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> |
c752f073 | 41 | #include <net/tcp_states.h> |
1da177e4 LT |
42 | #include <asm/uaccess.h> |
43 | #include <asm/ioctls.h> | |
44 | ||
45 | #include <linux/sunrpc/types.h> | |
ad06e4bd | 46 | #include <linux/sunrpc/clnt.h> |
1da177e4 LT |
47 | #include <linux/sunrpc/xdr.h> |
48 | #include <linux/sunrpc/svcsock.h> | |
49 | #include <linux/sunrpc/stats.h> | |
50 | ||
360d8738 | 51 | #define RPCDBG_FACILITY RPCDBG_SVCXPRT |
1da177e4 LT |
52 | |
53 | ||
54 | static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *, | |
6b174337 | 55 | int *errp, int flags); |
1da177e4 LT |
56 | static void svc_udp_data_ready(struct sock *, int); |
57 | static int svc_udp_recvfrom(struct svc_rqst *); | |
58 | static int svc_udp_sendto(struct svc_rqst *); | |
755cceab TT |
59 | static void svc_sock_detach(struct svc_xprt *); |
60 | static void svc_sock_free(struct svc_xprt *); | |
1da177e4 | 61 | |
b700cbb1 TT |
62 | static struct svc_xprt *svc_create_socket(struct svc_serv *, int, |
63 | struct sockaddr *, int, int); | |
ed07536e PZ |
64 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
65 | static struct lock_class_key svc_key[2]; | |
66 | static struct lock_class_key svc_slock_key[2]; | |
67 | ||
0f0257ea | 68 | static void svc_reclassify_socket(struct socket *sock) |
ed07536e PZ |
69 | { |
70 | struct sock *sk = sock->sk; | |
02b3d346 | 71 | BUG_ON(sock_owned_by_user(sk)); |
ed07536e PZ |
72 | switch (sk->sk_family) { |
73 | case AF_INET: | |
74 | sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD", | |
def13d74 TT |
75 | &svc_slock_key[0], |
76 | "sk_xprt.xpt_lock-AF_INET-NFSD", | |
77 | &svc_key[0]); | |
ed07536e PZ |
78 | break; |
79 | ||
80 | case AF_INET6: | |
81 | sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD", | |
def13d74 TT |
82 | &svc_slock_key[1], |
83 | "sk_xprt.xpt_lock-AF_INET6-NFSD", | |
84 | &svc_key[1]); | |
ed07536e PZ |
85 | break; |
86 | ||
87 | default: | |
88 | BUG(); | |
89 | } | |
90 | } | |
91 | #else | |
0f0257ea | 92 | static void svc_reclassify_socket(struct socket *sock) |
ed07536e PZ |
93 | { |
94 | } | |
95 | #endif | |
96 | ||
1da177e4 LT |
97 | /* |
98 | * Release an skbuff after use | |
99 | */ | |
5148bf4e | 100 | static void svc_release_skb(struct svc_rqst *rqstp) |
1da177e4 | 101 | { |
5148bf4e | 102 | struct sk_buff *skb = rqstp->rq_xprt_ctxt; |
1da177e4 LT |
103 | struct svc_deferred_req *dr = rqstp->rq_deferred; |
104 | ||
105 | if (skb) { | |
57b1d3ba TT |
106 | struct svc_sock *svsk = |
107 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
5148bf4e | 108 | rqstp->rq_xprt_ctxt = NULL; |
1da177e4 LT |
109 | |
110 | dprintk("svc: service %p, releasing skb %p\n", rqstp, skb); | |
57b1d3ba | 111 | skb_free_datagram(svsk->sk_sk, skb); |
1da177e4 LT |
112 | } |
113 | if (dr) { | |
114 | rqstp->rq_deferred = NULL; | |
115 | kfree(dr); | |
116 | } | |
117 | } | |
118 | ||
b92503b2 CL |
119 | union svc_pktinfo_u { |
120 | struct in_pktinfo pkti; | |
b92503b2 | 121 | struct in6_pktinfo pkti6; |
b92503b2 | 122 | }; |
bc375ea7 DM |
123 | #define SVC_PKTINFO_SPACE \ |
124 | CMSG_SPACE(sizeof(union svc_pktinfo_u)) | |
b92503b2 CL |
125 | |
126 | static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh) | |
127 | { | |
57b1d3ba TT |
128 | struct svc_sock *svsk = |
129 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
130 | switch (svsk->sk_sk->sk_family) { | |
b92503b2 CL |
131 | case AF_INET: { |
132 | struct in_pktinfo *pki = CMSG_DATA(cmh); | |
133 | ||
134 | cmh->cmsg_level = SOL_IP; | |
135 | cmh->cmsg_type = IP_PKTINFO; | |
136 | pki->ipi_ifindex = 0; | |
137 | pki->ipi_spec_dst.s_addr = rqstp->rq_daddr.addr.s_addr; | |
138 | cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); | |
139 | } | |
140 | break; | |
5a05ed73 | 141 | |
b92503b2 CL |
142 | case AF_INET6: { |
143 | struct in6_pktinfo *pki = CMSG_DATA(cmh); | |
144 | ||
145 | cmh->cmsg_level = SOL_IPV6; | |
146 | cmh->cmsg_type = IPV6_PKTINFO; | |
147 | pki->ipi6_ifindex = 0; | |
148 | ipv6_addr_copy(&pki->ipi6_addr, | |
149 | &rqstp->rq_daddr.addr6); | |
150 | cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); | |
151 | } | |
152 | break; | |
b92503b2 CL |
153 | } |
154 | return; | |
155 | } | |
156 | ||
1da177e4 LT |
157 | /* |
158 | * Generic sendto routine | |
159 | */ | |
0f0257ea | 160 | static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr) |
1da177e4 | 161 | { |
57b1d3ba TT |
162 | struct svc_sock *svsk = |
163 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
1da177e4 LT |
164 | struct socket *sock = svsk->sk_sock; |
165 | int slen; | |
bc375ea7 DM |
166 | union { |
167 | struct cmsghdr hdr; | |
168 | long all[SVC_PKTINFO_SPACE / sizeof(long)]; | |
169 | } buffer; | |
170 | struct cmsghdr *cmh = &buffer.hdr; | |
1da177e4 LT |
171 | int len = 0; |
172 | int result; | |
173 | int size; | |
174 | struct page **ppage = xdr->pages; | |
175 | size_t base = xdr->page_base; | |
176 | unsigned int pglen = xdr->page_len; | |
177 | unsigned int flags = MSG_MORE; | |
ad06e4bd | 178 | char buf[RPC_MAX_ADDRBUFLEN]; |
1da177e4 LT |
179 | |
180 | slen = xdr->len; | |
181 | ||
182 | if (rqstp->rq_prot == IPPROTO_UDP) { | |
b92503b2 CL |
183 | struct msghdr msg = { |
184 | .msg_name = &rqstp->rq_addr, | |
185 | .msg_namelen = rqstp->rq_addrlen, | |
186 | .msg_control = cmh, | |
187 | .msg_controllen = sizeof(buffer), | |
188 | .msg_flags = MSG_MORE, | |
189 | }; | |
190 | ||
191 | svc_set_cmsg_data(rqstp, cmh); | |
1da177e4 LT |
192 | |
193 | if (sock_sendmsg(sock, &msg, 0) < 0) | |
194 | goto out; | |
195 | } | |
196 | ||
197 | /* send head */ | |
198 | if (slen == xdr->head[0].iov_len) | |
199 | flags = 0; | |
44524359 N |
200 | len = kernel_sendpage(sock, rqstp->rq_respages[0], 0, |
201 | xdr->head[0].iov_len, flags); | |
1da177e4 LT |
202 | if (len != xdr->head[0].iov_len) |
203 | goto out; | |
204 | slen -= xdr->head[0].iov_len; | |
205 | if (slen == 0) | |
206 | goto out; | |
207 | ||
208 | /* send page data */ | |
209 | size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen; | |
