fix objcopy of PE images with .buildid section
authorJan Beulich <jbeulich@suse.com>
Fri, 21 Aug 2020 08:28:35 +0000 (10:28 +0200)
committerJan Beulich <jbeulich@suse.com>
Fri, 21 Aug 2020 08:28:35 +0000 (10:28 +0200)
Xen Project embeds a build ID in its hypervisor binary (including its
EFI variant), living in a standalone section. This usually gets placed
right after .rodata, and due to the rounding done on the (file) size of
.rodata the two sections appear to overlap (as far as e.g.
find_section_by_vma() is concerned). With the first byte "found" in
.rodata, nothing guarantees that the entire debug dir fits in that
section, leading to apparently random failure of objcopy on such an
image.

Possible alternatives to the solution chosen:
- make find_section_by_vma() honor virt_size,
- correct the recording of sizes elsewhere (ibfd has size == virt_size,
  while obfd doesn't),
- fix the linker to avoid producing apparently overlapping sections.

While touching the condition around and the contents of the disgnostic,
pull it up ahead of the bfd_malloc_and_get_section() call: There's no
point first obtaining the section contents, in order to then fail.

bfd/ChangeLog
bfd/peXXigen.c

index 670803ddfb14b0b69540b5333a379234fecbe756..62407f987aa2ca3ba4c3e102970b7b641dec2510 100644 (file)
@@ -1,3 +1,8 @@
+2020-08-21  Jan Beulich  <jbeulich@suse.com>
+
+       * peXXigen.c (_bfd_XX_bfd_copy_private_bfd_data_common): Check
+       last byte of debug dir, not first.
+
 2020-08-20  Nick Clifton  <nickc@redhat.com>
 
        PR 26428
index 3c3fa27e02004d944c75a60f2e6e40ace1aea563..646ad0f0bf55a135daa3fcab9935b6e8e24941b8 100644 (file)
@@ -2943,29 +2943,33 @@ _bfd_XX_bfd_copy_private_bfd_data_common (bfd * ibfd, bfd * obfd)
     {
       bfd_vma addr = ope->pe_opthdr.DataDirectory[PE_DEBUG_DATA].VirtualAddress
        + ope->pe_opthdr.ImageBase;
-      asection *section = find_section_by_vma (obfd, addr);
+      /* In particular a .buildid section may overlap (in VA space) with
+        whatever section comes ahead of it (largely because of section->size
+        representing s_size, not virt_size).  Therefore don't look for the
+        section containing the first byte, but for that covering the last
+        one.  */
+      bfd_vma last = addr + ope->pe_opthdr.DataDirectory[PE_DEBUG_DATA].Size - 1;
+      asection *section = find_section_by_vma (obfd, last);
       bfd_byte *data;
 
+      /* PR 17512: file: 0f15796a.  */
+      if (section && addr < section->vma)
+       {
+         /* xgettext:c-format */
+         _bfd_error_handler
+           (_("%pB: Data Directory (%lx bytes at %" PRIx64 ") "
+              "extends across section boundary at %" PRIx64),
+            obfd, ope->pe_opthdr.DataDirectory[PE_DEBUG_DATA].Size,
+            (uint64_t) addr, (uint64_t) section->vma);
+         return FALSE;
+       }
+
       if (section && bfd_malloc_and_get_section (obfd, section, &data))
        {
          unsigned int i;
          struct external_IMAGE_DEBUG_DIRECTORY *dd =
            (struct external_IMAGE_DEBUG_DIRECTORY *)(data + (addr - section->vma));
 
-         /* PR 17512: file: 0f15796a.  */
-         if ((unsigned long) ope->pe_opthdr.DataDirectory[PE_DEBUG_DATA].Size
-             > section->size - (addr - section->vma))
-           {
-             /* xgettext:c-format */
-             _bfd_error_handler
-               (_("%pB: Data Directory size (%lx) "
-                  "exceeds space left in section (%" PRIx64 ")"),
-                obfd, ope->pe_opthdr.DataDirectory[PE_DEBUG_DATA].Size,
-                (uint64_t) (section->size - (addr - section->vma)));
-             free (data);
-             return FALSE;
-           }
-
          for (i = 0; i < ope->pe_opthdr.DataDirectory[PE_DEBUG_DATA].Size
                 / sizeof (struct external_IMAGE_DEBUG_DIRECTORY); i++)
            {