blob: 0a148c9d2a5ce7b3915df8c10f5e5f0913b99c25 [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;
Chuck Lever58de2f82009-04-23 19:32:55 -0400294 case PF_INET6:
295 len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n",
Chuck Lever017cb472009-04-23 19:33:03 -0400296 proto_name,
297 &inet6_sk(sk)->rcv_saddr,
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000298 inet_sk(sk)->inet_num);
NeilBrown80212d52006-10-02 02:17:47 -0700299 break;
300 default:
Chuck Levere7942b92009-04-23 19:32:48 -0400301 len = snprintf(buf, remaining, "*unknown-%d*\n",
Chuck Lever017cb472009-04-23 19:33:03 -0400302 sk->sk_family);
NeilBrown80212d52006-10-02 02:17:47 -0700303 }
Chuck Levere7942b92009-04-23 19:32:48 -0400304
305 if (len >= remaining) {
306 *buf = '\0';
307 return -ENAMETOOLONG;
NeilBrown80212d52006-10-02 02:17:47 -0700308 }
309 return len;
310}
311
NeilBrown80212d52006-10-02 02:17:47 -0700312/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 * Check input queue length
314 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600315static int svc_recv_available(struct svc_sock *svsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 struct socket *sock = svsk->sk_sock;
318 int avail, err;
319
Sridhar Samudralae6242e92006-08-07 20:58:01 -0700320 err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321
322 return (err >= 0)? avail : err;
323}
324
325/*
326 * Generic recvfrom routine.
327 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600328static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr,
329 int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330{
Tom Tucker57b1d3b2007-12-30 21:08:22 -0600331 struct svc_sock *svsk =
332 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
Chuck Lever1ba95102007-02-12 00:53:31 -0800333 struct msghdr msg = {
334 .msg_flags = MSG_DONTWAIT,
335 };
336 int len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337
Tom Tucker260c1d12007-12-30 21:08:29 -0600338 rqstp->rq_xprt_hlen = 0;
339
Chuck Lever1ba95102007-02-12 00:53:31 -0800340 len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen,
341 msg.msg_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
Chuck Lever1ba95102007-02-12 00:53:31 -0800344 svsk, iov[0].iov_base, iov[0].iov_len, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 return len;
346}
347
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800348static int svc_partial_recvfrom(struct svc_rqst *rqstp,
349 struct kvec *iov, int nr,
350 int buflen, unsigned int base)
351{
352 size_t save_iovlen;
Trond Myklebust09acfea2012-03-11 15:22:54 -0400353 void *save_iovbase;
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800354 unsigned int i;
355 int ret;
356
357 if (base == 0)
358 return svc_recvfrom(rqstp, iov, nr, buflen);
359
360 for (i = 0; i < nr; i++) {
361 if (iov[i].iov_len > base)
362 break;
363 base -= iov[i].iov_len;
364 }
365 save_iovlen = iov[i].iov_len;
366 save_iovbase = iov[i].iov_base;
367 iov[i].iov_len -= base;
368 iov[i].iov_base += base;
369 ret = svc_recvfrom(rqstp, &iov[i], nr - i, buflen);
370 iov[i].iov_len = save_iovlen;
371 iov[i].iov_base = save_iovbase;
372 return ret;
373}
374
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375/*
376 * Set socket snd and rcv buffer lengths
377 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600378static void svc_sock_setbufsize(struct socket *sock, unsigned int snd,
379 unsigned int rcv)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380{
381#if 0
382 mm_segment_t oldfs;
383 oldfs = get_fs(); set_fs(KERNEL_DS);
384 sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
385 (char*)&snd, sizeof(snd));
386 sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
387 (char*)&rcv, sizeof(rcv));
388#else
389 /* sock_setsockopt limits use to sysctl_?mem_max,
390 * which isn't acceptable. Until that is made conditional
391 * on not having CAP_SYS_RESOURCE or similar, we go direct...
392 * DaveM said I could!
393 */
394 lock_sock(sock->sk);
395 sock->sk->sk_sndbuf = snd * 2;
396 sock->sk->sk_rcvbuf = rcv * 2;
Trond Myklebust47fcb032009-05-18 17:47:56 -0400397 sock->sk->sk_write_space(sock->sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398 release_sock(sock->sk);
399#endif
400}
401/*
402 * INET callback when data has been received on the socket.
403 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600404static void svc_udp_data_ready(struct sock *sk, int count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405{
Neil Brown939bb7e2005-09-13 01:25:39 -0700406 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
Eric Dumazeteaefd112011-02-18 03:26:36 +0000407 wait_queue_head_t *wq = sk_sleep(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408
Neil Brown939bb7e2005-09-13 01:25:39 -0700409 if (svsk) {
410 dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n",
Tom Tucker02fc6c32007-12-30 21:07:48 -0600411 svsk, sk, count,
412 test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
413 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
Tom Tuckerf6150c32007-12-30 21:07:57 -0600414 svc_xprt_enqueue(&svsk->sk_xprt);
Neil Brown939bb7e2005-09-13 01:25:39 -0700415 }
Eric Dumazeteaefd112011-02-18 03:26:36 +0000416 if (wq && waitqueue_active(wq))
417 wake_up_interruptible(wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418}
419
420/*
421 * INET callback when space is newly available on the socket.
422 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600423static void svc_write_space(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424{
425 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
Eric Dumazeteaefd112011-02-18 03:26:36 +0000426 wait_queue_head_t *wq = sk_sleep(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427
428 if (svsk) {
429 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
Tom Tucker02fc6c32007-12-30 21:07:48 -0600430 svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
Tom Tuckerf6150c32007-12-30 21:07:57 -0600431 svc_xprt_enqueue(&svsk->sk_xprt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 }
433
Eric Dumazeteaefd112011-02-18 03:26:36 +0000434 if (wq && waitqueue_active(wq)) {
Neil Brown939bb7e2005-09-13 01:25:39 -0700435 dprintk("RPC svc_write_space: someone sleeping on %p\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 svsk);
Eric Dumazeteaefd112011-02-18 03:26:36 +0000437 wake_up_interruptible(wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 }
439}
440
Trond Myklebust47fcb032009-05-18 17:47:56 -0400441static void svc_tcp_write_space(struct sock *sk)
442{
443 struct socket *sock = sk->sk_socket;
444
445 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && sock)
446 clear_bit(SOCK_NOSPACE, &sock->flags);
447 svc_write_space(sk);
448}
449
Tom Tucker9dbc2402007-12-30 21:08:12 -0600450/*
Chuck Lever7702ce42009-07-13 10:54:26 -0400451 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
452 */
453static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
454 struct cmsghdr *cmh)
455{
456 struct in_pktinfo *pki = CMSG_DATA(cmh);
Mi Jinlong849a1cf2011-08-30 17:18:41 +0800457 struct sockaddr_in *daddr = svc_daddr_in(rqstp);
458
Chuck Lever7702ce42009-07-13 10:54:26 -0400459 if (cmh->cmsg_type != IP_PKTINFO)
460 return 0;
Mi Jinlong849a1cf2011-08-30 17:18:41 +0800461
462 daddr->sin_family = AF_INET;
463 daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr;
Chuck Lever7702ce42009-07-13 10:54:26 -0400464 return 1;
465}
466
467/*
468 * See net/ipv6/datagram.c : datagram_recv_ctl
469 */
470static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
471 struct cmsghdr *cmh)
472{
473 struct in6_pktinfo *pki = CMSG_DATA(cmh);
Mi Jinlong849a1cf2011-08-30 17:18:41 +0800474 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
475
Chuck Lever7702ce42009-07-13 10:54:26 -0400476 if (cmh->cmsg_type != IPV6_PKTINFO)
477 return 0;
Mi Jinlong849a1cf2011-08-30 17:18:41 +0800478
479 daddr->sin6_family = AF_INET6;
Alexey Dobriyan4e3fd7a2011-11-21 03:39:03 +0000480 daddr->sin6_addr = pki->ipi6_addr;
Mi Jinlong849a1cf2011-08-30 17:18:41 +0800481 daddr->sin6_scope_id = pki->ipi6_ifindex;
Chuck Lever7702ce42009-07-13 10:54:26 -0400482 return 1;
483}
484
485/*
Tom Tucker9dbc2402007-12-30 21:08:12 -0600486 * Copy the UDP datagram's destination address to the rqstp structure.
