blob: 1413cdcc131c4a7ed49487e4a19691644e0ca65c [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/net/sunrpc/svcsock.c
3 *
4 * These are the RPC server socket internals.
5 *
6 * The server scheduling algorithm does not always distribute the load
7 * evenly when servicing a single client. May need to modify the
Tom Tuckerf6150c32007-12-30 21:07:57 -06008 * svc_xprt_enqueue procedure...
Linus Torvalds1da177e2005-04-16 15:20:36 -07009 *
10 * TCP support is largely untested and may be a little slow. The problem
11 * is that we currently do two separate recvfrom's, one for the 4-byte
12 * record length, and the second for the actual record. This could possibly
13 * be improved by always reading a minimum size of around 100 bytes and
14 * tucking any superfluous bytes away in a temporary store. Still, that
15 * leaves write requests out in the rain. An alternative may be to peek at
16 * the first skb in the queue, and if it matches the next TCP sequence
17 * number, to extract the record marker. Yuck.
18 *
19 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
20 */
21
Ilpo Järvinen172589c2007-08-28 15:50:33 -070022#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include <linux/sched.h>
Paul Gortmaker3a9a2312011-05-27 09:12:25 -040024#include <linux/module.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/errno.h>
26#include <linux/fcntl.h>
27#include <linux/net.h>
28#include <linux/in.h>
29#include <linux/inet.h>
30#include <linux/udp.h>
Andrew Morton91483c42005-08-09 20:20:07 -070031#include <linux/tcp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include <linux/unistd.h>
33#include <linux/slab.h>
34#include <linux/netdevice.h>
35#include <linux/skbuff.h>
NeilBrownb41b66d2006-10-02 02:17:48 -070036#include <linux/file.h>
Nigel Cunningham7dfb7102006-12-06 20:34:23 -080037#include <linux/freezer.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038#include <net/sock.h>
39#include <net/checksum.h>
40#include <net/ip.h>
Chuck Leverb92503b2007-02-12 00:53:36 -080041#include <net/ipv6.h>
Chuck Leverb7872fe2008-04-14 12:27:01 -040042#include <net/tcp.h>
Arnaldo Carvalho de Meloc752f072005-08-09 20:08:28 -070043#include <net/tcp_states.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070044#include <asm/uaccess.h>
45#include <asm/ioctls.h>
Eric Dumazet22911fc2012-06-27 00:23:44 +000046#include <trace/events/skb.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070047
48#include <linux/sunrpc/types.h>
Chuck Leverad06e4b2007-02-12 00:53:32 -080049#include <linux/sunrpc/clnt.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070050#include <linux/sunrpc/xdr.h>
Chuck Leverc0401ea2008-04-14 12:27:30 -040051#include <linux/sunrpc/msg_prot.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070052#include <linux/sunrpc/svcsock.h>
53#include <linux/sunrpc/stats.h>
Rahul Iyer4cfc7e62009-09-10 17:32:28 +030054#include <linux/sunrpc/xprt.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070055
H Hartley Sweeten177e4f92011-06-20 17:54:51 -070056#include "sunrpc.h"
57
Tom Tucker360d87382007-12-30 21:07:17 -060058#define RPCDBG_FACILITY RPCDBG_SVCXPRT
Linus Torvalds1da177e2005-04-16 15:20:36 -070059
60
61static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
J. Bruce Fields72c35372012-08-13 17:46:17 -040062 int flags);
David S. Miller676d2362014-04-11 16:15:36 -040063static void svc_udp_data_ready(struct sock *);
Linus Torvalds1da177e2005-04-16 15:20:36 -070064static int svc_udp_recvfrom(struct svc_rqst *);
65static int svc_udp_sendto(struct svc_rqst *);
Tom Tucker755ccea2007-12-30 21:07:27 -060066static void svc_sock_detach(struct svc_xprt *);
Trond Myklebust69b6ba32008-12-23 16:30:11 -050067static void svc_tcp_sock_detach(struct svc_xprt *);
Tom Tucker755ccea2007-12-30 21:07:27 -060068static void svc_sock_free(struct svc_xprt *);
Linus Torvalds1da177e2005-04-16 15:20:36 -070069
Tom Tuckerb700cbb2007-12-30 21:07:42 -060070static struct svc_xprt *svc_create_socket(struct svc_serv *, int,
Pavel Emelyanov62832c02010-09-29 16:04:18 +040071 struct net *, struct sockaddr *,
72 int, int);
Trond Myklebust9e00abc2011-07-13 19:20:49 -040073#if defined(CONFIG_SUNRPC_BACKCHANNEL)
Andy Adamson1f11a032011-01-06 02:04:26 +000074static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
75 struct net *, struct sockaddr *,
76 int, int);
77static void svc_bc_sock_free(struct svc_xprt *xprt);
Trond Myklebust9e00abc2011-07-13 19:20:49 -040078#endif /* CONFIG_SUNRPC_BACKCHANNEL */
Andy Adamson1f11a032011-01-06 02:04:26 +000079
Peter Zijlstraed075362006-12-06 20:35:24 -080080#ifdef CONFIG_DEBUG_LOCK_ALLOC
81static struct lock_class_key svc_key[2];
82static struct lock_class_key svc_slock_key[2];
83
Tom Tucker0f0257e2007-12-30 21:08:27 -060084static void svc_reclassify_socket(struct socket *sock)
Peter Zijlstraed075362006-12-06 20:35:24 -080085{
86 struct sock *sk = sock->sk;
Weston Andros Adamson1b7a1812012-10-23 10:43:39 -040087
88 WARN_ON_ONCE(sock_owned_by_user(sk));
89 if (sock_owned_by_user(sk))
90 return;
91
Peter Zijlstraed075362006-12-06 20:35:24 -080092 switch (sk->sk_family) {
93 case AF_INET:
94 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
Tom Tuckerdef13d72007-12-30 21:08:08 -060095 &svc_slock_key[0],
96 "sk_xprt.xpt_lock-AF_INET-NFSD",
97 &svc_key[0]);
Peter Zijlstraed075362006-12-06 20:35:24 -080098 break;
99
100 case AF_INET6:
101 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
Tom Tuckerdef13d72007-12-30 21:08:08 -0600102 &svc_slock_key[1],
103 "sk_xprt.xpt_lock-AF_INET6-NFSD",
104 &svc_key[1]);
Peter Zijlstraed075362006-12-06 20:35:24 -0800105 break;
106
107 default:
108 BUG();
109 }
110}
111#else
Tom Tucker0f0257e2007-12-30 21:08:27 -0600112static void svc_reclassify_socket(struct socket *sock)
Peter Zijlstraed075362006-12-06 20:35:24 -0800113{
114}
115#endif
116
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117/*
118 * Release an skbuff after use
119 */
Tom Tucker5148bf42007-12-30 21:07:25 -0600120static void svc_release_skb(struct svc_rqst *rqstp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121{
Tom Tucker5148bf42007-12-30 21:07:25 -0600122 struct sk_buff *skb = rqstp->rq_xprt_ctxt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123
124 if (skb) {
Tom Tucker57b1d3b2007-12-30 21:08:22 -0600125 struct svc_sock *svsk =
126 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
Tom Tucker5148bf42007-12-30 21:07:25 -0600127 rqstp->rq_xprt_ctxt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128
129 dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
Eric Dumazet9d410c72009-10-30 05:03:53 +0000130 skb_free_datagram_locked(svsk->sk_sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132}
133
Chuck Leverb92503b2007-02-12 00:53:36 -0800134union svc_pktinfo_u {
135 struct in_pktinfo pkti;
Chuck Leverb92503b2007-02-12 00:53:36 -0800136 struct in6_pktinfo pkti6;
Chuck Leverb92503b2007-02-12 00:53:36 -0800137};
David S. Millerbc375ea2007-04-12 13:35:59 -0700138#define SVC_PKTINFO_SPACE \
139 CMSG_SPACE(sizeof(union svc_pktinfo_u))
Chuck Leverb92503b2007-02-12 00:53:36 -0800140
141static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
142{
Tom Tucker57b1d3b2007-12-30 21:08:22 -0600143 struct svc_sock *svsk =
144 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
145 switch (svsk->sk_sk->sk_family) {
Chuck Leverb92503b2007-02-12 00:53:36 -0800146 case AF_INET: {
147 struct in_pktinfo *pki = CMSG_DATA(cmh);
148
149 cmh->cmsg_level = SOL_IP;
150 cmh->cmsg_type = IP_PKTINFO;
151 pki->ipi_ifindex = 0;
Mi Jinlong849a1cf2011-08-30 17:18:41 +0800152 pki->ipi_spec_dst.s_addr =
153 svc_daddr_in(rqstp)->sin_addr.s_addr;
Chuck Leverb92503b2007-02-12 00:53:36 -0800154 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
155 }
156 break;
NeilBrown5a05ed72007-03-06 01:42:22 -0800157
Chuck Leverb92503b2007-02-12 00:53:36 -0800158 case AF_INET6: {
159 struct in6_pktinfo *pki = CMSG_DATA(cmh);
Mi Jinlong849a1cf2011-08-30 17:18:41 +0800160 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
Chuck Leverb92503b2007-02-12 00:53:36 -0800161
162 cmh->cmsg_level = SOL_IPV6;
163 cmh->cmsg_type = IPV6_PKTINFO;
Mi Jinlong849a1cf2011-08-30 17:18:41 +0800164 pki->ipi6_ifindex = daddr->sin6_scope_id;
Alexey Dobriyan4e3fd7a2011-11-21 03:39:03 +0000165 pki->ipi6_addr = daddr->sin6_addr;
Chuck Leverb92503b2007-02-12 00:53:36 -0800166 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
167 }
168 break;
Chuck Leverb92503b2007-02-12 00:53:36 -0800169 }
Chuck Leverb92503b2007-02-12 00:53:36 -0800170}
171
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172/*
Rahul Iyer4cfc7e62009-09-10 17:32:28 +0300173 * send routine intended to be shared by the fore- and back-channel
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174 */
Rahul Iyer4cfc7e62009-09-10 17:32:28 +0300175int svc_send_common(struct socket *sock, struct xdr_buf *xdr,
176 struct page *headpage, unsigned long headoffset,
177 struct page *tailpage, unsigned long tailoffset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 int result;
180 int size;
181 struct page **ppage = xdr->pages;
182 size_t base = xdr->page_base;
183 unsigned int pglen = xdr->page_len;
Trond Myklebust226453d2015-10-07 13:54:42 -0400184 unsigned int flags = MSG_MORE | MSG_SENDPAGE_NOTLAST;
Rahul Iyer4cfc7e62009-09-10 17:32:28 +0300185 int slen;
186 int len = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187
