I just recently learned about the signalfd(2) system call, which was introduced to the Linux kernel back in 2007:

signalfd() creates a file descriptor that can be used to accept signals targeted at the caller. This provides an alternative to the use of a signal handler or sigwaitinfo(2), and has the advantage that the file descriptor may be monitored by select(2), poll(2), and epoll(7).

The traditional asynchronous delivery mechanism can be tricky to get right, whereas this provides a convenient fd interface that integrates nicely with your existing event-based code.

I was interested in using signalfd() in some Python code, but Python does not expose this system call through any of the standard libraries. There are a variety of ways one could add support, including:

  • Writing a Python module in C
  • Using the ctypes module (which I played with a few years ago)

However, I decided to use this as an excuse to learn about the cffi module. You can find the complete code in my python-signalfd repository and an explanation of the process below.

The cffi documentation lists a number of principles the project tries to follow; the first two read as follows:

  • The goal is to call C code from Python. You should be able to do so without learning a 3rd language: every alternative requires you to learn their own language (Cython, SWIG) or API (ctypes)…
  • Keep all the Python-related logic in Python so that you don’t need to write much C code (unlike CPython native C extensions).

In practice, what this means is that if the C API documentation for a function looks like this:

unsigned int sleep(unsigned int seconds);

Then you can make this function available in Python like this:

from cffi import FFI
ffi = FFI()
crt = ffi.dlopen(None)
ffi.cdef('unsigned int sleep(unsigned int seconds);')

And to use it:

crt.sleep(10)

While this works great for a simple function like sleep, it gets slightly more complicated when you function prototype looks like this:

   #include <sys/signalfd.h>
   int signalfd(int fd, const sigset_t *mask, int flags);

If you try what seems obvious given the above example:

ffi.cdef('''
       #include <sys/signalfd.h>
       int signalfd(int fd, const sigset_t *mask, int flags);
''')

You’ll run into an error:

cffi.api.CDefError: cannot parse "#include <sys/signalfd.h>"
:3: Directives not supported yet

You can try that without the #include statement, but you’ll just get a new error:

cffi.api.CDefError: cannot parse "int signalfd(int fd, const sigset_t *mask, int flags);"
:3:37: before: *

What all this means is that you need to translate sigset_t into standard C types. You could go digging through include files in /usr/include, but an easier method is to create a small C source file like this:

#include <sys/signalfd.h>

And then run it through the preprocessor:

$ gcc -E sourcefile.c

Inspecting the output of this command reveals that sigset_t is a typedef for __sigset_t, and that __sigset_t looks like this:

typedef struct
  {
    unsigned long int __val[(1024 / (8 * sizeof (unsigned long int)))];
  } __sigset_t;
typedef __sigset_t sigset_t;

If you plug this into your cdef:

ffi.cdef('''
typedef struct
{
  unsigned long int __val[(1024 / (8 * sizeof (unsigned long int)))];
} __sigset_t;
typedef __sigset_t sigset_t;

int signalfd(int fd, const sigset_t *mask, int flags);
''')

You end up with the following:

cffi.api.FFIError: unsupported non-constant or not immediately constant expression

This happens because of the sizeof() expression in the struct. We need to replace that with an actual size. We can use the ffi.sizeof() method to accomplish the same thing, like this:

ffi.cdef('''
typedef struct
{
  unsigned long int __val[%d];
} __sigset_t;
typedef __sigset_t sigset_t;

int signalfd (int fd, const sigset_t * mask, int flags);
''' % ( 1024 / (8 * ffi.sizeof('''unsigned long int''') )))

This will load without error. You can create a variable suitable for passing as the mask parameter to signalfd like this:

>>> mask = ffi.new('sigset_t *')
>>> mask
<cdata 'struct $__sigset_t *' owning 128 bytes>

The trick, of course, is populating that variable correctly. I ended up just implementing all of the sigsetops functions, which, having already set up the sigset_t structure, meant just adding this:

ffi.cdef('''
int sigemptyset(sigset_t *set);
int sigfillset(sigset_t *set);
int sigaddset(sigset_t *set, int signum);
int sigdelset(sigset_t *set, int signum);
int sigismember(const sigset_t *set, int signum);
int sigprocmask(int how, const sigset_t *set, sigset_t *oldset);
''')

Now we’re all set to call these functions through the crt variable we created ealier (by calling ffi.dlopen(None)):

>>> import signal
>>> mask = ffi.new('sigset_t *')
>>> crt.sigemptyset(mask)
0
>>> crt.sigismember(mask, signal.SIGINT)
0
>>> crt.sigaddset(mask, signal.SIGINT)
0
>>> crt.sigismember(mask, signal.SIGINT)
1

And finally, we can all signalfd():

>>> crt.sigprocmask(0, mask, ffi.NULL)
0
>>> fd = crt.signalfd(-1, mask, 0)
>>> from select import poll
>>> p = poll()
>>> p.register(fd)
>>> p.poll()
^C[(3, 1)]
>>> 

In case it’s not obvious from the above example, when I typed CONTROL-C on my keyboard, sending a SIGINT to the Python shell, it caused the p.poll() method to exit, reporting activity on fd 3 (which is the fd we were given by signalfd()). We call sigprocmask(2) to prevent the normal asynchronous delivery of signals, which would otherwise result in Python handling the SIGINT and generating a KeyboardInterrupt exception.

You can find this all packaged up nicely with a slightly more pythonic interface in my python-signalfd repository on GitHub.


Gabe’s Geek Log has an article about signalfd that is also worth reading.