Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/net/sunrpc/svcsock.c | |
3 | * | |
4 | * These are the RPC server socket internals. | |
5 | * | |
6 | * The server scheduling algorithm does not always distribute the load | |
7 | * evenly when servicing a single client. May need to modify the | |
f6150c3c | 8 | * svc_xprt_enqueue procedure... |
1da177e4 LT |
9 | * |
10 | * TCP support is largely untested and may be a little slow. The problem | |
11 | * is that we currently do two separate recvfrom's, one for the 4-byte | |
12 | * record length, and the second for the actual record. This could possibly | |
13 | * be improved by always reading a minimum size of around 100 bytes and | |
14 | * tucking any superfluous bytes away in a temporary store. Still, that | |
15 | * leaves write requests out in the rain. An alternative may be to peek at | |
16 | * the first skb in the queue, and if it matches the next TCP sequence | |
17 | * number, to extract the record marker. Yuck. | |
18 | * | |
19 | * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de> | |
20 | */ | |
21 | ||
172589cc | 22 | #include <linux/kernel.h> |
1da177e4 LT |
23 | #include <linux/sched.h> |
24 | #include <linux/errno.h> | |
25 | #include <linux/fcntl.h> | |
26 | #include <linux/net.h> | |
27 | #include <linux/in.h> | |
28 | #include <linux/inet.h> | |
29 | #include <linux/udp.h> | |
91483c4b | 30 | #include <linux/tcp.h> |
1da177e4 LT |
31 | #include <linux/unistd.h> |
32 | #include <linux/slab.h> | |
33 | #include <linux/netdevice.h> | |
34 | #include <linux/skbuff.h> | |
b41b66d6 | 35 | #include <linux/file.h> |
7dfb7103 | 36 | #include <linux/freezer.h> |
1da177e4 LT |
37 | #include <net/sock.h> |
38 | #include <net/checksum.h> | |
39 | #include <net/ip.h> | |
b92503b2 | 40 | #include <net/ipv6.h> |
b7872fe8 | 41 | #include <net/tcp.h> |
c752f073 | 42 | #include <net/tcp_states.h> |
1da177e4 LT |
43 | #include <asm/uaccess.h> |
44 | #include <asm/ioctls.h> | |
45 | ||
46 | #include <linux/sunrpc/types.h> | |
ad06e4bd | 47 | #include <linux/sunrpc/clnt.h> |
1da177e4 | 48 | #include <linux/sunrpc/xdr.h> |
c0401ea0 | 49 | #include <linux/sunrpc/msg_prot.h> |
1da177e4 LT |
50 | #include <linux/sunrpc/svcsock.h> |
51 | #include <linux/sunrpc/stats.h> | |
52 | ||
360d8738 | 53 | #define RPCDBG_FACILITY RPCDBG_SVCXPRT |
1da177e4 LT |
54 | |
55 | ||
56 | static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *, | |
6b174337 | 57 | int *errp, int flags); |
1da177e4 LT |
58 | static void svc_udp_data_ready(struct sock *, int); |
59 | static int svc_udp_recvfrom(struct svc_rqst *); | |
60 | static int svc_udp_sendto(struct svc_rqst *); | |
755cceab | 61 | static void svc_sock_detach(struct svc_xprt *); |
69b6ba37 | 62 | static void svc_tcp_sock_detach(struct svc_xprt *); |
755cceab | 63 | static void svc_sock_free(struct svc_xprt *); |
1da177e4 | 64 | |
b700cbb1 TT |
65 | static struct svc_xprt *svc_create_socket(struct svc_serv *, int, |
66 | struct sockaddr *, int, int); | |
ed07536e PZ |
67 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
68 | static struct lock_class_key svc_key[2]; | |
69 | static struct lock_class_key svc_slock_key[2]; | |
70 | ||
0f0257ea | 71 | static void svc_reclassify_socket(struct socket *sock) |
ed07536e PZ |
72 | { |
73 | struct sock *sk = sock->sk; | |
02b3d346 | 74 | BUG_ON(sock_owned_by_user(sk)); |
ed07536e PZ |
75 | switch (sk->sk_family) { |
76 | case AF_INET: | |
77 | sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD", | |
def13d74 TT |
78 | &svc_slock_key[0], |
79 | "sk_xprt.xpt_lock-AF_INET-NFSD", | |
80 | &svc_key[0]); | |
ed07536e PZ |
81 | break; |
82 | ||
83 | case AF_INET6: | |
84 | sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD", | |
def13d74 TT |
85 | &svc_slock_key[1], |
86 | "sk_xprt.xpt_lock-AF_INET6-NFSD", | |
87 | &svc_key[1]); | |
ed07536e PZ |
88 | break; |
89 | ||
90 | default: | |
91 | BUG(); | |
92 | } | |
93 | } | |
94 | #else | |
0f0257ea | 95 | static void svc_reclassify_socket(struct socket *sock) |
ed07536e PZ |
96 | { |
97 | } | |
98 | #endif | |
99 | ||
1da177e4 LT |
100 | /* |
101 | * Release an skbuff after use | |
102 | */ | |
5148bf4e | 103 | static void svc_release_skb(struct svc_rqst *rqstp) |
1da177e4 | 104 | { |
5148bf4e | 105 | struct sk_buff *skb = rqstp->rq_xprt_ctxt; |
1da177e4 LT |
106 | |
107 | if (skb) { | |
57b1d3ba TT |
108 | struct svc_sock *svsk = |
109 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
5148bf4e | 110 | rqstp->rq_xprt_ctxt = NULL; |
1da177e4 LT |
111 | |
112 | dprintk("svc: service %p, releasing skb %p\n", rqstp, skb); | |
57b1d3ba | 113 | skb_free_datagram(svsk->sk_sk, skb); |
1da177e4 | 114 | } |
1da177e4 LT |
115 | } |
116 | ||
b92503b2 CL |
117 | union svc_pktinfo_u { |
118 | struct in_pktinfo pkti; | |
b92503b2 | 119 | struct in6_pktinfo pkti6; |
b92503b2 | 120 | }; |
bc375ea7 DM |
121 | #define SVC_PKTINFO_SPACE \ |
122 | CMSG_SPACE(sizeof(union svc_pktinfo_u)) | |
b92503b2 CL |
123 | |
124 | static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh) | |
125 | { | |
57b1d3ba TT |
126 | struct svc_sock *svsk = |
127 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
128 | switch (svsk->sk_sk->sk_family) { | |
b92503b2 CL |
129 | case AF_INET: { |
130 | struct in_pktinfo *pki = CMSG_DATA(cmh); | |
131 | ||
132 | cmh->cmsg_level = SOL_IP; | |
133 | cmh->cmsg_type = IP_PKTINFO; | |
134 | pki->ipi_ifindex = 0; | |
135 | pki->ipi_spec_dst.s_addr = rqstp->rq_daddr.addr.s_addr; | |
136 | cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); | |
137 | } | |
138 | break; | |
5a05ed73 | 139 | |
b92503b2 CL |
140 | case AF_INET6: { |
141 | struct in6_pktinfo *pki = CMSG_DATA(cmh); | |
142 | ||
143 | cmh->cmsg_level = SOL_IPV6; | |
144 | cmh->cmsg_type = IPV6_PKTINFO; | |
145 | pki->ipi6_ifindex = 0; | |
146 | ipv6_addr_copy(&pki->ipi6_addr, | |
147 | &rqstp->rq_daddr.addr6); | |
148 | cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); | |
149 | } | |
150 | break; | |
b92503b2 CL |
151 | } |
152 | return; | |
153 | } | |
154 | ||
1da177e4 LT |
155 | /* |
156 | * Generic sendto routine | |
157 | */ | |
0f0257ea | 158 | static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr) |
1da177e4 | 159 | { |
57b1d3ba TT |
160 | struct svc_sock *svsk = |
161 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
1da177e4 LT |
162 | struct socket *sock = svsk->sk_sock; |
163 | int slen; | |
bc375ea7 DM |
164 | union { |
165 | struct cmsghdr hdr; | |
166 | long all[SVC_PKTINFO_SPACE / sizeof(long)]; | |
167 | } buffer; | |
168 | struct cmsghdr *cmh = &buffer.hdr; | |
1da177e4 LT |
169 | int len = 0; |
170 | int result; | |
171 | int size; | |
172 | struct page **ppage = xdr->pages; | |
173 | size_t base = xdr->page_base; | |
174 | unsigned int pglen = xdr->page_len; | |
