]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/net/sunrpc/svcsock.c | |
3 | * | |
4 | * These are the RPC server socket internals. | |
5 | * | |
6 | * The server scheduling algorithm does not always distribute the load | |
7 | * evenly when servicing a single client. May need to modify the | |
f6150c3c | 8 | * svc_xprt_enqueue procedure... |
1da177e4 LT |
9 | * |
10 | * TCP support is largely untested and may be a little slow. The problem | |
11 | * is that we currently do two separate recvfrom's, one for the 4-byte | |
12 | * record length, and the second for the actual record. This could possibly | |
13 | * be improved by always reading a minimum size of around 100 bytes and | |
14 | * tucking any superfluous bytes away in a temporary store. Still, that | |
15 | * leaves write requests out in the rain. An alternative may be to peek at | |
16 | * the first skb in the queue, and if it matches the next TCP sequence | |
17 | * number, to extract the record marker. Yuck. | |
18 | * | |
19 | * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de> | |
20 | */ | |
21 | ||
172589cc | 22 | #include <linux/kernel.h> |
1da177e4 LT |
23 | #include <linux/sched.h> |
24 | #include <linux/errno.h> | |
25 | #include <linux/fcntl.h> | |
26 | #include <linux/net.h> | |
27 | #include <linux/in.h> | |
28 | #include <linux/inet.h> | |
29 | #include <linux/udp.h> | |
91483c4b | 30 | #include <linux/tcp.h> |
1da177e4 LT |
31 | #include <linux/unistd.h> |
32 | #include <linux/slab.h> | |
33 | #include <linux/netdevice.h> | |
34 | #include <linux/skbuff.h> | |
b41b66d6 | 35 | #include <linux/file.h> |
7dfb7103 | 36 | #include <linux/freezer.h> |
1da177e4 LT |
37 | #include <net/sock.h> |
38 | #include <net/checksum.h> | |
39 | #include <net/ip.h> | |
b92503b2 | 40 | #include <net/ipv6.h> |
b7872fe8 | 41 | #include <net/tcp.h> |
c752f073 | 42 | #include <net/tcp_states.h> |
1da177e4 LT |
43 | #include <asm/uaccess.h> |
44 | #include <asm/ioctls.h> | |
45 | ||
46 | #include <linux/sunrpc/types.h> | |
ad06e4bd | 47 | #include <linux/sunrpc/clnt.h> |
1da177e4 | 48 | #include <linux/sunrpc/xdr.h> |
c0401ea0 | 49 | #include <linux/sunrpc/msg_prot.h> |
1da177e4 LT |
50 | #include <linux/sunrpc/svcsock.h> |
51 | #include <linux/sunrpc/stats.h> | |
4cfc7e60 | 52 | #include <linux/sunrpc/xprt.h> |
1da177e4 | 53 | |
360d8738 | 54 | #define RPCDBG_FACILITY RPCDBG_SVCXPRT |
1da177e4 LT |
55 | |
56 | ||
57 | static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *, | |
6b174337 | 58 | int *errp, int flags); |
1da177e4 LT |
59 | static void svc_udp_data_ready(struct sock *, int); |
60 | static int svc_udp_recvfrom(struct svc_rqst *); | |
61 | static int svc_udp_sendto(struct svc_rqst *); | |
755cceab | 62 | static void svc_sock_detach(struct svc_xprt *); |
69b6ba37 | 63 | static void svc_tcp_sock_detach(struct svc_xprt *); |
755cceab | 64 | static void svc_sock_free(struct svc_xprt *); |
1da177e4 | 65 | |
b700cbb1 | 66 | static struct svc_xprt *svc_create_socket(struct svc_serv *, int, |
62832c03 PE |
67 | struct net *, struct sockaddr *, |
68 | int, int); | |
1f11a034 AA |
69 | #if defined(CONFIG_NFS_V4_1) |
70 | static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int, | |
71 | struct net *, struct sockaddr *, | |
72 | int, int); | |
73 | static void svc_bc_sock_free(struct svc_xprt *xprt); | |
74 | #endif /* CONFIG_NFS_V4_1 */ | |
75 | ||
ed07536e PZ |
76 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
77 | static struct lock_class_key svc_key[2]; | |
78 | static struct lock_class_key svc_slock_key[2]; | |
79 | ||
0f0257ea | 80 | static void svc_reclassify_socket(struct socket *sock) |
ed07536e PZ |
81 | { |
82 | struct sock *sk = sock->sk; | |
02b3d346 | 83 | BUG_ON(sock_owned_by_user(sk)); |
ed07536e PZ |
84 | switch (sk->sk_family) { |
85 | case AF_INET: | |
86 | sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD", | |
def13d74 TT |
87 | &svc_slock_key[0], |
88 | "sk_xprt.xpt_lock-AF_INET-NFSD", | |
89 | &svc_key[0]); | |
ed07536e PZ |
90 | break; |
91 | ||
92 | case AF_INET6: | |
93 | sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD", | |
def13d74 TT |
94 | &svc_slock_key[1], |
95 | "sk_xprt.xpt_lock-AF_INET6-NFSD", | |
96 | &svc_key[1]); | |
ed07536e PZ |
97 | break; |
98 | ||
99 | default: | |
100 | BUG(); | |
101 | } | |
102 | } | |
103 | #else | |
0f0257ea | 104 | static void svc_reclassify_socket(struct socket *sock) |
ed07536e PZ |
105 | { |
106 | } | |
107 | #endif | |
108 | ||
1da177e4 LT |
109 | /* |
110 | * Release an skbuff after use | |
111 | */ | |
5148bf4e | 112 | static void svc_release_skb(struct svc_rqst *rqstp) |
1da177e4 | 113 | { |
5148bf4e | 114 | struct sk_buff *skb = rqstp->rq_xprt_ctxt; |
1da177e4 LT |
115 | |
116 | if (skb) { | |
57b1d3ba TT |
117 | struct svc_sock *svsk = |
118 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
5148bf4e | 119 | rqstp->rq_xprt_ctxt = NULL; |
1da177e4 LT |
120 | |
121 | dprintk("svc: service %p, releasing skb %p\n", rqstp, skb); | |
9d410c79 | 122 | skb_free_datagram_locked(svsk->sk_sk, skb); |
1da177e4 | 123 | } |
1da177e4 LT |
124 | } |
125 | ||
b92503b2 CL |
126 | union svc_pktinfo_u { |
127 | struct in_pktinfo pkti; | |
b92503b2 | 128 | struct in6_pktinfo pkti6; |
b92503b2 | 129 | }; |
bc375ea7 DM |
130 | #define SVC_PKTINFO_SPACE \ |
131 | CMSG_SPACE(sizeof(union svc_pktinfo_u)) | |
b92503b2 CL |
132 | |
133 | static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh) | |
134 | { | |
57b1d3ba TT |
135 | struct svc_sock *svsk = |
136 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
137 | switch (svsk->sk_sk->sk_family) { | |
b92503b2 CL |
138 | case AF_INET: { |
139 | struct in_pktinfo *pki = CMSG_DATA(cmh); | |
140 | ||
141 | cmh->cmsg_level = SOL_IP; | |
142 | cmh->cmsg_type = IP_PKTINFO; | |
143 | pki->ipi_ifindex = 0; | |
144 | pki->ipi_spec_dst.s_addr = rqstp->rq_daddr.addr.s_addr; | |
145 | cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); | |
146 | } | |
147 | break; | |
5a05ed73 | 148 | |
b92503b2 CL |
149 | case AF_INET6: { |
150 | struct in6_pktinfo *pki = CMSG_DATA(cmh); | |
151 | ||
152 | cmh->cmsg_level = SOL_IPV6; | |
153 | cmh->cmsg_type = IPV6_PKTINFO; | |
154 | pki->ipi6_ifindex = 0; | |
155 | ipv6_addr_copy(&pki->ipi6_addr, | |
156 | &rqstp->rq_daddr.addr6); | |
157 | cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); | |
158 | } | |
159 | break; | |
b92503b2 | 160 | } |
b92503b2 CL |
161 | } |
162 | ||
1da177e4 | 163 | /* |
4cfc7e60 | 164 | * send routine intended to be shared by the fore- and back-channel |
1da177e4 | 165 | */ |
4cfc7e60 RI |
166 | int svc_send_common(struct socket *sock, struct xdr_buf *xdr, |
167 | struct page *headpage, unsigned long headoffset, | |
168 | struct page *tailpage, unsigned long tailoffset) | |
1da177e4 | 169 | { |
1da177e4 LT |
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; | |
4cfc7e60 RI |
176 | int slen; |
177 | int len = 0; | |
1da177e4 LT |
178 | |
179 | slen = xdr->len; | |
180 | ||
1da177e4 LT |
181 | /* send head */ |
182 | if (slen == xdr->head[0].iov_len) | |
183 | flags = 0; | |
4cfc7e60 | 184 | len = kernel_sendpage(sock, headpage, headoffset, |
44524359 | 185 | xdr->head[0].iov_len, flags); |
1da177e4 LT |
186 | if (len != xdr->head[0].iov_len) |
187 | goto out; | |
188 | slen -= xdr->head[0].iov_len; | |
189 | if (slen == 0) | |
190 | goto out; | |
191 | ||
192 | /* send page data */ | |
193 | size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen; | |
194 | while (pglen > 0) { | |
195 | if (slen == size) | |
196 | flags = 0; | |
e6242e92 | 197 | result = kernel_sendpage(sock, *ppage, base, size, flags); |
1da177e4 LT |
198 | if (result > 0) |
199 | len += result; | |
200 | if (result != size) | |
201 | goto out; | |
202 | slen -= size; | |
203 | pglen -= size; | |
204 | size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen; | |
205 | base = 0; | |
206 | ppage++; | |
207 | } | |
4cfc7e60 | 208 | |
1da177e4 LT |
