svcsock.c 42 KB

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027102810291030103110321033103410351036103710381039104010411042104310441045104610471048104910501051105210531054105510561057105810591060106110621063106410651066106710681069107010711072107310741075107610771078107910801081108210831084108510861087108810891090109110921093109410951096109710981099110011011102110311041105110611071108110911101111111211131114111511161117111811191120112111221123112411251126112711281129113011311132113311341135113611371138113911401141114211431144114511461147114811491150115111521153115411551156115711581159116011611162116311641165116611671168116911701171117211731174117511761177117811791180118111821183118411851186118711881189119011911192119311941195119611971198119912001201120212031204120512061207120812091210121112121213121412151216121712181219122012211222122312241225122612271228122912301231123212331234123512361237123812391240124112421243124412451246124712481249125012511252125312541255125612571258125912601261126212631264126512661267126812691270127112721273127412751276127712781279128012811282128312841285128612871288128912901291129212931294129512961297129812991300130113021303130413051306130713081309131013111312131313141315131613171318131913201321132213231324132513261327132813291330133113321333133413351336133713381339134013411342134313441345134613471348134913501351135213531354135513561357135813591360136113621363136413651366136713681369137013711372137313741375137613771378137913801381138213831384138513861387138813891390139113921393139413951396139713981399140014011402140314041405140614071408140914101411141214131414141514161417141814191420142114221423142414251426142714281429143014311432143314341435143614371438143914401441144214431444144514461447144814491450145114521453145414551456145714581459146014611462146314641465146614671468146914701471147214731474147514761477147814791480148114821483148414851486148714881489149014911492149314941495149614971498149915001501150215031504150515061507150815091510151115121513151415151516151715181519152015211522152315241525152615271528152915301531153215331534153515361537153815391540154115421543154415451546154715481549155015511552155315541555155615571558155915601561156215631564156515661567156815691570157115721573157415751576157715781579158015811582158315841585158615871588158915901591159215931594159515961597159815991600160116021603160416051606160716081609161016111612161316141615
  1. /*
  2. * linux/net/sunrpc/svcsock.c
  3. *
  4. * These are the RPC server socket internals.
  5. *
  6. * The server scheduling algorithm does not always distribute the load
  7. * evenly when servicing a single client. May need to modify the
  8. * svc_xprt_enqueue procedure...
  9. *
  10. * TCP support is largely untested and may be a little slow. The problem
  11. * is that we currently do two separate recvfrom's, one for the 4-byte
  12. * record length, and the second for the actual record. This could possibly
  13. * be improved by always reading a minimum size of around 100 bytes and
  14. * tucking any superfluous bytes away in a temporary store. Still, that
  15. * leaves write requests out in the rain. An alternative may be to peek at
  16. * the first skb in the queue, and if it matches the next TCP sequence
  17. * number, to extract the record marker. Yuck.
  18. *
  19. * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
  20. */
  21. #include <linux/kernel.h>
  22. #include <linux/sched.h>
  23. #include <linux/errno.h>
  24. #include <linux/fcntl.h>
  25. #include <linux/net.h>
  26. #include <linux/in.h>
  27. #include <linux/inet.h>
  28. #include <linux/udp.h>
  29. #include <linux/tcp.h>
  30. #include <linux/unistd.h>
  31. #include <linux/slab.h>
  32. #include <linux/netdevice.h>
  33. #include <linux/skbuff.h>
  34. #include <linux/file.h>
  35. #include <linux/freezer.h>
  36. #include <net/sock.h>
  37. #include <net/checksum.h>
  38. #include <net/ip.h>
  39. #include <net/ipv6.h>
  40. #include <net/tcp.h>
  41. #include <net/tcp_states.h>
  42. #include <asm/uaccess.h>
  43. #include <asm/ioctls.h>
  44. #include <linux/sunrpc/types.h>
  45. #include <linux/sunrpc/clnt.h>
  46. #include <linux/sunrpc/xdr.h>
  47. #include <linux/sunrpc/msg_prot.h>
  48. #include <linux/sunrpc/svcsock.h>
  49. #include <linux/sunrpc/stats.h>
  50. #include <linux/sunrpc/xprt.h>
  51. #define RPCDBG_FACILITY RPCDBG_SVCXPRT
  52. static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
  53. int *errp, int flags);
  54. static void svc_udp_data_ready(struct sock *, int);
  55. static int svc_udp_recvfrom(struct svc_rqst *);
  56. static int svc_udp_sendto(struct svc_rqst *);
  57. static void svc_sock_detach(struct svc_xprt *);
  58. static void svc_tcp_sock_detach(struct svc_xprt *);
  59. static void svc_sock_free(struct svc_xprt *);
  60. static struct svc_xprt *svc_create_socket(struct svc_serv *, int,
  61. struct net *, struct sockaddr *,
  62. int, int);
  63. #if defined(CONFIG_NFS_V4_1)
  64. static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
  65. struct net *, struct sockaddr *,
  66. int, int);
  67. static void svc_bc_sock_free(struct svc_xprt *xprt);
  68. #endif /* CONFIG_NFS_V4_1 */
  69. #ifdef CONFIG_DEBUG_LOCK_ALLOC
  70. static struct lock_class_key svc_key[2];
  71. static struct lock_class_key svc_slock_key[2];
  72. static void svc_reclassify_socket(struct socket *sock)
  73. {
  74. struct sock *sk = sock->sk;
  75. BUG_ON(sock_owned_by_user(sk));
  76. switch (sk->sk_family) {
  77. case AF_INET:
  78. sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
  79. &svc_slock_key[0],
  80. "sk_xprt.xpt_lock-AF_INET-NFSD",
  81. &svc_key[0]);
  82. break;
  83. case AF_INET6:
  84. sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
  85. &svc_slock_key[1],
  86. "sk_xprt.xpt_lock-AF_INET6-NFSD",
  87. &svc_key[1]);
  88. break;
  89. default:
  90. BUG();
  91. }
  92. }
  93. #else
  94. static void svc_reclassify_socket(struct socket *sock)
  95. {
  96. }
  97. #endif
  98. /*
  99. * Release an skbuff after use
  100. */
  101. static void svc_release_skb(struct svc_rqst *rqstp)
  102. {
  103. struct sk_buff *skb = rqstp->rq_xprt_ctxt;
  104. if (skb) {
  105. struct svc_sock *svsk =
  106. container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
  107. rqstp->rq_xprt_ctxt = NULL;
  108. dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
  109. skb_free_datagram_locked(svsk->sk_sk, skb);
  110. }
  111. }
  112. union svc_pktinfo_u {
  113. struct in_pktinfo pkti;
  114. struct in6_pktinfo pkti6;
  115. };
  116. #define SVC_PKTINFO_SPACE \
  117. CMSG_SPACE(sizeof(union svc_pktinfo_u))
  118. static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
