tcp.c 81 KB

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027102810291030103110321033103410351036103710381039104010411042104310441045104610471048104910501051105210531054105510561057105810591060106110621063106410651066106710681069107010711072107310741075107610771078107910801081108210831084108510861087108810891090109110921093109410951096109710981099110011011102110311041105110611071108110911101111111211131114111511161117111811191120112111221123112411251126112711281129113011311132113311341135113611371138113911401141114211431144114511461147114811491150115111521153115411551156115711581159116011611162116311641165116611671168116911701171117211731174117511761177117811791180118111821183118411851186118711881189119011911192119311941195119611971198119912001201120212031204120512061207120812091210121112121213121412151216121712181219122012211222122312241225122612271228122912301231123212331234123512361237123812391240124112421243124412451246124712481249125012511252125312541255125612571258125912601261126212631264126512661267126812691270127112721273127412751276127712781279128012811282128312841285128612871288128912901291129212931294129512961297129812991300130113021303130413051306130713081309131013111312131313141315131613171318131913201321132213231324132513261327132813291330133113321333133413351336133713381339134013411342134313441345134613471348134913501351135213531354135513561357135813591360136113621363136413651366136713681369137013711372137313741375137613771378137913801381138213831384138513861387138813891390139113921393139413951396139713981399140014011402140314041405140614071408140914101411141214131414141514161417141814191420142114221423142414251426142714281429143014311432143314341435143614371438143914401441144214431444144514461447144814491450145114521453145414551456145714581459146014611462146314641465146614671468146914701471147214731474147514761477147814791480148114821483148414851486148714881489149014911492149314941495149614971498149915001501150215031504150515061507150815091510151115121513151415151516151715181519152015211522152315241525152615271528152915301531153215331534153515361537153815391540154115421543154415451546154715481549155015511552155315541555155615571558155915601561156215631564156515661567156815691570157115721573157415751576157715781579158015811582158315841585158615871588158915901591159215931594159515961597159815991600160116021603160416051606160716081609161016111612161316141615161616171618161916201621162216231624162516261627162816291630163116321633163416351636163716381639164016411642164316441645164616471648164916501651165216531654165516561657165816591660166116621663166416651666166716681669167016711672167316741675167616771678167916801681168216831684168516861687168816891690169116921693169416951696169716981699170017011702170317041705170617071708170917101711171217131714171517161717171817191720172117221723172417251726172717281729173017311732173317341735173617371738173917401741174217431744174517461747174817491750175117521753175417551756175717581759176017611762176317641765176617671768176917701771177217731774177517761777177817791780178117821783178417851786178717881789179017911792179317941795179617971798179918001801180218031804180518061807180818091810181118121813181418151816181718181819182018211822182318241825182618271828182918301831183218331834183518361837183818391840184118421843184418451846184718481849185018511852185318541855185618571858185918601861186218631864186518661867186818691870187118721873187418751876187718781879188018811882188318841885188618871888188918901891189218931894189518961897189818991900190119021903190419051906190719081909191019111912191319141915191619171918191919201921192219231924192519261927192819291930193119321933193419351936193719381939194019411942194319441945194619471948194919501951195219531954195519561957195819591960196119621963196419651966196719681969197019711972197319741975197619771978197919801981198219831984198519861987198819891990199119921993199419951996199719981999200020012002200320042005200620072008200920102011201220132014201520162017201820192020202120222023202420252026202720282029203020312032203320342035203620372038203920402041204220432044204520462047204820492050205120522053205420552056205720582059206020612062206320642065206620672068206920702071207220732074207520762077207820792080208120822083208420852086208720882089209020912092209320942095209620972098209921002101210221032104210521062107210821092110211121122113211421152116211721182119212021212122212321242125212621272128212921302131213221332134213521362137213821392140214121422143214421452146214721482149215021512152215321542155215621572158215921602161216221632164216521662167216821692170217121722173217421752176217721782179218021812182218321842185218621872188218921902191219221932194219521962197219821992200220122022203220422052206220722082209221022112212221322142215221622172218221922202221222222232224222522262227222822292230223122322233223422352236223722382239224022412242224322442245224622472248224922502251225222532254225522562257225822592260226122622263226422652266226722682269227022712272227322742275227622772278227922802281228222832284228522862287228822892290229122922293229422952296229722982299230023012302230323042305230623072308230923102311231223132314231523162317231823192320232123222323232423252326232723282329233023312332233323342335233623372338233923402341234223432344234523462347234823492350235123522353235423552356235723582359236023612362236323642365236623672368236923702371237223732374237523762377237823792380238123822383238423852386238723882389239023912392239323942395239623972398239924002401240224032404240524062407240824092410241124122413241424152416241724182419242024212422242324242425242624272428242924302431243224332434243524362437243824392440244124422443244424452446244724482449245024512452245324542455245624572458245924602461246224632464246524662467246824692470247124722473247424752476247724782479248024812482248324842485248624872488248924902491249224932494249524962497249824992500250125022503250425052506250725082509251025112512251325142515251625172518251925202521252225232524252525262527252825292530253125322533253425352536253725382539254025412542254325442545254625472548254925502551255225532554255525562557255825592560256125622563256425652566256725682569257025712572257325742575257625772578257925802581258225832584258525862587258825892590259125922593259425952596259725982599260026012602260326042605260626072608260926102611261226132614261526162617261826192620262126222623262426252626262726282629263026312632263326342635263626372638263926402641264226432644264526462647264826492650265126522653265426552656265726582659266026612662266326642665266626672668266926702671267226732674267526762677267826792680268126822683268426852686268726882689269026912692269326942695269626972698269927002701270227032704270527062707270827092710271127122713271427152716271727182719272027212722272327242725272627272728272927302731273227332734273527362737273827392740274127422743274427452746274727482749275027512752275327542755275627572758275927602761276227632764276527662767276827692770277127722773277427752776277727782779278027812782278327842785278627872788278927902791279227932794279527962797279827992800280128022803280428052806280728082809281028112812281328142815281628172818281928202821282228232824282528262827282828292830283128322833283428352836283728382839284028412842284328442845284628472848284928502851285228532854285528562857285828592860286128622863286428652866286728682869287028712872287328742875287628772878287928802881288228832884288528862887288828892890289128922893289428952896289728982899290029012902290329042905290629072908290929102911291229132914291529162917291829192920292129222923292429252926292729282929293029312932293329342935293629372938293929402941294229432944294529462947294829492950295129522953295429552956295729582959296029612962296329642965296629672968296929702971297229732974297529762977297829792980298129822983298429852986298729882989299029912992299329942995299629972998299930003001300230033004300530063007300830093010301130123013301430153016301730183019302030213022302330243025302630273028302930303031303230333034303530363037303830393040304130423043304430453046304730483049305030513052305330543055305630573058305930603061306230633064306530663067306830693070307130723073307430753076307730783079308030813082308330843085308630873088308930903091309230933094309530963097309830993100310131023103310431053106
  1. /*
  2. * INET An implementation of the TCP/IP protocol suite for the LINUX
  3. * operating system. INET is implemented using the BSD Socket
  4. * interface as the means of communication with the user level.
  5. *
  6. * Implementation of the Transmission Control Protocol(TCP).
  7. *
  8. * Authors: Ross Biro
  9. * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  10. * Mark Evans, <evansmp@uhura.aston.ac.uk>
  11. * Corey Minyard <wf-rch!minyard@relay.EU.net>
  12. * Florian La Roche, <flla@stud.uni-sb.de>
  13. * Charles Hedrick, <hedrick@klinzhai.rutgers.edu>
  14. * Linus Torvalds, <torvalds@cs.helsinki.fi>
  15. * Alan Cox, <gw4pts@gw4pts.ampr.org>
  16. * Matthew Dillon, <dillon@apollo.west.oic.com>
  17. * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
  18. * Jorge Cwik, <jorge@laser.satlink.net>
  19. *
  20. * Fixes:
  21. * Alan Cox : Numerous verify_area() calls
  22. * Alan Cox : Set the ACK bit on a reset
  23. * Alan Cox : Stopped it crashing if it closed while
  24. * sk->inuse=1 and was trying to connect
  25. * (tcp_err()).
  26. * Alan Cox : All icmp error handling was broken
  27. * pointers passed where wrong and the
  28. * socket was looked up backwards. Nobody
  29. * tested any icmp error code obviously.
  30. * Alan Cox : tcp_err() now handled properly. It
  31. * wakes people on errors. poll
  32. * behaves and the icmp error race
  33. * has gone by moving it into sock.c
  34. * Alan Cox : tcp_send_reset() fixed to work for
  35. * everything not just packets for
  36. * unknown sockets.
  37. * Alan Cox : tcp option processing.
  38. * Alan Cox : Reset tweaked (still not 100%) [Had
  39. * syn rule wrong]
  40. * Herp Rosmanith : More reset fixes
  41. * Alan Cox : No longer acks invalid rst frames.
  42. * Acking any kind of RST is right out.
  43. * Alan Cox : Sets an ignore me flag on an rst
  44. * receive otherwise odd bits of prattle
  45. * escape still
  46. * Alan Cox : Fixed another acking RST frame bug.
  47. * Should stop LAN workplace lockups.
  48. * Alan Cox : Some tidyups using the new skb list
  49. * facilities
  50. * Alan Cox : sk->keepopen now seems to work
  51. * Alan Cox : Pulls options out correctly on accepts
  52. * Alan Cox : Fixed assorted sk->rqueue->next errors
  53. * Alan Cox : PSH doesn't end a TCP read. Switched a
  54. * bit to skb ops.
  55. * Alan Cox : Tidied tcp_data to avoid a potential
  56. * nasty.
  57. * Alan Cox : Added some better commenting, as the
  58. * tcp is hard to follow
  59. * Alan Cox : Removed incorrect check for 20 * psh
  60. * Michael O'Reilly : ack < copied bug fix.
  61. * Johannes Stille : Misc tcp fixes (not all in yet).
  62. * Alan Cox : FIN with no memory -> CRASH
  63. * Alan Cox : Added socket option proto entries.
  64. * Also added awareness of them to accept.
  65. * Alan Cox : Added TCP options (SOL_TCP)
  66. * Alan Cox : Switched wakeup calls to callbacks,
  67. * so the kernel can layer network
  68. * sockets.
  69. * Alan Cox : Use ip_tos/ip_ttl settings.
  70. * Alan Cox : Handle FIN (more) properly (we hope).
  71. * Alan Cox : RST frames sent on unsynchronised
  72. * state ack error.
  73. * Alan Cox : Put in missing check for SYN bit.
  74. * Alan Cox : Added tcp_select_window() aka NET2E
  75. * window non shrink trick.
  76. * Alan Cox : Added a couple of small NET2E timer
  77. * fixes
  78. * Charles Hedrick : TCP fixes
  79. * Toomas Tamm : TCP window fixes
  80. * Alan Cox : Small URG fix to rlogin ^C ack fight
  81. * Charles Hedrick : Rewrote most of it to actually work
  82. * Linus : Rewrote tcp_read() and URG handling
  83. * completely
  84. * Gerhard Koerting: Fixed some missing timer handling
  85. * Matthew Dillon : Reworked TCP machine states as per RFC
  86. * Gerhard Koerting: PC/TCP workarounds
  87. * Adam Caldwell : Assorted timer/timing errors
  88. * Matthew Dillon : Fixed another RST bug
  89. * Alan Cox : Move to kernel side addressing changes.
  90. * Alan Cox : Beginning work on TCP fastpathing
  91. * (not yet usable)
  92. * Arnt Gulbrandsen: Turbocharged tcp_check() routine.
  93. * Alan Cox : TCP fast path debugging
  94. * Alan Cox : Window clamping
  95. * Michael Riepe : Bug in tcp_check()
  96. * Matt Dillon : More TCP improvements and RST bug fixes
  97. * Matt Dillon : Yet more small nasties remove from the
  98. * TCP code (Be very nice to this man if
  99. * tcp finally works 100%) 8)
  100. * Alan Cox : BSD accept semantics.
  101. * Alan Cox : Reset on closedown bug.
  102. * Peter De Schrijver : ENOTCONN check missing in tcp_sendto().
  103. * Michael Pall : Handle poll() after URG properly in
  104. * all cases.
  105. * Michael Pall : Undo the last fix in tcp_read_urg()
  106. * (multi URG PUSH broke rlogin).
  107. * Michael Pall : Fix the multi URG PUSH problem in
  108. * tcp_readable(), poll() after URG
  109. * works now.
  110. * Michael Pall : recv(...,MSG_OOB) never blocks in the
  111. * BSD api.
  112. * Alan Cox : Changed the semantics of sk->socket to
  113. * fix a race and a signal problem with
  114. * accept() and async I/O.
  115. * Alan Cox : Relaxed the rules on tcp_sendto().
  116. * Yury Shevchuk : Really fixed accept() blocking problem.
  117. * Craig I. Hagan : Allow for BSD compatible TIME_WAIT for
  118. * clients/servers which listen in on
  119. * fixed ports.
  120. * Alan Cox : Cleaned the above up and shrank it to
  121. * a sensible code size.
  122. * Alan Cox : Self connect lockup fix.
  123. * Alan Cox : No connect to multicast.
  124. * Ross Biro : Close unaccepted children on master
  125. * socket close.
  126. * Alan Cox : Reset tracing code.
  127. * Alan Cox : Spurious resets on shutdown.
  128. * Alan Cox : Giant 15 minute/60 second timer error
  129. * Alan Cox : Small whoops in polling before an
  130. * accept.
  131. * Alan Cox : Kept the state trace facility since
  132. * it's handy for debugging.
  133. * Alan Cox : More reset handler fixes.
  134. * Alan Cox : Started rewriting the code based on
  135. * the RFC's for other useful protocol
  136. * references see: Comer, KA9Q NOS, and
  137. * for a reference on the difference
  138. * between specifications and how BSD
  139. * works see the 4.4lite source.
  140. * A.N.Kuznetsov : Don't time wait on completion of tidy
  141. * close.
  142. * Linus Torvalds : Fin/Shutdown & copied_seq changes.
  143. * Linus Torvalds : Fixed BSD port reuse to work first syn
  144. * Alan Cox : Reimplemented timers as per the RFC
  145. * and using multiple timers for sanity.
  146. * Alan Cox : Small bug fixes, and a lot of new
  147. * comments.
  148. * Alan Cox : Fixed dual reader crash by locking
  149. * the buffers (much like datagram.c)
  150. * Alan Cox : Fixed stuck sockets in probe. A probe
  151. * now gets fed up of retrying without
  152. * (even a no space) answer.
  153. * Alan Cox : Extracted closing code better
  154. * Alan Cox : Fixed the closing state machine to
  155. * resemble the RFC.
  156. * Alan Cox : More 'per spec' fixes.
  157. * Jorge Cwik : Even faster checksumming.
  158. * Alan Cox : tcp_data() doesn't ack illegal PSH
  159. * only frames. At least one pc tcp stack
  160. * generates them.
  161. * Alan Cox : Cache last socket.
  162. * Alan Cox : Per route irtt.
  163. * Matt Day : poll()->select() match BSD precisely on error
  164. * Alan Cox : New buffers
  165. * Marc Tamsky : Various sk->prot->retransmits and
  166. * sk->retransmits misupdating fixed.
  167. * Fixed tcp_write_timeout: stuck close,
  168. * and TCP syn retries gets used now.
  169. * Mark Yarvis : In tcp_read_wakeup(), don't send an
  170. * ack if state is TCP_CLOSED.
  171. * Alan Cox : Look up device on a retransmit - routes may
  172. * change. Doesn't yet cope with MSS shrink right
  173. * but it's a start!
  174. * Marc Tamsky : Closing in closing fixes.
  175. * Mike Shaver : RFC1122 verifications.
  176. * Alan Cox : rcv_saddr errors.
  177. * Alan Cox : Block double connect().
