Per-inferior target_terminal state, fix PR gdb/13211, more
authorPedro Alves <palves@redhat.com>
Tue, 30 Jan 2018 14:23:51 +0000 (14:23 +0000)
committerPedro Alves <palves@redhat.com>
Tue, 30 Jan 2018 14:55:18 +0000 (14:55 +0000)
commite671cd59d74cec9f53e110ce887128d1eeadb7f2
treec2e6a7cd82327a763f8648770e9326a02fbc71ef
parent9c3a5d9319648db16b30a91253ad02d41d242cef
Per-inferior target_terminal state, fix PR gdb/13211, more

In my multi-target branch I ran into problems with GDB's terminal
handling that exist in master as well, with multi-inferior debugging.

This patch adds a testcase for said problems
(gdb.multi/multi-term-settings.exp), fixes the problems, fixes PR
gdb/13211 as well (and adds a testcase for that too,
gdb.base/interrupt-daemon.exp).

The basis of the problem I ran into is the following.  Consider a
scenario where you have:

 - inferior 1 - started with "attach", process is running on some
   other terminal.

 - inferior 2 - started with "run", process is sharing gdb's terminal.

In this scenario, when you stop/resume both inferiors, you want GDB to
save/restore the terminal settings of inferior 2, the one that is
sharing GDB's terminal.  I.e., you want inferior 2 to "own" the
terminal (in target_terminal::is_ours/target_terminal::is_inferior
sense).

Unfortunately, that's not what you get currently.  Because GDB doesn't
know whether an attached inferior is actually sharing GDB's terminal,
it tries to save/restore its settings anyway, ignoring errors.  In
this case, this is pointless, because inferior 1 is running on a
different terminal, but GDB doesn't know better.

And then, because it is only possible to have the terminal settings of
a single inferior be in effect at a time, or make one inferior/pgrp be
the terminal's foreground pgrp (aka, only one inferior can "own" the
terminal, ignoring fork children here), if GDB happens to try to
restore the terminal settings of inferior 1 first, then GDB never
restores the terminal settings of inferior 2.

This patch fixes that and a few things more along the way:

 - Moves enum target_terminal::terminal_state out of the
   target_terminal class (it's currently private) and makes it a
   scoped enum so that it can be easily used elsewhere.

 - Replaces the inflow.c:terminal_is_ours boolean with a
   target_terminal_state variable.  This allows distinguishing is_ours
   and is_ours_for_output states.  This allows finally making
   child_terminal_ours_1 do something with its "output_only"
   parameter.

 - Makes each inferior have its own copy of the
   is_ours/is_ours_for_output/is_inferior state.

 - Adds a way for GDB to tell whether the inferior is sharing GDB's
   terminal.  Works best on Linux and Solaris; the fallback works just
   as well as currently.

 - With that, we can remove the inf->attach_flag tests from
   child_terminal_inferior/child_terminal_ours.

 - Currently target_ops.to_ours is responsible for both saving the
   current inferior's terminal state, and restoring gdb's state.
   Because each inferior has its own terminal state (possibly handled
   by different targets in a multi-target world, even), we need to
   split the inferior-saving part from the gdb-restoring part.  The
   patch adds a new target_ops.to_save_inferior target method for
   that.

 - Adds a new target_terminal::save_inferior() function, so that
   sequences like:

     scoped_restore_terminal_state save_state;
     target_terminal::ours_for_output ();

   ... restore back inferiors that were
   target_terminal_state::is_inferior before back to is_inferior, and
   leaves inferiors that were is_ours alone.

 - Along the way, this adds a default implementation of
   target_pass_ctrlc to inflow.c (for inf-child.c), that handles
   passing the Ctrl-C to a process running on GDB's terminal or to
   some other process otherwise.

 - Similarly, adds a new target default implementation of
   target_interrupt, for the "interrupt" command.  The current
   implementation of this hook in inf-ptrace.c kills the whole process
   group, but that's incorrect/undesirable because we may not be
   attached to all processes in the process group.  And also, it's
   incorrect because inferior_process_group() doesn't really return
   the inferior's real process group id if the inferior is not a
   process group leader...  This is the cause of PR gdb/13211 [1],
   which this patch fixes.  While at it, that target method's "ptid"
   parameter is eliminated, because it's not really used.

 - A new test is included that exercises and fixes PR gdb/13211, and
   also fixes a GDB issue reported on stackoverflow that I ran into
   while working on this [2].  The problem is similar to PR gdb/13211,
   except that it also triggers with Ctrl-C.  When debugging a daemon
   (i.e., a process that disconnects from the controlling terminal and
   is not a process group leader, then Ctrl-C doesn't work, you just
   can't interrupt the inferior at all, resulting in a hung debug
   session.  The problem is that since the inferior is no longer
   associated with gdb's session / controlling terminal, then trying
   to put the inferior in the foreground fails.  And so Ctrl-C never
   reaches the inferior directly.  pass_signal is only used when the
   inferior is attached, but that is not the case here.  This is fixed
   by the new child_pass_ctrlc.  Without the fix, the new
   interrupt-daemon.exp testcase fails with timeout waiting for a
   SIGINT that never arrives.

