]>
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 | 66 | static struct svc_xprt *svc_create_socket(struct svc_serv *, int, |
62832c03 PE |
67 | struct net *, struct sockaddr *, |
68 | int, int); | |
ed07536e PZ |
69 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
70 | static struct lock_class_key svc_key[2]; | |
71 | static struct lock_class_key svc_slock_key[2]; | |
72 | ||
0f0257ea | 73 | static void svc_reclassify_socket(struct socket *sock) |
ed07536e PZ |
74 | { |
75 | struct sock *sk = sock->sk; | |
02b3d346 | 76 | BUG_ON(sock_owned_by_user(sk)); |
ed07536e PZ |
77 | switch (sk->sk_family) { |
78 | case AF_INET: | |
79 | sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD", | |
def13d74 TT |
80 | &svc_slock_key[0], |
81 | "sk_xprt.xpt_lock-AF_INET-NFSD", | |
82 | &svc_key[0]); | |
ed07536e PZ |
83 | break; |
84 | ||
85 | case AF_INET6: | |
86 | sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD", | |
def13d74 TT |
87 | &svc_slock_key[1], |
88 | "sk_xprt.xpt_lock-AF_INET6-NFSD", | |
89 | &svc_key[1]); | |
ed07536e PZ |
90 | break; |
91 | ||
92 | default: | |
93 | BUG(); | |
94 | } | |
95 | } | |
96 | #else | |
0f0257ea | 97 | static void svc_reclassify_socket(struct socket *sock) |
ed07536e PZ |
98 | { |
99 | } | |
100 | #endif | |
101 | ||
1da177e4 LT |
102 | /* |
103 | * Release an skbuff after use | |
104 | */ | |
5148bf4e | 105 | static void svc_release_skb(struct svc_rqst *rqstp) |
1da177e4 | 106 | { |
5148bf4e | 107 | struct sk_buff *skb = rqstp->rq_xprt_ctxt; |
1da177e4 LT |
108 | |
109 | if (skb) { | |
57b1d3ba TT |
110 | struct svc_sock *svsk = |
111 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
5148bf4e | 112 | rqstp->rq_xprt_ctxt = NULL; |
1da177e4 LT |
113 | |
114 | dprintk("svc: service %p, releasing skb %p\n", rqstp, skb); | |
9d410c79 | 115 | skb_free_datagram_locked(svsk->sk_sk, skb); |
1da177e4 | 116 | } |
1da177e4 LT |
117 | } |
118 | ||
b92503b2 CL |
119 | union svc_pktinfo_u { |
120 | struct in_pktinfo pkti; | |
b92503b2 | 121 | struct in6_pktinfo pkti6; |
b92503b2 | 122 | }; |
bc375ea7 DM |
123 | #define SVC_PKTINFO_SPACE \ |
124 | CMSG_SPACE(sizeof(union svc_pktinfo_u)) | |
b92503b2 CL |
125 | |
126 | static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh) | |
127 | { | |
57b1d3ba TT |
128 | struct svc_sock *svsk = |
129 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
130 | switch (svsk->sk_sk->sk_family) { | |
b92503b2 CL |
131 | case AF_INET: { |
132 | struct in_pktinfo *pki = CMSG_DATA(cmh); | |
133 | ||
134 | cmh->cmsg_level = SOL_IP; | |
135 | cmh->cmsg_type = IP_PKTINFO; | |
136 | pki->ipi_ifindex = 0; | |
137 | pki->ipi_spec_dst.s_addr = rqstp->rq_daddr.addr.s_addr; | |
138 | cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); | |
139 | } | |
140 | break; | |
5a05ed73 | 141 | |
b92503b2 CL |
142 | case AF_INET6: { |
143 | struct in6_pktinfo *pki = CMSG_DATA(cmh); | |
144 | ||
145 | cmh->cmsg_level = SOL_IPV6; | |
146 | cmh->cmsg_type = IPV6_PKTINFO; | |
147 | pki->ipi6_ifindex = 0; | |
148 | ipv6_addr_copy(&pki->ipi6_addr, | |
149 | &rqstp->rq_daddr.addr6); | |
150 | cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); | |
151 | } | |
152 | break; | |
b92503b2 | 153 | } |
b92503b2 CL |
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 | } |
aa395145 ED |
422 | if (sk_sleep(sk) && waitqueue_active(sk_sleep(sk))) |
423 | wake_up_interruptible(sk_sleep(sk)); | |
1da177e4 LT |
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 | ||
aa395145 | 439 | if (sk_sleep(sk) && waitqueue_active(sk_sleep(sk))) { |
939bb7ef | 440 | dprintk("RPC svc_write_space: someone sleeping on %p\n", |
1da177e4 | 441 | svsk); |
aa395145 | 442 | wake_up_interruptible(sk_sleep(sk)); |
1da177e4 LT |
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 | } |
05ed690e | 550 | return -EAGAIN; |
1da177e4 | 551 | } |
9dbc240f | 552 | len = svc_addr_len(svc_addr(rqstp)); |
abc5c44d CL |
553 | if (len == 0) |
554 | return -EAFNOSUPPORT; | |
9dbc240f | 555 | rqstp->rq_addrlen = len; |
b7aa0bf7 ED |
556 | if (skb->tstamp.tv64 == 0) { |
557 | skb->tstamp = ktime_get_real(); | |
cca5172a | 558 | /* Don't enable netstamp, sunrpc doesn't |
1da177e4 LT |
559 | need that much accuracy */ |
560 | } | |
b7aa0bf7 | 561 | svsk->sk_sk->sk_stamp = skb->tstamp; |
02fc6c36 | 562 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */ |
1da177e4 | 563 | |
1da177e4 LT |
564 | len = skb->len - sizeof(struct udphdr); |
565 | rqstp->rq_arg.len = len; | |
566 | ||
95756482 | 567 | rqstp->rq_prot = IPPROTO_UDP; |
27459f09 | 568 | |
7702ce40 | 569 | if (!svc_udp_get_dest_address(rqstp, cmh)) { |
7a37f578 | 570 | if (net_ratelimit()) |
7702ce40 CL |
571 | printk(KERN_WARNING |
572 | "svc: received unknown control message %d/%d; " | |
573 | "dropping RPC reply datagram\n", | |
574 | cmh->cmsg_level, cmh->cmsg_type); | |
9d410c79 | 575 | skb_free_datagram_locked(svsk->sk_sk, skb); |
7a37f578 N |
576 | return 0; |
577 | } | |
1da177e4 LT |
578 | |
579 | if (skb_is_nonlinear(skb)) { | |
580 | /* we have to copy */ | |
581 | local_bh_disable(); | |
582 | if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) { | |
583 | local_bh_enable(); | |
584 | /* checksum error */ | |
9d410c79 | 585 | skb_free_datagram_locked(svsk->sk_sk, skb); |
1da177e4 LT |
586 | return 0; |
587 | } | |
588 | local_bh_enable(); | |
9d410c79 | 589 | skb_free_datagram_locked(svsk->sk_sk, skb); |
1da177e4 LT |
590 | } else { |
