blob: b7d435c3f19ec537275cb8b413d6c929c4dc8fa6 [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>
Rahul Iyer4cfc7e62009-09-10 17:32:28 +030052#include <linux/sunrpc/xprt.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070053
Tom Tucker360d87382007-12-30 21:07:17 -060054#define RPCDBG_FACILITY RPCDBG_SVCXPRT
Linus Torvalds1da177e2005-04-16 15:20:36 -070055
56
57static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
Chuck Lever6b174332007-02-12 00:53:28 -080058 int *errp, int flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -070059static void svc_udp_data_ready(struct sock *, int);
60static int svc_udp_recvfrom(struct svc_rqst *);
61static int svc_udp_sendto(struct svc_rqst *);
Tom Tucker755ccea2007-12-30 21:07:27 -060062static void svc_sock_detach(struct svc_xprt *);
Trond Myklebust69b6ba32008-12-23 16:30:11 -050063static void svc_tcp_sock_detach(struct svc_xprt *);
Tom Tucker755ccea2007-12-30 21:07:27 -060064static void svc_sock_free(struct svc_xprt *);
Linus Torvalds1da177e2005-04-16 15:20:36 -070065
Tom Tuckerb700cbb2007-12-30 21:07:42 -060066static struct svc_xprt *svc_create_socket(struct svc_serv *, int,
Pavel Emelyanov62832c02010-09-29 16:04:18 +040067 struct net *, struct sockaddr *,
68 int, int);
Andy Adamson1f11a032011-01-06 02:04:26 +000069#if defined(CONFIG_NFS_V4_1)
70static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
71 struct net *, struct sockaddr *,
72 int, int);
73static void svc_bc_sock_free(struct svc_xprt *xprt);
74#endif /* CONFIG_NFS_V4_1 */
75
Peter Zijlstraed075362006-12-06 20:35:24 -080076#ifdef CONFIG_DEBUG_LOCK_ALLOC
77static struct lock_class_key svc_key[2];
78static struct lock_class_key svc_slock_key[2];
79
Tom Tucker0f0257e2007-12-30 21:08:27 -060080static void svc_reclassify_socket(struct socket *sock)
Peter Zijlstraed075362006-12-06 20:35:24 -080081{
82 struct sock *sk = sock->sk;
John Heffner02b3d342007-09-12 10:42:12 +020083 BUG_ON(sock_owned_by_user(sk));
Peter Zijlstraed075362006-12-06 20:35:24 -080084 switch (sk->sk_family) {
85 case AF_INET:
86 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
Tom Tuckerdef13d72007-12-30 21:08:08 -060087 &svc_slock_key[0],
88 "sk_xprt.xpt_lock-AF_INET-NFSD",
89 &svc_key[0]);
Peter Zijlstraed075362006-12-06 20:35:24 -080090 break;
91
92 case AF_INET6:
93 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
Tom Tuckerdef13d72007-12-30 21:08:08 -060094 &svc_slock_key[1],
95 "sk_xprt.xpt_lock-AF_INET6-NFSD",
96 &svc_key[1]);
Peter Zijlstraed075362006-12-06 20:35:24 -080097 break;
98
99 default:
100 BUG();
101 }
102}
103#else
Tom Tucker0f0257e2007-12-30 21:08:27 -0600104static void svc_reclassify_socket(struct socket *sock)
Peter Zijlstraed075362006-12-06 20:35:24 -0800105{
106}
107#endif
108
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109/*
110 * Release an skbuff after use
111 */
Tom Tucker5148bf42007-12-30 21:07:25 -0600112static void svc_release_skb(struct svc_rqst *rqstp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113{
Tom Tucker5148bf42007-12-30 21:07:25 -0600114 struct sk_buff *skb = rqstp->rq_xprt_ctxt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115
116 if (skb) {
Tom Tucker57b1d3b2007-12-30 21:08:22 -0600117 struct svc_sock *svsk =
118 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
Tom Tucker5148bf42007-12-30 21:07:25 -0600119 rqstp->rq_xprt_ctxt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120
121 dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
Eric Dumazet9d410c72009-10-30 05:03:53 +0000122 skb_free_datagram_locked(svsk->sk_sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124}
125
Chuck Leverb92503b2007-02-12 00:53:36 -0800126union svc_pktinfo_u {
127 struct in_pktinfo pkti;
Chuck Leverb92503b2007-02-12 00:53:36 -0800128 struct in6_pktinfo pkti6;
Chuck Leverb92503b2007-02-12 00:53:36 -0800129};
David S. Millerbc375ea2007-04-12 13:35:59 -0700130#define SVC_PKTINFO_SPACE \
131 CMSG_SPACE(sizeof(union svc_pktinfo_u))
Chuck Leverb92503b2007-02-12 00:53:36 -0800132
133static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
134{
Tom Tucker57b1d3b2007-12-30 21:08:22 -0600135 struct svc_sock *svsk =
136 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
137 switch (svsk->sk_sk->sk_family) {
Chuck Leverb92503b2007-02-12 00:53:36 -0800138 case AF_INET: {
139 struct in_pktinfo *pki = CMSG_DATA(cmh);
140
141 cmh->cmsg_level = SOL_IP;
142 cmh->cmsg_type = IP_PKTINFO;
143 pki->ipi_ifindex = 0;
144 pki->ipi_spec_dst.s_addr = rqstp->rq_daddr.addr.s_addr;
145 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
146 }
147 break;
NeilBrown5a05ed72007-03-06 01:42:22 -0800148
Chuck Leverb92503b2007-02-12 00:53:36 -0800149 case AF_INET6: {
150 struct in6_pktinfo *pki = CMSG_DATA(cmh);
151
152 cmh->cmsg_level = SOL_IPV6;
153 cmh->cmsg_type = IPV6_PKTINFO;
154 pki->ipi6_ifindex = 0;
155 ipv6_addr_copy(&pki->ipi6_addr,
156 &rqstp->rq_daddr.addr6);
157 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
158 }
159 break;
Chuck Leverb92503b2007-02-12 00:53:36 -0800160 }
Chuck Leverb92503b2007-02-12 00:53:36 -0800161}
162
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163/*
Rahul Iyer4cfc7e62009-09-10 17:32:28 +0300164 * send routine intended to be shared by the fore- and back-channel
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165 */
Rahul Iyer4cfc7e62009-09-10 17:32:28 +0300166int svc_send_common(struct socket *sock, struct xdr_buf *xdr,
167 struct page *headpage, unsigned long headoffset,
168 struct page *tailpage, unsigned long tailoffset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 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;
Rahul Iyer4cfc7e62009-09-10 17:32:28 +0300176 int slen;
177 int len = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178
179 slen = xdr->len;
180
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 /* send head */
182 if (slen == xdr->head[0].iov_len)
183 flags = 0;
Rahul Iyer4cfc7e62009-09-10 17:32:28 +0300184 len = kernel_sendpage(sock, headpage, headoffset,
NeilBrown44524352006-10-04 02:15:46 -0700185 xdr->head[0].iov_len, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 if (len != xdr->head[0].iov_len)
187 goto out;
188 slen -= xdr->head[0].iov_len;
189 if (slen == 0)
190 goto out;
191
192 /* send page data */
193 size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen;
194 while (pglen > 0) {
195 if (slen == size)
196 flags = 0;
Sridhar Samudralae6242e92006-08-07 20:58:01 -0700197 result = kernel_sendpage(sock, *ppage, base, size, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 if (result > 0)
199 len += result;
200 if (result != size)
201 goto out;
202 slen -= size;
203 pglen -= size;
204 size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen;
205 base = 0;
206 ppage++;
207 }
Rahul Iyer4cfc7e62009-09-10 17:32:28 +0300208
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 /* send tail */
210 if (xdr->tail[0].iov_len) {
Rahul Iyer4cfc7e62009-09-10 17:32:28 +0300211 result = kernel_sendpage(sock, tailpage, tailoffset,
212 xdr->tail[0].iov_len, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 if (result > 0)
214 len += result;
215 }
Rahul Iyer4cfc7e62009-09-10 17:32:28 +0300216
217out:
218 return len;
219}
220
221
222/*
223 * Generic sendto routine
224 */
225static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr)
226{
227 struct svc_sock *svsk =
228 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
229 struct socket *sock = svsk->sk_sock;
230 union {
231 struct cmsghdr hdr;
232 long all[SVC_PKTINFO_SPACE / sizeof(long)];
233 } buffer;
234 struct cmsghdr *cmh = &buffer.hdr;
235 int len = 0;
236 unsigned long tailoff;
237 unsigned long headoff;
238 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
239
240 if (rqstp->rq_prot == IPPROTO_UDP) {
241 struct msghdr msg = {
242 .msg_name = &rqstp->rq_addr,
243 .msg_namelen = rqstp->rq_addrlen,
244 .msg_control = cmh,
245 .msg_controllen = sizeof(buffer),
246 .msg_flags = MSG_MORE,
247 };
248
249 svc_set_cmsg_data(rqstp, cmh);
250
251 if (sock_sendmsg(sock, &msg, 0) < 0)
252 goto out;
253 }
254
255 tailoff = ((unsigned long)xdr->tail[0].iov_base) & (PAGE_SIZE-1);
256 headoff = 0;
257 len = svc_send_common(sock, xdr, rqstp->rq_respages[0], headoff,
258 rqstp->rq_respages[0], tailoff);
259
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260out:
Chuck Leverad06e4b2007-02-12 00:53:32 -0800261 dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n",
Tom Tucker57b1d3b2007-12-30 21:08:22 -0600262 svsk, xdr->head[0].iov_base, xdr->head[0].iov_len,
Chuck Leverad06e4b2007-02-12 00:53:32 -0800263 xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264
265 return len;
266}
267
268/*
NeilBrown80212d52006-10-02 02:17:47 -0700269 * Report socket names for nfsdfs
270 */
Chuck Levere7942b92009-04-23 19:32:48 -0400271static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining)
NeilBrown80212d52006-10-02 02:17:47 -0700272{
Chuck Lever017cb472009-04-23 19:33:03 -0400273 const struct sock *sk = svsk->sk_sk;
274 const char *proto_name = sk->sk_protocol == IPPROTO_UDP ?
