• v1.36.1 84ddd2d703

    LDAP WebGUI v1.36.1
    All checks were successful
    Tests / Scope (push) Successful in 1s
    Tests / pytest (newest CPython, warning only) (push) Has been skipped
    Tests / shellcheck (install.sh, update.sh, bootstrap-ldap.sh, webupdate.sh, get_ldap-webgui.sh, setup.sh, config-backup.sh, certtool.sh, accesstool.sh) (push) Successful in 34s
    Tests / pytest (Python 3.13) (push) Successful in 22m34s
    Tests / pytest (Python 3.14) (push) Successful in 23m59s
    Tests / Publish release (push) Successful in 3s
    Stable

    Lyra released this 2026-11-13 21:10:39 +01:00 | 0 commits to main since this release

    Fixed

    • A web update showed only some of the manual steps the update
      counts: the ACL rule for the TOTP secrets in description (1.25.4,
      docs/upgrade-1.25.md) and the move of nginx's certificate out of
      /etc/ssl/certs/ (1.27.0, docs/upgrade-1.27.md) appeared in the
      CLI summary only. Both now go to the status file as well, and
      Settings → Update shows each in its own box with the guide link —
      the certificate step with its --vhost-only command.

    Update without git access: download the source tar.gz below, extract it, then run the update.sh it contains:
    sudo bash <extracted-dir>/update.sh --archive <tarball>

    Downloads
  • v1.36.0 a0b81bcdf3

    LDAP WebGUI v1.36.0
    All checks were successful
    Tests / Scope (push) Successful in 1s
    Tests / pytest (newest CPython, warning only) (push) Has been skipped
    Tests / shellcheck (install.sh, update.sh, bootstrap-ldap.sh, webupdate.sh, get_ldap-webgui.sh, setup.sh, config-backup.sh, certtool.sh, accesstool.sh) (push) Successful in 30s
    Tests / pytest (Python 3.13) (push) Successful in 22m19s
    Tests / pytest (Python 3.14) (push) Successful in 24m2s
    Tests / Publish release (push) Successful in 3s
    Stable

    Lyra released this 2026-11-12 21:02:53 +01:00 | 1 commits to main since this release

    Upgrade notes

    • One manual step, needed only for Machines → Register: the host
      register lives in ou=hosts and needs one ACL rule for the bind
      account. Neither the update nor the web update touches slapd; without
      the step the machines pages work as before and only Register is
      disabled, pointing to the guide. The update counts it under "One
      manual step is left", and the web update shows it in its status box.
      Procedure, check and way back:
      docs/upgrade-1.36.md. Directories built by bootstrap-ldap.sh from
      this release on carry the OU and the rule already.
    • ldap.all_hosts_group in config.yaml is ignored now (see
      Removed); remove the line at your convenience.
    • The nginx vhost is unchanged in this release.

    Added

    • Host register (Machines → Register): one device entry per host
      below ldap.hosts_base (default ou=hosts,<base DN>), the short name
      exactly as hostname prints it, case-sensitive, plus an optional
      description. A registered host's spelling is the reference: sudoHost
      values spelled differently are flagged against it, on the machines
      pages and on the sudo form. Invalid entries in the OU are listed and
      otherwise ignored.
    • Writes from the host page: register a host, edit its description,
      delete it (refused while a sudo rule still names it, exact-case
      confirmation), add a registered host to an existing sudo rule, remove
      a host from a rule (also an unregistered one — that is how a typo
      leaves a rule), and Create a sudo rule for this host with sudoHost
      prefilled. Each sudo-rule edit is one value, never a replace of the
      rule's hosts; security log HOST_CREATED / HOST_UPDATED / HOST_DELETED
      and SUDO_ROLE_UPDATED with host_added= / host_removed=.
    • New group names may carry upper case ([A-Za-z_][A-Za-z0-9._-],
      at most 32), with the hint that AllowGroups on the host takes the
      name exactly as spelled. User and service-account names stay lower
      case.
    • Host names up to 63 characters (one RFC 1123 label; was 58).
    • ldap.hosts_base (config.yaml only) and a read-only Hosts Base row
      on the Settings page.

    Changed

    • The host page shows sudo only: every sudo rule matching the host,
      resolved to people with the path per person (via %group, via user). Login is the host's AllowGroups in sshd_config and is not
      modelled; docs/machines.md explains how to keep login groups and
      sudo groups apart.
    • The host list comes from the register and the sudoHost values; the
      Machines card on a group page shows sudo only.
    • bootstrap-ldap.sh creates ou=hosts and its ACL rule {8} (eleven
      rules); a resumed run on a tree with the previous ten-rule set
      upgrades it. docs/ldap-acl-setup.md and docs/initial-admin.ldif
      follow.

    Removed

    • The per-host login view: the AllowGroups line, the login columns
      and the group card's "Opens login on" / "Logs in everywhere" lines.
    • login-<host> groups as a host source — such a group is an ordinary
      posixGroup and names no host any more.
    • ldap.all_hosts_group — ignored, with one warning in the log while
      config.yaml still sets it; the Settings page's All-Hosts Group
      row.

    Fixed

    • A concurrent description change on a group or user is reported
      ("changed by someone else in the meantime — nothing was saved")
      instead of a raw LDAP error.
    • Creating a group that already exists in another case
      (Web vs. web, or a race past the check) answers with the stored
      spelling, or a plain "already exists", instead of the LDAP result
      text.
    • The group-name hint on the create form promised 64 characters and
      mixed case while the server allowed 32, lower case; it now states the
      actual rule.
    • docs/initial-admin.ldif pointed to ACL rule {7} for
      ou=service-accounts; it is {9} with the new rule.
    • The host page and docs/machines.md said sss_cache -E makes a sudo
      change immediate on the host. It does for removals and narrowed rules,
      not for a rule that newly applies to a user or host — that one waits
      for SSSD's smart refresh or needs an emptied SSSD cache (the steps are
      in docs/machines.md); sss_cache is in sssd-tools.

