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 TT |
61 | static void svc_sock_detach(struct svc_xprt *); |
62 | static void svc_sock_free(struct svc_xprt *); | |
1da177e4 | 63 | |
b700cbb1 TT |
64 | static struct svc_xprt *svc_create_socket(struct svc_serv *, int, |
65 | struct sockaddr *, int, int); | |
ed07536e PZ |
66 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
67 | static struct lock_class_key svc_key[2]; | |
68 | static struct lock_class_key svc_slock_key[2]; | |
69 | ||
0f0257ea | 70 | static void svc_reclassify_socket(struct socket *sock) |
ed07536e PZ |
71 | { |
72 | struct sock *sk = sock->sk; | |
02b3d346 | 73 | BUG_ON(sock_owned_by_user(sk)); |
ed07536e PZ |
74 | switch (sk->sk_family) { |
75 | case AF_INET: | |
76 | sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD", | |
def13d74 TT |
77 | &svc_slock_key[0], |
78 | "sk_xprt.xpt_lock-AF_INET-NFSD", | |
79 | &svc_key[0]); | |
ed07536e PZ |
80 | break; |
81 | ||
82 | case AF_INET6: | |
83 | sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD", | |
def13d74 TT |
84 | &svc_slock_key[1], |
85 | "sk_xprt.xpt_lock-AF_INET6-NFSD", | |
86 | &svc_key[1]); | |
ed07536e PZ |
87 | break; |
88 | ||
89 | default: | |
90 | BUG(); | |
91 | } | |
92 | } | |
93 | #else | |
0f0257ea | 94 | static void svc_reclassify_socket(struct socket *sock) |
ed07536e PZ |
95 | { |
96 | } | |
97 | #endif | |
98 | ||
1da177e4 LT |
99 | /* |
100 | * Release an skbuff after use | |
101 | */ | |
5148bf4e | 102 | static void svc_release_skb(struct svc_rqst *rqstp) |
1da177e4 | 103 | { |
5148bf4e | 104 | struct sk_buff *skb = rqstp->rq_xprt_ctxt; |
1da177e4 LT |
105 | struct svc_deferred_req *dr = rqstp->rq_deferred; |
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 LT |
114 | } |
115 | if (dr) { | |
116 | rqstp->rq_deferred = NULL; | |
117 | kfree(dr); | |
118 | } | |
119 | } | |
120 | ||
b92503b2 CL |
121 | union svc_pktinfo_u { |
122 | struct in_pktinfo pkti; | |
b92503b2 | 123 | struct in6_pktinfo pkti6; |
b92503b2 | 124 | }; |
bc375ea7 DM |
125 | #define SVC_PKTINFO_SPACE \ |
126 | CMSG_SPACE(sizeof(union svc_pktinfo_u)) | |
b92503b2 CL |
127 | |
128 | static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh) | |
129 | { | |
57b1d3ba TT |
130 | struct svc_sock *svsk = |
131 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
132 | switch (svsk->sk_sk->sk_family) { | |
b92503b2 CL |
133 | case AF_INET: { |
134 | struct in_pktinfo *pki = CMSG_DATA(cmh); | |
135 | ||
136 | cmh->cmsg_level = SOL_IP; | |
137 | cmh->cmsg_type = IP_PKTINFO; | |
138 | pki->ipi_ifindex = 0; | |
139 | pki->ipi_spec_dst.s_addr = rqstp->rq_daddr.addr.s_addr; | |
140 | cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); | |
141 | } | |
142 | break; | |
5a05ed73 | 143 | |
b92503b2 CL |
144 | case AF_INET6: { |
145 | struct in6_pktinfo *pki = CMSG_DATA(cmh); | |
146 | ||
147 | cmh->cmsg_level = SOL_IPV6; | |
148 | cmh->cmsg_type = IPV6_PKTINFO; | |
149 | pki->ipi6_ifindex = 0; | |
150 | ipv6_addr_copy(&pki->ipi6_addr, | |
151 | &rqstp->rq_daddr.addr6); | |
152 | cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); | |
153 | } | |
154 | break; | |
b92503b2 CL |
155 | } |
156 | return; | |
157 | } | |
158 | ||
1da177e4 LT |
159 | /* |
160 | * Generic sendto routine | |
161 | */ | |
0f0257ea | 162 | static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr) |
1da177e4 | 163 | { |
57b1d3ba TT |
164 | struct svc_sock *svsk = |
165 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
1da177e4 LT |
166 | struct socket *sock = svsk->sk_sock; |
167 | int slen; | |
bc375ea7 DM |
168 | union { |
169 | struct cmsghdr hdr; | |
170 | long all[SVC_PKTINFO_SPACE / sizeof(long)]; | |
171 | } buffer; | |
172 | struct cmsghdr *cmh = &buffer.hdr; | |
1da177e4 LT |
173 | int len = 0; |
174 | int result; | |
175 | int size; | |
176 | struct page **ppage = xdr->pages; | |
177 | size_t base = xdr->page_base; | |
178 | unsigned int pglen = xdr->page_len; | |
179 | unsigned int flags = MSG_MORE; | |
5216a8e7 | 180 | RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); |
1da177e4 LT |
181 | |
182 | slen = xdr->len; | |
183 | ||
184 | if (rqstp->rq_prot == IPPROTO_UDP) { | |
b92503b2 CL |
185 | struct msghdr msg = { |
186 | .msg_name = &rqstp->rq_addr, | |
187 | .msg_namelen = rqstp->rq_addrlen, | |
188 | .msg_control = cmh, | |
189 | .msg_controllen = sizeof(buffer), | |
190 | .msg_flags = MSG_MORE, | |
191 | }; | |
192 | ||
193 | svc_set_cmsg_data(rqstp, cmh); | |
1da177e4 LT |
194 | |
195 | if (sock_sendmsg(sock, &msg, 0) < 0) | |
196 | goto out; | |
197 | } | |
198 | ||
199 | /* send head */ | |
200 | if (slen == xdr->head[0].iov_len) | |
201 | flags = 0; | |
44524359 N |
202 | len = kernel_sendpage(sock, rqstp->rq_respages[0], 0, |
203 | xdr->head[0].iov_len, flags); | |
1da177e4 LT |
204 | if (len != xdr->head[0].iov_len) |
205 | goto out; | |
206 | slen -= xdr->head[0].iov_len; | |
207 | if (slen == 0) | |
208 | goto out; | |
209 | ||
210 | /* send page data */ | |
211 | size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen; | |
212 | while (pglen > 0) { | |
213 | if (slen == size) | |
214 | flags = 0; | |
e6242e92 | 215 | result = kernel_sendpage(sock, *ppage, base, size, flags); |
1da177e4 LT |
216 | if (result > 0) |
217 | len += result; | |
218 | if (result != size) | |
219 | goto out; | |
220 | slen -= size; | |
221 | pglen -= size; | |
222 | size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen; | |
223 | base = 0; | |
224 | ppage++; | |
225 | } | |
226 | /* send tail */ | |
227 | if (xdr->tail[0].iov_len) { | |
44524359 N |
228 | result = kernel_sendpage(sock, rqstp->rq_respages[0], |
229 | ((unsigned long)xdr->tail[0].iov_base) | |
cca5172a | 230 | & (PAGE_SIZE-1), |
1da177e4 LT |
231 | xdr->tail[0].iov_len, 0); |
