Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * linux/net/sunrpc/svcsock.c | |
4 | * | |
5 | * These are the RPC server socket internals. | |
6 | * | |
7 | * The server scheduling algorithm does not always distribute the load | |
8 | * evenly when servicing a single client. May need to modify the | |
f6150c3c | 9 | * svc_xprt_enqueue procedure... |
1da177e4 LT |
10 | * |
11 | * TCP support is largely untested and may be a little slow. The problem | |
12 | * is that we currently do two separate recvfrom's, one for the 4-byte | |
13 | * record length, and the second for the actual record. This could possibly | |
14 | * be improved by always reading a minimum size of around 100 bytes and | |
15 | * tucking any superfluous bytes away in a temporary store. Still, that | |
16 | * leaves write requests out in the rain. An alternative may be to peek at | |
17 | * the first skb in the queue, and if it matches the next TCP sequence | |
18 | * number, to extract the record marker. Yuck. | |
19 | * | |
20 | * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de> | |
21 | */ | |
22 | ||
172589cc | 23 | #include <linux/kernel.h> |
1da177e4 | 24 | #include <linux/sched.h> |
3a9a231d | 25 | #include <linux/module.h> |
1da177e4 LT |
26 | #include <linux/errno.h> |
27 | #include <linux/fcntl.h> | |
28 | #include <linux/net.h> | |
29 | #include <linux/in.h> | |
30 | #include <linux/inet.h> | |
31 | #include <linux/udp.h> | |
91483c4b | 32 | #include <linux/tcp.h> |
1da177e4 LT |
33 | #include <linux/unistd.h> |
34 | #include <linux/slab.h> | |
35 | #include <linux/netdevice.h> | |
36 | #include <linux/skbuff.h> | |
b41b66d6 | 37 | #include <linux/file.h> |
7dfb7103 | 38 | #include <linux/freezer.h> |
1da177e4 LT |
39 | #include <net/sock.h> |
40 | #include <net/checksum.h> | |
41 | #include <net/ip.h> | |
b92503b2 | 42 | #include <net/ipv6.h> |
850cbadd | 43 | #include <net/udp.h> |
b7872fe8 | 44 | #include <net/tcp.h> |
c752f073 | 45 | #include <net/tcp_states.h> |
7c0f6ba6 | 46 | #include <linux/uaccess.h> |
becd2014 | 47 | #include <linux/highmem.h> |
1da177e4 LT |
48 | #include <asm/ioctls.h> |
49 | ||
50 | #include <linux/sunrpc/types.h> | |
ad06e4bd | 51 | #include <linux/sunrpc/clnt.h> |
1da177e4 | 52 | #include <linux/sunrpc/xdr.h> |
c0401ea0 | 53 | #include <linux/sunrpc/msg_prot.h> |
1da177e4 LT |
54 | #include <linux/sunrpc/svcsock.h> |
55 | #include <linux/sunrpc/stats.h> | |
4cfc7e60 | 56 | #include <linux/sunrpc/xprt.h> |
1da177e4 | 57 | |
40e0b090 | 58 | #include <trace/events/sock.h> |
998024de CL |
59 | #include <trace/events/sunrpc.h> |
60 | ||
9e55eef4 | 61 | #include "socklib.h" |
177e4f99 HS |
62 | #include "sunrpc.h" |
63 | ||
360d8738 | 64 | #define RPCDBG_FACILITY RPCDBG_SVCXPRT |
1da177e4 LT |
65 | |
66 | ||
67 | static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *, | |
72c35376 | 68 | int flags); |
1da177e4 LT |
69 | static int svc_udp_recvfrom(struct svc_rqst *); |
70 | static int svc_udp_sendto(struct svc_rqst *); | |
755cceab | 71 | static void svc_sock_detach(struct svc_xprt *); |
69b6ba37 | 72 | static void svc_tcp_sock_detach(struct svc_xprt *); |
755cceab | 73 | static void svc_sock_free(struct svc_xprt *); |
1da177e4 | 74 | |
b700cbb1 | 75 | static struct svc_xprt *svc_create_socket(struct svc_serv *, int, |
62832c03 PE |
76 | struct net *, struct sockaddr *, |
77 | int, int); | |
ed07536e PZ |
78 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
79 | static struct lock_class_key svc_key[2]; | |
80 | static struct lock_class_key svc_slock_key[2]; | |
81 | ||
0f0257ea | 82 | static void svc_reclassify_socket(struct socket *sock) |
ed07536e PZ |
83 | { |
84 | struct sock *sk = sock->sk; | |
1b7a1819 | 85 | |
fafc4e1e | 86 | if (WARN_ON_ONCE(!sock_allow_reclassification(sk))) |
1b7a1819 WAA |
87 | return; |
88 | ||
ed07536e PZ |
89 | switch (sk->sk_family) { |
90 | case AF_INET: | |
91 | sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD", | |
def13d74 TT |
92 | &svc_slock_key[0], |
93 | "sk_xprt.xpt_lock-AF_INET-NFSD", | |
94 | &svc_key[0]); | |
ed07536e PZ |
95 | break; |
96 | ||
97 | case AF_INET6: | |
98 | sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD", | |
def13d74 TT |
99 | &svc_slock_key[1], |
100 | "sk_xprt.xpt_lock-AF_INET6-NFSD", | |
101 | &svc_key[1]); | |
ed07536e PZ |
102 | break; |
103 | ||
104 | default: | |
105 | BUG(); | |
106 | } | |
107 | } | |
108 | #else | |
0f0257ea | 109 | static void svc_reclassify_socket(struct socket *sock) |
ed07536e PZ |
110 | { |
111 | } | |
112 | #endif | |
113 | ||
cca557a5 CL |
114 | /** |
115 | * svc_tcp_release_rqst - Release transport-related resources | |
116 | * @rqstp: request structure with resources to be released | |
117 | * | |
1da177e4 | 118 | */ |
cca557a5 | 119 | static void svc_tcp_release_rqst(struct svc_rqst *rqstp) |
1da177e4 | 120 | { |
1da177e4 LT |
121 | } |
122 | ||
cca557a5 CL |
123 | /** |
124 | * svc_udp_release_rqst - Release transport-related resources | |
125 | * @rqstp: request structure with resources to be released | |
126 | * | |
127 | */ | |
128 | static void svc_udp_release_rqst(struct svc_rqst *rqstp) | |
850cbadd PA |
129 | { |
130 | struct sk_buff *skb = rqstp->rq_xprt_ctxt; | |
131 | ||
132 | if (skb) { | |
133 | rqstp->rq_xprt_ctxt = NULL; | |
850cbadd PA |
134 | consume_skb(skb); |
135 | } | |
136 | } | |
137 | ||
b92503b2 CL |
138 | union svc_pktinfo_u { |
139 | struct in_pktinfo pkti; | |
b92503b2 | 140 | struct in6_pktinfo pkti6; |
b92503b2 | 141 | }; |
bc375ea7 DM |
142 | #define SVC_PKTINFO_SPACE \ |
143 | CMSG_SPACE(sizeof(union svc_pktinfo_u)) | |
b92503b2 CL |
144 | |
145 | static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh) | |
146 | { | |
57b1d3ba TT |
147 | struct svc_sock *svsk = |
148 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
149 | switch (svsk->sk_sk->sk_family) { | |
b92503b2 CL |
150 | case AF_INET: { |
151 | struct in_pktinfo *pki = CMSG_DATA(cmh); | |
152 | ||
153 | cmh->cmsg_level = SOL_IP; | |
154 | cmh->cmsg_type = IP_PKTINFO; | |
155 | pki->ipi_ifindex = 0; | |
849a1cf1 MJ |
156 | pki->ipi_spec_dst.s_addr = |
157 | svc_daddr_in(rqstp)->sin_addr.s_addr; | |
b92503b2 CL |
158 | cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); |
159 | } | |
160 | break; | |
5a05ed73 | 161 | |
b92503b2 CL |
162 | case AF_INET6: { |
163 | struct in6_pktinfo *pki = CMSG_DATA(cmh); | |
849a1cf1 | 164 | struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp); |
b92503b2 CL |
165 | |
166 | cmh->cmsg_level = SOL_IPV6; | |
167 | cmh->cmsg_type = IPV6_PKTINFO; | |
849a1cf1 | 168 | pki->ipi6_ifindex = daddr->sin6_scope_id; |
4e3fd7a0 | 169 | pki->ipi6_addr = daddr->sin6_addr; |
b92503b2 CL |
170 | cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); |
171 | } | |
172 | break; | |
b92503b2 | 173 | } |
b92503b2 CL |
174 | } |
175 | ||
03493bca CL |
176 | static int svc_sock_result_payload(struct svc_rqst *rqstp, unsigned int offset, |
177 | unsigned int length) | |
41205539 CL |
178 | { |
179 | return 0; | |
180 | } | |
181 | ||
80212d59 N |
182 | /* |
183 | * Report socket names for nfsdfs | |
184 | */ | |
e7942b9f | 185 | static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining) |
80212d59 | 186 | { |
017cb47f CL |
187 | const struct sock *sk = svsk->sk_sk; |
188 | const char *proto_name = sk->sk_protocol == IPPROTO_UDP ? | |
189 | "udp" : "tcp"; | |
80212d59 N |
190 | int len; |
191 | ||
017cb47f | 192 | switch (sk->sk_family) { |
e7942b9f CL |
193 | case PF_INET: |
194 | len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n", | |
017cb47f | 195 | proto_name, |
c720c7e8 ED |
196 | &inet_sk(sk)->inet_rcv_saddr, |
