CHANGELOG.md

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

Unreleased

Mandatory Configuration Change

Datastores are no longer managed by matrix-media-repo internally, meaning you MUST specify a datastore ID on each of your configured datastores. If you're setting up matrix-media-repo for the first time then you can use whatever you want for a datastore ID (though it's recommended to stick to alphanumeric strings). If you're upgrading to this version however, you will need to pull the datastore IDs out of the matrix-media-repo and add them to your configuration.

For safety, the datastores table is not deleted from the database in this upgrade. A future version may drop the table, however.

Getting existing datastore IDs

Before upgrading, you can get your datastore IDs fairly easily. The best way might be to look at the startup log of your media repo:

INFO[2023-05-21 20:58:45.116 Z] Datastores:                                  
INFO[2023-05-21 20:58:45.116 Z]         file (e9ce13bbb062383ce1bcee76414058668877f2d51635810652335374336): /mnt/mmr-store/location4
INFO[2023-05-21 20:58:45.117 Z]         s3 (7669e2fb8ccaa0801e4255a417ad20884f76b8611659655069202644992): s3://redacted.r2.cloudflarestorage.com/redacted

This way, you're able to correlate locations to IDs. For example, the file datastore configured to put media at /mnt/mmr-store/location4 has ID e9ce13bbb062383ce1bcee76414058668877f2d51635810652335374336. Add this as id: "e9ce13bbb062383ce1bcee76414058668877f2d51635810652335374336" in your media repo config file.

Alternatively, you can use the admin API to get your datastores:

curl -s -X GET -H "Authorization: Bearer YOUR_ACCESS_TOKEN" https://example.org/_matrix/media/unstable/admin/datastores
{
  "e9ce13bbb062383ce1bcee76414058668877f2d51635810652335374336": {
    "type": "file",
    "uri": "/mnt/mmr-store/location4"
  },
  "7669e2fb8ccaa0801e4255a417ad20884f76b8611659655069202644992": {
    "type": "s3",
    "uri": "s3://redacted.r2.cloudflarestorage.com/redacted"
  }
}

The returned object is keyed by ID over the API.

In either case, take the ID and add it to the associated datastore in your config, similar to the following:

# Your specific configuration may be different
datastores:
  - type: file
    id: "e9ce13bbb062383ce1bcee76414058668877f2d51635810652335374336"  ## ADD THIS
    enabled: true
    forKinds: ["archives"]
    opts:
      path: "/mnt/mmr-store/location4"
  - type: s3
    id: "7669e2fb8ccaa0801e4255a417ad20884f76b8611659655069202644992"  ## ADD THIS
    enabled: true
    forKinds: ["all"]
    opts:
      ssl: true
      tempPath: "/mnt/mmr-store/s3-staging"
      endpoint: redacted.r2.cloudflarestorage.com
      accessKeyId: "redacted"
      accessSecret: "redacted"
      bucketName: "redacted"

Note: If matrix-media-repo detects that a datastore ID is used but not referenced in the config then it will refuse to start.

This new configuration style additionally allows for out-of-band datastore transfers. If you move all your data to a new path/server, for example, then you can simply update the path in the config for that datastore.

Deprecations

• The GET /_matrix/media/unstable/local_copy/:server/:mediaId (and unstable/io.t2bot.media variant) endpoint is deprecated and scheduled for removal. If you are using this endpoint, please comment on this issue to explain your use case.

Added

• Added a federation.ignoredHosts config option to block media from individual homeservers.
• Support for MSC2246 (async uploads) is added, with per-user quota limiting options.
• Support for MSC4034 (self-serve usage information) is added, alongside a new "maximum file count" quota limit.
• The GET /_synapse/admin/v1/statistics/users/media endpoint from Synapse is now supported at the same path for local server admins.
• Thumbnailing support for BMP images.
• Thumbnailing support for TIFF images.

Removed

• IPFS support has been removed due to maintenance burden.
• Exports initiated through the admin API no longer support ?include_data=false. Exports will always contain data.

Changed

• Mandatory configuration change: You must add datastore IDs to your datastore configuration, as matrix-media-repo will no longer manage datastores for you.
• Datastores no longer use the enabled flag set on them. Use forKinds: [] instead.
• Some admin endpoints for purging media, quarantining media, and background task information now require additional path components.

