blob: 983bfa9f310258afaff576ab72a2f3b367977442 [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>
24#include <linux/errno.h>
25#include <linux/fcntl.h>
26#include <linux/net.h>
27#include <linux/in.h>
28#include <linux/inet.h>
29#include <linux/udp.h>
Andrew Morton91483c42005-08-09 20:20:07 -070030#include <linux/tcp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include <linux/unistd.h>
32#include <linux/slab.h>
33#include <linux/netdevice.h>
34#include <linux/skbuff.h>
NeilBrownb41b66d2006-10-02 02:17:48 -070035#include <linux/file.h>
Nigel Cunningham7dfb7102006-12-06 20:34:23 -080036#include <linux/freezer.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include <net/sock.h>
38#include <net/checksum.h>
39#include <net/ip.h>
Chuck Leverb92503b2007-02-12 00:53:36 -080040#include <net/ipv6.h>
Chuck Leverb7872fe2008-04-14 12:27:01 -040041#include <net/tcp.h>
Arnaldo Carvalho de Meloc752f072005-08-09 20:08:28 -070042#include <net/tcp_states.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070043#include <asm/uaccess.h>
44#include <asm/ioctls.h>
45
46#include <linux/sunrpc/types.h>
Chuck Leverad06e4b2007-02-12 00:53:32 -080047#include <linux/sunrpc/clnt.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070048#include <linux/sunrpc/xdr.h>
Chuck Leverc0401ea2008-04-14 12:27:30 -040049#include <linux/sunrpc/msg_prot.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070050#include <linux/sunrpc/svcsock.h>
51#include <linux/sunrpc/stats.h>
52
Tom Tucker360d87382007-12-30 21:07:17 -060053#define RPCDBG_FACILITY RPCDBG_SVCXPRT
Linus Torvalds1da177e2005-04-16 15:20:36 -070054
55
56static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
Chuck Lever6b174332007-02-12 00:53:28 -080057 int *errp, int flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -070058static void svc_udp_data_ready(struct sock *, int);
59static int svc_udp_recvfrom(struct svc_rqst *);
60static int svc_udp_sendto(struct svc_rqst *);
Tom Tucker755ccea2007-12-30 21:07:27 -060061static void svc_sock_detach(struct svc_xprt *);
Trond Myklebust69b6ba32008-12-23 16:30:11 -050062static void svc_tcp_sock_detach(struct svc_xprt *);
Tom Tucker755ccea2007-12-30 21:07:27 -060063static void svc_sock_free(struct svc_xprt *);
Linus Torvalds1da177e2005-04-16 15:20:36 -070064
Tom Tuckerb700cbb2007-12-30 21:07:42 -060065static struct svc_xprt *svc_create_socket(struct svc_serv *, int,
66 struct sockaddr *, int, int);
Peter Zijlstraed075362006-12-06 20:35:24 -080067#ifdef CONFIG_DEBUG_LOCK_ALLOC
68static struct lock_class_key svc_key[2];
69static struct lock_class_key svc_slock_key[2];
70
Tom Tucker0f0257e2007-12-30 21:08:27 -060071static void svc_reclassify_socket(struct socket *sock)
Peter Zijlstraed075362006-12-06 20:35:24 -080072{
73 struct sock *sk = sock->sk;
John Heffner02b3d342007-09-12 10:42:12 +020074 BUG_ON(sock_owned_by_user(sk));
Peter Zijlstraed075362006-12-06 20:35:24 -080075 switch (sk->sk_family) {
76 case AF_INET:
77 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
Tom Tuckerdef13d72007-12-30 21:08:08 -060078 &svc_slock_key[0],
79 "sk_xprt.xpt_lock-AF_INET-NFSD",
80 &svc_key[0]);
Peter Zijlstraed075362006-12-06 20:35:24 -080081 break;
82
83 case AF_INET6:
84 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
Tom Tuckerdef13d72007-12-30 21:08:08 -060085 &svc_slock_key[1],
86 "sk_xprt.xpt_lock-AF_INET6-NFSD",
87 &svc_key[1]);
Peter Zijlstraed075362006-12-06 20:35:24 -080088 break;
89
90 default:
91 BUG();
92 }
93}
94#else
Tom Tucker0f0257e2007-12-30 21:08:27 -060095static void svc_reclassify_socket(struct socket *sock)
Peter Zijlstraed075362006-12-06 20:35:24 -080096{
97}
98#endif
99
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100/*
101 * Release an skbuff after use
102 */
Tom Tucker5148bf42007-12-30 21:07:25 -0600103static void svc_release_skb(struct svc_rqst *rqstp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104{
Tom Tucker5148bf42007-12-30 21:07:25 -0600105 struct sk_buff *skb = rqstp->rq_xprt_ctxt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106
107 if (skb) {
Tom Tucker57b1d3b2007-12-30 21:08:22 -0600108 struct svc_sock *svsk =
109 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
Tom Tucker5148bf42007-12-30 21:07:25 -0600110 rqstp->rq_xprt_ctxt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111
112 dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
Tom Tucker57b1d3b2007-12-30 21:08:22 -0600113 skb_free_datagram(svsk->sk_sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115}
116
Chuck Leverb92503b2007-02-12 00:53:36 -0800117union svc_pktinfo_u {
118 struct in_pktinfo pkti;
Chuck Leverb92503b2007-02-12 00:53:36 -0800119 struct in6_pktinfo pkti6;
Chuck Leverb92503b2007-02-12 00:53:36 -0800120};
David S. Millerbc375ea2007-04-12 13:35:59 -0700121#define SVC_PKTINFO_SPACE \
122 CMSG_SPACE(sizeof(union svc_pktinfo_u))
Chuck Leverb92503b2007-02-12 00:53:36 -0800123
124static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
125{
Tom Tucker57b1d3b2007-12-30 21:08:22 -0600126 struct svc_sock *svsk =
127 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
128 switch (svsk->sk_sk->sk_family) {
Chuck Leverb92503b2007-02-12 00:53:36 -0800129 case AF_INET: {
130 struct in_pktinfo *pki = CMSG_DATA(cmh);
131
132 cmh->cmsg_level = SOL_IP;
133 cmh->cmsg_type = IP_PKTINFO;
134 pki->ipi_ifindex = 0;
135 pki->ipi_spec_dst.s_addr = rqstp->rq_daddr.addr.s_addr;
136 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
137 }
138 break;
NeilBrown5a05ed72007-03-06 01:42:22 -0800139
Chuck Leverb92503b2007-02-12 00:53:36 -0800140 case AF_INET6: {
141 struct in6_pktinfo *pki = CMSG_DATA(cmh);
142
143 cmh->cmsg_level = SOL_IPV6;
144 cmh->cmsg_type = IPV6_PKTINFO;
145 pki->ipi6_ifindex = 0;
146 ipv6_addr_copy(&pki->ipi6_addr,
147 &rqstp->rq_daddr.addr6);
148 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
149 }
150 break;
Chuck Leverb92503b2007-02-12 00:53:36 -0800151 }
152 return;
153}
154
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155/*
156 * Generic sendto routine
157 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600158static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159{
Tom Tucker57b1d3b2007-12-30 21:08:22 -0600160 struct svc_sock *svsk =
161 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162 struct socket *sock = svsk->sk_sock;
163 int slen;
David S. Millerbc375ea2007-04-12 13:35:59 -0700164 union {
165 struct cmsghdr hdr;
166 long all[SVC_PKTINFO_SPACE / sizeof(long)];
167 } buffer;
168 struct cmsghdr *cmh = &buffer.hdr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169 int len = 0;
170 int result;
171 int size;
172 struct page **ppage = xdr->pages;