  178. * Alan Cox : Small hooks for enSKIP.
  179. * Alexey Kuznetsov: Path MTU discovery.
  180. * Alan Cox : Support soft errors.
  181. * Alan Cox : Fix MTU discovery pathological case
  182. * when the remote claims no mtu!
  183. * Marc Tamsky : TCP_CLOSE fix.
  184. * Colin (G3TNE) : Send a reset on syn ack replies in
  185. * window but wrong (fixes NT lpd problems)
  186. * Pedro Roque : Better TCP window handling, delayed ack.
  187. * Joerg Reuter : No modification of locked buffers in
  188. * tcp_do_retransmit()
  189. * Eric Schenk : Changed receiver side silly window
  190. * avoidance algorithm to BSD style
  191. * algorithm. This doubles throughput
  192. * against machines running Solaris,
  193. * and seems to result in general
  194. * improvement.
  195. * Stefan Magdalinski : adjusted tcp_readable() to fix FIONREAD
  196. * Willy Konynenberg : Transparent proxying support.
  197. * Mike McLagan : Routing by source
  198. * Keith Owens : Do proper merging with partial SKB's in
  199. * tcp_do_sendmsg to avoid burstiness.
  200. * Eric Schenk : Fix fast close down bug with
  201. * shutdown() followed by close().
  202. * Andi Kleen : Make poll agree with SIGIO
  203. * Salvatore Sanfilippo : Support SO_LINGER with linger == 1 and
  204. * lingertime == 0 (RFC 793 ABORT Call)
  205. * Hirokazu Takahashi : Use copy_from_user() instead of
  206. * csum_and_copy_from_user() if possible.
  207. *
  208. * This program is free software; you can redistribute it and/or
  209. * modify it under the terms of the GNU General Public License
  210. * as published by the Free Software Foundation; either version
  211. * 2 of the License, or(at your option) any later version.
  212. *
  213. * Description of States:
  214. *
  215. * TCP_SYN_SENT sent a connection request, waiting for ack
  216. *
  217. * TCP_SYN_RECV received a connection request, sent ack,
  218. * waiting for final ack in three-way handshake.
  219. *
  220. * TCP_ESTABLISHED connection established
  221. *
  222. * TCP_FIN_WAIT1 our side has shutdown, waiting to complete
  223. * transmission of remaining buffered data
  224. *
  225. * TCP_FIN_WAIT2 all buffered data sent, waiting for remote
  226. * to shutdown
  227. *
  228. * TCP_CLOSING both sides have shutdown but we still have
  229. * data we have to finish sending
  230. *
  231. * TCP_TIME_WAIT timeout to catch resent junk before entering
  232. * closed, can only be entered from FIN_WAIT2
  233. * or CLOSING. Required because the other end
  234. * may not have gotten our last ACK causing it
  235. * to retransmit the data packet (which we ignore)
  236. *
  237. * TCP_CLOSE_WAIT remote side has shutdown and is waiting for
  238. * us to finish writing our data and to shutdown
  239. * (we have to close() to move on to LAST_ACK)
  240. *
  241. * TCP_LAST_ACK out side has shutdown after remote has
  242. * shutdown. There may still be data in our
  243. * buffer that we have to finish sending
  244. *
  245. * TCP_CLOSE socket is finished
  246. */
  247. #define pr_fmt(fmt) "TCP: " fmt
  248. #include <linux/kernel.h>
  249. #include <linux/module.h>
  250. #include <linux/types.h>
  251. #include <linux/fcntl.h>
  252. #include <linux/poll.h>
  253. #include <linux/init.h>
  254. #include <linux/fs.h>
  255. #include <linux/skbuff.h>
  256. #include <linux/scatterlist.h>
  257. #include <linux/splice.h>
  258. #include <linux/net.h>
  259. #include <linux/socket.h>
  260. #include <linux/random.h>
  261. #include <linux/bootmem.h>
  262. #include <linux/highmem.h>
  263. #include <linux/swap.h>
  264. #include <linux/cache.h>
  265. #include <linux/err.h>
  266. #include <linux/crypto.h>
  267. #include <linux/time.h>
  268. #include <linux/slab.h>
  269. #include <net/icmp.h>
  270. #include <net/inet_common.h>
  271. #include <net/tcp.h>
  272. #include <net/xfrm.h>
  273. #include <net/ip.h>
  274. #include <net/sock.h>
  275. #include <asm/uaccess.h>
  276. #include <asm/ioctls.h>
  277. #include <net/busy_poll.h>
  278. int sysctl_tcp_fin_timeout __read_mostly = TCP_FIN_TIMEOUT;
  279. int sysctl_tcp_min_tso_segs __read_mostly = 2;
  280. int sysctl_tcp_autocorking __read_mostly = 1;
  281. struct percpu_counter tcp_orphan_count;
  282. EXPORT_SYMBOL_GPL(tcp_orphan_count);
  283. long sysctl_tcp_mem[3] __read_mostly;
  284. int sysctl_tcp_wmem[3] __read_mostly;
  285. int sysctl_tcp_rmem[3] __read_mostly;
  286. EXPORT_SYMBOL(sysctl_tcp_mem);
  287. EXPORT_SYMBOL(sysctl_tcp_rmem);
  288. EXPORT_SYMBOL(sysctl_tcp_wmem);
  289. atomic_long_t tcp_memory_allocated; /* Current allocated memory. */
  290. EXPORT_SYMBOL(tcp_memory_allocated);
  291. /*
  292. * Current number of TCP sockets.
  293. */
  294. struct percpu_counter tcp_sockets_allocated;
  295. EXPORT_SYMBOL(tcp_sockets_allocated);
  296. /*
  297. * TCP splice context
  298. */
  299. struct tcp_splice_state {
  300. struct pipe_inode_info *pipe;
  301. size_t len;
  302. unsigned int flags;
  303. };
  304. /*
  305. * Pressure flag: try to collapse.
  306. * Technical note: it is used by multiple contexts non atomically.
  307. * All the __sk_mem_schedule() is of this nature: accounting
  308. * is strict, actions are advisory and have some latency.
  309. */
  310. int tcp_memory_pressure __read_mostly;
  311. EXPORT_SYMBOL(tcp_memory_pressure);
  312. void tcp_enter_memory_pressure(struct sock *sk)
  313. {
  314. if (!tcp_memory_pressure) {
  315. NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMEMORYPRESSURES);
  316. tcp_memory_pressure = 1;
  317. }
  318. }
  319. EXPORT_SYMBOL(tcp_enter_memory_pressure);
  320. /* Convert seconds to retransmits based on initial and max timeout */
  321. static u8 secs_to_retrans(int seconds, int timeout, int rto_max)
  322. {
  323. u8 res = 0;
  324. if (seconds > 0) {
  325. int period = timeout;
  326. res = 1;
  327. while (seconds > period && res < 255) {
  328. res++;
  329. timeout <<= 1;
  330. if (timeout > rto_max)
  331. timeout = rto_max;
  332. period += timeout;
  333. }
  334. }
  335. return res;
  336. }
  337. /* Convert retransmits to seconds based on initial and max timeout */
  338. static int retrans_to_secs(u8 retrans, int timeout, int rto_max)
  339. {
  340. int period = 0;
  341. if (retrans > 0) {
  342. period = timeout;
  343. while (--retrans) {
  344. timeout <<= 1;
  345. if (timeout > rto_max)
  346. timeout = rto_max;
  347. period += timeout;
  348. }
  349. }
  350. return period;
  351. }
  352. /* Address-family independent initialization for a tcp_sock.
  353. *
  354. * NOTE: A lot of things set to zero explicitly by call to
  355. * sk_alloc() so need not be done here.
  356. */
  357. void tcp_init_sock(struct sock *sk)
  358. {
  359. struct inet_connection_sock *icsk = inet_csk(sk);
  360. struct tcp_sock *tp = tcp_sk(sk);
  361. __skb_queue_head_init(&tp->out_of_order_queue);
  362. tcp_init_xmit_timers(sk);
  363. tcp_prequeue_init(tp);
  364. INIT_LIST_HEAD(&tp->tsq_node);
  365. icsk->icsk_rto = TCP_TIMEOUT_INIT;
  366. tp->mdev_us = jiffies_to_usecs(TCP_TIMEOUT_INIT);
  367. /* So many TCP implementations out there (incorrectly) count the
  368. * initial SYN frame in their delayed-ACK and congestion control
  369. * algorithms that we must have the following bandaid to talk
  370. * efficiently to them. -DaveM
  371. */
  372. tp->snd_cwnd = TCP_INIT_CWND;
  373. /* See draft-stevens-tcpca-spec-01 for discussion of the
  374. * initialization of these values.
  375. */
  376. tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
  377. tp->snd_cwnd_clamp = ~0;
  378. tp->mss_cache = TCP_MSS_DEFAULT;
  379. tp->reordering = sysctl_tcp_reordering;
  380. tcp_enable_early_retrans(tp);
  381. tcp_assign_congestion_control(sk);
  382. tp->tsoffset = 0;
  383. sk->sk_state = TCP_CLOSE;
  384. sk->sk_write_space = sk_stream_write_space;
  385. sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
  386. icsk->icsk_sync_mss = tcp_sync_mss;
  387. sk->sk_sndbuf = sysctl_tcp_wmem[1];
  388. sk->sk_rcvbuf = sysctl_tcp_rmem[1];
  389. local_bh_disable();
  390. sock_update_memcg(sk);
  391. sk_sockets_allocated_inc(sk);
  392. local_bh_enable();
  393. }
  394. EXPORT_SYMBOL(tcp_init_sock);
  395. static void tcp_tx_timestamp(struct sock *sk, struct sk_buff *skb)
  396. {
  397. if (sk->sk_tsflags) {
  398. struct skb_shared_info *shinfo = skb_shinfo(skb);
  399. sock_tx_timestamp(sk, &shinfo->tx_flags);
  400. if (shinfo->tx_flags & SKBTX_ANY_TSTAMP)
  401. shinfo->tskey = TCP_SKB_CB(skb)->seq + skb->len - 1;
  402. }
  403. }
  404. /*
  405. * Wait for a TCP event.
  406. *
  407. * Note that we don't need to lock the socket, as the upper poll layers
  408. * take care of normal races (between the test and the event) and we don't
  409. * go look at any of the socket buffers directly.
  410. */
  411. unsigned int tcp_poll(struct file *file, struct socket *sock, poll_table *wait)
  412. {
  413. unsigned int mask;
  414. struct sock *sk = sock->sk;
  415. const struct tcp_sock *tp = tcp_sk(sk);
  416. sock_rps_record_flow(sk);
  417. sock_poll_wait(file, sk_sleep(sk), wait);
  418. if (sk->sk_state == TCP_LISTEN)
  419. return inet_csk_listen_poll(sk);
  420. /* Socket is not locked. We are protected from async events
  421. * by poll logic and correct handling of state changes
  422. * made by other threads is impossible in any case.
  423. */
  424. mask = 0;
  425. /*
  426. * POLLHUP is certainly not done right. But poll() doesn't
  427. * have a notion of HUP in just one direction, and for a
  428. * socket the read side is more interesting.
  429. *
  430. * Some poll() documentation says that POLLHUP is incompatible
  431. * with the POLLOUT/POLLWR flags, so somebody should check this
  432. * all. But careful, it tends to be safer to return too many
  433. * bits than too few, and you can easily break real applications
  434. * if you don't tell them that something has hung up!
  435. *
  436. * Check-me.
  437. *
  438. * Check number 1. POLLHUP is _UNMASKABLE_ event (see UNIX98 and
  439. * our fs/select.c). It means that after we received EOF,
  440. * poll always returns immediately, making impossible poll() on write()
  441. * in state CLOSE_WAIT. One solution is evident --- to set POLLHUP
  442. * if and only if shutdown has been made in both directions.
  443. * Actually, it is interesting to look how Solaris and DUX
  444. * solve this dilemma. I would prefer, if POLLHUP were maskable,
  445. * then we could set it on SND_SHUTDOWN. BTW examples given
  446. * in Stevens' books assume exactly this behaviour, it explains
  447. * why POLLHUP is incompatible with POLLOUT. --ANK
  448. *
  449. * NOTE. Check for TCP_CLOSE is added. The goal is to prevent
  450. * blocking on fresh not-connected or disconnected socket. --ANK
  451. */
  452. if (sk->sk_shutdown == SHUTDOWN_MASK || sk->sk_state == TCP_CLOSE)
  453. mask |= POLLHUP;
  454. if (sk->sk_shutdown & RCV_SHUTDOWN)
  455. mask |= POLLIN | POLLRDNORM | POLLRDHUP;
  456. /* Connected or passive Fast Open socket? */
  457. if (sk->sk_state != TCP_SYN_SENT &&
  458. (sk->sk_state != TCP_SYN_RECV || tp->fastopen_rsk != NULL)) {
  459. int target = sock_rcvlowat(sk, 0, INT_MAX);
  460. if (tp->urg_seq == tp->copied_seq &&
  461. !sock_flag(sk, SOCK_URGINLINE) &&
  462. tp->urg_data)
  463. target++;
  464. /* Potential race condition. If read of tp below will
  465. * escape above sk->sk_state, we can be illegally awaken
  466. * in SYN_* states. */
  467. if (tp->rcv_nxt - tp->copied_seq >= target)
  468. mask |= POLLIN | POLLRDNORM;
  469. if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
  470. if (sk_stream_is_writeable(sk)) {
  471. mask |= POLLOUT | POLLWRNORM;
  472. } else { /* send SIGIO later */
  473. set_bit(SOCK_ASYNC_NOSPACE,
  474. &sk->sk_socket->flags);
  475. set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
  476. /* Race breaker. If space is freed after
  477. * wspace test but before the flags are set,
  478. * IO signal will be lost.
  479. */
  480. if (sk_stream_is_writeable(sk))
  481. mask |= POLLOUT | POLLWRNORM;
  482. }
  483. } else
  484. mask |= POLLOUT | POLLWRNORM;
  485. if (tp->urg_data & TCP_URG_VALID)
  486. mask |= POLLPRI;
  487. }
  488. /* This barrier is coupled with smp_wmb() in tcp_reset() */
  489. smp_rmb();
  490. if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
  491. mask |= POLLERR;
  492. return mask;
  493. }
  494. EXPORT_SYMBOL(tcp_poll);
  495. int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg)
  496. {
  497. struct tcp_sock *tp = tcp_sk(sk);
  498. int answ;
  499. bool slow;
  500. switch (cmd) {
  501. case SIOCINQ:
  502. if (sk->sk_state == TCP_LISTEN)
  503. return -EINVAL;
  504. slow = lock_sock_fast(sk);
  505. if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))
  506. answ = 0;
  507. else if (sock_flag(sk, SOCK_URGINLINE) ||
  508. !tp->urg_data ||
  509. before(tp->urg_seq, tp->copied_seq) ||
  510. !before(tp->urg_seq, tp->rcv_nxt)) {
  511. answ = tp->rcv_nxt - tp->copied_seq;
  512. /* Subtract 1, if FIN was received */
  513. if (answ && sock_flag(sk, SOCK_DONE))
  514. answ--;
  515. } else
  516. answ = tp->urg_seq - tp->copied_seq;
  517. unlock_sock_fast(sk, slow);
  518. break;
  519. case SIOCATMARK:
  520. answ = tp->urg_data && tp->urg_seq == tp->copied_seq;
  521. break;
  522. case SIOCOUTQ:
  523. if (sk->sk_state == TCP_LISTEN)
  524. return -EINVAL;
  525. if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))
  526. answ = 0;
  527. else
  528. answ = tp->write_seq - tp->snd_una;
  529. break;
  530. case SIOCOUTQNSD:
  531. if (sk->sk_state == TCP_LISTEN)
  532. return -EINVAL;
  533. if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))
  534. answ = 0;
  535. else
  536. answ = tp->write_seq - tp->snd_nxt;
  537. break;
  538. default:
  539. return -ENOIOCTLCMD;
  540. }
  541. return put_user(answ, (int __user *)arg);
  542. }
  543. EXPORT_SYMBOL(tcp_ioctl);
  544. static inline void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb)
  545. {
  546. TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH;
  547. tp->pushed_seq = tp->write_seq;
  548. }
  549. static inline bool forced_push(const struct tcp_sock *tp)
  550. {
  551. return after(tp->write_seq, tp->pushed_seq + (tp->max_window >> 1));
  552. }
  553. static void skb_entail(struct sock *sk, struct sk_buff *skb)
  554. {
  555. struct tcp_sock *tp = tcp_sk(sk);
  556. struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
  557. skb->csum = 0;
  558. tcb->seq = tcb->end_seq = tp->write_seq;
  559. tcb->tcp_flags = TCPHDR_ACK;
  560. tcb->sacked = 0;
  561. __skb_header_release(skb);
  562. tcp_add_write_queue_tail(sk, skb);
  563. sk->sk_wmem_queued += skb->truesize;
  564. sk_mem_charge(sk, skb->truesize);
  565. if (tp->nonagle & TCP_NAGLE_PUSH)
  566. tp->nonagle &= ~TCP_NAGLE_PUSH;
  567. }
  568. static inline void tcp_mark_urg(struct tcp_sock *tp, int flags)
  569. {
  570. if (flags & MSG_OOB)
  571. tp->snd_up = tp->write_seq;
  572. }
  573. /* If a not yet filled skb is pushed, do not send it if
  574. * we have data packets in Qdisc or NIC queues :
  575. * Because TX completion will happen shortly, it gives a chance
  576. * to coalesce future sendmsg() payload into this skb, without
  577. * need for a timer, and with no latency trade off.