232 | ||
233 | if (result > 0) | |
234 | len += result; | |
235 | } | |
236 | out: | |
ad06e4bd | 237 | dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n", |
57b1d3ba | 238 | svsk, xdr->head[0].iov_base, xdr->head[0].iov_len, |
ad06e4bd | 239 | xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf))); |
1da177e4 LT |
240 | |
241 | return len; | |
242 | } | |
243 | ||
80212d59 N |
244 | /* |
245 | * Report socket names for nfsdfs | |
246 | */ | |
247 | static int one_sock_name(char *buf, struct svc_sock *svsk) | |
248 | { | |
249 | int len; | |
250 | ||
251 | switch(svsk->sk_sk->sk_family) { | |
252 | case AF_INET: | |
21454aaa HH |
253 | len = sprintf(buf, "ipv4 %s %pI4 %d\n", |
254 | svsk->sk_sk->sk_protocol == IPPROTO_UDP ? | |
80212d59 | 255 | "udp" : "tcp", |
21454aaa | 256 | &inet_sk(svsk->sk_sk)->rcv_saddr, |
80212d59 N |
257 | inet_sk(svsk->sk_sk)->num); |
258 | break; | |
259 | default: | |
260 | len = sprintf(buf, "*unknown-%d*\n", | |
261 | svsk->sk_sk->sk_family); | |
262 | } | |
263 | return len; | |
264 | } | |
265 | ||
266 | int | |
b41b66d6 | 267 | svc_sock_names(char *buf, struct svc_serv *serv, char *toclose) |
80212d59 | 268 | { |
b41b66d6 | 269 | struct svc_sock *svsk, *closesk = NULL; |
80212d59 N |
270 | int len = 0; |
271 | ||
272 | if (!serv) | |
273 | return 0; | |
aaf68cfb | 274 | spin_lock_bh(&serv->sv_lock); |
7a182083 | 275 | list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list) { |
80212d59 | 276 | int onelen = one_sock_name(buf+len, svsk); |
b41b66d6 N |
277 | if (toclose && strcmp(toclose, buf+len) == 0) |
278 | closesk = svsk; | |
279 | else | |
280 | len += onelen; | |
80212d59 | 281 | } |
aaf68cfb | 282 | spin_unlock_bh(&serv->sv_lock); |
b41b66d6 | 283 | if (closesk) |
5680c446 N |
284 | /* Should unregister with portmap, but you cannot |
285 | * unregister just one protocol... | |
286 | */ | |
7a182083 | 287 | svc_close_xprt(&closesk->sk_xprt); |
37a03472 N |
288 | else if (toclose) |
289 | return -ENOENT; | |
80212d59 N |
290 | return len; |
291 | } | |
292 | EXPORT_SYMBOL(svc_sock_names); | |
293 | ||
1da177e4 LT |
294 | /* |
295 | * Check input queue length | |
296 | */ | |
0f0257ea | 297 | static int svc_recv_available(struct svc_sock *svsk) |
1da177e4 | 298 | { |
1da177e4 LT |
299 | struct socket *sock = svsk->sk_sock; |
300 | int avail, err; | |
301 | ||
e6242e92 | 302 | err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail); |
1da177e4 LT |
303 | |
304 | return (err >= 0)? avail : err; | |
305 | } | |
306 | ||
307 | /* | |
308 | * Generic recvfrom routine. | |
309 | */ | |
0f0257ea TT |
310 | static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr, |
311 | int buflen) | |
1da177e4 | 312 | { |
57b1d3ba TT |
313 | struct svc_sock *svsk = |
314 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
1ba95105 CL |
315 | struct msghdr msg = { |
316 | .msg_flags = MSG_DONTWAIT, | |
317 | }; | |
318 | int len; | |
1da177e4 | 319 | |
260c1d12 TT |
320 | rqstp->rq_xprt_hlen = 0; |
321 | ||
1ba95105 CL |
322 | len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen, |
323 | msg.msg_flags); | |
1da177e4 | 324 | |
1da177e4 | 325 | dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n", |
1ba95105 | 326 | svsk, iov[0].iov_base, iov[0].iov_len, len); |
1da177e4 LT |
327 | return len; |
328 | } | |
329 | ||
330 | /* | |
331 | * Set socket snd and rcv buffer lengths | |
332 | */ | |
0f0257ea TT |
333 | static void svc_sock_setbufsize(struct socket *sock, unsigned int snd, |
334 | unsigned int rcv) | |
1da177e4 LT |
335 | { |
336 | #if 0 | |
337 | mm_segment_t oldfs; | |
338 | oldfs = get_fs(); set_fs(KERNEL_DS); | |
339 | sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF, | |
340 | (char*)&snd, sizeof(snd)); | |
341 | sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF, | |
342 | (char*)&rcv, sizeof(rcv)); | |
343 | #else | |
344 | /* sock_setsockopt limits use to sysctl_?mem_max, | |
345 | * which isn't acceptable. Until that is made conditional | |
346 | * on not having CAP_SYS_RESOURCE or similar, we go direct... | |
347 | * DaveM said I could! | |
348 | */ | |
349 | lock_sock(sock->sk); | |
350 | sock->sk->sk_sndbuf = snd * 2; | |
351 | sock->sk->sk_rcvbuf = rcv * 2; | |
352 | sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK|SOCK_RCVBUF_LOCK; | |
353 | release_sock(sock->sk); | |
354 | #endif | |
355 | } | |
356 | /* | |
357 | * INET callback when data has been received on the socket. | |
358 | */ | |
0f0257ea | 359 | static void svc_udp_data_ready(struct sock *sk, int count) |
1da177e4 | 360 | { |
939bb7ef | 361 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 | 362 | |
939bb7ef NB |
363 | if (svsk) { |
364 | dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n", | |
02fc6c36 TT |
365 | svsk, sk, count, |
366 | test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags)); | |
367 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); | |
f6150c3c | 368 | svc_xprt_enqueue(&svsk->sk_xprt); |
939bb7ef | 369 | } |
1da177e4 LT |
370 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
371 | wake_up_interruptible(sk->sk_sleep); | |
372 | } | |
373 | ||
374 | /* | |
375 | * INET callback when space is newly available on the socket. | |
376 | */ | |
0f0257ea | 377 | static void svc_write_space(struct sock *sk) |
1da177e4 LT |
378 | { |
379 | struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data); | |
380 | ||
381 | if (svsk) { | |
382 | dprintk("svc: socket %p(inet %p), write_space busy=%d\n", | |
02fc6c36 | 383 | svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags)); |
f6150c3c | 384 | svc_xprt_enqueue(&svsk->sk_xprt); |
1da177e4 LT |
385 | } |
386 | ||
387 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) { | |
939bb7ef | 388 | dprintk("RPC svc_write_space: someone sleeping on %p\n", |
