{"service":"corpus","corpus":"rfc","source":"RFC Editor","license":"IETF Trust / RFC Editor terms","record":{"id":"10069","title":"Managing Multiple Paths for a QUIC Connection","text":"﻿\n\n\n\nInternet Engineering Task Force (IETF)              刘彦梅 (Y. Liu), Ed.\nRequest for Comments: 10069                                 Alibaba Inc.\nCategory: Standards Track                                 马云飞 (Y. Ma)\nISSN: 2070-1721                                                Harvey AI\n                                                      Q. De Coninck, Ed.\n                                              University of Mons (UMONS)\n                                                          O. Bonaventure\n                                                     UCLouvain and WELRI\n                                                              C. Huitema\n                                                    Private Octopus Inc.\n                                                       M. Kühlewind, Ed.\n                                                                Ericsson\n                                                           November 2026\n\n\n             Managing Multiple Paths for a QUIC Connection\n\nAbstract\n\n   This document specifies a multipath extension for the QUIC protocol\n   to enable the simultaneous usage of multiple paths for a single\n   connection.  It introduces explicit path identifiers to create,\n   delete, and manage multiple paths.  This document does not specify\n   address discovery or management, nor how applications using QUIC\n   schedule traffic over multiple paths.\n\nStatus of This Memo\n\n   This is an Internet Standards Track document.\n\n   This document is a product of the Internet Engineering Task Force\n   (IETF).  It represents the consensus of the IETF community.  It has\n   received public review and has been approved for publication by the\n   Internet Engineering Steering Group (IESG).  Further information on\n   Internet Standards is available in Section 2 of RFC 7841.\n\n   Information about the current status of this document, any errata,\n   and how to provide feedback on it may be obtained at\n   https://www.rfc-editor.org/info/rfc10069.\n\nCopyright Notice\n\n   Copyright (c) 2026 IETF Trust and the persons identified as the\n   document authors.  All rights reserved.\n\n   This document is subject to BCP 78 and the IETF Trust's Legal\n   Provisions Relating to IETF Documents\n   (https://trustee.ietf.org/license-info) in effect on the date of\n   publication of this document.  Please review these documents\n   carefully, as they describe your rights and restrictions with respect\n   to this document.  Code Components extracted from this document must\n   include Revised BSD License text as described in Section 4.e of the\n   Trust Legal Provisions and are provided without warranty as described\n   in the Revised BSD License.\n\nTable of Contents\n\n   1.  Introduction\n     1.1.  Conventions and Definitions\n   2.  Connection Lifecycle and Packet Protection\n     2.1.  initial_max_path_id Transport Parameter\n     2.2.  Relation to Other Transport Parameters\n     2.3.  Handling ACK and PATH_ACK in 0-RTT and 1-RTT\n     2.4.  Nonce Calculation after Handshake Completion\n     2.5.  Key Phase Update Process\n     2.6.  Connection Closure\n   3.  Path Management\n     3.1.  Path Initiation and Validation\n       3.1.1.  Path Establishment Example\n       3.1.2.  Relation to Probing and Migration\n       3.1.3.  Address Validation Token\n     3.2.  Handling Connection IDs\n       3.2.1.  Issuing New Connection IDs\n       3.2.2.  Rotating and Retiring Connection IDs\n     3.3.  Path Status Management\n     3.4.  Path Close\n       3.4.1.  Path Closure Example\n       3.4.2.  Avoiding Spurious Stateless Resets\n       3.4.3.  Handling PATH_ACK for Abandoned Paths\n   4.  New Frames\n     4.1.  PATH_ACK Frame\n     4.2.  PATH_ABANDON Frame\n       4.2.1.  Error Codes\n     4.3.  PATH_STATUS_AVAILABLE and PATH_STATUS_BACKUP Frames\n     4.4.  PATH_NEW_CONNECTION_ID Frame\n     4.5.  PATH_RETIRE_CONNECTION_ID Frame\n     4.6.  MAX_PATH_ID Frame\n     4.7.  PATHS_BLOCKED and PATH_CIDS_BLOCKED Frames\n   5.  Implementation Considerations\n     5.1.  Connection ID Changes, Migration, and NAT Rebindings\n     5.2.  Using Multiple Paths on the Same 4-tuple\n     5.3.  Congestion Control\n     5.4.  Computing Path RTT\n     5.5.  Packet Scheduling\n     5.6.  Retransmissions\n     5.7.  PTO Expiration\n     5.8.  Paths Having Different PMTU Sizes\n     5.9.  Idle Timeout and Keep-Alives\n   6.  IANA Considerations\n   7.  Security Considerations\n     7.1.  Memory Allocation for Per-Path Resources\n     7.2.  Denial of Service with Multiple Paths\n     7.3.  Cryptographic Handshake and AEAD Nonce\n   8.  References\n     8.1.  Normative References\n     8.2.  Informative References\n   Acknowledgments\n   Authors' Addresses\n\n1.  Introduction\n\n   This document specifies an extension to QUIC version 1\n   [QUIC-TRANSPORT] to enable the simultaneous usage of multiple paths\n   for a single connection, using the same or different 4-tuples (of\n   source/destination port numbers and source/destination IP addresses).\n\n   Connection migration as specified in Section 9 of [QUIC-TRANSPORT]\n   directs a peer to switch sending through a new preferred path and, if\n   successful, to release resources associated with the old path.  The\n   multipath extension specified in this document builds on this\n   mechanism but introduces a path identifier, or path ID, to manage\n   connection IDs and packet number spaces per path, enabling the use of\n   multiple paths simultaneously.  As such, a path in this extension is\n   defined by its path ID.  Note that it therefore is possible to have\n   multiple paths / path IDs for the same 4-tuple.\n\n   The connection ID of a packet binds the packet to a path ID and\n   therefore to a packet number space.  That means each connection ID is\n   associated with exactly one path ID but multiple connection IDs are\n   usually issued for each path ID.  The same path ID is used in both\n   directions, starting with 0 for the initial path.  Path IDs are\n   monotonically increasing and cannot be reused.\n\n   This extension uses multiple packet number spaces, specifically one\n   for each path.  Each path ID-specific packet number space starts at\n   packet number 0.  As such, each path maintains distinct packet number\n   states for sending and receiving packets, as in [QUIC-TRANSPORT].\n   Using multiple packet number spaces enables direct use of the loss\n   detection and congestion control mechanisms defined in\n   [QUIC-RECOVERY] on a per-path basis.  However, use of multiple packet\n   number spaces requires non-zero connection IDs in order to identify\n   the path and the respective packet number space, as well as a\n   modified Authenticated Encryption with Associated Data (AEAD)\n   calculation including the path ID (see Section 2.4).\n\n   As such, this extension specifies a departure from the specification\n   of path management in Section 9 of [QUIC-TRANSPORT] and therefore\n   requires a new transport parameter, as specified in Section 2.1, to\n   indicate support of the multipath extension specified in this\n   document.\n\n   Further, this document specifies the needed path management\n   mechanisms for path initiation in Section 3.1, handling of per-path\n   connection IDs in Section 3.2, signaling of preferred path usage in\n   Section 3.3, and explicit removal of paths that have been abandoned\n   in Section 3.4.  Note that in this extension, a QUIC server does not\n   initiate the creation of a path, but it has to validate a new path\n   created by a client.\n\n   This extension does not cover address discovery and management.  The\n   extension specified in this document can be used as is if address\n   management and the actual decision to set up or tear down paths are\n   handled by the application.  This document does not prevent future\n   extensions from defining mechanisms to cope with the remaining\n   scenarios.\n\n   Further, this document does not specify scheduling algorithms that\n   define how multiple, simultaneously open paths are used to send\n   packets.  As these differ depending on application requirements, only\n   some basic implementation guidance is discussed in Section 5.  This\n   extension can be used with different scheduling algorithms that,\n   e.g., can range from support for failover to simultaneous use of the\n   aggregated capacity across all open paths.  There are currently no\n   IETF specifications that define scheduling algorithms for\n   simultaneously (i.e., concurrently) using multiple paths.\n\n   Specifically, while failover between Wi-Fi and mobile networks is a\n   well-known multipath use case, it only temporarily uses two paths at\n   the same time to avoid transmission pauses.  Simultaneous path usage\n   generally, however, needs more consideration than specified in this\n   document to avoid negative performance impacts, e.g., when stream\n   data is distributed over multiple paths with different delays.\n\n   The operational considerations for QUIC are addressed in [RFC9312].\n   They also apply to QUIC connections using the extensions defined in\n   this document.  An additional complexity is that applications might\n   use a combination of monitored and non-monitored paths, but that\n   complexity already exists when using path migration as defined in\n   [QUIC-TRANSPORT].\n\n1.1.  Conventions and Definitions\n\n   The key words \"MUST\", \"MUST NOT\", \"REQUIRED\", \"SHALL\", \"SHALL NOT\",\n   \"SHOULD\", \"SHOULD NOT\", \"RECOMMENDED\", \"NOT RECOMMENDED\", \"MAY\", and\n   \"OPTIONAL\" in this document are to be interpreted as described in BCP\n   14 [RFC2119] [RFC8174] when, and only when, they appear in all\n   capitals, as shown here.\n\n   This document uses the terminology defined in [QUIC-TRANSPORT].  When\n   this document uses the term \"path\", it refers to the notion of\n   \"network path\" used in [QUIC-TRANSPORT].\n\n   The packet diagrams in this document use the conventions defined in\n   Section 1.3 of [QUIC-TRANSPORT], including the notation (i) to denote\n   variable-length integers, encoded as specified in Section 16 of\n   [QUIC-TRANSPORT].\n\n2.  Connection Lifecycle and Packet Protection\n\n   This document defines a new transport parameter initial_max_path_id\n   to indicate the support of the multipath extension.  If either of the\n   endpoints does not advertise the initial_max_path_id transport\n   parameter, then both endpoints MUST NOT use any frame or mechanism\n   defined in this document.  If both endpoints advertise the\n   initial_max_path_id transport parameter, once the handshake is\n   completed, a new AEAD usage applies to all 1-RTT packets, as\n   specified in Section 2.4, and new paths can be used, as specified in\n   Section 3.  