sts-sponsors team mailing list archive

[Mauricio Faria de Oliveira <1947099@xxxxxxxxxxxxxxxxxx>]

Setting verification-done-bionic per comments 39/40.

** Tags removed: verification-needed verification-needed-bionic
** Tags added: verification-done verification-done-bionic

-- 
You received this bug notification because you are a member of STS
Sponsors, which is subscribed to the bug report.
https://bugs.launchpad.net/bugs/1947099

Title:
  ipconfig does not honour user-requested timeouts in some cases

Status in klibc package in Ubuntu:
  Won't Fix
Status in klibc source package in Bionic:
  Fix Released

Bug description:
  [Impact]
  In some cases, ipconfig can take a longer time than the user-specified timeouts, causing unexpected delays.

  [Test Plan]

  - Check that the ipconfig utility is able to obtain an IP through
  dhcp:

  # ip l set dev ens9 down
  # date; /usr/lib/klibc/bin/ipconfig ens9; date

  - Check if it respects the timeout (i.e. 2 seconds in the example
  below):

  # ip l set dev ens9 down
  # date; /usr/lib/klibc/bin/ipconfig -t 2 ens9; date

  - The original issue is that timeout is being ignored when not
  receiving any reply from a DHCP Server, so for this test you have to
  stop your DHCP Server (i.e. dnsmasq) and then test the timeouts:

  # ip l set dev ens9 down
  # date; /usr/lib/klibc/bin/ipconfig -t 2 ens9; date

  # ip l set dev ens9 down
  # date; /usr/lib/klibc/bin/ipconfig -t 5 ens9; date

  Make sure it times out respectivelly in 2 and 5 seconds.

  [racb: pending agreement with the SRU team; please see comment 37]

  Any situation where ipconfig encounters an error sending the DHCP
  packet, it will automatically set a delay of 10 seconds, which could
  be longer than the user-specified timeout. It can be reproduced by
  creating a dummy interface and attempting to run ipconfig on it with a
  timeout value of less than 10:

  # ip link add eth1 type dummy
  # date; /usr/lib/klibc/bin/ipconfig -t 2 eth1; date
  Thu Nov 18 04:46:13 EST 2021
  IP-Config: eth1 hardware address ae:e0:f5:9d:7e:00 mtu 1500 DHCP RARP
  IP-Config: no response after 2 secs - giving up
  Thu Nov 18 04:46:23 EST 2021

  ^ Notice above, ipconfig thinks that it waited 2 seconds, but the
  timestamps show an actual delay of 10 seconds.

  [Where problems could occur]
  Please see reproduction steps above. We are seeing this in production too (see comment #2).

  [Other Info]
  A patch to fix the issue is being proposed here. It is a safe fix - it only checks before going into sleep that the timeout never exceeds the user-requested value.

  [Original Description]

  In some cases, ipconfig can take longer than the user-specified
  timeouts, causing unexpected delays.

  in main.c, in function loop(), the process can go into
  process_timeout_event() (or process_receive_event() ) and if it
  encounters an error situation, will set an attempt to "try again
  later" at time equal now + 10 seconds by setting

  s->expire = now + 10;

  This can happen at any time during the main event loop, which can end
  up extending the user-specified timeout if "now + 10" is greater than
  "start_time + user-specified-timeout".

  I believe a patch like the following is needed to avoid this problem:

  --- a/usr/kinit/ipconfig/main.c
  +++ b/usr/kinit/ipconfig/main.c
  @@ -437,6 +437,13 @@ static int loop(void)

                          if (timeout > s->expire - now.tv_sec)
                                  timeout = s->expire - now.tv_sec;
  +
  +                       /* Compensate for already-lost time */
  +                       gettimeofday(&now, NULL);
  +                       if (now.tv_sec + timeout > start + loop_timeout) {
  +                               timeout = loop_timeout - (now.tv_sec - start);
  +                               printf("Lowered timeout to match user request = (%d s) \n", timeout);
  +                       }
                  }

  I believe the current behaviour is buggy. This is confirmed when the
  following line is executed:

                          if (loop_timeout >= 0 &&
                              now.tv_sec - start >= loop_timeout) {
                                  printf("IP-Config: no response after %d "
                                         "secs - giving up\n", loop_timeout);
                                  rc = -1;
                                  goto bail;
                          }

  'loop_timeout' is the user-specified time-out. With a value of 2, in
  case of error, this line prints:

  IP-Config: no response after 2 secs - giving up

  So it thinks that it waited 2 seconds - however, in reality it had
  actually waited for 10 seconds.

  The suggested code-change ensures that the timeout that is actually
  used never exceeds the user-specified timeout.

  [ Regression potential ]

  This change ensures that user-specified timeouts are never exceeded, which is a problem that appears to happen only in case of interface errors.
  It may be that someone is relying on current behaviour where they receive DHCP offers after their specified timeout (but within the 10-second error timeout). However, 1) that is buggy behaviour and should be exposed. Such a user would need to update their specified timeout to make it long enough to receive the DHCP offer (setting the timeout to 10 would keep the existing behaviour). 2) I think it is unlikely that such a scenario exists at all. The 10-second timeout problem happens when there are problems with the interface that prevent it from even sending out the DHCP request. I think it is very unlikely (or even, impossible) that DHCP offers would be received on a dead interface.

  Based on the above points, I consider the regression potential to be
  very low for this change. I do not expect anyone who is currently
  using ipconfig successfully to notice this change.

  I believe the only difference introduced by this is the reduction of
  delays caused by dead or problematic network interfaces. Those error
  delays are shortened such that they never exceeed user-specified
  timeouts.

To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/klibc/+bug/1947099/+subscriptions