blob: 9d504234af4a664f5f116171b9d6e02825d1ac4f [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 */
243static int one_sock_name(char *buf, struct svc_sock *svsk)
244{
245 int len;
246
247 switch(svsk->sk_sk->sk_family) {
248 case AF_INET:
Harvey Harrison21454aa2008-10-31 00:54:56 -0700249 len = sprintf(buf, "ipv4 %s %pI4 %d\n",
250 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;
255 default:
256 len = sprintf(buf, "*unknown-%d*\n",
257 svsk->sk_sk->sk_family);
258 }
259 return len;
260}
261
262int
NeilBrownb41b66d2006-10-02 02:17:48 -0700263svc_sock_names(char *buf, struct svc_serv *serv, char *toclose)
NeilBrown80212d52006-10-02 02:17:47 -0700264{
NeilBrownb41b66d2006-10-02 02:17:48 -0700265 struct svc_sock *svsk, *closesk = NULL;
NeilBrown80212d52006-10-02 02:17:47 -0700266 int len = 0;
267
268 if (!serv)
269 return 0;
NeilBrownaaf68cf2007-02-08 14:20:30 -0800270 spin_lock_bh(&serv->sv_lock);
Tom Tucker7a182082007-12-30 21:07:53 -0600271 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list) {
NeilBrown80212d52006-10-02 02:17:47 -0700272 int onelen = one_sock_name(buf+len, svsk);
NeilBrownb41b66d2006-10-02 02:17:48 -0700273 if (toclose && strcmp(toclose, buf+len) == 0)
274 closesk = svsk;
275 else
276 len += onelen;
NeilBrown80212d52006-10-02 02:17:47 -0700277 }
NeilBrownaaf68cf2007-02-08 14:20:30 -0800278 spin_unlock_bh(&serv->sv_lock);
NeilBrownb41b66d2006-10-02 02:17:48 -0700279 if (closesk)
NeilBrown5680c442006-10-04 02:15:45 -0700280 /* Should unregister with portmap, but you cannot
281 * unregister just one protocol...
282 */
Tom Tucker7a182082007-12-30 21:07:53 -0600283 svc_close_xprt(&closesk->sk_xprt);
NeilBrown37a03472006-10-04 02:15:44 -0700284 else if (toclose)
285 return -ENOENT;
NeilBrown80212d52006-10-02 02:17:47 -0700286 return len;
287}
Trond Myklebust24c37672008-12-23 16:30:12 -0500288EXPORT_SYMBOL_GPL(svc_sock_names);
NeilBrown80212d52006-10-02 02:17:47 -0700289
290/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 * Check input queue length
292 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600293static int svc_recv_available(struct svc_sock *svsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 struct socket *sock = svsk->sk_sock;
296 int avail, err;
297
Sridhar Samudralae6242e92006-08-07 20:58:01 -0700298 err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299
300 return (err >= 0)? avail : err;
301}
302
303/*
304 * Generic recvfrom routine.
305 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600306static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr,
307 int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308{
Tom Tucker57b1d3b2007-12-30 21:08:22 -0600309 struct svc_sock *svsk =
310 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
Chuck Lever1ba95102007-02-12 00:53:31 -0800311 struct msghdr msg = {
312 .msg_flags = MSG_DONTWAIT,
313 };
314 int len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315
Tom Tucker260c1d12007-12-30 21:08:29 -0600316 rqstp->rq_xprt_hlen = 0;
317
Chuck Lever1ba95102007-02-12 00:53:31 -0800318 len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen,
319 msg.msg_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
Chuck Lever1ba95102007-02-12 00:53:31 -0800322 svsk, iov[0].iov_base, iov[0].iov_len, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 return len;
324}
325
326/*
327 * Set socket snd and rcv buffer lengths
328 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600329static void svc_sock_setbufsize(struct socket *sock, unsigned int snd,
330 unsigned int rcv)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331{
332#if 0
333 mm_segment_t oldfs;
334 oldfs = get_fs(); set_fs(KERNEL_DS);
335 sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
336 (char*)&snd, sizeof(snd));
337 sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
338 (char*)&rcv, sizeof(rcv));
339#else
340 /* sock_setsockopt limits use to sysctl_?mem_max,
341 * which isn't acceptable. Until that is made conditional
342 * on not having CAP_SYS_RESOURCE or similar, we go direct...
343 * DaveM said I could!
344 */
345 lock_sock(sock->sk);
346 sock->sk->sk_sndbuf = snd * 2;
347 sock->sk->sk_rcvbuf = rcv * 2;
J. Bruce Fields7f421832009-05-27 18:51:06 -0400348 sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK|SOCK_RCVBUF_LOCK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 release_sock(sock->sk);
350#endif
351}
352/*
353 * INET callback when data has been received on the socket.
354 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600355static void svc_udp_data_ready(struct sock *sk, int count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356{
Neil Brown939bb7e2005-09-13 01:25:39 -0700357 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358
Neil Brown939bb7e2005-09-13 01:25:39 -0700359 if (svsk) {
360 dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n",
Tom Tucker02fc6c32007-12-30 21:07:48 -0600361 svsk, sk, count,
362 test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
363 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
Tom Tuckerf6150c32007-12-30 21:07:57 -0600364 svc_xprt_enqueue(&svsk->sk_xprt);
Neil Brown939bb7e2005-09-13 01:25:39 -0700365 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
367 wake_up_interruptible(sk->sk_sleep);
368}
369
370/*
371 * INET callback when space is newly available on the socket.
372 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600373static void svc_write_space(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374{
375 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
376
377 if (svsk) {
378 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
Tom Tucker02fc6c32007-12-30 21:07:48 -0600379 svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
Tom Tuckerf6150c32007-12-30 21:07:57 -0600380 svc_xprt_enqueue(&svsk->sk_xprt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 }
382
383 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) {
Neil Brown939bb7e2005-09-13 01:25:39 -0700384 dprintk("RPC svc_write_space: someone sleeping on %p\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 svsk);
386 wake_up_interruptible(sk->sk_sleep);
387 }
388}
389
Tom Tucker9dbc2402007-12-30 21:08:12 -0600390/*
391 * Copy the UDP datagram's destination address to the rqstp structure.