487 * The 'destination' address in this case is the address to which the
488 * peer sent the datagram, i.e. our local address. For multihomed
489 * hosts, this can change from msg to msg. Note that only the IP
490 * address changes, the port number should remain the same.
491 */
Chuck Lever7702ce42009-07-13 10:54:26 -0400492static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
493 struct cmsghdr *cmh)
Chuck Lever95756482007-02-12 00:53:38 -0800494{
Chuck Lever7702ce42009-07-13 10:54:26 -0400495 switch (cmh->cmsg_level) {
496 case SOL_IP:
497 return svc_udp_get_dest_address4(rqstp, cmh);
498 case SOL_IPV6:
499 return svc_udp_get_dest_address6(rqstp, cmh);
Chuck Lever95756482007-02-12 00:53:38 -0800500 }
Chuck Lever7702ce42009-07-13 10:54:26 -0400501
502 return 0;
Chuck Lever95756482007-02-12 00:53:38 -0800503}
504
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505/*
506 * Receive a datagram from a UDP socket.
507 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600508static int svc_udp_recvfrom(struct svc_rqst *rqstp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509{
Tom Tucker57b1d3b2007-12-30 21:08:22 -0600510 struct svc_sock *svsk =
511 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600512 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 struct sk_buff *skb;
David S. Millerbc375ea2007-04-12 13:35:59 -0700514 union {
515 struct cmsghdr hdr;
516 long all[SVC_PKTINFO_SPACE / sizeof(long)];
517 } buffer;
518 struct cmsghdr *cmh = &buffer.hdr;
NeilBrown7a37f572007-03-06 01:42:21 -0800519 struct msghdr msg = {
520 .msg_name = svc_addr(rqstp),
521 .msg_control = cmh,
522 .msg_controllen = sizeof(buffer),
523 .msg_flags = MSG_DONTWAIT,
524 };
Chuck Leverabc5c442009-04-23 19:31:25 -0400525 size_t len;
526 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527
Tom Tucker02fc6c32007-12-30 21:07:48 -0600528 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 /* udp sockets need large rcvbuf as all pending
530 * requests are still in that buffer. sndbuf must
531 * also be large enough that there is enough space
Greg Banks3262c812006-10-02 02:17:58 -0700532 * for one reply per thread. We count all threads
533 * rather than threads in a particular pool, which
534 * provides an upper bound on the number of threads
535 * which will access the socket.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 */
537 svc_sock_setbufsize(svsk->sk_sock,
NeilBrownc6b0a9f2006-10-06 00:44:05 -0700538 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
539 (serv->sv_nrthreads+3) * serv->sv_max_mesg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540
Tom Tucker02fc6c32007-12-30 21:07:48 -0600541 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
NeilBrown05ed6902007-05-09 02:34:55 -0700542 skb = NULL;
543 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
544 0, 0, MSG_PEEK | MSG_DONTWAIT);
545 if (err >= 0)
546 skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err);
547
548 if (skb == NULL) {
549 if (err != -EAGAIN) {
550 /* possibly an icmp error */
551 dprintk("svc: recvfrom returned error %d\n", -err);
Tom Tucker02fc6c32007-12-30 21:07:48 -0600552 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 }
J. Bruce Fields9f9d2eb2012-08-17 21:35:24 -0400554 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 }
Tom Tucker9dbc2402007-12-30 21:08:12 -0600556 len = svc_addr_len(svc_addr(rqstp));
Tom Tucker9dbc2402007-12-30 21:08:12 -0600557 rqstp->rq_addrlen = len;
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -0700558 if (skb->tstamp.tv64 == 0) {
559 skb->tstamp = ktime_get_real();
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -0800560 /* Don't enable netstamp, sunrpc doesn't
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561 need that much accuracy */
562 }
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -0700563 svsk->sk_sk->sk_stamp = skb->tstamp;
Tom Tucker02fc6c32007-12-30 21:07:48 -0600564 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 len = skb->len - sizeof(struct udphdr);
567 rqstp->rq_arg.len = len;
568
Chuck Lever95756482007-02-12 00:53:38 -0800569 rqstp->rq_prot = IPPROTO_UDP;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570
Chuck Lever7702ce42009-07-13 10:54:26 -0400571 if (!svc_udp_get_dest_address(rqstp, cmh)) {
Joe Perchese87cc472012-05-13 21:56:26 +0000572 net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n",
573 cmh->cmsg_level, cmh->cmsg_type);
J. Bruce Fieldsf23abfd2012-08-17 20:32:27 -0400574 goto out_free;
NeilBrown7a37f572007-03-06 01:42:21 -0800575 }
Mi Jinlong849a1cf2011-08-30 17:18:41 +0800576 rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577
578 if (skb_is_nonlinear(skb)) {
579 /* we have to copy */
580 local_bh_disable();
581 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
582 local_bh_enable();
583 /* checksum error */
Eric Dumazet22911fc2012-06-27 00:23:44 +0000584 goto out_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 }
586 local_bh_enable();
Eric Dumazet9d410c72009-10-30 05:03:53 +0000587 skb_free_datagram_locked(svsk->sk_sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 } else {
589 /* we can use it in-place */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600590 rqstp->rq_arg.head[0].iov_base = skb->data +
591 sizeof(struct udphdr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 rqstp->rq_arg.head[0].iov_len = len;
Eric Dumazet22911fc2012-06-27 00:23:44 +0000593 if (skb_checksum_complete(skb))
594 goto out_free;
Tom Tucker5148bf42007-12-30 21:07:25 -0600595 rqstp->rq_xprt_ctxt = skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 }
597
598 rqstp->rq_arg.page_base = 0;
599 if (len <= rqstp->rq_arg.head[0].iov_len) {
600 rqstp->rq_arg.head[0].iov_len = len;
601 rqstp->rq_arg.page_len = 0;
NeilBrown44524352006-10-04 02:15:46 -0700602 rqstp->rq_respages = rqstp->rq_pages+1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 } else {
604 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
NeilBrown44524352006-10-04 02:15:46 -0700605 rqstp->rq_respages = rqstp->rq_pages + 1 +
Ilpo Järvinen172589c2007-08-28 15:50:33 -0700606 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 }
J. Bruce Fieldsafc59402012-12-10 18:01:37 -0500608 rqstp->rq_next_page = rqstp->rq_respages+1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609
610 if (serv->sv_stats)
611 serv->sv_stats->netudpcnt++;
612
613 return len;
J. Bruce Fieldsf23abfd2012-08-17 20:32:27 -0400614out_free:
615 trace_kfree_skb(skb, svc_udp_recvfrom);
616 skb_free_datagram_locked(svsk->sk_sk, skb);
617 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618}
619
620static int
621svc_udp_sendto(struct svc_rqst *rqstp)
622{
623 int error;
624
625 error = svc_sendto(rqstp, &rqstp->rq_res);
626 if (error == -ECONNREFUSED)
627 /* ICMP error on earlier request. */
628 error = svc_sendto(rqstp, &rqstp->rq_res);
629
630 return error;
631}
632
Tom Tuckere831fe62007-12-30 21:07:29 -0600633static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
634{
635}
636
Tom Tucker323bee32007-12-30 21:07:31 -0600637static int svc_udp_has_wspace(struct svc_xprt *xprt)
638{
639 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600640 struct svc_serv *serv = xprt->xpt_server;
Tom Tucker323bee32007-12-30 21:07:31 -0600641 unsigned long required;
642
643 /*
644 * Set the SOCK_NOSPACE flag before checking the available
645 * sock space.
