blob: b6e59f0a9475d665ea1952a8428e56ed6b06b89e [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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
Tom Tuckerf6150c32007-12-30 21:07:57 -06008 * svc_xprt_enqueue procedure...
Linus Torvalds1da177e2005-04-16 15:20:36 -07009 *
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
Ilpo Järvinen172589c2007-08-28 15:50:33 -070022#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include <linux/sched.h>
Paul Gortmaker3a9a2312011-05-27 09:12:25 -040024#include <linux/module.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/errno.h>
26#include <linux/fcntl.h>
27#include <linux/net.h>
28#include <linux/in.h>
29#include <linux/inet.h>
30#include <linux/udp.h>
Andrew Morton91483c42005-08-09 20:20:07 -070031#include <linux/tcp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include <linux/unistd.h>
33#include <linux/slab.h>
34#include <linux/netdevice.h>
35#include <linux/skbuff.h>
NeilBrownb41b66d2006-10-02 02:17:48 -070036#include <linux/file.h>
Nigel Cunningham7dfb7102006-12-06 20:34:23 -080037#include <linux/freezer.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038#include <net/sock.h>
39#include <net/checksum.h>
40#include <net/ip.h>
Chuck Leverb92503b2007-02-12 00:53:36 -080041#include <net/ipv6.h>
Chuck Leverb7872fe2008-04-14 12:27:01 -040042#include <net/tcp.h>
Arnaldo Carvalho de Meloc752f072005-08-09 20:08:28 -070043#include <net/tcp_states.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070044#include <asm/uaccess.h>
45#include <asm/ioctls.h>
Eric Dumazet22911fc2012-06-27 00:23:44 +000046#include <trace/events/skb.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070047
48#include <linux/sunrpc/types.h>
Chuck Leverad06e4b2007-02-12 00:53:32 -080049#include <linux/sunrpc/clnt.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070050#include <linux/sunrpc/xdr.h>
Chuck Leverc0401ea2008-04-14 12:27:30 -040051#include <linux/sunrpc/msg_prot.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070052#include <linux/sunrpc/svcsock.h>
53#include <linux/sunrpc/stats.h>
Rahul Iyer4cfc7e62009-09-10 17:32:28 +030054#include <linux/sunrpc/xprt.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070055
H Hartley Sweeten177e4f92011-06-20 17:54:51 -070056#include "sunrpc.h"
57
Tom Tucker360d87382007-12-30 21:07:17 -060058#define RPCDBG_FACILITY RPCDBG_SVCXPRT
Linus Torvalds1da177e2005-04-16 15:20:36 -070059
60
61static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
J. Bruce Fields72c35372012-08-13 17:46:17 -040062 int flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -070063static void svc_udp_data_ready(struct sock *, int);
64static int svc_udp_recvfrom(struct svc_rqst *);
65static int svc_udp_sendto(struct svc_rqst *);
Tom Tucker755ccea2007-12-30 21:07:27 -060066static void svc_sock_detach(struct svc_xprt *);
Trond Myklebust69b6ba32008-12-23 16:30:11 -050067static void svc_tcp_sock_detach(struct svc_xprt *);
Tom Tucker755ccea2007-12-30 21:07:27 -060068static void svc_sock_free(struct svc_xprt *);
Linus Torvalds1da177e2005-04-16 15:20:36 -070069
Tom Tuckerb700cbb2007-12-30 21:07:42 -060070static struct svc_xprt *svc_create_socket(struct svc_serv *, int,
Pavel Emelyanov62832c02010-09-29 16:04:18 +040071 struct net *, struct sockaddr *,
72 int, int);
Trond Myklebust9e00abc2011-07-13 19:20:49 -040073#if defined(CONFIG_SUNRPC_BACKCHANNEL)
Andy Adamson1f11a032011-01-06 02:04:26 +000074static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
75 struct net *, struct sockaddr *,
76 int, int);
77static void svc_bc_sock_free(struct svc_xprt *xprt);
Trond Myklebust9e00abc2011-07-13 19:20:49 -040078#endif /* CONFIG_SUNRPC_BACKCHANNEL */
Andy Adamson1f11a032011-01-06 02:04:26 +000079
Peter Zijlstraed075362006-12-06 20:35:24 -080080#ifdef CONFIG_DEBUG_LOCK_ALLOC
81static struct lock_class_key svc_key[2];
82static struct lock_class_key svc_slock_key[2];
83
Tom Tucker0f0257e2007-12-30 21:08:27 -060084static void svc_reclassify_socket(struct socket *sock)
Peter Zijlstraed075362006-12-06 20:35:24 -080085{
86 struct sock *sk = sock->sk;
Weston Andros Adamson1b7a1812012-10-23 10:43:39 -040087
88 WARN_ON_ONCE(sock_owned_by_user(sk));
89 if (sock_owned_by_user(sk))
90 return;
91
Peter Zijlstraed075362006-12-06 20:35:24 -080092 switch (sk->sk_family) {
93 case AF_INET:
94 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
Tom Tuckerdef13d72007-12-30 21:08:08 -060095 &svc_slock_key[0],
96 "sk_xprt.xpt_lock-AF_INET-NFSD",
97 &svc_key[0]);
Peter Zijlstraed075362006-12-06 20:35:24 -080098 break;
99
100 case AF_INET6:
101 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
Tom Tuckerdef13d72007-12-30 21:08:08 -0600102 &svc_slock_key[1],
103 "sk_xprt.xpt_lock-AF_INET6-NFSD",
104 &svc_key[1]);
Peter Zijlstraed075362006-12-06 20:35:24 -0800105 break;
106
107 default:
108 BUG();
109 }
110}
111#else
Tom Tucker0f0257e2007-12-30 21:08:27 -0600112static void svc_reclassify_socket(struct socket *sock)
Peter Zijlstraed075362006-12-06 20:35:24 -0800113{
114}
115#endif
116
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117/*
118 * Release an skbuff after use
119 */
Tom Tucker5148bf42007-12-30 21:07:25 -0600120static void svc_release_skb(struct svc_rqst *rqstp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121{
Tom Tucker5148bf42007-12-30 21:07:25 -0600122 struct sk_buff *skb = rqstp->rq_xprt_ctxt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123
124 if (skb) {
Tom Tucker57b1d3b2007-12-30 21:08:22 -0600125 struct svc_sock *svsk =
126 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
Tom Tucker5148bf42007-12-30 21:07:25 -0600127 rqstp->rq_xprt_ctxt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128
129 dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
Eric Dumazet9d410c72009-10-30 05:03:53 +0000130 skb_free_datagram_locked(svsk->sk_sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132}
133
Chuck Leverb92503b2007-02-12 00:53:36 -0800134union svc_pktinfo_u {
135 struct in_pktinfo pkti;
Chuck Leverb92503b2007-02-12 00:53:36 -0800136 struct in6_pktinfo pkti6;
Chuck Leverb92503b2007-02-12 00:53:36 -0800137};
David S. Millerbc375ea2007-04-12 13:35:59 -0700138#define SVC_PKTINFO_SPACE \
139 CMSG_SPACE(sizeof(union svc_pktinfo_u))
Chuck Leverb92503b2007-02-12 00:53:36 -0800140
141static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
142{
Tom Tucker57b1d3b2007-12-30 21:08:22 -0600143 struct svc_sock *svsk =
144 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
145 switch (svsk->sk_sk->sk_family) {
Chuck Leverb92503b2007-02-12 00:53:36 -0800146 case AF_INET: {
147 struct in_pktinfo *pki = CMSG_DATA(cmh);
148
149 cmh->cmsg_level = SOL_IP;
150 cmh->cmsg_type = IP_PKTINFO;
151 pki->ipi_ifindex = 0;
Mi Jinlong849a1cf2011-08-30 17:18:41 +0800152 pki->ipi_spec_dst.s_addr =
153 svc_daddr_in(rqstp)->sin_addr.s_addr;
Chuck Leverb92503b2007-02-12 00:53:36 -0800154 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
155 }
156 break;
NeilBrown5a05ed72007-03-06 01:42:22 -0800157
Chuck Leverb92503b2007-02-12 00:53:36 -0800158 case AF_INET6: {
159 struct in6_pktinfo *pki = CMSG_DATA(cmh);
Mi Jinlong849a1cf2011-08-30 17:18:41 +0800160 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
Chuck Leverb92503b2007-02-12 00:53:36 -0800161
162 cmh->cmsg_level = SOL_IPV6;
163 cmh->cmsg_type = IPV6_PKTINFO;
Mi Jinlong849a1cf2011-08-30 17:18:41 +0800164 pki->ipi6_ifindex = daddr->sin6_scope_id;
Alexey Dobriyan4e3fd7a2011-11-21 03:39:03 +0000165 pki->ipi6_addr = daddr->sin6_addr;
Chuck Leverb92503b2007-02-12 00:53:36 -0800166 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
167 }
168 break;
Chuck Leverb92503b2007-02-12 00:53:36 -0800169 }
Chuck Leverb92503b2007-02-12 00:53:36 -0800170}
171
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172/*
Rahul Iyer4cfc7e62009-09-10 17:32:28 +0300173 * send routine intended to be shared by the fore- and back-channel
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174 */
Rahul Iyer4cfc7e62009-09-10 17:32:28 +0300175int svc_send_common(struct socket *sock, struct xdr_buf *xdr,
176 struct page *headpage, unsigned long headoffset,
177 struct page *tailpage, unsigned long tailoffset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 int result;
180 int size;
181 struct page **ppage = xdr->pages;
182 size_t base = xdr->page_base;
183 unsigned int pglen = xdr->page_len;
184 unsigned int flags = MSG_MORE;
Rahul Iyer4cfc7e62009-09-10 17:32:28 +0300185 int slen;
186 int len = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187
188 slen = xdr->len;
189
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 /* send head */
191 if (slen == xdr->head[0].iov_len)
192 flags = 0;
Rahul Iyer4cfc7e62009-09-10 17:32:28 +0300193 len = kernel_sendpage(sock, headpage, headoffset,
NeilBrown44524352006-10-04 02:15:46 -0700194 xdr->head[0].iov_len, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 if (len != xdr->head[0].iov_len)
