github owncloud/ocis v8.0.11

2 hours ago

Table of Contents

  • Changelog for 8.0.11

Changes in 8.0.11

Summary

  • Security - Fix signing-key to public share guests: #12332
  • Security - Do not leak internal error details in webdav error responses: #12398
  • Security - Harden public link brute force protection: #12944
  • Security - Enforce sharing permissions when updating links: #13447
  • Bugfix - Rate-limit the exportPersonalData endpoint: #12516
  • Bugfix - Apply the role allowlist on permission updates: #12540
  • Bugfix - Use a dedicated ICAP client per scan: #13386
  • Bugfix - Turn on the events TLS in the jsoncs3 share manager and in OCM: #13442
  • Bugfix - Fix slow builds with GNU Make 4.4: #13454
  • Bugfix - Remove stale group shares in the share manager: #13460
  • Bugfix - Filter spurious TLS handshake errors from CI test runner output: #13499
  • Bugfix - Wait for app registry before running WOPI cs3 validator tests: #13504
  • Bugfix - Reuse HTTP connections in the proxy and in the metadata client: #13539
  • Enhancement - Harden the Collabora deployment examples: #13389
  • Enhancement - Bump Web to 12.3.4: #13402
  • Enhancement - Update Mockery to 2.53.8: #13454
  • Enhancement - Bump Web to 12.3.5: #13487
  • Enhancement - Bump Web to 12.3.6: #13521

Details

  • Security - Fix signing-key to public share guests: #12332

    The /ocs/v[12].php/cloud/user/signing-key endpoint was reachable through a
    public share session. The endpoint public-token is no longer allowed by the
    public-share resource scope in reva.

    #12332
    owncloud/reva#608

  • Security - Do not leak internal error details in webdav error responses: #12398

    The webdav service returned raw internal error messages to the client when a
    server-side error occurred. For example, a thumbnail request with a null byte in
    the filename and ?preview=1 could trigger a 500 Internal Server Error whose
    body exposed the internal storage filesystem path (information disclosure).

    Server-side error handlers now return a generic, user-relevant message to the
    client, while the detailed error is only logged server-side.

    #12398

  • Security - Harden public link brute force protection: #12944

    We've hardened the brute force protection for password protected public links so
    that failed password attempts are reliably counted and the rate limit is
    consistently enforced.

    #12944

  • Security - Enforce sharing permissions when updating links: #13447

    Updating a link now applies the same permission checks that apply when creating
    a link.

    #13447

  • Bugfix - Rate-limit the exportPersonalData endpoint: #12516

    We've added a rate limit to the exportPersonalData endpoint to mitigate an
    authenticated application-level denial-of-service. Rate-limit per endpoint path
    carries the userID, so effectively per user. The endpoint is now limited to 5
    requests per minute.

    #12516

  • Bugfix - Apply the role allowlist on permission updates: #12540

    The PATCH permission and space-root permission handlers now consult the
    administrator role allowlist when validating a request, consistent with the
    invite handler.

    #12540

  • Bugfix - Use a dedicated ICAP client per scan: #13386

    The antivirus service now creates a dedicated ICAP client for every scan instead
    of sharing a single client between its workers. This only affects deployments
    using ANTIVIRUS_SCANNER_TYPE=icap.

    #13386

  • Bugfix - Turn on the events TLS in the jsoncs3 share manager and in OCM: #13442

    The sharing service now gives the OCIS_EVENTS_ENABLE_TLS setting to the
    jsoncs3 share manager. The ocm service now gives this setting to the sciencemesh
    and the ocmcore event connections. Before, these connections did not turn on TLS
    for the connection to the events broker. Thus they did not use the root CA
    certificate. The connection to a NATS server with a private CA failed. The
    services tried again for up to 15 minutes.

    #13442

  • Bugfix - Fix slow builds with GNU Make 4.4: #13454

    The file .make/recursion.mk now expands the GO, GOPATH, GOHOSTOS,
    GOHOSTARCH and GOHOSTARM variables one time only, before its export.
    Before, each use of these variables ran a shell command again. With GNU Make
    4.4, these commands ran again and again. A make call took about two minutes
    before it started the build, and it printed thousands of "env: 'GOHOSTOS': No
    such file or directory" errors. Ubuntu 26.04 and the ubuntu-latest GitHub
    runner have GNU Make 4.4.