188 slen = xdr->len;
189
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 /* send head */
191 if (slen == xdr->head[0].iov_len)
192 flags = 0;
Rahul Iyer4cfc7e62009-09-10 17:32:28 +0300193 len = kernel_sendpage(sock, headpage, headoffset,
NeilBrown44524352006-10-04 02:15:46 -0700194 xdr->head[0].iov_len, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 if (len != xdr->head[0].iov_len)
196 goto out;
197 slen -= xdr->head[0].iov_len;
198 if (slen == 0)
199 goto out;
200
201 /* send page data */
202 size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen;
203 while (pglen > 0) {
204 if (slen == size)
205 flags = 0;
Sridhar Samudralae6242e92006-08-07 20:58:01 -0700206 result = kernel_sendpage(sock, *ppage, base, size, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 if (result > 0)
208 len += result;
209 if (result != size)
210 goto out;
211 slen -= size;
212 pglen -= size;
213 size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen;
214 base = 0;
215 ppage++;
216 }
Rahul Iyer4cfc7e62009-09-10 17:32:28 +0300217
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 /* send tail */
219 if (xdr->tail[0].iov_len) {
Rahul Iyer4cfc7e62009-09-10 17:32:28 +0300220 result = kernel_sendpage(sock, tailpage, tailoffset,
221 xdr->tail[0].iov_len, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 if (result > 0)
223 len += result;
224 }
Rahul Iyer4cfc7e62009-09-10 17:32:28 +0300225
226out:
227 return len;
228}
229
230
231/*
232 * Generic sendto routine
233 */
234static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr)
235{
236 struct svc_sock *svsk =
237 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
238 struct socket *sock = svsk->sk_sock;
239 union {
240 struct cmsghdr hdr;
241 long all[SVC_PKTINFO_SPACE / sizeof(long)];
242 } buffer;
243 struct cmsghdr *cmh = &buffer.hdr;
244 int len = 0;
245 unsigned long tailoff;
246 unsigned long headoff;
247 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
248
249 if (rqstp->rq_prot == IPPROTO_UDP) {
250 struct msghdr msg = {
251 .msg_name = &rqstp->rq_addr,
252 .msg_namelen = rqstp->rq_addrlen,
253 .msg_control = cmh,
254 .msg_controllen = sizeof(buffer),
255 .msg_flags = MSG_MORE,
256 };
257
258 svc_set_cmsg_data(rqstp, cmh);
259
Al Virod8725c82014-12-11 00:02:50 -0500260 if (sock_sendmsg(sock, &msg) < 0)
Rahul Iyer4cfc7e62009-09-10 17:32:28 +0300261 goto out;
262 }
263
264 tailoff = ((unsigned long)xdr->tail[0].iov_base) & (PAGE_SIZE-1);
265 headoff = 0;
266 len = svc_send_common(sock, xdr, rqstp->rq_respages[0], headoff,
267 rqstp->rq_respages[0], tailoff);
268
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269out:
Chuck Leverad06e4b2007-02-12 00:53:32 -0800270 dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n",
Tom Tucker57b1d3b2007-12-30 21:08:22 -0600271 svsk, xdr->head[0].iov_base, xdr->head[0].iov_len,
Chuck Leverad06e4b2007-02-12 00:53:32 -0800272 xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273
274 return len;
275}
276
277/*
NeilBrown80212d52006-10-02 02:17:47 -0700278 * Report socket names for nfsdfs
279 */
Chuck Levere7942b92009-04-23 19:32:48 -0400280static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining)
NeilBrown80212d52006-10-02 02:17:47 -0700281{
Chuck Lever017cb472009-04-23 19:33:03 -0400282 const struct sock *sk = svsk->sk_sk;
283 const char *proto_name = sk->sk_protocol == IPPROTO_UDP ?
284 "udp" : "tcp";
NeilBrown80212d52006-10-02 02:17:47 -0700285 int len;
286
Chuck Lever017cb472009-04-23 19:33:03 -0400287 switch (sk->sk_family) {
Chuck Levere7942b92009-04-23 19:32:48 -0400288 case PF_INET:
289 len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n",
Chuck Lever017cb472009-04-23 19:33:03 -0400290 proto_name,
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000291 &inet_sk(sk)->inet_rcv_saddr,
292 inet_sk(sk)->inet_num);
NeilBrown80212d52006-10-02 02:17:47 -0700293 break;
Eric Dumazetc2bb06d2013-10-09 03:05:48 -0700294#if IS_ENABLED(CONFIG_IPV6)
Chuck Lever58de2f82009-04-23 19:32:55 -0400295 case PF_INET6:
296 len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n",
Chuck Lever017cb472009-04-23 19:33:03 -0400297 proto_name,
Eric Dumazetefe42082013-10-03 15:42:29 -0700298 &sk->sk_v6_rcv_saddr,
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000299 inet_sk(sk)->inet_num);
NeilBrown80212d52006-10-02 02:17:47 -0700300 break;
Eric Dumazetc2bb06d2013-10-09 03:05:48 -0700301#endif
NeilBrown80212d52006-10-02 02:17:47 -0700302 default:
Chuck Levere7942b92009-04-23 19:32:48 -0400303 len = snprintf(buf, remaining, "*unknown-%d*\n",
Chuck Lever017cb472009-04-23 19:33:03 -0400304 sk->sk_family);
NeilBrown80212d52006-10-02 02:17:47 -0700305 }
Chuck Levere7942b92009-04-23 19:32:48 -0400306
307 if (len >= remaining) {
308 *buf = '\0';
309 return -ENAMETOOLONG;
NeilBrown80212d52006-10-02 02:17:47 -0700310 }
311 return len;
312}
313
NeilBrown80212d52006-10-02 02:17:47 -0700314/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 * Generic recvfrom routine.
316 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600317static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr,
318 int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319{
Tom Tucker57b1d3b2007-12-30 21:08:22 -0600320 struct svc_sock *svsk =
321 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
Chuck Lever1ba95102007-02-12 00:53:31 -0800322 struct msghdr msg = {
323 .msg_flags = MSG_DONTWAIT,
324 };
325 int len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326
Tom Tucker260c1d12007-12-30 21:08:29 -0600327 rqstp->rq_xprt_hlen = 0;
328
Trond Myklebustf8d1ff42014-08-03 13:03:13 -0400329 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
Chuck Lever1ba95102007-02-12 00:53:31 -0800330 len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen,
331 msg.msg_flags);
Trond Myklebustf8d1ff42014-08-03 13:03:13 -0400332 /* If we read a full record, then assume there may be more
333 * data to read (stream based sockets only!)
334 */
335 if (len == buflen)
336 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
Chuck Lever1ba95102007-02-12 00:53:31 -0800339 svsk, iov[0].iov_base, iov[0].iov_len, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 return len;
341}
342
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800343static int svc_partial_recvfrom(struct svc_rqst *rqstp,
344 struct kvec *iov, int nr,
345 int buflen, unsigned int base)
346{
347 size_t save_iovlen;
Trond Myklebust09acfea2012-03-11 15:22:54 -0400348 void *save_iovbase;
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800349 unsigned int i;
350 int ret;
351
352 if (base == 0)
353 return svc_recvfrom(rqstp, iov, nr, buflen);
354
355 for (i = 0; i < nr; i++) {
356 if (iov[i].iov_len > base)
357 break;
358 base -= iov[i].iov_len;
359 }
360 save_iovlen = iov[i].iov_len;
361 save_iovbase = iov[i].iov_base;
362 iov[i].iov_len -= base;
363 iov[i].iov_base += base;
364 ret = svc_recvfrom(rqstp, &iov[i], nr - i, buflen);
365 iov[i].iov_len = save_iovlen;
366 iov[i].iov_base = save_iovbase;
367 return ret;
368}
369
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370/*
371 * Set socket snd and rcv buffer lengths
372 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600373static void svc_sock_setbufsize(struct socket *sock, unsigned int snd,
374 unsigned int rcv)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375{
376#if 0
377 mm_segment_t oldfs;
378 oldfs = get_fs(); set_fs(KERNEL_DS);
379 sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
380 (char*)&snd, sizeof(snd));
381 sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
382 (char*)&rcv, sizeof(rcv));
383#else
384 /* sock_setsockopt limits use to sysctl_?mem_max,
385 * which isn't acceptable. Until that is made conditional
386 * on not having CAP_SYS_RESOURCE or similar, we go direct...
387 * DaveM said I could!
388 */
389 lock_sock(sock->sk);
390 sock->sk->sk_sndbuf = snd * 2;
391 sock->sk->sk_rcvbuf = rcv * 2;
Trond Myklebust47fcb032009-05-18 17:47:56 -0400392 sock->sk->sk_write_space(sock->sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 release_sock(sock->sk);
394#endif
395}
Chuck Lever16e4d932014-05-19 13:40:22 -0400396
397static int svc_sock_secure_port(struct svc_rqst *rqstp)
398{
399 return svc_port_is_privileged(svc_addr(rqstp));
400}
401
J. Bruce Fields0442f142015-10-09 01:44:07 +0000402static bool sunrpc_waitqueue_active(wait_queue_head_t *wq)
403{
404 if (!wq)
405 return false;
406 /*
407 * There should normally be a memory * barrier here--see
408 * wq_has_sleeper().
409 *
410 * It appears that isn't currently necessary, though, basically
411 * because callers all appear to have sufficient memory barriers
412 * between the time the relevant change is made and the
413 * time they call these callbacks.
414 *
415 * The nfsd code itself doesn't actually explicitly wait on
416 * these waitqueues, but it may wait on them for example in
417 * sendpage() or sendmsg() calls. (And those may be the only
418 * places, since it it uses nonblocking reads.)
419 *
420 * Maybe we should add the memory barriers anyway, but these are
421 * hot paths so we'd need to be convinced there's no sigificant
422 * penalty.
423 */
424 return waitqueue_active(wq);
425}
426
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427/*
428 * INET callback when data has been received on the socket.