392 * The 'destination' address in this case is the address to which the
393 * peer sent the datagram, i.e. our local address. For multihomed
394 * hosts, this can change from msg to msg. Note that only the IP
395 * address changes, the port number should remain the same.
396 */
397static void svc_udp_get_dest_address(struct svc_rqst *rqstp,
398 struct cmsghdr *cmh)
Chuck Lever95756482007-02-12 00:53:38 -0800399{
Tom Tucker57b1d3b2007-12-30 21:08:22 -0600400 struct svc_sock *svsk =
401 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
402 switch (svsk->sk_sk->sk_family) {
Chuck Lever95756482007-02-12 00:53:38 -0800403 case AF_INET: {
NeilBrown7a37f572007-03-06 01:42:21 -0800404 struct in_pktinfo *pki = CMSG_DATA(cmh);
405 rqstp->rq_daddr.addr.s_addr = pki->ipi_spec_dst.s_addr;
Chuck Lever95756482007-02-12 00:53:38 -0800406 break;
NeilBrown7a37f572007-03-06 01:42:21 -0800407 }
Chuck Lever95756482007-02-12 00:53:38 -0800408 case AF_INET6: {
NeilBrown7a37f572007-03-06 01:42:21 -0800409 struct in6_pktinfo *pki = CMSG_DATA(cmh);
410 ipv6_addr_copy(&rqstp->rq_daddr.addr6, &pki->ipi6_addr);
Chuck Lever95756482007-02-12 00:53:38 -0800411 break;
NeilBrown7a37f572007-03-06 01:42:21 -0800412 }
Chuck Lever95756482007-02-12 00:53:38 -0800413 }
Chuck Lever95756482007-02-12 00:53:38 -0800414}
415
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416/*
417 * Receive a datagram from a UDP socket.
418 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600419static int svc_udp_recvfrom(struct svc_rqst *rqstp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420{
Tom Tucker57b1d3b2007-12-30 21:08:22 -0600421 struct svc_sock *svsk =
422 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600423 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 struct sk_buff *skb;
David S. Millerbc375ea2007-04-12 13:35:59 -0700425 union {
426 struct cmsghdr hdr;
427 long all[SVC_PKTINFO_SPACE / sizeof(long)];
428 } buffer;
429 struct cmsghdr *cmh = &buffer.hdr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 int err, len;
NeilBrown7a37f572007-03-06 01:42:21 -0800431 struct msghdr msg = {
432 .msg_name = svc_addr(rqstp),
433 .msg_control = cmh,
434 .msg_controllen = sizeof(buffer),
435 .msg_flags = MSG_DONTWAIT,
436 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437
Tom Tucker02fc6c32007-12-30 21:07:48 -0600438 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439 /* udp sockets need large rcvbuf as all pending
440 * requests are still in that buffer. sndbuf must
441 * also be large enough that there is enough space
Greg Banks3262c812006-10-02 02:17:58 -0700442 * for one reply per thread. We count all threads
443 * rather than threads in a particular pool, which
444 * provides an upper bound on the number of threads
445 * which will access the socket.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446 */
447 svc_sock_setbufsize(svsk->sk_sock,
NeilBrownc6b0a9f2006-10-06 00:44:05 -0700448 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
449 (serv->sv_nrthreads+3) * serv->sv_max_mesg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450
Tom Tucker02fc6c32007-12-30 21:07:48 -0600451 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
NeilBrown05ed6902007-05-09 02:34:55 -0700452 skb = NULL;
453 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
454 0, 0, MSG_PEEK | MSG_DONTWAIT);
455 if (err >= 0)
456 skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err);
457
458 if (skb == NULL) {
459 if (err != -EAGAIN) {
460 /* possibly an icmp error */
461 dprintk("svc: recvfrom returned error %d\n", -err);
Tom Tucker02fc6c32007-12-30 21:07:48 -0600462 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 }
Tom Tuckera6046f72007-12-30 21:08:01 -0600464 svc_xprt_received(&svsk->sk_xprt);
NeilBrown05ed6902007-05-09 02:34:55 -0700465 return -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 }
Tom Tucker9dbc2402007-12-30 21:08:12 -0600467 len = svc_addr_len(svc_addr(rqstp));
468 if (len < 0)
469 return len;
470 rqstp->rq_addrlen = len;
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -0700471 if (skb->tstamp.tv64 == 0) {
472 skb->tstamp = ktime_get_real();
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -0800473 /* Don't enable netstamp, sunrpc doesn't
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 need that much accuracy */
475 }
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -0700476 svsk->sk_sk->sk_stamp = skb->tstamp;
Tom Tucker02fc6c32007-12-30 21:07:48 -0600477 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478
479 /*
480 * Maybe more packets - kick another thread ASAP.