  -POST /_matrix/media/unstable/admin/purge/<server>/<media id>?access_token=your_access_token
  +POST /_matrix/media/unstable/admin/purge/media/<server>/<media id>?access_token=your_access_token
  
  -POST /_matrix/media/unstable/admin/quarantine/<server>/<media id>?access_token=your_access_token
  +POST /_matrix/media/unstable/admin/quarantine/media/<server>/<media id>?access_token=your_access_token
  
  -GET /_matrix/media/unstable/admin/tasks/<task ID>
  +GET /_matrix/media/unstable/admin/task/<task ID>

• Per-user upload quotas now do not allow users to exceed the maximum values, even by 1 byte. Previously, users could exceed the limits by a little bit.
• Updated to Go 1.19
• Updated to Go 1.20
• Logs are now less noisy by default.
• Connected homeservers must support at least Matrix 1.1 on the Client-Server API.

Fixed

• URL previews now follow redirects properly.
• Overall memory usage is improved, particularly during media uploads and API-initiated imports.
  • Note: If you use plugins then memory usage will still be somewhat high due to temporary caching of uploads.
• Fixed shutdown stall if the config was reloaded more than once while running.

1.2.13 - February 12, 2023

Deprecations

• In version 1.3.0, IPFS will no longer be supported as a datastore. Please migrate your data if you are using the IPFS support.

Added

• Added the Cross-Origin-Resource-Policy: cross-origin header to all downloads, as per MSC3828.
• Added metrics for tracking which S3 operations are performed against datastores.

Changed

• Swap out the HEIF library for better support towards ARM64 Docker Images.
• The development environment now uses Synapse as a homeserver. Test accounts will need recreating.
• Updated to Go 1.18
• Improved error message when thumbnailer cannot determine image dimensions.

Fixed

• Return default media attributes if none have been explicitly set.

1.2.12 - March 31, 2022

Fixed

• Fixed a permissions check issue on the new statistics endpoint released in v1.2.11

1.2.11 - March 31, 2022

Added

• New config option to set user agent when requesting URL previews.
• Added support for image/jxl thumbnailing.
• Built-in early support for content ranges (being able to skip around in audio and video). This is only available if caching is enabled.
• New config option for changing the log level.
• New (currently undocumented) binary s3_consistency_check to find objects in S3 which might not be referenced by the media repo database. Note that this can include uploads in progress.
• Admin endpoint to GET users' usage statistics for a server.

Removed

• Support for the in-memory cache has been removed. Redis or having no cache are now the only options.
• Support for the Redis config under features has been removed. It is now only available at the top level of the config. See the sample config for more details.

Fixed

• Fixed media being permanently lost when transferring to an (effectively) readonly S3 datastore.
• Purging non-existent files now won't cause errors.
• Fixed HEIF/HEIC thumbnailing. Note that this thumbnail type might cause increased memory usage.
• Ensure endpoints register in a stable way, making them predictably available.
• Reduced download hits to datastores when using Redis cache.

Changed

• Updated support for post-MSC3069 homeservers.
• Updated the built-in oEmbed providers.json

1.2.10 - December 23rd, 2021

Deprecation notices

In a future version (likely the next), the in-memory cache support will be removed. Instead, please use the Redis caching that is now supported properly by this release, or disable caching if not applicable for your deployment.

Added

• Added support for setting the Redis database number.

Fixed

• Fixed an issue with the Redis config not being recognized at the root level.

1.2.9 - December 22nd, 2021

Deprecation notices

In a future version (likely the next), the in-memory cache support will be removed. Instead, please use the Redis caching that is now supported properly by this release, or disable caching if not applicable for your deployment.

Added

• Added support for HEAD at the /healthz endpoint.
• Added X-Content-Security-Policy: sandbox in contexts where the normal CSP header would be served. This is a limited, pre-standard form of CSP supported by IE11, in order to have at least some mitigation of XSS attacks.
• Added support for the org.matrix.msc2705.animated query parameter.
• Added support for S3 storage classes (optional).
• Added support for listening on Matrix 1.1 endpoints (/_matrix/media/v3/*).

Changed

• Support the Redis config at the root level of the config, promoting it to a proper feature.

Fixed

• Improved performance of datastore selection when only one datastore is eligible to contain media.
• Fixed blurhash not enabling itself.
• Fixed blurhash implementation to match MSC.

1.2.8 - April 30th, 2021

Fixed

• Fixed crashes when internal workers encounter panics.

[1.2.7] - April 19th, 2021

Security advisories

This release includes a fix for CVE-2021-29453.

Server administrators are recommended to upgrade as soon as possible. This issue is considered to be exploited in the wild due to some deployments being affected unexpectedly.

Added

• Added support for structured logging (JSON).

Changed

• Turned color-coded logs off by default. This can be changed in the config.

Fixed

• Fixed memory exhaustion when thumbnailing maliciously crafted images.

1.2.6 - March 25th, 2021

Added

• Added ffmpeg and ImageMagick to Docker image to support specialized thumbnail types.