197 | inet_sk(sk)->inet_num); | |
80212d59 | 198 | break; |
c2bb06db | 199 | #if IS_ENABLED(CONFIG_IPV6) |
58de2f86 CL |
200 | case PF_INET6: |
201 | len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n", | |
017cb47f | 202 | proto_name, |
efe4208f | 203 | &sk->sk_v6_rcv_saddr, |
c720c7e8 | 204 | inet_sk(sk)->inet_num); |
80212d59 | 205 | break; |
c2bb06db | 206 | #endif |
80212d59 | 207 | default: |
e7942b9f | 208 | len = snprintf(buf, remaining, "*unknown-%d*\n", |
017cb47f | 209 | sk->sk_family); |
80212d59 | 210 | } |
e7942b9f CL |
211 | |
212 | if (len >= remaining) { | |
213 | *buf = '\0'; | |
214 | return -ENAMETOOLONG; | |
80212d59 N |
215 | } |
216 | return len; | |
217 | } | |
218 | ||
ca07eda3 CL |
219 | #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE |
220 | static void svc_flush_bvec(const struct bio_vec *bvec, size_t size, size_t seek) | |
221 | { | |
222 | struct bvec_iter bi = { | |
13a9a9d7 | 223 | .bi_size = size + seek, |
ca07eda3 CL |
224 | }; |
225 | struct bio_vec bv; | |
226 | ||
227 | bvec_iter_advance(bvec, &bi, seek & PAGE_MASK); | |
228 | for_each_bvec(bv, bvec, bi, bi) | |
229 | flush_dcache_page(bv.bv_page); | |
230 | } | |
231 | #else | |
232 | static inline void svc_flush_bvec(const struct bio_vec *bvec, size_t size, | |
233 | size_t seek) | |
234 | { | |
235 | } | |
236 | #endif | |
237 | ||
1da177e4 | 238 | /* |
ca07eda3 CL |
239 | * Read from @rqstp's transport socket. The incoming message fills whole |
240 | * pages in @rqstp's rq_pages array until the last page of the message | |
241 | * has been received into a partial page. | |
1da177e4 | 242 | */ |
ca07eda3 CL |
243 | static ssize_t svc_tcp_read_msg(struct svc_rqst *rqstp, size_t buflen, |
244 | size_t seek) | |
1da177e4 | 245 | { |
57b1d3ba TT |
246 | struct svc_sock *svsk = |
247 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
ca07eda3 | 248 | struct bio_vec *bvec = rqstp->rq_bvec; |
4c8e5537 | 249 | struct msghdr msg = { NULL }; |
ca07eda3 | 250 | unsigned int i; |
4c8e5537 | 251 | ssize_t len; |
ca07eda3 | 252 | size_t t; |
1da177e4 | 253 | |
f8d1ff47 | 254 | clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
ca07eda3 | 255 | |
9088151f CH |
256 | for (i = 0, t = 0; t < buflen; i++, t += PAGE_SIZE) |
257 | bvec_set_page(&bvec[i], rqstp->rq_pages[i], PAGE_SIZE, 0); | |
ca07eda3 CL |
258 | rqstp->rq_respages = &rqstp->rq_pages[i]; |
259 | rqstp->rq_next_page = rqstp->rq_respages + 1; | |
260 | ||
de4eda9d | 261 | iov_iter_bvec(&msg.msg_iter, ITER_DEST, bvec, i, buflen); |
ca07eda3 CL |
262 | if (seek) { |
263 | iov_iter_advance(&msg.msg_iter, seek); | |
264 | buflen -= seek; | |
4c8e5537 TM |
265 | } |
266 | len = sock_recvmsg(svsk->sk_sock, &msg, MSG_DONTWAIT); | |
ca07eda3 CL |
267 | if (len > 0) |
268 | svc_flush_bvec(bvec, len, seek); | |
269 | ||
f8d1ff47 TM |
270 | /* If we read a full record, then assume there may be more |
271 | * data to read (stream based sockets only!) | |
272 | */ | |
273 | if (len == buflen) | |
274 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); | |
1da177e4 | 275 | |
1da177e4 LT |
276 | return len; |
277 | } | |
278 | ||
279 | /* | |
280 | * Set socket snd and rcv buffer lengths | |
281 | */ | |
b7e5034c | 282 | static void svc_sock_setbufsize(struct svc_sock *svsk, unsigned int nreqs) |
1da177e4 | 283 | { |
b7e5034c BF |
284 | unsigned int max_mesg = svsk->sk_xprt.xpt_server->sv_max_mesg; |
285 | struct socket *sock = svsk->sk_sock; | |
286 | ||
287 | nreqs = min(nreqs, INT_MAX / 2 / max_mesg); | |
288 | ||
1da177e4 | 289 | lock_sock(sock->sk); |
b7e5034c BF |
290 | sock->sk->sk_sndbuf = nreqs * max_mesg * 2; |
291 | sock->sk->sk_rcvbuf = nreqs * max_mesg * 2; | |
47fcb03f | 292 | sock->sk->sk_write_space(sock->sk); |
1da177e4 | 293 | release_sock(sock->sk); |
1da177e4 | 294 | } |
16e4d93f | 295 | |
989f881e | 296 | static void svc_sock_secure_port(struct svc_rqst *rqstp) |
16e4d93f | 297 | { |
989f881e | 298 | if (svc_port_is_privileged(svc_addr(rqstp))) |
7827c81f | 299 | set_bit(RQ_SECURE, &rqstp->rq_flags); |
989f881e | 300 | else |
7827c81f | 301 | clear_bit(RQ_SECURE, &rqstp->rq_flags); |
16e4d93f CL |
302 | } |
303 | ||
1da177e4 LT |
304 | /* |
305 | * INET callback when data has been received on the socket. | |
306 | */ | |
fa9251af | 307 | static void svc_data_ready(struct sock *sk) |
1da177e4 | 308 | { |
939bb7ef | 309 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 | 310 | |
40e0b090 PY |
311 | trace_sk_data_ready(sk); |
312 | ||
939bb7ef | 313 | if (svsk) { |
eebe53e8 VL |
314 | /* Refer to svc_setup_socket() for details. */ |
315 | rmb(); | |
fa9251af | 316 | svsk->sk_odata(sk); |
998024de | 317 | trace_svcsock_data_ready(&svsk->sk_xprt, 0); |
4720b070 TM |
318 | if (!test_and_set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags)) |
319 | svc_xprt_enqueue(&svsk->sk_xprt); | |
939bb7ef | 320 | } |
1da177e4 LT |
321 | } |
322 | ||
323 | /* | |
324 | * INET callback when space is newly available on the socket. | |
325 | */ | |
0f0257ea | 326 | static void svc_write_space(struct sock *sk) |
1da177e4 LT |
327 | { |
328 | struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data); | |
329 | ||
330 | if (svsk) { | |
eebe53e8 VL |
331 | /* Refer to svc_setup_socket() for details. */ |
332 | rmb(); | |
998024de | 333 | trace_svcsock_write_space(&svsk->sk_xprt, 0); |
fa9251af | 334 | svsk->sk_owspace(sk); |
f6150c3c | 335 | svc_xprt_enqueue(&svsk->sk_xprt); |
1da177e4 | 336 | } |
1da177e4 LT |
337 | } |
338 | ||
c7fb3f06 TM |
339 | static int svc_tcp_has_wspace(struct svc_xprt *xprt) |
340 | { | |
637600f3 | 341 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); |
c7fb3f06 TM |
342 | |
343 | if (test_bit(XPT_LISTENER, &xprt->xpt_flags)) | |
344 | return 1; | |
637600f3 | 345 | return !test_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); |
51877680 TM |
346 | } |
347 | ||
ea08e392 SM |
348 | static void svc_tcp_kill_temp_xprt(struct svc_xprt *xprt) |
349 | { | |
c433594c CH |
350 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); |
351 | ||
352 | sock_no_linger(svsk->sk_sock->sk); | |
ea08e392 SM |
353 | } |
354 | ||
7702ce40 CL |
355 | /* |
356 | * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo | |
357 | */ | |
358 | static int svc_udp_get_dest_address4(struct svc_rqst *rqstp, | |
359 | struct cmsghdr *cmh) | |
360 | { | |
361 | struct in_pktinfo *pki = CMSG_DATA(cmh); | |
849a1cf1 MJ |
362 | struct sockaddr_in *daddr = svc_daddr_in(rqstp); |
363 | ||
7702ce40 CL |
364 | if (cmh->cmsg_type != IP_PKTINFO) |
365 | return 0; | |
849a1cf1 MJ |
366 | |
367 | daddr->sin_family = AF_INET; | |
368 | daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr; | |
7702ce40 CL |
369 | return 1; |
370 | } | |
371 | ||
372 | /* | |
73df66f8 | 373 | * See net/ipv6/datagram.c : ip6_datagram_recv_ctl |
7702ce40 CL |
374 | */ |
375 | static int svc_udp_get_dest_address6(struct svc_rqst *rqstp, | |
376 | struct cmsghdr *cmh) | |
377 | { | |
378 | struct in6_pktinfo *pki = CMSG_DATA(cmh); | |
849a1cf1 MJ |
379 | struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp); |
380 | ||
7702ce40 CL |
381 | if (cmh->cmsg_type != IPV6_PKTINFO) |
382 | return 0; | |
849a1cf1 MJ |
383 | |
384 | daddr->sin6_family = AF_INET6; | |
4e3fd7a0 | 385 | daddr->sin6_addr = pki->ipi6_addr; |
849a1cf1 | 386 | daddr->sin6_scope_id = pki->ipi6_ifindex; |
7702ce40 CL |
387 | return 1; |
388 | } | |
389 | ||
9dbc240f TT |
390 | /* |
391 | * Copy the UDP datagram's destination address to the rqstp structure. | |
392 | * The 'destination' address in this case is the address to which the | |