209 | /* send tail */ |
210 | if (xdr->tail[0].iov_len) { | |
4cfc7e60 RI |
211 | result = kernel_sendpage(sock, tailpage, tailoffset, |
212 | xdr->tail[0].iov_len, 0); | |
1da177e4 LT |
213 | if (result > 0) |
214 | len += result; | |
215 | } | |
4cfc7e60 RI |
216 | |
217 | out: | |
218 | return len; | |
219 | } | |
220 | ||
221 | ||
222 | /* | |
223 | * Generic sendto routine | |
224 | */ | |
225 | static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr) | |
226 | { | |
227 | struct svc_sock *svsk = | |
228 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
229 | struct socket *sock = svsk->sk_sock; | |
230 | union { | |
231 | struct cmsghdr hdr; | |
232 | long all[SVC_PKTINFO_SPACE / sizeof(long)]; | |
233 | } buffer; | |
234 | struct cmsghdr *cmh = &buffer.hdr; | |
235 | int len = 0; | |
236 | unsigned long tailoff; | |
237 | unsigned long headoff; | |
238 | RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); | |
239 | ||
240 | if (rqstp->rq_prot == IPPROTO_UDP) { | |
241 | struct msghdr msg = { | |
242 | .msg_name = &rqstp->rq_addr, | |
243 | .msg_namelen = rqstp->rq_addrlen, | |
244 | .msg_control = cmh, | |
245 | .msg_controllen = sizeof(buffer), | |
246 | .msg_flags = MSG_MORE, | |
247 | }; | |
248 | ||
249 | svc_set_cmsg_data(rqstp, cmh); | |
250 | ||
251 | if (sock_sendmsg(sock, &msg, 0) < 0) | |
252 | goto out; | |
253 | } | |
254 | ||
255 | tailoff = ((unsigned long)xdr->tail[0].iov_base) & (PAGE_SIZE-1); | |
256 | headoff = 0; | |
257 | len = svc_send_common(sock, xdr, rqstp->rq_respages[0], headoff, | |
258 | rqstp->rq_respages[0], tailoff); | |
259 | ||
1da177e4 | 260 | out: |
ad06e4bd | 261 | dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n", |
57b1d3ba | 262 | svsk, xdr->head[0].iov_base, xdr->head[0].iov_len, |
ad06e4bd | 263 | xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf))); |
1da177e4 LT |
264 | |
265 | return len; | |
266 | } | |
267 | ||
80212d59 N |
268 | /* |
269 | * Report socket names for nfsdfs | |
270 | */ | |
e7942b9f | 271 | static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining) |
80212d59 | 272 | { |
017cb47f CL |
273 | const struct sock *sk = svsk->sk_sk; |
274 | const char *proto_name = sk->sk_protocol == IPPROTO_UDP ? | |
275 | "udp" : "tcp"; | |
80212d59 N |
276 | int len; |
277 | ||
017cb47f | 278 | switch (sk->sk_family) { |
e7942b9f CL |
279 | case PF_INET: |
280 | len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n", | |
017cb47f | 281 | proto_name, |
c720c7e8 ED |
282 | &inet_sk(sk)->inet_rcv_saddr, |
283 | inet_sk(sk)->inet_num); | |
80212d59 | 284 | break; |
58de2f86 CL |
285 | case PF_INET6: |
286 | len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n", | |
017cb47f CL |
287 | proto_name, |
288 | &inet6_sk(sk)->rcv_saddr, | |
c720c7e8 | 289 | inet_sk(sk)->inet_num); |
80212d59 N |
290 | break; |
291 | default: | |
e7942b9f | 292 | len = snprintf(buf, remaining, "*unknown-%d*\n", |
017cb47f | 293 | sk->sk_family); |
80212d59 | 294 | } |
e7942b9f CL |
295 | |
296 | if (len >= remaining) { | |
297 | *buf = '\0'; | |
298 | return -ENAMETOOLONG; | |
80212d59 N |
299 | } |
300 | return len; | |
301 | } | |
302 | ||
8435d34d CL |
303 | /** |
304 | * svc_sock_names - construct a list of listener names in a string | |
305 | * @serv: pointer to RPC service | |
306 | * @buf: pointer to a buffer to fill in with socket names | |
307 | * @buflen: size of the buffer to be filled | |
308 | * @toclose: pointer to '\0'-terminated C string containing the name | |
309 | * of a listener to be closed | |
310 | * | |
311 | * Fills in @buf with a '\n'-separated list of names of listener | |
312 | * sockets. If @toclose is not NULL, the socket named by @toclose | |
313 | * is closed, and is not included in the output list. | |
314 | * | |
315 | * Returns positive length of the socket name string, or a negative | |
316 | * errno value on error. | |
317 | */ | |
318 | int svc_sock_names(struct svc_serv *serv, char *buf, const size_t buflen, | |
319 | const char *toclose) | |
80212d59 | 320 | { |
b41b66d6 | 321 | struct svc_sock *svsk, *closesk = NULL; |
80212d59 N |
322 | int len = 0; |
323 | ||
324 | if (!serv) | |
325 | return 0; | |
e7942b9f | 326 | |
aaf68cfb | 327 | spin_lock_bh(&serv->sv_lock); |
7a182083 | 328 | list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list) { |
e7942b9f CL |
329 | int onelen = svc_one_sock_name(svsk, buf + len, buflen - len); |
330 | if (onelen < 0) { | |
331 | len = onelen; | |
332 | break; | |
333 | } | |
3942302e | 334 | if (toclose && strcmp(toclose, buf + len) == 0) { |
b41b66d6 | 335 | closesk = svsk; |
3942302e N |
336 | svc_xprt_get(&closesk->sk_xprt); |
337 | } else | |
b41b66d6 | 338 | len += onelen; |
80212d59 | 339 | } |
aaf68cfb | 340 | spin_unlock_bh(&serv->sv_lock); |
e7942b9f | 341 | |
3942302e | 342 | if (closesk) { |
5680c446 N |
343 | /* Should unregister with portmap, but you cannot |
344 | * unregister just one protocol... | |
345 | */ | |
7a182083 | 346 | svc_close_xprt(&closesk->sk_xprt); |
3942302e N |
347 | svc_xprt_put(&closesk->sk_xprt); |
348 | } else if (toclose) | |
37a03472 | 349 | return -ENOENT; |
80212d59 N |
350 | return len; |
351 | } | |
24c3767e | 352 | EXPORT_SYMBOL_GPL(svc_sock_names); |
80212d59 | 353 | |
1da177e4 LT |
354 | /* |
355 | * Check input queue length | |
356 | */ | |
0f0257ea | 357 | static int svc_recv_available(struct svc_sock *svsk) |
1da177e4 | 358 | { |
1da177e4 LT |
359 | struct socket *sock = svsk->sk_sock; |
360 | int avail, err; | |
361 | ||
e6242e92 | 362 | err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail); |
1da177e4 LT |
363 | |
364 | return (err >= 0)? avail : err; | |
365 | } | |
366 | ||
367 | /* | |
368 | * Generic recvfrom routine. | |
369 | */ | |
0f0257ea TT |
370 | static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr, |
371 | int buflen) | |
1da177e4 | 372 | { |
57b1d3ba TT |
373 | struct svc_sock *svsk = |
374 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
1ba95105 CL |
375 | struct msghdr msg = { |
376 | .msg_flags = MSG_DONTWAIT, | |
377 | }; | |
378 | int len; | |
1da177e4 | 379 | |
260c1d12 TT |
380 | rqstp->rq_xprt_hlen = 0; |
381 | ||
1ba95105 CL |
382 | len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen, |
383 | msg.msg_flags); | |
1da177e4 | 384 | |
1da177e4 | 385 | dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n", |
1ba95105 | 386 | svsk, iov[0].iov_base, iov[0].iov_len, len); |
1da177e4 LT |
387 | return len; |
388 | } | |
389 | ||
390 | /* | |
391 | * Set socket snd and rcv buffer lengths | |
392 | */ | |
0f0257ea TT |
393 | static void svc_sock_setbufsize(struct socket *sock, unsigned int snd, |
394 | unsigned int rcv) | |
1da177e4 LT |
395 | { |
396 | #if 0 | |
397 | mm_segment_t oldfs; | |
398 | oldfs = get_fs(); set_fs(KERNEL_DS); | |
399 | sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF, | |
400 | (char*)&snd, sizeof(snd)); | |
401 | sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF, | |
402 | (char*)&rcv, sizeof(rcv)); | |
403 | #else | |
404 | /* sock_setsockopt limits use to sysctl_?mem_max, | |
405 | * which isn't acceptable. Until that is made conditional | |
406 | * on not having CAP_SYS_RESOURCE or similar, we go direct... | |
407 | * DaveM said I could! | |
408 | */ | |
409 | lock_sock(sock->sk); | |
410 | sock->sk->sk_sndbuf = snd * 2; | |
411 | sock->sk->sk_rcvbuf = rcv * 2; | |
7f421835 | 412 | sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK|SOCK_RCVBUF_LOCK; |
47fcb03f | 413 | sock->sk->sk_write_space(sock->sk); |
1da177e4 LT |
414 | release_sock(sock->sk); |
415 | #endif | |
416 | } | |
417 | /* | |
418 | * INET callback when data has been received on the socket. | |
419 | */ | |
0f0257ea | 420 | static void svc_udp_data_ready(struct sock *sk, int count) |
1da177e4 | 421 | { |
939bb7ef | 422 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
eaefd110 | 423 | wait_queue_head_t *wq = sk_sleep(sk); |
1da177e4 | 424 | |
939bb7ef NB |
