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 | |
998024de CL |
58 | #include <trace/events/sunrpc.h> |
59 | ||
9e55eef4 | 60 | #include "socklib.h" |
177e4f99 HS |
61 | #include "sunrpc.h" |
62 | ||
360d8738 | 63 | #define RPCDBG_FACILITY RPCDBG_SVCXPRT |
1da177e4 LT |
64 | |
65 | ||
66 | static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *, | |
72c35376 | 67 | int flags); |
1da177e4 LT |
68 | static int svc_udp_recvfrom(struct svc_rqst *); |
69 | static int svc_udp_sendto(struct svc_rqst *); | |
755cceab | 70 | static void svc_sock_detach(struct svc_xprt *); |
69b6ba37 | 71 | static void svc_tcp_sock_detach(struct svc_xprt *); |
755cceab | 72 | static void svc_sock_free(struct svc_xprt *); |
1da177e4 | 73 | |
b700cbb1 | 74 | static struct svc_xprt *svc_create_socket(struct svc_serv *, int, |
62832c03 PE |
75 | struct net *, struct sockaddr *, |
76 | int, int); | |
ed07536e PZ |
77 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
78 | static struct lock_class_key svc_key[2]; | |
79 | static struct lock_class_key svc_slock_key[2]; | |
80 | ||
0f0257ea | 81 | static void svc_reclassify_socket(struct socket *sock) |
ed07536e PZ |
82 | { |
83 | struct sock *sk = sock->sk; | |
1b7a1819 | 84 | |
fafc4e1e | 85 | if (WARN_ON_ONCE(!sock_allow_reclassification(sk))) |
1b7a1819 WAA |
86 | return; |
87 | ||
ed07536e PZ |
88 | switch (sk->sk_family) { |
89 | case AF_INET: | |
90 | sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD", | |
def13d74 TT |
91 | &svc_slock_key[0], |
92 | "sk_xprt.xpt_lock-AF_INET-NFSD", | |
93 | &svc_key[0]); | |
ed07536e PZ |
94 | break; |
95 | ||
96 | case AF_INET6: | |
97 | sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD", | |
def13d74 TT |
98 | &svc_slock_key[1], |
99 | "sk_xprt.xpt_lock-AF_INET6-NFSD", | |
100 | &svc_key[1]); | |
ed07536e PZ |
101 | break; |
102 | ||
103 | default: | |
104 | BUG(); | |
105 | } | |
106 | } | |
107 | #else | |
0f0257ea | 108 | static void svc_reclassify_socket(struct socket *sock) |
ed07536e PZ |
109 | { |
110 | } | |
111 | #endif | |
112 | ||
cca557a5 CL |
113 | /** |
114 | * svc_tcp_release_rqst - Release transport-related resources | |
115 | * @rqstp: request structure with resources to be released | |
116 | * | |
1da177e4 | 117 | */ |
cca557a5 | 118 | static void svc_tcp_release_rqst(struct svc_rqst *rqstp) |
1da177e4 | 119 | { |
5148bf4e | 120 | struct sk_buff *skb = rqstp->rq_xprt_ctxt; |
1da177e4 LT |
121 | |
122 | if (skb) { | |
57b1d3ba TT |
123 | struct svc_sock *svsk = |
124 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
1da177e4 | 125 | |
7dae1dd7 | 126 | rqstp->rq_xprt_ctxt = NULL; |
9d410c79 | 127 | skb_free_datagram_locked(svsk->sk_sk, skb); |
1da177e4 | 128 | } |
1da177e4 LT |
129 | } |
130 | ||
cca557a5 CL |
131 | /** |
132 | * svc_udp_release_rqst - Release transport-related resources | |
133 | * @rqstp: request structure with resources to be released | |
134 | * | |
135 | */ | |
136 | static void svc_udp_release_rqst(struct svc_rqst *rqstp) | |
850cbadd PA |
137 | { |
138 | struct sk_buff *skb = rqstp->rq_xprt_ctxt; | |
139 | ||
140 | if (skb) { | |
141 | rqstp->rq_xprt_ctxt = NULL; | |
850cbadd PA |
142 | consume_skb(skb); |
143 | } | |
144 | } | |
145 | ||
b92503b2 CL |
146 | union svc_pktinfo_u { |
147 | struct in_pktinfo pkti; | |
b92503b2 | 148 | struct in6_pktinfo pkti6; |
b92503b2 | 149 | }; |
bc375ea7 DM |
150 | #define SVC_PKTINFO_SPACE \ |
151 | CMSG_SPACE(sizeof(union svc_pktinfo_u)) | |
b92503b2 CL |
152 | |
153 | static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh) | |
154 | { | |
57b1d3ba TT |
155 | struct svc_sock *svsk = |
156 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
157 | switch (svsk->sk_sk->sk_family) { | |
b92503b2 CL |
158 | case AF_INET: { |
159 | struct in_pktinfo *pki = CMSG_DATA(cmh); | |
160 | ||
161 | cmh->cmsg_level = SOL_IP; | |
162 | cmh->cmsg_type = IP_PKTINFO; | |
163 | pki->ipi_ifindex = 0; | |
849a1cf1 MJ |
164 | pki->ipi_spec_dst.s_addr = |
165 | svc_daddr_in(rqstp)->sin_addr.s_addr; | |
b92503b2 CL |
166 | cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); |
167 | } | |
168 | break; | |
5a05ed73 | 169 | |
b92503b2 CL |
170 | case AF_INET6: { |
171 | struct in6_pktinfo *pki = CMSG_DATA(cmh); | |
849a1cf1 | 172 | struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp); |
b92503b2 CL |
173 | |
174 | cmh->cmsg_level = SOL_IPV6; | |
175 | cmh->cmsg_type = IPV6_PKTINFO; | |
849a1cf1 | 176 | pki->ipi6_ifindex = daddr->sin6_scope_id; |
4e3fd7a0 | 177 | pki->ipi6_addr = daddr->sin6_addr; |
b92503b2 CL |
178 | cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); |
179 | } | |
180 | break; | |
b92503b2 | 181 | } |
b92503b2 CL |
182 | } |
183 | ||
03493bca CL |
184 | static int svc_sock_result_payload(struct svc_rqst *rqstp, unsigned int offset, |
185 | unsigned int length) | |
41205539 CL |
186 | { |
187 | return 0; | |
188 | } | |
189 | ||
80212d59 N |
190 | /* |
191 | * Report socket names for nfsdfs | |
192 | */ | |
e7942b9f | 193 | static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining) |
80212d59 | 194 | { |
017cb47f CL |
195 | const struct sock *sk = svsk->sk_sk; |
196 | const char *proto_name = sk->sk_protocol == IPPROTO_UDP ? | |
197 | "udp" : "tcp"; | |
80212d59 N |
198 | int len; |
199 | ||
017cb47f | 200 | switch (sk->sk_family) { |
e7942b9f CL |
201 | case PF_INET: |
202 | len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n", | |
017cb47f | 203 | proto_name, |
c720c7e8 ED |
204 | &inet_sk(sk)->inet_rcv_saddr, |
205 | inet_sk(sk)->inet_num); | |
80212d59 | 206 | break; |
c2bb06db | 207 | #if IS_ENABLED(CONFIG_IPV6) |
58de2f86 CL |
208 | case PF_INET6: |
209 | len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n", | |
017cb47f | 210 | proto_name, |
efe4208f | 211 | &sk->sk_v6_rcv_saddr, |
c720c7e8 | 212 | inet_sk(sk)->inet_num); |
80212d59 | 213 | break; |
c2bb06db | 214 | #endif |
80212d59 | 215 | default: |
e7942b9f | 216 | len = snprintf(buf, remaining, "*unknown-%d*\n", |
017cb47f | 217 | sk->sk_family); |
80212d59 | 218 | } |
e7942b9f CL |
219 | |
220 | if (len >= remaining) { | |
221 | *buf = '\0'; | |
222 | return -ENAMETOOLONG; | |
80212d59 N |
223 | } |
224 | return len; | |
225 | } | |
226 | ||
ca07eda3 CL |
227 | #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE |
228 | static void svc_flush_bvec(const struct bio_vec *bvec, size_t size, size_t seek) | |
229 | { | |
230 | struct bvec_iter bi = { | |
13a9a9d7 | 231 | .bi_size = size + seek, |
ca07eda3 CL |
232 | }; |
233 | struct bio_vec bv; | |
234 | ||
235 | bvec_iter_advance(bvec, &bi, seek & PAGE_MASK); | |
236 | for_each_bvec(bv, bvec, bi, bi) | |
237 | flush_dcache_page(bv.bv_page); | |
238 | } | |
239 | #else | |
240 | static inline void svc_flush_bvec(const struct bio_vec *bvec, size_t size, | |
241 | size_t seek) | |
242 | { | |
243 | } | |
244 | #endif | |
245 | ||
1da177e4 | 246 | /* |
ca07eda3 CL |
247 | * Read from @rqstp's transport socket. The incoming message fills whole |
248 | * pages in @rqstp's rq_pages array until the last page of the message | |
249 | * has been received into a partial page. | |
1da177e4 | 250 | */ |
ca07eda3 CL |
251 | static ssize_t svc_tcp_read_msg(struct svc_rqst *rqstp, size_t buflen, |
252 | size_t seek) | |
1da177e4 | 253 | { |
57b1d3ba TT |
