PATCH: PR binutils/12639: nm/readelf failed to detect corrupted symbol table

PATCH: PR binutils/12639: nm/readelf failed to detect corrupted symbol table

[H.J. Lu]

Hi,

Some tools may generate corrupted symbol table such that global symbols
are treated as local.  Linker will complain undefined symbols while
outputs from nm/readelf show the symbols are defined and global.  This
patch adds corrupted symbol table check to nm/readelf.  OK to install?

Thanks.


H.J.
---
bfd/

2013-11-05  H.J. Lu  <hongjiu.lu@intel.com>

	PR binutils/12639
	* elfcode.h (elf_slurp_symbol_table): Check corrupted global
	symbols.

binutils/

2013-11-05  H.J. Lu  <hongjiu.lu@intel.com>

	PR binutils/12639
	* readelf.c (process_symbol_table): Detect corrupted symbol
	table.
diff --git a/bfd/elfcode.h b/bfd/elfcode.h
index e296c5c..cf45c27 100644
--- a/bfd/elfcode.h
+++ b/bfd/elfcode.h
@@ -1168,6 +1168,9 @@ elf_slurp_symbol_table (bfd *abfd, asymbol **symptrs, bfd_boolean dynamic)
     sym = symbase = NULL;
   else
     {
+      /* Start of global symbols */
+      Elf_Internal_Sym *start_global;
+
       isymbuf = bfd_elf_get_elf_syms (abfd, hdr, symcount, 0,
 				      NULL, NULL, NULL);
       if (isymbuf == NULL)
@@ -1212,6 +1215,9 @@ elf_slurp_symbol_table (bfd *abfd, asymbol **symptrs, bfd_boolean dynamic)
       if (xver != NULL)
 	++xver;
       isymend = isymbuf + symcount;
+      start_global = isymbuf;
+      if (!elf_bad_symtab (abfd))
+	start_global += hdr->sh_info;
       for (isym = isymbuf + 1, sym = symbase; isym < isymend; isym++, sym++)
 	{
 	  memcpy (&sym->internal_elf_sym, isym, sizeof (Elf_Internal_Sym));
@@ -1270,6 +1276,12 @@ elf_slurp_symbol_table (bfd *abfd, asymbol **symptrs, bfd_boolean dynamic)
 	  if ((abfd->flags & (EXEC_P | DYNAMIC)) != 0)
 	    sym->symbol.value -= sym->symbol.section->vma;
 
+	  if (isym < start_global
+	      && ELF_ST_BIND (isym->st_info) != STB_LOCAL)
+	    (*_bfd_error_handler)
+	      (_("%s: corrupted global symbol `%s' treated as local"),
+	       abfd->filename, sym->symbol.name);
+
 	  switch (ELF_ST_BIND (isym->st_info))
 	    {
 	    case STB_LOCAL:
diff --git a/binutils/readelf.c b/binutils/readelf.c
index 0389f14..9ae5b5d 100644
--- a/binutils/readelf.c
+++ b/binutils/readelf.c
@@ -9607,6 +9607,12 @@ process_symbol_table (FILE * file)
   else if (do_dyn_syms || (do_syms && !do_using_dynamic))
     {
       unsigned int i;
+      /* Irix 5 and 6 are broken.  Object file symbol tables are not
+	 always sorted correctly such that local symbols precede global
+	 symbols, and the sh_info field in the symbol table is not
+	 always right.  */
+      bfd_boolean check_corrupt_symtab
+	= elf_header.e_ident[EI_OSABI] != ELFOSABI_IRIX;
 
       for (i = 0, section = section_headers;
 	   i < elf_header.e_shnum;
@@ -9669,7 +9675,12 @@ process_symbol_table (FILE * file)
 	      putchar (' ');
 	      print_vma (psym->st_size, DEC_5);
 	      printf (" %-7s", get_symbol_type (ELF_ST_TYPE (psym->st_info)));
-	      printf (" %-6s", get_symbol_binding (ELF_ST_BIND (psym->st_info)));
+	      if (check_corrupt_symtab
+		  && si < section->sh_info
+		  && ELF_ST_BIND (psym->st_info) != STB_LOCAL)
+		printf (" %-6s", "<corrupt>");
+	      else
+		printf (" %-6s", get_symbol_binding (ELF_ST_BIND (psym->st_info)));
 	      printf (" %-7s", get_symbol_visibility (ELF_ST_VISIBILITY (psym->st_other)));
 	      /* Check to see if any other bits in the st_other field are set.
 	         Note - displaying this information disrupts the layout of the

Re: PATCH: PR binutils/12639: nm/readelf failed to detect corrupted symbol table

[Richard Sandiford]

"H.J. Lu" <hongjiu.lu@intel.com> writes:
> Some tools may generate corrupted symbol table such that global symbols
> are treated as local.  Linker will complain undefined symbols while
> outputs from nm/readelf show the symbols are defined and global.  This
> patch adds corrupted symbol table check to nm/readelf.  OK to install?

Hmm, what was the outcome of:

    https://sourceware.org/ml/binutils/2011-04/msg00088.html

which seemed to be the last word on the original patch?  FWIW I agree
that printing "<corrupt>" instead of the binding would be confusing.
You said:

    https://sourceware.org/ml/binutils/2011-04/msg00100.html

but we could be more helpful than "Bad symtab".  E.g.:

    Symbol %d is in the local area but is not STB_LOCAL

Or if you wanted to get really fancy you could stick a [*] at
the end of the line and say something like:

    [*] These symbols are in the local area but are not STB_LOCAL

at the end (all text just examples).

[BTW, sorry for butting in, but the MIPS reference caught my eye...]

Thanks,
Richard