Skip to content

Skycontroller 2 mpp-linux-1.0.7: services, boot and network exposure

Static read of the live Skycontroller 2’s system volume (mpp-linux-1.0.7) by one agent, 2026-10-04. Status: Agent-cited, with these claims re-checked by the commissioning session against the copy: the single inetd.conf line (FTP on tcp 21, root, -wS, rooted at /var/lib/ftp), no iptables reference in etc or usr/bin (control: the same search finds ftpd in inetd.conf), no hostapd.conf, dnsmasq.conf or usr/bin/dnsmasq (the binary is usr/sbin/dnsmasq), macgyver.conf auto_send_https_report = 0, the adbd string checks (ro.adb.secure present as a string, no setuid, setgid or capset, socket_inaddr_any_server present, the wildcard listener helper), the usbnetd strings (dnsmasq arguments, interfaces eth*,usb*), and the service count (22). Not re-derived: the service table details, the boot-stage order, the macgyverd internals and everything marked Inference. This describes what the image configures, not runtime state.

Difference from the nap image worth stating up front: this unit has no Wi-Fi access point at all (no hostapd, no AP-mode service); the phone link is USB. The earlier discussion of the AP’s WPA2 key applies to the NG image only. The exposure section describes configuration read from the system volume. Where a row says Inference, whether a service listens on a running unit was not observed.

Static analysis only. Image root: the extracted system volume (contains bin sbin lib usr etc). Paths below are relative to it. Nothing was run, no device or network host was contacted, nothing was written into the extract. Tags: Verified (file or bytes read), Agent-cited (taken from the task brief or a prior report, not re-derived here), Inference, Open. Every negative result was checked with /usr/bin/grep -a against a positive control (a string known to exist in the same search scope) before it was trusted. No Wi-Fi keys, serials, MACs or device ids appear in this report; the image holds none of them in the files read.

  1. Boot is boxinit (etc/boxinit.rc imports boxinit.dragonbase.rc, which imports the 17 files in etc/boxinit.d/). They declare 22 services in classes core (4), main (15) and mpp (3). console is the only service in boxinit.rc. Verified.
  2. etc/default.mppboard.prop:2-3 sets ro.debuggable=1 and service.adb.tcp.port=9050. That starts a root console on the UART (ttyPA3), adbd and ulogcatd (boxinit.rc:14-15, 50-adb.rc:2-3, 90-ulogcatd.rc:11-12). adbd imports no setuid, setgid or capset call, so it stays root. ro.adb.secure is set nowhere in the image. Verified for files; listening state is Inference.
  3. etc/inetd.conf:1 is the only inetd line: ftpd -wS /var/lib/ftp on tcp 21 as root, upload enabled. The nap image had two such lines; this one has no port 51 and no update FTP root. The macgyver crash report folder sits under this FTP root. Verified.
  4. There is no Wi-Fi access point and no filter or NAT firewall in this image. wifid runs STA only on the Broadcom bcm43526 (50-wifi.rc:1), with DHCP and a static fallback of 192.168.42.3. No hostapd exists and nothing invokes iptables. Verified.
  5. Phone link: usbnetd watches eth*,usb*, tries udhcpc first and, if no lease arrives, runs dnsmasq as a DHCP server (DNS off) at 192.168.N.1/24 with range .10 to .100 on that interface. aoad (Android Open Accessory, host side) and iosd (iAP2 serial gadget g_aea) carry ARSDK over USB. avabye announces mDNS on eth*,usb* only, not on the drone-facing wlan0. Verified (config and code); N derivation Open.
  6. macgyverd is the same Parrot crash reporter as on the Bebop 2 (different binary, same config shape). It runs at every boot as root, collects oops, crash dumps and generic reports, and keeps them in /var/lib/ftp/internal_000/Debug/crash_reports. auto_send_https_report = 0 in etc/macgyver.conf:15. The binary has no path that sends without that flag. Verified (config, strings); runtime Inference.
  7. usbctrl-mon resets the USB controller and PHY when it sees a dead host controller, and raises a generic report. memtester is started by a udev rule when a USB device with vendor 19cf and product ffff is plugged in, and holds the unit powered while it runs. Verified.
  8. Not in the image: firewall policy applied outside it, runtime property overrides, the AP-less drone credentials store, /factory, /data, /update, and any /customer_prod (nothing in the image references that name at all).

sbin/boxinit contains the stage strings early-init, keychord_init, post-fs, post-fs-data, early-boot, early-boot-main, queue_property_triggers, load_persist_props and reads /etc/default.prop, /etc/default.%s.prop, /etc/build.prop (Verified, strings). The full stage order is the same as in the nap report (Agent-cited, sc-services-network.md Boot stages); stage names used by this image’s rc files are early-init, early-fs, init, fs, post-fs, post-fs-data, early-boot, early-boot-main, boot, boot-main.

etc/default.mppboard.prop is the only default prop file, and its name implies ro.hardware=mppboard, which is also how libnetmon-bcm43526b.so builds /etc/wifi/%s_config.yaml from ro.hardware (Verified for the format string, Inference for the value).