173 size_t base = xdr->page_base;
174 unsigned int pglen = xdr->page_len;
175 unsigned int flags = MSG_MORE;
Pavel Emelyanov5216a8e2008-02-21 10:57:45 +0300176 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177
178 slen = xdr->len;
179
180 if (rqstp->rq_prot == IPPROTO_UDP) {
Chuck Leverb92503b2007-02-12 00:53:36 -0800181 struct msghdr msg = {
182 .msg_name = &rqstp->rq_addr,
183 .msg_namelen = rqstp->rq_addrlen,
184 .msg_control = cmh,
185 .msg_controllen = sizeof(buffer),
186 .msg_flags = MSG_MORE,
187 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188
Chuck Leverb92503b2007-02-12 00:53:36 -0800189 svc_set_cmsg_data(rqstp, cmh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190
191 if (sock_sendmsg(sock, &msg, 0) < 0)
192 goto out;
193 }
194
195 /* send head */
196 if (slen == xdr->head[0].iov_len)
197 flags = 0;
NeilBrown44524352006-10-04 02:15:46 -0700198 len = kernel_sendpage(sock, rqstp->rq_respages[0], 0,
199 xdr->head[0].iov_len, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 if (len != xdr->head[0].iov_len)
201 goto out;
202 slen -= xdr->head[0].iov_len;
203 if (slen == 0)
204 goto out;
205
206 /* send page data */
207 size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen;
208 while (pglen > 0) {
209 if (slen == size)
210 flags = 0;
Sridhar Samudralae6242e92006-08-07 20:58:01 -0700211 result = kernel_sendpage(sock, *ppage, base, size, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 if (result > 0)
213 len += result;
214 if (result != size)
215 goto out;
216 slen -= size;
217 pglen -= size;
218 size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen;
219 base = 0;
220 ppage++;
221 }
222 /* send tail */
223 if (xdr->tail[0].iov_len) {
NeilBrown44524352006-10-04 02:15:46 -0700224 result = kernel_sendpage(sock, rqstp->rq_respages[0],
225 ((unsigned long)xdr->tail[0].iov_base)
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -0800226 & (PAGE_SIZE-1),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 xdr->tail[0].iov_len, 0);
228
229 if (result > 0)
230 len += result;
231 }
232out:
Chuck Leverad06e4b2007-02-12 00:53:32 -0800233 dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n",
Tom Tucker57b1d3b2007-12-30 21:08:22 -0600234 svsk, xdr->head[0].iov_base, xdr->head[0].iov_len,
Chuck Leverad06e4b2007-02-12 00:53:32 -0800235 xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236
237 return len;
238}
239
240/*
NeilBrown80212d52006-10-02 02:17:47 -0700241 * Report socket names for nfsdfs
242 */
Chuck Levere7942b92009-04-23 19:32:48 -0400243static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining)
NeilBrown80212d52006-10-02 02:17:47 -0700244{
245 int len;
246
247 switch(svsk->sk_sk->sk_family) {
Chuck Levere7942b92009-04-23 19:32:48 -0400248 case PF_INET:
249 len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n",
Harvey Harrison21454aa2008-10-31 00:54:56 -0700250 svsk->sk_sk->sk_protocol == IPPROTO_UDP ?
NeilBrown80212d52006-10-02 02:17:47 -0700251 "udp" : "tcp",
Harvey Harrison21454aa2008-10-31 00:54:56 -0700252 &inet_sk(svsk->sk_sk)->rcv_saddr,
NeilBrown80212d52006-10-02 02:17:47 -0700253 inet_sk(svsk->sk_sk)->num);
254 break;
Chuck Lever58de2f82009-04-23 19:32:55 -0400255 case PF_INET6:
256 len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n",
257 svsk->sk_sk->sk_protocol == IPPROTO_UDP ?
258 "udp" : "tcp",
259 &inet6_sk(svsk->sk_sk)->rcv_saddr,
260 inet_sk(svsk->sk_sk)->num);
261 break;
NeilBrown80212d52006-10-02 02:17:47 -0700262 default:
Chuck Levere7942b92009-04-23 19:32:48 -0400263 len = snprintf(buf, remaining, "*unknown-%d*\n",
NeilBrown80212d52006-10-02 02:17:47 -0700264 svsk->sk_sk->sk_family);
265 }
Chuck Levere7942b92009-04-23 19:32:48 -0400266
267 if (len >= remaining) {
268 *buf = '\0';
269 return -ENAMETOOLONG;
270 }
NeilBrown80212d52006-10-02 02:17:47 -0700271 return len;
272}
273
Chuck Lever8435d342009-04-23 19:32:40 -0400274/**
275 * svc_sock_names - construct a list of listener names in a string
276 * @serv: pointer to RPC service
277 * @buf: pointer to a buffer to fill in with socket names
278 * @buflen: size of the buffer to be filled
279 * @toclose: pointer to '\0'-terminated C string containing the name
280 * of a listener to be closed
281 *
282 * Fills in @buf with a '\n'-separated list of names of listener
283 * sockets. If @toclose is not NULL, the socket named by @toclose
284 * is closed, and is not included in the output list.
285 *
286 * Returns positive length of the socket name string, or a negative
287 * errno value on error.
288 */
289int svc_sock_names(struct svc_serv *serv, char *buf, const size_t buflen,
290 const char *toclose)
NeilBrown80212d52006-10-02 02:17:47 -0700291{
NeilBrownb41b66d2006-10-02 02:17:48 -0700292 struct svc_sock *svsk, *closesk = NULL;
NeilBrown80212d52006-10-02 02:17:47 -0700293 int len = 0;
294
295 if (!serv)
296 return 0;
Chuck Levere7942b92009-04-23 19:32:48 -0400297
NeilBrownaaf68cf2007-02-08 14:20:30 -0800298 spin_lock_bh(&serv->sv_lock);
Tom Tucker7a182082007-12-30 21:07:53 -0600299 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list) {
Chuck Levere7942b92009-04-23 19:32:48 -0400300 int onelen = svc_one_sock_name(svsk, buf + len, buflen - len);
301 if (onelen < 0) {
302 len = onelen;
303 break;
304 }
305 if (toclose && strcmp(toclose, buf + len) == 0)
NeilBrownb41b66d2006-10-02 02:17:48 -0700306 closesk = svsk;
307 else
308 len += onelen;
NeilBrown80212d52006-10-02 02:17:47 -0700309 }
NeilBrownaaf68cf2007-02-08 14:20:30 -0800310 spin_unlock_bh(&serv->sv_lock);
Chuck Levere7942b92009-04-23 19:32:48 -0400311
NeilBrownb41b66d2006-10-02 02:17:48 -0700312 if (closesk)
NeilBrown5680c442006-10-04 02:15:45 -0700313 /* Should unregister with portmap, but you cannot
314 * unregister just one protocol...
315 */
Tom Tucker7a182082007-12-30 21:07:53 -0600316 svc_close_xprt(&closesk->sk_xprt);
NeilBrown37a03472006-10-04 02:15:44 -0700317 else if (toclose)
318 return -ENOENT;
NeilBrown80212d52006-10-02 02:17:47 -0700319 return len;
320}
Trond Myklebust24c37672008-12-23 16:30:12 -0500321EXPORT_SYMBOL_GPL(svc_sock_names);
NeilBrown80212d52006-10-02 02:17:47 -0700322
323/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 * Check input queue length
325 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600326static int svc_recv_available(struct svc_sock *svsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 struct socket *sock = svsk->sk_sock;
329 int avail, err;
330
Sridhar Samudralae6242e92006-08-07 20:58:01 -0700331 err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332
333 return (err >= 0)? avail : err;
334}
335
336/*
337 * Generic recvfrom routine.
