Re: [RFC][PATCH] GDB kills itself instead of interrupting inferior

[Pedro Alves <palves at redhat dot com>]

Hi!

Very sorry for the slow response on this.

On 02/15/2013 05:34 PM, Cyril Nikolaev wrote:
> Hi! When GDB is run with IO redirected to a pipe, 'interrupt' command
> causes it to kill its own process group instead of the inferior. The
> problem manifests itself in async mode:
> 
>     $ cat | gdb <file>
>     (gdb) set target-async on
>     (gdb) run &
>     (gdb) interrupt
>     A debugging session is active.
>     Inferior 1 [process 20584] will be killed.
>     Quit anyway? (y or n) [answered Y; input not from terminal]
> 
> In this case GDB tells that its stdin isn't a tty and doesn't save
> inferior process group in `inflow.c:terminal_init_inferior_with_pgrp`
> which is wrong. And then when it receives `interrupt` command
> it `kill`'s process group 0 in `inf-ptrace.c:inf_ptrace_stop` instead
> of inferior process group.
> 
> When GDB is used from SublimeGDB (debugging plugin for Sublime Text
> editor) that means killing its own process group including Sublime
> and possibly X session. There is a corresponding issue in SublimeGDB
> bug tracker: https://github.com/quarnster/SublimeGDB/issues/29.
> 
> I suppose GDB should save inferior pgid regardless of having its
> terminal as pgid is valuable not only to reset foreground process
> group, but also to interrupt inferior.
> 
> I attach a patch that is supposed to do that. I am very new to GDB
> code. Does it look ok?

Yep, thanks!  It just needed a regression test.  I went ahead and wrote
one.  I mulled over how to spawn GDB like in your reproducer (eventually
had a wrapper shell script, and pointed the tcl GDB variable to it),
but then realized that "set interactive-mode off" is enough to trigger
badness:

 ...
 (gdb) interrupt
 (gdb) Quit
 (gdb) interrupt
 (gdb) Quit
 ...

GDB doesn't quit, but that's immaterial.

Testing with GDBserver exposed a race, where "interrupt" right after
"continue&" would occasionally do nothing, leading to spurious fails.
This is related to target remote ctrl-c issues being addressed on
another thread, where ctrl-c is swallowed in some conditions.  I've
worked around them in the test (and added a comment).

I took the liberty of adjusting/expanding your email's text, and
using it as commit log.  Below's what I checked in.

(I noticed that in the current code, the inferior's process group
is _always_ the inferior's PID, so that could be simplified.
The "This is for Lynx" comment in terminal_init_inferior is no
longer relevant, given we don't support native Lynx debugging
anymore.)

--------
GDB kills itself instead of interrupting inferior

When GDB is run with IO redirected to a pipe, the 'interrupt' command
causes it to kill its own process group instead of the inferior's.
The problem manifests itself in async mode, native debugging:

    $ cat | gdb <file>
    (gdb) set target-async on
    (gdb) run &
    (gdb) interrupt
    A debugging session is active.
    Inferior 1 [process 20584] will be killed.
    Quit anyway? (y or n) [answered Y; input not from terminal]

In this case, GDB tells that its stdin isn't a tty and doesn't save
the inferior's process group in
inflow.c:terminal_init_inferior_with_pgrp.  The 'interrupt' command
tries to 'kill' the inferior's process group in
`inf-ptrace.c:inf_ptrace_stop`, but since that wasn't saved in the
first place, GDB kills process group 0, meaning, its own process
group.

When GDB is used from a frontend, that means killing its own process
group including the frontend and possibly the X session.  This was
originally seen with SublimeGDB:
  https://github.com/quarnster/SublimeGDB/issues/29.

The patch makes GDB save the inferior pgid regardless of having a
terminal, as pgid is used not only to reset foreground process group,
but also to interrupt the inferior process.  It also adds a regression
test.  Luckily, we can emulate not having a terminal with "set
interactive-mode off", avoiding the need of special magic to spawn gdb
with a pipe.

