Re: [msysGit] Re: [PATCH] compat: Add another rudimentary poll() emulation

From: Albert Dvornik <dvornik@gmail.com>
Date: May 27, 2010, 14:05 UTC
Message-ID: <AANLkTimyFxEYjQLefWvExBarhZAwoXNLDUum2lVj4Q3B@mail.gmail.com>
In-Reply-To: <AANLkTikezlVaX8ARkRw8kEk9wL9RL_5I6X3vK83nzFUl@mail.gmail.com>

On Thu, May 27, 2010 at 9:06 AM, Erik Faye-Lund <kusmabite@googlemail.com> wrote: [...]

Show 6 quoted lines

> But perhaps you should include a check along the lines of this:
>
> if (nfds > FD_SETSIZE)
>        return errno = EINVAL, error("poll: nfds must be below %d", FD_SETSIZE);
>
> Just so we can know when the code fails :)

If you're checking against FD_SETSIZE (which is IMO a good idea), you should consider that (a) on the one system I'm aware of where fd_set doesn't use a bitmap (Windows), FD_SETSIZE is a limit on the number of descriptors added to the set, but (b) on systems where fd_set uses a bitmap (i.e. Linux, perhaps all UNIXes, etc), FD_SETSIZE is a limit on *each descriptor value*. This is also what POSIX says.

So on the latter systems, we want something like this before each FD_SET:

if (ufds[i].fd >= FD_SETSIZE) {
    errno = EINVAL;
    return error("poll: each fd must be below %d", FD_SETSIZE);
}

(The reason to have it in the loop, rather than just check maxfd afterwards, is that FD_SET with an argument that's too big can trash the stack.)

Of course, on Windows this would impose a limitation that all descriptors be < 64, which is probably crazy. Which means that you'd have to actually distinguish the two types. Sigh.

--bert