[Bug gdb/14704] New: GDB fails to decompress relocatable files

[Bug gdb/14704] New: GDB fails to decompress relocatable files

[hjl.tools at gmail dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=14704

             Bug #: 14704
           Summary: GDB fails to decompress relocatable files
           Product: gdb
           Version: unknown
            Status: NEW
          Severity: normal
          Priority: P2
         Component: gdb
        AssignedTo: unassigned@sourceware.org
        ReportedBy: hjl.tools@gmail.com
    Classification: Unclassified


[hjl@gnu-tools-1 tmp]$ cat x.c
void bar ();

int
main ()
{
  bar ();
  return 0;
}
[hjl@gnu-tools-1 tmp]$ gcc -g x.c -c
[hjl@gnu-tools-1 tmp]$ objcopy --compress-debug-sections x.o
[hjl@gnu-tools-1 tmp]$ gdb --batch x.o
Dwarf Error: wrong version in compilation unit header (is 0, should be 2, 3, or
4) [in module /tmp/x.o]
[hjl@gnu-tools-1 tmp]$ objcopy --decompress-debug-sections x.o
[hjl@gnu-tools-1 tmp]$ gdb --batch x.o
[hjl@gnu-tools-1 tmp]$

-- 
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.

[Bug gdb/14704] GDB fails to decompress relocatable files

[dje at google dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=14704

dje at google dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |dje at google dot com,
                   |                            |tromey at redhat dot com

--- Comment #1 from dje at google dot com 2012-10-11 23:12:39 UTC ---
Filing this here for reference sake, having just tripped over this myself in a
different (non .o) context.

In 7.5 the check for whether a section is compressed appears before the check
for whether a section has relocs.  In cvs head the tests are (effectively)
reversed.

The change came in here:

2012-07-18  Tom Tromey  <tromey@redhat.com>

        * dwarf2read.c: Don't include zlib.h or sys/mman.h.
        (pagesize): Remove.
        (struct dwarf2_section_info) <map_addr, map_len>: Remove.
        (zlib_decompress_section): Remove.
        (dwarf2_read_section): Use gdb_bfd_map_section.
        (munmap_section_buffer): Remove.
        (free_dwo_file, dwarf2_per_objfile_free): Don't use
        munmap_section_buffer.
        * gdb_bfd.c: Include zlib.h, sys/mman.h.
        (struct gdb_bfd_section_data): New.
        (free_one_bfd_section): New function.
        (gdb_bfd_close_or_warn): Use free_one_bfd_section.
        (get_section_descriptor, zlib_decompress_section)
        (gdb_bfd_map_section): New functions.
        * gdb_bfd.h (gdb_bfd_map_section): Declare.

Also, I think there's a separate issue here, which is: Why does bfd attach
SEC_RELOC to a section if its .rela section is empty?

Seems like we want this, but of course I can't discount the possibility of some
obscure reason why things are the way they are:

--- elf.c=    2012-10-11 15:55:02.722802893 -0700
+++ elf.c    2012-10-11 15:55:02.722802893 -0700
@@ -1873,18 +1873,20 @@
        *hdr2 = *hdr;
        *p_hdr = hdr2;
        elf_elfsections (abfd)[shindex] = hdr2;
-       target_sect->reloc_count += NUM_SHDR_ENTRIES (hdr);
-       target_sect->flags |= SEC_RELOC;
-       target_sect->relocation = NULL;
-       target_sect->rel_filepos = hdr->sh_offset;
-       /* In the section to which the relocations apply, mark whether
-          its relocations are of the REL or RELA variety.  */
+       /* Don't mark the target section as having relocs if this
+          section is empty.  */
        if (hdr->sh_size != 0)
          {
+           target_sect->reloc_count += NUM_SHDR_ENTRIES (hdr);
+           target_sect->flags |= SEC_RELOC;
+           target_sect->relocation = NULL;
+           target_sect->rel_filepos = hdr->sh_offset;
+           /* In the section to which the relocations apply, mark whether
+              its relocations are of the REL or RELA variety.  */
            if (hdr->sh_type == SHT_RELA)
              target_sect->use_rela_p = 1;
+           abfd->flags |= HAS_RELOC;
          }
-       abfd->flags |= HAS_RELOC;
        return TRUE;
       }

-- 
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.