As specified in Section 4, the new frames defined in this\n   document can only be sent in 1-RTT packets.\n\n   The transport parameter negotiation enables incremental deployment of\n   implementations of the multipath extension.\n\n2.1.  initial_max_path_id Transport Parameter\n\n   The new transport parameter is defined as follows:\n\n   initial_max_path_id (parameter ID 0x3e):  Parameter value is a\n      variable-length integer specifying the maximum path ID an endpoint\n      is willing to maintain at connection initiation.  This value MUST\n      NOT exceed 2^32-1, which is the maximum allowed value for the path\n      ID due to restrictions on the nonce calculation (see Section 2.4).\n\n   The initial_max_path_id transport parameter limits the initial\n   maximum number of open paths that can be used during a connection.\n   For example, if initial_max_path_id is set to 1, only connection IDs\n   associated with path IDs 0 and 1 should be issued by the peer.  If an\n   endpoint receives an initial_max_path_id transport parameter with\n   value 0, the peer aims to enable the multipath extension without\n   allowing extra paths immediately.\n\n   Setting initial_max_path_id parameter is equivalent to sending a\n   MAX_PATH_ID frame (Section 4.6) with the same value.  As such to\n   allow for the use of more paths later, endpoints can send the\n   MAX_PATH_ID frame to increase the maximum allowed path ID.\n\n   If an initial_max_path_id transport parameter value higher than\n   2^32-1 is received, the receiver MUST close the connection with an\n   error of type TRANSPORT_PARAMETER_ERROR.\n\n   When advertising the initial_max_path_id transport parameter,\n   endpoints MUST use Source and Destination Connection IDs with non-\n   zero lengths.  If an initial_max_path_id transport parameter is\n   received and the carrying packet contains a zero-length connection\n   ID, the receiver MUST treat this as a connection error of type\n   PROTOCOL_VIOLATION and close the connection.\n\n   Cipher suites with a nonce shorter than 12 bytes cannot be used\n   together with the multipath extension.  If such a cipher suite is\n   selected and the use of the multipath extension is supported,\n   endpoints MUST abort the handshake with an error of type\n   TRANSPORT_PARAMETER_ERROR.\n\n   The initial_max_path_id parameter MUST NOT be remembered for use in a\n   subsequent connection (Section 7.4.1 of [QUIC-TRANSPORT]).\n\n2.2.  Relation to Other Transport Parameters\n\n   When the QUIC multipath extension is used, the\n   active_connection_id_limit transport parameter [QUIC-TRANSPORT]\n   limits the maximum number of active connection IDs per path.  As\n   defined in Section 5.1.1 of [QUIC-TRANSPORT], connection IDs that are\n   issued and not retired are considered active.\n\n   If an endpoint receives a disable_active_migration transport\n   parameter, it is forbidden to establish new paths to the peer's\n   handshake address.  However, establishment of additional paths to\n   other peer addresses (e.g., carried by peer's preferred_address) is\n   immediately valid.\n\n   If the server uses the preferred_address transport parameter, clients\n   cannot assume that the initial server address and the addresses\n   contained in this parameter can be simultaneously used for multipath\n   (Section 9.6.2 of [QUIC-TRANSPORT]).  Use of the preferred address\n   with the same local address is considered as a migration event that\n   does not change the path ID.  A such, the path ID for the connection\n   ID specified in the preferred_address transport parameter is 0.\n\n2.3.  Handling ACK and PATH_ACK in 0-RTT and 1-RTT\n\n   The PATH_ACK frame (see Section 4.1) is used to acknowledge 1-RTT\n   packets.  Compared to the ACK frame, as specified in Section 19.3 of\n   [QUIC-TRANSPORT], the PATH_ACK frame additionally contains the path\n   ID to identify the path-specific packet number space.  ACK frames\n   when used with the multipath extension acknowledge packets for the\n   path with path ID 0.  As multipath support is unknown during the\n   handshake, acknowledgments of Initial and Handshake packets are sent\n   using ACK frames.\n\n   After the handshake concludes with support for the multipath\n   extension, endpoints SHOULD use PATH_ACK frames instead of ACK\n   frames, including for so far unacknowledged 0-RTT packets using path\n   ID 0.  Endpoints MUST still process ACK frames that acknowledge 0-RTT\n   packets or 1-RTT packets.  For example, a sender could negotiate\n   multipath support for later use and keep only the initial path with\n   path ID 0 for a while.  During this single-path period, the sender\n   might prefer to send ACK frames.\n\n2.4.  Nonce Calculation after Handshake Completion\n\n   Section 5.3 of [QUIC-TLS] specifies AEAD usage, particularly the use\n   of a nonce, N, formed by combining the packet protection\n   Initialization Vector (IV) with the packet number.  When multiple\n   packet number spaces are used, the packet number alone would not\n   guarantee the uniqueness of the nonce.  Therefore, the nonce N is\n   calculated for 1-RTT packets if the multipath extension is used by\n   combining the packet protection IV with the packet number and with\n   the 32 bits of the path ID.  In order to guarantee the uniqueness of\n   the nonce, the path ID is limited to a max value of 2^32-1, as\n   specified in Section 2.1.\n\n   To calculate the nonce, a 96-bit Path-and-Packet-Number (PPN) is\n   composed of the 32 bits of the path ID in network byte order, two\n   zero bits, and the 62 bits of the reconstructed QUIC packet number in\n   network byte order, as illustrated in Figure 1.\n\n     PPN {\n       Path ID (32),\n       Zeroes (2) = 0b00,\n       Packet Number (62)\n     }\n\n                  Figure 1: 96 Bits Path-And-Packet-Number\n\n   The IV length is equal to the nonce length.  If the IV is larger than\n   96 bits, the path-and-packet-number is left-padded with zeros to the\n   size of the IV.  The exclusive OR of the padded packet number and the\n   IV forms the AEAD nonce.  An AEAD algorithm where the nonce length is\n   less than 12 bytes cannot be used with the QUIC multipath extension.\n   The following figure illustrates this for a 96-bit IV.\n\n   IV(12);\n   N(12) = IV xor PPN;\n\n                        Figure 2: Nonce Calculation\n\n   For example, assuming the IV value is 0x6b26114b9cba2b63a9e8dd4f, the\n   path ID is 3, and the packet number is 54321 (hex value 0xd431), the\n   nonce will be set to 0x6b2611489cba2b63a9e8097e, as illustrated below\n   (where ⊕ represents exclusive OR):\n\n      IV:      6b26114b9cba2b63a9e8dd4f\n   ⊕  PPN:     00000003000000000000d431\n   ------------------------------------\n      Nonce:   6b2611489cba2b63a9e8097e\n\n                    Figure 3: Example Nonce Calculation\n\n2.5.  Key Phase Update Process\n\n   The Key Phase bit update process is specified in Section 6 of\n   [QUIC-TLS].  The general principles of key update are not changed in\n   this specification.  Following [QUIC-TLS], the Key Phase bit is used\n   to indicate which packet protection keys are used to protect the\n   packet.  The Key Phase bit is toggled to signal each subsequent key\n   update.\n\n   Because of network delays, packets protected with the older key might\n   arrive later than the packets protected with the new key.  However,\n   receivers can solely rely on the Key Phase bit to determine the\n   corresponding packet protection key, assuming that there is a\n   sufficiently large interval between two consecutive key updates\n   (Section 6.5 of [QUIC-TLS]).\n\n   When this specification is used, endpoints SHOULD wait for at least\n   three times the largest Probe Timeout (PTO) (see Section 6.2 of\n   [QUIC-RECOVERY]) among all the paths before initiating a new key\n   update after receiving an acknowledgment that confirms the receipt of\n   the previous key update.  This interval is different from that in\n   [QUIC-TLS], which used three times the PTO of the single path.\n\n   Since packets that arrive after their decryption key has been\n   discarded will be dropped, the choice of three times the largest PTO\n   is a trade-off: Longer delays reduce the probability of losing\n   packets, but keeping old keys longer can negatively impact the\n   security of the protocol.  The use of three times the largest PTO\n   aims to minimize packet lost for all paths and therefore limits the\n   impact on performance.\n\n   Following Section 5.4 of [QUIC-TLS], the Key Phase bit is protected,\n   so sending multiple packets with the Key Phase bit flipping at the\n   same time should not cause activity across different paths to be\n   linkable by an observer.\n\n2.6.  Connection Closure\n\n   CONNECTION_CLOSE frames and their processing are unchanged from\n   [QUIC-TRANSPORT].  They can be sent on any open path.  Section 10.2\n   of [QUIC-TRANSPORT] specifies that the closing and draining\n   connection states \"SHOULD persist for at least three times the\n   current PTO\".  When this specification is used, these states SHOULD\n   instead persist for at least three times the largest PTO among all\n   paths.\n\n3.  Path Management\n\n   After completing the handshake indicating multipath support,\n   endpoints can start using multiple paths.  In accordance with\n   Section 9 of [QUIC-TRANSPORT], this extension only enables clients to\n   open a new path.  The client can open a new path when both endpoints\n   have issued available connection IDs for at least one unused, common\n   path ID, as the same path ID is used in both directions.\n\n   This document specifies path initiation (see Section 3.1), issuing\n   and retirement of per-path connection IDs (see Section 3.2), path\n   status management (see Section 3.3), and path closure (see\n   Section 3.4).  However, this document does not specify when a client\n   decides to initiate or close a path or how multiple open paths are\n   used for sending.\n\n   For path management, this extension specifies the following frames in\n   Section 4:\n\n   *  PATH_ABANDON (see Section 4.2)\n\n   *  PATH_STATUS_BACKUP (see Section 4.3)\n\n   *  PATH_STATUS_AVAILABLE (see Section 4.3)\n\n   *  PATH_NEW_CONNECTION_ID (see Section 4.4)\n\n   *  PATH_RETIRE_CONNECTION_ID (see Section 4.5)\n\n   *  MAX_PATH_ID (see Section 4.6)\n\n   *  PATHS_BLOCKED (see Section 4.7)\n\n   *  PATH_CIDS_BLOCKED (see Section 4.7)\n\n3.1.  Path Initiation and Validation\n\n   To open a new path, an endpoint MUST use a new connection ID\n   associated with an unused path ID.  An endpoint MUST use a connection\n   ID associated to the same path ID as used in the packet received by\n   the endpoint when it intends to send packets on the same path.\n\n   A client that wants to use a new path MUST validate the peer's\n   address as described in Section 8.2 of [QUIC-TRANSPORT], unless it\n   has previously validated the 4-tuple used for that path.