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> | |
52 | ||
360d8738 | 53 | #define RPCDBG_FACILITY RPCDBG_SVCXPRT |
1da177e4 LT |
54 | |
55 | ||
56 | static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *, | |
6b174337 | 57 | int *errp, int flags); |
1da177e4 LT |
58 | static void svc_udp_data_ready(struct sock *, int); |
59 | static int svc_udp_recvfrom(struct svc_rqst *); | |
60 | static int svc_udp_sendto(struct svc_rqst *); | |
755cceab | 61 | static void svc_sock_detach(struct svc_xprt *); |
69b6ba37 | 62 | static void svc_tcp_sock_detach(struct svc_xprt *); |
755cceab | 63 | static void svc_sock_free(struct svc_xprt *); |
1da177e4 | 64 | |
b700cbb1 TT |
65 | static struct svc_xprt *svc_create_socket(struct svc_serv *, int, |
66 | struct sockaddr *, int, int); | |
ed07536e PZ |
67 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
68 | static struct lock_class_key svc_key[2]; | |
69 | static struct lock_class_key svc_slock_key[2]; | |
70 | ||
0f0257ea | 71 | static void svc_reclassify_socket(struct socket *sock) |
ed07536e PZ |
72 | { |
73 | struct sock *sk = sock->sk; | |
02b3d346 | 74 | BUG_ON(sock_owned_by_user(sk)); |
ed07536e PZ |
75 | switch (sk->sk_family) { |
76 | case AF_INET: | |
77 | sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD", | |
def13d74 TT |
78 | &svc_slock_key[0], |
79 | "sk_xprt.xpt_lock-AF_INET-NFSD", | |
80 | &svc_key[0]); | |
ed07536e PZ |
81 | break; |
82 | ||
83 | case AF_INET6: | |
84 | sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD", | |
def13d74 TT |
85 | &svc_slock_key[1], |
86 | "sk_xprt.xpt_lock-AF_INET6-NFSD", | |
87 | &svc_key[1]); | |
ed07536e PZ |
88 | break; |
89 | ||
90 | default: | |
91 | BUG(); | |
92 | } | |
93 | } | |
94 | #else | |
0f0257ea | 95 | static void svc_reclassify_socket(struct socket *sock) |
ed07536e PZ |
96 | { |
97 | } | |
98 | #endif | |
99 | ||
1da177e4 LT |
100 | /* |
101 | * Release an skbuff after use | |
102 | */ | |
5148bf4e | 103 | static void svc_release_skb(struct svc_rqst *rqstp) |
1da177e4 | 104 | { |
5148bf4e | 105 | struct sk_buff *skb = rqstp->rq_xprt_ctxt; |
1da177e4 LT |
106 | |
107 | if (skb) { | |
57b1d3ba TT |
108 | struct svc_sock *svsk = |
109 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
5148bf4e | 110 | rqstp->rq_xprt_ctxt = NULL; |
1da177e4 LT |
111 | |
112 | dprintk("svc: service %p, releasing skb %p\n", rqstp, skb); | |
57b1d3ba | 113 | skb_free_datagram(svsk->sk_sk, skb); |
1da177e4 | 114 | } |
1da177e4 LT |
115 | } |
116 | ||
b92503b2 CL |
117 | union svc_pktinfo_u { |
118 | struct in_pktinfo pkti; | |
b92503b2 | 119 | struct in6_pktinfo pkti6; |
b92503b2 | 120 | }; |
bc375ea7 DM |
121 | #define SVC_PKTINFO_SPACE \ |
122 | CMSG_SPACE(sizeof(union svc_pktinfo_u)) | |
b92503b2 CL |
123 | |
124 | static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh) | |
125 | { | |
57b1d3ba TT |
126 | struct svc_sock *svsk = |
127 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
128 | switch (svsk->sk_sk->sk_family) { | |
b92503b2 CL |
129 | case AF_INET: { |
130 | struct in_pktinfo *pki = CMSG_DATA(cmh); | |
131 | ||
132 | cmh->cmsg_level = SOL_IP; | |
133 | cmh->cmsg_type = IP_PKTINFO; | |
134 | pki->ipi_ifindex = 0; | |
135 | pki->ipi_spec_dst.s_addr = rqstp->rq_daddr.addr.s_addr; | |
136 | cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); | |
137 | } | |
138 | break; | |
5a05ed73 | 139 | |
b92503b2 CL |
140 | case AF_INET6: { |
141 | struct in6_pktinfo *pki = CMSG_DATA(cmh); | |
142 | ||
143 | cmh->cmsg_level = SOL_IPV6; | |
144 | cmh->cmsg_type = IPV6_PKTINFO; | |
145 | pki->ipi6_ifindex = 0; | |
146 | ipv6_addr_copy(&pki->ipi6_addr, | |
147 | &rqstp->rq_daddr.addr6); | |
148 | cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); | |
149 | } | |
150 | break; | |
b92503b2 CL |
151 | } |
152 | return; | |
153 | } | |
154 | ||
1da177e4 LT |
155 | /* |
156 | * Generic sendto routine | |
157 | */ | |
0f0257ea | 158 | static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr) |
1da177e4 | 159 | { |
57b1d3ba TT |
160 | struct svc_sock *svsk = |
161 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
1da177e4 LT |
162 | struct socket *sock = svsk->sk_sock; |
163 | int slen; | |
bc375ea7 DM |
164 | union { |
165 | struct cmsghdr hdr; | |
166 | long all[SVC_PKTINFO_SPACE / sizeof(long)]; | |
167 | } buffer; | |
168 | struct cmsghdr *cmh = &buffer.hdr; | |
1da177e4 LT |
169 | int len = 0; |
170 | int result; | |
171 | int size; | |
172 | struct page **ppage = xdr->pages; | |
173 | size_t base = xdr->page_base; | |
174 | unsigned int pglen = xdr->page_len; | |
175 | unsigned int flags = MSG_MORE; | |
5216a8e7 | 176 | RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); |
1da177e4 LT |
177 | |
178 | slen = xdr->len; | |
179 | ||
180 | if (rqstp->rq_prot == IPPROTO_UDP) { | |
b92503b2 CL |
181 | struct msghdr msg = { |
182 | .msg_name = &rqstp->rq_addr, | |
183 | .msg_namelen = rqstp->rq_addrlen, | |
184 | .msg_control = cmh, | |
185 | .msg_controllen = sizeof(buffer), | |
186 | .msg_flags = MSG_MORE, | |
187 | }; | |
188 | ||
189 | svc_set_cmsg_data(rqstp, cmh); | |
1da177e4 LT |
190 | |
191 | if (sock_sendmsg(sock, &msg, 0) < 0) | |
192 | goto out; | |
193 | } | |
194 | ||
195 | /* send head */ | |
196 | if (slen == xdr->head[0].iov_len) | |
197 | flags = 0; | |
44524359 N |
198 | len = kernel_sendpage(sock, rqstp->rq_respages[0], 0, |
199 | xdr->head[0].iov_len, flags); | |
1da177e4 LT |
200 | if (len != xdr->head[0].iov_len) |
201 | goto out; | |
202 | slen -= xdr->head[0].iov_len; | |
203 | if (slen == 0) | |
204 | goto out; | |
205 | ||
206 | /* send page data */ | |
207 | size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen; | |
208 | while (pglen > 0) { | |
209 | if (slen == size) | |
210 | flags = 0; | |
e6242e92 | 211 | result = kernel_sendpage(sock, *ppage, base, size, flags); |
1da177e4 LT |
212 | if (result > 0) |
213 | len += result; | |
214 | if (result != size) | |
215 | goto out; | |
216 | slen -= size; | |
217 | pglen -= size; | |
218 | size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen; | |
219 | base = 0; | |
220 | ppage++; | |
221 | } | |
222 | /* send tail */ | |
223 | if (xdr->tail[0].iov_len) { | |
44524359 N |
224 | result = kernel_sendpage(sock, rqstp->rq_respages[0], |
225 | ((unsigned long)xdr->tail[0].iov_base) | |
