Re: [PATCH]: gdb/769 - segv fault on "info shared" on GDB 5.2.1 HPUX64 11.00

Re: [PATCH]: gdb/769 - segv fault on "info shared" on GDB 5.2.1 HPUX64 11.00

[Josh Martin]

> On Oct 9,  5:42pm, Josh Martin wrote:
> 
> > Here's the ChangeLog entry that I forgot to post.  Unfortunately I cannot 
get 
> > expect to work on my system, so I can neither create a test case for this 
bug, 
> > nor verify other tests after this patch.
> > 
> >  - Josh Martin
> >  
> > 2002-09-28  Josh Martin  <timeslice@iname.com>
> >  
> > 	* solib.c (info_sharedlibrary_command): Added catch for potential
> > 	dereference of NULL pointer (current_target_so_ops).
> > 	Fix PR gdb/769.
> >  	
> > > For platforms that aren't covered by the gdb/solib-foo.c files and the 
gdbarch
> > > platform dependancy files the "info sharedlibrary" command will cause a 
> > > segmentation fault by dereferencing a NULL pointer (current_target_so_ops) 
in 
> > > gdb/solib.c:update_solib_list.  The patch checks if current_target_so_ops 
is 
> > > NULL, and if so it responds with a "command not implemented" message.
> > > 
> > > What I really wanted to do was to implement support for HPUX 11.00 64-bit 
> > w/GCC. 
> > > It shouldn't be too difficult as 64-bit GCC in HPUX 11.00 uses GNU ld and 
the 
> > > "standard" elf64hppa object format. Unfortunately I had no idea how to 
proceed 
> > > or where to find the neccesary information, thus I stuck with this 
"band-aid" 
> > > patch.
>
> Date: Thu, 5 Dec 2002 17:54:05 -0700
> From: Kevin Buettner <kevinb@redhat.com>
> 
> I've been thinking about this patch some more.
> 
> What I'm wondering is why your hpux target uses solib.o without defining
> an appropriate solib-hpux.c (or whatever) file?
> 
> Anyway, with regard to catching potential dereferences to a NULL
> current_target_so_ops, I'm inclined to handle this either via
> a gdb_assert() or an explicit check (perhaps in the TARGET_SO_* macros)
> with a call to internal_error().  Because that's really what it is.  We
> shouldn't be in solib.c at all if an appropriate backend hasn't been
> defined.
> 
> Kevin

I'm not involved with the HPUX maintainance/port of GDB, so I'm not completely 
sure, but I think there already is a backend for the HPUX libarary format (which 
is in 32-bit), but on a 64-bit HP platform the libraries are written in a 64-bit 
ELF format, at least with gcc and GNU ld, which goes through pa64solib.c.  I 
also noticed that GDB isn't reading the core files properly, it is still trying 
to open them in the 32-bit format, but I'm not sure if this is a problem with 
GDB, or the BFD libraries.

 - Josh Martin

Re: [PATCH]: gdb/769 - segv fault on "info shared" on GDB 5.2.1 HPUX64 11.00

[Josh Martin]

I was not familiar with how gdb called errors, so I just looked at similar code 
from that source file. If what I used was not the correct method then by all 
means change the patch, or point me to the definition of error() so I can change 
it appropriately.

 - Josh Martin
 
> Date: Mon, 25 Nov 2002 16:08:20 -0700
> From: Kevin Buettner <kevinb@redhat.com>
> To: Josh Martin <Josh.Martin@abq.sc.philips.com>, 
gdb-patches@sources.redhat.com
> Subject: Re: [PATCH]: gdb/769 - segv fault on "info shared" on GDB 5.2.1 
HPUX64 11.00
> MIME-Version: 1.0
> 
> On Oct 9,  5:42pm, Josh Martin wrote:
> 
> > 2002-09-28  Josh Martin  <timeslice@iname.com>
> >  
> > 	* solib.c (info_sharedlibrary_command): Added catch for potential
> > 	dereference of NULL pointer (current_target_so_ops).
> > 	Fix PR gdb/769.
> 
> Josh,
> 
> Your patch looks reasonable.  I think I'd prefer to see a call to error()
> instead of printf_unfiltered() though:
> 
> > > +      printf_unfiltered("Shared library info is not yet implemented on 
this "
> > > +			"platform\n");
> 
> Sorry for taking so long on this.
> 
> Kevin

Re: [PATCH]: gdb/769 - segv fault on "info shared" on GDB 5.2.1 HPUX64 11.00

[Josh Martin]

Here's the ChangeLog entry that I forgot to post.  Unfortunately I cannot get 
expect to work on my system, so I can neither create a test case for this bug, 
nor verify other tests after this patch.

 - Josh Martin
 
2002-09-28  Josh Martin  <timeslice@iname.com>
 
	* solib.c (info_sharedlibrary_command): Added catch for potential
	dereference of NULL pointer (current_target_so_ops).
	Fix PR gdb/769.
 	
