blob: 824d32fb31214b5f433f439c439e7161fba28fb5 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/net/sunrpc/svcsock.c
3 *
4 * These are the RPC server socket internals.
5 *
6 * The server scheduling algorithm does not always distribute the load
7 * evenly when servicing a single client. May need to modify the
Tom Tuckerf6150c32007-12-30 21:07:57 -06008 * svc_xprt_enqueue procedure...
Linus Torvalds1da177e2005-04-16 15:20:36 -07009 *
10 * TCP support is largely untested and may be a little slow. The problem
11 * is that we currently do two separate recvfrom's, one for the 4-byte
12 * record length, and the second for the actual record. This could possibly
13 * be improved by always reading a minimum size of around 100 bytes and
14 * tucking any superfluous bytes away in a temporary store. Still, that
15 * leaves write requests out in the rain. An alternative may be to peek at
16 * the first skb in the queue, and if it matches the next TCP sequence
17 * number, to extract the record marker. Yuck.
18 *
19 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
20 */
21
Ilpo Järvinen172589c2007-08-28 15:50:33 -070022#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include <linux/sched.h>
Paul Gortmaker3a9a2312011-05-27 09:12:25 -040024#include <linux/module.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/errno.h>
26#include <linux/fcntl.h>
27#include <linux/net.h>
28#include <linux/in.h>
29#include <linux/inet.h>
30#include <linux/udp.h>
Andrew Morton91483c42005-08-09 20:20:07 -070031#include <linux/tcp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include <linux/unistd.h>
33#include <linux/slab.h>
34#include <linux/netdevice.h>
35#include <linux/skbuff.h>
NeilBrownb41b66d2006-10-02 02:17:48 -070036#include <linux/file.h>
Nigel Cunningham7dfb7102006-12-06 20:34:23 -080037#include <linux/freezer.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038#include <net/sock.h>
39#include <net/checksum.h>
40#include <net/ip.h>
Chuck Leverb92503b2007-02-12 00:53:36 -080041#include <net/ipv6.h>
Chuck Leverb7872fe2008-04-14 12:27:01 -040042#include <net/tcp.h>
Arnaldo Carvalho de Meloc752f072005-08-09 20:08:28 -070043#include <net/tcp_states.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070044#include <asm/uaccess.h>
45#include <asm/ioctls.h>
46
47#include <linux/sunrpc/types.h>
Chuck Leverad06e4b2007-02-12 00:53:32 -080048#include <linux/sunrpc/clnt.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070049#include <linux/sunrpc/xdr.h>
Chuck Leverc0401ea2008-04-14 12:27:30 -040050#include <linux/sunrpc/msg_prot.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070051#include <linux/sunrpc/svcsock.h>
52#include <linux/sunrpc/stats.h>
Rahul Iyer4cfc7e62009-09-10 17:32:28 +030053#include <linux/sunrpc/xprt.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070054
H Hartley Sweeten177e4f92011-06-20 17:54:51 -070055#include "sunrpc.h"
56
Tom Tucker360d8732007-12-30 21:07:17 -060057#define RPCDBG_FACILITY RPCDBG_SVCXPRT
Linus Torvalds1da177e2005-04-16 15:20:36 -070058
59
60static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
Chuck Lever6b174332007-02-12 00:53:28 -080061 int *errp, int flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -070062static void svc_udp_data_ready(struct sock *, int);
63static int svc_udp_recvfrom(struct svc_rqst *);
64static int svc_udp_sendto(struct svc_rqst *);
Tom Tucker755ccea2007-12-30 21:07:27 -060065static void svc_sock_detach(struct svc_xprt *);
Trond Myklebust69b6ba32008-12-23 16:30:11 -050066static void svc_tcp_sock_detach(struct svc_xprt *);
Tom Tucker755ccea2007-12-30 21:07:27 -060067static void svc_sock_free(struct svc_xprt *);
Linus Torvalds1da177e2005-04-16 15:20:36 -070068
Tom Tuckerb700cbb2007-12-30 21:07:42 -060069static struct svc_xprt *svc_create_socket(struct svc_serv *, int,
Pavel Emelyanov62832c02010-09-29 16:04:18 +040070 struct net *, struct sockaddr *,
71 int, int);
Trond Myklebust9e00abc2011-07-13 19:20:49 -040072#if defined(CONFIG_SUNRPC_BACKCHANNEL)
Andy Adamson1f11a032011-01-06 02:04:26 +000073static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
74 struct net *, struct sockaddr *,
75 int, int);
76static void svc_bc_sock_free(struct svc_xprt *xprt);
Trond Myklebust9e00abc2011-07-13 19:20:49 -040077#endif /* CONFIG_SUNRPC_BACKCHANNEL */
Andy Adamson1f11a032011-01-06 02:04:26 +000078
Peter Zijlstraed075362006-12-06 20:35:24 -080079#ifdef CONFIG_DEBUG_LOCK_ALLOC
80static struct lock_class_key svc_key[2];
81static struct lock_class_key svc_slock_key[2];
82
Tom Tucker0f0257e2007-12-30 21:08:27 -060083static void svc_reclassify_socket(struct socket *sock)
Peter Zijlstraed075362006-12-06 20:35:24 -080084{
85 struct sock *sk = sock->sk;
John Heffner02b3d342007-09-12 10:42:12 +020086 BUG_ON(sock_owned_by_user(sk));
Peter Zijlstraed075362006-12-06 20:35:24 -080087 switch (sk->sk_family) {
88 case AF_INET:
89 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
Tom Tuckerdef13d72007-12-30 21:08:08 -060090 &svc_slock_key[0],
91 "sk_xprt.xpt_lock-AF_INET-NFSD",
92 &svc_key[0]);
Peter Zijlstraed075362006-12-06 20:35:24 -080093 break;
94
95 case AF_INET6:
96 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
Tom Tuckerdef13d72007-12-30 21:08:08 -060097 &svc_slock_key[1],
98 "sk_xprt.xpt_lock-AF_INET6-NFSD",
99 &svc_key[1]);
Peter Zijlstraed075362006-12-06 20:35:24 -0800100 break;
101
102 default:
103 BUG();
104 }
105}
106#else
Tom Tucker0f0257e2007-12-30 21:08:27 -0600107static void svc_reclassify_socket(struct socket *sock)
Peter Zijlstraed075362006-12-06 20:35:24 -0800108{
109}
110#endif
111
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112/*
113 * Release an skbuff after use
114 */
Tom Tucker5148bf42007-12-30 21:07:25 -0600115static void svc_release_skb(struct svc_rqst *rqstp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116{
Tom Tucker5148bf42007-12-30 21:07:25 -0600117 struct sk_buff *skb = rqstp->rq_xprt_ctxt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118
119 if (skb) {
Tom Tucker57b1d3b2007-12-30 21:08:22 -0600120 struct svc_sock *svsk =
121 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
Tom Tucker5148bf42007-12-30 21:07:25 -0600122 rqstp->rq_xprt_ctxt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123
124 dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
Eric Dumazet9d410c72009-10-30 05:03:53 +0000125 skb_free_datagram_locked(svsk->sk_sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127}
128
Chuck Leverb92503b2007-02-12 00:53:36 -0800129union svc_pktinfo_u {
130 struct in_pktinfo pkti;
Chuck Leverb92503b2007-02-12 00:53:36 -0800131 struct in6_pktinfo pkti6;
Chuck Leverb92503b2007-02-12 00:53:36 -0800132};
David S. Millerbc375ea2007-04-12 13:35:59 -0700133#define SVC_PKTINFO_SPACE \
134 CMSG_SPACE(sizeof(union svc_pktinfo_u))
Chuck Leverb92503b2007-02-12 00:53:36 -0800135
136static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
137{
Tom Tucker57b1d3b2007-12-30 21:08:22 -0600138 struct svc_sock *svsk =
139 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
140 switch (svsk->sk_sk->sk_family) {
Chuck Leverb92503b2007-02-12 00:53:36 -0800141 case AF_INET: {
142 struct in_pktinfo *pki = CMSG_DATA(cmh);
143
144 cmh->cmsg_level = SOL_IP;
145 cmh->cmsg_type = IP_PKTINFO;
146 pki->ipi_ifindex = 0;
Mi Jinlong849a1cf2011-08-30 17:18:41 +0800147 pki->ipi_spec_dst.s_addr =
148 svc_daddr_in(rqstp)->sin_addr.s_addr;
Chuck Leverb92503b2007-02-12 00:53:36 -0800149 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
150 }
151 break;
NeilBrown5a05ed72007-03-06 01:42:22 -0800152
Chuck Leverb92503b2007-02-12 00:53:36 -0800153 case AF_INET6: {
154 struct in6_pktinfo *pki = CMSG_DATA(cmh);
Mi Jinlong849a1cf2011-08-30 17:18:41 +0800155 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
Chuck Leverb92503b2007-02-12 00:53:36 -0800156
157 cmh->cmsg_level = SOL_IPV6;
158 cmh->cmsg_type = IPV6_PKTINFO;
Mi Jinlong849a1cf2011-08-30 17:18:41 +0800159 pki->ipi6_ifindex = daddr->sin6_scope_id;
Alexey Dobriyan4e3fd7a2011-11-21 03:39:03 +0000160 pki->ipi6_addr = daddr->sin6_addr;
Chuck Leverb92503b2007-02-12 00:53:36 -0800161 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
162 }
163 break;
Chuck Leverb92503b2007-02-12 00:53:36 -0800164 }
Chuck Leverb92503b2007-02-12 00:53:36 -0800165}
166
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167/*
Rahul Iyer4cfc7e62009-09-10 17:32:28 +0300168 * send routine intended to be shared by the fore- and back-channel
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169 */
Rahul Iyer4cfc7e62009-09-10 17:32:28 +0300170int svc_send_common(struct socket *sock, struct xdr_buf *xdr,
171 struct page *headpage, unsigned long headoffset,
172 struct page *tailpage, unsigned long tailoffset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174 int result;
175 int size;
176 struct page **ppage = xdr->pages;
177 size_t base = xdr->page_base;
178 unsigned int pglen = xdr->page_len;
179 unsigned int flags = MSG_MORE;
Rahul Iyer4cfc7e62009-09-10 17:32:28 +0300180 int slen;
181 int len = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182
183 slen = xdr->len;
184
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 /* send head */
186 if (slen == xdr->head[0].iov_len)
187 flags = 0;
Rahul Iyer4cfc7e62009-09-10 17:32:28 +0300188 len = kernel_sendpage(sock, headpage, headoffset,
NeilBrown44524352006-10-04 02:15:46 -0700189 xdr->head[0].iov_len, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 if (len != xdr->head[0].iov_len)
191 goto out;
192 slen -= xdr->head[0].iov_len;
193 if (slen == 0)
194 goto out;
195
196 /* send page data */
