Messages in this thread | | | From | Eric Dumazet <> | Subject | Re: [PATCH 2.6.19-rc3] VFS: per-sb dentry lru list | Date | Mon, 30 Oct 2006 16:08:25 +0100 |
| |
On Monday 30 October 2006 15:24, Vasily Averin wrote: > Andrew Morton wrote: > > On Fri, 27 Oct 2006 18:05:50 +0400 > > > > Vasily Averin <vvs@sw.ru> wrote: > >> Virtuozzo/OpenVZ linux kernel team has discovered that umount/remount > >> can last for hours looping in shrink_dcache_sb() without much successes. > >> Since during shrinking s_umount semaphore is taken lots of other > >> unrelated operations like sync can stop working until shrink finished. > > > > Did you consider altering shrink_dcache_sb() so that it holds onto > > dcache_lock and moves all the to-be-pruned dentries onto a private list > > in a single pass, then prunes them all outside the lock? > > At the first glance it is wrong because of 2 reasons: > 1) it continues to check the whole global LRU list (we propose to use > per-sb LRU, it will provide very quick search)
Quick search maybe, but your patch adds 2 pointers to each dentry in the system... That's pretty expensive, as dentries are already using a *lot* of ram.
Maybe an alternative would be to not have anymore a global dentry_unused, but only per-sb unused dentries lists ?
> 2) we have not any guarantee that someone will add new unused dentries to > the list when we prune it outside the lock. And to the contrary, some of > unused dentries can be used again. As far as I understand we should hold > dcache_lock beginning at the removing dentry from unused_list until > dentry_iput() call. > > David did it inside shrink_dcache_for_umount() just because it have > guarantee that all the filesystem operations are finished and new ones > cannot be started.
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
| |