210 | while (pglen > 0) { | |
211 | if (slen == size) | |
212 | flags = 0; | |
e6242e92 | 213 | result = kernel_sendpage(sock, *ppage, base, size, flags); |
1da177e4 LT |
214 | if (result > 0) |
215 | len += result; | |
216 | if (result != size) | |
217 | goto out; | |
218 | slen -= size; | |
219 | pglen -= size; | |
220 | size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen; | |
221 | base = 0; | |
222 | ppage++; | |
223 | } | |
224 | /* send tail */ | |
225 | if (xdr->tail[0].iov_len) { | |
44524359 N |
226 | result = kernel_sendpage(sock, rqstp->rq_respages[0], |
227 | ((unsigned long)xdr->tail[0].iov_base) | |
cca5172a | 228 | & (PAGE_SIZE-1), |
1da177e4 LT |
229 | xdr->tail[0].iov_len, 0); |
230 | ||
231 | if (result > 0) | |
232 | len += result; | |
233 | } | |
234 | out: | |
ad06e4bd | 235 | dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n", |
57b1d3ba | 236 | svsk, xdr->head[0].iov_base, xdr->head[0].iov_len, |
ad06e4bd | 237 | xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf))); |
1da177e4 LT |
238 | |
239 | return len; | |
240 | } | |
241 | ||
80212d59 N |
242 | /* |
243 | * Report socket names for nfsdfs | |
244 | */ | |
245 | static int one_sock_name(char *buf, struct svc_sock *svsk) | |
246 | { | |
247 | int len; | |
248 | ||
249 | switch(svsk->sk_sk->sk_family) { | |
250 | case AF_INET: | |
251 | len = sprintf(buf, "ipv4 %s %u.%u.%u.%u %d\n", | |
252 | svsk->sk_sk->sk_protocol==IPPROTO_UDP? | |
253 | "udp" : "tcp", | |
254 | NIPQUAD(inet_sk(svsk->sk_sk)->rcv_saddr), | |
255 | inet_sk(svsk->sk_sk)->num); | |
256 | break; | |
257 | default: | |
258 | len = sprintf(buf, "*unknown-%d*\n", | |
259 | svsk->sk_sk->sk_family); | |
260 | } | |
261 | return len; | |
262 | } | |
263 | ||
264 | int | |
b41b66d6 | 265 | svc_sock_names(char *buf, struct svc_serv *serv, char *toclose) |
80212d59 | 266 | { |
b41b66d6 | 267 | struct svc_sock *svsk, *closesk = NULL; |
80212d59 N |
268 | int len = 0; |
269 | ||
270 | if (!serv) | |
271 | return 0; | |
aaf68cfb | 272 | spin_lock_bh(&serv->sv_lock); |
7a182083 | 273 | list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list) { |
80212d59 | 274 | int onelen = one_sock_name(buf+len, svsk); |
b41b66d6 N |
275 | if (toclose && strcmp(toclose, buf+len) == 0) |
276 | closesk = svsk; | |
277 | else | |
278 | len += onelen; | |
80212d59 | 279 | } |
aaf68cfb | 280 | spin_unlock_bh(&serv->sv_lock); |
b41b66d6 | 281 | if (closesk) |
5680c446 N |
282 | /* Should unregister with portmap, but you cannot |
283 | * unregister just one protocol... | |
284 | */ | |
7a182083 | 285 | svc_close_xprt(&closesk->sk_xprt); |
37a03472 N |
286 | else if (toclose) |
287 | return -ENOENT; | |
80212d59 N |
288 | return len; |
289 | } | |
290 | EXPORT_SYMBOL(svc_sock_names); | |
291 | ||
1da177e4 LT |
292 | /* |
293 | * Check input queue length | |
294 | */ | |
0f0257ea | 295 | static int svc_recv_available(struct svc_sock *svsk) |
1da177e4 | 296 | { |
1da177e4 LT |
297 | struct socket *sock = svsk->sk_sock; |
298 | int avail, err; | |
299 | ||
e6242e92 | 300 | err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail); |
1da177e4 LT |
301 | |
302 | return (err >= 0)? avail : err; | |
303 | } | |
304 | ||
305 | /* | |
306 | * Generic recvfrom routine. | |
307 | */ | |
0f0257ea TT |
308 | static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr, |
309 | int buflen) | |
1da177e4 | 310 | { |
57b1d3ba TT |
311 | struct svc_sock *svsk = |
312 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
1ba95105 CL |
313 | struct msghdr msg = { |
314 | .msg_flags = MSG_DONTWAIT, | |
315 | }; | |
316 | int len; | |
1da177e4 | 317 | |
260c1d12 TT |
318 | rqstp->rq_xprt_hlen = 0; |
319 | ||
1ba95105 CL |
320 | len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen, |
321 | msg.msg_flags); | |
1da177e4 | 322 | |
1da177e4 | 323 | dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n", |
1ba95105 | 324 | svsk, iov[0].iov_base, iov[0].iov_len, len); |
1da177e4 LT |
325 | return len; |
326 | } | |
327 | ||
328 | /* | |
329 | * Set socket snd and rcv buffer lengths | |
330 | */ | |
0f0257ea TT |
331 | static void svc_sock_setbufsize(struct socket *sock, unsigned int snd, |
332 | unsigned int rcv) | |
1da177e4 LT |
333 | { |
334 | #if 0 | |
335 | mm_segment_t oldfs; | |
336 | oldfs = get_fs(); set_fs(KERNEL_DS); | |
337 | sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF, | |
338 | (char*)&snd, sizeof(snd)); | |
339 | sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF, | |
340 | (char*)&rcv, sizeof(rcv)); | |
341 | #else | |
342 | /* sock_setsockopt limits use to sysctl_?mem_max, | |
343 | * which isn't acceptable. Until that is made conditional | |
344 | * on not having CAP_SYS_RESOURCE or similar, we go direct... | |
345 | * DaveM said I could! | |
346 | */ | |
347 | lock_sock(sock->sk); | |
348 | sock->sk->sk_sndbuf = snd * 2; | |
349 | sock->sk->sk_rcvbuf = rcv * 2; | |
350 | sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK|SOCK_RCVBUF_LOCK; | |
351 | release_sock(sock->sk); | |
352 | #endif | |
353 | } | |
354 | /* | |
355 | * INET callback when data has been received on the socket. | |
356 | */ | |
0f0257ea | 357 | static void svc_udp_data_ready(struct sock *sk, int count) |
1da177e4 | 358 | { |
939bb7ef | 359 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 | 360 | |
939bb7ef NB |
361 | if (svsk) { |
362 | dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n", | |
02fc6c36 TT |
363 | svsk, sk, count, |
364 | test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags)); | |
365 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); | |
f6150c3c | 366 | svc_xprt_enqueue(&svsk->sk_xprt); |
939bb7ef | 367 | } |
1da177e4 LT |
368 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
369 | wake_up_interruptible(sk->sk_sleep); | |
370 | } | |
371 | ||
372 | /* | |
373 | * INET callback when space is newly available on the socket. | |
374 | */ | |
0f0257ea | 375 | static void svc_write_space(struct sock *sk) |
1da177e4 LT |
376 | { |
