sock.c 65 KB

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027102810291030103110321033103410351036103710381039104010411042104310441045104610471048104910501051105210531054105510561057105810591060106110621063106410651066106710681069107010711072107310741075107610771078107910801081108210831084108510861087108810891090109110921093109410951096109710981099110011011102110311041105110611071108110911101111111211131114111511161117111811191120112111221123112411251126112711281129113011311132113311341135113611371138113911401141114211431144114511461147114811491150115111521153115411551156115711581159116011611162116311641165116611671168116911701171117211731174117511761177117811791180118111821183118411851186118711881189119011911192119311941195119611971198119912001201120212031204120512061207120812091210121112121213121412151216121712181219122012211222122312241225122612271228122912301231123212331234123512361237123812391240124112421243124412451246124712481249125012511252125312541255125612571258125912601261126212631264126512661267126812691270127112721273127412751276127712781279128012811282128312841285128612871288128912901291129212931294129512961297129812991300130113021303130413051306130713081309131013111312131313141315131613171318131913201321132213231324132513261327132813291330133113321333133413351336133713381339134013411342134313441345134613471348134913501351135213531354135513561357135813591360136113621363136413651366136713681369137013711372137313741375137613771378137913801381138213831384138513861387138813891390139113921393139413951396139713981399140014011402140314041405140614071408140914101411141214131414141514161417141814191420142114221423142414251426142714281429143014311432143314341435143614371438143914401441144214431444144514461447144814491450145114521453145414551456145714581459146014611462146314641465146614671468146914701471147214731474147514761477147814791480148114821483148414851486148714881489149014911492149314941495149614971498149915001501150215031504150515061507150815091510151115121513151415151516151715181519152015211522152315241525152615271528152915301531153215331534153515361537153815391540154115421543154415451546154715481549155015511552155315541555155615571558155915601561156215631564156515661567156815691570157115721573157415751576157715781579158015811582158315841585158615871588158915901591159215931594159515961597159815991600160116021603160416051606160716081609161016111612161316141615161616171618161916201621162216231624162516261627162816291630163116321633163416351636163716381639164016411642164316441645164616471648164916501651165216531654165516561657165816591660166116621663166416651666166716681669167016711672167316741675167616771678167916801681168216831684168516861687168816891690169116921693169416951696169716981699170017011702170317041705170617071708170917101711171217131714171517161717171817191720172117221723172417251726172717281729173017311732173317341735173617371738173917401741174217431744174517461747174817491750175117521753175417551756175717581759176017611762176317641765176617671768176917701771177217731774177517761777177817791780178117821783178417851786178717881789179017911792179317941795179617971798179918001801180218031804180518061807180818091810181118121813181418151816181718181819182018211822182318241825182618271828182918301831183218331834183518361837183818391840184118421843184418451846184718481849185018511852185318541855185618571858185918601861186218631864186518661867186818691870187118721873187418751876187718781879188018811882188318841885188618871888188918901891189218931894189518961897189818991900190119021903190419051906190719081909191019111912191319141915191619171918191919201921192219231924192519261927192819291930193119321933193419351936193719381939194019411942194319441945194619471948194919501951195219531954195519561957195819591960196119621963196419651966196719681969197019711972197319741975197619771978197919801981198219831984198519861987198819891990199119921993199419951996199719981999200020012002200320042005200620072008200920102011201220132014201520162017201820192020202120222023202420252026202720282029203020312032203320342035203620372038203920402041204220432044204520462047204820492050205120522053205420552056205720582059206020612062206320642065206620672068206920702071207220732074207520762077207820792080208120822083208420852086208720882089209020912092209320942095209620972098209921002101210221032104210521062107210821092110211121122113211421152116211721182119212021212122212321242125212621272128212921302131213221332134213521362137213821392140214121422143214421452146214721482149215021512152215321542155215621572158215921602161216221632164216521662167216821692170217121722173217421752176217721782179218021812182218321842185218621872188218921902191219221932194219521962197219821992200220122022203220422052206220722082209221022112212221322142215221622172218221922202221222222232224222522262227222822292230223122322233223422352236223722382239224022412242224322442245224622472248224922502251225222532254225522562257225822592260226122622263226422652266226722682269227022712272227322742275227622772278227922802281228222832284228522862287228822892290229122922293229422952296229722982299230023012302230323042305230623072308230923102311231223132314231523162317231823192320232123222323232423252326232723282329233023312332233323342335233623372338233923402341234223432344234523462347234823492350235123522353235423552356235723582359236023612362236323642365236623672368236923702371237223732374237523762377237823792380238123822383238423852386238723882389239023912392239323942395239623972398239924002401240224032404240524062407240824092410241124122413241424152416241724182419242024212422242324242425242624272428242924302431243224332434243524362437243824392440244124422443244424452446244724482449245024512452245324542455245624572458245924602461246224632464246524662467246824692470247124722473247424752476247724782479248024812482248324842485248624872488248924902491249224932494249524962497249824992500250125022503250425052506250725082509251025112512251325142515251625172518251925202521252225232524252525262527252825292530253125322533253425352536253725382539254025412542254325442545254625472548254925502551255225532554255525562557255825592560256125622563256425652566256725682569257025712572257325742575257625772578257925802581258225832584258525862587258825892590259125922593259425952596259725982599260026012602260326042605260626072608260926102611261226132614261526162617261826192620
  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. * Generic socket support routines. Memory allocators, socket lock/release
  7. * handler for protocols to use and generic option handler.
  8. *
  9. *
  10. * Authors: Ross Biro
  11. * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  12. * Florian La Roche, <flla@stud.uni-sb.de>
  13. * Alan Cox, <A.Cox@swansea.ac.uk>
  14. *
  15. * Fixes:
  16. * Alan Cox : Numerous verify_area() problems
  17. * Alan Cox : Connecting on a connecting socket
  18. * now returns an error for tcp.
  19. * Alan Cox : sock->protocol is set correctly.
  20. * and is not sometimes left as 0.
  21. * Alan Cox : connect handles icmp errors on a
  22. * connect properly. Unfortunately there
  23. * is a restart syscall nasty there. I
  24. * can't match BSD without hacking the C
  25. * library. Ideas urgently sought!
  26. * Alan Cox : Disallow bind() to addresses that are
  27. * not ours - especially broadcast ones!!
  28. * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
  29. * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
  30. * instead they leave that for the DESTROY timer.
  31. * Alan Cox : Clean up error flag in accept
  32. * Alan Cox : TCP ack handling is buggy, the DESTROY timer
  33. * was buggy. Put a remove_sock() in the handler
  34. * for memory when we hit 0. Also altered the timer
  35. * code. The ACK stuff can wait and needs major
  36. * TCP layer surgery.
  37. * Alan Cox : Fixed TCP ack bug, removed remove sock
  38. * and fixed timer/inet_bh race.
  39. * Alan Cox : Added zapped flag for TCP
  40. * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
  41. * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
  42. * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
  43. * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
  44. * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
  45. * Rick Sladkey : Relaxed UDP rules for matching packets.
  46. * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
  47. * Pauline Middelink : identd support
  48. * Alan Cox : Fixed connect() taking signals I think.
  49. * Alan Cox : SO_LINGER supported
  50. * Alan Cox : Error reporting fixes
  51. * Anonymous : inet_create tidied up (sk->reuse setting)
  52. * Alan Cox : inet sockets don't set sk->type!
  53. * Alan Cox : Split socket option code
  54. * Alan Cox : Callbacks
  55. * Alan Cox : Nagle flag for Charles & Johannes stuff
  56. * Alex : Removed restriction on inet fioctl
  57. * Alan Cox : Splitting INET from NET core
  58. * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
  59. * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
  60. * Alan Cox : Split IP from generic code
  61. * Alan Cox : New kfree_skbmem()
  62. * Alan Cox : Make SO_DEBUG superuser only.
  63. * Alan Cox : Allow anyone to clear SO_DEBUG
  64. * (compatibility fix)
  65. * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
  66. * Alan Cox : Allocator for a socket is settable.
  67. * Alan Cox : SO_ERROR includes soft errors.
  68. * Alan Cox : Allow NULL arguments on some SO_ opts
  69. * Alan Cox : Generic socket allocation to make hooks
  70. * easier (suggested by Craig Metz).
  71. * Michael Pall : SO_ERROR returns positive errno again
  72. * Steve Whitehouse: Added default destructor to free
  73. * protocol private data.
  74. * Steve Whitehouse: Added various other default routines
  75. * common to several socket families.
  76. * Chris Evans : Call suser() check last on F_SETOWN
  77. * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
  78. * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
  79. * Andi Kleen : Fix write_space callback
  80. * Chris Evans : Security fixes - signedness again
  81. * Arnaldo C. Melo : cleanups, use skb_queue_purge
  82. *
  83. * To Fix:
  84. *
  85. *
  86. * This program is free software; you can redistribute it and/or
  87. * modify it under the terms of the GNU General Public License
  88. * as published by the Free Software Foundation; either version
  89. * 2 of the License, or (at your option) any later version.
  90. */
  91. #include <linux/capability.h>
  92. #include <linux/errno.h>
  93. #include <linux/types.h>
  94. #include <linux/socket.h>
  95. #include <linux/in.h>
  96. #include <linux/kernel.h>
  97. #include <linux/module.h>
  98. #include <linux/proc_fs.h>
  99. #include <linux/seq_file.h>
  100. #include <linux/sched.h>
  101. #include <linux/timer.h>
  102. #include <linux/string.h>
  103. #include <linux/sockios.h>
  104. #include <linux/net.h>
  105. #include <linux/mm.h>
  106. #include <linux/slab.h>
  107. #include <linux/interrupt.h>
  108. #include <linux/poll.h>
  109. #include <linux/tcp.h>
  110. #include <linux/init.h>
  111. #include <linux/highmem.h>
  112. #include <linux/user_namespace.h>
  113. #include <asm/uaccess.h>
  114. #include <asm/system.h>
  115. #include <linux/netdevice.h>
  116. #include <net/protocol.h>
  117. #include <linux/skbuff.h>
  118. #include <net/net_namespace.h>
  119. #include <net/request_sock.h>
  120. #include <net/sock.h>
  121. #include <linux/net_tstamp.h>
  122. #include <net/xfrm.h>
  123. #include <linux/ipsec.h>
  124. #include <net/cls_cgroup.h>
  125. #include <net/netprio_cgroup.h>
  126. #include <linux/filter.h>
  127. #include <trace/events/sock.h>
  128. #ifdef CONFIG_INET
  129. #include <net/tcp.h>
  130. #endif
  131. /*
  132. * Each address family might have different locking rules, so we have
  133. * one slock key per address family:
  134. */
  135. static struct lock_class_key af_family_keys[AF_MAX];
  136. static struct lock_class_key af_family_slock_keys[AF_MAX];
  137. /*
  138. * Make lock validator output more readable. (we pre-construct these
  139. * strings build-time, so that runtime initialization of socket
  140. * locks is fast):
  141. */
  142. static const char *const af_family_key_strings[AF_MAX+1] = {
  143. "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
  144. "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
  145. "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
  146. "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
  147. "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
  148. "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
  149. "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
  150. "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
  151. "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
  152. "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
  153. "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
  154. "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
  155. "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
  156. "sk_lock-AF_NFC" , "sk_lock-AF_MAX"
  157. };
  158. static const char *const af_family_slock_key_strings[AF_MAX+1] = {
  159. "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
  160. "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
  161. "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
  162. "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
  163. "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
  164. "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
  165. "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
  166. "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
  167. "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
  168. "slock-27" , "slock-28" , "slock-AF_CAN" ,
  169. "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
  170. "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
  171. "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
  172. "slock-AF_NFC" , "slock-AF_MAX"
  173. };
  174. static const char *const af_family_clock_key_strings[AF_MAX+1] = {
  175. "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
  176. "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
  177. "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
  178. "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
  179. "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
  180. "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
  181. "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
  182. "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
  183. "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
  184. "clock-27" , "clock-28" , "clock-AF_CAN" ,
  185. "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
  186. "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
  187. "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
  188. "clock-AF_NFC" , "clock-AF_MAX"
  189. };
  190. /*
  191. * sk_callback_lock locking rules are per-address-family,
  192. * so split the lock classes by using a per-AF key:
  193. */
  194. static struct lock_class_key af_callback_keys[AF_MAX];
  195. /* Take into consideration the size of the struct sk_buff overhead in the
  196. * determination of these values, since that is non-constant across
  197. * platforms. This makes socket queueing behavior and performance
  198. * not depend upon such differences.