425 | if (svsk) { |
426 | dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n", | |
02fc6c36 TT |
427 | svsk, sk, count, |
428 | test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags)); | |
429 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); | |
f6150c3c | 430 | svc_xprt_enqueue(&svsk->sk_xprt); |
939bb7ef | 431 | } |
eaefd110 ED |
432 | if (wq && waitqueue_active(wq)) |
433 | wake_up_interruptible(wq); | |
1da177e4 LT |
434 | } |
435 | ||
436 | /* | |
437 | * INET callback when space is newly available on the socket. | |
438 | */ | |
0f0257ea | 439 | static void svc_write_space(struct sock *sk) |
1da177e4 LT |
440 | { |
441 | struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data); | |
eaefd110 | 442 | wait_queue_head_t *wq = sk_sleep(sk); |
1da177e4 LT |
443 | |
444 | if (svsk) { | |
445 | dprintk("svc: socket %p(inet %p), write_space busy=%d\n", | |
02fc6c36 | 446 | svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags)); |
f6150c3c | 447 | svc_xprt_enqueue(&svsk->sk_xprt); |
1da177e4 LT |
448 | } |
449 | ||
eaefd110 | 450 | if (wq && waitqueue_active(wq)) { |
939bb7ef | 451 | dprintk("RPC svc_write_space: someone sleeping on %p\n", |
1da177e4 | 452 | svsk); |
eaefd110 | 453 | wake_up_interruptible(wq); |
1da177e4 LT |
454 | } |
455 | } | |
456 | ||
47fcb03f TM |
457 | static void svc_tcp_write_space(struct sock *sk) |
458 | { | |
459 | struct socket *sock = sk->sk_socket; | |
460 | ||
461 | if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && sock) | |
462 | clear_bit(SOCK_NOSPACE, &sock->flags); | |
463 | svc_write_space(sk); | |
464 | } | |
465 | ||
7702ce40 CL |
466 | /* |
467 | * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo | |
468 | */ | |
469 | static int svc_udp_get_dest_address4(struct svc_rqst *rqstp, | |
470 | struct cmsghdr *cmh) | |
471 | { | |
472 | struct in_pktinfo *pki = CMSG_DATA(cmh); | |
473 | if (cmh->cmsg_type != IP_PKTINFO) | |
474 | return 0; | |
475 | rqstp->rq_daddr.addr.s_addr = pki->ipi_spec_dst.s_addr; | |
476 | return 1; | |
477 | } | |
478 | ||
479 | /* | |
480 | * See net/ipv6/datagram.c : datagram_recv_ctl | |
481 | */ | |
482 | static int svc_udp_get_dest_address6(struct svc_rqst *rqstp, | |
483 | struct cmsghdr *cmh) | |
484 | { | |
485 | struct in6_pktinfo *pki = CMSG_DATA(cmh); | |
486 | if (cmh->cmsg_type != IPV6_PKTINFO) | |
487 | return 0; | |
488 | ipv6_addr_copy(&rqstp->rq_daddr.addr6, &pki->ipi6_addr); | |
489 | return 1; | |
490 | } | |
491 | ||
9dbc240f TT |
492 | /* |
493 | * Copy the UDP datagram's destination address to the rqstp structure. | |
494 | * The 'destination' address in this case is the address to which the | |
495 | * peer sent the datagram, i.e. our local address. For multihomed | |
496 | * hosts, this can change from msg to msg. Note that only the IP | |
497 | * address changes, the port number should remain the same. | |
498 | */ | |
7702ce40 CL |
499 | static int svc_udp_get_dest_address(struct svc_rqst *rqstp, |
500 | struct cmsghdr *cmh) | |
95756482 | 501 | { |
7702ce40 CL |
502 | switch (cmh->cmsg_level) { |
503 | case SOL_IP: | |
504 | return svc_udp_get_dest_address4(rqstp, cmh); | |
505 | case SOL_IPV6: | |
506 | return svc_udp_get_dest_address6(rqstp, cmh); | |
95756482 | 507 | } |
7702ce40 CL |
508 | |
509 | return 0; | |
95756482 CL |
510 | } |
511 | ||
1da177e4 LT |
512 | /* |
513 | * Receive a datagram from a UDP socket. | |
514 | */ | |
0f0257ea | 515 | static int svc_udp_recvfrom(struct svc_rqst *rqstp) |
1da177e4 | 516 | { |
57b1d3ba TT |
517 | struct svc_sock *svsk = |
518 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
bb5cf160 | 519 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
1da177e4 | 520 | struct sk_buff *skb; |
bc375ea7 DM |
521 | union { |
522 | struct cmsghdr hdr; | |
523 | long all[SVC_PKTINFO_SPACE / sizeof(long)]; | |
524 | } buffer; | |
525 | struct cmsghdr *cmh = &buffer.hdr; | |
7a37f578 N |
526 | struct msghdr msg = { |
527 | .msg_name = svc_addr(rqstp), | |
528 | .msg_control = cmh, | |
529 | .msg_controllen = sizeof(buffer), | |
530 | .msg_flags = MSG_DONTWAIT, | |
531 | }; | |
abc5c44d CL |
532 | size_t len; |
533 | int err; | |
1da177e4 | 534 | |
02fc6c36 | 535 | if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags)) |
1da177e4 LT |
536 | /* udp sockets need large rcvbuf as all pending |
537 | * requests are still in that buffer. sndbuf must | |
538 | * also be large enough that there is enough space | |
3262c816 GB |
539 | * for one reply per thread. We count all threads |
540 | * rather than threads in a particular pool, which | |
541 | * provides an upper bound on the number of threads | |
542 | * which will access the socket. | |
1da177e4 LT |
543 | */ |
544 | svc_sock_setbufsize(svsk->sk_sock, | |
c6b0a9f8 N |
545 | (serv->sv_nrthreads+3) * serv->sv_max_mesg, |
546 | (serv->sv_nrthreads+3) * serv->sv_max_mesg); | |
1da177e4 | 547 | |
02fc6c36 | 548 | clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
05ed690e N |
549 | skb = NULL; |
550 | err = kernel_recvmsg(svsk->sk_sock, &msg, NULL, | |
551 | 0, 0, MSG_PEEK | MSG_DONTWAIT); | |
552 | if (err >= 0) | |
553 | skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err); | |
554 | ||
555 | if (skb == NULL) { | |
556 | if (err != -EAGAIN) { | |
557 | /* possibly an icmp error */ | |
558 | dprintk("svc: recvfrom returned error %d\n", -err); | |
02fc6c36 | 559 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 560 | } |
05ed690e | 561 | return -EAGAIN; |
1da177e4 | 562 | } |
9dbc240f | 563 | len = svc_addr_len(svc_addr(rqstp)); |
abc5c44d CL |
564 | if (len == 0) |
565 | return -EAFNOSUPPORT; | |
9dbc240f | 566 | rqstp->rq_addrlen = len; |
b7aa0bf7 ED |
567 | if (skb->tstamp.tv64 == 0) { |
568 | skb->tstamp = ktime_get_real(); | |
cca5172a | 569 | /* Don't enable netstamp, sunrpc doesn't |
1da177e4 LT |
570 | need that much accuracy */ |
571 | } | |
b7aa0bf7 | 572 | svsk->sk_sk->sk_stamp = skb->tstamp; |
02fc6c36 | 573 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */ |
1da177e4 | 574 | |
1da177e4 LT |
575 | len = skb->len - sizeof(struct udphdr); |
576 | rqstp->rq_arg.len = len; | |
577 | ||
95756482 | 578 | rqstp->rq_prot = IPPROTO_UDP; |
27459f09 | 579 | |
7702ce40 | 580 | if (!svc_udp_get_dest_address(rqstp, cmh)) { |
7a37f578 | 581 | if (net_ratelimit()) |
7702ce40 CL |
582 | printk(KERN_WARNING |
583 | "svc: received unknown control message %d/%d; " | |
584 | "dropping RPC reply datagram\n", | |
585 | cmh->cmsg_level, cmh->cmsg_type); | |
9d410c79 | 586 | skb_free_datagram_locked(svsk->sk_sk, skb); |
7a37f578 N |
587 | return 0; |
588 | } | |
1da177e4 LT |
589 | |
590 | if (skb_is_nonlinear(skb)) { | |
591 | /* we have to copy */ | |
592 | local_bh_disable(); | |
593 | if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) { | |
594 | local_bh_enable(); | |
595 | /* checksum error */ | |
9d410c79 | 596 | skb_free_datagram_locked(svsk->sk_sk, skb); |
1da177e4 LT |
597 | return 0; |
598 | } | |
599 | local_bh_enable(); | |
9d410c79 | 600 | skb_free_datagram_locked(svsk->sk_sk, skb); |
1da177e4 LT |
601 | } else { |
602 | /* we can use it in-place */ | |
0f0257ea TT |
603 | rqstp->rq_arg.head[0].iov_base = skb->data + |
604 | sizeof(struct udphdr); | |
1da177e4 | 605 | rqstp->rq_arg.head[0].iov_len = len; |
fb286bb2 | 606 | if (skb_checksum_complete(skb)) { |
9d410c79 | 607 | skb_free_datagram_locked(svsk->sk_sk, skb); |
fb286bb2 | 608 | return 0; |
1da177e4 | 609 | } |
5148bf4e | 610 | rqstp->rq_xprt_ctxt = skb; |
1da177e4 LT |
611 | } |
612 | ||
613 | rqstp->rq_arg.page_base = 0; | |
614 | if (len <= rqstp->rq_arg.head[0].iov_len) { | |
615 | rqstp->rq_arg.head[0].iov_len = len; | |
616 | rqstp->rq_arg.page_len = 0; | |
44524359 | 617 | rqstp->rq_respages = rqstp->rq_pages+1; |
1da177e4 LT |
618 | } else { |
619 | rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len; | |
44524359 | 620 | rqstp->rq_respages = rqstp->rq_pages + 1 + |