481 */
Tom Tuckera6046f72007-12-30 21:08:01 -0600482 svc_xprt_received(&svsk->sk_xprt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483
484 len = skb->len - sizeof(struct udphdr);
485 rqstp->rq_arg.len = len;
486
Chuck Lever95756482007-02-12 00:53:38 -0800487 rqstp->rq_prot = IPPROTO_UDP;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488
NeilBrown7a37f572007-03-06 01:42:21 -0800489 if (cmh->cmsg_level != IPPROTO_IP ||
490 cmh->cmsg_type != IP_PKTINFO) {
491 if (net_ratelimit())
492 printk("rpcsvc: received unknown control message:"
493 "%d/%d\n",
494 cmh->cmsg_level, cmh->cmsg_type);
495 skb_free_datagram(svsk->sk_sk, skb);
496 return 0;
497 }
498 svc_udp_get_dest_address(rqstp, cmh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499
500 if (skb_is_nonlinear(skb)) {
501 /* we have to copy */
502 local_bh_disable();
503 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
504 local_bh_enable();
505 /* checksum error */
506 skb_free_datagram(svsk->sk_sk, skb);
507 return 0;
508 }
509 local_bh_enable();
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -0800510 skb_free_datagram(svsk->sk_sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 } else {
512 /* we can use it in-place */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600513 rqstp->rq_arg.head[0].iov_base = skb->data +
514 sizeof(struct udphdr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 rqstp->rq_arg.head[0].iov_len = len;
Herbert Xufb286bb2005-11-10 13:01:24 -0800516 if (skb_checksum_complete(skb)) {
517 skb_free_datagram(svsk->sk_sk, skb);
518 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 }
Tom Tucker5148bf42007-12-30 21:07:25 -0600520 rqstp->rq_xprt_ctxt = skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 }
522
523 rqstp->rq_arg.page_base = 0;
524 if (len <= rqstp->rq_arg.head[0].iov_len) {
525 rqstp->rq_arg.head[0].iov_len = len;
526 rqstp->rq_arg.page_len = 0;
NeilBrown44524352006-10-04 02:15:46 -0700527 rqstp->rq_respages = rqstp->rq_pages+1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 } else {
529 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
NeilBrown44524352006-10-04 02:15:46 -0700530 rqstp->rq_respages = rqstp->rq_pages + 1 +
Ilpo Järvinen172589c2007-08-28 15:50:33 -0700531 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 }
533
534 if (serv->sv_stats)
535 serv->sv_stats->netudpcnt++;
536
537 return len;
538}
539
540static int
541svc_udp_sendto(struct svc_rqst *rqstp)
542{
543 int error;
544
545 error = svc_sendto(rqstp, &rqstp->rq_res);
546 if (error == -ECONNREFUSED)
547 /* ICMP error on earlier request. */
548 error = svc_sendto(rqstp, &rqstp->rq_res);
549
550 return error;
551}
552
Tom Tuckere831fe62007-12-30 21:07:29 -0600553static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
554{
555}
556
Tom Tucker323bee32007-12-30 21:07:31 -0600557static int svc_udp_has_wspace(struct svc_xprt *xprt)
558{
559 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600560 struct svc_serv *serv = xprt->xpt_server;
Tom Tucker323bee32007-12-30 21:07:31 -0600561 unsigned long required;
562
563 /*
564 * Set the SOCK_NOSPACE flag before checking the available
565 * sock space.
566 */
567 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
Tom Tucker7a90e8c2007-12-30 21:07:55 -0600568 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
Tom Tucker323bee32007-12-30 21:07:31 -0600569 if (required*2 > sock_wspace(svsk->sk_sk))
570 return 0;
571 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
572 return 1;
573}
574
Tom Tucker38a417c2007-12-30 21:07:36 -0600575static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
576{
577 BUG();
578 return NULL;
579}
580
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600581static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
582 struct sockaddr *sa, int salen,
583 int flags)
584{
585 return svc_create_socket(serv, IPPROTO_UDP, sa, salen, flags);
586}
587
Tom Tucker360d87382007-12-30 21:07:17 -0600588static struct svc_xprt_ops svc_udp_ops = {
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600589 .xpo_create = svc_udp_create,
Tom Tucker5d137992007-12-30 21:07:23 -0600590 .xpo_recvfrom = svc_udp_recvfrom,
591 .xpo_sendto = svc_udp_sendto,
Tom Tucker5148bf42007-12-30 21:07:25 -0600592 .xpo_release_rqst = svc_release_skb,
Tom Tucker755ccea2007-12-30 21:07:27 -0600593 .xpo_detach = svc_sock_detach,
594 .xpo_free = svc_sock_free,
Tom Tuckere831fe62007-12-30 21:07:29 -0600595 .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
Tom Tucker323bee32007-12-30 21:07:31 -0600596 .xpo_has_wspace = svc_udp_has_wspace,
Tom Tucker38a417c2007-12-30 21:07:36 -0600597 .xpo_accept = svc_udp_accept,
Tom Tucker360d87382007-12-30 21:07:17 -0600598};
599
600static struct svc_xprt_class svc_udp_class = {
601 .xcl_name = "udp",
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600602 .xcl_owner = THIS_MODULE,
Tom Tucker360d87382007-12-30 21:07:17 -0600603 .xcl_ops = &svc_udp_ops,
Tom Tucker49023152007-12-30 21:07:21 -0600604 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
Tom Tucker360d87382007-12-30 21:07:17 -0600605};
606
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600607static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608{
NeilBrown7a37f572007-03-06 01:42:21 -0800609 int one = 1;
610 mm_segment_t oldfs;
611
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600612 svc_xprt_init(&svc_udp_class, &svsk->sk_xprt, serv);
Tom Tuckerdef13d72007-12-30 21:08:08 -0600613 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 svsk->sk_sk->sk_data_ready = svc_udp_data_ready;
615 svsk->sk_sk->sk_write_space = svc_write_space;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616
617 /* initialise setting must have enough space to
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -0800618 * receive and respond to one request.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 * svc_udp_recvfrom will re-adjust if necessary
620 */
621 svc_sock_setbufsize(svsk->sk_sock,
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600622 3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
623 3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624
Tom Tucker0f0257e2007-12-30 21:08:27 -0600625 /* data might have come in before data_ready set up */
626 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
Tom Tucker02fc6c32007-12-30 21:07:48 -0600627 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
NeilBrown7a37f572007-03-06 01:42:21 -0800628
629 oldfs = get_fs();
630 set_fs(KERNEL_DS);
631 /* make sure we get destination address info */
632 svsk->sk_sock->ops->setsockopt(svsk->sk_sock, IPPROTO_IP, IP_PKTINFO,
633 (char __user *)&one, sizeof(one));
634 set_fs(oldfs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635}
636
637/*
638 * A data_ready event on a listening socket means there's a connection
639 * pending. Do not use state_change as a substitute for it.
640 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600641static void svc_tcp_listen_data_ready(struct sock *sk, int count_unused)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642{
Neil Brown939bb7e2005-09-13 01:25:39 -0700643 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644
645 dprintk("svc: socket %p TCP (listen) state change %d\n",
Neil Brown939bb7e2005-09-13 01:25:39 -0700646 sk, sk->sk_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647
Neil Brown939bb7e2005-09-13 01:25:39 -0700648 /*
649 * This callback may called twice when a new connection
650 * is established as a child socket inherits everything
651 * from a parent LISTEN socket.