393 | * peer sent the datagram, i.e. our local address. For multihomed | |
394 | * hosts, this can change from msg to msg. Note that only the IP | |
395 | * address changes, the port number should remain the same. | |
396 | */ | |
7702ce40 CL |
397 | static int svc_udp_get_dest_address(struct svc_rqst *rqstp, |
398 | struct cmsghdr *cmh) | |
95756482 | 399 | { |
7702ce40 CL |
400 | switch (cmh->cmsg_level) { |
401 | case SOL_IP: | |
402 | return svc_udp_get_dest_address4(rqstp, cmh); | |
403 | case SOL_IPV6: | |
404 | return svc_udp_get_dest_address6(rqstp, cmh); | |
95756482 | 405 | } |
7702ce40 CL |
406 | |
407 | return 0; | |
95756482 CL |
408 | } |
409 | ||
fff1ebb2 CL |
410 | /** |
411 | * svc_udp_recvfrom - Receive a datagram from a UDP socket. | |
412 | * @rqstp: request structure into which to receive an RPC Call | |
413 | * | |
414 | * Called in a loop when XPT_DATA has been set. | |
415 | * | |
416 | * Returns: | |
417 | * On success, the number of bytes in a received RPC Call, or | |
418 | * %0 if a complete RPC Call message was not ready to return | |
1da177e4 | 419 | */ |
0f0257ea | 420 | static int svc_udp_recvfrom(struct svc_rqst *rqstp) |
1da177e4 | 421 | { |
57b1d3ba TT |
422 | struct svc_sock *svsk = |
423 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
bb5cf160 | 424 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
1da177e4 | 425 | struct sk_buff *skb; |
bc375ea7 DM |
426 | union { |
427 | struct cmsghdr hdr; | |
428 | long all[SVC_PKTINFO_SPACE / sizeof(long)]; | |
429 | } buffer; | |
430 | struct cmsghdr *cmh = &buffer.hdr; | |
7a37f578 N |
431 | struct msghdr msg = { |
432 | .msg_name = svc_addr(rqstp), | |
433 | .msg_control = cmh, | |
434 | .msg_controllen = sizeof(buffer), | |
435 | .msg_flags = MSG_DONTWAIT, | |
436 | }; | |
abc5c44d CL |
437 | size_t len; |
438 | int err; | |
1da177e4 | 439 | |
02fc6c36 | 440 | if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags)) |
1da177e4 LT |
441 | /* udp sockets need large rcvbuf as all pending |
442 | * requests are still in that buffer. sndbuf must | |
443 | * also be large enough that there is enough space | |
3262c816 GB |
444 | * for one reply per thread. We count all threads |
445 | * rather than threads in a particular pool, which | |
446 | * provides an upper bound on the number of threads | |
447 | * which will access the socket. | |
1da177e4 | 448 | */ |
b7e5034c | 449 | svc_sock_setbufsize(svsk, serv->sv_nrthreads + 3); |
1da177e4 | 450 | |
02fc6c36 | 451 | clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
05ed690e N |
452 | err = kernel_recvmsg(svsk->sk_sock, &msg, NULL, |
453 | 0, 0, MSG_PEEK | MSG_DONTWAIT); | |
fff1ebb2 CL |
454 | if (err < 0) |
455 | goto out_recv_err; | |
ec095263 | 456 | skb = skb_recv_udp(svsk->sk_sk, MSG_DONTWAIT, &err); |
fff1ebb2 CL |
457 | if (!skb) |
458 | goto out_recv_err; | |
459 | ||
9dbc240f | 460 | len = svc_addr_len(svc_addr(rqstp)); |
9dbc240f | 461 | rqstp->rq_addrlen = len; |
2456e855 | 462 | if (skb->tstamp == 0) { |
b7aa0bf7 | 463 | skb->tstamp = ktime_get_real(); |
cca5172a | 464 | /* Don't enable netstamp, sunrpc doesn't |
1da177e4 LT |
465 | need that much accuracy */ |
466 | } | |
3a0ed3e9 | 467 | sock_write_timestamp(svsk->sk_sk, skb->tstamp); |
02fc6c36 | 468 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */ |
1da177e4 | 469 | |
fff1ebb2 | 470 | len = skb->len; |
1da177e4 | 471 | rqstp->rq_arg.len = len; |
fff1ebb2 | 472 | trace_svcsock_udp_recv(&svsk->sk_xprt, len); |
1da177e4 | 473 | |
95756482 | 474 | rqstp->rq_prot = IPPROTO_UDP; |
27459f09 | 475 | |
fff1ebb2 CL |
476 | if (!svc_udp_get_dest_address(rqstp, cmh)) |
477 | goto out_cmsg_err; | |
849a1cf1 | 478 | rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp)); |
1da177e4 LT |
479 | |
480 | if (skb_is_nonlinear(skb)) { | |
481 | /* we have to copy */ | |
482 | local_bh_disable(); | |
fff1ebb2 CL |
483 | if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) |
484 | goto out_bh_enable; | |
1da177e4 | 485 | local_bh_enable(); |
850cbadd | 486 | consume_skb(skb); |
1da177e4 LT |
487 | } else { |
488 | /* we can use it in-place */ | |
1da8c681 | 489 | rqstp->rq_arg.head[0].iov_base = skb->data; |
1da177e4 | 490 | rqstp->rq_arg.head[0].iov_len = len; |
22911fc5 ED |
491 | if (skb_checksum_complete(skb)) |
492 | goto out_free; | |
5148bf4e | 493 | rqstp->rq_xprt_ctxt = skb; |
1da177e4 LT |
494 | } |
495 | ||
496 | rqstp->rq_arg.page_base = 0; | |
497 | if (len <= rqstp->rq_arg.head[0].iov_len) { | |
498 | rqstp->rq_arg.head[0].iov_len = len; | |
499 | rqstp->rq_arg.page_len = 0; | |
44524359 | 500 | rqstp->rq_respages = rqstp->rq_pages+1; |
1da177e4 LT |
501 | } else { |
502 | rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len; | |
44524359 | 503 | rqstp->rq_respages = rqstp->rq_pages + 1 + |
172589cc | 504 | DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE); |
1da177e4 | 505 | } |
afc59400 | 506 | rqstp->rq_next_page = rqstp->rq_respages+1; |
1da177e4 LT |
507 | |
508 | if (serv->sv_stats) | |
509 | serv->sv_stats->netudpcnt++; | |
510 | ||
82011c80 | 511 | svc_xprt_received(rqstp->rq_xprt); |
1da177e4 | 512 | return len; |
fff1ebb2 CL |
513 | |
514 | out_recv_err: | |
515 | if (err != -EAGAIN) { | |
516 | /* possibly an icmp error */ | |
517 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); | |
518 | } | |
519 | trace_svcsock_udp_recv_err(&svsk->sk_xprt, err); | |
82011c80 | 520 | goto out_clear_busy; |
fff1ebb2 CL |
521 | out_cmsg_err: |
522 | net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n", | |
523 | cmh->cmsg_level, cmh->cmsg_type); | |
524 | goto out_free; | |
525 | out_bh_enable: | |
526 | local_bh_enable(); | |
f23abfdb | 527 | out_free: |
850cbadd | 528 | kfree_skb(skb); |
82011c80 CL |
529 | out_clear_busy: |
530 | svc_xprt_received(rqstp->rq_xprt); | |
f23abfdb | 531 | return 0; |
1da177e4 LT |
532 | } |
533 | ||
da1661b9 CL |
534 | /** |
535 | * svc_udp_sendto - Send out a reply on a UDP socket | |
536 | * @rqstp: completed svc_rqst | |
537 | * | |
ca4faf54 CL |
538 | * xpt_mutex ensures @rqstp's whole message is written to the socket |
539 | * without interruption. | |
540 | * | |
da1661b9 CL |
541 | * Returns the number of bytes sent, or a negative errno. |
542 | */ | |
543 | static int svc_udp_sendto(struct svc_rqst *rqstp) | |
1da177e4 | 544 | { |
da1661b9 CL |
545 | struct svc_xprt *xprt = rqstp->rq_xprt; |
546 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
547 | struct xdr_buf *xdr = &rqstp->rq_res; | |
548 | union { | |
549 | struct cmsghdr hdr; | |
550 | long all[SVC_PKTINFO_SPACE / sizeof(long)]; | |
551 | } buffer; | |
552 | struct cmsghdr *cmh = &buffer.hdr; | |
553 | struct msghdr msg = { | |
554 | .msg_name = &rqstp->rq_addr, | |
555 | .msg_namelen = rqstp->rq_addrlen, | |
556 | .msg_control = cmh, | |
557 | .msg_controllen = sizeof(buffer), | |
558 | }; | |
3f649ab7 | 559 | unsigned int sent; |
da1661b9 | 560 | int err; |
1da177e4 | 561 | |
cca557a5 | 562 | svc_udp_release_rqst(rqstp); |
23cf1ee1 | 563 | |
da1661b9 | 564 | svc_set_cmsg_data(rqstp, cmh); |
1da177e4 | 565 | |
ca4faf54 CL |
566 | mutex_lock(&xprt->xpt_mutex); |
567 | ||
568 | if (svc_xprt_is_dead(xprt)) | |
569 | goto out_notconn; | |
570 | ||
ff053dbb TM |
571 | err = xdr_alloc_bvec(xdr, GFP_KERNEL); |
572 | if (err < 0) | |
573 | goto out_unlock; | |
574 | ||
da1661b9 | 575 | err = xprt_sock_sendmsg(svsk->sk_sock, &msg, xdr, 0, 0, &sent); |
da1661b9 CL |
576 | if (err == -ECONNREFUSED) { |
577 | /* ICMP error on earlier request. */ | |
578 | err = xprt_sock_sendmsg(svsk->sk_sock, &msg, xdr, 0, 0, &sent); | |
da1661b9 | 579 | } |
ff053dbb | 580 | xdr_free_bvec(xdr); |
