Re: [RFA] Linux Checkpoints, take 3

[Michael Snyder <msnyder at redhat dot com>]

Mark Kettenis wrote:

> > +       /* Now save the 'state' (file position) of all open file descriptors.
> > + 	 Unfortunately fork does not take care of that for us...  */
>
>
> Hmm, fork(2) clones it file descroptors but both file descriptors
> refer to the same open file description.  Nasty.  Wonder whether it'd
> be possible to copy the open file descriptorions too.  Well, I guess
> the point was to do this without kernel support...

This time around, anyway.  I like the successive approximation approach.
So much the better if, on the next pass, we can do even better by using
kernel support.

> Anyway, a few problems that I spotted:
>
> 1. Please drop the "extern" from _initialize_linux_fork() (and the
>    associated  comment).  You'll need an explicit prototype again, like:
>
>    /* Prevent warning from -Wmissing-prototypes.  */
>    void _initialize_linux_fork (void);

I don't understand the motivation, but OK.

> 2. Please update the copyright notice to use (C) and the new FSF
>    address.

Good catch, thanks.

> 3. Replace the FORKS_EXISTS() macro with a forks_exists_p() function?
>    Or inline it if you don't think it's ever going to be changed.

'k.

> 4. Aren't the values used for SEEK_SET/SEEK_CUR the same on all Linux
>    ports?  In that case perhaps #defining LINUX_SEEK_SET 0, and using
>    LINUX_SEEK_SET would be a good solution.
>
> 5. Could you seperate out the "Detaching after fork from child
>    process" printing?

Sorry, don't understand the request.

> And from the nitpicking department:
>
> 1. The comments on the #includes are a bit silly.  I'd rather not see
>    any unless there's something really unobvious going on.  They get
>    out of data pretty soon anyway.

Ooooook, sure...

> 2. Please don't put the function name in comments above the function.
>    (call_lseek, linux_fork_killall, linux_fork_mourn_inferior, 

'k...

> 3. Really sorry to hear that you're suffering from a split personality
>    problem:
>
>
> > 2005-12-19  Michael Snyder  <msnyder@redhat.com>
> >
> > 2005-12-19  Michael Van Meter Snyder  <michsnyd@clwang-lnx.cisco.com>

Well, that's life on the Information Superhighway...

> Otherwise, I think it's ok.  It's a bit sad that this is a Linux-only
> implementation, but then it's mostly Linux-specific code. 

Well, it's intimately tied in with the linux follow-fork code.
It should be possible to do something similar with other systems
where follow-fork works -- but the only other one I'm aware of is
hpux.

> I'll see if
> I can rig up something for OpenBSD, and then we can sort out what bits
> are target-independent.

If you can intercept the fork system call, it should be doable.