254 | struct svc_sock *svsk = |
255 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
ca07eda3 | 256 | struct bio_vec *bvec = rqstp->rq_bvec; |
4c8e5537 | 257 | struct msghdr msg = { NULL }; |
ca07eda3 | 258 | unsigned int i; |
4c8e5537 | 259 | ssize_t len; |
ca07eda3 | 260 | size_t t; |
1da177e4 | 261 | |
260c1d12 TT |
262 | rqstp->rq_xprt_hlen = 0; |
263 | ||
f8d1ff47 | 264 | clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
ca07eda3 CL |
265 | |
266 | for (i = 0, t = 0; t < buflen; i++, t += PAGE_SIZE) { | |
267 | bvec[i].bv_page = rqstp->rq_pages[i]; | |
268 | bvec[i].bv_len = PAGE_SIZE; | |
269 | bvec[i].bv_offset = 0; | |
270 | } | |
271 | rqstp->rq_respages = &rqstp->rq_pages[i]; | |
272 | rqstp->rq_next_page = rqstp->rq_respages + 1; | |
273 | ||
274 | iov_iter_bvec(&msg.msg_iter, READ, bvec, i, buflen); | |
275 | if (seek) { | |
276 | iov_iter_advance(&msg.msg_iter, seek); | |
277 | buflen -= seek; | |
4c8e5537 TM |
278 | } |
279 | len = sock_recvmsg(svsk->sk_sock, &msg, MSG_DONTWAIT); | |
ca07eda3 CL |
280 | if (len > 0) |
281 | svc_flush_bvec(bvec, len, seek); | |
282 | ||
f8d1ff47 TM |
283 | /* If we read a full record, then assume there may be more |
284 | * data to read (stream based sockets only!) | |
285 | */ | |
286 | if (len == buflen) | |
287 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); | |
1da177e4 | 288 | |
1da177e4 LT |
289 | return len; |
290 | } | |
291 | ||
292 | /* | |
293 | * Set socket snd and rcv buffer lengths | |
294 | */ | |
b7e5034c | 295 | static void svc_sock_setbufsize(struct svc_sock *svsk, unsigned int nreqs) |
1da177e4 | 296 | { |
b7e5034c BF |
297 | unsigned int max_mesg = svsk->sk_xprt.xpt_server->sv_max_mesg; |
298 | struct socket *sock = svsk->sk_sock; | |
299 | ||
300 | nreqs = min(nreqs, INT_MAX / 2 / max_mesg); | |
301 | ||
1da177e4 | 302 | lock_sock(sock->sk); |
b7e5034c BF |
303 | sock->sk->sk_sndbuf = nreqs * max_mesg * 2; |
304 | sock->sk->sk_rcvbuf = nreqs * max_mesg * 2; | |
47fcb03f | 305 | sock->sk->sk_write_space(sock->sk); |
1da177e4 | 306 | release_sock(sock->sk); |
1da177e4 | 307 | } |
16e4d93f | 308 | |
989f881e | 309 | static void svc_sock_secure_port(struct svc_rqst *rqstp) |
16e4d93f | 310 | { |
989f881e CL |
311 | if (svc_port_is_privileged(svc_addr(rqstp))) |
312 | set_bit(RQ_SECURE, &rqstp->rq_flags); | |
313 | else | |
314 | clear_bit(RQ_SECURE, &rqstp->rq_flags); | |
16e4d93f CL |
315 | } |
316 | ||
1da177e4 LT |
317 | /* |
318 | * INET callback when data has been received on the socket. | |
319 | */ | |
fa9251af | 320 | static void svc_data_ready(struct sock *sk) |
1da177e4 | 321 | { |
939bb7ef | 322 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 | 323 | |
939bb7ef | 324 | if (svsk) { |
eebe53e8 VL |
325 | /* Refer to svc_setup_socket() for details. */ |
326 | rmb(); | |
fa9251af | 327 | svsk->sk_odata(sk); |
998024de | 328 | trace_svcsock_data_ready(&svsk->sk_xprt, 0); |
4720b070 TM |
329 | if (!test_and_set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags)) |
330 | svc_xprt_enqueue(&svsk->sk_xprt); | |
939bb7ef | 331 | } |
1da177e4 LT |
332 | } |
333 | ||
334 | /* | |
335 | * INET callback when space is newly available on the socket. | |
336 | */ | |
0f0257ea | 337 | static void svc_write_space(struct sock *sk) |
1da177e4 LT |
338 | { |
339 | struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data); | |
340 | ||
341 | if (svsk) { | |
eebe53e8 VL |
342 | /* Refer to svc_setup_socket() for details. */ |
343 | rmb(); | |
998024de | 344 | trace_svcsock_write_space(&svsk->sk_xprt, 0); |
fa9251af | 345 | svsk->sk_owspace(sk); |
f6150c3c | 346 | svc_xprt_enqueue(&svsk->sk_xprt); |
1da177e4 | 347 | } |
1da177e4 LT |
348 | } |
349 | ||
c7fb3f06 TM |
350 | static int svc_tcp_has_wspace(struct svc_xprt *xprt) |
351 | { | |
637600f3 | 352 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); |
c7fb3f06 TM |
353 | |
354 | if (test_bit(XPT_LISTENER, &xprt->xpt_flags)) | |
355 | return 1; | |
637600f3 | 356 | return !test_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); |
51877680 TM |
357 | } |
358 | ||
ea08e392 SM |
359 | static void svc_tcp_kill_temp_xprt(struct svc_xprt *xprt) |
360 | { | |
c433594c CH |
361 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); |
362 | ||
363 | sock_no_linger(svsk->sk_sock->sk); | |
ea08e392 SM |
364 | } |
365 | ||
7702ce40 CL |
366 | /* |
367 | * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo | |
368 | */ | |
369 | static int svc_udp_get_dest_address4(struct svc_rqst *rqstp, | |
370 | struct cmsghdr *cmh) | |
371 | { | |
372 | struct in_pktinfo *pki = CMSG_DATA(cmh); | |
849a1cf1 MJ |
373 | struct sockaddr_in *daddr = svc_daddr_in(rqstp); |
374 | ||
7702ce40 CL |
375 | if (cmh->cmsg_type != IP_PKTINFO) |
376 | return 0; | |
849a1cf1 MJ |
377 | |
378 | daddr->sin_family = AF_INET; | |
379 | daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr; | |
7702ce40 CL |
380 | return 1; |
381 | } | |
382 | ||
383 | /* | |
73df66f8 | 384 | * See net/ipv6/datagram.c : ip6_datagram_recv_ctl |
7702ce40 CL |
385 | */ |
386 | static int svc_udp_get_dest_address6(struct svc_rqst *rqstp, | |
387 | struct cmsghdr *cmh) | |
388 | { | |
389 | struct in6_pktinfo *pki = CMSG_DATA(cmh); | |
849a1cf1 MJ |
390 | struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp); |
391 | ||
7702ce40 CL |
392 | if (cmh->cmsg_type != IPV6_PKTINFO) |
393 | return 0; | |
849a1cf1 MJ |
394 | |
395 | daddr->sin6_family = AF_INET6; | |
4e3fd7a0 | 396 | daddr->sin6_addr = pki->ipi6_addr; |
849a1cf1 | 397 | daddr->sin6_scope_id = pki->ipi6_ifindex; |
7702ce40 CL |
398 | return 1; |
399 | } | |
400 | ||
9dbc240f TT |
401 | /* |
402 | * Copy the UDP datagram's destination address to the rqstp structure. | |
403 | * The 'destination' address in this case is the address to which the | |
404 | * peer sent the datagram, i.e. our local address. For multihomed | |
405 | * hosts, this can change from msg to msg. Note that only the IP | |
406 | * address changes, the port number should remain the same. | |
407 | */ | |
7702ce40 CL |
408 | static int svc_udp_get_dest_address(struct svc_rqst *rqstp, |
409 | struct cmsghdr *cmh) | |
95756482 | 410 | { |
7702ce40 CL |
411 | switch (cmh->cmsg_level) { |
412 | case SOL_IP: | |
413 | return svc_udp_get_dest_address4(rqstp, cmh); | |
414 | case SOL_IPV6: | |
415 | return svc_udp_get_dest_address6(rqstp, cmh); | |
95756482 | 416 | } |
7702ce40 CL |
417 | |
418 | return 0; | |
95756482 CL |
419 | } |
420 | ||
fff1ebb2 CL |
421 | /** |
422 | * svc_udp_recvfrom - Receive a datagram from a UDP socket. | |
423 | * @rqstp: request structure into which to receive an RPC Call | |
424 | * | |
425 | * Called in a loop when XPT_DATA has been set. | |
426 | * | |
427 | * Returns: | |
428 | * On success, the number of bytes in a received RPC Call, or | |
429 | * %0 if a complete RPC Call message was not ready to return | |
1da177e4 | 430 | */ |
0f0257ea | 431 | static int svc_udp_recvfrom(struct svc_rqst *rqstp) |
1da177e4 | 432 | { |
57b1d3ba TT |
433 | struct svc_sock *svsk = |
434 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
bb5cf160 | 435 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
1da177e4 | 436 | struct sk_buff *skb; |
bc375ea7 DM |
437 | union { |
438 | struct cmsghdr hdr; | |