652 * 1) data_ready method of the parent socket will be called
653 * when one of child sockets become ESTABLISHED.
654 * 2) data_ready method of the child socket may be called
655 * when it receives data before the socket is accepted.
656 * In case of 2, we should ignore it silently.
657 */
658 if (sk->sk_state == TCP_LISTEN) {
659 if (svsk) {
Tom Tucker02fc6c32007-12-30 21:07:48 -0600660 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
Tom Tuckerf6150c32007-12-30 21:07:57 -0600661 svc_xprt_enqueue(&svsk->sk_xprt);
Neil Brown939bb7e2005-09-13 01:25:39 -0700662 } else
663 printk("svc: socket %p: no user data\n", sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 }
Neil Brown939bb7e2005-09-13 01:25:39 -0700665
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
667 wake_up_interruptible_all(sk->sk_sleep);
668}
669
670/*
671 * A state change on a connected socket means it's dying or dead.
672 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600673static void svc_tcp_state_change(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674{
Neil Brown939bb7e2005-09-13 01:25:39 -0700675 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676
677 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
Neil Brown939bb7e2005-09-13 01:25:39 -0700678 sk, sk->sk_state, sk->sk_user_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679
Neil Brown939bb7e2005-09-13 01:25:39 -0700680 if (!svsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 printk("svc: socket %p: no user data\n", sk);
Neil Brown939bb7e2005-09-13 01:25:39 -0700682 else {
Tom Tucker02fc6c32007-12-30 21:07:48 -0600683 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
Tom Tuckerf6150c32007-12-30 21:07:57 -0600684 svc_xprt_enqueue(&svsk->sk_xprt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
687 wake_up_interruptible_all(sk->sk_sleep);
688}
689
Tom Tucker0f0257e2007-12-30 21:08:27 -0600690static void svc_tcp_data_ready(struct sock *sk, int count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691{
Neil Brown939bb7e2005-09-13 01:25:39 -0700692 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693
694 dprintk("svc: socket %p TCP data ready (svsk %p)\n",
Neil Brown939bb7e2005-09-13 01:25:39 -0700695 sk, sk->sk_user_data);
696 if (svsk) {
Tom Tucker02fc6c32007-12-30 21:07:48 -0600697 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
Tom Tuckerf6150c32007-12-30 21:07:57 -0600698 svc_xprt_enqueue(&svsk->sk_xprt);
Neil Brown939bb7e2005-09-13 01:25:39 -0700699 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
701 wake_up_interruptible(sk->sk_sleep);
702}
703
704/*
705 * Accept a TCP connection
706 */
Tom Tucker38a417c2007-12-30 21:07:36 -0600707static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708{
Tom Tucker38a417c2007-12-30 21:07:36 -0600709 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
akpm@linux-foundation.orgcdd88b92007-02-12 00:53:38 -0800710 struct sockaddr_storage addr;
711 struct sockaddr *sin = (struct sockaddr *) &addr;
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600712 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 struct socket *sock = svsk->sk_sock;
714 struct socket *newsock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 struct svc_sock *newsvsk;
716 int err, slen;
Pavel Emelyanov5216a8e2008-02-21 10:57:45 +0300717 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718
719 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
720 if (!sock)
Tom Tucker38a417c2007-12-30 21:07:36 -0600721 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722
Tom Tucker02fc6c32007-12-30 21:07:48 -0600723 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
Sridhar Samudralae6242e92006-08-07 20:58:01 -0700724 err = kernel_accept(sock, &newsock, O_NONBLOCK);
725 if (err < 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 if (err == -ENOMEM)
727 printk(KERN_WARNING "%s: no more sockets!\n",
728 serv->sv_name);
Sridhar Samudralae6242e92006-08-07 20:58:01 -0700729 else if (err != -EAGAIN && net_ratelimit())
730 printk(KERN_WARNING "%s: accept failed (err %d)!\n",
731 serv->sv_name, -err);
Tom Tucker38a417c2007-12-30 21:07:36 -0600732 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 }
Tom Tucker02fc6c32007-12-30 21:07:48 -0600734 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735
akpm@linux-foundation.orgcdd88b92007-02-12 00:53:38 -0800736 err = kernel_getpeername(newsock, sin, &slen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 if (err < 0) {
738 if (net_ratelimit())
739 printk(KERN_WARNING "%s: peername failed (err %d)!\n",
740 serv->sv_name, -err);
741 goto failed; /* aborted connection or whatever */
742 }
743
744 /* Ideally, we would want to reject connections from unauthorized
Chuck Leverad06e4b2007-02-12 00:53:32 -0800745 * hosts here, but when we get encryption, the IP of the host won't
746 * tell us anything. For now just warn about unpriv connections.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 */
akpm@linux-foundation.orgcdd88b92007-02-12 00:53:38 -0800748 if (!svc_port_is_privileged(sin)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749 dprintk(KERN_WARNING
Chuck Leverad06e4b2007-02-12 00:53:32 -0800750 "%s: connect from unprivileged port: %s\n",
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -0800751 serv->sv_name,
akpm@linux-foundation.orgcdd88b92007-02-12 00:53:38 -0800752 __svc_print_addr(sin, buf, sizeof(buf)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 }
Chuck Leverad06e4b2007-02-12 00:53:32 -0800754 dprintk("%s: connect from %s\n", serv->sv_name,
akpm@linux-foundation.orgcdd88b92007-02-12 00:53:38 -0800755 __svc_print_addr(sin, buf, sizeof(buf)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756
757 /* make sure that a write doesn't block forever when
758 * low on memory
759 */
760 newsock->sk->sk_sndtimeo = HZ*30;
761
Chuck Lever6b174332007-02-12 00:53:28 -0800762 if (!(newsvsk = svc_setup_socket(serv, newsock, &err,
763 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY))))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 goto failed;
Tom Tucker9dbc2402007-12-30 21:08:12 -0600765 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
Frank van Maarseveena9747692007-07-09 22:21:39 +0200766 err = kernel_getsockname(newsock, sin, &slen);
767 if (unlikely(err < 0)) {
768 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
769 slen = offsetof(struct sockaddr, sa_data);
770 }
Tom Tucker9dbc2402007-12-30 21:08:12 -0600771 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
Chuck Lever067d7812007-02-12 00:53:30 -0800772
Tom Tuckerf9f3cc42007-12-30 21:07:40 -0600773 if (serv->sv_stats)
774 serv->sv_stats->nettcpconn++;
775
776 return &newsvsk->sk_xprt;
777
778failed:
779 sock_release(newsock);
780 return NULL;
781}
782
783/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 * Receive data from a TCP socket.