175 | unsigned int flags = MSG_MORE; | |
5216a8e7 | 176 | RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); |
1da177e4 LT |
177 | |
178 | slen = xdr->len; | |
179 | ||
180 | if (rqstp->rq_prot == IPPROTO_UDP) { | |
b92503b2 CL |
181 | struct msghdr msg = { |
182 | .msg_name = &rqstp->rq_addr, | |
183 | .msg_namelen = rqstp->rq_addrlen, | |
184 | .msg_control = cmh, | |
185 | .msg_controllen = sizeof(buffer), | |
186 | .msg_flags = MSG_MORE, | |
187 | }; | |
188 | ||
189 | svc_set_cmsg_data(rqstp, cmh); | |
1da177e4 LT |
190 | |
191 | if (sock_sendmsg(sock, &msg, 0) < 0) | |
192 | goto out; | |
193 | } | |
194 | ||
195 | /* send head */ | |
196 | if (slen == xdr->head[0].iov_len) | |
197 | flags = 0; | |
44524359 N |
198 | len = kernel_sendpage(sock, rqstp->rq_respages[0], 0, |
199 | xdr->head[0].iov_len, flags); | |
1da177e4 LT |
200 | if (len != xdr->head[0].iov_len) |
201 | goto out; | |
202 | slen -= xdr->head[0].iov_len; | |
203 | if (slen == 0) | |
204 | goto out; | |
205 | ||
206 | /* send page data */ | |
207 | size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen; | |
208 | while (pglen > 0) { | |
209 | if (slen == size) | |
210 | flags = 0; | |
e6242e92 | 211 | result = kernel_sendpage(sock, *ppage, base, size, flags); |
1da177e4 LT |
212 | if (result > 0) |
213 | len += result; | |
214 | if (result != size) | |
215 | goto out; | |
216 | slen -= size; | |
217 | pglen -= size; | |
218 | size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen; | |
219 | base = 0; | |
220 | ppage++; | |
221 | } | |
222 | /* send tail */ | |
223 | if (xdr->tail[0].iov_len) { | |
44524359 N |
224 | result = kernel_sendpage(sock, rqstp->rq_respages[0], |
225 | ((unsigned long)xdr->tail[0].iov_base) | |
cca5172a | 226 | & (PAGE_SIZE-1), |
1da177e4 LT |
227 | xdr->tail[0].iov_len, 0); |
228 | ||
229 | if (result > 0) | |
230 | len += result; | |
231 | } | |
232 | out: | |
ad06e4bd | 233 | dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n", |
57b1d3ba | 234 | svsk, xdr->head[0].iov_base, xdr->head[0].iov_len, |
ad06e4bd | 235 | xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf))); |
1da177e4 LT |
236 | |
237 | return len; | |
238 | } | |
239 | ||
80212d59 N |
240 | /* |
241 | * Report socket names for nfsdfs | |
242 | */ | |
e7942b9f | 243 | static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining) |
80212d59 | 244 | { |
017cb47f CL |
245 | const struct sock *sk = svsk->sk_sk; |
246 | const char *proto_name = sk->sk_protocol == IPPROTO_UDP ? | |
247 | "udp" : "tcp"; | |
80212d59 N |
248 | int len; |
249 | ||
017cb47f | 250 | switch (sk->sk_family) { |
e7942b9f CL |
251 | case PF_INET: |
252 | len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n", | |
017cb47f CL |
253 | proto_name, |
254 | &inet_sk(sk)->rcv_saddr, | |
255 | inet_sk(sk)->num); | |
80212d59 | 256 | break; |
58de2f86 CL |
257 | case PF_INET6: |
258 | len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n", | |
017cb47f CL |
259 | proto_name, |
260 | &inet6_sk(sk)->rcv_saddr, | |
261 | inet_sk(sk)->num); | |
58de2f86 | 262 | break; |
80212d59 | 263 | default: |
e7942b9f | 264 | len = snprintf(buf, remaining, "*unknown-%d*\n", |
017cb47f | 265 | sk->sk_family); |
80212d59 | 266 | } |
e7942b9f CL |
267 | |
268 | if (len >= remaining) { | |
269 | *buf = '\0'; | |
270 | return -ENAMETOOLONG; | |
271 | } | |
80212d59 N |
272 | return len; |
273 | } | |
274 | ||
8435d34d CL |
275 | /** |
276 | * svc_sock_names - construct a list of listener names in a string | |
277 | * @serv: pointer to RPC service | |
278 | * @buf: pointer to a buffer to fill in with socket names | |
279 | * @buflen: size of the buffer to be filled | |
280 | * @toclose: pointer to '\0'-terminated C string containing the name | |
281 | * of a listener to be closed | |
282 | * | |
283 | * Fills in @buf with a '\n'-separated list of names of listener | |
284 | * sockets. If @toclose is not NULL, the socket named by @toclose | |
285 | * is closed, and is not included in the output list. | |
286 | * | |
287 | * Returns positive length of the socket name string, or a negative | |
288 | * errno value on error. | |
289 | */ | |
290 | int svc_sock_names(struct svc_serv *serv, char *buf, const size_t buflen, | |
291 | const char *toclose) | |
80212d59 | 292 | { |
b41b66d6 | 293 | struct svc_sock *svsk, *closesk = NULL; |
80212d59 N |
294 | int len = 0; |
295 | ||
296 | if (!serv) | |
297 | return 0; | |
e7942b9f | 298 | |
aaf68cfb | 299 | spin_lock_bh(&serv->sv_lock); |
7a182083 | 300 | list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list) { |
e7942b9f CL |
301 | int onelen = svc_one_sock_name(svsk, buf + len, buflen - len); |
302 | if (onelen < 0) { | |
303 | len = onelen; | |
304 | break; | |
305 | } | |
306 | if (toclose && strcmp(toclose, buf + len) == 0) | |
b41b66d6 N |
307 | closesk = svsk; |
308 | else | |
309 | len += onelen; | |
80212d59 | 310 | } |
aaf68cfb | 311 | spin_unlock_bh(&serv->sv_lock); |
e7942b9f | 312 | |
b41b66d6 | 313 | if (closesk) |
5680c446 N |
314 | /* Should unregister with portmap, but you cannot |
315 | * unregister just one protocol... | |
316 | */ | |
7a182083 | 317 | svc_close_xprt(&closesk->sk_xprt); |
37a03472 N |
318 | else if (toclose) |
319 | return -ENOENT; | |
80212d59 N |
320 | return len; |
321 | } | |
24c3767e | 322 | EXPORT_SYMBOL_GPL(svc_sock_names); |
80212d59 | 323 | |
1da177e4 LT |
324 | /* |
325 | * Check input queue length | |
326 | */ | |
0f0257ea | 327 | static int svc_recv_available(struct svc_sock *svsk) |
1da177e4 | 328 | { |
1da177e4 LT |
329 | struct socket *sock = svsk->sk_sock; |
330 | int avail, err; | |
331 | ||
e6242e92 | 332 | err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail); |
1da177e4 LT |
333 | |
334 | return (err >= 0)? avail : err; | |
335 | } | |
336 | ||
337 | /* | |
338 | * Generic recvfrom routine. | |
339 | */ | |
0f0257ea TT |
340 | static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr, |
341 | int buflen) | |
1da177e4 | 342 | { |
57b1d3ba TT |
343 | struct svc_sock *svsk = |
344 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
1ba95105 CL |
345 | struct msghdr msg = { |
346 | .msg_flags = MSG_DONTWAIT, | |
347 | }; | |
348 | int len; | |
1da177e4 | 349 | |
260c1d12 TT |
350 | rqstp->rq_xprt_hlen = 0; |
351 | ||
1ba95105 CL |
352 | len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen, |
353 | msg.msg_flags); | |
1da177e4 | 354 | |
1da177e4 | 355 | dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n", |
1ba95105 | 356 | svsk, iov[0].iov_base, iov[0].iov_len, len); |
1da177e4 LT |
357 | return len; |
358 | } | |
359 | ||
360 | /* | |
361 | * Set socket snd and rcv buffer lengths | |
362 | */ | |
0f0257ea TT |
363 | static void svc_sock_setbufsize(struct socket *sock, unsigned int snd, |
