blob: 3f959c681885ba41ccdf6300bb27a35745ec11be [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);
David S. Miller676d2362014-04-11 16:15:36 -040063static void svc_udp_data_ready(struct sock *);
Linus Torvalds1da177e2005-04-16 15:20:36 -070064static 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 * Generic recvfrom routine.
316 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600317static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr,
318 int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319{
Tom Tucker57b1d3b2007-12-30 21:08:22 -0600320 struct svc_sock *svsk =
321 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
Chuck Lever1ba95102007-02-12 00:53:31 -0800322 struct msghdr msg = {
323 .msg_flags = MSG_DONTWAIT,
324 };
325 int len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326
Tom Tucker260c1d12007-12-30 21:08:29 -0600327 rqstp->rq_xprt_hlen = 0;
328
Trond Myklebustf8d1ff42014-08-03 13:03:13 -0400329 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
Chuck Lever1ba95102007-02-12 00:53:31 -0800330 len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen,
331 msg.msg_flags);
Trond Myklebustf8d1ff42014-08-03 13:03:13 -0400332 /* If we read a full record, then assume there may be more
333 * data to read (stream based sockets only!)
334 */
335 if (len == buflen)
336 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
Chuck Lever1ba95102007-02-12 00:53:31 -0800339 svsk, iov[0].iov_base, iov[0].iov_len, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 return len;
341}
342
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800343static int svc_partial_recvfrom(struct svc_rqst *rqstp,
344 struct kvec *iov, int nr,
345 int buflen, unsigned int base)
346{
347 size_t save_iovlen;
Trond Myklebust09acfea2012-03-11 15:22:54 -0400348 void *save_iovbase;
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800349 unsigned int i;
350 int ret;
351
352 if (base == 0)
353 return svc_recvfrom(rqstp, iov, nr, buflen);
354
355 for (i = 0; i < nr; i++) {
356 if (iov[i].iov_len > base)
357 break;
358 base -= iov[i].iov_len;
359 }
360 save_iovlen = iov[i].iov_len;
361 save_iovbase = iov[i].iov_base;
362 iov[i].iov_len -= base;
363 iov[i].iov_base += base;
364 ret = svc_recvfrom(rqstp, &iov[i], nr - i, buflen);
365 iov[i].iov_len = save_iovlen;
366 iov[i].iov_base = save_iovbase;
367 return ret;
368}
369
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370/*
371 * Set socket snd and rcv buffer lengths
372 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600373static void svc_sock_setbufsize(struct socket *sock, unsigned int snd,
374 unsigned int rcv)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375{
376#if 0
377 mm_segment_t oldfs;
378 oldfs = get_fs(); set_fs(KERNEL_DS);
379 sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
380 (char*)&snd, sizeof(snd));
381 sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
382 (char*)&rcv, sizeof(rcv));
383#else
384 /* sock_setsockopt limits use to sysctl_?mem_max,
385 * which isn't acceptable. Until that is made conditional
386 * on not having CAP_SYS_RESOURCE or similar, we go direct...
387 * DaveM said I could!
388 */
389 lock_sock(sock->sk);
390 sock->sk->sk_sndbuf = snd * 2;
391 sock->sk->sk_rcvbuf = rcv * 2;
Trond Myklebust47fcb032009-05-18 17:47:56 -0400392 sock->sk->sk_write_space(sock->sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 release_sock(sock->sk);
394#endif
395}
Chuck Lever16e4d932014-05-19 13:40:22 -0400396
397static int svc_sock_secure_port(struct svc_rqst *rqstp)
398{
399 return svc_port_is_privileged(svc_addr(rqstp));
400}
401
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402/*
403 * INET callback when data has been received on the socket.
404 */
David S. Miller676d2362014-04-11 16:15:36 -0400405static void svc_udp_data_ready(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406{
Neil Brown939bb7e2005-09-13 01:25:39 -0700407 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
Eric Dumazeteaefd112011-02-18 03:26:36 +0000408 wait_queue_head_t *wq = sk_sleep(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409
Neil Brown939bb7e2005-09-13 01:25:39 -0700410 if (svsk) {
David S. Miller676d2362014-04-11 16:15:36 -0400411 dprintk("svc: socket %p(inet %p), busy=%d\n",
412 svsk, sk,
Tom Tucker02fc6c32007-12-30 21:07:48 -0600413 test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
414 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
Tom Tuckerf6150c32007-12-30 21:07:57 -0600415 svc_xprt_enqueue(&svsk->sk_xprt);
Neil Brown939bb7e2005-09-13 01:25:39 -0700416 }
Eric Dumazeteaefd112011-02-18 03:26:36 +0000417 if (wq && waitqueue_active(wq))
418 wake_up_interruptible(wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419}
420
421/*
422 * INET callback when space is newly available on the socket.
423 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600424static void svc_write_space(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425{
426 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
Eric Dumazeteaefd112011-02-18 03:26:36 +0000427 wait_queue_head_t *wq = sk_sleep(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428
429 if (svsk) {
430 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
Tom Tucker02fc6c32007-12-30 21:07:48 -0600431 svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
Tom Tuckerf6150c32007-12-30 21:07:57 -0600432 svc_xprt_enqueue(&svsk->sk_xprt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 }
434
Eric Dumazeteaefd112011-02-18 03:26:36 +0000435 if (wq && waitqueue_active(wq)) {
Neil Brown939bb7e2005-09-13 01:25:39 -0700436 dprintk("RPC svc_write_space: someone sleeping on %p\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 svsk);
Eric Dumazeteaefd112011-02-18 03:26:36 +0000438 wake_up_interruptible(wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439 }
440}
441
Trond Myklebustc7fb3f02014-07-24 23:59:32 -0400442static int svc_tcp_has_wspace(struct svc_xprt *xprt)
443{
444 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
445 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
446 int required;
447
448 if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
449 return 1;
450 required = atomic_read(&xprt->xpt_reserved) + serv->sv_max_mesg;
451 if (sk_stream_wspace(svsk->sk_sk) >= required ||
452 (sk_stream_min_wspace(svsk->sk_sk) == 0 &&
453 atomic_read(&xprt->xpt_reserved) == 0))
454 return 1;
455 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
456 return 0;
457}
458
Trond Myklebust47fcb032009-05-18 17:47:56 -0400459static void svc_tcp_write_space(struct sock *sk)
460{
Trond Myklebustc7fb3f02014-07-24 23:59:32 -0400461 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
Trond Myklebust47fcb032009-05-18 17:47:56 -0400462 struct socket *sock = sk->sk_socket;
463
Trond Myklebustc7fb3f02014-07-24 23:59:32 -0400464 if (!sk_stream_is_writeable(sk) || !sock)
465 return;
466 if (!svsk || svc_tcp_has_wspace(&svsk->sk_xprt))
Trond Myklebust47fcb032009-05-18 17:47:56 -0400467 clear_bit(SOCK_NOSPACE, &sock->flags);
468 svc_write_space(sk);
469}
470
Trond Myklebust51877682014-07-24 23:59:33 -0400471static void svc_tcp_adjust_wspace(struct svc_xprt *xprt)
472{
473 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
474
475 if (svc_tcp_has_wspace(xprt))
476 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
477}
478
Tom Tucker9dbc2402007-12-30 21:08:12 -0600479/*
Chuck Lever7702ce42009-07-13 10:54:26 -0400480 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
481 */
482static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
483 struct cmsghdr *cmh)
484{
485 struct in_pktinfo *pki = CMSG_DATA(cmh);
Mi Jinlong849a1cf2011-08-30 17:18:41 +0800486 struct sockaddr_in *daddr = svc_daddr_in(rqstp);
487
Chuck Lever7702ce42009-07-13 10:54:26 -0400488 if (cmh->cmsg_type != IP_PKTINFO)
489 return 0;
Mi Jinlong849a1cf2011-08-30 17:18:41 +0800490
491 daddr->sin_family = AF_INET;
492 daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr;
Chuck Lever7702ce42009-07-13 10:54:26 -0400493 return 1;
494}
495
496/*
Tom Parkin73df66f2013-01-31 01:02:24 +0000497 * See net/ipv6/datagram.c : ip6_datagram_recv_ctl
Chuck Lever7702ce42009-07-13 10:54:26 -0400498 */
499static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
500 struct cmsghdr *cmh)
501{
502 struct in6_pktinfo *pki = CMSG_DATA(cmh);
Mi Jinlong849a1cf2011-08-30 17:18:41 +0800503 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
504
Chuck Lever7702ce42009-07-13 10:54:26 -0400505 if (cmh->cmsg_type != IPV6_PKTINFO)
506 return 0;
Mi Jinlong849a1cf2011-08-30 17:18:41 +0800507
508 daddr->sin6_family = AF_INET6;
Alexey Dobriyan4e3fd7a2011-11-21 03:39:03 +0000509 daddr->sin6_addr = pki->ipi6_addr;
Mi Jinlong849a1cf2011-08-30 17:18:41 +0800510 daddr->sin6_scope_id = pki->ipi6_ifindex;
Chuck Lever7702ce42009-07-13 10:54:26 -0400511 return 1;
512}
513
514/*
Tom Tucker9dbc2402007-12-30 21:08:12 -0600515 * Copy the UDP datagram's destination address to the rqstp structure.