197 size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen;
198 while (pglen > 0) {
199 if (slen == size)
200 flags = 0;
Sridhar Samudralae6242e92006-08-07 20:58:01 -0700201 result = kernel_sendpage(sock, *ppage, base, size, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 if (result > 0)
203 len += result;
204 if (result != size)
205 goto out;
206 slen -= size;
207 pglen -= size;
208 size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen;
209 base = 0;
210 ppage++;
211 }
Rahul Iyer4cfc7e62009-09-10 17:32:28 +0300212
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 /* send tail */
214 if (xdr->tail[0].iov_len) {
Rahul Iyer4cfc7e62009-09-10 17:32:28 +0300215 result = kernel_sendpage(sock, tailpage, tailoffset,
216 xdr->tail[0].iov_len, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 if (result > 0)
218 len += result;
219 }
Rahul Iyer4cfc7e62009-09-10 17:32:28 +0300220
221out:
222 return len;
223}
224
225
226/*
227 * Generic sendto routine
228 */
229static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr)
230{
231 struct svc_sock *svsk =
232 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
233 struct socket *sock = svsk->sk_sock;
234 union {
235 struct cmsghdr hdr;
236 long all[SVC_PKTINFO_SPACE / sizeof(long)];
237 } buffer;
238 struct cmsghdr *cmh = &buffer.hdr;
239 int len = 0;
240 unsigned long tailoff;
241 unsigned long headoff;
242 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
243
244 if (rqstp->rq_prot == IPPROTO_UDP) {
245 struct msghdr msg = {
246 .msg_name = &rqstp->rq_addr,
247 .msg_namelen = rqstp->rq_addrlen,
248 .msg_control = cmh,
249 .msg_controllen = sizeof(buffer),
250 .msg_flags = MSG_MORE,
251 };
252
253 svc_set_cmsg_data(rqstp, cmh);
254
255 if (sock_sendmsg(sock, &msg, 0) < 0)
256 goto out;
257 }
258
259 tailoff = ((unsigned long)xdr->tail[0].iov_base) & (PAGE_SIZE-1);
260 headoff = 0;
261 len = svc_send_common(sock, xdr, rqstp->rq_respages[0], headoff,
262 rqstp->rq_respages[0], tailoff);
263
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264out:
Chuck Leverad06e4b2007-02-12 00:53:32 -0800265 dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n",
Tom Tucker57b1d3b2007-12-30 21:08:22 -0600266 svsk, xdr->head[0].iov_base, xdr->head[0].iov_len,
Chuck Leverad06e4b2007-02-12 00:53:32 -0800267 xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268
269 return len;
270}
271
272/*
NeilBrown80212d52006-10-02 02:17:47 -0700273 * Report socket names for nfsdfs
274 */
Chuck Levere7942b92009-04-23 19:32:48 -0400275static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining)
NeilBrown80212d52006-10-02 02:17:47 -0700276{
Chuck Lever017cb472009-04-23 19:33:03 -0400277 const struct sock *sk = svsk->sk_sk;
278 const char *proto_name = sk->sk_protocol == IPPROTO_UDP ?
279 "udp" : "tcp";
NeilBrown80212d52006-10-02 02:17:47 -0700280 int len;
281
Chuck Lever017cb472009-04-23 19:33:03 -0400282 switch (sk->sk_family) {
Chuck Levere7942b92009-04-23 19:32:48 -0400283 case PF_INET:
284 len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n",
Chuck Lever017cb472009-04-23 19:33:03 -0400285 proto_name,
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000286 &inet_sk(sk)->inet_rcv_saddr,
287 inet_sk(sk)->inet_num);
NeilBrown80212d52006-10-02 02:17:47 -0700288 break;
Chuck Lever58de2f82009-04-23 19:32:55 -0400289 case PF_INET6:
290 len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n",
Chuck Lever017cb472009-04-23 19:33:03 -0400291 proto_name,
292 &inet6_sk(sk)->rcv_saddr,
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000293 inet_sk(sk)->inet_num);
NeilBrown80212d52006-10-02 02:17:47 -0700294 break;
295 default:
Chuck Levere7942b92009-04-23 19:32:48 -0400296 len = snprintf(buf, remaining, "*unknown-%d*\n",
Chuck Lever017cb472009-04-23 19:33:03 -0400297 sk->sk_family);
NeilBrown80212d52006-10-02 02:17:47 -0700298 }
Chuck Levere7942b92009-04-23 19:32:48 -0400299
300 if (len >= remaining) {
301 *buf = '\0';
302 return -ENAMETOOLONG;
NeilBrown80212d52006-10-02 02:17:47 -0700303 }
304 return len;
305}
306
Chuck Lever8435d342009-04-23 19:32:40 -0400307/**
308 * svc_sock_names - construct a list of listener names in a string
309 * @serv: pointer to RPC service
310 * @buf: pointer to a buffer to fill in with socket names
311 * @buflen: size of the buffer to be filled
312 * @toclose: pointer to '\0'-terminated C string containing the name
313 * of a listener to be closed
314 *
315 * Fills in @buf with a '\n'-separated list of names of listener
316 * sockets. If @toclose is not NULL, the socket named by @toclose
317 * is closed, and is not included in the output list.
318 *
319 * Returns positive length of the socket name string, or a negative
320 * errno value on error.
321 */
322int svc_sock_names(struct svc_serv *serv, char *buf, const size_t buflen,
323 const char *toclose)
NeilBrown80212d52006-10-02 02:17:47 -0700324{
NeilBrownb41b66d2006-10-02 02:17:48 -0700325 struct svc_sock *svsk, *closesk = NULL;
NeilBrown80212d52006-10-02 02:17:47 -0700326 int len = 0;
327
328 if (!serv)
329 return 0;
Chuck Levere7942b92009-04-23 19:32:48 -0400330
NeilBrownaaf68cf2007-02-08 14:20:30 -0800331 spin_lock_bh(&serv->sv_lock);
Tom Tucker7a182082007-12-30 21:07:53 -0600332 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list) {
Chuck Levere7942b92009-04-23 19:32:48 -0400333 int onelen = svc_one_sock_name(svsk, buf + len, buflen - len);
334 if (onelen < 0) {
335 len = onelen;
336 break;
337 }
NeilBrown39423022010-11-15 11:27:01 +1100338 if (toclose && strcmp(toclose, buf + len) == 0) {
NeilBrownb41b66d2006-10-02 02:17:48 -0700339 closesk = svsk;
NeilBrown39423022010-11-15 11:27:01 +1100340 svc_xprt_get(&closesk->sk_xprt);
341 } else
NeilBrownb41b66d2006-10-02 02:17:48 -0700342 len += onelen;
NeilBrown80212d52006-10-02 02:17:47 -0700343 }
NeilBrownaaf68cf2007-02-08 14:20:30 -0800344 spin_unlock_bh(&serv->sv_lock);
Chuck Levere7942b92009-04-23 19:32:48 -0400345
NeilBrown39423022010-11-15 11:27:01 +1100346 if (closesk) {
NeilBrown5680c442006-10-04 02:15:45 -0700347 /* Should unregister with portmap, but you cannot
348 * unregister just one protocol...
349 */
Tom Tucker7a182082007-12-30 21:07:53 -0600350 svc_close_xprt(&closesk->sk_xprt);
NeilBrown39423022010-11-15 11:27:01 +1100351 svc_xprt_put(&closesk->sk_xprt);
352 } else if (toclose)
NeilBrown37a03472006-10-04 02:15:44 -0700353 return -ENOENT;
NeilBrown80212d52006-10-02 02:17:47 -0700354 return len;
355}
Trond Myklebust24c37672008-12-23 16:30:12 -0500356EXPORT_SYMBOL_GPL(svc_sock_names);
NeilBrown80212d52006-10-02 02:17:47 -0700357
358/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 * Check input queue length
360 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600361static int svc_recv_available(struct svc_sock *svsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363 struct socket *sock = svsk->sk_sock;
364 int avail, err;
365
Sridhar Samudralae6242e92006-08-07 20:58:01 -0700366 err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367
368 return (err >= 0)? avail : err;
369}
370
371/*
372 * Generic recvfrom routine.
373 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600374static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr,
375 int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376{
Tom Tucker57b1d3b2007-12-30 21:08:22 -0600377 struct svc_sock *svsk =
378 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
Chuck Lever1ba95102007-02-12 00:53:31 -0800379 struct msghdr msg = {
380 .msg_flags = MSG_DONTWAIT,
381 };
382 int len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383
Tom Tucker260c1d12007-12-30 21:08:29 -0600384 rqstp->rq_xprt_hlen = 0;
385
Chuck Lever1ba95102007-02-12 00:53:31 -0800386 len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen,
387 msg.msg_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
Chuck Lever1ba95102007-02-12 00:53:31 -0800390 svsk, iov[0].iov_base, iov[0].iov_len, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 return len;
392}
393
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800394static int svc_partial_recvfrom(struct svc_rqst *rqstp,
395 struct kvec *iov, int nr,
396 int buflen, unsigned int base)
397{
398 size_t save_iovlen;
Trond Myklebust09acfea2012-03-11 15:22:54 -0400399 void *save_iovbase;
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800400 unsigned int i;
401 int ret;
402
403 if (base == 0)
404 return svc_recvfrom(rqstp, iov, nr, buflen);
405
406 for (i = 0; i < nr; i++) {
407 if (iov[i].iov_len > base)
408 break;
409 base -= iov[i].iov_len;
410 }
411 save_iovlen = iov[i].iov_len;
412 save_iovbase = iov[i].iov_base;
413 iov[i].iov_len -= base;
414 iov[i].iov_base += base;
415 ret = svc_recvfrom(rqstp, &iov[i], nr - i, buflen);
416 iov[i].iov_len = save_iovlen;
417 iov[i].iov_base = save_iovbase;
418 return ret;
419}
420
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421/*
422 * Set socket snd and rcv buffer lengths
423 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600424static void svc_sock_setbufsize(struct socket *sock, unsigned int snd,
425 unsigned int rcv)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426{
427#if 0
428 mm_segment_t oldfs;
429 oldfs = get_fs(); set_fs(KERNEL_DS);
430 sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
431 (char*)&snd, sizeof(snd));
432 sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
433 (char*)&rcv, sizeof(rcv));
434#else
435 /* sock_setsockopt limits use to sysctl_?mem_max,
436 * which isn't acceptable. Until that is made conditional
437 * on not having CAP_SYS_RESOURCE or similar, we go direct...
