SEA: BlobDeserializer SIGSEGVs when fuse byte is set but no NODE_SEA_BLOB is present

src/node_sea_bin.cc の node::sea::FindSingleExecutableBlob() から開始し、BlobDeserializer が実行される前に postject_find_resource("NODE_SEA_BLOB", ...) の結果を追跡します。提供されている fuse-flipping コマンドで再現し、blob がない場合に SIGSEGV ではなく、明確な起動エラーが発生して終了コードが 0 以外になることを確認します。

Version

v26.1.0 (also reproduces on v25.6.0)

Platform

Linux arm64 (reproduced on Apple Silicon via Docker Desktop, but the SEGV is platform-independent)

Subsystem

sea

What steps will reproduce the bug?

Binaries where the postject fuse byte is set to 1 but NODE_SEA_BLOB cannot be located at runtime currently die with a NULL-deref SIGSEGV inside BlobDeserializer::ReadArithmetic, with no error message indicating the cause.

Take any Node binary, flip the fuse byte from 0 to 1 without injecting an actual SEA blob:

python3 -c "
sent = b'NODE_SEA_FUSE_fce680ab2cc467b6e072b8b5df1996b2'
with open('hello','rb') as f: buf = bytearray(f.read())
i = buf.find(sent)
buf[i + len(sent) + 1] = ord('1')
with open('hello','wb') as f: f.write(bytes(buf))
"
chmod +x hello
./hello --version    # → Segmentation fault, exit 139

This state arises naturally when postject is run against a host binary with no PT_NOTE program header — postject silently fails to inject the note but still flips the fuse byte. See https://github.com/nodejs/postject/issues/107 and https://github.com/nodejs/unofficial-builds/issues/200.

How often does it reproduce? Is there a required condition?

100% reproducible. Required condition: fuse byte set to 1 AND no NODE_SEA_BLOB discoverable via postject_find_resource().

What is the expected behavior? Why is that the expected behavior?

A clear error indicating that the SEA fuse is set but no blob is present, rather than a bare SIGSEGV at startup. The current behavior makes it look like a crash in OpenSSL or libc (because the SIGILLs from OpenSSL's ARM crypto-extension probes show up first under gdb), when the actual cause is much earlier and recoverable.

What do you see instead?

Program received signal SIGSEGV, Segmentation fault.
#0  memcpy ()
#1  node::BlobDeserializer<...>::ReadArithmetic<unsigned int>()
#2  node::sea::FindSingleExecutableResource()
#3  node::sea::FixupArgsForSEA(int, char**)
#4  node::Start(int, char**)

postject_find_resource("NODE_SEA_BLOB", &size, ...) returns NULL, then BlobDeserializer::ReadArithmetic calls memcpy(dst, NULL, sizeof(uint32_t)) → SIGSEGV.

Additional information

Suggested fix in node::sea::FindSingleExecutableBlob() (src/node_sea_bin.cc) — guard the deserialization on the resource lookup:

const char* blob = static_cast<const char*>(
    postject_find_resource("NODE_SEA_BLOB", &size, ...));
if (blob == nullptr) {
    fprintf(stderr,
            "node: SEA fuse is set but no NODE_SEA_BLOB resource was found "
            "in this binary. The host binary may be missing a PT_NOTE program "
            "header (run `readelf -lW <binary> | grep NOTE` to check).\n");
    exit(static_cast<int>(node::ExitCode::kGenericUserError));
}

Either that or CHECK_NOT_NULL(blob) — anything that surfaces a cause rather than a bare SEGV.

Related:

• https://github.com/nodejs/unofficial-builds/issues/200 — upstream cause (arm64-musl tarball with no PT_NOTE)
• https://github.com/nodejs/unofficial-builds/pull/233 — fix for that root cause
• https://github.com/nodejs/postject/issues/107 — companion postject issue (silent injection failure)