├── Debian9 └── x86_64 │ ├── linux-headers-4.14.129-bbrplus.deb │ └── linux-image-4.14.129-bbrplus.deb ├── LICENSE ├── README.md ├── centos7 └── x86_64 │ └── kernel-4.14.129-bbrplus.rpm ├── ok_bbrplus_centos.sh └── tcp_bbrplus.c /Debian9/x86_64/linux-headers-4.14.129-bbrplus.deb: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/cx9208/bbrplus/8e46a57d4ae51229b48d905f990e345c7b581ffb/Debian9/x86_64/linux-headers-4.14.129-bbrplus.deb -------------------------------------------------------------------------------- /Debian9/x86_64/linux-image-4.14.129-bbrplus.deb: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/cx9208/bbrplus/8e46a57d4ae51229b48d905f990e345c7b581ffb/Debian9/x86_64/linux-image-4.14.129-bbrplus.deb -------------------------------------------------------------------------------- /LICENSE: -------------------------------------------------------------------------------- 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 | # BBRplus 2 | 3 | 在https://blog.csdn.net/dog250/article/details/80629551 中， 4 | dog250大神提到了bbr初版的两个问题：bbr在高丢包率下易失速以及bbr收敛慢的问题， 5 | 提到了他个人与bbr作者对这两个问题的一些修正，并在文末给出了修正后的完整代码。 6 | 在这里我**只是将它编译出来（不是我写的）并做成了一键脚本**，我叫它bbr修正版，或者bbrplus。 7 | 它基于原版bbr，但修正了bbr存在的上述问题，尝试使其更好，减少排队和丢包。 8 | 9 | 由于编译修正后的模块需要4.14版的内核， 10 | 以及需要修改内核的部分源码，所以需要重新编译整个内核。 11 | 这里提供一个编译好并内置bbrplus的适用于centos7的内核，以及安装方法与编译供方法大家测试。 12 | 13 | **感谢dog250大神对bbr相关原理和代码的解析与分享！** 14 | 15 | **注意，这是一个实验性的修改，没有人对它的稳定性负责，也不担保它一定能产生正向的效果。 16 | 所以请酌情使用，at your own risk.** 17 | 18 | # 脚本安装方法： 19 | 20 | **不要在生产环境使用一键脚本，建议手动安装，进不了系统用vnc切内核** 21 | 22 | 一键脚本（全系统）： 23 | 见https://github.com/chiakge/Linux-NetSpeed 24 | 25 | 一键脚本（仅CentOS）： 26 | ```bash 27 | wget "https://github.com/cx9208/bbrplus/raw/master/ok_bbrplus_centos.sh" && chmod +x ok_bbrplus_centos.sh && ./ok_bbrplus_centos.sh 28 | ``` 29 | 安装后，执行uname -r，显示4.14.129-bbrplus则切换内核成功 30 | 执行lsmod | grep bbr，显示有bbrplus则开启成功 31 | 32 | # 手动安装方法： 33 | 1. 34 | 卸载本机的锐速（如果有） 35 | 36 | 2. 37 | 下载内核 38 | wget https://github.com/cx9208/bbrplus/raw/master/centos7/x86_64/kernel-4.14.129-bbrplus.rpm 39 | 40 | 3. 41 | 安装内核 42 | yum install -y kernel-4.14.129-bbrplus.rpm 43 | 44 | 4. 45 | 切换启动内核 46 | grub2-set-default 'CentOS Linux (4.14.129-bbrplus) 7 (Core)' 47 | 48 | 5. 49 | 设置fq 50 | echo "net.core.default_qdisc=fq" >> /etc/sysctl.conf 51 | 设置bbrplus 52 | echo "net.ipv4.tcp_congestion_control=bbrplus" >> /etc/sysctl.conf 53 | 54 | 6. 55 | 重启 56 | reboot 57 | 58 | 7. 59 | 检查内核版本 60 | uname -r 61 | 显示4.14.129-bbrplus则成功 62 | 63 | 检查bbrplus是否已经启动 64 | lsmod | grep bbrplus 65 | 显示有tcp_bbrplus则成功 66 | 67 | # 卸载方法： 68 | 安装别的内核bbrplus自动失效，卸载内核自行谷歌即可 69 | 70 | # 内核编译： 71 | 72 | 只能用于4.14.x内核，更高版本的tcp部分源码有改动，要移植到高版本内核得自己研究 73 | 74 | 下载4.14内核源码 75 | wget https://cdn.kernel.org/pub/linux/kernel/v4.x/linux-4.14.91.tar.xz 76 | 77 | 解压 78 | tar -Jxvf linux-4.14.91.tar.xz -C /root/ 79 | 80 | 修改linux-4.14.91/include/net/inet_connection_sock.h，139行 81 | u64 icsk_ca_priv[112 / sizeof(u64)]; 82 | #define ICSK_CA_PRIV_SIZE (14 * sizeof(u64)) 83 | 这两段数值改为112和14，如上 84 | 85 | 修改/net/ipv4/tcp_output.c#L，1823行 86 | tcp_snd_wnd_test函数大括号后} 87 | 换行添加EXPORT_SYMBOL(tcp_snd_wnd_test); 88 | 89 | 添加tcp_bbrplus.c，删除/net/ipv4/tcp_bbr.c 90 | 修改linux-4.14.91/net/ipv4/Makefile， 91 | obj-$(CONFIG_TCP_CONG_BBR) += tcp_bbrplus.o，bbr改为bbrplus 92 | 93 | 安装依赖 94 | centos 95 | yum -y groupinstall Development tools 96 | yum -y install ncurses-devel bc gcc gcc-c++ ncurses ncurses-devel cmake elfutils-libelf-devel openssl-devel rpm-build redhat-rpm-config asciidoc hmaccalc perl-ExtUtils-Embed xmlto audit-libs-devel binutils-devel elfutils-devel elfutils-libelf-devel newt-devel python-devel zlib-devel 97 | 98 | debian 99 | wget -qO- git.io/superupdate.sh | bash 100 | apt-get install build-essential libncurses5-dev 101 | apt-get build-dep linux 102 | 103 | 切换到目录 104 | cd /root/linux-4.14.91 105 | 106 | 配置 107 | make oldconfig 108 | 或者 109 | make menuconfig 110 | 111 | 确保CONFIG_TCP_CONG_BBR=m 112 | 113 | 禁用签名调试 114 | scripts/config --disable MODULE_SIG 115 | scripts/config --disable DEBUG_INFO 116 | 117 | 118 | 开始编译 119 | centos：make rpm-pkg 120 | 121 | debian：make deb-pkg 122 | 123 | -------------------------------------------------------------------------------- /centos7/x86_64/kernel-4.14.129-bbrplus.rpm: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/cx9208/bbrplus/8e46a57d4ae51229b48d905f990e345c7b581ffb/centos7/x86_64/kernel-4.14.129-bbrplus.rpm -------------------------------------------------------------------------------- /ok_bbrplus_centos.sh: -------------------------------------------------------------------------------- 1 | #!/usr/bin/env bash 2 | 3 | #脚本制作:cx9208 4 | kernel_version="4.14.129-bbrplus" 5 | if [[ ! -f /etc/redhat-release ]]; then 6 | echo -e "仅支持centos" 7 | exit 0 8 | fi 9 | 10 | if [[ "$(uname -r)" == "${kernel_version}" ]]; then 11 | echo -e "内核已经安装，无需重复执行。" 12 | exit 0 13 | fi 14 | 15 | #卸载原加速 16 | echo -e "卸载加速..." 17 | sed -i '/net.core.default_qdisc/d' /etc/sysctl.conf 18 | sed -i '/net.ipv4.tcp_congestion_control/d' /etc/sysctl.conf 19 | if [[ -e /appex/bin/serverSpeeder.sh ]]; then 20 | wget --no-check-certificate -O appex.sh https://raw.githubusercontent.com/0oVicero0/serverSpeeder_Install/master/appex.sh && chmod +x appex.sh && bash appex.sh uninstall 21 | rm -f appex.sh 22 | fi 23 | echo -e "下载内核..." 24 | wget https://github.com/cx9208/bbrplus/raw/master/centos7/x86_64/kernel-${kernel_version}.rpm 25 | echo -e "安装内核..." 26 | yum install -y kernel-${kernel_version}.rpm 27 | 28 | #检查内核是否安装成功 29 | list="$(awk -F\' '$1=="menuentry " {print i++ " : " $2}' /etc/grub2.cfg)" 30 | target="CentOS Linux (${kernel_version})" 31 | result=$(echo $list | grep "${target}") 32 | if [[ "$result" = "" ]]; then 33 | echo -e "内核安装失败" 34 | exit 1 35 | fi 36 | 37 | echo -e "切换内核..." 38 | grub2-set-default 'CentOS Linux (${kernel_version}) 7 (Core)' 39 | echo -e "启用模块..." 40 | echo "net.core.default_qdisc=fq" >> /etc/sysctl.conf 41 | echo "net.ipv4.tcp_congestion_control=bbrplus" >> /etc/sysctl.conf 42 | rm -f kernel-${kernel_version}.rpm 43 | 44 | read -p "bbrplus安装完成，现在重启 ? [Y/n] :" yn 45 | [ -z "${yn}" ] && yn="y" 46 | if [[ $yn == [Yy] ]]; then 47 | echo -e "重启中..." 48 | reboot 49 | fi 50 | -------------------------------------------------------------------------------- /tcp_bbrplus.c: -------------------------------------------------------------------------------- 1 | /* Bottleneck Bandwidth and RTT (BBR) congestion control 2 | * 3 | * BBR congestion control computes the sending rate based on the delivery 4 | * rate (throughput) estimated from ACKs. In a nutshell: 5 | * 6 | * On each ACK, update our model of the network path: 7 | * bottleneck_bandwidth = windowed_max(delivered / elapsed, 10 round trips) 8 | * min_rtt = windowed_min(rtt, 10 seconds) 9 | * pacing_rate = pacing_gain * bottleneck_bandwidth 10 | * cwnd = max(cwnd_gain * bottleneck_bandwidth * min_rtt, 4) 11 | * 12 | * The core algorithm does not react directly to packet losses or delays, 13 | * although BBR may adjust the size of next send per ACK when loss is 14 | * observed, or adjust the sending rate if it estimates there is a 15 | * traffic policer, in order to keep the drop rate reasonable. 