| Network Working Group | R. Hamilton |
| Internet-Draft | J. Iyengar |
| Intended status: Informational | I. Swett |
| Expires: July 16, 2016 | A. Wilk |
| January 13, 2016 |
QUIC: A UDP-Based Secure and Reliable Transport for HTTP/2
draft-tsvwg-quic-protocol-02
QUIC (Quick UDP Internet Connection) is a new multiplexed and secure transport atop UDP, designed from the ground up and optimized for HTTP/2 semantics. While built with HTTP/2 as the primary application protocol, QUIC builds on decades of transport and security experience, and implements mechanisms that make it attractive as a modern general-purpose transport. QUIC provides multiplexing and flow control equivalent to HTTP/2, security equivalent to TLS, and connection semantics, reliability, and congestion control equivalent to TCP.
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at http://datatracker.ietf.org/drafts/current/.
Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."
This Internet-Draft will expire on July 16, 2016.
Copyright (c) 2016 IETF Trust and the persons identified as the document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (http://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.
This protocol is the outcome of work by many engineers, not just the authors of this document. The design and rationale behind QUIC draw significantly from work by Jim Roskind. In alphabetical order, the contributors to the project are: Britt Cyr, Ryan Hamilton, Jana Iyengar, Fedor Kouranov, Charles Krasic, Jo Kulik, Adam Langley, Jim Roskind, Robbie Shade, Satyam Shekhar, Cherie Shi, Ian Swett, Raman Tenneti, Antonio Vicente, Patrik Westin, Alyssa Wilk, Dale Worley, Dan Zhang, Daniel Ziegler.
Special thanks are due to the following for helping shape QUIC and its deployment: Chris Bentzel, Misha Efimov, Roberto Peon, Alistair Riddoch, Siddharth Vijayakrishnan, and Assar Westerlund. QUIC has also benefited immensely from discussions with folks in private conversations and public ones on the proto-quic@chromium.org mailing list.
QUIC (Quick UDP Internet Connection) is a new multiplexed and secure transport atop UDP, designed from the ground up and optimized for HTTP/2 semantics. While built with HTTP/2 as the primary application protocol, QUIC builds on decades of transport and security experience, and implements mechanisms that make it attractive as a modern general-purpose transport. QUIC provides multiplexing and flow control equivalent to HTTP/2, security equivalent to TLS, and connection semantics, reliability, and congestion control equivalent to TCP.
QUIC operates entirely in userspace, and is currently shipped to users as a part of the Chromium browser, enabling rapid deployment and experimentation. As a userspace transport atop UDP, QUIC allows innovations which have proven difficult to deploy with existing protocols as they are hampered by legacy clients and middleboxes, or by prolonged Operating System development and deployment cycles.
An important goal for QUIC is to inform better transport design through rapid experimentation. As a result, we hope to inform and where possible migrate distilled changes into TCP and TLS, which tend to have much longer iteration cycles.
This document describes the conceptual design and the wire specification of the QUIC protocol. Accompanying documents describe the combined crypto and transport handshake [QUIC-CRYPTO], and loss recovery and congestion control [draft-quic-loss-recovery]. Additional resources, including a more detailed rationale document, are available on the Chromium QUIC webpage.
All integer values used in QUIC, including length, version, and type, are in little-endian byte order, and not in network byte order. QUIC does not enforce alignment of types in dynamically sized frames.
A few terms that are used throughout this document are defined below.
We now briefly describe QUIC's key mechanisms and benefits. QUIC is functionally equivalent to TCP+TLS+HTTP/2, but implemented on top of UDP. Key advantages of QUIC over TCP+TLS+HTTP/2 include:
QUIC combines the crypto and transport handshakes, reducing the number of roundtrips required for setting up a secure connection. QUIC connections are commonly 0-RTT, meaning that on most QUIC connections, data can be sent immediately without waiting for a reply from the server, as compared to the 1-3 roundtrips required for TCP+TLS before application data can be sent.
QUIC provides a dedicated stream (Stream ID 1) to be used for performing the handshake, but the details of this handshake protocol are out of this document's scope. For a complete description of the current handshake protocol, please see the QUIC Crypto Handshake document. QUIC current handshake will be replaced by TLS 1.3 in the future.