439 | long all[SVC_PKTINFO_SPACE / sizeof(long)]; | |
440 | } buffer; | |
441 | struct cmsghdr *cmh = &buffer.hdr; | |
7a37f578 N |
442 | struct msghdr msg = { |
443 | .msg_name = svc_addr(rqstp), | |
444 | .msg_control = cmh, | |
445 | .msg_controllen = sizeof(buffer), | |
446 | .msg_flags = MSG_DONTWAIT, | |
447 | }; | |
abc5c44d CL |
448 | size_t len; |
449 | int err; | |
1da177e4 | 450 | |
02fc6c36 | 451 | if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags)) |
1da177e4 LT |
452 | /* udp sockets need large rcvbuf as all pending |
453 | * requests are still in that buffer. sndbuf must | |
454 | * also be large enough that there is enough space | |
3262c816 GB |
455 | * for one reply per thread. We count all threads |
456 | * rather than threads in a particular pool, which | |
457 | * provides an upper bound on the number of threads | |
458 | * which will access the socket. | |
1da177e4 | 459 | */ |
b7e5034c | 460 | svc_sock_setbufsize(svsk, serv->sv_nrthreads + 3); |
1da177e4 | 461 | |
02fc6c36 | 462 | clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
05ed690e N |
463 | err = kernel_recvmsg(svsk->sk_sock, &msg, NULL, |
464 | 0, 0, MSG_PEEK | MSG_DONTWAIT); | |
fff1ebb2 CL |
465 | if (err < 0) |
466 | goto out_recv_err; | |
467 | skb = skb_recv_udp(svsk->sk_sk, 0, 1, &err); | |
468 | if (!skb) | |
469 | goto out_recv_err; | |
470 | ||
9dbc240f | 471 | len = svc_addr_len(svc_addr(rqstp)); |
9dbc240f | 472 | rqstp->rq_addrlen = len; |
2456e855 | 473 | if (skb->tstamp == 0) { |
b7aa0bf7 | 474 | skb->tstamp = ktime_get_real(); |
cca5172a | 475 | /* Don't enable netstamp, sunrpc doesn't |
1da177e4 LT |
476 | need that much accuracy */ |
477 | } | |
3a0ed3e9 | 478 | sock_write_timestamp(svsk->sk_sk, skb->tstamp); |
02fc6c36 | 479 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */ |
1da177e4 | 480 | |
fff1ebb2 | 481 | len = skb->len; |
1da177e4 | 482 | rqstp->rq_arg.len = len; |
fff1ebb2 | 483 | trace_svcsock_udp_recv(&svsk->sk_xprt, len); |
1da177e4 | 484 | |
95756482 | 485 | rqstp->rq_prot = IPPROTO_UDP; |
27459f09 | 486 | |
fff1ebb2 CL |
487 | if (!svc_udp_get_dest_address(rqstp, cmh)) |
488 | goto out_cmsg_err; | |
849a1cf1 | 489 | rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp)); |
1da177e4 LT |
490 | |
491 | if (skb_is_nonlinear(skb)) { | |
492 | /* we have to copy */ | |
493 | local_bh_disable(); | |
fff1ebb2 CL |
494 | if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) |
495 | goto out_bh_enable; | |
1da177e4 | 496 | local_bh_enable(); |
850cbadd | 497 | consume_skb(skb); |
1da177e4 LT |
498 | } else { |
499 | /* we can use it in-place */ | |
1da8c681 | 500 | rqstp->rq_arg.head[0].iov_base = skb->data; |
1da177e4 | 501 | rqstp->rq_arg.head[0].iov_len = len; |
22911fc5 ED |
502 | if (skb_checksum_complete(skb)) |
503 | goto out_free; | |
5148bf4e | 504 | rqstp->rq_xprt_ctxt = skb; |
1da177e4 LT |
505 | } |
506 | ||
507 | rqstp->rq_arg.page_base = 0; | |
508 | if (len <= rqstp->rq_arg.head[0].iov_len) { | |
509 | rqstp->rq_arg.head[0].iov_len = len; | |
510 | rqstp->rq_arg.page_len = 0; | |
44524359 | 511 | rqstp->rq_respages = rqstp->rq_pages+1; |
1da177e4 LT |
512 | } else { |
513 | rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len; | |
44524359 | 514 | rqstp->rq_respages = rqstp->rq_pages + 1 + |
172589cc | 515 | DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE); |
1da177e4 | 516 | } |
afc59400 | 517 | rqstp->rq_next_page = rqstp->rq_respages+1; |
1da177e4 LT |
518 | |
519 | if (serv->sv_stats) | |
520 | serv->sv_stats->netudpcnt++; | |
521 | ||
522 | return len; | |
fff1ebb2 CL |
523 | |
524 | out_recv_err: | |
525 | if (err != -EAGAIN) { | |
526 | /* possibly an icmp error */ | |
527 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); | |
528 | } | |
529 | trace_svcsock_udp_recv_err(&svsk->sk_xprt, err); | |
530 | return 0; | |
531 | out_cmsg_err: | |
532 | net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n", | |
533 | cmh->cmsg_level, cmh->cmsg_type); | |
534 | goto out_free; | |
535 | out_bh_enable: | |
536 | local_bh_enable(); | |
f23abfdb | 537 | out_free: |
850cbadd | 538 | kfree_skb(skb); |
f23abfdb | 539 | return 0; |
1da177e4 LT |
540 | } |
541 | ||
da1661b9 CL |
542 | /** |
543 | * svc_udp_sendto - Send out a reply on a UDP socket | |
544 | * @rqstp: completed svc_rqst | |
545 | * | |
ca4faf54 CL |
546 | * xpt_mutex ensures @rqstp's whole message is written to the socket |
547 | * without interruption. | |
548 | * | |
da1661b9 CL |
549 | * Returns the number of bytes sent, or a negative errno. |
550 | */ | |
551 | static int svc_udp_sendto(struct svc_rqst *rqstp) | |
1da177e4 | 552 | { |
da1661b9 CL |
553 | struct svc_xprt *xprt = rqstp->rq_xprt; |
554 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
555 | struct xdr_buf *xdr = &rqstp->rq_res; | |
556 | union { | |
557 | struct cmsghdr hdr; | |
558 | long all[SVC_PKTINFO_SPACE / sizeof(long)]; | |
559 | } buffer; | |
560 | struct cmsghdr *cmh = &buffer.hdr; | |
561 | struct msghdr msg = { | |
562 | .msg_name = &rqstp->rq_addr, | |
563 | .msg_namelen = rqstp->rq_addrlen, | |
564 | .msg_control = cmh, | |
565 | .msg_controllen = sizeof(buffer), | |
566 | }; | |
3f649ab7 | 567 | unsigned int sent; |
da1661b9 | 568 | int err; |
1da177e4 | 569 | |
cca557a5 | 570 | svc_udp_release_rqst(rqstp); |
23cf1ee1 | 571 | |
da1661b9 | 572 | svc_set_cmsg_data(rqstp, cmh); |
1da177e4 | 573 | |
ca4faf54 CL |
574 | mutex_lock(&xprt->xpt_mutex); |
575 | ||
576 | if (svc_xprt_is_dead(xprt)) | |
577 | goto out_notconn; | |
578 | ||
da1661b9 CL |
579 | err = xprt_sock_sendmsg(svsk->sk_sock, &msg, xdr, 0, 0, &sent); |
580 | xdr_free_bvec(xdr); | |
581 | if (err == -ECONNREFUSED) { | |
582 | /* ICMP error on earlier request. */ | |
583 | err = xprt_sock_sendmsg(svsk->sk_sock, &msg, xdr, 0, 0, &sent); | |
584 | xdr_free_bvec(xdr); | |
585 | } | |
998024de | 586 | trace_svcsock_udp_send(xprt, err); |
ca4faf54 CL |
587 | |
588 | mutex_unlock(&xprt->xpt_mutex); | |
da1661b9 CL |
589 | if (err < 0) |
590 | return err; | |
591 | return sent; | |
ca4faf54 CL |
592 | |
593 | out_notconn: | |
594 | mutex_unlock(&xprt->xpt_mutex); | |
595 | return -ENOTCONN; | |
1da177e4 LT |
596 | } |
597 | ||
323bee32 TT |
598 | static int svc_udp_has_wspace(struct svc_xprt *xprt) |
599 | { | |
600 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
bb5cf160 | 601 | struct svc_serv *serv = xprt->xpt_server; |
323bee32 TT |
602 | unsigned long required; |
603 | ||
604 | /* | |
605 | * Set the SOCK_NOSPACE flag before checking the available | |
606 | * sock space. | |
607 | */ | |
608 | set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
7a90e8cc | 609 | required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg; |
323bee32 TT |
610 | if (required*2 > sock_wspace(svsk->sk_sk)) |
611 | return 0; | |
612 | clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
613 | return 1; | |
614 | } | |
615 | ||
38a417cc TT |
616 | static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt) |
617 | { | |
618 | BUG(); | |
619 | return NULL; | |
620 | } | |
621 | ||
ea08e392 SM |
622 | static void svc_udp_kill_temp_xprt(struct svc_xprt *xprt) |
623 | { | |
624 | } | |
625 | ||
b700cbb1 | 626 | static struct svc_xprt *svc_udp_create(struct svc_serv *serv, |
62832c03 | 627 | struct net *net, |
b700cbb1 TT |