1da177e4 LT |
389 | svsk); |
390 | wake_up_interruptible(sk->sk_sleep); | |
391 | } | |
392 | } | |
393 | ||
9dbc240f TT |
394 | /* |
395 | * Copy the UDP datagram's destination address to the rqstp structure. | |
396 | * The 'destination' address in this case is the address to which the | |
397 | * peer sent the datagram, i.e. our local address. For multihomed | |
398 | * hosts, this can change from msg to msg. Note that only the IP | |
399 | * address changes, the port number should remain the same. | |
400 | */ | |
401 | static void svc_udp_get_dest_address(struct svc_rqst *rqstp, | |
402 | struct cmsghdr *cmh) | |
95756482 | 403 | { |
57b1d3ba TT |
404 | struct svc_sock *svsk = |
405 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
406 | switch (svsk->sk_sk->sk_family) { | |
95756482 | 407 | case AF_INET: { |
7a37f578 N |
408 | struct in_pktinfo *pki = CMSG_DATA(cmh); |
409 | rqstp->rq_daddr.addr.s_addr = pki->ipi_spec_dst.s_addr; | |
95756482 | 410 | break; |
7a37f578 | 411 | } |
95756482 | 412 | case AF_INET6: { |
7a37f578 N |
413 | struct in6_pktinfo *pki = CMSG_DATA(cmh); |
414 | ipv6_addr_copy(&rqstp->rq_daddr.addr6, &pki->ipi6_addr); | |
95756482 | 415 | break; |
7a37f578 | 416 | } |
95756482 | 417 | } |
95756482 CL |
418 | } |
419 | ||
1da177e4 LT |
420 | /* |
421 | * Receive a datagram from a UDP socket. | |
422 | */ | |
0f0257ea | 423 | static int svc_udp_recvfrom(struct svc_rqst *rqstp) |
1da177e4 | 424 | { |
57b1d3ba TT |
425 | struct svc_sock *svsk = |
426 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
bb5cf160 | 427 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
1da177e4 | 428 | struct sk_buff *skb; |
bc375ea7 DM |
429 | union { |
430 | struct cmsghdr hdr; | |
431 | long all[SVC_PKTINFO_SPACE / sizeof(long)]; | |
432 | } buffer; | |
433 | struct cmsghdr *cmh = &buffer.hdr; | |
1da177e4 | 434 | int err, len; |
7a37f578 N |
435 | struct msghdr msg = { |
436 | .msg_name = svc_addr(rqstp), | |
437 | .msg_control = cmh, | |
438 | .msg_controllen = sizeof(buffer), | |
439 | .msg_flags = MSG_DONTWAIT, | |
440 | }; | |
1da177e4 | 441 | |
02fc6c36 | 442 | if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags)) |
1da177e4 LT |
443 | /* udp sockets need large rcvbuf as all pending |
444 | * requests are still in that buffer. sndbuf must | |
445 | * also be large enough that there is enough space | |
3262c816 GB |
446 | * for one reply per thread. We count all threads |
447 | * rather than threads in a particular pool, which | |
448 | * provides an upper bound on the number of threads | |
449 | * which will access the socket. | |
1da177e4 LT |
450 | */ |
451 | svc_sock_setbufsize(svsk->sk_sock, | |
c6b0a9f8 N |
452 | (serv->sv_nrthreads+3) * serv->sv_max_mesg, |
453 | (serv->sv_nrthreads+3) * serv->sv_max_mesg); | |
1da177e4 | 454 | |
02fc6c36 | 455 | clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
05ed690e N |
456 | skb = NULL; |
457 | err = kernel_recvmsg(svsk->sk_sock, &msg, NULL, | |
458 | 0, 0, MSG_PEEK | MSG_DONTWAIT); | |
459 | if (err >= 0) | |
460 | skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err); | |
461 | ||
462 | if (skb == NULL) { | |
463 | if (err != -EAGAIN) { | |
464 | /* possibly an icmp error */ | |
465 | dprintk("svc: recvfrom returned error %d\n", -err); | |
02fc6c36 | 466 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 467 | } |
a6046f71 | 468 | svc_xprt_received(&svsk->sk_xprt); |
05ed690e | 469 | return -EAGAIN; |
1da177e4 | 470 | } |
9dbc240f TT |
471 | len = svc_addr_len(svc_addr(rqstp)); |
472 | if (len < 0) | |
473 | return len; | |
474 | rqstp->rq_addrlen = len; | |
b7aa0bf7 ED |
475 | if (skb->tstamp.tv64 == 0) { |
476 | skb->tstamp = ktime_get_real(); | |
cca5172a | 477 | /* Don't enable netstamp, sunrpc doesn't |
1da177e4 LT |
478 | need that much accuracy */ |
479 | } | |
b7aa0bf7 | 480 | svsk->sk_sk->sk_stamp = skb->tstamp; |
02fc6c36 | 481 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */ |
1da177e4 LT |
482 | |
483 | /* | |
484 | * Maybe more packets - kick another thread ASAP. | |
485 | */ | |
a6046f71 | 486 | svc_xprt_received(&svsk->sk_xprt); |
1da177e4 LT |
487 | |
488 | len = skb->len - sizeof(struct udphdr); | |
489 | rqstp->rq_arg.len = len; | |
490 | ||
95756482 | 491 | rqstp->rq_prot = IPPROTO_UDP; |
27459f09 | 492 | |
7a37f578 N |
493 | if (cmh->cmsg_level != IPPROTO_IP || |
494 | cmh->cmsg_type != IP_PKTINFO) { | |
495 | if (net_ratelimit()) | |
496 | printk("rpcsvc: received unknown control message:" | |
497 | "%d/%d\n", | |
498 | cmh->cmsg_level, cmh->cmsg_type); | |
499 | skb_free_datagram(svsk->sk_sk, skb); | |
500 | return 0; | |
501 | } | |
502 | svc_udp_get_dest_address(rqstp, cmh); | |
1da177e4 LT |
503 | |
504 | if (skb_is_nonlinear(skb)) { | |
505 | /* we have to copy */ | |
506 | local_bh_disable(); | |
507 | if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) { | |
508 | local_bh_enable(); | |
509 | /* checksum error */ | |
510 | skb_free_datagram(svsk->sk_sk, skb); | |
511 | return 0; | |
512 | } | |
513 | local_bh_enable(); | |
cca5172a | 514 | skb_free_datagram(svsk->sk_sk, skb); |
1da177e4 LT |
515 | } else { |
516 | /* we can use it in-place */ | |
0f0257ea TT |
517 | rqstp->rq_arg.head[0].iov_base = skb->data + |
518 | sizeof(struct udphdr); | |
1da177e4 | 519 | rqstp->rq_arg.head[0].iov_len = len; |
fb286bb2 HX |
520 | if (skb_checksum_complete(skb)) { |
521 | skb_free_datagram(svsk->sk_sk, skb); | |
522 | return 0; | |
1da177e4 | 523 | } |
5148bf4e | 524 | rqstp->rq_xprt_ctxt = skb; |
1da177e4 LT |
525 | } |
526 | ||
527 | rqstp->rq_arg.page_base = 0; | |
528 | if (len <= rqstp->rq_arg.head[0].iov_len) { | |
529 | rqstp->rq_arg.head[0].iov_len = len; | |
530 | rqstp->rq_arg.page_len = 0; | |