\n\n   After receiving packets from the client on a new path, if the server\n   decides to use the new path, the server MUST validate the peer's\n   address before sending any data as described in (Section 8.2 of\n   [QUIC-TRANSPORT]), unless it has previously validated the 4-tuple\n   used for that path.  Until the client's address is validated, the\n   anti-amplification limit from Section 8 of [QUIC-TRANSPORT] applies.\n\n   If an endpoint sends a PATH_RESPONSE (Section 19.18 of\n   [QUIC-TRANSPORT]), it MUST be sent on the same path as used by the\n   packet that contained the PATH_CHALLENGE frame (Section 19.17 of\n   [QUIC-TRANSPORT]) using a connection ID associated with the same path\n   ID.\n\n   The server might receive packets for a yet unused path ID that do not\n   contain a PATH_CHALLENGE frame.  Such packets are valid if they can\n   be properly decrypted given a valid connection ID.\n\n   Each endpoint MUST also validate that a minimum QUIC packet MTU of\n   1200 bytes is supported on the path.  This can be done during initial\n   path validation or separately later if the amplification limit\n   prevents it initially, as specified in Section 8.2.1 of\n   [QUIC-TRANSPORT].\n\n   An endpoint that receives packets on a new path and does not want to\n   establish this path is expected to close the path by sending a\n   PATH_ABANDON on another path, as specified in Section 3.4.\n\n   An endpoint that has no active connection ID for this path or lacks\n   other resources to immediately configure a new path could delay\n   sending the PATH_RESPONSE until sufficient resources are available.\n   Long delays might cause the peer to repeat the PATH_CHALLENGE and\n   eventually send a PATH_ABANDON, in which case the procedures\n   specified in Section 3.4 apply.\n\n   PATH_ACK frames (see Section 4.1) can be returned on any path.  If\n   the PATH_ACK is preferred to be sent on the same path as the\n   acknowledged packet (see Section 5.4 for further guidance), it can be\n   beneficial to bundle a PATH_ACK frame with the PATH_RESPONSE frame\n   during path validation.\n\n   If validation succeeds, the client can continue to use the path.  If\n   validation fails, the client MUST NOT use the path and can remove any\n   status associated to the path initiation attempt.  As the used path\n   ID is consumed either way, the endpoint MUST explicitly close the\n   path, as specified in Section 3.4.\n\n3.1.1.  Path Establishment Example\n\n   In the example below, it is assumed that both endpoints have\n   indicated an initial_max_path_id value of at least 2, which means\n   both endpoints can use path IDs 0, 1, and 2.  Note that path ID 0 is\n   already used for the initial path.\n\n      Client                                                  Server\n\n      (Provide new CIDs for path 1 on an existing path 0)\n      1-RTT[X]: DCID=S0, PATH_NEW_CONNECTION_ID[C1, Seq=0, PathID=1] -->\n              <-- 1-RTT[Y]: DCID=C0,\n                            PATH_NEW_CONNECTION_ID[S1, Seq=0, PathID=1],\n                            PATH_ACK[PathID=0, PN=X]\n              <-- 1-RTT[Y+1]: DCID=C0, PATH_NEW_CONNECTION_ID[S2, Seq=0,\n                                                               PathID=2]\n      ...\n      (start sending packets on a new path using path ID 1)\n      1-RTT[0]: DCID=S1, PATH_CHALLENGE[X] -->\n           <-- 1-RTT[0]: DCID=C1, PATH_RESPONSE[X], PATH_CHALLENGE[Y],\n                                                PATH_ACK[PathID=1, PN=0]\n      1-RTT[1]: DCID=S1, PATH_RESPONSE[Y],\n               PATH_ACK[PathID=1, PN=0], ... -->\n\n                Figure 4: Example of New Path Establishment\n\n   In Figure 4, the endpoints first exchange new available connection\n   IDs with the PATH_NEW_CONNECTION_ID frame, as further explained in\n   Section 3.2.  In this example, the client provides one connection ID\n   (C1 with path ID 1), and the server provides two connection IDs (S1\n   with path ID 1 and S2 with path ID 2).\n\n   Before the client opens a new path by sending a packet on that path\n   with a PATH_CHALLENGE frame, it has to check whether there is an\n   unused connection ID for the same unused path ID available for each\n   side.  In this example, the path ID 1 is used, which is the smallest\n   unused path ID available as recommended in Section 3.2.\n   Respectively, the client chooses the connection ID S1 as the\n   Destination Connection ID of the new path when sending the\n   PATH_CHALLENGE frame.  The server replies with a PATH_RESPONSE\n   bundled with the PATH_ACK using connection ID C1 associated with the\n   same path ID.\n\n3.1.2.  Relation to Probing and Migration\n\n   Section 9.1 of [QUIC-TRANSPORT] introduces the concept of \"probing\"\n   and \"non-probing\" frames.  A packet that contains at least one \"non-\n   probing\" frame is a \"non-probing\" packet.  Migration, as specified in\n   Section 9.2 of [QUIC-TRANSPORT], is initiated by sending packets\n   containing non-probing frames on a new (validated) path; however, it\n   uses the same path ID as on the old path.  When the multipath\n   extension is negotiated, the reception of any packet (no matter if\n   \"probing\" or \"non-probing\") on a new path with a new, so far unused\n   path ID does not impact the path status of any existing path.\n   Therefore, any frame can be sent on a new path with a new path ID at\n   any time as long as the anti-amplification limits (see\n   Section 21.1.1.1 of [QUIC-TRANSPORT]) and the congestion control\n   limits for this path are respected.\n\n   An endpoint could receive a packet with a connection ID associated to\n   an active path ID where the packet's 4-tuple does not match the\n   4-tuple currently used with that path ID.  This MUST be treated as\n   path migration, as specified in Section 9.3 of [QUIC-TRANSPORT], with\n   the constraint that all connection IDs used during path migration\n   MUST be associated with the current path ID of the path being\n   migrated.\n\n3.1.3.  Address Validation Token\n\n   As specified in Section 9.3 of [QUIC-TRANSPORT], the server is\n   expected to send a new address validation token to a client following\n   the successful validation of a new client address.  The client will\n   receive several tokens.  When considering using a token for\n   subsequent connections, it might be difficult for the client to pick\n   the \"right\" token among multiple tokens obtained in a previous\n   connection.  The client is likely to fall back to the strategy\n   specified in Section 8.1.3 of [QUIC-TRANSPORT], i.e., picking the\n   last received token.  To avoid issues when clients make the \"wrong\"\n   choice, a server SHOULD issue tokens that are capable of validating\n   any of the previously validated addresses.  Including more addresses\n   increases the probability that the token will be useful in the future\n   but at the cost of a larger token.  Further guidance on token usage\n   can be found in Section 8.1.3 of [QUIC-TRANSPORT].\n\n3.2.  Handling Connection IDs\n\n   When the multipath extension is used, endpoints have to use the\n   PATH_NEW_CONNECTION_ID and PATH_RETIRE_CONNECTION_ID frames to\n   indicate the respective path ID together with the associated sequence\n   number (see Section 5.1.1 of [QUIC-TRANSPORT]), at least for all\n   paths with a path ID other than 0.  Each path ID has its own\n   connection ID sequence number space whose initial value is 0.\n\n   Endpoints SHOULD also use PATH_NEW_CONNECTION_ID and\n   PATH_RETIRE_CONNECTION_ID for the initial path with path ID 0.\n   However, the use of NEW_CONNECTION_ID and RETIRE_CONNECTION_ID is\n   still valid and endpoints need to process these frames as\n   corresponding to path ID 0.\n\n3.2.1.  Issuing New Connection IDs\n\n   In order to let the peer open new paths, it is RECOMMENDED to\n   proactively issue at least one Connection ID for each unused path ID\n   up to the minimum of the peer's and the local maximum path ID limits.\n\n   If an endpoint does not want to issue connection IDs for all unused\n   path IDs for any reason, it SHOULD NOT introduce discontinuity in the\n   issuing of path IDs, as path initiation requires available connection\n   IDs for the same path ID on both sides.  For instance, if the maximum\n   path ID limit is 2 and the endpoint wants to provide connection IDs\n   for only one path ID inside range [1, 2], it should select path ID 1\n   (and not path ID 2).\n\n   Similarly, endpoints SHOULD consume path IDs in a continuous way,\n   i.e., when creating paths.  However, endpoints cannot expect to\n   receive new connection IDs or path initiation attempts with in-order\n   use of path IDs due to out-of-order delivery or path validation\n   failure.\n\n   Each endpoint maintains the set of connection IDs received from its\n   peer for each path, any of which it can use when sending packets on\n   that path; see also Section 5.1 of [QUIC-TRANSPORT].  Usually, it is\n   desired to provide at least one additional connection ID for all used\n   paths to allow for (unintentional) migration events (Section 9.5 of\n   [QUIC-TRANSPORT]).\n\n   As further specified in Section 5.1 of [QUIC-TRANSPORT], connection\n   IDs cannot be issued more than once on the same connection and\n   therefore are unique for the scope of the connection, regardless of\n   the associated path ID.\n\n   Endpoints MUST NOT issue new connection IDs with path IDs greater\n   than the Maximum Path Identifier field in MAX_PATH_ID frames (see\n   Section 4.6) or the value of initial_max_path_id transport parameter\n   if no MAX_PATH_ID frame was received yet.  Receipt of a frame with a\n   greater path ID is a connection error as specified in Section 4.\n\n   When an endpoint cannot open a path because there are no unused path\n   IDs, it SHOULD either send a MAX_PATH_ID frame to increase the\n   maximum path ID limit (when limited by the sender) or a PATHS_BLOCKED\n   frame (see Section 4.7) to inform the peer that its current limit\n   prevented the creation of the new path.\n\n3.2.2.  Rotating and Retiring Connection IDs\n\n   Section 5.1.2 of [QUIC-TRANSPORT] indicates that an endpoint can\n   change the connection ID it uses to another available one at any time\n   during the connection.  For the extension specified in this document,\n   endpoints MUST only rotate to another connection ID associated with\n   the same path ID.  Use of a connection ID associated with another\n   path ID will be considered as an attempt to open a new path instead.\n\n   An endpoint is supposed to retire any connection ID that is not being\n   used, and the server is expected to provide replacements, as\n   specified in Section 5.1.2 of [QUIC-TRANSPORT].  As such, when\n   receiving a PATH_RETIRE_CONNECTION_ID frame, an endpoint SHOULD\n   provide new connection IDs for that path, if still open, using\n   PATH_NEW_CONNECTION_ID frames.