[PATCH v5 13/14] locks: skip deadlock detection on FL_FILE_PVT locks

[J. Bruce Fields]

On Tue, Jan 14, 2014 at 04:21:53PM -0500, Richard Hipp wrote:
> SQLite uses only F_SETLK, never F_SETLKW.  Doesn't that mean that SQLite
> will work the same with or without deadlock detection?  Doesn't deadlock
> detection only come into play with F_SETLKW?

That's correct.

> > >> (Actually, what happens if you receive a signal which waiting on a file
> > lock?)
> > >
> > > Return -EINTR.
> >
> 
> Huh.  SQLite is not checking for EINTR if fcntl(F_SETLK,...) fails.  Should
> it be?  Or does EINTR only come up for F_SETLKW?

I don't know--I wonder if a distributed filesystem, for example, might
allow even a non-blocking lock request to be interrupted?  Might be
interesting to check what nfs does.

--b.

> 
> We do check for EINTR and retry for other system calls (read(), write(),
> fallocate(), ftruncate(), close(), chmod(), open(), maybe others too).