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 | */ | |
243 | static int one_sock_name(char *buf, struct svc_sock *svsk) | |
244 | { | |
245 | int len; | |
246 | ||
247 | switch(svsk->sk_sk->sk_family) { | |
248 | case AF_INET: | |
21454aaa HH |
249 | len = sprintf(buf, "ipv4 %s %pI4 %d\n", |
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: | |
256 | len = sprintf(buf, "*unknown-%d*\n", | |
257 | svsk->sk_sk->sk_family); | |
258 | } | |
259 | return len; | |
260 | } | |
261 | ||
262 | int | |
b41b66d6 | 263 | svc_sock_names(char *buf, struct svc_serv *serv, char *toclose) |
80212d59 | 264 | { |
b41b66d6 | 265 | struct svc_sock *svsk, *closesk = NULL; |
80212d59 N |
266 | int len = 0; |
267 | ||
268 | if (!serv) | |
269 | return 0; | |
aaf68cfb | 270 | spin_lock_bh(&serv->sv_lock); |
7a182083 | 271 | list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list) { |
80212d59 | 272 | int onelen = one_sock_name(buf+len, svsk); |
b41b66d6 N |
273 | if (toclose && strcmp(toclose, buf+len) == 0) |
274 | closesk = svsk; | |
275 | else | |
276 | len += onelen; | |
80212d59 | 277 | } |
aaf68cfb | 278 | spin_unlock_bh(&serv->sv_lock); |
b41b66d6 | 279 | if (closesk) |
5680c446 N |
280 | /* Should unregister with portmap, but you cannot |
281 | * unregister just one protocol... | |
282 | */ | |
7a182083 | 283 | svc_close_xprt(&closesk->sk_xprt); |
37a03472 N |
284 | else if (toclose) |
285 | return -ENOENT; | |
80212d59 N |
286 | return len; |
287 | } | |
24c3767e | 288 | EXPORT_SYMBOL_GPL(svc_sock_names); |
80212d59 | 289 | |
1da177e4 LT |
290 | /* |
291 | * Check input queue length | |
292 | */ | |
0f0257ea | 293 | static int svc_recv_available(struct svc_sock *svsk) |
1da177e4 | 294 | { |
1da177e4 LT |
295 | struct socket *sock = svsk->sk_sock; |
296 | int avail, err; | |
297 | ||
e6242e92 | 298 | err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail); |
1da177e4 LT |
299 | |
300 | return (err >= 0)? avail : err; | |
301 | } | |
302 | ||
303 | /* | |
304 | * Generic recvfrom routine. | |
305 | */ | |
0f0257ea TT |
306 | static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr, |
307 | int buflen) | |
1da177e4 | 308 | { |
57b1d3ba TT |
309 | struct svc_sock *svsk = |
310 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
1ba95105 CL |
311 | struct msghdr msg = { |
312 | .msg_flags = MSG_DONTWAIT, | |
313 | }; | |
314 | int len; | |
1da177e4 | 315 | |
260c1d12 TT |
316 | rqstp->rq_xprt_hlen = 0; |
317 | ||
1ba95105 CL |
318 | len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen, |
319 | msg.msg_flags); | |
1da177e4 | 320 | |
1da177e4 | 321 | dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n", |
1ba95105 | 322 | svsk, iov[0].iov_base, iov[0].iov_len, len); |
1da177e4 LT |
323 | return len; |
324 | } | |
325 | ||
326 | /* | |
327 | * Set socket snd and rcv buffer lengths | |
328 | */ | |
0f0257ea TT |
329 | static void svc_sock_setbufsize(struct socket *sock, unsigned int snd, |
330 | unsigned int rcv) | |
1da177e4 LT |
331 | { |
332 | #if 0 | |
333 | mm_segment_t oldfs; | |
334 | oldfs = get_fs(); set_fs(KERNEL_DS); | |
335 | sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF, | |
336 | (char*)&snd, sizeof(snd)); | |
337 | sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF, | |
338 | (char*)&rcv, sizeof(rcv)); | |
339 | #else | |
340 | /* sock_setsockopt limits use to sysctl_?mem_max, | |
341 | * which isn't acceptable. Until that is made conditional | |
342 | * on not having CAP_SYS_RESOURCE or similar, we go direct... | |
343 | * DaveM said I could! | |
344 | */ | |
345 | lock_sock(sock->sk); | |
346 | sock->sk->sk_sndbuf = snd * 2; | |
347 | sock->sk->sk_rcvbuf = rcv * 2; | |
1da177e4 LT |
348 | release_sock(sock->sk); |
349 | #endif | |
350 | } | |
351 | /* | |
352 | * INET callback when data has been received on the socket. | |
353 | */ | |
0f0257ea | 354 | static void svc_udp_data_ready(struct sock *sk, int count) |
1da177e4 | 355 | { |
939bb7ef | 356 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 | 357 | |
939bb7ef NB |
358 | if (svsk) { |
359 | dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n", | |
02fc6c36 TT |
360 | svsk, sk, count, |
361 | test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags)); | |
362 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); | |
f6150c3c | 363 | svc_xprt_enqueue(&svsk->sk_xprt); |
939bb7ef | 364 | } |
1da177e4 LT |
365 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
366 | wake_up_interruptible(sk->sk_sleep); | |
367 | } | |
368 | ||
369 | /* | |
370 | * INET callback when space is newly available on the socket. | |
371 | */ | |
0f0257ea | 372 | static void svc_write_space(struct sock *sk) |
1da177e4 LT |
373 | { |
374 | struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data); | |
375 | ||
376 | if (svsk) { | |
377 | dprintk("svc: socket %p(inet %p), write_space busy=%d\n", | |
02fc6c36 | 378 | svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags)); |
f6150c3c | 379 | svc_xprt_enqueue(&svsk->sk_xprt); |
1da177e4 LT |
380 | } |
381 | ||
382 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) { | |
939bb7ef | 383 | dprintk("RPC svc_write_space: someone sleeping on %p\n", |
1da177e4 LT |
384 | svsk); |