338 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600339static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr,
340 int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341{
Tom Tucker57b1d3b2007-12-30 21:08:22 -0600342 struct svc_sock *svsk =
343 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
Chuck Lever1ba95102007-02-12 00:53:31 -0800344 struct msghdr msg = {
345 .msg_flags = MSG_DONTWAIT,
346 };
347 int len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348
Tom Tucker260c1d12007-12-30 21:08:29 -0600349 rqstp->rq_xprt_hlen = 0;
350
Chuck Lever1ba95102007-02-12 00:53:31 -0800351 len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen,
352 msg.msg_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
Chuck Lever1ba95102007-02-12 00:53:31 -0800355 svsk, iov[0].iov_base, iov[0].iov_len, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 return len;
357}
358
359/*
360 * Set socket snd and rcv buffer lengths
361 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600362static void svc_sock_setbufsize(struct socket *sock, unsigned int snd,
363 unsigned int rcv)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364{
365#if 0
366 mm_segment_t oldfs;
367 oldfs = get_fs(); set_fs(KERNEL_DS);
368 sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
369 (char*)&snd, sizeof(snd));
370 sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
371 (char*)&rcv, sizeof(rcv));
372#else
373 /* sock_setsockopt limits use to sysctl_?mem_max,
374 * which isn't acceptable. Until that is made conditional
375 * on not having CAP_SYS_RESOURCE or similar, we go direct...
376 * DaveM said I could!
377 */
378 lock_sock(sock->sk);
379 sock->sk->sk_sndbuf = snd * 2;
380 sock->sk->sk_rcvbuf = rcv * 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 release_sock(sock->sk);
382#endif
383}
384/*
385 * INET callback when data has been received on the socket.
386 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600387static void svc_udp_data_ready(struct sock *sk, int count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388{
Neil Brown939bb7e2005-09-13 01:25:39 -0700389 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390
Neil Brown939bb7e2005-09-13 01:25:39 -0700391 if (svsk) {
392 dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n",
Tom Tucker02fc6c32007-12-30 21:07:48 -0600393 svsk, sk, count,
394 test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
395 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
Tom Tuckerf6150c32007-12-30 21:07:57 -0600396 svc_xprt_enqueue(&svsk->sk_xprt);
Neil Brown939bb7e2005-09-13 01:25:39 -0700397 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
399 wake_up_interruptible(sk->sk_sleep);
400}
401
402/*
403 * INET callback when space is newly available on the socket.
404 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600405static void svc_write_space(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406{
407 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
408
409 if (svsk) {
410 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
Tom Tucker02fc6c32007-12-30 21:07:48 -0600411 svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
Tom Tuckerf6150c32007-12-30 21:07:57 -0600412 svc_xprt_enqueue(&svsk->sk_xprt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 }
414
415 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) {
Neil Brown939bb7e2005-09-13 01:25:39 -0700416 dprintk("RPC svc_write_space: someone sleeping on %p\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417 svsk);
418 wake_up_interruptible(sk->sk_sleep);
419 }
420}
421
Tom Tucker9dbc2402007-12-30 21:08:12 -0600422/*
423 * Copy the UDP datagram's destination address to the rqstp structure.
424 * The 'destination' address in this case is the address to which the
425 * peer sent the datagram, i.e. our local address. For multihomed
426 * hosts, this can change from msg to msg. Note that only the IP
427 * address changes, the port number should remain the same.
428 */
429static void svc_udp_get_dest_address(struct svc_rqst *rqstp,
430 struct cmsghdr *cmh)
Chuck Lever95756482007-02-12 00:53:38 -0800431{
Tom Tucker57b1d3b2007-12-30 21:08:22 -0600432 struct svc_sock *svsk =
433 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
434 switch (svsk->sk_sk->sk_family) {
Chuck Lever95756482007-02-12 00:53:38 -0800435 case AF_INET: {
NeilBrown7a37f572007-03-06 01:42:21 -0800436 struct in_pktinfo *pki = CMSG_DATA(cmh);
437 rqstp->rq_daddr.addr.s_addr = pki->ipi_spec_dst.s_addr;
Chuck Lever95756482007-02-12 00:53:38 -0800438 break;
NeilBrown7a37f572007-03-06 01:42:21 -0800439 }
Chuck Lever95756482007-02-12 00:53:38 -0800440 case AF_INET6: {
NeilBrown7a37f572007-03-06 01:42:21 -0800441 struct in6_pktinfo *pki = CMSG_DATA(cmh);
442 ipv6_addr_copy(&rqstp->rq_daddr.addr6, &pki->ipi6_addr);
Chuck Lever95756482007-02-12 00:53:38 -0800443 break;
NeilBrown7a37f572007-03-06 01:42:21 -0800444 }
Chuck Lever95756482007-02-12 00:53:38 -0800445 }
Chuck Lever95756482007-02-12 00:53:38 -0800446}
447
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448/*
449 * Receive a datagram from a UDP socket.
450 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600451static int svc_udp_recvfrom(struct svc_rqst *rqstp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452{
Tom Tucker57b1d3b2007-12-30 21:08:22 -0600453 struct svc_sock *svsk =
454 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600455 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 struct sk_buff *skb;
David S. Millerbc375ea2007-04-12 13:35:59 -0700457 union {
458 struct cmsghdr hdr;
459 long all[SVC_PKTINFO_SPACE / sizeof(long)];
460 } buffer;
461 struct cmsghdr *cmh = &buffer.hdr;
NeilBrown7a37f572007-03-06 01:42:21 -0800462 struct msghdr msg = {
463 .msg_name = svc_addr(rqstp),
464 .msg_control = cmh,
465 .msg_controllen = sizeof(buffer),
466 .msg_flags = MSG_DONTWAIT,
467 };
Chuck Leverabc5c442009-04-23 19:31:25 -0400468 size_t len;
469 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470
Tom Tucker02fc6c32007-12-30 21:07:48 -0600471 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 /* udp sockets need large rcvbuf as all pending
473 * requests are still in that buffer. sndbuf must
474 * also be large enough that there is enough space
Greg Banks3262c812006-10-02 02:17:58 -0700475 * for one reply per thread. We count all threads
476 * rather than threads in a particular pool, which
477 * provides an upper bound on the number of threads
478 * which will access the socket.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 */
480 svc_sock_setbufsize(svsk->sk_sock,
NeilBrownc6b0a9f2006-10-06 00:44:05 -0700481 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
482 (serv->sv_nrthreads+3) * serv->sv_max_mesg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483
Tom Tucker02fc6c32007-12-30 21:07:48 -0600484 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
NeilBrown05ed6902007-05-09 02:34:55 -0700485 skb = NULL;
486 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
487 0, 0, MSG_PEEK | MSG_DONTWAIT);
488 if (err >= 0)
489 skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err);
490
491 if (skb == NULL) {
492 if (err != -EAGAIN) {
493 /* possibly an icmp error */
494 dprintk("svc: recvfrom returned error %d\n", -err);
Tom Tucker02fc6c32007-12-30 21:07:48 -0600495 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 }
Tom Tuckera6046f72007-12-30 21:08:01 -0600497 svc_xprt_received(&svsk->sk_xprt);
NeilBrown05ed6902007-05-09 02:34:55 -0700498 return -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 }
Tom Tucker9dbc2402007-12-30 21:08:12 -0600500 len = svc_addr_len(svc_addr(rqstp));
Chuck Leverabc5c442009-04-23 19:31:25 -0400501 if (len == 0)
502 return -EAFNOSUPPORT;
Tom Tucker9dbc2402007-12-30 21:08:12 -0600503 rqstp->rq_addrlen = len;
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -0700504 if (skb->tstamp.tv64 == 0) {
505 skb->tstamp = ktime_get_real();
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -0800506 /* Don't enable netstamp, sunrpc doesn't
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 need that much accuracy */
508 }
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -0700509 svsk->sk_sk->sk_stamp = skb->tstamp;
Tom Tucker02fc6c32007-12-30 21:07:48 -0600510 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511
512 /*
513 * Maybe more packets - kick another thread ASAP.
