├── .gitignore ├── .travis.yml ├── COPYING ├── README.md ├── cal_speed.py ├── checkack.py ├── docs ├── suft-state.graphml └── suft-state.png ├── examples ├── rap │ └── rap.go ├── suft-nc │ └── main.go └── tors │ ├── README.md │ ├── io.go │ ├── io_windows.go │ └── main.go ├── protocol ├── conn.go ├── conn_test.go ├── debug.go ├── endpoint.go ├── endpoint_test.go ├── linked.go ├── linked_test.go ├── netfd_unix.go ├── netfd_windows.go ├── netpoll.go ├── packet.go ├── state.go ├── stopwatch.go ├── timer.go ├── timer.s └── timing_test.go └── wireshark-plugin ├── README.md └── suft.lua /.gitignore: -------------------------------------------------------------------------------- 1 | *.exe 2 | *.bat 3 | *.sh 4 | *.pprof 5 | *.pdf 6 | *.log 7 | *.tmp 8 | *.dmp 9 | -------------------------------------------------------------------------------- /.travis.yml: -------------------------------------------------------------------------------- 1 | language: go 2 | go: 3 | - 1.7.3 4 | -------------------------------------------------------------------------------- /COPYING: -------------------------------------------------------------------------------- 1 | GNU GENERAL PUBLIC LICENSE 2 | Version 3, 29 June 2007 3 | 4 | Copyright (C) 2007 Free Software Foundation, Inc. 5 | Everyone is permitted to copy and distribute verbatim copies 6 | of this license document, but changing it is not allowed. 7 | 8 | Preamble 9 | 10 | The GNU General Public License is a free, copyleft license for 11 | software and other kinds of works. 12 | 13 | The licenses for most software and other practical works are designed 14 | to take away your freedom to share and change the works. By contrast, 15 | the GNU General Public License is intended to guarantee your freedom to 16 | share and change all versions of a program--to make sure it remains free 17 | software for all its users. We, the Free Software Foundation, use the 18 | GNU General Public License for most of our software; it applies also to 19 | any other work released this way by its authors. You can apply it to 20 | your programs, too. 21 | 22 | When we speak of free software, we are referring to freedom, not 23 | price. Our General Public Licenses are designed to make sure that you 24 | have the freedom to distribute copies of free software (and charge for 25 | them if you wish), that you receive source code or can get it if you 26 | want it, that you can change the software or use pieces of it in new 27 | free programs, and that you know you can do these things. 28 | 29 | To protect your rights, we need to prevent others from denying you 30 | these rights or asking you to surrender the rights. Therefore, you have 31 | certain responsibilities if you distribute copies of the software, or if 32 | you modify it: responsibilities to respect the freedom of others. 33 | 34 | For example, if you distribute copies of such a program, whether 35 | gratis or for a fee, you must pass on to the recipients the same 36 | freedoms that you received. You must make sure that they, too, receive 37 | or can get the source code. And you must show them these terms so they 38 | know their rights. 39 | 40 | Developers that use the GNU GPL protect your rights with two steps: 41 | (1) assert copyright on the software, and (2) offer you this License 42 | giving you legal permission to copy, distribute and/or modify it. 43 | 44 | For the developers' and authors' protection, the GPL clearly explains 45 | that there is no warranty for this free software. For both users' and 46 | authors' sake, the GPL requires that modified versions be marked as 47 | changed, so that their problems will not be attributed erroneously to 48 | authors of previous versions. 49 | 50 | Some devices are designed to deny users access to install or run 51 | modified versions of the software inside them, although the manufacturer 52 | can do so. This is fundamentally incompatible with the aim of 53 | protecting users' freedom to change the software. The systematic 54 | pattern of such abuse occurs in the area of products for individuals to 55 | use, which is precisely where it is most unacceptable. Therefore, we 56 | have designed this version of the GPL to prohibit the practice for those 57 | products. If such problems arise substantially in other domains, we 58 | stand ready to extend this provision to those domains in future versions 59 | of the GPL, as needed to protect the freedom of users. 60 | 61 | Finally, every program is threatened constantly by software patents. 62 | States should not allow patents to restrict development and use of 63 | software on general-purpose computers, but in those that do, we wish to 64 | avoid the special danger that patents applied to a free program could 65 | make it effectively proprietary. To prevent this, the GPL assures that 66 | patents cannot be used to render the program non-free. 67 | 68 | The precise terms and conditions for copying, distribution and 69 | modification follow. 70 | 71 | TERMS AND CONDITIONS 72 | 73 | 0. Definitions. 74 | 75 | "This License" refers to version 3 of the GNU General Public License. 76 | 77 | "Copyright" also means copyright-like laws that apply to other kinds of 78 | works, such as semiconductor masks. 79 | 80 | "The Program" refers to any copyrightable work licensed under this 81 | License. Each licensee is addressed as "you". "Licensees" and 82 | "recipients" may be individuals or organizations. 83 | 84 | To "modify" a work means to copy from or adapt all or part of the work 85 | in a fashion requiring copyright permission, other than the making of an 86 | exact copy. The resulting work is called a "modified version" of the 87 | earlier work or a work "based on" the earlier work. 88 | 89 | A "covered work" means either the unmodified Program or a work based 90 | on the Program. 91 | 92 | To "propagate" a work means to do anything with it that, without 93 | permission, would make you directly or secondarily liable for 94 | infringement under applicable copyright law, except executing it on a 95 | computer or modifying a private copy. Propagation includes copying, 96 | distribution (with or without modification), making available to the 97 | public, and in some countries other activities as well. 98 | 99 | To "convey" a work means any kind of propagation that enables other 100 | parties to make or receive copies. Mere interaction with a user through 101 | a computer network, with no transfer of a copy, is not conveying. 102 | 103 | An interactive user interface displays "Appropriate Legal Notices" 104 | to the extent that it includes a convenient and prominently visible 105 | feature that (1) displays an appropriate copyright notice, and (2) 106 | tells the user that there is no warranty for the work (except to the 107 | extent that warranties are provided), that licensees may convey the 108 | work under this License, and how to view a copy of this License. If 109 | the interface presents a list of user commands or options, such as a 110 | menu, a prominent item in the list meets this criterion. 111 | 112 | 1. Source Code. 113 | 114 | The "source code" for a work means the preferred form of the work 115 | for making modifications to it. "Object code" means any non-source 116 | form of a work. 117 | 118 | A "Standard Interface" means an interface that either is an official 119 | standard defined by a recognized standards body, or, in the case of 120 | interfaces specified for a particular programming language, one that 121 | is widely used among developers working in that language. 122 | 123 | The "System Libraries" of an executable work include anything, other 124 | than the work as a whole, that (a) is included in the normal form of 125 | packaging a Major Component, but which is not part of that Major 126 | Component, and (b) serves only to enable use of the work with that 127 | Major Component, or to implement a Standard Interface for which an 128 | implementation is available to the public in source code form. A 129 | "Major Component", in this context, means a major essential component 130 | (kernel, window system, and so on) of the specific operating system 131 | (if any) on which the executable work runs, or a compiler used to 132 | produce the work, or an object code interpreter used to run it. 133 | 134 | The "Corresponding Source" for a work in object code form means all 135 | the source code needed to generate, install, and (for an executable 136 | work) run the object code and to modify the work, including scripts to 137 | control those activities. However, it does not include the work's 138 | System Libraries, or general-purpose tools or generally available free 139 | programs which are used unmodified in performing those activities but 140 | which are not part of the work. For example, Corresponding Source 141 | includes interface definition files associated with source files for 142 | the work, and the source code for shared libraries and dynamically 143 | linked subprograms that the work is specifically designed to require, 144 | such as by intimate data communication or control flow between those 145 | subprograms and other parts of the work. 146 | 147 | The Corresponding Source need not include anything that users 148 | can regenerate automatically from other parts of the Corresponding 149 | Source. 150 | 151 | The Corresponding Source for a work in source code form is that 152 | same work. 153 | 154 | 2. Basic Permissions. 155 | 156 | All rights granted under this License are granted for the term of 157 | copyright on the Program, and are irrevocable provided the stated 158 | conditions are met. This License explicitly affirms your unlimited 159 | permission to run the unmodified Program. The output from running a 160 | covered work is covered by this License only if the output, given its 161 | content, constitutes a covered work. This License acknowledges your 162 | rights of fair use or other equivalent, as provided by copyright law. 163 | 164 | You may make, run and propagate covered works that you do not 165 | convey, without conditions so long as your license otherwise remains 166 | in force. You may convey covered works to others for the sole purpose 167 | of having them make modifications exclusively for you, or provide you 168 | with facilities for running those works, provided that you comply with 169 | the terms of this License in conveying all material for which you do 170 | not control copyright. Those thus making or running the covered works 171 | for you must do so exclusively on your behalf, under your direction 172 | and control, on terms that prohibit them from making any copies of 173 | your copyrighted material outside their relationship with you. 174 | 175 | Conveying under any other circumstances is permitted solely under 176 | the conditions stated below. Sublicensing is not allowed; section 10 177 | makes it unnecessary. 178 | 179 | 3. Protecting Users' Legal Rights From Anti-Circumvention Law. 180 | 181 | No covered work shall be deemed part of an effective technological 182 | measure under any applicable law fulfilling obligations under article 183 | 11 of the WIPO copyright treaty adopted on 20 December 1996, or 184 | similar laws prohibiting or restricting circumvention of such 185 | measures. 186 | 187 | When you convey a covered work, you waive any legal power to forbid 188 | circumvention of technological measures to the extent such circumvention 189 | is effected by exercising rights under this License with respect to 190 | the covered work, and you disclaim any intention to limit operation or 191 | modification of the work as a means of enforcing, against the work's 192 | users, your or third parties' legal rights to forbid circumvention of 193 | technological measures. 194 | 195 | 4. Conveying Verbatim Copies. 196 | 197 | You may convey verbatim copies of the Program's source code as you 198 | receive it, in any medium, provided that you conspicuously and 199 | appropriately publish on each copy an appropriate copyright notice; 200 | keep intact all notices stating that this License and any 201 | non-permissive terms added in accord with section 7 apply to the code; 202 | keep intact all notices of the absence of any warranty; and give all 203 | recipients a copy of this License along with the Program. 204 | 205 | You may charge any price or no price for each copy that you convey, 206 | and you may offer support or warranty protection for a fee. 207 | 208 | 5. Conveying Modified Source Versions. 209 | 210 | You may convey a work based on the Program, or the modifications to 211 | produce it from the Program, in the form of source code under the 212 | terms of section 4, provided that you also meet all of these conditions: 213 | 214 | a) The work must carry prominent notices stating that you modified 215 | it, and giving a relevant date. 216 | 217 | b) The work must carry prominent notices stating that it is 218 | released under this License and any conditions added under section 219 | 7. This requirement modifies the requirement in section 4 to 220 | "keep intact all notices". 221 | 222 | c) You must license the entire work, as a whole, under this 223 | License to anyone who comes into possession of a copy. This 224 | License will therefore apply, along with any applicable section 7 225 | additional terms, to the whole of the work, and all its parts, 226 | regardless of how they are packaged. This License gives no 227 | permission to license the work in any other way, but it does not 228 | invalidate such permission if you have separately received it. 229 | 230 | d) If the work has interactive user interfaces, each must display 231 | Appropriate Legal Notices; however, if the Program has interactive 232 | interfaces that do not display Appropriate Legal Notices, your 233 | work need not make them do so. 234 | 235 | A compilation of a covered work with other separate and independent 236 | works, which are not by their nature extensions of the covered work, 237 | and which are not combined with it such as to form a larger program, 238 | in or on a volume of a storage or distribution medium, is called an 239 | "aggregate" if the compilation and its resulting copyright are not 240 | used to limit the access or legal rights of the compilation's users 241 | beyond what the individual works permit. Inclusion of a covered work 242 | in an aggregate does not cause this License to apply to the other 243 | parts of the aggregate. 244 | 245 | 6. Conveying Non-Source Forms. 246 | 247 | You may convey a covered work in object code form under the terms 248 | of sections 4 and 5, provided that you also convey the 249 | machine-readable Corresponding Source under the terms of this License, 250 | in one of these ways: 251 | 252 | a) Convey the object code in, or embodied in, a physical product 253 | (including a physical distribution medium), accompanied by the 254 | Corresponding Source fixed on a durable physical medium 255 | customarily used for software interchange. 256 | 257 | b) Convey the object code in, or embodied in, a physical product 258 | (including a physical distribution medium), accompanied by a 259 | written offer, valid for at least three years and valid for as 260 | long as you offer spare parts or customer support for that product 261 | model, to give anyone who possesses the object code either (1) a 262 | copy of the Corresponding Source for all the software in the 263 | product that is covered by this License, on a durable physical 264 | medium customarily used for software interchange, for a price no 265 | more than your reasonable cost of physically performing this 266 | conveying of source, or (2) access to copy the 267 | Corresponding Source from a network server at no charge. 268 | 269 | c) Convey individual copies of the object code with a copy of the 270 | written offer to provide the Corresponding Source. This 271 | alternative is allowed only occasionally and noncommercially, and 272 | only if you received the object code with such an offer, in accord 273 | with subsection 6b. 274 | 275 | d) Convey the object code by offering access from a designated 276 | place (gratis or for a charge), and offer equivalent access to the 277 | Corresponding Source in the same way through the same place at no 278 | further charge. You need not require recipients to copy the 279 | Corresponding Source along with the object code. If the place to 280 | copy the object code is a network server, the Corresponding Source 281 | may be on a different server (operated by you or a third party) 282 | that supports equivalent copying facilities, provided you maintain 283 | clear directions next to the object code saying where to find the 284 | Corresponding Source. Regardless of what server hosts the 285 | Corresponding Source, you remain obligated to ensure that it is 286 | available for as long as needed to satisfy these requirements. 