Stage Action File:line Status
early-init export PATH, BOXINIT_DATA=/data, TERM, TMPDIR/TMP/TEMP=/tmp, HOME=/data/home/root etc/boxinit.dragonbase.rc:6-15 Verified
init TERM; loglevel 5; printk 1; overcommit 1/90; alignment 5; panic_on_oops 0; modprobe path; wmem_max 1 MB etc/boxinit.rc:44-65 Verified
init pulse the Wi-Fi chip reset gpio (RESET_WIFI 1, 70 ms, 0) etc/boxinit.rc:67-70 Verified
init write ulog_kmsgd 17 to the ulog main node etc/boxinit.d/90-kmsgd.rc:2-3 Verified
init clear /proc/sys/kernel/hotplug (udevd takes over) etc/boxinit.d/30-udev.rc:26-27 Verified
early-fs mkdir /dev/shm; mount tmpfs on it and on /tmp; create /tmp/{cache,lock,run,tmp,log,etc}; empty /tmp/etc/resolv.conf etc/boxinit.dragonbase.rc:23-34 Verified
fs mount ubifs ubi0:factory on /factory ro; ubi2:data on /data rw; ubi2:update on /update rw (all no_chk_data_crc nosuid nodev) etc/boxinit.rc:73-75 Verified
fs if /update/.factory_reset exists: run factory_reset then delete the flag etc/boxinit.rc:76-77 Verified
post-fs mkdir /tmp/avahi, /tmp/avahi/services (etc/avahi/services is a symlink to it) etc/boxinit.d/50-avabye.rc:1-3 Verified
post-fs-data mkdir /data/property 0700, /data/lib 0755 system, /data/home, /data/home/root etc/boxinit.dragonbase.rc:18-21 Verified
post-fs-data mkdir /data/lib 0777, /data/lib/ftp 0777, /data/lib/misc 0777; set persist.periodic.log=1, persist.wifi.periodic.log=1 etc/boxinit.rc:81-87 Verified
post-fs-data mkdir /var/lib/crashdump etc/boxinit.d/30-crashdump.rc:1-2 Verified
early-boot setprop ro.factory.{serial,pcb_serial,wifi.mac_address,bt.mac_address,js.centers,js.min_max} from /factory/*.txt via ${file:...} etc/boxinit.rc:19-29 Verified (values not read)
early-boot ro.product.model="Skycontroller 2", ro.product.model.id="0x090f", ro.updater.filename="mpp_update.plf", five ro.sensors.* names etc/boxinit.rc:31-41 Verified
early-boot-main wait for /dev/.udev_perms_done etc/boxinit.d/30-udev.rc:30-31 Verified
boot ifup lo, hostname mpp, domainname localdomain, class_start core etc/boxinit.rc:91-96 Verified
boot setprop load_g_aea 1 (iosd) ; mkdir /data/lib/ftp/internal_000/Debug/crash_reports 0777 (macgyver) etc/boxinit.d/98-iosd.rc:1-2, 90-macgyver.rc:2-5 Verified
boot-main class_start main, then class_start mpp etc/boxinit.rc:98-102 Verified
property ro.debuggable=1: start console, adbd, ulogcatd etc/boxinit.rc:14-15, 50-adb.rc:2-3, 90-ulogcatd.rc:11-12 Verified
property load_g_aea=1: insmod g_aea.ko app_protocol=com.parrot.dronesdk idProduct=0x5040, then reset the property etc/boxinit.d/98-iosd.rc:8-11 Verified
property ro.revision=0: stop iosd (comment: usb0 bug on hardware 0) etc/boxinit.d/98-iosd.rc:12-15 Verified
property wifi.watchdog.counter=*: stop wifid, pulse Wi-Fi and USB hub reset gpios, unbind and rebind ci_hdrc.1, start wifid and broadcom-report etc/boxinit.d/50-wifi.rc:13-25 Verified
property init.svc.memtester=stopped: stop memtester_led, mppd.led.enable=1 etc/boxinit.d/99-memtester.rc:11-13 Verified

Differences from the nap boot that matter. There is no mpp.services.start gate: class mpp starts straight at boot-main. There is no statexec of a factory reader, so ro.factory.* come from plain text files on /factory. There is no /customer_prod mount. boxinit.base.rc (which computes rw.debuggable) exists but nothing imports it: only boxinit.rc:3 and boxinit.dragonbase.rc:4 contain import, and neither names it (Verified, with the import lines as positive control). The triggers key on ro.debuggable, not rw.debuggable.

Properties set in rc files: ro.factory.* (6), ro.product.model, ro.product.model.id, ro.updater.filename, ro.sensors.* (5), persist.periodic.log, persist.wifi.periodic.log, load_g_aea, mppd.led.enable. Set outside the rc files by binaries: wifi.watchdog.counter (bcmdriver.so strings), iosd.charge.device and others (iosd strings). Read by rc triggers: ro.debuggable, load_g_aea, ro.revision, wifi.watchdog.counter, init.svc.memtester. Where ro.revision and ro.serialno are set is Open (not in any rc file).

22 services: 21 in 17 files under etc/boxinit.d/, plus console in etc/boxinit.rc. Columns follow the nap report. - means the keyword is absent. When user is absent boxinit’s default is Inference (root).