646 */
647 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
Tom Tucker7a90e8c2007-12-30 21:07:55 -0600648 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
Tom Tucker323bee32007-12-30 21:07:31 -0600649 if (required*2 > sock_wspace(svsk->sk_sk))
650 return 0;
651 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
652 return 1;
653}
654
Tom Tucker38a417c2007-12-30 21:07:36 -0600655static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
656{
657 BUG();
658 return NULL;
659}
660
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600661static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
Pavel Emelyanov62832c02010-09-29 16:04:18 +0400662 struct net *net,
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600663 struct sockaddr *sa, int salen,
664 int flags)
665{
Pavel Emelyanov62832c02010-09-29 16:04:18 +0400666 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600667}
668
Tom Tucker360d87382007-12-30 21:07:17 -0600669static struct svc_xprt_ops svc_udp_ops = {
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600670 .xpo_create = svc_udp_create,
Tom Tucker5d137992007-12-30 21:07:23 -0600671 .xpo_recvfrom = svc_udp_recvfrom,
672 .xpo_sendto = svc_udp_sendto,
Tom Tucker5148bf42007-12-30 21:07:25 -0600673 .xpo_release_rqst = svc_release_skb,
Tom Tucker755ccea2007-12-30 21:07:27 -0600674 .xpo_detach = svc_sock_detach,
675 .xpo_free = svc_sock_free,
Tom Tuckere831fe62007-12-30 21:07:29 -0600676 .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
Tom Tucker323bee32007-12-30 21:07:31 -0600677 .xpo_has_wspace = svc_udp_has_wspace,
Tom Tucker38a417c2007-12-30 21:07:36 -0600678 .xpo_accept = svc_udp_accept,
Tom Tucker360d87382007-12-30 21:07:17 -0600679};
680
681static struct svc_xprt_class svc_udp_class = {
682 .xcl_name = "udp",
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600683 .xcl_owner = THIS_MODULE,
Tom Tucker360d87382007-12-30 21:07:17 -0600684 .xcl_ops = &svc_udp_ops,
Tom Tucker49023152007-12-30 21:07:21 -0600685 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
Tom Tucker360d87382007-12-30 21:07:17 -0600686};
687
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600688static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689{
Chuck Lever7702ce42009-07-13 10:54:26 -0400690 int err, level, optname, one = 1;
NeilBrown7a37f572007-03-06 01:42:21 -0800691
Stanislav Kinsburskybd4620d2011-12-06 14:19:10 +0300692 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class,
693 &svsk->sk_xprt, serv);
Tom Tuckerdef13d72007-12-30 21:08:08 -0600694 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 svsk->sk_sk->sk_data_ready = svc_udp_data_ready;
696 svsk->sk_sk->sk_write_space = svc_write_space;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697
698 /* initialise setting must have enough space to
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -0800699 * receive and respond to one request.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 * svc_udp_recvfrom will re-adjust if necessary
701 */
702 svc_sock_setbufsize(svsk->sk_sock,
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600703 3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
704 3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705
Tom Tucker0f0257e2007-12-30 21:08:27 -0600706 /* data might have come in before data_ready set up */
707 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
Tom Tucker02fc6c32007-12-30 21:07:48 -0600708 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
NeilBrown7a37f572007-03-06 01:42:21 -0800709
NeilBrown7a37f572007-03-06 01:42:21 -0800710 /* make sure we get destination address info */
Chuck Lever7702ce42009-07-13 10:54:26 -0400711 switch (svsk->sk_sk->sk_family) {
712 case AF_INET:
713 level = SOL_IP;
714 optname = IP_PKTINFO;
715 break;
716 case AF_INET6:
717 level = SOL_IPV6;
718 optname = IPV6_RECVPKTINFO;
719 break;
720 default:
721 BUG();
722 }
723 err = kernel_setsockopt(svsk->sk_sock, level, optname,
724 (char *)&one, sizeof(one));
725 dprintk("svc: kernel_setsockopt returned %d\n", err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726}
727
728/*
729 * A data_ready event on a listening socket means there's a connection
730 * pending. Do not use state_change as a substitute for it.
731 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600732static void svc_tcp_listen_data_ready(struct sock *sk, int count_unused)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733{
Neil Brown939bb7e2005-09-13 01:25:39 -0700734 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
Eric Dumazeteaefd112011-02-18 03:26:36 +0000735 wait_queue_head_t *wq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736
737 dprintk("svc: socket %p TCP (listen) state change %d\n",
Neil Brown939bb7e2005-09-13 01:25:39 -0700738 sk, sk->sk_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739
Neil Brown939bb7e2005-09-13 01:25:39 -0700740 /*
741 * This callback may called twice when a new connection
742 * is established as a child socket inherits everything
743 * from a parent LISTEN socket.
744 * 1) data_ready method of the parent socket will be called
745 * when one of child sockets become ESTABLISHED.
746 * 2) data_ready method of the child socket may be called
747 * when it receives data before the socket is accepted.
748 * In case of 2, we should ignore it silently.
749 */
750 if (sk->sk_state == TCP_LISTEN) {
751 if (svsk) {
Tom Tucker02fc6c32007-12-30 21:07:48 -0600752 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
Tom Tuckerf6150c32007-12-30 21:07:57 -0600753 svc_xprt_enqueue(&svsk->sk_xprt);
Neil Brown939bb7e2005-09-13 01:25:39 -0700754 } else
755 printk("svc: socket %p: no user data\n", sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 }
Neil Brown939bb7e2005-09-13 01:25:39 -0700757
Eric Dumazeteaefd112011-02-18 03:26:36 +0000758 wq = sk_sleep(sk);
759 if (wq && waitqueue_active(wq))
760 wake_up_interruptible_all(wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761}
762
763/*
764 * A state change on a connected socket means it's dying or dead.
765 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600766static void svc_tcp_state_change(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767{
Neil Brown939bb7e2005-09-13 01:25:39 -0700768 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
Eric Dumazeteaefd112011-02-18 03:26:36 +0000769 wait_queue_head_t *wq = sk_sleep(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770
771 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
Neil Brown939bb7e2005-09-13 01:25:39 -0700772 sk, sk->sk_state, sk->sk_user_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773
Neil Brown939bb7e2005-09-13 01:25:39 -0700774 if (!svsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 printk("svc: socket %p: no user data\n", sk);
Neil Brown939bb7e2005-09-13 01:25:39 -0700776 else {
Tom Tucker02fc6c32007-12-30 21:07:48 -0600777 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
Tom Tuckerf6150c32007-12-30 21:07:57 -0600778 svc_xprt_enqueue(&svsk->sk_xprt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 }
Eric Dumazeteaefd112011-02-18 03:26:36 +0000780 if (wq && waitqueue_active(wq))
781 wake_up_interruptible_all(wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782}
783
Tom Tucker0f0257e2007-12-30 21:08:27 -0600784static void svc_tcp_data_ready(struct sock *sk, int count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785{
Neil Brown939bb7e2005-09-13 01:25:39 -0700786 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
Eric Dumazeteaefd112011-02-18 03:26:36 +0000787 wait_queue_head_t *wq = sk_sleep(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788
789 dprintk("svc: socket %p TCP data ready (svsk %p)\n",
Neil Brown939bb7e2005-09-13 01:25:39 -0700790 sk, sk->sk_user_data);
791 if (svsk) {
Tom Tucker02fc6c32007-12-30 21:07:48 -0600792 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
Tom Tuckerf6150c32007-12-30 21:07:57 -0600793 svc_xprt_enqueue(&svsk->sk_xprt);
Neil Brown939bb7e2005-09-13 01:25:39 -0700794 }
Eric Dumazeteaefd112011-02-18 03:26:36 +0000795 if (wq && waitqueue_active(wq))
796 wake_up_interruptible(wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797}
798
799/*
800 * Accept a TCP connection
801 */
Tom Tucker38a417c2007-12-30 21:07:36 -0600802static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803{
Tom Tucker38a417c2007-12-30 21:07:36 -0600804 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
akpm@linux-foundation.orgcdd88b92007-02-12 00:53:38 -0800805 struct sockaddr_storage addr;
806 struct sockaddr *sin = (struct sockaddr *) &addr;
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600807 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 struct socket *sock = svsk->sk_sock;
809 struct socket *newsock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 struct svc_sock *newsvsk;
811 int err, slen;
Pavel Emelyanov5216a8e2008-02-21 10:57:45 +0300812 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813
814 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
815 if (!sock)
Tom Tucker38a417c2007-12-30 21:07:36 -0600816 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817
