Comment 6 for bug 856072

> Here's a mirror (code is forked from netbsd):
> https://github.com/android/platform_bionic/blob/master/libc/netbsd/net/
> getaddrinfo.c
>
> fopen() is called inside _sethtent() and fclose() in _endhtent(). Both are
> called by _files_getaddrinfo() which is responsible for reading /etc/hosts.

Thanks Jim!

Obviously this code can crash on single cpu devices too, it just doesn't happen often. right? It seems worth trying to fix that. I don't really think process affinity is a replacement for a data safety model.

> I don't see a way to skip that

One thing we could do would be to cache the contents of the file in ram in a read-only threadsafe buffer. Update it under lock for the same conditions we call res_ninit() right now.

A slightly uglier approach would be to just put a lock around _files_get_addrinfo. If STDIO is really not threadsafe (independent of file) couldn't either of these still have races against the content process saving a download, for instance? Or is the issue that all the threads are reading /etc/hosts?

Either way means pretty much importing the c file you referenced and calling that version of getaddrinfo on android < honeycomb, but that actually looks relatively easy to do to me. (almost everything in there is static, so just drag it along and make the necessary changes).