516 * The 'destination' address in this case is the address to which the
517 * peer sent the datagram, i.e. our local address. For multihomed
518 * hosts, this can change from msg to msg. Note that only the IP
519 * address changes, the port number should remain the same.
520 */
Chuck Lever7702ce42009-07-13 10:54:26 -0400521static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
522 struct cmsghdr *cmh)
Chuck Lever95756482007-02-12 00:53:38 -0800523{
Chuck Lever7702ce42009-07-13 10:54:26 -0400524 switch (cmh->cmsg_level) {
525 case SOL_IP:
526 return svc_udp_get_dest_address4(rqstp, cmh);
527 case SOL_IPV6:
528 return svc_udp_get_dest_address6(rqstp, cmh);
Chuck Lever95756482007-02-12 00:53:38 -0800529 }
Chuck Lever7702ce42009-07-13 10:54:26 -0400530
531 return 0;
Chuck Lever95756482007-02-12 00:53:38 -0800532}
533
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534/*
535 * Receive a datagram from a UDP socket.
536 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600537static int svc_udp_recvfrom(struct svc_rqst *rqstp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538{
Tom Tucker57b1d3b2007-12-30 21:08:22 -0600539 struct svc_sock *svsk =
540 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600541 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 struct sk_buff *skb;
David S. Millerbc375ea2007-04-12 13:35:59 -0700543 union {
544 struct cmsghdr hdr;
545 long all[SVC_PKTINFO_SPACE / sizeof(long)];
546 } buffer;
547 struct cmsghdr *cmh = &buffer.hdr;
NeilBrown7a37f572007-03-06 01:42:21 -0800548 struct msghdr msg = {
549 .msg_name = svc_addr(rqstp),
550 .msg_control = cmh,
551 .msg_controllen = sizeof(buffer),
552 .msg_flags = MSG_DONTWAIT,
553 };
Chuck Leverabc5c442009-04-23 19:31:25 -0400554 size_t len;
555 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556
Tom Tucker02fc6c32007-12-30 21:07:48 -0600557 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 /* udp sockets need large rcvbuf as all pending
559 * requests are still in that buffer. sndbuf must
560 * also be large enough that there is enough space
Greg Banks3262c812006-10-02 02:17:58 -0700561 * for one reply per thread. We count all threads
562 * rather than threads in a particular pool, which
563 * provides an upper bound on the number of threads
564 * which will access the socket.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 */
566 svc_sock_setbufsize(svsk->sk_sock,
NeilBrownc6b0a9f2006-10-06 00:44:05 -0700567 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
568 (serv->sv_nrthreads+3) * serv->sv_max_mesg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569
Tom Tucker02fc6c32007-12-30 21:07:48 -0600570 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
NeilBrown05ed6902007-05-09 02:34:55 -0700571 skb = NULL;
572 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
573 0, 0, MSG_PEEK | MSG_DONTWAIT);
574 if (err >= 0)
575 skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err);
576
577 if (skb == NULL) {
578 if (err != -EAGAIN) {
579 /* possibly an icmp error */
580 dprintk("svc: recvfrom returned error %d\n", -err);
Tom Tucker02fc6c32007-12-30 21:07:48 -0600581 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 }
J. Bruce Fields9f9d2eb2012-08-17 21:35:24 -0400583 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 }
Tom Tucker9dbc2402007-12-30 21:08:12 -0600585 len = svc_addr_len(svc_addr(rqstp));
Tom Tucker9dbc2402007-12-30 21:08:12 -0600586 rqstp->rq_addrlen = len;
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -0700587 if (skb->tstamp.tv64 == 0) {
588 skb->tstamp = ktime_get_real();
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -0800589 /* Don't enable netstamp, sunrpc doesn't
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 need that much accuracy */
591 }
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -0700592 svsk->sk_sk->sk_stamp = skb->tstamp;
Tom Tucker02fc6c32007-12-30 21:07:48 -0600593 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 len = skb->len - sizeof(struct udphdr);
596 rqstp->rq_arg.len = len;
597
Chuck Lever95756482007-02-12 00:53:38 -0800598 rqstp->rq_prot = IPPROTO_UDP;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599
Chuck Lever7702ce42009-07-13 10:54:26 -0400600 if (!svc_udp_get_dest_address(rqstp, cmh)) {
Joe Perchese87cc472012-05-13 21:56:26 +0000601 net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n",
602 cmh->cmsg_level, cmh->cmsg_type);
J. Bruce Fieldsf23abfd2012-08-17 20:32:27 -0400603 goto out_free;
NeilBrown7a37f572007-03-06 01:42:21 -0800604 }
Mi Jinlong849a1cf2011-08-30 17:18:41 +0800605 rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606
607 if (skb_is_nonlinear(skb)) {
608 /* we have to copy */
609 local_bh_disable();
610 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
611 local_bh_enable();
612 /* checksum error */
Eric Dumazet22911fc2012-06-27 00:23:44 +0000613 goto out_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 }
615 local_bh_enable();
Eric Dumazet9d410c72009-10-30 05:03:53 +0000616 skb_free_datagram_locked(svsk->sk_sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 } else {
618 /* we can use it in-place */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600619 rqstp->rq_arg.head[0].iov_base = skb->data +
620 sizeof(struct udphdr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 rqstp->rq_arg.head[0].iov_len = len;
Eric Dumazet22911fc2012-06-27 00:23:44 +0000622 if (skb_checksum_complete(skb))
623 goto out_free;
Tom Tucker5148bf42007-12-30 21:07:25 -0600624 rqstp->rq_xprt_ctxt = skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 }
626
627 rqstp->rq_arg.page_base = 0;
628 if (len <= rqstp->rq_arg.head[0].iov_len) {
629 rqstp->rq_arg.head[0].iov_len = len;
630 rqstp->rq_arg.page_len = 0;
NeilBrown44524352006-10-04 02:15:46 -0700631 rqstp->rq_respages = rqstp->rq_pages+1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 } else {
633 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
NeilBrown44524352006-10-04 02:15:46 -0700634 rqstp->rq_respages = rqstp->rq_pages + 1 +
Ilpo Järvinen172589c2007-08-28 15:50:33 -0700635 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 }
J. Bruce Fieldsafc59402012-12-10 18:01:37 -0500637 rqstp->rq_next_page = rqstp->rq_respages+1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638
639 if (serv->sv_stats)
640 serv->sv_stats->netudpcnt++;
641
642 return len;
J. Bruce Fieldsf23abfd2012-08-17 20:32:27 -0400643out_free:
644 trace_kfree_skb(skb, svc_udp_recvfrom);
645 skb_free_datagram_locked(svsk->sk_sk, skb);
646 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647}
648
649static int
650svc_udp_sendto(struct svc_rqst *rqstp)
651{
652 int error;
653
654 error = svc_sendto(rqstp, &rqstp->rq_res);
655 if (error == -ECONNREFUSED)
656 /* ICMP error on earlier request. */
657 error = svc_sendto(rqstp, &rqstp->rq_res);
658
659 return error;
660}
661
Tom Tuckere831fe62007-12-30 21:07:29 -0600662static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
663{
664}
665
Tom Tucker323bee32007-12-30 21:07:31 -0600666static int svc_udp_has_wspace(struct svc_xprt *xprt)
667{
668 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600669 struct svc_serv *serv = xprt->xpt_server;
Tom Tucker323bee32007-12-30 21:07:31 -0600670 unsigned long required;
671
672 /*
673 * Set the SOCK_NOSPACE flag before checking the available
674 * sock space.