591 | /* we can use it in-place */ | |
0f0257ea TT |
592 | rqstp->rq_arg.head[0].iov_base = skb->data + |
593 | sizeof(struct udphdr); | |
1da177e4 | 594 | rqstp->rq_arg.head[0].iov_len = len; |
fb286bb2 | 595 | if (skb_checksum_complete(skb)) { |
9d410c79 | 596 | skb_free_datagram_locked(svsk->sk_sk, skb); |
fb286bb2 | 597 | return 0; |
1da177e4 | 598 | } |
5148bf4e | 599 | rqstp->rq_xprt_ctxt = skb; |
1da177e4 LT |
600 | } |
601 | ||
602 | rqstp->rq_arg.page_base = 0; | |
603 | if (len <= rqstp->rq_arg.head[0].iov_len) { | |
604 | rqstp->rq_arg.head[0].iov_len = len; | |
605 | rqstp->rq_arg.page_len = 0; | |
44524359 | 606 | rqstp->rq_respages = rqstp->rq_pages+1; |
1da177e4 LT |
607 | } else { |
608 | rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len; | |
44524359 | 609 | rqstp->rq_respages = rqstp->rq_pages + 1 + |
172589cc | 610 | DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE); |
1da177e4 LT |
611 | } |
612 | ||
613 | if (serv->sv_stats) | |
614 | serv->sv_stats->netudpcnt++; | |
615 | ||
616 | return len; | |
617 | } | |
618 | ||
619 | static int | |
620 | svc_udp_sendto(struct svc_rqst *rqstp) | |
621 | { | |
622 | int error; | |
623 | ||
624 | error = svc_sendto(rqstp, &rqstp->rq_res); | |
625 | if (error == -ECONNREFUSED) | |
626 | /* ICMP error on earlier request. */ | |
627 | error = svc_sendto(rqstp, &rqstp->rq_res); | |
628 | ||
629 | return error; | |
630 | } | |
631 | ||
e831fe65 TT |
632 | static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp) |
633 | { | |
634 | } | |
635 | ||
323bee32 TT |
636 | static int svc_udp_has_wspace(struct svc_xprt *xprt) |
637 | { | |
638 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
bb5cf160 | 639 | struct svc_serv *serv = xprt->xpt_server; |
323bee32 TT |
640 | unsigned long required; |
641 | ||
642 | /* | |
643 | * Set the SOCK_NOSPACE flag before checking the available | |
644 | * sock space. | |
645 | */ | |
646 | set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
7a90e8cc | 647 | required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg; |
323bee32 TT |
648 | if (required*2 > sock_wspace(svsk->sk_sk)) |
649 | return 0; | |
650 | clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
651 | return 1; | |
652 | } | |
653 | ||
38a417cc TT |
654 | static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt) |
655 | { | |
656 | BUG(); | |
657 | return NULL; | |
658 | } | |
659 | ||
b700cbb1 | 660 | static struct svc_xprt *svc_udp_create(struct svc_serv *serv, |
62832c03 | 661 | struct net *net, |
b700cbb1 TT |
662 | struct sockaddr *sa, int salen, |
663 | int flags) | |
664 | { | |
62832c03 | 665 | return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags); |
b700cbb1 TT |
666 | } |
667 | ||
360d8738 | 668 | static struct svc_xprt_ops svc_udp_ops = { |
b700cbb1 | 669 | .xpo_create = svc_udp_create, |
5d137990 TT |
670 | .xpo_recvfrom = svc_udp_recvfrom, |
671 | .xpo_sendto = svc_udp_sendto, | |
5148bf4e | 672 | .xpo_release_rqst = svc_release_skb, |
755cceab TT |
673 | .xpo_detach = svc_sock_detach, |
674 | .xpo_free = svc_sock_free, | |
e831fe65 | 675 | .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr, |
323bee32 | 676 | .xpo_has_wspace = svc_udp_has_wspace, |
38a417cc | 677 | .xpo_accept = svc_udp_accept, |
360d8738 TT |
678 | }; |
679 | ||
680 | static struct svc_xprt_class svc_udp_class = { | |
681 | .xcl_name = "udp", | |
b700cbb1 | 682 | .xcl_owner = THIS_MODULE, |
360d8738 | 683 | .xcl_ops = &svc_udp_ops, |
49023155 | 684 | .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP, |
360d8738 TT |
685 | }; |
686 | ||
bb5cf160 | 687 | static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv) |
1da177e4 | 688 | { |
7702ce40 | 689 | int err, level, optname, one = 1; |
7a37f578 | 690 | |
bb5cf160 | 691 | svc_xprt_init(&svc_udp_class, &svsk->sk_xprt, serv); |
def13d74 | 692 | clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
693 | svsk->sk_sk->sk_data_ready = svc_udp_data_ready; |
694 | svsk->sk_sk->sk_write_space = svc_write_space; | |
1da177e4 LT |
695 | |
696 | /* initialise setting must have enough space to | |
cca5172a | 697 | * receive and respond to one request. |
1da177e4 LT |
698 | * svc_udp_recvfrom will re-adjust if necessary |
699 | */ | |
700 | svc_sock_setbufsize(svsk->sk_sock, | |
bb5cf160 TT |
701 | 3 * svsk->sk_xprt.xpt_server->sv_max_mesg, |
702 | 3 * svsk->sk_xprt.xpt_server->sv_max_mesg); | |
1da177e4 | 703 | |
0f0257ea TT |
704 | /* data might have come in before data_ready set up */ |
705 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); | |
02fc6c36 | 706 | set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); |
7a37f578 | 707 | |
7a37f578 | 708 | /* make sure we get destination address info */ |
7702ce40 CL |
709 | switch (svsk->sk_sk->sk_family) { |
710 | case AF_INET: | |
711 | level = SOL_IP; | |
712 | optname = IP_PKTINFO; | |
713 | break; | |
714 | case AF_INET6: | |
715 | level = SOL_IPV6; | |
716 | optname = IPV6_RECVPKTINFO; | |
717 | break; | |
718 | default: | |
719 | BUG(); | |
720 | } | |
721 | err = kernel_setsockopt(svsk->sk_sock, level, optname, | |
722 | (char *)&one, sizeof(one)); | |
723 | dprintk("svc: kernel_setsockopt returned %d\n", err); | |
1da177e4 LT |
724 | } |
725 | ||
726 | /* | |
727 | * A data_ready event on a listening socket means there's a connection | |
728 | * pending. Do not use state_change as a substitute for it. | |
729 | */ | |