    Update without git access: download the source tar.gz below, extract it, then run the update.sh it contains:
    sudo bash <extracted-dir>/update.sh --archive <tarball>

    Downloads
  • v1.35.0 22cc96d1c0

    LDAP WebGUI v1.35.0
    All checks were successful
    Tests / Scope (push) Successful in 1s
    Tests / pytest (newest CPython, warning only) (push) Has been skipped
    Tests / shellcheck (install.sh, update.sh, bootstrap-ldap.sh, webupdate.sh, get_ldap-webgui.sh, setup.sh, config-backup.sh, certtool.sh, accesstool.sh) (push) Successful in 30s
    Tests / pytest (Python 3.13) (push) Successful in 22m16s
    Tests / pytest (Python 3.14) (push) Successful in 23m51s
    Tests / Publish release (push) Successful in 3s
    Stable

    Lyra released this 2026-11-12 00:20:18 +01:00 | 2 commits to main since this release

    Upgrade notes

    • One manual step for a directory with more than 500 entries in an
      area
      (users, groups or sudo rules) whose bind account has no
      olcLimits line yet. slapd counts its size limit over the whole paged
      search, so the WebGUI's bind account needs
      olcLimits: dn.exact="<bind DN>" size.prtotal=unlimited on the data
      database. On the directory host an update at a terminal shows the line,
      asks and adds it; the web update never writes to slapd and names
      sudo ldap-webgui-update --ldap-limits instead. With slapd on another
      host, the update to this release shows the ready ldapmodify once.
      An existing size=unlimited line (recommended since 1.32) stays valid
      — nothing to do; keep it if you may go back below this release.
      Smaller directories need nothing. docs/ldap-acl-setup.md → Size
      limit for the bind account
      .
    • The nginx vhost is unchanged in this release.

    Added

    • Areas of any size: every directory read — the user, group and
      sudo-rule lists, the Machines page, a group's primary members, the
      search for the next free uidNumber/gidNumber — runs as a paged search
      (pages of 250) and returns the whole area; filters always search all of
      it. Where slapd still cuts the result, the page fails with the exact
      olcLimits line for the configured bind DN instead of showing a
      partial list, and automatic IDs are refused, as before.
    • update.sh --ldap-limits and a check after every update: is this
      host the directory host (slapd active here, a database for
      ldap.base_dn, ldap.host one of this host's addresses)? Then the
      bind account's line is in place, or offered. A broader olcLimits
      value in front of it (users, *, a subtree …) wins over it in
      slapd — reported with the list in the right order, never reordered
      automatically. A failed check is a warning, never a failed update.
    • bootstrap-ldap.sh sets the line on the directory it builds; an
      existing value for the bind account is left as it is.

    Changed

    • The CSV export reads the users once instead of once per 200 rows.
    • docs/ldap-acl-setup.md: size.prtotal=unlimited is the recommended
      line; the pitfalls size.hard alone, size.pr, root over ldapi:///
      (not exempt) and the order of olcLimits values.

    Fixed

    • Detail pages no longer report a failed read as "not found". A user,
      group or sudo rule whose read hits a directory limit (or fails
      otherwise) now shows "Could not read …" with the cause; "not found"
      only when the entry really does not exist. Same for the sudo toggle on
      the user page.

    Update without git access: download the source tar.gz below, extract it, then run the update.sh it contains:
    sudo bash <extracted-dir>/update.sh --archive <tarball>

    Downloads
  • v1.34.0 992b856771

    LDAP WebGUI v1.34.0
    All checks were successful
    Tests / pytest (newest CPython, warning only) (push) Has been skipped
    Tests / Scope (push) Successful in 2s
    Tests / shellcheck (install.sh, update.sh, bootstrap-ldap.sh, webupdate.sh, get_ldap-webgui.sh, setup.sh, config-backup.sh, certtool.sh, accesstool.sh) (push) Successful in 33s
    Tests / pytest (Python 3.13) (push) Successful in 22m2s
    Tests / pytest (Python 3.14) (push) Successful in 23m30s
    Tests / Publish release (push) Successful in 4s
    Stable

    Lyra released this 2026-11-09 19:43:42 +01:00 | 3 commits to main since this release

    Upgrade notes

    • The nginx vhost changes, and the update rolls it out by itself — no
      manual step.
      What nginx does differently afterwards:
      • TLS 1.3 only. Browsers, curl and Zabbix's HTTP agent with
        OpenSSL 1.1.1 or newer connect as before; a poller on an older TLS
        library no longer does (docs/monitoring.md → Troubleshooting).
      • Stricter headers: a Content-Security-Policy without
        'unsafe-inline' and with object-src 'none';
        Cross-Origin-Opener-Policy and Cross-Origin-Resource-Policy
        same-origin; a wider Permissions-Policy; X-XSS-Protection is
        no longer sent.
      • Login rate limit in nginx: POST /login and POST /2fa/verify
        at 10 a minute per client address, a burst of 20 — over it nginx
        answers 429 with its own page. Opening the login page is not counted.
      • The app gets Host as nginx's normalised name, and
        X-Forwarded-For holds the client address only.
    • Downgrade to 1.33.0: its update renders 1.33's own vhost. If that
      rollout fails or is skipped (a lock still held, …), the 1.33 web UI
      runs under the new CSP, which blocks its inline scripts — it stays
      broken until sudo ldap-webgui-update --vhost-only.