998024de | 581 | trace_svcsock_udp_send(xprt, err); |
ff053dbb | 582 | out_unlock: |
ca4faf54 | 583 | mutex_unlock(&xprt->xpt_mutex); |
da1661b9 CL |
584 | if (err < 0) |
585 | return err; | |
586 | return sent; | |
ca4faf54 CL |
587 | |
588 | out_notconn: | |
589 | mutex_unlock(&xprt->xpt_mutex); | |
590 | return -ENOTCONN; | |
1da177e4 LT |
591 | } |
592 | ||
323bee32 TT |
593 | static int svc_udp_has_wspace(struct svc_xprt *xprt) |
594 | { | |
595 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
bb5cf160 | 596 | struct svc_serv *serv = xprt->xpt_server; |
323bee32 TT |
597 | unsigned long required; |
598 | ||
599 | /* | |
600 | * Set the SOCK_NOSPACE flag before checking the available | |
601 | * sock space. | |
602 | */ | |
603 | set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
7a90e8cc | 604 | required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg; |
323bee32 TT |
605 | if (required*2 > sock_wspace(svsk->sk_sk)) |
606 | return 0; | |
607 | clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
608 | return 1; | |
609 | } | |
610 | ||
38a417cc TT |
611 | static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt) |
612 | { | |
613 | BUG(); | |
614 | return NULL; | |
615 | } | |
616 | ||
ea08e392 SM |
617 | static void svc_udp_kill_temp_xprt(struct svc_xprt *xprt) |
618 | { | |
619 | } | |
620 | ||
b700cbb1 | 621 | static struct svc_xprt *svc_udp_create(struct svc_serv *serv, |
62832c03 | 622 | struct net *net, |
b700cbb1 TT |
623 | struct sockaddr *sa, int salen, |
624 | int flags) | |
625 | { | |
62832c03 | 626 | return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags); |
b700cbb1 TT |
627 | } |
628 | ||
2412e927 | 629 | static const struct svc_xprt_ops svc_udp_ops = { |
b700cbb1 | 630 | .xpo_create = svc_udp_create, |
5d137990 TT |
631 | .xpo_recvfrom = svc_udp_recvfrom, |
632 | .xpo_sendto = svc_udp_sendto, | |
03493bca | 633 | .xpo_result_payload = svc_sock_result_payload, |
cca557a5 | 634 | .xpo_release_rqst = svc_udp_release_rqst, |
755cceab TT |
635 | .xpo_detach = svc_sock_detach, |
636 | .xpo_free = svc_sock_free, | |
323bee32 | 637 | .xpo_has_wspace = svc_udp_has_wspace, |
38a417cc | 638 | .xpo_accept = svc_udp_accept, |
16e4d93f | 639 | .xpo_secure_port = svc_sock_secure_port, |
ea08e392 | 640 | .xpo_kill_temp_xprt = svc_udp_kill_temp_xprt, |
360d8738 TT |
641 | }; |
642 | ||
643 | static struct svc_xprt_class svc_udp_class = { | |
644 | .xcl_name = "udp", | |
b700cbb1 | 645 | .xcl_owner = THIS_MODULE, |
360d8738 | 646 | .xcl_ops = &svc_udp_ops, |
49023155 | 647 | .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP, |
3c45ddf8 | 648 | .xcl_ident = XPRT_TRANSPORT_UDP, |
360d8738 TT |
649 | }; |
650 | ||
bb5cf160 | 651 | static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv) |
1da177e4 | 652 | { |
bd4620dd SK |
653 | svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class, |
654 | &svsk->sk_xprt, serv); | |
def13d74 | 655 | clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags); |
fa9251af | 656 | svsk->sk_sk->sk_data_ready = svc_data_ready; |
1da177e4 | 657 | svsk->sk_sk->sk_write_space = svc_write_space; |
1da177e4 LT |
658 | |
659 | /* initialise setting must have enough space to | |
cca5172a | 660 | * receive and respond to one request. |
1da177e4 LT |
661 | * svc_udp_recvfrom will re-adjust if necessary |
662 | */ | |
b7e5034c | 663 | svc_sock_setbufsize(svsk, 3); |
1da177e4 | 664 | |
0f0257ea TT |
665 | /* data might have come in before data_ready set up */ |
666 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); | |
02fc6c36 | 667 | set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); |
7a37f578 | 668 | |
7a37f578 | 669 | /* make sure we get destination address info */ |
7702ce40 CL |
670 | switch (svsk->sk_sk->sk_family) { |
671 | case AF_INET: | |
c1f9ec57 | 672 | ip_sock_set_pktinfo(svsk->sk_sock->sk); |
7d7207c2 | 673 | break; |
7702ce40 | 674 | case AF_INET6: |
7d7207c2 | 675 | ip6_sock_set_recvpktinfo(svsk->sk_sock->sk); |
7702ce40 CL |
676 | break; |
677 | default: | |
678 | BUG(); | |
679 | } | |
1da177e4 LT |
680 | } |
681 | ||
682 | /* | |
683 | * A data_ready event on a listening socket means there's a connection | |
684 | * pending. Do not use state_change as a substitute for it. | |
685 | */ | |
676d2369 | 686 | static void svc_tcp_listen_data_ready(struct sock *sk) |
1da177e4 | 687 | { |
939bb7ef | 688 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 | 689 | |
40e0b090 PY |
690 | trace_sk_data_ready(sk); |
691 | ||
eebe53e8 VL |
692 | if (svsk) { |
693 | /* Refer to svc_setup_socket() for details. */ | |
694 | rmb(); | |
fa9251af | 695 | svsk->sk_odata(sk); |
eebe53e8 VL |
696 | } |
697 | ||
939bb7ef NB |
698 | /* |
699 | * This callback may called twice when a new connection | |
700 | * is established as a child socket inherits everything | |
701 | * from a parent LISTEN socket. | |
702 | * 1) data_ready method of the parent socket will be called | |
703 | * when one of child sockets become ESTABLISHED. | |
704 | * 2) data_ready method of the child socket may be called | |
705 | * when it receives data before the socket is accepted. | |
706 | * In case of 2, we should ignore it silently. | |
707 | */ | |
708 | if (sk->sk_state == TCP_LISTEN) { | |
709 | if (svsk) { | |
02fc6c36 | 710 | set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
f6150c3c | 711 | svc_xprt_enqueue(&svsk->sk_xprt); |
a0469f46 | 712 | } |
1da177e4 | 713 | } |
1da177e4 LT |
714 | } |
715 | ||
716 | /* | |
717 | * A state change on a connected socket means it's dying or dead. | |
718 | */ | |
0f0257ea | 719 | static void svc_tcp_state_change(struct sock *sk) |
1da177e4 | 720 | { |
939bb7ef | 721 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 | 722 | |
a0469f46 | 723 | if (svsk) { |
eebe53e8 VL |
724 | /* Refer to svc_setup_socket() for details. */ |
725 | rmb(); | |
fa9251af | 726 | svsk->sk_ostate(sk); |
a0469f46 | 727 | trace_svcsock_tcp_state(&svsk->sk_xprt, svsk->sk_sock); |
e844d307 CL |
728 | if (sk->sk_state != TCP_ESTABLISHED) |
729 | svc_xprt_deferred_close(&svsk->sk_xprt); | |
1da177e4 | 730 | } |
1da177e4 LT |
731 | } |
732 | ||
733 | /* | |
734 | * Accept a TCP connection | |
735 | */ | |
38a417cc | 736 | static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt) |
1da177e4 | 737 | { |
38a417cc | 738 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); |
cdd88b9f | 739 | struct sockaddr_storage addr; |
740 | struct sockaddr *sin = (struct sockaddr *) &addr; | |
bb5cf160 | 741 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
1da177e4 LT |
742 | struct socket *sock = svsk->sk_sock; |
743 | struct socket *newsock; | |
1da177e4 LT |
744 | struct svc_sock *newsvsk; |
745 | int err, slen; | |
746 | ||
1da177e4 | 747 | if (!sock) |
38a417cc | 748 | return NULL; |
1da177e4 | 749 | |
02fc6c36 | 750 | clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
e6242e92 SS |
751 | err = kernel_accept(sock, &newsock, O_NONBLOCK); |
752 | if (err < 0) { | |
1da177e4 LT |
753 | if (err == -ENOMEM) |
754 | printk(KERN_WARNING "%s: no more sockets!\n", | |
755 | serv->sv_name); | |
e87cc472 JP |
756 | else if (err != -EAGAIN) |
757 | net_warn_ratelimited("%s: accept failed (err %d)!\n", | |
758 | serv->sv_name, -err); | |
a0469f46 | 759 | trace_svcsock_accept_err(xprt, serv->sv_name, err); |
38a417cc | 760 | return NULL; |
1da177e4 | 761 | } |
02fc6c36 | 762 | set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 763 | |
9b2c45d4 | 764 | err = kernel_getpeername(newsock, sin); |
1da177e4 | 765 | if (err < 0) { |
a0469f46 | 766 | trace_svcsock_getpeername_err(xprt, serv->sv_name, err); |