16 | * 17 | * Here is a state transition diagram for BBR: 18 | * 19 | * | 20 | * V 21 | * +---> STARTUP ----+ 22 | * | | | 23 | * | V | 24 | * | DRAIN ----+ 25 | * | | | 26 | * | V | 27 | * +---> PROBE_BW ----+ 28 | * | ^ | | 29 | * | | | | 30 | * | +----+ | 31 | * | | 32 | * +---- PROBE_RTT <--+ 33 | * 34 | * A BBR flow starts in STARTUP, and ramps up its sending rate quickly. 35 | * When it estimates the pipe is full, it enters DRAIN to drain the queue. 36 | * In steady state a BBR flow only uses PROBE_BW and PROBE_RTT. 37 | * A long-lived BBR flow spends the vast majority of its time remaining 38 | * (repeatedly) in PROBE_BW, fully probing and utilizing the pipe's bandwidth 39 | * in a fair manner, with a small, bounded queue. *If* a flow has been 40 | * continuously sending for the entire min_rtt window, and hasn't seen an RTT 41 | * sample that matches or decreases its min_rtt estimate for 10 seconds, then 42 | * it briefly enters PROBE_RTT to cut inflight to a minimum value to re-probe 43 | * the path's two-way propagation delay (min_rtt). When exiting PROBE_RTT, if 44 | * we estimated that we reached the full bw of the pipe then we enter PROBE_BW; 45 | * otherwise we enter STARTUP to try to fill the pipe. 46 | * 47 | * BBR is described in detail in: 48 | * "BBR: Congestion-Based Congestion Control", 49 | * Neal Cardwell, Yuchung Cheng, C. Stephen Gunn, Soheil Hassas Yeganeh, 50 | * Van Jacobson. ACM Queue, Vol. 14 No. 5, September-October 2016. 51 | * 52 | * There is a public e-mail list for discussing BBR development and testing: 53 | * https://groups.google.com/forum/#!forum/bbr-dev 54 | * 55 | * NOTE: BBR might be used with the fq qdisc ("man tc-fq") with pacing enabled, 56 | * otherwise TCP stack falls back to an internal pacing using one high 57 | * resolution timer per TCP socket and may use more resources. 58 | */ 59 | #include 60 | #include 61 | #include 62 | #include 63 | #include 64 | #include 65 | 66 | /* Scale factor for rate in pkt/uSec unit to avoid truncation in bandwidth 67 | * estimation. The rate unit ~= (1500 bytes / 1 usec / 2^24) ~= 715 bps. 68 | * This handles bandwidths from 0.06pps (715bps) to 256Mpps (3Tbps) in a u32. 69 | * Since the minimum window is >=4 packets, the lower bound isn't 70 | * an issue. The upper bound isn't an issue with existing technologies. 71 | */ 72 | #define BW_SCALE 24 73 | #define BW_UNIT (1 << BW_SCALE) 74 | 75 | #define BBR_SCALE 8 /* scaling factor for fractions in BBR (e.g. gains) */ 76 | #define BBR_UNIT (1 << BBR_SCALE) 77 | 78 | /* BBR has the following modes for deciding how fast to send: */ 79 | enum bbr_mode { 80 | BBR_STARTUP, /* ramp up sending rate rapidly to fill pipe */ 81 | BBR_DRAIN, /* drain any queue created during startup */ 82 | BBR_PROBE_BW, /* discover, share bw: pace around estimated bw */ 83 | BBR_PROBE_RTT, /* cut inflight to min to probe min_rtt */ 84 | }; 85 | 86 | /* BBR congestion control block */ 87 | struct bbr { 88 | u32 min_rtt_us; /* min RTT in min_rtt_win_sec window */ 89 | u32 min_rtt_stamp; /* timestamp of min_rtt_us */ 90 | u32 probe_rtt_done_stamp; /* end time for BBR_PROBE_RTT mode */ 91 | struct minmax bw; /* Max recent delivery rate in pkts/uS << 24 */ 92 | u32 rtt_cnt; /* count of packet-timed rounds elapsed */ 93 | u32 next_rtt_delivered; /* scb->tx.delivered at end of round */ 94 | u64 cycle_mstamp; /* time of this cycle phase start */ 95 | u32 mode:3, /* current bbr_mode in state machine */ 96 | prev_ca_state:3, /* CA state on previous ACK */ 97 | packet_conservation:1, /* use packet conservation? */ 98 | restore_cwnd:1, /* decided to revert cwnd to old value */ 99 | round_start:1, /* start of packet-timed tx->ack round? */ 100 | cycle_len:4, /* phases in this PROBE_BW gain cycle */ 101 | tso_segs_goal:7, /* segments we want in each skb we send */ 102 | idle_restart:1, /* restarting after idle? */ 103 | probe_rtt_round_done:1, /* a BBR_PROBE_RTT round at 4 pkts? */ 104 | unused:8, 105 | lt_is_sampling:1, /* taking long-term ("LT") samples now? */ 106 | lt_rtt_cnt:7, /* round trips in long-term interval */ 107 | lt_use_bw:1; /* use lt_bw as our bw estimate? */ 108 | u32 lt_bw; /* LT est delivery rate in pkts/uS << 24 */ 109 | u32 lt_last_delivered; /* LT intvl start: tp->delivered */ 110 | u32 lt_last_stamp; /* LT intvl start: tp->delivered_mstamp */ 111 | u32 lt_last_lost; /* LT intvl start: tp->lost */ 112 | u32 pacing_gain:10, /* current gain for setting pacing rate */ 113 | cwnd_gain:10, /* current gain for setting cwnd */ 114 | full_bw_cnt:3, /* number of rounds without large bw gains */ 115 | cycle_idx:3, /* current index in pacing_gain cycle array */ 116 | has_seen_rtt:1, /* have we seen an RTT sample yet? */ 117 | unused_b:5; 118 | u32 prior_cwnd; /* prior cwnd upon entering loss recovery */ 119 | u32 full_bw; /* recent bw, to estimate if pipe is full */ 120 | /* For tracking ACK aggregation: */ 121 | u64 ack_epoch_mstamp; 122 | /* start of ACK sampling epoch */ 123 | u16 extra_acked[2]; 124 | /* max excess data ACKed in epoch */ 125 | u32 ack_epoch_acked:20, /* packets (S)ACKed in sampling epoch */ 126 | extra_acked_win_rtts:5, /* age of extra_acked, in round trips */ 127 | extra_acked_win_idx:1, /* current index in extra_acked array */ 128 | unused1:6; 129 | }; 130 | 131 | #define CYCLE_LEN 8 /* number of phases in a pacing gain cycle */ 132 | 133 | /* Window length of bw filter (in rounds): */ 134 | static const int bbr_bw_rtts = CYCLE_LEN + 2; 135 | /* Window length of min_rtt filter (in sec): */ 136 | static const u32 bbr_min_rtt_win_sec = 10; 137 | /* Minimum time (in ms) spent at bbr_cwnd_min_target in BBR_PROBE_RTT mode: */ 138 | static const u32 bbr_probe_rtt_mode_ms = 200; 139 | /* Skip TSO below the following bandwidth (bits/sec): */ 140 | static const int bbr_min_tso_rate = 1200000; 141 | 142 | /* We use a high_gain value of 2/ln(2) because it's the smallest pacing gain 143 | * that will allow a smoothly increasing pacing rate that will double each RTT 144 | * and send the same number of packets per RTT that an un-paced, slow-starting 145 | * Reno or CUBIC flow would: 146 | */ 147 | static const int bbr_high_gain = BBR_UNIT * 2885 / 1000 + 1; 148 | /* The pacing