438 * DaveM said I could!
439 */
440 lock_sock(sock->sk);
441 sock->sk->sk_sndbuf = snd * 2;
442 sock->sk->sk_rcvbuf = rcv * 2;
Trond Myklebust47fcb032009-05-18 17:47:56 -0400443 sock->sk->sk_write_space(sock->sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 release_sock(sock->sk);
445#endif
446}
447/*
448 * INET callback when data has been received on the socket.
449 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600450static void svc_udp_data_ready(struct sock *sk, int count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451{
Neil Brown939bb7e2005-09-13 01:25:39 -0700452 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
Eric Dumazeteaefd112011-02-18 03:26:36 +0000453 wait_queue_head_t *wq = sk_sleep(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454
Neil Brown939bb7e2005-09-13 01:25:39 -0700455 if (svsk) {
456 dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n",
Tom Tucker02fc6c32007-12-30 21:07:48 -0600457 svsk, sk, count,
458 test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
459 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
Tom Tuckerf6150c32007-12-30 21:07:57 -0600460 svc_xprt_enqueue(&svsk->sk_xprt);
Neil Brown939bb7e2005-09-13 01:25:39 -0700461 }
Eric Dumazeteaefd112011-02-18 03:26:36 +0000462 if (wq && waitqueue_active(wq))
463 wake_up_interruptible(wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464}
465
466/*
467 * INET callback when space is newly available on the socket.
468 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600469static void svc_write_space(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470{
471 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
Eric Dumazeteaefd112011-02-18 03:26:36 +0000472 wait_queue_head_t *wq = sk_sleep(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473
474 if (svsk) {
475 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
Tom Tucker02fc6c32007-12-30 21:07:48 -0600476 svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
Tom Tuckerf6150c32007-12-30 21:07:57 -0600477 svc_xprt_enqueue(&svsk->sk_xprt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 }
479
Eric Dumazeteaefd112011-02-18 03:26:36 +0000480 if (wq && waitqueue_active(wq)) {
Neil Brown939bb7e2005-09-13 01:25:39 -0700481 dprintk("RPC svc_write_space: someone sleeping on %p\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 svsk);
Eric Dumazeteaefd112011-02-18 03:26:36 +0000483 wake_up_interruptible(wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 }
485}
486
Trond Myklebust47fcb032009-05-18 17:47:56 -0400487static void svc_tcp_write_space(struct sock *sk)
488{
489 struct socket *sock = sk->sk_socket;
490
491 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && sock)
492 clear_bit(SOCK_NOSPACE, &sock->flags);
493 svc_write_space(sk);
494}
495
Tom Tucker9dbc2402007-12-30 21:08:12 -0600496/*
Chuck Lever7702ce42009-07-13 10:54:26 -0400497 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
498 */
499static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
500 struct cmsghdr *cmh)
501{
502 struct in_pktinfo *pki = CMSG_DATA(cmh);
Mi Jinlong849a1cf2011-08-30 17:18:41 +0800503 struct sockaddr_in *daddr = svc_daddr_in(rqstp);
504
Chuck Lever7702ce42009-07-13 10:54:26 -0400505 if (cmh->cmsg_type != IP_PKTINFO)
506 return 0;
Mi Jinlong849a1cf2011-08-30 17:18:41 +0800507
508 daddr->sin_family = AF_INET;
509 daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr;
Chuck Lever7702ce42009-07-13 10:54:26 -0400510 return 1;
511}
512
513/*
514 * See net/ipv6/datagram.c : datagram_recv_ctl
515 */
516static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
517 struct cmsghdr *cmh)
518{
519 struct in6_pktinfo *pki = CMSG_DATA(cmh);
Mi Jinlong849a1cf2011-08-30 17:18:41 +0800520 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
521
Chuck Lever7702ce42009-07-13 10:54:26 -0400522 if (cmh->cmsg_type != IPV6_PKTINFO)
523 return 0;
Mi Jinlong849a1cf2011-08-30 17:18:41 +0800524
525 daddr->sin6_family = AF_INET6;
Alexey Dobriyan4e3fd7a2011-11-21 03:39:03 +0000526 daddr->sin6_addr = pki->ipi6_addr;
Mi Jinlong849a1cf2011-08-30 17:18:41 +0800527 daddr->sin6_scope_id = pki->ipi6_ifindex;
Chuck Lever7702ce42009-07-13 10:54:26 -0400528 return 1;
529}
530
531/*
Tom Tucker9dbc2402007-12-30 21:08:12 -0600532 * Copy the UDP datagram's destination address to the rqstp structure.
533 * The 'destination' address in this case is the address to which the
534 * peer sent the datagram, i.e. our local address. For multihomed
535 * hosts, this can change from msg to msg. Note that only the IP
536 * address changes, the port number should remain the same.
537 */
Chuck Lever7702ce42009-07-13 10:54:26 -0400538static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
539 struct cmsghdr *cmh)
Chuck Lever95756482007-02-12 00:53:38 -0800540{
Chuck Lever7702ce42009-07-13 10:54:26 -0400541 switch (cmh->cmsg_level) {
542 case SOL_IP:
543 return svc_udp_get_dest_address4(rqstp, cmh);
544 case SOL_IPV6:
545 return svc_udp_get_dest_address6(rqstp, cmh);
Chuck Lever95756482007-02-12 00:53:38 -0800546 }
Chuck Lever7702ce42009-07-13 10:54:26 -0400547
548 return 0;
Chuck Lever95756482007-02-12 00:53:38 -0800549}
550
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551/*
552 * Receive a datagram from a UDP socket.
553 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600554static int svc_udp_recvfrom(struct svc_rqst *rqstp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555{
Tom Tucker57b1d3b2007-12-30 21:08:22 -0600556 struct svc_sock *svsk =
557 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600558 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 struct sk_buff *skb;
David S. Millerbc375ea2007-04-12 13:35:59 -0700560 union {
561 struct cmsghdr hdr;
562 long all[SVC_PKTINFO_SPACE / sizeof(long)];
563 } buffer;
564 struct cmsghdr *cmh = &buffer.hdr;
NeilBrown7a37f572007-03-06 01:42:21 -0800565 struct msghdr msg = {
566 .msg_name = svc_addr(rqstp),
567 .msg_control = cmh,
568 .msg_controllen = sizeof(buffer),
569 .msg_flags = MSG_DONTWAIT,
570 };
Chuck Leverabc5c442009-04-23 19:31:25 -0400571 size_t len;
572 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573
Tom Tucker02fc6c32007-12-30 21:07:48 -0600574 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 /* udp sockets need large rcvbuf as all pending
576 * requests are still in that buffer. sndbuf must
577 * also be large enough that there is enough space
Greg Banks3262c812006-10-02 02:17:58 -0700578 * for one reply per thread. We count all threads
579 * rather than threads in a particular pool, which
580 * provides an upper bound on the number of threads
581 * which will access the socket.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 */
583 svc_sock_setbufsize(svsk->sk_sock,
NeilBrownc6b0a9f2006-10-06 00:44:05 -0700584 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
585 (serv->sv_nrthreads+3) * serv->sv_max_mesg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586
Tom Tucker02fc6c32007-12-30 21:07:48 -0600587 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
NeilBrown05ed6902007-05-09 02:34:55 -0700588 skb = NULL;
589 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
590 0, 0, MSG_PEEK | MSG_DONTWAIT);
591 if (err >= 0)
592 skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err);
593
594 if (skb == NULL) {
595 if (err != -EAGAIN) {
596 /* possibly an icmp error */
597 dprintk("svc: recvfrom returned error %d\n", -err);
Tom Tucker02fc6c32007-12-30 21:07:48 -0600598 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 }
NeilBrown05ed6902007-05-09 02:34:55 -0700600 return -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 }
Tom Tucker9dbc2402007-12-30 21:08:12 -0600602 len = svc_addr_len(svc_addr(rqstp));
Chuck Leverabc5c442009-04-23 19:31:25 -0400603 if (len == 0)
604 return -EAFNOSUPPORT;
Tom Tucker9dbc2402007-12-30 21:08:12 -0600605 rqstp->rq_addrlen = len;
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -0700606 if (skb->tstamp.tv64 == 0) {
607 skb->tstamp = ktime_get_real();
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -0800608 /* Don't enable netstamp, sunrpc doesn't
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 need that much accuracy */
610 }
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -0700611 svsk->sk_sk->sk_stamp = skb->tstamp;
Tom Tucker02fc6c32007-12-30 21:07:48 -0600612 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 len = skb->len - sizeof(struct udphdr);
615 rqstp->rq_arg.len = len;
616
Chuck Lever95756482007-02-12 00:53:38 -0800617 rqstp->rq_prot = IPPROTO_UDP;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618
Chuck Lever7702ce42009-07-13 10:54:26 -0400619 if (!svc_udp_get_dest_address(rqstp, cmh)) {
NeilBrown7a37f572007-03-06 01:42:21 -0800620 if (net_ratelimit())
Chuck Lever7702ce42009-07-13 10:54:26 -0400621 printk(KERN_WARNING
622 "svc: received unknown control message %d/%d; "
623 "dropping RPC reply datagram\n",
