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 LT |
23 | #include <linux/sched.h> |
24 | #include <linux/errno.h> | |
25 | #include <linux/fcntl.h> | |
26 | #include <linux/net.h> | |
27 | #include <linux/in.h> | |
28 | #include <linux/inet.h> | |
29 | #include <linux/udp.h> | |
91483c4b | 30 | #include <linux/tcp.h> |
1da177e4 LT |
31 | #include <linux/unistd.h> |
32 | #include <linux/slab.h> | |
33 | #include <linux/netdevice.h> | |
34 | #include <linux/skbuff.h> | |
b41b66d6 | 35 | #include <linux/file.h> |
7dfb7103 | 36 | #include <linux/freezer.h> |
1da177e4 LT |
37 | #include <net/sock.h> |
38 | #include <net/checksum.h> | |
39 | #include <net/ip.h> | |
b92503b2 | 40 | #include <net/ipv6.h> |
b7872fe8 | 41 | #include <net/tcp.h> |
c752f073 | 42 | #include <net/tcp_states.h> |
1da177e4 LT |
43 | #include <asm/uaccess.h> |
44 | #include <asm/ioctls.h> | |
45 | ||
46 | #include <linux/sunrpc/types.h> | |
ad06e4bd | 47 | #include <linux/sunrpc/clnt.h> |
1da177e4 | 48 | #include <linux/sunrpc/xdr.h> |
c0401ea0 | 49 | #include <linux/sunrpc/msg_prot.h> |
1da177e4 LT |
50 | #include <linux/sunrpc/svcsock.h> |
51 | #include <linux/sunrpc/stats.h> | |
4cfc7e60 | 52 | #include <linux/sunrpc/xprt.h> |
1da177e4 | 53 | |
360d8738 | 54 | #define RPCDBG_FACILITY RPCDBG_SVCXPRT |
1da177e4 LT |
55 | |
56 | ||
57 | static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *, | |
6b174337 | 58 | int *errp, int flags); |
1da177e4 LT |
59 | static void svc_udp_data_ready(struct sock *, int); |
60 | static int svc_udp_recvfrom(struct svc_rqst *); | |
61 | static int svc_udp_sendto(struct svc_rqst *); | |
755cceab | 62 | static void svc_sock_detach(struct svc_xprt *); |
69b6ba37 | 63 | static void svc_tcp_sock_detach(struct svc_xprt *); |
755cceab | 64 | static void svc_sock_free(struct svc_xprt *); |
1da177e4 | 65 | |
b700cbb1 TT |
66 | static struct svc_xprt *svc_create_socket(struct svc_serv *, int, |
67 | struct sockaddr *, int, int); | |
ed07536e PZ |
68 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
69 | static struct lock_class_key svc_key[2]; | |
70 | static struct lock_class_key svc_slock_key[2]; | |
71 | ||
0f0257ea | 72 | static void svc_reclassify_socket(struct socket *sock) |
ed07536e PZ |
73 | { |
74 | struct sock *sk = sock->sk; | |
02b3d346 | 75 | BUG_ON(sock_owned_by_user(sk)); |
ed07536e PZ |
76 | switch (sk->sk_family) { |
77 | case AF_INET: | |
78 | sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD", | |
def13d74 TT |
79 | &svc_slock_key[0], |
80 | "sk_xprt.xpt_lock-AF_INET-NFSD", | |
81 | &svc_key[0]); | |
ed07536e PZ |
82 | break; |
83 | ||
84 | case AF_INET6: | |
85 | sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD", | |
def13d74 TT |
86 | &svc_slock_key[1], |
87 | "sk_xprt.xpt_lock-AF_INET6-NFSD", | |
88 | &svc_key[1]); | |
ed07536e PZ |
89 | break; |
90 | ||
91 | default: | |
92 | BUG(); | |
93 | } | |
94 | } | |
95 | #else | |
0f0257ea | 96 | static void svc_reclassify_socket(struct socket *sock) |
ed07536e PZ |
97 | { |
98 | } | |
99 | #endif | |
100 | ||
1da177e4 LT |
101 | /* |
102 | * Release an skbuff after use | |
103 | */ | |
5148bf4e | 104 | static void svc_release_skb(struct svc_rqst *rqstp) |
1da177e4 | 105 | { |
5148bf4e | 106 | struct sk_buff *skb = rqstp->rq_xprt_ctxt; |
1da177e4 LT |
107 | |
108 | if (skb) { | |
57b1d3ba TT |
109 | struct svc_sock *svsk = |
110 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
5148bf4e | 111 | rqstp->rq_xprt_ctxt = NULL; |
1da177e4 LT |
112 | |
113 | dprintk("svc: service %p, releasing skb %p\n", rqstp, skb); | |
9d410c79 | 114 | skb_free_datagram_locked(svsk->sk_sk, skb); |
1da177e4 | 115 | } |
1da177e4 LT |
116 | } |
117 | ||
b92503b2 CL |
118 | union svc_pktinfo_u { |
119 | struct in_pktinfo pkti; | |
b92503b2 | 120 | struct in6_pktinfo pkti6; |
b92503b2 | 121 | }; |
bc375ea7 DM |
122 | #define SVC_PKTINFO_SPACE \ |
123 | CMSG_SPACE(sizeof(union svc_pktinfo_u)) | |
b92503b2 CL |
124 | |
125 | static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh) | |
126 | { | |
57b1d3ba TT |
127 | struct svc_sock *svsk = |
128 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
129 | switch (svsk->sk_sk->sk_family) { | |
b92503b2 CL |
130 | case AF_INET: { |
131 | struct in_pktinfo *pki = CMSG_DATA(cmh); | |
132 | ||
133 | cmh->cmsg_level = SOL_IP; | |
134 | cmh->cmsg_type = IP_PKTINFO; | |
135 | pki->ipi_ifindex = 0; | |
136 | pki->ipi_spec_dst.s_addr = rqstp->rq_daddr.addr.s_addr; | |
137 | cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); | |
138 | } | |
139 | break; | |
5a05ed73 | 140 | |
b92503b2 CL |
141 | case AF_INET6: { |
142 | struct in6_pktinfo *pki = CMSG_DATA(cmh); | |
143 | ||
144 | cmh->cmsg_level = SOL_IPV6; | |
145 | cmh->cmsg_type = IPV6_PKTINFO; | |
146 | pki->ipi6_ifindex = 0; | |
147 | ipv6_addr_copy(&pki->ipi6_addr, | |
148 | &rqstp->rq_daddr.addr6); | |
149 | cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); | |
150 | } | |
151 | break; | |
b92503b2 CL |
152 | } |
153 | return; | |
154 | } | |
155 | ||
1da177e4 | 156 | /* |
4cfc7e60 | 157 | * send routine intended to be shared by the fore- and back-channel |
1da177e4 | 158 | */ |
4cfc7e60 RI |
159 | int svc_send_common(struct socket *sock, struct xdr_buf *xdr, |
160 | struct page *headpage, unsigned long headoffset, | |
161 | struct page *tailpage, unsigned long tailoffset) | |
1da177e4 | 162 | { |
1da177e4 LT |
163 | int result; |
164 | int size; | |
165 | struct page **ppage = xdr->pages; | |
166 | size_t base = xdr->page_base; | |
167 | unsigned int pglen = xdr->page_len; | |
168 | unsigned int flags = MSG_MORE; | |
4cfc7e60 RI |
169 | int slen; |
170 | int len = 0; | |
1da177e4 LT |
171 | |
172 | slen = xdr->len; | |
173 | ||
1da177e4 LT |
174 | /* send head */ |
175 | if (slen == xdr->head[0].iov_len) | |
176 | flags = 0; | |
4cfc7e60 | 177 | len = kernel_sendpage(sock, headpage, headoffset, |
44524359 | 178 | xdr->head[0].iov_len, flags); |
1da177e4 LT |
179 | if (len != xdr->head[0].iov_len) |
180 | goto out; | |
181 | slen -= xdr->head[0].iov_len; | |
182 | if (slen == 0) | |
183 | goto out; | |
184 | ||
185 | /* send page data */ | |
186 | size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen; | |
187 | while (pglen > 0) { | |
188 | if (slen == size) | |
189 | flags = 0; | |
e6242e92 | 190 | result = kernel_sendpage(sock, *ppage, base, size, flags); |
1da177e4 LT |
191 | if (result > 0) |
192 | len += result; | |
193 | if (result != size) | |
194 | goto out; | |
195 | slen -= size; | |
196 | pglen -= size; | |
197 | size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen; | |
198 | base = 0; | |
199 | ppage++; | |
200 | } | |
4cfc7e60 | 201 | |
1da177e4 LT |
202 | /* send tail */ |
203 | if (xdr->tail[0].iov_len) { | |
4cfc7e60 RI |
204 | result = kernel_sendpage(sock, tailpage, tailoffset, |
205 | xdr->tail[0].iov_len, 0); | |
1da177e4 LT |
206 | if (result > 0) |
207 | len += result; | |
208 | } | |
4cfc7e60 RI |
209 | |
210 | out: | |
211 | return len; | |
212 | } | |
213 | ||
214 | ||
215 | /* | |
216 | * Generic sendto routine | |
217 | */ | |
218 | static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr) | |
219 | { | |
220 | struct svc_sock *svsk = | |
221 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
222 | struct socket *sock = svsk->sk_sock; | |
223 | union { | |
224 | struct cmsghdr hdr; | |
