[PATCH common v2 0/2] cmd: run: do not leave the command behind on a caller's timeout

[Kefu Chai <k.chai@proxmox.com>]

Hi,

a colleague ran 'pvesm alloc' on an RBD pool whose placement groups
never became active: the storage lock gave up and got released, but
'rbd create' just kept running. Patch 2 kills and reaps the command
after run's eval if it is still around. Patch 1 stops run() from
cancelling the caller's alarm on its way out, which would let
'lvcreate' run without the lock's limit on shared LVM once 'vgs' ran
before it.

Changes since v1, from Wolfgang's review [1]:

- patch 1: disable the caller's alarm before installing our own ALRM
  handler, so it cannot fire into the new handler right before our
  timeout is armed, which would have been reported as run's timeout
- patch 1: run the flush of the last buffered stderr line in its own
  eval, since the caller's alarm can interrupt it now; its error is
  warned on a timeout and appended otherwise, with a new test

Patch 2 is unchanged. I replied to the review [1] why I kept it a
separate patch.

Link to v1: https://lore.proxmox.com/pve-devel/20261021110126.4024539-1-k.chai@proxmox.com/

[1] https://lore.proxmox.com/pve-devel/cgsng52kirt4qcaehzqlb75kqttxcgjirdgum6ndwp7ym6ag2d@p254xv6gnjv4/

Kefu Chai (2):
  cmd: run: do not cancel the caller's alarm
  cmd: run: kill the command when run dies

 src/PVE/Cmd.pm   | 90 ++++++++++++++++++++++++----------------------
 test/cmd-test.pl | 85 ++++++++++++++++++++++++++++++++++++++++++++-
 2 files changed, 132 insertions(+), 43 deletions(-)

--
2.47.3