linux-kernel - Re: [PATCH] btrfs: remove extent buffer's redundant `len` member field

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20250502103014.GN9140@suse.cz>
Date: Fri, 2 May 2025 12:30:14 +0200
From: David Sterba <dsterba@...e.cz>
To: Daniel Vacek <neelx@...e.com>
Cc: Chris Mason <clm@...com>, Josef Bacik <josef@...icpanda.com>,
	David Sterba <dsterba@...e.com>, linux-btrfs@...r.kernel.org,
	linux-kernel@...r.kernel.org
Subject: Re: [PATCH] btrfs: remove extent buffer's redundant `len` member
 field

On Wed, Apr 30, 2025 at 02:31:33PM +0200, Daniel Vacek wrote:
> On Wed, 30 Apr 2025 at 10:21, Daniel Vacek <neelx@...e.com> wrote:
> >
> > On Wed, 30 Apr 2025 at 10:03, David Sterba <dsterba@...e.cz> wrote:
> > >
> > > On Tue, Apr 29, 2025 at 05:17:57PM +0200, Daniel Vacek wrote:
> > > > Even super block nowadays uses nodesize for eb->len. This is since commits
> > > >
> > > > 551561c34663 ("btrfs: don't pass nodesize to __alloc_extent_buffer()")
> > > > da17066c4047 ("btrfs: pull node/sector/stripe sizes out of root and into fs_info")
> > > > ce3e69847e3e ("btrfs: sink parameter len to alloc_extent_buffer")
> > > > a83fffb75d09 ("btrfs: sink blocksize parameter to btrfs_find_create_tree_block")
> > > >
> > > > With these the eb->len is not really useful anymore. Let's use the nodesize
> > > > directly where applicable.
> > >
> > > I've had this patch in my local branch for some years from the times we
> > > were optimizing extent buffer size. The size on release config is 240
> > > bytes. The goal was to get it under 256 and keep it aligned.
> > >
> > > Removing eb->len does not change the structure size and leaves a hole
> > >
> > >  struct extent_buffer {
> > >         u64                        start;                /*     0     8 */
> > > -       u32                        len;                  /*     8     4 */
> > > -       u32                        folio_size;           /*    12     4 */
> > > +       u32                        folio_size;           /*     8     4 */
> > > +
> > > +       /* XXX 4 bytes hole, try to pack */
> > > +
> > >         long unsigned int          bflags;               /*    16     8 */
> > >         struct btrfs_fs_info *     fs_info;              /*    24     8 */
> > >         void *                     addr;                 /*    32     8 */
> > > @@ -5554,8 +5556,8 @@ struct extent_buffer {
> > >         struct rw_semaphore        lock;                 /*    72    40 */
> > >         struct folio *             folios[16];           /*   112   128 */
> > >
> > > -       /* size: 240, cachelines: 4, members: 14 */
> > > -       /* sum members: 238, holes: 1, sum holes: 2 */
> > > +       /* size: 240, cachelines: 4, members: 13 */
> > > +       /* sum members: 234, holes: 2, sum holes: 6 */
> > >         /* forced alignments: 1, forced holes: 1, sum forced holes: 2 */
> > >         /* last cacheline: 48 bytes */
> > >  } __attribute__((__aligned__(8)));
> > >
> > > The benefit of duplicating the length in each eb is that it's in the
> > > same cacheline as the other members that are used for offset
> > > calculations or bit manipulations.
> > >
> > > Going to the fs_info->nodesize may or may not hit a cache, also because
> > > it needs to do 2 pointer dereferences, so from that perspective I think
> > > it's making it worse.
> >
> > I was considering that. Since fs_info is shared for all ebs and other
> > stuff like transactions, etc. I think the cache is hot most of the
> > time and there will be hardly any performance difference observable.
> > Though without benchmarks this is just a speculation (on both sides).
> >
> > > I don't think we need to do the optimization right now, but maybe in the
> > > future if there's a need to add something to eb. Still we can use the
> > > remaining 16 bytes up to 256 without making things worse.
> >
> > This really depends on configuration. On my laptop (Debian -rt kernel)
> > the eb struct is actually 272 bytes as the rt_mutex is significantly
> > heavier than raw spin lock. And -rt is a first class citizen nowadays,
> > often used in Kubernetes deployments like 5G RAN telco, dpdk and such.
> > I think it would be nice to slim the struct below 256 bytes even there
> > if that's your aim.
> 
> Eventually we can get there by using ushort for bflags and moving
> log_index and folio_shift to fill the hole.
> Let me know what you think.

The bflags are atomic bits and this requires unsigned long. Also the
short int type is something we want to avoid because it's not a natural
type on many architectures and generates worse code. I don't think we
need to optimize for RT kernels, it's now part of mainline kernel but by
far not a common configuration.