675 */
676 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
Tom Tucker7a90e8c2007-12-30 21:07:55 -0600677 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
Tom Tucker323bee32007-12-30 21:07:31 -0600678 if (required*2 > sock_wspace(svsk->sk_sk))
679 return 0;
680 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
681 return 1;
682}
683
Tom Tucker38a417c2007-12-30 21:07:36 -0600684static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
685{
686 BUG();
687 return NULL;
688}
689
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600690static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
Pavel Emelyanov62832c02010-09-29 16:04:18 +0400691 struct net *net,
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600692 struct sockaddr *sa, int salen,
693 int flags)
694{
Pavel Emelyanov62832c02010-09-29 16:04:18 +0400695 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600696}
697
Tom Tucker360d87382007-12-30 21:07:17 -0600698static struct svc_xprt_ops svc_udp_ops = {
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600699 .xpo_create = svc_udp_create,
Tom Tucker5d137992007-12-30 21:07:23 -0600700 .xpo_recvfrom = svc_udp_recvfrom,
701 .xpo_sendto = svc_udp_sendto,
Tom Tucker5148bf42007-12-30 21:07:25 -0600702 .xpo_release_rqst = svc_release_skb,
Tom Tucker755ccea2007-12-30 21:07:27 -0600703 .xpo_detach = svc_sock_detach,
704 .xpo_free = svc_sock_free,
Tom Tuckere831fe62007-12-30 21:07:29 -0600705 .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
Tom Tucker323bee32007-12-30 21:07:31 -0600706 .xpo_has_wspace = svc_udp_has_wspace,
Tom Tucker38a417c2007-12-30 21:07:36 -0600707 .xpo_accept = svc_udp_accept,
Chuck Lever16e4d932014-05-19 13:40:22 -0400708 .xpo_secure_port = svc_sock_secure_port,
Tom Tucker360d87382007-12-30 21:07:17 -0600709};
710
711static struct svc_xprt_class svc_udp_class = {
712 .xcl_name = "udp",
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600713 .xcl_owner = THIS_MODULE,
Tom Tucker360d87382007-12-30 21:07:17 -0600714 .xcl_ops = &svc_udp_ops,
Tom Tucker49023152007-12-30 21:07:21 -0600715 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
Chuck Lever3c45ddf2014-07-16 15:38:32 -0400716 .xcl_ident = XPRT_TRANSPORT_UDP,
Tom Tucker360d87382007-12-30 21:07:17 -0600717};
718
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600719static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720{
Chuck Lever7702ce42009-07-13 10:54:26 -0400721 int err, level, optname, one = 1;
NeilBrown7a37f572007-03-06 01:42:21 -0800722
Stanislav Kinsburskybd4620d2011-12-06 14:19:10 +0300723 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class,
724 &svsk->sk_xprt, serv);
Tom Tuckerdef13d72007-12-30 21:08:08 -0600725 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 svsk->sk_sk->sk_data_ready = svc_udp_data_ready;
727 svsk->sk_sk->sk_write_space = svc_write_space;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728
729 /* initialise setting must have enough space to
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -0800730 * receive and respond to one request.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 * svc_udp_recvfrom will re-adjust if necessary
732 */
733 svc_sock_setbufsize(svsk->sk_sock,
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600734 3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
735 3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736
Tom Tucker0f0257e2007-12-30 21:08:27 -0600737 /* data might have come in before data_ready set up */
738 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
Tom Tucker02fc6c32007-12-30 21:07:48 -0600739 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
NeilBrown7a37f572007-03-06 01:42:21 -0800740
NeilBrown7a37f572007-03-06 01:42:21 -0800741 /* make sure we get destination address info */
Chuck Lever7702ce42009-07-13 10:54:26 -0400742 switch (svsk->sk_sk->sk_family) {
743 case AF_INET:
744 level = SOL_IP;
745 optname = IP_PKTINFO;
746 break;
747 case AF_INET6:
748 level = SOL_IPV6;
749 optname = IPV6_RECVPKTINFO;
750 break;
751 default:
752 BUG();
753 }
754 err = kernel_setsockopt(svsk->sk_sock, level, optname,
755 (char *)&one, sizeof(one));
756 dprintk("svc: kernel_setsockopt returned %d\n", err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757}
758
759/*
760 * A data_ready event on a listening socket means there's a connection
761 * pending. Do not use state_change as a substitute for it.
762 */
David S. Miller676d2362014-04-11 16:15:36 -0400763static void svc_tcp_listen_data_ready(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764{
Neil Brown939bb7e2005-09-13 01:25:39 -0700765 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
Eric Dumazeteaefd112011-02-18 03:26:36 +0000766 wait_queue_head_t *wq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767
768 dprintk("svc: socket %p TCP (listen) state change %d\n",
Neil Brown939bb7e2005-09-13 01:25:39 -0700769 sk, sk->sk_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770
Neil Brown939bb7e2005-09-13 01:25:39 -0700771 /*
772 * This callback may called twice when a new connection
773 * is established as a child socket inherits everything
774 * from a parent LISTEN socket.
775 * 1) data_ready method of the parent socket will be called
776 * when one of child sockets become ESTABLISHED.
777 * 2) data_ready method of the child socket may be called
778 * when it receives data before the socket is accepted.
779 * In case of 2, we should ignore it silently.
780 */
781 if (sk->sk_state == TCP_LISTEN) {
782 if (svsk) {
Tom Tucker02fc6c32007-12-30 21:07:48 -0600783 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
Tom Tuckerf6150c32007-12-30 21:07:57 -0600784 svc_xprt_enqueue(&svsk->sk_xprt);
Neil Brown939bb7e2005-09-13 01:25:39 -0700785 } else
786 printk("svc: socket %p: no user data\n", sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 }
Neil Brown939bb7e2005-09-13 01:25:39 -0700788
Eric Dumazeteaefd112011-02-18 03:26:36 +0000789 wq = sk_sleep(sk);
790 if (wq && waitqueue_active(wq))
791 wake_up_interruptible_all(wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792}
793
794/*
795 * A state change on a connected socket means it's dying or dead.
796 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600797static void svc_tcp_state_change(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798{
Neil Brown939bb7e2005-09-13 01:25:39 -0700799 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
Eric Dumazeteaefd112011-02-18 03:26:36 +0000800 wait_queue_head_t *wq = sk_sleep(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801
802 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
Neil Brown939bb7e2005-09-13 01:25:39 -0700803 sk, sk->sk_state, sk->sk_user_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804
Neil Brown939bb7e2005-09-13 01:25:39 -0700805 if (!svsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 printk("svc: socket %p: no user data\n", sk);
Neil Brown939bb7e2005-09-13 01:25:39 -0700807 else {
Tom Tucker02fc6c32007-12-30 21:07:48 -0600808 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
Tom Tuckerf6150c32007-12-30 21:07:57 -0600809 svc_xprt_enqueue(&svsk->sk_xprt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 }
Eric Dumazeteaefd112011-02-18 03:26:36 +0000811 if (wq && waitqueue_active(wq))
812 wake_up_interruptible_all(wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813}
814
David S. Miller676d2362014-04-11 16:15:36 -0400815static void svc_tcp_data_ready(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816{
Neil Brown939bb7e2005-09-13 01:25:39 -0700817 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
Eric Dumazeteaefd112011-02-18 03:26:36 +0000818 wait_queue_head_t *wq = sk_sleep(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819
820 dprintk("svc: socket %p TCP data ready (svsk %p)\n",
Neil Brown939bb7e2005-09-13 01:25:39 -0700821 sk, sk->sk_user_data);
822 if (svsk) {
Tom Tucker02fc6c32007-12-30 21:07:48 -0600823 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
Tom Tuckerf6150c32007-12-30 21:07:57 -0600824 svc_xprt_enqueue(&svsk->sk_xprt);
Neil Brown939bb7e2005-09-13 01:25:39 -0700825 }
Eric Dumazeteaefd112011-02-18 03:26:36 +0000826 if (wq && waitqueue_active(wq))
827 wake_up_interruptible(wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828}
829
830/*
831 * Accept a TCP connection
832 */
Tom Tucker38a417c2007-12-30 21:07:36 -0600833static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834{
Tom Tucker38a417c2007-12-30 21:07:36 -0600835 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
akpm@linux-foundation.orgcdd88b92007-02-12 00:53:38 -0800836 struct sockaddr_storage addr;
837 struct sockaddr *sin = (struct sockaddr *) &addr;
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600838 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839 struct socket *sock = svsk->sk_sock;
840 struct socket *newsock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 struct svc_sock *newsvsk;
842 int err, slen;
Pavel Emelyanov5216a8e2008-02-21 10:57:45 +0300843 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844
845 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
846 if (!sock)
Tom Tucker38a417c2007-12-30 21:07:36 -0600847 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848
Tom Tucker02fc6c32007-12-30 21:07:48 -0600849 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
