RMS_Update: restore bounded sudo prompt and raise UDP buffers to 16MB

[Cybis320]

Two related changes to RMS_Update.sh, kept as separate commits.

1. Restore the bounded sudo password prompt

sudoWithTimeout() was added in 52b085ec and removed in c5fd3ab5 ("Prevent corruption of config and mask file", a 384-line restructure). Nothing in that commit was about sudo and no reference to it survives, so this looks like collateral of the rewrite rather than a decision.

Without it, the fallback branch runs a bare sudo apt-get and waits on sudo's own prompt. Sudo's passwd_timeout defaults to 0 (no timeout), so an unattended run can block indefinitely — which matters because GRMSUpdater.sh stops all stations before calling this script.

Reinstated as ensure_sudo "<reason>": 30s read -t timeout, 3 attempts, background keep-alive to survive a long apt-get. install_missing_dependencies now goes through it, collapsing the apt-get loop that was previously written out twice.

Four deliberate differences from the original:

• the keep-alive is trapped (stop_sudo_keep_alive from emergency_cleanup and a new EXIT trap) — the original leaked it
• background refresh uses sudo -n -v so it can never compete for the terminal, plus a kill -0 "$$" guard
• printf instead of echo for the password, which echo can mangle on backslashes
• returns 1 without prompting when no terminal is attached, so non-interactive callers skip instead of hanging

2. Raise UDP buffers to 16MB during update

BufferedCapture.py asks rtspsrc for a 16MB udp-buffer-size, but the kernel clamps that to net.core.rmem_max. Where the limit is lower the request is a no-op and frames drop under load. UpdateBuffers.sh already sets this, but nothing ever called it, so stations only got it if run by hand.

• UpdateBuffers.sh gains --check (unprivileged, exit 0 adequate / 10 needs raising) and --yes; argument parsing moved above the root gate so --check works without sudo
• fixes a silent no-op: with no tty and no --yes, the confirm prompt used to read empty, print "No changes made" and exit 0 — indistinguishable from success to a caller. It now fails loudly
• applying verifies the values actually took instead of assuming
• RMS_Update.sh gains update_udp_buffers() after the dependency step, plus --skip-buffers

The check is unprivileged, so sudo is only requested when there is real work: one prompt per station, silent no-op on every update after. It always returns 0 — buffer sizing affects capture at runtime and should never fail a build.

Testing

Against a mock Linux sysctl (macOS has no net.core.*):

• --check at 208KB / 1MB / 16MB returns 10 / 10 / 0
• no tty without --yes exits 1; with --yes proceeds; already-adequate exits 0 without applying
• timeout path measured at exactly one 30s interval, not 3x
• all three update_udp_buffers paths return 0
• both scripts pass bash -n

Nothing was executed against /etc — the apply stage was stubbed in the test copy.

Note

GRMSUpdater.sh:595 now takes the skip path and prints the manual command rather than stalling with stations stopped. That also means multi-camera hosts won't get the buffer bump automatically and will need one manual sudo ./Scripts/UpdateBuffers.sh --yes. If that should be automatic, the cleaner fix is for GRMSUpdater to establish sudo up front, before stop_stations.

🤖 Generated with Claude Code