Forgot to add modified files for the last commit!

http://breakpad.appspot.com/33002

A=jim.blandy
R=nealsid


git-svn-id: http://google-breakpad.googlecode.com/svn/branches/linux-dwarf@415 4c0a9323-5329-0410-9bdc-e9ce6186880e

src/client/linux/handler/Makefile

@@ -1,55 +1,45 @@
 CXX=g++
 CC=gcc
 
-CXXFLAGS=-gstabs+ -I../../.. -Wall -D_REENTRANT
+CXXFLAGS=-gstabs+ -I../../../ -I../../../testing/gtest/include -I../../../testing/include -I../../../testing/gtest -D_REENTRANT -m32
+CFLAGS=$(CXXFLAGS)
 LDFLAGS=-lpthread
 
 OBJ_DIR=.
 BIN_DIR=.
 
-THREAD_SRC=linux_thread.cc
-SHARE_SRC=../../minidump_file_writer.cc\
-	  ../../../common/string_conversion.cc\
-	  ../../../common/linux/file_id.cc\
-	  minidump_generator.cc
-HANDLER_SRC=exception_handler.cc\
-	    ../../../common/linux/guid_creator.cc
-SHARE_C_SRC=../../../common/convert_UTF.c
+TEST_CC_SRC=exception_handler_unittest.cc \
+	    exception_handler.cc \
+	    ../../../testing/gtest/src/gtest-all.cc \
+	    ../../../common/linux/guid_creator.cc \
+	    ../minidump_writer/minidump_writer.cc \
+	    ../../minidump_file_writer.cc \
+	    ../minidump_writer/linux_dumper.cc \
+	    ../../../testing/gtest/src/gtest_main.cc \
+	    ../../../common/string_conversion.cc \
+	    ../minidump_writer/directory_reader_unittest.cc \
+	    ../minidump_writer/line_reader_unittest.cc \
+	    ../minidump_writer/linux_dumper_unittest.cc \
+	    ../minidump_writer/minidump_writer_unittest.cc
 
-THREAD_TEST_SRC=linux_thread_test.cc
-MINIDUMP_TEST_SRC=minidump_test.cc
-EXCEPTION_TEST_SRC=exception_handler_test.cc
+TEST_C_SRC = ../../../common/convert_UTF.c
 
-THREAD_OBJ=$(patsubst %.cc,$(OBJ_DIR)/%.o,$(THREAD_SRC))
-SHARE_OBJ=$(patsubst %.cc,$(OBJ_DIR)/%.o,$(SHARE_SRC))
-HANDLER_OBJ=$(patsubst %.cc,$(OBJ_DIR)/%.o,$(HANDLER_SRC))
-SHARE_C_OBJ=$(patsubst %.c,$(OBJ_DIR)/%.o,$(SHARE_C_SRC)) md5.o
-THREAD_TEST_OBJ=$(patsubst %.cc,$(OBJ_DIR)/%.o, $(THREAD_TEST_SRC))\
-		$(THREAD_OBJ)
-MINIDUMP_TEST_OBJ=$(patsubst %.cc,$(OBJ_DIR)/%.o, $(MINIDUMP_TEST_SRC))\
-		  $(THREAD_OBJ) $(SHARE_OBJ) $(SHARE_C_OBJ)
-EXCEPTION_TEST_OBJ=$(patsubst %.cc,$(OBJ_DIR)/%.o, $(EXCEPTION_TEST_SRC))\
-		   $(THREAD_OBJ) $(SHARE_OBJ) $(SHARE_C_OBJ) $(HANDLER_OBJ)
+TEST_CC_OBJ=$(patsubst %.cc, $(OBJ_DIR)/%.o,$(TEST_CC_SRC))
+TEST_C_OBJ=$(patsubst %.c, $(OBJ_DIR)/%.o, $(TEST_C_SRC))
 
-BIN=$(BIN_DIR)/minidump_test\
-    $(BIN_DIR)/linux_thread_test\
-    $(BIN_DIR)/exception_handler_test
+LINUX_CLIENT_BIN=$(BIN_DIR)/linux_client_test
+
+BIN=$(LINUX_CLIENT_BIN)
 
 .PHONY:all clean
 
+check:$(BIN)
+	$(LINUX_CLIENT_BIN)
+
 all:$(BIN)
 
-$(BIN_DIR)/linux_thread_test:$(THREAD_TEST_OBJ)
+$(BIN_DIR)/linux_client_test:$(TEST_CC_OBJ) $(TEST_C_OBJ)
 	$(CXX) $(CXXFLAGS) $(LDFLAGS) $^ -o $@
 
-$(BIN_DIR)/minidump_test:$(MINIDUMP_TEST_OBJ)
-	$(CXX) $(CXXFLAGS) $(LDFLAGS) $^ -o $@
-
-$(BIN_DIR)/exception_handler_test:$(EXCEPTION_TEST_OBJ)
-	$(CXX) $(CXXFLAGS) $(LDFLAGS) $^ -o $@
-
-md5.o:../../../common/md5.c
-	$(CC) $(CXXFLAGS) -c $^
-
 clean:
-	rm -f $(BIN) *.o *.dmp
+	rm -f $(BIN) $(TEST_CC_OBJ) $(TEST_C_OBJ)

src/client/linux/handler/exception_handler.cc

@@ -1,8 +1,6 @@
-// Copyright (c) 2006, Google Inc.
+// Copyright (c) 2009, Google Inc.
 // All rights reserved.
 //
-// Author: Li Liu
-//
 // Redistribution and use in source and binary forms, with or without
 // modification, are permitted provided that the following conditions are
 // met:
@@ -29,48 +27,87 @@
 // (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
 // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
-#include <signal.h>
-#include <sys/stat.h>
-#include <sys/types.h>
-#include <unistd.h>
+// The ExceptionHandler object installs signal handlers for a number of
+// signals. We rely on the signal handler running on the thread which crashed
+// in order to identify it. This is true of the synchronous signals (SEGV etc),
+// but not true of ABRT. Thus, if you send ABRT to yourself in a program which
+// uses ExceptionHandler, you need to use tgkill to direct it to the current
+// thread.
+//
+// The signal flow looks like this:
+//
+//   SignalHandler (uses a global stack of ExceptionHandler objects to find
+//        |         one to handle the signal. If the first rejects it, try
+//        |         the second etc...)
+//        V
+//   HandleSignal ----------------------------| (clones a new process which
+//        |                                   |  shares an address space with
+//   (wait for cloned                         |  the crashed process. This
+//     process)                               |  allows us to ptrace the crashed
+//        |                                   |  process)
+//        V                                   V
+//   (set signal handler to             ThreadEntry (static function to bounce
+//    SIG_DFL and rethrow,                    |      back into the object)
+//    killing the crashed                     |
+//    process)                                V
+//                                          DoDump  (writes minidump)
+//                                            |
+//                                            V
+//                                         sys_exit
+//
 
-#include <cassert>
-#include <cstdlib>
-#include <cstdio>
-#include <ctime>
-#include <linux/limits.h>
+// This code is a little fragmented. Different functions of the ExceptionHandler
+// class run in a number of different contexts. Some of them run in a normal
+// context and are easy to code, others run in a compromised context and the
+// restrictions at the top of minidump_writer.cc apply: no libc and use the
+// alternative malloc. Each function should have comment above it detailing the
+// context which it runs in.
 
 #include "client/linux/handler/exception_handler.h"
+
+#include <errno.h>
+#include <fcntl.h>
+#include <linux/limits.h>
+#include <sched.h>
+#include <signal.h>
+#include <stdio.h>
+#include <sys/mman.h>
+#include <sys/signal.h>
+#include <sys/syscall.h>
+#include <sys/ucontext.h>
+#include <sys/user.h>
+#include <sys/wait.h>
+#include <unistd.h>
+
+#include "common/linux/linux_libc_support.h"
+#include "common/linux/linux_syscall_support.h"
+#include "common/linux/memory.h"
+#include "client/linux/minidump_writer//minidump_writer.h"
 #include "common/linux/guid_creator.h"
-#include "google_breakpad/common/minidump_format.h"
+
+// A wrapper for the tgkill syscall: send a signal to a specific thread.
+static int tgkill(pid_t tgid, pid_t tid, int sig) {
+  syscall(__NR_tgkill, tgid, tid, sig);
+}
 
 namespace google_breakpad {
 
-// Signals that we are interested.
-int SigTable[] = {
-#if defined(SIGSEGV)
-  SIGSEGV,
-#endif
-#ifdef SIGABRT
-  SIGABRT,
-#endif
-#ifdef SIGFPE
-  SIGFPE,
-#endif
-#ifdef SIGILL
-  SIGILL,
-#endif
-#ifdef SIGBUS
-  SIGBUS,
-#endif
+// The list of signals which we consider to be crashes. The default action for
+// all these signals must be Core (see man 7 signal) because we rethrow the
+// signal after handling it and expect that it'll be fatal.
+static const int kExceptionSignals[] = {
+  SIGSEGV, SIGABRT, SIGFPE, SIGILL, SIGBUS, -1
 };
 
-std::vector<ExceptionHandler*> *ExceptionHandler::handler_stack_ = NULL;
-int ExceptionHandler::handler_stack_index_ = 0;
+// We can stack multiple exception handlers. In that case, this is the global
+// which holds the stack.
+std::vector<ExceptionHandler*>* ExceptionHandler::handler_stack_ = NULL;
+unsigned ExceptionHandler::handler_stack_index_ = 0;
 pthread_mutex_t ExceptionHandler::handler_stack_mutex_ =
-PTHREAD_MUTEX_INITIALIZER;
+    PTHREAD_MUTEX_INITIALIZER;
 
-ExceptionHandler::ExceptionHandler(const string &dump_path,
+// Runs before crashing: normal context.
+ExceptionHandler::ExceptionHandler(const std::string &dump_path,
                                    FilterCallback filter,
                                    MinidumpCallback callback,
                                    void *callback_context,
@@ -79,212 +116,76 @@ ExceptionHandler::ExceptionHandler(const string &dump_path,
       callback_(callback),
       callback_context_(callback_context),
       dump_path_(),
-      installed_handler_(install_handler) {
+      handler_installed_(install_handler),
+      crash_handler_(NULL) {
   set_dump_path(dump_path);
 
-  act_.sa_handler = HandleException;
-  act_.sa_flags = SA_ONSTACK;
-  sigemptyset(&act_.sa_mask);
-  // now, make sure we're blocking all the signals we are handling
-  // when we're handling any of them
-  for ( size_t i = 0; i < sizeof(SigTable) / sizeof(SigTable[0]); ++i) {
-    sigaddset(&act_.sa_mask, SigTable[i]);
-  }
-
   if (install_handler) {
-    SetupHandler();
+    InstallHandlers();
+
     pthread_mutex_lock(&handler_stack_mutex_);
-    if (handler_stack_ == NULL)
-      handler_stack_ = new std::vector<ExceptionHandler *>;
-    handler_stack_->push_back(this);
+      if (handler_stack_ == NULL)
+        handler_stack_ = new std::vector<ExceptionHandler *>;
+      handler_stack_->push_back(this);
     pthread_mutex_unlock(&handler_stack_mutex_);
   }
 }
 
+// Runs before crashing: normal context.
 ExceptionHandler::~ExceptionHandler() {
-  TeardownAllHandler();
-  pthread_mutex_lock(&handler_stack_mutex_);
-  if (handler_stack_->back() == this) {
-    handler_stack_->pop_back();
-  } else {
-    fprintf(stderr, "warning: removing Breakpad handler out of order\n");
-    for (std::vector<ExceptionHandler *>::iterator iterator =
-         handler_stack_->begin();
-         iterator != handler_stack_->end();
-         ++iterator) {
-      if (*iterator == this) {
-        handler_stack_->erase(iterator);
-      }
-    }
-  }
-
-  if (handler_stack_->empty()) {
-    // When destroying the last ExceptionHandler that installed a handler,
-    // clean up the handler stack.
-    delete handler_stack_;
-    handler_stack_ = NULL;
-  }
-  pthread_mutex_unlock(&handler_stack_mutex_);
+  UninstallHandlers();
 }
 
-bool ExceptionHandler::WriteMinidump() {
-  bool success = InternalWriteMinidump(0, 0, NULL);
-  UpdateNextID();
-  return success;
-}
+// Runs before crashing: normal context.
+bool ExceptionHandler::InstallHandlers() {
+  // We run the signal handlers on an alternative stack because we might have
+  // crashed because of a stack overflow.
 
-// static
-bool ExceptionHandler::WriteMinidump(const string &dump_path,
-                   MinidumpCallback callback,
-                   void *callback_context) {
-  ExceptionHandler handler(dump_path, NULL, callback,
-                           callback_context, false);
-  return handler.InternalWriteMinidump(0, 0, NULL);
-}
+  // We use this value rather than SIGSTKSZ because we would end up overrunning
+  // such a small stack.
+  static const unsigned kSigStackSize = 8192;
 
-void ExceptionHandler::SetupHandler() {
-  // Signal on a different stack to avoid using the stack
-  // of the crashing thread.
-  struct sigaltstack sig_stack;
-  sig_stack.ss_sp = malloc(MINSIGSTKSZ);
-  if (sig_stack.ss_sp == NULL)
-    return;
-  sig_stack.ss_size = MINSIGSTKSZ;
-  sig_stack.ss_flags = 0;
+  signal_stack = malloc(kSigStackSize);
+  stack_t stack;
+  memset(&stack, 0, sizeof(stack));
+  stack.ss_sp = signal_stack;
+  stack.ss_size = kSigStackSize;
 
-  if (sigaltstack(&sig_stack, NULL) < 0)
-    return;
-  for (size_t i = 0; i < sizeof(SigTable) / sizeof(SigTable[0]); ++i)
-    SetupHandler(SigTable[i]);
-}
-
-void ExceptionHandler::SetupHandler(int signo) {
-
-  // We're storing pointers to the old signal action
-  // structure, rather than copying the structure
-  // because we can't count on the sa_mask field to
-  // be scalar.
-  struct sigaction *old_act = &old_actions_[signo];
-
-  if (sigaction(signo, &act_, old_act) < 0)
-   return;
-}
-
-void ExceptionHandler::TeardownHandler(int signo) {
-  TeardownHandler(signo, NULL);
-}
-
-void ExceptionHandler::TeardownHandler(int signo, struct sigaction *final_handler) {
-  if (old_actions_[signo].sa_handler) {
-    struct sigaction *act = &old_actions_[signo];
-    sigaction(signo, act, final_handler);
-    memset(&old_actions_[signo], 0x0, sizeof(struct sigaction));
-  }
-}
-
-void ExceptionHandler::TeardownAllHandler() {
-  for (size_t i = 0; i < sizeof(SigTable) / sizeof(SigTable[0]); ++i) {
-    TeardownHandler(SigTable[i]);
-  }
-}
-
-// static
-void ExceptionHandler::HandleException(int signo) {
-  // In Linux, the context information about the signal is put on the stack of
-  // the signal handler frame as value parameter. For some reasons, the
-  // prototype of the handler doesn't declare this information as parameter, we
-  // will do it by hand. It is the second parameter above the signal number.
-  // However, if we are being called by another signal handler passing the
-  // signal up the chain, then we may not have this random extra parameter,
-  // so we may have to walk the stack to find it.  We do the actual work
-  // on another thread, where it's a little safer, but we want the ebp
-  // from this frame to find it.
-  uintptr_t current_ebp = 0;
-  asm volatile ("movl %%ebp, %0"
-                :"=m"(current_ebp));
-
-  pthread_mutex_lock(&handler_stack_mutex_);
-  ExceptionHandler *current_handler =
-    handler_stack_->at(handler_stack_->size() - ++handler_stack_index_);
-  pthread_mutex_unlock(&handler_stack_mutex_);
-
-  // Restore original handler.
-  struct sigaction old_action;
-  current_handler->TeardownHandler(signo, &old_action);
-
-  struct sigcontext *sig_ctx = NULL;
-  if (current_handler->InternalWriteMinidump(signo, current_ebp, &sig_ctx)) {
-    // Fully handled this exception, safe to exit.
-    exit(EXIT_FAILURE);
-  } else {
-    // Exception not fully handled, will call the next handler in stack to
-    // process it.
-    if (old_action.sa_handler != NULL && sig_ctx != NULL) {
-
-      // Have our own typedef, because of the comment above w.r.t signal
-      // context on the stack
-      typedef void (*SignalHandler)(int signo, struct sigcontext);
-
-      SignalHandler old_handler =
-          reinterpret_cast<SignalHandler>(old_action.sa_handler);
-
-      sigset_t old_set;
-      // Use SIG_BLOCK here because we don't want to unblock a signal
-      // that the signal handler we're currently in needs to block
-      sigprocmask(SIG_BLOCK, &old_action.sa_mask, &old_set);
-      old_handler(signo, *sig_ctx);
-      sigprocmask(SIG_SETMASK, &old_set, NULL);
-    }
-
-  }
-
-  pthread_mutex_lock(&handler_stack_mutex_);
-  current_handler->SetupHandler(signo);
-  --handler_stack_index_;
-  // All the handlers in stack have been invoked to handle the exception,
-  // normally the process should be terminated and should not reach here.
-  // In case we got here, ask the OS to handle it to avoid endless loop,
-  // normally the OS will generate a core and termiate the process. This
-  // may be desired to debug the program.
-  if (handler_stack_index_ == 0)
-    signal(signo, SIG_DFL);
-  pthread_mutex_unlock(&handler_stack_mutex_);
-}
-
-bool ExceptionHandler::InternalWriteMinidump(int signo,
-                                             uintptr_t sighandler_ebp,
-                                             struct sigcontext **sig_ctx) {
-  if (filter_ && !filter_(callback_context_))
+  if (sigaltstack(&stack, NULL) == -1)
     return false;
 
-  bool success = false;
-  // Block all the signals we want to process when writting minidump.
-  // We don't want it to be interrupted.
-  sigset_t sig_blocked, sig_old;
-  bool blocked = true;
-  sigfillset(&sig_blocked);
-  for (size_t i = 0; i < sizeof(SigTable) / sizeof(SigTable[0]); ++i)
-    sigdelset(&sig_blocked, SigTable[i]);
-  if (sigprocmask(SIG_BLOCK, &sig_blocked, &sig_old) != 0) {
-    blocked = false;
-    fprintf(stderr, "google_breakpad::ExceptionHandler::HandleException: "
-                    "failed to block signals.\n");
+  struct sigaction sa;
+  memset(&sa, 0, sizeof(sa));
+  sigemptyset(&sa.sa_mask);
+
+  // mask all exception signals when we're handling one of them.
+  for (unsigned i = 0; kExceptionSignals[i] != -1; ++i)
+    sigaddset(&sa.sa_mask, kExceptionSignals[i]);
+
+  sa.sa_sigaction = SignalHandler;
+  sa.sa_flags = SA_ONSTACK | SA_SIGINFO;
+
+  for (unsigned i = 0; kExceptionSignals[i] != -1; ++i) {
+    struct sigaction* old = new struct sigaction;
+    if (sigaction(kExceptionSignals[i], &sa, old) == -1)
+      return false;
+    old_handlers_.push_back(std::make_pair(kExceptionSignals[i], old));
   }
-
-  success = minidump_generator_.WriteMinidumpToFile(
-                     next_minidump_path_c_, signo, sighandler_ebp, sig_ctx);
-
-  // Unblock the signals.
-  if (blocked) {
-    sigprocmask(SIG_SETMASK, &sig_old, NULL);
-  }
-
-  if (callback_)
-    success = callback_(dump_path_c_, next_minidump_id_c_,
-                          callback_context_, success);
-  return success;
 }
 
+// Runs before crashing: normal context.
+void ExceptionHandler::UninstallHandlers() {
+  for (unsigned i = 0; i < old_handlers_.size(); ++i) {
+    struct sigaction *action =
+        reinterpret_cast<struct sigaction*>(old_handlers_[i].second);
+    sigaction(old_handlers_[i].first, action, NULL);
+    delete action;
+  }
+
+  old_handlers_.clear();
+}
+
+// Runs before crashing: normal context.
 void ExceptionHandler::UpdateNextID() {
   GUID guid;
   char guid_str[kGUIDStringLength + 1];
@@ -302,4 +203,120 @@ void ExceptionHandler::UpdateNextID() {
   }
 }
 
+// This function runs in a compromised context: see the top of the file.
+// Runs on the crashing thread.
+// static
+void ExceptionHandler::SignalHandler(int sig, siginfo_t* info, void* uc) {
+  // All the exception signals are blocked at this point.
+
+  pthread_mutex_lock(&handler_stack_mutex_);
+
+  if (!handler_stack_->size()) {
+    pthread_mutex_unlock(&handler_stack_mutex_);
+    return;
+  }
+
+  for (int i = handler_stack_->size() - 1; i >= 0; --i) {
+    if ((*handler_stack_)[i]->HandleSignal(sig, info, uc)) {
+      // successfully handled: We are in an invalid state since an exception
+      // signal has been delivered. We don't call the exit handlers because
+      // they could end up corrupting on-disk state.
+      break;
+    }
+  }
+
+  pthread_mutex_unlock(&handler_stack_mutex_);
+
+  // Terminate ourselves with the same signal so that our parent knows that we
+  // crashed. The default action for all the signals which we catch is Core, so
+  // this is the end of us.
+  signal(sig, SIG_DFL);
+  tgkill(getpid(), sys_gettid(), sig);
+
+  // not reached.
+}
+
+struct ThreadArgument {
+  pid_t pid;  // the crashing process
+  ExceptionHandler* handler;
+  const void* context;  // a CrashContext structure
+  size_t context_size;
+};
+
+// This is the entry function for the cloned process. We are in a compromised
+// context here: see the top of the file.
+// static
+int ExceptionHandler::ThreadEntry(void *arg) {
+  const ThreadArgument *thread_arg = reinterpret_cast<ThreadArgument*>(arg);
+  return thread_arg->handler->DoDump(thread_arg->pid, thread_arg->context,
+                                     thread_arg->context_size) == false;
+}
+
+// This function runs in a compromised context: see the top of the file.
+// Runs on the crashing thread.
+bool ExceptionHandler::HandleSignal(int sig, siginfo_t* info, void* uc) {
+  if (filter_ && !filter_(callback_context_))
+    return false;
+
+  // Allow ourselves to be dumped.
+  sys_prctl(PR_SET_DUMPABLE, 1);
+
+  CrashContext context;
+  memcpy(&context.siginfo, info, sizeof(siginfo_t));
+  memcpy(&context.context, uc, sizeof(struct ucontext));
+  memcpy(&context.float_state, ((struct ucontext *)uc)->uc_mcontext.fpregs,
+         sizeof(context.float_state));
+  context.tid = sys_gettid();
+
+  if (crash_handler_ && crash_handler_(&context, sizeof(context),
+                                       callback_context_))
+    return true;
+
+  static const unsigned kChildStackSize = 8000;
+  PageAllocator allocator;
+  uint8_t* stack = (uint8_t*) allocator.Alloc(kChildStackSize);
+  if (!stack)
+    return false;
+  // clone() needs the top-most address. (scrub just to be safe)
+  stack += kChildStackSize;
+  my_memset(stack - 16, 0, 16);
+
+  ThreadArgument thread_arg;
+  thread_arg.handler = this;
+  thread_arg.pid = getpid();
+  thread_arg.context = &context;
+  thread_arg.context_size = sizeof(context);
+
+  const pid_t child = sys_clone(
+      ThreadEntry, stack, CLONE_FILES | CLONE_FS | CLONE_UNTRACED,
+      &thread_arg, NULL, NULL, NULL);
+  int r, status;
+  do {
+    r = sys_waitpid(child, &status, __WALL);
+  } while (r == -1 && errno == EINTR);
+
+  if (r == -1) {
+    static const char msg[] = "ExceptionHandler::HandleSignal: waitpid failed:";
+    sys_write(2, msg, sizeof(msg) - 1);
+    sys_write(2, strerror(errno), strlen(strerror(errno)));
+    sys_write(2, "\n", 1);
+  }
+
+  bool success = r != -1 && WIFEXITED(status) && WEXITSTATUS(status) == 0;
+
+  if (callback_)
+    success = callback_(dump_path_c_, next_minidump_id_c_,
+                        callback_context_, success);
+
+  return success;
+}
+
+// This function runs in a compromised context: see the top of the file.
+// Runs on the cloned process.
+bool ExceptionHandler::DoDump(pid_t crashing_process, const void* context,
+                              size_t context_size) {
+  return google_breakpad::WriteMinidump(
+      next_minidump_path_c_, crashing_process, context, context_size);
+}
+
 }  // namespace google_breakpad

src/client/linux/handler/exception_handler.h

@@ -1,8 +1,6 @@
-// Copyright (c) 2006, Google Inc.
+// Copyright (c) 2009, Google Inc.
 // All rights reserved.
 //
-// Author: Li Liu
-//
 // Redistribution and use in source and binary forms, with or without
 // modification, are permitted provided that the following conditions are
 // met:
@@ -29,26 +27,16 @@
 // (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
 // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
-#ifndef CLIENT_LINUX_HANDLER_EXCEPTION_HANDLER_H__
-#define CLIENT_LINUX_HANDLER_EXCEPTION_HANDLER_H__
+#ifndef CLIENT_LINUX_HANDLER_EXCEPTION_HANDLER_H_
+#define CLIENT_LINUX_HANDLER_EXCEPTION_HANDLER_H_
 
-#include <pthread.h>
-
-#include <map>
-#include <string>
-#include <signal.h>
 #include <vector>
+#include <string>
 
-#include "client/linux/handler/minidump_generator.h"
-
-// Context information when exception occured.
-struct sigcontex;
+#include <signal.h>
 
 namespace google_breakpad {
 
-using std::string;
-
-//
 // ExceptionHandler
 //
 // ExceptionHandler can write a minidump file when an exception occurs,
@@ -73,7 +61,6 @@ using std::string;
 //
 // Caller should try to make the callbacks as crash-friendly as possible,
 // it should avoid use heap memory allocation as much as possible.
-//
 class ExceptionHandler {
  public:
   // A callback function to run before Breakpad performs any substantial
@@ -108,6 +95,15 @@ class ExceptionHandler {
                                    void *context,
                                    bool succeeded);
 
+  // In certain cases, a user may wish to handle the generation of the minidump
+  // themselves. In this case, they can install a handler callback which is
+  // called when a crash has occured. If this function returns true, no other
+  // processing of occurs and the process will shortly be crashed. If this
+  // returns false, the normal processing continues.
+  typedef bool (*HandlerCallback)(const void* crash_context,
+                                  size_t crash_context_size,
+                                  void* context);
+
   // Creates a new ExceptionHandler instance to handle writing minidumps.
   // Before writing a minidump, the optional filter callback will be called.
   // Its return value determines whether or not Breakpad should write a
@@ -116,111 +112,87 @@ class ExceptionHandler {
   // If install_handler is true, then a minidump will be written whenever
   // an unhandled exception occurs.  If it is false, minidumps will only
   // be written when WriteMinidump is called.
-  ExceptionHandler(const string &dump_path,
+  ExceptionHandler(const std::string &dump_path,
                    FilterCallback filter, MinidumpCallback callback,
                    void *callback_context,
                    bool install_handler);
   ~ExceptionHandler();
 
   // Get and set the minidump path.
-  string dump_path() const { return dump_path_; }
-  void set_dump_path(const string &dump_path) {
+  std::string dump_path() const { return dump_path_; }
+  void set_dump_path(const std::string &dump_path) {
     dump_path_ = dump_path;
     dump_path_c_ = dump_path_.c_str();
     UpdateNextID();
   }
 
+  void set_crash_handler(HandlerCallback callback) {
+    crash_handler_ = callback;
+  }
+
   // Writes a minidump immediately.  This can be used to capture the
   // execution state independently of a crash.  Returns true on success.
   bool WriteMinidump();
 
   // Convenience form of WriteMinidump which does not require an
   // ExceptionHandler instance.
-  static bool WriteMinidump(const string &dump_path,
+  static bool WriteMinidump(const std::string &dump_path,
                             MinidumpCallback callback,
                             void *callback_context);
 
+  // This structure is passed to minidump_writer.h:WriteMinidump via an opaque
+  // blob. It shouldn't be needed in any user code.
+  struct CrashContext {
+    siginfo_t siginfo;
+    pid_t tid;  // the crashing thread.
+    struct ucontext context;
+    struct _libc_fpstate float_state;
+  };
+
  private:
-  // Setup crash handler.
-  void SetupHandler();
-  // Setup signal handler for a signal.
-  void SetupHandler(int signo);
-  // Teardown the handler for a signal.
-  void TeardownHandler(int signo);
-  // Teardown the handler for a signal.
-  void TeardownHandler(int signo, struct sigaction *old);
-  // Teardown all handlers.
-  void TeardownAllHandler();
+  bool InstallHandlers();
+  void UninstallHandlers();
+  void PreresolveSymbols();
 
-  // Signal handler.
-  static void HandleException(int signo);
-
-  // If called from a signal handler, sighandler_ebp is the ebp of
-  // that signal handler's frame, and sig_ctx is an out parameter
-  // that will be set to point at the sigcontext that was placed
-  // on the stack by the kernel.  You can pass zero and NULL
-  // for the second and third parameters if you are not calling
-  // this from a signal handler.
-  bool InternalWriteMinidump(int signo, uintptr_t sighandler_ebp,
-                             struct sigcontext **sig_ctx);
-
-  // Generates a new ID and stores it in next_minidump_id, and stores the
-  // path of the next minidump to be written in next_minidump_path_.
   void UpdateNextID();
+  static void SignalHandler(int sig, siginfo_t* info, void* uc);
+  bool HandleSignal(int sig, siginfo_t* info, void* uc);
+  static int ThreadEntry(void* arg);
+  bool DoDump(pid_t crashing_process, const void* context,
+              size_t context_size);
 
- private:
-  FilterCallback filter_;
-  MinidumpCallback callback_;
-  void *callback_context_;
+  const FilterCallback filter_;
+  const MinidumpCallback callback_;
+  void* const callback_context_;
 
-  // The directory in which a minidump will be written, set by the dump_path
-  // argument to the constructor, or set_dump_path.
-  string dump_path_;
-
-  // The basename of the next minidump to be written, without the extension
-  string next_minidump_id_;
-
-  // The full pathname of the next minidump to be written, including the file
-  // extension
-  string next_minidump_path_;
+  std::string dump_path_;
+  std::string next_minidump_path_;
+  std::string next_minidump_id_;
 
   // Pointers to C-string representations of the above. These are set
   // when the above are set so we can avoid calling c_str during
   // an exception.
-  const char *dump_path_c_;
-  const char *next_minidump_id_c_;
-  const char *next_minidump_path_c_;
+  const char* dump_path_c_;
+  const char* next_minidump_path_c_;
+  const char* next_minidump_id_c_;
 
-  // True if the ExceptionHandler installed an unhandled exception filter
-  // when created (with an install_handler parameter set to true).
-  bool installed_handler_;
+  const bool handler_installed_;
+  void* signal_stack;  // the handler stack.
+  HandlerCallback crash_handler_;
 
   // The global exception handler stack. This is need becuase there may exist
   // multiple ExceptionHandler instances in a process. Each will have itself
   // registered in this stack.
-  static std::vector<ExceptionHandler *> *handler_stack_;
+  static std::vector<ExceptionHandler*> *handler_stack_;
   // The index of the handler that should handle the next exception.
-  static int handler_stack_index_;
+  static unsigned handler_stack_index_;
   static pthread_mutex_t handler_stack_mutex_;
 
-  // The minidump generator.
-  MinidumpGenerator minidump_generator_;
-
-  // disallow copy ctor and operator=
-  explicit ExceptionHandler(const ExceptionHandler &);
-  void operator=(const ExceptionHandler &);
-
-  // The sigactions structure we use for each signal
-  struct sigaction act_;
-
-
-  // Keep the previous handlers for the signal.
-  // We're wasting a bit of memory here since we only change
-  // the handler for some signals but i want to avoid allocating
-  // memory in the signal handler
-  struct sigaction old_actions_[NSIG];
+  // A vector of the old signal handlers. The void* is a pointer to a newly
+  // allocated sigaction structure to avoid pulling in too many includes.
+  std::vector<std::pair<int, void *> > old_handlers_;
 };
 
 }  // namespace google_breakpad
 
-#endif  // CLIENT_LINUX_HANDLER_EXCEPTION_HANDLER_H__
+#endif  // CLIENT_LINUX_HANDLER_EXCEPTION_HANDLER_H_

src/client/linux/handler/exception_handler_test.cc

@@ -1,124 +0,0 @@
-// Copyright (c) 2006, Google Inc.
-// All rights reserved.
-//
-// Author: Li Liu
-//
-// Redistribution and use in source and binary forms, with or without
-// modification, are permitted provided that the following conditions are
-// met:
-//
-//     * Redistributions of source code must retain the above copyright
-// notice, this list of conditions and the following disclaimer.
-//     * Redistributions in binary form must reproduce the above
-// copyright notice, this list of conditions and the following disclaimer
-// in the documentation and/or other materials provided with the
-// distribution.
-//     * Neither the name of Google Inc. nor the names of its
-// contributors may be used to endorse or promote products derived from
-// this software without specific prior written permission.
-//
-// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-
-#include <pthread.h>
-#include <unistd.h>
-
-#include <cassert>
-#include <cstdio>
-#include <cstdlib>
-#include <cstring>
-
-#include "client/linux/handler/exception_handler.h"
-#include "client/linux/handler/linux_thread.h"
-
-using namespace google_breakpad;
-
-// Thread use this to see if it should stop working.
-static bool should_exit = false;
-
-static int foo2(int arg) {
-  // Stack variable, used for debugging stack dumps.
-  /*DDDebug*/printf("%s:%d\n", __FUNCTION__, __LINE__);
-  int c = 0xcccccccc;
-  fprintf(stderr, "Thread trying to crash: %x\n", getpid());
-  c = *reinterpret_cast<int *>(0x5);
-  return c;
-}
-
-static int foo(int arg) {
-  // Stack variable, used for debugging stack dumps.
-  int b = 0xbbbbbbbb;
-  b = foo2(b);
-  return b;
-}
-
-static void *thread_crash(void *) {
-  // Stack variable, used for debugging stack dumps.
-  int a = 0xaaaaaaaa;
-  sleep(1);
-  a = foo(a);
-  printf("%x\n", a);
-  return NULL;
-}
-
-static void *thread_main(void *) {
-  while (!should_exit)
-    sleep(1);
-  return NULL;
-}
-
-static void CreateCrashThread() {
-  pthread_t h;
-  pthread_create(&h, NULL, thread_crash, NULL);
-  pthread_detach(h);
-}
-
-// Create working threads.
-static void CreateThread(int num) {
-  pthread_t h;
-  for (int i = 0; i < num; ++i) {
-    pthread_create(&h, NULL, thread_main, NULL);
-    pthread_detach(h);
-  }
-}
-
-// Callback when minidump written.
-static bool MinidumpCallback(const char *dump_path,
-                             const char *minidump_id,
-                             void *context,
-                             bool succeeded) {
-  int index = reinterpret_cast<int>(context);
-  printf("%d %s: %s is dumped\n", index, __FUNCTION__, minidump_id);
-  if (index == 0) {
-    should_exit = true;
-    return true;
-  }
-  // Don't process it.
-  return false;
-}
-
-int main(int argc, char *argv[]) {
-  int handler_index = 0;
-  ExceptionHandler handler_ignore(".", NULL, MinidumpCallback,
-                           (void*)handler_index, true);
-  ++handler_index;
-  ExceptionHandler handler_process(".", NULL, MinidumpCallback,
-                           (void*)handler_index, true);
-  CreateCrashThread();
-  CreateThread(10);
-
-  while (true)
-    sleep(1);
-  should_exit = true;
-
-  return 0;
-}

src/client/linux/handler/exception_handler_unittest.cc

@@ -0,0 +1,256 @@
+// Copyright (c) 2009, Google Inc.
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+#include <string>
+
+#include <stdint.h>
+#include <unistd.h>
+#include <signal.h>
+#include <sys/poll.h>
+#include <sys/socket.h>
+#include <sys/uio.h>
+
+#include "client/linux/handler//exception_handler.h"
+#include "client/linux/minidump_writer/minidump_writer.h"
+#include "common/linux/linux_libc_support.h"
+#include "common/linux/linux_syscall_support.h"
+#include "breakpad_googletest_includes.h"
+
+// This provides a wrapper around system calls which may be
+// interrupted by a signal and return EINTR. See man 7 signal.
+#define HANDLE_EINTR(x) ({ \
+  typeof(x) __eintr_result__; \
+  do { \
+    __eintr_result__ = x; \
+  } while (__eintr_result__ == -1 && errno == EINTR); \
+  __eintr_result__;\
+})
+
+using namespace google_breakpad;
+
+static void sigchld_handler(int signo) { }
+
+class ExceptionHandlerTest : public ::testing::Test {
+ protected:
+  void SetUp() {
+    // We need to be able to wait for children, so SIGCHLD cannot be SIG_IGN.
+    struct sigaction sa;
+    memset(&sa, 0, sizeof(sa));
+    sa.sa_handler = sigchld_handler;
+    ASSERT_NE(sigaction(SIGCHLD, &sa, &old_action), -1);
+  }
+
+  void TearDown() {
+    sigaction(SIGCHLD, &old_action, NULL);
+  }
+
+  struct sigaction old_action;
+};
+
+TEST(ExceptionHandlerTest, Simple) {
+  ExceptionHandler handler("/tmp", NULL, NULL, NULL, true);
+}
+
+static bool DoneCallback(const char* dump_path,
+                         const char* minidump_id,
+                         void* context,
+                         bool succeeded) {
+  if (!succeeded)
+    return succeeded;
+
+  int fd = (int) context;
+  uint32_t len = my_strlen(minidump_id);
+  HANDLE_EINTR(sys_write(fd, &len, sizeof(len)));
+  HANDLE_EINTR(sys_write(fd, minidump_id, len));
+  sys_close(fd);
+
+  return true;
+}
+
+TEST(ExceptionHandlerTest, ChildCrash) {
+  int fds[2];
+  ASSERT_NE(pipe(fds), -1);
+
+  const pid_t child = fork();
+  if (child == 0) {
+    close(fds[0]);
+    ExceptionHandler handler("/tmp", NULL, DoneCallback, (void*) fds[1],
+                             true);
+    *reinterpret_cast<int*>(NULL) = 0;
+  }
+  close(fds[1]);
+
+  int status;
+  ASSERT_NE(HANDLE_EINTR(waitpid(child, &status, 0)), -1);
+  ASSERT_TRUE(WIFSIGNALED(status));
+  ASSERT_EQ(WTERMSIG(status), SIGSEGV);
+
+  struct pollfd pfd;
+  memset(&pfd, 0, sizeof(pfd));
+  pfd.fd = fds[0];
+  pfd.events = POLLIN | POLLERR;
+
+  const int r = HANDLE_EINTR(poll(&pfd, 1, 0));
+  ASSERT_EQ(r, 1);
+  ASSERT_TRUE(pfd.revents & POLLIN);
+
+  uint32_t len;
+  ASSERT_EQ(read(fds[0], &len, sizeof(len)), sizeof(len));
+  ASSERT_LT(len, 2048);
+  char* filename = reinterpret_cast<char*>(malloc(len + 1));
+  ASSERT_EQ(read(fds[0], filename, len), len);
+  filename[len] = 0;
+  close(fds[0]);
+
+  const std::string minidump_filename = std::string("/tmp/") + filename +
+                                        ".dmp";
+
+  struct stat st;
+  ASSERT_EQ(stat(minidump_filename.c_str(), &st), 0);
+  ASSERT_GT(st.st_size, 0u);
+  unlink(minidump_filename.c_str());
+}
+
+static const unsigned kControlMsgSize =
+    CMSG_SPACE(sizeof(int)) + CMSG_SPACE(sizeof(struct ucred));
+
+static bool
+CrashHandler(const void* crash_context, size_t crash_context_size,
+             void* context) {
+  const int fd = (int) context;
+  int fds[2];
+  pipe(fds);
+
+  struct kernel_msghdr msg = {0};
+  struct kernel_iovec iov;
+  iov.iov_base = const_cast<void*>(crash_context);
+  iov.iov_len = crash_context_size;
+
+  msg.msg_iov = &iov;
+  msg.msg_iovlen = 1;
+  char cmsg[kControlMsgSize];
+  memset(cmsg, 0, kControlMsgSize);
+  msg.msg_control = cmsg;
+  msg.msg_controllen = sizeof(cmsg);
+
+  struct cmsghdr *hdr = CMSG_FIRSTHDR(&msg);
+  hdr->cmsg_level = SOL_SOCKET;
+  hdr->cmsg_type = SCM_RIGHTS;
+  hdr->cmsg_len = CMSG_LEN(sizeof(int));
+  *((int*) CMSG_DATA(hdr)) = fds[1];
+  hdr = CMSG_NXTHDR((struct msghdr*) &msg, hdr);
+  hdr->cmsg_level = SOL_SOCKET;
+  hdr->cmsg_type = SCM_CREDENTIALS;
+  hdr->cmsg_len = CMSG_LEN(sizeof(struct ucred));
+  struct ucred *cred = reinterpret_cast<struct ucred*>(CMSG_DATA(hdr));
+  cred->uid = getuid();
+  cred->gid = getgid();
+  cred->pid = getpid();
+
+  HANDLE_EINTR(sys_sendmsg(fd, &msg, 0));
+  sys_close(fds[1]);
+
+  char b;
+  HANDLE_EINTR(sys_read(fds[0], &b, 1));
+
+  return true;
+}
+
+TEST(ExceptionHandlerTest, ExternalDumper) {
+  int fds[2];
+  ASSERT_NE(socketpair(AF_UNIX, SOCK_DGRAM, 0, fds), -1);
+  static const int on = 1;
+  setsockopt(fds[0], SOL_SOCKET, SO_PASSCRED, &on, sizeof(on));
+  setsockopt(fds[1], SOL_SOCKET, SO_PASSCRED, &on, sizeof(on));
+
+  const pid_t child = fork();
+  if (child == 0) {
+    close(fds[0]);
+    ExceptionHandler handler("/tmp", NULL, NULL, (void*) fds[1], true);
+    handler.set_crash_handler(CrashHandler);
+    *reinterpret_cast<int*>(NULL) = 0;
+  }
+
+  close(fds[1]);
+  struct msghdr msg = {0};
+  struct iovec iov;
+  static const unsigned kCrashContextSize =
+      sizeof(ExceptionHandler::CrashContext);
+  char context[kCrashContextSize];
+  char control[kControlMsgSize];
+  iov.iov_base = context;
+  iov.iov_len = kCrashContextSize;
+  msg.msg_iov = &iov;
+  msg.msg_iovlen = 1;
+  msg.msg_control = control;
+  msg.msg_controllen = kControlMsgSize;
+
+  const ssize_t n = HANDLE_EINTR(recvmsg(fds[0], &msg, 0));
+  ASSERT_EQ(n, kCrashContextSize);
+  ASSERT_EQ(msg.msg_controllen, kControlMsgSize);
+  ASSERT_EQ(msg.msg_flags, 0);
+
+  pid_t crashing_pid = -1;
+  int signal_fd = -1;
+  for (struct cmsghdr *hdr = CMSG_FIRSTHDR(&msg); hdr;
+       hdr = CMSG_NXTHDR(&msg, hdr)) {
+    if (hdr->cmsg_level != SOL_SOCKET)
+      continue;
+    if (hdr->cmsg_type == SCM_RIGHTS) {
+      const unsigned len = hdr->cmsg_len -
+          (((uint8_t*)CMSG_DATA(hdr)) - (uint8_t*)hdr);
+      ASSERT_EQ(len, sizeof(int));
+      signal_fd = *((int *) CMSG_DATA(hdr));
+    } else if (hdr->cmsg_type == SCM_CREDENTIALS) {
+      const struct ucred *cred =
+          reinterpret_cast<struct ucred*>(CMSG_DATA(hdr));
+      crashing_pid = cred->pid;
+    }
+  }
+
+  ASSERT_NE(crashing_pid, -1);
+  ASSERT_NE(signal_fd, -1);
+
+  char templ[] = "/tmp/exception-handler-unittest-XXXXXX";
+  mktemp(templ);
+  ASSERT_TRUE(WriteMinidump(templ, crashing_pid, context,
+                            kCrashContextSize));
+  static const char b = 0;
+  HANDLE_EINTR(write(signal_fd, &b, 1));
+
+  int status;
+  ASSERT_NE(HANDLE_EINTR(waitpid(child, &status, 0)), -1);
+  ASSERT_TRUE(WIFSIGNALED(status));
+  ASSERT_EQ(WTERMSIG(status), SIGSEGV);
+
+  struct stat st;
+  ASSERT_EQ(stat(templ, &st), 0);
+  ASSERT_GT(st.st_size, 0u);
+  unlink(templ);
+}