172589cc | 621 | DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE); |
1da177e4 LT |
622 | } |
623 | ||
624 | if (serv->sv_stats) | |
625 | serv->sv_stats->netudpcnt++; | |
626 | ||
627 | return len; | |
628 | } | |
629 | ||
630 | static int | |
631 | svc_udp_sendto(struct svc_rqst *rqstp) | |
632 | { | |
633 | int error; | |
634 | ||
635 | error = svc_sendto(rqstp, &rqstp->rq_res); | |
636 | if (error == -ECONNREFUSED) | |
637 | /* ICMP error on earlier request. */ | |
638 | error = svc_sendto(rqstp, &rqstp->rq_res); | |
639 | ||
640 | return error; | |
641 | } | |
642 | ||
e831fe65 TT |
643 | static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp) |
644 | { | |
645 | } | |
646 | ||
323bee32 TT |
647 | static int svc_udp_has_wspace(struct svc_xprt *xprt) |
648 | { | |
649 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
bb5cf160 | 650 | struct svc_serv *serv = xprt->xpt_server; |
323bee32 TT |
651 | unsigned long required; |
652 | ||
653 | /* | |
654 | * Set the SOCK_NOSPACE flag before checking the available | |
655 | * sock space. | |
656 | */ | |
657 | set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
7a90e8cc | 658 | required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg; |
323bee32 TT |
659 | if (required*2 > sock_wspace(svsk->sk_sk)) |
660 | return 0; | |
661 | clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
662 | return 1; | |
663 | } | |
664 | ||
38a417cc TT |
665 | static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt) |
666 | { | |
667 | BUG(); | |
668 | return NULL; | |
669 | } | |
670 | ||
b700cbb1 | 671 | static struct svc_xprt *svc_udp_create(struct svc_serv *serv, |
62832c03 | 672 | struct net *net, |
b700cbb1 TT |
673 | struct sockaddr *sa, int salen, |
674 | int flags) | |
675 | { | |
62832c03 | 676 | return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags); |
b700cbb1 TT |
677 | } |
678 | ||
360d8738 | 679 | static struct svc_xprt_ops svc_udp_ops = { |
b700cbb1 | 680 | .xpo_create = svc_udp_create, |
5d137990 TT |
681 | .xpo_recvfrom = svc_udp_recvfrom, |
682 | .xpo_sendto = svc_udp_sendto, | |
5148bf4e | 683 | .xpo_release_rqst = svc_release_skb, |
755cceab TT |
684 | .xpo_detach = svc_sock_detach, |
685 | .xpo_free = svc_sock_free, | |
e831fe65 | 686 | .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr, |
323bee32 | 687 | .xpo_has_wspace = svc_udp_has_wspace, |
38a417cc | 688 | .xpo_accept = svc_udp_accept, |
360d8738 TT |
689 | }; |
690 | ||
691 | static struct svc_xprt_class svc_udp_class = { | |
692 | .xcl_name = "udp", | |
b700cbb1 | 693 | .xcl_owner = THIS_MODULE, |
360d8738 | 694 | .xcl_ops = &svc_udp_ops, |
49023155 | 695 | .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP, |
360d8738 TT |
696 | }; |
697 | ||
bb5cf160 | 698 | static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv) |
1da177e4 | 699 | { |
7702ce40 | 700 | int err, level, optname, one = 1; |
7a37f578 | 701 | |
bb5cf160 | 702 | svc_xprt_init(&svc_udp_class, &svsk->sk_xprt, serv); |
def13d74 | 703 | clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
704 | svsk->sk_sk->sk_data_ready = svc_udp_data_ready; |
705 | svsk->sk_sk->sk_write_space = svc_write_space; | |
1da177e4 LT |
706 | |
707 | /* initialise setting must have enough space to | |
cca5172a | 708 | * receive and respond to one request. |
1da177e4 LT |
709 | * svc_udp_recvfrom will re-adjust if necessary |
710 | */ | |
711 | svc_sock_setbufsize(svsk->sk_sock, | |
bb5cf160 TT |
712 | 3 * svsk->sk_xprt.xpt_server->sv_max_mesg, |
713 | 3 * svsk->sk_xprt.xpt_server->sv_max_mesg); | |
1da177e4 | 714 | |
0f0257ea TT |
715 | /* data might have come in before data_ready set up */ |
716 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); | |
02fc6c36 | 717 | set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); |
7a37f578 | 718 | |
7a37f578 | 719 | /* make sure we get destination address info */ |
7702ce40 CL |
720 | switch (svsk->sk_sk->sk_family) { |
721 | case AF_INET: | |
722 | level = SOL_IP; | |
723 | optname = IP_PKTINFO; | |
724 | break; | |
725 | case AF_INET6: | |
726 | level = SOL_IPV6; | |
727 | optname = IPV6_RECVPKTINFO; | |
728 | break; | |
729 | default: | |
730 | BUG(); | |
731 | } | |
732 | err = kernel_setsockopt(svsk->sk_sock, level, optname, | |
733 | (char *)&one, sizeof(one)); | |
734 | dprintk("svc: kernel_setsockopt returned %d\n", err); | |
1da177e4 LT |
735 | } |
736 | ||
737 | /* | |
738 | * A data_ready event on a listening socket means there's a connection | |
739 | * pending. Do not use state_change as a substitute for it. | |
740 | */ | |
0f0257ea | 741 | static void svc_tcp_listen_data_ready(struct sock *sk, int count_unused) |
1da177e4 | 742 | { |
939bb7ef | 743 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
eaefd110 | 744 | wait_queue_head_t *wq; |
1da177e4 LT |
745 | |
746 | dprintk("svc: socket %p TCP (listen) state change %d\n", | |
939bb7ef | 747 | sk, sk->sk_state); |
1da177e4 | 748 | |
939bb7ef NB |
749 | /* |
750 | * This callback may called twice when a new connection | |
751 | * is established as a child socket inherits everything | |
752 | * from a parent LISTEN socket. | |
753 | * 1) data_ready method of the parent socket will be called | |
754 | * when one of child sockets become ESTABLISHED. | |
755 | * 2) data_ready method of the child socket may be called | |
756 | * when it receives data before the socket is accepted. | |
757 | * In case of 2, we should ignore it silently. | |
758 | */ | |
759 | if (sk->sk_state == TCP_LISTEN) { | |
760 | if (svsk) { | |
02fc6c36 | 761 | set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
f6150c3c | 762 | svc_xprt_enqueue(&svsk->sk_xprt); |
939bb7ef NB |
763 | } else |
764 | printk("svc: socket %p: no user data\n", sk); | |
1da177e4 | 765 | } |
939bb7ef | 766 | |
eaefd110 ED |
767 | wq = sk_sleep(sk); |
768 | if (wq && waitqueue_active(wq)) | |
769 | wake_up_interruptible_all(wq); | |
1da177e4 LT |
770 | } |
771 | ||
772 | /* | |
773 | * A state change on a connected socket means it's dying or dead. | |
774 | */ | |
0f0257ea | 775 | static void svc_tcp_state_change(struct sock *sk) |
1da177e4 | 776 | { |
939bb7ef | 777 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
eaefd110 | 778 | wait_queue_head_t *wq = sk_sleep(sk); |
1da177e4 LT |
779 | |
780 | dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n", | |
939bb7ef | 781 | sk, sk->sk_state, sk->sk_user_data); |
1da177e4 | 782 | |
939bb7ef | 783 | if (!svsk) |
1da177e4 | 784 | printk("svc: socket %p: no user data\n", sk); |
939bb7ef | 785 | else { |
02fc6c36 | 786 | set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); |
f6150c3c | 787 | svc_xprt_enqueue(&svsk->sk_xprt); |
1da177e4 | 788 | } |
eaefd110 ED |
789 | if (wq && waitqueue_active(wq)) |
790 | wake_up_interruptible_all(wq); | |
1da177e4 LT |
791 | } |
792 | ||
0f0257ea | 793 | static void svc_tcp_data_ready(struct sock *sk, int count) |
1da177e4 | 794 | { |
939bb7ef | 795 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
eaefd110 | 796 | wait_queue_head_t *wq = sk_sleep(sk); |
1da177e4 LT |
797 | |
798 | dprintk("svc: socket %p TCP data ready (svsk %p)\n", | |
939bb7ef NB |
799 | sk, sk->sk_user_data); |
800 | if (svsk) { | |
02fc6c36 | 801 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
f6150c3c | 802 | svc_xprt_enqueue(&svsk->sk_xprt); |
939bb7ef | 803 | } |
eaefd110 ED |
804 | if (wq && waitqueue_active(wq)) |
805 | wake_up_interruptible(wq); | |
1da177e4 LT |
806 | } |
807 | ||
808 | /* | |
809 | * Accept a TCP connection | |
810 | */ | |
38a417cc | 811 | static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt) |
1da177e4 | 812 | { |
38a417cc | 813 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); |
cdd88b9f AM |
814 | struct sockaddr_storage addr; |
815 | struct sockaddr *sin = (struct sockaddr *) &addr; | |
bb5cf160 | 816 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
1da177e4 LT |
817 | struct socket *sock = svsk->sk_sock; |