377 | struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data); | |
378 | ||
379 | if (svsk) { | |
380 | dprintk("svc: socket %p(inet %p), write_space busy=%d\n", | |
02fc6c36 | 381 | svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags)); |
f6150c3c | 382 | svc_xprt_enqueue(&svsk->sk_xprt); |
1da177e4 LT |
383 | } |
384 | ||
385 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) { | |
939bb7ef | 386 | dprintk("RPC svc_write_space: someone sleeping on %p\n", |
1da177e4 LT |
387 | svsk); |
388 | wake_up_interruptible(sk->sk_sleep); | |
389 | } | |
390 | } | |
391 | ||
9dbc240f TT |
392 | /* |
393 | * Copy the UDP datagram's destination address to the rqstp structure. | |
394 | * The 'destination' address in this case is the address to which the | |
395 | * peer sent the datagram, i.e. our local address. For multihomed | |
396 | * hosts, this can change from msg to msg. Note that only the IP | |
397 | * address changes, the port number should remain the same. | |
398 | */ | |
399 | static void svc_udp_get_dest_address(struct svc_rqst *rqstp, | |
400 | struct cmsghdr *cmh) | |
95756482 | 401 | { |
57b1d3ba TT |
402 | struct svc_sock *svsk = |
403 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
404 | switch (svsk->sk_sk->sk_family) { | |
95756482 | 405 | case AF_INET: { |
7a37f578 N |
406 | struct in_pktinfo *pki = CMSG_DATA(cmh); |
407 | rqstp->rq_daddr.addr.s_addr = pki->ipi_spec_dst.s_addr; | |
95756482 | 408 | break; |
7a37f578 | 409 | } |
95756482 | 410 | case AF_INET6: { |
7a37f578 N |
411 | struct in6_pktinfo *pki = CMSG_DATA(cmh); |
412 | ipv6_addr_copy(&rqstp->rq_daddr.addr6, &pki->ipi6_addr); | |
95756482 | 413 | break; |
7a37f578 | 414 | } |
95756482 | 415 | } |
95756482 CL |
416 | } |
417 | ||
1da177e4 LT |
418 | /* |
419 | * Receive a datagram from a UDP socket. | |
420 | */ | |
0f0257ea | 421 | static int svc_udp_recvfrom(struct svc_rqst *rqstp) |
1da177e4 | 422 | { |
57b1d3ba TT |
423 | struct svc_sock *svsk = |
424 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
bb5cf160 | 425 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
1da177e4 | 426 | struct sk_buff *skb; |
bc375ea7 DM |
427 | union { |
428 | struct cmsghdr hdr; | |
429 | long all[SVC_PKTINFO_SPACE / sizeof(long)]; | |
430 | } buffer; | |
431 | struct cmsghdr *cmh = &buffer.hdr; | |
1da177e4 | 432 | int err, len; |
7a37f578 N |
433 | struct msghdr msg = { |
434 | .msg_name = svc_addr(rqstp), | |
435 | .msg_control = cmh, | |
436 | .msg_controllen = sizeof(buffer), | |
437 | .msg_flags = MSG_DONTWAIT, | |
438 | }; | |
1da177e4 | 439 | |
02fc6c36 | 440 | if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags)) |
1da177e4 LT |
441 | /* udp sockets need large rcvbuf as all pending |
442 | * requests are still in that buffer. sndbuf must | |
443 | * also be large enough that there is enough space | |
3262c816 GB |
444 | * for one reply per thread. We count all threads |
445 | * rather than threads in a particular pool, which | |
446 | * provides an upper bound on the number of threads | |
447 | * which will access the socket. | |
1da177e4 LT |
448 | */ |
449 | svc_sock_setbufsize(svsk->sk_sock, | |
c6b0a9f8 N |
450 | (serv->sv_nrthreads+3) * serv->sv_max_mesg, |
451 | (serv->sv_nrthreads+3) * serv->sv_max_mesg); | |
1da177e4 | 452 | |
02fc6c36 | 453 | clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
05ed690e N |
454 | skb = NULL; |
455 | err = kernel_recvmsg(svsk->sk_sock, &msg, NULL, | |
456 | 0, 0, MSG_PEEK | MSG_DONTWAIT); | |
457 | if (err >= 0) | |
458 | skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err); | |
459 | ||
460 | if (skb == NULL) { | |
461 | if (err != -EAGAIN) { | |
462 | /* possibly an icmp error */ | |
463 | dprintk("svc: recvfrom returned error %d\n", -err); | |
02fc6c36 | 464 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 465 | } |
a6046f71 | 466 | svc_xprt_received(&svsk->sk_xprt); |
05ed690e | 467 | return -EAGAIN; |
1da177e4 | 468 | } |
9dbc240f TT |
469 | len = svc_addr_len(svc_addr(rqstp)); |
470 | if (len < 0) | |
471 | return len; | |
472 | rqstp->rq_addrlen = len; | |
b7aa0bf7 ED |
473 | if (skb->tstamp.tv64 == 0) { |
474 | skb->tstamp = ktime_get_real(); | |
cca5172a | 475 | /* Don't enable netstamp, sunrpc doesn't |
1da177e4 LT |
476 | need that much accuracy */ |
477 | } | |
b7aa0bf7 | 478 | svsk->sk_sk->sk_stamp = skb->tstamp; |
02fc6c36 | 479 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */ |
1da177e4 LT |
480 | |
481 | /* | |
482 | * Maybe more packets - kick another thread ASAP. | |
483 | */ | |
a6046f71 | 484 | svc_xprt_received(&svsk->sk_xprt); |
1da177e4 LT |
485 | |
486 | len = skb->len - sizeof(struct udphdr); | |
487 | rqstp->rq_arg.len = len; | |
488 | ||
95756482 | 489 | rqstp->rq_prot = IPPROTO_UDP; |
27459f09 | 490 | |
7a37f578 N |
491 | if (cmh->cmsg_level != IPPROTO_IP || |
492 | cmh->cmsg_type != IP_PKTINFO) { | |
493 | if (net_ratelimit()) | |
494 | printk("rpcsvc: received unknown control message:" | |
495 | "%d/%d\n", | |
496 | cmh->cmsg_level, cmh->cmsg_type); | |
497 | skb_free_datagram(svsk->sk_sk, skb); | |
498 | return 0; | |
499 | } | |
500 | svc_udp_get_dest_address(rqstp, cmh); | |
1da177e4 LT |
501 | |
502 | if (skb_is_nonlinear(skb)) { | |
503 | /* we have to copy */ | |
504 | local_bh_disable(); | |
505 | if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) { | |
506 | local_bh_enable(); | |
507 | /* checksum error */ | |
508 | skb_free_datagram(svsk->sk_sk, skb); | |
509 | return 0; | |
510 | } | |
511 | local_bh_enable(); | |
cca5172a | 512 | skb_free_datagram(svsk->sk_sk, skb); |
1da177e4 LT |
513 | } else { |
514 | /* we can use it in-place */ | |
0f0257ea TT |
515 | rqstp->rq_arg.head[0].iov_base = skb->data + |
516 | sizeof(struct udphdr); | |
1da177e4 | 517 | rqstp->rq_arg.head[0].iov_len = len; |
fb286bb2 HX |
518 | if (skb_checksum_complete(skb)) { |
519 | skb_free_datagram(svsk->sk_sk, skb); | |
520 | return 0; | |
1da177e4 | 521 | } |
5148bf4e | 522 | rqstp->rq_xprt_ctxt = skb; |
1da177e4 LT |
523 | } |
524 | ||
525 | rqstp->rq_arg.page_base = 0; | |