  578. * As packets containing data payload have a bigger truesize
  579. * than pure acks (dataless) packets, the last checks prevent
  580. * autocorking if we only have an ACK in Qdisc/NIC queues,
  581. * or if TX completion was delayed after we processed ACK packet.
  582. */
  583. static bool tcp_should_autocork(struct sock *sk, struct sk_buff *skb,
  584. int size_goal)
  585. {
  586. return skb->len < size_goal &&
  587. sysctl_tcp_autocorking &&
  588. skb != tcp_write_queue_head(sk) &&
  589. atomic_read(&sk->sk_wmem_alloc) > skb->truesize;
  590. }
  591. static void tcp_push(struct sock *sk, int flags, int mss_now,
  592. int nonagle, int size_goal)
  593. {
  594. struct tcp_sock *tp = tcp_sk(sk);
  595. struct sk_buff *skb;
  596. if (!tcp_send_head(sk))
  597. return;
  598. skb = tcp_write_queue_tail(sk);
  599. if (!(flags & MSG_MORE) || forced_push(tp))
  600. tcp_mark_push(tp, skb);
  601. tcp_mark_urg(tp, flags);
  602. if (tcp_should_autocork(sk, skb, size_goal)) {
  603. /* avoid atomic op if TSQ_THROTTLED bit is already set */
  604. if (!test_bit(TSQ_THROTTLED, &tp->tsq_flags)) {
  605. NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAUTOCORKING);
  606. set_bit(TSQ_THROTTLED, &tp->tsq_flags);
  607. }
  608. /* It is possible TX completion already happened
  609. * before we set TSQ_THROTTLED.
  610. */
  611. if (atomic_read(&sk->sk_wmem_alloc) > skb->truesize)
  612. return;
  613. }
  614. if (flags & MSG_MORE)
  615. nonagle = TCP_NAGLE_CORK;
  616. __tcp_push_pending_frames(sk, mss_now, nonagle);
  617. }
  618. static int tcp_splice_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb,
  619. unsigned int offset, size_t len)
  620. {
  621. struct tcp_splice_state *tss = rd_desc->arg.data;
  622. int ret;
  623. ret = skb_splice_bits(skb, offset, tss->pipe, min(rd_desc->count, len),
  624. tss->flags);
  625. if (ret > 0)
  626. rd_desc->count -= ret;
  627. return ret;
  628. }
  629. static int __tcp_splice_read(struct sock *sk, struct tcp_splice_state *tss)
  630. {
  631. /* Store TCP splice context information in read_descriptor_t. */
  632. read_descriptor_t rd_desc = {
  633. .arg.data = tss,
  634. .count = tss->len,
  635. };
  636. return tcp_read_sock(sk, &rd_desc, tcp_splice_data_recv);
  637. }
  638. /**
  639. * tcp_splice_read - splice data from TCP socket to a pipe
  640. * @sock: socket to splice from
  641. * @ppos: position (not valid)
  642. * @pipe: pipe to splice to
  643. * @len: number of bytes to splice
  644. * @flags: splice modifier flags
  645. *
  646. * Description:
  647. * Will read pages from given socket and fill them into a pipe.
  648. *
  649. **/
  650. ssize_t tcp_splice_read(struct socket *sock, loff_t *ppos,
  651. struct pipe_inode_info *pipe, size_t len,
  652. unsigned int flags)
  653. {
  654. struct sock *sk = sock->sk;
  655. struct tcp_splice_state tss = {
  656. .pipe = pipe,
  657. .len = len,
  658. .flags = flags,
  659. };
  660. long timeo;
  661. ssize_t spliced;
  662. int ret;
  663. sock_rps_record_flow(sk);
  664. /*
  665. * We can't seek on a socket input
  666. */
  667. if (unlikely(*ppos))
  668. return -ESPIPE;
  669. ret = spliced = 0;
  670. lock_sock(sk);
  671. timeo = sock_rcvtimeo(sk, sock->file->f_flags & O_NONBLOCK);
  672. while (tss.len) {
  673. ret = __tcp_splice_read(sk, &tss);
  674. if (ret < 0)
  675. break;
  676. else if (!ret) {
  677. if (spliced)
  678. break;
  679. if (sock_flag(sk, SOCK_DONE))
  680. break;
  681. if (sk->sk_err) {
  682. ret = sock_error(sk);
  683. break;
  684. }
  685. if (sk->sk_shutdown & RCV_SHUTDOWN)
  686. break;
  687. if (sk->sk_state == TCP_CLOSE) {
  688. /*
  689. * This occurs when user tries to read
  690. * from never connected socket.
  691. */
  692. if (!sock_flag(sk, SOCK_DONE))
  693. ret = -ENOTCONN;
  694. break;
  695. }
  696. if (!timeo) {
  697. ret = -EAGAIN;
  698. break;
  699. }
  700. sk_wait_data(sk, &timeo);
  701. if (signal_pending(current)) {
  702. ret = sock_intr_errno(timeo);
  703. break;
  704. }
  705. continue;
  706. }
  707. tss.len -= ret;
  708. spliced += ret;
  709. if (!timeo)
  710. break;
  711. release_sock(sk);
  712. lock_sock(sk);
  713. if (sk->sk_err || sk->sk_state == TCP_CLOSE ||
  714. (sk->sk_shutdown & RCV_SHUTDOWN) ||
  715. signal_pending(current))
  716. break;
  717. }
  718. release_sock(sk);
  719. if (spliced)
  720. return spliced;
  721. return ret;
  722. }
  723. EXPORT_SYMBOL(tcp_splice_read);
  724. struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp)
  725. {
  726. struct sk_buff *skb;
  727. /* The TCP header must be at least 32-bit aligned. */
  728. size = ALIGN(size, 4);
  729. skb = alloc_skb_fclone(size + sk->sk_prot->max_header, gfp);
  730. if (skb) {
  731. if (sk_wmem_schedule(sk, skb->truesize)) {
  732. skb_reserve(skb, sk->sk_prot->max_header);
  733. /*
  734. * Make sure that we have exactly size bytes
  735. * available to the caller, no more, no less.
  736. */
  737. skb->reserved_tailroom = skb->end - skb->tail - size;
  738. return skb;
  739. }
  740. __kfree_skb(skb);
  741. } else {
  742. sk->sk_prot->enter_memory_pressure(sk);
  743. sk_stream_moderate_sndbuf(sk);
  744. }
  745. return NULL;
  746. }
  747. static unsigned int tcp_xmit_size_goal(struct sock *sk, u32 mss_now,
  748. int large_allowed)
  749. {
  750. struct tcp_sock *tp = tcp_sk(sk);
  751. u32 new_size_goal, size_goal, hlen;
  752. if (!large_allowed || !sk_can_gso(sk))
  753. return mss_now;
  754. /* Maybe we should/could use sk->sk_prot->max_header here ? */
  755. hlen = inet_csk(sk)->icsk_af_ops->net_header_len +
  756. inet_csk(sk)->icsk_ext_hdr_len +
  757. tp->tcp_header_len;
  758. new_size_goal = sk->sk_gso_max_size - 1 - hlen;
  759. new_size_goal = tcp_bound_to_half_wnd(tp, new_size_goal);
  760. /* We try hard to avoid divides here */
  761. size_goal = tp->gso_segs * mss_now;
  762. if (unlikely(new_size_goal < size_goal ||
  763. new_size_goal >= size_goal + mss_now)) {
  764. tp->gso_segs = min_t(u16, new_size_goal / mss_now,
  765. sk->sk_gso_max_segs);
  766. size_goal = tp->gso_segs * mss_now;
  767. }
  768. return max(size_goal, mss_now);
  769. }
  770. static int tcp_send_mss(struct sock *sk, int *size_goal, int flags)
  771. {
  772. int mss_now;
  773. mss_now = tcp_current_mss(sk);
  774. *size_goal = tcp_xmit_size_goal(sk, mss_now, !(flags & MSG_OOB));
  775. return mss_now;
  776. }
  777. static ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset,
  778. size_t size, int flags)
  779. {
  780. struct tcp_sock *tp = tcp_sk(sk);
  781. int mss_now, size_goal;
  782. int err;
  783. ssize_t copied;
  784. long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
  785. /* Wait for a connection to finish. One exception is TCP Fast Open
  786. * (passive side) where data is allowed to be sent before a connection
  787. * is fully established.
  788. */
  789. if (((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) &&
  790. !tcp_passive_fastopen(sk)) {
  791. if ((err = sk_stream_wait_connect(sk, &timeo)) != 0)
  792. goto out_err;
  793. }
  794. clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
  795. mss_now = tcp_send_mss(sk, &size_goal, flags);
  796. copied = 0;
  797. err = -EPIPE;
  798. if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
  799. goto out_err;
  800. while (size > 0) {
  801. struct sk_buff *skb = tcp_write_queue_tail(sk);
  802. int copy, i;
  803. bool can_coalesce;
  804. if (!tcp_send_head(sk) || (copy = size_goal - skb->len) <= 0) {
  805. new_segment:
  806. if (!sk_stream_memory_free(sk))
  807. goto wait_for_sndbuf;
  808. skb = sk_stream_alloc_skb(sk, 0, sk->sk_allocation);
  809. if (!skb)
  810. goto wait_for_memory;
  811. skb_entail(sk, skb);
  812. copy = size_goal;
  813. }
  814. if (copy > size)
  815. copy = size;
  816. i = skb_shinfo(skb)->nr_frags;
  817. can_coalesce = skb_can_coalesce(skb, i, page, offset);
  818. if (!can_coalesce && i >= MAX_SKB_FRAGS) {
  819. tcp_mark_push(tp, skb);
  820. goto new_segment;
  821. }
  822. if (!sk_wmem_schedule(sk, copy))
  823. goto wait_for_memory;
  824. if (can_coalesce) {
  825. skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
  826. } else {
  827. get_page(page);
  828. skb_fill_page_desc(skb, i, page, offset, copy);
  829. }
  830. skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
  831. skb->len += copy;
  832. skb->data_len += copy;
  833. skb->truesize += copy;
  834. sk->sk_wmem_queued += copy;
  835. sk_mem_charge(sk, copy);
  836. skb->ip_summed = CHECKSUM_PARTIAL;
  837. tp->write_seq += copy;
  838. TCP_SKB_CB(skb)->end_seq += copy;
  839. tcp_skb_pcount_set(skb, 0);
  840. if (!copied)
  841. TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH;
  842. copied += copy;
  843. offset += copy;
  844. if (!(size -= copy)) {
  845. tcp_tx_timestamp(sk, skb);
  846. goto out;
  847. }
  848. if (skb->len < size_goal || (flags & MSG_OOB))
  849. continue;
  850. if (forced_push(tp)) {
  851. tcp_mark_push(tp, skb);
  852. __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH);
  853. } else if (skb == tcp_send_head(sk))
  854. tcp_push_one(sk, mss_now);
  855. continue;
  856. wait_for_sndbuf:
  857. set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
  858. wait_for_memory:
  859. tcp_push(sk, flags & ~MSG_MORE, mss_now,
  860. TCP_NAGLE_PUSH, size_goal);
  861. if ((err = sk_stream_wait_memory(sk, &timeo)) != 0)
  862. goto do_error;
  863. mss_now = tcp_send_mss(sk, &size_goal, flags);
  864. }
  865. out:
  866. if (copied && !(flags & MSG_SENDPAGE_NOTLAST))
  867. tcp_push(sk, flags, mss_now, tp->nonagle, size_goal);
  868. return copied;
  869. do_error:
  870. if (copied)
  871. goto out;
  872. out_err:
  873. return sk_stream_error(sk, flags, err);
  874. }
  875. int tcp_sendpage(struct sock *sk, struct page *page, int offset,
  876. size_t size, int flags)
  877. {
  878. ssize_t res;
  879. if (!(sk->sk_route_caps & NETIF_F_SG) ||
  880. !(sk->sk_route_caps & NETIF_F_ALL_CSUM))
  881. return sock_no_sendpage(sk->sk_socket, page, offset, size,
  882. flags);
  883. lock_sock(sk);
  884. res = do_tcp_sendpages(sk, page, offset, size, flags);
  885. release_sock(sk);
  886. return res;
  887. }
  888. EXPORT_SYMBOL(tcp_sendpage);
  889. static inline int select_size(const struct sock *sk, bool sg)
  890. {
  891. const struct tcp_sock *tp = tcp_sk(sk);
  892. int tmp = tp->mss_cache;
  893. if (sg) {
  894. if (sk_can_gso(sk)) {
  895. /* Small frames wont use a full page:
  896. * Payload will immediately follow tcp header.
  897. */
  898. tmp = SKB_WITH_OVERHEAD(2048 - MAX_TCP_HEADER);
  899. } else {
  900. int pgbreak = SKB_MAX_HEAD(MAX_TCP_HEADER);
  901. if (tmp >= pgbreak &&
  902. tmp <= pgbreak + (MAX_SKB_FRAGS - 1) * PAGE_SIZE)
  903. tmp = pgbreak;
  904. }
  905. }
  906. return tmp;
  907. }
  908. void tcp_free_fastopen_req(struct tcp_sock *tp)
  909. {
  910. if (tp->fastopen_req != NULL) {
  911. kfree(tp->fastopen_req);
  912. tp->fastopen_req = NULL;
  913. }
  914. }
  915. static int tcp_sendmsg_fastopen(struct sock *sk, struct msghdr *msg,
  916. int *copied, size_t size)
  917. {
  918. struct tcp_sock *tp = tcp_sk(sk);
  919. int err, flags;
  920. if (!(sysctl_tcp_fastopen & TFO_CLIENT_ENABLE))
  921. return -EOPNOTSUPP;
  922. if (tp->fastopen_req != NULL)
  923. return -EALREADY; /* Another Fast Open is in progress */
  924. tp->fastopen_req = kzalloc(sizeof(struct tcp_fastopen_request),
  925. sk->sk_allocation);
  926. if (unlikely(tp->fastopen_req == NULL))
  927. return -ENOBUFS;
  928. tp->fastopen_req->data = msg;
  929. tp->fastopen_req->size = size;
  930. flags = (msg->msg_flags & MSG_DONTWAIT) ? O_NONBLOCK : 0;
  931. err = __inet_stream_connect(sk->sk_socket, msg->msg_name,
  932. msg->msg_namelen, flags);
  933. *copied = tp->fastopen_req->copied;
  934. tcp_free_fastopen_req(tp);
  935. return err;
  936. }
  937. int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
  938. size_t size)
  939. {
  940. const struct iovec *iov;
  941. struct tcp_sock *tp = tcp_sk(sk);
  942. struct sk_buff *skb;
  943. int iovlen, flags, err, copied = 0;
  944. int mss_now = 0, size_goal, copied_syn = 0, offset = 0;
  945. bool sg;
  946. long timeo;
  947. lock_sock(sk);
  948. flags = msg->msg_flags;
  949. if (flags & MSG_FASTOPEN) {
  950. err = tcp_sendmsg_fastopen(sk, msg, &copied_syn, size);
  951. if (err == -EINPROGRESS && copied_syn > 0)
  952. goto out;
  953. else if (err)
  954. goto out_err;
  955. offset = copied_syn;
  956. }
  957. timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
  958. /* Wait for a connection to finish. One exception is TCP Fast Open
  959. * (passive side) where data is allowed to be sent before a connection
  960. * is fully established.