0f0257ea | 730 | static void svc_tcp_listen_data_ready(struct sock *sk, int count_unused) |
1da177e4 | 731 | { |
939bb7ef | 732 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 LT |
733 | |
734 | dprintk("svc: socket %p TCP (listen) state change %d\n", | |
939bb7ef | 735 | sk, sk->sk_state); |
1da177e4 | 736 | |
939bb7ef NB |
737 | /* |
738 | * This callback may called twice when a new connection | |
739 | * is established as a child socket inherits everything | |
740 | * from a parent LISTEN socket. | |
741 | * 1) data_ready method of the parent socket will be called | |
742 | * when one of child sockets become ESTABLISHED. | |
743 | * 2) data_ready method of the child socket may be called | |
744 | * when it receives data before the socket is accepted. | |
745 | * In case of 2, we should ignore it silently. | |
746 | */ | |
747 | if (sk->sk_state == TCP_LISTEN) { | |
748 | if (svsk) { | |
02fc6c36 | 749 | set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
f6150c3c | 750 | svc_xprt_enqueue(&svsk->sk_xprt); |
939bb7ef NB |
751 | } else |
752 | printk("svc: socket %p: no user data\n", sk); | |
1da177e4 | 753 | } |
939bb7ef | 754 | |
aa395145 ED |
755 | if (sk_sleep(sk) && waitqueue_active(sk_sleep(sk))) |
756 | wake_up_interruptible_all(sk_sleep(sk)); | |
1da177e4 LT |
757 | } |
758 | ||
759 | /* | |
760 | * A state change on a connected socket means it's dying or dead. | |
761 | */ | |
0f0257ea | 762 | static void svc_tcp_state_change(struct sock *sk) |
1da177e4 | 763 | { |
939bb7ef | 764 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 LT |
765 | |
766 | dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n", | |
939bb7ef | 767 | sk, sk->sk_state, sk->sk_user_data); |
1da177e4 | 768 | |
939bb7ef | 769 | if (!svsk) |
1da177e4 | 770 | printk("svc: socket %p: no user data\n", sk); |
939bb7ef | 771 | else { |
02fc6c36 | 772 | set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); |
f6150c3c | 773 | svc_xprt_enqueue(&svsk->sk_xprt); |
1da177e4 | 774 | } |
aa395145 ED |
775 | if (sk_sleep(sk) && waitqueue_active(sk_sleep(sk))) |
776 | wake_up_interruptible_all(sk_sleep(sk)); | |
1da177e4 LT |
777 | } |
778 | ||
0f0257ea | 779 | static void svc_tcp_data_ready(struct sock *sk, int count) |
1da177e4 | 780 | { |
939bb7ef | 781 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 LT |
782 | |
783 | dprintk("svc: socket %p TCP data ready (svsk %p)\n", | |
939bb7ef NB |
784 | sk, sk->sk_user_data); |
785 | if (svsk) { | |
02fc6c36 | 786 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
f6150c3c | 787 | svc_xprt_enqueue(&svsk->sk_xprt); |
939bb7ef | 788 | } |
aa395145 ED |
789 | if (sk_sleep(sk) && waitqueue_active(sk_sleep(sk))) |
790 | wake_up_interruptible(sk_sleep(sk)); | |
1da177e4 LT |
791 | } |
792 | ||
793 | /* | |
794 | * Accept a TCP connection | |
795 | */ | |
38a417cc | 796 | static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt) |
1da177e4 | 797 | { |
38a417cc | 798 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); |
cdd88b9f AM |
799 | struct sockaddr_storage addr; |
800 | struct sockaddr *sin = (struct sockaddr *) &addr; | |
bb5cf160 | 801 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
1da177e4 LT |
802 | struct socket *sock = svsk->sk_sock; |
803 | struct socket *newsock; | |
1da177e4 LT |
804 | struct svc_sock *newsvsk; |
805 | int err, slen; | |
5216a8e7 | 806 | RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); |
1da177e4 LT |
807 | |
808 | dprintk("svc: tcp_accept %p sock %p\n", svsk, sock); | |
809 | if (!sock) | |
38a417cc | 810 | return NULL; |
1da177e4 | 811 | |
02fc6c36 | 812 | clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
e6242e92 SS |
813 | err = kernel_accept(sock, &newsock, O_NONBLOCK); |
814 | if (err < 0) { | |
1da177e4 LT |
815 | if (err == -ENOMEM) |
816 | printk(KERN_WARNING "%s: no more sockets!\n", | |
817 | serv->sv_name); | |
e6242e92 | 818 | else if (err != -EAGAIN && net_ratelimit()) |
1da177e4 LT |
819 | printk(KERN_WARNING "%s: accept failed (err %d)!\n", |
820 | serv->sv_name, -err); | |
38a417cc | 821 | return NULL; |
1da177e4 | 822 | } |
02fc6c36 | 823 | set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 824 | |
cdd88b9f | 825 | err = kernel_getpeername(newsock, sin, &slen); |
1da177e4 LT |
826 | if (err < 0) { |
827 | if (net_ratelimit()) | |
828 | printk(KERN_WARNING "%s: peername failed (err %d)!\n", | |
829 | serv->sv_name, -err); | |
830 | goto failed; /* aborted connection or whatever */ | |
831 | } | |
832 | ||
833 | /* Ideally, we would want to reject connections from unauthorized | |
ad06e4bd CL |
834 | * hosts here, but when we get encryption, the IP of the host won't |
835 | * tell us anything. For now just warn about unpriv connections. | |
1da177e4 | 836 | */ |
cdd88b9f | 837 | if (!svc_port_is_privileged(sin)) { |
1da177e4 | 838 | dprintk(KERN_WARNING |
ad06e4bd | 839 | "%s: connect from unprivileged port: %s\n", |
cca5172a | 840 | serv->sv_name, |
cdd88b9f | 841 | __svc_print_addr(sin, buf, sizeof(buf))); |
1da177e4 | 842 | } |
ad06e4bd | 843 | dprintk("%s: connect from %s\n", serv->sv_name, |
cdd88b9f | 844 | __svc_print_addr(sin, buf, sizeof(buf))); |
1da177e4 LT |
845 | |
846 | /* make sure that a write doesn't block forever when | |
847 | * low on memory | |
848 | */ | |
849 | newsock->sk->sk_sndtimeo = HZ*30; | |
850 | ||
6b174337 CL |
851 | if (!(newsvsk = svc_setup_socket(serv, newsock, &err, |