526 | if (len <= rqstp->rq_arg.head[0].iov_len) { | |
527 | rqstp->rq_arg.head[0].iov_len = len; | |
528 | rqstp->rq_arg.page_len = 0; | |
44524359 | 529 | rqstp->rq_respages = rqstp->rq_pages+1; |
1da177e4 LT |
530 | } else { |
531 | rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len; | |
44524359 | 532 | rqstp->rq_respages = rqstp->rq_pages + 1 + |
172589cc | 533 | DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE); |
1da177e4 LT |
534 | } |
535 | ||
536 | if (serv->sv_stats) | |
537 | serv->sv_stats->netudpcnt++; | |
538 | ||
539 | return len; | |
540 | } | |
541 | ||
542 | static int | |
543 | svc_udp_sendto(struct svc_rqst *rqstp) | |
544 | { | |
545 | int error; | |
546 | ||
547 | error = svc_sendto(rqstp, &rqstp->rq_res); | |
548 | if (error == -ECONNREFUSED) | |
549 | /* ICMP error on earlier request. */ | |
550 | error = svc_sendto(rqstp, &rqstp->rq_res); | |
551 | ||
552 | return error; | |
553 | } | |
554 | ||
e831fe65 TT |
555 | static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp) |
556 | { | |
557 | } | |
558 | ||
323bee32 TT |
559 | static int svc_udp_has_wspace(struct svc_xprt *xprt) |
560 | { | |
561 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
bb5cf160 | 562 | struct svc_serv *serv = xprt->xpt_server; |
323bee32 TT |
563 | unsigned long required; |
564 | ||
565 | /* | |
566 | * Set the SOCK_NOSPACE flag before checking the available | |
567 | * sock space. | |
568 | */ | |
569 | set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
7a90e8cc | 570 | required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg; |
323bee32 TT |
571 | if (required*2 > sock_wspace(svsk->sk_sk)) |
572 | return 0; | |
573 | clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
574 | return 1; | |
575 | } | |
576 | ||
38a417cc TT |
577 | static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt) |
578 | { | |
579 | BUG(); | |
580 | return NULL; | |
581 | } | |
582 | ||
b700cbb1 TT |
583 | static struct svc_xprt *svc_udp_create(struct svc_serv *serv, |
584 | struct sockaddr *sa, int salen, | |
585 | int flags) | |
586 | { | |
587 | return svc_create_socket(serv, IPPROTO_UDP, sa, salen, flags); | |
588 | } | |
589 | ||
360d8738 | 590 | static struct svc_xprt_ops svc_udp_ops = { |
b700cbb1 | 591 | .xpo_create = svc_udp_create, |
5d137990 TT |
592 | .xpo_recvfrom = svc_udp_recvfrom, |
593 | .xpo_sendto = svc_udp_sendto, | |
5148bf4e | 594 | .xpo_release_rqst = svc_release_skb, |
755cceab TT |
595 | .xpo_detach = svc_sock_detach, |
596 | .xpo_free = svc_sock_free, | |
e831fe65 | 597 | .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr, |
323bee32 | 598 | .xpo_has_wspace = svc_udp_has_wspace, |
38a417cc | 599 | .xpo_accept = svc_udp_accept, |
360d8738 TT |
600 | }; |
601 | ||
602 | static struct svc_xprt_class svc_udp_class = { | |
603 | .xcl_name = "udp", | |
b700cbb1 | 604 | .xcl_owner = THIS_MODULE, |
360d8738 | 605 | .xcl_ops = &svc_udp_ops, |
49023155 | 606 | .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP, |
360d8738 TT |
607 | }; |
608 | ||
bb5cf160 | 609 | static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv) |
1da177e4 | 610 | { |
7a37f578 N |
611 | int one = 1; |
612 | mm_segment_t oldfs; | |
613 | ||
bb5cf160 | 614 | svc_xprt_init(&svc_udp_class, &svsk->sk_xprt, serv); |
def13d74 | 615 | clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
616 | svsk->sk_sk->sk_data_ready = svc_udp_data_ready; |
617 | svsk->sk_sk->sk_write_space = svc_write_space; | |
1da177e4 LT |
618 | |
619 | /* initialise setting must have enough space to | |
cca5172a | 620 | * receive and respond to one request. |
1da177e4 LT |
621 | * svc_udp_recvfrom will re-adjust if necessary |
622 | */ | |
623 | svc_sock_setbufsize(svsk->sk_sock, | |
bb5cf160 TT |
624 | 3 * svsk->sk_xprt.xpt_server->sv_max_mesg, |
625 | 3 * svsk->sk_xprt.xpt_server->sv_max_mesg); | |
1da177e4 | 626 | |
0f0257ea TT |
627 | /* data might have come in before data_ready set up */ |
628 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); | |
02fc6c36 | 629 | set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); |
7a37f578 N |
630 | |
631 | oldfs = get_fs(); | |
632 | set_fs(KERNEL_DS); | |
633 | /* make sure we get destination address info */ | |
634 | svsk->sk_sock->ops->setsockopt(svsk->sk_sock, IPPROTO_IP, IP_PKTINFO, | |
635 | (char __user *)&one, sizeof(one)); | |
636 | set_fs(oldfs); | |
1da177e4 LT |
637 | } |
638 | ||
639 | /* | |
640 | * A data_ready event on a listening socket means there's a connection | |
641 | * pending. Do not use state_change as a substitute for it. | |
642 | */ | |
0f0257ea | 643 | static void svc_tcp_listen_data_ready(struct sock *sk, int count_unused) |
1da177e4 | 644 | { |
939bb7ef | 645 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 LT |
646 | |
647 | dprintk("svc: socket %p TCP (listen) state change %d\n", | |
939bb7ef | 648 | sk, sk->sk_state); |
1da177e4 | 649 | |
939bb7ef NB |
650 | /* |
651 | * This callback may called twice when a new connection | |
652 | * is established as a child socket inherits everything | |
653 | * from a parent LISTEN socket. | |
654 | * 1) data_ready method of the parent socket will be called | |
655 | * when one of child sockets become ESTABLISHED. | |
656 | * 2) data_ready method of the child socket may be called | |
657 | * when it receives data before the socket is accepted. | |
658 | * In case of 2, we should ignore it silently. | |
659 | */ | |
660 | if (sk->sk_state == TCP_LISTEN) { | |
661 | if (svsk) { | |
02fc6c36 | 662 | set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
f6150c3c | 663 | svc_xprt_enqueue(&svsk->sk_xprt); |
939bb7ef NB |
664 | } else |
665 | printk("svc: socket %p: no user data\n", sk); | |
1da177e4 | 666 | } |
939bb7ef | 667 | |
1da177e4 LT |
668 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
669 | wake_up_interruptible_all(sk->sk_sleep); | |
670 | } | |
671 | ||
672 | /* | |
673 | * A state change on a connected socket means it's dying or dead. | |
674 | */ | |
0f0257ea | 675 | static void svc_tcp_state_change(struct sock *sk) |
1da177e4 | 676 | { |
