[oferchen]

Summary

A daemon session's process now exits with the code its transfer ended on, and the daemon log carries log_exit()'s lines in place of an invented summary, as rsync 3.5.1's per-connection child does.

• Exit status (cleanup.c:213-218, :264):
  • A transfer that runs to its end maps io_error and got_xfer_error to its code via core::exit_code::io_error_exit_code, so a partial transfer exits 23.
  • A fatal transfer exits with the peer's MSG_ERROR_EXIT code if one arrived (io.c:1930), otherwise with its own RERR code.
  • The code goes into session_exit_code. The forked TCP child now exits with it; before, it dropped it via .map(|_| ()) and always exited 0. The stdio and inetd sessions already turned that field into their exit status.
• Daemon log (log.c:937-963, log.c:310-327):
  • A fatal error logs the cause line and the rsync error: ... (code N) trailer from fatal_report_lines, the same lines the client is sent. This replaces the invented transfer failed to HOST (IP): module=M error=E line.
  • The sent ... received ... total size line is written only for a clean exit or a sender (log.c:941).
  • A partial transfer's end-of-run exit logs no trailer (am_server = 2, cleanup.c:121-122).
• Abortive close on the receiver now keys on a fatal error rather than on a nonzero code, so a partial push (code 23) keeps the graceful close.

Evidence

All cells are measured against the 3.5.1 tarball build. Server exit statuses come from strace (TCP daemon) or an rsh shim (daemon mode via --rsync-path).

cell	3.5.1 server	oc before	oc after
pull of a missing file	exit 23, totals	exit 0, totals	exit 23, totals
clean pull	exit 0, totals	exit 0	exit 0, totals
push with an unreadable source subdir (receiver io_error)	exit 23, no totals, no trailer	exit 0, totals	exit 23, no totals, no trailer
push into a 0555 dir	exit 11, mkdir cause + trailer	exit 0, transfer failed to line	exits with the trailer's code, trailer logged (3, see residuals)

tests/daemon_session_exit_status.rs covers four cells in rsh daemon mode and asserts the server process's exit status and the daemon log. It is deterministic: the oc client reaps its rsh child before exiting. The unreadable-dir cell skips with a printed reason when run as root.

Mutants, all killed:

• A partial transfer exits 0: two cells.
• A receiver logs totals on code 23: the io_error push cell.
• Nothing is logged on a fatal error: the fatal push cell.
• session_exit_code is not set: three cells.
• A fatal error exits 1: the fatal push cell.

Residuals

• Push into a 0555 dir: oc reports code 3 where 3.5.1 reports 11 with the mkdir ... failed cause. That is fix: report a failed destination-root mkdir as RERR_FILEIO on every path #8344 (local) and task 611 (daemon); this PR makes the session exit with whatever code is reported.
• Trailer role tag: it says [generator= on the receiver; 3.5.1 says [Receiver= for the pre-fork mkdir.
• Sender totals on a fatal exit: oc's sender logs none, because it has no stats on the error path. 3.5.1 logs them even on error (log.c:941).
• Missing source: the oc daemon sender logs neither the link_stat ... failed line nor the real totals. That is task 280.
• Forked TCP child's exit status: a test can't observe it. The parent records it only at --debug=EXIT, and upstream discards it. The strace oracle cell above covers it, and the rsh cells exercise the same session_exit_code path.

Test plan

• python3 tools/ci/check_rustfmt_all.py
• Workspace clippy with all features, plus no-default-features clippy of daemon
• cargo nextest run -p daemon --all-features: 2152 passed
• Targeted -p bin (178), -p core and -p transfer (530) daemon, inetd and stdio tests
• 3.5.1 oracle: TCP strace exits and rsh daemon-mode statuses and logs