[Bug binutils/17466] New: objcopy corrupts Windows executables when adding a section

[gintautas.miliauskas at gmail dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=17466

            Bug ID: 17466
           Summary: objcopy corrupts Windows executables when adding a
                    section
           Product: binutils
           Version: 2.24
            Status: NEW
          Severity: normal
          Priority: P2
         Component: binutils
          Assignee: unassigned at sourceware dot org
          Reporter: gintautas.miliauskas at gmail dot com

I've been trying to add some additional information to a Windows .exe file
(compiled by mingw gcc) using objcopy, but whenever I add a new section using
objcopy, the binary fails to execute. Here's a summary script:

$ cd /tmp
$ echo 'int main() { return 0; }' > c.c
$ gcc -o c.exe c.c
$ ./c.exe && echo ok
ok

$ echo x > x
$ objcopy c.exe --add-section .debug_foo=x
$ ./c.exe
bash: ./c.exe: cannot execute binary file: Exec format error

$ objdump -j c.exe
  0 .debug_foo    00000002  0000000100000000  0000000100000000  00000458  2**0
                  CONTENTS, READONLY, DEBUGGING
  1 .text         00001d00  0000000000401000  0000000000401000  00000800  2**4
                  CONTENTS, ALLOC, LOAD, READONLY, CODE
  2 .data         000000a0  0000000000403000  0000000000403000  00002600  2**4
                  CONTENTS, ALLOC, LOAD, DATA
  3 .rdata        00000970  0000000000404000  0000000000404000  00002800  2**4
                  CONTENTS, ALLOC, LOAD, READONLY, DATA
  4 .pdata        00000234  0000000000405000  0000000000405000  00003200  2**2
                  CONTENTS, ALLOC, LOAD, READONLY, DATA
  5 .xdata        000001f8  0000000000406000  0000000000406000  00003600  2**2
                  CONTENTS, ALLOC, LOAD, READONLY, DATA
  6 .bss          00000a60  0000000000407000  0000000000407000  00000000  2**6
                  ALLOC
  7 .idata        000007ac  0000000000408000  0000000000408000  00003800  2**2
                  CONTENTS, ALLOC, LOAD, DATA
  8 .CRT          00000068  0000000000409000  0000000000409000  00004000  2**3
                  CONTENTS, ALLOC, LOAD, DATA
  9 .tls          00000068  000000000040a000  000000000040a000  00004200  2**5
                  CONTENTS, ALLOC, LOAD, DATA
 10 .debug_aranges 000003d0  000000000040b000  000000000040b000  00004400  2**0
                  CONTENTS, READONLY, DEBUGGING
 11 .debug_info   0000b2d4  000000000040c000  000000000040c000  00004800  2**0
                  CONTENTS, READONLY, DEBUGGING
 12 .debug_abbrev 000019cc  0000000000418000  0000000000418000  0000fc00  2**0
                  CONTENTS, READONLY, DEBUGGING
 13 .debug_line   000017c0  000000000041a000  000000000041a000  00011600  2**0
                  CONTENTS, READONLY, DEBUGGING
 14 .debug_frame  00000ac0  000000000041c000  000000000041c000  00012e00  2**3
                  CONTENTS, READONLY, DEBUGGING
 15 .debug_str    00000300  000000000041d000  000000000041d000  00013a00  2**0
                  CONTENTS, READONLY, DEBUGGING
 16 .debug_loc    00002e19  000000000041e000  000000000041e000  00013e00  2**0
                  CONTENTS, READONLY, DEBUGGING
 17 .debug_ranges 00000520  0000000000421000  0000000000421000  00016e00  2**0
                  CONTENTS, READONLY, DEBUGGING


Not sure why the section ends up first, but either way, if I run the binary
through objcopy again, the new section is moved to the end:

$ objcopy c.exe
$ objdump -h c.exe

c.exe:     file format pei-x86-64

Sections:
Idx Name          Size      VMA               LMA               File off  Algn
  0 .text         00001d00  0000000000401000  0000000000401000  00000600  2**4
                  CONTENTS, ALLOC, LOAD, READONLY, CODE
  1 .data         000000a0  0000000000403000  0000000000403000  00002400  2**4
                  CONTENTS, ALLOC, LOAD, DATA
  2 .rdata        00000970  0000000000404000  0000000000404000  00002600  2**4
                  CONTENTS, ALLOC, LOAD, READONLY, DATA
  3 .pdata        00000234  0000000000405000  0000000000405000  00003000  2**2
                  CONTENTS, ALLOC, LOAD, READONLY, DATA
  4 .xdata        000001f8  0000000000406000  0000000000406000  00003400  2**2
                  CONTENTS, ALLOC, LOAD, READONLY, DATA
  5 .bss          00000a60  0000000000407000  0000000000407000  00000000  2**6
                  ALLOC
  6 .idata        000007ac  0000000000408000  0000000000408000  00003600  2**2
                  CONTENTS, ALLOC, LOAD, DATA
  7 .CRT          00000068  0000000000409000  0000000000409000  00003e00  2**3
                  CONTENTS, ALLOC, LOAD, DATA
  8 .tls          00000068  000000000040a000  000000000040a000  00004000  2**5
                  CONTENTS, ALLOC, LOAD, DATA
  9 .debug_aranges 000003d0  000000000040b000  000000000040b000  00004200  2**0
                  CONTENTS, READONLY, DEBUGGING
 10 .debug_info   0000b2d4  000000000040c000  000000000040c000  00004600  2**0
                  CONTENTS, READONLY, DEBUGGING
 11 .debug_abbrev 000019cc  0000000000418000  0000000000418000  0000fa00  2**0
                  CONTENTS, READONLY, DEBUGGING
 12 .debug_line   000017c0  000000000041a000  000000000041a000  00011400  2**0
                  CONTENTS, READONLY, DEBUGGING
 13 .debug_frame  00000ac0  000000000041c000  000000000041c000  00012c00  2**3
                  CONTENTS, READONLY, DEBUGGING
 14 .debug_str    00000300  000000000041d000  000000000041d000  00013800  2**0
                  CONTENTS, READONLY, DEBUGGING
 15 .debug_loc    00002e19  000000000041e000  000000000041e000  00013c00  2**0
                  CONTENTS, READONLY, DEBUGGING
 16 .debug_ranges 00000520  0000000000421000  0000000000421000  00016c00  2**0
                  CONTENTS, READONLY, DEBUGGING
 17 .debug_foo    00000002  0000000100000000  0000000100000000  00017200  2**0
                  CONTENTS, READONLY, DEBUGGING

The corruption is not irreversible: if I remove the section using objcopy, the
binary starts working again:

$ objcopy --remove-section .debug_foo c.exe
$ ./c.exe && echo ok
ok

What's going on here? Is this use case of objcopy intrinsically not supported?
Should I be using something else?

-- 
You are receiving this mail because:
You are on the CC list for the bug.