    Added

    • HSTS, set on the command line only: install.sh --hsts 300|86400|31536000 / --no-hsts, or HSTS and HSTS_MAX_AGE in
      /etc/ldap-webgui/vhost.env plus --vhost-only. Never
      includeSubDomains or preload. Refused while the certificate does
      not verify
      for every server name against this host's trust store
      (self-signed refused even when trusted); an update then renders the
      vhost without HSTS and names the manual step. Settings → Certificates
      shows the state read-only. docs/upgrade-1.33.md, section 9: the
      lockout cases, the way back (browsers keep the policy until max-age
      runs out), the IP-address emergency route.
    • While HSTS is on, a new nginx certificate that does not verify is
      refused
      — by the certificate runner, and for an upload with its key
      already in Settings → Certificates, before the password is asked for.

    Changed

    • The web UI has nothing inline any more — no inline scripts, event
      handlers, <style> blocks or style= attributes — so the CSP needs
      neither 'unsafe-inline' nor hashes. Markdown tables (docs, release
      notes) align with align=.

    Update without git access: download the source tar.gz below, extract it, then run the update.sh it contains:
    sudo bash <extracted-dir>/update.sh --archive <tarball>

    Downloads
  • v1.33.0 b48ad7042d

    LDAP WebGUI v1.33.0
    All checks were successful
    Tests / Scope (push) Successful in 1s
    Tests / pytest (newest CPython, warning only) (push) Has been skipped
    Tests / shellcheck (install.sh, update.sh, bootstrap-ldap.sh, webupdate.sh, get_ldap-webgui.sh, setup.sh, config-backup.sh, certtool.sh, accesstool.sh) (push) Successful in 1m22s
    Tests / pytest (Python 3.13) (push) Successful in 21m9s
    Tests / pytest (Python 3.14) (push) Successful in 22m30s
    Tests / Publish release (push) Successful in 3s
    Stable

    Lyra released this 2026-11-08 19:02:50 +01:00 | 4 commits to main since this release

    Upgrade notes

    • The nginx vhost is now replaced by every update.
      /etc/nginx/sites-available/ldap-webgui is the application's file:
      rendered from the shipped template and the new host settings file
      /etc/ldap-webgui/vhost.env, checked, installed, confirmed served,
      and put back if anything fails. Local edits to the vhost are lost —
      host names, certificate paths and listen addresses belong in
      vhost.env, who may connect stays Settings → Access.
    • The first update migrates: it reads those values from the
      installed vhost and writes vhost.env; the previous file stays as
      /etc/nginx/sites-available/ldap-webgui.pre-<version>.
    • A vhost edited by hand is not touched — no rollout, a manual step
      that lists the differing lines, repeated by every update — until you
      carry over what you need and run
      sudo ldap-webgui-update --vhost-only --replace-vhost.
    • Git mode (sudo ldap-webgui-update) and
      ldap-webgui-update --archive need a second step:
      they run the
      previous release's update.sh, which does not know the vhost yet.
      Afterwards run sudo ldap-webgui-update --vhost-only (after
      --archive: sudo bash <unpacked release>/update.sh --vhost-only).
      The web update, --from-release and an unpacked release's own
      update.sh do it in one run.
    • The steps, how to check, how to back out and the downgrade:
      docs/upgrade-1.33.md (also under Settings → Documentation).

    Added

    • /etc/ldap-webgui/vhost.env (root:root 0640) — the vhost's host
      settings: SERVER_NAME, TLS_CERT, TLS_KEY, LISTEN. Parsed,
      never sourced; an invalid file or an unknown key skips the rollout
      with a manual step and never fails the update.
    • update.sh --vhost-only re-renders the vhost from vhost.env and
      the installed release — checked, nginx -t, reloaded, confirmed, the
      previous one put back on failure; no release change, no restart of the
      WebGUI. The way to apply an edited vhost.env.
    • A change of the listen addresses restarts nginx (the update,
      --vhost-only, install.sh): a reload cannot move nginx between
      "every address" and given ones — it keeps the old sockets and still
      reports success. Same addresses: a reload, as before.
    • update.sh --replace-vhost replaces a hand-edited vhost from
      before 1.33.0 (the previous file is kept).
    • install.sh --server-name "NAMES" and --listen "ADDRS" set
      SERVER_NAME and LISTEN at install time. With LISTEN set, nginx
      listens on those addresses plus 127.0.0.1 (the health checks use
      loopback).
    • /health answers with an X-LDAP-WebGUI-Vhost header — the hash of
      the rendered vhost, which the update and the installer use to confirm
      nginx serves the new one.

    Changed

    • The security headers reach the static files. The locations for
      /static/ and /health set headers of their own, and nginx then drops
      every header set at server level: CSP, X-Frame-Options,
      Referrer-Policy and Permissions-Policy were missing on stylesheets,
      scripts and images. Every such location now carries the full set.
    • /robots.txt sends one Content-Type instead of two.
    • The vhost template's comments are English, its header says it is
      overwritten by every update, the cipher list is labelled Mozilla
      intermediate; /health is proxied with Connection "" like the rest.
    • A re-run of install.sh replaces a differing vhost — an older
      release's or one edited by hand — after keeping it as
      ldap-webgui.pre-install-<timestamp>; host names, certificate paths
      and listen addresses are carried over into vhost.env. Up to 1.32 it
      kept any vhost that included the access list.
    • install.sh and the update take the access list lock and then the
      certificate lock before touching nginx, so an access list change or a
      certificate renewal from the web UI never runs into a vhost change.
    • The web update's One manual step is left box shows why the vhost
      was not replaced, the lines a hand edit added, and the command the
      update itself prints for that case (--replace-vhost, the ln -sfn
      line, or the lines that put the previous vhost back).
    • install.sh reports a vhost nginx does not serve as a failed check
      and ends INCOMPLETE (exit 1) instead of "All checks passed".
    • install.sh refuses a vhost.env that is not root's alone (a
      symlink, another owner, group- or world-writable) before installing
      anything — the update skips the rollout for it.
    • The vhost's structure check also refuses return, rewrite and if
      (they act before the access list) apart from the template's redirect
      and robots.txt answers, and error_page with = or a named target.