gain of 1/high_gain in BBR_DRAIN is calculated to typically drain 149 | * the queue created in BBR_STARTUP in a single round: 150 | */ 151 | static const int bbr_drain_gain = BBR_UNIT * 1000 / 2885; 152 | /* The gain for deriving steady-state cwnd tolerates delayed/stretched ACKs: */ 153 | static const int bbr_cwnd_gain = BBR_UNIT * 2; 154 | 155 | enum bbr_pacing_gain_phase { 156 | BBR_BW_PROBE_UP = 0, 157 | BBR_BW_PROBE_DOWN = 1, 158 | BBR_BW_PROBE_CRUISE = 2, 159 | }; 160 | 161 | 162 | /* The pacing_gain values for the PROBE_BW gain cycle, to discover/share bw: */ 163 | static const int bbr_pacing_gain[] = { 164 | BBR_UNIT * 5 / 4, /* probe for more available bw */ 165 | BBR_UNIT * 3 / 4, /* drain queue and/or yield bw to other flows */ 166 | BBR_UNIT, BBR_UNIT, BBR_UNIT, /* cruise at 1.0*bw to utilize pipe, */ 167 | BBR_UNIT, BBR_UNIT, BBR_UNIT /* without creating excess queue... */ 168 | }; 169 | /* Randomize the starting gain cycling phase over N phases: */ 170 | static const u32 bbr_cycle_rand = 7; 171 | 172 | /* Try to keep at least this many packets in flight, if things go smoothly. For 173 | * smooth functioning, a sliding window protocol ACKing every other packet 174 | * needs at least 4 packets in flight: 175 | */ 176 | static const u32 bbr_cwnd_min_target = 4; 177 | 178 | /* To estimate if BBR_STARTUP mode (i.e. high_gain) has filled pipe... */ 179 | /* If bw has increased significantly (1.25x), there may be more bw available: */ 180 | static const u32 bbr_full_bw_thresh = BBR_UNIT * 5 / 4; 181 | /* But after 3 rounds w/o significant bw growth, estimate pipe is full: */ 182 | static const u32 bbr_full_bw_cnt = 3; 183 | 184 | /* "long-term" ("LT") bandwidth estimator parameters... */ 185 | /* The minimum number of rounds in an LT bw sampling interval: */ 186 | static const u32 bbr_lt_intvl_min_rtts = 4; 187 | /* If lost/delivered ratio > 20%, interval is "lossy" and we may be policed: */ 188 | static const u32 bbr_lt_loss_thresh = 50; 189 | /* If 2 intervals have a bw ratio <= 1/8, their bw is "consistent": */ 190 | static const u32 bbr_lt_bw_ratio = BBR_UNIT / 8; 191 | /* If 2 intervals have a bw diff <= 4 Kbit/sec their bw is "consistent": */ 192 | static const u32 bbr_lt_bw_diff = 4000 / 8; 193 | /* If we estimate we're policed, use lt_bw for this many round trips: */ 194 | static const u32 bbr_lt_bw_max_rtts = 48; 195 | 196 | /* Gain factor for adding extra_acked to target cwnd: */ 197 | static const int bbr_extra_acked_gain = BBR_UNIT; 198 | /* Window length of extra_acked window. Max allowed val is 31. */ 199 | static const u32 bbr_extra_acked_win_rtts = 10; 200 | /* Max allowed val for ack_epoch_acked, after which sampling epoch is reset */ 201 | static const u32 bbr_ack_epoch_acked_reset_thresh = 1U << 20; 202 | /* Time period for clamping cwnd increment due to ack aggregation */ 203 | static const u32 bbr_extra_acked_max_us = 100 * 1000; 204 | 205 | /* Each cycle, try to hold sub-unity gain until inflight <= BDP. */ 206 | static const bool bbr_drain_to_target = true; /* default: enabled */ 207 | 208 | extern bool tcp_snd_wnd_test(const struct tcp_sock *tp, 209 | const struct sk_buff *skb, 210 | unsigned int cur_mss); 211 | 212 | /* Do we estimate that STARTUP filled the pipe? */ 213 | static bool bbr_full_bw_reached(const struct sock *sk) 214 | { 215 | const struct bbr *bbr = inet_csk_ca(sk); 216 | 217 | return bbr->full_bw_cnt >= bbr_full_bw_cnt; 218 | } 219 | 220 | static void bbr_set_cycle_idx(struct sock *sk, int cycle_idx) 221 | { 222 | struct bbr *bbr = inet_csk_ca(sk); 223 | bbr->cycle_idx = cycle_idx; 224 | bbr->pacing_gain = bbr->lt_use_bw ? 225 | BBR_UNIT : bbr_pacing_gain[bbr->cycle_idx]; 226 | } 227 | 228 | u32 bbr_max_bw(const struct sock *sk); 229 | u32 bbr_inflight(struct sock *sk, u32 bw, int gain); 230 | u32 bbr_max_bw(const struct sock *sk); 231 | 232 | static void bbr_drain_to_target_cycling(struct sock *sk, 233 | const struct rate_sample *rs) 234 | { 235 | struct tcp_sock *tp = tcp_sk(sk); 236 | struct bbr *bbr = inet_csk_ca(sk); 237 | u32 elapsed_us = 238 | tcp_stamp_us_delta(tp->delivered_mstamp, bbr->cycle_mstamp); 239 | u32 inflight, bw; 240 | if (bbr->mode != BBR_PROBE_BW) 241 | return; 242 | 243 | /* Always need to probe for bw before we forget good bw estimate. */ 244 | if (elapsed_us > bbr->cycle_len * bbr->min_rtt_us) { 245 | /* Start a new PROBE_BW probing cycle of [2 to 8] x min_rtt. */ 246 | bbr->cycle_mstamp = tp->delivered_mstamp; 247 | bbr->cycle_len = CYCLE_LEN - prandom_u32_max(bbr_cycle_rand); 248 | bbr_set_cycle_idx(sk, BBR_BW_PROBE_UP); /* probe bandwidth */ 249 | return; 250 | } 251 | /* The pacing_gain of 1.0 paces at the estimated bw to try to fully 252 | * use the pipe without increasing the queue. 253 | */ 254 | if (bbr->pacing_gain == BBR_UNIT) 255 | return; 256 | inflight = rs->prior_in_flight; /* what was in-flight before ACK? */ 257 | bw = bbr_max_bw(sk); 258 | /* A pacing_gain < 1.0 tries to drain extra queue we added if bw 259 | * probing didn't find more bw. If inflight falls to match BDP then we 260 | * estimate queue is drained; persisting would underutilize the pipe. 261 | */ 262 | if (bbr->pacing_gain < BBR_UNIT) { 263 | if (inflight <= bbr_inflight(sk, bw, BBR_UNIT)) 264 | bbr_set_cycle_idx(sk, BBR_BW_PROBE_CRUISE); /* cruise */ 265 | return; 266 | } 267 | /* A pacing_gain > 1.0 probes for bw by trying to raise inflight to at 268 | * least pacing_gain*BDP; this may take more than min_rtt if min_rtt is 269 | * small (e.g. on a LAN). We do not persist if packets are lost, since 270 | * a path with small buffers may not hold that much. Similarly we exit 271 | * if we were prevented by app/recv-win from reaching the target. 