QUIC has pluggable congestion control and richer signaling than TCP, which enables QUIC to provide richer information to congestion control algorithms than TCP. Currently, the default congestion control is a reimplementation of TCP Cubic; we are currently experimenting with alternative approaches.
One example of richer information is that each packet, both original and retransmitted, carries a new packet sequence number. This allows a QUIC sender to distinguish ACKs for retransmissions from ACKs for original transmissions, thus avoiding TCP's retransmission ambiguity problem. QUIC ACKs also explicitly carry the delay between the receipt of a packet and its acknowledgment being sent, and together with the monotonically-increasing packet numbers, this allows for precise roundtrip-time (RTT) calculation.
Finally, QUIC's ACK frames support up to 256 NACK ranges, so QUIC is more resilient to reordering than TCP (with SACK), as well as able to keep more bytes on the wire when there is reordering or loss. Both client and server have a more accurate picture of which packets the peer has received.
QUIC implements stream- and connection-level flow control, closely following HTTP/2's flow control. QUIC's stream-level flow control works as follows. A QUIC receiver advertises the absolute byte offset within each stream upto which the receiver is willing to receive data. As data is sent, received, and delivered on a particular stream, the receiver sends WINDOW_UPDATE frames that increase the advertised offset limit for that stream, allowing the peer to send more data on that stream.
In addition to per-stream flow control, QUIC implements connection-level flow control to limit the aggregate buffer that a QUIC receiver is willing to allocate to a connection. Connection flow control works in the same way as stream flow control, but the bytes delivered and highest received offset are all aggregates across all streams.
Similar to TCP's receive-window autotuning, QUIC implements autotuning of flow control credits for both stream and connection flow controllers. QUIC's autotuning increases the size of the credits sent per WINDOW_UPDATE frame if it appears to be limiting the sender's rate, and throttles the sender when the receiving application is slow.
HTTP/2 on TCP suffers from head-of-line blocking in TCP. Since HTTP/2 multiplexes many streams atop TCP's single-bytestream abstraction, a loss of a TCP segment results in blocking of all subsequent segments until a retransmission arrives, irrespective of the HTTP/2 stream that is encapsulated in subsequent segments.
Because QUIC is designed from the ground up for multiplexed operation, lost packets carrying data for an individual stream generally only impact that specific stream. Each stream frame can be immediately dispatched to that stream on arrival, so streams without loss can continue to be reassembled and make forward progress in the application.
Caveat: QUIC currently compresses HTTP headers via HTTP/2 HPACK header compression, which imposes head-of-line blocking for header frames only.
TCP headers appear in plaintext on the wire and not authenticated, causing a plethora of injection and header manipulation issues for TCP, such as receive-window manipulation and sequence-number overwriting. While some of these are active attacks, others are mechanisms used by middleboxes in the network sometimes in an attempt to transparently improve TCP performance. However, even "performance-enhancing" middleboxes still effectively limit the evolvability of the transport protocol, as has been observed in the design of MPTCP and in its subsequent deployability issues.
QUIC packets are always authenticated and typically the payload is fully encrypted. The parts of the packet header which are not encrypted are still authenticated by the receiver, so as to thwart any packet injection or manipulation by third parties. QUIC protects connections from witting or unwitting middlebox manipulation of end-to-end communication.
Caveat: PUBLIC_RESET packets that reset a connection are currently not authenticated.
In order to recover lost packets without waiting for a retransmission, QUIC currently employs a simple XOR-based FEC scheme. An FEC packet contains parity of the packets in the FEC group. If one of the packets in the group is lost, the contents of that packet can be recovered from the FEC packet and the remaining packets in the group. The sender may decide whether to send FEC packets to optimize specific scenarios (e.g., beginning and end of a request).
TCP connections are identified by a 4-tuple of source address, source port, destination address and destination port. A well-known problem with TCP is that connections do not survive IP address changes (for example, by switching from WiFi to cellular) or port number changes (when a client's NAT binding expires causing a change in the port number seen at the server). While MPTCP addresses the connection migration problem for TCP, it is still plagued by lack of middlebox support and lack of OS deployment.
