Netflix

ckrasic@netflix.com

Akamai Technologies

mbishop@evequefou.be

Facebook

afrind@fb.com

Transport QUIC This specification defines QPACK, a compression format for efficiently representing HTTP header fields, to be used in HTTP/3. This is a variation of HPACK header compression that seeks to reduce head-of-line blocking. Discussion of this draft takes place on the QUIC working group mailing list (quic@ietf.org), which is archived at https://mailarchive.ietf.org/arch/search/?email_list=quic. Working Group information can be found at https://github.com/quicwg; source code and issues list for this draft can be found at https://github.com/quicwg/base-drafts/labels/-qpack.

The QUIC transport protocol is designed to support HTTP semantics, and its design subsumes many of the features of HTTP/2 . HTTP/2 uses HPACK for header compression. If HPACK were used for HTTP/3 , it would induce head-of-line blocking due to built-in assumptions of a total ordering across frames on all streams. QPACK reuses core concepts from HPACK, but is redesigned to allow correctness in the presence of out-of-order delivery, with flexibility for implementations to balance between resilience against head-of-line blocking and optimal compression ratio. The design goals are to closely approach the compression ratio of HPACK with substantially less head-of-line blocking under the same loss conditions.

The key words “MUST”, “MUST NOT”, “REQUIRED”, “SHALL”, “SHALL NOT”, “SHOULD”, “SHOULD NOT”, “RECOMMENDED”, “NOT RECOMMENDED”, “MAY”, and “OPTIONAL” in this document are to be interpreted as described in BCP 14 when, and only when, they appear in all capitals, as shown here. Definitions of terms that are used in this document: A name-value pair sent as part of an HTTP message. An ordered collection of header fields associated with an HTTP message. A header list can contain multiple header fields with the same name. It can also contain duplicate header fields. The compressed representation of a header list. An implementation which transforms a header list into a header block. An implementation which transforms a header block into a header list. A unique index for each entry in the dynamic table. A reference point for relative and post-base indices. References to dynamic table entries in header blocks are relative to a Base. The total number of entries inserted in the dynamic table. QPACK is a name, not an acronym.

Diagrams use the format described in , with the following additional conventions: Indicates that x is A bits long Indicates that x uses the prefixed integer encoding defined in , beginning with an A-bit prefix. Indicates that x is variable-length and extends to the end of the region.

Like HPACK, QPACK uses two tables for associating header fields to indices. The static table (see ) is predefined and contains common header fields (some of them with an empty value). The dynamic table (see ) is built up over the course of the connection and can be used by the encoder to index header fields in the encoded header lists. QPACK defines unidirectional streams for sending instructions from encoder to decoder and vice versa.

An encoder converts a header list into a header block by emitting either an indexed or a literal representation for each header field in the list (see ). Indexed representations achieve high compression by replacing the literal name and possibly the value with an index to either the static or dynamic table. References to the static table and literal representations do not require any dynamic state and never risk head-of-line blocking. References to the dynamic table risk head-of-line blocking if the encoder has not received an acknowledgement indicating the entry is available at the decoder. An encoder MAY insert any entry in the dynamic table it chooses; it is not limited to header fields it is compressing. QPACK preserves the ordering of header fields within each header list. An encoder MUST emit header field representations in the order they appear in the input header list. QPACK is designed to contain the more complex state tracking to the encoder, while the decoder is relatively simple.

An encoder MUST ensure that a header block which references a dynamic table entry is not processed by the decoder after the referenced entry has been evicted. Hence the encoder needs to retain information about each compressed header block that references the dynamic table until that header block is acknowledged by the decoder (see ).

A dynamic table entry is considered blocking and cannot be evicted until its insertion has been acknowledged and there are no outstanding unacknowledged references to the entry. In particular, a dynamic table entry that has never been referenced can still be blocking. An encoder MUST NOT insert an entry into the dynamic table (or duplicate an existing entry) if doing so would evict a blocking entry. In order to avoid this, an encoder that uses the dynamic table has to keep track of blocking entries. A blocking entry is unrelated to a blocked stream, see .

To ensure that the encoder is not prevented from adding new entries, the encoder can avoid referencing entries that are close to eviction. Rather than reference such an entry, the encoder can emit a Duplicate instruction (see ), and reference the duplicate instead. Determining which entries are too close to eviction to reference is an encoder preference. One heuristic is to target a fixed amount of available space in the dynamic table: either unused space or space that can be reclaimed by evicting non-blocking entries. To achieve this, the encoder can maintain a draining index, which is the smallest absolute index (see ) in the dynamic table that it will emit a reference for. As new entries are inserted, the encoder increases the draining index to maintain the section of the table that it will not reference. If the encoder does not create new references to entries with an absolute index lower than the draining index, the number of unacknowledged references to those entries will eventually become zero, allowing them to be evicted.