  119. {
  120. struct svc_sock *svsk =
  121. container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
  122. switch (svsk->sk_sk->sk_family) {
  123. case AF_INET: {
  124. struct in_pktinfo *pki = CMSG_DATA(cmh);
  125. cmh->cmsg_level = SOL_IP;
  126. cmh->cmsg_type = IP_PKTINFO;
  127. pki->ipi_ifindex = 0;
  128. pki->ipi_spec_dst.s_addr = rqstp->rq_daddr.addr.s_addr;
  129. cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
  130. }
  131. break;
  132. case AF_INET6: {
  133. struct in6_pktinfo *pki = CMSG_DATA(cmh);
  134. cmh->cmsg_level = SOL_IPV6;
  135. cmh->cmsg_type = IPV6_PKTINFO;
  136. pki->ipi6_ifindex = 0;
  137. ipv6_addr_copy(&pki->ipi6_addr,
  138. &rqstp->rq_daddr.addr6);
  139. cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
  140. }
  141. break;
  142. }
  143. }
  144. /*
  145. * send routine intended to be shared by the fore- and back-channel
  146. */
  147. int svc_send_common(struct socket *sock, struct xdr_buf *xdr,
  148. struct page *headpage, unsigned long headoffset,
  149. struct page *tailpage, unsigned long tailoffset)
  150. {
  151. int result;
  152. int size;
  153. struct page **ppage = xdr->pages;
  154. size_t base = xdr->page_base;
  155. unsigned int pglen = xdr->page_len;
  156. unsigned int flags = MSG_MORE;
  157. int slen;
  158. int len = 0;
  159. slen = xdr->len;
  160. /* send head */
  161. if (slen == xdr->head[0].iov_len)
  162. flags = 0;
  163. len = kernel_sendpage(sock, headpage, headoffset,
  164. xdr->head[0].iov_len, flags);
  165. if (len != xdr->head[0].iov_len)
  166. goto out;
  167. slen -= xdr->head[0].iov_len;
  168. if (slen == 0)
  169. goto out;
  170. /* send page data */
  171. size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen;
  172. while (pglen > 0) {
  173. if (slen == size)
  174. flags = 0;
  175. result = kernel_sendpage(sock, *ppage, base, size, flags);
  176. if (result > 0)
  177. len += result;
  178. if (result != size)
  179. goto out;
  180. slen -= size;
  181. pglen -= size;
  182. size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen;
  183. base = 0;
  184. ppage++;
  185. }
  186. /* send tail */
  187. if (xdr->tail[0].iov_len) {
  188. result = kernel_sendpage(sock, tailpage, tailoffset,
  189. xdr->tail[0].iov_len, 0);
  190. if (result > 0)
  191. len += result;
  192. }
  193. out:
  194. return len;
  195. }
  196. /*
  197. * Generic sendto routine
  198. */
  199. static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr)
  200. {
  201. struct svc_sock *svsk =
  202. container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
  203. struct socket *sock = svsk->sk_sock;
  204. union {
  205. struct cmsghdr hdr;
  206. long all[SVC_PKTINFO_SPACE / sizeof(long)];
  207. } buffer;
  208. struct cmsghdr *cmh = &buffer.hdr;
  209. int len = 0;
  210. unsigned long tailoff;
  211. unsigned long headoff;
  212. RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
  213. if (rqstp->rq_prot == IPPROTO_UDP) {
  214. struct msghdr msg = {
  215. .msg_name = &rqstp->rq_addr,
  216. .msg_namelen = rqstp->rq_addrlen,
  217. .msg_control = cmh,
  218. .msg_controllen = sizeof(buffer),
  219. .msg_flags = MSG_MORE,
  220. };
  221. svc_set_cmsg_data(rqstp, cmh);
  222. if (sock_sendmsg(sock, &msg, 0) < 0)
  223. goto out;
  224. }
  225. tailoff = ((unsigned long)xdr->tail[0].iov_base) & (PAGE_SIZE-1);
  226. headoff = 0;
  227. len = svc_send_common(sock, xdr, rqstp->rq_respages[0], headoff,
  228. rqstp->rq_respages[0], tailoff);
  229. out:
  230. dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n",
  231. svsk, xdr->head[0].iov_base, xdr->head[0].iov_len,
  232. xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf)));
  233. return len;
  234. }
  235. /*
  236. * Report socket names for nfsdfs
  237. */
  238. static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining)
  239. {
  240. const struct sock *sk = svsk->sk_sk;
  241. const char *proto_name = sk->sk_protocol == IPPROTO_UDP ?
  242. "udp" : "tcp";
  243. int len;
  244. switch (sk->sk_family) {
  245. case PF_INET:
  246. len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n",
  247. proto_name,
  248. &inet_sk(sk)->inet_rcv_saddr,
  249. inet_sk(sk)->inet_num);
  250. break;
  251. case PF_INET6:
  252. len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n",
  253. proto_name,
  254. &inet6_sk(sk)->rcv_saddr,
  255. inet_sk(sk)->inet_num);
  256. break;
  257. default:
  258. len = snprintf(buf, remaining, "*unknown-%d*\n",
  259. sk->sk_family);
  260. }
  261. if (len >= remaining) {
  262. *buf = '\0';
  263. return -ENAMETOOLONG;
  264. }
  265. return len;
  266. }
  267. /**
  268. * svc_sock_names - construct a list of listener names in a string
  269. * @serv: pointer to RPC service
  270. * @buf: pointer to a buffer to fill in with socket names
  271. * @buflen: size of the buffer to be filled
  272. * @toclose: pointer to '\0'-terminated C string containing the name
  273. * of a listener to be closed
  274. *
  275. * Fills in @buf with a '\n'-separated list of names of listener
  276. * sockets. If @toclose is not NULL, the socket named by @toclose
  277. * is closed, and is not included in the output list.
  278. *
  279. * Returns positive length of the socket name string, or a negative
  280. * errno value on error.
  281. */
  282. int svc_sock_names(struct svc_serv *serv, char *buf, const size_t buflen,
  283. const char *toclose)
  284. {
  285. struct svc_sock *svsk, *closesk = NULL;
  286. int len = 0;
  287. if (!serv)
  288. return 0;
  289. spin_lock_bh(&serv->sv_lock);
  290. list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list) {
  291. int onelen = svc_one_sock_name(svsk, buf + len, buflen - len);
  292. if (onelen < 0) {
  293. len = onelen;
  294. break;
  295. }
  296. if (toclose && strcmp(toclose, buf + len) == 0) {
  297. closesk = svsk;
  298. svc_xprt_get(&closesk->sk_xprt);
  299. } else
  300. len += onelen;
  301. }
  302. spin_unlock_bh(&serv->sv_lock);
  303. if (closesk) {
  304. /* Should unregister with portmap, but you cannot
  305. * unregister just one protocol...
  306. */
  307. svc_close_xprt(&closesk->sk_xprt);
  308. svc_xprt_put(&closesk->sk_xprt);
  309. } else if (toclose)
  310. return -ENOENT;
  311. return len;
  312. }
  313. EXPORT_SYMBOL_GPL(svc_sock_names);
  314. /*
  315. * Check input queue length
  316. */
  317. static int svc_recv_available(struct svc_sock *svsk)
  318. {
  319. struct socket *sock = svsk->sk_sock;
  320. int avail, err;
  321. err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail);
  322. return (err >= 0)? avail : err;
  323. }
  324. /*
  325. * Generic recvfrom routine.