cca5172a | 226 | & (PAGE_SIZE-1), |
1da177e4 LT |
227 | xdr->tail[0].iov_len, 0); |
228 | ||
229 | if (result > 0) | |
230 | len += result; | |
231 | } | |
232 | out: | |
ad06e4bd | 233 | dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n", |
57b1d3ba | 234 | svsk, xdr->head[0].iov_base, xdr->head[0].iov_len, |
ad06e4bd | 235 | xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf))); |
1da177e4 LT |
236 | |
237 | return len; | |
238 | } | |
239 | ||
80212d59 N |
240 | /* |
241 | * Report socket names for nfsdfs | |
242 | */ | |
e7942b9f | 243 | static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining) |
80212d59 N |
244 | { |
245 | int len; | |
246 | ||
247 | switch(svsk->sk_sk->sk_family) { | |
e7942b9f CL |
248 | case PF_INET: |
249 | len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n", | |
21454aaa | 250 | svsk->sk_sk->sk_protocol == IPPROTO_UDP ? |
80212d59 | 251 | "udp" : "tcp", |
21454aaa | 252 | &inet_sk(svsk->sk_sk)->rcv_saddr, |
80212d59 N |
253 | inet_sk(svsk->sk_sk)->num); |
254 | break; | |
255 | default: | |
e7942b9f | 256 | len = snprintf(buf, remaining, "*unknown-%d*\n", |
80212d59 N |
257 | svsk->sk_sk->sk_family); |
258 | } | |
e7942b9f CL |
259 | |
260 | if (len >= remaining) { | |
261 | *buf = '\0'; | |
262 | return -ENAMETOOLONG; | |
263 | } | |
80212d59 N |
264 | return len; |
265 | } | |
266 | ||
8435d34d CL |
267 | /** |
268 | * svc_sock_names - construct a list of listener names in a string | |
269 | * @serv: pointer to RPC service | |
270 | * @buf: pointer to a buffer to fill in with socket names | |
271 | * @buflen: size of the buffer to be filled | |
272 | * @toclose: pointer to '\0'-terminated C string containing the name | |
273 | * of a listener to be closed | |
274 | * | |
275 | * Fills in @buf with a '\n'-separated list of names of listener | |
276 | * sockets. If @toclose is not NULL, the socket named by @toclose | |
277 | * is closed, and is not included in the output list. | |
278 | * | |
279 | * Returns positive length of the socket name string, or a negative | |
280 | * errno value on error. | |
281 | */ | |
282 | int svc_sock_names(struct svc_serv *serv, char *buf, const size_t buflen, | |
283 | const char *toclose) | |
80212d59 | 284 | { |
b41b66d6 | 285 | struct svc_sock *svsk, *closesk = NULL; |
80212d59 N |
286 | int len = 0; |
287 | ||
288 | if (!serv) | |
289 | return 0; | |
e7942b9f | 290 | |
aaf68cfb | 291 | spin_lock_bh(&serv->sv_lock); |
7a182083 | 292 | list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list) { |
e7942b9f CL |
293 | int onelen = svc_one_sock_name(svsk, buf + len, buflen - len); |
294 | if (onelen < 0) { | |
295 | len = onelen; | |
296 | break; | |
297 | } | |
298 | if (toclose && strcmp(toclose, buf + len) == 0) | |
b41b66d6 N |
299 | closesk = svsk; |
300 | else | |
301 | len += onelen; | |
80212d59 | 302 | } |
aaf68cfb | 303 | spin_unlock_bh(&serv->sv_lock); |
e7942b9f | 304 | |
b41b66d6 | 305 | if (closesk) |
5680c446 N |
306 | /* Should unregister with portmap, but you cannot |
307 | * unregister just one protocol... | |
308 | */ | |
7a182083 | 309 | svc_close_xprt(&closesk->sk_xprt); |
37a03472 N |
310 | else if (toclose) |
311 | return -ENOENT; | |
80212d59 N |
312 | return len; |
313 | } | |
24c3767e | 314 | EXPORT_SYMBOL_GPL(svc_sock_names); |
80212d59 | 315 | |
1da177e4 LT |
316 | /* |
317 | * Check input queue length | |
318 | */ | |
0f0257ea | 319 | static int svc_recv_available(struct svc_sock *svsk) |
1da177e4 | 320 | { |
1da177e4 LT |
321 | struct socket *sock = svsk->sk_sock; |
322 | int avail, err; | |
323 | ||
e6242e92 | 324 | err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail); |
1da177e4 LT |
325 | |
326 | return (err >= 0)? avail : err; | |
327 | } | |
328 | ||
329 | /* | |
330 | * Generic recvfrom routine. | |
331 | */ | |
0f0257ea TT |
332 | static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr, |
333 | int buflen) | |
1da177e4 | 334 | { |
57b1d3ba TT |
335 | struct svc_sock *svsk = |
336 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
1ba95105 CL |
337 | struct msghdr msg = { |
338 | .msg_flags = MSG_DONTWAIT, | |
339 | }; | |
340 | int len; | |
1da177e4 | 341 | |
260c1d12 TT |
342 | rqstp->rq_xprt_hlen = 0; |
343 | ||
1ba95105 CL |
344 | len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen, |
345 | msg.msg_flags); | |
1da177e4 | 346 | |
1da177e4 | 347 | dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n", |
1ba95105 | 348 | svsk, iov[0].iov_base, iov[0].iov_len, len); |
1da177e4 LT |
349 | return len; |
350 | } | |
351 | ||
352 | /* | |
353 | * Set socket snd and rcv buffer lengths | |
354 | */ | |
0f0257ea TT |
355 | static void svc_sock_setbufsize(struct socket *sock, unsigned int snd, |
356 | unsigned int rcv) | |
1da177e4 LT |
357 | { |
358 | #if 0 | |
359 | mm_segment_t oldfs; | |
360 | oldfs = get_fs(); set_fs(KERNEL_DS); | |
361 | sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF, | |
362 | (char*)&snd, sizeof(snd)); | |
363 | sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF, | |
364 | (char*)&rcv, sizeof(rcv)); | |
365 | #else | |
366 | /* sock_setsockopt limits use to sysctl_?mem_max, | |
367 | * which isn't acceptable. Until that is made conditional | |
368 | * on not having CAP_SYS_RESOURCE or similar, we go direct... | |
369 | * DaveM said I could! | |
370 | */ | |
371 | lock_sock(sock->sk); | |
372 | sock->sk->sk_sndbuf = snd * 2; | |
373 | sock->sk->sk_rcvbuf = rcv * 2; | |
1da177e4 LT |
374 | release_sock(sock->sk); |
375 | #endif | |
376 | } | |
377 | /* | |
378 | * INET callback when data has been received on the socket. | |
379 | */ | |
0f0257ea | 380 | static void svc_udp_data_ready(struct sock *sk, int count) |
1da177e4 | 381 | { |
939bb7ef | 382 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 | 383 | |
939bb7ef NB |
384 | if (svsk) { |
385 | dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n", | |
02fc6c36 TT |
386 | svsk, sk, count, |
387 | test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags)); | |
388 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); | |
f6150c3c | 389 | svc_xprt_enqueue(&svsk->sk_xprt); |
939bb7ef | 390 | } |
1da177e4 LT |
391 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
392 | wake_up_interruptible(sk->sk_sleep); | |
393 | } | |
394 | ||
395 | /* | |
