Mastodon is a free, open-source social network server based on ActivityPub. Mastodon versions before v4.3.18, v4.4.12, and v4.5.5 do not have a limit on the maximum number of poll options for remote posts, allowing attackers to create polls with a very large amount of options, greatly increasing resource consumption. Depending on the number of poll options, an attacker can cause disproportionate resource usage in both Mastodon servers and clients, potentially causing Denial of Service either server-side or client-side. Mastodon versions v4.5.5, v4.4.12, v4.3.18 are patched.
Mastodon is a free, open-source social network server based on ActivityPub. Prior to versions 4.5.5, 4.4.12, and 4.3.18, the server does not enforce a maximum length for the names of lists or filters, or for filter keywords, allowing any user to set an arbitrarily long string as the name or keyword. Any local user can abuse the list or filter fields to cause disproportionate storage and computing resource usage. They can additionally cause their own web interface to be unusable, although they must intentionally do this to themselves or unknowingly approve a malicious API client. Mastodon versions v4.5.5, v4.4.12, v4.3.18 are patched.
Mastodon is a free, open-source social network server based on ActivityPub. Prior to versions 4.5.5, 4.4.12, and 4.3.18, an insecure direct object reference in the web push subscription update endpoint lets any authenticated user update another user's push subscription by guessing or obtaining the numeric subscription id. This can be used to disrupt push notifications for other users and also leaks the web push subscription endpoint. Any user with a web push subscription is impacted, because another authenticated user can tamper with their push subscription settings if they can guess or obtain the subscription id. This allows an attacker to disrupt push notifications by changing the policy (whether to filter notifications from non-followers or non-followed users) and subscribed notification types of their victims. Additionally, the endpoint returns the subscription object, which includes the push notification endpoint for this subscription, but not its keypair. Mastodon versions v4.5.5, v4.4.12, v4.3.18 are patched.
Mastodon is a free, open-source social network server based on ActivityPub. Mastodon allows server administrators to suspend remote users to prevent interactions. However, some logic errors allow already-known posts from such suspended users to appear in timelines if boosted. Furthermore, under certain circumstances, previously-unknown posts from suspended users can be processed. This issue allows old posts from suspended users to occasionally end up on timelines on all Mastodon versions. Additionally, on Mastodon versions from v4.5.0 to v4.5.4, v4.4.5 to v4.4.11, v4.3.13 to v4.3.17, and v4.2.26 to v4.2.29, remote suspended users can partially bypass the suspension to get new posts in. Mastodon versions v4.5.5, v4.4.12, v4.3.18 are patched.
Copier is a library and CLI app for rendering project templates. Prior to version 9.11.2, Copier suggests that it's safe to generate a project from a safe template, i.e. one that doesn't use unsafe features like custom Jinja extensions which would require passing the `--UNSAFE,--trust` flag. As it turns out, a safe template can currently include arbitrary files/directories outside the local template clone location by using symlinks along with `_preserve_symlinks: false` (which is Copier's default setting). Version 9.11.2 patches the issue.
Copier is a library and CLI app for rendering project templates. Prior to version 9.11.2, Copier suggests that it's safe to generate a project from a safe template, i.e. one that doesn't use unsafe features like custom Jinja extensions which would require passing the `--UNSAFE,--trust` flag. As it turns out, a safe template can currently write to arbitrary directories outside the destination path by using directory a symlink along with `_preserve_symlinks: true` and a generated directory structure whose rendered path is inside the symlinked directory. This way, a malicious template author can create a template that overwrites arbitrary files (according to the user's write permissions), e.g., to cause havoc. Version 9.11.2 patches the issue.
vLLM is an inference and serving engine for large language models (LLMs). Starting in version 0.10.1 and prior to version 0.14.0, vLLM loads Hugging Face `auto_map` dynamic modules during model resolution without gating on `trust_remote_code`, allowing attacker-controlled Python code in a model repo/path to execute at server startup. An attacker who can influence the model repo/path (local directory or remote Hugging Face repo) can achieve arbitrary code execution on the vLLM host during model load. This happens before any request handling and does not require API access. Version 0.14.0 fixes the issue.
Saleor is an e-commerce platform. Starting in version 3.0.0 and prior to versions 3.20.108, 3.21.43, and 3.22.27, Saleor was allowing users to modify rich text fields with HTML without running any backend HTML cleaners thus allowing malicious actors to perform stored XSS attacks on dashboards and storefronts. Malicious staff members could craft script injections to target other staff members, possibly stealing their access and/or refresh tokens. This issue has been patched in versions 3.22.27, 3.21.43, and 3.20.108. In case of inability to upgrade straight away, a possible workaround is to use client-side cleaner.
Saleor is an e-commerce platform. Starting in version 3.0.0 and prior to versions 3.20.108, 3.21.43, and 3.22.27, Saleor allowed authenticated staff users or Apps to upload arbitrary files, including malicious HTML and SVG files containing Javascript. Depending on the deployment strategy, these files may be served from the same domain as the dashboard without any restrictions leading to the execution of malicious scripts in the context of the user's browser. Malicious staff members could craft script injections to target other staff members, possibly stealing their access and/or refresh tokens. Users are vulnerable if they host the media files inside the same domain as the dashboard, e.g., dashboard is at `example.com/dashboard/` and media are under `example.com/media/`. They are not impact if media files are hosted in a different domain, e.g., `media.example.com`. Users are impacted if they do not return a `Content-Disposition: attachment` header for the media files. Saleor Cloud users are not impacted. This issue has been patched in versions: 3.22.27, 3.21.43, and 3.20.108. Some workarounds are available for those unable to upgrade. Configure the servers hosting the media files (e.g., CDN or reverse proxy) to return the Content-Disposition: attachment header. This instructs browsers to download the file instead of rendering them in the browser. Prevent the servers from returning HTML and SVG files. Set-up a `Content-Security-Policy` for media files, such as `Content-Security-Policy: default-src 'none'; base-uri 'none'; frame-ancestors 'none'; form-action 'none';`.
CVAT is an open source interactive video and image annotation tool for computer vision. In versions 2.2.0 through 2.54.0, an attacker is able to execute arbitrary JavaScript in a victim user's CVAT UI session, provided that they are able to create a maliciously crafted label in a CVAT task or project, then get the victim user to either edit that label, or view a shape that refers to that label; and/or get the victim user to upload a maliciously crafted SVG image when configuring a skeleton. This gives the attacker temporary access to all CVAT resources that the victim user can access. Version 2.55.0 fixes the issue.