Use memcpy(3) to copy pci_conf data into compat structures instead of assigning
members one by one.

Good to see more Capsicumization efforts going on. I just had a few pedantic suggestions. Otherwise, this generally looks good.

In D48222#1101641, @markj wrote:

Another option might be to extend connectat(2) to support IP sockets. For instance, connectat(s, {AF_INET, 0.0.0.0, 0}), disallowed in capability, could permit sendto(s) in capability mode. Or, connectat(s, {AF_INET, 192.168.1.0}, 0) permits sendto() to 192.168.1.0/24. This would allow syslogd to transmit to any sockaddr, so weakens the sandbox a bit, but it makes userspace much simpler.

I do not like how complicated this is getting, but the code all looks good.

Thanks for finding and fixing all of these.

In D26456#1087673, @jfree wrote:

In D26456#1087672, @arrowd wrote:

Any progress on this?

Yes, I’ve been working on it. I will try to get them in this weekend. Sorry for the wait

Add missing error checking on memory allocation

• Fix sysctl(8) usage
• Do not use atf_skip(3) in test head - it causes failure

• Fix argument parsing in syslogd_start()
• Assign address to interface for host_action test

Use sendmsg(2) to send iovec list.

In D26456#1087672, @arrowd wrote:

Any progress on this?

Separate the "basic" test into three separate tests:

• unix
• inet
• inet6

• Use sysctl's -n flag to print value without variable name. Silly mistake on my part.
• Skip test if INET and INET6 are unsupported by kernel.

• Minor style changes
• Only call getsockopt() and getpeername() when Debug is non-zero

• Use calloc(3) instead of malloc(3) for array allocations.
• Add missing error checking for a memory allocation

Address Mark's comments: use calloc() instead of malloc() for
memory allocations. This is done to avoid potential integer overflow
in total allocation size.

Address Mark's less complicated comments

Bumping this

Bumping this

Bumping this

Bumping this

The ioctl subsystem section was not modified from NetBSD's manual page, so it may be inaccurate for FreeBSD. Any help adding FreeBSD-specific subsystem character codes would be great.

Address compatibility concerns pointed out by Warner

In D45755#1050226, @bz wrote:

Anyone? Anyone who can test the Intel debug on drm code?

Do not update pci_conf_old and pci_conf_old32 structures. These
should stay static for compatibility with binaries that were
compiled to use the PCIOCGETCONF_OLD and PCIOCGETCONF_OLD32 ioctls.