Performing Sqlite Database Forensics

Performing SQLite Database Forensics

Overview

SQLite is the most widely deployed database engine in the world, used by virtually every mobile application, web browser, and many desktop applications to store user data. In digital forensics, SQLite databases are critical evidence sources containing browser history, messaging records, call logs, GPS locations, application preferences, and cached content. Forensic analysis goes beyond simple SQL queries to examine the internal B-tree page structures, freelist pages containing deleted records, Write-Ahead Log (WAL) files preserving transaction history, and unallocated space within database pages where recoverable data may persist after deletion.

When to Use

• When conducting security assessments that involve performing sqlite database forensics
• When following incident response procedures for related security events
• When performing scheduled security testing or auditing activities
• When validating security controls through hands-on testing

Detection Gaps & Validation

A plain SELECT * on the live table is the shallowest possible SQLite exam and misses most recoverable evidence. Cover these or you will under-report:

• The deleted rows live outside the main tables. Querying the database with sqlite3 shows only active records. Deleted data persists in freelist pages, in unallocated space between the cell-pointer array and cell-content area, and in overflow pages. Parse these raw (as the freelist/slack code here does) — "no such message" from a SQL query is not proof it was never there.
• WAL and journal hold uncommitted/rolled-back state. A -wal file can contain newer rows not yet checkpointed into the main DB, and older versions of pages superseded by later frames; a -journal holds pre-transaction images. NEVER open the DB read-write (it triggers a checkpoint and destroys this evidence) — copy db, -wal, -shm, and -journal together, and parse WAL frames page-by-page for multiple versions of the same row.
• Secure-delete and VACUUM erase the freelist. If PRAGMA secure_delete was on, or the app ran VACUUM/auto-vacuum, deleted-record recovery yields little — note this as a coverage limitation, not a clean device. Carve surrounding unallocated disk space and check for prior DB copies/backups.
• Validate timestamps by epoch, then corroborate. Decode against the correct base: Chrome/WebKit (µs since 1601), Mozilla PRTime (µs since 1970), Unix (s/ms), Mac Absolute (s since 2001). A mis-decoded timestamp can shift events by decades. Cross-check recovered rows against _shm indices, app logs, and a second artifact before relying on them.
• Interpretation false positives. Carved freelist fragments may be stale/overwritten and mix bytes from unrelated records — a "recovered URL" can be a Frankenstein of two rows. Browser-history existence shows a page was rendered/visited, not that the user deliberately typed it (prefetch, redirects, ads). Confirm partial recoveries against record structure (valid serial types, rowid) before reporting them as fact.

Prerequisites

• DB Browser for SQLite (sqlitebrowser)
• SQLite command-line tools (sqlite3)
• Python 3.8+ with sqlite3 module
• Belkasoft Evidence Center or Axiom (commercial)
• Hex editor (HxD, 010 Editor) for manual page inspection
• Understanding of B-tree data structures

SQLite Internal Structure

Database Header (First 100 Bytes)

Page Types

Deleted Record Recovery

Method 1: Freelist Page Analysis

When records are deleted, SQLite may place their pages on the freelist rather than overwriting them immediately.

import struct
import sqlite3
import os


def analyze_freelist(db_path: str) -> dict:
    """Analyze SQLite freelist to identify pages containing deleted data."""
    with open(db_path, "rb") as f:
        # Read header
        header = f.read(100)
        page_size = struct.unpack(">H", header[16:18])[0]
        if page_size == 1:
            page_size = 65536
        first_freelist_page = struct.unpack(">I", header[32:36])[0]
        total_freelist_pages = struct.unpack(">I", header[36:40])[0]

        freelist_info = {
            "page_size": page_size,
            "first_freelist_page": first_freelist_page,
            "total_freelist_pages": total_freelist_pages,
            "trunk_pages": [],
            "leaf_pages": []
        }

        if first_freelist_page == 0:
            return freelist_info

        # Walk the freelist trunk chain
        trunk_page = first_freelist_page
        while trunk_page != 0:
            offset = (trunk_page - 1) * page_size
            f.seek(offset)
            page_data = f.read(page_size)

            next_trunk = struct.unpack(">I", page_data[0:4])[0]
            leaf_count = struct.unpack(">I", page_data[4:8])[0]

            leaves = []
            for i in range(leaf_count):
                leaf_page = struct.unpack(">I", page_data[8 + i * 4:12 + i * 4])[0]
                leaves.append(leaf_page)