225 | long all[SVC_PKTINFO_SPACE / sizeof(long)]; | |
226 | } buffer; | |
227 | struct cmsghdr *cmh = &buffer.hdr; | |
228 | int len = 0; | |
229 | unsigned long tailoff; | |
230 | unsigned long headoff; | |
231 | RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); | |
232 | ||
233 | if (rqstp->rq_prot == IPPROTO_UDP) { | |
234 | struct msghdr msg = { | |
235 | .msg_name = &rqstp->rq_addr, | |
236 | .msg_namelen = rqstp->rq_addrlen, | |
237 | .msg_control = cmh, | |
238 | .msg_controllen = sizeof(buffer), | |
239 | .msg_flags = MSG_MORE, | |
240 | }; | |
241 | ||
242 | svc_set_cmsg_data(rqstp, cmh); | |
243 | ||
244 | if (sock_sendmsg(sock, &msg, 0) < 0) | |
245 | goto out; | |
246 | } | |
247 | ||
248 | tailoff = ((unsigned long)xdr->tail[0].iov_base) & (PAGE_SIZE-1); | |
249 | headoff = 0; | |
250 | len = svc_send_common(sock, xdr, rqstp->rq_respages[0], headoff, | |
251 | rqstp->rq_respages[0], tailoff); | |
252 | ||
1da177e4 | 253 | out: |
ad06e4bd | 254 | dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n", |
57b1d3ba | 255 | svsk, xdr->head[0].iov_base, xdr->head[0].iov_len, |
ad06e4bd | 256 | xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf))); |
1da177e4 LT |
257 | |
258 | return len; | |
259 | } | |
260 | ||
80212d59 N |
261 | /* |
262 | * Report socket names for nfsdfs | |
263 | */ | |
e7942b9f | 264 | static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining) |
80212d59 | 265 | { |
017cb47f CL |
266 | const struct sock *sk = svsk->sk_sk; |
267 | const char *proto_name = sk->sk_protocol == IPPROTO_UDP ? | |
268 | "udp" : "tcp"; | |
80212d59 N |
269 | int len; |
270 | ||
017cb47f | 271 | switch (sk->sk_family) { |
e7942b9f CL |
272 | case PF_INET: |
273 | len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n", | |
017cb47f | 274 | proto_name, |
c720c7e8 ED |
275 | &inet_sk(sk)->inet_rcv_saddr, |
276 | inet_sk(sk)->inet_num); | |
80212d59 | 277 | break; |
58de2f86 CL |
278 | case PF_INET6: |
279 | len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n", | |
017cb47f CL |
280 | proto_name, |
281 | &inet6_sk(sk)->rcv_saddr, | |
c720c7e8 | 282 | inet_sk(sk)->inet_num); |
80212d59 N |
283 | break; |
284 | default: | |
e7942b9f | 285 | len = snprintf(buf, remaining, "*unknown-%d*\n", |
017cb47f | 286 | sk->sk_family); |
80212d59 | 287 | } |
e7942b9f CL |
288 | |
289 | if (len >= remaining) { | |
290 | *buf = '\0'; | |
291 | return -ENAMETOOLONG; | |
80212d59 N |
292 | } |
293 | return len; | |
294 | } | |
295 | ||
8435d34d CL |
296 | /** |
297 | * svc_sock_names - construct a list of listener names in a string | |
298 | * @serv: pointer to RPC service | |
299 | * @buf: pointer to a buffer to fill in with socket names | |
300 | * @buflen: size of the buffer to be filled | |
301 | * @toclose: pointer to '\0'-terminated C string containing the name | |
302 | * of a listener to be closed | |
303 | * | |
304 | * Fills in @buf with a '\n'-separated list of names of listener | |
305 | * sockets. If @toclose is not NULL, the socket named by @toclose | |
306 | * is closed, and is not included in the output list. | |
307 | * | |
308 | * Returns positive length of the socket name string, or a negative | |
309 | * errno value on error. | |
310 | */ | |
311 | int svc_sock_names(struct svc_serv *serv, char *buf, const size_t buflen, | |
312 | const char *toclose) | |
80212d59 | 313 | { |
b41b66d6 | 314 | struct svc_sock *svsk, *closesk = NULL; |
80212d59 N |
315 | int len = 0; |
316 | ||
317 | if (!serv) | |
318 | return 0; | |
e7942b9f | 319 | |
aaf68cfb | 320 | spin_lock_bh(&serv->sv_lock); |
7a182083 | 321 | list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list) { |
e7942b9f CL |
322 | int onelen = svc_one_sock_name(svsk, buf + len, buflen - len); |
323 | if (onelen < 0) { | |
324 | len = onelen; | |
325 | break; | |
326 | } | |
327 | if (toclose && strcmp(toclose, buf + len) == 0) | |
b41b66d6 N |
328 | closesk = svsk; |
329 | else | |
330 | len += onelen; | |
80212d59 | 331 | } |
aaf68cfb | 332 | spin_unlock_bh(&serv->sv_lock); |
e7942b9f | 333 | |
b41b66d6 | 334 | if (closesk) |
5680c446 N |
335 | /* Should unregister with portmap, but you cannot |
336 | * unregister just one protocol... | |
337 | */ | |
7a182083 | 338 | svc_close_xprt(&closesk->sk_xprt); |
37a03472 N |
339 | else if (toclose) |
340 | return -ENOENT; | |
80212d59 N |
341 | return len; |
342 | } | |
24c3767e | 343 | EXPORT_SYMBOL_GPL(svc_sock_names); |
80212d59 | 344 | |
1da177e4 LT |
345 | /* |
346 | * Check input queue length | |
347 | */ | |
0f0257ea | 348 | static int svc_recv_available(struct svc_sock *svsk) |
1da177e4 | 349 | { |
1da177e4 LT |
350 | struct socket *sock = svsk->sk_sock; |
351 | int avail, err; | |
352 | ||
e6242e92 | 353 | err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail); |
1da177e4 LT |
354 | |
355 | return (err >= 0)? avail : err; | |
356 | } | |
357 | ||
358 | /* | |
359 | * Generic recvfrom routine. | |
360 | */ | |
0f0257ea TT |
361 | static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr, |
362 | int buflen) | |
1da177e4 | 363 | { |
57b1d3ba TT |
364 | struct svc_sock *svsk = |
365 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
1ba95105 CL |
366 | struct msghdr msg = { |
367 | .msg_flags = MSG_DONTWAIT, | |
368 | }; | |
369 | int len; | |
1da177e4 | 370 | |
260c1d12 TT |
371 | rqstp->rq_xprt_hlen = 0; |
372 | ||
1ba95105 CL |
373 | len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen, |
374 | msg.msg_flags); | |
1da177e4 | 375 | |
1da177e4 | 376 | dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n", |
1ba95105 | 377 | svsk, iov[0].iov_base, iov[0].iov_len, len); |
1da177e4 LT |
378 | return len; |
379 | } | |
380 | ||
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; | |
7f421835 | 403 | sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK|SOCK_RCVBUF_LOCK; |
47fcb03f | 404 | sock->sk->sk_write_space(sock->sk); |
1da177e4 LT |
405 | release_sock(sock->sk); |
406 | #endif | |
407 | } | |
408 | /* | |
409 | * INET callback when data has been received on the socket. | |
410 | */ | |
0f0257ea | 411 | static void svc_udp_data_ready(struct sock *sk, int count) |
1da177e4 | 412 | { |
939bb7ef | 413 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 | 414 | |
939bb7ef NB |
415 | if (svsk) { |
416 | dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n", | |
02fc6c36 TT |
417 | svsk, sk, count, |
418 | test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags)); | |
419 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); | |
f6150c3c | 420 | svc_xprt_enqueue(&svsk->sk_xprt); |
939bb7ef | 421 | } |
1da177e4 LT |
422 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
423 | wake_up_interruptible(sk->sk_sleep); | |
424 | } | |
425 | ||
426 | /* | |
427 | * INET callback when space is newly available on the socket. | |
428 | */ | |
0f0257ea | 429 | static void svc_write_space(struct sock *sk) |
1da177e4 LT |
430 | { |
431 | struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data); | |
432 | ||
433 | if (svsk) { | |
434 | dprintk("svc: socket %p(inet %p), write_space busy=%d\n", | |
02fc6c36 | 435 | svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags)); |
f6150c3c | 436 | svc_xprt_enqueue(&svsk->sk_xprt); |
1da177e4 LT |
437 | } |
438 | ||
439 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) { | |
939bb7ef | 440 | dprintk("RPC svc_write_space: someone sleeping on %p\n", |