275 "udp" : "tcp";
NeilBrown80212d52006-10-02 02:17:47 -0700276 int len;
277
Chuck Lever017cb472009-04-23 19:33:03 -0400278 switch (sk->sk_family) {
Chuck Levere7942b92009-04-23 19:32:48 -0400279 case PF_INET:
280 len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n",
Chuck Lever017cb472009-04-23 19:33:03 -0400281 proto_name,
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000282 &inet_sk(sk)->inet_rcv_saddr,
283 inet_sk(sk)->inet_num);
NeilBrown80212d52006-10-02 02:17:47 -0700284 break;
Chuck Lever58de2f82009-04-23 19:32:55 -0400285 case PF_INET6:
286 len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n",
Chuck Lever017cb472009-04-23 19:33:03 -0400287 proto_name,
288 &inet6_sk(sk)->rcv_saddr,
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000289 inet_sk(sk)->inet_num);
NeilBrown80212d52006-10-02 02:17:47 -0700290 break;
291 default:
Chuck Levere7942b92009-04-23 19:32:48 -0400292 len = snprintf(buf, remaining, "*unknown-%d*\n",
Chuck Lever017cb472009-04-23 19:33:03 -0400293 sk->sk_family);
NeilBrown80212d52006-10-02 02:17:47 -0700294 }
Chuck Levere7942b92009-04-23 19:32:48 -0400295
296 if (len >= remaining) {
297 *buf = '\0';
298 return -ENAMETOOLONG;
NeilBrown80212d52006-10-02 02:17:47 -0700299 }
300 return len;
301}
302
Chuck Lever8435d342009-04-23 19:32:40 -0400303/**
304 * svc_sock_names - construct a list of listener names in a string
305 * @serv: pointer to RPC service
306 * @buf: pointer to a buffer to fill in with socket names
307 * @buflen: size of the buffer to be filled
308 * @toclose: pointer to '\0'-terminated C string containing the name
309 * of a listener to be closed
310 *
311 * Fills in @buf with a '\n'-separated list of names of listener
312 * sockets. If @toclose is not NULL, the socket named by @toclose
313 * is closed, and is not included in the output list.
314 *
315 * Returns positive length of the socket name string, or a negative
316 * errno value on error.
317 */
318int svc_sock_names(struct svc_serv *serv, char *buf, const size_t buflen,
319 const char *toclose)
NeilBrown80212d52006-10-02 02:17:47 -0700320{
NeilBrownb41b66d2006-10-02 02:17:48 -0700321 struct svc_sock *svsk, *closesk = NULL;
NeilBrown80212d52006-10-02 02:17:47 -0700322 int len = 0;
323
324 if (!serv)
325 return 0;
Chuck Levere7942b92009-04-23 19:32:48 -0400326
NeilBrownaaf68cf2007-02-08 14:20:30 -0800327 spin_lock_bh(&serv->sv_lock);
Tom Tucker7a182082007-12-30 21:07:53 -0600328 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list) {
Chuck Levere7942b92009-04-23 19:32:48 -0400329 int onelen = svc_one_sock_name(svsk, buf + len, buflen - len);
330 if (onelen < 0) {
331 len = onelen;
332 break;
333 }
NeilBrown39423022010-11-15 11:27:01 +1100334 if (toclose && strcmp(toclose, buf + len) == 0) {
NeilBrownb41b66d2006-10-02 02:17:48 -0700335 closesk = svsk;
NeilBrown39423022010-11-15 11:27:01 +1100336 svc_xprt_get(&closesk->sk_xprt);
337 } else
NeilBrownb41b66d2006-10-02 02:17:48 -0700338 len += onelen;
NeilBrown80212d52006-10-02 02:17:47 -0700339 }
NeilBrownaaf68cf2007-02-08 14:20:30 -0800340 spin_unlock_bh(&serv->sv_lock);
Chuck Levere7942b92009-04-23 19:32:48 -0400341
NeilBrown39423022010-11-15 11:27:01 +1100342 if (closesk) {
NeilBrown5680c442006-10-04 02:15:45 -0700343 /* Should unregister with portmap, but you cannot
344 * unregister just one protocol...
345 */
Tom Tucker7a182082007-12-30 21:07:53 -0600346 svc_close_xprt(&closesk->sk_xprt);
NeilBrown39423022010-11-15 11:27:01 +1100347 svc_xprt_put(&closesk->sk_xprt);
348 } else if (toclose)
NeilBrown37a03472006-10-04 02:15:44 -0700349 return -ENOENT;
NeilBrown80212d52006-10-02 02:17:47 -0700350 return len;
351}
Trond Myklebust24c37672008-12-23 16:30:12 -0500352EXPORT_SYMBOL_GPL(svc_sock_names);
NeilBrown80212d52006-10-02 02:17:47 -0700353
354/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 * Check input queue length
356 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600357static int svc_recv_available(struct svc_sock *svsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 struct socket *sock = svsk->sk_sock;
360 int avail, err;
361
Sridhar Samudralae6242e92006-08-07 20:58:01 -0700362 err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363
364 return (err >= 0)? avail : err;
365}
366
367/*
368 * Generic recvfrom routine.
369 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600370static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr,
371 int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372{
Tom Tucker57b1d3b2007-12-30 21:08:22 -0600373 struct svc_sock *svsk =
374 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
Chuck Lever1ba95102007-02-12 00:53:31 -0800375 struct msghdr msg = {
376 .msg_flags = MSG_DONTWAIT,
377 };
378 int len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379
Tom Tucker260c1d12007-12-30 21:08:29 -0600380 rqstp->rq_xprt_hlen = 0;
381
Chuck Lever1ba95102007-02-12 00:53:31 -0800382 len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen,
383 msg.msg_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
Chuck Lever1ba95102007-02-12 00:53:31 -0800386 svsk, iov[0].iov_base, iov[0].iov_len, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 return len;
388}
389
390/*
391 * Set socket snd and rcv buffer lengths
392 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600393static void svc_sock_setbufsize(struct socket *sock, unsigned int snd,
394 unsigned int rcv)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395{
396#if 0
397 mm_segment_t oldfs;
398 oldfs = get_fs(); set_fs(KERNEL_DS);
399 sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
400 (char*)&snd, sizeof(snd));
401 sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
402 (char*)&rcv, sizeof(rcv));
403#else
404 /* sock_setsockopt limits use to sysctl_?mem_max,
405 * which isn't acceptable. Until that is made conditional
406 * on not having CAP_SYS_RESOURCE or similar, we go direct...
407 * DaveM said I could!
408 */
409 lock_sock(sock->sk);
410 sock->sk->sk_sndbuf = snd * 2;
411 sock->sk->sk_rcvbuf = rcv * 2;
J. Bruce Fields7f421832009-05-27 18:51:06 -0400412 sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK|SOCK_RCVBUF_LOCK;
Trond Myklebust47fcb032009-05-18 17:47:56 -0400413 sock->sk->sk_write_space(sock->sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 release_sock(sock->sk);
415#endif
416}
417/*
418 * INET callback when data has been received on the socket.
419 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600420static void svc_udp_data_ready(struct sock *sk, int count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421{
Neil Brown939bb7e2005-09-13 01:25:39 -0700422 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
Eric Dumazeteaefd112011-02-18 03:26:36 +0000423 wait_queue_head_t *wq = sk_sleep(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424
Neil Brown939bb7e2005-09-13 01:25:39 -0700425 if (svsk) {
426 dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n",
Tom Tucker02fc6c32007-12-30 21:07:48 -0600427 svsk, sk, count,
428 test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
429 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
Tom Tuckerf6150c32007-12-30 21:07:57 -0600430 svc_xprt_enqueue(&svsk->sk_xprt);
Neil Brown939bb7e2005-09-13 01:25:39 -0700431 }
Eric Dumazeteaefd112011-02-18 03:26:36 +0000432 if (wq && waitqueue_active(wq))
433 wake_up_interruptible(wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434}
435
436/*
437 * INET callback when space is newly available on the socket.
