Re: [PATCH] btrfs: replace kcalloc() calls to kzalloc_objs()

[David Sterba <dsterba@suse.cz>]

On Tue, Feb 24, 2026 at 02:55:55PM +0000, David Laight wrote:
> On Tue, 24 Feb 2026 15:07:10 +1030
> Qu Wenruo <quwenruo.btrfs@gmx.com> wrote:
> 
> > 在 2026/2/24 10:14, Miquel Sabaté Solà 写道:
> > > Commit 2932ba8d9c99 ("slab: Introduce kmalloc_obj() and family")
> > > introduced, among many others, the kzalloc_objs() helper, which has some
> > > benefits over kcalloc().
> > > 
> > > Cc: Kees Cook <kees@kernel.org>
> > > Signed-off-by: Miquel Sabaté Solà <mssola@mssola.com>
> > > ---
> > >   fs/btrfs/block-group.c       | 2 +-
> > >   fs/btrfs/raid56.c            | 8 ++++----
> > >   fs/btrfs/tests/zoned-tests.c | 2 +-
> > >   fs/btrfs/volumes.c           | 6 ++----
> > >   fs/btrfs/zoned.c             | 5 ++---
> > >   5 files changed, 10 insertions(+), 13 deletions(-)
> > > 
> > > diff --git a/fs/btrfs/block-group.c b/fs/btrfs/block-group.c
> > > index 37bea850b3f0..8d85b4707690 100644
> > > --- a/fs/btrfs/block-group.c
> > > +++ b/fs/btrfs/block-group.c
> > > @@ -2239,7 +2239,7 @@ int btrfs_rmap_block(struct btrfs_fs_info *fs_info, u64 chunk_start,
> > >   	if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
> > >   		io_stripe_size = btrfs_stripe_nr_to_offset(nr_data_stripes(map));
> > >   
> > > -	buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
> > > +	buf = kzalloc_objs(*buf, map->num_stripes, GFP_NOFS);  
> > 
> > Not sure if we should use *buf for the type.
> > 
> > I still remember we had some bugs related to incorrect type usage.
> 
> The global change really ought to have used u64 to add the type-check.
> Otherwise it will have added 'very hard to find' bugs in the very code
> it is trying to make better.
> 
> Using *buf for the type might be a reasonable pattern for new code.

I find this a bit contradictory: I agree that using *buf as the argument
can cause bugs hard to find, yet the next sentence recommends to use it.

This kzalloc_obj way is new I'm analyzing what would be a good pattern
and so far I don't like the "*buf" style of 1st argument. As the
function is really a macro it does not dereference it but it still
appears as it does.

Writing the type explicitly looks still more like a C to me. Types in
arguments are in helpers like container_of or rb_entry and it makes it
obvious that there's something special while for the kzalloc_obj I need
to remember it.

The whole thing would read better as "allocate object of type", so I'm
probably going to convert it to this pattern in btrfs code.