On Sat, Oct 25, 2014 at 10:18 AM, Al Viro <viro@xxxxxxxxxxxxxxxxxx> wrote: > On Fri, Oct 24, 2014 at 09:24:45AM +0200, Miklos Szeredi wrote: > >> The reason I didn't do your "fix" is that it >> >> - adds more lines than it takes, >> >> - I wasn't sure at all if the lockless access is actually correct >> without the ACCESS_ONCE and all the memory barrier magic that might be >> necessary on weird architectures. > > _What_ lockless accesses? There is an extremely embarrassing bug in that > commit, all right, but it has nothing to do with barriers... All > barrier-related issues are taken care of by ovl_path_upper() (and without > that you'd have tons of worse problems). Fetching ->upperfile outside of > ->i_mutex is fine - in the worst case we'll fetch NULL, open the sucker > grab ->i_mutex and find out that it has already been taken care of. > In which case we fput() what we'd opened and move on (fput() under > ->i_mutex is fine - it's going to be delayed until return from syscall > anyway). Yes, but it's not about race with copy-up (which the ovl_path_upper() protects against), but race of two fsync calls with each other. If there's no synchronization between them, then that od->upperfile does indeed count as lockless access, no matter that the assignment was done under lock. Thanks, Miklos -- To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