818 | struct socket *newsock; | |
1da177e4 LT |
819 | struct svc_sock *newsvsk; |
820 | int err, slen; | |
5216a8e7 | 821 | RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); |
1da177e4 LT |
822 | |
823 | dprintk("svc: tcp_accept %p sock %p\n", svsk, sock); | |
824 | if (!sock) | |
38a417cc | 825 | return NULL; |
1da177e4 | 826 | |
02fc6c36 | 827 | clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
e6242e92 SS |
828 | err = kernel_accept(sock, &newsock, O_NONBLOCK); |
829 | if (err < 0) { | |
1da177e4 LT |
830 | if (err == -ENOMEM) |
831 | printk(KERN_WARNING "%s: no more sockets!\n", | |
832 | serv->sv_name); | |
e6242e92 | 833 | else if (err != -EAGAIN && net_ratelimit()) |
1da177e4 LT |
834 | printk(KERN_WARNING "%s: accept failed (err %d)!\n", |
835 | serv->sv_name, -err); | |
38a417cc | 836 | return NULL; |
1da177e4 | 837 | } |
02fc6c36 | 838 | set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 839 | |
cdd88b9f | 840 | err = kernel_getpeername(newsock, sin, &slen); |
1da177e4 LT |
841 | if (err < 0) { |
842 | if (net_ratelimit()) | |
843 | printk(KERN_WARNING "%s: peername failed (err %d)!\n", | |
844 | serv->sv_name, -err); | |
845 | goto failed; /* aborted connection or whatever */ | |
846 | } | |
847 | ||
848 | /* Ideally, we would want to reject connections from unauthorized | |
ad06e4bd CL |
849 | * hosts here, but when we get encryption, the IP of the host won't |
850 | * tell us anything. For now just warn about unpriv connections. | |
1da177e4 | 851 | */ |
cdd88b9f | 852 | if (!svc_port_is_privileged(sin)) { |
1da177e4 | 853 | dprintk(KERN_WARNING |
ad06e4bd | 854 | "%s: connect from unprivileged port: %s\n", |
cca5172a | 855 | serv->sv_name, |
cdd88b9f | 856 | __svc_print_addr(sin, buf, sizeof(buf))); |
1da177e4 | 857 | } |
ad06e4bd | 858 | dprintk("%s: connect from %s\n", serv->sv_name, |
cdd88b9f | 859 | __svc_print_addr(sin, buf, sizeof(buf))); |
1da177e4 LT |
860 | |
861 | /* make sure that a write doesn't block forever when | |
862 | * low on memory | |
863 | */ | |
864 | newsock->sk->sk_sndtimeo = HZ*30; | |
865 | ||
6b174337 CL |
866 | if (!(newsvsk = svc_setup_socket(serv, newsock, &err, |
867 | (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY)))) | |
1da177e4 | 868 | goto failed; |
9dbc240f | 869 | svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen); |
a9747692 FM |
870 | err = kernel_getsockname(newsock, sin, &slen); |
871 | if (unlikely(err < 0)) { | |
872 | dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err); | |
873 | slen = offsetof(struct sockaddr, sa_data); | |
874 | } | |
9dbc240f | 875 | svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen); |
067d7817 | 876 | |
f9f3cc4f TT |
877 | if (serv->sv_stats) |
878 | serv->sv_stats->nettcpconn++; | |
879 | ||
880 | return &newsvsk->sk_xprt; | |
881 | ||
882 | failed: | |
883 | sock_release(newsock); | |
884 | return NULL; | |
885 | } | |
886 | ||
1da177e4 | 887 | /* |
8f55f3c0 AB |
888 | * Receive data. |
889 | * If we haven't gotten the record length yet, get the next four bytes. | |
890 | * Otherwise try to gobble up as much as possible up to the complete | |
891 | * record length. | |
1da177e4 | 892 | */ |
8f55f3c0 | 893 | static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp) |
1da177e4 | 894 | { |
bb5cf160 | 895 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
5ee78d48 | 896 | unsigned int want; |
8f55f3c0 | 897 | int len; |
1da177e4 | 898 | |
7f421835 BF |
899 | if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags)) |
900 | /* sndbuf needs to have room for one request | |
901 | * per thread, otherwise we can stall even when the | |
902 | * network isn't a bottleneck. | |
903 | * | |
904 | * We count all threads rather than threads in a | |
905 | * particular pool, which provides an upper bound | |
906 | * on the number of threads which will access the socket. | |
907 | * | |
908 | * rcvbuf just needs to be able to hold a few requests. | |
909 | * Normally they will be removed from the queue | |
910 | * as soon a a complete request arrives. | |
911 | */ | |
912 | svc_sock_setbufsize(svsk->sk_sock, | |
913 | (serv->sv_nrthreads+3) * serv->sv_max_mesg, | |
914 | 3 * serv->sv_max_mesg); | |
915 | ||
02fc6c36 | 916 | clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 917 | |
c0401ea0 | 918 | if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) { |
1da177e4 LT |
919 | struct kvec iov; |
920 | ||
5ee78d48 | 921 | want = sizeof(rpc_fraghdr) - svsk->sk_tcplen; |
1da177e4 LT |
922 | iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen; |
923 | iov.iov_len = want; | |
924 | if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0) | |
925 | goto error; | |
926 | svsk->sk_tcplen += len; | |
927 | ||
928 | if (len < want) { | |
c0401ea0 CL |
929 | dprintk("svc: short recvfrom while reading record " |
930 | "length (%d of %d)\n", len, want); | |
8f55f3c0 | 931 | goto err_again; /* record header not complete */ |
1da177e4 LT |
932 | } |
933 | ||
934 | svsk->sk_reclen = ntohl(svsk->sk_reclen); | |
c0401ea0 | 935 | if (!(svsk->sk_reclen & RPC_LAST_STREAM_FRAGMENT)) { |
1da177e4 LT |
936 | /* FIXME: technically, a record can be fragmented, |
937 | * and non-terminal fragments will not have the top | |
938 | * bit set in the fragment length header. | |
939 | * But apparently no known nfs clients send fragmented | |
940 | * records. */ | |
34e9a63b | 941 | if (net_ratelimit()) |
c0401ea0 CL |
942 | printk(KERN_NOTICE "RPC: multiple fragments " |
943 | "per record not supported\n"); | |
1da177e4 LT |
944 | goto err_delete; |
945 | } | |
8f55f3c0 | 946 | |
c0401ea0 | 947 | svsk->sk_reclen &= RPC_FRAGMENT_SIZE_MASK; |
1da177e4 | 948 | dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen); |
c6b0a9f8 | 949 | if (svsk->sk_reclen > serv->sv_max_mesg) { |
34e9a63b | 950 | if (net_ratelimit()) |
c0401ea0 CL |
951 | printk(KERN_NOTICE "RPC: " |
952 | "fragment too large: 0x%08lx\n", | |
953 | (unsigned long)svsk->sk_reclen); | |
1da177e4 LT |
954 | goto err_delete; |
955 | } | |
956 | } | |
957 | ||
958 | /* Check whether enough data is available */ | |
959 | len = svc_recv_available(svsk); | |
960 | if (len < 0) | |
961 | goto error; | |
962 | ||
963 | if (len < svsk->sk_reclen) { | |
964 | dprintk("svc: incomplete TCP record (%d of %d)\n", | |
965 | len, svsk->sk_reclen); | |
8f55f3c0 | 966 | goto err_again; /* record not complete */ |
1da177e4 LT |
967 | } |
968 | len = svsk->sk_reclen; | |
1da177e4 | 969 | |
8f55f3c0 AB |
970 | return len; |
971 | error: | |
b48fa6b9 | 972 | if (len == -EAGAIN) |
8f55f3c0 | 973 | dprintk("RPC: TCP recv_record got EAGAIN\n"); |
8f55f3c0 AB |
974 | return len; |
975 | err_delete: | |
976 | set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); | |
977 | err_again: | |
978 | return -EAGAIN; | |
979 | } | |
980 | ||
4cfc7e60 RI |
981 | static int svc_process_calldir(struct svc_sock *svsk, struct svc_rqst *rqstp, |
982 | struct rpc_rqst **reqpp, struct kvec *vec) | |
983 | { | |
984 | struct rpc_rqst *req = NULL; | |
985 | u32 *p; | |
986 | u32 xid; | |
987 | u32 calldir; | |
988 | int len; | |
989 | ||
990 | len = svc_recvfrom(rqstp, vec, 1, 8); | |
991 | if (len < 0) | |
992 | goto error; | |
993 | ||
994 | p = (u32 *)rqstp->rq_arg.head[0].iov_base; | |
995 | xid = *p++; | |
996 | calldir = *p; | |
997 | ||
998 | if (calldir == 0) { | |
999 | /* REQUEST is the most common case */ | |
1000 | vec[0] = rqstp->rq_arg.head[0]; | |
1001 | } else { | |
1002 | /* REPLY */ | |
d75faea3 BF |
1003 | struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt; |
1004 | ||
1005 | if (bc_xprt) | |
1006 | req = xprt_lookup_rqst(bc_xprt, xid); | |
4cfc7e60 RI |
1007 | |
1008 | if (!req) { | |
1009 | printk(KERN_NOTICE | |
1010 | "%s: Got unrecognized reply: " | |
d75faea3 | 1011 | "calldir 0x%x xpt_bc_xprt %p xid %08x\n", |
4cfc7e60 | 1012 | __func__, ntohl(calldir), |
d75faea3 | 1013 | bc_xprt, xid); |
4cfc7e60 RI |
1014 | vec[0] = rqstp->rq_arg.head[0]; |
