CHANGELOG revision 55cd0b38
12019-04-12 2 - 1.21.0 3 - [FEATURE] Add qlog log module. 4 52019-04-01 6 - 1.20.0 7 - [FEATURE] Add support for Q046. 8 92019-03-19 10 - 1.19.6 11 - [BUGFIX] Ensure that Largest Observed does not decrease in ACKs we 12 send. ACK frames placed in packets in buffered queues (optimization 13 introduced in 1.17.15) can be preceded by an ACK frame generated later. 14 In this case, the older ACK frame should not be sent out, as Chromium- 15 based servers flags decrease in the ACK frame's Largest Observed value 16 as an error. 17 182019-03-05 19 - 1.19.5 20 - [BUGFIX] Use correct public key from PUBS based on KEXS index. 21 - [BUGFIX] Check flags before dispatching writes, avoiding assert. 22 - [BUGFIX] Set :scheme to "https" (instead of "HTTP"). 23 242019-02-25 25 - 1.19.4 26 - [BUGFIX] Check buffer bounds when looking up version in 0-RTT blob. 27 - [BUGFIX] http_client: don't fetch 0-rtt info if handshake failed. 28 - Log number of pacer calls at DEBUG, rather than NOTICE, level. 29 302019-02-18 31 - 1.19.3 32 - [BUGFIX] Q044: don't encode packet number in 6 bytes. Six-byte 33 packet number encoding does not exist in Q044. This fixes a 34 regression introduced in '[BUGFIX] Buffered packets can contain 35 ACK frames' -- we need to keep QUIC version in mind when selecting 36 the longest possible packet number encoding used for the buffered 37 packet that carries the ACK. 38 - [BUGFIX] Do not increase CWND when timeout occurs. 39 - http_client: support setting handshake timeout on command line. 40 Use -o handshake_to=timeout. 41 - http_client: use -k to connect UDP socket to pick up ICMP errors. 42 - http_client: allow pathless mode, when only handshake is performed 43 without issuing any requests. This can be done by simply not 44 specifying a -p flag on the command line. 45 462019-02-11 47 - 1.19.2 48 - [BUGFIX] Begin negotiation with version provided in 0-RTT info. 49 - [BUGFIX] Version checking in zero_rtt deserialize function. 50 512019-02-04 52 - 1.19.1 53 - [BUGFIX] Fix Windows build. 54 552019-02-04 56 - 1.19.0 57 - [FEATURE, API Change] 0-RTT support. Add function to export 0-RTT 58 information; it can be supplied to a subsequent connect() call. 59 - [FEATURE] Add -0 flag to http_client to exercise 0-RTT support. 60 - [BUGFIX] Resuscitate the Windows build. 61 - [BUGFIX] Send HTTP settings (max header list size) if necessary. 62 - [BUGFIX] Buffered packets can contain ACK frames. 63 - [BUGFIX] Make packet writeable once all STREAM frames are elided. 64 - [BUGFIX] Fix potential null dereference when realloc fails. 65 - cmake: simplify build configuration. 66 672019-01-28 68 - 1.18.0 69 - [API Change] Can specify clock granularity in engine settings. 70 - [BUGFIX] Pacer uses fixed clock granularity. Since the change on 71 2018-04-09, it is not necessary to try to approximate the next tick 72 time in the pacer: it can use fix clock granularity specified by 73 the user. 74 - [BUGFIX] Do not tick constantly before handshake is done. 75 - [BUGFIX] Do not exit tick prematurely: reuse packet with ACK. Even 76 if we cannot allocate *more* packets, we could still be able to write 77 to one already allocated. 78 - [BUGFIX] Do not schedule pacer if there are no lost packets. 79 802019-01-17 81 - 1.17.15 82 - [BUGFIX] http_client: make sure only one read per on_read() callback 83 is performed in the header conversion bypass (-B) mode. 84 - http_client: with -E, assign random priority when stream is created. 85 - [OPTIMIZATION] On immediate write, place an ACK frame into the first 86 buffered packet if an ACK is queued. This reduces the number of 87 standalone ACK packets. 88 - [OPTIMIZATION] Allow placing more than one STREAM frame from the same 89 stream into an outgoing packet. This change minimizes the number of 90 buffered packets required to store several small HTTP messages by 91 virtue of allowing more than one STREAM frame from HEADERS stream in 92 the same packet. 93 - [OPTIMIZATION] Flush headers when writing to buffered packets. This 94 causes the headers to be written to the same buffered packet queue, 95 thereby improving packet utilization, especially for small HTTP 96 messages. 