libdw: Handle .debug_loclists in dwarf_getlocation.

Handle all new DW_LLE opcodes in .debug_loclists in dwarf_getlocation.
__libdw_read_begin_end_pair_inc now also handles a default location
(which is simply the range [0,-1]). Since expression blocks can now
also come from the .debug_loclists section add a new fake_loclists_cu
necessary for checking bounds while parsing expression blocks.

Adapt varlocs test to handle debug-only files.
Test testfileranges5.debug and testfilesplitranges5.debug with it.

Signed-off-by: Mark Wielaard <mark@klomp.org>

8 files changed, 340 insertions, 16 deletions

diff --git a/libdw/ChangeLog b/libdw/ChangeLog
index 0db49bf8..22712f18 100644
--- a/libdw/ChangeLog
+++ b/libdw/ChangeLog
@@ -1,3 +1,20 @@
+2018-04-07  Mark Wielaard  <mark@klomp.org>
+
+	* libdwP.h (struct Dwarf_CU): Add locs_base.
+	(__libdw_cu_locs_base): New static inline function.
+	* libdw_findcu.c (__libdw_intern_next_unit): Initialize locs_base.
+	* dwarf_begin_elf.c (valid_p): Create fake_loclists_cu if necessary.
+	* dwarf_end.c (dwarf_end): Clean up fake_loclists_cu.
+	* dwarf_getlocation.c (initial_offset): Handle .debug_loclists.
+	(getlocations_addr): Likewise.
+	(dwarf_getlocation_addr): Likewise.
+	* dwarf_getlocation_attr.c (attr_form_cu): Use fake_loclists_cu for
+	DWARF5.
+	(initial_offset): Handle DW_FORM_loclistx.
+	* dwarf_ranges.c (__libdw_read_begin_end_pair_inc): Handle
+	.debug_loclists.
+	* libdwP.h (struct Dwarf): Add fake_loclists_cu.
+
 2018-04-12  Mark Wielaard  <mark@klomp.org>
 
 	* dwarf.h: Add DWARF5 location list entry DW_LLE encodings.
diff --git a/libdw/dwarf_begin_elf.c b/libdw/dwarf_begin_elf.c
index af5096fb..513af2b1 100644
--- a/libdw/dwarf_begin_elf.c
+++ b/libdw/dwarf_begin_elf.c
@@ -226,6 +226,9 @@ valid_p (Dwarf *result)
       result = NULL;
     }
 
+  /* For dwarf_location_attr () we need a "fake" CU to indicate
+     where the "fake" attribute data comes from.  This is a block
+     inside the .debug_loc or .debug_loclists section.  */
   if (result != NULL && result->sectiondata[IDX_debug_loc] != NULL)
     {
       result->fake_loc_cu = (Dwarf_CU *) calloc (1, sizeof (Dwarf_CU));
@@ -248,6 +251,29 @@ valid_p (Dwarf *result)
 	}
     }
 
