Informix Root vs Non-Root Installation
What the two installation models actually are, what the non-root model gives up, and why a container running as an unprivileged UID is not the same thing.
Informix® supports two different installation models on UNIX and Linux. The model is chosen when the product is installed, it cannot be changed afterwards, and it determines ownership, authentication, local connection behaviour, and which enterprise features are available at all.
This article is the practical overview. The deeper question — why the conventional engine still carries root-owned SUID binaries, why Enterprise Replication was eventually adapted to non-root operation while HDR, RSS and SDS were not, and what all of that means for OpenShift — is covered in the companion article, Informix Privilege Architecture: Root, Non-Root, HA, and Containers.
How to read this article
Much of what circulates about non-root Informix mixes product documentation with folklore. This article therefore labels the basis for its claims:
| OFFICIAL | Stated directly in IBM® or HCL product documentation. |
| HISTORICAL | Contemporaneous IBM or IIUG conference material. Useful for design intent; not a current product statement. |
| INFERRED | A conclusion drawn from documented behaviour. IBM and HCL have not stated it. |
Where this article infers something, it says so. An inference is never presented as the vendor’s rationale.
1. What IBM means by root-based and non-root installations
OFFICIAL. IBM documentation is explicit that the owner of the
database server depends on the privileges of the user who runs the installation
application. The standard way to install the server is as a superuser with
administrative privileges, and the installation application creates the user
informix. If the installer is run by a non-administrative user
instead, that user becomes the owner of the database server and user
informix is not created.
The key phrase is installation model. This is not a runtime switch, not a configuration parameter, and not a property of the process table. It is decided once, by who runs the installer.
| Area | Root-based installation | Non-root installation |
|---|---|---|
| Installer run by | root | Ordinary OS user |
| Installation owner | User informix is created and is the normal DBSA identity | The installing user is the owner and has DBSA privileges; informix is not created |
Programs in $INFORMIXDIR/bin | A defined set are SUID root and SGID informix | No SUID or SGID programs |
| High-availability clusters (HDR, RSS, SDS) | Supported | Not available |
| Enterprise Replication | Supported | Not available for servers that have different owners |
| ON-Bar | Supported | Not available |
| Auditing through role separation | Supported | Not available |
onperf | Supported | Not available |
| OS authentication | Normal system authentication available | Cannot authenticate users through OS-level system calls |
| Encrypted connections / CLE | Supported | Supported, provided the security-layer plug-in is installed separately |
| Platform | UNIX, Linux, Windows | UNIX, Linux, Mac OS X — not Windows |
| Later conversion | Neither model can be converted or upgraded into the other | |
OFFICIAL. On conversion, the documentation leaves no room: you cannot change the owner of the database server, and you cannot convert or upgrade a non-root installation to a standard, root-based installation, or the reverse.
2. Historical background: Private Server
HISTORICAL. The capability arrived in Informix 11.70 and was discussed at the time under the name Private Server (also “private instance” and “private installation”). It was presented at IIUG 2011 by Jonathan Leffler of IBM, in session C18, Installing and Running IDS 11.70 Without Using Root Privileges, and again at IIUG 2013 in session B10, All About Private Servers — Tips, Tricks and Traps.
The motivation given in that material was not primarily technical elegance. It
was that root privileges are rigorously controlled in many industries, finance
and health care being the examples cited; that some embedded systems do not use
root at all; and that having to add user informix and group
informix was itself an obstacle. The 2013 session adds a fourth
motivation that has aged well: auditors question SUID and SGID programs.
This history matters because it explains the shape of the feature. Non-root installation was not a reduced-price edition or a licensing tier. It was an engineering exercise in removing a specific dependency — and the things that did not survive that exercise are the things that needed the dependency.
3. Installation ownership and permissions
Ownership is the root of every other difference in this article. Both models establish an installation owner at install time; what differs is who that owner is, what the owner is trusted to do, and whether the owner’s group is trusted at all.