\n\n   While it is expected that the peer provides at least one unused\n   connection ID for all active paths using the PATH_NEW_CONNECTION_ID\n   after retirement of an old connection ID, an endpoint MAY send a\n   PATH_CIDS_BLOCKED (see Section 4.7) if it wants to change the\n   connection ID but no unused connection ID for a that path is\n   available.  Further, an endpoint MAY also send a PATH_CIDS_BLOCKED\n   frame if it wants to open a new path and has no connection IDs\n   available for an unused path ID, even though the Maximum Path\n   Identifier value would allow for more paths.\n\n   Retirement of connection IDs will not retire the path ID that\n   corresponds to the connection ID or any other path resources as the\n   packet number space is associated to the path ID.\n\n   The peer that sends the PATH_RETIRE_CONNECTION_ID frame can keep\n   sending data on the path that the retired connection ID was used on\n   but has to use a different connection ID for the same path ID when\n   doing so.\n\n3.3.  Path Status Management\n\n   An endpoint can send PATH_STATUS_BACKUP and PATH_STATUS_AVAILABLE\n   frames (see Section 4.3) to inform the peer that it should send\n   packets on the paths with the preference expressed by these frames.\n   Note that an endpoint might not follow the peer's advertisements, but\n   these frames are still a clear signal of the peer's preference of\n   path usage.\n\n   Each peer indicates its preference of path usage independently of the\n   other peer.  That means that peers could have different usage\n   preferences for the same path.  Depending on the data sender's\n   decisions, this might lead to usage of paths that have been indicated\n   as \"backup\" by the peer or non-usage of some locally available paths.\n\n   PATH_STATUS_AVAILABLE indicates that a path is \"available\", i.e., it\n   suggests to the peer to use its own logic to split traffic among\n   available paths.\n\n   PATH_STATUS_BACKUP suggests that a path should only be used as\n   backup, i.e., that no traffic should be sent on that path if another\n   path is available and usable.  If all established paths are indicated\n   as backup paths, no guidance is provided about which path should be\n   used.\n\n   Similarly, if no frame indicating a path usage preference was\n   received for a certain path, the preference of the peer is unknown\n   and the sender needs to decide based on its own local logic if the\n   path should be used.\n\n   If an endpoint starts using a backup path because it has detected\n   issues on the paths marked as \"available\", it is RECOMMENDED to\n   update its own path state signaling such that the peer avoids using\n   the broken path.  An endpoint that detects a path breakage can also\n   explicitly close the path by sending a PATH_ABANDON frame (see\n   Section 3.4) in order to avoid that its peer keeps using it and\n   enable faster switchover to a backup path.  If the endpoints do not\n   want to close the path immediately, as connectivity could be re-\n   established, PING frames can potentially be used to quickly detect\n   connectivity changes and switch back in a timely way.\n\n   The PATH_STATUS_AVAILABLE and PATH_STATUS_BACKUP frames share a\n   common, per-path sequence number space to detect and ignore outdated\n   information, as further described in Section 4.3.  This is needed as\n   they might arrive out of order, e.g., if sent using different paths.\n\n3.4.  Path Close\n\n   At any time in the connection, each endpoint can decide to abandon a\n   path, for example, following changes in local connectivity or local\n   preferences.  An endpoint that wants to abandon a path MUST\n   explicitly close the path by sending a PATH_ABANDON frame (see\n   Section 4.2).  This is true whether the decision to close the path\n   results from implicit signals such as an idle timeout (see\n   Section 5.9) or packet losses, as well as for any other reason such\n   as management of local resources.\n\n   Endpoints that send a PATH_ABANDON frame MUST treat all connection\n   IDs received from the peer for the path ID indicated in the\n   PATH_ABANDON as immediately retired and subsequently cannot send any\n   packet on that path anymore.  Note that while abandoning a path will\n   cause connection ID retirement, the inverse is not true: Retiring the\n   associated connection IDs does not indicate path abandonment (see\n   further Section 3.2).\n\n   PATH_ABANDON frames can be sent on any open path, not only on the\n   path that is intended to be closed.  It is RECOMMENDED to send the\n   PATH_ABANDON frames on another open path, especially if connectivity\n   on the to-be-abandoned path is expected to be broken.\n\n   When an endpoint receives a PATH_ABANDON frame, it MUST send a\n   corresponding PATH_ABANDON frame if it has not already done so and\n   respectively treat all connection IDs received from the peer for that\n   path as immediately retired.  While that means retired connection IDs\n   received from the peer cannot be used for sending anymore, packets\n   from the peer might still be in transit.  Therefore, knowledge of the\n   connection IDs issued to the peer and of the state of the number\n   space associated to the path SHOULD be retained for 3 PTOs after the\n   PATH_ABANDON frame has been received.  This avoids generating\n   spurious Stateless Reset packets, as discussed in Section 3.4.2, and\n   helps acknowledge any potentially reordered, outstanding packets from\n   the peer (see Section 3.4.3).\n\n   It is also possible that an endpoint will receive a PATH_ABANDON\n   frame before receiving or sending any traffic on a path.  For\n   example, if the client tries to initiate a path and the path cannot\n   be established, it will send a PATH_ABANDON frame (see Section 3.1).\n   An endpoint could also decide to abandon an unused path for any other\n   reason, for example, removing a hole from the sequence of path IDs in\n   use.  This is not an error.\n\n   If a peer sends a PATH_ABANDON frame but never receives a\n   corresponding PATH_ABANDON frame, it might not be able to remove path\n   state.  It is left to the implementation to handle this unexpected\n   behavior, as it does not impact interoperability.  If the endpoint is\n   no longer willing to process the issued connection IDs for the\n   abandoned path, it MAY close the connection but SHOULD wait at least\n   3 PTOs after sending the PATH_ABANDON frame.\n\n   After a path is abandoned, the path ID MUST NOT be reused for new\n   paths, as the path ID is part of the nonce calculation (see\n   Section 2.4).\n\n   If a PATH_ABANDON frame is received for the only open path of a QUIC\n   connection, the receiving peer SHOULD send a CONNECTION_CLOSE frame\n   and enter the closing state.  Alternatively, a client MAY instead try\n   to open a new path, if available, and only initiate connection\n   closure if path validation fails or a CONNECTION_CLOSE frame is\n   received from the server.  Similarly, the server MAY wait for a\n   short, limited time, such as one PTO, to see if a packet is received\n   on a new path before sending the CONNECTION_CLOSE frame.\n\n   Note that other explicit closing mechanisms of [QUIC-TRANSPORT] still\n   apply on the whole connection.  In particular, the reception of\n   either a CONNECTION_CLOSE (Section 10.2 of [QUIC-TRANSPORT]) or a\n   Stateless Reset (Section 10.3 of [QUIC-TRANSPORT]) closes the\n   connection.\n\n3.4.1.  Path Closure Example\n\n   In the example below, the client wants to close the path with path ID\n   0.  It sends the PATH_ABANDON frame to terminate the path with path\n   ID 0 on the path with path ID 1 using the connection ID S1.  After\n   receiving the PATH_ABANDON frame for path ID 0, the server also sends\n   a PATH_ABANDON frame with path ID 0 together with a PATH_ACK frame on\n   the same path using connection ID C1.\n\n   Client                                                      Server\n\n   (client tells server to abandon a path with path ID 0)\n   1-RTT[X]: DCID=S1 PATH_ABANDON[path ID=0]->\n                              (server tells client to abandon a path)\n                       <-1-RTT[Y]: DCID=C1 PATH_ABANDON[path ID=0],\n                                              PATH_ACK[PATH ID=1, PN=X]\n   1-RTT[U]: DCID=S1 PATH_ACK[path ID=1, PN=Y] ->\n\n                    Figure 5: Example of Closing a Path\n\n   Note that if the PATH_ABANDON frame is instead sent on the to-be-\n   abandoned path, the last acknowledgment still needs to be sent on a\n   different path as no further packets can be sent on the abandoned\n   path after the PATH_ABANDON frame.\n\n3.4.2.  Avoiding Spurious Stateless Resets\n\n   Due to network delays, packets sent on an abandoned path can arrive\n   well after the connection IDs have been retired.  If not recognized\n   as bound to the local connection, such a packet triggers the peer to\n   send a Stateless Reset packet.  The requirement to retain knowledge\n   of connection ID and about the packet number space for 3 PTOs after\n   receiving a PATH_ABANDON frame, as specified in Section 3.4 above, is\n   intended to reduce the risk of sending such spurious stateless\n   packets, but it cannot completely avoid that risk.\n\n   Section 10.3 of [QUIC-TRANSPORT] specified that the Stateless Reset\n   Tokens associated with retired connection IDs cannot be used to\n   identify Stateless Reset packets.  The immediate retirement of\n   connection IDs received from the peer for an abandoned path\n   guarantees that spurious Stateless Reset packets sent by the peer\n   will not cause the closure of the QUIC connection.\n\n3.4.3.  Handling PATH_ACK for Abandoned Paths\n\n   When an endpoint sends a PATH_ABANDON frame, there might still be\n   some packets in transit from the peer.  Further, if an endpoint\n   receives a PATH_ABANDON frame, it might still receive reordered\n   packets on the abandoned path.  Endpoints SHOULD promptly send\n   PATH_ACK frames for all unacknowledged packets received on an\n   abandoned path if path state is still retained to do so.\n\n   PATH_ACK frames have to be sent on a different path than the path\n   being abandoned after sending the PATH_ABANDON frame as connection\n   IDs are immediately retired.\n\n   When an endpoint finally deletes all state associated with the path,\n   the packets sent over the path and not yet acknowledged MUST be\n   considered lost.  PATH_ACK frames received with an abandoned path ID\n   are silently ignored, as specified in Section 4.\n\n4.  New Frames\n\n   All frames defined in this document MUST only be sent in 1-RTT\n   packets.  If an endpoint receives a multipath-specific frame in a\n   different packet type, it MUST close the connection with an error of\n   type PROTOCOL_VIOLATION.\n\n   Receipt of multipath-specific frames that use a path ID that is\n   greater than the announced Maximum Paths value in the MAX_PATH_ID\n   frame or in the initial_max_path_id transport parameter, if no\n   MAX_PATH_ID frame was received yet, MUST be treated as a connection\n   error of type PROTOCOL_VIOLATION.