272 | */ 273 | if (elapsed_us > bbr->min_rtt_us && 274 | (inflight >= bbr_inflight(sk, bw, bbr->pacing_gain) || 275 | rs->losses || /* perhaps pacing_gain*BDP won't fit */ 276 | rs->is_app_limited || /* previously app-limited */ 277 | !tcp_send_head(sk) || /* currently app/rwin-limited */ 278 | !tcp_snd_wnd_test(tp, tcp_send_head(sk), tp->mss_cache))) { 279 | bbr_set_cycle_idx(sk, BBR_BW_PROBE_DOWN); /* drain queue */ 280 | return; 281 | } 282 | } 283 | 284 | 285 | /* Return maximum extra acked in past k-2k round trips, 286 | * where k = bbr_extra_acked_win_rtts. 287 | */ 288 | static u16 bbr_extra_acked(const struct sock *sk) 289 | { 290 | struct bbr *bbr = inet_csk_ca(sk); 291 | return max(bbr->extra_acked[0], bbr->extra_acked[1]); 292 | } 293 | 294 | 295 | /* Return the windowed max recent bandwidth sample, in pkts/uS << BW_SCALE. */ 296 | u32 bbr_max_bw(const struct sock *sk) 297 | { 298 | struct bbr *bbr = inet_csk_ca(sk); 299 | 300 | return minmax_get(&bbr->bw); 301 | } 302 | 303 | /* Return the estimated bandwidth of the path, in pkts/uS << BW_SCALE. */ 304 | static u32 bbr_bw(const struct sock *sk) 305 | { 306 | struct bbr *bbr = inet_csk_ca(sk); 307 | 308 | return bbr->lt_use_bw ? bbr->lt_bw : bbr_max_bw(sk); 309 | } 310 | 311 | /* Return rate in bytes per second, optionally with a gain. 312 | * The order here is chosen carefully to avoid overflow of u64. This should 313 | * work for input rates of up to 2.9Tbit/sec and gain of 2.89x. 314 | */ 315 | static u64 bbr_rate_bytes_per_sec(struct sock *sk, u64 rate, int gain) 316 | { 317 | rate *= tcp_mss_to_mtu(sk, tcp_sk(sk)->mss_cache); 318 | rate *= gain; 319 | rate >>= BBR_SCALE; 320 | rate *= USEC_PER_SEC; 321 | return rate >> BW_SCALE; 322 | } 323 | 324 | /* Convert a BBR bw and gain factor to a pacing rate in bytes per second. */ 325 | static u32 bbr_bw_to_pacing_rate(struct sock *sk, u32 bw, int gain) 326 | { 327 | u64 rate = bw; 328 | 329 | rate = bbr_rate_bytes_per_sec(sk, rate, gain); 330 | rate = min_t(u64, rate, sk->sk_max_pacing_rate); 331 | return rate; 332 | } 333 | 334 | /* Initialize pacing rate to: high_gain * init_cwnd / RTT. */ 335 | static void bbr_init_pacing_rate_from_rtt(struct sock *sk) 336 | { 337 | struct tcp_sock *tp = tcp_sk(sk); 338 | struct bbr *bbr = inet_csk_ca(sk); 339 | u64 bw; 340 | u32 rtt_us; 341 | 342 | if (tp->srtt_us) { /* any RTT sample yet? */ 343 | rtt_us = max(tp->srtt_us >> 3, 1U); 344 | bbr->has_seen_rtt = 1; 345 | } else { /* no RTT sample yet */ 346 | rtt_us = USEC_PER_MSEC; /* use nominal default RTT */ 347 | } 348 | bw = (u64)tp->snd_cwnd * BW_UNIT; 349 | do_div(bw, rtt_us); 350 | sk->sk_pacing_rate = bbr_bw_to_pacing_rate(sk, bw, bbr_high_gain); 351 | } 352 | 353 | /* Pace using current bw estimate and a gain factor. In order to help drive the 354 | * network toward lower queues while maintaining high utilization and low 355 | * latency, the average pacing rate aims to be slightly (~1%) lower than the 356 | * estimated bandwidth. This is an important aspect of the design. In this 357 | * implementation this slightly lower pacing rate is achieved implicitly by not 358 | * including link-layer headers in the packet size used for the pacing rate. 359 | */ 360 | static void bbr_set_pacing_rate(struct sock *sk, u32 bw, int gain) 361 | { 362 | struct tcp_sock *tp = tcp_sk(sk); 363 | struct bbr *bbr = inet_csk_ca(sk); 364 | u32 rate = bbr_bw_to_pacing_rate(sk, bw, gain); 365 | 366 | if (unlikely(!bbr->has_seen_rtt && tp->srtt_us)) 367 | bbr_init_pacing_rate_from_rtt(sk); 368 | if (bbr_full_bw_reached(sk) || rate > sk->sk_pacing_rate) 369 | sk->sk_pacing_rate = rate; 370 | } 371 | 372 | /* Return count of segments we want in the skbs we send, or 0 for default. */ 373 | static u32 bbr_tso_segs_goal(struct sock *sk) 374 | { 375 | struct bbr *bbr = inet_csk_ca(sk); 376 | 377 | return bbr->tso_segs_goal; 378 | } 379 | 380 | static void bbr_set_tso_segs_goal(struct sock *sk) 381 | { 382 | struct tcp_sock *tp = tcp_sk(sk); 383 | struct bbr *bbr = inet_csk_ca(sk); 384 | u32 min_segs; 385 | 386 | min_segs = sk->sk_pacing_rate < (bbr_min_tso_rate >> 3) ? 1 : 2; 387 | bbr->tso_segs_goal = min(tcp_tso_autosize(sk, tp->mss_cache, min_segs), 388 | 0x7FU); 389 | } 390 | 391 | /* Save "last known good" cwnd so we can restore it after losses or PROBE_RTT */ 392 | static void bbr_save_cwnd(struct sock *sk) 393 | { 394 | struct tcp_sock *tp = tcp_sk(sk); 395 | struct bbr *bbr = inet_csk_ca(sk); 396 | 397 | if (bbr->prev_ca_state < TCP_CA_Recovery && bbr->mode != BBR_PROBE_RTT) 398 | bbr->prior_cwnd = tp->snd_cwnd; /* this cwnd is good enough */ 399 | else /* loss recovery or BBR_PROBE_RTT have temporarily cut cwnd */ 400 | bbr->prior_cwnd = max(bbr->prior_cwnd, tp->snd_cwnd); 401 | } 402 | 403 | static void bbr_cwnd_event(struct sock *sk, enum tcp_ca_event event) 404 | { 405 | struct tcp_sock *tp = tcp_sk(sk); 406 | struct bbr *bbr = inet_csk_ca(sk); 407 | 408 | if (event == CA_EVENT_TX_START && tp->app_limited) { 409 | bbr->idle_restart = 1; 410 | bbr->ack_epoch_mstamp = tp->tcp_mstamp; 411 | bbr->ack_epoch_acked = 0; 412 | 413 | /* Avoid pointless buffer overflows: pace at est. bw if we don't 414 | * need more speed (we're restarting from idle and app-limited). 415 | */ 416 | if (bbr->mode == BBR_PROBE_BW) 417 | bbr_set_pacing_rate(sk, bbr_bw(sk), BBR_UNIT); 418 | } 419 | } 420 | 421 | /* Find target cwnd. Right-size the cwnd based on min RTT and the 422 | * estimated bottleneck bandwidth: 423 | * 424 | * cwnd = bw * min_rtt * gain = BDP * gain 425 | * 426 | * The key factor, gain, controls the amount of queue. While a small gain 427 | * builds a smaller queue, it becomes more vulnerable to noise in RTT 428 | * measurements (e.g., delayed ACKs or other ACK compression effects). This 429 | * noise may cause BBR to under-estimate the rate. 430 | * 431 | * To achieve full performance in high-speed paths, we budget enough cwnd to 432 | * fit full-sized skbs in-flight on both end hosts to fully utilize the path: 433 | * - one skb in sending host Qdisc, 434 | * - one skb in sending host TSO/GSO engine 435 | * - one skb being received by receiver host LRO/GRO/delayed-ACK engine 436 | * Don't worry, at low rates (bbr_min_tso_rate) this won't bloat cwnd because 437 | * in such cases tso_segs_goal is 1. The minimum cwnd is 4 packets, 438 | * which allows 2 outstanding 2-packet sequences, to try to keep pipe 439 | * full even with ACK-every-other-packet delayed ACKs. 440 | */ 441 | static u32 bbr_bdp(struct sock *sk, u32 bw, int gain) 442 | { 443 | struct bbr *bbr = inet_csk_ca(sk); 444 | u32 bdp; 445 | u64 w; 446 | 447 | /* If we've never had a valid RTT sample, cap cwnd at the initial 448 | * default. This should only happen when the connection is not using TCP 449 | * timestamps and has retransmitted all of the SYN/SYNACK/data packets 450 | * ACKed so far. In this case, an RTO can cut cwnd to 1, in which 451 | * case we need to slow-start up toward something safe: TCP_INIT_CWND. 452 | */ 453 | if (unlikely(bbr->min_rtt_us == ~0U)) /* no valid RTT samples yet? */ 454 | return TCP_INIT_CWND; /* be safe: cap at default initial cwnd*/ 455 | 456 | w = (u64)bw * bbr->min_rtt_us; 457 | 458 | /* Apply a gain to the given value, then remove the BW_SCALE shift. */ 459 | bdp = (((w * gain) >> BBR_SCALE) + BW_UNIT - 1) / BW_UNIT; 460 | 461 | return bdp; 462 | } 463 | 464 | static u32 bbr_quantization_budget(struct sock *sk, u32 cwnd, int gain) 465 | { 466 | 467 | /* Allow enough full-sized skbs in flight to utilize end systems. */ 468 | cwnd += 3 * bbr_tso_segs_goal(sk); 469 | 470 | return cwnd; 471 | } 472 | 473 | /* Find inflight based on min RTT and the estimated bottleneck bandwidth. */ 474 | u32 bbr_inflight(struct sock *sk, u32 bw, int gain) 475 | { 476 | u32 inflight; 477 | inflight = bbr_bdp(sk, bw, gain); 478 | inflight = bbr_quantization_budget(sk, inflight, gain); 479 | return inflight; 480 | 481 | } 482 | 483 | /* Find the cwnd increment based on estimate of ack aggregation */ 484 | static u32 bbr_ack_aggregation_cwnd(struct sock *sk) 485 | { 486 | u32 max_aggr_cwnd, aggr_cwnd = 0; 487 | if (bbr_extra_acked_gain && bbr_full_bw_reached(sk)) { 488 | max_aggr_cwnd = ((u64)bbr_bw(sk) * bbr_extra_acked_max_us) 489 | / BW_UNIT; 490 | aggr_cwnd = (bbr_extra_acked_gain * bbr_extra_acked(sk)) 491 | >> BBR_SCALE; 492 | aggr_cwnd = min(aggr_cwnd, max_aggr_cwnd); 493 | } 494 | return aggr_cwnd; 495 | } 496 | 497 | 498 | /* An optimization in BBR to reduce losses: On the first round of recovery, we 499 | * follow the packet conservation principle: send P packets per P packets acked. 