2ping protocol
- Line protocol version: 4.1
- Document version: 20181225
Introduction
The 2ping protocol is a bi-directional ping system. This allows for several features, such as 3-way ping (akin to SYN, SYN/ACK, ACK), and the ability to determine whether packet loss was inbound or outbound. This is accomplished by the client and server each keeping state tables of needed incoming and outgoing requests. If packet loss occurs without a complete network failure, the client and server can eventually resume communication and compare notes about which sides received what requests.
Network transit
2ping operates over UDP, and is compatible with IPv4, IPv6 and possibly future Layer 3 protocols. The server/listener listens on port 15998. The client can use any high source port, and will send packets to the server on destination port 15998. The server will then respond to the client on the port the client originally used as a source port. The UDP message payload is currently completely binary data, and is designed with consistent and/or minimum space, consistent parsing, and forwards compatibility in mind. The protocol payload does not contain any IP-specific information, and is NAT/PAT safe.
Message format
The 2ping format is a variable length binary format, with the length determined by which opcodes are set. Opcode data lengths are in a standardized location, which allows for forward compatibility. A program utilizing the 2ping protocol can still parse a complete payload, even if the program does not understand a certain opcode.
All data is in network byte order (big endian), with one byte equal to 8 bits (1 octet). Bytes are most significant bit first, and multiple-byte sequences (flags and integers capable of being larger than 8 bits) are most significant byte first.
| Field | Length |
|---|---|
| Magic number 0x3250 | 2 bytes, required |
| Checksum | 2 bytes, field required, checksum optional |
| Message ID | 6 bytes, required |
| Opcode flags | 2 bytes, required |
| Opcode data | Variable, zero or more bytes, length depends on opcode flags |
| Padding | Variable, optional |
Padding may be added to a packet to bring it up to a desired minimum size. A 128 byte minimum packet size is recommended. Padding bytes should be null character bytes by default, but the contents of the padding is not important, as the packet parser should not use the contents of the padding for program operation.
The minimum valid length of a 2ping payload is 12 bytes (2-byte magic number + 2-byte checksum field + 6-byte message ID + 2-byte opcode flags with no flags enabled + zero padding).
Message IDs
Message IDs are 48 bits (6 bytes) of data to identify a specific message. The message ID must be 48 bits of pseudo-random data; do not use incrementing numbers or a discernible pattern. Likewise, do not attempt to analyze or devise logic from the message IDs that a peer sends.
Message IDs are unique within a socket's 5-tuple (local address, local port, peer address, peer port, protocol). For example, when implementing a listener, care should be taken to differentiate between message ID 0xF1E2D3C4B5A6 sent from peer 10.0.0.10, and message ID 0xF1E2D3C4B5A6 sent from peer 172.16.0.10, or even between the same message ID sent from the same host on different source ports. While the specification's random ID requirement and keyspace size makes the chance of a collision from two different peers unlikely, it is still possible.
It is understandable that a 48-bit random unique ID is not easy to read from an end user's perspective. The implementation may therefore wish to keep a local mapping between message IDs and a more human-friendly identifier, such as an incrementing integer.
Checksum
The checksum allows for verification of packet integrity from unintentional transit errors. (If cryptographic verification is desired to prevent intentional tampering in transit, see opcode 0x0080 below.) While UDP includes its own checksum, it is optional in IPv4 and cannot be relied upon. (IPv6 makes the UDP checksum mandatory.)
The 2ping protocol uses the following checksum method, as described in pseudocode:
checksum = 0
if length(input) is not even:
input.append(0)
for (a, b) in input as ((int1, int2), (int3, int4), ...):
checksum = checksum + (a << 8) + b
checksum = ((checksum & 0xffff) + (checksum >> 16))
checksum = ~checksum & 0xffff
if checksum == 0:
checksum = 0xffff
The input data is defined as the entire packet data payload, from the magic number to the end of the padding. The checksum field itself is zeroed for the purpose of the method input, and is replaced with the result of the method. An all zero transmitted checksum value means that the transmitter generated no checksum.