QUIC connections are identified by a 64-bit Connection ID, randomly generated by the client. QUIC can survive IP address changes and NAT re-bindings since the Connection ID remains the same across these migrations. QUIC also provides automatic cryptographic verification of a migrating client, since a migrating client continues to use the same session key for encrypting and decrypting packets.
QUIC has Special Packets and Regular Packets. There are two types of Special Packets: Version Negotiation Packets and Public Reset Packets, and two types of Regular Packets: Frame Packets, and FEC Packets. All QUIC packets should be sized to fit within the path's MTU to avoid IP fragmentation. Path MTU discovery is a work in progress, and the current QUIC implementation uses a 1350-byte maximum QUIC packet size for IPv6, 1370 for IPv4.
All QUIC packets on the wire begin with a public header sized between 2 and 19 bytes. The wire format for the public header is as follows:
0 1 2 3 4 8
+--------+--------+--------+--------+--------+--- ---+
| Public | Connection ID (0, 8, 32, or 64) ... | ->
|Flags(8)| (variable length) |
+--------+--------+--------+--------+--------+--- ---+
9 10 11 12
+--------+--------+--------+--------+
| QUIC Version (32) | ->
| (optional) |
+--------+--------+--------+--------+
13 14 15 16 17 18
+--------+--------+--------+--------+--------+--------+
| Packet Number (8, 16, 32, or 48) |
| (variable length) |
+--------+--------+--------+--------+--------+--------+
The payload may include various type-dependent header bytes as described below.
The fields in the public header are the following:
A Public Flags processing flowchart follows:
Check the public flags in public header
|
|
V
+--------------+
| Public Reset | YES
| flag set? |---------------> Public Reset Packet
+--------------+
|
| NO
V
+------------+ +-------------+
| Version | YES | Packet sent | YES
| flag set? |--------->| by server? |--------> Version Negotiation
+------------+ +-------------+ Packet
| |
| NO | NO
V V
Regular Packet Regular Packet with
QUIC Version present in header
A version negotiation packet is only sent by the server. Version Negotiation packets begin with an 8-bit public flags and 64-bit Connection ID. The public flags must set PUBLIC_FLAG_VERSION and indicate the 64-bit Connection ID. The rest of the Version Negotiation packet is a list of 4-byte versions which the server supports:
0 1 2 3 4 5 6 7 8
+--------+--------+--------+--------+--------+--------+--------+--------+--------+
| Public | Connection ID (64) | ->
|Flags(8)| |
+--------+--------+--------+--------+--------+--------+--------+--------+--------+
9 10 11 12 13 14 15 16 17
+--------+--------+--------+--------+--------+--------+--------+--------+---...--+
| 1st QUIC version supported | 2nd QUIC version supported | ...
| by server (32) | by server (32) |
+--------+--------+--------+--------+--------+--------+--------+--------+---...--+
A Public Reset packet begins with an 8-bit public flags and 64-bit Connection ID. The public flags must set PUBLIC_FLAG_RESET and indicate the 64-bit Connection ID. The rest of the Public Reset packet is encoded as if it were a crypto handshake message of the tag PRST (see [QUIC-CRYPTO]):
0 1 2 3 4 8
+--------+--------+--------+--------+--------+-- --+
| Public | Connection ID (64) ... | ->
|Flags(8)| |
+--------+--------+--------+--------+--------+-- --+
9 10 11 12 13 14
+--------+--------+--------+--------+--------+--------+---
| Quic Tag (32) | Tag value map ... ->
| (PRST) | (variable length)
+--------+--------+--------+--------+--------+--------+---
Tag value map: The tag value map contains the following tag-values:
(TODO: Public Reset packet should include authenticated (destination) server IP/port.)
Regular Packets are authenticated and encrypted. The Public Header is authenticated but not encrypted, and the rest of the packet starting with the Private Flags field is encrypted. Immediately following the Public Header, Regular Packets contain AEAD (authenticated encryption and associated data) data. This data must be decrypted in order for the contents to be interpreted. After decryption, the plaintext starts with a Private Header.