+----------+---------------------------------+--------+ | Draining | Referenceable | Unused | | Entries | Entries | Space | +----------+---------------------------------+--------+ ^ ^ ^ | | | Dropping Draining Index Insertion Point Point

Because QUIC does not guarantee order between data on different streams, a decoder might encounter a header block that references a dynamic table entry that it has not yet received. Each header block contains a Required Insert Count (see ), the lowest possible value for the Insert Count with which the header block can be decoded. For a header block with references to the dynamic table, the Required Insert Count is one larger than the largest absolute index of all referenced dynamic table entries. For a header block with no references to the dynamic table, the Required Insert Count is zero. When the decoder receives a header block with a Required Insert Count greater than its own Insert Count, the stream cannot be processed immediately, and is considered “blocked” (see ). The decoder specifies an upper bound on the number of streams which can be blocked using the SETTINGS_QPACK_BLOCKED_STREAMS setting (see ). An encoder MUST limit the number of streams which could become blocked to the value of SETTINGS_QPACK_BLOCKED_STREAMS at all times. If a decoder encounters more blocked streams than it promised to support, it MUST treat this as a connection error of type HTTP_QPACK_DECOMPRESSION_FAILED. Note that the decoder might not become blocked on every stream which risks becoming blocked. An encoder can decide whether to risk having a stream become blocked. If permitted by the value of SETTINGS_QPACK_BLOCKED_STREAMS, compression efficiency can often be improved by referencing dynamic table entries that are still in transit, but if there is loss or reordering the stream can become blocked at the decoder. An encoder can avoid the risk of blocking by only referencing dynamic table entries which have been acknowledged, but this could mean using literals. Since literals make the header block larger, this can result in the encoder becoming blocked on congestion or flow control limits.

The Known Received Count is the total number of dynamic table insertions and duplications acknowledged by the decoder. The encoder tracks the Known Received Count in order to identify which dynamic table entries can be referenced without potentially blocking a stream. The decoder tracks the Known Received Count in order to be able to send Insert Count Increment instructions (see ). If a header block was potentially blocking, the encoder infers from receiving a Header Acknowledgement instruction () that the decoder has received all dynamic table state necessary to process that header block. If the Required Insert Count of the acknowledged header block is greater than the current Known Received Count, the encoder updates the Known Received Count to the value of the Required Insert Count of the acknowledged header block. To acknowledge dynamic table entries which are not referenced by header blocks, for example because the encoder or the decoder have chosen not to risk blocked streams, the decoder sends an Insert Count Increment instruction (see ).

As in HPACK, the decoder processes header blocks and emits the corresponding header lists. It also processes instructions received on the encoder stream that modify the dynamic table. Note that header blocks and encoder stream instructions arrive on separate streams. This is unlike HPACK, where header blocks can contain instructions that modify the dynamic table, and there is no dedicated stream of HPACK instructions. The decoder MUST emit header fields in the order their representations appear in the input header block.

Upon receipt of a header block, the decoder examines the Required Insert Count. When the Required Insert Count is less than or equal to the decoder’s Insert Count, the header block can be processed immediately. Otherwise, the stream on which the header block was received becomes blocked. While blocked, header block data SHOULD remain in the blocked stream’s flow control window. A stream becomes unblocked when the Insert Count becomes greater than or equal to the Required Insert Count for all header blocks the decoder has started reading from the stream. When processing header blocks, the decoder expects the Required Insert Count to exactly match the value defined in . If it encounters a smaller value than expected, it MUST treat this as a connection error of type HTTP_QPACK_DECOMPRESSION_FAILED (see ). If it encounters a larger value than expected, it MAY treat this as a connection error of type HTTP_QPACK_DECOMPRESSION_FAILED.

The decoder signals the following events by emitting decoder instructions () on the decoder stream.

After the decoder finishes decoding a header block containing dynamic table references, it MUST emit a Header Acknowledgement instruction (). A stream may carry multiple header blocks in the case of intermediate responses, trailers, and pushed requests. The encoder interprets each Header Acknowledgement instruction as acknowledging the earliest unacknowledged header block containing dynamic table references sent on the given stream.

When an endpoint receives a stream reset before the end of a stream or before all header blocks are processed on that stream, or when it abandons reading of a stream, it generates a Stream Cancellation instruction (see ). This signals to the encoder that all references to the dynamic table on that stream are no longer outstanding. A decoder with a maximum dynamic table capacity equal to zero (see ) MAY omit sending Stream Cancellations, because the encoder cannot have any dynamic table references. An encoder cannot infer from this instruction that any updates to the dynamic table have been received. The Header Acknowledgement and Stream Cancellation instructions permit the encoder to remove references to entries in the dynamic table. When an entry with absolute index lower than the Known Received Count has zero references, then it is no longer considered blocking (see ).

After receiving new table entries on the encoder stream, the decoder chooses when to emit Insert Count Increment instructions (see ). Emitting this instruction after adding each new dynamic table entry will provide the timeliest feedback to the encoder, but could be redundant with other decoder feedback. By delaying an Insert Count Increment instruction, the decoder might be able to coalesce multiple Insert Count Increment instructions, or replace them entirely with Header Acknowledgements (see ). However, delaying too long may lead to compression inefficiencies if the encoder waits for an entry to be acknowledged before using it.