Tom Tucker02fc6c32007-12-30 21:07:48 -0600818 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
Sridhar Samudralae6242e92006-08-07 20:58:01 -0700819 err = kernel_accept(sock, &newsock, O_NONBLOCK);
820 if (err < 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 if (err == -ENOMEM)
822 printk(KERN_WARNING "%s: no more sockets!\n",
823 serv->sv_name);
Joe Perchese87cc472012-05-13 21:56:26 +0000824 else if (err != -EAGAIN)
825 net_warn_ratelimited("%s: accept failed (err %d)!\n",
826 serv->sv_name, -err);
Tom Tucker38a417c2007-12-30 21:07:36 -0600827 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 }
Tom Tucker02fc6c32007-12-30 21:07:48 -0600829 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830
akpm@linux-foundation.orgcdd88b92007-02-12 00:53:38 -0800831 err = kernel_getpeername(newsock, sin, &slen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832 if (err < 0) {
Joe Perchese87cc472012-05-13 21:56:26 +0000833 net_warn_ratelimited("%s: peername failed (err %d)!\n",
834 serv->sv_name, -err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 goto failed; /* aborted connection or whatever */
836 }
837
838 /* Ideally, we would want to reject connections from unauthorized
Chuck Leverad06e4b2007-02-12 00:53:32 -0800839 * hosts here, but when we get encryption, the IP of the host won't
840 * tell us anything. For now just warn about unpriv connections.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 */
akpm@linux-foundation.orgcdd88b92007-02-12 00:53:38 -0800842 if (!svc_port_is_privileged(sin)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 dprintk(KERN_WARNING
Chuck Leverad06e4b2007-02-12 00:53:32 -0800844 "%s: connect from unprivileged port: %s\n",
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -0800845 serv->sv_name,
akpm@linux-foundation.orgcdd88b92007-02-12 00:53:38 -0800846 __svc_print_addr(sin, buf, sizeof(buf)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847 }
Chuck Leverad06e4b2007-02-12 00:53:32 -0800848 dprintk("%s: connect from %s\n", serv->sv_name,
akpm@linux-foundation.orgcdd88b92007-02-12 00:53:38 -0800849 __svc_print_addr(sin, buf, sizeof(buf)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850
851 /* make sure that a write doesn't block forever when
852 * low on memory
853 */
854 newsock->sk->sk_sndtimeo = HZ*30;
855
J. Bruce Fields72c35372012-08-13 17:46:17 -0400856 newsvsk = svc_setup_socket(serv, newsock,
857 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY));
858 if (IS_ERR(newsvsk))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859 goto failed;
Tom Tucker9dbc2402007-12-30 21:08:12 -0600860 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
Frank van Maarseveena9747692007-07-09 22:21:39 +0200861 err = kernel_getsockname(newsock, sin, &slen);
862 if (unlikely(err < 0)) {
863 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
864 slen = offsetof(struct sockaddr, sa_data);
865 }
Tom Tucker9dbc2402007-12-30 21:08:12 -0600866 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
Chuck Lever067d7812007-02-12 00:53:30 -0800867
Tom Tuckerf9f3cc42007-12-30 21:07:40 -0600868 if (serv->sv_stats)
869 serv->sv_stats->nettcpconn++;
870
871 return &newsvsk->sk_xprt;
872
873failed:
874 sock_release(newsock);
875 return NULL;
876}
877
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800878static unsigned int svc_tcp_restore_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
879{
880 unsigned int i, len, npages;
881
J. Bruce Fields8af345f2012-12-03 16:45:35 -0500882 if (svsk->sk_datalen == 0)
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800883 return 0;
J. Bruce Fields8af345f2012-12-03 16:45:35 -0500884 len = svsk->sk_datalen;
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800885 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
886 for (i = 0; i < npages; i++) {
887 if (rqstp->rq_pages[i] != NULL)
888 put_page(rqstp->rq_pages[i]);
889 BUG_ON(svsk->sk_pages[i] == NULL);
890 rqstp->rq_pages[i] = svsk->sk_pages[i];
891 svsk->sk_pages[i] = NULL;
892 }
893 rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
894 return len;
895}
896
897static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
898{
899 unsigned int i, len, npages;
900
J. Bruce Fields8af345f2012-12-03 16:45:35 -0500901 if (svsk->sk_datalen == 0)
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800902 return;
J. Bruce Fields8af345f2012-12-03 16:45:35 -0500903 len = svsk->sk_datalen;
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800904 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
905 for (i = 0; i < npages; i++) {
906 svsk->sk_pages[i] = rqstp->rq_pages[i];
907 rqstp->rq_pages[i] = NULL;
908 }
909}
910
911static void svc_tcp_clear_pages(struct svc_sock *svsk)
912{
913 unsigned int i, len, npages;
914
J. Bruce Fields8af345f2012-12-03 16:45:35 -0500915 if (svsk->sk_datalen == 0)
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800916 goto out;
J. Bruce Fields8af345f2012-12-03 16:45:35 -0500917 len = svsk->sk_datalen;
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800918 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
919 for (i = 0; i < npages; i++) {
920 BUG_ON(svsk->sk_pages[i] == NULL);
921 put_page(svsk->sk_pages[i]);
922 svsk->sk_pages[i] = NULL;
923 }
924out:
925 svsk->sk_tcplen = 0;
J. Bruce Fields8af345f2012-12-03 16:45:35 -0500926 svsk->sk_datalen = 0;
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800927}
928
Tom Tuckerf9f3cc42007-12-30 21:07:40 -0600929/*
J. Bruce Fieldsad46ccf2012-12-03 16:30:42 -0500930 * Receive fragment record header.
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +0300931 * If we haven't gotten the record length yet, get the next four bytes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 */
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +0300933static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934{
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600935 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
Trond Myklebust5ee78d42009-05-18 17:47:56 -0400936 unsigned int want;
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +0300937 int len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938
Tom Tucker02fc6c32007-12-30 21:07:48 -0600939 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940
Chuck Leverc0401ea2008-04-14 12:27:30 -0400941 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 struct kvec iov;
943
Trond Myklebust5ee78d42009-05-18 17:47:56 -0400944 want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
946 iov.iov_len = want;
947 if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
948 goto error;
949 svsk->sk_tcplen += len;
950
951 if (len < want) {
Chuck Leverc0401ea2008-04-14 12:27:30 -0400952 dprintk("svc: short recvfrom while reading record "
953 "length (%d of %d)\n", len, want);
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800954 return -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955 }
956
J. Bruce Fieldscc248d42012-12-03 16:11:13 -0500957 dprintk("svc: TCP record, %d bytes\n", svc_sock_reclen(svsk));
J. Bruce Fields836fbad2012-12-03 15:50:38 -0500958 if (svc_sock_reclen(svsk) + svsk->sk_datalen >
959 serv->sv_max_mesg) {
J. Bruce Fields3a28e332012-12-10 18:15:07 -0500960 net_notice_ratelimited("RPC: fragment too large: %d\n",
961 svc_sock_reclen(svsk));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 goto err_delete;
963 }
964 }
965
J. Bruce Fieldscc248d42012-12-03 16:11:13 -0500966 return svc_sock_reclen(svsk);
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800967error:
968 dprintk("RPC: TCP recv_record got %d\n", len);
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +0300969 return len;
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800970err_delete:
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +0300971 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +0300972 return -EAGAIN;
973}
974
Trond Myklebust586c52c2009-05-18 17:47:56 -0400975static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
Rahul Iyer4cfc7e62009-09-10 17:32:28 +0300976{
Trond Myklebust586c52c2009-05-18 17:47:56 -0400977 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
Rahul Iyer4cfc7e62009-09-10 17:32:28 +0300978 struct rpc_rqst *req = NULL;
Trond Myklebust586c52c2009-05-18 17:47:56 -0400979 struct kvec *src, *dst;
980 __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
J. Bruce Fields48e65552011-02-14 14:52:03 -0500981 __be32 xid;
982 __be32 calldir;
Rahul Iyer4cfc7e62009-09-10 17:32:28 +0300983
Rahul Iyer4cfc7e62009-09-10 17:32:28 +0300984 xid = *p++;
985 calldir = *p;
986
Trond Myklebust586c52c2009-05-18 17:47:56 -0400987 if (bc_xprt)
988 req = xprt_lookup_rqst(bc_xprt, xid);
J. Bruce Fieldsd75faea2010-11-30 19:15:01 -0500989
Trond Myklebust586c52c2009-05-18 17:47:56 -0400990 if (!req) {
991 printk(KERN_NOTICE
992 "%s: Got unrecognized reply: "
993 "calldir 0x%x xpt_bc_xprt %p xid %08x\n",
994 __func__, ntohl(calldir),
995 bc_xprt, xid);
996 return -EAGAIN;
Rahul Iyer4cfc7e62009-09-10 17:32:28 +0300997 }
Trond Myklebust586c52c2009-05-18 17:47:56 -0400998
999 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
1000 /*
1001 * XXX!: cheating for now! Only copying HEAD.
