Red Hat PIE patch

Red Hat PIE patch

[Michael Snyder]

Does anyone know anything about a "Red Hat PIE patch"
for position independent executables?  Has it been merged?
If not, where can I find the most recent version?

Thanks,
Michael

Re: Red Hat PIE patch

[Joel Brobecker]

> Does anyone know anything about a "Red Hat PIE patch"
> for position independent executables?  Has it been merged?
> If not, where can I find the most recent version?

Jan re-submitted a series recently, and Tom reviewed most if not
all of them. You should be able to find the relevant messages
in the Nov archives.

-- 
Joel

Re: Red Hat PIE patch

[Tom Tromey]

>>>>> "Joel" == Joel Brobecker <brobecker@adacore.com> writes:

>> Does anyone know anything about a "Red Hat PIE patch"
>> for position independent executables?  Has it been merged?
>> If not, where can I find the most recent version?

Joel> Jan re-submitted a series recently, and Tom reviewed most if not
Joel> all of them. You should be able to find the relevant messages
Joel> in the Nov archives.

I have not yet reviewed numbers 9, 12, and 14.  These touched on things
that I am less familiar with.  I hope to get to them, but if someone
else wants to take a look, that is fine with me.

Tom

Re: Red Hat PIE patch

[Michael Snyder]

Tom Tromey wrote:
>>>>>> "Joel" == Joel Brobecker <brobecker@adacore.com> writes:
> 
>>> Does anyone know anything about a "Red Hat PIE patch"
>>> for position independent executables?  Has it been merged?
>>> If not, where can I find the most recent version?
> 
> Joel> Jan re-submitted a series recently, and Tom reviewed most if not
> Joel> all of them. You should be able to find the relevant messages
> Joel> in the Nov archives.
> 
> I have not yet reviewed numbers 9, 12, and 14.  These touched on things
> that I am less familiar with.  I hope to get to them, but if someone
> else wants to take a look, that is fine with me.

Just FYI, I've applied the patches and done basic "aliveness"
testing.  They do seem to work for me, and I will be doing more
with them so I may have further testing words later.

Re: Red Hat PIE patch

[Jan Kratochvil]

On Mon, 07 Dec 2009 23:06:33 +0100, Michael Snyder wrote:
> Tom Tromey wrote:
> > I have not yet reviewed numbers 9, 12, and 14.  These touched on things
> > that I am less familiar with.  I hope to get to them, but if someone
> > else wants to take a look, that is fine with me.

http://sourceware.org/ml/gdb-patches/2009-11/msg00169.html
Patch 2 should be probably dropped as the patch itself makes no sense.  There
there is IMO some non-stop race unrelated to the PIE patched but triggered by
the patchset (causing testsuite regression).

http://sourceware.org/ml/gdb-patches/2009-11/msg00170.html
Patch 3 has wrong gdbarch dealing but it should affect only Cell
frame-specific arches, if at all so did not get to it back.  Shouldn't be
rather `gdbarch' moved from `struct bp_location' to `struct address_space'?
But got lost in the details of gdbarch by Markus Deuling / Ulrich Weigand.


> Just FYI, I've applied the patches and done basic "aliveness"
> testing.  They do seem to work for me, and I will be doing more
> with them so I may have further testing words later.

Thanks, BTW there are now some add-ons to that series which are still not
properly split/integrated/reposted.
	http://cvs.fedora.redhat.com/viewvc/rpms/gdb/F-12/gdb-archer-pie-addons.patch?content-type=text%2Fplain&view=co


Regards,
Jan

Re: Red Hat PIE patch

[Jan Kratochvil]

On Tue, 08 Dec 2009 14:56:43 +0100, Jan Kratochvil wrote:
> there are now some add-ons to that series which are still not
> properly split/integrated/reposted.
> 	http://cvs.fedora.redhat.com/viewvc/rpms/gdb/F-12/gdb-archer-pie-addons.patch?content-type=text%2Fplain&view=co

Errata for FSF mailing list archives.


