unlinkat/EBUSY/hosed is back (Jan 2023) #17216
Description
Another variant of the "podman rm hoses everything" flake:
not ok 510 podman kube --network
...
$ podman-remote --url unix:/tmp/podman_tmp_X6wS pod rm -t 0 -f test_pod
Error: 2 errors occurred:
* removing container 0e3da7bfb50c6d7fb4f92ea81c6470baa918abf7e06d577eca57c6c8f659200b from pod 0c72f802ee245ebd7058f5f53d38563817d7837b7a0541b15e7b6bcf0ca290ee: removing container 0e3da7bfb50c6d7fb4f92ea81c6470baa918abf7e06d577eca57c6c8f659200b root filesystem: 1 error occurred:
* unlinkat /home/some26533dude/.local/share/containers/storage/overlay/0cf8c86cb0aa2c77d3bc9488ade58ca3cb56bc8183f98dc9090c822f58bc8192/merged: device or resource busy
* removing container 4dabd961b7beefe2d9354d3ca3a6f3e298e4c4fe3fbf554d74938ef6b0115388 from pod 0c72f802ee245ebd7058f5f53d38563817d7837b7a0541b15e7b6bcf0ca290ee: a container that depends on container 4dabd961b7beefe2d9354d3ca3a6f3e298e4c4fe3fbf554d74938ef6b0115388 could not be removed: container state improper
All subsequent tests fail, sometimes with timeout, or with unmarshal errors, or with EBUSY.
I'm like 99% sure that this is the same as #17042 (the "crun: executable not found") flake, because it's happening in the same place and leaves the system hosed in a similar way, but it's easier to merge than to unmerge so I'm filing separately.
Seen in f37 remote rootless.