The checksum field is required, though an implementation is not required to perform or verify checksums. However, if verification of checksums are performed, the verifier must be able to recognize the zero transmitted checksum value from a peer that did not compute a checksum.
Checksum computation and verification attempts are strongly recommend for IPv4 packets, but should not be necessary for IPv6 packets, as verification is already performed at the UDPv6 level, and malformed packets should be discarded by the time they reach the userland.
This specification originally specified a different checksum method (RFC 768's UDP checksum method). However, it was discovered that the only known 2ping implementation which computed checksums at the time was using an incorrect method, so the specification was changed to match the implementation to preserve compatibility.
Opcodes
Opcodes' data in the opcode data area is stored in sequence according to which flags are set in the opcode flags field, starting with the least significant bit. Thus if no opcode flags are set, the opcode data area does not exist.
For example, if 0x0001 and 0x0020 flags are set, the entire opcode data area consists of the opcode header and segment data for 0x0001, followed by the opcode header and segment data for 0x0020.
2 bytes of opcode flags allow for up to 16 opcodes. The final flag (0x8000) is a container for an extended segment format, and allows for an arbitrary number of segments.
Each opcode data segment is comprised of the following:
| Field | Length |
|---|---|
| Segment data length (not including length header itself) | 2 bytes, required |
| Segment data | Variable, zero or more bytes, determined by segment data length header above |
An opcode data segment may be between zero and 65,535 bytes long, plus 2 bytes for the segment header.
This layout is designed for forward compatibility with future protocol revisions. A 2ping implementation can simply stop processing once it reaches the last opcode flag it understands, or it can at least parse the opcodes it does not understand, as the opcode data segment header includes the length of the segment data -- using this information, the implementation can simply skip over the opcodes it does not understand.
If there is a numeric gap in the understood opcodes of a 2ping implementation, it must at least check for the opcode and parse the opcode data segment in order to skip over it. For example, a 2ping implementation might understand 0x0001, 0x0002 and 0x0008, but not 0x0004, 0x0010 or 0x0020. To be able to parse the data segment for 0x0008, it must still check whether 0x0004 is set and seek to its segment data length header, determine the length of the segment data for 0x0004, and skip over it to begin parsing 0x0008. Due to the standardized length headers, this is possible without knowing what 0x0004 does. Once it is finished parsing 0x0008, it may stop checking opcode flags altogether.
The minimum 2ping implementation must be able to understand and follow opcodes 0x0001 (reply requested) and 0x0002 (in reply to). Implementation of all other opcodes are optional (but highly recommended, as investigation opcodes are the core focus of 2ping).
0x0001 - Reply requested
| Field | Length |
|---|---|
| No segment data | 0 bytes |
The sending end requests a reply from the receiving end. If the receiving end does not send a reply back to the sending end, the sending end will consider it a lost packet. Note: No segment data is part of this opcode, but as all used opcode segments must have a header, there will still be a 2-byte length header, signifying zero bytes of data.
0x0002 - In reply to
| Field | Length |
|---|---|
| Replied message ID | 6 bytes, required |
This opcode signifies the reply to a packet that requested a reply. The original sender's message ID is enclosed.
0x0004 - RTT enclosed
| Field | Length |
|---|---|
| RTT in microseconds | 4 bytes, required |
If this packet is a reply packet and an RTT is enclosed, the packet being replied to was a successful ping, and the RTT is the round trip time of the previous operation, in microseconds. Up to 2^32-1 microseconds are possible, approximately 71 minutes. In a typical 3-way ping scenario, this is only sent in the third message in the sequence, from the client to the server. Thus the server will know both the server->client->server ping RTT, as well as the original client->server->client ping RTT.
0x0008 - Investigation complete, originally replied to as requested
| Field | Length |
|---|---|
| Number of message IDs enclosed | 2 bytes, required |
| Message ID | 6 bytes, optional |
| Message ID... | 6 bytes..., optional |
This is a response to "0x0020 - Expected reply never received, please investigate", please continue reading below. The message IDs listed in this opcode are packets that the responder knows about, and had responded to. Since the requester never received the response, the requester can assume that the packet loss occurred inbound (relative to the requester).
