fuzz(flatbuf): add fuzz harness for FunctionCall/FunctionCallResult decode by Devendrakumarsinha · Pull Request #1878 · hyperlight-dev/hyperlight

Devendrakumarsinha · 2026-10-04T05:36:21Z

Summary

Adds a libfuzzer-based fuzz target that directly exercises the FlatBuffers
decode path for FunctionCall and FunctionCallResult — the host-side
parsing of guest-controlled wire data arriving via the G2H virtqueue.

Motivation

Existing fuzzers (guest_call, host_call) exercise these decode paths
only indirectly through a full sandbox with a running hypervisor. This
harness isolates the parsing logic without requiring KVM or Hyper-V,
enabling faster iteration and broader input coverage.

Design

BoundedSource mirrors RecvChain semantics: take_bytes/take_chunks
reject reads beyond available data, matching production behavior and
avoiding false OOM positives from unbounded test allocators.

Results

Coverage: ~960 branch points
exec/s: ~110,000 (no hypervisor overhead)
No crashes found

Testing

cargo +nightly fuzz run flatbuf_decode -- -max_len=65536

Copilot

🟡 Changes recommended

The harness does not currently reproduce production control and external-payload boundaries or trailing-byte validation.

3 open findings

What changed in this PR

Adds a fast fuzz target for FlatBuffers function call decoding without a hypervisor.

Changes:

Adds bounded external-value fuzzing.
Registers the fuzz target and dependencies.
Updates fuzz-generated file exclusions.

File	Description
`fuzz/fuzz_targets/flatbuf_decode.rs`	Implements the decode harness.
`fuzz/Cargo.toml`	Registers the target and dependencies.
`fuzz/.gitignore`	Updates fuzz output exclusions.

🧠 Review effort: Balanced

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

andreiltd · 2026-10-12T16:48:16Z

Hey @Devendrakumarsinha , thanks for the contribution! This looks very reasonable although those few comments from copilot about mirroring production code is worth addressing 🙂

Devendrakumarsinha · 2026-10-13T17:03:58Z

Thanks @andreiltd ,addressed all three, mirroring the production G2H path in read_control:

Split control/external ; reads the 4-byte LE size prefix, takes SIZE_PREFIX_LEN + payload_len as the size-prefixed control FlatBuffer, and passes only the trailing bytes to BoundedSource as the external payload (via transport::{SIZE_PREFIX_LEN, size_prefix_payload_len}). Invalid layouts are rejected before decoding, like read_control.
finish rejects trailing external bytes ; mirrors finish_external_values.
coverage/ restored in .gitignore.

…ecode Add a libfuzzer-based fuzz target that directly exercises the FlatBuffers decode path for FunctionCall and FunctionCallResult — the host-side parsing of guest-controlled wire data arriving via the G2H virtqueue. Previously these decode paths had no direct fuzz coverage; existing fuzzers (guest_call, host_call) exercise them only indirectly through a full sandbox. This harness isolates the parsing logic without requiring a hypervisor. Uses a BoundedSource that mirrors RecvChain semantics: take_bytes/take_chunks reject reads beyond available data, matching production behavior. Initial run (~1M iterations): - Coverage: ~960 branch points - exec/s: ~110,000 - No crashes found Signed-off-by: Devendra Kumar Sinha <devendrasinha.me@gmail.com>

andreiltd · 2026-10-14T10:57:06Z

Hey, thanks! This looks very close: lock file need to be updated, also CI need to pick up new fuzzing target, so it need to be added to: ValidatePullRequest.yml, Fuzzing.yml and Justfile.

hyperlight-gh-bot · 2026-10-14T11:29:35Z

Benchmark Results

Measured commit: fd33f164c175
Baseline commit: 7d244de79713

kvm / amd (Linux) (➖ stable)

No benchmark improved or regressed.

