Congestion control API design
=============================
-This is mostly inspired by the MSQUIC congestion control API
-as it was the most isolated one among the libraries.
-
-The API is designed in a way to be later easily transformed into a
-a fetchable implementation API.
-
-The API is centered around two structures - the `OSSL_CC_METHOD`
-structure holding the API calls into the congestion control module
-and `OSSL_CC_DATA` opaque type that holds the data of needed
-for the operation of the module.
-
-Most of the information when the functions of the API are
-supposed to be called and how the API is designed to operate
-should be clear from the comments for the individual
-functions in [Appendix A](#appendix-a).
-
-Changeables
------------
-
-As some parameters that the congestion control algorithm needs might
-be updated during the lifetime of the connection these parameters
-are called changeables.
-
-The CC implementation will save the pointers to these parameters'
-data and will use the current value of the data at the pointer when
-it needs to recalculate the allowance and whether the sending is
-blocked or not.
-
-Some examples of changeables
-
- - `payload_length` - the maximum length of payload that can be sent
- in a single UDP packet (not counting UDP and IP headers).
- - `smoothed_rtt` - the current round trip time of the connection
- as computed from the acked packets.
- - `rtt_variance` - the variance of the `smoothed_rtt` value.
-
-Thread handling
----------------
-
-The `OSSL_CC_DATA` is created with the `new` function per each
-connection. The expectation is that there is only a single thread
-accessing the `ccdata` which should not be a limitation as the calls
-into the module will be done from the packetizer layer of the QUIC
-connection handling and there the eventual writes from multiple threads
-handling individual streams of the connection have to be synchronized
-already to create the packets.
-
-Congestion event handling
--------------------------
-
-The congestion state of a connection can change only after some event
-happens - i.e., a packet is considered lost meaning the `on_data_lost()`
-is called, or an ack arrives for a packet causing calls to
-`on_data_acked()` or `on_spurious_congestion_event()` functions.
-The congestion control does not produce any timer events by itself.
-
-Exemptions
-----------
-
-To facilitate probing and to avoid having to always special-case
-probing packets when considering congestion on sending, the
-`set_exemption()` function allows setting a number of packets that are
-allowed to be sent even when forbidden by the eventual congestion state.
-
-The exemptions must be used if and only if a packet (or multiple packets)
-has to be sent as required by the protocol regardless of the congestion state.
-
-Paths
------
-
-Initially the design expects that only a single path per-connection is
-actively sending data. In future when multiple active paths sending data
-shall be supported the instances of `OSSL_CC_DATA` would be per-path.
-
-There might need to be further adjustments needed in that case. However
-at least initially this API is intended to be internal to the
-OpenSSL library allowing any necessary changes of the API.
-
-Appendix A
-----------
-
-Proposed header file with comments explaining the individual
-functions. The API is meant to be internal initially so the method
-accessors to set the individual functions will be added later once
-the API is public. Alternatively this might be also implemented
-as fetchable dispatch API.
-
-```C
-/*
- * Copyright 2022 The OpenSSL Project Authors. All Rights Reserved.
- *
- * Licensed under the Apache License 2.0 (the "License"). You may not use
- * this file except in compliance with the License. You can obtain a copy
- * in the file LICENSE in the source distribution or at
- * https://www.openssl.org/source/license.html
- */
-
-#include <openssl/ssl.h>
-
-typedef struct ossl_cc_method_st OSSL_CC_METHOD;
-/*
- * An OSSL_CC_DATA is an externally defined opaque pointer holding
- * the internal data of the congestion control method
- */
-typedef struct ossl_cc_data_st OSSL_CC_DATA;
-
-/*
- * Once this becomes public API then we will need functions to create and
- * free an OSSL_CC_METHOD, as well as functions to get/set the various
- * function pointers....unless we make it fetchable.
- */
-struct ossl_cc_method_st {
- /*
- * Create a new OSSL_CC_DATA object to handle the congestion control
- * calculations.
- *
- * |settings| are mandatory settings that will cause the
- * new() call to fail if they are not understood).
- * |options| are optional settings that will not
- * cause the new() call to fail if they are not understood.
- * |changeables| contain additional parameters that the congestion
- * control algorithms need that can be updated during the
- * connection lifetime - for example size of the datagram payload.
- * To avoid calling a function with OSSL_PARAM array every time
- * these parameters are changed the addresses of these param
- * values are considered permanent and the values can be updated
- * any time.
- */
- OSSL_CC_DATA *(*new)(OSSL_PARAM *settings, OSSL_PARAM *options,
- OSSL_PARAM *changeables);
-
- /*
- * Release the OSSL_CC_DATA.
- */
- void (*free)(OSSL_CC_DATA *ccdata);
-
- /*
- * Reset the congestion control state.
- * |flags| to support different level of reset (partial/full).
- */
- void (*reset)(OSSL_CC_DATA *ccdata, int flags);
-
- /*
- * Set number of packets exempted from CC - used for probing
- * |numpackets| is a small value (2).
- * Returns 0 on error, 1 otherwise.
- */
- int (*set_exemption)(OSSL_CC_DATA *ccdata, int numpackets);
-
- /*
- * Get current number of packets exempted from CC.
- * Returns negative value on error, the number otherwise.
- */
- int (*get_exemption)(OSSL_CC_DATA *ccdata);
-
- /*
- * Returns 1 if sending is allowed, 0 otherwise.
- */
- int (*can_send)(OSSL_CC_DATA *ccdata);
-
- /*
- * Returns number of bytes allowed to be sent.
