USB Protocol: Difference between revisions

From Hackspire
Jump to navigation Jump to search
m (Typos)
Line 139: Line 139:


===USB traffic analysis===
===USB traffic analysis===
<span style="color: IndianRed">Any suggestions of a free USB analyzer for Microsoft Windows?</span>
<span style="color: IndianRed">Any suggestions of a ''free'' USB analyzer for Microsoft Windows?</span>


==TODO==
==TODO==

Revision as of 20:32, 27 August 2007

This article describes the protocol used for USB communication between the computer and the TI-Nspire. It has been documented from TI's official linking software Computer Link Software, an analysis of USB captures of data traffic, and results of tests with a custom host implementation. This documentation may be used to build a full-featured third-party implementation of the computer program, to use features provided by the link not available from Computer Link Software's GUI, to indirectly gather additional information on the TI-Nspire hardware and software, and to discover exploitable flaws of the TI-Nspire OS's implementation of the protocol.

The documentation is currently incomplete, sometimes vague or too restrictive, and may be wrong on some points. Feel free to contribute to its enhancement and refinement. Interrogations and information based on assumptions which needs to be confirmed by a deeper analysis, more tests with Computer Link Software or tests with a third-party host implementation are formatted in indian red. Once enough tests have been made to validate or correct these parts, please edit them and remove the highlighting.

The descriptions of the packets of the protocol let sometimes appear hard-coded value, for which it is not clear whether the corresponding field is constant or may vary under certain conditions. You may edit these descriptions to document additional logic found for them. Limit cases have not all been tested: most of the time the TI-Nspire's implementation of the protocol handles them and make use of error codes. These error codes have not all been documented yet.

Overview

We will call host the participant of the communication which is the USB host, i.e. the computer for a transfer between a computer and a TI-Nspire. The host of a transfer between two TI-Nspires is probably also the USB host, determined with the Host Negotiation Protocol of USB On-The-Go. The communication model of the TI-Nspire seems to allow multiple devices to communicate simultaneously with the same host, through not yet released USB hubs, similar to TI-Navigator available for former calculator models. The TI-Nspire and the computer thus have each a dynamically assigned address used to identify the source and destination of a packet. A message sent from a source host/device to a destination host/device will be called a packet in the rest of this documentation. It is actually a protocol packet which may be transparently split into several USB packets by the USB device and host stacks if its length is greater or equal than the maximum USB packet size supported by the TI-Nspire (64 bytes).

Packets are exchanged between services of the host and the device. A service is a software module which produce and interpret a subset of packet types. A service is identified by a two bytes identifier. Each packet contains both the source and destination service identifiers. A service of a participant sending a packet choose the target service it wants to talk with. Service IDs are similar to TCP ports, except that multiple source and destination services are used during the same session (more specifically for acknowledgment) as we will see.

We will use further in the documentation notations of the form: 6400:8001->6401:4060 where 6400 is here the source address, 8001 the source service, 6401 the destination address and 4060 the destination service.

After a connection reset (described further) and once a device has been declared to the host, transfers are initiated by the host and have this form (each line represents a packet):

Host: Request
Device: ACK (acknowledgment)
Device: Response
Host: ACK (acknowledgment)

What could be called a "functional request" (for example "transfer a file") may use serveral exchanges of this type. The host and the device have stateful sessions and can keep track of the current state between these exchanges (for example when listing the content of a directory, which requires one pair of request/response for each file, the device keeps track of the current file).

Packet format

Packets send by a host or a device have the same format. A packet has a 16 bytes header and may have an optional data part of variable length. A packet is at most 270 bytes long, so the data part is at most 254 bytes long.

54 FD SA SA SS SS DA DA DS DS DC DC SZ AK SQ CK [data part]

The different fields of the header part are:

  • 54 FD: constant
  • SA SA: source address
  • SS SS: source service ID
  • DA DA: destination address
  • DS DS: destination service ID
  • DC DC: checksum of the data part
  • SZ: size of the data part
  • AK: used for acknowledgment packets, 00 for other packets.
  • SQ: sequence number
  • CK: checksum of the header: the sum of the preceding bytes modulo 256

Here is additional information for the fields not obvious and not described before.

