Re: [PATCH] Fix of dcache race leading to busy inodes on umount

[Andrew Morton <akpm@osdl.org>]

Kirill Korotaev <dev@sw.ru> wrote:
>
> This patch fixes dcache race between shrink_dcache_XXX functions
> and dput().
> 
> Example race scenario:
> 
> 	CPU 0				CPU 1
> umount /dev/sda1
> generic_shutdown_super          shrink_dcache_memory()
> shrink_dcache_parent            dput dentry
> select_parent                   prune_one_dentry()
>                                  <<<< child is dead, locks are released,
>                                    but parent is still referenced!!! >>>>
> 
> skip dentry->parent,
> since it's d_count > 0
> 
> message: BUSY inodes after umount...
>                                  <<< parent is left on dentry_unused
> 				   list, referencing freed super block
> 
> We faced these messages about busy inodes constantly after some stress 
> testing with mount/umount operations parrallel with some other activity.
> This patch helped the problem.
> 
> The patch was heavilly tested on 2.6.8 during 2 months,
> this forward-ported version boots and works ok as well.
> 

You've provided no description of how the patch solves the problem.

> /* dcache_lock protects shrinker's list */
> static void shrink_dcache_racecheck(struct dentry *parent, int *racecheck)
> {
> 	struct super_block *sb;
> 	struct dcache_shrinker *ds;
> 
> 	sb = parent->d_sb;
> 	list_for_each_entry(ds, &sb->s_dshrinkers, list) {
> 		/* is one of dcache shrinkers working on the dentry? */
> 		if (ds->dentry == parent) {
> 			*racecheck = 1;
> 			break;
> 		}
> 	}
> }
> 
> 

This all looks awfully hacky.  Why is it done this way, and is there no
cleaner solution?