Tested on x86_64 Fedora 17.

gdb/
2013-07-26  Cyril Nikolaev  <cyril@nichtverstehen.de>

	* inflow.c (terminal_init_inferior_with_pgrp): Save inferior
	process group regardless of having tty on stdin.

gdb/testsuite/
2013-07-26  Pedro Alves  <palves@redhat.com>

	* gdb.base/interrupt-noterm.c, gdb.base/interrupt-noterm.exp: New
	files.
---
 gdb/inflow.c                                | 17 ++++---
 gdb/testsuite/gdb.base/interrupt-noterm.c   | 25 ++++++++++
 gdb/testsuite/gdb.base/interrupt-noterm.exp | 76 +++++++++++++++++++++++++++++
 3 files changed, 111 insertions(+), 7 deletions(-)
 create mode 100644 gdb/testsuite/gdb.base/interrupt-noterm.c
 create mode 100644 gdb/testsuite/gdb.base/interrupt-noterm.exp

diff --git a/gdb/inflow.c b/gdb/inflow.c
index 79a99d1..faf4888 100644
--- a/gdb/inflow.c
+++ b/gdb/inflow.c
@@ -217,19 +217,22 @@ static void terminal_ours_1 (int);
 void
 terminal_init_inferior_with_pgrp (int pgrp)
 {
+  struct inferior *inf = current_inferior ();
+  struct terminal_info *tinfo = get_inflow_inferior_data (inf);
+
+#ifdef PROCESS_GROUP_TYPE
+  /* Store the process group even without a terminal as it is used not
+     only to reset the tty foreground process group, but also to
+     interrupt the inferior.  */
+  tinfo->process_group = pgrp;
+#endif
+
   if (gdb_has_a_terminal ())
     {
-      struct inferior *inf = current_inferior ();
-      struct terminal_info *tinfo = get_inflow_inferior_data (inf);
-
       xfree (tinfo->ttystate);
       tinfo->ttystate = serial_copy_tty_state (stdin_serial,
 					       our_terminal_info.ttystate);
 
-#ifdef PROCESS_GROUP_TYPE
-      tinfo->process_group = pgrp;
-#endif
-
       /* Make sure that next time we call terminal_inferior (which will be
          before the program runs, as it needs to be), we install the new
          process group.  */
diff --git a/gdb/testsuite/gdb.base/interrupt-noterm.c b/gdb/testsuite/gdb.base/interrupt-noterm.c
new file mode 100644
index 0000000..264b0b0
--- /dev/null
+++ b/gdb/testsuite/gdb.base/interrupt-noterm.c
@@ -0,0 +1,25 @@
+/* This testcase is part of GDB, the GNU debugger.
+
+   Copyright (C) 2013 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 ()
+{
+  sleep (3);
+  return 0;
+}
diff --git a/gdb/testsuite/gdb.base/interrupt-noterm.exp b/gdb/testsuite/gdb.base/interrupt-noterm.exp
new file mode 100644
index 0000000..42d17b1
--- /dev/null
+++ b/gdb/testsuite/gdb.base/interrupt-noterm.exp
@@ -0,0 +1,76 @@
+#   Copyright (C) 2013 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/>.
+
+standard_testfile
+
+if [prepare_for_testing "failed to prepare for testing" \
+    ${testfile} ${srcfile} {debug}] {
+    return -1
+}
+
+# Pretend there's no terminal.
+gdb_test_no_output "set interactive-mode off"
+gdb_test_no_output "set target-async on"
+
+if ![runto main] {
+    fail "Can't run to main"
+    return -1
+}
+
+# Delete breakpoints so that the next resume is a plain continue,
+# instead of a step-over-breakpoint sequence just while GDB sends the
+# interrupt request.  If that's buggy on some targets (and it was on
+# target remote for a while, where a ctrl-c at the wrong time will get
+# lost), then it should get its own specific test.  Disable
+# confirmation, avoiding complications caused by the fact that we've
+# disabled the terminal -- GDB would auto-answer "yes", confusing
+# gdb_test_multiple.
+gdb_test_no_output "set confirm off"
+gdb_test_no_output "delete"
+gdb_test_no_output "set confirm on"
+
+set async_supported -1
+set test "continue &"
+gdb_test_multiple $test $test {
+    -re "Continuing\\.\r\n$gdb_prompt $" {
+	set async_supported 1
+	pass $test
+    }
+    -re ".*Asynchronous execution not supported on this target..*" {
+	unsupported $test
+    }
+}
+if { $async_supported < 0 } {
+    return 1
+}
+
+# With native debugging, and no terminal (emulated by interactive-mode
+# off, above), GDB had a bug where "interrupt" would send SIGINT to
+# its own process group, instead of the inferior's.
+set test "interrupt"
+gdb_test_multiple $test $test {
+    -re "interrupt\r\n$gdb_prompt " {
+	pass $test
+    }
+}
+
+set test "inferior received SIGINT"
+gdb_test_multiple "" $test {
+    -re "\r\nProgram received signal SIGINT.*" {
+	# This appears after the prompt, which was already consumed
+	# above.
+	pass $test
+    }
+}
-- 
1.7.11.7