Sridhar Samudralae6242e92006-08-07 20:58:01 -0700850 err = kernel_accept(sock, &newsock, O_NONBLOCK);
851 if (err < 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 if (err == -ENOMEM)
853 printk(KERN_WARNING "%s: no more sockets!\n",
854 serv->sv_name);
Joe Perchese87cc472012-05-13 21:56:26 +0000855 else if (err != -EAGAIN)
856 net_warn_ratelimited("%s: accept failed (err %d)!\n",
857 serv->sv_name, -err);
Tom Tucker38a417c2007-12-30 21:07:36 -0600858 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859 }
Tom Tucker02fc6c32007-12-30 21:07:48 -0600860 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861
akpm@linux-foundation.orgcdd88b92007-02-12 00:53:38 -0800862 err = kernel_getpeername(newsock, sin, &slen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 if (err < 0) {
Joe Perchese87cc472012-05-13 21:56:26 +0000864 net_warn_ratelimited("%s: peername failed (err %d)!\n",
865 serv->sv_name, -err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 goto failed; /* aborted connection or whatever */
867 }
868
869 /* Ideally, we would want to reject connections from unauthorized
Chuck Leverad06e4b2007-02-12 00:53:32 -0800870 * hosts here, but when we get encryption, the IP of the host won't
871 * tell us anything. For now just warn about unpriv connections.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 */
akpm@linux-foundation.orgcdd88b92007-02-12 00:53:38 -0800873 if (!svc_port_is_privileged(sin)) {
Kinglong Meea48fd0f2014-05-29 12:18:52 +0800874 dprintk("%s: connect from unprivileged port: %s\n",
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -0800875 serv->sv_name,
akpm@linux-foundation.orgcdd88b92007-02-12 00:53:38 -0800876 __svc_print_addr(sin, buf, sizeof(buf)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 }
Chuck Leverad06e4b2007-02-12 00:53:32 -0800878 dprintk("%s: connect from %s\n", serv->sv_name,
akpm@linux-foundation.orgcdd88b92007-02-12 00:53:38 -0800879 __svc_print_addr(sin, buf, sizeof(buf)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880
881 /* make sure that a write doesn't block forever when
882 * low on memory
883 */
884 newsock->sk->sk_sndtimeo = HZ*30;
885
J. Bruce Fields72c35372012-08-13 17:46:17 -0400886 newsvsk = svc_setup_socket(serv, newsock,
887 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY));
888 if (IS_ERR(newsvsk))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889 goto failed;
Tom Tucker9dbc2402007-12-30 21:08:12 -0600890 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
Frank van Maarseveena9747692007-07-09 22:21:39 +0200891 err = kernel_getsockname(newsock, sin, &slen);
892 if (unlikely(err < 0)) {
893 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
894 slen = offsetof(struct sockaddr, sa_data);
895 }
Tom Tucker9dbc2402007-12-30 21:08:12 -0600896 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
Chuck Lever067d7812007-02-12 00:53:30 -0800897
NeilBrownef11ce22014-05-12 11:22:47 +1000898 if (sock_is_loopback(newsock->sk))
899 set_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
900 else
901 clear_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
Tom Tuckerf9f3cc42007-12-30 21:07:40 -0600902 if (serv->sv_stats)
903 serv->sv_stats->nettcpconn++;
904
905 return &newsvsk->sk_xprt;
906
907failed:
908 sock_release(newsock);
909 return NULL;
910}
911
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800912static unsigned int svc_tcp_restore_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
913{
914 unsigned int i, len, npages;
915
J. Bruce Fields8af345f2012-12-03 16:45:35 -0500916 if (svsk->sk_datalen == 0)
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800917 return 0;
J. Bruce Fields8af345f2012-12-03 16:45:35 -0500918 len = svsk->sk_datalen;
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800919 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
920 for (i = 0; i < npages; i++) {
921 if (rqstp->rq_pages[i] != NULL)
922 put_page(rqstp->rq_pages[i]);
923 BUG_ON(svsk->sk_pages[i] == NULL);
924 rqstp->rq_pages[i] = svsk->sk_pages[i];
925 svsk->sk_pages[i] = NULL;
926 }
927 rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
928 return len;
929}
930
931static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
932{
933 unsigned int i, len, npages;
934
J. Bruce Fields8af345f2012-12-03 16:45:35 -0500935 if (svsk->sk_datalen == 0)
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800936 return;
J. Bruce Fields8af345f2012-12-03 16:45:35 -0500937 len = svsk->sk_datalen;
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800938 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
939 for (i = 0; i < npages; i++) {
940 svsk->sk_pages[i] = rqstp->rq_pages[i];
941 rqstp->rq_pages[i] = NULL;
942 }
943}
944
945static void svc_tcp_clear_pages(struct svc_sock *svsk)
946{
947 unsigned int i, len, npages;
948
J. Bruce Fields8af345f2012-12-03 16:45:35 -0500949 if (svsk->sk_datalen == 0)
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800950 goto out;
J. Bruce Fields8af345f2012-12-03 16:45:35 -0500951 len = svsk->sk_datalen;
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800952 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
953 for (i = 0; i < npages; i++) {
J. Bruce Fieldscf3aa022013-06-26 11:09:06 -0400954 if (svsk->sk_pages[i] == NULL) {
955 WARN_ON_ONCE(1);
956 continue;
957 }
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800958 put_page(svsk->sk_pages[i]);
959 svsk->sk_pages[i] = NULL;
960 }
961out:
962 svsk->sk_tcplen = 0;
J. Bruce Fields8af345f2012-12-03 16:45:35 -0500963 svsk->sk_datalen = 0;
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800964}
965
Tom Tuckerf9f3cc42007-12-30 21:07:40 -0600966/*
J. Bruce Fieldsad46ccf2012-12-03 16:30:42 -0500967 * Receive fragment record header.
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +0300968 * If we haven't gotten the record length yet, get the next four bytes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 */
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +0300970static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971{
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600972 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
Trond Myklebust5ee78d42009-05-18 17:47:56 -0400973 unsigned int want;
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +0300974 int len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975
Chuck Leverc0401ea2008-04-14 12:27:30 -0400976 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 struct kvec iov;
978
Trond Myklebust5ee78d42009-05-18 17:47:56 -0400979 want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
981 iov.iov_len = want;
982 if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
983 goto error;
984 svsk->sk_tcplen += len;
985
986 if (len < want) {
Chuck Leverc0401ea2008-04-14 12:27:30 -0400987 dprintk("svc: short recvfrom while reading record "
988 "length (%d of %d)\n", len, want);
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800989 return -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 }
991
J. Bruce Fieldscc248d42012-12-03 16:11:13 -0500992 dprintk("svc: TCP record, %d bytes\n", svc_sock_reclen(svsk));
J. Bruce Fields836fbad2012-12-03 15:50:38 -0500993 if (svc_sock_reclen(svsk) + svsk->sk_datalen >
994 serv->sv_max_mesg) {
J. Bruce Fields3a28e332012-12-10 18:15:07 -0500995 net_notice_ratelimited("RPC: fragment too large: %d\n",
996 svc_sock_reclen(svsk));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 goto err_delete;
998 }
999 }
1000
J. Bruce Fieldscc248d42012-12-03 16:11:13 -05001001 return svc_sock_reclen(svsk);
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001002error:
1003 dprintk("RPC: TCP recv_record got %d\n", len);
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +03001004 return len;
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001005err_delete:
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +03001006 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +03001007 return -EAGAIN;
1008}
1009
Trond Myklebust586c52c2009-05-18 17:47:56 -04001010static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
Rahul Iyer4cfc7e62009-09-10 17:32:28 +03001011{
Trond Myklebust586c52c2009-05-18 17:47:56 -04001012 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
Rahul Iyer4cfc7e62009-09-10 17:32:28 +03001013 struct rpc_rqst *req = NULL;
Trond Myklebust586c52c2009-05-18 17:47:56 -04001014 struct kvec *src, *dst;
1015 __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
J. Bruce Fields48e65552011-02-14 14:52:03 -05001016 __be32 xid;
1017 __be32 calldir;
Rahul Iyer4cfc7e62009-09-10 17:32:28 +03001018
Rahul Iyer4cfc7e62009-09-10 17:32:28 +03001019 xid = *p++;
1020 calldir = *p;
1021
Trond Myklebust586c52c2009-05-18 17:47:56 -04001022 if (bc_xprt)
1023 req = xprt_lookup_rqst(bc_xprt, xid);
J. Bruce Fieldsd75faea2010-11-30 19:15:01 -05001024
Trond Myklebust586c52c2009-05-18 17:47:56 -04001025 if (!req) {
1026 printk(KERN_NOTICE
1027 "%s: Got unrecognized reply: "
1028 "calldir 0x%x xpt_bc_xprt %p xid %08x\n",
1029 __func__, ntohl(calldir),
Chuck Lever71efecb2014-08-22 15:10:50 -04001030 bc_xprt, ntohl(xid));
Trond Myklebust586c52c2009-05-18 17:47:56 -04001031 return -EAGAIN;
Rahul Iyer4cfc7e62009-09-10 17:32:28 +03001032 }
Trond Myklebust586c52c2009-05-18 17:47:56 -04001033
1034 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
1035 /*
1036 * XXX!: cheating for now! Only copying HEAD.