In theory, the number of message IDs enclosed may be between zero and 65,535, but limits on the total byte length of the opcode data area will make this impossible to achieve. In addition, UDP, IP and Ethernet length restrictions will further limit the number of message IDs that can be enclosed. The number of message IDs enclosed may legally be zero, but if that is the case, it's better to just not set the 0x0008 opcode flag.
Investigation replies must not be unsolicited, and may only be sent in response to messages that requested a reply, and requested an investigation into specific message IDs. However, investigation responses by the requestee are not required, and can be ignored if packet payload space is limited or other problems arise. It is the responsibility of the requester to resend investigation requests if the requestee does not respond to them in its reply.
0x0010 - Investigation complete, request never received
| Field | Length |
|---|---|
| Number of message IDs enclosed | 2 bytes, required |
| Message ID | 6 bytes, optional |
| Message ID... | 6 bytes..., optional |
This is a response to "0x0020 - Expected reply never received, please investigate", please continue reading below. The message IDs listed in this opcode are packets that the responder does not know about. Since the responder never received the request, the requester can assume that the packet loss occurred outbound (relative to the requester).
In theory, the number of message IDs enclosed may be between zero and 65,535, but limits on the total byte length of the opcode data area will make this impossible to achieve. In addition, UDP, IP and Ethernet length restrictions will further limit the number of message IDs that can be enclosed. The number of message IDs enclosed may legally be zero, but if that is the case, it's better to just not set the 0x0010 opcode flag.
Investigation replies must not be unsolicited, and may only be sent in response to messages that requested a reply, and requested an investigation into specific message IDs. However, investigation responses by the requestee are not required, and can be ignored if packet payload space is limited or other problems arise. It is the responsibility of the requester to resend investigation requests if the requestee does not respond to them in its reply.
0x0020 - Expected reply never received, please investigate
| Field | Length |
|---|---|
| Number of message IDs enclosed | 2 bytes, required |
| Message ID | 6 bytes, optional |
| Message ID... | 6 bytes..., optional |
If the requester sends a packet to a remote party and indicates a reply is requested (0x0001), and a reply is never received, the requester can use this opcode to inquire about what happened to it. If replies to this inquiry come back in either "0x0008 - Investigation complete, originally replied to as requested" or "0x0010 - Investigation complete, request never received" opcodes (as explained above), the requester can use this information to determine whether the packet loss occurred inbound or outbound relative to the requester.
In theory, the number of message IDs enclosed may be between zero and 65,535, but limits on the total byte length of the opcode data area will make this impossible to achieve. In addition, UDP, IP and Ethernet length restrictions will further limit the number of message IDs that can be enclosed. The number of message IDs enclosed may legally be zero, but if that is the case, it's better to just not set the 0x0020 opcode flag.
Care should be taken not to begin inquiring too quickly. UDP packets may arrive delayed or out of order, so 10 seconds should be a good amount of time to wait before inquiring.
The requester may send inquiries multiple times. The requestee is not required to respond to a specific inquiry immediately, as UDP payload space may be limited. Thus it is the responsibility of the requester to continue to send inquiries until it receives a response, so the requestee does not need to keep a state table of unreplied inquiries.
0x0040 - Courtesy message ID expiration
| Field | Length |
|---|---|
| Number of message IDs enclosed | 2 bytes, required |
| Message ID | 6 bytes, optional |
| Message ID... | 6 bytes..., optional |
To facilitate bi-directional packet loss detection, it is necessary for each peer to maintain a set of state tables: messages expecting a reply and not yet received, and remote peer messages expecting a reply that the near end has replied to. Maintenance of the first cache is easy; simply remove a message ID once a reply or investigation has been received. However, the second case is trickier. In some cases, it may be possible to know if the far end has received an expected response and will not inquire about it. But in others, there may not be enough hints to be able to guarantee a peer will not ask for an investigation later.
In particular, in a 3-way ping (assuming "Peer 1" initiated the ping), Peer 2 can figure out that the near end is satisfied with the response to the first leg, since Peer 1 sent the third leg to Peer 2, and so Peer 2 can remove the first leg's message ID from its cache. However, Peer 1 cannot tell if Peer 2 received the third leg (a response to the second leg), since a response to the third leg does not exist. But Peer 2 could in the future inquire about the response to the second leg sometime in the future.