  961. */
  962. if (((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) &&
  963. !tcp_passive_fastopen(sk)) {
  964. if ((err = sk_stream_wait_connect(sk, &timeo)) != 0)
  965. goto do_error;
  966. }
  967. if (unlikely(tp->repair)) {
  968. if (tp->repair_queue == TCP_RECV_QUEUE) {
  969. copied = tcp_send_rcvq(sk, msg, size);
  970. goto out_nopush;
  971. }
  972. err = -EINVAL;
  973. if (tp->repair_queue == TCP_NO_QUEUE)
  974. goto out_err;
  975. /* 'common' sending to sendq */
  976. }
  977. /* This should be in poll */
  978. clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
  979. mss_now = tcp_send_mss(sk, &size_goal, flags);
  980. /* Ok commence sending. */
  981. iovlen = msg->msg_iter.nr_segs;
  982. iov = msg->msg_iter.iov;
  983. copied = 0;
  984. err = -EPIPE;
  985. if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
  986. goto out_err;
  987. sg = !!(sk->sk_route_caps & NETIF_F_SG);
  988. while (--iovlen >= 0) {
  989. size_t seglen = iov->iov_len;
  990. unsigned char __user *from = iov->iov_base;
  991. iov++;
  992. if (unlikely(offset > 0)) { /* Skip bytes copied in SYN */
  993. if (offset >= seglen) {
  994. offset -= seglen;
  995. continue;
  996. }
  997. seglen -= offset;
  998. from += offset;
  999. offset = 0;
  1000. }
  1001. while (seglen > 0) {
  1002. int copy = 0;
  1003. int max = size_goal;
  1004. skb = tcp_write_queue_tail(sk);
  1005. if (tcp_send_head(sk)) {
  1006. if (skb->ip_summed == CHECKSUM_NONE)
  1007. max = mss_now;
  1008. copy = max - skb->len;
  1009. }
  1010. if (copy <= 0) {
  1011. new_segment:
  1012. /* Allocate new segment. If the interface is SG,
  1013. * allocate skb fitting to single page.
  1014. */
  1015. if (!sk_stream_memory_free(sk))
  1016. goto wait_for_sndbuf;
  1017. skb = sk_stream_alloc_skb(sk,
  1018. select_size(sk, sg),
  1019. sk->sk_allocation);
  1020. if (!skb)
  1021. goto wait_for_memory;
  1022. /*
  1023. * Check whether we can use HW checksum.
  1024. */
  1025. if (sk->sk_route_caps & NETIF_F_ALL_CSUM)
  1026. skb->ip_summed = CHECKSUM_PARTIAL;
  1027. skb_entail(sk, skb);
  1028. copy = size_goal;
  1029. max = size_goal;
  1030. /* All packets are restored as if they have
  1031. * already been sent. skb_mstamp isn't set to
  1032. * avoid wrong rtt estimation.
  1033. */
  1034. if (tp->repair)
  1035. TCP_SKB_CB(skb)->sacked |= TCPCB_REPAIRED;
  1036. }
  1037. /* Try to append data to the end of skb. */
  1038. if (copy > seglen)
  1039. copy = seglen;
  1040. /* Where to copy to? */
  1041. if (skb_availroom(skb) > 0) {
  1042. /* We have some space in skb head. Superb! */
  1043. copy = min_t(int, copy, skb_availroom(skb));
  1044. err = skb_add_data_nocache(sk, skb, from, copy);
  1045. if (err)
  1046. goto do_fault;
  1047. } else {
  1048. bool merge = true;
  1049. int i = skb_shinfo(skb)->nr_frags;
  1050. struct page_frag *pfrag = sk_page_frag(sk);
  1051. if (!sk_page_frag_refill(sk, pfrag))
  1052. goto wait_for_memory;
  1053. if (!skb_can_coalesce(skb, i, pfrag->page,
  1054. pfrag->offset)) {
  1055. if (i == MAX_SKB_FRAGS || !sg) {
  1056. tcp_mark_push(tp, skb);
  1057. goto new_segment;
  1058. }
  1059. merge = false;
  1060. }
  1061. copy = min_t(int, copy, pfrag->size - pfrag->offset);
  1062. if (!sk_wmem_schedule(sk, copy))
  1063. goto wait_for_memory;
  1064. err = skb_copy_to_page_nocache(sk, from, skb,
  1065. pfrag->page,
  1066. pfrag->offset,
  1067. copy);
  1068. if (err)
  1069. goto do_error;
  1070. /* Update the skb. */
  1071. if (merge) {
  1072. skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
  1073. } else {
  1074. skb_fill_page_desc(skb, i, pfrag->page,
  1075. pfrag->offset, copy);
  1076. get_page(pfrag->page);
  1077. }
  1078. pfrag->offset += copy;
  1079. }
  1080. if (!copied)
  1081. TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH;
  1082. tp->write_seq += copy;
  1083. TCP_SKB_CB(skb)->end_seq += copy;
  1084. tcp_skb_pcount_set(skb, 0);
  1085. from += copy;
  1086. copied += copy;
  1087. if ((seglen -= copy) == 0 && iovlen == 0) {
  1088. tcp_tx_timestamp(sk, skb);
  1089. goto out;
  1090. }
  1091. if (skb->len < max || (flags & MSG_OOB) || unlikely(tp->repair))
  1092. continue;
  1093. if (forced_push(tp)) {
  1094. tcp_mark_push(tp, skb);
  1095. __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH);
  1096. } else if (skb == tcp_send_head(sk))
  1097. tcp_push_one(sk, mss_now);
  1098. continue;
  1099. wait_for_sndbuf:
  1100. set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
  1101. wait_for_memory:
  1102. if (copied)
  1103. tcp_push(sk, flags & ~MSG_MORE, mss_now,
  1104. TCP_NAGLE_PUSH, size_goal);
  1105. if ((err = sk_stream_wait_memory(sk, &timeo)) != 0)
  1106. goto do_error;
  1107. mss_now = tcp_send_mss(sk, &size_goal, flags);
  1108. }
  1109. }
  1110. out:
  1111. if (copied)
  1112. tcp_push(sk, flags, mss_now, tp->nonagle, size_goal);
  1113. out_nopush:
  1114. release_sock(sk);
  1115. return copied + copied_syn;
  1116. do_fault:
  1117. if (!skb->len) {
  1118. tcp_unlink_write_queue(skb, sk);
  1119. /* It is the one place in all of TCP, except connection
  1120. * reset, where we can be unlinking the send_head.
  1121. */
  1122. tcp_check_send_head(sk, skb);
  1123. sk_wmem_free_skb(sk, skb);
  1124. }
  1125. do_error:
  1126. if (copied + copied_syn)
  1127. goto out;
  1128. out_err:
  1129. err = sk_stream_error(sk, flags, err);
  1130. release_sock(sk);
  1131. return err;
  1132. }
  1133. EXPORT_SYMBOL(tcp_sendmsg);
  1134. /*
  1135. * Handle reading urgent data. BSD has very simple semantics for
  1136. * this, no blocking and very strange errors 8)
  1137. */
  1138. static int tcp_recv_urg(struct sock *sk, struct msghdr *msg, int len, int flags)
  1139. {
  1140. struct tcp_sock *tp = tcp_sk(sk);
  1141. /* No URG data to read. */
  1142. if (sock_flag(sk, SOCK_URGINLINE) || !tp->urg_data ||
  1143. tp->urg_data == TCP_URG_READ)
  1144. return -EINVAL; /* Yes this is right ! */
  1145. if (sk->sk_state == TCP_CLOSE && !sock_flag(sk, SOCK_DONE))
  1146. return -ENOTCONN;
  1147. if (tp->urg_data & TCP_URG_VALID) {
  1148. int err = 0;
  1149. char c = tp->urg_data;
  1150. if (!(flags & MSG_PEEK))
  1151. tp->urg_data = TCP_URG_READ;
  1152. /* Read urgent data. */
  1153. msg->msg_flags |= MSG_OOB;
  1154. if (len > 0) {
  1155. if (!(flags & MSG_TRUNC))
  1156. err = memcpy_to_msg(msg, &c, 1);
  1157. len = 1;
  1158. } else
  1159. msg->msg_flags |= MSG_TRUNC;
  1160. return err ? -EFAULT : len;
  1161. }
  1162. if (sk->sk_state == TCP_CLOSE || (sk->sk_shutdown & RCV_SHUTDOWN))
  1163. return 0;
  1164. /* Fixed the recv(..., MSG_OOB) behaviour. BSD docs and
  1165. * the available implementations agree in this case:
  1166. * this call should never block, independent of the
  1167. * blocking state of the socket.
  1168. * Mike <pall@rz.uni-karlsruhe.de>
  1169. */
  1170. return -EAGAIN;
  1171. }
  1172. static int tcp_peek_sndq(struct sock *sk, struct msghdr *msg, int len)
  1173. {
  1174. struct sk_buff *skb;
  1175. int copied = 0, err = 0;
  1176. /* XXX -- need to support SO_PEEK_OFF */
  1177. skb_queue_walk(&sk->sk_write_queue, skb) {
  1178. err = skb_copy_datagram_msg(skb, 0, msg, skb->len);
  1179. if (err)
  1180. break;
  1181. copied += skb->len;
  1182. }
  1183. return err ?: copied;
  1184. }
  1185. /* Clean up the receive buffer for full frames taken by the user,
  1186. * then send an ACK if necessary. COPIED is the number of bytes
  1187. * tcp_recvmsg has given to the user so far, it speeds up the
  1188. * calculation of whether or not we must ACK for the sake of
  1189. * a window update.
  1190. */
  1191. static void tcp_cleanup_rbuf(struct sock *sk, int copied)
  1192. {
  1193. struct tcp_sock *tp = tcp_sk(sk);
  1194. bool time_to_ack = false;
  1195. struct sk_buff *skb = skb_peek(&sk->sk_receive_queue);
  1196. WARN(skb && !before(tp->copied_seq, TCP_SKB_CB(skb)->end_seq),
  1197. "cleanup rbuf bug: copied %X seq %X rcvnxt %X\n",
  1198. tp->copied_seq, TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt);
  1199. if (inet_csk_ack_scheduled(sk)) {
  1200. const struct inet_connection_sock *icsk = inet_csk(sk);
  1201. /* Delayed ACKs frequently hit locked sockets during bulk
  1202. * receive. */
  1203. if (icsk->icsk_ack.blocked ||
  1204. /* Once-per-two-segments ACK was not sent by tcp_input.c */
  1205. tp->rcv_nxt - tp->rcv_wup > icsk->icsk_ack.rcv_mss ||
  1206. /*
  1207. * If this read emptied read buffer, we send ACK, if
  1208. * connection is not bidirectional, user drained
  1209. * receive buffer and there was a small segment
  1210. * in queue.
  1211. */
  1212. (copied > 0 &&
  1213. ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED2) ||
  1214. ((icsk->icsk_ack.pending & ICSK_ACK_PUSHED) &&
  1215. !icsk->icsk_ack.pingpong)) &&
  1216. !atomic_read(&sk->sk_rmem_alloc)))
  1217. time_to_ack = true;
  1218. }
  1219. /* We send an ACK if we can now advertise a non-zero window
  1220. * which has been raised "significantly".
  1221. *
  1222. * Even if window raised up to infinity, do not send window open ACK
  1223. * in states, where we will not receive more. It is useless.
  1224. */
  1225. if (copied > 0 && !time_to_ack && !(sk->sk_shutdown & RCV_SHUTDOWN)) {
  1226. __u32 rcv_window_now = tcp_receive_window(tp);
  1227. /* Optimize, __tcp_select_window() is not cheap. */
  1228. if (2*rcv_window_now <= tp->window_clamp) {
  1229. __u32 new_window = __tcp_select_window(sk);
  1230. /* Send ACK now, if this read freed lots of space
  1231. * in our buffer. Certainly, new_window is new window.
  1232. * We can advertise it now, if it is not less than current one.
  1233. * "Lots" means "at least twice" here.
  1234. */
  1235. if (new_window && new_window >= 2 * rcv_window_now)
  1236. time_to_ack = true;
  1237. }
  1238. }
  1239. if (time_to_ack)
  1240. tcp_send_ack(sk);
  1241. }
  1242. static void tcp_prequeue_process(struct sock *sk)
  1243. {
  1244. struct sk_buff *skb;
  1245. struct tcp_sock *tp = tcp_sk(sk);
  1246. NET_INC_STATS_USER(sock_net(sk), LINUX_MIB_TCPPREQUEUED);
  1247. /* RX process wants to run with disabled BHs, though it is not
  1248. * necessary */
  1249. local_bh_disable();
  1250. while ((skb = __skb_dequeue(&tp->ucopy.prequeue)) != NULL)
  1251. sk_backlog_rcv(sk, skb);
  1252. local_bh_enable();
  1253. /* Clear memory counter. */
  1254. tp->ucopy.memory = 0;
  1255. }
  1256. static struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off)
  1257. {
  1258. struct sk_buff *skb;
  1259. u32 offset;
  1260. while ((skb = skb_peek(&sk->sk_receive_queue)) != NULL) {
  1261. offset = seq - TCP_SKB_CB(skb)->seq;
  1262. if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)
  1263. offset--;
  1264. if (offset < skb->len || (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)) {
  1265. *off = offset;
  1266. return skb;
  1267. }
  1268. /* This looks weird, but this can happen if TCP collapsing
  1269. * splitted a fat GRO packet, while we released socket lock
  1270. * in skb_splice_bits()
  1271. */
  1272. sk_eat_skb(sk, skb);
  1273. }
  1274. return NULL;
  1275. }
  1276. /*
  1277. * This routine provides an alternative to tcp_recvmsg() for routines
  1278. * that would like to handle copying from skbuffs directly in 'sendfile'
  1279. * fashion.
  1280. * Note:
  1281. * - It is assumed that the socket was locked by the caller.
  1282. * - The routine does not block.
  1283. * - At present, there is no support for reading OOB data
  1284. * or for 'peeking' the socket using this routine
  1285. * (although both would be easy to implement).