1da177e4 LT |
767 | goto failed; /* aborted connection or whatever */ |
768 | } | |
9b2c45d4 | 769 | slen = err; |
1da177e4 | 770 | |
fa9251af TM |
771 | /* Reset the inherited callbacks before calling svc_setup_socket */ |
772 | newsock->sk->sk_state_change = svsk->sk_ostate; | |
773 | newsock->sk->sk_data_ready = svsk->sk_odata; | |
774 | newsock->sk->sk_write_space = svsk->sk_owspace; | |
775 | ||
1da177e4 LT |
776 | /* make sure that a write doesn't block forever when |
777 | * low on memory | |
778 | */ | |
779 | newsock->sk->sk_sndtimeo = HZ*30; | |
780 | ||
72c35376 BF |
781 | newsvsk = svc_setup_socket(serv, newsock, |
782 | (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY)); | |
783 | if (IS_ERR(newsvsk)) | |
1da177e4 | 784 | goto failed; |
9dbc240f | 785 | svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen); |
9b2c45d4 DV |
786 | err = kernel_getsockname(newsock, sin); |
787 | slen = err; | |
a0469f46 | 788 | if (unlikely(err < 0)) |
a9747692 | 789 | slen = offsetof(struct sockaddr, sa_data); |
9dbc240f | 790 | svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen); |
067d7817 | 791 | |
ef11ce24 N |
792 | if (sock_is_loopback(newsock->sk)) |
793 | set_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags); | |
794 | else | |
795 | clear_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags); | |
f9f3cc4f TT |
796 | if (serv->sv_stats) |
797 | serv->sv_stats->nettcpconn++; | |
798 | ||
799 | return &newsvsk->sk_xprt; | |
800 | ||
801 | failed: | |
802 | sock_release(newsock); | |
803 | return NULL; | |
804 | } | |
805 | ||
ca07eda3 CL |
806 | static size_t svc_tcp_restore_pages(struct svc_sock *svsk, |
807 | struct svc_rqst *rqstp) | |
31d68ef6 | 808 | { |
ca07eda3 CL |
809 | size_t len = svsk->sk_datalen; |
810 | unsigned int i, npages; | |
31d68ef6 | 811 | |
ca07eda3 | 812 | if (!len) |
31d68ef6 | 813 | return 0; |
31d68ef6 BF |
814 | npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
815 | for (i = 0; i < npages; i++) { | |
816 | if (rqstp->rq_pages[i] != NULL) | |
817 | put_page(rqstp->rq_pages[i]); | |
818 | BUG_ON(svsk->sk_pages[i] == NULL); | |
819 | rqstp->rq_pages[i] = svsk->sk_pages[i]; | |
820 | svsk->sk_pages[i] = NULL; | |
821 | } | |
822 | rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]); | |
823 | return len; | |
824 | } | |
825 | ||
826 | static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp) | |
827 | { | |
828 | unsigned int i, len, npages; | |
829 | ||
8af345f5 | 830 | if (svsk->sk_datalen == 0) |
31d68ef6 | 831 | return; |
8af345f5 | 832 | len = svsk->sk_datalen; |
31d68ef6 BF |
833 | npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
834 | for (i = 0; i < npages; i++) { | |
835 | svsk->sk_pages[i] = rqstp->rq_pages[i]; | |
836 | rqstp->rq_pages[i] = NULL; | |
837 | } | |
838 | } | |
839 | ||
840 | static void svc_tcp_clear_pages(struct svc_sock *svsk) | |
841 | { | |
842 | unsigned int i, len, npages; | |
843 | ||
8af345f5 | 844 | if (svsk->sk_datalen == 0) |
31d68ef6 | 845 | goto out; |
8af345f5 | 846 | len = svsk->sk_datalen; |
31d68ef6 BF |
847 | npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
848 | for (i = 0; i < npages; i++) { | |
cf3aa02c BF |
849 | if (svsk->sk_pages[i] == NULL) { |
850 | WARN_ON_ONCE(1); | |
851 | continue; | |
852 | } | |
31d68ef6 BF |
853 | put_page(svsk->sk_pages[i]); |
854 | svsk->sk_pages[i] = NULL; | |
855 | } | |
856 | out: | |
857 | svsk->sk_tcplen = 0; | |
8af345f5 | 858 | svsk->sk_datalen = 0; |
31d68ef6 BF |
859 | } |
860 | ||
1da177e4 | 861 | /* |
a5cda73e | 862 | * Receive fragment record header into sk_marker. |
1da177e4 | 863 | */ |
a5cda73e CL |
864 | static ssize_t svc_tcp_read_marker(struct svc_sock *svsk, |
865 | struct svc_rqst *rqstp) | |
1da177e4 | 866 | { |
a5cda73e | 867 | ssize_t want, len; |
1da177e4 | 868 | |
a5cda73e CL |
869 | /* If we haven't gotten the record length yet, |
870 | * get the next four bytes. | |
871 | */ | |
c0401ea0 | 872 | if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) { |
a5cda73e | 873 | struct msghdr msg = { NULL }; |
1da177e4 LT |
874 | struct kvec iov; |
875 | ||
5ee78d48 | 876 | want = sizeof(rpc_fraghdr) - svsk->sk_tcplen; |
02648908 | 877 | iov.iov_base = ((char *)&svsk->sk_marker) + svsk->sk_tcplen; |
1da177e4 | 878 | iov.iov_len = want; |
de4eda9d | 879 | iov_iter_kvec(&msg.msg_iter, ITER_DEST, &iov, 1, want); |
a5cda73e | 880 | len = sock_recvmsg(svsk->sk_sock, &msg, MSG_DONTWAIT); |
4c8e5537 | 881 | if (len < 0) |
a5cda73e | 882 | return len; |
1da177e4 | 883 | svsk->sk_tcplen += len; |
1da177e4 | 884 | if (len < want) { |
a5cda73e CL |
885 | /* call again to read the remaining bytes */ |
886 | goto err_short; | |
1da177e4 | 887 | } |
a5cda73e | 888 | trace_svcsock_marker(&svsk->sk_xprt, svsk->sk_marker); |
836fbadb | 889 | if (svc_sock_reclen(svsk) + svsk->sk_datalen > |
a5cda73e CL |
890 | svsk->sk_xprt.xpt_server->sv_max_mesg) |
891 | goto err_too_large; | |
1da177e4 | 892 | } |
cc248d4b | 893 | return svc_sock_reclen(svsk); |
a5cda73e CL |
894 | |
895 | err_too_large: | |
896 | net_notice_ratelimited("svc: %s %s RPC fragment too large: %d\n", | |
897 | __func__, svsk->sk_xprt.xpt_server->sv_name, | |
898 | svc_sock_reclen(svsk)); | |
e844d307 | 899 | svc_xprt_deferred_close(&svsk->sk_xprt); |
a5cda73e | 900 | err_short: |
8f55f3c0 AB |
901 | return -EAGAIN; |
902 | } | |
903 | ||
586c52cc | 904 | static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp) |
4cfc7e60 | 905 | { |
586c52cc | 906 | struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt; |
4cfc7e60 | 907 | struct rpc_rqst *req = NULL; |
586c52cc TM |
908 | struct kvec *src, *dst; |
909 | __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base; | |
48e6555c BF |
910 | __be32 xid; |
911 | __be32 calldir; | |
4cfc7e60 | 912 | |
4cfc7e60 RI |
913 | xid = *p++; |
914 | calldir = *p; | |
915 | ||
093a1468 | 916 | if (!bc_xprt) |
586c52cc | 917 | return -EAGAIN; |
75c84151 | 918 | spin_lock(&bc_xprt->queue_lock); |
093a1468 TM |
919 | req = xprt_lookup_rqst(bc_xprt, xid); |
920 | if (!req) | |
921 | goto unlock_notfound; | |
586c52cc TM |
922 | |
923 | memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf)); | |
924 | /* | |
925 | * XXX!: cheating for now! Only copying HEAD. | |
926 | * But we know this is good enough for now (in fact, for any | |
927 | * callback reply in the forseeable future). | |
928 | */ | |
929 | dst = &req->rq_private_buf.head[0]; | |
930 | src = &rqstp->rq_arg.head[0]; | |
931 | if (dst->iov_len < src->iov_len) | |
093a1468 | 932 | goto unlock_eagain; /* whatever; just giving up. */ |
586c52cc | 933 | memcpy(dst->iov_base, src->iov_base, src->iov_len); |
cc248d4b | 934 | xprt_complete_rqst(req->rq_task, rqstp->rq_arg.len); |
586c52cc | 935 | rqstp->rq_arg.len = 0; |
75c84151 | 936 | spin_unlock(&bc_xprt->queue_lock); |
586c52cc | 937 | return 0; |
093a1468 TM |
938 | unlock_notfound: |
939 | printk(KERN_NOTICE | |
940 | "%s: Got unrecognized reply: " | |
941 | "calldir 0x%x xpt_bc_xprt %p xid %08x\n", | |
942 | __func__, ntohl(calldir), | |
943 | bc_xprt, ntohl(xid)); | |
944 | unlock_eagain: | |
75c84151 | 945 | spin_unlock(&bc_xprt->queue_lock); |
093a1468 | 946 | return -EAGAIN; |
586c52cc TM |
947 | } |
948 | ||
836fbadb BF |
949 | static void svc_tcp_fragment_received(struct svc_sock *svsk) |
950 | { | |
951 | /* If we have more data, signal svc_xprt_enqueue() to try again */ | |
836fbadb | 952 | svsk->sk_tcplen = 0; |
02648908 | 953 | svsk->sk_marker = xdr_zero; |
836fbadb | 954 | } |
31d68ef6 | 955 | |
ca07eda3 CL |
956 | /** |
957 | * svc_tcp_recvfrom - Receive data from a TCP socket | |
