]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/net/sunrpc/svcsock.c | |
3 | * | |
4 | * These are the RPC server socket internals. | |
5 | * | |
6 | * The server scheduling algorithm does not always distribute the load | |
7 | * evenly when servicing a single client. May need to modify the | |
f6150c3c | 8 | * svc_xprt_enqueue procedure... |
1da177e4 LT |
9 | * |
10 | * TCP support is largely untested and may be a little slow. The problem | |
11 | * is that we currently do two separate recvfrom's, one for the 4-byte | |
12 | * record length, and the second for the actual record. This could possibly | |
13 | * be improved by always reading a minimum size of around 100 bytes and | |
14 | * tucking any superfluous bytes away in a temporary store. Still, that | |
15 | * leaves write requests out in the rain. An alternative may be to peek at | |
16 | * the first skb in the queue, and if it matches the next TCP sequence | |
17 | * number, to extract the record marker. Yuck. | |
18 | * | |
19 | * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de> | |
20 | */ | |
21 | ||
172589cc | 22 | #include <linux/kernel.h> |
1da177e4 | 23 | #include <linux/sched.h> |
3a9a231d | 24 | #include <linux/module.h> |
1da177e4 LT |
25 | #include <linux/errno.h> |
26 | #include <linux/fcntl.h> | |
27 | #include <linux/net.h> | |
28 | #include <linux/in.h> | |
29 | #include <linux/inet.h> | |
30 | #include <linux/udp.h> | |
91483c4b | 31 | #include <linux/tcp.h> |
1da177e4 LT |
32 | #include <linux/unistd.h> |
33 | #include <linux/slab.h> | |
34 | #include <linux/netdevice.h> | |
35 | #include <linux/skbuff.h> | |
b41b66d6 | 36 | #include <linux/file.h> |
7dfb7103 | 37 | #include <linux/freezer.h> |
1da177e4 LT |
38 | #include <net/sock.h> |
39 | #include <net/checksum.h> | |
40 | #include <net/ip.h> | |
b92503b2 | 41 | #include <net/ipv6.h> |
850cbadd | 42 | #include <net/udp.h> |
b7872fe8 | 43 | #include <net/tcp.h> |
c752f073 | 44 | #include <net/tcp_states.h> |
7c0f6ba6 | 45 | #include <linux/uaccess.h> |
1da177e4 | 46 | #include <asm/ioctls.h> |
22911fc5 | 47 | #include <trace/events/skb.h> |
1da177e4 LT |
48 | |
49 | #include <linux/sunrpc/types.h> | |
ad06e4bd | 50 | #include <linux/sunrpc/clnt.h> |
1da177e4 | 51 | #include <linux/sunrpc/xdr.h> |
c0401ea0 | 52 | #include <linux/sunrpc/msg_prot.h> |
1da177e4 LT |
53 | #include <linux/sunrpc/svcsock.h> |
54 | #include <linux/sunrpc/stats.h> | |
4cfc7e60 | 55 | #include <linux/sunrpc/xprt.h> |
1da177e4 | 56 | |
177e4f99 HS |
57 | #include "sunrpc.h" |
58 | ||
360d8738 | 59 | #define RPCDBG_FACILITY RPCDBG_SVCXPRT |
1da177e4 LT |
60 | |
61 | ||
62 | static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *, | |
72c35376 | 63 | int flags); |
1da177e4 LT |
64 | static int svc_udp_recvfrom(struct svc_rqst *); |
65 | static int svc_udp_sendto(struct svc_rqst *); | |
755cceab | 66 | static void svc_sock_detach(struct svc_xprt *); |
69b6ba37 | 67 | static void svc_tcp_sock_detach(struct svc_xprt *); |
755cceab | 68 | static void svc_sock_free(struct svc_xprt *); |
1da177e4 | 69 | |
b700cbb1 | 70 | static struct svc_xprt *svc_create_socket(struct svc_serv *, int, |
62832c03 PE |
71 | struct net *, struct sockaddr *, |
72 | int, int); | |
9e00abc3 | 73 | #if defined(CONFIG_SUNRPC_BACKCHANNEL) |
1f11a034 AA |
74 | static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int, |
75 | struct net *, struct sockaddr *, | |
76 | int, int); | |
77 | static void svc_bc_sock_free(struct svc_xprt *xprt); | |
9e00abc3 | 78 | #endif /* CONFIG_SUNRPC_BACKCHANNEL */ |
1f11a034 | 79 | |
ed07536e PZ |
80 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
81 | static struct lock_class_key svc_key[2]; | |
82 | static struct lock_class_key svc_slock_key[2]; | |
83 | ||
0f0257ea | 84 | static void svc_reclassify_socket(struct socket *sock) |
ed07536e PZ |
85 | { |
86 | struct sock *sk = sock->sk; | |
1b7a1819 | 87 | |
fafc4e1e | 88 | if (WARN_ON_ONCE(!sock_allow_reclassification(sk))) |
1b7a1819 WAA |
89 | return; |
90 | ||
ed07536e PZ |
91 | switch (sk->sk_family) { |
92 | case AF_INET: | |
93 | sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD", | |
def13d74 TT |
94 | &svc_slock_key[0], |
95 | "sk_xprt.xpt_lock-AF_INET-NFSD", | |
96 | &svc_key[0]); | |
ed07536e PZ |
97 | break; |
98 | ||
99 | case AF_INET6: | |
100 | sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD", | |
def13d74 TT |
101 | &svc_slock_key[1], |
102 | "sk_xprt.xpt_lock-AF_INET6-NFSD", | |
103 | &svc_key[1]); | |
ed07536e PZ |
104 | break; |
105 | ||
106 | default: | |
107 | BUG(); | |
108 | } | |
109 | } | |
110 | #else | |
0f0257ea | 111 | static void svc_reclassify_socket(struct socket *sock) |
ed07536e PZ |
112 | { |
113 | } | |
114 | #endif | |
115 | ||
1da177e4 LT |
116 | /* |
117 | * Release an skbuff after use | |
118 | */ | |
5148bf4e | 119 | static void svc_release_skb(struct svc_rqst *rqstp) |
1da177e4 | 120 | { |
5148bf4e | 121 | struct sk_buff *skb = rqstp->rq_xprt_ctxt; |
1da177e4 LT |
122 | |
123 | if (skb) { | |
57b1d3ba TT |
124 | struct svc_sock *svsk = |
125 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
5148bf4e | 126 | rqstp->rq_xprt_ctxt = NULL; |
1da177e4 LT |
127 | |
128 | dprintk("svc: service %p, releasing skb %p\n", rqstp, skb); | |
9d410c79 | 129 | skb_free_datagram_locked(svsk->sk_sk, skb); |
1da177e4 | 130 | } |
1da177e4 LT |
131 | } |
132 | ||
850cbadd PA |
133 | static void svc_release_udp_skb(struct svc_rqst *rqstp) |
134 | { | |
135 | struct sk_buff *skb = rqstp->rq_xprt_ctxt; | |
136 | ||
137 | if (skb) { | |
138 | rqstp->rq_xprt_ctxt = NULL; | |
139 | ||
140 | dprintk("svc: service %p, releasing skb %p\n", rqstp, skb); | |
141 | consume_skb(skb); | |
142 | } | |
143 | } | |
144 | ||
b92503b2 CL |
145 | union svc_pktinfo_u { |
146 | struct in_pktinfo pkti; | |
b92503b2 | 147 | struct in6_pktinfo pkti6; |
b92503b2 | 148 | }; |
bc375ea7 DM |
149 | #define SVC_PKTINFO_SPACE \ |
150 | CMSG_SPACE(sizeof(union svc_pktinfo_u)) | |
b92503b2 CL |
151 | |
152 | static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh) | |
153 | { | |
57b1d3ba TT |
154 | struct svc_sock *svsk = |
155 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
156 | switch (svsk->sk_sk->sk_family) { | |
b92503b2 CL |
157 | case AF_INET: { |
158 | struct in_pktinfo *pki = CMSG_DATA(cmh); | |
159 | ||
160 | cmh->cmsg_level = SOL_IP; | |
161 | cmh->cmsg_type = IP_PKTINFO; | |
162 | pki->ipi_ifindex = 0; | |
849a1cf1 MJ |
163 | pki->ipi_spec_dst.s_addr = |
164 | svc_daddr_in(rqstp)->sin_addr.s_addr; | |
b92503b2 CL |
165 | cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); |
166 | } | |
167 | break; | |
5a05ed73 | 168 | |
b92503b2 CL |
169 | case AF_INET6: { |
170 | struct in6_pktinfo *pki = CMSG_DATA(cmh); | |
849a1cf1 | 171 | struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp); |
b92503b2 CL |
172 | |
173 | cmh->cmsg_level = SOL_IPV6; | |
174 | cmh->cmsg_type = IPV6_PKTINFO; | |
849a1cf1 | 175 | pki->ipi6_ifindex = daddr->sin6_scope_id; |
4e3fd7a0 | 176 | pki->ipi6_addr = daddr->sin6_addr; |
b92503b2 CL |
177 | cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); |
178 | } | |
179 | break; | |
b92503b2 | 180 | } |
b92503b2 CL |
181 | } |
182 | ||
1da177e4 | 183 | /* |
4cfc7e60 | 184 | * send routine intended to be shared by the fore- and back-channel |
1da177e4 | 185 | */ |
4cfc7e60 RI |
186 | int svc_send_common(struct socket *sock, struct xdr_buf *xdr, |
187 | struct page *headpage, unsigned long headoffset, | |
188 | struct page *tailpage, unsigned long tailoffset) | |
1da177e4 | 189 | { |
1da177e4 LT |
190 | int result; |
191 | int size; | |
192 | struct page **ppage = xdr->pages; | |
193 | size_t base = xdr->page_base; | |
194 | unsigned int pglen = xdr->page_len; | |
226453d8 | 195 | unsigned int flags = MSG_MORE | MSG_SENDPAGE_NOTLAST; |
4cfc7e60 RI |
196 | int slen; |
197 | int len = 0; | |
1da177e4 LT |
198 | |
199 | slen = xdr->len; | |
200 | ||
1da177e4 LT |
201 | /* send head */ |
202 | if (slen == xdr->head[0].iov_len) | |
203 | flags = 0; | |
4cfc7e60 | 204 | len = kernel_sendpage(sock, headpage, headoffset, |