1da177e4 LT |
441 | svsk); |
442 | wake_up_interruptible(sk->sk_sleep); | |
443 | } | |
444 | } | |
445 | ||
47fcb03f TM |
446 | static void svc_tcp_write_space(struct sock *sk) |
447 | { | |
448 | struct socket *sock = sk->sk_socket; | |
449 | ||
450 | if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && sock) | |
451 | clear_bit(SOCK_NOSPACE, &sock->flags); | |
452 | svc_write_space(sk); | |
453 | } | |
454 | ||
7702ce40 CL |
455 | /* |
456 | * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo | |
457 | */ | |
458 | static int svc_udp_get_dest_address4(struct svc_rqst *rqstp, | |
459 | struct cmsghdr *cmh) | |
460 | { | |
461 | struct in_pktinfo *pki = CMSG_DATA(cmh); | |
462 | if (cmh->cmsg_type != IP_PKTINFO) | |
463 | return 0; | |
464 | rqstp->rq_daddr.addr.s_addr = pki->ipi_spec_dst.s_addr; | |
465 | return 1; | |
466 | } | |
467 | ||
468 | /* | |
469 | * See net/ipv6/datagram.c : datagram_recv_ctl | |
470 | */ | |
471 | static int svc_udp_get_dest_address6(struct svc_rqst *rqstp, | |
472 | struct cmsghdr *cmh) | |
473 | { | |
474 | struct in6_pktinfo *pki = CMSG_DATA(cmh); | |
475 | if (cmh->cmsg_type != IPV6_PKTINFO) | |
476 | return 0; | |
477 | ipv6_addr_copy(&rqstp->rq_daddr.addr6, &pki->ipi6_addr); | |
478 | return 1; | |
479 | } | |
480 | ||
9dbc240f TT |
481 | /* |
482 | * Copy the UDP datagram's destination address to the rqstp structure. | |
483 | * The 'destination' address in this case is the address to which the | |
484 | * peer sent the datagram, i.e. our local address. For multihomed | |
485 | * hosts, this can change from msg to msg. Note that only the IP | |
486 | * address changes, the port number should remain the same. | |
487 | */ | |
7702ce40 CL |
488 | static int svc_udp_get_dest_address(struct svc_rqst *rqstp, |
489 | struct cmsghdr *cmh) | |
95756482 | 490 | { |
7702ce40 CL |
491 | switch (cmh->cmsg_level) { |
492 | case SOL_IP: | |
493 | return svc_udp_get_dest_address4(rqstp, cmh); | |
494 | case SOL_IPV6: | |
495 | return svc_udp_get_dest_address6(rqstp, cmh); | |
95756482 | 496 | } |
7702ce40 CL |
497 | |
498 | return 0; | |
95756482 CL |
499 | } |
500 | ||
1da177e4 LT |
501 | /* |
502 | * Receive a datagram from a UDP socket. | |
503 | */ | |
0f0257ea | 504 | static int svc_udp_recvfrom(struct svc_rqst *rqstp) |
1da177e4 | 505 | { |
57b1d3ba TT |
506 | struct svc_sock *svsk = |
507 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
bb5cf160 | 508 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
1da177e4 | 509 | struct sk_buff *skb; |
bc375ea7 DM |
510 | union { |
511 | struct cmsghdr hdr; | |
512 | long all[SVC_PKTINFO_SPACE / sizeof(long)]; | |
513 | } buffer; | |
514 | struct cmsghdr *cmh = &buffer.hdr; | |
7a37f578 N |
515 | struct msghdr msg = { |
516 | .msg_name = svc_addr(rqstp), | |
517 | .msg_control = cmh, | |
518 | .msg_controllen = sizeof(buffer), | |
519 | .msg_flags = MSG_DONTWAIT, | |
520 | }; | |
abc5c44d CL |
521 | size_t len; |
522 | int err; | |
1da177e4 | 523 | |
02fc6c36 | 524 | if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags)) |
1da177e4 LT |
525 | /* udp sockets need large rcvbuf as all pending |
526 | * requests are still in that buffer. sndbuf must | |
527 | * also be large enough that there is enough space | |
3262c816 GB |
528 | * for one reply per thread. We count all threads |
529 | * rather than threads in a particular pool, which | |
530 | * provides an upper bound on the number of threads | |
531 | * which will access the socket. | |
1da177e4 LT |
532 | */ |
533 | svc_sock_setbufsize(svsk->sk_sock, | |
c6b0a9f8 N |
534 | (serv->sv_nrthreads+3) * serv->sv_max_mesg, |
535 | (serv->sv_nrthreads+3) * serv->sv_max_mesg); | |
1da177e4 | 536 | |
02fc6c36 | 537 | clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
05ed690e N |
538 | skb = NULL; |
539 | err = kernel_recvmsg(svsk->sk_sock, &msg, NULL, | |
540 | 0, 0, MSG_PEEK | MSG_DONTWAIT); | |
541 | if (err >= 0) | |
542 | skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err); | |
543 | ||
544 | if (skb == NULL) { | |
545 | if (err != -EAGAIN) { | |
546 | /* possibly an icmp error */ | |
547 | dprintk("svc: recvfrom returned error %d\n", -err); | |
02fc6c36 | 548 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 549 | } |
a6046f71 | 550 | svc_xprt_received(&svsk->sk_xprt); |
05ed690e | 551 | return -EAGAIN; |
1da177e4 | 552 | } |
9dbc240f | 553 | len = svc_addr_len(svc_addr(rqstp)); |
abc5c44d CL |
554 | if (len == 0) |
555 | return -EAFNOSUPPORT; | |
9dbc240f | 556 | rqstp->rq_addrlen = len; |
b7aa0bf7 ED |
557 | if (skb->tstamp.tv64 == 0) { |
558 | skb->tstamp = ktime_get_real(); | |
cca5172a | 559 | /* Don't enable netstamp, sunrpc doesn't |
1da177e4 LT |
560 | need that much accuracy */ |
561 | } | |
b7aa0bf7 | 562 | svsk->sk_sk->sk_stamp = skb->tstamp; |
02fc6c36 | 563 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */ |
1da177e4 LT |
564 | |
565 | /* | |
566 | * Maybe more packets - kick another thread ASAP. | |
567 | */ | |
a6046f71 | 568 | svc_xprt_received(&svsk->sk_xprt); |
1da177e4 LT |
569 | |
570 | len = skb->len - sizeof(struct udphdr); | |
571 | rqstp->rq_arg.len = len; | |
572 | ||
95756482 | 573 | rqstp->rq_prot = IPPROTO_UDP; |
27459f09 | 574 | |
7702ce40 | 575 | if (!svc_udp_get_dest_address(rqstp, cmh)) { |
7a37f578 | 576 | if (net_ratelimit()) |
7702ce40 CL |
577 | printk(KERN_WARNING |
578 | "svc: received unknown control message %d/%d; " | |
579 | "dropping RPC reply datagram\n", | |
580 | cmh->cmsg_level, cmh->cmsg_type); | |
9d410c79 | 581 | skb_free_datagram_locked(svsk->sk_sk, skb); |
7a37f578 N |
582 | return 0; |
583 | } | |
1da177e4 LT |
584 | |
585 | if (skb_is_nonlinear(skb)) { | |
586 | /* we have to copy */ | |
587 | local_bh_disable(); | |
588 | if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) { | |
589 | local_bh_enable(); | |
590 | /* checksum error */ | |
9d410c79 | 591 | skb_free_datagram_locked(svsk->sk_sk, skb); |
1da177e4 LT |
592 | return 0; |
593 | } | |
594 | local_bh_enable(); | |
9d410c79 | 595 | skb_free_datagram_locked(svsk->sk_sk, skb); |
1da177e4 LT |
596 | } else { |
597 | /* we can use it in-place */ | |
0f0257ea TT |
598 | rqstp->rq_arg.head[0].iov_base = skb->data + |
599 | sizeof(struct udphdr); | |
1da177e4 | 600 | rqstp->rq_arg.head[0].iov_len = len; |
fb286bb2 | 601 | if (skb_checksum_complete(skb)) { |
9d410c79 | 602 | skb_free_datagram_locked(svsk->sk_sk, skb); |
fb286bb2 | 603 | return 0; |
1da177e4 | 604 | } |
5148bf4e | 605 | rqstp->rq_xprt_ctxt = skb; |
1da177e4 LT |
606 | } |
607 | ||
608 | rqstp->rq_arg.page_base = 0; | |
609 | if (len <= rqstp->rq_arg.head[0].iov_len) { | |
610 | rqstp->rq_arg.head[0].iov_len = len; | |
611 | rqstp->rq_arg.page_len = 0; | |
44524359 | 612 | rqstp->rq_respages = rqstp->rq_pages+1; |
1da177e4 LT |
613 | } else { |
614 | rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len; | |
44524359 | 615 | rqstp->rq_respages = rqstp->rq_pages + 1 + |
172589cc | 616 | DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE); |
1da177e4 LT |
617 | } |
618 | ||
619 | if (serv->sv_stats) | |
620 | serv->sv_stats->netudpcnt++; | |
621 | ||
622 | return len; | |
623 | } | |
624 | ||
625 | static int | |
626 | svc_udp_sendto(struct svc_rqst *rqstp) | |
627 | { | |
628 | int error; | |
629 | ||
630 | error = svc_sendto(rqstp, &rqstp->rq_res); | |
631 | if (error == -ECONNREFUSED) | |