- * |time_since_last_send| is time since last send operation
- * in microseconds.
- * |time_valid| is 1 if the |time_since_last_send| holds
- * a meaningful value, 0 otherwise.
- */
- size_t (*get_send_allowance)(OSSL_CC_DATA *ccdata,
- uint64_t time_since_last_send,
- int time_valid);
-
- /*
- * Returns the maximum number of bytes allowed to be in flight.
- */
- size_t (*get_bytes_in_flight_max)(OSSL_CC_DATA *ccdata);
-
- /*
- * Returns the next time at which the CC will release more budget for
- * sending, or ossl_time_infinite().
- */
- OSSL_TIME (*get_next_credit_time)(OSSL_CC_DATA *ccdata);
-
- /*
- * To be called when a packet with retransmittable data was sent.
- * |num_retransmittable_bytes| is the number of bytes sent
- * in the packet that are retransmittable.
- * Returns 1 on success, 0 otherwise.
- */
- int (*on_data_sent)(OSSL_CC_DATA *ccdata,
- size_t num_retransmittable_bytes);
-
- /*
- * To be called when retransmittable data was invalidated.
- * I.E. they are not considered in-flight anymore but
- * are neither acknowledged nor lost. In particular used when
- * 0RTT data was rejected.
- * |num_retransmittable_bytes| is the number of bytes
- * of the invalidated data.
- * Returns 1 if sending is unblocked (can_send returns 1), 0
- * otherwise.
- */
- int (*on_data_invalidated)(OSSL_CC_DATA *ccdata,
- size_t num_retransmittable_bytes);
-
- /*
- * To be called when sent data was acked.
- * |time_now| is current time in microseconds.
- * |largest_pn_acked| is the largest packet number of the acked
- * packets.
- * |num_retransmittable_bytes| is the number of retransmittable
- * packet bytes that were newly acked.
- * Returns 1 if sending is unblocked (can_send returns 1), 0
- * otherwise.
- */
- int (*on_data_acked)(OSSL_CC_DATA *ccdata,
- uint64_t time_now,
- uint64_t last_pn_acked,
- size_t num_retransmittable_bytes);
-
- /*
- * To be called when sent data is considered lost.
- * |largest_pn_lost| is the largest packet number of the lost
- * packets.
- * |largest_pn_sent| is the largest packet number sent on this
- * connection.
- * |num_retransmittable_bytes| is the number of retransmittable
- * packet bytes that are newly considered lost.
- * |persistent_congestion| is 1 if the congestion is considered
- * persistent (see RFC 9002 Section 7.6), 0 otherwise.
- */
- void (*on_data_lost)(OSSL_CC_DATA *ccdata,
- uint64_t largest_pn_lost,
- uint64_t largest_pn_sent,
- size_t num_retransmittable_bytes,
- int persistent_congestion);
-
- /*
- * To be called when all lost data from the previous call to
- * on_data_lost() was actually acknowledged.
- * This reverts the size of the congestion window to the state
- * before the on_data_lost() call.
- * Returns 1 if sending is unblocked, 0 otherwise.
- */
- int (*on_spurious_congestion_event)(OSSL_CC_DATA *ccdata);
-};
-```
+We use an abstract interface for the QUIC congestion controller to facilitate
+use of pluggable QUIC congestion controllers in the future. The interface is
+based on interfaces suggested by RFC 9002 and MSQUIC's congestion control APIs.
+
+`OSSL_CC_METHOD` provides a vtable of function pointers to congestion controller
+methods. `OSSL_CC_DATA` is an opaque type representing a congestion controller
+instance.
+
+For details on the API, see the comments in `include/internal/quic_cc.h`.
+
+Congestion controllers are not thread safe; the caller is responsible for
+synchronisation.
+
+Congestion controllers may vary their state with respect to time. This is
+faciliated via the `get_wakeup_deadline` method and the `now` argument to the
+`new` method, which provides access to a clock. While no current congestion
+controller makes use of this facility, it can be used by future congestion
+controllers to implement packet pacing.
+
+Congestion controllers may expose integer configuration options via the
+`set_option_uint` and `get_option_uint` methods. These options may be specific
+to the congestion controller method, although there are some well known ones
+intended to be common to all congestion controllers. The use of strings for
+option names is avoided for performance reasons.
+
+Currently, the only dependency injected to a congestion controller is access to
+a clock. In the future it is likely that access at least to the statistics
+manager will be provided. Excessive futureproofing of the congestion controller
+interface has been avoided as this is currently an internal API for which no API
+stability guarantees are required; for example, no currently implemented
+congestion control algorithm requires access to the statistics manager, but such
+access can readily be added later as needed.
+
+QUIC congestion control state is per-path, per-connection. Currently we support
+only a single path per connection, so there is one congestion control instance
+per connection. This may change in future.
+
+While the congestion control API is roughly based around the arrangement of
+functions as described by the congestion control psuedocode in RFC 9002, there
+are some deliberate changes in order to obtain cleaner separation between the
+loss detection and congestion control functions. Where a literal option of RFC
+9002 psuedocode would require a congestion controller to access the ACK
+manager's internal state directly, the interface between the two has been
+changed to avoid this. This involves some small amounts of functionality which
+RFC 9002 considers part of the congestion controller being part of the ACK
+manager in our implementation. See the comments in `include/internal/quic_cc.h`
+and `ssl/quic/quic_ackm.c` for more information.
projects / thirdparty / openssl.git / blobdiff