196 goto out;
197 slen -= xdr->head[0].iov_len;
198 if (slen == 0)
199 goto out;
200
201 /* send page data */
202 size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen;
203 while (pglen > 0) {
204 if (slen == size)
205 flags = 0;
Sridhar Samudralae6242e92006-08-07 20:58:01 -0700206 result = kernel_sendpage(sock, *ppage, base, size, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 if (result > 0)
208 len += result;
209 if (result != size)
210 goto out;
211 slen -= size;
212 pglen -= size;
213 size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen;
214 base = 0;
215 ppage++;
216 }
Rahul Iyer4cfc7e62009-09-10 17:32:28 +0300217
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 /* send tail */
219 if (xdr->tail[0].iov_len) {
Rahul Iyer4cfc7e62009-09-10 17:32:28 +0300220 result = kernel_sendpage(sock, tailpage, tailoffset,
221 xdr->tail[0].iov_len, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 if (result > 0)
223 len += result;
224 }
Rahul Iyer4cfc7e62009-09-10 17:32:28 +0300225
226out:
227 return len;
228}
229
230
231/*
232 * Generic sendto routine
233 */
234static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr)
235{
236 struct svc_sock *svsk =
237 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
238 struct socket *sock = svsk->sk_sock;
239 union {
240 struct cmsghdr hdr;
241 long all[SVC_PKTINFO_SPACE / sizeof(long)];
242 } buffer;
243 struct cmsghdr *cmh = &buffer.hdr;
244 int len = 0;
245 unsigned long tailoff;
246 unsigned long headoff;
247 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
248
249 if (rqstp->rq_prot == IPPROTO_UDP) {
250 struct msghdr msg = {
251 .msg_name = &rqstp->rq_addr,
252 .msg_namelen = rqstp->rq_addrlen,
253 .msg_control = cmh,
254 .msg_controllen = sizeof(buffer),
255 .msg_flags = MSG_MORE,
256 };
257
258 svc_set_cmsg_data(rqstp, cmh);
259
260 if (sock_sendmsg(sock, &msg, 0) < 0)
261 goto out;
262 }
263
264 tailoff = ((unsigned long)xdr->tail[0].iov_base) & (PAGE_SIZE-1);
265 headoff = 0;
266 len = svc_send_common(sock, xdr, rqstp->rq_respages[0], headoff,
267 rqstp->rq_respages[0], tailoff);
268
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269out:
Chuck Leverad06e4b2007-02-12 00:53:32 -0800270 dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n",
Tom Tucker57b1d3b2007-12-30 21:08:22 -0600271 svsk, xdr->head[0].iov_base, xdr->head[0].iov_len,
Chuck Leverad06e4b2007-02-12 00:53:32 -0800272 xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273
274 return len;
275}
276
277/*
NeilBrown80212d52006-10-02 02:17:47 -0700278 * Report socket names for nfsdfs
279 */
Chuck Levere7942b92009-04-23 19:32:48 -0400280static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining)
NeilBrown80212d52006-10-02 02:17:47 -0700281{
Chuck Lever017cb472009-04-23 19:33:03 -0400282 const struct sock *sk = svsk->sk_sk;
283 const char *proto_name = sk->sk_protocol == IPPROTO_UDP ?
284 "udp" : "tcp";
NeilBrown80212d52006-10-02 02:17:47 -0700285 int len;
286
Chuck Lever017cb472009-04-23 19:33:03 -0400287 switch (sk->sk_family) {
Chuck Levere7942b92009-04-23 19:32:48 -0400288 case PF_INET:
289 len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n",
Chuck Lever017cb472009-04-23 19:33:03 -0400290 proto_name,
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000291 &inet_sk(sk)->inet_rcv_saddr,
292 inet_sk(sk)->inet_num);
NeilBrown80212d52006-10-02 02:17:47 -0700293 break;
Eric Dumazetc2bb06d2013-10-09 03:05:48 -0700294#if IS_ENABLED(CONFIG_IPV6)
Chuck Lever58de2f82009-04-23 19:32:55 -0400295 case PF_INET6:
296 len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n",
Chuck Lever017cb472009-04-23 19:33:03 -0400297 proto_name,
Eric Dumazetefe42082013-10-03 15:42:29 -0700298 &sk->sk_v6_rcv_saddr,
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000299 inet_sk(sk)->inet_num);
NeilBrown80212d52006-10-02 02:17:47 -0700300 break;
Eric Dumazetc2bb06d2013-10-09 03:05:48 -0700301#endif
NeilBrown80212d52006-10-02 02:17:47 -0700302 default:
Chuck Levere7942b92009-04-23 19:32:48 -0400303 len = snprintf(buf, remaining, "*unknown-%d*\n",
Chuck Lever017cb472009-04-23 19:33:03 -0400304 sk->sk_family);
NeilBrown80212d52006-10-02 02:17:47 -0700305 }
Chuck Levere7942b92009-04-23 19:32:48 -0400306
307 if (len >= remaining) {
308 *buf = '\0';
309 return -ENAMETOOLONG;
NeilBrown80212d52006-10-02 02:17:47 -0700310 }
311 return len;
312}
313
NeilBrown80212d52006-10-02 02:17:47 -0700314/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 * Check input queue length
316 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600317static int svc_recv_available(struct svc_sock *svsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319 struct socket *sock = svsk->sk_sock;
320 int avail, err;
321
Sridhar Samudralae6242e92006-08-07 20:58:01 -0700322 err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323
324 return (err >= 0)? avail : err;
325}
326
327/*
328 * Generic recvfrom routine.
329 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600330static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr,
331 int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332{
Tom Tucker57b1d3b2007-12-30 21:08:22 -0600333 struct svc_sock *svsk =
334 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
Chuck Lever1ba95102007-02-12 00:53:31 -0800335 struct msghdr msg = {
336 .msg_flags = MSG_DONTWAIT,
337 };
338 int len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339
Tom Tucker260c1d12007-12-30 21:08:29 -0600340 rqstp->rq_xprt_hlen = 0;
341
Chuck Lever1ba95102007-02-12 00:53:31 -0800342 len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen,
343 msg.msg_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
Chuck Lever1ba95102007-02-12 00:53:31 -0800346 svsk, iov[0].iov_base, iov[0].iov_len, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 return len;
348}
349
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800350static int svc_partial_recvfrom(struct svc_rqst *rqstp,
351 struct kvec *iov, int nr,
352 int buflen, unsigned int base)
353{
354 size_t save_iovlen;
Trond Myklebust09acfea2012-03-11 15:22:54 -0400355 void *save_iovbase;
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800356 unsigned int i;
357 int ret;
358
359 if (base == 0)
360 return svc_recvfrom(rqstp, iov, nr, buflen);
361
362 for (i = 0; i < nr; i++) {
363 if (iov[i].iov_len > base)
364 break;
365 base -= iov[i].iov_len;
366 }
367 save_iovlen = iov[i].iov_len;
368 save_iovbase = iov[i].iov_base;
369 iov[i].iov_len -= base;
370 iov[i].iov_base += base;
371 ret = svc_recvfrom(rqstp, &iov[i], nr - i, buflen);
372 iov[i].iov_len = save_iovlen;
373 iov[i].iov_base = save_iovbase;
374 return ret;
375}
376
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377/*
378 * Set socket snd and rcv buffer lengths
379 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600380static void svc_sock_setbufsize(struct socket *sock, unsigned int snd,
381 unsigned int rcv)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382{
383#if 0
384 mm_segment_t oldfs;
385 oldfs = get_fs(); set_fs(KERNEL_DS);
386 sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
387 (char*)&snd, sizeof(snd));
388 sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
389 (char*)&rcv, sizeof(rcv));
390#else
391 /* sock_setsockopt limits use to sysctl_?mem_max,
392 * which isn't acceptable. Until that is made conditional
393 * on not having CAP_SYS_RESOURCE or similar, we go direct...
394 * DaveM said I could!
395 */
396 lock_sock(sock->sk);
397 sock->sk->sk_sndbuf = snd * 2;
398 sock->sk->sk_rcvbuf = rcv * 2;
Trond Myklebust47fcb032009-05-18 17:47:56 -0400399 sock->sk->sk_write_space(sock->sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 release_sock(sock->sk);
401#endif
402}
403/*
404 * INET callback when data has been received on the socket.
405 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600406static void svc_udp_data_ready(struct sock *sk, int count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407{
Neil Brown939bb7e2005-09-13 01:25:39 -0700408 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
Eric Dumazeteaefd112011-02-18 03:26:36 +0000409 wait_queue_head_t *wq = sk_sleep(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410
Neil Brown939bb7e2005-09-13 01:25:39 -0700411 if (svsk) {
412 dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n",
Tom Tucker02fc6c32007-12-30 21:07:48 -0600413 svsk, sk, count,
414 test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
415 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
Tom Tuckerf6150c32007-12-30 21:07:57 -0600416 svc_xprt_enqueue(&svsk->sk_xprt);
Neil Brown939bb7e2005-09-13 01:25:39 -0700417 }
Eric Dumazeteaefd112011-02-18 03:26:36 +0000418 if (wq && waitqueue_active(wq))
419 wake_up_interruptible(wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420}
421
422/*
423 * INET callback when space is newly available on the socket.