  1286. */
  1287. int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
  1288. sk_read_actor_t recv_actor)
  1289. {
  1290. struct sk_buff *skb;
  1291. struct tcp_sock *tp = tcp_sk(sk);
  1292. u32 seq = tp->copied_seq;
  1293. u32 offset;
  1294. int copied = 0;
  1295. if (sk->sk_state == TCP_LISTEN)
  1296. return -ENOTCONN;
  1297. while ((skb = tcp_recv_skb(sk, seq, &offset)) != NULL) {
  1298. if (offset < skb->len) {
  1299. int used;
  1300. size_t len;
  1301. len = skb->len - offset;
  1302. /* Stop reading if we hit a patch of urgent data */
  1303. if (tp->urg_data) {
  1304. u32 urg_offset = tp->urg_seq - seq;
  1305. if (urg_offset < len)
  1306. len = urg_offset;
  1307. if (!len)
  1308. break;
  1309. }
  1310. used = recv_actor(desc, skb, offset, len);
  1311. if (used <= 0) {
  1312. if (!copied)
  1313. copied = used;
  1314. break;
  1315. } else if (used <= len) {
  1316. seq += used;
  1317. copied += used;
  1318. offset += used;
  1319. }
  1320. /* If recv_actor drops the lock (e.g. TCP splice
  1321. * receive) the skb pointer might be invalid when
  1322. * getting here: tcp_collapse might have deleted it
  1323. * while aggregating skbs from the socket queue.
  1324. */
  1325. skb = tcp_recv_skb(sk, seq - 1, &offset);
  1326. if (!skb)
  1327. break;
  1328. /* TCP coalescing might have appended data to the skb.
  1329. * Try to splice more frags
  1330. */
  1331. if (offset + 1 != skb->len)
  1332. continue;
  1333. }
  1334. if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) {
  1335. sk_eat_skb(sk, skb);
  1336. ++seq;
  1337. break;
  1338. }
  1339. sk_eat_skb(sk, skb);
  1340. if (!desc->count)
  1341. break;
  1342. tp->copied_seq = seq;
  1343. }
  1344. tp->copied_seq = seq;
  1345. tcp_rcv_space_adjust(sk);
  1346. /* Clean up data we have read: This will do ACK frames. */
  1347. if (copied > 0) {
  1348. tcp_recv_skb(sk, seq, &offset);
  1349. tcp_cleanup_rbuf(sk, copied);
  1350. }
  1351. return copied;
  1352. }
  1353. EXPORT_SYMBOL(tcp_read_sock);
  1354. /*
  1355. * This routine copies from a sock struct into the user buffer.
  1356. *
  1357. * Technical note: in 2.3 we work on _locked_ socket, so that
  1358. * tricks with *seq access order and skb->users are not required.
  1359. * Probably, code can be easily improved even more.
  1360. */
  1361. int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
  1362. size_t len, int nonblock, int flags, int *addr_len)
  1363. {
  1364. struct tcp_sock *tp = tcp_sk(sk);
  1365. int copied = 0;
  1366. u32 peek_seq;
  1367. u32 *seq;
  1368. unsigned long used;
  1369. int err;
  1370. int target; /* Read at least this many bytes */
  1371. long timeo;
  1372. struct task_struct *user_recv = NULL;
  1373. struct sk_buff *skb;
  1374. u32 urg_hole = 0;
  1375. if (unlikely(flags & MSG_ERRQUEUE))
  1376. return inet_recv_error(sk, msg, len, addr_len);
  1377. if (sk_can_busy_loop(sk) && skb_queue_empty(&sk->sk_receive_queue) &&
  1378. (sk->sk_state == TCP_ESTABLISHED))
  1379. sk_busy_loop(sk, nonblock);
  1380. lock_sock(sk);
  1381. err = -ENOTCONN;
  1382. if (sk->sk_state == TCP_LISTEN)
  1383. goto out;
  1384. timeo = sock_rcvtimeo(sk, nonblock);
  1385. /* Urgent data needs to be handled specially. */
  1386. if (flags & MSG_OOB)
  1387. goto recv_urg;
  1388. if (unlikely(tp->repair)) {
  1389. err = -EPERM;
  1390. if (!(flags & MSG_PEEK))
  1391. goto out;
  1392. if (tp->repair_queue == TCP_SEND_QUEUE)
  1393. goto recv_sndq;
  1394. err = -EINVAL;
  1395. if (tp->repair_queue == TCP_NO_QUEUE)
  1396. goto out;
  1397. /* 'common' recv queue MSG_PEEK-ing */
  1398. }
  1399. seq = &tp->copied_seq;
  1400. if (flags & MSG_PEEK) {
  1401. peek_seq = tp->copied_seq;
  1402. seq = &peek_seq;
  1403. }
  1404. target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
  1405. do {
  1406. u32 offset;
  1407. /* Are we at urgent data? Stop if we have read anything or have SIGURG pending. */
  1408. if (tp->urg_data && tp->urg_seq == *seq) {
  1409. if (copied)
  1410. break;
  1411. if (signal_pending(current)) {
  1412. copied = timeo ? sock_intr_errno(timeo) : -EAGAIN;
  1413. break;
  1414. }
  1415. }
  1416. /* Next get a buffer. */
  1417. skb_queue_walk(&sk->sk_receive_queue, skb) {
  1418. /* Now that we have two receive queues this
  1419. * shouldn't happen.
  1420. */
  1421. if (WARN(before(*seq, TCP_SKB_CB(skb)->seq),
  1422. "recvmsg bug: copied %X seq %X rcvnxt %X fl %X\n",
  1423. *seq, TCP_SKB_CB(skb)->seq, tp->rcv_nxt,
  1424. flags))
  1425. break;
  1426. offset = *seq - TCP_SKB_CB(skb)->seq;
  1427. if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)
  1428. offset--;
  1429. if (offset < skb->len)
  1430. goto found_ok_skb;
  1431. if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)
  1432. goto found_fin_ok;
  1433. WARN(!(flags & MSG_PEEK),
  1434. "recvmsg bug 2: copied %X seq %X rcvnxt %X fl %X\n",
  1435. *seq, TCP_SKB_CB(skb)->seq, tp->rcv_nxt, flags);
  1436. }
  1437. /* Well, if we have backlog, try to process it now yet. */
  1438. if (copied >= target && !sk->sk_backlog.tail)
  1439. break;
  1440. if (copied) {
  1441. if (sk->sk_err ||
  1442. sk->sk_state == TCP_CLOSE ||
  1443. (sk->sk_shutdown & RCV_SHUTDOWN) ||
  1444. !timeo ||
  1445. signal_pending(current))
  1446. break;
  1447. } else {
  1448. if (sock_flag(sk, SOCK_DONE))
  1449. break;
  1450. if (sk->sk_err) {
  1451. copied = sock_error(sk);
  1452. break;
  1453. }
  1454. if (sk->sk_shutdown & RCV_SHUTDOWN)
  1455. break;
  1456. if (sk->sk_state == TCP_CLOSE) {
  1457. if (!sock_flag(sk, SOCK_DONE)) {
  1458. /* This occurs when user tries to read
  1459. * from never connected socket.
  1460. */
  1461. copied = -ENOTCONN;
  1462. break;
  1463. }
  1464. break;
  1465. }
  1466. if (!timeo) {
  1467. copied = -EAGAIN;
  1468. break;
  1469. }
  1470. if (signal_pending(current)) {
  1471. copied = sock_intr_errno(timeo);
  1472. break;
  1473. }
  1474. }
  1475. tcp_cleanup_rbuf(sk, copied);
  1476. if (!sysctl_tcp_low_latency && tp->ucopy.task == user_recv) {
  1477. /* Install new reader */
  1478. if (!user_recv && !(flags & (MSG_TRUNC | MSG_PEEK))) {
  1479. user_recv = current;
  1480. tp->ucopy.task = user_recv;
  1481. tp->ucopy.msg = msg;
  1482. }
  1483. tp->ucopy.len = len;
  1484. WARN_ON(tp->copied_seq != tp->rcv_nxt &&
  1485. !(flags & (MSG_PEEK | MSG_TRUNC)));
  1486. /* Ugly... If prequeue is not empty, we have to
  1487. * process it before releasing socket, otherwise
  1488. * order will be broken at second iteration.
  1489. * More elegant solution is required!!!
  1490. *
  1491. * Look: we have the following (pseudo)queues:
  1492. *
  1493. * 1. packets in flight
  1494. * 2. backlog
  1495. * 3. prequeue
  1496. * 4. receive_queue
  1497. *
  1498. * Each queue can be processed only if the next ones
  1499. * are empty. At this point we have empty receive_queue.
  1500. * But prequeue _can_ be not empty after 2nd iteration,
  1501. * when we jumped to start of loop because backlog
  1502. * processing added something to receive_queue.
  1503. * We cannot release_sock(), because backlog contains
  1504. * packets arrived _after_ prequeued ones.
  1505. *
  1506. * Shortly, algorithm is clear --- to process all
  1507. * the queues in order. We could make it more directly,
  1508. * requeueing packets from backlog to prequeue, if
  1509. * is not empty. It is more elegant, but eats cycles,
  1510. * unfortunately.
  1511. */
  1512. if (!skb_queue_empty(&tp->ucopy.prequeue))
  1513. goto do_prequeue;
  1514. /* __ Set realtime policy in scheduler __ */
  1515. }
  1516. if (copied >= target) {
  1517. /* Do not sleep, just process backlog. */
  1518. release_sock(sk);
  1519. lock_sock(sk);
  1520. } else
  1521. sk_wait_data(sk, &timeo);
  1522. if (user_recv) {
  1523. int chunk;
  1524. /* __ Restore normal policy in scheduler __ */
  1525. if ((chunk = len - tp->ucopy.len) != 0) {
  1526. NET_ADD_STATS_USER(sock_net(sk), LINUX_MIB_TCPDIRECTCOPYFROMBACKLOG, chunk);
  1527. len -= chunk;
  1528. copied += chunk;
  1529. }
  1530. if (tp->rcv_nxt == tp->copied_seq &&
  1531. !skb_queue_empty(&tp->ucopy.prequeue)) {
  1532. do_prequeue:
  1533. tcp_prequeue_process(sk);
  1534. if ((chunk = len - tp->ucopy.len) != 0) {
  1535. NET_ADD_STATS_USER(sock_net(sk), LINUX_MIB_TCPDIRECTCOPYFROMPREQUEUE, chunk);
  1536. len -= chunk;
  1537. copied += chunk;
  1538. }
  1539. }
  1540. }
  1541. if ((flags & MSG_PEEK) &&
  1542. (peek_seq - copied - urg_hole != tp->copied_seq)) {
  1543. net_dbg_ratelimited("TCP(%s:%d): Application bug, race in MSG_PEEK\n",
  1544. current->comm,
  1545. task_pid_nr(current));
  1546. peek_seq = tp->copied_seq;
  1547. }
  1548. continue;
  1549. found_ok_skb:
  1550. /* Ok so how much can we use? */
  1551. used = skb->len - offset;
  1552. if (len < used)
  1553. used = len;
  1554. /* Do we have urgent data here? */
  1555. if (tp->urg_data) {
  1556. u32 urg_offset = tp->urg_seq - *seq;
  1557. if (urg_offset < used) {
  1558. if (!urg_offset) {
  1559. if (!sock_flag(sk, SOCK_URGINLINE)) {
  1560. ++*seq;
  1561. urg_hole++;
  1562. offset++;
  1563. used--;
  1564. if (!used)
  1565. goto skip_copy;
  1566. }
  1567. } else
  1568. used = urg_offset;
  1569. }
  1570. }
  1571. if (!(flags & MSG_TRUNC)) {
  1572. err = skb_copy_datagram_msg(skb, offset, msg, used);
  1573. if (err) {
  1574. /* Exception. Bailout! */
  1575. if (!copied)
  1576. copied = -EFAULT;
  1577. break;
  1578. }
  1579. }
  1580. *seq += used;
  1581. copied += used;
  1582. len -= used;
  1583. tcp_rcv_space_adjust(sk);
  1584. skip_copy:
  1585. if (tp->urg_data && after(tp->copied_seq, tp->urg_seq)) {
  1586. tp->urg_data = 0;
  1587. tcp_fast_path_check(sk);
  1588. }
  1589. if (used + offset < skb->len)
  1590. continue;
  1591. if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)
  1592. goto found_fin_ok;
  1593. if (!(flags & MSG_PEEK))
  1594. sk_eat_skb(sk, skb);
  1595. continue;
  1596. found_fin_ok:
  1597. /* Process the FIN. */
  1598. ++*seq;
  1599. if (!(flags & MSG_PEEK))
  1600. sk_eat_skb(sk, skb);
  1601. break;
  1602. } while (len > 0);
  1603. if (user_recv) {
  1604. if (!skb_queue_empty(&tp->ucopy.prequeue)) {
  1605. int chunk;
  1606. tp->ucopy.len = copied > 0 ? len : 0;
  1607. tcp_prequeue_process(sk);
  1608. if (copied > 0 && (chunk = len - tp->ucopy.len) != 0) {
  1609. NET_ADD_STATS_USER(sock_net(sk), LINUX_MIB_TCPDIRECTCOPYFROMPREQUEUE, chunk);
  1610. len -= chunk;
  1611. copied += chunk;
  1612. }
  1613. }
  1614. tp->ucopy.task = NULL;
  1615. tp->ucopy.len = 0;
  1616. }
  1617. /* According to UNIX98, msg_name/msg_namelen are ignored
  1618. * on connected socket. I was just happy when found this 8) --ANK
  1619. */
  1620. /* Clean up data we have read: This will do ACK frames. */
  1621. tcp_cleanup_rbuf(sk, copied);
  1622. release_sock(sk);
  1623. return copied;
  1624. out:
  1625. release_sock(sk);
  1626. return err;
  1627. recv_urg:
  1628. err = tcp_recv_urg(sk, msg, len, flags);
  1629. goto out;
  1630. recv_sndq:
  1631. err = tcp_peek_sndq(sk, msg, len);
  1632. goto out;
  1633. }
  1634. EXPORT_SYMBOL(tcp_recvmsg);
  1635. void tcp_set_state(struct sock *sk, int state)
  1636. {
  1637. int oldstate = sk->sk_state;
  1638. switch (state) {
  1639. case TCP_ESTABLISHED:
  1640. if (oldstate != TCP_ESTABLISHED)
  1641. TCP_INC_STATS(sock_net(sk), TCP_MIB_CURRESTAB);
  1642. break;
  1643. case TCP_CLOSE:
  1644. if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
  1645. TCP_INC_STATS(sock_net(sk), TCP_MIB_ESTABRESETS);
  1646. sk->sk_prot->unhash(sk);
  1647. if (inet_csk(sk)->icsk_bind_hash &&
  1648. !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
  1649. inet_put_port(sk);
  1650. /* fall through */
  1651. default:
  1652. if (oldstate == TCP_ESTABLISHED)
  1653. TCP_DEC_STATS(sock_net(sk), TCP_MIB_CURRESTAB);
  1654. }
  1655. /* Change state AFTER socket is unhashed to avoid closed
  1656. * socket sitting in hash tables.
  1657. */
  1658. sk->sk_state = state;
  1659. #ifdef STATE_TRACE
  1660. SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n", sk, statename[oldstate], statename[state]);
  1661. #endif
  1662. }
  1663. EXPORT_SYMBOL_GPL(tcp_set_state);
  1664. /*
  1665. * State processing on a close. This implements the state shift for
  1666. * sending our FIN frame. Note that we only send a FIN for some
  1667. * states. A shutdown() may have already sent the FIN, or we may be
  1668. * closed.