File:line Service Command line as written Class User Group Flags Capability, setenv, socket Trigger Status
boxinit.rc:6-11 console /usr/bin/askfirst /bin/sh -l core root log console, disabled - ro.debuggable=1 Verified
30-crashdump.rc:4-8 crashdumpd /usr/bin/crashdump -d main root - - capability SYS_PTRACE class main Verified
30-udev.rc:21-24 udevd /usr/bin/ulogwrapper /usr/bin/udevd core root inet - - class core Verified
40-jubamountd.rc:1-6 jubamountd /usr/bin/jubamountd core root - - capability SYS_ADMIN SETGID class core Verified
50-adb.rc:5-13 adbd /usr/bin/adbd core - - disabled socket adbd stream 660 0 0; 23 capabilities incl. SYS_MODULE, SYS_RAWIO, SYS_ADMIN, NET_ADMIN ro.debuggable=1 Verified
50-avabye.rc:5-7 avabye /usr/bin/avabye main root - - - class main Verified
50-usbnetd.rc:1-3 usbnetd /usr/bin/usbnetd main root - - - class main Verified
50-wifi.rc:1-5 wifid /usr/bin/wifid --mode STA --ip 192.168.42.3 --dhclient main root - - setenv WIFID_DRIVER bcmdriver, BCM_MODULES_DIR_NAME bcm43526-extra class main Verified
50-wifi.rc:7-11 broadcom-report /usr/bin/report-generic broadcom-watchdog main root - disabled, oneshot - started by the wifi.watchdog.counter trigger Verified
50-wifi.rc:27-31 wifid-dhclient /usr/bin/ulogwrapper /sbin/udhcpc -S -f -q -n -A10 main root - disabled, oneshot - started by wifid (@wifid-dhclient socket) Verified (rc); starter Inference
60-inetd.rc:1-4 inetd /usr/bin/ulogwrapper /usr/sbin/inetd -f main root root - - class main Verified
60-sensorsd-mpp.rc:1-3 sensorsd-mpp /usr/bin/sensorsd-mpp main root - - - class main Verified
90-aoad.rc:1-3 aoad /usr/bin/aoad mpp root - - - class mpp Verified
90-kmsgd.rc:5-11 kmsgd /usr/bin/kmsgd main root - - capability SYSLOG; setenv ULOG_DEVICE kmsgd class main Verified
90-macgyver.rc:7-10 macgyverd /usr/bin/macgyverd -f main root root - - class main Verified
90-ulogcatd.rc:2-8 ulogcatd /usr/bin/ulogcat -F /etc/ulogcatd.conf main root log inet disabled capability SYSLOG ro.debuggable=1 Verified
98-iosd.rc:4-6 iosd /usr/bin/iosd mpp root - - - class mpp; stopped when ro.revision=0 Verified
98-usbctrl-mon.rc:1-4 usbctrl-mon /usr/bin/usbctrl-mon main root - - - class main Verified
98-usbctrl-mon.rc:6-9 usbctrl-report /usr/bin/report-generic "usb ci_hdrc.0 fatal error" main root - disabled, oneshot - started by usbctrl-mon (ctl.start usbctrl-report, strings) Verified (rc); starter Verified by strings
99-memtester.rc:1-4 memtester /usr/bin/memtester.sh main root - disabled - udev rule 70-memtester.rules via pstart Verified
99-memtester.rc:6-9 memtester_led /usr/bin/memtester_led.sh main root - disabled - started by memtester.sh via pstart Verified
99-mppd.rc:1-3 mppd /usr/bin/mppd mpp root - - - class mpp Verified

Started automatically at boot (no disabled): udevd, jubamountd, crashdumpd, avabye, usbnetd, wifid, inetd, sensorsd-mpp, kmsgd, macgyverd, usbctrl-mon, aoad, iosd, mppd. Debug-gated: console, adbd, ulogcatd. Every user line says root, and adbd, which has no user line, inherits root (Inference for the default). No rc file drops to an unprivileged user for any daemon.

Diff against nap (SKYCTRL_NG, nap-linux-2.0.0-rc3)

Section titled “Diff against nap (SKYCTRL_NG, nap-linux-2.0.0-rc3)”

Only on the Skycontroller 2: usbnetd, macgyverd, iosd (plus module g_aea), usbctrl-mon and usbctrl-report, memtester and memtester_led, broadcom-report, and the Wi-Fi reset-on-watchdog trigger. Support files: etc/macgyver.conf, the Broadcom driver (usr/lib/wifid/bcmdriver.so, usr/lib/libnetmon-bcm43526b.so, lib/modules/extra/bcm43526*, lib/firmware/brcm/*), etc/wifi/*.yaml, usr/share/netmon/*.yaml, bottomboard.hex, dfu-programmer.

Only on nap: gpsd, btgpsdemux, hudd, the second Wi-Fi instance wifid-ap (AP on wlan1, 192.168.44.1), dnsmasq_wlan1 and its start_dnsmasq_wlan1 script (NAT of controller ports 121/151/161 to the drone’s FTP ports plus IP forwarding), the second FTP service on port 51 and its update_ftp root, the mpp.services.start gate, the factory reader at early-boot, TI/Atheros Wi-Fi modules. All from the nap report (Agent-cited).

Same on both: console, crashdumpd, udevd, jubamountd, adbd, avabye, wifid (STA), wifid-dhclient, inetd, sensorsd-mpp, aoad, kmsgd, ulogcatd, mppd, and the same ro.debuggable=1 plus TCP 9050 pair in the board prop file.

Correction to the task brief: usr/sbin/dnsmasq is present in this image. It is run only by usbnetd, with arguments built in code and no config file (etc/dnsmasq.conf does not exist). hostapd, gpsd and hudd are absent. Verified: ls usr/sbin shows dnsmasq and shows no hostapd, with ftpd, inetd, telnetd and iptables listed as controls.