\n\n   If an endpoint receives a multipath-specific frame with a path ID\n   that it cannot process anymore (e.g., because the path might have\n   been abandoned), it MUST silently ignore the frame.\n\n4.1.  PATH_ACK Frame\n\n   The PATH_ACK frame (types 0x3e and 0x3f) is an extension of the ACK\n   frame specified in Section 19.3 of [QUIC-TRANSPORT].  It is used to\n   acknowledge packets that were sent on different paths, as each path\n   has its own packet number space.  If the frame type is 0x3f, PATH_ACK\n   frames also contain the sum of QUIC packets with associated Explicit\n   Congestion Notification (ECN) marks received on the acknowledged\n   packet number space up to this point.\n\n   PATH_ACK frames are formatted as shown in Figure 6.\n\n     PATH_ACK Frame {\n       Type (i) = 0x3e..0x3f,\n       Path Identifier (i),\n       Largest Acknowledged (i),\n       ACK Delay (i),\n       ACK Range Count (i),\n       First ACK Range (i),\n       ACK Range (..) ...,\n       [ECN Counts (..)],\n     }\n\n                      Figure 6: PATH_ACK Frame Format\n\n   Compared to the ACK frame specified in [QUIC-TRANSPORT], the\n   following field is added:\n\n   Path Identifier:  The path ID associated with the packet number space\n      of the 0-RTT and 1-RTT packets that are acknowledged by the\n      PATH_ACK frame.\n\n4.2.  PATH_ABANDON Frame\n\n   The PATH_ABANDON frame informs the peer to abandon a path.  After the\n   PATH_ABANDON frame is sent on a path, the path can no longer be used\n   for sending.\n\n   PATH_ABANDON frames are formatted as shown in Figure 7.\n\n     PATH_ABANDON Frame {\n       Type (i) = 0x3e75,\n       Path Identifier (i),\n       Error Code (i),\n     }\n\n                    Figure 7: PATH_ABANDON Frame Format\n\n   PATH_ABANDON frames contain the following fields:\n\n   Path Identifier:  The path ID associated to the to-be-abandoned path.\n\n   Error Code:  A variable-length integer that indicates the reason for\n      abandoning this path.  NO_ERROR(0x0) indicates that the path is\n      being abandoned without any error being encountered.  Other error\n      codes can be found in Section 4.2.1.\n\n   PATH_ABANDON frames are ack-eliciting.  If a packet containing a\n   PATH_ABANDON frame is considered lost, the peer SHOULD repeat it.\n\n   Use of the PATH_ABANDON frame is specified in Section 3.4.\n\n4.2.1.  Error Codes\n\n   QUIC transport error codes are 62-bit unsigned integers (see\n   Section 20.1 of [QUIC-TRANSPORT]).  In addition to NO_ERROR(0x0), the\n   following QUIC error codes are defined for use in the PATH_ABANDON\n   frame:\n\n   APPLICATION_ABANDON_PATH (0x3e):  The endpoint is abandoning the path\n      at the request of the application.\n\n   PATH_RESOURCE_LIMIT_REACHED (0x3e75):  The endpoint is abandoning the\n      path because it cannot allocate sufficient resources to maintain\n      it.\n\n   PATH_UNSTABLE_OR_POOR (0x3e76):  The endpoint is abandoning the path\n      because the used interface is observed to be unstable or\n      performance is considered poor.  This condition can occur, e.g.,\n      due to frequent handover events during high-speed mobility or due\n      to a weak wireless signal.\n\n   NO_CID_AVAILABLE_FOR_PATH (0x3e77):  The endpoint is abandoning the\n      path due to the lack of a connection ID for this path.  This might\n      occur when the peer initiates a new path but has not provided a\n      corresponding connection ID for the path ID (or the packet\n      containing the connection IDs has not arrived yet).\n\n4.3.  PATH_STATUS_AVAILABLE and PATH_STATUS_BACKUP Frames\n\n   PATH_STATUS_AVAILABLE frames (type=0x3e77) are used by endpoints to\n   inform the peer that the indicated path is available for sending.\n\n   PATH_STATUS_AVAILABLE frames are formatted as shown in Figure 8.\n\n     PATH_STATUS_AVAILABLE Frame {\n       Type (i) = 0x3e77,\n       Path Identifier (i),\n       Path Status Sequence Number (i),\n     }\n\n                Figure 8: PATH_STATUS_AVAILABLE Frame Format\n\n   PATH_STATUS_BACKUP frames (type=0x3e76) are used by endpoints to\n   inform the peer about its preference to not use the indicated path\n   for sending.\n\n   PATH_STATUS_BACKUP frames are formatted as shown in Figure 9.\n\n     PATH_STATUS_BACKUP Frame {\n       Type (i) = 0x3e76\n       Path Identifier (i),\n       Path Status Sequence Number (i),\n     }\n\n                 Figure 9: PATH_STATUS_BACKUP Frame Format\n\n   Both PATH_STATUS_AVAILABLE and PATH_STATUS_BACKUP frames contain the\n   following fields:\n\n   Path Identifier:  The path ID that the status update corresponds to.\n      All path IDs below the maximum path ID limit can be indicated,\n      even if the path is not in active use yet.\n\n   Path Status Sequence Number:  A variable-length integer specifying\n      the per-path sequence number assigned for this frame.\n\n   The sequence number space is common to the two frame types, and\n   monotonically increasing values MUST be used when sending\n   PATH_STATUS_AVAILABLE or PATH_STATUS_BACKUP frames for a given path\n   ID.\n\n   Frames might be received out of order.  A peer MUST ignore an\n   incoming PATH_STATUS_AVAILABLE or PATH_STATUS_BACKUP frame if it\n   previously received another PATH_STATUS_BACKUP frame or\n   PATH_STATUS_AVAILABLE frame for the same path ID with a Path Status\n   Sequence Number equal to or higher than the Path Status Sequence\n   Number of the incoming frame.\n\n   The requirement of monotonically increasing sequence numbers is per\n   path.  Receivers could very well receive the same sequence number for\n   PATH_STATUS_AVAILABLE or PATH_STATUS_BACKUP frames on different\n   paths.  As such, the receiver of the PATH_STATUS_AVAILABLE or\n   PATH_STATUS_BACKUP frame needs to use and compare the sequence\n   numbers separately for each path ID.\n\n   PATH_STATUS_BACKUP and PATH_STATUS_AVAILABLE frames are ack-\n   eliciting.  If a packet containing a PATH_STATUS_BACKUP or\n   PATH_STATUS_AVAILABLE frame is considered lost, the peer SHOULD\n   resend the frame only if it contains the last status sent for that\n   path -- as indicated by the sequence number.\n\n   A PATH_STATUS_BACKUP or a PATH_STATUS_AVAILABLE frame MAY be bundled\n   with a PATH_NEW_CONNECTION_ID frame or a PATH_RESPONSE frame in order\n   to indicate the preferred path usage before or during path\n   initiation.\n\n4.4.  PATH_NEW_CONNECTION_ID Frame\n\n   The PATH_NEW_CONNECTION_ID frame (type=0x3e78) is an extension of the\n   NEW_CONNECTION_ID frame specified in Section 19.15 of\n   [QUIC-TRANSPORT].  It is used to provide its peer with alternative\n   connection IDs for 1-RTT packets for a specific path.  The peer can\n   then use a different connection ID on the same path to break\n   linkability when migrating on that path; see also Section 9.5 of\n   [QUIC-TRANSPORT].\n\n   PATH_NEW_CONNECTION_ID frames are formatted as shown in Figure 10.\n\n   PATH_NEW_CONNECTION_ID Frame {\n     Type (i) = 0x3e78,\n     Path Identifier (i),\n     Sequence Number (i),\n     Retire Prior To (i),\n     Length (8),\n     Connection ID (8..160),\n     Stateless Reset Token (128),\n   }\n\n               Figure 10: PATH_NEW_CONNECTION_ID Frame Format\n\n   Compared to the NEW_CONNECTION_ID frame specified in Section 19.15 of\n   [QUIC-TRANSPORT], the following field is added:\n\n   Path Identifier:  The path ID associated with the connection ID.\n      This means the provided connection ID can only be used on the\n      corresponding path.\n\n   Note that, other than for the NEW_CONNECTION_ID frame in\n   Section 19.15 of [QUIC-TRANSPORT], the sequence number applies on a\n   per-path context.  This means different connection IDs on different\n   paths might have the same sequence number value.\n\n   The Retire Prior To field indicates which connection IDs should be\n   retired among those that share the path ID in the Path Identifier\n   field.  Connection IDs associated with different path IDs are not\n   affected.\n\n   Note that the NEW_CONNECTION_ID frame can only be used to issue or\n   retire connection IDs for the initial path with path ID 0.\n\n   The last paragraph of Section 5.1.2 of [QUIC-TRANSPORT] specifies how\n   to verify the Retire Prior To field of an incoming NEW_CONNECTION_ID\n   frame.  The same rule applies for PATH_NEW_CONNECTION_ID frames, but\n   it applies per path.  If the multipath extension is used, the rule\n   for the NEW_CONNECTION_ID frame is only applied for path ID 0.\n\n4.5.  PATH_RETIRE_CONNECTION_ID Frame\n\n   The PATH_RETIRE_CONNECTION_ID frame (type=0x3e79) is an extension of\n   the RETIRE_CONNECTION_ID frame specified in Section 19.16 of\n   [QUIC-TRANSPORT].  It is used to indicate that an endpoint will no\n   longer use a connection ID for a specific path ID that was issued by\n   its peer.  To retire the connection ID used during the handshake on\n   the initial path, path ID 0 is used.  Sending a\n   PATH_RETIRE_CONNECTION_ID frame also serves as a request to the peer\n   to send additional connection IDs for this path (see also Section 5.1\n   of [QUIC-TRANSPORT]), unless the path specified by the path ID has\n   been abandoned.  New path-specific connection IDs can be delivered to\n   a peer using the PATH_NEW_CONNECTION_ID frame (see Section 4.4).\n\n   PATH_RETIRE_CONNECTION_ID frames are formatted as shown in Figure 11.\n\n   PATH_RETIRE_CONNECTION_ID Frame {\n     Type (i) = 0x3e79,\n     Path Identifier (i),\n     Sequence Number (i),\n   }\n\n             Figure 11: PATH_RETIRE_CONNECTION_ID Frame Format\n\n   Compared to the RETIRE_CONNECTION_ID frame specified in Section 19.16\n   of [QUIC-TRANSPORT], the following field is added:\n\n   Path Identifier:  The path ID associated with the connection ID to\n      retire.\n\n   Note that the RETIRE_CONNECTION_ID frame can only be used to retire\n   connection IDs for the initial path with path ID 0.\n\n   As the PATH_NEW_CONNECTION_ID frame applies the sequence number per\n   path, the sequence number in the PATH_RETIRE_CONNECTION_ID frame is\n   also per path.  The PATH_RETIRE_CONNECTION_ID frame retires the\n   Connection ID with the specified path ID and sequence number.\n\n   The processing of an incoming RETIRE_CONNECTION_ID frame is described\n   in Section 19.16 of [QUIC-TRANSPORT].  The same processing applies\n   for PATH_RETIRE_CONNECTION_ID frames per path, while with use of the\n   multipath extension, the processing of a RETIRE_CONNECTION_ID frame\n   is only applied for path ID 0.\n\n4.6.  MAX_PATH_ID Frame\n\n   A MAX_PATH_ID frame (type=0x3e7a) informs the peer of the maximum\n   path ID it is permitted to use.\n\n   MAX_PATH_ID frames are formatted as shown in Figure 12.\n\n   MAX_PATH_ID Frame {\n     Type (i) = 0x3e7a,\n     Maximum Path Identifier (i),\n   }\n\n                    Figure 12: MAX_PATH_ID Frame Format\n\n   MAX_PATH_ID frames contain the following field:\n\n   Maximum Path Identifier:  The maximum path ID that the sending\n      endpoint is willing to accept.  