624 cmh->cmsg_level, cmh->cmsg_type);
Eric Dumazet9d410c72009-10-30 05:03:53 +0000625 skb_free_datagram_locked(svsk->sk_sk, skb);
NeilBrown7a37f572007-03-06 01:42:21 -0800626 return 0;
627 }
Mi Jinlong849a1cf2011-08-30 17:18:41 +0800628 rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629
630 if (skb_is_nonlinear(skb)) {
631 /* we have to copy */
632 local_bh_disable();
633 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
634 local_bh_enable();
635 /* checksum error */
Eric Dumazet9d410c72009-10-30 05:03:53 +0000636 skb_free_datagram_locked(svsk->sk_sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 return 0;
638 }
639 local_bh_enable();
Eric Dumazet9d410c72009-10-30 05:03:53 +0000640 skb_free_datagram_locked(svsk->sk_sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 } else {
642 /* we can use it in-place */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600643 rqstp->rq_arg.head[0].iov_base = skb->data +
644 sizeof(struct udphdr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 rqstp->rq_arg.head[0].iov_len = len;
Herbert Xufb286bb2005-11-10 13:01:24 -0800646 if (skb_checksum_complete(skb)) {
Eric Dumazet9d410c72009-10-30 05:03:53 +0000647 skb_free_datagram_locked(svsk->sk_sk, skb);
Herbert Xufb286bb2005-11-10 13:01:24 -0800648 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 }
Tom Tucker5148bf42007-12-30 21:07:25 -0600650 rqstp->rq_xprt_ctxt = skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 }
652
653 rqstp->rq_arg.page_base = 0;
654 if (len <= rqstp->rq_arg.head[0].iov_len) {
655 rqstp->rq_arg.head[0].iov_len = len;
656 rqstp->rq_arg.page_len = 0;
NeilBrown44524352006-10-04 02:15:46 -0700657 rqstp->rq_respages = rqstp->rq_pages+1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 } else {
659 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
NeilBrown44524352006-10-04 02:15:46 -0700660 rqstp->rq_respages = rqstp->rq_pages + 1 +
Ilpo Järvinen172589c2007-08-28 15:50:33 -0700661 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 }
663
664 if (serv->sv_stats)
665 serv->sv_stats->netudpcnt++;
666
667 return len;
668}
669
670static int
671svc_udp_sendto(struct svc_rqst *rqstp)
672{
673 int error;
674
675 error = svc_sendto(rqstp, &rqstp->rq_res);
676 if (error == -ECONNREFUSED)
677 /* ICMP error on earlier request. */
678 error = svc_sendto(rqstp, &rqstp->rq_res);
679
680 return error;
681}
682
Tom Tuckere831fe62007-12-30 21:07:29 -0600683static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
684{
685}
686
Tom Tucker323bee32007-12-30 21:07:31 -0600687static int svc_udp_has_wspace(struct svc_xprt *xprt)
688{
689 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600690 struct svc_serv *serv = xprt->xpt_server;
Tom Tucker323bee32007-12-30 21:07:31 -0600691 unsigned long required;
692
693 /*
694 * Set the SOCK_NOSPACE flag before checking the available
695 * sock space.
696 */
697 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
Tom Tucker7a90e8c2007-12-30 21:07:55 -0600698 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
Tom Tucker323bee32007-12-30 21:07:31 -0600699 if (required*2 > sock_wspace(svsk->sk_sk))
700 return 0;
701 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
702 return 1;
703}
704
Tom Tucker38a417c2007-12-30 21:07:36 -0600705static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
706{
707 BUG();
708 return NULL;
709}
710
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600711static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
Pavel Emelyanov62832c02010-09-29 16:04:18 +0400712 struct net *net,
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600713 struct sockaddr *sa, int salen,
714 int flags)
715{
Pavel Emelyanov62832c02010-09-29 16:04:18 +0400716 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600717}
718
Tom Tucker360d8732007-12-30 21:07:17 -0600719static struct svc_xprt_ops svc_udp_ops = {
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600720 .xpo_create = svc_udp_create,
Tom Tucker5d137992007-12-30 21:07:23 -0600721 .xpo_recvfrom = svc_udp_recvfrom,
722 .xpo_sendto = svc_udp_sendto,
Tom Tucker5148bf42007-12-30 21:07:25 -0600723 .xpo_release_rqst = svc_release_skb,
Tom Tucker755ccea2007-12-30 21:07:27 -0600724 .xpo_detach = svc_sock_detach,
725 .xpo_free = svc_sock_free,
Tom Tuckere831fe62007-12-30 21:07:29 -0600726 .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
Tom Tucker323bee32007-12-30 21:07:31 -0600727 .xpo_has_wspace = svc_udp_has_wspace,
Tom Tucker38a417c2007-12-30 21:07:36 -0600728 .xpo_accept = svc_udp_accept,
Tom Tucker360d8732007-12-30 21:07:17 -0600729};
730
731static struct svc_xprt_class svc_udp_class = {
732 .xcl_name = "udp",
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600733 .xcl_owner = THIS_MODULE,
Tom Tucker360d8732007-12-30 21:07:17 -0600734 .xcl_ops = &svc_udp_ops,
Tom Tucker49023152007-12-30 21:07:21 -0600735 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
Tom Tucker360d8732007-12-30 21:07:17 -0600736};
737
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600738static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739{
Chuck Lever7702ce42009-07-13 10:54:26 -0400740 int err, level, optname, one = 1;
NeilBrown7a37f572007-03-06 01:42:21 -0800741
Stanislav Kinsburskybd4620d2011-12-06 14:19:10 +0300742 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class,
743 &svsk->sk_xprt, serv);
Tom Tuckerdef13d72007-12-30 21:08:08 -0600744 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 svsk->sk_sk->sk_data_ready = svc_udp_data_ready;
746 svsk->sk_sk->sk_write_space = svc_write_space;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747
748 /* initialise setting must have enough space to
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -0800749 * receive and respond to one request.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 * svc_udp_recvfrom will re-adjust if necessary
751 */
752 svc_sock_setbufsize(svsk->sk_sock,
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600753 3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
754 3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755
Tom Tucker0f0257e2007-12-30 21:08:27 -0600756 /* data might have come in before data_ready set up */
757 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
Tom Tucker02fc6c32007-12-30 21:07:48 -0600758 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
NeilBrown7a37f572007-03-06 01:42:21 -0800759
NeilBrown7a37f572007-03-06 01:42:21 -0800760 /* make sure we get destination address info */
Chuck Lever7702ce42009-07-13 10:54:26 -0400761 switch (svsk->sk_sk->sk_family) {
762 case AF_INET:
763 level = SOL_IP;
764 optname = IP_PKTINFO;
765 break;
766 case AF_INET6:
767 level = SOL_IPV6;
768 optname = IPV6_RECVPKTINFO;
769 break;
770 default:
771 BUG();
772 }
773 err = kernel_setsockopt(svsk->sk_sock, level, optname,
774 (char *)&one, sizeof(one));
775 dprintk("svc: kernel_setsockopt returned %d\n", err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776}
777
778/*
779 * A data_ready event on a listening socket means there's a connection
780 * pending. Do not use state_change as a substitute for it.
781 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600782static void svc_tcp_listen_data_ready(struct sock *sk, int count_unused)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783{
Neil Brown939bb7e2005-09-13 01:25:39 -0700784 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
Eric Dumazeteaefd112011-02-18 03:26:36 +0000785 wait_queue_head_t *wq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786
787 dprintk("svc: socket %p TCP (listen) state change %d\n",
Neil Brown939bb7e2005-09-13 01:25:39 -0700788 sk, sk->sk_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789
Neil Brown939bb7e2005-09-13 01:25:39 -0700790 /*
791 * This callback may called twice when a new connection
792 * is established as a child socket inherits everything
793 * from a parent LISTEN socket.
794 * 1) data_ready method of the parent socket will be called
795 * when one of child sockets become ESTABLISHED.
796 * 2) data_ready method of the child socket may be called
797 * when it receives data before the socket is accepted.
798 * In case of 2, we should ignore it silently.
799 */
800 if (sk->sk_state == TCP_LISTEN) {
801 if (svsk) {
Tom Tucker02fc6c32007-12-30 21:07:48 -0600802 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
Tom Tuckerf6150c32007-12-30 21:07:57 -0600803 svc_xprt_enqueue(&svsk->sk_xprt);
Neil Brown939bb7e2005-09-13 01:25:39 -0700804 } else
805 printk("svc: socket %p: no user data\n", sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 }
Neil Brown939bb7e2005-09-13 01:25:39 -0700807
Eric Dumazeteaefd112011-02-18 03:26:36 +0000808 wq = sk_sleep(sk);
809 if (wq && waitqueue_active(wq))
810 wake_up_interruptible_all(wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811}
812
813/*
814 * A state change on a connected socket means it's dying or dead.