385 | wake_up_interruptible(sk->sk_sleep); | |
386 | } | |
387 | } | |
388 | ||
9dbc240f TT |
389 | /* |
390 | * Copy the UDP datagram's destination address to the rqstp structure. | |
391 | * The 'destination' address in this case is the address to which the | |
392 | * peer sent the datagram, i.e. our local address. For multihomed | |
393 | * hosts, this can change from msg to msg. Note that only the IP | |
394 | * address changes, the port number should remain the same. | |
395 | */ | |
396 | static void svc_udp_get_dest_address(struct svc_rqst *rqstp, | |
397 | struct cmsghdr *cmh) | |
95756482 | 398 | { |
57b1d3ba TT |
399 | struct svc_sock *svsk = |
400 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
401 | switch (svsk->sk_sk->sk_family) { | |
95756482 | 402 | case AF_INET: { |
7a37f578 N |
403 | struct in_pktinfo *pki = CMSG_DATA(cmh); |
404 | rqstp->rq_daddr.addr.s_addr = pki->ipi_spec_dst.s_addr; | |
95756482 | 405 | break; |
7a37f578 | 406 | } |
95756482 | 407 | case AF_INET6: { |
7a37f578 N |
408 | struct in6_pktinfo *pki = CMSG_DATA(cmh); |
409 | ipv6_addr_copy(&rqstp->rq_daddr.addr6, &pki->ipi6_addr); | |
95756482 | 410 | break; |
7a37f578 | 411 | } |
95756482 | 412 | } |
95756482 CL |
413 | } |
414 | ||
1da177e4 LT |
415 | /* |
416 | * Receive a datagram from a UDP socket. | |
417 | */ | |
0f0257ea | 418 | static int svc_udp_recvfrom(struct svc_rqst *rqstp) |
1da177e4 | 419 | { |
57b1d3ba TT |
420 | struct svc_sock *svsk = |
421 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
bb5cf160 | 422 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
1da177e4 | 423 | struct sk_buff *skb; |
bc375ea7 DM |
424 | union { |
425 | struct cmsghdr hdr; | |
426 | long all[SVC_PKTINFO_SPACE / sizeof(long)]; | |
427 | } buffer; | |
428 | struct cmsghdr *cmh = &buffer.hdr; | |
1da177e4 | 429 | int err, len; |
7a37f578 N |
430 | struct msghdr msg = { |
431 | .msg_name = svc_addr(rqstp), | |
432 | .msg_control = cmh, | |
433 | .msg_controllen = sizeof(buffer), | |
434 | .msg_flags = MSG_DONTWAIT, | |
435 | }; | |
1da177e4 | 436 | |
02fc6c36 | 437 | if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags)) |
1da177e4 LT |
438 | /* udp sockets need large rcvbuf as all pending |
439 | * requests are still in that buffer. sndbuf must | |
440 | * also be large enough that there is enough space | |
3262c816 GB |
441 | * for one reply per thread. We count all threads |
442 | * rather than threads in a particular pool, which | |
443 | * provides an upper bound on the number of threads | |
444 | * which will access the socket. | |
1da177e4 LT |
445 | */ |
446 | svc_sock_setbufsize(svsk->sk_sock, | |
c6b0a9f8 N |
447 | (serv->sv_nrthreads+3) * serv->sv_max_mesg, |
448 | (serv->sv_nrthreads+3) * serv->sv_max_mesg); | |
1da177e4 | 449 | |
02fc6c36 | 450 | clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
05ed690e N |
451 | skb = NULL; |
452 | err = kernel_recvmsg(svsk->sk_sock, &msg, NULL, | |
453 | 0, 0, MSG_PEEK | MSG_DONTWAIT); | |
454 | if (err >= 0) | |
455 | skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err); | |
456 | ||
457 | if (skb == NULL) { | |
458 | if (err != -EAGAIN) { | |
459 | /* possibly an icmp error */ | |
460 | dprintk("svc: recvfrom returned error %d\n", -err); | |
02fc6c36 | 461 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 462 | } |
a6046f71 | 463 | svc_xprt_received(&svsk->sk_xprt); |
05ed690e | 464 | return -EAGAIN; |
1da177e4 | 465 | } |
9dbc240f TT |
466 | len = svc_addr_len(svc_addr(rqstp)); |
467 | if (len < 0) | |
468 | return len; | |
469 | rqstp->rq_addrlen = len; | |
b7aa0bf7 ED |
470 | if (skb->tstamp.tv64 == 0) { |
471 | skb->tstamp = ktime_get_real(); | |
cca5172a | 472 | /* Don't enable netstamp, sunrpc doesn't |
1da177e4 LT |
473 | need that much accuracy */ |
474 | } | |
b7aa0bf7 | 475 | svsk->sk_sk->sk_stamp = skb->tstamp; |
02fc6c36 | 476 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */ |
1da177e4 LT |
477 | |
478 | /* | |
479 | * Maybe more packets - kick another thread ASAP. | |
480 | */ | |
a6046f71 | 481 | svc_xprt_received(&svsk->sk_xprt); |
1da177e4 LT |
482 | |
483 | len = skb->len - sizeof(struct udphdr); | |
484 | rqstp->rq_arg.len = len; | |
485 | ||
95756482 | 486 | rqstp->rq_prot = IPPROTO_UDP; |
27459f09 | 487 | |
7a37f578 N |
488 | if (cmh->cmsg_level != IPPROTO_IP || |
489 | cmh->cmsg_type != IP_PKTINFO) { | |
490 | if (net_ratelimit()) | |
491 | printk("rpcsvc: received unknown control message:" | |
492 | "%d/%d\n", | |
493 | cmh->cmsg_level, cmh->cmsg_type); | |
494 | skb_free_datagram(svsk->sk_sk, skb); | |
495 | return 0; | |
496 | } | |
497 | svc_udp_get_dest_address(rqstp, cmh); | |
1da177e4 LT |
498 | |
499 | if (skb_is_nonlinear(skb)) { | |
500 | /* we have to copy */ | |
501 | local_bh_disable(); | |
502 | if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) { | |
503 | local_bh_enable(); | |
504 | /* checksum error */ | |
505 | skb_free_datagram(svsk->sk_sk, skb); | |
506 | return 0; | |
507 | } | |
508 | local_bh_enable(); | |
cca5172a | 509 | skb_free_datagram(svsk->sk_sk, skb); |
1da177e4 LT |
510 | } else { |
511 | /* we can use it in-place */ | |
0f0257ea TT |
512 | rqstp->rq_arg.head[0].iov_base = skb->data + |
513 | sizeof(struct udphdr); | |
