Re: [RFC 0/1] Check for pid re-use before killing domain process
I believe that pidfd syscalls were introduced in kernel 5.2. Judging by
our CI build setup, the oldest distrubution that we support is Alma
Linux 8, which still has kernel version 4.18.
On 10/6/22 4:33 AM, manish.mishra wrote:
Libvirt stores pid of domain(e.g Qemu) process when a domain process
is started
and same pid is used while destroying domain process. There is always a race
possible that before libvirt tries to kill domain process, actual domain process
is already dead and same pid is used for another process. In that case libvirt
may kill an innocent process.
With this patch we store start time of domain process when domain is started and
we match stime again while killing domain process if it does not match we can
assume domain process is already dead.
This patch series tries to create a generic interface which can be used for
non-domain processes too, even though initial patch series handles pid re-use
for domain process only.
Proposed changes:
1. While creating a daemon process through virExec, create a file which stores
start time of the process, along with the pid file. A Separate file for start
time is created instead of storing start time too in *.pid file so that
existing external user of <domain-id>*.pid file does not break.
2. Persist stime in domstatus in domain xml.
3. In virProcessKill before sending signal to process also verify start time of
process. For sending signal to process pidfd_send_signal is used avoid any
race between verifying start time and sending signal. Following steps are
taken while killing a process.
3.1 Open a pid-fd for given pid with pidfd_open.
3.2 Verify start time of pid with stored start time. If start time does not
match, assume process is already dead.
3.3 Send signal to pid-fd with pidfd_send_signal. Even if pid was re-used
just after verifying star time, signal will be sent to a pidfd instance
which was created before verifying timestamp.
manish.mishra (1):
Check for pid re-use before killing qemu process
src/conf/domain_conf.c | 14 ++-
src/conf/domain_conf.h | 1 +
src/libvirt_private.syms | 7 ++
src/qemu/qemu_domain.h | 1 +
src/qemu/qemu_process.c | 49 ++++++--
src/util/vircommand.c | 37 ++++++
src/util/vircommand.h | 3 +
src/util/virpidfile.c | 113 +++++++++++++-----
src/util/virpidfile.h | 17 +++
src/util/virprocess.c | 73 ++++++++++-
src/util/virprocess.h | 8 ++
.../qemustatusxml2xmldata/backup-pull-in.xml | 2 +-
.../blockjob-blockdev-in.xml | 2 +-
.../blockjob-mirror-in.xml | 2 +-
.../migration-in-params-in.xml | 2 +-
.../migration-out-nbd-bitmaps-in.xml | 2 +-
.../migration-out-nbd-in.xml | 2 +-
.../migration-out-nbd-out.xml | 2 +-
.../migration-out-nbd-tls-in.xml | 2 +-
.../migration-out-nbd-tls-out.xml | 2 +-
.../migration-out-params-in.xml | 2 +-
tests/qemustatusxml2xmldata/modern-in.xml | 2 +-
tests/qemustatusxml2xmldata/upgrade-in.xml | 2 +-
tests/qemustatusxml2xmldata/upgrade-out.xml | 2 +-
.../qemustatusxml2xmldata/vcpus-multi-in.xml | 2 +-
25 files changed, 295 insertions(+), 56 deletions(-)