815 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600816static void svc_tcp_state_change(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817{
Neil Brown939bb7e2005-09-13 01:25:39 -0700818 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
Eric Dumazeteaefd112011-02-18 03:26:36 +0000819 wait_queue_head_t *wq = sk_sleep(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820
821 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
Neil Brown939bb7e2005-09-13 01:25:39 -0700822 sk, sk->sk_state, sk->sk_user_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823
Neil Brown939bb7e2005-09-13 01:25:39 -0700824 if (!svsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 printk("svc: socket %p: no user data\n", sk);
Neil Brown939bb7e2005-09-13 01:25:39 -0700826 else {
Tom Tucker02fc6c32007-12-30 21:07:48 -0600827 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
Tom Tuckerf6150c32007-12-30 21:07:57 -0600828 svc_xprt_enqueue(&svsk->sk_xprt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 }
Eric Dumazeteaefd112011-02-18 03:26:36 +0000830 if (wq && waitqueue_active(wq))
831 wake_up_interruptible_all(wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832}
833
Tom Tucker0f0257e2007-12-30 21:08:27 -0600834static void svc_tcp_data_ready(struct sock *sk, int count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835{
Neil Brown939bb7e2005-09-13 01:25:39 -0700836 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
Eric Dumazeteaefd112011-02-18 03:26:36 +0000837 wait_queue_head_t *wq = sk_sleep(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838
839 dprintk("svc: socket %p TCP data ready (svsk %p)\n",
Neil Brown939bb7e2005-09-13 01:25:39 -0700840 sk, sk->sk_user_data);
841 if (svsk) {
Tom Tucker02fc6c32007-12-30 21:07:48 -0600842 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
Tom Tuckerf6150c32007-12-30 21:07:57 -0600843 svc_xprt_enqueue(&svsk->sk_xprt);
Neil Brown939bb7e2005-09-13 01:25:39 -0700844 }
Eric Dumazeteaefd112011-02-18 03:26:36 +0000845 if (wq && waitqueue_active(wq))
846 wake_up_interruptible(wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847}
848
849/*
850 * Accept a TCP connection
851 */
Tom Tucker38a417c2007-12-30 21:07:36 -0600852static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853{
Tom Tucker38a417c2007-12-30 21:07:36 -0600854 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
akpm@linux-foundation.orgcdd88b92007-02-12 00:53:38 -0800855 struct sockaddr_storage addr;
856 struct sockaddr *sin = (struct sockaddr *) &addr;
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600857 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 struct socket *sock = svsk->sk_sock;
859 struct socket *newsock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 struct svc_sock *newsvsk;
861 int err, slen;
Pavel Emelyanov5216a8e2008-02-21 10:57:45 +0300862 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863
864 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
865 if (!sock)
Tom Tucker38a417c2007-12-30 21:07:36 -0600866 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867
Tom Tucker02fc6c32007-12-30 21:07:48 -0600868 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
Sridhar Samudralae6242e92006-08-07 20:58:01 -0700869 err = kernel_accept(sock, &newsock, O_NONBLOCK);
870 if (err < 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 if (err == -ENOMEM)
872 printk(KERN_WARNING "%s: no more sockets!\n",
873 serv->sv_name);
Sridhar Samudralae6242e92006-08-07 20:58:01 -0700874 else if (err != -EAGAIN && net_ratelimit())
875 printk(KERN_WARNING "%s: accept failed (err %d)!\n",
876 serv->sv_name, -err);
Tom Tucker38a417c2007-12-30 21:07:36 -0600877 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 }
Tom Tucker02fc6c32007-12-30 21:07:48 -0600879 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880
akpm@linux-foundation.orgcdd88b92007-02-12 00:53:38 -0800881 err = kernel_getpeername(newsock, sin, &slen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 if (err < 0) {
883 if (net_ratelimit())
884 printk(KERN_WARNING "%s: peername failed (err %d)!\n",
885 serv->sv_name, -err);
886 goto failed; /* aborted connection or whatever */
887 }
888
889 /* Ideally, we would want to reject connections from unauthorized
Chuck Leverad06e4b2007-02-12 00:53:32 -0800890 * hosts here, but when we get encryption, the IP of the host won't
891 * tell us anything. For now just warn about unpriv connections.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892 */
akpm@linux-foundation.orgcdd88b92007-02-12 00:53:38 -0800893 if (!svc_port_is_privileged(sin)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894 dprintk(KERN_WARNING
Chuck Leverad06e4b2007-02-12 00:53:32 -0800895 "%s: connect from unprivileged port: %s\n",
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -0800896 serv->sv_name,
akpm@linux-foundation.orgcdd88b92007-02-12 00:53:38 -0800897 __svc_print_addr(sin, buf, sizeof(buf)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898 }
Chuck Leverad06e4b2007-02-12 00:53:32 -0800899 dprintk("%s: connect from %s\n", serv->sv_name,
akpm@linux-foundation.orgcdd88b92007-02-12 00:53:38 -0800900 __svc_print_addr(sin, buf, sizeof(buf)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901
902 /* make sure that a write doesn't block forever when
903 * low on memory
904 */
905 newsock->sk->sk_sndtimeo = HZ*30;
906
Chuck Lever6b174332007-02-12 00:53:28 -0800907 if (!(newsvsk = svc_setup_socket(serv, newsock, &err,
908 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY))))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 goto failed;
Tom Tucker9dbc2402007-12-30 21:08:12 -0600910 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
Frank van Maarseveena9747692007-07-09 22:21:39 +0200911 err = kernel_getsockname(newsock, sin, &slen);
912 if (unlikely(err < 0)) {
913 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
914 slen = offsetof(struct sockaddr, sa_data);
915 }
Tom Tucker9dbc2402007-12-30 21:08:12 -0600916 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
Chuck Lever067d7812007-02-12 00:53:30 -0800917
Tom Tuckerf9f3cc42007-12-30 21:07:40 -0600918 if (serv->sv_stats)
919 serv->sv_stats->nettcpconn++;
920
921 return &newsvsk->sk_xprt;
922
923failed:
924 sock_release(newsock);
925 return NULL;
926}
927
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800928static unsigned int svc_tcp_restore_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
929{
930 unsigned int i, len, npages;
931
932 if (svsk->sk_tcplen <= sizeof(rpc_fraghdr))
933 return 0;
934 len = svsk->sk_tcplen - sizeof(rpc_fraghdr);
935 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
936 for (i = 0; i < npages; i++) {
937 if (rqstp->rq_pages[i] != NULL)
938 put_page(rqstp->rq_pages[i]);
939 BUG_ON(svsk->sk_pages[i] == NULL);
940 rqstp->rq_pages[i] = svsk->sk_pages[i];
941 svsk->sk_pages[i] = NULL;
942 }
943 rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
944 return len;
945}
946
947static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
948{
949 unsigned int i, len, npages;
950
951 if (svsk->sk_tcplen <= sizeof(rpc_fraghdr))
952 return;
953 len = svsk->sk_tcplen - sizeof(rpc_fraghdr);
954 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
955 for (i = 0; i < npages; i++) {
956 svsk->sk_pages[i] = rqstp->rq_pages[i];
957 rqstp->rq_pages[i] = NULL;
958 }
959}
960
961static void svc_tcp_clear_pages(struct svc_sock *svsk)
962{
963 unsigned int i, len, npages;
964
965 if (svsk->sk_tcplen <= sizeof(rpc_fraghdr))
966 goto out;
967 len = svsk->sk_tcplen - sizeof(rpc_fraghdr);
968 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
969 for (i = 0; i < npages; i++) {
970 BUG_ON(svsk->sk_pages[i] == NULL);
971 put_page(svsk->sk_pages[i]);
972 svsk->sk_pages[i] = NULL;
973 }
974out:
975 svsk->sk_tcplen = 0;
976}
977
Tom Tuckerf9f3cc42007-12-30 21:07:40 -0600978/*
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +0300979 * Receive data.
980 * If we haven't gotten the record length yet, get the next four bytes.
981 * Otherwise try to gobble up as much as possible up to the complete
982 * record length.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 */
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +0300984static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985{
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600986 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
Trond Myklebust5ee78d42009-05-18 17:47:56 -0400987 unsigned int want;
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +0300988 int len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989
Tom Tucker02fc6c32007-12-30 21:07:48 -0600990 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991
Chuck Leverc0401ea2008-04-14 12:27:30 -0400992 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 struct kvec iov;
994
Trond Myklebust5ee78d42009-05-18 17:47:56 -0400995 want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
997 iov.iov_len = want;
998 if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
999 goto error;
1000 svsk->sk_tcplen += len;
1001
1002 if (len < want) {
Chuck Leverc0401ea2008-04-14 12:27:30 -04001003 dprintk("svc: short recvfrom while reading record "
1004 "length (%d of %d)\n", len, want);
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001005 return -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 }
1007
1008 svsk->sk_reclen = ntohl(svsk->sk_reclen);
Chuck Leverc0401ea2008-04-14 12:27:30 -04001009 if (!(svsk->sk_reclen & RPC_LAST_STREAM_FRAGMENT)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 /* FIXME: technically, a record can be fragmented,
1011 * and non-terminal fragments will not have the top
1012 * bit set in the fragment length header.
1013 * But apparently no known nfs clients send fragmented
1014 * records. */
NeilBrown34e9a632007-01-29 13:19:52 -08001015 if (net_ratelimit())
Chuck Leverc0401ea2008-04-14 12:27:30 -04001016 printk(KERN_NOTICE "RPC: multiple fragments "
1017 "per record not supported\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 goto err_delete;
1019 }
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +03001020
Chuck Leverc0401ea2008-04-14 12:27:30 -04001021 svsk->sk_reclen &= RPC_FRAGMENT_SIZE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen);
NeilBrownc6b0a9f2006-10-06 00:44:05 -07001023 if (svsk->sk_reclen > serv->sv_max_mesg) {
NeilBrown34e9a632007-01-29 13:19:52 -08001024 if (net_ratelimit())
Chuck Leverc0401ea2008-04-14 12:27:30 -04001025 printk(KERN_NOTICE "RPC: "
1026 "fragment too large: 0x%08lx\n",
1027 (unsigned long)svsk->sk_reclen);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 goto err_delete;
1029 }
1030 }
1031
J. Bruce Fieldscc6c2122011-02-14 15:03:35 -05001032 if (svsk->sk_reclen < 8)
1033 goto err_delete; /* client is nuts. */
1034
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 len = svsk->sk_reclen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +03001037 return len;
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001038error:
1039 dprintk("RPC: TCP recv_record got %d\n", len);
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +03001040 return len;
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001041err_delete:
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +03001042 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +03001043 return -EAGAIN;
1044}
1045
Trond Myklebust586c52c2009-05-18 17:47:56 -04001046static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
Rahul Iyer4cfc7e62009-09-10 17:32:28 +03001047{
Trond Myklebust586c52c2009-05-18 17:47:56 -04001048 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
Rahul Iyer4cfc7e62009-09-10 17:32:28 +03001049 struct rpc_rqst *req = NULL;
Trond Myklebust586c52c2009-05-18 17:47:56 -04001050 struct kvec *src, *dst;
1051 __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
J. Bruce Fields48e65552011-02-14 14:52:03 -05001052 __be32 xid;
1053 __be32 calldir;
Rahul Iyer4cfc7e62009-09-10 17:32:28 +03001054
Rahul Iyer4cfc7e62009-09-10 17:32:28 +03001055 xid = *p++;
1056 calldir = *p;
1057
Trond Myklebust586c52c2009-05-18 17:47:56 -04001058 if (bc_xprt)
1059 req = xprt_lookup_rqst(bc_xprt, xid);
J. Bruce Fieldsd75faea2010-11-30 19:15:01 -05001060
Trond Myklebust586c52c2009-05-18 17:47:56 -04001061 if (!req) {
1062 printk(KERN_NOTICE
1063 "%s: Got unrecognized reply: "
1064 "calldir 0x%x xpt_bc_xprt %p xid %08x\n",
1065 __func__, ntohl(calldir),
1066 bc_xprt, xid);
1067 return -EAGAIN;
Rahul Iyer4cfc7e62009-09-10 17:32:28 +03001068 }
Trond Myklebust586c52c2009-05-18 17:47:56 -04001069
1070 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
1071 /*
1072 * XXX!: cheating for now! Only copying HEAD.