  199. */
  200. #define _SK_MEM_PACKETS 256
  201. #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
  202. #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
  203. #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
  204. /* Run time adjustable parameters. */
  205. __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
  206. __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
  207. __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
  208. __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
  209. /* Maximal space eaten by iovec or ancillary data plus some space */
  210. int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
  211. EXPORT_SYMBOL(sysctl_optmem_max);
  212. #if defined(CONFIG_CGROUPS)
  213. #if !defined(CONFIG_NET_CLS_CGROUP)
  214. int net_cls_subsys_id = -1;
  215. EXPORT_SYMBOL_GPL(net_cls_subsys_id);
  216. #endif
  217. #if !defined(CONFIG_NETPRIO_CGROUP)
  218. int net_prio_subsys_id = -1;
  219. EXPORT_SYMBOL_GPL(net_prio_subsys_id);
  220. #endif
  221. #endif
  222. static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
  223. {
  224. struct timeval tv;
  225. if (optlen < sizeof(tv))
  226. return -EINVAL;
  227. if (copy_from_user(&tv, optval, sizeof(tv)))
  228. return -EFAULT;
  229. if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
  230. return -EDOM;
  231. if (tv.tv_sec < 0) {
  232. static int warned __read_mostly;
  233. *timeo_p = 0;
  234. if (warned < 10 && net_ratelimit()) {
  235. warned++;
  236. printk(KERN_INFO "sock_set_timeout: `%s' (pid %d) "
  237. "tries to set negative timeout\n",
  238. current->comm, task_pid_nr(current));
  239. }
  240. return 0;
  241. }
  242. *timeo_p = MAX_SCHEDULE_TIMEOUT;
  243. if (tv.tv_sec == 0 && tv.tv_usec == 0)
  244. return 0;
  245. if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
  246. *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
  247. return 0;
  248. }
  249. static void sock_warn_obsolete_bsdism(const char *name)
  250. {
  251. static int warned;
  252. static char warncomm[TASK_COMM_LEN];
  253. if (strcmp(warncomm, current->comm) && warned < 5) {
  254. strcpy(warncomm, current->comm);
  255. printk(KERN_WARNING "process `%s' is using obsolete "
  256. "%s SO_BSDCOMPAT\n", warncomm, name);
  257. warned++;
  258. }
  259. }
  260. #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
  261. static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
  262. {
  263. if (sk->sk_flags & flags) {
  264. sk->sk_flags &= ~flags;
  265. if (!(sk->sk_flags & SK_FLAGS_TIMESTAMP))
  266. net_disable_timestamp();
  267. }
  268. }
  269. int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
  270. {
  271. int err;
  272. int skb_len;
  273. unsigned long flags;
  274. struct sk_buff_head *list = &sk->sk_receive_queue;
  275. /* Cast sk->rcvbuf to unsigned... It's pointless, but reduces
  276. number of warnings when compiling with -W --ANK
  277. */
  278. if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
  279. (unsigned)sk->sk_rcvbuf) {
  280. atomic_inc(&sk->sk_drops);
  281. trace_sock_rcvqueue_full(sk, skb);
  282. return -ENOMEM;
  283. }
  284. err = sk_filter(sk, skb);
  285. if (err)
  286. return err;
  287. if (!sk_rmem_schedule(sk, skb->truesize)) {
  288. atomic_inc(&sk->sk_drops);
  289. return -ENOBUFS;
  290. }
  291. skb->dev = NULL;
  292. skb_set_owner_r(skb, sk);
  293. /* Cache the SKB length before we tack it onto the receive
  294. * queue. Once it is added it no longer belongs to us and
  295. * may be freed by other threads of control pulling packets
  296. * from the queue.
  297. */
  298. skb_len = skb->len;
  299. /* we escape from rcu protected region, make sure we dont leak
  300. * a norefcounted dst
  301. */
  302. skb_dst_force(skb);
  303. spin_lock_irqsave(&list->lock, flags);
  304. skb->dropcount = atomic_read(&sk->sk_drops);
  305. __skb_queue_tail(list, skb);
  306. spin_unlock_irqrestore(&list->lock, flags);
  307. if (!sock_flag(sk, SOCK_DEAD))
  308. sk->sk_data_ready(sk, skb_len);
  309. return 0;
  310. }
  311. EXPORT_SYMBOL(sock_queue_rcv_skb);
  312. int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
  313. {
  314. int rc = NET_RX_SUCCESS;
  315. if (sk_filter(sk, skb))
  316. goto discard_and_relse;
  317. skb->dev = NULL;
  318. if (sk_rcvqueues_full(sk, skb)) {
  319. atomic_inc(&sk->sk_drops);
  320. goto discard_and_relse;
  321. }
  322. if (nested)
  323. bh_lock_sock_nested(sk);
  324. else
  325. bh_lock_sock(sk);
  326. if (!sock_owned_by_user(sk)) {
  327. /*
  328. * trylock + unlock semantics:
  329. */
  330. mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
  331. rc = sk_backlog_rcv(sk, skb);
  332. mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
  333. } else if (sk_add_backlog(sk, skb)) {
  334. bh_unlock_sock(sk);
  335. atomic_inc(&sk->sk_drops);
  336. goto discard_and_relse;
  337. }
  338. bh_unlock_sock(sk);
  339. out:
  340. sock_put(sk);
  341. return rc;
  342. discard_and_relse:
  343. kfree_skb(skb);
  344. goto out;
  345. }
  346. EXPORT_SYMBOL(sk_receive_skb);
  347. void sk_reset_txq(struct sock *sk)
  348. {
  349. sk_tx_queue_clear(sk);
  350. }
  351. EXPORT_SYMBOL(sk_reset_txq);
  352. struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
  353. {
  354. struct dst_entry *dst = __sk_dst_get(sk);
  355. if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
  356. sk_tx_queue_clear(sk);
  357. RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
  358. dst_release(dst);
  359. return NULL;
  360. }
  361. return dst;
  362. }
  363. EXPORT_SYMBOL(__sk_dst_check);
  364. struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
  365. {
  366. struct dst_entry *dst = sk_dst_get(sk);
  367. if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
  368. sk_dst_reset(sk);
  369. dst_release(dst);
  370. return NULL;
  371. }
  372. return dst;
  373. }
  374. EXPORT_SYMBOL(sk_dst_check);
  375. static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
  376. {
  377. int ret = -ENOPROTOOPT;
  378. #ifdef CONFIG_NETDEVICES
  379. struct net *net = sock_net(sk);
  380. char devname[IFNAMSIZ];
  381. int index;
  382. /* Sorry... */
  383. ret = -EPERM;
  384. if (!capable(CAP_NET_RAW))
  385. goto out;
  386. ret = -EINVAL;
  387. if (optlen < 0)
  388. goto out;
  389. /* Bind this socket to a particular device like "eth0",
  390. * as specified in the passed interface name. If the
  391. * name is "" or the option length is zero the socket
  392. * is not bound.
  393. */
  394. if (optlen > IFNAMSIZ - 1)
  395. optlen = IFNAMSIZ - 1;
  396. memset(devname, 0, sizeof(devname));
  397. ret = -EFAULT;
  398. if (copy_from_user(devname, optval, optlen))
  399. goto out;
  400. index = 0;
  401. if (devname[0] != '\0') {
  402. struct net_device *dev;
  403. rcu_read_lock();
  404. dev = dev_get_by_name_rcu(net, devname);
  405. if (dev)
  406. index = dev->ifindex;
  407. rcu_read_unlock();
  408. ret = -ENODEV;
  409. if (!dev)
  410. goto out;
  411. }
  412. lock_sock(sk);
  413. sk->sk_bound_dev_if = index;
  414. sk_dst_reset(sk);
  415. release_sock(sk);
  416. ret = 0;
  417. out:
  418. #endif
  419. return ret;
  420. }
  421. static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
  422. {
  423. if (valbool)
  424. sock_set_flag(sk, bit);
  425. else
  426. sock_reset_flag(sk, bit);
  427. }
  428. /*
  429. * This is meant for all protocols to use and covers goings on
  430. * at the socket level. Everything here is generic.
  431. */
  432. int sock_setsockopt(struct socket *sock, int level, int optname,
  433. char __user *optval, unsigned int optlen)
  434. {
  435. struct sock *sk = sock->sk;
  436. int val;
  437. int valbool;
  438. struct linger ling;
  439. int ret = 0;
  440. /*
  441. * Options without arguments
  442. */
  443. if (optname == SO_BINDTODEVICE)
  444. return sock_bindtodevice(sk, optval, optlen);
  445. if (optlen < sizeof(int))
  446. return -EINVAL;
  447. if (get_user(val, (int __user *)optval))
  448. return -EFAULT;
  449. valbool = val ? 1 : 0;
  450. lock_sock(sk);
  451. switch (optname) {
  452. case SO_DEBUG:
  453. if (val && !capable(CAP_NET_ADMIN))
  454. ret = -EACCES;
  455. else
  456. sock_valbool_flag(sk, SOCK_DBG, valbool);
  457. break;
  458. case SO_REUSEADDR:
  459. sk->sk_reuse = valbool;
  460. break;
  461. case SO_TYPE:
  462. case SO_PROTOCOL:
  463. case SO_DOMAIN:
  464. case SO_ERROR:
  465. ret = -ENOPROTOOPT;
  466. break;
  467. case SO_DONTROUTE:
  468. sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
  469. break;
  470. case SO_BROADCAST:
  471. sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
  472. break;
  473. case SO_SNDBUF:
  474. /* Don't error on this BSD doesn't and if you think
  475. about it this is right. Otherwise apps have to
  476. play 'guess the biggest size' games. RCVBUF/SNDBUF
  477. are treated in BSD as hints */
  478. if (val > sysctl_wmem_max)
  479. val = sysctl_wmem_max;
  480. set_sndbuf:
  481. sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
  482. if ((val * 2) < SOCK_MIN_SNDBUF)
  483. sk->sk_sndbuf = SOCK_MIN_SNDBUF;
  484. else
  485. sk->sk_sndbuf = val * 2;
  486. /*
  487. * Wake up sending tasks if we
  488. * upped the value.