Regards,
Jan


--- a/gdb/dwarf2read.c
+++ b/gdb/dwarf2read.c
@@ -5754,7 +5754,12 @@ read_common_block (struct die_info *die, struct dwarf2_cu *cu)
 {
   struct attribute *attr;
   struct symbol *sym;
-  CORE_ADDR base = (CORE_ADDR) 0;
+  struct objfile *objfile = cu->objfile;
+  CORE_ADDR baseaddr = ANOFFSET (objfile->section_offsets,
+				 SECT_OFF_TEXT (objfile));
+  /* This is used only for DW_AT_data_member_location entries.  */
+  CORE_ADDR base = 0;
+  int base_p = 0;
 
   attr = dwarf2_attr (die, DW_AT_location, cu);
   if (attr)
@@ -5763,6 +5768,7 @@ read_common_block (struct die_info *die, struct dwarf2_cu *cu)
       if (attr_form_is_block (attr))
 	{
 	  base = decode_locdesc (DW_BLOCK (attr), cu);
+	  base_p = 1;
 	}
       else if (attr_form_is_section_offset (attr))
 	{
@@ -5824,12 +5830,15 @@ read_common_block (struct die_info *die, struct dwarf2_cu *cu)
 	      else
 		dwarf2_complex_location_expr_complaint ();
 
-	      SYMBOL_VALUE_ADDRESS (sym) = base + byte_offset;
+	      if (!base_p)
+		dwarf2_invalid_attrib_class_complaint
+		  ("DW_AT_data_member_location", "common block member");
+	      SYMBOL_VALUE_ADDRESS (sym) = base + byte_offset + baseaddr;
 	      add_symbol_to_list (sym, &global_symbols);
 	    }
 
 	  if (SYMBOL_CLASS (sym) == LOC_STATIC)
-	    SET_FIELD_PHYSADDR (*field, SYMBOL_VALUE_ADDRESS (sym));
+	    SET_FIELD_PHYSADDR (*field, SYMBOL_VALUE_ADDRESS (sym) - baseaddr);
 	  else
 	    SET_FIELD_PHYSNAME (*field, SYMBOL_LINKAGE_NAME (sym));
 	  FIELD_TYPE (*field) = SYMBOL_TYPE (sym);
@@ -5843,7 +5852,7 @@ read_common_block (struct die_info *die, struct dwarf2_cu *cu)
 
       sym = new_symbol (die, type, cu);
       /* SYMBOL_VALUE_ADDRESS never gets used as all its fields are static.  */
-      SYMBOL_VALUE_ADDRESS (sym) = base;
+      SYMBOL_VALUE_ADDRESS (sym) = base + baseaddr;
 
       set_die_type (die, type, cu);
     }
--- a/gdb/exec.c
+++ b/gdb/exec.c
@@ -32,6 +32,7 @@
 #include "exec.h"
 #include "observer.h"
 #include "arch-utils.h"
+#include "solib.h"
 
 #include <fcntl.h>
 #include "readline/readline.h"
@@ -220,6 +221,10 @@ exec_file_attach (char *filename, int from_tty)
       char *scratch_pathname;
       int scratch_chan;
       struct target_section *sections = NULL, *sections_end = NULL;
+      struct target_section *p;
+      int addr_bit;
+      CORE_ADDR mask = CORE_ADDR_MAX;
+      CORE_ADDR displacement;
 
       scratch_chan = openp (getenv ("PATH"), OPF_TRY_CWD_FIRST, filename,
 		   write_files ? O_RDWR | O_BINARY : O_RDONLY | O_BINARY,
@@ -288,12 +293,23 @@ exec_file_attach (char *filename, int from_tty)
 		 scratch_pathname, bfd_errmsg (bfd_get_error ()));
 	}
 
+      set_gdbarch_from_file (exec_bfd);
+
+      addr_bit = gdbarch_addr_bit (target_gdbarch);
+      if (addr_bit < (sizeof (CORE_ADDR) * HOST_CHAR_BIT))
+	mask = ((CORE_ADDR) 1 << addr_bit) - 1;
+
+      displacement = solib_exec_displacement ();
+      for (p = sections; p < sections_end; p++)
+	{
+	  p->addr = (p->addr + displacement) & mask;
+	  p->endaddr = (p->endaddr + displacement) & mask;
+	}
+
       exec_bfd_mtime = bfd_get_mtime (exec_bfd);
 
       validate_files ();
 