632 | /* ICMP error on earlier request. */ | |
633 | error = svc_sendto(rqstp, &rqstp->rq_res); | |
634 | ||
635 | return error; | |
636 | } | |
637 | ||
e831fe65 TT |
638 | static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp) |
639 | { | |
640 | } | |
641 | ||
323bee32 TT |
642 | static int svc_udp_has_wspace(struct svc_xprt *xprt) |
643 | { | |
644 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
bb5cf160 | 645 | struct svc_serv *serv = xprt->xpt_server; |
323bee32 TT |
646 | unsigned long required; |
647 | ||
648 | /* | |
649 | * Set the SOCK_NOSPACE flag before checking the available | |
650 | * sock space. | |
651 | */ | |
652 | set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
7a90e8cc | 653 | required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg; |
323bee32 TT |
654 | if (required*2 > sock_wspace(svsk->sk_sk)) |
655 | return 0; | |
656 | clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
657 | return 1; | |
658 | } | |
659 | ||
38a417cc TT |
660 | static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt) |
661 | { | |
662 | BUG(); | |
663 | return NULL; | |
664 | } | |
665 | ||
b700cbb1 TT |
666 | static struct svc_xprt *svc_udp_create(struct svc_serv *serv, |
667 | struct sockaddr *sa, int salen, | |
668 | int flags) | |
669 | { | |
670 | return svc_create_socket(serv, IPPROTO_UDP, sa, salen, flags); | |
671 | } | |
672 | ||
360d8738 | 673 | static struct svc_xprt_ops svc_udp_ops = { |
b700cbb1 | 674 | .xpo_create = svc_udp_create, |
5d137990 TT |
675 | .xpo_recvfrom = svc_udp_recvfrom, |
676 | .xpo_sendto = svc_udp_sendto, | |
5148bf4e | 677 | .xpo_release_rqst = svc_release_skb, |
755cceab TT |
678 | .xpo_detach = svc_sock_detach, |
679 | .xpo_free = svc_sock_free, | |
e831fe65 | 680 | .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr, |
323bee32 | 681 | .xpo_has_wspace = svc_udp_has_wspace, |
38a417cc | 682 | .xpo_accept = svc_udp_accept, |
360d8738 TT |
683 | }; |
684 | ||
685 | static struct svc_xprt_class svc_udp_class = { | |
686 | .xcl_name = "udp", | |
b700cbb1 | 687 | .xcl_owner = THIS_MODULE, |
360d8738 | 688 | .xcl_ops = &svc_udp_ops, |
49023155 | 689 | .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP, |
360d8738 TT |
690 | }; |
691 | ||
bb5cf160 | 692 | static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv) |
1da177e4 | 693 | { |
7702ce40 | 694 | int err, level, optname, one = 1; |
7a37f578 | 695 | |
bb5cf160 | 696 | svc_xprt_init(&svc_udp_class, &svsk->sk_xprt, serv); |
def13d74 | 697 | clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
698 | svsk->sk_sk->sk_data_ready = svc_udp_data_ready; |
699 | svsk->sk_sk->sk_write_space = svc_write_space; | |
1da177e4 LT |
700 | |
701 | /* initialise setting must have enough space to | |
cca5172a | 702 | * receive and respond to one request. |
1da177e4 LT |
703 | * svc_udp_recvfrom will re-adjust if necessary |
704 | */ | |
705 | svc_sock_setbufsize(svsk->sk_sock, | |
bb5cf160 TT |
706 | 3 * svsk->sk_xprt.xpt_server->sv_max_mesg, |
707 | 3 * svsk->sk_xprt.xpt_server->sv_max_mesg); | |
1da177e4 | 708 | |
0f0257ea TT |
709 | /* data might have come in before data_ready set up */ |
710 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); | |
02fc6c36 | 711 | set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); |
7a37f578 | 712 | |
7a37f578 | 713 | /* make sure we get destination address info */ |
7702ce40 CL |
714 | switch (svsk->sk_sk->sk_family) { |
715 | case AF_INET: | |
716 | level = SOL_IP; | |
717 | optname = IP_PKTINFO; | |
718 | break; | |
719 | case AF_INET6: | |
720 | level = SOL_IPV6; | |
721 | optname = IPV6_RECVPKTINFO; | |
722 | break; | |
723 | default: | |
724 | BUG(); | |
725 | } | |
726 | err = kernel_setsockopt(svsk->sk_sock, level, optname, | |
727 | (char *)&one, sizeof(one)); | |
728 | dprintk("svc: kernel_setsockopt returned %d\n", err); | |
1da177e4 LT |
729 | } |
730 | ||
731 | /* | |
732 | * A data_ready event on a listening socket means there's a connection | |
733 | * pending. Do not use state_change as a substitute for it. | |
734 | */ | |
0f0257ea | 735 | static void svc_tcp_listen_data_ready(struct sock *sk, int count_unused) |
1da177e4 | 736 | { |
939bb7ef | 737 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 LT |
738 | |
739 | dprintk("svc: socket %p TCP (listen) state change %d\n", | |
939bb7ef | 740 | sk, sk->sk_state); |
1da177e4 | 741 | |
939bb7ef NB |
742 | /* |
743 | * This callback may called twice when a new connection | |
744 | * is established as a child socket inherits everything | |
745 | * from a parent LISTEN socket. | |
746 | * 1) data_ready method of the parent socket will be called | |
747 | * when one of child sockets become ESTABLISHED. | |
748 | * 2) data_ready method of the child socket may be called | |
749 | * when it receives data before the socket is accepted. | |
750 | * In case of 2, we should ignore it silently. | |
751 | */ | |
752 | if (sk->sk_state == TCP_LISTEN) { | |
753 | if (svsk) { | |
02fc6c36 | 754 | set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
f6150c3c | 755 | svc_xprt_enqueue(&svsk->sk_xprt); |
939bb7ef NB |
756 | } else |
757 | printk("svc: socket %p: no user data\n", sk); | |
1da177e4 | 758 | } |
939bb7ef | 759 | |
1da177e4 LT |
760 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
761 | wake_up_interruptible_all(sk->sk_sleep); | |
762 | } | |
763 | ||
764 | /* | |
765 | * A state change on a connected socket means it's dying or dead. | |
766 | */ | |
0f0257ea | 767 | static void svc_tcp_state_change(struct sock *sk) |
1da177e4 | 768 | { |
939bb7ef | 769 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 LT |
770 | |
771 | dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n", | |
939bb7ef | 772 | sk, sk->sk_state, sk->sk_user_data); |
1da177e4 | 773 | |
939bb7ef | 774 | if (!svsk) |
1da177e4 | 775 | printk("svc: socket %p: no user data\n", sk); |
939bb7ef | 776 | else { |
02fc6c36 | 777 | set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); |
f6150c3c | 778 | svc_xprt_enqueue(&svsk->sk_xprt); |
1da177e4 | 779 | } |
1da177e4 LT |
780 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
781 | wake_up_interruptible_all(sk->sk_sleep); | |
782 | } | |
783 | ||
0f0257ea | 784 | static void svc_tcp_data_ready(struct sock *sk, int count) |
1da177e4 | 785 | { |
939bb7ef | 786 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 LT |
787 | |
788 | dprintk("svc: socket %p TCP data ready (svsk %p)\n", | |
939bb7ef NB |
789 | sk, sk->sk_user_data); |
790 | if (svsk) { | |
02fc6c36 | 791 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
f6150c3c | 792 | svc_xprt_enqueue(&svsk->sk_xprt); |
939bb7ef | 793 | } |
1da177e4 LT |
794 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
795 | wake_up_interruptible(sk->sk_sleep); | |
796 | } | |
797 | ||
798 | /* | |
799 | * Accept a TCP connection | |
800 | */ | |
38a417cc | 801 | static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt) |
1da177e4 | 802 | { |
38a417cc | 803 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); |
cdd88b9f | 804 | struct sockaddr_storage addr; |
805 | struct sockaddr *sin = (struct sockaddr *) &addr; | |
bb5cf160 | 806 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
1da177e4 LT |
807 | struct socket *sock = svsk->sk_sock; |
808 | struct socket *newsock; | |
1da177e4 LT |
809 | struct svc_sock *newsvsk; |
