Details

Reviewers

kib
menyy_mellanox.com
ae

Group Reviewers

network

Commits

rS333624: MFC r333362:
rS333362: Fix for missing network interface address event when adding the default IPv6

Summary

After adding the default IPv6 default link-local address, the kernel is not informed about the new network interface address. Move the EVENTHANDLER_INVOKE() call to a more appropriate location.

This event is needed by ibcore which regularly listen for network address changes.
The issue manifests itself when creating a VLAN network interface and configuring a IPv4 address only which also brings up the IPv6 link-local address, but the ifaddr_event is not invoked for the latter and then ibcore will not see the IPv6 link-local address added only the IPv4 one.

Diff Detail

Repository

rS FreeBSD src repository - subversion

Lint

Lint Skipped

Unit

Tests Skipped

Build Status

Buildable 16475

Event Timeline

• hselasky created this revision.Dec 7 2017, 2:44 PM

Herald added a subscriber: imp. · View Herald TranscriptDec 7 2017, 2:44 PM

For my understanding of this stuff. Is some lock which guarantees consistency for the addresses configuration is still held while the eventhandler is invoked ? What is the lock ?

netinet6/in6_ifattach.c
550 ↗	(On Diff #36337)	The comment is actually useless. It is quite obvious for any reader that an eventhandler is called, so the text repeats the code.

@kib: Other places where "EVENTHANDLER_INVOKE(ifaddr_event, ifp)" is used there are no locks involved from what I can see. Correct me if I'm wrong.

In D13407#280108, @hselasky wrote:

@kib: Other places where "EVENTHANDLER_INVOKE(ifaddr_event, ifp)" is used there are no locks involved from what I can see. Correct me if I'm wrong.

Then how e.g. validity of ifp is ensured for the duration of the eventhandler callback ? Also, do eventhandlers re-check the condition they are called for ?

If not, this is quite hideous.

Then how e.g. validity of ifp is ensured for the duration of the eventhandler callback ? Also, do eventhandlers re-check the condition they are called for ?

If not, this is quite hideous.

The "ifp" pointer is already used several times inside the same function, so I assume it is safe to call the eventhandler with this network interface pointer aswell.

Remove useless comment as per KIB's suggestion.

Herald added a subscriber: ae. · View Herald TranscriptMay 4 2018, 11:13 AM

Harbormaster completed remote builds in B16427: Diff 42134.May 4 2018, 11:13 AM

• hselasky added a reviewer: menyy_mellanox.com.May 6 2018, 5:35 PM

I've checked the callers of in6_ifattach_linklocal(), and there appears to be no fast path involved, which means among others that sleeping is OK inside the eventhandler callbacks.

• hselasky removed reviewers: glebius, gnn.May 8 2018, 8:25 AM

kib accepted this revision.May 8 2018, 8:48 AM

This revision is now accepted and ready to land.May 8 2018, 8:48 AM

• hselasky added a reviewer: ae.May 8 2018, 8:52 AM