97 982019-01-16 99 - 1.17.14 100 - [FEATURE] http_client can now collect stats: time to connect, TTFB, 101 time to download, requests per seconds, and bandwidth. See -t flag. 102 - [BUGFIX] http_client: -B, -K, and -r can be used at the same time. 103 - [BUGFIX] http_client: do not display cert chain before each request. 104 Now this is only done once per connection, if the handshake is 105 successful and -a option is given. 106 - [BUGFIX] Do not wait to dispatch on_close() after reading. If a 107 stream is scheduled to be closed after on_read() callback returns, 108 close it immediately instead of waiting until the end of the tick. 109 If client creates new request from on_close() event, they will be 110 processed in the same tick instead of waiting for the next one. 111 - [BUGFIX] HEADERS stream is critical: always treat it with highest 112 priority. 113 1142019-01-10 115 - 1.17.12 116 - [FEATURE] http_client can now issue parallel requests in the context 117 of a single connection. See -w option. 118 1192019-01-03 120 - 1.17.11 121 - Fix strict aliasing warning in optimized compilation. 122 1232018-12-27 124 - 1.17.10 125 - Fix the example program to be able to use parallel connections 126 again. (See the -n argument.) 127 1282018-12-18 129 - 1.17.9 130 - [BUGFIX] Engine: reduce minimum batch size from 256 to 4 131 1322018-12-10 133 - 1.17.8 134 - [BUGFIX] Fix compilation on FreeBSD and 32-bit Linux 135 1362018-12-03 137 - 1.17.7 138 - [BUGFIX] Do not unset PING alarm before ringing expired alarms. 139 This prevented PING from ever being sent. 140 1412018-11-29 142 - 1.17.6 143 - Add failsafe: resume sending packets after some time 144 145 The change puts a 1-second limit on the amount of time the engine 146 will not send packets after some packets are delayed. This makes 147 the library robust in case the user does not unblock the engine 148 explicitly using lsquic_engine_send_unsent_packets() call. 149 150 - [BUGFIX] Handle corner cases in send controller when packets are 151 a) delayed or b) dropped during repackaging. 152 - [BUGFIX] Memory leak: destroy buffered packets during controller 153 cleanup. 154 1552018-11-16 156 - 1.17.3 157 - [BUGFIX] Do not send STOP_WAITING frames when using Q044 158 1592018-10-19 160 - 1.17.2 161 - [BUGFIX] Memory leak in test_frame_rw unit test. 162 - [BUGFIX] Parsing packets with short IETF header: do not overwrite 163 flags. (Only a problem in unit tests -- benign otherwise.) 164 1652018-10-16 166 - 1.17.0 167 - [API Change] Packet out Memory Interface (PMI) update: 168 - Split PMI pool return into pmi_release and pmi_return 169 - PMI callbacks take peer_ctx and is_ipv6 arguments 170 - [BUGFIX] Fix use-after-free when certificate is updated 171 - Silence gcc warning in optimized mode by performing useless 172 initialization 173 - cmake: use the standard variable CMAKE_BUILD_TYPE instead of 174 DEVEL_MODE 175 1762018-10-03 177 - 1.16.0 178 - [API Change] Add lsquic_conn_n_avail_streams() 179 - [BUGFIX] only dispatch crypto stream read events if WANT_READ is on 180 1812018-09-27 182 - 1.15.0 183 - [API Change] Add LSCONN_ST_PEER_GOING_AWAY to the list of conn statuses 184 - [BUGFIX] free uncompressed headers correctly when error occurs 185 1862018-09-12 187 - 1.14.3 188 - [BUGFIX] Do not abort conn on STREAM frame for a reset stream 189 - [BUGFIX] Drop packets that would become empty due to repackaging. 190 Packets on the scheduled queue may be marked for repackaging. 191 Frames such as ACK frame that are never resent are removed from 192 repackaged packets. We must check that the newly repackaged packet 193 would not be empty. If it would be, it is destroyed instead and 194 the next packet on the scheduled queue is used. Note that this 195 change only affects the logic to return the next packet to be sent. 196 Lost packets that are being rescheduled are already processed in 197 this fashion. 198 - Byteswap CID before logging it - this makes it match Chrome CIDs. 199 (Except Q035, which is the last little-endian GQUIC version we 200 support.) 