]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/net/sunrpc/svcsock.c | |
3 | * | |
4 | * These are the RPC server socket internals. | |
5 | * | |
6 | * The server scheduling algorithm does not always distribute the load | |
7 | * evenly when servicing a single client. May need to modify the | |
f6150c3c | 8 | * svc_xprt_enqueue procedure... |
1da177e4 LT |
9 | * |
10 | * TCP support is largely untested and may be a little slow. The problem | |
11 | * is that we currently do two separate recvfrom's, one for the 4-byte | |
12 | * record length, and the second for the actual record. This could possibly | |
13 | * be improved by always reading a minimum size of around 100 bytes and | |
14 | * tucking any superfluous bytes away in a temporary store. Still, that | |
15 | * leaves write requests out in the rain. An alternative may be to peek at | |
16 | * the first skb in the queue, and if it matches the next TCP sequence | |
17 | * number, to extract the record marker. Yuck. | |
18 | * | |
19 | * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de> | |
20 | */ | |
21 | ||
172589cc | 22 | #include <linux/kernel.h> |
1da177e4 | 23 | #include <linux/sched.h> |
3a9a231d | 24 | #include <linux/module.h> |
1da177e4 LT |
25 | #include <linux/errno.h> |
26 | #include <linux/fcntl.h> | |
27 | #include <linux/net.h> | |
28 | #include <linux/in.h> | |
29 | #include <linux/inet.h> | |
30 | #include <linux/udp.h> | |
91483c4b | 31 | #include <linux/tcp.h> |
1da177e4 LT |
32 | #include <linux/unistd.h> |
33 | #include <linux/slab.h> | |
34 | #include <linux/netdevice.h> | |
35 | #include <linux/skbuff.h> | |
b41b66d6 | 36 | #include <linux/file.h> |
7dfb7103 | 37 | #include <linux/freezer.h> |
1da177e4 LT |
38 | #include <net/sock.h> |
39 | #include <net/checksum.h> | |
40 | #include <net/ip.h> | |
b92503b2 | 41 | #include <net/ipv6.h> |
b7872fe8 | 42 | #include <net/tcp.h> |
c752f073 | 43 | #include <net/tcp_states.h> |
1da177e4 LT |
44 | #include <asm/uaccess.h> |
45 | #include <asm/ioctls.h> | |
22911fc5 | 46 | #include <trace/events/skb.h> |
1da177e4 LT |
47 | |
48 | #include <linux/sunrpc/types.h> | |
ad06e4bd | 49 | #include <linux/sunrpc/clnt.h> |
1da177e4 | 50 | #include <linux/sunrpc/xdr.h> |
c0401ea0 | 51 | #include <linux/sunrpc/msg_prot.h> |
1da177e4 LT |
52 | #include <linux/sunrpc/svcsock.h> |
53 | #include <linux/sunrpc/stats.h> | |
4cfc7e60 | 54 | #include <linux/sunrpc/xprt.h> |
1da177e4 | 55 | |
177e4f99 HS |
56 | #include "sunrpc.h" |
57 | ||
360d8738 | 58 | #define RPCDBG_FACILITY RPCDBG_SVCXPRT |
1da177e4 LT |
59 | |
60 | ||
61 | static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *, | |
72c35376 | 62 | int flags); |
1da177e4 LT |
63 | static void svc_udp_data_ready(struct sock *, int); |
64 | static int svc_udp_recvfrom(struct svc_rqst *); | |
65 | static int svc_udp_sendto(struct svc_rqst *); | |
755cceab | 66 | static void svc_sock_detach(struct svc_xprt *); |
69b6ba37 | 67 | static void svc_tcp_sock_detach(struct svc_xprt *); |
755cceab | 68 | static void svc_sock_free(struct svc_xprt *); |
1da177e4 | 69 | |
b700cbb1 | 70 | static struct svc_xprt *svc_create_socket(struct svc_serv *, int, |
62832c03 PE |
71 | struct net *, struct sockaddr *, |
72 | int, int); | |
9e00abc3 | 73 | #if defined(CONFIG_SUNRPC_BACKCHANNEL) |
1f11a034 AA |
74 | static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int, |
75 | struct net *, struct sockaddr *, | |
76 | int, int); | |
77 | static void svc_bc_sock_free(struct svc_xprt *xprt); | |
9e00abc3 | 78 | #endif /* CONFIG_SUNRPC_BACKCHANNEL */ |
1f11a034 | 79 | |
ed07536e PZ |
80 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
81 | static struct lock_class_key svc_key[2]; | |
82 | static struct lock_class_key svc_slock_key[2]; | |
83 | ||
0f0257ea | 84 | static void svc_reclassify_socket(struct socket *sock) |
ed07536e PZ |
85 | { |
86 | struct sock *sk = sock->sk; | |
02b3d346 | 87 | BUG_ON(sock_owned_by_user(sk)); |
ed07536e PZ |
88 | switch (sk->sk_family) { |
89 | case AF_INET: | |
90 | sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD", | |
def13d74 TT |
91 | &svc_slock_key[0], |
92 | "sk_xprt.xpt_lock-AF_INET-NFSD", | |
93 | &svc_key[0]); | |
ed07536e PZ |
94 | break; |
95 | ||
96 | case AF_INET6: | |
97 | sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD", | |
def13d74 TT |
98 | &svc_slock_key[1], |
99 | "sk_xprt.xpt_lock-AF_INET6-NFSD", | |
100 | &svc_key[1]); | |
ed07536e PZ |
101 | break; |
102 | ||
103 | default: | |
104 | BUG(); | |
105 | } | |
106 | } | |
107 | #else | |
0f0257ea | 108 | static void svc_reclassify_socket(struct socket *sock) |
ed07536e PZ |
109 | { |
110 | } | |
111 | #endif | |
112 | ||
1da177e4 LT |
113 | /* |
114 | * Release an skbuff after use | |
115 | */ | |
5148bf4e | 116 | static void svc_release_skb(struct svc_rqst *rqstp) |
1da177e4 | 117 | { |
5148bf4e | 118 | struct sk_buff *skb = rqstp->rq_xprt_ctxt; |
1da177e4 LT |
119 | |
120 | if (skb) { | |
57b1d3ba TT |
121 | struct svc_sock *svsk = |
122 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
5148bf4e | 123 | rqstp->rq_xprt_ctxt = NULL; |
1da177e4 LT |
124 | |
125 | dprintk("svc: service %p, releasing skb %p\n", rqstp, skb); | |
9d410c79 | 126 | skb_free_datagram_locked(svsk->sk_sk, skb); |
1da177e4 | 127 | } |
1da177e4 LT |
128 | } |
129 | ||
b92503b2 CL |
130 | union svc_pktinfo_u { |
131 | struct in_pktinfo pkti; | |
b92503b2 | 132 | struct in6_pktinfo pkti6; |
b92503b2 | 133 | }; |
bc375ea7 DM |
134 | #define SVC_PKTINFO_SPACE \ |
135 | CMSG_SPACE(sizeof(union svc_pktinfo_u)) | |
b92503b2 CL |
136 | |
137 | static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh) | |
138 | { | |
57b1d3ba TT |
139 | struct svc_sock *svsk = |
140 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
141 | switch (svsk->sk_sk->sk_family) { | |
b92503b2 CL |
142 | case AF_INET: { |
143 | struct in_pktinfo *pki = CMSG_DATA(cmh); | |
144 | ||
145 | cmh->cmsg_level = SOL_IP; | |
146 | cmh->cmsg_type = IP_PKTINFO; | |
147 | pki->ipi_ifindex = 0; | |
849a1cf1 MJ |
148 | pki->ipi_spec_dst.s_addr = |
149 | svc_daddr_in(rqstp)->sin_addr.s_addr; | |
b92503b2 CL |
150 | cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); |
151 | } | |
152 | break; | |
5a05ed73 | 153 | |
b92503b2 CL |
154 | case AF_INET6: { |
155 | struct in6_pktinfo *pki = CMSG_DATA(cmh); | |
849a1cf1 | 156 | struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp); |
b92503b2 CL |
157 | |
158 | cmh->cmsg_level = SOL_IPV6; | |
159 | cmh->cmsg_type = IPV6_PKTINFO; | |
849a1cf1 | 160 | pki->ipi6_ifindex = daddr->sin6_scope_id; |
4e3fd7a0 | 161 | pki->ipi6_addr = daddr->sin6_addr; |
b92503b2 CL |
162 | cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); |
163 | } | |
164 | break; | |
b92503b2 | 165 | } |
b92503b2 CL |
166 | } |
167 | ||
1da177e4 | 168 | /* |
4cfc7e60 | 169 | * send routine intended to be shared by the fore- and back-channel |
1da177e4 | 170 | */ |
4cfc7e60 RI |
171 | int svc_send_common(struct socket *sock, struct xdr_buf *xdr, |
172 | struct page *headpage, unsigned long headoffset, | |
173 | struct page *tailpage, unsigned long tailoffset) | |
1da177e4 | 174 | { |
1da177e4 LT |
175 | int result; |
176 | int size; | |
177 | struct page **ppage = xdr->pages; | |
178 | size_t base = xdr->page_base; | |
179 | unsigned int pglen = xdr->page_len; | |
180 | unsigned int flags = MSG_MORE; | |
4cfc7e60 RI |
181 | int slen; |
182 | int len = 0; | |
1da177e4 LT |
183 | |
184 | slen = xdr->len; | |
185 | ||
1da177e4 LT |
186 | /* send head */ |
187 | if (slen == xdr->head[0].iov_len) | |
188 | flags = 0; | |
4cfc7e60 | 189 | len = kernel_sendpage(sock, headpage, headoffset, |
44524359 | 190 | xdr->head[0].iov_len, flags); |
1da177e4 LT |
191 | if (len != xdr->head[0].iov_len) |
192 | goto out; | |
193 | slen -= xdr->head[0].iov_len; | |
194 | if (slen == 0) | |
195 | goto out; | |
196 | ||
197 | /* send page data */ | |
198 | size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen; | |
199 | while (pglen > 0) { | |
200 | if (slen == size) | |
201 | flags = 0; | |
e6242e92 | 202 | result = kernel_sendpage(sock, *ppage, base, size, flags); |
1da177e4 LT |
203 | if (result > 0) |
204 | len += result; | |