In this situation, it would be best to do a periodic cleanup of the state tables, but ideally the age of the message ID should be long enough (recommended default is 10 minutes), since it may take the peer awhile to request an investigation. In a session with one or two pings per second that's not a lot, but in a flood mode, thousands or millions of message IDs could get stored in the state table during that 10 minute period.
To combat this, the courtesy opcode contains a list of message IDs that a peer is no longer concerned about, and can be removed from the remote peer's cache. Courtesy opcodes can be sent any time a packet is being sent from one peer to another (unlike investigation requests, which may only be sent when requesting a reply, or investigation results, which are in the immediate response to the investigation request).
This opcode is optional (but recommended), and due to its opportunistic nature, it is not guaranteed the message will reach the remote peer. So it is still up to an implementation to do periodic maintenance on its caches, since it cannot rely on courtesy responses to prune its caches.
Peers may send unsolicited messages with courtesy opcodes (a message not requesting a reply, not replying to another message, and containing the courtesy opcode), but in the normal course of operation, it should not be necessary.
0x0080 - Message authentication code (MAC)
| Field | Length |
|---|---|
| Digest type index | 2 bytes, required |
| Computed hash value | Length depends on digest type, required |
The message authentication code field is allowed to provide cryptographic data integrity and authentication of the remote side. This helps to prevent injection and replay attacks during transit.
The payload data, combined with a shared secret key, is hashed and is verified by the other peer. The digest types supported are:
| Index | Digest Type | Computed hash size |
|---|---|---|
| 0 | Private / locally reserved | Variable |
| 1 | HMAC-MD5 | 16 bytes |
| 2 | HMAC-SHA1 | 20 bytes |
| 3 | HMAC-SHA256 | 32 bytes |
| 4 | HMAC-CRC32 | 4 bytes |
| 5 | HMAC-SHA512 | 64 bytes |
The two peers must use the same key and digest type. If a peer is instructed to hash its messages, it must not accept replies that are not hashed with the same digest type and key.
Index 0 may be used by clients for digest types not defined by this specification, and implementations must ignore index 0 hashes if a non-standard digest has not been defined locally. Additional digest types may be added to this specification in the future.
The entire payload is hashed, from the magic number to the padding, inclusive. When computing the MAC hash, the hash value itself is filled with zeroed bytes, and replaced with the computed hash. The MAC hash is computed before the payload checksum, therefore the checksum area must also be zeroed before the MAC hash is computed.
Use of this opcode is optional and its parameters (whether to use hashing, the digest type and the shared secret) must be coordinated between the two sides ahead of time. An implementation is not required to implement all digest types.
HMAC-CRC32 is included in this specification mostly as a joke, and a programming exercise for the reader. Please don't use this (and expect any sort of cryptographic data integrity). Endianness is not specified by CRC32; use network byte order (big endian). The HMAC specification requires aligning the key to the digest function's block size; CRC32 does not use a block size, so assume a block size of 64 bytes (same as MD5, SHA1, SHA256 and SHA512).
0x0100 - Host processing latency
| Field | Length |
|---|---|
| Delay in microseconds | 4 bytes, required |
The time between receiving a message packet and sending a reply packet may be non-trivial, due to host processing. This opcode can be used by an implementation, when sending a reply (0x0002), to inform a peer how much time progressed between a receive and a send. The peer may then subtract this latency time to better calculate network transit time. Up to 2^32-1 microseconds are possible, approximately 71 minutes.
The delta should be computed from immediately after the original message packet has been received, to as soon as possible before the reply packet has been sent. Note that due to optional MAC hash and checksum calculation processing, which must be computed after the rest of the opcode data area, there still could be some host processing latency not accounted for in the reported latency time.
This opcode should only be used when combined with a monotonic clock, as using a wall clock may result in processing being greater than the RTT, less than zero, etc.