1037 * But we know this is good enough for now (in fact, for any
1038 * callback reply in the forseeable future).
1039 */
1040 dst = &req->rq_private_buf.head[0];
1041 src = &rqstp->rq_arg.head[0];
1042 if (dst->iov_len < src->iov_len)
1043 return -EAGAIN; /* whatever; just giving up. */
1044 memcpy(dst->iov_base, src->iov_base, src->iov_len);
J. Bruce Fieldscc248d42012-12-03 16:11:13 -05001045 xprt_complete_rqst(req->rq_task, rqstp->rq_arg.len);
Trond Myklebust586c52c2009-05-18 17:47:56 -04001046 rqstp->rq_arg.len = 0;
1047 return 0;
1048}
1049
1050static int copy_pages_to_kvecs(struct kvec *vec, struct page **pages, int len)
1051{
1052 int i = 0;
1053 int t = 0;
1054
1055 while (t < len) {
1056 vec[i].iov_base = page_address(pages[i]);
1057 vec[i].iov_len = PAGE_SIZE;
1058 i++;
1059 t += PAGE_SIZE;
1060 }
1061 return i;
Rahul Iyer4cfc7e62009-09-10 17:32:28 +03001062}
1063
J. Bruce Fields836fbad2012-12-03 15:50:38 -05001064static void svc_tcp_fragment_received(struct svc_sock *svsk)
1065{
1066 /* If we have more data, signal svc_xprt_enqueue() to try again */
J. Bruce Fields836fbad2012-12-03 15:50:38 -05001067 dprintk("svc: TCP %s record (%d bytes)\n",
1068 svc_sock_final_rec(svsk) ? "final" : "nonfinal",
1069 svc_sock_reclen(svsk));
1070 svsk->sk_tcplen = 0;
1071 svsk->sk_reclen = 0;
1072}
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001073
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +03001074/*
1075 * Receive data from a TCP socket.
1076 */
1077static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
1078{
1079 struct svc_sock *svsk =
1080 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
1081 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1082 int len;
1083 struct kvec *vec;
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001084 unsigned int want, base;
Trond Myklebust586c52c2009-05-18 17:47:56 -04001085 __be32 *p;
1086 __be32 calldir;
Trond Myklebust5ee78d42009-05-18 17:47:56 -04001087 int pnum;
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +03001088
1089 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1090 svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
1091 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
1092 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1093
1094 len = svc_tcp_recv_record(svsk, rqstp);
1095 if (len < 0)
1096 goto error;
1097
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001098 base = svc_tcp_restore_pages(svsk, rqstp);
J. Bruce Fields836fbad2012-12-03 15:50:38 -05001099 want = svc_sock_reclen(svsk) - (svsk->sk_tcplen - sizeof(rpc_fraghdr));
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001100
NeilBrown3cc03b12006-10-04 02:15:47 -07001101 vec = rqstp->rq_vec;
Rahul Iyer4cfc7e62009-09-10 17:32:28 +03001102
Trond Myklebust586c52c2009-05-18 17:47:56 -04001103 pnum = copy_pages_to_kvecs(&vec[0], &rqstp->rq_pages[0],
J. Bruce Fields836fbad2012-12-03 15:50:38 -05001104 svsk->sk_datalen + want);
Rahul Iyer4cfc7e62009-09-10 17:32:28 +03001105
NeilBrown44524352006-10-04 02:15:46 -07001106 rqstp->rq_respages = &rqstp->rq_pages[pnum];
J. Bruce Fieldsafc59402012-12-10 18:01:37 -05001107 rqstp->rq_next_page = rqstp->rq_respages + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108
1109 /* Now receive data */
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001110 len = svc_partial_recvfrom(rqstp, vec, pnum, want, base);
J. Bruce Fields8af345f2012-12-03 16:45:35 -05001111 if (len >= 0) {
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001112 svsk->sk_tcplen += len;
J. Bruce Fields8af345f2012-12-03 16:45:35 -05001113 svsk->sk_datalen += len;
1114 }
J. Bruce Fields836fbad2012-12-03 15:50:38 -05001115 if (len != want || !svc_sock_final_rec(svsk)) {
J. Bruce Fieldsbe1e4442012-08-09 18:12:28 -04001116 svc_tcp_save_pages(svsk, rqstp);
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001117 if (len < 0 && len != -EAGAIN)
J. Bruce Fieldsad46ccf2012-12-03 16:30:42 -05001118 goto err_delete;
J. Bruce Fields836fbad2012-12-03 15:50:38 -05001119 if (len == want)
1120 svc_tcp_fragment_received(svsk);
1121 else
J. Bruce Fields3a28e332012-12-10 18:15:07 -05001122 dprintk("svc: incomplete TCP record (%d of %d)\n",
1123 (int)(svsk->sk_tcplen - sizeof(rpc_fraghdr)),
J. Bruce Fields836fbad2012-12-03 15:50:38 -05001124 svc_sock_reclen(svsk));
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001125 goto err_noclose;
1126 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127
J. Bruce Fields1f691b02013-06-26 10:55:40 -04001128 if (svsk->sk_datalen < 8) {
J. Bruce Fieldscf3aa022013-06-26 11:09:06 -04001129 svsk->sk_datalen = 0;
J. Bruce Fieldsad46ccf2012-12-03 16:30:42 -05001130 goto err_delete; /* client is nuts. */
J. Bruce Fieldscf3aa022013-06-26 11:09:06 -04001131 }
J. Bruce Fieldsad46ccf2012-12-03 16:30:42 -05001132
J. Bruce Fields836fbad2012-12-03 15:50:38 -05001133 rqstp->rq_arg.len = svsk->sk_datalen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134 rqstp->rq_arg.page_base = 0;
Trond Myklebust5ee78d42009-05-18 17:47:56 -04001135 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1136 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137 rqstp->rq_arg.page_len = 0;
Trond Myklebust5ee78d42009-05-18 17:47:56 -04001138 } else
1139 rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140
Tom Tucker5148bf42007-12-30 21:07:25 -06001141 rqstp->rq_xprt_ctxt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142 rqstp->rq_prot = IPPROTO_TCP;
NeilBrownef11ce22014-05-12 11:22:47 +10001143 rqstp->rq_local = !!test_bit(XPT_LOCAL, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144
Trond Myklebust586c52c2009-05-18 17:47:56 -04001145 p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1146 calldir = p[1];
J. Bruce Fields8985ef02011-04-09 10:03:10 -04001147 if (calldir)
Trond Myklebust586c52c2009-05-18 17:47:56 -04001148 len = receive_cb_reply(svsk, rqstp);
Trond Myklebust586c52c2009-05-18 17:47:56 -04001149
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 /* Reset TCP read info */
J. Bruce Fields8af345f2012-12-03 16:45:35 -05001151 svsk->sk_datalen = 0;
J. Bruce Fields836fbad2012-12-03 15:50:38 -05001152 svc_tcp_fragment_received(svsk);
Trond Myklebust0601f792009-05-18 17:47:56 -04001153
J. Bruce Fields8985ef02011-04-09 10:03:10 -04001154 if (len < 0)
1155 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156
Tom Tucker9dbc2402007-12-30 21:08:12 -06001157 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 if (serv->sv_stats)
1159 serv->sv_stats->nettcpcnt++;
1160
Trond Myklebust5ee78d42009-05-18 17:47:56 -04001161 return rqstp->rq_arg.len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +03001163error:
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001164 if (len != -EAGAIN)
J. Bruce Fieldsad46ccf2012-12-03 16:30:42 -05001165 goto err_delete;
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001166 dprintk("RPC: TCP recvfrom got EAGAIN\n");
J. Bruce Fields9f9d2eb2012-08-17 21:35:24 -04001167 return 0;
J. Bruce Fieldsad46ccf2012-12-03 16:30:42 -05001168err_delete:
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001169 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1170 svsk->sk_xprt.xpt_server->sv_name, -len);
1171 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1172err_noclose:
J. Bruce Fields9f9d2eb2012-08-17 21:35:24 -04001173 return 0; /* record not complete */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174}
1175
1176/*
1177 * Send out data on TCP socket.