4. Standard installation ownership model
OFFICIAL. User informix is required for root-based
installations because it has the unique UID used to manage and maintain database
server instances and databases on the host. User informix is a
member of group informix, and the documentation states that user
informix should be the only member of that group.
OFFICIAL. A defined set of utilities is installed SUID root and
SGID informix:
SUID root, SGID informix:
onaudit onbar_d ondblog onedcu oninit onmode
onshowaudit onsmsync onsnmp onsrvapd ontape snmpdm
SGID informix:
oncheck onedpdu onload onlog onparams
onpload onspaces onstat onunload xtree
The default mode of oninit is documented indirectly but
unambiguously: to let DBSA-group users other than informix run it,
IBM instructs you to log in as root and change its permissions in
$INFORMIXDIR/bin from 6754 to 6755. The leading
6 is the SUID and SGID pair.
HISTORICAL. The 2013 IIUG session describes the consequence in
plain terms: in a regular server some virtual processors run with root
privileges, and the server can use those privileges to create files in a
user’s home directory, to run the commands for a SYSTEM
statement as the invoking user, and to make changes that only root is permitted
to make.
5. Non-root installation ownership model
OFFICIAL. If you run the installation application as a
non-administrative user, that user is the non-root owner of the installation and
has DBSA privileges over the database server. The non-root server runs without
user root or user informix privileges.
HISTORICAL. The 2011 session sets out what the installer does
and, more tellingly, what it does not do: it checks that you want a private
install, creates $INFORMIXDIR owned by the installing user, and
cannot do anything else. Only that user can run the server. There are no
setuid programs and no setgid programs.
The same material draws out a distinction that is easy to miss and has real consequences for permissions:
The group is not trusted
In a conventional installation, group
informixis completely trusted. In a private server the owner is an ordinary user who may belong to an ordinary group such asstaff— and that user should certainly not trust every member ofstaff.
HISTORICAL. This is why chunk permissions change from 660 to
600, why the main shared-memory segment is created with no group or public
access, and why group write permission is revoked on directories and files. The
2013 session recommends checking the result with onsecurity -p,
which performs the startup security checks appropriately for a private server.
6. Authentication differences
OFFICIAL. The non-root server cannot authenticate users based on system calls to the OS level. The DBSA of the installation must create internal users to grant database server access to other users.
HISTORICAL. The reason is mundane and decisive:
/etc/shadow is not readable by non-root users. Verifying a Unix
password requires privilege the private server does not have. The 2011 session
notes that PAM and GSSCSM continue to work as before — those delegate
authentication elsewhere — and that this is precisely why mapped users
become important in a non-root deployment.
OFFICIAL. Informix supports CREATE USER, and
database users can be given operating-system properties — UID, GID and home
directory — which the server then uses when it interacts with the OS on
their behalf, for example for SPL SYSTEM() or
SET EXPLAIN. From 11.70.xC3 the OS administrator can also use
/etc/informix/allowed.surrogates to control which operating-system
users and groups may act as surrogates for mapped users.
For an environment whose identity model is PAM, Active Directory, trusted hosts and conventional Unix accounts, this is an architectural difference rather than a configuration detail. The non-root DBSA owns the user list.
7. Local and shared-memory communication differences
This is the single most common practical failure with non-root installations, and the documentation describing it is genuinely confusing, so it is worth setting out carefully.
OFFICIAL. For root installations, the communication files
required for shared-memory (onipcshm) and stream-pipe
(onipcstr) connections are stored in and accessed from
/INFORMIXTMP — a location a non-root installation does not
have write permission for. The cfd option in the sqlhosts file sets
the directory where the communication files are stored instead:
server_shm onipcshm host service cfd=/path/writable/by/owner
OFFICIAL. Clients check /INFORMIXTMP
first, and only then $INFORMIXDIR/etc. If the client and the
server are in different locations, the communication files are in neither
directory and the connection fails; in that case you must specify the
cfd value in sqlhosts for the connection to succeed.
