mquire

mquire, a play on the memory and inquire words, is a memory querying tool inspired by osquery.

Key advantage: No external debug symbols needed

mquire can analyze Linux kernel memory snapshots without requiring external debug symbols.

Everything needed for analysis is already embedded in the memory dump itself. This means you can analyze:

• Unknown or custom kernels you've never seen before
• Any Linux distribution without preparation
• Memory snapshots where external debug symbols are unavailable or lost

Requirements

Kernel version requirements:

• BTF support: Kernel 4.18 or newer with BTF enabled (most modern distributions enable it by default)
• Kallsyms support: Kernel 6.4 or newer (due to changes in scripts/kallsyms.c format)

How it works

mquire analyzes kernel memory by reading two types of information that are embedded in modern Linux kernels:

1. Type information from BTF (BPF Type Format) - Describes the structure and layout of kernel data types. BTF data is parsed using the btfparse crate.
2. Symbol information from Kallsyms - Provides the memory locations of kernel symbols (same data used by /proc/kallsyms)

By combining type information with symbol locations, mquire can find and read complex kernel data structures like:

• Process memory mappings (using maple tree structures)
• Cached file data (using XArray structures)
• Kernel log messages

This makes it possible to extract files directly from the kernel's file cache, even if they've been deleted from disk.

Compatibility notes

The Kallsyms scanner depends on the data format from scripts/kallsyms.c in the kernel source. If future kernel versions change this format, the scanner heuristics may need updates.

Capabilities

Tables

mquire provides SQL tables to query different aspects of the system or the state of the tool itself.

mquire is not a database. Each query reconstructs kernel data structures by scanning memory and following pointers. There are no precomputed indexes or cached results: every table access is a traversal of kernel data. Use AS MATERIALIZED to avoid redundant scans (see Query Optimization), and provide constraints like task when querying per-process tables like task_open_files and memory_mappings to limit the scan to a single process.

Design principle: virtual addresses as join keys. Tables use virtual_address (the kernel address of the underlying data structure) as the canonical join key: not pid or other user-visible identifiers. This is intentional, because the same PID can appear multiple times across different discovery sources and root tasks, while a virtual address uniquely identifies a specific kernel object. Both the SQL tables and the underlying LinuxOperatingSystem API are built around this convention.

System information

• os_version - Kernel version and architecture
• system_info - Hostname and domain name
• boot_time - System boot time
• kallsyms - Kernel symbol addresses (same data as /proc/kallsyms)
• dmesg - Kernel ring buffer messages (same data as dmesg command)

Process information

• tasks - Running processes with command lines and binary paths. Each task is discovered via multiple independent sources, which is useful for rootkit detection. See Comparing task enumeration methods for rootkit detection and Deduplicated process list.
• task_open_files - Files opened by each process (provide a task constraint for targeted analysis, or query all tasks at once)
• memory_mappings - Memory regions mapped by each process (provide a task constraint for targeted analysis, or query all tasks at once)
• task_capabilities - Linux capability sets (effective, permitted, inheritable, bounding, ambient) for a task. Requires a task constraint, so join it against tasks/processes (e.g. JOIN task_capabilities c ON c.task = p.virtual_address).
• task_ptrace_flags - Decodes the ptrace field of a given task. Requires a task constraint.

Kernel modules

• kernel_modules - Loaded kernel modules with metadata (name, state, version, parameters, taint flags)
• kernel_module_mem_entries - The mem entries from a kernel module object. Requires a kernel_module constraint, so join it against kernel_modules (e.g. JOIN kernel_module_mem_entries r ON r.kernel_module = m.virtual_address).
• ftrace_ops - A list of struct ftrace_ops nodes, walked from the ftrace_ops_list symbol by default. Constrain virtual_address to read a single node, or start_vaddr (optionally bounded by end_vaddr) to walk from an arbitrary node.