44524359 | 205 | xdr->head[0].iov_len, flags); |
1da177e4 LT |
206 | if (len != xdr->head[0].iov_len) |
207 | goto out; | |
208 | slen -= xdr->head[0].iov_len; | |
209 | if (slen == 0) | |
210 | goto out; | |
211 | ||
212 | /* send page data */ | |
213 | size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen; | |
214 | while (pglen > 0) { | |
215 | if (slen == size) | |
216 | flags = 0; | |
e6242e92 | 217 | result = kernel_sendpage(sock, *ppage, base, size, flags); |
1da177e4 LT |
218 | if (result > 0) |
219 | len += result; | |
220 | if (result != size) | |
221 | goto out; | |
222 | slen -= size; | |
223 | pglen -= size; | |
224 | size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen; | |
225 | base = 0; | |
226 | ppage++; | |
227 | } | |
4cfc7e60 | 228 | |
1da177e4 LT |
229 | /* send tail */ |
230 | if (xdr->tail[0].iov_len) { | |
4cfc7e60 RI |
231 | result = kernel_sendpage(sock, tailpage, tailoffset, |
232 | xdr->tail[0].iov_len, 0); | |
1da177e4 LT |
233 | if (result > 0) |
234 | len += result; | |
235 | } | |
4cfc7e60 RI |
236 | |
237 | out: | |
238 | return len; | |
239 | } | |
240 | ||
241 | ||
242 | /* | |
243 | * Generic sendto routine | |
244 | */ | |
245 | static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr) | |
246 | { | |
247 | struct svc_sock *svsk = | |
248 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
249 | struct socket *sock = svsk->sk_sock; | |
250 | union { | |
251 | struct cmsghdr hdr; | |
252 | long all[SVC_PKTINFO_SPACE / sizeof(long)]; | |
253 | } buffer; | |
254 | struct cmsghdr *cmh = &buffer.hdr; | |
255 | int len = 0; | |
256 | unsigned long tailoff; | |
257 | unsigned long headoff; | |
258 | RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); | |
259 | ||
260 | if (rqstp->rq_prot == IPPROTO_UDP) { | |
261 | struct msghdr msg = { | |
262 | .msg_name = &rqstp->rq_addr, | |
263 | .msg_namelen = rqstp->rq_addrlen, | |
264 | .msg_control = cmh, | |
265 | .msg_controllen = sizeof(buffer), | |
266 | .msg_flags = MSG_MORE, | |
267 | }; | |
268 | ||
269 | svc_set_cmsg_data(rqstp, cmh); | |
270 | ||
d8725c86 | 271 | if (sock_sendmsg(sock, &msg) < 0) |
4cfc7e60 RI |
272 | goto out; |
273 | } | |
274 | ||
275 | tailoff = ((unsigned long)xdr->tail[0].iov_base) & (PAGE_SIZE-1); | |
276 | headoff = 0; | |
277 | len = svc_send_common(sock, xdr, rqstp->rq_respages[0], headoff, | |
278 | rqstp->rq_respages[0], tailoff); | |
279 | ||
1da177e4 | 280 | out: |
ad06e4bd | 281 | dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n", |
57b1d3ba | 282 | svsk, xdr->head[0].iov_base, xdr->head[0].iov_len, |
ad06e4bd | 283 | xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf))); |
1da177e4 LT |
284 | |
285 | return len; | |
286 | } | |
287 | ||
80212d59 N |
288 | /* |
289 | * Report socket names for nfsdfs | |
290 | */ | |
e7942b9f | 291 | static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining) |
80212d59 | 292 | { |
017cb47f CL |
293 | const struct sock *sk = svsk->sk_sk; |
294 | const char *proto_name = sk->sk_protocol == IPPROTO_UDP ? | |
295 | "udp" : "tcp"; | |
80212d59 N |
296 | int len; |
297 | ||
017cb47f | 298 | switch (sk->sk_family) { |
e7942b9f CL |
299 | case PF_INET: |
300 | len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n", | |
017cb47f | 301 | proto_name, |
c720c7e8 ED |
302 | &inet_sk(sk)->inet_rcv_saddr, |
303 | inet_sk(sk)->inet_num); | |
80212d59 | 304 | break; |
c2bb06db | 305 | #if IS_ENABLED(CONFIG_IPV6) |
58de2f86 CL |
306 | case PF_INET6: |
307 | len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n", | |
017cb47f | 308 | proto_name, |
efe4208f | 309 | &sk->sk_v6_rcv_saddr, |
c720c7e8 | 310 | inet_sk(sk)->inet_num); |
80212d59 | 311 | break; |
c2bb06db | 312 | #endif |
80212d59 | 313 | default: |
e7942b9f | 314 | len = snprintf(buf, remaining, "*unknown-%d*\n", |
017cb47f | 315 | sk->sk_family); |
80212d59 | 316 | } |
e7942b9f CL |
317 | |
318 | if (len >= remaining) { | |
319 | *buf = '\0'; | |
320 | return -ENAMETOOLONG; | |
80212d59 N |
321 | } |
322 | return len; | |
323 | } | |
324 | ||
1da177e4 LT |
325 | /* |
326 | * Generic recvfrom routine. | |
327 | */ | |
0f0257ea TT |
328 | static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr, |
329 | int buflen) | |
1da177e4 | 330 | { |
57b1d3ba TT |
331 | struct svc_sock *svsk = |
332 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
1ba95105 CL |
333 | struct msghdr msg = { |
334 | .msg_flags = MSG_DONTWAIT, | |
335 | }; | |
336 | int len; | |
1da177e4 | 337 | |
260c1d12 TT |
338 | rqstp->rq_xprt_hlen = 0; |
339 | ||
f8d1ff47 | 340 | clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
1ba95105 CL |
341 | len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen, |
342 | msg.msg_flags); | |
f8d1ff47 TM |
343 | /* If we read a full record, then assume there may be more |
344 | * data to read (stream based sockets only!) | |
345 | */ | |
346 | if (len == buflen) | |
347 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); | |
1da177e4 | 348 | |
1da177e4 | 349 | dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n", |
1ba95105 | 350 | svsk, iov[0].iov_base, iov[0].iov_len, len); |
1da177e4 LT |
351 | return len; |
352 | } | |
353 | ||
31d68ef6 BF |
354 | static int svc_partial_recvfrom(struct svc_rqst *rqstp, |
355 | struct kvec *iov, int nr, | |
356 | int buflen, unsigned int base) | |
357 | { | |
358 | size_t save_iovlen; | |
09acfea5 | 359 | void *save_iovbase; |
31d68ef6 BF |
360 | unsigned int i; |
361 | int ret; | |
362 | ||
363 | if (base == 0) | |
364 | return svc_recvfrom(rqstp, iov, nr, buflen); | |
365 | ||
366 | for (i = 0; i < nr; i++) { | |
367 | if (iov[i].iov_len > base) | |
368 | break; | |
369 | base -= iov[i].iov_len; | |
370 | } | |
371 | save_iovlen = iov[i].iov_len; | |
372 | save_iovbase = iov[i].iov_base; | |
373 | iov[i].iov_len -= base; | |
374 | iov[i].iov_base += base; | |
375 | ret = svc_recvfrom(rqstp, &iov[i], nr - i, buflen); | |
376 | iov[i].iov_len = save_iovlen; | |
377 | iov[i].iov_base = save_iovbase; | |
378 | return ret; | |
379 | } | |
380 | ||
1da177e4 LT |
381 | /* |
382 | * Set socket snd and rcv buffer lengths | |
383 | */ | |
0f0257ea TT |
384 | static void svc_sock_setbufsize(struct socket *sock, unsigned int snd, |
385 | unsigned int rcv) | |
1da177e4 LT |
386 | { |
387 | #if 0 | |
388 | mm_segment_t oldfs; | |
389 | oldfs = get_fs(); set_fs(KERNEL_DS); | |
390 | sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF, | |
391 | (char*)&snd, sizeof(snd)); | |
392 | sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF, | |
393 | (char*)&rcv, sizeof(rcv)); | |
394 | #else | |
395 | /* sock_setsockopt limits use to sysctl_?mem_max, | |
396 | * which isn't acceptable. Until that is made conditional | |
397 | * on not having CAP_SYS_RESOURCE or similar, we go direct... | |
398 | * DaveM said I could! | |
399 | */ | |
400 | lock_sock(sock->sk); | |
401 | sock->sk->sk_sndbuf = snd * 2; | |
402 | sock->sk->sk_rcvbuf = rcv * 2; | |
47fcb03f | 403 | sock->sk->sk_write_space(sock->sk); |
1da177e4 LT |
404 | release_sock(sock->sk); |
405 | #endif | |
406 | } | |
16e4d93f CL |
407 | |
408 | static int svc_sock_secure_port(struct svc_rqst *rqstp) | |
409 | { | |
410 | return svc_port_is_privileged(svc_addr(rqstp)); | |
411 | } | |
412 | ||
1da177e4 LT |
413 | /* |
414 | * INET callback when data has been received on the socket. | |
415 | */ | |
fa9251af | 416 | static void svc_data_ready(struct sock *sk) |
1da177e4 | 417 | { |
939bb7ef | 418 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 | 419 | |
939bb7ef | 420 | if (svsk) { |
676d2369 DM |
421 | dprintk("svc: socket %p(inet %p), busy=%d\n", |
422 | svsk, sk, | |
02fc6c36 | 423 | test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags)); |
fa9251af | 424 | svsk->sk_odata(sk); |
4720b070 TM |
425 | if (!test_and_set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags)) |
426 | svc_xprt_enqueue(&svsk->sk_xprt); | |
939bb7ef | 427 | } |
1da177e4 LT |
428 | } |
429 | ||
430 | /* | |
431 | * INET callback when space is newly available on the socket. | |
432 | */ | |