src/client/linux/handler/linux_thread.cc

@@ -1,411 +0,0 @@
-// Copyright (c) 2006, Google Inc.
-// All rights reserved.
-//
-// Author: Li Liu
-//
-// Redistribution and use in source and binary forms, with or without
-// modification, are permitted provided that the following conditions are
-// met:
-//
-//     * Redistributions of source code must retain the above copyright
-// notice, this list of conditions and the following disclaimer.
-//     * Redistributions in binary form must reproduce the above
-// copyright notice, this list of conditions and the following disclaimer
-// in the documentation and/or other materials provided with the
-// distribution.
-//     * Neither the name of Google Inc. nor the names of its
-// contributors may be used to endorse or promote products derived from
-// this software without specific prior written permission.
-//
-// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-//
-#include <errno.h>
-#include <dirent.h>
-#include <fcntl.h>
-#include <sys/ptrace.h>
-#include <sys/stat.h>
-#include <sys/types.h>
-#include <unistd.h>
-#include <sys/wait.h>
-#include <string.h>
-
-#include <algorithm>
-#include <cassert>
-#include <cstdio>
-#include <cstdlib>
-#include <functional>
-
-#include "client/linux/handler/linux_thread.h"
-
-using namespace google_breakpad;
-
-// This unamed namespace contains helper function.
-namespace {
-
-// Context information for the callbacks when validating address by listing
-// modules.
-struct AddressValidatingContext {
-  uintptr_t address;
-  bool is_mapped;
-
-  AddressValidatingContext() : address(0UL), is_mapped(false) {
-  }
-};
-
-// Convert from string to int.
-bool LocalAtoi(char *s, int *r) {
-  assert(s != NULL);
-  assert(r != NULL);
-  char *endptr = NULL;
-  int ret = strtol(s, &endptr, 10);
-  if (endptr == s)
-    return false;
-  *r = ret;
-  return true;
-}
-
-// Fill the proc path of a thread given its id.
-void FillProcPath(int pid, char *path, int path_size) {
-  char pid_str[32];
-  snprintf(pid_str, sizeof(pid_str), "%d", pid);
-  snprintf(path, path_size, "/proc/%s/", pid_str);
-}
-
-// Read thread info from /proc/$pid/status.
-bool ReadThreadInfo(int pid, ThreadInfo *info) {
-  assert(info != NULL);
-  char status_path[80];
-  // Max size we want to read from status file.
-  static const int kStatusMaxSize = 1024;
-  char status_content[kStatusMaxSize];
-
-  FillProcPath(pid, status_path, sizeof(status_path));
-  strcat(status_path, "status");
-  int fd = open(status_path, O_RDONLY, 0);
-  if (fd < 0)
-    return false;
-
-  int num_read = read(fd, status_content, kStatusMaxSize - 1);
-  if (num_read < 0) {
-    close(fd);
-    return false;
-  }
-  close(fd);
-  status_content[num_read] = '\0';
-
-  char *tgid_start = strstr(status_content, "Tgid:");
-  if (tgid_start)
-    sscanf(tgid_start, "Tgid:\t%d\n", &(info->tgid));
-  else
-    // tgid not supported by kernel??
-    info->tgid = 0;
-
-  tgid_start = strstr(status_content, "Pid:");
-  if (tgid_start) {
-    sscanf(tgid_start, "Pid:\t%d\n" "PPid:\t%d\n", &(info->pid),
-           &(info->ppid));
-    return true;
-  }
-  return false;
-}
-
-// Callback invoked for each mapped module.
-// It use the module's adderss range to validate the address.
-bool IsAddressInModuleCallback(const ModuleInfo &module_info,
-                               void *context) {
-  AddressValidatingContext *addr =
-    reinterpret_cast<AddressValidatingContext *>(context);
-  addr->is_mapped = ((addr->address >= module_info.start_addr) &&
-                     (addr->address <= module_info.start_addr +
-                      module_info.size));
-  return !addr->is_mapped;
-}
-
-#if defined(__i386__) && !defined(NO_FRAME_POINTER)
-void *GetNextFrame(void **last_ebp) {
-  void *sp = *last_ebp;
-  if ((unsigned long)sp == (unsigned long)last_ebp)
-    return NULL;
-  if ((unsigned long)sp & (sizeof(void *) - 1))
-    return NULL;
-  if ((unsigned long)sp - (unsigned long)last_ebp > 100000)
-    return NULL;
-  return sp;
-}
-#else
-void *GetNextFrame(void **last_ebp) {
-  return reinterpret_cast<void*>(last_ebp);
-}
-#endif
-
-// Suspend a thread by attaching to it.
-bool SuspendThread(int pid, void *context) {
-  // This may fail if the thread has just died or debugged.
-  errno = 0;
-  if (ptrace(PTRACE_ATTACH, pid, NULL, NULL) != 0 &&
-      errno != 0) {
-    return false;
-  }
-  while (waitpid(pid, NULL, __WALL) < 0) {
-    if (errno != EINTR) {
-      ptrace(PTRACE_DETACH, pid, NULL, NULL);
-      return false;
-    }
-  }
-  return true;
-}
-
-// Resume a thread by detaching from it.
-bool ResumeThread(int pid, void *context) {
-  return ptrace(PTRACE_DETACH, pid, NULL, NULL) >= 0;
-}
-
-// Callback to get the thread information.
-// Will be called for each thread found.
-bool ThreadInfoCallback(int pid, void *context) {
-  CallbackParam<ThreadCallback> *thread_callback =
-    reinterpret_cast<CallbackParam<ThreadCallback> *>(context);
-  ThreadInfo thread_info;
-  if (ReadThreadInfo(pid, &thread_info) && thread_callback) {
-    // Invoke callback from caller.
-    return (thread_callback->call_back)(thread_info, thread_callback->context);
-  }
-  return false;
-}
-
-}  // namespace
-
-namespace google_breakpad {
-
-LinuxThread::LinuxThread(int pid) : pid_(pid) , threads_suspened_(false) {
-}
-
-LinuxThread::~LinuxThread() {
-  if (threads_suspened_)
-    ResumeAllThreads();
-}
-
-int LinuxThread::SuspendAllThreads() {
-  CallbackParam<PidCallback> callback_param(SuspendThread, NULL);
-  int thread_count = 0;
-  if ((thread_count = IterateProcSelfTask(pid_, &callback_param)) > 0)
-    threads_suspened_ = true;
-  return thread_count;
-}
-
-void LinuxThread::ResumeAllThreads() const {
-  CallbackParam<PidCallback> callback_param(ResumeThread, NULL);
-  IterateProcSelfTask(pid_, &callback_param);
-}
-
-int LinuxThread::GetThreadCount() const {
-  return IterateProcSelfTask(pid_, NULL);
-}
-
-int LinuxThread::ListThreads(
-    CallbackParam<ThreadCallback> *thread_callback_param) const {
-  CallbackParam<PidCallback> callback_param(ThreadInfoCallback,
-                                            thread_callback_param);
-  return IterateProcSelfTask(pid_, &callback_param);
-}
-
-bool LinuxThread::GetRegisters(int pid, user_regs_struct *regs) const {
-  assert(regs);
-  return (regs != NULL &&
-          (ptrace(PTRACE_GETREGS, pid, NULL, regs) == 0) &&
-          errno == 0);
-}
-
-// Get the floating-point registers of a thread.
-// The caller must get the thread pid by ListThreads.
-bool LinuxThread::GetFPRegisters(int pid, user_fpregs_struct *regs) const {
-  assert(regs);
-  return (regs != NULL &&
-          (ptrace(PTRACE_GETREGS, pid, NULL, regs) ==0) &&
-          errno == 0);
-}
-
-bool LinuxThread::GetFPXRegisters(int pid, user_fpxregs_struct *regs) const {
-  assert(regs);
-  return (regs != NULL &&
-          (ptrace(PTRACE_GETFPREGS, pid, NULL, regs) != 0) &&
-          errno == 0);
-}
-
-bool LinuxThread::GetDebugRegisters(int pid, DebugRegs *regs) const {
-  assert(regs);
-
-#define GET_DR(name, num)\
-  name->dr##num = ptrace(PTRACE_PEEKUSER, pid,\
-                         offsetof(struct user, u_debugreg[num]), NULL)
-  GET_DR(regs, 0);
-  GET_DR(regs, 1);
-  GET_DR(regs, 2);
-  GET_DR(regs, 3);
-  GET_DR(regs, 4);
-  GET_DR(regs, 5);
-  GET_DR(regs, 6);
-  GET_DR(regs, 7);
-  return true;
-}
-
-int LinuxThread::GetThreadStackDump(uintptr_t current_ebp,
-                                    uintptr_t current_esp,
-                                    void *buf,
-                                    int buf_size) const {
-  assert(buf);
-  assert(buf_size > 0);
-
-  uintptr_t stack_bottom = GetThreadStackBottom(current_ebp);
-  int size = stack_bottom - current_esp;
-  size = buf_size > size ? size : buf_size;
-  if (size > 0)
-    memcpy(buf, reinterpret_cast<void*>(current_esp), size);
-  return size;
-}
-
-// Get the stack bottom of a thread by stack walking. It works
-// unless the stack has been corrupted or the frame pointer has been omited.
-// This is just a temporary solution before we get better ideas about how
-// this can be done.
-//
-// We will check each frame address by checking into module maps.
-// TODO(liuli): Improve it.
-uintptr_t LinuxThread::GetThreadStackBottom(uintptr_t current_ebp) const {
-  void **sp = reinterpret_cast<void **>(current_ebp);
-  void **previous_sp = sp;
-  while (sp && IsAddressMapped((uintptr_t)sp)) {
-    previous_sp = sp;
-    sp = reinterpret_cast<void **>(GetNextFrame(sp));
-  }
-  return (uintptr_t)previous_sp;
-}
-
-int LinuxThread::GetModuleCount() const {
-  return ListModules(NULL);
-}
-
-int LinuxThread::ListModules(
-    CallbackParam<ModuleCallback> *callback_param) const {
-  char line[512];
-  const char *maps_path = "/proc/self/maps";
-
-  int module_count = 0;
-  FILE *fp = fopen(maps_path, "r");
-  if (fp == NULL)
-    return -1;
-
-  uintptr_t start_addr;
-  uintptr_t end_addr;
-  while (fgets(line, sizeof(line), fp) != NULL) {
-    if (sscanf(line, "%x-%x", &start_addr, &end_addr) == 2) {
-      ModuleInfo module;
-      memset(&module, 0, sizeof(module));
-      module.start_addr = start_addr;
-      module.size = end_addr - start_addr;
-      char *name = NULL;
-      assert(module.size > 0);
-      // Only copy name if the name is a valid path name.
-      if ((name = strchr(line, '/')) != NULL) {
-        // Get rid of the last '\n' in line
-        char *last_return = strchr(line, '\n');
-        if (last_return != NULL)
-          *last_return = '\0';
-        // Keep a space for the ending 0.
-        strncpy(module.name, name, sizeof(module.name) - 1);
-        ++module_count;
-      }
-      if (callback_param &&
-          !(callback_param->call_back(module, callback_param->context)))
-        break;
-    }
-  }
-  fclose(fp);
-  return module_count;
-}
-
-// Parse /proc/$pid/tasks to list all the threads of the process identified by
-// pid.
-int LinuxThread::IterateProcSelfTask(int pid,
-                          CallbackParam<PidCallback> *callback_param) const {
-  char task_path[80];
-  FillProcPath(pid, task_path, sizeof(task_path));
-  strcat(task_path, "task");
-
-  DIR *dir = opendir(task_path);
-  if (dir == NULL)
-    return -1;
-
-  int pid_number = 0;
-  // Record the last pid we've found. This is used for duplicated thread
-  // removal. Duplicated thread information can be found in /proc/$pid/tasks.
-  int last_pid = -1;
-  struct dirent *entry = NULL;
-  while ((entry = readdir(dir)) != NULL) {
-    if (strcmp(entry->d_name, ".") &&
-        strcmp(entry->d_name, "..")) {
-      int tpid = 0;
-      if (LocalAtoi(entry->d_name, &tpid) &&
-          last_pid != tpid) {
-        last_pid = tpid;
-        ++pid_number;
-        // Invoke the callback.
-        if (callback_param &&
-            !(callback_param->call_back)(tpid, callback_param->context))
-          break;
-      }
-    }
-  }
-  closedir(dir);
-  return pid_number;
-}
-
-// Check if the address is a valid virtual address.
-// If the address is in any of the mapped modules, we take it as valid.
-// Otherwise it is invalid.
-bool LinuxThread::IsAddressMapped(uintptr_t address) const {
-  AddressValidatingContext addr;
-  addr.address = address;
-  CallbackParam<ModuleCallback> callback_param(IsAddressInModuleCallback,
-                                               &addr);
-  ListModules(&callback_param);
-  return addr.is_mapped;
-}
-
-bool LinuxThread::FindSigContext(uintptr_t sighandler_ebp,
-                                 struct sigcontext **sig_ctx) {
-  uintptr_t previous_ebp;
-  const int MAX_STACK_DEPTH = 10;
-  int depth_counter = 0;
-
-  do {
-    // We're looking for a |struct sigcontext| as the second parameter
-    // to a signal handler function call.  Luckily, the sigcontext
-    // has an ebp member which should match the ebp pointed to
-    // by the ebp of the signal handler frame.
-    previous_ebp = reinterpret_cast<uintptr_t>(GetNextFrame(
-                                  reinterpret_cast<void**>(sighandler_ebp)));
-    // The stack looks like this:
-    // | previous ebp | previous eip | first param | second param |,
-    // so we need to offset by 3 to get to the second parameter.
-    *sig_ctx = reinterpret_cast<struct sigcontext*>(sighandler_ebp +
-                                                    3 * sizeof(uintptr_t));
-    sighandler_ebp = previous_ebp;
-    depth_counter++;
-  } while(previous_ebp != (*sig_ctx)->ebp && sighandler_ebp != 0 &&
-          IsAddressMapped(sighandler_ebp) && depth_counter < MAX_STACK_DEPTH);
-
-  return previous_ebp == (*sig_ctx)->ebp && previous_ebp != 0;
-}
-
-}  // namespace google_breakpad

src/client/linux/handler/linux_thread.h

@@ -1,204 +0,0 @@
-// Copyright (c) 2006, Google Inc.
-// All rights reserved.
-//
-// Author: Li Liu
-//
-// Redistribution and use in source and binary forms, with or without
-// modification, are permitted provided that the following conditions are
-// met:
-//
-//     * Redistributions of source code must retain the above copyright
-// notice, this list of conditions and the following disclaimer.
-//     * Redistributions in binary form must reproduce the above
-// copyright notice, this list of conditions and the following disclaimer
-// in the documentation and/or other materials provided with the
-// distribution.
-//     * Neither the name of Google Inc. nor the names of its
-// contributors may be used to endorse or promote products derived from
-// this software without specific prior written permission.
-//
-// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-//
-#ifndef CLIENT_LINUX_HANDLER_LINUX_THREAD_H__
-#define CLIENT_LINUX_HANDLER_LINUX_THREAD_H__
-
-#include <stdint.h>
-#include <sys/user.h>
-
-namespace google_breakpad {
-
-// Max module path name length.
-#define kMaxModuleNameLength 256
-
-// Holding information about a thread in the process.
-struct ThreadInfo {
-  // Id of the thread group.
-  int tgid;
-  // Id of the thread.
-  int pid;
-  // Id of the parent process.
-  int ppid;
-};
-
-// Holding infomaton about a module in the process.
-struct ModuleInfo {
-  char name[kMaxModuleNameLength];
-  uintptr_t start_addr;
-  int size;
-};
-
-// Holding debug registers.
-struct DebugRegs {
-  int dr0;
-  int dr1;
-  int dr2;
-  int dr3;
-  int dr4;
-  int dr5;
-  int dr6;
-  int dr7;
-};
-
-// A callback to run when got a thread in the process.
-// Return true will go on to the next thread while return false will stop the
-// iteration.
-typedef bool (*ThreadCallback)(const ThreadInfo &thread_info, void *context);
-
-// A callback to run when a new module is found in the process.
-// Return true will go on to the next module while return false will stop the
-// iteration.
-typedef bool (*ModuleCallback)(const ModuleInfo &module_info, void *context);
-
-// Holding the callback information.
-template<class CallbackFunc>
-struct CallbackParam {
-  // Callback function address.
-  CallbackFunc call_back;
-  // Callback context;
-  void *context;
-
-  CallbackParam() : call_back(NULL), context(NULL) {
-  }
-
-  CallbackParam(CallbackFunc func, void *func_context) :
-    call_back(func), context(func_context) {
-  }
-};
-
-///////////////////////////////////////////////////////////////////////////////
-
-//
-// LinuxThread
-//
-// Provides handy support for operation on linux threads.
-// It uses ptrace to get thread registers. Since ptrace only works in a
-// different process other than the one being ptraced, user of this class
-// should create another process before using the class.
-//
-// The process should be created in the following way:
-//    int cloned_pid = clone(ProcessEntryFunction, stack_address,
-//                           CLONE_VM | CLONE_FILES | CLONE_FS | CLONE_UNTRACED,
-//                           (void*)&arguments);
-//    waitpid(cloned_pid, NULL, __WALL);
-//
-// If CLONE_VM is not used, GetThreadStackBottom, GetThreadStackDump
-// will not work since it just use memcpy to get the stack dump.
-//
-class LinuxThread {
- public:
-  // Create a LinuxThread instance to list all the threads in a process.
-  explicit LinuxThread(int pid);
-  ~LinuxThread();
-
-  // Stop all the threads in the process.
-  // Return the number of stopped threads in the process.
-  // Return -1 means failed to stop threads.
-  int SuspendAllThreads();
-
-  // Resume all the suspended threads.
-  void ResumeAllThreads() const;
-
-  // Get the count of threads in the process.
-  // Return -1 means error.
-  int GetThreadCount() const;
-
-  // List the threads of process.
-  // Whenever there is a thread found, the callback will be invoked to process
-  // the information.
-  // Return number of threads listed.
-  int ListThreads(CallbackParam<ThreadCallback> *thread_callback_param) const;
-
-  // Get the general purpose registers of a thread.
-  // The caller must get the thread pid by ListThreads.
-  bool GetRegisters(int pid, user_regs_struct *regs) const;
-
-  // Get the floating-point registers of a thread.
-  // The caller must get the thread pid by ListThreads.
-  bool GetFPRegisters(int pid, user_fpregs_struct *regs) const;
-
-  // Get all the extended floating-point registers. May not work on all
-  // machines.
-  // The caller must get the thread pid by ListThreads.
-  bool GetFPXRegisters(int pid, user_fpxregs_struct *regs) const;
-
-  // Get the debug registers.
-  // The caller must get the thread pid by ListThreads.
-  bool GetDebugRegisters(int pid, DebugRegs *regs) const;
-
-  // Get the stack memory dump.
-  int GetThreadStackDump(uintptr_t current_ebp,
-                         uintptr_t current_esp,
-                         void *buf,
-                         int buf_size) const;
-
-  // Get the module count of the current process.
-  int GetModuleCount() const;
-
-  // Get the mapped modules in the address space.
-  // Whenever a module is found, the callback will be invoked to process the
-  // information.
-  // Return how may modules are found.
-  int ListModules(CallbackParam<ModuleCallback> *callback_param) const;
-
-  // Get the bottom of the stack from ebp.
-  uintptr_t GetThreadStackBottom(uintptr_t current_ebp) const;
-
-  // Finds a sigcontext on the stack given the ebp of our signal handler.
-  bool FindSigContext(uintptr_t sighandler_ebp, struct sigcontext **sig_ctx);
-
- private:
-  // This callback will run when a new thread has been found.
-  typedef bool (*PidCallback)(int pid, void *context);
-
-  // Read thread information from /proc/$pid/task.
-  // Whenever a thread has been found, and callback will be invoked with
-  // the pid of the thread.
-  // Return number of threads found.
-  // Return -1 means the directory doesn't exist.
-  int IterateProcSelfTask(int pid,
-                          CallbackParam<PidCallback> *callback_param) const;
-
-  // Check if the address is a valid virtual address.
-  bool IsAddressMapped(uintptr_t address) const;
-
- private:
-  // The pid of the process we are listing threads.
-  int pid_;
-
-  // Mark if we have suspended the threads.
-  bool threads_suspened_;
-};
-
-}  // namespace google_breakpad
-
-#endif  // CLIENT_LINUX_HANDLER_LINUX_THREAD_H__

src/client/linux/handler/linux_thread_test.cc

@@ -1,224 +0,0 @@
-// Copyright (c) 2006, Google Inc.
-// All rights reserved.
-//
-// Author: Li Liu
-//
-// Redistribution and use in source and binary forms, with or without
-// modification, are permitted provided that the following conditions are
-// met:
-//
-//     * Redistributions of source code must retain the above copyright
-// notice, this list of conditions and the following disclaimer.
-//     * Redistributions in binary form must reproduce the above
-// copyright notice, this list of conditions and the following disclaimer
-// in the documentation and/or other materials provided with the
-// distribution.
-//     * Neither the name of Google Inc. nor the names of its
-// contributors may be used to endorse or promote products derived from
-// this software without specific prior written permission.
-//
-// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-
-#include <pthread.h>
-#include <sys/types.h>
-#include <unistd.h>
-#include <sys/wait.h>
-
-#include <cstdio>
-#include <cstdlib>
-#include <cstring>
-
-#include "client/linux/handler/linux_thread.h"
-
-using namespace google_breakpad;
-
-// Thread use this to see if it should stop working.
-static bool should_exit = false;
-
-static void foo2(int *a) {
-  // Stack variable, used for debugging stack dumps.
-  int c = 0xcccccccc;
-  c = c;
-  while (!should_exit)
-    sleep(1);
-}
-
-static void foo() {
-  // Stack variable, used for debugging stack dumps.
-  int a = 0xaaaaaaaa;
-  foo2(&a);
-}
-
-static void *thread_main(void *) {
-  // Stack variable, used for debugging stack dumps.
-  int b = 0xbbbbbbbb;
-  b = b;
-  while (!should_exit) {
-    foo();
-  }
-  return NULL;
-}
-
-static void CreateThreads(int num) {
-  pthread_t handle;
-  for (int i = 0; i < num; i++) {
-    if (0 != pthread_create(&handle, NULL, thread_main, NULL))
-      fprintf(stderr, "Failed to create thread.\n");
-    else
-      pthread_detach(handle);
-  }
-}
-
-static bool ProcessOneModule(const struct ModuleInfo &module_info,
-                             void *context) {
-  printf("0x%x[%8d]         %s\n", module_info.start_addr, module_info.size,
-         module_info.name);
-  return true;
-}
-
-static bool ProcessOneThread(const struct ThreadInfo &thread_info,
-                             void *context) {
-  printf("\n\nPID: %d, TGID: %d, PPID: %d\n",
-         thread_info.pid,
-         thread_info.tgid,
-         thread_info.ppid);
-
-  struct user_regs_struct regs;
-  struct user_fpregs_struct fp_regs;
-  struct user_fpxregs_struct fpx_regs;
-  struct DebugRegs dbg_regs;
-
-  LinuxThread *threads = reinterpret_cast<LinuxThread *>(context);
-  memset(&regs, 0, sizeof(regs));
-  if (threads->GetRegisters(thread_info.pid, &regs)) {
-    printf("  gs                           = 0x%lx\n", regs.xgs);
-    printf("  fs                           = 0x%lx\n", regs.xfs);
-    printf("  es                           = 0x%lx\n", regs.xes);
-    printf("  ds                           = 0x%lx\n", regs.xds);
-    printf("  edi                          = 0x%lx\n", regs.edi);
-    printf("  esi                          = 0x%lx\n", regs.esi);
-    printf("  ebx                          = 0x%lx\n", regs.ebx);
-    printf("  edx                          = 0x%lx\n", regs.edx);
-    printf("  ecx                          = 0x%lx\n", regs.ecx);
-    printf("  eax                          = 0x%lx\n", regs.eax);
-    printf("  ebp                          = 0x%lx\n", regs.ebp);
-    printf("  eip                          = 0x%lx\n", regs.eip);
-    printf("  cs                           = 0x%lx\n", regs.xcs);
-    printf("  eflags                       = 0x%lx\n", regs.eflags);
-    printf("  esp                          = 0x%lx\n", regs.esp);
-    printf("  ss                           = 0x%lx\n", regs.xss);
-  } else {
-    fprintf(stderr, "ERROR: Failed to get general purpose registers\n");
-  }
-  memset(&fp_regs, 0, sizeof(fp_regs));
-  if (threads->GetFPRegisters(thread_info.pid, &fp_regs)) {
-    printf("\n Floating point registers:\n");
-    printf("  fctl                         = 0x%lx\n", fp_regs.cwd);
-    printf("  fstat                        = 0x%lx\n", fp_regs.swd);
-    printf("  ftag                         = 0x%lx\n", fp_regs.twd);
-    printf("  fioff                        = 0x%lx\n", fp_regs.fip);
-    printf("  fiseg                        = 0x%lx\n", fp_regs.fcs);
-    printf("  fooff                        = 0x%lx\n", fp_regs.foo);
-    printf("  foseg                        = 0x%lx\n", fp_regs.fos);
-    int st_space_size = sizeof(fp_regs.st_space) / sizeof(fp_regs.st_space[0]);
-    printf("  st_space[%2d]                 = 0x", st_space_size);
-    for (int i = 0; i < st_space_size; ++i)
-      printf("%02lx", fp_regs.st_space[i]);
-    printf("\n");
-  } else {
-    fprintf(stderr, "ERROR: Failed to get floating-point registers\n");
-  }
-  memset(&fpx_regs, 0, sizeof(fpx_regs));
-  if (threads->GetFPXRegisters(thread_info.pid, &fpx_regs)) {
-    printf("\n Extended floating point registers:\n");
-    printf("  fctl                         = 0x%x\n", fpx_regs.cwd);
-    printf("  fstat                        = 0x%x\n", fpx_regs.swd);
-    printf("  ftag                         = 0x%x\n", fpx_regs.twd);
-    printf("  fioff                        = 0x%lx\n", fpx_regs.fip);
-    printf("  fiseg                        = 0x%lx\n", fpx_regs.fcs);
-    printf("  fooff                        = 0x%lx\n", fpx_regs.foo);
-    printf("  foseg                        = 0x%lx\n", fpx_regs.fos);
-    printf("  fop                          = 0x%x\n", fpx_regs.fop);
-    printf("  mxcsr                        = 0x%lx\n", fpx_regs.mxcsr);
-    int space_size = sizeof(fpx_regs.st_space) / sizeof(fpx_regs.st_space[0]);
-    printf("  st_space[%2d]                 = 0x", space_size);
-    for (int i = 0; i < space_size; ++i)
-      printf("%02lx", fpx_regs.st_space[i]);
-    printf("\n");
-    space_size = sizeof(fpx_regs.xmm_space) / sizeof(fpx_regs.xmm_space[0]);
-    printf("  xmm_space[%2d]                = 0x", space_size);
-    for (int i = 0; i < space_size; ++i)
-      printf("%02lx", fpx_regs.xmm_space[i]);
-    printf("\n");
-  }
-  if (threads->GetDebugRegisters(thread_info.pid, &dbg_regs)) {
-    printf("\n Debug registers:\n");
-    printf("  dr0                          = 0x%x\n", dbg_regs.dr0);
-    printf("  dr1                          = 0x%x\n", dbg_regs.dr1);
-    printf("  dr2                          = 0x%x\n", dbg_regs.dr2);
-    printf("  dr3                          = 0x%x\n", dbg_regs.dr3);
-    printf("  dr4                          = 0x%x\n", dbg_regs.dr4);
-    printf("  dr5                          = 0x%x\n", dbg_regs.dr5);
-    printf("  dr6                          = 0x%x\n", dbg_regs.dr6);
-    printf("  dr7                          = 0x%x\n", dbg_regs.dr7);
-    printf("\n");
-  }
-  if (regs.esp != 0) {
-    // Print the stack content.
-    int size = 1024 * 2;
-    char *buf = new char[size];
-    size = threads->GetThreadStackDump(regs.ebp,
-                                       regs.esp,
-                                      (void*)buf, size);
-    printf(" Stack content:                 = 0x");
-    size /= sizeof(unsigned long);
-    unsigned long *p_buf = (unsigned long *)(buf);
-    for (int i = 0; i < size; i += 1)
-      printf("%.8lx ", p_buf[i]);
-    delete []buf;
-    printf("\n");
-  }
-  return true;
-}
-
-static int PrintAllThreads(void *argument) {
-  int pid = (int)argument;
-
-  LinuxThread threads(pid);
-  int total_thread = threads.SuspendAllThreads();
-  printf("There are %d threads in the process: %d\n", total_thread, pid);
-  int total_module = threads.GetModuleCount();
-  printf("There are %d modules in the process: %d\n", total_module, pid);
-  CallbackParam<ModuleCallback> module_callback(ProcessOneModule, &threads);
-  threads.ListModules(&module_callback);
-  CallbackParam<ThreadCallback> thread_callback(ProcessOneThread, &threads);
-  threads.ListThreads(&thread_callback);
-  return 0;
-}
-
-int main(int argc, char **argv) {
-  int pid = getpid();
-  printf("Main thread is %d\n", pid);
-  CreateThreads(1);
-  // Create stack for the process.
-  char *stack = new char[1024 * 100];
-  int cloned_pid = clone(PrintAllThreads, stack + 1024 * 100,
-                           CLONE_VM | CLONE_FILES | CLONE_FS | CLONE_UNTRACED,
-                           (void*)getpid());
-  waitpid(cloned_pid, NULL, __WALL);
-  should_exit = true;
-  printf("Test finished.\n");
-
-  delete []stack;
-  return 0;
-}