0 1
+--------+--------+
|Private | FEC (8)|
|Flags(8)| (opt) |
+--------+--------+
The fields in the private header are the following:
(TODO: Document the inputs to encryption and decryption and describe trial decryption.)
Beyond the Private Header, Frame Packets have a payload that is a series of type-prefixed frames. The format of frame types is defined later in this document, but the general format of a Frame Packet is as follows:
+--------+---...---+--------+---...---+ | Type | Payload | Type | Payload | +--------+---...---+--------+---...---+
FEC packets (those packets with FLAG_FEC set) have a payload that simply contains an XOR of the null-padded payload of each Data Packet in the FEC group. FEC packets must also have FLAG_FEC_GROUP set.
+-----...----+ | Redundancy | +-----...----+
A QUIC client is the endpoint that initiates a connection. QUIC's connection establishment intertwines version negotiation with the crypto and transport handshakes to reduce connection establishment latency. We first describe version negotiation below.
Each of the initial packets sent from the client to the server must set the version flag, and must specify the version of the protocol being used. Every packet sent by the client must have the version flag on, until it receives a packet from the server with the version flag off. After the server receives the first packet from the client with the version flag off, it must ignore any (possibly delayed) packets with the version flag on.
When the server receives a packet with a Connection ID for a new connection, it will compare the client’s version to the versions it supports. If the client’s version is acceptable to the server, the server will use this protocol version for the lifetime of the connection. In this case, all packets sent by the server will have the version flag off.
If the client’s version is not acceptable to the server, a 1-RTT delay will be incurred. The server will send a Version Negotiation Packet to the client. This packet will have the version flag set and will include the server's set of supported versions.
When the client receives a Version Negotiation Packet from the server, it will select an acceptable protocol version and resend all packets using this version. These packet must continue to have the version flag set and must include the new negotiated protocol version. Eventually, the client receives the first Regular Packet (i.e. not a Version Negotiation Packet) from the server indicating the end of version negotiation, and the client now sends all subsequent packets with the version flag off.
In order to avoid downgrade attacks, the version of the protocol that the client specified in the first packet and the set of versions supported by the server must be included in the crypto handshake data. The client needs to verify that the server’s version list from the handshake matches the list of versions in the Version Negotiation Packet. The server needs to verify that the client’s version from the handshake represents a version of the protocol that it does not actually support.
The rest of the connection establishment is described in the handshake document [QUIC-CRYPTO]. The crypto handshake is performed over the dedicated crypto stream (Stream ID 1).
During connection establishment, the handshake must negotiate various transport parameters. The currently defined transport parameters are described later in the document.
QUIC implements connection reliability, congestion control, and flow control. QUIC flow control closely follows HTTP/2's flow control. QUIC reliability and congestion control are described in an accompanying document. A QUIC connection uses a single packet sequence number space for shared congestion control and loss recovery across the connection.
All data transferred in a QUIC connection, including the crypto handshake, is sent as data inside streams, but the ACKs acknowledge QUIC Packets.
This section conceptually describes the use of streams for data transfer within a QUIC connection. The various frames that are mentioned in this section are described in the section on Frame Types and Formats.
Streams are independent sequences of bi-directional data cut into stream frames. Streams can be created either by the client or the server, can concurrently send data interleaved with other streams, and can be cancelled. QUIC's stream lifetime is modeled closely after HTTP/2's [RFC7540]. (HTTP/2's usage of QUIC streams is described in more detail later in the document.)
Stream creation is done implicitly, by sending a STREAM frame for a given stream. To avoid stream ID collision, the Stream-ID must be even if the server initiates the stream, and odd if the client initiates the stream. 0 is not a valid Stream-ID. Stream 1 is reserved for the crypto handshake, which should be the first client-initiated stream. When using HTTP/2 over QUIC, Stream 3 is reserved for transmitting compressed headers for all other streams, ensuring reliable in-order delivery and processing of headers.
Stream-IDs from each side of the connection must increase monotonically as new streams are created. E.g. Stream 2 may be created after stream 3, but stream 7 must not be created after stream 9. The peer may receive streams out of order. For example, if a server receives packet 10 including frames for stream 9 before it receives packet 9 including frames for stream 7, it should handle this gracefully.