628 | struct sockaddr *sa, int salen, |
629 | int flags) | |
630 | { | |
62832c03 | 631 | return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags); |
b700cbb1 TT |
632 | } |
633 | ||
2412e927 | 634 | static const struct svc_xprt_ops svc_udp_ops = { |
b700cbb1 | 635 | .xpo_create = svc_udp_create, |
5d137990 TT |
636 | .xpo_recvfrom = svc_udp_recvfrom, |
637 | .xpo_sendto = svc_udp_sendto, | |
03493bca | 638 | .xpo_result_payload = svc_sock_result_payload, |
cca557a5 | 639 | .xpo_release_rqst = svc_udp_release_rqst, |
755cceab TT |
640 | .xpo_detach = svc_sock_detach, |
641 | .xpo_free = svc_sock_free, | |
323bee32 | 642 | .xpo_has_wspace = svc_udp_has_wspace, |
38a417cc | 643 | .xpo_accept = svc_udp_accept, |
16e4d93f | 644 | .xpo_secure_port = svc_sock_secure_port, |
ea08e392 | 645 | .xpo_kill_temp_xprt = svc_udp_kill_temp_xprt, |
360d8738 TT |
646 | }; |
647 | ||
648 | static struct svc_xprt_class svc_udp_class = { | |
649 | .xcl_name = "udp", | |
b700cbb1 | 650 | .xcl_owner = THIS_MODULE, |
360d8738 | 651 | .xcl_ops = &svc_udp_ops, |
49023155 | 652 | .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP, |
3c45ddf8 | 653 | .xcl_ident = XPRT_TRANSPORT_UDP, |
360d8738 TT |
654 | }; |
655 | ||
bb5cf160 | 656 | static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv) |
1da177e4 | 657 | { |
bd4620dd SK |
658 | svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class, |
659 | &svsk->sk_xprt, serv); | |
def13d74 | 660 | clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags); |
fa9251af | 661 | svsk->sk_sk->sk_data_ready = svc_data_ready; |
1da177e4 | 662 | svsk->sk_sk->sk_write_space = svc_write_space; |
1da177e4 LT |
663 | |
664 | /* initialise setting must have enough space to | |
cca5172a | 665 | * receive and respond to one request. |
1da177e4 LT |
666 | * svc_udp_recvfrom will re-adjust if necessary |
667 | */ | |
b7e5034c | 668 | svc_sock_setbufsize(svsk, 3); |
1da177e4 | 669 | |
0f0257ea TT |
670 | /* data might have come in before data_ready set up */ |
671 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); | |
02fc6c36 | 672 | set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); |
7a37f578 | 673 | |
7a37f578 | 674 | /* make sure we get destination address info */ |
7702ce40 CL |
675 | switch (svsk->sk_sk->sk_family) { |
676 | case AF_INET: | |
c1f9ec57 | 677 | ip_sock_set_pktinfo(svsk->sk_sock->sk); |
7d7207c2 | 678 | break; |
7702ce40 | 679 | case AF_INET6: |
7d7207c2 | 680 | ip6_sock_set_recvpktinfo(svsk->sk_sock->sk); |
7702ce40 CL |
681 | break; |
682 | default: | |
683 | BUG(); | |
684 | } | |
1da177e4 LT |
685 | } |
686 | ||
687 | /* | |
688 | * A data_ready event on a listening socket means there's a connection | |
689 | * pending. Do not use state_change as a substitute for it. | |
690 | */ | |
676d2369 | 691 | static void svc_tcp_listen_data_ready(struct sock *sk) |
1da177e4 | 692 | { |
939bb7ef | 693 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 | 694 | |
eebe53e8 VL |
695 | if (svsk) { |
696 | /* Refer to svc_setup_socket() for details. */ | |
697 | rmb(); | |
fa9251af | 698 | svsk->sk_odata(sk); |
eebe53e8 VL |
699 | } |
700 | ||
939bb7ef NB |
701 | /* |
702 | * This callback may called twice when a new connection | |
703 | * is established as a child socket inherits everything | |
704 | * from a parent LISTEN socket. | |
705 | * 1) data_ready method of the parent socket will be called | |
706 | * when one of child sockets become ESTABLISHED. | |
707 | * 2) data_ready method of the child socket may be called | |
708 | * when it receives data before the socket is accepted. | |
709 | * In case of 2, we should ignore it silently. | |
710 | */ | |
711 | if (sk->sk_state == TCP_LISTEN) { | |
712 | if (svsk) { | |
02fc6c36 | 713 | set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
f6150c3c | 714 | svc_xprt_enqueue(&svsk->sk_xprt); |
a0469f46 | 715 | } |
1da177e4 | 716 | } |
1da177e4 LT |
717 | } |
718 | ||
719 | /* | |
720 | * A state change on a connected socket means it's dying or dead. | |
721 | */ | |
0f0257ea | 722 | static void svc_tcp_state_change(struct sock *sk) |
1da177e4 | 723 | { |
939bb7ef | 724 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 | 725 | |
a0469f46 | 726 | if (svsk) { |
eebe53e8 VL |
727 | /* Refer to svc_setup_socket() for details. */ |
728 | rmb(); | |
fa9251af | 729 | svsk->sk_ostate(sk); |
a0469f46 | 730 | trace_svcsock_tcp_state(&svsk->sk_xprt, svsk->sk_sock); |
b2f21f7d TM |
731 | if (sk->sk_state != TCP_ESTABLISHED) { |
732 | set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); | |
733 | svc_xprt_enqueue(&svsk->sk_xprt); | |
734 | } | |
1da177e4 | 735 | } |
1da177e4 LT |
736 | } |
737 | ||
738 | /* | |
739 | * Accept a TCP connection | |
740 | */ | |
38a417cc | 741 | static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt) |
1da177e4 | 742 | { |
38a417cc | 743 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); |
cdd88b9f | 744 | struct sockaddr_storage addr; |
745 | struct sockaddr *sin = (struct sockaddr *) &addr; | |
bb5cf160 | 746 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
1da177e4 LT |
747 | struct socket *sock = svsk->sk_sock; |
748 | struct socket *newsock; | |
1da177e4 LT |
749 | struct svc_sock *newsvsk; |
750 | int err, slen; | |
751 | ||
1da177e4 | 752 | if (!sock) |
38a417cc | 753 | return NULL; |
1da177e4 | 754 | |
02fc6c36 | 755 | clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
e6242e92 SS |
756 | err = kernel_accept(sock, &newsock, O_NONBLOCK); |
757 | if (err < 0) { | |
1da177e4 LT |
758 | if (err == -ENOMEM) |
759 | printk(KERN_WARNING "%s: no more sockets!\n", | |
760 | serv->sv_name); | |
e87cc472 JP |
761 | else if (err != -EAGAIN) |
762 | net_warn_ratelimited("%s: accept failed (err %d)!\n", | |
763 | serv->sv_name, -err); | |
a0469f46 | 764 | trace_svcsock_accept_err(xprt, serv->sv_name, err); |
38a417cc | 765 | return NULL; |
1da177e4 | 766 | } |
02fc6c36 | 767 | set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 768 | |
9b2c45d4 | 769 | err = kernel_getpeername(newsock, sin); |
1da177e4 | 770 | if (err < 0) { |
a0469f46 | 771 | trace_svcsock_getpeername_err(xprt, serv->sv_name, err); |
1da177e4 LT |
772 | goto failed; /* aborted connection or whatever */ |
773 | } | |
9b2c45d4 | 774 | slen = err; |
1da177e4 | 775 | |
fa9251af TM |
776 | /* Reset the inherited callbacks before calling svc_setup_socket */ |
777 | newsock->sk->sk_state_change = svsk->sk_ostate; | |
778 | newsock->sk->sk_data_ready = svsk->sk_odata; | |
779 | newsock->sk->sk_write_space = svsk->sk_owspace; | |
780 | ||
1da177e4 LT |
781 | /* make sure that a write doesn't block forever when |
782 | * low on memory | |
783 | */ | |
784 | newsock->sk->sk_sndtimeo = HZ*30; | |
785 | ||
72c35376 BF |
786 | newsvsk = svc_setup_socket(serv, newsock, |
787 | (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY)); | |
788 | if (IS_ERR(newsvsk)) | |
1da177e4 | 789 | goto failed; |
9dbc240f | 790 | svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen); |
9b2c45d4 DV |
791 | err = kernel_getsockname(newsock, sin); |
792 | slen = err; | |
a0469f46 | 793 | if (unlikely(err < 0)) |
a9747692 | 794 | slen = offsetof(struct sockaddr, sa_data); |
9dbc240f | 795 | svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen); |
067d7817 | 796 | |