  489. */
  490. sk->sk_write_space(sk);
  491. break;
  492. case SO_SNDBUFFORCE:
  493. if (!capable(CAP_NET_ADMIN)) {
  494. ret = -EPERM;
  495. break;
  496. }
  497. goto set_sndbuf;
  498. case SO_RCVBUF:
  499. /* Don't error on this BSD doesn't and if you think
  500. about it this is right. Otherwise apps have to
  501. play 'guess the biggest size' games. RCVBUF/SNDBUF
  502. are treated in BSD as hints */
  503. if (val > sysctl_rmem_max)
  504. val = sysctl_rmem_max;
  505. set_rcvbuf:
  506. sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
  507. /*
  508. * We double it on the way in to account for
  509. * "struct sk_buff" etc. overhead. Applications
  510. * assume that the SO_RCVBUF setting they make will
  511. * allow that much actual data to be received on that
  512. * socket.
  513. *
  514. * Applications are unaware that "struct sk_buff" and
  515. * other overheads allocate from the receive buffer
  516. * during socket buffer allocation.
  517. *
  518. * And after considering the possible alternatives,
  519. * returning the value we actually used in getsockopt
  520. * is the most desirable behavior.
  521. */
  522. if ((val * 2) < SOCK_MIN_RCVBUF)
  523. sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
  524. else
  525. sk->sk_rcvbuf = val * 2;
  526. break;
  527. case SO_RCVBUFFORCE:
  528. if (!capable(CAP_NET_ADMIN)) {
  529. ret = -EPERM;
  530. break;
  531. }
  532. goto set_rcvbuf;
  533. case SO_KEEPALIVE:
  534. #ifdef CONFIG_INET
  535. if (sk->sk_protocol == IPPROTO_TCP)
  536. tcp_set_keepalive(sk, valbool);
  537. #endif
  538. sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
  539. break;
  540. case SO_OOBINLINE:
  541. sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
  542. break;
  543. case SO_NO_CHECK:
  544. sk->sk_no_check = valbool;
  545. break;
  546. case SO_PRIORITY:
  547. if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
  548. sk->sk_priority = val;
  549. else
  550. ret = -EPERM;
  551. break;
  552. case SO_LINGER:
  553. if (optlen < sizeof(ling)) {
  554. ret = -EINVAL; /* 1003.1g */
  555. break;
  556. }
  557. if (copy_from_user(&ling, optval, sizeof(ling))) {
  558. ret = -EFAULT;
  559. break;
  560. }
  561. if (!ling.l_onoff)
  562. sock_reset_flag(sk, SOCK_LINGER);
  563. else {
  564. #if (BITS_PER_LONG == 32)
  565. if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
  566. sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
  567. else
  568. #endif
  569. sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
  570. sock_set_flag(sk, SOCK_LINGER);
  571. }
  572. break;
  573. case SO_BSDCOMPAT:
  574. sock_warn_obsolete_bsdism("setsockopt");
  575. break;
  576. case SO_PASSCRED:
  577. if (valbool)
  578. set_bit(SOCK_PASSCRED, &sock->flags);
  579. else
  580. clear_bit(SOCK_PASSCRED, &sock->flags);
  581. break;
  582. case SO_TIMESTAMP:
  583. case SO_TIMESTAMPNS:
  584. if (valbool) {
  585. if (optname == SO_TIMESTAMP)
  586. sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
  587. else
  588. sock_set_flag(sk, SOCK_RCVTSTAMPNS);
  589. sock_set_flag(sk, SOCK_RCVTSTAMP);
  590. sock_enable_timestamp(sk, SOCK_TIMESTAMP);
  591. } else {
  592. sock_reset_flag(sk, SOCK_RCVTSTAMP);
  593. sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
  594. }
  595. break;
  596. case SO_TIMESTAMPING:
  597. if (val & ~SOF_TIMESTAMPING_MASK) {
  598. ret = -EINVAL;
  599. break;
  600. }
  601. sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
  602. val & SOF_TIMESTAMPING_TX_HARDWARE);
  603. sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
  604. val & SOF_TIMESTAMPING_TX_SOFTWARE);
  605. sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
  606. val & SOF_TIMESTAMPING_RX_HARDWARE);
  607. if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
  608. sock_enable_timestamp(sk,
  609. SOCK_TIMESTAMPING_RX_SOFTWARE);
  610. else
  611. sock_disable_timestamp(sk,
  612. (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
  613. sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
  614. val & SOF_TIMESTAMPING_SOFTWARE);
  615. sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
  616. val & SOF_TIMESTAMPING_SYS_HARDWARE);
  617. sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
  618. val & SOF_TIMESTAMPING_RAW_HARDWARE);
  619. break;
  620. case SO_RCVLOWAT:
  621. if (val < 0)
  622. val = INT_MAX;
  623. sk->sk_rcvlowat = val ? : 1;
  624. break;
  625. case SO_RCVTIMEO:
  626. ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
  627. break;
  628. case SO_SNDTIMEO:
  629. ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
  630. break;
  631. case SO_ATTACH_FILTER:
  632. ret = -EINVAL;
  633. if (optlen == sizeof(struct sock_fprog)) {
  634. struct sock_fprog fprog;
  635. ret = -EFAULT;
  636. if (copy_from_user(&fprog, optval, sizeof(fprog)))
  637. break;
  638. ret = sk_attach_filter(&fprog, sk);
  639. }
  640. break;
  641. case SO_DETACH_FILTER:
  642. ret = sk_detach_filter(sk);
  643. break;
  644. case SO_PASSSEC:
  645. if (valbool)
  646. set_bit(SOCK_PASSSEC, &sock->flags);
  647. else
  648. clear_bit(SOCK_PASSSEC, &sock->flags);
  649. break;
  650. case SO_MARK:
  651. if (!capable(CAP_NET_ADMIN))
  652. ret = -EPERM;
  653. else
  654. sk->sk_mark = val;
  655. break;
  656. /* We implement the SO_SNDLOWAT etc to
  657. not be settable (1003.1g 5.3) */
  658. case SO_RXQ_OVFL:
  659. sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
  660. break;
  661. case SO_WIFI_STATUS:
  662. sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
  663. break;
  664. default:
  665. ret = -ENOPROTOOPT;
  666. break;
  667. }
  668. release_sock(sk);
  669. return ret;
  670. }
  671. EXPORT_SYMBOL(sock_setsockopt);
  672. void cred_to_ucred(struct pid *pid, const struct cred *cred,
  673. struct ucred *ucred)
  674. {
  675. ucred->pid = pid_vnr(pid);
  676. ucred->uid = ucred->gid = -1;
  677. if (cred) {
  678. struct user_namespace *current_ns = current_user_ns();
  679. ucred->uid = user_ns_map_uid(current_ns, cred, cred->euid);
  680. ucred->gid = user_ns_map_gid(current_ns, cred, cred->egid);
  681. }
  682. }
  683. EXPORT_SYMBOL_GPL(cred_to_ucred);
  684. int sock_getsockopt(struct socket *sock, int level, int optname,
  685. char __user *optval, int __user *optlen)
  686. {
  687. struct sock *sk = sock->sk;
  688. union {
  689. int val;
  690. struct linger ling;
  691. struct timeval tm;
  692. } v;
  693. int lv = sizeof(int);
  694. int len;
  695. if (get_user(len, optlen))
  696. return -EFAULT;
  697. if (len < 0)
  698. return -EINVAL;
  699. memset(&v, 0, sizeof(v));
  700. switch (optname) {
  701. case SO_DEBUG:
  702. v.val = sock_flag(sk, SOCK_DBG);
  703. break;
  704. case SO_DONTROUTE:
  705. v.val = sock_flag(sk, SOCK_LOCALROUTE);
  706. break;
  707. case SO_BROADCAST:
  708. v.val = !!sock_flag(sk, SOCK_BROADCAST);
  709. break;
  710. case SO_SNDBUF:
  711. v.val = sk->sk_sndbuf;
  712. break;
  713. case SO_RCVBUF:
  714. v.val = sk->sk_rcvbuf;
  715. break;
  716. case SO_REUSEADDR:
  717. v.val = sk->sk_reuse;
  718. break;
  719. case SO_KEEPALIVE:
  720. v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
  721. break;
  722. case SO_TYPE:
  723. v.val = sk->sk_type;
  724. break;
  725. case SO_PROTOCOL:
  726. v.val = sk->sk_protocol;
  727. break;
  728. case SO_DOMAIN:
  729. v.val = sk->sk_family;
  730. break;
  731. case SO_ERROR:
  732. v.val = -sock_error(sk);
  733. if (v.val == 0)
  734. v.val = xchg(&sk->sk_err_soft, 0);
  735. break;
  736. case SO_OOBINLINE:
  737. v.val = !!sock_flag(sk, SOCK_URGINLINE);
  738. break;
  739. case SO_NO_CHECK:
  740. v.val = sk->sk_no_check;
  741. break;
  742. case SO_PRIORITY:
  743. v.val = sk->sk_priority;
  744. break;
  745. case SO_LINGER:
  746. lv = sizeof(v.ling);
  747. v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
  748. v.ling.l_linger = sk->sk_lingertime / HZ;
  749. break;
  750. case SO_BSDCOMPAT:
  751. sock_warn_obsolete_bsdism("getsockopt");
  752. break;
  753. case SO_TIMESTAMP:
  754. v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
  755. !sock_flag(sk, SOCK_RCVTSTAMPNS);
  756. break;
  757. case SO_TIMESTAMPNS:
  758. v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
  759. break;
  760. case SO_TIMESTAMPING:
  761. v.val = 0;
  762. if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
  763. v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
  764. if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
  765. v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
  766. if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
  767. v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
  768. if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
  769. v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
  770. if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
  771. v.val |= SOF_TIMESTAMPING_SOFTWARE;
  772. if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
  773. v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
  774. if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
  775. v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
  776. break;
  777. case SO_RCVTIMEO:
  778. lv = sizeof(struct timeval);
  779. if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
  780. v.tm.tv_sec = 0;
  781. v.tm.tv_usec = 0;
  782. } else {
  783. v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
  784. v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
  785. }
  786. break;
  787. case SO_SNDTIMEO:
  788. lv = sizeof(struct timeval);
  789. if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
  790. v.tm.tv_sec = 0;
  791. v.tm.tv_usec = 0;
  792. } else {
  793. v.tm.tv_sec = sk->sk_sndtimeo / HZ;
  794. v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
  795. }
  796. break;
  797. case SO_RCVLOWAT:
  798. v.val = sk->sk_rcvlowat;
  799. break;
  800. case SO_SNDLOWAT:
  801. v.val = 1;
  802. break;
  803. case SO_PASSCRED:
  804. v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
  805. break;
  806. case SO_PEERCRED:
  807. {
  808. struct ucred peercred;
  809. if (len > sizeof(peercred))
  810. len = sizeof(peercred);
  811. cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
  812. if (copy_to_user(optval, &peercred, len))
  813. return -EFAULT;
  814. goto lenout;
  815. }
  816. case SO_PEERNAME:
  817. {
  818. char address[128];
  819. if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
  820. return -ENOTCONN;
  821. if (lv < len)
  822. return -EINVAL;
  823. if (copy_to_user(optval, address, len))
  824. return -EFAULT;
  825. goto lenout;
  826. }
  827. /* Dubious BSD thing... Probably nobody even uses it, but
  828. * the UNIX standard wants it for whatever reason... -DaveM
  829. */
  830. case SO_ACCEPTCONN:
  831. v.val = sk->sk_state == TCP_LISTEN;
  832. break;
  833. case SO_PASSSEC:
  834. v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
  835. break;
  836. case SO_PEERSEC:
  837. return security_socket_getpeersec_stream(sock, optval, optlen, len);
  838. case SO_MARK:
  839. v.val = sk->sk_mark;
  840. break;
  841. case SO_RXQ_OVFL:
  842. v.val = !!sock_flag(sk, SOCK_RXQ_OVFL);
  843. break;
  844. case SO_WIFI_STATUS:
  845. v.val = !!sock_flag(sk, SOCK_WIFI_STATUS);
  846. break;
  847. default:
  848. return -ENOPROTOOPT;
  849. }
  850. if (len > lv)
  851. len = lv;
  852. if (copy_to_user(optval, &v, len))
  853. return -EFAULT;
  854. lenout:
  855. if (put_user(len, optlen))
  856. return -EFAULT;
  857. return 0;
  858. }
  859. /*
  860. * Initialize an sk_lock.
