On Tue, Apr 09, 2019 at 05:45:54PM -0700, Eric Biggers wrote: > On Wed, Apr 10, 2019 at 01:33:46AM +0100, Al Viro wrote: > > On Tue, Apr 09, 2019 at 04:35:44PM -0700, Eric Biggers wrote: > > > From: Eric Biggers <ebiggers@xxxxxxxxxx> > > > > > > Path lookups that traverse encrypted symlink(s) are very slow because > > > each encrypted symlink needs to be decrypted each time it's followed. > > > > > > Make encrypted symlinks faster by caching the decrypted symlink target > > > in ->i_link. The first call to ->get_link() sets it; later calls simply > > > return it. ->symlink() also sets it when the symlink is created. > > > > > > When the inode's ->i_crypt_info is freed, ->i_link is freed too. > > > > > > Note: RCU-delayed freeing of ->i_link is not yet implemented. > > > Therefore, for now even when ->i_link is set, path lookups must continue > > > to drop out of RCU-walk mode when following an encrypted symlink. > > > > And how the devil would they continue to do that, if I might ask? > > ->get_link() is *NOT* called if ->i_link is non-NULL, period. > > You're right, I didn't notice that ->get_link() isn't called when > ->i_link is non-NULL. > > But that being the case, what's the point of simple_get_link()? Non-NULL ->get_link() => DCACHE_SYMLINK_TYPE in ->d_flags => d_is_symlink() true => step_into() progresses to pick_link(). IOW, non-NULL ->get_link() is what tells you that we have a symlink there.