[TriLUG] Fwd: Why do File IO event seem to be out-of-order in strace?

[Scott Lambdin via TriLUG]

Yes, that is my tentative conclusion so far, fcntl locking does not play
well with stdio.h file calls.

http://man7.org/linux/man-pages/man2/fcntl.2.html
       Because of the buffering performed by the stdio(3) library, the use
       of record locking with routines in that package should be avoided;
       use read(2) and write(2) instead.

I'm just trying to analyze the failure (bored at work due to temporary
conditions) and the fix will be done by people above my pay grade.

Now if I can trace the stdio.h system calls through the filesystem driver,
I will be a smart ^@#$%.

Thanks for the new link.  And the clean program.

--Scott



On Tue, Sep 15, 2015 at 1:56 PM, Igor Partola via TriLUG <trilug at trilug.org>
wrote:

> A deadlock is simply a condition where process A is waiting on a resource
> locked by process B and process B is waiting on a resource locked by
> process A. They are trivially avoided by having every code path lock
> resources in the same order.
>
> I really don't think that fork is necessary in this case. After all, you
> want a system level lock to work between two unrelated processes the same
> way as two forks. fnctl() does not care. I cleaned up your example and put
> it here:
>
> https://gist.github.com/ipartola/2b7d6ecc3e6c0fbdb668
>
> Compile it via `gcc -Wall main.c` and run it via `./a.out & ./a.out`. You
> will see that only one process writes to the file at once.
>
> You are correct in that fcntl() can lock a byte range. However, there be
> dragons, and your program correctly sets fl.l_len to 0 which means "lock
> the whole file". Check out http://0pointer.de/blog/projects/locking.html
> for some gotchas with file locking in general.
>
> If you want portability and to write slightly less code, lockf() from
> unistd.h might be a better choice for you:
>
> lockf(fd, F_LOCK, 0); // lock
> lockf(fd, F_ULOCK, 0); // unlock
>
> I am not sure if this helps you in any way with your original problem, but
> this test program should work fine for you.
>
> Igor
>
> On Tue, Sep 15, 2015 at 1:29 PM, Scott Lambdin via TriLUG <
> trilug at trilug.org
> > wrote:
>
> > Hi -
> >
> > Just letting you know that I found out the forks are required for the
> test
> > program.  Each process can only lock one byte range, in order to avoid
> > something called "Dreadlocks"
> >
> > --Scott
> >
> > ---------- Forwarded message ----------
> > From: Igor Partola via TriLUG <trilug at trilug.org>
> > Date: Tue, Sep 8, 2015 at 4:33 PM
> > Subject: Re: [TriLUG] Why do File IO event seem to be out-of-order in
> > strace?
> > To: Triangle Linux Users Group General Discussion <trilug at trilug.org>
> >
> >
> > Looks to me like you are doing fork() incorrectly:
> >
> >  if(childPID >= 0) // fork was successful
> >     {
> >         if(childPID == 0) // child process
> >         {
> >
> > This code after the above statement will run for both parent and child.
> > Since you are doing this in a loop the second process you fork will
> inherit
> > the parent's lock.
> >
> > The correct way to write this would be like so:
> >
> > childPID = fork();
> >
> > if (childPID < 0) {
> >     puts("Could not fork!");
> >     abort(1);
> > }
> >
> > if (childPID == 0) {
> >     // Do the lock test.
> > }
> >
> > if (childPID > 0) {
> >      // Spawn more children or wait on existing children to exit.
> > }
> >
> >
> > For your test, you don't need to fork(). Simply have a program that does
> > the locking with a 10 second sleep before unlocking it. Fire off this
> same
> > program twice and watch the second one take longer than 10 seconds while
> it
> > waits for the lock.
> >
> > Igor
> > --
> > This message was sent to: Scott Lambdin <lopaki at gmail.com>
> > To unsubscribe, send a blank message to trilug-leave at trilug.org from
> that
> > address.
> > TriLUG mailing list : http://www.trilug.org/mailman/listinfo/trilug
> > Unsubscribe or edit options on the web  :
> > http://www.trilug.org/mailman/options/trilug/lopaki%40gmail.com
> > Welcome to TriLUG: http://trilug.org/welcome
> >
> >
> >
> > --
> >
> > Eat like you give a damn.  Go vegan.
> > --
> > This message was sent to: Igor Partola <igor at igorpartola.com>
> > To unsubscribe, send a blank message to trilug-leave at trilug.org from
> that
> > address.
> > TriLUG mailing list : http://www.trilug.org/mailman/listinfo/trilug
> > Unsubscribe or edit options on the web  :
> > http://www.trilug.org/mailman/options/trilug/igor%40igorpartola.com
> > Welcome to TriLUG: http://trilug.org/welcome
> >
> --
> This message was sent to: Scott Lambdin <lopaki at gmail.com>
> To unsubscribe, send a blank message to trilug-leave at trilug.org from that
> address.
> TriLUG mailing list : http://www.trilug.org/mailman/listinfo/trilug
> Unsubscribe or edit options on the web  :
> http://www.trilug.org/mailman/options/trilug/lopaki%40gmail.com
> Welcome to TriLUG: http://trilug.org/welcome
>



-- 

Eat like you give a damn.  Go vegan.