0x0200 - Encrypted packet
| Field | Length |
|---|---|
| Method index | 2 bytes, required |
| Encrypted data | Variable, required |
This opcode acts as a container for a shared-secret encrypted 2ping packet. A complete valid 2ping packet, from magic number and checksum through optional padding, is encrypted using a known method and contained in this opcode in a stub 2ping packet.
The stub packet, which is what is sent along the the wire, is also a valid 2ping packet, but contains only a single opcode, 0x0200. When the encrypted payload in the stub packet is decrypted on the receiving end, the decrypted packet is used in place of the stub packet and is parsed normally. The stub packet's message ID is not used and should be pseudo-random. Implementations must not re-use the encrypted packet's message ID for the unencrypted stub's message ID.
When one end specifies it is using encryption, it must not accept replies from the other end if the message is not encrypted, the method index does not match, or the encrypted payload does not decrypt to a valid 2ping packet.
As of this protocol revision, a single method is specified, HKDF-AES256-CBC (index 1). Implementations may use a locally-reserved method by specifying index 0, as long as both ends have implemented the method.
Method index 1 - HKDF-AES256-CBC
- Shared secret
- HKDF (RFC 5869) key derivation function, extract + expand rounds
- Input key material: shared secret (UTF-8)
- Salt: 16 byte (128 bit) per-message pseudo-random value
- Output key length: 32 bytes (256 bits)
- Expand round info value: 0xd889ac93aceba1f398d0c69bc8c6a7aa + 8 byte (64 bit) per-session pseudo-random value (concatenated)
- AES encryption
- Cipher mode: AES-CBC
- Key: 32 byte (256 bit) output of HKDF extract + expand rounds above
- Initialization vector (IV): Same as HKDF salt above
- Input: Complete unencrypted 2ping packet
The session ID, IV/salt and the result of the AES encryption are concatenated to form the encrypted data field. To decrypt, extract the session ID from the first 8 bytes of the encrypted data field, the IV/salt from the following 16 bytes, and use the method to AES decrypt the remaining bytes.
AES has a 128 bit block size, which data must be padded to when encrypting. As trailing padding is a core feature of 2ping, no special padding method is required. The implementation only needs to make sure the unencrypted packet length is a multiple of 16 bytes before encryption.
0x8000 - Extended segments
This opcode data segment extends the 2ping format, allowing for a nearly unlimited number of available segments. Within the opcode segment data area is a series of sub-segments, referred to as extended segments, consisting of a 32-bit extID, a length of the extended segment data, and the extended segment data itself. Each part is built as so:
| Field | Length |
|---|---|
| Extended segment ID | 4 bytes, required |
| Extended segment data length | 2 bytes, required |
| Extended segment data | Variable, zero or more bytes, determined by extended segment data length header above |
Multiple extended segment parts are chained together to form the data area of a 0x8000 opcode. The opcode data area must include a maximum of one instance of any individual extended segment.
Parsing this opcode's data area is similar to parsing the opcodes as a whole: read the extended segment ID and determine if it's a known ID. If so, parse its contents by reading the extended segment data length and extended segment data. If not, read the extended segment data length to determine how far to skip over the extended segment data. If no extended segments are to be included with the packet payload, do not set the opcode.
Defined extended segments
Extended segment IDs are 32 bits, allowing for approximately 4.2 billion different pieces of functionality. When defining a local extension, please choose a random ID to avoid the possibility of conflicting extensions. If the extension is useful to the public as a whole, please consider submitting it for inclusion in this specification.
The following are registered extended segments. An implementation is not required to implement any extended segments (or the extended segment opcode itself, for that matter), but you are strongly encouraged to do so.
0x3250564e - Program version
| Field | Length |
|---|---|
| Program version text (UTF-8) | Variable length |
The human-readable text version of the program or firmware generating the packet, with optional information such as architecture, etc. An example could be "Network Tools 3.0-1distro2 (x86_64-linux)".
As the 2ping protocol is designed to be backwards and forwards compatible, this field must not be used by an implementation to determine functionality. It is recommended that this field be sent with every packet, but received segments should not be displayed to the user unless in a verbose/debug/etc mode.
