]>
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 | |
8 | * svc_sock_enqueue procedure... | |
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 | ||
22 | #include <linux/sched.h> | |
23 | #include <linux/errno.h> | |
24 | #include <linux/fcntl.h> | |
25 | #include <linux/net.h> | |
26 | #include <linux/in.h> | |
27 | #include <linux/inet.h> | |
28 | #include <linux/udp.h> | |
91483c4b | 29 | #include <linux/tcp.h> |
1da177e4 LT |
30 | #include <linux/unistd.h> |
31 | #include <linux/slab.h> | |
32 | #include <linux/netdevice.h> | |
33 | #include <linux/skbuff.h> | |
b41b66d6 | 34 | #include <linux/file.h> |
1da177e4 LT |
35 | #include <net/sock.h> |
36 | #include <net/checksum.h> | |
37 | #include <net/ip.h> | |
c752f073 | 38 | #include <net/tcp_states.h> |
1da177e4 LT |
39 | #include <asm/uaccess.h> |
40 | #include <asm/ioctls.h> | |
41 | ||
42 | #include <linux/sunrpc/types.h> | |
43 | #include <linux/sunrpc/xdr.h> | |
44 | #include <linux/sunrpc/svcsock.h> | |
45 | #include <linux/sunrpc/stats.h> | |
46 | ||
47 | /* SMP locking strategy: | |
48 | * | |
3262c816 GB |
49 | * svc_pool->sp_lock protects most of the fields of that pool. |
50 | * svc_serv->sv_lock protects sv_tempsocks, sv_permsocks, sv_tmpcnt. | |
51 | * when both need to be taken (rare), svc_serv->sv_lock is first. | |
52 | * BKL protects svc_serv->sv_nrthread. | |
1a68d952 | 53 | * svc_sock->sk_defer_lock protects the svc_sock->sk_deferred list |
c081a0c7 | 54 | * svc_sock->sk_flags.SK_BUSY prevents a svc_sock being enqueued multiply. |
1da177e4 LT |
55 | * |
56 | * Some flags can be set to certain values at any time | |
57 | * providing that certain rules are followed: | |
58 | * | |
1da177e4 LT |
59 | * SK_CONN, SK_DATA, can be set or cleared at any time. |
60 | * after a set, svc_sock_enqueue must be called. | |
61 | * after a clear, the socket must be read/accepted | |
62 | * if this succeeds, it must be set again. | |
63 | * SK_CLOSE can set at any time. It is never cleared. | |
64 | * | |
65 | */ | |
66 | ||
67 | #define RPCDBG_FACILITY RPCDBG_SVCSOCK | |
68 | ||
69 | ||
70 | static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *, | |
71 | int *errp, int pmap_reg); | |
72 | static void svc_udp_data_ready(struct sock *, int); | |
73 | static int svc_udp_recvfrom(struct svc_rqst *); | |
74 | static int svc_udp_sendto(struct svc_rqst *); | |
75 | ||
76 | static struct svc_deferred_req *svc_deferred_dequeue(struct svc_sock *svsk); | |
77 | static int svc_deferred_recv(struct svc_rqst *rqstp); | |
78 | static struct cache_deferred_req *svc_defer(struct cache_req *req); | |
79 | ||
36bdfc8b GB |
80 | /* apparently the "standard" is that clients close |
81 | * idle connections after 5 minutes, servers after | |
82 | * 6 minutes | |
83 | * http://www.connectathon.org/talks96/nfstcp.pdf | |
84 | */ | |
85 | static int svc_conn_age_period = 6*60; | |
86 | ||
1da177e4 | 87 | /* |
3262c816 | 88 | * Queue up an idle server thread. Must have pool->sp_lock held. |
1da177e4 | 89 | * Note: this is really a stack rather than a queue, so that we only |
3262c816 | 90 | * use as many different threads as we need, and the rest don't pollute |
1da177e4 LT |
91 | * the cache. |
92 | */ | |
93 | static inline void | |
3262c816 | 94 | svc_thread_enqueue(struct svc_pool *pool, struct svc_rqst *rqstp) |
1da177e4 | 95 | { |
3262c816 | 96 | list_add(&rqstp->rq_list, &pool->sp_threads); |
1da177e4 LT |
97 | } |
98 | ||
99 | /* | |
3262c816 | 100 | * Dequeue an nfsd thread. Must have pool->sp_lock held. |
1da177e4 LT |
101 | */ |
102 | static inline void | |
3262c816 | 103 | svc_thread_dequeue(struct svc_pool *pool, struct svc_rqst *rqstp) |
1da177e4 LT |
104 | { |
105 | list_del(&rqstp->rq_list); | |
106 | } | |
107 | ||
108 | /* | |
109 | * Release an skbuff after use | |
110 | */ | |
111 | static inline void | |
112 | svc_release_skb(struct svc_rqst *rqstp) | |
113 | { | |
114 | struct sk_buff *skb = rqstp->rq_skbuff; | |
115 | struct svc_deferred_req *dr = rqstp->rq_deferred; | |
116 | ||
117 | if (skb) { | |
118 | rqstp->rq_skbuff = NULL; | |
119 | ||
120 | dprintk("svc: service %p, releasing skb %p\n", rqstp, skb); | |
121 | skb_free_datagram(rqstp->rq_sock->sk_sk, skb); | |
122 | } | |
123 | if (dr) { | |
124 | rqstp->rq_deferred = NULL; | |
125 | kfree(dr); | |
126 | } | |
127 | } | |
128 | ||
129 | /* | |
130 | * Any space to write? | |
131 | */ | |
132 | static inline unsigned long | |
133 | svc_sock_wspace(struct svc_sock *svsk) | |
134 | { | |
135 | int wspace; | |
136 | ||
137 | if (svsk->sk_sock->type == SOCK_STREAM) | |
138 | wspace = sk_stream_wspace(svsk->sk_sk); | |
139 | else | |
140 | wspace = sock_wspace(svsk->sk_sk); | |
141 | ||
142 | return wspace; | |
143 | } | |
144 | ||
145 | /* | |
146 | * Queue up a socket with data pending. If there are idle nfsd | |
147 | * processes, wake 'em up. | |
148 | * | |
149 | */ | |
150 | static void | |
151 | svc_sock_enqueue(struct svc_sock *svsk) | |
152 | { | |
153 | struct svc_serv *serv = svsk->sk_server; | |
bfd24160 | 154 | struct svc_pool *pool; |
1da177e4 | 155 | struct svc_rqst *rqstp; |
bfd24160 | 156 | int cpu; |
1da177e4 LT |
157 | |
158 | if (!(svsk->sk_flags & | |
159 | ( (1<<SK_CONN)|(1<<SK_DATA)|(1<<SK_CLOSE)|(1<<SK_DEFERRED)) )) | |
160 | return; | |
161 | if (test_bit(SK_DEAD, &svsk->sk_flags)) | |
162 | return; | |
163 | ||
bfd24160 GB |
164 | cpu = get_cpu(); |
165 | pool = svc_pool_for_cpu(svsk->sk_server, cpu); | |
166 | put_cpu(); | |
167 | ||
3262c816 | 168 | spin_lock_bh(&pool->sp_lock); |
1da177e4 | 169 | |
3262c816 GB |
170 | if (!list_empty(&pool->sp_threads) && |
171 | !list_empty(&pool->sp_sockets)) | |
1da177e4 LT |
172 | printk(KERN_ERR |
173 | "svc_sock_enqueue: threads and sockets both waiting??\n"); | |
174 | ||
175 | if (test_bit(SK_DEAD, &svsk->sk_flags)) { | |
176 | /* Don't enqueue dead sockets */ | |
177 | dprintk("svc: socket %p is dead, not enqueued\n", svsk->sk_sk); | |
178 | goto out_unlock; | |
179 | } | |
180 | ||
c081a0c7 GB |
181 | /* Mark socket as busy. It will remain in this state until the |
182 | * server has processed all pending data and put the socket back | |
183 | * on the idle list. We update SK_BUSY atomically because | |
184 | * it also guards against trying to enqueue the svc_sock twice. | |
185 | */ | |
186 | if (test_and_set_bit(SK_BUSY, &svsk->sk_flags)) { | |
187 | /* Don't enqueue socket while already enqueued */ | |
1da177e4 LT |
188 | dprintk("svc: socket %p busy, not enqueued\n", svsk->sk_sk); |
189 | goto out_unlock; | |
190 | } | |
3262c816 GB |
191 | BUG_ON(svsk->sk_pool != NULL); |
192 | svsk->sk_pool = pool; | |
1da177e4 LT |
193 | |
194 | set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
5685f0fa | 195 | if (((atomic_read(&svsk->sk_reserved) + serv->sv_bufsz)*2 |