1002 * But we know this is good enough for now (in fact, for any
1003 * callback reply in the forseeable future).
1004 */
1005 dst = &req->rq_private_buf.head[0];
1006 src = &rqstp->rq_arg.head[0];
1007 if (dst->iov_len < src->iov_len)
1008 return -EAGAIN; /* whatever; just giving up. */
1009 memcpy(dst->iov_base, src->iov_base, src->iov_len);
J. Bruce Fieldscc248d42012-12-03 16:11:13 -05001010 xprt_complete_rqst(req->rq_task, rqstp->rq_arg.len);
Trond Myklebust586c52c2009-05-18 17:47:56 -04001011 rqstp->rq_arg.len = 0;
1012 return 0;
1013}
1014
1015static int copy_pages_to_kvecs(struct kvec *vec, struct page **pages, int len)
1016{
1017 int i = 0;
1018 int t = 0;
1019
1020 while (t < len) {
1021 vec[i].iov_base = page_address(pages[i]);
1022 vec[i].iov_len = PAGE_SIZE;
1023 i++;
1024 t += PAGE_SIZE;
1025 }
1026 return i;
Rahul Iyer4cfc7e62009-09-10 17:32:28 +03001027}
1028
J. Bruce Fields836fbad2012-12-03 15:50:38 -05001029static void svc_tcp_fragment_received(struct svc_sock *svsk)
1030{
1031 /* If we have more data, signal svc_xprt_enqueue() to try again */
1032 if (svc_recv_available(svsk) > sizeof(rpc_fraghdr))
1033 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1034 dprintk("svc: TCP %s record (%d bytes)\n",
1035 svc_sock_final_rec(svsk) ? "final" : "nonfinal",
1036 svc_sock_reclen(svsk));
1037 svsk->sk_tcplen = 0;
1038 svsk->sk_reclen = 0;
1039}
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001040
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +03001041/*
1042 * Receive data from a TCP socket.
1043 */
1044static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
1045{
1046 struct svc_sock *svsk =
1047 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
1048 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1049 int len;
1050 struct kvec *vec;
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001051 unsigned int want, base;
Trond Myklebust586c52c2009-05-18 17:47:56 -04001052 __be32 *p;
1053 __be32 calldir;
Trond Myklebust5ee78d42009-05-18 17:47:56 -04001054 int pnum;
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +03001055
1056 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1057 svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
1058 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
1059 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1060
1061 len = svc_tcp_recv_record(svsk, rqstp);
1062 if (len < 0)
1063 goto error;
1064
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001065 base = svc_tcp_restore_pages(svsk, rqstp);
J. Bruce Fields836fbad2012-12-03 15:50:38 -05001066 want = svc_sock_reclen(svsk) - (svsk->sk_tcplen - sizeof(rpc_fraghdr));
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001067
NeilBrown3cc03b12006-10-04 02:15:47 -07001068 vec = rqstp->rq_vec;
Rahul Iyer4cfc7e62009-09-10 17:32:28 +03001069
Trond Myklebust586c52c2009-05-18 17:47:56 -04001070 pnum = copy_pages_to_kvecs(&vec[0], &rqstp->rq_pages[0],
J. Bruce Fields836fbad2012-12-03 15:50:38 -05001071 svsk->sk_datalen + want);
Rahul Iyer4cfc7e62009-09-10 17:32:28 +03001072
NeilBrown44524352006-10-04 02:15:46 -07001073 rqstp->rq_respages = &rqstp->rq_pages[pnum];
J. Bruce Fieldsafc59402012-12-10 18:01:37 -05001074 rqstp->rq_next_page = rqstp->rq_respages + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075
1076 /* Now receive data */
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001077 len = svc_partial_recvfrom(rqstp, vec, pnum, want, base);
J. Bruce Fields8af345f2012-12-03 16:45:35 -05001078 if (len >= 0) {
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001079 svsk->sk_tcplen += len;
J. Bruce Fields8af345f2012-12-03 16:45:35 -05001080 svsk->sk_datalen += len;
1081 }
J. Bruce Fields836fbad2012-12-03 15:50:38 -05001082 if (len != want || !svc_sock_final_rec(svsk)) {
J. Bruce Fieldsbe1e4442012-08-09 18:12:28 -04001083 svc_tcp_save_pages(svsk, rqstp);
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001084 if (len < 0 && len != -EAGAIN)
J. Bruce Fieldsad46ccf2012-12-03 16:30:42 -05001085 goto err_delete;
J. Bruce Fields836fbad2012-12-03 15:50:38 -05001086 if (len == want)
1087 svc_tcp_fragment_received(svsk);
1088 else
J. Bruce Fields3a28e332012-12-10 18:15:07 -05001089 dprintk("svc: incomplete TCP record (%d of %d)\n",
1090 (int)(svsk->sk_tcplen - sizeof(rpc_fraghdr)),
J. Bruce Fields836fbad2012-12-03 15:50:38 -05001091 svc_sock_reclen(svsk));
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001092 goto err_noclose;
1093 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094
J. Bruce Fieldsad46ccf2012-12-03 16:30:42 -05001095 if (svc_sock_reclen(svsk) < 8)
1096 goto err_delete; /* client is nuts. */
1097
J. Bruce Fields836fbad2012-12-03 15:50:38 -05001098 rqstp->rq_arg.len = svsk->sk_datalen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 rqstp->rq_arg.page_base = 0;
Trond Myklebust5ee78d42009-05-18 17:47:56 -04001100 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1101 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 rqstp->rq_arg.page_len = 0;
Trond Myklebust5ee78d42009-05-18 17:47:56 -04001103 } else
1104 rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105
Tom Tucker5148bf42007-12-30 21:07:25 -06001106 rqstp->rq_xprt_ctxt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107 rqstp->rq_prot = IPPROTO_TCP;
1108
Trond Myklebust586c52c2009-05-18 17:47:56 -04001109 p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1110 calldir = p[1];
J. Bruce Fields8985ef02011-04-09 10:03:10 -04001111 if (calldir)
Trond Myklebust586c52c2009-05-18 17:47:56 -04001112 len = receive_cb_reply(svsk, rqstp);
Trond Myklebust586c52c2009-05-18 17:47:56 -04001113
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114 /* Reset TCP read info */
J. Bruce Fields8af345f2012-12-03 16:45:35 -05001115 svsk->sk_datalen = 0;
J. Bruce Fields836fbad2012-12-03 15:50:38 -05001116 svc_tcp_fragment_received(svsk);
Trond Myklebust0601f792009-05-18 17:47:56 -04001117
J. Bruce Fields8985ef02011-04-09 10:03:10 -04001118 if (len < 0)
1119 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120
Tom Tucker9dbc2402007-12-30 21:08:12 -06001121 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 if (serv->sv_stats)
1123 serv->sv_stats->nettcpcnt++;
1124
Trond Myklebust5ee78d42009-05-18 17:47:56 -04001125 return rqstp->rq_arg.len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +03001127error:
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001128 if (len != -EAGAIN)
J. Bruce Fieldsad46ccf2012-12-03 16:30:42 -05001129 goto err_delete;
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001130 dprintk("RPC: TCP recvfrom got EAGAIN\n");
J. Bruce Fields9f9d2eb2012-08-17 21:35:24 -04001131 return 0;
J. Bruce Fieldsad46ccf2012-12-03 16:30:42 -05001132err_delete:
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001133 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1134 svsk->sk_xprt.xpt_server->sv_name, -len);
1135 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1136err_noclose:
J. Bruce Fields9f9d2eb2012-08-17 21:35:24 -04001137 return 0; /* record not complete */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138}
1139
1140/*
1141 * Send out data on TCP socket.