424 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600425static void svc_write_space(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426{
427 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
Eric Dumazeteaefd112011-02-18 03:26:36 +0000428 wait_queue_head_t *wq = sk_sleep(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429
430 if (svsk) {
431 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
Tom Tucker02fc6c32007-12-30 21:07:48 -0600432 svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
Tom Tuckerf6150c32007-12-30 21:07:57 -0600433 svc_xprt_enqueue(&svsk->sk_xprt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434 }
435
Eric Dumazeteaefd112011-02-18 03:26:36 +0000436 if (wq && waitqueue_active(wq)) {
Neil Brown939bb7e2005-09-13 01:25:39 -0700437 dprintk("RPC svc_write_space: someone sleeping on %p\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 svsk);
Eric Dumazeteaefd112011-02-18 03:26:36 +0000439 wake_up_interruptible(wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440 }
441}
442
Trond Myklebust47fcb032009-05-18 17:47:56 -0400443static void svc_tcp_write_space(struct sock *sk)
444{
445 struct socket *sock = sk->sk_socket;
446
Eric Dumazet64dc6132013-07-22 20:26:31 -0700447 if (sk_stream_is_writeable(sk) && sock)
Trond Myklebust47fcb032009-05-18 17:47:56 -0400448 clear_bit(SOCK_NOSPACE, &sock->flags);
449 svc_write_space(sk);
450}
451
Tom Tucker9dbc2402007-12-30 21:08:12 -0600452/*
Chuck Lever7702ce42009-07-13 10:54:26 -0400453 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
454 */
455static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
456 struct cmsghdr *cmh)
457{
458 struct in_pktinfo *pki = CMSG_DATA(cmh);
Mi Jinlong849a1cf2011-08-30 17:18:41 +0800459 struct sockaddr_in *daddr = svc_daddr_in(rqstp);
460
Chuck Lever7702ce42009-07-13 10:54:26 -0400461 if (cmh->cmsg_type != IP_PKTINFO)
462 return 0;
Mi Jinlong849a1cf2011-08-30 17:18:41 +0800463
464 daddr->sin_family = AF_INET;
465 daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr;
Chuck Lever7702ce42009-07-13 10:54:26 -0400466 return 1;
467}
468
469/*
Tom Parkin73df66f2013-01-31 01:02:24 +0000470 * See net/ipv6/datagram.c : ip6_datagram_recv_ctl
Chuck Lever7702ce42009-07-13 10:54:26 -0400471 */
472static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
473 struct cmsghdr *cmh)
474{
475 struct in6_pktinfo *pki = CMSG_DATA(cmh);
Mi Jinlong849a1cf2011-08-30 17:18:41 +0800476 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
477
Chuck Lever7702ce42009-07-13 10:54:26 -0400478 if (cmh->cmsg_type != IPV6_PKTINFO)
479 return 0;
Mi Jinlong849a1cf2011-08-30 17:18:41 +0800480
481 daddr->sin6_family = AF_INET6;
Alexey Dobriyan4e3fd7a2011-11-21 03:39:03 +0000482 daddr->sin6_addr = pki->ipi6_addr;
Mi Jinlong849a1cf2011-08-30 17:18:41 +0800483 daddr->sin6_scope_id = pki->ipi6_ifindex;
Chuck Lever7702ce42009-07-13 10:54:26 -0400484 return 1;
485}
486
487/*
Tom Tucker9dbc2402007-12-30 21:08:12 -0600488 * Copy the UDP datagram's destination address to the rqstp structure.
489 * The 'destination' address in this case is the address to which the
490 * peer sent the datagram, i.e. our local address. For multihomed
491 * hosts, this can change from msg to msg. Note that only the IP
492 * address changes, the port number should remain the same.
493 */
Chuck Lever7702ce42009-07-13 10:54:26 -0400494static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
495 struct cmsghdr *cmh)
Chuck Lever95756482007-02-12 00:53:38 -0800496{
Chuck Lever7702ce42009-07-13 10:54:26 -0400497 switch (cmh->cmsg_level) {
498 case SOL_IP:
499 return svc_udp_get_dest_address4(rqstp, cmh);
500 case SOL_IPV6:
501 return svc_udp_get_dest_address6(rqstp, cmh);
Chuck Lever95756482007-02-12 00:53:38 -0800502 }
Chuck Lever7702ce42009-07-13 10:54:26 -0400503
504 return 0;
Chuck Lever95756482007-02-12 00:53:38 -0800505}
506
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507/*
508 * Receive a datagram from a UDP socket.
509 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600510static int svc_udp_recvfrom(struct svc_rqst *rqstp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511{
Tom Tucker57b1d3b2007-12-30 21:08:22 -0600512 struct svc_sock *svsk =
513 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600514 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 struct sk_buff *skb;
David S. Millerbc375ea2007-04-12 13:35:59 -0700516 union {
517 struct cmsghdr hdr;
518 long all[SVC_PKTINFO_SPACE / sizeof(long)];
519 } buffer;
520 struct cmsghdr *cmh = &buffer.hdr;
NeilBrown7a37f572007-03-06 01:42:21 -0800521 struct msghdr msg = {
522 .msg_name = svc_addr(rqstp),
523 .msg_control = cmh,
524 .msg_controllen = sizeof(buffer),
525 .msg_flags = MSG_DONTWAIT,
526 };
Chuck Leverabc5c442009-04-23 19:31:25 -0400527 size_t len;
528 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529
Tom Tucker02fc6c32007-12-30 21:07:48 -0600530 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 /* udp sockets need large rcvbuf as all pending
532 * requests are still in that buffer. sndbuf must
533 * also be large enough that there is enough space
Greg Banks3262c812006-10-02 02:17:58 -0700534 * for one reply per thread. We count all threads
535 * rather than threads in a particular pool, which
536 * provides an upper bound on the number of threads
537 * which will access the socket.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 */
539 svc_sock_setbufsize(svsk->sk_sock,
NeilBrownc6b0a9f2006-10-06 00:44:05 -0700540 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
541 (serv->sv_nrthreads+3) * serv->sv_max_mesg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542
Tom Tucker02fc6c32007-12-30 21:07:48 -0600543 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
NeilBrown05ed6902007-05-09 02:34:55 -0700544 skb = NULL;
545 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
546 0, 0, MSG_PEEK | MSG_DONTWAIT);
547 if (err >= 0)
548 skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err);
549
550 if (skb == NULL) {
551 if (err != -EAGAIN) {
552 /* possibly an icmp error */
553 dprintk("svc: recvfrom returned error %d\n", -err);
Tom Tucker02fc6c32007-12-30 21:07:48 -0600554 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 }
J. Bruce Fields9f9d2eb2012-08-17 21:35:24 -0400556 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 }
Tom Tucker9dbc2402007-12-30 21:08:12 -0600558 len = svc_addr_len(svc_addr(rqstp));
Tom Tucker9dbc2402007-12-30 21:08:12 -0600559 rqstp->rq_addrlen = len;
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -0700560 if (skb->tstamp.tv64 == 0) {
561 skb->tstamp = ktime_get_real();
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -0800562 /* Don't enable netstamp, sunrpc doesn't
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 need that much accuracy */
564 }
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -0700565 svsk->sk_sk->sk_stamp = skb->tstamp;
Tom Tucker02fc6c32007-12-30 21:07:48 -0600566 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 len = skb->len - sizeof(struct udphdr);
569 rqstp->rq_arg.len = len;
570
Chuck Lever95756482007-02-12 00:53:38 -0800571 rqstp->rq_prot = IPPROTO_UDP;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572
Chuck Lever7702ce42009-07-13 10:54:26 -0400573 if (!svc_udp_get_dest_address(rqstp, cmh)) {
Joe Perchese87cc472012-05-13 21:56:26 +0000574 net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n",
575 cmh->cmsg_level, cmh->cmsg_type);
J. Bruce Fieldsf23abfd2012-08-17 20:32:27 -0400576 goto out_free;
NeilBrown7a37f572007-03-06 01:42:21 -0800577 }
Mi Jinlong849a1cf2011-08-30 17:18:41 +0800578 rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579
580 if (skb_is_nonlinear(skb)) {
581 /* we have to copy */
582 local_bh_disable();
583 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
584 local_bh_enable();
585 /* checksum error */
Eric Dumazet22911fc2012-06-27 00:23:44 +0000586 goto out_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 }
588 local_bh_enable();
Eric Dumazet9d410c72009-10-30 05:03:53 +0000589 skb_free_datagram_locked(svsk->sk_sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 } else {
591 /* we can use it in-place */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600592 rqstp->rq_arg.head[0].iov_base = skb->data +
593 sizeof(struct udphdr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 rqstp->rq_arg.head[0].iov_len = len;
Eric Dumazet22911fc2012-06-27 00:23:44 +0000595 if (skb_checksum_complete(skb))
596 goto out_free;
Tom Tucker5148bf42007-12-30 21:07:25 -0600597 rqstp->rq_xprt_ctxt = skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 }
599
600 rqstp->rq_arg.page_base = 0;
601 if (len <= rqstp->rq_arg.head[0].iov_len) {
602 rqstp->rq_arg.head[0].iov_len = len;
603 rqstp->rq_arg.page_len = 0;
NeilBrown44524352006-10-04 02:15:46 -0700604 rqstp->rq_respages = rqstp->rq_pages+1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 } else {
606 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
NeilBrown44524352006-10-04 02:15:46 -0700607 rqstp->rq_respages = rqstp->rq_pages + 1 +
Ilpo Järvinen172589c2007-08-28 15:50:33 -0700608 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 }
J. Bruce Fieldsafc59402012-12-10 18:01:37 -0500610 rqstp->rq_next_page = rqstp->rq_respages+1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611
612 if (serv->sv_stats)
613 serv->sv_stats->netudpcnt++;
614
615 return len;
J. Bruce Fieldsf23abfd2012-08-17 20:32:27 -0400616out_free:
617 trace_kfree_skb(skb, svc_udp_recvfrom);
618 skb_free_datagram_locked(svsk->sk_sk, skb);
619 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620}
621
622static int
623svc_udp_sendto(struct svc_rqst *rqstp)
624{
625 int error;
626
627 error = svc_sendto(rqstp, &rqstp->rq_res);
628 if (error == -ECONNREFUSED)
629 /* ICMP error on earlier request. */
630 error = svc_sendto(rqstp, &rqstp->rq_res);
631
632 return error;
633}
634
Tom Tuckere831fe62007-12-30 21:07:29 -0600635static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
636{
637}
638
Tom Tucker323bee32007-12-30 21:07:31 -0600639static int svc_udp_has_wspace(struct svc_xprt *xprt)
640{
641 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600642 struct svc_serv *serv = xprt->xpt_server;
Tom Tucker323bee32007-12-30 21:07:31 -0600643 unsigned long required;
644
645 /*
646 * Set the SOCK_NOSPACE flag before checking the available
647 * sock space.