205 | if (result != size) | |
206 | goto out; | |
207 | slen -= size; | |
208 | pglen -= size; | |
209 | size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen; | |
210 | base = 0; | |
211 | ppage++; | |
212 | } | |
4cfc7e60 | 213 | |
1da177e4 LT |
214 | /* send tail */ |
215 | if (xdr->tail[0].iov_len) { | |
4cfc7e60 RI |
216 | result = kernel_sendpage(sock, tailpage, tailoffset, |
217 | xdr->tail[0].iov_len, 0); | |
1da177e4 LT |
218 | if (result > 0) |
219 | len += result; | |
220 | } | |
4cfc7e60 RI |
221 | |
222 | out: | |
223 | return len; | |
224 | } | |
225 | ||
226 | ||
227 | /* | |
228 | * Generic sendto routine | |
229 | */ | |
230 | static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr) | |
231 | { | |
232 | struct svc_sock *svsk = | |
233 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
234 | struct socket *sock = svsk->sk_sock; | |
235 | union { | |
236 | struct cmsghdr hdr; | |
237 | long all[SVC_PKTINFO_SPACE / sizeof(long)]; | |
238 | } buffer; | |
239 | struct cmsghdr *cmh = &buffer.hdr; | |
240 | int len = 0; | |
241 | unsigned long tailoff; | |
242 | unsigned long headoff; | |
243 | RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); | |
244 | ||
245 | if (rqstp->rq_prot == IPPROTO_UDP) { | |
246 | struct msghdr msg = { | |
247 | .msg_name = &rqstp->rq_addr, | |
248 | .msg_namelen = rqstp->rq_addrlen, | |
249 | .msg_control = cmh, | |
250 | .msg_controllen = sizeof(buffer), | |
251 | .msg_flags = MSG_MORE, | |
252 | }; | |
253 | ||
254 | svc_set_cmsg_data(rqstp, cmh); | |
255 | ||
256 | if (sock_sendmsg(sock, &msg, 0) < 0) | |
257 | goto out; | |
258 | } | |
259 | ||
260 | tailoff = ((unsigned long)xdr->tail[0].iov_base) & (PAGE_SIZE-1); | |
261 | headoff = 0; | |
262 | len = svc_send_common(sock, xdr, rqstp->rq_respages[0], headoff, | |
263 | rqstp->rq_respages[0], tailoff); | |
264 | ||
1da177e4 | 265 | out: |
ad06e4bd | 266 | dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n", |
57b1d3ba | 267 | svsk, xdr->head[0].iov_base, xdr->head[0].iov_len, |
ad06e4bd | 268 | xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf))); |
1da177e4 LT |
269 | |
270 | return len; | |
271 | } | |
272 | ||
80212d59 N |
273 | /* |
274 | * Report socket names for nfsdfs | |
275 | */ | |
e7942b9f | 276 | static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining) |
80212d59 | 277 | { |
017cb47f CL |
278 | const struct sock *sk = svsk->sk_sk; |
279 | const char *proto_name = sk->sk_protocol == IPPROTO_UDP ? | |
280 | "udp" : "tcp"; | |
80212d59 N |
281 | int len; |
282 | ||
017cb47f | 283 | switch (sk->sk_family) { |
e7942b9f CL |
284 | case PF_INET: |
285 | len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n", | |
017cb47f | 286 | proto_name, |
c720c7e8 ED |
287 | &inet_sk(sk)->inet_rcv_saddr, |
288 | inet_sk(sk)->inet_num); | |
80212d59 | 289 | break; |
58de2f86 CL |
290 | case PF_INET6: |
291 | len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n", | |
017cb47f CL |
292 | proto_name, |
293 | &inet6_sk(sk)->rcv_saddr, | |
c720c7e8 | 294 | inet_sk(sk)->inet_num); |
80212d59 N |
295 | break; |
296 | default: | |
e7942b9f | 297 | len = snprintf(buf, remaining, "*unknown-%d*\n", |
017cb47f | 298 | sk->sk_family); |
80212d59 | 299 | } |
e7942b9f CL |
300 | |
301 | if (len >= remaining) { | |
302 | *buf = '\0'; | |
303 | return -ENAMETOOLONG; | |
80212d59 N |
304 | } |
305 | return len; | |
306 | } | |
307 | ||
8435d34d CL |
308 | /** |
309 | * svc_sock_names - construct a list of listener names in a string | |
310 | * @serv: pointer to RPC service | |
311 | * @buf: pointer to a buffer to fill in with socket names | |
312 | * @buflen: size of the buffer to be filled | |
313 | * @toclose: pointer to '\0'-terminated C string containing the name | |
314 | * of a listener to be closed | |
315 | * | |
316 | * Fills in @buf with a '\n'-separated list of names of listener | |
317 | * sockets. If @toclose is not NULL, the socket named by @toclose | |
318 | * is closed, and is not included in the output list. | |
319 | * | |
320 | * Returns positive length of the socket name string, or a negative | |
321 | * errno value on error. | |
322 | */ | |
323 | int svc_sock_names(struct svc_serv *serv, char *buf, const size_t buflen, | |
324 | const char *toclose) | |
80212d59 | 325 | { |
b41b66d6 | 326 | struct svc_sock *svsk, *closesk = NULL; |
80212d59 N |
327 | int len = 0; |
328 | ||
329 | if (!serv) | |
330 | return 0; | |
e7942b9f | 331 | |
aaf68cfb | 332 | spin_lock_bh(&serv->sv_lock); |
7a182083 | 333 | list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list) { |
e7942b9f CL |
334 | int onelen = svc_one_sock_name(svsk, buf + len, buflen - len); |
335 | if (onelen < 0) { | |
336 | len = onelen; | |
337 | break; | |
338 | } | |
3942302e | 339 | if (toclose && strcmp(toclose, buf + len) == 0) { |
b41b66d6 | 340 | closesk = svsk; |
3942302e N |
341 | svc_xprt_get(&closesk->sk_xprt); |
342 | } else | |
b41b66d6 | 343 | len += onelen; |
80212d59 | 344 | } |
aaf68cfb | 345 | spin_unlock_bh(&serv->sv_lock); |
e7942b9f | 346 | |
3942302e | 347 | if (closesk) { |
5680c446 N |
348 | /* Should unregister with portmap, but you cannot |
349 | * unregister just one protocol... | |
350 | */ | |
7a182083 | 351 | svc_close_xprt(&closesk->sk_xprt); |
3942302e N |
352 | svc_xprt_put(&closesk->sk_xprt); |
353 | } else if (toclose) | |
37a03472 | 354 | return -ENOENT; |
80212d59 N |
355 | return len; |
356 | } | |
24c3767e | 357 | EXPORT_SYMBOL_GPL(svc_sock_names); |
80212d59 | 358 | |
1da177e4 LT |
359 | /* |
360 | * Check input queue length | |
361 | */ | |
0f0257ea | 362 | static int svc_recv_available(struct svc_sock *svsk) |
1da177e4 | 363 | { |
1da177e4 LT |
364 | struct socket *sock = svsk->sk_sock; |
365 | int avail, err; | |
366 | ||
e6242e92 | 367 | err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail); |
1da177e4 LT |
368 | |
369 | return (err >= 0)? avail : err; | |
370 | } | |
371 | ||
372 | /* | |
373 | * Generic recvfrom routine. | |
374 | */ | |
0f0257ea TT |
375 | static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr, |
376 | int buflen) | |
1da177e4 | 377 | { |
57b1d3ba TT |
378 | struct svc_sock *svsk = |
379 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
1ba95105 CL |
380 | struct msghdr msg = { |
381 | .msg_flags = MSG_DONTWAIT, | |
382 | }; | |
383 | int len; | |
1da177e4 | 384 | |
260c1d12 TT |
385 | rqstp->rq_xprt_hlen = 0; |
386 | ||
1ba95105 CL |
387 | len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen, |
388 | msg.msg_flags); | |
1da177e4 | 389 | |
1da177e4 | 390 | dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n", |
1ba95105 | 391 | svsk, iov[0].iov_base, iov[0].iov_len, len); |
1da177e4 LT |
392 | return len; |
393 | } | |
394 | ||
31d68ef6 BF |
395 | static int svc_partial_recvfrom(struct svc_rqst *rqstp, |
396 | struct kvec *iov, int nr, | |
397 | int buflen, unsigned int base) | |
398 | { | |
399 | size_t save_iovlen; | |
09acfea5 | 400 | void *save_iovbase; |
31d68ef6 BF |
401 | unsigned int i; |
402 | int ret; | |
403 | ||
404 | if (base == 0) | |
405 | return svc_recvfrom(rqstp, iov, nr, buflen); | |
406 | ||
407 | for (i = 0; i < nr; i++) { | |
408 | if (iov[i].iov_len > base) | |
409 | break; | |
410 | base -= iov[i].iov_len; | |
411 | } | |
412 | save_iovlen = iov[i].iov_len; | |
413 | save_iovbase = iov[i].iov_base; | |
414 | iov[i].iov_len -= base; | |
415 | iov[i].iov_base += base; | |
416 | ret = svc_recvfrom(rqstp, &iov[i], nr - i, buflen); | |
417 | iov[i].iov_len = save_iovlen; | |
418 | iov[i].iov_base = save_iovbase; | |
419 | return ret; | |
420 | } | |
421 | ||
1da177e4 LT |
422 | /* |
423 | * Set socket snd and rcv buffer lengths | |
424 | */ | |
0f0257ea TT |
425 | static void svc_sock_setbufsize(struct socket *sock, unsigned int snd, |
426 | unsigned int rcv) | |
1da177e4 LT |
427 | { |
428 | #if 0 | |
429 | mm_segment_t oldfs; | |
430 | oldfs = get_fs(); set_fs(KERNEL_DS); | |
431 | sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF, | |
432 | (char*)&snd, sizeof(snd)); | |
433 | sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF, | |
434 | (char*)&rcv, sizeof(rcv)); | |
435 | #else | |
436 | /* sock_setsockopt limits use to sysctl_?mem_max, | |