1da177e4 | 514 | rqstp->rq_arg.head[0].iov_len = len; |
fb286bb2 HX |
515 | if (skb_checksum_complete(skb)) { |
516 | skb_free_datagram(svsk->sk_sk, skb); | |
517 | return 0; | |
1da177e4 | 518 | } |
5148bf4e | 519 | rqstp->rq_xprt_ctxt = skb; |
1da177e4 LT |
520 | } |
521 | ||
522 | rqstp->rq_arg.page_base = 0; | |
523 | if (len <= rqstp->rq_arg.head[0].iov_len) { | |
524 | rqstp->rq_arg.head[0].iov_len = len; | |
525 | rqstp->rq_arg.page_len = 0; | |
44524359 | 526 | rqstp->rq_respages = rqstp->rq_pages+1; |
1da177e4 LT |
527 | } else { |
528 | rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len; | |
44524359 | 529 | rqstp->rq_respages = rqstp->rq_pages + 1 + |
172589cc | 530 | DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE); |
1da177e4 LT |
531 | } |
532 | ||
533 | if (serv->sv_stats) | |
534 | serv->sv_stats->netudpcnt++; | |
535 | ||
536 | return len; | |
537 | } | |
538 | ||
539 | static int | |
540 | svc_udp_sendto(struct svc_rqst *rqstp) | |
541 | { | |
542 | int error; | |
543 | ||
544 | error = svc_sendto(rqstp, &rqstp->rq_res); | |
545 | if (error == -ECONNREFUSED) | |
546 | /* ICMP error on earlier request. */ | |
547 | error = svc_sendto(rqstp, &rqstp->rq_res); | |
548 | ||
549 | return error; | |
550 | } | |
551 | ||
e831fe65 TT |
552 | static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp) |
553 | { | |
554 | } | |
555 | ||
323bee32 TT |
556 | static int svc_udp_has_wspace(struct svc_xprt *xprt) |
557 | { | |
558 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
bb5cf160 | 559 | struct svc_serv *serv = xprt->xpt_server; |
323bee32 TT |
560 | unsigned long required; |
561 | ||
562 | /* | |
563 | * Set the SOCK_NOSPACE flag before checking the available | |
564 | * sock space. | |
565 | */ | |
566 | set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
7a90e8cc | 567 | required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg; |
323bee32 TT |
568 | if (required*2 > sock_wspace(svsk->sk_sk)) |
569 | return 0; | |
570 | clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
571 | return 1; | |
572 | } | |
573 | ||
38a417cc TT |
574 | static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt) |
575 | { | |
576 | BUG(); | |
577 | return NULL; | |
578 | } | |
579 | ||
b700cbb1 TT |
580 | static struct svc_xprt *svc_udp_create(struct svc_serv *serv, |
581 | struct sockaddr *sa, int salen, | |
582 | int flags) | |
583 | { | |
584 | return svc_create_socket(serv, IPPROTO_UDP, sa, salen, flags); | |
585 | } | |
586 | ||
360d8738 | 587 | static struct svc_xprt_ops svc_udp_ops = { |
b700cbb1 | 588 | .xpo_create = svc_udp_create, |
5d137990 TT |
589 | .xpo_recvfrom = svc_udp_recvfrom, |
590 | .xpo_sendto = svc_udp_sendto, | |
5148bf4e | 591 | .xpo_release_rqst = svc_release_skb, |
755cceab TT |
592 | .xpo_detach = svc_sock_detach, |
593 | .xpo_free = svc_sock_free, | |
e831fe65 | 594 | .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr, |
323bee32 | 595 | .xpo_has_wspace = svc_udp_has_wspace, |
38a417cc | 596 | .xpo_accept = svc_udp_accept, |
360d8738 TT |
597 | }; |
598 | ||
599 | static struct svc_xprt_class svc_udp_class = { | |
600 | .xcl_name = "udp", | |
b700cbb1 | 601 | .xcl_owner = THIS_MODULE, |
360d8738 | 602 | .xcl_ops = &svc_udp_ops, |
49023155 | 603 | .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP, |
360d8738 TT |
604 | }; |
605 | ||
bb5cf160 | 606 | static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv) |
1da177e4 | 607 | { |
7a37f578 N |
608 | int one = 1; |
609 | mm_segment_t oldfs; | |
610 | ||
bb5cf160 | 611 | svc_xprt_init(&svc_udp_class, &svsk->sk_xprt, serv); |
def13d74 | 612 | clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
613 | svsk->sk_sk->sk_data_ready = svc_udp_data_ready; |
614 | svsk->sk_sk->sk_write_space = svc_write_space; | |
1da177e4 LT |
615 | |
616 | /* initialise setting must have enough space to | |
cca5172a | 617 | * receive and respond to one request. |
1da177e4 LT |
618 | * svc_udp_recvfrom will re-adjust if necessary |
619 | */ | |
620 | svc_sock_setbufsize(svsk->sk_sock, | |
bb5cf160 TT |
621 | 3 * svsk->sk_xprt.xpt_server->sv_max_mesg, |
622 | 3 * svsk->sk_xprt.xpt_server->sv_max_mesg); | |
1da177e4 | 623 | |
0f0257ea TT |
624 | /* data might have come in before data_ready set up */ |
625 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); | |
02fc6c36 | 626 | set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); |
7a37f578 N |
627 | |
628 | oldfs = get_fs(); | |
629 | set_fs(KERNEL_DS); | |
630 | /* make sure we get destination address info */ | |
631 | svsk->sk_sock->ops->setsockopt(svsk->sk_sock, IPPROTO_IP, IP_PKTINFO, | |
632 | (char __user *)&one, sizeof(one)); | |
633 | set_fs(oldfs); | |
1da177e4 LT |
634 | } |
635 | ||
636 | /* | |
637 | * A data_ready event on a listening socket means there's a connection | |
638 | * pending. Do not use state_change as a substitute for it. | |
639 | */ | |
0f0257ea | 640 | static void svc_tcp_listen_data_ready(struct sock *sk, int count_unused) |
1da177e4 | 641 | { |
939bb7ef | 642 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 LT |
643 | |
644 | dprintk("svc: socket %p TCP (listen) state change %d\n", | |
939bb7ef | 645 | sk, sk->sk_state); |
