[fuse-devel] Interrupt vs. original request races

[Aaron J. <ja...@go...>]

Hi fuse-devel,

fuse.txt in the kernel documentation has this handy section about the protocol
for interrupts:

    https://github.com/torvalds/linux/blob/6c09e94a32e1094983d39879b9e8ccf9ffdcfa36/Documentation/filesystems/fuse.txt#L148-L186

If I'm reading the three cases enumerated near the top correctly, it should be
impossible for an interrupt op to be sent to userspace before the original
request. In particular, 2) says that the interrupt is queued only after the
original request is sent. (Please correct me if my reading is wrong.)

If this is the case, why does the part starting at line 173 talk about a race
between the original request and the interrupt request? Is this applicable only
if the userspace code processes requests concurrently?

That is, if I read requests on only one thread am I guaranteed that that thread
sees the original request before the interrupt?

Thanks,
Aaron