This value MUST NOT exceed 2^32-1,\n      which is the maximum allowed value for the path ID due to\n      restrictions on the nonce calculation (see Section 2.4).  The\n      Maximum Path Identifier value MUST NOT be lower than the value\n      advertised in the initial_max_path_id transport parameter.\n\n   Receipt of an invalid Maximum Path Identifier value MUST be treated\n   as a connection error of type PROTOCOL_VIOLATION.\n\n   Loss or reordering can cause an endpoint to receive a MAX_PATH_ID\n   frame with a smaller Maximum Path Identifier value than was\n   previously received.  MAX_PATH_ID frames that do not increase the\n   path limit MUST be ignored.\n\n   MAX_PATH_ID frames are ack-eliciting and SHOULD be retransmitted when\n   lost and no more recent MAX_PATH_ID frame has been sent in the\n   meantime.\n\n4.7.  PATHS_BLOCKED and PATH_CIDS_BLOCKED Frames\n\n   A sender can send a PATHS_BLOCKED frame (type=0x3e7b) when it wishes\n   to open a path but is unable to do so due to the maximum path ID\n   limit set by its peer.\n\n   A sender can send a PATH_CIDS_BLOCKED frame (type=0x3e7c) when it\n   wishes to open a path with a valid path ID or change the connection\n   ID on an established path but is unable to do so because there are no\n   unused connection IDs available for the corresponding path ID.\n\n   Note that PATHS_BLOCKED and PATH_CIDS_BLOCKED frames are\n   informational.  Sending a PATHS_BLOCKED or a PATH_CIDS_BLOCKED frame\n   does not imply a particular action from the peer like sending a\n   MAX_PATH_ID frame with a new Maximum Path Identifier value but\n   informs the peer that the maximum path ID limit or the absence of\n   unused connection IDs prevented the creation or the usage of paths.\n   If the successful reception of a PATHS_BLOCKED/PATH_CIDS_BLOCKED\n   frame was acknowledged, but no action is taken by the peer, this is\n   likely a deliberate decision by the peer and repeating the\n   PATHS_BLOCKED/PATH_CIDS_BLOCKED frame will not change that.\n\n   PATHS_BLOCKED frames are formatted as shown in Figure 13.\n\n   PATHS_BLOCKED Frame {\n     Type (i) = 0x3e7b,\n     Maximum Path Identifier (i),\n   }\n\n                   Figure 13: PATHS_BLOCKED Frame Format\n\n   PATHS_BLOCKED frames contain the following field:\n\n   Maximum Path Identifier:  A variable-length integer indicating the\n      maximum path ID that was allowed at the time the frame was sent.\n      If the received value is lower than the currently allowed maximum\n      value, this frame can be ignored.\n\n   PATH_CIDS_BLOCKED frames are formatted as shown in Figure 14.\n\n   PATH_CIDS_BLOCKED Frame {\n     Type (i) = 0x3e7c,\n     Path Identifier (i),\n     Next Sequence Number (i),\n   }\n\n                 Figure 14: PATH_CIDS_BLOCKED Frame Format\n\n   PATH_CIDS_BLOCKED frames contain the following fields:\n\n   Path Identifier:  Identifier of the path for which unused connection\n      IDs are not available.\n\n   Next Sequence Number:  The next sequence number that is expected to\n      be issued for a connection ID for this path by the peer.\n\n   Receipt of a value of Maximum Path Identifier or Path Identifier that\n   is higher than the local maximum value MUST be treated as a\n   connection error of type PROTOCOL_VIOLATION.\n\n   Receipt of a value of Next Sequence Number that is higher than the\n   sequence number of the next expected to be issued connection ID for\n   this path MUST be treated as a connection error of type\n   PROTOCOL_VIOLATION.\n\n   PATHS_BLOCKED and PATH_CIDS_BLOCKED frames are ack-eliciting and MAY\n   be retransmitted if the path is still blocked when the loss is\n   detected.\n\n5.  Implementation Considerations\n\n   This section provides informational guidance for implementors.\n\n5.1.  Connection ID Changes, Migration, and NAT Rebindings\n\n   With the multipath extension, each path uses a separate packet number\n   space.  This is a major difference from [QUIC-TRANSPORT], which only\n   defines three number spaces (Initial, Handshake, and Application data\n   packets).\n\n   For any given path, connection ID rotation, NAT rebinding, or client-\n   initiated migration as specified in [QUIC-TRANSPORT] might occur like\n   on a single path.  These events do not change the path ID and do not\n   affect the packet number space associated with the path.\n\n   It is generally preferable to use multipath mechanisms, such as\n   creating a new path and later abandoning the old path rather than\n   doing migration of a single path as specified in [QUIC-TRANSPORT].\n   This enables a smoother handover and allows a more controlled\n   migration handling at the server side.  However, migration of a\n   single path cannot be avoided in case of NAT rebinding or if the\n   server requests migration to a \"preferred address\" during the\n   handshake.\n\n   Section 9.3 of [QUIC-TRANSPORT] allows an endpoint to skip validation\n   of a peer address if that address has been seen recently.  However,\n   when the multipath extension is used and an endpoint has multiple\n   addresses that could lead to switching between different paths, it\n   should rather maintain multiple open paths instead.\n\n   Servers observing a 4-tuple change will perform path validation (see\n   Section 9 of [QUIC-TRANSPORT]).  If path validation process succeeds,\n   the endpoints reset the path's congestion controller and round-trip\n   time estimator according to Section 9.4 of [QUIC-TRANSPORT].\n\n5.2.  Using Multiple Paths on the Same 4-tuple\n\n   It is possible to create paths that refer to the same 4-tuple.  For\n   example, endpoints might want to create paths that use different\n   Differentiated Service [RFC2475] markings.  This could be done in\n   conjunction with scheduling algorithms that match streams to paths so\n   that, for example, data frames for low priority streams are sent over\n   low priority paths.  Since these paths use different path IDs, they\n   can be managed independently to suit the needs of the application.\n   (The application would need to manage how client and server use\n   differentiated services on a path.  This is not specified in this\n   document.)\n\n   There might be cases in which paths are created with different\n   4-tuples but end up using the same 4-tuples as a consequence of path\n   migrations.  Consider the following example where all paths use the\n   same source and destination ports:\n\n   *  Client starts path 1 from address 192.0.2.1 to server address\n      198.51.100.1.\n\n   *  Client starts path 2 from address 192.0.2.2 to server address\n      198.51.100.1.\n\n   *  Both paths are used for a while.\n\n   *  Server sends packet from address 198.51.100.1 to client address\n      192.0.2.1 with Connection ID indicating path ID 2.\n\n   *  Client receives the packet, recognizes a path migration, and\n      updates the source address of path 2 to 192.0.2.1.\n\n   (For simplicity, this example uses IPv4 addresses but would work\n   similarly for IPv6 addresses.)\n\n   Such unintentional use of the same 4-tuple on different paths ought\n   to be rare.  When they happen, the two paths would be redundant, and\n   the endpoint could want to close one of them.\n\n5.3.  Congestion Control\n\n   When the QUIC multipath extension is used, senders manage per-path\n   congestion status as required in Section 9.4 of [QUIC-TRANSPORT].\n   However, in [QUIC-TRANSPORT], only one active path is assumed and, as\n   such, the requirement is to reset the congestion control status on\n   path migration.  With the multipath extension, multiple paths can be\n   used simultaneously.  Therefore, separate congestion control state is\n   maintained for each path.  This means a sender is not allowed to send\n   more data on a given path than congestion control for that path\n   indicates.\n\n   When a Multipath QUIC connection uses two or more paths, there is no\n   guarantee that these paths are fully disjoint.  When two (or more\n   paths) share the same bottleneck, using a standard congestion control\n   scheme could result in an unfair distribution of the bandwidth with\n   the multipath connection getting more bandwidth than competing single\n   paths connections.  Multipath TCP uses the Linked Increases Algorithm\n   (LIA) congestion control scheme specified in [RFC6356] to solve this\n   problem.  This scheme can immediately be adapted to Multipath QUIC.\n   Other coupled congestion control schemes have been proposed for\n   Multipath TCP, such as [OLIA].  Designers of congestion control\n   algorithms specialized for Multipath QUIC are advised to follow BCP\n   133; see Section 7.10 of [RFC9743].\n\n   Section 5.1.2 of [QUIC-TRANSPORT] indicates that an endpoint can\n   change the connection ID it uses to another available one at any time\n   during the connection.  As such, a change of the Connection ID\n   without any change in the address does not indicate a path change and\n   the endpoint can keep the same congestion control and RTT measurement\n   state.\n\n5.4.  Computing Path RTT\n\n   PATH_ACK frames indicate which path the acknowledged packets were\n   sent on, but they could be received through any open path.  If\n   successive acknowledgments are received on different paths, the\n   measured RTT samples can fluctuate widely, which could result in poor\n   performance depending on, for example, the congestion control\n   algorithm.\n\n   Congestion control state as defined in [QUIC-RECOVERY] is kept per\n   path ID.  However, depending on which path acknowledgments are sent,\n   the actual RTT of a path cannot be calculated or might not be the\n   right value to be used.\n\n   Instead of using the real RTT of a path, it is recommended to\n   consider the sum of two one-way delays: the delay on the packet\n   sending path and the delay on the return path chosen for the\n   acknowledgments.  When different paths have different\n   characteristics, the delays can vary widely.  For example, consider a\n   multipath transmission using both a terrestrial path with a latency\n   of 50ms in each direction and a geostationary satellite path with a\n   latency of 300ms in each direction.  The sum of the two one-way\n   delays will depend on the combination of paths used for the packet\n   transmission and the acknowledgment transmission, as shown in\n   Table 1.\n\n            +======================+=============+===========+\n            | ACK Path \\ Data Path | Terrestrial | Satellite |\n            +======================+=============+===========+\n            | Terrestrial          | 100ms       | 350ms     |\n            +----------------------+-------------+-----------+\n            | Satellite            | 350ms       | 600ms     |\n            +----------------------+-------------+-----------+\n\n              Table 1: Example of ACK Delays Using Multiple\n                                  Paths\n\n   The computed values reflect both the state of the network path and\n   the scheduling decisions of the acknowledgment sender.  