287 | 288 | e) Convey the object code using peer-to-peer transmission, provided 289 | you inform other peers where the object code and Corresponding 290 | Source of the work are being offered to the general public at no 291 | charge under subsection 6d. 292 | 293 | A separable portion of the object code, whose source code is excluded 294 | from the Corresponding Source as a System Library, need not be 295 | included in conveying the object code work. 296 | 297 | A "User Product" is either (1) a "consumer product", which means any 298 | tangible personal property which is normally used for personal, family, 299 | or household purposes, or (2) anything designed or sold for incorporation 300 | into a dwelling. In determining whether a product is a consumer product, 301 | doubtful cases shall be resolved in favor of coverage. For a particular 302 | product received by a particular user, "normally used" refers to a 303 | typical or common use of that class of product, regardless of the status 304 | of the particular user or of the way in which the particular user 305 | actually uses, or expects or is expected to use, the product. A product 306 | is a consumer product regardless of whether the product has substantial 307 | commercial, industrial or non-consumer uses, unless such uses represent 308 | the only significant mode of use of the product. 309 | 310 | "Installation Information" for a User Product means any methods, 311 | procedures, authorization keys, or other information required to install 312 | and execute modified versions of a covered work in that User Product from 313 | a modified version of its Corresponding Source. The information must 314 | suffice to ensure that the continued functioning of the modified object 315 | code is in no case prevented or interfered with solely because 316 | modification has been made. 317 | 318 | If you convey an object code work under this section in, or with, or 319 | specifically for use in, a User Product, and the conveying occurs as 320 | part of a transaction in which the right of possession and use of the 321 | User Product is transferred to the recipient in perpetuity or for a 322 | fixed term (regardless of how the transaction is characterized), the 323 | Corresponding Source conveyed under this section must be accompanied 324 | by the Installation Information. But this requirement does not apply 325 | if neither you nor any third party retains the ability to install 326 | modified object code on the User Product (for example, the work has 327 | been installed in ROM). 328 | 329 | The requirement to provide Installation Information does not include a 330 | requirement to continue to provide support service, warranty, or updates 331 | for a work that has been modified or installed by the recipient, or for 332 | the User Product in which it has been modified or installed. Access to a 333 | network may be denied when the modification itself materially and 334 | adversely affects the operation of the network or violates the rules and 335 | protocols for communication across the network. 336 | 337 | Corresponding Source conveyed, and Installation Information provided, 338 | in accord with this section must be in a format that is publicly 339 | documented (and with an implementation available to the public in 340 | source code form), and must require no special password or key for 341 | unpacking, reading or copying. 342 | 343 | 7. Additional Terms. 344 | 345 | "Additional permissions" are terms that supplement the terms of this 346 | License by making exceptions from one or more of its conditions. 347 | Additional permissions that are applicable to the entire Program shall 348 | be treated as though they were included in this License, to the extent 349 | that they are valid under applicable law. If additional permissions 350 | apply only to part of the Program, that part may be used separately 351 | under those permissions, but the entire Program remains governed by 352 | this License without regard to the additional permissions. 353 | 354 | When you convey a copy of a covered work, you may at your option 355 | remove any additional permissions from that copy, or from any part of 356 | it. (Additional permissions may be written to require their own 357 | removal in certain cases when you modify the work.) You may place 358 | additional permissions on material, added by you to a covered work, 359 | for which you have or can give appropriate copyright permission. 360 | 361 | Notwithstanding any other provision of this License, for material you 362 | add to a covered work, you may (if authorized by the copyright holders of 363 | that material) supplement the terms of this License with terms: 364 | 365 | a) Disclaiming warranty or limiting liability differently from the 366 | terms of sections 15 and 16 of this License; or 367 | 368 | b) Requiring preservation of specified reasonable legal notices or 369 | author attributions in that material or in the Appropriate Legal 370 | Notices displayed by works containing it; or 371 | 372 | c) Prohibiting misrepresentation of the origin of that material, or 373 | requiring that modified versions of such material be marked in 374 | reasonable ways as different from the original version; or 375 | 376 | d) Limiting the use for publicity purposes of names of licensors or 377 | authors of the material; or 378 | 379 | e) Declining to grant rights under trademark law for use of some 380 | trade names, trademarks, or service marks; or 381 | 382 | f) Requiring indemnification of licensors and authors of that 383 | material by anyone who conveys the material (or modified versions of 384 | it) with contractual assumptions of liability to the recipient, for 385 | any liability that these contractual assumptions directly impose on 386 | those licensors and authors. 387 | 388 | All other non-permissive additional terms are considered "further 389 | restrictions" within the meaning of section 10. If the Program as you 390 | received it, or any part of it, contains a notice stating that it is 391 | governed by this License along with a term that is a further 392 | restriction, you may remove that term. If a license document contains 393 | a further restriction but permits relicensing or conveying under this 394 | License, you may add to a covered work material governed by the terms 395 | of that license document, provided that the further restriction does 396 | not survive such relicensing or conveying. 397 | 398 | If you add terms to a covered work in accord with this section, you 399 | must place, in the relevant source files, a statement of the 400 | additional terms that apply to those files, or a notice indicating 401 | where to find the applicable terms. 402 | 403 | Additional terms, permissive or non-permissive, may be stated in the 404 | form of a separately written license, or stated as exceptions; 405 | the above requirements apply either way. 406 | 407 | 8. Termination. 408 | 409 | You may not propagate or modify a covered work except as expressly 410 | provided under this License. Any attempt otherwise to propagate or 411 | modify it is void, and will automatically terminate your rights under 412 | this License (including any patent licenses granted under the third 413 | paragraph of section 11). 414 | 415 | However, if you cease all violation of this License, then your 416 | license from a particular copyright holder is reinstated (a) 417 | provisionally, unless and until the copyright holder explicitly and 418 | finally terminates your license, and (b) permanently, if the copyright 419 | holder fails to notify you of the violation by some reasonable means 420 | prior to 60 days after the cessation. 421 | 422 | Moreover, your license from a particular copyright holder is 423 | reinstated permanently if the copyright holder notifies you of the 424 | violation by some reasonable means, this is the first time you have 425 | received notice of violation of this License (for any work) from that 426 | copyright holder, and you cure the violation prior to 30 days after 427 | your receipt of the notice. 428 | 429 | Termination of your rights under this section does not terminate the 430 | licenses of parties who have received copies or rights from you under 431 | this License. If your rights have been terminated and not permanently 432 | reinstated, you do not qualify to receive new licenses for the same 433 | material under section 10. 434 | 435 | 9. Acceptance Not Required for Having Copies. 436 | 437 | You are not required to accept this License in order to receive or 438 | run a copy of the Program. Ancillary propagation of a covered work 439 | occurring solely as a consequence of using peer-to-peer transmission 440 | to receive a copy likewise does not require acceptance. However, 441 | nothing other than this License grants you permission to propagate or 442 | modify any covered work. These actions infringe copyright if you do 443 | not accept this License. Therefore, by modifying or propagating a 444 | covered work, you indicate your acceptance of this License to do so. 445 | 446 | 10. Automatic Licensing of Downstream Recipients. 447 | 448 | Each time you convey a covered work, the recipient automatically 449 | receives a license from the original licensors, to run, modify and 450 | propagate that work, subject to this License. You are not responsible 451 | for enforcing compliance by third parties with this License. 452 | 453 | An "entity transaction" is a transaction transferring control of an 454 | organization, or substantially all assets of one, or subdividing an 455 | organization, or merging organizations. If propagation of a covered 456 | work results from an entity transaction, each party to that 457 | transaction who receives a copy of the work also receives whatever 458 | licenses to the work the party's predecessor in interest had or could 459 | give under the previous paragraph, plus a right to possession of the 460 | Corresponding Source of the work from the predecessor in interest, if 461 | the predecessor has it or can get it with reasonable efforts. 462 | 463 | You may not impose any further restrictions on the exercise of the 464 | rights granted or affirmed under this License. For example, you may 465 | not impose a license fee, royalty, or other charge for exercise of 466 | rights granted under this License, and you may not initiate litigation 467 | (including a cross-claim or counterclaim in a lawsuit) alleging that 468 | any patent claim is infringed by making, using, selling, offering for 469 | sale, or importing the Program or any portion of it. 470 | 471 | 11. Patents. 472 | 473 | A "contributor" is a copyright holder who authorizes use under this 474 | License of the Program or a work on which the Program is based. The 475 | work thus licensed is called the contributor's "contributor version". 476 | 477 | A contributor's "essential patent claims" are all patent claims 478 | owned or controlled by the contributor, whether already acquired or 479 | hereafter acquired, that would be infringed by some manner, permitted 480 | by this License, of making, using, or selling its contributor version, 481 | but do not include claims that would be infringed only as a 482 | consequence of further modification of the contributor version. For 483 | purposes of this definition, "control" includes the right to grant 484 | patent sublicenses in a manner consistent with the requirements of 485 | this License. 486 | 487 | Each contributor grants you a non-exclusive, worldwide, royalty-free 488 | patent license under the contributor's essential patent claims, to 489 | make, use, sell, offer for sale, import and otherwise run, modify and 490 | propagate the contents of its contributor version. 491 | 492 | In the following three paragraphs, a "patent license" is any express 493 | agreement or commitment, however denominated, not to enforce a patent 494 | (such as an express permission to practice a patent or covenant not to 495 | sue for patent infringement). To "grant" such a patent license to a 496 | party means to make such an agreement or commitment not to enforce a 497 | patent against the party. 498 | 499 | If you convey a covered work, knowingly relying on a patent license, 500 | and the Corresponding Source of the work is not available for anyone 501 | to copy, free of charge and under the terms of this License, through a 502 | publicly available network server or other readily accessible means, 503 | then you must either (1) cause the Corresponding Source to be so 504 | available, or (2) arrange to deprive yourself of the benefit of the 505 | patent license for this particular work, or (3) arrange, in a manner 506 | consistent with the requirements of this License, to extend the patent 507 | license to downstream recipients. "Knowingly relying" means you have 508 | actual knowledge that, but for the patent license, your conveying the 509 | covered work in a country, or your recipient's use of the covered work 510 | in a country, would infringe one or more identifiable patents in that 511 | country that you have reason to believe are valid. 512 | 513 | If, pursuant to or in connection with a single transaction or 514 | arrangement, you convey, or propagate by procuring conveyance of, a 515 | covered work, and grant a patent license to some of the parties 516 | receiving the covered work authorizing them to use, propagate, modify 517 | or convey a specific copy of the covered work, then the patent license 518 | you grant is automatically extended to all recipients of the covered 519 | work and works based on it. 520 | 521 | A patent license is "discriminatory" if it does not include within 522 | the scope of its coverage, prohibits the exercise of, or is 523 | conditioned on the non-exercise of one or more of the rights that are 524 | specifically granted under this License. You may not convey a covered 525 | work if you are a party to an arrangement with a third party that is 526 | in the business of distributing software, under which you make payment 527 | to the third party based on the extent of your activity of conveying 528 | the work, and under which the third party grants, to any of the 529 | parties who would receive the covered work from you, a discriminatory 530 | patent license (a) in connection with copies of the covered work 531 | conveyed by you (or copies made from those copies), or (b) primarily 532 | for and in connection with specific products or compilations that 533 | contain the covered work, unless you entered into that arrangement, 534 | or that patent license was granted, prior to 28 March 2007. 535 | 536 | Nothing in this License shall be construed as excluding or limiting 537 | any implied license or other defenses to infringement that may 538 | otherwise be available to you under applicable patent law. 539 | 540 | 12. No Surrender of Others' Freedom. 541 | 542 | If conditions are imposed on you (whether by court order, agreement or 543 | otherwise) that contradict the conditions of this License, they do not 544 | excuse you from the conditions of this License. If you cannot convey a 545 | covered work so as to satisfy simultaneously your obligations under this 546 | License and any other pertinent obligations, then as a consequence you may 547 | not convey it at all. For example, if you agree to terms that obligate you 548 | to collect a royalty for further conveying from those to whom you convey 549 | the Program, the only way you could satisfy both those terms and this 550 | License would be to refrain entirely from conveying the Program. 551 | 552 | 13. Use with the GNU Affero General Public License. 553 | 554 | Notwithstanding any other provision of this License, you have 555 | permission to link or combine any covered work with a work licensed 556 | under version 3 of the GNU Affero General Public License into a single 557 | combined work, and to convey the resulting work. The terms of this 558 | License will continue to apply to the part which is the covered work, 559 | but the special requirements of the GNU Affero General Public License, 560 | section 13, concerning interaction through a network will apply to the 561 | combination as such. 562 | 563 | 14. Revised Versions of this License. 564 | 565 | The Free Software Foundation may publish revised and/or new versions of 566 | the GNU General Public License from time to time. Such new versions will 567 | be similar in spirit to the present version, but may differ in detail to 568 | address new problems or concerns. 569 | 570 | Each version is given a distinguishing version number. If the 571 | Program specifies that a certain numbered version of the GNU General 572 | Public License "or any later version" applies to it, you have the 573 | option of following the terms and conditions either of that numbered 574 | version or of any later version published by the Free Software 575 | Foundation. If the Program does not specify a version number of the 576 | GNU General Public License, you may choose any version ever published 577 | by the Free Software Foundation. 578 | 579 | If the Program specifies that a proxy can decide which future 580 | versions of the GNU General Public License can be used, that proxy's 581 | public statement of acceptance of a version permanently authorizes you 582 | to choose that version for the Program. 583 | 584 | Later license versions may give you additional or different 585 | permissions. However, no additional obligations are imposed on any 586 | author or copyright holder as a result of your choosing to follow a 587 | later version. 588 | 589 | 15. Disclaimer of Warranty. 