[1] PR gdb/13211 - Async / Process group and interrupt not working
https://sourceware.org/bugzilla/show_bug.cgi?id=13211

[2] GDB not reacting Ctrl-C when after fork() and setsid()
https://stackoverflow.com/questions/46101292/gdb-not-reacting-ctrl-c-when-after-fork-and-setsid

Note this patch does _not_ fix:

 - PR gdb/14559 - The 'interrupt' command does not work if sigwait is in use
   https://sourceware.org/bugzilla/show_bug.cgi?id=14559

 - PR gdb/9425 - When using "sigwait" GDB doesn't trap SIGINT. Ctrl+C terminates program when should break gdb.
   https://sourceware.org/bugzilla/show_bug.cgi?id=9425

The only way to fix that that I know of (without changing the kernel)
is to make GDB put inferiors in a separate session (create a
pseudo-tty master/slave pair, make the inferior run with the slave as
its terminal, and have gdb pump output/input on the master end).

gdb/ChangeLog:
2018-01-30  Pedro Alves  <palves@redhat.com>

PR gdb/13211
* config.in, configure: Regenerate.
* configure.ac: Check for getpgid.
* go32-nat.c (go32_pass_ctrlc): New.
(go32_target): Install it.
* inf-child.c (inf_child_target): Install
child_terminal_save_inferior, child_pass_ctrlc and
child_interrupt.
* inf-ptrace.c (inf_ptrace_interrupt): Delete.
(inf_ptrace_target): No longer install it.
* infcmd.c (interrupt_target_1): Adjust.
* inferior.h (child_terminal_save_inferior, child_pass_ctrlc)
(child_interrupt): Declare.
(inferior::terminal_state): New.
* inflow.c (struct terminal_info): Update comments.
(inferior_process_group): Delete.
(terminal_is_ours): Delete.
(gdb_tty_state): New.
(child_terminal_init): Adjust.
(is_gdb_terminal, sharing_input_terminal_1)
(sharing_input_terminal): New functions.
(child_terminal_inferior): Adjust.  Use sharing_input_terminal.
Set the process's actual process group in the foreground if
possible.  Handle is_ours_for_output/is_ours distinction.  Don't
mark terminal as the inferior's if not sharing GDB's terminal.
Don't check attach_flag.
(child_terminal_ours_for_output, child_terminal_ours): Adjust to
pass down a target_terminal_state.
(child_terminal_save_inferior): New, factored out from ...
(child_terminal_ours_1): ... this.  Handle
target_terminal_state::is_ours_for_output.
(child_interrupt, child_pass_ctrlc): New.
(inflow_inferior_exit): Clear the inferior's terminal_state.
(copy_terminal_info): Copy the inferior's terminal state.
(_initialize_inflow): Remove reference to terminal_is_ours.
* inflow.h (inferior_process_group): Delete.
* nto-procfs.c (nto_handle_sigint, procfs_interrupt): Adjust.
* procfs.c (procfs_target): Don't install procfs_interrupt.
(procfs_interrupt): Delete.
* remote.c (remote_serial_quit_handler): Adjust.
(remote_interrupt): Remove ptid parameter.  Adjust.
* target-delegates.c: Regenerate.
* target.c: Include "terminal.h".
(target_terminal::terminal_state): Rename to ...
(target_terminal::m_terminal_state): ... this.
(target_terminal::init): Adjust.
(target_terminal::inferior): Adjust to per-inferior
terminal_state.
(target_terminal::restore_inferior, target_terminal_is_ours_kind): New.
(target_terminal::ours, target_terminal::ours_for_output): Use
target_terminal_is_ours_kind.
(target_interrupt): Remove ptid parameter.  Adjust.
(default_target_pass_ctrlc): Adjust.
* target.h (target_ops::to_terminal_save_inferior): New field.
(target_ops::to_interrupt): Remove ptid_t parameter.
(target_interrupt): Remove ptid_t parameter.  Update comment.
(target_pass_ctrlc): Update comment.
* target/target.h (target_terminal_state): New scoped enum,
factored out of ...
(target_terminal::terminal_state): ... here.
(target_terminal::inferior): Update comments.
(target_terminal::restore_inferior): New.
(target_terminal::is_inferior, target_terminal::is_ours)
(target_terminal::is_ours_for_output): Adjust.
(target_terminal::scoped_restore_terminal_state): Adjust to
rename, and call restore_inferior() instead of inferior().
(target_terminal::scoped_restore_terminal_state::m_state): Change
type.
(target_terminal::terminal_state): Rename to ...
(target_terminal::m_terminal_state): ... this and change type.

gdb/gdbserver/ChangeLog:
2018-01-30  Pedro Alves  <palves@redhat.com>

PR gdb/13211
* target.c (target_terminal::terminal_state): Rename to ...
(target_terminal::m_terminal_state): ... this.

gdb/testsuite/ChangeLog:
2018-01-30  Pedro Alves  <palves@redhat.com>

PR gdb/13211
* gdb.base/interrupt-daemon.c: New.
* gdb.base/interrupt-daemon.exp: New.
* gdb.multi/multi-term-settings.c: New.
* gdb.multi/multi-term-settings.exp: New.
25 files changed:
gdb/ChangeLog
gdb/config.in
gdb/configure
gdb/configure.ac
gdb/gdbserver/ChangeLog
gdb/gdbserver/target.c
gdb/go32-nat.c
gdb/inf-child.c
gdb/inf-ptrace.c
gdb/infcmd.c
gdb/inferior.h
gdb/inflow.c
gdb/inflow.h
gdb/nto-procfs.c
gdb/procfs.c
gdb/remote.c
gdb/target-delegates.c
gdb/target.c
gdb/target.h
gdb/target/target.h
gdb/testsuite/ChangeLog
gdb/testsuite/gdb.base/interrupt-daemon.c [new file with mode: 0644]
gdb/testsuite/gdb.base/interrupt-daemon.exp [new file with mode: 0644]
gdb/testsuite/gdb.multi/multi-term-settings.c [new file with mode: 0644]
gdb/testsuite/gdb.multi/multi-term-settings.exp [new file with mode: 0644]