785 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600786static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787{
Tom Tucker57b1d3b2007-12-30 21:08:22 -0600788 struct svc_sock *svsk =
789 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600790 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 int len;
NeilBrown3cc03b12006-10-04 02:15:47 -0700792 struct kvec *vec;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 int pnum, vlen;
794
795 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
Tom Tucker02fc6c32007-12-30 21:07:48 -0600796 svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
797 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
798 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799
J. Bruce Fields7f421832009-05-27 18:51:06 -0400800 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
801 /* sndbuf needs to have room for one request
802 * per thread, otherwise we can stall even when the
803 * network isn't a bottleneck.
804 *
805 * We count all threads rather than threads in a
806 * particular pool, which provides an upper bound
807 * on the number of threads which will access the socket.
808 *
809 * rcvbuf just needs to be able to hold a few requests.
810 * Normally they will be removed from the queue
811 * as soon a a complete request arrives.
812 */
813 svc_sock_setbufsize(svsk->sk_sock,
814 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
815 3 * serv->sv_max_mesg);
816
Tom Tucker02fc6c32007-12-30 21:07:48 -0600817 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818
819 /* Receive data. If we haven't got the record length yet, get
820 * the next four bytes. Otherwise try to gobble up as much as
821 * possible up to the complete record length.
822 */
Chuck Leverc0401ea2008-04-14 12:27:30 -0400823 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
824 int want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 struct kvec iov;
826
827 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
828 iov.iov_len = want;
829 if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
830 goto error;
831 svsk->sk_tcplen += len;
832
833 if (len < want) {
Chuck Leverc0401ea2008-04-14 12:27:30 -0400834 dprintk("svc: short recvfrom while reading record "
835 "length (%d of %d)\n", len, want);
Tom Tuckera6046f72007-12-30 21:08:01 -0600836 svc_xprt_received(&svsk->sk_xprt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 return -EAGAIN; /* record header not complete */
838 }
839
840 svsk->sk_reclen = ntohl(svsk->sk_reclen);
Chuck Leverc0401ea2008-04-14 12:27:30 -0400841 if (!(svsk->sk_reclen & RPC_LAST_STREAM_FRAGMENT)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 /* FIXME: technically, a record can be fragmented,
843 * and non-terminal fragments will not have the top
844 * bit set in the fragment length header.
845 * But apparently no known nfs clients send fragmented
846 * records. */
NeilBrown34e9a632007-01-29 13:19:52 -0800847 if (net_ratelimit())
Chuck Leverc0401ea2008-04-14 12:27:30 -0400848 printk(KERN_NOTICE "RPC: multiple fragments "
849 "per record not supported\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 goto err_delete;
851 }
Chuck Leverc0401ea2008-04-14 12:27:30 -0400852 svsk->sk_reclen &= RPC_FRAGMENT_SIZE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen);
NeilBrownc6b0a9f2006-10-06 00:44:05 -0700854 if (svsk->sk_reclen > serv->sv_max_mesg) {
NeilBrown34e9a632007-01-29 13:19:52 -0800855 if (net_ratelimit())
Chuck Leverc0401ea2008-04-14 12:27:30 -0400856 printk(KERN_NOTICE "RPC: "
857 "fragment too large: 0x%08lx\n",
858 (unsigned long)svsk->sk_reclen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859 goto err_delete;
860 }
861 }
862
863 /* Check whether enough data is available */
864 len = svc_recv_available(svsk);
865 if (len < 0)
866 goto error;
867
868 if (len < svsk->sk_reclen) {
869 dprintk("svc: incomplete TCP record (%d of %d)\n",
870 len, svsk->sk_reclen);
Tom Tuckera6046f72007-12-30 21:08:01 -0600871 svc_xprt_received(&svsk->sk_xprt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 return -EAGAIN; /* record not complete */
873 }
874 len = svsk->sk_reclen;
Tom Tucker02fc6c32007-12-30 21:07:48 -0600875 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876
NeilBrown3cc03b12006-10-04 02:15:47 -0700877 vec = rqstp->rq_vec;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 vec[0] = rqstp->rq_arg.head[0];
879 vlen = PAGE_SIZE;
880 pnum = 1;
881 while (vlen < len) {
NeilBrown44524352006-10-04 02:15:46 -0700882 vec[pnum].iov_base = page_address(rqstp->rq_pages[pnum]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 vec[pnum].iov_len = PAGE_SIZE;
884 pnum++;
885 vlen += PAGE_SIZE;
886 }
NeilBrown44524352006-10-04 02:15:46 -0700887 rqstp->rq_respages = &rqstp->rq_pages[pnum];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888
889 /* Now receive data */
890 len = svc_recvfrom(rqstp, vec, pnum, len);
891 if (len < 0)
892 goto error;
893
894 dprintk("svc: TCP complete record (%d bytes)\n", len);
895 rqstp->rq_arg.len = len;
896 rqstp->rq_arg.page_base = 0;
897 if (len <= rqstp->rq_arg.head[0].iov_len) {
898 rqstp->rq_arg.head[0].iov_len = len;
899 rqstp->rq_arg.page_len = 0;
900 } else {
901 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
902 }
903
Tom Tucker5148bf42007-12-30 21:07:25 -0600904 rqstp->rq_xprt_ctxt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905 rqstp->rq_prot = IPPROTO_TCP;
906
907 /* Reset TCP read info */
908 svsk->sk_reclen = 0;
909 svsk->sk_tcplen = 0;
910
Tom Tucker9dbc2402007-12-30 21:08:12 -0600911 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
Tom Tuckera6046f72007-12-30 21:08:01 -0600912 svc_xprt_received(&svsk->sk_xprt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913 if (serv->sv_stats)
914 serv->sv_stats->nettcpcnt++;
915
916 return len;
917
918 err_delete:
Tom Tucker02fc6c32007-12-30 21:07:48 -0600919 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 return -EAGAIN;
921
922 error:
923 if (len == -EAGAIN) {
924 dprintk("RPC: TCP recvfrom got EAGAIN\n");
Tom Tuckera6046f72007-12-30 21:08:01 -0600925 svc_xprt_received(&svsk->sk_xprt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926 } else {
927 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600928 svsk->sk_xprt.xpt_server->sv_name, -len);
Olaf Kirch93fbf1a2006-01-06 00:19:56 -0800929 goto err_delete;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930 }
931
932 return len;
933}
934
935/*
936 * Send out data on TCP socket.