1da177e4 | 646 | |
939bb7ef NB |
647 | /* |
648 | * This callback may called twice when a new connection | |
649 | * is established as a child socket inherits everything | |
650 | * from a parent LISTEN socket. | |
651 | * 1) data_ready method of the parent socket will be called | |
652 | * when one of child sockets become ESTABLISHED. | |
653 | * 2) data_ready method of the child socket may be called | |
654 | * when it receives data before the socket is accepted. | |
655 | * In case of 2, we should ignore it silently. | |
656 | */ | |
657 | if (sk->sk_state == TCP_LISTEN) { | |
658 | if (svsk) { | |
02fc6c36 | 659 | set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
f6150c3c | 660 | svc_xprt_enqueue(&svsk->sk_xprt); |
939bb7ef NB |
661 | } else |
662 | printk("svc: socket %p: no user data\n", sk); | |
1da177e4 | 663 | } |
939bb7ef | 664 | |
1da177e4 LT |
665 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
666 | wake_up_interruptible_all(sk->sk_sleep); | |
667 | } | |
668 | ||
669 | /* | |
670 | * A state change on a connected socket means it's dying or dead. | |
671 | */ | |
0f0257ea | 672 | static void svc_tcp_state_change(struct sock *sk) |
1da177e4 | 673 | { |
939bb7ef | 674 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 LT |
675 | |
676 | dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n", | |
939bb7ef | 677 | sk, sk->sk_state, sk->sk_user_data); |
1da177e4 | 678 | |
939bb7ef | 679 | if (!svsk) |
1da177e4 | 680 | printk("svc: socket %p: no user data\n", sk); |
939bb7ef | 681 | else { |
02fc6c36 | 682 | set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); |
f6150c3c | 683 | svc_xprt_enqueue(&svsk->sk_xprt); |
1da177e4 | 684 | } |
1da177e4 LT |
685 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
686 | wake_up_interruptible_all(sk->sk_sleep); | |
687 | } | |
688 | ||
0f0257ea | 689 | static void svc_tcp_data_ready(struct sock *sk, int count) |
1da177e4 | 690 | { |
939bb7ef | 691 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 LT |
692 | |
693 | dprintk("svc: socket %p TCP data ready (svsk %p)\n", | |
939bb7ef NB |
694 | sk, sk->sk_user_data); |
695 | if (svsk) { | |
02fc6c36 | 696 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
f6150c3c | 697 | svc_xprt_enqueue(&svsk->sk_xprt); |
939bb7ef | 698 | } |
1da177e4 LT |
699 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
700 | wake_up_interruptible(sk->sk_sleep); | |
701 | } | |
702 | ||
703 | /* | |
704 | * Accept a TCP connection | |
705 | */ | |
38a417cc | 706 | static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt) |
1da177e4 | 707 | { |
38a417cc | 708 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); |
cdd88b9f | 709 | struct sockaddr_storage addr; |
710 | struct sockaddr *sin = (struct sockaddr *) &addr; | |
bb5cf160 | 711 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
1da177e4 LT |
712 | struct socket *sock = svsk->sk_sock; |
713 | struct socket *newsock; | |
1da177e4 LT |
714 | struct svc_sock *newsvsk; |
715 | int err, slen; | |
5216a8e7 | 716 | RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); |
1da177e4 LT |
717 | |
718 | dprintk("svc: tcp_accept %p sock %p\n", svsk, sock); | |
719 | if (!sock) | |
38a417cc | 720 | return NULL; |
1da177e4 | 721 | |
02fc6c36 | 722 | clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
e6242e92 SS |
723 | err = kernel_accept(sock, &newsock, O_NONBLOCK); |
724 | if (err < 0) { | |
1da177e4 LT |
725 | if (err == -ENOMEM) |
726 | printk(KERN_WARNING "%s: no more sockets!\n", | |
727 | serv->sv_name); | |
e6242e92 | 728 | else if (err != -EAGAIN && net_ratelimit()) |
1da177e4 LT |
729 | printk(KERN_WARNING "%s: accept failed (err %d)!\n", |
730 | serv->sv_name, -err); | |
38a417cc | 731 | return NULL; |
1da177e4 | 732 | } |
02fc6c36 | 733 | set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 734 | |
cdd88b9f | 735 | err = kernel_getpeername(newsock, sin, &slen); |
1da177e4 LT |
736 | if (err < 0) { |
737 | if (net_ratelimit()) | |
738 | printk(KERN_WARNING "%s: peername failed (err %d)!\n", | |
739 | serv->sv_name, -err); | |
740 | goto failed; /* aborted connection or whatever */ | |
741 | } | |
742 | ||
743 | /* Ideally, we would want to reject connections from unauthorized | |
ad06e4bd CL |
744 | * hosts here, but when we get encryption, the IP of the host won't |
745 | * tell us anything. For now just warn about unpriv connections. | |
1da177e4 | 746 | */ |
cdd88b9f | 747 | if (!svc_port_is_privileged(sin)) { |
1da177e4 | 748 | dprintk(KERN_WARNING |
ad06e4bd | 749 | "%s: connect from unprivileged port: %s\n", |
cca5172a | 750 | serv->sv_name, |
cdd88b9f | 751 | __svc_print_addr(sin, buf, sizeof(buf))); |
1da177e4 | 752 | } |
ad06e4bd | 753 | dprintk("%s: connect from %s\n", serv->sv_name, |
cdd88b9f | 754 | __svc_print_addr(sin, buf, sizeof(buf))); |
1da177e4 LT |
755 | |
756 | /* make sure that a write doesn't block forever when | |
757 | * low on memory | |
758 | */ | |
759 | newsock->sk->sk_sndtimeo = HZ*30; | |
760 | ||
6b174337 CL |
761 | if (!(newsvsk = svc_setup_socket(serv, newsock, &err, |
762 | (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY)))) | |
1da177e4 | 763 | goto failed; |
9dbc240f | 764 | svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen); |