ef11ce24 N |
797 | if (sock_is_loopback(newsock->sk)) |
798 | set_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags); | |
799 | else | |
800 | clear_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags); | |
f9f3cc4f TT |
801 | if (serv->sv_stats) |
802 | serv->sv_stats->nettcpconn++; | |
803 | ||
804 | return &newsvsk->sk_xprt; | |
805 | ||
806 | failed: | |
807 | sock_release(newsock); | |
808 | return NULL; | |
809 | } | |
810 | ||
ca07eda3 CL |
811 | static size_t svc_tcp_restore_pages(struct svc_sock *svsk, |
812 | struct svc_rqst *rqstp) | |
31d68ef6 | 813 | { |
ca07eda3 CL |
814 | size_t len = svsk->sk_datalen; |
815 | unsigned int i, npages; | |
31d68ef6 | 816 | |
ca07eda3 | 817 | if (!len) |
31d68ef6 | 818 | return 0; |
31d68ef6 BF |
819 | npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
820 | for (i = 0; i < npages; i++) { | |
821 | if (rqstp->rq_pages[i] != NULL) | |
822 | put_page(rqstp->rq_pages[i]); | |
823 | BUG_ON(svsk->sk_pages[i] == NULL); | |
824 | rqstp->rq_pages[i] = svsk->sk_pages[i]; | |
825 | svsk->sk_pages[i] = NULL; | |
826 | } | |
827 | rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]); | |
828 | return len; | |
829 | } | |
830 | ||
831 | static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp) | |
832 | { | |
833 | unsigned int i, len, npages; | |
834 | ||
8af345f5 | 835 | if (svsk->sk_datalen == 0) |
31d68ef6 | 836 | return; |
8af345f5 | 837 | len = svsk->sk_datalen; |
31d68ef6 BF |
838 | npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
839 | for (i = 0; i < npages; i++) { | |
840 | svsk->sk_pages[i] = rqstp->rq_pages[i]; | |
841 | rqstp->rq_pages[i] = NULL; | |
842 | } | |
843 | } | |
844 | ||
845 | static void svc_tcp_clear_pages(struct svc_sock *svsk) | |
846 | { | |
847 | unsigned int i, len, npages; | |
848 | ||
8af345f5 | 849 | if (svsk->sk_datalen == 0) |
31d68ef6 | 850 | goto out; |
8af345f5 | 851 | len = svsk->sk_datalen; |
31d68ef6 BF |
852 | npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
853 | for (i = 0; i < npages; i++) { | |
cf3aa02c BF |
854 | if (svsk->sk_pages[i] == NULL) { |
855 | WARN_ON_ONCE(1); | |
856 | continue; | |
857 | } | |
31d68ef6 BF |
858 | put_page(svsk->sk_pages[i]); |
859 | svsk->sk_pages[i] = NULL; | |
860 | } | |
861 | out: | |
862 | svsk->sk_tcplen = 0; | |
8af345f5 | 863 | svsk->sk_datalen = 0; |
31d68ef6 BF |
864 | } |
865 | ||
1da177e4 | 866 | /* |
a5cda73e | 867 | * Receive fragment record header into sk_marker. |
1da177e4 | 868 | */ |
a5cda73e CL |
869 | static ssize_t svc_tcp_read_marker(struct svc_sock *svsk, |
870 | struct svc_rqst *rqstp) | |
1da177e4 | 871 | { |
a5cda73e | 872 | ssize_t want, len; |
1da177e4 | 873 | |
a5cda73e CL |
874 | /* If we haven't gotten the record length yet, |
875 | * get the next four bytes. | |
876 | */ | |
c0401ea0 | 877 | if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) { |
a5cda73e | 878 | struct msghdr msg = { NULL }; |
1da177e4 LT |
879 | struct kvec iov; |
880 | ||
5ee78d48 | 881 | want = sizeof(rpc_fraghdr) - svsk->sk_tcplen; |
02648908 | 882 | iov.iov_base = ((char *)&svsk->sk_marker) + svsk->sk_tcplen; |
1da177e4 | 883 | iov.iov_len = want; |
a5cda73e CL |
884 | iov_iter_kvec(&msg.msg_iter, READ, &iov, 1, want); |
885 | len = sock_recvmsg(svsk->sk_sock, &msg, MSG_DONTWAIT); | |
4c8e5537 | 886 | if (len < 0) |
a5cda73e | 887 | return len; |
1da177e4 | 888 | svsk->sk_tcplen += len; |
1da177e4 | 889 | if (len < want) { |
a5cda73e CL |
890 | /* call again to read the remaining bytes */ |
891 | goto err_short; | |
1da177e4 | 892 | } |
a5cda73e | 893 | trace_svcsock_marker(&svsk->sk_xprt, svsk->sk_marker); |
836fbadb | 894 | if (svc_sock_reclen(svsk) + svsk->sk_datalen > |
a5cda73e CL |
895 | svsk->sk_xprt.xpt_server->sv_max_mesg) |
896 | goto err_too_large; | |
1da177e4 | 897 | } |
cc248d4b | 898 | return svc_sock_reclen(svsk); |
a5cda73e CL |
899 | |
900 | err_too_large: | |
901 | net_notice_ratelimited("svc: %s %s RPC fragment too large: %d\n", | |
902 | __func__, svsk->sk_xprt.xpt_server->sv_name, | |
903 | svc_sock_reclen(svsk)); | |
8f55f3c0 | 904 | set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); |
a5cda73e | 905 | err_short: |
8f55f3c0 AB |
906 | return -EAGAIN; |
907 | } | |
908 | ||
586c52cc | 909 | static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp) |
4cfc7e60 | 910 | { |
586c52cc | 911 | struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt; |
4cfc7e60 | 912 | struct rpc_rqst *req = NULL; |
586c52cc TM |
913 | struct kvec *src, *dst; |
914 | __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base; | |
48e6555c BF |
915 | __be32 xid; |
916 | __be32 calldir; | |
4cfc7e60 | 917 | |
4cfc7e60 RI |
918 | xid = *p++; |
919 | calldir = *p; | |
920 | ||
093a1468 | 921 | if (!bc_xprt) |
586c52cc | 922 | return -EAGAIN; |
75c84151 | 923 | spin_lock(&bc_xprt->queue_lock); |
093a1468 TM |
924 | req = xprt_lookup_rqst(bc_xprt, xid); |
925 | if (!req) | |
926 | goto unlock_notfound; | |
586c52cc TM |
927 | |
928 | memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf)); | |
929 | /* | |
930 | * XXX!: cheating for now! Only copying HEAD. | |
931 | * But we know this is good enough for now (in fact, for any | |
932 | * callback reply in the forseeable future). | |
933 | */ | |
934 | dst = &req->rq_private_buf.head[0]; | |
935 | src = &rqstp->rq_arg.head[0]; | |
936 | if (dst->iov_len < src->iov_len) | |
093a1468 | 937 | goto unlock_eagain; /* whatever; just giving up. */ |
586c52cc | 938 | memcpy(dst->iov_base, src->iov_base, src->iov_len); |
cc248d4b | 939 | xprt_complete_rqst(req->rq_task, rqstp->rq_arg.len); |
586c52cc | 940 | rqstp->rq_arg.len = 0; |
75c84151 | 941 | spin_unlock(&bc_xprt->queue_lock); |
586c52cc | 942 | return 0; |
093a1468 TM |
943 | unlock_notfound: |
944 | printk(KERN_NOTICE | |
945 | "%s: Got unrecognized reply: " | |
946 | "calldir 0x%x xpt_bc_xprt %p xid %08x\n", | |
947 | __func__, ntohl(calldir), | |
948 | bc_xprt, ntohl(xid)); | |
949 | unlock_eagain: | |
75c84151 | 950 | spin_unlock(&bc_xprt->queue_lock); |
093a1468 | 951 | return -EAGAIN; |
586c52cc TM |
952 | } |
953 | ||
836fbadb BF |
954 | static void svc_tcp_fragment_received(struct svc_sock *svsk) |
955 | { | |
956 | /* If we have more data, signal svc_xprt_enqueue() to try again */ | |
836fbadb | 957 | svsk->sk_tcplen = 0; |
02648908 | 958 | svsk->sk_marker = xdr_zero; |
836fbadb | 959 | } |
31d68ef6 | 960 | |
ca07eda3 CL |
961 | /** |
962 | * svc_tcp_recvfrom - Receive data from a TCP socket | |
963 | * @rqstp: request structure into which to receive an RPC Call | |
964 | * | |
965 | * Called in a loop when XPT_DATA has been set. | |
966 | * | |
967 | * Read the 4-byte stream record marker, then use the record length | |
968 | * in that marker to set up exactly the resources needed to receive | |
969 | * the next RPC message into @rqstp. | |
970 | * | |
971 | * Returns: | |
972 | * On success, the number of bytes in a received RPC Call, or | |
973 | * %0 if a complete RPC Call message was not ready to return | |
974 | * | |
975 | * The zero return case handles partial receives and callback Replies. | |
976 | * The state of a partial receive is preserved in the svc_sock for | |
977 | * the next call to svc_tcp_recvfrom. | |
8f55f3c0 AB |
978 | */ |
979 | static int svc_tcp_recvfrom(struct svc_rqst *rqstp) | |