    #13454

  • Bugfix - Remove stale group shares in the share manager: #13460

    A stale group share is a share ID in the group cache that has no share. On each
    listing of the received shares, the share manager found the stale ID again and
    uploaded the received.json file of the user again with the same content. The
    stale ID stayed, so this repeated on each listing.

    The share manager now removes a stale group share from the group cache. It does
    this only when it read the share file of the space. It reads each share from the
    same copy of that file, so a parallel listing cannot make valid shares look
    stale. A Remove no longer uploads a file when the share is not in it. A listing
    of the created shares now also reads each share from one copy of the share file.
    Before, a parallel listing could hide some shares of a space from the creator.

    An update of a share now reads the share again under the space lock. Before, a
    parallel failed upload in the same space could remove the space from memory. The
    update then reported success, but it did not write the change. An update now
    also returns NotFound for a share that a parallel unshare removed. Before, the
    update could write the removed share again.

    When the upload of a Remove fails, the share manager now reads the file again on
    the next call. Before, it kept the share out of memory, but the share stayed in
    the file. A retry then found nothing to remove, did not upload, and reported
    success. In the received.json file of a user, this also occurred inside one
    Remove after a temporary upload error. The same applies to an Add. Before, a
    failed upload of an Add kept the new entry in memory. For example, a client got
    an error when it accepted a share, but this pod showed the share as accepted.

    An unshare and an update of a share now read the share file of the space again
    when the share is not in memory. Before, a listing that got a temporary NotFound
    left an empty copy of the space in memory. An unshare then did not remove the
    share from the file, but it reported success. An update returned NotFound for a
    share that existed.

    The share manager removes a stale entry only after a grace period of 5 minutes.
    During a rolling upgrade, an older version can write the grantee entry before
    the share. Without the grace period, a listing in this window removed the entry,
    and no group member saw the share. A listing that finds the share again restarts
    the grace period. Thus two short losses do not add up to one long loss. Each pod
    counts the grace period itself. A pod remembers a stale entry for 1 hour after
    the last listing that found it. Thus the grace period also ends on a pod that
    finds the entry only once in some minutes. A pod looks for the entries to forget
    at most once a minute. Thus a listing that finds many stale entries does not
    hold the lock for a long time. When the removal of a stale entry fails, the next
    listing tries the removal again at once. The pod forgets the entry only after
    the removal.

    The share manager now also removes a stale entry when the share file of the
    space does not exist. A failed first share in a space leaves such an entry. A
    download that gets NotFound is not enough proof, because the NotFound can be
    temporary. The share manager checks the file a second time.

    The share manager now writes a share before the entry for the grantee. When a
    write fails, it removes the entries that it wrote, also the entry for the
    grantee. The removal also runs when the request was cancelled. Each removal has
    its own timeout, so a slow removal does not stop the next one. Thus a failed
    share does not block a new share with the same grantee. The share manager
    removes the share and the entry of the creator at the same time, so a failed
    share waits for at most two timeouts. An unshare removes the share also when the
    entry for the grantee cannot be removed, and it reports success. Before, the
    client got an error for a share that was gone. The next listing of the grantee
    removes the stale entry.

    The share manager now removes many shares together, with one upload for each
    document. A purge of a space with 20 group shares did 41 uploads and now does 3.
    A listing that finds 20 expired shares did 60 uploads and now does 3. A listing
    removes the expired shares of all spaces together. The setting
    SHARING_USER_JSONCS3_MAX_CONCURRENCY limits the parallel removals. Its default
    does not change and is 1. The removals always use a limit of at least 3. Before,
    the removal of one share updated its 3 documents in parallel. With a limit of 1,
    an unshare needed one more round trip to the storage.

    The share cleanup of the ocis shares commands now removes all orphaned shares
    together, with one upload for each document. Before, it removed the shares one
    at a time. 20 orphaned shares in one space did 60 uploads and now do 3. A Stat
    error no longer stops the cleanup with a panic. The cleanup skips that share and
    checks the next share. The cleanup also stops with an error log when it cannot
    get a gateway client. Before, it stopped with a panic.

    The share manager now sends one ShareExpired event for each expired share.
    Before, each listing that found the expired share sent an event, also when the
    removal failed. Thus the users got many expiry notifications.

    The share manager no longer keeps an empty space entry in the created cache and
    in the group cache. Before, each listing read the share file of a deleted space
    again.