src/client/linux/handler/minidump_generator.cc

@@ -1,816 +0,0 @@
-// Copyright (c) 2006, Google Inc.
-// All rights reserved.
-//
-// Author: Li Liu
-//
-// Redistribution and use in source and binary forms, with or without
-// modification, are permitted provided that the following conditions are
-// met:
-//
-//     * Redistributions of source code must retain the above copyright
-// notice, this list of conditions and the following disclaimer.
-//     * Redistributions in binary form must reproduce the above
-// copyright notice, this list of conditions and the following disclaimer
-// in the documentation and/or other materials provided with the
-// distribution.
-//     * Neither the name of Google Inc. nor the names of its
-// contributors may be used to endorse or promote products derived from
-// this software without specific prior written permission.
-//
-// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-
-#include <fcntl.h>
-#include <pthread.h>
-#include <signal.h>
-#include <sys/stat.h>
-#include <sys/types.h>
-#include <unistd.h>
-#include <sys/utsname.h>
-#include <sys/wait.h>
-
-#include <cstdlib>
-#include <cstdio>
-#include <ctime>
-#include <string.h>
-
-#include "common/linux/file_id.h"
-#include "client/linux/handler/linux_thread.h"
-#include "client/minidump_file_writer.h"
-#include "client/minidump_file_writer-inl.h"
-#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 {
-
-using namespace google_breakpad;
-
-// Argument for the writer function.
-struct WriterArgument {
-  MinidumpFileWriter *minidump_writer;
-
-  // Context for the callback.
-  void *version_context;
-
-  // Pid of the thread who called WriteMinidumpToFile
-  int requester_pid;
-
-  // The stack bottom of the thread which caused the dump.
-  // Mainly used to find the thread id of the crashed thread since signal
-  // handler may not be called in the thread who caused it.
-  uintptr_t crashed_stack_bottom;
-
-  // Pid of the crashing thread.
-  int crashed_pid;
-
-  // Signal number when crash happed. Can be 0 if this is a requested dump.
-  int signo;
-
-  // The ebp of the signal handler frame.  Can be zero if this
-  // is a requested dump.
-  uintptr_t sighandler_ebp;
-
-  // Signal context when crash happed. Can be NULL if this is a requested dump.
-  // This is actually an out parameter, but it will be filled in at the start
-  // of the writer thread.
-  struct sigcontext *sig_ctx;
-
-  // Used to get information about the threads.
-  LinuxThread *thread_lister;
-};
-
-// Holding context information for the callback of finding the crashing thread.
-struct FindCrashThreadContext {
-  const LinuxThread *thread_lister;
-  uintptr_t crashing_stack_bottom;
-  int crashing_thread_pid;
-
-  FindCrashThreadContext() :
-    thread_lister(NULL),
-    crashing_stack_bottom(0UL),
-    crashing_thread_pid(-1) {
-  }
-};
-
-// Callback for list threads.
-// It will compare the stack bottom of the provided thread with the stack
-// bottom of the crashed thread, it they are eqaul, this is thread is the one
-// who crashed.
-bool IsThreadCrashedCallback(const ThreadInfo &thread_info, void *context) {
-  FindCrashThreadContext *crashing_context =
-    static_cast<FindCrashThreadContext *>(context);
-  const LinuxThread *thread_lister = crashing_context->thread_lister;
-  struct user_regs_struct regs;
-  if (thread_lister->GetRegisters(thread_info.pid, &regs)) {
-    uintptr_t last_ebp = regs.ebp;
-    uintptr_t stack_bottom = thread_lister->GetThreadStackBottom(last_ebp);
-    if (stack_bottom > last_ebp &&
-        stack_bottom == crashing_context->crashing_stack_bottom) {
-      // Got it. Stop iteration.
-      crashing_context->crashing_thread_pid = thread_info.pid;
-      return false;
-    }
-  }
-  return true;
-}
-
-// Find the crashing thread id.
-// This is done based on stack bottom comparing.
-int FindCrashingThread(uintptr_t crashing_stack_bottom,
-                       int requester_pid,
-                       const LinuxThread *thread_lister) {
-  FindCrashThreadContext context;
-  context.thread_lister = thread_lister;
-  context.crashing_stack_bottom = crashing_stack_bottom;
-  CallbackParam<ThreadCallback> callback_param(IsThreadCrashedCallback,
-                                               &context);
-  thread_lister->ListThreads(&callback_param);
-  return context.crashing_thread_pid;
-}
-
-// Write the thread stack info minidump.
-bool WriteThreadStack(uintptr_t last_ebp,
-                      uintptr_t last_esp,
-                      const LinuxThread *thread_lister,
-                      UntypedMDRVA *memory,
-                      MDMemoryDescriptor *loc) {
-  // Maximum stack size for a thread.
-  uintptr_t stack_bottom = thread_lister->GetThreadStackBottom(last_ebp);
-  if (stack_bottom > last_esp) {
-    int size = stack_bottom - last_esp;
-    if (size > 0) {
-      if (!memory->Allocate(size))
-        return false;
-      memory->Copy(reinterpret_cast<void*>(last_esp), size);
-      loc->start_of_memory_range = 0 | last_esp;
-      loc->memory = memory->location();
-    }
-    return true;
-  }
-  return false;
-}
-
-// Write CPU context based on signal context.
-bool WriteContext(MDRawContextX86 *context, const struct sigcontext *sig_ctx,
-                  const DebugRegs *debug_regs) {
-  assert(sig_ctx != NULL);
-  context->context_flags = MD_CONTEXT_X86_FULL;
-  context->gs = sig_ctx->gs;
-  context->fs = sig_ctx->fs;
-  context->es = sig_ctx->es;
-  context->ds = sig_ctx->ds;
-  context->cs = sig_ctx->cs;
-  context->ss = sig_ctx->ss;
-  context->edi = sig_ctx->edi;
-  context->esi = sig_ctx->esi;
-  context->ebp = sig_ctx->ebp;
-  context->esp = sig_ctx->esp;
-  context->ebx = sig_ctx->ebx;
-  context->edx = sig_ctx->edx;
-  context->ecx = sig_ctx->ecx;
-  context->eax = sig_ctx->eax;
-  context->eip = sig_ctx->eip;
-  context->eflags = sig_ctx->eflags;
-  if (sig_ctx->fpstate != NULL) {
-    context->context_flags = MD_CONTEXT_X86_FULL |
-      MD_CONTEXT_X86_FLOATING_POINT;
-    context->float_save.control_word = sig_ctx->fpstate->cw;
-    context->float_save.status_word = sig_ctx->fpstate->sw;
-    context->float_save.tag_word = sig_ctx->fpstate->tag;
-    context->float_save.error_offset = sig_ctx->fpstate->ipoff;
-    context->float_save.error_selector = sig_ctx->fpstate->cssel;
-    context->float_save.data_offset = sig_ctx->fpstate->dataoff;
-    context->float_save.data_selector = sig_ctx->fpstate->datasel;
-    memcpy(context->float_save.register_area, sig_ctx->fpstate->_st,
-           sizeof(context->float_save.register_area));
-  }
-
-  if (debug_regs != NULL) {
-    context->context_flags |= MD_CONTEXT_X86_DEBUG_REGISTERS;
-    context->dr0 = debug_regs->dr0;
-    context->dr1 = debug_regs->dr1;
-    context->dr2 = debug_regs->dr2;
-    context->dr3 = debug_regs->dr3;
-    context->dr6 = debug_regs->dr6;
-    context->dr7 = debug_regs->dr7;
-  }
-  return true;
-}
-
-// Write CPU context based on provided registers.
-bool WriteContext(MDRawContextX86 *context,
-                  const struct user_regs_struct *regs,
-                  const struct user_fpregs_struct *fp_regs,
-                  const DebugRegs *dbg_regs) {
-  if (!context || !regs)
-    return false;
-
-  context->context_flags = MD_CONTEXT_X86_FULL;
-
-  context->cs = regs->xcs;
-  context->ds = regs->xds;
-  context->es = regs->xes;
-  context->fs = regs->xfs;
-  context->gs = regs->xgs;
-  context->ss = regs->xss;
-  context->edi = regs->edi;
-  context->esi = regs->esi;
-  context->ebx = regs->ebx;
-  context->edx = regs->edx;
-  context->ecx = regs->ecx;
-  context->eax = regs->eax;
-  context->ebp = regs->ebp;
-  context->eip = regs->eip;
-  context->esp = regs->esp;
-  context->eflags = regs->eflags;
-
-  if (dbg_regs != NULL) {
-    context->context_flags |= MD_CONTEXT_X86_DEBUG_REGISTERS;
-    context->dr0 = dbg_regs->dr0;
-    context->dr1 = dbg_regs->dr1;
-    context->dr2 = dbg_regs->dr2;
-    context->dr3 = dbg_regs->dr3;
-    context->dr6 = dbg_regs->dr6;
-    context->dr7 = dbg_regs->dr7;
-  }
-
-  if (fp_regs != NULL) {
-    context->context_flags |= MD_CONTEXT_X86_FLOATING_POINT;
-    context->float_save.control_word = fp_regs->cwd;
-    context->float_save.status_word = fp_regs->swd;
-    context->float_save.tag_word = fp_regs->twd;
-    context->float_save.error_offset = fp_regs->fip;
-    context->float_save.error_selector = fp_regs->fcs;
-    context->float_save.data_offset = fp_regs->foo;
-    context->float_save.data_selector = fp_regs->fos;
-    context->float_save.data_selector = fp_regs->fos;
-
-    memcpy(context->float_save.register_area, fp_regs->st_space,
-           sizeof(context->float_save.register_area));
-  }
-  return true;
-}
-
-// Write information about a crashed thread.
-// When a thread crash, kernel will write something on the stack for processing
-// signal. This makes the current stack not reliable, and our stack walker
-// won't figure out the whole call stack for this. So we write the stack at the
-// time of the crash into the minidump file, not the current stack.
-bool WriteCrashedThreadStream(MinidumpFileWriter *minidump_writer,
-                       const WriterArgument *writer_args,
-                       const ThreadInfo &thread_info,
-                       MDRawThread *thread) {
-  assert(writer_args->sig_ctx != NULL);
-
-  thread->thread_id = thread_info.pid;
-
-  UntypedMDRVA memory(minidump_writer);
-  if (!WriteThreadStack(writer_args->sig_ctx->ebp,
-                        writer_args->sig_ctx->esp,
-                        writer_args->thread_lister,
-                        &memory,
-                        &thread->stack))
-    return false;
-
-  TypedMDRVA<MDRawContextX86> context(minidump_writer);
-  if (!context.Allocate())
-    return false;
-  thread->thread_context = context.location();
-  memset(context.get(), 0, sizeof(MDRawContextX86));
-  return WriteContext(context.get(), writer_args->sig_ctx, NULL);
-}
-
-// Write information about a thread.
-// This function only processes thread running normally at the crash.
-bool WriteThreadStream(MinidumpFileWriter *minidump_writer,
-                       const LinuxThread *thread_lister,
-                       const ThreadInfo &thread_info,
-                       MDRawThread *thread) {
-  thread->thread_id = thread_info.pid;
-
-  struct user_regs_struct regs;
-  memset(&regs, 0, sizeof(regs));
-  if (!thread_lister->GetRegisters(thread_info.pid, &regs)) {
-    perror(NULL);
-    return false;
-  }
-
-  UntypedMDRVA memory(minidump_writer);
-  if (!WriteThreadStack(regs.ebp,
-                   regs.esp,
-                   thread_lister,
-                   &memory,
-                   &thread->stack))
-    return false;
-
-  struct user_fpregs_struct fp_regs;
-  DebugRegs dbg_regs;
-  memset(&fp_regs, 0, sizeof(fp_regs));
-  // Get all the registers.
-  thread_lister->GetFPRegisters(thread_info.pid, &fp_regs);
-  thread_lister->GetDebugRegisters(thread_info.pid, &dbg_regs);
-
-  // Write context
-  TypedMDRVA<MDRawContextX86> context(minidump_writer);
-  if (!context.Allocate())
-    return false;
-  thread->thread_context = context.location();
-  memset(context.get(), 0, sizeof(MDRawContextX86));
-  return WriteContext(context.get(), &regs, &fp_regs, &dbg_regs);
-}
-
-bool WriteCPUInformation(MDRawSystemInfo *sys_info) {
-  const char *proc_cpu_path = "/proc/cpuinfo";
-  char line[128];
-  char vendor_id[13];
-  const char vendor_id_name[] = "vendor_id";
-  const size_t vendor_id_name_length = sizeof(vendor_id_name) - 1;
-
-  struct CpuInfoEntry {
-    const char *info_name;
-    int value;
-  } cpu_info_table[] = {
-    { "processor", -1 },
-    { "model", 0 },
-    { "stepping",  0 },
-    { "cpuid level", 0 },
-    { NULL, -1 },
-  };
-
-  memset(vendor_id, 0, sizeof(vendor_id));
-
-  FILE *fp = fopen(proc_cpu_path, "r");
-  if (fp != NULL) {
-    while (fgets(line, sizeof(line), fp)) {
-      CpuInfoEntry *entry = &cpu_info_table[0];
-      while (entry->info_name != NULL) {
-        if (!strncmp(line, entry->info_name, strlen(entry->info_name))) {
-          char *value = strchr(line, ':');
-          value++;
-          if (value != NULL)
-            sscanf(value, " %d", &(entry->value));
-        }
-        entry++;
-      }
-
-      // special case for vendor_id
-      if (!strncmp(line, vendor_id_name, vendor_id_name_length)) {
-        char *value = strchr(line, ':');
-        if (value == NULL)
-          continue;
-
-        value++;
-        while (*value && isspace(*value))
-          value++;
-        if (*value) {
-          size_t length = strlen(value);
-          // we don't want the trailing newline
-          if (value[length - 1] == '\n')
-            length--;
-          // ensure we have space for the value
-          if (length < sizeof(vendor_id))
-            strncpy(vendor_id, value, length);
-        }
-      }
-    }
-    fclose(fp);
-  }
-
-  // /proc/cpuinfo contains cpu id, change it into number by adding one.
-  cpu_info_table[0].value++;
-
-  sys_info->number_of_processors = cpu_info_table[0].value;
-  sys_info->processor_level      = cpu_info_table[3].value;
-  sys_info->processor_revision   = cpu_info_table[1].value << 8 |
-                                   cpu_info_table[2].value;
-
-  sys_info->processor_architecture = MD_CPU_ARCHITECTURE_UNKNOWN;
-  struct utsname uts;
-  if (uname(&uts) == 0) {
-    // Match i*86 and x86* as X86 architecture.
-    if ((strstr(uts.machine, "x86") == uts.machine) ||
-        (strlen(uts.machine) == 4 &&
-         uts.machine[0] == 'i' &&
-         uts.machine[2] == '8' &&
-         uts.machine[3] == '6')) {
-      sys_info->processor_architecture = MD_CPU_ARCHITECTURE_X86;
-      if (vendor_id[0] != '\0')
-        memcpy(sys_info->cpu.x86_cpu_info.vendor_id, vendor_id,
-               sizeof(sys_info->cpu.x86_cpu_info.vendor_id));
-    }
-  }
-  return true;
-}
-
-bool WriteOSInformation(MinidumpFileWriter *minidump_writer,
-                        MDRawSystemInfo *sys_info) {
-  sys_info->platform_id = MD_OS_LINUX;
-
-  struct utsname uts;
-  if (uname(&uts) == 0) {
-    char os_version[512];
-    size_t space_left = sizeof(os_version);
-    memset(os_version, 0, space_left);
-    const char *os_info_table[] = {
-      uts.sysname,
-      uts.release,
-      uts.version,
-      uts.machine,
-      "GNU/Linux",
-      NULL
-    };
-    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) {
-        strcat(os_version, " ");
-        space_left--;
-      }
-      if (space_left > strlen(*cur_os_info)) {
-        strcat(os_version, *cur_os_info);
-        space_left -= strlen(*cur_os_info);
-      } else {
-        break;
-      }
-    }
-
-    MDLocationDescriptor location;
-    if (!minidump_writer->WriteString(os_version, 0, &location))
-      return false;
-    sys_info->csd_version_rva = location.rva;
-  }
-  return true;
-}
-
-// Callback context for get writting thread information.
-struct ThreadInfoCallbackCtx {
-  MinidumpFileWriter *minidump_writer;
-  const WriterArgument *writer_args;
-  TypedMDRVA<MDRawThreadList> *list;
-  int thread_index;
-};
-
-// Callback run for writing threads information in the process.
-bool ThreadInfomationCallback(const ThreadInfo &thread_info,
-                                 void *context) {
-  ThreadInfoCallbackCtx *callback_context =
-    static_cast<ThreadInfoCallbackCtx *>(context);
-  bool success = true;
-  MDRawThread thread;
-  memset(&thread, 0, sizeof(MDRawThread));
-  if (thread_info.pid != callback_context->writer_args->crashed_pid ||
-      callback_context->writer_args->sig_ctx == NULL) {
-    success = WriteThreadStream(callback_context->minidump_writer,
-                           callback_context->writer_args->thread_lister,
-                           thread_info, &thread);
-  } else {
-    success = WriteCrashedThreadStream(callback_context->minidump_writer,
-                                       callback_context->writer_args,
-                                       thread_info, &thread);
-  }
-  if (success) {
-    callback_context->list->CopyIndexAfterObject(
-        callback_context->thread_index++,
-        &thread, sizeof(MDRawThread));
-  }
-  return success;
-}
-
-// Stream writers
-bool WriteThreadListStream(MinidumpFileWriter *minidump_writer,
-                           const WriterArgument *writer_args,
-                           MDRawDirectory *dir) {
-  // Get the thread information.
-  const LinuxThread *thread_lister = writer_args->thread_lister;
-  int thread_count = thread_lister->GetThreadCount();
-  if (thread_count < 0)
-    return false;
-  TypedMDRVA<MDRawThreadList> list(minidump_writer);
-  if (!list.AllocateObjectAndArray(thread_count, sizeof(MDRawThread)))
-    return false;
-  dir->stream_type = MD_THREAD_LIST_STREAM;
-  dir->location = list.location();
-  list.get()->number_of_threads = thread_count;
-
-  ThreadInfoCallbackCtx context;
-  context.minidump_writer = minidump_writer;
-  context.writer_args = writer_args;
-  context.list = &list;
-  context.thread_index = 0;
-  CallbackParam<ThreadCallback> callback_param(ThreadInfomationCallback,
-                                               &context);
-  int written = thread_lister->ListThreads(&callback_param);
-  return written == thread_count;
-}
-
-bool WriteCVRecord(MinidumpFileWriter *minidump_writer,
-                   MDRawModule *module,
-                   const char *module_path) {
-  TypedMDRVA<MDCVInfoPDB70> cv(minidump_writer);
-
-  // Only return the last path component of the full module path
-  const char *module_name = strrchr(module_path, '/');
-  // Increment past the slash
-  if (module_name)
-    ++module_name;
-  else
-    module_name = "<Unknown>";
-
-  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, 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;
-
-  // Get the module identifier
-  FileID file_id(module_path);
-  unsigned char identifier[16];
-
-  if (file_id.ElfFileIdentifier(identifier)) {
-    cv_ptr->signature.data1 = (uint32_t)identifier[0] << 24 |
-      (uint32_t)identifier[1] << 16 | (uint32_t)identifier[2] << 8 |
-      (uint32_t)identifier[3];
-    cv_ptr->signature.data2 = (uint32_t)identifier[4] << 8 | identifier[5];
-    cv_ptr->signature.data3 = (uint32_t)identifier[6] << 8 | identifier[7];
-    cv_ptr->signature.data4[0] = identifier[8];
-    cv_ptr->signature.data4[1] = identifier[9];
-    cv_ptr->signature.data4[2] = identifier[10];
-    cv_ptr->signature.data4[3] = identifier[11];
-    cv_ptr->signature.data4[4] = identifier[12];
-    cv_ptr->signature.data4[5] = identifier[13];
-    cv_ptr->signature.data4[6] = identifier[14];
-    cv_ptr->signature.data4[7] = identifier[15];
-  }
-  return true;
-}
-
-struct ModuleInfoCallbackCtx {
-  MinidumpFileWriter *minidump_writer;
-  const WriterArgument *writer_args;
-  TypedMDRVA<MDRawModuleList> *list;
-  int module_index;
-};
-
-bool ModuleInfoCallback(const ModuleInfo &module_info,
-                           void *context) {
-  ModuleInfoCallbackCtx *callback_context =
-    static_cast<ModuleInfoCallbackCtx *>(context);
-  // Skip those modules without name, or those that are not modules.
-  if (strlen(module_info.name) == 0 ||
-      !strchr(module_info.name, '/'))
-    return true;
-
-  MDRawModule module;
-  memset(&module, 0, sizeof(module));
-  MDLocationDescriptor loc;
-  if (!callback_context->minidump_writer->WriteString(module_info.name, 0,
-                                                      &loc))
-    return false;
-  module.base_of_image = (u_int64_t)module_info.start_addr;
-  module.size_of_image = module_info.size;
-  module.module_name_rva = loc.rva;
-
-  if (!WriteCVRecord(callback_context->minidump_writer, &module,
-                     module_info.name))
-    return false;
-  callback_context->list->CopyIndexAfterObject(
-      callback_context->module_index++, &module, MD_MODULE_SIZE);
-  return true;
-}
-
-bool WriteModuleListStream(MinidumpFileWriter *minidump_writer,
-                           const WriterArgument *writer_args,
-                           MDRawDirectory *dir) {
-  TypedMDRVA<MDRawModuleList> list(minidump_writer);
-  int module_count  = writer_args->thread_lister->GetModuleCount();
-  if (module_count <= 0 ||
-      !list.AllocateObjectAndArray(module_count, MD_MODULE_SIZE))
-    return false;
-  dir->stream_type = MD_MODULE_LIST_STREAM;
-  dir->location = list.location();
-  list.get()->number_of_modules = module_count;
-  ModuleInfoCallbackCtx context;
-  context.minidump_writer = minidump_writer;
-  context.writer_args = writer_args;
-  context.list = &list;
-  context.module_index = 0;
-  CallbackParam<ModuleCallback> callback(ModuleInfoCallback, &context);
-  return writer_args->thread_lister->ListModules(&callback) == module_count;
-}
-
-bool WriteSystemInfoStream(MinidumpFileWriter *minidump_writer,
-                           const WriterArgument *writer_args,
-                           MDRawDirectory *dir) {
-  TypedMDRVA<MDRawSystemInfo> sys_info(minidump_writer);
-  if (!sys_info.Allocate())
-    return false;
-  dir->stream_type = MD_SYSTEM_INFO_STREAM;
-  dir->location = sys_info.location();
-
-  return WriteCPUInformation(sys_info.get()) &&
-         WriteOSInformation(minidump_writer, sys_info.get());
-}
-
-bool WriteExceptionStream(MinidumpFileWriter *minidump_writer,
-                           const WriterArgument *writer_args,
-                           MDRawDirectory *dir) {
-  // This happenes when this is not a crash, but a requested dump.
-  if (writer_args->sig_ctx == NULL)
-    return false;
-
-  TypedMDRVA<MDRawExceptionStream> exception(minidump_writer);
-  if (!exception.Allocate())
-    return false;
-
-  dir->stream_type = MD_EXCEPTION_STREAM;
-  dir->location = exception.location();
-  exception.get()->thread_id = writer_args->crashed_pid;
-  exception.get()->exception_record.exception_code = writer_args->signo;
-  exception.get()->exception_record.exception_flags = 0;
-  if (writer_args->sig_ctx != NULL) {
-    exception.get()->exception_record.exception_address =
-      writer_args->sig_ctx->eip;
-  } else {
-    return true;
-  }
-
-  // Write context of the exception.
-  TypedMDRVA<MDRawContextX86> context(minidump_writer);
-  if (!context.Allocate())
-    return false;
-  exception.get()->thread_context = context.location();
-  memset(context.get(), 0, sizeof(MDRawContextX86));
-  return WriteContext(context.get(), writer_args->sig_ctx, NULL);
-}
-
-bool WriteMiscInfoStream(MinidumpFileWriter *minidump_writer,
-                           const WriterArgument *writer_args,
-                           MDRawDirectory *dir) {
-  TypedMDRVA<MDRawMiscInfo> info(minidump_writer);
-  if (!info.Allocate())
-    return false;
-
-  dir->stream_type = MD_MISC_INFO_STREAM;
-  dir->location = info.location();
-  info.get()->size_of_info = sizeof(MDRawMiscInfo);
-  info.get()->flags1 = MD_MISCINFO_FLAGS1_PROCESS_ID;
-  info.get()->process_id = writer_args->requester_pid;
-
-  return true;
-}
-
-bool WriteBreakpadInfoStream(MinidumpFileWriter *minidump_writer,
-                           const WriterArgument *writer_args,
-                           MDRawDirectory *dir) {
-  TypedMDRVA<MDRawBreakpadInfo> info(minidump_writer);
-  if (!info.Allocate())
-    return false;
-
-  dir->stream_type = MD_BREAKPAD_INFO_STREAM;
-  dir->location = info.location();
-
-  info.get()->validity = MD_BREAKPAD_INFO_VALID_DUMP_THREAD_ID |
-                        MD_BREAKPAD_INFO_VALID_REQUESTING_THREAD_ID;
-  info.get()->dump_thread_id = getpid();
-  info.get()->requesting_thread_id = writer_args->requester_pid;
-  return true;
-}
-
-// Prototype of writer functions.
-typedef bool (*WriteStringFN)(MinidumpFileWriter *,
-                              const WriterArgument *,
-                              MDRawDirectory *);
-
-// Function table to writer a full minidump.
-WriteStringFN writers[] = {
-  WriteThreadListStream,
-  WriteModuleListStream,
-  WriteSystemInfoStream,
-  WriteExceptionStream,
-  WriteMiscInfoStream,
-  WriteBreakpadInfoStream,
-};
-
-// Will call each writer function in the writers table.
-// It runs in a different process from the crashing process, but sharing
-// the same address space. This enables it to use ptrace functions.
-int Write(void *argument) {
-  WriterArgument *writer_args =
-    static_cast<WriterArgument *>(argument);
-
-  if (!writer_args->thread_lister->SuspendAllThreads())
-    return -1;
-
-  if (writer_args->sighandler_ebp != 0 &&
-      writer_args->thread_lister->FindSigContext(writer_args->sighandler_ebp,
-                                                 &writer_args->sig_ctx)) {
-    writer_args->crashed_stack_bottom =
-      writer_args->thread_lister->GetThreadStackBottom(
-                                             writer_args->sig_ctx->ebp);
-    int crashed_pid =  FindCrashingThread(writer_args->crashed_stack_bottom,
-                                          writer_args->requester_pid,
-                                          writer_args->thread_lister);
-    if (crashed_pid > 0)
-      writer_args->crashed_pid = crashed_pid;
-  }
-
-
-  MinidumpFileWriter *minidump_writer = writer_args->minidump_writer;
-  TypedMDRVA<MDRawHeader> header(minidump_writer);
-  TypedMDRVA<MDRawDirectory> dir(minidump_writer);
-  if (!header.Allocate())
-    return 0;
-
-  int writer_count = sizeof(writers) / sizeof(writers[0]);
-  // Need directory space for all writers.
-  if (!dir.AllocateArray(writer_count))
-    return 0;
-  header.get()->signature = MD_HEADER_SIGNATURE;
-  header.get()->version = MD_HEADER_VERSION;
-  header.get()->time_date_stamp = time(NULL);
-  header.get()->stream_count = writer_count;
-  header.get()->stream_directory_rva = dir.position();
-
-  int dir_index = 0;
-  MDRawDirectory local_dir;
-  for (int i = 0; i < writer_count; ++i) {
-    if (writers[i](minidump_writer, writer_args, &local_dir))
-      dir.CopyIndex(dir_index++, &local_dir);
-  }
-
-  writer_args->thread_lister->ResumeAllThreads();
-  return 0;
-}
-
-}  // namespace
-
-namespace google_breakpad {
-
-MinidumpGenerator::MinidumpGenerator() {
-  AllocateStack();
-}
-
-MinidumpGenerator::~MinidumpGenerator() {
-}
-
-void MinidumpGenerator::AllocateStack() {
-  stack_.reset(new char[kStackSize]);
-}
-
-bool MinidumpGenerator::WriteMinidumpToFile(const char *file_pathname,
-                                   int signo,
-                                   uintptr_t sighandler_ebp,
-                                   struct sigcontext **sig_ctx) const {
-  assert(file_pathname != NULL);
-  assert(stack_ != NULL);
-
-  if (stack_ == NULL || file_pathname == NULL)
-    return false;
-
-  MinidumpFileWriter minidump_writer;
-  if (minidump_writer.Open(file_pathname)) {
-    WriterArgument argument;
-    memset(&argument, 0, sizeof(argument));
-    LinuxThread thread_lister(getpid());
-    argument.thread_lister = &thread_lister;
-    argument.minidump_writer = &minidump_writer;
-    argument.requester_pid = getpid();
-    argument.crashed_pid = getpid();
-    argument.signo = signo;
-    argument.sighandler_ebp = sighandler_ebp;
-    argument.sig_ctx = NULL;
-
-    int cloned_pid = clone(Write, stack_.get() + kStackSize,
-                           CLONE_VM | CLONE_FILES | CLONE_FS | CLONE_UNTRACED,
-                           (void*)&argument);
-    waitpid(cloned_pid, NULL, __WALL);
-    if (sig_ctx != NULL)
-        *sig_ctx = argument.sig_ctx;
-    return true;
-  }
-
-  return false;
-}
-
-}  // namespace google_breakpad

src/client/linux/minidump_writer/directory_reader.h

@@ -0,0 +1,105 @@
+// Copyright (c) 2009, Google Inc.
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+#ifndef CLIENT_LINUX_MINIDUMP_WRITER_DIRECTORY_READER_H_
+#define CLIENT_LINUX_MINIDUMP_WRITER_DIRECTORY_READER_H_
+
+#include <stdint.h>
+#include <unistd.h>
+#include <limits.h>
+#include <assert.h>
+#include <errno.h>
+#include <string.h>
+
+#include "common/linux/linux_syscall_support.h"
+
+namespace google_breakpad {
+
+// A class for enumerating a directory without using diropen/readdir or other
+// functions which may allocate memory.
+class DirectoryReader {
+ public:
+  DirectoryReader(int fd)
+      : fd_(fd),
+        buf_used_(0) {
+  }
+
+  // Return the next entry from the directory
+  //   name: (output) the NUL terminated entry name
+  //
+  // Returns true iff successful (false on EOF).
+  //
+  // After calling this, one must call |PopEntry| otherwise you'll get the same
+  // entry over and over.
+  bool GetNextEntry(const char** name) {
+    struct kernel_dirent* const dent =
+      reinterpret_cast<kernel_dirent*>(buf_);
+
+    if (buf_used_ == 0) {
+      // need to read more entries.
+      const int n = sys_getdents(fd_, dent, sizeof(buf_));
+      if (n < 0) {
+        return false;
+      } else if (n == 0) {
+        hit_eof_ = true;
+      } else {
+        buf_used_ += n;
+      }
+    }
+
+    if (buf_used_ == 0 && hit_eof_)
+      return false;
+
+    assert(buf_used_ > 0);
+
+    *name = dent->d_name;
+    return true;
+  }
+
+  void PopEntry() {
+    if (!buf_used_)
+      return;
+
+    const struct kernel_dirent* const dent =
+      reinterpret_cast<kernel_dirent*>(buf_);
+
+    buf_used_ -= dent->d_reclen;
+    memmove(buf_, buf_ + dent->d_reclen, buf_used_);
+  }
+
+ private:
+  const int fd_;
+  bool hit_eof_;
+  unsigned buf_used_;
+  uint8_t buf_[sizeof(struct kernel_dirent) + NAME_MAX + 1];
+};
+
+}  // namespace google_breakpad
+
+#endif  // CLIENT_LINUX_MINIDUMP_WRITER_DIRECTORY_READER_H_

src/client/linux/minidump_writer/directory_reader_unittest.cc

@@ -0,0 +1,77 @@
+// Copyright (c) 2009, Google Inc.
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+#include <set>
+#include <string>
+
+#include <dirent.h>
+#include <fcntl.h>
+#include <sys/types.h>
+
+#include "client/linux/minidump_writer/directory_reader.h"
+#include "breakpad_googletest_includes.h"
+
+using namespace google_breakpad;
+
+namespace {
+typedef testing::Test DirectoryReaderTest;
+}
+
+TEST(DirectoryReaderTest, CompareResults) {
+  std::set<std::string> dent_set;
+
+  DIR *const dir = opendir("/proc/self");
+  ASSERT_TRUE(dir != NULL);
+
+  struct dirent* dent;
+  while ((dent = readdir(dir)))
+    dent_set.insert(dent->d_name);
+
+  closedir(dir);
+
+  const int fd = open("/proc/self", O_DIRECTORY | O_RDONLY);
+  ASSERT_GE(fd, 0);
+
+  DirectoryReader dir_reader(fd);
+  unsigned seen = 0;
+
+  const char* name;
+  while (dir_reader.GetNextEntry(&name)) {
+    ASSERT_TRUE(dent_set.find(name) != dent_set.end());
+    seen++;
+    dir_reader.PopEntry();
+  }
+
+  ASSERT_TRUE(dent_set.find("status") != dent_set.end());
+  ASSERT_TRUE(dent_set.find("stat") != dent_set.end());
+  ASSERT_TRUE(dent_set.find("cmdline") != dent_set.end());
+
+  ASSERT_EQ(dent_set.size(), seen);
+  close(fd);
+}

src/client/linux/minidump_writer/line_reader.h

@@ -0,0 +1,130 @@
+// Copyright (c) 2009, Google Inc.
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+#ifndef CLIENT_LINUX_MINIDUMP_WRITER_LINE_READER_H_
+#define CLIENT_LINUX_MINIDUMP_WRITER_LINE_READER_H_
+
+#include <stdint.h>
+#include <assert.h>
+#include <string.h>
+
+#include "common/linux/linux_syscall_support.h"
+
+namespace google_breakpad {
+
+// A class for reading a file, line by line, without using fopen/fgets or other
+// functions which may allocate memory.
+class LineReader {
+ public:
+  LineReader(int fd)
+      : fd_(fd),
+        hit_eof_(false),
+        buf_used_(0) {
+  }
+
+  // The maximum length of a line.
+  static const size_t kMaxLineLen = 512;
+
+  // Return the next line from the file.
+  //   line: (output) a pointer to the start of the line. The line is NUL
+  //     terminated.
+  //   len: (output) the length of the line (not inc the NUL byte)
+  //
+  // Returns true iff successful (false on EOF).
+  //
+  // One must call |PopLine| after this function, otherwise you'll continue to
+  // get the same line over and over.
+  bool GetNextLine(const char **line, unsigned *len) {
+    for (;;) {
+      if (buf_used_ == 0 && hit_eof_)
+        return false;
+
+      for (unsigned i = 0; i < buf_used_; ++i) {
+        if (buf_[i] == '\n' || buf_[i] == 0) {
+          buf_[i] = 0;
+          *len = i;
+          *line = buf_;
+          return true;
+        }
+      }
+
+      if (buf_used_ == sizeof(buf_)) {
+        // we scanned the whole buffer and didn't find an end-of-line marker.
+        // This line is too long to process.
+        return false;
+      }
+
+      // We didn't find any end-of-line terminators in the buffer. However, if
+      // this is the last line in the file it might not have one:
+      if (hit_eof_) {
+        assert(buf_used_);
+        // There's room for the NUL because of the buf_used_ == sizeof(buf_)
+        // check above.
+        buf_[buf_used_] = 0;
+        *len = buf_used_;
+        buf_used_ += 1;  // since we appended the NUL.
+        *line = buf_;
+        return true;
+      }
+
+      // Otherwise, we should pull in more data from the file
+      const ssize_t n = sys_read(fd_, buf_ + buf_used_,
+                                 sizeof(buf_) - buf_used_);
+      if (n < 0) {
+        return false;
+      } else if (n == 0) {
+        hit_eof_ = true;
+      } else {
+        buf_used_ += n;
+      }
+
+      // At this point, we have either set the hit_eof_ flag, or we have more
+      // data to process...
+    }
+  }
+
+  void PopLine(unsigned len) {
+    // len doesn't include the NUL byte at the end.
+
+    assert(buf_used_ >= len + 1);
+    buf_used_ -= len + 1;
+    memmove(buf_, buf_ + len + 1, buf_used_);
+  }
+
+ private:
+  const int fd_;
+
+  bool hit_eof_;
+  unsigned buf_used_;
+  char buf_[kMaxLineLen];
+};
+
+}  // namespace google_breakpad
+
+#endif  // CLIENT_LINUX_MINIDUMP_WRITER_LINE_READER_H_

src/client/linux/minidump_writer/line_reader_unittest.cc

@@ -0,0 +1,184 @@
+// Copyright (c) 2009, Google Inc.
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+#include <stdlib.h>
+#include <unistd.h>
+#include <sys/types.h>
+
+#include "client/linux/minidump_writer/line_reader.h"
+#include "breakpad_googletest_includes.h"
+
+using namespace google_breakpad;
+
+static int TemporaryFile() {
+  static const char templ[] = "/tmp/line-reader-unittest-XXXXXX";
+  char templ_copy[sizeof(templ)];
+  memcpy(templ_copy, templ, sizeof(templ));
+  const int fd = mkstemp(templ_copy);
+  if (fd >= 0)
+    unlink(templ_copy);
+
+  return fd;
+}
+
+namespace {
+typedef testing::Test LineReaderTest;
+}
+
+TEST(LineReaderTest, EmptyFile) {
+  const int fd = TemporaryFile();
+  LineReader reader(fd);
+
+  const char *line;
+  unsigned len;
+  ASSERT_FALSE(reader.GetNextLine(&line, &len));
+
+  close(fd);
+}
+
+TEST(LineReaderTest, OneLineTerminated) {
+  const int fd = TemporaryFile();
+  write(fd, "a\n", 2);
+  lseek(fd, 0, SEEK_SET);
+  LineReader reader(fd);
+
+  const char *line;
+  unsigned len;
+  ASSERT_TRUE(reader.GetNextLine(&line, &len));
+  ASSERT_EQ(len, 1);
+  ASSERT_EQ(line[0], 'a');
+  ASSERT_EQ(line[1], 0);
+  reader.PopLine(len);
+
+  ASSERT_FALSE(reader.GetNextLine(&line, &len));
+
+  close(fd);
+}
+
+TEST(LineReaderTest, OneLine) {
+  const int fd = TemporaryFile();
+  write(fd, "a", 1);
+  lseek(fd, 0, SEEK_SET);
+  LineReader reader(fd);
+
+  const char *line;
+  unsigned len;
+  ASSERT_TRUE(reader.GetNextLine(&line, &len));
+  ASSERT_EQ(len, 1);
+  ASSERT_EQ(line[0], 'a');
+  ASSERT_EQ(line[1], 0);
+  reader.PopLine(len);
+
+  ASSERT_FALSE(reader.GetNextLine(&line, &len));
+
+  close(fd);
+}
+
+TEST(LineReaderTest, TwoLinesTerminated) {
+  const int fd = TemporaryFile();
+  write(fd, "a\nb\n", 4);
+  lseek(fd, 0, SEEK_SET);
+  LineReader reader(fd);
+
+  const char *line;
+  unsigned len;
+  ASSERT_TRUE(reader.GetNextLine(&line, &len));
+  ASSERT_EQ(len, 1);
+  ASSERT_EQ(line[0], 'a');
+  ASSERT_EQ(line[1], 0);
+  reader.PopLine(len);
+
+  ASSERT_TRUE(reader.GetNextLine(&line, &len));
+  ASSERT_EQ(len, 1);
+  ASSERT_EQ(line[0], 'b');
+  ASSERT_EQ(line[1], 0);
+  reader.PopLine(len);
+
+  ASSERT_FALSE(reader.GetNextLine(&line, &len));
+
+  close(fd);
+}
+
+TEST(LineReaderTest, TwoLines) {
+  const int fd = TemporaryFile();
+  write(fd, "a\nb", 3);
+  lseek(fd, 0, SEEK_SET);
+  LineReader reader(fd);
+
+  const char *line;
+  unsigned len;
+  ASSERT_TRUE(reader.GetNextLine(&line, &len));
+  ASSERT_EQ(len, 1);
+  ASSERT_EQ(line[0], 'a');
+  ASSERT_EQ(line[1], 0);
+  reader.PopLine(len);
+
+  ASSERT_TRUE(reader.GetNextLine(&line, &len));
+  ASSERT_EQ(len, 1);
+  ASSERT_EQ(line[0], 'b');
+  ASSERT_EQ(line[1], 0);
+  reader.PopLine(len);
+
+  ASSERT_FALSE(reader.GetNextLine(&line, &len));
+
+  close(fd);
+}
+
+TEST(LineReaderTest, MaxLength) {
+  const int fd = TemporaryFile();
+  char l[LineReader::kMaxLineLen - 1];
+  memset(l, 'a', sizeof(l));
+  write(fd, l, sizeof(l));
+  lseek(fd, 0, SEEK_SET);
+  LineReader reader(fd);
+
+  const char *line;
+  unsigned len;
+  ASSERT_TRUE(reader.GetNextLine(&line, &len));
+  ASSERT_EQ(len, sizeof(l));
+  ASSERT_TRUE(memcmp(l, line, sizeof(l)) == 0);
+  ASSERT_EQ(line[len], 0);
+
+  close(fd);
+}
+
+TEST(LineReaderTest, TooLong) {
+  const int fd = TemporaryFile();
+  char l[LineReader::kMaxLineLen];
+  memset(l, 'a', sizeof(l));
+  write(fd, l, sizeof(l));
+  lseek(fd, 0, SEEK_SET);
+  LineReader reader(fd);
+
+  const char *line;
+  unsigned len;
+  ASSERT_FALSE(reader.GetNextLine(&line, &len));
+
+  close(fd);
+}

src/client/linux/minidump_writer/linux_dumper.cc

@@ -0,0 +1,419 @@
+// Copyright (c) 2009, Google Inc.
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// This code deals with the mechanics of getting information about a crashed
+// process. Since this code may run in a compromised address space, the same
+// rules apply as detailed at the top of minidump_writer.h: no libc calls and
+// use the alternative allocator.
+
+#include "client/linux/minidump_writer/linux_dumper.h"
+
+#include <assert.h>
+#include <limits.h>
+#include <stddef.h>
+#include <stdlib.h>
+#include <stdio.h>
+#include <string.h>
+
+#include <unistd.h>
+#include <errno.h>
+#include <fcntl.h>
+
+#include <sys/types.h>
+#include <sys/ptrace.h>
+#include <sys/wait.h>
+
+#include "client/linux/minidump_writer/directory_reader.h"
+#include "client/linux/minidump_writer/line_reader.h"
+#include "common/linux/linux_libc_support.h"
+#include "common/linux/linux_syscall_support.h"
+
+// Suspend a thread by attaching to it.
+static bool SuspendThread(pid_t pid) {
+  // This may fail if the thread has just died or debugged.
+  errno = 0;
+  if (sys_ptrace(PTRACE_ATTACH, pid, NULL, NULL) != 0 &&
+      errno != 0) {
+    return false;
+  }
+  while (sys_waitpid(pid, NULL, __WALL) < 0) {
+    if (errno != EINTR) {
+      sys_ptrace(PTRACE_DETACH, pid, NULL, NULL);
+      return false;
+    }
+  }
+  return true;
+}
+
+// Resume a thread by detaching from it.
+static bool ResumeThread(pid_t pid) {
+  return sys_ptrace(PTRACE_DETACH, pid, NULL, NULL) >= 0;
+}
+
+namespace google_breakpad {
+
+LinuxDumper::LinuxDumper(int pid)
+    : pid_(pid),
+      threads_suspened_(false),
+      threads_(&allocator_, 8),
+      mappings_(&allocator_) {
+}
+
+bool LinuxDumper::Init() {
+  return EnumerateThreads(&threads_) &&
+         EnumerateMappings(&mappings_);
+}
+
+bool LinuxDumper::ThreadsSuspend() {
+  if (threads_suspened_)
+    return true;
+  bool good = true;
+  for (size_t i = 0; i < threads_.size(); ++i)
+    good &= SuspendThread(threads_[i]);
+  threads_suspened_ = true;
+  return good;
+}
+
+bool LinuxDumper::ThreadsResume() {
+  if (!threads_suspened_)
+    return false;
+  bool good = true;
+  for (size_t i = 0; i < threads_.size(); ++i)
+    good &= ResumeThread(threads_[i]);
+  threads_suspened_ = false;
+  return good;
+}
+
+void
+LinuxDumper::BuildProcPath(char* path, pid_t pid, const char* node) const {
+  assert(path);
+  if (!path) {
+    return;
+  }
+
+  path[0] = '\0';
+
+  const unsigned pid_len = my_int_len(pid);
+
+  assert(node);
+  if (!node) {
+    return;
+  }
+
+  size_t node_len = my_strlen(node);
+  assert(node_len < NAME_MAX);
+  if (node_len >= NAME_MAX) {
+    return;
+  }
+
+  assert(node_len > 0);
+  if (node_len == 0) {
+    return;
+  }
+
+  assert(pid > 0);
+  if (pid <= 0) {
+    return;
+  }
+
+  const size_t total_length = 6 + pid_len + 1 + node_len;
+
+  assert(total_length < NAME_MAX);
+  if (total_length >= NAME_MAX) {
+    return;
+  }
+
+  memcpy(path, "/proc/", 6);
+  my_itos(path + 6, pid, pid_len);
+  memcpy(path + 6 + pid_len, "/", 1);
+  memcpy(path + 6 + pid_len + 1, node, node_len);
+  memcpy(path + total_length, "\0", 1);
+}
+
+void*
+LinuxDumper::FindBeginningOfLinuxGateSharedLibrary(const pid_t pid) const {
+  char auxv_path[80];
+  BuildProcPath(auxv_path, pid, "auxv");
+
+  // If BuildProcPath errors out due to invalid input, we'll handle it when
+  // we try to sys_open the file.
+
+  // Find the AT_SYSINFO_EHDR entry for linux-gate.so
+  // See http://www.trilithium.com/johan/2005/08/linux-gate/ for more
+  // information.
+  int fd = sys_open(auxv_path, O_RDONLY, 0);
+  if (fd < 0) {
+    return NULL;
+  }
+
+  elf_aux_entry one_aux_entry;
+  while (sys_read(fd,
+                  &one_aux_entry,
+                  sizeof(elf_aux_entry)) == sizeof(elf_aux_entry) &&
+         one_aux_entry.a_type != AT_NULL) {
+    if (one_aux_entry.a_type == AT_SYSINFO_EHDR) {
+      close(fd);
+      return reinterpret_cast<void*>(one_aux_entry.a_un.a_val);
+    }
+  }
+  close(fd);
+  return NULL;
+}
+
+bool
+LinuxDumper::EnumerateMappings(wasteful_vector<MappingInfo*>* result) const {
+  char maps_path[80];
+  BuildProcPath(maps_path, pid_, "maps");
+
+  // linux_gate_loc is the beginning of the kernel's mapping of
+  // linux-gate.so in the process.  It doesn't actually show up in the
+  // maps list as a filename, so we use the aux vector to find it's
+  // load location and special case it's entry when creating the list
+  // of mappings.
+  const void* linux_gate_loc;
+  linux_gate_loc = FindBeginningOfLinuxGateSharedLibrary(pid_);
+
+  const int fd = sys_open(maps_path, O_RDONLY, 0);
+  if (fd < 0)
+    return false;
+  LineReader* const line_reader = new(allocator_) LineReader(fd);
+
+  const char* line;
+  unsigned line_len;
+  while (line_reader->GetNextLine(&line, &line_len)) {
+    uintptr_t start_addr, end_addr, offset;
+
+    const char* i1 = my_read_hex_ptr(&start_addr, line);
+    if (*i1 == '-') {
+      const char* i2 = my_read_hex_ptr(&end_addr, i1 + 1);
+      if (*i2 == ' ') {
+        const char* i3 = my_read_hex_ptr(&offset, i2 + 6 /* skip ' rwxp ' */);
+        if (*i3 == ' ') {
+          MappingInfo* const module = new(allocator_) MappingInfo;
+          memset(module, 0, sizeof(MappingInfo));
+          module->start_addr = start_addr;
+          module->size = end_addr - start_addr;
+          module->offset = offset;
+          const char* name = NULL;
+          // Only copy name if the name is a valid path name, or if
+          // we've found the VDSO image
+          if ((name = my_strchr(line, '/')) != NULL) {
+            const unsigned l = my_strlen(name);
+            if (l < sizeof(module->name))
+              memcpy(module->name, name, l);
+          } else if (linux_gate_loc &&
+                     reinterpret_cast<void*>(module->start_addr) ==
+                     linux_gate_loc) {
+            memcpy(module->name,
+                   kLinuxGateLibraryName,
+                   my_strlen(kLinuxGateLibraryName));
+            module->offset = 0;
+          }
+          result->push_back(module);
+        }
+      }
+    }
+    line_reader->PopLine(line_len);
+  }
+
+  sys_close(fd);
+
+  return result->size() > 0;
+}
+
+// Parse /proc/$pid/task to list all the threads of the process identified by
+// pid.
+bool LinuxDumper::EnumerateThreads(wasteful_vector<pid_t>* result) const {
+  char task_path[80];
+  BuildProcPath(task_path, pid_, "task");
+
+  const int fd = sys_open(task_path, O_RDONLY | O_DIRECTORY, 0);
+  if (fd < 0)
+    return false;
+  DirectoryReader* dir_reader = new(allocator_) DirectoryReader(fd);
+
+  // The directory may contain duplicate entries which we filter by assuming
+  // that they are consecutive.
+  int last_tid = -1;
+  const char* dent_name;
+  while (dir_reader->GetNextEntry(&dent_name)) {
+    if (my_strcmp(dent_name, ".") &&
+        my_strcmp(dent_name, "..")) {
+      int tid = 0;
+      if (my_strtoui(&tid, dent_name) &&
+          last_tid != tid) {
+        last_tid = tid;
+        result->push_back(tid);
+      }
+    }
+    dir_reader->PopEntry();
+  }
+
+  sys_close(fd);
+  return true;
+}
+
+// Read thread info from /proc/$pid/status.
+// Fill out the |tgid|, |ppid| and |pid| members of |info|. If unavailible,
+// these members are set to -1. Returns true iff all three members are
+// availible.
+bool LinuxDumper::ThreadInfoGet(pid_t tid, ThreadInfo* info) {
+  assert(info != NULL);
+  char status_path[80];
+  BuildProcPath(status_path, tid, "status");
+
+  const int fd = open(status_path, O_RDONLY);
+  if (fd < 0)
+    return false;
+
+  LineReader* const line_reader = new(allocator_) LineReader(fd);
+  const char* line;
+  unsigned line_len;
+
+  info->ppid = info->tgid = -1;
+
+  while (line_reader->GetNextLine(&line, &line_len)) {
+    if (my_strncmp("Tgid:\t", line, 6) == 0) {
+      my_strtoui(&info->tgid, line + 6);
+    } else if (my_strncmp("PPid:\t", line, 6) == 0) {
+      my_strtoui(&info->ppid, line + 6);
+    }
+
+    line_reader->PopLine(line_len);
+  }
+
+  if (info->ppid == -1 || info->tgid == -1)
+    return false;
+
+  if (sys_ptrace(PTRACE_GETREGS, tid, NULL, &info->regs) == -1 ||
+      sys_ptrace(PTRACE_GETFPREGS, tid, NULL, &info->fpregs) == -1) {
+    return false;
+  }
+
+#if defined(__i386) || defined(__x86_64)
+  if (sys_ptrace(PTRACE_GETFPXREGS, tid, NULL, &info->fpxregs) == -1)
+    return false;
+
+  for (unsigned i = 0; i < ThreadInfo::kNumDebugRegisters; ++i) {
+    if (sys_ptrace(
+        PTRACE_PEEKUSER, tid,
+        reinterpret_cast<void*> (offsetof(struct user,
+                                          u_debugreg[0]) + i *
+                                 sizeof(debugreg_t)),
+        &info->dregs[i]) == -1) {
+      return false;
+    }
+  }
+#endif
+
+  const uint8_t* stack_pointer;
+#if defined(__i386)
+  memcpy(&stack_pointer, &info->regs.esp, sizeof(info->regs.esp));
+#elif defined(__x86_64)
+  memcpy(&stack_pointer, &info->regs.rsp, sizeof(info->regs.rsp));
+#else
+#error "This code hasn't been ported to your platform yet."
+#endif
+
+  if (!GetStackInfo(&info->stack, &info->stack_len,
+                    (uintptr_t) stack_pointer))
+    return false;
+
+  return true;
+}
+
+// Get information about the stack, given the stack pointer. We don't try to
+// walk the stack since we might not have all the information needed to do
+// unwind. So we just grab, up to, 32k of stack.
+bool LinuxDumper::GetStackInfo(const void** stack, size_t* stack_len,
+                               uintptr_t int_stack_pointer) {
+#if defined(__i386) || defined(__x86_64)
+  static const bool stack_grows_down = true;
+  static const uintptr_t page_size = 4096;
+#else
+#error "This code has not been ported to your platform yet."
+#endif
+  // Move the stack pointer to the bottom of the page that it's in.
+  uint8_t* const stack_pointer =
+      reinterpret_cast<uint8_t*>(int_stack_pointer & ~(page_size - 1));
+
+  // The number of bytes of stack which we try to capture.
+  static unsigned kStackToCapture = 32 * 1024;
+
+  const MappingInfo* mapping = FindMapping(stack_pointer);
+  if (!mapping)
+    return false;
+  if (stack_grows_down) {
+    const ptrdiff_t offset = stack_pointer - (uint8_t*) mapping->start_addr;
+    const ptrdiff_t distance_to_end =
+        static_cast<ptrdiff_t>(mapping->size) - offset;
+    *stack_len = distance_to_end > kStackToCapture ?
+                 kStackToCapture : distance_to_end;
+    *stack = stack_pointer;
+  } else {
+    const ptrdiff_t offset = stack_pointer - (uint8_t*) mapping->start_addr;
+    *stack_len = offset > kStackToCapture ? kStackToCapture : offset;
+    *stack = stack_pointer - *stack_len;
+  }
+
+  return true;
+}
+
+// static
+void LinuxDumper::CopyFromProcess(void* dest, pid_t child, const void* src,
+                                  size_t length) {
+  unsigned long tmp;
+  size_t done = 0;
+  static const size_t word_size = sizeof(tmp);
+  uint8_t* const local = (uint8_t*) dest;
+  uint8_t* const remote = (uint8_t*) src;
+
+  while (done < length) {
+    const size_t l = length - done > word_size ? word_size : length - done;
+    if (sys_ptrace(PTRACE_PEEKDATA, child, remote + done, &tmp) == -1)
+      tmp = 0;
+    memcpy(local + done, &tmp, l);
+    done += l;
+  }
+}
+
+// Find the mapping which the given memory address falls in.
+const MappingInfo* LinuxDumper::FindMapping(const void* address) const {
+  const uintptr_t addr = (uintptr_t) address;
+
+  for (size_t i = 0; i < mappings_.size(); ++i) {
+    const uintptr_t start = static_cast<uintptr_t>(mappings_[i]->start_addr);
+    if (addr >= start && addr - start < mappings_[i]->size)
+      return mappings_[i];
+  }
+
+  return NULL;
+}
+
+}  // namespace google_breakpad