980 | { | |
981 | struct svc_sock *svsk = | |
982 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
983 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; | |
ca07eda3 CL |
984 | size_t want, base; |
985 | ssize_t len; | |
586c52cc TM |
986 | __be32 *p; |
987 | __be32 calldir; | |
8f55f3c0 | 988 | |
a5cda73e CL |
989 | clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
990 | len = svc_tcp_read_marker(svsk, rqstp); | |
8f55f3c0 AB |
991 | if (len < 0) |
992 | goto error; | |
993 | ||
31d68ef6 | 994 | base = svc_tcp_restore_pages(svsk, rqstp); |
a5cda73e | 995 | want = len - (svsk->sk_tcplen - sizeof(rpc_fraghdr)); |
ca07eda3 | 996 | len = svc_tcp_read_msg(rqstp, base + want, base); |
8af345f5 | 997 | if (len >= 0) { |
7dae1dd7 | 998 | trace_svcsock_tcp_recv(&svsk->sk_xprt, len); |
31d68ef6 | 999 | svsk->sk_tcplen += len; |
8af345f5 BF |
1000 | svsk->sk_datalen += len; |
1001 | } | |
6be8c594 CL |
1002 | if (len != want || !svc_sock_final_rec(svsk)) |
1003 | goto err_incomplete; | |
1004 | if (svsk->sk_datalen < 8) | |
1005 | goto err_nuts; | |
ad46ccf0 | 1006 | |
836fbadb | 1007 | rqstp->rq_arg.len = svsk->sk_datalen; |
1da177e4 | 1008 | rqstp->rq_arg.page_base = 0; |
5ee78d48 TM |
1009 | if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) { |
1010 | rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len; | |
1da177e4 | 1011 | rqstp->rq_arg.page_len = 0; |
5ee78d48 TM |
1012 | } else |
1013 | rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len; | |
1da177e4 | 1014 | |
5148bf4e | 1015 | rqstp->rq_xprt_ctxt = NULL; |
1da177e4 | 1016 | rqstp->rq_prot = IPPROTO_TCP; |
7501cc2b JL |
1017 | if (test_bit(XPT_LOCAL, &svsk->sk_xprt.xpt_flags)) |
1018 | set_bit(RQ_LOCAL, &rqstp->rq_flags); | |
1019 | else | |
1020 | clear_bit(RQ_LOCAL, &rqstp->rq_flags); | |
1da177e4 | 1021 | |
586c52cc TM |
1022 | p = (__be32 *)rqstp->rq_arg.head[0].iov_base; |
1023 | calldir = p[1]; | |
8985ef0b | 1024 | if (calldir) |
586c52cc | 1025 | len = receive_cb_reply(svsk, rqstp); |
586c52cc | 1026 | |
1da177e4 | 1027 | /* Reset TCP read info */ |
8af345f5 | 1028 | svsk->sk_datalen = 0; |
836fbadb | 1029 | svc_tcp_fragment_received(svsk); |
0601f793 | 1030 | |
8985ef0b BF |
1031 | if (len < 0) |
1032 | goto error; | |
1da177e4 | 1033 | |
9dbc240f | 1034 | svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt); |
1da177e4 LT |
1035 | if (serv->sv_stats) |
1036 | serv->sv_stats->nettcpcnt++; | |
1037 | ||
5ee78d48 | 1038 | return rqstp->rq_arg.len; |
1da177e4 | 1039 | |
6be8c594 CL |
1040 | err_incomplete: |
1041 | svc_tcp_save_pages(svsk, rqstp); | |
1042 | if (len < 0 && len != -EAGAIN) | |
1043 | goto err_delete; | |
1044 | if (len == want) | |
1045 | svc_tcp_fragment_received(svsk); | |
1046 | else | |
1047 | trace_svcsock_tcp_recv_short(&svsk->sk_xprt, | |
1048 | svc_sock_reclen(svsk), | |
1049 | svsk->sk_tcplen - sizeof(rpc_fraghdr)); | |
1050 | goto err_noclose; | |
8f55f3c0 | 1051 | error: |
31d68ef6 | 1052 | if (len != -EAGAIN) |
ad46ccf0 | 1053 | goto err_delete; |
6be8c594 | 1054 | trace_svcsock_tcp_recv_eagain(&svsk->sk_xprt, 0); |
9f9d2ebe | 1055 | return 0; |
6be8c594 CL |
1056 | err_nuts: |
1057 | svsk->sk_datalen = 0; | |
ad46ccf0 | 1058 | err_delete: |
6be8c594 | 1059 | trace_svcsock_tcp_recv_err(&svsk->sk_xprt, len); |
31d68ef6 BF |
1060 | set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); |
1061 | err_noclose: | |
9f9d2ebe | 1062 | return 0; /* record not complete */ |
1da177e4 LT |
1063 | } |
1064 | ||
4a85a6a3 CL |
1065 | static int svc_tcp_send_kvec(struct socket *sock, const struct kvec *vec, |
1066 | int flags) | |
1067 | { | |
1068 | return kernel_sendpage(sock, virt_to_page(vec->iov_base), | |
1069 | offset_in_page(vec->iov_base), | |
1070 | vec->iov_len, flags); | |
1071 | } | |
1072 | ||
1073 | /* | |
1074 | * kernel_sendpage() is used exclusively to reduce the number of | |
1075 | * copy operations in this path. Therefore the caller must ensure | |
1076 | * that the pages backing @xdr are unchanging. | |
1077 | * | |
1078 | * In addition, the logic assumes that * .bv_len is never larger | |
1079 | * than PAGE_SIZE. | |
1080 | */ | |
4d12b727 CL |
1081 | static int svc_tcp_sendmsg(struct socket *sock, struct xdr_buf *xdr, |
1082 | rpc_fraghdr marker, unsigned int *sentp) | |
4a85a6a3 CL |
1083 | { |
1084 | const struct kvec *head = xdr->head; | |
1085 | const struct kvec *tail = xdr->tail; | |
1086 | struct kvec rm = { | |
1087 | .iov_base = &marker, | |
1088 | .iov_len = sizeof(marker), | |
1089 | }; | |
4d12b727 CL |
1090 | struct msghdr msg = { |
1091 | .msg_flags = 0, | |
1092 | }; | |
987c7b1d | 1093 | int ret; |
4a85a6a3 CL |
1094 | |
1095 | *sentp = 0; | |
1096 | xdr_alloc_bvec(xdr, GFP_KERNEL); | |
1097 | ||
4d12b727 | 1098 | ret = kernel_sendmsg(sock, &msg, &rm, 1, rm.iov_len); |
4a85a6a3 CL |
1099 | if (ret < 0) |
1100 | return ret; | |
1101 | *sentp += ret; | |
1102 | if (ret != rm.iov_len) | |
1103 | return -EAGAIN; | |
1104 | ||
987c7b1d | 1105 | ret = svc_tcp_send_kvec(sock, head, 0); |
4a85a6a3 CL |
1106 | if (ret < 0) |
1107 | return ret; | |
1108 | *sentp += ret; | |
1109 | if (ret != head->iov_len) | |
1110 | goto out; | |
1111 | ||
1112 | if (xdr->page_len) { | |
1113 | unsigned int offset, len, remaining; | |
1114 | struct bio_vec *bvec; | |
1115 | ||
bad4c6eb CL |
1116 | bvec = xdr->bvec + (xdr->page_base >> PAGE_SHIFT); |
1117 | offset = offset_in_page(xdr->page_base); | |
4a85a6a3 | 1118 | remaining = xdr->page_len; |
4a85a6a3 | 1119 | while (remaining > 0) { |
bad4c6eb | 1120 | len = min(remaining, bvec->bv_len - offset); |
4a85a6a3 CL |
1121 | ret = kernel_sendpage(sock, bvec->bv_page, |
1122 | bvec->bv_offset + offset, | |
987c7b1d | 1123 | len, 0); |
4a85a6a3 CL |
1124 | if (ret < 0) |
1125 | return ret; | |
1126 | *sentp += ret; | |
1127 | if (ret != len) | |
1128 | goto out; | |
1129 | remaining -= len; | |
1130 | offset = 0; | |
1131 | bvec++; | |
1132 | } | |
1133 | } | |
1134 | ||
1135 | if (tail->iov_len) { | |
1136 | ret = svc_tcp_send_kvec(sock, tail, 0); | |
1137 | if (ret < 0) | |
1138 | return ret; | |
1139 | *sentp += ret; | |
1140 | } | |
1141 | ||
1142 | out: | |
1143 | return 0; | |
1144 | } | |
1145 | ||
da1661b9 CL |
1146 | /** |
1147 | * svc_tcp_sendto - Send out a reply on a TCP socket | |
1148 | * @rqstp: completed svc_rqst | |
1149 | * | |
ca4faf54 CL |
1150 | * xpt_mutex ensures @rqstp's whole message is written to the socket |
1151 | * without interruption. | |
1152 | * | |
da1661b9 | 1153 | * Returns the number of bytes sent, or a negative errno. |
1da177e4 | 1154 | */ |
0f0257ea | 1155 | static int svc_tcp_sendto(struct svc_rqst *rqstp) |
1da177e4 | 1156 | { |
da1661b9 CL |
1157 | struct svc_xprt *xprt = rqstp->rq_xprt; |
1158 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
1159 | struct xdr_buf *xdr = &rqstp->rq_res; | |
1160 | rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | | |
1161 | (u32)xdr->len); | |
3f649ab7 | 1162 | unsigned int sent; |
da1661b9 | 1163 | int err; |
1da177e4 | 1164 | |
cca557a5 | 1165 | svc_tcp_release_rqst(rqstp); |
23cf1ee1 | 1166 | |
e0a912e8 | 1167 | atomic_inc(&svsk->sk_sendqlen); |
ca4faf54 CL |
1168 | mutex_lock(&xprt->xpt_mutex); |
1169 | if (svc_xprt_is_dead(xprt)) | |
1170 | goto out_notconn; | |