1da177e4 LT |
196 | > svc_sock_wspace(svsk)) |
197 | && !test_bit(SK_CLOSE, &svsk->sk_flags) | |
198 | && !test_bit(SK_CONN, &svsk->sk_flags)) { | |
199 | /* Don't enqueue while not enough space for reply */ | |
200 | dprintk("svc: socket %p no space, %d*2 > %ld, not enqueued\n", | |
5685f0fa | 201 | svsk->sk_sk, atomic_read(&svsk->sk_reserved)+serv->sv_bufsz, |
1da177e4 | 202 | svc_sock_wspace(svsk)); |
3262c816 | 203 | svsk->sk_pool = NULL; |
c081a0c7 | 204 | clear_bit(SK_BUSY, &svsk->sk_flags); |
1da177e4 LT |
205 | goto out_unlock; |
206 | } | |
207 | clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
208 | ||
1da177e4 | 209 | |
3262c816 GB |
210 | if (!list_empty(&pool->sp_threads)) { |
211 | rqstp = list_entry(pool->sp_threads.next, | |
1da177e4 LT |
212 | struct svc_rqst, |
213 | rq_list); | |
214 | dprintk("svc: socket %p served by daemon %p\n", | |
215 | svsk->sk_sk, rqstp); | |
3262c816 | 216 | svc_thread_dequeue(pool, rqstp); |
1da177e4 LT |
217 | if (rqstp->rq_sock) |
218 | printk(KERN_ERR | |
219 | "svc_sock_enqueue: server %p, rq_sock=%p!\n", | |
220 | rqstp, rqstp->rq_sock); | |
221 | rqstp->rq_sock = svsk; | |
c45c357d | 222 | atomic_inc(&svsk->sk_inuse); |
1da177e4 | 223 | rqstp->rq_reserved = serv->sv_bufsz; |
5685f0fa | 224 | atomic_add(rqstp->rq_reserved, &svsk->sk_reserved); |
3262c816 | 225 | BUG_ON(svsk->sk_pool != pool); |
1da177e4 LT |
226 | wake_up(&rqstp->rq_wait); |
227 | } else { | |
228 | dprintk("svc: socket %p put into queue\n", svsk->sk_sk); | |
3262c816 GB |
229 | list_add_tail(&svsk->sk_ready, &pool->sp_sockets); |
230 | BUG_ON(svsk->sk_pool != pool); | |
1da177e4 LT |
231 | } |
232 | ||
233 | out_unlock: | |
3262c816 | 234 | spin_unlock_bh(&pool->sp_lock); |
1da177e4 LT |
235 | } |
236 | ||
237 | /* | |
3262c816 | 238 | * Dequeue the first socket. Must be called with the pool->sp_lock held. |
1da177e4 LT |
239 | */ |
240 | static inline struct svc_sock * | |
3262c816 | 241 | svc_sock_dequeue(struct svc_pool *pool) |
1da177e4 LT |
242 | { |
243 | struct svc_sock *svsk; | |
244 | ||
3262c816 | 245 | if (list_empty(&pool->sp_sockets)) |
1da177e4 LT |
246 | return NULL; |
247 | ||
3262c816 | 248 | svsk = list_entry(pool->sp_sockets.next, |
1da177e4 LT |
249 | struct svc_sock, sk_ready); |
250 | list_del_init(&svsk->sk_ready); | |
251 | ||
252 | dprintk("svc: socket %p dequeued, inuse=%d\n", | |
c45c357d | 253 | svsk->sk_sk, atomic_read(&svsk->sk_inuse)); |
1da177e4 LT |
254 | |
255 | return svsk; | |
256 | } | |
257 | ||
258 | /* | |
259 | * Having read something from a socket, check whether it | |
260 | * needs to be re-enqueued. | |
261 | * Note: SK_DATA only gets cleared when a read-attempt finds | |
262 | * no (or insufficient) data. | |
263 | */ | |
264 | static inline void | |
265 | svc_sock_received(struct svc_sock *svsk) | |
266 | { | |
3262c816 | 267 | svsk->sk_pool = NULL; |
1da177e4 LT |
268 | clear_bit(SK_BUSY, &svsk->sk_flags); |
269 | svc_sock_enqueue(svsk); | |
270 | } | |
271 | ||
272 | ||
273 | /** | |
274 | * svc_reserve - change the space reserved for the reply to a request. | |
275 | * @rqstp: The request in question | |
276 | * @space: new max space to reserve | |
277 | * | |
278 | * Each request reserves some space on the output queue of the socket | |
279 | * to make sure the reply fits. This function reduces that reserved | |
280 | * space to be the amount of space used already, plus @space. | |
281 | * | |
282 | */ | |
283 | void svc_reserve(struct svc_rqst *rqstp, int space) | |
284 | { | |
285 | space += rqstp->rq_res.head[0].iov_len; | |
286 | ||
287 | if (space < rqstp->rq_reserved) { | |
288 | struct svc_sock *svsk = rqstp->rq_sock; | |
5685f0fa | 289 | atomic_sub((rqstp->rq_reserved - space), &svsk->sk_reserved); |
1da177e4 | 290 | rqstp->rq_reserved = space; |
1da177e4 LT |
291 | |
292 | svc_sock_enqueue(svsk); | |
293 | } | |
294 | } | |
295 | ||
296 | /* | |
297 | * Release a socket after use. | |
298 | */ | |
299 | static inline void | |
300 | svc_sock_put(struct svc_sock *svsk) | |
301 | { | |
c45c357d | 302 | if (atomic_dec_and_test(&svsk->sk_inuse) && test_bit(SK_DEAD, &svsk->sk_flags)) { |
1da177e4 LT |
303 | dprintk("svc: releasing dead socket\n"); |
304 | sock_release(svsk->sk_sock); | |
305 | kfree(svsk); | |
306 | } | |
1da177e4 LT |
307 | } |
308 | ||
309 | static void | |
310 | svc_sock_release(struct svc_rqst *rqstp) | |
311 | { | |
312 | struct svc_sock *svsk = rqstp->rq_sock; | |
313 | ||
314 | svc_release_skb(rqstp); | |
315 | ||
316 | svc_free_allpages(rqstp); | |
317 | rqstp->rq_res.page_len = 0; | |
318 | rqstp->rq_res.page_base = 0; | |
319 | ||
320 | ||
321 | /* Reset response buffer and release | |
322 | * the reservation. | |
323 | * But first, check that enough space was reserved | |
324 | * for the reply, otherwise we have a bug! | |
325 | */ | |
326 | if ((rqstp->rq_res.len) > rqstp->rq_reserved) | |
327 | printk(KERN_ERR "RPC request reserved %d but used %d\n", | |
328 | rqstp->rq_reserved, | |
329 | rqstp->rq_res.len); | |
330 | ||
331 | rqstp->rq_res.head[0].iov_len = 0; | |
332 | svc_reserve(rqstp, 0); | |
333 | rqstp->rq_sock = NULL; | |
334 | ||
335 | svc_sock_put(svsk); | |
336 | } | |
337 | ||
338 | /* | |
339 | * External function to wake up a server waiting for data | |
3262c816 GB |
340 | * This really only makes sense for services like lockd |
341 | * which have exactly one thread anyway. | |
1da177e4 LT |
342 | */ |
343 | void | |
344 | svc_wake_up(struct svc_serv *serv) | |
345 | { | |
346 | struct svc_rqst *rqstp; | |
3262c816 GB |
347 | unsigned int i; |
348 | struct svc_pool *pool; | |
349 | ||
350 | for (i = 0; i < serv->sv_nrpools; i++) { | |
351 | pool = &serv->sv_pools[i]; | |
352 | ||
353 | spin_lock_bh(&pool->sp_lock); | |
354 | if (!list_empty(&pool->sp_threads)) { | |
355 | rqstp = list_entry(pool->sp_threads.next, | |
356 | struct svc_rqst, | |
357 | rq_list); | |
358 | dprintk("svc: daemon %p woken up.\n", rqstp); | |
359 | /* | |
360 | svc_thread_dequeue(pool, rqstp); | |
361 | rqstp->rq_sock = NULL; | |
362 | */ | |
363 | wake_up(&rqstp->rq_wait); | |
364 | } | |
365 | spin_unlock_bh(&pool->sp_lock); | |
1da177e4 | 366 | } |
1da177e4 LT |
367 | } |
368 | ||
369 | /* | |
370 | * Generic sendto routine | |
371 | */ | |
372 | static int | |
373 | svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr) | |
374 | { | |
375 | struct svc_sock *svsk = rqstp->rq_sock; | |
376 | struct socket *sock = svsk->sk_sock; | |
377 | int slen; | |
378 | char buffer[CMSG_SPACE(sizeof(struct in_pktinfo))]; | |
379 | struct cmsghdr *cmh = (struct cmsghdr *)buffer; | |
380 | struct in_pktinfo *pki = (struct in_pktinfo *)CMSG_DATA(cmh); | |
381 | int len = 0; | |
382 | int result; | |
383 | int size; | |
384 | struct page **ppage = xdr->pages; | |