0f0257ea | 433 | static void svc_write_space(struct sock *sk) |
1da177e4 LT |
434 | { |
435 | struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data); | |
436 | ||
437 | if (svsk) { | |
438 | dprintk("svc: socket %p(inet %p), write_space busy=%d\n", | |
02fc6c36 | 439 | svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags)); |
fa9251af | 440 | svsk->sk_owspace(sk); |
f6150c3c | 441 | svc_xprt_enqueue(&svsk->sk_xprt); |
1da177e4 | 442 | } |
1da177e4 LT |
443 | } |
444 | ||
c7fb3f06 TM |
445 | static int svc_tcp_has_wspace(struct svc_xprt *xprt) |
446 | { | |
637600f3 | 447 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); |
c7fb3f06 TM |
448 | |
449 | if (test_bit(XPT_LISTENER, &xprt->xpt_flags)) | |
450 | return 1; | |
637600f3 | 451 | return !test_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); |
51877680 TM |
452 | } |
453 | ||
ea08e392 SM |
454 | static void svc_tcp_kill_temp_xprt(struct svc_xprt *xprt) |
455 | { | |
456 | struct svc_sock *svsk; | |
457 | struct socket *sock; | |
458 | struct linger no_linger = { | |
459 | .l_onoff = 1, | |
460 | .l_linger = 0, | |
461 | }; | |
462 | ||
463 | svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
464 | sock = svsk->sk_sock; | |
465 | kernel_setsockopt(sock, SOL_SOCKET, SO_LINGER, | |
466 | (char *)&no_linger, sizeof(no_linger)); | |
467 | } | |
468 | ||
7702ce40 CL |
469 | /* |
470 | * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo | |
471 | */ | |
472 | static int svc_udp_get_dest_address4(struct svc_rqst *rqstp, | |
473 | struct cmsghdr *cmh) | |
474 | { | |
475 | struct in_pktinfo *pki = CMSG_DATA(cmh); | |
849a1cf1 MJ |
476 | struct sockaddr_in *daddr = svc_daddr_in(rqstp); |
477 | ||
7702ce40 CL |
478 | if (cmh->cmsg_type != IP_PKTINFO) |
479 | return 0; | |
849a1cf1 MJ |
480 | |
481 | daddr->sin_family = AF_INET; | |
482 | daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr; | |
7702ce40 CL |
483 | return 1; |
484 | } | |
485 | ||
486 | /* | |
73df66f8 | 487 | * See net/ipv6/datagram.c : ip6_datagram_recv_ctl |
7702ce40 CL |
488 | */ |
489 | static int svc_udp_get_dest_address6(struct svc_rqst *rqstp, | |
490 | struct cmsghdr *cmh) | |
491 | { | |
492 | struct in6_pktinfo *pki = CMSG_DATA(cmh); | |
849a1cf1 MJ |
493 | struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp); |
494 | ||
7702ce40 CL |
495 | if (cmh->cmsg_type != IPV6_PKTINFO) |
496 | return 0; | |
849a1cf1 MJ |
497 | |
498 | daddr->sin6_family = AF_INET6; | |
4e3fd7a0 | 499 | daddr->sin6_addr = pki->ipi6_addr; |
849a1cf1 | 500 | daddr->sin6_scope_id = pki->ipi6_ifindex; |
7702ce40 CL |
501 | return 1; |
502 | } | |
503 | ||
9dbc240f TT |
504 | /* |
505 | * Copy the UDP datagram's destination address to the rqstp structure. | |
506 | * The 'destination' address in this case is the address to which the | |
507 | * peer sent the datagram, i.e. our local address. For multihomed | |
508 | * hosts, this can change from msg to msg. Note that only the IP | |
509 | * address changes, the port number should remain the same. | |
510 | */ | |
7702ce40 CL |
511 | static int svc_udp_get_dest_address(struct svc_rqst *rqstp, |
512 | struct cmsghdr *cmh) | |
95756482 | 513 | { |
7702ce40 CL |
514 | switch (cmh->cmsg_level) { |
515 | case SOL_IP: | |
516 | return svc_udp_get_dest_address4(rqstp, cmh); | |
517 | case SOL_IPV6: | |
518 | return svc_udp_get_dest_address6(rqstp, cmh); | |
95756482 | 519 | } |
7702ce40 CL |
520 | |
521 | return 0; | |
95756482 CL |
522 | } |
523 | ||
1da177e4 LT |
524 | /* |
525 | * Receive a datagram from a UDP socket. | |
526 | */ | |
0f0257ea | 527 | static int svc_udp_recvfrom(struct svc_rqst *rqstp) |
1da177e4 | 528 | { |
57b1d3ba TT |
529 | struct svc_sock *svsk = |
530 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
bb5cf160 | 531 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
1da177e4 | 532 | struct sk_buff *skb; |
bc375ea7 DM |
533 | union { |
534 | struct cmsghdr hdr; | |
535 | long all[SVC_PKTINFO_SPACE / sizeof(long)]; | |
536 | } buffer; | |
537 | struct cmsghdr *cmh = &buffer.hdr; | |
7a37f578 N |
538 | struct msghdr msg = { |
539 | .msg_name = svc_addr(rqstp), | |
540 | .msg_control = cmh, | |
541 | .msg_controllen = sizeof(buffer), | |
542 | .msg_flags = MSG_DONTWAIT, | |
543 | }; | |
abc5c44d CL |
544 | size_t len; |
545 | int err; | |
1da177e4 | 546 | |
02fc6c36 | 547 | if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags)) |
1da177e4 LT |
548 | /* udp sockets need large rcvbuf as all pending |
549 | * requests are still in that buffer. sndbuf must | |
550 | * also be large enough that there is enough space | |
3262c816 GB |
551 | * for one reply per thread. We count all threads |
552 | * rather than threads in a particular pool, which | |
553 | * provides an upper bound on the number of threads | |
554 | * which will access the socket. | |
1da177e4 LT |
555 | */ |
556 | svc_sock_setbufsize(svsk->sk_sock, | |
c6b0a9f8 N |
557 | (serv->sv_nrthreads+3) * serv->sv_max_mesg, |
558 | (serv->sv_nrthreads+3) * serv->sv_max_mesg); | |
1da177e4 | 559 | |
02fc6c36 | 560 | clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
05ed690e N |
561 | skb = NULL; |
562 | err = kernel_recvmsg(svsk->sk_sock, &msg, NULL, | |
563 | 0, 0, MSG_PEEK | MSG_DONTWAIT); | |
564 | if (err >= 0) | |
7c13f97f | 565 | skb = skb_recv_udp(svsk->sk_sk, 0, 1, &err); |
05ed690e N |
566 | |
567 | if (skb == NULL) { | |
568 | if (err != -EAGAIN) { | |
569 | /* possibly an icmp error */ | |
570 | dprintk("svc: recvfrom returned error %d\n", -err); | |
02fc6c36 | 571 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 572 | } |
9f9d2ebe | 573 | return 0; |
1da177e4 | 574 | } |
9dbc240f | 575 | len = svc_addr_len(svc_addr(rqstp)); |
9dbc240f | 576 | rqstp->rq_addrlen = len; |
b7aa0bf7 ED |
577 | if (skb->tstamp.tv64 == 0) { |
578 | skb->tstamp = ktime_get_real(); | |
cca5172a | 579 | /* Don't enable netstamp, sunrpc doesn't |
1da177e4 LT |
580 | need that much accuracy */ |
581 | } | |
b7aa0bf7 | 582 | svsk->sk_sk->sk_stamp = skb->tstamp; |
02fc6c36 | 583 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */ |
1da177e4 | 584 | |
1da8c681 | 585 | len = skb->len; |
1da177e4 LT |
586 | rqstp->rq_arg.len = len; |
587 | ||
95756482 | 588 | rqstp->rq_prot = IPPROTO_UDP; |
27459f09 | 589 | |
7702ce40 | 590 | if (!svc_udp_get_dest_address(rqstp, cmh)) { |
e87cc472 JP |
591 | net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n", |
592 | cmh->cmsg_level, cmh->cmsg_type); | |
f23abfdb | 593 | goto out_free; |
7a37f578 | 594 | } |
849a1cf1 | 595 | rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp)); |
1da177e4 LT |
596 | |
597 | if (skb_is_nonlinear(skb)) { | |
598 | /* we have to copy */ | |
599 | local_bh_disable(); | |
600 | if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) { | |
601 | local_bh_enable(); | |
602 | /* checksum error */ | |
22911fc5 | 603 | goto out_free; |
1da177e4 LT |
604 | } |
605 | local_bh_enable(); | |
850cbadd | 606 | consume_skb(skb); |
1da177e4 LT |
607 | } else { |
608 | /* we can use it in-place */ | |
1da8c681 | 609 | rqstp->rq_arg.head[0].iov_base = skb->data; |
1da177e4 | 610 | rqstp->rq_arg.head[0].iov_len = len; |
22911fc5 ED |
611 | if (skb_checksum_complete(skb)) |
612 | goto out_free; | |
5148bf4e | 613 | rqstp->rq_xprt_ctxt = skb; |
1da177e4 LT |
614 | } |
615 | ||
616 | rqstp->rq_arg.page_base = 0; | |
617 | if (len <= rqstp->rq_arg.head[0].iov_len) { | |
618 | rqstp->rq_arg.head[0].iov_len = len; | |
619 | rqstp->rq_arg.page_len = 0; | |
44524359 | 620 | rqstp->rq_respages = rqstp->rq_pages+1; |
1da177e4 LT |
621 | } else { |
622 | rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len; | |
44524359 | 623 | rqstp->rq_respages = rqstp->rq_pages + 1 + |
172589cc | 624 | DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE); |
1da177e4 | 625 | } |
afc59400 | 626 | rqstp->rq_next_page = rqstp->rq_respages+1; |
1da177e4 LT |
627 | |
628 | if (serv->sv_stats) | |
629 | serv->sv_stats->netudpcnt++; | |
630 | ||
631 | return len; | |
f23abfdb | 632 | out_free: |
850cbadd | 633 | kfree_skb(skb); |