Benchmark Results

function_call_codec

	`encode_control`	`decode_vec_bytes_copy`
`byte_chunks`	`753.14 ns` (➖ 1.04x faster)
`vec_bytes`	`589.87 ns` (➖ 1.00x slower)
		`373.55 µs` (➖ 1.00x faster)

payload_allocation

	`slot_pool_segmented`
`262144`	`524.34 ns` (➖ 1.00x faster)
`65536`	`149.34 ns` (➖ 1.05x slower)

sandboxes

	`create_initialized_and_drop`
`medium`	`82.19 ms` (➖ 1.01x slower)

slot_pool

	`alloc_dealloc_1500`	`alloc_dealloc_4096`	`alloc_dealloc_128`
	`7.81 ns` (➖ 1.00x slower)	`7.71 ns` (➖ 1.01x faster)	`7.74 ns` (➖ 1.00x faster)

snapshot_files

	`load_snapshot_unverified`
`small`	`92.49 µs` (➖ 1.01x faster)

virtq_readonly

	`slot_pool_segmented_fragmented`	`slot_pool_segmented`
`262144`	`7.54 µs` (➖ 1.06x slower)	`7.12 µs` (➖ 1.00x slower)
`65536`	`2.06 µs` (➖ 1.06x slower)	`1.97 µs` (➖ 1.07x faster)

virtq_readwrite

	`slot_pool_segmented_fragmented`	`slot_pool_segmented`
`65536`	`6.32 µs` (➖ 1.01x faster)	`6.34 µs` (➖ 1.00x slower)
`8192`	`1.11 µs` (➖ 1.07x slower)	`1.08 µs` (➖ 1.05x slower)
`262144`		`27.19 µs` (➖ 1.06x faster)

kvm / intel (Linux) (➖ stable)

No benchmark improved or regressed.

Benchmark Results

function_call_codec

	`encode_control`	`decode_vec_bytes_copy`
`byte_chunks`	`663.09 ns` (➖ 1.03x faster)
`vec_bytes`	`495.76 ns` (➖ 1.05x faster)
		`668.53 µs` (➖ 1.00x slower)

payload_allocation

	`slot_pool_segmented`
`262144`	`497.30 ns` (➖ 1.10x faster)
`65536`	`134.63 ns` (➖ 1.03x faster)

sandboxes

	`create_initialized_and_drop`
`medium`	`73.05 ms` (➖ 1.05x faster)

slot_pool

	`alloc_dealloc_1500`	`alloc_dealloc_4096`	`alloc_dealloc_128`
	`6.97 ns` (➖ 1.01x slower)	`6.91 ns` (➖ 1.00x slower)	`6.89 ns` (➖ 1.01x faster)

snapshot_files

	`load_snapshot_unverified`
`small`	`45.23 µs` (➖ 1.01x slower)

virtq_readonly

	`slot_pool_segmented_fragmented`	`slot_pool_segmented`
`262144`	`7.66 µs` (➖ 1.02x slower)	`7.60 µs` (➖ 1.01x slower)
`65536`	`2.15 µs` (➖ 1.01x slower)	`2.14 µs` (➖ 1.02x slower)

virtq_readwrite

	`slot_pool_segmented_fragmented`	`slot_pool_segmented`
`65536`	`7.10 µs` (➖ 1.00x slower)	`7.08 µs` (➖ 1.01x faster)
`8192`	`773.16 ns` (➖ 1.04x slower)	`754.39 ns` (➖ 1.06x slower)
`262144`		`30.16 µs` (➖ 1.04x slower)

mshv3 / amd (Linux) (➖ stable)

No benchmark improved or regressed.

Benchmark Results

function_call_codec

	`encode_control`	`decode_vec_bytes_copy`
`byte_chunks`	`948.92 ns` (➖ 1.02x slower)
`vec_bytes`	`720.44 ns` (➖ 1.04x slower)
		`318.99 µs` (➖ 1.18x slower)

payload_allocation

	`slot_pool_segmented`
`262144`	`718.19 ns` (➖ 1.02x faster)
`65536`	`197.15 ns` (➖ 1.02x slower)

sandboxes

	`create_initialized_and_drop`
`medium`	`60.60 ms` (➖ 1.00x faster)

slot_pool

	`alloc_dealloc_1500`	`alloc_dealloc_4096`	`alloc_dealloc_128`
	`9.44 ns` (➖ 1.03x faster)	`9.66 ns` (➖ 1.01x faster)	`9.50 ns` (➖ 1.02x faster)

snapshot_files

	`load_snapshot_unverified`
`small`	`83.35 µs` (➖ 1.01x slower)

virtq_readonly

	`slot_pool_segmented_fragmented`	`slot_pool_segmented`
`262144`	`10.03 µs` (➖ 1.01x slower)	`9.77 µs` (➖ 1.04x faster)
`65536`	`2.39 µs` (➖ 1.04x faster)	`2.24 µs` (➖ 1.01x slower)

virtq_readwrite

	`slot_pool_segmented_fragmented`	`slot_pool_segmented`
`65536`	`8.59 µs` (➖ 1.13x faster)	`8.39 µs` (➖ 1.15x faster)
`8192`	`1.30 µs` (➖ 1.07x faster)	`1.39 µs` (➖ 1.14x slower)
`262144`		`36.84 µs` (➖ 1.03x faster)

mshv3 / intel (Linux) (➖ stable)

No benchmark improved or regressed.