This segment's number, 0x3250564e, evaluates to ASCII "2PVN" ("2ping version number"), and was chosen before the decision to recommend segment numbers be randomly assigned to avoid collisions with unregistered extended segments. All other extended segment numbers in this specification were randomly chosen.
0x2ff6ad68 - Random data
| Field | Length |
|---|---|
| Flags | 2 bytes, required |
| Random data | Variable length |
Random data as generated by the host's random number generator. The following flags may be present:
- 0x0001 - Whether the data was generated by a hardware random number generator.
- 0x0002 - Whether the data was generated by the operating system's random number generator.
If this segment is to be used for cryptographic or trusted purposes, it should be combined with "0x0080 - Message authentication code (MAC)" or "0x0200 - Encrypted packet" to ensure end-to-end integrity.
0x64f69319 - Wall clock
| Field | Length |
|---|---|
| Time in microseconds | 8 bytes, required |
Host time (wall clock) of the sender, in microseconds since 1970-01-01 00:00:00 UTC (Unix epoch).
0x771d8dfb - Monotonic clock
| Field | Length |
|---|---|
| Generation ID | 2 bytes, required |
| Time in microseconds | 8 bytes, required |
Monotonic clock time, in microseconds since an epoch. The epoch is arbitrary; set the generation ID to a random value and the epoch to a random offset from the host's monotonic clock. The generation ID and epoch may occasionally be regenerated together (for example, every few hours or days), to avoid leaking information about wall clock or program start time.
A peer may compare two successive values by making sure the generation IDs match and the later time is greater than the earlier time.
This segment must only be sent if the host is capable of using a monotonic, high-precision clock.
0x88a1f7c7 - Battery levels
| Field | Length |
|---|---|
| Number of batteries enclosed | 2 bytes, required |
| Battery ID | 2 bytes, optional |
| Battery level | 2 bytes, optional |
| Battery ID... | 2 bytes..., optional |
| Battery level... | 2 bytes..., optional |
If the host is a device which includes batteries, this may be used to report their levels. Multiple batteries may be reported using different battery IDs. The level is indicated as a percentage between 0x0000 (completely empty) and 0xffff (completely full).
The number of batteries enclosed may legally be zero, but if that is the case, it's better to just not include the 0x88a1f7c7 segment.
0xa837b44e - Notice text
| Field | Length |
|---|---|
| Notice text (UTF-8) | Variable length |
Arbitrary text to be sent with the packet. This text should be defined by the user by a program flag, UI option, etc. It is designed to be human-readable; do not use this segment to pass machine-parsed data between the client and listener. Instead, if extended data transfer is desired between the client and listener, simply define a new extended segment ID.
The implementation may display this text to the user, but it is not guaranteed the user will see it.
Procedures
In the following pseudocode examples, message IDs are represented as zero-padded incrementing numbers. This is for the purpose of illustrating reply chains in these examples only. In real life, message IDs MUST be randomly-generated 48-bit (6 byte) identifiers. The program implementing the 2ping protocol may choose to locally associate a better identifier for the user (such as an incrementing integer), but the protocol message ID must be random and pseudo-unique.
As mentioned above, the simplest 2ping packet is a 12-byte payload: 2 bytes for the magic number (always 0x3250), 2 bytes for the checksum field, 6 bytes for the message ID, and 2 blank opcode flag bytes:
Example 1
CLIENT: 00000000a001, no opcodes
Of course, this isn't very useful, and would be analogous to a NOOP. The simplest ping would require the "reply requested" opcode:
Example 2
CLIENT: 00000000a001, reply requested
SERVER: 00000000b001, in reply to 00000000a001
A 3-way ping further extends this.
Example 3
CLIENT: 00000000a001, reply requested
SERVER: 00000000b001, reply requested, in reply to 00000000a001
CLIENT: 00000000a002, in reply to 00000000b001, successful ping rtt 12345 µs
The client successfully received the response from the server and was able to measure an RTT of 12345 µs. The client then sends a reply back to the server, referencing the original reply, and letting it know the RTT it measured. The server is then able to determine its RTT between the second and third leg (say, 11804 µs), and also know the RTT between the first and second leg.