A contradiction in the documentation worth knowing about
The same IBM page states both that non-root installations “now store communication files in the
$INFORMIXDIR/etcdirectory by default” and that communication files “are not written to the$INFORMIXDIR/etcdirectory if no communication files directory is specified as an option in the sqlhosts file”. Those two sentences cannot both be true. The page opens by saying you must includecfdto use shared-memory and stream-pipe connections at all, which is the reading that matches field experience.The safe operational rule is therefore: on a non-root installation, set
cfdexplicitly and do not rely on a default.
The usual failure mode. A local connection fails while the
database server is running perfectly well, because the client and the server
disagree about where the communication file lives. The cfd value has
to be consistent on both sides — setting it only on the server is not
sufficient.
HISTORICAL. Shared-memory and stream-pipe connections were not available to non-root installations at first. Informix 11.70.xC2 supported TCP only; the 2011 session records that the fix was scheduled for 11.70.xC3 and required Client SDK changes as well as server changes. IBM’s 11.70.xC3 feature material confirms the delivery.
OFFICIAL. One further constraint on onipcstr: the
installation directory must be on a local file system, because a named domain
socket cannot be created on a non-local file system.
8. Feature restrictions in non-root installations
OFFICIAL. If you install the database server without root privileges, you cannot use the following features and tools:
- Enterprise Replication for servers that have different owners
- High-availability clusters
- Auditing through role separation
- The ON-Bar utility for backing up and restoring data
- The
onperfutility to monitor database server performance
OFFICIAL. Encrypted connections and column-level encryption are
not on that list. They are supported, provided the required
security-layer plug-in is installed separately on the host — the GSKit
installer in $INFORMIXDIR/gskit. This is a packaging step, not a
capability loss, and it is frequently misreported.
HISTORICAL. Two further restrictions appear in the conference
material rather than in the installation-owner list. The 2013 session states
plainly that private servers are not supported on Windows, and that the server
cannot create files as other users and cannot necessarily read all files —
which is what breaks SET EXPLAIN to arbitrary paths,
SET DEBUG FILE, LOTOFILE, and
FILETOBLOB/FILETOCLOB reads.
9. Enterprise Replication and the same-owner restriction
HISTORICAL. In 11.70, Enterprise Replication was simply not available to private servers. The 2011 session lists it in the same “Not Supported (Yet)” block as MACH 11, I-Star, ON-Bar with ISM, role separation and SNMP.
OFFICIAL. Informix 12.10 changed that. Enterprise Replication is supported among non-root servers, and the restriction that survives into current documentation is narrower and specific: ER is unavailable “for servers that have different owners”. Same owner, and ER works.
OFFICIAL. The supporting mechanism is trusted-server
configuration that does not depend on files only root can manage.
REMOTE_USERS_CFG specifies an alternative to ~/.rhosts
for listing trusted users; REMOTE_SERVER_CFG specifies an
alternative to /etc/hosts.equiv for listing trusted remote hosts.
OFFICIAL. The same two parameters also enable distributed queries on non-root installations — another capability the 2013 conference material still described as “problematic”, and which the current documentation treats as supported once those parameters are set.
This progression is the most informative thing in the whole subject, and it is taken up in detail in the companion article.
10. The high-availability cluster restriction
OFFICIAL. High-availability clusters are unavailable in a non-root installation. This covers the MACH 11 family — HDR secondary, RS secondary (RSS) and SD secondary (SDS) servers.
HISTORICAL. The 2011 session lists MACH 11 — HDR, SDS, RSS — under the heading Not Supported (Yet). The parenthesis is the implementing engineer’s, not ours. The 2013 session softens the wording to “ER, HDR, RSS, SDS are problematic”.
INFERRED. ER was in that same bucket in 2011 and came out of it in 12.10; HDR, RSS and SDS did not. Because the trusted-server mechanisms that unblocked ER apply equally to cluster connections, remote authentication alone does not appear to explain the surviving restriction. Neither IBM nor HCL has published a statement of what does.
For planning purposes the practical consequence is not in doubt: if the topology needs HDR, RSS or SDS, the installation model must be root-based. Only the explanation is open.
