On Sun, Jun 10, 2018 at 8:02 AM, Al Viro <viro@xxxxxxxxxxxxxxxxxx> wrote: > On Sun, Jun 10, 2018 at 06:49:10AM +0100, Al Viro wrote: >> On Tue, May 29, 2018 at 04:41:41PM +0200, Miklos Szeredi wrote: >> > From: Miklos Szeredi <miklos@xxxxxxxxxx> >> > >> > Split out common helper for race free insertion of an already allocated >> > inode into the cache. Use this from iget5_locked() and >> > insert_inode_locked4(). Make iget5_locked() use new_inode()/iput() instead >> > of alloc_inode()/destroy_inode() directly. >> >> ... thus hitting the sucker with ->evict_inode(), in condition that is quite >> likely to be unfit to be seen by that. >> >> NAK. > > To clarify: objection here is against the switch to new_inode/iput. The rest is > sane. What makes new_inode() better here, anyway? Umm, got to look into this. The patch has already been pulled by Linus, but I hope it's salvageable. Thanks, Miklos