500 | * After that, we slow-start and send at most 2*P packets per P packets acked. 501 | * After recovery finishes, or upon undo, we restore the cwnd we had when 502 | * recovery started (capped by the target cwnd based on estimated BDP). 503 | * 504 | * TODO(ycheng/ncardwell): implement a rate-based approach. 505 | */ 506 | static bool bbr_set_cwnd_to_recover_or_restore( 507 | struct sock *sk, const struct rate_sample *rs, u32 acked, u32 *new_cwnd) 508 | { 509 | struct tcp_sock *tp = tcp_sk(sk); 510 | struct bbr *bbr = inet_csk_ca(sk); 511 | u8 prev_state = bbr->prev_ca_state, state = inet_csk(sk)->icsk_ca_state; 512 | u32 cwnd = tp->snd_cwnd; 513 | 514 | /* An ACK for P pkts should release at most 2*P packets. We do this 515 | * in two steps. First, here we deduct the number of lost packets. 516 | * Then, in bbr_set_cwnd() we slow start up toward the target cwnd. 517 | */ 518 | if (rs->losses > 0) 519 | cwnd = max_t(s32, cwnd - rs->losses, 1); 520 | 521 | if (state == TCP_CA_Recovery && prev_state != TCP_CA_Recovery) { 522 | /* Starting 1st round of Recovery, so do packet conservation. */ 523 | bbr->packet_conservation = 1; 524 | bbr->next_rtt_delivered = tp->delivered; /* start round now */ 525 | /* Cut unused cwnd from app behavior, TSQ, or TSO deferral: */ 526 | cwnd = tcp_packets_in_flight(tp) + acked; 527 | } else if (prev_state >= TCP_CA_Recovery && state < TCP_CA_Recovery) { 528 | /* Exiting loss recovery; restore cwnd saved before recovery. */ 529 | bbr->restore_cwnd = 1; 530 | bbr->packet_conservation = 0; 531 | } 532 | bbr->prev_ca_state = state; 533 | 534 | if (bbr->restore_cwnd) { 535 | /* Restore cwnd after exiting loss recovery or PROBE_RTT. */ 536 | cwnd = max(cwnd, bbr->prior_cwnd); 537 | bbr->restore_cwnd = 0; 538 | } 539 | 540 | if (bbr->packet_conservation) { 541 | *new_cwnd = max(cwnd, tcp_packets_in_flight(tp) + acked); 542 | return true; /* yes, using packet conservation */ 543 | } 544 | *new_cwnd = cwnd; 545 | return false; 546 | } 547 | 548 | /* Slow-start up toward target cwnd (if bw estimate is growing, or packet loss 549 | * has drawn us down below target), or snap down to target if we're above it. 550 | */ 551 | static void bbr_set_cwnd(struct sock *sk, const struct rate_sample *rs, 552 | u32 acked, u32 bw, int gain) 553 | { 554 | struct tcp_sock *tp = tcp_sk(sk); 555 | struct bbr *bbr = inet_csk_ca(sk); 556 | u32 cwnd = 0, target_cwnd = 0; 557 | 558 | if (!acked) 559 | return; 560 | 561 | if (bbr_set_cwnd_to_recover_or_restore(sk, rs, acked, &cwnd)) 562 | goto done; 563 | 564 | /* If we're below target cwnd, slow start cwnd toward target cwnd. */ 565 | target_cwnd = bbr_bdp(sk, bw, gain); 566 | //// 567 | /* Increment the cwnd to account for excess ACKed data that seems 568 | * due to aggregation (of data and/or ACKs) visible in the ACK stream. 569 | */ 570 | target_cwnd += bbr_ack_aggregation_cwnd(sk); 571 | //// 572 | target_cwnd = bbr_quantization_budget(sk, target_cwnd, gain); 573 | if (bbr_full_bw_reached(sk)) /* only cut cwnd if we filled the pipe */ 574 | cwnd = min(cwnd + acked, target_cwnd); 575 | else if (cwnd < target_cwnd || tp->delivered < TCP_INIT_CWND) 576 | cwnd = cwnd + acked; 577 | cwnd = max(cwnd, bbr_cwnd_min_target); 578 | 579 | done: 580 | tp->snd_cwnd = min(cwnd, tp->snd_cwnd_clamp); /* apply global cap */ 581 | if (bbr->mode == BBR_PROBE_RTT) /* drain queue, refresh min_rtt */ 582 | tp->snd_cwnd = min(tp->snd_cwnd, bbr_cwnd_min_target); 583 | } 584 | 585 | /* End cycle phase if it's time and/or we hit the phase's in-flight target. */ 586 | static bool bbr_is_next_cycle_phase(struct sock *sk, 587 | const struct rate_sample *rs) 588 | { 589 | struct tcp_sock *tp = tcp_sk(sk); 590 | struct bbr *bbr = inet_csk_ca(sk); 591 | bool is_full_length = 592 | tcp_stamp_us_delta(tp->delivered_mstamp, bbr->cycle_mstamp) > 593 | bbr->min_rtt_us; 594 | u32 inflight, bw; 595 | 596 | /* The pacing_gain of 1.0 paces at the estimated bw to try to fully 597 | * use the pipe without increasing the queue. 598 | */ 599 | if (bbr->pacing_gain == BBR_UNIT) 600 | return is_full_length; /* just use wall clock time */ 601 | 602 | inflight = rs->prior_in_flight; /* what was in-flight before ACK? */ 603 | bw = bbr_max_bw(sk); 604 | 605 | /* A pacing_gain > 1.0 probes for bw by trying to raise inflight to at 606 | * least pacing_gain*BDP; this may take more than min_rtt if min_rtt is 607 | * small (e.g. on a LAN). We do not persist if packets are lost, since 608 | * a path with small buffers may not hold that much. 609 | */ 610 | if (bbr->pacing_gain > BBR_UNIT) 611 | return is_full_length && 612 | (rs->losses || /* perhaps pacing_gain*BDP won't fit */ 613 | inflight >= bbr_inflight(sk, bw, bbr->pacing_gain)); 614 | 615 | /* A pacing_gain < 1.0 tries to drain extra queue we added if bw 616 | * probing didn't find more bw. If inflight falls to match BDP then we 617 | * estimate queue is drained; persisting would underutilize the pipe. 618 | */ 619 | return is_full_length || 620 | inflight <= bbr_inflight(sk, bw, BBR_UNIT); 621 | } 622 | 623 | static void bbr_advance_cycle_phase(struct sock *sk) 624 | { 625 | struct tcp_sock *tp = tcp_sk(sk); 626 | struct bbr *bbr = inet_csk_ca(sk); 627 | 628 | 629 | bbr->cycle_idx = (bbr->cycle_idx + 1) & (CYCLE_LEN - 1); 630 | bbr->cycle_mstamp = tp->delivered_mstamp; 631 | bbr->pacing_gain = bbr_pacing_gain[bbr->cycle_idx]; 632 | } 633 | 634 | /* Gain cycling: cycle pacing gain to converge to fair share of available bw. */ 635 | static void bbr_update_cycle_phase(struct sock *sk, 636 | const struct rate_sample *rs) 637 | { 638 | struct bbr *bbr = inet_csk_ca(sk); 639 | 640 | if (bbr_drain_to_target) { 641 | bbr_drain_to_target_cycling(sk, rs); 642 | return; 643 | } 644 | 645 | if ((bbr->mode == BBR_PROBE_BW) && !bbr->lt_use_bw && 646 | bbr_is_next_cycle_phase(sk, rs)) 647 | bbr_advance_cycle_phase(sk); 648 | } 649 | 650 | static void bbr_reset_startup_mode(struct sock *sk) 651 | { 652 | struct bbr *bbr = inet_csk_ca(sk); 653 | 654 | bbr->mode = BBR_STARTUP; 655 | bbr->pacing_gain = bbr_high_gain; 656 | bbr->cwnd_gain = bbr_high_gain; 657 | } 658 | 659 | static void bbr_reset_probe_bw_mode(struct sock *sk) 660 | { 661 | struct bbr *bbr = inet_csk_ca(sk); 662 | 663 | bbr->mode = BBR_PROBE_BW; 664 | bbr->pacing_gain = BBR_UNIT; 665 | bbr->cwnd_gain = bbr_cwnd_gain; 666 | bbr->cycle_idx = CYCLE_LEN - 1 - prandom_u32_max(bbr_cycle_rand); 667 | bbr_advance_cycle_phase(sk); /* flip to next phase of gain cycle */ 668 | } 669 | 670 | static void bbr_reset_mode(struct sock *sk) 671 | { 672 | if (!bbr_full_bw_reached(sk)) 673 | bbr_reset_startup_mode(sk); 674 | else 675 | bbr_reset_probe_bw_mode(sk); 676 | } 677 | 678 | /* Start a new long-term sampling interval. */ 679 | static void bbr_reset_lt_bw_sampling_interval(struct sock *sk) 680 | { 681 | struct tcp_sock *tp = tcp_sk(sk); 682 | struct bbr *bbr = inet_csk_ca(sk); 683 | 684 | bbr->lt_last_stamp = div_u64(tp->delivered_mstamp, USEC_PER_MSEC); 685 | bbr->lt_last_delivered = tp->delivered; 686 | bbr->lt_last_lost = tp->lost; 687 | bbr->lt_rtt_cnt = 0; 688 | } 689 | 690 | /* Completely reset long-term bandwidth sampling. */ 691 | static void bbr_reset_lt_bw_sampling(struct sock *sk) 692 | { 693 | struct bbr *bbr = inet_csk_ca(sk); 694 | 695 | bbr->lt_bw = 0; 696 | bbr->lt_use_bw = 0; 697 | bbr->lt_is_sampling = false; 698 | bbr_reset_lt_bw_sampling_interval(sk); 699 | } 700 | 701 | /* Long-term bw sampling interval is done. Estimate whether we're policed. */ 702 | static void bbr_lt_bw_interval_done(struct sock *sk, u32 bw) 703 | { 704 | struct bbr *bbr = inet_csk_ca(sk); 705 | u32 diff; 706 | 707 | if (bbr->lt_bw) { /* do we have bw from a previous interval? */ 708 | /* Is new bw close to the lt_bw from the previous interval? */ 709 | diff = abs(bw - bbr->lt_bw); 710 | if ((diff * BBR_UNIT <= bbr_lt_bw_ratio * bbr->lt_bw) || 711 | (bbr_rate_bytes_per_sec(sk, diff, BBR_UNIT) <= 712 | bbr_lt_bw_diff)) { 713 | /* All criteria are met; estimate we're policed. */ 714 | bbr->lt_bw = (bw + bbr->lt_bw) >> 1; /* avg 2 intvls */ 715 | bbr->lt_use_bw = 1; 716 | bbr->pacing_gain = BBR_UNIT; /* try to avoid drops */ 717 | bbr->lt_rtt_cnt = 0; 718 | return; 719 | } 720 | } 721 | bbr->lt_bw = bw; 722 | bbr_reset_lt_bw_sampling_interval(sk); 723 | } 724 | 725 | /* Token-bucket traffic policers are common (see "An Internet-Wide Analysis of 726 | * Traffic Policing", SIGCOMM 2016). BBR detects token-bucket policers and 727 | * explicitly models their policed rate, to reduce unnecessary losses. We 728 | * estimate that we're policed if we see 2 consecutive sampling intervals with 729 | * consistent throughput and high packet loss. If we think we're being policed, 730 | * set lt_bw to the "long-term" average delivery rate from those 2 intervals. 731 | */ 732 | static void bbr_lt_bw_sampling(struct sock *sk, const struct rate_sample *rs) 733 | { 734 | struct tcp_sock *tp = tcp_sk(sk); 735 | struct bbr *bbr = inet_csk_ca(sk); 736 | u32 lost, delivered; 737 | u64 bw; 738 | u32 t; 739 | 740 | if (bbr->lt_use_bw) { /* already using long-term rate, lt_bw? */ 741 | if (bbr->mode == BBR_PROBE_BW && bbr->round_start && 742 | ++bbr->lt_rtt_cnt >= bbr_lt_bw_max_rtts) { 743 | bbr_reset_lt_bw_sampling(sk); /* stop using lt_bw */ 744 | bbr_reset_probe_bw_mode(sk); /* restart gain cycling */ 745 | } 746 | return; 747 | } 748 | 749 | /* Wait for the first loss before sampling, to let the policer exhaust 750 | * its tokens and estimate the steady-state rate allowed by the policer. 751 | * Starting samples earlier includes bursts that over-estimate the bw. 752 | */ 753 | if (!bbr->lt_is_sampling) { 754 | if (!rs->losses) 755 | return; 756 | bbr_reset_lt_bw_sampling_interval(sk); 757 | bbr->lt_is_sampling = true; 758 | } 759 | 760 | /* To avoid underestimates, reset sampling if we run out of data. */ 761 | if (rs->is_app_limited) { 762 | bbr_reset_lt_bw_sampling(sk); 763 | return; 764 | } 765 | 766 | if (bbr->round_start) 767 | bbr->lt_rtt_cnt++; /* count round trips in this interval */ 768 | if (bbr->lt_rtt_cnt < bbr_lt_intvl_min_rtts) 769 | return; /* sampling interval needs to be longer */ 770 | if (bbr->lt_rtt_cnt > 4 * bbr_lt_intvl_min_rtts) { 771 | bbr_reset_lt_bw_sampling(sk); /* interval is too long */ 772 | return; 773 | } 774 | 775 | /* End sampling interval when a packet is lost, so we estimate the 776 | * policer tokens were exhausted. Stopping the sampling before the 777 | * tokens are exhausted under-estimates the policed rate. 778 | */ 779 | if (!rs->losses) 780 | return; 781 | 782 | /* Calculate packets lost and delivered in sampling interval. */ 783 | lost = tp->lost - bbr->lt_last_lost; 784 | delivered = tp->delivered - bbr->lt_last_delivered; 785 | /* Is loss rate (lost/delivered) >= lt_loss_thresh? If not, wait. */ 786 | if (!delivered || (lost << BBR_SCALE) < bbr_lt_loss_thresh * delivered) 787 | return; 788 | 789 | /* Find average delivery rate in this sampling interval. */ 790 | t = div_u64(tp->delivered_mstamp, USEC_PER_MSEC) - bbr->lt_last_stamp; 791 | if ((s32)t < 1) 792 | return; /* interval is less than one ms, so wait */ 793 | /* Check if can multiply without overflow */ 794 | if (t >= ~0U / USEC_PER_MSEC) { 795 | bbr_reset_lt_bw_sampling(sk); /* interval too long; reset */ 796 | return; 797 | } 798 | t *= USEC_PER_MSEC; 799 | bw = (u64)delivered * BW_UNIT; 800 | do_div(bw, t); 801 | bbr_lt_bw_interval_done(sk, bw); 802 | } 803 | 804 | /* Estimate the bandwidth based on how fast packets are delivered */ 805 | static void bbr_update_bw(struct sock *sk, const struct rate_sample *rs) 806 | { 807 | struct tcp_sock *tp = tcp_sk(sk); 808 | struct bbr *bbr = inet_csk_ca(sk); 809 | u64 bw; 810 | 811 | bbr->round_start = 0; 812 | if (rs->delivered < 0 || rs->interval_us <= 0) 813 | return; /* Not a valid observation */ 814 | 815 | /* See if we've reached the next RTT */ 816 | if (!before(rs->prior_delivered, bbr->next_rtt_delivered)) { 817 | bbr->next_rtt_delivered = tp->delivered; 818 | bbr->rtt_cnt++; 819 | bbr->round_start = 1; 820 | bbr->packet_conservation = 0; 821 | } 822 | 823 | bbr_lt_bw_sampling(sk, rs); 824 | 825 | /* Divide delivered by the interval to find a (lower bound) bottleneck 826 | * bandwidth sample. Delivered is in packets and interval_us in uS and 827 | * ratio will be <<1 for most connections. So delivered is first scaled. 