Now let's say the reply to the original client ping never came back. The client can start inquiring about whether the server saw the original request, and the server can provide info at the same time it is responding to a new ping request:
Example 4
CLIENT: 00000000a001, reply requested
CLIENT: 00000000a002, reply requested, did not receive reply to 00000000a001
SERVER: 00000000b002, reply requested, in reply to 00000000a002, received and replied to 00000000a001
CLIENT: 00000000a003, in reply to 00000000b002, successful ping rtt 12345 µs
Now the client can tell that the server received the original request, and replied, which suggests inbound packet loss. Note from the example message IDs, the server most likely originally replied to 00000000a001 with 00000000b001. However, the specific message ID the server originally replied with is not important and not tracked, and hence is not indicated in the investigation reply, only that it was originally received and replied to.
Conversely, for a request for reply that the server never actually received:
Example 5
CLIENT: 00000000a001, reply requested
CLIENT: 00000000a002, reply requested, did not receive reply to 00000000a001
SERVER: 00000000b001, reply requested, in reply to 00000000a002, never received 00000000a001
CLIENT: 00000000a003, in reply to 00000000b001, successful ping rtt 12345 µs
This suggests outbound packet loss. As mentioned before, the client should not immediately start asking about replies never received. An actual sequence of events may look something like this:
Example 6
CLIENT: 00000000a001, reply requested
CLIENT: 00000000a002, reply requested
CLIENT: 00000000a003, reply requested
SERVER: 00000000b002, reply requested, in reply to 00000000a003
CLIENT: 00000000a004, in reply to 00000000b002, successful ping rtt 12823 µs
... etc
CLIENT: 00000000a00a, reply requested, did not receive reply to 00000000a001 or 00000000a002
SERVER: 00000000b006, reply requested, in reply to 00000000a00a, received and replied to 00000000a001, never received 00000000a002, did not receive reply to 00000000b002
CLIENT: 00000000a00b, in reply to 00000000b006, successful ping rtt 13112 µs, received and replied to 00000000b002
The client waited a few seconds before asking about 00000000a001 and 00000000a002. The server replied that it knew about 00000000a001, but never received 00000000a002, indicating there is some packet loss both inbound and outbound.
Also notice that in the same packet where the server replied with info about the client's lost packets inquiry, it also inquired about its own lost packet. In this case, the server never received the last segment of the 3-way ping, 00000000b002. The client then tells the server it did originally receive and respond to the packet in question.
This raises an interesting point, that there is essentially no difference between a server and a client in 2ping. The "server" is expected to listen for the initial datagrams, and the "client" is expected to initiate ping requests at regular intervals. But if the "server" decides to randomly initiate a ping request of its own, the "client" is expected to respond appropriately, as a server would do.
Reference packet dumps
Example 1
CLIENT: 32 50 2d ae 00 00 00 00 a0 01 00 00
Example 2
CLIENT: 32 50 2d ad 00 00 00 00 a0 01 00 01 00 00
SERVER: 32 50 7d a4 00 00 00 00 b0 01 00 02 00 06 00 00 00 00 a0 01
Example 3
CLIENT: 32 50 2d ad 00 00 00 00 a0 01 00 01 00 00
SERVER: 32 50 7d a3 00 00 00 00 b0 01 00 03 00 00 00 06 00 00 00 00 a0 01
CLIENT: 32 50 4d 62 00 00 00 00 a0 02 00 06 00 06 00 00 00 00 b0 01 00 04 00 00 30 39
Example 4
CLIENT: 32 50 2d ad 00 00 00 00 a0 01 00 01 00 00
