CHANGELOG revision bfc7bfd8
12018-02-26 2 - [API Change] lsquic_engine_connect() returns pointer to the connection 3 object. 4 - [API Change] Add lsquic_conn_get_engine() to get engine object from 5 connection object. 6 - [API Change] Add lsquic_conn_status() to query connection status. 7 - [API Change] Add add lsquic_conn_set_ctx(). 8 - [API Change] Add new timestamp format, e.g. 2017-03-21 13:43:46.671345 9 - [OPTIMIZATION] Process handshake STREAM frames as soon as packet 10 arrives. 11 - [OPTIMIZATION] Do not compile expensive send controller sanity check 12 by default. 13 - [OPTIMIZATION] Add fast path to gquic_be_gen_reg_pkt_header. 14 - [OPTIMIZATION] Only make squeeze function call if necessary. 15 - [OPTIMIZATION] Speed up Q039 ACK frame parsing. 16 - [OPTIMIZATION] Fit most used elements of packet_out into first 64 bytes. 17 - [OPTIMIZATION] Keep track of scheduled bytes instead of calculating. 18 - [OPTIMIZATION] Prefetch next unacked packet when processing ACK. 19 - [OPTIMIZATION] Leverage fact that ACK ranges and unacked list are. 20 ordered. 21 - [OPTIMIZATION] Reduce function pointer use for STREAM frame generation 22 - Fix: reset incoming streams that arrive after we send GOAWAY. 23 - Fix: delay client on_new_conn() call until connection is fully set up. 24 - Fixes to buffered packets logic: splitting, STREAM frame elision. 25 - Fix: do not dispatch on_write callback if no packets are available. 26 - Fix WINDOW_UPDATE send and resend logic. 27 - Fix STREAM frame extension code. 28 - Fix: Drop unflushed data when stream is reset. 29 - Switch to tracking CWND using bytes rather than packets. 30 - Fix TCP friendly adjustment in cubic. 31 - Fix: do not generate invalid STOP_WAITING frames during high packet 32 loss. 33 - Pacer fixes. 34 352017-12-18 36 37 - Fix: better follow cubic curve after idle period 38 - Fix: add missing parts to outgoing packet splitting code 39 - Fix: compilation using gcc 4.8.4 40 412017-10-31 42 43 - Add APIs.txt -- describes LSQUIC APIs 44 452017-10-31 46 47 - [API Change] Sendfile-like functionality is gone. The stream no 48 longer opens files and deals with file descriptors. (Among other 49 things, this makes the code more portable.) Three writing functions 50 are provided: 51 52 lsquic_stream_write 53 lsquic_stream_writev 54 lsquic_stream_writef (NEW) 55 56 lsquic_stream_writef() is given an abstract reader that has function 57 pointers for size() and read() functions which the user can implement. 58 This is the most flexible way. lsquic_stream_write() and 59 lsquic_stream_writev() are now both implemented as wrappers around 60 lsquic_stream_writef(). 61 62 - [OPTIMIZATION] When writing to stream, be it within or without the 63 on_write() callback, place data directly into packet buffer, 64 bypassing auxiliary data structures. This reduces amount of memory 65 required, for the amount of data that can be written is limited 66 by the congestion window. 67 68 To support writes outside the on_write() callback, we keep N 69 outgoing packet buffers per connection which can be written to 70 by any stream. One half of these are reserved for the highest 71 priority stream(s), the other half for all other streams. This way, 72 low-priority streams cannot write instead of high-priority streams 73 and, on the other hand, low-priority streams get a chance to send 74 their packets out. 75 76 The algorithm is as follows: 77 78 - When user writes to stream outside of the callback: 79 - If this is the highest priority stream, place it onto the 80 reserved N/2 queue or fail. 81 (The actual size of this queue is dynamic -- MAX(N/2, CWND) -- 82 rather than N/2, allowing high-priority streams to write as 83 much as can be sent.) 84 - If the stream is not the highest priority, try to place the 85 data onto the reserved N/2 queue or fail. 86 - When tick occurs *and* more packets can be scheduled: 87 - Transfer packets from the high N/2 queue to the scheduled 88 queue. 89 - If more scheduling is allowed: 90 - Call on_write callbacks for highest-priority streams, 91 placing resulting packets directly onto the scheduled queue. 92 - If more scheduling is allowed: 93 - Transfer packets from the low N/2 queue to the scheduled 94 queue. 95 - If more scheduling is allowed: 96 - Call on_write callbacks for non-highest-priority streams, 97 placing resulting packets directly onto the scheduled queue 98 99 The number N is currently 20, but it could be varied based on 100 resource usage. 101 102 - If stream is created due to incoming headers, make headers readable 103 from on_new. 104 105 - Outgoing packets are no longer marked non-writeable to prevent placing 106 more than one STREAM frame from the same stream into a single packet. 107 This property is maintained via code flow and an explicit check. 108 Packets for stream data are allocated using a special function. 109 110 - STREAM frame elision is cheaper, as we only perform it if a reset 111 stream has outgoing packets referencing it. 112 113 - lsquic_packet_out_t is smaller, as stream_rec elements are now 114 inside a union. 115 1162017-10-12 117 118 - Do not send RST_STREAM when stream is closed for reading 119 - Raise maximum header size from 4K to 64K 120 - Check header name and value lengths against maximum imposed by HPACK 121 - Fix NULL dereference in stream flow controller 122 1232017-10-09 124 125 - Hide handshake implementation behind a set of function pointers 126 - Use monotonically increasing clock 127 - Make sure that retx delay is not larger than the max of 60 seconds 128 1292017-09-29 130 131 - A few fixes to code and README 132 1332017-09-28 134 135 - Add support for Q041; drop support for Q040 136 1372017-09-27 138 139 - Fix CMakeLists.txt: BoringSSL include and lib was mixed up 140 1412017-09-26 142 143 - Add support for Mac OS 144 - Add support for Raspberry Pi 145 - Fix BoringSSL compilation: include <openssl/hmac.h> explicitly 146 1472017-09-22 148 149 - Initial release 150