Re: [Libguestfs] [nbdkit PATCH 8/8] rate: Atomically set CLOEXEC on fds

On 7/31/19 4:31 PM, Eric Blake wrote: > The rate filter is potentially opening fds in one thread while another > thread is processing a fork() in the plugin. Although the file is not > open for long, we MUST atomically use CLOEXEC to avoid fd leaks. This > one is a bit harder to observe using only the sh plugin, because the > window is small; you'll have better success at catching the leak by > using gdb or recompiling code to insert strategic sleeps. In fact, I have to tweak this commit message: you CAN'T observe this one with the sh plugin unless you recompile it to use #define THREAD_MODEL NBDKIT_THREAD_MODEL_SERIALIZE_REQUESTS, as well as introducing the timing hacks mentioned above (that's because with our current SERIALIZE_ALL_REQUESTS, there is never more than one thread in filter/plugin code at a time).

But it does raise an interesting point - if we hit platforms that are unable to support atomic CLOEXEC, one possibility is a patch that forces SERIALIZE_ALL_REQUESTS as the maximum parallelism allowed on that platform (while remaining at our goal of PARALLEL on more competent systems) - once we do that, the lacking systems will be serialized to the point that there is no race window where one thread can fork() while another is obtaining an fd.