  861. *
  862. * (We also register the sk_lock with the lock validator.)
  863. */
  864. static inline void sock_lock_init(struct sock *sk)
  865. {
  866. sock_lock_init_class_and_name(sk,
  867. af_family_slock_key_strings[sk->sk_family],
  868. af_family_slock_keys + sk->sk_family,
  869. af_family_key_strings[sk->sk_family],
  870. af_family_keys + sk->sk_family);
  871. }
  872. /*
  873. * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
  874. * even temporarly, because of RCU lookups. sk_node should also be left as is.
  875. * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
  876. */
  877. static void sock_copy(struct sock *nsk, const struct sock *osk)
  878. {
  879. #ifdef CONFIG_SECURITY_NETWORK
  880. void *sptr = nsk->sk_security;
  881. #endif
  882. memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
  883. memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
  884. osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
  885. #ifdef CONFIG_SECURITY_NETWORK
  886. nsk->sk_security = sptr;
  887. security_sk_clone(osk, nsk);
  888. #endif
  889. }
  890. /*
  891. * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
  892. * un-modified. Special care is taken when initializing object to zero.
  893. */
  894. static inline void sk_prot_clear_nulls(struct sock *sk, int size)
  895. {
  896. if (offsetof(struct sock, sk_node.next) != 0)
  897. memset(sk, 0, offsetof(struct sock, sk_node.next));
  898. memset(&sk->sk_node.pprev, 0,
  899. size - offsetof(struct sock, sk_node.pprev));
  900. }
  901. void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
  902. {
  903. unsigned long nulls1, nulls2;
  904. nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
  905. nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
  906. if (nulls1 > nulls2)
  907. swap(nulls1, nulls2);
  908. if (nulls1 != 0)
  909. memset((char *)sk, 0, nulls1);
  910. memset((char *)sk + nulls1 + sizeof(void *), 0,
  911. nulls2 - nulls1 - sizeof(void *));
  912. memset((char *)sk + nulls2 + sizeof(void *), 0,
  913. size - nulls2 - sizeof(void *));
  914. }
  915. EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
  916. static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
  917. int family)
  918. {
  919. struct sock *sk;
  920. struct kmem_cache *slab;
  921. slab = prot->slab;
  922. if (slab != NULL) {
  923. sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
  924. if (!sk)
  925. return sk;
  926. if (priority & __GFP_ZERO) {
  927. if (prot->clear_sk)
  928. prot->clear_sk(sk, prot->obj_size);
  929. else
  930. sk_prot_clear_nulls(sk, prot->obj_size);
  931. }
  932. } else
  933. sk = kmalloc(prot->obj_size, priority);
  934. if (sk != NULL) {
  935. kmemcheck_annotate_bitfield(sk, flags);
  936. if (security_sk_alloc(sk, family, priority))
  937. goto out_free;
  938. if (!try_module_get(prot->owner))
  939. goto out_free_sec;
  940. sk_tx_queue_clear(sk);
  941. }
  942. return sk;
  943. out_free_sec:
  944. security_sk_free(sk);
  945. out_free:
  946. if (slab != NULL)
  947. kmem_cache_free(slab, sk);
  948. else
  949. kfree(sk);
  950. return NULL;
  951. }
  952. static void sk_prot_free(struct proto *prot, struct sock *sk)
  953. {
  954. struct kmem_cache *slab;
  955. struct module *owner;
  956. owner = prot->owner;
  957. slab = prot->slab;
  958. security_sk_free(sk);
  959. if (slab != NULL)
  960. kmem_cache_free(slab, sk);
  961. else
  962. kfree(sk);
  963. module_put(owner);
  964. }
  965. #ifdef CONFIG_CGROUPS
  966. void sock_update_classid(struct sock *sk)
  967. {
  968. u32 classid;
  969. rcu_read_lock(); /* doing current task, which cannot vanish. */
  970. classid = task_cls_classid(current);
  971. rcu_read_unlock();
  972. if (classid && classid != sk->sk_classid)
  973. sk->sk_classid = classid;
  974. }
  975. EXPORT_SYMBOL(sock_update_classid);
  976. void sock_update_netprioidx(struct sock *sk)
  977. {
  978. struct cgroup_netprio_state *state;
  979. if (in_interrupt())
  980. return;
  981. rcu_read_lock();
  982. state = task_netprio_state(current);
  983. sk->sk_cgrp_prioidx = state ? state->prioidx : 0;
  984. rcu_read_unlock();
  985. }
  986. EXPORT_SYMBOL_GPL(sock_update_netprioidx);
  987. #endif
  988. /**
  989. * sk_alloc - All socket objects are allocated here
  990. * @net: the applicable net namespace
  991. * @family: protocol family
  992. * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
  993. * @prot: struct proto associated with this new sock instance
  994. */
  995. struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
  996. struct proto *prot)
  997. {
  998. struct sock *sk;
  999. sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
  1000. if (sk) {
  1001. sk->sk_family = family;
  1002. /*
  1003. * See comment in struct sock definition to understand
  1004. * why we need sk_prot_creator -acme
  1005. */
  1006. sk->sk_prot = sk->sk_prot_creator = prot;
  1007. sock_lock_init(sk);
  1008. sock_net_set(sk, get_net(net));
  1009. atomic_set(&sk->sk_wmem_alloc, 1);
  1010. sock_update_classid(sk);
  1011. sock_update_netprioidx(sk);
  1012. }
  1013. return sk;
  1014. }
  1015. EXPORT_SYMBOL(sk_alloc);
  1016. static void __sk_free(struct sock *sk)
  1017. {
  1018. struct sk_filter *filter;
  1019. if (sk->sk_destruct)
  1020. sk->sk_destruct(sk);
  1021. filter = rcu_dereference_check(sk->sk_filter,
  1022. atomic_read(&sk->sk_wmem_alloc) == 0);
  1023. if (filter) {
  1024. sk_filter_uncharge(sk, filter);
  1025. RCU_INIT_POINTER(sk->sk_filter, NULL);
  1026. }
  1027. sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
  1028. if (atomic_read(&sk->sk_omem_alloc))
  1029. printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
  1030. __func__, atomic_read(&sk->sk_omem_alloc));
  1031. if (sk->sk_peer_cred)
  1032. put_cred(sk->sk_peer_cred);
  1033. put_pid(sk->sk_peer_pid);
  1034. put_net(sock_net(sk));
  1035. sk_prot_free(sk->sk_prot_creator, sk);
  1036. }
  1037. void sk_free(struct sock *sk)
  1038. {
  1039. /*
  1040. * We subtract one from sk_wmem_alloc and can know if
  1041. * some packets are still in some tx queue.
  1042. * If not null, sock_wfree() will call __sk_free(sk) later
  1043. */
  1044. if (atomic_dec_and_test(&sk->sk_wmem_alloc))
  1045. __sk_free(sk);
  1046. }
  1047. EXPORT_SYMBOL(sk_free);
  1048. /*
  1049. * Last sock_put should drop reference to sk->sk_net. It has already
  1050. * been dropped in sk_change_net. Taking reference to stopping namespace
  1051. * is not an option.
  1052. * Take reference to a socket to remove it from hash _alive_ and after that
  1053. * destroy it in the context of init_net.
  1054. */
  1055. void sk_release_kernel(struct sock *sk)
  1056. {
  1057. if (sk == NULL || sk->sk_socket == NULL)
  1058. return;
  1059. sock_hold(sk);
  1060. sock_release(sk->sk_socket);
  1061. release_net(sock_net(sk));
  1062. sock_net_set(sk, get_net(&init_net));
  1063. sock_put(sk);
  1064. }
  1065. EXPORT_SYMBOL(sk_release_kernel);
  1066. /**
  1067. * sk_clone_lock - clone a socket, and lock its clone
  1068. * @sk: the socket to clone
  1069. * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
  1070. *
  1071. * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
  1072. */
  1073. struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
  1074. {
  1075. struct sock *newsk;
  1076. newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
  1077. if (newsk != NULL) {
  1078. struct sk_filter *filter;
  1079. sock_copy(newsk, sk);
  1080. /* SANITY */
  1081. get_net(sock_net(newsk));
  1082. sk_node_init(&newsk->sk_node);
  1083. sock_lock_init(newsk);
  1084. bh_lock_sock(newsk);
  1085. newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
  1086. newsk->sk_backlog.len = 0;
  1087. atomic_set(&newsk->sk_rmem_alloc, 0);
  1088. /*
  1089. * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
  1090. */
  1091. atomic_set(&newsk->sk_wmem_alloc, 1);
  1092. atomic_set(&newsk->sk_omem_alloc, 0);
  1093. skb_queue_head_init(&newsk->sk_receive_queue);
  1094. skb_queue_head_init(&newsk->sk_write_queue);
  1095. #ifdef CONFIG_NET_DMA
  1096. skb_queue_head_init(&newsk->sk_async_wait_queue);
  1097. #endif
  1098. spin_lock_init(&newsk->sk_dst_lock);
  1099. rwlock_init(&newsk->sk_callback_lock);
  1100. lockdep_set_class_and_name(&newsk->sk_callback_lock,
  1101. af_callback_keys + newsk->sk_family,
  1102. af_family_clock_key_strings[newsk->sk_family]);
  1103. newsk->sk_dst_cache = NULL;
  1104. newsk->sk_wmem_queued = 0;
  1105. newsk->sk_forward_alloc = 0;
  1106. newsk->sk_send_head = NULL;
  1107. newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
  1108. sock_reset_flag(newsk, SOCK_DONE);
  1109. skb_queue_head_init(&newsk->sk_error_queue);
  1110. filter = rcu_dereference_protected(newsk->sk_filter, 1);
  1111. if (filter != NULL)
  1112. sk_filter_charge(newsk, filter);
  1113. if (unlikely(xfrm_sk_clone_policy(newsk))) {
  1114. /* It is still raw copy of parent, so invalidate
  1115. * destructor and make plain sk_free() */
  1116. newsk->sk_destruct = NULL;
  1117. bh_unlock_sock(newsk);
  1118. sk_free(newsk);
  1119. newsk = NULL;
  1120. goto out;
  1121. }
  1122. newsk->sk_err = 0;
  1123. newsk->sk_priority = 0;
  1124. /*
  1125. * Before updating sk_refcnt, we must commit prior changes to memory
  1126. * (Documentation/RCU/rculist_nulls.txt for details)
  1127. */
  1128. smp_wmb();
  1129. atomic_set(&newsk->sk_refcnt, 2);
  1130. /*
  1131. * Increment the counter in the same struct proto as the master
  1132. * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
  1133. * is the same as sk->sk_prot->socks, as this field was copied
  1134. * with memcpy).
