Hi Brian,
On Wed 01-11-23 15:01:22, Brian Foster wrote:
> On Wed, Nov 01, 2023 at 06:43:06PM +0100, Jan Kara wrote:
> > Convert bcachefs to use bdev_open_by_path() and pass the handle around.
> >
> > CC: Kent Overstreet <kent.overstreet@xxxxxxxxx>
> > CC: Brian Foster <bfoster@xxxxxxxxxx>
> > CC: linux-bcachefs@xxxxxxxxxxxxxxx
> > Signed-off-by: Jan Kara <jack@xxxxxxx>
> > ---
> > fs/bcachefs/super-io.c | 19 ++++++++++---------
> > fs/bcachefs/super_types.h | 1 +
> > 2 files changed, 11 insertions(+), 9 deletions(-)
> >
> > diff --git a/fs/bcachefs/super-io.c b/fs/bcachefs/super-io.c
> > index 332d41e1c0a3..01a32c41a540 100644
> > --- a/fs/bcachefs/super-io.c
> > +++ b/fs/bcachefs/super-io.c
> ...
> > @@ -685,21 +685,22 @@ int bch2_read_super(const char *path, struct bch_opts *opts,
> > if (!opt_get(*opts, nochanges))
> > sb->mode |= BLK_OPEN_WRITE;
> >
> > - sb->bdev = blkdev_get_by_path(path, sb->mode, sb->holder, &bch2_sb_handle_bdev_ops);
> > - if (IS_ERR(sb->bdev) &&
> > - PTR_ERR(sb->bdev) == -EACCES &&
> > + sb->bdev_handle = bdev_open_by_path(path, sb->mode, sb->holder, &bch2_sb_handle_bdev_ops);
> > + if (IS_ERR(sb->bdev_handle) &&
> > + PTR_ERR(sb->bdev_handle) == -EACCES &&
> > opt_get(*opts, read_only)) {
> > sb->mode &= ~BLK_OPEN_WRITE;
> >
> > - sb->bdev = blkdev_get_by_path(path, sb->mode, sb->holder, &bch2_sb_handle_bdev_ops);
> > - if (!IS_ERR(sb->bdev))
> > + sb->bdev_handle = bdev_open_by_path(path, sb->mode, sb->holder, &bch2_sb_handle_bdev_ops);
> > + if (!IS_ERR(sb->bdev_handle))
> > opt_set(*opts, nochanges, true);
> > }
> >
> > - if (IS_ERR(sb->bdev)) {
> > - ret = PTR_ERR(sb->bdev);
> > + if (IS_ERR(sb->bdev_handle)) {
> > + ret = PTR_ERR(sb->bdev_handle);
> > goto out;
> > }
> > + sb->bdev = sb->bdev_handle->bdev;
> >
> > ret = bch2_sb_realloc(sb, 0);
> > if (ret) {
>
> This all seems reasonable to me, but should bcachefs use sb_open_mode()
> somewhere in here to involve use of the restrict writes flag in the
> first place? It looks like bcachefs sort of open codes bits of
> mount_bdev() so it isn't clear the flag would be used anywhere...
Yes, so AFAICS bcachefs will not get the write restriction from the changes
in the generic code. Using sb_open_mode() in bcachefs would fix that but
given the 'noexcl' and 'nochanges' mount options it will not be completely
seamless anyway. I guess once the generic changes are in, bcachefs can
decide how exactly it wants to set the BLK_OPEN_RESTRICT_WRITES flag. Or if
you already have an opinion, we can just add the patch to this series.
Honza
--
Jan Kara <jack@xxxxxxxx>
SUSE Labs, CR