429 */
David S. Miller676d2362014-04-11 16:15:36 -0400430static void svc_udp_data_ready(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431{
Neil Brown939bb7e2005-09-13 01:25:39 -0700432 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
Eric Dumazeteaefd112011-02-18 03:26:36 +0000433 wait_queue_head_t *wq = sk_sleep(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434
Neil Brown939bb7e2005-09-13 01:25:39 -0700435 if (svsk) {
David S. Miller676d2362014-04-11 16:15:36 -0400436 dprintk("svc: socket %p(inet %p), busy=%d\n",
437 svsk, sk,
Tom Tucker02fc6c32007-12-30 21:07:48 -0600438 test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
439 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
Tom Tuckerf6150c32007-12-30 21:07:57 -0600440 svc_xprt_enqueue(&svsk->sk_xprt);
Neil Brown939bb7e2005-09-13 01:25:39 -0700441 }
J. Bruce Fields0442f142015-10-09 01:44:07 +0000442 if (sunrpc_waitqueue_active(wq))
Eric Dumazeteaefd112011-02-18 03:26:36 +0000443 wake_up_interruptible(wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444}
445
446/*
447 * INET callback when space is newly available on the socket.
448 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600449static void svc_write_space(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450{
451 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
Eric Dumazeteaefd112011-02-18 03:26:36 +0000452 wait_queue_head_t *wq = sk_sleep(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453
454 if (svsk) {
455 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
Tom Tucker02fc6c32007-12-30 21:07:48 -0600456 svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
Tom Tuckerf6150c32007-12-30 21:07:57 -0600457 svc_xprt_enqueue(&svsk->sk_xprt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 }
459
J. Bruce Fields0442f142015-10-09 01:44:07 +0000460 if (sunrpc_waitqueue_active(wq)) {
Neil Brown939bb7e2005-09-13 01:25:39 -0700461 dprintk("RPC svc_write_space: someone sleeping on %p\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 svsk);
Eric Dumazeteaefd112011-02-18 03:26:36 +0000463 wake_up_interruptible(wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 }
465}
466
Trond Myklebustc7fb3f02014-07-24 23:59:32 -0400467static int svc_tcp_has_wspace(struct svc_xprt *xprt)
468{
469 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
470 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
471 int required;
472
473 if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
474 return 1;
475 required = atomic_read(&xprt->xpt_reserved) + serv->sv_max_mesg;
476 if (sk_stream_wspace(svsk->sk_sk) >= required ||
477 (sk_stream_min_wspace(svsk->sk_sk) == 0 &&
478 atomic_read(&xprt->xpt_reserved) == 0))
479 return 1;
480 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
481 return 0;
482}
483
Trond Myklebust47fcb032009-05-18 17:47:56 -0400484static void svc_tcp_write_space(struct sock *sk)
485{
Trond Myklebustc7fb3f02014-07-24 23:59:32 -0400486 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
Trond Myklebust47fcb032009-05-18 17:47:56 -0400487 struct socket *sock = sk->sk_socket;
488
Trond Myklebustc7fb3f02014-07-24 23:59:32 -0400489 if (!sk_stream_is_writeable(sk) || !sock)
490 return;
491 if (!svsk || svc_tcp_has_wspace(&svsk->sk_xprt))
Trond Myklebust47fcb032009-05-18 17:47:56 -0400492 clear_bit(SOCK_NOSPACE, &sock->flags);
493 svc_write_space(sk);
494}
495
Trond Myklebust51877682014-07-24 23:59:33 -0400496static void svc_tcp_adjust_wspace(struct svc_xprt *xprt)
497{
498 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
499
500 if (svc_tcp_has_wspace(xprt))
501 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
502}
503
Tom Tucker9dbc2402007-12-30 21:08:12 -0600504/*
Chuck Lever7702ce42009-07-13 10:54:26 -0400505 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
506 */
507static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
508 struct cmsghdr *cmh)
509{
510 struct in_pktinfo *pki = CMSG_DATA(cmh);
Mi Jinlong849a1cf2011-08-30 17:18:41 +0800511 struct sockaddr_in *daddr = svc_daddr_in(rqstp);
512
Chuck Lever7702ce42009-07-13 10:54:26 -0400513 if (cmh->cmsg_type != IP_PKTINFO)
514 return 0;
Mi Jinlong849a1cf2011-08-30 17:18:41 +0800515
516 daddr->sin_family = AF_INET;
517 daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr;
Chuck Lever7702ce42009-07-13 10:54:26 -0400518 return 1;
519}
520
521/*
Tom Parkin73df66f2013-01-31 01:02:24 +0000522 * See net/ipv6/datagram.c : ip6_datagram_recv_ctl
Chuck Lever7702ce42009-07-13 10:54:26 -0400523 */
524static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
525 struct cmsghdr *cmh)
526{
527 struct in6_pktinfo *pki = CMSG_DATA(cmh);
Mi Jinlong849a1cf2011-08-30 17:18:41 +0800528 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
529
Chuck Lever7702ce42009-07-13 10:54:26 -0400530 if (cmh->cmsg_type != IPV6_PKTINFO)
531 return 0;
Mi Jinlong849a1cf2011-08-30 17:18:41 +0800532
533 daddr->sin6_family = AF_INET6;
Alexey Dobriyan4e3fd7a2011-11-21 03:39:03 +0000534 daddr->sin6_addr = pki->ipi6_addr;
Mi Jinlong849a1cf2011-08-30 17:18:41 +0800535 daddr->sin6_scope_id = pki->ipi6_ifindex;
Chuck Lever7702ce42009-07-13 10:54:26 -0400536 return 1;
537}
538
539/*
Tom Tucker9dbc2402007-12-30 21:08:12 -0600540 * Copy the UDP datagram's destination address to the rqstp structure.
541 * The 'destination' address in this case is the address to which the
542 * peer sent the datagram, i.e. our local address. For multihomed
543 * hosts, this can change from msg to msg. Note that only the IP
544 * address changes, the port number should remain the same.
545 */
Chuck Lever7702ce42009-07-13 10:54:26 -0400546static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
547 struct cmsghdr *cmh)
Chuck Lever95756482007-02-12 00:53:38 -0800548{
Chuck Lever7702ce42009-07-13 10:54:26 -0400549 switch (cmh->cmsg_level) {
550 case SOL_IP:
551 return svc_udp_get_dest_address4(rqstp, cmh);
552 case SOL_IPV6:
553 return svc_udp_get_dest_address6(rqstp, cmh);
Chuck Lever95756482007-02-12 00:53:38 -0800554 }
Chuck Lever7702ce42009-07-13 10:54:26 -0400555
556 return 0;
Chuck Lever95756482007-02-12 00:53:38 -0800557}
558
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559/*
560 * Receive a datagram from a UDP socket.
561 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600562static int svc_udp_recvfrom(struct svc_rqst *rqstp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563{
Tom Tucker57b1d3b2007-12-30 21:08:22 -0600564 struct svc_sock *svsk =
565 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600566 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 struct sk_buff *skb;
David S. Millerbc375ea2007-04-12 13:35:59 -0700568 union {
569 struct cmsghdr hdr;
570 long all[SVC_PKTINFO_SPACE / sizeof(long)];
571 } buffer;
572 struct cmsghdr *cmh = &buffer.hdr;
NeilBrown7a37f572007-03-06 01:42:21 -0800573 struct msghdr msg = {
574 .msg_name = svc_addr(rqstp),
575 .msg_control = cmh,
576 .msg_controllen = sizeof(buffer),
577 .msg_flags = MSG_DONTWAIT,
578 };
Chuck Leverabc5c442009-04-23 19:31:25 -0400579 size_t len;
580 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581
Tom Tucker02fc6c32007-12-30 21:07:48 -0600582 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 /* udp sockets need large rcvbuf as all pending
584 * requests are still in that buffer. sndbuf must
585 * also be large enough that there is enough space
Greg Banks3262c812006-10-02 02:17:58 -0700586 * for one reply per thread. We count all threads
587 * rather than threads in a particular pool, which
588 * provides an upper bound on the number of threads
589 * which will access the socket.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 */
591 svc_sock_setbufsize(svsk->sk_sock,
NeilBrownc6b0a9f2006-10-06 00:44:05 -0700592 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
593 (serv->sv_nrthreads+3) * serv->sv_max_mesg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594
Tom Tucker02fc6c32007-12-30 21:07:48 -0600595 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
NeilBrown05ed6902007-05-09 02:34:55 -0700596 skb = NULL;
597 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
598 0, 0, MSG_PEEK | MSG_DONTWAIT);
599 if (err >= 0)
600 skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err);
601
602 if (skb == NULL) {
603 if (err != -EAGAIN) {
604 /* possibly an icmp error */
605 dprintk("svc: recvfrom returned error %d\n", -err);
Tom Tucker02fc6c32007-12-30 21:07:48 -0600606 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 }
J. Bruce Fields9f9d2eb2012-08-17 21:35:24 -0400608 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 }
Tom Tucker9dbc2402007-12-30 21:08:12 -0600610 len = svc_addr_len(svc_addr(rqstp));
Tom Tucker9dbc2402007-12-30 21:08:12 -0600611 rqstp->rq_addrlen = len;
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -0700612 if (skb->tstamp.tv64 == 0) {
613 skb->tstamp = ktime_get_real();
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -0800614 /* Don't enable netstamp, sunrpc doesn't
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 need that much accuracy */
616 }
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -0700617 svsk->sk_sk->sk_stamp = skb->tstamp;
Tom Tucker02fc6c32007-12-30 21:07:48 -0600618 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 len = skb->len - sizeof(struct udphdr);
621 rqstp->rq_arg.len = len;
622
Chuck Lever95756482007-02-12 00:53:38 -0800623 rqstp->rq_prot = IPPROTO_UDP;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624
Chuck Lever7702ce42009-07-13 10:54:26 -0400625 if (!svc_udp_get_dest_address(rqstp, cmh)) {
Joe Perchese87cc472012-05-13 21:56:26 +0000626 net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n",
627 cmh->cmsg_level, cmh->cmsg_type);
J. Bruce Fieldsf23abfd2012-08-17 20:32:27 -0400628 goto out_free;
NeilBrown7a37f572007-03-06 01:42:21 -0800629 }
Mi Jinlong849a1cf2011-08-30 17:18:41 +0800630 rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631
632 if (skb_is_nonlinear(skb)) {
633 /* we have to copy */
634 local_bh_disable();
635 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
636 local_bh_enable();
637 /* checksum error */
Eric Dumazet22911fc2012-06-27 00:23:44 +0000638 goto out_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 }
640 local_bh_enable();
Eric Dumazet9d410c72009-10-30 05:03:53 +0000641 skb_free_datagram_locked(svsk->sk_sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 } else {
643 /* we can use it in-place */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600644 rqstp->rq_arg.head[0].iov_base = skb->data +
645 sizeof(struct udphdr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 rqstp->rq_arg.head[0].iov_len = len;
Eric Dumazet22911fc2012-06-27 00:23:44 +0000647 if (skb_checksum_complete(skb))
648 goto out_free;
Tom Tucker5148bf42007-12-30 21:07:25 -0600649 rqstp->rq_xprt_ctxt = skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 }
651
652 rqstp->rq_arg.page_base = 0;
653 if (len <= rqstp->rq_arg.head[0].iov_len) {
654 rqstp->rq_arg.head[0].iov_len = len;
655 rqstp->rq_arg.page_len = 0;
NeilBrown44524352006-10-04 02:15:46 -0700656 rqstp->rq_respages = rqstp->rq_pages+1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 } else {
658 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
NeilBrown44524352006-10-04 02:15:46 -0700659 rqstp->rq_respages = rqstp->rq_pages + 1 +
Ilpo Järvinen172589c2007-08-28 15:50:33 -0700660 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 }
J. Bruce Fieldsafc59402012-12-10 18:01:37 -0500662 rqstp->rq_next_page = rqstp->rq_respages+1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663
664 if (serv->sv_stats)
665 serv->sv_stats->netudpcnt++;
666
667 return len;
J. Bruce Fieldsf23abfd2012-08-17 20:32:27 -0400668out_free:
669 trace_kfree_skb(skb, svc_udp_recvfrom);
670 skb_free_datagram_locked(svsk->sk_sk, skb);
671 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672}
673
674static int
675svc_udp_sendto(struct svc_rqst *rqstp)
676{
677 int error;
678
679 error = svc_sendto(rqstp, &rqstp->rq_res);
680 if (error == -ECONNREFUSED)
681 /* ICMP error on earlier request. */
682 error = svc_sendto(rqstp, &rqstp->rq_res);
683
684 return error;
685}
686
Tom Tuckere831fe62007-12-30 21:07:29 -0600687static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
688{
689}
690
Tom Tucker323bee32007-12-30 21:07:31 -0600691static int svc_udp_has_wspace(struct svc_xprt *xprt)
692{
693 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600694 struct svc_serv *serv = xprt->xpt_server;
Tom Tucker323bee32007-12-30 21:07:31 -0600695 unsigned long required;
696
697 /*
698 * Set the SOCK_NOSPACE flag before checking the available
699 * sock space.