  1135. *
  1136. * This _changes_ the previous behaviour, where
  1137. * tcp_create_openreq_child always was incrementing the
  1138. * equivalent to tcp_prot->socks (inet_sock_nr), so this have
  1139. * to be taken into account in all callers. -acme
  1140. */
  1141. sk_refcnt_debug_inc(newsk);
  1142. sk_set_socket(newsk, NULL);
  1143. newsk->sk_wq = NULL;
  1144. if (newsk->sk_prot->sockets_allocated)
  1145. sk_sockets_allocated_inc(newsk);
  1146. if (newsk->sk_flags & SK_FLAGS_TIMESTAMP)
  1147. net_enable_timestamp();
  1148. }
  1149. out:
  1150. return newsk;
  1151. }
  1152. EXPORT_SYMBOL_GPL(sk_clone_lock);
  1153. void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
  1154. {
  1155. __sk_dst_set(sk, dst);
  1156. sk->sk_route_caps = dst->dev->features;
  1157. if (sk->sk_route_caps & NETIF_F_GSO)
  1158. sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
  1159. sk->sk_route_caps &= ~sk->sk_route_nocaps;
  1160. if (sk_can_gso(sk)) {
  1161. if (dst->header_len) {
  1162. sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
  1163. } else {
  1164. sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
  1165. sk->sk_gso_max_size = dst->dev->gso_max_size;
  1166. }
  1167. }
  1168. }
  1169. EXPORT_SYMBOL_GPL(sk_setup_caps);
  1170. void __init sk_init(void)
  1171. {
  1172. if (totalram_pages <= 4096) {
  1173. sysctl_wmem_max = 32767;
  1174. sysctl_rmem_max = 32767;
  1175. sysctl_wmem_default = 32767;
  1176. sysctl_rmem_default = 32767;
  1177. } else if (totalram_pages >= 131072) {
  1178. sysctl_wmem_max = 131071;
  1179. sysctl_rmem_max = 131071;
  1180. }
  1181. }
  1182. /*
  1183. * Simple resource managers for sockets.
  1184. */
  1185. /*
  1186. * Write buffer destructor automatically called from kfree_skb.
  1187. */
  1188. void sock_wfree(struct sk_buff *skb)
  1189. {
  1190. struct sock *sk = skb->sk;
  1191. unsigned int len = skb->truesize;
  1192. if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
  1193. /*
  1194. * Keep a reference on sk_wmem_alloc, this will be released
  1195. * after sk_write_space() call
  1196. */
  1197. atomic_sub(len - 1, &sk->sk_wmem_alloc);
  1198. sk->sk_write_space(sk);
  1199. len = 1;
  1200. }
  1201. /*
  1202. * if sk_wmem_alloc reaches 0, we must finish what sk_free()
  1203. * could not do because of in-flight packets
  1204. */
  1205. if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
  1206. __sk_free(sk);
  1207. }
  1208. EXPORT_SYMBOL(sock_wfree);
  1209. /*
  1210. * Read buffer destructor automatically called from kfree_skb.
  1211. */
  1212. void sock_rfree(struct sk_buff *skb)
  1213. {
  1214. struct sock *sk = skb->sk;
  1215. unsigned int len = skb->truesize;
  1216. atomic_sub(len, &sk->sk_rmem_alloc);
  1217. sk_mem_uncharge(sk, len);
  1218. }
  1219. EXPORT_SYMBOL(sock_rfree);
  1220. int sock_i_uid(struct sock *sk)
  1221. {
  1222. int uid;
  1223. read_lock_bh(&sk->sk_callback_lock);
  1224. uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
  1225. read_unlock_bh(&sk->sk_callback_lock);
  1226. return uid;
  1227. }
  1228. EXPORT_SYMBOL(sock_i_uid);
  1229. unsigned long sock_i_ino(struct sock *sk)
  1230. {
  1231. unsigned long ino;
  1232. read_lock_bh(&sk->sk_callback_lock);
  1233. ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
  1234. read_unlock_bh(&sk->sk_callback_lock);
  1235. return ino;
  1236. }
  1237. EXPORT_SYMBOL(sock_i_ino);
  1238. /*
  1239. * Allocate a skb from the socket's send buffer.
  1240. */
  1241. struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
  1242. gfp_t priority)
  1243. {
  1244. if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
  1245. struct sk_buff *skb = alloc_skb(size, priority);
  1246. if (skb) {
  1247. skb_set_owner_w(skb, sk);
  1248. return skb;
  1249. }
  1250. }
  1251. return NULL;
  1252. }
  1253. EXPORT_SYMBOL(sock_wmalloc);
  1254. /*
  1255. * Allocate a skb from the socket's receive buffer.
  1256. */
  1257. struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
  1258. gfp_t priority)
  1259. {
  1260. if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
  1261. struct sk_buff *skb = alloc_skb(size, priority);
  1262. if (skb) {
  1263. skb_set_owner_r(skb, sk);
  1264. return skb;
  1265. }
  1266. }
  1267. return NULL;
  1268. }
  1269. /*
  1270. * Allocate a memory block from the socket's option memory buffer.
  1271. */
  1272. void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
  1273. {
  1274. if ((unsigned)size <= sysctl_optmem_max &&
  1275. atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
  1276. void *mem;
  1277. /* First do the add, to avoid the race if kmalloc
  1278. * might sleep.
  1279. */
  1280. atomic_add(size, &sk->sk_omem_alloc);
  1281. mem = kmalloc(size, priority);
  1282. if (mem)
  1283. return mem;
  1284. atomic_sub(size, &sk->sk_omem_alloc);
  1285. }
  1286. return NULL;
  1287. }
  1288. EXPORT_SYMBOL(sock_kmalloc);
  1289. /*
  1290. * Free an option memory block.
  1291. */
  1292. void sock_kfree_s(struct sock *sk, void *mem, int size)
  1293. {
  1294. kfree(mem);
  1295. atomic_sub(size, &sk->sk_omem_alloc);
  1296. }
  1297. EXPORT_SYMBOL(sock_kfree_s);
  1298. /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
  1299. I think, these locks should be removed for datagram sockets.
  1300. */
  1301. static long sock_wait_for_wmem(struct sock *sk, long timeo)
  1302. {
  1303. DEFINE_WAIT(wait);
  1304. clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
  1305. for (;;) {
  1306. if (!timeo)
  1307. break;
  1308. if (signal_pending(current))
  1309. break;
  1310. set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
  1311. prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
  1312. if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
  1313. break;
  1314. if (sk->sk_shutdown & SEND_SHUTDOWN)
  1315. break;
  1316. if (sk->sk_err)
  1317. break;
  1318. timeo = schedule_timeout(timeo);
  1319. }
  1320. finish_wait(sk_sleep(sk), &wait);
  1321. return timeo;
  1322. }
  1323. /*
  1324. * Generic send/receive buffer handlers
  1325. */
  1326. struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
  1327. unsigned long data_len, int noblock,
  1328. int *errcode)
  1329. {
  1330. struct sk_buff *skb;
  1331. gfp_t gfp_mask;
  1332. long timeo;
  1333. int err;
  1334. gfp_mask = sk->sk_allocation;
  1335. if (gfp_mask & __GFP_WAIT)
  1336. gfp_mask |= __GFP_REPEAT;
  1337. timeo = sock_sndtimeo(sk, noblock);
  1338. while (1) {
  1339. err = sock_error(sk);
  1340. if (err != 0)
  1341. goto failure;
  1342. err = -EPIPE;
  1343. if (sk->sk_shutdown & SEND_SHUTDOWN)
  1344. goto failure;
  1345. if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
  1346. skb = alloc_skb(header_len, gfp_mask);
  1347. if (skb) {
  1348. int npages;
  1349. int i;
  1350. /* No pages, we're done... */
  1351. if (!data_len)
  1352. break;
  1353. npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
  1354. skb->truesize += data_len;
  1355. skb_shinfo(skb)->nr_frags = npages;
  1356. for (i = 0; i < npages; i++) {
  1357. struct page *page;
  1358. page = alloc_pages(sk->sk_allocation, 0);
  1359. if (!page) {
  1360. err = -ENOBUFS;
  1361. skb_shinfo(skb)->nr_frags = i;
  1362. kfree_skb(skb);
  1363. goto failure;
  1364. }
  1365. __skb_fill_page_desc(skb, i,
  1366. page, 0,
  1367. (data_len >= PAGE_SIZE ?
  1368. PAGE_SIZE :
  1369. data_len));
  1370. data_len -= PAGE_SIZE;
  1371. }
  1372. /* Full success... */
  1373. break;
  1374. }
  1375. err = -ENOBUFS;
  1376. goto failure;
  1377. }
  1378. set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
  1379. set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
  1380. err = -EAGAIN;
  1381. if (!timeo)
  1382. goto failure;
  1383. if (signal_pending(current))
  1384. goto interrupted;
  1385. timeo = sock_wait_for_wmem(sk, timeo);
  1386. }
  1387. skb_set_owner_w(skb, sk);
  1388. return skb;
  1389. interrupted:
  1390. err = sock_intr_errno(timeo);
  1391. failure:
  1392. *errcode = err;
  1393. return NULL;
  1394. }
  1395. EXPORT_SYMBOL(sock_alloc_send_pskb);
  1396. struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
  1397. int noblock, int *errcode)
  1398. {
  1399. return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
  1400. }
  1401. EXPORT_SYMBOL(sock_alloc_send_skb);
  1402. static void __lock_sock(struct sock *sk)
  1403. __releases(&sk->sk_lock.slock)
  1404. __acquires(&sk->sk_lock.slock)
  1405. {
  1406. DEFINE_WAIT(wait);
  1407. for (;;) {
  1408. prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
  1409. TASK_UNINTERRUPTIBLE);
  1410. spin_unlock_bh(&sk->sk_lock.slock);
  1411. schedule();
  1412. spin_lock_bh(&sk->sk_lock.slock);
  1413. if (!sock_owned_by_user(sk))
  1414. break;
  1415. }
  1416. finish_wait(&sk->sk_lock.wq, &wait);
  1417. }
  1418. static void __release_sock(struct sock *sk)
  1419. __releases(&sk->sk_lock.slock)
  1420. __acquires(&sk->sk_lock.slock)
  1421. {
  1422. struct sk_buff *skb = sk->sk_backlog.head;
  1423. do {
  1424. sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
  1425. bh_unlock_sock(sk);
  1426. do {
  1427. struct sk_buff *next = skb->next;
  1428. WARN_ON_ONCE(skb_dst_is_noref(skb));
  1429. skb->next = NULL;
  1430. sk_backlog_rcv(sk, skb);
  1431. /*
  1432. * We are in process context here with softirqs
  1433. * disabled, use cond_resched_softirq() to preempt.