810 | int err, slen; | |
5216a8e7 | 811 | RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); |
1da177e4 LT |
812 | |
813 | dprintk("svc: tcp_accept %p sock %p\n", svsk, sock); | |
814 | if (!sock) | |
38a417cc | 815 | return NULL; |
1da177e4 | 816 | |
02fc6c36 | 817 | clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
e6242e92 SS |
818 | err = kernel_accept(sock, &newsock, O_NONBLOCK); |
819 | if (err < 0) { | |
1da177e4 LT |
820 | if (err == -ENOMEM) |
821 | printk(KERN_WARNING "%s: no more sockets!\n", | |
822 | serv->sv_name); | |
e6242e92 | 823 | else if (err != -EAGAIN && net_ratelimit()) |
1da177e4 LT |
824 | printk(KERN_WARNING "%s: accept failed (err %d)!\n", |
825 | serv->sv_name, -err); | |
38a417cc | 826 | return NULL; |
1da177e4 | 827 | } |
02fc6c36 | 828 | set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 829 | |
cdd88b9f | 830 | err = kernel_getpeername(newsock, sin, &slen); |
1da177e4 LT |
831 | if (err < 0) { |
832 | if (net_ratelimit()) | |
833 | printk(KERN_WARNING "%s: peername failed (err %d)!\n", | |
834 | serv->sv_name, -err); | |
835 | goto failed; /* aborted connection or whatever */ | |
836 | } | |
837 | ||
838 | /* Ideally, we would want to reject connections from unauthorized | |
ad06e4bd CL |
839 | * hosts here, but when we get encryption, the IP of the host won't |
840 | * tell us anything. For now just warn about unpriv connections. | |
1da177e4 | 841 | */ |
cdd88b9f | 842 | if (!svc_port_is_privileged(sin)) { |
1da177e4 | 843 | dprintk(KERN_WARNING |
ad06e4bd | 844 | "%s: connect from unprivileged port: %s\n", |
cca5172a | 845 | serv->sv_name, |
cdd88b9f | 846 | __svc_print_addr(sin, buf, sizeof(buf))); |
1da177e4 | 847 | } |
ad06e4bd | 848 | dprintk("%s: connect from %s\n", serv->sv_name, |
cdd88b9f | 849 | __svc_print_addr(sin, buf, sizeof(buf))); |
1da177e4 LT |
850 | |
851 | /* make sure that a write doesn't block forever when | |
852 | * low on memory | |
853 | */ | |
854 | newsock->sk->sk_sndtimeo = HZ*30; | |
855 | ||
6b174337 CL |
856 | if (!(newsvsk = svc_setup_socket(serv, newsock, &err, |
857 | (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY)))) | |
1da177e4 | 858 | goto failed; |
9dbc240f | 859 | svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen); |
a9747692 FM |
860 | err = kernel_getsockname(newsock, sin, &slen); |
861 | if (unlikely(err < 0)) { | |
862 | dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err); | |
863 | slen = offsetof(struct sockaddr, sa_data); | |
864 | } | |
9dbc240f | 865 | svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen); |
067d7817 | 866 | |
f9f3cc4f TT |
867 | if (serv->sv_stats) |
868 | serv->sv_stats->nettcpconn++; | |
869 | ||
870 | return &newsvsk->sk_xprt; | |
871 | ||
872 | failed: | |
873 | sock_release(newsock); | |
874 | return NULL; | |
875 | } | |
876 | ||
1da177e4 | 877 | /* |
8f55f3c0 AB |
878 | * Receive data. |
879 | * If we haven't gotten the record length yet, get the next four bytes. | |
880 | * Otherwise try to gobble up as much as possible up to the complete | |
881 | * record length. | |
1da177e4 | 882 | */ |
8f55f3c0 | 883 | static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp) |
1da177e4 | 884 | { |
bb5cf160 | 885 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
8f55f3c0 | 886 | int len; |
1da177e4 | 887 | |
7f421835 BF |
888 | if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags)) |
889 | /* sndbuf needs to have room for one request | |
890 | * per thread, otherwise we can stall even when the | |
891 | * network isn't a bottleneck. | |
892 | * | |
893 | * We count all threads rather than threads in a | |
894 | * particular pool, which provides an upper bound | |
895 | * on the number of threads which will access the socket. | |
896 | * | |
897 | * rcvbuf just needs to be able to hold a few requests. | |
898 | * Normally they will be removed from the queue | |
899 | * as soon a a complete request arrives. | |
900 | */ | |
901 | svc_sock_setbufsize(svsk->sk_sock, | |
902 | (serv->sv_nrthreads+3) * serv->sv_max_mesg, | |
903 | 3 * serv->sv_max_mesg); | |
904 | ||
02fc6c36 | 905 | clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 906 | |
c0401ea0 CL |
907 | if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) { |
908 | int want = sizeof(rpc_fraghdr) - svsk->sk_tcplen; | |
1da177e4 LT |
909 | struct kvec iov; |
910 | ||
911 | iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen; | |
912 | iov.iov_len = want; | |
913 | if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0) | |
914 | goto error; | |
915 | svsk->sk_tcplen += len; | |
916 | ||
917 | if (len < want) { | |
c0401ea0 CL |
918 | dprintk("svc: short recvfrom while reading record " |
919 | "length (%d of %d)\n", len, want); | |
a6046f71 | 920 | svc_xprt_received(&svsk->sk_xprt); |
8f55f3c0 | 921 | goto err_again; /* record header not complete */ |
1da177e4 LT |
922 | } |
923 | ||
924 | svsk->sk_reclen = ntohl(svsk->sk_reclen); | |
c0401ea0 | 925 | if (!(svsk->sk_reclen & RPC_LAST_STREAM_FRAGMENT)) { |
1da177e4 LT |
926 | /* FIXME: technically, a record can be fragmented, |
927 | * and non-terminal fragments will not have the top | |
928 | * bit set in the fragment length header. | |
929 | * But apparently no known nfs clients send fragmented | |
930 | * records. */ | |
34e9a63b | 931 | if (net_ratelimit()) |
c0401ea0 CL |
932 | printk(KERN_NOTICE "RPC: multiple fragments " |
933 | "per record not supported\n"); | |
1da177e4 LT |
934 | goto err_delete; |
935 | } | |
8f55f3c0 | 936 | |
c0401ea0 | 937 | svsk->sk_reclen &= RPC_FRAGMENT_SIZE_MASK; |
1da177e4 | 938 | dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen); |
c6b0a9f8 | 939 | if (svsk->sk_reclen > serv->sv_max_mesg) { |
34e9a63b | 940 | if (net_ratelimit()) |
c0401ea0 CL |
941 | printk(KERN_NOTICE "RPC: " |
942 | "fragment too large: 0x%08lx\n", | |
943 | (unsigned long)svsk->sk_reclen); | |
1da177e4 LT |
944 | goto err_delete; |
945 | } | |
946 | } | |
947 | ||
948 | /* Check whether enough data is available */ | |
949 | len = svc_recv_available(svsk); | |
950 | if (len < 0) | |
951 | goto error; | |
952 | ||
953 | if (len < svsk->sk_reclen) { | |
954 | dprintk("svc: incomplete TCP record (%d of %d)\n", | |
955 | len, svsk->sk_reclen); | |
a6046f71 | 956 | svc_xprt_received(&svsk->sk_xprt); |
8f55f3c0 | 957 | goto err_again; /* record not complete */ |
1da177e4 LT |
958 | } |
959 | len = svsk->sk_reclen; | |
02fc6c36 | 960 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 961 | |
8f55f3c0 AB |
962 | return len; |
963 | error: | |
964 | if (len == -EAGAIN) { | |
965 | dprintk("RPC: TCP recv_record got EAGAIN\n"); | |
966 | svc_xprt_received(&svsk->sk_xprt); | |
967 | } | |
968 | return len; | |
969 | err_delete: | |
970 | set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); | |
301e99ce | 971 | svc_xprt_received(&svsk->sk_xprt); |
8f55f3c0 AB |
972 | err_again: |
973 | return -EAGAIN; | |
974 | } | |
975 | ||
4cfc7e60 RI |
976 | static int svc_process_calldir(struct svc_sock *svsk, struct svc_rqst *rqstp, |
977 | struct rpc_rqst **reqpp, struct kvec *vec) | |
978 | { | |
979 | struct rpc_rqst *req = NULL; | |
980 | u32 *p; | |
981 | u32 xid; | |
982 | u32 calldir; | |
983 | int len; | |
984 | ||
985 | len = svc_recvfrom(rqstp, vec, 1, 8); | |
986 | if (len < 0) | |
987 | goto error; | |
988 | ||