Network information

• network_connections - Active network connections (TCP sockets)
• network_interfaces - Network interfaces with IP addresses and MAC addresses

File system

• syslog_file - System logs read from the kernel's file cache (works even if log files are deleted or unavailable, as long as they're cached in memory)

Debugging

• mquire_diagnostics - Internal mquire logs showing analysis progress, warnings, and errors

Commands

mquire provides three main commands:

• mquire shell - Start an interactive SQL shell to query memory snapshots
• mquire query - Execute a single SQL query and output results (supports JSON or table format)
• mquire command - Execute custom commands on memory snapshots (e.g., .task_tree, .system_version, .dump)

Dot Commands

mquire provides special commands prefixed with a dot (.) to distinguish them from SQL queries.

Built-in Commands

These commands work in the interactive shell and with mquire query:

• .tables - List all available tables
• .schema - Show schema for all tables
• .schema <table> - Show schema for a specific table
• .commands - List all available custom commands
• .exit - Exit the interactive shell (shell only)

Custom Commands

These commands work in the interactive shell and with mquire command:

Use --help with any command to see available options and usage information. For example: .task_tree --help

.system_version

Display the operating system version information.

This is a convenience command equivalent to SELECT * FROM os_version, but with formatted output.

.task_tree

Display a hierarchical tree of running processes and threads, similar to the pstree command on Linux.

Options:

• --show-threads - Include threads in addition to processes. When enabled, displays both TGID and TID for each entry.
• --use-real-parent - Use the real_parent field instead of parent for building the tree structure. The real_parent field shows the original parent process before any reparenting (useful for tracking process creation chains even after parent processes exit).

Notes:

• The format is [TGID TID] when showing threads, or [TGID] when threads are hidden. TGID (Thread Group ID) is what's commonly called PID. For main threads (where TGID == TID), both values will be the same.

.carve

Carve a region of virtual memory to disk. This command extracts raw memory content from a specific virtual address range using a given page table, useful for extracting process memory, heap contents, or other memory regions.

Arguments:

• ROOT_PAGE_TABLE - The physical address of the root page table (hex string with optional 0x prefix). This determines the address space to use for translation.
• VIRTUAL_ADDRESS - The virtual address to start carving from (hex string with optional 0x prefix).
• SIZE - Number of bytes to carve.
• DESTINATION_PATH - Output file path where the carved memory will be written.

Notes:

• The command shows a summary of mapped vs unmapped regions before writing.
• Unmapped regions are filled with zeros in the output file.

.dump

Extract files from the kernel's file cache to recover files directly from memory. This command iterates through all tasks and their open file descriptors, extracting file contents from the page cache.

Arguments:

• OUTPUT - Output directory for extracted files. Files are organized by TGID (e.g., tgid_1234/path/to/file).

Notes:

• Currently works with files opened through file descriptors (from the process file descriptor table).
• Does not yet support extracting data from memory-mapped files.
• Empty files (no data in page cache) are skipped.
• Regions with read errors are zero-padded in the output.

Use cases

mquire is designed for:

• Forensic analysis - Analyze memory snapshots from compromised systems to understand what was running and what files were accessed
• Incident response - Quickly query memory dumps to find evidence of malicious activity
• Security research - Study kernel internals and process behavior from memory snapshots
• Malware analysis - Examine running processes and their file operations without detection
• Custom tooling - Build your own analysis tools using the mquire library crate, which provides a reusable API for kernel memory analysis

Building and installation

Pre-built packages from CI

Pre-built packages are available as artifacts from CI runs. You can download them from the Actions tab by selecting a successful workflow run and downloading the artifacts. The following package formats are available:

• linux-deb-package - Debian/Ubuntu .deb package
• linux-rpm-package - Fedora/RHEL/CentOS .rpm package
• linux-tgz-package - Generic Linux .tar.gz archive

Building from source

mquire is written in Rust. To build it:

# Clone the repository
git clone https://github.com/trailofbits/mquire
cd mquire

# Build the project
cargo build --release

# The binary will be in target/release/
# - mquire: Unified tool with shell, query, and command modes

Supported snapshot formats

mquire supports the following memory snapshot formats, detected by file extension:

Acquiring a memory snapshot

Using AVML (recommended)

We recommend AVML for acquiring memory snapshots from live Linux systems. LiME was previously suggested but is no longer actively maintained.

sudo avml output.lime

Important: Do not use --compress when acquiring snapshots for mquire. mquire does not support compressed AVML snapshots. If you have a compressed snapshot, use avml-convert to decompress it first: avml-convert compressed.lime uncompressed.lime

See the AVML documentation for additional options.

