Open race-condition bugs in glibc

06 Mar 2014 19:10:09 GMT

As part of the freeze announcement for musl 1.0, we mentioned longstanding open race condition bugs in glibc. Shortly after the announcement went out on Phoronix, I got a request for details on which bugs we were referring to, so I put together a list.

The most critical (in my opinion) open race bugs are the ones I described on this blog back in 2012; they are reported in glibc issue 12683:

They make it virtually impossible to use thread cancellation safely; at the very least you would have to block cancellation around all cancellation points which allocate or free resources.

Since the original post on EWONTFIX was written, attempts to follow up with the Austin Group seem to have clarified (see issue 614, and issue 529 which was referenced in the response to it) that the requirements on side effects in the event of cancellation are as I interpreted them. A related glibc issue (symptom of the same design problem) is:

A few other race-related issues present in glibc (and which musl's implementation of pthreads avoids) are:

And one which musl shares (musl's POSIX aio is very immature and due for an overhaul):

The above are all bugs which I reported; in addition to these, there are at least a few other race-related bugs open against glibc:

In many ways, the above bugs are what inspired the creation of EWONTFIX, and while I haven't gotten a chance to write a detailed analysis of them all, it's nice to have them at least catalogued. I hope this list can serve as a catalyst for interest in getting them fixed.