If the decoder encounters a reference in a header block representation to a dynamic table entry which has already been evicted or which has an absolute index greater than or equal to the declared Required Insert Count (see ), it MUST treat this as a connection error of type HTTP_QPACK_DECOMPRESSION_FAILED. If the decoder encounters a reference in an encoder instruction to a dynamic table entry which has already been evicted, it MUST treat this as a connection error of type HTTP_QPACK_ENCODER_STREAM_ERROR.

Unlike in HPACK, entries in the QPACK static and dynamic tables are addressed separately. The following sections describe how entries in each table are addressed.

The static table consists of a predefined static list of header fields, each of which has a fixed index over time. Its entries are defined in . All entries in the static table have a name and a value. However, values can be empty (that is, have a length of 0). Each entry is identified by a unique index. Note that the QPACK static table is indexed from 0, whereas the HPACK static table is indexed from 1. When the decoder encounters an invalid static table index in a header block representation it MUST treat this as a connection error of type HTTP_QPACK_DECOMPRESSION_FAILED. If this index is received on the encoder stream, this MUST be treated as a connection error of type HTTP_QPACK_ENCODER_STREAM_ERROR.

The dynamic table consists of a list of header fields maintained in first-in, first-out order. Each HTTP/3 endpoint holds a dynamic table that is initially empty. Entries are added by encoder instructions received on the encoder stream (see ). The dynamic table can contain duplicate entries (i.e., entries with the same name and same value). Therefore, duplicate entries MUST NOT be treated as an error by the decoder. Dynamic table entries can have empty values.

The size of the dynamic table is the sum of the size of its entries. The size of an entry is the sum of its name’s length in bytes, its value’s length in bytes, and 32. The size of an entry is calculated using the length of its name and value without Huffman encoding applied.

The encoder sets the capacity of the dynamic table, which serves as the upper limit on its size. The initial capacity of the dynamic table is zero. The encoder sends a Set Dynamic Table Capacity instruction () with a non-zero capacity to begin using the dynamic table. Before a new entry is added to the dynamic table, entries are evicted from the end of the dynamic table until the size of the dynamic table is less than or equal to (table capacity - size of new entry). The encoder MUST NOT cause a blocking dynamic table entry to be evicted (see ). The new entry is then added to the table. It is an error if the encoder attempts to add an entry that is larger than the dynamic table capacity; the decoder MUST treat this as a connection error of type HTTP_QPACK_ENCODER_STREAM_ERROR. A new entry can reference an entry in the dynamic table that will be evicted when adding this new entry into the dynamic table. Implementations are cautioned to avoid deleting the referenced name or value if the referenced entry is evicted from the dynamic table prior to inserting the new entry. Whenever the dynamic table capacity is reduced by the encoder (see ), entries are evicted from the end of the dynamic table until the size of the dynamic table is less than or equal to the new table capacity. This mechanism can be used to completely clear entries from the dynamic table by setting a capacity of 0, which can subsequently be restored.

To bound the memory requirements of the decoder, the decoder limits the maximum value the encoder is permitted to set for the dynamic table capacity. In HTTP/3, this limit is determined by the value of SETTINGS_QPACK_MAX_TABLE_CAPACITY sent by the decoder (see ). The encoder MUST not set a dynamic table capacity that exceeds this maximum, but it can choose to use a lower dynamic table capacity (see ). For clients using 0-RTT data in HTTP/3, the server’s maximum table capacity is the remembered value of the setting, or zero if the value was not previously sent. When the client’s 0-RTT value of the SETTING is zero, the server MAY set it to a non-zero value in its SETTINGS frame. If the remembered value is non-zero, the server MUST send the same non-zero value in its SETTINGS frame. If it specifies any other value, or omits SETTINGS_QPACK_MAX_TABLE_CAPACITY from SETTINGS, the encoder must treat this as a connection error of type HTTP_QPACK_DECODER_STREAM_ERROR. For HTTP/3 servers and HTTP/3 clients when 0-RTT is not attempted or is rejected, the maximum table capacity is 0 until the encoder processes a SETTINGS frame with a non-zero value of SETTINGS_QPACK_MAX_TABLE_CAPACITY. When the maximum table capacity is zero, the encoder MUST NOT insert entries into the dynamic table, and MUST NOT send any encoder instructions on the encoder stream.

Each entry possesses an absolute index which is fixed for the lifetime of that entry. The first entry inserted has an absolute index of “0”; indices increase by one with each insertion.

Relative indices begin at zero and increase in the opposite direction from the absolute index. Determining which entry has a relative index of “0” depends on the context of the reference. In encoder instructions (see ), a relative index of “0” refers to the most recently inserted value in the dynamic table. Note that this means the entry referenced by a given relative index will change while interpreting instructions on the encoder stream.