937 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600938static int svc_tcp_sendto(struct svc_rqst *rqstp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939{
940 struct xdr_buf *xbufp = &rqstp->rq_res;
941 int sent;
Alexey Dobriyand8ed0292006-09-26 22:29:38 -0700942 __be32 reclen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943
944 /* Set up the first element of the reply kvec.
945 * Any other kvecs that may be in use have been taken
946 * care of by the server implementation itself.
947 */
948 reclen = htonl(0x80000000|((xbufp->len ) - 4));
949 memcpy(xbufp->head[0].iov_base, &reclen, 4);
950
Tom Tucker57b1d3b2007-12-30 21:08:22 -0600951 if (test_bit(XPT_DEAD, &rqstp->rq_xprt->xpt_flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 return -ENOTCONN;
953
954 sent = svc_sendto(rqstp, &rqstp->rq_res);
955 if (sent != xbufp->len) {
Tom Tucker0f0257e2007-12-30 21:08:27 -0600956 printk(KERN_NOTICE
957 "rpc-srv/tcp: %s: %s %d when sending %d bytes "
958 "- shutting down socket\n",
Tom Tucker57b1d3b2007-12-30 21:08:22 -0600959 rqstp->rq_xprt->xpt_server->sv_name,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 (sent<0)?"got error":"sent only",
961 sent, xbufp->len);
Tom Tucker57b1d3b2007-12-30 21:08:22 -0600962 set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
Tom Tuckerf6150c32007-12-30 21:07:57 -0600963 svc_xprt_enqueue(rqstp->rq_xprt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964 sent = -EAGAIN;
965 }
966 return sent;
967}
968
Tom Tuckere831fe62007-12-30 21:07:29 -0600969/*
970 * Setup response header. TCP has a 4B record length field.
971 */
972static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
973{
974 struct kvec *resv = &rqstp->rq_res.head[0];
975
976 /* tcp needs a space for the record length... */
977 svc_putnl(resv, 0);
978}
979
Tom Tucker323bee32007-12-30 21:07:31 -0600980static int svc_tcp_has_wspace(struct svc_xprt *xprt)
981{
Tom Tucker57b1d3b2007-12-30 21:08:22 -0600982 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600983 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
Tom Tucker323bee32007-12-30 21:07:31 -0600984 int required;
985 int wspace;
986
987 /*
988 * Set the SOCK_NOSPACE flag before checking the available
989 * sock space.
990 */
991 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
Tom Tucker7a90e8c2007-12-30 21:07:55 -0600992 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
Tom Tucker323bee32007-12-30 21:07:31 -0600993 wspace = sk_stream_wspace(svsk->sk_sk);
994
995 if (wspace < sk_stream_min_wspace(svsk->sk_sk))
996 return 0;
997 if (required * 2 > wspace)
998 return 0;
999
1000 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
1001 return 1;
1002}
1003
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001004static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1005 struct sockaddr *sa, int salen,
1006 int flags)
1007{
1008 return svc_create_socket(serv, IPPROTO_TCP, sa, salen, flags);
1009}
1010
Tom Tucker360d87382007-12-30 21:07:17 -06001011static struct svc_xprt_ops svc_tcp_ops = {
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001012 .xpo_create = svc_tcp_create,
Tom Tucker5d137992007-12-30 21:07:23 -06001013 .xpo_recvfrom = svc_tcp_recvfrom,
1014 .xpo_sendto = svc_tcp_sendto,
Tom Tucker5148bf42007-12-30 21:07:25 -06001015 .xpo_release_rqst = svc_release_skb,
Trond Myklebust69b6ba32008-12-23 16:30:11 -05001016 .xpo_detach = svc_tcp_sock_detach,
Tom Tucker755ccea2007-12-30 21:07:27 -06001017 .xpo_free = svc_sock_free,
Tom Tuckere831fe62007-12-30 21:07:29 -06001018 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
Tom Tucker323bee32007-12-30 21:07:31 -06001019 .xpo_has_wspace = svc_tcp_has_wspace,
Tom Tucker38a417c2007-12-30 21:07:36 -06001020 .xpo_accept = svc_tcp_accept,
Tom Tucker360d87382007-12-30 21:07:17 -06001021};
1022
1023static struct svc_xprt_class svc_tcp_class = {
1024 .xcl_name = "tcp",
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001025 .xcl_owner = THIS_MODULE,
Tom Tucker360d87382007-12-30 21:07:17 -06001026 .xcl_ops = &svc_tcp_ops,
Tom Tucker49023152007-12-30 21:07:21 -06001027 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
Tom Tucker360d87382007-12-30 21:07:17 -06001028};
1029
1030void svc_init_xprt_sock(void)
1031{
1032 svc_reg_xprt_class(&svc_tcp_class);
1033 svc_reg_xprt_class(&svc_udp_class);
1034}
1035
1036void svc_cleanup_xprt_sock(void)
1037{
1038 svc_unreg_xprt_class(&svc_tcp_class);
1039 svc_unreg_xprt_class(&svc_udp_class);
1040}
1041
Tom Tuckerbb5cf162007-12-30 21:07:50 -06001042static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043{
1044 struct sock *sk = svsk->sk_sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045
Tom Tuckerbb5cf162007-12-30 21:07:50 -06001046 svc_xprt_init(&svc_tcp_class, &svsk->sk_xprt, serv);
Tom Tuckerdef13d72007-12-30 21:08:08 -06001047 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 if (sk->sk_state == TCP_LISTEN) {
1049 dprintk("setting up TCP socket for listening\n");
Tom Tucker02fc6c32007-12-30 21:07:48 -06001050 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 sk->sk_data_ready = svc_tcp_listen_data_ready;
Tom Tucker02fc6c32007-12-30 21:07:48 -06001052 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053 } else {
1054 dprintk("setting up TCP socket for reading\n");
1055 sk->sk_state_change = svc_tcp_state_change;
1056 sk->sk_data_ready = svc_tcp_data_ready;
1057 sk->sk_write_space = svc_write_space;
1058
1059 svsk->sk_reclen = 0;
1060 svsk->sk_tcplen = 0;
1061
Chuck Leverb7872fe2008-04-14 12:27:01 -04001062 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063
J. Bruce Fields7f421832009-05-27 18:51:06 -04001064 /* initialise setting must have enough space to
1065 * receive and respond to one request.
