Skip to content

kormax/apple-enhanced-contactless-polling

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

73 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Apple Enhanced Contactless Polling (ECP)

![ECP Access Home with PN532] ![ECP Transit Clipper with ST25R3916 on Flipper Zero] ![ECP EMV-Transit Ventra with with PN5180] ![ECP EMV-Transit Maestro with with PN5180]

Overview

Enhanced Contactless Polling (ECP) is a proprietary extension to the ISO/IEC 14443 (A/B) standard developed by Apple.

It defines a custom data frame that a contactless reader has to transmit during the polling sequence, providing an end device with contextual info about the reader field, allowing it to decide if it wants to resolve routing to a particular applet or system feature even before any back and forth communication starts.

This extension:

  • Helps to make sure that the end device will only start communication with the reader if it has something useful to do with it, avoiding error beeps and card clash;
  • Increases privacy and security as it complicates brute force scanning for available passes on the device in a single tap.
  • Allows automatic usage of non ISO7816-compliant passes:
    • DESFire in native mode and on card-level instead of app-level;
    • Passes without application ID (Mifare Plus in some cases).
  • Helps with conflict resolution when there are multiple passes with the same ISO7816 AID:
    • Allows to differentiate between Gymkit and ISO18013 even though both use NDEF AID for BLE handover.
    • Resolves routing issues between access credentials from the same system manufacturer (HID, AssaAbloy, WaveLynx, Brivo, Kisi, etc.) even if they use the same AIDs and the underlying technology does not support subcredentials at a higher lever.
  • May serve as a form of NFC DRM, requiring reader manufacturers to pay licensing fees in order to be able to use this feature and provide a better experience for Apple users.

ECP is also sometimes referred to as Enhanced Contactless Protocol. For explanation, look into extras section

Use cases

Express mode for most passes (apart from FeliCa and CATHAY) is implemented using ECP. That includes:

  • Credit cards (For transit fallback);
  • Transit cards;
  • Access passes:
    • University;
    • Office badges;
    • Venue (Theme parks);
    • Apartment (Multi-family homes);
    • Hotel.
  • Keys ("shareable" access passes):
    • Car;
    • Home;
    • Access.

Other features use ECP as well:

  • Value Added Services (VAS):
    Allows the reader to select the VAS applet and try to get pass in advance (although failing to do so), causing a pass to appear on a screen for authentication or under a payment card if one is selected.
    ![Image showing VAS] ![Image showing VAS] ![Image showing VAS]
  • GymKit:
    Makes Apple Watch act as an NDEF tag for BLE handover in order to connect to supported gym equipment.
    ![Image showing GymKit setup]
  • Identity:
    Makes Apple device act as an NDEF tag for BLE handover in order to connect to a ISO18013 verfier.

    ![ISO18013 Request prompt]

  • CarKey Setup:
    Tells the device what car brand it is, causing a car key setup popup to appear on a screen.

    ![Image showing IOS16 CarKey pairing prompt] ![Image showing IOS17 CarKey pairing prompt]

  • Field ignore:
    Makes Apple devices not react (by react meaning displaying a default payment card) to a field generated by other apple devices.
  • AirDrop:
    Replaces field ignore in IOS17 for background reading, used to negotiate an AirDrop session. NameDrop is a special case of AirDrop. Triggers a warp animation.

    ![AirDrop warp animation preview]

  • HomeKit:
    Allows appliances with an NFC reader that lack card emulation mode to convey pairing info and bring up a pairing prompt when a user device is brought near to it.

Device support

Reader side:

  • Can be implemented in software on most devices, provided that low-level access to NFC hardware is available. In some cases, it is required to re-implement parts of the protocol stack in the software when doing so.
    HALs/Libraries for most popular chips contain separate confidential versions that include ECP support and are given to approved partners only, but a homebrew solution is easy to implement.
    Proof of concept was successfully tested using PN532, PN5180, ST25R3916(B), MFRC552 chips and PCSCV2 readers;
    For information about those chips and how it can be implemented, visit the Examples page;
  • iOS has special reader APIs that make the device emit specific ECP frames:
    • NFCVASReaderSession, PaymentCardReaderSession for VAS;
    • MobileDocumentReaderSession for Identity;
    • When using other derivatives of NFCReaderSession, the device emits the Ignore frame so that other Apple devices don't react to it;
  • Some Android-based handheld reader manufacturers have implemented this feature in their software.

