once.cpp: call_once: suffers from lack of uniqueness and synchronization handle manipulations

[noloader@…]

call_once attempts to form a unique mutex name by concatenating 2AC1A572DB6944B0A65C38C4140AF2F4 with the PID.

As Wagner and Goldberg taught us, PIDs are a useless source of entropy (or part thereof). See ​http://www.cs.berkeley.edu/~daw/papers/ddj-netscape.html.

An attacker using Win32 can manipulate synchronization objects by (1) squatting the name 'a priori', (2) creating a different type of object using the same name, and (3) creating the expected object but with lesser permissions than the program expects (ie, remove the ability to 'wait' on the mutex). See ​http://www.softwareintegrity.com/Documents/Old-Dogs-and-New-Tricks.pdf.

In addition, two different routines within the same program, each expecting to be 'run once', will suffer contention.

[viboes]

Do you have any suggestion to improve the current state?

[viboes]

How is this related to #5752 boost::call_once() is unreliable on some platforms?

[viboes]

Could you point where in the code this is happening? Moved to support request until clarified.

[viboes]

Closed as it seems this doesn't corresponds to the current Boost.Thread implementation.