    When a space is deleted, the share manager now logs the errors of the share
    cleanup. Before, it sent them to a disabled logger.

    Each log line of a share listing now has the fields of one share only. Before, a
    later line had the fields of all earlier shares, and a log search found the
    wrong share.

    A Remove on the created cache or on the group cache no longer panics when the
    user or group has no entry for the space of the share, or when the share ID has
    no delimiter. Before, the panic stopped the sharing service, also in an unshare
    and in the cleanup of orphaned shares.

    #13460
    owncloud/reva#813

  • Bugfix - Filter spurious TLS handshake errors from CI test runner output: #13499

    Go's net/http server emits TLS handshake error from <addr>: ... to stderr
    whenever a client connects to an HTTPS port but does not complete the handshake.
    This happens routinely in acceptance test runs: health-check probes, Playwright
    browser initialisation, and curl readiness polls all trigger the message even
    though no test ever fails because of it. The messages accumulate in the hundreds
    per run and push genuinely interesting output far down the CI log.

    The five Python test-runner scripts (run-github.py, run-litmus.py,
    run-cs3api.py, run-wopi.py) now pipe each long-lived ocis/ociswrapper
    subprocess's stderr through a background thread that drops lines containing TLS handshake error and forwards everything else to the runner's stderr unchanged.
    Real errors remain fully visible.

    The two Bash scripts (tests/ldap-smoke/run.sh,
    tests/ldap-smoke-distributed/run.sh) use a Bash process substitution (`2>

    (grep -v "TLS handshake error" >&2)`) for the same effect without requiring a
    log file.

    #13499

  • Bugfix - Wait for app registry before running WOPI cs3 validator tests: #13504

    The WOPI cs3 validator occasionally failed with HTTP 500 on OpenInApp calls
    during the EditFlows test group. The gateway logged error finding app providers — meaning the app-provider service had not yet finished registering
    its mime-type entries with the gateway, even though the ocis_healthy() check
    (a graph API call) had already passed.

    A new app_registry_ready() probe is added to run-wopi.py that calls GET /app/list and verifies the response contains at least one mime-type entry. This
    probe runs after ocis_healthy() and before the WOPI service is started,
    ensuring the gateway's app provider list is populated before the validator
    begins issuing OpenInApp requests.

    #13504

  • Bugfix - Reuse HTTP connections in the proxy and in the metadata client: #13539

    Under high load, oCIS opened and closed many TCP connections each second. The
    host ran out of free ports. This caused failed logins and failed requests.

    The proxy now keeps up to 100 idle connections for each backend. Before, it kept
    2. The proxy has no global limit for idle connections. The OIDC client of the
    proxy now keeps its connections alive. Before, it opened a new connection with a
    new TLS handshake for each call.

    The metadata client and the HTTP clients of reva use the same settings.

    The proxy and the reva clients now close an idle connection after 30 seconds.
    Before, the proxy used 90 seconds. A load balancer or an ingress controller
    often closes an idle connection after 60 seconds. If the client closes first, it
    never sends a request on a connection that the other side already closed. The
    transport does not repeat such a request for methods like PUT, POST or PROPFIND.

    #13539
    owncloud/reva#824

  • Enhancement - Harden the Collabora deployment examples: #13389

    The ocis_full and ocis_ocm deployment examples now ship a hardened Collabora
    configuration. Collabora is restricted to the hosts it needs, runs on its own
    internal network that only the proxy and the collaboration service share, and
    the admin console requires explicitly configured credentials
    (COLLABORA_ADMIN_USER and COLLABORA_ADMIN_PASSWORD). If you copy these
    examples, make sure to set the credentials in your .env file.

    #13389

  • Enhancement - Bump Web to 12.3.4: #13402

    #13402
    https://github.com/owncloud/web/releases/tag/v12.3.4

  • Enhancement - Update Mockery to 2.53.8: #13454

    We updated the mockery tool in .bingo/mockery.mod from v2.53.5 to v2.53.8.

    #13454

  • Enhancement - Bump Web to 12.3.5: #13487

    #13487
    https://github.com/owncloud/web/releases/tag/v12.3.5

  • Enhancement - Bump Web to 12.3.6: #13521

    #13521
    https://github.com/owncloud/web/releases/tag/v12.3.6

Don't miss a new ocis release

NewReleases is sending notifications on new releases.