1073 * But we know this is good enough for now (in fact, for any
1074 * callback reply in the forseeable future).
1075 */
1076 dst = &req->rq_private_buf.head[0];
1077 src = &rqstp->rq_arg.head[0];
1078 if (dst->iov_len < src->iov_len)
1079 return -EAGAIN; /* whatever; just giving up. */
1080 memcpy(dst->iov_base, src->iov_base, src->iov_len);
1081 xprt_complete_rqst(req->rq_task, svsk->sk_reclen);
1082 rqstp->rq_arg.len = 0;
1083 return 0;
1084}
1085
1086static int copy_pages_to_kvecs(struct kvec *vec, struct page **pages, int len)
1087{
1088 int i = 0;
1089 int t = 0;
1090
1091 while (t < len) {
1092 vec[i].iov_base = page_address(pages[i]);
1093 vec[i].iov_len = PAGE_SIZE;
1094 i++;
1095 t += PAGE_SIZE;
1096 }
1097 return i;
Rahul Iyer4cfc7e62009-09-10 17:32:28 +03001098}
1099
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001100
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +03001101/*
1102 * Receive data from a TCP socket.
1103 */
1104static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
1105{
1106 struct svc_sock *svsk =
1107 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
1108 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1109 int len;
1110 struct kvec *vec;
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001111 unsigned int want, base;
Trond Myklebust586c52c2009-05-18 17:47:56 -04001112 __be32 *p;
1113 __be32 calldir;
Trond Myklebust5ee78d42009-05-18 17:47:56 -04001114 int pnum;
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +03001115
1116 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1117 svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
1118 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
1119 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1120
1121 len = svc_tcp_recv_record(svsk, rqstp);
1122 if (len < 0)
1123 goto error;
1124
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001125 base = svc_tcp_restore_pages(svsk, rqstp);
1126 want = svsk->sk_reclen - base;
1127
NeilBrown3cc03b12006-10-04 02:15:47 -07001128 vec = rqstp->rq_vec;
Rahul Iyer4cfc7e62009-09-10 17:32:28 +03001129
Trond Myklebust586c52c2009-05-18 17:47:56 -04001130 pnum = copy_pages_to_kvecs(&vec[0], &rqstp->rq_pages[0],
1131 svsk->sk_reclen);
Rahul Iyer4cfc7e62009-09-10 17:32:28 +03001132
NeilBrown44524352006-10-04 02:15:46 -07001133 rqstp->rq_respages = &rqstp->rq_pages[pnum];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134
1135 /* Now receive data */
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001136 len = svc_partial_recvfrom(rqstp, vec, pnum, want, base);
1137 if (len >= 0)
1138 svsk->sk_tcplen += len;
1139 if (len != want) {
1140 if (len < 0 && len != -EAGAIN)
1141 goto err_other;
1142 svc_tcp_save_pages(svsk, rqstp);
1143 dprintk("svc: incomplete TCP record (%d of %d)\n",
1144 svsk->sk_tcplen, svsk->sk_reclen);
1145 goto err_noclose;
1146 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147
Trond Myklebust5ee78d42009-05-18 17:47:56 -04001148 rqstp->rq_arg.len = svsk->sk_reclen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149 rqstp->rq_arg.page_base = 0;
Trond Myklebust5ee78d42009-05-18 17:47:56 -04001150 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1151 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 rqstp->rq_arg.page_len = 0;
Trond Myklebust5ee78d42009-05-18 17:47:56 -04001153 } else
1154 rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155
Tom Tucker5148bf42007-12-30 21:07:25 -06001156 rqstp->rq_xprt_ctxt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 rqstp->rq_prot = IPPROTO_TCP;
1158
Trond Myklebust586c52c2009-05-18 17:47:56 -04001159 p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1160 calldir = p[1];
J. Bruce Fields8985ef02011-04-09 10:03:10 -04001161 if (calldir)
Trond Myklebust586c52c2009-05-18 17:47:56 -04001162 len = receive_cb_reply(svsk, rqstp);
Trond Myklebust586c52c2009-05-18 17:47:56 -04001163
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 /* Reset TCP read info */
1165 svsk->sk_reclen = 0;
1166 svsk->sk_tcplen = 0;
Trond Myklebust0601f792009-05-18 17:47:56 -04001167 /* If we have more data, signal svc_xprt_enqueue() to try again */
1168 if (svc_recv_available(svsk) > sizeof(rpc_fraghdr))
1169 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1170
J. Bruce Fields8985ef02011-04-09 10:03:10 -04001171 if (len < 0)
1172 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173
Tom Tucker9dbc2402007-12-30 21:08:12 -06001174 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 if (serv->sv_stats)
1176 serv->sv_stats->nettcpcnt++;
1177
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001178 dprintk("svc: TCP complete record (%d bytes)\n", rqstp->rq_arg.len);
Trond Myklebust5ee78d42009-05-18 17:47:56 -04001179 return rqstp->rq_arg.len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +03001181error:
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001182 if (len != -EAGAIN)
1183 goto err_other;
1184 dprintk("RPC: TCP recvfrom got EAGAIN\n");
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +03001185 return -EAGAIN;
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001186err_other:
1187 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1188 svsk->sk_xprt.xpt_server->sv_name, -len);
1189 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1190err_noclose:
1191 return -EAGAIN; /* record not complete */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192}
1193
1194/*
1195 * Send out data on TCP socket.
1196 */
Tom Tucker0f0257e2007-12-30 21:08:27 -06001197static int svc_tcp_sendto(struct svc_rqst *rqstp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198{
1199 struct xdr_buf *xbufp = &rqstp->rq_res;
1200 int sent;
Alexey Dobriyand8ed0292006-09-26 22:29:38 -07001201 __be32 reclen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202
1203 /* Set up the first element of the reply kvec.
1204 * Any other kvecs that may be in use have been taken
1205 * care of by the server implementation itself.
1206 */
1207 reclen = htonl(0x80000000|((xbufp->len ) - 4));
1208 memcpy(xbufp->head[0].iov_base, &reclen, 4);
1209
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 sent = svc_sendto(rqstp, &rqstp->rq_res);
1211 if (sent != xbufp->len) {
Tom Tucker0f0257e2007-12-30 21:08:27 -06001212 printk(KERN_NOTICE
1213 "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1214 "- shutting down socket\n",
Tom Tucker57b1d3b2007-12-30 21:08:22 -06001215 rqstp->rq_xprt->xpt_server->sv_name,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 (sent<0)?"got error":"sent only",
1217 sent, xbufp->len);
Tom Tucker57b1d3b2007-12-30 21:08:22 -06001218 set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
Tom Tuckerf6150c32007-12-30 21:07:57 -06001219 svc_xprt_enqueue(rqstp->rq_xprt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 sent = -EAGAIN;
1221 }
1222 return sent;
1223}
1224
Tom Tuckere831fe62007-12-30 21:07:29 -06001225/*
1226 * Setup response header. TCP has a 4B record length field.