f23abfdb | 634 | return 0; |
1da177e4 LT |
635 | } |
636 | ||
637 | static int | |
638 | svc_udp_sendto(struct svc_rqst *rqstp) | |
639 | { | |
640 | int error; | |
641 | ||
642 | error = svc_sendto(rqstp, &rqstp->rq_res); | |
643 | if (error == -ECONNREFUSED) | |
644 | /* ICMP error on earlier request. */ | |
645 | error = svc_sendto(rqstp, &rqstp->rq_res); | |
646 | ||
647 | return error; | |
648 | } | |
649 | ||
e831fe65 TT |
650 | static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp) |
651 | { | |
652 | } | |
653 | ||
323bee32 TT |
654 | static int svc_udp_has_wspace(struct svc_xprt *xprt) |
655 | { | |
656 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
bb5cf160 | 657 | struct svc_serv *serv = xprt->xpt_server; |
323bee32 TT |
658 | unsigned long required; |
659 | ||
660 | /* | |
661 | * Set the SOCK_NOSPACE flag before checking the available | |
662 | * sock space. | |
663 | */ | |
664 | set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
7a90e8cc | 665 | required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg; |
323bee32 TT |
666 | if (required*2 > sock_wspace(svsk->sk_sk)) |
667 | return 0; | |
668 | clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
669 | return 1; | |
670 | } | |
671 | ||
38a417cc TT |
672 | static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt) |
673 | { | |
674 | BUG(); | |
675 | return NULL; | |
676 | } | |
677 | ||
ea08e392 SM |
678 | static void svc_udp_kill_temp_xprt(struct svc_xprt *xprt) |
679 | { | |
680 | } | |
681 | ||
b700cbb1 | 682 | static struct svc_xprt *svc_udp_create(struct svc_serv *serv, |
62832c03 | 683 | struct net *net, |
b700cbb1 TT |
684 | struct sockaddr *sa, int salen, |
685 | int flags) | |
686 | { | |
62832c03 | 687 | return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags); |
b700cbb1 TT |
688 | } |
689 | ||
360d8738 | 690 | static struct svc_xprt_ops svc_udp_ops = { |
b700cbb1 | 691 | .xpo_create = svc_udp_create, |
5d137990 TT |
692 | .xpo_recvfrom = svc_udp_recvfrom, |
693 | .xpo_sendto = svc_udp_sendto, | |
850cbadd | 694 | .xpo_release_rqst = svc_release_udp_skb, |
755cceab TT |
695 | .xpo_detach = svc_sock_detach, |
696 | .xpo_free = svc_sock_free, | |
e831fe65 | 697 | .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr, |
323bee32 | 698 | .xpo_has_wspace = svc_udp_has_wspace, |
38a417cc | 699 | .xpo_accept = svc_udp_accept, |
16e4d93f | 700 | .xpo_secure_port = svc_sock_secure_port, |
ea08e392 | 701 | .xpo_kill_temp_xprt = svc_udp_kill_temp_xprt, |
360d8738 TT |
702 | }; |
703 | ||
704 | static struct svc_xprt_class svc_udp_class = { | |
705 | .xcl_name = "udp", | |
b700cbb1 | 706 | .xcl_owner = THIS_MODULE, |
360d8738 | 707 | .xcl_ops = &svc_udp_ops, |
49023155 | 708 | .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP, |
3c45ddf8 | 709 | .xcl_ident = XPRT_TRANSPORT_UDP, |
360d8738 TT |
710 | }; |
711 | ||
bb5cf160 | 712 | static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv) |
1da177e4 | 713 | { |
7702ce40 | 714 | int err, level, optname, one = 1; |
7a37f578 | 715 | |
bd4620dd SK |
716 | svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class, |
717 | &svsk->sk_xprt, serv); | |
def13d74 | 718 | clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags); |
fa9251af | 719 | svsk->sk_sk->sk_data_ready = svc_data_ready; |
1da177e4 | 720 | svsk->sk_sk->sk_write_space = svc_write_space; |
1da177e4 LT |
721 | |
722 | /* initialise setting must have enough space to | |
cca5172a | 723 | * receive and respond to one request. |
1da177e4 LT |
724 | * svc_udp_recvfrom will re-adjust if necessary |
725 | */ | |
726 | svc_sock_setbufsize(svsk->sk_sock, | |
bb5cf160 TT |
727 | 3 * svsk->sk_xprt.xpt_server->sv_max_mesg, |
728 | 3 * svsk->sk_xprt.xpt_server->sv_max_mesg); | |
1da177e4 | 729 | |
0f0257ea TT |
730 | /* data might have come in before data_ready set up */ |
731 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); | |
02fc6c36 | 732 | set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); |
7a37f578 | 733 | |
7a37f578 | 734 | /* make sure we get destination address info */ |
7702ce40 CL |
735 | switch (svsk->sk_sk->sk_family) { |
736 | case AF_INET: | |
737 | level = SOL_IP; | |
738 | optname = IP_PKTINFO; | |
739 | break; | |
740 | case AF_INET6: | |
741 | level = SOL_IPV6; | |
742 | optname = IPV6_RECVPKTINFO; | |
743 | break; | |
744 | default: | |
745 | BUG(); | |
746 | } | |
747 | err = kernel_setsockopt(svsk->sk_sock, level, optname, | |
748 | (char *)&one, sizeof(one)); | |
749 | dprintk("svc: kernel_setsockopt returned %d\n", err); | |
1da177e4 LT |
750 | } |
751 | ||
752 | /* | |
753 | * A data_ready event on a listening socket means there's a connection | |
754 | * pending. Do not use state_change as a substitute for it. | |
755 | */ | |
676d2369 | 756 | static void svc_tcp_listen_data_ready(struct sock *sk) |
1da177e4 | 757 | { |
939bb7ef | 758 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 LT |
759 | |
760 | dprintk("svc: socket %p TCP (listen) state change %d\n", | |
939bb7ef | 761 | sk, sk->sk_state); |
1da177e4 | 762 | |
fa9251af TM |
763 | if (svsk) |
764 | svsk->sk_odata(sk); | |
939bb7ef NB |
765 | /* |
766 | * This callback may called twice when a new connection | |
767 | * is established as a child socket inherits everything | |
768 | * from a parent LISTEN socket. | |
769 | * 1) data_ready method of the parent socket will be called | |
770 | * when one of child sockets become ESTABLISHED. | |
771 | * 2) data_ready method of the child socket may be called | |
772 | * when it receives data before the socket is accepted. | |
773 | * In case of 2, we should ignore it silently. | |
774 | */ | |
775 | if (sk->sk_state == TCP_LISTEN) { | |
776 | if (svsk) { | |
02fc6c36 | 777 | set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
f6150c3c | 778 | svc_xprt_enqueue(&svsk->sk_xprt); |
939bb7ef NB |
779 | } else |
780 | printk("svc: socket %p: no user data\n", sk); | |
1da177e4 | 781 | } |
1da177e4 LT |
782 | } |
783 | ||
784 | /* | |
785 | * A state change on a connected socket means it's dying or dead. | |
786 | */ | |
0f0257ea | 787 | static void svc_tcp_state_change(struct sock *sk) |
1da177e4 | 788 | { |
939bb7ef | 789 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 LT |
790 | |
791 | dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n", | |
939bb7ef | 792 | sk, sk->sk_state, sk->sk_user_data); |
1da177e4 | 793 | |
939bb7ef | 794 | if (!svsk) |
1da177e4 | 795 | printk("svc: socket %p: no user data\n", sk); |
939bb7ef | 796 | else { |
fa9251af | 797 | svsk->sk_ostate(sk); |
b2f21f7d TM |
798 | if (sk->sk_state != TCP_ESTABLISHED) { |
799 | set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); | |
800 | svc_xprt_enqueue(&svsk->sk_xprt); | |
801 | } | |
1da177e4 | 802 | } |
1da177e4 LT |
803 | } |
804 | ||
805 | /* | |
806 | * Accept a TCP connection | |
807 | */ | |
38a417cc | 808 | static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt) |
1da177e4 | 809 | { |
38a417cc | 810 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); |
cdd88b9f AM |
811 | struct sockaddr_storage addr; |
812 | struct sockaddr *sin = (struct sockaddr *) &addr; | |
bb5cf160 | 813 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
1da177e4 LT |
814 | struct socket *sock = svsk->sk_sock; |
815 | struct socket *newsock; | |
1da177e4 LT |
816 | struct svc_sock *newsvsk; |
817 | int err, slen; | |
5216a8e7 | 818 | RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); |
1da177e4 LT |
819 | |
820 | dprintk("svc: tcp_accept %p sock %p\n", svsk, sock); | |
821 | if (!sock) | |
38a417cc | 822 | return NULL; |
1da177e4 | 823 | |
02fc6c36 | 824 | clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
e6242e92 SS |
825 | err = kernel_accept(sock, &newsock, O_NONBLOCK); |
826 | if (err < 0) { | |
1da177e4 LT |
827 | if (err == -ENOMEM) |
828 | printk(KERN_WARNING "%s: no more sockets!\n", | |
829 | serv->sv_name); | |
e87cc472 JP |
830 | else if (err != -EAGAIN) |
831 | net_warn_ratelimited("%s: accept failed (err %d)!\n", | |
832 | serv->sv_name, -err); | |
38a417cc | 833 | return NULL; |