648 */
649 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
Tom Tucker7a90e8c2007-12-30 21:07:55 -0600650 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
Tom Tucker323bee32007-12-30 21:07:31 -0600651 if (required*2 > sock_wspace(svsk->sk_sk))
652 return 0;
653 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
654 return 1;
655}
656
Tom Tucker38a417c2007-12-30 21:07:36 -0600657static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
658{
659 BUG();
660 return NULL;
661}
662
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600663static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
Pavel Emelyanov62832c02010-09-29 16:04:18 +0400664 struct net *net,
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600665 struct sockaddr *sa, int salen,
666 int flags)
667{
Pavel Emelyanov62832c02010-09-29 16:04:18 +0400668 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600669}
670
Tom Tucker360d87382007-12-30 21:07:17 -0600671static struct svc_xprt_ops svc_udp_ops = {
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600672 .xpo_create = svc_udp_create,
Tom Tucker5d137992007-12-30 21:07:23 -0600673 .xpo_recvfrom = svc_udp_recvfrom,
674 .xpo_sendto = svc_udp_sendto,
Tom Tucker5148bf42007-12-30 21:07:25 -0600675 .xpo_release_rqst = svc_release_skb,
Tom Tucker755ccea2007-12-30 21:07:27 -0600676 .xpo_detach = svc_sock_detach,
677 .xpo_free = svc_sock_free,
Tom Tuckere831fe62007-12-30 21:07:29 -0600678 .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
Tom Tucker323bee32007-12-30 21:07:31 -0600679 .xpo_has_wspace = svc_udp_has_wspace,
Tom Tucker38a417c2007-12-30 21:07:36 -0600680 .xpo_accept = svc_udp_accept,
Tom Tucker360d87382007-12-30 21:07:17 -0600681};
682
683static struct svc_xprt_class svc_udp_class = {
684 .xcl_name = "udp",
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600685 .xcl_owner = THIS_MODULE,
Tom Tucker360d87382007-12-30 21:07:17 -0600686 .xcl_ops = &svc_udp_ops,
Tom Tucker49023152007-12-30 21:07:21 -0600687 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
Tom Tucker360d87382007-12-30 21:07:17 -0600688};
689
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600690static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691{
Chuck Lever7702ce42009-07-13 10:54:26 -0400692 int err, level, optname, one = 1;
NeilBrown7a37f572007-03-06 01:42:21 -0800693
Stanislav Kinsburskybd4620d2011-12-06 14:19:10 +0300694 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class,
695 &svsk->sk_xprt, serv);
Tom Tuckerdef13d72007-12-30 21:08:08 -0600696 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 svsk->sk_sk->sk_data_ready = svc_udp_data_ready;
698 svsk->sk_sk->sk_write_space = svc_write_space;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699
700 /* initialise setting must have enough space to
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -0800701 * receive and respond to one request.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 * svc_udp_recvfrom will re-adjust if necessary
703 */
704 svc_sock_setbufsize(svsk->sk_sock,
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600705 3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
706 3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707
Tom Tucker0f0257e2007-12-30 21:08:27 -0600708 /* data might have come in before data_ready set up */
709 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
Tom Tucker02fc6c32007-12-30 21:07:48 -0600710 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
NeilBrown7a37f572007-03-06 01:42:21 -0800711
NeilBrown7a37f572007-03-06 01:42:21 -0800712 /* make sure we get destination address info */
Chuck Lever7702ce42009-07-13 10:54:26 -0400713 switch (svsk->sk_sk->sk_family) {
714 case AF_INET:
715 level = SOL_IP;
716 optname = IP_PKTINFO;
717 break;
718 case AF_INET6:
719 level = SOL_IPV6;
720 optname = IPV6_RECVPKTINFO;
721 break;
722 default:
723 BUG();
724 }
725 err = kernel_setsockopt(svsk->sk_sock, level, optname,
726 (char *)&one, sizeof(one));
727 dprintk("svc: kernel_setsockopt returned %d\n", err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728}
729
730/*
731 * A data_ready event on a listening socket means there's a connection
732 * pending. Do not use state_change as a substitute for it.
733 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600734static void svc_tcp_listen_data_ready(struct sock *sk, int count_unused)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735{
Neil Brown939bb7e2005-09-13 01:25:39 -0700736 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
Eric Dumazeteaefd112011-02-18 03:26:36 +0000737 wait_queue_head_t *wq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738
739 dprintk("svc: socket %p TCP (listen) state change %d\n",
Neil Brown939bb7e2005-09-13 01:25:39 -0700740 sk, sk->sk_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741
Neil Brown939bb7e2005-09-13 01:25:39 -0700742 /*
743 * This callback may called twice when a new connection
744 * is established as a child socket inherits everything
745 * from a parent LISTEN socket.
746 * 1) data_ready method of the parent socket will be called
747 * when one of child sockets become ESTABLISHED.
748 * 2) data_ready method of the child socket may be called
749 * when it receives data before the socket is accepted.
750 * In case of 2, we should ignore it silently.
751 */
752 if (sk->sk_state == TCP_LISTEN) {
753 if (svsk) {
Tom Tucker02fc6c32007-12-30 21:07:48 -0600754 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
Tom Tuckerf6150c32007-12-30 21:07:57 -0600755 svc_xprt_enqueue(&svsk->sk_xprt);
Neil Brown939bb7e2005-09-13 01:25:39 -0700756 } else
757 printk("svc: socket %p: no user data\n", sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 }
Neil Brown939bb7e2005-09-13 01:25:39 -0700759
Eric Dumazeteaefd112011-02-18 03:26:36 +0000760 wq = sk_sleep(sk);
761 if (wq && waitqueue_active(wq))
762 wake_up_interruptible_all(wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763}
764
765/*
766 * A state change on a connected socket means it's dying or dead.
767 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600768static void svc_tcp_state_change(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769{
Neil Brown939bb7e2005-09-13 01:25:39 -0700770 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
Eric Dumazeteaefd112011-02-18 03:26:36 +0000771 wait_queue_head_t *wq = sk_sleep(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772
773 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
Neil Brown939bb7e2005-09-13 01:25:39 -0700774 sk, sk->sk_state, sk->sk_user_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775
Neil Brown939bb7e2005-09-13 01:25:39 -0700776 if (!svsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 printk("svc: socket %p: no user data\n", sk);
Neil Brown939bb7e2005-09-13 01:25:39 -0700778 else {
Tom Tucker02fc6c32007-12-30 21:07:48 -0600779 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
Tom Tuckerf6150c32007-12-30 21:07:57 -0600780 svc_xprt_enqueue(&svsk->sk_xprt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 }
Eric Dumazeteaefd112011-02-18 03:26:36 +0000782 if (wq && waitqueue_active(wq))
783 wake_up_interruptible_all(wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784}
785
Tom Tucker0f0257e2007-12-30 21:08:27 -0600786static void svc_tcp_data_ready(struct sock *sk, int count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787{
Neil Brown939bb7e2005-09-13 01:25:39 -0700788 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
Eric Dumazeteaefd112011-02-18 03:26:36 +0000789 wait_queue_head_t *wq = sk_sleep(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790
791 dprintk("svc: socket %p TCP data ready (svsk %p)\n",
Neil Brown939bb7e2005-09-13 01:25:39 -0700792 sk, sk->sk_user_data);
793 if (svsk) {
Tom Tucker02fc6c32007-12-30 21:07:48 -0600794 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
Tom Tuckerf6150c32007-12-30 21:07:57 -0600795 svc_xprt_enqueue(&svsk->sk_xprt);
Neil Brown939bb7e2005-09-13 01:25:39 -0700796 }
Eric Dumazeteaefd112011-02-18 03:26:36 +0000797 if (wq && waitqueue_active(wq))
798 wake_up_interruptible(wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799}
800
801/*
802 * Accept a TCP connection
803 */
Tom Tucker38a417c2007-12-30 21:07:36 -0600804static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805{
Tom Tucker38a417c2007-12-30 21:07:36 -0600806 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
akpm@linux-foundation.orgcdd88b92007-02-12 00:53:38 -0800807 struct sockaddr_storage addr;
808 struct sockaddr *sin = (struct sockaddr *) &addr;
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600809 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 struct socket *sock = svsk->sk_sock;
811 struct socket *newsock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 struct svc_sock *newsvsk;
813 int err, slen;
Pavel Emelyanov5216a8e2008-02-21 10:57:45 +0300814 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815
816 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
817 if (!sock)
Tom Tucker38a417c2007-12-30 21:07:36 -0600818 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819
Tom Tucker02fc6c32007-12-30 21:07:48 -0600820 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
Sridhar Samudralae6242e92006-08-07 20:58:01 -0700821 err = kernel_accept(sock, &newsock, O_NONBLOCK);
822 if (err < 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 if (err == -ENOMEM)
824 printk(KERN_WARNING "%s: no more sockets!\n",
825 serv->sv_name);
Joe Perchese87cc472012-05-13 21:56:26 +0000826 else if (err != -EAGAIN)
827 net_warn_ratelimited("%s: accept failed (err %d)!\n",
828 serv->sv_name, -err);
Tom Tucker38a417c2007-12-30 21:07:36 -0600829 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 }
Tom Tucker02fc6c32007-12-30 21:07:48 -0600831 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832
akpm@linux-foundation.orgcdd88b92007-02-12 00:53:38 -0800833 err = kernel_getpeername(newsock, sin, &slen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 if (err < 0) {