44524359 | 531 | rqstp->rq_respages = rqstp->rq_pages+1; |
1da177e4 LT |
532 | } else { |
533 | rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len; | |
44524359 | 534 | rqstp->rq_respages = rqstp->rq_pages + 1 + |
172589cc | 535 | DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE); |
1da177e4 LT |
536 | } |
537 | ||
538 | if (serv->sv_stats) | |
539 | serv->sv_stats->netudpcnt++; | |
540 | ||
541 | return len; | |
542 | } | |
543 | ||
544 | static int | |
545 | svc_udp_sendto(struct svc_rqst *rqstp) | |
546 | { | |
547 | int error; | |
548 | ||
549 | error = svc_sendto(rqstp, &rqstp->rq_res); | |
550 | if (error == -ECONNREFUSED) | |
551 | /* ICMP error on earlier request. */ | |
552 | error = svc_sendto(rqstp, &rqstp->rq_res); | |
553 | ||
554 | return error; | |
555 | } | |
556 | ||
e831fe65 TT |
557 | static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp) |
558 | { | |
559 | } | |
560 | ||
323bee32 TT |
561 | static int svc_udp_has_wspace(struct svc_xprt *xprt) |
562 | { | |
563 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
bb5cf160 | 564 | struct svc_serv *serv = xprt->xpt_server; |
323bee32 TT |
565 | unsigned long required; |
566 | ||
567 | /* | |
568 | * Set the SOCK_NOSPACE flag before checking the available | |
569 | * sock space. | |
570 | */ | |
571 | set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
7a90e8cc | 572 | required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg; |
323bee32 TT |
573 | if (required*2 > sock_wspace(svsk->sk_sk)) |
574 | return 0; | |
575 | clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
576 | return 1; | |
577 | } | |
578 | ||
38a417cc TT |
579 | static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt) |
580 | { | |
581 | BUG(); | |
582 | return NULL; | |
583 | } | |
584 | ||
b700cbb1 TT |
585 | static struct svc_xprt *svc_udp_create(struct svc_serv *serv, |
586 | struct sockaddr *sa, int salen, | |
587 | int flags) | |
588 | { | |
589 | return svc_create_socket(serv, IPPROTO_UDP, sa, salen, flags); | |
590 | } | |
591 | ||
360d8738 | 592 | static struct svc_xprt_ops svc_udp_ops = { |
b700cbb1 | 593 | .xpo_create = svc_udp_create, |
5d137990 TT |
594 | .xpo_recvfrom = svc_udp_recvfrom, |
595 | .xpo_sendto = svc_udp_sendto, | |
5148bf4e | 596 | .xpo_release_rqst = svc_release_skb, |
755cceab TT |
597 | .xpo_detach = svc_sock_detach, |
598 | .xpo_free = svc_sock_free, | |
e831fe65 | 599 | .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr, |
323bee32 | 600 | .xpo_has_wspace = svc_udp_has_wspace, |
38a417cc | 601 | .xpo_accept = svc_udp_accept, |
360d8738 TT |
602 | }; |
603 | ||
604 | static struct svc_xprt_class svc_udp_class = { | |
605 | .xcl_name = "udp", | |
b700cbb1 | 606 | .xcl_owner = THIS_MODULE, |
360d8738 | 607 | .xcl_ops = &svc_udp_ops, |
49023155 | 608 | .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP, |
360d8738 TT |
609 | }; |
610 | ||
bb5cf160 | 611 | static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv) |
1da177e4 | 612 | { |
7a37f578 N |
613 | int one = 1; |
614 | mm_segment_t oldfs; | |
615 | ||
bb5cf160 | 616 | svc_xprt_init(&svc_udp_class, &svsk->sk_xprt, serv); |
def13d74 | 617 | clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
618 | svsk->sk_sk->sk_data_ready = svc_udp_data_ready; |
619 | svsk->sk_sk->sk_write_space = svc_write_space; | |
1da177e4 LT |
620 | |
621 | /* initialise setting must have enough space to | |
cca5172a | 622 | * receive and respond to one request. |
1da177e4 LT |
623 | * svc_udp_recvfrom will re-adjust if necessary |
624 | */ | |
625 | svc_sock_setbufsize(svsk->sk_sock, | |
bb5cf160 TT |
626 | 3 * svsk->sk_xprt.xpt_server->sv_max_mesg, |
627 | 3 * svsk->sk_xprt.xpt_server->sv_max_mesg); | |
1da177e4 | 628 | |
0f0257ea TT |
629 | /* data might have come in before data_ready set up */ |
630 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); | |
02fc6c36 | 631 | set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); |
7a37f578 N |
632 | |
633 | oldfs = get_fs(); | |
634 | set_fs(KERNEL_DS); | |
635 | /* make sure we get destination address info */ | |
636 | svsk->sk_sock->ops->setsockopt(svsk->sk_sock, IPPROTO_IP, IP_PKTINFO, | |
637 | (char __user *)&one, sizeof(one)); | |
638 | set_fs(oldfs); | |
1da177e4 LT |
639 | } |
640 | ||
641 | /* | |
642 | * A data_ready event on a listening socket means there's a connection | |
643 | * pending. Do not use state_change as a substitute for it. | |
644 | */ | |
0f0257ea | 645 | static void svc_tcp_listen_data_ready(struct sock *sk, int count_unused) |
1da177e4 | 646 | { |
939bb7ef | 647 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 LT |
648 | |
649 | dprintk("svc: socket %p TCP (listen) state change %d\n", | |
939bb7ef | 650 | sk, sk->sk_state); |
1da177e4 | 651 | |
939bb7ef NB |
652 | /* |
653 | * This callback may called twice when a new connection | |
654 | * is established as a child socket inherits everything | |
655 | * from a parent LISTEN socket. | |
656 | * 1) data_ready method of the parent socket will be called | |
657 | * when one of child sockets become ESTABLISHED. | |
658 | * 2) data_ready method of the child socket may be called | |
659 | * when it receives data before the socket is accepted. | |
660 | * In case of 2, we should ignore it silently. | |
661 | */ | |
662 | if (sk->sk_state == TCP_LISTEN) { | |
663 | if (svsk) { | |
02fc6c36 | 664 | set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
f6150c3c | 665 | svc_xprt_enqueue(&svsk->sk_xprt); |
939bb7ef NB |
666 | } else |
667 | printk("svc: socket %p: no user data\n", sk); | |
1da177e4 | 668 | } |
939bb7ef | 669 | |
1da177e4 LT |
670 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
671 | wake_up_interruptible_all(sk->sk_sleep); | |
672 | } | |
673 | ||
674 | /* | |