590 | 591 | THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY 592 | APPLICABLE LAW. EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT 593 | HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY 594 | OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, 595 | THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 596 | PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM 597 | IS WITH YOU. SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF 598 | ALL NECESSARY SERVICING, REPAIR OR CORRECTION. 599 | 600 | 16. Limitation of Liability. 601 | 602 | IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING 603 | WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS 604 | THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY 605 | GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE 606 | USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF 607 | DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD 608 | PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS), 609 | EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF 610 | SUCH DAMAGES. 611 | 612 | 17. Interpretation of Sections 15 and 16. 613 | 614 | If the disclaimer of warranty and limitation of liability provided 615 | above cannot be given local legal effect according to their terms, 616 | reviewing courts shall apply local law that most closely approximates 617 | an absolute waiver of all civil liability in connection with the 618 | Program, unless a warranty or assumption of liability accompanies a 619 | copy of the Program in return for a fee. 620 | 621 | END OF TERMS AND CONDITIONS 622 | 623 | How to Apply These Terms to Your New Programs 624 | 625 | If you develop a new program, and you want it to be of the greatest 626 | possible use to the public, the best way to achieve this is to make it 627 | free software which everyone can redistribute and change under these terms. 628 | 629 | To do so, attach the following notices to the program. It is safest 630 | to attach them to the start of each source file to most effectively 631 | state the exclusion of warranty; and each file should have at least 632 | the "copyright" line and a pointer to where the full notice is found. 633 | 634 | 635 | Copyright (C) 636 | 637 | This program is free software: you can redistribute it and/or modify 638 | it under the terms of the GNU General Public License as published by 639 | the Free Software Foundation, either version 3 of the License, or 640 | (at your option) any later version. 641 | 642 | This program is distributed in the hope that it will be useful, 643 | but WITHOUT ANY WARRANTY; without even the implied warranty of 644 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 645 | GNU General Public License for more details. 646 | 647 | You should have received a copy of the GNU General Public License 648 | along with this program. If not, see . 649 | 650 | Also add information on how to contact you by electronic and paper mail. 651 | 652 | If the program does terminal interaction, make it output a short 653 | notice like this when it starts in an interactive mode: 654 | 655 | Copyright (C) 656 | This program comes with ABSOLUTELY NO WARRANTY; for details type `show w'. 657 | This is free software, and you are welcome to redistribute it 658 | under certain conditions; type `show c' for details. 659 | 660 | The hypothetical commands `show w' and `show c' should show the appropriate 661 | parts of the General Public License. Of course, your program's commands 662 | might be different; for a GUI interface, you would use an "about box". 663 | 664 | You should also get your employer (if you work as a programmer) or school, 665 | if any, to sign a "copyright disclaimer" for the program, if necessary. 666 | For more information on this, and how to apply and follow the GNU GPL, see 667 | . 668 | 669 | The GNU General Public License does not permit incorporating your program 670 | into proprietary programs. If your program is a subroutine library, you 671 | may consider it more useful to permit linking proprietary applications with 672 | the library. If this is what you want to do, use the GNU Lesser General 673 | Public License instead of this License. But first, please read 674 | . 675 | -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | # Introduction 2 | 3 | [![Build Status](https://travis-ci.org/spance/suft.svg)](https://travis-ci.org/spance/suft) 4 | [![GoDoc](https://godoc.org/github.com/spance/suft/protocol?status.svg)](https://godoc.org/github.com/spance/suft/protocol) 5 | 6 | The SUFT (Small-scale UDP Fast Transmission) Protocol is an application layer transmission protocol based on UDP and implemented in Golang. It has lower latency than TCP and provides reliable and ordered delivery of a stream of octets under plain congestion control. 7 | 8 | The protocol seeks for maximized throughput and minimizes impact of lost packets on throughput. It is only for small/medium-scale communication or some situations where TCP is not applicable. 9 | 10 | # Goals & Features 11 | 12 | - Transmitting model has predictable performance. 13 | - Fast retransmission mode does better on lossy link. 14 | - Minimum retransmission mode doesn't waste traffic. 15 | - No resource consumption while the connection is idle. 16 | - Special modes for certain situations. 17 | 18 | # Protocol APIs 19 | 20 | SUFT implements the Golang: [net.Conn](https://golang.org/pkg/net/#Conn) and [net.Listener](https://golang.org/pkg/net/#Listener) interfaces completely. 21 | 22 | ```go 23 | import "github.com/spance/suft/protocol" 24 | 25 | e, err := suft.NewEndpoint(p *suft.Params) 26 | // for server 27 | conn := e.Listen() // or e.Accept() 28 | // for client 29 | conn, err := e.Dial(rAddr string) 30 | // your business ... 31 | conn.Close() 32 | e.Close() 33 | ``` 34 | 35 | # Basic Theories 36 | 37 | scnt(sender: the count of unique data packets) and dups(sender: retransmission counter) 38 | 39 | ``` 40 | Lose Rate = dups/(scnt + dups) 41 | 42 | Retransmit Rate = dups/scnt 43 | ``` 44 | 45 | latency, window and traffic speed 46 | 47 | ``` 48 | Speed = (1000 / latency) * mss * win 49 | ``` 50 | 51 | # License 52 | 53 | GPL version 3 or any later version 54 | 55 | SUFT Protocol 56 | Copyright (C) 2016 spance, l2dy 57 | 58 | This program is free software: you can redistribute it and/or modify 59 | it under the terms of the GNU General Public License as published by 60 | the Free Software Foundation, either version 3 of the License, or 61 | (at your option) any later version. 62 | 63 | This program is distributed in the hope that it will be useful, 64 | but WITHOUT ANY WARRANTY; without even the implied warranty of 65 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 66 | GNU General Public License for more details. 67 | 68 | You should have received a copy of the GNU General Public License 69 | along with this program. If not, see . 70 | 71 | # Tool Usage 72 | 73 | "suft-nc" package include a tool for testing, similar to netcat (nc). 74 | 75 | Build with `go get -u -v github.com/spance/suft/examples/suft-nc` 76 | 77 | ``` 78 | ./suft-nc [-l addr:port] [-r addr:port] [-s] [-b 10] [-fr] < [send_file] > [recv_file] 79 | 80 | -l: local bind address, e.g. localhost:9090 or :8080 81 | -r: remote address (for client), e.g. 8.8.8.8:9090 or examples.com:8080 82 | -b: max bandwidth of sending in mbps (be careful, see Notes#2) 83 | -s: for server 84 | -fr: enable fast retransmission (useful for lossy link) 85 | -ft: flat traffic (slow down bursty traffic, useful when sender has more bandwidth than receiver) 86 | ``` 87 | 88 | Examples: 89 | 90 | ``` 91 | // send my_file to remote in 10mbps 92 | remote# ./suft-nc -l :9090 -s > recv_file 93 | local# ./suft-nc -r remote:9090 -b 10 -fr < my_file 94 | ``` 95 | 96 | ``` 97 | // recv my_file from remote in 50mbps 98 | remote# ./suft-nc -l :9090 -s -b 50 -fr < my_file 99 | local# ./suft-nc -r remote:9090 > recv_file 100 | ``` 101 | 102 | ``` 103 | // simple chat room 104 | remote# ./suft-nc -l :9090 -s 105 | local# ./suft-nc -r remote:9090 106 | ``` 107 | 108 | Notes: 109 | 110 | 1. The target to be connected shouldn't be behind NAT (or should use port mapping). 111 | 2. Use improper bandwidth(-b) may waste huge bandwidth and may be suspected of carrying out flood attack. 112 | 113 | # How to test? 114 | 115 | ``` 116 | // remote, send 100MB stream in 50mbps (your bandwidth bottleneck) 117 | remote# dd if=/dev/zero bs=1M count=100 status=none | ./suft-nc -s -l :9090 -b 50 -fr -ft 118 | // status=none or status=noxfer 119 | 120 | // local console#1, monitor traffic(for simplicity, we can use bmon) 121 | local# bmon -p eth0 122 | // local console#2, recv stream 123 | local# ./suft-nc -r remote:9090 > /dev/null 124 | ``` 125 | -------------------------------------------------------------------------------- /cal_speed.py: -------------------------------------------------------------------------------- 1 | #!/usr/bin/python 2 | import sys 3 | 4 | def speed(latency, win): 5 | return (1000.0/latency)*win*1400.0/1024.0 6 | 7 | if len(sys.argv) == 3: 8 | latency = int(sys.argv[1]) 9 | win = int(sys.argv[2]) 10 | print "Theoretical speed %.2fKB/s" % speed(latency, win) 11 | else: 12 | print "%s [latency] [win]" % sys.argv[0] 13 | 14 | -------------------------------------------------------------------------------- /checkack.py: -------------------------------------------------------------------------------- 1 | #!/usr/bin/python 2 | import sys 3 | 4 | args=sys.argv[1:] 5 | if len(args) < 2: 6 | print 'args: ackNo bitmap...' 7 | sys.exit(1) 8 | 9 | on=int(args[0]) 10 | 11 | args=''.join(args[1:]) 12 | args=args.replace('[', '') 13 | args=args.replace(']', '') 14 | print args 15 | 16 | def getBits(): 17 | global args 18 | bits=list(args[:16]) 19 | args=args[16:] 20 | for i in range(8): 21 | bits[i]=bits[14-i] 22 | bits[i+1]=bits[15-i] 23 | return int(''.join(bits), 16) 24 | 25 | while len(args)>0: 26 | bits=getBits() 27 | for i in range(64): 28 | print '%d %s' % (on, '.' if bits&1 else 'miss') 29 | bits>>=1 30 | on+=1 31 | 32 | -------------------------------------------------------------------------------- /docs/suft-state.graphml: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17 | 18 | 19 | 20 | 21 | 22 | 23 | FIN 24 | 25 | 26 | 27 | 28 | 29 | 30 | 31 | 32 | 33 | 34 | 35 | 36 | 37 | 38 | 39 | 40 | SYN0 41 | 42 | 43 | 44 | 45 | 46 | 47 | 48 | 49 | 50 | 51 | 52 | 53 | 54 | 55 | 56 | 57 | SYN1 58 | 59 | 60 | 61 | 62 | 63 | 64 | 65 | 66 | 67 | 68 | 69 | 70 | 71 | 72 | 73 | 74 | SYN1 75 | 76 | 77 | 78 | 79 | 80 | 81 | 82 | 83 | 84 | 85 | 86 | 87 | 88 | 89 | 90 | 91 | EST1 92 | 93 | 94 | 95 | 96 | 97 | 98 | 99 | 100 | 101 | 102 | 103 | 104 | 105 | 106 | 107 | 108 | EST1 109 | 110 | 111 | 112 | 113 | 114 | 115 | 116 | 117 | 118 | 119 | 120 | 121 | 122 | 123 | 124 | 125 | Working 126 | 127 | 128 | 129 | 130 | 131 | 132 | 133 | 134 | 135 | 136 | 137 | 138 | 139 | 140 | 141 | 142 | FIN0 143 | 144 | 145 | 146 | 147 | 148 | 149 | 150 | 151 | 152 | 153 | 154 | 155 | 156 | 157 | 158 | 159 | FIN1 160 | 161 | 162 | 163 | 164 | 165 | 166 | 167 | 168 | 169 | 170 | 171 | 172 | 173 | 174 | 175 | 176 | FIN 177 | 178 | 179 | 180 | 181 | 182 | 183 | 184 | 185 | 186 | 187 | 188 | 189 | 190 | 191 | 192 | 193 | FIN1 194 | 195 | 196 | 197 | 198 | 199 | 200 | 201 | 202 | 203 | 204 | 205 | 206 | 207 | 208 | 209 | 210 | FIN1 211 | 212 | 213 | 214 | 215 | 216 | 217 | 218 | 219 | 220 | 221 | 222 | 223 | 224 | 225 | 226 | 227 | FIN0 228 | 229 | 230 | 231 | 232 | 233 | 234 | 235 | 236 | 237 | 238 | 239 | 240 | 241 | 242 | 243 | 244 | FIN1 245 | 246 | 247 | 248 | 249 | 250 | 251 | 252 | 253 | 254 | 255 | 256 | 257 | 258 | 259 | 260 | 261 | Listen 262 | 263 | 264 | 265 | 266 | 267 | 268 | 269 | 270 | 271 | 272 | 273 | 274 | 275 | 276 | 277 | 278 | Dial 279 | 280 | 281 | 282 | 283 | 284 | 285 | 286 | 287 | 288 | 289 | 290 | 291 | 292 | 293 | 294 | 295 | 296 | send syn, waiting syn+ack 297 | 298 | 299 | 300 | 301 | 302 | 303 | 304 | 305 | 306 | 307 | 308 | 309 | 310 | 311 | 312 | 313 | 314 | reply syn+ack, waiting ack 315 | 316 | 317 | 318 | 319 | 320 | 321 | 322 | 323 | 324 | 325 | 326 | 327 | 328 | 329 | 330 | 331 | 332 | send ack 333 | 334 | 335 | 336 | 337 | 338 | 339 | 340 | 341 | 342 | 343 | 344 | 345 | 346 | 347 | 348 | 349 | 350 | 351 | 352 | 353 | 354 | 355 | 356 | 357 | 358 | 359 | 360 | 361 | 362 | 363 | 364 | 365 | 366 | 367 | 368 | 369 | 370 | active close 371 | 372 | 373 | 374 | 375 | 376 | 377 | 378 | 379 | 380 | 381 | 382 | 383 | 384 | 385 | 386 | 387 | 388 | reply ack 389 | 390 | 391 | 392 | 393 | 394 | 395 | 396 | 397 | 398 | 399 | 400 | 401 | 402 | 403 | 404 | 405 | 406 | passsive close 407 | 408 | 409 | 410 | 411 | 412 | 413 | 414 | 415 | 416 | 417 | 418 | 419 | 420 | 421 | 422 | 423 | 424 | recv ack 425 | 426 | 427 | 428 | 429 | 430 | 431 | 432 | 433 | 434 | 435 | 436 | 437 | 438 | 439 | 440 | 441 | 442 | Close/W 443 | send fin, waiting ack 444 | 445 | 446 | 447 | 448 | 449 | 450 | 451 | 452 | 453 | 454 | 455 | 456 | 457 | 458 | 459 | 460 | 461 | recv fin 462 | Close/R 463 | 464 | 465 | 466 | 467 | 468 | 469 | 470 | 471 | 472 | 473 | 474 | 475 | 476 | 477 | 478 | 479 | 480 | recv fin 481 | Close/R 482 | reply ack 483 | 484 | 485 | 486 | 487 | 488 | 489 | 490 | 491 | 492 | 493 | 494 | 495 | 496 | 497 | 498 | 499 | 500 | Close/W 501 | send fin, waiting ack 502 | 503 | 504 | 505 | 506 | 507 | 508 | 509 | 510 | 511 | 512 | 513 | 514 | 515 | 516 | 517 | 518 | 519 | 520 | 521 | 522 | 523 | 524 | 525 | 526 | 527 | 528 | 529 | recv syn 530 | 531 | 532 | 533 | 534 | 535 | 536 | 537 | 538 | 539 | 540 | 541 | 542 | 543 | 544 | 545 | 546 | -------------------------------------------------------------------------------- /docs/suft-state.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/spance/suft/358fdb24d82d1292cd7e9f824e367936de3ecfee/docs/suft-state.png -------------------------------------------------------------------------------- /examples/rap/rap.go: -------------------------------------------------------------------------------- 1 | package main 2 | 3 | import ( 4 | "flag" 5 | "io" 6 | "log" 7 | "net" 8 | "os" 9 | "time" 10 | 11 | "github.com/spance/suft/protocol" 12 | ) 13 | 14 | func init() { 15 | log.SetOutput(os.Stderr) 16 | log.SetFlags(log.Flags() | log.Lmicroseconds) 17 | } 18 | 19 | func main() { 20 | var raddr string 21 | var p suft.Params 22 | 23 | flag.StringVar(&p.LocalAddr, "l", "", "local") 24 | flag.StringVar(&raddr, "r", "", "AtServer => backend_tcp_peer, AtClient => remote_suft_peer") 25 | flag.BoolVar(&p.IsServ, "s", false, "is server") 26 | flag.BoolVar(&p.FastRetransmit, "fr", true, "FastRetransmit") 27 | flag.BoolVar(&p.FlatTraffic, "ft", true, "FlatTraffic") 28 | flag.Int64Var(&p.Bandwidth, "b", 4, "bandwidth in mbps") 29 | 30 | flag.IntVar(&p.Debug, "debug", 0, "debug") 31 | flag.BoolVar(&p.EnablePprof, "pprof", false, "pprof") 32 | flag.BoolVar(&p.Stacktrace, "stacktrace", false, "stacktrace") 33 | flag.Parse() 34 | 35 | if !p.IsServ && raddr == "" { 36 | log.Fatalln("missing -r") 37 | } 38 | 39 | e, err := suft.NewEndpoint(&p) 40 | checkErr(err) 41 | defer e.Close() 42 | log.Println("start", e.Addr()) 43 | 44 | var suConn *suft.Conn 45 | if !p.IsServ { // client 46 | ln, err := net.Listen("tcp", p.LocalAddr) 47 | checkErr(err) 48 | defer ln.Close() 49 | 50 | for { 51 | local, err := ln.Accept() 52 | checkErr(err) 53 | suConn, err = e.Dial(raddr) 54 | if checkWarn(err) { 55 | log.Printf("connected to %s for %s", suConn.RemoteAddr(), local.RemoteAddr()) 56 | go duplexPipe(suConn, local) 57 | } else { 58 | safeClose(suConn) 59 | } 60 | } 61 | } else { 62 | for { 63 | suConn = e.Listen() 64 | backend, err := net.Dial("tcp", raddr) 65 | if checkWarn(err) { 66 | log.Printf("connected from %s to %s", suConn.RemoteAddr(), backend.RemoteAddr()) 67 | go duplexPipe(suConn, backend) 68 | } else { 69 | safeClose(backend) 70 | } 71 | } 72 | } 73 | } 74 | 75 | func checkErr(e error) { 76 | if e != nil { 77 | log.Panicln(e) 78 | } 79 | } 80 | 81 | func checkWarn(e error) bool { 82 | if e != nil { 83 | log.Println(e) 84 | return false 85 | } 86 | return true 87 | } 88 | 89 | func safeClose(c io.Closer) { 90 | if c != nil { 91 | c.Close() 92 | } 93 | } 94 | 95 | func duplexPipe(s *suft.Conn, t net.Conn) { 96 | const BUF_SIZE = 1 << 20 97 | t.