201 2022018-09-06 203 - 1.14.0 204 - [API Change] Disable packet sending if full batch cannot be sent 205 If lsquic_packets_out_f() cannot send the whole batch, disable 206 packet sending until lsquic_engine_send_unsent_packets() is called. 207 - [BUGFIX] Handle case when STREAM frame does not fit. 208 - [BUGFIX] Always allow incoming STREAM frames to overlap. Peers 209 may send overlapping STREAM frames even if using versions older 210 than Q043. 211 - Custom header set fixes: 212 - set "FIN reached" flag when custom header with FIN flag is 213 claimed; 214 - do not return custom header set for a reset stream. 215 2162018-08-27 217 218 - 1.13.0 219 - [FEATURE, API Change] Add ability to create custom header set 220 objects via callbacks. This avoids reading and re-parsing 221 headers from the stream. 222 2232018-08-27 224 225 - 1.12.4 226 - Fix memory leak when engine is destroyed 227 - Fix memory leak in http_client 228 - Fix gcc warning in unit tests 229 2302018-08-22 231 232 - 1.12.3 233 - [BUGFIX] Fix duplicate STREAM frame detection 234 2352018-08-20 236 237 - 1.12.2 238 - [BUGFIX] Update count of scheduled bytes when adjusting size of 239 an already-scheduled packet. 240 - Emit info instead of warning messages when stream is used in 241 unexpected ways. 242 2432018-08-17 244 245 - 1.12.0 246 - [FEATURE, API Change] Add support for certificate verification 247 2482018-08-16 249 250 - 1.11.1 251 - [BUGFIX] Fix STOP_WAITING frame processing in the new Q044 code 252 2532018-08-15 254 255 - 1.11.0 256 - [FEATURE] Add support for Q044. 257 2582018-08-09 259 260 - 1.10.2 261 - [BUGFIX] Don't go over limit when creating delayed streams 262 2632018-07-10 264 265 - 1.10.1 266 - [BUGFIX] process connections after each batch of packets is read 267 This avoids a problem of accumulating a very large list of packets 268 (possible when speeds are high and socket's receive buffer is large) 269 and processing it all at once. 270 - If glibc is older than 2.17, link with rt. This is necessary for 271 clock_getres(2). 272 - Add version macros to lsquic.h; remove unnecessary includes. 273 2742018-06-13 275 276 - [BUGFIX] allow multiple parallel connections by default 277 278 Use the original method of tracking connections by CIDs by default. 279 If zero-sized CID support is turned on, connections are tracked by 280 the address. A new connection is not created if another connection 281 is using the same network address 282 2832018-05-30 284 285 - [FEATURE, API CHANGE] Support zero-sized CIDs in received packets 286 2872018-05-24 288 289 - Close connection properly when packet encryption fails 290 2912018-05-23 292 293 - [BUGFIX] Do not produce packet sequence gaps due to delayed packets 294 2952018-05-21 296 297 - [API Change] Add optional callback to call when handshake is done 298 - [API Change, BUGFIX] After send failure, wait until transport available 299 3002018-05-18 301 302 - [API] Expose useful lsquic_ver2str[] in lsquic.h 303 - [BUGFIX] Do not produce packet sequence gaps due to STREAM frame elision 304 - Improve checks of number of incoming streams limit and associated 305 error reporting. 306 - [BUGFIX] Command-line option `-6` now works correctly. 307 3082018-05-16 309 310 - [FEATURE] DNS resolution 311 - [BUGFIX] Frame insertion mis-ID as overlap instead of dup 312 - http_client: fix priority range generated by -E flag 313 3142018-05-09 315 316 - [FEATURE] Add support for Q043. 317 - Support for versions Q037, Q038, Q041, and Q042 has been removed. 318 - Fix typo in debug message. 319 - Fix code indentation. 320 - Add /* fallthru */ comment to stop newer gcc from complaining. 321 - Logger: fix compilation of optimized Windows build. 322 3232018-05-04 324 325 - [FEATURE] Add support for Q042. 326 - Remove comment: MSPC is obsolete (no code changes) 327 - Prog: use lsquic_str2ver() when processing -o version flag 328 - Remove unused CTIM and SRBF transport parameters 329 - Disable QUIC versions Q037 and Q038 by default 330 - Fix Windows compilation by including compat header file in 331 lshpack.c 332 - Address warnings produced by newer versions of gcc 333 - Future-proof: turn off -Werror 334 3352018-05-02 336 337 - [BUGFIX] Make lsquic_conn_quic_version() available 338 - Switch to using ls-hpack 1.1 339 - [BUGFIX] Do not ignore stream resets after receiving FIN 340 3412018-04-27 342 343 - HPACK: do not allow header block to end with table size update. 