Device side:

  • Implemented using a customized CRS applet;
  • Can be implemented on chips that allow reading raw frames in emulation mode even before selection;
  • Can be reproduced on select mobile handsets running the Android 15 operating system, with the introduction of Observe Mode, which allows Android devices to look at the polling frames sent by the contactless reader even before starting communication.

ECP and Express Mode behavior

Overview

Upon entering a loop, the device does not answer to the first polling frame it sees, instead opting to wait and see what other technologies does the field poll for, allowing it to make a fully informed decision on what applet or feature to trigger later.

When the device makes a decision, it is mostly, although not in all cases (excluding keys) signified by a card image appearing along with a spinner.

![Image showing express mode animation after decision]

Even though ECP is sent during the polling loop, device does not answer to it directly. Instead, it responds to a polling frame related to the technology of the pass that the device had decided to use.

Polling

When device enters the loop initially:

  • In case of a full polling loop (A,B,F) it waits through one full iteration before making a decision on what applet to select:
(ENTRY) -> A -> ECP_A -> B -> ECP_B -> F -> (DECISION) -> A -> (RESPONSE)
A -> ECP_A -> (ENTRY) -> B -> ECP_B -> F -> A -> ECP_A -> (DECISION) -> B -> (RESPONSE)
  • In case of partial or wierdly-ordered polling loop, behavior is different. For example:
(ENTRY) -> A -> ECP_A -> A -> ECP_A -> (DECISION) -> A -> (RESPONSE)
(ENTRY) -> F -> B -> ECP_B -> A -> F -> B -> (DECISION) -> ECP_B -> A -> (RESPONSE)
(ENTRY) -> A -> ECP_B -> F -> A -> ECP_B -> (DECISION) -> F -> A -> (RESPONSE)
(ENTRY) -> F -> F -> F -> (DECISION) -> F -> (RESPONSE)
(ENTRY) -> A -> A -> A -> (DECISION)
(ENTRY) -> A -> ECP_A -> F -> A -> ECP_A -> F -> (DECISION) -> A -> (RESPONSE)

Characters A, B, and F were used in examples as a shorthand for full polling frame names: WUPA, WUPB, SENSF_REQ respectively. ECP frame has different values depending on a use case _A/B suffix refers to the modulation used.
Tests were conducted using very big intervals between polling frames. IRL if polling is faster device might respond after more frames than shown, presumably because of internal processing delay.

In conclusion, if the reader is polling for:

  • 1 technology, the decision is made after the third poll, the response is given on the fourth;
  • 2 technologies, the decision is made after the second polling loop, while the response is given on the third.
  • 3 technologies, the decision is made after the first loop, response is given on the second.

Field detection grace period

Apple devices seem to have a grace period after leaving an NFC field, where if a device re-enters the field in time, it will still consider it to be the same field.
It was done in order to better support express mode with devices that turn off the NFC field in between polling iterations in order to conserve energy.
This way, an end device wouldn't think that it entered a new field on each polling iteration, which would've prevented express mode from being triggered, as it needs at least two loops for selection.

According to tests, the grace period duration depends on what pass types are enabled for express mode on the device. The grace period duration setting seems to be separate for each NFC technology, with the worst/maximum value taken from enabled pass categories:

  1. ECP (NFC-A/B) + CATHAY (NFC-A):
    1. (EMV) Payment card: 300 ms;
    2. Transit card: 750 ms.
  2. FeliCa (NFC-F):
    1. Transit card: 750 ms;

For improved stability and increased polling performance, it is advised to do polling in 250-100 ms intervals or less, keeping in mind the field activity duty cycle.