396 | * INET callback when space is newly available on the socket. | |
397 | */ | |
0f0257ea | 398 | static void svc_write_space(struct sock *sk) |
1da177e4 LT |
399 | { |
400 | struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data); | |
401 | ||
402 | if (svsk) { | |
403 | dprintk("svc: socket %p(inet %p), write_space busy=%d\n", | |
02fc6c36 | 404 | svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags)); |
f6150c3c | 405 | svc_xprt_enqueue(&svsk->sk_xprt); |
1da177e4 LT |
406 | } |
407 | ||
408 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) { | |
939bb7ef | 409 | dprintk("RPC svc_write_space: someone sleeping on %p\n", |
1da177e4 LT |
410 | svsk); |
411 | wake_up_interruptible(sk->sk_sleep); | |
412 | } | |
413 | } | |
414 | ||
9dbc240f TT |
415 | /* |
416 | * Copy the UDP datagram's destination address to the rqstp structure. | |
417 | * The 'destination' address in this case is the address to which the | |
418 | * peer sent the datagram, i.e. our local address. For multihomed | |
419 | * hosts, this can change from msg to msg. Note that only the IP | |
420 | * address changes, the port number should remain the same. | |
421 | */ | |
422 | static void svc_udp_get_dest_address(struct svc_rqst *rqstp, | |
423 | struct cmsghdr *cmh) | |
95756482 | 424 | { |
57b1d3ba TT |
425 | struct svc_sock *svsk = |
426 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
427 | switch (svsk->sk_sk->sk_family) { | |
95756482 | 428 | case AF_INET: { |
7a37f578 N |
429 | struct in_pktinfo *pki = CMSG_DATA(cmh); |
430 | rqstp->rq_daddr.addr.s_addr = pki->ipi_spec_dst.s_addr; | |
95756482 | 431 | break; |
7a37f578 | 432 | } |
95756482 | 433 | case AF_INET6: { |
7a37f578 N |
434 | struct in6_pktinfo *pki = CMSG_DATA(cmh); |
435 | ipv6_addr_copy(&rqstp->rq_daddr.addr6, &pki->ipi6_addr); | |
95756482 | 436 | break; |
7a37f578 | 437 | } |
95756482 | 438 | } |
95756482 CL |
439 | } |
440 | ||
1da177e4 LT |
441 | /* |
442 | * Receive a datagram from a UDP socket. | |
443 | */ | |
0f0257ea | 444 | static int svc_udp_recvfrom(struct svc_rqst *rqstp) |
1da177e4 | 445 | { |
57b1d3ba TT |
446 | struct svc_sock *svsk = |
447 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
bb5cf160 | 448 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
1da177e4 | 449 | struct sk_buff *skb; |
bc375ea7 DM |
450 | union { |
451 | struct cmsghdr hdr; | |
452 | long all[SVC_PKTINFO_SPACE / sizeof(long)]; | |
453 | } buffer; | |
454 | struct cmsghdr *cmh = &buffer.hdr; | |
7a37f578 N |
455 | struct msghdr msg = { |
456 | .msg_name = svc_addr(rqstp), | |
457 | .msg_control = cmh, | |
458 | .msg_controllen = sizeof(buffer), | |
459 | .msg_flags = MSG_DONTWAIT, | |
460 | }; | |
abc5c44d CL |
461 | size_t len; |
462 | int err; | |
1da177e4 | 463 | |
02fc6c36 | 464 | if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags)) |
1da177e4 LT |
465 | /* udp sockets need large rcvbuf as all pending |
466 | * requests are still in that buffer. sndbuf must | |
467 | * also be large enough that there is enough space | |
3262c816 GB |
468 | * for one reply per thread. We count all threads |
469 | * rather than threads in a particular pool, which | |
470 | * provides an upper bound on the number of threads | |
471 | * which will access the socket. | |
1da177e4 LT |
472 | */ |
473 | svc_sock_setbufsize(svsk->sk_sock, | |
c6b0a9f8 N |
474 | (serv->sv_nrthreads+3) * serv->sv_max_mesg, |
475 | (serv->sv_nrthreads+3) * serv->sv_max_mesg); | |
1da177e4 | 476 | |
02fc6c36 | 477 | clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
05ed690e N |
478 | skb = NULL; |
479 | err = kernel_recvmsg(svsk->sk_sock, &msg, NULL, | |
480 | 0, 0, MSG_PEEK | MSG_DONTWAIT); | |
481 | if (err >= 0) | |
482 | skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err); | |
483 | ||
484 | if (skb == NULL) { | |
485 | if (err != -EAGAIN) { | |
486 | /* possibly an icmp error */ | |
487 | dprintk("svc: recvfrom returned error %d\n", -err); | |
02fc6c36 | 488 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 489 | } |
a6046f71 | 490 | svc_xprt_received(&svsk->sk_xprt); |
05ed690e | 491 | return -EAGAIN; |
1da177e4 | 492 | } |
9dbc240f | 493 | len = svc_addr_len(svc_addr(rqstp)); |
abc5c44d CL |
494 | if (len == 0) |
495 | return -EAFNOSUPPORT; | |
9dbc240f | 496 | rqstp->rq_addrlen = len; |
b7aa0bf7 ED |
497 | if (skb->tstamp.tv64 == 0) { |
498 | skb->tstamp = ktime_get_real(); | |
cca5172a | 499 | /* Don't enable netstamp, sunrpc doesn't |
1da177e4 LT |
500 | need that much accuracy */ |
501 | } | |
b7aa0bf7 | 502 | svsk->sk_sk->sk_stamp = skb->tstamp; |
02fc6c36 | 503 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */ |
1da177e4 LT |
504 | |
505 | /* | |
506 | * Maybe more packets - kick another thread ASAP. | |
507 | */ | |
a6046f71 | 508 | svc_xprt_received(&svsk->sk_xprt); |
1da177e4 LT |
509 | |
510 | len = skb->len - sizeof(struct udphdr); | |
511 | rqstp->rq_arg.len = len; | |
512 | ||
95756482 | 513 | rqstp->rq_prot = IPPROTO_UDP; |
27459f09 | 514 | |
7a37f578 N |
515 | if (cmh->cmsg_level != IPPROTO_IP || |
516 | cmh->cmsg_type != IP_PKTINFO) { | |
517 | if (net_ratelimit()) | |
518 | printk("rpcsvc: received unknown control message:" | |
519 | "%d/%d\n", | |
520 | cmh->cmsg_level, cmh->cmsg_type); | |
521 | skb_free_datagram(svsk->sk_sk, skb); | |
522 | return 0; | |
523 | } | |
524 | svc_udp_get_dest_address(rqstp, cmh); | |
1da177e4 LT |
525 | |
526 | if (skb_is_nonlinear(skb)) { | |
527 | /* we have to copy */ | |
528 | local_bh_disable(); | |
529 | if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) { | |
530 | local_bh_enable(); | |
531 | /* checksum error */ | |
532 | skb_free_datagram(svsk->sk_sk, skb); | |
533 | return 0; | |
534 | } | |
535 | local_bh_enable(); | |
cca5172a | 536 | skb_free_datagram(svsk->sk_sk, skb); |
1da177e4 LT |
537 | } else { |
538 | /* we can use it in-place */ | |
0f0257ea TT |
539 | rqstp->rq_arg.head[0].iov_base = skb->data + |
540 | sizeof(struct udphdr); | |
1da177e4 | 541 | rqstp->rq_arg.head[0].iov_len = len; |
fb286bb2 HX |
542 | if (skb_checksum_complete(skb)) { |
543 | skb_free_datagram(svsk->sk_sk, skb); | |
544 | return 0; | |
1da177e4 | 545 | } |
5148bf4e | 546 | rqstp->rq_xprt_ctxt = skb; |
1da177e4 LT |
547 | } |
548 | ||
549 | rqstp->rq_arg.page_base = 0; | |
550 | if (len <= rqstp->rq_arg.head[0].iov_len) { | |
