We have something very odd in the end of path_openat():
if (error == -EOPENSTALE) {
if (flags & LOOKUP_RCU)
error = -ECHILD;
else
error = -ESTALE;
}
Note that *nothing* that may return EOPENSTALE is ever called in
RCU mode, pretty much by definition - we must have done something
blocking to have gotten that and in RCU mode that's not going to
happen.
This check does not look for RCU mode, though - it checks whether
we'd *started* in RCU mode, ignoring any successful unlazy that
might have landed us in non-RCU mode. So this ECHILD is not
a dead code.
It really looks like it ought to have been - there's nothing
a retry in non-RCU mode would have solved; the checks on ->d_seq
should've guaranteed that there had been possible timings for
just that tree traversal in non-RCU mode, hitting exact same
dentries. If they hadn't, we have a much worse problem there...
Miklos, could you recall what was the original intent of that?
Do we want to keep that logics there, or should it just turn into
"map -ENOPENSTALE to -ESTALE??