438 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600439static void svc_write_space(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440{
441 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
Eric Dumazeteaefd112011-02-18 03:26:36 +0000442 wait_queue_head_t *wq = sk_sleep(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443
444 if (svsk) {
445 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
Tom Tucker02fc6c32007-12-30 21:07:48 -0600446 svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
Tom Tuckerf6150c32007-12-30 21:07:57 -0600447 svc_xprt_enqueue(&svsk->sk_xprt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 }
449
Eric Dumazeteaefd112011-02-18 03:26:36 +0000450 if (wq && waitqueue_active(wq)) {
Neil Brown939bb7e2005-09-13 01:25:39 -0700451 dprintk("RPC svc_write_space: someone sleeping on %p\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 svsk);
Eric Dumazeteaefd112011-02-18 03:26:36 +0000453 wake_up_interruptible(wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 }
455}
456
Trond Myklebust47fcb032009-05-18 17:47:56 -0400457static void svc_tcp_write_space(struct sock *sk)
458{
459 struct socket *sock = sk->sk_socket;
460
461 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && sock)
462 clear_bit(SOCK_NOSPACE, &sock->flags);
463 svc_write_space(sk);
464}
465
Tom Tucker9dbc2402007-12-30 21:08:12 -0600466/*
Chuck Lever7702ce42009-07-13 10:54:26 -0400467 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
468 */
469static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
470 struct cmsghdr *cmh)
471{
472 struct in_pktinfo *pki = CMSG_DATA(cmh);
473 if (cmh->cmsg_type != IP_PKTINFO)
474 return 0;
475 rqstp->rq_daddr.addr.s_addr = pki->ipi_spec_dst.s_addr;
476 return 1;
477}
478
479/*
480 * See net/ipv6/datagram.c : datagram_recv_ctl
481 */
482static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
483 struct cmsghdr *cmh)
484{
485 struct in6_pktinfo *pki = CMSG_DATA(cmh);
486 if (cmh->cmsg_type != IPV6_PKTINFO)
487 return 0;
488 ipv6_addr_copy(&rqstp->rq_daddr.addr6, &pki->ipi6_addr);
489 return 1;
490}
491
492/*
Tom Tucker9dbc2402007-12-30 21:08:12 -0600493 * Copy the UDP datagram's destination address to the rqstp structure.
494 * The 'destination' address in this case is the address to which the
495 * peer sent the datagram, i.e. our local address. For multihomed
496 * hosts, this can change from msg to msg. Note that only the IP
497 * address changes, the port number should remain the same.
498 */
Chuck Lever7702ce42009-07-13 10:54:26 -0400499static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
500 struct cmsghdr *cmh)
Chuck Lever95756482007-02-12 00:53:38 -0800501{
Chuck Lever7702ce42009-07-13 10:54:26 -0400502 switch (cmh->cmsg_level) {
503 case SOL_IP:
504 return svc_udp_get_dest_address4(rqstp, cmh);
505 case SOL_IPV6:
506 return svc_udp_get_dest_address6(rqstp, cmh);
Chuck Lever95756482007-02-12 00:53:38 -0800507 }
Chuck Lever7702ce42009-07-13 10:54:26 -0400508
509 return 0;
Chuck Lever95756482007-02-12 00:53:38 -0800510}
511
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512/*
513 * Receive a datagram from a UDP socket.
514 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600515static int svc_udp_recvfrom(struct svc_rqst *rqstp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516{
Tom Tucker57b1d3b2007-12-30 21:08:22 -0600517 struct svc_sock *svsk =
518 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600519 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 struct sk_buff *skb;
David S. Millerbc375ea2007-04-12 13:35:59 -0700521 union {
522 struct cmsghdr hdr;
523 long all[SVC_PKTINFO_SPACE / sizeof(long)];
524 } buffer;
525 struct cmsghdr *cmh = &buffer.hdr;
NeilBrown7a37f572007-03-06 01:42:21 -0800526 struct msghdr msg = {
527 .msg_name = svc_addr(rqstp),
528 .msg_control = cmh,
529 .msg_controllen = sizeof(buffer),
530 .msg_flags = MSG_DONTWAIT,
531 };
Chuck Leverabc5c442009-04-23 19:31:25 -0400532 size_t len;
533 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534
Tom Tucker02fc6c32007-12-30 21:07:48 -0600535 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 /* udp sockets need large rcvbuf as all pending
537 * requests are still in that buffer. sndbuf must
538 * also be large enough that there is enough space
Greg Banks3262c812006-10-02 02:17:58 -0700539 * for one reply per thread. We count all threads
540 * rather than threads in a particular pool, which
541 * provides an upper bound on the number of threads
542 * which will access the socket.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543 */
544 svc_sock_setbufsize(svsk->sk_sock,
NeilBrownc6b0a9f2006-10-06 00:44:05 -0700545 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
546 (serv->sv_nrthreads+3) * serv->sv_max_mesg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547
Tom Tucker02fc6c32007-12-30 21:07:48 -0600548 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
NeilBrown05ed6902007-05-09 02:34:55 -0700549 skb = NULL;
550 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
551 0, 0, MSG_PEEK | MSG_DONTWAIT);
552 if (err >= 0)
553 skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err);
554
555 if (skb == NULL) {
556 if (err != -EAGAIN) {
557 /* possibly an icmp error */
558 dprintk("svc: recvfrom returned error %d\n", -err);
Tom Tucker02fc6c32007-12-30 21:07:48 -0600559 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 }
NeilBrown05ed6902007-05-09 02:34:55 -0700561 return -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 }
Tom Tucker9dbc2402007-12-30 21:08:12 -0600563 len = svc_addr_len(svc_addr(rqstp));
Chuck Leverabc5c442009-04-23 19:31:25 -0400564 if (len == 0)
565 return -EAFNOSUPPORT;
Tom Tucker9dbc2402007-12-30 21:08:12 -0600566 rqstp->rq_addrlen = len;
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -0700567 if (skb->tstamp.tv64 == 0) {
568 skb->tstamp = ktime_get_real();
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -0800569 /* Don't enable netstamp, sunrpc doesn't
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 need that much accuracy */
571 }
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -0700572 svsk->sk_sk->sk_stamp = skb->tstamp;
Tom Tucker02fc6c32007-12-30 21:07:48 -0600573 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 len = skb->len - sizeof(struct udphdr);
576 rqstp->rq_arg.len = len;
577
Chuck Lever95756482007-02-12 00:53:38 -0800578 rqstp->rq_prot = IPPROTO_UDP;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579
Chuck Lever7702ce42009-07-13 10:54:26 -0400580 if (!svc_udp_get_dest_address(rqstp, cmh)) {
NeilBrown7a37f572007-03-06 01:42:21 -0800581 if (net_ratelimit())
Chuck Lever7702ce42009-07-13 10:54:26 -0400582 printk(KERN_WARNING
583 "svc: received unknown control message %d/%d; "
584 "dropping RPC reply datagram\n",
585 cmh->cmsg_level, cmh->cmsg_type);
Eric Dumazet9d410c72009-10-30 05:03:53 +0000586 skb_free_datagram_locked(svsk->sk_sk, skb);
NeilBrown7a37f572007-03-06 01:42:21 -0800587 return 0;
588 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589
590 if (skb_is_nonlinear(skb)) {
591 /* we have to copy */
592 local_bh_disable();
593 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
594 local_bh_enable();
595 /* checksum error */
Eric Dumazet9d410c72009-10-30 05:03:53 +0000596 skb_free_datagram_locked(svsk->sk_sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 return 0;
598 }
599 local_bh_enable();
Eric Dumazet9d410c72009-10-30 05:03:53 +0000600 skb_free_datagram_locked(svsk->sk_sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 } else {
602 /* we can use it in-place */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600603 rqstp->rq_arg.head[0].iov_base = skb->data +
604 sizeof(struct udphdr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 rqstp->rq_arg.head[0].iov_len = len;
Herbert Xufb286bb2005-11-10 13:01:24 -0800606 if (skb_checksum_complete(skb)) {
Eric Dumazet9d410c72009-10-30 05:03:53 +0000607 skb_free_datagram_locked(svsk->sk_sk, skb);
Herbert Xufb286bb2005-11-10 13:01:24 -0800608 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 }
Tom Tucker5148bf42007-12-30 21:07:25 -0600610 rqstp->rq_xprt_ctxt = skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 }
612
613 rqstp->rq_arg.page_base = 0;
614 if (len <= rqstp->rq_arg.head[0].iov_len) {
615 rqstp->rq_arg.head[0].iov_len = len;
616 rqstp->rq_arg.page_len = 0;
NeilBrown44524352006-10-04 02:15:46 -0700617 rqstp->rq_respages = rqstp->rq_pages+1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 } else {
619 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
NeilBrown44524352006-10-04 02:15:46 -0700620 rqstp->rq_respages = rqstp->rq_pages + 1 +
Ilpo Järvinen172589c2007-08-28 15:50:33 -0700621 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 }
623
624 if (serv->sv_stats)
625 serv->sv_stats->netudpcnt++;
626
627 return len;
628}
629
630static int
631svc_udp_sendto(struct svc_rqst *rqstp)
632{
633 int error;
634
635 error = svc_sendto(rqstp, &rqstp->rq_res);
636 if (error == -ECONNREFUSED)
637 /* ICMP error on earlier request. */
638 error = svc_sendto(rqstp, &rqstp->rq_res);
639
640 return error;
641}
642
Tom Tuckere831fe62007-12-30 21:07:29 -0600643static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
644{
645}
646
Tom Tucker323bee32007-12-30 21:07:31 -0600647static int svc_udp_has_wspace(struct svc_xprt *xprt)
648{
649 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600650 struct svc_serv *serv = xprt->xpt_server;
Tom Tucker323bee32007-12-30 21:07:31 -0600651 unsigned long required;
652
653 /*
654 * Set the SOCK_NOSPACE flag before checking the available
655 * sock space.