src/client/linux/minidump_writer/linux_dumper.h

@@ -0,0 +1,142 @@
+// Copyright (c) 2009, Google Inc.
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+#ifndef CLIENT_LINUX_MINIDUMP_WRITER_LINUX_DUMPER_H_
+#define CLIENT_LINUX_MINIDUMP_WRITER_LINUX_DUMPER_H_
+
+#include <elf.h>
+#include <stdint.h>
+#include <sys/user.h>
+#include <linux/limits.h>
+
+#include "common/linux/memory.h"
+
+namespace google_breakpad {
+
+typedef typeof(((struct user*) 0)->u_debugreg[0]) debugreg_t;
+
+// Typedef for our parsing of the auxv variables in /proc/pid/auxv.
+#if defined(__i386)
+typedef Elf32_auxv_t elf_aux_entry;
+#elif defined(__x86_64__)
+typedef Elf64_auxv_t elf_aux_entry;
+#endif
+// When we find the VDSO mapping in the process's address space, this
+// is the name we use for it when writing it to the minidump.
+// This should always be less than NAME_MAX!
+const char kLinuxGateLibraryName[] = "linux-gate.so";
+
+// We produce one of these structures for each thread in the crashed process.
+struct ThreadInfo {
+  pid_t tgid;   // thread group id
+  pid_t ppid;   // parent process
+
+  // Even on platforms where the stack grows down, the following will point to
+  // the smallest address in the stack.
+  const void* stack;  // pointer to the stack area
+  size_t stack_len;  // length of the stack to copy
+
+  user_regs_struct regs;
+  user_fpregs_struct fpregs;
+#if defined(__i386) || defined(__x86_64)
+  user_fpxregs_struct fpxregs;
+
+  static const unsigned kNumDebugRegisters = 8;
+  debugreg_t dregs[8];
+#endif
+};
+
+// One of these is produced for each mapping in the process (i.e. line in
+// /proc/$x/maps).
+struct MappingInfo {
+  uintptr_t start_addr;
+  size_t size;
+  size_t offset;  // offset into the backed file.
+  char name[NAME_MAX];
+};
+
+class LinuxDumper {
+ public:
+  explicit LinuxDumper(pid_t pid);
+
+  // Parse the data for |threads| and |mappings|.
+  bool Init();
+
+  // Suspend/resume all threads in the given process.
+  bool ThreadsSuspend();
+  bool ThreadsResume();
+
+  // Read information about the given thread. Returns true on success. One must
+  // have called |ThreadsSuspend| first.
+  bool ThreadInfoGet(pid_t tid, ThreadInfo* info);
+
+  // These are only valid after a call to |Init|.
+  const wasteful_vector<pid_t> &threads() { return threads_; }
+  const wasteful_vector<MappingInfo*> &mappings() { return mappings_; }
+  const MappingInfo* FindMapping(const void* address) const;
+
+  // Find a block of memory to take as the stack given the top of stack pointer.
+  //   stack: (output) the lowest address in the memory area
+  //   stack_len: (output) the length of the memory area
+  //   stack_top: the current top of the stack
+  bool GetStackInfo(const void** stack, size_t* stack_len, uintptr_t stack_top);
+
+  PageAllocator* allocator() { return &allocator_; }
+
+  // memcpy from a remote process.
+  static void CopyFromProcess(void* dest, pid_t child, const void* src,
+                              size_t length);
+
+  // Builds a proc path for a certain pid for a node.  path is a
+  // character array that is overwritten, and node is the final node
+  // without any slashes.
+  void BuildProcPath(char* path, pid_t pid, const char* node) const;
+
+  // Utility method to find the location of where the kernel has
+  // mapped linux-gate.so in memory(shows up in /proc/pid/maps as
+  // [vdso], but we can't guarantee that it's the only virtual dynamic
+  // shared object.  Parsing the auxilary vector for AT_SYSINFO_EHDR
+  // is the safest way to go.)
+  void* FindBeginningOfLinuxGateSharedLibrary(const pid_t pid) const;
+ private:
+  bool EnumerateMappings(wasteful_vector<MappingInfo*>* result) const;
+  bool EnumerateThreads(wasteful_vector<pid_t>* result) const;
+
+  const pid_t pid_;
+
+  mutable PageAllocator allocator_;
+
+  bool threads_suspened_;
+  wasteful_vector<pid_t> threads_;  // the ids of all the threads
+  wasteful_vector<MappingInfo*> mappings_;  // info from /proc/<pid>/maps
+};
+
+}  // namespace google_breakpad
+
+#endif  // CLIENT_LINUX_HANDLER_LINUX_DUMPER_H_

src/client/linux/minidump_writer/linux_dumper_unittest.cc

@@ -0,0 +1,118 @@
+// Copyright (c) 2009, Google Inc.
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+#include <unistd.h>
+
+#include "client/linux/minidump_writer/linux_dumper.h"
+#include "breakpad_googletest_includes.h"
+
+using namespace google_breakpad;
+
+namespace {
+typedef testing::Test LinuxDumperTest;
+}
+
+TEST(LinuxDumperTest, Setup) {
+  LinuxDumper dumper(getpid());
+}
+
+TEST(LinuxDumperTest, FindMappings) {
+  LinuxDumper dumper(getpid());
+  ASSERT_TRUE(dumper.Init());
+
+  ASSERT_TRUE(dumper.FindMapping(reinterpret_cast<void*>(getpid)));
+  ASSERT_TRUE(dumper.FindMapping(reinterpret_cast<void*>(printf)));
+  ASSERT_FALSE(dumper.FindMapping(NULL));
+}
+
+TEST(LinuxDumperTest, ThreadList) {
+  LinuxDumper dumper(getpid());
+  ASSERT_TRUE(dumper.Init());
+
+  ASSERT_GE(dumper.threads().size(), 1);
+  bool found = false;
+  for (size_t i = 0; i < dumper.threads().size(); ++i) {
+    if (dumper.threads()[i] == getpid()) {
+      found = true;
+      break;
+    }
+  }
+}
+
+TEST(LinuxDumperTest, BuildProcPath) {
+  const pid_t pid = getpid();
+  LinuxDumper dumper(pid);
+
+  char maps_path[256] = "dummymappath";
+  char maps_path_expected[256];
+  snprintf(maps_path_expected, sizeof(maps_path_expected),
+           "/proc/%d/maps", pid);
+  dumper.BuildProcPath(maps_path, pid, "maps");
+  ASSERT_STREQ(maps_path, maps_path_expected);
+
+  // In release mode, we expect BuildProcPath to handle the invalid
+  // parameters correctly and fill map_path with an empty
+  // NULL-terminated string.
+#ifdef NDEBUG
+  snprintf(maps_path, sizeof(maps_path), "dummymappath");
+  dumper.BuildProcPath(maps_path, 0, "maps");
+  EXPECT_STREQ(maps_path, "");
+
+  snprintf(maps_path, sizeof(maps_path), "dummymappath");
+  dumper.BuildProcPath(maps_path, getpid(), "");
+  EXPECT_STREQ(maps_path, "");
+
+  snprintf(maps_path, sizeof(maps_path), "dummymappath");
+  dumper.BuildProcPath(maps_path, getpid(), NULL);
+  EXPECT_STREQ(maps_path, "");
+#endif
+}
+
+TEST(LinuxDumperTest, MappingsIncludeLinuxGate) {
+  LinuxDumper dumper(getpid());
+  ASSERT_TRUE(dumper.Init());
+
+  void* linux_gate_loc = dumper.FindBeginningOfLinuxGateSharedLibrary(getpid());
+  if (linux_gate_loc) {
+    bool found_linux_gate = false;
+
+    const wasteful_vector<MappingInfo*> mappings = dumper.mappings();
+    const MappingInfo* mapping;
+    for (unsigned i = 0; i < mappings.size(); ++i) {
+      mapping = mappings[i];
+      if (!strcmp(mapping->name, kLinuxGateLibraryName)) {
+        found_linux_gate = true;
+        break;
+      }
+    }
+    EXPECT_TRUE(found_linux_gate);
+    EXPECT_EQ(linux_gate_loc, reinterpret_cast<void*>(mapping->start_addr));
+    EXPECT_EQ(0, memcmp(linux_gate_loc, ELFMAG, SELFMAG));
+  }
+}

src/client/linux/minidump_writer/minidump_writer.cc

@@ -0,0 +1,872 @@
+// Copyright (c) 2009, Google Inc.
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// This code writes out minidump files:
+//   http://msdn.microsoft.com/en-us/library/ms680378(VS.85,loband).aspx
+//
+// Minidumps are a Microsoft format which Breakpad uses for recording crash
+// dumps. This code has to run in a compromised environment (the address space
+// may have received SIGSEGV), thus the following rules apply:
+//   * You may not enter the dynamic linker. This means that we cannot call
+//     any symbols in a shared library (inc libc). Because of this we replace
+//     libc functions in linux_libc_support.h.
+//   * You may not call syscalls via the libc wrappers. This rule is a subset
+//     of the first rule but it bears repeating. We have direct wrappers
+//     around the system calls in linux_syscall_support.h.
+//   * You may not malloc. There's an alternative allocator in memory.h and
+//     a canonical instance in the LinuxDumper object. We use the placement
+//     new form to allocate objects and we don't delete them.
+
+#include "client/linux/minidump_writer/minidump_writer.h"
+#include "client/minidump_file_writer-inl.h"
+
+#include <errno.h>
+#include <fcntl.h>
+#include <stdio.h>
+#include <unistd.h>
+#include <sys/ucontext.h>
+#include <sys/user.h>
+#include <sys/utsname.h>
+
+#include "client/minidump_file_writer.h"
+#include "google_breakpad/common/minidump_format.h"
+#include "google_breakpad/common/minidump_cpu_amd64.h"
+#include "google_breakpad/common/minidump_cpu_x86.h"
+
+#include "client/linux/handler/exception_handler.h"
+#include "client/linux/minidump_writer/line_reader.h"
+#include "client/linux/minidump_writer//linux_dumper.h"
+#include "common/linux/linux_libc_support.h"
+#include "common/linux/linux_syscall_support.h"
+
+// These are additional minidump stream values which are specific to the linux
+// breakpad implementation.
+enum {
+  MD_LINUX_CPU_INFO              = 0x47670003,    /* /proc/cpuinfo    */
+  MD_LINUX_PROC_STATUS           = 0x47670004,    /* /proc/$x/status  */
+  MD_LINUX_LSB_RELEASE           = 0x47670005,    /* /etc/lsb-release */
+  MD_LINUX_CMD_LINE              = 0x47670006,    /* /proc/$x/cmdline */
+  MD_LINUX_ENVIRON               = 0x47670007,    /* /proc/$x/environ */
+  MD_LINUX_AUXV                  = 0x47670008,    /* /proc/$x/auxv    */
+};
+
+// Minidump defines register structures which are different from the raw
+// structures which we get from the kernel. These are platform specific
+// functions to juggle the ucontext and user structures into minidump format.
+#if defined(__i386)
+typedef MDRawContextX86 RawContextCPU;
+
+// Write a uint16_t to memory
+//   out: memory location to write to
+//   v: value to write.
+static void U16(void* out, uint16_t v) {
+  memcpy(out, &v, sizeof(v));
+}
+
+// Write a uint32_t to memory
+//   out: memory location to write to
+//   v: value to write.
+static void U32(void* out, uint32_t v) {
+  memcpy(out, &v, sizeof(v));
+}
+
+// Juggle an x86 user_(fp|fpx|)regs_struct into minidump format
+//   out: the minidump structure
+//   info: the collection of register structures.
+static void CPUFillFromThreadInfo(MDRawContextX86 *out,
+                                  const google_breakpad::ThreadInfo &info) {
+  out->context_flags = MD_CONTEXT_X86_ALL;
+
+  out->dr0 = info.dregs[0];
+  out->dr1 = info.dregs[1];
+  out->dr2 = info.dregs[2];
+  out->dr3 = info.dregs[3];
+  // 4 and 5 deliberatly omitted because they aren't included in the minidump
+  // format.
+  out->dr6 = info.dregs[6];
+  out->dr7 = info.dregs[7];
+
+  out->gs = info.regs.xgs;
+  out->fs = info.regs.xfs;
+  out->es = info.regs.xes;
+  out->ds = info.regs.xds;
+
+  out->edi = info.regs.edi;
+  out->esi = info.regs.esi;
+  out->ebx = info.regs.ebx;
+  out->edx = info.regs.edx;
+  out->ecx = info.regs.ecx;
+  out->eax = info.regs.eax;
+
+  out->ebp = info.regs.ebp;
+  out->eip = info.regs.eip;
+  out->cs = info.regs.xcs;
+  out->eflags = info.regs.eflags;
+  out->esp = info.regs.esp;
+  out->ss = info.regs.xss;
+
+  out->float_save.control_word = info.fpregs.cwd;
+  out->float_save.status_word = info.fpregs.swd;
+  out->float_save.tag_word = info.fpregs.twd;
+  out->float_save.error_offset = info.fpregs.fip;
+  out->float_save.error_selector = info.fpregs.fcs;
+  out->float_save.data_offset = info.fpregs.foo;
+  out->float_save.data_selector = info.fpregs.fos;
+
+  // 8 registers * 10 bytes per register.
+  memcpy(out->float_save.register_area, info.fpregs.st_space, 10 * 8);
+
+  // This matches the Intel fpsave format.
+  U16(out->extended_registers + 0, info.fpregs.cwd);
+  U16(out->extended_registers + 2, info.fpregs.swd);
+  U16(out->extended_registers + 4, info.fpregs.twd);
+  U16(out->extended_registers + 6, info.fpxregs.fop);
+  U32(out->extended_registers + 8, info.fpxregs.fip);
+  U16(out->extended_registers + 12, info.fpxregs.fcs);
+  U32(out->extended_registers + 16, info.fpregs.foo);
+  U16(out->extended_registers + 20, info.fpregs.fos);
+  U32(out->extended_registers + 24, info.fpxregs.mxcsr);
+
+  memcpy(out->extended_registers + 32, &info.fpxregs.st_space, 128);
+  memcpy(out->extended_registers + 160, &info.fpxregs.xmm_space, 128);
+}
+
+// Juggle an x86 ucontext into minidump format
+//   out: the minidump structure
+//   info: the collection of register structures.
+static void CPUFillFromUContext(MDRawContextX86 *out, const ucontext *uc,
+                                const struct _libc_fpstate* fp) {
+  const greg_t* regs = uc->uc_mcontext.gregs;
+
+  out->context_flags = MD_CONTEXT_X86_FULL |
+                       MD_CONTEXT_X86_FLOATING_POINT;
+
+  out->gs = regs[REG_GS];
+  out->fs = regs[REG_FS];
+  out->es = regs[REG_ES];
+  out->ds = regs[REG_DS];
+
+  out->edi = regs[REG_EDI];
+  out->esi = regs[REG_ESI];
+  out->ebx = regs[REG_EBX];
+  out->edx = regs[REG_EDX];
+  out->ecx = regs[REG_ECX];
+  out->eax = regs[REG_EAX];
+
+  out->ebp = regs[REG_EBP];
+  out->eip = regs[REG_EIP];
+  out->cs = regs[REG_CS];
+  out->eflags = regs[REG_EFL];
+  out->esp = regs[REG_UESP];
+  out->ss = regs[REG_SS];
+
+  out->float_save.control_word = fp->cw;
+  out->float_save.status_word = fp->sw;
+  out->float_save.tag_word = fp->tag;
+  out->float_save.error_offset = fp->ipoff;
+  out->float_save.error_selector = fp->cssel;
+  out->float_save.data_offset = fp->dataoff;
+  out->float_save.data_selector = fp->datasel;
+
+  // 8 registers * 10 bytes per register.
+  memcpy(out->float_save.register_area, fp->_st, 10 * 8);
+}
+
+#elif defined(__x86_64)
+typedef MDRawContextAMD64 RawContextCPU;
+
+static void CPUFillFromThreadInfo(MDRawContextAMD64 *out,
+                                  const google_breakpad::ThreadInfo &info) {
+  out->context_flags = MD_CONTEXT_AMD64_FULL |
+                       MD_CONTEXT_AMD64_SEGMENTS;
+
+  out->cs = info.regs.cs;
+
+  out->ds = info.regs.ds;
+  out->es = info.regs.es;
+  out->fs = info.regs.fs;
+  out->gs = info.regs.gs;
+
+  out->ss = info.regs.ss;
+  out->eflags = info.regs.eflags;
+
+  out->dr0 = info.dregs[0];
+  out->dr1 = info.dregs[1];
+  out->dr2 = info.dregs[2];
+  out->dr3 = info.dregs[3];
+  // 4 and 5 deliberatly omitted because they aren't included in the minidump
+  // format.
+  out->dr6 = info.dregs[6];
+  out->dr7 = info.dregs[7];
+
+  out->rax = info.regs.rax;
+  out->rcx = info.regs.rcx;
+  out->rdx = info.regs.rdx;
+  out->rbx = info.regs.rbx;
+
+  out->rsp = info.regs.rsp;
+
+  out->rbp = info.regs.rbp;
+  out->rsi = info.regs.rsi;
+  out->rdi = info.regs.rdi;
+  out->r8 = info.regs.r8;
+  out->r9 = info.regs.r9;
+  out->r10 = info.regs.r10;
+  out->r11 = info.regs.r11;
+  out->r12 = info.regs.r12;
+  out->r13 = info.regs.r13;
+  out->r14 = info.regs.r14;
+  out->r15 = info.regs.r15;
+
+  out->rip = info.regs.rip;
+
+  out->flt_save.control_word = info.fpregs.cwd;
+  out->flt_save.status_word = info.fpregs.swd;
+  out->flt_save.tag_word = info.fpregs.twd;
+  out->flt_save.error_opcode = info.fpregs.fop;
+  out->flt_save.error_offset = info.fpregs.rip;
+  out->flt_save.error_selector = 0; // We don't have this.
+  out->flt_save.data_offset = info.fpregs.rdp;
+  out->flt_save.data_selector = 0;  // We don't have this.
+  out->flt_save.mx_csr = info.fpregs.mxcsr;
+  out->flt_save.mx_csr_mask = info.fpregs.mxcsr_mask;
+  memcpy(&out->flt_save.float_registers, &info.fpregs.st_space, 8 * 16);
+  memcpy(&out->flt_save.xmm_registers, &info.fpregs.xmm_space, 16 * 16);
+}
+
+static void CPUFillFromUContext(MDRawContextAMD64 *out, const ucontext *uc,
+                                const struct _libc_fpstate* fpregs) {
+  const greg_t* regs = uc->gregs;
+
+  out->context_flags = MD_CONTEXT_AMD64_FULL;
+
+  out->cs = regs[REG_CSGSFS] & 0xffff;
+
+  out->fs = (regs[REG_CSGSFS] >> 32) & 0xffff;
+  out->gs = (regs[REG_CSGSFS] >> 16) & 0xffff;
+
+  out->eflags = regs[REG_EFL];
+
+  out->rax = regs[REG_RAX];
+  out->rcx = regs[REG_RCX];
+  out->rdx = regs[REG_RDX];
+  out->rbx = regs[REG_RBX];
+
+  out->rsp = regs[REG_RSP];
+  out->rbp = regs[REG_RBP];
+  out->rsi = regs[REG_RSI];
+  out->rdi = regs[REG_RDI];
+  out->r8 = regs[REG_R8];
+  out->r9 = regs[REG_R9];
+  out->r10 = regs[REG_R10];
+  out->r11 = regs[REG_R11];
+  out->r12 = regs[REG_R12];
+  out->r13 = regs[REG_R13];
+  out->r14 = regs[REG_R14];
+  out->r15 = regs[REG_R15];
+
+  out->rip = regs[REG_RIP];
+
+  out->flt_save.control_word = fpregs->cwd;
+  out->flt_save.status_word = fpregs->swd;
+  out->flt_save.tag_word = fpregs->ftw;
+  out->flt_save.error_opcode = fpregs->fop;
+  out->flt_save.error_offset = fpregs->rip;
+  out->flt_save.data_offset = fpregs->rdp;
+  out->flt_save.error_selector = 0; // We don't have this.
+  out->flt_save.data_selector = 0;  // We don't have this.
+  out->flt_save.mx_csr = fpregs->mxcsr;
+  out->flt_save.mx_csr_mask = fpregs->mxcsr_mask;
+  memcpy(&out->flt_save.float_registers, &fpregs->_st, 8 * 16);
+  memcpy(&out->flt_save.xmm_registers, &fpregs->_xmm, 16 * 16);
+}
+
+#else
+#error "This code has not been ported to your platform yet."
+#endif
+
+namespace google_breakpad {
+
+class MinidumpWriter {
+ public:
+  MinidumpWriter(const char* filename,
+                 pid_t crashing_pid,
+                 const ExceptionHandler::CrashContext* context)
+      : filename_(filename),
+        siginfo_(&context->siginfo),
+        ucontext_(&context->context),
+        float_state_(&context->float_state),
+        crashing_tid_(context->tid),
+        dumper_(crashing_pid) {
+  }
+
+  bool Init() {
+    return dumper_.Init() && minidump_writer_.Open(filename_) &&
+           dumper_.ThreadsSuspend();
+  }
+
+  ~MinidumpWriter() {
+    minidump_writer_.Close();
+    dumper_.ThreadsResume();
+  }
+
+  bool Dump() {
+    // A minidump file contains a number of tagged streams. This is the number
+    // of stream which we write.
+    static const unsigned kNumWriters = 11;
+
+    TypedMDRVA<MDRawHeader> header(&minidump_writer_);
+    TypedMDRVA<MDRawDirectory> dir(&minidump_writer_);
+    if (!header.Allocate())
+      return false;
+    if (!dir.AllocateArray(kNumWriters))
+      return false;
+    memset(header.get(), 0, sizeof(MDRawHeader));
+
+    header.get()->signature = MD_HEADER_SIGNATURE;
+    header.get()->version = MD_HEADER_VERSION;
+    header.get()->time_date_stamp = time(NULL);
+    header.get()->stream_count = kNumWriters;
+    header.get()->stream_directory_rva = dir.position();
+
+    unsigned dir_index = 0;
+    MDRawDirectory dirent;
+
+    if (!WriteThreadListStream(&dirent))
+      return false;
+    dir.CopyIndex(dir_index++, &dirent);
+
+    if (!WriteMappings(&dirent))
+      return false;
+    dir.CopyIndex(dir_index++, &dirent);
+
+    if (!WriteExceptionStream(&dirent))
+      return false;
+    dir.CopyIndex(dir_index++, &dirent);
+
+    if (!WriteSystemInfoStream(&dirent))
+      return false;
+    dir.CopyIndex(dir_index++, &dirent);
+
+    dirent.stream_type = MD_LINUX_CPU_INFO;
+    if (!WriteFile(&dirent.location, "/proc/cpuinfo"))
+      NullifyDirectoryEntry(&dirent);
+    dir.CopyIndex(dir_index++, &dirent);
+
+    dirent.stream_type = MD_LINUX_PROC_STATUS;
+    if (!WriteProcFile(&dirent.location, crashing_tid_, "status"))
+      NullifyDirectoryEntry(&dirent);
+    dir.CopyIndex(dir_index++, &dirent);
+
+    dirent.stream_type = MD_LINUX_LSB_RELEASE;
+    if (!WriteFile(&dirent.location, "/etc/lsb-release"))
+      NullifyDirectoryEntry(&dirent);
+    dir.CopyIndex(dir_index++, &dirent);
+
+    dirent.stream_type = MD_LINUX_CMD_LINE;
+    if (!WriteProcFile(&dirent.location, crashing_tid_, "cmdline"))
+      NullifyDirectoryEntry(&dirent);
+    dir.CopyIndex(dir_index++, &dirent);
+
+    dirent.stream_type = MD_LINUX_ENVIRON;
+    if (!WriteProcFile(&dirent.location, crashing_tid_, "environ"))
+      NullifyDirectoryEntry(&dirent);
+    dir.CopyIndex(dir_index++, &dirent);
+
+    dirent.stream_type = MD_LINUX_AUXV;
+    if (!WriteProcFile(&dirent.location, crashing_tid_, "auxv"))
+      NullifyDirectoryEntry(&dirent);
+    dir.CopyIndex(dir_index++, &dirent);
+
+    dirent.stream_type = MD_LINUX_AUXV;
+    if (!WriteProcFile(&dirent.location, crashing_tid_, "maps"))
+      NullifyDirectoryEntry(&dirent);
+    dir.CopyIndex(dir_index++, &dirent);
+
+    // If you add more directory entries, don't forget to update kNumWriters,
+    // above.
+
+    dumper_.ThreadsResume();
+    return true;
+  }
+
+  // Write information about the threads.
+  bool WriteThreadListStream(MDRawDirectory* dirent) {
+    const unsigned num_threads = dumper_.threads().size();
+
+    TypedMDRVA<uint32_t> list(&minidump_writer_);
+    if (!list.AllocateObjectAndArray(num_threads, sizeof(MDRawThread)))
+      return false;
+
+    dirent->stream_type = MD_THREAD_LIST_STREAM;
+    dirent->location = list.location();
+
+    *list.get() = num_threads;
+
+    for (unsigned i = 0; i < num_threads; ++i) {
+      MDRawThread thread;
+      my_memset(&thread, 0, sizeof(thread));
+      thread.thread_id = dumper_.threads()[i];
+      // We have a different source of information for the crashing thread. If
+      // we used the actual state of the thread we would find it running in the
+      // signal handler with the alternative stack, which would be deeply
+      // unhelpful.
+      if (thread.thread_id == crashing_tid_) {
+        const void* stack;
+        size_t stack_len;
+        if (!dumper_.GetStackInfo(&stack, &stack_len, GetStackPointer()))
+          return false;
+        UntypedMDRVA memory(&minidump_writer_);
+        if (!memory.Allocate(stack_len))
+          return false;
+        uint8_t* stack_copy = (uint8_t*) dumper_.allocator()->Alloc(stack_len);
+        dumper_.CopyFromProcess(stack_copy, thread.thread_id, stack, stack_len);
+        memory.Copy(stack_copy, stack_len);
+        thread.stack.start_of_memory_range = (uintptr_t) (stack);
+        thread.stack.memory = memory.location();
+        TypedMDRVA<RawContextCPU> cpu(&minidump_writer_);
+        if (!cpu.Allocate())
+          return false;
+        my_memset(cpu.get(), 0, sizeof(RawContextCPU));
+        CPUFillFromUContext(cpu.get(), ucontext_, float_state_);
+        thread.thread_context = cpu.location();
+        crashing_thread_context_ = cpu.location();
+      } else {
+        ThreadInfo info;
+        if (!dumper_.ThreadInfoGet(dumper_.threads()[i], &info))
+          return false;
+        UntypedMDRVA memory(&minidump_writer_);
+        if (!memory.Allocate(info.stack_len))
+          return false;
+        uint8_t* stack_copy =
+            (uint8_t*) dumper_.allocator()->Alloc(info.stack_len);
+        dumper_.CopyFromProcess(stack_copy, thread.thread_id, info.stack,
+                                info.stack_len);
+        memory.Copy(stack_copy, info.stack_len);
+        thread.stack.start_of_memory_range = (uintptr_t)(info.stack);
+        thread.stack.memory = memory.location();
+        TypedMDRVA<RawContextCPU> cpu(&minidump_writer_);
+        if (!cpu.Allocate())
+          return false;
+        my_memset(cpu.get(), 0, sizeof(RawContextCPU));
+        CPUFillFromThreadInfo(cpu.get(), info);
+        thread.thread_context = cpu.location();
+      }
+
+      list.CopyIndexAfterObject(i, &thread, sizeof(thread));
+    }
+
+    return true;
+  }
+
+  static bool ShouldIncludeMapping(const MappingInfo& mapping) {
+    if (mapping.name[0] == 0 || // we only want modules with filenames.
+        mapping.offset || // we only want to include one mapping per shared lib.
+        mapping.size < 4096) {  // too small to get a signature for.
+      return false;
+    }
+
+    return true;
+  }
+
+  // Write information about the mappings in effect. Because we are using the
+  // minidump format, the information about the mappings is pretty limited.
+  // Because of this, we also include the full, unparsed, /proc/$x/maps file in
+  // another stream in the file.
+  bool WriteMappings(MDRawDirectory* dirent) {
+    const unsigned num_mappings = dumper_.mappings().size();
+    unsigned num_output_mappings = 0;
+
+    for (unsigned i = 0; i < dumper_.mappings().size(); ++i) {
+      const MappingInfo& mapping = *dumper_.mappings()[i];
+      if (ShouldIncludeMapping(mapping))
+        num_output_mappings++;
+    }
+
+    TypedMDRVA<uint32_t> list(&minidump_writer_);
+    if (!list.AllocateObjectAndArray(num_output_mappings, sizeof(MDRawModule)))
+      return false;
+
+    dirent->stream_type = MD_MODULE_LIST_STREAM;
+    dirent->location = list.location();
+    *list.get() = num_output_mappings;
+
+    for (unsigned i = 0, j = 0; i < num_mappings; ++i) {
+      const MappingInfo& mapping = *dumper_.mappings()[i];
+      if (!ShouldIncludeMapping(mapping))
+        continue;
+
+      MDRawModule mod;
+      my_memset(&mod, 0, sizeof(mod));
+      mod.base_of_image = mapping.start_addr;
+      mod.size_of_image = mapping.size;
+      const size_t filepath_len = my_strlen(mapping.name);
+
+      // Figure out file name from path
+      const char* filename_ptr = mapping.name + filepath_len - 1;
+      while (filename_ptr >= mapping.name) {
+        if (*filename_ptr == '/')
+          break;
+        filename_ptr--;
+      }
+      filename_ptr++;
+      const size_t filename_len = mapping.name + filepath_len - filename_ptr;
+
+      uint8_t cv_buf[MDCVInfoPDB70_minsize + NAME_MAX];
+      uint8_t* cv_ptr = cv_buf;
+      UntypedMDRVA cv(&minidump_writer_);
+      if (!cv.Allocate(MDCVInfoPDB70_minsize + filename_len + 1))
+        return false;
+
+      const uint32_t cv_signature = MD_CVINFOPDB70_SIGNATURE;
+      memcpy(cv_ptr, &cv_signature, sizeof(cv_signature));
+      cv_ptr += sizeof(cv_signature);
+
+      {
+        // We XOR the first page of the file to get a signature for it.
+        uint8_t xor_buf[sizeof(MDGUID)];
+        size_t done = 0;
+        uint8_t* signature = cv_ptr;
+        cv_ptr += sizeof(xor_buf);
+
+        my_memset(signature, 0, sizeof(xor_buf));
+        while (done < 4096) {
+          dumper_.CopyFromProcess(xor_buf, crashing_tid_,
+                                  (void *) (mod.base_of_image + done),
+                                  sizeof(xor_buf));
+          for (unsigned i = 0; i < sizeof(xor_buf); ++i)
+            signature[i] ^= xor_buf[i];
+          done += sizeof(xor_buf);
+        }
+        my_memset(cv_ptr, 0, sizeof(uint32_t));  // Set age to 0 on Linux.
+        cv_ptr += sizeof(uint32_t);
+      }
+
+      // Write pdb_file_name
+      memcpy(cv_ptr, filename_ptr, filename_len + 1);
+      cv.Copy(cv_buf, MDCVInfoPDB70_minsize + filename_len + 1);
+
+      mod.cv_record = cv.location();
+
+      MDLocationDescriptor ld;
+      if (!minidump_writer_.WriteString(mapping.name, filepath_len, &ld))
+        return false;
+      mod.module_name_rva = ld.rva;
+
+      list.CopyIndexAfterObject(j++, &mod, sizeof(mod));
+    }
+
+    return true;
+  }
+
+  bool WriteExceptionStream(MDRawDirectory* dirent) {
+    TypedMDRVA<MDRawExceptionStream> exc(&minidump_writer_);
+    if (!exc.Allocate())
+      return false;
+    my_memset(exc.get(), 0, sizeof(MDRawExceptionStream));
+
+    dirent->stream_type = MD_EXCEPTION_STREAM;
+    dirent->location = exc.location();
+
+    exc.get()->thread_id = crashing_tid_;
+    exc.get()->exception_record.exception_code = siginfo_->si_signo;
+    exc.get()->exception_record.exception_address =
+        (uintptr_t) siginfo_->si_addr;
+    exc.get()->thread_context = crashing_thread_context_;
+
+    return true;
+  }
+
+  bool WriteSystemInfoStream(MDRawDirectory* dirent) {
+    TypedMDRVA<MDRawSystemInfo> si(&minidump_writer_);
+    if (!si.Allocate())
+      return false;
+    my_memset(si.get(), 0, sizeof(MDRawSystemInfo));
+
+    dirent->stream_type = MD_SYSTEM_INFO_STREAM;
+    dirent->location = si.location();
+
+    WriteCPUInformation(si.get());
+    WriteOSInformation(si.get());
+
+    return true;
+  }
+
+ private:
+#if defined(__i386)
+  uintptr_t GetStackPointer() {
+    return ucontext_->uc_mcontext.gregs[REG_ESP];
+  }
+#elif defined(__x86_64)
+  uintptr_t GetStackPointer() {
+    return ucontext_->uc_mcontext.gregs[REG_RSP];
+  }
+#else
+#error "This code has not been ported to your platform yet."
+#endif
+
+  void NullifyDirectoryEntry(MDRawDirectory* dirent) {
+    dirent->stream_type = 0;
+    dirent->location.data_size = 0;
+    dirent->location.rva = 0;
+  }
+
+  bool WriteCPUInformation(MDRawSystemInfo* sys_info) {
+    char vendor_id[sizeof(sys_info->cpu.x86_cpu_info.vendor_id) + 1] = {0};
+    static const char vendor_id_name[] = "vendor_id";
+    static const size_t vendor_id_name_length = sizeof(vendor_id_name) - 1;
+
+    struct CpuInfoEntry {
+      const char* info_name;
+      int value;
+      bool found;
+    } cpu_info_table[] = {
+      { "processor", -1, false },
+      { "model", 0, false },
+      { "stepping",  0, false },
+      { "cpuid level", 0, false },
+    };
+
+    // processor_architecture should always be set, do this first
+    sys_info->processor_architecture =
+#if defined(__i386)
+        MD_CPU_ARCHITECTURE_X86;
+#elif defined(__x86_64)
+        MD_CPU_ARCHITECTURE_AMD64;
+#else
+#error "Unknown CPU arch"
+#endif
+
+    const int fd = sys_open("/proc/cpuinfo", O_RDONLY, 0);
+    if (fd < 0)
+      return false;
+
+    {
+      PageAllocator allocator;
+      LineReader* const line_reader = new(allocator) LineReader(fd);
+      const char* line;
+      unsigned line_len;
+      while (line_reader->GetNextLine(&line, &line_len)) {
+        for (size_t i = 0;
+             i < sizeof(cpu_info_table) / sizeof(cpu_info_table[0]);
+             i++) {
+          CpuInfoEntry* entry = &cpu_info_table[i];
+          if (entry->found)
+            continue;
+          if (!strncmp(line, entry->info_name, strlen(entry->info_name))) {
+            const char* value = strchr(line, ':');
+            if (!value)
+              continue;
+
+            // the above strncmp only matches the prefix, it might be the wrong
+            // line. i.e. we matched "model name" instead of "model".
+            // check and make sure there is only spaces between the prefix and
+            // the colon.
+            const char* space_ptr = line + strlen(entry->info_name);
+            for (; space_ptr < value; space_ptr++) {
+              if (!isspace(*space_ptr)) {
+                break;
+              }
+            }
+            if (space_ptr != value)
+              continue;
+
+            sscanf(++value, " %d", &(entry->value));
+            entry->found = true;
+          }
+        }
+
+        // special case for vendor_id
+        if (!strncmp(line, vendor_id_name, vendor_id_name_length)) {
+          const char* value = strchr(line, ':');
+          if (!value)
+            goto popline;
+
+          // skip ':" and all the spaces that follows
+          do {
+            value++;
+          } while (isspace(*value));
+
+          if (*value) {
+            size_t length = strlen(value);
+            if (length == 0)
+              goto popline;
+            // we don't want the trailing newline
+            if (value[length - 1] == '\n')
+              length--;
+            // ensure we have space for the value
+            if (length < sizeof(vendor_id))
+              strncpy(vendor_id, value, length);
+          }
+        }
+
+popline:
+        line_reader->PopLine(line_len);
+      }
+      sys_close(fd);
+    }
+
+    // make sure we got everything we wanted
+    for (size_t i = 0;
+         i < sizeof(cpu_info_table) / sizeof(cpu_info_table[0]);
+         i++) {
+      if (!cpu_info_table[i].found) {
+        return false;
+      }
+    }
+    // /proc/cpuinfo contains cpu id, change it into number by adding one.
+    cpu_info_table[0].value++;
+
+    sys_info->number_of_processors = cpu_info_table[0].value;
+    sys_info->processor_level      = cpu_info_table[3].value;
+    sys_info->processor_revision   = cpu_info_table[1].value << 8 |
+                                     cpu_info_table[2].value;
+
+    if (vendor_id[0] != '\0') {
+      memcpy(sys_info->cpu.x86_cpu_info.vendor_id, vendor_id,
+             sizeof(sys_info->cpu.x86_cpu_info.vendor_id));
+    }
+    return true;
+  }
+
+  bool WriteFile(MDLocationDescriptor* result, const char* filename) {
+    const int fd = sys_open(filename, O_RDONLY, 0);
+    if (fd < 0)
+      return false;
+
+    // We can't stat the files because several of the files that we want to
+    // read are kernel seqfiles, which always have a length of zero. So we have
+    // to read as much as we can into a buffer.
+    static const unsigned kMaxFileSize = 1024;
+    uint8_t* data = (uint8_t*) dumper_.allocator()->Alloc(kMaxFileSize);
+
+    size_t done = 0;
+    while (done < kMaxFileSize) {
+      ssize_t r;
+      do {
+        r = sys_read(fd, data + done, kMaxFileSize - done);
+      } while (r == -1 && errno == EINTR);
+
+      if (r < 1)
+        break;
+      done += r;
+    }
+    sys_close(fd);
+
+    if (!done)
+      return false;
+
+    UntypedMDRVA memory(&minidump_writer_);
+    if (!memory.Allocate(done))
+      return false;
+    memory.Copy(data, done);
+    *result = memory.location();
+    return true;
+  }
+
+  bool WriteOSInformation(MDRawSystemInfo* sys_info) {
+    sys_info->platform_id = MD_OS_LINUX;
+
+    struct utsname uts;
+    if (uname(&uts))
+      return false;
+
+    static const size_t buf_len = 512;
+    char buf[buf_len] = {0};
+    size_t space_left = buf_len - 1;
+    const char* info_table[] = {
+      uts.sysname,
+      uts.release,
+      uts.version,
+      uts.machine,
+      NULL
+    };
+    bool first_item = true;
+    for (const char** cur_info = info_table; *cur_info; cur_info++) {
+      static const char* separator = " ";
+      size_t separator_len = strlen(separator);
+      size_t info_len = strlen(*cur_info);
+      if (info_len == 0)
+        continue;
+
+      if (space_left < info_len + (first_item ? 0 : separator_len))
+        break;
+
+      if (!first_item) {
+        strcat(buf, separator);
+        space_left -= separator_len;
+      }
+
+      first_item = false;
+      strcat(buf, *cur_info);
+      space_left -= info_len;
+    }
+
+    MDLocationDescriptor location;
+    if (!minidump_writer_.WriteString(buf, 0, &location))
+      return false;
+    sys_info->csd_version_rva = location.rva;
+
+    return true;
+  }
+
+  bool WriteProcFile(MDLocationDescriptor* result, pid_t pid,
+                     const char* filename) {
+    char buf[80];
+    memcpy(buf, "/proc/", 6);
+    const unsigned pid_len = my_int_len(pid);
+    my_itos(buf + 6, pid, pid_len);
+    buf[6 + pid_len] = '/';
+    memcpy(buf + 6 + pid_len + 1, filename, my_strlen(filename) + 1);
+    return WriteFile(result, buf);
+  }
+
+  const char* const filename_;  // output filename
+  const siginfo_t* const siginfo_;  // from the signal handler (see sigaction)
+  const struct ucontext* const ucontext_;  // also from the signal handler
+  const struct _libc_fpstate* const float_state_;  // ditto
+  const pid_t crashing_tid_;  // the process which actually crashed
+  LinuxDumper dumper_;
+  MinidumpFileWriter minidump_writer_;
+  MDLocationDescriptor crashing_thread_context_;
+};
+
+bool WriteMinidump(const char* filename, pid_t crashing_process,
+                   const void* blob, size_t blob_size) {
+  if (blob_size != sizeof(ExceptionHandler::CrashContext))
+    return false;
+  const ExceptionHandler::CrashContext* context =
+      reinterpret_cast<const ExceptionHandler::CrashContext*>(blob);
+  MinidumpWriter writer(filename, crashing_process, context);
+  if (!writer.Init())
+    return false;
+  return writer.Dump();
+}
+
+}  // namespace google_breakpad

src/client/linux/minidump_writer/minidump_writer.h

@@ -1,8 +1,6 @@
-// Copyright (c) 2006, Google Inc.
+// Copyright (c) 2009, Google Inc.
 // All rights reserved.
 //
-// Author: Li Liu
-//
 // Redistribution and use in source and binary forms, with or without
 // modification, are permitted provided that the following conditions are
 // met:
@@ -29,45 +27,27 @@
 // (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
 // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
-#ifndef CLIENT_LINUX_HANDLER_MINIDUMP_GENERATOR_H__
-#define CLIENT_LINUX_HANDLER_MINIDUMP_GENERATOR_H__
+#ifndef CLIENT_LINUX_MINIDUMP_WRITER_MINIDUMP_WRITER_H_
+#define CLIENT_LINUX_MINIDUMP_WRITER_MINIDUMP_WRITER_H_
 