  1669. */
  1670. static const unsigned char new_state[16] = {
  1671. /* current state: new state: action: */
  1672. /* (Invalid) */ TCP_CLOSE,
  1673. /* TCP_ESTABLISHED */ TCP_FIN_WAIT1 | TCP_ACTION_FIN,
  1674. /* TCP_SYN_SENT */ TCP_CLOSE,
  1675. /* TCP_SYN_RECV */ TCP_FIN_WAIT1 | TCP_ACTION_FIN,
  1676. /* TCP_FIN_WAIT1 */ TCP_FIN_WAIT1,
  1677. /* TCP_FIN_WAIT2 */ TCP_FIN_WAIT2,
  1678. /* TCP_TIME_WAIT */ TCP_CLOSE,
  1679. /* TCP_CLOSE */ TCP_CLOSE,
  1680. /* TCP_CLOSE_WAIT */ TCP_LAST_ACK | TCP_ACTION_FIN,
  1681. /* TCP_LAST_ACK */ TCP_LAST_ACK,
  1682. /* TCP_LISTEN */ TCP_CLOSE,
  1683. /* TCP_CLOSING */ TCP_CLOSING,
  1684. };
  1685. static int tcp_close_state(struct sock *sk)
  1686. {
  1687. int next = (int)new_state[sk->sk_state];
  1688. int ns = next & TCP_STATE_MASK;
  1689. tcp_set_state(sk, ns);
  1690. return next & TCP_ACTION_FIN;
  1691. }
  1692. /*
  1693. * Shutdown the sending side of a connection. Much like close except
  1694. * that we don't receive shut down or sock_set_flag(sk, SOCK_DEAD).
  1695. */
  1696. void tcp_shutdown(struct sock *sk, int how)
  1697. {
  1698. /* We need to grab some memory, and put together a FIN,
  1699. * and then put it into the queue to be sent.
  1700. * Tim MacKenzie(tym@dibbler.cs.monash.edu.au) 4 Dec '92.
  1701. */
  1702. if (!(how & SEND_SHUTDOWN))
  1703. return;
  1704. /* If we've already sent a FIN, or it's a closed state, skip this. */
  1705. if ((1 << sk->sk_state) &
  1706. (TCPF_ESTABLISHED | TCPF_SYN_SENT |
  1707. TCPF_SYN_RECV | TCPF_CLOSE_WAIT)) {
  1708. /* Clear out any half completed packets. FIN if needed. */
  1709. if (tcp_close_state(sk))
  1710. tcp_send_fin(sk);
  1711. }
  1712. }
  1713. EXPORT_SYMBOL(tcp_shutdown);
  1714. bool tcp_check_oom(struct sock *sk, int shift)
  1715. {
  1716. bool too_many_orphans, out_of_socket_memory;
  1717. too_many_orphans = tcp_too_many_orphans(sk, shift);
  1718. out_of_socket_memory = tcp_out_of_memory(sk);
  1719. if (too_many_orphans)
  1720. net_info_ratelimited("too many orphaned sockets\n");
  1721. if (out_of_socket_memory)
  1722. net_info_ratelimited("out of memory -- consider tuning tcp_mem\n");
  1723. return too_many_orphans || out_of_socket_memory;
  1724. }
  1725. void tcp_close(struct sock *sk, long timeout)
  1726. {
  1727. struct sk_buff *skb;
  1728. int data_was_unread = 0;
  1729. int state;
  1730. lock_sock(sk);
  1731. sk->sk_shutdown = SHUTDOWN_MASK;
  1732. if (sk->sk_state == TCP_LISTEN) {
  1733. tcp_set_state(sk, TCP_CLOSE);
  1734. /* Special case. */
  1735. inet_csk_listen_stop(sk);
  1736. goto adjudge_to_death;
  1737. }
  1738. /* We need to flush the recv. buffs. We do this only on the
  1739. * descriptor close, not protocol-sourced closes, because the
  1740. * reader process may not have drained the data yet!
  1741. */
  1742. while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) {
  1743. u32 len = TCP_SKB_CB(skb)->end_seq - TCP_SKB_CB(skb)->seq;
  1744. if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)
  1745. len--;
  1746. data_was_unread += len;
  1747. __kfree_skb(skb);
  1748. }
  1749. sk_mem_reclaim(sk);
  1750. /* If socket has been already reset (e.g. in tcp_reset()) - kill it. */
  1751. if (sk->sk_state == TCP_CLOSE)
  1752. goto adjudge_to_death;
  1753. /* As outlined in RFC 2525, section 2.17, we send a RST here because
  1754. * data was lost. To witness the awful effects of the old behavior of
  1755. * always doing a FIN, run an older 2.1.x kernel or 2.0.x, start a bulk
  1756. * GET in an FTP client, suspend the process, wait for the client to
  1757. * advertise a zero window, then kill -9 the FTP client, wheee...
  1758. * Note: timeout is always zero in such a case.
  1759. */
  1760. if (unlikely(tcp_sk(sk)->repair)) {
  1761. sk->sk_prot->disconnect(sk, 0);
  1762. } else if (data_was_unread) {
  1763. /* Unread data was tossed, zap the connection. */
  1764. NET_INC_STATS_USER(sock_net(sk), LINUX_MIB_TCPABORTONCLOSE);
  1765. tcp_set_state(sk, TCP_CLOSE);
  1766. tcp_send_active_reset(sk, sk->sk_allocation);
  1767. } else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) {
  1768. /* Check zero linger _after_ checking for unread data. */
  1769. sk->sk_prot->disconnect(sk, 0);
  1770. NET_INC_STATS_USER(sock_net(sk), LINUX_MIB_TCPABORTONDATA);
  1771. } else if (tcp_close_state(sk)) {
  1772. /* We FIN if the application ate all the data before
  1773. * zapping the connection.
  1774. */
  1775. /* RED-PEN. Formally speaking, we have broken TCP state
  1776. * machine. State transitions:
  1777. *
  1778. * TCP_ESTABLISHED -> TCP_FIN_WAIT1
  1779. * TCP_SYN_RECV -> TCP_FIN_WAIT1 (forget it, it's impossible)
  1780. * TCP_CLOSE_WAIT -> TCP_LAST_ACK
  1781. *
  1782. * are legal only when FIN has been sent (i.e. in window),
  1783. * rather than queued out of window. Purists blame.
  1784. *
  1785. * F.e. "RFC state" is ESTABLISHED,
  1786. * if Linux state is FIN-WAIT-1, but FIN is still not sent.
  1787. *
  1788. * The visible declinations are that sometimes
  1789. * we enter time-wait state, when it is not required really
  1790. * (harmless), do not send active resets, when they are
  1791. * required by specs (TCP_ESTABLISHED, TCP_CLOSE_WAIT, when
  1792. * they look as CLOSING or LAST_ACK for Linux)
  1793. * Probably, I missed some more holelets.
  1794. * --ANK
  1795. * XXX (TFO) - To start off we don't support SYN+ACK+FIN
  1796. * in a single packet! (May consider it later but will
  1797. * probably need API support or TCP_CORK SYN-ACK until
  1798. * data is written and socket is closed.)
  1799. */
  1800. tcp_send_fin(sk);
  1801. }
  1802. sk_stream_wait_close(sk, timeout);
  1803. adjudge_to_death:
  1804. state = sk->sk_state;
  1805. sock_hold(sk);
  1806. sock_orphan(sk);
  1807. /* It is the last release_sock in its life. It will remove backlog. */
  1808. release_sock(sk);
  1809. /* Now socket is owned by kernel and we acquire BH lock
  1810. to finish close. No need to check for user refs.
  1811. */
  1812. local_bh_disable();
  1813. bh_lock_sock(sk);
  1814. WARN_ON(sock_owned_by_user(sk));
  1815. percpu_counter_inc(sk->sk_prot->orphan_count);
  1816. /* Have we already been destroyed by a softirq or backlog? */
  1817. if (state != TCP_CLOSE && sk->sk_state == TCP_CLOSE)
  1818. goto out;
  1819. /* This is a (useful) BSD violating of the RFC. There is a
  1820. * problem with TCP as specified in that the other end could
  1821. * keep a socket open forever with no application left this end.
  1822. * We use a 1 minute timeout (about the same as BSD) then kill
  1823. * our end. If they send after that then tough - BUT: long enough
  1824. * that we won't make the old 4*rto = almost no time - whoops
  1825. * reset mistake.
  1826. *
  1827. * Nope, it was not mistake. It is really desired behaviour
  1828. * f.e. on http servers, when such sockets are useless, but
  1829. * consume significant resources. Let's do it with special
  1830. * linger2 option. --ANK
  1831. */
  1832. if (sk->sk_state == TCP_FIN_WAIT2) {
  1833. struct tcp_sock *tp = tcp_sk(sk);
  1834. if (tp->linger2 < 0) {
  1835. tcp_set_state(sk, TCP_CLOSE);
  1836. tcp_send_active_reset(sk, GFP_ATOMIC);
  1837. NET_INC_STATS_BH(sock_net(sk),
  1838. LINUX_MIB_TCPABORTONLINGER);
  1839. } else {
  1840. const int tmo = tcp_fin_time(sk);
  1841. if (tmo > TCP_TIMEWAIT_LEN) {
  1842. inet_csk_reset_keepalive_timer(sk,
  1843. tmo - TCP_TIMEWAIT_LEN);
  1844. } else {
  1845. tcp_time_wait(sk, TCP_FIN_WAIT2, tmo);
  1846. goto out;
  1847. }
  1848. }
  1849. }
  1850. if (sk->sk_state != TCP_CLOSE) {
  1851. sk_mem_reclaim(sk);
  1852. if (tcp_check_oom(sk, 0)) {
  1853. tcp_set_state(sk, TCP_CLOSE);
  1854. tcp_send_active_reset(sk, GFP_ATOMIC);
  1855. NET_INC_STATS_BH(sock_net(sk),
  1856. LINUX_MIB_TCPABORTONMEMORY);
  1857. }
  1858. }
  1859. if (sk->sk_state == TCP_CLOSE) {
  1860. struct request_sock *req = tcp_sk(sk)->fastopen_rsk;
  1861. /* We could get here with a non-NULL req if the socket is
  1862. * aborted (e.g., closed with unread data) before 3WHS
  1863. * finishes.
  1864. */
  1865. if (req != NULL)
  1866. reqsk_fastopen_remove(sk, req, false);
  1867. inet_csk_destroy_sock(sk);
  1868. }
  1869. /* Otherwise, socket is reprieved until protocol close. */
  1870. out:
  1871. bh_unlock_sock(sk);
  1872. local_bh_enable();
  1873. sock_put(sk);
  1874. }
  1875. EXPORT_SYMBOL(tcp_close);
  1876. /* These states need RST on ABORT according to RFC793 */
  1877. static inline bool tcp_need_reset(int state)
  1878. {
  1879. return (1 << state) &
  1880. (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_FIN_WAIT1 |
  1881. TCPF_FIN_WAIT2 | TCPF_SYN_RECV);
  1882. }
  1883. int tcp_disconnect(struct sock *sk, int flags)
  1884. {
  1885. struct inet_sock *inet = inet_sk(sk);
  1886. struct inet_connection_sock *icsk = inet_csk(sk);
  1887. struct tcp_sock *tp = tcp_sk(sk);
  1888. int err = 0;
  1889. int old_state = sk->sk_state;
  1890. if (old_state != TCP_CLOSE)
  1891. tcp_set_state(sk, TCP_CLOSE);
  1892. /* ABORT function of RFC793 */
  1893. if (old_state == TCP_LISTEN) {
  1894. inet_csk_listen_stop(sk);
  1895. } else if (unlikely(tp->repair)) {
  1896. sk->sk_err = ECONNABORTED;
  1897. } else if (tcp_need_reset(old_state) ||
  1898. (tp->snd_nxt != tp->write_seq &&
  1899. (1 << old_state) & (TCPF_CLOSING | TCPF_LAST_ACK))) {
  1900. /* The last check adjusts for discrepancy of Linux wrt. RFC
  1901. * states
  1902. */
  1903. tcp_send_active_reset(sk, gfp_any());
  1904. sk->sk_err = ECONNRESET;
  1905. } else if (old_state == TCP_SYN_SENT)
  1906. sk->sk_err = ECONNRESET;
  1907. tcp_clear_xmit_timers(sk);
  1908. __skb_queue_purge(&sk->sk_receive_queue);
  1909. tcp_write_queue_purge(sk);
  1910. __skb_queue_purge(&tp->out_of_order_queue);
  1911. inet->inet_dport = 0;
  1912. if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
  1913. inet_reset_saddr(sk);
  1914. sk->sk_shutdown = 0;
  1915. sock_reset_flag(sk, SOCK_DONE);
  1916. tp->srtt_us = 0;
  1917. if ((tp->write_seq += tp->max_window + 2) == 0)
  1918. tp->write_seq = 1;
  1919. icsk->icsk_backoff = 0;
  1920. tp->snd_cwnd = 2;
  1921. icsk->icsk_probes_out = 0;
  1922. tp->packets_out = 0;
  1923. tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
  1924. tp->snd_cwnd_cnt = 0;
  1925. tp->window_clamp = 0;
  1926. tcp_set_ca_state(sk, TCP_CA_Open);
  1927. tcp_clear_retrans(tp);
  1928. inet_csk_delack_init(sk);
  1929. tcp_init_send_head(sk);
  1930. memset(&tp->rx_opt, 0, sizeof(tp->rx_opt));
  1931. __sk_dst_reset(sk);
  1932. WARN_ON(inet->inet_num && !icsk->icsk_bind_hash);
  1933. sk->sk_error_report(sk);
  1934. return err;
  1935. }
  1936. EXPORT_SYMBOL(tcp_disconnect);
  1937. void tcp_sock_destruct(struct sock *sk)
  1938. {
  1939. inet_sock_destruct(sk);
  1940. kfree(inet_csk(sk)->icsk_accept_queue.fastopenq);
  1941. }
  1942. static inline bool tcp_can_repair_sock(const struct sock *sk)
  1943. {
  1944. return ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN) &&
  1945. ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_ESTABLISHED));
  1946. }
  1947. static int tcp_repair_options_est(struct tcp_sock *tp,
  1948. struct tcp_repair_opt __user *optbuf, unsigned int len)
  1949. {
  1950. struct tcp_repair_opt opt;
  1951. while (len >= sizeof(opt)) {
  1952. if (copy_from_user(&opt, optbuf, sizeof(opt)))
  1953. return -EFAULT;
  1954. optbuf++;
  1955. len -= sizeof(opt);
  1956. switch (opt.opt_code) {
  1957. case TCPOPT_MSS:
  1958. tp->rx_opt.mss_clamp = opt.opt_val;
  1959. break;
  1960. case TCPOPT_WINDOW:
  1961. {
  1962. u16 snd_wscale = opt.opt_val & 0xFFFF;
  1963. u16 rcv_wscale = opt.opt_val >> 16;
  1964. if (snd_wscale > 14 || rcv_wscale > 14)
  1965. return -EFBIG;
  1966. tp->rx_opt.snd_wscale = snd_wscale;
  1967. tp->rx_opt.rcv_wscale = rcv_wscale;
  1968. tp->rx_opt.wscale_ok = 1;
  1969. }
  1970. break;
  1971. case TCPOPT_SACK_PERM:
  1972. if (opt.opt_val != 0)
  1973. return -EINVAL;
  1974. tp->rx_opt.sack_ok |= TCP_SACK_SEEN;
  1975. if (sysctl_tcp_fack)
  1976. tcp_enable_fack(tp);
  1977. break;
  1978. case TCPOPT_TIMESTAMP:
  1979. if (opt.opt_val != 0)
  1980. return -EINVAL;
  1981. tp->rx_opt.tstamp_ok = 1;
  1982. break;
  1983. }
  1984. }
  1985. return 0;
  1986. }
  1987. /*
  1988. * Socket option code for TCP.