344 3452018-04-25 346 347 - [BUGFIX] Do not create gap in sent packnos when squeezing delayed 348 packets. 349 - [BUGFIX] sendctl checks for all unacked bytes, not just retx bytes. 350 - [BUGFIX] connections with blocked scheduled packets are not tickable 351 for sending. 352 - [BUGFIX] Conn is tickable if it wants to send a connection-level 353 frame. 354 3552018-04-23 356 357 - Fix busy loop: tickable must make progress. When connection is 358 self-reporting as tickable, it must make progress when ticked. There 359 are two issues: 360 1. If there are buffered packets, the connection is only tickable if 361 they can be sent out. 362 2. A connection is tickable if there are streams on the servicing 363 queue. When the tick occurs, we must service the stream 364 independent of whether any packets are sent. 365 - Fix assertion in pacer which can be incorrect under some 366 conditions. 367 - cmake: do not turn on address sanitizer if in Travis. 368 3692018-04-20 370 371 - [BUGFIX] Fix bug in lsquic_engine_connect() exposed by yesterday's 372 changes. 373 3742018-04-19 375 376 - [BUGFIX] Add connection to Tickable Queue on stream write 377 - cmake: use MSVC variable instead of trying to detect 378 - engine: improve connection incref/decref logging 379 - stream: don't ignore errors that may occur on triggered flush 380 - connection: remove obsolete method 381 - engine: indicate connection as tickable if previous call went 382 over threshold 383 3842018-04-09 385 386 [API Change, OPTIMIZATION] Only process conns that need to be processed 387 388 The API is simplified: do not expose the user code to several 389 queues. A "connection queue" is now an internal concept. 390 The user processes connections using the single function 391 lsquic_engine_process_conns(). When this function is called, 392 only those connections are processed that need to be processed. 393 A connection needs to be processed when: 394 395 1. New incoming packets have been fed to the connection. 396 2. User wants to read from a stream that is readable. 397 3. User wants to write to a stream that is writeable. 398 4. There are buffered packets that can be sent out. (This 399 means that the user wrote to a stream outside of the 400 lsquic library callback.) 401 5. A control frame (such as BLOCKED) needs to be sent out. 402 6. A stream needs to be serviced or delayed stream needs to 403 be created. 404 7. An alarm rings. 405 8. Pacer timer expires. 406 407 To achieve this, the library places the connections into two 408 priority queues (min heaps): 409 410 1. Tickable Queue; and 411 2. Advisory Tick Time queue (ATTQ). 412 413 Each time lsquic_engine_process_conns() is called, the Tickable 414 Queue is emptied. After the connections have been ticked, they are 415 queried again: if a connection is not being closed, it is placed 416 either in the Tickable Queue if it is ready to be ticked again or 417 it is placed in the Advisory Tick Time Queue. It is assumed that 418 a connection always has at least one timer set (the idle alarm). 419 420 The connections in the Tickable Queue are arranged in the least 421 recently ticked order. This lets connections that have been quiet 422 longer to get their packets scheduled first. 423 424 This change means that the library no longer needs to be ticked 425 periodically. The user code can query the library when is the 426 next tick event and schedule it exactly. When connections are 427 processed, only the tickable connections are processed, not *all* 428 the connections. When there are no tick events, it means that no 429 timer event is necessary -- only the file descriptor READ event 430 is active. 431 432 The following are improvements and simplifications that have 433 been triggered: 434 435 - Queue of connections with incoming packets is gone. 436 - "Pending Read/Write Events" Queue is gone (along with its 437 history and progress checks). This queue has become the 438 Tickable Queue. 