1066 * svc_tcp_recvfrom will re-adjust if necessary
1067 */
1068 svc_sock_setbufsize(svsk->sk_sock,
1069 3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
1070 3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
1071
1072 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
Tom Tucker02fc6c32007-12-30 21:07:48 -06001073 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -08001074 if (sk->sk_state != TCP_ESTABLISHED)
Tom Tucker02fc6c32007-12-30 21:07:48 -06001075 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 }
1077}
1078
Tom Tucker0f0257e2007-12-30 21:08:27 -06001079void svc_sock_update_bufs(struct svc_serv *serv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080{
1081 /*
1082 * The number of server threads has changed. Update
1083 * rcvbuf and sndbuf accordingly on all sockets
1084 */
1085 struct list_head *le;
1086
1087 spin_lock_bh(&serv->sv_lock);
1088 list_for_each(le, &serv->sv_permsocks) {
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -08001089 struct svc_sock *svsk =
Tom Tucker7a182082007-12-30 21:07:53 -06001090 list_entry(le, struct svc_sock, sk_xprt.xpt_list);
Tom Tucker02fc6c32007-12-30 21:07:48 -06001091 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092 }
1093 list_for_each(le, &serv->sv_tempsocks) {
1094 struct svc_sock *svsk =
Tom Tucker7a182082007-12-30 21:07:53 -06001095 list_entry(le, struct svc_sock, sk_xprt.xpt_list);
Tom Tucker02fc6c32007-12-30 21:07:48 -06001096 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 }
1098 spin_unlock_bh(&serv->sv_lock);
1099}
Trond Myklebust24c37672008-12-23 16:30:12 -05001100EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101
1102/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 * Initialize socket for RPC use and create svc_sock struct
1104 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1105 */
Chuck Lever6b174332007-02-12 00:53:28 -08001106static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1107 struct socket *sock,
1108 int *errp, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109{
1110 struct svc_sock *svsk;
1111 struct sock *inet;
Chuck Lever6b174332007-02-12 00:53:28 -08001112 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113
1114 dprintk("svc: svc_setup_socket %p\n", sock);
Panagiotis Issaris0da974f2006-07-21 14:51:30 -07001115 if (!(svsk = kzalloc(sizeof(*svsk), GFP_KERNEL))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 *errp = -ENOMEM;
1117 return NULL;
1118 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119
1120 inet = sock->sk;
1121
1122 /* Register socket with portmapper */
1123 if (*errp >= 0 && pmap_register)
Chuck Leverbaf01ca2009-03-18 20:46:13 -04001124 *errp = svc_register(serv, inet->sk_family, inet->sk_protocol,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125 ntohs(inet_sk(inet)->sport));
1126
1127 if (*errp < 0) {
1128 kfree(svsk);
1129 return NULL;
1130 }
1131
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 inet->sk_user_data = svsk;
1133 svsk->sk_sock = sock;
1134 svsk->sk_sk = inet;
1135 svsk->sk_ostate = inet->sk_state_change;
1136 svsk->sk_odata = inet->sk_data_ready;
1137 svsk->sk_owspace = inet->sk_write_space;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138
1139 /* Initialize the socket */
1140 if (sock->type == SOCK_DGRAM)
Tom Tuckerbb5cf162007-12-30 21:07:50 -06001141 svc_udp_init(svsk, serv);
J. Bruce Fields7f421832009-05-27 18:51:06 -04001142 else
Tom Tuckerbb5cf162007-12-30 21:07:50 -06001143 svc_tcp_init(svsk, serv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1146 svsk, svsk->sk_sk);
1147
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148 return svsk;
1149}
1150
NeilBrownb41b66d2006-10-02 02:17:48 -07001151int svc_addsock(struct svc_serv *serv,
1152 int fd,
Chuck Lever29373912008-10-03 17:15:38 -04001153 char *name_return)
NeilBrownb41b66d2006-10-02 02:17:48 -07001154{
1155 int err = 0;
1156 struct socket *so = sockfd_lookup(fd, &err);
1157 struct svc_sock *svsk = NULL;
1158
1159 if (!so)
1160 return err;
1161 if (so->sk->sk_family != AF_INET)
1162 err = -EAFNOSUPPORT;
1163 else if (so->sk->sk_protocol != IPPROTO_TCP &&
1164 so->sk->sk_protocol != IPPROTO_UDP)
1165 err = -EPROTONOSUPPORT;
1166 else if (so->state > SS_UNCONNECTED)
1167 err = -EISCONN;
1168 else {
Tom Tucker2da2c212008-11-23 09:58:08 -06001169 if (!try_module_get(THIS_MODULE))
1170 err = -ENOENT;
1171 else
1172 svsk = svc_setup_socket(serv, so, &err,
1173 SVC_SOCK_DEFAULTS);
NeilBrowne79eff12007-02-12 00:53:30 -08001174 if (svsk) {
Tom Tucker9dbc2402007-12-30 21:08:12 -06001175 struct sockaddr_storage addr;
1176 struct sockaddr *sin = (struct sockaddr *)&addr;
1177 int salen;
1178 if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0)
1179 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
Tom Tucker4e5caaa2007-12-30 21:08:20 -06001180 clear_bit(XPT_TEMP, &svsk->sk_xprt.xpt_flags);
1181 spin_lock_bh(&serv->sv_lock);
1182 list_add(&svsk->sk_xprt.xpt_list, &serv->sv_permsocks);
1183 spin_unlock_bh(&serv->sv_lock);
Tom Tuckera6046f72007-12-30 21:08:01 -06001184 svc_xprt_received(&svsk->sk_xprt);
NeilBrownb41b66d2006-10-02 02:17:48 -07001185 err = 0;
Tom Tucker2da2c212008-11-23 09:58:08 -06001186 } else
1187 module_put(THIS_MODULE);
NeilBrownb41b66d2006-10-02 02:17:48 -07001188 }
1189 if (err) {
1190 sockfd_put(so);
1191 return err;
1192 }
NeilBrownb41b66d2006-10-02 02:17:48 -07001193 return one_sock_name(name_return, svsk);
1194}
1195EXPORT_SYMBOL_GPL(svc_addsock);
1196
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197/*
1198 * Create socket for RPC service.