  1434. * This is safe to do because we've taken the backlog
  1435. * queue private:
  1436. */
  1437. cond_resched_softirq();
  1438. skb = next;
  1439. } while (skb != NULL);
  1440. bh_lock_sock(sk);
  1441. } while ((skb = sk->sk_backlog.head) != NULL);
  1442. /*
  1443. * Doing the zeroing here guarantee we can not loop forever
  1444. * while a wild producer attempts to flood us.
  1445. */
  1446. sk->sk_backlog.len = 0;
  1447. }
  1448. /**
  1449. * sk_wait_data - wait for data to arrive at sk_receive_queue
  1450. * @sk: sock to wait on
  1451. * @timeo: for how long
  1452. *
  1453. * Now socket state including sk->sk_err is changed only under lock,
  1454. * hence we may omit checks after joining wait queue.
  1455. * We check receive queue before schedule() only as optimization;
  1456. * it is very likely that release_sock() added new data.
  1457. */
  1458. int sk_wait_data(struct sock *sk, long *timeo)
  1459. {
  1460. int rc;
  1461. DEFINE_WAIT(wait);
  1462. prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
  1463. set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
  1464. rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
  1465. clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
  1466. finish_wait(sk_sleep(sk), &wait);
  1467. return rc;
  1468. }
  1469. EXPORT_SYMBOL(sk_wait_data);
  1470. /**
  1471. * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
  1472. * @sk: socket
  1473. * @size: memory size to allocate
  1474. * @kind: allocation type
  1475. *
  1476. * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
  1477. * rmem allocation. This function assumes that protocols which have
  1478. * memory_pressure use sk_wmem_queued as write buffer accounting.
  1479. */
  1480. int __sk_mem_schedule(struct sock *sk, int size, int kind)
  1481. {
  1482. struct proto *prot = sk->sk_prot;
  1483. int amt = sk_mem_pages(size);
  1484. long allocated;
  1485. sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
  1486. allocated = sk_memory_allocated_add(sk, amt);
  1487. /* Under limit. */
  1488. if (allocated <= sk_prot_mem_limits(sk, 0)) {
  1489. sk_leave_memory_pressure(sk);
  1490. return 1;
  1491. }
  1492. /* Under pressure. */
  1493. if (allocated > sk_prot_mem_limits(sk, 1))
  1494. sk_enter_memory_pressure(sk);
  1495. /* Over hard limit. */
  1496. if (allocated > sk_prot_mem_limits(sk, 2))
  1497. goto suppress_allocation;
  1498. /* guarantee minimum buffer size under pressure */
  1499. if (kind == SK_MEM_RECV) {
  1500. if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
  1501. return 1;
  1502. } else { /* SK_MEM_SEND */
  1503. if (sk->sk_type == SOCK_STREAM) {
  1504. if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
  1505. return 1;
  1506. } else if (atomic_read(&sk->sk_wmem_alloc) <
  1507. prot->sysctl_wmem[0])
  1508. return 1;
  1509. }
  1510. if (sk_has_memory_pressure(sk)) {
  1511. int alloc;
  1512. if (!sk_under_memory_pressure(sk))
  1513. return 1;
  1514. alloc = sk_sockets_allocated_read_positive(sk);
  1515. if (sk_prot_mem_limits(sk, 2) > alloc *
  1516. sk_mem_pages(sk->sk_wmem_queued +
  1517. atomic_read(&sk->sk_rmem_alloc) +
  1518. sk->sk_forward_alloc))
  1519. return 1;
  1520. }
  1521. suppress_allocation:
  1522. if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
  1523. sk_stream_moderate_sndbuf(sk);
  1524. /* Fail only if socket is _under_ its sndbuf.
  1525. * In this case we cannot block, so that we have to fail.
  1526. */
  1527. if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
  1528. return 1;
  1529. }
  1530. trace_sock_exceed_buf_limit(sk, prot, allocated);
  1531. /* Alas. Undo changes. */
  1532. sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
  1533. sk_memory_allocated_sub(sk, amt);
  1534. return 0;
  1535. }
  1536. EXPORT_SYMBOL(__sk_mem_schedule);
  1537. /**
  1538. * __sk_reclaim - reclaim memory_allocated
  1539. * @sk: socket
  1540. */
  1541. void __sk_mem_reclaim(struct sock *sk)
  1542. {
  1543. sk_memory_allocated_sub(sk,
  1544. sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT);
  1545. sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
  1546. if (sk_under_memory_pressure(sk) &&
  1547. (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
  1548. sk_leave_memory_pressure(sk);
  1549. }
  1550. EXPORT_SYMBOL(__sk_mem_reclaim);
  1551. /*
  1552. * Set of default routines for initialising struct proto_ops when
  1553. * the protocol does not support a particular function. In certain
  1554. * cases where it makes no sense for a protocol to have a "do nothing"
  1555. * function, some default processing is provided.
  1556. */
  1557. int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
  1558. {
  1559. return -EOPNOTSUPP;
  1560. }
  1561. EXPORT_SYMBOL(sock_no_bind);
  1562. int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
  1563. int len, int flags)
  1564. {
  1565. return -EOPNOTSUPP;
  1566. }
  1567. EXPORT_SYMBOL(sock_no_connect);
  1568. int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
  1569. {
  1570. return -EOPNOTSUPP;
  1571. }
  1572. EXPORT_SYMBOL(sock_no_socketpair);
  1573. int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
  1574. {
  1575. return -EOPNOTSUPP;
  1576. }
  1577. EXPORT_SYMBOL(sock_no_accept);
  1578. int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
  1579. int *len, int peer)
  1580. {
  1581. return -EOPNOTSUPP;
  1582. }
  1583. EXPORT_SYMBOL(sock_no_getname);
  1584. unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
  1585. {
  1586. return 0;
  1587. }
  1588. EXPORT_SYMBOL(sock_no_poll);
  1589. int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
  1590. {
  1591. return -EOPNOTSUPP;
  1592. }
  1593. EXPORT_SYMBOL(sock_no_ioctl);
  1594. int sock_no_listen(struct socket *sock, int backlog)
  1595. {
  1596. return -EOPNOTSUPP;
  1597. }
  1598. EXPORT_SYMBOL(sock_no_listen);
  1599. int sock_no_shutdown(struct socket *sock, int how)
  1600. {
  1601. return -EOPNOTSUPP;
  1602. }
  1603. EXPORT_SYMBOL(sock_no_shutdown);
  1604. int sock_no_setsockopt(struct socket *sock, int level, int optname,
  1605. char __user *optval, unsigned int optlen)
  1606. {
  1607. return -EOPNOTSUPP;
  1608. }
  1609. EXPORT_SYMBOL(sock_no_setsockopt);
  1610. int sock_no_getsockopt(struct socket *sock, int level, int optname,
  1611. char __user *optval, int __user *optlen)
  1612. {
  1613. return -EOPNOTSUPP;
  1614. }
  1615. EXPORT_SYMBOL(sock_no_getsockopt);
  1616. int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
  1617. size_t len)
  1618. {
  1619. return -EOPNOTSUPP;
  1620. }
  1621. EXPORT_SYMBOL(sock_no_sendmsg);
  1622. int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
  1623. size_t len, int flags)
  1624. {
  1625. return -EOPNOTSUPP;
  1626. }
  1627. EXPORT_SYMBOL(sock_no_recvmsg);
  1628. int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
  1629. {
  1630. /* Mirror missing mmap method error code */
  1631. return -ENODEV;
  1632. }
  1633. EXPORT_SYMBOL(sock_no_mmap);
  1634. ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
  1635. {
  1636. ssize_t res;
  1637. struct msghdr msg = {.msg_flags = flags};
  1638. struct kvec iov;
  1639. char *kaddr = kmap(page);
  1640. iov.iov_base = kaddr + offset;
  1641. iov.iov_len = size;
  1642. res = kernel_sendmsg(sock, &msg, &iov, 1, size);
  1643. kunmap(page);
  1644. return res;
  1645. }
  1646. EXPORT_SYMBOL(sock_no_sendpage);
  1647. /*
  1648. * Default Socket Callbacks
  1649. */
  1650. static void sock_def_wakeup(struct sock *sk)
  1651. {
  1652. struct socket_wq *wq;
  1653. rcu_read_lock();
  1654. wq = rcu_dereference(sk->sk_wq);
  1655. if (wq_has_sleeper(wq))
  1656. wake_up_interruptible_all(&wq->wait);
  1657. rcu_read_unlock();
  1658. }
  1659. static void sock_def_error_report(struct sock *sk)
  1660. {
  1661. struct socket_wq *wq;
  1662. rcu_read_lock();
  1663. wq = rcu_dereference(sk->sk_wq);
  1664. if (wq_has_sleeper(wq))
  1665. wake_up_interruptible_poll(&wq->wait, POLLERR);
  1666. sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
  1667. rcu_read_unlock();
  1668. }
  1669. static void sock_def_readable(struct sock *sk, int len)
  1670. {
  1671. struct socket_wq *wq;
  1672. rcu_read_lock();
  1673. wq = rcu_dereference(sk->sk_wq);
  1674. if (wq_has_sleeper(wq))
  1675. wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
  1676. POLLRDNORM | POLLRDBAND);
  1677. sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
  1678. rcu_read_unlock();
  1679. }
  1680. static void sock_def_write_space(struct sock *sk)
  1681. {
  1682. struct socket_wq *wq;
  1683. rcu_read_lock();
  1684. /* Do not wake up a writer until he can make "significant"
  1685. * progress. --DaveM
  1686. */
  1687. if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
  1688. wq = rcu_dereference(sk->sk_wq);
  1689. if (wq_has_sleeper(wq))
  1690. wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
  1691. POLLWRNORM | POLLWRBAND);
  1692. /* Should agree with poll, otherwise some programs break */
  1693. if (sock_writeable(sk))
  1694. sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
  1695. }
  1696. rcu_read_unlock();
  1697. }
  1698. static void sock_def_destruct(struct sock *sk)
  1699. {
  1700. kfree(sk->sk_protinfo);
  1701. }
  1702. void sk_send_sigurg(struct sock *sk)
  1703. {