Section titled “Drone link (wifid, bcm driver, yaml files)”

wifid (etc/boxinit.d/50-wifi.rc:1) runs --mode STA --ip 192.168.42.3 --dhclient: it joins the drone’s network as a client, asks for a lease with udhcpc and falls back to static 192.168.42.3 if that fails (usage text and strings in usr/bin/wifid; Verified). The drone side is 192.168.42.1, a literal in mppd (Verified). The Broadcom path is WIFID_DRIVER=bcmdriver (loads usr/lib/wifid/bcmdriver.so, which links libnetmon-bcm43526b.so) and BCM_MODULES_DIR_NAME=bcm43526-extra (selects lib/modules/extra/bcm43526-extra/{bcm_dbus,wl}.ko; a second, different wl.ko exists in bcm43526/). Which of the two is loaded is decided by that env name (Inference). Firmware and calibration: lib/firmware/brcm/bcm43526-firmware.bin and bcm43526-ffff-ffff.nvm.

Settings that wifid reads and writes live in /var/lib/libshsettings/wifi.cfg (writable data partition, so a runtime copy may differ). The image default etc/libshsettings/wifi.cfg:1-9: outdoor = true, autocountry = false, country = "US", autoselect = true, band = 4, channel = -1. Meaning of band = 4 is Open. mppd stores known drones and their Wi-Fi keys through its config module (strings known_drones, wifi_key, drone protected: wifi_key=... given by mobile); etc/mppd/drone_manager.cfg:1 is known_drones = ( ); in the image and any real entries would live in a runtime copy (Verified for the empty default, Open for the store).

bcmdriver.so has netmon_infra_security_set_wpa2 and netmon_infra_security_set_open, so it can join either a WPA2 or an open network (Verified, symbols); which one is used follows what the phone tells mppd (SET_SSID, WPA2, WPA2_PRO strings in mppd; Inference).

etc/wifi/*.yaml is looked up as /etc/wifi/%s_config.yaml with %s = ro.hardware (libnetmon-bcm43526b.so strings; Verified). Per file:

File Describes Settings (no secrets) Status
etc/wifi/mppboard_config.yaml:1-19 This unit (board name matches default.mppboard.prop; text says “on mpp”) ampdu_rx_wsize: 2, rts_enable: 1, custom Broadcom regulatory version 44 mapped to country US (comment: forbids 5 GHz in the US) Verified (content); applies to this unit Inference
etc/wifi/evinrudeboard_config.yaml:1-4 Another Parrot board, name only Broadcom version 1 mapped to country JP Verified (content); product Open
etc/wifi/mykonos4_config.yaml:1-18 A bcm43602-based board (comment names 43602) rx/tx chains 7, slot_time_us: 20, A-MPDU rx/tx and window sizes -1 Verified (content); product Open
etc/wifi/paros_config.yaml:1-18 Identical to mykonos4 (bcm43602) same as mykonos4 Verified (content); product Open

Only the first applies to a bcm43526 unit, since the other three name other boards and a different chip; they are shipped in the same package (Inference). usr/share/netmon/device_perf.yaml has per-device gain and power tables for Bebop (old name), mykonos3board and SkyController (Verified, device: lines 2, 50, 97), and country_regul.yaml holds the channel authorisation per country, with US, CA and IL marked autolock: 1 in netmon_config.yaml.

wifi.watchdog.counter is set by the driver when its watchdog fails (strings “watchdog KO starting recovery”, “Restart Broadcom module”); the rc trigger then resets the Wi-Fi chip and the USB hub, rebinds the USB controller ci_hdrc.1, restarts wifid and files a generic report named broadcom-watchdog (Verified).

mppd is the ARSDK proxy between phone and drone. Its strings show four proxy roles: proxy_drone, proxy_mpp (the controller as a device), proxy_mobile_mux and proxy_mobile_net (Verified, symbols). It creates an ARSDK net backend (arsdk_backend_net_new, start_listen) with Avahi discovery and publishing (arsdk_publisher_avahi_new, arsdk_discovery_new) and a mux backend with a mux publisher over USB (arsdk_backend_mux_new). It accepts or rejects a phone peer by name and id (peer connection requested: name=..., id=..., rejected peer). What decides acceptance is Open.

mDNS: etc/avabye.conf:4 is INTERFACE=eth*,usb* and etc/avabye.conf:8 is SERVICE_DIR=/etc/avahi/services (symlink to /tmp/avahi/services). avabye has no wlan pattern, so it does not answer or announce on the drone-facing wlan0 (Verified, config; the interface match is by pattern against interface names, Inference). The service file is written at runtime by arsdk (libarsdk.so has /etc/avahi/services/arsdk-ng.service and the format _arsdk-%04x._udp; Verified). With ro.product.model.id = 0x090f that gives _arsdk-090f._udp, which matches the live announcement (Agent-cited, the brief). The port, 44444, is Agent-cited from the live unit: the literal 44444 does occur in usr/bin/mppd, but its only reference is a source-line constant in a log call (movw 0x4c4, line number, fcn.00024504), so it is not evidence of a port number and the listen port is Open in the image.

Other mppd network facts (Verified, strings): a video restream path, with an HTTP request socket (GET /video) and an SDP reply that names 192.168.53.1, RTP to an IP taken from the property mppd.restream.ip, ports 55004 and 55005 as client ports in a JSON blob, and 5004, 5005, 51 and 21 as the drone’s stream and FTP ports in another JSON blob. A debug server is gated on the property mppd.mobile.debug and the string says it listens on 127.0.0.1 only (“DEBUG enabled: listen on 127.0.0.1:%d”; Verified for the string, runtime Inference).