939bb7ef | 677 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 LT |
678 | |
679 | dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n", | |
939bb7ef | 680 | sk, sk->sk_state, sk->sk_user_data); |
1da177e4 | 681 | |
939bb7ef | 682 | if (!svsk) |
1da177e4 | 683 | printk("svc: socket %p: no user data\n", sk); |
939bb7ef | 684 | else { |
02fc6c36 | 685 | set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); |
f6150c3c | 686 | svc_xprt_enqueue(&svsk->sk_xprt); |
1da177e4 | 687 | } |
1da177e4 LT |
688 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
689 | wake_up_interruptible_all(sk->sk_sleep); | |
690 | } | |
691 | ||
0f0257ea | 692 | static void svc_tcp_data_ready(struct sock *sk, int count) |
1da177e4 | 693 | { |
939bb7ef | 694 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 LT |
695 | |
696 | dprintk("svc: socket %p TCP data ready (svsk %p)\n", | |
939bb7ef NB |
697 | sk, sk->sk_user_data); |
698 | if (svsk) { | |
02fc6c36 | 699 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
f6150c3c | 700 | svc_xprt_enqueue(&svsk->sk_xprt); |
939bb7ef | 701 | } |
1da177e4 LT |
702 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
703 | wake_up_interruptible(sk->sk_sleep); | |
704 | } | |
705 | ||
706 | /* | |
707 | * Accept a TCP connection | |
708 | */ | |
38a417cc | 709 | static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt) |
1da177e4 | 710 | { |
38a417cc | 711 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); |
cdd88b9f AM |
712 | struct sockaddr_storage addr; |
713 | struct sockaddr *sin = (struct sockaddr *) &addr; | |
bb5cf160 | 714 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
1da177e4 LT |
715 | struct socket *sock = svsk->sk_sock; |
716 | struct socket *newsock; | |
1da177e4 LT |
717 | struct svc_sock *newsvsk; |
718 | int err, slen; | |
ad06e4bd | 719 | char buf[RPC_MAX_ADDRBUFLEN]; |
1da177e4 LT |
720 | |
721 | dprintk("svc: tcp_accept %p sock %p\n", svsk, sock); | |
722 | if (!sock) | |
38a417cc | 723 | return NULL; |
1da177e4 | 724 | |
02fc6c36 | 725 | clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
e6242e92 SS |
726 | err = kernel_accept(sock, &newsock, O_NONBLOCK); |
727 | if (err < 0) { | |
1da177e4 LT |
728 | if (err == -ENOMEM) |
729 | printk(KERN_WARNING "%s: no more sockets!\n", | |
730 | serv->sv_name); | |
e6242e92 | 731 | else if (err != -EAGAIN && net_ratelimit()) |
1da177e4 LT |
732 | printk(KERN_WARNING "%s: accept failed (err %d)!\n", |
733 | serv->sv_name, -err); | |
38a417cc | 734 | return NULL; |
1da177e4 | 735 | } |
02fc6c36 | 736 | set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 737 | |
cdd88b9f | 738 | err = kernel_getpeername(newsock, sin, &slen); |
1da177e4 LT |
739 | if (err < 0) { |
740 | if (net_ratelimit()) | |
741 | printk(KERN_WARNING "%s: peername failed (err %d)!\n", | |
742 | serv->sv_name, -err); | |
743 | goto failed; /* aborted connection or whatever */ | |
744 | } | |
745 | ||
746 | /* Ideally, we would want to reject connections from unauthorized | |
ad06e4bd CL |
747 | * hosts here, but when we get encryption, the IP of the host won't |
748 | * tell us anything. For now just warn about unpriv connections. | |
1da177e4 | 749 | */ |
cdd88b9f | 750 | if (!svc_port_is_privileged(sin)) { |
1da177e4 | 751 | dprintk(KERN_WARNING |
ad06e4bd | 752 | "%s: connect from unprivileged port: %s\n", |
cca5172a | 753 | serv->sv_name, |
cdd88b9f | 754 | __svc_print_addr(sin, buf, sizeof(buf))); |
1da177e4 | 755 | } |
ad06e4bd | 756 | dprintk("%s: connect from %s\n", serv->sv_name, |
cdd88b9f | 757 | __svc_print_addr(sin, buf, sizeof(buf))); |
1da177e4 LT |
758 | |
759 | /* make sure that a write doesn't block forever when | |
760 | * low on memory | |
761 | */ | |
762 | newsock->sk->sk_sndtimeo = HZ*30; | |
763 | ||
6b174337 CL |
764 | if (!(newsvsk = svc_setup_socket(serv, newsock, &err, |
765 | (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY)))) | |
1da177e4 | 766 | goto failed; |
9dbc240f | 767 | svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen); |
a9747692 FM |
768 | err = kernel_getsockname(newsock, sin, &slen); |
769 | if (unlikely(err < 0)) { | |
770 | dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err); | |
771 | slen = offsetof(struct sockaddr, sa_data); | |
772 | } | |
9dbc240f | 773 | svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen); |
067d7817 | 774 | |
f9f3cc4f TT |
775 | if (serv->sv_stats) |
776 | serv->sv_stats->nettcpconn++; | |
777 | ||
778 | return &newsvsk->sk_xprt; | |
779 | ||
780 | failed: | |
781 | sock_release(newsock); | |
782 | return NULL; | |
783 | } | |
784 | ||
1da177e4 LT |
785 | /* |
786 | * Receive data from a TCP socket. | |
787 | */ | |
0f0257ea | 788 | static int svc_tcp_recvfrom(struct svc_rqst *rqstp) |
1da177e4 | 789 | { |
57b1d3ba TT |
790 | struct svc_sock *svsk = |
791 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
bb5cf160 | 792 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
1da177e4 | 793 | int len; |
3cc03b16 | 794 | struct kvec *vec; |
1da177e4 LT |
795 | int pnum, vlen; |
796 | ||
797 | dprintk("svc: tcp_recv %p data %d conn %d close %d\n", | |
02fc6c36 TT |
798 | svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags), |
799 | test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags), | |
800 | test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags)); | |
1da177e4 | 801 | |
02fc6c36 | 802 | if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags)) |
1da177e4 LT |
803 | /* sndbuf needs to have room for one request |
804 | * per thread, otherwise we can stall even when the | |
805 | * network isn't a bottleneck. | |
3262c816 GB |
806 | * |
807 | * We count all threads rather than threads in a | |
808 | * particular pool, which provides an upper bound | |
809 | * on the number of threads which will access the socket. | |
810 | * | |
1da177e4 | 811 | * rcvbuf just needs to be able to hold a few requests. |
cca5172a | 812 | * Normally they will be removed from the queue |
1da177e4 LT |
813 | * as soon a a complete request arrives. |