[Bug gdb/14704] GDB fails to decompress relocatable files

[tromey at redhat dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=14704

--- Comment #2 from Tom Tromey <tromey at redhat dot com> 2012-10-15 17:45:15 UTC ---
FWIW, 7.5 decompressed the section but didn't actually apply relocations.
That seems wrong.

There are comments in dwarf2read (now in gdb_bfd.c) dating back to 2008
about this:

1.255        (csilvers 19-Apr-08):   /* When debugging .o files, we may need to
apply relocations; see
1.255        (csilvers 19-Apr-08):     
http://sourceware.org/ml/gdb-patches/2002-04/msg00136.html .
1.255        (csilvers 19-Apr-08):      We never compress sections in .o files,
so we only need to
1.255        (csilvers 19-Apr-08):      try this when the section is not
compressed.  */


I assume that what should happen instead is that the section be decompressed,
then relocations applied.

-- 
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.

[Bug gdb/14704] GDB fails to decompress relocatable files

[tromey at redhat dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=14704

--- Comment #3 from Tom Tromey <tromey at redhat dot com> 2012-10-15 18:15:37 UTC ---
We arrive in bfd_get_full_section_contents with
sec->compress_status == COMPRESS_SECTION_NONE.
This is because BFD_DECOMPRESS is not set when calling
_bfd_elf_make_section_from_shdr.

Maybe having gdb set BFD_DECOMPRESS is the right thing to do.
It is not clear to me; BFD says that this is "for BFD use only",
but then I see uses of it in ld and binutils.

If this is the correct approach we could easily set the flag
in the various wrapper functions in gdb_bfd.c.

-- 
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.

[Bug gdb/14704] GDB fails to decompress relocatable files

[hjl.tools at gmail dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=14704

--- Comment #4 from H.J. Lu <hjl.tools at gmail dot com> 2012-10-15 20:45:31 UTC ---
(In reply to comment #3)
> We arrive in bfd_get_full_section_contents with
> sec->compress_status == COMPRESS_SECTION_NONE.
> This is because BFD_DECOMPRESS is not set when calling
> _bfd_elf_make_section_from_shdr.
> 
> Maybe having gdb set BFD_DECOMPRESS is the right thing to do.
> It is not clear to me; BFD says that this is "for BFD use only",
> but then I see uses of it in ld and binutils.
> 
> If this is the correct approach we could easily set the flag
> in the various wrapper functions in gdb_bfd.c.

Yes, it is the right thing to do.

-- 
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.

[Bug gdb/14704] GDB fails to decompress relocatable files

[hjl.tools at gmail dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=14704

--- Comment #5 from H.J. Lu <hjl.tools at gmail dot com> 2012-10-15 20:49:47 UTC ---
(In reply to comment #3)
>
> Maybe having gdb set BFD_DECOMPRESS is the right thing to do.
> It is not clear to me; BFD says that this is "for BFD use only",
> but then I see uses of it in ld and binutils.
> 

It means BFD_DECOMPRESS is used by BFD, not used to describe
the input file:

binutils/objdump.c:  printf (_("flags 0x%08x:\n"), abfd->flags &
~BFD_FLAGS_FOR_BFD_USE_MASK);

We don't want to display those bits in objdump.

-- 
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.

[Bug gdb/14704] GDB fails to decompress relocatable files

[tromey at redhat dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=14704

--- Comment #6 from Tom Tromey <tromey at redhat dot com> 2012-10-16 18:06:18 UTC ---
While trying a cleanup here I found out that
bfd_is_section_compressed does not work if
BFD_DECOMPRESS is set.  This is because bfd_get_section_contents
fails for compressed sections.