439 - The connection hash no longer needs to track the connection 440 insertion order. 441 4422018-04-02 443 444 - [FEATURE] Windows support 445 446 - Reduce stack use -- outgoing packet batch is now allocated on the heap. 447 4482018-03-09 449 450 - [OPTIMIZATION] Merge series of ACKs if possible 451 452 Parsed single-range ACK frames (that is the majority of frames) are 453 saved in the connection and their processing is deferred until the 454 connection is ticked. If several ACKs come in a series between 455 adjacent ticks, we check whether the latest ACK is a strict superset 456 of the saved ACK. If it is, the older ACK is not processed. 457 458 If ACK frames can be merged, they are merged and only one of them is 459 either processed or saved. 460 461 - [OPTIMIZATION] Speed up ACK verification by simplifying send history. 462 463 Never generate a gap in the sent packet number sequence. This reduces 464 the send history to a single number instead of potentially a series of 465 packet ranges and thereby speeds up ACK verification. 466 467 By default, detecting a gap in the send history is not fatal: only a 468 single warning is generated per connection. The connection can continue 469 to operate even if the ACK verification code is not able to detect some 470 inconsistencies. 471 472 - [OPTIMIZATION] Rearrange the lsquic_send_ctl struct 473 474 The first part of struct lsquic_send_ctl now consists of members that 475 are used in lsquic_send_ctl_got_ack() (in the absense of packet loss, 476 which is the normal case). To speed up reads and writes, we no longer 477 try to save space by using 8- and 16-bit integers. Use regular integer 478 width for everything. 479 480 - [OPTIMIZATION] Cache size of sent packet. 481 482 - [OPTIMIZATION] Keep track of the largest ACKed in packet_out 483 484 Instead of parsing our own ACK frames when packet has been acked, 485 use the value saved in the packet_out structure when the ACK frame 486 was generated. 487 488 - [OPTIMIZATION] Take RTT sampling conditional out of ACK loop 489 490 - [OPTIMIZATION] ACK processing: only call clock_gettime() if needed 491 492 - [OPTIMIZATION] Several code-level optimizations to ACK processing. 493 494 - Fix: http_client: fix -I flag; switch assert() to abort() 495 4962018-02-26 497 - [API Change] lsquic_engine_connect() returns pointer to the connection 498 object. 499 - [API Change] Add lsquic_conn_get_engine() to get engine object from 500 connection object. 501 - [API Change] Add lsquic_conn_status() to query connection status. 502 - [API Change] Add add lsquic_conn_set_ctx(). 503 - [API Change] Add new timestamp format, e.g. 2017-03-21 13:43:46.671345 504 - [OPTIMIZATION] Process handshake STREAM frames as soon as packet 505 arrives. 506 - [OPTIMIZATION] Do not compile expensive send controller sanity check 507 by default. 508 - [OPTIMIZATION] Add fast path to gquic_be_gen_reg_pkt_header. 509 - [OPTIMIZATION] Only make squeeze function call if necessary. 510 - [OPTIMIZATION] Speed up Q039 ACK frame parsing. 511 - [OPTIMIZATION] Fit most used elements of packet_out into first 64 bytes. 512 - [OPTIMIZATION] Keep track of scheduled bytes instead of calculating. 513 - [OPTIMIZATION] Prefetch next unacked packet when processing ACK. 514 - [OPTIMIZATION] Leverage fact that ACK ranges and unacked list are. 515 ordered. 516 - [OPTIMIZATION] Reduce function pointer use for STREAM frame generation 517 - Fix: reset incoming streams that arrive after we send GOAWAY. 518 - Fix: delay client on_new_conn() call until connection is fully set up. 519 - Fixes to buffered packets logic: splitting, STREAM frame elision. 520 - Fix: do not dispatch on_write callback if no packets are available. 521 - Fix WINDOW_UPDATE send and resend logic. 522 - Fix STREAM frame extension code. 523 - Fix: Drop unflushed data when stream is reset. 524 - Switch to tracking CWND using bytes rather than packets. 525 - Fix TCP friendly adjustment in cubic. 526 - Fix: do not generate invalid STOP_WAITING frames during high packet 527 loss. 528 - Pacer fixes. 