1142 */
Tom Tucker0f0257e2007-12-30 21:08:27 -06001143static int svc_tcp_sendto(struct svc_rqst *rqstp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144{
1145 struct xdr_buf *xbufp = &rqstp->rq_res;
1146 int sent;
Alexey Dobriyand8ed0292006-09-26 22:29:38 -07001147 __be32 reclen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148
1149 /* Set up the first element of the reply kvec.
1150 * Any other kvecs that may be in use have been taken
1151 * care of by the server implementation itself.
1152 */
1153 reclen = htonl(0x80000000|((xbufp->len ) - 4));
1154 memcpy(xbufp->head[0].iov_base, &reclen, 4);
1155
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 sent = svc_sendto(rqstp, &rqstp->rq_res);
1157 if (sent != xbufp->len) {
Tom Tucker0f0257e2007-12-30 21:08:27 -06001158 printk(KERN_NOTICE
1159 "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1160 "- shutting down socket\n",
Tom Tucker57b1d3b2007-12-30 21:08:22 -06001161 rqstp->rq_xprt->xpt_server->sv_name,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162 (sent<0)?"got error":"sent only",
1163 sent, xbufp->len);
Tom Tucker57b1d3b2007-12-30 21:08:22 -06001164 set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
Tom Tuckerf6150c32007-12-30 21:07:57 -06001165 svc_xprt_enqueue(rqstp->rq_xprt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 sent = -EAGAIN;
1167 }
1168 return sent;
1169}
1170
Tom Tuckere831fe62007-12-30 21:07:29 -06001171/*
1172 * Setup response header. TCP has a 4B record length field.
1173 */
1174static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
1175{
1176 struct kvec *resv = &rqstp->rq_res.head[0];
1177
1178 /* tcp needs a space for the record length... */
1179 svc_putnl(resv, 0);
1180}
1181
Tom Tucker323bee32007-12-30 21:07:31 -06001182static int svc_tcp_has_wspace(struct svc_xprt *xprt)
1183{
Tom Tucker57b1d3b2007-12-30 21:08:22 -06001184 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
Trond Myklebust47fcb032009-05-18 17:47:56 -04001185 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
Tom Tucker323bee32007-12-30 21:07:31 -06001186 int required;
Tom Tucker323bee32007-12-30 21:07:31 -06001187
Trond Myklebust47fcb032009-05-18 17:47:56 -04001188 if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
1189 return 1;
1190 required = atomic_read(&xprt->xpt_reserved) + serv->sv_max_mesg;
1191 if (sk_stream_wspace(svsk->sk_sk) >= required)
1192 return 1;
Tom Tucker323bee32007-12-30 21:07:31 -06001193 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
Trond Myklebust47fcb032009-05-18 17:47:56 -04001194 return 0;
Tom Tucker323bee32007-12-30 21:07:31 -06001195}
1196
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001197static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
Pavel Emelyanov62832c02010-09-29 16:04:18 +04001198 struct net *net,
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001199 struct sockaddr *sa, int salen,
1200 int flags)
1201{
Pavel Emelyanov62832c02010-09-29 16:04:18 +04001202 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001203}
1204
Trond Myklebust9e00abc2011-07-13 19:20:49 -04001205#if defined(CONFIG_SUNRPC_BACKCHANNEL)
Andy Adamson1f11a032011-01-06 02:04:26 +00001206static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
1207 struct net *, struct sockaddr *,
1208 int, int);
1209static void svc_bc_sock_free(struct svc_xprt *xprt);
1210
1211static struct svc_xprt *svc_bc_tcp_create(struct svc_serv *serv,
1212 struct net *net,
1213 struct sockaddr *sa, int salen,
1214 int flags)
1215{
1216 return svc_bc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1217}
1218
1219static void svc_bc_tcp_sock_detach(struct svc_xprt *xprt)
1220{
1221}
1222
1223static struct svc_xprt_ops svc_tcp_bc_ops = {
1224 .xpo_create = svc_bc_tcp_create,
1225 .xpo_detach = svc_bc_tcp_sock_detach,
1226 .xpo_free = svc_bc_sock_free,
1227 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1228};
1229
1230static struct svc_xprt_class svc_tcp_bc_class = {
1231 .xcl_name = "tcp-bc",
1232 .xcl_owner = THIS_MODULE,
1233 .xcl_ops = &svc_tcp_bc_ops,
1234 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1235};
Andy Adamson16b2d1e2011-01-06 02:04:27 +00001236
1237static void svc_init_bc_xprt_sock(void)
1238{
1239 svc_reg_xprt_class(&svc_tcp_bc_class);
1240}
1241
1242static void svc_cleanup_bc_xprt_sock(void)
1243{
1244 svc_unreg_xprt_class(&svc_tcp_bc_class);
1245}
Trond Myklebust9e00abc2011-07-13 19:20:49 -04001246#else /* CONFIG_SUNRPC_BACKCHANNEL */
Andy Adamson16b2d1e2011-01-06 02:04:27 +00001247static void svc_init_bc_xprt_sock(void)
1248{
1249}
1250
1251static void svc_cleanup_bc_xprt_sock(void)
1252{
1253}
Trond Myklebust9e00abc2011-07-13 19:20:49 -04001254#endif /* CONFIG_SUNRPC_BACKCHANNEL */
Andy Adamson1f11a032011-01-06 02:04:26 +00001255
Tom Tucker360d87382007-12-30 21:07:17 -06001256static struct svc_xprt_ops svc_tcp_ops = {
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001257 .xpo_create = svc_tcp_create,
Tom Tucker5d137992007-12-30 21:07:23 -06001258 .xpo_recvfrom = svc_tcp_recvfrom,
1259 .xpo_sendto = svc_tcp_sendto,
Tom Tucker5148bf42007-12-30 21:07:25 -06001260 .xpo_release_rqst = svc_release_skb,
Trond Myklebust69b6ba32008-12-23 16:30:11 -05001261 .xpo_detach = svc_tcp_sock_detach,
Tom Tucker755ccea2007-12-30 21:07:27 -06001262 .xpo_free = svc_sock_free,
Tom Tuckere831fe62007-12-30 21:07:29 -06001263 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
Tom Tucker323bee32007-12-30 21:07:31 -06001264 .xpo_has_wspace = svc_tcp_has_wspace,
Tom Tucker38a417c2007-12-30 21:07:36 -06001265 .xpo_accept = svc_tcp_accept,
Tom Tucker360d87382007-12-30 21:07:17 -06001266};
1267
1268static struct svc_xprt_class svc_tcp_class = {
1269 .xcl_name = "tcp",
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001270 .xcl_owner = THIS_MODULE,
Tom Tucker360d87382007-12-30 21:07:17 -06001271 .xcl_ops = &svc_tcp_ops,
Tom Tucker49023152007-12-30 21:07:21 -06001272 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
Tom Tucker360d87382007-12-30 21:07:17 -06001273};
1274
1275void svc_init_xprt_sock(void)
1276{
1277 svc_reg_xprt_class(&svc_tcp_class);
1278 svc_reg_xprt_class(&svc_udp_class);
Andy Adamson16b2d1e2011-01-06 02:04:27 +00001279 svc_init_bc_xprt_sock();
Tom Tucker360d87382007-12-30 21:07:17 -06001280}
1281
1282void svc_cleanup_xprt_sock(void)
1283{
1284 svc_unreg_xprt_class(&svc_tcp_class);
1285 svc_unreg_xprt_class(&svc_udp_class);
Andy Adamson16b2d1e2011-01-06 02:04:27 +00001286 svc_cleanup_bc_xprt_sock();
Tom Tucker360d87382007-12-30 21:07:17 -06001287}
1288
Tom Tuckerbb5cf162007-12-30 21:07:50 -06001289static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290{
1291 struct sock *sk = svsk->sk_sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292
Stanislav Kinsburskybd4620d2011-12-06 14:19:10 +03001293 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1294 &svsk->sk_xprt, serv);
Tom Tuckerdef13d72007-12-30 21:08:08 -06001295 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 if (sk->sk_state == TCP_LISTEN) {
1297 dprintk("setting up TCP socket for listening\n");
