suspend/resume time-gap and expiration timers in network stack

[Hiroki Sato]

Hi,

 I recently committed s/time_second/time_uptime/ replacement in
 sys/netinet6 which was done in sys/netinet some years ago.  It works
 well and should be more robust against time_second deviation (by NTP,
 for example).  However, IPv6 default routers and prefixes do not
 expire as before if the machine is suspended and then resumed after a
 long time, say, 3 days.

 I think the impact is limited because suspend/resume causes to reset
 all of the physical network interfaces in some way, but in some cases
 the decayed entries can persist.  So I would like your comments about
 the following idea to solve this:

 1. Record RTC value upon suspend.

 2. Calculate the time-gap by using the recorded value and the current
    one upon resume.

 3. Store the calculated time-gap in time_uptime_gap variable and use
    (time_uptime + time_uptime_gap) for expiration timers.

 If there is no suspend/resume event, nothing happens.  If any, the
 time-gap is added to the expiration time.  The time_uptime_gap
 variable is always positive, so (time_uptime + time_uptime_gap) never
 goes backwards.

 A experimental patch is attached.  Changes include the following:

====
- Add clock_suspend_stamp() and clock_resume_fixup() functions to
  calculate suspend/resume time-gap.  clock_suspend_stamp() record
  an RTC value just before disabling timer, and clock_resume_fixup()
  store the time-gap into time_uptime_gap variable.  Ideally,
  time_uptime + time_uptime_gap becomes equal to the elapsed time
  since boottime.

- Add time-gap calculation in ACPI suspend/resume.

- Add kern.clock_suspend_stamp and kern.clock_resume_fixup sysctls to
  allow time-gap calculation from userland.  These are write-only, and
  setting them to non-zero calls the functions.  Typically,
  kern.clock_suspend_stamp should be set to 1 just before suspend,
  and kern.clock_resume_fixup should be set to 1 just after resume.
  kern.clock_resume_fixup is idempotent.

- Add TIME_UPTIME_CALED macro.  This is defined as (time_uptime +
  time_uptime_gap).

- Use TIME_UPTIME_CALED for expiration timers in sys/netinet6.
====

 It is not so smart and I do not think it is commit-ready, but it
 helps you to understand what I want to do.  While this implementation
 cares about only ACPI suspend/resume and timers in sys/netinet6, this
 can be applied to VM suspend/resume and sys/net and sys/netinet
 expiration timers like one for ARP table.

-- Hiroki
-------------- next part --------------
A non-text attachment was scrubbed...
Name: uptime_gap.20130817-1.diff
Type: text/x-patch
Size: 22063 bytes
Desc: not available
URL: <http://lists.freebsd.org/pipermail/freebsd-arch/attachments/20130817/10774d58/attachment.bin>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 196 bytes
Desc: not available
URL: <http://lists.freebsd.org/pipermail/freebsd-arch/attachments/20130817/10774d58/attachment.sig>