656 */
657 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
Tom Tucker7a90e8c2007-12-30 21:07:55 -0600658 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
Tom Tucker323bee32007-12-30 21:07:31 -0600659 if (required*2 > sock_wspace(svsk->sk_sk))
660 return 0;
661 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
662 return 1;
663}
664
Tom Tucker38a417c2007-12-30 21:07:36 -0600665static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
666{
667 BUG();
668 return NULL;
669}
670
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600671static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
Pavel Emelyanov62832c02010-09-29 16:04:18 +0400672 struct net *net,
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600673 struct sockaddr *sa, int salen,
674 int flags)
675{
Pavel Emelyanov62832c02010-09-29 16:04:18 +0400676 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600677}
678
Tom Tucker360d87382007-12-30 21:07:17 -0600679static struct svc_xprt_ops svc_udp_ops = {
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600680 .xpo_create = svc_udp_create,
Tom Tucker5d137992007-12-30 21:07:23 -0600681 .xpo_recvfrom = svc_udp_recvfrom,
682 .xpo_sendto = svc_udp_sendto,
Tom Tucker5148bf42007-12-30 21:07:25 -0600683 .xpo_release_rqst = svc_release_skb,
Tom Tucker755ccea2007-12-30 21:07:27 -0600684 .xpo_detach = svc_sock_detach,
685 .xpo_free = svc_sock_free,
Tom Tuckere831fe62007-12-30 21:07:29 -0600686 .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
Tom Tucker323bee32007-12-30 21:07:31 -0600687 .xpo_has_wspace = svc_udp_has_wspace,
Tom Tucker38a417c2007-12-30 21:07:36 -0600688 .xpo_accept = svc_udp_accept,
Tom Tucker360d87382007-12-30 21:07:17 -0600689};
690
691static struct svc_xprt_class svc_udp_class = {
692 .xcl_name = "udp",
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600693 .xcl_owner = THIS_MODULE,
Tom Tucker360d87382007-12-30 21:07:17 -0600694 .xcl_ops = &svc_udp_ops,
Tom Tucker49023152007-12-30 21:07:21 -0600695 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
Tom Tucker360d87382007-12-30 21:07:17 -0600696};
697
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600698static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699{
Chuck Lever7702ce42009-07-13 10:54:26 -0400700 int err, level, optname, one = 1;
NeilBrown7a37f572007-03-06 01:42:21 -0800701
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600702 svc_xprt_init(&svc_udp_class, &svsk->sk_xprt, serv);
Tom Tuckerdef13d72007-12-30 21:08:08 -0600703 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 svsk->sk_sk->sk_data_ready = svc_udp_data_ready;
705 svsk->sk_sk->sk_write_space = svc_write_space;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706
707 /* initialise setting must have enough space to
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -0800708 * receive and respond to one request.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 * svc_udp_recvfrom will re-adjust if necessary
710 */
711 svc_sock_setbufsize(svsk->sk_sock,
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600712 3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
713 3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714
Tom Tucker0f0257e2007-12-30 21:08:27 -0600715 /* data might have come in before data_ready set up */
716 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
Tom Tucker02fc6c32007-12-30 21:07:48 -0600717 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
NeilBrown7a37f572007-03-06 01:42:21 -0800718
NeilBrown7a37f572007-03-06 01:42:21 -0800719 /* make sure we get destination address info */
Chuck Lever7702ce42009-07-13 10:54:26 -0400720 switch (svsk->sk_sk->sk_family) {
721 case AF_INET:
722 level = SOL_IP;
723 optname = IP_PKTINFO;
724 break;
725 case AF_INET6:
726 level = SOL_IPV6;
727 optname = IPV6_RECVPKTINFO;
728 break;
729 default:
730 BUG();
731 }
732 err = kernel_setsockopt(svsk->sk_sock, level, optname,
733 (char *)&one, sizeof(one));
734 dprintk("svc: kernel_setsockopt returned %d\n", err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735}
736
737/*
738 * A data_ready event on a listening socket means there's a connection
739 * pending. Do not use state_change as a substitute for it.
740 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600741static void svc_tcp_listen_data_ready(struct sock *sk, int count_unused)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742{
Neil Brown939bb7e2005-09-13 01:25:39 -0700743 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
Eric Dumazeteaefd112011-02-18 03:26:36 +0000744 wait_queue_head_t *wq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745
746 dprintk("svc: socket %p TCP (listen) state change %d\n",
Neil Brown939bb7e2005-09-13 01:25:39 -0700747 sk, sk->sk_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748
Neil Brown939bb7e2005-09-13 01:25:39 -0700749 /*
750 * This callback may called twice when a new connection
751 * is established as a child socket inherits everything
752 * from a parent LISTEN socket.
753 * 1) data_ready method of the parent socket will be called
754 * when one of child sockets become ESTABLISHED.
755 * 2) data_ready method of the child socket may be called
756 * when it receives data before the socket is accepted.
757 * In case of 2, we should ignore it silently.
758 */
759 if (sk->sk_state == TCP_LISTEN) {
760 if (svsk) {
Tom Tucker02fc6c32007-12-30 21:07:48 -0600761 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
Tom Tuckerf6150c32007-12-30 21:07:57 -0600762 svc_xprt_enqueue(&svsk->sk_xprt);
Neil Brown939bb7e2005-09-13 01:25:39 -0700763 } else
764 printk("svc: socket %p: no user data\n", sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 }
Neil Brown939bb7e2005-09-13 01:25:39 -0700766
Eric Dumazeteaefd112011-02-18 03:26:36 +0000767 wq = sk_sleep(sk);
768 if (wq && waitqueue_active(wq))
769 wake_up_interruptible_all(wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770}
771
772/*
773 * A state change on a connected socket means it's dying or dead.
774 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600775static void svc_tcp_state_change(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776{
Neil Brown939bb7e2005-09-13 01:25:39 -0700777 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
Eric Dumazeteaefd112011-02-18 03:26:36 +0000778 wait_queue_head_t *wq = sk_sleep(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779
780 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
Neil Brown939bb7e2005-09-13 01:25:39 -0700781 sk, sk->sk_state, sk->sk_user_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782
Neil Brown939bb7e2005-09-13 01:25:39 -0700783 if (!svsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 printk("svc: socket %p: no user data\n", sk);
Neil Brown939bb7e2005-09-13 01:25:39 -0700785 else {
Tom Tucker02fc6c32007-12-30 21:07:48 -0600786 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
Tom Tuckerf6150c32007-12-30 21:07:57 -0600787 svc_xprt_enqueue(&svsk->sk_xprt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 }
Eric Dumazeteaefd112011-02-18 03:26:36 +0000789 if (wq && waitqueue_active(wq))
790 wake_up_interruptible_all(wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791}
792
Tom Tucker0f0257e2007-12-30 21:08:27 -0600793static void svc_tcp_data_ready(struct sock *sk, int count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794{
Neil Brown939bb7e2005-09-13 01:25:39 -0700795 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
Eric Dumazeteaefd112011-02-18 03:26:36 +0000796 wait_queue_head_t *wq = sk_sleep(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797
798 dprintk("svc: socket %p TCP data ready (svsk %p)\n",
Neil Brown939bb7e2005-09-13 01:25:39 -0700799 sk, sk->sk_user_data);
800 if (svsk) {
Tom Tucker02fc6c32007-12-30 21:07:48 -0600801 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
Tom Tuckerf6150c32007-12-30 21:07:57 -0600802 svc_xprt_enqueue(&svsk->sk_xprt);
Neil Brown939bb7e2005-09-13 01:25:39 -0700803 }
Eric Dumazeteaefd112011-02-18 03:26:36 +0000804 if (wq && waitqueue_active(wq))
805 wake_up_interruptible(wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806}
807
808/*
809 * Accept a TCP connection
810 */
Tom Tucker38a417c2007-12-30 21:07:36 -0600811static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812{
Tom Tucker38a417c2007-12-30 21:07:36 -0600813 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
akpm@linux-foundation.orgcdd88b92007-02-12 00:53:38 -0800814 struct sockaddr_storage addr;
815 struct sockaddr *sin = (struct sockaddr *) &addr;
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600816 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 struct socket *sock = svsk->sk_sock;
818 struct socket *newsock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819 struct svc_sock *newsvsk;
820 int err, slen;
Pavel Emelyanov5216a8e2008-02-21 10:57:45 +0300821 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822
823 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
824 if (!sock)
Tom Tucker38a417c2007-12-30 21:07:36 -0600825 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826
Tom Tucker02fc6c32007-12-30 21:07:48 -0600827 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