551 | rqstp->rq_arg.head[0].iov_len = len; | |
552 | rqstp->rq_arg.page_len = 0; | |
44524359 | 553 | rqstp->rq_respages = rqstp->rq_pages+1; |
1da177e4 LT |
554 | } else { |
555 | rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len; | |
44524359 | 556 | rqstp->rq_respages = rqstp->rq_pages + 1 + |
172589cc | 557 | DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE); |
1da177e4 LT |
558 | } |
559 | ||
560 | if (serv->sv_stats) | |
561 | serv->sv_stats->netudpcnt++; | |
562 | ||
563 | return len; | |
564 | } | |
565 | ||
566 | static int | |
567 | svc_udp_sendto(struct svc_rqst *rqstp) | |
568 | { | |
569 | int error; | |
570 | ||
571 | error = svc_sendto(rqstp, &rqstp->rq_res); | |
572 | if (error == -ECONNREFUSED) | |
573 | /* ICMP error on earlier request. */ | |
574 | error = svc_sendto(rqstp, &rqstp->rq_res); | |
575 | ||
576 | return error; | |
577 | } | |
578 | ||
e831fe65 TT |
579 | static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp) |
580 | { | |
581 | } | |
582 | ||
323bee32 TT |
583 | static int svc_udp_has_wspace(struct svc_xprt *xprt) |
584 | { | |
585 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
bb5cf160 | 586 | struct svc_serv *serv = xprt->xpt_server; |
323bee32 TT |
587 | unsigned long required; |
588 | ||
589 | /* | |
590 | * Set the SOCK_NOSPACE flag before checking the available | |
591 | * sock space. | |
592 | */ | |
593 | set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
7a90e8cc | 594 | required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg; |
323bee32 TT |
595 | if (required*2 > sock_wspace(svsk->sk_sk)) |
596 | return 0; | |
597 | clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
598 | return 1; | |
599 | } | |
600 | ||
38a417cc TT |
601 | static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt) |
602 | { | |
603 | BUG(); | |
604 | return NULL; | |
605 | } | |
606 | ||
b700cbb1 TT |
607 | static struct svc_xprt *svc_udp_create(struct svc_serv *serv, |
608 | struct sockaddr *sa, int salen, | |
609 | int flags) | |
610 | { | |
611 | return svc_create_socket(serv, IPPROTO_UDP, sa, salen, flags); | |
612 | } | |
613 | ||
360d8738 | 614 | static struct svc_xprt_ops svc_udp_ops = { |
b700cbb1 | 615 | .xpo_create = svc_udp_create, |
5d137990 TT |
616 | .xpo_recvfrom = svc_udp_recvfrom, |
617 | .xpo_sendto = svc_udp_sendto, | |
5148bf4e | 618 | .xpo_release_rqst = svc_release_skb, |
755cceab TT |
619 | .xpo_detach = svc_sock_detach, |
620 | .xpo_free = svc_sock_free, | |
e831fe65 | 621 | .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr, |
323bee32 | 622 | .xpo_has_wspace = svc_udp_has_wspace, |
38a417cc | 623 | .xpo_accept = svc_udp_accept, |
360d8738 TT |
624 | }; |
625 | ||
626 | static struct svc_xprt_class svc_udp_class = { | |
627 | .xcl_name = "udp", | |
b700cbb1 | 628 | .xcl_owner = THIS_MODULE, |
360d8738 | 629 | .xcl_ops = &svc_udp_ops, |
49023155 | 630 | .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP, |
360d8738 TT |
631 | }; |
632 | ||
bb5cf160 | 633 | static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv) |
1da177e4 | 634 | { |
7a37f578 N |
635 | int one = 1; |
636 | mm_segment_t oldfs; | |
637 | ||
bb5cf160 | 638 | svc_xprt_init(&svc_udp_class, &svsk->sk_xprt, serv); |
def13d74 | 639 | clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
640 | svsk->sk_sk->sk_data_ready = svc_udp_data_ready; |
641 | svsk->sk_sk->sk_write_space = svc_write_space; | |
1da177e4 LT |
642 | |
643 | /* initialise setting must have enough space to | |
cca5172a | 644 | * receive and respond to one request. |
1da177e4 LT |
645 | * svc_udp_recvfrom will re-adjust if necessary |
646 | */ | |
647 | svc_sock_setbufsize(svsk->sk_sock, | |
bb5cf160 TT |
648 | 3 * svsk->sk_xprt.xpt_server->sv_max_mesg, |
649 | 3 * svsk->sk_xprt.xpt_server->sv_max_mesg); | |
1da177e4 | 650 | |
0f0257ea TT |
651 | /* data might have come in before data_ready set up */ |
652 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); | |
02fc6c36 | 653 | set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); |
7a37f578 N |
654 | |
655 | oldfs = get_fs(); | |
656 | set_fs(KERNEL_DS); | |
657 | /* make sure we get destination address info */ | |
658 | svsk->sk_sock->ops->setsockopt(svsk->sk_sock, IPPROTO_IP, IP_PKTINFO, | |
659 | (char __user *)&one, sizeof(one)); | |
660 | set_fs(oldfs); | |
1da177e4 LT |
661 | } |
662 | ||
663 | /* | |
664 | * A data_ready event on a listening socket means there's a connection | |
665 | * pending. Do not use state_change as a substitute for it. | |
666 | */ | |
0f0257ea | 667 | static void svc_tcp_listen_data_ready(struct sock *sk, int count_unused) |
1da177e4 | 668 | { |
939bb7ef | 669 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 LT |
670 | |
671 | dprintk("svc: socket %p TCP (listen) state change %d\n", | |
939bb7ef | 672 | sk, sk->sk_state); |
1da177e4 | 673 | |
939bb7ef NB |
674 | /* |
675 | * This callback may called twice when a new connection | |
676 | * is established as a child socket inherits everything | |
677 | * from a parent LISTEN socket. | |
678 | * 1) data_ready method of the parent socket will be called | |
679 | * when one of child sockets become ESTABLISHED. | |
680 | * 2) data_ready method of the child socket may be called | |
681 | * when it receives data before the socket is accepted. | |
682 | * In case of 2, we should ignore it silently. | |
683 | */ | |
684 | if (sk->sk_state == TCP_LISTEN) { | |
685 | if (svsk) { | |
02fc6c36 | 686 | set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
f6150c3c | 687 | svc_xprt_enqueue(&svsk->sk_xprt); |
939bb7ef NB |
688 | } else |
689 | printk("svc: socket %p: no user data\n", sk); | |
1da177e4 | 690 | } |
939bb7ef | 691 | |
1da177e4 LT |
692 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
693 | wake_up_interruptible_all(sk->sk_sleep); | |
694 | } | |
695 | ||
696 | /* | |
697 | * A state change on a connected socket means it's dying or dead. | |
698 | */ | |
0f0257ea | 699 | static void svc_tcp_state_change(struct sock *sk) |
1da177e4 | 700 | { |
939bb7ef | 701 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 LT |
702 | |
703 | dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n", | |
939bb7ef | 704 | sk, sk->sk_state, sk->sk_user_data); |
1da177e4 | 705 | |
939bb7ef | 706 | if (!svsk) |