If we assume\n   that the PATH_ACK will be sent over the terrestrial link, because\n   this decision provides the best response time, the computed RTT value\n   for the satellite path will be about 350ms.  This is lower than the\n   600ms that would be measured if the PATH_ACK came over the satellite\n   channel, but it is still the right value for computing, for example,\n   the PTO timeout: If a PATH_ACK is not received after more than 350ms,\n   either the packet or its PATH_ACK were probably lost.\n\n   The simplest implementation is to use the delays measured when\n   receiving new packet acknowledgments to compute smoothed_rtt and\n   rttvar per Section 5.3 of [QUIC-RECOVERY] regardless of the path\n   through which PATH_ACK frames are received.  This approach will\n   provide good results as long as acknowledgments are sent consistently\n   over one path.  If at any time the acknowledgment sender revisits its\n   sending preferences, this can also change the paths that are used to\n   send acknowledgments.  However, this is not very different from route\n   changes on a single path.  The RTT, RTT variance, and PTO estimates\n   will rapidly converge to reflect the new conditions.  There is one\n   exception: the minimum RTT, which is also a known challenge when\n   route changes occurs on a single path.  An acknowledgment receiver\n   can, however, remember the path over which the PATH_ACK that produced\n   the minimum RTT was received and restart the minimum RTT computation\n   if that acknowledgment path changes or is abandoned.  If\n   acknowledgments are not sent consistently over one path, the\n   acknowledgment receiver can monitor over which path acknowledgments\n   are received and only use samples for acknowledgments received on the\n   same path on which the data was sent, if any.\n\n   Further, congestion control functions that rely on delay estimates\n   need to consider cases where acknowledgments are sent over multiple\n   paths with different delays explicitly.\n\n5.5.  Packet Scheduling\n\n   The transmission of packets containing data is limited by the arrival\n   of data from the application and by congestion control.  Generally,\n   QUIC packets that increase the number of bytes in flight can only be\n   sent when the congestion window for the selected path allows it.\n\n   Most frames, including control frames (PATH_CHALLENGE and\n   PATH_RESPONSE being the notable exceptions), can be sent and received\n   on any open path.  As such, a packet scheduler is needed to decide\n   which path to use for sending the next packet, among those paths with\n   an open congestion window.  If multiple paths are used to send data\n   frames belonging to the same stream, data delivery will experience\n   the maximum delay of all used paths due to in-order delivery.  The\n   scheduling is a local decision based on the preferences of the\n   application and the implementation.\n\n   This implies that an endpoint might send and receive PATH_ACK frames\n   on a path different from the one that carried the acknowledged\n   packets.  As noted in Section 5.4, RTT estimates computed using the\n   standard algorithm reflect both the characteristics of the path and\n   the scheduling algorithm of PATH_ACK frames.  The estimates will\n   converge faster if the scheduling strategy of PATH_ACK frames is\n   stable.  Implementations can choose different strategies such as, for\n   instance, sending PATH_ACK frames either simply on the path where the\n   acknowledged packets were received or alternatively the shortest\n   path, which results in shorter control loops and potentially better\n   performance.\n\n   Since packets that only carry PATH_ACK frames are not congestion\n   controlled (see Section 7 of [QUIC-RECOVERY]), senders should\n   carefully consider the load induced by these packets, especially if\n   the capacity is unknown on that path, e.g., when that path is not\n   used for sending data frames.\n\n5.6.  Retransmissions\n\n   Simultaneous use of multiple paths enables different retransmission\n   strategies to cope with losses such as a) retransmitting lost frames\n   over the same path, b) retransmitting lost frames on a different or\n   dedicated path, and c) duplicate lost frames on several paths (not\n   recommended for general purpose use due to the network overhead).\n   While this document does not preclude a specific strategy, more\n   detailed specification is out of scope.\n\n   As noted in Section 2.2 of [QUIC-TRANSPORT], STREAM frame boundaries\n   are not expected to be preserved when data is retransmitted.\n   Especially when STREAM frames have to be retransmitted over a\n   different path with a smaller MTU limit, smaller STREAM frames might\n   need to be sent instead.\n\n5.7.  PTO Expiration\n\n   An implementation should follow the mechanism specified in\n   [QUIC-RECOVERY] for detecting packet loss on each individual path.  A\n   special case happens when the PTO timer expires.  According to\n   [QUIC-RECOVERY], no packet will be declared lost until either the\n   packet sender receives a new acknowledgment for this path or the path\n   itself is finally declared broken.  This cautious process minimizes\n   the risk of spurious retransmissions, but it might cause significant\n   delivery delay for the frames contained in these \"lost packets\".\n\n   Endpoints could take advantage of the multipath extension and\n   retransmit the content of the delayed packets on other available\n   paths if the congestion control window on these paths allows.\n\n5.8.  Paths Having Different PMTU Sizes\n\n   An implementation should take care to handle different Path MTU\n   (PMTU) sizes across multiple paths.  As specified in Section 14.3 of\n   [QUIC-TRANSPORT], the Datagram Packetization Layer PMTU Discovery\n   (DPLPMTUD) Maximum Packet Size (MPS) is maintained for each\n   combination of local and remote IP addresses.  Note that with the\n   multipath extension, multiple paths could use the same 4-tuple but\n   might have different MPS due to other factors (see Section 5.2).\n\n   One simple option, if the PMTUs are similar, is to apply the minimum\n   PMTU of all paths to each path, which could also help to simplify\n   retransmission processing.\n\n5.9.  Idle Timeout and Keep-Alives\n\n   [QUIC-TRANSPORT] defines an idle timeout for closing the connection,\n   which applies in case of multipath usage if no packet is received on\n   any path for the duration of the idle timeout.\n\n   This document does not specify per-path idle timeouts.  An endpoint\n   can decide to close a path at any time, whether the path is in active\n   use or not.  For example, an endpoint might wait to send the initial\n   PATH_ABANDON frame until it sends another frame anyway.  Note that\n   the receiver of an initial PATH_ABANDON frame is, however, required\n   to immediately reply (see Section 3.4).\n\n   If a path is not actively used for a while, it might not be usable\n   anymore, e.g., due to middlebox timeouts.  To avoid such path\n   breakage, endpoints can send ack-eliciting packets, such as packets\n   containing PING frames (Section 19.2 of [QUIC-TRANSPORT]) on that\n   path to keep it alive.  However, this specification does not\n   recommend sending keep-alives, as it can create unnecessary overhead,\n   especially if there are other actively used paths.\n\n   Section 5.3 of [QUIC-TRANSPORT] defines an optional keep-alive\n   process.  This process can be applied to each path separately\n   depending on application needs.  Some applications could decide to\n   not keep any not-actively used path alive or keep only one additional\n   path or multiple paths (e.g., for more redundancy) alive.  As\n   discussed in Section 10.1.2 of [QUIC-TRANSPORT], the keep-alive\n   interval needs to incorporate timeouts in middleboxes on the path.\n\n   If a path was not actively used for a while and no keep-alives have\n   been sent, an endpoint can probe it before switching to active use if\n   there are still other paths that are currently usable.\n\n6.  IANA Considerations\n\n   This document defines a new transport parameter to enable\n   simultaneous use of multiple paths within one QUIC connection.\n   Further, it specifies new frame types for path management and new\n   error codes when a path is abandoned.\n\n   IANA has allocated short values with permanent as the status, IETF as\n   the change controller, and the QUIC WG as the contact to the\n   respective registries under <https://www.iana.org/assignments/quic>.\n\n   The following parameter defined in Table 2 has been added to the\n   \"QUIC Transport Parameters\" registry.\n\n              +=======+=====================+===============+\n              | Value | Parameter Name      | Specification |\n              +=======+=====================+===============+\n              | 0x3e  | initial_max_path_id | Section 2.1   |\n              +-------+---------------------+---------------+\n\n                    Table 2: Addition to QUIC Transport\n                             Parameters Entries\n\n   The following frame types defined in Table 3 have been added to the\n   \"QUIC Frame Types\" registry.\n\n         +===========+===========================+===============+\n         | Value     | Frame Type Name           | Specification |\n         +===========+===========================+===============+\n         | 0x3e-0x3f | PATH_ACK                  | Section 4.1   |\n         +-----------+---------------------------+---------------+\n         | 0x3e75    | PATH_ABANDON              | Section 4.2   |\n         +-----------+---------------------------+---------------+\n         | 0x3e76    | PATH_STATUS_BACKUP        | Section 4.3   |\n         +-----------+---------------------------+---------------+\n         | 0x3e77    | PATH_STATUS_AVAILABLE     | Section 4.3   |\n         +-----------+---------------------------+---------------+\n         | 0x3e78    | PATH_NEW_CONNECTION_ID    | Section 4.4   |\n         +-----------+---------------------------+---------------+\n         | 0x3e79    | PATH_RETIRE_CONNECTION_ID | Section 4.5   |\n         +-----------+---------------------------+---------------+\n         | 0x3e7a    | MAX_PATH_ID               | Section 4.6   |\n         +-----------+---------------------------+---------------+\n         | 0x3e7b    | PATHS_BLOCKED             | Section 4.7   |\n         +-----------+---------------------------+---------------+\n         | 0x3e7c    | PATH_CIDS_BLOCKED         | Section 4.7   |\n         +-----------+---------------------------+---------------+\n\n               Table 3: Addition to QUIC Frame Types Entries\n\n   The following transport error codes defined in Table 4 have been\n   added to the \"QUIC Transport Error Codes\" registry.