USB to a phone:

  • Android: aoad (class mpp) watches udev for USB devices, probes for the Android Open Accessory protocol and keeps a per-device socket @/com/parrot/aoad/<bus>/<dev> (strings). mppd switches an eligible device into accessory mode, one device at a time, and persist.aoa.uri plus a literal Google Play URL for the FreeFlight 3 app are the accessory URI the phone is shown (strings). The controller is USB host on this path. Verified (strings); the phone shows the URI, the controller fetches nothing.
  • iOS: iosd (class mpp) loads gadget module g_aea (Apple External Accessory iAP2) with app_protocol=com.parrot.dronesdk. It presents a USB serial pair (ttyGS0, ttyGS1; mppd opens /dev/ttyGS1). The module has no Ethernet function strings. iosd also negotiates USB role switch and sets the charge current for the iOS device (strings). Verified (strings); iosd stops on ro.revision=0.
  • Wi-Fi AP: none. No hostapd, no AP service in rc, and mppd prints “No wifi_ap on this platform” (Verified for the string; absence searches above).

etc/inetd.conf:1: 21 stream tcp nowait root /usr/bin/ulogwrapper /usr/bin/ulogwrapper /usr/sbin/ftpd -wS /var/lib/ftp (Verified). One line only.

Port Handler User Root dir Notes
tcp 21 busybox ftpd -w -S root (nowait root) /var/lib/ftp, which is /data/lib/ftp if /var/lib is the symlink the nap image uses (/var is not in this extract, so Inference) upload enabled, syslog logging, directory created 0777 (boxinit.rc:82)

Binding: inetd binds the wildcard address with no address in the config line; service.adb-style interface restriction is absent. So the service answers on every interface that has an address: wlan0 (drone network), usb* and eth* (Inference for the wildcard bind; Verified that the line has no bind address).

BusyBox is v1.25.0 (Verified, strings). Its ftpd text calls itself an “Anonymous FTP server”, takes only -w, -v, -S, -t, -T and DIR, and the image has no /etc/shadow or /etc/ftpusers. The ftp user in etc/passwd:11 is not named on the inetd line. So the server asks for no credential the image could check (Inference from the usage text; the login code was not disassembled).

etc/default.mppboard.prop:1-3 (Verified): ro.boot.console=ttyPA3, ro.debuggable=1, service.adb.tcp.port=9050.

adbd (usr/bin/adbd, an AOSP-derived build with packages/adb/adb/adb.cpp and adbd_auth.cpp paths in its strings): it contains service.adb.tcp.port, persist.adb.tcp.port, service.adb.root, ro.adb.secure, auth_required, /data/misc/adb/adb_keys, tcp:%d, socket_inaddr_any_server, socket_loopback_server, and the key-auth messages (Verified, strings). It does not contain ro.secure, ro.debuggable, setuid, setgid, setgroups, capset or prctl (Verified: each searched with /usr/bin/grep -c -a -w, with ro.adb.secure and listen as positive controls). So it has no code to drop privileges and runs with what boxinit gives it, which is root with the capability list in 50-adb.rc:9-13. ro.adb.secure is set in no file of the image (search of etc usr bin sbin lib returns only usr/bin/adbd, Verified). Whether adbd treats an unset ro.adb.secure as “no authentication” is Inference (AOSP behaviour for that era); adb key storage is /data/misc/adb/adb_keys, which is not in the image.

Binding: with service.adb.tcp.port set, AOSP-style adbd listens on tcp:9050 on all addresses (Inference from the inaddr_any symbol and the AOSP code path). It also has a local adbd unix socket created by boxinit (50-adb.rc:8). USB adb: adbd expects /dev/usb-ffs/adb/ep0 or /dev/android_adb; nothing in etc/ mounts functionfs or configures an adb gadget (usb-ffs appears in adbd and nowhere under etc, Verified), and the only gadget this image loads is g_aea. So adb over USB is not configured by this image (Inference; kernel and runtime not seen).

console: root shell via askfirst /bin/sh -l on ttyPA3 when ro.debuggable=1 (Verified, rc). askfirst is a small helper that waits for a key; no password step is in the rc (Inference that the shell is then exec’d directly). Physical access only.

ulogcatd (debug-gated): etc/ulogcatd.conf:7,17,21: kernel and ulog messages in long format to /update/log, two files of 16 MB. The -p 7016 TCP serve line (line 10) and the control socket line (line 27) are commented out (Verified).

iptables: the binaries exist (usr/sbin/iptables, ip6tables, -save, -restore link to xtables-multi), plus modules nf_conntrack_ftp.ko and nf_nat_ftp.ko under lib/modules/3.4.11+/kernel/net. No file under etc, usr/bin, usr/lib/wifid, bin, sbin/boxinit mentions iptables, ip_forward, nf_conntrack or nf_nat (Verified: search over those paths returned only libip4tc.so and libip6tc.so for iptables, with dnsmasq returning boxinit.rc, wifid and usbnetd as the positive control). Not searched in the same pass: the large binaries mppd, libarsdk.so and busybox for the netfilter terms. mppd and wifid were read as strings and show no iptables call. So no rule is installed by anything in the image: no INPUT filter, no NAT, no IP forwarding (Verified for the image, Inference for runtime, Open for outside).