1227 */
1228static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
1229{
1230 struct kvec *resv = &rqstp->rq_res.head[0];
1231
1232 /* tcp needs a space for the record length... */
1233 svc_putnl(resv, 0);
1234}
1235
Tom Tucker323bee32007-12-30 21:07:31 -06001236static int svc_tcp_has_wspace(struct svc_xprt *xprt)
1237{
Tom Tucker57b1d3b2007-12-30 21:08:22 -06001238 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
Trond Myklebust47fcb032009-05-18 17:47:56 -04001239 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
Tom Tucker323bee32007-12-30 21:07:31 -06001240 int required;
Tom Tucker323bee32007-12-30 21:07:31 -06001241
Trond Myklebust47fcb032009-05-18 17:47:56 -04001242 if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
1243 return 1;
1244 required = atomic_read(&xprt->xpt_reserved) + serv->sv_max_mesg;
1245 if (sk_stream_wspace(svsk->sk_sk) >= required)
1246 return 1;
Tom Tucker323bee32007-12-30 21:07:31 -06001247 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
Trond Myklebust47fcb032009-05-18 17:47:56 -04001248 return 0;
Tom Tucker323bee32007-12-30 21:07:31 -06001249}
1250
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001251static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
Pavel Emelyanov62832c02010-09-29 16:04:18 +04001252 struct net *net,
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001253 struct sockaddr *sa, int salen,
1254 int flags)
1255{
Pavel Emelyanov62832c02010-09-29 16:04:18 +04001256 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001257}
1258
Trond Myklebust9e00abc2011-07-13 19:20:49 -04001259#if defined(CONFIG_SUNRPC_BACKCHANNEL)
Andy Adamson1f11a032011-01-06 02:04:26 +00001260static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
1261 struct net *, struct sockaddr *,
1262 int, int);
1263static void svc_bc_sock_free(struct svc_xprt *xprt);
1264
1265static struct svc_xprt *svc_bc_tcp_create(struct svc_serv *serv,
1266 struct net *net,
1267 struct sockaddr *sa, int salen,
1268 int flags)
1269{
1270 return svc_bc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1271}
1272
1273static void svc_bc_tcp_sock_detach(struct svc_xprt *xprt)
1274{
1275}
1276
1277static struct svc_xprt_ops svc_tcp_bc_ops = {
1278 .xpo_create = svc_bc_tcp_create,
1279 .xpo_detach = svc_bc_tcp_sock_detach,
1280 .xpo_free = svc_bc_sock_free,
1281 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1282};
1283
1284static struct svc_xprt_class svc_tcp_bc_class = {
1285 .xcl_name = "tcp-bc",
1286 .xcl_owner = THIS_MODULE,
1287 .xcl_ops = &svc_tcp_bc_ops,
1288 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1289};
Andy Adamson16b2d1e2011-01-06 02:04:27 +00001290
1291static void svc_init_bc_xprt_sock(void)
1292{
1293 svc_reg_xprt_class(&svc_tcp_bc_class);
1294}
1295
1296static void svc_cleanup_bc_xprt_sock(void)
1297{
1298 svc_unreg_xprt_class(&svc_tcp_bc_class);
1299}
Trond Myklebust9e00abc2011-07-13 19:20:49 -04001300#else /* CONFIG_SUNRPC_BACKCHANNEL */
Andy Adamson16b2d1e2011-01-06 02:04:27 +00001301static void svc_init_bc_xprt_sock(void)
1302{
1303}
1304
1305static void svc_cleanup_bc_xprt_sock(void)
1306{
1307}
Trond Myklebust9e00abc2011-07-13 19:20:49 -04001308#endif /* CONFIG_SUNRPC_BACKCHANNEL */
Andy Adamson1f11a032011-01-06 02:04:26 +00001309
Tom Tucker360d8732007-12-30 21:07:17 -06001310static struct svc_xprt_ops svc_tcp_ops = {
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001311 .xpo_create = svc_tcp_create,
Tom Tucker5d137992007-12-30 21:07:23 -06001312 .xpo_recvfrom = svc_tcp_recvfrom,
1313 .xpo_sendto = svc_tcp_sendto,
Tom Tucker5148bf42007-12-30 21:07:25 -06001314 .xpo_release_rqst = svc_release_skb,
Trond Myklebust69b6ba32008-12-23 16:30:11 -05001315 .xpo_detach = svc_tcp_sock_detach,
Tom Tucker755ccea2007-12-30 21:07:27 -06001316 .xpo_free = svc_sock_free,
Tom Tuckere831fe62007-12-30 21:07:29 -06001317 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
Tom Tucker323bee32007-12-30 21:07:31 -06001318 .xpo_has_wspace = svc_tcp_has_wspace,
Tom Tucker38a417c2007-12-30 21:07:36 -06001319 .xpo_accept = svc_tcp_accept,
Tom Tucker360d8732007-12-30 21:07:17 -06001320};
1321
1322static struct svc_xprt_class svc_tcp_class = {
1323 .xcl_name = "tcp",
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001324 .xcl_owner = THIS_MODULE,
Tom Tucker360d8732007-12-30 21:07:17 -06001325 .xcl_ops = &svc_tcp_ops,
Tom Tucker49023152007-12-30 21:07:21 -06001326 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
Tom Tucker360d8732007-12-30 21:07:17 -06001327};
1328
1329void svc_init_xprt_sock(void)
1330{
1331 svc_reg_xprt_class(&svc_tcp_class);
1332 svc_reg_xprt_class(&svc_udp_class);
Andy Adamson16b2d1e2011-01-06 02:04:27 +00001333 svc_init_bc_xprt_sock();
Tom Tucker360d8732007-12-30 21:07:17 -06001334}
1335
1336void svc_cleanup_xprt_sock(void)
1337{
1338 svc_unreg_xprt_class(&svc_tcp_class);
1339 svc_unreg_xprt_class(&svc_udp_class);
Andy Adamson16b2d1e2011-01-06 02:04:27 +00001340 svc_cleanup_bc_xprt_sock();
Tom Tucker360d8732007-12-30 21:07:17 -06001341}
1342
Tom Tuckerbb5cf162007-12-30 21:07:50 -06001343static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344{
1345 struct sock *sk = svsk->sk_sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346
Stanislav Kinsburskybd4620d2011-12-06 14:19:10 +03001347 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1348 &svsk->sk_xprt, serv);
Tom Tuckerdef13d72007-12-30 21:08:08 -06001349 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350 if (sk->sk_state == TCP_LISTEN) {
1351 dprintk("setting up TCP socket for listening\n");
Tom Tucker02fc6c32007-12-30 21:07:48 -06001352 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353 sk->sk_data_ready = svc_tcp_listen_data_ready;
Tom Tucker02fc6c32007-12-30 21:07:48 -06001354 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 } else {
1356 dprintk("setting up TCP socket for reading\n");
1357 sk->sk_state_change = svc_tcp_state_change;
1358 sk->sk_data_ready = svc_tcp_data_ready;
Trond Myklebust47fcb032009-05-18 17:47:56 -04001359 sk->sk_write_space = svc_tcp_write_space;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360
1361 svsk->sk_reclen = 0;
1362 svsk->sk_tcplen = 0;
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001363 memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364
Chuck Leverb7872fe2008-04-14 12:27:01 -04001365 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366
Tom Tucker02fc6c32007-12-30 21:07:48 -06001367 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -08001368 if (sk->sk_state != TCP_ESTABLISHED)
Tom Tucker02fc6c32007-12-30 21:07:48 -06001369 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 }
1371}
1372
Tom Tucker0f0257e2007-12-30 21:08:27 -06001373void svc_sock_update_bufs(struct svc_serv *serv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374{
1375 /*
1376 * The number of server threads has changed. Update
1377 * rcvbuf and sndbuf accordingly on all sockets
1378 */
Pavel Emelyanov8f3a6de2010-10-05 23:30:19 +04001379 struct svc_sock *svsk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380
1381 spin_lock_bh(&serv->sv_lock);
Pavel Emelyanov8f3a6de2010-10-05 23:30:19 +04001382 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
Tom Tucker02fc6c32007-12-30 21:07:48 -06001383 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384 spin_unlock_bh(&serv->sv_lock);
1385}
Trond Myklebust24c37672008-12-23 16:30:12 -05001386EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387
1388/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389 * Initialize socket for RPC use and create svc_sock struct
1390 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1391 */
Chuck Lever6b174332007-02-12 00:53:28 -08001392static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1393 struct socket *sock,
1394 int *errp, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395{
1396 struct svc_sock *svsk;
1397 struct sock *inet;
Chuck Lever6b174332007-02-12 00:53:28 -08001398 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399
1400 dprintk("svc: svc_setup_socket %p\n", sock);
Panagiotis Issaris0da974f2006-07-21 14:51:30 -07001401 if (!(svsk = kzalloc(sizeof(*svsk), GFP_KERNEL))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 *errp = -ENOMEM;
1403 return NULL;
1404 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405
1406 inet = sock->sk;
1407
1408 /* Register socket with portmapper */
1409 if (*errp >= 0 && pmap_register)
Trond Myklebustd3b773e2012-01-23 13:22:45 -05001410 *errp = svc_register(serv, sock_net(sock->sk), inet->sk_family,
Stanislav Kinsbursky5247fab2012-01-13 14:02:48 +04001411 inet->sk_protocol,
Eric Dumazetc720c7e2009-10-15 06:30:45 +00001412 ntohs(inet_sk(inet)->inet_sport));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413
1414 if (*errp < 0) {
1415 kfree(svsk);
1416 return NULL;
1417 }
1418
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 inet->sk_user_data = svsk;
1420 svsk->sk_sock = sock;
1421 svsk->sk_sk = inet;
1422 svsk->sk_ostate = inet->sk_state_change;
1423 svsk->sk_odata = inet->sk_data_ready;
1424 svsk->sk_owspace = inet->sk_write_space;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425
1426 /* Initialize the socket */
1427 if (sock->type == SOCK_DGRAM)
Tom Tuckerbb5cf162007-12-30 21:07:50 -06001428 svc_udp_init(svsk, serv);
Olga Kornievskaia96604392008-10-21 14:13:47 -04001429 else {
1430 /* initialise setting must have enough space to
1431 * receive and respond to one request.
1432 */
1433 svc_sock_setbufsize(svsk->sk_sock, 4 * serv->sv_max_mesg,
1434 4 * serv->sv_max_mesg);
Tom Tuckerbb5cf162007-12-30 21:07:50 -06001435 svc_tcp_init(svsk, serv);
Olga Kornievskaia96604392008-10-21 14:13:47 -04001436 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438 dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1439 svsk, svsk->sk_sk);
1440
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 return svsk;
1442}
1443
Chuck Leverbfba9ab2009-04-23 19:32:33 -04001444/**
1445 * svc_addsock - add a listener socket to an RPC service
1446 * @serv: pointer to RPC service to which to add a new listener
1447 * @fd: file descriptor of the new listener
1448 * @name_return: pointer to buffer to fill in with name of listener
1449 * @len: size of the buffer
1450 *
1451 * Fills in socket name and returns positive length of name if successful.
1452 * Name is terminated with '\n'. On error, returns a negative errno
1453 * value.