437 | * which isn't acceptable. Until that is made conditional | |
438 | * on not having CAP_SYS_RESOURCE or similar, we go direct... | |
439 | * DaveM said I could! | |
440 | */ | |
441 | lock_sock(sock->sk); | |
442 | sock->sk->sk_sndbuf = snd * 2; | |
443 | sock->sk->sk_rcvbuf = rcv * 2; | |
47fcb03f | 444 | sock->sk->sk_write_space(sock->sk); |
1da177e4 LT |
445 | release_sock(sock->sk); |
446 | #endif | |
447 | } | |
448 | /* | |
449 | * INET callback when data has been received on the socket. | |
450 | */ | |
0f0257ea | 451 | static void svc_udp_data_ready(struct sock *sk, int count) |
1da177e4 | 452 | { |
939bb7ef | 453 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
eaefd110 | 454 | wait_queue_head_t *wq = sk_sleep(sk); |
1da177e4 | 455 | |
939bb7ef NB |
456 | if (svsk) { |
457 | dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n", | |
02fc6c36 TT |
458 | svsk, sk, count, |
459 | test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags)); | |
460 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); | |
f6150c3c | 461 | svc_xprt_enqueue(&svsk->sk_xprt); |
939bb7ef | 462 | } |
eaefd110 ED |
463 | if (wq && waitqueue_active(wq)) |
464 | wake_up_interruptible(wq); | |
1da177e4 LT |
465 | } |
466 | ||
467 | /* | |
468 | * INET callback when space is newly available on the socket. | |
469 | */ | |
0f0257ea | 470 | static void svc_write_space(struct sock *sk) |
1da177e4 LT |
471 | { |
472 | struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data); | |
eaefd110 | 473 | wait_queue_head_t *wq = sk_sleep(sk); |
1da177e4 LT |
474 | |
475 | if (svsk) { | |
476 | dprintk("svc: socket %p(inet %p), write_space busy=%d\n", | |
02fc6c36 | 477 | svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags)); |
f6150c3c | 478 | svc_xprt_enqueue(&svsk->sk_xprt); |
1da177e4 LT |
479 | } |
480 | ||
eaefd110 | 481 | if (wq && waitqueue_active(wq)) { |
939bb7ef | 482 | dprintk("RPC svc_write_space: someone sleeping on %p\n", |
1da177e4 | 483 | svsk); |
eaefd110 | 484 | wake_up_interruptible(wq); |
1da177e4 LT |
485 | } |
486 | } | |
487 | ||
47fcb03f TM |
488 | static void svc_tcp_write_space(struct sock *sk) |
489 | { | |
490 | struct socket *sock = sk->sk_socket; | |
491 | ||
492 | if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && sock) | |
493 | clear_bit(SOCK_NOSPACE, &sock->flags); | |
494 | svc_write_space(sk); | |
495 | } | |
496 | ||
7702ce40 CL |
497 | /* |
498 | * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo | |
499 | */ | |
500 | static int svc_udp_get_dest_address4(struct svc_rqst *rqstp, | |
501 | struct cmsghdr *cmh) | |
502 | { | |
503 | struct in_pktinfo *pki = CMSG_DATA(cmh); | |
849a1cf1 MJ |
504 | struct sockaddr_in *daddr = svc_daddr_in(rqstp); |
505 | ||
7702ce40 CL |
506 | if (cmh->cmsg_type != IP_PKTINFO) |
507 | return 0; | |
849a1cf1 MJ |
508 | |
509 | daddr->sin_family = AF_INET; | |
510 | daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr; | |
7702ce40 CL |
511 | return 1; |
512 | } | |
513 | ||
514 | /* | |
515 | * See net/ipv6/datagram.c : datagram_recv_ctl | |
516 | */ | |
517 | static int svc_udp_get_dest_address6(struct svc_rqst *rqstp, | |
518 | struct cmsghdr *cmh) | |
519 | { | |
520 | struct in6_pktinfo *pki = CMSG_DATA(cmh); | |
849a1cf1 MJ |
521 | struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp); |
522 | ||
7702ce40 CL |
523 | if (cmh->cmsg_type != IPV6_PKTINFO) |
524 | return 0; | |
849a1cf1 MJ |
525 | |
526 | daddr->sin6_family = AF_INET6; | |
4e3fd7a0 | 527 | daddr->sin6_addr = pki->ipi6_addr; |
849a1cf1 | 528 | daddr->sin6_scope_id = pki->ipi6_ifindex; |
7702ce40 CL |
529 | return 1; |
530 | } | |
531 | ||
9dbc240f TT |
532 | /* |
533 | * Copy the UDP datagram's destination address to the rqstp structure. | |
534 | * The 'destination' address in this case is the address to which the | |
535 | * peer sent the datagram, i.e. our local address. For multihomed | |
536 | * hosts, this can change from msg to msg. Note that only the IP | |
537 | * address changes, the port number should remain the same. | |
538 | */ | |
7702ce40 CL |
539 | static int svc_udp_get_dest_address(struct svc_rqst *rqstp, |
540 | struct cmsghdr *cmh) | |
95756482 | 541 | { |
7702ce40 CL |
542 | switch (cmh->cmsg_level) { |
543 | case SOL_IP: | |
544 | return svc_udp_get_dest_address4(rqstp, cmh); | |
545 | case SOL_IPV6: | |
546 | return svc_udp_get_dest_address6(rqstp, cmh); | |
95756482 | 547 | } |
7702ce40 CL |
548 | |
549 | return 0; | |
95756482 CL |
550 | } |
551 | ||
1da177e4 LT |
552 | /* |
553 | * Receive a datagram from a UDP socket. | |
554 | */ | |
0f0257ea | 555 | static int svc_udp_recvfrom(struct svc_rqst *rqstp) |
1da177e4 | 556 | { |
57b1d3ba TT |
557 | struct svc_sock *svsk = |
558 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
bb5cf160 | 559 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
1da177e4 | 560 | struct sk_buff *skb; |
bc375ea7 DM |
561 | union { |
562 | struct cmsghdr hdr; | |
563 | long all[SVC_PKTINFO_SPACE / sizeof(long)]; | |
564 | } buffer; | |
565 | struct cmsghdr *cmh = &buffer.hdr; | |
7a37f578 N |
566 | struct msghdr msg = { |
567 | .msg_name = svc_addr(rqstp), | |
568 | .msg_control = cmh, | |
569 | .msg_controllen = sizeof(buffer), | |
570 | .msg_flags = MSG_DONTWAIT, | |
571 | }; | |
abc5c44d CL |
572 | size_t len; |
573 | int err; | |
1da177e4 | 574 | |
02fc6c36 | 575 | if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags)) |
1da177e4 LT |
576 | /* udp sockets need large rcvbuf as all pending |
577 | * requests are still in that buffer. sndbuf must | |
578 | * also be large enough that there is enough space | |
3262c816 GB |
579 | * for one reply per thread. We count all threads |
580 | * rather than threads in a particular pool, which | |
581 | * provides an upper bound on the number of threads | |
582 | * which will access the socket. | |
1da177e4 LT |
583 | */ |
584 | svc_sock_setbufsize(svsk->sk_sock, | |
c6b0a9f8 N |
585 | (serv->sv_nrthreads+3) * serv->sv_max_mesg, |
586 | (serv->sv_nrthreads+3) * serv->sv_max_mesg); | |
1da177e4 | 587 | |
02fc6c36 | 588 | clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
05ed690e N |
589 | skb = NULL; |
590 | err = kernel_recvmsg(svsk->sk_sock, &msg, NULL, | |
591 | 0, 0, MSG_PEEK | MSG_DONTWAIT); | |
592 | if (err >= 0) | |
593 | skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err); | |
594 | ||
595 | if (skb == NULL) { | |
596 | if (err != -EAGAIN) { | |
597 | /* possibly an icmp error */ | |
598 | dprintk("svc: recvfrom returned error %d\n", -err); | |
02fc6c36 | 599 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 600 | } |
05ed690e | 601 | return -EAGAIN; |
1da177e4 | 602 | } |
9dbc240f | 603 | len = svc_addr_len(svc_addr(rqstp)); |
abc5c44d CL |
604 | if (len == 0) |
605 | return -EAFNOSUPPORT; | |
9dbc240f | 606 | rqstp->rq_addrlen = len; |
b7aa0bf7 ED |
607 | if (skb->tstamp.tv64 == 0) { |
608 | skb->tstamp = ktime_get_real(); | |
cca5172a | 609 | /* Don't enable netstamp, sunrpc doesn't |
1da177e4 LT |
610 | need that much accuracy */ |
611 | } | |
b7aa0bf7 | 612 | svsk->sk_sk->sk_stamp = skb->tstamp; |
02fc6c36 | 613 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */ |
1da177e4 | 614 | |
1da177e4 LT |
615 | len = skb->len - sizeof(struct udphdr); |
616 | rqstp->rq_arg.len = len; | |
617 | ||
95756482 | 618 | rqstp->rq_prot = IPPROTO_UDP; |
27459f09 | 619 | |
7702ce40 | 620 | if (!svc_udp_get_dest_address(rqstp, cmh)) { |
e87cc472 JP |
621 | net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n", |
622 | cmh->cmsg_level, cmh->cmsg_type); | |
22911fc5 ED |
623 | out_free: |
624 | trace_kfree_skb(skb, svc_udp_recvfrom); | |
9d410c79 | 625 | skb_free_datagram_locked(svsk->sk_sk, skb); |
7a37f578 N |
626 | return 0; |
627 | } | |
849a1cf1 | 628 | rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp)); |
1da177e4 LT |
629 | |
630 | if (skb_is_nonlinear(skb)) { | |
631 | /* we have to copy */ | |
632 | local_bh_disable(); | |
633 | if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) { | |
634 | local_bh_enable(); | |
635 | /* checksum error */ | |
22911fc5 | 636 | goto out_free; |
1da177e4 LT |
637 | } |
638 | local_bh_enable(); | |