700 */
701 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
Tom Tucker7a90e8c2007-12-30 21:07:55 -0600702 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
Tom Tucker323bee32007-12-30 21:07:31 -0600703 if (required*2 > sock_wspace(svsk->sk_sk))
704 return 0;
705 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
706 return 1;
707}
708
Tom Tucker38a417c2007-12-30 21:07:36 -0600709static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
710{
711 BUG();
712 return NULL;
713}
714
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600715static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
Pavel Emelyanov62832c02010-09-29 16:04:18 +0400716 struct net *net,
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600717 struct sockaddr *sa, int salen,
718 int flags)
719{
Pavel Emelyanov62832c02010-09-29 16:04:18 +0400720 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600721}
722
Tom Tucker360d87382007-12-30 21:07:17 -0600723static struct svc_xprt_ops svc_udp_ops = {
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600724 .xpo_create = svc_udp_create,
Tom Tucker5d137992007-12-30 21:07:23 -0600725 .xpo_recvfrom = svc_udp_recvfrom,
726 .xpo_sendto = svc_udp_sendto,
Tom Tucker5148bf42007-12-30 21:07:25 -0600727 .xpo_release_rqst = svc_release_skb,
Tom Tucker755ccea2007-12-30 21:07:27 -0600728 .xpo_detach = svc_sock_detach,
729 .xpo_free = svc_sock_free,
Tom Tuckere831fe62007-12-30 21:07:29 -0600730 .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
Tom Tucker323bee32007-12-30 21:07:31 -0600731 .xpo_has_wspace = svc_udp_has_wspace,
Tom Tucker38a417c2007-12-30 21:07:36 -0600732 .xpo_accept = svc_udp_accept,
Chuck Lever16e4d932014-05-19 13:40:22 -0400733 .xpo_secure_port = svc_sock_secure_port,
Tom Tucker360d87382007-12-30 21:07:17 -0600734};
735
736static struct svc_xprt_class svc_udp_class = {
737 .xcl_name = "udp",
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600738 .xcl_owner = THIS_MODULE,
Tom Tucker360d87382007-12-30 21:07:17 -0600739 .xcl_ops = &svc_udp_ops,
Tom Tucker49023152007-12-30 21:07:21 -0600740 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
Chuck Lever3c45ddf2014-07-16 15:38:32 -0400741 .xcl_ident = XPRT_TRANSPORT_UDP,
Tom Tucker360d87382007-12-30 21:07:17 -0600742};
743
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600744static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745{
Chuck Lever7702ce42009-07-13 10:54:26 -0400746 int err, level, optname, one = 1;
NeilBrown7a37f572007-03-06 01:42:21 -0800747
Stanislav Kinsburskybd4620d2011-12-06 14:19:10 +0300748 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class,
749 &svsk->sk_xprt, serv);
Tom Tuckerdef13d72007-12-30 21:08:08 -0600750 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751 svsk->sk_sk->sk_data_ready = svc_udp_data_ready;
752 svsk->sk_sk->sk_write_space = svc_write_space;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753
754 /* initialise setting must have enough space to
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -0800755 * receive and respond to one request.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 * svc_udp_recvfrom will re-adjust if necessary
757 */
758 svc_sock_setbufsize(svsk->sk_sock,
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600759 3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
760 3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761
Tom Tucker0f0257e2007-12-30 21:08:27 -0600762 /* data might have come in before data_ready set up */
763 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
Tom Tucker02fc6c32007-12-30 21:07:48 -0600764 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
NeilBrown7a37f572007-03-06 01:42:21 -0800765
NeilBrown7a37f572007-03-06 01:42:21 -0800766 /* make sure we get destination address info */
Chuck Lever7702ce42009-07-13 10:54:26 -0400767 switch (svsk->sk_sk->sk_family) {
768 case AF_INET:
769 level = SOL_IP;
770 optname = IP_PKTINFO;
771 break;
772 case AF_INET6:
773 level = SOL_IPV6;
774 optname = IPV6_RECVPKTINFO;
775 break;
776 default:
777 BUG();
778 }
779 err = kernel_setsockopt(svsk->sk_sock, level, optname,
780 (char *)&one, sizeof(one));
781 dprintk("svc: kernel_setsockopt returned %d\n", err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782}
783
784/*
785 * A data_ready event on a listening socket means there's a connection
786 * pending. Do not use state_change as a substitute for it.
787 */
David S. Miller676d2362014-04-11 16:15:36 -0400788static void svc_tcp_listen_data_ready(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789{
Neil Brown939bb7e2005-09-13 01:25:39 -0700790 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
Eric Dumazeteaefd112011-02-18 03:26:36 +0000791 wait_queue_head_t *wq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792
793 dprintk("svc: socket %p TCP (listen) state change %d\n",
Neil Brown939bb7e2005-09-13 01:25:39 -0700794 sk, sk->sk_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795
Neil Brown939bb7e2005-09-13 01:25:39 -0700796 /*
797 * This callback may called twice when a new connection
798 * is established as a child socket inherits everything
799 * from a parent LISTEN socket.
800 * 1) data_ready method of the parent socket will be called
801 * when one of child sockets become ESTABLISHED.
802 * 2) data_ready method of the child socket may be called
803 * when it receives data before the socket is accepted.
804 * In case of 2, we should ignore it silently.
805 */
806 if (sk->sk_state == TCP_LISTEN) {
807 if (svsk) {
Tom Tucker02fc6c32007-12-30 21:07:48 -0600808 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
Tom Tuckerf6150c32007-12-30 21:07:57 -0600809 svc_xprt_enqueue(&svsk->sk_xprt);
Neil Brown939bb7e2005-09-13 01:25:39 -0700810 } else
811 printk("svc: socket %p: no user data\n", sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 }
Neil Brown939bb7e2005-09-13 01:25:39 -0700813
Eric Dumazeteaefd112011-02-18 03:26:36 +0000814 wq = sk_sleep(sk);
J. Bruce Fields0442f142015-10-09 01:44:07 +0000815 if (sunrpc_waitqueue_active(wq))
Eric Dumazeteaefd112011-02-18 03:26:36 +0000816 wake_up_interruptible_all(wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817}
818
819/*
820 * A state change on a connected socket means it's dying or dead.