1da177e4 | 834 | } |
02fc6c36 | 835 | set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 836 | |
cdd88b9f | 837 | err = kernel_getpeername(newsock, sin, &slen); |
1da177e4 | 838 | if (err < 0) { |
e87cc472 JP |
839 | net_warn_ratelimited("%s: peername failed (err %d)!\n", |
840 | serv->sv_name, -err); | |
1da177e4 LT |
841 | goto failed; /* aborted connection or whatever */ |
842 | } | |
843 | ||
844 | /* Ideally, we would want to reject connections from unauthorized | |
ad06e4bd CL |
845 | * hosts here, but when we get encryption, the IP of the host won't |
846 | * tell us anything. For now just warn about unpriv connections. | |
1da177e4 | 847 | */ |
cdd88b9f | 848 | if (!svc_port_is_privileged(sin)) { |
a48fd0f9 | 849 | dprintk("%s: connect from unprivileged port: %s\n", |
cca5172a | 850 | serv->sv_name, |
cdd88b9f | 851 | __svc_print_addr(sin, buf, sizeof(buf))); |
1da177e4 | 852 | } |
ad06e4bd | 853 | dprintk("%s: connect from %s\n", serv->sv_name, |
cdd88b9f | 854 | __svc_print_addr(sin, buf, sizeof(buf))); |
1da177e4 | 855 | |
fa9251af TM |
856 | /* Reset the inherited callbacks before calling svc_setup_socket */ |
857 | newsock->sk->sk_state_change = svsk->sk_ostate; | |
858 | newsock->sk->sk_data_ready = svsk->sk_odata; | |
859 | newsock->sk->sk_write_space = svsk->sk_owspace; | |
860 | ||
1da177e4 LT |
861 | /* make sure that a write doesn't block forever when |
862 | * low on memory | |
863 | */ | |
864 | newsock->sk->sk_sndtimeo = HZ*30; | |
865 | ||
72c35376 BF |
866 | newsvsk = svc_setup_socket(serv, newsock, |
867 | (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY)); | |
868 | if (IS_ERR(newsvsk)) | |
1da177e4 | 869 | goto failed; |
9dbc240f | 870 | svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen); |
a9747692 FM |
871 | err = kernel_getsockname(newsock, sin, &slen); |
872 | if (unlikely(err < 0)) { | |
873 | dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err); | |
874 | slen = offsetof(struct sockaddr, sa_data); | |
875 | } | |
9dbc240f | 876 | svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen); |
067d7817 | 877 | |
ef11ce24 N |
878 | if (sock_is_loopback(newsock->sk)) |
879 | set_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags); | |
880 | else | |
881 | clear_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags); | |
f9f3cc4f TT |
882 | if (serv->sv_stats) |
883 | serv->sv_stats->nettcpconn++; | |
884 | ||
885 | return &newsvsk->sk_xprt; | |
886 | ||
887 | failed: | |
888 | sock_release(newsock); | |
889 | return NULL; | |
890 | } | |
891 | ||
31d68ef6 BF |
892 | static unsigned int svc_tcp_restore_pages(struct svc_sock *svsk, struct svc_rqst *rqstp) |
893 | { | |
894 | unsigned int i, len, npages; | |
895 | ||
8af345f5 | 896 | if (svsk->sk_datalen == 0) |
31d68ef6 | 897 | return 0; |
8af345f5 | 898 | len = svsk->sk_datalen; |
31d68ef6 BF |
899 | npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
900 | for (i = 0; i < npages; i++) { | |
901 | if (rqstp->rq_pages[i] != NULL) | |
902 | put_page(rqstp->rq_pages[i]); | |
903 | BUG_ON(svsk->sk_pages[i] == NULL); | |
904 | rqstp->rq_pages[i] = svsk->sk_pages[i]; | |
905 | svsk->sk_pages[i] = NULL; | |
906 | } | |
907 | rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]); | |
908 | return len; | |
909 | } | |
910 | ||
911 | static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp) | |
912 | { | |
913 | unsigned int i, len, npages; | |
914 | ||
8af345f5 | 915 | if (svsk->sk_datalen == 0) |
31d68ef6 | 916 | return; |
8af345f5 | 917 | len = svsk->sk_datalen; |
31d68ef6 BF |
918 | npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
919 | for (i = 0; i < npages; i++) { | |
920 | svsk->sk_pages[i] = rqstp->rq_pages[i]; | |
921 | rqstp->rq_pages[i] = NULL; | |
922 | } | |
923 | } | |
924 | ||
925 | static void svc_tcp_clear_pages(struct svc_sock *svsk) | |
926 | { | |
927 | unsigned int i, len, npages; | |
928 | ||
8af345f5 | 929 | if (svsk->sk_datalen == 0) |
31d68ef6 | 930 | goto out; |
8af345f5 | 931 | len = svsk->sk_datalen; |
31d68ef6 BF |
932 | npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
933 | for (i = 0; i < npages; i++) { | |
cf3aa02c BF |
934 | if (svsk->sk_pages[i] == NULL) { |
935 | WARN_ON_ONCE(1); | |
936 | continue; | |
937 | } | |
31d68ef6 BF |
938 | put_page(svsk->sk_pages[i]); |
939 | svsk->sk_pages[i] = NULL; | |
940 | } | |
941 | out: | |
942 | svsk->sk_tcplen = 0; | |
8af345f5 | 943 | svsk->sk_datalen = 0; |
31d68ef6 BF |
944 | } |
945 | ||
1da177e4 | 946 | /* |
ad46ccf0 | 947 | * Receive fragment record header. |
8f55f3c0 | 948 | * If we haven't gotten the record length yet, get the next four bytes. |
1da177e4 | 949 | */ |
8f55f3c0 | 950 | static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp) |
1da177e4 | 951 | { |
bb5cf160 | 952 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
5ee78d48 | 953 | unsigned int want; |
8f55f3c0 | 954 | int len; |
1da177e4 | 955 | |
c0401ea0 | 956 | if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) { |
1da177e4 LT |
957 | struct kvec iov; |
958 | ||
5ee78d48 | 959 | want = sizeof(rpc_fraghdr) - svsk->sk_tcplen; |
1da177e4 LT |
960 | iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen; |
961 | iov.iov_len = want; | |
962 | if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0) | |
963 | goto error; | |
964 | svsk->sk_tcplen += len; | |
965 | ||
966 | if (len < want) { | |
c0401ea0 CL |
967 | dprintk("svc: short recvfrom while reading record " |
968 | "length (%d of %d)\n", len, want); | |
31d68ef6 | 969 | return -EAGAIN; |
1da177e4 LT |
970 | } |
971 | ||
cc248d4b | 972 | dprintk("svc: TCP record, %d bytes\n", svc_sock_reclen(svsk)); |
836fbadb BF |
973 | if (svc_sock_reclen(svsk) + svsk->sk_datalen > |
974 | serv->sv_max_mesg) { | |
3a28e331 BF |
975 | net_notice_ratelimited("RPC: fragment too large: %d\n", |
976 | svc_sock_reclen(svsk)); | |
1da177e4 LT |
977 | goto err_delete; |
978 | } | |
979 | } | |
980 | ||
cc248d4b | 981 | return svc_sock_reclen(svsk); |
31d68ef6 BF |
982 | error: |
983 | dprintk("RPC: TCP recv_record got %d\n", len); | |
8f55f3c0 | 984 | return len; |
31d68ef6 | 985 | err_delete: |
8f55f3c0 | 986 | set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); |
8f55f3c0 AB |
987 | return -EAGAIN; |
988 | } | |
989 | ||
586c52cc | 990 | static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp) |
4cfc7e60 | 991 | { |
586c52cc | 992 | struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt; |
4cfc7e60 | 993 | struct rpc_rqst *req = NULL; |
586c52cc TM |
994 | struct kvec *src, *dst; |
995 | __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base; | |
48e6555c BF |
996 | __be32 xid; |
997 | __be32 calldir; | |
4cfc7e60 | 998 | |
4cfc7e60 RI |
999 | xid = *p++; |
1000 | calldir = *p; | |
1001 | ||
093a1468 | 1002 | if (!bc_xprt) |
586c52cc | 1003 | return -EAGAIN; |
093a1468 TM |
1004 | spin_lock_bh(&bc_xprt->transport_lock); |
1005 | req = xprt_lookup_rqst(bc_xprt, xid); | |
1006 | if (!req) | |
1007 | goto unlock_notfound; | |
586c52cc TM |
1008 | |
1009 | memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf)); | |
1010 | /* | |
1011 | * XXX!: cheating for now! Only copying HEAD. | |
1012 | * But we know this is good enough for now (in fact, for any | |
1013 | * callback reply in the forseeable future). | |
1014 | */ | |
1015 | dst = &req->rq_private_buf.head[0]; | |
1016 | src = &rqstp->rq_arg.head[0]; | |
1017 | if (dst->iov_len < src->iov_len) | |
093a1468 | 1018 | goto unlock_eagain; /* whatever; just giving up. */ |
586c52cc | 1019 | memcpy(dst->iov_base, src->iov_base, src->iov_len); |
cc248d4b | 1020 | xprt_complete_rqst(req->rq_task, rqstp->rq_arg.len); |
586c52cc | 1021 | rqstp->rq_arg.len = 0; |
093a1468 | 1022 | spin_unlock_bh(&bc_xprt->transport_lock); |
586c52cc | 1023 | return 0; |
093a1468 TM |
1024 | unlock_notfound: |
1025 | printk(KERN_NOTICE | |
1026 | "%s: Got unrecognized reply: " | |
1027 | "calldir 0x%x xpt_bc_xprt %p xid %08x\n", | |
1028 | __func__, ntohl(calldir), | |
1029 | bc_xprt, ntohl(xid)); | |
1030 | unlock_eagain: | |
1031 | spin_unlock_bh(&bc_xprt->transport_lock); | |
