test/units/testsuite-04.bsod.sh

   1 #!/usr/bin/env bash
   2 # SPDX-License-Identifier: LGPL-2.1-or-later
   3 set -eux
   4 set -o pipefail
   5
   6 if systemd-detect-virt -cq; then
   7     echo "This test requires a VM, skipping the test"
   8     exit 0
   9 fi
  10
  11 # shellcheck disable=SC2317
  12 at_exit() {
  13     local EC=$?
  14
  15     if [[ $EC -ne 0 ]] && [[ -e /tmp/console.dump ]]; then
  16         cat /tmp/console.dump
  17     fi
  18
  19     if mountpoint -q /var/log/journal; then
  20         journalctl --relinquish-var
  21         umount /var/log/journal
  22         journalctl --flush
  23     fi
  24
  25     return 0
  26 }
  27
  28 vcs_dump_and_check() {
  29     local expected_message="${1:?}"
  30
  31     # It might take a while before the systemd-bsod stuff appears on the VCS,
  32     # so try it a couple of times
  33     for _ in {0..9}; do
  34         setterm --term linux --dump --file /tmp/console.dump
  35         if grep -aq "Press any key to exit" /tmp/console.dump
  36             grep -aq "$expected_message" /tmp/console.dump
  37             grep -aq "The current boot has failed" /tmp/console.dump; then
  38
  39             return 0
  40         fi
  41
  42         sleep .5
  43     done
  44
  45     return 1
  46 }
  47
  48 # Since systemd-bsod always fetches only the first emergency message from the
  49 # current boot, let's temporarily overmount /var/log/journal with a tmpfs,
  50 # as we're going to wipe it multiple times, but we need to keep the original
  51 # journal intact for the other tests to work correctly.
  52 trap at_exit EXIT
  53 mount -t tmpfs tmpfs /var/log/journal
  54 systemctl restart systemd-journald
  55
  56 systemctl stop systemd-bsod
  57
  58 # Since we just wiped the journal, there should be no emergency messages and
  59 # systemd-bsod should be just a no-op
  60 timeout 10s /usr/lib/systemd/systemd-bsod
  61 setterm --term linux --dump --file /tmp/console.dump
  62 (! grep "The current boot has failed" /tmp/console.dump)
  63
  64 # systemd-bsod should pick up emergency messages only with UID=0, so let's check
  65 # that as well
  66 systemd-run --user --machine testuser@ --wait --pipe systemd-cat -p emerg echo "User emergency message"
  67 systemd-cat -p emerg echo "Root emergency message"
  68 journalctl --sync
  69 # Set $SYSTEMD_COLORS so systemd-bsod also prints out the QR code
  70 SYSTEMD_COLORS=256 /usr/lib/systemd/systemd-bsod &
  71 PID=$!
  72 vcs_dump_and_check "Root emergency message"
  73 grep -aq "Scan the QR code" /tmp/console.dump
  74 # TODO: check if systemd-bsod exits on a key press (didn't figure this one out yet)
  75 kill $PID
  76 timeout 10 bash -c "while kill -0 $PID; do sleep .5; done"
  77
  78 # Wipe the journal
  79 journalctl --vacuum-size=1 --rotate
  80 (! journalctl -q -b -p emerg --grep .)
  81
  82 # Check the systemd-bsod.service as well
  83 # Note: the systemd-bsod.service unit has ConditionVirtualization=no, so let's
  84 # temporarily override it just for the test
  85 mkdir /run/systemd/system/systemd-bsod.service.d
  86 printf '[Unit]\nConditionVirtualization=\n' >/run/systemd/system/systemd-bsod.service.d/99-override.conf
  87 systemctl daemon-reload
  88 systemctl start systemd-bsod
  89 systemd-cat -p emerg echo "Service emergency message"
  90 vcs_dump_and_check "Service emergency message"
  91 systemctl stop systemd-bsod
  92
  93 # Wipe the journal
  94 journalctl --vacuum-size=1 --rotate
  95 (! journalctl -q -b -p emerg --grep .)
  96
  97 # Same as above, but make sure the service responds to signals even when there are
  98 # no "emerg" messages, see systemd/systemd#30084
  99 (! systemctl is-active systemd-bsod)
 100 systemctl start systemd-bsod
 101 timeout 5s bash -xec 'until systemctl is-active systemd-bsod; do sleep .5; done'
 102 timeout 5s systemctl stop systemd-bsod
 103 timeout 5s bash -xec 'while systemctl is-active systemd-bsod; do sleep .5; done'