814 | */ | |
815 | svc_sock_setbufsize(svsk->sk_sock, | |
c6b0a9f8 N |
816 | (serv->sv_nrthreads+3) * serv->sv_max_mesg, |
817 | 3 * serv->sv_max_mesg); | |
1da177e4 | 818 | |
02fc6c36 | 819 | clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
820 | |
821 | /* Receive data. If we haven't got the record length yet, get | |
822 | * the next four bytes. Otherwise try to gobble up as much as | |
823 | * possible up to the complete record length. | |
824 | */ | |
825 | if (svsk->sk_tcplen < 4) { | |
826 | unsigned long want = 4 - svsk->sk_tcplen; | |
827 | struct kvec iov; | |
828 | ||
829 | iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen; | |
830 | iov.iov_len = want; | |
831 | if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0) | |
832 | goto error; | |
833 | svsk->sk_tcplen += len; | |
834 | ||
835 | if (len < want) { | |
836 | dprintk("svc: short recvfrom while reading record length (%d of %lu)\n", | |
cca5172a | 837 | len, want); |
a6046f71 | 838 | svc_xprt_received(&svsk->sk_xprt); |
1da177e4 LT |
839 | return -EAGAIN; /* record header not complete */ |
840 | } | |
841 | ||
842 | svsk->sk_reclen = ntohl(svsk->sk_reclen); | |
843 | if (!(svsk->sk_reclen & 0x80000000)) { | |
844 | /* FIXME: technically, a record can be fragmented, | |
845 | * and non-terminal fragments will not have the top | |
846 | * bit set in the fragment length header. | |
847 | * But apparently no known nfs clients send fragmented | |
848 | * records. */ | |
34e9a63b N |
849 | if (net_ratelimit()) |
850 | printk(KERN_NOTICE "RPC: bad TCP reclen 0x%08lx" | |
851 | " (non-terminal)\n", | |
852 | (unsigned long) svsk->sk_reclen); | |
1da177e4 LT |
853 | goto err_delete; |
854 | } | |
855 | svsk->sk_reclen &= 0x7fffffff; | |
856 | dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen); | |
c6b0a9f8 | 857 | if (svsk->sk_reclen > serv->sv_max_mesg) { |
34e9a63b N |
858 | if (net_ratelimit()) |
859 | printk(KERN_NOTICE "RPC: bad TCP reclen 0x%08lx" | |
860 | " (large)\n", | |
861 | (unsigned long) svsk->sk_reclen); | |
1da177e4 LT |
862 | goto err_delete; |
863 | } | |
864 | } | |
865 | ||
866 | /* Check whether enough data is available */ | |
867 | len = svc_recv_available(svsk); | |
868 | if (len < 0) | |
869 | goto error; | |
870 | ||
871 | if (len < svsk->sk_reclen) { | |
872 | dprintk("svc: incomplete TCP record (%d of %d)\n", | |
873 | len, svsk->sk_reclen); | |
a6046f71 | 874 | svc_xprt_received(&svsk->sk_xprt); |
1da177e4 LT |
875 | return -EAGAIN; /* record not complete */ |
876 | } | |
877 | len = svsk->sk_reclen; | |
02fc6c36 | 878 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 879 | |
3cc03b16 | 880 | vec = rqstp->rq_vec; |
1da177e4 LT |
881 | vec[0] = rqstp->rq_arg.head[0]; |
882 | vlen = PAGE_SIZE; | |
883 | pnum = 1; | |
884 | while (vlen < len) { | |
44524359 | 885 | vec[pnum].iov_base = page_address(rqstp->rq_pages[pnum]); |
1da177e4 LT |
886 | vec[pnum].iov_len = PAGE_SIZE; |
887 | pnum++; | |
888 | vlen += PAGE_SIZE; | |
889 | } | |
44524359 | 890 | rqstp->rq_respages = &rqstp->rq_pages[pnum]; |
1da177e4 LT |
891 | |
892 | /* Now receive data */ | |
893 | len = svc_recvfrom(rqstp, vec, pnum, len); | |
894 | if (len < 0) | |
895 | goto error; | |
896 | ||
897 | dprintk("svc: TCP complete record (%d bytes)\n", len); | |
898 | rqstp->rq_arg.len = len; | |
899 | rqstp->rq_arg.page_base = 0; | |
900 | if (len <= rqstp->rq_arg.head[0].iov_len) { | |
901 | rqstp->rq_arg.head[0].iov_len = len; | |
902 | rqstp->rq_arg.page_len = 0; | |
903 | } else { | |
904 | rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len; | |
905 | } | |
906 | ||
5148bf4e | 907 | rqstp->rq_xprt_ctxt = NULL; |
1da177e4 LT |
908 | rqstp->rq_prot = IPPROTO_TCP; |
909 | ||
910 | /* Reset TCP read info */ | |
911 | svsk->sk_reclen = 0; | |
912 | svsk->sk_tcplen = 0; | |
913 | ||
9dbc240f | 914 | svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt); |
a6046f71 | 915 | svc_xprt_received(&svsk->sk_xprt); |
1da177e4 LT |
916 | if (serv->sv_stats) |
917 | serv->sv_stats->nettcpcnt++; | |
918 | ||
919 | return len; | |
920 | ||
921 | err_delete: | |
02fc6c36 | 922 | set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
923 | return -EAGAIN; |
924 | ||
925 | error: | |
926 | if (len == -EAGAIN) { | |
927 | dprintk("RPC: TCP recvfrom got EAGAIN\n"); | |
a6046f71 | 928 | svc_xprt_received(&svsk->sk_xprt); |
1da177e4 LT |
929 | } else { |
930 | printk(KERN_NOTICE "%s: recvfrom returned errno %d\n", | |
bb5cf160 | 931 | svsk->sk_xprt.xpt_server->sv_name, -len); |
93fbf1a5 | 932 | goto err_delete; |
1da177e4 LT |
933 | } |
934 | ||
935 | return len; | |
936 | } | |
937 | ||
938 | /* | |
939 | * Send out data on TCP socket. | |
940 | */ | |
0f0257ea | 941 | static int svc_tcp_sendto(struct svc_rqst *rqstp) |
1da177e4 LT |
942 | { |
943 | struct xdr_buf *xbufp = &rqstp->rq_res; | |
944 | int sent; | |
d8ed029d | 945 | __be32 reclen; |
1da177e4 LT |
946 | |
947 | /* Set up the first element of the reply kvec. | |
948 | * Any other kvecs that may be in use have been taken | |
949 | * care of by the server implementation itself. | |
950 | */ | |
951 | reclen = htonl(0x80000000|((xbufp->len ) - 4)); | |
952 | memcpy(xbufp->head[0].iov_base, &reclen, 4); | |
953 | ||
57b1d3ba | 954 | if (test_bit(XPT_DEAD, &rqstp->rq_xprt->xpt_flags)) |
1da177e4 LT |
955 | return -ENOTCONN; |
956 | ||
957 | sent = svc_sendto(rqstp, &rqstp->rq_res); | |
958 | if (sent != xbufp->len) { | |
0f0257ea TT |
959 | printk(KERN_NOTICE |
960 | "rpc-srv/tcp: %s: %s %d when sending %d bytes " | |
961 | "- shutting down socket\n", | |
57b1d3ba | 962 | rqstp->rq_xprt->xpt_server->sv_name, |
1da177e4 LT |
963 | (sent<0)?"got error":"sent only", |
964 | sent, xbufp->len); | |
57b1d3ba | 965 | set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags); |
f6150c3c | 966 | svc_xprt_enqueue(rqstp->rq_xprt); |
1da177e4 LT |
967 | sent = -EAGAIN; |
968 | } | |
969 | return sent; | |
970 | } | |
971 | ||
e831fe65 TT |
972 | /* |
973 | * Setup response header. TCP has a 4B record length field. | |
974 | */ | |