            freelist_info["trunk_pages"].append({
                "page_number": trunk_page,
                "next_trunk": next_trunk,
                "leaf_count": leaf_count,
                "leaf_pages": leaves
            })
            freelist_info["leaf_pages"].extend(leaves)
            trunk_page = next_trunk

    return freelist_info


def extract_freelist_content(db_path: str, output_dir: str):
    """Extract raw content from freelist pages for analysis."""
    info = analyze_freelist(db_path)
    os.makedirs(output_dir, exist_ok=True)

    with open(db_path, "rb") as f:
        page_size = info["page_size"]
        for page_num in info["leaf_pages"]:
            offset = (page_num - 1) * page_size
            f.seek(offset)
            page_data = f.read(page_size)
            output_file = os.path.join(output_dir, f"freelist_page_{page_num}.bin")
            with open(output_file, "wb") as out:
                out.write(page_data)

    return len(info["leaf_pages"])

Method 2: WAL (Write-Ahead Log) Analysis

The WAL file contains pending transactions that have not yet been checkpointed back to the main database.

def parse_wal_header(wal_path: str) -> dict:
    """Parse SQLite WAL file header and frame inventory."""
    with open(wal_path, "rb") as f:
        header = f.read(32)
        magic = struct.unpack(">I", header[0:4])[0]
        file_format = struct.unpack(">I", header[4:8])[0]
        page_size = struct.unpack(">I", header[8:12])[0]
        checkpoint_seq = struct.unpack(">I", header[12:16])[0]
        salt1 = struct.unpack(">I", header[16:20])[0]
        salt2 = struct.unpack(">I", header[20:24])[0]

        wal_info = {
            "magic": hex(magic),
            "format": file_format,
            "page_size": page_size,
            "checkpoint_sequence": checkpoint_seq,
            "frames": []
        }

        # Parse frames (24-byte header + page_size data each)
        frame_offset = 32
        frame_num = 0
        file_size = os.path.getsize(wal_path)

        while frame_offset + 24 + page_size <= file_size:
            f.seek(frame_offset)
            frame_header = f.read(24)
            page_number = struct.unpack(">I", frame_header[0:4])[0]
            db_size_after = struct.unpack(">I", frame_header[4:8])[0]

            wal_info["frames"].append({
                "frame_number": frame_num,
                "page_number": page_number,
                "db_size_pages": db_size_after,
                "offset": frame_offset
            })
            frame_offset += 24 + page_size
            frame_num += 1

    return wal_info

Method 3: Unallocated Space Within Pages

Deleted cells within active B-tree pages leave data in the unallocated region between the cell pointer array and the cell content area.

def analyze_unallocated_space(db_path: str, page_number: int) -> dict:
    """Analyze unallocated space within a specific B-tree page."""
    with open(db_path, "rb") as f:
        header = f.read(100)
        page_size = struct.unpack(">H", header[16:18])[0]
        if page_size == 1:
            page_size = 65536

        offset = (page_number - 1) * page_size
        f.seek(offset)
        page_data = f.read(page_size)

        # Parse page header (8 or 12 bytes depending on type)
        page_type = page_data[0]
        first_freeblock = struct.unpack(">H", page_data[1:3])[0]
        cell_count = struct.unpack(">H", page_data[3:5])[0]
        cell_content_offset = struct.unpack(">H", page_data[5:7])[0]
        if cell_content_offset == 0:
            cell_content_offset = 65536

        header_size = 12 if page_type in (0x02, 0x05) else 8
        cell_pointer_end = header_size + cell_count * 2

        unallocated_start = cell_pointer_end
        unallocated_end = cell_content_offset
        unallocated_size = unallocated_end - unallocated_start

        return {
            "page_number": page_number,
            "page_type": hex(page_type),
            "cell_count": cell_count,
            "unallocated_start": unallocated_start,
            "unallocated_end": unallocated_end,
            "unallocated_size": unallocated_size,
            "unallocated_data": page_data[unallocated_start:unallocated_end].hex()
        }

Common Forensic Databases

Timestamp Decoding

from datetime import datetime, timedelta

def decode_chrome_timestamp(chrome_ts: int) -> datetime:
    """Convert Chrome/WebKit timestamp to datetime (microseconds since 1601-01-01)."""
    epoch_delta = 11644473600
    return datetime.utcfromtimestamp((chrome_ts / 1000000) - epoch_delta)

def decode_unix_timestamp(unix_ts: int) -> datetime:
    """Convert Unix timestamp to datetime."""
    return datetime.utcfromtimestamp(unix_ts)

def decode_mac_absolute_time(mac_ts: float) -> datetime:
    """Convert Mac Absolute Time (seconds since 2001-01-01)."""
    mac_epoch = datetime(2001, 1, 1)
    return mac_epoch + timedelta(seconds=mac_ts)

def decode_mozilla_timestamp(moz_ts: int) -> datetime:
    """Convert Mozilla PRTime (microseconds since Unix epoch)."""
    return datetime.utcfromtimestamp(moz_ts / 1000000)

