From ca1344d04a9c1804234417dcfbd868524abc7ce6 Mon Sep 17 00:00:00 2001

From: =?UTF-8?q?Zbigniew=20J=C4=99drzejewski-Szmek?= <zbyszek@in.waw.pl>

Date: Tue, 9 Jan 2024 11:28:04 +0100

Subject: [PATCH] journal: again create user journals for users with high uids

This effectively reverts a change in 115d5145a257c1a27330acf9f063b5f4d910ca4d

'journald: move uid_for_system_journal() to uid-alloc-range.h', which slipped

in an additional check of uid_is_container(uid). The problem is that that change

is not backwards-compatible at all and very hard for users to handle.

There is no common agreement on mappings of high-range uids. Systemd declares

ownership of a large range for container uids in https://systemd.io/UIDS-GIDS/,

but this is only a recent change and various sites allocated those ranges

in a different way, in particular FreeIPA uses (used?) uids from this range

for human users. On big sites with lots of users changing uids is obviously a

hard problem. We generally assume that uids cannot be "freed" and/or changed

and/or reused safely, so we shouldn't demand the same from others.

This is somewhat similar to the situation with SYSTEM_ALLOC_UID_MIN /

SYSTEM_UID_MAX, which we tried to define to a fixed value in our code, causing

huge problems for existing systems with were created with a different

definition and couldn't be easily updated. For that case, we added a

configuration time switch and we now parse /etc/login.defs to actually use the

value that is appropriate for the local system.

Unfortunately, login.defs doesn't have a concept of container allocation ranges

(and we don't have code to parse and use those nonexistent names either), so we

can't tell users to adjust logind.defs to work around the changed definition.

login.defs has SUB_UID_{MIN,MAX}, but those aren't really the same thing,

because they are used to define where the add allocations for subuids, which is

generally a much smaller range. Maybe we should talk with other folks about

the appropriate allocation ranges and define some new settings in login.defs.

But this would require discussion and coordination with other projects first.

Actualy, it seems that this change was needed at all. The code in the container

does not log to the outside journal. It talks to its own journald, which does

journal splitting using its internal logic based on shifted uids. So let's

revert the change to fix user systems.

Fixes https://bugzilla.redhat.com/show_bug.cgi?id=2251843.

---

 src/basic/uid-classification.c | 2 +-

 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/basic/uid-classification.c b/src/basic/uid-classification.c

index e2d2cebc6d..2c8b06c0d3 100644

--- a/src/basic/uid-classification.c

+++ b/src/basic/uid-classification.c

@@ -127,5 +127,5 @@ bool uid_for_system_journal(uid_t uid) {

         /* Returns true if the specified UID shall get its data stored in the system journal. */

-        return uid_is_system(uid) || uid_is_dynamic(uid) || uid == UID_NOBODY || uid_is_container(uid);

+        return uid_is_system(uid) || uid_is_dynamic(uid) || uid == UID_NOBODY;

 }