1178 */
Tom Tucker0f0257e2007-12-30 21:08:27 -06001179static int svc_tcp_sendto(struct svc_rqst *rqstp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180{
1181 struct xdr_buf *xbufp = &rqstp->rq_res;
1182 int sent;
Alexey Dobriyand8ed0292006-09-26 22:29:38 -07001183 __be32 reclen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184
1185 /* Set up the first element of the reply kvec.
1186 * Any other kvecs that may be in use have been taken
1187 * care of by the server implementation itself.
1188 */
1189 reclen = htonl(0x80000000|((xbufp->len ) - 4));
1190 memcpy(xbufp->head[0].iov_base, &reclen, 4);
1191
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192 sent = svc_sendto(rqstp, &rqstp->rq_res);
1193 if (sent != xbufp->len) {
Tom Tucker0f0257e2007-12-30 21:08:27 -06001194 printk(KERN_NOTICE
1195 "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1196 "- shutting down socket\n",
Tom Tucker57b1d3b2007-12-30 21:08:22 -06001197 rqstp->rq_xprt->xpt_server->sv_name,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198 (sent<0)?"got error":"sent only",
1199 sent, xbufp->len);
Tom Tucker57b1d3b2007-12-30 21:08:22 -06001200 set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
Tom Tuckerf6150c32007-12-30 21:07:57 -06001201 svc_xprt_enqueue(rqstp->rq_xprt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202 sent = -EAGAIN;
1203 }
1204 return sent;
1205}
1206
Tom Tuckere831fe62007-12-30 21:07:29 -06001207/*
1208 * Setup response header. TCP has a 4B record length field.
1209 */
1210static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
1211{
1212 struct kvec *resv = &rqstp->rq_res.head[0];
1213
1214 /* tcp needs a space for the record length... */
1215 svc_putnl(resv, 0);
1216}
1217
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001218static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
Pavel Emelyanov62832c02010-09-29 16:04:18 +04001219 struct net *net,
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001220 struct sockaddr *sa, int salen,
1221 int flags)
1222{
Pavel Emelyanov62832c02010-09-29 16:04:18 +04001223 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001224}
1225
Trond Myklebust9e00abc2011-07-13 19:20:49 -04001226#if defined(CONFIG_SUNRPC_BACKCHANNEL)
Andy Adamson1f11a032011-01-06 02:04:26 +00001227static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
1228 struct net *, struct sockaddr *,
1229 int, int);
1230static void svc_bc_sock_free(struct svc_xprt *xprt);
1231
1232static struct svc_xprt *svc_bc_tcp_create(struct svc_serv *serv,
1233 struct net *net,
1234 struct sockaddr *sa, int salen,
1235 int flags)
1236{
1237 return svc_bc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1238}
1239
1240static void svc_bc_tcp_sock_detach(struct svc_xprt *xprt)
1241{
1242}
1243
1244static struct svc_xprt_ops svc_tcp_bc_ops = {
1245 .xpo_create = svc_bc_tcp_create,
1246 .xpo_detach = svc_bc_tcp_sock_detach,
1247 .xpo_free = svc_bc_sock_free,
1248 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
Chuck Lever16e4d932014-05-19 13:40:22 -04001249 .xpo_secure_port = svc_sock_secure_port,
Andy Adamson1f11a032011-01-06 02:04:26 +00001250};
1251
1252static struct svc_xprt_class svc_tcp_bc_class = {
1253 .xcl_name = "tcp-bc",
1254 .xcl_owner = THIS_MODULE,
1255 .xcl_ops = &svc_tcp_bc_ops,
1256 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1257};
Andy Adamson16b2d1e2011-01-06 02:04:27 +00001258
1259static void svc_init_bc_xprt_sock(void)
1260{
1261 svc_reg_xprt_class(&svc_tcp_bc_class);
1262}
1263
1264static void svc_cleanup_bc_xprt_sock(void)
1265{
1266 svc_unreg_xprt_class(&svc_tcp_bc_class);
1267}
Trond Myklebust9e00abc2011-07-13 19:20:49 -04001268#else /* CONFIG_SUNRPC_BACKCHANNEL */
Andy Adamson16b2d1e2011-01-06 02:04:27 +00001269static void svc_init_bc_xprt_sock(void)
1270{
1271}
1272
1273static void svc_cleanup_bc_xprt_sock(void)
1274{
1275}
Trond Myklebust9e00abc2011-07-13 19:20:49 -04001276#endif /* CONFIG_SUNRPC_BACKCHANNEL */
Andy Adamson1f11a032011-01-06 02:04:26 +00001277
Tom Tucker360d87382007-12-30 21:07:17 -06001278static struct svc_xprt_ops svc_tcp_ops = {
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001279 .xpo_create = svc_tcp_create,
Tom Tucker5d137992007-12-30 21:07:23 -06001280 .xpo_recvfrom = svc_tcp_recvfrom,
1281 .xpo_sendto = svc_tcp_sendto,
Tom Tucker5148bf42007-12-30 21:07:25 -06001282 .xpo_release_rqst = svc_release_skb,
Trond Myklebust69b6ba32008-12-23 16:30:11 -05001283 .xpo_detach = svc_tcp_sock_detach,
Tom Tucker755ccea2007-12-30 21:07:27 -06001284 .xpo_free = svc_sock_free,
Tom Tuckere831fe62007-12-30 21:07:29 -06001285 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
Tom Tucker323bee32007-12-30 21:07:31 -06001286 .xpo_has_wspace = svc_tcp_has_wspace,
Tom Tucker38a417c2007-12-30 21:07:36 -06001287 .xpo_accept = svc_tcp_accept,
Chuck Lever16e4d932014-05-19 13:40:22 -04001288 .xpo_secure_port = svc_sock_secure_port,
Trond Myklebust51877682014-07-24 23:59:33 -04001289 .xpo_adjust_wspace = svc_tcp_adjust_wspace,
Tom Tucker360d87382007-12-30 21:07:17 -06001290};
1291
1292static struct svc_xprt_class svc_tcp_class = {
1293 .xcl_name = "tcp",
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001294 .xcl_owner = THIS_MODULE,
Tom Tucker360d87382007-12-30 21:07:17 -06001295 .xcl_ops = &svc_tcp_ops,
Tom Tucker49023152007-12-30 21:07:21 -06001296 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
Chuck Lever3c45ddf2014-07-16 15:38:32 -04001297 .xcl_ident = XPRT_TRANSPORT_TCP,
Tom Tucker360d87382007-12-30 21:07:17 -06001298};
1299
1300void svc_init_xprt_sock(void)
1301{
1302 svc_reg_xprt_class(&svc_tcp_class);
1303 svc_reg_xprt_class(&svc_udp_class);
Andy Adamson16b2d1e2011-01-06 02:04:27 +00001304 svc_init_bc_xprt_sock();
Tom Tucker360d87382007-12-30 21:07:17 -06001305}
1306
1307void svc_cleanup_xprt_sock(void)
1308{
1309 svc_unreg_xprt_class(&svc_tcp_class);
1310 svc_unreg_xprt_class(&svc_udp_class);
Andy Adamson16b2d1e2011-01-06 02:04:27 +00001311 svc_cleanup_bc_xprt_sock();
Tom Tucker360d87382007-12-30 21:07:17 -06001312}
1313
Tom Tuckerbb5cf162007-12-30 21:07:50 -06001314static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315{
1316 struct sock *sk = svsk->sk_sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317
Stanislav Kinsburskybd4620d2011-12-06 14:19:10 +03001318 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1319 &svsk->sk_xprt, serv);
Tom Tuckerdef13d72007-12-30 21:08:08 -06001320 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321 if (sk->sk_state == TCP_LISTEN) {
1322 dprintk("setting up TCP socket for listening\n");
Tom Tucker02fc6c32007-12-30 21:07:48 -06001323 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324 sk->sk_data_ready = svc_tcp_listen_data_ready;
Tom Tucker02fc6c32007-12-30 21:07:48 -06001325 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326 } else {
1327 dprintk("setting up TCP socket for reading\n");
1328 sk->sk_state_change = svc_tcp_state_change;
1329 sk->sk_data_ready = svc_tcp_data_ready;