 #include <stdint.h>
-
-#include "google_breakpad/common/breakpad_types.h"
-#include "processor/scoped_ptr.h"
-
-struct sigcontext;
+#include <unistd.h>
 
 namespace google_breakpad {
 
+// Write a minidump to the filesystem. This function does not malloc nor use
+// libc functions which may. Thus, it can be used in contexts where the state
+// of the heap may be corrupt.
+//   filename: the filename to write to. This is opened O_EXCL and fails if
+//     open fails.
+//   crashing_process: the pid of the crashing process. This must be trusted.
+//   blob: a blob of data from the crashing process. See exception_handler.h
+//   blob_size: the length of |blob|, in bytes
 //
-// MinidumpGenerator
-//
-// Write a minidump to file based on the signo and sig_ctx.
-// A minidump generator should be created before any exception happen.
-//
-class MinidumpGenerator {
-  public:
-   MinidumpGenerator();
-
-   ~MinidumpGenerator();
-
-   // Write minidump.
-   bool WriteMinidumpToFile(const char *file_pathname,
-                            int signo,
-                            uintptr_t sighandler_ebp,
-                            struct sigcontext **sig_ctx) const;
-  private:
-   // Allocate memory for stack.
-   void AllocateStack();
-
-  private:
-   // Stack size of the writer thread.
-   static const int kStackSize = 1024 * 1024;
-   scoped_array<char> stack_;
-};
+// Returns true iff successful.
+bool WriteMinidump(const char* filename, pid_t crashing_process,
+                   const void* blob, size_t blob_size);
 
 }  // namespace google_breakpad
 
-#endif   // CLIENT_LINUX_HANDLER_MINIDUMP_GENERATOR_H__
+#endif  // CLIENT_LINUX_MINIDUMP_WRITER_MINIDUMP_WRITER_H_

src/client/linux/minidump_writer/minidump_writer_unittest.cc

@@ -1,8 +1,6 @@
-// Copyright (c) 2006, Google Inc.
+// Copyright (c) 2009, Google Inc.
 // All rights reserved.
 //
-// Author: Li Liu
-//
 // Redistribution and use in source and binary forms, with or without
 // modification, are permitted provided that the following conditions are
 // met:
@@ -29,58 +27,53 @@
 // (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
 // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
-#include <pthread.h>
 #include <unistd.h>
+#include <sys/syscall.h>
 
-#include <cassert>
-#include <cstdio>
-#include <cstdlib>
-#include <cstring>
-
-#include "client/linux/handler/minidump_generator.h"
+#include "client/linux/handler/exception_handler.h"
+#include "client/linux/minidump_writer/minidump_writer.h"
+#include "breakpad_googletest_includes.h"
 
 using namespace google_breakpad;
 
-// Thread use this to see if it should stop working.
-static bool should_exit = false;
+// This provides a wrapper around system calls which may be
+// interrupted by a signal and return EINTR. See man 7 signal.
+#define HANDLE_EINTR(x) ({ \
+  typeof(x) __eintr_result__; \
+  do { \
+    __eintr_result__ = x; \
+  } while (__eintr_result__ == -1 && errno == EINTR); \
+  __eintr_result__;\
+})
 
-static void foo2(int arg) {
-  // Stack variable, used for debugging stack dumps.
-  int c = arg;
-  c = 0xcccccccc;
-  while (!should_exit)
-    sleep(1);
+namespace {
+typedef testing::Test MinidumpWriterTest;
 }
 
-static void foo(int arg) {
-  // Stack variable, used for debugging stack dumps.
-  int b = arg;
-  b = 0xbbbbbbbb;
-  foo2(b);
-}
+TEST(MinidumpWriterTest, Setup) {
+  int fds[2];
+  ASSERT_NE(-1, pipe(fds));
 
-static void *thread_main(void *) {
-  // Stack variable, used for debugging stack dumps.
-  int a = 0xaaaaaaaa;
-  foo(a);
-  return NULL;
-}
-
-static void CreateThread(int num) {
-  pthread_t h;
-  for (int i = 0; i < num; ++i) {
-    pthread_create(&h, NULL, thread_main, NULL);
-    pthread_detach(h);
+  const pid_t child = fork();
+  if (child == 0) {
+    close(fds[1]);
+    char b;
+    HANDLE_EINTR(read(fds[0], &b, sizeof(b)));
+    close(fds[0]);
+    syscall(__NR_exit);
   }
-}
+  close(fds[0]);
 
-int main(int argc, char *argv[]) {
-  CreateThread(10);
-  google_breakpad::MinidumpGenerator mg;
-  if (mg.WriteMinidumpToFile("minidump_test.out", -1, 0, NULL))
-    printf("Succeeded written minidump\n");
-  else
-    printf("Failed to write minidump\n");
-  should_exit = true;
-  return 0;
+  ExceptionHandler::CrashContext context;
+  memset(&context, 0, sizeof(context));
+
+  char templ[] = "/tmp/minidump-writer-unittest-XXXXXX";
+  mktemp(templ);
+  ASSERT_TRUE(WriteMinidump(templ, child, &context, sizeof(context)));
+  struct stat st;
+  ASSERT_EQ(stat(templ, &st), 0);
+  ASSERT_GT(st.st_size, 0u);
+  unlink(templ);
+
+  close(fds[1]);
 }

src/client/mac/handler/minidump_test.xcodeproj/project.pbxproj

@@ -133,9 +133,9 @@
 		F9721F380E8B0CFC00D7E813 /* dump_syms.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; name = dump_syms.h; path = ../../../common/mac/dump_syms.h; sourceTree = SOURCE_ROOT; };
 		F9721F390E8B0D0D00D7E813 /* dump_syms.mm */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.objcpp; name = dump_syms.mm; path = ../../../common/mac/dump_syms.mm; sourceTree = SOURCE_ROOT; };
 		F9721F6B0E8B0D7000D7E813 /* Cocoa.framework */ = {isa = PBXFileReference; lastKnownFileType = wrapper.framework; name = Cocoa.framework; path = /System/Library/Frameworks/Cocoa.framework; sourceTree = "<absolute>"; };
-		F9721F760E8B0DC700D7E813 /* bytereader.cc */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; name = bytereader.cc; path = ../../../common/mac/dwarf/bytereader.cc; sourceTree = SOURCE_ROOT; };
-		F9721F770E8B0DC700D7E813 /* dwarf2reader.cc */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; name = dwarf2reader.cc; path = ../../../common/mac/dwarf/dwarf2reader.cc; sourceTree = SOURCE_ROOT; };
-		F9721F780E8B0DC700D7E813 /* functioninfo.cc */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; name = functioninfo.cc; path = ../../../common/mac/dwarf/functioninfo.cc; sourceTree = SOURCE_ROOT; };
+		F9721F760E8B0DC700D7E813 /* bytereader.cc */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; name = bytereader.cc; path = ../../../common/dwarf/bytereader.cc; sourceTree = SOURCE_ROOT; };
+		F9721F770E8B0DC700D7E813 /* dwarf2reader.cc */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; name = dwarf2reader.cc; path = ../../../common/dwarf/dwarf2reader.cc; sourceTree = SOURCE_ROOT; };
+		F9721F780E8B0DC700D7E813 /* functioninfo.cc */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; name = functioninfo.cc; path = ../../../common/dwarf/functioninfo.cc; sourceTree = SOURCE_ROOT; };
 		F9721FA10E8B0E2300D7E813 /* SenTestingKit.framework */ = {isa = PBXFileReference; lastKnownFileType = wrapper.framework; name = SenTestingKit.framework; path = /System/Library/Frameworks/SenTestingKit.framework; sourceTree = "<absolute>"; };
 		F9721FA80E8B0E4800D7E813 /* md5.c */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.c; name = md5.c; path = ../../../common/md5.c; sourceTree = SOURCE_ROOT; };
 		F982089A0DB3280D0017AECA /* breakpad_nlist_test.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = breakpad_nlist_test.h; sourceTree = "<group>"; };

src/client/mac/sender/crash_report_sender.m

@@ -112,9 +112,12 @@ NSString *const kDefaultServerType = @"google";
 @implementation NSTextField (ResizabilityExtentions)
 - (float)breakpad_adjustHeightToFit {
   NSRect oldFrame = [self frame];
+  // Starting with the 10.5 SDK, height won't grow, so make it huge to start.
+  NSRect presizeFrame = oldFrame;
+  presizeFrame.size.height = MAXFLOAT;
   // sizeToFit will blow out the width rather than making the field taller, so
   // we do it manually.
-  NSSize newSize = [[self cell] cellSizeForBounds:oldFrame];
+  NSSize newSize = [[self cell] cellSizeForBounds:presizeFrame];
   NSRect newFrame = NSMakeRect(oldFrame.origin.x, oldFrame.origin.y,
                                NSWidth(oldFrame), newSize.height);
   [self setFrame:newFrame];

src/client/minidump_file_writer.cc

@@ -37,6 +37,8 @@
 #include <string.h>
 #include <unistd.h>
 
+#include "common/linux/linux_syscall_support.h"
+#include "common/linux/linux_libc_support.h"
 #include "client/minidump_file_writer-inl.h"
 #include "common/string_conversion.h"
 
@@ -53,7 +55,11 @@ MinidumpFileWriter::~MinidumpFileWriter() {
 
 bool MinidumpFileWriter::Open(const char *path) {
   assert(file_ == -1);
+#if __linux__
+  file_ = sys_open(path, O_WRONLY | O_CREAT | O_EXCL, 0600);
+#else
   file_ = open(path, O_WRONLY | O_CREAT | O_EXCL, 0600);
+#endif
 
   return file_ != -1;
 }
@@ -63,7 +69,11 @@ bool MinidumpFileWriter::Close() {
 
   if (file_ != -1) {
     ftruncate(file_, position_);
+#if __linux__
+    result = (sys_close(file_) == 0);
+#else
     result = (close(file_) == 0);
+#endif
     file_ = -1;
   }
 
@@ -227,9 +237,16 @@ bool MinidumpFileWriter::Copy(MDRVA position, const void *src, ssize_t size) {
     return false;
 
   // Seek and write the data
-  if (lseek(file_, position, SEEK_SET) == static_cast<off_t>(position))
-    if (write(file_, src, size) == size)
+#if __linux__
+  if (sys_lseek(file_, position, SEEK_SET) == static_cast<off_t>(position)) {
+    if (sys_write(file_, src, size) == size) {
+#else
+  if (lseek(file_, position, SEEK_SET) == static_cast<off_t>(position)) {
+    if (write(file_, src, size) == size) {
+#endif
       return true;
+    }
+  }
 
   return false;
 }

src/common/dwarf/bytereader-inl.h

@@ -29,7 +29,9 @@
 #ifndef UTIL_DEBUGINFO_BYTEREADER_INL_H__
 #define UTIL_DEBUGINFO_BYTEREADER_INL_H__
 
-#include "common/mac/dwarf/bytereader.h"
+#include <cassert>
+
+#include "common/dwarf/bytereader.h"
 
 namespace dwarf2reader {
 
@@ -37,9 +39,11 @@ inline uint8 ByteReader::ReadOneByte(const char* buffer) const {
   return buffer[0];
 }
 
-inline uint16 ByteReader::ReadTwoBytes(const char* buffer) const {
-  const uint16 buffer0 = static_cast<uint16>(buffer[0]);
-  const uint16 buffer1 = static_cast<uint16>(buffer[1]);
+inline uint16 ByteReader::ReadTwoBytes(const char* signed_buffer) const {
+  const unsigned char *buffer
+    = reinterpret_cast<const unsigned char *>(signed_buffer);
+  const uint16 buffer0 = buffer[0];
+  const uint16 buffer1 = buffer[1];
   if (endian_ == ENDIANNESS_LITTLE) {
     return buffer0 | buffer1 << 8;
   } else {
@@ -47,11 +51,13 @@ inline uint16 ByteReader::ReadTwoBytes(const char* buffer) const {
   }
 }
 
-inline uint64 ByteReader::ReadFourBytes(const char* buffer) const {
-  const uint32 buffer0 = static_cast<uint32>(buffer[0]);
-  const uint32 buffer1 = static_cast<uint32>(buffer[1]);
-  const uint32 buffer2 = static_cast<uint32>(buffer[2]);
-  const uint32 buffer3 = static_cast<uint32>(buffer[3]);
+inline uint64 ByteReader::ReadFourBytes(const char* signed_buffer) const {
+  const unsigned char *buffer
+    = reinterpret_cast<const unsigned char *>(signed_buffer);
+  const uint32 buffer0 = buffer[0];
+  const uint32 buffer1 = buffer[1];
+  const uint32 buffer2 = buffer[2];
+  const uint32 buffer3 = buffer[3];
   if (endian_ == ENDIANNESS_LITTLE) {
     return buffer0 | buffer1 << 8 | buffer2 << 16 | buffer3 << 24;
   } else {
@@ -59,15 +65,17 @@ inline uint64 ByteReader::ReadFourBytes(const char* buffer) const {
   }
 }
 
-inline uint64 ByteReader::ReadEightBytes(const char* buffer) const {
-  const uint64 buffer0 = static_cast<uint64>(buffer[0]);
-  const uint64 buffer1 = static_cast<uint64>(buffer[1]);
-  const uint64 buffer2 = static_cast<uint64>(buffer[2]);
-  const uint64 buffer3 = static_cast<uint64>(buffer[3]);
-  const uint64 buffer4 = static_cast<uint64>(buffer[4]);
-  const uint64 buffer5 = static_cast<uint64>(buffer[5]);
-  const uint64 buffer6 = static_cast<uint64>(buffer[6]);
-  const uint64 buffer7 = static_cast<uint64>(buffer[7]);
+inline uint64 ByteReader::ReadEightBytes(const char* signed_buffer) const {
+  const unsigned char *buffer
+    = reinterpret_cast<const unsigned char *>(signed_buffer);
+  const uint64 buffer0 = buffer[0];
+  const uint64 buffer1 = buffer[1];
+  const uint64 buffer2 = buffer[2];
+  const uint64 buffer3 = buffer[3];
+  const uint64 buffer4 = buffer[4];
+  const uint64 buffer5 = buffer[5];
+  const uint64 buffer6 = buffer[6];
+  const uint64 buffer7 = buffer[7];
   if (endian_ == ENDIANNESS_LITTLE) {
     return buffer0 | buffer1 << 8 | buffer2 << 16 | buffer3 << 24 |
       buffer4 << 32 | buffer5 << 40 | buffer6 << 48 | buffer7 << 56;

src/common/dwarf/bytereader.cc

@@ -26,9 +26,9 @@
 // (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
 // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
-#include "common/mac/dwarf/bytereader-inl.h"
+#include "common/dwarf/bytereader-inl.h"
 
-#include "common/mac/dwarf/bytereader.h"
+#include "common/dwarf/bytereader.h"
 
 namespace dwarf2reader {
 

src/common/dwarf/bytereader.h

@@ -26,11 +26,11 @@
 // (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
 // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
-#ifndef COMMON_MAC_DWARF_BYTEREADER_H__
-#define COMMON_MAC_DWARF_BYTEREADER_H__
+#ifndef COMMON_DWARF_BYTEREADER_H__
+#define COMMON_DWARF_BYTEREADER_H__
 
 #include <string>
-#include "common/mac/dwarf/types.h"
+#include "common/dwarf/types.h"
 
 namespace dwarf2reader {
 
@@ -129,4 +129,4 @@ class ByteReader {
 
 }  // namespace dwarf2reader
 
-#endif  // COMMON_MAC_DWARF_BYTEREADER_H__
+#endif  // COMMON_DWARF_BYTEREADER_H__

src/common/dwarf/dwarf2diehandler.cc

@@ -0,0 +1,171 @@
+// Copyright 2009 Google Inc. All Rights Reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// Implementation of dwarf2reader::DieDispatcher class.
+
+#include <cassert>
+
+#include "common/dwarf/dwarf2diehandler.h"
+
+namespace dwarf2reader {
+
+DIEDispatcher::~DIEDispatcher() {
+  while (! die_handlers_.empty()) {
+    HandlerStack &entry = die_handlers_.top();
+    if (entry.handler_ != root_handler_)
+      delete entry.handler_;
+    die_handlers_.pop();
+  }
+}
+
+bool DIEDispatcher::StartCompilationUnit(uint64 offset, uint8 address_size,
+                                         uint8 offset_size, uint64 cu_length,
+                                         uint8 dwarf_version) {
+  return root_handler_->StartCompilationUnit(offset, address_size,
+                                             offset_size, cu_length,
+                                             dwarf_version);
+}
+
+bool DIEDispatcher::StartDIE(uint64 offset, enum DwarfTag tag,
+                             const AttributeList& attrs) {
+  // The stack entry for the parent of this DIE, if there is one.
+  HandlerStack *parent = die_handlers_.empty() ? NULL : &die_handlers_.top();
+
+  // Does this call indicate that we're done receiving the parent's
+  // attributes' values?  If so, call its EndAttributes member function.
+  if (parent && parent->handler_ && ! parent->reported_attributes_end_) {
+    parent->reported_attributes_end_ = true;
+    if (! parent->handler_->EndAttributes()) {
+      // Finish off this handler now. and edit *PARENT to indicate that
+      // we don't want to visit any of the children.
+      parent->handler_->Finish();
+      if (parent->handler_ != root_handler_) delete parent->handler_;
+      parent->handler_ = NULL;
+      return false;
+    }
+  }
+
+  // Find a handler for this DIE.
+  DIEHandler *handler;
+  if (parent) {
+    if (parent->handler_)
+      // Ask the parent to find a handler.
+      handler = parent->handler_->FindChildHandler(offset, tag, attrs);
+    else
+      // No parent handler means we're not interested in any of our
+      // children.
+      handler = NULL;
+  } else {
+    // This is the root DIE.  For a non-root DIE, the parent's handler
+    // decides whether to visit it, but the root DIE has no parent
+    // handler, so we have a special method on the root DIE handler
+    // itself to decide.
+    if (root_handler_->StartRootDIE(offset, tag, attrs))
+      handler = root_handler_;
+    else
+      handler = NULL;
+  }
+
+  // Push a handler stack entry for this new handler.  As an
+  // optimization, we don't push NULL-handler entries on top of other
+  // NULL-handler entries; we just let the oldest such entry stand for
+  // the whole subtree.
+  if (handler || (parent && parent->handler_)) {
+    HandlerStack entry;
+    entry.offset_ = offset;
+    entry.handler_ = handler;
+    entry.reported_attributes_end_ = false;
+    die_handlers_.push(entry);
+  }
+
+  return handler != NULL;
+}
+
+void DIEDispatcher::EndDIE(uint64 offset) {
+  assert(! die_handlers_.empty());
+  HandlerStack *entry = &die_handlers_.top();
+  if (entry->handler_) {
+    // This entry had better be the handler for this DIE.
+    assert(entry->offset_ == offset);
+    // If a DIE has no children, this EndDIE call indicates that we're
+    // done receiving its attributes' values.
+    if (! entry->reported_attributes_end_)
+      entry->handler_->EndAttributes(); // Ignore return value: no children.
+    entry->handler_->Finish();
+    if (entry->handler_ != root_handler_) delete entry->handler_;
+  } else {
+    // If this DIE is within a tree we're ignoring, then don't pop the
+    // handler stack: that entry stands for the whole tree.
+    if (entry->offset_ != offset)
+      return;
+  }
+  die_handlers_.pop();
+}
+
+void DIEDispatcher::ProcessAttributeUnsigned(uint64 offset,
+                                             enum DwarfAttribute attr,
+                                             enum DwarfForm form,
+                                             uint64 data) {
+  HandlerStack &current = die_handlers_.top();
+  // This had better be an attribute of the DIE we were meant to handle.
+  assert(offset == current.offset_);
+  current.handler_->ProcessAttributeUnsigned(attr, form, data);
+}
+
+void DIEDispatcher::ProcessAttributeSigned(uint64 offset,
+                                           enum DwarfAttribute attr,
+                                           enum DwarfForm form,
+                                           int64 data) {
+  HandlerStack &current = die_handlers_.top();
+  // This had better be an attribute of the DIE we were meant to handle.
+  assert(offset == current.offset_);
+  current.handler_->ProcessAttributeSigned(attr, form, data);
+}
+
+void DIEDispatcher::ProcessAttributeBuffer(uint64 offset,
+                                           enum DwarfAttribute attr,
+                                           enum DwarfForm form,
+                                           const char* data,
+                                           uint64 len) {
+  HandlerStack &current = die_handlers_.top();
+  // This had better be an attribute of the DIE we were meant to handle.
+  assert(offset == current.offset_);
+  current.handler_->ProcessAttributeBuffer(attr, form, data, len);
+}
+
+void DIEDispatcher::ProcessAttributeString(uint64 offset,
+                                           enum DwarfAttribute attr,
+                                           enum DwarfForm form,
+                                           const string& data) {
+  HandlerStack &current = die_handlers_.top();
+  // This had better be an attribute of the DIE we were meant to handle.
+  assert(offset == current.offset_);
+  current.handler_->ProcessAttributeString(attr, form, data);
+}
+
+} // namespace dwarf2reader

src/common/dwarf/dwarf2diehandler.h

@@ -0,0 +1,323 @@
+// Copyright 2009 Google Inc. All Rights Reserved.  -*- mode: c++ -*-
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// dwarf2reader::CompilationUnit is a simple and direct parser for
+// DWARF data, but its handler interface is not convenient to use.  In
+// particular:
+//
+// - CompilationUnit calls Dwarf2Handler's member functions to report
+//   every attribute's value, regardless of what sort of DIE it is.
+//   As a result, the ProcessAttributeX functions end up looking like
+//   this:
+//
+//     switch (parent_die_tag) {
+//       case DW_TAG_x:
+//         switch (attribute_name) {
+//           case DW_AT_y:
+//             handle attribute y of DIE type x
+//           ...
+//         } break;
+//       ...
+//     } 
+//
+//   In C++ it's much nicer to use virtual function dispatch to find
+//   the right code for a given case than to switch on the DIE tag
+//   like this.
+//
+// - Processing different kinds of DIEs requires different sets of
+//   data.  It would be nice to be able to have separate classes for
+//   separate kinds of DIEs, each with the members appropriate to its
+//   role, instead of having one handler class that needs to hold data
+//   for all every DIE type.
+//
+// - It would be nice to have separate handler objects for separate
+//   DIEs, instead of a single handler instance required to keep track
+//   of everything.
+//
+// - It's not convenient to take some action after all attributes have
+//   been seen, but before visiting any children.  The only indication
+//   you have that a DIE's attribute list is complete is that you get
+//   either a StartDIE or an EndDIE call.
+//
+// - It's not convenient to make use of the tree structure of the
+//   DIEs.  Skipping all the children of a given die requires
+//   maintaining state and returning false from StartDIE until we get
+//   an EndDIE call with the appropriate offset.  And it's not
+//   convenient to maintain the stack of contexts for the DIEs we have
+//   decided to enter.
+//
+// This interface tries to take care of all that.  (How'd you guess?)
+//
+// Using the classes here, you provide an initial handler for the root
+// DIE of the compilation unit.  Each handler receives its DIE's
+// attributes, and provides fresh handler objects for children of
+// interest, if any.
+//
+// You use them as follows:
+//
+// - Define handler classes specialized to particular DIE types of
+//   interest.  These handler classes must inherit from the DIEHandler
+//   class, defined below.  Thus:
+//
+//     class My_DW_TAG_X_Handler: public DIEHandler { ... };
+//     class My_DW_TAG_Y_Handler: public DIEHandler { ... };
+//
+//   DIEHandler subclasses needn't correspond exactly to single DIE
+//   types, as shown here; the point is that you can write different
+//   classes for different kinds of DIEs.
+//
+// - In particular, define a DIE handler class for the compilation
+//   unit's root DIE, that inherits from the RootDIEHandler class
+//
+//     class My_DW_TAG_compile_unit_Handler: public RootDIEHandler { ... };
+//
+//   RootDIEHandler inherits from DIEHandler, adding a few additional
+//   member functions for examining the compilation unit as a whole,
+//   and other quirks of rootness.
+//
+// - Then, create a DIEDispatcher instance, passing it an instance of
+//   your root DIE handler, and use that as the
+//   dwarf2reader::CompilationUnit's handler:
+//
+//     My_DW_TAG_compile_unit_Handler root_die_handler(...);
+//     DIEDispatcher die_dispatcher(&root_die_handler);
+//     CompilationUnit reader(sections, offset, bytereader, &die_dispatcher);
+//
+//   Here, 'die_dispatcher' acts as a shim between 'reader' and the
+//   various DIE-specific handlers.
+//
+// - When you call reader.Start(), die_dispatcher behaves as follows,
+//   starting with your root die handler and the compilation unit's
+//   root DIE:
+//
+//   - It calls the handler's ProcessAttributeX member functions for
+//     each of the DIE's attributes.
+//
+//   - It calls the handler's EndAttributes member function.  This
+//     should return true if any of the DIE's children should be
+//     visited, in which case:
+//
+//     - die_dispatcher calls the handler's FindChildHandler member
+//       function.  If that returns NULL, die_dispatcher ignores that
+//       child and its descendants.  Otherwise, FindChildHandler
+//       returns a pointer to a DIEHandler instance; die_dispatcher
+//       uses that handler to process the child, using this procedure
+//       recursively.
+// 
+//   - When die_dispatcher has finished processing all the DIE's
+//     children, it invokes the handler's Finish() member function,
+//     and destroys the handler.  (As a special case, it doesn't
+//     destroy the root DIE handler.)
+// 
+// This allows the code for handling a particular kind of DIE to be
+// gathered together in a single class, makes it easy to skip all the
+// children or individual children of a particular DIE, and provides
+// appropriate parental context for each die.
+
+#ifndef COMMON_DWARF_DWARF2DIEHANDLER_H__
+#define COMMON_DWARF_DWARF2DIEHANDLER_H__
+
+#include <stack>
+
+#include "common/dwarf/types.h"
+#include "common/dwarf/dwarf2enums.h"
+#include "common/dwarf/dwarf2reader.h"
+
+namespace dwarf2reader {
+
+// A parent class for handlers for specific DIE types.
+// The series of calls made on a DIE handler is as follows:
+//
+// - construction, by the parent DIE's handler
+// - for each attribute of the DIE:
+//   - ProcessAttributeX()
+// - EndAttributes()
+// - if that returned true, then for each child:
+//   - FindChildHandler()
+//   - if that returns non-NULL:
+//     - this sequence inserted recursively
+// - Finish()
+class DIEHandler {
+ public:
+  DIEHandler() { }
+  virtual ~DIEHandler() { }
+
+  // When we visit a DIE, we first use these member functions to
+  // report the DIE's attributes and their values.  These have the
+  // same restrictions as the corresponding member functions of
+  // dwarf2reader::Dwarf2Handler.
+  //
+  // The default definitions ignore the values they are passed.
+  virtual void ProcessAttributeUnsigned(enum DwarfAttribute attr,
+                                        enum DwarfForm form,
+                                        uint64 data) { }
+  virtual void ProcessAttributeSigned(enum DwarfAttribute attr,
+                                      enum DwarfForm form,
+                                      int64 data) { }
+  virtual void ProcessAttributeBuffer(enum DwarfAttribute attr,
+                                      enum DwarfForm form,
+                                      const char* data,
+                                      uint64 len) { }
+  virtual void ProcessAttributeString(enum DwarfAttribute attr,
+                                      enum DwarfForm form,
+                                      const string& data) { }
+
+  // Once we have reported all the DIE's attributes' values, we call
+  // this member function.  If it returns false, we skip all the DIE's
+  // children.  If it returns true, we call FindChildHandler on each
+  // child.  If that returns a handler object, we use that to visit
+  // the child; otherwise, we skip it.
+  //
+  // The default definition applies FindChildHandler to all children.
+  virtual bool EndAttributes() { return true; }
+
+  // If EndAttributes returns true to indicate that some of the DIE's
+  // children might be of interest, then we apply this function to
+  // each of the DIE's children.  If it returns a handler object, then
+  // we use that to visit the child.  If it returns NULL, we skip that
+  // child (and its children).
+  //
+  // OFFSET is the offset of the child; TAG indicates what kind of DIE
+  // it is; and ATTRS is the list of attributes the DIE will have, and
+  // their forms (their values are not provided).
+  //
+  // The default definition doesn't handle any children.
+  virtual DIEHandler *FindChildHandler(uint64 offset, enum DwarfTag tag,
+                                       const AttributeList &attrs) {
+    return NULL;
+  }
+
+  // When we are done processing a DIE, we call this member function.
+  // This happens after the EndAttributes call, all FindChildHandler
+  // calls (if any), and all operations on the children themselves (if
+  // any).
+  virtual void Finish() { };
+};
+
+// A subclass of DIEHandler, with additional kludges for handling the
+// compilation unit's root die.
+class RootDIEHandler: public DIEHandler {
+ public:
+  RootDIEHandler() { }
+  virtual ~RootDIEHandler() { }
+
+  // We pass the values reported via StartCompilationUnit to this
+  // member function, and skip the entire compilation unit if it
+  // returns false.  So the root DIE handler is actually also a
+  // compilation unit handler.  The default definition always visits
+  // the compilation unit.
+  virtual bool StartCompilationUnit(uint64 offset, uint8 address_size,
+                                    uint8 offset_size, uint64 cu_length,
+                                    uint8 dwarf_version) { return true; }
+
+  // For the root DIE handler only, we pass the offset, tag and
+  // attributes of the compilation unit's root DIE.  This is the only
+  // way the root DIE handler can find the root DIE's tag.  If this
+  // function returns true, we will visit the root DIE using the usual
+  // DIEHandler methods; otherwise, we skip the entire compilation
+  // unit.
+  //
+  // The default definition elects to visit the root DIE.
+  virtual bool StartRootDIE(uint64 offset, enum DwarfTag tag,
+                            const AttributeList& attrs) { return true; }
+};
+
+class DIEDispatcher: public Dwarf2Handler {
+ public:
+  // Create a Dwarf2Handler which uses ROOT_HANDLER as the handler for
+  // the compilation unit's root die, as described for the DIEHandler
+  // class.
+  DIEDispatcher(RootDIEHandler *root_handler) :
+      root_handler_(root_handler) { }
+  // Destroying a DIEDispatcher destroys all active handler objects
+  // except the root handler.
+  ~DIEDispatcher();
+  bool StartCompilationUnit(uint64 offset, uint8 address_size,
+                            uint8 offset_size, uint64 cu_length,
+                            uint8 dwarf_version);
+  bool StartDIE(uint64 offset, enum DwarfTag tag,
+                const AttributeList &attrs);
+  void ProcessAttributeUnsigned(uint64 offset,
+                                enum DwarfAttribute attr,
+                                enum DwarfForm form,
+                                uint64 data);
+  void ProcessAttributeSigned(uint64 offset,
+                              enum DwarfAttribute attr,
+                              enum DwarfForm form,
+                              int64 data);
+  void ProcessAttributeBuffer(uint64 offset,
+                              enum DwarfAttribute attr,
+                              enum DwarfForm form,
+                              const char* data,
+                              uint64 len);
+  void ProcessAttributeString(uint64 offset,
+                              enum DwarfAttribute attr,
+                              enum DwarfForm form,
+                              const string &data);
+  void EndDIE(uint64 offset);
+
+ private:
+
+  // The type of a handler stack entry.  This includes some fields
+  // which don't really need to be on the stack --- they could just be
+  // single data members of DIEDispatcher --- but putting them here
+  // makes it easy to see that the code is correct.
+  struct HandlerStack {
+    // The offset of the DIE for this handler stack entry.
+    uint64 offset_;
+
+    // The object interested in this DIE's attributes and children.
+    // If NULL, we're not interested in either.
+    DIEHandler *handler_;
+
+    // Have we reported the end of this DIE's attributes to the handler?
+    bool reported_attributes_end_;
+  };
+
+  // Stack of DIE attribute handlers.  At StartDIE(D), the top of the
+  // stack is the handler of D's parent, whom we may ask for a handler
+  // for D itself.  At EndDIE(D), the top of the stack is D's handler.
+  // Special cases:
+  //
+  // - Before we've seen the compilation unit's root DIE, the stack is
+  //   empty; we'll call root_handler_'s special member functions, and
+  //   perhaps push root_handler_ on the stack to look at the root's
+  //   immediate children.
+  //
+  // - When we decide to ignore a subtree, we only push an entry on
+  //   the stack for the root of the tree being ignored, rather than
+  //   pushing lots of stack entries with ignore_children_ set.
+  stack<HandlerStack> die_handlers_;
+
+  // The root handler.  We don't push it on die_handlers_ until we
+  // actually get the StartDIE call for the root.
+  RootDIEHandler *root_handler_;
+};
+
+} // namespace dwarf2reader
+#endif  // COMMON_DWARF_DWARF2DIEHANDLER_H__

src/common/dwarf/dwarf2enums.h

@@ -26,8 +26,8 @@
 // (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
 // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
-#ifndef COMMON_MAC_DWARF_DWARF2ENUMS_H__
-#define COMMON_MAC_DWARF_DWARF2ENUMS_H__
+#ifndef COMMON_DWARF_DWARF2ENUMS_H__
+#define COMMON_DWARF_DWARF2ENUMS_H__
 
 namespace dwarf2reader {
 
@@ -486,5 +486,40 @@ enum DwarfOpcode {
   DW_OP_GNU_push_tls_address         =0xe0
 };
 
+// Source languages.  These are values for DW_AT_language.
+enum DwarfLanguage
+  {
+    DW_LANG_none                     =0x0000,
+    DW_LANG_C89                      =0x0001,
+    DW_LANG_C                        =0x0002,
+    DW_LANG_Ada83                    =0x0003,
+    DW_LANG_C_plus_plus              =0x0004,
+    DW_LANG_Cobol74                  =0x0005,
+    DW_LANG_Cobol85                  =0x0006,
+    DW_LANG_Fortran77                =0x0007,
+    DW_LANG_Fortran90                =0x0008,
+    DW_LANG_Pascal83                 =0x0009,
+    DW_LANG_Modula2                  =0x000a,
+    DW_LANG_Java                     =0x000b,
+    DW_LANG_C99                      =0x000c,
+    DW_LANG_Ada95                    =0x000d,
+    DW_LANG_Fortran95                =0x000e,
+    DW_LANG_PLI                      =0x000f,
+    DW_LANG_ObjC                     =0x0010,
+    DW_LANG_ObjC_plus_plus           =0x0011,
+    DW_LANG_UPC                      =0x0012,
+    DW_LANG_D                        =0x0013,
+    // Implementation-defined language code range.
+    DW_LANG_lo_user = 0x8000,
+    DW_LANG_hi_user = 0xffff,
+
+    // Extensions.
+
+    // MIPS assembly language.  The GNU toolchain uses this for all
+    // assembly languages, since there's no generic DW_LANG_ value for that.
+    DW_LANG_Mips_Assembler           =0x8001,
+    DW_LANG_Upc                      =0x8765 // Unified Parallel C
+  };
+
 }  // namespace dwarf2reader
-#endif  // COMMON_MAC_DWARF_DWARF2ENUMS_H__
+#endif  // COMMON_DWARF_DWARF2ENUMS_H__

src/common/dwarf/dwarf2reader.cc

@@ -26,25 +26,15 @@
 // (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
 // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
-#include <ext/hash_map>
 #include <stack>
 #include <utility>
+#include <memory>
+#include <cstring>
 
-#include "common/mac/dwarf/bytereader-inl.h"
-#include "common/mac/dwarf/dwarf2reader.h"
-#include "common/mac/dwarf/bytereader.h"
-#include "common/mac/dwarf/line_state_machine.h"
-
-namespace __gnu_cxx
-{
-  template<> struct hash< std::string >
-  {
-    size_t operator()( const std::string& x ) const
-    {
-      return hash< const char* >()( x.c_str() );
-    }
-  };
-}
+#include "common/dwarf/bytereader-inl.h"
+#include "common/dwarf/dwarf2reader.h"
+#include "common/dwarf/bytereader.h"
+#include "common/dwarf/line_state_machine.h"
 
 namespace dwarf2reader {
 
@@ -87,8 +77,12 @@ void CompilationUnit::ReadAbbrevs() {
   if (abbrevs_)
     return;
 
-  // First get the debug_abbrev section
-  SectionMap::const_iterator iter = sections_.find("__debug_abbrev");
+  // First get the debug_abbrev section.  ".debug_abbrev" is the name
+  // recommended in the DWARF spec, and used on Linux;
+  // "__debug_abbrev" is the name used in Mac OS X Mach-O files.
+  SectionMap::const_iterator iter = sections_.find(".debug_abbrev");
+  if (iter == sections_.end())
+    iter = sections_.find("__debug_abbrev");
   assert(iter != sections_.end());
 
   abbrevs_ = new vector<Abbrev>;
@@ -100,7 +94,9 @@ void CompilationUnit::ReadAbbrevs() {
   const char* abbrev_start = iter->second.first +
                                       header_.abbrev_offset;
   const char* abbrevptr = abbrev_start;
+#ifndef NDEBUG
   const uint64 abbrev_length = iter->second.second - header_.abbrev_offset;
+#endif
 
   while (1) {
     CompilationUnit::Abbrev abbrev;
@@ -273,8 +269,12 @@ void CompilationUnit::ReadHeader() {
 }
 
 uint64 CompilationUnit::Start() {
-  // First get the debug_info section
-  SectionMap::const_iterator iter = sections_.find("__debug_info");
+  // First get the debug_info section.  ".debug_info" is the name
+  // recommended in the DWARF spec, and used on Linux; "__debug_info"
+  // is the name used in Mac OS X Mach-O files.
+  SectionMap::const_iterator iter = sections_.find(".debug_info");
+  if (iter == sections_.end())
+    iter = sections_.find("__debug_info");
   assert(iter != sections_.end());
 
   // Set up our buffer
@@ -304,8 +304,12 @@ uint64 CompilationUnit::Start() {
   // Otherwise, continue by reading our abbreviation entries.
   ReadAbbrevs();
 
-  // Set the string section if we have one.
-  iter = sections_.find("__debug_str");
+  // Set the string section if we have one.  ".debug_str" is the name
+  // recommended in the DWARF spec, and used on Linux; "__debug_str"
+  // is the name used in Mac OS X Mach-O files.
+  iter = sections_.find(".debug_str");
+  if (iter == sections_.end())
+    iter = sections_.find("__debug_str");
   if (iter != sections_.end()) {
     string_buffer_ = iter->second.first;
     string_buffer_length_ = iter->second.second;
@@ -476,7 +480,7 @@ void CompilationUnit::ProcessDIEs() {
   // we need semantics of boost scoped_ptr here - no intention of trasnferring
   // ownership of the stack.  use const, but then we limit ourselves to not
   // ever being able to call .reset() on the smart pointer.
-  auto_ptr<stack<uint64> > const die_stack(new stack<uint64>);
+  std::auto_ptr<stack<uint64> > const die_stack(new stack<uint64>);
 
   while (dieptr < (lengthstart + header_.length)) {
     // We give the user the absolute offset from the beginning of
@@ -811,16 +815,18 @@ void LineInfo::ReadLines() {
     lengthstart += 4;
 
   const char* lineptr = after_header_;
+  lsm.Reset(header_.default_is_stmt);
   while (lineptr < lengthstart + header_.total_length) {
-    lsm.Reset(header_.default_is_stmt);
-    while (!lsm.end_sequence) {
-      size_t oplength;
-      bool add_line = ProcessOneOpcode(reader_, handler_, header_,
-                                       lineptr, &lsm, &oplength, (uintptr_t)-1, NULL);
-      if (add_line)
-        handler_->AddLine(lsm.address, lsm.file_num, lsm.line_num,
-                          lsm.column_num);
-      lineptr += oplength;
+    size_t oplength;
+    bool add_line = ProcessOneOpcode(reader_, handler_, header_,
+                                     lineptr, &lsm, &oplength, (uintptr_t)-1, NULL);
+    lineptr += oplength;
+    if (lsm.end_sequence) {
+      handler_->EndSequence(lsm.address);
+      lsm.Reset(header_.default_is_stmt);      
+    } else if (add_line) {
+      handler_->AddLine(lsm.address, lsm.file_num, lsm.line_num,
+                        lsm.column_num);
     }
   }
 

src/common/dwarf/dwarf2reader.h

@@ -33,20 +33,19 @@
 // reading if you wish to modify the implementation.
 // Only a cursory attempt is made to explain terminology that is
 // used here, as it is much better explained in the standard documents
-#ifndef COMMON_MAC_DWARF_DWARF2READER_H__
-#define COMMON_MAC_DWARF_DWARF2READER_H__
+#ifndef COMMON_DWARF_DWARF2READER_H__
+#define COMMON_DWARF_DWARF2READER_H__
 
-#include <ext/hash_map>
 #include <list>
+#include <map>
 #include <string>
 #include <utility>
 #include <vector>
 
-#include "common/mac/dwarf/dwarf2enums.h"
-#include "common/mac/dwarf/types.h"
+#include "common/dwarf/dwarf2enums.h"
+#include "common/dwarf/types.h"
 
 using namespace std;
-using namespace __gnu_cxx;
 
 namespace dwarf2reader {
 struct LineStateMachine;
@@ -56,7 +55,7 @@ class LineInfoHandler;
 
 // This maps from a string naming a section to a pair containing a
 // the data for the section, and the size of the section.
-typedef hash_map<string, pair<const char*, uint64> > SectionMap;
+typedef map<string, pair<const char*, uint64> > SectionMap;
 typedef list<pair<enum DwarfAttribute, enum DwarfForm> > AttributeList;
 typedef AttributeList::iterator AttributeIterator;
 typedef AttributeList::const_iterator ConstAttributeIterator;
@@ -174,6 +173,16 @@ class LineInfoHandler {
   // starts at, if we know it (0 otherwise).
   virtual void AddLine(uint64 address, uint32 file_num, uint32 line_num,
                        uint32 column_num) { }
+
+  // Called at the end of a sequence of lines.  ADDRESS is the address
+  // of the first byte after the final machine instruction of the
+  // sequence.
+  // 
+  // Note that this is *not* necessarily the end of the line data for
+  // the compilation unit: a single compilation unit's line program
+  // may contain several "end of sequence" markers, to describe (for
+  // example) several discontiguous regions of code.
+  virtual void EndSequence(uint64 address) { }
 };
 
 // The base of DWARF2/3 debug info is a DIE (Debugging Information
@@ -225,8 +234,11 @@ class CompilationUnit {
 
   // Begin reading a Dwarf2 compilation unit, and calling the
   // callbacks in the Dwarf2Handler
-  // Return the offset of the end of the compilation unit - the passed
-  // in offset.
+
+  // Return the full length of the compilation unit, including
+  // headers.  This plus the starting offset passed to the constructor
+  // is the offset of the end of the compilation unit --- the start of
+  // the next compilation unit, if there is one.
   uint64 Start();
 
  private:

src/common/dwarf/functioninfo.cc

@@ -34,19 +34,9 @@
 #include <vector>
 
 
-#include "common/mac/dwarf/functioninfo.h"
+#include "common/dwarf/functioninfo.h"
 
-#include "common/mac/dwarf/bytereader.h"
-
-
-namespace __gnu_cxx 
-{
-  template<> 
-    struct hash<std::string> 
-    {
-      size_t operator()(const std::string& k) const;
-    };
-}
+#include "common/dwarf/bytereader.h"
 
 
 namespace dwarf2reader {
@@ -131,6 +121,17 @@ void CULineInfoHandler::AddLine(uint64 address, uint32 file_num,
   }
 }
 
+void CULineInfoHandler::EndSequence(uint64 address) { 
+  // To preserve this code's behavior prior to the addition of the
+  // EndSequence member function, we duplicate the previous line,
+  // changing its address to ADDRESS.
+  LineMap::iterator it = linemap_->lower_bound(address);
+  if (it != linemap_->begin()) {
+    it--;
+    linemap_->insert(make_pair(address, it->second));
+  }
+}
+
 bool CUFunctionInfoHandler::StartCompilationUnit(uint64 offset,
                                                  uint8 address_size,
                                                  uint8 offset_size,

src/common/dwarf/functioninfo.h

@@ -31,15 +31,15 @@
 // collector that uses the DWARF2/3 reader interface to build a mapping
 // of addresses to files, lines, and functions.
 