821 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600822static void svc_tcp_state_change(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823{
Neil Brown939bb7e2005-09-13 01:25:39 -0700824 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
Eric Dumazeteaefd112011-02-18 03:26:36 +0000825 wait_queue_head_t *wq = sk_sleep(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826
827 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
Neil Brown939bb7e2005-09-13 01:25:39 -0700828 sk, sk->sk_state, sk->sk_user_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829
Neil Brown939bb7e2005-09-13 01:25:39 -0700830 if (!svsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 printk("svc: socket %p: no user data\n", sk);
Neil Brown939bb7e2005-09-13 01:25:39 -0700832 else {
Tom Tucker02fc6c32007-12-30 21:07:48 -0600833 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
Tom Tuckerf6150c32007-12-30 21:07:57 -0600834 svc_xprt_enqueue(&svsk->sk_xprt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 }
J. Bruce Fields0442f142015-10-09 01:44:07 +0000836 if (sunrpc_waitqueue_active(wq))
Eric Dumazeteaefd112011-02-18 03:26:36 +0000837 wake_up_interruptible_all(wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838}
839
David S. Miller676d2362014-04-11 16:15:36 -0400840static void svc_tcp_data_ready(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841{
Neil Brown939bb7e2005-09-13 01:25:39 -0700842 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
Eric Dumazeteaefd112011-02-18 03:26:36 +0000843 wait_queue_head_t *wq = sk_sleep(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844
845 dprintk("svc: socket %p TCP data ready (svsk %p)\n",
Neil Brown939bb7e2005-09-13 01:25:39 -0700846 sk, sk->sk_user_data);
847 if (svsk) {
Tom Tucker02fc6c32007-12-30 21:07:48 -0600848 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
Tom Tuckerf6150c32007-12-30 21:07:57 -0600849 svc_xprt_enqueue(&svsk->sk_xprt);
Neil Brown939bb7e2005-09-13 01:25:39 -0700850 }
J. Bruce Fields0442f142015-10-09 01:44:07 +0000851 if (sunrpc_waitqueue_active(wq))
Eric Dumazeteaefd112011-02-18 03:26:36 +0000852 wake_up_interruptible(wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853}
854
855/*
856 * Accept a TCP connection
857 */
Tom Tucker38a417c2007-12-30 21:07:36 -0600858static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859{
Tom Tucker38a417c2007-12-30 21:07:36 -0600860 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
akpm@linux-foundation.orgcdd88b92007-02-12 00:53:38 -0800861 struct sockaddr_storage addr;
862 struct sockaddr *sin = (struct sockaddr *) &addr;
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600863 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 struct socket *sock = svsk->sk_sock;
865 struct socket *newsock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 struct svc_sock *newsvsk;
867 int err, slen;
Pavel Emelyanov5216a8e2008-02-21 10:57:45 +0300868 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869
870 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
871 if (!sock)
Tom Tucker38a417c2007-12-30 21:07:36 -0600872 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873
Tom Tucker02fc6c32007-12-30 21:07:48 -0600874 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
Sridhar Samudralae6242e92006-08-07 20:58:01 -0700875 err = kernel_accept(sock, &newsock, O_NONBLOCK);
876 if (err < 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 if (err == -ENOMEM)
878 printk(KERN_WARNING "%s: no more sockets!\n",
879 serv->sv_name);
Joe Perchese87cc472012-05-13 21:56:26 +0000880 else if (err != -EAGAIN)
881 net_warn_ratelimited("%s: accept failed (err %d)!\n",
882 serv->sv_name, -err);
Tom Tucker38a417c2007-12-30 21:07:36 -0600883 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 }
Tom Tucker02fc6c32007-12-30 21:07:48 -0600885 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886
akpm@linux-foundation.orgcdd88b92007-02-12 00:53:38 -0800887 err = kernel_getpeername(newsock, sin, &slen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 if (err < 0) {
Joe Perchese87cc472012-05-13 21:56:26 +0000889 net_warn_ratelimited("%s: peername failed (err %d)!\n",
890 serv->sv_name, -err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 goto failed; /* aborted connection or whatever */
892 }
893
894 /* Ideally, we would want to reject connections from unauthorized
Chuck Leverad06e4b2007-02-12 00:53:32 -0800895 * hosts here, but when we get encryption, the IP of the host won't
896 * tell us anything. For now just warn about unpriv connections.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 */
akpm@linux-foundation.orgcdd88b92007-02-12 00:53:38 -0800898 if (!svc_port_is_privileged(sin)) {
Kinglong Meea48fd0f2014-05-29 12:18:52 +0800899 dprintk("%s: connect from unprivileged port: %s\n",
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -0800900 serv->sv_name,
akpm@linux-foundation.orgcdd88b92007-02-12 00:53:38 -0800901 __svc_print_addr(sin, buf, sizeof(buf)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 }
Chuck Leverad06e4b2007-02-12 00:53:32 -0800903 dprintk("%s: connect from %s\n", serv->sv_name,
akpm@linux-foundation.orgcdd88b92007-02-12 00:53:38 -0800904 __svc_print_addr(sin, buf, sizeof(buf)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905
906 /* make sure that a write doesn't block forever when
907 * low on memory
908 */
909 newsock->sk->sk_sndtimeo = HZ*30;
910
J. Bruce Fields72c35372012-08-13 17:46:17 -0400911 newsvsk = svc_setup_socket(serv, newsock,
912 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY));
913 if (IS_ERR(newsvsk))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 goto failed;
Tom Tucker9dbc2402007-12-30 21:08:12 -0600915 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
Frank van Maarseveena9747692007-07-09 22:21:39 +0200916 err = kernel_getsockname(newsock, sin, &slen);
917 if (unlikely(err < 0)) {
918 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
919 slen = offsetof(struct sockaddr, sa_data);
920 }
Tom Tucker9dbc2402007-12-30 21:08:12 -0600921 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
Chuck Lever067d7812007-02-12 00:53:30 -0800922
NeilBrownef11ce22014-05-12 11:22:47 +1000923 if (sock_is_loopback(newsock->sk))
924 set_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
925 else
926 clear_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
Tom Tuckerf9f3cc42007-12-30 21:07:40 -0600927 if (serv->sv_stats)
928 serv->sv_stats->nettcpconn++;
929
930 return &newsvsk->sk_xprt;
931
932failed:
933 sock_release(newsock);
934 return NULL;
935}
936
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800937static unsigned int svc_tcp_restore_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
938{
939 unsigned int i, len, npages;
940
J. Bruce Fields8af345f2012-12-03 16:45:35 -0500941 if (svsk->sk_datalen == 0)
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800942 return 0;
J. Bruce Fields8af345f2012-12-03 16:45:35 -0500943 len = svsk->sk_datalen;
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800944 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
945 for (i = 0; i < npages; i++) {
946 if (rqstp->rq_pages[i] != NULL)
947 put_page(rqstp->rq_pages[i]);
948 BUG_ON(svsk->sk_pages[i] == NULL);
949 rqstp->rq_pages[i] = svsk->sk_pages[i];
950 svsk->sk_pages[i] = NULL;
951 }
952 rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
953 return len;
954}
955
956static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
957{
958 unsigned int i, len, npages;
959
J. Bruce Fields8af345f2012-12-03 16:45:35 -0500960 if (svsk->sk_datalen == 0)
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800961 return;
J. Bruce Fields8af345f2012-12-03 16:45:35 -0500962 len = svsk->sk_datalen;
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800963 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
964 for (i = 0; i < npages; i++) {
965 svsk->sk_pages[i] = rqstp->rq_pages[i];
966 rqstp->rq_pages[i] = NULL;
967 }
968}
969
970static void svc_tcp_clear_pages(struct svc_sock *svsk)
971{
972 unsigned int i, len, npages;
973
J. Bruce Fields8af345f2012-12-03 16:45:35 -0500974 if (svsk->sk_datalen == 0)
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800975 goto out;
J. Bruce Fields8af345f2012-12-03 16:45:35 -0500976 len = svsk->sk_datalen;
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800977 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
978 for (i = 0; i < npages; i++) {
J. Bruce Fieldscf3aa022013-06-26 11:09:06 -0400979 if (svsk->sk_pages[i] == NULL) {
980 WARN_ON_ONCE(1);
981 continue;
982 }
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800983 put_page(svsk->sk_pages[i]);
984 svsk->sk_pages[i] = NULL;
985 }
986out:
987 svsk->sk_tcplen = 0;
J. Bruce Fields8af345f2012-12-03 16:45:35 -0500988 svsk->sk_datalen = 0;
J. Bruce Fields31d68ef2011-02-24 11:25:33 -0800989}
990
Tom Tuckerf9f3cc42007-12-30 21:07:40 -0600991/*
J. Bruce Fieldsad46ccf2012-12-03 16:30:42 -0500992 * Receive fragment record header.
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +0300993 * If we haven't gotten the record length yet, get the next four bytes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 */
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +0300995static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996{
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600997 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
Trond Myklebust5ee78d42009-05-18 17:47:56 -0400998 unsigned int want;
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +0300999 int len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000
Chuck Leverc0401ea2008-04-14 12:27:30 -04001001 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 struct kvec iov;
1003
Trond Myklebust5ee78d42009-05-18 17:47:56 -04001004 want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
1006 iov.iov_len = want;
1007 if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
1008 goto error;
1009 svsk->sk_tcplen += len;
1010
1011 if (len < want) {
Chuck Leverc0401ea2008-04-14 12:27:30 -04001012 dprintk("svc: short recvfrom while reading record "
1013 "length (%d of %d)\n", len, want);
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001014 return -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 }
1016
J. Bruce Fieldscc248d42012-12-03 16:11:13 -05001017 dprintk("svc: TCP record, %d bytes\n", svc_sock_reclen(svsk));
J. Bruce Fields836fbad2012-12-03 15:50:38 -05001018 if (svc_sock_reclen(svsk) + svsk->sk_datalen >
1019 serv->sv_max_mesg) {
J. Bruce Fields3a28e332012-12-10 18:15:07 -05001020 net_notice_ratelimited("RPC: fragment too large: %d\n",
1021 svc_sock_reclen(svsk));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 goto err_delete;
1023 }
1024 }
1025
J. Bruce Fieldscc248d42012-12-03 16:11:13 -05001026 return svc_sock_reclen(svsk);
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001027error:
1028 dprintk("RPC: TCP recv_record got %d\n", len);
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +03001029 return len;
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001030err_delete:
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +03001031 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +03001032 return -EAGAIN;
1033}
1034
Trond Myklebust586c52c2009-05-18 17:47:56 -04001035static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
Rahul Iyer4cfc7e62009-09-10 17:32:28 +03001036{
Trond Myklebust586c52c2009-05-18 17:47:56 -04001037 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
Rahul Iyer4cfc7e62009-09-10 17:32:28 +03001038 struct rpc_rqst *req = NULL;
Trond Myklebust586c52c2009-05-18 17:47:56 -04001039 struct kvec *src, *dst;
1040 __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
J. Bruce Fields48e65552011-02-14 14:52:03 -05001041 __be32 xid;
1042 __be32 calldir;
Rahul Iyer4cfc7e62009-09-10 17:32:28 +03001043
Rahul Iyer4cfc7e62009-09-10 17:32:28 +03001044 xid = *p++;
1045 calldir = *p;
1046
Trond Myklebust093a1462014-11-12 18:04:04 -05001047 if (!bc_xprt)
Trond Myklebust586c52c2009-05-18 17:47:56 -04001048 return -EAGAIN;
Trond Myklebust093a1462014-11-12 18:04:04 -05001049 spin_lock_bh(&bc_xprt->transport_lock);
1050 req = xprt_lookup_rqst(bc_xprt, xid);
1051 if (!req)
1052 goto unlock_notfound;
Trond Myklebust586c52c2009-05-18 17:47:56 -04001053
1054 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
1055 /*
1056 * XXX!: cheating for now! Only copying HEAD.
1057 * But we know this is good enough for now (in fact, for any
1058 * callback reply in the forseeable future).