11. ON-Bar, auditing, onperf and other unavailable facilities
OFFICIAL. ON-Bar, role-separated auditing and
onperf are unavailable. Note where the SUID list in section 4 and
this list overlap: onbar_d, onaudit,
onshowaudit, onsnmp and snmpdm are all
SUID root in a conventional installation, and all of the corresponding facilities
are missing from the non-root model. The correlation is not subtle.
HISTORICAL. The 2011 material is more precise about ON-Bar than the current documentation. It lists “ON-Bar with ISM” as unsupported because ISM requires root privileges, and adds that storage managers not needing root “should be OK”. That is a 2011 statement about a then-current bundled storage manager, and it is not a supported configuration claim today — but it does indicate that the ON-Bar restriction was understood as a storage-manager privilege problem rather than a backup-engine one.
HISTORICAL. Other operations the 2011 session records as not
working, or as failing silently, in a private server: processor affinity; the
resident shared-memory option; changing owner or group on files; and switching
UID or GID, which cannot be done without root privileges — so the
SYSTEM statement runs as the server owner rather than as the
invoking user. The 2013 session summarises the general behaviour usefully: some
root-privileged operations fail, but the server keeps running.
ON-Tape, by contrast, works normally, as does SQL generally.
12. Filesystem and IPC differences at a glance
| Item | Root-based | Non-root |
|---|---|---|
| Chunk permissions | 660 — group informix is trusted | 600 — the owner’s group is not trusted |
| Main shared-memory segment | Conventional permissions | No group or public access |
| Communication files | /INFORMIXTMP | Directory named by cfd in sqlhosts |
| Trusted users / hosts | ~/.rhosts, /etc/hosts.equiv | REMOTE_USERS_CFG, REMOTE_SERVER_CFG |
| Password authentication | System authentication available | Internal or mapped users; PAM and GSSCSM still available |
onipcstr | No additional constraint | $INFORMIXDIR must be on a local file system |
| Security check | onsecurity | onsecurity -p |
13. Containers and OpenShift: two different questions
Container security introduces a second axis, and conflating it with the installation model is the most consequential mistake in this whole subject. The installation model and the runtime UID are independent decisions.
| Question | What it determines |
|---|---|
| How was Informix installed? | Whether this is a conventional root-based installation or a genuine Informix non-root installation — and therefore which features exist at all. |
| What UID and capabilities does the container run with? | The runtime security context imposed by Docker, Kubernetes or OpenShift — and therefore what the installed product is permitted to do. |
Taking a conventional $INFORMIXDIR and setting
runAsUser: 1001 does not produce an Informix non-root installation.
It produces a root-based installation running under a UID that may not own its
own files. Equally, a conventional image running as UID 0 is still the
conventional installation model; the UID is a runtime security decision, not an
installation type.
Oninit® recommendation. For an environment that needs HDR, RSS, SDS, Enterprise Replication across differently-owned servers, Connection Manager, conventional enterprise authentication, or ON-Bar-based recovery: install Informix conventionally, with root, at image build time. Then reduce container privileges separately, and only as far as is verified safe.
Do not use the non-root installation model as an SCC workaround
Surrendering HA, ON-Bar and role-separated auditing in order to satisfy an OpenShift
runAsUserpolicy solves the wrong problem. The engineering task is to establish the minimum runtime UID and capabilities a conventional installation actually requires, and then tighten the SCC orsecurityContextaround that verified requirement.
The mechanics — arbitrary project UIDs, allowPrivilegeEscalation
against SUID oninit, SELinux, and persistent volume ownership —
are covered in the
companion
article.