385 | size_t base = xdr->page_base; | |
386 | unsigned int pglen = xdr->page_len; | |
387 | unsigned int flags = MSG_MORE; | |
388 | ||
389 | slen = xdr->len; | |
390 | ||
391 | if (rqstp->rq_prot == IPPROTO_UDP) { | |
392 | /* set the source and destination */ | |
393 | struct msghdr msg; | |
394 | msg.msg_name = &rqstp->rq_addr; | |
395 | msg.msg_namelen = sizeof(rqstp->rq_addr); | |
396 | msg.msg_iov = NULL; | |
397 | msg.msg_iovlen = 0; | |
398 | msg.msg_flags = MSG_MORE; | |
399 | ||
400 | msg.msg_control = cmh; | |
401 | msg.msg_controllen = sizeof(buffer); | |
402 | cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); | |
403 | cmh->cmsg_level = SOL_IP; | |
404 | cmh->cmsg_type = IP_PKTINFO; | |
405 | pki->ipi_ifindex = 0; | |
406 | pki->ipi_spec_dst.s_addr = rqstp->rq_daddr; | |
407 | ||
408 | if (sock_sendmsg(sock, &msg, 0) < 0) | |
409 | goto out; | |
410 | } | |
411 | ||
412 | /* send head */ | |
413 | if (slen == xdr->head[0].iov_len) | |
414 | flags = 0; | |
e6242e92 | 415 | len = kernel_sendpage(sock, rqstp->rq_respages[0], 0, xdr->head[0].iov_len, flags); |
1da177e4 LT |
416 | if (len != xdr->head[0].iov_len) |
417 | goto out; | |
418 | slen -= xdr->head[0].iov_len; | |
419 | if (slen == 0) | |
420 | goto out; | |
421 | ||
422 | /* send page data */ | |
423 | size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen; | |
424 | while (pglen > 0) { | |
425 | if (slen == size) | |
426 | flags = 0; | |
e6242e92 | 427 | result = kernel_sendpage(sock, *ppage, base, size, flags); |
1da177e4 LT |
428 | if (result > 0) |
429 | len += result; | |
430 | if (result != size) | |
431 | goto out; | |
432 | slen -= size; | |
433 | pglen -= size; | |
434 | size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen; | |
435 | base = 0; | |
436 | ppage++; | |
437 | } | |
438 | /* send tail */ | |
439 | if (xdr->tail[0].iov_len) { | |
e6242e92 | 440 | result = kernel_sendpage(sock, rqstp->rq_respages[rqstp->rq_restailpage], |
1da177e4 LT |
441 | ((unsigned long)xdr->tail[0].iov_base)& (PAGE_SIZE-1), |
442 | xdr->tail[0].iov_len, 0); | |
443 | ||
444 | if (result > 0) | |
445 | len += result; | |
446 | } | |
447 | out: | |
448 | dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %x)\n", | |
449 | rqstp->rq_sock, xdr->head[0].iov_base, xdr->head[0].iov_len, xdr->len, len, | |
450 | rqstp->rq_addr.sin_addr.s_addr); | |
451 | ||
452 | return len; | |
453 | } | |
454 | ||
80212d59 N |
455 | /* |
456 | * Report socket names for nfsdfs | |
457 | */ | |
458 | static int one_sock_name(char *buf, struct svc_sock *svsk) | |
459 | { | |
460 | int len; | |
461 | ||
462 | switch(svsk->sk_sk->sk_family) { | |
463 | case AF_INET: | |
464 | len = sprintf(buf, "ipv4 %s %u.%u.%u.%u %d\n", | |
465 | svsk->sk_sk->sk_protocol==IPPROTO_UDP? | |
466 | "udp" : "tcp", | |
467 | NIPQUAD(inet_sk(svsk->sk_sk)->rcv_saddr), | |
468 | inet_sk(svsk->sk_sk)->num); | |
469 | break; | |
470 | default: | |
471 | len = sprintf(buf, "*unknown-%d*\n", | |
472 | svsk->sk_sk->sk_family); | |
473 | } | |
474 | return len; | |
475 | } | |
476 | ||
477 | int | |
b41b66d6 | 478 | svc_sock_names(char *buf, struct svc_serv *serv, char *toclose) |
80212d59 | 479 | { |
b41b66d6 | 480 | struct svc_sock *svsk, *closesk = NULL; |
80212d59 N |
481 | int len = 0; |
482 | ||
483 | if (!serv) | |
484 | return 0; | |
485 | spin_lock(&serv->sv_lock); | |
486 | list_for_each_entry(svsk, &serv->sv_permsocks, sk_list) { | |
487 | int onelen = one_sock_name(buf+len, svsk); | |
b41b66d6 N |
488 | if (toclose && strcmp(toclose, buf+len) == 0) |
489 | closesk = svsk; | |
490 | else | |
491 | len += onelen; | |
80212d59 N |
492 | } |
493 | spin_unlock(&serv->sv_lock); | |
b41b66d6 N |
494 | if (closesk) |
495 | svc_delete_socket(closesk); | |
37a03472 N |
496 | else if (toclose) |
497 | return -ENOENT; | |
80212d59 N |
498 | return len; |
499 | } | |
500 | EXPORT_SYMBOL(svc_sock_names); | |
501 | ||
1da177e4 LT |
502 | /* |
503 | * Check input queue length | |
504 | */ | |
505 | static int | |
506 | svc_recv_available(struct svc_sock *svsk) | |
507 | { | |
1da177e4 LT |
508 | struct socket *sock = svsk->sk_sock; |
509 | int avail, err; | |
510 | ||
e6242e92 | 511 | err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail); |
1da177e4 LT |
512 | |
513 | return (err >= 0)? avail : err; | |
514 | } | |
515 | ||
516 | /* | |
517 | * Generic recvfrom routine. | |
518 | */ | |
519 | static int | |
520 | svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr, int buflen) | |
521 | { | |
522 | struct msghdr msg; | |
523 | struct socket *sock; | |
524 | int len, alen; | |
525 | ||
526 | rqstp->rq_addrlen = sizeof(rqstp->rq_addr); | |
527 | sock = rqstp->rq_sock->sk_sock; | |
528 | ||
529 | msg.msg_name = &rqstp->rq_addr; | |
530 | msg.msg_namelen = sizeof(rqstp->rq_addr); | |
531 | msg.msg_control = NULL; | |
532 | msg.msg_controllen = 0; | |
533 | ||
534 | msg.msg_flags = MSG_DONTWAIT; | |
535 | ||
536 | len = kernel_recvmsg(sock, &msg, iov, nr, buflen, MSG_DONTWAIT); | |
537 | ||
538 | /* sock_recvmsg doesn't fill in the name/namelen, so we must.. | |
539 | * possibly we should cache this in the svc_sock structure | |
540 | * at accept time. FIXME | |
541 | */ | |
542 | alen = sizeof(rqstp->rq_addr); | |
e6242e92 | 543 | kernel_getpeername(sock, (struct sockaddr *)&rqstp->rq_addr, &alen); |
1da177e4 LT |
544 | |
545 | dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n", | |
546 | rqstp->rq_sock, iov[0].iov_base, iov[0].iov_len, len); | |
547 | ||
548 | return len; | |
549 | } | |
550 | ||
551 | /* | |
552 | * Set socket snd and rcv buffer lengths | |
553 | */ | |
554 | static inline void | |
555 | svc_sock_setbufsize(struct socket *sock, unsigned int snd, unsigned int rcv) | |
556 | { | |
557 | #if 0 | |
558 | mm_segment_t oldfs; | |
559 | oldfs = get_fs(); set_fs(KERNEL_DS); | |
560 | sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF, | |
561 | (char*)&snd, sizeof(snd)); | |
562 | sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF, | |
563 | (char*)&rcv, sizeof(rcv)); | |
564 | #else | |
565 | /* sock_setsockopt limits use to sysctl_?mem_max, | |
566 | * which isn't acceptable. Until that is made conditional | |
567 | * on not having CAP_SYS_RESOURCE or similar, we go direct... | |
568 | * DaveM said I could! | |
569 | */ | |
570 | lock_sock(sock->sk); | |
571 | sock->sk->sk_sndbuf = snd * 2; | |
572 | sock->sk->sk_rcvbuf = rcv * 2; | |
573 | sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK|SOCK_RCVBUF_LOCK; | |
574 | release_sock(sock->sk); | |
575 | #endif | |
576 | } | |
577 | /* | |
578 | * INET callback when data has been received on the socket. | |
579 | */ | |
580 | static void | |
581 | svc_udp_data_ready(struct sock *sk, int count) | |
582 | { | |