958 | * @rqstp: request structure into which to receive an RPC Call | |
959 | * | |
960 | * Called in a loop when XPT_DATA has been set. | |
961 | * | |
962 | * Read the 4-byte stream record marker, then use the record length | |
963 | * in that marker to set up exactly the resources needed to receive | |
964 | * the next RPC message into @rqstp. | |
965 | * | |
966 | * Returns: | |
967 | * On success, the number of bytes in a received RPC Call, or | |
968 | * %0 if a complete RPC Call message was not ready to return | |
969 | * | |
970 | * The zero return case handles partial receives and callback Replies. | |
971 | * The state of a partial receive is preserved in the svc_sock for | |
972 | * the next call to svc_tcp_recvfrom. | |
8f55f3c0 AB |
973 | */ |
974 | static int svc_tcp_recvfrom(struct svc_rqst *rqstp) | |
975 | { | |
976 | struct svc_sock *svsk = | |
977 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
978 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; | |
ca07eda3 CL |
979 | size_t want, base; |
980 | ssize_t len; | |
586c52cc TM |
981 | __be32 *p; |
982 | __be32 calldir; | |
8f55f3c0 | 983 | |
a5cda73e CL |
984 | clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
985 | len = svc_tcp_read_marker(svsk, rqstp); | |
8f55f3c0 AB |
986 | if (len < 0) |
987 | goto error; | |
988 | ||
31d68ef6 | 989 | base = svc_tcp_restore_pages(svsk, rqstp); |
a5cda73e | 990 | want = len - (svsk->sk_tcplen - sizeof(rpc_fraghdr)); |
ca07eda3 | 991 | len = svc_tcp_read_msg(rqstp, base + want, base); |
8af345f5 | 992 | if (len >= 0) { |
7dae1dd7 | 993 | trace_svcsock_tcp_recv(&svsk->sk_xprt, len); |
31d68ef6 | 994 | svsk->sk_tcplen += len; |
8af345f5 BF |
995 | svsk->sk_datalen += len; |
996 | } | |
6be8c594 CL |
997 | if (len != want || !svc_sock_final_rec(svsk)) |
998 | goto err_incomplete; | |
999 | if (svsk->sk_datalen < 8) | |
1000 | goto err_nuts; | |
ad46ccf0 | 1001 | |
836fbadb | 1002 | rqstp->rq_arg.len = svsk->sk_datalen; |
1da177e4 | 1003 | rqstp->rq_arg.page_base = 0; |
5ee78d48 TM |
1004 | if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) { |
1005 | rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len; | |
1da177e4 | 1006 | rqstp->rq_arg.page_len = 0; |
5ee78d48 TM |
1007 | } else |
1008 | rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len; | |
1da177e4 | 1009 | |
5148bf4e | 1010 | rqstp->rq_xprt_ctxt = NULL; |
1da177e4 | 1011 | rqstp->rq_prot = IPPROTO_TCP; |
7501cc2b | 1012 | if (test_bit(XPT_LOCAL, &svsk->sk_xprt.xpt_flags)) |
7827c81f | 1013 | set_bit(RQ_LOCAL, &rqstp->rq_flags); |
7501cc2b | 1014 | else |
7827c81f | 1015 | clear_bit(RQ_LOCAL, &rqstp->rq_flags); |
1da177e4 | 1016 | |
586c52cc TM |
1017 | p = (__be32 *)rqstp->rq_arg.head[0].iov_base; |
1018 | calldir = p[1]; | |
8985ef0b | 1019 | if (calldir) |
586c52cc | 1020 | len = receive_cb_reply(svsk, rqstp); |
586c52cc | 1021 | |
1da177e4 | 1022 | /* Reset TCP read info */ |
8af345f5 | 1023 | svsk->sk_datalen = 0; |
836fbadb | 1024 | svc_tcp_fragment_received(svsk); |
0601f793 | 1025 | |
8985ef0b BF |
1026 | if (len < 0) |
1027 | goto error; | |
1da177e4 | 1028 | |
9dbc240f | 1029 | svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt); |
1da177e4 LT |
1030 | if (serv->sv_stats) |
1031 | serv->sv_stats->nettcpcnt++; | |
1032 | ||
82011c80 | 1033 | svc_xprt_received(rqstp->rq_xprt); |
5ee78d48 | 1034 | return rqstp->rq_arg.len; |
1da177e4 | 1035 | |
6be8c594 CL |
1036 | err_incomplete: |
1037 | svc_tcp_save_pages(svsk, rqstp); | |
1038 | if (len < 0 && len != -EAGAIN) | |
1039 | goto err_delete; | |
1040 | if (len == want) | |
1041 | svc_tcp_fragment_received(svsk); | |
1042 | else | |
1043 | trace_svcsock_tcp_recv_short(&svsk->sk_xprt, | |
1044 | svc_sock_reclen(svsk), | |
1045 | svsk->sk_tcplen - sizeof(rpc_fraghdr)); | |
1046 | goto err_noclose; | |
8f55f3c0 | 1047 | error: |
31d68ef6 | 1048 | if (len != -EAGAIN) |
ad46ccf0 | 1049 | goto err_delete; |
6be8c594 | 1050 | trace_svcsock_tcp_recv_eagain(&svsk->sk_xprt, 0); |
82011c80 | 1051 | goto err_noclose; |
6be8c594 CL |
1052 | err_nuts: |
1053 | svsk->sk_datalen = 0; | |
ad46ccf0 | 1054 | err_delete: |
6be8c594 | 1055 | trace_svcsock_tcp_recv_err(&svsk->sk_xprt, len); |
e844d307 | 1056 | svc_xprt_deferred_close(&svsk->sk_xprt); |
31d68ef6 | 1057 | err_noclose: |
82011c80 | 1058 | svc_xprt_received(rqstp->rq_xprt); |
9f9d2ebe | 1059 | return 0; /* record not complete */ |
1da177e4 LT |
1060 | } |
1061 | ||
4a85a6a3 CL |
1062 | static int svc_tcp_send_kvec(struct socket *sock, const struct kvec *vec, |
1063 | int flags) | |
1064 | { | |
1065 | return kernel_sendpage(sock, virt_to_page(vec->iov_base), | |
1066 | offset_in_page(vec->iov_base), | |
1067 | vec->iov_len, flags); | |
1068 | } | |
1069 | ||
1070 | /* | |
1071 | * kernel_sendpage() is used exclusively to reduce the number of | |
1072 | * copy operations in this path. Therefore the caller must ensure | |
1073 | * that the pages backing @xdr are unchanging. | |
1074 | * | |
1075 | * In addition, the logic assumes that * .bv_len is never larger | |
1076 | * than PAGE_SIZE. | |
1077 | */ | |
4d12b727 CL |
1078 | static int svc_tcp_sendmsg(struct socket *sock, struct xdr_buf *xdr, |
1079 | rpc_fraghdr marker, unsigned int *sentp) | |
4a85a6a3 CL |
1080 | { |
1081 | const struct kvec *head = xdr->head; | |
1082 | const struct kvec *tail = xdr->tail; | |
1083 | struct kvec rm = { | |
1084 | .iov_base = &marker, | |
1085 | .iov_len = sizeof(marker), | |
1086 | }; | |
4d12b727 CL |
1087 | struct msghdr msg = { |
1088 | .msg_flags = 0, | |
1089 | }; | |
987c7b1d | 1090 | int ret; |
4a85a6a3 CL |
1091 | |
1092 | *sentp = 0; | |
b056fa07 TM |
1093 | ret = xdr_alloc_bvec(xdr, GFP_KERNEL); |
1094 | if (ret < 0) | |
1095 | return ret; | |
4a85a6a3 | 1096 | |
4d12b727 | 1097 | ret = kernel_sendmsg(sock, &msg, &rm, 1, rm.iov_len); |
4a85a6a3 CL |
1098 | if (ret < 0) |
1099 | return ret; | |
1100 | *sentp += ret; | |
1101 | if (ret != rm.iov_len) | |
1102 | return -EAGAIN; | |
1103 | ||
987c7b1d | 1104 | ret = svc_tcp_send_kvec(sock, head, 0); |
4a85a6a3 CL |
1105 | if (ret < 0) |
1106 | return ret; | |
1107 | *sentp += ret; | |
1108 | if (ret != head->iov_len) | |
1109 | goto out; | |
1110 | ||
1111 | if (xdr->page_len) { | |
1112 | unsigned int offset, len, remaining; | |
1113 | struct bio_vec *bvec; | |
1114 | ||
bad4c6eb CL |
1115 | bvec = xdr->bvec + (xdr->page_base >> PAGE_SHIFT); |
1116 | offset = offset_in_page(xdr->page_base); | |
4a85a6a3 | 1117 | remaining = xdr->page_len; |
4a85a6a3 | 1118 | while (remaining > 0) { |
bad4c6eb | 1119 | len = min(remaining, bvec->bv_len - offset); |
4a85a6a3 CL |
1120 | ret = kernel_sendpage(sock, bvec->bv_page, |
1121 | bvec->bv_offset + offset, | |
987c7b1d | 1122 | len, 0); |
4a85a6a3 CL |
1123 | if (ret < 0) |
1124 | return ret; | |
1125 | *sentp += ret; | |
1126 | if (ret != len) | |
1127 | goto out; | |
1128 | remaining -= len; | |
1129 | offset = 0; | |
1130 | bvec++; | |
1131 | } | |
1132 | } | |
1133 | ||
1134 | if (tail->iov_len) { | |
1135 | ret = svc_tcp_send_kvec(sock, tail, 0); | |
1136 | if (ret < 0) | |
1137 | return ret; | |
1138 | *sentp += ret; | |
1139 | } | |
1140 | ||
1141 | out: | |
1142 | return 0; | |
1143 | } | |
1144 | ||
da1661b9 CL |
1145 | /** |
1146 | * svc_tcp_sendto - Send out a reply on a TCP socket | |
1147 | * @rqstp: completed svc_rqst | |
1148 | * | |
ca4faf54 CL |
1149 | * xpt_mutex ensures @rqstp's whole message is written to the socket |
1150 | * without interruption. | |
1151 | * | |
da1661b9 | 1152 | * Returns the number of bytes sent, or a negative errno. |
1da177e4 | 1153 | */ |