Joe Perchese87cc472012-05-13 21:56:26 +0000835 net_warn_ratelimited("%s: peername failed (err %d)!\n",
836 serv->sv_name, -err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 goto failed; /* aborted connection or whatever */
838 }
839
840 /* Ideally, we would want to reject connections from unauthorized
Chuck Leverad06e4b2007-02-12 00:53:32 -0800841 * hosts here, but when we get encryption, the IP of the host won't
842 * tell us anything. For now just warn about unpriv connections.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 */
akpm@linux-foundation.orgcdd88b92007-02-12 00:53:38 -0800844 if (!svc_port_is_privileged(sin)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 dprintk(KERN_WARNING
Chuck Leverad06e4b2007-02-12 00:53:32 -0800846 "%s: connect from unprivileged port: %s\n",
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -0800847 serv->sv_name,
akpm@linux-foundation.orgcdd88b92007-02-12 00:53:38 -0800848 __svc_print_addr(sin, buf, sizeof(buf)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 }
Chuck Leverad06e4b2007-02-12 00:53:32 -0800850 dprintk("%s: connect from %s\n", serv->sv_name,
akpm@linux-foundation.orgcdd88b92007-02-12 00:53:38 -0800851 __svc_print_addr(sin, buf, sizeof(buf)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852
853 /* make sure that a write doesn't block forever when
854 * low on memory
855 */
856 newsock->sk->sk_sndtimeo = HZ*30;
857
J. Bruce Fields72c35372012-08-13 17:46:17 -0400858 newsvsk = svc_setup_socket(serv, newsock,
859 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY));
860 if (IS_ERR(newsvsk))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 goto failed;
Tom Tucker9dbc2402007-12-30 21:08:12 -0600862 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
Frank van Maarseveena9747692007-07-09 22:21:39 +0200863 err = kernel_getsockname(newsock, sin, &slen);
864 if (unlikely(err < 0)) {
865 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
866 slen = offsetof(struct sockaddr, sa_data);
867 }
Tom Tucker9dbc2402007-12-30 21:08:12 -0600868 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
Chuck Lever067d7812007-02-12 00:53:30 -0800869
Tom Tuckerf9f3cc42007-12-30 21:07:40 -0600870 if (serv->sv_stats)
871 serv->sv_stats->nettcpconn++;
872
873 return &newsvsk->sk_xprt;
874
875failed:
876 sock_release(newsock);
877 return NULL;
878}
879
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800880static unsigned int svc_tcp_restore_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
881{
882 unsigned int i, len, npages;
883
J. Bruce Fields8af345f2012-12-03 16:45:35 -0500884 if (svsk->sk_datalen == 0)
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800885 return 0;
J. Bruce Fields8af345f2012-12-03 16:45:35 -0500886 len = svsk->sk_datalen;
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800887 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
888 for (i = 0; i < npages; i++) {
889 if (rqstp->rq_pages[i] != NULL)
890 put_page(rqstp->rq_pages[i]);
891 BUG_ON(svsk->sk_pages[i] == NULL);
892 rqstp->rq_pages[i] = svsk->sk_pages[i];
893 svsk->sk_pages[i] = NULL;
894 }
895 rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
896 return len;
897}
898
899static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
900{
901 unsigned int i, len, npages;
902
J. Bruce Fields8af345f2012-12-03 16:45:35 -0500903 if (svsk->sk_datalen == 0)
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800904 return;
J. Bruce Fields8af345f2012-12-03 16:45:35 -0500905 len = svsk->sk_datalen;
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800906 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
907 for (i = 0; i < npages; i++) {
908 svsk->sk_pages[i] = rqstp->rq_pages[i];
909 rqstp->rq_pages[i] = NULL;
910 }
911}
912
913static void svc_tcp_clear_pages(struct svc_sock *svsk)
914{
915 unsigned int i, len, npages;
916
J. Bruce Fields8af345f2012-12-03 16:45:35 -0500917 if (svsk->sk_datalen == 0)
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800918 goto out;
J. Bruce Fields8af345f2012-12-03 16:45:35 -0500919 len = svsk->sk_datalen;
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800920 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
921 for (i = 0; i < npages; i++) {
J. Bruce Fieldscf3aa022013-06-26 11:09:06 -0400922 if (svsk->sk_pages[i] == NULL) {
923 WARN_ON_ONCE(1);
924 continue;
925 }
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800926 put_page(svsk->sk_pages[i]);
927 svsk->sk_pages[i] = NULL;
928 }
929out:
930 svsk->sk_tcplen = 0;
J. Bruce Fields8af345f2012-12-03 16:45:35 -0500931 svsk->sk_datalen = 0;
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800932}
933
Tom Tuckerf9f3cc42007-12-30 21:07:40 -0600934/*
J. Bruce Fieldsad46ccf2012-12-03 16:30:42 -0500935 * Receive fragment record header.
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +0300936 * If we haven't gotten the record length yet, get the next four bytes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 */
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +0300938static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939{
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600940 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
Trond Myklebust5ee78d42009-05-18 17:47:56 -0400941 unsigned int want;
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +0300942 int len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943
Tom Tucker02fc6c32007-12-30 21:07:48 -0600944 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945
Chuck Leverc0401ea2008-04-14 12:27:30 -0400946 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 struct kvec iov;
948
Trond Myklebust5ee78d42009-05-18 17:47:56 -0400949 want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
951 iov.iov_len = want;
952 if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
953 goto error;
954 svsk->sk_tcplen += len;
955
956 if (len < want) {
Chuck Leverc0401ea2008-04-14 12:27:30 -0400957 dprintk("svc: short recvfrom while reading record "
958 "length (%d of %d)\n", len, want);
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800959 return -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 }
961
J. Bruce Fieldscc248d42012-12-03 16:11:13 -0500962 dprintk("svc: TCP record, %d bytes\n", svc_sock_reclen(svsk));
J. Bruce Fields836fbad2012-12-03 15:50:38 -0500963 if (svc_sock_reclen(svsk) + svsk->sk_datalen >
964 serv->sv_max_mesg) {
J. Bruce Fields3a28e332012-12-10 18:15:07 -0500965 net_notice_ratelimited("RPC: fragment too large: %d\n",
966 svc_sock_reclen(svsk));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967 goto err_delete;
968 }
969 }
970
J. Bruce Fieldscc248d42012-12-03 16:11:13 -0500971 return svc_sock_reclen(svsk);
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800972error:
973 dprintk("RPC: TCP recv_record got %d\n", len);
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +0300974 return len;
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800975err_delete:
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +0300976 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +0300977 return -EAGAIN;
978}
979
Trond Myklebust586c52c2009-05-18 17:47:56 -0400980static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
Rahul Iyer4cfc7e62009-09-10 17:32:28 +0300981{
Trond Myklebust586c52c2009-05-18 17:47:56 -0400982 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
Rahul Iyer4cfc7e62009-09-10 17:32:28 +0300983 struct rpc_rqst *req = NULL;
Trond Myklebust586c52c2009-05-18 17:47:56 -0400984 struct kvec *src, *dst;
985 __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
J. Bruce Fields48e65552011-02-14 14:52:03 -0500986 __be32 xid;
987 __be32 calldir;
Rahul Iyer4cfc7e62009-09-10 17:32:28 +0300988
Rahul Iyer4cfc7e62009-09-10 17:32:28 +0300989 xid = *p++;
990 calldir = *p;
991
Trond Myklebust586c52c2009-05-18 17:47:56 -0400992 if (bc_xprt)
993 req = xprt_lookup_rqst(bc_xprt, xid);
J. Bruce Fieldsd75faea2010-11-30 19:15:01 -0500994
Trond Myklebust586c52c2009-05-18 17:47:56 -0400995 if (!req) {
996 printk(KERN_NOTICE
997 "%s: Got unrecognized reply: "
998 "calldir 0x%x xpt_bc_xprt %p xid %08x\n",
999 __func__, ntohl(calldir),
1000 bc_xprt, xid);
1001 return -EAGAIN;
Rahul Iyer4cfc7e62009-09-10 17:32:28 +03001002 }
Trond Myklebust586c52c2009-05-18 17:47:56 -04001003
1004 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
1005 /*
1006 * XXX!: cheating for now! Only copying HEAD.
1007 * But we know this is good enough for now (in fact, for any
1008 * callback reply in the forseeable future).
1009 */
1010 dst = &req->rq_private_buf.head[0];
1011 src = &rqstp->rq_arg.head[0];
1012 if (dst->iov_len < src->iov_len)
1013 return -EAGAIN; /* whatever; just giving up. */
1014 memcpy(dst->iov_base, src->iov_base, src->iov_len);
J. Bruce Fieldscc248d42012-12-03 16:11:13 -05001015 xprt_complete_rqst(req->rq_task, rqstp->rq_arg.len);
Trond Myklebust586c52c2009-05-18 17:47:56 -04001016 rqstp->rq_arg.len = 0;
1017 return 0;
1018}
1019
1020static int copy_pages_to_kvecs(struct kvec *vec, struct page **pages, int len)
1021{
1022 int i = 0;
1023 int t = 0;
1024
1025 while (t < len) {
1026 vec[i].iov_base = page_address(pages[i]);
1027 vec[i].iov_len = PAGE_SIZE;
1028 i++;
1029 t += PAGE_SIZE;
1030 }
1031 return i;
Rahul Iyer4cfc7e62009-09-10 17:32:28 +03001032}
1033
J. Bruce Fields836fbad2012-12-03 15:50:38 -05001034static void svc_tcp_fragment_received(struct svc_sock *svsk)
1035{
1036 /* If we have more data, signal svc_xprt_enqueue() to try again */
1037 if (svc_recv_available(svsk) > sizeof(rpc_fraghdr))
1038 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1039 dprintk("svc: TCP %s record (%d bytes)\n",
1040 svc_sock_final_rec(svsk) ? "final" : "nonfinal",
1041 svc_sock_reclen(svsk));
1042 svsk->sk_tcplen = 0;
1043 svsk->sk_reclen = 0;
1044}
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001045
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +03001046/*
1047 * Receive data from a TCP socket.