Service identifiers

The host which always initiate the transfers (except address assignment during described further) choose the service ID it uses for standard (not acknowledgment) packets. It will be the source service ID that appears in the packets sent to the device, and the destination service ID of the responses of the device. Any non-standard service ID may be chosen. Computer link Software uses service ID greater or equal than 0x8001. The service ID must be chosen before starting to exchange packets with a new service of the device. When using another service of the device, the host must disconnect itself from the previous service (this process is described further), and choose a new local service ID for the subsequent packets for the new target service. Each time a new host service ID is chosen, it must be different that the previous ones. The device won't answer to packets with a source service ID which matches a host service which has previously asked to disconnect itself. The same service IDs can be reused only once the connection with the device is reset. At each generation Computer Link Software increments by one the service ID to ensures unique generation, but this is actually not mandatory.

The identifiers (in hexadecimal) of the currently known standard services on the device side are:

  • 00FF: packet reception acknowledgment
  • 4003: device address request
  • 4004: service tuning (?)
  • 4020: device information
  • 4021: screen capture
  • 4041: service activity (?)
  • 4050: authentification (?)
  • 4054: hub connection (?)
  • 4060: file management
  • 4080: OS installation
  • To be completed

The identifiers of the services on the host side are:

  • 00FF: packet reception acknowledgment
  • 4003: device address assignment
  • 40DE: service deconnection
  • To be completed

Sequence number

Sequence numbers are solely used for proper acknowledgment of packets. Each participant manages itself the generation of the sequence numbers for the packets it sends (a communication between a computer and a TI-Nspire brings into play two sequences). Each packet (except acknowledgment packets, see further, and except in the case of an overflow) has it own sequence number. The number is incremented each time a packet is sent (do they really need to be consecutive?). The generation is common for all the services of a participant, i.e. two consecutive packets sent by/to different services will have consecutive sequence numbers. The sequence number is reinitialized to 1 after a connection reset (does the host really have to make it start from 1?). The sequence number which follows 255 is 1, 0 is never used (is there really a reason for this are does it still work with 0?).

Acknowledgment

Whether send by the host or a device, an acknowledgment packet always have 0x00FF as source service ID. The destination service ID must be the source service ID of the previously received packet the participant acknowledges. The sequence number must be the same as the one of the packet acknowledged. The field AK of the header must contain 0x0A (is there a special value for NACK packets? Do NACK packets really exist?). The 2 byte-long data part must contain the destination service ID of the packet acknowledged (i.e. a local service ID). Here is an example of an acknowledgment packet sent by a TI-Nspire to a computer:

6401:00FF->6400:8001 AK=0A SQ=02 
data part: 40 60

Here the device acknowledges the reception of a packet sent by the service 8001 of the host, which had 02 as sequence number, and was sent to the service 4060 of the TI-Nspire.

The TI-Nspire will acknowledge the reception of invalid packets with minor errors (for instance with a bad header or data checksum, or with a duplicate sequence number), but the target service won't take the packet into account.

Data part checksum

The 2 byte-long checksum is more or less a CRC checksum. The checksum is 00 00 if there is no data part in the packet. Here is an implementation in Python of the algorithm:

def checksum(data):
    acc = 0
    for byte in data:
        first = (ord(byte) << 8) | acc >> 8
        acc = acc & 0xFF
         second = (((acc & 0xF) << 4) ^ acc) << 8
       third = second >> 5
        acc = third >> 7
        acc = (acc ^ first ^ second ^ third) & 0xFFFF
    return acc

# A few examples
assert checksum('\x00\xFF') == 0xFF00
assert checksum('\x05\x00\x00') == 0x57AD
assert checksum('\x20\x00\x00\x00\x00\x00\x00\x00\x05\x01\x00') == 0xA095

Data part

The format and interpretation of the data part depends on the source and destination services for the packet, and on the current state of the exchange. Multiple-bytes integers that appear in the data part (for example sizes are service identifiers) are in big endian (most significant byte first). Strings are terminated with a null byte (the examples shown in this documentation will always make the null byte appear to prevent from forgetting it).

Common packets with a data part are those which indicates the success or failure of the processing initiated by the previous request packet (or response packet in the case of a processing by the host). These packets should not be confused with packets acknowledging the reception: for example a request packet can be received successfully, but its data part can contain incorrect data - the device would here return an ACK packet and then a failure packet. These success and failure packets are 2 bytes long, starts with 0xFF and are followed by 0x00 in case of success, or an error code in case of failure. The error code is not specific to a type of request (but is it specific to a service?), some codes are reused when they have the same meaning (for example in the case of the file management service).