(*net.TCPConn).SetNoDelay(true) 98 | defer t.Close() 99 | defer s.Close() 100 | 101 | var wait = make(chan byte, 2) 102 | go func() { 103 | buf := make([]byte, BUF_SIZE) 104 | io.CopyBuffer(s, t, buf) 105 | wait <- 1 106 | }() 107 | go func() { 108 | buf := make([]byte, BUF_SIZE) 109 | io.CopyBuffer(t, s, buf) 110 | wait <- 1 111 | }() 112 | // one channel closed 113 | <-wait 114 | select { 115 | case <-wait: // second 116 | case <-time.After(10e9): 117 | } 118 | } 119 | -------------------------------------------------------------------------------- /examples/suft-nc/main.go: -------------------------------------------------------------------------------- 1 | package main 2 | 3 | import ( 4 | "flag" 5 | "fmt" 6 | "io" 7 | "log" 8 | "os" 9 | "time" 10 | 11 | "github.com/spance/suft/protocol" 12 | ) 13 | 14 | func init() { 15 | log.SetOutput(os.Stderr) 16 | log.SetFlags(log.Flags() | log.Lmicroseconds) 17 | } 18 | 19 | var timeWaiting int64 20 | var waiting = make(chan *eofStatus, 2) 21 | 22 | func main() { 23 | var raddr string 24 | var p suft.Params 25 | flag.StringVar(&p.LocalAddr, "l", "", "local") 26 | flag.StringVar(&raddr, "r", "", "remote") 27 | flag.BoolVar(&p.IsServ, "s", false, "is server") 28 | flag.BoolVar(&p.FastRetransmit, "fr", true, "FastRetransmit") 29 | flag.BoolVar(&p.FlatTraffic, "ft", true, "FlatTraffic") 30 | flag.Int64Var(&p.Bandwidth, "b", 2, "bandwidth in mbps") 31 | flag.IntVar(&p.Debug, "debug", 0, "debug") 32 | flag.BoolVar(&p.EnablePprof, "pprof", false, "pprof") 33 | flag.BoolVar(&p.Stacktrace, "stacktrace", false, "stacktrace") 34 | flag.Int64Var(&timeWaiting, "w", 0, "Timeout waiting for half-closed connection") 35 | flag.Parse() 36 | 37 | if !p.IsServ && raddr == "" { 38 | log.Fatalln("missing -r") 39 | } 40 | 41 | e, err := suft.NewEndpoint(&p) 42 | checkErr(err) 43 | defer e.Close() 44 | 45 | log.Println("start", e.Addr()) 46 | var conn *suft.Conn 47 | if !p.IsServ { // client 48 | conn, err = e.Dial(raddr) 49 | checkErr(err) 50 | log.Println("connected to", conn.RemoteAddr()) 51 | go writeOut(conn) 52 | go readIn(conn) 53 | } else { 54 | conn = e.Listen() 55 | log.Println("connected from", conn.RemoteAddr()) 56 | go writeOut(conn) 57 | go readIn(conn) 58 | } 59 | 60 | var eof1, eof2 *eofStatus 61 | // there is one direction channel has been released 62 | eof1 = <-waiting 63 | log.Println(eof1.status) 64 | 65 | if timeWaiting > 0 { 66 | for1: 67 | for i, v := range [2]int64{1, timeWaiting} { 68 | select { 69 | case eof2 = <-waiting: 70 | // stdin reader released 71 | break for1 72 | case <-time.After(time.Duration(v * 1e9)): 73 | if i == 0 { // first second for preparing countdown 74 | log.Printf("the countdown to %c has started", "RW"[(eof1.channel+1)%2]) 75 | } else { // kill 76 | conn.Close() 77 | } 78 | } 79 | } 80 | } else { 81 | for2: 82 | for { // special case for close stdin 83 | select { 84 | case eof2 = <-waiting: 85 | // stdin reader released 86 | break for2 87 | case <-time.After(1e9): 88 | // conn is closed but reading stdin was still blocking 89 | if conn.IsClosed() { 90 | break for2 91 | } 92 | } 93 | } 94 | } 95 | 96 | if eof2 != nil { 97 | log.Println(eof2.status) 98 | } 99 | conn.PrintState() 100 | } 101 | 102 | func readIn(c *suft.Conn) { 103 | var ( 104 | n int64 105 | err1, err2 error 106 | ) 107 | wa := suft.StartWatch("R") 108 | n, err1 = io.Copy(c, os.Stdin) 109 | wa.Stop(int(n)) 110 | 111 | if timeWaiting <= 0 { 112 | err2 = c.Close() 113 | } 114 | waiting <- &eofStatus{ 115 | channel: 0, 116 | status: fmt.Sprint("readIn done", err1, err2), 117 | } 118 | } 119 | 120 | func writeOut(c *suft.Conn) { 121 | var ( 122 | n int64 123 | err1, err2 error 124 | ) 125 | wa := suft.StartWatch("W") 126 | n, err1 = io.Copy(os.Stdout, c) 127 | os.Stdout.Sync() 128 | wa.Stop(int(n)) 129 | 130 | if timeWaiting <= 0 { 131 | err2 = c.Close() 132 | } 133 | waiting <- &eofStatus{ 134 | channel: 1, 135 | status: fmt.Sprint("writeOut done", err1, err2), 136 | } 137 | } 138 | 139 | func checkErr(e error) { 140 | if e != nil { 141 | log.Fatalln(e) 142 | } 143 | } 144 | 145 | type eofStatus struct { 146 | channel int 147 | status string 148 | } 149 | -------------------------------------------------------------------------------- /examples/tors/README.md: -------------------------------------------------------------------------------- 1 | # Usage 2 | 3 | 1. for server 4 | 5 | ``` 6 | ./tors -l : -b 7 | ``` 8 | 9 | 2. for client - pull file from remote 10 | 11 | ``` 12 | rsync -P --blocking-io -e './tors -r :' x: . 13 | ``` 14 | 15 | 3. for client - push file to remote 16 | 17 | ``` 18 | rsync -P --blocking-io -e './tors -r : -b ' x: 19 | ``` 20 | -------------------------------------------------------------------------------- /examples/tors/io.go: -------------------------------------------------------------------------------- 1 | // +build !windows 2 | 3 | package main 4 | 5 | import ( 6 | "io" 7 | "os" 8 | ) 9 | 10 | func getStdin() io.Reader { 11 | return os.Stdin 12 | } 13 | 14 | func getStdout() io.Writer { 15 | return os.Stdout 16 | } 17 | -------------------------------------------------------------------------------- /examples/tors/io_windows.go: -------------------------------------------------------------------------------- 1 | package main 2 | 3 | import ( 4 | "io" 5 | "log" 6 | "os" 7 | "syscall" 8 | "unsafe" 9 | ) 10 | 11 | var ( 12 | _CreateEvent *syscall.Proc 13 | _GetOverlappedResult *syscall.Proc 14 | ) 15 | 16 | func init() { 17 | wk, _ := syscall.LoadDLL("kernel32.dll") 18 | _CreateEvent = wk.MustFindProc("CreateEventW") 19 | _GetOverlappedResult = wk.MustFindProc("GetOverlappedResult") 20 | } 21 | 22 | func getStdin() io.Reader { 23 | return newWinAsyncIOFile(os.Stdin.Fd()) 24 | } 25 | 26 | func getStdout() io.Writer { 27 | return newWinAsyncIOFile(os.Stdout.Fd()) 28 | } 29 | 30 | type winAsyncIOFile struct { 31 | hd syscall.Handle 32 | he syscall.Handle 33 | overlapped syscall.Overlapped 34 | } 35 | 36 | func newWinAsyncIOFile(h uintptr) *winAsyncIOFile { 37 | s := &winAsyncIOFile{hd: syscall.Handle(h)} 38 | s.he = createEvent() 39 | s.overlapped.HEvent = s.he 40 | return s 41 | } 42 | 43 | func createEvent() syscall.Handle { 44 | // https://msdn.microsoft.com/en-us/library/windows/desktop/ms682396(v=vs.85).aspx 45 | h, _, e := _CreateEvent.Call(0, 0, 0, 0) 46 | if h == 0 { // return NULL 47 | log.Fatalln("_CreateEventW", e) 48 | } 49 | return syscall.Handle(h) 50 | } 51 | 52 | func (s *winAsyncIOFile) getOverlappedResult(done *uint32) error { 53 | // https://msdn.microsoft.com/en-us/library/windows/desktop/ms683209(v=vs.85).aspx 54 | r, _, en := syscall.Syscall6(_GetOverlappedResult.Addr(), 4, 55 | uintptr(s.hd), 56 | uintptr(unsafe.Pointer(&s.overlapped)), 57 | uintptr(unsafe.Pointer(done)), 58 | 1, 59 | 0, 0) 60 | if r != 0 { 61 | return nil 62 | } else { 63 | // 995=The I/O operation has been aborted because of either a thread exit or an application request. 64 | if en == syscall.ERROR_OPERATION_ABORTED { 65 | return io.EOF 66 | } else { 67 | return en 68 | } 69 | } 70 | } 71 | 72 | func (s *winAsyncIOFile) Read(b []byte) (int, error) { 73 | var done uint32 74 | e := syscall.ReadFile(s.hd, b, &done, &s.overlapped) 75 | if en, y := e.(syscall.Errno); y { 76 | if en == syscall.ERROR_IO_PENDING { 77 | e = s.getOverlappedResult(&done) 78 | } 79 | } 80 | return int(done), e 81 | } 82 | 83 | func (s *winAsyncIOFile) Write(b []byte) (int, error) { 84 | var done uint32 85 | e := syscall.WriteFile(s.hd, b, &done, &s.overlapped) 86 | if en, y := e.(syscall.Errno); y { 87 | if en == syscall.ERROR_IO_PENDING { 88 | e = s.getOverlappedResult(&done) 89 | } 90 | } 91 | return int(done), e 92 | } 93 | 94 | func (s *winAsyncIOFile) Close() error { 95 | return syscall.CancelIoEx(s.hd, &s.overlapped) 96 | } 97 | -------------------------------------------------------------------------------- /examples/tors/main.go: -------------------------------------------------------------------------------- 1 | package main 2 | 3 | import ( 4 | "encoding/binary" 5 | "flag" 6 | "io" 7 | "log" 8 | "net" 9 | "os" 10 | "os/exec" 11 | "strings" 12 | 13 | "github.com/spance/suft/protocol" 14 | ) 15 | 16 | var ( 17 | transMode int 18 | endpoint *suft.Endpoint 19 | ) 20 | 21 | func init() { 22 | log.SetFlags(log.LstdFlags | log.Lshortfile) 23 | log.SetOutput(os.Stderr) 24 | } 25 | 26 | func main() { 27 | var lAddr, rAddr, trans string 28 | var bw int64 29 | flag.StringVar(&lAddr, "l", "", "listen addr for server") 30 | flag.StringVar(&rAddr, "r", "", "remote addr for client") 31 | flag.StringVar(&trans, "t", "suft", "transport layer") 32 | flag.Int64Var(&bw, "b", 4, "bandwidth in mbps") 33 | flag.Parse() 34 | 35 | if (lAddr != "" && rAddr != "") || (lAddr == "" && rAddr == "") { 36 | log.Fatalln("-l for server and -r for client") 37 | } 38 | var isServ bool 39 | var err error 40 | isServ = !(rAddr != "" && flag.NArg() >= 2) 41 | 42 | switch trans { 43 | case "suft": 44 | transMode = 'S' 45 | endpoint, err = suft.NewEndpoint(&suft.Params{ 46 | LocalAddr: lAddr, 47 | Bandwidth: bw, 48 | FastRetransmit: true, 49 | FlatTraffic: true, 50 | IsServ: isServ, 51 | }) 52 | abortIf(err) 53 | 54 | case "tcp": 55 | transMode = 'T' 56 | 57 | default: 58 | log.Fatalln("Unknown -t", trans) 59 | } 60 | 61 | if isServ { 62 | listen(lAddr) 63 | } else { 64 | connect(rAddr, flag.Args()) 65 | } 66 | 67 | } 68 | 69 | func netListen(addr string) net.Listener { 70 | if transMode == 'T' { 71 | ln, err := net.Listen("tcp", addr) 72 | abortIf(err) 73 | return ln 74 | } else { 75 | return endpoint 76 | } 77 | } 78 | 79 | func netDial(addr string) net.Conn { 80 | var conn net.Conn 81 | var err error 82 | if transMode == 'T' { 83 | conn, err = net.Dial("tcp", addr) 84 | } else { 85 | conn, err = endpoint.Dial(addr) 86 | } 87 | abortIf(err) 88 | return conn 89 | } 90 | 91 | func listen(addr string) { 92 | ln := netListen(addr) 93 | defer ln.Close() 94 | log.Println("listen at", ln.Addr()) 95 | 96 | for { 97 | conn, err := ln.Accept() 98 | if err == nil { 99 | log.Println("Client from", conn.RemoteAddr()) 100 | go sessionHandle(conn) 101 | } else if isTemporaryError(err) { 102 | continue 103 | } else { 104 | break 105 | } 106 | } 107 | } 108 | 109 | func sessionHandle(conn net.Conn) { 110 | defer conn.Close() 111 | 112 | cmdline := readInitialHeader(conn) 113 | if !(len(cmdline) > 1 && cmdline[0] == "rsync") { 114 | return 115 | } 116 | log.Printf("RUN %s", cmdline) 117 | 118 | cmd := exec.Command(cmdline[0], cmdline[1:]...) 119 | stdin, err := cmd.StdinPipe() 120 | abortIf(err) 121 | stdout, err := cmd.StdoutPipe() 122 | abortIf(err) 123 | stderr, err := cmd.StderrPipe() 124 | abortIf(err) 125 | abortIf(cmd.Start()) 126 | 127 | go func() { 128 | // conn -> cmd.stdin 129 | n, e := io.Copy(stdin, conn) 130 | e = consumeClosedError(e) 131 | log.Println("net/Rx", n, e) 132 | }() 133 | go func() { 134 | // cmd.stdout -> conn 135 | n, e := io.Copy(conn, stdout) 136 | log.Println("net/Tx", n, e) 137 | }() 138 | // cmd.stderr -> current.stderr 139 | go io.Copy(os.Stderr, stderr) 140 | 141 | err = cmd.Wait() 142 | conn.Close() 143 | log.Println("finished", err) 144 | } 145 | 146 | func connect(addr string, args []string) { 147 | conn := netDial(addr) 148 | defer conn.Close() 149 | writeInitialHeader(conn, args[1:]) 150 | 151 | var done = make(chan int, 1) 152 | var stdin = getStdin() 153 | 154 | go func() { 155 | // stdin -> conn 156 | n, e := io.Copy(conn, stdin) 157 | log.Println("net/Tx", n, e) 158 | done <- 1 159 | }() 160 | go func() { 161 | // conn -> stdout 162 | n, e := io.Copy(getStdout(), conn) 163 | log.Println("net/Rx", n, e) 164 | done <- 2 165 | }() 166 | <-done 167 | } 168 | 169 | func isTemporaryError(err error) bool { 170 | if err != nil { 171 | if ne, y := err.(net.Error); y { 172 | return ne.Temporary() 173 | } 174 | } 175 | return false 176 | } 177 | 178 | func abortIf(err error) { 179 | if err != nil { 180 | log.Fatalln(err) 181 | } 182 | } 183 | 184 | func readInitialHeader(rd net.Conn) []string { 185 | var l4 = make([]byte, 4) 186 | _, err := io.ReadFull(rd, l4) 187 | abortIf(err) 188 | nlen := binary.BigEndian.Uint32(l4) 189 | var cmd = make([]byte, nlen) 190 | _, err = io.ReadFull(rd, cmd) 191 | abortIf(err) 192 | return strings.Split(string(cmd), "\x00") 193 | } 194 | 195 | func writeInitialHeader(rd net.Conn, args []string) { 196 | var cmdline = strings.Join(args, "\x00") 197 | var l4 = make([]byte, 4) 198 | binary.BigEndian.PutUint32(l4, uint32(len(cmdline))) 199 | _, err := rd.Write(l4) 200 | abortIf(err) 201 | _, err = rd.Write([]byte(cmdline)) 202 | abortIf(err) 203 | } 204 | 205 | func consumeClosedError(e error) error { 206 | if e != nil && strings.Contains(e.Error(), "closed") { 207 | return nil 208 | } else { 209 | return e 210 | } 211 | } 212 | -------------------------------------------------------------------------------- /protocol/conn.go: -------------------------------------------------------------------------------- 1 | package suft 2 | 3 | import ( 4 | "encoding/binary" 5 | "fmt" 6 | "io" 7 | "log" 8 | "net" 9 | "time" 10 | ) 11 | 12 | const ( 13 | _MAX_RETRIES = 6 14 | _MIN_RTT = 8 15 | _MIN_RTO = 30 16 | _MIN_ATO = 2 17 | _MAX_ATO = 10 18 | _MIN_SWND = 10 19 | _MAX_SWND = 960 20 | ) 21 | 22 | const ( 23 | _VACK_SCHED = iota + 1 24 | _VACK_QUICK 25 | _VACK_MUST 26 | _VSWND_ACTIVE 27 | _VRETR_IMMED 28 | ) 29 | 30 | const ( 31 | _RETR_REST = -1 32 | _CLOSE = 0xff 33 | ) 34 | 35 | var debug int 36 | 37 | func nodeOf(pk *packet) *qNode { 38 | return &qNode{packet: pk} 39 | } 40 | 41 | func (c *Conn) internalRecvLoop() { 42 | defer func() { 43 | // avoid send to closed channel while some replaying 44 | // data packets were received in shutting down. 45 | _ = recover() 46 | }() 47 | var buf, body []byte 48 | for { 49 | select { 50 | case buf = <-c.evRecv: 51 | if buf != nil { 52 | body = buf[_TH_SIZE:] 53 | } else { // shutdown 54 | return 55 | } 56 | } 57 | pk := new(packet) 58 | // keep the original buffer, so we could recycle it in future 59 | pk.buffer = buf 60 | unmarshall(pk, body) 61 | if pk.flag&_F_SACK != 0 { 62 | c.processSAck(pk) 63 | continue 64 | } 65 | if pk.flag&_F_ACK != 0 { 66 | c.processAck(pk) 67 | } 68 | if pk.flag&_F_DATA != 0 { 69 | c.insertData(pk) 70 | } else if pk.flag&_F_FIN != 0 { 71 | if pk.flag&_F_RESET != 0 { 72 | go c.forceShutdownWithLock() 73 | } else { 74 | go c.closeR(pk) 75 | } 76 | } 77 | } 78 | } 79 | 80 | func (c *Conn) internalSendLoop() { 81 | var timer = newTimer(c.rtt) 82 | for { 83 | select { 84 | case v := <-c.evSWnd: 85 | switch v { 86 | case _VRETR_IMMED: 87 | c.outlock.Lock() 88 | c.retransmit2() 89 | c.outlock.Unlock() 90 | case _VSWND_ACTIVE: 91 | timer.TryActive(c.rtt) 92 | case _CLOSE: 93 | return 94 | } 95 | case <-timer.C: // timeout yet 96 | var notifySender bool 97 | c.outlock.Lock() 98 | rest, _ := c.retransmit() 99 | switch rest { 100 | case _RETR_REST, 0: // nothing to send 101 | if c.outQ.size() > 0 { 102 | timer.Reset(c.rtt) 103 | } else { 104 | timer.Stop() 105 | // avoid sender blocking 106 | notifySender = true 107 | } 108 | default: // recent rto point 109 | timer.Reset(minI64(rest, c.rtt)) 110 | } 111 | c.outlock.Unlock() 112 | if notifySender { 113 | select { 114 | case c.evSend <- 1: 115 | default: 116 | } 117 | } 118 | } 119 | } 120 | } 121 | 122 | func (c *Conn) internalAckLoop() { 123 | var ackTimer = newTimer(c.ato) 124 | var lastAckState byte 125 | for { 126 | var v byte 127 | select { 128 | case <-ackTimer.C: 129 | // may cause sending duplicated ack if ato>rtt 130 | v = _VACK_QUICK 131 | case v = <-c.evAck: 132 | ackTimer.TryActive(c.ato) 133 | state := lastAckState 134 | lastAckState = v 135 | if state != v { 136 | if v == _CLOSE { 137 | return 138 | } 139 | v = _VACK_MUST 140 | } 141 | } 142 | c.inlock.Lock() 143 | if pkAck := c.makeAck(v); pkAck != nil { 144 | c.internalWrite(nodeOf(pkAck)) 145 | } 146 | c.inlock.Unlock() 147 | } 148 | } 149 | 150 | func (c *Conn) retransmit() (rest int64, count int32) { 151 | var now, rto = Now(), c.rto 152 | var limit = c.cwnd 153 | for item := c.outQ.head; item != nil && limit > 0; item = item.next { 154 | if item.scnt != _SENT_OK { // ACKed has scnt==-1 155 | diff := now - item.sent 156 | if diff > rto { // already rto 157 | c.internalWrite(item) 158 | count++ 159 | } else { 160 | // continue search next min rto duration 161 | if rest > 0 { 162 | rest = minI64(rest, rto-diff+1) 163 | } else { 164 | rest = rto - diff + 1 165 | } 166 | limit-- 167 | } 168 | } 169 | } 170 | c.outDupCnt += int(count) 171 | if count > 0 { 172 | shrcond := (c.fastRetransmit && count > maxI32(c.cwnd>>5, 4)) || (!c.fastRetransmit && count > c.cwnd>>3) 173 | if shrcond && now-c.lastShrink > c.rto { 174 | log.Printf("shrink cwnd from=%d to=%d s/4=%d", c.cwnd, c.cwnd>>1, c.swnd>>2) 175 | c.lastShrink = now 176 | // shrink cwnd and ensure cwnd >= swnd/4 177 | if c.cwnd > c.swnd>>1 { 178 | c.cwnd >>= 1 179 | } 180 | } 181 | } 182 | if c.outQ.size() > 0 { 183 | return 184 | } 185 | return _RETR_REST, 0 186 | } 187 | 188 | func (c *Conn) retransmit2() (count int32) { 189 | var limit, now = minI32(c.outPending>>4, 8), Now() 190 | var fRtt = c.rtt 191 | if now-c.lastShrink > c.rto { 192 | fRtt += maxI64(c.rtt>>4, 1) 193 | } else { 194 | fRtt += maxI64(c.rtt>>1, 2) 195 | } 196 | for item := c.outQ.head; item != nil && count < limit; item = item.next { 197 | if item.scnt != _SENT_OK { // ACKed has scnt==-1 198 | if item.miss >= 3 && now-item.sent >= fRtt { 199 | item.miss = 0 200 | c.internalWrite(item) 201 | count++ 202 | } 203 | } 204 | } 205 | c.fRCnt += int(count) 206 | c.outDupCnt += int(count) 207 | return 208 | } 209 | 210 | func (c *Conn) inputAndSend(pk *packet) error { 211 | item := &qNode{packet: pk} 212 | if c.mySeq&3 == 1 { 213 | c.tSlotT0 = NowNS() 214 | } 215 | c.outlock.Lock() 216 | // inflight packets exceeds cwnd 217 | // inflight includes: 1, unacked; 2, missed 218 | for c.outPending >= c.cwnd+c.missed { 219 | c.outlock.Unlock() 220 | if c.wtmo > 0 { 221 | var tmo int64 222 | tmo, c.wtmo = c.wtmo, 0 223 | select { 224 | case v := <-c.evSend: 225 | if v == _CLOSE { 226 | return io.EOF 227 | } 228 | case <-NewTimerChan(tmo): 229 | return ErrIOTimeout 230 | } 231 | } else { 232 | if v := <-c.evSend; v == _CLOSE { 233 | return io.EOF 234 | } 235 | } 236 | c.outlock.Lock() 237 | } 238 | c.outPending++ 239 | c.outPkCnt++ 240 | c.mySeq++ 241 | pk.seq = c.mySeq 242 | c.outQ.appendTail(item) 243 | c.internalWrite(item) 244 | c.outlock.Unlock() 245 | // active resending timer, must blocking 246 | c.evSWnd <- _VSWND_ACTIVE 247 | if c.mySeq&3 == 0 && c.flatTraffic { 248 | // calculate time error bewteen tslot with actual usage. 