    Removed

    • The update's vhost drift warning ("The installed nginx vhost differs
      from the one this release ships"
      ) — the update replaces the vhost
      instead.
    • The NGINX_VHOST_FILE environment override of update.sh; the vhost
      is always /etc/nginx/sites-available/ldap-webgui.

    Update without git access: download the source tar.gz below, extract it, then run the update.sh it contains:
    sudo bash <extracted-dir>/update.sh --archive <tarball>

    Downloads
  • v1.32.2 bf8485300f

    LDAP WebGUI v1.32.2
    All checks were successful
    Tests / pytest (newest CPython, warning only) (push) Has been skipped
    Tests / shellcheck (install.sh, update.sh, bootstrap-ldap.sh, webupdate.sh, get_ldap-webgui.sh, setup.sh, config-backup.sh, certtool.sh, accesstool.sh) (push) Successful in 25s
    Tests / Scope (push) Successful in 1s
    Tests / pytest (Python 3.14) (push) Successful in 22m21s
    Tests / pytest (Python 3.13) (push) Successful in 23m54s
    Tests / Publish release (push) Successful in 3s
    Stable

    Lyra released this 2026-11-08 01:42:40 +01:00 | 5 commits to main since this release

    Added

    • docs/firewall-ufw.md — a manual guide to a host firewall with
      ufw: which ports to open for whom (SSH and the WebGUI from the admin
      network, the monitoring host, LDAPS for the sssd clients, 389 only for
      StartTLS, port 80 redirect-only or closed). The scripts never touch the
      firewall. In the documentation viewer and linked from the README's
      install section.

    Changed

    • Supported Python is 3.13+ (Debian 13, the only target platform).
      CI and the release gate run the test suite on Python 3.13 and 3.14;
      3.11 is no longer tested.

    Fixed

    • A fresh install no longer checks the example configuration's LDAP
      host.
      install.sh enables and starts the status refresh timer, and a
      timer started long after boot fires at once — before anyone started
      the WebGUI, while config.yaml was still the example. The first line
      of app.log was then certificate status: LDAP server
      (ldap.example.com:636 (LDAPS)) unreadable — invalid server address
      .
      The refresh now runs only while the WebGUI runs
      (ConditionPathIsDirectory=/run/ldap-webgui, the WebGUI's runtime
      directory); otherwise the run is skipped, and the timer keeps its
      interval.
    • The setup's hint for a missing curl (and python3) reads
      apt-get update && apt-get install ….
      A plain apt-get install
      fails on a host whose package lists are stale.

    Upgrade notes

    • No vhost change, no new Python pin, nothing to do by hand. The update
      reinstalls the status refresh service unit with the new condition.

    Update without git access: download the source tar.gz below, extract it, then run the update.sh it contains:
    sudo bash <extracted-dir>/update.sh --archive <tarball>

    Downloads
  • v1.32.1 776f372834

    LDAP WebGUI v1.32.1
    All checks were successful
    Tests / pytest (newest CPython, warning only) (push) Has been skipped
    Tests / Scope (push) Successful in 3s
    Tests / shellcheck (install.sh, update.sh, bootstrap-ldap.sh, webupdate.sh, get_ldap-webgui.sh, setup.sh, config-backup.sh, certtool.sh, accesstool.sh) (push) Successful in 25s
    Tests / pytest (Python 3.11) (push) Successful in 19m38s
    Tests / pytest (Python 3.13) (push) Successful in 20m42s
    Tests / Publish release (push) Successful in 4s
    Stable

    Lyra released this 2026-11-07 09:44:05 +01:00 | 6 commits to main since this release

    Security

    • A password re-check that binds on another entry ends the session.
      Change-password, 2FA disable/replace and the admin re-checks confirm
      the current password by a bind, but the password write then went to
      the DN the login had bound with. Had another tool moved the account
      by modrdn after the login and put a different entry at the old DN,
      the new password landed on that entry. The re-check now compares the
      DN it bound with against the session's (normalised: case and spaces
      around separators aside); if they differ, the session ends, nothing
      is written, and REAUTH_IDENTITY_CHANGED is logged — without
      counting toward the re-authentication limit. Sessions from before
      1.32.1 carry no DN and are not compared.

    • A uid or cn in the path addresses exactly its entry. 1.32.0
      stopped the path guards from rewriting a name (/users/a%2Cb had
      shown and saved ab, a 65-character uid its 64-character prefix) by
      refusing every name they would have changed with 400. The name is now
      used as it came: the LDAP layer looks it up escaped and answers only
      for the entry that carries exactly that name (case aside), so a,b
      opens and edits a,b, never ab, through every per-entry route
      (edit, delete, password reset, SSH keys, sudo toggle, group members,
      sudo rules). A name no entry can carry (control characters, over 256
      characters) is still a 400.

    • A name the directory folds onto another entry no longer reaches it.
      slapd matches uid and cn ignoring leading, trailing and repeated
      spaces and compatibility characters, so POST /users/%20alice/reset-password
      (or alice%20) set alice's password, logged uid= alice, and for
      a b left the sessions of a b alive. Every lookup of an existing
      account, group or sudo rule now answers only for the entry whose own
      name is the one asked for (ASCII case aside — the Kelvin sign K no
      longer counts as k), the password reset has no fallback to a
      constructed DN, and group membership checks compare memberUid
      values exactly (the server folds spaces there too). The password
      reset, user delete and sudo grant/revoke log the entry's own uid, and
      the reset ends the sessions keyed on it.

    • A trailing newline no longer passes the name validators. They used
      re.match with $, which also matches before a final \n — abc\n
      was a valid group name. Every anchored validator now uses fullmatch
      (uid, cn, group cn, email, the login username, settings, the docs
      viewer, the update/access/certificate runner inputs, the monitoring
      token hash).

