diff options
author | Pedro Alves <palves@redhat.com> | 2013-07-26 11:15:45 +0000 |
---|---|---|
committer | Pedro Alves <palves@redhat.com> | 2013-07-26 11:15:45 +0000 |
commit | 6f64ef53c0a77b34717741d4a5453760ff2a2cab (patch) | |
tree | eeca176fc65659893945ef5ea5bfac2f6d5cb248 /gdb/inflow.c | |
parent | cc3d0a6f351b2005ed19bf7f33db8f6b2f43863b (diff) | |
download | gdb-6f64ef53c0a77b34717741d4a5453760ff2a2cab.zip gdb-6f64ef53c0a77b34717741d4a5453760ff2a2cab.tar.gz gdb-6f64ef53c0a77b34717741d4a5453760ff2a2cab.tar.bz2 |
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.
Diffstat (limited to 'gdb/inflow.c')
-rw-r--r-- | gdb/inflow.c | 17 |
1 files changed, 10 insertions, 7 deletions
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. */ |