* feat: add reverse DNS resolution for IP addresses
- Add --resolve-dns flag to enable background DNS resolution
- Add --show-ptr-lookups flag to show/hide PTR lookup connections
- Create dns.rs module with async resolver and LRU cache
- Display hostnames in UI with 'd' key toggle
- Include hostname fields in JSON log output when resolved
Closes#97
* feat(ui): add network health visualization to Graph tab
- Add RTT/latency tracking via TCP SYN-ACK timing
- Add packet loss percentage tracking from retransmit counts
- Add Network Health chart with dual-axis (RTT + loss)
- Add TCP States panel showing connection state distribution
- Add per-connection RTT display in Details tab
Show helpful message when clipboard fails due to sandbox restrictions,
informing users to use --no-sandbox flag. Also adds wl-copy fallback
for Wayland environments where arboard fails.
- Add traffic history tracking with 60-second ring buffer
- Add Graph tab with traffic and connection charts
- Add sparklines to Interface Stats on Overview
- Add Tab/Shift+Tab navigation between tabs
Use periodic procfs PID cache to resolve thread names (e.g. 'Socket Thread')
to main process names (e.g. 'firefox'). Falls back to eBPF name for
short-lived processes that have already exited.
* feat: add Landlock sandbox and capability dropping for Linux
- Restrict filesystem access to /proc only after initialization
- Block TCP bind/connect on kernel 6.4+ (network sandbox)
- Drop CAP_NET_RAW after pcap handle opened
- Add --no-sandbox and --sandbox-strict CLI options
- Show privilege info on non-Linux platforms in UI
- Add SECURITY.md documentation
* fix: remove unused set_sandbox_info and hide Landlock line on non-Linux
* fix: gate SandboxInfo to Linux only to fix clippy warnings
* fix: add is_admin() function for Windows builds
The Windows build was failing because ui.rs called crate::is_admin()
but the function didn't exist. Added the implementation using Windows
Security API to check if the process has elevated privileges.
Also added Win32_Security feature to windows crate dependencies.
* fix: add is_admin() to main.rs for Windows binary crate
The previous fix added is_admin() to lib.rs but ui.rs is compiled
as part of the binary crate (main.rs), not the library crate.
Added the function to main.rs so crate::is_admin() resolves correctly.
* feat: reorganize platform code into per-platform directories
- Move platform files into linux/, macos/, windows/, freebsd/ subdirectories
- Unify create_process_lookup() API with _use_pktap parameter across all platforms
- Update build.rs paths for eBPF program location
- Reduce cfg attributes in main mod.rs from ~42 to 8
* fix: widen tolerance for test_sliding_window_no_skip_first_sample
Increase acceptable range from 9000-11000 to 5000-15000 to account
for timing variability on macOS ARM CI runners.
* docs: update Linux build dependencies and remove EBPF_BUILD.md
- Add missing build-essential, pkg-config, zlib1g-dev to documentation
- Update rust.yml CI with complete dependencies
- Remove EBPF_BUILD.md (info already in INSTALL.md)
- Update references in README.md and ARCHITECTURE.md
Adds real-time TCP connection quality monitoring:
- Retransmission detection via sequence number analysis
- Out-of-order packet tracking
- Fast retransmit detection (RFC 2581)
- Per-connection and aggregate statistics
- Network Stats panel in UI
- Platform-independent implementation using pcap data
Statistics show both active (current connections) and total
(cumulative) counts. Individual connection details display
per-connection metrics for troubleshooting network issues.
Changed bandwidth sorting to use the sum of upload and download speeds
instead of separate sorting for each direction. This provides a simpler
way to identify connections with the highest total bandwidth usage.
- Replace BandwidthDown/BandwidthUp with single BandwidthTotal enum
- Update sort logic to calculate combined rates
- Simplify UI to show "Down/Up ↓/↑" indicator
- Update documentation and tests
- Change sliding window from 5s to 10s for stable measurements
- Increase max_samples from 100 to 20,000 to preserve full time window
- Fix minimum time span threshold from 100ms to 1s for stability
- Remove skip(1) bug that excluded first sample from calculations
- Remove artificial decay logic causing rate drops during transfers
- Add idle connection detection for proper zero-rate display
- Update tests for 1s minimum requirement and add edge case coverage
Remove CAP_NET_ADMIN requirement and eliminate need for CAP_SYS_ADMIN on
modern kernels by using non-promiscuous mode for packet capture. This
significantly reduces security surface by following principle of least privilege.