Communication start delay

Even though NFC polling can be really fast, it still takes some time for a device to analyse polling frames and make a decision on which routing to activate in order to begin communication.
During tests, following delays have been measured:

  1. ECP (NFC-A/B):
    1. PC + PN532: 100ms;
    2. Proxmark3: 1031640 / 13560000 * 1000 ~= 76 ms or ~45 polling iterations;
  2. FeliCa (NFC-F):
    1. PC + PN532: 50 ms.
    2. Proxmark3: 547584 / 13560000 * 1000 ~= 40 ms or 3 polling iterations;
  3. CATHAY (NFC-A):
    1. Proxmark3, no delay:1042912 / 13560000 * 1000 ~= 77 ms or ~150 polling iterations;
    2. Proxmark3, 5 ms delay: 979680 / 13560000 * 1000 ~= 72 ms or ~12 polling iterations;

It's important to understand that the hardware used for measuring might have some operational delay itself.
PC + PN532 method yields a bigger measurement error as it runs on a non-real-time system. Proxmark3 method gives more accurate results as the measurement is done on a separate device, which is intended specifically for NFC analysis.

As a result of the tests, we have the following findings:

  • FeliCa has from 50% to 100% less processing delay in comparison to NFC-A/B, which is quite interesting.
    • One assumption is that additional delay may be caused by ECP part of the resolution stack inside of CRS, which does not impact FeliCa resolution, as it has a native system-code-based method of resolving applets for express mode.
    • The other assumption is that this delay is arbitrary, and is used to prevent situations when communication is started "just as" the device barely enters the field, when signal strength is poor and long commands can be interrupted.
  • With NFC-A, if the polling frame is sent really fast without any arbitrary delays, the device might actually respond a little bit slower (up to 82 ms).
    The sweet spot for response speed is about 5 ms guard time between polling attempts, which can reduce delay down to about 70 ms.

Express Mode priority

Although not possible during normal operation, if a reader is polling for multiple cards using express mode that use different technology qualifiers for selection, the following technology priority will be applied:

  1. ECP (ecp):
    1. Primary (ecp.2.tci);
    2. EMV Fallback (ecp.2.open_loop);
  2. FeliCa (felica.*);
  3. CATHAY (generic.type_a).

There could be more valid express mode qualifiers available, but they have to be verified to be used on real passes before being listed here. If you have any info, feel free to create a PR.

Quirks and bugs:

  • If polling for both ECP and FeliCa, the device will sometimes display the FeliCa pass in animation while actually selecting and emulating the ECP-triggered pass. The frequency of this bug depends on the polling loop interval.
  • In IOS17 the new NameDrop frame does not follow the aforementioned rules fully, as the device reacts to it with an animation on the first iteration. Although the response itself is still returned as presented in examples.

Structure

Frame format

Each ECP frame consists of a header, version, payload and CRC:

     6A         XX          XX...     XX XX
  [Header]  [Version]  [Payload (n)]  [CRC]
  • Header byte has a constant value of (HEX) 6A;
  • Version number can be either 0x01 or 0x02;
  • Payload: Version-dependent;
  • CRC (Calculated via ISO14443A/B algorithm, according to the modulation used).

Payload

For V1 payload consists only of a single TCI:

  XX XX XX
   [TCI]  
  • TCI is a 3 byte long identifier. More info below.

For V2 payload contains terminal configuration, terminal type, terminal subtype, and data:

     XX       XX        XX         XX...
  [Config]  [Type]  [Subtype]  [Data (n)]
  • Configuration byte. Responsible for feature flags and data length. More info below;
  • Type contains terminal type:
    • 0x01: Transit;
    • 0x02: Access;
    • 0x03: Identity (Handoff);
    • 0x05: AirDrop.
  • Subtype depends on type. In most cases it has a value of 0x00;
  • Data. Its content and availability depend on terminal type and subtype. Detailed description below.

Configuration byte

Configuration byte is a mandatory part of a V2 ECP payload. It consists of 4 flag bits and a length nibble:

Bit 07 06 05 04 03 02 01 00
Function Automatic pass presentment Authentication not required Unknown flag 1 Unknown flag 2 Length
  • Bit 07. Automatic pass presentment. The default value is 1. A matching pass does not appear on a screen if this value is set to 0, which also disables express mode as a result. ECP frames with this value being 0 are not found IRL.
  • Bit 06. Authentication not required. The default value is 1. Disables express mode if the value is 0, the device will bring up a matching pass for manual authentication instead. Access readers may set it when auth is required. Also set to 0 for Identity and AirDrop/NameDrop.
  • Bit 05-04. Meaning unknown. The default value is 0;
  • Bits 03-00. Defines the length of the data part (0-15 bytes). If the length does not match, the device will ignore this ECP frame.

