fifos: delegate unhandled kqueue filters to underlying filesystem

[kevans]

Summary

This gives the vfs layer a chance to provide handling for EVFILT_VNODE,
for instance. Change pipe_specops to use the default vop_kqfilter to
accommodate fifoops that don't specify the method (i.e. all in-tree).

Based on a patch by Jan Kokemüller.

PR: 225934

[kevans]

vop_stdkqfilter is the default.

[kevans]

sys/fs/fifofs/fifo_vnops.c

86

I had considered adding a small fifo_kqfilter that asserts we don't get any EVFILT_READ/EVFILT_WRITE since these should either be rejected by or handled by pipe_kqfilter, but I'm not sure.

[markj]

I wonder if it would be useful to support the same thing for bound unix sockets.

sys/fs/fifofs/fifo_vnops.c

86

Perhaps instead modify vfs_kqfilter() to assert vp->v_type != VFIFO || kn->kn_filter == EVFILT_VNODE. I don't have any particular objection to adding an INVARIANTS-only VOP though.

[kevans]

sys/fs/fifofs/fifo_vnops.c

86

I guess that raises another question- should this have been scoped to just passing through EVFILT_VNODE? It doesn't seem like there's anything in particular preventing us from passing along a (bogus) EVFILT_PROCDESC or EVFILT_EMPTY

[markj]

sys/fs/fifofs/fifo_vnops.c

86

It seems a bit more maintainable to simply pass unknown filters through to vop_kqfilter(), rather than trying to second-guess the implementation, especially since VOP_KQFILTER could be overridden by a fs that implements custom filters. (OneFS does exactly this, in fact.) The assertion can simply verify that the filter is not EVFILT_READ or _WRITE.