  326. */
  327. static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr,
  328. int buflen)
  329. {
  330. struct svc_sock *svsk =
  331. container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
  332. struct msghdr msg = {
  333. .msg_flags = MSG_DONTWAIT,
  334. };
  335. int len;
  336. rqstp->rq_xprt_hlen = 0;
  337. len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen,
  338. msg.msg_flags);
  339. dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
  340. svsk, iov[0].iov_base, iov[0].iov_len, len);
  341. return len;
  342. }
  343. /*
  344. * Set socket snd and rcv buffer lengths
  345. */
  346. static void svc_sock_setbufsize(struct socket *sock, unsigned int snd,
  347. unsigned int rcv)
  348. {
  349. #if 0
  350. mm_segment_t oldfs;
  351. oldfs = get_fs(); set_fs(KERNEL_DS);
  352. sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
  353. (char*)&snd, sizeof(snd));
  354. sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
  355. (char*)&rcv, sizeof(rcv));
  356. #else
  357. /* sock_setsockopt limits use to sysctl_?mem_max,
  358. * which isn't acceptable. Until that is made conditional
  359. * on not having CAP_SYS_RESOURCE or similar, we go direct...
  360. * DaveM said I could!
  361. */
  362. lock_sock(sock->sk);
  363. sock->sk->sk_sndbuf = snd * 2;
  364. sock->sk->sk_rcvbuf = rcv * 2;
  365. sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK|SOCK_RCVBUF_LOCK;
  366. sock->sk->sk_write_space(sock->sk);
  367. release_sock(sock->sk);
  368. #endif
  369. }
  370. /*
  371. * INET callback when data has been received on the socket.
  372. */
  373. static void svc_udp_data_ready(struct sock *sk, int count)
  374. {
  375. struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
  376. wait_queue_head_t *wq = sk_sleep(sk);
  377. if (svsk) {
  378. dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n",
  379. svsk, sk, count,
  380. test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
  381. set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
  382. svc_xprt_enqueue(&svsk->sk_xprt);
  383. }
  384. if (wq && waitqueue_active(wq))
  385. wake_up_interruptible(wq);
  386. }
  387. /*
  388. * INET callback when space is newly available on the socket.
  389. */
  390. static void svc_write_space(struct sock *sk)
  391. {
  392. struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
  393. wait_queue_head_t *wq = sk_sleep(sk);
  394. if (svsk) {
  395. dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
  396. svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
  397. svc_xprt_enqueue(&svsk->sk_xprt);
  398. }
  399. if (wq && waitqueue_active(wq)) {
  400. dprintk("RPC svc_write_space: someone sleeping on %p\n",
  401. svsk);
  402. wake_up_interruptible(wq);
  403. }
  404. }
  405. static void svc_tcp_write_space(struct sock *sk)
  406. {
  407. struct socket *sock = sk->sk_socket;
  408. if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && sock)
  409. clear_bit(SOCK_NOSPACE, &sock->flags);
  410. svc_write_space(sk);
  411. }
  412. /*
  413. * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
  414. */
  415. static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
  416. struct cmsghdr *cmh)
  417. {
  418. struct in_pktinfo *pki = CMSG_DATA(cmh);
  419. if (cmh->cmsg_type != IP_PKTINFO)
  420. return 0;
  421. rqstp->rq_daddr.addr.s_addr = pki->ipi_spec_dst.s_addr;
  422. return 1;
  423. }
  424. /*
  425. * See net/ipv6/datagram.c : datagram_recv_ctl
  426. */
  427. static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
  428. struct cmsghdr *cmh)
  429. {
  430. struct in6_pktinfo *pki = CMSG_DATA(cmh);
  431. if (cmh->cmsg_type != IPV6_PKTINFO)
  432. return 0;
  433. ipv6_addr_copy(&rqstp->rq_daddr.addr6, &pki->ipi6_addr);
  434. return 1;
  435. }
  436. /*
  437. * Copy the UDP datagram's destination address to the rqstp structure.
  438. * The 'destination' address in this case is the address to which the
  439. * peer sent the datagram, i.e. our local address. For multihomed
  440. * hosts, this can change from msg to msg. Note that only the IP
  441. * address changes, the port number should remain the same.
  442. */
  443. static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
  444. struct cmsghdr *cmh)
  445. {
  446. switch (cmh->cmsg_level) {
  447. case SOL_IP:
  448. return svc_udp_get_dest_address4(rqstp, cmh);
  449. case SOL_IPV6:
  450. return svc_udp_get_dest_address6(rqstp, cmh);
  451. }
  452. return 0;
  453. }
  454. /*
  455. * Receive a datagram from a UDP socket.
  456. */
  457. static int svc_udp_recvfrom(struct svc_rqst *rqstp)
  458. {
  459. struct svc_sock *svsk =
  460. container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
  461. struct svc_serv *serv = svsk->sk_xprt.xpt_server;
  462. struct sk_buff *skb;
  463. union {
  464. struct cmsghdr hdr;
  465. long all[SVC_PKTINFO_SPACE / sizeof(long)];
  466. } buffer;
  467. struct cmsghdr *cmh = &buffer.hdr;
  468. struct msghdr msg = {
  469. .msg_name = svc_addr(rqstp),
  470. .msg_control = cmh,
  471. .msg_controllen = sizeof(buffer),
  472. .msg_flags = MSG_DONTWAIT,
  473. };
  474. size_t len;
  475. int err;
  476. if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
  477. /* udp sockets need large rcvbuf as all pending
  478. * requests are still in that buffer. sndbuf must
  479. * also be large enough that there is enough space
  480. * for one reply per thread. We count all threads
  481. * rather than threads in a particular pool, which
  482. * provides an upper bound on the number of threads
  483. * which will access the socket.
