Submit #1016280: R6410418 Jackrong-llm-finetuning-guide commit ef2b17f2798c Path Traversalinfo

TitleR6410418 Jackrong-llm-finetuning-guide commit ef2b17f2798c Path Traversal
DescriptionA vulnerability was found in Jackrong-llm-finetuning-guide commit ef2b17f2798c and classified as medium 5.4. Affected is the validate_indexed_tensors function of the qwen-mtp-gguf pipeline component. The manipulation of the argument weight_map in model.safetensors.index.json leads to a path traversal. It is possible to launch the attack remotely. User interaction is required because the operator must run the conversion pipeline on a model repo prepared by the attacker. Technical Details - Affected file/function: qwen-mtp-gguf/scripts/qwen_mtp_gguf_pipeline.py / validate_indexed_tensors() (lines 544-559), reached from model_has_valid_mtp() and inject_mtp_index() - Vulnerable parameter: weight_map values of model.safetensors.index.json - Attack vector: Network - Privileges required: None - Trigger condition: the pipeline joins each weight_map value onto the model directory (shard = model_dir / filename at line 552) and opens it with safetensors.safe_open; an absolute value discards the model-dir prefix, so the pipeline opens and validates files at arbitrary absolute paths outside the model package, and line 554 echoes the joined path in FileNotFoundError, giving an existence/format/key-name oracle for arbitrary local paths Impact - Confidentiality: Low (existence, safetensors parseability and tensor key names of arbitrary local files are disclosed; tensor values are not exported at this sink) - Integrity: Low (the attacker-written index passes the pipeline validation gate and enters downstream conversion stages) - Availability: None CVSS v3.1 Score: 5.4 (Medium) Vector: CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:U/C:L/I:L/A:N Timeline - Discovered: 2026-10-05 - Vendor notified: [pending] - Patch released: [pending] - Public disclosure: [pending] Countermeasure Reject absolute paths and `..` segments in index-supplied shard names, verify resolved-path containment (is_relative_to the model directory) before opening, and apply the same check at the sibling join in extract_mtp_heads() (line 610).
Source⚠️ https://github.com/R6410418/Jackrong-llm-finetuning-guide/issues/15
User
 Dem0000000 (UID 98743)
Submission10/06/2026 18:04 (14 days ago)
Moderation10/20/2026 18:28 (14 days later)
StatusAccepted
VulDB entry422909 [R6410418 Jackrong-llm-finetuning-guide up to ef2b17f2798cd18cd85375080c35a58397ad0d09 qwen_mtp_gguf_pipeline.py validate_indexed_tensors weight_map path traversal]
Points20

Do you need the next level of professionalism?

Upgrade your account now!