364 | unsigned int rcv) | |
1da177e4 LT |
365 | { |
366 | #if 0 | |
367 | mm_segment_t oldfs; | |
368 | oldfs = get_fs(); set_fs(KERNEL_DS); | |
369 | sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF, | |
370 | (char*)&snd, sizeof(snd)); | |
371 | sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF, | |
372 | (char*)&rcv, sizeof(rcv)); | |
373 | #else | |
374 | /* sock_setsockopt limits use to sysctl_?mem_max, | |
375 | * which isn't acceptable. Until that is made conditional | |
376 | * on not having CAP_SYS_RESOURCE or similar, we go direct... | |
377 | * DaveM said I could! | |
378 | */ | |
379 | lock_sock(sock->sk); | |
380 | sock->sk->sk_sndbuf = snd * 2; | |
381 | sock->sk->sk_rcvbuf = rcv * 2; | |
7f421835 | 382 | sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK|SOCK_RCVBUF_LOCK; |
1da177e4 LT |
383 | release_sock(sock->sk); |
384 | #endif | |
385 | } | |
386 | /* | |
387 | * INET callback when data has been received on the socket. | |
388 | */ | |
0f0257ea | 389 | static void svc_udp_data_ready(struct sock *sk, int count) |
1da177e4 | 390 | { |
939bb7ef | 391 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 | 392 | |
939bb7ef NB |
393 | if (svsk) { |
394 | dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n", | |
02fc6c36 TT |
395 | svsk, sk, count, |
396 | test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags)); | |
397 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); | |
f6150c3c | 398 | svc_xprt_enqueue(&svsk->sk_xprt); |
939bb7ef | 399 | } |
1da177e4 LT |
400 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
401 | wake_up_interruptible(sk->sk_sleep); | |
402 | } | |
403 | ||
404 | /* | |
405 | * INET callback when space is newly available on the socket. | |
406 | */ | |
0f0257ea | 407 | static void svc_write_space(struct sock *sk) |
1da177e4 LT |
408 | { |
409 | struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data); | |
410 | ||
411 | if (svsk) { | |
412 | dprintk("svc: socket %p(inet %p), write_space busy=%d\n", | |
02fc6c36 | 413 | svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags)); |
f6150c3c | 414 | svc_xprt_enqueue(&svsk->sk_xprt); |
1da177e4 LT |
415 | } |
416 | ||
417 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) { | |
939bb7ef | 418 | dprintk("RPC svc_write_space: someone sleeping on %p\n", |
1da177e4 LT |
419 | svsk); |
420 | wake_up_interruptible(sk->sk_sleep); | |
421 | } | |
422 | } | |
423 | ||
9dbc240f TT |
424 | /* |
425 | * Copy the UDP datagram's destination address to the rqstp structure. | |
426 | * The 'destination' address in this case is the address to which the | |
427 | * peer sent the datagram, i.e. our local address. For multihomed | |
428 | * hosts, this can change from msg to msg. Note that only the IP | |
429 | * address changes, the port number should remain the same. | |
430 | */ | |
431 | static void svc_udp_get_dest_address(struct svc_rqst *rqstp, | |
432 | struct cmsghdr *cmh) | |
95756482 | 433 | { |
57b1d3ba TT |
434 | struct svc_sock *svsk = |
435 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
436 | switch (svsk->sk_sk->sk_family) { | |
95756482 | 437 | case AF_INET: { |
7a37f578 N |
438 | struct in_pktinfo *pki = CMSG_DATA(cmh); |
439 | rqstp->rq_daddr.addr.s_addr = pki->ipi_spec_dst.s_addr; | |
95756482 | 440 | break; |
7a37f578 | 441 | } |
95756482 | 442 | case AF_INET6: { |
7a37f578 N |
443 | struct in6_pktinfo *pki = CMSG_DATA(cmh); |
444 | ipv6_addr_copy(&rqstp->rq_daddr.addr6, &pki->ipi6_addr); | |
95756482 | 445 | break; |
7a37f578 | 446 | } |
95756482 | 447 | } |
95756482 CL |
448 | } |
449 | ||
1da177e4 LT |
450 | /* |
451 | * Receive a datagram from a UDP socket. | |
452 | */ | |
0f0257ea | 453 | static int svc_udp_recvfrom(struct svc_rqst *rqstp) |
1da177e4 | 454 | { |
57b1d3ba TT |
455 | struct svc_sock *svsk = |
456 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
bb5cf160 | 457 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
1da177e4 | 458 | struct sk_buff *skb; |
bc375ea7 DM |
459 | union { |
460 | struct cmsghdr hdr; | |
461 | long all[SVC_PKTINFO_SPACE / sizeof(long)]; | |
462 | } buffer; | |
463 | struct cmsghdr *cmh = &buffer.hdr; | |
7a37f578 N |
464 | struct msghdr msg = { |
465 | .msg_name = svc_addr(rqstp), | |
466 | .msg_control = cmh, | |
467 | .msg_controllen = sizeof(buffer), | |
468 | .msg_flags = MSG_DONTWAIT, | |
469 | }; | |
abc5c44d CL |
470 | size_t len; |
471 | int err; | |
1da177e4 | 472 | |
02fc6c36 | 473 | if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags)) |
1da177e4 LT |
474 | /* udp sockets need large rcvbuf as all pending |
475 | * requests are still in that buffer. sndbuf must | |
476 | * also be large enough that there is enough space | |
3262c816 GB |
477 | * for one reply per thread. We count all threads |
478 | * rather than threads in a particular pool, which | |
479 | * provides an upper bound on the number of threads | |
480 | * which will access the socket. | |
1da177e4 LT |
481 | */ |
482 | svc_sock_setbufsize(svsk->sk_sock, | |
c6b0a9f8 N |
483 | (serv->sv_nrthreads+3) * serv->sv_max_mesg, |
484 | (serv->sv_nrthreads+3) * serv->sv_max_mesg); | |
1da177e4 | 485 | |
02fc6c36 | 486 | clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
05ed690e N |
487 | skb = NULL; |
488 | err = kernel_recvmsg(svsk->sk_sock, &msg, NULL, | |
489 | 0, 0, MSG_PEEK | MSG_DONTWAIT); | |
490 | if (err >= 0) | |
491 | skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err); | |
492 | ||
493 | if (skb == NULL) { | |
494 | if (err != -EAGAIN) { | |
495 | /* possibly an icmp error */ | |
496 | dprintk("svc: recvfrom returned error %d\n", -err); | |
02fc6c36 | 497 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 498 | } |
a6046f71 | 499 | svc_xprt_received(&svsk->sk_xprt); |
05ed690e | 500 | return -EAGAIN; |
1da177e4 | 501 | } |
9dbc240f | 502 | len = svc_addr_len(svc_addr(rqstp)); |
abc5c44d CL |
503 | if (len == 0) |
504 | return -EAFNOSUPPORT; | |
9dbc240f | 505 | rqstp->rq_addrlen = len; |
b7aa0bf7 ED |
506 | if (skb->tstamp.tv64 == 0) { |
507 | skb->tstamp = ktime_get_real(); | |
cca5172a | 508 | /* Don't enable netstamp, sunrpc doesn't |
1da177e4 LT |
509 | need that much accuracy */ |
510 | } | |
b7aa0bf7 | 511 | svsk->sk_sk->sk_stamp = skb->tstamp; |
02fc6c36 | 512 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */ |
1da177e4 LT |
513 | |
514 | /* | |
515 | * Maybe more packets - kick another thread ASAP. | |
516 | */ | |
a6046f71 | 517 | svc_xprt_received(&svsk->sk_xprt); |
1da177e4 LT |
518 | |
519 | len = skb->len - sizeof(struct udphdr); | |
520 | rqstp->rq_arg.len = len; | |
521 | ||
95756482 | 522 | rqstp->rq_prot = IPPROTO_UDP; |
27459f09 | 523 | |
7a37f578 N |
524 | if (cmh->cmsg_level != IPPROTO_IP || |