+-----+---------------+-------+ | n-1 | ... | d | Absolute Index + - - +---------------+ - - - + | 0 | ... | n-d-1 | Relative Index +-----+---------------+-------+ ^ | | V Insertion Point Dropping Point n = count of entries inserted d = count of entries dropped

Unlike in encoder instructions, relative indices in header block representations are relative to the Base at the beginning of the header block (see ). This ensures that references are stable even if header blocks and dynamic table updates are processed out of order. In a header block a relative index of “0” refers to the entry with absolute index equal to Base - 1.

Base | V +-----+-----+-----+-----+-------+ | n-1 | n-2 | n-3 | ... | d | Absolute Index +-----+-----+ - +-----+ - + | 0 | ... | n-d-3 | Relative Index +-----+-----+-------+ n = count of entries inserted d = count of entries dropped In this example, Base = n - 2

Post-Base indices are used in header block instructions for entries with absolute indices greater than or equal to Base, starting at 0 for the entry with absolute index equal to Base, and increasing in the same direction as the absolute index. Post-Base indices allow an encoder to process a header block in a single pass and include references to entries added while processing this (or other) header blocks.

Base | V +-----+-----+-----+-----+-----+ | n-1 | n-2 | n-3 | ... | d | Absolute Index +-----+-----+-----+-----+-----+ | 1 | 0 | Post-Base Index +-----+-----+ n = count of entries inserted d = count of entries dropped In this example, Base = n - 2

The prefixed integer from is used heavily throughout this document. The format from is used unmodified. Note, however, that QPACK uses some prefix sizes not actually used in HPACK. QPACK implementations MUST be able to decode integers up to and including 62 bits long.

The string literal defined by is also used throughout. This string format includes optional Huffman encoding. HPACK defines string literals to begin on a byte boundary. They begin with a single bit flag, denoted as ‘H’ in this document (indicating whether the string is Huffman-coded), followed by the Length encoded as a 7-bit prefix integer, and finally Length bytes of data. When Huffman encoding is enabled, the Huffman table from is used without modification. This document expands the definition of string literals and permits them to begin other than on a byte boundary. An “N-bit prefix string literal” begins with the same Huffman flag, followed by the length encoded as an (N-1)-bit prefix integer. The prefix size, N, can have a value between 2 and 8 inclusive. The remainder of the string literal is unmodified. A string literal without a prefix length noted is an 8-bit prefix string literal and follows the definitions in without modification.

QPACK defines two unidirectional stream types: An encoder stream is a unidirectional stream of type 0x02. It carries an unframed sequence of encoder instructions from encoder to decoder. A decoder stream is a unidirectional stream of type 0x03. It carries an unframed sequence of decoder instructions from decoder to encoder. HTTP/3 endpoints contain a QPACK encoder and decoder. Each endpoint MUST initiate at most one encoder stream and at most one decoder stream. Receipt of a second instance of either stream type MUST be treated as a connection error of type HTTP_STREAM_CREATION_ERROR. These streams MUST NOT be closed. Closure of either unidirectional stream type MUST be treated as a connection error of type HTTP_CLOSED_CRITICAL_STREAM. An endpoint MAY avoid creating an encoder stream if it’s not going to be used (for example if its encoder doesn’t wish to use the dynamic table, or if the maximum size of the dynamic table permitted by the peer is zero). An endpoint MAY avoid creating a decoder stream if its decoder sets the maximum capacity of the dynamic table to zero. An endpoint MUST allow its peer to create an encoder stream and a decoder stream even if the connection’s settings prevent their use.

An encoder sends encoder instructions on the encoder stream to set the capacity of the dynamic table and add dynamic table entries. Instructions adding table entries can use existing entries to avoid transmitting redundant information. The name can be transmitted as a reference to an existing entry in the static or the dynamic table or as a string literal. For entries which already exist in the dynamic table, the full entry can also be used by reference, creating a duplicate entry. This section specifies the following encoder instructions.

An encoder informs the decoder of a change to the dynamic table capacity using an instruction which begins with the ‘001’ three-bit pattern. This is followed by the new dynamic table capacity represented as an integer with a 5-bit prefix (see ).

0 1 2 3 4 5 6 7 +---+---+---+---+---+---+---+---+ | 0 | 0 | 1 | Capacity (5+) | +---+---+---+-------------------+

The new capacity MUST be lower than or equal to the limit described in . In HTTP/3, this limit is the value of the SETTINGS_QPACK_MAX_TABLE_CAPACITY parameter (see ) received from the decoder. The decoder MUST treat a new dynamic table capacity value that exceeds this limit as a connection error of type HTTP_QPACK_ENCODER_STREAM_ERROR. Reducing the dynamic table capacity can cause entries to be evicted (see ). This MUST NOT cause the eviction of blocking entries (see ). Changing the capacity of the dynamic table is not acknowledged as this instruction does not insert an entry.

An encoder adds an entry to the dynamic table where the header field name matches the header field name of an entry stored in the static or the dynamic table using an instruction that starts with the ‘1’ one-bit pattern. The second (‘T’) bit indicates whether the reference is to the static or dynamic table. The 6-bit prefix integer (see ) that follows is used to locate the table entry for the header name. When T=1, the number represents the static table index; when T=0, the number is the relative index of the entry in the dynamic table. The header name reference is followed by the header field value represented as a string literal (see ).