-      set_gdbarch_from_file (exec_bfd);
-
       /* Add the executable's sections to the current address spaces'
 	 list of sections.  */
       add_target_sections (sections, sections_end);
--- a/gdb/gdbtypes.h
+++ b/gdb/gdbtypes.h
@@ -966,6 +966,7 @@ extern void allocate_cplus_struct_type (struct type *);
 #define FIELD_LOC_KIND(thisfld) ((thisfld).loc_kind)
 #define FIELD_BITPOS(thisfld) ((thisfld).loc.bitpos)
 #define FIELD_STATIC_PHYSNAME(thisfld) ((thisfld).loc.physname)
+/* This address is unrelocated by the objfile's ANOFFSET.  */
 #define FIELD_STATIC_PHYSADDR(thisfld) ((thisfld).loc.physaddr)
 #define FIELD_DWARF_BLOCK(thisfld) ((thisfld).loc.dwarf_block)
 #define SET_FIELD_BITPOS(thisfld, bitpos)			\
@@ -974,6 +975,7 @@ extern void allocate_cplus_struct_type (struct type *);
 #define SET_FIELD_PHYSNAME(thisfld, name)			\
   (FIELD_LOC_KIND (thisfld) = FIELD_LOC_KIND_PHYSNAME,		\
    FIELD_STATIC_PHYSNAME (thisfld) = (name))
+/* This address is unrelocated by the objfile's ANOFFSET.  */
 #define SET_FIELD_PHYSADDR(thisfld, addr)			\
   (FIELD_LOC_KIND (thisfld) = FIELD_LOC_KIND_PHYSADDR,		\
    FIELD_STATIC_PHYSADDR (thisfld) = (addr))
@@ -989,6 +991,7 @@ extern void allocate_cplus_struct_type (struct type *);
 #define TYPE_FIELD_LOC_KIND(thistype, n) FIELD_LOC_KIND (TYPE_FIELD (thistype, n))
 #define TYPE_FIELD_BITPOS(thistype, n) FIELD_BITPOS (TYPE_FIELD (thistype, n))
 #define TYPE_FIELD_STATIC_PHYSNAME(thistype, n) FIELD_STATIC_PHYSNAME (TYPE_FIELD (thistype, n))
+/* This address is unrelocated by the objfile's ANOFFSET.  */
 #define TYPE_FIELD_STATIC_PHYSADDR(thistype, n) FIELD_STATIC_PHYSADDR (TYPE_FIELD (thistype, n))
 #define TYPE_FIELD_DWARF_BLOCK(thistype, n) FIELD_DWARF_BLOCK (TYPE_FIELD (thistype, n))
 #define TYPE_FIELD_ARTIFICIAL(thistype, n) FIELD_ARTIFICIAL(TYPE_FIELD(thistype,n))
--- a/gdb/jv-lang.c
+++ b/gdb/jv-lang.c
@@ -416,7 +416,8 @@ java_link_class_type (struct gdbarch *gdbarch,
 