Benchmark Results

function_call_codec

	`encode_control`	`decode_vec_bytes_copy`
`byte_chunks`	`941.76 ns` (➖ 1.02x slower)
`vec_bytes`	`695.81 ns` (➖ 1.07x slower)
		`774.76 µs` (➖ 1.02x faster)

payload_allocation

	`slot_pool_segmented`
`262144`	`656.77 ns` (➖ 1.04x slower)
`65536`	`168.82 ns` (➖ 1.01x slower)

sandboxes

	`create_initialized_and_drop`
`medium`	`68.92 ms` (➖ 1.03x slower)

slot_pool

	`alloc_dealloc_1500`	`alloc_dealloc_4096`	`alloc_dealloc_128`
	`10.07 ns` (➖ 1.14x slower)	`9.28 ns` (➖ 1.06x slower)	`9.22 ns` (➖ 1.05x slower)

snapshot_files

	`load_snapshot_unverified`
`small`	`46.19 µs` (➖ 1.01x slower)

virtq_readonly

	`slot_pool_segmented_fragmented`	`slot_pool_segmented`
`262144`	`7.99 µs` (➖ 1.01x slower)	`7.92 µs` (➖ 1.02x faster)
`65536`	`2.27 µs` (➖ 1.00x faster)	`2.26 µs` (➖ 1.00x slower)

virtq_readwrite

	`slot_pool_segmented_fragmented`	`slot_pool_segmented`
`65536`	`7.70 µs` (➖ 1.02x slower)	`7.49 µs` (➖ 1.01x faster)
`8192`	`919.83 ns` (➖ 1.02x faster)	`930.84 ns` (➖ 1.02x slower)
`262144`		`38.26 µs` (➖ 1.14x faster)

hyperv-ws2025 / amd (Windows) (➖ stable)

No benchmark improved or regressed.

Benchmark Results

function_call_codec

	`encode_control`	`decode_vec_bytes_copy`
`byte_chunks`	`1.12 µs` (➖ 1.03x faster)
`vec_bytes`	`760.97 ns` (➖ 1.01x slower)
		`2.15 ms` (➖ 1.04x faster)

payload_allocation

	`slot_pool_segmented`
`262144`	`822.97 ns` (➖ 1.31x faster)
`65536`	`233.43 ns` (➖ 1.26x faster)

sandboxes

	`create_initialized_and_drop`
`medium`	`77.31 ms` (➖ 1.06x faster)

slot_pool

	`alloc_dealloc_1500`	`alloc_dealloc_4096`	`alloc_dealloc_128`
	`10.88 ns` (➖ 1.02x slower)	`11.07 ns` (➖ 1.02x faster)	`10.89 ns` (➖ 1.01x slower)

snapshot_files

	`load_snapshot_unverified`
`small`	`766.39 µs` (➖ 1.13x slower)

virtq_readonly

	`slot_pool_segmented_fragmented`	`slot_pool_segmented`
`262144`	`8.57 µs` (➖ 1.02x faster)	`8.58 µs` (➖ 1.00x faster)
`65536`	`2.54 µs` (➖ 1.00x slower)	`2.54 µs` (➖ 1.00x slower)

virtq_readwrite

	`slot_pool_segmented_fragmented`	`slot_pool_segmented`
`65536`	`8.75 µs` (➖ 1.01x faster)	`9.51 µs` (➖ 1.08x slower)
`8192`	`1.41 µs` (➖ 1.04x faster)	`1.43 µs` (➖ 1.05x slower)
`262144`		`39.25 µs` (➖ 1.01x slower)

hyperv-ws2025 / intel (Windows) (➖ stable)

No benchmark improved or regressed.