Other sockets (strings, so Inference for runtime): crashdump @crashdump abstract unix socket, wifid @wifid and @wifid-dhclient, jubamountd @/org/parrot/jubamountd, aoad @/com/parrot/aoad/.., sensorsd-mpp local socket. telnetd, ntpd, tcpdump, iperf, strace and htop binaries are present, but no rc file or inetd line starts telnetd or ntpd (Verified: search of etc for telnetd returns only inetd.conf for ftpd with ftpd as control and nothing for telnetd).

usbnetd (class main, root, 12 KB) is a netlink-driven interface manager. Verified from its strings and from disassembly of its three key functions (radare2, static):

  1. Interface patterns: INTERFACES=eth*,usb* is the built-in default, overridable by /etc/usbnetd.conf (variables INTERFACES and DHCLIENT_TIMEOUT). etc/usbnetd.conf does not exist in the image, so the defaults apply. It enumerates /sys/class/net at start and follows netlink events for added or removed interfaces; an interface that is down is brought up.
  2. For each matching interface it first forks udhcpc with arguments -i <if> -t<DHCLIENT_TIMEOUT> -T1 -n -q -f -S and waits. If udhcpc obtains a lease, it logs “Interface %s managed by dhclient” and does nothing more: the controller is then a DHCP client of whatever is on the cable. The lease script is usr/share/udhcpc/default.script, which sets the address and installs default routes and DNS from the lease (Verified for the script head; reading the rest was cursory).
  3. If udhcpc fails, usbnetd sets the interface address to 192.168.N.1/24 through a netlink address request (prefix 24), then execs dnsmasq with --dhcp-leasefile=/var/lib/misc/dnsmasq-N.leases --dhcp-range=192.168.N.10,192.168.N.100,infinite --interface=<if> --bind-interfaces --dhcp-authoritative --port=0 --pid-file=/dev/null --dhcp-option=3 --dhcp-option=6. So the controller runs the DHCP server on that interface only, with DNS switched off, and hands out no gateway and no DNS (options 3 and 6 are passed empty). The log line is “Started DHCP server on %s with address 192.168.%d.1”. Verified.

Which side runs the server: the controller does, when nobody else on the cable answers DHCP; the client role is tried first. The value N is an integer stored in the interface’s table slot, 4 bytes after the 16-byte name; whether it is the interface index or a counter is Open (not traced to its writer). The drone-side subnet is not touched by this service.

What the phone needs for this path: an eth* or usb* interface must appear on the controller. The kernel module rndis_host.ko is shipped (Verified), which fits an Android phone in USB tethering mode, and the controller would be the DHCP client first. The g_aea gadget has no network function (Verified), so an iPhone does not create such an interface. Whether the controller ever shows up as a network device to a USB host is not found: there is no gadget Ethernet module in the list (g_aea and rndis_host only, plus pl2303, usbtouchscreen).

What it is: the Parrot on-board crash reporter, same name and same config shape as the Bebop 2 one (docs/notes/macgyverd-crash-reporter.md in bebop-2, Agent-cited). The binary is not identical to the Bebop 2 4.7.1 one (different SHA-256, Verified), and the config differs only in three lines: crash dump folder /var/lib/crashdump (Bebop: /data/lib/crashdump), report folder under /var/lib/ftp (Bebop: /data/ftp), and max_report_stored = 10 (Bebop: 20). Verified by diff of the two files with comments stripped.

Start: etc/boxinit.d/90-macgyver.rc:7-10, macgyverd -f, class main, root. It is not disabled, so it runs at every boot; the Bebop 2 page notes disabled plus an rcS start, which does not apply here (Verified for this image).

etc/macgyver.conf (Verified, with line numbers):

  • line 6 is_kernel_oops_allowed = 1, line 9 is_crash_dump_allowed = 1.
  • line 12 is_java_exception_allowed = 1 and line 21 path_to_java_exception_folder are not options this binary knows: the option names is_java_exception_allowed and path_to_java_exception_folder occur 0 times in usr/bin/macgyverd, while is_generic_report_allowed and auto_send_https_report occur once each (Verified, positive controls). The binary’s own option list is is_kernel_oops_allowed, is_crash_dump_allowed, is_generic_report_allowed, auto_send_https_report, path_to_crash_dump_folder, path_to_pstore_folder, path_to_generic_report_folder, path_to_report_folder, oops_partition_name, max_report_stored, path_to_build_properties, path_to_product_info_xml. So lines 12 and 21 are leftovers from the Android-era template and have no effect (Inference).
  • line 15 auto_send_https_report = 0.
  • line 18 path_to_crash_dump_folder = /var/lib/crashdump; line 24 path_to_report_folder = /var/lib/ftp/internal_000/Debug/crash_reports; line 27 /etc/build.prop; line 31 oops_partition_name = "" (use RAM oops); line 34 max_report_stored = 10.

What it collects (strings; Verified): kernel oops from ramoops or an MTD oops partition, crash_* directories from crashdumpd, and generic_* files written by report-generic (default folder /data/generic). libcrashdump.so is loaded into every process through etc/ld.so.preload:1, so a crash in any program reaches crashdumpd (Verified for the preload; the hand-off is Inference). A report is a numbered directory report_NNN with subject.txt and gzip files: crashdump.txt.gz, generic.txt.gz, ulogcat.txt.gz, klog.txt.gz (kernel log through klogctl), logcat, and info.txt.gz. The info file carries the ro.parrot.build.* properties, uname, uptime and memory figures, the kernel command line, the init.svc.* service states, the boxinit property list, and “factory” and filesystem figures read from /factory and /factory/production_info.xml. That means a report can carry product identity and possibly the serial (Inference: the code reads /factory and ro.factory.serial is a property; the exact fields were not decoded). Dedup is by SHA-1.