828 | */ 829 | bw = (u64)rs->delivered * BW_UNIT; 830 | do_div(bw, rs->interval_us); 831 | 832 | /* If this sample is application-limited, it is likely to have a very 833 | * low delivered count that represents application behavior rather than 834 | * the available network rate. Such a sample could drag down estimated 835 | * bw, causing needless slow-down. Thus, to continue to send at the 836 | * last measured network rate, we filter out app-limited samples unless 837 | * they describe the path bw at least as well as our bw model. 838 | * 839 | * So the goal during app-limited phase is to proceed with the best 840 | * network rate no matter how long. We automatically leave this 841 | * phase when app writes faster than the network can deliver :) 842 | */ 843 | if (!rs->is_app_limited || bw >= bbr_max_bw(sk)) { 844 | /* Incorporate new sample into our max bw filter. */ 845 | minmax_running_max(&bbr->bw, bbr_bw_rtts, bbr->rtt_cnt, bw); 846 | } 847 | } 848 | 849 | /* Estimate when the pipe is full, using the change in delivery rate: BBR 850 | * estimates that STARTUP filled the pipe if the estimated bw hasn't changed by 851 | * at least bbr_full_bw_thresh (25%) after bbr_full_bw_cnt (3) non-app-limited 852 | * rounds. Why 3 rounds: 1: rwin autotuning grows the rwin, 2: we fill the 853 | * higher rwin, 3: we get higher delivery rate samples. Or transient 854 | * cross-traffic or radio noise can go away. CUBIC Hystart shares a similar 855 | * design goal, but uses delay and inter-ACK spacing instead of bandwidth. 856 | */ 857 | static void bbr_check_full_bw_reached(struct sock *sk, 858 | const struct rate_sample *rs) 859 | { 860 | struct bbr *bbr = inet_csk_ca(sk); 861 | u32 bw_thresh; 862 | 863 | if (bbr_full_bw_reached(sk) || !bbr->round_start || rs->is_app_limited) 864 | return; 865 | 866 | bw_thresh = (u64)bbr->full_bw * bbr_full_bw_thresh >> BBR_SCALE; 867 | if (bbr_max_bw(sk) >= bw_thresh) { 868 | bbr->full_bw = bbr_max_bw(sk); 869 | bbr->full_bw_cnt = 0; 870 | return; 871 | } 872 | ++bbr->full_bw_cnt; 873 | } 874 | 875 | /* If pipe is probably full, drain the queue and then enter steady-state. */ 876 | static void bbr_check_drain(struct sock *sk, const struct rate_sample *rs) 877 | { 878 | struct bbr *bbr = inet_csk_ca(sk); 879 | 880 | if (bbr->mode == BBR_STARTUP && bbr_full_bw_reached(sk)) { 881 | bbr->mode = BBR_DRAIN; /* drain queue we created */ 882 | bbr->pacing_gain = bbr_drain_gain; /* pace slow to drain */ 883 | bbr->cwnd_gain = bbr_high_gain; /* maintain cwnd */ 884 | } /* fall through to check if in-flight is already small: */ 885 | if (bbr->mode == BBR_DRAIN && 886 | tcp_packets_in_flight(tcp_sk(sk)) <= bbr_inflight(sk, bbr_max_bw(sk), BBR_UNIT)) 887 | bbr_reset_probe_bw_mode(sk); /* we estimate queue is drained */ 888 | } 889 | 890 | 891 | /* Estimates the windowed max degree of ack aggregation. 892 | * This is used to provision extra in-flight data to keep sending during 893 | * inter-ACK silences. 894 | * 895 | * Degree of ack aggregation is estimated as extra data acked beyond expected. 896 | * 897 | * max_extra_acked = "maximum recent excess data ACKed beyond max_bw * interval" 898 | * cwnd += max_extra_acked 899 | * 900 | * Max extra_acked is clamped by cwnd and bw * bbr_extra_acked_max_us (100 ms). 901 | * Max filter is an approximate sliding window of 10-20 (packet timed) round 902 | * trips. 903 | */ 904 | static void bbr_update_ack_aggregation(struct sock *sk, 905 | const struct rate_sample *rs) 906 | { 907 | u32 epoch_us, expected_acked, extra_acked; 908 | struct bbr *bbr = inet_csk_ca(sk); 909 | struct tcp_sock *tp = tcp_sk(sk); 910 | if (!bbr_extra_acked_gain || rs->acked_sacked <= 0 || 911 | rs->delivered < 0 || rs->interval_us <= 0) 912 | return; 913 | if (bbr->round_start) { 914 | bbr->extra_acked_win_rtts = min(0x1F, 915 | bbr->extra_acked_win_rtts + 1); 916 | if (bbr->extra_acked_win_rtts >= bbr_extra_acked_win_rtts) { 917 | bbr->extra_acked_win_rtts = 0; 918 | bbr->extra_acked_win_idx = bbr->extra_acked_win_idx ?0 : 1; 919 | bbr->extra_acked[bbr->extra_acked_win_idx] = 0; 920 | } 921 | } 922 | /* Compute how many packets we expected to be delivered over epoch. */ 923 | epoch_us = tcp_stamp_us_delta(tp->delivered_mstamp, 924 | bbr->ack_epoch_mstamp); 925 | expected_acked = ((u64)bbr_bw(sk) * epoch_us) / BW_UNIT; 926 | /* Reset the aggregation epoch if ACK rate is below expected rate or 927 | * significantly large no. of ack received since epoch (potentially 928 | * quite old epoch). 929 | */ 930 | if (bbr->ack_epoch_acked <= expected_acked || 931 | (bbr->ack_epoch_acked + rs->acked_sacked >= 932 | bbr_ack_epoch_acked_reset_thresh)) { 933 | bbr->ack_epoch_acked = 0; 934 | bbr->ack_epoch_mstamp = tp->delivered_mstamp; 935 | expected_acked = 0; 936 | } 937 | /* Compute excess data delivered, beyond what was expected. */ 938 | bbr->ack_epoch_acked = min(0xFFFFFU, 939 | bbr->ack_epoch_acked + rs->acked_sacked); 940 | extra_acked = bbr->ack_epoch_acked - expected_acked; 941 | extra_acked = min(extra_acked, tp->snd_cwnd); 942 | if (extra_acked > bbr->extra_acked[bbr->extra_acked_win_idx]) 943 | bbr->extra_acked[bbr->extra_acked_win_idx] = extra_acked; 944 | } 945 | 946 | /* The goal of PROBE_RTT mode is to have BBR flows cooperatively and 947 | * periodically drain the bottleneck queue, to converge to measure the true 948 | * min_rtt (unloaded propagation delay). This allows the flows to keep queues 949 | * small (reducing queuing delay and packet loss) and achieve fairness among 950 | * BBR flows. 951 | * 952 | * The min_rtt filter window is 10 seconds. When the min_rtt estimate expires, 953 | * we enter PROBE_RTT mode and cap the cwnd at bbr_cwnd_min_target=4 packets. 954 | * After at least bbr_probe_rtt_mode_ms=200ms and at least one packet-timed 955 | * round trip elapsed with that flight size <= 4, we leave PROBE_RTT mode and 956 | * re-enter the previous mode. BBR uses 200ms to approximately bound the 957 | * performance penalty of PROBE_RTT's cwnd capping to roughly 2% (200ms/10s). 