e0a912e8 | 1171 | tcp_sock_set_cork(svsk->sk_sk, true); |
4d12b727 | 1172 | err = svc_tcp_sendmsg(svsk->sk_sock, xdr, marker, &sent); |
da1661b9 | 1173 | xdr_free_bvec(xdr); |
998024de | 1174 | trace_svcsock_tcp_send(xprt, err < 0 ? err : sent); |
da1661b9 CL |
1175 | if (err < 0 || sent != (xdr->len + sizeof(marker))) |
1176 | goto out_close; | |
e0a912e8 TM |
1177 | if (atomic_dec_and_test(&svsk->sk_sendqlen)) |
1178 | tcp_sock_set_cork(svsk->sk_sk, false); | |
ca4faf54 | 1179 | mutex_unlock(&xprt->xpt_mutex); |
1da177e4 | 1180 | return sent; |
da1661b9 | 1181 | |
ca4faf54 | 1182 | out_notconn: |
e0a912e8 | 1183 | atomic_dec(&svsk->sk_sendqlen); |
ca4faf54 CL |
1184 | mutex_unlock(&xprt->xpt_mutex); |
1185 | return -ENOTCONN; | |
da1661b9 CL |
1186 | out_close: |
1187 | pr_notice("rpc-srv/tcp: %s: %s %d when sending %d bytes - shutting down socket\n", | |
1188 | xprt->xpt_server->sv_name, | |
1189 | (err < 0) ? "got error" : "sent", | |
1190 | (err < 0) ? err : sent, xdr->len); | |
1191 | set_bit(XPT_CLOSE, &xprt->xpt_flags); | |
1192 | svc_xprt_enqueue(xprt); | |
e0a912e8 | 1193 | atomic_dec(&svsk->sk_sendqlen); |
ca4faf54 | 1194 | mutex_unlock(&xprt->xpt_mutex); |
da1661b9 | 1195 | return -EAGAIN; |
1da177e4 LT |
1196 | } |
1197 | ||
b700cbb1 | 1198 | static struct svc_xprt *svc_tcp_create(struct svc_serv *serv, |
62832c03 | 1199 | struct net *net, |
b700cbb1 TT |
1200 | struct sockaddr *sa, int salen, |
1201 | int flags) | |
1202 | { | |
62832c03 | 1203 | return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags); |
b700cbb1 TT |
1204 | } |
1205 | ||
2412e927 | 1206 | static const struct svc_xprt_ops svc_tcp_ops = { |
b700cbb1 | 1207 | .xpo_create = svc_tcp_create, |
5d137990 TT |
1208 | .xpo_recvfrom = svc_tcp_recvfrom, |
1209 | .xpo_sendto = svc_tcp_sendto, | |
03493bca | 1210 | .xpo_result_payload = svc_sock_result_payload, |
cca557a5 | 1211 | .xpo_release_rqst = svc_tcp_release_rqst, |
69b6ba37 | 1212 | .xpo_detach = svc_tcp_sock_detach, |
755cceab | 1213 | .xpo_free = svc_sock_free, |
323bee32 | 1214 | .xpo_has_wspace = svc_tcp_has_wspace, |
38a417cc | 1215 | .xpo_accept = svc_tcp_accept, |
16e4d93f | 1216 | .xpo_secure_port = svc_sock_secure_port, |
ea08e392 | 1217 | .xpo_kill_temp_xprt = svc_tcp_kill_temp_xprt, |
360d8738 TT |
1218 | }; |
1219 | ||
1220 | static struct svc_xprt_class svc_tcp_class = { | |
1221 | .xcl_name = "tcp", | |
b700cbb1 | 1222 | .xcl_owner = THIS_MODULE, |
360d8738 | 1223 | .xcl_ops = &svc_tcp_ops, |
49023155 | 1224 | .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP, |
3c45ddf8 | 1225 | .xcl_ident = XPRT_TRANSPORT_TCP, |
360d8738 TT |
1226 | }; |
1227 | ||
1228 | void svc_init_xprt_sock(void) | |
1229 | { | |
1230 | svc_reg_xprt_class(&svc_tcp_class); | |
1231 | svc_reg_xprt_class(&svc_udp_class); | |
1232 | } | |
1233 | ||
1234 | void svc_cleanup_xprt_sock(void) | |
1235 | { | |
1236 | svc_unreg_xprt_class(&svc_tcp_class); | |
1237 | svc_unreg_xprt_class(&svc_udp_class); | |
1238 | } | |
1239 | ||
bb5cf160 | 1240 | static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv) |
1da177e4 LT |
1241 | { |
1242 | struct sock *sk = svsk->sk_sk; | |
1da177e4 | 1243 | |
bd4620dd SK |
1244 | svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class, |
1245 | &svsk->sk_xprt, serv); | |
def13d74 | 1246 | set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags); |
362142b2 | 1247 | set_bit(XPT_CONG_CTRL, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 1248 | if (sk->sk_state == TCP_LISTEN) { |
ece200dd | 1249 | strcpy(svsk->sk_xprt.xpt_remotebuf, "listener"); |
02fc6c36 | 1250 | set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 1251 | sk->sk_data_ready = svc_tcp_listen_data_ready; |
02fc6c36 | 1252 | set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 1253 | } else { |
1da177e4 | 1254 | sk->sk_state_change = svc_tcp_state_change; |
fa9251af | 1255 | sk->sk_data_ready = svc_data_ready; |
637600f3 | 1256 | sk->sk_write_space = svc_write_space; |
1da177e4 | 1257 | |
02648908 | 1258 | svsk->sk_marker = xdr_zero; |
1da177e4 | 1259 | svsk->sk_tcplen = 0; |
8af345f5 | 1260 | svsk->sk_datalen = 0; |
31d68ef6 | 1261 | memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages)); |
1da177e4 | 1262 | |
e0a912e8 | 1263 | tcp_sock_set_nodelay(sk); |
1da177e4 | 1264 | |
02fc6c36 | 1265 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
b2f21f7d TM |
1266 | switch (sk->sk_state) { |
1267 | case TCP_SYN_RECV: | |
1268 | case TCP_ESTABLISHED: | |
1269 | break; | |
1270 | default: | |
02fc6c36 | 1271 | set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); |
b2f21f7d | 1272 | } |
1da177e4 LT |
1273 | } |
1274 | } | |
1275 | ||
0f0257ea | 1276 | void svc_sock_update_bufs(struct svc_serv *serv) |
1da177e4 LT |
1277 | { |
1278 | /* | |
1279 | * The number of server threads has changed. Update | |
1280 | * rcvbuf and sndbuf accordingly on all sockets | |
1281 | */ | |
8f3a6de3 | 1282 | struct svc_sock *svsk; |
1da177e4 LT |
1283 | |
1284 | spin_lock_bh(&serv->sv_lock); | |
8f3a6de3 | 1285 | list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list) |
02fc6c36 | 1286 | set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
1287 | spin_unlock_bh(&serv->sv_lock); |
1288 | } | |
24c3767e | 1289 | EXPORT_SYMBOL_GPL(svc_sock_update_bufs); |
1da177e4 | 1290 | |
1da177e4 LT |
1291 | /* |
1292 | * Initialize socket for RPC use and create svc_sock struct | |
1da177e4 | 1293 | */ |
6b174337 CL |
1294 | static struct svc_sock *svc_setup_socket(struct svc_serv *serv, |
1295 | struct socket *sock, | |
72c35376 | 1296 | int flags) |
1da177e4 LT |
1297 | { |
1298 | struct svc_sock *svsk; | |
1299 | struct sock *inet; | |
6b174337 | 1300 | int pmap_register = !(flags & SVC_SOCK_ANONYMOUS); |
72c35376 | 1301 | int err = 0; |
1da177e4 | 1302 | |
72c35376 BF |
1303 | svsk = kzalloc(sizeof(*svsk), GFP_KERNEL); |
1304 | if (!svsk) | |
1305 | return ERR_PTR(-ENOMEM); | |
1da177e4 LT |
1306 | |
1307 | inet = sock->sk; | |
1308 | ||
1309 | /* Register socket with portmapper */ | |
72c35376 BF |
1310 | if (pmap_register) |
1311 | err = svc_register(serv, sock_net(sock->sk), inet->sk_family, | |
5247fab5 | 1312 | inet->sk_protocol, |
c720c7e8 | 1313 | ntohs(inet_sk(inet)->inet_sport)); |
1da177e4 | 1314 | |
72c35376 | 1315 | if (err < 0) { |
1da177e4 | 1316 | kfree(svsk); |
72c35376 | 1317 | return ERR_PTR(err); |
1da177e4 LT |
1318 | } |
1319 | ||
1da177e4 LT |
1320 | svsk->sk_sock = sock; |
1321 | svsk->sk_sk = inet; | |
1322 | svsk->sk_ostate = inet->sk_state_change; | |
1323 | svsk->sk_odata = inet->sk_data_ready; | |
1324 | svsk->sk_owspace = inet->sk_write_space; | |
eebe53e8 VL |
1325 | /* |
1326 | * This barrier is necessary in order to prevent race condition | |
1327 | * with svc_data_ready(), svc_listen_data_ready() and others | |
1328 | * when calling callbacks above. | |
1329 | */ | |
1330 | wmb(); | |
1331 | inet->sk_user_data = svsk; | |
1da177e4 LT |
1332 | |
1333 | /* Initialize the socket */ | |
1334 | if (sock->type == SOCK_DGRAM) | |
bb5cf160 | 1335 | svc_udp_init(svsk, serv); |
637600f3 | 1336 | else |
bb5cf160 | 1337 | svc_tcp_init(svsk, serv); |
1da177e4 | 1338 | |
998024de | 1339 | trace_svcsock_new_socket(sock); |
1da177e4 LT |
1340 | return svsk; |
1341 | } | |
1342 | ||
30646394 SK |