1059 */
1060 dst = &req->rq_private_buf.head[0];
1061 src = &rqstp->rq_arg.head[0];
1062 if (dst->iov_len < src->iov_len)
Trond Myklebust093a1462014-11-12 18:04:04 -05001063 goto unlock_eagain; /* whatever; just giving up. */
Trond Myklebust586c52c2009-05-18 17:47:56 -04001064 memcpy(dst->iov_base, src->iov_base, src->iov_len);
J. Bruce Fieldscc248d42012-12-03 16:11:13 -05001065 xprt_complete_rqst(req->rq_task, rqstp->rq_arg.len);
Trond Myklebust586c52c2009-05-18 17:47:56 -04001066 rqstp->rq_arg.len = 0;
Trond Myklebust093a1462014-11-12 18:04:04 -05001067 spin_unlock_bh(&bc_xprt->transport_lock);
Trond Myklebust586c52c2009-05-18 17:47:56 -04001068 return 0;
Trond Myklebust093a1462014-11-12 18:04:04 -05001069unlock_notfound:
1070 printk(KERN_NOTICE
1071 "%s: Got unrecognized reply: "
1072 "calldir 0x%x xpt_bc_xprt %p xid %08x\n",
1073 __func__, ntohl(calldir),
1074 bc_xprt, ntohl(xid));
1075unlock_eagain:
1076 spin_unlock_bh(&bc_xprt->transport_lock);
1077 return -EAGAIN;
Trond Myklebust586c52c2009-05-18 17:47:56 -04001078}
1079
1080static int copy_pages_to_kvecs(struct kvec *vec, struct page **pages, int len)
1081{
1082 int i = 0;
1083 int t = 0;
1084
1085 while (t < len) {
1086 vec[i].iov_base = page_address(pages[i]);
1087 vec[i].iov_len = PAGE_SIZE;
1088 i++;
1089 t += PAGE_SIZE;
1090 }
1091 return i;
Rahul Iyer4cfc7e62009-09-10 17:32:28 +03001092}
1093
J. Bruce Fields836fbad2012-12-03 15:50:38 -05001094static void svc_tcp_fragment_received(struct svc_sock *svsk)
1095{
1096 /* If we have more data, signal svc_xprt_enqueue() to try again */
J. Bruce Fields836fbad2012-12-03 15:50:38 -05001097 dprintk("svc: TCP %s record (%d bytes)\n",
1098 svc_sock_final_rec(svsk) ? "final" : "nonfinal",
1099 svc_sock_reclen(svsk));
1100 svsk->sk_tcplen = 0;
1101 svsk->sk_reclen = 0;
1102}
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001103
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +03001104/*
1105 * Receive data from a TCP socket.
1106 */
1107static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
1108{
1109 struct svc_sock *svsk =
1110 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
1111 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1112 int len;
1113 struct kvec *vec;
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001114 unsigned int want, base;
Trond Myklebust586c52c2009-05-18 17:47:56 -04001115 __be32 *p;
1116 __be32 calldir;
Trond Myklebust5ee78d42009-05-18 17:47:56 -04001117 int pnum;
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +03001118
1119 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1120 svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
1121 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
1122 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1123
1124 len = svc_tcp_recv_record(svsk, rqstp);
1125 if (len < 0)
1126 goto error;
1127
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001128 base = svc_tcp_restore_pages(svsk, rqstp);
J. Bruce Fields836fbad2012-12-03 15:50:38 -05001129 want = svc_sock_reclen(svsk) - (svsk->sk_tcplen - sizeof(rpc_fraghdr));
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001130
NeilBrown3cc03b12006-10-04 02:15:47 -07001131 vec = rqstp->rq_vec;
Rahul Iyer4cfc7e62009-09-10 17:32:28 +03001132
Trond Myklebust586c52c2009-05-18 17:47:56 -04001133 pnum = copy_pages_to_kvecs(&vec[0], &rqstp->rq_pages[0],
J. Bruce Fields836fbad2012-12-03 15:50:38 -05001134 svsk->sk_datalen + want);
Rahul Iyer4cfc7e62009-09-10 17:32:28 +03001135
NeilBrown44524352006-10-04 02:15:46 -07001136 rqstp->rq_respages = &rqstp->rq_pages[pnum];
J. Bruce Fieldsafc59402012-12-10 18:01:37 -05001137 rqstp->rq_next_page = rqstp->rq_respages + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138
1139 /* Now receive data */
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001140 len = svc_partial_recvfrom(rqstp, vec, pnum, want, base);
J. Bruce Fields8af345f2012-12-03 16:45:35 -05001141 if (len >= 0) {
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001142 svsk->sk_tcplen += len;
J. Bruce Fields8af345f2012-12-03 16:45:35 -05001143 svsk->sk_datalen += len;
1144 }
J. Bruce Fields836fbad2012-12-03 15:50:38 -05001145 if (len != want || !svc_sock_final_rec(svsk)) {
J. Bruce Fieldsbe1e4442012-08-09 18:12:28 -04001146 svc_tcp_save_pages(svsk, rqstp);
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001147 if (len < 0 && len != -EAGAIN)
J. Bruce Fieldsad46ccf2012-12-03 16:30:42 -05001148 goto err_delete;
J. Bruce Fields836fbad2012-12-03 15:50:38 -05001149 if (len == want)
1150 svc_tcp_fragment_received(svsk);
1151 else
J. Bruce Fields3a28e332012-12-10 18:15:07 -05001152 dprintk("svc: incomplete TCP record (%d of %d)\n",
1153 (int)(svsk->sk_tcplen - sizeof(rpc_fraghdr)),
J. Bruce Fields836fbad2012-12-03 15:50:38 -05001154 svc_sock_reclen(svsk));
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001155 goto err_noclose;
1156 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157
J. Bruce Fields1f691b02013-06-26 10:55:40 -04001158 if (svsk->sk_datalen < 8) {
J. Bruce Fieldscf3aa022013-06-26 11:09:06 -04001159 svsk->sk_datalen = 0;
J. Bruce Fieldsad46ccf2012-12-03 16:30:42 -05001160 goto err_delete; /* client is nuts. */
J. Bruce Fieldscf3aa022013-06-26 11:09:06 -04001161 }
J. Bruce Fieldsad46ccf2012-12-03 16:30:42 -05001162
J. Bruce Fields836fbad2012-12-03 15:50:38 -05001163 rqstp->rq_arg.len = svsk->sk_datalen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 rqstp->rq_arg.page_base = 0;
Trond Myklebust5ee78d42009-05-18 17:47:56 -04001165 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1166 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167 rqstp->rq_arg.page_len = 0;
Trond Myklebust5ee78d42009-05-18 17:47:56 -04001168 } else
1169 rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170
Tom Tucker5148bf42007-12-30 21:07:25 -06001171 rqstp->rq_xprt_ctxt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 rqstp->rq_prot = IPPROTO_TCP;
Jeff Layton7501cc22014-11-19 07:51:15 -05001173 if (test_bit(XPT_LOCAL, &svsk->sk_xprt.xpt_flags))
1174 set_bit(RQ_LOCAL, &rqstp->rq_flags);
1175 else
1176 clear_bit(RQ_LOCAL, &rqstp->rq_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177
Trond Myklebust586c52c2009-05-18 17:47:56 -04001178 p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1179 calldir = p[1];
J. Bruce Fields8985ef02011-04-09 10:03:10 -04001180 if (calldir)
Trond Myklebust586c52c2009-05-18 17:47:56 -04001181 len = receive_cb_reply(svsk, rqstp);
Trond Myklebust586c52c2009-05-18 17:47:56 -04001182
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183 /* Reset TCP read info */
J. Bruce Fields8af345f2012-12-03 16:45:35 -05001184 svsk->sk_datalen = 0;
J. Bruce Fields836fbad2012-12-03 15:50:38 -05001185 svc_tcp_fragment_received(svsk);
Trond Myklebust0601f792009-05-18 17:47:56 -04001186
J. Bruce Fields8985ef02011-04-09 10:03:10 -04001187 if (len < 0)
1188 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189
Tom Tucker9dbc2402007-12-30 21:08:12 -06001190 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191 if (serv->sv_stats)
1192 serv->sv_stats->nettcpcnt++;
1193
Trond Myklebust5ee78d42009-05-18 17:47:56 -04001194 return rqstp->rq_arg.len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195
Alexandros Batsakis8f55f3c2009-08-20 03:34:19 +03001196error:
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001197 if (len != -EAGAIN)
J. Bruce Fieldsad46ccf2012-12-03 16:30:42 -05001198 goto err_delete;
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001199 dprintk("RPC: TCP recvfrom got EAGAIN\n");
J. Bruce Fields9f9d2eb2012-08-17 21:35:24 -04001200 return 0;
J. Bruce Fieldsad46ccf2012-12-03 16:30:42 -05001201err_delete:
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001202 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1203 svsk->sk_xprt.xpt_server->sv_name, -len);
1204 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1205err_noclose:
J. Bruce Fields9f9d2eb2012-08-17 21:35:24 -04001206 return 0; /* record not complete */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207}
1208
1209/*
1210 * Send out data on TCP socket.
1211 */
Tom Tucker0f0257e2007-12-30 21:08:27 -06001212static int svc_tcp_sendto(struct svc_rqst *rqstp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213{
1214 struct xdr_buf *xbufp = &rqstp->rq_res;
1215 int sent;
Alexey Dobriyand8ed0292006-09-26 22:29:38 -07001216 __be32 reclen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217
1218 /* Set up the first element of the reply kvec.
1219 * Any other kvecs that may be in use have been taken
1220 * care of by the server implementation itself.
1221 */
1222 reclen = htonl(0x80000000|((xbufp->len ) - 4));
1223 memcpy(xbufp->head[0].iov_base, &reclen, 4);
1224
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225 sent = svc_sendto(rqstp, &rqstp->rq_res);
1226 if (sent != xbufp->len) {
Tom Tucker0f0257e2007-12-30 21:08:27 -06001227 printk(KERN_NOTICE
1228 "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1229 "- shutting down socket\n",
Tom Tucker57b1d3b2007-12-30 21:08:22 -06001230 rqstp->rq_xprt->xpt_server->sv_name,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 (sent<0)?"got error":"sent only",
1232 sent, xbufp->len);
Tom Tucker57b1d3b2007-12-30 21:08:22 -06001233 set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
Tom Tuckerf6150c32007-12-30 21:07:57 -06001234 svc_xprt_enqueue(rqstp->rq_xprt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 sent = -EAGAIN;
1236 }
1237 return sent;
1238}
1239
Tom Tuckere831fe62007-12-30 21:07:29 -06001240/*
1241 * Setup response header. TCP has a 4B record length field.