525 | cmh->cmsg_type != IP_PKTINFO) { | |
526 | if (net_ratelimit()) | |
527 | printk("rpcsvc: received unknown control message:" | |
528 | "%d/%d\n", | |
529 | cmh->cmsg_level, cmh->cmsg_type); | |
530 | skb_free_datagram(svsk->sk_sk, skb); | |
531 | return 0; | |
532 | } | |
533 | svc_udp_get_dest_address(rqstp, cmh); | |
1da177e4 LT |
534 | |
535 | if (skb_is_nonlinear(skb)) { | |
536 | /* we have to copy */ | |
537 | local_bh_disable(); | |
538 | if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) { | |
539 | local_bh_enable(); | |
540 | /* checksum error */ | |
541 | skb_free_datagram(svsk->sk_sk, skb); | |
542 | return 0; | |
543 | } | |
544 | local_bh_enable(); | |
cca5172a | 545 | skb_free_datagram(svsk->sk_sk, skb); |
1da177e4 LT |
546 | } else { |
547 | /* we can use it in-place */ | |
0f0257ea TT |
548 | rqstp->rq_arg.head[0].iov_base = skb->data + |
549 | sizeof(struct udphdr); | |
1da177e4 | 550 | rqstp->rq_arg.head[0].iov_len = len; |
fb286bb2 HX |
551 | if (skb_checksum_complete(skb)) { |
552 | skb_free_datagram(svsk->sk_sk, skb); | |
553 | return 0; | |
1da177e4 | 554 | } |
5148bf4e | 555 | rqstp->rq_xprt_ctxt = skb; |
1da177e4 LT |
556 | } |
557 | ||
558 | rqstp->rq_arg.page_base = 0; | |
559 | if (len <= rqstp->rq_arg.head[0].iov_len) { | |
560 | rqstp->rq_arg.head[0].iov_len = len; | |
561 | rqstp->rq_arg.page_len = 0; | |
44524359 | 562 | rqstp->rq_respages = rqstp->rq_pages+1; |
1da177e4 LT |
563 | } else { |
564 | rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len; | |
44524359 | 565 | rqstp->rq_respages = rqstp->rq_pages + 1 + |
172589cc | 566 | DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE); |
1da177e4 LT |
567 | } |
568 | ||
569 | if (serv->sv_stats) | |
570 | serv->sv_stats->netudpcnt++; | |
571 | ||
572 | return len; | |
573 | } | |
574 | ||
575 | static int | |
576 | svc_udp_sendto(struct svc_rqst *rqstp) | |
577 | { | |
578 | int error; | |
579 | ||
580 | error = svc_sendto(rqstp, &rqstp->rq_res); | |
581 | if (error == -ECONNREFUSED) | |
582 | /* ICMP error on earlier request. */ | |
583 | error = svc_sendto(rqstp, &rqstp->rq_res); | |
584 | ||
585 | return error; | |
586 | } | |
587 | ||
e831fe65 TT |
588 | static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp) |
589 | { | |
590 | } | |
591 | ||
323bee32 TT |
592 | static int svc_udp_has_wspace(struct svc_xprt *xprt) |
593 | { | |
594 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
bb5cf160 | 595 | struct svc_serv *serv = xprt->xpt_server; |
323bee32 TT |
596 | unsigned long required; |
597 | ||
598 | /* | |
599 | * Set the SOCK_NOSPACE flag before checking the available | |
600 | * sock space. | |
601 | */ | |
602 | set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
7a90e8cc | 603 | required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg; |
323bee32 TT |
604 | if (required*2 > sock_wspace(svsk->sk_sk)) |
605 | return 0; | |
606 | clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
607 | return 1; | |
608 | } | |
609 | ||
38a417cc TT |
610 | static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt) |
611 | { | |
612 | BUG(); | |
613 | return NULL; | |
614 | } | |
615 | ||
b700cbb1 TT |
616 | static struct svc_xprt *svc_udp_create(struct svc_serv *serv, |
617 | struct sockaddr *sa, int salen, | |
618 | int flags) | |
619 | { | |
620 | return svc_create_socket(serv, IPPROTO_UDP, sa, salen, flags); | |
621 | } | |
622 | ||
360d8738 | 623 | static struct svc_xprt_ops svc_udp_ops = { |
b700cbb1 | 624 | .xpo_create = svc_udp_create, |
5d137990 TT |
625 | .xpo_recvfrom = svc_udp_recvfrom, |
626 | .xpo_sendto = svc_udp_sendto, | |
5148bf4e | 627 | .xpo_release_rqst = svc_release_skb, |
755cceab TT |
628 | .xpo_detach = svc_sock_detach, |
629 | .xpo_free = svc_sock_free, | |
e831fe65 | 630 | .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr, |
323bee32 | 631 | .xpo_has_wspace = svc_udp_has_wspace, |
38a417cc | 632 | .xpo_accept = svc_udp_accept, |
360d8738 TT |
633 | }; |
634 | ||
635 | static struct svc_xprt_class svc_udp_class = { | |
636 | .xcl_name = "udp", | |
b700cbb1 | 637 | .xcl_owner = THIS_MODULE, |
360d8738 | 638 | .xcl_ops = &svc_udp_ops, |
49023155 | 639 | .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP, |
360d8738 TT |
640 | }; |
641 | ||
bb5cf160 | 642 | static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv) |
1da177e4 | 643 | { |
7a37f578 N |
644 | int one = 1; |
645 | mm_segment_t oldfs; | |
646 | ||
bb5cf160 | 647 | svc_xprt_init(&svc_udp_class, &svsk->sk_xprt, serv); |
def13d74 | 648 | clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
649 | svsk->sk_sk->sk_data_ready = svc_udp_data_ready; |
650 | svsk->sk_sk->sk_write_space = svc_write_space; | |
1da177e4 LT |
651 | |
652 | /* initialise setting must have enough space to | |
cca5172a | 653 | * receive and respond to one request. |
1da177e4 LT |
654 | * svc_udp_recvfrom will re-adjust if necessary |
655 | */ | |
656 | svc_sock_setbufsize(svsk->sk_sock, | |
bb5cf160 TT |
657 | 3 * svsk->sk_xprt.xpt_server->sv_max_mesg, |
658 | 3 * svsk->sk_xprt.xpt_server->sv_max_mesg); | |
1da177e4 | 659 | |
0f0257ea TT |
660 | /* data might have come in before data_ready set up */ |
661 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); | |
02fc6c36 | 662 | set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); |
7a37f578 N |
663 | |
664 | oldfs = get_fs(); | |
665 | set_fs(KERNEL_DS); | |
666 | /* make sure we get destination address info */ | |
667 | svsk->sk_sock->ops->setsockopt(svsk->sk_sock, IPPROTO_IP, IP_PKTINFO, | |
668 | (char __user *)&one, sizeof(one)); | |
669 | set_fs(oldfs); | |
1da177e4 LT |
670 | } |
671 | ||
672 | /* | |
673 | * A data_ready event on a listening socket means there's a connection | |
674 | * pending. Do not use state_change as a substitute for it. | |
675 | */ | |
0f0257ea | 676 | static void svc_tcp_listen_data_ready(struct sock *sk, int count_unused) |
1da177e4 | 677 | { |
939bb7ef | 678 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 LT |
679 | |
680 | dprintk("svc: socket %p TCP (listen) state change %d\n", | |
939bb7ef | 681 | sk, sk->sk_state); |
1da177e4 | 682 | |
939bb7ef NB |
683 | /* |
684 | * This callback may called twice when a new connection | |
685 | * is established as a child socket inherits everything | |
686 | * from a parent LISTEN socket. | |
687 | * 1) data_ready method of the parent socket will be called | |
688 | * when one of child sockets become ESTABLISHED. | |
689 | * 2) data_ready method of the child socket may be called | |
690 | * when it receives data before the socket is accepted. | |
691 | * In case of 2, we should ignore it silently. | |
692 | */ | |
693 | if (sk->sk_state == TCP_LISTEN) { | |
694 | if (svsk) { | |