\n\n   +======+=============================+==============+===============+\n   |Value | Code                        |Description   | Specification |\n   +======+=============================+==============+===============+\n   |0x3e  | APPLICATION_ABANDON_PATH    |Path          | Section 4.2.1 |\n   |      |                             |abandoned at  |               |\n   |      |                             |the           |               |\n   |      |                             |application's |               |\n   |      |                             |request       |               |\n   +------+-----------------------------+--------------+---------------+\n   |0x3e75| PATH_RESOURCE_LIMIT_REACHED |Path          | Section 4.2.1 |\n   |      |                             |abandoned due |               |\n   |      |                             |to resource   |               |\n   |      |                             |limitations   |               |\n   |      |                             |in the        |               |\n   |      |                             |transport     |               |\n   +------+-----------------------------+--------------+---------------+\n   |0x3e76| PATH_UNSTABLE_OR_POOR       |Path          | Section 4.2.1 |\n   |      |                             |abandoned due |               |\n   |      |                             |to unstable   |               |\n   |      |                             |interfaces or |               |\n   |      |                             |poor          |               |\n   |      |                             |performance   |               |\n   +------+-----------------------------+--------------+---------------+\n   |0x3e77| NO_CID_AVAILABLE_FOR_PATH   |Path          | Section 4.2.1 |\n   |      |                             |abandoned due |               |\n   |      |                             |to no         |               |\n   |      |                             |available     |               |\n   |      |                             |connection    |               |\n   |      |                             |IDs for the   |               |\n   |      |                             |path          |               |\n   +------+-----------------------------+--------------+---------------+\n\n          Table 4: Addition to QUIC Transport Error Codes Entries\n\n7.  Security Considerations\n\n   The multipath extension retains all security properties of\n   [QUIC-TRANSPORT] and [QUIC-TLS] but requires some additional\n   consideration regarding:\n\n   *  potential additional resource usage for per-path connection IDs\n      and multiple concurrent path contexts,\n\n   *  a potentially increased amplification risk for denial-of-service\n      attacks if multiple paths are used simultaneously, and\n\n   *  changes to the nonce calculation due to the use of multiple packet\n      number spaces.\n\n7.1.  Memory Allocation for Per-Path Resources\n\n   The maximum path ID limit in initial_max_path_id or the MAX_PATH_ID\n   frame limits the number of paths an endpoint is willing to maintain\n   and thereby also limits the associated path resources.  Furthermore,\n   as connection IDs have to be issued by both endpoints for the same\n   path ID before an endpoint can open a path, each endpoint could also\n   control the per-path resource usage by only issuing connection IDs\n   for a limited number of paths.  However, using the maximum path ID\n   limit in initial_max_path_id or the MAX_PATH_ID frame is preferred.\n\n   To avoid unnecessary resource usage that could be exploited in a\n   resource exhaustion attack, endpoints SHOULD allocate additional path\n   resources, such as, e.g., for packet number handling, only after path\n   validation has successfully completed.\n\n7.2.  Denial of Service with Multiple Paths\n\n   Path validation as specified in Section 8.2 of [QUIC-TRANSPORT] for\n   migration is used unchanged for path initiation in this extension.\n   Further, the multipath extension allows for the creation of multiple\n   paths, which means that in addition to the security considerations on\n   source address spoofing outlined in Section 21.5.4 of\n   [QUIC-TRANSPORT], there is a risk of amplified denial-of-service\n   attacks through simultaneous opening or migration of multiple paths.\n   For example, an attacker could set or spoof the 4-tuples used in\n   multiple paths so that packets sent by the server would travel\n   through common network paths in an attempt to overwhelm a target.\n\n   [QUIC-TRANSPORT] only allows the use of one path and the number of\n   concurrent path validation attempts is limited by the number of\n   issued connection IDs.  This extension allows for multiple open paths\n   that could in theory be migrated all at the same time, still limited\n   by the number of issued connection IDs for each path.  Further,\n   multiple paths could be initialized simultaneously, limited by the\n   maximum number of allowed paths.\n\n   Therefore, endpoints are advised to keep the maximum number of paths\n   small and might consider additional measures to limit the number of\n   concurrent path validation processes, e.g., by pacing them out or\n   limiting the number of path initiation attempts over a certain time\n   period.\n\n   However, the use of multipath does not change the anti-amplification\n   limits as specified in Section 8 of [QUIC-TRANSPORT].  The attacker\n   would need to send a separate challenge on each path used in the\n   attack, and the amplification limits would apply to the amount of\n   data sent by the server on that path.  The attacker could get the\n   same effect by opening many QUIC connections and conducting the\n   attack on each of them without negotiating the multipath option.\n\n7.3.  Cryptographic Handshake and AEAD Nonce\n\n   The multipath extension as specified in this document is only enabled\n   after a successful handshake when both endpoints indicate support for\n   this extension.  All new frames defined in this extension are only\n   used in 1-RTT packets.\n\n   As the handshake is not changed by this extension, the transport\n   security mechanisms as specified in [QUIC-TLS], such as encryption\n   key exchange and peer authentication, remain unchanged.  As such, the\n   security considerations in [QUIC-TLS] apply unaltered.\n\n   The limits as discussed on Appendix B of [QUIC-TLS] apply to the\n   total number of packets sent on all paths, not each path separately.\n\n   This specification changes the AEAD nonce calculation by including\n   the path ID as part of the calculation (see Section 2.4).  To ensure\n   unique nonces, path IDs are limited to 32 bits and cannot be reused\n   for another path of the same connection.\n\n8.  References\n\n8.1.  Normative References\n\n   [QUIC-RECOVERY]\n              Iyengar, J., Ed. and I. Swett, Ed., \"QUIC Loss Detection\n              and Congestion Control\", RFC 9002, DOI 10.17487/RFC9002,\n              May 2021, <https://www.rfc-editor.org/info/rfc9002>.\n\n   [QUIC-TLS] Thomson, M., Ed. and S. Turner, Ed., \"Using TLS to Secure\n              QUIC\", RFC 9001, DOI 10.17487/RFC9001, May 2021,\n              <https://www.rfc-editor.org/info/rfc9001>.\n\n   [QUIC-TRANSPORT]\n              Iyengar, J., Ed. and M. Thomson, Ed., \"QUIC: A UDP-Based\n              Multiplexed and Secure Transport\", RFC 9000,\n              DOI 10.17487/RFC9000, May 2021,\n              <https://www.rfc-editor.org/info/rfc9000>.\n\n   [RFC2119]  Bradner, S., \"Key words for use in RFCs to Indicate\n              Requirement Levels\", BCP 14, RFC 2119,\n              DOI 10.17487/RFC2119, March 1997,\n              <https://www.rfc-editor.org/info/rfc2119>.\n\n   [RFC8174]  Leiba, B., \"Ambiguity of Uppercase vs Lowercase in RFC\n              2119 Key Words\", BCP 14, RFC 8174, DOI 10.17487/RFC8174,\n              May 2017, <https://www.rfc-editor.org/info/rfc8174>.\n\n8.2.  Informative References\n\n   [OLIA]     Khalili, R., Gast, N., Popovic, M., Upadhyay, U., and J.\n              Le Boudec, \"MPTCP is not pareto-optimal: performance\n              issues and a possible solution\", CoNEXT '12: Proceedings\n              of the 8th international conference on Emerging networking\n              experiments and technologies, pp. 1-12,\n              DOI 10.1145/2413176.2413178, December 2012,\n              <https://doi.org/10.1145/2413176.2413178>.\n\n   [RFC2475]  Blake, S., Black, D., Carlson, M., Davies, E., Wang, Z.,\n              and W. Weiss, \"An Architecture for Differentiated\n              Services\", RFC 2475, DOI 10.17487/RFC2475, December 1998,\n              <https://www.rfc-editor.org/info/rfc2475>.\n\n   [RFC6356]  Raiciu, C., Handley, M., and D. Wischik, \"Coupled\n              Congestion Control for Multipath Transport Protocols\",\n              RFC 6356, DOI 10.17487/RFC6356, October 2011,\n              <https://www.rfc-editor.org/info/rfc6356>.\n\n   [RFC9312]  Kühlewind, M. and B. Trammell, \"Manageability of the QUIC\n              Transport Protocol\", RFC 9312, DOI 10.17487/RFC9312,\n              September 2022, <https://www.rfc-editor.org/info/rfc9312>.\n\n   [RFC9743]  Duke, M., Ed. and G. Fairhurst, Ed., \"Specifying New\n              Congestion Control Algorithms\", BCP 133, RFC 9743,\n              DOI 10.17487/RFC9743, March 2025,\n              <https://www.rfc-editor.org/info/rfc9743>.\n\nAcknowledgments\n\n   This document is a collaboration of authors that combines work from\n   three proposals.  Further authors of one of the original proposals\n   are Qing An and Zhenyu Li.\n\n   Thanks to Marten Seemann, Kazuho Oku, Martin Thomson, Magnus\n   Westerlund, Mike Bishop, Lucas Pardue, Michael Eriksson, Yu Zhu,\n   Gorry Fairhurst, Tilmann Zäschke, and Tommy Pauly for their thorough\n   reviews and valuable contributions.\n\n   Thanks for IETF Last Call reviews to Carsten Bormann (artart),\n   Antoine Fressancourt (intdir), Adrian Farrel (opsdir), and Meral\n   Shirazipour (genart).\n\nAuthors' Addresses\n\n   Yanmei Liu (editor)\n   Alibaba Inc.\n   Email: miaoji.lym@alibaba-inc.com\n\n   Additional contact information:\n\n      刘彦梅 (editor)\n      Alibaba Inc.\n\n\n   Yunfei Ma\n   Harvey AI\n   Email: yfmascgy@gmail.com\n\n   Additional contact information:\n\n      马云飞\n      Harvey AI\n\n\n   Quentin De Coninck (editor)\n   University of Mons (UMONS)\n   Email: quentin.deconinck@umons.ac.be\n\n\n   Olivier Bonaventure\n   UCLouvain and WELRI\n   Email: olivier.bonaventure@uclouvain.be\n\n\n   Christian Huitema\n   Private Octopus Inc.\n   Email: huitema@huitema.net\n\n\n   Mirja Kühlewind (editor)\n   Ericsson\n   Email: mirja.kuehlewind@ericsson.com\n","snippet":"","snippet_marked":"","source":"RFC Editor","citation":"https://www.rfc-editor.org/rfc/rfc10069","license":"IETF Trust / RFC Editor terms","year":"","url":"/record/10069","fields":{"status":"PROPOSED STANDARD","stream":"IETF","wg":"quic","date":"November 2026","obsoletes":"","updates":""}},"generated_at":"2026-11-22T20:18:34Z"}