1da177e4 | 707 | printk("svc: socket %p: no user data\n", sk); |
939bb7ef | 708 | else { |
02fc6c36 | 709 | set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); |
f6150c3c | 710 | svc_xprt_enqueue(&svsk->sk_xprt); |
1da177e4 | 711 | } |
1da177e4 LT |
712 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
713 | wake_up_interruptible_all(sk->sk_sleep); | |
714 | } | |
715 | ||
0f0257ea | 716 | static void svc_tcp_data_ready(struct sock *sk, int count) |
1da177e4 | 717 | { |
939bb7ef | 718 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 LT |
719 | |
720 | dprintk("svc: socket %p TCP data ready (svsk %p)\n", | |
939bb7ef NB |
721 | sk, sk->sk_user_data); |
722 | if (svsk) { | |
02fc6c36 | 723 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
f6150c3c | 724 | svc_xprt_enqueue(&svsk->sk_xprt); |
939bb7ef | 725 | } |
1da177e4 LT |
726 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
727 | wake_up_interruptible(sk->sk_sleep); | |
728 | } | |
729 | ||
730 | /* | |
731 | * Accept a TCP connection | |
732 | */ | |
38a417cc | 733 | static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt) |
1da177e4 | 734 | { |
38a417cc | 735 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); |
cdd88b9f AM |
736 | struct sockaddr_storage addr; |
737 | struct sockaddr *sin = (struct sockaddr *) &addr; | |
bb5cf160 | 738 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
1da177e4 LT |
739 | struct socket *sock = svsk->sk_sock; |
740 | struct socket *newsock; | |
1da177e4 LT |
741 | struct svc_sock *newsvsk; |
742 | int err, slen; | |
5216a8e7 | 743 | RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); |
1da177e4 LT |
744 | |
745 | dprintk("svc: tcp_accept %p sock %p\n", svsk, sock); | |
746 | if (!sock) | |
38a417cc | 747 | return NULL; |
1da177e4 | 748 | |
02fc6c36 | 749 | clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
e6242e92 SS |
750 | err = kernel_accept(sock, &newsock, O_NONBLOCK); |
751 | if (err < 0) { | |
1da177e4 LT |
752 | if (err == -ENOMEM) |
753 | printk(KERN_WARNING "%s: no more sockets!\n", | |
754 | serv->sv_name); | |
e6242e92 | 755 | else if (err != -EAGAIN && net_ratelimit()) |
1da177e4 LT |
756 | printk(KERN_WARNING "%s: accept failed (err %d)!\n", |
757 | serv->sv_name, -err); | |
38a417cc | 758 | return NULL; |
1da177e4 | 759 | } |
02fc6c36 | 760 | set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 761 | |
cdd88b9f | 762 | err = kernel_getpeername(newsock, sin, &slen); |
1da177e4 LT |
763 | if (err < 0) { |
764 | if (net_ratelimit()) | |
765 | printk(KERN_WARNING "%s: peername failed (err %d)!\n", | |
766 | serv->sv_name, -err); | |
767 | goto failed; /* aborted connection or whatever */ | |
768 | } | |
769 | ||
770 | /* Ideally, we would want to reject connections from unauthorized | |
ad06e4bd CL |
771 | * hosts here, but when we get encryption, the IP of the host won't |
772 | * tell us anything. For now just warn about unpriv connections. | |
1da177e4 | 773 | */ |
cdd88b9f | 774 | if (!svc_port_is_privileged(sin)) { |
1da177e4 | 775 | dprintk(KERN_WARNING |
ad06e4bd | 776 | "%s: connect from unprivileged port: %s\n", |
cca5172a | 777 | serv->sv_name, |
cdd88b9f | 778 | __svc_print_addr(sin, buf, sizeof(buf))); |
1da177e4 | 779 | } |
ad06e4bd | 780 | dprintk("%s: connect from %s\n", serv->sv_name, |
cdd88b9f | 781 | __svc_print_addr(sin, buf, sizeof(buf))); |
1da177e4 LT |
782 | |
783 | /* make sure that a write doesn't block forever when | |
784 | * low on memory | |
785 | */ | |
786 | newsock->sk->sk_sndtimeo = HZ*30; | |
787 | ||
6b174337 CL |
788 | if (!(newsvsk = svc_setup_socket(serv, newsock, &err, |
789 | (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY)))) | |
1da177e4 | 790 | goto failed; |
9dbc240f | 791 | svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen); |
a9747692 FM |
792 | err = kernel_getsockname(newsock, sin, &slen); |
793 | if (unlikely(err < 0)) { | |
794 | dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err); | |
795 | slen = offsetof(struct sockaddr, sa_data); | |
796 | } | |
9dbc240f | 797 | svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen); |
067d7817 | 798 | |
f9f3cc4f TT |
799 | if (serv->sv_stats) |
800 | serv->sv_stats->nettcpconn++; | |
801 | ||
802 | return &newsvsk->sk_xprt; | |
803 | ||
804 | failed: | |
805 | sock_release(newsock); | |
806 | return NULL; | |
807 | } | |
808 | ||
1da177e4 LT |
809 | /* |
810 | * Receive data from a TCP socket. | |
811 | */ | |
0f0257ea | 812 | static int svc_tcp_recvfrom(struct svc_rqst *rqstp) |
1da177e4 | 813 | { |
57b1d3ba TT |
814 | struct svc_sock *svsk = |
815 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
bb5cf160 | 816 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
1da177e4 | 817 | int len; |
3cc03b16 | 818 | struct kvec *vec; |
1da177e4 LT |
819 | int pnum, vlen; |
820 | ||
821 | dprintk("svc: tcp_recv %p data %d conn %d close %d\n", | |
02fc6c36 TT |
822 | svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags), |
823 | test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags), | |
824 | test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags)); | |
1da177e4 | 825 | |
02fc6c36 | 826 | clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
827 | |
828 | /* Receive data. If we haven't got the record length yet, get | |
829 | * the next four bytes. Otherwise try to gobble up as much as | |
830 | * possible up to the complete record length. | |
831 | */ | |
c0401ea0 CL |
832 | if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) { |
833 | int want = sizeof(rpc_fraghdr) - svsk->sk_tcplen; | |
1da177e4 LT |
834 | struct kvec iov; |
835 | ||
836 | iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen; | |
837 | iov.iov_len = want; | |
838 | if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0) | |
839 | goto error; | |
840 | svsk->sk_tcplen += len; | |
841 | ||
842 | if (len < want) { | |
c0401ea0 CL |
843 | dprintk("svc: short recvfrom while reading record " |
844 | "length (%d of %d)\n", len, want); | |
a6046f71 | 845 | svc_xprt_received(&svsk->sk_xprt); |
1da177e4 LT |
846 | return -EAGAIN; /* record header not complete */ |
847 | } | |
848 | ||
849 | svsk->sk_reclen = ntohl(svsk->sk_reclen); | |
c0401ea0 | 850 | if (!(svsk->sk_reclen & RPC_LAST_STREAM_FRAGMENT)) { |
1da177e4 LT |
851 | /* FIXME: technically, a record can be fragmented, |