0 1 2 3 4 5 6 7 +---+---+---+---+---+---+---+---+ | 1 | T | Name Index (6+) | +---+---+-----------------------+ | H | Value Length (7+) | +---+---------------------------+ | Value String (Length bytes) | +-------------------------------+

An encoder adds an entry to the dynamic table where both the header field name and the header field value are represented as string literals using an instruction that starts with the ‘01’ two-bit pattern. This is followed by the name represented as a 6-bit prefix string literal, and the value represented as an 8-bit prefix string literal (see ).

0 1 2 3 4 5 6 7 +---+---+---+---+---+---+---+---+ | 0 | 1 | H | Name Length (5+) | +---+---+---+-------------------+ | Name String (Length bytes) | +---+---------------------------+ | H | Value Length (7+) | +---+---------------------------+ | Value String (Length bytes) | +-------------------------------+

An encoder duplicates an existing entry in the dynamic table using an instruction that begins with the ‘000’ three-bit pattern. This is followed by the relative index of the existing entry represented as an integer with a 5-bit prefix (see .

0 1 2 3 4 5 6 7 +---+---+---+---+---+---+---+---+ | 0 | 0 | 0 | Index (5+) | +---+---+---+-------------------+

The existing entry is re-inserted into the dynamic table without resending either the name or the value. This is useful to avoid adding a reference to an older entry, which might block inserting new entries.

A decoder sends decoder instructions on the decoder stream to inform the encoder about the processing of header blocks and table updates to ensure consistency of the dynamic table. This section specifies the following decoder instructions.

After processing a header block whose declared Required Insert Count is not zero, the decoder emits a Header Acknowledgement instruction. The instruction begins with the ‘1’ one-bit pattern which is followed by the header block’s associated stream ID encoded as a 7-bit prefix integer (see ). This instruction is used as described in and in .

0 1 2 3 4 5 6 7 +---+---+---+---+---+---+---+---+ | 1 | Stream ID (7+) | +---+---------------------------+

If an encoder receives a Header Acknowledgement instruction referring to a stream on which every header block with a non-zero Required Insert Count has already been acknowledged, that MUST be treated as a connection error of type HTTP_QPACK_DECODER_STREAM_ERROR.

When a stream is reset or reading is abandoned, the decoder emits a Stream Cancellation instruction. The instruction begins with the ‘01’ two-bit pattern, which is followed by the stream ID of the affected stream encoded as a 6-bit prefix integer. This instruction is used as described in .

0 1 2 3 4 5 6 7 +---+---+---+---+---+---+---+---+ | 0 | 1 | Stream ID (6+) | +---+---+-----------------------+

The Insert Count Increment instruction begins with the ‘00’ two-bit pattern, followed by the Increment encoded as a 6-bit prefix integer. The value of the Increment is the total number of dynamic table insertions and duplications processed by the decoder since the last time it sent a Header Acknowledgement instruction that increased the Known Received Count (see ) or an Insert Count Increment instruction. The encoder uses this value to update the Known Received Count, as described in .

0 1 2 3 4 5 6 7 +---+---+---+---+---+---+---+---+ | 0 | 0 | Increment (6+) | +---+---+-----------------------+

An encoder that receives an Increment field equal to zero, or one that increases the Known Received Count beyond what the encoder has sent MUST treat this as a connection error of type HTTP_QPACK_DECODER_STREAM_ERROR.

A header block consists of a prefix and a possibly empty sequence of representations defined in this section. Each representation corresponds to a single header field. These representations reference the static table or the dynamic table in a particular state, but do not modify that state. Header blocks are carried in frames on streams defined by the enclosing protocol.

Each header block is prefixed with two integers. The Required Insert Count is encoded as an integer with an 8-bit prefix after the encoding described in ). The Base is encoded as a sign bit (‘S’) and a Delta Base value with a 7-bit prefix (see ).

0 1 2 3 4 5 6 7 +---+---+---+---+---+---+---+---+ | Required Insert Count (8+) | +---+---------------------------+ | S | Delta Base (7+) | +---+---------------------------+ | Compressed Headers ... +-------------------------------+

Required Insert Count identifies the state of the dynamic table needed to process the header block. Blocking decoders use the Required Insert Count to determine when it is safe to process the rest of the block. The encoder transforms the Required Insert Count as follows before encoding:

if ReqInsertCount == 0: EncInsertCount = 0 else: EncInsertCount = (ReqInsertCount mod (2 * MaxEntries)) + 1

Here MaxEntries is the maximum number of entries that the dynamic table can have. The smallest entry has empty name and value strings and has the size of 32. Hence MaxEntries is calculated as

MaxEntries = floor( MaxTableCapacity / 32 )

MaxTableCapacity is the maximum capacity of the dynamic table as specified by the decoder (see ). This encoding limits the length of the prefix on long-lived connections. The decoder can reconstruct the Required Insert Count using an algorithm such as the following. If the decoder encounters a value of EncodedInsertCount that could not have been produced by a conformant encoder, it MUST treat this as a connection error of type HTTP_QPACK_DECOMPRESSION_FAILED. TotalNumberOfInserts is the total number of inserts into the decoder’s dynamic table.