1032 | return -EAGAIN; | |
586c52cc TM |
1033 | } |
1034 | ||
1035 | static int copy_pages_to_kvecs(struct kvec *vec, struct page **pages, int len) | |
1036 | { | |
1037 | int i = 0; | |
1038 | int t = 0; | |
1039 | ||
1040 | while (t < len) { | |
1041 | vec[i].iov_base = page_address(pages[i]); | |
1042 | vec[i].iov_len = PAGE_SIZE; | |
1043 | i++; | |
1044 | t += PAGE_SIZE; | |
1045 | } | |
1046 | return i; | |
4cfc7e60 RI |
1047 | } |
1048 | ||
836fbadb BF |
1049 | static void svc_tcp_fragment_received(struct svc_sock *svsk) |
1050 | { | |
1051 | /* If we have more data, signal svc_xprt_enqueue() to try again */ | |
836fbadb BF |
1052 | dprintk("svc: TCP %s record (%d bytes)\n", |
1053 | svc_sock_final_rec(svsk) ? "final" : "nonfinal", | |
1054 | svc_sock_reclen(svsk)); | |
1055 | svsk->sk_tcplen = 0; | |
1056 | svsk->sk_reclen = 0; | |
1057 | } | |
31d68ef6 | 1058 | |
8f55f3c0 AB |
1059 | /* |
1060 | * Receive data from a TCP socket. | |
1061 | */ | |
1062 | static int svc_tcp_recvfrom(struct svc_rqst *rqstp) | |
1063 | { | |
1064 | struct svc_sock *svsk = | |
1065 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
1066 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; | |
1067 | int len; | |
1068 | struct kvec *vec; | |
31d68ef6 | 1069 | unsigned int want, base; |
586c52cc TM |
1070 | __be32 *p; |
1071 | __be32 calldir; | |
5ee78d48 | 1072 | int pnum; |
8f55f3c0 AB |
1073 | |
1074 | dprintk("svc: tcp_recv %p data %d conn %d close %d\n", | |
1075 | svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags), | |
1076 | test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags), | |
1077 | test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags)); | |
1078 | ||
1079 | len = svc_tcp_recv_record(svsk, rqstp); | |
1080 | if (len < 0) | |
1081 | goto error; | |
1082 | ||
31d68ef6 | 1083 | base = svc_tcp_restore_pages(svsk, rqstp); |
836fbadb | 1084 | want = svc_sock_reclen(svsk) - (svsk->sk_tcplen - sizeof(rpc_fraghdr)); |
31d68ef6 | 1085 | |
3cc03b16 | 1086 | vec = rqstp->rq_vec; |
4cfc7e60 | 1087 | |
586c52cc | 1088 | pnum = copy_pages_to_kvecs(&vec[0], &rqstp->rq_pages[0], |
836fbadb | 1089 | svsk->sk_datalen + want); |
586c52cc | 1090 | |
44524359 | 1091 | rqstp->rq_respages = &rqstp->rq_pages[pnum]; |
afc59400 | 1092 | rqstp->rq_next_page = rqstp->rq_respages + 1; |
1da177e4 LT |
1093 | |
1094 | /* Now receive data */ | |
31d68ef6 | 1095 | len = svc_partial_recvfrom(rqstp, vec, pnum, want, base); |
8af345f5 | 1096 | if (len >= 0) { |
31d68ef6 | 1097 | svsk->sk_tcplen += len; |
8af345f5 BF |
1098 | svsk->sk_datalen += len; |
1099 | } | |
836fbadb | 1100 | if (len != want || !svc_sock_final_rec(svsk)) { |
be1e4444 | 1101 | svc_tcp_save_pages(svsk, rqstp); |
31d68ef6 | 1102 | if (len < 0 && len != -EAGAIN) |
ad46ccf0 | 1103 | goto err_delete; |
836fbadb BF |
1104 | if (len == want) |
1105 | svc_tcp_fragment_received(svsk); | |
1106 | else | |
3a28e331 BF |
1107 | dprintk("svc: incomplete TCP record (%d of %d)\n", |
1108 | (int)(svsk->sk_tcplen - sizeof(rpc_fraghdr)), | |
836fbadb | 1109 | svc_sock_reclen(svsk)); |
31d68ef6 BF |
1110 | goto err_noclose; |
1111 | } | |
1da177e4 | 1112 | |
1f691b07 | 1113 | if (svsk->sk_datalen < 8) { |
cf3aa02c | 1114 | svsk->sk_datalen = 0; |
ad46ccf0 | 1115 | goto err_delete; /* client is nuts. */ |
cf3aa02c | 1116 | } |
ad46ccf0 | 1117 | |
836fbadb | 1118 | rqstp->rq_arg.len = svsk->sk_datalen; |
1da177e4 | 1119 | rqstp->rq_arg.page_base = 0; |
5ee78d48 TM |
1120 | if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) { |
1121 | rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len; | |
1da177e4 | 1122 | rqstp->rq_arg.page_len = 0; |
5ee78d48 TM |
1123 | } else |
1124 | rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len; | |
1da177e4 | 1125 | |
5148bf4e | 1126 | rqstp->rq_xprt_ctxt = NULL; |
1da177e4 | 1127 | rqstp->rq_prot = IPPROTO_TCP; |
7501cc2b JL |
1128 | if (test_bit(XPT_LOCAL, &svsk->sk_xprt.xpt_flags)) |
1129 | set_bit(RQ_LOCAL, &rqstp->rq_flags); | |
1130 | else | |
1131 | clear_bit(RQ_LOCAL, &rqstp->rq_flags); | |
1da177e4 | 1132 | |
586c52cc TM |
1133 | p = (__be32 *)rqstp->rq_arg.head[0].iov_base; |
1134 | calldir = p[1]; | |
8985ef0b | 1135 | if (calldir) |
586c52cc | 1136 | len = receive_cb_reply(svsk, rqstp); |
586c52cc | 1137 | |
1da177e4 | 1138 | /* Reset TCP read info */ |
8af345f5 | 1139 | svsk->sk_datalen = 0; |
836fbadb | 1140 | svc_tcp_fragment_received(svsk); |
0601f793 | 1141 | |
8985ef0b BF |
1142 | if (len < 0) |
1143 | goto error; | |
1da177e4 | 1144 | |
9dbc240f | 1145 | svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt); |
1da177e4 LT |
1146 | if (serv->sv_stats) |
1147 | serv->sv_stats->nettcpcnt++; | |
1148 | ||
5ee78d48 | 1149 | return rqstp->rq_arg.len; |
1da177e4 | 1150 | |
8f55f3c0 | 1151 | error: |
31d68ef6 | 1152 | if (len != -EAGAIN) |
ad46ccf0 | 1153 | goto err_delete; |
31d68ef6 | 1154 | dprintk("RPC: TCP recvfrom got EAGAIN\n"); |
9f9d2ebe | 1155 | return 0; |
ad46ccf0 | 1156 | err_delete: |
31d68ef6 BF |
1157 | printk(KERN_NOTICE "%s: recvfrom returned errno %d\n", |
1158 | svsk->sk_xprt.xpt_server->sv_name, -len); | |
1159 | set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); | |
1160 | err_noclose: | |
9f9d2ebe | 1161 | return 0; /* record not complete */ |
1da177e4 LT |
1162 | } |
1163 | ||
1164 | /* | |
1165 | * Send out data on TCP socket. | |
1166 | */ | |
0f0257ea | 1167 | static int svc_tcp_sendto(struct svc_rqst *rqstp) |
1da177e4 LT |
1168 | { |
1169 | struct xdr_buf *xbufp = &rqstp->rq_res; | |
1170 | int sent; | |
d8ed029d | 1171 | __be32 reclen; |
1da177e4 LT |
1172 | |
1173 | /* Set up the first element of the reply kvec. | |
1174 | * Any other kvecs that may be in use have been taken | |
1175 | * care of by the server implementation itself. | |
1176 | */ | |
1177 | reclen = htonl(0x80000000|((xbufp->len ) - 4)); | |
1178 | memcpy(xbufp->head[0].iov_base, &reclen, 4); | |
1179 | ||
1da177e4 LT |
1180 | sent = svc_sendto(rqstp, &rqstp->rq_res); |
1181 | if (sent != xbufp->len) { | |
0f0257ea TT |
1182 | printk(KERN_NOTICE |
1183 | "rpc-srv/tcp: %s: %s %d when sending %d bytes " | |
1184 | "- shutting down socket\n", | |
57b1d3ba | 1185 | rqstp->rq_xprt->xpt_server->sv_name, |
1da177e4 LT |
1186 | (sent<0)?"got error":"sent only", |
1187 | sent, xbufp->len); | |
57b1d3ba | 1188 | set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags); |
f6150c3c | 1189 | svc_xprt_enqueue(rqstp->rq_xprt); |
1da177e4 LT |
1190 | sent = -EAGAIN; |
1191 | } | |
1192 | return sent; | |
1193 | } | |
1194 | ||
e831fe65 TT |
1195 | /* |
1196 | * Setup response header. TCP has a 4B record length field. | |
1197 | */ | |
1198 | static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp) | |
1199 | { | |
1200 | struct kvec *resv = &rqstp->rq_res.head[0]; | |
1201 | ||
1202 | /* tcp needs a space for the record length... */ | |
1203 | svc_putnl(resv, 0); | |
1204 | } | |
1205 | ||
b700cbb1 | 1206 | static struct svc_xprt *svc_tcp_create(struct svc_serv *serv, |
62832c03 | 1207 | struct net *net, |
b700cbb1 TT |
1208 | struct sockaddr *sa, int salen, |
1209 | int flags) | |
1210 | { | |
62832c03 | 1211 | return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags); |
b700cbb1 TT |
1212 | } |
1213 | ||
9e00abc3 | 1214 | #if defined(CONFIG_SUNRPC_BACKCHANNEL) |
1f11a034 AA |
1215 | static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int, |
1216 | struct net *, struct sockaddr *, | |
1217 | int, int); | |
1218 | static void svc_bc_sock_free(struct svc_xprt *xprt); | |
1219 | ||
1220 | static struct svc_xprt *svc_bc_tcp_create(struct svc_serv *serv, | |
1221 | struct net *net, | |
1222 | struct sockaddr *sa, int salen, | |
1223 | int flags) | |
1224 | { | |
1225 | return svc_bc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags); | |
1226 | } | |
1227 | ||
1228 | static void svc_bc_tcp_sock_detach(struct svc_xprt *xprt) | |
1229 | { | |
1230 | } | |
1231 | ||
1232 | static struct svc_xprt_ops svc_tcp_bc_ops = { | |
1233 | .xpo_create = svc_bc_tcp_create, | |
1234 | .xpo_detach = svc_bc_tcp_sock_detach, | |