+  if (result != NULL && result->sectiondata[IDX_debug_loclists] != NULL)
+    {
+      result->fake_loclists_cu = (Dwarf_CU *) calloc (1, sizeof (Dwarf_CU));
+      if (unlikely (result->fake_loclists_cu == NULL))
+	{
+	  Dwarf_Sig8_Hash_free (&result->sig8_hash);
+	  __libdw_seterrno (DWARF_E_NOMEM);
+	  free (result->fake_loc_cu);
+	  free (result);
+	  result = NULL;
+	}
+      else
+	{
+	  result->fake_loclists_cu->sec_idx = IDX_debug_loclists;
+	  result->fake_loclists_cu->dbg = result;
+	  result->fake_loclists_cu->startp
+	    = result->sectiondata[IDX_debug_loclists]->d_buf;
+	  result->fake_loclists_cu->endp
+	    = (result->sectiondata[IDX_debug_loclists]->d_buf
+	       + result->sectiondata[IDX_debug_loclists]->d_size);
+	}
+    }
+
   /* For DW_OP_constx/GNU_const_index and DW_OP_addrx/GNU_addr_index
      the dwarf_location_attr () will need a "fake" address CU to
      indicate where the attribute data comes from.  This is a just
@@ -260,6 +286,7 @@ valid_p (Dwarf *result)
 	  Dwarf_Sig8_Hash_free (&result->sig8_hash);
 	  __libdw_seterrno (DWARF_E_NOMEM);
 	  free (result->fake_loc_cu);
+	  free (result->fake_loclists_cu);
 	  free (result);
 	  result = NULL;
 	}
diff --git a/libdw/dwarf_end.c b/libdw/dwarf_end.c
index 19546741..23a50a0b 100644
--- a/libdw/dwarf_end.c
+++ b/libdw/dwarf_end.c
@@ -113,6 +113,11 @@ dwarf_end (Dwarf *dwarf)
 	  cu_free (dwarf->fake_loc_cu);
 	  free (dwarf->fake_loc_cu);
 	}
+      if (dwarf->fake_loclists_cu != NULL)
+	{
+	  cu_free (dwarf->fake_loclists_cu);
+	  free (dwarf->fake_loclists_cu);
+	}
       if (dwarf->fake_addr_cu != NULL)
 	{
 	  cu_free (dwarf->fake_addr_cu);
diff --git a/libdw/dwarf_getlocation.c b/libdw/dwarf_getlocation.c
index d4b8effe..d293e75d 100644
--- a/libdw/dwarf_getlocation.c
+++ b/libdw/dwarf_getlocation.c
@@ -696,13 +696,77 @@ __libdw_cu_base_address (Dwarf_CU *cu)
 static int
 initial_offset (Dwarf_Attribute *attr, ptrdiff_t *offset)
 {
-  size_t secidx = IDX_debug_loc;
+  size_t secidx = (attr->cu->version < 5
+		   ? IDX_debug_loc : IDX_debug_loclists);
 
   Dwarf_Word start_offset;
-  if (__libdw_formptr (attr, secidx,
-		       DWARF_E_NO_DEBUG_LOC,
-		       NULL, &start_offset) == NULL)
-    return -1;
+  if (attr->form == DW_FORM_loclistx)
+    {
+      Dwarf_Word idx;
+      Dwarf_CU *cu = attr->cu;
+      const unsigned char *datap = attr->valp;
+      const unsigned char *endp = cu->endp;
+      if (datap >= endp)
+	{
+	  __libdw_seterrno (DWARF_E_INVALID_DWARF);
+	  return -1;
+	}
+      get_uleb128 (idx, datap, endp);
+
+      Elf_Data *data = cu->dbg->sectiondata[secidx];
+      if (data == NULL && cu->unit_type == DW_UT_split_compile)
+	{
+	  cu = __libdw_find_split_unit (cu);
+	  if (cu != NULL)
+	    data = cu->dbg->sectiondata[secidx];
+	}
+
+      if (data == NULL)
+	{
+	  __libdw_seterrno (secidx == IDX_debug_loc
+                            ? DWARF_E_NO_DEBUG_LOC
+                            : DWARF_E_NO_DEBUG_LOCLISTS);
+	  return -1;
+	}
+
+      Dwarf_Off loc_base_off = __libdw_cu_locs_base (cu);
+
+      /* The section should at least contain room for one offset.  */
+      size_t sec_size = cu->dbg->sectiondata[secidx]->d_size;
+      size_t offset_size = cu->offset_size;
+      if (offset_size > sec_size)
+	{
+	invalid_offset:
+	  __libdw_seterrno (DWARF_E_INVALID_OFFSET);
+	  return -1;
+	}
+
+      /* And the base offset should be at least inside the section.  */
+      if (loc_base_off > (sec_size - offset_size))
+	goto invalid_offset;
+
+      size_t max_idx = (sec_size - offset_size - loc_base_off) / offset_size;
+      if (idx > max_idx)
+	goto invalid_offset;
+
+      datap = (cu->dbg->sectiondata[secidx]->d_buf
+	       + loc_base_off + (idx * offset_size));
+      if (offset_size == 4)
+	start_offset = read_4ubyte_unaligned (cu->dbg, datap);
+      else
+	start_offset = read_8ubyte_unaligned (cu->dbg, datap);
+
+      start_offset += loc_base_off;
+    }
+  else
+    {
+      if (__libdw_formptr (attr, secidx,
+			   (secidx == IDX_debug_loc
+			    ? DWARF_E_NO_DEBUG_LOC
+			    : DWARF_E_NO_DEBUG_LOCLISTS),
+			    NULL, &start_offset) == NULL)
+	return -1;
+    }
 