675 | * A state change on a connected socket means it's dying or dead. | |
676 | */ | |
0f0257ea | 677 | static void svc_tcp_state_change(struct sock *sk) |
1da177e4 | 678 | { |
939bb7ef | 679 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 LT |
680 | |
681 | dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n", | |
939bb7ef | 682 | sk, sk->sk_state, sk->sk_user_data); |
1da177e4 | 683 | |
939bb7ef | 684 | if (!svsk) |
1da177e4 | 685 | printk("svc: socket %p: no user data\n", sk); |
939bb7ef | 686 | else { |
02fc6c36 | 687 | set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); |
f6150c3c | 688 | svc_xprt_enqueue(&svsk->sk_xprt); |
1da177e4 | 689 | } |
1da177e4 LT |
690 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
691 | wake_up_interruptible_all(sk->sk_sleep); | |
692 | } | |
693 | ||
0f0257ea | 694 | static void svc_tcp_data_ready(struct sock *sk, int count) |
1da177e4 | 695 | { |
939bb7ef | 696 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 LT |
697 | |
698 | dprintk("svc: socket %p TCP data ready (svsk %p)\n", | |
939bb7ef NB |
699 | sk, sk->sk_user_data); |
700 | if (svsk) { | |
02fc6c36 | 701 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
f6150c3c | 702 | svc_xprt_enqueue(&svsk->sk_xprt); |
939bb7ef | 703 | } |
1da177e4 LT |
704 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
705 | wake_up_interruptible(sk->sk_sleep); | |
706 | } | |
707 | ||
708 | /* | |
709 | * Accept a TCP connection | |
710 | */ | |
38a417cc | 711 | static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt) |
1da177e4 | 712 | { |
38a417cc | 713 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); |
cdd88b9f AM |
714 | struct sockaddr_storage addr; |
715 | struct sockaddr *sin = (struct sockaddr *) &addr; | |
bb5cf160 | 716 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
1da177e4 LT |
717 | struct socket *sock = svsk->sk_sock; |
718 | struct socket *newsock; | |
1da177e4 LT |
719 | struct svc_sock *newsvsk; |
720 | int err, slen; | |
5216a8e7 | 721 | RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); |
1da177e4 LT |
722 | |
723 | dprintk("svc: tcp_accept %p sock %p\n", svsk, sock); | |
724 | if (!sock) | |
38a417cc | 725 | return NULL; |
1da177e4 | 726 | |
02fc6c36 | 727 | clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
e6242e92 SS |
728 | err = kernel_accept(sock, &newsock, O_NONBLOCK); |
729 | if (err < 0) { | |
1da177e4 LT |
730 | if (err == -ENOMEM) |
731 | printk(KERN_WARNING "%s: no more sockets!\n", | |
732 | serv->sv_name); | |
e6242e92 | 733 | else if (err != -EAGAIN && net_ratelimit()) |
1da177e4 LT |
734 | printk(KERN_WARNING "%s: accept failed (err %d)!\n", |
735 | serv->sv_name, -err); | |
38a417cc | 736 | return NULL; |
1da177e4 | 737 | } |
02fc6c36 | 738 | set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 739 | |
cdd88b9f | 740 | err = kernel_getpeername(newsock, sin, &slen); |
1da177e4 LT |
741 | if (err < 0) { |
742 | if (net_ratelimit()) | |
743 | printk(KERN_WARNING "%s: peername failed (err %d)!\n", | |
744 | serv->sv_name, -err); | |
745 | goto failed; /* aborted connection or whatever */ | |
746 | } | |
747 | ||
748 | /* Ideally, we would want to reject connections from unauthorized | |
ad06e4bd CL |
749 | * hosts here, but when we get encryption, the IP of the host won't |
750 | * tell us anything. For now just warn about unpriv connections. | |
1da177e4 | 751 | */ |
cdd88b9f | 752 | if (!svc_port_is_privileged(sin)) { |
1da177e4 | 753 | dprintk(KERN_WARNING |
ad06e4bd | 754 | "%s: connect from unprivileged port: %s\n", |
cca5172a | 755 | serv->sv_name, |
cdd88b9f | 756 | __svc_print_addr(sin, buf, sizeof(buf))); |
1da177e4 | 757 | } |
ad06e4bd | 758 | dprintk("%s: connect from %s\n", serv->sv_name, |
cdd88b9f | 759 | __svc_print_addr(sin, buf, sizeof(buf))); |
1da177e4 LT |
760 | |
761 | /* make sure that a write doesn't block forever when | |
762 | * low on memory | |
763 | */ | |
764 | newsock->sk->sk_sndtimeo = HZ*30; | |
765 | ||
6b174337 CL |
766 | if (!(newsvsk = svc_setup_socket(serv, newsock, &err, |
767 | (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY)))) | |
1da177e4 | 768 | goto failed; |
9dbc240f | 769 | svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen); |
a9747692 FM |
770 | err = kernel_getsockname(newsock, sin, &slen); |
771 | if (unlikely(err < 0)) { | |
772 | dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err); | |
773 | slen = offsetof(struct sockaddr, sa_data); | |
774 | } | |
9dbc240f | 775 | svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen); |
067d7817 | 776 | |
f9f3cc4f TT |
777 | if (serv->sv_stats) |
778 | serv->sv_stats->nettcpconn++; | |
779 | ||
780 | return &newsvsk->sk_xprt; | |
781 | ||
782 | failed: | |
783 | sock_release(newsock); | |
784 | return NULL; | |
785 | } | |
786 | ||
1da177e4 LT |
787 | /* |
788 | * Receive data from a TCP socket. | |
789 | */ | |
0f0257ea | 790 | static int svc_tcp_recvfrom(struct svc_rqst *rqstp) |
1da177e4 | 791 | { |
57b1d3ba TT |
792 | struct svc_sock *svsk = |
793 | container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); | |
bb5cf160 | 794 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
1da177e4 | 795 | int len; |
3cc03b16 | 796 | struct kvec *vec; |
1da177e4 LT |
797 | int pnum, vlen; |
798 | ||
799 | dprintk("svc: tcp_recv %p data %d conn %d close %d\n", | |
02fc6c36 TT |
800 | svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags), |
801 | test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags), | |
802 | test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags)); | |
1da177e4 | 803 | |
02fc6c36 | 804 | if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags)) |
1da177e4 LT |
805 | /* sndbuf needs to have room for one request |
806 | * per thread, otherwise we can stall even when the | |
