mirror of
https://github.com/ansible-collections/community.general.git
synced 2025-04-26 12:21:26 -07:00
50ceb8d7d8
open(path) throws an error when called on a directory or UNIX socket, and therefore a check to ensure that the path is absent will always succeed when there is a directory or file located there. This updates the check to use os.access(path, os.F_OK) instead, which instead just checks that the path exists instead of trying to open it as a file, and therefore properly handles directories and sockets. This causes a slight semantic change in how permissions are handled. The existing code will fail to work correctly if the user running the module on the managed host has no read access to the path specified. The new code will work correctly in that situation. Both versions fail if the user can't traverse the parent directory. I've also added a check to the try block to catch OSError. I've seen this call fail with an OSError on rare occasion in the face of odd extended permissions (usually MAC configuration) in cases where it should technically return False. In such cases, the file is functionally inaccessible to the user making the call, so it's essentially not there, but it can't be created by them either. I've documented this, as well as the fact that the bug this change fixes exists, and a rather nasty inconsistency involving symbloic handling that I stumbled across while testing this change. Fixes: #20870 |
||
|---|---|---|
| .. | ||
| cli | ||
| compat | ||
| config | ||
| errors | ||
| executor | ||
| galaxy | ||
| inventory | ||
| module_utils | ||
| modules | ||
| parsing | ||
| playbook | ||
| plugins | ||
| template | ||
| utils | ||
| vars | ||
| __init__.py | ||
| constants.py | ||
| release.py | ||