Where it sends or stores: stored locally under the FTP root (/var/lib/ftp/internal_000/Debug/crash_reports, created 0777 by 90-macgyver.rc:3-5). The binary has an HTTPS client (libssl 1.0.0, SSLv23_client_method, multipart POST with User-Agent: MacGyver to /cgi-bin/macgyver/reportcrash) and the host name crashreport.parrot.biz, plus an embedded self-signed certificate it adds to its own trust store; certificate subject is O=Parrot Automotive SAS, OU=OS Team, CN=crashreport.parrot.biz, validity 1969-01-01 to 2032-12-31 (the contact address in the certificate subject is omitted here on purpose). That is a Parrot service by name and organisation. Verified (strings, certificate decode); same host and certificate class as the Bebop 2 page (Agent-cited).

Is anything sent without a user action? The sending function logs “send report disabled!” and “send report %03zu denied!” (strings) and the only enabling switch the binary reads is auto_send_https_report, which is 0. There is no control socket: macgyverd imports no pomp or socket-server code, only inotify on the report folders, timerfd, and the outbound SSL socket (Verified: imports listed). report-generic only writes a file and checks that init.svc.macgyverd is running; it does not send (Verified, strings). So in this image no report leaves the unit unless auto_send_https_report is changed to 1 or the binary is run with different options; there is no on-device “send now” path that I found (Verified for config, Inference for runtime, since a runtime-edited conf could differ). Whether a phone-side ARSDK command can trigger a send is Open (no such string found in mppd). The Bebop 2 page says the hostname did not resolve on 2026-10-02 (Agent-cited); that was not rechecked, since no network access is allowed here.

Exposure side effect: the report folder is under the FTP root served on tcp 21 (see inetd). Anyone who can reach that FTP service can list and read the stored reports, and with -w can also upload into the same tree (Verified for the paths and flags; reading is Inference for the anonymous behaviour).

Service What it does Status
usbctrl-mon Polls /sys/devices/platform/ci_hdrc.0/usb1/status through a periodic timer. If the host controller reports dead, it stops iosd (ctl.stop), removes g_aea, writes the USB PHY over I2C (/dev/i2c-0), unbinds and rebinds ci_hdrc.0, reloads g_aea through load_g_aea, restarts iosd, then starts usbctrl-report, which files a generic report “usb ci_hdrc.0 fatal error”. It logs status changes between running, dead and unknown, and sets a system property when needed. Verified (strings and rc); sequencing Inference
memtester and memtester_led Not started at boot. udev rule lib/udev/rules.d/70-memtester.rules:3-9: on a USB device add with vendor 19cf and product ffff it sets MEMTESTER=1, runs mpp_bb_cli stayon 1 (keeps the unit powered through the bottom-board MCU) and pstart memtester. On any USB remove with MEMTESTER=1 it runs stayon 0 and pstop memtester (lines 11-15). memtester.sh sets mppd.led.enable 0, starts the LED helper, lights magenta, then loops memtester 20M forever and logs and stops the blinking if a line contains FAILURE. memtester_led.sh blinks magenta. Effect: a factory burn-in mode. Anyone who can plug a USB device with that id into the controller’s USB host port puts the unit in a continuous memory-test loop and holds it on until the device is removed. Vendor 19cf is the id the nap image uses for its own gadget, so it is a Parrot id (Inference). Verified
Bottom-board rules 52-bottomboard.rules: a USB device with vendor 03eb product 2fd9 (Atmel DFU boot id) runs bottom_board_update, which flashes /lib/firmware/bottomboard.hex (or the .dv.hex legacy file for hardware revision 4 or lower) with dfu-programmer atxmega32c4, sets POWER_KEEP=1 and launches the new firmware. Product 2406 runs gamepad_plugged, which writes POWER_KEEP=0. The flashed content comes from the image, not from the device. Verified
jubamountd and the juba udev rules Auto-mounts USB mass storage and SD cards (etc/jubamountd.conf:31,34: sd* vfat and ntfs, uid and gid 1000, mounted under /tmp/.mnt), notifies clients, and tags gamepad, bottom board, iPod and generic USB devices for aoad and mppd. Verified
sensorsd-mpp, kmsgd, crashdumpd, udevd Sensor daemon (IMU icm20608, magnetometer ak8963), kernel log to ulog, crash capture (writes /var/lib/crashdump), udev. Verified (rc), behaviour Inference
etc/udev_perms.conf:20 /dev/ulog_* are mode 0666 root:log; the comment at line 19 says only group log can read. Verified; same mismatch noted in the nap report

Scope: what this image configures. “Verified” means a config file or rc was read, “Inference” means it depends on runtime behaviour the image alone does not show. The unit has no Wi-Fi access point, so there is no network that “a person on Wi-Fi the controller creates” can join; the controller is a Wi-Fi client of the drone, and a USB-side network only appears when a phone or other USB device creates an eth* or usb* interface. The question for a person on any network the controller creates therefore reduces to the USB network below.