807 | * network isn't a bottleneck. | |
3262c816 GB |
808 | * |
809 | * We count all threads rather than threads in a | |
810 | * particular pool, which provides an upper bound | |
811 | * on the number of threads which will access the socket. | |
812 | * | |
1da177e4 | 813 | * rcvbuf just needs to be able to hold a few requests. |
cca5172a | 814 | * Normally they will be removed from the queue |
1da177e4 LT |
815 | * as soon a a complete request arrives. |
816 | */ | |
817 | svc_sock_setbufsize(svsk->sk_sock, | |
c6b0a9f8 N |
818 | (serv->sv_nrthreads+3) * serv->sv_max_mesg, |
819 | 3 * serv->sv_max_mesg); | |
1da177e4 | 820 | |
02fc6c36 | 821 | clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
822 | |
823 | /* Receive data. If we haven't got the record length yet, get | |
824 | * the next four bytes. Otherwise try to gobble up as much as | |
825 | * possible up to the complete record length. | |
826 | */ | |
c0401ea0 CL |
827 | if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) { |
828 | int want = sizeof(rpc_fraghdr) - svsk->sk_tcplen; | |
1da177e4 LT |
829 | struct kvec iov; |
830 | ||
831 | iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen; | |
832 | iov.iov_len = want; | |
833 | if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0) | |
834 | goto error; | |
835 | svsk->sk_tcplen += len; | |
836 | ||
837 | if (len < want) { | |
c0401ea0 CL |
838 | dprintk("svc: short recvfrom while reading record " |
839 | "length (%d of %d)\n", len, want); | |
a6046f71 | 840 | svc_xprt_received(&svsk->sk_xprt); |
1da177e4 LT |
841 | return -EAGAIN; /* record header not complete */ |
842 | } | |
843 | ||
844 | svsk->sk_reclen = ntohl(svsk->sk_reclen); | |
c0401ea0 | 845 | if (!(svsk->sk_reclen & RPC_LAST_STREAM_FRAGMENT)) { |
1da177e4 LT |
846 | /* FIXME: technically, a record can be fragmented, |
847 | * and non-terminal fragments will not have the top | |
848 | * bit set in the fragment length header. | |
849 | * But apparently no known nfs clients send fragmented | |
850 | * records. */ | |
34e9a63b | 851 | if (net_ratelimit()) |
c0401ea0 CL |
852 | printk(KERN_NOTICE "RPC: multiple fragments " |
853 | "per record not supported\n"); | |
1da177e4 LT |
854 | goto err_delete; |
855 | } | |
c0401ea0 | 856 | svsk->sk_reclen &= RPC_FRAGMENT_SIZE_MASK; |
1da177e4 | 857 | dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen); |
c6b0a9f8 | 858 | if (svsk->sk_reclen > serv->sv_max_mesg) { |
34e9a63b | 859 | if (net_ratelimit()) |
c0401ea0 CL |
860 | printk(KERN_NOTICE "RPC: " |
861 | "fragment too large: 0x%08lx\n", | |
862 | (unsigned long)svsk->sk_reclen); | |
1da177e4 LT |
863 | goto err_delete; |
864 | } | |
865 | } | |
866 | ||
867 | /* Check whether enough data is available */ | |
868 | len = svc_recv_available(svsk); | |
869 | if (len < 0) | |
870 | goto error; | |
871 | ||
872 | if (len < svsk->sk_reclen) { | |
873 | dprintk("svc: incomplete TCP record (%d of %d)\n", | |
874 | len, svsk->sk_reclen); | |
a6046f71 | 875 | svc_xprt_received(&svsk->sk_xprt); |
1da177e4 LT |
876 | return -EAGAIN; /* record not complete */ |
877 | } | |
878 | len = svsk->sk_reclen; | |
02fc6c36 | 879 | set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); |
1da177e4 | 880 | |
3cc03b16 | 881 | vec = rqstp->rq_vec; |
1da177e4 LT |
882 | vec[0] = rqstp->rq_arg.head[0]; |
883 | vlen = PAGE_SIZE; | |
884 | pnum = 1; | |
885 | while (vlen < len) { | |
44524359 | 886 | vec[pnum].iov_base = page_address(rqstp->rq_pages[pnum]); |
1da177e4 LT |
887 | vec[pnum].iov_len = PAGE_SIZE; |
888 | pnum++; | |
889 | vlen += PAGE_SIZE; | |
890 | } | |
44524359 | 891 | rqstp->rq_respages = &rqstp->rq_pages[pnum]; |
1da177e4 LT |
892 | |
893 | /* Now receive data */ | |
894 | len = svc_recvfrom(rqstp, vec, pnum, len); | |
895 | if (len < 0) | |
896 | goto error; | |
897 | ||
898 | dprintk("svc: TCP complete record (%d bytes)\n", len); | |
899 | rqstp->rq_arg.len = len; | |
900 | rqstp->rq_arg.page_base = 0; | |
901 | if (len <= rqstp->rq_arg.head[0].iov_len) { | |
902 | rqstp->rq_arg.head[0].iov_len = len; | |
903 | rqstp->rq_arg.page_len = 0; | |
904 | } else { | |
905 | rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len; | |
906 | } | |
907 | ||
5148bf4e | 908 | rqstp->rq_xprt_ctxt = NULL; |
1da177e4 LT |
909 | rqstp->rq_prot = IPPROTO_TCP; |
910 | ||
911 | /* Reset TCP read info */ | |
912 | svsk->sk_reclen = 0; | |
913 | svsk->sk_tcplen = 0; | |
914 | ||
9dbc240f | 915 | svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt); |
a6046f71 | 916 | svc_xprt_received(&svsk->sk_xprt); |
1da177e4 LT |
917 | if (serv->sv_stats) |
918 | serv->sv_stats->nettcpcnt++; | |
919 | ||
920 | return len; | |
921 | ||
922 | err_delete: | |
02fc6c36 | 923 | set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); |
1da177e4 LT |
924 | return -EAGAIN; |
925 | ||
926 | error: | |
927 | if (len == -EAGAIN) { | |
928 | dprintk("RPC: TCP recvfrom got EAGAIN\n"); | |
a6046f71 | 929 | svc_xprt_received(&svsk->sk_xprt); |
1da177e4 LT |
930 | } else { |
931 | printk(KERN_NOTICE "%s: recvfrom returned errno %d\n", | |
bb5cf160 | 932 | svsk->sk_xprt.xpt_server->sv_name, -len); |
93fbf1a5 | 933 | goto err_delete; |
1da177e4 LT |
934 | } |
935 | ||
936 | return len; | |
937 | } | |
938 | ||
939 | /* | |
940 | * Send out data on TCP socket. | |
941 | */ | |
0f0257ea | 942 | static int svc_tcp_sendto(struct svc_rqst *rqstp) |
1da177e4 LT |
943 | { |
944 | struct xdr_buf *xbufp = &rqstp->rq_res; | |
945 | int sent; | |
d8ed029d | 946 | __be32 reclen; |
1da177e4 LT |
947 | |
948 | /* Set up the first element of the reply kvec. | |
949 | * Any other kvecs that may be in use have been taken | |
950 | * care of by the server implementation itself. | |
951 | */ | |