939bb7ef | 583 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 | 584 | |
939bb7ef NB |
585 | if (svsk) { |
586 | dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n", | |
587 | svsk, sk, count, test_bit(SK_BUSY, &svsk->sk_flags)); | |
588 | set_bit(SK_DATA, &svsk->sk_flags); | |
589 | svc_sock_enqueue(svsk); | |
590 | } | |
1da177e4 LT |
591 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
592 | wake_up_interruptible(sk->sk_sleep); | |
593 | } | |
594 | ||
595 | /* | |
596 | * INET callback when space is newly available on the socket. | |
597 | */ | |
598 | static void | |
599 | svc_write_space(struct sock *sk) | |
600 | { | |
601 | struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data); | |
602 | ||
603 | if (svsk) { | |
604 | dprintk("svc: socket %p(inet %p), write_space busy=%d\n", | |
605 | svsk, sk, test_bit(SK_BUSY, &svsk->sk_flags)); | |
606 | svc_sock_enqueue(svsk); | |
607 | } | |
608 | ||
609 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) { | |
939bb7ef | 610 | dprintk("RPC svc_write_space: someone sleeping on %p\n", |
1da177e4 LT |
611 | svsk); |
612 | wake_up_interruptible(sk->sk_sleep); | |
613 | } | |
614 | } | |
615 | ||
616 | /* | |
617 | * Receive a datagram from a UDP socket. | |
618 | */ | |
1da177e4 LT |
619 | static int |
620 | svc_udp_recvfrom(struct svc_rqst *rqstp) | |
621 | { | |
622 | struct svc_sock *svsk = rqstp->rq_sock; | |
623 | struct svc_serv *serv = svsk->sk_server; | |
624 | struct sk_buff *skb; | |
625 | int err, len; | |
626 | ||
627 | if (test_and_clear_bit(SK_CHNGBUF, &svsk->sk_flags)) | |
628 | /* udp sockets need large rcvbuf as all pending | |
629 | * requests are still in that buffer. sndbuf must | |
630 | * also be large enough that there is enough space | |
3262c816 GB |
631 | * for one reply per thread. We count all threads |
632 | * rather than threads in a particular pool, which | |
633 | * provides an upper bound on the number of threads | |
634 | * which will access the socket. | |
1da177e4 LT |
635 | */ |
636 | svc_sock_setbufsize(svsk->sk_sock, | |
637 | (serv->sv_nrthreads+3) * serv->sv_bufsz, | |
638 | (serv->sv_nrthreads+3) * serv->sv_bufsz); | |
639 | ||
640 | if ((rqstp->rq_deferred = svc_deferred_dequeue(svsk))) { | |
641 | svc_sock_received(svsk); | |
642 | return svc_deferred_recv(rqstp); | |
643 | } | |
644 | ||
645 | clear_bit(SK_DATA, &svsk->sk_flags); | |
646 | while ((skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err)) == NULL) { | |
647 | if (err == -EAGAIN) { | |
648 | svc_sock_received(svsk); | |
649 | return err; | |
650 | } | |
651 | /* possibly an icmp error */ | |
652 | dprintk("svc: recvfrom returned error %d\n", -err); | |
653 | } | |
a61bbcf2 PM |
654 | if (skb->tstamp.off_sec == 0) { |
655 | struct timeval tv; | |
656 | ||
657 | tv.tv_sec = xtime.tv_sec; | |
4bcde03d | 658 | tv.tv_usec = xtime.tv_nsec / NSEC_PER_USEC; |
a61bbcf2 | 659 | skb_set_timestamp(skb, &tv); |
1da177e4 LT |
660 | /* Don't enable netstamp, sunrpc doesn't |
661 | need that much accuracy */ | |
662 | } | |
a61bbcf2 | 663 | skb_get_timestamp(skb, &svsk->sk_sk->sk_stamp); |
1da177e4 LT |
664 | set_bit(SK_DATA, &svsk->sk_flags); /* there may be more data... */ |
665 | ||
666 | /* | |
667 | * Maybe more packets - kick another thread ASAP. | |
668 | */ | |
669 | svc_sock_received(svsk); | |
670 | ||
671 | len = skb->len - sizeof(struct udphdr); | |
672 | rqstp->rq_arg.len = len; | |
673 | ||
674 | rqstp->rq_prot = IPPROTO_UDP; | |
675 | ||
676 | /* Get sender address */ | |
677 | rqstp->rq_addr.sin_family = AF_INET; | |
678 | rqstp->rq_addr.sin_port = skb->h.uh->source; | |
679 | rqstp->rq_addr.sin_addr.s_addr = skb->nh.iph->saddr; | |
680 | rqstp->rq_daddr = skb->nh.iph->daddr; | |
681 | ||
682 | if (skb_is_nonlinear(skb)) { | |
683 | /* we have to copy */ | |
684 | local_bh_disable(); | |
685 | if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) { | |
686 | local_bh_enable(); | |
687 | /* checksum error */ | |
688 | skb_free_datagram(svsk->sk_sk, skb); | |
689 | return 0; | |
690 | } | |
691 | local_bh_enable(); | |
692 | skb_free_datagram(svsk->sk_sk, skb); | |
693 | } else { | |
694 | /* we can use it in-place */ | |
695 | rqstp->rq_arg.head[0].iov_base = skb->data + sizeof(struct udphdr); | |
696 | rqstp->rq_arg.head[0].iov_len = len; | |
fb286bb2 HX |
697 | if (skb_checksum_complete(skb)) { |
698 | skb_free_datagram(svsk->sk_sk, skb); | |
699 | return 0; | |
1da177e4 LT |
700 | } |
701 | rqstp->rq_skbuff = skb; | |
702 | } | |
703 | ||
704 | rqstp->rq_arg.page_base = 0; | |
705 | if (len <= rqstp->rq_arg.head[0].iov_len) { | |
706 | rqstp->rq_arg.head[0].iov_len = len; | |
707 | rqstp->rq_arg.page_len = 0; | |
708 | } else { | |
709 | rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len; | |
710 | rqstp->rq_argused += (rqstp->rq_arg.page_len + PAGE_SIZE - 1)/ PAGE_SIZE; | |
711 | } | |
712 | ||
713 | if (serv->sv_stats) | |
714 | serv->sv_stats->netudpcnt++; | |
715 | ||
716 | return len; | |
717 | } | |
718 | ||
719 | static int | |
720 | svc_udp_sendto(struct svc_rqst *rqstp) | |
721 | { | |
722 | int error; | |
723 | ||
724 | error = svc_sendto(rqstp, &rqstp->rq_res); | |
725 | if (error == -ECONNREFUSED) | |
726 | /* ICMP error on earlier request. */ | |
727 | error = svc_sendto(rqstp, &rqstp->rq_res); | |
728 | ||
729 | return error; | |
730 | } | |
731 | ||
732 | static void | |
733 | svc_udp_init(struct svc_sock *svsk) | |
734 | { | |
735 | svsk->sk_sk->sk_data_ready = svc_udp_data_ready; | |
736 | svsk->sk_sk->sk_write_space = svc_write_space; | |
737 | svsk->sk_recvfrom = svc_udp_recvfrom; | |
738 | svsk->sk_sendto = svc_udp_sendto; | |
739 | ||
740 | /* initialise setting must have enough space to | |
741 | * receive and respond to one request. | |
742 | * svc_udp_recvfrom will re-adjust if necessary | |
743 | */ | |
744 | svc_sock_setbufsize(svsk->sk_sock, | |
745 | 3 * svsk->sk_server->sv_bufsz, | |
746 | 3 * svsk->sk_server->sv_bufsz); | |
747 | ||
748 | set_bit(SK_DATA, &svsk->sk_flags); /* might have come in before data_ready set up */ | |
749 | set_bit(SK_CHNGBUF, &svsk->sk_flags); | |
750 | } | |
751 | ||
752 | /* | |
753 | * A data_ready event on a listening socket means there's a connection | |
754 | * pending. Do not use state_change as a substitute for it. | |
755 | */ | |
756 | static void | |
757 | svc_tcp_listen_data_ready(struct sock *sk, int count_unused) | |
758 | { | |
939bb7ef | 759 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 LT |
760 | |
761 | dprintk("svc: socket %p TCP (listen) state change %d\n", | |
939bb7ef | 762 | sk, sk->sk_state); |
1da177e4 | 763 | |
939bb7ef NB |
764 | /* |
765 | * This callback may called twice when a new connection | |
766 | * is established as a child socket inherits everything | |
767 | * from a parent LISTEN socket. | |