  484. */
  485. svc_sock_setbufsize(svsk->sk_sock,
  486. (serv->sv_nrthreads+3) * serv->sv_max_mesg,
  487. (serv->sv_nrthreads+3) * serv->sv_max_mesg);
  488. clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
  489. skb = NULL;
  490. err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
  491. 0, 0, MSG_PEEK | MSG_DONTWAIT);
  492. if (err >= 0)
  493. skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err);
  494. if (skb == NULL) {
  495. if (err != -EAGAIN) {
  496. /* possibly an icmp error */
  497. dprintk("svc: recvfrom returned error %d\n", -err);
  498. set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
  499. }
  500. return -EAGAIN;
  501. }
  502. len = svc_addr_len(svc_addr(rqstp));
  503. if (len == 0)
  504. return -EAFNOSUPPORT;
  505. rqstp->rq_addrlen = len;
  506. if (skb->tstamp.tv64 == 0) {
  507. skb->tstamp = ktime_get_real();
  508. /* Don't enable netstamp, sunrpc doesn't
  509. need that much accuracy */
  510. }
  511. svsk->sk_sk->sk_stamp = skb->tstamp;
  512. set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
  513. len = skb->len - sizeof(struct udphdr);
  514. rqstp->rq_arg.len = len;
  515. rqstp->rq_prot = IPPROTO_UDP;
  516. if (!svc_udp_get_dest_address(rqstp, cmh)) {
  517. if (net_ratelimit())
  518. printk(KERN_WARNING
  519. "svc: received unknown control message %d/%d; "
  520. "dropping RPC reply datagram\n",
  521. cmh->cmsg_level, cmh->cmsg_type);
  522. skb_free_datagram_locked(svsk->sk_sk, skb);
  523. return 0;
  524. }
  525. if (skb_is_nonlinear(skb)) {
  526. /* we have to copy */
  527. local_bh_disable();
  528. if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
  529. local_bh_enable();
  530. /* checksum error */
  531. skb_free_datagram_locked(svsk->sk_sk, skb);
  532. return 0;
  533. }
  534. local_bh_enable();
  535. skb_free_datagram_locked(svsk->sk_sk, skb);
  536. } else {
  537. /* we can use it in-place */
  538. rqstp->rq_arg.head[0].iov_base = skb->data +
  539. sizeof(struct udphdr);
  540. rqstp->rq_arg.head[0].iov_len = len;
  541. if (skb_checksum_complete(skb)) {
  542. skb_free_datagram_locked(svsk->sk_sk, skb);
  543. return 0;
  544. }
  545. rqstp->rq_xprt_ctxt = skb;
  546. }
  547. rqstp->rq_arg.page_base = 0;
  548. if (len <= rqstp->rq_arg.head[0].iov_len) {
  549. rqstp->rq_arg.head[0].iov_len = len;
  550. rqstp->rq_arg.page_len = 0;
  551. rqstp->rq_respages = rqstp->rq_pages+1;
  552. } else {
  553. rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
  554. rqstp->rq_respages = rqstp->rq_pages + 1 +
  555. DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
  556. }
  557. if (serv->sv_stats)
  558. serv->sv_stats->netudpcnt++;
  559. return len;
  560. }
  561. static int
  562. svc_udp_sendto(struct svc_rqst *rqstp)
  563. {
  564. int error;
  565. error = svc_sendto(rqstp, &rqstp->rq_res);
  566. if (error == -ECONNREFUSED)
  567. /* ICMP error on earlier request. */
  568. error = svc_sendto(rqstp, &rqstp->rq_res);
  569. return error;
  570. }
  571. static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
  572. {
  573. }
  574. static int svc_udp_has_wspace(struct svc_xprt *xprt)
  575. {
  576. struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
  577. struct svc_serv *serv = xprt->xpt_server;
  578. unsigned long required;
  579. /*
  580. * Set the SOCK_NOSPACE flag before checking the available
  581. * sock space.
  582. */
  583. set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
  584. required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
  585. if (required*2 > sock_wspace(svsk->sk_sk))
  586. return 0;
  587. clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
  588. return 1;
  589. }
  590. static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
  591. {
  592. BUG();
  593. return NULL;
  594. }
  595. static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
  596. struct net *net,
  597. struct sockaddr *sa, int salen,
  598. int flags)
  599. {
  600. return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
  601. }
  602. static struct svc_xprt_ops svc_udp_ops = {
  603. .xpo_create = svc_udp_create,
  604. .xpo_recvfrom = svc_udp_recvfrom,
  605. .xpo_sendto = svc_udp_sendto,
  606. .xpo_release_rqst = svc_release_skb,
  607. .xpo_detach = svc_sock_detach,
  608. .xpo_free = svc_sock_free,
  609. .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
  610. .xpo_has_wspace = svc_udp_has_wspace,
  611. .xpo_accept = svc_udp_accept,
  612. };
  613. static struct svc_xprt_class svc_udp_class = {
  614. .xcl_name = "udp",
  615. .xcl_owner = THIS_MODULE,
  616. .xcl_ops = &svc_udp_ops,
  617. .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
  618. };
  619. static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
  620. {
  621. int err, level, optname, one = 1;
  622. svc_xprt_init(&svc_udp_class, &svsk->sk_xprt, serv);
  623. clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
  624. svsk->sk_sk->sk_data_ready = svc_udp_data_ready;
  625. svsk->sk_sk->sk_write_space = svc_write_space;
  626. /* initialise setting must have enough space to
  627. * receive and respond to one request.
  628. * svc_udp_recvfrom will re-adjust if necessary
  629. */
  630. svc_sock_setbufsize(svsk->sk_sock,
  631. 3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
  632. 3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
  633. /* data might have come in before data_ready set up */
  634. set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
  635. set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
  636. /* make sure we get destination address info */
  637. switch (svsk->sk_sk->sk_family) {
  638. case AF_INET:
  639. level = SOL_IP;
  640. optname = IP_PKTINFO;
  641. break;
  642. case AF_INET6:
  643. level = SOL_IPV6;
  644. optname = IPV6_RECVPKTINFO;
  645. break;
  646. default:
  647. BUG();
  648. }
  649. err = kernel_setsockopt(svsk->sk_sock, level, optname,
  650. (char *)&one, sizeof(one));
  651. dprintk("svc: kernel_setsockopt returned %d\n", err);
  652. }
  653. /*
  654. * A data_ready event on a listening socket means there's a connection
  655. * pending. Do not use state_change as a substitute for it.
  656. */
  657. static void svc_tcp_listen_data_ready(struct sock *sk, int count_unused)
  658. {
  659. struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
  660. wait_queue_head_t *wq;
  661. dprintk("svc: socket %p TCP (listen) state change %d\n",
  662. sk, sk->sk_state);
  663. /*
  664. * This callback may called twice when a new connection
  665. * is established as a child socket inherits everything
  666. * from a parent LISTEN socket.
  667. * 1) data_ready method of the parent socket will be called
  668. * when one of child sockets become ESTABLISHED.
  669. * 2) data_ready method of the child socket may be called
  670. * when it receives data before the socket is accepted.
  671. * In case of 2, we should ignore it silently.
  672. */
  673. if (sk->sk_state == TCP_LISTEN) {
  674. if (svsk) {
  675. set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
  676. svc_xprt_enqueue(&svsk->sk_xprt);
  677. } else
  678. printk("svc: socket %p: no user data\n", sk);
  679. }
  680. wq = sk_sleep(sk);
  681. if (wq && waitqueue_active(wq))
  682. wake_up_interruptible_all(wq);
  683. }
  684. /*
  685. * A state change on a connected socket means it's dying or dead.