9d410c79 | 639 | skb_free_datagram_locked(svsk->sk_sk, skb); |
1da177e4 LT |
640 | } else { |
641 | /* we can use it in-place */ | |
0f0257ea TT |
642 | rqstp->rq_arg.head[0].iov_base = skb->data + |
643 | sizeof(struct udphdr); | |
1da177e4 | 644 | rqstp->rq_arg.head[0].iov_len = len; |
22911fc5 ED |
645 | if (skb_checksum_complete(skb)) |
646 | goto out_free; | |
5148bf4e | 647 | rqstp->rq_xprt_ctxt = skb; |
1da177e4 LT |
648 | } |
649 | ||
650 | rqstp->rq_arg.page_base = 0; | |
651 | if (len <= rqstp->rq_arg.head[0].iov_len) { | |
652 | rqstp->rq_arg.head[0].iov_len = len; | |
653 | rqstp->rq_arg.page_len = 0; | |
44524359 | 654 | rqstp->rq_respages = rqstp->rq_pages+1; |
1da177e4 LT |
655 | } else { |
656 | rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len; | |
44524359 | 657 | rqstp->rq_respages = rqstp->rq_pages + 1 + |
172589cc | 658 | DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE); |
1da177e4 LT |
659 | } |
660 | ||
661 | if (serv->sv_stats) | |
662 | serv->sv_stats->netudpcnt++; | |
663 | ||
664 | return len; | |
665 | } | |
666 | ||
667 | static int | |
668 | svc_udp_sendto(struct svc_rqst *rqstp) | |
669 | { | |
670 | int error; | |
671 | ||
672 | error = svc_sendto(rqstp, &rqstp->rq_res); | |
673 | if (error == -ECONNREFUSED) | |
674 | /* ICMP error on earlier request. */ | |
675 | error = svc_sendto(rqstp, &rqstp->rq_res); | |
676 | ||
677 | return error; | |
678 | } | |
679 | ||
e831fe65 TT |
680 | static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp) |
681 | { | |
682 | } | |
683 | ||
323bee32 TT |
684 | static int svc_udp_has_wspace(struct svc_xprt *xprt) |
685 | { | |
686 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
bb5cf160 | 687 | struct svc_serv *serv = xprt->xpt_server; |
323bee32 TT |
688 | unsigned long required; |
689 | ||
690 | /* | |
691 | * Set the SOCK_NOSPACE flag before checking the available | |
692 | * sock space. | |
693 | */ | |
694 | set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
7a90e8cc | 695 | required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg; |
323bee32 TT |
696 | if (required*2 > sock_wspace(svsk->sk_sk)) |
697 | return 0; | |
698 | clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
699 | return 1; | |
700 | } | |
701 | ||
38a417cc TT |
702 | static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt) |
703 | { | |
704 | BUG(); | |
705 | return NULL; | |
706 | } | |
707 | ||
b700cbb1 | 708 | static struct svc_xprt *svc_udp_create(struct svc_serv *serv, |
62832c03 | 709 | struct net *net, |
b700cbb1 TT |
710 | struct sockaddr *sa, int salen, |
711 | int flags) | |
712 | { | |
62832c03 | 713 | return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags); |
b700cbb1 TT |
714 | } |
715 | ||
360d8738 | 716 | static struct svc_xprt_ops svc_udp_ops = { |
b700cbb1 | 717 | .xpo_create = svc_udp_create, |
5d137990 TT |
718 | .xpo_recvfrom = svc_udp_recvfrom, |
719 | .xpo_sendto = svc_udp_sendto, | |
5148bf4e | 720 | .xpo_release_rqst = svc_release_skb, |
755cceab TT |
721 | .xpo_detach = svc_sock_detach, |
722 | .xpo_free = svc_sock_free, | |
e831fe65 | 723 | .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr, |
323bee32 | 724 | .xpo_has_wspace = svc_udp_has_wspace, |
38a417cc | 725 | .xpo_accept = svc_udp_accept, |
360d8738 TT |
726 | }; |
727 | ||
728 | static struct svc_xprt_class svc_udp_class = { | |
729 | .xcl_name = "udp", | |
b700cbb1 | 730 | .xcl_owner = THIS_MODULE, |
360d8738 | 731 | .xcl_ops = &svc_udp_ops, |
49023155 | 732 | .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP, |
360d8738 TT |
733 | }; |
734 | ||
bb5cf160 | 735 | static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv) |
1da177e4 | 736 | { |
7702ce40 | 737 | int err, level, optname, one = 1; |
7a37f578 | 738 | |
bd4620dd SK |
739 | svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class, |
740 | &svsk->sk_xprt, serv); | |
def13d74 | 741 | clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
742 | svsk->sk_sk->sk_data_ready = svc_udp_data_ready; |
743 | svsk->sk_sk->sk_write_space = svc_write_space; | |
1da177e4 LT |
744 | |
745 | /* initialise setting must have enough space to | |
cca5172a | 746 | * receive and respond to one request. |
1da177e4 LT |
747 | * svc_udp_recvfrom will re-adjust if necessary |
748 | */ | |
749 | svc_sock_setbufsize(svsk->sk_sock, | |
bb5cf160 TT |
750 | 3 * svsk->sk_xprt.xpt_server->sv_max_mesg, |
751 | 3 * svsk->sk_xprt.xpt_server->sv_max_mesg); | |
1da177e4 | 752 | |
0f0257ea TT |
753 | /* data might have come in before data_ready set up */ |
754 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); | |
02fc6c36 | 755 | set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); |
7a37f578 | 756 | |
7a37f578 | 757 | /* make sure we get destination address info */ |
7702ce40 CL |
758 | switch (svsk->sk_sk->sk_family) { |
759 | case AF_INET: | |
760 | level = SOL_IP; | |
761 | optname = IP_PKTINFO; | |
762 | break; | |
763 | case AF_INET6: | |
764 | level = SOL_IPV6; | |
765 | optname = IPV6_RECVPKTINFO; | |
766 | break; | |
767 | default: | |
768 | BUG(); | |
769 | } | |
770 | err = kernel_setsockopt(svsk->sk_sock, level, optname, | |
771 | (char *)&one, sizeof(one)); | |
772 | dprintk("svc: kernel_setsockopt returned %d\n", err); | |
1da177e4 LT |
773 | } |
774 | ||
775 | /* | |
776 | * A data_ready event on a listening socket means there's a connection | |
777 | * pending. Do not use state_change as a substitute for it. | |
778 | */ | |
0f0257ea | 779 | static void svc_tcp_listen_data_ready(struct sock *sk, int count_unused) |
1da177e4 | 780 | { |
939bb7ef | 781 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
eaefd110 | 782 | wait_queue_head_t *wq; |
1da177e4 LT |
783 | |
784 | dprintk("svc: socket %p TCP (listen) state change %d\n", | |
939bb7ef | 785 | sk, sk->sk_state); |
1da177e4 | 786 | |
939bb7ef NB |
787 | /* |
788 | * This callback may called twice when a new connection | |
789 | * is established as a child socket inherits everything | |
790 | * from a parent LISTEN socket. | |
791 | * 1) data_ready method of the parent socket will be called | |
792 | * when one of child sockets become ESTABLISHED. | |
793 | * 2) data_ready method of the child socket may be called | |
794 | * when it receives data before the socket is accepted. | |
795 | * In case of 2, we should ignore it silently. | |
796 | */ | |
797 | if (sk->sk_state == TCP_LISTEN) { | |
798 | if (svsk) { | |
02fc6c36 | 799 | set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
f6150c3c | 800 | svc_xprt_enqueue(&svsk->sk_xprt); |
939bb7ef NB |
801 | } else |
802 | printk("svc: socket %p: no user data\n", sk); | |
1da177e4 | 803 | } |
939bb7ef | 804 | |
eaefd110 ED |
805 | wq = sk_sleep(sk); |
806 | if (wq && waitqueue_active(wq)) | |
807 | wake_up_interruptible_all(wq); | |
1da177e4 LT |
808 | } |
809 | ||
810 | /* | |
811 | * A state change on a connected socket means it's dying or dead. | |
812 | */ | |
0f0257ea | 813 | static void svc_tcp_state_change(struct sock *sk) |
1da177e4 | 814 | { |
939bb7ef | 815 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
eaefd110 | 816 | wait_queue_head_t *wq = sk_sleep(sk); |
1da177e4 LT |
817 | |
818 | dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n", | |
939bb7ef | 819 | sk, sk->sk_state, sk->sk_user_data); |
1da177e4 | 820 | |
939bb7ef | 821 | if (!svsk) |
1da177e4 | 822 | printk("svc: socket %p: no user data\n", sk); |
939bb7ef | 823 | else { |
02fc6c36 | 824 | set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); |
f6150c3c | 825 | svc_xprt_enqueue(&svsk->sk_xprt); |
1da177e4 | 826 | } |
eaefd110 ED |
827 | if (wq && waitqueue_active(wq)) |
828 | wake_up_interruptible_all(wq); | |
1da177e4 LT |
829 | } |
830 | ||
0f0257ea | 831 | static void svc_tcp_data_ready(struct sock *sk, int count) |
1da177e4 | 832 | { |
939bb7ef | 833 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
eaefd110 | 834 | wait_queue_head_t *wq = sk_sleep(sk); |
1da177e4 LT |
835 | |
836 | dprintk("svc: socket %p TCP data ready (svsk %p)\n", | |
939bb7ef NB |
837 | sk, sk->sk_user_data); |
838 | if (svsk) { | |
02fc6c36 | 839 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
f6150c3c | 840 | svc_xprt_enqueue(&svsk->sk_xprt); |
