On Thu, Apr 08, 2021 at 01:19:33PM +0100, Christoph Hellwig wrote: > On Wed, Apr 07, 2021 at 06:01:14PM -0700, Darrick J. Wong wrote: > > From: Darrick J. Wong <djwong@xxxxxxxxxx> > > > > Now that the scrub context stores a pointer to the file that was used to > > invoke the scrub call, the struct xfs_inode pointer that we passed to > > all the setup functions is no longer necessary. This is only ever used > > if the caller wants us to scrub the metadata of the open file. > > Even before we had the xfs_inode in struct xfs_scrub, so why detour > through struct file? xfs_scrub.fil[ep] is the file corresponding to the fd that userspace used to call scrub. xfs_scrub.ip is the file that we're actively checking, if applicable. This can be a different file than filp if we're checking file-based metadata such as the rt bitmap and summary files; or if the caller is doing scrub-by-handle for a non-regular file (e.g. open the root directory and pass in the ino/gen of a symlink file to check that the symlink is ok). Let me respin this with added struct commentary in scrub.h. --D