1235 | .xpo_free = svc_bc_sock_free, | |
1236 | .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr, | |
16e4d93f | 1237 | .xpo_secure_port = svc_sock_secure_port, |
1f11a034 AA |
1238 | }; |
1239 | ||
1240 | static struct svc_xprt_class svc_tcp_bc_class = { | |
1241 | .xcl_name = "tcp-bc", | |
1242 | .xcl_owner = THIS_MODULE, | |
1243 | .xcl_ops = &svc_tcp_bc_ops, | |
1244 | .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP, | |
1245 | }; | |
16b2d1e1 AA |
1246 | |
1247 | static void svc_init_bc_xprt_sock(void) | |
1248 | { | |
1249 | svc_reg_xprt_class(&svc_tcp_bc_class); | |
1250 | } | |
1251 | ||
1252 | static void svc_cleanup_bc_xprt_sock(void) | |
1253 | { | |
1254 | svc_unreg_xprt_class(&svc_tcp_bc_class); | |
1255 | } | |
9e00abc3 | 1256 | #else /* CONFIG_SUNRPC_BACKCHANNEL */ |
16b2d1e1 AA |
1257 | static void svc_init_bc_xprt_sock(void) |
1258 | { | |
1259 | } | |
1260 | ||
1261 | static void svc_cleanup_bc_xprt_sock(void) | |
1262 | { | |
1263 | } | |
9e00abc3 | 1264 | #endif /* CONFIG_SUNRPC_BACKCHANNEL */ |
1f11a034 | 1265 | |
360d8738 | 1266 | static struct svc_xprt_ops svc_tcp_ops = { |
b700cbb1 | 1267 | .xpo_create = svc_tcp_create, |
5d137990 TT |
1268 | .xpo_recvfrom = svc_tcp_recvfrom, |
1269 | .xpo_sendto = svc_tcp_sendto, | |
5148bf4e | 1270 | .xpo_release_rqst = svc_release_skb, |
69b6ba37 | 1271 | .xpo_detach = svc_tcp_sock_detach, |
755cceab | 1272 | .xpo_free = svc_sock_free, |
e831fe65 | 1273 | .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr, |
323bee32 | 1274 | .xpo_has_wspace = svc_tcp_has_wspace, |
38a417cc | 1275 | .xpo_accept = svc_tcp_accept, |
16e4d93f | 1276 | .xpo_secure_port = svc_sock_secure_port, |
ea08e392 | 1277 | .xpo_kill_temp_xprt = svc_tcp_kill_temp_xprt, |
360d8738 TT |
1278 | }; |
1279 | ||
1280 | static struct svc_xprt_class svc_tcp_class = { | |
1281 | .xcl_name = "tcp", | |
b700cbb1 | 1282 | .xcl_owner = THIS_MODULE, |
360d8738 | 1283 | .xcl_ops = &svc_tcp_ops, |
49023155 | 1284 | .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP, |
3c45ddf8 | 1285 | .xcl_ident = XPRT_TRANSPORT_TCP, |
360d8738 TT |
1286 | }; |
1287 | ||
1288 | void svc_init_xprt_sock(void) | |
1289 | { | |
1290 | svc_reg_xprt_class(&svc_tcp_class); | |
1291 | svc_reg_xprt_class(&svc_udp_class); | |
16b2d1e1 | 1292 | svc_init_bc_xprt_sock(); |
360d8738 TT |
1293 | } |
1294 | ||
1295 | void svc_cleanup_xprt_sock(void) | |
1296 | { | |
1297 | svc_unreg_xprt_class(&svc_tcp_class); | |
1298 | svc_unreg_xprt_class(&svc_udp_class); | |
16b2d1e1 | 1299 | svc_cleanup_bc_xprt_sock(); |
360d8738 TT |
1300 | } |
1301 | ||
bb5cf160 | 1302 | static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv) |
1da177e4 LT |
1303 | { |
1304 | struct sock *sk = svsk->sk_sk; | |
1da177e4 | 1305 | |
bd4620dd SK |
1306 | svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class, |
1307 | &svsk->sk_xprt, serv); | |
def13d74 | 1308 | set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
1309 | if (sk->sk_state == TCP_LISTEN) { |
1310 | dprintk("setting up TCP socket for listening\n"); | |
02fc6c36 | 1311 | set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 1312 | sk->sk_data_ready = svc_tcp_listen_data_ready; |
02fc6c36 | 1313 | set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
1314 | } else { |
1315 | dprintk("setting up TCP socket for reading\n"); | |
1316 | sk->sk_state_change = svc_tcp_state_change; | |
fa9251af | 1317 | sk->sk_data_ready = svc_data_ready; |
637600f3 | 1318 | sk->sk_write_space = svc_write_space; |
1da177e4 LT |
1319 | |
1320 | svsk->sk_reclen = 0; | |
1321 | svsk->sk_tcplen = 0; | |
8af345f5 | 1322 | svsk->sk_datalen = 0; |
31d68ef6 | 1323 | memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages)); |
1da177e4 | 1324 | |
b7872fe8 | 1325 | tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF; |
1da177e4 | 1326 | |
02fc6c36 | 1327 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
b2f21f7d TM |
1328 | switch (sk->sk_state) { |
1329 | case TCP_SYN_RECV: | |
1330 | case TCP_ESTABLISHED: | |
1331 | break; | |
1332 | default: | |
02fc6c36 | 1333 | set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); |
b2f21f7d | 1334 | } |
1da177e4 LT |
1335 | } |
1336 | } | |
1337 | ||
0f0257ea | 1338 | void svc_sock_update_bufs(struct svc_serv *serv) |
1da177e4 LT |
1339 | { |
1340 | /* | |
1341 | * The number of server threads has changed. Update | |
1342 | * rcvbuf and sndbuf accordingly on all sockets | |
1343 | */ | |
8f3a6de3 | 1344 | struct svc_sock *svsk; |
1da177e4 LT |
1345 | |
1346 | spin_lock_bh(&serv->sv_lock); | |
8f3a6de3 | 1347 | list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list) |
02fc6c36 | 1348 | set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
1349 | spin_unlock_bh(&serv->sv_lock); |
1350 | } | |
24c3767e | 1351 | EXPORT_SYMBOL_GPL(svc_sock_update_bufs); |
1da177e4 | 1352 | |
1da177e4 LT |
1353 | /* |
1354 | * Initialize socket for RPC use and create svc_sock struct | |
1da177e4 | 1355 | */ |
6b174337 CL |
1356 | static struct svc_sock *svc_setup_socket(struct svc_serv *serv, |
1357 | struct socket *sock, | |
72c35376 | 1358 | int flags) |
1da177e4 LT |
1359 | { |
1360 | struct svc_sock *svsk; | |
1361 | struct sock *inet; | |
6b174337 | 1362 | int pmap_register = !(flags & SVC_SOCK_ANONYMOUS); |
72c35376 | 1363 | int err = 0; |
1da177e4 LT |
1364 | |
1365 | dprintk("svc: svc_setup_socket %p\n", sock); | |
72c35376 BF |
1366 | svsk = kzalloc(sizeof(*svsk), GFP_KERNEL); |
1367 | if (!svsk) | |
1368 | return ERR_PTR(-ENOMEM); | |
1da177e4 LT |
1369 | |
1370 | inet = sock->sk; | |
1371 | ||
1372 | /* Register socket with portmapper */ | |
72c35376 BF |
1373 | if (pmap_register) |
1374 | err = svc_register(serv, sock_net(sock->sk), inet->sk_family, | |
5247fab5 | 1375 | inet->sk_protocol, |
c720c7e8 | 1376 | ntohs(inet_sk(inet)->inet_sport)); |
1da177e4 | 1377 | |
72c35376 | 1378 | if (err < 0) { |
1da177e4 | 1379 | kfree(svsk); |
72c35376 | 1380 | return ERR_PTR(err); |
1da177e4 LT |
1381 | } |
1382 | ||
1da177e4 LT |
1383 | inet->sk_user_data = svsk; |
1384 | svsk->sk_sock = sock; | |
1385 | svsk->sk_sk = inet; | |
1386 | svsk->sk_ostate = inet->sk_state_change; | |
1387 | svsk->sk_odata = inet->sk_data_ready; | |
1388 | svsk->sk_owspace = inet->sk_write_space; | |
1da177e4 LT |
1389 | |
1390 | /* Initialize the socket */ | |
1391 | if (sock->type == SOCK_DGRAM) | |
bb5cf160 | 1392 | svc_udp_init(svsk, serv); |
637600f3 | 1393 | else |
bb5cf160 | 1394 | svc_tcp_init(svsk, serv); |
1da177e4 | 1395 | |
c7995f8a TM |
1396 | dprintk("svc: svc_setup_socket created %p (inet %p), " |
1397 | "listen %d close %d\n", | |
1398 | svsk, svsk->sk_sk, | |
1399 | test_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags), | |
1400 | test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags)); | |
1da177e4 | 1401 | |
1da177e4 LT |
1402 | return svsk; |
1403 | } | |
1404 | ||
30646394 SK |
1405 | bool svc_alien_sock(struct net *net, int fd) |
1406 | { | |
1407 | int err; | |
1408 | struct socket *sock = sockfd_lookup(fd, &err); | |
1409 | bool ret = false; | |
1410 | ||
1411 | if (!sock) | |
1412 | goto out; | |
1413 | if (sock_net(sock->sk) != net) | |
1414 | ret = true; | |
1415 | sockfd_put(sock); | |
1416 | out: | |
1417 | return ret; | |
1418 | } | |
1419 | EXPORT_SYMBOL_GPL(svc_alien_sock); | |
1420 | ||
bfba9ab4 CL |
1421 | /** |
1422 | * svc_addsock - add a listener socket to an RPC service | |
1423 | * @serv: pointer to RPC service to which to add a new listener | |
1424 | * @fd: file descriptor of the new listener | |
1425 | * @name_return: pointer to buffer to fill in with name of listener | |
1426 | * @len: size of the buffer | |
1427 | * | |
1428 | * Fills in socket name and returns positive length of name if successful. | |
1429 | * Name is terminated with '\n'. On error, returns a negative errno | |
1430 | * value. | |
1431 | */ | |
1432 | int svc_addsock(struct svc_serv *serv, const int fd, char *name_return, | |
1433 | const size_t len) | |
b41b66d6 N |
1434 | { |
1435 | int err = 0; | |
1436 | struct socket *so = sockfd_lookup(fd, &err); | |