Services

Now that we have seen the common parts of the protocol, let's review the functions offered by each service. The descriptions of the various IN (from the device to the host) and OUT (from the host to the device) packets will be presented in this format (random values are used here):

OUT: 6400:8001->6401:4060
FF 00

In this example the service 0x8001 of the host whose address is 0x6400 sends to the service 0x4060 of the device with address 0x6401 the 2 byte-long sequence FF 00 in the data part of the packet (which is a success packet). The examples in the following sections will most of time use for common packets 0x6400 as host address, 0x6401 as device address and 0x8001 as generated host service ID. Values which are not example values will appear in bold. Multi-bytes fields in the data part of some packets which are not constant values will be surrounded with square brackets. If the field has a constant length, its length will appear just after the opening bracket. Fields with no specified length have a variable length.

[2 device address]: a 2 byte-long field containing the device address
[directory path] 00: a field of variable length containing the path as a null-terminated string

Mandatory IN and OUT acknowledgment packets will never be shown, only request and response packets will be described. Exchanges which don't fit in this type of description will be presented with enough details to fully understand their special features.

Address assignment

Before starting any communication with the device, whether the host program has already been started and the device has just been plugged in, or the host program has just been started and the device was already plugged, the host must first:

  1. reset the connection
  2. assign an address to the device

The connection reset is mandatory because it forces the device to forget the identifiers of the host services and sequence numbers used in the previous communication. Not resetting the connection may cause the reuse of these unique elements, and thus the absence of response to the host requests. A connection reset is triggered by issuing the IOCTL_INTERNAL_USB_RESET_PORT kernel-mode I/O request on Microsoft Windows.

Just after the reset, the device will spontaneously send an address request packet:

IN: 0000:4003->0000:4003

The host must response with an address assignment packet:

OUT: 6400:4003->6401:4003
[2 device address] FF 00

In this example, the address which appears in the data part of the packet would be 6401 (the same as the destination address in the header). Note that the host should use in the source address field of the header the address it has chosen for itself, and keep it for all the subsequent packets. Computer Link Software always choose 6400 as host address and assigns the address 6401 to the device (what if more than one TI-Nspire is plugged to the computer?). The two bytes FF 00 mean that the host accepts the address request.

In the special case of address assignment, the device will not send an acknowledgment packet for the address assignment packet.

The host can now use whatever device service he would like to. It has to choose before a service identifier, and will do this before starting to use any service.

Service disconnection

Nothing special needs to be sent by the host to connect to a service of a device before using it. Simply using the correct destination service ID in the request packets it sends is enough.

Once the host has started to use a device service, it needs to disconnect from the current service before using a different service. The request is:

OUT: 6400:40DE->6401:[id of the device service from which to disconnect]
[id of the host service which was used until then to communicate with the device service]

For example if the host uses the device service 4060 with the generated host service ID 8001 and wants to disconnect from it:

(... The host uses the service. Some requests with the header:
  OUT: 6400:8001->6401:4060
...)
OUT: 6400:40DE->6401:4060
8001

Note that Computer Link Software disconnects very often from the services it uses, even between requests to the same service (for example between a directory listing and a file attributes request to the file management service). This is not necessary.

Testing

Computer Link Software

Computer Link Software's functions can be tested on the command line using the testing code embedded in the Java binding NavNet.jar for the low-level connector. TESTNUM is the number of the test to run (remove the echo at the beginning and the file redirection at the end the first time to understand how it works). The command line to use with Cygwin on Microsoft Windows is:

$ echo $TESTNUM | java -cp "C:\Program Files\TI Education\TI-Nspire Computer Link\lib\navnet.jar" com.ti.eps.navnet.main  >logs 2>&1

The logs of both the native NavNet connector and the Java testing program will be written to the output file. The verbosity of the native logs can be tuned in Main.class:

NavNet.init("-c X -d X -p connectors"); // X (initially 4) should be between 0 (no logs) and 5 (finest)

USB traffic analysis

Any suggestions of a free USB analyzer for Microsoft Windows?

TODO

  • Move the section on USB descriptors here
  • Find the differences between PC/TI-Nspire transfers and TI-Nspire/TI-Nspire transfers.
  • Test and document the services available