1242 */
1243static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
1244{
1245 struct kvec *resv = &rqstp->rq_res.head[0];
1246
1247 /* tcp needs a space for the record length... */
1248 svc_putnl(resv, 0);
1249}
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,
Chuck Lever16e4d932014-05-19 13:40:22 -04001282 .xpo_secure_port = svc_sock_secure_port,
Andy Adamson1f11a032011-01-06 02:04:26 +00001283};
1284
1285static struct svc_xprt_class svc_tcp_bc_class = {
1286 .xcl_name = "tcp-bc",
1287 .xcl_owner = THIS_MODULE,
1288 .xcl_ops = &svc_tcp_bc_ops,
1289 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1290};
Andy Adamson16b2d1e2011-01-06 02:04:27 +00001291
1292static void svc_init_bc_xprt_sock(void)
1293{
1294 svc_reg_xprt_class(&svc_tcp_bc_class);
1295}
1296
1297static void svc_cleanup_bc_xprt_sock(void)
1298{
1299 svc_unreg_xprt_class(&svc_tcp_bc_class);
1300}
Trond Myklebust9e00abc2011-07-13 19:20:49 -04001301#else /* CONFIG_SUNRPC_BACKCHANNEL */
Andy Adamson16b2d1e2011-01-06 02:04:27 +00001302static void svc_init_bc_xprt_sock(void)
1303{
1304}
1305
1306static void svc_cleanup_bc_xprt_sock(void)
1307{
1308}
Trond Myklebust9e00abc2011-07-13 19:20:49 -04001309#endif /* CONFIG_SUNRPC_BACKCHANNEL */
Andy Adamson1f11a032011-01-06 02:04:26 +00001310
Tom Tucker360d87382007-12-30 21:07:17 -06001311static struct svc_xprt_ops svc_tcp_ops = {
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001312 .xpo_create = svc_tcp_create,
Tom Tucker5d137992007-12-30 21:07:23 -06001313 .xpo_recvfrom = svc_tcp_recvfrom,
1314 .xpo_sendto = svc_tcp_sendto,
Tom Tucker5148bf42007-12-30 21:07:25 -06001315 .xpo_release_rqst = svc_release_skb,
Trond Myklebust69b6ba32008-12-23 16:30:11 -05001316 .xpo_detach = svc_tcp_sock_detach,
Tom Tucker755ccea2007-12-30 21:07:27 -06001317 .xpo_free = svc_sock_free,
Tom Tuckere831fe62007-12-30 21:07:29 -06001318 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
Tom Tucker323bee32007-12-30 21:07:31 -06001319 .xpo_has_wspace = svc_tcp_has_wspace,
Tom Tucker38a417c2007-12-30 21:07:36 -06001320 .xpo_accept = svc_tcp_accept,
Chuck Lever16e4d932014-05-19 13:40:22 -04001321 .xpo_secure_port = svc_sock_secure_port,
Trond Myklebust51877682014-07-24 23:59:33 -04001322 .xpo_adjust_wspace = svc_tcp_adjust_wspace,
Tom Tucker360d87382007-12-30 21:07:17 -06001323};
1324
1325static struct svc_xprt_class svc_tcp_class = {
1326 .xcl_name = "tcp",
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001327 .xcl_owner = THIS_MODULE,
Tom Tucker360d87382007-12-30 21:07:17 -06001328 .xcl_ops = &svc_tcp_ops,
Tom Tucker49023152007-12-30 21:07:21 -06001329 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
Chuck Lever3c45ddf2014-07-16 15:38:32 -04001330 .xcl_ident = XPRT_TRANSPORT_TCP,
Tom Tucker360d87382007-12-30 21:07:17 -06001331};
1332
1333void svc_init_xprt_sock(void)
1334{
1335 svc_reg_xprt_class(&svc_tcp_class);
1336 svc_reg_xprt_class(&svc_udp_class);
Andy Adamson16b2d1e2011-01-06 02:04:27 +00001337 svc_init_bc_xprt_sock();
Tom Tucker360d87382007-12-30 21:07:17 -06001338}
1339
1340void svc_cleanup_xprt_sock(void)
1341{
1342 svc_unreg_xprt_class(&svc_tcp_class);
1343 svc_unreg_xprt_class(&svc_udp_class);
Andy Adamson16b2d1e2011-01-06 02:04:27 +00001344 svc_cleanup_bc_xprt_sock();
Tom Tucker360d87382007-12-30 21:07:17 -06001345}
1346
Tom Tuckerbb5cf162007-12-30 21:07:50 -06001347static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348{
1349 struct sock *sk = svsk->sk_sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350
Stanislav Kinsburskybd4620d2011-12-06 14:19:10 +03001351 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1352 &svsk->sk_xprt, serv);
Tom Tuckerdef13d72007-12-30 21:08:08 -06001353 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 if (sk->sk_state == TCP_LISTEN) {
1355 dprintk("setting up TCP socket for listening\n");
Tom Tucker02fc6c32007-12-30 21:07:48 -06001356 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357 sk->sk_data_ready = svc_tcp_listen_data_ready;
Tom Tucker02fc6c32007-12-30 21:07:48 -06001358 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359 } else {
1360 dprintk("setting up TCP socket for reading\n");
1361 sk->sk_state_change = svc_tcp_state_change;
1362 sk->sk_data_ready = svc_tcp_data_ready;
Trond Myklebust47fcb032009-05-18 17:47:56 -04001363 sk->sk_write_space = svc_tcp_write_space;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364
1365 svsk->sk_reclen = 0;
1366 svsk->sk_tcplen = 0;
J. Bruce Fields8af345f2012-12-03 16:45:35 -05001367 svsk->sk_datalen = 0;
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001368 memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369
Chuck Leverb7872fe2008-04-14 12:27:01 -04001370 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371
Tom Tucker02fc6c32007-12-30 21:07:48 -06001372 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -08001373 if (sk->sk_state != TCP_ESTABLISHED)
Tom Tucker02fc6c32007-12-30 21:07:48 -06001374 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375 }
1376}
1377
Tom Tucker0f0257e2007-12-30 21:08:27 -06001378void svc_sock_update_bufs(struct svc_serv *serv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379{
1380 /*
1381 * The number of server threads has changed. Update
1382 * rcvbuf and sndbuf accordingly on all sockets
1383 */
Pavel Emelyanov8f3a6de2010-10-05 23:30:19 +04001384 struct svc_sock *svsk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385
1386 spin_lock_bh(&serv->sv_lock);
Pavel Emelyanov8f3a6de2010-10-05 23:30:19 +04001387 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
Tom Tucker02fc6c32007-12-30 21:07:48 -06001388 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389 spin_unlock_bh(&serv->sv_lock);
1390}
Trond Myklebust24c37672008-12-23 16:30:12 -05001391EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392
1393/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394 * Initialize socket for RPC use and create svc_sock struct
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 */
Chuck Lever6b174332007-02-12 00:53:28 -08001396static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1397 struct socket *sock,
J. Bruce Fields72c35372012-08-13 17:46:17 -04001398 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399{
1400 struct svc_sock *svsk;
1401 struct sock *inet;
Chuck Lever6b174332007-02-12 00:53:28 -08001402 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
J. Bruce Fields72c35372012-08-13 17:46:17 -04001403 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404
1405 dprintk("svc: svc_setup_socket %p\n", sock);
J. Bruce Fields72c35372012-08-13 17:46:17 -04001406 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1407 if (!svsk)
1408 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409
1410 inet = sock->sk;
1411
1412 /* Register socket with portmapper */
J. Bruce Fields72c35372012-08-13 17:46:17 -04001413 if (pmap_register)
1414 err = svc_register(serv, sock_net(sock->sk), inet->sk_family,
Stanislav Kinsbursky5247fab2012-01-13 14:02:48 +04001415 inet->sk_protocol,
Eric Dumazetc720c7e2009-10-15 06:30:45 +00001416 ntohs(inet_sk(inet)->inet_sport));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417
J. Bruce Fields72c35372012-08-13 17:46:17 -04001418 if (err < 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 kfree(svsk);
J. Bruce Fields72c35372012-08-13 17:46:17 -04001420 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 }
1422
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 inet->sk_user_data = svsk;
1424 svsk->sk_sock = sock;
1425 svsk->sk_sk = inet;
1426 svsk->sk_ostate = inet->sk_state_change;
1427 svsk->sk_odata = inet->sk_data_ready;
1428 svsk->sk_owspace = inet->sk_write_space;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429
1430 /* Initialize the socket */
1431 if (sock->type == SOCK_DGRAM)
Tom Tuckerbb5cf162007-12-30 21:07:50 -06001432 svc_udp_init(svsk, serv);
Olga Kornievskaia96604392008-10-21 14:13:47 -04001433 else {
1434 /* initialise setting must have enough space to
1435 * receive and respond to one request.
1436 */
1437 svc_sock_setbufsize(svsk->sk_sock, 4 * serv->sv_max_mesg,
1438 4 * serv->sv_max_mesg);
Tom Tuckerbb5cf162007-12-30 21:07:50 -06001439 svc_tcp_init(svsk, serv);
Olga Kornievskaia96604392008-10-21 14:13:47 -04001440 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442 dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1443 svsk, svsk->sk_sk);
1444
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445 return svsk;
1446}
1447
Stanislav Kinsbursky30646392014-02-26 16:50:01 +03001448bool svc_alien_sock(struct net *net, int fd)
1449{
1450 int err;
1451 struct socket *sock = sockfd_lookup(fd, &err);
1452 bool ret = false;
1453
1454 if (!sock)
1455 goto out;
1456 if (sock_net(sock->sk) != net)
1457 ret = true;
1458 sockfd_put(sock);
1459out:
1460 return ret;
1461}
1462EXPORT_SYMBOL_GPL(svc_alien_sock);
1463
Chuck Leverbfba9ab2009-04-23 19:32:33 -04001464/**
1465 * svc_addsock - add a listener socket to an RPC service
1466 * @serv: pointer to RPC service to which to add a new listener
1467 * @fd: file descriptor of the new listener
1468 * @name_return: pointer to buffer to fill in with name of listener
1469 * @len: size of the buffer
1470 *
1471 * Fills in socket name and returns positive length of name if successful.
1472 * Name is terminated with '\n'. On error, returns a negative errno
1473 * value.