514 */
Tom Tuckera6046f72007-12-30 21:08:01 -0600515 svc_xprt_received(&svsk->sk_xprt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516
517 len = skb->len - sizeof(struct udphdr);
518 rqstp->rq_arg.len = len;
519
Chuck Lever95756482007-02-12 00:53:38 -0800520 rqstp->rq_prot = IPPROTO_UDP;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521
NeilBrown7a37f572007-03-06 01:42:21 -0800522 if (cmh->cmsg_level != IPPROTO_IP ||
523 cmh->cmsg_type != IP_PKTINFO) {
524 if (net_ratelimit())
525 printk("rpcsvc: received unknown control message:"
526 "%d/%d\n",
527 cmh->cmsg_level, cmh->cmsg_type);
528 skb_free_datagram(svsk->sk_sk, skb);
529 return 0;
530 }
531 svc_udp_get_dest_address(rqstp, cmh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532
533 if (skb_is_nonlinear(skb)) {
534 /* we have to copy */
535 local_bh_disable();
536 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
537 local_bh_enable();
538 /* checksum error */
539 skb_free_datagram(svsk->sk_sk, skb);
540 return 0;
541 }
542 local_bh_enable();
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -0800543 skb_free_datagram(svsk->sk_sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544 } else {
545 /* we can use it in-place */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600546 rqstp->rq_arg.head[0].iov_base = skb->data +
547 sizeof(struct udphdr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 rqstp->rq_arg.head[0].iov_len = len;
Herbert Xufb286bb2005-11-10 13:01:24 -0800549 if (skb_checksum_complete(skb)) {
550 skb_free_datagram(svsk->sk_sk, skb);
551 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 }
Tom Tucker5148bf42007-12-30 21:07:25 -0600553 rqstp->rq_xprt_ctxt = skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554 }
555
556 rqstp->rq_arg.page_base = 0;
557 if (len <= rqstp->rq_arg.head[0].iov_len) {
558 rqstp->rq_arg.head[0].iov_len = len;
559 rqstp->rq_arg.page_len = 0;
NeilBrown44524352006-10-04 02:15:46 -0700560 rqstp->rq_respages = rqstp->rq_pages+1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561 } else {
562 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
NeilBrown44524352006-10-04 02:15:46 -0700563 rqstp->rq_respages = rqstp->rq_pages + 1 +
Ilpo Järvinen172589c2007-08-28 15:50:33 -0700564 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 }
566
567 if (serv->sv_stats)
568 serv->sv_stats->netudpcnt++;
569
570 return len;
571}
572
573static int
574svc_udp_sendto(struct svc_rqst *rqstp)
575{
576 int error;
577
578 error = svc_sendto(rqstp, &rqstp->rq_res);
579 if (error == -ECONNREFUSED)
580 /* ICMP error on earlier request. */
581 error = svc_sendto(rqstp, &rqstp->rq_res);
582
583 return error;
584}
585
Tom Tuckere831fe62007-12-30 21:07:29 -0600586static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
587{
588}
589
Tom Tucker323bee32007-12-30 21:07:31 -0600590static int svc_udp_has_wspace(struct svc_xprt *xprt)
591{
592 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600593 struct svc_serv *serv = xprt->xpt_server;
Tom Tucker323bee32007-12-30 21:07:31 -0600594 unsigned long required;
595
596 /*
597 * Set the SOCK_NOSPACE flag before checking the available
598 * sock space.
599 */
600 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
Tom Tucker7a90e8c2007-12-30 21:07:55 -0600601 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
Tom Tucker323bee32007-12-30 21:07:31 -0600602 if (required*2 > sock_wspace(svsk->sk_sk))
603 return 0;
604 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
605 return 1;
606}
607
Tom Tucker38a417c2007-12-30 21:07:36 -0600608static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
609{
610 BUG();
611 return NULL;
612}
613
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600614static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
615 struct sockaddr *sa, int salen,
616 int flags)
617{
618 return svc_create_socket(serv, IPPROTO_UDP, sa, salen, flags);
619}
620
Tom Tucker360d87382007-12-30 21:07:17 -0600621static struct svc_xprt_ops svc_udp_ops = {
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600622 .xpo_create = svc_udp_create,
Tom Tucker5d137992007-12-30 21:07:23 -0600623 .xpo_recvfrom = svc_udp_recvfrom,
624 .xpo_sendto = svc_udp_sendto,
Tom Tucker5148bf42007-12-30 21:07:25 -0600625 .xpo_release_rqst = svc_release_skb,
Tom Tucker755ccea2007-12-30 21:07:27 -0600626 .xpo_detach = svc_sock_detach,
627 .xpo_free = svc_sock_free,
Tom Tuckere831fe62007-12-30 21:07:29 -0600628 .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
Tom Tucker323bee32007-12-30 21:07:31 -0600629 .xpo_has_wspace = svc_udp_has_wspace,
Tom Tucker38a417c2007-12-30 21:07:36 -0600630 .xpo_accept = svc_udp_accept,
Tom Tucker360d87382007-12-30 21:07:17 -0600631};
632
633static struct svc_xprt_class svc_udp_class = {
634 .xcl_name = "udp",
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600635 .xcl_owner = THIS_MODULE,
Tom Tucker360d87382007-12-30 21:07:17 -0600636 .xcl_ops = &svc_udp_ops,
Tom Tucker49023152007-12-30 21:07:21 -0600637 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
Tom Tucker360d87382007-12-30 21:07:17 -0600638};
639
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600640static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641{
NeilBrown7a37f572007-03-06 01:42:21 -0800642 int one = 1;
643 mm_segment_t oldfs;
644
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600645 svc_xprt_init(&svc_udp_class, &svsk->sk_xprt, serv);
Tom Tuckerdef13d72007-12-30 21:08:08 -0600646 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 svsk->sk_sk->sk_data_ready = svc_udp_data_ready;
648 svsk->sk_sk->sk_write_space = svc_write_space;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649
650 /* initialise setting must have enough space to
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -0800651 * receive and respond to one request.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 * svc_udp_recvfrom will re-adjust if necessary
653 */
654 svc_sock_setbufsize(svsk->sk_sock,
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600655 3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
656 3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657
Tom Tucker0f0257e2007-12-30 21:08:27 -0600658 /* data might have come in before data_ready set up */
659 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
Tom Tucker02fc6c32007-12-30 21:07:48 -0600660 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
NeilBrown7a37f572007-03-06 01:42:21 -0800661
662 oldfs = get_fs();
663 set_fs(KERNEL_DS);
664 /* make sure we get destination address info */
665 svsk->sk_sock->ops->setsockopt(svsk->sk_sock, IPPROTO_IP, IP_PKTINFO,
666 (char __user *)&one, sizeof(one));
667 set_fs(oldfs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668}
669
670/*
671 * A data_ready event on a listening socket means there's a connection
672 * pending. Do not use state_change as a substitute for it.
673 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600674static void svc_tcp_listen_data_ready(struct sock *sk, int count_unused)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675{
Neil Brown939bb7e2005-09-13 01:25:39 -0700676 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677
678 dprintk("svc: socket %p TCP (listen) state change %d\n",
Neil Brown939bb7e2005-09-13 01:25:39 -0700679 sk, sk->sk_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680
Neil Brown939bb7e2005-09-13 01:25:39 -0700681 /*
682 * This callback may called twice when a new connection
683 * is established as a child socket inherits everything
684 * from a parent LISTEN socket.
