[Bug binutils/5299] New: [bfd] after bfd_check_format_matches() we can have duplicated sections

[darkjames at darkjames dot ath dot cx]

When we build bfd with --enable-targets=all

and test objdump on sample PE file:

$ ../src/binutils/objdump -h ./a.exe    

We've got:
#v+
./a.exe:     file format pei-i386

Sections:
Idx Name          Size      VMA               LMA               File off  Algn
  0 .text         0013412a  00401000  00401000  00001000  2**2
                  CONTENTS, ALLOC, LOAD, READONLY, CODE
  1 .rdata        0005775f  00536000  00536000  00136000  2**2
                  CONTENTS, ALLOC, LOAD, READONLY, DATA
  2 .data         00007000  0058e000  0058e000  0018e000  2**2
                  CONTENTS, ALLOC, LOAD, DATA
  3 .tls          00000029  005a9000  005a9000  00195000  2**2
                  CONTENTS, ALLOC, LOAD, DATA
  4 .rsrc         0000c170  005aa000  005aa000  00196000  2**2
                  CONTENTS, ALLOC, LOAD, READONLY, DATA
  5 .text         0013412a  00401000  00401000  00001000  2**4
                  CONTENTS, ALLOC, LOAD, READONLY, CODE
  6 .rdata        0005775f  00536000  00536000  00136000  2**4
                  CONTENTS, ALLOC, LOAD, READONLY, DATA
  7 .data         00007000  0058e000  0058e000  0018e000  2**4
                  CONTENTS, ALLOC, LOAD, DATA
  8 .tls          00000029  005a9000  005a9000  00195000  2**2
                  CONTENTS, ALLOC, LOAD, DATA
  9 .rsrc         0000c170  005aa000  005aa000  00196000  2**2
                  CONTENTS, ALLOC, LOAD, READONLY, DATA
#v-

Both: pei-i386 and efi-app-ia32 adds sections, however because of check in
pe_bfd_object_p():
#v+
  target = coff_object_p (abfd);       /* !!! here: we add sections to abfd */
  if (target) {
      bfd_boolean efi = i->Subsystem == IMAGE_SUBSYSTEM_EFI_APPLICATION;
      /* Pseudo language:
             check if we've got proper target. When efi is set: it must starts
with efi-
             else it must starts with pei-
     */
#v-

Although it doesn't matter if efi is set or no. When we build pei-i386 and
efi-app-ia32 (for x86 app) we always would get duplications.

efi- or pei- target is discared with: bfd_error_wrong_format (but sections which
was added in coff_object_p() isn't discared!)

So in bfd_check_format_matches() we have only one proper target.
So programmer after successed bfd_check_format_matches()
continue his work, and he thinks everything is fine.

However after first run of bfd_check_format_matches() we already know proper
target, so we can once again execute objdump with -b set

#v+
$ ../src/binutils/objdump -h ./a.exe -b pei-i386

./a.exe:     file format pei-i386

Sections:
Idx Name          Size      VMA               LMA               File off  Algn
  0 .text         0013412a  00401000  00401000  00001000  2**4
                  CONTENTS, ALLOC, LOAD, READONLY, CODE
  1 .rdata        0005775f  00536000  00536000  00136000  2**4
                  CONTENTS, ALLOC, LOAD, READONLY, DATA
  2 .data         00007000  0058e000  0058e000  0018e000  2**4
                  CONTENTS, ALLOC, LOAD, DATA
  3 .tls          00000029  005a9000  005a9000  00195000  2**2
                  CONTENTS, ALLOC, LOAD, DATA
  4 .rsrc         0000c170  005aa000  005aa000  00196000  2**2
                  CONTENTS, ALLOC, LOAD, READONLY, DATA
#v-

and it's good.

So workaround for now is:

When bfd_check_format_matches() success with only one target.
Save target name. Close bfd, bfd_set_default_target() to that target. Pass to
bfd_openr() that target. And hope it's ok.

But I think it's hack, and bfd_check_format_matches() should matter about this
cases.

For example: when checking target don't use passed bfd. But only copy of it.
Or second option: check if target is good before operating on bfd (use only
private struct?) (I found this bug in two targets, don't know if it's more this
kind bugs)

When target is good, and it's only one: 
   run again check only for that target (to read sections)
   or use saved copy (if we save bfd's among test)

-- 
           Summary: [bfd] after bfd_check_format_matches() we can have
                    duplicated sections
           Product: binutils
           Version: unspecified
            Status: NEW
          Severity: normal
          Priority: P2
         Component: binutils
        AssignedTo: unassigned at sources dot redhat dot com
        ReportedBy: darkjames at darkjames dot ath dot cx
                CC: bug-binutils at gnu dot org
 GCC build triplet: x86_64-pc-linux-gnu
  GCC host triplet: x86_64-pc-linux-gnu
GCC target triplet: --enable-targets=all (but connected with pei-i386 <==>
                    efi-app-i


http://sourceware.org/bugzilla/show_bug.cgi?id=5299

------- You are receiving this mail because: -------
You are on the CC list for the bug, or are watching someone who is.