Sridhar Samudralae6242e92006-08-07 20:58:01 -0700828 err = kernel_accept(sock, &newsock, O_NONBLOCK);
829 if (err < 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 if (err == -ENOMEM)
831 printk(KERN_WARNING "%s: no more sockets!\n",
832 serv->sv_name);
Sridhar Samudralae6242e92006-08-07 20:58:01 -0700833 else if (err != -EAGAIN && net_ratelimit())
834 printk(KERN_WARNING "%s: accept failed (err %d)!\n",
835 serv->sv_name, -err);
Tom Tucker38a417c2007-12-30 21:07:36 -0600836 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 }
Tom Tucker02fc6c32007-12-30 21:07:48 -0600838 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839
akpm@linux-foundation.orgcdd88b92007-02-12 00:53:38 -0800840 err = kernel_getpeername(newsock, sin, &slen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 if (err < 0) {
842 if (net_ratelimit())
843 printk(KERN_WARNING "%s: peername failed (err %d)!\n",
844 serv->sv_name, -err);
845 goto failed; /* aborted connection or whatever */
846 }
847
848 /* Ideally, we would want to reject connections from unauthorized
Chuck Leverad06e4b2007-02-12 00:53:32 -0800849 * hosts here, but when we get encryption, the IP of the host won't
850 * tell us anything. For now just warn about unpriv connections.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 */
akpm@linux-foundation.orgcdd88b92007-02-12 00:53:38 -0800852 if (!svc_port_is_privileged(sin)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 dprintk(KERN_WARNING
Chuck Leverad06e4b2007-02-12 00:53:32 -0800854 "%s: connect from unprivileged port: %s\n",
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -0800855 serv->sv_name,
akpm@linux-foundation.orgcdd88b92007-02-12 00:53:38 -0800856 __svc_print_addr(sin, buf, sizeof(buf)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 }
Chuck Leverad06e4b2007-02-12 00:53:32 -0800858 dprintk("%s: connect from %s\n", serv->sv_name,
akpm@linux-foundation.orgcdd88b92007-02-12 00:53:38 -0800859 __svc_print_addr(sin, buf, sizeof(buf)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860
861 /* make sure that a write doesn't block forever when
862 * low on memory
863 */
864 newsock->sk->sk_sndtimeo = HZ*30;
865
Chuck Lever6b174332007-02-12 00:53:28 -0800866 if (!(newsvsk = svc_setup_socket(serv, newsock, &err,
867 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY))))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 goto failed;
Tom Tucker9dbc2402007-12-30 21:08:12 -0600869 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
Frank van Maarseveena9747692007-07-09 22:21:39 +0200870 err = kernel_getsockname(newsock, sin, &slen);
871 if (unlikely(err < 0)) {
872 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
873 slen = offsetof(struct sockaddr, sa_data);
874 }
Tom Tucker9dbc2402007-12-30 21:08:12 -0600875 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
Chuck Lever067d7812007-02-12 00:53:30 -0800876
Tom Tuckerf9f3cc42007-12-30 21:07:40 -0600877 if (serv->sv_stats)
878 serv->sv_stats->nettcpconn++;
879
880 return &newsvsk->sk_xprt;
881
882failed:
883 sock_release(newsock);
884 return NULL;
885}
886
887/*
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +0300888 * Receive data.
889 * If we haven't gotten the record length yet, get the next four bytes.
890 * Otherwise try to gobble up as much as possible up to the complete
891 * record length.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892 */
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +0300893static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894{
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600895 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +0300896 int len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897
J. Bruce Fields7f421832009-05-27 18:51:06 -0400898 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
899 /* sndbuf needs to have room for one request
900 * per thread, otherwise we can stall even when the
901 * network isn't a bottleneck.
902 *
903 * We count all threads rather than threads in a
904 * particular pool, which provides an upper bound
905 * on the number of threads which will access the socket.
906 *
907 * rcvbuf just needs to be able to hold a few requests.
908 * Normally they will be removed from the queue
909 * as soon a a complete request arrives.
910 */
911 svc_sock_setbufsize(svsk->sk_sock,
912 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
913 3 * serv->sv_max_mesg);
914
Tom Tucker02fc6c32007-12-30 21:07:48 -0600915 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916
Chuck Leverc0401ea2008-04-14 12:27:30 -0400917 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
918 int want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 struct kvec iov;
920
921 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
922 iov.iov_len = want;
923 if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
924 goto error;
925 svsk->sk_tcplen += len;
926
927 if (len < want) {
Chuck Leverc0401ea2008-04-14 12:27:30 -0400928 dprintk("svc: short recvfrom while reading record "
929 "length (%d of %d)\n", len, want);
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +0300930 goto err_again; /* record header not complete */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931 }
932
933 svsk->sk_reclen = ntohl(svsk->sk_reclen);
Chuck Leverc0401ea2008-04-14 12:27:30 -0400934 if (!(svsk->sk_reclen & RPC_LAST_STREAM_FRAGMENT)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 /* FIXME: technically, a record can be fragmented,
936 * and non-terminal fragments will not have the top
937 * bit set in the fragment length header.
938 * But apparently no known nfs clients send fragmented
939 * records. */
NeilBrown34e9a632007-01-29 13:19:52 -0800940 if (net_ratelimit())
Chuck Leverc0401ea2008-04-14 12:27:30 -0400941 printk(KERN_NOTICE "RPC: multiple fragments "
942 "per record not supported\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943 goto err_delete;
944 }
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +0300945
Chuck Leverc0401ea2008-04-14 12:27:30 -0400946 svsk->sk_reclen &= RPC_FRAGMENT_SIZE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen);
NeilBrownc6b0a9f2006-10-06 00:44:05 -0700948 if (svsk->sk_reclen > serv->sv_max_mesg) {
NeilBrown34e9a632007-01-29 13:19:52 -0800949 if (net_ratelimit())
Chuck Leverc0401ea2008-04-14 12:27:30 -0400950 printk(KERN_NOTICE "RPC: "
951 "fragment too large: 0x%08lx\n",
952 (unsigned long)svsk->sk_reclen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 goto err_delete;
954 }
955 }
956
957 /* Check whether enough data is available */
958 len = svc_recv_available(svsk);
959 if (len < 0)
960 goto error;
961
962 if (len < svsk->sk_reclen) {
963 dprintk("svc: incomplete TCP record (%d of %d)\n",
964 len, svsk->sk_reclen);
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +0300965 goto err_again; /* record not complete */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 }
967 len = svsk->sk_reclen;
Tom Tucker02fc6c32007-12-30 21:07:48 -0600968 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +0300970 return len;
971 error:
Neil Brownb48fa6b2010-03-01 16:51:14 +1100972 if (len == -EAGAIN)
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +0300973 dprintk("RPC: TCP recv_record got EAGAIN\n");
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +0300974 return len;
975 err_delete:
976 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
977 err_again:
978 return -EAGAIN;
979}
980
Rahul Iyer4cfc7e62009-09-10 17:32:28 +0300981static int svc_process_calldir(struct svc_sock *svsk, struct svc_rqst *rqstp,
982 struct rpc_rqst **reqpp, struct kvec *vec)
983{
984 struct rpc_rqst *req = NULL;
985 u32 *p;
986 u32 xid;
987 u32 calldir;
988 int len;
989
990 len = svc_recvfrom(rqstp, vec, 1, 8);
991 if (len < 0)
992 goto error;
993
994 p = (u32 *)rqstp->rq_arg.head[0].iov_base;
995 xid = *p++;
996 calldir = *p;
997
998 if (calldir == 0) {
999 /* REQUEST is the most common case */
1000 vec[0] = rqstp->rq_arg.head[0];
1001 } else {
1002 /* REPLY */
J. Bruce Fieldsd75faea2010-11-30 19:15:01 -05001003 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
1004
1005 if (bc_xprt)
1006 req = xprt_lookup_rqst(bc_xprt, xid);
Rahul Iyer4cfc7e62009-09-10 17:32:28 +03001007
1008 if (!req) {
1009 printk(KERN_NOTICE
1010 "%s: Got unrecognized reply: "
J. Bruce Fieldsd75faea2010-11-30 19:15:01 -05001011 "calldir 0x%x xpt_bc_xprt %p xid %08x\n",
Rahul Iyer4cfc7e62009-09-10 17:32:28 +03001012 __func__, ntohl(calldir),
J. Bruce Fieldsd75faea2010-11-30 19:15:01 -05001013 bc_xprt, xid);
Rahul Iyer4cfc7e62009-09-10 17:32:28 +03001014 vec[0] = rqstp->rq_arg.head[0];
1015 goto out;
1016 }
1017
1018 memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
1019 sizeof(struct xdr_buf));
1020 /* copy the xid and call direction */
1021 memcpy(req->rq_private_buf.head[0].iov_base,
1022 rqstp->rq_arg.head[0].iov_base, 8);
1023 vec[0] = req->rq_private_buf.head[0];
1024 }
1025 out:
1026 vec[0].iov_base += 8;
1027 vec[0].iov_len -= 8;
1028 len = svsk->sk_reclen - 8;
1029 error:
1030 *reqpp = req;
1031 return len;
1032}
1033
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +03001034/*
1035 * Receive data from a TCP socket.