1474 */
1475int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1476 const size_t len)
NeilBrownb41b66d2006-10-02 02:17:48 -07001477{
1478 int err = 0;
1479 struct socket *so = sockfd_lookup(fd, &err);
1480 struct svc_sock *svsk = NULL;
J. Bruce Fieldsa8e10072012-08-13 18:01:03 -04001481 struct sockaddr_storage addr;
1482 struct sockaddr *sin = (struct sockaddr *)&addr;
1483 int salen;
NeilBrownb41b66d2006-10-02 02:17:48 -07001484
1485 if (!so)
1486 return err;
J. Bruce Fieldsa8e10072012-08-13 18:01:03 -04001487 err = -EAFNOSUPPORT;
Aime Le Rouzic205ba422010-01-26 14:03:56 -05001488 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
J. Bruce Fieldsa8e10072012-08-13 18:01:03 -04001489 goto out;
1490 err = -EPROTONOSUPPORT;
1491 if (so->sk->sk_protocol != IPPROTO_TCP &&
NeilBrownb41b66d2006-10-02 02:17:48 -07001492 so->sk->sk_protocol != IPPROTO_UDP)
J. Bruce Fieldsa8e10072012-08-13 18:01:03 -04001493 goto out;
1494 err = -EISCONN;
1495 if (so->state > SS_UNCONNECTED)
1496 goto out;
1497 err = -ENOENT;
1498 if (!try_module_get(THIS_MODULE))
1499 goto out;
1500 svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS);
1501 if (IS_ERR(svsk)) {
1502 module_put(THIS_MODULE);
1503 err = PTR_ERR(svsk);
1504 goto out;
NeilBrownb41b66d2006-10-02 02:17:48 -07001505 }
J. Bruce Fieldsa8e10072012-08-13 18:01:03 -04001506 if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0)
1507 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
J. Bruce Fields39b55302012-08-14 15:50:34 -04001508 svc_add_new_perm_xprt(serv, &svsk->sk_xprt);
Chuck Levere7942b92009-04-23 19:32:48 -04001509 return svc_one_sock_name(svsk, name_return, len);
J. Bruce Fieldsa8e10072012-08-13 18:01:03 -04001510out:
1511 sockfd_put(so);
1512 return err;
NeilBrownb41b66d2006-10-02 02:17:48 -07001513}
1514EXPORT_SYMBOL_GPL(svc_addsock);
1515
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516/*
1517 * Create socket for RPC service.
1518 */
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001519static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1520 int protocol,
Pavel Emelyanov62832c02010-09-29 16:04:18 +04001521 struct net *net,
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001522 struct sockaddr *sin, int len,
1523 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524{
1525 struct svc_sock *svsk;
1526 struct socket *sock;
1527 int error;
1528 int type;
Tom Tucker9dbc2402007-12-30 21:08:12 -06001529 struct sockaddr_storage addr;
1530 struct sockaddr *newsin = (struct sockaddr *)&addr;
1531 int newlen;
Trond Myklebustc69da772009-03-30 18:59:17 -04001532 int family;
1533 int val;
Pavel Emelyanov5216a8e2008-02-21 10:57:45 +03001534 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535
Chuck Leverad06e4b2007-02-12 00:53:32 -08001536 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1537 serv->sv_program->pg_name, protocol,
Chuck Lever77f1f672007-02-12 00:53:39 -08001538 __svc_print_addr(sin, buf, sizeof(buf)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539
1540 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1541 printk(KERN_WARNING "svc: only UDP and TCP "
1542 "sockets supported\n");
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001543 return ERR_PTR(-EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545
Trond Myklebustc69da772009-03-30 18:59:17 -04001546 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1547 switch (sin->sa_family) {
1548 case AF_INET6:
1549 family = PF_INET6;
1550 break;
1551 case AF_INET:
1552 family = PF_INET;
1553 break;
1554 default:
1555 return ERR_PTR(-EINVAL);
1556 }
1557
Pavel Emelyanov14ec63c2010-09-29 16:06:57 +04001558 error = __sock_create(net, family, type, protocol, &sock, 1);
Chuck Lever77f1f672007-02-12 00:53:39 -08001559 if (error < 0)
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001560 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561
Peter Zijlstraed075362006-12-06 20:35:24 -08001562 svc_reclassify_socket(sock);
1563
Trond Myklebustc69da772009-03-30 18:59:17 -04001564 /*
1565 * If this is an PF_INET6 listener, we want to avoid
1566 * getting requests from IPv4 remotes. Those should
1567 * be shunted to a PF_INET listener via rpcbind.
1568 */
1569 val = 1;
1570 if (family == PF_INET6)
1571 kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
1572 (char *)&val, sizeof(val));
1573
Eric Sesterhenn18114742006-09-28 14:37:07 -07001574 if (type == SOCK_STREAM)
Pavel Emelyanov4a17fd52012-04-19 03:39:36 +00001575 sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */
Chuck Lever77f1f672007-02-12 00:53:39 -08001576 error = kernel_bind(sock, sin, len);
Eric Sesterhenn18114742006-09-28 14:37:07 -07001577 if (error < 0)
1578 goto bummer;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579
Tom Tucker9dbc2402007-12-30 21:08:12 -06001580 newlen = len;
1581 error = kernel_getsockname(sock, newsin, &newlen);
1582 if (error < 0)
1583 goto bummer;
1584
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585 if (protocol == IPPROTO_TCP) {
Sridhar Samudralae6242e92006-08-07 20:58:01 -07001586 if ((error = kernel_listen(sock, 64)) < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587 goto bummer;
1588 }
1589
J. Bruce Fields72c35372012-08-13 17:46:17 -04001590 svsk = svc_setup_socket(serv, sock, flags);
J. Bruce Fieldsa8e10072012-08-13 18:01:03 -04001591 if (IS_ERR(svsk)) {
1592 error = PTR_ERR(svsk);
1593 goto bummer;
NeilBrowne79eff12007-02-12 00:53:30 -08001594 }
J. Bruce Fieldsa8e10072012-08-13 18:01:03 -04001595 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1596 return (struct svc_xprt *)svsk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597bummer:
1598 dprintk("svc: svc_create_socket error = %d\n", -error);
1599 sock_release(sock);
Tom Tuckerb700cbb2007-12-30 21:07:42 -06001600 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601}
1602
1603/*
Tom Tucker755ccea2007-12-30 21:07:27 -06001604 * Detach the svc_sock from the socket so that no
1605 * more callbacks occur.
1606 */
1607static void svc_sock_detach(struct svc_xprt *xprt)
1608{
1609 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1610 struct sock *sk = svsk->sk_sk;
Eric Dumazeteaefd112011-02-18 03:26:36 +00001611 wait_queue_head_t *wq;
Tom Tucker755ccea2007-12-30 21:07:27 -06001612
1613 dprintk("svc: svc_sock_detach(%p)\n", svsk);
1614
1615 /* put back the old socket callbacks */
1616 sk->sk_state_change = svsk->sk_ostate;
1617 sk->sk_data_ready = svsk->sk_odata;
1618 sk->sk_write_space = svsk->sk_owspace;
Trond Myklebust69b6ba32008-12-23 16:30:11 -05001619
Eric Dumazeteaefd112011-02-18 03:26:36 +00001620 wq = sk_sleep(sk);
J. Bruce Fields0442f142015-10-09 01:44:07 +00001621 if (sunrpc_waitqueue_active(wq))
Eric Dumazeteaefd112011-02-18 03:26:36 +00001622 wake_up_interruptible(wq);
Trond Myklebust69b6ba32008-12-23 16:30:11 -05001623}
1624
1625/*
1626 * Disconnect the socket, and reset the callbacks
1627 */
1628static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1629{
1630 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1631
1632 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
1633
1634 svc_sock_detach(xprt);
1635
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001636 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1637 svc_tcp_clear_pages(svsk);
Trond Myklebust69b6ba32008-12-23 16:30:11 -05001638 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
J. Bruce Fields31d68ef2011-02-24 11:25:33 -08001639 }
Tom Tucker755ccea2007-12-30 21:07:27 -06001640}
1641
1642/*
1643 * Free the svc_sock's socket resources and the svc_sock itself.
1644 */
1645static void svc_sock_free(struct svc_xprt *xprt)
1646{
1647 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1648 dprintk("svc: svc_sock_free(%p)\n", svsk);
1649
Tom Tucker755ccea2007-12-30 21:07:27 -06001650 if (svsk->sk_sock->file)
1651 sockfd_put(svsk->sk_sock);
1652 else
1653 sock_release(svsk->sk_sock);
1654 kfree(svsk);
1655}
Benny Halevy7652e5a2009-04-01 09:23:09 -04001656
Trond Myklebust9e00abc2011-07-13 19:20:49 -04001657#if defined(CONFIG_SUNRPC_BACKCHANNEL)
Benny Halevy7652e5a2009-04-01 09:23:09 -04001658/*
Andy Adamson1f11a032011-01-06 02:04:26 +00001659 * Create a back channel svc_xprt which shares the fore channel socket.
Benny Halevy7652e5a2009-04-01 09:23:09 -04001660 */
Andy Adamson1f11a032011-01-06 02:04:26 +00001661static struct svc_xprt *svc_bc_create_socket(struct svc_serv *serv,
1662 int protocol,
1663 struct net *net,
1664 struct sockaddr *sin, int len,
1665 int flags)
Benny Halevy7652e5a2009-04-01 09:23:09 -04001666{
1667 struct svc_sock *svsk;
Andy Adamson1f11a032011-01-06 02:04:26 +00001668 struct svc_xprt *xprt;
Benny Halevy7652e5a2009-04-01 09:23:09 -04001669
Andy Adamson1f11a032011-01-06 02:04:26 +00001670 if (protocol != IPPROTO_TCP) {
1671 printk(KERN_WARNING "svc: only TCP sockets"
1672 " supported on shared back channel\n");
1673 return ERR_PTR(-EINVAL);
1674 }
1675
Benny Halevy7652e5a2009-04-01 09:23:09 -04001676 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1677 if (!svsk)
Andy Adamson1f11a032011-01-06 02:04:26 +00001678 return ERR_PTR(-ENOMEM);
Benny Halevy7652e5a2009-04-01 09:23:09 -04001679
1680 xprt = &svsk->sk_xprt;
Stanislav Kinsburskybd4620d2011-12-06 14:19:10 +03001681 svc_xprt_init(net, &svc_tcp_bc_class, xprt, serv);
Andy Adamson1f11a032011-01-06 02:04:26 +00001682
Andy Adamson4a19de02011-01-06 02:04:35 +00001683 serv->sv_bc_xprt = xprt;
Andy Adamson1f11a032011-01-06 02:04:26 +00001684
Benny Halevy7652e5a2009-04-01 09:23:09 -04001685 return xprt;
1686}
Benny Halevy7652e5a2009-04-01 09:23:09 -04001687
1688/*
Andy Adamson1f11a032011-01-06 02:04:26 +00001689 * Free a back channel svc_sock.
Benny Halevy7652e5a2009-04-01 09:23:09 -04001690 */
Andy Adamson1f11a032011-01-06 02:04:26 +00001691static void svc_bc_sock_free(struct svc_xprt *xprt)
Benny Halevy7652e5a2009-04-01 09:23:09 -04001692{
Andy Adamson778be232011-01-25 15:38:01 +00001693 if (xprt)
Benny Halevy7652e5a2009-04-01 09:23:09 -04001694 kfree(container_of(xprt, struct svc_sock, sk_xprt));
1695}
Trond Myklebust9e00abc2011-07-13 19:20:49 -04001696#endif /* CONFIG_SUNRPC_BACKCHANNEL */