a9747692 FM |
765 | err = kernel_getsockname(newsock, sin, &slen); |
766 | if (unlikely(err < 0)) { | |
767 | dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err); | |
768 | slen = offsetof(struct sockaddr, sa_data); | |
769 | } | |
9dbc240f | 770 | svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen); |
067d7817 | 771 | |
f9f3cc4f TT |
772 | if (serv->sv_stats) |
773 | serv->sv_stats->nettcpconn++; | |
774 | ||
775 | return &newsvsk->sk_xprt; | |
776 | ||
777 | failed: | |
778 | sock_release(newsock); | |
779 | return NULL; | |
780 | } | |
781 | ||
1da177e4 LT |
782 | /* |
783 | * Receive data from a TCP socket. | |
784 | */ | |
0f0257ea | 785 | static int svc_tcp_recvfrom(struct svc_rqst *rqstp) |
1da177e4 | 786 | { |
57b1d3ba TT |
787 | struct svc_sock *svsk = |
788 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
bb5cf160 | 789 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
1da177e4 | 790 | int len; |
3cc03b16 | 791 | struct kvec *vec; |
1da177e4 LT |
792 | int pnum, vlen; |
793 | ||
794 | dprintk("svc: tcp_recv %p data %d conn %d close %d\n", | |
02fc6c36 TT |
795 | svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags), |
796 | test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags), | |
797 | test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags)); | |
1da177e4 | 798 | |
02fc6c36 | 799 | clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
800 | |
801 | /* Receive data. If we haven't got the record length yet, get | |
802 | * the next four bytes. Otherwise try to gobble up as much as | |
803 | * possible up to the complete record length. | |
804 | */ | |
c0401ea0 CL |
805 | if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) { |
806 | int want = sizeof(rpc_fraghdr) - svsk->sk_tcplen; | |
1da177e4 LT |
807 | struct kvec iov; |
808 | ||
809 | iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen; | |
810 | iov.iov_len = want; | |
811 | if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0) | |
812 | goto error; | |
813 | svsk->sk_tcplen += len; | |
814 | ||
815 | if (len < want) { | |
c0401ea0 CL |
816 | dprintk("svc: short recvfrom while reading record " |
817 | "length (%d of %d)\n", len, want); | |
a6046f71 | 818 | svc_xprt_received(&svsk->sk_xprt); |
1da177e4 LT |
819 | return -EAGAIN; /* record header not complete */ |
820 | } | |
821 | ||
822 | svsk->sk_reclen = ntohl(svsk->sk_reclen); | |
c0401ea0 | 823 | if (!(svsk->sk_reclen & RPC_LAST_STREAM_FRAGMENT)) { |
1da177e4 LT |
824 | /* FIXME: technically, a record can be fragmented, |
825 | * and non-terminal fragments will not have the top | |
826 | * bit set in the fragment length header. | |
827 | * But apparently no known nfs clients send fragmented | |
828 | * records. */ | |
34e9a63b | 829 | if (net_ratelimit()) |
c0401ea0 CL |
830 | printk(KERN_NOTICE "RPC: multiple fragments " |
831 | "per record not supported\n"); | |
1da177e4 LT |
832 | goto err_delete; |
833 | } | |
c0401ea0 | 834 | svsk->sk_reclen &= RPC_FRAGMENT_SIZE_MASK; |
1da177e4 | 835 | dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen); |
c6b0a9f8 | 836 | if (svsk->sk_reclen > serv->sv_max_mesg) { |
34e9a63b | 837 | if (net_ratelimit()) |
c0401ea0 CL |
838 | printk(KERN_NOTICE "RPC: " |
839 | "fragment too large: 0x%08lx\n", | |
840 | (unsigned long)svsk->sk_reclen); | |
1da177e4 LT |
841 | goto err_delete; |
842 | } | |
843 | } | |
844 | ||
845 | /* Check whether enough data is available */ | |
846 | len = svc_recv_available(svsk); | |
847 | if (len < 0) | |
848 | goto error; | |
849 | ||
850 | if (len < svsk->sk_reclen) { | |
851 | dprintk("svc: incomplete TCP record (%d of %d)\n", | |
852 | len, svsk->sk_reclen); | |
a6046f71 | 853 | svc_xprt_received(&svsk->sk_xprt); |
1da177e4 LT |
854 | return -EAGAIN; /* record not complete */ |
855 | } | |
856 | len = svsk->sk_reclen; | |
02fc6c36 | 857 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 858 | |
3cc03b16 | 859 | vec = rqstp->rq_vec; |
1da177e4 LT |
860 | vec[0] = rqstp->rq_arg.head[0]; |
861 | vlen = PAGE_SIZE; | |
862 | pnum = 1; | |
863 | while (vlen < len) { | |
44524359 | 864 | vec[pnum].iov_base = page_address(rqstp->rq_pages[pnum]); |
1da177e4 LT |
865 | vec[pnum].iov_len = PAGE_SIZE; |
866 | pnum++; | |
867 | vlen += PAGE_SIZE; | |
868 | } | |
44524359 | 869 | rqstp->rq_respages = &rqstp->rq_pages[pnum]; |
1da177e4 LT |
870 | |
871 | /* Now receive data */ | |
872 | len = svc_recvfrom(rqstp, vec, pnum, len); | |
873 | if (len < 0) | |
874 | goto error; | |
875 | ||
876 | dprintk("svc: TCP complete record (%d bytes)\n", len); | |
877 | rqstp->rq_arg.len = len; | |
878 | rqstp->rq_arg.page_base = 0; | |
879 | if (len <= rqstp->rq_arg.head[0].iov_len) { | |
880 | rqstp->rq_arg.head[0].iov_len = len; | |
881 | rqstp->rq_arg.page_len = 0; | |
882 | } else { | |
883 | rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len; | |
884 | } | |
885 | ||
5148bf4e | 886 | rqstp->rq_xprt_ctxt = NULL; |
1da177e4 LT |
887 | rqstp->rq_prot = IPPROTO_TCP; |
888 | ||
889 | /* Reset TCP read info */ | |
890 | svsk->sk_reclen = 0; | |
891 | svsk->sk_tcplen = 0; | |
892 | ||
9dbc240f | 893 | svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt); |
