_eth_reader logs errors when read(2) hits EINTR and EAGAIN

[eschaton]

When running a VAX simulator with networking via tap from the latest SIMH on NetBSD 9.3, I get periodic _eth_reader errors that the read would block.

This appears to be from the code in the HAVE_TAP_NETWORK calling read(2) and not handling the case where it returns <0 with errno set to EAGAIN or EINTR. The proper thing to do in those cases would be to treat the read(2) as having received 0 bytes, and retrying the loop.

I've worked around this centrally by wrapping the part that increases the error count and calls _eth_error() to avoid doing so when errno is set to EAGAIN or EINTR.

[pkoning2]

The code is a bit hard to read, but... does it set non-blocking mode on the file descriptor? If not, EAGAIN would not be expected. And why would EINTR be returned?

[eschaton]

It does appear to set non-blocking (look for FIONBIO).

[eschaton]

Oh, and any UNIX system call can return EINTR at any time, Google "PC losering problem…"