Data

Data is a part of the payload in V2, it contains TCIs and extra data:

  XX XX XX...        XX..
  [TCIs (n)]   [Extra data (n)]
  • TCIs define an array of 3-byte long identifiers. Standard allows for 0-n long TCI arrays to be conveyed depending on terminal type and subtype;
  • Extra data contents depend on terminal type, subtype, and TCIs:
    • For transit it contains a mask of supported EMV payment networks for fallback;
    • For access/key readers it may contain a 8-byte long unique reader group identifier, which allows differentiating between them for passes of the same type;
    • For HomeKit it contains pairing information;
    • For NameDrop it carries a 6-byte long BLE MAC address;
    • For AirDrop it carries a 6-byte long zeroed-out value.

TCI

Terminal Capabilities Identifier (TCI), is an arbitrary three-byte-long value that establishes reader relation to a particular pass type (Home key, Car key, Transit) or system feature (Ignore, GymKit, AirDrop, NameDrop).

The following restrictions apply to the use of TCI:

  • Some TCIs are bound to a reader with a particular type and subtype (which requires V2), while others trigger for all types (support V1). It is not known if this behavior is a bug or was intentional;
  • TCIs intended for V1 cannot be used with V2.

TCI format is arbitrary, although several patterns related to the grouping of similar functionality can be established:

  • VAS: grouped with the last byte having a value of 0x00, 0x01, 0x02, 0x03 depending on mode;
  • Access (Car/Home/University/Office/Venue): The first byte is always 0x02, other two link to a particular pass provider;
  • Transit: The first byte is always 0x03, other two link to a particular transit agency (and their pass);
  • CarKey: The first byte is always 0x01. The next three nibbles serve as a manufacturer identifier. The last nibble serves as the reader location index. This can be seen in wallet configuration JSON hosted at smp-device-content.apple.com.

Configuration examples

Note that CRC A/B, ECP Header, Configuration bytes are omitted from this table.

NA - not applicable; XX - any; ?? - unknown

Name Version Type Subtype TCI Data Source Description
VAS or payment 01 NA NA 00 00 00 NA Sniffing VAS ECP configurations are sometimes regarded to as VASUP-A(B)
VAS and payment 01 NA NA 00 00 01 NA Bruteforce
VAS only 01 NA NA 00 00 02 NA Bruteforce
Payment only 01 NA NA 00 00 03 NA Bruteforce As all other VAS frames, also serves as anti-CATHAY
GymKit 01 NA NA c3 00 00 NA Bruteforce The only way of triggering GymKit on an apple watch. There are other frames that trigger GymKit too, but they also trigger express transit for iPhones
Ignore 01 NA NA cf 00 00 NA Sniffing
Transit 02 01 00 XX XX XX XX XX XX XX XX Bruteforce based on TFL example TCI refers to a transit agency, Data is a mask of allowed EMV payment networks for fallback
Transit: Ventra 02 01 00 03 00 00 ?? ?? ?? ?? ?? Bruteforce
Transit: HOP Fastpass 02 01 00 03 04 00 ?? ?? ?? ?? ?? Bruteforce
Transit: WMATA 02 01 00 03 00 01 ?? ?? ?? ?? ?? Bruteforce Will select a Smart Trip card
Transit: TFL 02 01 00 03 00 02 79 00 00 00 00 Sniffing: Payment Village/Proxmark community Allows Amex, Visa, Mastercard, Maestro, VPay
Transit: LA Tap 02 01 00 03 00 05 ?? ?? ?? ?? ?? Bruteforce
Transit: Clipper 02 01 00 03 00 07 ?? ?? ?? ?? ?? Bruteforce
Transit: Navigo 02 01 00 03 09 5a ?? ?? ?? ?? ?? Bruteforce
Access 02 02 XX XX XX XX NA/XX XX XX XX XX XX XX XX Assumption based on other data TCI refers to a pass provider, Data is reader group identifier
Access: Venue 02 02 00 XX XX XX NA HID Reader configuration manual
Access: Hotel: Hilton 02 02 00 02 ff ff NA File TCI might be a wildcard in case a booking is not made, needs further testing
Access: Home Key 02 02 06 02 11 00 XX XX XX XX XX XX XX XX Sniffing/File Having more than one key breaks usual ECP logic
Access: Car Pairing 02 02 09 XX XX XX NA Bruteforce TCI refers to a combination of car manufacturer + reader position. Too many combinations involved. For rest refer to smp-device-content
Access: Car Pairing: BMW/MINI 02 02 09 01 00 01 NA Configuration
Access: Car Pairing: Mercedes 02 02 09 01 02 01 NA Configuration
Access: Car Pairing: Genesis 02 02 09 01 00 51 NA Configuration
Access: Car Pairing: KIA 02 02 09 01 00 41 NA Configuration
Access: Car Pairing: Hyundai 02 02 09 01 03 01 NA Configuration
Access: Car Pairing: BYD 02 02 09 01 07 01 NA Configuration
Access: Car Pairing: Denza 02 02 09 01 02 41 NA Configuration
Access: Car Pairing: YW 02 02 09 01 02 B1 NA Configuration
Access: Car Pairing: FCB 02 02 09 01 02 D1 NA Configuration
Access: Car Pairing: Lotus 02 02 09 01 00 91 NA Configuration
Identity 02 03 00 NA/00 NA/00 Sniffing Only ECP frame found IRL that lacks a full TCI. Could this mean that TCI length is variable or it could be missing and the extra byte is data instead?
AirDrop 02 05 00 01 00 00 00 00 00 00 00 00 Sniffing Sent only after device sees a NameDrop frame
NameDrop 02 05 00 01 00 01 XX XX XX XX XX XX Sniffing Data part contains a BLE MAC-address