249 | // consider last sleep time error 250 | t1 := NowNS() 251 | terr := c.tSlot<<2 - c.lastSErr - (t1 - c.tSlotT0) 252 | // rest terr/2 if current time usage less than tslot of 100us. 253 | if terr > 1e5 { // 100us 254 | time.Sleep(time.Duration(terr >> 1)) 255 | c.lastSErr = maxI64(NowNS()-t1-terr, 0) 256 | } else { 257 | c.lastSErr >>= 1 258 | } 259 | } 260 | return nil 261 | } 262 | 263 | func (c *Conn) internalWrite(item *qNode) { 264 | if item.scnt >= 20 { 265 | // no exception of sending fin 266 | if item.flag&_F_FIN != 0 { 267 | c.fakeShutdown() 268 | c.dest = nil 269 | return 270 | } else { 271 | log.Println("Warn: too many retries", item) 272 | if c.urgent > 0 { // abort 273 | c.forceShutdown() 274 | return 275 | } else { // continue to retry 10 276 | c.urgent++ 277 | item.scnt = 10 278 | } 279 | } 280 | } 281 | // update current sent time and prev sent time 282 | item.sent, item.sent_1 = Now(), item.sent 283 | item.scnt++ 284 | buf := item.marshall(c.connID) 285 | if debug >= 3 { 286 | var pkType = packetTypeNames[item.flag] 287 | if item.flag&_F_SACK != 0 { 288 | log.Printf("send %s trp=%d on=%d %x", pkType, item.seq, item.ack, buf[_AH_SIZE+4:]) 289 | } else { 290 | log.Printf("send %s seq=%d ack=%d scnt=%d len=%d", pkType, item.seq, item.ack, item.scnt, len(buf)-_TH_SIZE) 291 | } 292 | } 293 | c.sock.WriteToUDP(buf, c.dest) 294 | } 295 | 296 | func (c *Conn) logAck(ack uint32) { 297 | c.lastAck = ack 298 | c.lastAckTime = Now() 299 | } 300 | 301 | func (c *Conn) makeLastAck() (pk *packet) { 302 | c.inlock.Lock() 303 | defer c.inlock.Unlock() 304 | if Now()-c.lastAckTime < c.rtt { 305 | return nil 306 | } 307 | pk = &packet{ 308 | ack: maxU32(c.lastAck, c.inQ.maxCtnSeq), 309 | flag: _F_ACK, 310 | } 311 | c.logAck(pk.ack) 312 | return 313 | } 314 | 315 | func (c *Conn) makeAck(level byte) (pk *packet) { 316 | now := Now() 317 | if level < _VACK_MUST && now-c.lastAckTime < c.ato { 318 | if level < _VACK_QUICK || now-c.lastAckTime < minI64(c.ato>>2, 1) { 319 | return 320 | } 321 | } 322 | // ready Q <-| 323 | // |-> outQ start (or more right) 324 | // |-> bitmap start 325 | // [predecessor] [predecessor+1] [predecessor+2] ..... 326 | var fakeSAck bool 327 | var predecessor = c.inQ.maxCtnSeq 328 | bmap, tbl := c.inQ.makeHolesBitmap(predecessor) 329 | if len(bmap) <= 0 { // fake sack 330 | bmap = make([]uint64, 1) 331 | bmap[0], tbl = 1, 1 332 | fakeSAck = true 333 | } 334 | // head 4-byte: TBL:1 | SCNT:1 | DELAY:2 335 | buf := make([]byte, len(bmap)*8+4) 336 | pk = &packet{ 337 | ack: predecessor + 1, 338 | flag: _F_SACK, 339 | payload: buf, 340 | } 341 | if fakeSAck { 342 | pk.ack-- 343 | } 344 | buf[0] = byte(tbl) 345 | // mark delayed time according to the time reference point 346 | if trp := c.inQ.lastIns; trp != nil { 347 | delayed := now - trp.sent 348 | if delayed < c.rtt { 349 | pk.seq = trp.seq 350 | pk.flag |= _F_TIME 351 | buf[1] = trp.scnt 352 | if delayed <= 0 { 353 | delayed = 1 354 | } 355 | binary.BigEndian.PutUint16(buf[2:], uint16(delayed)) 356 | } 357 | } 358 | buf1 := buf[4:] 359 | for i, b := range bmap { 360 | binary.BigEndian.PutUint64(buf1[i*8:], b) 361 | } 362 | c.logAck(predecessor) 363 | return 364 | } 365 | 366 | func unmarshallSAck(data []byte) (bmap []uint64, tbl uint32, delayed uint16, scnt uint8) { 367 | if len(data) > 0 { 368 | bmap = make([]uint64, len(data)>>3) 369 | } else { 370 | return 371 | } 372 | tbl = uint32(data[0]) 373 | scnt = data[1] 374 | delayed = binary.BigEndian.Uint16(data[2:]) 375 | data = data[4:] 376 | for i := 0; i < len(bmap); i++ { 377 | bmap[i] = binary.BigEndian.Uint64(data[i*8:]) 378 | } 379 | return 380 | } 381 | 382 | func calSwnd(bandwidth, rtt int64) int32 { 383 | w := int32(bandwidth * rtt / (8000 * _MSS)) 384 | if w <= _MAX_SWND { 385 | if w >= _MIN_SWND { 386 | return w 387 | } else { 388 | return _MIN_SWND 389 | } 390 | } else { 391 | return _MAX_SWND 392 | } 393 | } 394 | 395 | func (c *Conn) measure(seq uint32, delayed int64, scnt uint8) { 396 | target := c.outQ.get(seq) 397 | if target != nil { 398 | var lastSent int64 399 | switch target.scnt - scnt { 400 | case 0: 401 | // not sent again since this ack was sent out 402 | lastSent = target.sent 403 | case 1: 404 | // sent again once since this ack was sent out 405 | // then use prev sent time 406 | lastSent = target.sent_1 407 | default: 408 | // can't measure here because the packet was sent too many times 409 | return 410 | } 411 | // real-time rtt 412 | rtt := Now() - lastSent - delayed 413 | // reject these abnormal measures: 414 | // 1. rtt too small -> rtt/8 415 | // 2. backlogging too long 416 | if rtt < maxI64(c.rtt>>3, 1) || delayed > c.rtt>>1 { 417 | return 418 | } 419 | // srtt: update 1/8 420 | err := rtt - (c.srtt >> 3) 421 | c.srtt += err 422 | c.rtt = c.srtt >> 3 423 | if c.rtt < _MIN_RTT { 424 | c.rtt = _MIN_RTT 425 | } 426 | // s-swnd: update 1/4 427 | swnd := c.swnd<<3 - c.swnd + calSwnd(c.bandwidth, c.rtt) 428 | c.swnd = swnd >> 3 429 | c.tSlot = c.rtt * 1e6 / int64(c.swnd) 430 | c.ato = c.rtt >> 4 431 | if c.ato < _MIN_ATO { 432 | c.ato = _MIN_ATO 433 | } else if c.ato > _MAX_ATO { 434 | c.ato = _MAX_ATO 435 | } 436 | if err < 0 { 437 | err = -err 438 | err -= c.mdev >> 2 439 | if err > 0 { 440 | err >>= 3 441 | } 442 | } else { 443 | err -= c.mdev >> 2 444 | } 445 | // mdev: update 1/4 446 | c.mdev += err 447 | rto := c.rtt + maxI64(c.rtt<<1, c.mdev) 448 | if rto >= c.rto { 449 | c.rto = rto 450 | } else { 451 | c.rto = (c.rto + rto) >> 1 452 | } 453 | if c.rto < _MIN_RTO { 454 | c.rto = _MIN_RTO 455 | } 456 | if debug >= 1 { 457 | log.Printf("--- rtt=%d srtt=%d rto=%d swnd=%d", c.rtt, c.srtt, c.rto, c.swnd) 458 | } 459 | } 460 | } 461 | 462 | func (c *Conn) processSAck(pk *packet) { 463 | c.outlock.Lock() 464 | bmap, tbl, delayed, scnt := unmarshallSAck(pk.payload) 465 | if bmap == nil { // bad packet 466 | c.outlock.Unlock() 467 | return 468 | } 469 | if pk.flag&_F_TIME != 0 { 470 | c.measure(pk.seq, int64(delayed), scnt) 471 | } 472 | deleted, missed, continuous := c.outQ.deleteByBitmap(bmap, pk.ack, tbl) 473 | if deleted > 0 { 474 | c.ackHit(deleted, missed) 475 | // lock is released 476 | } else { 477 | c.outlock.Unlock() 478 | } 479 | if c.fastRetransmit && !continuous { 480 | // peer Q is uncontinuous, then trigger FR 481 | if deleted == 0 { 482 | c.evSWnd <- _VRETR_IMMED 483 | } else { 484 | select { 485 | case c.evSWnd <- _VRETR_IMMED: 486 | default: 487 | } 488 | } 489 | } 490 | if debug >= 2 { 491 | log.Printf("SACK qhead=%d deleted=%d outPending=%d on=%d %016x", 492 | c.outQ.distanceOfHead(0), deleted, c.outPending, pk.ack, bmap) 493 | } 494 | } 495 | 496 | func (c *Conn) processAck(pk *packet) { 497 | c.outlock.Lock() 498 | if end := c.outQ.get(pk.ack); end != nil { // ack hit 499 | _, deleted := c.outQ.deleteBefore(end) 500 | c.ackHit(deleted, 0) // lock is released 501 | if debug >= 2 { 502 | log.Printf("ACK hit on=%d", pk.ack) 503 | } 504 | // special case: ack the FIN 505 | if pk.seq == _FIN_ACK_SEQ { 506 | select { 507 | case c.evClose <- _S_FIN0: 508 | default: 509 | } 510 | } 511 | } else { // duplicated ack 512 | if debug >= 2 { 513 | log.Printf("ACK miss on=%d", pk.ack) 514 | } 515 | if pk.flag&_F_SYN != 0 { // No.3 Ack lost 516 | if pkAck := c.makeLastAck(); pkAck != nil { 517 | c.internalWrite(nodeOf(pkAck)) 518 | } 519 | } 520 | c.outlock.Unlock() 521 | } 522 | } 523 | 524 | func (c *Conn) ackHit(deleted, missed int32) { 525 | // must in outlock 526 | c.outPending -= deleted 527 | now := Now() 528 | if c.cwnd < c.swnd && now-c.lastShrink > c.rto { 529 | if c.cwnd < c.swnd>>1 { 530 | c.cwnd <<= 1 531 | } else { 532 | c.cwnd += deleted << 1 533 | } 534 | } 535 | if c.cwnd > c.swnd { 536 | c.cwnd = c.swnd 537 | } 538 | if now-c.lastRstMis > c.ato { 539 | c.lastRstMis = now 540 | c.missed = missed 541 | } else { 542 | c.missed = c.missed>>1 + missed 543 | } 544 | if qswnd := c.swnd >> 4; c.missed > qswnd { 545 | c.missed = qswnd 546 | } 547 | c.outlock.Unlock() 548 | select { 549 | case c.evSend <- 1: 550 | default: 551 | } 552 | } 553 | 554 | func (c *Conn) insertData(pk *packet) { 555 | c.inlock.Lock() 556 | defer c.inlock.Unlock() 557 | exists := c.inQ.contains(pk.seq) 558 | // duplicated with already queued or history 559 | // means: last ACK were lost 560 | if exists || pk.seq <= c.inQ.maxCtnSeq { 561 | // then send ACK for dups 562 | select { 563 | case c.evAck <- _VACK_MUST: 564 | default: 565 | } 566 | if debug >= 2 { 567 | dumpQ(fmt.Sprint("duplicated ", pk.seq), c.inQ) 568 | } 569 | c.inDupCnt++ 570 | return 571 | } 572 | // record current time in sent and regard as received time 573 | item := &qNode{packet: pk, sent: Now()} 574 | dis := c.inQ.searchInsert(item, c.lastReadSeq) 575 | if debug >= 3 { 576 | log.Printf("\t\t\trecv DATA seq=%d dis=%d maxCtn=%d lastReadSeq=%d", item.seq, dis, c.inQ.maxCtnSeq, c.lastReadSeq) 577 | } 578 | 579 | var ackState byte = _VACK_MUST 580 | var available bool 581 | switch dis { 582 | case 0: // impossible 583 | return 584 | case 1: 585 | if c.inQDirty { 586 | available = c.inQ.updateContinuous(item) 587 | if c.inQ.isWholeContinuous() { // whole Q is ordered 588 | c.inQDirty = false 589 | } else { //those holes still exists. 590 | ackState = _VACK_QUICK 591 | } 592 | } else { 593 | // here is an ideal situation 594 | c.inQ.maxCtnSeq = pk.seq 595 | available = true 596 | ackState = _VACK_SCHED 597 | } 598 | 599 | default: // there is an unordered packet, hole occurred here. 600 | if !c.inQDirty { 601 | c.inQDirty = true 602 | } 603 | } 604 | 605 | // write valid received count 606 | c.inPkCnt++ 607 | c.inQ.lastIns = item 608 | // try notify ack 609 | select { 610 | case c.evAck <- ackState: 611 | default: 612 | } 613 | if available { // try notify reader 614 | select { 615 | case c.evRead <- 1: 616 | default: 617 | } 618 | } 619 | } 620 | 621 | func (c *Conn) readInQ() bool { 622 | c.inlock.Lock() 623 | defer c.inlock.Unlock() 624 | // read already <-|-> expected Q 625 | // [lastReadSeq] | [lastReadSeq+1] [lastReadSeq+2] ...... 626 | if c.inQ.isEqualsHead(c.lastReadSeq+1) && c.lastReadSeq < c.inQ.maxCtnSeq { 627 | c.lastReadSeq = c.inQ.maxCtnSeq 628 | availabled := c.inQ.get(c.inQ.maxCtnSeq) 629 | availabled, _ = c.inQ.deleteBefore(availabled) 630 | for i := availabled; i != nil; i = i.next { 631 | c.inQReady = append(c.inQReady, i.payload...) 632 | // data was copied, then could recycle memory 633 | bpool.Put(i.buffer) 634 | i.payload = nil 635 | i.buffer = nil 636 | } 637 | return true 638 | } 639 | return false 640 | } 641 | 642 | // should not call this function concurrently. 643 | func (c *Conn) Read(buf []byte) (nr int, err error) { 644 | for { 645 | if len(c.inQReady) > 0 { 646 | n := copy(buf, c.inQReady) 647 | c.inQReady = c.inQReady[n:] 648 | return n, nil 649 | } 650 | if !c.readInQ() { 651 | if c.rtmo > 0 { 652 | var tmo int64 653 | tmo, c.rtmo = c.rtmo, 0 654 | select { 655 | case _, y := <-c.evRead: 656 | if !y && len(c.inQReady) == 0 { 657 | return 0, io.EOF 658 | } 659 | case <-NewTimerChan(tmo): 660 | return 0, ErrIOTimeout 661 | } 662 | } else { 663 | // only when evRead is closed and inQReady is empty 664 | // then could reply eof 665 | if _, y := <-c.evRead; !y && len(c.inQReady) == 0 { 666 | return 0, io.EOF 667 | } 668 | } 669 | } 670 | } 671 | } 672 | 673 | // should not call this function concurrently. 674 | func (c *Conn) Write(data []byte) (nr int, err error) { 675 | for len(data) > 0 && err == nil { 676 | //buf := make([]byte, _MSS+_AH_SIZE) 677 | buf := bpool.Get(c.mss + _AH_SIZE) 678 | body := buf[_TH_SIZE+_CH_SIZE:] 679 | n := copy(body, data) 680 | nr += n 681 | data = data[n:] 682 | pk := &packet{flag: _F_DATA, payload: body[:n], buffer: buf[:_AH_SIZE+n]} 683 | err = c.inputAndSend(pk) 684 | } 685 | return 686 | } 687 | 688 | func (c *Conn) LocalAddr() net.Addr { 689 | return c.sock.LocalAddr() 690 | } 691 | 692 | func (c *Conn) RemoteAddr() net.Addr { 693 | return c.dest 694 | } 695 | 696 | func (c *Conn) SetDeadline(t time.Time) error { 697 | c.SetReadDeadline(t) 698 | c.SetWriteDeadline(t) 699 | return nil 700 | } 701 | 702 | func (c *Conn) SetReadDeadline(t time.Time) error { 703 | if d := t.UnixNano()/Millisecond - Now(); d > 0 { 704 | c.rtmo = d 705 | } 706 | return nil 707 | } 708 | 709 | func (c *Conn) SetWriteDeadline(t time.Time) error { 710 | if d := t.UnixNano()/Millisecond - Now(); d > 0 { 711 | c.wtmo = d 712 | } 713 | return nil 714 | } 715 | -------------------------------------------------------------------------------- /protocol/conn_test.go: -------------------------------------------------------------------------------- 1 | package suft 2 | 3 | import ( 4 | "math/rand" 5 | "sort" 6 | "testing" 7 | ) 8 | 9 | var conn *Conn 10 | 11 | func init() { 12 | conn = &Conn{ 13 | outQ: newLinkedMap(_QModeOut), 14 | inQ: newLinkedMap(_QModeIn), 15 | } 16 | } 17 | 18 | func assert(cond bool, t testing.TB, format string, args ...interface{}) { 19 | if !cond { 20 | t.Errorf(format, args...) 21 | panic("last error") 22 | } 23 | } 24 | 25 | func Test_ordered_insert(t *testing.T) { 26 | data := []byte{1} 27 | var pk *packet 28 | for i := int32(1); i < 33; i++ { 29 | pk = &packet{ 30 | seq: uint32(i), 31 | payload: data, 32 | } 33 | conn.insertData(pk) 34 | assert(conn.inQ.size() == i, t, "len inQ=%d", conn.inQ.size()) 35 | assert(conn.inQ.maxCtnSeq == pk.seq, t, "lastCtnIn") 36 | assert(!conn.inQDirty, t, "dirty") 37 | } 38 | } 39 | 40 | func Test_unordered_insert(t *testing.T) { 41 | conn.inQ.reset() 42 | 43 | data := []byte{1} 44 | var pk *packet 45 | var seqs = make([]int, 0xfff) 46 | // unordered insert, and assert size 47 | for i := 1; i < len(seqs); i++ { 48 | var seq uint32 49 | for conn.inQ.contains(seq) || seq == 0 { 50 | seq = uint32(rand.Int31n(0xFFffff)) 51 | } 52 | seqs[i] = int(seq) 53 | pk = &packet{ 54 | seq: seq, 55 | payload: data, 56 | } 57 | conn.insertData(pk) 58 | assert(conn.inQ.size() == int32(i), t, "i=%d inQ.len=%d", i, conn.inQ.size()) 59 | } 60 | // assert lastCtnSeq 61 | sort.Ints(seqs) 62 | var zero = 0 63 | var last *int 64 | for i := 0; i < len(seqs); i++ { 65 | if i == 0 && seqs[0] != 0 { 66 | last = &zero 67 | break 68 | } 69 | if last != nil && seqs[i]-*last > 1 { 70 | if i == 1 { 71 | last = &zero 72 | } 73 | break 74 | } 75 | last = &seqs[i] 76 | } 77 | if *last != int(conn.inQ.maxCtnSeq) { 78 | for i, j := range seqs { 79 | if i < 10 { 80 | t.Logf("seq %d", j) 81 | } 82 | } 83 | } 84 | assert(*last == int(conn.inQ.maxCtnSeq), t, "lastCtnSeq=%d but expected=%d", conn.inQ.maxCtnSeq, *last) 85 | t.Logf("lastCtnSeq=%d dirty=%v", conn.inQ.maxCtnSeq, conn.inQDirty) 86 | } 87 | -------------------------------------------------------------------------------- /protocol/debug.go: -------------------------------------------------------------------------------- 1 | package suft 2 | 3 | import ( 4 | "encoding/hex" 5 | "fmt" 6 | "log" 7 | "os" 8 | "os/signal" 9 | "runtime" 10 | "runtime/pprof" 11 | "sync/atomic" 12 | "syscall" 13 | ) 14 | 15 | var enable_pprof bool 16 | var enable_stacktrace bool 17 | var debug_inited int32 18 | 19 | func set_debug_params(p *Params) { 20 | if atomic.CompareAndSwapInt32(&debug_inited, 0, 1) { 21 | debug = p.Debug 22 | enable_pprof = p.EnablePprof 23 | enable_stacktrace = p.Stacktrace 24 | if enable_pprof { 25 | f, err := os.Create("suft.pprof") 26 | if err != nil { 27 | log.Fatalln(err) 28 | } 29 | pprof.StartCPUProfile(f) 30 | } 31 | } 32 | } 33 | 34 | func (c *Conn) PrintState() { 35 | log.Printf("inQ=%d inQReady=%d outQ=%d", c.inQ.size(), len(c.inQReady), c.outQ.size()) 36 | log.Printf("inMaxCtnSeq=%d lastAck=%d lastReadSeq=%d", c.inQ.maxCtnSeq, c.lastAck, c.lastReadSeq) 37 | if c.inPkCnt > 0 { 38 | log.Printf("Rx pcnt=%d dups=%d %%d=%f%%", c.inPkCnt, c.inDupCnt, 100*float32(c.inDupCnt)/float32(c.inPkCnt)) 39 | } 40 | if c.outPkCnt > 0 { 41 | log.Printf("Tx pcnt=%d dups=%d %%d=%f%%", c.outPkCnt, c.outDupCnt, 100*float32(c.outDupCnt)/float32(c.outPkCnt)) 42 | } 43 | log.Printf("current-rtt=%d FastRetransmit=%d", c.rtt, c.fRCnt) 44 | if enable_stacktrace { 45 | var buf = make([]byte, 6400) 46 | for i := 0; i < 3; i++ { 47 | n := runtime.Stack(buf, true) 48 | if n >= len(buf) { 49 | buf = make([]byte, len(buf)<<1) 50 | } else { 51 | buf = buf[:n] 52 | break 53 | } 54 | } 55 | fmt.Println(string(buf)) 56 | } 57 | if enable_pprof { 58 | pprof.StopCPUProfile() 59 | } 60 | } 61 | 62 | func (c *Conn) internal_state() { 63 | ev := make(chan os.Signal, 10) 64 | signal.Notify(ev, syscall.Signal(12), syscall.SIGINT) 65 | for v := range ev { 66 | c.PrintState() 67 | if v == syscall.SIGINT { 68 | os.Exit(1) 69 | } 70 | } 71 | } 72 | 73 | func printBits(b uint64, j, s, d uint32) { 74 | fmt.Printf("bits=%064b j=%d seq=%d dis=%d\n", b, j, s, d) 75 | } 76 | 77 | func dumpb(label string, buf []byte) { 78 | log.Println(label, "\n", hex.Dump(buf)) 79 | } 80 | 81 | func dumpQ(s string, q *linkedMap) { 82 | var seqs = make([]uint32, 0, 20) 83 | n := q.head 84 | for i, m := int32(0), q.size(); i < m && n != nil; i++ { 85 | seqs = append(seqs, n.seq) 86 | n = n.next 87 | if len(seqs) == 20 { 88 | log.Printf("%s: Q=%d", s, seqs) 89 | seqs = seqs[:0] 90 | } 91 | } 92 | if len(seqs) > 0 { 93 | log.Printf("%s: Q=%d", s, seqs) 94 | } 95 | } 96 | -------------------------------------------------------------------------------- /protocol/endpoint.go: -------------------------------------------------------------------------------- 1 | package suft 2 | 3 | import ( 4 | "encoding/binary" 5 | "fmt" 6 | "io" 7 | "log" 8 | "math/rand" 9 | "net" 10 | "sort" 11 | "sync" 12 | "sync/atomic" 13 | "time" 14 | 15 | "github.com/cloudflare/golibs/bytepool" 16 | ) 17 | 18 | const ( 19 | _SO_BUF_SIZE = 8 << 20 20 | ) 21 | 22 | var ( 23 | bpool bytepool.BytePool 24 | ) 25 | 26 | type Params struct { 27 | LocalAddr string 28 | Bandwidth int64 29 | Mtu int 30 | IsServ bool 31 | FastRetransmit bool 32 | FlatTraffic bool 33 | EnablePprof bool 34 | Stacktrace bool 35 | Debug int 36 | } 37 | 38 | type connID struct { 39 | lid uint32 40 | rid uint32 41 | } 42 | 43 | type Endpoint struct { 44 | udpconn *net.UDPConn 45 | state int32 46 | idSeq uint32 47 | isServ bool 48 | listenChan chan *Conn 49 | lRegistry map[uint32]*Conn 50 | rRegistry map[string][]uint32 51 | mlock sync.RWMutex 52 | timeout *iTimer 53 | params Params 54 | } 55 | 56 | func (c *connID) setRid(b []byte) { 57 | c.rid = binary.BigEndian.Uint32(b[_MAGIC_SIZE+6:]) 58 | } 59 | 60 | func init() { 61 | bpool.Init(0, 2000) 62 | rand.Seed(NowNS()) 63 | } 64 | 65 | func NewEndpoint(p *Params) (*Endpoint, error) { 66 | set_debug_params(p) 67 | if p.Bandwidth <= 0 || p.Bandwidth > 100 { 68 | return nil, fmt.Errorf("bw->(0,100]") 69 | } 70 | conn, err := net.ListenPacket("udp", p.LocalAddr) 71 | if err != nil { 72 | return nil, err 73 | } 74 | e := &Endpoint{ 75 | udpconn: conn.