-#ifndef COMMON_MAC_DWARF_FUNCTIONINFO_H__
-#define COMMON_MAC_DWARF_FUNCTIONINFO_H__
+#ifndef COMMON_DWARF_FUNCTIONINFO_H__
+#define COMMON_DWARF_FUNCTIONINFO_H__
 
 #include <map>
 #include <string>
 #include <utility>
 #include <vector>
 
-#include "common/mac/dwarf/dwarf2reader.h"
+#include "common/dwarf/dwarf2reader.h"
 
 
 namespace dwarf2reader {
@@ -99,6 +99,16 @@ class CULineInfoHandler: public LineInfoHandler {
   virtual void AddLine(uint64 address, uint32 file_num, uint32 line_num,
                        uint32 column_num);
 
+  // Called at the end of a sequence of lines.  ADDRESS is the address
+  // of the first byte after the final machine instruction of the
+  // sequence.
+  // 
+  // Note that this is *not* necessarily the end of the line data for
+  // the compilation unit: a single compilation unit's line program
+  // may contain several "end of sequence" markers, to describe (for
+  // example) several discontiguous regions of code.
+  virtual void EndSequence(uint64 address);
+
 
  private:
   LineMap* linemap_;
@@ -175,4 +185,4 @@ class CUFunctionInfoHandler: public Dwarf2Handler {
 };
 
 }  // namespace dwarf2reader
-#endif  // COMMON_MAC_DWARF_FUNCTIONINFO_H__
+#endif  // COMMON_DWARF_FUNCTIONINFO_H__

src/common/dwarf/line_state_machine.h

@@ -27,8 +27,8 @@
 // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 
-#ifndef COMMON_MAC_DWARF_LINE_STATE_MACHINE_H__
-#define COMMON_MAC_DWARF_LINE_STATE_MACHINE_H__
+#ifndef COMMON_DWARF_LINE_STATE_MACHINE_H__
+#define COMMON_DWARF_LINE_STATE_MACHINE_H__
 
 namespace dwarf2reader {
 
@@ -58,4 +58,4 @@ struct LineStateMachine {
 }  // namespace dwarf2reader
 
 
-#endif  // COMMON_MAC_DWARF_LINE_STATE_MACHINE_H__
+#endif  // COMMON_DWARF_LINE_STATE_MACHINE_H__

src/common/dwarf/types.h

@@ -30,8 +30,8 @@
 // This file contains some typedefs for basic types
 
 
-#ifndef _COMMON_MAC_DWARF_TYPES_H__
-#define _COMMON_MAC_DWARF_TYPES_H__
+#ifndef _COMMON_DWARF_TYPES_H__
+#define _COMMON_DWARF_TYPES_H__
 
 typedef signed char         int8;
 typedef short               int16;
@@ -43,4 +43,4 @@ typedef unsigned short     uint16;
 typedef unsigned int       uint32;
 typedef unsigned long long uint64;
 
-#endif // _COMMON_MAC_DWARF_TYPES_H__
+#endif // _COMMON_DWARF_TYPES_H__

src/common/linux/dump_symbols.cc

@@ -27,147 +27,50 @@
 // (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
 // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
-#include <a.out.h>
-#include <cstdarg>
-#include <cstdlib>
-#include <cstdio>
+#include <assert.h>
 #include <cxxabi.h>
 #include <elf.h>
 #include <errno.h>
 #include <fcntl.h>
 #include <link.h>
+#include <string.h>
 #include <sys/mman.h>
-#include <stab.h>
 #include <sys/stat.h>
 #include <sys/types.h>
 #include <unistd.h>
-#include <algorithm>
 
-#include <string>
+#include <algorithm>
+#include <cstdarg>
+#include <cstdio>
+#include <cstdlib>
+#include <cstring>
 #include <functional>
 #include <list>
-#include <vector>
 #include <map>
-#include <string.h>
+#include <string>
+#include <vector>
 
 #include "common/linux/dump_symbols.h"
 #include "common/linux/file_id.h"
 #include "common/linux/guid_creator.h"
+#include "common/linux/module.h"
+#include "common/linux/stabs_reader.h"
 #include "processor/scoped_ptr.h"
 
 // This namespace contains helper functions.
 namespace {
 
-struct SourceFileInfo;
-
-// Infomation of a line.
-struct LineInfo {
-  // Offset from start of the function.
-  // Load from stab symbol.
-  ElfW(Off) rva_to_func;
-  // Offset from base of the loading binary.
-  ElfW(Off) rva_to_base;
-  // Size of the line.
-  // It is the difference of the starting address of the line and starting
-  // address of the next N_SLINE, N_FUN or N_SO.
-  uint32_t size;
-  // Line number.
-  uint32_t line_num;
-  // The source file this line belongs to.
-  SourceFileInfo *file;
-};
-
-typedef std::list<struct LineInfo> LineInfoList;
-
-// Information of a function.
-struct FuncInfo {
-  // Name of the function.
-  std::string name;
-  // Offset from the base of the loading address.
-  ElfW(Off) rva_to_base;
-  // Virtual address of the function.
-  // Load from stab symbol.
-  ElfW(Addr) addr;
-  // Size of the function.
-  // It is the difference of the starting address of the function and starting
-  // address of the next N_FUN or N_SO.
-  uint32_t size;
-  // Total size of stack parameters.
-  uint32_t stack_param_size;
-  // Line information array.
-  LineInfoList line_info;
-};
-
-typedef std::list<struct FuncInfo> FuncInfoList;
-
-// Information of a source file.
-struct SourceFileInfo {
-  // Name of the source file.
-  const char *name;
-  // Starting address of the source file.
-  ElfW(Addr) addr;
-  // Id of the source file.
-  int source_id;
-  // Functions information.
-  FuncInfoList func_info;
-};
-
-// A simple std::list of pointers to SourceFileInfo structures, that
-// owns the structures pointed to: destroying the list destroys them,
-// as well.
-class SourceFileInfoList : public std::list<SourceFileInfo *> {
- public:
-  ~SourceFileInfoList() {
-    for (iterator it = this->begin(); it != this->end(); it++)
-      delete *it;
-  }
-};
-
-typedef std::map<const char *, SourceFileInfo *> NameToFileMap;
-
-// Information of a symbol table.
-// This is the root of all types of symbol.
-struct SymbolInfo {
-  // The main files used in this module.  This does not include header
-  // files; it includes only files that were provided as the primary
-  // source file for the compilation unit.  In STABS, these are files
-  // named in 'N_SO' entries.
-  SourceFileInfoList main_files;
-
-  // Map from file names to source file structures.  Note that this
-  // map's keys are compared as pointers, not strings, so if the same
-  // name appears at two different addresses in stabstr, the map will
-  // treat that as two different names.  If the linker didn't unify
-  // names in .stabstr (which it does), this would result in duplicate
-  // FILE lines, which is benign.
-  NameToFileMap name_to_file;
-
-  // An array of some addresses at which a file boundary occurs.
-  //
-  // The STABS information describing a compilation unit gives the
-  // unit's start address, but not its ending address or size.  Those
-  // must be inferred by finding the start address of the next file.
-  // For the last compilation unit, or when one compilation unit ends
-  // before the next one starts, STABS includes an N_SO entry whose
-  // filename is the empty string; such an entry's address serves
-  // simply to mark the end of the preceding compilation unit.  Rather
-  // than create FuncInfoList for such entries, we record their
-  // addresses here.  These are not necessarily sorted.
-  std::vector<ElfW(Addr)> file_boundaries;
-
-  // The current source file, for line number information.  This is
-  // persistent across functions.
-  SourceFileInfo *current_source_file;
-};
+using google_breakpad::Module;
+using std::vector;
 
 // Stab section name.
 static const char *kStabName = ".stab";
 
 // Demangle using abi call.
 // Older GCC may not support it.
-static std::string Demangle(const char *mangled) {
+static std::string Demangle(const std::string &mangled) {
   int status = 0;
-  char *demangled = abi::__cxa_demangle(mangled, NULL, NULL, &status);
+  char *demangled = abi::__cxa_demangle(mangled.c_str(), NULL, NULL, &status);
   if (status == 0 && demangled != NULL) {
     std::string str(demangled);
     free(demangled);
@@ -220,285 +123,227 @@ static const ElfW(Shdr) *FindSectionByName(const char *name,
 
   for (int i = 0; i < nsection; ++i) {
     const char *section_name =
-      (char*)(strtab->sh_offset + sections[i].sh_name);
+      reinterpret_cast<char*>(strtab->sh_offset + sections[i].sh_name);
     if (!strncmp(name, section_name, name_len))
       return sections + i;
   }
   return NULL;
 }
 
-// Return the SourceFileInfo for the file named NAME in SYMBOLS, as
-// recorden in the name_to_file map.  If none exists, create a new
-// one.
-//
-// If the file is a main file, it is the caller's responsibility to
-// set its address and add it to the list of main files.
-//
-// When creating a new file, this function does not make a copy of
-// NAME; NAME must stay alive for as long as the symbol table does.
-static SourceFileInfo *FindSourceFileInfo(SymbolInfo *symbols,
-                                          const char *name) {
-  SourceFileInfo **map_entry = &symbols->name_to_file[name];
-  SourceFileInfo *file;
-  if (*map_entry)
-    file = *map_entry;
-  else {
-    file = new SourceFileInfo;
-    file->name = name;
-    file->source_id = -1;
-    file->addr = 0;
-    *map_entry = file;
-  }
-  return file;
-}
+// Our handler class for STABS data.
+class DumpStabsHandler: public google_breakpad::StabsHandler {
+ public:
+  DumpStabsHandler(Module *module) :
+      module_(module),
+      comp_unit_base_address_(0),
+      current_function_(NULL),
+      current_source_file_(NULL),
+      current_source_file_name_(NULL) { }
 
-static int LoadLineInfo(struct nlist *list,
-                        struct nlist *list_end,
-                        SymbolInfo *symbols,
-                        struct SourceFileInfo *source_file_info,
-                        struct FuncInfo *func_info,
-                        const ElfW(Shdr) *stabstr_section) {
-  struct nlist *cur_list = list;
-  // The name of the file any subsequent lines would belong to.
-  const char *last_source_name = symbols->current_source_file->name;
-  do {
-    // Skip non line information.
-    while (cur_list < list_end && cur_list->n_type != N_SLINE) {
-      // Only exit when got another function, or source file.
-      if (cur_list->n_type == N_FUN || cur_list->n_type == N_SO)
-        return cur_list - list;
-      // N_SOL means source lines following it will be from another
-      // source file.  But don't actually create a file entry yet;
-      // wait until we see executable code attributed to the file.
-      if (cur_list->n_type == N_SOL
-          && cur_list->n_un.n_strx > 0)
-        last_source_name = reinterpret_cast<char *>(cur_list->n_un.n_strx
-                                                 + stabstr_section->sh_offset);
-      ++cur_list;
-    }
-    struct LineInfo line;
-    while (cur_list < list_end && cur_list->n_type == N_SLINE) {
-      // If this line is attributed to a new file, create its entry now.
-      if (last_source_name != symbols->current_source_file->name) {
-        symbols->current_source_file
-          = FindSourceFileInfo(symbols, last_source_name);
-      }
-      line.file = symbols->current_source_file;
-      line.rva_to_func = cur_list->n_value;
-      // n_desc is a signed short
-      line.line_num = (unsigned short)cur_list->n_desc;
-      // We will compute these later.  For now, pacify compiler warnings.
-      line.size = 0;
-      line.rva_to_base = 0;
-      func_info->line_info.push_back(line);
-      ++cur_list;
-    }
-  } while (list < list_end);
+  bool StartCompilationUnit(const char *name, uint64_t address,
+                            const char *build_directory);
+  bool EndCompilationUnit(uint64_t address);
+  bool StartFunction(const std::string &name, uint64_t address);
+  bool EndFunction(uint64_t address);
+  bool Line(uint64_t address, const char *name, int number);
+  void Warning(const char *format, ...);
 
-  return cur_list - list;
-}
+  // Do any final processing necessary to make module_ contain all the
+  // data provided by the STABS reader.
+  //
+  // Because STABS does not provide reliable size information for
+  // functions and lines, we need to make a pass over the data after
+  // processing all the STABS to compute those sizes.  We take care of
+  // that here.
+  void Finalize();
 
-static int LoadFuncSymbols(struct nlist *list,
-                           struct nlist *list_end,
-                           SymbolInfo *symbols,
-                           struct SourceFileInfo *source_file_info,
-                           const ElfW(Shdr) *stabstr_section) {
-  struct nlist *cur_list = list;
-  assert(cur_list->n_type == N_SO);
-  ++cur_list;
-  source_file_info->func_info.clear();
-  while (cur_list < list_end) {
-    // Go until the function symbol.
-    while (cur_list < list_end && cur_list->n_type != N_FUN) {
-      if (cur_list->n_type == N_SO) {
-        return cur_list - list;
-      }
-      ++cur_list;
-      continue;
-    }
-    if (cur_list->n_type == N_FUN) {
-      struct FuncInfo func_info;
-      // The STABS data for an N_FUN entry is the function's (mangled)
-      // name, followed by a colon, followed by type information.  We
-      // want to retain the name only.
-      const char *stabs_name
-        = reinterpret_cast<char *>(cur_list->n_un.n_strx +
-                                   stabstr_section->sh_offset);
-      const char *name_end = strchr(stabs_name, ':');
-      if (! name_end)
-        name_end = stabs_name + strlen(stabs_name);
-      func_info.name = std::string(stabs_name, name_end - stabs_name);
-      func_info.addr = cur_list->n_value;
-      func_info.rva_to_base = 0;
-      func_info.size = 0;
-      func_info.stack_param_size = 0;
-      cur_list++;
+ private:
 
-      // Line info.
-      cur_list += LoadLineInfo(cur_list,
-                               list_end,
-                               symbols,
-                               source_file_info,
-                               &func_info,
-                               stabstr_section);
+  // An arbitrary, but very large, size to use for functions whose
+  // size we can't compute properly.
+  static const uint64_t kFallbackSize = 0x10000000;
 
-      // 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);
-      }
-    }
-  }
-  return cur_list - list;
-}
+  // The module we're contributing debugging info to.
+  Module *module_;
 
-// Compute size and rva information based on symbols loaded from stab section.
-static bool ComputeSizeAndRVA(ElfW(Addr) loading_addr,
-                              struct SymbolInfo *symbols) {
-  SourceFileInfoList::iterator file_it;
-  FuncInfoList::iterator func_it;
-  LineInfoList::iterator line_it;
+  // The functions we've generated so far.  We don't add these to
+  // module_ as we parse them.  Instead, we wait until we've computed
+  // their ending address, and their lines' ending addresses.
+  //
+  // We could just stick them in module_ from the outset, but if
+  // module_ already contains data gathered from other debugging
+  // formats, that would complicate the size computation.
+  vector<Module::Function *> functions_;
 
-  // A table of all the addresses at which files and functions start
-  // or end.  We build this from the file boundary list and our lists
-  // of files and functions, sort it, and then use it to find the ends
-  // of functions and source lines for which we have no size
-  // information.
-  std::vector<ElfW(Addr)> boundaries = symbols->file_boundaries;
-  for (file_it = symbols->main_files.begin();
-       file_it != symbols->main_files.end(); file_it++) {
-    boundaries.push_back((*file_it)->addr);
-    for (func_it = (*file_it)->func_info.begin();
-         func_it != (*file_it)->func_info.end(); func_it++)
-      boundaries.push_back(func_it->addr);
-  }
-  std::sort(boundaries.begin(), boundaries.end());
+  // Boundary addresses.  STABS doesn't necessarily supply sizes for
+  // functions and lines, so we need to compute them ourselves by
+  // finding the next object.
+  vector<Module::Address> boundaries_;
 
-  int no_next_addr_count = 0;
-  for (file_it = symbols->main_files.begin();
-       file_it != symbols->main_files.end(); file_it++) {
-    for (func_it = (*file_it)->func_info.begin();
-         func_it != (*file_it)->func_info.end(); func_it++) {
-      struct FuncInfo &func_info = *func_it;
-      assert(func_info.addr >= loading_addr);
-      func_info.rva_to_base = func_info.addr - loading_addr;
-      func_info.size = 0;
-      std::vector<ElfW(Addr)>::iterator boundary
-        = std::upper_bound(boundaries.begin(), boundaries.end(),
-                           func_info.addr);
-      ElfW(Addr) next_addr = (boundary == boundaries.end()) ? 0 : *boundary;
-      // I've noticed functions with an address bigger than any other functions
-      // and source files modules, this is probably the last function in the
-      // 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;
-      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;
-        }
+  // The base address of the current compilation unit.  We use this to
+  // recognize functions we should omit from the symbol file.  (If you
+  // know the details of why we omit these, please patch this
+  // comment.)
+  Module::Address comp_unit_base_address_;
 
-        no_next_addr_count++;
-        func_info.size = kDefaultSize;
-      }
-      // Compute line size.
-      for (line_it = func_info.line_info.begin(); 
-	   line_it != func_info.line_info.end(); line_it++) {
-        struct LineInfo &line_info = *line_it;
-	LineInfoList::iterator next_line_it = line_it;
-	next_line_it++;
-        line_info.size = 0;
-        if (next_line_it != func_info.line_info.end()) {
-          line_info.size =
-            next_line_it->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;
-      }  // for each line.
-    }  // for each function.
-  } // for each source file.
+  // The function we're currently contributing lines to.
+  Module::Function *current_function_;
+
+  // The last Module::File we got a line number in.
+  Module::File *current_source_file_;
+
+  // The pointer in the .stabstr section of the name that
+  // current_source_file_ is built from.  This allows us to quickly
+  // recognize when the current line is in the same file as the
+  // previous one (which it usually is).
+  const char *current_source_file_name_;
+};
+    
+bool DumpStabsHandler::StartCompilationUnit(const char *name, uint64_t address,
+                                            const char *build_directory) {
+  assert(! comp_unit_base_address_);
+  current_source_file_name_ = name;
+  current_source_file_ = module_->FindFile(name);
+  comp_unit_base_address_ = address;
+  boundaries_.push_back(static_cast<Module::Address>(address));
   return true;
 }
 
+bool DumpStabsHandler::EndCompilationUnit(uint64_t address) {
+  assert(comp_unit_base_address_);
+  comp_unit_base_address_ = 0;
+  current_source_file_ = NULL;
+  current_source_file_name_ = NULL;
+  if (address)
+    boundaries_.push_back(static_cast<Module::Address>(address));
+  return true;
+}
+
+bool DumpStabsHandler::StartFunction(const std::string &name,
+                                     uint64_t address) {
+  assert(! current_function_);
+  Module::Function *f = new Module::Function;
+  f->name_ = Demangle(name);
+  f->address_ = address;
+  f->size_ = 0;           // We compute this in DumpStabsHandler::Finalize().
+  f->parameter_size_ = 0; // We don't provide this information.
+  current_function_ = f;
+  boundaries_.push_back(static_cast<Module::Address>(address));
+  return true;
+}
+
+bool DumpStabsHandler::EndFunction(uint64_t address) {
+  assert(current_function_);
+  // Functions in this compilation unit should have address bigger
+  // than the compilation unit's starting address.  There may be a lot
+  // of duplicated entries for functions in the STABS data; only one
+  // entry can meet this requirement.
+  //
+  // (I don't really understand the above comment; just bringing it
+  // along from the previous code, and leaving the behaivor unchanged.
+  // If you know the whole story, please patch this comment.  --jimb)
+  if (current_function_->address_ >= comp_unit_base_address_)
+    functions_.push_back(current_function_);
+  else
+    delete current_function_;
+  current_function_ = NULL;
+  if (address)
+    boundaries_.push_back(static_cast<Module::Address>(address));
+  return true;
+}
+
+bool DumpStabsHandler::Line(uint64_t address, const char *name, int number) {
+  assert(current_function_);
+  assert(current_source_file_);
+  if (name != current_source_file_name_) {
+    current_source_file_ = module_->FindFile(name);
+    current_source_file_name_ = name;
+  }
+  Module::Line line;
+  line.address_ = address;
+  line.size_ = 0;  // We compute this in DumpStabsHandler::Finalize().
+  line.file_ = current_source_file_;
+  line.number_ = number;
+  current_function_->lines_.push_back(line);
+  return true;
+}
+
+void DumpStabsHandler::Warning(const char *format, ...) {
+  va_list args;
+  va_start(args, format);
+  vfprintf(stderr, format, args);
+  va_end(args);
+}
+
+void DumpStabsHandler::Finalize() {
+  // Sort our boundary list, so we can search it quickly.
+  sort(boundaries_.begin(), boundaries_.end());
+  // Sort all functions by address, just for neatness.
+  sort(functions_.begin(), functions_.end(),
+       Module::Function::CompareByAddress);
+  for (vector<Module::Function *>::iterator func_it = functions_.begin();
+       func_it != functions_.end();
+       func_it++) {
+    Module::Function *f = *func_it;
+    // Compute the function f's size.
+    vector<Module::Address>::iterator boundary
+        = std::upper_bound(boundaries_.begin(), boundaries_.end(), f->address_);
+    if (boundary != boundaries_.end())
+      f->size_ = *boundary - f->address_;
+    else
+      // If this is the last function in the module, and the STABS
+      // reader was unable to give us its ending address, then assign
+      // it a bogus, very large value.  This will happen at most once
+      // per module: since we've added all functions' addresses to the
+      // boundary table, only one can be the last.
+      f->size_ = kFallbackSize;
+
+    // Compute sizes for each of the function f's lines --- if it has any.
+    if (! f->lines_.empty()) {
+      stable_sort(f->lines_.begin(), f->lines_.end(),
+                  Module::Line::CompareByAddress);
+      vector<Module::Line>::iterator last_line = f->lines_.end() - 1;
+      for (vector<Module::Line>::iterator line_it = f->lines_.begin();
+           line_it != last_line; line_it++)
+        line_it[0].size_ = line_it[1].address_ - line_it[0].address_;
+      // Compute the size of the last line from f's end address.
+      last_line->size_ = (f->address_ + f->size_) - last_line->address_;
+    }
+  }
+  // Now that everything has a size, add our functions to the module, and
+  // dispose of our private list.
+  module_->AddFunctions(functions_.begin(), functions_.end());
+  functions_.clear();
+}
+
 static bool LoadSymbols(const ElfW(Shdr) *stab_section,
                         const ElfW(Shdr) *stabstr_section,
-                        ElfW(Addr) loading_addr,
-                        struct SymbolInfo *symbols) {
+                        Module *module) {
   if (stab_section == NULL || stabstr_section == NULL)
     return false;
 
-  struct nlist *lists =
-    reinterpret_cast<struct nlist *>(stab_section->sh_offset);
-  int nstab = stab_section->sh_size / sizeof(struct nlist);
-  // First pass, load all symbols from the object file.
-  for (int i = 0; i < nstab; ) {
-    int step = 1;
-    struct nlist *cur_list = lists + i;
-    if (cur_list->n_type == N_SO) {
-      if (cur_list->n_un.n_strx) {
-        const char *name = reinterpret_cast<char *>(cur_list->n_un.n_strx
-                                                 + stabstr_section->sh_offset);
-        struct SourceFileInfo *source_file_info
-            = FindSourceFileInfo(symbols, name);
-        // Add it to the list; use ADDR to tell whether we've already done so.
-        if (! source_file_info->addr)
-          symbols->main_files.push_back(source_file_info);
-        source_file_info->addr = cur_list->n_value;
-        symbols->current_source_file = source_file_info;
-        step = LoadFuncSymbols(cur_list, lists + nstab, symbols,
-                               source_file_info, stabstr_section);
-      } else {
-        // N_SO entries with no name mark file boundary addresses.
-        symbols->file_boundaries.push_back(cur_list->n_value);
-      }
-    }
-    i += step;
-  }
-
-  // Second pass, compute the size of functions and lines.
-  return ComputeSizeAndRVA(loading_addr, symbols);
+  // A callback object to handle data from the STABS reader.
+  DumpStabsHandler handler(module);
+  // Find the addresses of the STABS data, and create a STABS reader object.
+  uint8_t *stabs = reinterpret_cast<uint8_t *>(stab_section->sh_offset);
+  uint8_t *stabstr = reinterpret_cast<uint8_t *>(stabstr_section->sh_offset);
+  google_breakpad::StabsReader reader(stabs, stab_section->sh_size,
+                                      stabstr, stabstr_section->sh_size,
+                                      &handler);
+  // Read the STABS data, and do post-processing.
+  if (! reader.Process())
+    return false;
+  handler.Finalize();
+  return true;
 }
 
-static bool LoadSymbols(ElfW(Ehdr) *elf_header, struct SymbolInfo *symbols) {
+static bool LoadSymbols(ElfW(Ehdr) *elf_header, Module *module) {
   // Translate all offsets in section headers into address.
   FixAddress(elf_header);
   ElfW(Addr) loading_addr = GetLoadingAddress(
       reinterpret_cast<ElfW(Phdr) *>(elf_header->e_phoff),
       elf_header->e_phnum);
+  module->SetLoadAddress(loading_addr);
 
   const ElfW(Shdr) *sections =
     reinterpret_cast<ElfW(Shdr) *>(elf_header->e_shoff);
@@ -512,153 +357,7 @@ static bool LoadSymbols(ElfW(Ehdr) *elf_header, struct SymbolInfo *symbols) {
   const ElfW(Shdr) *stabstr_section = stab_section->sh_link + sections;
 
   // Load symbols.
-  return LoadSymbols(stab_section, stabstr_section, loading_addr, symbols);
-}
-
-static bool WriteModuleInfo(FILE *file,
-                            ElfW(Half) arch,
-                            const std::string &obj_file) {
-  const char *arch_name = NULL;
-  if (arch == EM_386)
-    arch_name = "x86";
-  else if (arch == EM_X86_64)
-    arch_name = "x86_64";
-  else
-    return false;
-
-  unsigned char identifier[16];
-  google_breakpad::FileID file_id(obj_file.c_str());
-  if (file_id.ElfFileIdentifier(identifier)) {
-    char identifier_str[40];
-    file_id.ConvertIdentifierToString(identifier,
-                                      identifier_str, sizeof(identifier_str));
-    char id_no_dash[40];
-    int id_no_dash_len = 0;
-    memset(id_no_dash, 0, sizeof(id_no_dash));
-    for (int i = 0; identifier_str[i] != '\0'; ++i)
-      if (identifier_str[i] != '-')
-        id_no_dash[id_no_dash_len++] = identifier_str[i];
-    // Add an extra "0" by the end.
-    id_no_dash[id_no_dash_len++] = '0';
-    std::string filename = obj_file;
-    size_t slash_pos = obj_file.find_last_of("/");
-    if (slash_pos != std::string::npos)
-      filename = obj_file.substr(slash_pos + 1);
-    return 0 <= fprintf(file, "MODULE Linux %s %s %s\n", arch_name,
-                        id_no_dash, filename.c_str());
-  }
-  return false;
-}
-
-// Set *INCLUDED_FILES to the list of included files in SYMBOLS,
-// ordered appropriately for output.  Included files should appear in
-// the order in which they are first referenced by source line info.
-// Assign these files source id numbers starting with NEXT_SOURCE_ID.
-//
-// Note that the name_to_file map may contain #included files that are
-// unreferenced; these are the result of LoadFuncSymbols omitting
-// functions from the list whose addresses fall outside the address
-// range of the file that contains them.
-static void CollectIncludedFiles(const struct SymbolInfo &symbols,
-                                 std::vector<SourceFileInfo *> *included_files,
-                                 int next_source_id) {
-  for (SourceFileInfoList::const_iterator file_it = symbols.main_files.begin();
-       file_it != symbols.main_files.end(); file_it++) {
-    for (FuncInfoList::const_iterator func_it = (*file_it)->func_info.begin();
-         func_it != (*file_it)->func_info.end(); func_it++) {
-      for (LineInfoList::const_iterator line_it = func_it->line_info.begin();
-           line_it != func_it->line_info.end(); line_it++) {
-        SourceFileInfo *file = line_it->file;
-        if (file->source_id == -1) {
-          file->source_id = next_source_id++;
-          // Here we use the source id as a mark, ensuring that each
-          // file appears in the list only once.
-          included_files->push_back(file);
-        }
-      }
-    }
-  }
-}
-
-// Write 'FILE' lines for all source files in SYMBOLS to FILE.  We
-// assign source id numbers to files here.
-static bool WriteSourceFileInfo(FILE *file, struct SymbolInfo &symbols) {
-  int next_source_id = 0;
-  // Assign source id numbers to main files, and write them out to the file.
-  for (SourceFileInfoList::iterator file_it = symbols.main_files.begin();
-       file_it != symbols.main_files.end(); file_it++) {
-    SourceFileInfo *file_info = *file_it;
-    assert(file_info->addr);
-    // We only output 'FILE' lines for main files if their names
-    // contain '.'.  The extensionless C++ header files are #included,
-    // not main files, so it wouldn't affect them.  If you know the
-    // story, please patch this comment.
-    if (strchr(file_info->name, '.')) {
-      file_info->source_id = next_source_id++;
-      if (0 > fprintf(file, "FILE %d %s\n",
-                      file_info->source_id, file_info->name))
-        return false;
-    }
-  }
-  // Compute the list of included files, and write them out.
-  // Can't use SourceFileInfoList here, because that owns the files it
-  // points to.
-  std::vector<SourceFileInfo *> included_files;
-  std::vector<SourceFileInfo *>::const_iterator file_it;
-  CollectIncludedFiles(symbols, &included_files, next_source_id);
-  for (file_it = included_files.begin(); file_it != included_files.end();
-       file_it++) {
-    if (0 > fprintf(file, "FILE %d %s\n",
-                    (*file_it)->source_id, (*file_it)->name))
-      return false;
-  }
-  return true;
-}
-
-static bool WriteOneFunction(FILE *file,
-                             const struct FuncInfo &func_info){
-  std::string func_name = Demangle(func_info.name.c_str());
-
-  if (func_info.size <= 0)
-    return true;
-
-  if (0 <= fprintf(file, "FUNC %lx %lx %d %s\n",
-                   (unsigned long) func_info.rva_to_base,
-                   (unsigned long) func_info.size,
-                   func_info.stack_param_size,
-                   func_name.c_str())) {
-    for (LineInfoList::const_iterator it = func_info.line_info.begin();
-	 it != func_info.line_info.end(); it++) {
-      const struct LineInfo &line_info = *it;
-      if (0 > fprintf(file, "%lx %lx %d %d\n",
-                      (unsigned long) line_info.rva_to_base,
-                      (unsigned long) line_info.size,
-                      line_info.line_num,
-                      line_info.file->source_id))
-        return false;
-    }
-    return true;
-  }
-  return false;
-}
-
-static bool WriteFunctionInfo(FILE *file, const struct SymbolInfo &symbols) {
-  for (SourceFileInfoList::const_iterator it = symbols.main_files.begin();
-       it != symbols.main_files.end(); it++) {
-    const struct SourceFileInfo &file_info = **it;
-    for (FuncInfoList::const_iterator fiIt = file_info.func_info.begin(); 
-	 fiIt != file_info.func_info.end(); fiIt++) {
-      const struct FuncInfo &func_info = *fiIt;
-      if (!WriteOneFunction(file, func_info))
-        return false;
-    }
-  }
-  return true;
-}
-
-static bool DumpStabSymbols(FILE *file, struct SymbolInfo &symbols) {
-  return WriteSourceFileInfo(file, symbols) &&
-    WriteFunctionInfo(file, symbols);
+  return LoadSymbols(stab_section, stabstr_section, module);
 }
 
 //
@@ -713,6 +412,48 @@ class MmapWrapper {
    size_t size_;
 };
 
+// Return the breakpad symbol file identifier for the architecture of
+// ELF_HEADER.
+const char *ElfArchitecture(const ElfW(Ehdr) *elf_header) {
+  ElfW(Half) arch = elf_header->e_machine;
+  if (arch == EM_386)
+    return "x86";
+  else if (arch == EM_X86_64)
+    return "x86_64";
+  else
+    return NULL;
+}
+
+// Format the Elf file identifier in IDENTIFIER as a UUID with the
+// dashes removed.
+std::string FormatIdentifier(unsigned char identifier[16]) {
+  char identifier_str[40];
+  google_breakpad::FileID::ConvertIdentifierToString(
+      identifier,
+      identifier_str,
+      sizeof(identifier_str));
+  std::string id_no_dash;
+  for (int i = 0; identifier_str[i] != '\0'; ++i)
+    if (identifier_str[i] != '-')
+      id_no_dash += identifier_str[i];
+  // Add an extra "0" by the end.  PDB files on Windows have an 'age'
+  // number appended to the end of the file identifier; this isn't
+  // really used or necessary on other platforms, but let's preserve
+  // the pattern.
+  id_no_dash += '0';
+  return id_no_dash;
+}
+
+// Return the non-directory portion of FILENAME: the portion after the
+// last slash, or the whole filename if there are no slashes.
+std::string BaseFileName(const std::string &filename) {
+  // Lots of copies!  basename's behavior is less than ideal.
+  char *c_filename = strdup(filename.c_str());
+  std::string base = basename(c_filename);
+  free(c_filename);
+  return base;
+}
+
 }  // namespace
 
 namespace google_breakpad {
@@ -734,16 +475,27 @@ bool DumpSymbols::WriteSymbolFile(const std::string &obj_file,
   ElfW(Ehdr) *elf_header = reinterpret_cast<ElfW(Ehdr) *>(obj_base);
   if (!IsValidElf(elf_header))
     return false;
-  struct SymbolInfo symbols;
 
-  if (!LoadSymbols(elf_header, &symbols))
-     return false;
-  // Write to symbol file.
-  if (WriteModuleInfo(sym_file, elf_header->e_machine, obj_file) &&
-      DumpStabSymbols(sym_file, symbols))
-    return true;
+  unsigned char identifier[16];
+  google_breakpad::FileID file_id(obj_file.c_str());
+  if (! file_id.ElfFileIdentifier(identifier))
+    return false;
 
-  return false;
+  const char *architecture = ElfArchitecture(elf_header);
+  if (! architecture)
+    return false;
+
+  std::string name = BaseFileName(obj_file);
+  std::string os = "Linux";
+  std::string id = FormatIdentifier(identifier);
+
+  Module module(name, os, architecture, id);
+  if (!LoadSymbols(elf_header, &module))
+    return false;
+  if (!module.Write(sym_file))
+    return false;
+
+  return true;
 }
 
 }  // namespace google_breakpad

src/common/linux/file_id.cc

@@ -32,104 +32,74 @@
 // See file_id.h for documentation
 //
 
-#include <cassert>
-#include <cstdio>
+#include "common/linux/file_id.h"
+
+#include <arpa/inet.h>
 #include <elf.h>
 #include <fcntl.h>
 #include <link.h>
-#include <sys/mman.h>
 #include <string.h>
+#include <sys/mman.h>
 #include <unistd.h>
 
-#include "common/linux/file_id.h"
-#include "common/md5.h"
+#include <cassert>
+#include <cstdio>
 
 namespace google_breakpad {
 
-static bool FindElfTextSection(const void *elf_mapped_base,
-                               const void **text_start,
-                               int *text_size) {
-  assert(elf_mapped_base);
-  assert(text_start);
-  assert(text_size);
-
-  const unsigned char *elf_base =
-    static_cast<const unsigned char *>(elf_mapped_base);
-  const ElfW(Ehdr) *elf_header =
-    reinterpret_cast<const ElfW(Ehdr) *>(elf_base);
-  if (memcmp(elf_header, ELFMAG, SELFMAG) != 0)
-    return false;
-  *text_start = NULL;
-  *text_size = 0;
-  const ElfW(Shdr) *sections =
-    reinterpret_cast<const ElfW(Shdr) *>(elf_base + elf_header->e_shoff);
-  const char *text_section_name = ".text";
-  int name_len = strlen(text_section_name);
-  const ElfW(Shdr) *string_section = sections + elf_header->e_shstrndx;
-  const ElfW(Shdr) *text_section = NULL;
-  for (int i = 0; i < elf_header->e_shnum; ++i) {
-    if (sections[i].sh_type == SHT_PROGBITS) {
-      const char *section_name = (char*)(elf_base +
-                                         string_section->sh_offset +
-                                         sections[i].sh_name);
-      if (!strncmp(section_name, text_section_name, name_len)) {
-        text_section = &sections[i];
-        break;
-      }
-    }
-  }
-  if (text_section != NULL && text_section->sh_size > 0) {
-    int text_section_size = text_section->sh_size;
-    *text_start = elf_base + text_section->sh_offset;
-    *text_size = text_section_size;
-  }
-  return true;
-}
-
-FileID::FileID(const char *path) {
+FileID::FileID(const char* path) {
   strncpy(path_, path, sizeof(path_));
 }
 