952 | reclen = htonl(0x80000000|((xbufp->len ) - 4)); | |
953 | memcpy(xbufp->head[0].iov_base, &reclen, 4); | |
954 | ||
57b1d3ba | 955 | if (test_bit(XPT_DEAD, &rqstp->rq_xprt->xpt_flags)) |
1da177e4 LT |
956 | return -ENOTCONN; |
957 | ||
958 | sent = svc_sendto(rqstp, &rqstp->rq_res); | |
959 | if (sent != xbufp->len) { | |
0f0257ea TT |
960 | printk(KERN_NOTICE |
961 | "rpc-srv/tcp: %s: %s %d when sending %d bytes " | |
962 | "- shutting down socket\n", | |
57b1d3ba | 963 | rqstp->rq_xprt->xpt_server->sv_name, |
1da177e4 LT |
964 | (sent<0)?"got error":"sent only", |
965 | sent, xbufp->len); | |
57b1d3ba | 966 | set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags); |
f6150c3c | 967 | svc_xprt_enqueue(rqstp->rq_xprt); |
1da177e4 LT |
968 | sent = -EAGAIN; |
969 | } | |
970 | return sent; | |
971 | } | |
972 | ||
e831fe65 TT |
973 | /* |
974 | * Setup response header. TCP has a 4B record length field. | |
975 | */ | |
976 | static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp) | |
977 | { | |
978 | struct kvec *resv = &rqstp->rq_res.head[0]; | |
979 | ||
980 | /* tcp needs a space for the record length... */ | |
981 | svc_putnl(resv, 0); | |
982 | } | |
983 | ||
323bee32 TT |
984 | static int svc_tcp_has_wspace(struct svc_xprt *xprt) |
985 | { | |
57b1d3ba | 986 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); |
bb5cf160 | 987 | struct svc_serv *serv = svsk->sk_xprt.xpt_server; |
323bee32 TT |
988 | int required; |
989 | int wspace; | |
990 | ||
991 | /* | |
992 | * Set the SOCK_NOSPACE flag before checking the available | |
993 | * sock space. | |
994 | */ | |
995 | set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
7a90e8cc | 996 | required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg; |
323bee32 TT |
997 | wspace = sk_stream_wspace(svsk->sk_sk); |
998 | ||
999 | if (wspace < sk_stream_min_wspace(svsk->sk_sk)) | |
1000 | return 0; | |
1001 | if (required * 2 > wspace) | |
1002 | return 0; | |
1003 | ||
1004 | clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
1005 | return 1; | |
1006 | } | |
1007 | ||
b700cbb1 TT |
1008 | static struct svc_xprt *svc_tcp_create(struct svc_serv *serv, |
1009 | struct sockaddr *sa, int salen, | |
1010 | int flags) | |
1011 | { | |
1012 | return svc_create_socket(serv, IPPROTO_TCP, sa, salen, flags); | |
1013 | } | |
1014 | ||
360d8738 | 1015 | static struct svc_xprt_ops svc_tcp_ops = { |
b700cbb1 | 1016 | .xpo_create = svc_tcp_create, |
5d137990 TT |
1017 | .xpo_recvfrom = svc_tcp_recvfrom, |
1018 | .xpo_sendto = svc_tcp_sendto, | |
5148bf4e | 1019 | .xpo_release_rqst = svc_release_skb, |
755cceab TT |
1020 | .xpo_detach = svc_sock_detach, |
1021 | .xpo_free = svc_sock_free, | |
e831fe65 | 1022 | .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr, |
323bee32 | 1023 | .xpo_has_wspace = svc_tcp_has_wspace, |
38a417cc | 1024 | .xpo_accept = svc_tcp_accept, |
360d8738 TT |
1025 | }; |
1026 | ||
1027 | static struct svc_xprt_class svc_tcp_class = { | |
1028 | .xcl_name = "tcp", | |
b700cbb1 | 1029 | .xcl_owner = THIS_MODULE, |
360d8738 | 1030 | .xcl_ops = &svc_tcp_ops, |
49023155 | 1031 | .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP, |
360d8738 TT |
1032 | }; |
1033 | ||
1034 | void svc_init_xprt_sock(void) | |
1035 | { | |
1036 | svc_reg_xprt_class(&svc_tcp_class); | |
1037 | svc_reg_xprt_class(&svc_udp_class); | |
1038 | } | |
1039 | ||
1040 | void svc_cleanup_xprt_sock(void) | |
1041 | { | |
1042 | svc_unreg_xprt_class(&svc_tcp_class); | |
1043 | svc_unreg_xprt_class(&svc_udp_class); | |
1044 | } | |
1045 | ||
bb5cf160 | 1046 | static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv) |
1da177e4 LT |
1047 | { |
1048 | struct sock *sk = svsk->sk_sk; | |
1da177e4 | 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 | ||
b7872fe8 | 1066 | tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF; |
1da177e4 LT |
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 | } | |
23d42ee2 | 1104 | EXPORT_SYMBOL(svc_sock_update_bufs); |
1da177e4 | 1105 | |
1da177e4 LT |
1106 | /* |
1107 | * Initialize socket for RPC use and create svc_sock struct | |
1108 | * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF. | |
1109 | */ | |
6b174337 CL |
1110 | static struct svc_sock *svc_setup_socket(struct svc_serv *serv, |
1111 | struct socket *sock, | |
1112 | int *errp, int flags) | |
1da177e4 LT |
1113 | { |
1114 | struct svc_sock *svsk; | |
1115 | struct sock *inet; | |
6b174337 | 1116 | int pmap_register = !(flags & SVC_SOCK_ANONYMOUS); |
b6632339 | 1117 | int val; |
1da177e4 LT |
1118 | |
1119 | dprintk("svc: svc_setup_socket %p\n", sock); | |
0da974f4 | 1120 | if (!(svsk = kzalloc(sizeof(*svsk), GFP_KERNEL))) { |
1da177e4 LT |
1121 | *errp = -ENOMEM; |
1122 | return NULL; | |
1123 | } | |
1da177e4 LT |
1124 | |
1125 | inet = sock->sk; | |
1126 | ||
1127 | /* Register socket with portmapper */ | |
1128 | if (*errp >= 0 && pmap_register) | |
1129 | *errp = svc_register(serv, inet->sk_protocol, | |
1130 | ntohs(inet_sk(inet)->sport)); | |
1131 | ||
1132 | if (*errp < 0) { | |
1133 | kfree(svsk); | |
1134 | return NULL; | |
1135 | } | |
1136 | ||
1da177e4 LT |
1137 | inet->sk_user_data = svsk; |
1138 | svsk->sk_sock = sock; | |
1139 | svsk->sk_sk = inet; | |
1140 | svsk->sk_ostate = inet->sk_state_change; | |
1141 | svsk->sk_odata = inet->sk_data_ready; | |
1142 | svsk->sk_owspace = inet->sk_write_space; | |
1da177e4 LT |
1143 | |
1144 | /* Initialize the socket */ | |
1145 | if (sock->type == SOCK_DGRAM) | |
bb5cf160 | 1146 | svc_udp_init(svsk, serv); |
1da177e4 | 1147 | else |
bb5cf160 | 1148 | svc_tcp_init(svsk, serv); |
1da177e4 | 1149 | |
b6632339 CL |
1150 | /* |
1151 | * We start one listener per sv_serv. We want AF_INET | |
1152 | * requests to be automatically shunted to our AF_INET6 | |
