[PATCH] Fix "until LINE" in main, when "until" runs into longjmp

[Andrew Burgess]

Pedro Alves <pedro@palves.net> writes:

> With a test like this:
>
> 1       #include <dlfcn.h>
> 2       int
> 3       main ()
> 4       {
> 5          dlsym (RTLD_DEFAULT, "FOO");
> 6          return 0;
> 7       }
>
> and then "start" followed by "until 6", GDB currently incorrectly
> stops inside the runtime loader, instead of line 6.  Vis:
>
>   ...
>   Temporary breakpoint 1, main () at until.c:5
>   4       {
>   (gdb) until 6
>   0x00007ffff7f0a90d in __GI__dl_catch_exception (exception=exception@entry=0x7fffffffdb00, operate=<optimized out>, args=0x7ffff7f0a90d <__GI__dl_catch_exception+109>) at dl-error-skeleton.c:206
>   206     dl-error-skeleton.c: No such file or directory.
>   (gdb)
>
> The problem is related to longjmp handling -- dlsym internally
> longjmps on error.  The testcase can be reduced to this:
>
> 1       #include <setjmp.h>
> 2       void func () {
> 3         jmp_buf buf;
> 4         if (setjmp (buf) == 0)
> 5           longjmp (buf, 1);
> 6       }
> 7
> 8       int main () {
> 9         func ();
> 10        return 0; /* until to here */
> 11      }
>
> and then with "start" followed by "until 10", GDB currently
> incorrectly stops at line 4 (returning from setjmp), instead of line
> 10.
>
> The problem is that the BPSTAT_WHAT_CLEAR_LONGJMP_RESUME code in
> infrun.c fails to find the initiating frame, and so infrun thinks that
> the longjmp jumped somewhere outer to "until"'s originating frame.
>
> Here:
>
>     case BPSTAT_WHAT_CLEAR_LONGJMP_RESUME:
>       {
> 	struct frame_info *init_frame;
>
> 	/* There are several cases to consider.
>
> 	   1. The initiating frame no longer exists.  In this case we
> 	   must stop, because the exception or longjmp has gone too
> 	   far.
>
>         ...
>
> 	init_frame = frame_find_by_id (ecs->event_thread->initiating_frame);
>
> 	if (init_frame)   // this is NULL!
> 	  {
> 	     ...
> 	  }
>
> 	/* For Cases 1 and 2, remove the step-resume breakpoint, if it
> 	   exists.  */
> 	delete_step_resume_breakpoint (ecs->event_thread);
>
> 	end_stepping_range (ecs);   // case 1., so we stop.
>       }
>
> The initiating frame is set by until_break_command ->
> set_longjmp_breakpoint.  The initiating frame is supposed to be the
> frame that is selected when the command was issued, but
> until_break_command instead passes the frame id of the _caller_ frame
> by mistake.  When the "until LINE" command is issued from main, the
> caller frame is the caller of main.  When later infrun tries to find
> that frame by id, it fails to find it, because frame_find_by_id
> doesn't unwind past main.
>
> The bug is that we passed the caller frame's id to
> set_longjmp_breakpoint.  We should have passed the selected frame's id
> instead.

LGTM.

Thanks,
Andrew

>
> Change-Id: Iaae1af7cdddf296b7c5af82c3b5b7d9b66755b1c
> ---
>  gdb/breakpoint.c                              |  2 +-
>  .../gdb.base/longjmp-until-in-main.c          | 34 ++++++++++++++
>  .../gdb.base/longjmp-until-in-main.exp        | 44 +++++++++++++++++++
>  3 files changed, 79 insertions(+), 1 deletion(-)
>  create mode 100644 gdb/testsuite/gdb.base/longjmp-until-in-main.c
>  create mode 100644 gdb/testsuite/gdb.base/longjmp-until-in-main.exp
>
> diff --git a/gdb/breakpoint.c b/gdb/breakpoint.c
> index a3be12557f6..74f53368464 100644
> --- a/gdb/breakpoint.c
> +++ b/gdb/breakpoint.c
> @@ -10489,7 +10489,7 @@ until_break_command (const char *arg, int from_tty, int anywhere)
>  				    caller_frame_id, bp_until);
>        breakpoints.emplace_back (std::move (caller_breakpoint));
>  
> -      set_longjmp_breakpoint (tp, caller_frame_id);
> +      set_longjmp_breakpoint (tp, stack_frame_id);
>        lj_deleter.emplace (thread);
>      }
>  
> diff --git a/gdb/testsuite/gdb.base/longjmp-until-in-main.c b/gdb/testsuite/gdb.base/longjmp-until-in-main.c
> new file mode 100644
> index 00000000000..3b9ef945a34
> --- /dev/null
> +++ b/gdb/testsuite/gdb.base/longjmp-until-in-main.c
> @@ -0,0 +1,34 @@
> +/* This testcase is part of GDB, the GNU debugger.
> +
> +   Copyright 2022 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 <setjmp.h>
> +
> +void
> +func ()
> +{
> +  jmp_buf buf;
> +
> +  if (setjmp (buf) == 0)
> +    longjmp (buf, 1);
> +}
> +
> +int
> +main ()
> +{
> +  func ();
> +  return 0; /* until to here */
> +}
> diff --git a/gdb/testsuite/gdb.base/longjmp-until-in-main.exp b/gdb/testsuite/gdb.base/longjmp-until-in-main.exp
> new file mode 100644
> index 00000000000..33a907fb83e
> --- /dev/null
> +++ b/gdb/testsuite/gdb.base/longjmp-until-in-main.exp
> @@ -0,0 +1,44 @@
> +# Copyright 2008-2022 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 "until LINE", started in the "main()" frame, where the until
> +# command runs into a longjmp that lands in a frame that is inner than
> +# main.  GDB internally intercepts the longjmp, sets a breakpoint at
> +# the jump destination, and once there, decides whether to stop or
> +# ignore the breakpoint hit depending on whether the initiating frame
> +# is present on the frame chain.  GDB used to have a bug where it
> +# recorded the frame of the caller of main instead of the frame of
> +# main as the initiating frame, and then later on when deciding
> +# whether the longjmp landed somewhere inner than main, since
> +# unwinding normally stops at main, GDB would fail to find the
> +# initiating frame.
> +
> +standard_testfile
> +
> +if {[prepare_for_testing "failed to prepare" ${testfile} ${srcfile}]} {
> +    return
> +}
> +
> +if {![runto_main]} {
> +    return
> +}
> +
> +delete_breakpoints
> +
> +set until_to_line [gdb_get_line_number "until to here"]
> +
> +gdb_test "until $until_to_line" \
> +    " until to here .*" \
> +    "until \$line, in main"
>
> base-commit: dd4c046506cd4da46b439a2b4f8b6d933ecbb961
> -- 
> 2.36.0