Fix bug of issue 135, now can dump symbols for dynamic libraries.

Also some minor fixes to make it compile with more strict compiling options.



git-svn-id: http://google-breakpad.googlecode.com/svn/trunk@133 4c0a9323-5329-0410-9bdc-e9ce6186880e

src/client/linux/handler/exception_handler_test.cc

@@ -107,7 +107,7 @@ static bool MinidumpCallback(const char *dump_path,
 }
 
 int main(int argc, char *argv[]) {
-  int handler_index = 1;
+  int handler_index = 0;
   ExceptionHandler handler_ignore(".", NULL, MinidumpCallback,
                            (void*)handler_index, true);
   ++handler_index;

src/client/linux/handler/linux_thread.cc

@@ -297,7 +297,7 @@ int LinuxThread::GetModuleCount() const {
 int LinuxThread::ListModules(
     CallbackParam<ModuleCallback> *callback_param) const {
   char line[512];
-  char *maps_path = "/proc/self/maps";
+  const char *maps_path = "/proc/self/maps";
 
   int module_count = 0;
   FILE *fp = fopen(maps_path, "r");

src/client/linux/handler/minidump_generator.cc

@@ -48,6 +48,10 @@
 #include "google_breakpad/common/minidump_format.h"
 #include "client/linux/handler/minidump_generator.h"
 
+#ifndef CLONE_UNTRACED
+#define CLONE_UNTRACED 0x00800000
+#endif
+
 // This unnamed namespace contains helper functions.
 namespace {
 
@@ -322,11 +326,11 @@ bool WriteThreadStream(MinidumpFileWriter *minidump_writer,
 }
 
 bool WriteCPUInformation(MDRawSystemInfo *sys_info) {
-  char *proc_cpu_path = "/proc/cpuinfo";
+  const char *proc_cpu_path = "/proc/cpuinfo";
   char line[128];
 
   struct CpuInfoEntry {
-    char *info_name;
+    const char *info_name;
     int value;
   } cpu_info_table[] = {
     { "processor", -1 },
@@ -384,7 +388,7 @@ bool WriteOSInformation(MinidumpFileWriter *minidump_writer,
     char os_version[512];
     size_t space_left = sizeof(os_version);
     memset(os_version, 0, space_left);
-    char *os_info_table[] = {
+    const char *os_info_table[] = {
       uts.sysname,
       uts.release,
       uts.version,
@@ -392,7 +396,7 @@ bool WriteOSInformation(MinidumpFileWriter *minidump_writer,
       "GNU/Linux",
       NULL
     };
-    for (char **cur_os_info = os_info_table;
+    for (const char **cur_os_info = os_info_table;
          *cur_os_info != NULL;
          cur_os_info++) {
       if (cur_os_info != os_info_table && space_left > 1) {
@@ -472,7 +476,8 @@ bool WriteThreadListStream(MinidumpFileWriter *minidump_writer,
   context.thread_index = 0;
   CallbackParam<ThreadCallback> callback_param(ThreadInfomationCallback,
                                                &context);
-  return thread_lister->ListThreads(&callback_param) == thread_count;
+  int written = thread_lister->ListThreads(&callback_param);
+  return written == thread_count;
 }
 
 bool WriteCVRecord(MinidumpFileWriter *minidump_writer,
@@ -481,7 +486,7 @@ bool WriteCVRecord(MinidumpFileWriter *minidump_writer,
   TypedMDRVA<MDCVInfoPDB70> cv(minidump_writer);
 
   // Only return the last path component of the full module path
-  char *module_name = strrchr(module_path, '/');
+  const char *module_name = strrchr(module_path, '/');
   // Increment past the slash
   if (module_name)
     ++module_name;
@@ -491,11 +496,13 @@ bool WriteCVRecord(MinidumpFileWriter *minidump_writer,
   size_t module_name_length = strlen(module_name);
   if (!cv.AllocateObjectAndArray(module_name_length + 1, sizeof(u_int8_t)))
     return false;
-  if (!cv.CopyIndexAfterObject(0, module_name, module_name_length))
+  if (!cv.CopyIndexAfterObject(0, const_cast<char *>(module_name),
+                               module_name_length))
     return false;
 
   module->cv_record = cv.location();
   MDCVInfoPDB70 *cv_ptr = cv.get();
+  memset(cv_ptr, 0, sizeof(MDCVInfoPDB70));
   cv_ptr->cv_signature = MD_CVINFOPDB70_SIGNATURE;
   cv_ptr->age = 0;
 
