On Wed, Jul 12, 2023 at 04:01:59PM -0700, Darrick J. Wong wrote:
> On Tue, Jun 20, 2023 at 02:32:14PM -0700, Omar Sandoval wrote:
> > From: Omar Sandoval <osandov@xxxxxx>
> >
> > xfs_rtallocate_extent_near() calls xfs_rtallocate_extent_block() with
> > the minlen and maxlen that were passed to it.
> > xfs_rtallocate_extent_block() then scans the bitmap block looking for a
> > free range of size maxlen. If there is none, it has to scan the whole
> > bitmap block before returning the largest range of at least size minlen.
> > For a fragmented realtime device and a large allocation request, it's
> > almost certain that this will have to search the whole bitmap block,
> > leading to high CPU usage.
> >
> > However, the realtime summary tells us the maximum size available in the
> > bitmap block. We can limit the search in xfs_rtallocate_extent_block()
> > to that size and often stop before scanning the whole bitmap block.
> >
> > Signed-off-by: Omar Sandoval <osandov@xxxxxx>
> > ---
> > fs/xfs/xfs_rtalloc.c | 8 +++++---
> > 1 file changed, 5 insertions(+), 3 deletions(-)
> >
> > diff --git a/fs/xfs/xfs_rtalloc.c b/fs/xfs/xfs_rtalloc.c
> > index ba7d42e0090f..d079dfb77c73 100644
> > --- a/fs/xfs/xfs_rtalloc.c
> > +++ b/fs/xfs/xfs_rtalloc.c
> > @@ -488,6 +488,8 @@ xfs_rtallocate_extent_near(
> > * allocating one.
> > */
> > if (maxlog >= 0) {
> > + xfs_extlen_t maxavail =
> > + min(maxlen, ((xfs_extlen_t)1 << (maxlog + 1)) - 1);
>
> There can be up to 2^52rtx (realtime extents) in the filesystem, right?
> xfs_extlen_t is a u32, which will overflow this calculation if the
> realtime volume is seriously huge. IOWs, doesn't this need to be:
>
> xfs_extlen_t maxavail = max_t(xfs_rtblock_t, maxlen,
> (1ULL << (maxlog + 1)) - 1);
>
> (The rest of the patch looks ok)
min_t instead of max_t, but good catch, fixed.
