linprocfs(5): Add /proc/net/route

[john.grafton_runbox.com]

Summary

Some Linux programs require it to exist before installation.

Test Plan

Load the new linprocfs module, mount it and cat {mountpoint}/proc/net/route.

The content should be very similar to Linux proc/net/route.

I've tested this module in both 14.0-CURRENT and 13.1-RELEASE with success.

Verified with GitHub pre-commit checks here: https://github.com/freebsd/freebsd-src/pull/616

[dchagin]

sys/compat/linprocfs/linprocfs.c

1605

after malloc?

1611

please, move decl before code

1625

malloc in case of M_WAITOK can't return NULL

1627

error = kern_sysctl
if (error != 0)

goto out;

1646

out before free()

1651

return (error);

[dchagin]

sys/compat/linprocfs/linprocfs.c

1608

btw, Is that enough? rtsock messages can be up to 64k

[john.grafton_runbox.com]

Updated source from dchagin comments

[john.grafton_runbox.com]

Ideally, I'd like to only malloc only the amount of memory needed for the sysctl results. This works in user sysctl by passing a NULL to oldp but the kernel_sysctl function doesn't appear to work the same way.

[dchagin]

looks ok to me, Alex, could you take a look at network part of code?

[melifaro]

sys/compat/linprocfs/linprocfs.c

1614

I’d rather use ‘rib_walk(fibnum, af, linux_route_print)’ public KPI instead of using it indirectly via sysctl.

[john.grafton_runbox.com]

sys/compat/linprocfs/linprocfs.c

1614

Ah, thank you. rib_walk was my first inclination but I wasn't sure if it'd return the information I needed. netstat(1) uses sysctl and thus provided an example on how to decode the results.

[melifaro]

sys/compat/linprocfs/linprocfs.c

1614

It does.

You'll get 'struct rtentry' in the iterator, which allows extracting prefix data via
rt_get_inet_prefix_pmask().
Nexthop is slightly more tricky, as there can be multiple paths attached to a single prefix, so you need something like sysctl_dumpentry() code.
Finally, you'll get the nexthop pointer, where you can easily get ifp (nh_ifp), mtu (nh_mtu), and gateway (gw_sa). Note that the gateway is only relevant if (nh_flags & NHF_GATEWAY) is true. Also, the gateway can be IPv6 for 4over6 routes.

[john.grafton_runbox.com]

working version utilizing rib_walk

[john.grafton_runbox.com]

• support metric

[john.grafton_runbox.com]

Thanks again for the reviews. The rib_walk function is indeed much better than a kernel sysctl! Not sure if Linux outputs multipath routes in proc/net/route, I'll take a look at its source.

[melifaro]

Thank you for providing the updated version w/ rib_walk!
Looks good to me, please see some minor comments inline. Once addressed, it should be good to land.

sys/compat/linprocfs/linprocfs.c

94

You shouldn’t include it, that’s a private header for the routing code.

1557

Probably worth using ‘struct in_addr’ here

1575

Re multipath: to be safe probably worth using nhop_select(raw_nhop, 0) to ensure it doesnt crash with multipath routes

1578

1587

I’d suggest defining 0xF explicitely as a flag combination

1588

You shouldn’t access rtentry fields directly, use rt_get_weight().

1605

Worth using c11 initialisers:

1620

You don’t need to hold ifnet lock, but you need to be in network epoch (NET_EPOCH_ENTER())

[john.grafton_runbox.com]

sys/compat/linprocfs/linprocfs.c

1620

Without the ifnet lock, linux_ifname() panics due to IFNET_RLOCK_ASSERT()

panic: Lock (sx) ifnet_sx not locked

[john.grafton_runbox.com]

• updated from reviewer notes

[melifaro]

@dchagin: want to commit it, or should I do do it?

sys/compat/linprocfs/linprocfs.c

1557

Nit: any reason not to have dst and mask as stack variables in linux_route_print() ?

1592

[john.grafton_runbox.com]

sys/compat/linprocfs/linprocfs.c

1557

No good reason. :)

[john.grafton_runbox.com]

• dst and mask as stack variables
• simpler flag calculation

[dchagin]

In D36949#841553, @melifaro wrote:

@dchagin: want to commit it, or should I do do it?

@melifaro, as you say, I will build universe in the night, can commit it.