1454 */
1455int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1456 const size_t len)
NeilBrownb41b66d2006-10-02 02:17:48 -07001457{
1458 int err = 0;
1459 struct socket *so = sockfd_lookup(fd, &err);
1460 struct svc_sock *svsk = NULL;
1461
1462 if (!so)
1463 return err;
Aime Le Rouzic205ba422010-01-26 14:03:56 -05001464 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
NeilBrownb41b66d2006-10-02 02:17:48 -07001465 err = -EAFNOSUPPORT;
1466 else if (so->sk->sk_protocol != IPPROTO_TCP &&
1467 so->sk->sk_protocol != IPPROTO_UDP)
1468 err = -EPROTONOSUPPORT;
1469 else if (so->state > SS_UNCONNECTED)
1470 err = -EISCONN;
1471 else {
Tom Tucker2da2c212008-11-23 09:58:08 -06001472 if (!try_module_get(THIS_MODULE))
1473 err = -ENOENT;
1474 else
1475 svsk = svc_setup_socket(serv, so, &err,
1476 SVC_SOCK_DEFAULTS);
NeilBrowne79eff12007-02-12 00:53:30 -08001477 if (svsk) {
Tom Tucker9dbc2402007-12-30 21:08:12 -06001478 struct sockaddr_storage addr;
1479 struct sockaddr *sin = (struct sockaddr *)&addr;
1480 int salen;
1481 if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0)
1482 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
Tom Tucker4e5caaa2007-12-30 21:08:20 -06001483 clear_bit(XPT_TEMP, &svsk->sk_xprt.xpt_flags);
1484 spin_lock_bh(&serv->sv_lock);
1485 list_add(&svsk->sk_xprt.xpt_list, &serv->sv_permsocks);
1486 spin_unlock_bh(&serv->sv_lock);
Tom Tuckera6046f72007-12-30 21:08:01 -06001487 svc_xprt_received(&svsk->sk_xprt);
NeilBrownb41b66d2006-10-02 02:17:48 -07001488 err = 0;
Tom Tucker2da2c212008-11-23 09:58:08 -06001489 } else
1490 module_put(THIS_MODULE);
NeilBrownb41b66d2006-10-02 02:17:48 -07001491 }
1492 if (err) {
1493 sockfd_put(so);
1494 return err;
1495 }
Chuck Levere7942b92009-04-23 19:32:48 -04001496 return svc_one_sock_name(svsk, name_return, len);
NeilBrownb41b66d2006-10-02 02:17:48 -07001497}
1498EXPORT_SYMBOL_GPL(svc_addsock);
1499
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500/*
1501 * Create socket for RPC service.
1502 */
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001503static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1504 int protocol,
Pavel Emelyanov62832c02010-09-29 16:04:18 +04001505 struct net *net,
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001506 struct sockaddr *sin, int len,
1507 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508{
1509 struct svc_sock *svsk;
1510 struct socket *sock;
1511 int error;
1512 int type;
Tom Tucker9dbc2402007-12-30 21:08:12 -06001513 struct sockaddr_storage addr;
1514 struct sockaddr *newsin = (struct sockaddr *)&addr;
1515 int newlen;
Trond Myklebustc69da772009-03-30 18:59:17 -04001516 int family;
1517 int val;
Pavel Emelyanov5216a8e2008-02-21 10:57:45 +03001518 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519
Chuck Leverad06e4b2007-02-12 00:53:32 -08001520 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1521 serv->sv_program->pg_name, protocol,
Chuck Lever77f1f672007-02-12 00:53:39 -08001522 __svc_print_addr(sin, buf, sizeof(buf)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523
1524 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1525 printk(KERN_WARNING "svc: only UDP and TCP "
1526 "sockets supported\n");
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001527 return ERR_PTR(-EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529
Trond Myklebustc69da772009-03-30 18:59:17 -04001530 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1531 switch (sin->sa_family) {
1532 case AF_INET6:
1533 family = PF_INET6;
1534 break;
1535 case AF_INET:
1536 family = PF_INET;
1537 break;
1538 default:
1539 return ERR_PTR(-EINVAL);
1540 }
1541
Pavel Emelyanov14ec63c2010-09-29 16:06:57 +04001542 error = __sock_create(net, family, type, protocol, &sock, 1);
Chuck Lever77f1f672007-02-12 00:53:39 -08001543 if (error < 0)
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001544 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545
Peter Zijlstraed075362006-12-06 20:35:24 -08001546 svc_reclassify_socket(sock);
1547
Trond Myklebustc69da772009-03-30 18:59:17 -04001548 /*
1549 * If this is an PF_INET6 listener, we want to avoid
1550 * getting requests from IPv4 remotes. Those should
1551 * be shunted to a PF_INET listener via rpcbind.
1552 */
1553 val = 1;
1554 if (family == PF_INET6)
1555 kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
1556 (char *)&val, sizeof(val));
1557
Eric Sesterhenn18114742006-09-28 14:37:07 -07001558 if (type == SOCK_STREAM)
Chuck Lever77f1f672007-02-12 00:53:39 -08001559 sock->sk->sk_reuse = 1; /* allow address reuse */
1560 error = kernel_bind(sock, sin, len);
Eric Sesterhenn18114742006-09-28 14:37:07 -07001561 if (error < 0)
1562 goto bummer;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563
Tom Tucker9dbc2402007-12-30 21:08:12 -06001564 newlen = len;
1565 error = kernel_getsockname(sock, newsin, &newlen);
1566 if (error < 0)
1567 goto bummer;
1568
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 if (protocol == IPPROTO_TCP) {
Sridhar Samudralae6242e92006-08-07 20:58:01 -07001570 if ((error = kernel_listen(sock, 64)) < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571 goto bummer;
1572 }
1573
NeilBrowne79eff12007-02-12 00:53:30 -08001574 if ((svsk = svc_setup_socket(serv, sock, &error, flags)) != NULL) {
Tom Tucker9dbc2402007-12-30 21:08:12 -06001575 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001576 return (struct svc_xprt *)svsk;
NeilBrowne79eff12007-02-12 00:53:30 -08001577 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578
1579bummer:
1580 dprintk("svc: svc_create_socket error = %d\n", -error);
1581 sock_release(sock);
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001582 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583}
1584
1585/*
Tom Tucker755ccea2007-12-30 21:07:27 -06001586 * Detach the svc_sock from the socket so that no
1587 * more callbacks occur.
1588 */
1589static void svc_sock_detach(struct svc_xprt *xprt)
1590{
1591 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1592 struct sock *sk = svsk->sk_sk;
Eric Dumazeteaefd112011-02-18 03:26:36 +00001593 wait_queue_head_t *wq;
Tom Tucker755ccea2007-12-30 21:07:27 -06001594
1595 dprintk("svc: svc_sock_detach(%p)\n", svsk);
1596
1597 /* put back the old socket callbacks */
1598 sk->sk_state_change = svsk->sk_ostate;
1599 sk->sk_data_ready = svsk->sk_odata;
1600 sk->sk_write_space = svsk->sk_owspace;
Trond Myklebust69b6ba32008-12-23 16:30:11 -05001601
Eric Dumazeteaefd112011-02-18 03:26:36 +00001602 wq = sk_sleep(sk);
1603 if (wq && waitqueue_active(wq))
1604 wake_up_interruptible(wq);
Trond Myklebust69b6ba32008-12-23 16:30:11 -05001605}
1606
1607/*
1608 * Disconnect the socket, and reset the callbacks
1609 */
1610static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1611{
1612 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1613
1614 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
1615
1616 svc_sock_detach(xprt);
1617
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001618 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1619 svc_tcp_clear_pages(svsk);
Trond Myklebust69b6ba32008-12-23 16:30:11 -05001620 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001621 }
Tom Tucker755ccea2007-12-30 21:07:27 -06001622}
1623
1624/*
1625 * Free the svc_sock's socket resources and the svc_sock itself.
1626 */
1627static void svc_sock_free(struct svc_xprt *xprt)
1628{
1629 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1630 dprintk("svc: svc_sock_free(%p)\n", svsk);
1631
Tom Tucker755ccea2007-12-30 21:07:27 -06001632 if (svsk->sk_sock->file)
1633 sockfd_put(svsk->sk_sock);
1634 else
1635 sock_release(svsk->sk_sock);
1636 kfree(svsk);
1637}
Benny Halevy7652e5a2009-04-01 09:23:09 -04001638
Trond Myklebust9e00abc2011-07-13 19:20:49 -04001639#if defined(CONFIG_SUNRPC_BACKCHANNEL)
Benny Halevy7652e5a2009-04-01 09:23:09 -04001640/*
Andy Adamson1f11a032011-01-06 02:04:26 +00001641 * Create a back channel svc_xprt which shares the fore channel socket.
Benny Halevy7652e5a2009-04-01 09:23:09 -04001642 */
Andy Adamson1f11a032011-01-06 02:04:26 +00001643static struct svc_xprt *svc_bc_create_socket(struct svc_serv *serv,
1644 int protocol,
1645 struct net *net,
1646 struct sockaddr *sin, int len,
1647 int flags)
Benny Halevy7652e5a2009-04-01 09:23:09 -04001648{
1649 struct svc_sock *svsk;
Andy Adamson1f11a032011-01-06 02:04:26 +00001650 struct svc_xprt *xprt;
Benny Halevy7652e5a2009-04-01 09:23:09 -04001651
Andy Adamson1f11a032011-01-06 02:04:26 +00001652 if (protocol != IPPROTO_TCP) {
1653 printk(KERN_WARNING "svc: only TCP sockets"
1654 " supported on shared back channel\n");
1655 return ERR_PTR(-EINVAL);
1656 }
1657
Benny Halevy7652e5a2009-04-01 09:23:09 -04001658 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1659 if (!svsk)
Andy Adamson1f11a032011-01-06 02:04:26 +00001660 return ERR_PTR(-ENOMEM);
Benny Halevy7652e5a2009-04-01 09:23:09 -04001661
1662 xprt = &svsk->sk_xprt;
Stanislav Kinsburskybd4620d2011-12-06 14:19:10 +03001663 svc_xprt_init(net, &svc_tcp_bc_class, xprt, serv);
Andy Adamson1f11a032011-01-06 02:04:26 +00001664
Andy Adamson4a19de02011-01-06 02:04:35 +00001665 serv->sv_bc_xprt = xprt;
Andy Adamson1f11a032011-01-06 02:04:26 +00001666
Benny Halevy7652e5a2009-04-01 09:23:09 -04001667 return xprt;
1668}
Benny Halevy7652e5a2009-04-01 09:23:09 -04001669
1670/*
Andy Adamson1f11a032011-01-06 02:04:26 +00001671 * Free a back channel svc_sock.
Benny Halevy7652e5a2009-04-01 09:23:09 -04001672 */
Andy Adamson1f11a032011-01-06 02:04:26 +00001673static void svc_bc_sock_free(struct svc_xprt *xprt)
Benny Halevy7652e5a2009-04-01 09:23:09 -04001674{
Andy Adamson778be232011-01-25 15:38:01 +00001675 if (xprt)
Benny Halevy7652e5a2009-04-01 09:23:09 -04001676 kfree(container_of(xprt, struct svc_sock, sk_xprt));
1677}
Trond Myklebust9e00abc2011-07-13 19:20:49 -04001678#endif /* CONFIG_SUNRPC_BACKCHANNEL */