685 * 1) data_ready method of the parent socket will be called
686 * when one of child sockets become ESTABLISHED.
687 * 2) data_ready method of the child socket may be called
688 * when it receives data before the socket is accepted.
689 * In case of 2, we should ignore it silently.
690 */
691 if (sk->sk_state == TCP_LISTEN) {
692 if (svsk) {
Tom Tucker02fc6c32007-12-30 21:07:48 -0600693 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
Tom Tuckerf6150c32007-12-30 21:07:57 -0600694 svc_xprt_enqueue(&svsk->sk_xprt);
Neil Brown939bb7e2005-09-13 01:25:39 -0700695 } else
696 printk("svc: socket %p: no user data\n", sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 }
Neil Brown939bb7e2005-09-13 01:25:39 -0700698
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
700 wake_up_interruptible_all(sk->sk_sleep);
701}
702
703/*
704 * A state change on a connected socket means it's dying or dead.
705 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600706static void svc_tcp_state_change(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707{
Neil Brown939bb7e2005-09-13 01:25:39 -0700708 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709
710 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
Neil Brown939bb7e2005-09-13 01:25:39 -0700711 sk, sk->sk_state, sk->sk_user_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712
Neil Brown939bb7e2005-09-13 01:25:39 -0700713 if (!svsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 printk("svc: socket %p: no user data\n", sk);
Neil Brown939bb7e2005-09-13 01:25:39 -0700715 else {
Tom Tucker02fc6c32007-12-30 21:07:48 -0600716 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
Tom Tuckerf6150c32007-12-30 21:07:57 -0600717 svc_xprt_enqueue(&svsk->sk_xprt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
720 wake_up_interruptible_all(sk->sk_sleep);
721}
722
Tom Tucker0f0257e2007-12-30 21:08:27 -0600723static void svc_tcp_data_ready(struct sock *sk, int count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724{
Neil Brown939bb7e2005-09-13 01:25:39 -0700725 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726
727 dprintk("svc: socket %p TCP data ready (svsk %p)\n",
Neil Brown939bb7e2005-09-13 01:25:39 -0700728 sk, sk->sk_user_data);
729 if (svsk) {
Tom Tucker02fc6c32007-12-30 21:07:48 -0600730 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
Tom Tuckerf6150c32007-12-30 21:07:57 -0600731 svc_xprt_enqueue(&svsk->sk_xprt);
Neil Brown939bb7e2005-09-13 01:25:39 -0700732 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
734 wake_up_interruptible(sk->sk_sleep);
735}
736
737/*
738 * Accept a TCP connection
739 */
Tom Tucker38a417c2007-12-30 21:07:36 -0600740static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741{
Tom Tucker38a417c2007-12-30 21:07:36 -0600742 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
akpm@linux-foundation.orgcdd88b92007-02-12 00:53:38 -0800743 struct sockaddr_storage addr;
744 struct sockaddr *sin = (struct sockaddr *) &addr;
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600745 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 struct socket *sock = svsk->sk_sock;
747 struct socket *newsock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 struct svc_sock *newsvsk;
749 int err, slen;
Pavel Emelyanov5216a8e2008-02-21 10:57:45 +0300750 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751
752 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
753 if (!sock)
Tom Tucker38a417c2007-12-30 21:07:36 -0600754 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755
Tom Tucker02fc6c32007-12-30 21:07:48 -0600756 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
Sridhar Samudralae6242e92006-08-07 20:58:01 -0700757 err = kernel_accept(sock, &newsock, O_NONBLOCK);
758 if (err < 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 if (err == -ENOMEM)
760 printk(KERN_WARNING "%s: no more sockets!\n",
761 serv->sv_name);
Sridhar Samudralae6242e92006-08-07 20:58:01 -0700762 else if (err != -EAGAIN && net_ratelimit())
763 printk(KERN_WARNING "%s: accept failed (err %d)!\n",
764 serv->sv_name, -err);
Tom Tucker38a417c2007-12-30 21:07:36 -0600765 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 }
Tom Tucker02fc6c32007-12-30 21:07:48 -0600767 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768
akpm@linux-foundation.orgcdd88b92007-02-12 00:53:38 -0800769 err = kernel_getpeername(newsock, sin, &slen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 if (err < 0) {
771 if (net_ratelimit())
772 printk(KERN_WARNING "%s: peername failed (err %d)!\n",
773 serv->sv_name, -err);
774 goto failed; /* aborted connection or whatever */
775 }
776
777 /* Ideally, we would want to reject connections from unauthorized
Chuck Leverad06e4b2007-02-12 00:53:32 -0800778 * hosts here, but when we get encryption, the IP of the host won't
779 * tell us anything. For now just warn about unpriv connections.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 */
akpm@linux-foundation.orgcdd88b92007-02-12 00:53:38 -0800781 if (!svc_port_is_privileged(sin)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 dprintk(KERN_WARNING
Chuck Leverad06e4b2007-02-12 00:53:32 -0800783 "%s: connect from unprivileged port: %s\n",
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -0800784 serv->sv_name,
akpm@linux-foundation.orgcdd88b92007-02-12 00:53:38 -0800785 __svc_print_addr(sin, buf, sizeof(buf)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 }
Chuck Leverad06e4b2007-02-12 00:53:32 -0800787 dprintk("%s: connect from %s\n", serv->sv_name,
akpm@linux-foundation.orgcdd88b92007-02-12 00:53:38 -0800788 __svc_print_addr(sin, buf, sizeof(buf)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789
790 /* make sure that a write doesn't block forever when
791 * low on memory
792 */
793 newsock->sk->sk_sndtimeo = HZ*30;
794
Chuck Lever6b174332007-02-12 00:53:28 -0800795 if (!(newsvsk = svc_setup_socket(serv, newsock, &err,
796 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY))))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797 goto failed;
Tom Tucker9dbc2402007-12-30 21:08:12 -0600798 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
Frank van Maarseveena9747692007-07-09 22:21:39 +0200799 err = kernel_getsockname(newsock, sin, &slen);
800 if (unlikely(err < 0)) {
801 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
802 slen = offsetof(struct sockaddr, sa_data);
803 }
Tom Tucker9dbc2402007-12-30 21:08:12 -0600804 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
Chuck Lever067d7812007-02-12 00:53:30 -0800805
Tom Tuckerf9f3cc42007-12-30 21:07:40 -0600806 if (serv->sv_stats)
807 serv->sv_stats->nettcpconn++;
808
809 return &newsvsk->sk_xprt;
810
811failed:
812 sock_release(newsock);
813 return NULL;
814}
815
816/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 * Receive data from a TCP socket.
818 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600819static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820{
Tom Tucker57b1d3b2007-12-30 21:08:22 -0600821 struct svc_sock *svsk =
822 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600823 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824 int len;
NeilBrown3cc03b12006-10-04 02:15:47 -0700825 struct kvec *vec;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 int pnum, vlen;
827
828 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
Tom Tucker02fc6c32007-12-30 21:07:48 -0600829 svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
830 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
831 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832
Tom Tucker02fc6c32007-12-30 21:07:48 -0600833 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834
835 /* Receive data. If we haven't got the record length yet, get
836 * the next four bytes. Otherwise try to gobble up as much as
837 * possible up to the complete record length.