  1989. */
  1990. static int do_tcp_setsockopt(struct sock *sk, int level,
  1991. int optname, char __user *optval, unsigned int optlen)
  1992. {
  1993. struct tcp_sock *tp = tcp_sk(sk);
  1994. struct inet_connection_sock *icsk = inet_csk(sk);
  1995. int val;
  1996. int err = 0;
  1997. /* These are data/string values, all the others are ints */
  1998. switch (optname) {
  1999. case TCP_CONGESTION: {
  2000. char name[TCP_CA_NAME_MAX];
  2001. if (optlen < 1)
  2002. return -EINVAL;
  2003. val = strncpy_from_user(name, optval,
  2004. min_t(long, TCP_CA_NAME_MAX-1, optlen));
  2005. if (val < 0)
  2006. return -EFAULT;
  2007. name[val] = 0;
  2008. lock_sock(sk);
  2009. err = tcp_set_congestion_control(sk, name);
  2010. release_sock(sk);
  2011. return err;
  2012. }
  2013. default:
  2014. /* fallthru */
  2015. break;
  2016. }
  2017. if (optlen < sizeof(int))
  2018. return -EINVAL;
  2019. if (get_user(val, (int __user *)optval))
  2020. return -EFAULT;
  2021. lock_sock(sk);
  2022. switch (optname) {
  2023. case TCP_MAXSEG:
  2024. /* Values greater than interface MTU won't take effect. However
  2025. * at the point when this call is done we typically don't yet
  2026. * know which interface is going to be used */
  2027. if (val < TCP_MIN_MSS || val > MAX_TCP_WINDOW) {
  2028. err = -EINVAL;
  2029. break;
  2030. }
  2031. tp->rx_opt.user_mss = val;
  2032. break;
  2033. case TCP_NODELAY:
  2034. if (val) {
  2035. /* TCP_NODELAY is weaker than TCP_CORK, so that
  2036. * this option on corked socket is remembered, but
  2037. * it is not activated until cork is cleared.
  2038. *
  2039. * However, when TCP_NODELAY is set we make
  2040. * an explicit push, which overrides even TCP_CORK
  2041. * for currently queued segments.
  2042. */
  2043. tp->nonagle |= TCP_NAGLE_OFF|TCP_NAGLE_PUSH;
  2044. tcp_push_pending_frames(sk);
  2045. } else {
  2046. tp->nonagle &= ~TCP_NAGLE_OFF;
  2047. }
  2048. break;
  2049. case TCP_THIN_LINEAR_TIMEOUTS:
  2050. if (val < 0 || val > 1)
  2051. err = -EINVAL;
  2052. else
  2053. tp->thin_lto = val;
  2054. break;
  2055. case TCP_THIN_DUPACK:
  2056. if (val < 0 || val > 1)
  2057. err = -EINVAL;
  2058. else {
  2059. tp->thin_dupack = val;
  2060. if (tp->thin_dupack)
  2061. tcp_disable_early_retrans(tp);
  2062. }
  2063. break;
  2064. case TCP_REPAIR:
  2065. if (!tcp_can_repair_sock(sk))
  2066. err = -EPERM;
  2067. else if (val == 1) {
  2068. tp->repair = 1;
  2069. sk->sk_reuse = SK_FORCE_REUSE;
  2070. tp->repair_queue = TCP_NO_QUEUE;
  2071. } else if (val == 0) {
  2072. tp->repair = 0;
  2073. sk->sk_reuse = SK_NO_REUSE;
  2074. tcp_send_window_probe(sk);
  2075. } else
  2076. err = -EINVAL;
  2077. break;
  2078. case TCP_REPAIR_QUEUE:
  2079. if (!tp->repair)
  2080. err = -EPERM;
  2081. else if (val < TCP_QUEUES_NR)
  2082. tp->repair_queue = val;
  2083. else
  2084. err = -EINVAL;
  2085. break;
  2086. case TCP_QUEUE_SEQ:
  2087. if (sk->sk_state != TCP_CLOSE)
  2088. err = -EPERM;
  2089. else if (tp->repair_queue == TCP_SEND_QUEUE)
  2090. tp->write_seq = val;
  2091. else if (tp->repair_queue == TCP_RECV_QUEUE)
  2092. tp->rcv_nxt = val;
  2093. else
  2094. err = -EINVAL;
  2095. break;
  2096. case TCP_REPAIR_OPTIONS:
  2097. if (!tp->repair)
  2098. err = -EINVAL;
  2099. else if (sk->sk_state == TCP_ESTABLISHED)
  2100. err = tcp_repair_options_est(tp,
  2101. (struct tcp_repair_opt __user *)optval,
  2102. optlen);
  2103. else
  2104. err = -EPERM;
  2105. break;
  2106. case TCP_CORK:
  2107. /* When set indicates to always queue non-full frames.
  2108. * Later the user clears this option and we transmit
  2109. * any pending partial frames in the queue. This is
  2110. * meant to be used alongside sendfile() to get properly
  2111. * filled frames when the user (for example) must write
  2112. * out headers with a write() call first and then use
  2113. * sendfile to send out the data parts.
  2114. *
  2115. * TCP_CORK can be set together with TCP_NODELAY and it is
  2116. * stronger than TCP_NODELAY.
  2117. */
  2118. if (val) {
  2119. tp->nonagle |= TCP_NAGLE_CORK;
  2120. } else {
  2121. tp->nonagle &= ~TCP_NAGLE_CORK;
  2122. if (tp->nonagle&TCP_NAGLE_OFF)
  2123. tp->nonagle |= TCP_NAGLE_PUSH;
  2124. tcp_push_pending_frames(sk);
  2125. }
  2126. break;
  2127. case TCP_KEEPIDLE:
  2128. if (val < 1 || val > MAX_TCP_KEEPIDLE)
  2129. err = -EINVAL;
  2130. else {
  2131. tp->keepalive_time = val * HZ;
  2132. if (sock_flag(sk, SOCK_KEEPOPEN) &&
  2133. !((1 << sk->sk_state) &
  2134. (TCPF_CLOSE | TCPF_LISTEN))) {
  2135. u32 elapsed = keepalive_time_elapsed(tp);
  2136. if (tp->keepalive_time > elapsed)
  2137. elapsed = tp->keepalive_time - elapsed;
  2138. else
  2139. elapsed = 0;
  2140. inet_csk_reset_keepalive_timer(sk, elapsed);
  2141. }
  2142. }
  2143. break;
  2144. case TCP_KEEPINTVL:
  2145. if (val < 1 || val > MAX_TCP_KEEPINTVL)
  2146. err = -EINVAL;
  2147. else
  2148. tp->keepalive_intvl = val * HZ;
  2149. break;
  2150. case TCP_KEEPCNT:
  2151. if (val < 1 || val > MAX_TCP_KEEPCNT)
  2152. err = -EINVAL;
  2153. else
  2154. tp->keepalive_probes = val;
  2155. break;
  2156. case TCP_SYNCNT:
  2157. if (val < 1 || val > MAX_TCP_SYNCNT)
  2158. err = -EINVAL;
  2159. else
  2160. icsk->icsk_syn_retries = val;
  2161. break;
  2162. case TCP_LINGER2:
  2163. if (val < 0)
  2164. tp->linger2 = -1;
  2165. else if (val > sysctl_tcp_fin_timeout / HZ)
  2166. tp->linger2 = 0;
  2167. else
  2168. tp->linger2 = val * HZ;
  2169. break;
  2170. case TCP_DEFER_ACCEPT:
  2171. /* Translate value in seconds to number of retransmits */
  2172. icsk->icsk_accept_queue.rskq_defer_accept =
  2173. secs_to_retrans(val, TCP_TIMEOUT_INIT / HZ,
  2174. TCP_RTO_MAX / HZ);
  2175. break;
  2176. case TCP_WINDOW_CLAMP:
  2177. if (!val) {
  2178. if (sk->sk_state != TCP_CLOSE) {
  2179. err = -EINVAL;
  2180. break;
  2181. }
  2182. tp->window_clamp = 0;
  2183. } else
  2184. tp->window_clamp = val < SOCK_MIN_RCVBUF / 2 ?
  2185. SOCK_MIN_RCVBUF / 2 : val;
  2186. break;
  2187. case TCP_QUICKACK:
  2188. if (!val) {
  2189. icsk->icsk_ack.pingpong = 1;
  2190. } else {
  2191. icsk->icsk_ack.pingpong = 0;
  2192. if ((1 << sk->sk_state) &
  2193. (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT) &&
  2194. inet_csk_ack_scheduled(sk)) {
  2195. icsk->icsk_ack.pending |= ICSK_ACK_PUSHED;
  2196. tcp_cleanup_rbuf(sk, 1);
  2197. if (!(val & 1))
  2198. icsk->icsk_ack.pingpong = 1;
  2199. }
  2200. }
  2201. break;
  2202. #ifdef CONFIG_TCP_MD5SIG
  2203. case TCP_MD5SIG:
  2204. /* Read the IP->Key mappings from userspace */
  2205. err = tp->af_specific->md5_parse(sk, optval, optlen);
  2206. break;
  2207. #endif
  2208. case TCP_USER_TIMEOUT:
  2209. /* Cap the max time in ms TCP will retry or probe the window
  2210. * before giving up and aborting (ETIMEDOUT) a connection.
  2211. */
  2212. if (val < 0)
  2213. err = -EINVAL;
  2214. else
  2215. icsk->icsk_user_timeout = msecs_to_jiffies(val);
  2216. break;
  2217. case TCP_FASTOPEN:
  2218. if (val >= 0 && ((1 << sk->sk_state) & (TCPF_CLOSE |
  2219. TCPF_LISTEN)))
  2220. err = fastopen_init_queue(sk, val);
  2221. else
  2222. err = -EINVAL;
  2223. break;
  2224. case TCP_TIMESTAMP:
  2225. if (!tp->repair)
  2226. err = -EPERM;
  2227. else
  2228. tp->tsoffset = val - tcp_time_stamp;
  2229. break;
  2230. case TCP_NOTSENT_LOWAT:
  2231. tp->notsent_lowat = val;
  2232. sk->sk_write_space(sk);
  2233. break;
  2234. default:
  2235. err = -ENOPROTOOPT;
  2236. break;
  2237. }
  2238. release_sock(sk);
  2239. return err;
  2240. }
  2241. int tcp_setsockopt(struct sock *sk, int level, int optname, char __user *optval,
  2242. unsigned int optlen)
  2243. {
  2244. const struct inet_connection_sock *icsk = inet_csk(sk);
  2245. if (level != SOL_TCP)
  2246. return icsk->icsk_af_ops->setsockopt(sk, level, optname,
  2247. optval, optlen);
  2248. return do_tcp_setsockopt(sk, level, optname, optval, optlen);
  2249. }
  2250. EXPORT_SYMBOL(tcp_setsockopt);
  2251. #ifdef CONFIG_COMPAT
  2252. int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
  2253. char __user *optval, unsigned int optlen)
  2254. {
  2255. if (level != SOL_TCP)
  2256. return inet_csk_compat_setsockopt(sk, level, optname,
  2257. optval, optlen);
  2258. return do_tcp_setsockopt(sk, level, optname, optval, optlen);
  2259. }
  2260. EXPORT_SYMBOL(compat_tcp_setsockopt);
  2261. #endif
  2262. /* Return information about state of tcp endpoint in API format. */
  2263. void tcp_get_info(const struct sock *sk, struct tcp_info *info)
  2264. {
  2265. const struct tcp_sock *tp = tcp_sk(sk);
  2266. const struct inet_connection_sock *icsk = inet_csk(sk);
  2267. u32 now = tcp_time_stamp;
  2268. memset(info, 0, sizeof(*info));
  2269. info->tcpi_state = sk->sk_state;
  2270. info->tcpi_ca_state = icsk->icsk_ca_state;
  2271. info->tcpi_retransmits = icsk->icsk_retransmits;
  2272. info->tcpi_probes = icsk->icsk_probes_out;
  2273. info->tcpi_backoff = icsk->icsk_backoff;
  2274. if (tp->rx_opt.tstamp_ok)
  2275. info->tcpi_options |= TCPI_OPT_TIMESTAMPS;
  2276. if (tcp_is_sack(tp))
  2277. info->tcpi_options |= TCPI_OPT_SACK;
  2278. if (tp->rx_opt.wscale_ok) {
  2279. info->tcpi_options |= TCPI_OPT_WSCALE;
  2280. info->tcpi_snd_wscale = tp->rx_opt.snd_wscale;
  2281. info->tcpi_rcv_wscale = tp->rx_opt.rcv_wscale;
  2282. }
  2283. if (tp->ecn_flags & TCP_ECN_OK)
  2284. info->tcpi_options |= TCPI_OPT_ECN;
  2285. if (tp->ecn_flags & TCP_ECN_SEEN)
  2286. info->tcpi_options |= TCPI_OPT_ECN_SEEN;
  2287. if (tp->syn_data_acked)
  2288. info->tcpi_options |= TCPI_OPT_SYN_DATA;
  2289. info->tcpi_rto = jiffies_to_usecs(icsk->icsk_rto);
  2290. info->tcpi_ato = jiffies_to_usecs(icsk->icsk_ack.ato);
  2291. info->tcpi_snd_mss = tp->mss_cache;
  2292. info->tcpi_rcv_mss = icsk->icsk_ack.rcv_mss;
  2293. if (sk->sk_state == TCP_LISTEN) {
  2294. info->tcpi_unacked = sk->sk_ack_backlog;
  2295. info->tcpi_sacked = sk->sk_max_ack_backlog;
  2296. } else {
  2297. info->tcpi_unacked = tp->packets_out;
  2298. info->tcpi_sacked = tp->sacked_out;
  2299. }
  2300. info->tcpi_lost = tp->lost_out;
  2301. info->tcpi_retrans = tp->retrans_out;
  2302. info->tcpi_fackets = tp->fackets_out;
  2303. info->tcpi_last_data_sent = jiffies_to_msecs(now - tp->lsndtime);
  2304. info->tcpi_last_data_recv = jiffies_to_msecs(now - icsk->icsk_ack.lrcvtime);
  2305. info->tcpi_last_ack_recv = jiffies_to_msecs(now - tp->rcv_tstamp);
  2306. info->tcpi_pmtu = icsk->icsk_pmtu_cookie;
  2307. info->tcpi_rcv_ssthresh = tp->rcv_ssthresh;
  2308. info->tcpi_rtt = tp->srtt_us >> 3;
  2309. info->tcpi_rttvar = tp->mdev_us >> 2;
  2310. info->tcpi_snd_ssthresh = tp->snd_ssthresh;
  2311. info->tcpi_snd_cwnd = tp->snd_cwnd;
  2312. info->tcpi_advmss = tp->advmss;
  2313. info->tcpi_reordering = tp->reordering;
  2314. info->tcpi_rcv_rtt = jiffies_to_usecs(tp->rcv_rtt_est.rtt)>>3;
  2315. info->tcpi_rcv_space = tp->rcvq_space.space;
  2316. info->tcpi_total_retrans = tp->total_retrans;
  2317. info->tcpi_pacing_rate = sk->sk_pacing_rate != ~0U ?
  2318. sk->sk_pacing_rate : ~0ULL;
  2319. info->tcpi_max_pacing_rate = sk->sk_max_pacing_rate != ~0U ?