975 | static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp) | |
976 | { | |
977 | struct kvec *resv = &rqstp->rq_res.head[0]; | |
978 | ||
979 | /* tcp needs a space for the record length... */ | |
980 | svc_putnl(resv, 0); | |
981 | } | |
982 | ||
323bee32 TT |
983 | static int svc_tcp_has_wspace(struct svc_xprt *xprt) |
984 | { | |
57b1d3ba | 985 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); |
bb5cf160 | 986 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
323bee32 TT |
987 | int required; |
988 | int wspace; | |
989 | ||
990 | /* | |
991 | * Set the SOCK_NOSPACE flag before checking the available | |
992 | * sock space. | |
993 | */ | |
994 | set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
7a90e8cc | 995 | required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg; |
323bee32 TT |
996 | wspace = sk_stream_wspace(svsk->sk_sk); |
997 | ||
998 | if (wspace < sk_stream_min_wspace(svsk->sk_sk)) | |
999 | return 0; | |
1000 | if (required * 2 > wspace) | |
1001 | return 0; | |
1002 | ||
1003 | clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
1004 | return 1; | |
1005 | } | |
1006 | ||
b700cbb1 TT |
1007 | static struct svc_xprt *svc_tcp_create(struct svc_serv *serv, |
1008 | struct sockaddr *sa, int salen, | |
1009 | int flags) | |
1010 | { | |
1011 | return svc_create_socket(serv, IPPROTO_TCP, sa, salen, flags); | |
1012 | } | |
1013 | ||
360d8738 | 1014 | static struct svc_xprt_ops svc_tcp_ops = { |
b700cbb1 | 1015 | .xpo_create = svc_tcp_create, |
5d137990 TT |
1016 | .xpo_recvfrom = svc_tcp_recvfrom, |
1017 | .xpo_sendto = svc_tcp_sendto, | |
5148bf4e | 1018 | .xpo_release_rqst = svc_release_skb, |
755cceab TT |
1019 | .xpo_detach = svc_sock_detach, |
1020 | .xpo_free = svc_sock_free, | |
e831fe65 | 1021 | .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr, |
323bee32 | 1022 | .xpo_has_wspace = svc_tcp_has_wspace, |
38a417cc | 1023 | .xpo_accept = svc_tcp_accept, |
360d8738 TT |
1024 | }; |
1025 | ||
1026 | static struct svc_xprt_class svc_tcp_class = { | |
1027 | .xcl_name = "tcp", | |
b700cbb1 | 1028 | .xcl_owner = THIS_MODULE, |
360d8738 | 1029 | .xcl_ops = &svc_tcp_ops, |
49023155 | 1030 | .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP, |
360d8738 TT |
1031 | }; |
1032 | ||
1033 | void svc_init_xprt_sock(void) | |
1034 | { | |
1035 | svc_reg_xprt_class(&svc_tcp_class); | |
1036 | svc_reg_xprt_class(&svc_udp_class); | |
1037 | } | |
1038 | ||
1039 | void svc_cleanup_xprt_sock(void) | |
1040 | { | |
1041 | svc_unreg_xprt_class(&svc_tcp_class); | |
1042 | svc_unreg_xprt_class(&svc_udp_class); | |
1043 | } | |
1044 | ||
bb5cf160 | 1045 | static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv) |
1da177e4 LT |
1046 | { |
1047 | struct sock *sk = svsk->sk_sk; | |
1048 | struct tcp_sock *tp = tcp_sk(sk); | |
1049 | ||
bb5cf160 | 1050 | svc_xprt_init(&svc_tcp_class, &svsk->sk_xprt, serv); |
def13d74 | 1051 | set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
1052 | if (sk->sk_state == TCP_LISTEN) { |
1053 | dprintk("setting up TCP socket for listening\n"); | |
02fc6c36 | 1054 | set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 1055 | sk->sk_data_ready = svc_tcp_listen_data_ready; |
02fc6c36 | 1056 | set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
1057 | } else { |
1058 | dprintk("setting up TCP socket for reading\n"); | |
1059 | sk->sk_state_change = svc_tcp_state_change; | |
1060 | sk->sk_data_ready = svc_tcp_data_ready; | |
1061 | sk->sk_write_space = svc_write_space; | |
1062 | ||
1063 | svsk->sk_reclen = 0; | |
1064 | svsk->sk_tcplen = 0; | |
1065 | ||
1066 | tp->nonagle = 1; /* disable Nagle's algorithm */ | |
1067 | ||
1068 | /* initialise setting must have enough space to | |
cca5172a | 1069 | * receive and respond to one request. |
1da177e4 LT |
1070 | * svc_tcp_recvfrom will re-adjust if necessary |
1071 | */ | |
1072 | svc_sock_setbufsize(svsk->sk_sock, | |
bb5cf160 TT |
1073 | 3 * svsk->sk_xprt.xpt_server->sv_max_mesg, |
1074 | 3 * svsk->sk_xprt.xpt_server->sv_max_mesg); | |
1da177e4 | 1075 | |
02fc6c36 TT |
1076 | set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); |
1077 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); | |
cca5172a | 1078 | if (sk->sk_state != TCP_ESTABLISHED) |
02fc6c36 | 1079 | set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
1080 | } |
1081 | } | |
1082 | ||
0f0257ea | 1083 | void svc_sock_update_bufs(struct svc_serv *serv) |
1da177e4 LT |
1084 | { |
1085 | /* | |
1086 | * The number of server threads has changed. Update | |
1087 | * rcvbuf and sndbuf accordingly on all sockets | |
1088 | */ | |
1089 | struct list_head *le; | |
1090 | ||
1091 | spin_lock_bh(&serv->sv_lock); | |
1092 | list_for_each(le, &serv->sv_permsocks) { | |
cca5172a | 1093 | struct svc_sock *svsk = |
7a182083 | 1094 | list_entry(le, struct svc_sock, sk_xprt.xpt_list); |
02fc6c36 | 1095 | set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
1096 | } |
1097 | list_for_each(le, &serv->sv_tempsocks) { | |
1098 | struct svc_sock *svsk = | |
7a182083 | 1099 | list_entry(le, struct svc_sock, sk_xprt.xpt_list); |
02fc6c36 | 1100 | set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
1101 | } |
1102 | spin_unlock_bh(&serv->sv_lock); | |
1103 | } | |
1104 | ||
1da177e4 LT |
1105 | /* |
1106 | * Initialize socket for RPC use and create svc_sock struct | |
1107 | * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF. | |
1108 | */ | |
6b174337 CL |
1109 | static struct svc_sock *svc_setup_socket(struct svc_serv *serv, |
1110 | struct socket *sock, | |
1111 | int *errp, int flags) | |
1da177e4 LT |
1112 | { |
1113 | struct svc_sock *svsk; | |
1114 | struct sock *inet; | |
6b174337 | 1115 | int pmap_register = !(flags & SVC_SOCK_ANONYMOUS); |
1da177e4 LT |
1116 | |
1117 | dprintk("svc: svc_setup_socket %p\n", sock); | |
0da974f4 | 1118 | if (!(svsk = kzalloc(sizeof(*svsk), GFP_KERNEL))) { |
1da177e4 LT |
1119 | *errp = -ENOMEM; |
1120 | return NULL; | |
1121 | } | |
1da177e4 LT |
1122 | |
1123 | inet = sock->sk; | |
1124 | ||
