๐ Select Language | ้ๆฉ่ฏญ่จ
fsmon is a real-time Linux filesystem change monitor powered by fanotify. It watches files and directories, captures every event (create, modify, delete, move, attribute change, etc.), and attributes each change back to the process that caused it โ including the PID, command name, user, parent PID, thread group ID, and optional full process ancestry chain.
Features
- Real-time Monitoring: Captures 14 fanotify event types (default: 8 core events; use
--types allfor all 14) - Process Attribution: Tracks PID, command name, user, PPID, and TGID for every file change โ even short-lived processes like
touch,rm,mv - Process Tree Tracking (
<CMD>positional arg): Pinpoint a specific process (e.g.,openclaw) and fsmon will track it plus all its descendants (fork/exec children), building a complete ancestry chain per event. - Recursive Monitoring: Watch entire directory trees with automatic tracking of newly created subdirectories
- Complete Deletion Capture: Captures every file deleted during
rm -rfvia persistent directory handle cache - High Performance: Rust + Tokio, <5MB memory footprint, zero-copy FID event parsing, binary-search log querying
- Capture-time Filtering: Filter by event type and file size โ in-process, nanosecond-fast, no fork.
- Live Updates: Add/remove paths while daemon runs โ no restart needed.
Quick Start
Prerequisites
- OS: Linux 5.9+ (requires fanotify FID mode)
- Tested Filesystems: ext4, XFS, btrfs
- Build: Rust toolchain (
cargo)
# Verify kernel version
# Install Rust if needed
|
Installation
# Build from source
# Or install from crates.io
Fanotify requires root privileges for the daemon:
A Complete Walkthrough
Monitor a web project directory, see what gets logged, then use standard Unix tools to filter and clean.
# Terminal 1: start the daemon (sudo for fanotify)
&
# Terminal 1 (or another): add paths to monitor
# Monitor /var/www/myapp recursively, MODIFY + CREATE events only,
# tracking the nginx and vim processes.
# List what's being monitored
# {"cmd":"nginx","paths":{"/var/www/myapp":{"recursive":true,"types":["MODIFY","CREATE"]}}}
# {"cmd":"vim","paths":{"/var/www/myapp":{"recursive":true,"types":["MODIFY","CREATE"]}}}
Now trigger some real file changes:
# Terminal 2: simulate real usage
Look at what fsmon captured:
# The raw log โ one JSONL line per event
# โ {"time":"2026-05-07T10:00:01+00:00","event_type":"CREATE","path":"/var/www/myapp/index.html","pid":1234,"cmd":"nginx","user":"www-data","file_size":0,"ppid":1,"tgid":1234}
# โ {"time":"2026-05-07T10:00:01+00:00","event_type":"CLOSE_WRITE","path":"/var/www/myapp/index.html","pid":1234,"cmd":"nginx","user":"www-data","file_size":21,"ppid":1,"tgid":1234}
# โ {"time":"2026-05-07T10:00:03+00:00","event_type":"DELETE","path":"/var/www/myapp/index.html","pid":5678,"cmd":"rm","user":"deploy","file_size":0,"ppid":1234,"tgid":5678}
# โ {"time":"2026-05-07T10:00:05+00:00","event_type":"CREATE","path":"/var/www/myapp/.config.json.swp","pid":9012,"cmd":"vim","user":"dev","file_size":4096,"ppid":5678,"tgid":9012,"chain":"9012|vim|dev;5678|sh|deploy;1234|openclaw|root;1|systemd|root"}
Every event includes ppid (parent PID) and tgid (thread group ID). When a <CMD> is specified on add, matching events also include chain โ a compact process ancestry string tracing back to PID 1.
Query with pipe
# What did nginx do in the last hour?
|
# What files were deleted?
|
# Who made the biggest changes?
|
# Real-time tail with filter (watch for deployments)
|
No built-in --pid, --cmd, --user, --sort flags needed โ jq does it all.
Clean with safety
# Preview what would be deleted (config default: keep 30 days)
# Actually clean with custom retention
# Or just use Unix tools directly on the files
for; do
&&
done
# Stop the daemon
File Locations
| Purpose | Path | Format |
|---|---|---|
| Infrastructure config | ~/.config/fsmon/fsmon.toml |
TOML (optional โ defaults without it) |
| Monitored paths database | ~/.local/share/fsmon/monitored.jsonl |
JSONL (grouped by cmd, paths as map keys) |
| Event logs | ~/.local/state/fsmon/*_log.jsonl |
JSONL (one event per line) |
| Unix socket | /tmp/fsmon-<UID>.sock |
TOML over stream |
Both the store path and log directory are configurable in ~/.config/fsmon/fsmon.toml
(see [monitored].path and [logging].path).
The daemon runs as root (via sudo) but resolves your original user's home directory
via SUDO_UID + getpwuid_r, so it writes to /home/<you>/... not /root/....
Note for vfat/exfat/NFS users: The daemon tries to chown log files back to your user. Filesystems without standard Unix ownership (vfat, exfat, NFS with no_root_squash off) don't support this. Logs remain owned by root. If
fsmon cleanfails as a normal user, runsudo fsmon cleanor use the Unix tools directly on the.jsonlfiles.