-bool FileID::ElfFileIdentifier(unsigned char identifier[16]) {
+bool FileID::ElfFileIdentifier(uint8_t identifier[kMDGUIDSize]) {
+  const ssize_t mapped_len = 4096;  // Page size (matches WriteMappings())
   int fd = open(path_, O_RDONLY);
   if (fd < 0)
     return false;
   struct stat st;
-  if (fstat(fd, &st) != 0 && st.st_size <= 0) {
+  if (fstat(fd, &st) != 0 || st.st_size <= mapped_len) {
     close(fd);
     return false;
   }
-  void *base = mmap(NULL, st.st_size,
+  void* base = mmap(NULL, mapped_len,
                     PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
-  if (base == MAP_FAILED) {
-    close(fd);
+  close(fd);
+  if (base == MAP_FAILED)
     return false;
-  }
-  bool success = false;
-  const void *text_section = NULL;
-  int text_size = 0;
-  if (FindElfTextSection(base, &text_section, &text_size) && (text_size > 0)) {
-    struct MD5Context md5;
-    MD5Init(&md5);
-    MD5Update(&md5,
-              static_cast<const unsigned char*>(text_section),
-              text_size);
-    MD5Final(identifier, &md5);
-    success = true;
+
+  memset(identifier, 0, kMDGUIDSize);
+  uint8_t* ptr = reinterpret_cast<uint8_t*>(base);
+  uint8_t* ptr_end = ptr + mapped_len;
+  while (ptr < ptr_end) {
+    for (unsigned i = 0; i < kMDGUIDSize; i++)
+      identifier[i] ^= ptr[i];
+    ptr += kMDGUIDSize;
   }
 
-  close(fd);
-  munmap(base, st.st_size);
-  return success;
+  munmap(base, mapped_len);
+  return true;
 }
 
 // static
-void FileID::ConvertIdentifierToString(const unsigned char identifier[16],
-                                       char *buffer, int buffer_length) {
+void FileID::ConvertIdentifierToString(const uint8_t identifier[kMDGUIDSize],
+                                       char* buffer, int buffer_length) {
+  uint8_t identifier_swapped[kMDGUIDSize];
+
+  // Endian-ness swap to match dump processor expectation.
+  memcpy(identifier_swapped, identifier, kMDGUIDSize);
+  uint32_t* data1 = reinterpret_cast<uint32_t*>(identifier_swapped);
+  *data1 = htonl(*data1);
+  uint16_t* data2 = reinterpret_cast<uint16_t*>(identifier_swapped + 4);
+  *data2 = htons(*data2);
+  uint16_t* data3 = reinterpret_cast<uint16_t*>(identifier_swapped + 6);
+  *data3 = htons(*data3);
+
   int buffer_idx = 0;
-  for (int idx = 0; (buffer_idx < buffer_length) && (idx < 16); ++idx) {
-    int hi = (identifier[idx] >> 4) & 0x0F;
-    int lo = (identifier[idx]) & 0x0F;
+  for (unsigned int idx = 0;
+       (buffer_idx < buffer_length) && (idx < kMDGUIDSize);
+       ++idx) {
+    int hi = (identifier_swapped[idx] >> 4) & 0x0F;
+    int lo = (identifier_swapped[idx]) & 0x0F;
 
     if (idx == 4 || idx == 6 || idx == 8 || idx == 10)
       buffer[buffer_idx++] = '-';

src/common/linux/file_id.h

@@ -35,25 +35,30 @@
 
 #include <limits.h>
 
+#include "common/linux/guid_creator.h"
+
 namespace google_breakpad {
 
+static const size_t kMDGUIDSize = sizeof(MDGUID);
+
 class FileID {
  public:
-  FileID(const char *path);
-  ~FileID() {};
+  explicit FileID(const char* path);
+  ~FileID() {}
 
   // Load the identifier for the elf file path specified in the constructor into
   // |identifier|.  Return false if the identifier could not be created for the
   // file.
-  // The current implementation will return the MD5 hash of the file's bytes.
-  bool ElfFileIdentifier(unsigned char identifier[16]);
+  // The current implementation will XOR the first page of data to generate an
+  // identifier.
+  bool ElfFileIdentifier(uint8_t identifier[kMDGUIDSize]);
 
   // Convert the |identifier| data to a NULL terminated string.  The string will
   // be formatted as a UUID (e.g., 22F065BB-FC9C-49F7-80FE-26A7CEBD7BCE).
   // The |buffer| should be at least 37 bytes long to receive all of the data
   // and termination.  Shorter buffers will contain truncated data.
-  static void ConvertIdentifierToString(const unsigned char identifier[16],
-                                        char *buffer, int buffer_length);
+  static void ConvertIdentifierToString(const uint8_t identifier[kMDGUIDSize],
+                                        char* buffer, int buffer_length);
 
  private:
   // Storage for the path specified
@@ -63,4 +68,3 @@ class FileID {
 }  // namespace google_breakpad
 
 #endif  // COMMON_LINUX_FILE_ID_H__
-

src/common/linux/linux_libc_support.h

@@ -0,0 +1,178 @@
+// Copyright (c) 2009, Google Inc.
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// This header provides replacements for libc functions that we need. We if
+// call the libc functions directly we risk crashing in the dynamic linker as
+// it tries to resolve uncached PLT entries.
+
+#ifndef CLIENT_LINUX_LINUX_LIBC_SUPPORT_H_
+#define CLIENT_LINUX_LINUX_LIBC_SUPPORT_H_
+
+#include <stdint.h>
+#include <limits.h>
+#include <sys/types.h>
+
+extern "C" {
+
+static inline size_t
+my_strlen(const char* s) {
+  size_t len = 0;
+  while (*s++) len++;
+  return len;
+}
+
+static inline int
+my_strcmp(const char* a, const char* b) {
+  for (;;) {
+    if (*a < *b)
+      return -1;
+    else if (*a > *b)
+      return 1;
+    else if (*a == 0)
+      return 0;
+    a++;
+    b++;
+  }
+}
+
+static inline int
+my_strncmp(const char* a, const char* b, size_t len) {
+  for (size_t i = 0; i < len; ++i) {
+    if (*a < *b)
+      return -1;
+    else if (*a > *b)
+      return 1;
+    else if (*a == 0)
+      return 0;
+    a++;
+    b++;
+  }
+
+  return 0;
+}
+
+// Parse a non-negative integer.
+//   result: (output) the resulting non-negative integer
+//   s: a NUL terminated string
+// Return true iff successful.
+static inline bool
+my_strtoui(int* result, const char* s) {
+  if (*s == 0)
+    return false;
+  int r = 0;
+  for (;; s++) {
+    if (*s == 0)
+      break;
+    const int old_r = r;
+    r *= 10;
+    if (*s < '0' || *s > '9')
+      return false;
+    r += *s - '0';
+    if (r < old_r)
+      return false;
+  }
+
+  *result = r;
+  return true;
+}
+
+// Return the length of the given, non-negative integer when expressed in base
+// 10.
+static inline unsigned
+my_int_len(int i) {
+  if (!i)
+    return 1;
+
+  int len = 0;
+  while (i) {
+    len++;
+    i /= 10;
+  }
+
+  return len;
+}
+
+// Convert a non-negative integer to a string
+//   output: (output) the resulting string is written here. This buffer must be
+//     large enough to hold the resulting string. Call |my_int_len| to get the
+//     required length.
+//   i: the non-negative integer to serialise.
+//   i_len: the length of the integer in base 10 (see |my_int_len|).
+static inline void
+my_itos(char* output, int i, unsigned i_len) {
+  for (unsigned index = i_len; index; --index, i /= 10)
+    output[index - 1] = '0' + (i % 10);
+}
+
+static inline const char*
+my_strchr(const char* haystack, char needle) {
+  while (*haystack && *haystack != needle)
+    haystack++;
+  if (*haystack == needle)
+    return haystack;
+  return (const char*) 0;
+}
+
+// Read a hex value
+//   result: (output) the resulting value
+//   s: a string
+// Returns a pointer to the first invalid charactor.
+static inline const char*
+my_read_hex_ptr(uintptr_t* result, const char* s) {
+  uintptr_t r = 0;
+
+  for (;; ++s) {
+    if (*s >= '0' && *s <= '9') {
+      r <<= 4;
+      r += *s - '0';
+    } else if (*s >= 'a' && *s <= 'f') {
+      r <<= 4;
+      r += (*s - 'a') + 10;
+    } else if (*s >= 'A' && *s <= 'F') {
+      r <<= 4;
+      r += (*s - 'A') + 10;
+    } else {
+      break;
+    }
+  }
+
+  *result = r;
+  return s;
+}
+
+static inline void
+my_memset(void* ip, char c, size_t len) {
+  char* p = (char *) ip;
+  while (len--)
+    *p++ = c;
+}
+
+}  // extern "C"
+
+#endif  // CLIENT_LINUX_LINUX_LIBC_SUPPORT_H_

src/common/linux/linux_libc_support_unittest.cc

@@ -0,0 +1,153 @@
+// Copyright (c) 2009, Google Inc.
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+#include "breakpad/linux/linux_libc_support.h"
+#include "testing/gtest/include/gtest/gtest.h"
+
+namespace {
+typedef testing::Test LinuxLibcSupportTest;
+}
+
+TEST(LinuxLibcSupportTest, strlen) {
+  static const char* test_data[] = { "", "a", "aa", "aaa", "aabc", NULL };
+  for (unsigned i = 0; ; ++i) {
+    if (!test_data[i])
+      break;
+    ASSERT_EQ(strlen(test_data[i]), my_strlen(test_data[i]));
+  }
+}
+
+TEST(LinuxLibcSupportTest, strcmp) {
+  static const char* test_data[] = {
+    "", "",
+    "a", "",
+    "", "a",
+    "a", "b",
+    "a", "a",
+    "ab", "aa",
+    "abc", "ab",
+    "abc", "abc",
+    NULL,
+  };
+
+  for (unsigned i = 0; ; ++i) {
+    if (!test_data[i*2])
+      break;
+    ASSERT_EQ(my_strcmp(test_data[i*2], test_data[i*2 + 1]),
+              strcmp(test_data[i*2], test_data[i*2 + 1]));
+  }
+}
+
+TEST(LinuxLibcSupportTest, strtoui) {
+  int result;
+
+  ASSERT_FALSE(my_strtoui(&result, ""));
+  ASSERT_FALSE(my_strtoui(&result, "-1"));
+  ASSERT_FALSE(my_strtoui(&result, "-"));
+  ASSERT_FALSE(my_strtoui(&result, "a"));
+  ASSERT_FALSE(my_strtoui(&result, "23472893472938472987987398472398"));
+
+  ASSERT_TRUE(my_strtoui(&result, "0"));
+  ASSERT_EQ(result, 0);
+  ASSERT_TRUE(my_strtoui(&result, "1"));
+  ASSERT_EQ(result, 1);
+  ASSERT_TRUE(my_strtoui(&result, "12"));
+  ASSERT_EQ(result, 12);
+  ASSERT_TRUE(my_strtoui(&result, "123"));
+  ASSERT_EQ(result, 123);
+  ASSERT_TRUE(my_strtoui(&result, "0123"));
+  ASSERT_EQ(result, 123);
+}
+
+TEST(LinuxLibcSupportTest, int_len) {
+  ASSERT_EQ(my_int_len(0), 1);
+  ASSERT_EQ(my_int_len(2), 1);
+  ASSERT_EQ(my_int_len(5), 1);
+  ASSERT_EQ(my_int_len(9), 1);
+  ASSERT_EQ(my_int_len(10), 2);
+  ASSERT_EQ(my_int_len(99), 2);
+  ASSERT_EQ(my_int_len(100), 3);
+  ASSERT_EQ(my_int_len(101), 3);
+  ASSERT_EQ(my_int_len(1000), 4);
+}
+
+TEST(LinuxLibcSupportTest, itos) {
+  char buf[10];
+
+  my_itos(buf, 0, 1);
+  ASSERT_EQ(0, memcmp(buf, "0", 1));
+
+  my_itos(buf, 1, 1);
+  ASSERT_EQ(0, memcmp(buf, "1", 1));
+
+  my_itos(buf, 10, 2);
+  ASSERT_EQ(0, memcmp(buf, "10", 2));
+
+  my_itos(buf, 63, 2);
+  ASSERT_EQ(0, memcmp(buf, "63", 2));
+
+  my_itos(buf, 101, 3);
+  ASSERT_EQ(0, memcmp(buf, "101", 2));
+}
+
+TEST(LinuxLibcSupportTest, strchr) {
+  ASSERT_EQ(NULL, my_strchr("abc", 'd'));
+  ASSERT_EQ(NULL, my_strchr("", 'd'));
+  ASSERT_EQ(NULL, my_strchr("efghi", 'd'));
+
+  ASSERT_TRUE(my_strchr("a", 'a'));
+  ASSERT_TRUE(my_strchr("abc", 'a'));
+  ASSERT_TRUE(my_strchr("bcda", 'a'));
+  ASSERT_TRUE(my_strchr("sdfasdf", 'a'));
+}
+
+TEST(LinuxLibcSupportTest, read_hex_ptr) {
+  uintptr_t result;
+  const char* last;
+
+  last = my_read_hex_ptr(&result, "");
+  ASSERT_EQ(result, 0);
+  ASSERT_EQ(*last, 0);
+
+  last = my_read_hex_ptr(&result, "0");
+  ASSERT_EQ(result, 0);
+  ASSERT_EQ(*last, 0);
+
+  last = my_read_hex_ptr(&result, "0123");
+  ASSERT_EQ(result, 0x123);
+  ASSERT_EQ(*last, 0);
+
+  last = my_read_hex_ptr(&result, "0123a");
+  ASSERT_EQ(result, 0x123a);
+  ASSERT_EQ(*last, 0);
+
+  last = my_read_hex_ptr(&result, "0123a-");
+  ASSERT_EQ(result, 0x123a);
+  ASSERT_EQ(*last, '-');
+}

src/common/linux/memory.h

@@ -0,0 +1,176 @@
+// Copyright (c) 2009, Google Inc.
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+#ifndef CLIENT_LINUX_HANDLER_MEMORY_H_
+#define CLIENT_LINUX_HANDLER_MEMORY_H_
+
+#include <stdint.h>
+#include <stdlib.h>
+#include <unistd.h>
+#include <sys/mman.h>
+
+#include "common/linux/linux_syscall_support.h"
+
+namespace google_breakpad {
+
+// This is very simple allocator which fetches pages from the kernel directly.
+// Thus, it can be used even when the heap may be corrupted.
+//
+// There is no free operation. The pages are only freed when the object is
+// destroyed.
+class PageAllocator {
+ public:
+  PageAllocator()
+      : page_size_(getpagesize()),
+        last_(NULL),
+        current_page_(NULL),
+        page_offset_(0) {
+  }
+
+  ~PageAllocator() {
+    FreeAll();
+  }
+
+  void *Alloc(unsigned bytes) {
+    if (!bytes)
+      return NULL;
+
+    if (current_page_ && page_size_ - page_offset_ >= bytes) {
+      uint8_t *const ret = current_page_ + page_offset_;
+      page_offset_ += bytes;
+      if (page_offset_ == page_size_) {
+        page_offset_ = 0;
+        current_page_ = NULL;
+      }
+
+      return ret;
+    }
+
+    const unsigned pages =
+        (bytes + sizeof(PageHeader) + page_size_ - 1) / page_size_;
+    uint8_t *const ret = GetNPages(pages);
+    if (!ret)
+      return NULL;
+
+    page_offset_ = (page_size_ - (page_size_ * pages - (bytes + sizeof(PageHeader)))) % page_size_;
+    current_page_ = page_offset_ ? ret + page_size_ * (pages - 1) : NULL;
+
+    return ret + sizeof(PageHeader);
+  }
+
+ private:
+  uint8_t *GetNPages(unsigned num_pages) {
+    void *a = sys_mmap2(NULL, page_size_ * num_pages, PROT_READ | PROT_WRITE,
+                        MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
+    if (a == MAP_FAILED)
+      return NULL;
+
+    struct PageHeader *header = reinterpret_cast<PageHeader*>(a);
+    header->next = last_;
+    header->num_pages = num_pages;
+    last_ = header;
+
+    return reinterpret_cast<uint8_t*>(a);
+  }
+
+  void FreeAll() {
+    PageHeader *next;
+
+    for (PageHeader *cur = last_; cur; cur = next) {
+      next = cur->next;
+      sys_munmap(cur, cur->num_pages * page_size_);
+    }
+  }
+
+  struct PageHeader {
+    PageHeader *next;  // pointer to the start of the next set of pages.
+    unsigned num_pages;  // the number of pages in this set.
+  };
+
+  const unsigned page_size_;
+  PageHeader *last_;
+  uint8_t *current_page_;
+  unsigned page_offset_;
+};
+
+// A wasteful vector is like a normal std::vector, except that it's very much
+// simplier and it allocates memory from a PageAllocator. It's wasteful
+// because, when resizing, it always allocates a whole new array since the
+// PageAllocator doesn't support realloc.
+template<class T>
+class wasteful_vector {
+ public:
+  wasteful_vector(PageAllocator *allocator, unsigned size_hint = 16)
+      : allocator_(allocator),
+        a_((T*) allocator->Alloc(sizeof(T) * size_hint)),
+        allocated_(size_hint),
+        used_(0) {
+  }
+
+  void push_back(const T& new_element) {
+    if (used_ == allocated_)
+      Realloc(allocated_ * 2);
+    a_[used_++] = new_element;
+  }
+
+  size_t size() const {
+    return used_;
+  }
+
+  T& operator[](size_t index) {
+    return a_[index];
+  }
+
+  const T& operator[](size_t index) const {
+    return a_[index];
+  }
+
+ private:
+  void Realloc(unsigned new_size) {
+    T *new_array =
+        reinterpret_cast<T*>(allocator_->Alloc(sizeof(T) * new_size));
+    memcpy(new_array, a_, used_ * sizeof(T));
+    a_ = new_array;
+    allocated_ = new_size;
+  }
+
+  PageAllocator *const allocator_;
+  T *a_;  // pointer to an array of |allocated_| elements.
+  unsigned allocated_;  // size of |a_|, in elements.
+  unsigned used_;  // number of used slots in |a_|.
+};
+
+}  // namespace google_breakpad
+
+inline void* operator new(size_t nbytes,
+                          google_breakpad::PageAllocator& allocator) {
+   return allocator.Alloc(nbytes);
+}
+
+#endif  // CLIENT_LINUX_HANDLER_MEMORY_H_

src/common/linux/memory_unittest.cc

@@ -0,0 +1,84 @@
+// Copyright (c) 2009, Google Inc.
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+#include "breakpad/linux/memory.h"
+#include "testing/gtest/include/gtest/gtest.h"
+
+using namespace google_breakpad;
+
+namespace {
+typedef testing::Test PageAllocatorTest;
+}
+
+TEST(PageAllocatorTest, Setup) {
+  PageAllocator allocator;
+}
+
+TEST(PageAllocatorTest, SmallObjects) {
+  PageAllocator allocator;
+
+  for (unsigned i = 1; i < 1024; ++i) {
+    uint8_t *p = reinterpret_cast<uint8_t*>(allocator.Alloc(i));
+    ASSERT_FALSE(p == NULL);
+    memset(p, 0, i);
+  }
+}
+
+TEST(PageAllocatorTest, LargeObject) {
+  PageAllocator allocator;
+
+  uint8_t *p = reinterpret_cast<uint8_t*>(allocator.Alloc(10000));
+  ASSERT_FALSE(p == NULL);
+  for (unsigned i = 1; i < 10; ++i) {
+    uint8_t *p = reinterpret_cast<uint8_t*>(allocator.Alloc(i));
+    ASSERT_FALSE(p == NULL);
+    memset(p, 0, i);
+  }
+}
+
+namespace {
+typedef testing::Test WastefulVectorTest;
+}
+
+TEST(WastefulVectorTest, Setup) {
+  PageAllocator allocator_;
+  wasteful_vector<int> v(&allocator_);
+  ASSERT_EQ(v.size(), 0u);
+}
+
+TEST(WastefulVectorTest, Simple) {
+  PageAllocator allocator_;
+  wasteful_vector<int> v(&allocator_);
+
+  for (unsigned i = 0; i < 256; ++i)
+    v.push_back(i);
+  ASSERT_EQ(v.size(), 256u);
+  for (unsigned i = 0; i < 256; ++i)
+    ASSERT_EQ(v[i], i);
+}

src/common/linux/module.cc

@@ -0,0 +1,167 @@
+// Copyright (c) 2009, Google Inc.
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+#include <cerrno>
+#include <cstring>
+#include "common/linux/module.h"
+
+namespace google_breakpad {
+
+Module::Module(const string &name, const string &os,
+               const string &architecture, const string &id) :
+    name_(name),
+    os_(os),
+    architecture_(architecture),
+    id_(id),
+    load_address_(0) { }
+
+Module::~Module() {
+  for (FileByNameMap::iterator it = files_.begin(); it != files_.end(); it++)
+    delete it->second;
+  for (vector<Function *>::iterator it = functions_.begin();
+       it != functions_.end(); it++)
+    delete *it;
+}
+
+void Module::SetLoadAddress(Address address) {
+  load_address_ = address;
+}
+
+void Module::AddFunction(Function *function) {
+  functions_.push_back(function);
+}
+
+void Module::AddFunctions(vector<Function *>::iterator begin,
+                          vector<Function *>::iterator end) {
+  functions_.insert(functions_.end(), begin, end);
+}
+
+Module::File *Module::FindFile(const string &name) {
+  // A tricky bit here.  The key of each map entry needs to be a
+  // pointer to the entry's File's name string.  This means that we
+  // can't do the initial lookup with any operation that would create
+  // an empty entry for us if the name isn't found (like, say,
+  // operator[] or insert do), because such a created entry's key will
+  // be a pointer the string passed as our argument.  Since the key of
+  // a map's value type is const, we can't fix it up once we've
+  // created our file.  lower_bound does the lookup without doing an
+  // insertion, and returns a good hint iterator to pass to insert.
+  // Our "destiny" is where we belong, whether we're there or not now.
+  FileByNameMap::iterator destiny = files_.lower_bound(&name);
+  if (destiny == files_.end()
+      || *destiny->first != name) {  // Repeated string comparison, boo hoo.
+    File *file = new File;
+    file->name_ = name;
+    file->source_id_ = -1;
+    destiny = files_.insert(destiny,
+                            FileByNameMap::value_type(&file->name_, file));
+  }
+  return destiny->second;
+}
+
+Module::File *Module::FindFile(const char *name) {
+  string name_string = name;
+  return FindFile(name_string);
+}
+
+void Module::AssignSourceIds() {
+  // First, give every source file an id of -1.
+  for (FileByNameMap::iterator file_it = files_.begin();
+       file_it != files_.end(); file_it++)
+    file_it->second->source_id_ = -1;
+
+  // Next, mark all files actually cited by our functions' line number
+  // info, by setting each one's source id to zero.
+  for (vector<Function *>::const_iterator func_it = functions_.begin();
+       func_it != functions_.end(); func_it++) {
+    Function *func = *func_it;
+    for (vector<Line>::iterator line_it = func->lines_.begin();
+         line_it != func->lines_.end(); line_it++)
+      line_it->file_->source_id_ = 0;
+  }
+
+  // Finally, assign source ids to those files that have been marked.
+  // We could have just assigned source id numbers while traversing
+  // the line numbers, but doing it this way numbers the files in
+  // lexicographical order by name, which is neat.
+  int next_source_id = 0;
+  for (FileByNameMap::iterator file_it = files_.begin();
+       file_it != files_.end(); file_it++)
+    if (! file_it->second->source_id_)
+      file_it->second->source_id_ = next_source_id++;
+}
+
+bool Module::ReportError() {
+  fprintf(stderr, "error writing symbol file: %s\n",
+          strerror (errno));
+  return false;
+}
+
+bool Module::Write(FILE *stream) {
+  if (0 > fprintf(stream, "MODULE %s %s %s %s\n",
+                  os_.c_str(), architecture_.c_str(), id_.c_str(),
+                  name_.c_str()))
+    return ReportError();
+
+  // Write out files.
+  AssignSourceIds();
+  for (FileByNameMap::iterator file_it = files_.begin();
+       file_it != files_.end(); file_it++) {
+    File *file = file_it->second;
+    if (file->source_id_ >= 0) {
+      if (0 > fprintf(stream, "FILE %d %s\n",
+                      file->source_id_, file->name_.c_str()))
+        return ReportError();
+    }
+  }
+
+  // Write out functions and their lines.
+  for (vector<Function *>::const_iterator func_it = functions_.begin();
+       func_it != functions_.end(); func_it++) {
+    Function *func = *func_it;
+    if (0 > fprintf(stream, "FUNC %lx %lx %lu %s\n",
+                    (unsigned long) (func->address_ - load_address_),
+                    (unsigned long) func->size_,
+                    (unsigned long) func->parameter_size_,
+                    func->name_.c_str()))
+      return ReportError();
+    for (vector<Line>::iterator line_it = func->lines_.begin();
+         line_it != func->lines_.end(); line_it++)
+      if (0 > fprintf(stream, "%lx %lx %d %d\n",
+                      (unsigned long) (line_it->address_ - load_address_),
+                      (unsigned long) line_it->size_,
+                      line_it->number_,
+                      line_it->file_->source_id_))
+        return ReportError();
+  }
+
+  return true;
+}
+
+} // namespace google_breakpad

src/common/linux/module.h

@@ -0,0 +1,193 @@
+// Copyright (c) 2009, Google Inc.             -*- mode: c++ -*-
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// module.h: defines google_breakpad::Module, for writing breakpad symbol files
+
+#ifndef COMMON_LINUX_MODULE_H__
+#define COMMON_LINUX_MODULE_H__
+
+#include <map>
+#include <string>
+#include <vector>
+#include <cstdio>
+
+#include "google_breakpad/common/breakpad_types.h"
+
+namespace google_breakpad {
+
+using std::string;
+using std::vector;
+using std::map;
+
+// A Module represents the contents of a module, and supports methods
+// for adding information produced by parsing STABS or DWARF data
+// --- possibly both from the same file --- and then writing out the
+// unified contents as a Breakpad-format symbol file.
+class Module {
+ public:
+  // The type of addresses and sizes in a symbol table.
+  typedef u_int64_t Address;
+  struct File;
+  struct Function;
+  struct Line;
+
+  // Addresses appearing in File, Function, and Line structures are
+  // absolute, not relative to the the module's load address.  That
+  // is, if the module were loaded at its nominal load address, the
+  // addresses would be correct.
+
+  // A source file.
+  struct File {
+    // The name of the source file.
+    string name_;
+
+    // The file's source id.  The Write member function clears this
+    // field and assigns source ids a fresh, so any value placed here
+    // before calling Write will be lost.
+    int source_id_;
+  };
+
+  // A function.
+  struct Function {
+    // For sorting by address.  (Not style-guide compliant, but it's
+    // stupid not to put this in the struct.)
+    static bool CompareByAddress(const Function *x, const Function *y) {
+      return x->address_ < y->address_;
+    }
+
+    // The function's name.
+    string name_;
+    
+    // The start address and length of the function's code.
+    Address address_, size_;
+
+    // The function's parameter size.
+    Address parameter_size_;
+
+    // Source lines belonging to this function, sorted by increasing
+    // address.
+    vector<Line> lines_;
+  };
+
+  // A source line.
+  struct Line {
+    // For sorting by address.  (Not style-guide compliant, but it's
+    // stupid not to put this in the struct.)
+    static bool CompareByAddress(const Module::Line &x, const Module::Line &y) {
+      return x.address_ < y.address_;
+    }
+
+    Address address_, size_;    // The address and size of the line's code.
+    File *file_;                // The source file.
+    int number_;                // The source line number.
+  };
+    
+  // Create a new module with the given name, operating system,
+  // architecture, and ID string.
+  Module(const string &name, const string &os, const string &architecture, 
+         const string &id);
+  ~Module();
+
+  // Set the module's load address to LOAD_ADDRESS; addresses given
+  // for functions and lines will be written to the Breakpad symbol
+  // file as offsets from this address.  Construction initializes this
+  // module's load address to zero: addresses written to the symbol
+  // file will be the same as they appear in the File and Line
+  // structures.
+  void SetLoadAddress(Address load_address);
+
+  // Add FUNCTION to the module.
+  // Destroying this module frees all Function objects that have been
+  // added with this function.
+  void AddFunction(Function *function);
+
+  // Add all the functions in [BEGIN,END) to the module.
+  // Destroying this module frees all Function objects that have been
+  // added with this function.
+  void AddFunctions(vector<Function *>::iterator begin,
+                    vector<Function *>::iterator end);
+
+  // If this module has a file named NAME, return a pointer to it.  If
+  // it has none, then create one and return a pointer to the new
+  // file.  Destroying this module frees all File objects that have
+  // been created using this function, or with Insert.
+  File *FindFile(const string &name);
+  File *FindFile(const char *name);
+
+  // Write this module to STREAM in the breakpad symbol format.
+  // Return true if all goes well, or false if an error occurs.  This
+  // method writes out:
+  // - a header based on the values given to the constructor,
+  // - the source files added via FindFile, and finally
+  // - the functions added via AddFunctions, each with its lines.
+  // Addresses in the output are all relative to the load address
+  // established by SetLoadAddress.
+  bool Write(FILE *stream);
+
+private:
+
+  // Find those files in this module that are actually referred to by
+  // functions' line number data, and assign them source id numbers.
+  // Set the source id numbers for all other files --- unused by the
+  // source line data --- to -1.  We do this before writing out the
+  // symbol file, at which point we omit any unused files.
+  void AssignSourceIds();
+
+  // Report an error that has occurred writing the symbol file, using
+  // errno to find the appropriate cause.  Return false.
+  static bool ReportError();
+
+  // Module header entries.
+  string name_, os_, architecture_, id_;
+
+  // The module's nominal load address.  Addresses for functions and
+  // lines are absolute, assuming the module is loaded at this
+  // address.
+  Address load_address_;
+
+  // Relation for maps whose keys are strings shared with some other
+  // structure.
+  struct CompareStringPtrs {
+    bool operator()(const string *x, const string *y) { return *x < *y; };
+  };
+
+  // A map from filenames to File structures.  The map's keys are
+  // pointers to the Files' names.
+  typedef map<const string *, File *, CompareStringPtrs> FileByNameMap;
+
+  // The module owns all the files and functions that have been added
+  // to it; destroying the module frees the Files and Functions these
+  // point to.
+  FileByNameMap files_;                 // This module's source files.  
+  vector<Function *> functions_;        // This module's functions.
+};
+
+} // namespace google_breakpad
+
+#endif  // COMMON_LINUX_MODULE_H__

src/common/linux/stabs_reader.cc

@@ -0,0 +1,200 @@
+// Copyright 2009 Google Inc. All Rights Reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// This file implements the google_breakpad::StabsReader class.
+
+#include <a.out.h>
+#include <stab.h>
+#include <cstring>
+#include <cassert>
+
+#include "common/linux/stabs_reader.h"
+
+namespace google_breakpad {
+
+StabsReader::StabsReader(const uint8_t *stab,    size_t stab_size,
+                         const uint8_t *stabstr, size_t stabstr_size,
+                         StabsHandler *handler) :
+    stabstr_(stabstr),
+    stabstr_size_(stabstr_size),
+    handler_(handler),
+    symbol_(NULL),
+    current_source_file_(NULL) { 
+  symbols_ = reinterpret_cast<const struct nlist *>(stab);
+  symbols_end_ = symbols_ + (stab_size / sizeof (*symbols_));
+}
+
+const char *StabsReader::SymbolString() {
+  ptrdiff_t offset = symbol_->n_un.n_strx;
+  if (offset < 0 || (size_t) offset >= stabstr_size_) {
+    handler_->Warning("symbol %d: name offset outside the string section",
+                      symbol_ - symbols_);
+    // Return our null string, to keep our promise about all names being
+    // taken from the string section.
+    offset = 0;
+  }
+  return reinterpret_cast<const char *>(stabstr_ + offset);
+}
+
+bool StabsReader::Process() {
+  symbol_ = symbols_;
+  while (symbol_ < symbols_end_) {
+    if (symbol_->n_type == N_SO) {
+      if (! ProcessCompilationUnit())
+        return false;
+    } else
+      symbol_++;
+  }
+  return true;
+}
+
+bool StabsReader::ProcessCompilationUnit() {
+  assert(symbol_ < symbols_end_ && symbol_->n_type == N_SO);
+
+  // There may be an N_SO entry whose name ends with a slash,
+  // indicating the directory in which the compilation occurred.
+  // The build directory defaults to NULL.
+  const char *build_directory = NULL;  
+  {
+    const char *name = SymbolString();
+    if (name[0] && name[strlen(name) - 1] == '/') {
+      build_directory = name;
+      symbol_++;
+    }
+  }
+      
+  // We expect to see an N_SO entry with a filename next, indicating
+  // the start of the compilation unit.
+  {
+    if (symbol_ >= symbols_end_ || symbol_->n_type != N_SO)
+      return true;
+    const char *name = SymbolString();
+    if (name[0] == '\0') {
+      // This seems to be a stray end-of-compilation-unit marker;
+      // consume it, but don't report the end, since we didn't see a
+      // beginning.
+      symbol_++;
+      return true;
+    }
+    current_source_file_ = name;
+  }
+
+  if (! handler_->StartCompilationUnit(current_source_file_,
+                                       SymbolValue(),
+                                       build_directory))
+    return false;
+
+  symbol_++;
+
+  // The STABS documentation says that some compilers may emit
+  // additional N_SO entries with names immediately following the
+  // first, and that they should be ignored.  However, the original
+  // Breakpad STABS reader doesn't ignore them, so we won't either.
+
+  // Process the body of the compilation unit, up to the next N_SO.
+  while (symbol_ < symbols_end_ && symbol_->n_type != N_SO) {
+    if (symbol_->n_type == N_FUN) {
+      if (! ProcessFunction())
+        return false;
+    } else
+      // Ignore anything else.
+      symbol_++;
+  }
+
+  // An N_SO with an empty name indicates the end of the compilation
+  // unit.  Default to zero.
+  uint64_t ending_address = 0;
+  if (symbol_ < symbols_end_) {
+    assert(symbol_->n_type == N_SO);
+    const char *name = SymbolString();
+    if (name[0] == '\0') {
+      ending_address = SymbolValue();
+      symbol_++;
+    }
+  }
+
+  if (! handler_->EndCompilationUnit(ending_address))
+    return false;
+
+  return true;
+}          
+
+bool StabsReader::ProcessFunction() {
+  assert(symbol_ < symbols_end_ && symbol_->n_type == N_FUN);
+
+  uint64_t function_address = SymbolValue();
+  // The STABS string for an N_FUN entry is the name of the function,
+  // followed by a colon, followed by type information for the
+  // function.  We want to pass the name alone to StartFunction.
+  const char *stab_string = SymbolString();
+  const char *name_end = strchr(stab_string, ':');
+  if (! name_end)
+    name_end = stab_string + strlen(stab_string);
+  std::string name(stab_string, name_end - stab_string);
+  if (! handler_->StartFunction(name, function_address))
+    return false;
+  symbol_++;
+
+  while (symbol_ < symbols_end_) {
+    if (symbol_->n_type == N_SO || symbol_->n_type == N_FUN)
+      break;
+    else if (symbol_->n_type == N_SLINE) {
+      // The value of an N_SLINE entry is the offset of the line from
+      // the function's start address.
+      uint64_t line_address = function_address + SymbolValue();
+      // The n_desc of a N_SLINE entry is the line number.  It's a
+      // signed 16-bit field; line numbers from 32768 to 65535 are
+      // stored as n-65536.
+      uint16_t line_number = symbol_->n_desc;
+      if (! handler_->Line(line_address, current_source_file_, line_number))
+        return false;
+      symbol_++;
+    } else if (symbol_->n_type == N_SOL) {
+      current_source_file_ = SymbolString();
+      symbol_++;
+    } else
+      // Ignore anything else.
+      symbol_++;
+  }
+
+  // If there is a subsequent N_SO or N_FUN entry, its address is our
+  // end address.
+  uint64_t ending_address = 0;
+  if (symbol_ < symbols_end_) {
+    assert(symbol_->n_type == N_SO || symbol_->n_type == N_FUN);
+    ending_address = SymbolValue();
+    // Note: we do not increment symbol_ here, since we haven't consumed it.
+  }
+
+  if (! handler_->EndFunction(ending_address))
+    return false;
+
+  return true;
+}
+
+} // namespace google_breakpad

src/common/linux/stabs_reader.h

@@ -0,0 +1,193 @@
+// Copyright 2009 Google Inc. All Rights Reserved.  -*- mode: c++ -*-
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// This file contains definitions related to the STABS reader and
+// its handler interfaces.
+// A description of the STABS debugging format can be found at
+// http://sourceware.org/gdb/current/onlinedocs/stabs_toc.html
+// The comments here assume you understand the format.
+//
+// This reader assumes that the system's <a.out.h> and <stab.h>
+// headers accurately describe the layout of the STABS data; this code
+// is not cross-platform safe.
+
+#ifndef COMMON_LINUX_STABS_READER_H__
+#define COMMON_LINUX_STABS_READER_H__
+
+#include <stdint.h>
+#include <cstddef>
+#include <a.out.h>
+
+#include <string>
+
+namespace google_breakpad {
+
+class StabsHandler;
+
+class StabsReader {
+ public:
+  // Create a reader for the STABS debug information whose .stab
+  // section is the STAB_SIZE bytes at STAB, and whose .stabstr
+  // section is the STABSTR_SIZE bytes at STABSTR.  The reader will
+  // call the member functions of HANDLER to report the information it
+  // finds, when the reader's 'Process' member function is called.
+  // 
+  // Note that, in ELF, the .stabstr section should be found using the
+  // 'sh_link' field of the .stab section header, not by name.
+  StabsReader(const uint8_t *stab,    size_t stab_size,
+              const uint8_t *stabstr, size_t stabstr_size,
+              StabsHandler *handler);
+
+  // Process the STABS data, calling the handler's member functions to
+  // report what we find.  While the handler functions return true,
+  // continue to process until we reach the end of the section.  If we
+  // processed the entire section and all handlers returned true,
+  // return true.  If any handler returned false, return false.
+  bool Process();
+
+ private:
+  // Return the name of the current symbol.
+  const char *SymbolString();
+
+  // Return the value of the current symbol.
+  const uint64_t SymbolValue() {
+    return symbol_->n_value;
+  }
+
+  // Process a compilation unit starting at symbol_.  Return true
+  // to continue processing, or false to abort.
+  bool ProcessCompilationUnit();
+
+  // Process a function in current_source_file_ starting at symbol_.
+  // Return true to continue processing, or false to abort.
+  bool ProcessFunction();
+
+  // The debugging information we're reading.
+  const struct nlist *symbols_, *symbols_end_;
+  const uint8_t *stabstr_;
+  size_t stabstr_size_;
+
+  StabsHandler *handler_;
+
+  // The current symbol we're processing.
+  const struct nlist *symbol_;
+
+  // The current source file name.
+  const char *current_source_file_;
+};
+
+// Consumer-provided callback structure for the STABS reader.  Clients
+// of the STABS reader provide an instance of this structure.  The
+// reader then invokes the member functions of that instance to report
+// the information it finds.
+//
+// The default definitions of the member functions do nothing, and return
+// true so processing will continue.
+class StabsHandler {
+ public:
+  StabsHandler() { }
+  virtual ~StabsHandler() { }
+
+  // Some general notes about the handler callback functions:
+
+  // Processing proceeds until the end of the .stabs section, or until
+  // one of these functions returns false.
+
+  // The addresses given are as reported in the STABS info, without
+  // regard for whether the module may be loaded at different
+  // addresses at different times (a shared library, say).  When
+  // processing STABS from an ELF shared library, the addresses given
+  // all assume the library is loaded at its nominal load address.
+  // They are *not* offsets from the nominal load address.  If you
+  // want offsets, you must subtract off the library's nominal load
+  // address.
+
+  // The arguments to these functions named FILENAME are all
+  // references to strings stored in the .stabstr section.  Because
+  // both the Linux and Solaris linkers factor out duplicate strings
+  // from the .stabstr section, the consumer can assume that if two
+  // FILENAME values are different addresses, they represent different
+  // file names.
+  //
+  // Thus, it's safe to use (say) std::map<char *, ...>, which does
+  // string address comparisons, not string content comparisons.
+  // Since all the strings are in same array of characters --- the
+  // .stabstr section --- comparing their addresses produces
+  // predictable, if not lexicographically meaningful, results.
+
+  // Begin processing a compilation unit whose main source file is
+  // named FILENAME, and whose base address is ADDRESS.  If
+  // BUILD_DIRECTORY is non-NULL, it is the name of the build
+  // directory in which the compilation occurred.
+  virtual bool StartCompilationUnit(const char *filename, uint64_t address,
+                                    const char *build_directory) {
+    return true;
+  }
+
+  // Finish processing the compilation unit.  If ADDRESS is non-zero,
+  // it is the ending address of the compilation unit.  If ADDRESS is
+  // zero, then the compilation unit's ending address is not
+  // available, and the consumer must infer it by other means.
+  virtual bool EndCompilationUnit(uint64_t address) { return true; }
+
+  // Begin processing a function named NAME, whose starting address is
+  // ADDRESS.  This function belongs to the compilation unit that was
+  // most recently started but not ended.
+  //
+  // Note that, unlike filenames, NAME is not a pointer into the
+  // .stabstr section; this is because the name as it appears in the
+  // STABS data is followed by type information.  The value passed to
+  // StartFunction is the function name alone.
+  //
+  // In languages that use name mangling, like C++, NAME is mangled.
+  virtual bool StartFunction(const std::string &name, uint64_t address) {
+    return true;
+  }
+
+  // Finish processing the function.  If ADDRESS is non-zero, it is
+  // the ending address for the function.  If ADDRESS is zero, then
+  // the function's ending address is not available, and the consumer
+  // must infer it by other means.
+  virtual bool EndFunction(uint64_t address) { return true; }
+  
+  // Report that the code at ADDRESS is attributable to line NUMBER of
+  // the source file named FILENAME.  The caller must infer the ending
+  // address of the line.
+  virtual bool Line(uint64_t address, const char *filename, int number) {
+    return true;
+  }
+
+  // Report a warning.  FORMAT is a printf-like format string,
+  // specifying how to format the subsequent arguments.  By default,
+  // print the message to the standard error output.
+  virtual void Warning(const char *format, ...) = 0;
+};
+
+} // namespace google_breakpad
+
+#endif  // COMMON_LINUX_STABS_READER_H__

src/common/mac/dump_syms.h

@@ -33,11 +33,11 @@
 
 #import <Foundation/Foundation.h>
 #include <mach-o/loader.h>
-#include "common/mac/dwarf/dwarf2reader.h"
+#include "common/dwarf/dwarf2reader.h"
 