1125 | /* Register socket with portmapper */ | |
1126 | if (*errp >= 0 && pmap_register) | |
1127 | *errp = svc_register(serv, inet->sk_protocol, | |
1128 | ntohs(inet_sk(inet)->sport)); | |
1129 | ||
1130 | if (*errp < 0) { | |
1131 | kfree(svsk); | |
1132 | return NULL; | |
1133 | } | |
1134 | ||
1da177e4 LT |
1135 | inet->sk_user_data = svsk; |
1136 | svsk->sk_sock = sock; | |
1137 | svsk->sk_sk = inet; | |
1138 | svsk->sk_ostate = inet->sk_state_change; | |
1139 | svsk->sk_odata = inet->sk_data_ready; | |
1140 | svsk->sk_owspace = inet->sk_write_space; | |
1da177e4 LT |
1141 | |
1142 | /* Initialize the socket */ | |
1143 | if (sock->type == SOCK_DGRAM) | |
bb5cf160 | 1144 | svc_udp_init(svsk, serv); |
1da177e4 | 1145 | else |
bb5cf160 | 1146 | svc_tcp_init(svsk, serv); |
1da177e4 | 1147 | |
1da177e4 LT |
1148 | dprintk("svc: svc_setup_socket created %p (inet %p)\n", |
1149 | svsk, svsk->sk_sk); | |
1150 | ||
1da177e4 LT |
1151 | return svsk; |
1152 | } | |
1153 | ||
b41b66d6 N |
1154 | int svc_addsock(struct svc_serv *serv, |
1155 | int fd, | |
1156 | char *name_return, | |
1157 | int *proto) | |
1158 | { | |
1159 | int err = 0; | |
1160 | struct socket *so = sockfd_lookup(fd, &err); | |
1161 | struct svc_sock *svsk = NULL; | |
1162 | ||
1163 | if (!so) | |
1164 | return err; | |
1165 | if (so->sk->sk_family != AF_INET) | |
1166 | err = -EAFNOSUPPORT; | |
1167 | else if (so->sk->sk_protocol != IPPROTO_TCP && | |
1168 | so->sk->sk_protocol != IPPROTO_UDP) | |
1169 | err = -EPROTONOSUPPORT; | |
1170 | else if (so->state > SS_UNCONNECTED) | |
1171 | err = -EISCONN; | |
1172 | else { | |
6b174337 | 1173 | svsk = svc_setup_socket(serv, so, &err, SVC_SOCK_DEFAULTS); |
e79eff1f | 1174 | if (svsk) { |
9dbc240f TT |
1175 | struct sockaddr_storage addr; |
1176 | struct sockaddr *sin = (struct sockaddr *)&addr; | |
1177 | int salen; | |
1178 | if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0) | |
1179 | svc_xprt_set_local(&svsk->sk_xprt, sin, salen); | |
4e5caaa5 TT |
1180 | clear_bit(XPT_TEMP, &svsk->sk_xprt.xpt_flags); |
1181 | spin_lock_bh(&serv->sv_lock); | |
1182 | list_add(&svsk->sk_xprt.xpt_list, &serv->sv_permsocks); | |
1183 | spin_unlock_bh(&serv->sv_lock); | |
a6046f71 | 1184 | svc_xprt_received(&svsk->sk_xprt); |
b41b66d6 | 1185 | err = 0; |
e79eff1f | 1186 | } |
b41b66d6 N |
1187 | } |
1188 | if (err) { | |
1189 | sockfd_put(so); | |
1190 | return err; | |
1191 | } | |
1192 | if (proto) *proto = so->sk->sk_protocol; | |
1193 | return one_sock_name(name_return, svsk); | |
1194 | } | |
1195 | EXPORT_SYMBOL_GPL(svc_addsock); | |
1196 | ||
1da177e4 LT |
1197 | /* |
1198 | * Create socket for RPC service. | |
1199 | */ | |
b700cbb1 TT |
1200 | static struct svc_xprt *svc_create_socket(struct svc_serv *serv, |
1201 | int protocol, | |
1202 | struct sockaddr *sin, int len, | |
1203 | int flags) | |
1da177e4 LT |
1204 | { |
1205 | struct svc_sock *svsk; | |
1206 | struct socket *sock; | |
1207 | int error; | |
1208 | int type; | |
ad06e4bd | 1209 | char buf[RPC_MAX_ADDRBUFLEN]; |
9dbc240f TT |
1210 | struct sockaddr_storage addr; |
1211 | struct sockaddr *newsin = (struct sockaddr *)&addr; | |
1212 | int newlen; | |
1da177e4 | 1213 | |
ad06e4bd CL |
1214 | dprintk("svc: svc_create_socket(%s, %d, %s)\n", |
1215 | serv->sv_program->pg_name, protocol, | |
77f1f67a | 1216 | __svc_print_addr(sin, buf, sizeof(buf))); |
1da177e4 LT |
1217 | |
1218 | if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) { | |
1219 | printk(KERN_WARNING "svc: only UDP and TCP " | |
1220 | "sockets supported\n"); | |
b700cbb1 | 1221 | return ERR_PTR(-EINVAL); |
1da177e4 LT |
1222 | } |
1223 | type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM; | |
1224 | ||
77f1f67a CL |
1225 | error = sock_create_kern(sin->sa_family, type, protocol, &sock); |
1226 | if (error < 0) | |
b700cbb1 | 1227 | return ERR_PTR(error); |
1da177e4 | 1228 | |
ed07536e PZ |
1229 | svc_reclassify_socket(sock); |
1230 | ||
18114746 | 1231 | if (type == SOCK_STREAM) |
77f1f67a CL |
1232 | sock->sk->sk_reuse = 1; /* allow address reuse */ |
1233 | error = kernel_bind(sock, sin, len); | |
18114746 ES |
1234 | if (error < 0) |
1235 | goto bummer; | |
1da177e4 | 1236 | |
9dbc240f TT |
1237 | newlen = len; |
1238 | error = kernel_getsockname(sock, newsin, &newlen); | |
1239 | if (error < 0) | |
1240 | goto bummer; | |
1241 | ||
1da177e4 | 1242 | if (protocol == IPPROTO_TCP) { |
e6242e92 | 1243 | if ((error = kernel_listen(sock, 64)) < 0) |
1da177e4 LT |
1244 | goto bummer; |
1245 | } | |
1246 | ||
e79eff1f | 1247 | if ((svsk = svc_setup_socket(serv, sock, &error, flags)) != NULL) { |
9dbc240f | 1248 | svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen); |
b700cbb1 | 1249 | return (struct svc_xprt *)svsk; |
e79eff1f | 1250 | } |
1da177e4 LT |
1251 | |
1252 | bummer: | |
1253 | dprintk("svc: svc_create_socket error = %d\n", -error); | |
1254 | sock_release(sock); | |
b700cbb1 | 1255 | return ERR_PTR(error); |
1da177e4 LT |
1256 | } |
1257 | ||
755cceab TT |
1258 | /* |
1259 | * Detach the svc_sock from the socket so that no | |
1260 | * more callbacks occur. | |
1261 | */ | |
1262 | static void svc_sock_detach(struct svc_xprt *xprt) | |
1263 | { | |
1264 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
1265 | struct sock *sk = svsk->sk_sk; | |
1266 | ||
1267 | dprintk("svc: svc_sock_detach(%p)\n", svsk); | |
1268 | ||
1269 | /* put back the old socket callbacks */ | |
1270 | sk->sk_state_change = svsk->sk_ostate; | |
1271 | sk->sk_data_ready = svsk->sk_odata; | |
1272 | sk->sk_write_space = svsk->sk_owspace; | |
1273 | } | |
1274 | ||
1275 | /* | |
1276 | * Free the svc_sock's socket resources and the svc_sock itself. | |
1277 | */ | |
1278 | static void svc_sock_free(struct svc_xprt *xprt) | |
1279 | { | |
1280 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
1281 | dprintk("svc: svc_sock_free(%p)\n", svsk); | |
1282 | ||
755cceab TT |
1283 | if (svsk->sk_sock->file) |
1284 | sockfd_put(svsk->sk_sock); | |
1285 | else | |
1286 | sock_release(svsk->sk_sock); | |
1287 | kfree(svsk); | |
1288 | } |