Auto-start on Boot (Optional)
fsmon does not install a systemd service. The daemon requires sudo (root) for fanotify. To start automatically on login:
&
Note: Use
sudo crontab -e(root's crontab) โ the daemon needs root privileges. Add thefsmoncommand to sudoers with NOPASSWD if using a user crontab instead.
Complete Commands
daemon
Start the fsmon daemon โ requires sudo for fanotify.
sudo fsmon daemon # Start daemon in foreground
sudo fsmon daemon & # Start daemon in background
sudo fsmon daemon --debug # Enable debug output (event matching + cache stats)
sudo fsmon daemon --cache-dir-cap N # Directory handle cache capacity (default: 100000)
sudo fsmon daemon --cache-dir-ttl N # Directory handle cache TTL in seconds (default: 3600)
sudo fsmon daemon --cache-file-size N # File size cache capacity (default: 10000)
sudo fsmon daemon --cache-proc-ttl N # Process cache TTL in seconds (default: 600)
sudo fsmon daemon --cache-stats-interval N # Cache stats log interval in debug mode (default: 60, 0=off)
sudo fsmon daemon --buffer-size N # Fanotify read buffer in bytes (default: 32768)
add
Add a path (optionally with process tracking) to the monitoring list. No sudo needed.
fsmon add nginx --path /var/www/myapp -r # Track nginx on /myapp recursively
fsmon add nginx --path /var/www/myapp # Track nginx on /myapp (non-recursive)
fsmon add _global --path /home -r # Monitor all events on /home (global)
fsmon add _global --path /home --types MODIFY # Filter by event types
fsmon add _global --path /home --types all # All 14 event types
fsmon add _global --path /home --size '>=1MB' # Minimum file size filter
Modes:
| Mode | Example | Behavior |
|---|---|---|
| CMD + --path | fsmon add openclaw --path /home |
Track openclaw (and descendants) on /home. Matching events include chain. |
| Global (_global) | fsmon add _global --path /home |
All events on /home captured. Each event has ppid/tgid. |
<CMD>(positional arg) enables process tree tracking: fork/exec children are automatically included. Matching events get achainfield (e.g.,"102|touch|root;101|sh|root;100|openclaw|root;1|systemd|root").- Multiple entries with different
<CMD>values can be added (OR logic per entry). --pathis required. Use_globalas CMD for global monitoring (all processes).
remove
Remove one or more paths from the monitoring list. No sudo needed.
fsmon remove _global # Remove entire global cmd group
fsmon remove nginx # Remove entire nginx cmd group
fsmon remove nginx --path /home # Remove /home from nginx group
fsmon remove _global --path /home # Remove /home from global group
monitored
List all monitored paths with their filtering configuration (JSONL).
fsmon monitored # Show all monitored path groups
Each line is a JSON object with cmd and paths fields. Pipe to jq for filtering.
query
Query historical events from log files. Output is JSONL โ pipe to jq for filtering.
fsmon query _global # Query global log
fsmon query nginx # Query nginx log only
fsmon query _global -t '>1h' # Events from last hour
fsmon query _global -t '>=2026-05-01' # From absolute time
fsmon query _global -t '<30m' # Events until 30 minutes ago
fsmon query _global -t '>1h' -t '<now' # Time range (since + until)
fsmon query _global --path /tmp # Filter events by path prefix
Examples with jq:
# Search by process (ppid/tgid always present)
|
# Search by ancestry chain (only when --cmd was used on add)
|
# Traditional cmd/user filtering
|
|
|
clean
Clean log files for a specific cmd group. Defaults from fsmon.toml: keep_days=30, size==>=1GB.
fsmon clean _global # Clean global log (defaults)
fsmon clean nginx --time '>7d' # Keep last 7 days of nginx events
fsmon clean nginx --size '>=500MB' # Size limit for nginx log
fsmon clean _global --dry-run # Preview without deleting
Priority: CLI arg > fsmon.toml > code default (keep_days=30, size=>=1GB)
You can also clean the raw log files directly without fsmon clean:
# Keep only last 500 lines per log file
for; do
&&
done
# Delete logs older than 30 days by mtime
Note: Native
fsmon cleanparses JSONL accurately (won't cut mid-line) and handles both time and size constraints. Raw Unix tools are simpler but may produce partial lines.
init
Initialize fsmon data directories. Creates log dir and monitored data dir. Does NOT write a config file โ config is optional, defaults apply without it.
fsmon init
cd
Open a subshell in the log directory. Type exit to return:
fsmon cd
ls _global_log.jsonl
Configuration
Config file is optional โ defaults apply without it.