852 | (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY)))) | |
1da177e4 | 853 | goto failed; |
9dbc240f | 854 | svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen); |
a9747692 FM |
855 | err = kernel_getsockname(newsock, sin, &slen); |
856 | if (unlikely(err < 0)) { | |
857 | dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err); | |
858 | slen = offsetof(struct sockaddr, sa_data); | |
859 | } | |
9dbc240f | 860 | svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen); |
067d7817 | 861 | |
f9f3cc4f TT |
862 | if (serv->sv_stats) |
863 | serv->sv_stats->nettcpconn++; | |
864 | ||
865 | return &newsvsk->sk_xprt; | |
866 | ||
867 | failed: | |
868 | sock_release(newsock); | |
869 | return NULL; | |
870 | } | |
871 | ||
1da177e4 | 872 | /* |
8f55f3c0 AB |
873 | * Receive data. |
874 | * If we haven't gotten the record length yet, get the next four bytes. | |
875 | * Otherwise try to gobble up as much as possible up to the complete | |
876 | * record length. | |
1da177e4 | 877 | */ |
8f55f3c0 | 878 | static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp) |
1da177e4 | 879 | { |
bb5cf160 | 880 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
8f55f3c0 | 881 | int len; |
1da177e4 | 882 | |
7f421835 BF |
883 | if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags)) |
884 | /* sndbuf needs to have room for one request | |
885 | * per thread, otherwise we can stall even when the | |
886 | * network isn't a bottleneck. | |
887 | * | |
888 | * We count all threads rather than threads in a | |
889 | * particular pool, which provides an upper bound | |
890 | * on the number of threads which will access the socket. | |
891 | * | |
892 | * rcvbuf just needs to be able to hold a few requests. | |
893 | * Normally they will be removed from the queue | |
894 | * as soon a a complete request arrives. | |
895 | */ | |
896 | svc_sock_setbufsize(svsk->sk_sock, | |
897 | (serv->sv_nrthreads+3) * serv->sv_max_mesg, | |
898 | 3 * serv->sv_max_mesg); | |
899 | ||
02fc6c36 | 900 | clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 901 | |
c0401ea0 CL |
902 | if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) { |
903 | int want = sizeof(rpc_fraghdr) - svsk->sk_tcplen; | |
1da177e4 LT |
904 | struct kvec iov; |
905 | ||
906 | iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen; | |
907 | iov.iov_len = want; | |
908 | if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0) | |
909 | goto error; | |
910 | svsk->sk_tcplen += len; | |
911 | ||
912 | if (len < want) { | |
c0401ea0 CL |
913 | dprintk("svc: short recvfrom while reading record " |
914 | "length (%d of %d)\n", len, want); | |
8f55f3c0 | 915 | goto err_again; /* record header not complete */ |
1da177e4 LT |
916 | } |
917 | ||
918 | svsk->sk_reclen = ntohl(svsk->sk_reclen); | |
c0401ea0 | 919 | if (!(svsk->sk_reclen & RPC_LAST_STREAM_FRAGMENT)) { |
1da177e4 LT |
920 | /* FIXME: technically, a record can be fragmented, |
921 | * and non-terminal fragments will not have the top | |
922 | * bit set in the fragment length header. | |
923 | * But apparently no known nfs clients send fragmented | |
924 | * records. */ | |
34e9a63b | 925 | if (net_ratelimit()) |
c0401ea0 CL |
926 | printk(KERN_NOTICE "RPC: multiple fragments " |
927 | "per record not supported\n"); | |
1da177e4 LT |
928 | goto err_delete; |
929 | } | |
8f55f3c0 | 930 | |
c0401ea0 | 931 | svsk->sk_reclen &= RPC_FRAGMENT_SIZE_MASK; |
1da177e4 | 932 | dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen); |
c6b0a9f8 | 933 | if (svsk->sk_reclen > serv->sv_max_mesg) { |
34e9a63b | 934 | if (net_ratelimit()) |
c0401ea0 CL |
935 | printk(KERN_NOTICE "RPC: " |
936 | "fragment too large: 0x%08lx\n", | |
937 | (unsigned long)svsk->sk_reclen); | |
1da177e4 LT |
938 | goto err_delete; |
939 | } | |
940 | } | |
941 | ||
942 | /* Check whether enough data is available */ | |
943 | len = svc_recv_available(svsk); | |
944 | if (len < 0) | |
945 | goto error; | |
946 | ||
947 | if (len < svsk->sk_reclen) { | |
948 | dprintk("svc: incomplete TCP record (%d of %d)\n", | |
949 | len, svsk->sk_reclen); | |
8f55f3c0 | 950 | goto err_again; /* record not complete */ |
1da177e4 LT |
951 | } |
952 | len = svsk->sk_reclen; | |
02fc6c36 | 953 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 954 | |
8f55f3c0 AB |
955 | return len; |
956 | error: | |
b48fa6b9 | 957 | if (len == -EAGAIN) |
8f55f3c0 | 958 | dprintk("RPC: TCP recv_record got EAGAIN\n"); |
8f55f3c0 AB |
959 | return len; |
960 | err_delete: | |
961 | set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); | |
962 | err_again: | |
963 | return -EAGAIN; | |
964 | } | |
965 | ||
4cfc7e60 RI |
966 | static int svc_process_calldir(struct svc_sock *svsk, struct svc_rqst *rqstp, |
967 | struct rpc_rqst **reqpp, struct kvec *vec) | |
968 | { | |
969 | struct rpc_rqst *req = NULL; | |
970 | u32 *p; | |
971 | u32 xid; | |
972 | u32 calldir; | |
973 | int len; | |
974 | ||
975 | len = svc_recvfrom(rqstp, vec, 1, 8); | |
976 | if (len < 0) | |
977 | goto error; | |
978 | ||
979 | p = (u32 *)rqstp->rq_arg.head[0].iov_base; | |
980 | xid = *p++; | |
981 | calldir = *p; | |
982 | ||
983 | if (calldir == 0) { | |
984 | /* REQUEST is the most common case */ | |
985 | vec[0] = rqstp->rq_arg.head[0]; | |
986 | } else { | |
987 | /* REPLY */ | |
988 | if (svsk->sk_bc_xprt) | |
989 | req = xprt_lookup_rqst(svsk->sk_bc_xprt, xid); | |
990 | ||
991 | if (!req) { | |
992 | printk(KERN_NOTICE | |
