More exec --preserve-fds flakes: boltdb, and bad file descriptor

[edsantiago]

Two new kinds of flakes in CI, all in the 'podman exec preserve fds sanity check' test:

podman [options] exec --preserve-fds 1 test1 ls
...
time="2020-04-08T14:45:15-04:00" level=error msg="unable to close socket: \"close unixpacket @->/tmp/podman_test358810366/socket/638a99e7f003410c5f5e27a943c9cbfbf76dd9a4fce6adc5521aa38a67823b8f/attach: bad file descriptor\""
Error: read unixpacket @->/tmp/podman_test358810366/socket/638a99e7f003410c5f5e27a943c9cbfbf76dd9a4fce6adc5521aa38a67823b8f/attach: read: bad file descriptor

Source: #5767, 1 instance

Next error:

podman [options] exec --preserve-fds 1 test1 ls
...
2020/04/08 14:45:18 bolt.Close(): funlock error: bad file descriptor
time="2020-04-08T14:45:18-04:00" level=error msg="failed to close libpod db: \"db file close: close /tmp/podman_test054218976/crio/libpod/bolt_state.db: bad file descriptor\""
Error: write /tmp/podman_test054218976/crio/libpod/bolt_state.db: bad file descriptor

Seen this one in #5671 instance 1 and instance 2

Also just seeing it in #5767 instance 1, instance 2

There is clearly something still broken in --preserve-fds

[edsantiago]

I'm seeing a new error that looks a lot like this, except it's in run instead of exec (but with --preserve-fds):

Error: error attaching to container ff1e67ec16aa697a4c217508b579af24dd8bd169bbcd068cf11b402c592f51a9: read unixpacket @->/tmp/podman_test136344526/socket/ff1e67ec16aa697a4c217508b579af24dd8bd169bbcd068cf11b402c592f51a9/attach: read: bad file descriptor

Log: https://api.cirrus-ci.com/v1/task/6481256420474880/logs/integration_test.log#t--podman-run---preserve-fds