Add advisory for physis_pro: remote error body can permanently disable a shared LLM provider - #3175
Open
ilPez00 wants to merge 1 commit into
Open
Add advisory for physis_pro: remote error body can permanently disable a shared LLM provider#3175ilPez00 wants to merge 1 commit into
ilPez00 wants to merge 1 commit into
Conversation
…e a shared LLM provider Filed by the crate author. Fixed in physis_pro 0.1.9; 0.1.6 through 0.1.8 are yanked.
Member
|
Please use the pull request template, and don't have your agent write the PR description. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Filed by the crate author, which per CONTRIBUTING waives the upstream-report and coordinated-disclosure prerequisites. The embargo is lifted: the fix, the affected range and the mechanism are already public in the crate's own repository and changelog.
Summary.
physis_proroutes completions through a cascade of LLM providers. In0.1.6–0.1.8a provider was removed from the cascade by substring-matching the text of a failed HTTP response. OpenAI-compatible endpoints quote the offending request back insideerror.message, and that request carries the caller's prompt — so a prompt containing a phrase likedoes not existthat provokes any 4xx would match. The removal was permanent (mark_unavailable()wrote a flag onlyProvider::new()ever cleared) and the cascade is process-wide, so the loss hit every caller rather than the one who caused it. Walking the cascade reachesNo provider availablefor the life of the process.Availability only — no disclosure or modification of data. Rated
AC:Hbecause success depends on the upstream provider echoing request content in its error bodies: common among OpenAI-compatible gateways, but a property of the vendor rather than of this crate.Fixed in
0.1.9. Detection now reads only the machine-writtenerror.code/error.typefields against a fixed set of vendor codes, never the message prose, and skips bodies that are not JSON. The park became a bounded 15-minute cooldown rather than a permanent latch.Versions.
patched = [">= 0.1.9"],unaffected = ["< 0.1.6"]. The lower bound was taken from the published tarballs rather than from git history —0.1.5carries only a pre-existing401/403latch, which keys on a status code rather than on caller-influenced text, so it is not vulnerable.0.1.6,0.1.7and0.1.8have been yanked.rustsec-admin lintreportsLinted okfor this file.Full write-up: https://github.com/ilPez00/physis-pro/blob/main/security/advisories/physis_pro-provider-availability.md