993 | "%s: Got unrecognized reply: " | |
994 | "calldir 0x%x sk_bc_xprt %p xid %08x\n", | |
995 | __func__, ntohl(calldir), | |
996 | svsk->sk_bc_xprt, xid); | |
997 | vec[0] = rqstp->rq_arg.head[0]; | |
998 | goto out; | |
999 | } | |
1000 | ||
1001 | memcpy(&req->rq_private_buf, &req->rq_rcv_buf, | |
1002 | sizeof(struct xdr_buf)); | |
1003 | /* copy the xid and call direction */ | |
1004 | memcpy(req->rq_private_buf.head[0].iov_base, | |
1005 | rqstp->rq_arg.head[0].iov_base, 8); | |
1006 | vec[0] = req->rq_private_buf.head[0]; | |
1007 | } | |
1008 | out: | |
1009 | vec[0].iov_base += 8; | |
1010 | vec[0].iov_len -= 8; | |
1011 | len = svsk->sk_reclen - 8; | |
1012 | error: | |
1013 | *reqpp = req; | |
1014 | return len; | |
1015 | } | |
1016 | ||
8f55f3c0 AB |
1017 | /* |
1018 | * Receive data from a TCP socket. | |
1019 | */ | |
1020 | static int svc_tcp_recvfrom(struct svc_rqst *rqstp) | |
1021 | { | |
1022 | struct svc_sock *svsk = | |
1023 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
1024 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; | |
1025 | int len; | |
1026 | struct kvec *vec; | |
1027 | int pnum, vlen; | |
4cfc7e60 | 1028 | struct rpc_rqst *req = NULL; |
8f55f3c0 AB |
1029 | |
1030 | dprintk("svc: tcp_recv %p data %d conn %d close %d\n", | |
1031 | svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags), | |
1032 | test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags), | |
1033 | test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags)); | |
1034 | ||
1035 | len = svc_tcp_recv_record(svsk, rqstp); | |
1036 | if (len < 0) | |
1037 | goto error; | |
1038 | ||
3cc03b16 | 1039 | vec = rqstp->rq_vec; |
1da177e4 LT |
1040 | vec[0] = rqstp->rq_arg.head[0]; |
1041 | vlen = PAGE_SIZE; | |
4cfc7e60 RI |
1042 | |
1043 | /* | |
1044 | * We have enough data for the whole tcp record. Let's try and read the | |
1045 | * first 8 bytes to get the xid and the call direction. We can use this | |
1046 | * to figure out if this is a call or a reply to a callback. If | |
1047 | * sk_reclen is < 8 (xid and calldir), then this is a malformed packet. | |
1048 | * In that case, don't bother with the calldir and just read the data. | |
1049 | * It will be rejected in svc_process. | |
1050 | */ | |
1051 | if (len >= 8) { | |
1052 | len = svc_process_calldir(svsk, rqstp, &req, vec); | |
1053 | if (len < 0) | |
1054 | goto err_again; | |
1055 | vlen -= 8; | |
1056 | } | |
1057 | ||
1da177e4 LT |
1058 | pnum = 1; |
1059 | while (vlen < len) { | |
4cfc7e60 RI |
1060 | vec[pnum].iov_base = (req) ? |
1061 | page_address(req->rq_private_buf.pages[pnum - 1]) : | |
1062 | page_address(rqstp->rq_pages[pnum]); | |
1da177e4 LT |
1063 | vec[pnum].iov_len = PAGE_SIZE; |
1064 | pnum++; | |
1065 | vlen += PAGE_SIZE; | |
1066 | } | |
44524359 | 1067 | rqstp->rq_respages = &rqstp->rq_pages[pnum]; |
1da177e4 LT |
1068 | |
1069 | /* Now receive data */ | |
1070 | len = svc_recvfrom(rqstp, vec, pnum, len); | |
1071 | if (len < 0) | |
8f55f3c0 | 1072 | goto err_again; |
1da177e4 | 1073 | |
4cfc7e60 RI |
1074 | /* |
1075 | * Account for the 8 bytes we read earlier | |
1076 | */ | |
1077 | len += 8; | |
1078 | ||
1079 | if (req) { | |
1080 | xprt_complete_rqst(req->rq_task, len); | |
1081 | len = 0; | |
1082 | goto out; | |
1083 | } | |
1da177e4 LT |
1084 | dprintk("svc: TCP complete record (%d bytes)\n", len); |
1085 | rqstp->rq_arg.len = len; | |
1086 | rqstp->rq_arg.page_base = 0; | |
1087 | if (len <= rqstp->rq_arg.head[0].iov_len) { | |
1088 | rqstp->rq_arg.head[0].iov_len = len; | |
1089 | rqstp->rq_arg.page_len = 0; | |
1090 | } else { | |
1091 | rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len; | |
1092 | } | |
1093 | ||
5148bf4e | 1094 | rqstp->rq_xprt_ctxt = NULL; |
1da177e4 LT |
1095 | rqstp->rq_prot = IPPROTO_TCP; |
1096 | ||
4cfc7e60 | 1097 | out: |
1da177e4 LT |
1098 | /* Reset TCP read info */ |
1099 | svsk->sk_reclen = 0; | |
1100 | svsk->sk_tcplen = 0; | |
1101 | ||
9dbc240f | 1102 | svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt); |
1da177e4 LT |
1103 | if (serv->sv_stats) |
1104 | serv->sv_stats->nettcpcnt++; | |
1105 | ||
1106 | return len; | |
1107 | ||
8f55f3c0 | 1108 | err_again: |
1da177e4 LT |
1109 | if (len == -EAGAIN) { |
1110 | dprintk("RPC: TCP recvfrom got EAGAIN\n"); | |
8f55f3c0 AB |
1111 | return len; |
1112 | } | |
1113 | error: | |
1114 | if (len != -EAGAIN) { | |
1da177e4 | 1115 | printk(KERN_NOTICE "%s: recvfrom returned errno %d\n", |
bb5cf160 | 1116 | svsk->sk_xprt.xpt_server->sv_name, -len); |
8f55f3c0 | 1117 | set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 1118 | } |
8f55f3c0 | 1119 | return -EAGAIN; |
1da177e4 LT |
1120 | } |
1121 | ||
1122 | /* | |
1123 | * Send out data on TCP socket. | |
1124 | */ | |
0f0257ea | 1125 | static int svc_tcp_sendto(struct svc_rqst *rqstp) |
1da177e4 LT |
1126 | { |
1127 | struct xdr_buf *xbufp = &rqstp->rq_res; | |
1128 | int sent; | |
d8ed029d | 1129 | __be32 reclen; |
1da177e4 LT |
1130 | |
1131 | /* Set up the first element of the reply kvec. | |
1132 | * Any other kvecs that may be in use have been taken | |
1133 | * care of by the server implementation itself. | |
1134 | */ | |
1135 | reclen = htonl(0x80000000|((xbufp->len ) - 4)); | |
1136 | memcpy(xbufp->head[0].iov_base, &reclen, 4); | |
1137 | ||
57b1d3ba | 1138 | if (test_bit(XPT_DEAD, &rqstp->rq_xprt->xpt_flags)) |
1da177e4 LT |
1139 | return -ENOTCONN; |
1140 | ||