a6046f71 | 894 | svc_xprt_received(&svsk->sk_xprt); |
1da177e4 LT |
895 | if (serv->sv_stats) |
896 | serv->sv_stats->nettcpcnt++; | |
897 | ||
898 | return len; | |
899 | ||
900 | err_delete: | |
02fc6c36 | 901 | set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
902 | return -EAGAIN; |
903 | ||
904 | error: | |
905 | if (len == -EAGAIN) { | |
906 | dprintk("RPC: TCP recvfrom got EAGAIN\n"); | |
a6046f71 | 907 | svc_xprt_received(&svsk->sk_xprt); |
1da177e4 LT |
908 | } else { |
909 | printk(KERN_NOTICE "%s: recvfrom returned errno %d\n", | |
bb5cf160 | 910 | svsk->sk_xprt.xpt_server->sv_name, -len); |
93fbf1a5 | 911 | goto err_delete; |
1da177e4 LT |
912 | } |
913 | ||
914 | return len; | |
915 | } | |
916 | ||
917 | /* | |
918 | * Send out data on TCP socket. | |
919 | */ | |
0f0257ea | 920 | static int svc_tcp_sendto(struct svc_rqst *rqstp) |
1da177e4 LT |
921 | { |
922 | struct xdr_buf *xbufp = &rqstp->rq_res; | |
923 | int sent; | |
d8ed029d | 924 | __be32 reclen; |
1da177e4 LT |
925 | |
926 | /* Set up the first element of the reply kvec. | |
927 | * Any other kvecs that may be in use have been taken | |
928 | * care of by the server implementation itself. | |
929 | */ | |
930 | reclen = htonl(0x80000000|((xbufp->len ) - 4)); | |
931 | memcpy(xbufp->head[0].iov_base, &reclen, 4); | |
932 | ||
57b1d3ba | 933 | if (test_bit(XPT_DEAD, &rqstp->rq_xprt->xpt_flags)) |
1da177e4 LT |
934 | return -ENOTCONN; |
935 | ||
936 | sent = svc_sendto(rqstp, &rqstp->rq_res); | |
937 | if (sent != xbufp->len) { | |
0f0257ea TT |
938 | printk(KERN_NOTICE |
939 | "rpc-srv/tcp: %s: %s %d when sending %d bytes " | |
940 | "- shutting down socket\n", | |
57b1d3ba | 941 | rqstp->rq_xprt->xpt_server->sv_name, |
1da177e4 LT |
942 | (sent<0)?"got error":"sent only", |
943 | sent, xbufp->len); | |
57b1d3ba | 944 | set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags); |
f6150c3c | 945 | svc_xprt_enqueue(rqstp->rq_xprt); |
1da177e4 LT |
946 | sent = -EAGAIN; |
947 | } | |
948 | return sent; | |
949 | } | |
950 | ||
e831fe65 TT |
951 | /* |
952 | * Setup response header. TCP has a 4B record length field. | |
953 | */ | |
954 | static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp) | |
955 | { | |
956 | struct kvec *resv = &rqstp->rq_res.head[0]; | |
957 | ||
958 | /* tcp needs a space for the record length... */ | |
959 | svc_putnl(resv, 0); | |
960 | } | |
961 | ||
323bee32 TT |
962 | static int svc_tcp_has_wspace(struct svc_xprt *xprt) |
963 | { | |
57b1d3ba | 964 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); |
bb5cf160 | 965 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
323bee32 TT |
966 | int required; |
967 | int wspace; | |
968 | ||
969 | /* | |
970 | * Set the SOCK_NOSPACE flag before checking the available | |
971 | * sock space. | |
972 | */ | |
973 | set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
7a90e8cc | 974 | required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg; |
323bee32 TT |
975 | wspace = sk_stream_wspace(svsk->sk_sk); |
976 | ||
977 | if (wspace < sk_stream_min_wspace(svsk->sk_sk)) | |
978 | return 0; | |
979 | if (required * 2 > wspace) | |
980 | return 0; | |
981 | ||
982 | clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
983 | return 1; | |
984 | } | |
985 | ||
b700cbb1 TT |
986 | static struct svc_xprt *svc_tcp_create(struct svc_serv *serv, |
987 | struct sockaddr *sa, int salen, | |
988 | int flags) | |
989 | { | |
990 | return svc_create_socket(serv, IPPROTO_TCP, sa, salen, flags); | |
991 | } | |
992 | ||
360d8738 | 993 | static struct svc_xprt_ops svc_tcp_ops = { |
b700cbb1 | 994 | .xpo_create = svc_tcp_create, |
5d137990 TT |
995 | .xpo_recvfrom = svc_tcp_recvfrom, |
996 | .xpo_sendto = svc_tcp_sendto, | |
5148bf4e | 997 | .xpo_release_rqst = svc_release_skb, |
69b6ba37 | 998 | .xpo_detach = svc_tcp_sock_detach, |
755cceab | 999 | .xpo_free = svc_sock_free, |
e831fe65 | 1000 | .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr, |
323bee32 | 1001 | .xpo_has_wspace = svc_tcp_has_wspace, |
38a417cc | 1002 | .xpo_accept = svc_tcp_accept, |
360d8738 TT |
1003 | }; |
1004 | ||
1005 | static struct svc_xprt_class svc_tcp_class = { | |
1006 | .xcl_name = "tcp", | |
b700cbb1 | 1007 | .xcl_owner = THIS_MODULE, |
360d8738 | 1008 | .xcl_ops = &svc_tcp_ops, |
49023155 | 1009 | .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP, |
360d8738 TT |
1010 | }; |
1011 | ||
1012 | void svc_init_xprt_sock(void) | |
1013 | { | |
1014 | svc_reg_xprt_class(&svc_tcp_class); | |
1015 | svc_reg_xprt_class(&svc_udp_class); | |
1016 | } | |
1017 | ||
1018 | void svc_cleanup_xprt_sock(void) | |
1019 | { | |
1020 | svc_unreg_xprt_class(&svc_tcp_class); | |
1021 | svc_unreg_xprt_class(&svc_udp_class); | |
1022 | } | |
1023 | ||
bb5cf160 | 1024 | static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv) |
1da177e4 LT |
1025 | { |
1026 | struct sock *sk = svsk->sk_sk; | |
1da177e4 | 1027 | |
bb5cf160 | 1028 | svc_xprt_init(&svc_tcp_class, &svsk->sk_xprt, serv); |
def13d74 | 1029 | set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
1030 | if (sk->sk_state == TCP_LISTEN) { |
1031 | dprintk("setting up TCP socket for listening\n"); | |