1437 | struct svc_sock *svsk = NULL; | |
a8e10078 BF |
1438 | struct sockaddr_storage addr; |
1439 | struct sockaddr *sin = (struct sockaddr *)&addr; | |
1440 | int salen; | |
b41b66d6 N |
1441 | |
1442 | if (!so) | |
1443 | return err; | |
a8e10078 | 1444 | err = -EAFNOSUPPORT; |
205ba423 | 1445 | if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6)) |
a8e10078 BF |
1446 | goto out; |
1447 | err = -EPROTONOSUPPORT; | |
1448 | if (so->sk->sk_protocol != IPPROTO_TCP && | |
b41b66d6 | 1449 | so->sk->sk_protocol != IPPROTO_UDP) |
a8e10078 BF |
1450 | goto out; |
1451 | err = -EISCONN; | |
1452 | if (so->state > SS_UNCONNECTED) | |
1453 | goto out; | |
1454 | err = -ENOENT; | |
1455 | if (!try_module_get(THIS_MODULE)) | |
1456 | goto out; | |
1457 | svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS); | |
1458 | if (IS_ERR(svsk)) { | |
1459 | module_put(THIS_MODULE); | |
1460 | err = PTR_ERR(svsk); | |
1461 | goto out; | |
b41b66d6 | 1462 | } |
a8e10078 BF |
1463 | if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0) |
1464 | svc_xprt_set_local(&svsk->sk_xprt, sin, salen); | |
39b55301 | 1465 | svc_add_new_perm_xprt(serv, &svsk->sk_xprt); |
e7942b9f | 1466 | return svc_one_sock_name(svsk, name_return, len); |
a8e10078 BF |
1467 | out: |
1468 | sockfd_put(so); | |
1469 | return err; | |
b41b66d6 N |
1470 | } |
1471 | EXPORT_SYMBOL_GPL(svc_addsock); | |
1472 | ||
1da177e4 LT |
1473 | /* |
1474 | * Create socket for RPC service. | |
1475 | */ | |
b700cbb1 TT |
1476 | static struct svc_xprt *svc_create_socket(struct svc_serv *serv, |
1477 | int protocol, | |
62832c03 | 1478 | struct net *net, |
b700cbb1 TT |
1479 | struct sockaddr *sin, int len, |
1480 | int flags) | |
1da177e4 LT |
1481 | { |
1482 | struct svc_sock *svsk; | |
1483 | struct socket *sock; | |
1484 | int error; | |
1485 | int type; | |
9dbc240f TT |
1486 | struct sockaddr_storage addr; |
1487 | struct sockaddr *newsin = (struct sockaddr *)&addr; | |
1488 | int newlen; | |
c69da774 TM |
1489 | int family; |
1490 | int val; | |
5216a8e7 | 1491 | RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); |
1da177e4 | 1492 | |
ad06e4bd CL |
1493 | dprintk("svc: svc_create_socket(%s, %d, %s)\n", |
1494 | serv->sv_program->pg_name, protocol, | |
77f1f67a | 1495 | __svc_print_addr(sin, buf, sizeof(buf))); |
1da177e4 LT |
1496 | |
1497 | if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) { | |
1498 | printk(KERN_WARNING "svc: only UDP and TCP " | |
1499 | "sockets supported\n"); | |
b700cbb1 | 1500 | return ERR_PTR(-EINVAL); |
1da177e4 | 1501 | } |
c69da774 | 1502 | |
1da177e4 | 1503 | type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM; |
c69da774 TM |
1504 | switch (sin->sa_family) { |
1505 | case AF_INET6: | |
1506 | family = PF_INET6; | |
1507 | break; | |
1508 | case AF_INET: | |
1509 | family = PF_INET; | |
1510 | break; | |
1511 | default: | |
1512 | return ERR_PTR(-EINVAL); | |
1513 | } | |
1da177e4 | 1514 | |
14ec63c3 | 1515 | error = __sock_create(net, family, type, protocol, &sock, 1); |
77f1f67a | 1516 | if (error < 0) |
b700cbb1 | 1517 | return ERR_PTR(error); |
1da177e4 | 1518 | |
ed07536e PZ |
1519 | svc_reclassify_socket(sock); |
1520 | ||
c69da774 TM |
1521 | /* |
1522 | * If this is an PF_INET6 listener, we want to avoid | |
1523 | * getting requests from IPv4 remotes. Those should | |
1524 | * be shunted to a PF_INET listener via rpcbind. | |
1525 | */ | |
1526 | val = 1; | |
1527 | if (family == PF_INET6) | |
1528 | kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY, | |
1529 | (char *)&val, sizeof(val)); | |
1530 | ||
18114746 | 1531 | if (type == SOCK_STREAM) |
4a17fd52 | 1532 | sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */ |
77f1f67a | 1533 | error = kernel_bind(sock, sin, len); |
18114746 ES |
1534 | if (error < 0) |
1535 | goto bummer; | |
1da177e4 | 1536 | |
9dbc240f TT |
1537 | newlen = len; |
1538 | error = kernel_getsockname(sock, newsin, &newlen); | |
1539 | if (error < 0) | |
1540 | goto bummer; | |
1541 | ||
1da177e4 | 1542 | if (protocol == IPPROTO_TCP) { |
e6242e92 | 1543 | if ((error = kernel_listen(sock, 64)) < 0) |
1da177e4 LT |
1544 | goto bummer; |
1545 | } | |
1546 | ||
72c35376 | 1547 | svsk = svc_setup_socket(serv, sock, flags); |
a8e10078 BF |
1548 | if (IS_ERR(svsk)) { |
1549 | error = PTR_ERR(svsk); | |
1550 | goto bummer; | |
e79eff1f | 1551 | } |
a8e10078 BF |
1552 | svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen); |
1553 | return (struct svc_xprt *)svsk; | |
1da177e4 LT |
1554 | bummer: |
1555 | dprintk("svc: svc_create_socket error = %d\n", -error); | |
1556 | sock_release(sock); | |
b700cbb1 | 1557 | return ERR_PTR(error); |
1da177e4 LT |
1558 | } |
1559 | ||
755cceab TT |
1560 | /* |
1561 | * Detach the svc_sock from the socket so that no | |
1562 | * more callbacks occur. | |
1563 | */ | |
1564 | static void svc_sock_detach(struct svc_xprt *xprt) | |
1565 | { | |
1566 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
1567 | struct sock *sk = svsk->sk_sk; | |
1568 | ||
1569 | dprintk("svc: svc_sock_detach(%p)\n", svsk); | |
1570 | ||
1571 | /* put back the old socket callbacks */ | |
069c225b | 1572 | lock_sock(sk); |
755cceab TT |
1573 | sk->sk_state_change = svsk->sk_ostate; |
1574 | sk->sk_data_ready = svsk->sk_odata; | |
1575 | sk->sk_write_space = svsk->sk_owspace; | |
069c225b TM |
1576 | sk->sk_user_data = NULL; |
1577 | release_sock(sk); | |
69b6ba37 TM |
1578 | } |
1579 | ||
1580 | /* | |
1581 | * Disconnect the socket, and reset the callbacks | |
1582 | */ | |
1583 | static void svc_tcp_sock_detach(struct svc_xprt *xprt) | |
1584 | { | |
1585 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
1586 | ||
1587 | dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk); | |
1588 | ||
1589 | svc_sock_detach(xprt); | |
1590 | ||
31d68ef6 BF |
1591 | if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) { |
1592 | svc_tcp_clear_pages(svsk); | |
69b6ba37 | 1593 | kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR); |
31d68ef6 | 1594 | } |
755cceab TT |
1595 | } |
1596 | ||
1597 | /* | |
1598 | * Free the svc_sock's socket resources and the svc_sock itself. | |
1599 | */ | |
1600 | static void svc_sock_free(struct svc_xprt *xprt) | |
1601 | { | |
1602 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
1603 | dprintk("svc: svc_sock_free(%p)\n", svsk); | |
1604 | ||
755cceab TT |
1605 | if (svsk->sk_sock->file) |
1606 | sockfd_put(svsk->sk_sock); | |
1607 | else | |
1608 | sock_release(svsk->sk_sock); | |
1609 | kfree(svsk); | |
1610 | } | |
7652e5a0 | 1611 | |
9e00abc3 | 1612 | #if defined(CONFIG_SUNRPC_BACKCHANNEL) |
7652e5a0 | 1613 | /* |
1f11a034 | 1614 | * Create a back channel svc_xprt which shares the fore channel socket. |
7652e5a0 | 1615 | */ |
1f11a034 AA |
1616 | static struct svc_xprt *svc_bc_create_socket(struct svc_serv *serv, |
1617 | int protocol, | |
1618 | struct net *net, | |
1619 | struct sockaddr *sin, int len, | |
1620 | int flags) | |
7652e5a0 BH |
1621 | { |
1622 | struct svc_sock *svsk; | |
1f11a034 AA |
1623 | struct svc_xprt *xprt; |
1624 | ||
1625 | if (protocol != IPPROTO_TCP) { | |
1626 | printk(KERN_WARNING "svc: only TCP sockets" | |
1627 | " supported on shared back channel\n"); | |
1628 | return ERR_PTR(-EINVAL); | |
1629 | } | |
7652e5a0 | 1630 | |
7652e5a0 BH |
1631 | svsk = kzalloc(sizeof(*svsk), GFP_KERNEL); |
1632 | if (!svsk) | |
1f11a034 | 1633 | return ERR_PTR(-ENOMEM); |
7652e5a0 BH |
1634 | |
1635 | xprt = &svsk->sk_xprt; | |
bd4620dd | 1636 | svc_xprt_init(net, &svc_tcp_bc_class, xprt, serv); |
1f11a034 | 1637 | |
4a19de0f | 1638 | serv->sv_bc_xprt = xprt; |
1f11a034 | 1639 | |
7652e5a0 BH |
1640 | return xprt; |
1641 | } | |
7652e5a0 BH |
1642 | |
1643 | /* | |
1f11a034 | 1644 | * Free a back channel svc_sock. |
7652e5a0 | 1645 | */ |
1f11a034 | 1646 | static void svc_bc_sock_free(struct svc_xprt *xprt) |
7652e5a0 | 1647 | { |
778be232 | 1648 | if (xprt) |
7652e5a0 BH |
1649 | kfree(container_of(xprt, struct svc_sock, sk_xprt)); |
1650 | } | |
9e00abc3 | 1651 | #endif /* CONFIG_SUNRPC_BACKCHANNEL */ |