939bb7ef | 841 | } |
eaefd110 ED |
842 | if (wq && waitqueue_active(wq)) |
843 | wake_up_interruptible(wq); | |
1da177e4 LT |
844 | } |
845 | ||
846 | /* | |
847 | * Accept a TCP connection | |
848 | */ | |
38a417cc | 849 | static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt) |
1da177e4 | 850 | { |
38a417cc | 851 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); |
cdd88b9f AM |
852 | struct sockaddr_storage addr; |
853 | struct sockaddr *sin = (struct sockaddr *) &addr; | |
bb5cf160 | 854 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
1da177e4 LT |
855 | struct socket *sock = svsk->sk_sock; |
856 | struct socket *newsock; | |
1da177e4 LT |
857 | struct svc_sock *newsvsk; |
858 | int err, slen; | |
5216a8e7 | 859 | RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); |
1da177e4 LT |
860 | |
861 | dprintk("svc: tcp_accept %p sock %p\n", svsk, sock); | |
862 | if (!sock) | |
38a417cc | 863 | return NULL; |
1da177e4 | 864 | |
02fc6c36 | 865 | clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
e6242e92 SS |
866 | err = kernel_accept(sock, &newsock, O_NONBLOCK); |
867 | if (err < 0) { | |
1da177e4 LT |
868 | if (err == -ENOMEM) |
869 | printk(KERN_WARNING "%s: no more sockets!\n", | |
870 | serv->sv_name); | |
e87cc472 JP |
871 | else if (err != -EAGAIN) |
872 | net_warn_ratelimited("%s: accept failed (err %d)!\n", | |
873 | serv->sv_name, -err); | |
38a417cc | 874 | return NULL; |
1da177e4 | 875 | } |
02fc6c36 | 876 | set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 877 | |
cdd88b9f | 878 | err = kernel_getpeername(newsock, sin, &slen); |
1da177e4 | 879 | if (err < 0) { |
e87cc472 JP |
880 | net_warn_ratelimited("%s: peername failed (err %d)!\n", |
881 | serv->sv_name, -err); | |
1da177e4 LT |
882 | goto failed; /* aborted connection or whatever */ |
883 | } | |
884 | ||
885 | /* Ideally, we would want to reject connections from unauthorized | |
ad06e4bd CL |
886 | * hosts here, but when we get encryption, the IP of the host won't |
887 | * tell us anything. For now just warn about unpriv connections. | |
1da177e4 | 888 | */ |
cdd88b9f | 889 | if (!svc_port_is_privileged(sin)) { |
1da177e4 | 890 | dprintk(KERN_WARNING |
ad06e4bd | 891 | "%s: connect from unprivileged port: %s\n", |
cca5172a | 892 | serv->sv_name, |
cdd88b9f | 893 | __svc_print_addr(sin, buf, sizeof(buf))); |
1da177e4 | 894 | } |
ad06e4bd | 895 | dprintk("%s: connect from %s\n", serv->sv_name, |
cdd88b9f | 896 | __svc_print_addr(sin, buf, sizeof(buf))); |
1da177e4 LT |
897 | |
898 | /* make sure that a write doesn't block forever when | |
899 | * low on memory | |
900 | */ | |
901 | newsock->sk->sk_sndtimeo = HZ*30; | |
902 | ||
72c35376 BF |
903 | newsvsk = svc_setup_socket(serv, newsock, |
904 | (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY)); | |
905 | if (IS_ERR(newsvsk)) | |
1da177e4 | 906 | goto failed; |
9dbc240f | 907 | svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen); |
a9747692 FM |
908 | err = kernel_getsockname(newsock, sin, &slen); |
909 | if (unlikely(err < 0)) { | |
910 | dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err); | |
911 | slen = offsetof(struct sockaddr, sa_data); | |
912 | } | |
9dbc240f | 913 | svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen); |
067d7817 | 914 | |
f9f3cc4f TT |
915 | if (serv->sv_stats) |
916 | serv->sv_stats->nettcpconn++; | |
917 | ||
918 | return &newsvsk->sk_xprt; | |
919 | ||
920 | failed: | |
921 | sock_release(newsock); | |
922 | return NULL; | |
923 | } | |
924 | ||
31d68ef6 BF |
925 | static unsigned int svc_tcp_restore_pages(struct svc_sock *svsk, struct svc_rqst *rqstp) |
926 | { | |
927 | unsigned int i, len, npages; | |
928 | ||
929 | if (svsk->sk_tcplen <= sizeof(rpc_fraghdr)) | |
930 | return 0; | |
931 | len = svsk->sk_tcplen - sizeof(rpc_fraghdr); | |
932 | npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
933 | for (i = 0; i < npages; i++) { | |
934 | if (rqstp->rq_pages[i] != NULL) | |
935 | put_page(rqstp->rq_pages[i]); | |
936 | BUG_ON(svsk->sk_pages[i] == NULL); | |
937 | rqstp->rq_pages[i] = svsk->sk_pages[i]; | |
938 | svsk->sk_pages[i] = NULL; | |
939 | } | |
940 | rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]); | |
941 | return len; | |
942 | } | |
943 | ||
944 | static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp) | |
945 | { | |
946 | unsigned int i, len, npages; | |
947 | ||
948 | if (svsk->sk_tcplen <= sizeof(rpc_fraghdr)) | |
949 | return; | |
950 | len = svsk->sk_tcplen - sizeof(rpc_fraghdr); | |
951 | npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
952 | for (i = 0; i < npages; i++) { | |
953 | svsk->sk_pages[i] = rqstp->rq_pages[i]; | |
954 | rqstp->rq_pages[i] = NULL; | |
955 | } | |
956 | } | |
957 | ||
958 | static void svc_tcp_clear_pages(struct svc_sock *svsk) | |
959 | { | |
960 | unsigned int i, len, npages; | |
961 | ||
962 | if (svsk->sk_tcplen <= sizeof(rpc_fraghdr)) | |
963 | goto out; | |
964 | len = svsk->sk_tcplen - sizeof(rpc_fraghdr); | |
965 | npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
966 | for (i = 0; i < npages; i++) { | |
967 | BUG_ON(svsk->sk_pages[i] == NULL); | |
968 | put_page(svsk->sk_pages[i]); | |
969 | svsk->sk_pages[i] = NULL; | |
970 | } | |
971 | out: | |
972 | svsk->sk_tcplen = 0; | |
973 | } | |
974 | ||
1da177e4 | 975 | /* |
8f55f3c0 AB |
976 | * Receive data. |
977 | * If we haven't gotten the record length yet, get the next four bytes. | |
978 | * Otherwise try to gobble up as much as possible up to the complete | |
979 | * record length. | |
1da177e4 | 980 | */ |
8f55f3c0 | 981 | static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp) |
1da177e4 | 982 | { |
bb5cf160 | 983 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
5ee78d48 | 984 | unsigned int want; |
8f55f3c0 | 985 | int len; |
1da177e4 | 986 | |
02fc6c36 | 987 | clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 988 | |
c0401ea0 | 989 | if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) { |
1da177e4 LT |
990 | struct kvec iov; |
991 | ||
5ee78d48 | 992 | want = sizeof(rpc_fraghdr) - svsk->sk_tcplen; |
1da177e4 LT |
993 | iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen; |
994 | iov.iov_len = want; | |
995 | if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0) | |
996 | goto error; | |
997 | svsk->sk_tcplen += len; | |
998 | ||
999 | if (len < want) { | |
c0401ea0 CL |
1000 | dprintk("svc: short recvfrom while reading record " |
1001 | "length (%d of %d)\n", len, want); | |
31d68ef6 | 1002 | return -EAGAIN; |
1da177e4 LT |
1003 | } |
1004 | ||
1005 | svsk->sk_reclen = ntohl(svsk->sk_reclen); | |
c0401ea0 | 1006 | if (!(svsk->sk_reclen & RPC_LAST_STREAM_FRAGMENT)) { |
1da177e4 LT |
1007 | /* FIXME: technically, a record can be fragmented, |
1008 | * and non-terminal fragments will not have the top | |
1009 | * bit set in the fragment length header. | |
1010 | * But apparently no known nfs clients send fragmented | |
1011 | * records. */ | |
e87cc472 | 1012 | net_notice_ratelimited("RPC: multiple fragments per record not supported\n"); |
1da177e4 LT |
1013 | goto err_delete; |
1014 | } | |
8f55f3c0 | 1015 | |
c0401ea0 | 1016 | svsk->sk_reclen &= RPC_FRAGMENT_SIZE_MASK; |
1da177e4 | 1017 | dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen); |
c6b0a9f8 | 1018 | if (svsk->sk_reclen > serv->sv_max_mesg) { |
e87cc472 JP |
1019 | net_notice_ratelimited("RPC: fragment too large: 0x%08lx\n", |
1020 | (unsigned long)svsk->sk_reclen); | |
1da177e4 LT |
1021 | goto err_delete; |
1022 | } | |
1023 | } | |
1024 | ||
cc6c2127 BF |
1025 | if (svsk->sk_reclen < 8) |
1026 | goto err_delete; /* client is nuts. */ | |
1027 | ||
1da177e4 | 1028 | len = svsk->sk_reclen; |
1da177e4 | 1029 | |
8f55f3c0 | 1030 | return len; |
31d68ef6 BF |
1031 | error: |
1032 | dprintk("RPC: TCP recv_record got %d\n", len); | |
8f55f3c0 | 1033 | return len; |
31d68ef6 | 1034 | err_delete: |
8f55f3c0 | 1035 | set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); |
8f55f3c0 AB |
1036 | return -EAGAIN; |
1037 | } | |
1038 | ||
586c52cc | 1039 | static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp) |
4cfc7e60 | 1040 | { |
586c52cc | 1041 | struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt; |