Source: Bruteforce - found by going over all config combinations; Sniffing - sniffed from a real reader; File - retrieved from pass file; Configuration - retreived from smp-device-content configuration json
Frames found via brute force may be working but not actually used. In case you have a real samle - let us know if it is different or the same

Full frame examples

Examples contain frames without CRC, which needs to be calculated according to the modulation used;

  • VAS or payment:
    6a01000000

         6a         01      000000 
      [Header]  [Version]   [TCI]  
    
  • GymKit:
    6a01c30000

         6a         01      c30000 
      [Header]  [Version]   [TCI]  
    
  • Ignore
    6a01cf0000

         6a         01      cf0000   
      [Header]  [Version]   [TCI]  
    
  • Access: Hotel: Hilton:
    6a02c3020002ffff

         6a         02        c3       02      00      02ffff  
      [Header]  [Version]  [Config]  [Type] [Subtype]  [TCI]     
    
  • NameDrop:
    6a02890500010001deadbeef6969

         6a         02        89       05      00      010001  deadbeef6969 
      [Header]  [Version]  [Config]  [Type] [Subtype]  [TCI]     [Data]     
    
      1000        1001
      [NA]  [Payload length]
    
  • Access: Car Pairing: Mercedes:
    6a02c30209010201

         6a         02        c3       02      09      010201 
      [Header]  [Version]  [Config]  [Type] [Subtype]  [TCI]  
    

    Replace TCI for any car TCI given in the table to get pairing prompt for other brand

Note that for examples to work 8-bit byte setting should be set in case of NFC-A, 2-byte CRC has to be appended beforehand.

If you have a Proxmark3, you can test those frames using commands hf 14a raw -akc for NFC-A and hf 14b raw -kc -d for NFC-B. After sending the frame try polling the device using the hf 14a reader -sk or hf 14b reader command.

Extras

EMV Transit Fallback

Express mode for EMV is triggered as a fallback in case a pass for a particular transit TCI has not been found in a system. EMV brand support mask is contained in the last 5 bytes of the frame's data. At the current moment, only the first byte is known to be used IRL.

The following table presents an encoding scheme for the first byte (bit 00 is the rightmost), with 1 (true) in a particular position signifying support:

Bit 07 06 05 04 03 02 01 00
Brand ?? Possibly VPAY/Electron VISA MAESTRO MASTERCARD ?? ?? Possibly AMEX

To find a bit responsible for your card network, you can modify a particular bit inside the mask. Afterward, having activated a specific card brand for express mode on your device, observe if the express mode will activate when brought near to a test reader.