If the endpoint receiving a STREAM frame does not want to accept the stream, it can immediately respond with a RST_STREAM frame (described below). Note, however, that the initiating endpoint may have already sent data on the stream as well; this data must be ignored.
Once a stream is created, it can be used to send and receive data. This means that a series of stream frames can be sent by a QUIC endpoint on a stream until the stream is terminated in that direction.
Either QUIC endpoint can terminate a stream normally. There are three ways that streams can be terminated:
Connections should remain open until they become idle for a pre-negotiated period of time. When a server decides to terminate an idle connection, it should not notify the client to avoid waking up the radio on mobile devices. A QUIC connection, once established, can be terminated in one of two ways:
An endpoint may also send a PUBLIC_RESET packet at any time during the connection to abruptly terminate an active connection. A PUBLIC_RESET is the QUIC equivalent of a TCP RST.
QUIC Frame Packets are populated by frames. which have a Frame Type byte, which itself has a type-dependent interpretation, followed by type-dependent frame header fields. All frames are contained within single QUIC Packets and no frame can span across a QUIC Packet boundary.
There are two interpretations for the Frame Type byte, resulting in two frame types: Special Frame Types, and Regular Frame Types. Special Frame Types encode both a Frame Type and corresponding flags all in the Frame Type byte, while Regular Frame Types use the Frame Type byte simply.
Currently defined Special Frame Types are:
+------------------+-----------------------------+ | Type-field value | Control Frame-type | +------------------+-----------------------------+ | 1fdooossB | STREAM | | 01ntllmmB | ACK | | 001xxxxxB | CONGESTION_FEEDBACK | +------------------+-----------------------------+
Currently defined Regular Frame Types are:
+------------------+-----------------------------+ | Type-field value | Control Frame-type | +------------------+-----------------------------+ | 00000000B (0x00) | PADDING | | 00000001B (0x01) | RST_STREAM | | 00000010B (0x02) | CONNECTION_CLOSE | | 00000011B (0x03) | GOAWAY | | 00000100B (0x04) | WINDOW_UPDATE | | 00000101B (0x05) | BLOCKED | | 00000110B (0x06) | STOP_WAITING | | 00000111B (0x07) | PING | +------------------+-----------------------------+
The STREAM frame is used to both implicitly create a stream and to send data on it, and is as follows:
0 1 … SLEN
+--------+--------+--------+--------+--------+
|Type (8)| Stream ID (8, 16, 24, or 32 bits) |
| | (Variable length SLEN bytes) |
+--------+--------+--------+--------+--------+
SLEN+1 SLEN+2 … SLEN+OLEN
+--------+--------+--------+--------+--------+--------+--------+--------+
| Offset (0, 16, 24, 32, 40, 48, 56, or 64 bits) (variable length) |
| (Variable length: OLEN bytes) |
+--------+--------+--------+--------+--------+--------+--------+--------+
SLEN+OLEN+1 SLEN+OLEN+2
+-------------+-------------+
| Data length (0 or 16 bits)|
| Optional(maybe 0 bytes) |
+------------+--------------+
The fields in the STREAM frame header are as follows:
A stream frame must always have either non-zero data length or the FIN bit set.