1048 */
1049static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
1050{
1051 struct svc_sock *svsk =
1052 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
1053 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1054 int len;
1055 struct kvec *vec;
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001056 unsigned int want, base;
Trond Myklebust586c52c2009-05-18 17:47:56 -04001057 __be32 *p;
1058 __be32 calldir;
Trond Myklebust5ee78d42009-05-18 17:47:56 -04001059 int pnum;
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +03001060
1061 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1062 svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
1063 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
1064 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1065
1066 len = svc_tcp_recv_record(svsk, rqstp);
1067 if (len < 0)
1068 goto error;
1069
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001070 base = svc_tcp_restore_pages(svsk, rqstp);
J. Bruce Fields836fbad2012-12-03 15:50:38 -05001071 want = svc_sock_reclen(svsk) - (svsk->sk_tcplen - sizeof(rpc_fraghdr));
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001072
NeilBrown3cc03b12006-10-04 02:15:47 -07001073 vec = rqstp->rq_vec;
Rahul Iyer4cfc7e62009-09-10 17:32:28 +03001074
Trond Myklebust586c52c2009-05-18 17:47:56 -04001075 pnum = copy_pages_to_kvecs(&vec[0], &rqstp->rq_pages[0],
J. Bruce Fields836fbad2012-12-03 15:50:38 -05001076 svsk->sk_datalen + want);
Rahul Iyer4cfc7e62009-09-10 17:32:28 +03001077
NeilBrown44524352006-10-04 02:15:46 -07001078 rqstp->rq_respages = &rqstp->rq_pages[pnum];
J. Bruce Fieldsafc59402012-12-10 18:01:37 -05001079 rqstp->rq_next_page = rqstp->rq_respages + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080
1081 /* Now receive data */
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001082 len = svc_partial_recvfrom(rqstp, vec, pnum, want, base);
J. Bruce Fields8af345f2012-12-03 16:45:35 -05001083 if (len >= 0) {
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001084 svsk->sk_tcplen += len;
J. Bruce Fields8af345f2012-12-03 16:45:35 -05001085 svsk->sk_datalen += len;
1086 }
J. Bruce Fields836fbad2012-12-03 15:50:38 -05001087 if (len != want || !svc_sock_final_rec(svsk)) {
J. Bruce Fieldsbe1e4442012-08-09 18:12:28 -04001088 svc_tcp_save_pages(svsk, rqstp);
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001089 if (len < 0 && len != -EAGAIN)
J. Bruce Fieldsad46ccf2012-12-03 16:30:42 -05001090 goto err_delete;
J. Bruce Fields836fbad2012-12-03 15:50:38 -05001091 if (len == want)
1092 svc_tcp_fragment_received(svsk);
1093 else
J. Bruce Fields3a28e332012-12-10 18:15:07 -05001094 dprintk("svc: incomplete TCP record (%d of %d)\n",
1095 (int)(svsk->sk_tcplen - sizeof(rpc_fraghdr)),
J. Bruce Fields836fbad2012-12-03 15:50:38 -05001096 svc_sock_reclen(svsk));
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001097 goto err_noclose;
1098 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099
J. Bruce Fields1f691b02013-06-26 10:55:40 -04001100 if (svsk->sk_datalen < 8) {
J. Bruce Fieldscf3aa022013-06-26 11:09:06 -04001101 svsk->sk_datalen = 0;
J. Bruce Fieldsad46ccf2012-12-03 16:30:42 -05001102 goto err_delete; /* client is nuts. */
J. Bruce Fieldscf3aa022013-06-26 11:09:06 -04001103 }
J. Bruce Fieldsad46ccf2012-12-03 16:30:42 -05001104
J. Bruce Fields836fbad2012-12-03 15:50:38 -05001105 rqstp->rq_arg.len = svsk->sk_datalen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106 rqstp->rq_arg.page_base = 0;
Trond Myklebust5ee78d42009-05-18 17:47:56 -04001107 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1108 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109 rqstp->rq_arg.page_len = 0;
Trond Myklebust5ee78d42009-05-18 17:47:56 -04001110 } else
1111 rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112
Tom Tucker5148bf42007-12-30 21:07:25 -06001113 rqstp->rq_xprt_ctxt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114 rqstp->rq_prot = IPPROTO_TCP;
1115
Trond Myklebust586c52c2009-05-18 17:47:56 -04001116 p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1117 calldir = p[1];
J. Bruce Fields8985ef02011-04-09 10:03:10 -04001118 if (calldir)
Trond Myklebust586c52c2009-05-18 17:47:56 -04001119 len = receive_cb_reply(svsk, rqstp);
Trond Myklebust586c52c2009-05-18 17:47:56 -04001120
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 /* Reset TCP read info */
J. Bruce Fields8af345f2012-12-03 16:45:35 -05001122 svsk->sk_datalen = 0;
J. Bruce Fields836fbad2012-12-03 15:50:38 -05001123 svc_tcp_fragment_received(svsk);
Trond Myklebust0601f792009-05-18 17:47:56 -04001124
J. Bruce Fields8985ef02011-04-09 10:03:10 -04001125 if (len < 0)
1126 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127
Tom Tucker9dbc2402007-12-30 21:08:12 -06001128 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129 if (serv->sv_stats)
1130 serv->sv_stats->nettcpcnt++;
1131
Trond Myklebust5ee78d42009-05-18 17:47:56 -04001132 return rqstp->rq_arg.len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +03001134error:
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001135 if (len != -EAGAIN)
J. Bruce Fieldsad46ccf2012-12-03 16:30:42 -05001136 goto err_delete;
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001137 dprintk("RPC: TCP recvfrom got EAGAIN\n");
J. Bruce Fields9f9d2eb2012-08-17 21:35:24 -04001138 return 0;
J. Bruce Fieldsad46ccf2012-12-03 16:30:42 -05001139err_delete:
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001140 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1141 svsk->sk_xprt.xpt_server->sv_name, -len);
1142 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1143err_noclose:
J. Bruce Fields9f9d2eb2012-08-17 21:35:24 -04001144 return 0; /* record not complete */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145}
1146
1147/*
1148 * Send out data on TCP socket.
1149 */
Tom Tucker0f0257e2007-12-30 21:08:27 -06001150static int svc_tcp_sendto(struct svc_rqst *rqstp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151{
1152 struct xdr_buf *xbufp = &rqstp->rq_res;
1153 int sent;
Alexey Dobriyand8ed0292006-09-26 22:29:38 -07001154 __be32 reclen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155
1156 /* Set up the first element of the reply kvec.
1157 * Any other kvecs that may be in use have been taken
1158 * care of by the server implementation itself.
1159 */
1160 reclen = htonl(0x80000000|((xbufp->len ) - 4));
1161 memcpy(xbufp->head[0].iov_base, &reclen, 4);
1162
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163 sent = svc_sendto(rqstp, &rqstp->rq_res);
1164 if (sent != xbufp->len) {
Tom Tucker0f0257e2007-12-30 21:08:27 -06001165 printk(KERN_NOTICE
1166 "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1167 "- shutting down socket\n",
Tom Tucker57b1d3b2007-12-30 21:08:22 -06001168 rqstp->rq_xprt->xpt_server->sv_name,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 (sent<0)?"got error":"sent only",
1170 sent, xbufp->len);
Tom Tucker57b1d3b2007-12-30 21:08:22 -06001171 set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
Tom Tuckerf6150c32007-12-30 21:07:57 -06001172 svc_xprt_enqueue(rqstp->rq_xprt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 sent = -EAGAIN;
1174 }
1175 return sent;
1176}
1177
Tom Tuckere831fe62007-12-30 21:07:29 -06001178/*
1179 * Setup response header. TCP has a 4B record length field.
