Skip to content

Network

  • Access point Bebop2-XXXXXX, open, 2.4 GHz channel 10 at first contact.
  • BSSID OUI A0:14:3D, registered to Parrot SA. All five Parrot SA OUIs (verified against /usr/share/hwdata/oui.txt): 00:12:1C, 00:26:7E, 90:03:B7, 90:3A:E6, A0:14:3D. bebop-2 scan matches by these or by an SSID starting with “bebop”.
  • Drone at 192.168.42.1, DHCP gave the workstation 192.168.42.36.
  • Services listed by rf-lab DEVICE_ACCESS.md (not yet checked by us): telnet 23 (after 4 presses), FTP 21 (root /data/ftp), ADB 9050.
  • Long press toggles 2.4/5 GHz (longpress_0.sh).
  • Default route on br0 (ethernet, 10.22.22.1, metric 100). wlan0 was on starla as a secondary link (metric 300).
  • NetworkManager profile bebop2: ipv4.never-default yes, ipv6.method disabled, connection.autoconnect no. Joining it moves wlan0 off starla without touching the default route.
  • Back to normal: nmcli con up starla.
  • Regulatory domain 00 (world). Passive scan covers 2.4 GHz 1 to 13 and 5 GHz including DFS, so a Bebop on any channel would be heard.
  • First power-up: not beaconing for at least a minute, LED state unknown. After a power cycle it beaconed within about 30 s. Q: why.

Debug mode opens two unauthenticated root paths, not one

Section titled “Debug mode opens two unauthenticated root paths, not one”

Four short presses on the power button open telnet on 23 and ADB on 9050 together. Both give root with no password on an open Wi-Fi network: telnet’s login is exec /bin/sh -l, and adb shell reports uid=0(root). Verified on the live aircraft 2026-10-02.

The press works while the drone is flying a controller session; it did not disturb the ARSDK link that was open at the time.

Worth stating plainly for anything built on top: ADB bypasses any restriction a client puts on the telnet path. mcbebop’s shell_read allows eleven read-only command names and refuses shell metacharacters, and that is a useful guard against a careless call, but it is not a security boundary while 9050 is open. The boundary is the button press, and it lapses at the next reboot.

Multiple Wi-Fi clients: yes. Multiple controllers: still unknown.

Section titled “Multiple Wi-Fi clients: yes. Multiple controllers: still unknown.”

These are two different questions and it is worth keeping them apart.

Wi-Fi association is not exclusive, and we demonstrated it on 2026-10-01 without setting out to. During the FreeFlight 6 test the lab phone was associated to this drone’s access point on 192.168.42.78 and pinging it at 2 to 5 ms, while this workstation was associated at the same time on 192.168.42.36. Two clients, both reachable, neither disturbed. The access point hands out DHCP leases normally and behaves like any other AP.

Two ARSDK control sessions: tested 2026-10-02, and the common claim is wrong. The drone does not refuse a second controller. It accepts the handshake and then silently redirects its telemetry to the newcomer, leaving the first session starved while that session still believes it is connected.

Measured on the aircraft, landed, props off:

session A session B
before B connects 193 keys, last event 0.05 s ago, 264 frames in
3 s after B connects 193 keys, last event 3.6 s ago, 264 frames in (frozen) 21 keys, last event 0.01 s ago, 164 frames in

A’s connected flag stayed True throughout. Its socket is fine; the drone simply stopped sending to it, because the handshake’s d2c_port tells the drone where to deliver and the second handshake overwrites that.

This is the worst of the three possible behaviours for a client author, because it fails quietly. A refusal would be loud and obvious. Instead the losing session keeps its connection object, its stored values and its optimism, and only an aging timestamp gives it away. Any client that trusts a connected boolean rather than telemetry freshness will report a healthy link to a drone that is no longer talking to it.

So “the drone serves one controller at a time” is wrong as usually stated. It serves one telemetry destination at a time, last writer wins. That phrase originates in pyparrot’s error text, was carried into mcbebop’s tool descriptions, and both were wrong; corrected after this test.

Practical consequences:

  • Freshness is the only honest health check. mcbebop’s connection_status reports last_update_age for exactly this reason, and it is the field to watch, not connected.
  • A second machine may associate to the Wi-Fi, use FTP, ADB or telnet freely without disturbing a control link. Only the ARSDK handshake takes it over.
  • The takeover is recoverable: the starved session reconnects normally, which simply takes the link back.

Our simulator modelled a refusal (status: 1), with a comment claiming that is what the aircraft does. It was encoding the same unverified assumption as the client, which is why no test caught it. The simulator now needs a takeover mode to match reality. TODO.

Comments

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