Strong customer authentication

Some applets have the ability to modify behavior based on the conditions that the transaction has been started under:

  • Unified Access applet potentially has the ability to guarantee to the reader, if requested, that a transaction has been initiated via manual authentication, which protects against exciting a device with an express ECP frame and using it on a reader that mandates SCA;
  • When EMV is used with express mode, it modifies transaction parameters that need to be used for a device to produce a successful transaction + check mark.

Requirements regarding the EMV transaction seem to be dependent on a payment network brand (as each has a separate applet implementation):

  • Mastercard/Maestro:
    • MCC is in a transit category
    • TC/ARQC + CDA
  • Visa:
    • Offline data authentication for online authorizations enabled
    • TC/ARQC + CDA

Other card brands may have different success conditions and behavior changes. If you have any info, feel free to create a PR.

Enhanced Contactless Protocol

When first researching the topic of ECP, in some rare situations I noticed that some brochures refer to ECP as "Enhanced Contactless Protocol". The first assumption was either made to put potential researchers off the track or that it was a simple mistake when creating the material.

When looking into some promotional documents, "Enhanced Contactless Protocol" arose once again, this time in the context of "DESFire ECP compatibility mode", which rang a bell.
After a bit of analysis, it turned out that DESFire protocol indeed has a special command created specifically for Apple devices, the sole purpose of which is to notify a device that a transaction has been done successfully.
This leads to the thought that ECP (Polling) and ECP (Protocol) are indeed two different terms when used by Apple and/or their partners. In conclusion, a new explanation for ECP (Protocol) has been formulated.

Enhanced Contactless Protocol is a protocol that implements commands that allow to explicitly notify an end device about the state of a transaction, without resorting to making any assumptions about a particular command sequence or operations that need to be done for a transaction to be deemed successful (or not).

Success conditions

An NFC transaction is deemed successful when a device produces a checkmark on a screen, followed by a notification in special cases.
Failure condition is displayed with an exclamation mark and an optional error popup/device vibration.

NFC protocols can be divided into two categories, depending on how a UX success condition is determined:

  • Explicit (Enhanced Contactless Protocol).
  • Implicit;

Following protocols are considered "enhanced" as they implement explicit status commands.

Protocol name Success condition command Failure condition Notes
Mifare DESFire NOTIFY_TRANSACTION_SUCCESS(0xEE) DESELECT/TRESET without the command
Unified Access (all CarKey, HomeKey, AccessKey, Aliro variants) OP_CONTROL_FLOW(0x3C) with success flags or DESELECT after attestation exchange OP_CONTROL_FLOW(0x3C) with failure flags or DESELECT/TRESET before it

DESFire command name was made up by me as it's newely discovered, no info about it online.

Following protocols have implicit transaction status detection:

Protocol name Success condition Failure condition Notes
T-Union DESELECT/TRESET after AID selection DESELECT/TRESET before AID selection
FeliCa 0.5 second delay after TRESET if REQUEST_SERVICE(0x02) has been used. 5 second delay after TRESET otherwise READ/WRITE commands with invalid keys Current success condition causes lots of confusion for users, as a top-up machine may do a TRESET and an additional POLLING for data verification, but it causes a check mark to appear thus misleading users, making them think that they can take the device out prematurely.
EMV Cryptogram generation (EMV mode) or magstripe data read (MAG compatability mode) DSELECT/TRESET before success condition
VAS Successful GET_DATA followed with a DESELECT/TRESET DSELECT/TRESET after failed GET_DATA

Other protocols supported by Apple Wallet, such as:

  • BMAC, SPTCC (Legacy China Transit);
  • Seos.

Were not researched due to lack of samples to do tests on. If you have access to any of them (on a device), feel free to add info in a PR.

Service mode

Service mode, also internally called "Plastic card mode" is a feature that is available on some contactless passes. It is intended to be used when you need to give a device to someone to conduct service/help operations with the card.