958 | * 959 | * Note that flows need only pay 2% if they are busy sending over the last 10 960 | * seconds. Interactive applications (e.g., Web, RPCs, video chunks) often have 961 | * natural silences or low-rate periods within 10 seconds where the rate is low 962 | * enough for long enough to drain its queue in the bottleneck. We pick up 963 | * these min RTT measurements opportunistically with our min_rtt filter. :-) 964 | */ 965 | 966 | static void bbr_update_min_rtt(struct sock *sk, const struct rate_sample *rs) 967 | { 968 | struct tcp_sock *tp = tcp_sk(sk); 969 | struct bbr *bbr = inet_csk_ca(sk); 970 | bool filter_expired; 971 | 972 | /* Track min RTT seen in the min_rtt_win_sec filter window: */ 973 | filter_expired = after(tcp_jiffies32, 974 | bbr->min_rtt_stamp + bbr_min_rtt_win_sec * HZ); 975 | if (rs->rtt_us >= 0 && 976 | (rs->rtt_us <= bbr->min_rtt_us || filter_expired)) { 977 | bbr->min_rtt_us = rs->rtt_us; 978 | bbr->min_rtt_stamp = tcp_jiffies32; 979 | } 980 | 981 | if (bbr_probe_rtt_mode_ms > 0 && filter_expired && 982 | !bbr->idle_restart && bbr->mode != BBR_PROBE_RTT) { 983 | bbr->mode = BBR_PROBE_RTT; /* dip, drain queue */ 984 | bbr->pacing_gain = BBR_UNIT; 985 | bbr->cwnd_gain = BBR_UNIT; 986 | bbr_save_cwnd(sk); /* note cwnd so we can restore it */ 987 | bbr->probe_rtt_done_stamp = 0; 988 | } 989 | 990 | if (bbr->mode == BBR_PROBE_RTT) { 991 | /* Ignore low rate samples during this mode. */ 992 | tp->app_limited = 993 | (tp->delivered + tcp_packets_in_flight(tp)) ? : 1; 994 | /* Maintain min packets in flight for max(200 ms, 1 round). */ 995 | if (!bbr->probe_rtt_done_stamp && 996 | tcp_packets_in_flight(tp) <= bbr_cwnd_min_target) { 997 | bbr->probe_rtt_done_stamp = tcp_jiffies32 + 998 | msecs_to_jiffies(bbr_probe_rtt_mode_ms); 999 | bbr->probe_rtt_round_done = 0; 1000 | bbr->next_rtt_delivered = tp->delivered; 1001 | } else if (bbr->probe_rtt_done_stamp) { 1002 | if (bbr->round_start) 1003 | bbr->probe_rtt_round_done = 1; 1004 | if (bbr->probe_rtt_round_done && 1005 | after(tcp_jiffies32, bbr->probe_rtt_done_stamp)) { 1006 | bbr->min_rtt_stamp = tcp_jiffies32; 1007 | bbr->restore_cwnd = 1; /* snap to prior_cwnd */ 1008 | bbr_reset_mode(sk); 1009 | } 1010 | } 1011 | } 1012 | bbr->idle_restart = 0; 1013 | } 1014 | 1015 | static void bbr_update_model(struct sock *sk, const struct rate_sample *rs) 1016 | { 1017 | bbr_update_bw(sk, rs); 1018 | bbr_update_ack_aggregation(sk, rs); 1019 | bbr_update_cycle_phase(sk, rs); 1020 | bbr_check_full_bw_reached(sk, rs); 1021 | bbr_check_drain(sk, rs); 1022 | bbr_update_min_rtt(sk, rs); 1023 | } 1024 | 1025 | static void bbr_main(struct sock *sk, const struct rate_sample *rs) 1026 | { 1027 | struct bbr *bbr = inet_csk_ca(sk); 1028 | u32 bw; 1029 | 1030 | bbr_update_model(sk, rs); 1031 | 1032 | bw = bbr_bw(sk); 1033 | bbr_set_pacing_rate(sk, bw, bbr->pacing_gain); 1034 | bbr_set_tso_segs_goal(sk); 1035 | bbr_set_cwnd(sk, rs, rs->acked_sacked, bw, bbr->cwnd_gain); 1036 | } 1037 | 1038 | static void bbr_init(struct sock *sk) 1039 | { 1040 | struct tcp_sock *tp = tcp_sk(sk); 1041 | struct bbr *bbr = inet_csk_ca(sk); 1042 | 1043 | bbr->prior_cwnd = 0; 1044 | bbr->tso_segs_goal = 0; /* default segs per skb until first ACK */ 1045 | bbr->rtt_cnt = 0; 1046 | bbr->next_rtt_delivered = 0; 1047 | bbr->prev_ca_state = TCP_CA_Open; 1048 | bbr->packet_conservation = 0; 1049 | 1050 | bbr->probe_rtt_done_stamp = 0; 1051 | bbr->probe_rtt_round_done = 0; 1052 | bbr->min_rtt_us = tcp_min_rtt(tp); 1053 | bbr->min_rtt_stamp = tcp_jiffies32; 1054 | 1055 | minmax_reset(&bbr->bw, bbr->rtt_cnt, 0); /* init max bw to 0 */ 1056 | 1057 | bbr->has_seen_rtt = 0; 1058 | bbr_init_pacing_rate_from_rtt(sk); 1059 | 1060 | bbr->restore_cwnd = 0; 1061 | bbr->round_start = 0; 1062 | bbr->idle_restart = 0; 1063 | bbr->full_bw = 0; 1064 | bbr->full_bw_cnt = 0; 1065 | bbr->cycle_mstamp = 0; 1066 | bbr->cycle_idx = 0; 1067 | bbr->cycle_len = 0; 1068 | bbr_reset_lt_bw_sampling(sk); 1069 | bbr_reset_startup_mode(sk); 1070 | bbr->ack_epoch_mstamp = tp->tcp_mstamp; 1071 | bbr->ack_epoch_acked = 0; 1072 | bbr->extra_acked_win_rtts = 0; 1073 | bbr->extra_acked_win_idx = 0; 1074 | bbr->extra_acked[0] = 0; 1075 | bbr->extra_acked[1] = 0; 1076 | 1077 | cmpxchg(&sk->sk_pacing_status, SK_PACING_NONE, SK_PACING_NEEDED); 1078 | } 1079 | 1080 | static u32 bbr_sndbuf_expand(struct sock *sk) 1081 | { 1082 | /* Provision 3 * cwnd since BBR may slow-start even during recovery. */ 1083 | return 3; 1084 | } 1085 | 1086 | /* In theory BBR does not need to undo the cwnd since it does not 1087 | * always reduce cwnd on losses (see bbr_main()). Keep it for now. 1088 | */ 1089 | static u32 bbr_undo_cwnd(struct sock *sk) 1090 | { 1091 | return tcp_sk(sk)->snd_cwnd; 1092 | } 1093 | 1094 | /* Entering loss recovery, so save cwnd for when we exit or undo recovery. */ 1095 | static u32 bbr_ssthresh(struct sock *sk) 1096 | { 1097 | bbr_save_cwnd(sk); 1098 | return TCP_INFINITE_SSTHRESH; /* BBR does not use ssthresh */ 1099 | } 1100 | 1101 | static size_t bbr_get_info(struct sock *sk, u32 ext, int *attr, 1102 | union tcp_cc_info *info) 1103 | { 1104 | if (ext & (1 << (INET_DIAG_BBRINFO - 1)) || 1105 | ext & (1 << (INET_DIAG_VEGASINFO - 1))) { 1106 | struct tcp_sock *tp = tcp_sk(sk); 1107 | struct bbr *bbr = inet_csk_ca(sk); 1108 | u64 bw = bbr_bw(sk); 1109 | 1110 | bw = bw * tp->mss_cache * USEC_PER_SEC >> BW_SCALE; 1111 | memset(&info->bbr, 0, sizeof(info->bbr)); 1112 | info->bbr.bbr_bw_lo = (u32)bw; 1113 | info->bbr.bbr_bw_hi = (u32)(bw >> 32); 1114 | info->bbr.bbr_min_rtt = bbr->min_rtt_us; 1115 | info->bbr.bbr_pacing_gain = bbr->pacing_gain; 1116 | info->bbr.bbr_cwnd_gain = bbr->cwnd_gain; 1117 | *attr = INET_DIAG_BBRINFO; 1118 | return sizeof(info->bbr); 1119 | } 1120 | return 0; 1121 | } 1122 | 1123 | static void bbr_set_state(struct sock *sk, u8 new_state) 1124 | { 1125 | struct bbr *bbr = inet_csk_ca(sk); 1126 | 1127 | if (new_state == TCP_CA_Loss) { 1128 | struct rate_sample rs = { .losses = 1 }; 1129 | 1130 | bbr->prev_ca_state = TCP_CA_Loss; 1131 | bbr->full_bw = 0; 1132 | bbr->round_start = 1; /* treat RTO like end of a round */ 1133 | bbr_lt_bw_sampling(sk, &rs); 1134 | } 1135 | } 1136 | 1137 | static struct tcp_congestion_ops tcp_bbr_cong_ops __read_mostly = { 1138 | .flags = TCP_CONG_NON_RESTRICTED, 1139 | .name = "bbrplus", 1140 | .owner = THIS_MODULE, 1141 | .init = bbr_init, 1142 | .cong_control = bbr_main, 1143 | .sndbuf_expand = bbr_sndbuf_expand, 1144 | .undo_cwnd = bbr_undo_cwnd, 1145 | .cwnd_event = bbr_cwnd_event, 1146 | .ssthresh = bbr_ssthresh, 1147 | .tso_segs_goal = bbr_tso_segs_goal, 1148 | .get_info = bbr_get_info, 1149 | .set_state = bbr_set_state, 1150 | }; 1151 | 1152 | static int __init bbr_register(void) 1153 | { 1154 | BUILD_BUG_ON(sizeof(struct bbr) > ICSK_CA_PRIV_SIZE); 1155 | return tcp_register_congestion_control(&tcp_bbr_cong_ops); 1156 | } 1157 | 1158 | static void __exit bbr_unregister(void) 1159 | { 1160 | tcp_unregister_congestion_control(&tcp_bbr_cong_ops); 1161 | } 1162 | 1163 | module_init(bbr_register); 1164 | module_exit(bbr_unregister); 1165 | 1166 | MODULE_AUTHOR("Van Jacobson "); 1167 | MODULE_AUTHOR("Neal Cardwell "); 1168 | MODULE_AUTHOR("Yuchung Cheng "); 1169 | MODULE_AUTHOR("Soheil Hassas Yeganeh "); 1170 | MODULE_LICENSE("Dual BSD/GPL"); 1171 | MODULE_DESCRIPTION("TCP BBR (Bottleneck Bandwidth and RTT)"); --------------------------------------------------------------------------------