Using virsh (virtual machines)

For libvirt/KVM virtual machines, use virsh dump to produce an ELF core dump:

virsh -c qemu:///system dump <domain> output.elf --memory-only --format elf

The VM is paused during the dump and resumed after. Add --live to avoid pausing, at the cost of snapshot consistency.

Getting started

Once you have a memory snapshot, you can interact with it using SQL queries and custom commands. mquire provides three ways to interact with snapshots:

Interactive shell

Start an interactive SQL shell:

mquire shell /path/to/memory.raw

This opens a prompt where you can run both SQL queries and commands interactively:

mquire> .tables                      # List all available tables
mquire> .schema tasks                # Show schema for a specific table
mquire> SELECT * FROM tasks;         # Run SQL queries
mquire> .task_tree --show-threads    # Run custom commands
mquire> .exit                        # Exit the shell

One-off SQL queries

Execute a single SQL query or built-in command from the command line:

# Output as JSON (default)
mquire query /path/to/memory.raw "SELECT * FROM os_version"

# Output as table format
mquire query /path/to/memory.raw "SELECT * FROM tasks" --format table

# Built-in commands work too
mquire query /path/to/memory.raw ".tables"
mquire query /path/to/memory.raw ".schema tasks"

Execute custom commands

Run custom commands for specialized analysis:

# List all available commands (default behavior)
mquire command /path/to/memory.raw

# Display system version
mquire command /path/to/memory.raw ".system_version"

# Show process tree
mquire command /path/to/memory.raw ".task_tree"

# Show process tree with threads
mquire command /path/to/memory.raw ".task_tree --show-threads"

# Get help for a command
mquire command /path/to/memory.raw ".task_tree --help"

Autostart SQL Files

mquire automatically loads and executes SQL files from $HOME/.config/trailofbits/mquire/autostart/ on startup. Files are organized by operating system and architecture:

autostart/
  common/common/    # All platforms and architectures
  common/{arch}/    # All platforms, specific architecture
  {os}/common/      # Specific platform, all architectures
  {os}/{arch}/      # Specific platform and architecture

Files within each directory are sorted alphabetically. Directories are scanned in the order shown above.

Features:

• Files must have a .sql extension
• Errors are logged to mquire_diagnostics but don't block execution
• Works with both mquire shell and mquire query commands

Shipped views

mquire ships reusable SQL views in the sql/views/ directory. Install them with just install-views. See the views README for the full list, numbering convention, and directory structure.

Linux views (sql/views/linux/common/):

• 000_processes.sql - Deduplicated process list across all discovery sources, filtered to user-space process leaders. Query with SELECT * FROM processes.
• 100_process_network_connections.sql - Maps network connections to owning processes by joining through file descriptors. Query with SELECT * FROM process_network_connections WHERE comm = 'sshd'.
• 120_process_capabilities.sql - Quick-peek capability overview, one row per process with each set as a compact list (ALL, '', or cap names; NULL if unreadable). Query with SELECT * FROM process_capabilities.
• - Quick-peek ptrace overview, one row per process with the set flags as a compact list ( when not traced) plus the raw value. Query with .

Query Optimization

mquire queries require reconstructing kernel data structures from virtual memory by dereferencing pointers using embedded type information and debug symbols. This processing can be expensive, so use query optimization techniques to improve performance dramatically.

Materialization with AS MATERIALIZED

Use the AS MATERIALIZED hint to cache table results when tables are used in JOINs or accessed multiple times.