References

• SQLite File Format: https://www.sqlite.org/fileformat2.html
• Belkasoft SQLite Analysis: https://belkasoft.com/sqlite-analysis
• Spyder Forensics SQLite Training: https://www.spyderforensics.com/sqlite-forensic-fundamentals-2025/
• Forensic Analysis of Damaged SQLite Databases: https://www.forensicfocus.com/articles/forensic-analysis-of-damaged-sqlite-databases/

Example Output

$ python3 sqlite_forensics.py --db /evidence/chrome/Default/History \
    --wal /evidence/chrome/Default/History-wal \
    --journal /evidence/chrome/Default/History-journal \
    --output /analysis/sqlite_report

SQLite Database Forensic Analyzer v2.0
========================================
Database:    /evidence/chrome/Default/History
Size:        48.2 MB
SQLite Ver:  3.39.5
Page Size:   4096 bytes
Total Pages: 12,345
Encoding:    UTF-8

[+] Analyzing WAL (Write-Ahead Log)...
    WAL file:       History-wal (2.1 MB)
    WAL frames:     512
    Checkpointed:   No (contains uncommitted data)
    Recoverable rows from WAL: 234

[+] Analyzing journal file...
    Journal file:   History-journal (0 bytes - rolled back)

[+] Scanning for deleted records (freelist pages)...
    Freelist pages:     456
    Deleted records recovered: 1,892

[+] Analyzing table: urls
    Active rows:     12,456
    Deleted rows:    1,234 (recovered from freelist)
    WAL-only rows:   89

--- Recovered Deleted URLs (Last 10) ---
Row ID | URL                                              | Title                    | Visit Count | Last Visit (UTC)
-------|--------------------------------------------------|--------------------------|-------------|---------------------
89234  | https://mega.nz/folder/xYz123#key=AbCdEf        | MEGA                     | 5           | 2024-01-16 03:20:00
89235  | https://transfer.sh/abc123/data.7z               | transfer.sh              | 1           | 2024-01-16 03:25:00
89240  | https://temp-mail.org/en/                        | Temp Mail                | 3           | 2024-01-15 13:00:00
89241  | https://browserleaks.com/ip                      | IP Leak Test             | 1           | 2024-01-15 12:55:00
89245  | https://www.virustotal.com/gui/file/a1b2c3...    | VirusTotal               | 2           | 2024-01-15 14:30:00
89250  | https://github.com/gentilkiwi/mimikatz/releases  | Mimikatz Releases        | 1           | 2024-01-15 16:00:00
89260  | https://raw.githubusercontent.com/.../payload.ps1| GitHub Raw               | 1           | 2024-01-15 14:34:00
89270  | https://pastebin.com/edit/kL9mN2pQ               | Pastebin - Edit          | 2           | 2024-01-15 14:42:00
89280  | https://duckduckgo.com/?q=clear+browser+history  | DuckDuckGo               | 1           | 2024-01-17 22:00:00
89285  | https://duckduckgo.com/?q=anti+forensics+tools   | DuckDuckGo               | 1           | 2024-01-17 22:05:00

[+] Analyzing table: downloads
    Active rows:     234
    Deleted rows:    12 (recovered)

--- Recovered Deleted Downloads ---
Row ID | Filename               | URL                                    | Size      | Start Time (UTC)
-------|------------------------|----------------------------------------|-----------|---------------------
5012   | payload.ps1            | https://raw.githubusercontent.com/...  | 4,096     | 2024-01-15 14:34:00
5015   | mimikatz_trunk.zip     | https://github.com/.../releases/...    | 1,892,352 | 2024-01-15 16:00:00
5018   | netscan_portable.zip   | https://www.softperfect.com/...        | 5,242,880 | 2024-01-15 15:05:00

[+] Slack space analysis...
    Pages with slack space data: 234
    Partial strings recovered:   67 fragments

Summary:
  Total records analyzed:  14,578 (active) + 3,126 (deleted/WAL)
  Evidence-relevant URLs:  23 (flagged)
  Deleted downloads:       12 (3 tool-related)
  Anti-forensics evidence: Browser history deletion detected
  Report: /analysis/sqlite_report/sqlite_forensics.json
  Recovered DB: /analysis/sqlite_report/History_recovered.db