[mgradwohl]

Description

Part of #2728 (Linux helper coverage): with the helper held, the Processes table's Handles/FDs, I/O Read and I/O Write now fill in for other users' processes. Until now they stayed blank because /proc/[pid]/fd and /proc/[pid]/io refuse a non-root reader.

• Protocol v4: a batched ReadProcessCounters request (39) and ProcessCountersResult (56). One round trip carries every target, capped at MAX_COUNTER_TARGETS (16384). Each entry has a flags byte (bit 1 = handle count read, bit 2 = I/O read); any other bit is refused.
• Helper: the new ProcessCountersRead.h opens /proc/[pid], checks the start time, counts FDs with ProcFdScan, and parses io with ProcIoParser.h. That parser is now shared with LinuxProcessProbe, which drops its own copy. A target that has gone or been reused is left out of the reply.
• Mapping: Readers::processCounters / handleProcessCounters. A helper without the reader answers ErrorReply. A reader that throws or finds nothing leaves out just that target.
• Probe: BrokeredProcessProbe::enumerate() sends one callIfIdle batch (1 s timeout), and only for processes whose counts were refused. It only asks a helper that is already running, so it never prompts. After an ErrorReply (an older helper) it stops asking.
• Windows: the pipe server classifies the new request as a read (its static_assert count is now 13). Windows doesn't serve it yet.
• Other updates: fuzz corpus moved to v4 with two new seeds and an updated dict; tools/helper-handshake.py knows v4's Hello; docs updated in tasksmack.md and the FAQ.

PR Size

🔴 Large (~610 added / 76 removed across 42 files). Most of it is tests, the 20 regenerated one-byte-version corpus seeds, and the protocol's encode/decode boilerplate. The behavior change is the helper reader plus the probe's single batch.

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   # all pass

New tests:

• BrokerProtocolTest: round trip, target cap, and refused flag bits.
• BrokerRequestMappingTest: no reader, a throw, nullopt.
• BrokeredProcessActionsTest: the probe fills refused counters, and stops after an ErrorReply.
• ProcessCountersReadTest and ProcIoParserTest.
• PkexecBrokerTest.TheHelperReadsProcessCountersAndLeavesOutAStaleTarget: the real helper binary over its socketpair.

WSL's kernel has no task I/O accounting (/proc/self/io is missing). The I/O assertions therefore expect ioRead exactly when /proc/self/io exists, so they hold on both kinds of kernel.

Additional Notes

v0.10.0 work, so no backport. The rest of #2728's Linux list (network attribution, hidepid listing, service actions, the title bar's text) follows in separate PRs.

🤖 Generated with Claude Code

[github-actions]

Dependency Review

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

Scanned Files

None