@@ -517,9 +524,8 @@ bool WriteCVRecord(MinidumpFileWriter *minidump_writer,
     cv_ptr->signature.data4[5] = identifier[13];
     cv_ptr->signature.data4[6] = identifier[14];
     cv_ptr->signature.data4[7] = identifier[15];
-    return true;
   }
-  return false;
+  return true;
 }
 
 struct ModuleInfoCallbackCtx {

src/common/linux/dump_symbols.cc

@@ -138,15 +138,15 @@ void FixAddress(void *obj_base) {
 }
 
 // Find the prefered loading address of the binary.
-// It is the PT_LOAD segment with offset to zero.
 ElfW(Addr) GetLoadingAddress(const ElfW(Phdr) *program_headers, int nheader) {
   for (int i = 0; i < nheader; ++i) {
     const ElfW(Phdr) &header = program_headers[i];
+    // For executable, it is the PT_LOAD segment with offset to zero.
     if (header.p_type == PT_LOAD &&
         header.p_offset == 0)
       return header.p_vaddr;
   }
-  assert(!"Should get a valid loading address");
+  // For other types of ELF, return 0.
   return 0;
 }
 
@@ -263,6 +263,8 @@ int LoadFuncSymbols(struct nlist *list,
       cur_list += LoadLineInfo(cur_list, list_end, &func_info);
       // Functions in this module should have address bigger than the module
       // startring address.
+      // There maybe a lot of duplicated entry for a function in the symbol,
+      // only one of them can met this.
       if (func_info.addr >= source_file_info->addr) {
         source_file_info->func_info.push_back(func_info);
       }
@@ -344,10 +346,40 @@ bool ComputeSizeAndRVA(ElfW(Addr) loading_addr, struct SymbolInfo *symbols) {
       ElfW(Addr) next_addr = NextAddress(&sorted_functions,
                                          &sorted_files,
                                          func_info);
-      if (next_addr == 0)
+      // I've noticed functions with an address bigger than any other functions
-        continue;
+      // and source files modules, this is probably the last function in the
-      func_info.size = next_addr - func_info.addr;
+      // module, due to limitions of Linux stab symbol, it is impossible to get
+      // the exact size of this kind of function, thus we give it a default
+      // very big value. This should be safe since this is the last function.
+      // But it is a ugly hack.....
+      // The following code can reproduce the case:
+      // template<class T>
+      // void Foo(T value) {
+      // }
+      //
+      // int main(void) {
+      //   Foo(10);
+      //   Foo(std::string("hello"));
+      //   return 0;
+      // }
+      // TODO(liuli): Find a better solution.
+      static const int kDefaultSize = 0x10000000;
+      static int no_next_addr_count = 0;
+      if (next_addr != 0) {
+        func_info.size = next_addr - func_info.addr;
+      } else {
+        if (no_next_addr_count > 1) {
+          fprintf(stderr, "Got more than one funtion without the \
+                  following symbol. Igore this function.\n");
+          fprintf(stderr, "The dumped symbol may not correct.\n");
+          assert(!"This should not happen!\n");
+          func_info.size = 0;
+          continue;
+        }
 
+        no_next_addr_count++;
+        func_info.size = kDefaultSize;
+      }
       // Compute line size.
       for (size_t k = 0; k < func_info.line_info.size(); ++k) {
         struct LineInfo &line_info = func_info.line_info[k];
@@ -356,9 +388,18 @@ bool ComputeSizeAndRVA(ElfW(Addr) loading_addr, struct SymbolInfo *symbols) {
           line_info.size =
             func_info.line_info[k + 1].rva_to_func - line_info.rva_to_func;
         } else {
+          // The last line in the function.
+          // If we can find a function or source file symbol immediately
+          // following the line, we can get the size of the line by computing
+          // the difference of the next address to the starting address of this
+          // line.
+          // Otherwise, we need to set a default big enough value. This occurs
+          // mostly because the this function is the last one in the module.
           if (next_addr != 0) {
             ElfW(Off) next_addr_offset = next_addr - func_info.addr;
             line_info.size = next_addr_offset - line_info.rva_to_func;
+          } else {
+            line_info.size = kDefaultSize;
           }
         }
         line_info.rva_to_base = line_info.rva_to_func + func_info.rva_to_base;
@@ -409,8 +450,6 @@ bool LoadSymbols(ElfW(Ehdr) *elf_header, struct SymbolInfo *symbols) {
   ElfW(Addr) loading_addr = GetLoadingAddress(
       reinterpret_cast<ElfW(Phdr) *>(elf_header->e_phoff),
       elf_header->e_phnum);
-  if (loading_addr == 0)
-    return false;
 
   const ElfW(Shdr) *sections =
     reinterpret_cast<ElfW(Shdr) *>(elf_header->e_shoff);

src/common/string_conversion.cc

@@ -31,11 +31,11 @@
 #include "processor/scoped_ptr.h"
 #include "common/string_conversion.h"
 
+namespace google_breakpad {
+
 using std::string;
 using std::vector;
 
-namespace google_breakpad {
-
 void UTF8ToUTF16(const char *in, vector<u_int16_t> *out) {
   size_t source_length = strlen(in);
   const UTF8 *source_ptr = reinterpret_cast<const UTF8 *>(in);