 // This will map from an architecture string to a SectionMap, which
 // will contain the offsets for all the sections in the dictionary
-typedef hash_map<string, dwarf2reader::SectionMap *> ArchSectionMap;
+typedef map<string, dwarf2reader::SectionMap *> ArchSectionMap;
 
 @interface DumpSymbols : NSObject {
  @protected

src/common/mac/dump_syms.mm

@@ -47,9 +47,9 @@
 #import "dump_syms.h"
 #import "common/mac/file_id.h"
 #import "common/mac/macho_utilities.h"
-#import "common/mac/dwarf/dwarf2reader.h"
-#import "common/mac/dwarf/functioninfo.h"
-#import "common/mac/dwarf/bytereader.h"
+#import "common/dwarf/dwarf2reader.h"
+#import "common/dwarf/functioninfo.h"
+#import "common/dwarf/bytereader.h"
 
 using google_breakpad::FileID;
 
@@ -68,15 +68,6 @@ static NSString *kHeaderCPUTypeKey = @"cpuType";
 // for pruning out extraneous non-function symbols.
 static const int kTextSection = 1;
 
-namespace __gnu_cxx {
-template<> 
-  struct hash<std::string> {
-    size_t operator()(const std::string& k) const {
-      return hash< const char* >()( k.c_str() );
-  }
-};
-}
-
 // Dump FunctionMap to stdout.  Print address, function name, file
 // name, line number, lowpc, and highpc if available.
 void DumpFunctionMap(const dwarf2reader::FunctionMap function_map) {

src/google_breakpad/common/minidump_exception_win32.h

@@ -94,8 +94,11 @@ typedef enum {
       /* EXCEPTION_PRIV_INSTRUCTION */
   MD_EXCEPTION_CODE_WIN_STACK_OVERFLOW           = 0xc00000fd,
       /* EXCEPTION_STACK_OVERFLOW */
-  MD_EXCEPTION_CODE_WIN_POSSIBLE_DEADLOCK        = 0xc0000194
+  MD_EXCEPTION_CODE_WIN_POSSIBLE_DEADLOCK        = 0xc0000194,
       /* EXCEPTION_POSSIBLE_DEADLOCK */
+  MD_EXCEPTION_CODE_WIN_UNHANDLED_CPP_EXCEPTION  = 0xe06d7363
+      /* Per http://support.microsoft.com/kb/185294,
+         generated by Visual C++ compiler */
 } MDExceptionCodeWin;
 
 

src/google_breakpad/processor/basic_source_line_resolver.h

@@ -33,28 +33,14 @@
 #ifndef GOOGLE_BREAKPAD_PROCESSOR_BASIC_SOURCE_LINE_RESOLVER_H__
 #define GOOGLE_BREAKPAD_PROCESSOR_BASIC_SOURCE_LINE_RESOLVER_H__
 
-// TODO: Platforms that have no hash_map can use map, at the likely cost of
-// performance.
-#ifdef __SUNPRO_CC
-#define BSLR_NO_HASH_MAP
-#endif  // __SUNPRO_CC
-
-#ifdef BSLR_NO_HASH_MAP
 #include <map>
-#else  // BSLR_NO_HASH_MAP
-#include <ext/hash_map>
-#endif  // BSLR_NO_HASH_MAP
 
 #include "google_breakpad/processor/source_line_resolver_interface.h"
 
 namespace google_breakpad {
 
 using std::string;
-#ifdef BSLR_NO_HASH_MAP
 using std::map;
-#else  // BSLR_NO_HASH_MAP
-using __gnu_cxx::hash_map;
-#endif  // BSLR_NO_HASH_MAP
 
 class BasicSourceLineResolver : public SourceLineResolverInterface {
  public:
@@ -85,23 +71,13 @@ class BasicSourceLineResolver : public SourceLineResolverInterface {
   struct Function;
   struct PublicSymbol;
   struct File;
-#ifdef BSLR_NO_HASH_MAP
   struct CompareString {
     bool operator()(const string &s1, const string &s2) const;
   };
-#else  // BSLR_NO_HASH_MAP
-  struct HashString {
-    size_t operator()(const string &s) const;
-  };
-#endif  // BSLR_NO_HASH_MAP
   class Module;
 
   // All of the modules we've loaded
-#ifdef BSLR_NO_HASH_MAP
   typedef map<string, Module*, CompareString> ModuleMap;
-#else  // BSLR_NO_HASH_MAP
-  typedef hash_map<string, Module*, HashString> ModuleMap;
-#endif  // BSLR_NO_HASH_MAP
   ModuleMap *modules_;
 
   // Disallow unwanted copy ctor and assignment operator

src/processor/basic_source_line_resolver.cc

@@ -51,9 +51,6 @@
 using std::map;
 using std::vector;
 using std::make_pair;
-#ifndef BSLR_NO_HASH_MAP
-using __gnu_cxx::hash;
-#endif  // BSLR_NO_HASH_MAP
 
 namespace google_breakpad {
 
@@ -125,11 +122,7 @@ class BasicSourceLineResolver::Module {
 
  private:
   friend class BasicSourceLineResolver;
-#ifdef BSLR_NO_HASH_MAP
   typedef map<int, string> FileMap;
-#else  // BSLR_NO_HASH_MAP
-  typedef hash_map<int, string> FileMap;
-#endif  // BSLR_NO_HASH_MAP
 
   // The types for stack_info_.  This is equivalent to MS DIA's
   // StackFrameTypeEnum.  Each identifies a different type of frame
@@ -702,15 +695,9 @@ bool BasicSourceLineResolver::Module::ParseStackInfo(char *stack_info_line) {
   return true;
 }
 
-#ifdef BSLR_NO_HASH_MAP
 bool BasicSourceLineResolver::CompareString::operator()(
     const string &s1, const string &s2) const {
   return strcmp(s1.c_str(), s2.c_str()) < 0;
 }
-#else  // BSLR_NO_HASH_MAP
-size_t BasicSourceLineResolver::HashString::operator()(const string &s) const {
-  return hash<const char*>()(s.c_str());
-}
-#endif  // BSLR_NO_HASH_MAP
 
 }  // namespace google_breakpad

src/processor/minidump_processor.cc

@@ -28,6 +28,7 @@
 // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #include <cassert>
+#include <cstdio>
 
 #include "google_breakpad/processor/minidump_processor.h"
 #include "google_breakpad/processor/call_stack.h"
@@ -753,6 +754,9 @@ string MinidumpProcessor::GetCrashReason(Minidump *dump, u_int64_t *address) {
         case MD_EXCEPTION_CODE_WIN_POSSIBLE_DEADLOCK:
           reason = "EXCEPTION_POSSIBLE_DEADLOCK";
           break;
+        case MD_EXCEPTION_CODE_WIN_UNHANDLED_CPP_EXCEPTION:
+	  reason = "Unhandled C++ Exception";
+	  break;
         default:
           BPLOG(INFO) << "Unknown exception reason " << reason;
           break;

src/processor/pathname_stripper_unittest.cc

@@ -27,6 +27,8 @@
 // (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
 // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
+#include <cstdio>
+
 #include "processor/pathname_stripper.h"
 #include "processor/logging.h"
 

src/processor/postfix_evaluator-inl.h

@@ -38,6 +38,7 @@
 #define PROCESSOR_POSTFIX_EVALUATOR_INL_H__
 
 
+#include <cstdio>
 #include <sstream>
 
 #include "processor/postfix_evaluator.h"

src/tools/linux/dump_syms/Makefile

@@ -16,22 +16,40 @@ BIN=dump_syms
 
 all:$(BIN)
 
-DUMP_OBJ=dump_symbols.o guid_creator.o dump_syms.o file_id.o md5.o
+TOP = ../../../..
+VPATH  = $(TOP)/src/common/linux
+VPATH += $(TOP)/src/common
+VPATH += $(TOP)/src/common/dwarf
+
+DUMP_OBJ =					\
+	dump_symbols.o				\
+	dump_syms.o				\
+	dwarf2diehandler.o			\
+	file_id.o				\
+	guid_creator.o				\
+	md5.o					\
+	module.o				\
+	stabs_reader.o				\
+	$(NULL)
 
 dump_syms:$(DUMP_OBJ)
 	$(CXX) $(CPPFLAGS) $(CXXFLAGS) -o $@ $^
 
-dump_symbols.o:../../../common/linux/dump_symbols.cc
-	$(CXX) $(CPPFLAGS) $(CXXFLAGS) -c $^
+# From this directory:
+dump_syms.o: dump_syms.cc
 
-guid_creator.o:../../../common/linux/guid_creator.cc
-	$(CXX) $(CPPFLAGS) $(CXXFLAGS) -c $^
+# From src/common/linux:
+dump_symbols.o: dump_symbols.cc
+stabs_reader.o: stabs_reader.cc
+module.o: module.cc
+guid_creator.o: guid_creator.cc
+file_id.o: file_id.cc
 
-file_id.o:../../../common/linux/file_id.cc
-	$(CXX) $(CPPFLAGS) $(CXXFLAGS) -c $^
+# From src/common:
+md5.o: md5.c
 
-md5.o:../../../common/md5.c
-	$(CC) $(CPPFLAGS) $(CXXFLAGS) -c $^
+# From src/common/dwarf:
+dwarf2diehandler.o: dwarf2diehandler.cc
 
 clean:
 	rm -f $(BIN) $(DUMP_OBJ)

src/tools/linux/md2core/minidump-2-core.cc

@@ -0,0 +1,602 @@
+// Copyright (c) 2009, Google Inc.
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// Converts a minidump file to a core file which gdb can read.
+// Large parts lifted from the userspace core dumper:
+//   http://code.google.com/p/google-coredumper/
+//
+// Usage: minidump-2-core 1234.dmp > core
+
+#include <vector>
+
+#include <stdio.h>
+#include <string.h>
+
+#include <elf.h>
+#include <errno.h>
+#include <unistd.h>
+#include <fcntl.h>
+#include <sys/user.h>
+#include <sys/mman.h>
+
+#include "google_breakpad/common/minidump_format.h"
+#include "google_breakpad/common/minidump_cpu_x86.h"
+#include "breakpad/linux/minidump_format_linux.h"
+
+#if __WORDSIZE == 64
+  #define ELF_CLASS ELFCLASS64
+  #define Ehdr      Elf64_Ehdr
+  #define Phdr      Elf64_Phdr
+  #define Shdr      Elf64_Shdr
+  #define Nhdr      Elf64_Nhdr
+  #define auxv_t    Elf64_auxv_t
+#else
+  #define ELF_CLASS ELFCLASS32
+  #define Ehdr      Elf32_Ehdr
+  #define Phdr      Elf32_Phdr
+  #define Shdr      Elf32_Shdr
+  #define Nhdr      Elf32_Nhdr
+  #define auxv_t    Elf32_auxv_t
+#endif
+
+
+#if defined(__x86_64__)
+  #define ELF_ARCH  EM_X86_64
+#elif defined(__i386__)
+  #define ELF_ARCH  EM_386
+#elif defined(__ARM_ARCH_3__)
+  #define ELF_ARCH  EM_ARM
+#elif defined(__mips__)
+  #define ELF_ARCH  EM_MIPS
+#endif
+
+static int usage(const char* argv0) {
+  fprintf(stderr, "Usage: %s <minidump file>\n", argv0);
+  return 1;
+}
+
+// Write all of the given buffer, handling short writes and EINTR. Return true
+// iff successful.
+static bool
+writea(int fd, const void* idata, size_t length) {
+  const uint8_t* data = (const uint8_t*) idata;
+
+  size_t done = 0;
+  while (done < length) {
+    ssize_t r;
+    do {
+      r = write(fd, data + done, length - done);
+    } while (r == -1 && errno == EINTR);
+
+    if (r < 1)
+      return false;
+    done += r;
+  }
+
+  return true;
+}
+
+// A range of a mmaped file.
+class MMappedRange {
+ public:
+  MMappedRange(const void* data, size_t length)
+      : data_(reinterpret_cast<const uint8_t*>(data)),
+        length_(length) {
+  }
+
+  // Get an object of |length| bytes at |offset| and return a pointer to it
+  // unless it's out of bounds.
+  const void* GetObject(size_t offset, size_t length) {
+    if (offset + length < offset)
+      return NULL;
+    if (offset + length > length_)
+      return NULL;
+    return data_ + offset;
+  }
+
+  // Get element |index| of an array of objects of length |length| starting at
+  // |offset| bytes. Return NULL if out of bounds.
+  const void* GetArrayElement(size_t offset, size_t length, unsigned index) {
+    const size_t element_offset = offset + index * length;
+    return GetObject(element_offset, length);
+  }
+
+  // Return a new range which is a subset of this range.
+  MMappedRange Subrange(const MDLocationDescriptor& location) const {
+    if (location.rva > length_ ||
+        location.rva + location.data_size < location.rva ||
+        location.rva + location.data_size > length_) {
+      return MMappedRange(NULL, 0);
+    }
+
+    return MMappedRange(data_ + location.rva, location.data_size);
+  }
+
+  const uint8_t* data() const { return data_; }
+  size_t length() const { return length_; }
+
+ private:
+  const uint8_t* const data_;
+  const size_t length_;
+};
+
+/* Dynamically determines the byte sex of the system. Returns non-zero
+ * for big-endian machines.
+ */
+static inline int sex() {
+  int probe = 1;
+  return !*(char *)&probe;
+}
+
+typedef struct elf_timeval {    /* Time value with microsecond resolution    */
+  long tv_sec;                  /* Seconds                                   */
+  long tv_usec;                 /* Microseconds                              */
+} elf_timeval;
+
+typedef struct elf_siginfo {    /* Information about signal (unused)         */
+  int32_t si_signo;             /* Signal number                             */
+  int32_t si_code;              /* Extra code                                */
+  int32_t si_errno;             /* Errno                                     */
+} elf_siginfo;
+
+typedef struct prstatus {       /* Information about thread; includes CPU reg*/
+  elf_siginfo    pr_info;       /* Info associated with signal               */
+  uint16_t       pr_cursig;     /* Current signal                            */
+  unsigned long  pr_sigpend;    /* Set of pending signals                    */
+  unsigned long  pr_sighold;    /* Set of held signals                       */
+  pid_t          pr_pid;        /* Process ID                                */
+  pid_t          pr_ppid;       /* Parent's process ID                       */
+  pid_t          pr_pgrp;       /* Group ID                                  */
+  pid_t          pr_sid;        /* Session ID                                */
+  elf_timeval    pr_utime;      /* User time                                 */
+  elf_timeval    pr_stime;      /* System time                               */
+  elf_timeval    pr_cutime;     /* Cumulative user time                      */
+  elf_timeval    pr_cstime;     /* Cumulative system time                    */
+  user_regs_struct pr_reg;      /* CPU registers                             */
+  uint32_t       pr_fpvalid;    /* True if math co-processor being used      */
+} prstatus;
+
+typedef struct prpsinfo {       /* Information about process                 */
+  unsigned char  pr_state;      /* Numeric process state                     */
+  char           pr_sname;      /* Char for pr_state                         */
+  unsigned char  pr_zomb;       /* Zombie                                    */
+  signed char    pr_nice;       /* Nice val                                  */
+  unsigned long  pr_flag;       /* Flags                                     */
+#if defined(__x86_64__) || defined(__mips__)
+  uint32_t       pr_uid;        /* User ID                                   */
+  uint32_t       pr_gid;        /* Group ID                                  */
+#else
+  uint16_t       pr_uid;        /* User ID                                   */
+  uint16_t       pr_gid;        /* Group ID                                  */
+#endif
+  pid_t          pr_pid;        /* Process ID                                */
+  pid_t          pr_ppid;       /* Parent's process ID                       */
+  pid_t          pr_pgrp;       /* Group ID                                  */
+  pid_t          pr_sid;        /* Session ID                                */
+  char           pr_fname[16];  /* Filename of executable                    */
+  char           pr_psargs[80]; /* Initial part of arg list                  */
+} prpsinfo;
+
+// We parse the minidump file and keep the parsed information in this structure.
+struct CrashedProcess {
+  CrashedProcess()
+      : crashing_tid(-1),
+        auxv(NULL),
+        auxv_length(0) {
+    memset(&prps, 0, sizeof(prps));
+    prps.pr_sname = 'R';
+  }
+
+  struct Mapping {
+    uint64_t start_address, end_address;
+  };
+  std::vector<Mapping> mappings;
+
+  pid_t crashing_tid;
+  int fatal_signal;
+
+  struct Thread {
+    pid_t tid;
+    user_regs_struct regs;
+    user_fpregs_struct fpregs;
+    user_fpxregs_struct fpxregs;
+    uintptr_t stack_addr;
+    const uint8_t* stack;
+    size_t stack_length;
+  };
+  std::vector<Thread> threads;
+
+  const uint8_t* auxv;
+  size_t auxv_length;
+
+  prpsinfo prps;
+};
+
+static uint32_t
+U32(const uint8_t* data) {
+  uint32_t v;
+  memcpy(&v, data, sizeof(v));
+  return v;
+}
+
+static uint16_t
+U16(const uint8_t* data) {
+  uint16_t v;
+  memcpy(&v, data, sizeof(v));
+  return v;
+}
+
+#if defined(__i386__)
+static void
+ParseThreadRegisters(CrashedProcess::Thread* thread, MMappedRange range) {
+  const MDRawContextX86* rawregs =
+      (const MDRawContextX86*) range.GetObject(0, sizeof(MDRawContextX86));
+
+  thread->regs.ebx = rawregs->ebx;
+  thread->regs.ecx = rawregs->ecx;
+  thread->regs.edx = rawregs->edx;
+  thread->regs.esi = rawregs->esi;
+  thread->regs.edi = rawregs->edi;
+  thread->regs.ebp = rawregs->ebp;
+  thread->regs.eax = rawregs->eax;
+  thread->regs.xds = rawregs->ds;
+  thread->regs.xes = rawregs->es;
+  thread->regs.xfs = rawregs->fs;
+  thread->regs.xgs = rawregs->gs;
+  thread->regs.orig_eax = rawregs->eax;
+  thread->regs.eip = rawregs->eip;
+  thread->regs.xcs = rawregs->cs;
+  thread->regs.eflags = rawregs->eflags;
+  thread->regs.esp = rawregs->esp;
+  thread->regs.xss = rawregs->ss;
+
+  thread->fpregs.cwd = rawregs->float_save.control_word;
+  thread->fpregs.swd = rawregs->float_save.status_word;
+  thread->fpregs.twd = rawregs->float_save.tag_word;
+  thread->fpregs.fip = rawregs->float_save.error_offset;
+  thread->fpregs.fcs = rawregs->float_save.error_selector;
+  thread->fpregs.foo = rawregs->float_save.data_offset;
+  thread->fpregs.fos = rawregs->float_save.data_selector;
+  memcpy(thread->fpregs.st_space, rawregs->float_save.register_area,
+         10 * 8);
+
+  thread->fpxregs.cwd = rawregs->float_save.control_word;
+  thread->fpxregs.swd = rawregs->float_save.status_word;
+  thread->fpxregs.twd = rawregs->float_save.tag_word;
+  thread->fpxregs.fop = U16(rawregs->extended_registers + 6);
+  thread->fpxregs.fip = U16(rawregs->extended_registers + 8);
+  thread->fpxregs.fcs = U16(rawregs->extended_registers + 12);
+  thread->fpxregs.foo = U16(rawregs->extended_registers + 16);
+  thread->fpxregs.fos = U16(rawregs->extended_registers + 20);
+  thread->fpxregs.mxcsr = U32(rawregs->extended_registers + 24);
+  memcpy(thread->fpxregs.st_space, rawregs->extended_registers + 32, 128);
+  memcpy(thread->fpxregs.xmm_space, rawregs->extended_registers + 160, 128);
+}
+#else
+#error "This code has not been ported to your platform yet"
+#endif
+
+static void
+ParseThreadList(CrashedProcess* crashinfo, MMappedRange range,
+                const MMappedRange& full_file) {
+  const uint32_t num_threads =
+      *(const uint32_t*) range.GetObject(0, sizeof(uint32_t));
+  for (unsigned i = 0; i < num_threads; ++i) {
+    CrashedProcess::Thread thread;
+    memset(&thread, 0, sizeof(thread));
+    const MDRawThread* rawthread =
+        (MDRawThread*) range.GetArrayElement(sizeof(uint32_t),
+                                             sizeof(MDRawThread), i);
+    thread.tid = rawthread->thread_id;
+    thread.stack_addr = rawthread->stack.start_of_memory_range;
+    MMappedRange stack_range = full_file.Subrange(rawthread->stack.memory);
+    thread.stack = stack_range.data();
+    thread.stack_length = rawthread->stack.memory.data_size;
+
+    ParseThreadRegisters(&thread,
+                         full_file.Subrange(rawthread->thread_context));
+
+    crashinfo->threads.push_back(thread);
+  }
+}
+
+static void
+ParseAuxVector(CrashedProcess* crashinfo, MMappedRange range) {
+  crashinfo->auxv = range.data();
+  crashinfo->auxv_length = range.length();
+}
+
+static void
+ParseCmdLine(CrashedProcess* crashinfo, MMappedRange range) {
+  const char* cmdline = (const char*) range.data();
+  for (size_t i = 0; i < range.length(); ++i) {
+    if (cmdline[i] == 0) {
+      static const size_t fname_len = sizeof(crashinfo->prps.pr_fname) - 1;
+      static const size_t args_len = sizeof(crashinfo->prps.pr_psargs) - 1;
+      memset(crashinfo->prps.pr_fname, 0, fname_len + 1);
+      memset(crashinfo->prps.pr_psargs, 0, args_len + 1);
+      const char* binary_name = strrchr(cmdline, '/');
+      if (binary_name) {
+        binary_name++;
+        const unsigned len = strlen(binary_name);
+        memcpy(crashinfo->prps.pr_fname, binary_name,
+               len > fname_len ? fname_len : len);
+      } else {
+        memcpy(crashinfo->prps.pr_fname, cmdline,
+               i > fname_len ? fname_len : i);
+      }
+
+      const unsigned len = range.length() > args_len ?
+                           args_len : range.length();
+      memcpy(crashinfo->prps.pr_psargs, cmdline, len);
+      for (unsigned i = 0; i < len; ++i) {
+        if (crashinfo->prps.pr_psargs[i] == 0)
+          crashinfo->prps.pr_psargs[i] = ' ';
+      }
+    }
+  }
+}
+
+static void
+ParseExceptionStream(CrashedProcess* crashinfo, MMappedRange range) {
+  const MDRawExceptionStream* exp =
+      (MDRawExceptionStream*) range.GetObject(0, sizeof(MDRawExceptionStream));
+  crashinfo->crashing_tid = exp->thread_id;
+  crashinfo->fatal_signal = (int) exp->exception_record.exception_code;
+}
+
+static bool
+WriteThread(const CrashedProcess::Thread& thread, int fatal_signal) {
+  struct prstatus pr;
+  memset(&pr, 0, sizeof(pr));
+
+  pr.pr_info.si_signo = fatal_signal;
+  pr.pr_cursig = fatal_signal;
+  pr.pr_pid = thread.tid;
+  memcpy(&pr.pr_reg, &thread.regs, sizeof(user_regs_struct));
+
+  Nhdr nhdr;
+  memset(&nhdr, 0, sizeof(nhdr));
+  nhdr.n_namesz = 5;
+  nhdr.n_descsz = sizeof(struct prstatus);
+  nhdr.n_type = NT_PRSTATUS;
+  if (!writea(1, &nhdr, sizeof(nhdr)) ||
+      !writea(1, "CORE\0\0\0\0", 8) ||
+      !writea(1, &pr, sizeof(struct prstatus))) {
+    return false;
+  }
+
+  nhdr.n_descsz = sizeof(user_fpregs_struct);
+  nhdr.n_type = NT_FPREGSET;
+  if (!writea(1, &nhdr, sizeof(nhdr)) ||
+      !writea(1, "CORE\0\0\0\0", 8) ||
+      !writea(1, &thread.fpregs, sizeof(user_fpregs_struct))) {
+    return false;
+  }
+
+  nhdr.n_descsz = sizeof(user_fpxregs_struct);
+  nhdr.n_type = NT_PRXFPREG;
+  if (!writea(1, &nhdr, sizeof(nhdr)) ||
+      !writea(1, "LINUX\0\0\0", 8) ||
+      !writea(1, &thread.fpxregs, sizeof(user_fpxregs_struct))) {
+    return false;
+  }
+
+  return true;
+}
+
+static void
+ParseModuleStream(CrashedProcess* crashinfo, MMappedRange range) {
+  const uint32_t num_mappings =
+      *(const uint32_t*) range.GetObject(0, sizeof(uint32_t));
+  for (unsigned i = 0; i < num_mappings; ++i) {
+    CrashedProcess::Mapping mapping;
+    const MDRawModule* rawmodule =
+        (MDRawModule*) range.GetArrayElement(sizeof(uint32_t),
+                                             sizeof(MDRawModule), i);
+    mapping.start_address = rawmodule->base_of_image;
+    mapping.end_address = rawmodule->size_of_image + rawmodule->base_of_image;
+
+    crashinfo->mappings.push_back(mapping);
+  }
+}
+
+int
+main(int argc, char** argv) {
+  if (argc != 2)
+    return usage(argv[0]);
+
+  const int fd = open(argv[1], O_RDONLY);
+  if (fd < 0)
+    return usage(argv[0]);
+
+  struct stat st;
+  fstat(fd, &st);
+
+  const void* bytes = mmap(NULL, st.st_size, PROT_READ, MAP_SHARED, fd, 0);
+  close(fd);
+  if (bytes == MAP_FAILED) {
+    perror("Failed to mmap dump file");
+    return 1;
+  }
+
+  MMappedRange dump(bytes, st.st_size);
+
+  const MDRawHeader* header =
+      (const MDRawHeader*) dump.GetObject(0, sizeof(MDRawHeader));
+
+  CrashedProcess crashinfo;
+
+  for (unsigned i = 0; i < header->stream_count; ++i) {
+    const MDRawDirectory* dirent =
+        (const MDRawDirectory*) dump.GetArrayElement(
+            header->stream_directory_rva, sizeof(MDRawDirectory), i);
+    switch (dirent->stream_type) {
+      case MD_THREAD_LIST_STREAM:
+        ParseThreadList(&crashinfo, dump.Subrange(dirent->location), dump);
+        break;
+      case MD_LINUX_AUXV:
+        ParseAuxVector(&crashinfo, dump.Subrange(dirent->location));
+        break;
+      case MD_LINUX_CMD_LINE:
+        ParseCmdLine(&crashinfo, dump.Subrange(dirent->location));
+        break;
+      case MD_EXCEPTION_STREAM:
+        ParseExceptionStream(&crashinfo, dump.Subrange(dirent->location));
+        break;
+      case MD_MODULE_LIST_STREAM:
+        ParseModuleStream(&crashinfo, dump.Subrange(dirent->location));
+      default:
+        fprintf(stderr, "Skipping %x\n", dirent->stream_type);
+    }
+  }
+
+  // Write the ELF header. The file will look like:
+  //   ELF header
+  //   Phdr for the PT_NOTE
+  //   Phdr for each of the thread stacks
+  //   PT_NOTE
+  //   each of the thread stacks
+  Ehdr ehdr;
+  memset(&ehdr, 0, sizeof(Ehdr));
+  ehdr.e_ident[0] = ELFMAG0;
+  ehdr.e_ident[1] = ELFMAG1;
+  ehdr.e_ident[2] = ELFMAG2;
+  ehdr.e_ident[3] = ELFMAG3;
+  ehdr.e_ident[4] = ELF_CLASS;
+  ehdr.e_ident[5] = sex() ? ELFDATA2MSB : ELFDATA2LSB;
+  ehdr.e_ident[6] = EV_CURRENT;
+  ehdr.e_type     = ET_CORE;
+  ehdr.e_machine  = ELF_ARCH;
+  ehdr.e_version  = EV_CURRENT;
+  ehdr.e_phoff    = sizeof(Ehdr);
+  ehdr.e_ehsize   = sizeof(Ehdr);
+  ehdr.e_phentsize= sizeof(Phdr);
+  ehdr.e_phnum    = 1 + crashinfo.threads.size() + crashinfo.mappings.size();
+  ehdr.e_shentsize= sizeof(Shdr);
+  if (!writea(1, &ehdr, sizeof(Ehdr)))
+    return 1;
+
+  size_t offset = sizeof(Ehdr) +
+                  (1 + crashinfo.threads.size() +
+                   crashinfo.mappings.size()) * sizeof(Phdr);
+  size_t filesz = sizeof(Nhdr) + 8 + sizeof(prpsinfo) +
+                  // sizeof(Nhdr) + 8 + sizeof(user) +
+                  sizeof(Nhdr) + 8 + crashinfo.auxv_length +
+                  crashinfo.threads.size() * (
+                    (sizeof(Nhdr) + 8 + sizeof(prstatus)) +
+                     sizeof(Nhdr) + 8 + sizeof(user_fpregs_struct) +
+                     sizeof(Nhdr) + 8 + sizeof(user_fpxregs_struct));
+
+  Phdr phdr;
+  memset(&phdr, 0, sizeof(Phdr));
+  phdr.p_type = PT_NOTE;
+  phdr.p_offset = offset;
+  phdr.p_filesz = filesz;
+  if (!writea(1, &phdr, sizeof(phdr)))
+    return 1;
+
+  phdr.p_type = PT_LOAD;
+  phdr.p_align = getpagesize();
+  size_t note_align = phdr.p_align - ((offset+filesz) % phdr.p_align);
+  if (note_align == phdr.p_align)
+    note_align = 0;
+  offset += note_align;
+
+  for (unsigned i = 0; i < crashinfo.threads.size(); ++i) {
+    const CrashedProcess::Thread& thread = crashinfo.threads[i];
+    offset += filesz;
+    filesz = thread.stack_length;
+    phdr.p_offset = offset;
+    phdr.p_vaddr = thread.stack_addr;
+    phdr.p_filesz = phdr.p_memsz = filesz;
+    phdr.p_flags = PF_R | PF_W;
+    if (!writea(1, &phdr, sizeof(phdr)))
+      return 1;
+  }
+
+  for (unsigned i = 0; i < crashinfo.mappings.size(); ++i) {
+    const CrashedProcess::Mapping& mapping = crashinfo.mappings[i];
+    phdr.p_offset = 0;
+    phdr.p_vaddr = mapping.start_address;
+    phdr.p_filesz = 0;
+    phdr.p_flags = PF_R;
+    phdr.p_memsz = mapping.end_address - mapping.start_address;
+    if (!writea(1, &phdr, sizeof(phdr)))
+      return 1;
+  }
+
+  Nhdr nhdr;
+  memset(&nhdr, 0, sizeof(nhdr));
+  nhdr.n_namesz = 5;
+  nhdr.n_descsz = sizeof(prpsinfo);
+  nhdr.n_type = NT_PRPSINFO;
+  if (!writea(1, &nhdr, sizeof(nhdr)) ||
+      !writea(1, "CORE\0\0\0\0", 8) ||
+      !writea(1, &crashinfo.prps, sizeof(prpsinfo))) {
+    return 1;
+  }
+
+  nhdr.n_descsz = crashinfo.auxv_length;
+  nhdr.n_type = NT_AUXV;
+  if (!writea(1, &nhdr, sizeof(nhdr)) ||
+      !writea(1, "CORE\0\0\0\0", 8) ||
+      !writea(1, &crashinfo.auxv, crashinfo.auxv_length)) {
+    return 1;
+  }
+
+  for (unsigned i = 0; i < crashinfo.threads.size(); ++i) {
+    if (crashinfo.threads[i].tid == crashinfo.crashing_tid) {
+      WriteThread(crashinfo.threads[i], crashinfo.fatal_signal);
+      break;
+    }
+  }
+
+  for (unsigned i = 0; i < crashinfo.threads.size(); ++i) {
+    if (crashinfo.threads[i].tid != crashinfo.crashing_tid)
+      WriteThread(crashinfo.threads[i], 0);
+  }
+
+  if (note_align) {
+    char scratch[note_align];
+    memset(scratch, 0, sizeof(scratch));
+    if (!writea(1, scratch, sizeof(scratch)))
+      return 1;
+  }
+
+  for (unsigned i = 0; i < crashinfo.threads.size(); ++i) {
+    const CrashedProcess::Thread& thread = crashinfo.threads[i];
+    if (!writea(1, thread.stack, thread.stack_length))
+      return 1;
+  }
+
+  munmap(const_cast<void*>(bytes), st.st_size);
+
+  return 0;
+}

src/tools/mac/crash_report/crash_report.xcodeproj/project.pbxproj

@@ -130,9 +130,9 @@
 		9BE650AF0B52FE3000611104 /* macho_id.h */ = {isa = PBXFileReference; fileEncoding = 30; lastKnownFileType = sourcecode.c.h; name = macho_id.h; path = ../../../common/mac/macho_id.h; sourceTree = SOURCE_ROOT; };
 		9BE650B00B52FE3000611104 /* macho_walker.cc */ = {isa = PBXFileReference; fileEncoding = 30; lastKnownFileType = sourcecode.cpp.cpp; name = macho_walker.cc; path = ../../../common/mac/macho_walker.cc; sourceTree = SOURCE_ROOT; };
 		9BE650B10B52FE3000611104 /* macho_walker.h */ = {isa = PBXFileReference; fileEncoding = 30; lastKnownFileType = sourcecode.c.h; name = macho_walker.h; path = ../../../common/mac/macho_walker.h; sourceTree = SOURCE_ROOT; };
-		F9C7ECE20E8ABCA600E953AD /* bytereader.cc */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; name = bytereader.cc; path = ../../../common/mac/dwarf/bytereader.cc; sourceTree = SOURCE_ROOT; };
-		F9C7ECE30E8ABCA600E953AD /* dwarf2reader.cc */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; name = dwarf2reader.cc; path = ../../../common/mac/dwarf/dwarf2reader.cc; sourceTree = SOURCE_ROOT; };
-		F9C7ECE40E8ABCA600E953AD /* functioninfo.cc */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; name = functioninfo.cc; path = ../../../common/mac/dwarf/functioninfo.cc; sourceTree = SOURCE_ROOT; };
+		F9C7ECE20E8ABCA600E953AD /* bytereader.cc */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; name = bytereader.cc; path = ../../../common/dwarf/bytereader.cc; sourceTree = SOURCE_ROOT; };
+		F9C7ECE30E8ABCA600E953AD /* dwarf2reader.cc */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; name = dwarf2reader.cc; path = ../../../common/dwarf/dwarf2reader.cc; sourceTree = SOURCE_ROOT; };
+		F9C7ECE40E8ABCA600E953AD /* functioninfo.cc */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; name = functioninfo.cc; path = ../../../common/dwarf/functioninfo.cc; sourceTree = SOURCE_ROOT; };
 		FD6625C40CF4D438004AC844 /* stackwalker_amd64.cc */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; name = stackwalker_amd64.cc; path = ../../../processor/stackwalker_amd64.cc; sourceTree = SOURCE_ROOT; };
 		FD6625C50CF4D438004AC844 /* stackwalker_amd64.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; name = stackwalker_amd64.h; path = ../../../processor/stackwalker_amd64.h; sourceTree = SOURCE_ROOT; };
 		FD8EDEAC0CADDAD400A5EDF1 /* stackwalker_sparc.cc */ = {isa = PBXFileReference; fileEncoding = 30; lastKnownFileType = sourcecode.cpp.cpp; name = stackwalker_sparc.cc; path = ../../../processor/stackwalker_sparc.cc; sourceTree = SOURCE_ROOT; };

src/tools/mac/dump_syms/dump_syms.xcodeproj/project.pbxproj

@@ -45,15 +45,15 @@
 		9BE650440B52F6D800611104 /* macho_id.h */ = {isa = PBXFileReference; fileEncoding = 30; lastKnownFileType = sourcecode.c.h; name = macho_id.h; path = ../../../common/mac/macho_id.h; sourceTree = SOURCE_ROOT; };
 		9BE650450B52F6D800611104 /* macho_walker.cc */ = {isa = PBXFileReference; fileEncoding = 30; lastKnownFileType = sourcecode.cpp.cpp; name = macho_walker.cc; path = ../../../common/mac/macho_walker.cc; sourceTree = SOURCE_ROOT; };
 		9BE650460B52F6D800611104 /* macho_walker.h */ = {isa = PBXFileReference; fileEncoding = 30; lastKnownFileType = sourcecode.c.h; name = macho_walker.h; path = ../../../common/mac/macho_walker.h; sourceTree = SOURCE_ROOT; };
-		F95B422B0E0E22D100DBDE83 /* bytereader-inl.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; name = "bytereader-inl.h"; path = "../../../common/mac/dwarf/bytereader-inl.h"; sourceTree = SOURCE_ROOT; };
-		F95B422C0E0E22D100DBDE83 /* bytereader.cc */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; name = bytereader.cc; path = ../../../common/mac/dwarf/bytereader.cc; sourceTree = SOURCE_ROOT; };
-		F95B422D0E0E22D100DBDE83 /* bytereader.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; name = bytereader.h; path = ../../../common/mac/dwarf/bytereader.h; sourceTree = SOURCE_ROOT; };
-		F95B422E0E0E22D100DBDE83 /* dwarf2enums.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; name = dwarf2enums.h; path = ../../../common/mac/dwarf/dwarf2enums.h; sourceTree = SOURCE_ROOT; };
-		F95B422F0E0E22D100DBDE83 /* dwarf2reader.cc */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; name = dwarf2reader.cc; path = ../../../common/mac/dwarf/dwarf2reader.cc; sourceTree = SOURCE_ROOT; };
-		F95B42300E0E22D100DBDE83 /* dwarf2reader.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; name = dwarf2reader.h; path = ../../../common/mac/dwarf/dwarf2reader.h; sourceTree = SOURCE_ROOT; };
-		F95B42310E0E22D100DBDE83 /* line_state_machine.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; name = line_state_machine.h; path = ../../../common/mac/dwarf/line_state_machine.h; sourceTree = SOURCE_ROOT; };
-		F9C7ED420E8AD93000E953AD /* functioninfo.cc */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; name = functioninfo.cc; path = ../../../common/mac/dwarf/functioninfo.cc; sourceTree = SOURCE_ROOT; };
-		F9F5344D0E7C902C0012363F /* functioninfo.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; name = functioninfo.h; path = ../../../common/mac/dwarf/functioninfo.h; sourceTree = SOURCE_ROOT; };
+		F95B422B0E0E22D100DBDE83 /* bytereader-inl.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; name = "bytereader-inl.h"; path = "../../../common/dwarf/bytereader-inl.h"; sourceTree = SOURCE_ROOT; };
+		F95B422C0E0E22D100DBDE83 /* bytereader.cc */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; name = bytereader.cc; path = ../../../common/dwarf/bytereader.cc; sourceTree = SOURCE_ROOT; };
+		F95B422D0E0E22D100DBDE83 /* bytereader.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; name = bytereader.h; path = ../../../common/dwarf/bytereader.h; sourceTree = SOURCE_ROOT; };
+		F95B422E0E0E22D100DBDE83 /* dwarf2enums.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; name = dwarf2enums.h; path = ../../../common/dwarf/dwarf2enums.h; sourceTree = SOURCE_ROOT; };
+		F95B422F0E0E22D100DBDE83 /* dwarf2reader.cc */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; name = dwarf2reader.cc; path = ../../../common/dwarf/dwarf2reader.cc; sourceTree = SOURCE_ROOT; };
+		F95B42300E0E22D100DBDE83 /* dwarf2reader.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; name = dwarf2reader.h; path = ../../../common/dwarf/dwarf2reader.h; sourceTree = SOURCE_ROOT; };
+		F95B42310E0E22D100DBDE83 /* line_state_machine.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; name = line_state_machine.h; path = ../../../common/dwarf/line_state_machine.h; sourceTree = SOURCE_ROOT; };
+		F9C7ED420E8AD93000E953AD /* functioninfo.cc */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; name = functioninfo.cc; path = ../../../common/dwarf/functioninfo.cc; sourceTree = SOURCE_ROOT; };
+		F9F5344D0E7C902C0012363F /* functioninfo.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; name = functioninfo.h; path = ../../../common/dwarf/functioninfo.h; sourceTree = SOURCE_ROOT; };
 /* End PBXFileReference section */
 
 /* Begin PBXFrameworksBuildPhase section */