FullRange = 2 * MaxEntries if EncodedInsertCount == 0: ReqInsertCount = 0 else: if EncodedInsertCount > FullRange: Error MaxValue = TotalNumberOfInserts + MaxEntries # MaxWrapped is the largest possible value of # ReqInsertCount that is 0 mod 2*MaxEntries MaxWrapped = floor(MaxValue / FullRange) * FullRange ReqInsertCount = MaxWrapped + EncodedInsertCount - 1 # If ReqInsertCount exceeds MaxValue, the Encoder's value # must have wrapped one fewer time if ReqInsertCount > MaxValue: if ReqInsertCount <= FullRange: Error ReqInsertCount -= FullRange # Value of 0 must be encoded as 0. if ReqInsertCount == 0: Error

For example, if the dynamic table is 100 bytes, then the Required Insert Count will be encoded modulo 6. If a decoder has received 10 inserts, then an encoded value of 3 indicates that the Required Insert Count is 9 for the header block.

The Base is used to resolve references in the dynamic table as described in . To save space, the Base is encoded relative to the Required Insert Count using a one-bit sign (‘S’) and the Delta Base value. A sign bit of 0 indicates that the Base is greater than or equal to the value of the Required Insert Count; the decoder adds the value of Delta Base to the Required Insert Count to determine the value of the Base. A sign bit of 1 indicates that the Base is less than the Required Insert Count; the decoder subtracts the value of Delta Base from the Required Insert Count and also subtracts one to determine the value of the Base. That is:

if S == 0: Base = ReqInsertCount + DeltaBase else: Base = ReqInsertCount - DeltaBase - 1

A single-pass encoder determines the Base before encoding a header block. If the encoder inserted entries in the dynamic table while encoding the header block, Required Insert Count will be greater than the Base, so the encoded difference is negative and the sign bit is set to 1. If the header block did not reference the most recent entry in the table and did not insert any new entries, the Base will be greater than the Required Insert Count, so the delta will be positive and the sign bit is set to 0. An encoder that produces table updates before encoding a header block might set Base to the value of Required Insert Count. In such case, both the sign bit and the Delta Base will be set to zero. A header block that does not reference the dynamic table can use any value for the Base; setting Delta Base to zero is one of the most efficient encodings. For example, with a Required Insert Count of 9, a decoder receives an S bit of 1 and a Delta Base of 2. This sets the Base to 6 and enables post-base indexing for three entries. In this example, a relative index of 1 refers to the 5th entry that was added to the table; a post-base index of 1 refers to the 8th entry.

An indexed header field representation identifies an entry in the static table, or an entry in the dynamic table with an absolute index less than the Base.

0 1 2 3 4 5 6 7 +---+---+---+---+---+---+---+---+ | 1 | T | Index (6+) | +---+---+-----------------------+

This representation starts with the ‘1’ 1-bit pattern, followed by the ‘T’ bit indicating whether the reference is into the static or dynamic table. The 6-bit prefix integer (see ) that follows is used to locate the table entry for the header field. When T=1, the number represents the static table index; when T=0, the number is the relative index of the entry in the dynamic table.

An indexed header field with post-base index representation identifies an entry in the dynamic table with an absolute index greater than or equal to the Base.

0 1 2 3 4 5 6 7 +---+---+---+---+---+---+---+---+ | 0 | 0 | 0 | 1 | Index (4+) | +---+---+---+---+---------------+

This representation starts with the ‘0001’ 4-bit pattern. This is followed by the post-base index (see ) of the matching header field, represented as an integer with a 4-bit prefix (see ).

A literal header field with name reference representation encodes a header field where the header field name matches the header field name of an entry in the static table, or the header field name of an entry in the dynamic table with an absolute index less than the Base.

0 1 2 3 4 5 6 7 +---+---+---+---+---+---+---+---+ | 0 | 1 | N | T |Name Index (4+)| +---+---+---+---+---------------+ | H | Value Length (7+) | +---+---------------------------+ | Value String (Length bytes) | +-------------------------------+

This representation starts with the ‘01’ two-bit pattern. The following bit, ‘N’, indicates whether an intermediary is permitted to add this header to the dynamic header table on subsequent hops. When the ‘N’ bit is set, the encoded header MUST always be encoded with a literal representation. In particular, when a peer sends a header field that it received represented as a literal header field with the ‘N’ bit set, it MUST use a literal representation to forward this header field. This bit is intended for protecting header field values that are not to be put at risk by compressing them (see for more details). The fourth (‘T’) bit indicates whether the reference is to the static or dynamic table. The 4-bit prefix integer (see ) that follows is used to locate the table entry for the header name. When T=1, the number represents the static table index; when T=0, the number is the relative index of the entry in the dynamic table. Only the header field name is taken from the dynamic table entry; the header field value is encoded as an 8-bit prefix string literal (see ).

