Update onnx upper bound to allow 1.21.0 (CVE-2026-27489)#29
Open
wjhrdy wants to merge 1 commit into
Open
Conversation
onnx < 1.21.0 is affected by CVE-2026-27489 (CVSS 8.7 HIGH), a path traversal vulnerability via symlink in external data loading. Relax the upper bound from <=1.19.0 to <=1.21.0 so downstream users can install the patched version. Fixes amd#28 Signed-off-by: Will Hardy <whardy@redhat.com>
Deepali1999
pushed a commit
to Deepali1999/vllm-cpu
that referenced
this pull request
Jul 7, 2026
This commit removes amd-quark>=0.8.99 from requirements/rocm.txt to allow upgrading onnx from 1.19.0 to 1.21.0, which is required to fix CVE-2026-27489. ## Problem amd-quark 0.11.2 (latest version) pins onnx<=1.19.0, creating an impossible constraint conflict: - amd-quark requires: onnx<=1.19.0,>=1.16.0 - CVE fix requires: onnx>=1.21.0 - No version of onnx satisfies both constraints Build failure: ``` ERROR: Unable to resolve requirement specifier onnx<=1.19.0,>=1.16.0 with constraint onnx==1.21.0 ``` ## Evidence that amd-quark is not a hard dependency 1. **Already validated in rhai/0.18.0 branch:** - Commit: f178b3e92fb493b05ad7770c04ee72b5cc19edaa - PR red-hat-data-services#469: "Remove amd-quark from ROCm requirements for onnx CVE fix" - Full accept-sync run 24567391901: ALL TESTS PASSED * Build ✅ * Image ✅ * LM-EVAL (solo + duo) ✅ * GuideLLM (6 models including 2x FP8-dynamic) ✅ * Quantized Models test partition ✅ 2. **Only MXFP4/MXFP6 quantized models affected:** - FP8 W8A8 models use vLLM native Fp8Config, not QuarkConfig - Standard models completely unaffected - Only affects specific quantization formats (MXFP4/MXFP6) 3. **Temporary removal until upstream fix:** - Upstream PR pending: amd/Quark#29 - Will re-add once amd/Quark bumps their onnx pin - This is a temporary workaround to unblock the CVE fix ## Related Tickets - INFERENG-7300: Update onnx to 1.21.0 in ROCm variant to fix CVE - INFERENG-6173: Original tracker for amd-quark removal - AIPCC-14368: AIPCC pipeline onnx CVE fix ## Downstream Impact This change enables pipeline to upgrade onnx to 1.21.0 without constraint conflicts. RHAIIS 3.5-EA1 uses vLLM 0.19.1 from main branch, so removing amd-quark here unblocks the downstream fix. Signed-off-by: Tahmid Muttaki <tmuttaki@redhat.com>
Deepali1999
pushed a commit
to Deepali1999/vllm-cpu
that referenced
this pull request
Jul 7, 2026
…hat-data-services#522) ## Summary Removes `amd-quark>=0.8.99` from `requirements/rocm.txt` to unblock the onnx upgrade required for CVE-2026-27489 fix. ## Problem **Constraint conflict preventing onnx CVE fix:** ``` ERROR: Unable to resolve requirement specifier onnx<=1.19.0,>=1.16.0 with constraint onnx==1.21.0 ``` - **amd-quark 0.11.2** (latest) requires: `onnx<=1.19.0,>=1.16.0` - **CVE fix** requires: `onnx>=1.21.0` - **No version of onnx** satisfies both constraints ## Evidence that amd-quark is NOT a hard dependency ### ✅ Already validated in rhai/0.18.0 branch - **Commit:** `f178b3e92fb493b05ad7770c04ee72b5cc19edaa` - **PR:** red-hat-data-services#469 - "Remove amd-quark from ROCm requirements for onnx CVE fix (rhai/0.18.0)" - **Validation run:** [24567391901](https://github.com/neuralmagic/nm-cicd/actions/runs/24567391901) **All tests passed:** - ✅ Build - ✅ Image - ✅ LM-EVAL (solo + duo) - ✅ GuideLLM (6 models including 2x FP8-dynamic) - ✅ Quantized Models test partition ### 📋 Impact Assessment **Not affected:** - ✅ FP8 W8A8 models (use vLLM native `Fp8Config`, not `QuarkConfig`) - ✅ Standard unquantized models - ✅ All standard quantization formats **Affected (limited scope):** -⚠️ MXFP4/MXFP6 quantized models only ### 🔄 Temporary Removal This is a **temporary removal** until upstream fixes the onnx pin: - **Upstream PR:** amd/Quark#29 - **Re-add when:** amd/Quark bumps their onnx constraint - **Tracking:** Will monitor upstream PR and re-enable once available ## Related Tickets - **INFERENG-7300:** Update onnx to 1.21.0 in ROCm variant to fix CVE - **INFERENG-6173:** Original tracker for amd-quark removal validation - **AIPCC-14368:** AIPCC pipeline onnx CVE fix ## Downstream Impact **RHAIIS 3.5-EA1** uses vLLM 0.19.1 from the main branch. This change unblocks: 1. Pipeline MR to upgrade onnx==1.21.0 in ROCm constraints 2. Downstream CVE remediation for RHAIIS releases 3. Container image security scans passing ## Testing Plan - [ ] Run accept-sync workflow on this branch - [ ] Verify build completes successfully - [ ] Confirm lm-eval tests pass - [ ] Validate guidellm benchmarks - [ ] Check quantized model tests (excluding MXFP4/MXFP6) --- 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
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
1 participant
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.
Summary
Relax the onnx version constraint in
requirements.txtfrom<=1.19.0to<=1.21.0to allow installation of onnx 1.21.0, which fixes CVE-2026-27489 (CVSS 8.7 HIGH — path traversal via symlink in external data loading).Changes
Context
Downstream projects that depend on Quark inherit this constraint and are currently unable to upgrade to the patched onnx version.
Fixes #28