On Mon, Oct 21, 2024 at 11:43:13PM +0100, Al Viro wrote:
> On Mon, Oct 21, 2024 at 06:09:10PM +0100, Al Viro wrote:
> > On Mon, Oct 21, 2024 at 02:39:58PM +0200, Christian Brauner wrote:
> >
> > > > See #getname.fixup; on top of #base.getname and IMO worth folding into it.
> > >
> > > Yes, please fold so I can rebase my series on top of it.
> >
> > OK... What I have is #base.getname-fixed, with two commits - trivial
> > "teach filename_lookup() to accept NULL" and introducing getname_maybe_null(),
> > with fix folded in.
> >
> > #work.xattr2 and #work.statx2 are on top of that.
>
> BTW, speaking of statx() - I would rather lift the call of cp_statx() out
> of do_statx() and do_statx_fd() into the callers. Yes, that needs making
> it non-static, due to io_uring; not a problem, IMO - it fits into the
> "how do we copy internal objects to userland ones" family of helpers.
>
> Another fun issue: for by-pathname case vfs_fstatat() ends up hitting
> the same vfs_statx_path() as statx(2); however, for by-descriptor case
> they do vfs_getattr() and vfs_statx_path() resp.
>
> The difference is, vfs_statx_path() has
> if (request_mask & STATX_MNT_ID_UNIQUE) {
> stat->mnt_id = real_mount(path->mnt)->mnt_id_unique;
> stat->result_mask |= STATX_MNT_ID_UNIQUE;
> } else {
> stat->mnt_id = real_mount(path->mnt)->mnt_id;
> stat->result_mask |= STATX_MNT_ID;
> }
>
> if (path_mounted(path))
> stat->attributes |= STATX_ATTR_MOUNT_ROOT;
> stat->attributes_mask |= STATX_ATTR_MOUNT_ROOT;
>
> /*
> * If this is a block device inode, override the filesystem
> * attributes with the block device specific parameters that need to be
> * obtained from the bdev backing inode.
> */
> if (S_ISBLK(stat->mode))
> bdev_statx(path, stat, request_mask);
> done after vfs_getattr(). Questions:
>
> 1) why is STATX_MNT_ID set without checking if it's in the mask passed to
> the damn thing?
If you look at the history you'll find that STATX_MNT_ID has always been
set unconditionally, i.e., userspace didn't have to request it
explicitly. And that's fine. It's not like it's expensive.
>
> 2) why, in the name of everything unholy, does statx() on /dev/weird_shite
> trigger modprobe on that thing? Without even a permission check on it...
Block people should know. But afaict, that's a block device thing which
happens with CONFIG_BLOCK_LEGACY_AUTOLOAD enabled which is supposedly
deprecated.