A literal header field with post-base name reference representation encodes a header field where the header field name matches the header field name of a dynamic table entry with an absolute index greater than or equal to the Base.

0 1 2 3 4 5 6 7 +---+---+---+---+---+---+---+---+ | 0 | 0 | 0 | 0 | N |NameIdx(3+)| +---+---+---+---+---+-----------+ | H | Value Length (7+) | +---+---------------------------+ | Value String (Length bytes) | +-------------------------------+

This representation starts with the ‘0000’ four-bit pattern. The fifth bit is the ‘N’ bit as described in . This is followed by a post-base index of the dynamic table entry (see ) encoded as an integer with a 3-bit prefix (see ). Only the header field name is taken from the dynamic table entry; the header field value is encoded as an 8-bit prefix string literal (see ).

The literal header field without name reference representation encodes a header field name and a header field value as string literals.

0 1 2 3 4 5 6 7 +---+---+---+---+---+---+---+---+ | 0 | 0 | 1 | N | H |NameLen(3+)| +---+---+---+---+---+-----------+ | Name String (Length bytes) | +---+---------------------------+ | H | Value Length (7+) | +---+---------------------------+ | Value String (Length bytes) | +-------------------------------+

This representation begins with the ‘001’ three-bit pattern. The fourth bit is the ‘N’ bit as described in . The name follows, represented as a 4-bit prefix string literal, then the value, represented as an 8-bit prefix string literal (see ).

QPACK defines two settings which are included in the HTTP/3 SETTINGS frame. The default value is zero. See for usage. This is the equivalent of the SETTINGS_HEADER_TABLE_SIZE from HTTP/2. The default value is zero. See .

The following error codes are defined for HTTP/3 to indicate failures of QPACK which prevent the connection from continuing: The decoder failed to interpret a header block and is not able to continue decoding that header block. The decoder failed to interpret an encoder instruction received on the encoder stream. The encoder failed to interpret a decoder instruction received on the decoder stream.

TBD. Also see . While the negotiated limit on the dynamic table size accounts for much of the memory that can be consumed by a QPACK implementation, data which cannot be immediately sent due to flow control is not affected by this limit. Implementations should limit the size of unsent data, especially on the decoder stream where flexibility to choose what to send is limited. Possible responses to an excess of unsent data might include limiting the ability of the peer to open new streams, reading only from the encoder stream, or closing the connection.

This document specifies two settings. The entries in the following table are registered in the “HTTP/3 Settings” registry established in . Setting Name Code Specification Default QPACK_MAX_TABLE_CAPACITY 0x1 0 QPACK_BLOCKED_STREAMS 0x7 0

This document specifies two stream types. The entries in the following table are registered in the “HTTP/3 Stream Type” registry established in . Stream Type Code Specification Sender QPACK Encoder Stream 0x02 Both QPACK Decoder Stream 0x03 Both

This document specifies three error codes. The entries in the following table are registered in the “HTTP/3 Error Code” registry established in . Name Code Description Specification HTTP_QPACK_DECOMPRESSION_FAILED 0x200 Decompression of a header block failed HTTP_QPACK_ENCODER_STREAM_ERROR 0x201 Error on the encoder stream HTTP_QPACK_DECODER_STREAM_ERROR 0x202 Error on the decoder stream

Akamai Technologies Fastly Mozilla

Index Name Value 0 :authority   1 :path / 2 age 0 3 content-disposition   4 content-length 0 5 cookie   6 date   7 etag   8 if-modified-since   9 if-none-match   10 last-modified   11 link   12 location   13 referer   14 set-cookie   15 :method CONNECT 16 :method DELETE 17 :method GET 18 :method HEAD 19 :method OPTIONS 20 :method POST 21 :method PUT 22 :scheme http 23 :scheme https 24 :status 103 25 :status 200 26 :status 304 27 :status 404 28 :status 503 29 accept */* 30 accept application/dns-message 31 accept-encoding gzip, deflate, br 32 accept-ranges bytes 33 access-control-allow-headers cache-control 34 access-control-allow-headers content-type 35 access-control-allow-origin * 36 cache-control max-age=0 37 cache-control max-age=2592000 38 cache-control max-age=604800 39 cache-control no-cache 40 cache-control no-store 41 cache-control public, max-age=31536000 42 content-encoding br 43 content-encoding gzip 44 content-type application/dns-message 45 content-type application/javascript 46 content-type application/json 47 content-type application/x-www-form-urlencoded 48 content-type image/gif 49 content-type image/jpeg 50 content-type image/png 51 content-type text/css 52 content-type text/html; charset=utf-8 53 content-type text/plain 54 content-type text/plain;charset=utf-8 55 range bytes=0- 56 strict-transport-security max-age=31536000 57 strict-transport-security max-age=31536000; includesubdomains 58 strict-transport-security max-age=31536000; includesubdomains; preload 59 vary accept-encoding 60 vary origin 61 x-content-type-options nosniff 62 x-xss-protection 1; mode=block 63 :status 100 64 :status 204 65 :status 206 66 :status 302 67 :status 400 68 :status 403 69 :status 421 70 :status 425 71 :status 500 72 accept-language   73 access-control-allow-credentials FALSE 74 access-control-allow-credentials TRUE 75 access-control-allow-headers * 76 access-control-allow-methods get 77 access-control-allow-methods get, post, options 78 access-control-allow-methods options 79 access-control-expose-headers content-length 80 access-control-request-headers content-type 81 access-control-request-method get 82 access-control-request-method post 83 alt-svc clear 84 authorization   85 content-security-policy script-src 'none'; object-src 'none'; base-uri 'none' 86 early-data 1 87 expect-ct   88 forwarded   89 if-range   90 origin   91 purpose prefetch 92 server   93 timing-allow-origin * 94 upgrade-insecure-requests 1 95 user-agent   96 x-forwarded-for   97 x-frame-options deny 98 x-frame-options sameorigin