Tom Tucker02fc6c32007-12-30 21:07:48 -06001298 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299 sk->sk_data_ready = svc_tcp_listen_data_ready;
Tom Tucker02fc6c32007-12-30 21:07:48 -06001300 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301 } else {
1302 dprintk("setting up TCP socket for reading\n");
1303 sk->sk_state_change = svc_tcp_state_change;
1304 sk->sk_data_ready = svc_tcp_data_ready;
Trond Myklebust47fcb032009-05-18 17:47:56 -04001305 sk->sk_write_space = svc_tcp_write_space;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306
1307 svsk->sk_reclen = 0;
1308 svsk->sk_tcplen = 0;
J. Bruce Fields8af345f2012-12-03 16:45:35 -05001309 svsk->sk_datalen = 0;
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001310 memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311
Chuck Leverb7872fe2008-04-14 12:27:01 -04001312 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313
Tom Tucker02fc6c32007-12-30 21:07:48 -06001314 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -08001315 if (sk->sk_state != TCP_ESTABLISHED)
Tom Tucker02fc6c32007-12-30 21:07:48 -06001316 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317 }
1318}
1319
Tom Tucker0f0257e2007-12-30 21:08:27 -06001320void svc_sock_update_bufs(struct svc_serv *serv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321{
1322 /*
1323 * The number of server threads has changed. Update
1324 * rcvbuf and sndbuf accordingly on all sockets
1325 */
Pavel Emelyanov8f3a6de2010-10-05 23:30:19 +04001326 struct svc_sock *svsk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327
1328 spin_lock_bh(&serv->sv_lock);
Pavel Emelyanov8f3a6de2010-10-05 23:30:19 +04001329 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
Tom Tucker02fc6c32007-12-30 21:07:48 -06001330 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331 spin_unlock_bh(&serv->sv_lock);
1332}
Trond Myklebust24c37672008-12-23 16:30:12 -05001333EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334
1335/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 * Initialize socket for RPC use and create svc_sock struct
1337 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1338 */
Chuck Lever6b174332007-02-12 00:53:28 -08001339static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1340 struct socket *sock,
J. Bruce Fields72c35372012-08-13 17:46:17 -04001341 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342{
1343 struct svc_sock *svsk;
1344 struct sock *inet;
Chuck Lever6b174332007-02-12 00:53:28 -08001345 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
J. Bruce Fields72c35372012-08-13 17:46:17 -04001346 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347
1348 dprintk("svc: svc_setup_socket %p\n", sock);
J. Bruce Fields72c35372012-08-13 17:46:17 -04001349 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1350 if (!svsk)
1351 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352
1353 inet = sock->sk;
1354
1355 /* Register socket with portmapper */
J. Bruce Fields72c35372012-08-13 17:46:17 -04001356 if (pmap_register)
1357 err = svc_register(serv, sock_net(sock->sk), inet->sk_family,
Stanislav Kinsbursky5247fab2012-01-13 14:02:48 +04001358 inet->sk_protocol,
Eric Dumazetc720c7e2009-10-15 06:30:45 +00001359 ntohs(inet_sk(inet)->inet_sport));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360
J. Bruce Fields72c35372012-08-13 17:46:17 -04001361 if (err < 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362 kfree(svsk);
J. Bruce Fields72c35372012-08-13 17:46:17 -04001363 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364 }
1365
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366 inet->sk_user_data = svsk;
1367 svsk->sk_sock = sock;
1368 svsk->sk_sk = inet;
1369 svsk->sk_ostate = inet->sk_state_change;
1370 svsk->sk_odata = inet->sk_data_ready;
1371 svsk->sk_owspace = inet->sk_write_space;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372
1373 /* Initialize the socket */
1374 if (sock->type == SOCK_DGRAM)
Tom Tuckerbb5cf162007-12-30 21:07:50 -06001375 svc_udp_init(svsk, serv);
Olga Kornievskaia96604392008-10-21 14:13:47 -04001376 else {
1377 /* initialise setting must have enough space to
1378 * receive and respond to one request.
1379 */
1380 svc_sock_setbufsize(svsk->sk_sock, 4 * serv->sv_max_mesg,
1381 4 * serv->sv_max_mesg);
Tom Tuckerbb5cf162007-12-30 21:07:50 -06001382 svc_tcp_init(svsk, serv);
Olga Kornievskaia96604392008-10-21 14:13:47 -04001383 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1386 svsk, svsk->sk_sk);
1387
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 return svsk;
1389}
1390
Chuck Leverbfba9ab2009-04-23 19:32:33 -04001391/**
1392 * svc_addsock - add a listener socket to an RPC service
1393 * @serv: pointer to RPC service to which to add a new listener
1394 * @fd: file descriptor of the new listener
1395 * @name_return: pointer to buffer to fill in with name of listener
1396 * @len: size of the buffer
1397 *
1398 * Fills in socket name and returns positive length of name if successful.
1399 * Name is terminated with '\n'. On error, returns a negative errno
1400 * value.
1401 */
1402int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1403 const size_t len)
NeilBrownb41b66d2006-10-02 02:17:48 -07001404{
1405 int err = 0;
1406 struct socket *so = sockfd_lookup(fd, &err);
1407 struct svc_sock *svsk = NULL;
J. Bruce Fieldsa8e10072012-08-13 18:01:03 -04001408 struct sockaddr_storage addr;
1409 struct sockaddr *sin = (struct sockaddr *)&addr;
1410 int salen;
NeilBrownb41b66d2006-10-02 02:17:48 -07001411
1412 if (!so)
1413 return err;
J. Bruce Fieldsa8e10072012-08-13 18:01:03 -04001414 err = -EAFNOSUPPORT;
Aime Le Rouzic205ba422010-01-26 14:03:56 -05001415 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
J. Bruce Fieldsa8e10072012-08-13 18:01:03 -04001416 goto out;
1417 err = -EPROTONOSUPPORT;
1418 if (so->sk->sk_protocol != IPPROTO_TCP &&
NeilBrownb41b66d2006-10-02 02:17:48 -07001419 so->sk->sk_protocol != IPPROTO_UDP)
J. Bruce Fieldsa8e10072012-08-13 18:01:03 -04001420 goto out;
1421 err = -EISCONN;
1422 if (so->state > SS_UNCONNECTED)
1423 goto out;
1424 err = -ENOENT;
1425 if (!try_module_get(THIS_MODULE))
1426 goto out;
1427 svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS);
1428 if (IS_ERR(svsk)) {
1429 module_put(THIS_MODULE);
1430 err = PTR_ERR(svsk);
1431 goto out;
NeilBrownb41b66d2006-10-02 02:17:48 -07001432 }
J. Bruce Fieldsa8e10072012-08-13 18:01:03 -04001433 if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0)
1434 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
J. Bruce Fields39b55302012-08-14 15:50:34 -04001435 svc_add_new_perm_xprt(serv, &svsk->sk_xprt);
Chuck Levere7942b92009-04-23 19:32:48 -04001436 return svc_one_sock_name(svsk, name_return, len);
J. Bruce Fieldsa8e10072012-08-13 18:01:03 -04001437out:
1438 sockfd_put(so);
1439 return err;
NeilBrownb41b66d2006-10-02 02:17:48 -07001440}
1441EXPORT_SYMBOL_GPL(svc_addsock);
1442
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443/*
1444 * Create socket for RPC service.