# Who Claim Status
1 Any host on the drone’s network (wlan0) The controller is a Wi-Fi client of that network, address from DHCP or static 192.168.42.3. Anything that can reach that address can try tcp 21 (anonymous-style FTP as root, /data/lib/ftp tree, upload enabled, crash reports inside it) and tcp 9050 (adbd), because inetd binds the wildcard and no firewall is installed by the image Verified (config); wildcard bind and reachability Inference
2 Same The drone itself, or any other client of the drone’s access point, is in that position. The controller’s FTP, adbd and (if the live port is reachable) ARSDK service sit on the drone-facing address Inference
3 Same avabye announces _arsdk-090f._udp only on eth*,usb*, so mDNS discovery is not offered on wlan0; the ARSDK TCP/UDP listener (live port 44444, Agent-cited) is not shown to be restricted to a USB interface Verified (avabye config); listener binding Open
4 A USB host or phone that creates eth* or usb* The controller first asks that interface for a DHCP lease. If none answers it takes 192.168.N.1/24 and runs a DHCP server on that interface, with no gateway, no DNS and an infinite lease for .10 to .100. The phone then gets an address, discovers _arsdk-090f._udp and can talk ARSDK Verified (code); N Open
5 Same The same wildcard services, tcp 21 and tcp 9050 (adbd), answer on the USB network too, so a phone on the cable can reach the FTP tree and adbd; without a firewall in the image, nothing separates this from item 1 Verified (config); Inference (runtime)
6 Same adbd on 9050: ro.debuggable=1 is shipped, ro.adb.secure is set nowhere and adbd contains no privilege drop, so an unauthenticated session would be a root shell. Whether the live unit enforces key auth is not shown Verified (files, imports); auth behaviour Inference
7 A person with physical access UART root console on ttyPA3 (askfirst), gated on ro.debuggable=1; a USB device with vendor 19cf and product ffff starts a memory-test loop and holds the unit powered; a USB device with the Atmel DFU id 03eb:2fd9 triggers a flash of the bottom-board firmware from the image; USB storage is auto-mounted by jubamountd as root Verified (rc and rules)
8 A phone with ARSDK access mppd can write /update/.factory_reset (string “Unable to create factory-reset file”), which at next boot runs factory_reset (rm -rf /data/*). Which ARSDK command or button triggers it is not identified; this is the only remote-reachable wipe path found Verified (strings, rc, script); trigger Open
9 A client able to read FTP Crash reports (oops, crash dumps, kernel log, ulog, build and factory info) sit in the FTP-served folder with 0777 directories; with -w the tree is also writable, and files dropped there are scanned by macgyverd through inotify (Inference for the scan, since the watch is on the report folder) Verified (paths, flags); behaviour Inference
10 Remote parties No outbound crash upload without auto_send_https_report=1 (config 0); no inbound telnetd, ntpd or sshd; no ulogcat TCP port (commented out); mppd debug server only on 127.0.0.1 Verified (config and strings, with positive controls); runtime Inference
11 All of the above Every network-facing daemon found (inetd and its ftpd, adbd, wifid, mppd, avabye, usbnetd and dnsmasq) runs as root or has no user line. No rc file drops privileges Verified (rc)

The firewall policy and routing set outside the image (the kernel, any later iptables run by a different process, or a vendor tool); runtime property overrides (persist.* files under /data/property, kernel command line, ro.hardware, ro.revision, ro.serialno, ro.boot.debuggable); /factory, /data, /update contents; /customer_prod, which no file in the image names at all; the kernel image and its netfilter and USB gadget support; BusyBox ftpd login code (disassembly); mppd accept and reject logic for a phone peer and the real ARSDK listen port; the second half of usr/share/udhcpc/default.script; the full usbnetd main loop beyond the three functions read; libshsettings schema; button mapping files etc/mppd/mapping_*.cfg; Qt or HUD assets; the Broadcom firmware and wl.ko (not diffed beyond size and a byte compare showing they differ); the content of the two .nvm or .bin blobs. Nothing was run and no device was contacted.

  1. What listen port does mppd’s ARSDK net backend use, and on which addresses does it bind? The brief says live 44444; no port constant was found in the image (the only 44444 literal is a log line number).
  2. What is the value N in 192.168.N.1/24 (interface index, slot number, or something else), and is it stable across plug cycles?
  3. Does the live unit enforce adb key authentication on 9050 (ro.adb.secure unset in the image, key store on /data), and is adbd reachable on wlan0 as well as on the USB network?
  4. Does busybox 1.25.0 ftpd accept any user and password, and does the chroot reach the symlinked /var/lib/ftp path as expected?
  5. Which ARSDK command or UI action makes mppd create /update/.factory_reset?
  6. How does mppd decide to accept or reject a peer (peer connection requested: name=, id=): is there pairing, or an allow-list?
  7. Is /var/lib really a symlink to /data/lib on this image? /var is not in the extract; the rc files and macgyver.conf are consistent with it, but the link itself was not seen.
  8. Are the wifi_key values of known drones stored in clear in a runtime copy of drone_manager.cfg under /data? The image copy is empty.
  9. Do the bcm43526 and bcm43526-extra module sets differ in a way that matters (the env var selects the second), and which board revision uses which?
  10. What sets ro.revision and ro.serialno, and does a hardware revision 0 unit lose the iOS link by design (98-iosd.rc:12-15)?

Comments

Spotted a mistake, or tried this on your own Bebop? Leave a note. Comments are kept on this site's own server.