  686. */
  687. static void svc_tcp_state_change(struct sock *sk)
  688. {
  689. struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
  690. wait_queue_head_t *wq = sk_sleep(sk);
  691. dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
  692. sk, sk->sk_state, sk->sk_user_data);
  693. if (!svsk)
  694. printk("svc: socket %p: no user data\n", sk);
  695. else {
  696. set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
  697. svc_xprt_enqueue(&svsk->sk_xprt);
  698. }
  699. if (wq && waitqueue_active(wq))
  700. wake_up_interruptible_all(wq);
  701. }
  702. static void svc_tcp_data_ready(struct sock *sk, int count)
  703. {
  704. struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
  705. wait_queue_head_t *wq = sk_sleep(sk);
  706. dprintk("svc: socket %p TCP data ready (svsk %p)\n",
  707. sk, sk->sk_user_data);
  708. if (svsk) {
  709. set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
  710. svc_xprt_enqueue(&svsk->sk_xprt);
  711. }
  712. if (wq && waitqueue_active(wq))
  713. wake_up_interruptible(wq);
  714. }
  715. /*
  716. * Accept a TCP connection
  717. */
  718. static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
  719. {
  720. struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
  721. struct sockaddr_storage addr;
  722. struct sockaddr *sin = (struct sockaddr *) &addr;
  723. struct svc_serv *serv = svsk->sk_xprt.xpt_server;
  724. struct socket *sock = svsk->sk_sock;
  725. struct socket *newsock;
  726. struct svc_sock *newsvsk;
  727. int err, slen;
  728. RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
  729. dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
  730. if (!sock)
  731. return NULL;
  732. clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
  733. err = kernel_accept(sock, &newsock, O_NONBLOCK);
  734. if (err < 0) {
  735. if (err == -ENOMEM)
  736. printk(KERN_WARNING "%s: no more sockets!\n",
  737. serv->sv_name);
  738. else if (err != -EAGAIN && net_ratelimit())
  739. printk(KERN_WARNING "%s: accept failed (err %d)!\n",
  740. serv->sv_name, -err);
  741. return NULL;
  742. }
  743. set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
  744. err = kernel_getpeername(newsock, sin, &slen);
  745. if (err < 0) {
  746. if (net_ratelimit())
  747. printk(KERN_WARNING "%s: peername failed (err %d)!\n",
  748. serv->sv_name, -err);
  749. goto failed; /* aborted connection or whatever */
  750. }
  751. /* Ideally, we would want to reject connections from unauthorized
  752. * hosts here, but when we get encryption, the IP of the host won't
  753. * tell us anything. For now just warn about unpriv connections.
  754. */
  755. if (!svc_port_is_privileged(sin)) {
  756. dprintk(KERN_WARNING
  757. "%s: connect from unprivileged port: %s\n",
  758. serv->sv_name,
  759. __svc_print_addr(sin, buf, sizeof(buf)));
  760. }
  761. dprintk("%s: connect from %s\n", serv->sv_name,
  762. __svc_print_addr(sin, buf, sizeof(buf)));
  763. /* make sure that a write doesn't block forever when
  764. * low on memory
  765. */
  766. newsock->sk->sk_sndtimeo = HZ*30;
  767. if (!(newsvsk = svc_setup_socket(serv, newsock, &err,
  768. (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY))))
  769. goto failed;
  770. svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
  771. err = kernel_getsockname(newsock, sin, &slen);
  772. if (unlikely(err < 0)) {
  773. dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
  774. slen = offsetof(struct sockaddr, sa_data);
  775. }
  776. svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
  777. if (serv->sv_stats)
  778. serv->sv_stats->nettcpconn++;
  779. return &newsvsk->sk_xprt;
  780. failed:
  781. sock_release(newsock);
  782. return NULL;
  783. }
  784. /*
  785. * Receive data.
  786. * If we haven't gotten the record length yet, get the next four bytes.
  787. * Otherwise try to gobble up as much as possible up to the complete
  788. * record length.
  789. */
  790. static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp)
  791. {
  792. struct svc_serv *serv = svsk->sk_xprt.xpt_server;
  793. unsigned int want;
  794. int len;
  795. if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
  796. /* sndbuf needs to have room for one request
  797. * per thread, otherwise we can stall even when the
  798. * network isn't a bottleneck.
  799. *
  800. * We count all threads rather than threads in a
  801. * particular pool, which provides an upper bound
  802. * on the number of threads which will access the socket.
  803. *
  804. * rcvbuf just needs to be able to hold a few requests.
  805. * Normally they will be removed from the queue
  806. * as soon a a complete request arrives.
  807. */
  808. svc_sock_setbufsize(svsk->sk_sock,
  809. (serv->sv_nrthreads+3) * serv->sv_max_mesg,
  810. 3 * serv->sv_max_mesg);
  811. clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
  812. if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
  813. struct kvec iov;
  814. want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
  815. iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
  816. iov.iov_len = want;
  817. if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
  818. goto error;
  819. svsk->sk_tcplen += len;
  820. if (len < want) {
  821. dprintk("svc: short recvfrom while reading record "
  822. "length (%d of %d)\n", len, want);
  823. goto err_again; /* record header not complete */
  824. }
  825. svsk->sk_reclen = ntohl(svsk->sk_reclen);
  826. if (!(svsk->sk_reclen & RPC_LAST_STREAM_FRAGMENT)) {
  827. /* FIXME: technically, a record can be fragmented,
  828. * and non-terminal fragments will not have the top
  829. * bit set in the fragment length header.
  830. * But apparently no known nfs clients send fragmented
  831. * records. */
  832. if (net_ratelimit())
  833. printk(KERN_NOTICE "RPC: multiple fragments "
  834. "per record not supported\n");
  835. goto err_delete;
  836. }
  837. svsk->sk_reclen &= RPC_FRAGMENT_SIZE_MASK;
  838. dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen);
  839. if (svsk->sk_reclen > serv->sv_max_mesg) {
  840. if (net_ratelimit())
  841. printk(KERN_NOTICE "RPC: "
  842. "fragment too large: 0x%08lx\n",
  843. (unsigned long)svsk->sk_reclen);
  844. goto err_delete;
  845. }
  846. }
  847. if (svsk->sk_reclen < 8)
  848. goto err_delete; /* client is nuts. */
  849. /* Check whether enough data is available */
  850. len = svc_recv_available(svsk);
  851. if (len < 0)
  852. goto error;
  853. if (len < svsk->sk_reclen) {
  854. dprintk("svc: incomplete TCP record (%d of %d)\n",
  855. len, svsk->sk_reclen);
  856. goto err_again; /* record not complete */
  857. }
  858. len = svsk->sk_reclen;
  859. return len;
  860. error:
  861. if (len == -EAGAIN)
  862. dprintk("RPC: TCP recv_record got EAGAIN\n");
  863. return len;
  864. err_delete:
  865. set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
  866. err_again:
  867. return -EAGAIN;
  868. }
  869. static int svc_process_calldir(struct svc_sock *svsk, struct svc_rqst *rqstp,
  870. struct rpc_rqst **reqpp, struct kvec *vec)
  871. {
  872. struct rpc_rqst *req = NULL;
  873. __be32 *p;
  874. __be32 xid;
  875. __be32 calldir;
  876. int len;
  877. len = svc_recvfrom(rqstp, vec, 1, 8);
  878. if (len < 0)
  879. goto error;
  880. p = (u32 *)rqstp->rq_arg.head[0].iov_base;
  881. xid = *p++;
  882. calldir = *p;
  883. if (calldir == 0) {
  884. /* REQUEST is the most common case */
  885. vec[0] = rqstp->rq_arg.head[0];
  886. } else {
  887. /* REPLY */
  888. struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
  889. if (bc_xprt)
  890. req = xprt_lookup_rqst(bc_xprt, xid);
  891. if (!req) {
  892. printk(KERN_NOTICE
  893. "%s: Got unrecognized reply: "
  894. "calldir 0x%x xpt_bc_xprt %p xid %08x\n",
  895. __func__, ntohl(calldir),
  896. bc_xprt, xid);
  897. vec[0] = rqstp->rq_arg.head[0];
  898. goto out;
  899. }
  900. memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
  901. sizeof(struct xdr_buf));
  902. /* copy the xid and call direction */
  903. memcpy(req->rq_private_buf.head[0].iov_base,
  904. rqstp->rq_arg.head[0].iov_base, 8);
  905. vec[0] = req->rq_private_buf.head[0];
  906. }
  907. out:
  908. vec[0].iov_base += 8;
  909. vec[0].iov_len -= 8;
  910. len = svsk->sk_reclen - 8;
  911. error:
  912. *reqpp = req;
  913. return len;
  914. }
  915. /*
  916. * Receive data from a TCP socket.