    • Control characters in sudo rules are refused. A NUL, ESC, DEL or
      any other control character inside a sudoUser, sudoHost,
      sudoCommand, sudoRunAsUser, sudoRunAsGroup or sudoOption value
      was stored — the textarea is split at line breaks, and NUL, ESC or DEL
      are none — and reached sssd and sudo as part of the rule. Each value
      now has a length bound (sudoCommand 4096, sudoHost 255, sudoOption
      1024, the others 256) and each attribute a count (128 values,
      sudoOption 64). The checks run in the LDAP layer too, not only in the
      form.

    • description is bounded on users, service accounts, groups and sudo
      rules
      : at most 1024 characters per value (the schema's bound),
      control characters other than tab and line breaks refused. A value
      starting with the prefix the two-factor configuration is stored under
      is now refused on create too, not only on edit.

    • A refused login name is cut in the log. The name that failed the
      username check went into app.log whole, before the login limiter
      counted the attempt — one request could write up to the body limit.
      It is now cut at 64 characters with control characters masked. An
      X-Real-IP that is not an IP address is ignored in favour of the
      peer address, so it cannot carry a long value or a forged | key=…
      into the security log either.

    • Request bodies are limited to 1 MiB. Werkzeug applies its form
      memory limit only to multipart forms, so every other form field
      reached the application at nginx's 10 MB. A larger request now gets a
      413 page ("too large — nothing was saved"); an htmx request is sent
      back to its page with that message. The rest of a refused body (up to
      nginx's 10 MB) is read and discarded before the answer, so nginx gets
      the 413 instead of a reset connection and a 502; a larger or chunked
      body is not read and the connection is closed.

    • A fresh install no longer downgrades Pillow to 11.1.0. Since
      v0.8.1 install.sh installed pyotp, qrcode, Pillow==11.1.0 and
      ruamel.yaml a second time with its own versions after
      requirements.txt, which pulled Pillow back from the pinned 13.0.0 to
      a release with many published advisories — and the dependency checks,
      which read requirements.txt, could not see it. The installer now
      only checks that these packages import; requirements.txt is the only
      place a version is named, and a test keeps every script free of
      package versions. Operators: a host installed with install.sh
      since v0.8.1 and not updated since runs Pillow 11.1.0 — check with
      /opt/ldap-webgui/venv/bin/pip show pillow; an update (Settings →
      Update, or sudo ldap-webgui-update) installs 13.0.0.

    • New passwords have an upper bound and no NUL. At most 511 bytes
      (UTF-8) — the longest answer Linux-PAM passes on, so longer ones
      worked in the WebGUI but at no login prompt — and no NUL character,
      for the own change, a reset, a create and the CSV import. The login
      takes any length but refuses a NUL before the bind.

    Fixed

    • Entries with an unusual uid or cn open again. A uid such as (x),
      #lead, fz inner, one with umlauts or other non-ASCII characters,
      or one longer than 64 characters — written by another tool — was
      listed but could not be opened: before 1.32.0 its link answered
      "User not found" for a mangled name, 1.32.0 refused it with 400. The
      detail, edit and delete pages and every route behind them reach it
      again; new names keep the strict charset.

    • Line breaks and malformed values in single-line account fields are
      refused.
      User and service-account create and the user edit stored
      CR/LF in givenName, sn, mail, title, telephoneNumber,
      homeDirectory and loginShell, any string as loginShell (10 000
      characters, no path), and any length as homeDirectory. The forms and
      the LDAP layer now require one line, loginShell an absolute path
      without whitespace (max 255), homeDirectory an absolute path (max
      1024). A value already stored that the form sends back unchanged does
      not block an edit.

    • gidNumber on accounts is range-checked. Create and edit stored
      -1, 2147483648 or 10^20; a non-numeric value became 1000 on
      create and was ignored on edit. Both now answer "gidNumber must be a
      number between 1 and 65534", and an empty field on create still means
      1000. 0 (root's group) is refused as a new value — see Upgrade notes.

    • A non-ASCII sudoHost/sudoCommand/sudoOption value is refused
      by the form
      with the field's name and the reason (IA5String), instead
      of the raw ldap3 "invalid per syntax" message. sudoUser,
      sudoRunAsUser and sudoRunAsGroup are DirectoryString in the sudo
      schema and take any character but a control character, so a name with
      an umlaut can be referred to.

    • Removing many group members at once answers with one message per
      outcome.
      Selected values that are not members are reported together
      and cost no LDAP write (1000 of them made 1000 "check LDAP logs"
      flashes and 1000 modifies); failed removals are summarised too, and
      long name lists are cut after ten.

    • A gidNumber of thousands of digits no longer causes a 500. The
      user create and edit forms passed it to int(), which raises past
      4300 digits; a sudoUser of # plus such a number (or #²) did the
      same on the Machines pages. IDs are now read as at most ten ASCII
      digits and refused otherwise.

    • The CSV preview checks what the import will check. A row with a
      line break in title, an over-long name, a homeDirectory containing
      a newline or a loginShell with a tab was green in the preview and
      failed at commit, after the other rows had been written. The preview
      now applies the same account rules as the import, for new and for
      updated accounts. Control characters, for every single-line field,
      include C1 characters (U+0080–U+009F) and U+2028/U+2029.

    • Automatic ID assignment names the size limit. When the scan for
      the next free uidNumber/gidNumber is cut by the directory's size or
      time limit, the create (and the CSV import) now says so and points at
      olcLimits in docs/ldap-acl-setup.md, instead of "check the read
      permissions and the ID range".

    • An admin that is no posixAccount can sign in and change their
      password again.
      The exact lookup earlier in this release only finds
      posixAccounts, so an inetOrgPerson admin listed as member of a
      groupOfNames admin group was refused (or read-only) and
      /change-password answered "not found". The login now keeps the DN
      its bind succeeded with in the session; the group check and the own
      password change use it. Sessions from before the update keep working.
      A password reset of another account still resolves exactly.