(*net.UDPConn), 76 | idSeq: 1, 77 | isServ: p.IsServ, 78 | listenChan: make(chan *Conn, 1), 79 | lRegistry: make(map[uint32]*Conn), 80 | rRegistry: make(map[string][]uint32), 81 | timeout: newTimer(0), 82 | params: *p, 83 | } 84 | if e.isServ { 85 | e.state = _S_EST0 86 | } else { // client 87 | e.state = _S_EST1 88 | e.idSeq = uint32(rand.Int31()) 89 | } 90 | e.params.Bandwidth = p.Bandwidth << 20 // mbps to bps 91 | e.udpconn.SetReadBuffer(_SO_BUF_SIZE) 92 | go e.internal_listen() 93 | return e, nil 94 | } 95 | 96 | func (e *Endpoint) internal_listen() { 97 | const rtmo = 30 * 1e9 98 | var id connID 99 | var pdCtx = getPollCtx(e.udpconn) 100 | for { 101 | //var buf = make([]byte, 1600) 102 | var buf = bpool.Get(1600) 103 | net_pollSetDeadline(pdCtx, rtmo+runtimeNano(), 'r') 104 | n, addr, err := e.udpconn.ReadFromUDP(buf) 105 | if err == nil && n >= _AH_SIZE { 106 | buf = buf[:n] 107 | e.getConnID(&id, buf) 108 | 109 | switch id.lid { 110 | case 0: // new connection 111 | if e.isServ { 112 | go e.acceptNewConn(id, addr, buf) 113 | } else { 114 | dumpb("drop", buf) 115 | } 116 | 117 | case _INVALID_SEQ: 118 | dumpb("drop invalid", buf) 119 | 120 | default: // old connection 121 | e.mlock.RLock() 122 | conn := e.lRegistry[id.lid] 123 | e.mlock.RUnlock() 124 | if conn != nil { 125 | e.dispatch(conn, buf) 126 | } else { 127 | e.resetPeer(addr, id) 128 | dumpb("drop null", buf) 129 | } 130 | } 131 | 132 | } else if err != nil { 133 | // idle process 134 | if nerr, y := err.(net.Error); y && nerr.Timeout() { 135 | e.idleProcess() 136 | continue 137 | } 138 | // other errors 139 | if atomic.LoadInt32(&e.state) == _S_FIN { 140 | return 141 | } else { 142 | log.Println("Error: read sock", err) 143 | } 144 | } 145 | } 146 | } 147 | 148 | func (e *Endpoint) idleProcess() { 149 | // recycle/shrink memory 150 | bpool.Drain() 151 | e.mlock.Lock() 152 | defer e.mlock.Unlock() 153 | // reset urgent 154 | for _, c := range e.lRegistry { 155 | c.outlock.Lock() 156 | if c.outQ.size() == 0 && c.urgent != 0 { 157 | c.urgent = 0 158 | } 159 | c.outlock.Unlock() 160 | } 161 | } 162 | 163 | func (e *Endpoint) Dial(addr string) (*Conn, error) { 164 | rAddr, err := net.ResolveUDPAddr("udp", addr) 165 | if err != nil { 166 | return nil, err 167 | } 168 | e.mlock.Lock() 169 | e.idSeq++ 170 | id := connID{e.idSeq, 0} 171 | conn := NewConn(e, rAddr, id) 172 | e.lRegistry[id.lid] = conn 173 | e.mlock.Unlock() 174 | if atomic.LoadInt32(&e.state) != _S_FIN { 175 | err = conn.initConnection(nil) 176 | return conn, err 177 | } 178 | return nil, io.EOF 179 | } 180 | 181 | func (e *Endpoint) acceptNewConn(id connID, addr *net.UDPAddr, buf []byte) { 182 | rKey := addr.String() 183 | e.mlock.Lock() 184 | // map: remoteAddr => remoteConnID 185 | // filter duplicated syn packets 186 | if newArr := insertRid(e.rRegistry[rKey], id.rid); newArr != nil { 187 | e.rRegistry[rKey] = newArr 188 | } else { 189 | e.mlock.Unlock() 190 | log.Println("Warn: duplicated connection", addr) 191 | return 192 | } 193 | e.idSeq++ 194 | id.lid = e.idSeq 195 | conn := NewConn(e, addr, id) 196 | e.lRegistry[id.lid] = conn 197 | e.mlock.Unlock() 198 | err := conn.initConnection(buf) 199 | if err == nil { 200 | select { 201 | case e.listenChan <- conn: 202 | case <-time.After(_10ms): 203 | log.Println("Warn: no listener") 204 | } 205 | } else { 206 | e.removeConn(id, addr) 207 | log.Println("Error: init_connection", addr, err) 208 | } 209 | } 210 | 211 | func (e *Endpoint) removeConn(id connID, addr *net.UDPAddr) { 212 | e.mlock.Lock() 213 | delete(e.lRegistry, id.lid) 214 | rKey := addr.String() 215 | if newArr := deleteRid(e.rRegistry[rKey], id.rid); newArr != nil { 216 | if len(newArr) > 0 { 217 | e.rRegistry[rKey] = newArr 218 | } else { 219 | delete(e.rRegistry, rKey) 220 | } 221 | } 222 | e.mlock.Unlock() 223 | } 224 | 225 | // net.Listener 226 | func (e *Endpoint) Close() error { 227 | state := atomic.LoadInt32(&e.state) 228 | if state > 0 && atomic.CompareAndSwapInt32(&e.state, state, _S_FIN) { 229 | err := e.udpconn.Close() 230 | e.lRegistry = nil 231 | e.rRegistry = nil 232 | select { // release listeners 233 | case e.listenChan <- nil: 234 | default: 235 | } 236 | return err 237 | } 238 | return nil 239 | } 240 | 241 | // net.Listener 242 | func (e *Endpoint) Addr() net.Addr { 243 | return e.udpconn.LocalAddr() 244 | } 245 | 246 | // net.Listener 247 | func (e *Endpoint) Accept() (net.Conn, error) { 248 | if atomic.LoadInt32(&e.state) == _S_EST0 { 249 | return <-e.listenChan, nil 250 | } else { 251 | return nil, io.EOF 252 | } 253 | } 254 | 255 | func (e *Endpoint) Listen() *Conn { 256 | if atomic.LoadInt32(&e.state) == _S_EST0 { 257 | return <-e.listenChan 258 | } else { 259 | return nil 260 | } 261 | } 262 | 263 | // tmo in MS 264 | func (e *Endpoint) ListenTimeout(tmo int64) *Conn { 265 | if tmo <= 0 { 266 | return e.Listen() 267 | } 268 | if atomic.LoadInt32(&e.state) == _S_EST0 { 269 | select { 270 | case c := <-e.listenChan: 271 | return c 272 | case <-NewTimerChan(tmo): 273 | } 274 | } 275 | return nil 276 | } 277 | 278 | func (e *Endpoint) getConnID(idPtr *connID, buf []byte) { 279 | // TODO determine magic header 280 | magicAndLen := binary.BigEndian.Uint64(buf) 281 | if int(magicAndLen&0xFFff) == len(buf) { 282 | id := binary.BigEndian.Uint64(buf[_MAGIC_SIZE+2:]) 283 | idPtr.lid = uint32(id >> 32) 284 | idPtr.rid = uint32(id) 285 | } else { 286 | idPtr.lid = _INVALID_SEQ 287 | } 288 | } 289 | 290 | func (e *Endpoint) dispatch(c *Conn, buf []byte) { 291 | e.timeout.Reset(30) 292 | select { 293 | case c.evRecv <- buf: 294 | case <-e.timeout.C: 295 | log.Println("Warn: dispatch packet failed") 296 | } 297 | } 298 | 299 | func (e *Endpoint) resetPeer(addr *net.UDPAddr, id connID) { 300 | pk := &packet{flag: _F_FIN | _F_RESET} 301 | buf := nodeOf(pk).marshall(id) 302 | e.udpconn.WriteToUDP(buf, addr) 303 | } 304 | 305 | type u32Slice []uint32 306 | 307 | func (p u32Slice) Len() int { return len(p) } 308 | func (p u32Slice) Less(i, j int) bool { return p[i] < p[j] } 309 | func (p u32Slice) Swap(i, j int) { p[i], p[j] = p[j], p[i] } 310 | 311 | // if the rid is not existed in array then insert it return new array 312 | func insertRid(array []uint32, rid uint32) []uint32 { 313 | if len(array) > 0 { 314 | pos := sort.Search(len(array), func(n int) bool { 315 | return array[n] >= rid 316 | }) 317 | if pos < len(array) && array[pos] == rid { 318 | return nil 319 | } 320 | } 321 | array = append(array, rid) 322 | sort.Sort(u32Slice(array)) 323 | return array 324 | } 325 | 326 | // if rid was existed in array then delete it return new array 327 | func deleteRid(array []uint32, rid uint32) []uint32 { 328 | if len(array) > 0 { 329 | pos := sort.Search(len(array), func(n int) bool { 330 | return array[n] >= rid 331 | }) 332 | if pos < len(array) && array[pos] == rid { 333 | newArray := make([]uint32, len(array)-1) 334 | n := copy(newArray, array[:pos]) 335 | copy(newArray[n:], array[pos+1:]) 336 | return newArray 337 | } 338 | } 339 | return nil 340 | } 341 | -------------------------------------------------------------------------------- /protocol/endpoint_test.go: -------------------------------------------------------------------------------- 1 | package suft 2 | 3 | import ( 4 | "fmt" 5 | "math/rand" 6 | "sort" 7 | "testing" 8 | ) 9 | 10 | func Test_insert_delete_rid(t *testing.T) { 11 | var a []uint32 12 | var b = make([]uint32, 0, 1e3) 13 | var uniq = make(map[uint32]int) 14 | // insert into a with random 15 | for i := 0; i < cap(b); i++ { 16 | n := uint32(rand.Int31()) 17 | if _, y := uniq[n]; !y { 18 | b = append(b, n) 19 | uniq[n] = 1 20 | } 21 | dups := 1 22 | if i&0xf == 0xf { 23 | dups = 3 24 | } 25 | for j := 0; j < dups; j++ { 26 | if aa := insertRid(a, n); aa != nil { 27 | a = aa 28 | } 29 | } 30 | } 31 | sort.Sort(u32Slice(b)) 32 | bStr := fmt.Sprintf("%d", b) 33 | aStr := fmt.Sprintf("%d", a) 34 | assert(aStr == bStr, t, "a!=b") 35 | 36 | for i := 0; i < len(b); i++ { 37 | if aa := deleteRid(a, b[i]); aa != nil { 38 | a = aa 39 | } 40 | } 41 | assert(len(a) == 0, t, "a!=0") 42 | } 43 | -------------------------------------------------------------------------------- /protocol/linked.go: -------------------------------------------------------------------------------- 1 | package suft 2 | 3 | type qNode struct { 4 | *packet 5 | prev *qNode 6 | next *qNode 7 | sent int64 // last sent time 8 | sent_1 int64 // prev sent time 9 | miss int // sack miss count 10 | } 11 | 12 | type linkedMap struct { 13 | head *qNode 14 | tail *qNode 15 | qmap map[uint32]*qNode 16 | lastIns *qNode 17 | maxCtnSeq uint32 18 | mode int 19 | } 20 | 21 | const ( 22 | _QModeIn = 1 23 | _QModeOut = 2 24 | ) 25 | 26 | func newLinkedMap(qmode int) *linkedMap { 27 | return &linkedMap{ 28 | qmap: make(map[uint32]*qNode), 29 | mode: qmode, 30 | } 31 | } 32 | 33 | func (l *linkedMap) get(seq uint32) (i *qNode) { 34 | i = l.qmap[seq] 35 | return 36 | } 37 | 38 | func (l *linkedMap) contains(seq uint32) bool { 39 | _, y := l.qmap[seq] 40 | return y 41 | } 42 | 43 | func (l *linkedMap) size() int32 { 44 | return int32(len(l.qmap)) 45 | } 46 | 47 | func (l *linkedMap) reset() { 48 | l.head = nil 49 | l.tail = nil 50 | l.lastIns = nil 51 | l.maxCtnSeq = 0 52 | l.qmap = make(map[uint32]*qNode) 53 | } 54 | 55 | func (l *linkedMap) isEqualsHead(seq uint32) bool { 56 | return l.head != nil && seq == l.head.seq 57 | } 58 | 59 | func (l *linkedMap) distanceOfHead(seq uint32) int32 { 60 | if l.head != nil { 61 | return int32(seq - l.head.seq) 62 | } else { 63 | return -1 64 | } 65 | } 66 | 67 | func (l *linkedMap) appendTail(one *qNode) { 68 | if l.tail != nil { 69 | l.tail.next = one 70 | one.prev = l.tail 71 | l.tail = one 72 | } else { 73 | l.head = one 74 | l.tail = one 75 | } 76 | l.qmap[one.seq] = one 77 | } 78 | 79 | // xxx - n - yyy 80 | // xxx - yyy 81 | func (l *linkedMap) deleteAt(n *qNode) { 82 | x, y := n.prev, n.next 83 | if x != nil { 84 | x.next = y 85 | } else { 86 | l.head = y 87 | } 88 | if y != nil { 89 | y.prev = x 90 | } else { 91 | l.tail = x 92 | } 93 | n.prev, n.next = nil, nil 94 | delete(l.qmap, n.seq) 95 | } 96 | 97 | // delete with n <- ...n | 98 | func (l *linkedMap) deleteBefore(n *qNode) (left *qNode, deleted int32) { 99 | for i := n; i != nil; i = i.prev { 100 | delete(l.qmap, i.seq) 101 | if i.scnt != _SENT_OK { 102 | deleted++ 103 | // only outQ could delete at here 104 | if l.mode == _QModeOut { 105 | bpool.Put(i.buffer) 106 | i.buffer = nil 107 | } 108 | } 109 | } 110 | left = l.head 111 | l.head = n.next 112 | n.next = nil 113 | if l.head != nil { 114 | l.head.prev = nil 115 | } else { // n.next is the tail and is nil 116 | l.tail = nil 117 | } 118 | return 119 | } 120 | 121 | // xxx - ref 122 | // xxx - one - ref 123 | func (l *linkedMap) insertBefore(ref, one *qNode) { 124 | x := ref.prev 125 | if x != nil { 126 | x.next = one 127 | one.prev = x 128 | } else { 129 | l.head = one 130 | } 131 | ref.prev = one 132 | one.next = ref 133 | l.qmap[one.seq] = one 134 | } 135 | 136 | // ref - zzz 137 | // ref - one - zzz 138 | func (l *linkedMap) insertAfter(ref, one *qNode) { 139 | z := ref.next 140 | if z == nil { // append 141 | ref.next = one 142 | l.tail = one 143 | } else { // insert mid 144 | z.prev = one 145 | ref.next = one 146 | } 147 | one.prev = ref 148 | one.next = z 149 | l.qmap[one.seq] = one 150 | } 151 | 152 | // baseHead: the left outside boundary 153 | // if inserted, return the distance between newNode with baseHead 154 | func (l *linkedMap) searchInsert(one *qNode, baseHead uint32) (dis int64) { 155 | for i := l.tail; i != nil; i = i.prev { 156 | dis = int64(one.seq) - int64(i.seq) 157 | if dis > 0 { 158 | l.insertAfter(i, one) 159 | return 160 | } else if dis == 0 { 161 | // duplicated 162 | return 163 | } 164 | } 165 | if one.seq <= baseHead { 166 | return 0 167 | } 168 | if l.head != nil { 169 | l.insertBefore(l.head, one) 170 | } else { 171 | l.head = one 172 | l.tail = one 173 | l.qmap[one.seq] = one 174 | } 175 | dis = int64(one.seq) - int64(baseHead) 176 | return 177 | } 178 | 179 | func (l *linkedMap) updateContinuous(i *qNode) bool { 180 | var lastCtnSeq = l.maxCtnSeq 181 | for ; i != nil; i = i.next { 182 | if i.seq-lastCtnSeq == 1 { 183 | lastCtnSeq = i.seq 184 | } else { 185 | break 186 | } 187 | } 188 | if lastCtnSeq != l.maxCtnSeq { 189 | l.maxCtnSeq = lastCtnSeq 190 | return true 191 | } 192 | return false 193 | } 194 | 195 | func (l *linkedMap) isWholeContinuous() bool { 196 | return l.tail != nil && l.maxCtnSeq == l.tail.seq 197 | } 198 | 199 | /* 200 | func (l *linkedMap) searchMaxContinued(baseHead uint32) (*qNode, bool) { 201 | var last *qNode 202 | for i := l.head; i != nil; i = i.next { 203 | if last != nil { 204 | if i.seq-last.seq > 1 { 205 | return last, true 206 | } 207 | } else { 208 | if i.seq != baseHead { 209 | return nil, false 210 | } 211 | } 212 | last = i 213 | } 214 | if last != nil { 215 | return last, true 216 | } else { 217 | return nil, false 218 | } 219 | } 220 | */ 221 | 222 | func (q *qNode) forward(n int) *qNode { 223 | for ; n > 0 && q != nil; n-- { 224 | q = q.next 225 | } 226 | return q 227 | } 228 | 229 | // prev of bitmap start point 230 | func (l *linkedMap) makeHolesBitmap(prev uint32) ([]uint64, uint32) { 231 | var start *qNode 232 | var bits uint64 233 | var j uint32 234 | for i := l.head; i != nil; i = i.next { 235 | if i.seq >= prev+1 { 236 | start = i 237 | break 238 | } 239 | } 240 | var bitmap []uint64 241 | // search start which is the recent successor of [prev] 242 | Iterator: 243 | for i := start; i != nil; i = i.next { 244 | dis := i.seq - prev 245 | prev = i.seq 246 | j += dis // j is next bit index 247 | for j >= 65 { 248 | if len(bitmap) >= 20 { // bitmap too long 249 | break Iterator 250 | } 251 | bitmap = append(bitmap, bits) 252 | bits = 0 253 | j -= 64 254 | } 255 | bits |= 1 << (j - 1) 256 | } 257 | if j > 0 { 258 | // j -> (0, 64] 259 | bitmap = append(bitmap, bits) 260 | } 261 | return bitmap, j 262 | } 263 | 264 | // from= the bitmap start point 265 | func (l *linkedMap) deleteByBitmap(bmap []uint64, from uint32, tailBitsLen uint32) (deleted, missed int32, lastContinued bool) { 266 | var start = l.qmap[from] 267 | if start != nil { 268 | // delete predecessors 269 | if pred := start.prev; pred != nil { 270 | _, n := l.deleteBefore(pred) 271 | deleted += n 272 | } 273 | } else { 274 | // [from] is out of bounds 275 | return 276 | } 277 | var j, bitsLen uint32 278 | var bits uint64 279 | bits, bmap = bmap[0], bmap[1:] 280 | if len(bmap) > 0 { 281 | bitsLen = 64 282 | } else { 283 | // bmap.len==1, tail is here 284 | bitsLen = tailBitsLen 285 | } 286 | // maxContinued will save the max continued node (from [start]) which could be deleted safely. 