1199 */
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001200static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1201 int protocol,
1202 struct sockaddr *sin, int len,
1203 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204{
1205 struct svc_sock *svsk;
1206 struct socket *sock;
1207 int error;
1208 int type;
Tom Tucker9dbc2402007-12-30 21:08:12 -06001209 struct sockaddr_storage addr;
1210 struct sockaddr *newsin = (struct sockaddr *)&addr;
1211 int newlen;
Trond Myklebustc69da772009-03-30 18:59:17 -04001212 int family;
1213 int val;
Pavel Emelyanov5216a8e2008-02-21 10:57:45 +03001214 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215
Chuck Leverad06e4b2007-02-12 00:53:32 -08001216 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1217 serv->sv_program->pg_name, protocol,
Chuck Lever77f1f672007-02-12 00:53:39 -08001218 __svc_print_addr(sin, buf, sizeof(buf)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219
1220 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1221 printk(KERN_WARNING "svc: only UDP and TCP "
1222 "sockets supported\n");
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001223 return ERR_PTR(-EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225
Trond Myklebustc69da772009-03-30 18:59:17 -04001226 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1227 switch (sin->sa_family) {
1228 case AF_INET6:
1229 family = PF_INET6;
1230 break;
1231 case AF_INET:
1232 family = PF_INET;
1233 break;
1234 default:
1235 return ERR_PTR(-EINVAL);
1236 }
1237
1238 error = sock_create_kern(family, type, protocol, &sock);
Chuck Lever77f1f672007-02-12 00:53:39 -08001239 if (error < 0)
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001240 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241
Peter Zijlstraed075362006-12-06 20:35:24 -08001242 svc_reclassify_socket(sock);
1243
Trond Myklebustc69da772009-03-30 18:59:17 -04001244 /*
1245 * If this is an PF_INET6 listener, we want to avoid
1246 * getting requests from IPv4 remotes. Those should
1247 * be shunted to a PF_INET listener via rpcbind.
1248 */
1249 val = 1;
1250 if (family == PF_INET6)
1251 kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
1252 (char *)&val, sizeof(val));
1253
Eric Sesterhenn18114742006-09-28 14:37:07 -07001254 if (type == SOCK_STREAM)
Chuck Lever77f1f672007-02-12 00:53:39 -08001255 sock->sk->sk_reuse = 1; /* allow address reuse */
1256 error = kernel_bind(sock, sin, len);
Eric Sesterhenn18114742006-09-28 14:37:07 -07001257 if (error < 0)
1258 goto bummer;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259
Tom Tucker9dbc2402007-12-30 21:08:12 -06001260 newlen = len;
1261 error = kernel_getsockname(sock, newsin, &newlen);
1262 if (error < 0)
1263 goto bummer;
1264
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265 if (protocol == IPPROTO_TCP) {
Sridhar Samudralae6242e92006-08-07 20:58:01 -07001266 if ((error = kernel_listen(sock, 64)) < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267 goto bummer;
1268 }
1269
NeilBrowne79eff12007-02-12 00:53:30 -08001270 if ((svsk = svc_setup_socket(serv, sock, &error, flags)) != NULL) {
Tom Tucker9dbc2402007-12-30 21:08:12 -06001271 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001272 return (struct svc_xprt *)svsk;
NeilBrowne79eff12007-02-12 00:53:30 -08001273 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274
1275bummer:
1276 dprintk("svc: svc_create_socket error = %d\n", -error);
1277 sock_release(sock);
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001278 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279}
1280
1281/*
Tom Tucker755ccea2007-12-30 21:07:27 -06001282 * Detach the svc_sock from the socket so that no
1283 * more callbacks occur.
1284 */
1285static void svc_sock_detach(struct svc_xprt *xprt)
1286{
1287 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1288 struct sock *sk = svsk->sk_sk;
1289
1290 dprintk("svc: svc_sock_detach(%p)\n", svsk);
1291
1292 /* put back the old socket callbacks */
1293 sk->sk_state_change = svsk->sk_ostate;
1294 sk->sk_data_ready = svsk->sk_odata;
1295 sk->sk_write_space = svsk->sk_owspace;
Trond Myklebust69b6ba32008-12-23 16:30:11 -05001296
1297 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1298 wake_up_interruptible(sk->sk_sleep);
1299}
1300
1301/*
1302 * Disconnect the socket, and reset the callbacks
1303 */
1304static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1305{
1306 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1307
1308 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
1309
1310 svc_sock_detach(xprt);
1311
1312 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags))
1313 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
Tom Tucker755ccea2007-12-30 21:07:27 -06001314}
1315
1316/*
1317 * Free the svc_sock's socket resources and the svc_sock itself.
1318 */
1319static void svc_sock_free(struct svc_xprt *xprt)
1320{
1321 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1322 dprintk("svc: svc_sock_free(%p)\n", svsk);
1323
Tom Tucker755ccea2007-12-30 21:07:27 -06001324 if (svsk->sk_sock->file)
1325 sockfd_put(svsk->sk_sock);
1326 else
1327 sock_release(svsk->sk_sock);
1328 kfree(svsk);
1329}