   fields = NULL;
   nfields--;			/* First set up dummy "class" field. */
-  SET_FIELD_PHYSADDR (TYPE_FIELD (type, nfields), value_address (clas));
+  SET_FIELD_PHYSADDR (TYPE_FIELD (type, nfields), value_address (clas)
+    - (TYPE_OBJFILE (type) == NULL ? 0 : ANOFFSET (TYPE_OBJFILE (type)->section_offsets, SECT_OFF_TEXT (TYPE_OBJFILE (type)))));
   TYPE_FIELD_NAME (type, nfields) = "class";
   TYPE_FIELD_TYPE (type, nfields) = value_type (clas);
   SET_TYPE_FIELD_PRIVATE (type, nfields);
@@ -462,7 +463,8 @@ java_link_class_type (struct gdbarch *gdbarch,
 	  SET_TYPE_FIELD_PROTECTED (type, i);
 	}
       if (accflags & 0x0008)	/* ACC_STATIC */
-	SET_FIELD_PHYSADDR (TYPE_FIELD (type, i), boffset);
+	SET_FIELD_PHYSADDR (TYPE_FIELD (type, i), boffset
+	  - (TYPE_OBJFILE (type) == NULL ? 0 : ANOFFSET (TYPE_OBJFILE (type)->section_offsets, SECT_OFF_TEXT (TYPE_OBJFILE (type)))));
       else
 	TYPE_FIELD_BITPOS (type, i) = 8 * boffset;
       if (accflags & 0x8000)	/* FIELD_UNRESOLVED_FLAG */
--- a/gdb/solib-svr4.c
+++ b/gdb/solib-svr4.c
@@ -1621,7 +1621,10 @@ svr4_exec_displacement (void)
   if (target_auxv_search (&current_target, AT_ENTRY, &entry_point) == 1)
     return entry_point - exec_entry_point (exec_bfd, &current_target);
 
-  return svr4_static_exec_displacement ();
+  if (!ptid_equal (inferior_ptid, null_ptid))
+    return svr4_static_exec_displacement ();
+
+  return 0;
 }
 
 /* Relocate the main executable.  This function should be called upon
@@ -1632,7 +1635,7 @@ svr4_exec_displacement (void)
 static void
 svr4_relocate_main_executable (void)
 {
-  CORE_ADDR displacement = svr4_exec_displacement ();
+  CORE_ADDR displacement = solib_exec_displacement ();
 
   /* Even if DISPLACEMENT is 0 still try to relocate it as this is a new
      difference of in-memory vs. in-file addresses and we could already
@@ -1975,6 +1978,7 @@ _initialize_svr4_solib (void)
   svr4_so_ops.free_so = svr4_free_so;
   svr4_so_ops.clear_solib = svr4_clear_solib;
   svr4_so_ops.solib_create_inferior_hook = svr4_solib_create_inferior_hook;
+  svr4_so_ops.exec_displacement = svr4_exec_displacement;
   svr4_so_ops.special_symbol_handling = svr4_special_symbol_handling;
   svr4_so_ops.current_sos = svr4_current_sos;
   svr4_so_ops.open_symbol_file_object = open_symbol_file_object;
--- a/gdb/solib.c
+++ b/gdb/solib.c
@@ -1014,6 +1014,19 @@ solib_create_inferior_hook (int from_tty)
   ops->solib_create_inferior_hook (from_tty);
 }
 
+/* Query the difference of in-memory VMA addresses vs. exec_bfd VMAs.  */
+
+CORE_ADDR
+solib_exec_displacement (void)
+{
+  struct target_so_ops *ops = solib_ops (target_gdbarch);
+
+  if (ops->exec_displacement != NULL)
+    return (*ops->exec_displacement) ();
+  else
+    return 0;
+}
+
 /* GLOBAL FUNCTION
 
    in_solib_dynsym_resolve_code -- check to see if an address is in
--- a/gdb/solib.h
+++ b/gdb/solib.h
@@ -43,6 +43,8 @@ extern int solib_read_symbols (struct so_list *, int);
 
 extern void solib_create_inferior_hook (int from_tty);
 
+extern CORE_ADDR solib_exec_displacement (void);
+
 /* If ADDR lies in a shared library, return its name.  */
 
 extern char *solib_name_from_address (CORE_ADDR);
--- a/gdb/solist.h
+++ b/gdb/solist.h
@@ -89,6 +89,9 @@ struct target_so_ops
     /* Target dependent code to run after child process fork.  */
     void (*solib_create_inferior_hook) (int from_tty);
 