Trond Myklebust47fcb032009-05-18 17:47:56 -04001330 sk->sk_write_space = svc_tcp_write_space;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331
1332 svsk->sk_reclen = 0;
1333 svsk->sk_tcplen = 0;
J. Bruce Fields8af345f2012-12-03 16:45:35 -05001334 svsk->sk_datalen = 0;
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001335 memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336
Chuck Leverb7872fe2008-04-14 12:27:01 -04001337 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338
Tom Tucker02fc6c32007-12-30 21:07:48 -06001339 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -08001340 if (sk->sk_state != TCP_ESTABLISHED)
Tom Tucker02fc6c32007-12-30 21:07:48 -06001341 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 }
1343}
1344
Tom Tucker0f0257e2007-12-30 21:08:27 -06001345void svc_sock_update_bufs(struct svc_serv *serv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346{
1347 /*
1348 * The number of server threads has changed. Update
1349 * rcvbuf and sndbuf accordingly on all sockets
1350 */
Pavel Emelyanov8f3a6de2010-10-05 23:30:19 +04001351 struct svc_sock *svsk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352
1353 spin_lock_bh(&serv->sv_lock);
Pavel Emelyanov8f3a6de2010-10-05 23:30:19 +04001354 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
Tom Tucker02fc6c32007-12-30 21:07:48 -06001355 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 spin_unlock_bh(&serv->sv_lock);
1357}
Trond Myklebust24c37672008-12-23 16:30:12 -05001358EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359
1360/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 * Initialize socket for RPC use and create svc_sock struct
1362 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1363 */
Chuck Lever6b174332007-02-12 00:53:28 -08001364static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1365 struct socket *sock,
J. Bruce Fields72c35372012-08-13 17:46:17 -04001366 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367{
1368 struct svc_sock *svsk;
1369 struct sock *inet;
Chuck Lever6b174332007-02-12 00:53:28 -08001370 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
J. Bruce Fields72c35372012-08-13 17:46:17 -04001371 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372
1373 dprintk("svc: svc_setup_socket %p\n", sock);
J. Bruce Fields72c35372012-08-13 17:46:17 -04001374 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1375 if (!svsk)
1376 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377
1378 inet = sock->sk;
1379
1380 /* Register socket with portmapper */
J. Bruce Fields72c35372012-08-13 17:46:17 -04001381 if (pmap_register)
1382 err = svc_register(serv, sock_net(sock->sk), inet->sk_family,
Stanislav Kinsbursky5247fab2012-01-13 14:02:48 +04001383 inet->sk_protocol,
Eric Dumazetc720c7e2009-10-15 06:30:45 +00001384 ntohs(inet_sk(inet)->inet_sport));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385
J. Bruce Fields72c35372012-08-13 17:46:17 -04001386 if (err < 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 kfree(svsk);
J. Bruce Fields72c35372012-08-13 17:46:17 -04001388 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389 }
1390
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391 inet->sk_user_data = svsk;
1392 svsk->sk_sock = sock;
1393 svsk->sk_sk = inet;
1394 svsk->sk_ostate = inet->sk_state_change;
1395 svsk->sk_odata = inet->sk_data_ready;
1396 svsk->sk_owspace = inet->sk_write_space;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397
1398 /* Initialize the socket */
1399 if (sock->type == SOCK_DGRAM)
Tom Tuckerbb5cf162007-12-30 21:07:50 -06001400 svc_udp_init(svsk, serv);
Olga Kornievskaia96604392008-10-21 14:13:47 -04001401 else {
1402 /* initialise setting must have enough space to
1403 * receive and respond to one request.
1404 */
1405 svc_sock_setbufsize(svsk->sk_sock, 4 * serv->sv_max_mesg,
1406 4 * serv->sv_max_mesg);
Tom Tuckerbb5cf162007-12-30 21:07:50 -06001407 svc_tcp_init(svsk, serv);
Olga Kornievskaia96604392008-10-21 14:13:47 -04001408 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1411 svsk, svsk->sk_sk);
1412
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 return svsk;
1414}
1415
Stanislav Kinsbursky30646392014-02-26 16:50:01 +03001416bool svc_alien_sock(struct net *net, int fd)
1417{
1418 int err;
1419 struct socket *sock = sockfd_lookup(fd, &err);
1420 bool ret = false;
1421
1422 if (!sock)
1423 goto out;
1424 if (sock_net(sock->sk) != net)
1425 ret = true;
1426 sockfd_put(sock);
1427out:
1428 return ret;
1429}
1430EXPORT_SYMBOL_GPL(svc_alien_sock);
1431
Chuck Leverbfba9ab2009-04-23 19:32:33 -04001432/**
1433 * svc_addsock - add a listener socket to an RPC service
1434 * @serv: pointer to RPC service to which to add a new listener
1435 * @fd: file descriptor of the new listener
1436 * @name_return: pointer to buffer to fill in with name of listener
1437 * @len: size of the buffer
1438 *
1439 * Fills in socket name and returns positive length of name if successful.
1440 * Name is terminated with '\n'. On error, returns a negative errno
1441 * value.
1442 */
1443int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1444 const size_t len)
NeilBrownb41b66d2006-10-02 02:17:48 -07001445{
1446 int err = 0;
1447 struct socket *so = sockfd_lookup(fd, &err);
1448 struct svc_sock *svsk = NULL;
J. Bruce Fieldsa8e10072012-08-13 18:01:03 -04001449 struct sockaddr_storage addr;
1450 struct sockaddr *sin = (struct sockaddr *)&addr;
1451 int salen;
NeilBrownb41b66d2006-10-02 02:17:48 -07001452
1453 if (!so)
1454 return err;
J. Bruce Fieldsa8e10072012-08-13 18:01:03 -04001455 err = -EAFNOSUPPORT;
Aime Le Rouzic205ba422010-01-26 14:03:56 -05001456 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
J. Bruce Fieldsa8e10072012-08-13 18:01:03 -04001457 goto out;
1458 err = -EPROTONOSUPPORT;
1459 if (so->sk->sk_protocol != IPPROTO_TCP &&
NeilBrownb41b66d2006-10-02 02:17:48 -07001460 so->sk->sk_protocol != IPPROTO_UDP)
J. Bruce Fieldsa8e10072012-08-13 18:01:03 -04001461 goto out;
1462 err = -EISCONN;
1463 if (so->state > SS_UNCONNECTED)
1464 goto out;
1465 err = -ENOENT;
1466 if (!try_module_get(THIS_MODULE))
1467 goto out;
1468 svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS);
1469 if (IS_ERR(svsk)) {
1470 module_put(THIS_MODULE);
1471 err = PTR_ERR(svsk);
1472 goto out;
NeilBrownb41b66d2006-10-02 02:17:48 -07001473 }
J. Bruce Fieldsa8e10072012-08-13 18:01:03 -04001474 if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0)
1475 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
J. Bruce Fields39b55302012-08-14 15:50:34 -04001476 svc_add_new_perm_xprt(serv, &svsk->sk_xprt);
Chuck Levere7942b92009-04-23 19:32:48 -04001477 return svc_one_sock_name(svsk, name_return, len);
J. Bruce Fieldsa8e10072012-08-13 18:01:03 -04001478out:
1479 sockfd_put(so);
1480 return err;
NeilBrownb41b66d2006-10-02 02:17:48 -07001481}
1482EXPORT_SYMBOL_GPL(svc_addsock);
1483
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484/*
1485 * Create socket for RPC service.
