On Fri, Jun 24, 2011 at 01:58:27PM -0700, Justin TerAvest wrote: > fs_excl is a poor man's priority inheritance for filesystems to hint to > the block layer that an operation is important. It was never clearly > specified, not widely adopted, and will not prevent starvation in many > cases (like across cgroups). > > I talked to Ted Ts'o about this, and he said that it used to used more > frequently in the 2.4 and prior versions of Linux, back when we were > first converting from the Big Kernel Lock to having subsystem level > locks, and so it made sense to use fs_excl when a process owned the > global fs mutex and was waiting for an I/O to complete, but it's no > longer used much at all, and filesystems have better ways to mark an I/O > request as high priority. That's not quite true, it was added in Linux 2.6.13 in commit 22e2c507c301c3dbbcf91b4948b88f78842ee6c9: [PATCH] Update cfq io scheduler to time sliced design The users back then where the same as today: a few reiserfs journal callsites and lock_super. In addition to the lock_super uses in various fringe filesystems still left today it was also used around ->put_super (aka umount) and ->write_super, which at the point had already lost the grunt work of sync action to ->sync_fs. That beeing said I never liked it and asked for a removal a while ago, but Jens still wanted to keep it. As far as I'm concerned we should kill it gently, and if any regressions arise fix them with a bio flag. -- To unsubscribe from this list: send the line "unsubscribe reiserfs-devel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