# fsmon configuration file
#
# Infrastructure paths for fsmon. Monitored paths are added via
# 'fsmon add' / 'fsmon remove' and persisted in [monitored].path.
# All paths support ~ expansion. <UID> is replaced with the numeric UID at runtime.
[]
# Path to the auto-monitored monitored paths database.
= "~/.local/share/fsmon/monitored.jsonl"
[]
# Path to the event log directory (per-cmd *_log.jsonl files).
= "~/.local/state/fsmon"
# Defaults for 'fsmon clean' (no auto-clean; use cron/timer).
= 30
= ">=1GB"
[]
# Unix socket path for daemon-CLI live communication.
= "/tmp/fsmon-<UID>.sock"
[]
# Directory handle cache capacity (default: 100000, ~15-20MB).
# Each entry maps a kernel file handle to a directory path.
# Lower on memory-constrained systems; raise when monitoring
# large directory trees (>100k dirs) to reduce handle re-resolution.
= 100000
# Directory handle cache TTL in seconds (default: 3600 = 1 hour).
# Shorter TTL frees memory faster for volatile directory structures;
# longer TTL reduces handle re-resolution for stable directories.
= 3600
# File size cache capacity (default: 10000, ~1MB).
# Avoids stat() calls for files with known sizes.
# Raise for high-file-volume workloads (git checkout, npm install).
= 10000
# Process cache TTL in seconds (default: 600 = 10 minutes).
# Applies to both process info cache (PIDโcmd/user/ppid/tgid) and
# process tree cache (PIDโparent for ancestor chain tracking).
# Shorter TTL cleans up zombie process entries faster;
# longer TTL reduces /proc reads for long-lived processes.
= 600
# Cache stats log interval in seconds in debug mode (default: 60).
# Set to 0 to disable periodic cache stats output.
= 60
Override priority
CLI arguments (--cache-dir-cap, --cache-dir-ttl, --cache-file-size, --cache-proc-ttl, --cache-stats-interval, --buffer-size)
> fsmon.toml [cache] section
> code defaults
CLI flags override both config file and defaults:
# Override dir_cache capacity and fanotify buffer size at startup
&
Event Types
Default captures 8 core events. Use --types all for all 14.
Default (8): CLOSE_WRITE, ATTRIB, CREATE, DELETE, DELETE_SELF, MOVED_FROM, MOVED_TO, MOVE_SELF
All 14 (via --types all): + ACCESS, MODIFY, OPEN, OPEN_EXEC, CLOSE_NOWRITE, FS_ERROR
FS_ERROR only works with filesystem-level marks (requires a filesystem that supports it).
Log Format
Every event is a single JSON line. All fields are always present.
When <CMD> was specified on add and the event matches: chain is also included.
The chain format: pid|cmd|user per entry, ;-separated from the event process up to PID 1 (root).
Architecture
Linux Kernel (fanotify FID mode)
โ Raw # FID events pushed to kernel queue
โ tokio reads events asynchronously
โ fid_parser: resolves paths (two-pass + # DashMap dir handle cache)
โ filters: event type, size, recursive/non-recursive scope
โ (if <CMD> was specified) process tree check:
โ not in tracked tree โ drop immediately (zero /proc reads)
โ in tracked tree โ build ancestry chain โ append to event
โ write # JSONL โ per-cmd log file (<cmd>_log.jsonl)
Process tree (proc connector):
Fork/Exec/Exit events from netlink connector socket
โ # DashMap: pid โ {cmd, ppid, user, tgid, start_time}
On daemon start: /proc/*/stat snapshot seeds existing processes
is_descendant(pid, "openclaw") โ O(depth) # DashMap lookups
User pipe:
tail -f *.jsonl | jq 'select(...)'
Clean:
fsmon clean โ parse # JSONL, apply time/size filters, truncate
Source Tree
src/
โโโ bin/
โ โโโ fsmon.rs # CLI entry: main(), argument structs, arg tests
โ โโโ commands/
โ โโโ mod.rs # run() dispatch, parse_path_entries helper
โ โโโ daemon.rs # Daemon: load store, # Monitor::new(), run()
โ โโโ add.rs # CLI add: path normalization, store + socket
โ โโโ remove.rs # CLI remove: store + socket
โ โโโ monitored.rs # CLI monitored: # JSONL output
โ โโโ query.rs # CLI query: time filter, execute query
โ โโโ clean.rs # CLI clean: parser delegation
โ โโโ init_cd.rs # CLI init, cd
โ
โโโ lib.rs # FileEvent, EventType, # DaemonLock (flock singleton)
โโโ clean.rs # Log cleanup engine: time/size trim, tail-offset
โโโ config.rs # TOML config, # SUDO_UID home resolution
โโโ monitored.rs # Monitored paths database (JSONL store)
โโโ monitor.rs # Fanotify loop, socket handler, add/remove/events
โโโ fid_parser.rs # FID event parsing, two-pass path recovery
โโโ filters.rs # PathOptions, event/size filters, path matching
โโโ dir_cache.rs # Directory handle cache (DashMap + HandleKey)
โโโ proc_cache.rs # hNetlink proc connector: # Fork/Exec/Exit, build_chain
โโโ query.rs # Binary-search log query on sorted JSONL
โโโ socket.rs # Unix socket protocol (TOML req/resp)
โโโ utils.rs # Size/time parsing, process info lookup, chown
โโโ help.rs # Help text constants