Benchmark Results

function_call_codec

	`encode_control`	`decode_vec_bytes_copy`
`byte_chunks`	`1.28 µs` (➖ 1.05x slower)
`vec_bytes`	`770.69 ns` (➖ 1.04x faster)
		`2.95 ms` (➖ 1.03x faster)

payload_allocation

	`slot_pool_segmented`
`262144`	`737.13 ns` (➖ 1.05x faster)
`65536`	`213.41 ns` (➖ 1.08x faster)

sandboxes

	`create_initialized_and_drop`
`medium`	`101.93 ms` (➖ 1.14x faster)

slot_pool

	`alloc_dealloc_1500`	`alloc_dealloc_4096`	`alloc_dealloc_128`
	`10.31 ns` (➖ 1.01x slower)	`9.98 ns` (➖ 1.01x faster)	`9.89 ns` (➖ 1.07x faster)

snapshot_files

	`load_snapshot_unverified`
`small`	`532.61 µs` (➖ 1.10x faster)

virtq_readonly

	`slot_pool_segmented_fragmented`	`slot_pool_segmented`
`262144`	`7.74 µs` (➖ 1.02x faster)	`7.75 µs` (➖ 1.06x faster)
`65536`	`2.33 µs` (➖ 1.04x faster)	`2.37 µs` (➖ 1.03x faster)

virtq_readwrite

	`slot_pool_segmented_fragmented`	`slot_pool_segmented`
`65536`	`8.75 µs` (➖ 1.02x slower)	`8.43 µs` (➖ 1.03x faster)
`8192`	`1.25 µs` (➖ 1.12x faster)	`1.09 µs` (➖ 1.26x faster)
`262144`		`41.15 µs` (➖ 1.01x faster)

_{Reported by cargo ci bench-report --candidate run:38438487611 --config-file bench_report.toml.}

Devendrakumarsinha requested review from andreiltd, danbugs, dblnz, devigned, jprendes, jsturtevant, ludfjig, simongdavies, squillace and syntactically as code owners October 4, 2026 05:36

Devendrakumarsinha force-pushed the fuzz/flatbuf-decode-harness branch 2 times, most recently from 5d2b524 to 4c0441d Compare October 4, 2026 06:05

andreiltd requested a balanced review from Copilot October 12, 2026 16:27

Copilot started reviewing on behalf of andreiltd October 12, 2026 16:28 View session

Copilot AI reviewed Oct 12, 2026

View reviewed changes

Comment thread fuzz/fuzz_targets/flatbuf_decode.rs

Comment thread fuzz/fuzz_targets/flatbuf_decode.rs

Comment thread fuzz/.gitignore

This comment has been minimized.

Sign in to view

Devendrakumarsinha force-pushed the fuzz/flatbuf-decode-harness branch from 4c0441d to b32d581 Compare October 13, 2026 16:58

Devendrakumarsinha force-pushed the fuzz/flatbuf-decode-harness branch from b32d581 to fd33f16 Compare October 13, 2026 17:25

andreiltd added kind/enhancement For PRs adding features, improving functionality, docs, tests, etc. area/fuzzing Fuzzing related test failures created from weekly CI labels Oct 14, 2026

Conversation

Devendrakumarsinha commented Oct 4, 2026

Summary

Motivation

Design

Results

Testing

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

🟡 Changes recommended

Uh oh!

Uh oh!

Uh oh!

Uh oh!

andreiltd commented Oct 12, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

This comment has been minimized.

Devendrakumarsinha commented Oct 13, 2026

Uh oh!

andreiltd commented Oct 14, 2026

Uh oh!

hyperlight-gh-bot Bot commented Oct 14, 2026

Benchmark Results

Benchmark Results

function_call_codec

payload_allocation

sandboxes

slot_pool

snapshot_files

virtq_readonly

virtq_readwrite

Benchmark Results

function_call_codec

payload_allocation

sandboxes

slot_pool

snapshot_files

virtq_readonly

virtq_readwrite

Benchmark Results

function_call_codec

payload_allocation

sandboxes

slot_pool

snapshot_files

virtq_readonly

virtq_readwrite

Benchmark Results

function_call_codec

payload_allocation

sandboxes

slot_pool

snapshot_files

virtq_readonly

virtq_readwrite

Benchmark Results

function_call_codec

payload_allocation

sandboxes

slot_pool

snapshot_files

virtq_readonly

virtq_readwrite

Benchmark Results

function_call_codec

payload_allocation

sandboxes

slot_pool

snapshot_files

virtq_readonly

virtq_readwrite

Uh oh!

This branch has not been deployed

Reviewers

Assignees

Labels

Projects

Milestone

andreiltd commented Oct 12, 2026 •

edited

Loading