Add --json-log flag to output connection events as JSON lines.
Logs new_connection and connection_closed events with IPs, ports,
protocol, DPI info, and traffic statistics for SIEM tools.
The procfs-based process lookup was triggering a full scan on every cache
miss instead of relying on periodic refresh. This caused 50+ full procfs
scans per enrichment cycle when multiple connections lacked process info.
Changed get_process_for_connection() to do simple cache lookups only.
Periodic refresh (every 5s) is already handled by the enrichment thread.
Also added PROFILING.md with flamegraph profiling guide.
* feat: add TUN/TAP interface support
Add comprehensive support for TUN/TAP virtual network interfaces by
refactoring link layer parsing into modular components.
New modules:
- link_layer: Modular packet parsing (ethernet, raw_ip, linux_sll, tun_tap)
- protocol: Dedicated TCP/UDP/ICMP parsers
Changes:
- Remove TUN/TAP interface exclusions in capture.rs
- Add TUN/TAP detection and parsing support
- macOS PKTAP support with conditional compilation
Platform compatibility:
- Linux: Full TUN/TAP support
- macOS: TUN (utun*) and TAP support
- Windows: No breaking changes
Fixes#39
- Add Windows process lookup using GetExtendedTcpTable/GetExtendedUdpTable
- Resolve process names via OpenProcess and QueryFullProcessImageNameW
- Support TCP/UDP IPv4 and IPv6 connections
- Implement time-based caching with 2-second TTL
- Fix port byte order conversion from network to host order
- Migrate from winapi to windows crate (v0.59)
- Add debug logging for process lookup operations
- Update documentation in ROADMAP.md and README.md
Closes#36
* feat: detect insufficient privileges before network interface access
- Add privilege detection module for Linux, macOS, and Windows
- Check privileges before TUI initialization for visible errors
- Provide platform-specific instructions (sudo, setcap, Docker flags)
- Detect container environments and provide Docker-specific guidance
Fixes#34 - Extract actual packet size from IP header Total Length field
instead of using captured buffer length (limited by snaplen). This fixes
severe undercounting for large packets (NFS, jumbo frames).
* Remove musl targets to simplify Linux builds and fix cross-compilation issues
* Enable eBPF by default on Linux via linux-default feature for better packet capture
* Add macOS code signing and notarization support with graceful fallback for unsigned builds
* Fix Windows MSI packaging with improved WiX configuration and Npcap library linking
* Auto-extract changelog content from CHANGELOG.md into GitHub release notes
* Fix ARM cross-compilation (aarch64, armv7) with proper library paths and eBPF support
* Add comprehensive installation documentation for DMG, MSI, DEB, and RPM packages
* Allow re-running releases with --clobber flag for artifact uploads
- Add GitHub Actions workflow for automated releases
- Create .deb, .rpm, AppImage, .dmg, and .msi packages
- Generate shell completions and manpages in build.rs
- Add platform-specific icons from rustnet3.svg
- Include eBPF dependencies for Linux packages
- Support Windows 32-bit and 64-bit builds
- Extract shared CLI module to prevent duplication
* feat: Add experimental eBPF support for enhanced socket tracking
- Implement eBPF-based socket tracker for Linux with CO-RE support
- Add minimal vmlinux header (5.5KB) instead of full 3.4MB file
- Create graceful fallback mechanism to procfs when eBPF unavailable
- Add comprehensive eBPF build documentation
- Integrate libbpf-rs for eBPF program loading and management
- Support both IPv4 and IPv6 socket tracking
- Add capability checking for required permissions
The eBPF feature is optional and disabled by default. When enabled,
it provides faster and more accurate process-to-socket mapping on
Linux systems with appropriate permissions.