1141 | sent = svc_sendto(rqstp, &rqstp->rq_res); | |
1142 | if (sent != xbufp->len) { | |
0f0257ea TT |
1143 | printk(KERN_NOTICE |
1144 | "rpc-srv/tcp: %s: %s %d when sending %d bytes " | |
1145 | "- shutting down socket\n", | |
57b1d3ba | 1146 | rqstp->rq_xprt->xpt_server->sv_name, |
1da177e4 LT |
1147 | (sent<0)?"got error":"sent only", |
1148 | sent, xbufp->len); | |
57b1d3ba | 1149 | set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags); |
f6150c3c | 1150 | svc_xprt_enqueue(rqstp->rq_xprt); |
1da177e4 LT |
1151 | sent = -EAGAIN; |
1152 | } | |
1153 | return sent; | |
1154 | } | |
1155 | ||
e831fe65 TT |
1156 | /* |
1157 | * Setup response header. TCP has a 4B record length field. | |
1158 | */ | |
1159 | static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp) | |
1160 | { | |
1161 | struct kvec *resv = &rqstp->rq_res.head[0]; | |
1162 | ||
1163 | /* tcp needs a space for the record length... */ | |
1164 | svc_putnl(resv, 0); | |
1165 | } | |
1166 | ||
323bee32 TT |
1167 | static int svc_tcp_has_wspace(struct svc_xprt *xprt) |
1168 | { | |
57b1d3ba | 1169 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); |
47fcb03f | 1170 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
323bee32 | 1171 | int required; |
323bee32 | 1172 | |
47fcb03f TM |
1173 | if (test_bit(XPT_LISTENER, &xprt->xpt_flags)) |
1174 | return 1; | |
1175 | required = atomic_read(&xprt->xpt_reserved) + serv->sv_max_mesg; | |
1176 | if (sk_stream_wspace(svsk->sk_sk) >= required) | |
1177 | return 1; | |
323bee32 | 1178 | set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); |
47fcb03f | 1179 | return 0; |
323bee32 TT |
1180 | } |
1181 | ||
b700cbb1 | 1182 | static struct svc_xprt *svc_tcp_create(struct svc_serv *serv, |
62832c03 | 1183 | struct net *net, |
b700cbb1 TT |
1184 | struct sockaddr *sa, int salen, |
1185 | int flags) | |
1186 | { | |
62832c03 | 1187 | return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags); |
b700cbb1 TT |
1188 | } |
1189 | ||
360d8738 | 1190 | static struct svc_xprt_ops svc_tcp_ops = { |
b700cbb1 | 1191 | .xpo_create = svc_tcp_create, |
5d137990 TT |
1192 | .xpo_recvfrom = svc_tcp_recvfrom, |
1193 | .xpo_sendto = svc_tcp_sendto, | |
5148bf4e | 1194 | .xpo_release_rqst = svc_release_skb, |
69b6ba37 | 1195 | .xpo_detach = svc_tcp_sock_detach, |
755cceab | 1196 | .xpo_free = svc_sock_free, |
e831fe65 | 1197 | .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr, |
323bee32 | 1198 | .xpo_has_wspace = svc_tcp_has_wspace, |
38a417cc | 1199 | .xpo_accept = svc_tcp_accept, |
360d8738 TT |
1200 | }; |
1201 | ||
1202 | static struct svc_xprt_class svc_tcp_class = { | |
1203 | .xcl_name = "tcp", | |
b700cbb1 | 1204 | .xcl_owner = THIS_MODULE, |
360d8738 | 1205 | .xcl_ops = &svc_tcp_ops, |
49023155 | 1206 | .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP, |
360d8738 TT |
1207 | }; |
1208 | ||
1209 | void svc_init_xprt_sock(void) | |
1210 | { | |
1211 | svc_reg_xprt_class(&svc_tcp_class); | |
1212 | svc_reg_xprt_class(&svc_udp_class); | |
1213 | } | |
1214 | ||
1215 | void svc_cleanup_xprt_sock(void) | |
1216 | { | |
1217 | svc_unreg_xprt_class(&svc_tcp_class); | |
1218 | svc_unreg_xprt_class(&svc_udp_class); | |
1219 | } | |
1220 | ||
bb5cf160 | 1221 | static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv) |
1da177e4 LT |
1222 | { |
1223 | struct sock *sk = svsk->sk_sk; | |
1da177e4 | 1224 | |
bb5cf160 | 1225 | svc_xprt_init(&svc_tcp_class, &svsk->sk_xprt, serv); |
def13d74 | 1226 | set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
1227 | if (sk->sk_state == TCP_LISTEN) { |
1228 | dprintk("setting up TCP socket for listening\n"); | |
02fc6c36 | 1229 | set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 1230 | sk->sk_data_ready = svc_tcp_listen_data_ready; |
02fc6c36 | 1231 | set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
1232 | } else { |
1233 | dprintk("setting up TCP socket for reading\n"); | |
1234 | sk->sk_state_change = svc_tcp_state_change; | |
1235 | sk->sk_data_ready = svc_tcp_data_ready; | |
47fcb03f | 1236 | sk->sk_write_space = svc_tcp_write_space; |
1da177e4 LT |
1237 | |
1238 | svsk->sk_reclen = 0; | |
1239 | svsk->sk_tcplen = 0; | |
1240 | ||
b7872fe8 | 1241 | tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF; |
1da177e4 | 1242 | |
7f421835 BF |
1243 | /* initialise setting must have enough space to |
1244 | * receive and respond to one request. | |
1245 | * svc_tcp_recvfrom will re-adjust if necessary | |
1246 | */ | |
1247 | svc_sock_setbufsize(svsk->sk_sock, | |
1248 | 3 * svsk->sk_xprt.xpt_server->sv_max_mesg, | |
1249 | 3 * svsk->sk_xprt.xpt_server->sv_max_mesg); | |
1250 | ||
1251 | set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); | |
02fc6c36 | 1252 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
cca5172a | 1253 | if (sk->sk_state != TCP_ESTABLISHED) |
02fc6c36 | 1254 | set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
1255 | } |
1256 | } | |
1257 | ||
0f0257ea | 1258 | void svc_sock_update_bufs(struct svc_serv *serv) |
1da177e4 LT |
1259 | { |
1260 | /* | |
1261 | * The number of server threads has changed. Update | |
1262 | * rcvbuf and sndbuf accordingly on all sockets | |
1263 | */ | |
1264 | struct list_head *le; | |
1265 | ||
1266 | spin_lock_bh(&serv->sv_lock); | |
1267 | list_for_each(le, &serv->sv_permsocks) { | |
cca5172a | 1268 | struct svc_sock *svsk = |
7a182083 | 1269 | list_entry(le, struct svc_sock, sk_xprt.xpt_list); |