1445 */
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001446static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1447 int protocol,
Pavel Emelyanov62832c02010-09-29 16:04:18 +04001448 struct net *net,
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001449 struct sockaddr *sin, int len,
1450 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451{
1452 struct svc_sock *svsk;
1453 struct socket *sock;
1454 int error;
1455 int type;
Tom Tucker9dbc2402007-12-30 21:08:12 -06001456 struct sockaddr_storage addr;
1457 struct sockaddr *newsin = (struct sockaddr *)&addr;
1458 int newlen;
Trond Myklebustc69da772009-03-30 18:59:17 -04001459 int family;
1460 int val;
Pavel Emelyanov5216a8e2008-02-21 10:57:45 +03001461 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462
Chuck Leverad06e4b2007-02-12 00:53:32 -08001463 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1464 serv->sv_program->pg_name, protocol,
Chuck Lever77f1f672007-02-12 00:53:39 -08001465 __svc_print_addr(sin, buf, sizeof(buf)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466
1467 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1468 printk(KERN_WARNING "svc: only UDP and TCP "
1469 "sockets supported\n");
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001470 return ERR_PTR(-EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472
Trond Myklebustc69da772009-03-30 18:59:17 -04001473 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1474 switch (sin->sa_family) {
1475 case AF_INET6:
1476 family = PF_INET6;
1477 break;
1478 case AF_INET:
1479 family = PF_INET;
1480 break;
1481 default:
1482 return ERR_PTR(-EINVAL);
1483 }
1484
Pavel Emelyanov14ec63c2010-09-29 16:06:57 +04001485 error = __sock_create(net, family, type, protocol, &sock, 1);
Chuck Lever77f1f672007-02-12 00:53:39 -08001486 if (error < 0)
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001487 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488
Peter Zijlstraed075362006-12-06 20:35:24 -08001489 svc_reclassify_socket(sock);
1490
Trond Myklebustc69da772009-03-30 18:59:17 -04001491 /*
1492 * If this is an PF_INET6 listener, we want to avoid
1493 * getting requests from IPv4 remotes. Those should
1494 * be shunted to a PF_INET listener via rpcbind.
1495 */
1496 val = 1;
1497 if (family == PF_INET6)
1498 kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
1499 (char *)&val, sizeof(val));
1500
Eric Sesterhenn18114742006-09-28 14:37:07 -07001501 if (type == SOCK_STREAM)
Pavel Emelyanov4a17fd52012-04-19 03:39:36 +00001502 sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */
Chuck Lever77f1f672007-02-12 00:53:39 -08001503 error = kernel_bind(sock, sin, len);
Eric Sesterhenn18114742006-09-28 14:37:07 -07001504 if (error < 0)
1505 goto bummer;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506
Tom Tucker9dbc2402007-12-30 21:08:12 -06001507 newlen = len;
1508 error = kernel_getsockname(sock, newsin, &newlen);
1509 if (error < 0)
1510 goto bummer;
1511
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512 if (protocol == IPPROTO_TCP) {
Sridhar Samudralae6242e92006-08-07 20:58:01 -07001513 if ((error = kernel_listen(sock, 64)) < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514 goto bummer;
1515 }
1516
J. Bruce Fields72c35372012-08-13 17:46:17 -04001517 svsk = svc_setup_socket(serv, sock, flags);
J. Bruce Fieldsa8e10072012-08-13 18:01:03 -04001518 if (IS_ERR(svsk)) {
1519 error = PTR_ERR(svsk);
1520 goto bummer;
NeilBrowne79eff12007-02-12 00:53:30 -08001521 }
J. Bruce Fieldsa8e10072012-08-13 18:01:03 -04001522 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1523 return (struct svc_xprt *)svsk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524bummer:
1525 dprintk("svc: svc_create_socket error = %d\n", -error);
1526 sock_release(sock);
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001527 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528}
1529
1530/*
Tom Tucker755ccea2007-12-30 21:07:27 -06001531 * Detach the svc_sock from the socket so that no
1532 * more callbacks occur.
1533 */
1534static void svc_sock_detach(struct svc_xprt *xprt)
1535{
1536 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1537 struct sock *sk = svsk->sk_sk;
Eric Dumazeteaefd112011-02-18 03:26:36 +00001538 wait_queue_head_t *wq;
Tom Tucker755ccea2007-12-30 21:07:27 -06001539
1540 dprintk("svc: svc_sock_detach(%p)\n", svsk);
1541
1542 /* put back the old socket callbacks */
1543 sk->sk_state_change = svsk->sk_ostate;
1544 sk->sk_data_ready = svsk->sk_odata;
1545 sk->sk_write_space = svsk->sk_owspace;
Trond Myklebust69b6ba32008-12-23 16:30:11 -05001546
Eric Dumazeteaefd112011-02-18 03:26:36 +00001547 wq = sk_sleep(sk);
1548 if (wq && waitqueue_active(wq))
1549 wake_up_interruptible(wq);
Trond Myklebust69b6ba32008-12-23 16:30:11 -05001550}
1551
1552/*
1553 * Disconnect the socket, and reset the callbacks
1554 */
1555static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1556{
1557 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1558
1559 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
1560
1561 svc_sock_detach(xprt);
1562
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001563 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1564 svc_tcp_clear_pages(svsk);
Trond Myklebust69b6ba32008-12-23 16:30:11 -05001565 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001566 }
Tom Tucker755ccea2007-12-30 21:07:27 -06001567}
1568
1569/*
1570 * Free the svc_sock's socket resources and the svc_sock itself.
1571 */
1572static void svc_sock_free(struct svc_xprt *xprt)
1573{
1574 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1575 dprintk("svc: svc_sock_free(%p)\n", svsk);
1576
Tom Tucker755ccea2007-12-30 21:07:27 -06001577 if (svsk->sk_sock->file)
1578 sockfd_put(svsk->sk_sock);
1579 else
1580 sock_release(svsk->sk_sock);
1581 kfree(svsk);
1582}
Benny Halevy7652e5a2009-04-01 09:23:09 -04001583
Trond Myklebust9e00abc2011-07-13 19:20:49 -04001584#if defined(CONFIG_SUNRPC_BACKCHANNEL)
Benny Halevy7652e5a2009-04-01 09:23:09 -04001585/*
Andy Adamson1f11a032011-01-06 02:04:26 +00001586 * Create a back channel svc_xprt which shares the fore channel socket.
Benny Halevy7652e5a2009-04-01 09:23:09 -04001587 */
Andy Adamson1f11a032011-01-06 02:04:26 +00001588static struct svc_xprt *svc_bc_create_socket(struct svc_serv *serv,
1589 int protocol,
1590 struct net *net,
1591 struct sockaddr *sin, int len,
1592 int flags)
Benny Halevy7652e5a2009-04-01 09:23:09 -04001593{
1594 struct svc_sock *svsk;
Andy Adamson1f11a032011-01-06 02:04:26 +00001595 struct svc_xprt *xprt;
Benny Halevy7652e5a2009-04-01 09:23:09 -04001596
Andy Adamson1f11a032011-01-06 02:04:26 +00001597 if (protocol != IPPROTO_TCP) {
1598 printk(KERN_WARNING "svc: only TCP sockets"
1599 " supported on shared back channel\n");
1600 return ERR_PTR(-EINVAL);
1601 }
1602
Benny Halevy7652e5a2009-04-01 09:23:09 -04001603 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1604 if (!svsk)
Andy Adamson1f11a032011-01-06 02:04:26 +00001605 return ERR_PTR(-ENOMEM);
Benny Halevy7652e5a2009-04-01 09:23:09 -04001606
1607 xprt = &svsk->sk_xprt;
Stanislav Kinsburskybd4620d2011-12-06 14:19:10 +03001608 svc_xprt_init(net, &svc_tcp_bc_class, xprt, serv);
Andy Adamson1f11a032011-01-06 02:04:26 +00001609
Andy Adamson4a19de02011-01-06 02:04:35 +00001610 serv->sv_bc_xprt = xprt;
Andy Adamson1f11a032011-01-06 02:04:26 +00001611
Benny Halevy7652e5a2009-04-01 09:23:09 -04001612 return xprt;
1613}
Benny Halevy7652e5a2009-04-01 09:23:09 -04001614
1615/*
Andy Adamson1f11a032011-01-06 02:04:26 +00001616 * Free a back channel svc_sock.
Benny Halevy7652e5a2009-04-01 09:23:09 -04001617 */
Andy Adamson1f11a032011-01-06 02:04:26 +00001618static void svc_bc_sock_free(struct svc_xprt *xprt)
Benny Halevy7652e5a2009-04-01 09:23:09 -04001619{
Andy Adamson778be232011-01-25 15:38:01 +00001620 if (xprt)
Benny Halevy7652e5a2009-04-01 09:23:09 -04001621 kfree(container_of(xprt, struct svc_sock, sk_xprt));
1622}
Trond Myklebust9e00abc2011-07-13 19:20:49 -04001623#endif /* CONFIG_SUNRPC_BACKCHANNEL */