    • A CSV export re-imported as an upsert no longer fails on old
      values.
      A title over 128 or a telephoneNumber over 64 characters
      written by another tool failed the preview and the commit although
      nothing would have been written. Update rows are now checked on what
      differs from the stored entry, as the edit form does; new accounts
      are still checked in full.

    • A failed account-type lookup at login is logged as such. It is
      still refused, but as LOGIN_FAILURE | reason=account_type_unreadable
      instead of reason=service_account.

    • A user or group edit request without a field no longer clears it.
      A POST that did not carry mail, title, telephoneNumber or
      description removed those attributes and reported success; only a
      field sent empty clears its attribute now.

    • A CSV file with a cell over 128 KiB no longer causes a 500. The
      csv module's error is answered with "The file could not be read as
      CSV", without quoting the file.

    • Group gidNumber and sudoOrder are read as plain decimals. Both
      went through int(), which takes 1_000, +5 and Arabic-Indic
      digits. The gid range is also checked by the LDAP layer's group
      create, not only by the form.

    • A sudo rule with a sudoOrder above 999 can be edited again. Rules
      written by cvtsudoers or another tool often carry one; the form
      pre-filled it and the save refused it. The range applies to a new or
      changed value only, and so do the new value rules above — a rule
      with a longer command keeps saving as long as that command is
      unchanged.

    • An account with an SSH key another tool wrote can be edited
      again.
      A stored key with command= or no-pty failed every save of
      the key list, because the whole set was checked; keys the account
      already holds are no longer checked again. A new key still accepts
      from= only.

    • A CSV import refused as a whole is logged with its rows as failed.
      An import stopped before the first write (the uidNumber scan at the
      size limit, no free uidNumber) logged CSV_IMPORT_COMMIT with
      failed=0; the rows that were not written now count as failed=.

    Upgrade notes

    • Pillow: a host never updated since its install.sh run has
      Pillow 11.1.0 (see Security); this update replaces it.
    • No vhost change. nginx keeps client_max_body_size 10m; the
      application refuses anything over 1 MiB itself (the largest legitimate
      request is the CSV import, 512 KiB).
    • New account and group names — the create forms, the CSV import's
      new rows, service accounts, manage.py create-admin — must be at most
      32 characters of lower-case letters, digits, ., _ and -,
      starting with a letter or _. That is what useradd/groupadd
      accept by default (NAME_REGEX), utmp's 32-byte ut_user is where
      longer names get cut in who/last, and an @ is read as
      user@domain by sssd. Existing accounts and groups are untouched:
      they sign in, open, are edited, added as members and named in sudo
      rules as before, upper case and @ included.
    • gidNumber 0 (root's group) can no longer be given to an account
      as its primary group — on create, edit or in the CSV import. An
      account that already has it stays editable as long as the gidNumber
      is left as it is. Groups with gidNumber 0 are not affected.
    • New passwords are limited to 511 bytes (UTF-8) and must not
      contain a NUL. Existing passwords are untouched and still sign in,
      whatever their length.

    Update without git access: download the source tar.gz below, extract it, then run the update.sh it contains:
    sudo bash <extracted-dir>/update.sh --archive <tarball>

    Downloads
  • v1.32.0 af40f71e0f

    LDAP WebGUI v1.32.0
    All checks were successful
    Tests / Scope (push) Successful in 1s
    Tests / pytest (newest CPython, warning only) (push) Has been skipped
    Tests / shellcheck (install.sh, update.sh, bootstrap-ldap.sh, webupdate.sh, get_ldap-webgui.sh, setup.sh, config-backup.sh, certtool.sh, accesstool.sh) (push) Successful in 30s
    Tests / pytest (Python 3.11) (push) Successful in 19m22s
    Tests / pytest (Python 3.13) (push) Successful in 20m37s
    Tests / Publish release (push) Successful in 4s
    Stable

    Lyra released this 2026-11-06 20:29:37 +01:00 | 7 commits to main since this release

    Added

    • Machines — a read-only host register. A new Machines entry in
      the navigation (after Sudo Rules, both roles) answers per host who
      may log in over SSH and who has sudo there, derived from the directory
      as it is: a host is every name a sudoHost value gives (dotted values
      by their first label, negated ones included) and every posixGroup
      login-<host>. Nothing is stored, nothing is written to LDAP.
      /machines/ lists the hosts (server-side ?q= filter) with their
      source, the people in login-<host>, the rules naming them, a flag
      when only FQDN values name a host, and an inconsistent spelling flag
      listing the spellings found; above the table what applies everywhere,
      below it the rules that may apply on every host. /machines/<host> (any case) shows the line for that host's
      sshd_config with a copy button — AllowGroups login-<host>, or with
      the all-hosts group in front — the people it lets in with the path
      each came by (memberUid and primary group, service accounts marked),
      the sudo rules that apply, are excluded by a negated value, match only
      by FQDN or may apply, the people they resolve to one level deep, the
      members of the sudo group, and the caveats: the WebGUI cannot see
      sshd_config or the local sudoers, so every answer is labelled
      if the host has the line. Guide: docs/machines.md.
    • ldap.all_hosts_group (config.yaml, optional, empty by default):
      the cn of a posixGroup whose members may log in on every host. When
      set, it leads every host's AllowGroups line and its members are
      listed on every host page; a cn that names no posixGroup with a
      gidNumber gives a warning there. An invalid name is logged at startup
      and treated as unset. Shown read-only under Settings → LDAP
      Connection
      . See docs/settings.md.
    • A Machines card on every group page: "Opens login on " for
      a login-<host> group, "Logs in everywhere" for the all-hosts group,
      the local-sudoers line for auth.sudo_group, and each sudo rule
      naming the group (%<cn>, %#<gid>) with the hosts it names, linked
      ("every host except …" when ALL comes with negated values). When the directory cannot be read for it, the
      card says unavailable and the rest of the page is unchanged.
    • Annotated sudoHost values on a sudo rule's page: each value with
      its class (fqdn, opaque, wildcard, negated), host names linked
      to their host page. The sudoHost field on the create and edit forms
      carries a permanent hint that the spelling is case-sensitive — write
      it exactly as hostname prints it, because SSSD fetches a host's
      rules by an exact-case match. A wildcard value (web*, db[0-9]) always shows
      the hint that it applies only on clients with
      ldap_sudo_include_regexp = true in sssd.conf; it is allowed, never
      refused.
    • Warnings after saving a sudo rule (create and edit), one flash per
      finding: a wildcard sudoHost, and a host name whose spelling differs
      only in case from another spelling of the same host in the directory —
      only one of them can reach the host. Upper case alone never warns. The
      save is never blocked or undone; an unchanged save still warns, and
      when the directory cannot be read for the comparison only the wildcard
      check runs.