  917. */
  918. static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
  919. {
  920. struct svc_sock *svsk =
  921. container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
  922. struct svc_serv *serv = svsk->sk_xprt.xpt_server;
  923. int len;
  924. struct kvec *vec;
  925. struct rpc_rqst *req = NULL;
  926. unsigned int vlen;
  927. int pnum;
  928. dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
  929. svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
  930. test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
  931. test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
  932. len = svc_tcp_recv_record(svsk, rqstp);
  933. if (len < 0)
  934. goto error;
  935. vec = rqstp->rq_vec;
  936. vec[0] = rqstp->rq_arg.head[0];
  937. vlen = PAGE_SIZE;
  938. len = svc_process_calldir(svsk, rqstp, &req, vec);
  939. if (len < 0)
  940. goto err_again;
  941. vlen -= 8;
  942. pnum = 1;
  943. while (vlen < svsk->sk_reclen - 8) {
  944. vec[pnum].iov_base = (req) ?
  945. page_address(req->rq_private_buf.pages[pnum - 1]) :
  946. page_address(rqstp->rq_pages[pnum]);
  947. vec[pnum].iov_len = PAGE_SIZE;
  948. pnum++;
  949. vlen += PAGE_SIZE;
  950. }
  951. rqstp->rq_respages = &rqstp->rq_pages[pnum];
  952. /* Now receive data */
  953. len = svc_recvfrom(rqstp, vec, pnum, svsk->sk_reclen - 8);
  954. if (len < 0)
  955. goto err_again;
  956. if (req) {
  957. xprt_complete_rqst(req->rq_task, svsk->sk_reclen);
  958. rqstp->rq_arg.len = 0;
  959. goto out;
  960. }
  961. dprintk("svc: TCP complete record (%d bytes)\n", svsk->sk_reclen);
  962. rqstp->rq_arg.len = svsk->sk_reclen;
  963. rqstp->rq_arg.page_base = 0;
  964. if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
  965. rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
  966. rqstp->rq_arg.page_len = 0;
  967. } else
  968. rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
  969. rqstp->rq_xprt_ctxt = NULL;
  970. rqstp->rq_prot = IPPROTO_TCP;
  971. out:
  972. /* Reset TCP read info */
  973. svsk->sk_reclen = 0;
  974. svsk->sk_tcplen = 0;
  975. /* If we have more data, signal svc_xprt_enqueue() to try again */
  976. if (svc_recv_available(svsk) > sizeof(rpc_fraghdr))
  977. set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
  978. svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
  979. if (serv->sv_stats)
  980. serv->sv_stats->nettcpcnt++;
  981. return rqstp->rq_arg.len;
  982. err_again:
  983. if (len == -EAGAIN) {
  984. dprintk("RPC: TCP recvfrom got EAGAIN\n");
  985. return len;
  986. }
  987. error:
  988. if (len != -EAGAIN) {
  989. printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
  990. svsk->sk_xprt.xpt_server->sv_name, -len);
  991. set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
  992. }
  993. return -EAGAIN;
  994. }
  995. /*
  996. * Send out data on TCP socket.
  997. */
  998. static int svc_tcp_sendto(struct svc_rqst *rqstp)
  999. {
  1000. struct xdr_buf *xbufp = &rqstp->rq_res;
  1001. int sent;
  1002. __be32 reclen;
  1003. /* Set up the first element of the reply kvec.
  1004. * Any other kvecs that may be in use have been taken
  1005. * care of by the server implementation itself.
  1006. */
  1007. reclen = htonl(0x80000000|((xbufp->len ) - 4));
  1008. memcpy(xbufp->head[0].iov_base, &reclen, 4);
  1009. sent = svc_sendto(rqstp, &rqstp->rq_res);
  1010. if (sent != xbufp->len) {
  1011. printk(KERN_NOTICE
  1012. "rpc-srv/tcp: %s: %s %d when sending %d bytes "
  1013. "- shutting down socket\n",
  1014. rqstp->rq_xprt->xpt_server->sv_name,
  1015. (sent<0)?"got error":"sent only",
  1016. sent, xbufp->len);
  1017. set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
  1018. svc_xprt_enqueue(rqstp->rq_xprt);
  1019. sent = -EAGAIN;
  1020. }
  1021. return sent;
  1022. }
  1023. /*
  1024. * Setup response header. TCP has a 4B record length field.
