From a3a7922243d09225db7b33fcb554e8e527baa280 Mon Sep 17 00:00:00 2001 From: ShuaixinHuang <1836656514@qq.com> Date: Thu, 19 Nov 2026 19:09:30 +0800 Subject: [PATCH] perf(ntfs): skip long zero runs in offline USN journals (#12274) Co-authored-by: ShuaixinHuang <197200901+ShuaixinHuang@users.noreply.github.com> --- src/domain/windows/ntfsUsnJournal.test.ts | 42 +++++++++++++++++++++++ src/domain/windows/ntfsUsnJournal.ts | 26 +++++++++++--- 2 files changed, 64 insertions(+), 4 deletions(-) diff --git a/src/domain/windows/ntfsUsnJournal.test.ts b/src/domain/windows/ntfsUsnJournal.test.ts index db3e7bc5eb..5f2fb1d537 100644 --- a/src/domain/windows/ntfsUsnJournal.test.ts +++ b/src/domain/windows/ntfsUsnJournal.test.ts @@ -44,6 +44,48 @@ it("preserves full references, flags, offsets and UTF-16 names after sparse padd ]); }); +it.each([2 * 64 * 1024, 2 * 64 * 1024 + 8, 2 * 64 * 1024 + 24])( + "keeps the exact record boundary after a %i-byte zero run on a resumed page", + (paddingBytes) => { + const startFileOffset = 72; + const recordOffset = startFileOffset + paddingBytes; + const bytes = Buffer.concat([ + journalRecord(), + Buffer.alloc(paddingBytes), + journalRecord(), + Buffer.alloc(3), + ]); + expect([...readUsnJournal(bytes, startFileOffset)]).toMatchObject([ + { + ok: false, + error: { + kind: "padding", + fileOffset: startFileOffset, + bytes: paddingBytes, + }, + }, + { ok: true, value: { fileOffset: recordOffset, name: "a😃b" } }, + { + ok: false, + error: { kind: "truncated", fileOffset: recordOffset + 72 }, + }, + ]); + }, +); + +it("does not skip a malformed header with a zero length and nonzero version", () => { + const paddingBytes = 2 * 64 * 1024 + 24; + const bytes = Buffer.alloc(paddingBytes + 8); + bytes.writeUInt16LE(2, paddingBytes + 4); + expect([...readUsnJournal(bytes)]).toEqual([ + { + ok: false, + error: { kind: "padding", fileOffset: 0, bytes: paddingBytes }, + }, + { ok: false, error: { kind: "malformed", fileOffset: paddingBytes } }, + ]); +}); + it("uses a validated record length to skip unsupported and malformed records", () => { const unsupported = journalRecord(); unsupported.writeUInt16LE(9, 4); diff --git a/src/domain/windows/ntfsUsnJournal.ts b/src/domain/windows/ntfsUsnJournal.ts index b3f7a0545c..b3a942c2cc 100644 --- a/src/domain/windows/ntfsUsnJournal.ts +++ b/src/domain/windows/ntfsUsnJournal.ts @@ -2,6 +2,7 @@ import { BinaryReader } from "../binaryContainer.js"; import { err, ok, type Result } from "../result.js"; const COMMON_HEADER_BYTES = 8; +const ZERO_PADDING_BLOCK = Buffer.alloc(64 * 1024); /** Bound extent objects in each inline event independently of source record length. */ export const USN_INLINE_EXTENT_LIMIT = 16; @@ -55,12 +56,29 @@ export function* readUsnJournal( const length = journal.u32(fileOffset); if (length === 0 && journal.u32(fileOffset + 4) === 0) { const start = fileOffset; - do { - fileOffset += COMMON_HEADER_BYTES; - } while ( + fileOffset += COMMON_HEADER_BYTES; + if ( fileOffset + COMMON_HEADER_BYTES <= bytes.length && journal.word(fileOffset, 64) === 0n - ); + ) { + fileOffset += COMMON_HEADER_BYTES; + // Compare long zero runs in native code, then resolve the exact end + // on record boundaries without consuming a short trailing header. + while ( + ZERO_PADDING_BLOCK.length <= bytes.length - fileOffset && + ZERO_PADDING_BLOCK.compare( + bytes, + fileOffset, + fileOffset + ZERO_PADDING_BLOCK.length, + ) === 0 + ) + fileOffset += ZERO_PADDING_BLOCK.length; + while ( + fileOffset + COMMON_HEADER_BYTES <= bytes.length && + journal.word(fileOffset, 64) === 0n + ) + fileOffset += COMMON_HEADER_BYTES; + } yield err({ kind: "padding", fileOffset: start,