[PATCH 2/2 v2] Generalize cu_sec_idx

[PATCH 2/2 v2] Generalize cu_sec_idx

[Ulf Hermann]

Apparently CUs can appear in other sections than IDX_debug_info and
IDX_types. Rather than relying on the indirect indication provided by
type_offset we compare the addresses directly to figure out which section
a given CU belongs to.

This fixes the dwarf-getmacros test.

(Signed-off instead of Change-Id ...)

Signed-off-by: Ulf Hermann <ulf.hermann@qt.io>
---
  libdw/ChangeLog |  4 ++++
  libdw/libdwP.h  | 12 +++++++++++-
  2 files changed, 15 insertions(+), 1 deletion(-)

diff --git a/libdw/ChangeLog b/libdw/ChangeLog
index 996cd2e..508bf9c 100644
--- a/libdw/ChangeLog
+++ b/libdw/ChangeLog
@@ -1,3 +1,7 @@
+2017-12-08  Ulf Hermann  <ulf.hermann@qt.io>
+
+	* libdwP.h: Generalize cu_sec_idx to check all sections.
+
  2017-05-09  Ulf Hermann  <ulf.hermann@qt.io>
   	* libdwP.h: Fix check for the upper border of the range in 
__libdw_in_section.
diff --git a/libdw/libdwP.h b/libdw/libdwP.h
index e092d8e..8f3a95c 100644
--- a/libdw/libdwP.h
+++ b/libdw/libdwP.h
@@ -715,7 +715,17 @@ __libdw_read_offset (Dwarf *dbg, Dwarf *dbg_ret,
  static inline size_t
  cu_sec_idx (struct Dwarf_CU *cu)
  {
-  return cu->type_offset == 0 ? IDX_debug_info : IDX_debug_types;
+  for (int sec_index = IDX_debug_info; sec_index < IDX_last; ++sec_index)
+    {
+      Elf_Data *data = cu->dbg->sectiondata[sec_index];
+      if (data != NULL && data->d_buf != NULL
+          && cu->startp >= data->d_buf
+          && cu->startp < data->d_buf + data->d_size)
+        {
+          return sec_index;
+        }
+    }
+  return IDX_last;
  }
   static inline bool
-- 
2.8.1.windows.1

Re: [PATCH 2/2 v2] Generalize cu_sec_idx

[Mark Wielaard]

On Fri, 2017-12-08 at 16:06 +0100, Ulf Hermann wrote:
> Apparently CUs can appear in other sections than IDX_debug_info and
> IDX_types. Rather than relying on the indirect indication provided by
> type_offset we compare the addresses directly to figure out which section
> a given CU belongs to.

This is clever and indeed cu_sec_idx () is not generic enough.
But this is also somewhat inefficient. I am working on DWARF5 support
and there a CU can come from even more different sections (or file). So
I am changing Dwarf_CU to have an explicit section to which is it is
associated. This can then also be used by the "fake" CUs like created
in dwarf_getmacros.

Thanks,

Mark

Re: [PATCH 2/2 v2] Generalize cu_sec_idx

[Ulf Hermann]

On 12/14/2017 02:51 PM, Mark Wielaard wrote:
> This is clever and indeed cu_sec_idx () is not generic enough.
> But this is also somewhat inefficient. I am working on DWARF5 support
> and there a CU can come from even more different sections (or file). So
> I am changing Dwarf_CU to have an explicit section to which is it is
> associated. This can then also be used by the "fake" CUs like created
> in dwarf_getmacros.

Mind that the two most common cases are 0 and 1. In fact nothing else was supported before this change. So, most of the time this will not do a lot of iteration.

regards,
Ulf

Re: [PATCH 2/2 v2] Generalize cu_sec_idx

[Mark Wielaard]

[-- Attachment #1: Type: text/plain, Size: 987 bytes --]

On Thu, 2017-12-14 at 14:53 +0100, Ulf Hermann wrote:
> On 12/14/2017 02:51 PM, Mark Wielaard wrote:
> > This is clever and indeed cu_sec_idx () is not generic enough.
> > But this is also somewhat inefficient. I am working on DWARF5 support
> > and there a CU can come from even more different sections (or file). So
> > I am changing Dwarf_CU to have an explicit section to which is it is
> > associated. This can then also be used by the "fake" CUs like created
> > in dwarf_getmacros.
> 
> Mind that the two most common cases are 0 and 1. In fact nothing else
> was supported before this change. So, most of the time this will not
> do a lot of iteration.

yes, but the original code really was not correct. The attached patch
fixes it by adding an explicit sec_idx field to the Dwarf_CU struct
that is set whenever a struct Dwarf_CU is created, so that we never
have to guess.

This patch combined with the overflow fix makes all testcases PASS.

Cheers,

Mark

[-- Attachment #2: 0001-libdw-Add-explicit-section-index-to-struct-Dwarf_CU.patch --]
[-- Type: text/x-patch, Size: 4023 bytes --]

From 51a7292b7ec7ddebcd2abddc7efff9d604494d44 Mon Sep 17 00:00:00 2001
From: Mark Wielaard <mark@klomp.org>
Date: Wed, 20 Dec 2017 16:50:57 +0100
Subject: [PATCH 1/2] libdw: Add explicit section index to struct Dwarf_CU.

The DIE (attribute) data might come from either the main .debug_info
section or for DWARFv4 from a separate .debug_types section. Or in
case of the fake_loc_cu from the .debug_loc section and in the case
of macros from the .debug_macinfo or .debug_macro section.

We didn't handle the last two "fake" CU cases correctly when sanity
checking offsets in __libdw_read_address and __libdw_read_offset.

Add an explicit sec_idx field to struct Dwarf_CU that is always set
to the actual section that the data came from.