1343 | bool svc_alien_sock(struct net *net, int fd) |
1344 | { | |
1345 | int err; | |
1346 | struct socket *sock = sockfd_lookup(fd, &err); | |
1347 | bool ret = false; | |
1348 | ||
1349 | if (!sock) | |
1350 | goto out; | |
1351 | if (sock_net(sock->sk) != net) | |
1352 | ret = true; | |
1353 | sockfd_put(sock); | |
1354 | out: | |
1355 | return ret; | |
1356 | } | |
1357 | EXPORT_SYMBOL_GPL(svc_alien_sock); | |
1358 | ||
bfba9ab4 CL |
1359 | /** |
1360 | * svc_addsock - add a listener socket to an RPC service | |
1361 | * @serv: pointer to RPC service to which to add a new listener | |
1362 | * @fd: file descriptor of the new listener | |
1363 | * @name_return: pointer to buffer to fill in with name of listener | |
1364 | * @len: size of the buffer | |
4df493a2 | 1365 | * @cred: credential |
bfba9ab4 CL |
1366 | * |
1367 | * Fills in socket name and returns positive length of name if successful. | |
1368 | * Name is terminated with '\n'. On error, returns a negative errno | |
1369 | * value. | |
1370 | */ | |
1371 | int svc_addsock(struct svc_serv *serv, const int fd, char *name_return, | |
4df493a2 | 1372 | const size_t len, const struct cred *cred) |
b41b66d6 N |
1373 | { |
1374 | int err = 0; | |
1375 | struct socket *so = sockfd_lookup(fd, &err); | |
1376 | struct svc_sock *svsk = NULL; | |
a8e10078 BF |
1377 | struct sockaddr_storage addr; |
1378 | struct sockaddr *sin = (struct sockaddr *)&addr; | |
1379 | int salen; | |
b41b66d6 N |
1380 | |
1381 | if (!so) | |
1382 | return err; | |
a8e10078 | 1383 | err = -EAFNOSUPPORT; |
205ba423 | 1384 | if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6)) |
a8e10078 BF |
1385 | goto out; |
1386 | err = -EPROTONOSUPPORT; | |
1387 | if (so->sk->sk_protocol != IPPROTO_TCP && | |
b41b66d6 | 1388 | so->sk->sk_protocol != IPPROTO_UDP) |
a8e10078 BF |
1389 | goto out; |
1390 | err = -EISCONN; | |
1391 | if (so->state > SS_UNCONNECTED) | |
1392 | goto out; | |
1393 | err = -ENOENT; | |
1394 | if (!try_module_get(THIS_MODULE)) | |
1395 | goto out; | |
1396 | svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS); | |
1397 | if (IS_ERR(svsk)) { | |
1398 | module_put(THIS_MODULE); | |
1399 | err = PTR_ERR(svsk); | |
1400 | goto out; | |
b41b66d6 | 1401 | } |
9b2c45d4 DV |
1402 | salen = kernel_getsockname(svsk->sk_sock, sin); |
1403 | if (salen >= 0) | |
a8e10078 | 1404 | svc_xprt_set_local(&svsk->sk_xprt, sin, salen); |
4df493a2 | 1405 | svsk->sk_xprt.xpt_cred = get_cred(cred); |
39b55301 | 1406 | svc_add_new_perm_xprt(serv, &svsk->sk_xprt); |
e7942b9f | 1407 | return svc_one_sock_name(svsk, name_return, len); |
a8e10078 BF |
1408 | out: |
1409 | sockfd_put(so); | |
1410 | return err; | |
b41b66d6 N |
1411 | } |
1412 | EXPORT_SYMBOL_GPL(svc_addsock); | |
1413 | ||
1da177e4 LT |
1414 | /* |
1415 | * Create socket for RPC service. | |
1416 | */ | |
b700cbb1 TT |
1417 | static struct svc_xprt *svc_create_socket(struct svc_serv *serv, |
1418 | int protocol, | |
62832c03 | 1419 | struct net *net, |
b700cbb1 TT |
1420 | struct sockaddr *sin, int len, |
1421 | int flags) | |
1da177e4 LT |
1422 | { |
1423 | struct svc_sock *svsk; | |
1424 | struct socket *sock; | |
1425 | int error; | |
1426 | int type; | |
9dbc240f TT |
1427 | struct sockaddr_storage addr; |
1428 | struct sockaddr *newsin = (struct sockaddr *)&addr; | |
1429 | int newlen; | |
c69da774 | 1430 | int family; |
1da177e4 LT |
1431 | |
1432 | if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) { | |
1433 | printk(KERN_WARNING "svc: only UDP and TCP " | |
1434 | "sockets supported\n"); | |
b700cbb1 | 1435 | return ERR_PTR(-EINVAL); |
1da177e4 | 1436 | } |
c69da774 | 1437 | |
1da177e4 | 1438 | type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM; |
c69da774 TM |
1439 | switch (sin->sa_family) { |
1440 | case AF_INET6: | |
1441 | family = PF_INET6; | |
1442 | break; | |
1443 | case AF_INET: | |
1444 | family = PF_INET; | |
1445 | break; | |
1446 | default: | |
1447 | return ERR_PTR(-EINVAL); | |
1448 | } | |
1da177e4 | 1449 | |
14ec63c3 | 1450 | error = __sock_create(net, family, type, protocol, &sock, 1); |
77f1f67a | 1451 | if (error < 0) |
b700cbb1 | 1452 | return ERR_PTR(error); |
1da177e4 | 1453 | |
ed07536e PZ |
1454 | svc_reclassify_socket(sock); |
1455 | ||
c69da774 TM |
1456 | /* |
1457 | * If this is an PF_INET6 listener, we want to avoid | |
1458 | * getting requests from IPv4 remotes. Those should | |
1459 | * be shunted to a PF_INET listener via rpcbind. | |
1460 | */ | |
c69da774 | 1461 | if (family == PF_INET6) |
9b115749 | 1462 | ip6_sock_set_v6only(sock->sk); |
18114746 | 1463 | if (type == SOCK_STREAM) |
4a17fd52 | 1464 | sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */ |
77f1f67a | 1465 | error = kernel_bind(sock, sin, len); |
18114746 ES |
1466 | if (error < 0) |
1467 | goto bummer; | |
1da177e4 | 1468 | |
9b2c45d4 | 1469 | error = kernel_getsockname(sock, newsin); |
9dbc240f TT |
1470 | if (error < 0) |
1471 | goto bummer; | |
9b2c45d4 | 1472 | newlen = error; |
9dbc240f | 1473 | |
1da177e4 | 1474 | if (protocol == IPPROTO_TCP) { |
e6242e92 | 1475 | if ((error = kernel_listen(sock, 64)) < 0) |
1da177e4 LT |
1476 | goto bummer; |
1477 | } | |
1478 | ||
72c35376 | 1479 | svsk = svc_setup_socket(serv, sock, flags); |
a8e10078 BF |
1480 | if (IS_ERR(svsk)) { |
1481 | error = PTR_ERR(svsk); | |
1482 | goto bummer; | |
e79eff1f | 1483 | } |
a8e10078 BF |
1484 | svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen); |
1485 | return (struct svc_xprt *)svsk; | |
1da177e4 | 1486 | bummer: |
1da177e4 | 1487 | sock_release(sock); |
b700cbb1 | 1488 | return ERR_PTR(error); |
1da177e4 LT |
1489 | } |
1490 | ||
755cceab TT |
1491 | /* |
1492 | * Detach the svc_sock from the socket so that no | |
1493 | * more callbacks occur. | |
1494 | */ | |
1495 | static void svc_sock_detach(struct svc_xprt *xprt) | |
1496 | { | |
1497 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
1498 | struct sock *sk = svsk->sk_sk; | |
1499 | ||
755cceab | 1500 | /* put back the old socket callbacks */ |
069c225b | 1501 | lock_sock(sk); |
755cceab TT |
1502 | sk->sk_state_change = svsk->sk_ostate; |
1503 | sk->sk_data_ready = svsk->sk_odata; | |
1504 | sk->sk_write_space = svsk->sk_owspace; | |
069c225b TM |
1505 | sk->sk_user_data = NULL; |
1506 | release_sock(sk); | |
69b6ba37 TM |
1507 | } |
1508 | ||
1509 | /* | |
1510 | * Disconnect the socket, and reset the callbacks | |
1511 | */ | |
1512 | static void svc_tcp_sock_detach(struct svc_xprt *xprt) | |
1513 | { | |
1514 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
1515 | ||
69b6ba37 TM |
1516 | svc_sock_detach(xprt); |
1517 | ||
31d68ef6 BF |
1518 | if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) { |
1519 | svc_tcp_clear_pages(svsk); | |
69b6ba37 | 1520 | kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR); |
31d68ef6 | 1521 | } |
755cceab TT |
1522 | } |
1523 | ||
1524 | /* | |
1525 | * Free the svc_sock's socket resources and the svc_sock itself. | |
1526 | */ | |
1527 | static void svc_sock_free(struct svc_xprt *xprt) | |
1528 | { | |
1529 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
755cceab | 1530 | |
755cceab TT |
1531 | if (svsk->sk_sock->file) |
1532 | sockfd_put(svsk->sk_sock); | |
1533 | else | |
1534 | sock_release(svsk->sk_sock); | |
1535 | kfree(svsk); | |
1536 | } |