1015 | goto out; | |
1016 | } | |
1017 | ||
1018 | memcpy(&req->rq_private_buf, &req->rq_rcv_buf, | |
1019 | sizeof(struct xdr_buf)); | |
1020 | /* copy the xid and call direction */ | |
1021 | memcpy(req->rq_private_buf.head[0].iov_base, | |
1022 | rqstp->rq_arg.head[0].iov_base, 8); | |
1023 | vec[0] = req->rq_private_buf.head[0]; | |
1024 | } | |
1025 | out: | |
1026 | vec[0].iov_base += 8; | |
1027 | vec[0].iov_len -= 8; | |
1028 | len = svsk->sk_reclen - 8; | |
1029 | error: | |
1030 | *reqpp = req; | |
1031 | return len; | |
1032 | } | |
1033 | ||
8f55f3c0 AB |
1034 | /* |
1035 | * Receive data from a TCP socket. | |
1036 | */ | |
1037 | static int svc_tcp_recvfrom(struct svc_rqst *rqstp) | |
1038 | { | |
1039 | struct svc_sock *svsk = | |
1040 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
1041 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; | |
1042 | int len; | |
1043 | struct kvec *vec; | |
4cfc7e60 | 1044 | struct rpc_rqst *req = NULL; |
5ee78d48 TM |
1045 | unsigned int vlen; |
1046 | int pnum; | |
8f55f3c0 AB |
1047 | |
1048 | dprintk("svc: tcp_recv %p data %d conn %d close %d\n", | |
1049 | svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags), | |
1050 | test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags), | |
1051 | test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags)); | |
1052 | ||
1053 | len = svc_tcp_recv_record(svsk, rqstp); | |
1054 | if (len < 0) | |
1055 | goto error; | |
1056 | ||
3cc03b16 | 1057 | vec = rqstp->rq_vec; |
1da177e4 LT |
1058 | vec[0] = rqstp->rq_arg.head[0]; |
1059 | vlen = PAGE_SIZE; | |
4cfc7e60 RI |
1060 | |
1061 | /* | |
1062 | * We have enough data for the whole tcp record. Let's try and read the | |
1063 | * first 8 bytes to get the xid and the call direction. We can use this | |
1064 | * to figure out if this is a call or a reply to a callback. If | |
1065 | * sk_reclen is < 8 (xid and calldir), then this is a malformed packet. | |
1066 | * In that case, don't bother with the calldir and just read the data. | |
1067 | * It will be rejected in svc_process. | |
1068 | */ | |
1069 | if (len >= 8) { | |
1070 | len = svc_process_calldir(svsk, rqstp, &req, vec); | |
1071 | if (len < 0) | |
1072 | goto err_again; | |
1073 | vlen -= 8; | |
1074 | } | |
1075 | ||
1da177e4 | 1076 | pnum = 1; |
5ee78d48 | 1077 | while (vlen < svsk->sk_reclen - 8) { |
4cfc7e60 RI |
1078 | vec[pnum].iov_base = (req) ? |
1079 | page_address(req->rq_private_buf.pages[pnum - 1]) : | |
1080 | page_address(rqstp->rq_pages[pnum]); | |
1da177e4 LT |
1081 | vec[pnum].iov_len = PAGE_SIZE; |
1082 | pnum++; | |
1083 | vlen += PAGE_SIZE; | |
1084 | } | |
44524359 | 1085 | rqstp->rq_respages = &rqstp->rq_pages[pnum]; |
1da177e4 LT |
1086 | |
1087 | /* Now receive data */ | |
5ee78d48 | 1088 | len = svc_recvfrom(rqstp, vec, pnum, svsk->sk_reclen - 8); |
1da177e4 | 1089 | if (len < 0) |
8f55f3c0 | 1090 | goto err_again; |
1da177e4 | 1091 | |
4cfc7e60 | 1092 | if (req) { |
5ee78d48 TM |
1093 | xprt_complete_rqst(req->rq_task, svsk->sk_reclen); |
1094 | rqstp->rq_arg.len = 0; | |
4cfc7e60 RI |
1095 | goto out; |
1096 | } | |
5ee78d48 TM |
1097 | dprintk("svc: TCP complete record (%d bytes)\n", svsk->sk_reclen); |
1098 | rqstp->rq_arg.len = svsk->sk_reclen; | |
1da177e4 | 1099 | rqstp->rq_arg.page_base = 0; |
5ee78d48 TM |
1100 | if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) { |
1101 | rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len; | |
1da177e4 | 1102 | rqstp->rq_arg.page_len = 0; |
5ee78d48 TM |
1103 | } else |
1104 | rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len; | |
1da177e4 | 1105 | |
5148bf4e | 1106 | rqstp->rq_xprt_ctxt = NULL; |
1da177e4 LT |
1107 | rqstp->rq_prot = IPPROTO_TCP; |
1108 | ||
4cfc7e60 | 1109 | out: |
1da177e4 LT |
1110 | /* Reset TCP read info */ |
1111 | svsk->sk_reclen = 0; | |
1112 | svsk->sk_tcplen = 0; | |
0601f793 TM |
1113 | /* If we have more data, signal svc_xprt_enqueue() to try again */ |
1114 | if (svc_recv_available(svsk) > sizeof(rpc_fraghdr)) | |
1115 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); | |
1116 | ||
1da177e4 | 1117 | |
9dbc240f | 1118 | svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt); |
1da177e4 LT |
1119 | if (serv->sv_stats) |
1120 | serv->sv_stats->nettcpcnt++; | |
1121 | ||
5ee78d48 | 1122 | return rqstp->rq_arg.len; |
1da177e4 | 1123 | |
8f55f3c0 | 1124 | err_again: |
1da177e4 LT |
1125 | if (len == -EAGAIN) { |
1126 | dprintk("RPC: TCP recvfrom got EAGAIN\n"); | |
8f55f3c0 AB |
1127 | return len; |
1128 | } | |
1129 | error: | |
1130 | if (len != -EAGAIN) { | |
1da177e4 | 1131 | printk(KERN_NOTICE "%s: recvfrom returned errno %d\n", |
bb5cf160 | 1132 | svsk->sk_xprt.xpt_server->sv_name, -len); |
8f55f3c0 | 1133 | set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 1134 | } |
8f55f3c0 | 1135 | return -EAGAIN; |
1da177e4 LT |
1136 | } |
1137 | ||
1138 | /* | |
1139 | * Send out data on TCP socket. | |
1140 | */ | |
0f0257ea | 1141 | static int svc_tcp_sendto(struct svc_rqst *rqstp) |
1da177e4 LT |
1142 | { |
1143 | struct xdr_buf *xbufp = &rqstp->rq_res; | |
1144 | int sent; | |
d8ed029d | 1145 | __be32 reclen; |
1da177e4 LT |
1146 | |
1147 | /* Set up the first element of the reply kvec. | |
1148 | * Any other kvecs that may be in use have been taken | |
1149 | * care of by the server implementation itself. | |
1150 | */ | |
1151 | reclen = htonl(0x80000000|((xbufp->len ) - 4)); | |
1152 | memcpy(xbufp->head[0].iov_base, &reclen, 4); | |
1153 | ||
1da177e4 LT |
1154 | sent = svc_sendto(rqstp, &rqstp->rq_res); |
1155 | if (sent != xbufp->len) { | |
0f0257ea TT |
1156 | printk(KERN_NOTICE |
1157 | "rpc-srv/tcp: %s: %s %d when sending %d bytes " | |
1158 | "- shutting down socket\n", | |
57b1d3ba | 1159 | rqstp->rq_xprt->xpt_server->sv_name, |
1da177e4 LT |
1160 | (sent<0)?"got error":"sent only", |
1161 | sent, xbufp->len); | |
57b1d3ba | 1162 | set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags); |
f6150c3c | 1163 | svc_xprt_enqueue(rqstp->rq_xprt); |
1da177e4 LT |
1164 | sent = -EAGAIN; |
1165 | } | |
1166 | return sent; | |
1167 | } | |
1168 | ||
e831fe65 TT |
1169 | /* |
1170 | * Setup response header. TCP has a 4B record length field. | |
1171 | */ | |
1172 | static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp) | |
1173 | { | |
1174 | struct kvec *resv = &rqstp->rq_res.head[0]; | |
1175 | ||
1176 | /* tcp needs a space for the record length... */ | |
1177 | svc_putnl(resv, 0); | |
1178 | } | |
1179 | ||
323bee32 TT |
1180 | static int svc_tcp_has_wspace(struct svc_xprt *xprt) |
1181 | { | |
57b1d3ba | 1182 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); |
47fcb03f | 1183 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
323bee32 | 1184 | int required; |
323bee32 | 1185 | |
47fcb03f TM |
1186 | if (test_bit(XPT_LISTENER, &xprt->xpt_flags)) |
1187 | return 1; | |
1188 | required = atomic_read(&xprt->xpt_reserved) + serv->sv_max_mesg; | |
1189 | if (sk_stream_wspace(svsk->sk_sk) >= required) | |
1190 | return 1; | |
323bee32 | 1191 | set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); |
47fcb03f | 1192 | return 0; |
323bee32 TT |
1193 | } |
1194 | ||
b700cbb1 | 1195 | static struct svc_xprt *svc_tcp_create(struct svc_serv *serv, |
62832c03 | 1196 | struct net *net, |
b700cbb1 TT |
1197 | struct sockaddr *sa, int salen, |
1198 | int flags) | |
1199 | { | |
62832c03 | 1200 | return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags); |
b700cbb1 TT |
1201 | } |
1202 | ||
1f11a034 AA |
1203 | #if defined(CONFIG_NFS_V4_1) |
1204 | static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int, | |
1205 | struct net *, struct sockaddr *, | |
1206 | int, int); | |
1207 | static void svc_bc_sock_free(struct svc_xprt *xprt); | |
1208 | ||
1209 | static struct svc_xprt *svc_bc_tcp_create(struct svc_serv *serv, | |
1210 | struct net *net, | |
1211 | struct sockaddr *sa, int salen, | |
1212 | int flags) | |