1486 */
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001487static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1488 int protocol,
Pavel Emelyanov62832c02010-09-29 16:04:18 +04001489 struct net *net,
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001490 struct sockaddr *sin, int len,
1491 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492{
1493 struct svc_sock *svsk;
1494 struct socket *sock;
1495 int error;
1496 int type;
Tom Tucker9dbc2402007-12-30 21:08:12 -06001497 struct sockaddr_storage addr;
1498 struct sockaddr *newsin = (struct sockaddr *)&addr;
1499 int newlen;
Trond Myklebustc69da772009-03-30 18:59:17 -04001500 int family;
1501 int val;
Pavel Emelyanov5216a8e2008-02-21 10:57:45 +03001502 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503
Chuck Leverad06e4b2007-02-12 00:53:32 -08001504 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1505 serv->sv_program->pg_name, protocol,
Chuck Lever77f1f672007-02-12 00:53:39 -08001506 __svc_print_addr(sin, buf, sizeof(buf)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507
1508 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1509 printk(KERN_WARNING "svc: only UDP and TCP "
1510 "sockets supported\n");
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001511 return ERR_PTR(-EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513
Trond Myklebustc69da772009-03-30 18:59:17 -04001514 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1515 switch (sin->sa_family) {
1516 case AF_INET6:
1517 family = PF_INET6;
1518 break;
1519 case AF_INET:
1520 family = PF_INET;
1521 break;
1522 default:
1523 return ERR_PTR(-EINVAL);
1524 }
1525
Pavel Emelyanov14ec63c2010-09-29 16:06:57 +04001526 error = __sock_create(net, family, type, protocol, &sock, 1);
Chuck Lever77f1f672007-02-12 00:53:39 -08001527 if (error < 0)
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001528 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529
Peter Zijlstraed075362006-12-06 20:35:24 -08001530 svc_reclassify_socket(sock);
1531
Trond Myklebustc69da772009-03-30 18:59:17 -04001532 /*
1533 * If this is an PF_INET6 listener, we want to avoid
1534 * getting requests from IPv4 remotes. Those should
1535 * be shunted to a PF_INET listener via rpcbind.
1536 */
1537 val = 1;
1538 if (family == PF_INET6)
1539 kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
1540 (char *)&val, sizeof(val));
1541
Eric Sesterhenn18114742006-09-28 14:37:07 -07001542 if (type == SOCK_STREAM)
Pavel Emelyanov4a17fd52012-04-19 03:39:36 +00001543 sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */
Chuck Lever77f1f672007-02-12 00:53:39 -08001544 error = kernel_bind(sock, sin, len);
Eric Sesterhenn18114742006-09-28 14:37:07 -07001545 if (error < 0)
1546 goto bummer;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547
Tom Tucker9dbc2402007-12-30 21:08:12 -06001548 newlen = len;
1549 error = kernel_getsockname(sock, newsin, &newlen);
1550 if (error < 0)
1551 goto bummer;
1552
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553 if (protocol == IPPROTO_TCP) {
Sridhar Samudralae6242e92006-08-07 20:58:01 -07001554 if ((error = kernel_listen(sock, 64)) < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555 goto bummer;
1556 }
1557
J. Bruce Fields72c35372012-08-13 17:46:17 -04001558 svsk = svc_setup_socket(serv, sock, flags);
J. Bruce Fieldsa8e10072012-08-13 18:01:03 -04001559 if (IS_ERR(svsk)) {
1560 error = PTR_ERR(svsk);
1561 goto bummer;
NeilBrowne79eff12007-02-12 00:53:30 -08001562 }
J. Bruce Fieldsa8e10072012-08-13 18:01:03 -04001563 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1564 return (struct svc_xprt *)svsk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565bummer:
1566 dprintk("svc: svc_create_socket error = %d\n", -error);
1567 sock_release(sock);
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001568 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569}
1570
1571/*
Tom Tucker755ccea2007-12-30 21:07:27 -06001572 * Detach the svc_sock from the socket so that no
1573 * more callbacks occur.
1574 */
1575static void svc_sock_detach(struct svc_xprt *xprt)
1576{
1577 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1578 struct sock *sk = svsk->sk_sk;
Eric Dumazeteaefd112011-02-18 03:26:36 +00001579 wait_queue_head_t *wq;
Tom Tucker755ccea2007-12-30 21:07:27 -06001580
1581 dprintk("svc: svc_sock_detach(%p)\n", svsk);
1582
1583 /* put back the old socket callbacks */
1584 sk->sk_state_change = svsk->sk_ostate;
1585 sk->sk_data_ready = svsk->sk_odata;
1586 sk->sk_write_space = svsk->sk_owspace;
Trond Myklebust69b6ba32008-12-23 16:30:11 -05001587
Eric Dumazeteaefd112011-02-18 03:26:36 +00001588 wq = sk_sleep(sk);
1589 if (wq && waitqueue_active(wq))
1590 wake_up_interruptible(wq);
Trond Myklebust69b6ba32008-12-23 16:30:11 -05001591}
1592
1593/*
1594 * Disconnect the socket, and reset the callbacks
1595 */
1596static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1597{
1598 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1599
1600 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
1601
1602 svc_sock_detach(xprt);
1603
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001604 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1605 svc_tcp_clear_pages(svsk);
Trond Myklebust69b6ba32008-12-23 16:30:11 -05001606 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001607 }
Tom Tucker755ccea2007-12-30 21:07:27 -06001608}
1609
1610/*
1611 * Free the svc_sock's socket resources and the svc_sock itself.
1612 */
1613static void svc_sock_free(struct svc_xprt *xprt)
1614{
1615 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1616 dprintk("svc: svc_sock_free(%p)\n", svsk);
1617
Tom Tucker755ccea2007-12-30 21:07:27 -06001618 if (svsk->sk_sock->file)
1619 sockfd_put(svsk->sk_sock);
1620 else
1621 sock_release(svsk->sk_sock);
1622 kfree(svsk);
1623}
Benny Halevy7652e5a2009-04-01 09:23:09 -04001624
Trond Myklebust9e00abc2011-07-13 19:20:49 -04001625#if defined(CONFIG_SUNRPC_BACKCHANNEL)
Benny Halevy7652e5a2009-04-01 09:23:09 -04001626/*
Andy Adamson1f11a032011-01-06 02:04:26 +00001627 * Create a back channel svc_xprt which shares the fore channel socket.
Benny Halevy7652e5a2009-04-01 09:23:09 -04001628 */
Andy Adamson1f11a032011-01-06 02:04:26 +00001629static struct svc_xprt *svc_bc_create_socket(struct svc_serv *serv,
1630 int protocol,
1631 struct net *net,
1632 struct sockaddr *sin, int len,
1633 int flags)
Benny Halevy7652e5a2009-04-01 09:23:09 -04001634{
1635 struct svc_sock *svsk;
Andy Adamson1f11a032011-01-06 02:04:26 +00001636 struct svc_xprt *xprt;
Benny Halevy7652e5a2009-04-01 09:23:09 -04001637
Andy Adamson1f11a032011-01-06 02:04:26 +00001638 if (protocol != IPPROTO_TCP) {
1639 printk(KERN_WARNING "svc: only TCP sockets"
1640 " supported on shared back channel\n");
1641 return ERR_PTR(-EINVAL);
1642 }
1643
Benny Halevy7652e5a2009-04-01 09:23:09 -04001644 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1645 if (!svsk)
Andy Adamson1f11a032011-01-06 02:04:26 +00001646 return ERR_PTR(-ENOMEM);
Benny Halevy7652e5a2009-04-01 09:23:09 -04001647
1648 xprt = &svsk->sk_xprt;
Stanislav Kinsburskybd4620d2011-12-06 14:19:10 +03001649 svc_xprt_init(net, &svc_tcp_bc_class, xprt, serv);
Andy Adamson1f11a032011-01-06 02:04:26 +00001650
Andy Adamson4a19de02011-01-06 02:04:35 +00001651 serv->sv_bc_xprt = xprt;
Andy Adamson1f11a032011-01-06 02:04:26 +00001652
Benny Halevy7652e5a2009-04-01 09:23:09 -04001653 return xprt;
1654}
Benny Halevy7652e5a2009-04-01 09:23:09 -04001655
1656/*
Andy Adamson1f11a032011-01-06 02:04:26 +00001657 * Free a back channel svc_sock.
Benny Halevy7652e5a2009-04-01 09:23:09 -04001658 */
Andy Adamson1f11a032011-01-06 02:04:26 +00001659static void svc_bc_sock_free(struct svc_xprt *xprt)
Benny Halevy7652e5a2009-04-01 09:23:09 -04001660{
Andy Adamson778be232011-01-25 15:38:01 +00001661 if (xprt)
Benny Halevy7652e5a2009-04-01 09:23:09 -04001662 kfree(container_of(xprt, struct svc_sock, sk_xprt));
1663}
Trond Myklebust9e00abc2011-07-13 19:20:49 -04001664#endif /* CONFIG_SUNRPC_BACKCHANNEL */