When to materialize:

• Tables that are expensive to generate (e.g., tasks requires walking linked lists of process structures, dereferencing multiple pointers per process)
• Tables used in JOINs (accessed multiple times during query execution)
• Tables referenced multiple times in the same query

Example:

-- Find network connections for a specific process using materialization
WITH
  target_tasks AS MATERIALIZED (
    SELECT * FROM tasks WHERE comm = 'sshd' AND type = 'thread_group_leader'
  ),

  network_connections_mat AS MATERIALIZED (
    SELECT * FROM network_connections
  )

SELECT
  t.tgid,
  t.comm,
  nc.local_address,
  nc.local_port,
  nc.remote_address,
  nc.remote_port,
  nc.state,
  nc.protocol
FROM target_tasks t
JOIN task_open_files tof ON tof.task = t.virtual_address
JOIN network_connections_mat nc ON nc.inode = tof.inode;

Note: The task_open_files and memory_mappings tables use the task column as a generator input. When joined with the tasks table, SQLite automatically passes the constraint via nested loop joins, making direct JOINs efficient.

Performance impact: Materialization can provide significant speedup for queries with JOINs (typically 2-5x faster)

Example benchmark results:

Test performed on an Ubuntu 24.04 snapshot (kernel 6.8.0-63), 351 processes, 50 connections, 2142 open files. Performance will vary based on snapshot size, kernel version, and hardware.

JOIN Order Optimization

Start with the smallest table and JOIN toward larger tables to minimize rows processed early in the query pipeline.

Typical table sizes:

• network_connections: smallest - only processes with network activity
• tasks: medium - all processes
• task_open_files: largest - all open file descriptors

Optimal order:

Start with the filtered tasks table and join toward larger tables:

FROM target_tasks t                                       -- filtered tasks
JOIN task_open_files tof ON tof.task = t.virtual_address  -- open files
JOIN network_connections_mat nc ON nc.inode = tof.inode   -- matching connections

Understanding Query Execution

Use EXPLAIN QUERY PLAN to see how SQLite executes your query:

EXPLAIN QUERY PLAN
SELECT ...
FROM target_tasks t
JOIN task_open_files tof ON tof.task = t.virtual_address
JOIN network_connections_mat nc ON nc.inode = tof.inode;

Look for:

• BLOOM FILTER: SQLite's optimization for large JOINs
• AUTOMATIC COVERING INDEX: Temporary indexes created for lookups
• SCAN: Full table scan (expected for the driving table)
• SEARCH: Index-based lookup (efficient)

Best Practices

1. Always materialize expensive tables used in JOINs
2. Start with the smallest table as your driving table
3. Use multiline SQL in scripts for readability
4. Check query plans with EXPLAIN QUERY PLAN for complex queries
5. Avoid SELECT * in production - specify only needed columns

File extraction

Extract files from memory to disk:

mquire command /path/to/memory.raw ".dump /output/directory"

Example queries

All queries use standard SQL syntax.

System version

$ mquire shell ubuntu2404_6.14.0-37-generic.lime
mquire> SELECT * FROM os_version;
arch:"x86_64" kernel_version:"6.14.0-37-generic" system_version:"#37~24.04.1-Ubuntu SMP PREEMPT_DYNAMIC Thu Nov 20 10:25:38 UTC 2"

System information

$ mquire shell ubuntu2404_6.14.0-37-generic.lime
mquire> SELECT * FROM system_info;
domain:"(none)" hostname:"ubuntu2404"

Kernel modules

$ mquire shell ubuntu2404_6.14.0-37-generic.lime
mquire> SELECT name, state, src_version, parameters FROM kernel_modules LIMIT 5;
name:"snd_seq_dummy" state:"live" src_version:"7A40E0FD47A0746D1C9CD85" parameters:"ump (perm: 0o444), duplex (perm: 0o444), ports (perm: 0o444)" 
name:"snd_hrtimer" state:"live" src_version:"81EE6D58896E2C2E63E252D" parameters:"<null>" 
name:"qrtr" state:"live" src_version:"473C5AB47E04ECEA0106681" parameters:"<null>" 
name:"virtio_rng" state:"live" src_version:"0852940240D554836D22CB2" parameters:"<null>" 
name:"intel_rapl_msr" state:"live" src_version:"34853C4F5EB8FCAD28ACFB3" parameters:"<null>" 