287 | // keep the queue smallest 288 | var maxContinued *qNode 289 | lastContinued = true 290 | 291 | for i := start; i != nil; j++ { 292 | if j >= bitsLen { 293 | if len(bmap) > 0 { 294 | j = 0 295 | bits, bmap = bmap[0], bmap[1:] 296 | if len(bmap) > 0 { 297 | bitsLen = 64 298 | } else { 299 | bitsLen = tailBitsLen 300 | } 301 | } else { 302 | // no more pages 303 | goto finished 304 | } 305 | } 306 | if bits&1 == 1 { 307 | if lastContinued { 308 | maxContinued = i 309 | } 310 | if i.scnt != _SENT_OK { 311 | // no mark means first deleting 312 | deleted++ 313 | } 314 | // don't delete, just mark it 315 | i.scnt = _SENT_OK 316 | } else { 317 | // known it may be lost 318 | if i.miss == 0 { 319 | missed++ 320 | } 321 | i.miss++ 322 | lastContinued = false 323 | } 324 | bits >>= 1 325 | i = i.next 326 | } 327 | 328 | finished: 329 | if maxContinued != nil { 330 | l.deleteBefore(maxContinued) 331 | } 332 | return 333 | } 334 | -------------------------------------------------------------------------------- /protocol/linked_test.go: -------------------------------------------------------------------------------- 1 | package suft 2 | 3 | import ( 4 | "fmt" 5 | "math/rand" 6 | "testing" 7 | ) 8 | 9 | var lmap *linkedMap 10 | 11 | func init() { 12 | lmap = newLinkedMap(_QModeIn) 13 | } 14 | 15 | func node(seq int) *qNode { 16 | return &qNode{packet: &packet{seq: uint32(seq)}} 17 | } 18 | 19 | func Test_get(t *testing.T) { 20 | var i, n *qNode 21 | assert(!lmap.contains(0), t, "0 nil") 22 | n = node(0) 23 | lmap.head = n 24 | lmap.tail = n 25 | lmap.qmap[0] = n 26 | i = lmap.get(0) 27 | assert(i == n, t, "0=n") 28 | } 29 | 30 | func Test_insert(t *testing.T) { 31 | lmap.reset() 32 | n := node(1) 33 | // appendTail 34 | lmap.appendTail(n) 35 | assert(lmap.head == n, t, "head n") 36 | assert(lmap.tail == n, t, "head n") 37 | n = node(2) 38 | lmap.appendTail(n) 39 | assert(lmap.head != n, t, "head n") 40 | assert(lmap.tail == n, t, "head n") 41 | assert(lmap.size() == 2, t, "size") 42 | } 43 | 44 | func Test_insertAfter(t *testing.T) { 45 | n1 := lmap.get(1) 46 | n2 := n1.next 47 | n3 := node(3) 48 | lmap.insertAfter(n1, n3) 49 | assert(n1.next == n3, t, "n3") 50 | assert(n1 == n3.prev, t, "left n3") 51 | assert(n2 == n3.next, t, "n3 right") 52 | } 53 | 54 | func Test_insertBefore(t *testing.T) { 55 | n3 := lmap.get(3) 56 | n2 := n3.next 57 | n4 := node(4) 58 | lmap.insertAfter(n3, n4) 59 | assert(n3.next == n4, t, "n4") 60 | assert(n3 == n4.prev, t, "left n4") 61 | assert(n2 == n4.next, t, "n4 right") 62 | } 63 | 64 | func Test_deleteBefore(t *testing.T) { 65 | lmap.reset() 66 | for i := 1; i < 10; i++ { 67 | n := node(i) 68 | lmap.appendTail(n) 69 | } 70 | 71 | var assertRangeEquals = func(n *qNode, start, wantCount int) { 72 | var last *qNode 73 | var count int 74 | for ; n != nil; n = n.next { 75 | assert(int(n.seq) == start, t, "nseq=%d start=%d", n.seq, start) 76 | last = n 77 | start++ 78 | count++ 79 | } 80 | assert(last.next == nil, t, "tail nil") 81 | assert(count == wantCount, t, "count") 82 | } 83 | assertRangeEquals(lmap.head, 1, 9) 84 | var n *qNode 85 | n = lmap.get(3) 86 | n, _ = lmap.deleteBefore(n) 87 | assertRangeEquals(n, 1, 3) 88 | assert(lmap.head.seq == 4, t, "head") 89 | 90 | n = lmap.get(8) 91 | n, _ = lmap.deleteBefore(n) 92 | assertRangeEquals(n, 4, 5) 93 | assert(lmap.head.seq == 9, t, "head") 94 | 95 | n = lmap.get(9) 96 | n, _ = lmap.deleteBefore(n) 97 | assertRangeEquals(n, 9, 1) 98 | 99 | assert(lmap.size() == 0, t, "size 0") 100 | assert(lmap.head == nil, t, "head nil") 101 | assert(lmap.tail == nil, t, "tail nil") 102 | } 103 | 104 | func testBitmap(t *testing.T, bmap []uint64, prev uint32) { 105 | var j uint 106 | var k int 107 | bits := bmap[k] 108 | t.Logf("test-%d %016x", k, bits) 109 | var checkNextPage = func() { 110 | if j >= 64 { 111 | j = 0 112 | k++ 113 | bits = bmap[k] 114 | t.Logf("test-%d %016x", k, bits) 115 | } 116 | } 117 | for i := lmap.head; i != nil && k < len(bmap); i = i.next { 118 | checkNextPage() 119 | dis := i.seq - prev 120 | prev = i.seq 121 | if dis == 1 { 122 | bit := (bits >> j) & 1 123 | assert(bit == 1, t, "1 bit=%d j=%d", bit, j) 124 | j++ 125 | } else { 126 | for ; dis > 0; dis-- { 127 | checkNextPage() 128 | bit := (bits >> j) & 1 129 | want := uint64(0) 130 | if dis == 1 { 131 | want = 1 132 | } 133 | assert(bit == want, t, "?=%d bit=%d j=%d", want, bit, j) 134 | j++ 135 | } 136 | } 137 | } 138 | // remains bits should be 0 139 | for i := j & 63; i > 0; i-- { 140 | bit := (bits >> j) & 1 141 | assert(bit == 0, t, "00 bit=%d j=%d", bit, j) 142 | j++ 143 | } 144 | } 145 | 146 | func Test_bitmap(t *testing.T) { 147 | var prev uint32 148 | var head uint32 = prev + 1 149 | 150 | lmap.reset() 151 | // test 66-%3 and record holes 152 | var holes = make([]uint32, 0, 50) 153 | for i := head; i < 366; i++ { 154 | if i%3 == 0 { 155 | holes = append(holes, i) 156 | continue 157 | } 158 | n := node(int(i)) 159 | lmap.appendTail(n) 160 | } 161 | bmap, tbl := lmap.makeHolesBitmap(prev) 162 | testBitmap(t, bmap, prev) 163 | 164 | lmap.reset() 165 | // full 66, do deleteByBitmap then compare 166 | for i := head; i < 366; i++ { 167 | n := node(int(i)) 168 | lmap.appendTail(n) 169 | } 170 | 171 | lmap.deleteByBitmap(bmap, head, tbl) 172 | var holesResult = make([]uint32, 0, 50) 173 | for i := lmap.head; i != nil; i = i.next { 174 | if i.scnt != _SENT_OK { 175 | holesResult = append(holesResult, i.seq) 176 | } 177 | } 178 | a := fmt.Sprintf("%x", holes) 179 | b := fmt.Sprintf("%x", holesResult) 180 | assert(a == b, t, "deleteByBitmap \na=%s \nb=%s", a, b) 181 | 182 | lmap.reset() 183 | // test stride across page 1 184 | for i := head; i < 69; i++ { 185 | if i >= 63 && i <= 65 { 186 | continue 187 | } 188 | n := node(int(i)) 189 | lmap.appendTail(n) 190 | } 191 | bmap, _ = lmap.makeHolesBitmap(prev) 192 | testBitmap(t, bmap, prev) 193 | 194 | lmap.reset() 195 | prev = 65 196 | head = prev + 1 197 | // test stride across page 0 198 | for i := head; i < 68; i++ { 199 | n := node(int(i)) 200 | lmap.appendTail(n) 201 | } 202 | bmap, _ = lmap.makeHolesBitmap(prev) 203 | testBitmap(t, bmap, prev) 204 | } 205 | 206 | var ackbitmap []uint64 207 | 208 | func init_benchmark_map() { 209 | if lmap.size() != 640 { 210 | lmap.reset() 211 | for i := 1; i <= 640; i++ { 212 | lmap.appendTail(node(i)) 213 | } 214 | ackbitmap = make([]uint64, 10) 215 | for i := 0; i < len(ackbitmap); i++ { 216 | n := rand.Int63() 217 | ackbitmap[i] = uint64(n) << 1 218 | } 219 | } 220 | } 221 | 222 | func Benchmark_make_bitmap(b *testing.B) { 223 | init_benchmark_map() 224 | 225 | for i := 0; i < b.N; i++ { 226 | lmap.makeHolesBitmap(0) 227 | } 228 | } 229 | 230 | func Benchmark_apply_bitmap(b *testing.B) { 231 | init_benchmark_map() 232 | 233 | for i := 0; i < b.N; i++ { 234 | lmap.deleteByBitmap(ackbitmap, 1, 64) 235 | } 236 | } 237 | -------------------------------------------------------------------------------- /protocol/netfd_unix.go: -------------------------------------------------------------------------------- 1 | // +build darwin dragonfly freebsd linux nacl netbsd openbsd solaris 2 | 3 | package suft 4 | 5 | import ( 6 | "net" 7 | ) 8 | 9 | // Network file descriptor. 10 | type netFD struct { 11 | // locking/lifetime of sysfd + serialize access to Read and Write methods 12 | fdmu fdMutex 13 | 14 | // immutable until Close 15 | sysfd int 16 | family int 17 | sotype int 18 | isConnected bool 19 | net string 20 | laddr net.Addr 21 | raddr net.Addr 22 | 23 | // wait server 24 | pd pollDesc 25 | } 26 | -------------------------------------------------------------------------------- /protocol/netfd_windows.go: -------------------------------------------------------------------------------- 1 | package suft 2 | 3 | import ( 4 | "net" 5 | "syscall" 6 | ) 7 | 8 | // Network file descriptor. 9 | type netFD struct { 10 | // locking/lifetime of sysfd + serialize access to Read and Write methods 11 | fdmu fdMutex 12 | 13 | // immutable until Close 14 | sysfd syscall.Handle 15 | family int 16 | sotype int 17 | isConnected bool 18 | skipSyncNotif bool 19 | net string 20 | laddr net.Addr 21 | raddr net.Addr 22 | 23 | rop operation // read operation 24 | wop operation // write operation 25 | 26 | // wait server 27 | pd pollDesc 28 | } 29 | 30 | // operation contains superset of data necessary to perform all async IO. 31 | type operation struct { 32 | // Used by IOCP interface, it must be first field 33 | // of the struct, as our code rely on it. 34 | o syscall.Overlapped 35 | 36 | // fields used by runtime.netpoll 37 | runtimeCtx uintptr 38 | mode int32 39 | errno int32 40 | qty uint32 41 | 42 | // fields used only by net package 43 | fd *netFD 44 | errc chan error 45 | buf syscall.WSABuf 46 | sa syscall.Sockaddr 47 | rsa *syscall.RawSockaddrAny 48 | rsan int32 49 | handle syscall.Handle 50 | flags uint32 51 | } 52 | -------------------------------------------------------------------------------- /protocol/netpoll.go: -------------------------------------------------------------------------------- 1 | package suft 2 | 3 | import ( 4 | "net" 5 | "unsafe" 6 | ) 7 | 8 | func net_pollSetDeadline(ctx uintptr, d int64, mode int) 9 | 10 | func popcnt(src uint64) int 11 | 12 | type fdMutex struct { 13 | state uint64 14 | rsema uint32 15 | wsema uint32 16 | } 17 | 18 | type pollDesc struct { 19 | runtimeCtx uintptr 20 | } 21 | 22 | type netConn struct { 23 | fd *netFD 24 | } 25 | 26 | func getPollCtx(c *net.UDPConn) uintptr { 27 | connPtr := (*netConn)(unsafe.Pointer(c)) 28 | return connPtr.fd.pd.runtimeCtx 29 | } 30 | -------------------------------------------------------------------------------- /protocol/packet.go: -------------------------------------------------------------------------------- 1 | package suft 2 | 3 | import ( 4 | "encoding/binary" 5 | "fmt" 6 | ) 7 | 8 | const ( 9 | _F_NIL = 0 10 | _F_SYN = 1 11 | _F_ACK = 1 << 1 12 | _F_SACK = 1 << 2 13 | _F_TIME = 1 << 3 14 | _F_DATA = 1 << 4 15 | // reserved = 1 << 5 16 | _F_RESET = 1 << 6 17 | _F_FIN = 1 << 7 18 | ) 19 | 20 | var packetTypeNames = map[byte]string{ 21 | 0: "NOOP", 22 | 1: "SYN", 23 | 2: "ACK", 24 | 3: "SYN+ACK", 25 | 4: "SACK", 26 | 8: "TIME", 27 | 12: "SACK+TIME", 28 | 16: "DATA", 29 | 64: "RESET", 30 | 128: "FIN", 31 | 192: "FIN+RESET", 32 | } 33 | 34 | const ( 35 | _S_FIN = iota 36 | _S_FIN0 37 | _S_FIN1 38 | _S_SYN0 39 | _S_SYN1 40 | _S_EST0 41 | _S_EST1 42 | ) 43 | 44 | // Magic-6 | TH-10 | CH-10 | payload 45 | const ( 46 | _MAGIC_SIZE = 6 47 | _TH_SIZE = 10 + _MAGIC_SIZE 48 | _CH_SIZE = 10 49 | _AH_SIZE = _TH_SIZE + _CH_SIZE 50 | ) 51 | 52 | const ( 53 | // Max UDP payload: 1500 MTU - 20 IP hdr - 8 UDP hdr = 1472 bytes 54 | // Then: MSS = 1472-26 = 1446 55 | // And For ADSL: 1446-8 = 1438 56 | _MSS = 1438 57 | ) 58 | 59 | const ( 60 | _SENT_OK = 0xff 61 | ) 62 | 63 | type packet struct { 64 | seq uint32 65 | ack uint32 66 | flag uint8 67 | scnt uint8 68 | payload []byte 69 | buffer []byte 70 | } 71 | 72 | func (p *packet) marshall(id connID) []byte { 73 | buf := p.buffer 74 | if buf == nil { 75 | buf = make([]byte, _AH_SIZE+len(p.payload)) 76 | copy(buf[_TH_SIZE+10:], p.payload) 77 | } 78 | binary.BigEndian.PutUint16(buf[_MAGIC_SIZE:], uint16(len(buf))) 79 | binary.BigEndian.PutUint32(buf[_MAGIC_SIZE+2:], id.rid) 80 | binary.BigEndian.PutUint32(buf[_MAGIC_SIZE+6:], id.lid) 81 | binary.BigEndian.PutUint32(buf[_TH_SIZE:], p.seq) 82 | binary.BigEndian.PutUint32(buf[_TH_SIZE+4:], p.ack) 83 | buf[_TH_SIZE+8] = p.flag 84 | buf[_TH_SIZE+9] = p.scnt 85 | return buf 86 | } 87 | 88 | func unmarshall(pk *packet, buf []byte) { 89 | if len(buf) >= _CH_SIZE { 90 | pk.seq = binary.BigEndian.Uint32(buf) 91 | pk.ack = binary.BigEndian.Uint32(buf[4:]) 92 | pk.flag = buf[8] 93 | pk.scnt = buf[9] 94 | pk.payload = buf[10:] 95 | } 96 | } 97 | 98 | func (n *qNode) String() string { 99 | now := Now() 100 | return fmt.Sprintf("type=%s seq=%d scnt=%d sndtime~=%d,%d miss=%d", 101 | packetTypeNames[n.flag], n.seq, n.scnt, n.sent-now, n.sent_1-now, n.miss) 102 | } 103 | 104 | func maxI64(a, b int64) int64 { 105 | if a >= b { 106 | return a 107 | } else { 108 | return b 109 | } 110 | } 111 | 112 | func maxU32(a, b uint32) uint32 { 113 | if a >= b { 114 | return a 115 | } else { 116 | return b 117 | } 118 | } 119 | 120 | func minI64(a, b int64) int64 { 121 | if a <= b { 122 | return a 123 | } else { 124 | return b 125 | } 126 | } 127 | 128 | func maxI32(a, b int32) int32 { 129 | if a >= b { 130 | return a 131 | } else { 132 | return b 133 | } 134 | } 135 | 136 | func minI32(a, b int32) int32 { 137 | if a <= b { 138 | return a 139 | } else { 140 | return b 141 | } 142 | } 143 | -------------------------------------------------------------------------------- /protocol/state.go: -------------------------------------------------------------------------------- 1 | package suft 2 | 3 | import ( 4 | "errors" 5 | "log" 6 | "net" 7 | "sync" 8 | "sync/atomic" 9 | "time" 10 | ) 11 | 12 | const ( 13 | _10ms = time.Millisecond * 10 14 | _100ms = time.Millisecond * 100 15 | ) 16 | 17 | const ( 18 | _FIN_ACK_SEQ uint32 = 0xffFF0000 19 | _INVALID_SEQ uint32 = 0xffFFffFF 20 | ) 21 | 22 | var ( 23 | ErrIOTimeout error = &TimeoutError{} 24 | ErrUnknown = errors.New("Unknown error") 25 | ErrInexplicableData = errors.New("Inexplicable data") 26 | ErrTooManyAttempts = errors.New("Too many attempts to connect") 27 | ) 28 | 29 | type TimeoutError struct{} 30 | 31 | func (e *TimeoutError) Error() string { return "i/o timeout" } 32 | func (e *TimeoutError) Timeout() bool { return true } 33 | func (e *TimeoutError) Temporary() bool { return true } 34 | 35 | type Conn struct { 36 | sock *net.UDPConn 37 | dest *net.UDPAddr 38 | edp *Endpoint 39 | connID connID // 8 bytes 40 | // events 41 | evRecv chan []byte 42 | evRead chan byte 43 | evSend chan byte 44 | evSWnd chan byte 45 | evAck chan byte 46 | evClose chan byte 47 | // protocol state 48 | inlock sync.Mutex 49 | outlock sync.Mutex 50 | state int32 51 | mySeq uint32 52 | swnd int32 53 | cwnd int32 54 | missed int32 55 | outPending int32 56 | lastAck uint32 57 | lastAckTime int64 58 | lastAckTime2 int64 59 | lastShrink int64 60 | lastRstMis int64 61 | ato int64 62 | rto int64 63 | rtt int64 64 | srtt int64 65 | mdev int64 66 | rtmo int64 67 | wtmo int64 68 | tSlot int64 69 | tSlotT0 int64 70 | lastSErr int64 71 | // queue 72 | outQ *linkedMap 73 | inQ *linkedMap 74 | inQReady []byte 75 | inQDirty bool 76 | lastReadSeq uint32 // last user read seq 77 | // params 78 | bandwidth int64 79 | fastRetransmit bool 80 | flatTraffic bool 81 | mss int 82 | // statistics 83 | urgent int 84 | inPkCnt int 85 | inDupCnt int 86 | outPkCnt int 87 | outDupCnt int 88 | fRCnt int 89 | } 90 | 91 | func NewConn(e *Endpoint, dest *net.UDPAddr, id connID) *Conn { 92 | c := &Conn{ 93 | sock: e.udpconn, 94 | dest: dest, 95 | edp: e, 96 | connID: id, 97 | evRecv: make(chan []byte, 128), 98 | evRead: make(chan byte, 1), 99 | evSWnd: make(chan byte, 2), 100 | evSend: make(chan byte, 4), 101 | evAck: make(chan byte, 1), 102 | evClose: make(chan byte, 2), 103 | outQ: newLinkedMap(_QModeOut), 104 | inQ: newLinkedMap(_QModeIn), 105 | } 106 | p := e.params 107 | c.bandwidth = p.Bandwidth 108 | c.fastRetransmit = p.FastRetransmit 109 | c.flatTraffic = p.FlatTraffic 110 | c.mss = _MSS 111 | if dest.IP.To4() == nil { 112 | // typical ipv6 header length=40 113 | c.mss -= 20 114 | } 115 | return c 116 | } 117 | 118 | func (c *Conn) initConnection(buf []byte) (err error) { 119 | if buf == nil { 120 | err = c.initDialing() 121 | } else { //server 122 | err = c.acceptConnection(buf[_TH_SIZE:]) 123 | } 124 | if err != nil { 125 | return 126 | } 127 | if c.state == _S_EST1 { 128 | c.lastReadSeq = c.lastAck 129 | c.inQ.maxCtnSeq = c.lastAck 130 | c.rtt = maxI64(c.rtt, _MIN_RTT) 131 | c.mdev = c.rtt << 1 132 | c.srtt = c.rtt << 3 133 | c.rto = maxI64(c.rtt*2, _MIN_RTO) 134 | c.ato = maxI64(c.rtt>>4, _MIN_ATO) 135 | c.ato = minI64(c.ato, _MAX_ATO) 136 | // initial cwnd 137 | c.swnd = calSwnd(c.bandwidth, c.rtt) >> 1 138 | c.cwnd = 8 139 | go c.internalRecvLoop() 140 | go c.internalSendLoop() 141 | go c.internalAckLoop() 142 | if debug >= 0 { 143 | go c.internal_state() 144 | } 145 | return nil 146 | } else { 147 | return ErrUnknown 148 | } 149 | } 150 | 151 | func (c *Conn) initDialing() error { 152 | // first syn 153 | pk := &packet{ 154 | seq: c.mySeq, 155 | flag: _F_SYN, 156 | } 157 | item := nodeOf(pk) 158 | var buf []byte 159 | c.state = _S_SYN0 160 | t0 := Now() 161 | for i := 0; i < _MAX_RETRIES && c.state == _S_SYN0; i++ { 162 | // send syn 163 | c.internalWrite(item) 164 | select { 165 | case buf = <-c.evRecv: 166 | c.rtt = Now() - t0 167 | c.state = _S_SYN1 168 | c.connID.setRid(buf) 169 | buf = buf[_TH_SIZE:] 170 | case <-time.After(time.Second): 171 | continue 172 | } 173 | } 174 | if c.state == _S_SYN0 { 175 | return ErrTooManyAttempts 176 | } 177 | 178 | unmarshall(pk, buf) 179 | // expected syn+ack 180 | if pk.flag == _F_SYN|_F_ACK && pk.ack == c.mySeq { 181 | if scnt := pk.scnt - 1; scnt > 0 { 182 | c.rtt -= int64(scnt) * 1e3 183 | } 184 | log.Println("rtt", c.rtt) 185 | c.state = _S_EST0 186 | // build ack3 187 | pk.scnt = 0 188 | pk.ack = pk.seq 189 | pk.flag = _F_ACK 190 | item := nodeOf(pk) 191 | // send ack3 192 | c.internalWrite(item) 193 | // update lastAck 194 | c.logAck(pk.ack) 195 | c.state = _S_EST1 196 | return nil 197 | } else { 198 | return ErrInexplicableData 199 | } 200 | } 201 | 202 | func (c *Conn) acceptConnection(buf []byte) error { 203 | var pk = new(packet) 204 | var item *qNode 205 | unmarshall(pk, buf) 206 | // expected syn 207 | if pk.flag == _F_SYN { 208 | c.state = _S_SYN1 209 | // build syn+ack 210 | pk.ack = pk.seq 211 | pk.seq = c.mySeq 212 | pk.flag |= _F_ACK 213 | // update lastAck 214 | c.logAck(pk.ack) 215 | item = nodeOf(pk) 216 | item.scnt = pk.scnt - 1 217 | } else { 218 | dumpb("Syn1 ?", buf) 219 | return ErrInexplicableData 220 | } 221 | for i := 0; i < 5 && c.state == _S_SYN1; i++ { 222 | t0 := Now() 223 | // reply syn+ack 224 | c.internalWrite(item) 225 | // recv ack3 226 | select { 227 | case buf = <-c.evRecv: 228 | c.state = _S_EST0 229 | c.rtt = Now() - t0 230 | buf = buf[_TH_SIZE:] 231 | log.Println("rtt", c.rtt) 232 | case <-time.After(time.Second): 233 | continue 234 | } 235 | } 236 | if c.state == _S_SYN1 { 237 | return ErrTooManyAttempts 238 | } 239 | 240 | pk = new(packet) 241 | unmarshall(pk, buf) 242 | // expected ack3 243 | if pk.flag == _F_ACK && pk.ack == c.mySeq { 244 | c.state = _S_EST1 245 | } else { 246 | // if ack3 lost, resend syn+ack 3-times 247 | // and drop these coming data 248 | if pk.flag&_F_DATA != 0 && pk.seq > c.lastAck { 249 | c.internalWrite(item) 250 | c.state = _S_EST1 251 | } else { 252 | dumpb("Ack3 ?", buf) 253 | return ErrInexplicableData 254 | } 255 | } 256 | return nil 257 | } 258 | 259 | // 20,20,20,20, 100,100,100,100, 1s,1s,1s,1s 260 | func selfSpinWait(fn func() bool) error { 261 | const _MAX_SPIN = 12 262 | for i := 0; i < _MAX_SPIN; i++ { 263 | if fn() { 264 | return nil 265 | } else if i <= 3 { 266 | time.Sleep(_10ms * 2) 267 | } else if i <= 7 { 268 | time.Sleep(_100ms) 269 | } else { 270 | time.Sleep(time.Second) 271 | } 272 | } 273 | return ErrIOTimeout 274 | } 275 | 276 | func (c *Conn) IsClosed() bool { 277 | return atomic.LoadInt32(&c.state) <= _S_FIN1 278 | } 279 | 280 | /* 281 | active close: 282 | 1 <- send fin-W: closeW() 283 | before sending, ensure all outQ items has beed sent out and all of them has been acked. 