14. Decision guide
| Requirement | Installation model |
|---|---|
| HDR / RSS / SDS / any HA cluster topology | Root-based — no alternative |
| Enterprise Replication between servers with different owners | Root-based |
| Enterprise Replication between servers with the same owner | Either; non-root is supported from 12.10 |
| ON-Bar and storage-manager backup | Root-based |
| Role-separated auditing | Root-based |
onperf | Root-based |
| Enterprise OS authentication using local system accounts | Root-based |
| PAM or SSO authentication | Either — both work under non-root |
| Encrypted connections / column-level encryption | Either — install the security-layer plug-in separately under non-root |
| Windows | Root-based — non-root is UNIX, Linux and Mac OS X only |
| Production OpenShift needing full Informix capability | Root-based install; solve runtime privilege separately |
| Embedded deployment or virtual appliance | Non-root is appropriate |
| Developer-contained database, one per user, no central DBSA | Non-root is appropriate |
| No root access on the host, reduced capability acceptable | Non-root is appropriate |
One planning point deserves emphasis because the decision is irreversible: since neither model can be converted into the other, a non-root installation that later turns out to need HDR requires a fresh root-based installation and a migration, not a permissions change.
15. Common misconceptions
| Claim | Status |
|---|---|
| “Informix should never run as root.” | Misleading. A conventional installation deliberately ships SUID-root utilities, oninit among them, and some virtual processors legitimately hold root privileges. |
“If informix starts the engine, every oninit process must run as informix.” | Incorrect. SUID-root semantics mean the master oninit may legitimately appear as root in ps. |
| “A conventional install can be made non-root by changing ownership.” | Incorrect. Conversion is explicitly unsupported in both directions. |
| “Removing the SUID bit produces a supported non-root server.” | Incorrect. It produces a root-based installation with altered security properties and no support statement. |
“OpenShift runAsNonRoot is equivalent to an Informix non-root installation.” | Incorrect. Different layer, different decision, different consequences. |
| “UID 0 in a container means a privileged container.” | Incorrect. UID and container capabilities are separate controls; anyuid is not privileged. |
| “ER works non-root, so HA should too.” | Not supported by the evidence. The two operate at different architectural layers, and only ER was adapted. |
| “Non-root means you lose encryption.” | Incorrect. Encrypted connections and CLE are supported; the security-layer plug-in is installed separately. |
| “Non-root is available everywhere Informix is.” | Incorrect. It is not supported on Windows. |
16. References
Product documentation
- Installation owner — IBM Informix 14.10 — the standard and non-root installation models, the unsupported-feature list, and the statement that conversion between models is not possible. Also published for 12.10 and mirrored in the HCL Informix documentation.
- Connections to a non-root installation (UNIX, Linux) — IBM Informix 12.10 — the
cfdoption,/INFORMIXTMP, communication-file search order, and distributed queries on non-root installations. - Users and group membership for running utilities — IBM Informix 14.10 — the SUID root and SGID
informixutility lists. - The oninit utility — IBM Informix 14.10 — who may run
oninit, and the documented 6754 to 6755 permission change for DBSA-group users. - REMOTE_SERVER_CFG configuration parameter and REMOTE_USERS_CFG configuration parameter — IBM Informix 12.10.
Historical and supporting material
- Jonathan Leffler, IBM. Installing and Running IDS 11.70 Without Using Root Privileges. IIUG 2011 conference, session C18, 18 May 2011. The primary contemporaneous account of the Private Server design, including the “Not Supported (Yet)” feature list and the behaviour of the installer.
- Jonathan Leffler, IBM. All About Private Servers — Tips, Tricks and Traps. IIUG 2013 conference, session B10. Regular versus private server comparison, permissions guidance, and user management.
- Rakesh Naik, IBM. Test Drive some of the NEW Informix 11.70 features. IIUG 2012 conference, session H04. Records 11.70.xC3 support for shared-memory and stream-pipe connections on non-root installations, and
/etc/informix/allowed.surrogates. - IIUG 2016 Informix 12.10 certification preparation material — Enterprise Replication among non-root servers, and trusted-server configuration parameters.
Oninit® technical interpretation: this article separates the Informix product installation model from container runtime privilege policy, so that enterprise database capability is not sacrificed to satisfy an orchestration security constraint that a different control already addresses. Where the reasoning behind a documented restriction is not published, that is stated rather than filled in.