989 | p = (u32 *)rqstp->rq_arg.head[0].iov_base; | |
990 | xid = *p++; | |
991 | calldir = *p; | |
992 | ||
993 | if (calldir == 0) { | |
994 | /* REQUEST is the most common case */ | |
995 | vec[0] = rqstp->rq_arg.head[0]; | |
996 | } else { | |
997 | /* REPLY */ | |
998 | if (svsk->sk_bc_xprt) | |
999 | req = xprt_lookup_rqst(svsk->sk_bc_xprt, xid); | |
1000 | ||
1001 | if (!req) { | |
1002 | printk(KERN_NOTICE | |
1003 | "%s: Got unrecognized reply: " | |
1004 | "calldir 0x%x sk_bc_xprt %p xid %08x\n", | |
1005 | __func__, ntohl(calldir), | |
1006 | svsk->sk_bc_xprt, xid); | |
1007 | vec[0] = rqstp->rq_arg.head[0]; | |
1008 | goto out; | |
1009 | } | |
1010 | ||
1011 | memcpy(&req->rq_private_buf, &req->rq_rcv_buf, | |
1012 | sizeof(struct xdr_buf)); | |
1013 | /* copy the xid and call direction */ | |
1014 | memcpy(req->rq_private_buf.head[0].iov_base, | |
1015 | rqstp->rq_arg.head[0].iov_base, 8); | |
1016 | vec[0] = req->rq_private_buf.head[0]; | |
1017 | } | |
1018 | out: | |
1019 | vec[0].iov_base += 8; | |
1020 | vec[0].iov_len -= 8; | |
1021 | len = svsk->sk_reclen - 8; | |
1022 | error: | |
1023 | *reqpp = req; | |
1024 | return len; | |
1025 | } | |
1026 | ||
8f55f3c0 AB |
1027 | /* |
1028 | * Receive data from a TCP socket. | |
1029 | */ | |
1030 | static int svc_tcp_recvfrom(struct svc_rqst *rqstp) | |
1031 | { | |
1032 | struct svc_sock *svsk = | |
1033 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
1034 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; | |
1035 | int len; | |
1036 | struct kvec *vec; | |
1037 | int pnum, vlen; | |
4cfc7e60 | 1038 | struct rpc_rqst *req = NULL; |
8f55f3c0 AB |
1039 | |
1040 | dprintk("svc: tcp_recv %p data %d conn %d close %d\n", | |
1041 | svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags), | |
1042 | test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags), | |
1043 | test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags)); | |
1044 | ||
1045 | len = svc_tcp_recv_record(svsk, rqstp); | |
1046 | if (len < 0) | |
1047 | goto error; | |
1048 | ||
3cc03b16 | 1049 | vec = rqstp->rq_vec; |
1da177e4 LT |
1050 | vec[0] = rqstp->rq_arg.head[0]; |
1051 | vlen = PAGE_SIZE; | |
4cfc7e60 RI |
1052 | |
1053 | /* | |
1054 | * We have enough data for the whole tcp record. Let's try and read the | |
1055 | * first 8 bytes to get the xid and the call direction. We can use this | |
1056 | * to figure out if this is a call or a reply to a callback. If | |
1057 | * sk_reclen is < 8 (xid and calldir), then this is a malformed packet. | |
1058 | * In that case, don't bother with the calldir and just read the data. | |
1059 | * It will be rejected in svc_process. | |
1060 | */ | |
1061 | if (len >= 8) { | |
1062 | len = svc_process_calldir(svsk, rqstp, &req, vec); | |
1063 | if (len < 0) | |
1064 | goto err_again; | |
1065 | vlen -= 8; | |
1066 | } | |
1067 | ||
1da177e4 LT |
1068 | pnum = 1; |
1069 | while (vlen < len) { | |
4cfc7e60 RI |
1070 | vec[pnum].iov_base = (req) ? |
1071 | page_address(req->rq_private_buf.pages[pnum - 1]) : | |
1072 | page_address(rqstp->rq_pages[pnum]); | |
1da177e4 LT |
1073 | vec[pnum].iov_len = PAGE_SIZE; |
1074 | pnum++; | |
1075 | vlen += PAGE_SIZE; | |
1076 | } | |
44524359 | 1077 | rqstp->rq_respages = &rqstp->rq_pages[pnum]; |
1da177e4 LT |
1078 | |
1079 | /* Now receive data */ | |
1080 | len = svc_recvfrom(rqstp, vec, pnum, len); | |
1081 | if (len < 0) | |
8f55f3c0 | 1082 | goto err_again; |
1da177e4 | 1083 | |
4cfc7e60 RI |
1084 | /* |
1085 | * Account for the 8 bytes we read earlier | |
1086 | */ | |
1087 | len += 8; | |
1088 | ||
1089 | if (req) { | |
1090 | xprt_complete_rqst(req->rq_task, len); | |
1091 | len = 0; | |
1092 | goto out; | |
1093 | } | |
1da177e4 LT |
1094 | dprintk("svc: TCP complete record (%d bytes)\n", len); |
1095 | rqstp->rq_arg.len = len; | |
1096 | rqstp->rq_arg.page_base = 0; | |
1097 | if (len <= rqstp->rq_arg.head[0].iov_len) { | |
1098 | rqstp->rq_arg.head[0].iov_len = len; | |
1099 | rqstp->rq_arg.page_len = 0; | |
1100 | } else { | |
1101 | rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len; | |
1102 | } | |
1103 | ||
5148bf4e | 1104 | rqstp->rq_xprt_ctxt = NULL; |
1da177e4 LT |
1105 | rqstp->rq_prot = IPPROTO_TCP; |
1106 | ||
4cfc7e60 | 1107 | out: |
1da177e4 LT |
1108 | /* Reset TCP read info */ |
1109 | svsk->sk_reclen = 0; | |
1110 | svsk->sk_tcplen = 0; | |
1111 | ||
9dbc240f | 1112 | svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt); |
a6046f71 | 1113 | svc_xprt_received(&svsk->sk_xprt); |
1da177e4 LT |
1114 | if (serv->sv_stats) |
1115 | serv->sv_stats->nettcpcnt++; | |
1116 | ||
1117 | return len; | |
1118 | ||
8f55f3c0 | 1119 | err_again: |
1da177e4 LT |
1120 | if (len == -EAGAIN) { |
1121 | dprintk("RPC: TCP recvfrom got EAGAIN\n"); | |
a6046f71 | 1122 | svc_xprt_received(&svsk->sk_xprt); |
8f55f3c0 AB |
1123 | return len; |
1124 | } | |
1125 | error: | |
1126 | if (len != -EAGAIN) { | |
1da177e4 | 1127 | printk(KERN_NOTICE "%s: recvfrom returned errno %d\n", |
bb5cf160 | 1128 | svsk->sk_xprt.xpt_server->sv_name, -len); |
8f55f3c0 | 1129 | set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 1130 | } |
8f55f3c0 | 1131 | return -EAGAIN; |
1da177e4 LT |
1132 | } |
1133 | ||
1134 | /* | |
1135 | * Send out data on TCP socket. | |
1136 | */ | |
0f0257ea | 1137 | static int svc_tcp_sendto(struct svc_rqst *rqstp) |
1da177e4 LT |
1138 | { |
1139 | struct xdr_buf *xbufp = &rqstp->rq_res; | |
1140 | int sent; | |
d8ed029d | 1141 | __be32 reclen; |
1da177e4 LT |
1142 | |
1143 | /* Set up the first element of the reply kvec. | |
1144 | * Any other kvecs that may be in use have been taken | |
1145 | * care of by the server implementation itself. | |
1146 | */ | |
1147 | reclen = htonl(0x80000000|((xbufp->len ) - 4)); | |
1148 | memcpy(xbufp->head[0].iov_base, &reclen, 4); | |
1149 | ||
57b1d3ba | 1150 | if (test_bit(XPT_DEAD, &rqstp->rq_xprt->xpt_flags)) |
1da177e4 LT |
1151 | return -ENOTCONN; |
1152 | ||
1153 | sent = svc_sendto(rqstp, &rqstp->rq_res); | |
1154 | if (sent != xbufp->len) { | |
0f0257ea TT |
1155 | printk(KERN_NOTICE |
1156 | "rpc-srv/tcp: %s: %s %d when sending %d bytes " | |
1157 | "- shutting down socket\n", | |
57b1d3ba | 1158 | rqstp->rq_xprt->xpt_server->sv_name, |
1da177e4 LT |
1159 | (sent<0)?"got error":"sent only", |
1160 | sent, xbufp->len); | |
57b1d3ba | 1161 | set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags); |
f6150c3c | 1162 | svc_xprt_enqueue(rqstp->rq_xprt); |
1da177e4 LT |
1163 | sent = -EAGAIN; |
1164 | } | |
1165 | return sent; | |
1166 | } | |
1167 | ||
e831fe65 TT |
1168 | /* |
1169 | * Setup response header. TCP has a 4B record length field. | |
1170 | */ | |
1171 | static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp) | |
1172 | { | |
1173 | struct kvec *resv = &rqstp->rq_res.head[0]; | |
1174 | ||
1175 | /* tcp needs a space for the record length... */ | |
1176 | svc_putnl(resv, 0); | |
1177 | } | |
1178 | ||
323bee32 TT |
1179 | static int svc_tcp_has_wspace(struct svc_xprt *xprt) |
1180 | { | |
57b1d3ba | 1181 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); |
47fcb03f | 1182 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
323bee32 | 1183 | int required; |
323bee32 | 1184 | |
47fcb03f TM |
1185 | if (test_bit(XPT_LISTENER, &xprt->xpt_flags)) |
1186 | return 1; | |
1187 | required = atomic_read(&xprt->xpt_reserved) + serv->sv_max_mesg; | |