Running tasks

$ mquire shell ubuntu2404_6.14.0-37-generic.lime
mquire> SELECT comm, binary_path, command_line FROM tasks WHERE command_line NOT NULL AND comm LIKE "%systemd%";
comm:"systemd" binary_path:"/usr/lib/systemd/systemd" command_line:"/sbin/init splash"
comm:"systemd-oomd" binary_path:"/usr/lib/systemd/systemd-oomd" command_line:"/usr/lib/systemd/systemd-oomd"
comm:"systemd-resolve" binary_path:"/usr/lib/systemd/systemd-resolved" command_line:"/usr/lib/systemd/systemd-resolved"
comm:"systemd-udevd" binary_path:"/usr/bin/udevadm" command_line:"/usr/lib/systemd/systemd-udevd"
comm:"systemd" binary_path:"/usr/lib/systemd/systemd" command_line:"/usr/lib/systemd/systemd --user"
comm:"systemd-logind" binary_path:"/usr/lib/systemd/systemd-logind" command_line:"/usr/lib/systemd/systemd-logind"
comm:"systemd-journal" binary_path:"/usr/lib/systemd/systemd-journald" command_line:"/usr/lib/systemd/systemd-journald"
comm:"systemd-timesyn" binary_path:"/usr/lib/systemd/systemd-timesyncd" command_line:"/usr/lib/systemd/systemd-timesyncd"

Connections

Find network connections for a specific process by joining tasks, task_open_files, and network_connections.

$ mquire shell ubuntu2404_6.14.0-37-generic.lime
mquire> SELECT
  t.tgid,
  t.comm,
  nc.protocol,
  nc.local_address,
  nc.local_port,
  nc.remote_address,
  nc.remote_port,
  nc.state
FROM tasks t
JOIN task_open_files tof ON tof.task = t.virtual_address
JOIN network_connections nc ON nc.inode = tof.inode
WHERE t.comm = 'sshd';
tgid:"1134" comm:"sshd" protocol:"tcp" local_address:"0.0.0.0" local_port:"22" remote_address:"<null>" remote_port:"<null>" state:"listen"
tgid:"1134" comm:"sshd" protocol:"tcp" local_address:"::" local_port:"22" remote_address:"<null>" remote_port:"<null>" state:"listen"

Task open files

List open files for specific processes by joining tasks with task_open_files:

$ mquire shell ubuntu2404_6.14.0-37-generic.lime
mquire> SELECT t.comm, tof.path
FROM tasks t
JOIN task_open_files tof ON tof.task = t.virtual_address
WHERE t.comm LIKE '%systemd%'
LIMIT 10;
comm:"systemd" path:"/null"
comm:"systemd" path:"/null"
comm:"systemd" path:"/null"
comm:"systemd" path:"/kmsg"
comm:"systemd" path:"[eventpoll]"
comm:"systemd" path:"[signalfd]"
comm:"systemd" path:"inotify"
comm:"systemd" path:"/"
comm:"systemd" path:"[timerfd]"
comm:"systemd" path:"/usr/lib/systemd/systemd-executor"

Command-line query examples

JSON output (default)

$ mquire query --format=json ubuntu2404_6.14.0-37-generic.lime "SELECT * FROM os_version"
[
  {
    "arch": "x86_64",
    "kernel_version": "6.14.0-37-generic",
    "system_version": "#37~24.04.1-Ubuntu SMP PREEMPT_DYNAMIC Thu Nov 20 10:25:38 UTC 2"
  }
]

Table output

$ mquire query --format=table ubuntu2404_6.14.0-37-generic.lime "SELECT * FROM os_version"
arch:"x86_64" kernel_version:"6.14.0-37-generic" system_version:"#37~24.04.1-Ubuntu SMP PREEMPT_DYNAMIC Thu Nov 20 10:25:38 UTC 2"

