On Thu, 2015-10-29 at 00:15 +0000, Al Viro wrote:
> On Wed, Oct 28, 2015 at 04:08:29PM -0700, Eric Dumazet wrote:
> > > > Except for legacy stuff and stdin/stdout/stderr games, I really doubt
> > > > lot of applications absolutely rely on the POSIX thing...
> > >
> > > We obviously can't turn that into default behaviour, though. BTW, what
> > > distribution do you have in mind for those random descriptors? Uniform
> > > on [0,INT_MAX] is a bad idea for obvious reasons - you'll blow the
> > > memory footprint pretty soon...
> >
> > Simply [0 , fdt->max_fds] is working well in most cases.
>
> Umm... So first you dup2() to establish the ->max_fds you want, then
> do such opens?
Yes, dup2() is done at program startup, knowing the expected max load
(in term of concurrent fd) + ~10 % (actual fd array size can be more
than this because of power of two rounding in alloc_fdtable() )
But this is an optimization : If you do not use the initial dup2(), the
fd array can be automatically expanded if needed (all slots are in use)
> What used/unused ratio do you expect to deal with?
> And what kind of locking are you going to use? Keep in mind that
> e.g. dup2() is dependent on the lack of allocations while it's working,
> so it's not as simple as "we don't need no stinkin' ->files_lock"...
No locking change. files->file_lock is still taken.
We only want to minimize time to find an empty slot.
The trick is to not start bitmap search at files->next_fd, but a random
point. This is a win if we assume there are enough holes.
low = start;
if (low < files->next_fd)
low = files->next_fd;
res = -1;
if (flags & O_FD_FASTALLOC) {
random_point = pick_random_between(low, fdt->max_fds);
res = find_next_zero_bit(fdt->open_fds, fdt->max_fds,
random_point);
/* No empty slot found, try the other range */
if (res >= fdt->max_fds) {
res = find_next_zero_bit(fdt->open_fds,
low, random_point);
if (res >= random_point)
res = -1;
}
}
...
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