02fc6c36 | 1270 | set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
1271 | } |
1272 | list_for_each(le, &serv->sv_tempsocks) { | |
1273 | struct svc_sock *svsk = | |
7a182083 | 1274 | list_entry(le, struct svc_sock, sk_xprt.xpt_list); |
02fc6c36 | 1275 | set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
1276 | } |
1277 | spin_unlock_bh(&serv->sv_lock); | |
1278 | } | |
24c3767e | 1279 | EXPORT_SYMBOL_GPL(svc_sock_update_bufs); |
1da177e4 | 1280 | |
1da177e4 LT |
1281 | /* |
1282 | * Initialize socket for RPC use and create svc_sock struct | |
1283 | * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF. | |
1284 | */ | |
6b174337 CL |
1285 | static struct svc_sock *svc_setup_socket(struct svc_serv *serv, |
1286 | struct socket *sock, | |
1287 | int *errp, int flags) | |
1da177e4 LT |
1288 | { |
1289 | struct svc_sock *svsk; | |
1290 | struct sock *inet; | |
6b174337 | 1291 | int pmap_register = !(flags & SVC_SOCK_ANONYMOUS); |
1da177e4 LT |
1292 | |
1293 | dprintk("svc: svc_setup_socket %p\n", sock); | |
0da974f4 | 1294 | if (!(svsk = kzalloc(sizeof(*svsk), GFP_KERNEL))) { |
1da177e4 LT |
1295 | *errp = -ENOMEM; |
1296 | return NULL; | |
1297 | } | |
1da177e4 LT |
1298 | |
1299 | inet = sock->sk; | |
1300 | ||
1301 | /* Register socket with portmapper */ | |
1302 | if (*errp >= 0 && pmap_register) | |
baf01caf | 1303 | *errp = svc_register(serv, inet->sk_family, inet->sk_protocol, |
c720c7e8 | 1304 | ntohs(inet_sk(inet)->inet_sport)); |
1da177e4 LT |
1305 | |
1306 | if (*errp < 0) { | |
1307 | kfree(svsk); | |
1308 | return NULL; | |
1309 | } | |
1310 | ||
1da177e4 LT |
1311 | inet->sk_user_data = svsk; |
1312 | svsk->sk_sock = sock; | |
1313 | svsk->sk_sk = inet; | |
1314 | svsk->sk_ostate = inet->sk_state_change; | |
1315 | svsk->sk_odata = inet->sk_data_ready; | |
1316 | svsk->sk_owspace = inet->sk_write_space; | |
1da177e4 LT |
1317 | |
1318 | /* Initialize the socket */ | |
1319 | if (sock->type == SOCK_DGRAM) | |
bb5cf160 | 1320 | svc_udp_init(svsk, serv); |
7f421835 | 1321 | else |
bb5cf160 | 1322 | svc_tcp_init(svsk, serv); |
1da177e4 | 1323 | |
1da177e4 LT |
1324 | dprintk("svc: svc_setup_socket created %p (inet %p)\n", |
1325 | svsk, svsk->sk_sk); | |
1326 | ||
1da177e4 LT |
1327 | return svsk; |
1328 | } | |
1329 | ||
bfba9ab4 CL |
1330 | /** |
1331 | * svc_addsock - add a listener socket to an RPC service | |
1332 | * @serv: pointer to RPC service to which to add a new listener | |
1333 | * @fd: file descriptor of the new listener | |
1334 | * @name_return: pointer to buffer to fill in with name of listener | |
1335 | * @len: size of the buffer | |
1336 | * | |
1337 | * Fills in socket name and returns positive length of name if successful. | |
1338 | * Name is terminated with '\n'. On error, returns a negative errno | |
1339 | * value. | |
1340 | */ | |
1341 | int svc_addsock(struct svc_serv *serv, const int fd, char *name_return, | |
1342 | const size_t len) | |
b41b66d6 N |
1343 | { |
1344 | int err = 0; | |
1345 | struct socket *so = sockfd_lookup(fd, &err); | |
1346 | struct svc_sock *svsk = NULL; | |
1347 | ||
1348 | if (!so) | |
1349 | return err; | |
205ba423 | 1350 | if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6)) |
b41b66d6 N |
1351 | err = -EAFNOSUPPORT; |
1352 | else if (so->sk->sk_protocol != IPPROTO_TCP && | |
1353 | so->sk->sk_protocol != IPPROTO_UDP) | |
1354 | err = -EPROTONOSUPPORT; | |
1355 | else if (so->state > SS_UNCONNECTED) | |
1356 | err = -EISCONN; | |
1357 | else { | |
2da2c21d TT |
1358 | if (!try_module_get(THIS_MODULE)) |
1359 | err = -ENOENT; | |
1360 | else | |
1361 | svsk = svc_setup_socket(serv, so, &err, | |
1362 | SVC_SOCK_DEFAULTS); | |
e79eff1f | 1363 | if (svsk) { |
9dbc240f TT |
1364 | struct sockaddr_storage addr; |
1365 | struct sockaddr *sin = (struct sockaddr *)&addr; | |
1366 | int salen; | |
1367 | if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0) | |
1368 | svc_xprt_set_local(&svsk->sk_xprt, sin, salen); | |
4e5caaa5 TT |
1369 | clear_bit(XPT_TEMP, &svsk->sk_xprt.xpt_flags); |
1370 | spin_lock_bh(&serv->sv_lock); | |
1371 | list_add(&svsk->sk_xprt.xpt_list, &serv->sv_permsocks); | |
1372 | spin_unlock_bh(&serv->sv_lock); | |
a6046f71 | 1373 | svc_xprt_received(&svsk->sk_xprt); |
b41b66d6 | 1374 | err = 0; |
2da2c21d TT |
1375 | } else |
1376 | module_put(THIS_MODULE); | |
b41b66d6 N |
1377 | } |
1378 | if (err) { | |
1379 | sockfd_put(so); | |
1380 | return err; | |
1381 | } | |
e7942b9f | 1382 | return svc_one_sock_name(svsk, name_return, len); |
b41b66d6 N |
1383 | } |
1384 | EXPORT_SYMBOL_GPL(svc_addsock); | |
1385 | ||
1da177e4 LT |
1386 | /* |
1387 | * Create socket for RPC service. | |
1388 | */ | |
b700cbb1 TT |
1389 | static struct svc_xprt *svc_create_socket(struct svc_serv *serv, |
1390 | int protocol, | |
62832c03 | 1391 | struct net *net, |
b700cbb1 TT |
1392 | struct sockaddr *sin, int len, |
1393 | int flags) | |
1da177e4 LT |
1394 | { |
1395 | struct svc_sock *svsk; | |
1396 | struct socket *sock; | |
1397 | int error; | |
1398 | int type; | |
9dbc240f TT |
1399 | struct sockaddr_storage addr; |
1400 | struct sockaddr *newsin = (struct sockaddr *)&addr; | |
1401 | int newlen; | |
c69da774 TM |
1402 | int family; |
1403 | int val; | |
5216a8e7 | 1404 | RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); |
1da177e4 | 1405 | |
ad06e4bd CL |
1406 | dprintk("svc: svc_create_socket(%s, %d, %s)\n", |