   *offset = start_offset;
   return 0;
@@ -716,7 +780,7 @@ getlocations_addr (Dwarf_Attribute *attr, ptrdiff_t offset,
 {
   Dwarf_CU *cu = attr->cu;
   Dwarf *dbg = cu->dbg;
-  size_t secidx = IDX_debug_loc;
+  size_t secidx = cu->version < 5 ? IDX_debug_loc : IDX_debug_loclists;
   const unsigned char *readp = locs->d_buf + offset;
   const unsigned char *readendp = locs->d_buf + locs->d_size;
 
@@ -741,13 +805,22 @@ getlocations_addr (Dwarf_Attribute *attr, ptrdiff_t offset,
 
   /* We have a location expression.  */
   Dwarf_Block block;
-  if (readendp - readp < 2)
+  if (secidx == IDX_debug_loc)
     {
-    invalid:
-      __libdw_seterrno (DWARF_E_INVALID_DWARF);
-      return -1;
+      if (readendp - readp < 2)
+	{
+	invalid:
+	  __libdw_seterrno (DWARF_E_INVALID_DWARF);
+	  return -1;
+	}
+      block.length = read_2ubyte_unaligned_inc (dbg, readp);
+    }
+  else
+    {
+      if (readendp - readp < 1)
+	goto invalid;
+      get_uleb128 (block.length, readp, readendp);
     }
-  block.length = read_2ubyte_unaligned_inc (dbg, readp);
   block.data = (unsigned char *) readp;
   if (readendp - readp < (ptrdiff_t) block.length)
     goto invalid;
@@ -815,7 +888,7 @@ dwarf_getlocation_addr (Dwarf_Attribute *attr, Dwarf_Addr address,
   if (initial_offset (attr, &off) != 0)
     return -1;
 
-  size_t secidx = IDX_debug_loc;
+  size_t secidx = attr->cu->version < 5 ? IDX_debug_loc : IDX_debug_loclists;
   const Elf_Data *d = attr->cu->dbg->sectiondata[secidx];
 
   while (got < maxlocs
@@ -896,7 +969,7 @@ dwarf_getlocations (Dwarf_Attribute *attr, ptrdiff_t offset, Dwarf_Addr *basep,
 	return -1;
     }
 
-  size_t secidx = IDX_debug_loc;
+  size_t secidx = attr->cu->version < 5 ? IDX_debug_loc : IDX_debug_loclists;
   const Elf_Data *d = attr->cu->dbg->sectiondata[secidx];
 
   return getlocations_addr (attr, offset, basep, startp, endp,
diff --git a/libdw/dwarf_getlocation_attr.c b/libdw/dwarf_getlocation_attr.c
index 9d7fd4b5..875fc5d7 100644
--- a/libdw/dwarf_getlocation_attr.c
+++ b/libdw/dwarf_getlocation_attr.c
@@ -38,7 +38,7 @@ attr_form_cu (Dwarf_Attribute *attr)
 {
   /* If the attribute has block/expr form the data comes from the
      .debug_info from the same cu as the attr.  Otherwise it comes from
-     the .debug_loc data section.  */
+     the .debug_loc or .debug_loclists data section.  */
   switch (attr->form)
     {
     case DW_FORM_block1:
@@ -48,7 +48,9 @@ attr_form_cu (Dwarf_Attribute *attr)
     case DW_FORM_exprloc:
       return attr->cu;
     default:
-      return attr->cu->dbg->fake_loc_cu;
+      return (attr->cu->version < 5
+	      ? attr->cu->dbg->fake_loc_cu
+	      : attr->cu->dbg->fake_loclists_cu);
     }
 }
 
diff --git a/libdw/dwarf_ranges.c b/libdw/dwarf_ranges.c
index fa65e5c6..0f3ee6b5 100644
--- a/libdw/dwarf_ranges.c
+++ b/libdw/dwarf_ranges.c
@@ -36,6 +36,7 @@
 