02fc6c36 | 1032 | set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 1033 | sk->sk_data_ready = svc_tcp_listen_data_ready; |
02fc6c36 | 1034 | set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
1035 | } else { |
1036 | dprintk("setting up TCP socket for reading\n"); | |
1037 | sk->sk_state_change = svc_tcp_state_change; | |
1038 | sk->sk_data_ready = svc_tcp_data_ready; | |
1039 | sk->sk_write_space = svc_write_space; | |
1040 | ||
1041 | svsk->sk_reclen = 0; | |
1042 | svsk->sk_tcplen = 0; | |
1043 | ||
b7872fe8 | 1044 | tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF; |
1da177e4 | 1045 | |
02fc6c36 | 1046 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
cca5172a | 1047 | if (sk->sk_state != TCP_ESTABLISHED) |
02fc6c36 | 1048 | set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
1049 | } |
1050 | } | |
1051 | ||
0f0257ea | 1052 | void svc_sock_update_bufs(struct svc_serv *serv) |
1da177e4 LT |
1053 | { |
1054 | /* | |
1055 | * The number of server threads has changed. Update | |
1056 | * rcvbuf and sndbuf accordingly on all sockets | |
1057 | */ | |
1058 | struct list_head *le; | |
1059 | ||
1060 | spin_lock_bh(&serv->sv_lock); | |
1061 | list_for_each(le, &serv->sv_permsocks) { | |
cca5172a | 1062 | struct svc_sock *svsk = |
7a182083 | 1063 | list_entry(le, struct svc_sock, sk_xprt.xpt_list); |
02fc6c36 | 1064 | set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
1065 | } |
1066 | list_for_each(le, &serv->sv_tempsocks) { | |
1067 | struct svc_sock *svsk = | |
7a182083 | 1068 | list_entry(le, struct svc_sock, sk_xprt.xpt_list); |
02fc6c36 | 1069 | set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
1070 | } |
1071 | spin_unlock_bh(&serv->sv_lock); | |
1072 | } | |
24c3767e | 1073 | EXPORT_SYMBOL_GPL(svc_sock_update_bufs); |
1da177e4 | 1074 | |
1da177e4 LT |
1075 | /* |
1076 | * Initialize socket for RPC use and create svc_sock struct | |
1077 | * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF. | |
1078 | */ | |
6b174337 CL |
1079 | static struct svc_sock *svc_setup_socket(struct svc_serv *serv, |
1080 | struct socket *sock, | |
1081 | int *errp, int flags) | |
1da177e4 LT |
1082 | { |
1083 | struct svc_sock *svsk; | |
1084 | struct sock *inet; | |
6b174337 | 1085 | int pmap_register = !(flags & SVC_SOCK_ANONYMOUS); |
1da177e4 LT |
1086 | |
1087 | dprintk("svc: svc_setup_socket %p\n", sock); | |
0da974f4 | 1088 | if (!(svsk = kzalloc(sizeof(*svsk), GFP_KERNEL))) { |
1da177e4 LT |
1089 | *errp = -ENOMEM; |
1090 | return NULL; | |
1091 | } | |
1da177e4 LT |
1092 | |
1093 | inet = sock->sk; | |
1094 | ||
1095 | /* Register socket with portmapper */ | |
1096 | if (*errp >= 0 && pmap_register) | |
baf01caf | 1097 | *errp = svc_register(serv, inet->sk_family, inet->sk_protocol, |
1da177e4 LT |
1098 | ntohs(inet_sk(inet)->sport)); |
1099 | ||
1100 | if (*errp < 0) { | |
1101 | kfree(svsk); | |
1102 | return NULL; | |
1103 | } | |
1104 | ||
1da177e4 LT |
1105 | inet->sk_user_data = svsk; |
1106 | svsk->sk_sock = sock; | |
1107 | svsk->sk_sk = inet; | |
1108 | svsk->sk_ostate = inet->sk_state_change; | |
1109 | svsk->sk_odata = inet->sk_data_ready; | |
1110 | svsk->sk_owspace = inet->sk_write_space; | |
1da177e4 LT |
1111 | |
1112 | /* Initialize the socket */ | |
1113 | if (sock->type == SOCK_DGRAM) | |
bb5cf160 | 1114 | svc_udp_init(svsk, serv); |
47a14ef1 OK |
1115 | else { |
1116 | /* initialise setting must have enough space to | |
1117 | * receive and respond to one request. | |
1118 | */ | |
1119 | svc_sock_setbufsize(svsk->sk_sock, 4 * serv->sv_max_mesg, | |
1120 | 4 * serv->sv_max_mesg); | |
bb5cf160 | 1121 | svc_tcp_init(svsk, serv); |
47a14ef1 | 1122 | } |
1da177e4 | 1123 | |
1da177e4 LT |
1124 | dprintk("svc: svc_setup_socket created %p (inet %p)\n", |
1125 | svsk, svsk->sk_sk); | |
1126 | ||
1da177e4 LT |
1127 | return svsk; |
1128 | } | |
1129 | ||
b41b66d6 N |
1130 | int svc_addsock(struct svc_serv *serv, |
1131 | int fd, | |
29373913 | 1132 | char *name_return) |
b41b66d6 N |
1133 | { |
1134 | int err = 0; | |
1135 | struct socket *so = sockfd_lookup(fd, &err); | |
1136 | struct svc_sock *svsk = NULL; | |
1137 | ||
1138 | if (!so) | |
1139 | return err; | |
1140 | if (so->sk->sk_family != AF_INET) | |
1141 | err = -EAFNOSUPPORT; | |
1142 | else if (so->sk->sk_protocol != IPPROTO_TCP && | |
1143 | so->sk->sk_protocol != IPPROTO_UDP) | |
1144 | err = -EPROTONOSUPPORT; | |
1145 | else if (so->state > SS_UNCONNECTED) | |
1146 | err = -EISCONN; | |
1147 | else { | |
2da2c21d TT |
1148 | if (!try_module_get(THIS_MODULE)) |
1149 | err = -ENOENT; | |
1150 | else | |
1151 | svsk = svc_setup_socket(serv, so, &err, | |
1152 | SVC_SOCK_DEFAULTS); | |
e79eff1f | 1153 | if (svsk) { |
9dbc240f TT |
1154 | struct sockaddr_storage addr; |
1155 | struct sockaddr *sin = (struct sockaddr *)&addr; | |
1156 | int salen; | |
1157 | if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0) | |
1158 | svc_xprt_set_local(&svsk->sk_xprt, sin, salen); | |
4e5caaa5 TT |
1159 | clear_bit(XPT_TEMP, &svsk->sk_xprt.xpt_flags); |
1160 | spin_lock_bh(&serv->sv_lock); | |
1161 | list_add(&svsk->sk_xprt.xpt_list, &serv->sv_permsocks); | |
1162 | spin_unlock_bh(&serv->sv_lock); | |
a6046f71 | 1163 | svc_xprt_received(&svsk->sk_xprt); |
b41b66d6 | 1164 | err = 0; |