02fc6c36 | 695 | set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
f6150c3c | 696 | svc_xprt_enqueue(&svsk->sk_xprt); |
939bb7ef NB |
697 | } else |
698 | printk("svc: socket %p: no user data\n", sk); | |
1da177e4 | 699 | } |
939bb7ef | 700 | |
1da177e4 LT |
701 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
702 | wake_up_interruptible_all(sk->sk_sleep); | |
703 | } | |
704 | ||
705 | /* | |
706 | * A state change on a connected socket means it's dying or dead. | |
707 | */ | |
0f0257ea | 708 | static void svc_tcp_state_change(struct sock *sk) |
1da177e4 | 709 | { |
939bb7ef | 710 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 LT |
711 | |
712 | dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n", | |
939bb7ef | 713 | sk, sk->sk_state, sk->sk_user_data); |
1da177e4 | 714 | |
939bb7ef | 715 | if (!svsk) |
1da177e4 | 716 | printk("svc: socket %p: no user data\n", sk); |
939bb7ef | 717 | else { |
02fc6c36 | 718 | set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); |
f6150c3c | 719 | svc_xprt_enqueue(&svsk->sk_xprt); |
1da177e4 | 720 | } |
1da177e4 LT |
721 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
722 | wake_up_interruptible_all(sk->sk_sleep); | |
723 | } | |
724 | ||
0f0257ea | 725 | static void svc_tcp_data_ready(struct sock *sk, int count) |
1da177e4 | 726 | { |
939bb7ef | 727 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 LT |
728 | |
729 | dprintk("svc: socket %p TCP data ready (svsk %p)\n", | |
939bb7ef NB |
730 | sk, sk->sk_user_data); |
731 | if (svsk) { | |
02fc6c36 | 732 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
f6150c3c | 733 | svc_xprt_enqueue(&svsk->sk_xprt); |
939bb7ef | 734 | } |
1da177e4 LT |
735 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
736 | wake_up_interruptible(sk->sk_sleep); | |
737 | } | |
738 | ||
739 | /* | |
740 | * Accept a TCP connection | |
741 | */ | |
38a417cc | 742 | static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt) |
1da177e4 | 743 | { |
38a417cc | 744 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); |
cdd88b9f AM |
745 | struct sockaddr_storage addr; |
746 | struct sockaddr *sin = (struct sockaddr *) &addr; | |
bb5cf160 | 747 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
1da177e4 LT |
748 | struct socket *sock = svsk->sk_sock; |
749 | struct socket *newsock; | |
1da177e4 LT |
750 | struct svc_sock *newsvsk; |
751 | int err, slen; | |
5216a8e7 | 752 | RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); |
1da177e4 LT |
753 | |
754 | dprintk("svc: tcp_accept %p sock %p\n", svsk, sock); | |
755 | if (!sock) | |
38a417cc | 756 | return NULL; |
1da177e4 | 757 | |
02fc6c36 | 758 | clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
e6242e92 SS |
759 | err = kernel_accept(sock, &newsock, O_NONBLOCK); |
760 | if (err < 0) { | |
1da177e4 LT |
761 | if (err == -ENOMEM) |
762 | printk(KERN_WARNING "%s: no more sockets!\n", | |
763 | serv->sv_name); | |
e6242e92 | 764 | else if (err != -EAGAIN && net_ratelimit()) |
1da177e4 LT |
765 | printk(KERN_WARNING "%s: accept failed (err %d)!\n", |
766 | serv->sv_name, -err); | |
38a417cc | 767 | return NULL; |
1da177e4 | 768 | } |
02fc6c36 | 769 | set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 770 | |
cdd88b9f | 771 | err = kernel_getpeername(newsock, sin, &slen); |
1da177e4 LT |
772 | if (err < 0) { |
773 | if (net_ratelimit()) | |
774 | printk(KERN_WARNING "%s: peername failed (err %d)!\n", | |
775 | serv->sv_name, -err); | |
776 | goto failed; /* aborted connection or whatever */ | |
777 | } | |
778 | ||
779 | /* Ideally, we would want to reject connections from unauthorized | |
ad06e4bd CL |
780 | * hosts here, but when we get encryption, the IP of the host won't |
781 | * tell us anything. For now just warn about unpriv connections. | |
1da177e4 | 782 | */ |
cdd88b9f | 783 | if (!svc_port_is_privileged(sin)) { |
1da177e4 | 784 | dprintk(KERN_WARNING |
ad06e4bd | 785 | "%s: connect from unprivileged port: %s\n", |
cca5172a | 786 | serv->sv_name, |
cdd88b9f | 787 | __svc_print_addr(sin, buf, sizeof(buf))); |
1da177e4 | 788 | } |
ad06e4bd | 789 | dprintk("%s: connect from %s\n", serv->sv_name, |
cdd88b9f | 790 | __svc_print_addr(sin, buf, sizeof(buf))); |
1da177e4 LT |
791 | |
792 | /* make sure that a write doesn't block forever when | |
793 | * low on memory | |
794 | */ | |
795 | newsock->sk->sk_sndtimeo = HZ*30; | |
796 | ||
6b174337 CL |
797 | if (!(newsvsk = svc_setup_socket(serv, newsock, &err, |
798 | (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY)))) | |
1da177e4 | 799 | goto failed; |
9dbc240f | 800 | svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen); |
a9747692 FM |
801 | err = kernel_getsockname(newsock, sin, &slen); |
802 | if (unlikely(err < 0)) { | |
803 | dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err); | |
804 | slen = offsetof(struct sockaddr, sa_data); | |
805 | } | |
9dbc240f | 806 | svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen); |
067d7817 | 807 | |
f9f3cc4f TT |
808 | if (serv->sv_stats) |
809 | serv->sv_stats->nettcpconn++; | |
810 | ||
811 | return &newsvsk->sk_xprt; | |
812 | ||
813 | failed: | |
814 | sock_release(newsock); | |
815 | return NULL; | |
816 | } | |
817 | ||
1da177e4 LT |
818 | /* |
819 | * Receive data from a TCP socket. | |
820 | */ | |
0f0257ea | 821 | static int svc_tcp_recvfrom(struct svc_rqst *rqstp) |
1da177e4 | 822 | { |
57b1d3ba TT |
823 | struct svc_sock *svsk = |
824 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
bb5cf160 | 825 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
1da177e4 | 826 | int len; |
3cc03b16 | 827 | struct kvec *vec; |
1da177e4 LT |
828 | int pnum, vlen; |
829 | ||
830 | dprintk("svc: tcp_recv %p data %d conn %d close %d\n", | |
02fc6c36 TT |
831 | svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags), |
832 | test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags), | |
833 | test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags)); | |
1da177e4 | 834 | |
7f421835 BF |
835 | if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags)) |
836 | /* sndbuf needs to have room for one request | |
837 | * per thread, otherwise we can stall even when the | |
838 | * network isn't a bottleneck. | |
839 | * | |
840 | * We count all threads rather than threads in a | |
841 | * particular pool, which provides an upper bound | |
842 | * on the number of threads which will access the socket. | |
843 | * | |
844 | * rcvbuf just needs to be able to hold a few requests. | |
845 | * Normally they will be removed from the queue | |
846 | * as soon a a complete request arrives. | |
847 | */ | |
848 | svc_sock_setbufsize(svsk->sk_sock, | |
849 | (serv->sv_nrthreads+3) * serv->sv_max_mesg, | |
850 | 3 * serv->sv_max_mesg); | |
851 | ||