> For platforms that aren't covered by the gdb/solib-foo.c files and the gdbarch
> platform dependancy files the "info sharedlibrary" command will cause a 
> segmentation fault by dereferencing a NULL pointer (current_target_so_ops) in 
> gdb/solib.c:update_solib_list.  The patch checks if current_target_so_ops is 
> NULL, and if so it responds with a "command not implemented" message.
> 
> What I really wanted to do was to implement support for HPUX 11.00 64-bit 
w/GCC. 
> It shouldn't be too difficult as 64-bit GCC in HPUX 11.00 uses GNU ld and the 
> "standard" elf64hppa object format. Unfortunately I had no idea how to proceed 
> or where to find the neccesary information, thus I stuck with this "band-aid" 
> patch.
> 
> What follows is a diff -up patch for gdb-5.2.1/gdb/solib.c
> 
>  - Josh Martin
> 
>  
> --- solib.c~	Tue Feb 26 18:40:35 2002
> +++ solib.c	Sat Sep 28 08:43:03 2002
> @@ -619,6 +619,19 @@ info_sharedlibrary_command (char *ignore
>    int addr_width;
>    char *addr_fmt;
>  
> +  /*
> +    This is usually defined in a platform specific solib-foo.c/.h
> +    file.  Eventually the definition should be migrated to the
> +    gdbarch files.  In the mean time keep this from core dumping on
> +    platforms where there is no implementation.
> +  */
> +  if (current_target_so_ops == NULL)
> +    {
> +      printf_unfiltered("Shared library info is not yet implemented on this "
> +			"platform\n");
> +      return;
> +    }
> +
>    if (TARGET_PTR_BIT == 32)
>      {
>        addr_width = 8 + 4;
> 

Re: [PATCH]: gdb/769 - segv fault on "info shared" on GDB 5.2.1 HPUX64 11.00

[Kevin Buettner]

On Oct 9,  5:42pm, Josh Martin wrote:

> 2002-09-28  Josh Martin  <timeslice@iname.com>
>  
> 	* solib.c (info_sharedlibrary_command): Added catch for potential
> 	dereference of NULL pointer (current_target_so_ops).
> 	Fix PR gdb/769.

Josh,

Your patch looks reasonable.  I think I'd prefer to see a call to error()
instead of printf_unfiltered() though:

> > +      printf_unfiltered("Shared library info is not yet implemented on this "
> > +			"platform\n");

Sorry for taking so long on this.

Kevin

Re: [PATCH]: gdb/769 - segv fault on "info shared" on GDB 5.2.1 HPUX64 11.00

[Kevin Buettner]

On Oct 9,  5:42pm, Josh Martin wrote:

> Here's the ChangeLog entry that I forgot to post.  Unfortunately I cannot get 
> expect to work on my system, so I can neither create a test case for this bug, 
> nor verify other tests after this patch.
> 
>  - Josh Martin
>  
> 2002-09-28  Josh Martin  <timeslice@iname.com>
>  
> 	* solib.c (info_sharedlibrary_command): Added catch for potential
> 	dereference of NULL pointer (current_target_so_ops).
> 	Fix PR gdb/769.
>  	
> > For platforms that aren't covered by the gdb/solib-foo.c files and the gdbarch
> > platform dependancy files the "info sharedlibrary" command will cause a 
> > segmentation fault by dereferencing a NULL pointer (current_target_so_ops) in 
> > gdb/solib.c:update_solib_list.  The patch checks if current_target_so_ops is 
> > NULL, and if so it responds with a "command not implemented" message.
> > 
> > What I really wanted to do was to implement support for HPUX 11.00 64-bit 
> w/GCC. 
> > It shouldn't be too difficult as 64-bit GCC in HPUX 11.00 uses GNU ld and the 
> > "standard" elf64hppa object format. Unfortunately I had no idea how to proceed 
> > or where to find the neccesary information, thus I stuck with this "band-aid" 
> > patch.

I've been thinking about this patch some more.

What I'm wondering is why your hpux target uses solib.o without defining
an appropriate solib-hpux.c (or whatever) file?

Anyway, with regard to catching potential dereferences to a NULL
current_target_so_ops, I'm inclined to handle this either via
a gdb_assert() or an explicit check (perhaps in the TARGET_SO_* macros)
with a call to internal_error().  Because that's really what it is.  We
shouldn't be in solib.c at all if an appropriate backend hasn't been
defined.

Kevin

[PATCH]: gdb/769 - segv fault on "info shared" on GDB 5.2.1 HPUX64 11.00

[Josh Martin]

For platforms that aren't covered by the gdb/solib-foo.c files and the gdbarch
platform dependancy files the "info sharedlibrary" command will cause a 
segmentation fault by dereferencing a NULL pointer (current_target_so_ops) in 
gdb/solib.c:update_solib_list.  The patch checks if current_target_so_ops is 
NULL, and if so it responds with a "command not implemented" message.

What I really wanted to do was to implement support for HPUX 11.00 64-bit w/GCC. 
It shouldn't be too difficult as 64-bit GCC in HPUX 11.00 uses GNU ld and the 
"standard" elf64hppa object format. Unfortunately I had no idea how to proceed 
or where to find the neccesary information, thus I stuck with this "band-aid" 
patch.

What follows is a diff -up patch for gdb-5.2.1/gdb/solib.c

 - Josh Martin

 
--- solib.c~	Tue Feb 26 18:40:35 2002
+++ solib.c	Sat Sep 28 08:43:03 2002
@@ -619,6 +619,19 @@ info_sharedlibrary_command (char *ignore
   int addr_width;
   char *addr_fmt;
 
+  /*
+    This is usually defined in a platform specific solib-foo.c/.h
+    file.  Eventually the definition should be migrated to the
+    gdbarch files.  In the mean time keep this from core dumping on
+    platforms where there is no implementation.
+  */
+  if (current_target_so_ops == NULL)
+    {
+      printf_unfiltered("Shared library info is not yet implemented on this "
+			"platform\n");
+      return;
+    }
+
   if (TARGET_PTR_BIT == 32)
     {
       addr_width = 8 + 4;