[Security] Add missing blob-size assertion in ReadAllToBF16 weight loader#924
Merged
copybara-service[bot] merged 2 commits intoJun 2, 2026
Conversation
…ader MapAll (weights.cc:557) and MakeBatches (weights.cc:645) both assert that a tensor blob's on-disk size equals the buffer size derived from the tensor's declared shape (range.bytes == mat.PackedBytes()). ReadAllToBF16 was the only load mode missing this check. Without it, a crafted .sbs weight file with correct Rows/Cols metadata but an inflated blob bytes field passes all structural validation and reaches the keep_type branch, where tensor.range.bytes bytes are written into a mat.PackedBytes()-sized buffer — a heap OOB write with attacker- controlled content. The DecompressToBF16 branch has a sibling OOB read when the blob is undersized. Add the same HWY_ASSERT_M at the top of the per-tensor lambda, before either read path executes, mirroring the existing checks in MapAll and MakeBatches.
Member
|
Thanks for making the fix :) We're only able to merge on the dev branch, would you mind retargeting to that? |
Author
Done! |
jan-wassenberg
approved these changes
Jun 2, 2026
jan-wassenberg
added
the
copybara-import
Author
|
Thanks! Can it be closed? |
Member
|
FYI the logic here is a bit off. We only want to check this if keep_type == true, i.e. move the code down by one line. I'll fix that soon. |
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
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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
ReadAllToBF16inweights.ccis the only one of three weight-load modes missing therange.bytes == mat.PackedBytes()assertion. The other two modes (MapAllat line ~557,MakeBatchesat line ~645) both assert this invariant. Without the check, a crafted.sbsfile with correctRows/Colsmetadata but an inflated blobbytesfield passes all structural validation and causes a heap OOB write (keep_typebranch) or OOB read (DecompressToBF16branch).Fix
Add
HWY_ASSERT_M(tensor.range.bytes == mat.PackedBytes(), mat.Name())at the top of the per-tensor lambda inReadAllToBF16, before either read path executes — mirroring the existing checks inMapAllandMakeBatches.Test plan
.sbsfile with inflated blobbytesfor one tensor is rejected with an assertion failure inkReadBF16mode (--to_bf16ordecode_qbatch_size >= 128)