1213 | { | |
1214 | return svc_bc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags); | |
1215 | } | |
1216 | ||
1217 | static void svc_bc_tcp_sock_detach(struct svc_xprt *xprt) | |
1218 | { | |
1219 | } | |
1220 | ||
1221 | static struct svc_xprt_ops svc_tcp_bc_ops = { | |
1222 | .xpo_create = svc_bc_tcp_create, | |
1223 | .xpo_detach = svc_bc_tcp_sock_detach, | |
1224 | .xpo_free = svc_bc_sock_free, | |
1225 | .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr, | |
1226 | }; | |
1227 | ||
1228 | static struct svc_xprt_class svc_tcp_bc_class = { | |
1229 | .xcl_name = "tcp-bc", | |
1230 | .xcl_owner = THIS_MODULE, | |
1231 | .xcl_ops = &svc_tcp_bc_ops, | |
1232 | .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP, | |
1233 | }; | |
16b2d1e1 AA |
1234 | |
1235 | static void svc_init_bc_xprt_sock(void) | |
1236 | { | |
1237 | svc_reg_xprt_class(&svc_tcp_bc_class); | |
1238 | } | |
1239 | ||
1240 | static void svc_cleanup_bc_xprt_sock(void) | |
1241 | { | |
1242 | svc_unreg_xprt_class(&svc_tcp_bc_class); | |
1243 | } | |
1244 | #else /* CONFIG_NFS_V4_1 */ | |
1245 | static void svc_init_bc_xprt_sock(void) | |
1246 | { | |
1247 | } | |
1248 | ||
1249 | static void svc_cleanup_bc_xprt_sock(void) | |
1250 | { | |
1251 | } | |
1f11a034 AA |
1252 | #endif /* CONFIG_NFS_V4_1 */ |
1253 | ||
360d8738 | 1254 | static struct svc_xprt_ops svc_tcp_ops = { |
b700cbb1 | 1255 | .xpo_create = svc_tcp_create, |
5d137990 TT |
1256 | .xpo_recvfrom = svc_tcp_recvfrom, |
1257 | .xpo_sendto = svc_tcp_sendto, | |
5148bf4e | 1258 | .xpo_release_rqst = svc_release_skb, |
69b6ba37 | 1259 | .xpo_detach = svc_tcp_sock_detach, |
755cceab | 1260 | .xpo_free = svc_sock_free, |
e831fe65 | 1261 | .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr, |
323bee32 | 1262 | .xpo_has_wspace = svc_tcp_has_wspace, |
38a417cc | 1263 | .xpo_accept = svc_tcp_accept, |
360d8738 TT |
1264 | }; |
1265 | ||
1266 | static struct svc_xprt_class svc_tcp_class = { | |
1267 | .xcl_name = "tcp", | |
b700cbb1 | 1268 | .xcl_owner = THIS_MODULE, |
360d8738 | 1269 | .xcl_ops = &svc_tcp_ops, |
49023155 | 1270 | .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP, |
360d8738 TT |
1271 | }; |
1272 | ||
1273 | void svc_init_xprt_sock(void) | |
1274 | { | |
1275 | svc_reg_xprt_class(&svc_tcp_class); | |
1276 | svc_reg_xprt_class(&svc_udp_class); | |
16b2d1e1 | 1277 | svc_init_bc_xprt_sock(); |
360d8738 TT |
1278 | } |
1279 | ||
1280 | void svc_cleanup_xprt_sock(void) | |
1281 | { | |
1282 | svc_unreg_xprt_class(&svc_tcp_class); | |
1283 | svc_unreg_xprt_class(&svc_udp_class); | |
16b2d1e1 | 1284 | svc_cleanup_bc_xprt_sock(); |
360d8738 TT |
1285 | } |
1286 | ||
bb5cf160 | 1287 | static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv) |
1da177e4 LT |
1288 | { |
1289 | struct sock *sk = svsk->sk_sk; | |
1da177e4 | 1290 | |
bb5cf160 | 1291 | svc_xprt_init(&svc_tcp_class, &svsk->sk_xprt, serv); |
def13d74 | 1292 | set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
1293 | if (sk->sk_state == TCP_LISTEN) { |
1294 | dprintk("setting up TCP socket for listening\n"); | |
02fc6c36 | 1295 | set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 1296 | sk->sk_data_ready = svc_tcp_listen_data_ready; |
02fc6c36 | 1297 | set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
1298 | } else { |
1299 | dprintk("setting up TCP socket for reading\n"); | |
1300 | sk->sk_state_change = svc_tcp_state_change; | |
1301 | sk->sk_data_ready = svc_tcp_data_ready; | |
47fcb03f | 1302 | sk->sk_write_space = svc_tcp_write_space; |
1da177e4 LT |
1303 | |
1304 | svsk->sk_reclen = 0; | |
1305 | svsk->sk_tcplen = 0; | |
1306 | ||
b7872fe8 | 1307 | tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF; |
1da177e4 | 1308 | |
7f421835 BF |
1309 | /* initialise setting must have enough space to |
1310 | * receive and respond to one request. | |
1311 | * svc_tcp_recvfrom will re-adjust if necessary | |
1312 | */ | |
1313 | svc_sock_setbufsize(svsk->sk_sock, | |
1314 | 3 * svsk->sk_xprt.xpt_server->sv_max_mesg, | |
1315 | 3 * svsk->sk_xprt.xpt_server->sv_max_mesg); | |
1316 | ||
1317 | set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); | |
02fc6c36 | 1318 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
cca5172a | 1319 | if (sk->sk_state != TCP_ESTABLISHED) |
02fc6c36 | 1320 | set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
1321 | } |
1322 | } | |
1323 | ||
0f0257ea | 1324 | void svc_sock_update_bufs(struct svc_serv *serv) |
1da177e4 LT |
1325 | { |
1326 | /* | |
1327 | * The number of server threads has changed. Update | |
1328 | * rcvbuf and sndbuf accordingly on all sockets | |
1329 | */ | |
8f3a6de3 | 1330 | struct svc_sock *svsk; |
1da177e4 LT |
1331 | |
1332 | spin_lock_bh(&serv->sv_lock); | |
8f3a6de3 | 1333 | list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list) |
02fc6c36 | 1334 | set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); |
8f3a6de3 | 1335 | list_for_each_entry(svsk, &serv->sv_tempsocks, sk_xprt.xpt_list) |
02fc6c36 | 1336 | set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
1337 | spin_unlock_bh(&serv->sv_lock); |
1338 | } | |
24c3767e | 1339 | EXPORT_SYMBOL_GPL(svc_sock_update_bufs); |
1da177e4 | 1340 | |
1da177e4 LT |
1341 | /* |
1342 | * Initialize socket for RPC use and create svc_sock struct | |
1343 | * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF. | |
1344 | */ | |
6b174337 CL |
1345 | static struct svc_sock *svc_setup_socket(struct svc_serv *serv, |
1346 | struct socket *sock, | |
1347 | int *errp, int flags) | |
1da177e4 LT |
1348 | { |
1349 | struct svc_sock *svsk; | |
1350 | struct sock *inet; | |
6b174337 | 1351 | int pmap_register = !(flags & SVC_SOCK_ANONYMOUS); |
1da177e4 LT |
1352 | |
1353 | dprintk("svc: svc_setup_socket %p\n", sock); | |
0da974f4 | 1354 | if (!(svsk = kzalloc(sizeof(*svsk), GFP_KERNEL))) { |
1da177e4 LT |
1355 | *errp = -ENOMEM; |
1356 | return NULL; | |
1357 | } | |
1da177e4 LT |
1358 | |
1359 | inet = sock->sk; | |
1360 | ||
1361 | /* Register socket with portmapper */ | |
1362 | if (*errp >= 0 && pmap_register) | |
baf01caf | 1363 | *errp = svc_register(serv, inet->sk_family, inet->sk_protocol, |
c720c7e8 | 1364 | ntohs(inet_sk(inet)->inet_sport)); |
1da177e4 LT |
1365 | |
1366 | if (*errp < 0) { | |
1367 | kfree(svsk); | |
1368 | return NULL; | |
1369 | } | |
1370 | ||
1da177e4 LT |
1371 | inet->sk_user_data = svsk; |
1372 | svsk->sk_sock = sock; | |
1373 | svsk->sk_sk = inet; | |
1374 | svsk->sk_ostate = inet->sk_state_change; | |
1375 | svsk->sk_odata = inet->sk_data_ready; | |
1376 | svsk->sk_owspace = inet->sk_write_space; | |
1da177e4 LT |
1377 | |
1378 | /* Initialize the socket */ | |
1379 | if (sock->type == SOCK_DGRAM) | |
bb5cf160 | 1380 | svc_udp_init(svsk, serv); |
7f421835 | 1381 | else |
bb5cf160 | 1382 | svc_tcp_init(svsk, serv); |
1da177e4 | 1383 | |
1da177e4 LT |
1384 | dprintk("svc: svc_setup_socket created %p (inet %p)\n", |
1385 | svsk, svsk->sk_sk); | |
1386 | ||
1da177e4 LT |
1387 | return svsk; |
1388 | } | |
1389 | ||
bfba9ab4 CL |
1390 | /** |
1391 | * svc_addsock - add a listener socket to an RPC service | |
1392 | * @serv: pointer to RPC service to which to add a new listener | |
1393 | * @fd: file descriptor of the new listener | |
1394 | * @name_return: pointer to buffer to fill in with name of listener | |
1395 | * @len: size of the buffer | |
1396 | * | |
1397 | * Fills in socket name and returns positive length of name if successful. | |
1398 | * Name is terminated with '\n'. On error, returns a negative errno | |
1399 | * value. | |
1400 | */ | |
1401 | int svc_addsock(struct svc_serv *serv, const int fd, char *name_return, | |
1402 | const size_t len) | |
b41b66d6 N |
1403 | { |
1404 | int err = 0; | |
1405 | struct socket *so = sockfd_lookup(fd, &err); | |
1406 | struct svc_sock *svsk = NULL; | |
1407 | ||
1408 | if (!so) | |
1409 | return err; | |
205ba423 | 1410 | if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6)) |