1036 */
1037static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
1038{
1039 struct svc_sock *svsk =
1040 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
1041 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1042 int len;
1043 struct kvec *vec;
1044 int pnum, vlen;
Rahul Iyer4cfc7e62009-09-10 17:32:28 +03001045 struct rpc_rqst *req = NULL;
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +03001046
1047 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1048 svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
1049 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
1050 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1051
1052 len = svc_tcp_recv_record(svsk, rqstp);
1053 if (len < 0)
1054 goto error;
1055
NeilBrown3cc03b12006-10-04 02:15:47 -07001056 vec = rqstp->rq_vec;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 vec[0] = rqstp->rq_arg.head[0];
1058 vlen = PAGE_SIZE;
Rahul Iyer4cfc7e62009-09-10 17:32:28 +03001059
1060 /*
1061 * We have enough data for the whole tcp record. Let's try and read the
1062 * first 8 bytes to get the xid and the call direction. We can use this
1063 * to figure out if this is a call or a reply to a callback. If
1064 * sk_reclen is < 8 (xid and calldir), then this is a malformed packet.
1065 * In that case, don't bother with the calldir and just read the data.
1066 * It will be rejected in svc_process.
1067 */
1068 if (len >= 8) {
1069 len = svc_process_calldir(svsk, rqstp, &req, vec);
1070 if (len < 0)
1071 goto err_again;
1072 vlen -= 8;
1073 }
1074
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 pnum = 1;
1076 while (vlen < len) {
Rahul Iyer4cfc7e62009-09-10 17:32:28 +03001077 vec[pnum].iov_base = (req) ?
1078 page_address(req->rq_private_buf.pages[pnum - 1]) :
1079 page_address(rqstp->rq_pages[pnum]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080 vec[pnum].iov_len = PAGE_SIZE;
1081 pnum++;
1082 vlen += PAGE_SIZE;
1083 }
NeilBrown44524352006-10-04 02:15:46 -07001084 rqstp->rq_respages = &rqstp->rq_pages[pnum];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085
1086 /* Now receive data */
1087 len = svc_recvfrom(rqstp, vec, pnum, len);
1088 if (len < 0)
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +03001089 goto err_again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090
Rahul Iyer4cfc7e62009-09-10 17:32:28 +03001091 /*
1092 * Account for the 8 bytes we read earlier
1093 */
1094 len += 8;
1095
1096 if (req) {
1097 xprt_complete_rqst(req->rq_task, len);
1098 len = 0;
1099 goto out;
1100 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 dprintk("svc: TCP complete record (%d bytes)\n", len);
1102 rqstp->rq_arg.len = len;
1103 rqstp->rq_arg.page_base = 0;
1104 if (len <= rqstp->rq_arg.head[0].iov_len) {
1105 rqstp->rq_arg.head[0].iov_len = len;
1106 rqstp->rq_arg.page_len = 0;
1107 } else {
1108 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
1109 }
1110
Tom Tucker5148bf42007-12-30 21:07:25 -06001111 rqstp->rq_xprt_ctxt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112 rqstp->rq_prot = IPPROTO_TCP;
1113
Rahul Iyer4cfc7e62009-09-10 17:32:28 +03001114out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115 /* Reset TCP read info */
1116 svsk->sk_reclen = 0;
1117 svsk->sk_tcplen = 0;
1118
Tom Tucker9dbc2402007-12-30 21:08:12 -06001119 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120 if (serv->sv_stats)
1121 serv->sv_stats->nettcpcnt++;
1122
1123 return len;
1124
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +03001125err_again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 if (len == -EAGAIN) {
1127 dprintk("RPC: TCP recvfrom got EAGAIN\n");
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +03001128 return len;
1129 }
1130error:
1131 if (len != -EAGAIN) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
Tom Tuckerbb5cf162007-12-30 21:07:50 -06001133 svsk->sk_xprt.xpt_server->sv_name, -len);
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +03001134 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 }
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +03001136 return -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137}
1138
1139/*
1140 * Send out data on TCP socket.
1141 */
Tom Tucker0f0257e2007-12-30 21:08:27 -06001142static int svc_tcp_sendto(struct svc_rqst *rqstp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143{
1144 struct xdr_buf *xbufp = &rqstp->rq_res;
1145 int sent;
Alexey Dobriyand8ed0292006-09-26 22:29:38 -07001146 __be32 reclen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147
1148 /* Set up the first element of the reply kvec.
1149 * Any other kvecs that may be in use have been taken
1150 * care of by the server implementation itself.
1151 */
1152 reclen = htonl(0x80000000|((xbufp->len ) - 4));
1153 memcpy(xbufp->head[0].iov_base, &reclen, 4);
1154
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155 sent = svc_sendto(rqstp, &rqstp->rq_res);
1156 if (sent != xbufp->len) {
Tom Tucker0f0257e2007-12-30 21:08:27 -06001157 printk(KERN_NOTICE
1158 "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1159 "- shutting down socket\n",
Tom Tucker57b1d3b2007-12-30 21:08:22 -06001160 rqstp->rq_xprt->xpt_server->sv_name,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 (sent<0)?"got error":"sent only",
1162 sent, xbufp->len);
Tom Tucker57b1d3b2007-12-30 21:08:22 -06001163 set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
Tom Tuckerf6150c32007-12-30 21:07:57 -06001164 svc_xprt_enqueue(rqstp->rq_xprt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165 sent = -EAGAIN;
1166 }
1167 return sent;
1168}
1169
Tom Tuckere831fe62007-12-30 21:07:29 -06001170/*
1171 * Setup response header. TCP has a 4B record length field.
1172 */
1173static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
1174{
1175 struct kvec *resv = &rqstp->rq_res.head[0];
1176
1177 /* tcp needs a space for the record length... */
1178 svc_putnl(resv, 0);
1179}
1180
Tom Tucker323bee32007-12-30 21:07:31 -06001181static int svc_tcp_has_wspace(struct svc_xprt *xprt)
1182{
Tom Tucker57b1d3b2007-12-30 21:08:22 -06001183 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
Trond Myklebust47fcb032009-05-18 17:47:56 -04001184 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
Tom Tucker323bee32007-12-30 21:07:31 -06001185 int required;
Tom Tucker323bee32007-12-30 21:07:31 -06001186
Trond Myklebust47fcb032009-05-18 17:47:56 -04001187 if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
1188 return 1;
1189 required = atomic_read(&xprt->xpt_reserved) + serv->sv_max_mesg;
1190 if (sk_stream_wspace(svsk->sk_sk) >= required)
1191 return 1;
Tom Tucker323bee32007-12-30 21:07:31 -06001192 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
Trond Myklebust47fcb032009-05-18 17:47:56 -04001193 return 0;
Tom Tucker323bee32007-12-30 21:07:31 -06001194}
1195
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001196static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
Pavel Emelyanov62832c02010-09-29 16:04:18 +04001197 struct net *net,
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001198 struct sockaddr *sa, int salen,
1199 int flags)
1200{
Pavel Emelyanov62832c02010-09-29 16:04:18 +04001201 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001202}
1203
Andy Adamson1f11a032011-01-06 02:04:26 +00001204#if defined(CONFIG_NFS_V4_1)
1205static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
1206 struct net *, struct sockaddr *,
1207 int, int);
1208static void svc_bc_sock_free(struct svc_xprt *xprt);
1209
1210static struct svc_xprt *svc_bc_tcp_create(struct svc_serv *serv,
1211 struct net *net,
1212 struct sockaddr *sa, int salen,
1213 int flags)
1214{
1215 return svc_bc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1216}
1217
1218static void svc_bc_tcp_sock_detach(struct svc_xprt *xprt)
1219{
1220}
1221
1222static struct svc_xprt_ops svc_tcp_bc_ops = {
1223 .xpo_create = svc_bc_tcp_create,
1224 .xpo_detach = svc_bc_tcp_sock_detach,
1225 .xpo_free = svc_bc_sock_free,
1226 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1227};
1228
1229static struct svc_xprt_class svc_tcp_bc_class = {
1230 .xcl_name = "tcp-bc",
1231 .xcl_owner = THIS_MODULE,
1232 .xcl_ops = &svc_tcp_bc_ops,
1233 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1234};
Andy Adamson16b2d1e2011-01-06 02:04:27 +00001235
1236static void svc_init_bc_xprt_sock(void)
1237{
1238 svc_reg_xprt_class(&svc_tcp_bc_class);
1239}
1240
1241static void svc_cleanup_bc_xprt_sock(void)
1242{
1243 svc_unreg_xprt_class(&svc_tcp_bc_class);
1244}
1245#else /* CONFIG_NFS_V4_1 */
1246static void svc_init_bc_xprt_sock(void)
1247{
1248}
1249
1250static void svc_cleanup_bc_xprt_sock(void)
1251{
1252}
Andy Adamson1f11a032011-01-06 02:04:26 +00001253#endif /* CONFIG_NFS_V4_1 */
1254
Tom Tucker360d87382007-12-30 21:07:17 -06001255static struct svc_xprt_ops svc_tcp_ops = {
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001256 .xpo_create = svc_tcp_create,
Tom Tucker5d137992007-12-30 21:07:23 -06001257 .xpo_recvfrom = svc_tcp_recvfrom,