02fc6c36 | 852 | clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
853 | |
854 | /* Receive data. If we haven't got the record length yet, get | |
855 | * the next four bytes. Otherwise try to gobble up as much as | |
856 | * possible up to the complete record length. | |
857 | */ | |
c0401ea0 CL |
858 | if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) { |
859 | int want = sizeof(rpc_fraghdr) - svsk->sk_tcplen; | |
1da177e4 LT |
860 | struct kvec iov; |
861 | ||
862 | iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen; | |
863 | iov.iov_len = want; | |
864 | if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0) | |
865 | goto error; | |
866 | svsk->sk_tcplen += len; | |
867 | ||
868 | if (len < want) { | |
c0401ea0 CL |
869 | dprintk("svc: short recvfrom while reading record " |
870 | "length (%d of %d)\n", len, want); | |
a6046f71 | 871 | svc_xprt_received(&svsk->sk_xprt); |
1da177e4 LT |
872 | return -EAGAIN; /* record header not complete */ |
873 | } | |
874 | ||
875 | svsk->sk_reclen = ntohl(svsk->sk_reclen); | |
c0401ea0 | 876 | if (!(svsk->sk_reclen & RPC_LAST_STREAM_FRAGMENT)) { |
1da177e4 LT |
877 | /* FIXME: technically, a record can be fragmented, |
878 | * and non-terminal fragments will not have the top | |
879 | * bit set in the fragment length header. | |
880 | * But apparently no known nfs clients send fragmented | |
881 | * records. */ | |
34e9a63b | 882 | if (net_ratelimit()) |
c0401ea0 CL |
883 | printk(KERN_NOTICE "RPC: multiple fragments " |
884 | "per record not supported\n"); | |
1da177e4 LT |
885 | goto err_delete; |
886 | } | |
c0401ea0 | 887 | svsk->sk_reclen &= RPC_FRAGMENT_SIZE_MASK; |
1da177e4 | 888 | dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen); |
c6b0a9f8 | 889 | if (svsk->sk_reclen > serv->sv_max_mesg) { |
34e9a63b | 890 | if (net_ratelimit()) |
c0401ea0 CL |
891 | printk(KERN_NOTICE "RPC: " |
892 | "fragment too large: 0x%08lx\n", | |
893 | (unsigned long)svsk->sk_reclen); | |
1da177e4 LT |
894 | goto err_delete; |
895 | } | |
896 | } | |
897 | ||
898 | /* Check whether enough data is available */ | |
899 | len = svc_recv_available(svsk); | |
900 | if (len < 0) | |
901 | goto error; | |
902 | ||
903 | if (len < svsk->sk_reclen) { | |
904 | dprintk("svc: incomplete TCP record (%d of %d)\n", | |
905 | len, svsk->sk_reclen); | |
a6046f71 | 906 | svc_xprt_received(&svsk->sk_xprt); |
1da177e4 LT |
907 | return -EAGAIN; /* record not complete */ |
908 | } | |
909 | len = svsk->sk_reclen; | |
02fc6c36 | 910 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 911 | |
3cc03b16 | 912 | vec = rqstp->rq_vec; |
1da177e4 LT |
913 | vec[0] = rqstp->rq_arg.head[0]; |
914 | vlen = PAGE_SIZE; | |
915 | pnum = 1; | |
916 | while (vlen < len) { | |
44524359 | 917 | vec[pnum].iov_base = page_address(rqstp->rq_pages[pnum]); |
1da177e4 LT |
918 | vec[pnum].iov_len = PAGE_SIZE; |
919 | pnum++; | |
920 | vlen += PAGE_SIZE; | |
921 | } | |
44524359 | 922 | rqstp->rq_respages = &rqstp->rq_pages[pnum]; |
1da177e4 LT |
923 | |
924 | /* Now receive data */ | |
925 | len = svc_recvfrom(rqstp, vec, pnum, len); | |
926 | if (len < 0) | |
927 | goto error; | |
928 | ||
929 | dprintk("svc: TCP complete record (%d bytes)\n", len); | |
930 | rqstp->rq_arg.len = len; | |
931 | rqstp->rq_arg.page_base = 0; | |
932 | if (len <= rqstp->rq_arg.head[0].iov_len) { | |
933 | rqstp->rq_arg.head[0].iov_len = len; | |
934 | rqstp->rq_arg.page_len = 0; | |
935 | } else { | |
936 | rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len; | |
937 | } | |
938 | ||
5148bf4e | 939 | rqstp->rq_xprt_ctxt = NULL; |
1da177e4 LT |
940 | rqstp->rq_prot = IPPROTO_TCP; |
941 | ||
942 | /* Reset TCP read info */ | |
943 | svsk->sk_reclen = 0; | |
944 | svsk->sk_tcplen = 0; | |
945 | ||
9dbc240f | 946 | svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt); |
a6046f71 | 947 | svc_xprt_received(&svsk->sk_xprt); |
1da177e4 LT |
948 | if (serv->sv_stats) |
949 | serv->sv_stats->nettcpcnt++; | |
950 | ||
951 | return len; | |
952 | ||
953 | err_delete: | |
02fc6c36 | 954 | set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
955 | return -EAGAIN; |
956 | ||
957 | error: | |
958 | if (len == -EAGAIN) { | |
959 | dprintk("RPC: TCP recvfrom got EAGAIN\n"); | |
a6046f71 | 960 | svc_xprt_received(&svsk->sk_xprt); |
1da177e4 LT |
961 | } else { |
962 | printk(KERN_NOTICE "%s: recvfrom returned errno %d\n", | |
bb5cf160 | 963 | svsk->sk_xprt.xpt_server->sv_name, -len); |
93fbf1a5 | 964 | goto err_delete; |
1da177e4 LT |
965 | } |
966 | ||
967 | return len; | |
968 | } | |
969 | ||
970 | /* | |
971 | * Send out data on TCP socket. | |
972 | */ | |
0f0257ea | 973 | static int svc_tcp_sendto(struct svc_rqst *rqstp) |
1da177e4 LT |
974 | { |
975 | struct xdr_buf *xbufp = &rqstp->rq_res; | |
976 | int sent; | |
d8ed029d | 977 | __be32 reclen; |
1da177e4 LT |
978 | |
979 | /* Set up the first element of the reply kvec. | |
980 | * Any other kvecs that may be in use have been taken | |
981 | * care of by the server implementation itself. | |
982 | */ | |
983 | reclen = htonl(0x80000000|((xbufp->len ) - 4)); | |
984 | memcpy(xbufp->head[0].iov_base, &reclen, 4); | |
985 | ||
57b1d3ba | 986 | if (test_bit(XPT_DEAD, &rqstp->rq_xprt->xpt_flags)) |
1da177e4 LT |
987 | return -ENOTCONN; |
988 | ||
989 | sent = svc_sendto(rqstp, &rqstp->rq_res); | |
990 | if (sent != xbufp->len) { | |
0f0257ea TT |
991 | printk(KERN_NOTICE |
992 | "rpc-srv/tcp: %s: %s %d when sending %d bytes " | |
993 | "- shutting down socket\n", | |
57b1d3ba | 994 | rqstp->rq_xprt->xpt_server->sv_name, |
1da177e4 LT |
995 | (sent<0)?"got error":"sent only", |
996 | sent, xbufp->len); | |
57b1d3ba | 997 | set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags); |
f6150c3c | 998 | svc_xprt_enqueue(rqstp->rq_xprt); |
1da177e4 LT |
999 | sent = -EAGAIN; |
1000 | } | |
1001 | return sent; | |
1002 | } | |
1003 | ||
e831fe65 TT |
1004 | /* |
1005 | * Setup response header. TCP has a 4B record length field. | |
1006 | */ | |
1007 | static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp) | |
1008 | { | |
1009 | struct kvec *resv = &rqstp->rq_res.head[0]; | |
1010 | ||
1011 | /* tcp needs a space for the record length... */ | |
1012 | svc_putnl(resv, 0); | |
1013 | } | |
1014 | ||
323bee32 TT |
1015 | static int svc_tcp_has_wspace(struct svc_xprt *xprt) |
1016 | { | |
57b1d3ba | 1017 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); |
bb5cf160 | 1018 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
323bee32 TT |
1019 | int required; |
1020 | int wspace; | |
1021 | ||
1022 | /* | |
1023 | * Set the SOCK_NOSPACE flag before checking the available | |
1024 | * sock space. | |
1025 | */ | |
1026 | set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