0f0257ea | 1154 | static int svc_tcp_sendto(struct svc_rqst *rqstp) |
1da177e4 | 1155 | { |
da1661b9 CL |
1156 | struct svc_xprt *xprt = rqstp->rq_xprt; |
1157 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
1158 | struct xdr_buf *xdr = &rqstp->rq_res; | |
1159 | rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | | |
1160 | (u32)xdr->len); | |
3f649ab7 | 1161 | unsigned int sent; |
da1661b9 | 1162 | int err; |
1da177e4 | 1163 | |
cca557a5 | 1164 | svc_tcp_release_rqst(rqstp); |
23cf1ee1 | 1165 | |
e0a912e8 | 1166 | atomic_inc(&svsk->sk_sendqlen); |
ca4faf54 CL |
1167 | mutex_lock(&xprt->xpt_mutex); |
1168 | if (svc_xprt_is_dead(xprt)) | |
1169 | goto out_notconn; | |
e0a912e8 | 1170 | tcp_sock_set_cork(svsk->sk_sk, true); |
4d12b727 | 1171 | err = svc_tcp_sendmsg(svsk->sk_sock, xdr, marker, &sent); |
da1661b9 | 1172 | xdr_free_bvec(xdr); |
cb579086 | 1173 | trace_svcsock_tcp_send(xprt, err < 0 ? (long)err : sent); |
da1661b9 CL |
1174 | if (err < 0 || sent != (xdr->len + sizeof(marker))) |
1175 | goto out_close; | |
e0a912e8 TM |
1176 | if (atomic_dec_and_test(&svsk->sk_sendqlen)) |
1177 | tcp_sock_set_cork(svsk->sk_sk, false); | |
ca4faf54 | 1178 | mutex_unlock(&xprt->xpt_mutex); |
1da177e4 | 1179 | return sent; |
da1661b9 | 1180 | |
ca4faf54 | 1181 | out_notconn: |
e0a912e8 | 1182 | atomic_dec(&svsk->sk_sendqlen); |
ca4faf54 CL |
1183 | mutex_unlock(&xprt->xpt_mutex); |
1184 | return -ENOTCONN; | |
da1661b9 CL |
1185 | out_close: |
1186 | pr_notice("rpc-srv/tcp: %s: %s %d when sending %d bytes - shutting down socket\n", | |
1187 | xprt->xpt_server->sv_name, | |
1188 | (err < 0) ? "got error" : "sent", | |
1189 | (err < 0) ? err : sent, xdr->len); | |
e844d307 | 1190 | svc_xprt_deferred_close(xprt); |
e0a912e8 | 1191 | atomic_dec(&svsk->sk_sendqlen); |
ca4faf54 | 1192 | mutex_unlock(&xprt->xpt_mutex); |
da1661b9 | 1193 | return -EAGAIN; |
1da177e4 LT |
1194 | } |
1195 | ||
b700cbb1 | 1196 | static struct svc_xprt *svc_tcp_create(struct svc_serv *serv, |
62832c03 | 1197 | struct net *net, |
b700cbb1 TT |
1198 | struct sockaddr *sa, int salen, |
1199 | int flags) | |
1200 | { | |
62832c03 | 1201 | return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags); |
b700cbb1 TT |
1202 | } |
1203 | ||
2412e927 | 1204 | static const struct svc_xprt_ops svc_tcp_ops = { |
b700cbb1 | 1205 | .xpo_create = svc_tcp_create, |
5d137990 TT |
1206 | .xpo_recvfrom = svc_tcp_recvfrom, |
1207 | .xpo_sendto = svc_tcp_sendto, | |
03493bca | 1208 | .xpo_result_payload = svc_sock_result_payload, |
cca557a5 | 1209 | .xpo_release_rqst = svc_tcp_release_rqst, |
69b6ba37 | 1210 | .xpo_detach = svc_tcp_sock_detach, |
755cceab | 1211 | .xpo_free = svc_sock_free, |
323bee32 | 1212 | .xpo_has_wspace = svc_tcp_has_wspace, |
38a417cc | 1213 | .xpo_accept = svc_tcp_accept, |
16e4d93f | 1214 | .xpo_secure_port = svc_sock_secure_port, |
ea08e392 | 1215 | .xpo_kill_temp_xprt = svc_tcp_kill_temp_xprt, |
360d8738 TT |
1216 | }; |
1217 | ||
1218 | static struct svc_xprt_class svc_tcp_class = { | |
1219 | .xcl_name = "tcp", | |
b700cbb1 | 1220 | .xcl_owner = THIS_MODULE, |
360d8738 | 1221 | .xcl_ops = &svc_tcp_ops, |
49023155 | 1222 | .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP, |
3c45ddf8 | 1223 | .xcl_ident = XPRT_TRANSPORT_TCP, |
360d8738 TT |
1224 | }; |
1225 | ||
1226 | void svc_init_xprt_sock(void) | |
1227 | { | |
1228 | svc_reg_xprt_class(&svc_tcp_class); | |
1229 | svc_reg_xprt_class(&svc_udp_class); | |
1230 | } | |
1231 | ||
1232 | void svc_cleanup_xprt_sock(void) | |
1233 | { | |
1234 | svc_unreg_xprt_class(&svc_tcp_class); | |
1235 | svc_unreg_xprt_class(&svc_udp_class); | |
1236 | } | |
1237 | ||
bb5cf160 | 1238 | static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv) |
1da177e4 LT |
1239 | { |
1240 | struct sock *sk = svsk->sk_sk; | |
1da177e4 | 1241 | |
bd4620dd SK |
1242 | svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class, |
1243 | &svsk->sk_xprt, serv); | |
def13d74 | 1244 | set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags); |
362142b2 | 1245 | set_bit(XPT_CONG_CTRL, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 1246 | if (sk->sk_state == TCP_LISTEN) { |
ece200dd | 1247 | strcpy(svsk->sk_xprt.xpt_remotebuf, "listener"); |
02fc6c36 | 1248 | set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 1249 | sk->sk_data_ready = svc_tcp_listen_data_ready; |
02fc6c36 | 1250 | set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 1251 | } else { |
1da177e4 | 1252 | sk->sk_state_change = svc_tcp_state_change; |
fa9251af | 1253 | sk->sk_data_ready = svc_data_ready; |
637600f3 | 1254 | sk->sk_write_space = svc_write_space; |
1da177e4 | 1255 | |
02648908 | 1256 | svsk->sk_marker = xdr_zero; |
1da177e4 | 1257 | svsk->sk_tcplen = 0; |
8af345f5 | 1258 | svsk->sk_datalen = 0; |
31d68ef6 | 1259 | memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages)); |
1da177e4 | 1260 | |
e0a912e8 | 1261 | tcp_sock_set_nodelay(sk); |
1da177e4 | 1262 | |
02fc6c36 | 1263 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
b2f21f7d TM |
1264 | switch (sk->sk_state) { |
1265 | case TCP_SYN_RECV: | |
1266 | case TCP_ESTABLISHED: | |
1267 | break; | |
1268 | default: | |
e844d307 | 1269 | svc_xprt_deferred_close(&svsk->sk_xprt); |
b2f21f7d | 1270 | } |
1da177e4 LT |
1271 | } |
1272 | } | |
1273 | ||
0f0257ea | 1274 | void svc_sock_update_bufs(struct svc_serv *serv) |
1da177e4 LT |
1275 | { |
1276 | /* | |
1277 | * The number of server threads has changed. Update | |
1278 | * rcvbuf and sndbuf accordingly on all sockets | |
1279 | */ | |
8f3a6de3 | 1280 | struct svc_sock *svsk; |
1da177e4 LT |
1281 | |
1282 | spin_lock_bh(&serv->sv_lock); | |
8f3a6de3 | 1283 | list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list) |
02fc6c36 | 1284 | set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
1285 | spin_unlock_bh(&serv->sv_lock); |
1286 | } | |
24c3767e | 1287 | EXPORT_SYMBOL_GPL(svc_sock_update_bufs); |
1da177e4 | 1288 | |
1da177e4 LT |
1289 | /* |
1290 | * Initialize socket for RPC use and create svc_sock struct | |
1da177e4 | 1291 | */ |
6b174337 CL |
1292 | static struct svc_sock *svc_setup_socket(struct svc_serv *serv, |
1293 | struct socket *sock, | |
72c35376 | 1294 | int flags) |
1da177e4 LT |
1295 | { |
1296 | struct svc_sock *svsk; | |
1297 | struct sock *inet; | |
6b174337 | 1298 | int pmap_register = !(flags & SVC_SOCK_ANONYMOUS); |
72c35376 | 1299 | int err = 0; |
1da177e4 | 1300 | |
72c35376 BF |
1301 | svsk = kzalloc(sizeof(*svsk), GFP_KERNEL); |
1302 | if (!svsk) | |
1303 | return ERR_PTR(-ENOMEM); | |
1da177e4 LT |
1304 | |
1305 | inet = sock->sk; | |
1306 | ||
1307 | /* Register socket with portmapper */ | |
72c35376 BF |
1308 | if (pmap_register) |
1309 | err = svc_register(serv, sock_net(sock->sk), inet->sk_family, | |
5247fab5 | 1310 | inet->sk_protocol, |
c720c7e8 | 1311 | ntohs(inet_sk(inet)->inet_sport)); |
1da177e4 | 1312 | |
72c35376 | 1313 | if (err < 0) { |
1da177e4 | 1314 | kfree(svsk); |
72c35376 | 1315 | return ERR_PTR(err); |
1da177e4 LT |
1316 | } |
1317 | ||
1da177e4 LT |
1318 | svsk->sk_sock = sock; |
1319 | svsk->sk_sk = inet; | |
1320 | svsk->sk_ostate = inet->sk_state_change; | |
1321 | svsk->sk_odata = inet->sk_data_ready; | |
1322 | svsk->sk_owspace = inet->sk_write_space; | |
eebe53e8 VL |
1323 | /* |
1324 | * This barrier is necessary in order to prevent race condition | |
1325 | * with svc_data_ready(), svc_listen_data_ready() and others | |
1326 | * when calling callbacks above. | |
1327 | */ | |
1328 | wmb(); | |
1329 | inet->sk_user_data = svsk; | |
1da177e4 LT |
1330 | |
1331 | /* Initialize the socket */ | |