768 | * 1) data_ready method of the parent socket will be called | |
769 | * when one of child sockets become ESTABLISHED. | |
770 | * 2) data_ready method of the child socket may be called | |
771 | * when it receives data before the socket is accepted. | |
772 | * In case of 2, we should ignore it silently. | |
773 | */ | |
774 | if (sk->sk_state == TCP_LISTEN) { | |
775 | if (svsk) { | |
776 | set_bit(SK_CONN, &svsk->sk_flags); | |
777 | svc_sock_enqueue(svsk); | |
778 | } else | |
779 | printk("svc: socket %p: no user data\n", sk); | |
1da177e4 | 780 | } |
939bb7ef | 781 | |
1da177e4 LT |
782 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
783 | wake_up_interruptible_all(sk->sk_sleep); | |
784 | } | |
785 | ||
786 | /* | |
787 | * A state change on a connected socket means it's dying or dead. | |
788 | */ | |
789 | static void | |
790 | svc_tcp_state_change(struct sock *sk) | |
791 | { | |
939bb7ef | 792 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 LT |
793 | |
794 | dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n", | |
939bb7ef | 795 | sk, sk->sk_state, sk->sk_user_data); |
1da177e4 | 796 | |
939bb7ef | 797 | if (!svsk) |
1da177e4 | 798 | printk("svc: socket %p: no user data\n", sk); |
939bb7ef NB |
799 | else { |
800 | set_bit(SK_CLOSE, &svsk->sk_flags); | |
801 | svc_sock_enqueue(svsk); | |
1da177e4 | 802 | } |
1da177e4 LT |
803 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
804 | wake_up_interruptible_all(sk->sk_sleep); | |
805 | } | |
806 | ||
807 | static void | |
808 | svc_tcp_data_ready(struct sock *sk, int count) | |
809 | { | |
939bb7ef | 810 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 LT |
811 | |
812 | dprintk("svc: socket %p TCP data ready (svsk %p)\n", | |
939bb7ef NB |
813 | sk, sk->sk_user_data); |
814 | if (svsk) { | |
815 | set_bit(SK_DATA, &svsk->sk_flags); | |
816 | svc_sock_enqueue(svsk); | |
817 | } | |
1da177e4 LT |
818 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
819 | wake_up_interruptible(sk->sk_sleep); | |
820 | } | |
821 | ||
822 | /* | |
823 | * Accept a TCP connection | |
824 | */ | |
825 | static void | |
826 | svc_tcp_accept(struct svc_sock *svsk) | |
827 | { | |
828 | struct sockaddr_in sin; | |
829 | struct svc_serv *serv = svsk->sk_server; | |
830 | struct socket *sock = svsk->sk_sock; | |
831 | struct socket *newsock; | |
1da177e4 LT |
832 | struct svc_sock *newsvsk; |
833 | int err, slen; | |
834 | ||
835 | dprintk("svc: tcp_accept %p sock %p\n", svsk, sock); | |
836 | if (!sock) | |
837 | return; | |
838 | ||
e6242e92 SS |
839 | clear_bit(SK_CONN, &svsk->sk_flags); |
840 | err = kernel_accept(sock, &newsock, O_NONBLOCK); | |
841 | if (err < 0) { | |
1da177e4 LT |
842 | if (err == -ENOMEM) |
843 | printk(KERN_WARNING "%s: no more sockets!\n", | |
844 | serv->sv_name); | |
e6242e92 | 845 | else if (err != -EAGAIN && net_ratelimit()) |
1da177e4 LT |
846 | printk(KERN_WARNING "%s: accept failed (err %d)!\n", |
847 | serv->sv_name, -err); | |
e6242e92 | 848 | return; |
1da177e4 | 849 | } |
e6242e92 | 850 | |
1da177e4 LT |
851 | set_bit(SK_CONN, &svsk->sk_flags); |
852 | svc_sock_enqueue(svsk); | |
853 | ||
854 | slen = sizeof(sin); | |
e6242e92 | 855 | err = kernel_getpeername(newsock, (struct sockaddr *) &sin, &slen); |
1da177e4 LT |
856 | if (err < 0) { |
857 | if (net_ratelimit()) | |
858 | printk(KERN_WARNING "%s: peername failed (err %d)!\n", | |
859 | serv->sv_name, -err); | |
860 | goto failed; /* aborted connection or whatever */ | |
861 | } | |
862 | ||
863 | /* Ideally, we would want to reject connections from unauthorized | |
864 | * hosts here, but when we get encription, the IP of the host won't | |
865 | * tell us anything. For now just warn about unpriv connections. | |
866 | */ | |
867 | if (ntohs(sin.sin_port) >= 1024) { | |
868 | dprintk(KERN_WARNING | |
869 | "%s: connect from unprivileged port: %u.%u.%u.%u:%d\n", | |
870 | serv->sv_name, | |
871 | NIPQUAD(sin.sin_addr.s_addr), ntohs(sin.sin_port)); | |
872 | } | |
873 | ||
874 | dprintk("%s: connect from %u.%u.%u.%u:%04x\n", serv->sv_name, | |
875 | NIPQUAD(sin.sin_addr.s_addr), ntohs(sin.sin_port)); | |
876 | ||
877 | /* make sure that a write doesn't block forever when | |
878 | * low on memory | |
879 | */ | |
880 | newsock->sk->sk_sndtimeo = HZ*30; | |
881 | ||
882 | if (!(newsvsk = svc_setup_socket(serv, newsock, &err, 0))) | |
883 | goto failed; | |
884 | ||
885 | ||
886 | /* make sure that we don't have too many active connections. | |
887 | * If we have, something must be dropped. | |
888 | * | |
889 | * There's no point in trying to do random drop here for | |
890 | * DoS prevention. The NFS clients does 1 reconnect in 15 | |
891 | * seconds. An attacker can easily beat that. | |
892 | * | |
893 | * The only somewhat efficient mechanism would be if drop | |
894 | * old connections from the same IP first. But right now | |
895 | * we don't even record the client IP in svc_sock. | |
896 | */ | |
897 | if (serv->sv_tmpcnt > (serv->sv_nrthreads+3)*20) { | |
898 | struct svc_sock *svsk = NULL; | |
899 | spin_lock_bh(&serv->sv_lock); | |
900 | if (!list_empty(&serv->sv_tempsocks)) { | |
901 | if (net_ratelimit()) { | |
902 | /* Try to help the admin */ | |
903 | printk(KERN_NOTICE "%s: too many open TCP " | |
904 | "sockets, consider increasing the " | |
905 | "number of nfsd threads\n", | |
906 | serv->sv_name); | |
907 | printk(KERN_NOTICE "%s: last TCP connect from " | |
908 | "%u.%u.%u.%u:%d\n", | |
909 | serv->sv_name, | |
910 | NIPQUAD(sin.sin_addr.s_addr), | |
911 | ntohs(sin.sin_port)); | |
912 | } | |
913 | /* | |
914 | * Always select the oldest socket. It's not fair, | |
915 | * but so is life | |
916 | */ | |
917 | svsk = list_entry(serv->sv_tempsocks.prev, | |
918 | struct svc_sock, | |
919 | sk_list); | |
920 | set_bit(SK_CLOSE, &svsk->sk_flags); | |
c45c357d | 921 | atomic_inc(&svsk->sk_inuse); |
1da177e4 LT |
922 | } |
923 | spin_unlock_bh(&serv->sv_lock); | |
924 | ||
925 | if (svsk) { | |
926 | svc_sock_enqueue(svsk); | |
927 | svc_sock_put(svsk); | |
928 | } | |
929 | ||
930 | } | |
931 | ||
932 | if (serv->sv_stats) | |
933 | serv->sv_stats->nettcpconn++; | |
934 | ||
935 | return; | |
936 | ||
937 | failed: | |
938 | sock_release(newsock); | |
939 | return; | |
940 | } | |
941 | ||
942 | /* | |
943 | * Receive data from a TCP socket. | |
944 | */ | |
945 | static int | |
946 | svc_tcp_recvfrom(struct svc_rqst *rqstp) | |
947 | { | |
948 | struct svc_sock *svsk = rqstp->rq_sock; | |
949 | struct svc_serv *serv = svsk->sk_server; | |
950 | int len; | |
951 | struct kvec vec[RPCSVC_MAXPAGES]; | |
952 | int pnum, vlen; | |
953 | ||
954 | dprintk("svc: tcp_recv %p data %d conn %d close %d\n", | |
955 | svsk, test_bit(SK_DATA, &svsk->sk_flags), | |