1188 | if (sk_stream_wspace(svsk->sk_sk) >= required) | |
1189 | return 1; | |
323bee32 | 1190 | set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); |
47fcb03f | 1191 | return 0; |
323bee32 TT |
1192 | } |
1193 | ||
b700cbb1 TT |
1194 | static struct svc_xprt *svc_tcp_create(struct svc_serv *serv, |
1195 | struct sockaddr *sa, int salen, | |
1196 | int flags) | |
1197 | { | |
1198 | return svc_create_socket(serv, IPPROTO_TCP, sa, salen, flags); | |
1199 | } | |
1200 | ||
360d8738 | 1201 | static struct svc_xprt_ops svc_tcp_ops = { |
b700cbb1 | 1202 | .xpo_create = svc_tcp_create, |
5d137990 TT |
1203 | .xpo_recvfrom = svc_tcp_recvfrom, |
1204 | .xpo_sendto = svc_tcp_sendto, | |
5148bf4e | 1205 | .xpo_release_rqst = svc_release_skb, |
69b6ba37 | 1206 | .xpo_detach = svc_tcp_sock_detach, |
755cceab | 1207 | .xpo_free = svc_sock_free, |
e831fe65 | 1208 | .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr, |
323bee32 | 1209 | .xpo_has_wspace = svc_tcp_has_wspace, |
38a417cc | 1210 | .xpo_accept = svc_tcp_accept, |
360d8738 TT |
1211 | }; |
1212 | ||
1213 | static struct svc_xprt_class svc_tcp_class = { | |
1214 | .xcl_name = "tcp", | |
b700cbb1 | 1215 | .xcl_owner = THIS_MODULE, |
360d8738 | 1216 | .xcl_ops = &svc_tcp_ops, |
49023155 | 1217 | .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP, |
360d8738 TT |
1218 | }; |
1219 | ||
1220 | void svc_init_xprt_sock(void) | |
1221 | { | |
1222 | svc_reg_xprt_class(&svc_tcp_class); | |
1223 | svc_reg_xprt_class(&svc_udp_class); | |
1224 | } | |
1225 | ||
1226 | void svc_cleanup_xprt_sock(void) | |
1227 | { | |
1228 | svc_unreg_xprt_class(&svc_tcp_class); | |
1229 | svc_unreg_xprt_class(&svc_udp_class); | |
1230 | } | |
1231 | ||
bb5cf160 | 1232 | static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv) |
1da177e4 LT |
1233 | { |
1234 | struct sock *sk = svsk->sk_sk; | |
1da177e4 | 1235 | |
bb5cf160 | 1236 | svc_xprt_init(&svc_tcp_class, &svsk->sk_xprt, serv); |
def13d74 | 1237 | set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
1238 | if (sk->sk_state == TCP_LISTEN) { |
1239 | dprintk("setting up TCP socket for listening\n"); | |
02fc6c36 | 1240 | set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 1241 | sk->sk_data_ready = svc_tcp_listen_data_ready; |
02fc6c36 | 1242 | set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
1243 | } else { |
1244 | dprintk("setting up TCP socket for reading\n"); | |
1245 | sk->sk_state_change = svc_tcp_state_change; | |
1246 | sk->sk_data_ready = svc_tcp_data_ready; | |
47fcb03f | 1247 | sk->sk_write_space = svc_tcp_write_space; |
1da177e4 LT |
1248 | |
1249 | svsk->sk_reclen = 0; | |
1250 | svsk->sk_tcplen = 0; | |
1251 | ||
b7872fe8 | 1252 | tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF; |
1da177e4 | 1253 | |
7f421835 BF |
1254 | /* initialise setting must have enough space to |
1255 | * receive and respond to one request. | |
1256 | * svc_tcp_recvfrom will re-adjust if necessary | |
1257 | */ | |
1258 | svc_sock_setbufsize(svsk->sk_sock, | |
1259 | 3 * svsk->sk_xprt.xpt_server->sv_max_mesg, | |
1260 | 3 * svsk->sk_xprt.xpt_server->sv_max_mesg); | |
1261 | ||
1262 | set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); | |
02fc6c36 | 1263 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
cca5172a | 1264 | if (sk->sk_state != TCP_ESTABLISHED) |
02fc6c36 | 1265 | set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
1266 | } |
1267 | } | |
1268 | ||
0f0257ea | 1269 | void svc_sock_update_bufs(struct svc_serv *serv) |
1da177e4 LT |
1270 | { |
1271 | /* | |
1272 | * The number of server threads has changed. Update | |
1273 | * rcvbuf and sndbuf accordingly on all sockets | |
1274 | */ | |
1275 | struct list_head *le; | |
1276 | ||
1277 | spin_lock_bh(&serv->sv_lock); | |
1278 | list_for_each(le, &serv->sv_permsocks) { | |
cca5172a | 1279 | struct svc_sock *svsk = |
7a182083 | 1280 | list_entry(le, struct svc_sock, sk_xprt.xpt_list); |
02fc6c36 | 1281 | set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
1282 | } |
1283 | list_for_each(le, &serv->sv_tempsocks) { | |
1284 | struct svc_sock *svsk = | |
7a182083 | 1285 | list_entry(le, struct svc_sock, sk_xprt.xpt_list); |
02fc6c36 | 1286 | set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
1287 | } |
1288 | spin_unlock_bh(&serv->sv_lock); | |
1289 | } | |
24c3767e | 1290 | EXPORT_SYMBOL_GPL(svc_sock_update_bufs); |
1da177e4 | 1291 | |
1da177e4 LT |
1292 | /* |
1293 | * Initialize socket for RPC use and create svc_sock struct | |
1294 | * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF. | |
1295 | */ | |
6b174337 CL |
1296 | static struct svc_sock *svc_setup_socket(struct svc_serv *serv, |
1297 | struct socket *sock, | |
1298 | int *errp, int flags) | |
1da177e4 LT |
1299 | { |
1300 | struct svc_sock *svsk; | |
1301 | struct sock *inet; | |
6b174337 | 1302 | int pmap_register = !(flags & SVC_SOCK_ANONYMOUS); |
1da177e4 LT |
1303 | |
1304 | dprintk("svc: svc_setup_socket %p\n", sock); | |
0da974f4 | 1305 | if (!(svsk = kzalloc(sizeof(*svsk), GFP_KERNEL))) { |
1da177e4 LT |
1306 | *errp = -ENOMEM; |
1307 | return NULL; | |
1308 | } | |
1da177e4 LT |
1309 | |
1310 | inet = sock->sk; | |
1311 | ||
1312 | /* Register socket with portmapper */ | |
1313 | if (*errp >= 0 && pmap_register) | |
baf01caf | 1314 | *errp = svc_register(serv, inet->sk_family, inet->sk_protocol, |
c720c7e8 | 1315 | ntohs(inet_sk(inet)->inet_sport)); |
1da177e4 LT |
1316 | |
1317 | if (*errp < 0) { | |
1318 | kfree(svsk); | |
1319 | return NULL; | |
1320 | } | |
1321 | ||
1da177e4 LT |
1322 | inet->sk_user_data = svsk; |
1323 | svsk->sk_sock = sock; | |
1324 | svsk->sk_sk = inet; | |
1325 | svsk->sk_ostate = inet->sk_state_change; | |
1326 | svsk->sk_odata = inet->sk_data_ready; | |
1327 | svsk->sk_owspace = inet->sk_write_space; | |
1da177e4 LT |
1328 | |
1329 | /* Initialize the socket */ | |
1330 | if (sock->type == SOCK_DGRAM) | |
bb5cf160 | 1331 | svc_udp_init(svsk, serv); |
7f421835 | 1332 | else |
bb5cf160 | 1333 | svc_tcp_init(svsk, serv); |
1da177e4 | 1334 | |
1da177e4 LT |
1335 | dprintk("svc: svc_setup_socket created %p (inet %p)\n", |
1336 | svsk, svsk->sk_sk); | |
1337 | ||
1da177e4 LT |
1338 | return svsk; |
1339 | } | |
1340 | ||
bfba9ab4 CL |
1341 | /** |
1342 | * svc_addsock - add a listener socket to an RPC service | |
1343 | * @serv: pointer to RPC service to which to add a new listener | |
1344 | * @fd: file descriptor of the new listener | |
1345 | * @name_return: pointer to buffer to fill in with name of listener | |
1346 | * @len: size of the buffer | |
1347 | * | |
1348 | * Fills in socket name and returns positive length of name if successful. | |
1349 | * Name is terminated with '\n'. On error, returns a negative errno | |
1350 | * value. | |
1351 | */ | |
1352 | int svc_addsock(struct svc_serv *serv, const int fd, char *name_return, | |
1353 | const size_t len) | |
b41b66d6 N |
1354 | { |
1355 | int err = 0; | |
1356 | struct socket *so = sockfd_lookup(fd, &err); | |
1357 | struct svc_sock *svsk = NULL; | |
1358 | ||
1359 | if (!so) | |
1360 | return err; | |
205ba423 | 1361 | if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6)) |
b41b66d6 N |
1362 | err = -EAFNOSUPPORT; |
1363 | else if (so->sk->sk_protocol != IPPROTO_TCP && | |
1364 | so->sk->sk_protocol != IPPROTO_UDP) | |
1365 | err = -EPROTONOSUPPORT; | |