  1025. */
  1026. static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
  1027. {
  1028. struct kvec *resv = &rqstp->rq_res.head[0];
  1029. /* tcp needs a space for the record length... */
  1030. svc_putnl(resv, 0);
  1031. }
  1032. static int svc_tcp_has_wspace(struct svc_xprt *xprt)
  1033. {
  1034. struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
  1035. struct svc_serv *serv = svsk->sk_xprt.xpt_server;
  1036. int required;
  1037. if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
  1038. return 1;
  1039. required = atomic_read(&xprt->xpt_reserved) + serv->sv_max_mesg;
  1040. if (sk_stream_wspace(svsk->sk_sk) >= required)
  1041. return 1;
  1042. set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
  1043. return 0;
  1044. }
  1045. static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
  1046. struct net *net,
  1047. struct sockaddr *sa, int salen,
  1048. int flags)
  1049. {
  1050. return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
  1051. }
  1052. #if defined(CONFIG_NFS_V4_1)
  1053. static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
  1054. struct net *, struct sockaddr *,
  1055. int, int);
  1056. static void svc_bc_sock_free(struct svc_xprt *xprt);
  1057. static struct svc_xprt *svc_bc_tcp_create(struct svc_serv *serv,
  1058. struct net *net,
  1059. struct sockaddr *sa, int salen,
  1060. int flags)
  1061. {
  1062. return svc_bc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
  1063. }
  1064. static void svc_bc_tcp_sock_detach(struct svc_xprt *xprt)
  1065. {
  1066. }
  1067. static struct svc_xprt_ops svc_tcp_bc_ops = {
  1068. .xpo_create = svc_bc_tcp_create,
  1069. .xpo_detach = svc_bc_tcp_sock_detach,
  1070. .xpo_free = svc_bc_sock_free,
  1071. .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
  1072. };
  1073. static struct svc_xprt_class svc_tcp_bc_class = {
  1074. .xcl_name = "tcp-bc",
  1075. .xcl_owner = THIS_MODULE,
  1076. .xcl_ops = &svc_tcp_bc_ops,
  1077. .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
  1078. };
  1079. static void svc_init_bc_xprt_sock(void)
  1080. {
  1081. svc_reg_xprt_class(&svc_tcp_bc_class);
  1082. }
  1083. static void svc_cleanup_bc_xprt_sock(void)
  1084. {
  1085. svc_unreg_xprt_class(&svc_tcp_bc_class);
  1086. }
  1087. #else /* CONFIG_NFS_V4_1 */
  1088. static void svc_init_bc_xprt_sock(void)
  1089. {
  1090. }
  1091. static void svc_cleanup_bc_xprt_sock(void)
  1092. {
  1093. }
  1094. #endif /* CONFIG_NFS_V4_1 */
  1095. static struct svc_xprt_ops svc_tcp_ops = {
  1096. .xpo_create = svc_tcp_create,
  1097. .xpo_recvfrom = svc_tcp_recvfrom,
  1098. .xpo_sendto = svc_tcp_sendto,
  1099. .xpo_release_rqst = svc_release_skb,
  1100. .xpo_detach = svc_tcp_sock_detach,
  1101. .xpo_free = svc_sock_free,
  1102. .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
  1103. .xpo_has_wspace = svc_tcp_has_wspace,
  1104. .xpo_accept = svc_tcp_accept,
  1105. };
  1106. static struct svc_xprt_class svc_tcp_class = {
  1107. .xcl_name = "tcp",
  1108. .xcl_owner = THIS_MODULE,
  1109. .xcl_ops = &svc_tcp_ops,
  1110. .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
  1111. };
  1112. void svc_init_xprt_sock(void)
  1113. {
  1114. svc_reg_xprt_class(&svc_tcp_class);
  1115. svc_reg_xprt_class(&svc_udp_class);
  1116. svc_init_bc_xprt_sock();
  1117. }
  1118. void svc_cleanup_xprt_sock(void)
  1119. {
  1120. svc_unreg_xprt_class(&svc_tcp_class);
  1121. svc_unreg_xprt_class(&svc_udp_class);
  1122. svc_cleanup_bc_xprt_sock();
  1123. }
  1124. static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
  1125. {
  1126. struct sock *sk = svsk->sk_sk;
  1127. svc_xprt_init(&svc_tcp_class, &svsk->sk_xprt, serv);
  1128. set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
  1129. if (sk->sk_state == TCP_LISTEN) {
  1130. dprintk("setting up TCP socket for listening\n");
  1131. set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
  1132. sk->sk_data_ready = svc_tcp_listen_data_ready;
  1133. set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
  1134. } else {
  1135. dprintk("setting up TCP socket for reading\n");
  1136. sk->sk_state_change = svc_tcp_state_change;
  1137. sk->sk_data_ready = svc_tcp_data_ready;
  1138. sk->sk_write_space = svc_tcp_write_space;
  1139. svsk->sk_reclen = 0;
  1140. svsk->sk_tcplen = 0;
  1141. tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
  1142. /* initialise setting must have enough space to
  1143. * receive and respond to one request.
  1144. * svc_tcp_recvfrom will re-adjust if necessary
  1145. */
  1146. svc_sock_setbufsize(svsk->sk_sock,
  1147. 3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
  1148. 3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
  1149. set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
  1150. set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
  1151. if (sk->sk_state != TCP_ESTABLISHED)
  1152. set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
  1153. }
  1154. }
  1155. void svc_sock_update_bufs(struct svc_serv *serv)
  1156. {
  1157. /*
  1158. * The number of server threads has changed. Update
  1159. * rcvbuf and sndbuf accordingly on all sockets
  1160. */
  1161. struct svc_sock *svsk;
  1162. spin_lock_bh(&serv->sv_lock);
  1163. list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
  1164. set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
  1165. list_for_each_entry(svsk, &serv->sv_tempsocks, sk_xprt.xpt_list)
  1166. set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
  1167. spin_unlock_bh(&serv->sv_lock);
  1168. }
  1169. EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
  1170. /*
  1171. * Initialize socket for RPC use and create svc_sock struct
  1172. * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
  1173. */
  1174. static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
  1175. struct socket *sock,
  1176. int *errp, int flags)
  1177. {
  1178. struct svc_sock *svsk;
  1179. struct sock *inet;
  1180. int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
  1181. dprintk("svc: svc_setup_socket %p\n", sock);
  1182. if (!(svsk = kzalloc(sizeof(*svsk), GFP_KERNEL))) {
  1183. *errp = -ENOMEM;
  1184. return NULL;
  1185. }
  1186. inet = sock->sk;
  1187. /* Register socket with portmapper */
  1188. if (*errp >= 0 && pmap_register)
  1189. *errp = svc_register(serv, inet->sk_family, inet->sk_protocol,
  1190. ntohs(inet_sk(inet)->inet_sport));
  1191. if (*errp < 0) {
  1192. kfree(svsk);
  1193. return NULL;
  1194. }
  1195. inet->sk_user_data = svsk;
  1196. svsk->sk_sock = sock;
  1197. svsk->sk_sk = inet;
  1198. svsk->sk_ostate = inet->sk_state_change;
  1199. svsk->sk_odata = inet->sk_data_ready;
  1200. svsk->sk_owspace = inet->sk_write_space;
  1201. /* Initialize the socket */
  1202. if (sock->type == SOCK_DGRAM)
  1203. svc_udp_init(svsk, serv);
  1204. else
  1205. svc_tcp_init(svsk, serv);
  1206. dprintk("svc: svc_setup_socket created %p (inet %p)\n",
  1207. svsk, svsk->sk_sk);
  1208. return svsk;
  1209. }
  1210. /**
  1211. * svc_addsock - add a listener socket to an RPC service
  1212. * @serv: pointer to RPC service to which to add a new listener
  1213. * @fd: file descriptor of the new listener
  1214. * @name_return: pointer to buffer to fill in with name of listener
  1215. * @len: size of the buffer
  1216. *
  1217. * Fills in socket name and returns positive length of name if successful.
  1218. * Name is terminated with '\n'. On error, returns a negative errno
  1219. * value.