Signed-off-by: Mark Wielaard <mark@klomp.org>
---
 libdw/ChangeLog         | 11 +++++++++++
 libdw/dwarf_begin_elf.c |  1 +
 libdw/dwarf_getmacros.c |  1 +
 libdw/libdwP.h          |  4 +++-
 libdw/libdw_findcu.c    |  5 +++--
 5 files changed, 19 insertions(+), 3 deletions(-)

diff --git a/libdw/ChangeLog b/libdw/ChangeLog
index 350230e..22b7bf4 100644
--- a/libdw/ChangeLog
+++ b/libdw/ChangeLog
@@ -1,3 +1,14 @@
+2017-12-20  Mark Wielaard  <mark@klomp.org>
+
+	* libdwP.h (struct Dwarf_CU): Add sec_idx field.
+	(cu_sec_idx): Return cu->sec_idx.
+	* libdw_findcu.c (__libdw_intern_next_unit): Set cu sec_idx to
+	IDX_debug_info or IDX_debug_types.
+	* dwarf_begin_elf.c (valid_p): Set fake_loc_cu->sec_idx to
+	IDX_debug_loc.
+	* dwarf_getmacros.c (read_macros): Set fake_cu->sec_idx to
+	IDX_debug_macro or IDX_debug_macinfo.
+
 2017-12-12  Mark Wielaard  <mark@klomp.org>
 
 	* dwarf_aggregate_size.c (dwarf_aggregate_size): Don't peel the
diff --git a/libdw/dwarf_begin_elf.c b/libdw/dwarf_begin_elf.c
index afa15ce..7c3fe10 100644
--- a/libdw/dwarf_begin_elf.c
+++ b/libdw/dwarf_begin_elf.c
@@ -201,6 +201,7 @@ valid_p (Dwarf *result)
 	}
       else
 	{
+	  result->fake_loc_cu->sec_idx = IDX_debug_loc;
 	  result->fake_loc_cu->dbg = result;
 	  result->fake_loc_cu->startp
 	    = result->sectiondata[IDX_debug_loc]->d_buf;
diff --git a/libdw/dwarf_getmacros.c b/libdw/dwarf_getmacros.c
index db6582b..c456051 100644
--- a/libdw/dwarf_getmacros.c
+++ b/libdw/dwarf_getmacros.c
@@ -360,6 +360,7 @@ read_macros (Dwarf *dbg, int sec_index,
 	 Version 4 for the old GNU extension, version 5 for DWARF5.  */
       Dwarf_CU fake_cu = {
 	.dbg = dbg,
+	.sec_idx = sec_index,
 	.version = table->version,
 	.offset_size = table->is_64bit ? 8 : 4,
 	.startp = (void *) startp + offset,
diff --git a/libdw/libdwP.h b/libdw/libdwP.h
index 78c0013..f524347 100644
--- a/libdw/libdwP.h
+++ b/libdw/libdwP.h
@@ -293,6 +293,8 @@ struct Dwarf_CU
   uint8_t offset_size;
   uint16_t version;
 
+  size_t sec_idx; /* Normally .debug_info, could be .debug_type or "fake". */
+
   /* Zero if this is a normal CU.  Nonzero if it is a type unit.  */
   size_t type_offset;
   uint64_t type_sig8;
@@ -714,7 +716,7 @@ __libdw_read_offset (Dwarf *dbg, Dwarf *dbg_ret,
 static inline size_t
 cu_sec_idx (struct Dwarf_CU *cu)
 {
-  return cu->type_offset == 0 ? IDX_debug_info : IDX_debug_types;
+  return cu->sec_idx;
 }
 
 static inline bool
diff --git a/libdw/libdw_findcu.c b/libdw/libdw_findcu.c
index 082307b..4e025e2 100644
--- a/libdw/libdw_findcu.c
+++ b/libdw/libdw_findcu.c
@@ -93,8 +93,8 @@ __libdw_intern_next_unit (Dwarf *dbg, bool debug_types)
     }
 
   /* Invalid or truncated debug section data?  */
-  Elf_Data *data = dbg->sectiondata[debug_types
-				    ? IDX_debug_types : IDX_debug_info];
+  size_t sec_idx = debug_types ? IDX_debug_types : IDX_debug_info;
+  Elf_Data *data = dbg->sectiondata[sec_idx];
   if (unlikely (*offsetp > data->d_size))
     *offsetp = data->d_size;
 
@@ -102,6 +102,7 @@ __libdw_intern_next_unit (Dwarf *dbg, bool debug_types)
   struct Dwarf_CU *newp = libdw_typed_alloc (dbg, struct Dwarf_CU);
 
   newp->dbg = dbg;
+  newp->sec_idx = sec_idx;
   newp->start = oldoff;
   newp->end = *offsetp;
   newp->address_size = address_size;
-- 
1.8.3.1

Re: [PATCH 2/2 v2] Generalize cu_sec_idx

[Ulf Hermann]

> yes, but the original code really was not correct. The attached patch
> fixes it by adding an explicit sec_idx field to the Dwarf_CU struct
> that is set whenever a struct Dwarf_CU is created, so that we never
> have to guess.
>
> This patch combined with the overflow fix makes all testcases PASS.

Looks good to me.

Ulf

Re: [PATCH 2/2 v2] Generalize cu_sec_idx

[Mark Wielaard]

On Thu, Dec 21, 2017 at 02:48:15PM +0100, Ulf Hermann wrote:
> > yes, but the original code really was not correct. The attached patch
> > fixes it by adding an explicit sec_idx field to the Dwarf_CU struct
> > that is set whenever a struct Dwarf_CU is created, so that we never
> > have to guess.
> > 
> > This patch combined with the overflow fix makes all testcases PASS.
> 
> Looks good to me.

Thanks again. Pushed to master.