Fixed

• Handle guest accounts properly. Previously they were still declined, though by coincidence.

1.2.5 - March 17th, 2021

Added

• Added a -verify mode to imports to determine if large imports were successful.
• Added optional support for Sentry (error reporting).

Changed

• Content-Disposition of plain text files now defaults to inline.

Fixed

• Fixed rich oEmbed URL previews (Twitter).
• Fixed photo oEmbed URL previews (Giphy).
• Fixed orientation parsing for some thumbnails.
• Fixed file name being incorrect on the first download from remote servers.
• Fixed a download inefficiency where remote downloads could use extra bandwidth.
• Fixed a problem where secondary imports can never finish.
• Fixed imports not handling duplicate media IDs.
• Fixed some database connection errors not being handled correctly.

1.2.4 - March 5th, 2021

Fixed

• Fixed build error for modern versions of Go, improving IPFS implementation.

1.2.3 - March 4th, 2021

Added

• Introduced early plugin support (only for antispam for now).
  • Includes a simple OCR plugin to help mitigate text-based image spam.
• Added an X-Robots-Tag header to help prevent indexing. Thanks @jellykells!

Fixed

• Fixed crash when generating some thumbnails of audio.
• Fixed various artifact problems with APNG and GIF thumbnails. Thanks @Sorunome!
• Fixed a missing "unlimited size" check for thumbnails. Thanks @Sorunome!

1.2.2 - December 8th, 2020

Fixed

• Generate JPEG thumbnails for JPEG for reduced file size. Thanks @Sorunome!
• Strip charset parameter off binary media for better compatibility with other homeservers.

1.2.1 - October 27th, 2020

Added

• Added a new tool, export_synapse_for_import, which can be used to do an offline import from Synapse.
  • After running this tool, use the gdpr_import tool to bring the export into the media repo.
• Added thumbnailing support for some audio waveforms (MP3, WAV, OGG, and FLAC).
• Added audio metadata (duration, etc) to the unstable /info endpoint. Aligns with MSC2380.
• Added simple thumbnailing for MP4 videos.
• Added an asAttachment query parameter to download requests per MSC2702.

Fixed

• Fixed thumbnails for invalid JPEGs.
• Fixed incorrect metrics being published when using the Redis cache.
• Fixed errors generating thumbnails when bad EXIF headers were provided.
• Use r0 instead of v1 for federation requests. No changes should be needed to configurations or routing - it'll just work.

1.2.0 - August 2nd, 2020

Upgrade notes

This release contains a database change which might take a while. In order to support quotas, this release tracks how much a user has uploaded, which might take a while to initially calculate. If you have a large database (more than about 100k uploaded files), run the following steps before upgrading:

1. The PostgreSQL script described here. This can be run while the server is running.
2. If you have no intention of using stats or quotas, you're done (the stats table will be inaccurate). If you do plan on using either, run INSERT INTO user_stats SELECT user_id, SUM(size_bytes) FROM media GROUP BY user_id; which may take a while.
3. Change the owner of the table and function to your media repo's postgresql user. For example, if your postgres user is media, then run:

   ALTER TABLE user_stats OWNER TO media;
   ALTER FUNCTION track_update_user_media() OWNER TO media; 

Added

• Add webp image support. Thanks @Sorunome!
• Add apng image support. Thanks @Sorunome!
• Experimental support for Redis as a cache (in preparation for proper load balancing/HA support).
• Added oEmbed URL preview support.
• Added support for dynamic thumbnails.
• Added a way to prevent certain media from being quarantined (attributes API).
• Added support for quotas.

Changed

• Remove deprecated support for restricting uploads to certain mime types.
• Remove deprecated support for forUploads.
• Clarified what uploads.minBytes is intended to be used for.

Fixed

• GIFs now thumbnail correctly. Thanks @Sorunome!
• Fixed empty Content-Type header on retrieved remote media. Thanks @silkeh!
• Fixed various issues with IPv6 handling. Thanks @silkeh!
• Fixed high database usage for uploads when only one datastore is present.
• Fixed incorrect HTTP status codes for bad thumbnail requests.
• Fixed dimension checking on thumbnails.
• Fixed handling of EXIF metadata. Thanks @sorunome!
• Fixed handling of URL previews for some encodings.
• Fixed Cache-Control headers being present on errors.

1.1.3 - July 15th, 2020

Added

• Added options to cache access tokens for users. This prevents excessive calls to /account/whoami on your homeserver, particularly for appservices.
• Documentation on how to set up delegation with the media repo and Traefik. Thanks @derEisele!

Changed