  1704. if (sk->sk_socket && sk->sk_socket->file)
  1705. if (send_sigurg(&sk->sk_socket->file->f_owner))
  1706. sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
  1707. }
  1708. EXPORT_SYMBOL(sk_send_sigurg);
  1709. void sk_reset_timer(struct sock *sk, struct timer_list* timer,
  1710. unsigned long expires)
  1711. {
  1712. if (!mod_timer(timer, expires))
  1713. sock_hold(sk);
  1714. }
  1715. EXPORT_SYMBOL(sk_reset_timer);
  1716. void sk_stop_timer(struct sock *sk, struct timer_list* timer)
  1717. {
  1718. if (timer_pending(timer) && del_timer(timer))
  1719. __sock_put(sk);
  1720. }
  1721. EXPORT_SYMBOL(sk_stop_timer);
  1722. void sock_init_data(struct socket *sock, struct sock *sk)
  1723. {
  1724. skb_queue_head_init(&sk->sk_receive_queue);
  1725. skb_queue_head_init(&sk->sk_write_queue);
  1726. skb_queue_head_init(&sk->sk_error_queue);
  1727. #ifdef CONFIG_NET_DMA
  1728. skb_queue_head_init(&sk->sk_async_wait_queue);
  1729. #endif
  1730. sk->sk_send_head = NULL;
  1731. init_timer(&sk->sk_timer);
  1732. sk->sk_allocation = GFP_KERNEL;
  1733. sk->sk_rcvbuf = sysctl_rmem_default;
  1734. sk->sk_sndbuf = sysctl_wmem_default;
  1735. sk->sk_state = TCP_CLOSE;
  1736. sk_set_socket(sk, sock);
  1737. sock_set_flag(sk, SOCK_ZAPPED);
  1738. if (sock) {
  1739. sk->sk_type = sock->type;
  1740. sk->sk_wq = sock->wq;
  1741. sock->sk = sk;
  1742. } else
  1743. sk->sk_wq = NULL;
  1744. spin_lock_init(&sk->sk_dst_lock);
  1745. rwlock_init(&sk->sk_callback_lock);
  1746. lockdep_set_class_and_name(&sk->sk_callback_lock,
  1747. af_callback_keys + sk->sk_family,
  1748. af_family_clock_key_strings[sk->sk_family]);
  1749. sk->sk_state_change = sock_def_wakeup;
  1750. sk->sk_data_ready = sock_def_readable;
  1751. sk->sk_write_space = sock_def_write_space;
  1752. sk->sk_error_report = sock_def_error_report;
  1753. sk->sk_destruct = sock_def_destruct;
  1754. sk->sk_sndmsg_page = NULL;
  1755. sk->sk_sndmsg_off = 0;
  1756. sk->sk_peer_pid = NULL;
  1757. sk->sk_peer_cred = NULL;
  1758. sk->sk_write_pending = 0;
  1759. sk->sk_rcvlowat = 1;
  1760. sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
  1761. sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
  1762. sk->sk_stamp = ktime_set(-1L, 0);
  1763. /*
  1764. * Before updating sk_refcnt, we must commit prior changes to memory
  1765. * (Documentation/RCU/rculist_nulls.txt for details)
  1766. */
  1767. smp_wmb();
  1768. atomic_set(&sk->sk_refcnt, 1);
  1769. atomic_set(&sk->sk_drops, 0);
  1770. }
  1771. EXPORT_SYMBOL(sock_init_data);
  1772. void lock_sock_nested(struct sock *sk, int subclass)
  1773. {
  1774. might_sleep();
  1775. spin_lock_bh(&sk->sk_lock.slock);
  1776. if (sk->sk_lock.owned)
  1777. __lock_sock(sk);
  1778. sk->sk_lock.owned = 1;
  1779. spin_unlock(&sk->sk_lock.slock);
  1780. /*
  1781. * The sk_lock has mutex_lock() semantics here:
  1782. */
  1783. mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
  1784. local_bh_enable();
  1785. }
  1786. EXPORT_SYMBOL(lock_sock_nested);
  1787. void release_sock(struct sock *sk)
  1788. {
  1789. /*
  1790. * The sk_lock has mutex_unlock() semantics:
  1791. */
  1792. mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
  1793. spin_lock_bh(&sk->sk_lock.slock);
  1794. if (sk->sk_backlog.tail)
  1795. __release_sock(sk);
  1796. sk->sk_lock.owned = 0;
  1797. if (waitqueue_active(&sk->sk_lock.wq))
  1798. wake_up(&sk->sk_lock.wq);
  1799. spin_unlock_bh(&sk->sk_lock.slock);
  1800. }
  1801. EXPORT_SYMBOL(release_sock);
  1802. /**
  1803. * lock_sock_fast - fast version of lock_sock
  1804. * @sk: socket
  1805. *
  1806. * This version should be used for very small section, where process wont block
  1807. * return false if fast path is taken
  1808. * sk_lock.slock locked, owned = 0, BH disabled
  1809. * return true if slow path is taken
  1810. * sk_lock.slock unlocked, owned = 1, BH enabled
  1811. */
  1812. bool lock_sock_fast(struct sock *sk)
  1813. {
  1814. might_sleep();
  1815. spin_lock_bh(&sk->sk_lock.slock);
  1816. if (!sk->sk_lock.owned)
  1817. /*
  1818. * Note : We must disable BH
  1819. */
  1820. return false;
  1821. __lock_sock(sk);
  1822. sk->sk_lock.owned = 1;
  1823. spin_unlock(&sk->sk_lock.slock);
  1824. /*
  1825. * The sk_lock has mutex_lock() semantics here:
  1826. */
  1827. mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
  1828. local_bh_enable();
  1829. return true;
  1830. }
  1831. EXPORT_SYMBOL(lock_sock_fast);
  1832. int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
  1833. {
  1834. struct timeval tv;
  1835. if (!sock_flag(sk, SOCK_TIMESTAMP))
  1836. sock_enable_timestamp(sk, SOCK_TIMESTAMP);
  1837. tv = ktime_to_timeval(sk->sk_stamp);
  1838. if (tv.tv_sec == -1)
  1839. return -ENOENT;
  1840. if (tv.tv_sec == 0) {
  1841. sk->sk_stamp = ktime_get_real();
  1842. tv = ktime_to_timeval(sk->sk_stamp);
  1843. }
  1844. return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
  1845. }
  1846. EXPORT_SYMBOL(sock_get_timestamp);
  1847. int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
  1848. {
  1849. struct timespec ts;
  1850. if (!sock_flag(sk, SOCK_TIMESTAMP))
  1851. sock_enable_timestamp(sk, SOCK_TIMESTAMP);
  1852. ts = ktime_to_timespec(sk->sk_stamp);
  1853. if (ts.tv_sec == -1)
  1854. return -ENOENT;
  1855. if (ts.tv_sec == 0) {
  1856. sk->sk_stamp = ktime_get_real();
  1857. ts = ktime_to_timespec(sk->sk_stamp);
  1858. }
  1859. return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
  1860. }
  1861. EXPORT_SYMBOL(sock_get_timestampns);
  1862. void sock_enable_timestamp(struct sock *sk, int flag)
  1863. {
  1864. if (!sock_flag(sk, flag)) {
  1865. unsigned long previous_flags = sk->sk_flags;
  1866. sock_set_flag(sk, flag);
  1867. /*
  1868. * we just set one of the two flags which require net
  1869. * time stamping, but time stamping might have been on
  1870. * already because of the other one
  1871. */
  1872. if (!(previous_flags & SK_FLAGS_TIMESTAMP))
  1873. net_enable_timestamp();
  1874. }
  1875. }
  1876. /*
  1877. * Get a socket option on an socket.
  1878. *
  1879. * FIX: POSIX 1003.1g is very ambiguous here. It states that
  1880. * asynchronous errors should be reported by getsockopt. We assume
  1881. * this means if you specify SO_ERROR (otherwise whats the point of it).
  1882. */
  1883. int sock_common_getsockopt(struct socket *sock, int level, int optname,
  1884. char __user *optval, int __user *optlen)
  1885. {
  1886. struct sock *sk = sock->sk;
  1887. return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
  1888. }
  1889. EXPORT_SYMBOL(sock_common_getsockopt);
  1890. #ifdef CONFIG_COMPAT
  1891. int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
  1892. char __user *optval, int __user *optlen)
  1893. {
  1894. struct sock *sk = sock->sk;
  1895. if (sk->sk_prot->compat_getsockopt != NULL)
  1896. return sk->sk_prot->compat_getsockopt(sk, level, optname,
  1897. optval, optlen);
  1898. return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
  1899. }
  1900. EXPORT_SYMBOL(compat_sock_common_getsockopt);
  1901. #endif
  1902. int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
  1903. struct msghdr *msg, size_t size, int flags)
  1904. {
  1905. struct sock *sk = sock->sk;
  1906. int addr_len = 0;
  1907. int err;
  1908. err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
  1909. flags & ~MSG_DONTWAIT, &addr_len);
  1910. if (err >= 0)
  1911. msg->msg_namelen = addr_len;
  1912. return err;
  1913. }
  1914. EXPORT_SYMBOL(sock_common_recvmsg);
  1915. /*
  1916. * Set socket options on an inet socket.
  1917. */
  1918. int sock_common_setsockopt(struct socket *sock, int level, int optname,
  1919. char __user *optval, unsigned int optlen)
  1920. {
  1921. struct sock *sk = sock->sk;
  1922. return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
  1923. }
  1924. EXPORT_SYMBOL(sock_common_setsockopt);
  1925. #ifdef CONFIG_COMPAT
  1926. int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
  1927. char __user *optval, unsigned int optlen)
  1928. {
  1929. struct sock *sk = sock->sk;
  1930. if (sk->sk_prot->compat_setsockopt != NULL)
  1931. return sk->sk_prot->compat_setsockopt(sk, level, optname,
  1932. optval, optlen);
  1933. return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
  1934. }
  1935. EXPORT_SYMBOL(compat_sock_common_setsockopt);
  1936. #endif
  1937. void sk_common_release(struct sock *sk)
  1938. {
  1939. if (sk->sk_prot->destroy)
  1940. sk->sk_prot->destroy(sk);
  1941. /*
  1942. * Observation: when sock_common_release is called, processes have
  1943. * no access to socket. But net still has.
  1944. * Step one, detach it from networking:
  1945. *
  1946. * A. Remove from hash tables.
  1947. */
  1948. sk->sk_prot->unhash(sk);
  1949. /*
  1950. * In this point socket cannot receive new packets, but it is possible
  1951. * that some packets are in flight because some CPU runs receiver and
  1952. * did hash table lookup before we unhashed socket. They will achieve
  1953. * receive queue and will be purged by socket destructor.
  1954. *
  1955. * Also we still have packets pending on receive queue and probably,
  1956. * our own packets waiting in device queues. sock_destroy will drain
  1957. * receive queue, but transmitted packets will delay socket destruction
  1958. * until the last reference will be released.
