Re: [GIT PULL] Btrfs updates for 6.11

[David Sterba <dsterba@suse.cz>]

On Fri, Jul 19, 2024 at 01:35:53PM +0200, Geert Uytterhoeven wrote:
> Hi David,
> 
> On Fri, Jul 19, 2024 at 1:25 PM Geert Uytterhoeven <geert@linux-m68k.org> wrote:
> > On Mon, Jul 15, 2024 at 8:12 PM David Sterba <dsterba@suse.com> wrote:
> > > please pull the changes described below. The hilights are new logic
> > > behind background block group reclaim, automatic removal of qgroup
> > > after removing a subvolume and new 'rescue=' mount options. The rest is
> > > optimizations, cleanups and refactoring.
> > >
> > > There's a merge conflict caused by the latency fixes from last week in
> > > extent_map.c:btrfs_scan_inode(), related commits 4e660ca3a98d931809734
> > > and b3ebb9b7e92a928344a. Resolved in branch for-6.11-merged and that's
> > > been in linux-next for a few days.
> >
> > FTR, this is broken on 32-bit (doesn't build, good ;-) and on big-endian
> > (compiler warnings, no idea how it behaves :-(, so you better don't
> > trust your data to it in the latter case...

Internet forums are full of such quick and wrong conclusions, you don't
need to write more.

> I cannot find any other report of this, and don't know yet where it
> was introduced, but the bots started reporting this last May:
> 
>     3    fs/btrfs/inode.c:5711:5: warning: ‘location.type’ may be used
> uninitialized in this function [-Wmaybe-uninitialized]
>     3    fs/btrfs/inode.c:5640:9: warning: ‘location.objectid’ may be
> used uninitialized in this function [-Wmaybe-uninitialized]
> 
> https://lore.kernel.org/all/6655b55f.170a0220.406f9.2e0e@mx.google.com/
> 
> and I'm seeing failures in e.g. my m68k allmodconfig builds with
> gcc 9.5 due to CONFIG_WERROR=y.

Older compilers like 9.5 could not be able to reason about variable
validity in case it's passed by address, as is in this case:

 5707         ret = btrfs_inode_by_name(BTRFS_I(dir), dentry, &location, &di_type);
 5708         if (ret < 0)
 5709                 return ERR_PTR(ret);

and btrfs_inode_by_name() returns either a valid 'location' or an error that
the caller handles and does not use the variable.

> I suspect the big-endian accessors in fs/btrfs/accessors.h lack some
> initializations?

There are no special accessors on big endian hosts, same code, same
bytes in memory only a different order.

We fix warnings caused -Wmaybe-uninitialized even if it's because of old
compilers, but it's hard to notice reports if they're burried in some
mailinglist.

I do read your build reports after each -rcN but there are only some
modpost warnings in 6.10-rc7

https://lore.kernel.org/all/20240710083744.2885335-1-geert@linux-m68k.org/