1153 | * listener using a mapped IPv4 address. Make sure | |
1154 | * no-one starts an equivalent IPv4 listener, which | |
1155 | * would steal our incoming connections. | |
1156 | */ | |
1157 | val = 0; | |
1158 | if (serv->sv_family == AF_INET6) | |
1159 | kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY, | |
1160 | (char *)&val, sizeof(val)); | |
1161 | ||
1da177e4 LT |
1162 | dprintk("svc: svc_setup_socket created %p (inet %p)\n", |
1163 | svsk, svsk->sk_sk); | |
1164 | ||
1da177e4 LT |
1165 | return svsk; |
1166 | } | |
1167 | ||
b41b66d6 N |
1168 | int svc_addsock(struct svc_serv *serv, |
1169 | int fd, | |
29373913 | 1170 | char *name_return) |
b41b66d6 N |
1171 | { |
1172 | int err = 0; | |
1173 | struct socket *so = sockfd_lookup(fd, &err); | |
1174 | struct svc_sock *svsk = NULL; | |
1175 | ||
1176 | if (!so) | |
1177 | return err; | |
1178 | if (so->sk->sk_family != AF_INET) | |
1179 | err = -EAFNOSUPPORT; | |
1180 | else if (so->sk->sk_protocol != IPPROTO_TCP && | |
1181 | so->sk->sk_protocol != IPPROTO_UDP) | |
1182 | err = -EPROTONOSUPPORT; | |
1183 | else if (so->state > SS_UNCONNECTED) | |
1184 | err = -EISCONN; | |
1185 | else { | |
2da2c21d TT |
1186 | if (!try_module_get(THIS_MODULE)) |
1187 | err = -ENOENT; | |
1188 | else | |
1189 | svsk = svc_setup_socket(serv, so, &err, | |
1190 | SVC_SOCK_DEFAULTS); | |
e79eff1f | 1191 | if (svsk) { |
9dbc240f TT |
1192 | struct sockaddr_storage addr; |
1193 | struct sockaddr *sin = (struct sockaddr *)&addr; | |
1194 | int salen; | |
1195 | if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0) | |
1196 | svc_xprt_set_local(&svsk->sk_xprt, sin, salen); | |
4e5caaa5 TT |
1197 | clear_bit(XPT_TEMP, &svsk->sk_xprt.xpt_flags); |
1198 | spin_lock_bh(&serv->sv_lock); | |
1199 | list_add(&svsk->sk_xprt.xpt_list, &serv->sv_permsocks); | |
1200 | spin_unlock_bh(&serv->sv_lock); | |
a6046f71 | 1201 | svc_xprt_received(&svsk->sk_xprt); |
b41b66d6 | 1202 | err = 0; |
2da2c21d TT |
1203 | } else |
1204 | module_put(THIS_MODULE); | |
b41b66d6 N |
1205 | } |
1206 | if (err) { | |
1207 | sockfd_put(so); | |
1208 | return err; | |
1209 | } | |
b41b66d6 N |
1210 | return one_sock_name(name_return, svsk); |
1211 | } | |
1212 | EXPORT_SYMBOL_GPL(svc_addsock); | |
1213 | ||
1da177e4 LT |
1214 | /* |
1215 | * Create socket for RPC service. | |
1216 | */ | |
b700cbb1 TT |
1217 | static struct svc_xprt *svc_create_socket(struct svc_serv *serv, |
1218 | int protocol, | |
1219 | struct sockaddr *sin, int len, | |
1220 | int flags) | |
1da177e4 LT |
1221 | { |
1222 | struct svc_sock *svsk; | |
1223 | struct socket *sock; | |
1224 | int error; | |
1225 | int type; | |
9dbc240f TT |
1226 | struct sockaddr_storage addr; |
1227 | struct sockaddr *newsin = (struct sockaddr *)&addr; | |
1228 | int newlen; | |
5216a8e7 | 1229 | RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); |
1da177e4 | 1230 | |
ad06e4bd CL |
1231 | dprintk("svc: svc_create_socket(%s, %d, %s)\n", |
1232 | serv->sv_program->pg_name, protocol, | |
77f1f67a | 1233 | __svc_print_addr(sin, buf, sizeof(buf))); |
1da177e4 LT |
1234 | |
1235 | if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) { | |
1236 | printk(KERN_WARNING "svc: only UDP and TCP " | |
1237 | "sockets supported\n"); | |
b700cbb1 | 1238 | return ERR_PTR(-EINVAL); |
1da177e4 LT |
1239 | } |
1240 | type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM; | |
1241 | ||
77f1f67a CL |
1242 | error = sock_create_kern(sin->sa_family, type, protocol, &sock); |
1243 | if (error < 0) | |
b700cbb1 | 1244 | return ERR_PTR(error); |
1da177e4 | 1245 | |
ed07536e PZ |
1246 | svc_reclassify_socket(sock); |
1247 | ||
18114746 | 1248 | if (type == SOCK_STREAM) |
77f1f67a CL |
1249 | sock->sk->sk_reuse = 1; /* allow address reuse */ |
1250 | error = kernel_bind(sock, sin, len); | |
18114746 ES |
1251 | if (error < 0) |
1252 | goto bummer; | |
1da177e4 | 1253 | |
9dbc240f TT |
1254 | newlen = len; |
1255 | error = kernel_getsockname(sock, newsin, &newlen); | |
1256 | if (error < 0) | |
1257 | goto bummer; | |
1258 | ||
1da177e4 | 1259 | if (protocol == IPPROTO_TCP) { |
e6242e92 | 1260 | if ((error = kernel_listen(sock, 64)) < 0) |
1da177e4 LT |
1261 | goto bummer; |
1262 | } | |
1263 | ||
e79eff1f | 1264 | if ((svsk = svc_setup_socket(serv, sock, &error, flags)) != NULL) { |
9dbc240f | 1265 | svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen); |
b700cbb1 | 1266 | return (struct svc_xprt *)svsk; |
e79eff1f | 1267 | } |
1da177e4 LT |
1268 | |
1269 | bummer: | |
1270 | dprintk("svc: svc_create_socket error = %d\n", -error); | |
1271 | sock_release(sock); | |
b700cbb1 | 1272 | return ERR_PTR(error); |
1da177e4 LT |
1273 | } |
1274 | ||
755cceab TT |
1275 | /* |
1276 | * Detach the svc_sock from the socket so that no | |
1277 | * more callbacks occur. | |
1278 | */ | |
1279 | static void svc_sock_detach(struct svc_xprt *xprt) | |
1280 | { | |
1281 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
1282 | struct sock *sk = svsk->sk_sk; | |
1283 | ||
1284 | dprintk("svc: svc_sock_detach(%p)\n", svsk); | |
1285 | ||
1286 | /* put back the old socket callbacks */ | |
1287 | sk->sk_state_change = svsk->sk_ostate; | |
1288 | sk->sk_data_ready = svsk->sk_odata; | |
1289 | sk->sk_write_space = svsk->sk_owspace; | |
1290 | } | |
1291 | ||
1292 | /* | |
1293 | * Free the svc_sock's socket resources and the svc_sock itself. | |
1294 | */ | |
1295 | static void svc_sock_free(struct svc_xprt *xprt) | |
1296 | { | |
1297 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
1298 | dprintk("svc: svc_sock_free(%p)\n", svsk); | |
1299 | ||
755cceab TT |
1300 | if (svsk->sk_sock->file) |
1301 | sockfd_put(svsk->sk_sock); | |
1302 | else | |
1303 | sock_release(svsk->sk_sock); | |
1304 | kfree(svsk); | |
1305 | } |