Pseudo-code for single pass encoding, excluding handling of duplicates, non-blocking mode, and reference tracking.

baseIndex = dynamicTable.baseIndex largestReference = 0 for header in headers: staticIdx = staticTable.getIndex(header) if staticIdx: encodeIndexReference(streamBuffer, staticIdx) continue dynamicIdx = dynamicTable.getIndex(header) if !dynamicIdx: # No matching entry. Either insert+index or encode literal nameIdx = getNameIndex(header) if shouldIndex(header) and dynamicTable.canIndex(header): encodeLiteralWithIncrementalIndex(controlBuffer, nameIdx, header) dynamicTable.add(header) dynamicIdx = dynamicTable.baseIndex if !dynamicIdx: # Couldn't index it, literal if nameIdx <= staticTable.size: encodeLiteral(streamBuffer, nameIndex, header) else: # encode literal, possibly with nameIdx above baseIndex encodeDynamicLiteral(streamBuffer, nameIndex, baseIndex, header) largestReference = max(largestReference, dynamicTable.toAbsolute(nameIdx)) else: # Dynamic index reference assert(dynamicIdx) largestReference = max(largestReference, dynamicIdx) # Encode dynamicIdx, possibly with dynamicIdx above baseIndex encodeDynamicIndexReference(streamBuffer, dynamicIdx, baseIndex) # encode the prefix encodeInteger(prefixBuffer, 0x00, largestReference, 8) if baseIndex >= largestReference: encodeInteger(prefixBuffer, 0, baseIndex - largestReference, 7) else: encodeInteger(prefixBuffer, 0x80, largestReference - baseIndex, 7) return controlBuffer, prefixBuffer + streamBuffer

RFC Editor’s Note: Please remove this section prior to publication of a final version of this document.

Editorial changes

Decoders MUST emit Header Acknowledgements (#2939) Updated error code for multiple encoder or decoder streams (#2970) Added explicit defaults for new SETTINGS (#2974)

Endpoints are permitted to create encoder and decoder streams even if they can’t use them (#2100, #2529) Maximum values for settings removed (#2766, #2767)

Clarify initial dynamic table capacity maximums (#2276, #2330, #2330)

Introduced the terms dynamic table capacity and maximum dynamic table capacity. Renamed SETTINGS_HEADER_TABLE_SIZE to SETTINGS_QPACK_MAX_TABLE_CAPACITY.

Changed calculation of Delta Base Index to avoid an illegal value (#2002, #2005)

Change HTTP settings defaults (#2038) Substantial editorial reorganization

Largest Reference encoded modulo MaxEntries (#1763) New Static Table (#1355) Table Size Update with Insert Count=0 is a connection error (#1762) Stream Cancellations are optional when SETTINGS_HEADER_TABLE_SIZE=0 (#1761) Implementations must handle 62 bit integers (#1760) Different error types for each QPACK stream, other changes to error handling (#1726) Preserve header field order (#1725) Initial table size is the maximum permitted when table is first usable (#1642)

Only header blocks that reference the dynamic table are acknowledged (#1603, #1605)

Renumbered instructions for consistency (#1471, #1472) Decoder is allowed to validate largest reference (#1404, #1469) Header block acknowledgments also acknowledge the associated largest reference (#1370, #1400) Added an acknowledgment for unread streams (#1371, #1400) Removed framing from encoder stream (#1361,#1467) Control streams use typed unidirectional streams rather than fixed stream IDs (#910,#1359)

Separate instruction sets for table updates and header blocks (#1235, #1142, #1141) Reworked indexing scheme (#1176, #1145, #1136, #1130, #1125, #1314) Added mechanisms that support one-pass encoding (#1138, #1320) Added a setting to control the number of blocked decoders (#238, #1140, #1143) Moved table updates and acknowledgments to dedicated streams (#1121, #1122, #1238)

This draft draws heavily on the text of . The indirect input of those authors is gratefully acknowledged, as well as ideas from: Ryan Hamilton Patrick McManus Kazuho Oku Biren Roy Ian Swett Dmitri Tikhonov Buck’s contribution was supported by Google during his employment there. A substantial portion of Mike’s contribution was supported by Microsoft during his employment there.