4cfc7e60 | 1042 | struct rpc_rqst *req = NULL; |
586c52cc TM |
1043 | struct kvec *src, *dst; |
1044 | __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base; | |
48e6555c BF |
1045 | __be32 xid; |
1046 | __be32 calldir; | |
4cfc7e60 | 1047 | |
4cfc7e60 RI |
1048 | xid = *p++; |
1049 | calldir = *p; | |
1050 | ||
586c52cc TM |
1051 | if (bc_xprt) |
1052 | req = xprt_lookup_rqst(bc_xprt, xid); | |
4cfc7e60 | 1053 | |
586c52cc TM |
1054 | if (!req) { |
1055 | printk(KERN_NOTICE | |
1056 | "%s: Got unrecognized reply: " | |
1057 | "calldir 0x%x xpt_bc_xprt %p xid %08x\n", | |
1058 | __func__, ntohl(calldir), | |
1059 | bc_xprt, xid); | |
1060 | return -EAGAIN; | |
4cfc7e60 | 1061 | } |
586c52cc TM |
1062 | |
1063 | memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf)); | |
1064 | /* | |
1065 | * XXX!: cheating for now! Only copying HEAD. | |
1066 | * But we know this is good enough for now (in fact, for any | |
1067 | * callback reply in the forseeable future). | |
1068 | */ | |
1069 | dst = &req->rq_private_buf.head[0]; | |
1070 | src = &rqstp->rq_arg.head[0]; | |
1071 | if (dst->iov_len < src->iov_len) | |
1072 | return -EAGAIN; /* whatever; just giving up. */ | |
1073 | memcpy(dst->iov_base, src->iov_base, src->iov_len); | |
1074 | xprt_complete_rqst(req->rq_task, svsk->sk_reclen); | |
1075 | rqstp->rq_arg.len = 0; | |
1076 | return 0; | |
1077 | } | |
1078 | ||
1079 | static int copy_pages_to_kvecs(struct kvec *vec, struct page **pages, int len) | |
1080 | { | |
1081 | int i = 0; | |
1082 | int t = 0; | |
1083 | ||
1084 | while (t < len) { | |
1085 | vec[i].iov_base = page_address(pages[i]); | |
1086 | vec[i].iov_len = PAGE_SIZE; | |
1087 | i++; | |
1088 | t += PAGE_SIZE; | |
1089 | } | |
1090 | return i; | |
4cfc7e60 RI |
1091 | } |
1092 | ||
31d68ef6 | 1093 | |
8f55f3c0 AB |
1094 | /* |
1095 | * Receive data from a TCP socket. | |
1096 | */ | |
1097 | static int svc_tcp_recvfrom(struct svc_rqst *rqstp) | |
1098 | { | |
1099 | struct svc_sock *svsk = | |
1100 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
1101 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; | |
1102 | int len; | |
1103 | struct kvec *vec; | |
31d68ef6 | 1104 | unsigned int want, base; |
586c52cc TM |
1105 | __be32 *p; |
1106 | __be32 calldir; | |
5ee78d48 | 1107 | int pnum; |
8f55f3c0 AB |
1108 | |
1109 | dprintk("svc: tcp_recv %p data %d conn %d close %d\n", | |
1110 | svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags), | |
1111 | test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags), | |
1112 | test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags)); | |
1113 | ||
1114 | len = svc_tcp_recv_record(svsk, rqstp); | |
1115 | if (len < 0) | |
1116 | goto error; | |
1117 | ||
31d68ef6 BF |
1118 | base = svc_tcp_restore_pages(svsk, rqstp); |
1119 | want = svsk->sk_reclen - base; | |
1120 | ||
3cc03b16 | 1121 | vec = rqstp->rq_vec; |
4cfc7e60 | 1122 | |
586c52cc TM |
1123 | pnum = copy_pages_to_kvecs(&vec[0], &rqstp->rq_pages[0], |
1124 | svsk->sk_reclen); | |
1125 | ||
44524359 | 1126 | rqstp->rq_respages = &rqstp->rq_pages[pnum]; |
1da177e4 LT |
1127 | |
1128 | /* Now receive data */ | |
31d68ef6 BF |
1129 | len = svc_partial_recvfrom(rqstp, vec, pnum, want, base); |
1130 | if (len >= 0) | |
1131 | svsk->sk_tcplen += len; | |
1132 | if (len != want) { | |
be1e4444 | 1133 | svc_tcp_save_pages(svsk, rqstp); |
31d68ef6 BF |
1134 | if (len < 0 && len != -EAGAIN) |
1135 | goto err_other; | |
31d68ef6 BF |
1136 | dprintk("svc: incomplete TCP record (%d of %d)\n", |
1137 | svsk->sk_tcplen, svsk->sk_reclen); | |
1138 | goto err_noclose; | |
1139 | } | |
1da177e4 | 1140 | |
5ee78d48 | 1141 | rqstp->rq_arg.len = svsk->sk_reclen; |
1da177e4 | 1142 | rqstp->rq_arg.page_base = 0; |
5ee78d48 TM |
1143 | if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) { |
1144 | rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len; | |
1da177e4 | 1145 | rqstp->rq_arg.page_len = 0; |
5ee78d48 TM |
1146 | } else |
1147 | rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len; | |
1da177e4 | 1148 | |
5148bf4e | 1149 | rqstp->rq_xprt_ctxt = NULL; |
1da177e4 LT |
1150 | rqstp->rq_prot = IPPROTO_TCP; |
1151 | ||
586c52cc TM |
1152 | p = (__be32 *)rqstp->rq_arg.head[0].iov_base; |
1153 | calldir = p[1]; | |
8985ef0b | 1154 | if (calldir) |
586c52cc | 1155 | len = receive_cb_reply(svsk, rqstp); |
586c52cc | 1156 | |
1da177e4 LT |
1157 | /* Reset TCP read info */ |
1158 | svsk->sk_reclen = 0; | |
1159 | svsk->sk_tcplen = 0; | |
0601f793 TM |
1160 | /* If we have more data, signal svc_xprt_enqueue() to try again */ |
1161 | if (svc_recv_available(svsk) > sizeof(rpc_fraghdr)) | |
1162 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); | |
1163 | ||
8985ef0b BF |
1164 | if (len < 0) |
1165 | goto error; | |
1da177e4 | 1166 | |
9dbc240f | 1167 | svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt); |
1da177e4 LT |
1168 | if (serv->sv_stats) |
1169 | serv->sv_stats->nettcpcnt++; | |
1170 | ||
31d68ef6 | 1171 | dprintk("svc: TCP complete record (%d bytes)\n", rqstp->rq_arg.len); |
5ee78d48 | 1172 | return rqstp->rq_arg.len; |
1da177e4 | 1173 | |
8f55f3c0 | 1174 | error: |
31d68ef6 BF |
1175 | if (len != -EAGAIN) |
1176 | goto err_other; | |
1177 | dprintk("RPC: TCP recvfrom got EAGAIN\n"); | |
8f55f3c0 | 1178 | return -EAGAIN; |
31d68ef6 BF |
1179 | err_other: |
1180 | printk(KERN_NOTICE "%s: recvfrom returned errno %d\n", | |
1181 | svsk->sk_xprt.xpt_server->sv_name, -len); | |
1182 | set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); | |
1183 | err_noclose: | |
1184 | return -EAGAIN; /* record not complete */ | |
1da177e4 LT |
1185 | } |
1186 | ||
1187 | /* | |
1188 | * Send out data on TCP socket. | |
1189 | */ | |
0f0257ea | 1190 | static int svc_tcp_sendto(struct svc_rqst *rqstp) |
1da177e4 LT |
1191 | { |
1192 | struct xdr_buf *xbufp = &rqstp->rq_res; | |
1193 | int sent; | |
d8ed029d | 1194 | __be32 reclen; |
1da177e4 LT |
1195 | |
1196 | /* Set up the first element of the reply kvec. | |
1197 | * Any other kvecs that may be in use have been taken | |
1198 | * care of by the server implementation itself. | |
1199 | */ | |
1200 | reclen = htonl(0x80000000|((xbufp->len ) - 4)); | |
1201 | memcpy(xbufp->head[0].iov_base, &reclen, 4); | |
1202 | ||
1da177e4 LT |
1203 | sent = svc_sendto(rqstp, &rqstp->rq_res); |
1204 | if (sent != xbufp->len) { | |
0f0257ea TT |
1205 | printk(KERN_NOTICE |
1206 | "rpc-srv/tcp: %s: %s %d when sending %d bytes " | |
1207 | "- shutting down socket\n", | |
57b1d3ba | 1208 | rqstp->rq_xprt->xpt_server->sv_name, |
1da177e4 LT |
1209 | (sent<0)?"got error":"sent only", |
1210 | sent, xbufp->len); | |
57b1d3ba | 1211 | set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags); |
f6150c3c | 1212 | svc_xprt_enqueue(rqstp->rq_xprt); |
1da177e4 LT |
1213 | sent = -EAGAIN; |
1214 | } | |
1215 | return sent; | |
1216 | } | |
1217 | ||
e831fe65 TT |
1218 | /* |
1219 | * Setup response header. TCP has a 4B record length field. | |
1220 | */ | |
1221 | static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp) | |
1222 | { | |
1223 | struct kvec *resv = &rqstp->rq_res.head[0]; | |
1224 | ||
1225 | /* tcp needs a space for the record length... */ | |
1226 | svc_putnl(resv, 0); | |
1227 | } | |
1228 | ||
323bee32 TT |
1229 | static int svc_tcp_has_wspace(struct svc_xprt *xprt) |
1230 | { | |
57b1d3ba | 1231 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); |
47fcb03f | 1232 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
323bee32 | 1233 | int required; |
323bee32 | 1234 | |
47fcb03f TM |
1235 | if (test_bit(XPT_LISTENER, &xprt->xpt_flags)) |
1236 | return 1; | |
1237 | required = atomic_read(&xprt->xpt_reserved) + serv->sv_max_mesg; | |
1238 | if (sk_stream_wspace(svsk->sk_sk) >= required) | |
1239 | return 1; | |
323bee32 | 1240 | set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); |
47fcb03f | 1241 | return 0; |
323bee32 TT |
1242 | } |
1243 | ||
b700cbb1 | 1244 | static struct svc_xprt *svc_tcp_create(struct svc_serv *serv, |
62832c03 | 1245 | struct net *net, |
b700cbb1 TT |
1246 | struct sockaddr *sa, int salen, |
1247 | int flags) | |
1248 | { | |
