* Tollef Fog Heen <tfheen@err.no> [2025-03-03 06:20]:
This sounds like a bug in sbuild – it must reset the value of TMP/TMPDIR >>when changing UIDs.
I tend to disagree here. sbuild is not changing to a different user
but to a different UID of the same user.
So resetting TMPDIR would mean that sbuild would not respect any
TMPDIR and I think that would be wrong. Instead I see two options:
1. sbuild sets acls such that subuids have access to the TMPDIR.
I think both options are rather suboptimal and there are more tools
running into the same problem, like mmdebstrap in #1052471. So instead
of patching every tool to work around the specifics of libpam-tmpdir I
would prefer if libpam-tmpdir would learn about subuids.
]] Jochen Sprickerhof
* Tollef Fog Heen <tfheen@err.no> [2025-03-03 06:20]:
This sounds like a bug in sbuild – it must reset the value of TMP/TMPDIR >>>when changing UIDs.
I tend to disagree here. sbuild is not changing to a different user
but to a different UID of the same user.
How does this work with other resources that are linked to that
particular user, whether ephemereal or not? Say, do they share the
systemd --user instance, ssh or gpg agents? What about $HOME, or >/run/user/$UID? Does sbuilt open a new PAM session when switching to a >subuid? (If not, why not?)
Is there an in-depth description of what subuids really are somewhere?
A quick search on the net did not find a design doc or explanation of >tradeoffs around the questions asked above.
So resetting TMPDIR would mean that sbuild would not respect any
TMPDIR and I think that would be wrong. Instead I see two options:
1. sbuild sets acls such that subuids have access to the TMPDIR.
Does it need to share files between different subuids?
Sysop: | Keyop |
---|---|
Location: | Huddersfield, West Yorkshire, UK |
Users: | 482 |
Nodes: | 16 (2 / 14) |
Uptime: | 38:29:03 |
Calls: | 9,566 |
Calls today: | 26 |
Files: | 13,656 |
D/L today: |
2 files (941K bytes) |
Messages: | 6,141,656 |