838 */
Chuck Leverc0401ea2008-04-14 12:27:30 -0400839 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
840 int want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 struct kvec iov;
842
843 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
844 iov.iov_len = want;
845 if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
846 goto error;
847 svsk->sk_tcplen += len;
848
849 if (len < want) {
Chuck Leverc0401ea2008-04-14 12:27:30 -0400850 dprintk("svc: short recvfrom while reading record "
851 "length (%d of %d)\n", len, want);
Tom Tuckera6046f72007-12-30 21:08:01 -0600852 svc_xprt_received(&svsk->sk_xprt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 return -EAGAIN; /* record header not complete */
854 }
855
856 svsk->sk_reclen = ntohl(svsk->sk_reclen);
Chuck Leverc0401ea2008-04-14 12:27:30 -0400857 if (!(svsk->sk_reclen & RPC_LAST_STREAM_FRAGMENT)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 /* FIXME: technically, a record can be fragmented,
859 * and non-terminal fragments will not have the top
860 * bit set in the fragment length header.
861 * But apparently no known nfs clients send fragmented
862 * records. */
NeilBrown34e9a632007-01-29 13:19:52 -0800863 if (net_ratelimit())
Chuck Leverc0401ea2008-04-14 12:27:30 -0400864 printk(KERN_NOTICE "RPC: multiple fragments "
865 "per record not supported\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 goto err_delete;
867 }
Chuck Leverc0401ea2008-04-14 12:27:30 -0400868 svsk->sk_reclen &= RPC_FRAGMENT_SIZE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen);
NeilBrownc6b0a9f2006-10-06 00:44:05 -0700870 if (svsk->sk_reclen > serv->sv_max_mesg) {
NeilBrown34e9a632007-01-29 13:19:52 -0800871 if (net_ratelimit())
Chuck Leverc0401ea2008-04-14 12:27:30 -0400872 printk(KERN_NOTICE "RPC: "
873 "fragment too large: 0x%08lx\n",
874 (unsigned long)svsk->sk_reclen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 goto err_delete;
876 }
877 }
878
879 /* Check whether enough data is available */
880 len = svc_recv_available(svsk);
881 if (len < 0)
882 goto error;
883
884 if (len < svsk->sk_reclen) {
885 dprintk("svc: incomplete TCP record (%d of %d)\n",
886 len, svsk->sk_reclen);
Tom Tuckera6046f72007-12-30 21:08:01 -0600887 svc_xprt_received(&svsk->sk_xprt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 return -EAGAIN; /* record not complete */
889 }
890 len = svsk->sk_reclen;
Tom Tucker02fc6c32007-12-30 21:07:48 -0600891 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892
NeilBrown3cc03b12006-10-04 02:15:47 -0700893 vec = rqstp->rq_vec;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894 vec[0] = rqstp->rq_arg.head[0];
895 vlen = PAGE_SIZE;
896 pnum = 1;
897 while (vlen < len) {
NeilBrown44524352006-10-04 02:15:46 -0700898 vec[pnum].iov_base = page_address(rqstp->rq_pages[pnum]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 vec[pnum].iov_len = PAGE_SIZE;
900 pnum++;
901 vlen += PAGE_SIZE;
902 }
NeilBrown44524352006-10-04 02:15:46 -0700903 rqstp->rq_respages = &rqstp->rq_pages[pnum];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904
905 /* Now receive data */
906 len = svc_recvfrom(rqstp, vec, pnum, len);
907 if (len < 0)
908 goto error;
909
910 dprintk("svc: TCP complete record (%d bytes)\n", len);
911 rqstp->rq_arg.len = len;
912 rqstp->rq_arg.page_base = 0;
913 if (len <= rqstp->rq_arg.head[0].iov_len) {
914 rqstp->rq_arg.head[0].iov_len = len;
915 rqstp->rq_arg.page_len = 0;
916 } else {
917 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
918 }
919
Tom Tucker5148bf42007-12-30 21:07:25 -0600920 rqstp->rq_xprt_ctxt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 rqstp->rq_prot = IPPROTO_TCP;
922
923 /* Reset TCP read info */
924 svsk->sk_reclen = 0;
925 svsk->sk_tcplen = 0;
926
Tom Tucker9dbc2402007-12-30 21:08:12 -0600927 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
Tom Tuckera6046f72007-12-30 21:08:01 -0600928 svc_xprt_received(&svsk->sk_xprt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929 if (serv->sv_stats)
930 serv->sv_stats->nettcpcnt++;
931
932 return len;
933
934 err_delete:
Tom Tucker02fc6c32007-12-30 21:07:48 -0600935 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 return -EAGAIN;
937
938 error:
939 if (len == -EAGAIN) {
940 dprintk("RPC: TCP recvfrom got EAGAIN\n");
Tom Tuckera6046f72007-12-30 21:08:01 -0600941 svc_xprt_received(&svsk->sk_xprt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 } else {
943 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600944 svsk->sk_xprt.xpt_server->sv_name, -len);
Olaf Kirch93fbf1a2006-01-06 00:19:56 -0800945 goto err_delete;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946 }
947
948 return len;
949}
950
951/*
952 * Send out data on TCP socket.
953 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600954static int svc_tcp_sendto(struct svc_rqst *rqstp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955{
956 struct xdr_buf *xbufp = &rqstp->rq_res;
957 int sent;
Alexey Dobriyand8ed0292006-09-26 22:29:38 -0700958 __be32 reclen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959
960 /* Set up the first element of the reply kvec.
961 * Any other kvecs that may be in use have been taken
962 * care of by the server implementation itself.
963 */
964 reclen = htonl(0x80000000|((xbufp->len ) - 4));
965 memcpy(xbufp->head[0].iov_base, &reclen, 4);
966
Tom Tucker57b1d3b2007-12-30 21:08:22 -0600967 if (test_bit(XPT_DEAD, &rqstp->rq_xprt->xpt_flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 return -ENOTCONN;
969
970 sent = svc_sendto(rqstp, &rqstp->rq_res);
971 if (sent != xbufp->len) {
Tom Tucker0f0257e2007-12-30 21:08:27 -0600972 printk(KERN_NOTICE
973 "rpc-srv/tcp: %s: %s %d when sending %d bytes "
974 "- shutting down socket\n",
Tom Tucker57b1d3b2007-12-30 21:08:22 -0600975 rqstp->rq_xprt->xpt_server->sv_name,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976 (sent<0)?"got error":"sent only",
977 sent, xbufp->len);
Tom Tucker57b1d3b2007-12-30 21:08:22 -0600978 set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
Tom Tuckerf6150c32007-12-30 21:07:57 -0600979 svc_xprt_enqueue(rqstp->rq_xprt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980 sent = -EAGAIN;
981 }
982 return sent;
983}
984
Tom Tuckere831fe62007-12-30 21:07:29 -0600985/*
986 * Setup response header. TCP has a 4B record length field.
987 */
988static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
989{
990 struct kvec *resv = &rqstp->rq_res.head[0];
991
992 /* tcp needs a space for the record length... */
993 svc_putnl(resv, 0);
994}
995
Tom Tucker323bee32007-12-30 21:07:31 -0600996static int svc_tcp_has_wspace(struct svc_xprt *xprt)
997{
Tom Tucker57b1d3b2007-12-30 21:08:22 -0600998 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600999 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
Tom Tucker323bee32007-12-30 21:07:31 -06001000 int required;
1001 int wspace;
1002
1003 /*
1004 * Set the SOCK_NOSPACE flag before checking the available
1005 * sock space.