1407 | serv->sv_program->pg_name, protocol, | |
77f1f67a | 1408 | __svc_print_addr(sin, buf, sizeof(buf))); |
1da177e4 LT |
1409 | |
1410 | if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) { | |
1411 | printk(KERN_WARNING "svc: only UDP and TCP " | |
1412 | "sockets supported\n"); | |
b700cbb1 | 1413 | return ERR_PTR(-EINVAL); |
1da177e4 | 1414 | } |
c69da774 | 1415 | |
1da177e4 | 1416 | type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM; |
c69da774 TM |
1417 | switch (sin->sa_family) { |
1418 | case AF_INET6: | |
1419 | family = PF_INET6; | |
1420 | break; | |
1421 | case AF_INET: | |
1422 | family = PF_INET; | |
1423 | break; | |
1424 | default: | |
1425 | return ERR_PTR(-EINVAL); | |
1426 | } | |
1da177e4 | 1427 | |
c69da774 | 1428 | error = sock_create_kern(family, type, protocol, &sock); |
77f1f67a | 1429 | if (error < 0) |
b700cbb1 | 1430 | return ERR_PTR(error); |
1da177e4 | 1431 | |
ed07536e PZ |
1432 | svc_reclassify_socket(sock); |
1433 | ||
c69da774 TM |
1434 | /* |
1435 | * If this is an PF_INET6 listener, we want to avoid | |
1436 | * getting requests from IPv4 remotes. Those should | |
1437 | * be shunted to a PF_INET listener via rpcbind. | |
1438 | */ | |
1439 | val = 1; | |
1440 | if (family == PF_INET6) | |
1441 | kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY, | |
1442 | (char *)&val, sizeof(val)); | |
1443 | ||
18114746 | 1444 | if (type == SOCK_STREAM) |
77f1f67a CL |
1445 | sock->sk->sk_reuse = 1; /* allow address reuse */ |
1446 | error = kernel_bind(sock, sin, len); | |
18114746 ES |
1447 | if (error < 0) |
1448 | goto bummer; | |
1da177e4 | 1449 | |
9dbc240f TT |
1450 | newlen = len; |
1451 | error = kernel_getsockname(sock, newsin, &newlen); | |
1452 | if (error < 0) | |
1453 | goto bummer; | |
1454 | ||
1da177e4 | 1455 | if (protocol == IPPROTO_TCP) { |
e6242e92 | 1456 | if ((error = kernel_listen(sock, 64)) < 0) |
1da177e4 LT |
1457 | goto bummer; |
1458 | } | |
1459 | ||
e79eff1f | 1460 | if ((svsk = svc_setup_socket(serv, sock, &error, flags)) != NULL) { |
9dbc240f | 1461 | svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen); |
b700cbb1 | 1462 | return (struct svc_xprt *)svsk; |
e79eff1f | 1463 | } |
1da177e4 LT |
1464 | |
1465 | bummer: | |
1466 | dprintk("svc: svc_create_socket error = %d\n", -error); | |
1467 | sock_release(sock); | |
b700cbb1 | 1468 | return ERR_PTR(error); |
1da177e4 LT |
1469 | } |
1470 | ||
755cceab TT |
1471 | /* |
1472 | * Detach the svc_sock from the socket so that no | |
1473 | * more callbacks occur. | |
1474 | */ | |
1475 | static void svc_sock_detach(struct svc_xprt *xprt) | |
1476 | { | |
1477 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
1478 | struct sock *sk = svsk->sk_sk; | |
1479 | ||
1480 | dprintk("svc: svc_sock_detach(%p)\n", svsk); | |
1481 | ||
1482 | /* put back the old socket callbacks */ | |
1483 | sk->sk_state_change = svsk->sk_ostate; | |
1484 | sk->sk_data_ready = svsk->sk_odata; | |
1485 | sk->sk_write_space = svsk->sk_owspace; | |
69b6ba37 | 1486 | |
aa395145 ED |
1487 | if (sk_sleep(sk) && waitqueue_active(sk_sleep(sk))) |
1488 | wake_up_interruptible(sk_sleep(sk)); | |
69b6ba37 TM |
1489 | } |
1490 | ||
1491 | /* | |
1492 | * Disconnect the socket, and reset the callbacks | |
1493 | */ | |
1494 | static void svc_tcp_sock_detach(struct svc_xprt *xprt) | |
1495 | { | |
1496 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
1497 | ||
1498 | dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk); | |
1499 | ||
1500 | svc_sock_detach(xprt); | |
1501 | ||
1502 | if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) | |
1503 | kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR); | |
755cceab TT |
1504 | } |
1505 | ||
1506 | /* | |
1507 | * Free the svc_sock's socket resources and the svc_sock itself. | |
1508 | */ | |
1509 | static void svc_sock_free(struct svc_xprt *xprt) | |
1510 | { | |
1511 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
1512 | dprintk("svc: svc_sock_free(%p)\n", svsk); | |
1513 | ||
755cceab TT |
1514 | if (svsk->sk_sock->file) |
1515 | sockfd_put(svsk->sk_sock); | |
1516 | else | |
1517 | sock_release(svsk->sk_sock); | |
1518 | kfree(svsk); | |
1519 | } | |
7652e5a0 BH |
1520 | |
1521 | /* | |
1522 | * Create a svc_xprt. | |
1523 | * | |
1524 | * For internal use only (e.g. nfsv4.1 backchannel). | |
1525 | * Callers should typically use the xpo_create() method. | |
1526 | */ | |
1527 | struct svc_xprt *svc_sock_create(struct svc_serv *serv, int prot) | |
1528 | { | |
1529 | struct svc_sock *svsk; | |
1530 | struct svc_xprt *xprt = NULL; | |
1531 | ||
1532 | dprintk("svc: %s\n", __func__); | |
1533 | svsk = kzalloc(sizeof(*svsk), GFP_KERNEL); | |
1534 | if (!svsk) | |
1535 | goto out; | |
1536 | ||
1537 | xprt = &svsk->sk_xprt; | |
1538 | if (prot == IPPROTO_TCP) | |
1539 | svc_xprt_init(&svc_tcp_class, xprt, serv); | |
1540 | else if (prot == IPPROTO_UDP) | |
1541 | svc_xprt_init(&svc_udp_class, xprt, serv); | |
1542 | else | |
1543 | BUG(); | |
1544 | out: | |
1545 | dprintk("svc: %s return %p\n", __func__, xprt); | |
1546 | return xprt; | |
1547 | } | |
1548 | EXPORT_SYMBOL_GPL(svc_sock_create); | |
1549 | ||
1550 | /* | |
1551 | * Destroy a svc_sock. | |
1552 | */ | |
1553 | void svc_sock_destroy(struct svc_xprt *xprt) | |
1554 | { | |
1555 | if (xprt) | |
1556 | kfree(container_of(xprt, struct svc_sock, sk_xprt)); | |
1557 | } | |
1558 | EXPORT_SYMBOL_GPL(svc_sock_destroy); |