After some experimentation, the following changes to the behavior have been noticed when this mode is activated:

  • Device increases the grace period before it ends the NFC authorization;
  • TRESET/DESELECT, implicit, explicit transaction end conditions do not end the NFC session;
  • Device generates a transaction notification/receipt after the service mode period ends (No SE notifications to the SEP?).

Judging from this information, it is safe to assume that what this mode does is disables regular transaction status tracking and end condition fulfillment, allowing even an incompatible/slow/unique reader to conduct any NFC transaction sequences imaginable, with DESELECTs/TRESETs and so on. Extra time in comparison to regular NFC auth is also given in order to accommodate for all possible service processes and communication delays.

Polling Loop Filters

Android 15 introduced two brand-new concepts to the operating system's NFC stack:

Observe mode works by suppressing the ability of a device to respond to external NFC polling frames by default, instead passing the information about the polling loop down to the operating system and application level, allowing the former to decide on what to do next.

The following steps could include any of the following, based on characteristics of the polling loop as deemed fit by the device:

  • Releasing the NFC lock, allowing the transaction to go through automatically (this is the way it worked before anyway).
  • Requiring a user intervention for a specific action:
    • Authentication. For instance, this would entirely fix an issue where an Android phone would fail EMV transaction due to not being unlocked properly, or due to an unlock happening too early.
    • Credential sub-selection. For instance, a non-EMV reader could be detected via the usage of REQA frames (instead of WUPA), and a user could be asked to choose an appropriate access pass if one has multiple.

As a side effect, Observe Mode enables an additional feature, called "Polling Loop Filters". It allows any NFC-powered application to define a specific polling data pattern so that the operating system delivers control of the NFC communication to that application regardless of which app is set as a primary wallet application.

Those cases partially cover the feature set that ECP has, as described in Overview section, meaning that anticollision-level credential selection and NFC polling loop augmentation is now available on both Apple and Android devices.

Contributing

Missing pieces

The best way to help is to provide more samples of ECP frames and TCIs.
Especially interesting (missing) are the following:

  • TCIs of transit agencies that use EMV only (transit agency list at smp-device-content):
    • United Britain:
      • London (Possibly different for buses?);
    • United States:
      • New York (MTA);
    • Australia:
      • Sydney (NSW Transport);
    • France:
      • Dreux;
      • Lyon;
      • Montargis;
      • Nevers;
      • Tarbes-Lourdes.
    • Estonia:
      • Tartu.
    • Finland:
      • Turku.
    • Sweeden:
      • Malmo.
    • Belarus:
      • Minsk.
    • Ukraine:
      • Kyiv.
  • Access passes (There might be many unknown variations, so any samples would be welcome):
    • University;
    • Office;
    • Venues;
    • Hotels;
    • Multi-family housing;
  • Keys:
    • Car (Real device);
    • Access;
  • HomeKit pairing;
  • Identity (Standalone reader). Be careful not to get on a no-fly list if you dare to go and try sniffing data from it.

Meanwhile, some other unanswered questions remain, which require further analysis and testing, such as:

  • If more than one TCI can be sent by a reader;
  • If TCI has a format, at least in some situations;
  • If a TCI is 3 bytes long, or it could have a variable length (such as for transit), this question arose after seeing an "Identity" frame;
  • If the data part is actually considered a second TCI;
  • If the terminal subtype has some special encoding rules;
  • What the unknown configuration byte bits are responsible for (pairing or other flags?).

If you have any findings or thoughts on this matter, feel free to discuss them in the issues section.

Collecting information

Analysing pass files

The easiest way of retrieving useful information that doesn't require any special tools is .pkpass file analysis.
It can be done in the following way:

  1. Extract pass files:
    1. On MacOS, from ~/Library/Passes/Cards/;
    2. On jailbroken iPhone from ~/Library/Passes/Cards/;
    3. On a non-jailbroken iPhone, via an encrypted backup extracted at path HomeDomain/Library/Passes/Cards (Whoever patched it, send my regards).
      File extraction can be done using an app such as iMazing, iOSBackup or similar tools.
  2. Enter the folder with passes;
  3. Click on any .pkpass file, select "Show Package Contents"
  4. Open the pass.json file.
  5. Inside the file, search for keywords tci, openloop, ecp, transit, automatic, selection, express, as in example:
    {
      "formatVersion":1,
      "passTypeIdentifier":"paymentpass.com.apple",
      "teamIdentifier":"Apple Inc.",
      "paymentApplications":[
        {
            "secureElementIdentifier":"133713371337",
            "applicationIdentifier":"69696969696969696969",
            "applicationDescription":"Mastercard",
            "defaultPaymentApplication":true,
            "supportsOptionalAuthentication":1,
            "automaticSelectionCriteria":[
                {
                    "type":"ecp.2.open_loop",
                    "supportsExpress":true,
                    "openLoopExpressMask":"0800000000"
                }
            ],
            "supportedTransitNetworkIdentifiers":[],
        }
      ]
    }
    
    Other fields removed to reduce space taken
    Here we see that a Mastercard has automatic selection criteria 0800000000 which corresponds to 00001000 in binary for the first byte of transit data mask.
    All cards can be analyzed the same way.

Sniffing

The second way of collecting information about ECP is via sniffing.

Android

Thanks to Observe Mode feature introduced in Android 15, most flagship mobile handsets now have an ability to sniff polling frames without the use of any external hardware.

Android 15 Observe Mode Demo App can be used to collect data about the polling loop, including all custom polling frames such as ECP.

Proxmark

Sniffing can also be done using the functionality of a device like Proxmark (Easy or RDV2/4) connected to a Proxmark client inside of Termux running on an Android phone.
A couple of tidbits encountered:

  • The first time using the app I encountered an issue connecting to Proxmark3 directly as Termux did not connect a device, TCPUART app had to be installed to forward serial connection over the local network to be used in Proxmark client inside of Termux;
  • Some Android phones won't power Proxmark properly through direct connection. Connecting via a USB-C to USB-A dongle can help to overcome this issue.

More info on installing and running Proxmark client on your Android device here.

The command needed to collect traces is hf 14a sniff, after activating the command hold the Proxmark near a reader for a couple of seconds. In some cases, it is needed to tap/touch the reader in order to wake it up as it might not poll to save energy.

After that, press a button on a device, and traces will be downloaded and can be viewed with a hf 14a list command. You'll know which ones are the ones.

Some other devices might also be able to sniff the frames, but due to a lack of personal experience, I cannot recommend any.

Bruteforce

For the most dedicated people, there is a third method of finding useful information.

It is possible to find ECP frames using sheer brute force.

To do that, you have to program an NFC reader that does the following in a loop:

  1. Turn off the RF field;
  2. Wait 5-10 seconds;
  3. Turn on the RF field;
  4. Attempt to increment/modify one of the values in an ECP frame;
  5. Send the ECP frame 3 times in 0.5-second intervals;
  6. Attempt wake-up with anticollision; If successful continue, otherwise return to 1);
  7. Observe results, save useful info (uid, ats, sak, etc) try doing PPSE, DESFire functions, selecting basic AIDs.

A couple of tips:

  • If you plan on running the brute force while not nearby, you can record a video of your device.
  • Before brute forcing, decide what is your goal, what feature or pass you might want to activate, this will define which values you'll have to try changing.
  • Do not try to bruteforce all values mindlessly. From the info available we see that most combinations are done using changing terminal type, subtype, and first and last bytes of TCI.

Notes

  • If you find any mistakes/typos or have extra information to add, feel free to raise an issue or create a pull request;
  • Material provided here has been collected using publicly available methods, without access to the original specification. Do not consider the provided material as the only source of truth, as some assumptions and descriptions made might end up being false or not fully correct. This repository intends to get as close as possible to the truth, which might take some time and collaboration.
  • Information provided in this repository is intended for educational, research, and personal use. Its use in any other way is not encouraged.
  • Beware that ECP, just like any other proprietary technology, might be subject to legal protections depending on the jurisdiction. The mere fact of it being reverse-engineered does not always mean that it can be used in a commercial product as-is without causing an infringement. For use in commercial applications, you should contact Apple through official channels in order to get approval.
  • For tips on implementation, code examples, an overview of verified modules, refer to Examples directory;
  • For a code-friendly database of ECP frames, refer to the Resources directory.

References