    Security

    • An edit through an entry's link could write a different entry.
      The uid/cn in the URL was cleaned by removing every character
      outside the plain set and cutting at 64, and the page and every form
      behind it then used the cleaned name: /users/a,b showed and saved
      the account ab, a 65-character uid its 64-character prefix, and
      /sudo/fz-s,alias the rule fz-salias — edit, delete, password
      reset, SSH keys, sudo grant, group members and sudo rules alike. A
      URL whose uid or cn the cleaning would change is now refused with
      400 before anything is read or written. Accounts, groups and sudo
      rules with such names (a comma, a space, non-ASCII, more than 64
      characters) are still listed but cannot be opened in the web GUI
      until 1.32.1; manage them with the LDAP tools until then.

    Fixed

    • Lists were silently cut at the directory's size limit. ldap3 does
      not raise on sizeLimitExceeded or timeLimitExceeded, so a search
      that hit slapd's limit (500 entries per search by default) returned
      the first entries as if they were all of them: user, group and sudo
      lists, the CSV import's uid set, the Machines page, the reference
      scan before deleting an account and the scan for the next free
      uidNumber/gidNumber all reported success on a partial answer. Every
      search now checks the result code and fails visibly instead — the
      page shows "The directory returned more entries than its size limit
      allows (N) — raise olcSizeLimit/olcLimits for the bind account" (or
      the time-limit equivalent), and an account delete or create is
      refused. Existence checks (does this uid, group or sudo rule exist,
      is the account in the sudo group) still accept a cut answer that
      contains a match.
    • The dashboard counted a failed search as 0. With the size-limit
      check above, a directory over slapd's limit made the user, group and
      sudo lists fail — and the dashboard (and its Refresh) turned each
      failure into 0, next to "Healthy". A count that cannot be read now
      shows "—" with the reason as its tooltip; the group count depends on
      the full user list (primary members), so a cut user search takes it
      along. A sudoers base that does not exist is still 0 without an
      error — the sudo schema is optional — and the sudo rules page says so
      as a warning instead of the raw noSuchObject error.
    • One account with two uid values broke every group page. The web
      GUI never writes a second uid or cn, but other tools can, and such
      an entry came back with the value as a list: every group page failed
      with an internal server error, the user list did too, and other
      pages showed and linked the list's Python repr (the CSV export both
      values in one cell). Users, groups and sudo
      rules now always carry one uid/cn — the first value; a user page
      opened by its second uid shows that one.

    Upgrade notes

    • No vhost change.
    • Directories with more entries than slapd's size limit (500 per
      search by default) in ou=people, the groups or the sudoers subtree:
      after the update the lists and the Machines page show "The directory
      returned more entries than its size limit allows" instead of a
      silently cut list, the dashboard shows "—" for the count, and
      creating an account or group with
      an automatically assigned ID is refused, until the limit is raised for
      the WebGUI's bind DN — one olcLimits line, see
      docs/ldap-acl-setup.md, Size limit for the bind account. Smaller
      directories need nothing.

    Update without git access: download the source tar.gz below, extract it, then run the update.sh it contains:
    sudo bash <extracted-dir>/update.sh --archive <tarball>

    Downloads
  • v1.31.0 8915756f9f

    LDAP WebGUI v1.31.0
    All checks were successful
    Tests / Scope (push) Successful in 1s
    Tests / pytest (newest CPython, warning only) (push) Has been skipped
    Tests / shellcheck (install.sh, update.sh, bootstrap-ldap.sh, webupdate.sh, get_ldap-webgui.sh, setup.sh, config-backup.sh, certtool.sh, accesstool.sh) (push) Successful in 27s
    Tests / pytest (Python 3.11) (push) Successful in 19m11s
    Tests / pytest (Python 3.13) (push) Successful in 20m20s
    Tests / Publish release (push) Successful in 3s
    Stable

    Lyra released this 2026-11-05 19:14:56 +01:00 | 8 commits to main since this release