  2320. sk->sk_max_pacing_rate : ~0ULL;
  2321. }
  2322. EXPORT_SYMBOL_GPL(tcp_get_info);
  2323. static int do_tcp_getsockopt(struct sock *sk, int level,
  2324. int optname, char __user *optval, int __user *optlen)
  2325. {
  2326. struct inet_connection_sock *icsk = inet_csk(sk);
  2327. struct tcp_sock *tp = tcp_sk(sk);
  2328. int val, len;
  2329. if (get_user(len, optlen))
  2330. return -EFAULT;
  2331. len = min_t(unsigned int, len, sizeof(int));
  2332. if (len < 0)
  2333. return -EINVAL;
  2334. switch (optname) {
  2335. case TCP_MAXSEG:
  2336. val = tp->mss_cache;
  2337. if (!val && ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)))
  2338. val = tp->rx_opt.user_mss;
  2339. if (tp->repair)
  2340. val = tp->rx_opt.mss_clamp;
  2341. break;
  2342. case TCP_NODELAY:
  2343. val = !!(tp->nonagle&TCP_NAGLE_OFF);
  2344. break;
  2345. case TCP_CORK:
  2346. val = !!(tp->nonagle&TCP_NAGLE_CORK);
  2347. break;
  2348. case TCP_KEEPIDLE:
  2349. val = keepalive_time_when(tp) / HZ;
  2350. break;
  2351. case TCP_KEEPINTVL:
  2352. val = keepalive_intvl_when(tp) / HZ;
  2353. break;
  2354. case TCP_KEEPCNT:
  2355. val = keepalive_probes(tp);
  2356. break;
  2357. case TCP_SYNCNT:
  2358. val = icsk->icsk_syn_retries ? : sysctl_tcp_syn_retries;
  2359. break;
  2360. case TCP_LINGER2:
  2361. val = tp->linger2;
  2362. if (val >= 0)
  2363. val = (val ? : sysctl_tcp_fin_timeout) / HZ;
  2364. break;
  2365. case TCP_DEFER_ACCEPT:
  2366. val = retrans_to_secs(icsk->icsk_accept_queue.rskq_defer_accept,
  2367. TCP_TIMEOUT_INIT / HZ, TCP_RTO_MAX / HZ);
  2368. break;
  2369. case TCP_WINDOW_CLAMP:
  2370. val = tp->window_clamp;
  2371. break;
  2372. case TCP_INFO: {
  2373. struct tcp_info info;
  2374. if (get_user(len, optlen))
  2375. return -EFAULT;
  2376. tcp_get_info(sk, &info);
  2377. len = min_t(unsigned int, len, sizeof(info));
  2378. if (put_user(len, optlen))
  2379. return -EFAULT;
  2380. if (copy_to_user(optval, &info, len))
  2381. return -EFAULT;
  2382. return 0;
  2383. }
  2384. case TCP_QUICKACK:
  2385. val = !icsk->icsk_ack.pingpong;
  2386. break;
  2387. case TCP_CONGESTION:
  2388. if (get_user(len, optlen))
  2389. return -EFAULT;
  2390. len = min_t(unsigned int, len, TCP_CA_NAME_MAX);
  2391. if (put_user(len, optlen))
  2392. return -EFAULT;
  2393. if (copy_to_user(optval, icsk->icsk_ca_ops->name, len))
  2394. return -EFAULT;
  2395. return 0;
  2396. case TCP_THIN_LINEAR_TIMEOUTS:
  2397. val = tp->thin_lto;
  2398. break;
  2399. case TCP_THIN_DUPACK:
  2400. val = tp->thin_dupack;
  2401. break;
  2402. case TCP_REPAIR:
  2403. val = tp->repair;
  2404. break;
  2405. case TCP_REPAIR_QUEUE:
  2406. if (tp->repair)
  2407. val = tp->repair_queue;
  2408. else
  2409. return -EINVAL;
  2410. break;
  2411. case TCP_QUEUE_SEQ:
  2412. if (tp->repair_queue == TCP_SEND_QUEUE)
  2413. val = tp->write_seq;
  2414. else if (tp->repair_queue == TCP_RECV_QUEUE)
  2415. val = tp->rcv_nxt;
  2416. else
  2417. return -EINVAL;
  2418. break;
  2419. case TCP_USER_TIMEOUT:
  2420. val = jiffies_to_msecs(icsk->icsk_user_timeout);
  2421. break;
  2422. case TCP_FASTOPEN:
  2423. if (icsk->icsk_accept_queue.fastopenq != NULL)
  2424. val = icsk->icsk_accept_queue.fastopenq->max_qlen;
  2425. else
  2426. val = 0;
  2427. break;
  2428. case TCP_TIMESTAMP:
  2429. val = tcp_time_stamp + tp->tsoffset;
  2430. break;
  2431. case TCP_NOTSENT_LOWAT:
  2432. val = tp->notsent_lowat;
  2433. break;
  2434. default:
  2435. return -ENOPROTOOPT;
  2436. }
  2437. if (put_user(len, optlen))
  2438. return -EFAULT;
  2439. if (copy_to_user(optval, &val, len))
  2440. return -EFAULT;
  2441. return 0;
  2442. }
  2443. int tcp_getsockopt(struct sock *sk, int level, int optname, char __user *optval,
  2444. int __user *optlen)
  2445. {
  2446. struct inet_connection_sock *icsk = inet_csk(sk);
  2447. if (level != SOL_TCP)
  2448. return icsk->icsk_af_ops->getsockopt(sk, level, optname,
  2449. optval, optlen);
  2450. return do_tcp_getsockopt(sk, level, optname, optval, optlen);
  2451. }
  2452. EXPORT_SYMBOL(tcp_getsockopt);
  2453. #ifdef CONFIG_COMPAT
  2454. int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
  2455. char __user *optval, int __user *optlen)
  2456. {
  2457. if (level != SOL_TCP)
  2458. return inet_csk_compat_getsockopt(sk, level, optname,
  2459. optval, optlen);
  2460. return do_tcp_getsockopt(sk, level, optname, optval, optlen);
  2461. }
  2462. EXPORT_SYMBOL(compat_tcp_getsockopt);
  2463. #endif
  2464. #ifdef CONFIG_TCP_MD5SIG
  2465. static DEFINE_PER_CPU(struct tcp_md5sig_pool, tcp_md5sig_pool);
  2466. static DEFINE_MUTEX(tcp_md5sig_mutex);
  2467. static bool tcp_md5sig_pool_populated = false;
  2468. static void __tcp_alloc_md5sig_pool(void)
  2469. {
  2470. int cpu;
  2471. for_each_possible_cpu(cpu) {
  2472. if (!per_cpu(tcp_md5sig_pool, cpu).md5_desc.tfm) {
  2473. struct crypto_hash *hash;
  2474. hash = crypto_alloc_hash("md5", 0, CRYPTO_ALG_ASYNC);
  2475. if (IS_ERR_OR_NULL(hash))
  2476. return;
  2477. per_cpu(tcp_md5sig_pool, cpu).md5_desc.tfm = hash;
  2478. }
  2479. }
  2480. /* before setting tcp_md5sig_pool_populated, we must commit all writes
  2481. * to memory. See smp_rmb() in tcp_get_md5sig_pool()
  2482. */
  2483. smp_wmb();
  2484. tcp_md5sig_pool_populated = true;
  2485. }
  2486. bool tcp_alloc_md5sig_pool(void)
  2487. {
  2488. if (unlikely(!tcp_md5sig_pool_populated)) {
  2489. mutex_lock(&tcp_md5sig_mutex);
  2490. if (!tcp_md5sig_pool_populated)
  2491. __tcp_alloc_md5sig_pool();
  2492. mutex_unlock(&tcp_md5sig_mutex);
  2493. }
  2494. return tcp_md5sig_pool_populated;
  2495. }
  2496. EXPORT_SYMBOL(tcp_alloc_md5sig_pool);
  2497. /**
  2498. * tcp_get_md5sig_pool - get md5sig_pool for this user
  2499. *
  2500. * We use percpu structure, so if we succeed, we exit with preemption
  2501. * and BH disabled, to make sure another thread or softirq handling
  2502. * wont try to get same context.
  2503. */
  2504. struct tcp_md5sig_pool *tcp_get_md5sig_pool(void)
  2505. {
  2506. local_bh_disable();
  2507. if (tcp_md5sig_pool_populated) {
  2508. /* coupled with smp_wmb() in __tcp_alloc_md5sig_pool() */
  2509. smp_rmb();
  2510. return this_cpu_ptr(&tcp_md5sig_pool);
  2511. }
  2512. local_bh_enable();
  2513. return NULL;
  2514. }
  2515. EXPORT_SYMBOL(tcp_get_md5sig_pool);
  2516. int tcp_md5_hash_header(struct tcp_md5sig_pool *hp,
  2517. const struct tcphdr *th)
  2518. {
  2519. struct scatterlist sg;
  2520. struct tcphdr hdr;
  2521. int err;
  2522. /* We are not allowed to change tcphdr, make a local copy */
  2523. memcpy(&hdr, th, sizeof(hdr));
  2524. hdr.check = 0;
  2525. /* options aren't included in the hash */
  2526. sg_init_one(&sg, &hdr, sizeof(hdr));
  2527. err = crypto_hash_update(&hp->md5_desc, &sg, sizeof(hdr));
  2528. return err;
  2529. }
  2530. EXPORT_SYMBOL(tcp_md5_hash_header);
  2531. int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *hp,
  2532. const struct sk_buff *skb, unsigned int header_len)
  2533. {
  2534. struct scatterlist sg;
  2535. const struct tcphdr *tp = tcp_hdr(skb);
  2536. struct hash_desc *desc = &hp->md5_desc;
  2537. unsigned int i;
  2538. const unsigned int head_data_len = skb_headlen(skb) > header_len ?
  2539. skb_headlen(skb) - header_len : 0;
  2540. const struct skb_shared_info *shi = skb_shinfo(skb);
  2541. struct sk_buff *frag_iter;
  2542. sg_init_table(&sg, 1);
  2543. sg_set_buf(&sg, ((u8 *) tp) + header_len, head_data_len);
  2544. if (crypto_hash_update(desc, &sg, head_data_len))
  2545. return 1;
  2546. for (i = 0; i < shi->nr_frags; ++i) {
  2547. const struct skb_frag_struct *f = &shi->frags[i];
  2548. unsigned int offset = f->page_offset;
  2549. struct page *page = skb_frag_page(f) + (offset >> PAGE_SHIFT);
  2550. sg_set_page(&sg, page, skb_frag_size(f),
  2551. offset_in_page(offset));
  2552. if (crypto_hash_update(desc, &sg, skb_frag_size(f)))
  2553. return 1;
  2554. }
  2555. skb_walk_frags(skb, frag_iter)
  2556. if (tcp_md5_hash_skb_data(hp, frag_iter, 0))
  2557. return 1;
  2558. return 0;
  2559. }
  2560. EXPORT_SYMBOL(tcp_md5_hash_skb_data);
  2561. int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, const struct tcp_md5sig_key *key)
  2562. {
  2563. struct scatterlist sg;
  2564. sg_init_one(&sg, key->key, key->keylen);
  2565. return crypto_hash_update(&hp->md5_desc, &sg, key->keylen);
  2566. }
  2567. EXPORT_SYMBOL(tcp_md5_hash_key);
  2568. #endif
  2569. void tcp_done(struct sock *sk)
  2570. {
  2571. struct request_sock *req = tcp_sk(sk)->fastopen_rsk;
  2572. if (sk->sk_state == TCP_SYN_SENT || sk->sk_state == TCP_SYN_RECV)
  2573. TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_ATTEMPTFAILS);
  2574. tcp_set_state(sk, TCP_CLOSE);
  2575. tcp_clear_xmit_timers(sk);
  2576. if (req != NULL)
  2577. reqsk_fastopen_remove(sk, req, false);
  2578. sk->sk_shutdown = SHUTDOWN_MASK;
  2579. if (!sock_flag(sk, SOCK_DEAD))
  2580. sk->sk_state_change(sk);
  2581. else
  2582. inet_csk_destroy_sock(sk);
  2583. }
  2584. EXPORT_SYMBOL_GPL(tcp_done);
  2585. extern struct tcp_congestion_ops tcp_reno;
  2586. static __initdata unsigned long thash_entries;
  2587. static int __init set_thash_entries(char *str)
  2588. {
  2589. ssize_t ret;
  2590. if (!str)
  2591. return 0;
  2592. ret = kstrtoul(str, 0, &thash_entries);
  2593. if (ret)
  2594. return 0;
  2595. return 1;
  2596. }
  2597. __setup("thash_entries=", set_thash_entries);
  2598. static void __init tcp_init_mem(void)
  2599. {
  2600. unsigned long limit = nr_free_buffer_pages() / 8;
  2601. limit = max(limit, 128UL);
  2602. sysctl_tcp_mem[0] = limit / 4 * 3;
  2603. sysctl_tcp_mem[1] = limit;
  2604. sysctl_tcp_mem[2] = sysctl_tcp_mem[0] * 2;
  2605. }
  2606. void __init tcp_init(void)
  2607. {
  2608. struct sk_buff *skb = NULL;
  2609. unsigned long limit;
  2610. int max_rshare, max_wshare, cnt;
  2611. unsigned int i;
  2612. BUILD_BUG_ON(sizeof(struct tcp_skb_cb) > sizeof(skb->cb));
  2613. percpu_counter_init(&tcp_sockets_allocated, 0, GFP_KERNEL);
  2614. percpu_counter_init(&tcp_orphan_count, 0, GFP_KERNEL);
  2615. tcp_hashinfo.bind_bucket_cachep =
  2616. kmem_cache_create("tcp_bind_bucket",
  2617. sizeof(struct inet_bind_bucket), 0,
  2618. SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
  2619. /* Size and allocate the main established and bind bucket
  2620. * hash tables.
  2621. *
  2622. * The methodology is similar to that of the buffer cache.
  2623. */
  2624. tcp_hashinfo.ehash =
  2625. alloc_large_system_hash("TCP established",
  2626. sizeof(struct inet_ehash_bucket),
  2627. thash_entries,
  2628. 17, /* one slot per 128 KB of memory */
  2629. 0,
  2630. NULL,
  2631. &tcp_hashinfo.ehash_mask,
  2632. 0,
  2633. thash_entries ? 0 : 512 * 1024);
  2634. for (i = 0; i <= tcp_hashinfo.ehash_mask; i++)
  2635. INIT_HLIST_NULLS_HEAD(&tcp_hashinfo.ehash[i].chain, i);
  2636. if (inet_ehash_locks_alloc(&tcp_hashinfo))
  2637. panic("TCP: failed to alloc ehash_locks");
  2638. tcp_hashinfo.bhash =
  2639. alloc_large_system_hash("TCP bind",
  2640. sizeof(struct inet_bind_hashbucket),
  2641. tcp_hashinfo.ehash_mask + 1,
  2642. 17, /* one slot per 128 KB of memory */
  2643. 0,
  2644. &tcp_hashinfo.bhash_size,
  2645. NULL,
  2646. 0,
  2647. 64 * 1024);
  2648. tcp_hashinfo.bhash_size = 1U << tcp_hashinfo.bhash_size;
  2649. for (i = 0; i < tcp_hashinfo.bhash_size; i++) {
  2650. spin_lock_init(&tcp_hashinfo.bhash[i].lock);
  2651. INIT_HLIST_HEAD(&tcp_hashinfo.bhash[i].chain);
  2652. }
  2653. cnt = tcp_hashinfo.ehash_mask + 1;
  2654. tcp_death_row.sysctl_max_tw_buckets = cnt / 2;
  2655. sysctl_tcp_max_orphans = cnt / 2;
  2656. sysctl_max_syn_backlog = max(128, cnt / 256);
  2657. tcp_init_mem();
  2658. /* Set per-socket limits to no more than 1/128 the pressure threshold */
  2659. limit = nr_free_buffer_pages() << (PAGE_SHIFT - 7);
  2660. max_wshare = min(4UL*1024*1024, limit);
  2661. max_rshare = min(6UL*1024*1024, limit);
  2662. sysctl_tcp_wmem[0] = SK_MEM_QUANTUM;
  2663. sysctl_tcp_wmem[1] = 16*1024;
  2664. sysctl_tcp_wmem[2] = max(64*1024, max_wshare);
  2665. sysctl_tcp_rmem[0] = SK_MEM_QUANTUM;
  2666. sysctl_tcp_rmem[1] = 87380;
  2667. sysctl_tcp_rmem[2] = max(87380, max_rshare);
  2668. pr_info("Hash tables configured (established %u bind %u)\n",
  2669. tcp_hashinfo.ehash_mask + 1, tcp_hashinfo.bhash_size);
  2670. tcp_metrics_init();
  2671. BUG_ON(tcp_register_congestion_control(&tcp_reno) != 0);
  2672. tcp_tasklet_init();
  2673. }