1180 */
1181static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
1182{
1183 struct kvec *resv = &rqstp->rq_res.head[0];
1184
1185 /* tcp needs a space for the record length... */
1186 svc_putnl(resv, 0);
1187}
1188
Tom Tucker323bee32007-12-30 21:07:31 -06001189static int svc_tcp_has_wspace(struct svc_xprt *xprt)
1190{
Tom Tucker57b1d3b2007-12-30 21:08:22 -06001191 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
Trond Myklebust47fcb032009-05-18 17:47:56 -04001192 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
Tom Tucker323bee32007-12-30 21:07:31 -06001193 int required;
Tom Tucker323bee32007-12-30 21:07:31 -06001194
Trond Myklebust47fcb032009-05-18 17:47:56 -04001195 if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
1196 return 1;
1197 required = atomic_read(&xprt->xpt_reserved) + serv->sv_max_mesg;
NeilBrown447383d2013-07-25 11:30:23 +10001198 if (sk_stream_wspace(svsk->sk_sk) >= required ||
1199 (sk_stream_min_wspace(svsk->sk_sk) == 0 &&
1200 atomic_read(&xprt->xpt_reserved) == 0))
Trond Myklebust47fcb032009-05-18 17:47:56 -04001201 return 1;
Tom Tucker323bee32007-12-30 21:07:31 -06001202 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
Trond Myklebust47fcb032009-05-18 17:47:56 -04001203 return 0;
Tom Tucker323bee32007-12-30 21:07:31 -06001204}
1205
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001206static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
Pavel Emelyanov62832c02010-09-29 16:04:18 +04001207 struct net *net,
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001208 struct sockaddr *sa, int salen,
1209 int flags)
1210{
Pavel Emelyanov62832c02010-09-29 16:04:18 +04001211 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001212}
1213
Trond Myklebust9e00abc2011-07-13 19:20:49 -04001214#if defined(CONFIG_SUNRPC_BACKCHANNEL)
Andy Adamson1f11a032011-01-06 02:04:26 +00001215static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
1216 struct net *, struct sockaddr *,
1217 int, int);
1218static void svc_bc_sock_free(struct svc_xprt *xprt);
1219
1220static struct svc_xprt *svc_bc_tcp_create(struct svc_serv *serv,
1221 struct net *net,
1222 struct sockaddr *sa, int salen,
1223 int flags)
1224{
1225 return svc_bc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1226}
1227
1228static void svc_bc_tcp_sock_detach(struct svc_xprt *xprt)
1229{
1230}
1231
1232static struct svc_xprt_ops svc_tcp_bc_ops = {
1233 .xpo_create = svc_bc_tcp_create,
1234 .xpo_detach = svc_bc_tcp_sock_detach,
1235 .xpo_free = svc_bc_sock_free,
1236 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1237};
1238
1239static struct svc_xprt_class svc_tcp_bc_class = {
1240 .xcl_name = "tcp-bc",
1241 .xcl_owner = THIS_MODULE,
1242 .xcl_ops = &svc_tcp_bc_ops,
1243 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1244};
Andy Adamson16b2d1e2011-01-06 02:04:27 +00001245
1246static void svc_init_bc_xprt_sock(void)
1247{
1248 svc_reg_xprt_class(&svc_tcp_bc_class);
1249}
1250
1251static void svc_cleanup_bc_xprt_sock(void)
1252{
1253 svc_unreg_xprt_class(&svc_tcp_bc_class);
1254}
Trond Myklebust9e00abc2011-07-13 19:20:49 -04001255#else /* CONFIG_SUNRPC_BACKCHANNEL */
Andy Adamson16b2d1e2011-01-06 02:04:27 +00001256static void svc_init_bc_xprt_sock(void)
1257{
1258}
1259
1260static void svc_cleanup_bc_xprt_sock(void)
1261{
1262}
Trond Myklebust9e00abc2011-07-13 19:20:49 -04001263#endif /* CONFIG_SUNRPC_BACKCHANNEL */
Andy Adamson1f11a032011-01-06 02:04:26 +00001264
Tom Tucker360d87382007-12-30 21:07:17 -06001265static struct svc_xprt_ops svc_tcp_ops = {
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001266 .xpo_create = svc_tcp_create,
Tom Tucker5d137992007-12-30 21:07:23 -06001267 .xpo_recvfrom = svc_tcp_recvfrom,
1268 .xpo_sendto = svc_tcp_sendto,
Tom Tucker5148bf42007-12-30 21:07:25 -06001269 .xpo_release_rqst = svc_release_skb,
Trond Myklebust69b6ba32008-12-23 16:30:11 -05001270 .xpo_detach = svc_tcp_sock_detach,
Tom Tucker755ccea2007-12-30 21:07:27 -06001271 .xpo_free = svc_sock_free,
Tom Tuckere831fe62007-12-30 21:07:29 -06001272 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
Tom Tucker323bee32007-12-30 21:07:31 -06001273 .xpo_has_wspace = svc_tcp_has_wspace,
Tom Tucker38a417c2007-12-30 21:07:36 -06001274 .xpo_accept = svc_tcp_accept,
Tom Tucker360d87382007-12-30 21:07:17 -06001275};
1276
1277static struct svc_xprt_class svc_tcp_class = {
1278 .xcl_name = "tcp",
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001279 .xcl_owner = THIS_MODULE,
Tom Tucker360d87382007-12-30 21:07:17 -06001280 .xcl_ops = &svc_tcp_ops,
Tom Tucker49023152007-12-30 21:07:21 -06001281 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
Tom Tucker360d87382007-12-30 21:07:17 -06001282};
1283
1284void svc_init_xprt_sock(void)
1285{
1286 svc_reg_xprt_class(&svc_tcp_class);
1287 svc_reg_xprt_class(&svc_udp_class);
Andy Adamson16b2d1e2011-01-06 02:04:27 +00001288 svc_init_bc_xprt_sock();
Tom Tucker360d87382007-12-30 21:07:17 -06001289}
1290
1291void svc_cleanup_xprt_sock(void)
1292{
1293 svc_unreg_xprt_class(&svc_tcp_class);
1294 svc_unreg_xprt_class(&svc_udp_class);
Andy Adamson16b2d1e2011-01-06 02:04:27 +00001295 svc_cleanup_bc_xprt_sock();
Tom Tucker360d87382007-12-30 21:07:17 -06001296}
1297
Tom Tuckerbb5cf162007-12-30 21:07:50 -06001298static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299{
1300 struct sock *sk = svsk->sk_sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301
Stanislav Kinsburskybd4620d2011-12-06 14:19:10 +03001302 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1303 &svsk->sk_xprt, serv);
Tom Tuckerdef13d72007-12-30 21:08:08 -06001304 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 if (sk->sk_state == TCP_LISTEN) {
1306 dprintk("setting up TCP socket for listening\n");
Tom Tucker02fc6c32007-12-30 21:07:48 -06001307 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308 sk->sk_data_ready = svc_tcp_listen_data_ready;
Tom Tucker02fc6c32007-12-30 21:07:48 -06001309 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310 } else {
1311 dprintk("setting up TCP socket for reading\n");
1312 sk->sk_state_change = svc_tcp_state_change;
1313 sk->sk_data_ready = svc_tcp_data_ready;
Trond Myklebust47fcb032009-05-18 17:47:56 -04001314 sk->sk_write_space = svc_tcp_write_space;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315
1316 svsk->sk_reclen = 0;
1317 svsk->sk_tcplen = 0;
J. Bruce Fields8af345f2012-12-03 16:45:35 -05001318 svsk->sk_datalen = 0;
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001319 memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320
Chuck Leverb7872fe2008-04-14 12:27:01 -04001321 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322
Tom Tucker02fc6c32007-12-30 21:07:48 -06001323 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -08001324 if (sk->sk_state != TCP_ESTABLISHED)
Tom Tucker02fc6c32007-12-30 21:07:48 -06001325 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326 }
1327}
1328
Tom Tucker0f0257e2007-12-30 21:08:27 -06001329void svc_sock_update_bufs(struct svc_serv *serv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330{
1331 /*
1332 * The number of server threads has changed. Update
1333 * rcvbuf and sndbuf accordingly on all sockets
1334 */
Pavel Emelyanov8f3a6de2010-10-05 23:30:19 +04001335 struct svc_sock *svsk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336
1337 spin_lock_bh(&serv->sv_lock);
Pavel Emelyanov8f3a6de2010-10-05 23:30:19 +04001338 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
Tom Tucker02fc6c32007-12-30 21:07:48 -06001339 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 spin_unlock_bh(&serv->sv_lock);
1341}
Trond Myklebust24c37672008-12-23 16:30:12 -05001342EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343
1344/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345 * Initialize socket for RPC use and create svc_sock struct
1346 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1347 */
Chuck Lever6b174332007-02-12 00:53:28 -08001348static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1349 struct socket *sock,
J. Bruce Fields72c35372012-08-13 17:46:17 -04001350 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351{
1352 struct svc_sock *svsk;
1353 struct sock *inet;
Chuck Lever6b174332007-02-12 00:53:28 -08001354 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
J. Bruce Fields72c35372012-08-13 17:46:17 -04001355 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356
1357 dprintk("svc: svc_setup_socket %p\n", sock);
J. Bruce Fields72c35372012-08-13 17:46:17 -04001358 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1359 if (!svsk)
1360 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361
1362 inet = sock->sk;
1363
1364 /* Register socket with portmapper */
J. Bruce Fields72c35372012-08-13 17:46:17 -04001365 if (pmap_register)
1366 err = svc_register(serv, sock_net(sock->sk), inet->sk_family,
Stanislav Kinsbursky5247fab2012-01-13 14:02:48 +04001367 inet->sk_protocol,
Eric Dumazetc720c7e2009-10-15 06:30:45 +00001368 ntohs(inet_sk(inet)->inet_sport));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369
J. Bruce Fields72c35372012-08-13 17:46:17 -04001370 if (err < 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 kfree(svsk);
J. Bruce Fields72c35372012-08-13 17:46:17 -04001372 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 }
1374
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375 inet->sk_user_data = svsk;
1376 svsk->sk_sock = sock;
1377 svsk->sk_sk = inet;
1378 svsk->sk_ostate = inet->sk_state_change;
1379 svsk->sk_odata = inet->sk_data_ready;
1380 svsk->sk_owspace = inet->sk_write_space;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381
1382 /* Initialize the socket */
1383 if (sock->type == SOCK_DGRAM)
Tom Tuckerbb5cf162007-12-30 21:07:50 -06001384 svc_udp_init(svsk, serv);
Olga Kornievskaia96604392008-10-21 14:13:47 -04001385 else {
1386 /* initialise setting must have enough space to
1387 * receive and respond to one request.
1388 */
1389 svc_sock_setbufsize(svsk->sk_sock, 4 * serv->sv_max_mesg,
1390 4 * serv->sv_max_mesg);
Tom Tuckerbb5cf162007-12-30 21:07:50 -06001391 svc_tcp_init(svsk, serv);
Olga Kornievskaia96604392008-10-21 14:13:47 -04001392 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394 dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1395 svsk, svsk->sk_sk);
1396
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 return svsk;
1398}
1399
Chuck Leverbfba9ab2009-04-23 19:32:33 -04001400/**
1401 * svc_addsock - add a listener socket to an RPC service
1402 * @serv: pointer to RPC service to which to add a new listener
1403 * @fd: file descriptor of the new listener
1404 * @name_return: pointer to buffer to fill in with name of listener
1405 * @len: size of the buffer
1406 *
1407 * Fills in socket name and returns positive length of name if successful.
1408 * Name is terminated with '\n'. On error, returns a negative errno
1409 * value.
1410 */
1411int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1412 const size_t len)
NeilBrownb41b66d2006-10-02 02:17:48 -07001413{
1414 int err = 0;
1415 struct socket *so = sockfd_lookup(fd, &err);
1416 struct svc_sock *svsk = NULL;
J. Bruce Fieldsa8e10072012-08-13 18:01:03 -04001417 struct sockaddr_storage addr;
1418 struct sockaddr *sin = (struct sockaddr *)&addr;
1419 int salen;
NeilBrownb41b66d2006-10-02 02:17:48 -07001420
1421 if (!so)
1422 return err;
J. Bruce Fieldsa8e10072012-08-13 18:01:03 -04001423 err = -EAFNOSUPPORT;
Aime Le Rouzic205ba422010-01-26 14:03:56 -05001424 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
J. Bruce Fieldsa8e10072012-08-13 18:01:03 -04001425 goto out;
1426 err = -EPROTONOSUPPORT;
1427 if (so->sk->sk_protocol != IPPROTO_TCP &&
NeilBrownb41b66d2006-10-02 02:17:48 -07001428 so->sk->sk_protocol != IPPROTO_UDP)
J. Bruce Fieldsa8e10072012-08-13 18:01:03 -04001429 goto out;
1430 err = -EISCONN;
1431 if (so->state > SS_UNCONNECTED)
1432 goto out;
1433 err = -ENOENT;
1434 if (!try_module_get(THIS_MODULE))
1435 goto out;
1436 svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS);
1437 if (IS_ERR(svsk)) {
1438 module_put(THIS_MODULE);
1439 err = PTR_ERR(svsk);
1440 goto out;
NeilBrownb41b66d2006-10-02 02:17:48 -07001441 }
J. Bruce Fieldsa8e10072012-08-13 18:01:03 -04001442 if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0)
1443 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
J. Bruce Fields39b55302012-08-14 15:50:34 -04001444 svc_add_new_perm_xprt(serv, &svsk->sk_xprt);
Chuck Levere7942b92009-04-23 19:32:48 -04001445 return svc_one_sock_name(svsk, name_return, len);
J. Bruce Fieldsa8e10072012-08-13 18:01:03 -04001446out:
1447 sockfd_put(so);
1448 return err;
NeilBrownb41b66d2006-10-02 02:17:48 -07001449}
1450EXPORT_SYMBOL_GPL(svc_addsock);
1451
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452/*
1453 * Create socket for RPC service.