    Added

    • Monitoring through Zabbix. GET /health with
      Authorization: Bearer <token> answers a read-only JSON status report
      (schema 1): the live LDAP service bind, certificate expiry (HTTPS, LDAP
      server, LDAP CA), update available, the last update run (web and
      ldap-webgui-update), lockout counts, the daily config backup — each
      cached part with its age. No addresses, paths or configuration values
      but the LDAP host:port: errors are fixed words (timeout, refused,
      tls, …), the full text stays in app.log. No session, no cookie; 200
      whenever the report could be built, the state is in the body. Without the header
      /health answers exactly as before.
      Only loopback, the admin hosts
      and the monitoring hosts reach it — no nginx change. Guide:
      docs/monitoring.md.
    • A Zabbix 7.0 template, docs/zabbix/ldap_webgui_by_http.yaml
      (also on the host under /opt/ldap-webgui/docs/zabbix/): one HTTP
      agent request per minute, dependent items, certificate discovery, and
      triggers for an LDAP bind failure, certificates at 30/15/7 days and
      expired, an available update, a failing update check, a failed update
      or rollback, permanent 2FA locks, login lockouts, a failed or stalled
      config backup, a stalled status refresh and a report that stops
      answering. TLS verification is on — the Zabbix server/proxy needs the
      CA of the WebGUI's certificate.
    • Settings → Monitoring (below Access): generate or revoke the
      token (full admins, password re-entry; the token is shown once, only
      its SHA-256 is kept in config.yaml), the URL and header for Zabbix,
      the number of monitoring hosts, the template download (both roles).
      On the console: manage.py monitoring-token [--revoke]. Security log:
      MONITORING_TOKEN_SET, MONITORING_TOKEN_REVOKED,
      MONITORING_REAUTH_FAILED, and MONITORING_AUTH_FAILED for a refused
      poll (throttled; a successful poll is never logged).
    • ldap-webgui-status-refresh.timer, every 15 minutes as the service
      user (manage.py refresh-status): refreshes the certificate data and
      the update check, each only when its own interval has run out, so a
      poll never waits for a handshake or the release source. Installed and
      enabled by install.sh; update.sh installs it and enables it only
      where the timer is new on the host — a disabled timer stays disabled, a
      masked one is left alone.

    Changed

    • update.sh records every real run — success, noop, failed,
      rolled back, rollback failed; from the web UI or the console, not a dry
      run or a preflight — in /var/lib/ldap-webgui-update/last-update.json.
    • config-backup.sh records each run in
      /var/lib/ldap-webgui-backup/status.json (time, result, newest copy —
      nothing of the config); the backup unit may write that directory.
    • /health without a token unbinds its service connection after the
      check.

    Upgrade notes

    • No vhost change. Updated from the web UI, with
      ldap-webgui-update --from-release, or with the update.sh of an
      unpacked 1.31.0 archive, the refresh timer arrives installed and
      enabled; nothing to do unless you want Zabbix to watch the host.
    • To use the report: put the Zabbix server — or its proxy — into the
      monitoring hosts (Settings → Access when ldap-webgui-access.path is
      enabled, else install.sh --allow-monitoring LIST, or the file
      /etc/nginx/ldap-webgui-monitoring.conf plus
      nginx -t && systemctl reload nginx), generate a token under
      Settings → Monitoring and import the template
      (docs/monitoring.md).
    • ldap-webgui-update without --from-release (git mode, or
      --archive) runs the clone's update.sh — still 1.30.0's, unless you
      pulled the clone by hand — and that one carries out the whole update:
      it knows neither the refresh timer nor the last-update record. Only
      release mode switches to 1.31.0's own update.sh before changing
      anything. After such an update re-apply 1.31.0 once with its own
      update.sh — ldap-webgui-update --reinstall (git mode; the clone is
      at 1.31.0 by then), or on a host without git access
      bash <unpacked 1.31.0>/update.sh --archive <file> --reinstall: it
      installs and enables the timer and writes the first record.

    Update without git access: download the source tar.gz below, extract it, then run the update.sh it contains:
    sudo bash <extracted-dir>/update.sh --archive <tarball>

    Downloads
  • v1.30.0 492003bdf5

    LDAP WebGUI v1.30.0
    All checks were successful
    Tests / shellcheck (install.sh, update.sh, bootstrap-ldap.sh, webupdate.sh, get_ldap-webgui.sh, setup.sh, config-backup.sh, certtool.sh, accesstool.sh) (push) Successful in 24s
    Tests / pytest (Python 3.11) (push) Successful in 18m55s
    Tests / pytest (Python 3.13) (push) Successful in 20m6s
    Tests / Publish release (push) Successful in 3s
    Tests / pytest (newest CPython, warning only) (push) Has been skipped
    Tests / Scope (push) Successful in 3s
    Stable

    Lyra released this 2026-11-05 16:22:31 +01:00 | 9 commits to main since this release

    Added

    • Change who may reach the WebGUI from the web UI (Settings →
      Access, full admins, password re-entry): the admin list and the
      monitoring list, one list per save, with the same rules as
      install.sh --allow and at most 64 entries. A list that would shut out
      your own address is refused before anything is written; any needs a
      second step in which you type any again, and the unrestricted banner
      then shows on every page. A list changed by someone else since you
      loaded the page is refused. A root-owned runner writes the file, runs
      nginx -t, reloads nginx and confirms /health still answers —
      otherwise the previous file goes back; after a change that went
      through it stays next to the live one as <file>.prev for a manual
      back-out. A request the runner has not picked up within 10 minutes is
      dropped and shown as such, and install.sh --allow waits for a running
      change instead of interleaving with it. Every request and outcome is in
      the security log with the list before and after (ACCESS_LIST_*,
      ACCESS_REAUTH_FAILED). The runner (ldap-webgui-access.path /
      .service) is off until you enable it.

    Changed

    • The access list files' header names both writers: Written by the LDAP
      WebGUI (Settings → Access) or install.sh
      . Existing files keep their
      header until they are next written; both are read the same.
    • CI: the release job moved into .forgejo/workflows/test.yml
      (release.yml is gone). A release is published only after pytest
      (3.11, 3.13) and shellcheck passed on the tagged tree; the release
      repository no longer runs the suite twice per release, the dev
      repository no longer on tag pushes. The newest-CPython run is its own
      job that nothing depends on.

    Upgrade notes

    • No vhost change. To change the access lists from the web UI, enable the
      runner once, as root: systemctl enable --now ldap-webgui-access.path
      (docs/settings.md → Changing the lists from the web UI). Without
      it the Access card stays read-only, as before.

    Update without git access: download the source tar.gz below, extract it, then run the update.sh it contains:
    sudo bash <extracted-dir>/update.sh --archive <tarball>

    Downloads