7a90e8cc | 1027 | required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg; |
323bee32 TT |
1028 | wspace = sk_stream_wspace(svsk->sk_sk); |
1029 | ||
1030 | if (wspace < sk_stream_min_wspace(svsk->sk_sk)) | |
1031 | return 0; | |
1032 | if (required * 2 > wspace) | |
1033 | return 0; | |
1034 | ||
1035 | clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
1036 | return 1; | |
1037 | } | |
1038 | ||
b700cbb1 TT |
1039 | static struct svc_xprt *svc_tcp_create(struct svc_serv *serv, |
1040 | struct sockaddr *sa, int salen, | |
1041 | int flags) | |
1042 | { | |
1043 | return svc_create_socket(serv, IPPROTO_TCP, sa, salen, flags); | |
1044 | } | |
1045 | ||
360d8738 | 1046 | static struct svc_xprt_ops svc_tcp_ops = { |
b700cbb1 | 1047 | .xpo_create = svc_tcp_create, |
5d137990 TT |
1048 | .xpo_recvfrom = svc_tcp_recvfrom, |
1049 | .xpo_sendto = svc_tcp_sendto, | |
5148bf4e | 1050 | .xpo_release_rqst = svc_release_skb, |
69b6ba37 | 1051 | .xpo_detach = svc_tcp_sock_detach, |
755cceab | 1052 | .xpo_free = svc_sock_free, |
e831fe65 | 1053 | .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr, |
323bee32 | 1054 | .xpo_has_wspace = svc_tcp_has_wspace, |
38a417cc | 1055 | .xpo_accept = svc_tcp_accept, |
360d8738 TT |
1056 | }; |
1057 | ||
1058 | static struct svc_xprt_class svc_tcp_class = { | |
1059 | .xcl_name = "tcp", | |
b700cbb1 | 1060 | .xcl_owner = THIS_MODULE, |
360d8738 | 1061 | .xcl_ops = &svc_tcp_ops, |
49023155 | 1062 | .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP, |
360d8738 TT |
1063 | }; |
1064 | ||
1065 | void svc_init_xprt_sock(void) | |
1066 | { | |
1067 | svc_reg_xprt_class(&svc_tcp_class); | |
1068 | svc_reg_xprt_class(&svc_udp_class); | |
1069 | } | |
1070 | ||
1071 | void svc_cleanup_xprt_sock(void) | |
1072 | { | |
1073 | svc_unreg_xprt_class(&svc_tcp_class); | |
1074 | svc_unreg_xprt_class(&svc_udp_class); | |
1075 | } | |
1076 | ||
bb5cf160 | 1077 | static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv) |
1da177e4 LT |
1078 | { |
1079 | struct sock *sk = svsk->sk_sk; | |
1da177e4 | 1080 | |
bb5cf160 | 1081 | svc_xprt_init(&svc_tcp_class, &svsk->sk_xprt, serv); |
def13d74 | 1082 | set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
1083 | if (sk->sk_state == TCP_LISTEN) { |
1084 | dprintk("setting up TCP socket for listening\n"); | |
02fc6c36 | 1085 | set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 1086 | sk->sk_data_ready = svc_tcp_listen_data_ready; |
02fc6c36 | 1087 | set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
1088 | } else { |
1089 | dprintk("setting up TCP socket for reading\n"); | |
1090 | sk->sk_state_change = svc_tcp_state_change; | |
1091 | sk->sk_data_ready = svc_tcp_data_ready; | |
1092 | sk->sk_write_space = svc_write_space; | |
1093 | ||
1094 | svsk->sk_reclen = 0; | |
1095 | svsk->sk_tcplen = 0; | |
1096 | ||
b7872fe8 | 1097 | tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF; |
1da177e4 | 1098 | |
7f421835 BF |
1099 | /* initialise setting must have enough space to |
1100 | * receive and respond to one request. | |
1101 | * svc_tcp_recvfrom will re-adjust if necessary | |
1102 | */ | |
1103 | svc_sock_setbufsize(svsk->sk_sock, | |
1104 | 3 * svsk->sk_xprt.xpt_server->sv_max_mesg, | |
1105 | 3 * svsk->sk_xprt.xpt_server->sv_max_mesg); | |
1106 | ||
1107 | set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); | |
02fc6c36 | 1108 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
cca5172a | 1109 | if (sk->sk_state != TCP_ESTABLISHED) |
02fc6c36 | 1110 | set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
1111 | } |
1112 | } | |
1113 | ||
0f0257ea | 1114 | void svc_sock_update_bufs(struct svc_serv *serv) |
1da177e4 LT |
1115 | { |
1116 | /* | |
1117 | * The number of server threads has changed. Update | |
1118 | * rcvbuf and sndbuf accordingly on all sockets | |
1119 | */ | |
1120 | struct list_head *le; | |
1121 | ||
1122 | spin_lock_bh(&serv->sv_lock); | |
1123 | list_for_each(le, &serv->sv_permsocks) { | |
cca5172a | 1124 | struct svc_sock *svsk = |
7a182083 | 1125 | list_entry(le, struct svc_sock, sk_xprt.xpt_list); |
02fc6c36 | 1126 | set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
1127 | } |
1128 | list_for_each(le, &serv->sv_tempsocks) { | |
1129 | struct svc_sock *svsk = | |
7a182083 | 1130 | list_entry(le, struct svc_sock, sk_xprt.xpt_list); |
02fc6c36 | 1131 | set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
1132 | } |
1133 | spin_unlock_bh(&serv->sv_lock); | |
1134 | } | |
24c3767e | 1135 | EXPORT_SYMBOL_GPL(svc_sock_update_bufs); |
1da177e4 | 1136 | |
1da177e4 LT |
1137 | /* |
1138 | * Initialize socket for RPC use and create svc_sock struct | |
1139 | * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF. | |
1140 | */ | |
6b174337 CL |
1141 | static struct svc_sock *svc_setup_socket(struct svc_serv *serv, |
1142 | struct socket *sock, | |
1143 | int *errp, int flags) | |
1da177e4 LT |
1144 | { |
1145 | struct svc_sock *svsk; | |
1146 | struct sock *inet; | |
6b174337 | 1147 | int pmap_register = !(flags & SVC_SOCK_ANONYMOUS); |
1da177e4 LT |
1148 | |
1149 | dprintk("svc: svc_setup_socket %p\n", sock); | |
0da974f4 | 1150 | if (!(svsk = kzalloc(sizeof(*svsk), GFP_KERNEL))) { |
1da177e4 LT |
1151 | *errp = -ENOMEM; |
1152 | return NULL; | |
1153 | } | |
1da177e4 LT |
1154 | |
1155 | inet = sock->sk; | |
1156 | ||
1157 | /* Register socket with portmapper */ | |
1158 | if (*errp >= 0 && pmap_register) | |
baf01caf | 1159 | *errp = svc_register(serv, inet->sk_family, inet->sk_protocol, |
1da177e4 LT |
1160 | ntohs(inet_sk(inet)->sport)); |
1161 | ||
1162 | if (*errp < 0) { | |
1163 | kfree(svsk); | |
1164 | return NULL; | |
1165 | } | |
1166 | ||
1da177e4 LT |
1167 | inet->sk_user_data = svsk; |
1168 | svsk->sk_sock = sock; | |
1169 | svsk->sk_sk = inet; | |
1170 | svsk->sk_ostate = inet->sk_state_change; | |
1171 | svsk->sk_odata = inet->sk_data_ready; | |
1172 | svsk->sk_owspace = inet->sk_write_space; | |
1da177e4 LT |
1173 | |
1174 | /* Initialize the socket */ | |
1175 | if (sock->type == SOCK_DGRAM) | |
bb5cf160 | 1176 | svc_udp_init(svsk, serv); |
7f421835 | 1177 | else |
bb5cf160 | 1178 | svc_tcp_init(svsk, serv); |
1da177e4 | 1179 | |
1da177e4 LT |
1180 | dprintk("svc: svc_setup_socket created %p (inet %p)\n", |
1181 | svsk, svsk->sk_sk); | |
1182 | ||
1da177e4 LT |
1183 | return svsk; |
1184 | } | |
1185 | ||
bfba9ab4 CL |
1186 | /** |
1187 | * svc_addsock - add a listener socket to an RPC service | |
1188 | * @serv: pointer to RPC service to which to add a new listener | |
1189 | * @fd: file descriptor of the new listener | |
1190 | * @name_return: pointer to buffer to fill in with name of listener | |
1191 | * @len: size of the buffer | |
1192 | * | |
1193 | * Fills in socket name and returns positive length of name if successful. | |