  1220. */
  1221. int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
  1222. const size_t len)
  1223. {
  1224. int err = 0;
  1225. struct socket *so = sockfd_lookup(fd, &err);
  1226. struct svc_sock *svsk = NULL;
  1227. if (!so)
  1228. return err;
  1229. if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
  1230. err = -EAFNOSUPPORT;
  1231. else if (so->sk->sk_protocol != IPPROTO_TCP &&
  1232. so->sk->sk_protocol != IPPROTO_UDP)
  1233. err = -EPROTONOSUPPORT;
  1234. else if (so->state > SS_UNCONNECTED)
  1235. err = -EISCONN;
  1236. else {
  1237. if (!try_module_get(THIS_MODULE))
  1238. err = -ENOENT;
  1239. else
  1240. svsk = svc_setup_socket(serv, so, &err,
  1241. SVC_SOCK_DEFAULTS);
  1242. if (svsk) {
  1243. struct sockaddr_storage addr;
  1244. struct sockaddr *sin = (struct sockaddr *)&addr;
  1245. int salen;
  1246. if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0)
  1247. svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
  1248. clear_bit(XPT_TEMP, &svsk->sk_xprt.xpt_flags);
  1249. spin_lock_bh(&serv->sv_lock);
  1250. list_add(&svsk->sk_xprt.xpt_list, &serv->sv_permsocks);
  1251. spin_unlock_bh(&serv->sv_lock);
  1252. svc_xprt_received(&svsk->sk_xprt);
  1253. err = 0;
  1254. } else
  1255. module_put(THIS_MODULE);
  1256. }
  1257. if (err) {
  1258. sockfd_put(so);
  1259. return err;
  1260. }
  1261. return svc_one_sock_name(svsk, name_return, len);
  1262. }
  1263. EXPORT_SYMBOL_GPL(svc_addsock);
  1264. /*
  1265. * Create socket for RPC service.
  1266. */
  1267. static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
  1268. int protocol,
  1269. struct net *net,
  1270. struct sockaddr *sin, int len,
  1271. int flags)
  1272. {
  1273. struct svc_sock *svsk;
  1274. struct socket *sock;
  1275. int error;
  1276. int type;
  1277. struct sockaddr_storage addr;
  1278. struct sockaddr *newsin = (struct sockaddr *)&addr;
  1279. int newlen;
  1280. int family;
  1281. int val;
  1282. RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
  1283. dprintk("svc: svc_create_socket(%s, %d, %s)\n",
  1284. serv->sv_program->pg_name, protocol,
  1285. __svc_print_addr(sin, buf, sizeof(buf)));
  1286. if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
  1287. printk(KERN_WARNING "svc: only UDP and TCP "
  1288. "sockets supported\n");
  1289. return ERR_PTR(-EINVAL);
  1290. }
  1291. type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
  1292. switch (sin->sa_family) {
  1293. case AF_INET6:
  1294. family = PF_INET6;
  1295. break;
  1296. case AF_INET:
  1297. family = PF_INET;
  1298. break;
  1299. default:
  1300. return ERR_PTR(-EINVAL);
  1301. }
  1302. error = __sock_create(net, family, type, protocol, &sock, 1);
  1303. if (error < 0)
  1304. return ERR_PTR(error);
  1305. svc_reclassify_socket(sock);
  1306. /*
  1307. * If this is an PF_INET6 listener, we want to avoid
  1308. * getting requests from IPv4 remotes. Those should
  1309. * be shunted to a PF_INET listener via rpcbind.
  1310. */
  1311. val = 1;
  1312. if (family == PF_INET6)
  1313. kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
  1314. (char *)&val, sizeof(val));
  1315. if (type == SOCK_STREAM)
  1316. sock->sk->sk_reuse = 1; /* allow address reuse */
  1317. error = kernel_bind(sock, sin, len);
  1318. if (error < 0)
  1319. goto bummer;
  1320. newlen = len;
  1321. error = kernel_getsockname(sock, newsin, &newlen);
  1322. if (error < 0)
  1323. goto bummer;
  1324. if (protocol == IPPROTO_TCP) {
  1325. if ((error = kernel_listen(sock, 64)) < 0)
  1326. goto bummer;
  1327. }
  1328. if ((svsk = svc_setup_socket(serv, sock, &error, flags)) != NULL) {
  1329. svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
  1330. return (struct svc_xprt *)svsk;
  1331. }
  1332. bummer:
  1333. dprintk("svc: svc_create_socket error = %d\n", -error);
  1334. sock_release(sock);
  1335. return ERR_PTR(error);
  1336. }
  1337. /*
  1338. * Detach the svc_sock from the socket so that no
  1339. * more callbacks occur.
  1340. */
  1341. static void svc_sock_detach(struct svc_xprt *xprt)
  1342. {
  1343. struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
  1344. struct sock *sk = svsk->sk_sk;
  1345. wait_queue_head_t *wq;
  1346. dprintk("svc: svc_sock_detach(%p)\n", svsk);
  1347. /* put back the old socket callbacks */
  1348. sk->sk_state_change = svsk->sk_ostate;
  1349. sk->sk_data_ready = svsk->sk_odata;
  1350. sk->sk_write_space = svsk->sk_owspace;
  1351. wq = sk_sleep(sk);
  1352. if (wq && waitqueue_active(wq))
  1353. wake_up_interruptible(wq);
  1354. }
  1355. /*
  1356. * Disconnect the socket, and reset the callbacks
  1357. */
  1358. static void svc_tcp_sock_detach(struct svc_xprt *xprt)
  1359. {
  1360. struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
  1361. dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
  1362. svc_sock_detach(xprt);
  1363. if (!test_bit(XPT_LISTENER, &xprt->xpt_flags))
  1364. kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
  1365. }
  1366. /*
  1367. * Free the svc_sock's socket resources and the svc_sock itself.
  1368. */
  1369. static void svc_sock_free(struct svc_xprt *xprt)
  1370. {
  1371. struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
  1372. dprintk("svc: svc_sock_free(%p)\n", svsk);
  1373. if (svsk->sk_sock->file)
  1374. sockfd_put(svsk->sk_sock);
  1375. else
  1376. sock_release(svsk->sk_sock);
  1377. kfree(svsk);
  1378. }
  1379. #if defined(CONFIG_NFS_V4_1)
  1380. /*
  1381. * Create a back channel svc_xprt which shares the fore channel socket.
  1382. */
  1383. static struct svc_xprt *svc_bc_create_socket(struct svc_serv *serv,
  1384. int protocol,
  1385. struct net *net,
  1386. struct sockaddr *sin, int len,
  1387. int flags)
  1388. {
  1389. struct svc_sock *svsk;
  1390. struct svc_xprt *xprt;
  1391. if (protocol != IPPROTO_TCP) {
  1392. printk(KERN_WARNING "svc: only TCP sockets"
  1393. " supported on shared back channel\n");
  1394. return ERR_PTR(-EINVAL);
  1395. }
  1396. svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
  1397. if (!svsk)
  1398. return ERR_PTR(-ENOMEM);
  1399. xprt = &svsk->sk_xprt;
  1400. svc_xprt_init(&svc_tcp_bc_class, xprt, serv);
  1401. serv->sv_bc_xprt = xprt;
  1402. return xprt;
  1403. }
  1404. /*
  1405. * Free a back channel svc_sock.
  1406. */
  1407. static void svc_bc_sock_free(struct svc_xprt *xprt)
  1408. {
  1409. if (xprt)
  1410. kfree(container_of(xprt, struct svc_sock, sk_xprt));
  1411. }
  1412. #endif /* CONFIG_NFS_V4_1 */