62832c03 | 1249 | return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags); |
b700cbb1 TT |
1250 | } |
1251 | ||
9e00abc3 | 1252 | #if defined(CONFIG_SUNRPC_BACKCHANNEL) |
1f11a034 AA |
1253 | static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int, |
1254 | struct net *, struct sockaddr *, | |
1255 | int, int); | |
1256 | static void svc_bc_sock_free(struct svc_xprt *xprt); | |
1257 | ||
1258 | static struct svc_xprt *svc_bc_tcp_create(struct svc_serv *serv, | |
1259 | struct net *net, | |
1260 | struct sockaddr *sa, int salen, | |
1261 | int flags) | |
1262 | { | |
1263 | return svc_bc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags); | |
1264 | } | |
1265 | ||
1266 | static void svc_bc_tcp_sock_detach(struct svc_xprt *xprt) | |
1267 | { | |
1268 | } | |
1269 | ||
1270 | static struct svc_xprt_ops svc_tcp_bc_ops = { | |
1271 | .xpo_create = svc_bc_tcp_create, | |
1272 | .xpo_detach = svc_bc_tcp_sock_detach, | |
1273 | .xpo_free = svc_bc_sock_free, | |
1274 | .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr, | |
1275 | }; | |
1276 | ||
1277 | static struct svc_xprt_class svc_tcp_bc_class = { | |
1278 | .xcl_name = "tcp-bc", | |
1279 | .xcl_owner = THIS_MODULE, | |
1280 | .xcl_ops = &svc_tcp_bc_ops, | |
1281 | .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP, | |
1282 | }; | |
16b2d1e1 AA |
1283 | |
1284 | static void svc_init_bc_xprt_sock(void) | |
1285 | { | |
1286 | svc_reg_xprt_class(&svc_tcp_bc_class); | |
1287 | } | |
1288 | ||
1289 | static void svc_cleanup_bc_xprt_sock(void) | |
1290 | { | |
1291 | svc_unreg_xprt_class(&svc_tcp_bc_class); | |
1292 | } | |
9e00abc3 | 1293 | #else /* CONFIG_SUNRPC_BACKCHANNEL */ |
16b2d1e1 AA |
1294 | static void svc_init_bc_xprt_sock(void) |
1295 | { | |
1296 | } | |
1297 | ||
1298 | static void svc_cleanup_bc_xprt_sock(void) | |
1299 | { | |
1300 | } | |
9e00abc3 | 1301 | #endif /* CONFIG_SUNRPC_BACKCHANNEL */ |
1f11a034 | 1302 | |
360d8738 | 1303 | static struct svc_xprt_ops svc_tcp_ops = { |
b700cbb1 | 1304 | .xpo_create = svc_tcp_create, |
5d137990 TT |
1305 | .xpo_recvfrom = svc_tcp_recvfrom, |
1306 | .xpo_sendto = svc_tcp_sendto, | |
5148bf4e | 1307 | .xpo_release_rqst = svc_release_skb, |
69b6ba37 | 1308 | .xpo_detach = svc_tcp_sock_detach, |
755cceab | 1309 | .xpo_free = svc_sock_free, |
e831fe65 | 1310 | .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr, |
323bee32 | 1311 | .xpo_has_wspace = svc_tcp_has_wspace, |
38a417cc | 1312 | .xpo_accept = svc_tcp_accept, |
360d8738 TT |
1313 | }; |
1314 | ||
1315 | static struct svc_xprt_class svc_tcp_class = { | |
1316 | .xcl_name = "tcp", | |
b700cbb1 | 1317 | .xcl_owner = THIS_MODULE, |
360d8738 | 1318 | .xcl_ops = &svc_tcp_ops, |
49023155 | 1319 | .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP, |
360d8738 TT |
1320 | }; |
1321 | ||
1322 | void svc_init_xprt_sock(void) | |
1323 | { | |
1324 | svc_reg_xprt_class(&svc_tcp_class); | |
1325 | svc_reg_xprt_class(&svc_udp_class); | |
16b2d1e1 | 1326 | svc_init_bc_xprt_sock(); |
360d8738 TT |
1327 | } |
1328 | ||
1329 | void svc_cleanup_xprt_sock(void) | |
1330 | { | |
1331 | svc_unreg_xprt_class(&svc_tcp_class); | |
1332 | svc_unreg_xprt_class(&svc_udp_class); | |
16b2d1e1 | 1333 | svc_cleanup_bc_xprt_sock(); |
360d8738 TT |
1334 | } |
1335 | ||
bb5cf160 | 1336 | static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv) |
1da177e4 LT |
1337 | { |
1338 | struct sock *sk = svsk->sk_sk; | |
1da177e4 | 1339 | |
bd4620dd SK |
1340 | svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class, |
1341 | &svsk->sk_xprt, serv); | |
def13d74 | 1342 | set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
1343 | if (sk->sk_state == TCP_LISTEN) { |
1344 | dprintk("setting up TCP socket for listening\n"); | |
02fc6c36 | 1345 | set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 1346 | sk->sk_data_ready = svc_tcp_listen_data_ready; |
02fc6c36 | 1347 | set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
1348 | } else { |
1349 | dprintk("setting up TCP socket for reading\n"); | |
1350 | sk->sk_state_change = svc_tcp_state_change; | |
1351 | sk->sk_data_ready = svc_tcp_data_ready; | |
47fcb03f | 1352 | sk->sk_write_space = svc_tcp_write_space; |
1da177e4 LT |
1353 | |
1354 | svsk->sk_reclen = 0; | |
1355 | svsk->sk_tcplen = 0; | |
31d68ef6 | 1356 | memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages)); |
1da177e4 | 1357 | |
b7872fe8 | 1358 | tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF; |
1da177e4 | 1359 | |
02fc6c36 | 1360 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
cca5172a | 1361 | if (sk->sk_state != TCP_ESTABLISHED) |
02fc6c36 | 1362 | set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
1363 | } |
1364 | } | |
1365 | ||
0f0257ea | 1366 | void svc_sock_update_bufs(struct svc_serv *serv) |
1da177e4 LT |
1367 | { |
1368 | /* | |
1369 | * The number of server threads has changed. Update | |
1370 | * rcvbuf and sndbuf accordingly on all sockets | |
1371 | */ | |
8f3a6de3 | 1372 | struct svc_sock *svsk; |
1da177e4 LT |
1373 | |
1374 | spin_lock_bh(&serv->sv_lock); | |
8f3a6de3 | 1375 | list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list) |
02fc6c36 | 1376 | set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
1377 | spin_unlock_bh(&serv->sv_lock); |
1378 | } | |
24c3767e | 1379 | EXPORT_SYMBOL_GPL(svc_sock_update_bufs); |
1da177e4 | 1380 | |
1da177e4 LT |
1381 | /* |
1382 | * Initialize socket for RPC use and create svc_sock struct | |
1383 | * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF. | |
1384 | */ | |
6b174337 CL |
1385 | static struct svc_sock *svc_setup_socket(struct svc_serv *serv, |
1386 | struct socket *sock, | |
72c35376 | 1387 | int flags) |
1da177e4 LT |
1388 | { |
1389 | struct svc_sock *svsk; | |
1390 | struct sock *inet; | |
6b174337 | 1391 | int pmap_register = !(flags & SVC_SOCK_ANONYMOUS); |
72c35376 | 1392 | int err = 0; |
1da177e4 LT |
1393 | |
1394 | dprintk("svc: svc_setup_socket %p\n", sock); | |
72c35376 BF |
1395 | svsk = kzalloc(sizeof(*svsk), GFP_KERNEL); |
1396 | if (!svsk) | |
1397 | return ERR_PTR(-ENOMEM); | |
1da177e4 LT |
1398 | |
1399 | inet = sock->sk; | |
1400 | ||
1401 | /* Register socket with portmapper */ | |
72c35376 BF |
1402 | if (pmap_register) |
1403 | err = svc_register(serv, sock_net(sock->sk), inet->sk_family, | |
5247fab5 | 1404 | inet->sk_protocol, |
c720c7e8 | 1405 | ntohs(inet_sk(inet)->inet_sport)); |
1da177e4 | 1406 | |
72c35376 | 1407 | if (err < 0) { |
1da177e4 | 1408 | kfree(svsk); |
72c35376 | 1409 | return ERR_PTR(err); |
1da177e4 LT |
1410 | } |
1411 | ||
1da177e4 LT |
1412 | inet->sk_user_data = svsk; |
1413 | svsk->sk_sock = sock; | |
1414 | svsk->sk_sk = inet; | |
1415 | svsk->sk_ostate = inet->sk_state_change; | |
1416 | svsk->sk_odata = inet->sk_data_ready; | |
1417 | svsk->sk_owspace = inet->sk_write_space; | |
1da177e4 LT |
1418 | |
1419 | /* Initialize the socket */ | |
1420 | if (sock->type == SOCK_DGRAM) | |
bb5cf160 | 1421 | svc_udp_init(svsk, serv); |
96604398 OK |
1422 | else { |
1423 | /* initialise setting must have enough space to | |
1424 | * receive and respond to one request. | |
1425 | */ | |
1426 | svc_sock_setbufsize(svsk->sk_sock, 4 * serv->sv_max_mesg, | |
1427 | 4 * serv->sv_max_mesg); | |
bb5cf160 | 1428 | svc_tcp_init(svsk, serv); |
96604398 | 1429 | } |
1da177e4 | 1430 | |
1da177e4 LT |
1431 | dprintk("svc: svc_setup_socket created %p (inet %p)\n", |
1432 | svsk, svsk->sk_sk); | |
1433 | ||
1da177e4 LT |
1434 | return svsk; |
1435 | } | |
1436 | ||
bfba9ab4 CL |
1437 | /** |
1438 | * svc_addsock - add a listener socket to an RPC service | |
1439 | * @serv: pointer to RPC service to which to add a new listener | |
1440 | * @fd: file descriptor of the new listener | |
1441 | * @name_return: pointer to buffer to fill in with name of listener | |
1442 | * @len: size of the buffer | |
1443 | * | |
1444 | * Fills in socket name and returns positive length of name if successful. | |
1445 | * Name is terminated with '\n'. On error, returns a negative errno | |
1446 | * value. | |
1447 | */ | |
1448 | int svc_addsock(struct svc_serv *serv, const int fd, char *name_return, | |
1449 | const size_t len) | |
b41b66d6 N |
1450 | { |
1451 | int err = 0; | |