The ACK frame is sent to inform the peer which packets have been received, as well as which packets are still considered missing by the receiver (the contents of missing packets may need to be resent). The design of QUIC's ACK frame is different from TCP's and SCTP's SACK representations in that QUIC ACKs indicate the largest packet number observed thus far followed by a list of missing packet, or NACK, ranges indicating gaps in packets received below this packet number. To limit the NACK ranges to the ones that haven't yet been communicated to the peer, the peer periodically sends STOP_WAITING frames that signal the receiver to stop waiting for packets below a specified sequence number, raising the "least unacked" packet number at the receiver. A sender of an ACK frame thus reports only those NACK ranges between the received least unacked and the reported largest observed packet numbers. The frame is as follows:
0 1 N
+--------+--------+---------------------------------------------------+
| Type |Received| Largest Observed |
| (8) |Entropy | (8, 16, 32, or 48 bits) |
+--------+--------+---------------------------------------------------+
N+1 N+2 N+3 N+4 N+8
+--------+--------+---------+--------+--------------------------------+
| Ack Delay | Num | Delta | First Timestamp |
| Time (16) |Timestamp|Largest | (32 bits) |
| | (8) |Observed| |
+--------+--------+---------+--------+--------------------------------+
N+9 N+11 - X
+--------+-------------------+
| Delta | Time Since |
|Largest | Previous Timestamp| <-- Repeat (NumTimestamp - 1) times
|Observed| (16 bits) |
+--------+-------------------+
X X+1 - Y Y+1
+--------+-------------------------------------------------+--------+
| Number | Missing Packet Sequence Number Delta | Range |
| Ranges | (8, 16, 32, or 48 bits) | Length |
| (opt) | (repeats Number Ranges times) |(Repeat)|
+--------+-------------------------------------------------+--------+
Y+2 Y+3 - Z
+--------+-----------------------------------------------------+
| Number | Revived Packet Number |
| Revived| (8, 16, 32, or 48 bits, same as Largest Observed) |
| (opt) | (repeats Number Revived times) |
+--------+-----------------------------------------------------+
The fields in the ACK frame are as follows:
The entropy bits for a subset of packets (known to a receiver or sender) are accumulated into an 8 bit unsigned value, and similarly presented in both a STOP_WAITING frame and an ACK frame. If we defined E(k) to be the FLAG_ENTROPY bit present in packet number k, then the k'th packet's contribution C(k) is defined to be E(k) left shifted by k mod 8 bits. The accumulated entropy is then the bitwise-XOR sum of the contributions C(k), for all packets in the desired subset.
The STOP_WAITING frame is sent to inform the peer that it should not continue to wait for packets with packet numbers lower than a specified value. The packet number is encoded in 1, 2, 4 or 6 bytes, using the same coding length as is specified for the packet number for the enclosing packet's header (specified in the QUIC Frame Packet's Public Flags field.) The frame is as follows:
0 1 2 3 4 5 6 7
+--------+--------+--------+--------+--------+--------+-------+-------+
|Type (8)|Sent | Least unacked delta (8, 16, 32, or 48 bits) |
| |Entropy | (variable length) |
+--------+--------+--------+--------+--------+--------+-------+-------+
The fields in the STOP_WAITING frame are as follows:
The WINDOW_UPDATE frame is used to inform the peer of an increase in an endpoint's flow control receive window. The stream ID can be 0, indicating this WINDOW_UPDATE applies to the connection level flow control window, or > 0 indicating that the specified stream should increase its flow control window. The frame is as follows:
An absolute byte offset is specified, and the receiver of a WINDOW_UPDATE frame may only send up to that number of bytes on the specified stream. Violating flow control by sending further bytes will result in the receiving endpoint closing the connection.
On receipt of multiple WINDOW_UPDATE frames for a specific stream ID, it is only necessary to keep track of the maximum byte offset.
Both stream and session windows start with a default value of 16 KB, but this is typically increased during the handshake. To do this, an endpoint should negotiate the SFCW (Stream Flow Control Window) and CFCW (Connection/Session Flow Control Window) parameters in the handshake. The value associated with each tag should be the number of bytes for initial stream window and initial connection window respectively.
The frame is as follows:
0 1 4 5 12
+--------+--------+-- ... --+-------+--------+-- ... --+-------+
|Type(8) | Stream ID (32 bits) | Byte offset (64 bits) |
+--------+--------+-- ... --+-------+--------+-- ... --+-------+
The fields in the WINDOW_UPDATE frame are as follows:
The BLOCKED frame is used to indicate to the remote endpoint that this endpoint is ready to send data (and has data to send), but is currently flow control blocked. This is a purely informational frame, which is extremely useful for debugging purposes. A receiver of a BLOCKED frame should simply discard it (after possibly printing a helpful log message). The frame is as follows:
0 1 2 3 4
+--------+--------+--------+--------+--------+
|Type(8) | Stream ID (32 bits) |
+--------+--------+--------+--------+--------+
The fields in the BLOCKED frame are as follows:
The CONGESTION_FEEDBACK frame is an experimental frame currently not used. It is intended to provide extra congestion feedback information outside the scope of the standard ack frame. A CONGESTION_FEEDBACK frame must have the first three bits of the Frame Type set to 001. The last 5 bits of the Frame Type field are reserved for future use.