529 5302017-12-18 531 532 - Fix: better follow cubic curve after idle period 533 - Fix: add missing parts to outgoing packet splitting code 534 - Fix: compilation using gcc 4.8.4 535 5362017-10-31 537 538 - Add APIs.txt -- describes LSQUIC APIs 539 5402017-10-31 541 542 - [API Change] Sendfile-like functionality is gone. The stream no 543 longer opens files and deals with file descriptors. (Among other 544 things, this makes the code more portable.) Three writing functions 545 are provided: 546 547 lsquic_stream_write 548 lsquic_stream_writev 549 lsquic_stream_writef (NEW) 550 551 lsquic_stream_writef() is given an abstract reader that has function 552 pointers for size() and read() functions which the user can implement. 553 This is the most flexible way. lsquic_stream_write() and 554 lsquic_stream_writev() are now both implemented as wrappers around 555 lsquic_stream_writef(). 556 557 - [OPTIMIZATION] When writing to stream, be it within or without the 558 on_write() callback, place data directly into packet buffer, 559 bypassing auxiliary data structures. This reduces amount of memory 560 required, for the amount of data that can be written is limited 561 by the congestion window. 562 563 To support writes outside the on_write() callback, we keep N 564 outgoing packet buffers per connection which can be written to 565 by any stream. One half of these are reserved for the highest 566 priority stream(s), the other half for all other streams. This way, 567 low-priority streams cannot write instead of high-priority streams 568 and, on the other hand, low-priority streams get a chance to send 569 their packets out. 570 571 The algorithm is as follows: 572 573 - When user writes to stream outside of the callback: 574 - If this is the highest priority stream, place it onto the 575 reserved N/2 queue or fail. 576 (The actual size of this queue is dynamic -- MAX(N/2, CWND) -- 577 rather than N/2, allowing high-priority streams to write as 578 much as can be sent.) 579 - If the stream is not the highest priority, try to place the 580 data onto the reserved N/2 queue or fail. 581 - When tick occurs *and* more packets can be scheduled: 582 - Transfer packets from the high N/2 queue to the scheduled 583 queue. 584 - If more scheduling is allowed: 585 - Call on_write callbacks for highest-priority streams, 586 placing resulting packets directly onto the scheduled queue. 587 - If more scheduling is allowed: 588 - Transfer packets from the low N/2 queue to the scheduled 589 queue. 590 - If more scheduling is allowed: 591 - Call on_write callbacks for non-highest-priority streams, 592 placing resulting packets directly onto the scheduled queue 593 594 The number N is currently 20, but it could be varied based on 595 resource usage. 596 597 - If stream is created due to incoming headers, make headers readable 598 from on_new. 599 600 - Outgoing packets are no longer marked non-writeable to prevent placing 601 more than one STREAM frame from the same stream into a single packet. 602 This property is maintained via code flow and an explicit check. 603 Packets for stream data are allocated using a special function. 604 605 - STREAM frame elision is cheaper, as we only perform it if a reset 606 stream has outgoing packets referencing it. 607 608 - lsquic_packet_out_t is smaller, as stream_rec elements are now 609 inside a union. 610 6112017-10-12 612 613 - Do not send RST_STREAM when stream is closed for reading 614 - Raise maximum header size from 4K to 64K 615 - Check header name and value lengths against maximum imposed by HPACK 616 - Fix NULL dereference in stream flow controller 617 6182017-10-09 619 620 - Hide handshake implementation behind a set of function pointers 621 - Use monotonically increasing clock 622 - Make sure that retx delay is not larger than the max of 60 seconds 623 6242017-09-29 625 626 - A few fixes to code and README 627 6282017-09-28 629 630 - Add support for Q041; drop support for Q040 631 6322017-09-27 633 634 - Fix CMakeLists.txt: BoringSSL include and lib was mixed up 635 6362017-09-26 637 638 - Add support for Mac OS 639 - Add support for Raspberry Pi 640 - Fix BoringSSL compilation: include <openssl/hmac.h> explicitly 641 6422017-09-22 643 644 - Initial release 645