1454 */
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001455static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1456 int protocol,
Pavel Emelyanov62832c02010-09-29 16:04:18 +04001457 struct net *net,
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001458 struct sockaddr *sin, int len,
1459 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460{
1461 struct svc_sock *svsk;
1462 struct socket *sock;
1463 int error;
1464 int type;
Tom Tucker9dbc2402007-12-30 21:08:12 -06001465 struct sockaddr_storage addr;
1466 struct sockaddr *newsin = (struct sockaddr *)&addr;
1467 int newlen;
Trond Myklebustc69da772009-03-30 18:59:17 -04001468 int family;
1469 int val;
Pavel Emelyanov5216a8e2008-02-21 10:57:45 +03001470 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471
Chuck Leverad06e4b2007-02-12 00:53:32 -08001472 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1473 serv->sv_program->pg_name, protocol,
Chuck Lever77f1f672007-02-12 00:53:39 -08001474 __svc_print_addr(sin, buf, sizeof(buf)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475
1476 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1477 printk(KERN_WARNING "svc: only UDP and TCP "
1478 "sockets supported\n");
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001479 return ERR_PTR(-EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481
Trond Myklebustc69da772009-03-30 18:59:17 -04001482 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1483 switch (sin->sa_family) {
1484 case AF_INET6:
1485 family = PF_INET6;
1486 break;
1487 case AF_INET:
1488 family = PF_INET;
1489 break;
1490 default:
1491 return ERR_PTR(-EINVAL);
1492 }
1493
Pavel Emelyanov14ec63c2010-09-29 16:06:57 +04001494 error = __sock_create(net, family, type, protocol, &sock, 1);
Chuck Lever77f1f672007-02-12 00:53:39 -08001495 if (error < 0)
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001496 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497
Peter Zijlstraed075362006-12-06 20:35:24 -08001498 svc_reclassify_socket(sock);
1499
Trond Myklebustc69da772009-03-30 18:59:17 -04001500 /*
1501 * If this is an PF_INET6 listener, we want to avoid
1502 * getting requests from IPv4 remotes. Those should
1503 * be shunted to a PF_INET listener via rpcbind.
1504 */
1505 val = 1;
1506 if (family == PF_INET6)
1507 kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
1508 (char *)&val, sizeof(val));
1509
Eric Sesterhenn18114742006-09-28 14:37:07 -07001510 if (type == SOCK_STREAM)
Pavel Emelyanov4a17fd52012-04-19 03:39:36 +00001511 sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */
Chuck Lever77f1f672007-02-12 00:53:39 -08001512 error = kernel_bind(sock, sin, len);
Eric Sesterhenn18114742006-09-28 14:37:07 -07001513 if (error < 0)
1514 goto bummer;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515
Tom Tucker9dbc2402007-12-30 21:08:12 -06001516 newlen = len;
1517 error = kernel_getsockname(sock, newsin, &newlen);
1518 if (error < 0)
1519 goto bummer;
1520
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521 if (protocol == IPPROTO_TCP) {
Sridhar Samudralae6242e92006-08-07 20:58:01 -07001522 if ((error = kernel_listen(sock, 64)) < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523 goto bummer;
1524 }
1525
J. Bruce Fields72c35372012-08-13 17:46:17 -04001526 svsk = svc_setup_socket(serv, sock, flags);
J. Bruce Fieldsa8e10072012-08-13 18:01:03 -04001527 if (IS_ERR(svsk)) {
1528 error = PTR_ERR(svsk);
1529 goto bummer;
NeilBrowne79eff12007-02-12 00:53:30 -08001530 }
J. Bruce Fieldsa8e10072012-08-13 18:01:03 -04001531 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1532 return (struct svc_xprt *)svsk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533bummer:
1534 dprintk("svc: svc_create_socket error = %d\n", -error);
1535 sock_release(sock);
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001536 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537}
1538
1539/*
Tom Tucker755ccea2007-12-30 21:07:27 -06001540 * Detach the svc_sock from the socket so that no
1541 * more callbacks occur.
1542 */
1543static void svc_sock_detach(struct svc_xprt *xprt)
1544{
1545 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1546 struct sock *sk = svsk->sk_sk;
Eric Dumazeteaefd112011-02-18 03:26:36 +00001547 wait_queue_head_t *wq;
Tom Tucker755ccea2007-12-30 21:07:27 -06001548
1549 dprintk("svc: svc_sock_detach(%p)\n", svsk);
1550
1551 /* put back the old socket callbacks */
1552 sk->sk_state_change = svsk->sk_ostate;
1553 sk->sk_data_ready = svsk->sk_odata;
1554 sk->sk_write_space = svsk->sk_owspace;
Trond Myklebust69b6ba32008-12-23 16:30:11 -05001555
Eric Dumazeteaefd112011-02-18 03:26:36 +00001556 wq = sk_sleep(sk);
1557 if (wq && waitqueue_active(wq))
1558 wake_up_interruptible(wq);
Trond Myklebust69b6ba32008-12-23 16:30:11 -05001559}
1560
1561/*
1562 * Disconnect the socket, and reset the callbacks
1563 */
1564static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1565{
1566 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1567
1568 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
1569
1570 svc_sock_detach(xprt);
1571
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001572 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1573 svc_tcp_clear_pages(svsk);
Trond Myklebust69b6ba32008-12-23 16:30:11 -05001574 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001575 }
Tom Tucker755ccea2007-12-30 21:07:27 -06001576}
1577
1578/*
1579 * Free the svc_sock's socket resources and the svc_sock itself.
1580 */
1581static void svc_sock_free(struct svc_xprt *xprt)
1582{
1583 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1584 dprintk("svc: svc_sock_free(%p)\n", svsk);
1585
Tom Tucker755ccea2007-12-30 21:07:27 -06001586 if (svsk->sk_sock->file)
1587 sockfd_put(svsk->sk_sock);
1588 else
1589 sock_release(svsk->sk_sock);
1590 kfree(svsk);
1591}
Benny Halevy7652e5a2009-04-01 09:23:09 -04001592
Trond Myklebust9e00abc2011-07-13 19:20:49 -04001593#if defined(CONFIG_SUNRPC_BACKCHANNEL)
Benny Halevy7652e5a2009-04-01 09:23:09 -04001594/*
Andy Adamson1f11a032011-01-06 02:04:26 +00001595 * Create a back channel svc_xprt which shares the fore channel socket.
Benny Halevy7652e5a2009-04-01 09:23:09 -04001596 */
Andy Adamson1f11a032011-01-06 02:04:26 +00001597static struct svc_xprt *svc_bc_create_socket(struct svc_serv *serv,
1598 int protocol,
1599 struct net *net,
1600 struct sockaddr *sin, int len,
1601 int flags)
Benny Halevy7652e5a2009-04-01 09:23:09 -04001602{
1603 struct svc_sock *svsk;
Andy Adamson1f11a032011-01-06 02:04:26 +00001604 struct svc_xprt *xprt;
Benny Halevy7652e5a2009-04-01 09:23:09 -04001605
Andy Adamson1f11a032011-01-06 02:04:26 +00001606 if (protocol != IPPROTO_TCP) {
1607 printk(KERN_WARNING "svc: only TCP sockets"
1608 " supported on shared back channel\n");
1609 return ERR_PTR(-EINVAL);
1610 }
1611
Benny Halevy7652e5a2009-04-01 09:23:09 -04001612 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1613 if (!svsk)
Andy Adamson1f11a032011-01-06 02:04:26 +00001614 return ERR_PTR(-ENOMEM);
Benny Halevy7652e5a2009-04-01 09:23:09 -04001615
1616 xprt = &svsk->sk_xprt;
Stanislav Kinsburskybd4620d2011-12-06 14:19:10 +03001617 svc_xprt_init(net, &svc_tcp_bc_class, xprt, serv);
Andy Adamson1f11a032011-01-06 02:04:26 +00001618
Andy Adamson4a19de02011-01-06 02:04:35 +00001619 serv->sv_bc_xprt = xprt;
Andy Adamson1f11a032011-01-06 02:04:26 +00001620
Benny Halevy7652e5a2009-04-01 09:23:09 -04001621 return xprt;
1622}
Benny Halevy7652e5a2009-04-01 09:23:09 -04001623
1624/*
Andy Adamson1f11a032011-01-06 02:04:26 +00001625 * Free a back channel svc_sock.
Benny Halevy7652e5a2009-04-01 09:23:09 -04001626 */
Andy Adamson1f11a032011-01-06 02:04:26 +00001627static void svc_bc_sock_free(struct svc_xprt *xprt)
Benny Halevy7652e5a2009-04-01 09:23:09 -04001628{
Andy Adamson778be232011-01-25 15:38:01 +00001629 if (xprt)
Benny Halevy7652e5a2009-04-01 09:23:09 -04001630 kfree(container_of(xprt, struct svc_sock, sk_xprt));
1631}
Trond Myklebust9e00abc2011-07-13 19:20:49 -04001632#endif /* CONFIG_SUNRPC_BACKCHANNEL */