CLIENT: 32 50 8d 81 00 00 00 00 a0 02 00 21 00 00 00 08 00 01 00 00 00 00 a0 01
SERVER: 32 50 dd 8e 00 00 00 00 b0 02 00 0b 00 00 00 06 00 00 00 00 a0 02 00 08 00 01 00 00 00 00 a0 01
CLIENT: 32 50 4d 60 00 00 00 00 a0 03 00 06 00 06 00 00 00 00 b0 02 00 04 00 00 30 39
Example 5
CLIENT: 32 50 2d ad 00 00 00 00 a0 01 00 01 00 00
CLIENT: 32 50 8d 81 00 00 00 00 a0 02 00 21 00 00 00 08 00 01 00 00 00 00 a0 01
SERVER: 32 50 dd 87 00 00 00 00 b0 01 00 13 00 00 00 06 00 00 00 00 a0 02 00 08 00 01 00 00 00 00 a0 01
CLIENT: 32 50 4d 61 00 00 00 00 a0 03 00 06 00 06 00 00 00 00 b0 01 00 04 00 00 30 39
Example 6
CLIENT: 32 50 2d ad 00 00 00 00 a0 01 00 01 00 00
CLIENT: 32 50 2d ac 00 00 00 00 a0 02 00 01 00 00
CLIENT: 32 50 2d ab 00 00 00 00 a0 03 00 01 00 00
SERVER: 32 50 7d a0 00 00 00 00 b0 02 00 03 00 00 00 06 00 00 00 00 a0 03
CLIENT: 32 50 4b 81 00 00 00 00 a0 04 00 06 00 06 00 00 00 00 b0 02 00 04 00 00 32 17
... etc
CLIENT: 32 50 ed 6f 00 00 00 00 a0 0a 00 21 00 00 00 0e 00 02 00 00 00 00 a0 01 00 00 00 00 a0 02
SERVER: 32 50 8d 3b 00 00 00 00 b0 06 00 3b 00 00 00 06 00 00 00 00 a0 0a 00 08 00 01 00 00 00 00 a0 01 00 08 00 01 00 00 00 00 a0 02 00 08 00 01 00 00 00 00 b0 02
CLIENT: 32 50 9a 41 00 00 00 00 a0 0b 00 0e 00 06 00 00 00 00 b0 06 00 04 00 00 33 38 00 08 00 01 00 00 00 00 b0 02
Changelog
4.1 (20181225)
- Moved all opcode-related information to the Opcodes section.
- Program version text, notice text and HKDF input key material are explicitly specified as being UTF-8 encoded.
- Added note that host processing latency should only be used with a monotonic clock.
- Clarified monotonic clock best practices.
- Added note about the meaning of extended segment number 0x3250564e, and how it shouldn't be used as a guide.
- Changed all instances of "octet" to "byte", and explicitly defined a byte as 8 bits.
- General cleanup and minor clarifications.
4.0 (20170806)
- Added opcode 0x0200 - Encrypted packet
3.2 (20170722)
- Added HMAC-SHA512 (index 5) to 0x0080 MAC digest types.
- Added the following registered extended segments:
- 0x88a1f7c7: Battery levels
- Clarified that only one instance of a specific ExtID may be present in a packet.
3.1 (20160210)
- Added the following registered extended segments:
- 0x2ff6ad68: Random data
- 0x64f69319: Wall clock
- 0x771d8dfb: Monotonic clock
3.0 (20151025)
- Changed the checksum method from RFC 768 to a custom method with example pseudocode. This creates a functional incompatibility with previous versions of the specification. However, it was discovered that the only known 2ping implementation which computed checksums at the time was using an incorrect method, so the specification was changed to match the implementation to preserve compatibility.
- Fixed a typo in the extended segment table, changing "Extended segment ID" from 8 bytes to 4 bytes. This clarifies the previous (correct) assertion that the segment ID is 32 bits (4 bytes).
- Populated checksums of example packet dumps.
- Corrected the 5th client line of Example 6's packet dump (pseudocode version was correct, but example dump had 2 incorrect bytes added).
- Adjusted wording of psuedocode examples to clarify "reply requested" opcode comes before "in reply to" opcode.
2.0 (20120422)
- Protocol versions 1.0 and 2.0 are backwards and forwards compatible with each other.
- Changed recommended default minimum packet size from 64 bytes to 128 bytes.
- Added an extended segment container at opcode 0x8000.
- Added the following registered extended segments:
- 0x3250564e: Program version
- 0xa837b44e: Notice text
1.0 (20101020)
- Finalized initial release.
Copyright
Copyright (C) 2010-2021 Ryan Finnie
This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.