1006 */
1007 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
Tom Tucker7a90e8c2007-12-30 21:07:55 -06001008 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
Tom Tucker323bee32007-12-30 21:07:31 -06001009 wspace = sk_stream_wspace(svsk->sk_sk);
1010
1011 if (wspace < sk_stream_min_wspace(svsk->sk_sk))
1012 return 0;
1013 if (required * 2 > wspace)
1014 return 0;
1015
1016 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
1017 return 1;
1018}
1019
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001020static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1021 struct sockaddr *sa, int salen,
1022 int flags)
1023{
1024 return svc_create_socket(serv, IPPROTO_TCP, sa, salen, flags);
1025}
1026
Tom Tucker360d87382007-12-30 21:07:17 -06001027static struct svc_xprt_ops svc_tcp_ops = {
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001028 .xpo_create = svc_tcp_create,
Tom Tucker5d137992007-12-30 21:07:23 -06001029 .xpo_recvfrom = svc_tcp_recvfrom,
1030 .xpo_sendto = svc_tcp_sendto,
Tom Tucker5148bf42007-12-30 21:07:25 -06001031 .xpo_release_rqst = svc_release_skb,
Trond Myklebust69b6ba32008-12-23 16:30:11 -05001032 .xpo_detach = svc_tcp_sock_detach,
Tom Tucker755ccea2007-12-30 21:07:27 -06001033 .xpo_free = svc_sock_free,
Tom Tuckere831fe62007-12-30 21:07:29 -06001034 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
Tom Tucker323bee32007-12-30 21:07:31 -06001035 .xpo_has_wspace = svc_tcp_has_wspace,
Tom Tucker38a417c2007-12-30 21:07:36 -06001036 .xpo_accept = svc_tcp_accept,
Tom Tucker360d87382007-12-30 21:07:17 -06001037};
1038
1039static struct svc_xprt_class svc_tcp_class = {
1040 .xcl_name = "tcp",
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001041 .xcl_owner = THIS_MODULE,
Tom Tucker360d87382007-12-30 21:07:17 -06001042 .xcl_ops = &svc_tcp_ops,
Tom Tucker49023152007-12-30 21:07:21 -06001043 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
Tom Tucker360d87382007-12-30 21:07:17 -06001044};
1045
1046void svc_init_xprt_sock(void)
1047{
1048 svc_reg_xprt_class(&svc_tcp_class);
1049 svc_reg_xprt_class(&svc_udp_class);
1050}
1051
1052void svc_cleanup_xprt_sock(void)
1053{
1054 svc_unreg_xprt_class(&svc_tcp_class);
1055 svc_unreg_xprt_class(&svc_udp_class);
1056}
1057
Tom Tuckerbb5cf162007-12-30 21:07:50 -06001058static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059{
1060 struct sock *sk = svsk->sk_sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061
Tom Tuckerbb5cf162007-12-30 21:07:50 -06001062 svc_xprt_init(&svc_tcp_class, &svsk->sk_xprt, serv);
Tom Tuckerdef13d72007-12-30 21:08:08 -06001063 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 if (sk->sk_state == TCP_LISTEN) {
1065 dprintk("setting up TCP socket for listening\n");
Tom Tucker02fc6c32007-12-30 21:07:48 -06001066 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 sk->sk_data_ready = svc_tcp_listen_data_ready;
Tom Tucker02fc6c32007-12-30 21:07:48 -06001068 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069 } else {
1070 dprintk("setting up TCP socket for reading\n");
1071 sk->sk_state_change = svc_tcp_state_change;
1072 sk->sk_data_ready = svc_tcp_data_ready;
1073 sk->sk_write_space = svc_write_space;
1074
1075 svsk->sk_reclen = 0;
1076 svsk->sk_tcplen = 0;
1077
Chuck Leverb7872fe2008-04-14 12:27:01 -04001078 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079
Tom Tucker02fc6c32007-12-30 21:07:48 -06001080 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -08001081 if (sk->sk_state != TCP_ESTABLISHED)
Tom Tucker02fc6c32007-12-30 21:07:48 -06001082 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083 }
1084}
1085
Tom Tucker0f0257e2007-12-30 21:08:27 -06001086void svc_sock_update_bufs(struct svc_serv *serv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087{
1088 /*
1089 * The number of server threads has changed. Update
1090 * rcvbuf and sndbuf accordingly on all sockets
1091 */
1092 struct list_head *le;
1093
1094 spin_lock_bh(&serv->sv_lock);
1095 list_for_each(le, &serv->sv_permsocks) {
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -08001096 struct svc_sock *svsk =
Tom Tucker7a182082007-12-30 21:07:53 -06001097 list_entry(le, struct svc_sock, sk_xprt.xpt_list);
Tom Tucker02fc6c32007-12-30 21:07:48 -06001098 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 }
1100 list_for_each(le, &serv->sv_tempsocks) {
1101 struct svc_sock *svsk =
Tom Tucker7a182082007-12-30 21:07:53 -06001102 list_entry(le, struct svc_sock, sk_xprt.xpt_list);
Tom Tucker02fc6c32007-12-30 21:07:48 -06001103 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104 }
1105 spin_unlock_bh(&serv->sv_lock);
1106}
Trond Myklebust24c37672008-12-23 16:30:12 -05001107EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108
1109/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 * Initialize socket for RPC use and create svc_sock struct
1111 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1112 */
Chuck Lever6b174332007-02-12 00:53:28 -08001113static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1114 struct socket *sock,
1115 int *errp, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116{
1117 struct svc_sock *svsk;
1118 struct sock *inet;
Chuck Lever6b174332007-02-12 00:53:28 -08001119 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120
1121 dprintk("svc: svc_setup_socket %p\n", sock);
Panagiotis Issaris0da974f2006-07-21 14:51:30 -07001122 if (!(svsk = kzalloc(sizeof(*svsk), GFP_KERNEL))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 *errp = -ENOMEM;
1124 return NULL;
1125 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126
1127 inet = sock->sk;
1128
1129 /* Register socket with portmapper */
1130 if (*errp >= 0 && pmap_register)
Chuck Leverbaf01ca2009-03-18 20:46:13 -04001131 *errp = svc_register(serv, inet->sk_family, inet->sk_protocol,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 ntohs(inet_sk(inet)->sport));
1133
1134 if (*errp < 0) {
1135 kfree(svsk);
1136 return NULL;
1137 }
1138
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139 inet->sk_user_data = svsk;
1140 svsk->sk_sock = sock;
1141 svsk->sk_sk = inet;
1142 svsk->sk_ostate = inet->sk_state_change;
1143 svsk->sk_odata = inet->sk_data_ready;
1144 svsk->sk_owspace = inet->sk_write_space;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145
1146 /* Initialize the socket */
1147 if (sock->type == SOCK_DGRAM)
Tom Tuckerbb5cf162007-12-30 21:07:50 -06001148 svc_udp_init(svsk, serv);
Olga Kornievskaia47a14ef2008-10-21 14:13:47 -04001149 else {
1150 /* initialise setting must have enough space to
1151 * receive and respond to one request.
1152 */
1153 svc_sock_setbufsize(svsk->sk_sock, 4 * serv->sv_max_mesg,
1154 4 * serv->sv_max_mesg);
Tom Tuckerbb5cf162007-12-30 21:07:50 -06001155 svc_tcp_init(svsk, serv);
Olga Kornievskaia47a14ef2008-10-21 14:13:47 -04001156 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1159 svsk, svsk->sk_sk);
1160
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 return svsk;
1162}
1163
Chuck Leverbfba9ab2009-04-23 19:32:33 -04001164/**
1165 * svc_addsock - add a listener socket to an RPC service
1166 * @serv: pointer to RPC service to which to add a new listener
1167 * @fd: file descriptor of the new listener
1168 * @name_return: pointer to buffer to fill in with name of listener
1169 * @len: size of the buffer
1170 *
1171 * Fills in socket name and returns positive length of name if successful.