 /* Read up begin/end pair and increment read pointer.
     - If it's normal range record, set up `*beginp' and `*endp' and return 0.
+    - If it's a default location, set `*beginp' (0), `*endp' (-1) and return 0.
     - If it's base address selection record, set up `*basep' and return 1.
     - If it's end of rangelist, don't set anything and return 2
     - If an error occurs, don't set anything and return -1.  */
@@ -197,6 +198,119 @@ __libdw_read_begin_end_pair_inc (Dwarf_CU *cu, int sec_index,
 	  goto invalid;
 	}
     }
+  else if (sec_index == IDX_debug_loclists)
+    {
+      const unsigned char *addr = *addrp;
+      if (addrend - addr < 1)
+	goto invalid;
+
+      const char code = *addr++;
+      uint64_t begin = 0, end = 0, base = *basep, addr_idx;
+      switch (code)
+	{
+	case DW_LLE_end_of_list:
+	  *addrp = addr;
+	  return 2;
+
+	case DW_LLE_base_addressx:
+	  if (addrend - addr < 1)
+	    goto invalid;
+	  get_uleb128 (addr_idx, addr, addrend);
+	  if (__libdw_addrx (cu, addr_idx, &base) != 0)
+	    return -1;
+
+	  *basep = base;
+	  *addrp = addr;
+	  return 1;
+
+	case DW_LLE_startx_endx:
+	  if (addrend - addr < 1)
+	    goto invalid;
+	  get_uleb128 (addr_idx, addr, addrend);
+	  if (__libdw_addrx (cu, addr_idx, &begin) != 0)
+	    return -1;
+	  if (addrend - addr < 1)
+	    goto invalid;
+	  get_uleb128 (addr_idx, addr, addrend);
+	  if (__libdw_addrx (cu, addr_idx, &end) != 0)
+	    return -1;
+
+	  *beginp = begin;
+	  *endp = end;
+	  *addrp = addr;
+	  return 0;
+
+	case DW_LLE_startx_length:
+	  if (addrend - addr < 1)
+	    goto invalid;
+	  get_uleb128 (addr_idx, addr, addrend);
+	  if (__libdw_addrx (cu, addr_idx, &begin) != 0)
+	    return -1;
+	  if (addrend - addr < 1)
+	    goto invalid;
+	  get_uleb128 (end, addr, addrend);
+
+	  *beginp = begin;
+	  *endp = begin + end;
+	  *addrp = addr;
+	  return 0;
+
+	case DW_LLE_offset_pair:
+	  if (addrend - addr < 1)
+	    goto invalid;
+	  get_uleb128 (begin, addr, addrend);
+	  if (addrend - addr < 1)
+	    goto invalid;
+	  get_uleb128 (end, addr, addrend);
+
+	  *beginp = begin + base;
+	  *endp = end + base;
+	  *addrp = addr;
+	  return 0;
+
+	case DW_LLE_default_location:
+	  *beginp = 0;
+	  *endp = (Dwarf_Addr) -1;
+	  *addrp = addr;
+	  return 0;
+
+	case DW_LLE_base_address:
+	  if (addrend - addr < width)
+	    goto invalid;
+	  __libdw_read_address_inc (dbg, sec_index, &addr, width, &base);
+
+	  *basep = base;
+	  *addrp = addr;
+	  return 1;
+
+	case DW_LLE_start_end:
+	  if (addrend - addr < 2 * width)
+	    goto invalid;
+	  __libdw_read_address_inc (dbg, sec_index, &addr, width, &begin);
+	  __libdw_read_address_inc (dbg, sec_index, &addr, width, &end);
+
+	  *beginp = begin;
+	  *endp = end;
+	  *addrp = addr;
+	  return 0;
+
+	case DW_LLE_start_length:
+	  if (addrend - addr < width)
+	    goto invalid;
+	  __libdw_read_address_inc (dbg, sec_index, &addr, width, &begin);
+	  if (addrend - addr < 1)
+	    goto invalid;
+	  get_uleb128 (end, addr, addrend);
+
+	  *beginp = begin;
+	  *endp = begin + end;
+	  *addrp = addr;
+	  return 0;
+
+	default:
+	  goto invalid;
+	}
+    }
   else
     {
       __libdw_seterrno (DWARF_E_INVALID_DWARF);
diff --git a/libdw/libdwP.h b/libdw/libdwP.h
index f99ea584..dd47009b 100644
--- a/libdw/libdwP.h
+++ b/libdw/libdwP.h
@@ -206,8 +206,11 @@ struct Dwarf
   struct Dwarf_CFI_s *cfi;
 
   /* Fake loc CU.  Used when synthesizing attributes for Dwarf_Ops that
-     came from a location list entry in dwarf_getlocation_attr.  */
+     came from a location list entry in dwarf_getlocation_attr.
+     Depending on version this is the .debug_loc or .debug_loclists
+     section (could be both if mixing CUs with different DWARF versions).  */
   struct Dwarf_CU *fake_loc_cu;
+  struct Dwarf_CU *fake_loclists_cu;
 