1366 | else if (so->state > SS_UNCONNECTED) | |
1367 | err = -EISCONN; | |
1368 | else { | |
2da2c21d TT |
1369 | if (!try_module_get(THIS_MODULE)) |
1370 | err = -ENOENT; | |
1371 | else | |
1372 | svsk = svc_setup_socket(serv, so, &err, | |
1373 | SVC_SOCK_DEFAULTS); | |
e79eff1f | 1374 | if (svsk) { |
9dbc240f TT |
1375 | struct sockaddr_storage addr; |
1376 | struct sockaddr *sin = (struct sockaddr *)&addr; | |
1377 | int salen; | |
1378 | if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0) | |
1379 | svc_xprt_set_local(&svsk->sk_xprt, sin, salen); | |
4e5caaa5 TT |
1380 | clear_bit(XPT_TEMP, &svsk->sk_xprt.xpt_flags); |
1381 | spin_lock_bh(&serv->sv_lock); | |
1382 | list_add(&svsk->sk_xprt.xpt_list, &serv->sv_permsocks); | |
1383 | spin_unlock_bh(&serv->sv_lock); | |
a6046f71 | 1384 | svc_xprt_received(&svsk->sk_xprt); |
b41b66d6 | 1385 | err = 0; |
2da2c21d TT |
1386 | } else |
1387 | module_put(THIS_MODULE); | |
b41b66d6 N |
1388 | } |
1389 | if (err) { | |
1390 | sockfd_put(so); | |
1391 | return err; | |
1392 | } | |
e7942b9f | 1393 | return svc_one_sock_name(svsk, name_return, len); |
b41b66d6 N |
1394 | } |
1395 | EXPORT_SYMBOL_GPL(svc_addsock); | |
1396 | ||
1da177e4 LT |
1397 | /* |
1398 | * Create socket for RPC service. | |
1399 | */ | |
b700cbb1 TT |
1400 | static struct svc_xprt *svc_create_socket(struct svc_serv *serv, |
1401 | int protocol, | |
1402 | struct sockaddr *sin, int len, | |
1403 | int flags) | |
1da177e4 LT |
1404 | { |
1405 | struct svc_sock *svsk; | |
1406 | struct socket *sock; | |
1407 | int error; | |
1408 | int type; | |
9dbc240f TT |
1409 | struct sockaddr_storage addr; |
1410 | struct sockaddr *newsin = (struct sockaddr *)&addr; | |
1411 | int newlen; | |
c69da774 TM |
1412 | int family; |
1413 | int val; | |
5216a8e7 | 1414 | RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); |
1da177e4 | 1415 | |
ad06e4bd CL |
1416 | dprintk("svc: svc_create_socket(%s, %d, %s)\n", |
1417 | serv->sv_program->pg_name, protocol, | |
77f1f67a | 1418 | __svc_print_addr(sin, buf, sizeof(buf))); |
1da177e4 LT |
1419 | |
1420 | if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) { | |
1421 | printk(KERN_WARNING "svc: only UDP and TCP " | |
1422 | "sockets supported\n"); | |
b700cbb1 | 1423 | return ERR_PTR(-EINVAL); |
1da177e4 | 1424 | } |
c69da774 | 1425 | |
1da177e4 | 1426 | type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM; |
c69da774 TM |
1427 | switch (sin->sa_family) { |
1428 | case AF_INET6: | |
1429 | family = PF_INET6; | |
1430 | break; | |
1431 | case AF_INET: | |
1432 | family = PF_INET; | |
1433 | break; | |
1434 | default: | |
1435 | return ERR_PTR(-EINVAL); | |
1436 | } | |
1da177e4 | 1437 | |
c69da774 | 1438 | error = sock_create_kern(family, type, protocol, &sock); |
77f1f67a | 1439 | if (error < 0) |
b700cbb1 | 1440 | return ERR_PTR(error); |
1da177e4 | 1441 | |
ed07536e PZ |
1442 | svc_reclassify_socket(sock); |
1443 | ||
c69da774 TM |
1444 | /* |
1445 | * If this is an PF_INET6 listener, we want to avoid | |
1446 | * getting requests from IPv4 remotes. Those should | |
1447 | * be shunted to a PF_INET listener via rpcbind. | |
1448 | */ | |
1449 | val = 1; | |
1450 | if (family == PF_INET6) | |
1451 | kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY, | |
1452 | (char *)&val, sizeof(val)); | |
1453 | ||
18114746 | 1454 | if (type == SOCK_STREAM) |
77f1f67a CL |
1455 | sock->sk->sk_reuse = 1; /* allow address reuse */ |
1456 | error = kernel_bind(sock, sin, len); | |
18114746 ES |
1457 | if (error < 0) |
1458 | goto bummer; | |
1da177e4 | 1459 | |
9dbc240f TT |
1460 | newlen = len; |
1461 | error = kernel_getsockname(sock, newsin, &newlen); | |
1462 | if (error < 0) | |
1463 | goto bummer; | |
1464 | ||
1da177e4 | 1465 | if (protocol == IPPROTO_TCP) { |
e6242e92 | 1466 | if ((error = kernel_listen(sock, 64)) < 0) |
1da177e4 LT |
1467 | goto bummer; |
1468 | } | |
1469 | ||
e79eff1f | 1470 | if ((svsk = svc_setup_socket(serv, sock, &error, flags)) != NULL) { |
9dbc240f | 1471 | svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen); |
b700cbb1 | 1472 | return (struct svc_xprt *)svsk; |
e79eff1f | 1473 | } |
1da177e4 LT |
1474 | |
1475 | bummer: | |
1476 | dprintk("svc: svc_create_socket error = %d\n", -error); | |
1477 | sock_release(sock); | |
b700cbb1 | 1478 | return ERR_PTR(error); |
1da177e4 LT |
1479 | } |
1480 | ||
755cceab TT |
1481 | /* |
1482 | * Detach the svc_sock from the socket so that no | |
1483 | * more callbacks occur. | |
1484 | */ | |
1485 | static void svc_sock_detach(struct svc_xprt *xprt) | |
1486 | { | |
1487 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
1488 | struct sock *sk = svsk->sk_sk; | |
1489 | ||
1490 | dprintk("svc: svc_sock_detach(%p)\n", svsk); | |
1491 | ||
1492 | /* put back the old socket callbacks */ | |
1493 | sk->sk_state_change = svsk->sk_ostate; | |
1494 | sk->sk_data_ready = svsk->sk_odata; | |
1495 | sk->sk_write_space = svsk->sk_owspace; | |
69b6ba37 TM |
1496 | |
1497 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) | |
1498 | wake_up_interruptible(sk->sk_sleep); | |
1499 | } | |
1500 | ||
1501 | /* | |
1502 | * Disconnect the socket, and reset the callbacks | |
1503 | */ | |
1504 | static void svc_tcp_sock_detach(struct svc_xprt *xprt) | |
1505 | { | |
1506 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
1507 | ||
1508 | dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk); | |
1509 | ||
1510 | svc_sock_detach(xprt); | |
1511 | ||
1512 | if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) | |
1513 | kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR); | |
755cceab TT |
1514 | } |
1515 | ||
1516 | /* | |
1517 | * Free the svc_sock's socket resources and the svc_sock itself. | |
1518 | */ | |
1519 | static void svc_sock_free(struct svc_xprt *xprt) | |
1520 | { | |
1521 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
1522 | dprintk("svc: svc_sock_free(%p)\n", svsk); | |
1523 | ||
755cceab TT |
1524 | if (svsk->sk_sock->file) |
1525 | sockfd_put(svsk->sk_sock); | |
1526 | else | |
1527 | sock_release(svsk->sk_sock); | |
1528 | kfree(svsk); | |
1529 | } | |
7652e5a0 BH |
1530 | |
1531 | /* | |
1532 | * Create a svc_xprt. | |
1533 | * | |
1534 | * For internal use only (e.g. nfsv4.1 backchannel). | |
1535 | * Callers should typically use the xpo_create() method. | |
1536 | */ | |
1537 | struct svc_xprt *svc_sock_create(struct svc_serv *serv, int prot) | |
1538 | { | |
1539 | struct svc_sock *svsk; | |
1540 | struct svc_xprt *xprt = NULL; | |
1541 | ||
1542 | dprintk("svc: %s\n", __func__); | |
1543 | svsk = kzalloc(sizeof(*svsk), GFP_KERNEL); | |
1544 | if (!svsk) | |
1545 | goto out; | |
1546 | ||
1547 | xprt = &svsk->sk_xprt; | |
1548 | if (prot == IPPROTO_TCP) | |
1549 | svc_xprt_init(&svc_tcp_class, xprt, serv); | |
1550 | else if (prot == IPPROTO_UDP) | |
1551 | svc_xprt_init(&svc_udp_class, xprt, serv); | |
1552 | else | |
1553 | BUG(); | |
1554 | out: | |
1555 | dprintk("svc: %s return %p\n", __func__, xprt); | |
1556 | return xprt; | |
1557 | } | |
1558 | EXPORT_SYMBOL_GPL(svc_sock_create); | |
1559 | ||
1560 | /* | |
1561 | * Destroy a svc_sock. | |
1562 | */ | |
1563 | void svc_sock_destroy(struct svc_xprt *xprt) | |
1564 | { | |
1565 | if (xprt) | |
1566 | kfree(container_of(xprt, struct svc_sock, sk_xprt)); | |
1567 | } | |
1568 | EXPORT_SYMBOL_GPL(svc_sock_destroy); |