b41b66d6 N |
1411 | err = -EAFNOSUPPORT; |
1412 | else if (so->sk->sk_protocol != IPPROTO_TCP && | |
1413 | so->sk->sk_protocol != IPPROTO_UDP) | |
1414 | err = -EPROTONOSUPPORT; | |
1415 | else if (so->state > SS_UNCONNECTED) | |
1416 | err = -EISCONN; | |
1417 | else { | |
2da2c21d TT |
1418 | if (!try_module_get(THIS_MODULE)) |
1419 | err = -ENOENT; | |
1420 | else | |
1421 | svsk = svc_setup_socket(serv, so, &err, | |
1422 | SVC_SOCK_DEFAULTS); | |
e79eff1f | 1423 | if (svsk) { |
9dbc240f TT |
1424 | struct sockaddr_storage addr; |
1425 | struct sockaddr *sin = (struct sockaddr *)&addr; | |
1426 | int salen; | |
1427 | if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0) | |
1428 | svc_xprt_set_local(&svsk->sk_xprt, sin, salen); | |
4e5caaa5 TT |
1429 | clear_bit(XPT_TEMP, &svsk->sk_xprt.xpt_flags); |
1430 | spin_lock_bh(&serv->sv_lock); | |
1431 | list_add(&svsk->sk_xprt.xpt_list, &serv->sv_permsocks); | |
1432 | spin_unlock_bh(&serv->sv_lock); | |
a6046f71 | 1433 | svc_xprt_received(&svsk->sk_xprt); |
b41b66d6 | 1434 | err = 0; |
2da2c21d TT |
1435 | } else |
1436 | module_put(THIS_MODULE); | |
b41b66d6 N |
1437 | } |
1438 | if (err) { | |
1439 | sockfd_put(so); | |
1440 | return err; | |
1441 | } | |
e7942b9f | 1442 | return svc_one_sock_name(svsk, name_return, len); |
b41b66d6 N |
1443 | } |
1444 | EXPORT_SYMBOL_GPL(svc_addsock); | |
1445 | ||
1da177e4 LT |
1446 | /* |
1447 | * Create socket for RPC service. | |
1448 | */ | |
b700cbb1 TT |
1449 | static struct svc_xprt *svc_create_socket(struct svc_serv *serv, |
1450 | int protocol, | |
62832c03 | 1451 | struct net *net, |
b700cbb1 TT |
1452 | struct sockaddr *sin, int len, |
1453 | int flags) | |
1da177e4 LT |
1454 | { |
1455 | struct svc_sock *svsk; | |
1456 | struct socket *sock; | |
1457 | int error; | |
1458 | int type; | |
9dbc240f TT |
1459 | struct sockaddr_storage addr; |
1460 | struct sockaddr *newsin = (struct sockaddr *)&addr; | |
1461 | int newlen; | |
c69da774 TM |
1462 | int family; |
1463 | int val; | |
5216a8e7 | 1464 | RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); |
1da177e4 | 1465 | |
ad06e4bd CL |
1466 | dprintk("svc: svc_create_socket(%s, %d, %s)\n", |
1467 | serv->sv_program->pg_name, protocol, | |
77f1f67a | 1468 | __svc_print_addr(sin, buf, sizeof(buf))); |
1da177e4 LT |
1469 | |
1470 | if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) { | |
1471 | printk(KERN_WARNING "svc: only UDP and TCP " | |
1472 | "sockets supported\n"); | |
b700cbb1 | 1473 | return ERR_PTR(-EINVAL); |
1da177e4 | 1474 | } |
c69da774 | 1475 | |
1da177e4 | 1476 | type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM; |
c69da774 TM |
1477 | switch (sin->sa_family) { |
1478 | case AF_INET6: | |
1479 | family = PF_INET6; | |
1480 | break; | |
1481 | case AF_INET: | |
1482 | family = PF_INET; | |
1483 | break; | |
1484 | default: | |
1485 | return ERR_PTR(-EINVAL); | |
1486 | } | |
1da177e4 | 1487 | |
14ec63c3 | 1488 | error = __sock_create(net, family, type, protocol, &sock, 1); |
77f1f67a | 1489 | if (error < 0) |
b700cbb1 | 1490 | return ERR_PTR(error); |
1da177e4 | 1491 | |
ed07536e PZ |
1492 | svc_reclassify_socket(sock); |
1493 | ||
c69da774 TM |
1494 | /* |
1495 | * If this is an PF_INET6 listener, we want to avoid | |
1496 | * getting requests from IPv4 remotes. Those should | |
1497 | * be shunted to a PF_INET listener via rpcbind. | |
1498 | */ | |
1499 | val = 1; | |
1500 | if (family == PF_INET6) | |
1501 | kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY, | |
1502 | (char *)&val, sizeof(val)); | |
1503 | ||
18114746 | 1504 | if (type == SOCK_STREAM) |
77f1f67a CL |
1505 | sock->sk->sk_reuse = 1; /* allow address reuse */ |
1506 | error = kernel_bind(sock, sin, len); | |
18114746 ES |
1507 | if (error < 0) |
1508 | goto bummer; | |
1da177e4 | 1509 | |
9dbc240f TT |
1510 | newlen = len; |
1511 | error = kernel_getsockname(sock, newsin, &newlen); | |
1512 | if (error < 0) | |
1513 | goto bummer; | |
1514 | ||
1da177e4 | 1515 | if (protocol == IPPROTO_TCP) { |
e6242e92 | 1516 | if ((error = kernel_listen(sock, 64)) < 0) |
1da177e4 LT |
1517 | goto bummer; |
1518 | } | |
1519 | ||
e79eff1f | 1520 | if ((svsk = svc_setup_socket(serv, sock, &error, flags)) != NULL) { |
9dbc240f | 1521 | svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen); |
b700cbb1 | 1522 | return (struct svc_xprt *)svsk; |
e79eff1f | 1523 | } |
1da177e4 LT |
1524 | |
1525 | bummer: | |
1526 | dprintk("svc: svc_create_socket error = %d\n", -error); | |
1527 | sock_release(sock); | |
b700cbb1 | 1528 | return ERR_PTR(error); |
1da177e4 LT |
1529 | } |
1530 | ||
755cceab TT |
1531 | /* |
1532 | * Detach the svc_sock from the socket so that no | |
1533 | * more callbacks occur. | |
1534 | */ | |
1535 | static void svc_sock_detach(struct svc_xprt *xprt) | |
1536 | { | |
1537 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
1538 | struct sock *sk = svsk->sk_sk; | |
eaefd110 | 1539 | wait_queue_head_t *wq; |
755cceab TT |
1540 | |
1541 | dprintk("svc: svc_sock_detach(%p)\n", svsk); | |
1542 | ||
1543 | /* put back the old socket callbacks */ | |
1544 | sk->sk_state_change = svsk->sk_ostate; | |
1545 | sk->sk_data_ready = svsk->sk_odata; | |
1546 | sk->sk_write_space = svsk->sk_owspace; | |
69b6ba37 | 1547 | |
eaefd110 ED |
1548 | wq = sk_sleep(sk); |
1549 | if (wq && waitqueue_active(wq)) | |
1550 | wake_up_interruptible(wq); | |
69b6ba37 TM |
1551 | } |
1552 | ||
1553 | /* | |
1554 | * Disconnect the socket, and reset the callbacks | |
1555 | */ | |
1556 | static void svc_tcp_sock_detach(struct svc_xprt *xprt) | |
1557 | { | |
1558 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
1559 | ||
1560 | dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk); | |
1561 | ||
1562 | svc_sock_detach(xprt); | |
1563 | ||
1564 | if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) | |
1565 | kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR); | |
755cceab TT |
1566 | } |
1567 | ||
1568 | /* | |
1569 | * Free the svc_sock's socket resources and the svc_sock itself. | |
1570 | */ | |
1571 | static void svc_sock_free(struct svc_xprt *xprt) | |
1572 | { | |
1573 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
1574 | dprintk("svc: svc_sock_free(%p)\n", svsk); | |
1575 | ||
755cceab TT |
1576 | if (svsk->sk_sock->file) |
1577 | sockfd_put(svsk->sk_sock); | |
1578 | else | |
1579 | sock_release(svsk->sk_sock); | |
1580 | kfree(svsk); | |
1581 | } | |
7652e5a0 | 1582 | |
1f11a034 | 1583 | #if defined(CONFIG_NFS_V4_1) |
7652e5a0 | 1584 | /* |
1f11a034 | 1585 | * Create a back channel svc_xprt which shares the fore channel socket. |
7652e5a0 | 1586 | */ |
1f11a034 AA |
1587 | static struct svc_xprt *svc_bc_create_socket(struct svc_serv *serv, |
1588 | int protocol, | |
1589 | struct net *net, | |
1590 | struct sockaddr *sin, int len, | |
1591 | int flags) | |
7652e5a0 BH |
1592 | { |
1593 | struct svc_sock *svsk; | |
1f11a034 AA |
1594 | struct svc_xprt *xprt; |
1595 | ||
1596 | if (protocol != IPPROTO_TCP) { | |
1597 | printk(KERN_WARNING "svc: only TCP sockets" | |
1598 | " supported on shared back channel\n"); | |
1599 | return ERR_PTR(-EINVAL); | |
1600 | } | |
7652e5a0 | 1601 | |
7652e5a0 BH |
1602 | svsk = kzalloc(sizeof(*svsk), GFP_KERNEL); |
1603 | if (!svsk) | |
1f11a034 | 1604 | return ERR_PTR(-ENOMEM); |
7652e5a0 BH |
1605 | |
1606 | xprt = &svsk->sk_xprt; | |
1f11a034 AA |
1607 | svc_xprt_init(&svc_tcp_bc_class, xprt, serv); |
1608 | ||
4a19de0f | 1609 | serv->sv_bc_xprt = xprt; |
1f11a034 | 1610 | |
7652e5a0 BH |
1611 | return xprt; |
1612 | } | |
7652e5a0 BH |
1613 | |
1614 | /* | |
1f11a034 | 1615 | * Free a back channel svc_sock. |
7652e5a0 | 1616 | */ |
1f11a034 | 1617 | static void svc_bc_sock_free(struct svc_xprt *xprt) |
7652e5a0 | 1618 | { |
778be232 | 1619 | if (xprt) |
7652e5a0 BH |
1620 | kfree(container_of(xprt, struct svc_sock, sk_xprt)); |
1621 | } | |
1f11a034 | 1622 | #endif /* CONFIG_NFS_V4_1 */ |