+    /* Query the difference of in-memory VMA addresses vs. exec_bfd VMAs.  */
+    CORE_ADDR (*exec_displacement) (void);
+
     /* Do additional symbol handling, lookup, etc. after symbols
        for a shared object have been loaded.  */
     void (*special_symbol_handling) (void);
--- a/gdb/symfile.c
+++ b/gdb/symfile.c
@@ -873,15 +873,36 @@ syms_from_objfile (struct objfile *objfile,
      if an error occurs during symbol reading.  */
   old_chain = make_cleanup_free_objfile (objfile);
 
-  /* If ADDRS and OFFSETS are both NULL, put together a dummy address
-     list.  We now establish the convention that an addr of zero means
-     no load address was specified. */
+  /* If ADDRS and OFFSETS are both NULL, put together a dummy offset list.  */
+
   if (! addrs && ! offsets)
     {
-      local_addr
-	= alloc_section_addr_info (bfd_count_sections (objfile->obfd));
-      make_cleanup (xfree, local_addr);
-      addrs = local_addr;
+      /* Relocateble files have an exception in default_symfile_offsets which
+	 applies only for ADDRS.  But calling solib_exec_displacement is more
+	 suitable for OFFSETS.  Fortunately we never need the both
+	 functionalities simultaneously and in other cases zeroed ADDRS and
+	 zeroed OFFSETS are equivalent.  */
+
+      if ((bfd_get_file_flags (objfile->obfd) & (EXEC_P | DYNAMIC)) == 0)
+	{
+	  local_addr
+		 = alloc_section_addr_info (bfd_count_sections (objfile->obfd));
+	  make_cleanup (xfree, local_addr);
+	  addrs = local_addr;
+	}
+      else
+	{
+	  CORE_ADDR displacement = 0;
+	  int i;
+
+	  if (mainline)
+	    displacement = solib_exec_displacement ();
+
+	  num_offsets = bfd_count_sections (objfile->obfd);
+	  offsets = alloca (SIZEOF_N_SECTION_OFFSETS (num_offsets));
+	  for (i = 0; i < num_offsets; i++)
+	    offsets->offsets[i] = displacement;
+	}
     }
 
   /* Now either addrs or offsets is non-zero.  */
--- a/gdb/value.c
+++ b/gdb/value.c
@@ -1890,7 +1890,8 @@ value_static_field (struct type *type, int fieldno)
   if (TYPE_FIELD_LOC_KIND (type, fieldno) == FIELD_LOC_KIND_PHYSADDR)
     {
       retval = value_at (TYPE_FIELD_TYPE (type, fieldno),
-			 TYPE_FIELD_STATIC_PHYSADDR (type, fieldno));
+			 TYPE_FIELD_STATIC_PHYSADDR (type, fieldno)
+			   + (TYPE_OBJFILE (type) == NULL ? 0 : ANOFFSET (TYPE_OBJFILE (type)->section_offsets, SECT_OFF_TEXT (TYPE_OBJFILE (type)))));
     }
   else
     {
@@ -1920,7 +1921,8 @@ value_static_field (struct type *type, int fieldno)
  	}
       if (retval && VALUE_LVAL (retval) == lval_memory)
 	SET_FIELD_PHYSADDR (TYPE_FIELD (type, fieldno),
-			    value_address (retval));
+			    value_address (retval)
+			      - (TYPE_OBJFILE (type) == NULL ? 0 : ANOFFSET (TYPE_OBJFILE (type)->section_offsets, SECT_OFF_TEXT (TYPE_OBJFILE (type)))));
     }
   return retval;
 }

Re: Red Hat PIE patch

[Michael Snyder]

Jan Kratochvil wrote:
> 
>> Just FYI, I've applied the patches and done basic "aliveness"
>> testing.  They do seem to work for me, and I will be doing more
>> with them so I may have further testing words later.
> 
> Thanks, BTW there are now some add-ons to that series which are still not
> properly split/integrated/reposted.
> 	http://cvs.fedora.redhat.com/viewvc/rpms/gdb/F-12/gdb-archer-pie-addons.patch?content-type=text%2Fplain&view=co


Out of pure self-interest, I would love to see this work
finalized and merged / checked in.  I suddenly have a user
base who depend heavily on PIE.

Michael (mmmmmmm.... PIE good....)