852 | * and non-terminal fragments will not have the top | |
853 | * bit set in the fragment length header. | |
854 | * But apparently no known nfs clients send fragmented | |
855 | * records. */ | |
34e9a63b | 856 | if (net_ratelimit()) |
c0401ea0 CL |
857 | printk(KERN_NOTICE "RPC: multiple fragments " |
858 | "per record not supported\n"); | |
1da177e4 LT |
859 | goto err_delete; |
860 | } | |
c0401ea0 | 861 | svsk->sk_reclen &= RPC_FRAGMENT_SIZE_MASK; |
1da177e4 | 862 | dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen); |
c6b0a9f8 | 863 | if (svsk->sk_reclen > serv->sv_max_mesg) { |
34e9a63b | 864 | if (net_ratelimit()) |
c0401ea0 CL |
865 | printk(KERN_NOTICE "RPC: " |
866 | "fragment too large: 0x%08lx\n", | |
867 | (unsigned long)svsk->sk_reclen); | |
1da177e4 LT |
868 | goto err_delete; |
869 | } | |
870 | } | |
871 | ||
872 | /* Check whether enough data is available */ | |
873 | len = svc_recv_available(svsk); | |
874 | if (len < 0) | |
875 | goto error; | |
876 | ||
877 | if (len < svsk->sk_reclen) { | |
878 | dprintk("svc: incomplete TCP record (%d of %d)\n", | |
879 | len, svsk->sk_reclen); | |
a6046f71 | 880 | svc_xprt_received(&svsk->sk_xprt); |
1da177e4 LT |
881 | return -EAGAIN; /* record not complete */ |
882 | } | |
883 | len = svsk->sk_reclen; | |
02fc6c36 | 884 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 885 | |
3cc03b16 | 886 | vec = rqstp->rq_vec; |
1da177e4 LT |
887 | vec[0] = rqstp->rq_arg.head[0]; |
888 | vlen = PAGE_SIZE; | |
889 | pnum = 1; | |
890 | while (vlen < len) { | |
44524359 | 891 | vec[pnum].iov_base = page_address(rqstp->rq_pages[pnum]); |
1da177e4 LT |
892 | vec[pnum].iov_len = PAGE_SIZE; |
893 | pnum++; | |
894 | vlen += PAGE_SIZE; | |
895 | } | |
44524359 | 896 | rqstp->rq_respages = &rqstp->rq_pages[pnum]; |
1da177e4 LT |
897 | |
898 | /* Now receive data */ | |
899 | len = svc_recvfrom(rqstp, vec, pnum, len); | |
900 | if (len < 0) | |
901 | goto error; | |
902 | ||
903 | dprintk("svc: TCP complete record (%d bytes)\n", len); | |
904 | rqstp->rq_arg.len = len; | |
905 | rqstp->rq_arg.page_base = 0; | |
906 | if (len <= rqstp->rq_arg.head[0].iov_len) { | |
907 | rqstp->rq_arg.head[0].iov_len = len; | |
908 | rqstp->rq_arg.page_len = 0; | |
909 | } else { | |
910 | rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len; | |
911 | } | |
912 | ||
5148bf4e | 913 | rqstp->rq_xprt_ctxt = NULL; |
1da177e4 LT |
914 | rqstp->rq_prot = IPPROTO_TCP; |
915 | ||
916 | /* Reset TCP read info */ | |
917 | svsk->sk_reclen = 0; | |
918 | svsk->sk_tcplen = 0; | |
919 | ||
9dbc240f | 920 | svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt); |
a6046f71 | 921 | svc_xprt_received(&svsk->sk_xprt); |
1da177e4 LT |
922 | if (serv->sv_stats) |
923 | serv->sv_stats->nettcpcnt++; | |
924 | ||
925 | return len; | |
926 | ||
927 | err_delete: | |
02fc6c36 | 928 | set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
929 | return -EAGAIN; |
930 | ||
931 | error: | |
932 | if (len == -EAGAIN) { | |
933 | dprintk("RPC: TCP recvfrom got EAGAIN\n"); | |
a6046f71 | 934 | svc_xprt_received(&svsk->sk_xprt); |
1da177e4 LT |
935 | } else { |
936 | printk(KERN_NOTICE "%s: recvfrom returned errno %d\n", | |
bb5cf160 | 937 | svsk->sk_xprt.xpt_server->sv_name, -len); |
93fbf1a5 | 938 | goto err_delete; |
1da177e4 LT |
939 | } |
940 | ||
941 | return len; | |
942 | } | |
943 | ||
944 | /* | |
945 | * Send out data on TCP socket. | |
946 | */ | |
0f0257ea | 947 | static int svc_tcp_sendto(struct svc_rqst *rqstp) |
1da177e4 LT |
948 | { |
949 | struct xdr_buf *xbufp = &rqstp->rq_res; | |
950 | int sent; | |
d8ed029d | 951 | __be32 reclen; |
1da177e4 LT |
952 | |
953 | /* Set up the first element of the reply kvec. | |
954 | * Any other kvecs that may be in use have been taken | |
955 | * care of by the server implementation itself. | |
956 | */ | |
957 | reclen = htonl(0x80000000|((xbufp->len ) - 4)); | |
958 | memcpy(xbufp->head[0].iov_base, &reclen, 4); | |
959 | ||
57b1d3ba | 960 | if (test_bit(XPT_DEAD, &rqstp->rq_xprt->xpt_flags)) |
1da177e4 LT |
961 | return -ENOTCONN; |
962 | ||
963 | sent = svc_sendto(rqstp, &rqstp->rq_res); | |
964 | if (sent != xbufp->len) { | |
0f0257ea TT |
965 | printk(KERN_NOTICE |
966 | "rpc-srv/tcp: %s: %s %d when sending %d bytes " | |
967 | "- shutting down socket\n", | |
57b1d3ba | 968 | rqstp->rq_xprt->xpt_server->sv_name, |
1da177e4 LT |
969 | (sent<0)?"got error":"sent only", |
970 | sent, xbufp->len); | |
57b1d3ba | 971 | set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags); |
f6150c3c | 972 | svc_xprt_enqueue(rqstp->rq_xprt); |
1da177e4 LT |
973 | sent = -EAGAIN; |
974 | } | |
975 | return sent; | |
976 | } | |
977 | ||
e831fe65 TT |
978 | /* |
979 | * Setup response header. TCP has a 4B record length field. | |
980 | */ | |
981 | static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp) | |
982 | { | |
983 | struct kvec *resv = &rqstp->rq_res.head[0]; | |
984 | ||
985 | /* tcp needs a space for the record length... */ | |
986 | svc_putnl(resv, 0); | |
987 | } | |
988 | ||
323bee32 TT |
989 | static int svc_tcp_has_wspace(struct svc_xprt *xprt) |
990 | { | |
57b1d3ba | 991 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); |
bb5cf160 | 992 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
323bee32 TT |
993 | int required; |
994 | int wspace; | |
995 | ||
996 | /* | |
997 | * Set the SOCK_NOSPACE flag before checking the available | |
998 | * sock space. | |
999 | */ | |
1000 | set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
7a90e8cc | 1001 | required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg; |
323bee32 TT |
1002 | wspace = sk_stream_wspace(svsk->sk_sk); |
1003 | ||
1004 | if (wspace < sk_stream_min_wspace(svsk->sk_sk)) | |
1005 | return 0; | |
1006 | if (required * 2 > wspace) | |
1007 | return 0; | |
1008 | ||
1009 | clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
1010 | return 1; | |
1011 | } | |
1012 | ||
b700cbb1 TT |
1013 | static struct svc_xprt *svc_tcp_create(struct svc_serv *serv, |
1014 | struct sockaddr *sa, int salen, | |
1015 | int flags) | |
1016 | { | |
1017 | return svc_create_socket(serv, IPPROTO_TCP, sa, salen, flags); | |
1018 | } | |
1019 | ||
360d8738 | 1020 | static struct svc_xprt_ops svc_tcp_ops = { |
b700cbb1 | 1021 | .xpo_create = svc_tcp_create, |