284 | 2 -> wait to recv ack{fin-W} 285 | then trigger closeR, including send fin-R and wait to recv ack{fin-R} 286 | 287 | passive close: 288 | -> fin: 289 | if outQ is not empty then self-spin wait. 290 | if outQ empty, send ack{fin-W} then goto closeW(). 291 | */ 292 | func (c *Conn) Close() (err error) { 293 | if !atomic.CompareAndSwapInt32(&c.state, _S_EST1, _S_FIN0) { 294 | return selfSpinWait(func() bool { 295 | return atomic.LoadInt32(&c.state) == _S_FIN 296 | }) 297 | } 298 | var err0 error 299 | err0 = c.closeW() 300 | // waiting for fin-2 of peer 301 | err = selfSpinWait(func() bool { 302 | select { 303 | case v := <-c.evClose: 304 | if v == _S_FIN { 305 | return true 306 | } else { 307 | time.AfterFunc(_100ms, func() { c.evClose <- v }) 308 | } 309 | default: 310 | } 311 | return false 312 | }) 313 | defer c.afterShutdown() 314 | if err != nil { 315 | // backup path for wait ack(finW) timeout 316 | c.closeR(nil) 317 | } 318 | if err0 != nil { 319 | return err0 320 | } else { 321 | return 322 | } 323 | } 324 | 325 | func (c *Conn) beforeCloseW() (err error) { 326 | // check outQ was empty and all has been acked. 327 | // self-spin waiting 328 | for i := 0; i < 2; i++ { 329 | err = selfSpinWait(func() bool { 330 | return atomic.LoadInt32(&c.outPending) <= 0 331 | }) 332 | if err == nil { 333 | break 334 | } 335 | } 336 | // send fin, reliably 337 | c.outlock.Lock() 338 | c.mySeq++ 339 | c.outPending++ 340 | pk := &packet{seq: c.mySeq, flag: _F_FIN} 341 | item := nodeOf(pk) 342 | c.outQ.appendTail(item) 343 | c.internalWrite(item) 344 | c.outlock.Unlock() 345 | c.evSWnd <- _VSWND_ACTIVE 346 | return 347 | } 348 | 349 | func (c *Conn) closeW() (err error) { 350 | // close resource of sending 351 | defer c.afterCloseW() 352 | // send fin 353 | err = c.beforeCloseW() 354 | var closed bool 355 | var max = 20 356 | if c.rtt > 200 { 357 | max = int(c.rtt) / 10 358 | } 359 | // waiting for outQ means: 360 | // 1. all outQ has been acked, for passive 361 | // 2. fin has been acked, for active 362 | for i := 0; i < max && (atomic.LoadInt32(&c.outPending) > 0 || !closed); i++ { 363 | select { 364 | case v := <-c.evClose: 365 | if v == _S_FIN0 { 366 | // namely, last fin has been acked. 367 | closed = true 368 | } else { 369 | time.AfterFunc(_100ms, func() { c.evClose <- v }) 370 | } 371 | case <-time.After(_100ms): 372 | } 373 | } 374 | if closed || err != nil { 375 | return 376 | } else { 377 | return ErrIOTimeout 378 | } 379 | } 380 | 381 | func (c *Conn) afterCloseW() { 382 | // can't close(c.evRecv), avoid endpoint dispatch exception 383 | // stop pending inputAndSend 384 | select { 385 | case c.evSend <- _CLOSE: 386 | default: 387 | } 388 | // stop internalSendLoop 389 | c.evSWnd <- _CLOSE 390 | } 391 | 392 | // called by active and passive close() 393 | func (c *Conn) afterShutdown() { 394 | // stop internalRecvLoop 395 | c.evRecv <- nil 396 | // remove registry 397 | c.edp.removeConn(c.connID, c.dest) 398 | log.Println("shutdown", c.state) 399 | } 400 | 401 | // trigger by reset 402 | func (c *Conn) forceShutdownWithLock() { 403 | c.outlock.Lock() 404 | defer c.outlock.Unlock() 405 | c.forceShutdown() 406 | } 407 | 408 | // called by: 409 | // 1/ send exception 410 | // 2/ recv reset 411 | // drop outQ and force shutdown 412 | func (c *Conn) forceShutdown() { 413 | if atomic.CompareAndSwapInt32(&c.state, _S_EST1, _S_FIN) { 414 | defer c.afterShutdown() 415 | // stop sender 416 | for i := 0; i < cap(c.evSend); i++ { 417 | select { 418 | case <-c.evSend: 419 | default: 420 | } 421 | } 422 | select { 423 | case c.evSend <- _CLOSE: 424 | default: 425 | } 426 | c.outQ.reset() 427 | // stop reader 428 | close(c.evRead) 429 | c.inQ.reset() 430 | // stop internalLoops 431 | c.evSWnd <- _CLOSE 432 | c.evAck <- _CLOSE 433 | //log.Println("force shutdown") 434 | } 435 | } 436 | 437 | // for sending fin failed 438 | func (c *Conn) fakeShutdown() { 439 | select { 440 | case c.evClose <- _S_FIN0: 441 | default: 442 | } 443 | } 444 | 445 | func (c *Conn) closeR(pk *packet) { 446 | var passive = true 447 | for { 448 | state := atomic.LoadInt32(&c.state) 449 | switch state { 450 | case _S_FIN: 451 | return 452 | case _S_FIN1: // multiple FIN, maybe lost 453 | c.passiveCloseReply(pk, false) 454 | return 455 | case _S_FIN0: // active close preformed 456 | passive = false 457 | } 458 | if !atomic.CompareAndSwapInt32(&c.state, state, _S_FIN1) { 459 | continue 460 | } 461 | c.passiveCloseReply(pk, true) 462 | break 463 | } 464 | // here, R is closed. 465 | // ^^^^^^^^^^^^^^^^^^^^^ 466 | if passive { 467 | // passive closing call closeW contains sending fin and recv ack 468 | // may the ack of fin-2 was lost, then the closeW will timeout 469 | c.closeW() 470 | } 471 | // here, R,W both were closed. 472 | // ^^^^^^^^^^^^^^^^^^^^^ 473 | atomic.StoreInt32(&c.state, _S_FIN) 474 | // stop internalAckLoop 475 | c.evAck <- _CLOSE 476 | 477 | if passive { 478 | // close evRecv within here 479 | c.afterShutdown() 480 | } else { 481 | // notify active close thread 482 | select { 483 | case c.evClose <- _S_FIN: 484 | default: 485 | } 486 | } 487 | } 488 | 489 | func (c *Conn) passiveCloseReply(pk *packet, first bool) { 490 | if pk != nil && pk.flag&_F_FIN != 0 { 491 | if first { 492 | c.checkInQ(pk) 493 | close(c.evRead) 494 | } 495 | // ack the FIN 496 | pk = &packet{seq: _FIN_ACK_SEQ, ack: pk.seq, flag: _F_ACK} 497 | item := nodeOf(pk) 498 | c.internalWrite(item) 499 | } 500 | } 501 | 502 | // check inQ ends orderly, and copy queue data to user space 503 | func (c *Conn) checkInQ(pk *packet) { 504 | if nil != selfSpinWait(func() bool { 505 | return c.inQ.maxCtnSeq+1 == pk.seq 506 | }) { // timeout for waiting inQ to finish 507 | return 508 | } 509 | c.inlock.Lock() 510 | defer c.inlock.Unlock() 511 | if c.inQ.size() > 0 { 512 | for i := c.inQ.head; i != nil; i = i.next { 513 | c.inQReady = append(c.inQReady, i.payload...) 514 | } 515 | } 516 | } 517 | -------------------------------------------------------------------------------- /protocol/stopwatch.go: -------------------------------------------------------------------------------- 1 | package suft 2 | 3 | import ( 4 | "fmt" 5 | "os" 6 | "time" 7 | ) 8 | 9 | type watch struct { 10 | label string 11 | t1 time.Time 12 | } 13 | 14 | func StartWatch(s string) *watch { 15 | return &watch{ 16 | label: s, 17 | t1: time.Now(), 18 | } 19 | } 20 | 21 | func (w *watch) StopLoops(loop int, size int) { 22 | tu := time.Now().Sub(w.t1).Nanoseconds() 23 | timePerLoop := float64(tu) / float64(loop) 24 | throughput := float64(loop*size) * 1e6 / float64(tu) 25 | tu_ms := float64(tu) / 1e6 26 | fmt.Fprintf(os.Stderr, "%s tu=%.2f ms tpl=%.0f ns throughput=%.2f K/s\n", w.label, tu_ms, timePerLoop, throughput) 27 | } 28 | 29 | var _kt = float64(1e9 / 1024) 30 | 31 | func (w *watch) Stop(size int) { 32 | tu := time.Now().Sub(w.t1).Nanoseconds() 33 | throughput := float64(size) * _kt / float64(tu) 34 | tu_ms := float64(tu) / 1e6 35 | fmt.Fprintf(os.Stderr, "%s tu=%.2f ms throughput=%.2f K/s\n", w.label, tu_ms, throughput) 36 | } 37 | -------------------------------------------------------------------------------- /protocol/timer.go: -------------------------------------------------------------------------------- 1 | package suft 2 | 3 | const Millisecond = 1e6 4 | 5 | type iTimer struct { 6 | C chan byte 7 | r runtimeTimer 8 | } 9 | 10 | type runtimeTimer struct { 11 | i int 12 | when int64 13 | period int64 14 | f func(c interface{}, seq uintptr) // NOTE: must not be closure 15 | arg interface{} 16 | seq uintptr 17 | } 18 | 19 | func startTimer(*runtimeTimer) 20 | func stopTimer(*runtimeTimer) bool 21 | func runtimeNano() int64 22 | func currentTime() (sec int64, nsec int32) 23 | 24 | func Now() int64 { 25 | s, n := currentTime() 26 | return s*1e3 + int64(n/1e6) 27 | } 28 | 29 | func NowNS() int64 { 30 | s, n := currentTime() 31 | return s*1e9 + int64(n) 32 | } 33 | 34 | func newTimer(d int64) *iTimer { 35 | t := iTimer{ 36 | C: make(chan byte, 1), 37 | r: runtimeTimer{ 38 | i: -1, 39 | when: d*Millisecond + runtimeNano(), 40 | f: timerCallback, 41 | }, 42 | } 43 | t.r.arg = t.C 44 | return &t 45 | } 46 | 47 | func timerCallback(c interface{}, seq uintptr) { 48 | select { 49 | case c.(chan byte) <- 1: 50 | default: 51 | } 52 | } 53 | 54 | func (t *iTimer) Stop() { 55 | stopTimer(&t.r) 56 | } 57 | 58 | func (t *iTimer) Reset(d int64) { 59 | stopTimer(&t.r) 60 | select { 61 | case <-t.C: 62 | default: 63 | } 64 | t.r.when = d*Millisecond + runtimeNano() 65 | startTimer(&t.r) 66 | } 67 | 68 | func (t *iTimer) TryActive(d int64) { 69 | if t.r.i < 0 { 70 | select { 71 | case <-t.C: 72 | default: 73 | } 74 | t.r.when = d*Millisecond + runtimeNano() 75 | startTimer(&t.r) 76 | } 77 | } 78 | 79 | func NewTimerChan(d int64) <-chan byte { 80 | t := iTimer{ 81 | C: make(chan byte, 1), 82 | r: runtimeTimer{ 83 | i: -1, 84 | when: d*Millisecond + runtimeNano(), 85 | f: timerCallback, 86 | }, 87 | } 88 | t.r.arg = t.C 89 | startTimer(&t.r) 90 | return t.C 91 | } 92 | -------------------------------------------------------------------------------- /protocol/timer.s: -------------------------------------------------------------------------------- 1 | #include "textflag.h" 2 | 3 | TEXT ·currentTime(SB),NOSPLIT,$0 4 | JMP time·now(SB) 5 | 6 | TEXT ·startTimer(SB),NOSPLIT,$0 7 | JMP time·startTimer(SB) 8 | 9 | TEXT ·stopTimer(SB),NOSPLIT,$0 10 | JMP time·stopTimer(SB) 11 | 12 | TEXT ·runtimeNano(SB),NOSPLIT,$0 13 | JMP time·runtimeNano(SB) 14 | 15 | TEXT ·net_pollSetDeadline(SB),NOSPLIT,$0 16 | JMP net·runtime_pollSetDeadline(SB) 17 | 18 | -------------------------------------------------------------------------------- /protocol/timing_test.go: -------------------------------------------------------------------------------- 1 | package suft 2 | 3 | import ( 4 | "math" 5 | "testing" 6 | "time" 7 | ) 8 | 9 | func Test_sleep(t *testing.T) { 10 | const loops = 10 11 | var intervals = [...]int64{1, 1e3, 1e4, 1e5, 1e6, 1e7} 12 | var ret [len(intervals)][loops]int64 13 | for j := 0; j < len(intervals); j++ { 14 | v := time.Duration(intervals[j]) 15 | for i := 0; i < loops; i++ { 16 | t0 := NowNS() 17 | time.Sleep(v) 18 | ret[j][i] = NowNS() - t0 19 | } 20 | } 21 | for j := 0; j < len(intervals); j++ { 22 | var exp, sum, stdev float64 23 | exp = float64(intervals[j]) 24 | for _, v := range ret[j] { 25 | sum += float64(v) 26 | stdev += math.Pow(float64(v)-exp, 2) 27 | } 28 | stdev /= float64(loops) 29 | stdev = math.Sqrt(stdev) 30 | t.Logf("interval=%s sleeping=%s stdev/k=%.2f", time.Duration(intervals[j]), time.Duration(sum/loops), stdev/1e3) 31 | } 32 | } 33 | -------------------------------------------------------------------------------- /wireshark-plugin/README.md: -------------------------------------------------------------------------------- 1 | # Usage 2 | 3 | Run `wireshark -X lua_script:suft.lua`. 4 | 5 | # Install 6 | 7 | Copy `suft.lua` to wireshark/plugins/. 8 | 9 | Append `dofile(DATA_DIR.."/plugins/suft.lua")` at the end of wireshark/init.lua. 10 | -------------------------------------------------------------------------------- /wireshark-plugin/suft.lua: -------------------------------------------------------------------------------- 1 | -- create suft_p protocol and its fields 2 | suft_p = Proto.new("suft","SUFT Protocol") 3 | 4 | -- ProtoField.new(name, abbr, type, [voidstring], [base], [mask], [descr]) 5 | local f_margic = ProtoField.new("Magic Header", "suft.magic", ftypes.BYTES) 6 | local f_len = ProtoField.new("Length", "suft.len", ftypes.UINT16) 7 | local f_rcid = ProtoField.new("Remote ConnId", "suft.rcid", ftypes.UINT32) 8 | local f_lcid = ProtoField.new("Local ConnId", "suft.lcid", ftypes.UINT32) 9 | local f_seq = ProtoField.new("Seq", "suft.seq", ftypes.UINT32) 10 | local f_ack = ProtoField.new("Ack", "suft.ack", ftypes.UINT32) 11 | local f_flags = ProtoField.new("Flags", "suft.flags", ftypes.UINT8, nil, base.HEX) 12 | local f_flag_syn = ProtoField.new("SYN", "suft.flags.syn", ftypes.UINT8, nil, base.HEX, 1) 13 | local f_flag_ack = ProtoField.new("ACK", "suft.flags.ack", ftypes.UINT8, nil, base.HEX, 2) 14 | local f_flag_sack = ProtoField.new("SACK", "suft.flags.sack", ftypes.UINT8, nil, base.HEX, 4) 15 | local f_flag_time = ProtoField.new("TIME", "suft.flags.time", ftypes.UINT8, nil, base.HEX, 8) 16 | local f_flag_data = ProtoField.new("DATA", "suft.flags.data", ftypes.UINT8, nil, base.HEX, 16) 17 | local f_flag_reset = ProtoField.new("RESET", "suft.flags.reset", ftypes.UINT8, nil, base.HEX, 64) 18 | local f_flag_fin = ProtoField.new("FIN", "suft.flags.fin", ftypes.UINT8, nil, base.HEX, 128) 19 | local f_scnt = ProtoField.new("Sent Count", "suft.scnt", ftypes.UINT8) 20 | local f_payload = ProtoField.new("Payload", "suft.payload", ftypes.BYTES) 21 | 22 | local f_sack_tbl = ProtoField.new("SACK TBL", "suft.sack.tbl", ftypes.UINT8) 23 | local f_sack_scnt = ProtoField.new("SACK Sent Count", "suft.sack.scnt", ftypes.UINT8) 24 | local f_sack_delayed = ProtoField.new("SACK Delayed", "suft.sack.delayed", ftypes.UINT16) 25 | local f_sack_bitmap = ProtoField.new("SACK Bitmap", "suft.sack.bitmap", ftypes.UINT64, nil, base.HEX) 26 | 27 | suft_p.fields = { 28 | f_margic,f_len,f_rcid,f_lcid,f_seq,f_ack,f_flags, 29 | f_flag_syn,f_flag_ack,f_flag_sack,f_flag_time,f_flag_data,f_flag_reset,f_flag_fin, 30 | f_scnt,f_payload,f_sack_tbl,f_sack_scnt,f_sack_delayed,f_sack_bitmap, 31 | } 32 | 33 | local packet_type = { 34 | [1] = "SYN", 35 | [2] = "ACK", 36 | [3] = "SYN+ACK", 37 | [4] = "SACK", 38 | [12] = "SACK+TIME", 39 | [16] = "DATA", 40 | [128] = "FIN", 41 | [192] = "FIN+RESET", 42 | } 43 | 44 | -- suft_p dissector function 45 | function suft_p.dissector (buf, pinfo, root) 46 | -- validate packet length is adequate, otherwise quit 47 | if buf:len() < 26 then return end 48 | pinfo.cols.protocol = suft_p.name 49 | 50 | -- create subtree for suft_p 51 | local subtree = root:add(suft_p, buf()) 52 | -- add protocol fields to subtree 53 | subtree:add(f_margic, buf(0,6)) 54 | subtree:add(f_len, buf(6,2)) 55 | subtree:add(f_rcid, buf(8,4)) 56 | subtree:add(f_lcid, buf(12,4)) 57 | local v_seq, v_ack = buf(16,4), buf(20,4) 58 | local t_seq = subtree:add(f_seq, v_seq) 59 | subtree:add(f_ack, v_ack) 60 | 61 | local flags = buf(24,1) 62 | local ft = subtree:add(f_flags, flags) 63 | ft:add(f_flag_syn, flags) 64 | ft:add(f_flag_ack, flags) 65 | ft:add(f_flag_sack, flags) 66 | ft:add(f_flag_time, flags) 67 | ft:add(f_flag_data, flags) 68 | ft:add(f_flag_reset, flags) 69 | ft:add(f_flag_fin, flags) 70 | local v_flag = flags:uint() 71 | local ptype = packet_type[v_flag] 72 | local v_scnt = buf(25,1) 73 | if ptype then 74 | local fmt = "[%s] seq=%u ack=%u scnt=%u" 75 | if v_flag == 1 then 76 | fmt = fmt .. " ###### New Connection ######" 77 | end 78 | pinfo.cols.info = string.format(fmt, ptype, v_seq:uint(), v_ack:uint(), v_scnt:uint()) 79 | ft:append_text(string.format(" [%s]", ptype)) 80 | end 81 | 82 | subtree:add(f_scnt, v_scnt) 83 | local payload = subtree:add(f_payload, buf(26)) 84 | if bit32.btest(v_flag, 4) then -- sack 85 | t_seq:append_text(" [TRP]") 86 | local sack_tbl,sack_scnt,sack_delayed = buf(26,1),buf(27,1),buf(28,2) 87 | payload:add(f_sack_tbl, sack_tbl) 88 | payload:add(f_sack_scnt, sack_scnt) 89 | payload:add(f_sack_delayed, sack_delayed) 90 | local i, max_i = 0, buf:len() - 8 91 | for i=30,max_i,8 do 92 | payload:add(f_sack_bitmap, buf(i,8)):append_text(string.format(" [%u]", (i-30)/8)) 93 | end 94 | elseif bit32.btest(v_flag, 16) then -- data 95 | payload:append_text(string.format(" [Len %u]", buf:len()-26)) 96 | end 97 | end 98 | 99 | 100 | local dissector_table = DissectorTable.get("udp.port") 101 | dissector_table:add(9090, suft_p) 102 | dissector_table:add(9008, suft_p) 103 | dissector_table:add(5001, suft_p) 104 | --------------------------------------------------------------------------------