1258 .xpo_sendto = svc_tcp_sendto,
Tom Tucker5148bf42007-12-30 21:07:25 -06001259 .xpo_release_rqst = svc_release_skb,
Trond Myklebust69b6ba32008-12-23 16:30:11 -05001260 .xpo_detach = svc_tcp_sock_detach,
Tom Tucker755ccea2007-12-30 21:07:27 -06001261 .xpo_free = svc_sock_free,
Tom Tuckere831fe62007-12-30 21:07:29 -06001262 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
Tom Tucker323bee32007-12-30 21:07:31 -06001263 .xpo_has_wspace = svc_tcp_has_wspace,
Tom Tucker38a417c2007-12-30 21:07:36 -06001264 .xpo_accept = svc_tcp_accept,
Tom Tucker360d87382007-12-30 21:07:17 -06001265};
1266
1267static struct svc_xprt_class svc_tcp_class = {
1268 .xcl_name = "tcp",
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001269 .xcl_owner = THIS_MODULE,
Tom Tucker360d87382007-12-30 21:07:17 -06001270 .xcl_ops = &svc_tcp_ops,
Tom Tucker49023152007-12-30 21:07:21 -06001271 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
Tom Tucker360d87382007-12-30 21:07:17 -06001272};
1273
1274void svc_init_xprt_sock(void)
1275{
1276 svc_reg_xprt_class(&svc_tcp_class);
1277 svc_reg_xprt_class(&svc_udp_class);
Andy Adamson16b2d1e2011-01-06 02:04:27 +00001278 svc_init_bc_xprt_sock();
Tom Tucker360d87382007-12-30 21:07:17 -06001279}
1280
1281void svc_cleanup_xprt_sock(void)
1282{
1283 svc_unreg_xprt_class(&svc_tcp_class);
1284 svc_unreg_xprt_class(&svc_udp_class);
Andy Adamson16b2d1e2011-01-06 02:04:27 +00001285 svc_cleanup_bc_xprt_sock();
Tom Tucker360d87382007-12-30 21:07:17 -06001286}
1287
Tom Tuckerbb5cf162007-12-30 21:07:50 -06001288static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289{
1290 struct sock *sk = svsk->sk_sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291
Tom Tuckerbb5cf162007-12-30 21:07:50 -06001292 svc_xprt_init(&svc_tcp_class, &svsk->sk_xprt, serv);
Tom Tuckerdef13d72007-12-30 21:08:08 -06001293 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294 if (sk->sk_state == TCP_LISTEN) {
1295 dprintk("setting up TCP socket for listening\n");
Tom Tucker02fc6c32007-12-30 21:07:48 -06001296 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297 sk->sk_data_ready = svc_tcp_listen_data_ready;
Tom Tucker02fc6c32007-12-30 21:07:48 -06001298 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299 } else {
1300 dprintk("setting up TCP socket for reading\n");
1301 sk->sk_state_change = svc_tcp_state_change;
1302 sk->sk_data_ready = svc_tcp_data_ready;
Trond Myklebust47fcb032009-05-18 17:47:56 -04001303 sk->sk_write_space = svc_tcp_write_space;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304
1305 svsk->sk_reclen = 0;
1306 svsk->sk_tcplen = 0;
1307
Chuck Leverb7872fe2008-04-14 12:27:01 -04001308 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309
J. Bruce Fields7f421832009-05-27 18:51:06 -04001310 /* initialise setting must have enough space to
1311 * receive and respond to one request.
1312 * svc_tcp_recvfrom will re-adjust if necessary
1313 */
1314 svc_sock_setbufsize(svsk->sk_sock,
1315 3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
1316 3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
1317
1318 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
Tom Tucker02fc6c32007-12-30 21:07:48 -06001319 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -08001320 if (sk->sk_state != TCP_ESTABLISHED)
Tom Tucker02fc6c32007-12-30 21:07:48 -06001321 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322 }
1323}
1324
Tom Tucker0f0257e2007-12-30 21:08:27 -06001325void svc_sock_update_bufs(struct svc_serv *serv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326{
1327 /*
1328 * The number of server threads has changed. Update
1329 * rcvbuf and sndbuf accordingly on all sockets
1330 */
Pavel Emelyanov8f3a6de2010-10-05 23:30:19 +04001331 struct svc_sock *svsk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332
1333 spin_lock_bh(&serv->sv_lock);
Pavel Emelyanov8f3a6de2010-10-05 23:30:19 +04001334 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
Tom Tucker02fc6c32007-12-30 21:07:48 -06001335 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
Pavel Emelyanov8f3a6de2010-10-05 23:30:19 +04001336 list_for_each_entry(svsk, &serv->sv_tempsocks, sk_xprt.xpt_list)
Tom Tucker02fc6c32007-12-30 21:07:48 -06001337 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338 spin_unlock_bh(&serv->sv_lock);
1339}
Trond Myklebust24c37672008-12-23 16:30:12 -05001340EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341
1342/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 * Initialize socket for RPC use and create svc_sock struct
1344 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1345 */
Chuck Lever6b174332007-02-12 00:53:28 -08001346static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1347 struct socket *sock,
1348 int *errp, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349{
1350 struct svc_sock *svsk;
1351 struct sock *inet;
Chuck Lever6b174332007-02-12 00:53:28 -08001352 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353
1354 dprintk("svc: svc_setup_socket %p\n", sock);
Panagiotis Issaris0da974f2006-07-21 14:51:30 -07001355 if (!(svsk = kzalloc(sizeof(*svsk), GFP_KERNEL))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 *errp = -ENOMEM;
1357 return NULL;
1358 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359
1360 inet = sock->sk;
1361
1362 /* Register socket with portmapper */
1363 if (*errp >= 0 && pmap_register)
Chuck Leverbaf01ca2009-03-18 20:46:13 -04001364 *errp = svc_register(serv, inet->sk_family, inet->sk_protocol,
Eric Dumazetc720c7e2009-10-15 06:30:45 +00001365 ntohs(inet_sk(inet)->inet_sport));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366
1367 if (*errp < 0) {
1368 kfree(svsk);
1369 return NULL;
1370 }
1371
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 inet->sk_user_data = svsk;
1373 svsk->sk_sock = sock;
1374 svsk->sk_sk = inet;
1375 svsk->sk_ostate = inet->sk_state_change;
1376 svsk->sk_odata = inet->sk_data_ready;
1377 svsk->sk_owspace = inet->sk_write_space;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378
1379 /* Initialize the socket */
1380 if (sock->type == SOCK_DGRAM)
Tom Tuckerbb5cf162007-12-30 21:07:50 -06001381 svc_udp_init(svsk, serv);
J. Bruce Fields7f421832009-05-27 18:51:06 -04001382 else
Tom Tuckerbb5cf162007-12-30 21:07:50 -06001383 svc_tcp_init(svsk, serv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1386 svsk, svsk->sk_sk);
1387
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 return svsk;
1389}
1390
Chuck Leverbfba9ab2009-04-23 19:32:33 -04001391/**
1392 * svc_addsock - add a listener socket to an RPC service
1393 * @serv: pointer to RPC service to which to add a new listener
1394 * @fd: file descriptor of the new listener
1395 * @name_return: pointer to buffer to fill in with name of listener
1396 * @len: size of the buffer
1397 *
1398 * Fills in socket name and returns positive length of name if successful.
1399 * Name is terminated with '\n'. On error, returns a negative errno
1400 * value.
1401 */
1402int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1403 const size_t len)
NeilBrownb41b66d2006-10-02 02:17:48 -07001404{
1405 int err = 0;
1406 struct socket *so = sockfd_lookup(fd, &err);
1407 struct svc_sock *svsk = NULL;
1408
1409 if (!so)
1410 return err;
Aime Le Rouzic205ba422010-01-26 14:03:56 -05001411 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
NeilBrownb41b66d2006-10-02 02:17:48 -07001412 err = -EAFNOSUPPORT;
1413 else if (so->sk->sk_protocol != IPPROTO_TCP &&
1414 so->sk->sk_protocol != IPPROTO_UDP)
1415 err = -EPROTONOSUPPORT;
1416 else if (so->state > SS_UNCONNECTED)
1417 err = -EISCONN;
1418 else {
Tom Tucker2da2c212008-11-23 09:58:08 -06001419 if (!try_module_get(THIS_MODULE))
1420 err = -ENOENT;
1421 else
1422 svsk = svc_setup_socket(serv, so, &err,
1423 SVC_SOCK_DEFAULTS);
NeilBrowne79eff12007-02-12 00:53:30 -08001424 if (svsk) {
Tom Tucker9dbc2402007-12-30 21:08:12 -06001425 struct sockaddr_storage addr;
1426 struct sockaddr *sin = (struct sockaddr *)&addr;
1427 int salen;
1428 if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0)
1429 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
Tom Tucker4e5caaa2007-12-30 21:08:20 -06001430 clear_bit(XPT_TEMP, &svsk->sk_xprt.xpt_flags);
1431 spin_lock_bh(&serv->sv_lock);
1432 list_add(&svsk->sk_xprt.xpt_list, &serv->sv_permsocks);
1433 spin_unlock_bh(&serv->sv_lock);
Tom Tuckera6046f72007-12-30 21:08:01 -06001434 svc_xprt_received(&svsk->sk_xprt);
NeilBrownb41b66d2006-10-02 02:17:48 -07001435 err = 0;
Tom Tucker2da2c212008-11-23 09:58:08 -06001436 } else
1437 module_put(THIS_MODULE);
NeilBrownb41b66d2006-10-02 02:17:48 -07001438 }
1439 if (err) {
1440 sockfd_put(so);
1441 return err;
1442 }
Chuck Levere7942b92009-04-23 19:32:48 -04001443 return svc_one_sock_name(svsk, name_return, len);
NeilBrownb41b66d2006-10-02 02:17:48 -07001444}
1445EXPORT_SYMBOL_GPL(svc_addsock);
1446
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447/*
1448 * Create socket for RPC service.