-- 
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.

[Bug gdb/14704] GDB fails to decompress relocatable files

[tromey at redhat dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=14704

Tom Tromey <tromey at redhat dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|unassigned at sourceware    |tromey at redhat dot com
                   |dot org                     |

--- Comment #7 from Tom Tromey <tromey at redhat dot com> 2012-10-16 18:16:57 UTC ---
Testing a patch.

-- 
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.

[Bug gdb/14704] GDB fails to decompress relocatable files

[cvs-commit at gcc dot gnu.org]

http://sourceware.org/bugzilla/show_bug.cgi?id=14704

--- Comment #8 from cvs-commit at gcc dot gnu.org <cvs-commit at gcc dot gnu.org> 2012-11-08 19:52:47 UTC ---
CVSROOT:    /cvs/src
Module name:    src
Changes by:    tromey@sourceware.org    2012-11-08 19:52:42

Modified files:
    gdb            : ChangeLog gdb_bfd.c osabi.c xcoffread.c 
    gdb/contrib    : cc-with-tweaks.sh 
    gdb/python     : py-auto-load.c 
    gdb/testsuite  : ChangeLog 
Added files:
    gdb/testsuite/gdb.base: comprdebug.exp 

Log message:
    PR gdb/14704:
    * gdb_bfd.c (gdb_bfd_ref): Set BFD_DECOMPRESS.
    (zlib_decompress_section): Remove.
    (gdb_bfd_map_section): Only check for compressed section
    in mmap case.  Use bfd_get_full_section_contents.
    * osabi.c (check_note): Add 'sectsize' argument.  Read
    section data.
    (generic_elf_osabi_sniff_abi_tag_sections): Don't read
    section data.  Update for check_note change.
    * xcoffread.c (xcoff_initial_scan): Use
    bfd_get_full_section_contents.
    * py-auto-load.c (auto_load_section_scripts): Use
    bfd_get_full_section_contents.
    * contrib/cc-with-tweaks.sh: Add -Z option.
    testsuite
    * gdb.base/comprdebug.exp: New file.

Patches:
http://sourceware.org/cgi-bin/cvsweb.cgi/src/gdb/ChangeLog.diff?cvsroot=src&r1=1.14800&r2=1.14801
http://sourceware.org/cgi-bin/cvsweb.cgi/src/gdb/gdb_bfd.c.diff?cvsroot=src&r1=1.12&r2=1.13
http://sourceware.org/cgi-bin/cvsweb.cgi/src/gdb/osabi.c.diff?cvsroot=src&r1=1.58&r2=1.59
http://sourceware.org/cgi-bin/cvsweb.cgi/src/gdb/xcoffread.c.diff?cvsroot=src&r1=1.97&r2=1.98
http://sourceware.org/cgi-bin/cvsweb.cgi/src/gdb/contrib/cc-with-tweaks.sh.diff?cvsroot=src&r1=1.4&r2=1.5
http://sourceware.org/cgi-bin/cvsweb.cgi/src/gdb/python/py-auto-load.c.diff?cvsroot=src&r1=1.24&r2=1.25
http://sourceware.org/cgi-bin/cvsweb.cgi/src/gdb/testsuite/ChangeLog.diff?cvsroot=src&r1=1.3443&r2=1.3444
http://sourceware.org/cgi-bin/cvsweb.cgi/src/gdb/testsuite/gdb.base/comprdebug.exp.diff?cvsroot=src&r1=NONE&r2=1.1

-- 
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.

[Bug gdb/14704] GDB fails to decompress relocatable files

[tromey at redhat dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=14704

Tom Tromey <tromey at redhat dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|                            |FIXED
   Target Milestone|---                         |7.6

--- Comment #9 from Tom Tromey <tromey at redhat dot com> 2012-11-08 19:53:40 UTC ---
Fixed.

-- 
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.