On Tue, Jan 02, 2024 at 05:15:11PM -0800, Darrick J. Wong wrote: > On Tue, Jan 02, 2024 at 02:33:34AM -0800, Christoph Hellwig wrote: > > On Sun, Dec 31, 2023 at 12:17:28PM -0800, Darrick J. Wong wrote: > > > From: Darrick J. Wong <djwong@xxxxxxxxxx> > > > > > > Certain btree flags never change for the life of a btree cursor because > > > they describe the geometry of the btree itself. Encode these in the > > > btree ops structure and reduce the amount of code required in each btree > > > type's init_cursor functions. > > > > I like the idea, but why are the geom_flags mirrored into bc_flags > > instead of beeing kept entirely separate and accessed as > > cur->bc_ops->geom_flags which would be a lot easier to follow? > > Oh! That hadn't occurred to me. Let me take a look at that. Eeeeyugh, this became kind of a mess. These XFS_BTREE_ flags describe btree geometry, are set in the bc_ops->geom_flags, and never change: 1. XFS_BTREE_LONG_PTRS 2. XFS_BTREE_ROOT_IN_INODE 3. XFS_BTREE_IROOT_RECORDS /* rt rmap patchset */ 4. XFS_BTREE_IN_XFILE 5. XFS_BTREE_OVERLAPPING This one flag describes geometry but is set dynamically by xfs_btree_alloc_cursor. Some of the geom_flags (rmap, refcount) can set it directly too, since they don't exist in V4 filesystems: 6. XFS_BTREE_CRC_BLOCKS This one flag doesn't describe btree geometry but never changes and could be set in bc_ops->geom_flags: 7. XFS_BTREE_LASTREC_UPDATE The remaining flag actually describes per-cursor state: 8. XFS_BTREE_STAGING Flags 1-5 can be referenced directly from geom_flags. Flag 6 could be replaced by an xfs_has_crc call, though I'd bet it's cheaper to test a cursor variable than to walk to the xfs_mount and test_bit. But this feels weird. Flag 7 is set in geom_flags as it should be. Flag 8 is really a runtime flag, so it can stay in bc_flags. *or* I could rename geom_flags to default_bcflags and make it clearer that it's used to seed cur->bc_flags? --D
