New NaCl zygote implementation 2

base/process_util.h

 // Copyright (c) 2011 The Chromium Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 // This file/namespace contains utility functions for enumerating, ending and
 // computing statistics of processes.
 
 #ifndef BASE_PROCESS_UTIL_H_
 #define BASE_PROCESS_UTIL_H_
 #pragma once
@@ skipping 268 matching lines @@
 // Similar to the above two methods, but starts the child process in a process
 // group of its own, instead of allowing it to inherit the parent's process
 // group. The pgid of the child process will be the same as its pid.
 BASE_API bool LaunchAppInNewProcessGroup(
     const std::vector<std::string>& argv,
     const environment_vector& environ,
     const file_handle_mapping_vector& fds_to_remap,
     bool wait,
     ProcessHandle* process_handle);
 
+#if defined(OS_LINUX)
+// Similar to LaunchApp variants above except uses clone(.. clone_flags ..)
+// rather than fork(). This is useful for work inside the setuid sandbox.
+BASE_API bool LaunchAppWithClone(const std::vector<std::string>& argv,

[Mark Seaborn, 2011/06/15 16:09:04]

Please bear in mind Brett Wilson's post from January on chromium-dev: 'Base is
"done"';
http://groups.google.com/a/chromium.org/group/chromium-dev/browse_thread/thre....

Calling clone()+execve() is simple enough that you probably don't need to add a
new function to 'base' for it.

[Brad Chen, 2011/06/15 18:47:47]

Hmm... I'd have to mimic/copy ~100 lines of code from LaunchAppImpl() if I
didn't create this new variant. That seems much worse than this change.

On 2011/06/15 16:09:04, Mark Seaborn wrote:

[Brad Chen, 2011/06/20 22:09:47]

Today I spent some time trying to use clone/execvp directly. I am getting
mysterious failures in another thread, process_ not defined. After copying a
bunch of code from LaunchAppImpl() it appears to me it is not a simple matter of
clone+execvp, and it would be better to add LaunchAppWithClone() rather than
complicated/duplicate launching functionality in the caller. Of course I will
clear this with an OWNER on base before committing. 

On 2011/06/15 18:47:47, Brad Chen wrote:

+                                 const file_handle_mapping_vector& fds_to_remap,
+                                 bool wait, ProcessHandle* process_handle,
+                                 int clone_flags);
+#endif
+
 // AlterEnvironment returns a modified environment vector, constructed from the
 // given environment and the list of changes given in |changes|. Each key in
 // the environment is matched against the first element of the pairs. In the
 // event of a match, the value is replaced by the second of the pair, unless
 // the second is empty, in which case the key-value is removed.
 //
 // The returned array is allocated using new[] and must be freed by the caller.
 BASE_API char** AlterEnvironment(const environment_vector& changes,
                                  const char* const* const env);
 #endif  // defined(OS_POSIX)
@@ skipping 376 matching lines @@
 // instance running inside the parent.  The parent's Breakpad instance should
 // not handle the child's exceptions.  Calling RestoreDefaultExceptionHandler
 // in the child after forking will restore the standard exception handler.
 // See http://crbug.com/20371/ for more details.
 void RestoreDefaultExceptionHandler();
 #endif  // defined(OS_MACOSX)
 
 }  // namespace base
 
 #endif  // BASE_PROCESS_UTIL_H_