The PADDING frame pads a packet with 0x00 bytes. When this frame is encountered, the rest of the packet is expected to be padding bytes. The frame contains 0x00 bytes and extends to the end of the QUIC packet. A PADDING frame only has a Frame Type field, and must have the 8-bit Frame Type field set to 0x00.
The RST_STREAM frame allows for abnormal termination of a stream. When sent by the creator of a stream, it indicates the creator wishes to cancel the stream. When sent by the receiver of a stream, it indicates an error or that the receiver did not want to accept the stream, so the stream should be closed. The frame is as follows:
0 1 4 5 12 8 16
+-------+--------+-- ... ----+--------+-- ... ------+-------+-- ... ------+
|Type(8)| StreamID (32 bits) | Byte offset (64 bits)| Error code (32 bits)|
+-------+--------+-- ... ----+--------+-- ... ------+-------+-- ... ------+
The fields in a RST_STREAM frame are as follows:
The PING frame can be used by an endpoint to verify that a peer is still alive. The PING frame contains no payload. The receiver of a PING frame simply needs to ACK the packet containing this frame. The PING frame should be used to keep a connection alive when a stream is open. The default is to do this after 15 seconds of quiescence, which is much shorter than most NATs time out. A PING frame only has a Frame Type field, and must have the 8-bit Frame Type field set to 0x07.
The CONNECTION_CLOSE frame allows for notification that the connection is being closed. If there are streams in flight, those streams are all implicitly closed when the connection is closed. (Ideally, a GOAWAY frame would be sent with enough time that all streams are torn down.) The frame is as follows:
0 1 4 5 6 7
+--------+--------+-- ... -----+--------+--------+--------+----- ...
|Type(8) | Error code (32 bits)| Reason phrase | Reason phrase
| | | length (16 bits)|(variable length)
+--------+--------+-- ... -----+--------+--------+--------+----- ...
The fields of a CONNECTION_CLOSE frame are as follows:
The GOAWAY frame allows for notification that the connection should stop being used, and will likely be aborted in the future. Any active streams will continue to be processed, but the sender of the GOAWAY will not initiate any additional streams, and will not accept any new streams. The frame is as follows:
0 1 4 5 6 7 8
+--------+--------+-- ... -----+-------+-------+-------+------+
|Type(8) | Error code (32 bits)| Last Good Stream ID (32 bits)| ->
+--------+--------+-- ... -----+-------+-------+-------+------+
9 10 11
+--------+--------+--------+----- ...
| Reason phrase | Reason phrase
| length (16 bits)|(variable length)
+--------+--------+--------+----- ...
The fields of a GOAWAY frame are as follows:
The handshake is responsible for negotiating a variety of transport parameters for a QUIC connection.
The number to code mappings for QuicErrorCodes are currently defined in the Chromium source code in src/net/quic/quic_protocol.h. (TODO: hardcode numbers and add them here)
(TODO: implement)
QUIC will use the HTTP/2 prioritization mechanism. Roughly, a stream may be dependent on another stream. In this situation, the "parent" stream should effectively starve the "child" stream. In addition, parent streams have an explicit priority. Parent streams should not starve other parent streams, but should make progress proportional to their relative priority.
Since QUIC integrates various HTTP/2 mechanisms with transport mechanisms, QUIC implements a number of features that are also specified in HTTP/2. As a result, QUIC allows HTTP/2 mechanisms to be replaced by QUIC's implementation, reducing complexity in the HTTP/2 protocol. This section briefly describes how HTTP/2 semantics can be offered over a QUIC implementation.
When HTTP/2 headers and data are sent over QUIC, the QUIC layer handles most of the stream management. HTTP/2 Stream IDs are replaced by QUIC Stream IDs. HTTP/2 does not need to do any explicit stream framing when using QUIC---data sent over a QUIC stream simply consists of HTTP/2 headers or body. Requests and responses are considered complete when the QUIC stream is closed in the corresponding direction.
