python / cpython

The Python programming language
https://www.python.org
Other
62.59k stars 30.03k forks source link

`os.path.splitroot()` splits "Global" and "GLOBALROOT" device paths incorrectly #103593

Open barneygale opened 1 year ago

barneygale commented 1 year ago

Low priority issue, follow-up from #78079.

On Windows, os.path.splitroot() produces incorrect results for all of the following paths:

>>> from ntpath import splitroot
>>> splitroot('//./Global/Global/UNC')
('//./Global', '/', 'Global/UNC')
>>> splitroot('//./Global/Z:/')
('//./Global', '/', 'Z:/')
>>> splitroot('//?/Global/Z:/')
('//?/Global', '/', 'Z:/')
>>> splitroot('//./Global/UNC')
('//./Global', '/', 'UNC')
>>> splitroot('//?/Global/UNC/server/share/dir')
('//?/Global', '/', 'UNC/server/share/dir')
>>> splitroot('//?/GLOBALROOT/ContainerMappedDirectories/93AA221C-4061-460F-8C4E-A2EAC1BF3324')
('//?/GLOBALROOT', '/', 'ContainerMappedDirectories/93AA221C-4061-460F-8C4E-A2EAC1BF3324')
>>> splitroot('//?/GLOBALROOT/Device/LanManRedirector/abc/xyz')
('//?/GLOBALROOT', '/', 'Device/LanManRedirector/abc/xyz')
>>> splitroot('//?/GLOBALROOT/Device/Mup/abc/xyz')
('//?/GLOBALROOT', '/', 'Device/Mup/abc/xyz')
>>> splitroot('//?/GLOBALROOT/Device/WebDavRedirector/abc/xyz')
('//?/GLOBALROOT', '/', 'Device/WebDavRedirector/abc/xyz')
>>> splitroot('//?/GLOBALROOT/DosDevices/UNC/abc/xyz')
('//?/GLOBALROOT', '/', 'DosDevices/UNC/abc/xyz')
>>> splitroot('//?/GLOBALROOT/??/UNC/abc/xyz')
('//?/GLOBALROOT', '/', '??/UNC/abc/xyz')
eryksun commented 1 year ago

The "Global" qualifier would be consumed as part of the "\\.\" or "\\?\" device prefix. Using "Global" ensures that only system device names are found, bypassing local device names created in the current logon session via WinAPI DefineDosDeviceW() or NTAPI NtCreateSymbolicLinkObject(). Repeated "Global" qualifiers don't have to be supported.

"GlobalRoot" can be handled like the "UNC" device, by consuming up to two subsequent components, e.g "\\.\GlobalRoot\Device\HarddiskVolume2". An exception could be made to consume up to two more components if the path is on the multiple UNC device (MUP), e.g. "\\?\GlobalRoot\Device\Mup\server\share". Such a path might me returned by os.readlink() if the symlink is absolute (no relative flag) and directly targets "\Device\Mup" instead of "\??\UNC". Or maybe a script is manipulating NT paths from GetFinalPathNameByHandleW(), for example:

>>> h = CreateFile('//localhost/C$/Windows/py.exe', 0, 0, None, 3, 0, None)
>>> p = GetFinalPathNameByHandle(h, VOLUME_NAME_NT)
>>> p = '\\\\?\\GlobalRoot' + p
>>> p
'\\\\?\\GlobalRoot\\Device\\Mup\\localhost\\C$\\Windows\\py.exe'
>>> h2 = CreateFile(p, 0, 0, None, 3, 0, None)
>>> GetFinalPathNameByHandle(h2, VOLUME_NAME_DOS)
'\\\\?\\UNC\\localhost\\C$\\Windows\\py.exe'