Be lenient when reading CodeView records of unknown types (#110). r=bryner

 - Read (but don't use) CodeView records with unknown signature values
   instead of failing.  This allows the module list to be read in, and
   will result in better stack traces in affected dumps.

http://groups.google.com/group/airbag-dev/browse_thread/thread/2c7d3e3b1fd6ea96


git-svn-id: http://google-breakpad.googlecode.com/svn/trunk@104 4c0a9323-5329-0410-9bdc-e9ce6186880e

2 files changed, 38 insertions, 9 deletions

diff --git a/src/google_airbag/common/minidump_format.h b/src/google_airbag/common/minidump_format.h
index 977cbc3c..535a6626 100644
--- a/src/google_airbag/common/minidump_format.h
+++ b/src/google_airbag/common/minidump_format.h
@@ -590,6 +590,23 @@ typedef struct {
 
 #define MD_CVINFOPDB70_SIGNATURE 0x53445352  /* cvSignature = 'SDSR' */
 
+/* In addition to the two CodeView record formats above, used for linking
+ * to external pdb files, it is possible for debugging data to be carried
+ * directly in the CodeView record itself.  These signature values will
+ * be found in the first 4 bytes of the CodeView record.  Additional values
+ * not commonly experienced in the wild are given by "Microsoft Symbol and
+ * Type Information", http://www.x86.org/ftp/manuals/tools/sym.pdf, section
+ * 7.2.  An in-depth description of the CodeView 4.1 format is given by
+ * "Undocumented Windows 2000 Secrets", Windows 2000 Debugging Support/
+ * Microsoft Symbol File Internals/CodeView Subsections,
+ * http://www.rawol.com/features/undocumented/sbs-w2k-1-windows-2000-debugging-support.pdf
+ */
+#define MD_CVINFOCV41_SIGNATURE 0x3930424e  /* '90BN', CodeView 4.10. */
+#define MD_CVINFOCV50_SIGNATURE 0x3131424e  /* '11BN', CodeView 5.0,
+                                             * MS C7-format (/Z7). */
+
+#define MD_CVINFOUNKNOWN_SIGNATURE 0xffffffff  /* An unlikely value. */
+
 /* (MDRawModule).miscRecord can reference MDImageDebugMisc.  The Windows
  * structure is actually defined in WinNT.h.  This structure is effectively
  * obsolete with modules built by recent toolchains. */
diff --git a/src/google_airbag/processor/minidump.h b/src/google_airbag/processor/minidump.h
index f2700119..6aa87c21 100644
--- a/src/google_airbag/processor/minidump.h
+++ b/src/google_airbag/processor/minidump.h
@@ -369,15 +369,22 @@ class MinidumpModule : public MinidumpObject,
 
   // The CodeView record, which contains information to locate the module's
   // debugging information (pdb).  This is returned as u_int8_t* because
-  // the data can be one of two types: MDCVInfoPDB20* or MDCVInfoPDB70*.
-  // Check the record's signature in the first four bytes to differentiate.
-  // Current toolchains generate modules which carry MDCVInfoPDB70.
-  const u_int8_t* GetCVRecord();
+  // the data can be of types MDCVInfoPDB20* or MDCVInfoPDB70*, or it may be
+  // of a type unknown to Airbag, in which case the raw data will still be
+  // returned but no byte-swapping will have been performed.  Check the
+  // record's signature in the first four bytes to differentiate between
+  // the various types.  Current toolchains generate modules which carry
+  // MDCVInfoPDB70 by default.  Returns a pointer to the CodeView record on
+  // success, and NULL on failure.  On success, the optional |size| argument
+  // is set to the size of the CodeView record.
+  const u_int8_t* GetCVRecord(u_int32_t* size);
 
   // The miscellaneous debug record, which is obsolete.  Current toolchains
   // do not generate this type of debugging information (dbg), and this
-  // field is not expected to be present.
-  const MDImageDebugMisc* GetMiscRecord();
+  // field is not expected to be present.  Returns a pointer to the debugging
+  // record on success, and NULL on failure.  On success, the optional |size|
+  // argument is set to the size of the debugging record.
+  const MDImageDebugMisc* GetMiscRecord(u_int32_t* size);
 
   // Print a human-readable representation of the object to stdout.
   void Print();
@@ -414,11 +421,16 @@ class MinidumpModule : public MinidumpObject,
   const string*     name_;
 
   // Cached CodeView record - this is MDCVInfoPDB20 or (likely)
-  // MDCVInfoPDB70.  Stored as a u_int8_t because the structure contains
-  // a variable-sized string and its exact size cannot be known until it
-  // is processed.
+  // MDCVInfoPDB70, or possibly something else entirely.  Stored as a u_int8_t
+  // because the structure contains a variable-sized string and its exact
+  // size cannot be known until it is processed.
   vector<u_int8_t>* cv_record_;
 
+  // If cv_record_ is present, cv_record_signature_ contains a copy of the
+  // CodeView record's first four bytes, for ease of determinining the
+  // type of structure that cv_record_ contains.
+  u_int32_t cv_record_signature_;
+
   // Cached MDImageDebugMisc (usually not present), stored as u_int8_t
   // because the structure contains a variable-sized string and its exact
   // size cannot be known until it is processed.