5d137990 TT |
1022 | .xpo_recvfrom = svc_tcp_recvfrom, |
1023 | .xpo_sendto = svc_tcp_sendto, | |
5148bf4e | 1024 | .xpo_release_rqst = svc_release_skb, |
69b6ba37 | 1025 | .xpo_detach = svc_tcp_sock_detach, |
755cceab | 1026 | .xpo_free = svc_sock_free, |
e831fe65 | 1027 | .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr, |
323bee32 | 1028 | .xpo_has_wspace = svc_tcp_has_wspace, |
38a417cc | 1029 | .xpo_accept = svc_tcp_accept, |
360d8738 TT |
1030 | }; |
1031 | ||
1032 | static struct svc_xprt_class svc_tcp_class = { | |
1033 | .xcl_name = "tcp", | |
b700cbb1 | 1034 | .xcl_owner = THIS_MODULE, |
360d8738 | 1035 | .xcl_ops = &svc_tcp_ops, |
49023155 | 1036 | .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP, |
360d8738 TT |
1037 | }; |
1038 | ||
1039 | void svc_init_xprt_sock(void) | |
1040 | { | |
1041 | svc_reg_xprt_class(&svc_tcp_class); | |
1042 | svc_reg_xprt_class(&svc_udp_class); | |
1043 | } | |
1044 | ||
1045 | void svc_cleanup_xprt_sock(void) | |
1046 | { | |
1047 | svc_unreg_xprt_class(&svc_tcp_class); | |
1048 | svc_unreg_xprt_class(&svc_udp_class); | |
1049 | } | |
1050 | ||
bb5cf160 | 1051 | static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv) |
1da177e4 LT |
1052 | { |
1053 | struct sock *sk = svsk->sk_sk; | |
1da177e4 | 1054 | |
bb5cf160 | 1055 | svc_xprt_init(&svc_tcp_class, &svsk->sk_xprt, serv); |
def13d74 | 1056 | set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
1057 | if (sk->sk_state == TCP_LISTEN) { |
1058 | dprintk("setting up TCP socket for listening\n"); | |
02fc6c36 | 1059 | set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 1060 | sk->sk_data_ready = svc_tcp_listen_data_ready; |
02fc6c36 | 1061 | set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
1062 | } else { |
1063 | dprintk("setting up TCP socket for reading\n"); | |
1064 | sk->sk_state_change = svc_tcp_state_change; | |
1065 | sk->sk_data_ready = svc_tcp_data_ready; | |
1066 | sk->sk_write_space = svc_write_space; | |
1067 | ||
1068 | svsk->sk_reclen = 0; | |
1069 | svsk->sk_tcplen = 0; | |
1070 | ||
b7872fe8 | 1071 | tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF; |
1da177e4 | 1072 | |
02fc6c36 | 1073 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
cca5172a | 1074 | if (sk->sk_state != TCP_ESTABLISHED) |
02fc6c36 | 1075 | set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
1076 | } |
1077 | } | |
1078 | ||
0f0257ea | 1079 | void svc_sock_update_bufs(struct svc_serv *serv) |
1da177e4 LT |
1080 | { |
1081 | /* | |
1082 | * The number of server threads has changed. Update | |
1083 | * rcvbuf and sndbuf accordingly on all sockets | |
1084 | */ | |
1085 | struct list_head *le; | |
1086 | ||
1087 | spin_lock_bh(&serv->sv_lock); | |
1088 | list_for_each(le, &serv->sv_permsocks) { | |
cca5172a | 1089 | struct svc_sock *svsk = |
7a182083 | 1090 | list_entry(le, struct svc_sock, sk_xprt.xpt_list); |
02fc6c36 | 1091 | set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
1092 | } |
1093 | list_for_each(le, &serv->sv_tempsocks) { | |
1094 | struct svc_sock *svsk = | |
7a182083 | 1095 | list_entry(le, struct svc_sock, sk_xprt.xpt_list); |
02fc6c36 | 1096 | set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
1097 | } |
1098 | spin_unlock_bh(&serv->sv_lock); | |
1099 | } | |
24c3767e | 1100 | EXPORT_SYMBOL_GPL(svc_sock_update_bufs); |
1da177e4 | 1101 | |
1da177e4 LT |
1102 | /* |
1103 | * Initialize socket for RPC use and create svc_sock struct | |
1104 | * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF. | |
1105 | */ | |
6b174337 CL |
1106 | static struct svc_sock *svc_setup_socket(struct svc_serv *serv, |
1107 | struct socket *sock, | |
1108 | int *errp, int flags) | |
1da177e4 LT |
1109 | { |
1110 | struct svc_sock *svsk; | |
1111 | struct sock *inet; | |
6b174337 | 1112 | int pmap_register = !(flags & SVC_SOCK_ANONYMOUS); |
1da177e4 LT |
1113 | |
1114 | dprintk("svc: svc_setup_socket %p\n", sock); | |
0da974f4 | 1115 | if (!(svsk = kzalloc(sizeof(*svsk), GFP_KERNEL))) { |
1da177e4 LT |
1116 | *errp = -ENOMEM; |
1117 | return NULL; | |
1118 | } | |
1da177e4 LT |
1119 | |
1120 | inet = sock->sk; | |
1121 | ||
1122 | /* Register socket with portmapper */ | |
1123 | if (*errp >= 0 && pmap_register) | |
baf01caf | 1124 | *errp = svc_register(serv, inet->sk_family, inet->sk_protocol, |
1da177e4 LT |
1125 | ntohs(inet_sk(inet)->sport)); |
1126 | ||
1127 | if (*errp < 0) { | |
1128 | kfree(svsk); | |
1129 | return NULL; | |
1130 | } | |
1131 | ||
1da177e4 LT |
1132 | inet->sk_user_data = svsk; |
1133 | svsk->sk_sock = sock; | |
1134 | svsk->sk_sk = inet; | |
1135 | svsk->sk_ostate = inet->sk_state_change; | |
1136 | svsk->sk_odata = inet->sk_data_ready; | |
1137 | svsk->sk_owspace = inet->sk_write_space; | |
1da177e4 LT |
1138 | |
1139 | /* Initialize the socket */ | |
1140 | if (sock->type == SOCK_DGRAM) | |
bb5cf160 | 1141 | svc_udp_init(svsk, serv); |
47a14ef1 OK |
1142 | else { |
1143 | /* initialise setting must have enough space to | |
1144 | * receive and respond to one request. | |
1145 | */ | |
1146 | svc_sock_setbufsize(svsk->sk_sock, 4 * serv->sv_max_mesg, | |
1147 | 4 * serv->sv_max_mesg); | |
bb5cf160 | 1148 | svc_tcp_init(svsk, serv); |
47a14ef1 | 1149 | } |
1da177e4 | 1150 | |
1da177e4 LT |
1151 | dprintk("svc: svc_setup_socket created %p (inet %p)\n", |
1152 | svsk, svsk->sk_sk); | |
1153 | ||
1da177e4 LT |
1154 | return svsk; |
1155 | } | |
1156 | ||
bfba9ab4 CL |
1157 | /** |
1158 | * svc_addsock - add a listener socket to an RPC service | |
1159 | * @serv: pointer to RPC service to which to add a new listener | |
1160 | * @fd: file descriptor of the new listener | |
1161 | * @name_return: pointer to buffer to fill in with name of listener | |
1162 | * @len: size of the buffer | |
1163 | * | |
1164 | * Fills in socket name and returns positive length of name if successful. | |
1165 | * Name is terminated with '\n'. On error, returns a negative errno | |
1166 | * value. | |
1167 | */ | |
1168 | int svc_addsock(struct svc_serv *serv, const int fd, char *name_return, | |
1169 | const size_t len) | |
b41b66d6 N |
1170 | { |
1171 | int err = 0; | |
1172 | struct socket *so = sockfd_lookup(fd, &err); | |
1173 | struct svc_sock *svsk = NULL; | |
1174 | ||
1175 | if (!so) | |
1176 | return err; | |
