[mgradwohl]

Description

Part of #2728 (Linux helper coverage, after #2938): with the helper held, the Processes table's Net Sent and Net Received now cover other users' processes. The probe attributes each connection by mapping socket inodes to PIDs through /proc/<pid>/fd links. Another user's links are refused, so those connections stayed unowned and the columns read unavailable.

• Protocol v5: ProcessCountersRequest::withSockets (one byte, 0 or 1). Each ProcessCountersEntry carries socketInodes behind a third flags bit (4). Any other flag bit, or inodes without that bit, is refused. Inodes are capped at MAX_COUNTER_SOCKET_INODES (2^18) across the whole reply; an entry that would pass the cap goes back without its sockets but keeps its other figures.
• Helper: readProcessCountersAt(..., withSockets) reads every fd link through the same /proc/<pid> handle, after the start-time check. socketsRead is set only when the listing finished and no link was refused.
• Probe: BrokeredProcessProbe asks for sockets at the inner inode map's cadence (INODE_PID_CACHE_TTL_MS), not on every refresh, and only for processes whose network counters were refused (and only when the probe has network counters). readSocketTraffic() credits the connections the inner map left unowned to the helper's owners, with their start time. Between asks, the last read's processes keep their network columns. Like feat(linux): tasksmack-helper reads other users' FD counts and I/O bytes for the Processes table (#2728) #2938, it only uses a helper that's already running, never prompts, and stops asking after an ErrorReply.
• Other updates: fuzz seeds moved to v5 (the two counters seeds rewritten for the new layout), the dict, helper-handshake.py's v5 Hello, and docs in tasksmack.md and the FAQ.

PR Size

⚠️ Medium (~285 lines added, 40 files). Most of the file count is the 24 one-byte fuzz seed version bumps.

Type of Change

• New feature (non-breaking change that adds functionality)

Checklist

• I have read the CONTRIBUTING guidelines
• My code follows the project's coding standards
• I have run clang-tidy and addressed any warnings
• I have run clang-format on my changes
• I have run pre-commit run --all-files or installed pre-commit hooks
• I have added tests that prove my fix/feature works
• All new and existing tests pass
• I have updated documentation as needed

Testing

cmake --build --preset debug -j 4
TASKSMACK_REQUIRE_DISPLAY=1 TASKSMACK_REQUIRE_SDBUS=1 xvfb-run -a env -u WAYLAND_DISPLAY ctest --preset debug -j 4   # 682/682

New or changed tests:

• BrokerProtocolTest: the withSockets round trip and its 0/1 byte; the sockets flag bit; inodes refused without it; AnEntryWhoseSocketsWouldPassTheCapGoesWithoutThem.
• ProcessCountersReadTest.WithSocketsListsTheInodeOfASocketItHolds, which opens a socket and finds its st_ino.
• BrokeredProcessProbeTest.TheHelpersSocketInodesAttributeConnectionsTheProbeLeftUnowned: one ask, attribution of the matching inode only, no second ask before the cadence, and the attribution still standing.
• PkexecBrokerTest...ProcessCounters now asks for sockets from the real helper under its enforcing seccomp filter ([Security] tasksmack-helper: add a seccomp allow-list (and maybe Landlock) after the capability drop; UID 0 + CAP_SYS_PTRACE isn't containment #2752), so the link reads are covered by the allow-list.

Additional Notes

v0.10.0, no backport. Still to do for #2728 on Linux: hidepid listing through the helper, service actions, and the title bar's text.

🤖 Generated with Claude Code

[github-actions]

Dependency Review

✅ No vulnerabilities or license issues or OpenSSF Scorecard issues found.

Scanned Files

None