Re: [PATCH 1/2] Debugging without a binary (regression)

[Pedro Alves <palves@redhat.com>]

On 02/24/2016 02:26 PM, Luis Machado wrote:
> When we attempt to debug a process using GDBserver in standard remote mode
> without a symbol file on GDB's end, we may run into an issue where GDB cuts
> the connection attempt short due to an error. The error is caused by not
> being able to open a symbol file, like so:
> 
> --
> 
> (gdb) set sysroot
> (gdb) tar rem :2345
> Remote debugging using :2345
> /proc/23769/exe: Permission denied.
> (gdb) i r
> The program has no registers now.
> (gdb)
> 
> It should've been like this:
> 
> (gdb) set sysroot
> (gdb) tar rem :2345
> Remote debugging using :2345
> 0xf7ddb2d0 in ?? ()
> (gdb) i r
> eax            0x0  0
> ecx            0x0  0
> edx            0x0  0
> ebx            0x0  0
> esp            0xffffdfa0 0xffffdfa0
> ebp            0x0  0x0
> esi            0x0  0
> edi            0x0  0
> eip            0xf7ddb2d0 0xf7ddb2d0
> eflags         0x200  [ IF ]
> cs             0x33 51
> ss             0x2b 43
> ds             0x0  0
> es             0x0  0
> fs             0x0  0
> gs             0x0  0
> (gdb)
> 
> This is caused by a couple of function calls within exec_file_locate_attach
> that can potentially throw errors.
> 
> The following patch guards both exec_file_attach and symbol_file_add_main to
> prevent the errors from disrupting the connection process.
> 
> There was also a case where native GDB tripped on this problem, but it was
> mostly fixed by bf74e428bca61022bd5cdf6bf28789a184748b4d.
> 
> Regression-tested on x86-64/Ubuntu.
> 
> gdb/ChangeLog:
> 
> 2016-02-24  Luis Machado  <lgustavo@codesourcery.com>
> 
> 	* exec.c (exec_file_locate_attach): Guard a couple functions
> 	that can throw errors.
> ---
>  gdb/exec.c | 31 +++++++++++++++++++++++++++++--
>  1 file changed, 29 insertions(+), 2 deletions(-)
> 
> diff --git a/gdb/exec.c b/gdb/exec.c
> index 90811c0..3f77b3d 100644
> --- a/gdb/exec.c
> +++ b/gdb/exec.c
> @@ -176,8 +176,35 @@ exec_file_locate_attach (int pid, int from_tty)
>  
>    old_chain = make_cleanup (xfree, full_exec_path);
>  
> -  exec_file_attach (full_exec_path, from_tty);
> -  symbol_file_add_main (full_exec_path, from_tty);
> +  /* exec_file_attach and symbol_file_add_main may throw an error if the file
> +     cannot be opened either locally or remotely.  This happens when, for
> +     example, GDB connects to a GDBserver that is running on a different
> +     filesystem and the sysroot is set to non-target-based (no "target:").

The second sentence no longer makes sense after Gary's change.

I think it should read:

  This happens for example, when the file is first found in the local
  sysroot (above), and then disappears (a TOCTOU race), or when it doesn't
  exist in the target filesystem, or when the file does exist, but
  is not readable.

> +
> +     Then GDB will neither load the binary from the target nor be able to
> +     load a binary from the local filesystem (it may not exist in the local
> +     filesystem in the same path as in the remote filesystem).

No longer makes sense either.

> +
> +     Even without a binary, the remote-based debugging session should
> +     continue normally instead of ending abruptly.  Hence we catch thrown
> +     errors/exceptions in the following code.  */
> +  TRY
> +    {
> +      exec_file_attach (full_exec_path, from_tty);
> +    }
> +  CATCH (err, RETURN_MASK_ALL)

This should be RETURN_MASK_ERROR instead.  Silently swallowing ctrl-c is
rarely the right thing to do, if ever.  A ctrl-c during the
initial remote connection is supposed to bring down the connection
immediately:

~~~
  /* Start the remote connection.  If error() or QUIT, discard this
     target (we'd otherwise be in an inconsistent state) and then
     propogate the error on up the exception chain.  This ensures that
     the caller doesn't stumble along blindly assuming that the
     function succeeded.
~~~

static void
remote_start_remote (int from_tty, struct target_ops *target, int extended_p)
{
  struct remote_state *rs = get_remote_state ();
  struct packet_config *noack_config;
  char *wait_status = NULL;

  immediate_quit++;		/* Allow user to interrupt it.  */
  QUIT;
~~~


We should probably similarly completely tear down an incomplete
attach interrupted by ctrl-c too, but that's a separate issue.

> +    {
> +    }
> +  END_CATCH
> +
> +  TRY
> +    {
> +      symbol_file_add_main (full_exec_path, from_tty);
> +    }
> +  CATCH (err, RETURN_MASK_ALL)
> +    {
> +    }

Throwing is bad, but, should we warn, including
the exception message?

> +  END_CATCH
>  
>    do_cleanups (old_chain);
>  }
> 

Thanks,
Pedro Alves