1449 */
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001450static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1451 int protocol,
Pavel Emelyanov62832c02010-09-29 16:04:18 +04001452 struct net *net,
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001453 struct sockaddr *sin, int len,
1454 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455{
1456 struct svc_sock *svsk;
1457 struct socket *sock;
1458 int error;
1459 int type;
Tom Tucker9dbc2402007-12-30 21:08:12 -06001460 struct sockaddr_storage addr;
1461 struct sockaddr *newsin = (struct sockaddr *)&addr;
1462 int newlen;
Trond Myklebustc69da772009-03-30 18:59:17 -04001463 int family;
1464 int val;
Pavel Emelyanov5216a8e2008-02-21 10:57:45 +03001465 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466
Chuck Leverad06e4b2007-02-12 00:53:32 -08001467 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1468 serv->sv_program->pg_name, protocol,
Chuck Lever77f1f672007-02-12 00:53:39 -08001469 __svc_print_addr(sin, buf, sizeof(buf)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470
1471 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1472 printk(KERN_WARNING "svc: only UDP and TCP "
1473 "sockets supported\n");
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001474 return ERR_PTR(-EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476
Trond Myklebustc69da772009-03-30 18:59:17 -04001477 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1478 switch (sin->sa_family) {
1479 case AF_INET6:
1480 family = PF_INET6;
1481 break;
1482 case AF_INET:
1483 family = PF_INET;
1484 break;
1485 default:
1486 return ERR_PTR(-EINVAL);
1487 }
1488
Pavel Emelyanov14ec63c2010-09-29 16:06:57 +04001489 error = __sock_create(net, family, type, protocol, &sock, 1);
Chuck Lever77f1f672007-02-12 00:53:39 -08001490 if (error < 0)
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001491 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492
Peter Zijlstraed075362006-12-06 20:35:24 -08001493 svc_reclassify_socket(sock);
1494
Trond Myklebustc69da772009-03-30 18:59:17 -04001495 /*
1496 * If this is an PF_INET6 listener, we want to avoid
1497 * getting requests from IPv4 remotes. Those should
1498 * be shunted to a PF_INET listener via rpcbind.
1499 */
1500 val = 1;
1501 if (family == PF_INET6)
1502 kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
1503 (char *)&val, sizeof(val));
1504
Eric Sesterhenn18114742006-09-28 14:37:07 -07001505 if (type == SOCK_STREAM)
Chuck Lever77f1f672007-02-12 00:53:39 -08001506 sock->sk->sk_reuse = 1; /* allow address reuse */
1507 error = kernel_bind(sock, sin, len);
Eric Sesterhenn18114742006-09-28 14:37:07 -07001508 if (error < 0)
1509 goto bummer;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510
Tom Tucker9dbc2402007-12-30 21:08:12 -06001511 newlen = len;
1512 error = kernel_getsockname(sock, newsin, &newlen);
1513 if (error < 0)
1514 goto bummer;
1515
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516 if (protocol == IPPROTO_TCP) {
Sridhar Samudralae6242e92006-08-07 20:58:01 -07001517 if ((error = kernel_listen(sock, 64)) < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518 goto bummer;
1519 }
1520
NeilBrowne79eff12007-02-12 00:53:30 -08001521 if ((svsk = svc_setup_socket(serv, sock, &error, flags)) != NULL) {
Tom Tucker9dbc2402007-12-30 21:08:12 -06001522 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001523 return (struct svc_xprt *)svsk;
NeilBrowne79eff12007-02-12 00:53:30 -08001524 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525
1526bummer:
1527 dprintk("svc: svc_create_socket error = %d\n", -error);
1528 sock_release(sock);
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001529 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530}
1531
1532/*
Tom Tucker755ccea2007-12-30 21:07:27 -06001533 * Detach the svc_sock from the socket so that no
1534 * more callbacks occur.
1535 */
1536static void svc_sock_detach(struct svc_xprt *xprt)
1537{
1538 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1539 struct sock *sk = svsk->sk_sk;
Eric Dumazeteaefd112011-02-18 03:26:36 +00001540 wait_queue_head_t *wq;
Tom Tucker755ccea2007-12-30 21:07:27 -06001541
1542 dprintk("svc: svc_sock_detach(%p)\n", svsk);
1543
1544 /* put back the old socket callbacks */
1545 sk->sk_state_change = svsk->sk_ostate;
1546 sk->sk_data_ready = svsk->sk_odata;
1547 sk->sk_write_space = svsk->sk_owspace;
Trond Myklebust69b6ba32008-12-23 16:30:11 -05001548
Eric Dumazeteaefd112011-02-18 03:26:36 +00001549 wq = sk_sleep(sk);
1550 if (wq && waitqueue_active(wq))
1551 wake_up_interruptible(wq);
Trond Myklebust69b6ba32008-12-23 16:30:11 -05001552}
1553
1554/*
1555 * Disconnect the socket, and reset the callbacks
1556 */
1557static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1558{
1559 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1560
1561 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
1562
1563 svc_sock_detach(xprt);
1564
1565 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags))
1566 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
Tom Tucker755ccea2007-12-30 21:07:27 -06001567}
1568
1569/*
1570 * Free the svc_sock's socket resources and the svc_sock itself.
1571 */
1572static void svc_sock_free(struct svc_xprt *xprt)
1573{
1574 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1575 dprintk("svc: svc_sock_free(%p)\n", svsk);
1576
Tom Tucker755ccea2007-12-30 21:07:27 -06001577 if (svsk->sk_sock->file)
1578 sockfd_put(svsk->sk_sock);
1579 else
1580 sock_release(svsk->sk_sock);
1581 kfree(svsk);
1582}
Benny Halevy7652e5a2009-04-01 09:23:09 -04001583
Andy Adamson1f11a032011-01-06 02:04:26 +00001584#if defined(CONFIG_NFS_V4_1)
Benny Halevy7652e5a2009-04-01 09:23:09 -04001585/*
Andy Adamson1f11a032011-01-06 02:04:26 +00001586 * Create a back channel svc_xprt which shares the fore channel socket.
Benny Halevy7652e5a2009-04-01 09:23:09 -04001587 */
Andy Adamson1f11a032011-01-06 02:04:26 +00001588static struct svc_xprt *svc_bc_create_socket(struct svc_serv *serv,
1589 int protocol,
1590 struct net *net,
1591 struct sockaddr *sin, int len,
1592 int flags)
Benny Halevy7652e5a2009-04-01 09:23:09 -04001593{
1594 struct svc_sock *svsk;
Andy Adamson1f11a032011-01-06 02:04:26 +00001595 struct svc_xprt *xprt;
Benny Halevy7652e5a2009-04-01 09:23:09 -04001596
Andy Adamson1f11a032011-01-06 02:04:26 +00001597 if (protocol != IPPROTO_TCP) {
1598 printk(KERN_WARNING "svc: only TCP sockets"
1599 " supported on shared back channel\n");
1600 return ERR_PTR(-EINVAL);
1601 }
1602
Benny Halevy7652e5a2009-04-01 09:23:09 -04001603 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1604 if (!svsk)
Andy Adamson1f11a032011-01-06 02:04:26 +00001605 return ERR_PTR(-ENOMEM);
Benny Halevy7652e5a2009-04-01 09:23:09 -04001606
1607 xprt = &svsk->sk_xprt;
Andy Adamson1f11a032011-01-06 02:04:26 +00001608 svc_xprt_init(&svc_tcp_bc_class, xprt, serv);
1609
Andy Adamson4a19de02011-01-06 02:04:35 +00001610 serv->sv_bc_xprt = xprt;
Andy Adamson1f11a032011-01-06 02:04:26 +00001611
Benny Halevy7652e5a2009-04-01 09:23:09 -04001612 return xprt;
1613}
Benny Halevy7652e5a2009-04-01 09:23:09 -04001614
1615/*
Andy Adamson1f11a032011-01-06 02:04:26 +00001616 * Free a back channel svc_sock.
Benny Halevy7652e5a2009-04-01 09:23:09 -04001617 */
Andy Adamson1f11a032011-01-06 02:04:26 +00001618static void svc_bc_sock_free(struct svc_xprt *xprt)
Benny Halevy7652e5a2009-04-01 09:23:09 -04001619{
Andy Adamson778be232011-01-25 15:38:01 +00001620 if (xprt)
Benny Halevy7652e5a2009-04-01 09:23:09 -04001621 kfree(container_of(xprt, struct svc_sock, sk_xprt));
1622}
Andy Adamson1f11a032011-01-06 02:04:26 +00001623#endif /* CONFIG_NFS_V4_1 */