1177 | if (so->sk->sk_family != AF_INET) | |
1178 | err = -EAFNOSUPPORT; | |
1179 | else if (so->sk->sk_protocol != IPPROTO_TCP && | |
1180 | so->sk->sk_protocol != IPPROTO_UDP) | |
1181 | err = -EPROTONOSUPPORT; | |
1182 | else if (so->state > SS_UNCONNECTED) | |
1183 | err = -EISCONN; | |
1184 | else { | |
2da2c21d TT |
1185 | if (!try_module_get(THIS_MODULE)) |
1186 | err = -ENOENT; | |
1187 | else | |
1188 | svsk = svc_setup_socket(serv, so, &err, | |
1189 | SVC_SOCK_DEFAULTS); | |
e79eff1f | 1190 | if (svsk) { |
9dbc240f TT |
1191 | struct sockaddr_storage addr; |
1192 | struct sockaddr *sin = (struct sockaddr *)&addr; | |
1193 | int salen; | |
1194 | if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0) | |
1195 | svc_xprt_set_local(&svsk->sk_xprt, sin, salen); | |
4e5caaa5 TT |
1196 | clear_bit(XPT_TEMP, &svsk->sk_xprt.xpt_flags); |
1197 | spin_lock_bh(&serv->sv_lock); | |
1198 | list_add(&svsk->sk_xprt.xpt_list, &serv->sv_permsocks); | |
1199 | spin_unlock_bh(&serv->sv_lock); | |
a6046f71 | 1200 | svc_xprt_received(&svsk->sk_xprt); |
b41b66d6 | 1201 | err = 0; |
2da2c21d TT |
1202 | } else |
1203 | module_put(THIS_MODULE); | |
b41b66d6 N |
1204 | } |
1205 | if (err) { | |
1206 | sockfd_put(so); | |
1207 | return err; | |
1208 | } | |
e7942b9f | 1209 | return svc_one_sock_name(svsk, name_return, len); |
b41b66d6 N |
1210 | } |
1211 | EXPORT_SYMBOL_GPL(svc_addsock); | |
1212 | ||
1da177e4 LT |
1213 | /* |
1214 | * Create socket for RPC service. | |
1215 | */ | |
b700cbb1 TT |
1216 | static struct svc_xprt *svc_create_socket(struct svc_serv *serv, |
1217 | int protocol, | |
1218 | struct sockaddr *sin, int len, | |
1219 | int flags) | |
1da177e4 LT |
1220 | { |
1221 | struct svc_sock *svsk; | |
1222 | struct socket *sock; | |
1223 | int error; | |
1224 | int type; | |
9dbc240f TT |
1225 | struct sockaddr_storage addr; |
1226 | struct sockaddr *newsin = (struct sockaddr *)&addr; | |
1227 | int newlen; | |
c69da774 TM |
1228 | int family; |
1229 | int val; | |
5216a8e7 | 1230 | RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); |
1da177e4 | 1231 | |
ad06e4bd CL |
1232 | dprintk("svc: svc_create_socket(%s, %d, %s)\n", |
1233 | serv->sv_program->pg_name, protocol, | |
77f1f67a | 1234 | __svc_print_addr(sin, buf, sizeof(buf))); |
1da177e4 LT |
1235 | |
1236 | if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) { | |
1237 | printk(KERN_WARNING "svc: only UDP and TCP " | |
1238 | "sockets supported\n"); | |
b700cbb1 | 1239 | return ERR_PTR(-EINVAL); |
1da177e4 | 1240 | } |
c69da774 | 1241 | |
1da177e4 | 1242 | type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM; |
c69da774 TM |
1243 | switch (sin->sa_family) { |
1244 | case AF_INET6: | |
1245 | family = PF_INET6; | |
1246 | break; | |
1247 | case AF_INET: | |
1248 | family = PF_INET; | |
1249 | break; | |
1250 | default: | |
1251 | return ERR_PTR(-EINVAL); | |
1252 | } | |
1da177e4 | 1253 | |
c69da774 | 1254 | error = sock_create_kern(family, type, protocol, &sock); |
77f1f67a | 1255 | if (error < 0) |
b700cbb1 | 1256 | return ERR_PTR(error); |
1da177e4 | 1257 | |
ed07536e PZ |
1258 | svc_reclassify_socket(sock); |
1259 | ||
c69da774 TM |
1260 | /* |
1261 | * If this is an PF_INET6 listener, we want to avoid | |
1262 | * getting requests from IPv4 remotes. Those should | |
1263 | * be shunted to a PF_INET listener via rpcbind. | |
1264 | */ | |
1265 | val = 1; | |
1266 | if (family == PF_INET6) | |
1267 | kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY, | |
1268 | (char *)&val, sizeof(val)); | |
1269 | ||
18114746 | 1270 | if (type == SOCK_STREAM) |
77f1f67a CL |
1271 | sock->sk->sk_reuse = 1; /* allow address reuse */ |
1272 | error = kernel_bind(sock, sin, len); | |
18114746 ES |
1273 | if (error < 0) |
1274 | goto bummer; | |
1da177e4 | 1275 | |
9dbc240f TT |
1276 | newlen = len; |
1277 | error = kernel_getsockname(sock, newsin, &newlen); | |
1278 | if (error < 0) | |
1279 | goto bummer; | |
1280 | ||
1da177e4 | 1281 | if (protocol == IPPROTO_TCP) { |
e6242e92 | 1282 | if ((error = kernel_listen(sock, 64)) < 0) |
1da177e4 LT |
1283 | goto bummer; |
1284 | } | |
1285 | ||
e79eff1f | 1286 | if ((svsk = svc_setup_socket(serv, sock, &error, flags)) != NULL) { |
9dbc240f | 1287 | svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen); |
b700cbb1 | 1288 | return (struct svc_xprt *)svsk; |
e79eff1f | 1289 | } |
1da177e4 LT |
1290 | |
1291 | bummer: | |
1292 | dprintk("svc: svc_create_socket error = %d\n", -error); | |
1293 | sock_release(sock); | |
b700cbb1 | 1294 | return ERR_PTR(error); |
1da177e4 LT |
1295 | } |
1296 | ||
755cceab TT |
1297 | /* |
1298 | * Detach the svc_sock from the socket so that no | |
1299 | * more callbacks occur. | |
1300 | */ | |
1301 | static void svc_sock_detach(struct svc_xprt *xprt) | |
1302 | { | |
1303 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
1304 | struct sock *sk = svsk->sk_sk; | |
1305 | ||
1306 | dprintk("svc: svc_sock_detach(%p)\n", svsk); | |
1307 | ||
1308 | /* put back the old socket callbacks */ | |
1309 | sk->sk_state_change = svsk->sk_ostate; | |
1310 | sk->sk_data_ready = svsk->sk_odata; | |
1311 | sk->sk_write_space = svsk->sk_owspace; | |
69b6ba37 TM |
1312 | |
1313 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) | |
1314 | wake_up_interruptible(sk->sk_sleep); | |
1315 | } | |
1316 | ||
1317 | /* | |
1318 | * Disconnect the socket, and reset the callbacks | |
1319 | */ | |
1320 | static void svc_tcp_sock_detach(struct svc_xprt *xprt) | |
1321 | { | |
1322 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
1323 | ||
1324 | dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk); | |
1325 | ||
1326 | svc_sock_detach(xprt); | |
1327 | ||
1328 | if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) | |
1329 | kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR); | |
755cceab TT |
1330 | } |
1331 | ||
1332 | /* | |
1333 | * Free the svc_sock's socket resources and the svc_sock itself. | |
1334 | */ | |
1335 | static void svc_sock_free(struct svc_xprt *xprt) | |
1336 | { | |
1337 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
1338 | dprintk("svc: svc_sock_free(%p)\n", svsk); | |
1339 | ||
755cceab TT |
1340 | if (svsk->sk_sock->file) |
1341 | sockfd_put(svsk->sk_sock); | |
1342 | else | |
1343 | sock_release(svsk->sk_sock); | |
1344 | kfree(svsk); | |
1345 | } |