1194 | * Name is terminated with '\n'. On error, returns a negative errno | |
1195 | * value. | |
1196 | */ | |
1197 | int svc_addsock(struct svc_serv *serv, const int fd, char *name_return, | |
1198 | const size_t len) | |
b41b66d6 N |
1199 | { |
1200 | int err = 0; | |
1201 | struct socket *so = sockfd_lookup(fd, &err); | |
1202 | struct svc_sock *svsk = NULL; | |
1203 | ||
1204 | if (!so) | |
1205 | return err; | |
1206 | if (so->sk->sk_family != AF_INET) | |
1207 | err = -EAFNOSUPPORT; | |
1208 | else if (so->sk->sk_protocol != IPPROTO_TCP && | |
1209 | so->sk->sk_protocol != IPPROTO_UDP) | |
1210 | err = -EPROTONOSUPPORT; | |
1211 | else if (so->state > SS_UNCONNECTED) | |
1212 | err = -EISCONN; | |
1213 | else { | |
2da2c21d TT |
1214 | if (!try_module_get(THIS_MODULE)) |
1215 | err = -ENOENT; | |
1216 | else | |
1217 | svsk = svc_setup_socket(serv, so, &err, | |
1218 | SVC_SOCK_DEFAULTS); | |
e79eff1f | 1219 | if (svsk) { |
9dbc240f TT |
1220 | struct sockaddr_storage addr; |
1221 | struct sockaddr *sin = (struct sockaddr *)&addr; | |
1222 | int salen; | |
1223 | if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0) | |
1224 | svc_xprt_set_local(&svsk->sk_xprt, sin, salen); | |
4e5caaa5 TT |
1225 | clear_bit(XPT_TEMP, &svsk->sk_xprt.xpt_flags); |
1226 | spin_lock_bh(&serv->sv_lock); | |
1227 | list_add(&svsk->sk_xprt.xpt_list, &serv->sv_permsocks); | |
1228 | spin_unlock_bh(&serv->sv_lock); | |
a6046f71 | 1229 | svc_xprt_received(&svsk->sk_xprt); |
b41b66d6 | 1230 | err = 0; |
2da2c21d TT |
1231 | } else |
1232 | module_put(THIS_MODULE); | |
b41b66d6 N |
1233 | } |
1234 | if (err) { | |
1235 | sockfd_put(so); | |
1236 | return err; | |
1237 | } | |
e7942b9f | 1238 | return svc_one_sock_name(svsk, name_return, len); |
b41b66d6 N |
1239 | } |
1240 | EXPORT_SYMBOL_GPL(svc_addsock); | |
1241 | ||
1da177e4 LT |
1242 | /* |
1243 | * Create socket for RPC service. | |
1244 | */ | |
b700cbb1 TT |
1245 | static struct svc_xprt *svc_create_socket(struct svc_serv *serv, |
1246 | int protocol, | |
1247 | struct sockaddr *sin, int len, | |
1248 | int flags) | |
1da177e4 LT |
1249 | { |
1250 | struct svc_sock *svsk; | |
1251 | struct socket *sock; | |
1252 | int error; | |
1253 | int type; | |
9dbc240f TT |
1254 | struct sockaddr_storage addr; |
1255 | struct sockaddr *newsin = (struct sockaddr *)&addr; | |
1256 | int newlen; | |
c69da774 TM |
1257 | int family; |
1258 | int val; | |
5216a8e7 | 1259 | RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); |
1da177e4 | 1260 | |
ad06e4bd CL |
1261 | dprintk("svc: svc_create_socket(%s, %d, %s)\n", |
1262 | serv->sv_program->pg_name, protocol, | |
77f1f67a | 1263 | __svc_print_addr(sin, buf, sizeof(buf))); |
1da177e4 LT |
1264 | |
1265 | if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) { | |
1266 | printk(KERN_WARNING "svc: only UDP and TCP " | |
1267 | "sockets supported\n"); | |
b700cbb1 | 1268 | return ERR_PTR(-EINVAL); |
1da177e4 | 1269 | } |
c69da774 | 1270 | |
1da177e4 | 1271 | type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM; |
c69da774 TM |
1272 | switch (sin->sa_family) { |
1273 | case AF_INET6: | |
1274 | family = PF_INET6; | |
1275 | break; | |
1276 | case AF_INET: | |
1277 | family = PF_INET; | |
1278 | break; | |
1279 | default: | |
1280 | return ERR_PTR(-EINVAL); | |
1281 | } | |
1da177e4 | 1282 | |
c69da774 | 1283 | error = sock_create_kern(family, type, protocol, &sock); |
77f1f67a | 1284 | if (error < 0) |
b700cbb1 | 1285 | return ERR_PTR(error); |
1da177e4 | 1286 | |
ed07536e PZ |
1287 | svc_reclassify_socket(sock); |
1288 | ||
c69da774 TM |
1289 | /* |
1290 | * If this is an PF_INET6 listener, we want to avoid | |
1291 | * getting requests from IPv4 remotes. Those should | |
1292 | * be shunted to a PF_INET listener via rpcbind. | |
1293 | */ | |
1294 | val = 1; | |
1295 | if (family == PF_INET6) | |
1296 | kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY, | |
1297 | (char *)&val, sizeof(val)); | |
1298 | ||
18114746 | 1299 | if (type == SOCK_STREAM) |
77f1f67a CL |
1300 | sock->sk->sk_reuse = 1; /* allow address reuse */ |
1301 | error = kernel_bind(sock, sin, len); | |
18114746 ES |
1302 | if (error < 0) |
1303 | goto bummer; | |
1da177e4 | 1304 | |
9dbc240f TT |
1305 | newlen = len; |
1306 | error = kernel_getsockname(sock, newsin, &newlen); | |
1307 | if (error < 0) | |
1308 | goto bummer; | |
1309 | ||
1da177e4 | 1310 | if (protocol == IPPROTO_TCP) { |
e6242e92 | 1311 | if ((error = kernel_listen(sock, 64)) < 0) |
1da177e4 LT |
1312 | goto bummer; |
1313 | } | |
1314 | ||
e79eff1f | 1315 | if ((svsk = svc_setup_socket(serv, sock, &error, flags)) != NULL) { |
9dbc240f | 1316 | svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen); |
b700cbb1 | 1317 | return (struct svc_xprt *)svsk; |
e79eff1f | 1318 | } |
1da177e4 LT |
1319 | |
1320 | bummer: | |
1321 | dprintk("svc: svc_create_socket error = %d\n", -error); | |
1322 | sock_release(sock); | |
b700cbb1 | 1323 | return ERR_PTR(error); |
1da177e4 LT |
1324 | } |
1325 | ||
755cceab TT |
1326 | /* |
1327 | * Detach the svc_sock from the socket so that no | |
1328 | * more callbacks occur. | |
1329 | */ | |
1330 | static void svc_sock_detach(struct svc_xprt *xprt) | |
1331 | { | |
1332 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
1333 | struct sock *sk = svsk->sk_sk; | |
1334 | ||
1335 | dprintk("svc: svc_sock_detach(%p)\n", svsk); | |
1336 | ||
1337 | /* put back the old socket callbacks */ | |
1338 | sk->sk_state_change = svsk->sk_ostate; | |
1339 | sk->sk_data_ready = svsk->sk_odata; | |
1340 | sk->sk_write_space = svsk->sk_owspace; | |
69b6ba37 TM |
1341 | |
1342 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) | |
1343 | wake_up_interruptible(sk->sk_sleep); | |
1344 | } | |
1345 | ||
1346 | /* | |
1347 | * Disconnect the socket, and reset the callbacks | |
1348 | */ | |
1349 | static void svc_tcp_sock_detach(struct svc_xprt *xprt) | |
1350 | { | |
1351 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
1352 | ||
1353 | dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk); | |
1354 | ||
1355 | svc_sock_detach(xprt); | |
1356 | ||
1357 | if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) | |
1358 | kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR); | |
755cceab TT |
1359 | } |
1360 | ||
1361 | /* | |
1362 | * Free the svc_sock's socket resources and the svc_sock itself. | |
1363 | */ | |
1364 | static void svc_sock_free(struct svc_xprt *xprt) | |
1365 | { | |
1366 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
1367 | dprintk("svc: svc_sock_free(%p)\n", svsk); | |
1368 | ||
755cceab TT |
1369 | if (svsk->sk_sock->file) |
1370 | sockfd_put(svsk->sk_sock); | |
1371 | else | |
1372 | sock_release(svsk->sk_sock); | |
1373 | kfree(svsk); | |
1374 | } |