2da2c21d TT |
1165 | } else |
1166 | module_put(THIS_MODULE); | |
b41b66d6 N |
1167 | } |
1168 | if (err) { | |
1169 | sockfd_put(so); | |
1170 | return err; | |
1171 | } | |
b41b66d6 N |
1172 | return one_sock_name(name_return, svsk); |
1173 | } | |
1174 | EXPORT_SYMBOL_GPL(svc_addsock); | |
1175 | ||
1da177e4 LT |
1176 | /* |
1177 | * Create socket for RPC service. | |
1178 | */ | |
b700cbb1 TT |
1179 | static struct svc_xprt *svc_create_socket(struct svc_serv *serv, |
1180 | int protocol, | |
1181 | struct sockaddr *sin, int len, | |
1182 | int flags) | |
1da177e4 LT |
1183 | { |
1184 | struct svc_sock *svsk; | |
1185 | struct socket *sock; | |
1186 | int error; | |
1187 | int type; | |
9dbc240f TT |
1188 | struct sockaddr_storage addr; |
1189 | struct sockaddr *newsin = (struct sockaddr *)&addr; | |
1190 | int newlen; | |
c69da774 TM |
1191 | int family; |
1192 | int val; | |
5216a8e7 | 1193 | RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); |
1da177e4 | 1194 | |
ad06e4bd CL |
1195 | dprintk("svc: svc_create_socket(%s, %d, %s)\n", |
1196 | serv->sv_program->pg_name, protocol, | |
77f1f67a | 1197 | __svc_print_addr(sin, buf, sizeof(buf))); |
1da177e4 LT |
1198 | |
1199 | if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) { | |
1200 | printk(KERN_WARNING "svc: only UDP and TCP " | |
1201 | "sockets supported\n"); | |
b700cbb1 | 1202 | return ERR_PTR(-EINVAL); |
1da177e4 | 1203 | } |
c69da774 | 1204 | |
1da177e4 | 1205 | type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM; |
c69da774 TM |
1206 | switch (sin->sa_family) { |
1207 | case AF_INET6: | |
1208 | family = PF_INET6; | |
1209 | break; | |
1210 | case AF_INET: | |
1211 | family = PF_INET; | |
1212 | break; | |
1213 | default: | |
1214 | return ERR_PTR(-EINVAL); | |
1215 | } | |
1da177e4 | 1216 | |
c69da774 | 1217 | error = sock_create_kern(family, type, protocol, &sock); |
77f1f67a | 1218 | if (error < 0) |
b700cbb1 | 1219 | return ERR_PTR(error); |
1da177e4 | 1220 | |
ed07536e PZ |
1221 | svc_reclassify_socket(sock); |
1222 | ||
c69da774 TM |
1223 | /* |
1224 | * If this is an PF_INET6 listener, we want to avoid | |
1225 | * getting requests from IPv4 remotes. Those should | |
1226 | * be shunted to a PF_INET listener via rpcbind. | |
1227 | */ | |
1228 | val = 1; | |
1229 | if (family == PF_INET6) | |
1230 | kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY, | |
1231 | (char *)&val, sizeof(val)); | |
1232 | ||
18114746 | 1233 | if (type == SOCK_STREAM) |
77f1f67a CL |
1234 | sock->sk->sk_reuse = 1; /* allow address reuse */ |
1235 | error = kernel_bind(sock, sin, len); | |
18114746 ES |
1236 | if (error < 0) |
1237 | goto bummer; | |
1da177e4 | 1238 | |
9dbc240f TT |
1239 | newlen = len; |
1240 | error = kernel_getsockname(sock, newsin, &newlen); | |
1241 | if (error < 0) | |
1242 | goto bummer; | |
1243 | ||
1da177e4 | 1244 | if (protocol == IPPROTO_TCP) { |
e6242e92 | 1245 | if ((error = kernel_listen(sock, 64)) < 0) |
1da177e4 LT |
1246 | goto bummer; |
1247 | } | |
1248 | ||
e79eff1f | 1249 | if ((svsk = svc_setup_socket(serv, sock, &error, flags)) != NULL) { |
9dbc240f | 1250 | svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen); |
b700cbb1 | 1251 | return (struct svc_xprt *)svsk; |
e79eff1f | 1252 | } |
1da177e4 LT |
1253 | |
1254 | bummer: | |
1255 | dprintk("svc: svc_create_socket error = %d\n", -error); | |
1256 | sock_release(sock); | |
b700cbb1 | 1257 | return ERR_PTR(error); |
1da177e4 LT |
1258 | } |
1259 | ||
755cceab TT |
1260 | /* |
1261 | * Detach the svc_sock from the socket so that no | |
1262 | * more callbacks occur. | |
1263 | */ | |
1264 | static void svc_sock_detach(struct svc_xprt *xprt) | |
1265 | { | |
1266 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
1267 | struct sock *sk = svsk->sk_sk; | |
1268 | ||
1269 | dprintk("svc: svc_sock_detach(%p)\n", svsk); | |
1270 | ||
1271 | /* put back the old socket callbacks */ | |
1272 | sk->sk_state_change = svsk->sk_ostate; | |
1273 | sk->sk_data_ready = svsk->sk_odata; | |
1274 | sk->sk_write_space = svsk->sk_owspace; | |
69b6ba37 TM |
1275 | |
1276 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) | |
1277 | wake_up_interruptible(sk->sk_sleep); | |
1278 | } | |
1279 | ||
1280 | /* | |
1281 | * Disconnect the socket, and reset the callbacks | |
1282 | */ | |
1283 | static void svc_tcp_sock_detach(struct svc_xprt *xprt) | |
1284 | { | |
1285 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
1286 | ||
1287 | dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk); | |
1288 | ||
1289 | svc_sock_detach(xprt); | |
1290 | ||
1291 | if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) | |
1292 | kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR); | |
755cceab TT |
1293 | } |
1294 | ||
1295 | /* | |
1296 | * Free the svc_sock's socket resources and the svc_sock itself. | |
1297 | */ | |
1298 | static void svc_sock_free(struct svc_xprt *xprt) | |
1299 | { | |
1300 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
1301 | dprintk("svc: svc_sock_free(%p)\n", svsk); | |
1302 | ||
755cceab TT |
1303 | if (svsk->sk_sock->file) |
1304 | sockfd_put(svsk->sk_sock); | |
1305 | else | |
1306 | sock_release(svsk->sk_sock); | |
1307 | kfree(svsk); | |
1308 | } |