956 | test_bit(SK_CONN, &svsk->sk_flags), | |
957 | test_bit(SK_CLOSE, &svsk->sk_flags)); | |
958 | ||
959 | if ((rqstp->rq_deferred = svc_deferred_dequeue(svsk))) { | |
960 | svc_sock_received(svsk); | |
961 | return svc_deferred_recv(rqstp); | |
962 | } | |
963 | ||
964 | if (test_bit(SK_CLOSE, &svsk->sk_flags)) { | |
965 | svc_delete_socket(svsk); | |
966 | return 0; | |
967 | } | |
968 | ||
969 | if (test_bit(SK_CONN, &svsk->sk_flags)) { | |
970 | svc_tcp_accept(svsk); | |
971 | svc_sock_received(svsk); | |
972 | return 0; | |
973 | } | |
974 | ||
975 | if (test_and_clear_bit(SK_CHNGBUF, &svsk->sk_flags)) | |
976 | /* sndbuf needs to have room for one request | |
977 | * per thread, otherwise we can stall even when the | |
978 | * network isn't a bottleneck. | |
3262c816 GB |
979 | * |
980 | * We count all threads rather than threads in a | |
981 | * particular pool, which provides an upper bound | |
982 | * on the number of threads which will access the socket. | |
983 | * | |
1da177e4 LT |
984 | * rcvbuf just needs to be able to hold a few requests. |
985 | * Normally they will be removed from the queue | |
986 | * as soon a a complete request arrives. | |
987 | */ | |
988 | svc_sock_setbufsize(svsk->sk_sock, | |
989 | (serv->sv_nrthreads+3) * serv->sv_bufsz, | |
990 | 3 * serv->sv_bufsz); | |
991 | ||
992 | clear_bit(SK_DATA, &svsk->sk_flags); | |
993 | ||
994 | /* Receive data. If we haven't got the record length yet, get | |
995 | * the next four bytes. Otherwise try to gobble up as much as | |
996 | * possible up to the complete record length. | |
997 | */ | |
998 | if (svsk->sk_tcplen < 4) { | |
999 | unsigned long want = 4 - svsk->sk_tcplen; | |
1000 | struct kvec iov; | |
1001 | ||
1002 | iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen; | |
1003 | iov.iov_len = want; | |
1004 | if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0) | |
1005 | goto error; | |
1006 | svsk->sk_tcplen += len; | |
1007 | ||
1008 | if (len < want) { | |
1009 | dprintk("svc: short recvfrom while reading record length (%d of %lu)\n", | |
1010 | len, want); | |
1011 | svc_sock_received(svsk); | |
1012 | return -EAGAIN; /* record header not complete */ | |
1013 | } | |
1014 | ||
1015 | svsk->sk_reclen = ntohl(svsk->sk_reclen); | |
1016 | if (!(svsk->sk_reclen & 0x80000000)) { | |
1017 | /* FIXME: technically, a record can be fragmented, | |
1018 | * and non-terminal fragments will not have the top | |
1019 | * bit set in the fragment length header. | |
1020 | * But apparently no known nfs clients send fragmented | |
1021 | * records. */ | |
1022 | printk(KERN_NOTICE "RPC: bad TCP reclen 0x%08lx (non-terminal)\n", | |
1023 | (unsigned long) svsk->sk_reclen); | |
1024 | goto err_delete; | |
1025 | } | |
1026 | svsk->sk_reclen &= 0x7fffffff; | |
1027 | dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen); | |
1028 | if (svsk->sk_reclen > serv->sv_bufsz) { | |
1029 | printk(KERN_NOTICE "RPC: bad TCP reclen 0x%08lx (large)\n", | |
1030 | (unsigned long) svsk->sk_reclen); | |
1031 | goto err_delete; | |
1032 | } | |
1033 | } | |
1034 | ||
1035 | /* Check whether enough data is available */ | |
1036 | len = svc_recv_available(svsk); | |
1037 | if (len < 0) | |
1038 | goto error; | |
1039 | ||
1040 | if (len < svsk->sk_reclen) { | |
1041 | dprintk("svc: incomplete TCP record (%d of %d)\n", | |
1042 | len, svsk->sk_reclen); | |
1043 | svc_sock_received(svsk); | |
1044 | return -EAGAIN; /* record not complete */ | |
1045 | } | |
1046 | len = svsk->sk_reclen; | |
1047 | set_bit(SK_DATA, &svsk->sk_flags); | |
1048 | ||
1049 | vec[0] = rqstp->rq_arg.head[0]; | |
1050 | vlen = PAGE_SIZE; | |
1051 | pnum = 1; | |
1052 | while (vlen < len) { | |
1053 | vec[pnum].iov_base = page_address(rqstp->rq_argpages[rqstp->rq_argused++]); | |
1054 | vec[pnum].iov_len = PAGE_SIZE; | |
1055 | pnum++; | |
1056 | vlen += PAGE_SIZE; | |
1057 | } | |
1058 | ||
1059 | /* Now receive data */ | |
1060 | len = svc_recvfrom(rqstp, vec, pnum, len); | |
1061 | if (len < 0) | |
1062 | goto error; | |
1063 | ||
1064 | dprintk("svc: TCP complete record (%d bytes)\n", len); | |
1065 | rqstp->rq_arg.len = len; | |
1066 | rqstp->rq_arg.page_base = 0; | |
1067 | if (len <= rqstp->rq_arg.head[0].iov_len) { | |
1068 | rqstp->rq_arg.head[0].iov_len = len; | |
1069 | rqstp->rq_arg.page_len = 0; | |
1070 | } else { | |
1071 | rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len; | |
1072 | } | |
1073 | ||
1074 | rqstp->rq_skbuff = NULL; | |
1075 | rqstp->rq_prot = IPPROTO_TCP; | |
1076 | ||
1077 | /* Reset TCP read info */ | |
1078 | svsk->sk_reclen = 0; | |
1079 | svsk->sk_tcplen = 0; | |
1080 | ||
1081 | svc_sock_received(svsk); | |
1082 | if (serv->sv_stats) | |
1083 | serv->sv_stats->nettcpcnt++; | |
1084 | ||
1085 | return len; | |
1086 | ||
1087 | err_delete: | |
1088 | svc_delete_socket(svsk); | |
1089 | return -EAGAIN; | |
1090 | ||
1091 | error: | |
1092 | if (len == -EAGAIN) { | |
1093 | dprintk("RPC: TCP recvfrom got EAGAIN\n"); | |
1094 | svc_sock_received(svsk); | |
1095 | } else { | |
1096 | printk(KERN_NOTICE "%s: recvfrom returned errno %d\n", | |
1097 | svsk->sk_server->sv_name, -len); | |
93fbf1a5 | 1098 | goto err_delete; |
1da177e4 LT |
1099 | } |
1100 | ||
1101 | return len; | |
1102 | } | |
1103 | ||
1104 | /* | |
1105 | * Send out data on TCP socket. | |
1106 | */ | |
1107 | static int | |
1108 | svc_tcp_sendto(struct svc_rqst *rqstp) | |
1109 | { | |
1110 | struct xdr_buf *xbufp = &rqstp->rq_res; | |
1111 | int sent; | |
d8ed029d | 1112 | __be32 reclen; |
1da177e4 LT |
1113 | |
1114 | /* Set up the first element of the reply kvec. | |
1115 | * Any other kvecs that may be in use have been taken | |
1116 | * care of by the server implementation itself. | |
1117 | */ | |
1118 | reclen = htonl(0x80000000|((xbufp->len ) - 4)); | |
1119 | memcpy(xbufp->head[0].iov_base, &reclen, 4); | |
1120 | ||
1121 | if (test_bit(SK_DEAD, &rqstp->rq_sock->sk_flags)) | |
1122 | return -ENOTCONN; | |
1123 | ||
1124 | sent = svc_sendto(rqstp, &rqstp->rq_res); | |
1125 | if (sent != xbufp->len) { | |
1126 | printk(KERN_NOTICE "rpc-srv/tcp: %s: %s %d when sending %d bytes - shutting down socket\n", | |
1127 | rqstp->rq_sock->sk_server->sv_name, | |
1128 | (sent<0)?"got error":"sent only", | |
1129 | sent, xbufp->len); | |
1130 | svc_delete_socket(rqstp->rq_sock); | |
1131 | sent = -EAGAIN; | |
1132 | } | |
1133 | return sent; | |
1134 | } | |
1135 | ||
1136 | static void | |
1137 | svc_tcp_init(struct svc_sock *svsk) | |
1138 | { | |
1139 | struct sock *sk = svsk->sk_sk; | |
1140 | struct tcp_sock *tp = tcp_sk(sk); | |
1141 | ||
1142 | svsk->sk_recvfrom = svc_tcp_recvfrom; | |
1143 | svsk->sk_sendto = svc_tcp_sendto; | |
1144 | ||
1145 | if (sk->sk_state == TCP_LISTEN) { | |
1146 | dprintk("setting up TCP socket for listening\n"); | |