  1959. */
  1960. sock_orphan(sk);
  1961. xfrm_sk_free_policy(sk);
  1962. sk_refcnt_debug_release(sk);
  1963. sock_put(sk);
  1964. }
  1965. EXPORT_SYMBOL(sk_common_release);
  1966. static DEFINE_RWLOCK(proto_list_lock);
  1967. static LIST_HEAD(proto_list);
  1968. #ifdef CONFIG_PROC_FS
  1969. #define PROTO_INUSE_NR 64 /* should be enough for the first time */
  1970. struct prot_inuse {
  1971. int val[PROTO_INUSE_NR];
  1972. };
  1973. static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
  1974. #ifdef CONFIG_NET_NS
  1975. void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
  1976. {
  1977. __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
  1978. }
  1979. EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
  1980. int sock_prot_inuse_get(struct net *net, struct proto *prot)
  1981. {
  1982. int cpu, idx = prot->inuse_idx;
  1983. int res = 0;
  1984. for_each_possible_cpu(cpu)
  1985. res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
  1986. return res >= 0 ? res : 0;
  1987. }
  1988. EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
  1989. static int __net_init sock_inuse_init_net(struct net *net)
  1990. {
  1991. net->core.inuse = alloc_percpu(struct prot_inuse);
  1992. return net->core.inuse ? 0 : -ENOMEM;
  1993. }
  1994. static void __net_exit sock_inuse_exit_net(struct net *net)
  1995. {
  1996. free_percpu(net->core.inuse);
  1997. }
  1998. static struct pernet_operations net_inuse_ops = {
  1999. .init = sock_inuse_init_net,
  2000. .exit = sock_inuse_exit_net,
  2001. };
  2002. static __init int net_inuse_init(void)
  2003. {
  2004. if (register_pernet_subsys(&net_inuse_ops))
  2005. panic("Cannot initialize net inuse counters");
  2006. return 0;
  2007. }
  2008. core_initcall(net_inuse_init);
  2009. #else
  2010. static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
  2011. void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
  2012. {
  2013. __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
  2014. }
  2015. EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
  2016. int sock_prot_inuse_get(struct net *net, struct proto *prot)
  2017. {
  2018. int cpu, idx = prot->inuse_idx;
  2019. int res = 0;
  2020. for_each_possible_cpu(cpu)
  2021. res += per_cpu(prot_inuse, cpu).val[idx];
  2022. return res >= 0 ? res : 0;
  2023. }
  2024. EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
  2025. #endif
  2026. static void assign_proto_idx(struct proto *prot)
  2027. {
  2028. prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
  2029. if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
  2030. printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
  2031. return;
  2032. }
  2033. set_bit(prot->inuse_idx, proto_inuse_idx);
  2034. }
  2035. static void release_proto_idx(struct proto *prot)
  2036. {
  2037. if (prot->inuse_idx != PROTO_INUSE_NR - 1)
  2038. clear_bit(prot->inuse_idx, proto_inuse_idx);
  2039. }
  2040. #else
  2041. static inline void assign_proto_idx(struct proto *prot)
  2042. {
  2043. }
  2044. static inline void release_proto_idx(struct proto *prot)
  2045. {
  2046. }
  2047. #endif
  2048. int proto_register(struct proto *prot, int alloc_slab)
  2049. {
  2050. if (alloc_slab) {
  2051. prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
  2052. SLAB_HWCACHE_ALIGN | prot->slab_flags,
  2053. NULL);
  2054. if (prot->slab == NULL) {
  2055. printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
  2056. prot->name);
  2057. goto out;
  2058. }
  2059. if (prot->rsk_prot != NULL) {
  2060. prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
  2061. if (prot->rsk_prot->slab_name == NULL)
  2062. goto out_free_sock_slab;
  2063. prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
  2064. prot->rsk_prot->obj_size, 0,
  2065. SLAB_HWCACHE_ALIGN, NULL);
  2066. if (prot->rsk_prot->slab == NULL) {
  2067. printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
  2068. prot->name);
  2069. goto out_free_request_sock_slab_name;
  2070. }
  2071. }
  2072. if (prot->twsk_prot != NULL) {
  2073. prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
  2074. if (prot->twsk_prot->twsk_slab_name == NULL)
  2075. goto out_free_request_sock_slab;
  2076. prot->twsk_prot->twsk_slab =
  2077. kmem_cache_create(prot->twsk_prot->twsk_slab_name,
  2078. prot->twsk_prot->twsk_obj_size,
  2079. 0,
  2080. SLAB_HWCACHE_ALIGN |
  2081. prot->slab_flags,
  2082. NULL);
  2083. if (prot->twsk_prot->twsk_slab == NULL)
  2084. goto out_free_timewait_sock_slab_name;
  2085. }
  2086. }
  2087. write_lock(&proto_list_lock);
  2088. list_add(&prot->node, &proto_list);
  2089. assign_proto_idx(prot);
  2090. write_unlock(&proto_list_lock);
  2091. return 0;
  2092. out_free_timewait_sock_slab_name:
  2093. kfree(prot->twsk_prot->twsk_slab_name);
  2094. out_free_request_sock_slab:
  2095. if (prot->rsk_prot && prot->rsk_prot->slab) {
  2096. kmem_cache_destroy(prot->rsk_prot->slab);
  2097. prot->rsk_prot->slab = NULL;
  2098. }
  2099. out_free_request_sock_slab_name:
  2100. if (prot->rsk_prot)
  2101. kfree(prot->rsk_prot->slab_name);
  2102. out_free_sock_slab:
  2103. kmem_cache_destroy(prot->slab);
  2104. prot->slab = NULL;
  2105. out:
  2106. return -ENOBUFS;
  2107. }
  2108. EXPORT_SYMBOL(proto_register);
  2109. void proto_unregister(struct proto *prot)
  2110. {
  2111. write_lock(&proto_list_lock);
  2112. release_proto_idx(prot);
  2113. list_del(&prot->node);
  2114. write_unlock(&proto_list_lock);
  2115. if (prot->slab != NULL) {
  2116. kmem_cache_destroy(prot->slab);
  2117. prot->slab = NULL;
  2118. }
  2119. if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
  2120. kmem_cache_destroy(prot->rsk_prot->slab);
  2121. kfree(prot->rsk_prot->slab_name);
  2122. prot->rsk_prot->slab = NULL;
  2123. }
  2124. if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
  2125. kmem_cache_destroy(prot->twsk_prot->twsk_slab);
  2126. kfree(prot->twsk_prot->twsk_slab_name);
  2127. prot->twsk_prot->twsk_slab = NULL;
  2128. }
  2129. }
  2130. EXPORT_SYMBOL(proto_unregister);
  2131. #ifdef CONFIG_PROC_FS
  2132. static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
  2133. __acquires(proto_list_lock)
  2134. {
  2135. read_lock(&proto_list_lock);
  2136. return seq_list_start_head(&proto_list, *pos);
  2137. }
  2138. static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
  2139. {
  2140. return seq_list_next(v, &proto_list, pos);
  2141. }
  2142. static void proto_seq_stop(struct seq_file *seq, void *v)
  2143. __releases(proto_list_lock)
  2144. {
  2145. read_unlock(&proto_list_lock);
  2146. }
  2147. static char proto_method_implemented(const void *method)
  2148. {
  2149. return method == NULL ? 'n' : 'y';
  2150. }
  2151. static long sock_prot_memory_allocated(struct proto *proto)
  2152. {
  2153. return proto->memory_allocated != NULL ? proto_memory_allocated(proto): -1L;
  2154. }
  2155. static char *sock_prot_memory_pressure(struct proto *proto)
  2156. {
  2157. return proto->memory_pressure != NULL ?
  2158. proto_memory_pressure(proto) ? "yes" : "no" : "NI";
  2159. }
  2160. static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
  2161. {
  2162. seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
  2163. "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
  2164. proto->name,
  2165. proto->obj_size,
  2166. sock_prot_inuse_get(seq_file_net(seq), proto),
  2167. sock_prot_memory_allocated(proto),
  2168. sock_prot_memory_pressure(proto),
  2169. proto->max_header,
  2170. proto->slab == NULL ? "no" : "yes",
  2171. module_name(proto->owner),
  2172. proto_method_implemented(proto->close),
  2173. proto_method_implemented(proto->connect),
  2174. proto_method_implemented(proto->disconnect),
  2175. proto_method_implemented(proto->accept),
  2176. proto_method_implemented(proto->ioctl),
  2177. proto_method_implemented(proto->init),
  2178. proto_method_implemented(proto->destroy),
  2179. proto_method_implemented(proto->shutdown),
  2180. proto_method_implemented(proto->setsockopt),
  2181. proto_method_implemented(proto->getsockopt),
  2182. proto_method_implemented(proto->sendmsg),
  2183. proto_method_implemented(proto->recvmsg),
  2184. proto_method_implemented(proto->sendpage),
  2185. proto_method_implemented(proto->bind),
  2186. proto_method_implemented(proto->backlog_rcv),
  2187. proto_method_implemented(proto->hash),
  2188. proto_method_implemented(proto->unhash),
  2189. proto_method_implemented(proto->get_port),
  2190. proto_method_implemented(proto->enter_memory_pressure));
  2191. }
  2192. static int proto_seq_show(struct seq_file *seq, void *v)
  2193. {
  2194. if (v == &proto_list)
  2195. seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
  2196. "protocol",
  2197. "size",
  2198. "sockets",
  2199. "memory",
  2200. "press",
  2201. "maxhdr",
  2202. "slab",
  2203. "module",
  2204. "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
  2205. else
  2206. proto_seq_printf(seq, list_entry(v, struct proto, node));
  2207. return 0;
  2208. }
  2209. static const struct seq_operations proto_seq_ops = {
  2210. .start = proto_seq_start,
  2211. .next = proto_seq_next,
  2212. .stop = proto_seq_stop,
  2213. .show = proto_seq_show,
  2214. };
  2215. static int proto_seq_open(struct inode *inode, struct file *file)
  2216. {
  2217. return seq_open_net(inode, file, &proto_seq_ops,
  2218. sizeof(struct seq_net_private));
  2219. }
  2220. static const struct file_operations proto_seq_fops = {
  2221. .owner = THIS_MODULE,
  2222. .open = proto_seq_open,
  2223. .read = seq_read,
  2224. .llseek = seq_lseek,
  2225. .release = seq_release_net,
  2226. };
  2227. static __net_init int proto_init_net(struct net *net)
  2228. {
  2229. if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
  2230. return -ENOMEM;
  2231. return 0;
  2232. }
  2233. static __net_exit void proto_exit_net(struct net *net)
  2234. {
  2235. proc_net_remove(net, "protocols");
  2236. }
  2237. static __net_initdata struct pernet_operations proto_net_ops = {
  2238. .init = proto_init_net,
  2239. .exit = proto_exit_net,
  2240. };
  2241. static int __init proto_init(void)
  2242. {
  2243. return register_pernet_subsys(&proto_net_ops);
  2244. }
  2245. subsys_initcall(proto_init);
  2246. #endif /* PROC_FS */