1332 | if (sock->type == SOCK_DGRAM) | |
bb5cf160 | 1333 | svc_udp_init(svsk, serv); |
637600f3 | 1334 | else |
bb5cf160 | 1335 | svc_tcp_init(svsk, serv); |
1da177e4 | 1336 | |
998024de | 1337 | trace_svcsock_new_socket(sock); |
1da177e4 LT |
1338 | return svsk; |
1339 | } | |
1340 | ||
30646394 SK |
1341 | bool svc_alien_sock(struct net *net, int fd) |
1342 | { | |
1343 | int err; | |
1344 | struct socket *sock = sockfd_lookup(fd, &err); | |
1345 | bool ret = false; | |
1346 | ||
1347 | if (!sock) | |
1348 | goto out; | |
1349 | if (sock_net(sock->sk) != net) | |
1350 | ret = true; | |
1351 | sockfd_put(sock); | |
1352 | out: | |
1353 | return ret; | |
1354 | } | |
1355 | EXPORT_SYMBOL_GPL(svc_alien_sock); | |
1356 | ||
bfba9ab4 CL |
1357 | /** |
1358 | * svc_addsock - add a listener socket to an RPC service | |
1359 | * @serv: pointer to RPC service to which to add a new listener | |
1360 | * @fd: file descriptor of the new listener | |
1361 | * @name_return: pointer to buffer to fill in with name of listener | |
1362 | * @len: size of the buffer | |
4df493a2 | 1363 | * @cred: credential |
bfba9ab4 CL |
1364 | * |
1365 | * Fills in socket name and returns positive length of name if successful. | |
1366 | * Name is terminated with '\n'. On error, returns a negative errno | |
1367 | * value. | |
1368 | */ | |
1369 | int svc_addsock(struct svc_serv *serv, const int fd, char *name_return, | |
4df493a2 | 1370 | const size_t len, const struct cred *cred) |
b41b66d6 N |
1371 | { |
1372 | int err = 0; | |
1373 | struct socket *so = sockfd_lookup(fd, &err); | |
1374 | struct svc_sock *svsk = NULL; | |
a8e10078 BF |
1375 | struct sockaddr_storage addr; |
1376 | struct sockaddr *sin = (struct sockaddr *)&addr; | |
1377 | int salen; | |
b41b66d6 N |
1378 | |
1379 | if (!so) | |
1380 | return err; | |
a8e10078 | 1381 | err = -EAFNOSUPPORT; |
205ba423 | 1382 | if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6)) |
a8e10078 BF |
1383 | goto out; |
1384 | err = -EPROTONOSUPPORT; | |
1385 | if (so->sk->sk_protocol != IPPROTO_TCP && | |
b41b66d6 | 1386 | so->sk->sk_protocol != IPPROTO_UDP) |
a8e10078 BF |
1387 | goto out; |
1388 | err = -EISCONN; | |
1389 | if (so->state > SS_UNCONNECTED) | |
1390 | goto out; | |
1391 | err = -ENOENT; | |
1392 | if (!try_module_get(THIS_MODULE)) | |
1393 | goto out; | |
1394 | svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS); | |
1395 | if (IS_ERR(svsk)) { | |
1396 | module_put(THIS_MODULE); | |
1397 | err = PTR_ERR(svsk); | |
1398 | goto out; | |
b41b66d6 | 1399 | } |
9b2c45d4 DV |
1400 | salen = kernel_getsockname(svsk->sk_sock, sin); |
1401 | if (salen >= 0) | |
a8e10078 | 1402 | svc_xprt_set_local(&svsk->sk_xprt, sin, salen); |
4df493a2 | 1403 | svsk->sk_xprt.xpt_cred = get_cred(cred); |
39b55301 | 1404 | svc_add_new_perm_xprt(serv, &svsk->sk_xprt); |
e7942b9f | 1405 | return svc_one_sock_name(svsk, name_return, len); |
a8e10078 BF |
1406 | out: |
1407 | sockfd_put(so); | |
1408 | return err; | |
b41b66d6 N |
1409 | } |
1410 | EXPORT_SYMBOL_GPL(svc_addsock); | |
1411 | ||
1da177e4 LT |
1412 | /* |
1413 | * Create socket for RPC service. | |
1414 | */ | |
b700cbb1 TT |
1415 | static struct svc_xprt *svc_create_socket(struct svc_serv *serv, |
1416 | int protocol, | |
62832c03 | 1417 | struct net *net, |
b700cbb1 TT |
1418 | struct sockaddr *sin, int len, |
1419 | int flags) | |
1da177e4 LT |
1420 | { |
1421 | struct svc_sock *svsk; | |
1422 | struct socket *sock; | |
1423 | int error; | |
1424 | int type; | |
9dbc240f TT |
1425 | struct sockaddr_storage addr; |
1426 | struct sockaddr *newsin = (struct sockaddr *)&addr; | |
1427 | int newlen; | |
c69da774 | 1428 | int family; |
1da177e4 LT |
1429 | |
1430 | if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) { | |
1431 | printk(KERN_WARNING "svc: only UDP and TCP " | |
1432 | "sockets supported\n"); | |
b700cbb1 | 1433 | return ERR_PTR(-EINVAL); |
1da177e4 | 1434 | } |
c69da774 | 1435 | |
1da177e4 | 1436 | type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM; |
c69da774 TM |
1437 | switch (sin->sa_family) { |
1438 | case AF_INET6: | |
1439 | family = PF_INET6; | |
1440 | break; | |
1441 | case AF_INET: | |
1442 | family = PF_INET; | |
1443 | break; | |
1444 | default: | |
1445 | return ERR_PTR(-EINVAL); | |
1446 | } | |
1da177e4 | 1447 | |
14ec63c3 | 1448 | error = __sock_create(net, family, type, protocol, &sock, 1); |
77f1f67a | 1449 | if (error < 0) |
b700cbb1 | 1450 | return ERR_PTR(error); |
1da177e4 | 1451 | |
ed07536e PZ |
1452 | svc_reclassify_socket(sock); |
1453 | ||
c69da774 TM |
1454 | /* |
1455 | * If this is an PF_INET6 listener, we want to avoid | |
1456 | * getting requests from IPv4 remotes. Those should | |
1457 | * be shunted to a PF_INET listener via rpcbind. | |
1458 | */ | |
c69da774 | 1459 | if (family == PF_INET6) |
9b115749 | 1460 | ip6_sock_set_v6only(sock->sk); |
18114746 | 1461 | if (type == SOCK_STREAM) |
4a17fd52 | 1462 | sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */ |
77f1f67a | 1463 | error = kernel_bind(sock, sin, len); |
18114746 ES |
1464 | if (error < 0) |
1465 | goto bummer; | |
1da177e4 | 1466 | |
9b2c45d4 | 1467 | error = kernel_getsockname(sock, newsin); |
9dbc240f TT |
1468 | if (error < 0) |
1469 | goto bummer; | |
9b2c45d4 | 1470 | newlen = error; |
9dbc240f | 1471 | |
1da177e4 | 1472 | if (protocol == IPPROTO_TCP) { |
e6242e92 | 1473 | if ((error = kernel_listen(sock, 64)) < 0) |
1da177e4 LT |
1474 | goto bummer; |
1475 | } | |
1476 | ||
72c35376 | 1477 | svsk = svc_setup_socket(serv, sock, flags); |
a8e10078 BF |
1478 | if (IS_ERR(svsk)) { |
1479 | error = PTR_ERR(svsk); | |
1480 | goto bummer; | |
e79eff1f | 1481 | } |
a8e10078 BF |
1482 | svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen); |
1483 | return (struct svc_xprt *)svsk; | |
1da177e4 | 1484 | bummer: |
1da177e4 | 1485 | sock_release(sock); |
b700cbb1 | 1486 | return ERR_PTR(error); |
1da177e4 LT |
1487 | } |
1488 | ||
755cceab TT |
1489 | /* |
1490 | * Detach the svc_sock from the socket so that no | |
1491 | * more callbacks occur. | |
1492 | */ | |
1493 | static void svc_sock_detach(struct svc_xprt *xprt) | |
1494 | { | |
1495 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
1496 | struct sock *sk = svsk->sk_sk; | |
1497 | ||
755cceab | 1498 | /* put back the old socket callbacks */ |
069c225b | 1499 | lock_sock(sk); |
755cceab TT |
1500 | sk->sk_state_change = svsk->sk_ostate; |
1501 | sk->sk_data_ready = svsk->sk_odata; | |
1502 | sk->sk_write_space = svsk->sk_owspace; | |
069c225b TM |
1503 | sk->sk_user_data = NULL; |
1504 | release_sock(sk); | |
69b6ba37 TM |
1505 | } |
1506 | ||
1507 | /* | |
1508 | * Disconnect the socket, and reset the callbacks | |
1509 | */ | |
1510 | static void svc_tcp_sock_detach(struct svc_xprt *xprt) | |
1511 | { | |
1512 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
1513 | ||
69b6ba37 TM |
1514 | svc_sock_detach(xprt); |
1515 | ||
31d68ef6 BF |
1516 | if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) { |
1517 | svc_tcp_clear_pages(svsk); | |
69b6ba37 | 1518 | kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR); |
31d68ef6 | 1519 | } |
755cceab TT |
1520 | } |
1521 | ||
1522 | /* | |
1523 | * Free the svc_sock's socket resources and the svc_sock itself. | |
1524 | */ | |
1525 | static void svc_sock_free(struct svc_xprt *xprt) | |
1526 | { | |
1527 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
755cceab | 1528 | |
755cceab TT |
1529 | if (svsk->sk_sock->file) |
1530 | sockfd_put(svsk->sk_sock); | |
1531 | else | |
1532 | sock_release(svsk->sk_sock); | |
1533 | kfree(svsk); | |
1534 | } |