Custom command examples

List available commands

$ mquire command ubuntu2404_6.14.0-37-generic.lime
Available commands:
  .carve               Carve a region of virtual memory to disk
  .dump                Dump all open files from tasks to disk
  .system_version      Display the operating system version
  .task_tree           Display a hierarchical task tree

Display system version

$ mquire command ubuntu2404_6.14.0-37-generic.lime ".system_version"
System Version: #37~24.04.1-Ubuntu SMP PREEMPT_DYNAMIC Thu Nov 20 10:25:38 UTC 2
Kernel Version: 6.14.0-37-generic
Architecture: x86_64

Display process tree

$ mquire command ubuntu2404_6.8.0-63-generic.lime .task_tree | head -n 10
Parent: task_struct::parent
Threads: Disabled
Page Table: paddr(0x0000000001a60000)

└─ [0] (ffffffff90c0fcc0) swapper/0
   ╎   ↳ [0] (ffff982a00e33518) \xef\xbf\xbd\xef\xbf\xbd\xef\xbf\xbd,)\xef\xbf\xbd\xef\xbf\xbd\xef\xbf\xbd\x0e
   ├─ [1] (ffff982a0084a8c0) systemd
   │  ├─ [430] (ffff982a0d27a8c0) systemd-journal
   │  ├─ [495] (ffff982a08a88000) systemd-udevd
   │  ├─ [786] (ffff982a07a60000) systemd-oomd

Note: When multiple task_struct entries exist with the same TID (Thread ID, which can occur due to memory corruption or snapshot timing), duplicate entries are displayed with the continuation symbol ╎ ↳ indented under the primary entry. The format is [TGID] (virtual_address) name when threads are hidden, or [TGID TID] when showing threads (where TGID is the Thread Group ID, commonly known as PID).

Extract files from memory

$ mquire command ubuntu2404_6.14.0-37-generic.lime ".dump ./extracted_files"
Legend: SK = skipped, OK = all good, ER = errored

Summary:
  Total files processed: 1234
  Successfully dumped: 1156
  Skipped: 45
  Errors: 33

File Status:
  OK /usr/lib/systemd/systemd (TGID 1)
  OK /etc/passwd (TGID 1)
  SK /dev/null (TGID 1)
  ...

Troubleshooting

Use the --debug flag to enable verbose debug messages during initialization and analysis:

mquire --debug command /path/to/memory.raw ".system_version"

• shell and query modes: Debug messages are stored in the mquire_diagnostics SQL table. Query them with SELECT * FROM mquire_diagnostics;
• command mode: Debug messages are printed directly to stderr.

For initialization issues that prevent mquire from successfully loading the snapshot, using command mode with a simple command like .system_version is recommended, as it prints debug output to stderr immediately without needing to query the mquire_diagnostics table.

Configuration

mquire can be configured via a TOML file at $HOME/.config/trailofbits/mquire/config.toml. If the file does not exist, default values are used.

Available options and default values

[database]
# Maximum number of entries retained in the mquire_diagnostics table.
# During initialization, all entries are kept regardless of this limit.
# Once initialization completes, new log entries trigger eviction of the
# oldest entries when the total exceeds this value.
mquire_diagnostics_max_entries = 1000

Development

This project uses just as a command runner. Run just to see available commands:

SQL query integration tests

These tests verify mquire produces correct output for queries against memory snapshots.

• just integration-test - Run tests and compare output to expected JSON files
• just integration-update - Update expected JSON files with actual output (use when changing table schemas)

After running integration-update, review the git diff to ensure changes match your expectations before committing.

Adding new tests: Create a .sql file and matching .json file in the appropriate snapshot directory, then run just integration-update to populate the expected output.

Contributing

Contributions are welcome! When contributing, please follow these guidelines:

1. Test your changes - Make sure your changes work correctly before submitting
2. Keep dependencies minimal - Only add new dependencies when absolutely necessary
3. Avoid caching volatile data - Do not cache values that could move or change within kernel objects. Only cache stable references like:
   • Kallsyms location
   • init_task virtual address
   • BTF data

License

This project is licensed under the Apache License 2.0. See the LICENSE file for details.