Stream flow control is handled by QUIC, and does not need to be re-implemented in HTTP/2. QUIC's flow controller replaces the two levels of poorly matched flow controllers in current HTTP/2 deployments---one at the HTTP/2 level, and the other at the TCP level.
QUIC implements HPACK header compression for HTTP/2, which unfortunately introduces some Head-of-Line blocking since HTTP/2 header blocks must be decompressed in the order they were compressed.
Since streams may be processed in arbitrary order at a receiver, strict ordering across headers is enforced by sending all headers on a dedicated headers stream, with Stream ID 3. An HTTP/2 receiver using QUIC would thus process data from a stream only after receiving the corresponding header on the headers stream.
Future work will tweak the compressor and decompressor in QUIC so that the compressed output does not depend on unacked previous compressed state. This could be done, perhaps, by creating "checkpoints" of HPACK state which are updated when headers have been acked. When compressing headers QUIC would only compress relative to the previous "checkpoint".
Bytes sent on the dedicated headers stream are simply HTTP/2 HEADERS frames. The exact layout of these frames is described in RFC 7540.
Unlike when using TCP, the underlying connection for QUIC is guaranteed to be persistent. The HTTP "Connection" header is therefore does not apply. For best performance, it is expected that clients will not close a QUIC connection until the user navigates away from all web pages using that connection, or until the server closes the connection.
The Alternate-Protocol header is used to negotiate use of QUIC on future HTTP requests. To specify QUIC as an alternate protocol available on port 123, a server uses:
"Alternate-Protocol: 123:quic"
When a client receives a Alternate-Protocol header advertising QUIC, it can then attempt to use QUIC for future secure connections on that domain. Since middleboxes and/or firewalls can block QUIC and/or UDP communication, a client should implement a graceful fallback to TCP when QUIC reachability is broken.
Note that the server may reply with multiple field values or a comma-separated field value for Alternate-Protocol to indicate the various transports it supports.
A server can also send a header to notify that QUIC should not be used on this domain. If it sends the alternate-protocol-required header, the client should remember to not use QUIC on that domain in future, and not do any UDP probing to see if QUIC is available.
QUIC provides a dedicated stream (Stream ID 1) to be used for performing a combined connection and security handshake, but the details of this handshake protocol are out of this document's scope. However, QUIC does impose a number of requirements on any such handshake protocol. The following list of requirements documents properties of the current prototype handshake which should be provided by any future handshake protocol.
The QUIC handshake protocol manages to successfully achieve 0-RTT for most connections, and is critical to QUIC's latency improvements.
TCP verifies the client’s address by burning a round trip on the SYN, SYN_ACK exchange. QUIC uses a source address token delivered by the server in a previous connection.
QUIC servers store a number of pieces of data in the source address token, for use on a subsequent connection from the same client. This includes recently used source addresses, measured bandwidth to the client, and server-designated connection IDs (for Stateless REJs). An alternative handshake protocol’s analog of a source address token needs to be (i) opaque at the client, and (ii) large enough to permit these bits of information to be stored. Alternatively, the handshake protocol should have a different method to store this information at the client.
In addition to negotiating crypto parameters, the QUIC handshake also negotiates QUIC and HTTP/2 level parameters, including max open QUIC streams and other QUIC connection options.
The QUIC handshake compresses certificates so that an REJ, including the common Google certificate chain, is able to fit into two 1350 byte packets. This helps to reduce the amplification attack footprint of QUIC without reducing 0-RTT rate.
QUIC uses a Server Config Update (SCUP) message to refresh the source-address token (STK) and server config mid-connection, extending the period over which 0-RTT connections can be established by the client.
.
| [RFC2119] | Bradner, S., "Key Words for use in RFCs to Indicate Requirement Levels", March 1997. |
| [RFC7540] | Belshe, M., Peon, R. and M. Thomson, "Hypertext Transfer Protocol Version 2 (HTTP/2)", May 2015. |
| [QUIC-CRYPTO] | Langley, A. and W. Chang, "QUIC Crypto", June 2015. |
| [QUIC-CC] | Iyengar, J. and I. Swett, "QUIC Loss Recovery and Congestion Control", December 2015. |