1147 | sk->sk_data_ready = svc_tcp_listen_data_ready; | |
1148 | set_bit(SK_CONN, &svsk->sk_flags); | |
1149 | } else { | |
1150 | dprintk("setting up TCP socket for reading\n"); | |
1151 | sk->sk_state_change = svc_tcp_state_change; | |
1152 | sk->sk_data_ready = svc_tcp_data_ready; | |
1153 | sk->sk_write_space = svc_write_space; | |
1154 | ||
1155 | svsk->sk_reclen = 0; | |
1156 | svsk->sk_tcplen = 0; | |
1157 | ||
1158 | tp->nonagle = 1; /* disable Nagle's algorithm */ | |
1159 | ||
1160 | /* initialise setting must have enough space to | |
1161 | * receive and respond to one request. | |
1162 | * svc_tcp_recvfrom will re-adjust if necessary | |
1163 | */ | |
1164 | svc_sock_setbufsize(svsk->sk_sock, | |
1165 | 3 * svsk->sk_server->sv_bufsz, | |
1166 | 3 * svsk->sk_server->sv_bufsz); | |
1167 | ||
1168 | set_bit(SK_CHNGBUF, &svsk->sk_flags); | |
1169 | set_bit(SK_DATA, &svsk->sk_flags); | |
1170 | if (sk->sk_state != TCP_ESTABLISHED) | |
1171 | set_bit(SK_CLOSE, &svsk->sk_flags); | |
1172 | } | |
1173 | } | |
1174 | ||
1175 | void | |
1176 | svc_sock_update_bufs(struct svc_serv *serv) | |
1177 | { | |
1178 | /* | |
1179 | * The number of server threads has changed. Update | |
1180 | * rcvbuf and sndbuf accordingly on all sockets | |
1181 | */ | |
1182 | struct list_head *le; | |
1183 | ||
1184 | spin_lock_bh(&serv->sv_lock); | |
1185 | list_for_each(le, &serv->sv_permsocks) { | |
1186 | struct svc_sock *svsk = | |
1187 | list_entry(le, struct svc_sock, sk_list); | |
1188 | set_bit(SK_CHNGBUF, &svsk->sk_flags); | |
1189 | } | |
1190 | list_for_each(le, &serv->sv_tempsocks) { | |
1191 | struct svc_sock *svsk = | |
1192 | list_entry(le, struct svc_sock, sk_list); | |
1193 | set_bit(SK_CHNGBUF, &svsk->sk_flags); | |
1194 | } | |
1195 | spin_unlock_bh(&serv->sv_lock); | |
1196 | } | |
1197 | ||
1198 | /* | |
3262c816 GB |
1199 | * Receive the next request on any socket. This code is carefully |
1200 | * organised not to touch any cachelines in the shared svc_serv | |
1201 | * structure, only cachelines in the local svc_pool. | |
1da177e4 LT |
1202 | */ |
1203 | int | |
6fb2b47f | 1204 | svc_recv(struct svc_rqst *rqstp, long timeout) |
1da177e4 LT |
1205 | { |
1206 | struct svc_sock *svsk =NULL; | |
6fb2b47f | 1207 | struct svc_serv *serv = rqstp->rq_server; |
3262c816 | 1208 | struct svc_pool *pool = rqstp->rq_pool; |
1da177e4 LT |
1209 | int len; |
1210 | int pages; | |
1211 | struct xdr_buf *arg; | |
1212 | DECLARE_WAITQUEUE(wait, current); | |
1213 | ||
1214 | dprintk("svc: server %p waiting for data (to = %ld)\n", | |
1215 | rqstp, timeout); | |
1216 | ||
1217 | if (rqstp->rq_sock) | |
1218 | printk(KERN_ERR | |
1219 | "svc_recv: service %p, socket not NULL!\n", | |
1220 | rqstp); | |
1221 | if (waitqueue_active(&rqstp->rq_wait)) | |
1222 | printk(KERN_ERR | |
1223 | "svc_recv: service %p, wait queue active!\n", | |
1224 | rqstp); | |
1225 | ||
1226 | /* Initialize the buffers */ | |
1227 | /* first reclaim pages that were moved to response list */ | |
1228 | svc_pushback_allpages(rqstp); | |
1229 | ||
1230 | /* now allocate needed pages. If we get a failure, sleep briefly */ | |
1231 | pages = 2 + (serv->sv_bufsz + PAGE_SIZE -1) / PAGE_SIZE; | |
1232 | while (rqstp->rq_arghi < pages) { | |
1233 | struct page *p = alloc_page(GFP_KERNEL); | |
1234 | if (!p) { | |
121caf57 | 1235 | schedule_timeout_uninterruptible(msecs_to_jiffies(500)); |
1da177e4 LT |
1236 | continue; |
1237 | } | |
1238 | rqstp->rq_argpages[rqstp->rq_arghi++] = p; | |
1239 | } | |
1240 | ||
1241 | /* Make arg->head point to first page and arg->pages point to rest */ | |
1242 | arg = &rqstp->rq_arg; | |
1243 | arg->head[0].iov_base = page_address(rqstp->rq_argpages[0]); | |
1244 | arg->head[0].iov_len = PAGE_SIZE; | |
1245 | rqstp->rq_argused = 1; | |
1246 | arg->pages = rqstp->rq_argpages + 1; | |
1247 | arg->page_base = 0; | |
1248 | /* save at least one page for response */ | |
1249 | arg->page_len = (pages-2)*PAGE_SIZE; | |
1250 | arg->len = (pages-1)*PAGE_SIZE; | |
1251 | arg->tail[0].iov_len = 0; | |
3e1d1d28 CL |
1252 | |
1253 | try_to_freeze(); | |
1887b935 | 1254 | cond_resched(); |
1da177e4 LT |
1255 | if (signalled()) |
1256 | return -EINTR; | |
1257 | ||
3262c816 GB |
1258 | spin_lock_bh(&pool->sp_lock); |
1259 | if ((svsk = svc_sock_dequeue(pool)) != NULL) { | |
1da177e4 | 1260 | rqstp->rq_sock = svsk; |
c45c357d | 1261 | atomic_inc(&svsk->sk_inuse); |
1da177e4 | 1262 | rqstp->rq_reserved = serv->sv_bufsz; |
5685f0fa | 1263 | atomic_add(rqstp->rq_reserved, &svsk->sk_reserved); |
1da177e4 LT |
1264 | } else { |
1265 | /* No data pending. Go to sleep */ | |
3262c816 | 1266 | svc_thread_enqueue(pool, rqstp); |
1da177e4 LT |
1267 | |
1268 | /* | |
1269 | * We have to be able to interrupt this wait | |
1270 | * to bring down the daemons ... | |
1271 | */ | |
1272 | set_current_state(TASK_INTERRUPTIBLE); | |
1273 | add_wait_queue(&rqstp->rq_wait, &wait); | |
3262c816 | 1274 | spin_unlock_bh(&pool->sp_lock); |
1da177e4 LT |
1275 | |
1276 | schedule_timeout(timeout); | |
1277 | ||
3e1d1d28 | 1278 | try_to_freeze(); |
1da177e4 | 1279 | |
3262c816 | 1280 | spin_lock_bh(&pool->sp_lock); |
1da177e4 LT |
1281 | remove_wait_queue(&rqstp->rq_wait, &wait); |
1282 | ||
1283 | if (!(svsk = rqstp->rq_sock)) { | |
3262c816 GB |
1284 | svc_thread_dequeue(pool, rqstp); |
1285 | spin_unlock_bh(&pool->sp_lock); | |
1da177e4 LT |
1286 | dprintk("svc: server %p, no data yet\n", rqstp); |
1287 | return signalled()? -EINTR : -EAGAIN; | |
1288 | } | |
1289 | } | |
3262c816 | 1290 | spin_unlock_bh(&pool->sp_lock); |
1da177e4 | 1291 | |
3262c816 GB |
1292 | dprintk("svc: server %p, pool %u, socket %p, inuse=%d\n", |
1293 | rqstp, pool->sp_id, svsk, atomic_read(&svsk->sk_inuse)); | |
1da177e4 LT |
1294 | len = svsk->sk_recvfrom(rqstp); |
1295 | dprintk("svc: got len=%d\n", len); | |
1296 | ||
1297 | /* No data, incomplete (TCP) read, or accept() */ | |
1298 | if (len == 0 || len == -EAGAIN) { | |
1299 | rqstp->rq_res.len = 0; | |
1300 | svc_sock_release(rqstp); | |
1301 | return -EAGAIN; | |
1302 | } | |
1303 | svsk->sk_lastrecv = get_seconds(); | |
36bdfc8b | 1304 | clear_bit(SK_OLD, &svsk->sk_flags); |
1da177e4 LT |
1305 | |
1306 | rqstp->rq_secure = ntohs(rqstp->rq_addr.sin_port) < 1024; | |
1307 | rqstp->rq_chandle.defer = svc_defer; | |
1308 | ||
1309 | if (serv->sv_stats) | |
1310 | serv->sv_stats->netcnt++; | |
1311 | return len; | |
1312 | } | |
1313 | ||
1314 | /* | |
1315 | * Drop request | |
1316 | */ | |
1317 | void | |
1318 | svc_drop(struct svc_rqst *rqstp) | |
1319 | { | |
1320 | dprintk("svc: socket %p dropped request\n", rqstp->rq_sock); | |
1321 | svc_sock_release(rqstp); | |
1322 | } | |
1323 | ||
1324 | /* | |
1325 | * Return reply to client. | |
1326 | */ | |