   /* Similar for addrx/constx, which will come from .debug_addr section.  */
   struct Dwarf_CU *fake_addr_cu;
@@ -365,6 +368,10 @@ struct Dwarf_CU
      __libdw_cu_ranges_base.  */
   Dwarf_Off ranges_base;
 
+  /* The start of the offset table in .debug_loclists.
+     Don't access directly, call __libdw_cu_locs_base.  */
+  Dwarf_Off locs_base;
+
   /* Memory boundaries of this CU.  */
   void *startp;
   void *endp;
@@ -1177,6 +1184,84 @@ __libdw_cu_ranges_base (Dwarf_CU *cu)
 }
 
 
+/* The start of the offset table in .debug_loclists for DWARF5.  */
+static inline Dwarf_Off
+__libdw_cu_locs_base (Dwarf_CU *cu)
+{
+  if (cu->locs_base == (Dwarf_Off) -1)
+    {
+      Dwarf_Off offset = 0;
+      Dwarf_Die cu_die = CUDIE(cu);
+      Dwarf_Attribute attr;
+      if (dwarf_attr (&cu_die, DW_AT_loclists_base, &attr) != NULL)
+	{
+	  Dwarf_Word off;
+	  if (dwarf_formudata (&attr, &off) == 0)
+	    offset = off;
+	}
+
+      /* There wasn't an loclists_base, if the Dwarf does have a
+	 .debug_loclists section, then it might be we need the
+	 base after the first header. */
+      Elf_Data *data = cu->dbg->sectiondata[IDX_debug_loclists];
+      if (offset == 0 && data != NULL)
+	{
+	  Dwarf *dbg = cu->dbg;
+	  const unsigned char *readp = data->d_buf;
+	  const unsigned char *const dataend
+	    = (unsigned char *) data->d_buf + data->d_size;
+
+	  uint64_t unit_length = read_4ubyte_unaligned_inc (dbg, readp);
+	  unsigned int offset_size = 4;
+	  if (unlikely (unit_length == 0xffffffff))
+	    {
+	      if (unlikely (readp > dataend - 8))
+		goto no_header;
+
+	      unit_length = read_8ubyte_unaligned_inc (dbg, readp);
+	      offset_size = 8;
+	    }
+
+	  if (readp > dataend - 8
+	      || unit_length < 8
+	      || unit_length > (uint64_t) (dataend - readp))
+	    goto no_header;
+
+	  uint16_t version = read_2ubyte_unaligned_inc (dbg, readp);
+	  if (version != 5)
+	    goto no_header;
+
+	  uint8_t address_size = *readp++;
+	  if (address_size != 4 && address_size != 8)
+	    goto no_header;
+
+	  uint8_t segment_size = *readp++;
+	  if (segment_size != 0)
+	    goto no_header;
+
+	  uint32_t offset_entry_count;
+	  offset_entry_count = read_4ubyte_unaligned_inc (dbg, readp);
+
+	  const unsigned char *offset_array_start = readp;
+	  if (offset_entry_count <= 0)
+	    goto no_header;
+
+	  uint64_t needed = offset_entry_count * offset_size;
+	  if (unit_length - 8 < needed)
+	    goto no_header;
+
+	  offset = (Dwarf_Off) (offset_array_start
+				- (unsigned char *) data->d_buf);
+	}
+
+    no_header:
+      cu->locs_base = offset;
+    }
+
+  return cu->locs_base;
+}
+
+
 /* Link skeleton and split compile units.  */
 static inline void
 __libdw_link_skel_split (Dwarf_CU *skel, Dwarf_CU *split)
diff --git a/libdw/libdw_findcu.c b/libdw/libdw_findcu.c
index 83c2eb14..9d231999 100644
--- a/libdw/libdw_findcu.c
+++ b/libdw/libdw_findcu.c
@@ -122,6 +122,7 @@ __libdw_intern_next_unit (Dwarf *dbg, bool debug_types)
   newp->addr_base = (Dwarf_Off) -1;
   newp->str_off_base = (Dwarf_Off) -1;
   newp->ranges_base = (Dwarf_Off) -1;
+  newp->locs_base = (Dwarf_Off) -1;
 
   newp->startp = data->d_buf + newp->start;
   newp->endp = data->d_buf + newp->end;