1172 * Name is terminated with '\n'. On error, returns a negative errno
1173 * value.
1174 */
1175int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1176 const size_t len)
NeilBrownb41b66d2006-10-02 02:17:48 -07001177{
1178 int err = 0;
1179 struct socket *so = sockfd_lookup(fd, &err);
1180 struct svc_sock *svsk = NULL;
1181
1182 if (!so)
1183 return err;
1184 if (so->sk->sk_family != AF_INET)
1185 err = -EAFNOSUPPORT;
1186 else if (so->sk->sk_protocol != IPPROTO_TCP &&
1187 so->sk->sk_protocol != IPPROTO_UDP)
1188 err = -EPROTONOSUPPORT;
1189 else if (so->state > SS_UNCONNECTED)
1190 err = -EISCONN;
1191 else {
Tom Tucker2da2c212008-11-23 09:58:08 -06001192 if (!try_module_get(THIS_MODULE))
1193 err = -ENOENT;
1194 else
1195 svsk = svc_setup_socket(serv, so, &err,
1196 SVC_SOCK_DEFAULTS);
NeilBrowne79eff12007-02-12 00:53:30 -08001197 if (svsk) {
Tom Tucker9dbc2402007-12-30 21:08:12 -06001198 struct sockaddr_storage addr;
1199 struct sockaddr *sin = (struct sockaddr *)&addr;
1200 int salen;
1201 if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0)
1202 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
Tom Tucker4e5caaa2007-12-30 21:08:20 -06001203 clear_bit(XPT_TEMP, &svsk->sk_xprt.xpt_flags);
1204 spin_lock_bh(&serv->sv_lock);
1205 list_add(&svsk->sk_xprt.xpt_list, &serv->sv_permsocks);
1206 spin_unlock_bh(&serv->sv_lock);
Tom Tuckera6046f72007-12-30 21:08:01 -06001207 svc_xprt_received(&svsk->sk_xprt);
NeilBrownb41b66d2006-10-02 02:17:48 -07001208 err = 0;
Tom Tucker2da2c212008-11-23 09:58:08 -06001209 } else
1210 module_put(THIS_MODULE);
NeilBrownb41b66d2006-10-02 02:17:48 -07001211 }
1212 if (err) {
1213 sockfd_put(so);
1214 return err;
1215 }
Chuck Levere7942b92009-04-23 19:32:48 -04001216 return svc_one_sock_name(svsk, name_return, len);
NeilBrownb41b66d2006-10-02 02:17:48 -07001217}
1218EXPORT_SYMBOL_GPL(svc_addsock);
1219
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220/*
1221 * Create socket for RPC service.
1222 */
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001223static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1224 int protocol,
1225 struct sockaddr *sin, int len,
1226 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227{
1228 struct svc_sock *svsk;
1229 struct socket *sock;
1230 int error;
1231 int type;
Tom Tucker9dbc2402007-12-30 21:08:12 -06001232 struct sockaddr_storage addr;
1233 struct sockaddr *newsin = (struct sockaddr *)&addr;
1234 int newlen;
Trond Myklebustc69da772009-03-30 18:59:17 -04001235 int family;
1236 int val;
Pavel Emelyanov5216a8e2008-02-21 10:57:45 +03001237 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238
Chuck Leverad06e4b2007-02-12 00:53:32 -08001239 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1240 serv->sv_program->pg_name, protocol,
Chuck Lever77f1f672007-02-12 00:53:39 -08001241 __svc_print_addr(sin, buf, sizeof(buf)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242
1243 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1244 printk(KERN_WARNING "svc: only UDP and TCP "
1245 "sockets supported\n");
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001246 return ERR_PTR(-EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248
Trond Myklebustc69da772009-03-30 18:59:17 -04001249 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1250 switch (sin->sa_family) {
1251 case AF_INET6:
1252 family = PF_INET6;
1253 break;
1254 case AF_INET:
1255 family = PF_INET;
1256 break;
1257 default:
1258 return ERR_PTR(-EINVAL);
1259 }
1260
1261 error = sock_create_kern(family, type, protocol, &sock);
Chuck Lever77f1f672007-02-12 00:53:39 -08001262 if (error < 0)
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001263 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264
Peter Zijlstraed075362006-12-06 20:35:24 -08001265 svc_reclassify_socket(sock);
1266
Trond Myklebustc69da772009-03-30 18:59:17 -04001267 /*
1268 * If this is an PF_INET6 listener, we want to avoid
1269 * getting requests from IPv4 remotes. Those should
1270 * be shunted to a PF_INET listener via rpcbind.
1271 */
1272 val = 1;
1273 if (family == PF_INET6)
1274 kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
1275 (char *)&val, sizeof(val));
1276
Eric Sesterhenn18114742006-09-28 14:37:07 -07001277 if (type == SOCK_STREAM)
Chuck Lever77f1f672007-02-12 00:53:39 -08001278 sock->sk->sk_reuse = 1; /* allow address reuse */
1279 error = kernel_bind(sock, sin, len);
Eric Sesterhenn18114742006-09-28 14:37:07 -07001280 if (error < 0)
1281 goto bummer;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282
Tom Tucker9dbc2402007-12-30 21:08:12 -06001283 newlen = len;
1284 error = kernel_getsockname(sock, newsin, &newlen);
1285 if (error < 0)
1286 goto bummer;
1287
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288 if (protocol == IPPROTO_TCP) {
Sridhar Samudralae6242e92006-08-07 20:58:01 -07001289 if ((error = kernel_listen(sock, 64)) < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290 goto bummer;
1291 }
1292
NeilBrowne79eff12007-02-12 00:53:30 -08001293 if ((svsk = svc_setup_socket(serv, sock, &error, flags)) != NULL) {
Tom Tucker9dbc2402007-12-30 21:08:12 -06001294 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001295 return (struct svc_xprt *)svsk;
NeilBrowne79eff12007-02-12 00:53:30 -08001296 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297
1298bummer:
1299 dprintk("svc: svc_create_socket error = %d\n", -error);
1300 sock_release(sock);
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001301 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302}
1303
1304/*
Tom Tucker755ccea2007-12-30 21:07:27 -06001305 * Detach the svc_sock from the socket so that no
1306 * more callbacks occur.
1307 */
1308static void svc_sock_detach(struct svc_xprt *xprt)
1309{
1310 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1311 struct sock *sk = svsk->sk_sk;
1312
1313 dprintk("svc: svc_sock_detach(%p)\n", svsk);
1314
1315 /* put back the old socket callbacks */
1316 sk->sk_state_change = svsk->sk_ostate;
1317 sk->sk_data_ready = svsk->sk_odata;
1318 sk->sk_write_space = svsk->sk_owspace;
Trond Myklebust69b6ba32008-12-23 16:30:11 -05001319
1320 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1321 wake_up_interruptible(sk->sk_sleep);
1322}
1323
1324/*
1325 * Disconnect the socket, and reset the callbacks
1326 */
1327static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1328{
1329 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1330
1331 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
1332
1333 svc_sock_detach(xprt);
1334
1335 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags))
1336 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
Tom Tucker755ccea2007-12-30 21:07:27 -06001337}
1338
1339/*
1340 * Free the svc_sock's socket resources and the svc_sock itself.
1341 */
1342static void svc_sock_free(struct svc_xprt *xprt)
1343{
1344 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1345 dprintk("svc: svc_sock_free(%p)\n", svsk);
1346
Tom Tucker755ccea2007-12-30 21:07:27 -06001347 if (svsk->sk_sock->file)
1348 sockfd_put(svsk->sk_sock);
1349 else
1350 sock_release(svsk->sk_sock);
1351 kfree(svsk);
1352}