Messages in this thread | | | Date | Thu, 3 Oct 2013 07:14:49 +0100 | From | Al Viro <> | Subject | Re: [rfc][possible solution] RCU vfsmounts |
| |
On Mon, Sep 30, 2013 at 08:49:21PM +0100, Al Viro wrote: > * btrfs: wants btrfs_root_readonly(BTRFS_I(inode)->root) usable in > ->permission(). Delayed freeing of struct btrfs_root, perhaps?
Not needed, actually - it's only checked with MAY_WRITE, and we don't pass that in RCU mode.
Anyway, I've slapped what looks like a sufficient set of synchronize_rcu() in affected filesystems and re-pushed. Result is at least supposed to cover all of them. It still might bugger your memory, chew the disks, etc., so it's only for testing on scratch boxen at that point.
Still, review and (if you are really brave) testing would be very much appreciated.
I'll post the patchset on top of the current mainline in a few.
| |