char *args, int from_tty)
   target_terminal::ours ();
 }
 
+void
+windows_nat_target::break_out_process_thread (bool &process_alive)
+{
+  /* This is called when the process_thread thread is blocked in
+     WaitForDebugEvent (unless it already returned some event we
+     haven't consumed yet), and we need to unblock it so that we can
+     have it call DebugActiveProcessStop.
+
+     To make WaitForDebugEvent return, we need to force some event in
+     the inferior.  Any method that lets us do that (without
+     disturbing the other threads), injects a new thread in the
+     inferior.
+
+     We don't use DebugBreakProcess for this, because that injects a
+     thread that ends up executing a breakpoint instruction.  We can't
+     let the injected thread hit that breakpoint _after_ we've
+     detached.  Consuming events until we see a breakpoint trap isn't
+     100% reliable, because we can't distinguish it from some other
+     thread itself deciding to call int3 while we're detaching, unless
+     we temporarily suspend all threads.  It's just a lot of
+     complication, and there's an easier way.
+
+     Important observation: the thread creation event for the newly
+     injected thread is sufficient to unblock WaitForDebugEvent.
+
+     Instead of DebugBreakProcess, we can instead use
+     CreateRemoteThread to control the code that the injected thread
+     runs ourselves.  We could consider pointing the injected thread
+     at some side-effect-free Win32 function as entry point.  However,
+     finding the address of such a function requires having at least
+     minimal symbols loaded for ntdll.dll.  Having a way that avoids
+     that is better, so that detach always works correctly even when
+     we don't have any symbols loaded.
+
+     So what we do is inject a thread that doesn't actually run ANY
+     userspace code, because we force-terminate it as soon as we see
+     its corresponding thread creation event.  CreateRemoteThread
+     gives us the new thread's ID, which we can match with the thread
+     associated with the CREATE_THREAD_DEBUG_EVENT event.  */
+
+  DWORD injected_thread_id = 0;
+  HANDLE injected_thread_handle
+    = CreateRemoteThread (windows_process.handle, NULL,
+			  0, (LPTHREAD_START_ROUTINE) 0,
+			  NULL, 0, &injected_thread_id);
+
+  if (injected_thread_handle == NULL)
+    {
+      DWORD err = GetLastError ();
+
+      DEBUG_EVENTS ("CreateRemoteThread failed with %u", err);
+
+      if (err == ERROR_ACCESS_DENIED)
+	{
+	  /* Creating the remote thread fails with ERROR_ACCESS_DENIED
+	     if the process exited before we had a chance to inject
+	     the thread.  Continue with the loop below and consume the
+	     process exit event anyhow, so that our caller can always
+	     call windows_continue.  */
+	}
+      else
+	throw_winerror_with_name (_("Can't detach from running process.  "
+				    "Interrupt it first."),
+				  err);
+    }
+
+  process_alive = true;
+
+  /* At this point, the user has declared that they want to detach, so
+     any event that happens from this point on should be forwarded to
+     the inferior.  */
+
+  for (;;)
+    {
+      DEBUG_EVENT current_event;
+      wait_for_debug_event_main_thread (&current_event);
+
+      if (current_event.dwDebugEventCode == EXIT_PROCESS_DEBUG_EVENT)
+	{
+	  DEBUG_EVENTS ("got EXIT_PROCESS_DEBUG_EVENT");
+	  process_alive = false;
+	  break;
+	}
+
+      if (current_event.dwDebugEventCode == CREATE_THREAD_DEBUG_EVENT
+	  && current_event.dwThreadId == injected_thread_id)
+	{
+	  DEBUG_EVENTS ("got CREATE_THREAD_DEBUG_EVENT for injected thread");
+
+	  /* Terminate the injected thread, so it doesn't run any code
+	     at all.  All we wanted was some event, and
+	     CREATE_THREAD_DEBUG_EVENT is sufficient.  */
+	  CHECK (TerminateThread (injected_thread_handle, 0));
+	  break;
+	}
+
+      DEBUG_EVENTS ("got unrelated event, code %u",
+		    current_event.dwDebugEventCode);
+      windows_continue (DBG_CONTINUE, -1, 0);
+    }
+
+  if (injected_thread_handle != NULL)
+    CHECK (CloseHandle (injected_thread_handle));
+}
+
 void
 windows_nat_target::detach (inferior *inf, int from_tty)
 {
+  /* If we see the process exit while unblocking the process_thread
+     helper thread, then we should skip the actual
+     DebugActiveProcessStop call.  But don't report an error.  Just
+     pretend the process exited shortly after the detach.  */
+  bool process_alive = true;
+
+  /* The process_thread helper thread will be blocked in
+     WaitForDebugEvent waiting for events if we've resumed the target
+     before we get here, e.g., with "attach&" or "c&".  We need to
+     unblock it so that we can have it call DebugActiveProcessStop
+     below, in the do_synchronously block.  */
+  if (m_continued)
+    break_out_process_thread (process_alive);
+
   windows_continue (DBG_CONTINUE, -1, 0, true);
 
   std::optional<unsigned> err;
-  do_synchronously ([&] ()
-    {
-      if (!DebugActiveProcessStop (windows_process.current_event.dwProcessId))
-	err = (unsigned) GetLastError ();
-      else
-	DebugSetProcessKillOnExit (FALSE);
-      return false;
-    });
+  if (process_alive)
+    do_synchronously ([&] ()
+      {
+	if (!DebugActiveProcessStop (windows_process.current_event.dwProcessId))
+	  err = (unsigned) GetLastError ();
+	else
+	  DebugSetProcessKillOnExit (FALSE);
+	return false;
+      });
 
   if (err.has_value ())
     {