1452 | struct socket *so = sockfd_lookup(fd, &err); | |
1453 | struct svc_sock *svsk = NULL; | |
a8e10078 BF |
1454 | struct sockaddr_storage addr; |
1455 | struct sockaddr *sin = (struct sockaddr *)&addr; | |
1456 | int salen; | |
b41b66d6 N |
1457 | |
1458 | if (!so) | |
1459 | return err; | |
a8e10078 | 1460 | err = -EAFNOSUPPORT; |
205ba423 | 1461 | if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6)) |
a8e10078 BF |
1462 | goto out; |
1463 | err = -EPROTONOSUPPORT; | |
1464 | if (so->sk->sk_protocol != IPPROTO_TCP && | |
b41b66d6 | 1465 | so->sk->sk_protocol != IPPROTO_UDP) |
a8e10078 BF |
1466 | goto out; |
1467 | err = -EISCONN; | |
1468 | if (so->state > SS_UNCONNECTED) | |
1469 | goto out; | |
1470 | err = -ENOENT; | |
1471 | if (!try_module_get(THIS_MODULE)) | |
1472 | goto out; | |
1473 | svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS); | |
1474 | if (IS_ERR(svsk)) { | |
1475 | module_put(THIS_MODULE); | |
1476 | err = PTR_ERR(svsk); | |
1477 | goto out; | |
b41b66d6 | 1478 | } |
a8e10078 BF |
1479 | if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0) |
1480 | svc_xprt_set_local(&svsk->sk_xprt, sin, salen); | |
39b55301 | 1481 | svc_add_new_perm_xprt(serv, &svsk->sk_xprt); |
e7942b9f | 1482 | return svc_one_sock_name(svsk, name_return, len); |
a8e10078 BF |
1483 | out: |
1484 | sockfd_put(so); | |
1485 | return err; | |
b41b66d6 N |
1486 | } |
1487 | EXPORT_SYMBOL_GPL(svc_addsock); | |
1488 | ||
1da177e4 LT |
1489 | /* |
1490 | * Create socket for RPC service. | |
1491 | */ | |
b700cbb1 TT |
1492 | static struct svc_xprt *svc_create_socket(struct svc_serv *serv, |
1493 | int protocol, | |
62832c03 | 1494 | struct net *net, |
b700cbb1 TT |
1495 | struct sockaddr *sin, int len, |
1496 | int flags) | |
1da177e4 LT |
1497 | { |
1498 | struct svc_sock *svsk; | |
1499 | struct socket *sock; | |
1500 | int error; | |
1501 | int type; | |
9dbc240f TT |
1502 | struct sockaddr_storage addr; |
1503 | struct sockaddr *newsin = (struct sockaddr *)&addr; | |
1504 | int newlen; | |
c69da774 TM |
1505 | int family; |
1506 | int val; | |
5216a8e7 | 1507 | RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); |
1da177e4 | 1508 | |
ad06e4bd CL |
1509 | dprintk("svc: svc_create_socket(%s, %d, %s)\n", |
1510 | serv->sv_program->pg_name, protocol, | |
77f1f67a | 1511 | __svc_print_addr(sin, buf, sizeof(buf))); |
1da177e4 LT |
1512 | |
1513 | if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) { | |
1514 | printk(KERN_WARNING "svc: only UDP and TCP " | |
1515 | "sockets supported\n"); | |
b700cbb1 | 1516 | return ERR_PTR(-EINVAL); |
1da177e4 | 1517 | } |
c69da774 | 1518 | |
1da177e4 | 1519 | type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM; |
c69da774 TM |
1520 | switch (sin->sa_family) { |
1521 | case AF_INET6: | |
1522 | family = PF_INET6; | |
1523 | break; | |
1524 | case AF_INET: | |
1525 | family = PF_INET; | |
1526 | break; | |
1527 | default: | |
1528 | return ERR_PTR(-EINVAL); | |
1529 | } | |
1da177e4 | 1530 | |
14ec63c3 | 1531 | error = __sock_create(net, family, type, protocol, &sock, 1); |
77f1f67a | 1532 | if (error < 0) |
b700cbb1 | 1533 | return ERR_PTR(error); |
1da177e4 | 1534 | |
ed07536e PZ |
1535 | svc_reclassify_socket(sock); |
1536 | ||
c69da774 TM |
1537 | /* |
1538 | * If this is an PF_INET6 listener, we want to avoid | |
1539 | * getting requests from IPv4 remotes. Those should | |
1540 | * be shunted to a PF_INET listener via rpcbind. | |
1541 | */ | |
1542 | val = 1; | |
1543 | if (family == PF_INET6) | |
1544 | kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY, | |
1545 | (char *)&val, sizeof(val)); | |
1546 | ||
18114746 | 1547 | if (type == SOCK_STREAM) |
4a17fd52 | 1548 | sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */ |
77f1f67a | 1549 | error = kernel_bind(sock, sin, len); |
18114746 ES |
1550 | if (error < 0) |
1551 | goto bummer; | |
1da177e4 | 1552 | |
9dbc240f TT |
1553 | newlen = len; |
1554 | error = kernel_getsockname(sock, newsin, &newlen); | |
1555 | if (error < 0) | |
1556 | goto bummer; | |
1557 | ||
1da177e4 | 1558 | if (protocol == IPPROTO_TCP) { |
e6242e92 | 1559 | if ((error = kernel_listen(sock, 64)) < 0) |
1da177e4 LT |
1560 | goto bummer; |
1561 | } | |
1562 | ||
72c35376 | 1563 | svsk = svc_setup_socket(serv, sock, flags); |
a8e10078 BF |
1564 | if (IS_ERR(svsk)) { |
1565 | error = PTR_ERR(svsk); | |
1566 | goto bummer; | |
e79eff1f | 1567 | } |
a8e10078 BF |
1568 | svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen); |
1569 | return (struct svc_xprt *)svsk; | |
1da177e4 LT |
1570 | bummer: |
1571 | dprintk("svc: svc_create_socket error = %d\n", -error); | |
1572 | sock_release(sock); | |
b700cbb1 | 1573 | return ERR_PTR(error); |
1da177e4 LT |
1574 | } |
1575 | ||
755cceab TT |
1576 | /* |
1577 | * Detach the svc_sock from the socket so that no | |
1578 | * more callbacks occur. | |
1579 | */ | |
1580 | static void svc_sock_detach(struct svc_xprt *xprt) | |
1581 | { | |
1582 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
1583 | struct sock *sk = svsk->sk_sk; | |
eaefd110 | 1584 | wait_queue_head_t *wq; |
755cceab TT |
1585 | |
1586 | dprintk("svc: svc_sock_detach(%p)\n", svsk); | |
1587 | ||
1588 | /* put back the old socket callbacks */ | |
1589 | sk->sk_state_change = svsk->sk_ostate; | |
1590 | sk->sk_data_ready = svsk->sk_odata; | |
1591 | sk->sk_write_space = svsk->sk_owspace; | |
69b6ba37 | 1592 | |
eaefd110 ED |
1593 | wq = sk_sleep(sk); |
1594 | if (wq && waitqueue_active(wq)) | |
1595 | wake_up_interruptible(wq); | |
69b6ba37 TM |
1596 | } |
1597 | ||
1598 | /* | |
1599 | * Disconnect the socket, and reset the callbacks | |
1600 | */ | |
1601 | static void svc_tcp_sock_detach(struct svc_xprt *xprt) | |
1602 | { | |
1603 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
1604 | ||
1605 | dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk); | |
1606 | ||
1607 | svc_sock_detach(xprt); | |
1608 | ||
31d68ef6 BF |
1609 | if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) { |
1610 | svc_tcp_clear_pages(svsk); | |
69b6ba37 | 1611 | kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR); |
31d68ef6 | 1612 | } |
755cceab TT |
1613 | } |
1614 | ||
1615 | /* | |
1616 | * Free the svc_sock's socket resources and the svc_sock itself. | |
1617 | */ | |
1618 | static void svc_sock_free(struct svc_xprt *xprt) | |
1619 | { | |
1620 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
1621 | dprintk("svc: svc_sock_free(%p)\n", svsk); | |
1622 | ||
755cceab TT |
1623 | if (svsk->sk_sock->file) |
1624 | sockfd_put(svsk->sk_sock); | |
1625 | else | |
1626 | sock_release(svsk->sk_sock); | |
1627 | kfree(svsk); | |
1628 | } | |
7652e5a0 | 1629 | |
9e00abc3 | 1630 | #if defined(CONFIG_SUNRPC_BACKCHANNEL) |
7652e5a0 | 1631 | /* |
1f11a034 | 1632 | * Create a back channel svc_xprt which shares the fore channel socket. |
7652e5a0 | 1633 | */ |
1f11a034 AA |
1634 | static struct svc_xprt *svc_bc_create_socket(struct svc_serv *serv, |
1635 | int protocol, | |
1636 | struct net *net, | |
1637 | struct sockaddr *sin, int len, | |
1638 | int flags) | |
7652e5a0 BH |
1639 | { |
1640 | struct svc_sock *svsk; | |
1f11a034 AA |
1641 | struct svc_xprt *xprt; |
1642 | ||
1643 | if (protocol != IPPROTO_TCP) { | |
1644 | printk(KERN_WARNING "svc: only TCP sockets" | |
1645 | " supported on shared back channel\n"); | |
1646 | return ERR_PTR(-EINVAL); | |
1647 | } | |
7652e5a0 | 1648 | |
7652e5a0 BH |
1649 | svsk = kzalloc(sizeof(*svsk), GFP_KERNEL); |
1650 | if (!svsk) | |
1f11a034 | 1651 | return ERR_PTR(-ENOMEM); |
7652e5a0 BH |
1652 | |
1653 | xprt = &svsk->sk_xprt; | |
bd4620dd | 1654 | svc_xprt_init(net, &svc_tcp_bc_class, xprt, serv); |
1f11a034 | 1655 | |
4a19de0f | 1656 | serv->sv_bc_xprt = xprt; |
1f11a034 | 1657 | |
7652e5a0 BH |
1658 | return xprt; |
1659 | } | |
7652e5a0 BH |
1660 | |
1661 | /* | |
1f11a034 | 1662 | * Free a back channel svc_sock. |
7652e5a0 | 1663 | */ |
1f11a034 | 1664 | static void svc_bc_sock_free(struct svc_xprt *xprt) |
7652e5a0 | 1665 | { |
778be232 | 1666 | if (xprt) |
7652e5a0 BH |
1667 | kfree(container_of(xprt, struct svc_sock, sk_xprt)); |
1668 | } | |
9e00abc3 | 1669 | #endif /* CONFIG_SUNRPC_BACKCHANNEL */ |