1327 | int | |
1328 | svc_send(struct svc_rqst *rqstp) | |
1329 | { | |
1330 | struct svc_sock *svsk; | |
1331 | int len; | |
1332 | struct xdr_buf *xb; | |
1333 | ||
1334 | if ((svsk = rqstp->rq_sock) == NULL) { | |
1335 | printk(KERN_WARNING "NULL socket pointer in %s:%d\n", | |
1336 | __FILE__, __LINE__); | |
1337 | return -EFAULT; | |
1338 | } | |
1339 | ||
1340 | /* release the receive skb before sending the reply */ | |
1341 | svc_release_skb(rqstp); | |
1342 | ||
1343 | /* calculate over-all length */ | |
1344 | xb = & rqstp->rq_res; | |
1345 | xb->len = xb->head[0].iov_len + | |
1346 | xb->page_len + | |
1347 | xb->tail[0].iov_len; | |
1348 | ||
57b47a53 IM |
1349 | /* Grab svsk->sk_mutex to serialize outgoing data. */ |
1350 | mutex_lock(&svsk->sk_mutex); | |
1da177e4 LT |
1351 | if (test_bit(SK_DEAD, &svsk->sk_flags)) |
1352 | len = -ENOTCONN; | |
1353 | else | |
1354 | len = svsk->sk_sendto(rqstp); | |
57b47a53 | 1355 | mutex_unlock(&svsk->sk_mutex); |
1da177e4 LT |
1356 | svc_sock_release(rqstp); |
1357 | ||
1358 | if (len == -ECONNREFUSED || len == -ENOTCONN || len == -EAGAIN) | |
1359 | return 0; | |
1360 | return len; | |
1361 | } | |
1362 | ||
36bdfc8b GB |
1363 | /* |
1364 | * Timer function to close old temporary sockets, using | |
1365 | * a mark-and-sweep algorithm. | |
1366 | */ | |
1367 | static void | |
1368 | svc_age_temp_sockets(unsigned long closure) | |
1369 | { | |
1370 | struct svc_serv *serv = (struct svc_serv *)closure; | |
1371 | struct svc_sock *svsk; | |
1372 | struct list_head *le, *next; | |
1373 | LIST_HEAD(to_be_aged); | |
1374 | ||
1375 | dprintk("svc_age_temp_sockets\n"); | |
1376 | ||
1377 | if (!spin_trylock_bh(&serv->sv_lock)) { | |
1378 | /* busy, try again 1 sec later */ | |
1379 | dprintk("svc_age_temp_sockets: busy\n"); | |
1380 | mod_timer(&serv->sv_temptimer, jiffies + HZ); | |
1381 | return; | |
1382 | } | |
1383 | ||
1384 | list_for_each_safe(le, next, &serv->sv_tempsocks) { | |
1385 | svsk = list_entry(le, struct svc_sock, sk_list); | |
1386 | ||
1387 | if (!test_and_set_bit(SK_OLD, &svsk->sk_flags)) | |
1388 | continue; | |
c45c357d | 1389 | if (atomic_read(&svsk->sk_inuse) || test_bit(SK_BUSY, &svsk->sk_flags)) |
36bdfc8b | 1390 | continue; |
c45c357d | 1391 | atomic_inc(&svsk->sk_inuse); |
36bdfc8b GB |
1392 | list_move(le, &to_be_aged); |
1393 | set_bit(SK_CLOSE, &svsk->sk_flags); | |
1394 | set_bit(SK_DETACHED, &svsk->sk_flags); | |
1395 | } | |
1396 | spin_unlock_bh(&serv->sv_lock); | |
1397 | ||
1398 | while (!list_empty(&to_be_aged)) { | |
1399 | le = to_be_aged.next; | |
1400 | /* fiddling the sk_list node is safe 'cos we're SK_DETACHED */ | |
1401 | list_del_init(le); | |
1402 | svsk = list_entry(le, struct svc_sock, sk_list); | |
1403 | ||
1404 | dprintk("queuing svsk %p for closing, %lu seconds old\n", | |
1405 | svsk, get_seconds() - svsk->sk_lastrecv); | |
1406 | ||
1407 | /* a thread will dequeue and close it soon */ | |
1408 | svc_sock_enqueue(svsk); | |
1409 | svc_sock_put(svsk); | |
1410 | } | |
1411 | ||
1412 | mod_timer(&serv->sv_temptimer, jiffies + svc_conn_age_period * HZ); | |
1413 | } | |
1414 | ||
1da177e4 LT |
1415 | /* |
1416 | * Initialize socket for RPC use and create svc_sock struct | |
1417 | * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF. | |
1418 | */ | |
1419 | static struct svc_sock * | |
1420 | svc_setup_socket(struct svc_serv *serv, struct socket *sock, | |
1421 | int *errp, int pmap_register) | |
1422 | { | |
1423 | struct svc_sock *svsk; | |
1424 | struct sock *inet; | |
1425 | ||
1426 | dprintk("svc: svc_setup_socket %p\n", sock); | |
0da974f4 | 1427 | if (!(svsk = kzalloc(sizeof(*svsk), GFP_KERNEL))) { |
1da177e4 LT |
1428 | *errp = -ENOMEM; |
1429 | return NULL; | |
1430 | } | |
1da177e4 LT |
1431 | |
1432 | inet = sock->sk; | |
1433 | ||
1434 | /* Register socket with portmapper */ | |
1435 | if (*errp >= 0 && pmap_register) | |
1436 | *errp = svc_register(serv, inet->sk_protocol, | |
1437 | ntohs(inet_sk(inet)->sport)); | |
1438 | ||
1439 | if (*errp < 0) { | |
1440 | kfree(svsk); | |
1441 | return NULL; | |
1442 | } | |
1443 | ||
1444 | set_bit(SK_BUSY, &svsk->sk_flags); | |
1445 | inet->sk_user_data = svsk; | |
1446 | svsk->sk_sock = sock; | |
1447 | svsk->sk_sk = inet; | |
1448 | svsk->sk_ostate = inet->sk_state_change; | |
1449 | svsk->sk_odata = inet->sk_data_ready; | |
1450 | svsk->sk_owspace = inet->sk_write_space; | |
1451 | svsk->sk_server = serv; | |
c45c357d | 1452 | atomic_set(&svsk->sk_inuse, 0); |
1da177e4 | 1453 | svsk->sk_lastrecv = get_seconds(); |
1a68d952 | 1454 | spin_lock_init(&svsk->sk_defer_lock); |
1da177e4 LT |
1455 | INIT_LIST_HEAD(&svsk->sk_deferred); |
1456 | INIT_LIST_HEAD(&svsk->sk_ready); | |
57b47a53 | 1457 | mutex_init(&svsk->sk_mutex); |
1da177e4 LT |
1458 | |
1459 | /* Initialize the socket */ | |
1460 | if (sock->type == SOCK_DGRAM) | |
1461 | svc_udp_init(svsk); | |
1462 | else | |
1463 | svc_tcp_init(svsk); | |
1464 | ||
1465 | spin_lock_bh(&serv->sv_lock); | |
1466 | if (!pmap_register) { | |
1467 | set_bit(SK_TEMP, &svsk->sk_flags); | |
1468 | list_add(&svsk->sk_list, &serv->sv_tempsocks); | |
1469 | serv->sv_tmpcnt++; | |
36bdfc8b GB |
1470 | if (serv->sv_temptimer.function == NULL) { |
1471 | /* setup timer to age temp sockets */ | |
1472 | setup_timer(&serv->sv_temptimer, svc_age_temp_sockets, | |
1473 | (unsigned long)serv); | |
1474 | mod_timer(&serv->sv_temptimer, | |
1475 | jiffies + svc_conn_age_period * HZ); | |
1476 | } | |
1da177e4 LT |
1477 | } else { |
1478 | clear_bit(SK_TEMP, &svsk->sk_flags); | |
1479 | list_add(&svsk->sk_list, &serv->sv_permsocks); | |
1480 | } | |
1481 | spin_unlock_bh(&serv->sv_lock); | |
1482 | ||
1483 | dprintk("svc: svc_setup_socket created %p (inet %p)\n", | |
1484 | svsk, svsk->sk_sk); | |
1485 | ||
1486 | clear_bit(SK_BUSY, &svsk->sk_flags); | |
1487 | svc_sock_enqueue(svsk); | |
1488 | return svsk; | |
1489 | } | |
1490 | ||
b41b66d6 N |
1491 | int svc_addsock(struct svc_serv *serv, |
1492 | int fd, | |
1493 | char *name_return, | |
1494 | int *proto) | |
1495 | { | |
1496 | int err = 0; | |
1497 | struct socket *so = sockfd_lookup(fd, &err); | |
1498 | struct svc_sock *svsk = NULL; | |
1499 | ||
1500 | if (!so) | |
1501 | return err; | |
1502 | if (so->sk->sk_family != AF_INET) | |
1503 | err = -EAFNOSUPPORT; | |
1504 | else if (so->sk->sk_protocol != IPPROTO_TCP && | |
1505 | so->sk->sk_protocol != IPPROTO_UDP) | |
1506 | err = -EPROTONOSUPPORT; | |
1507 | else if (so->state > SS_UNCONNECTED) | |
1508 | err = -EISCONN; | |
1509 | else { | |
1510 | svsk = svc_setup_socket(serv, so, &err, 1); | |
1511 | if (svsk) | |
1512 | err = 0; | |
1513 | } | |
1514 | if (err) { | |
1515 | sockfd_put(so); | |
1516 | return err; | |
1517 | } | |
1518 | if (proto) *proto = so->sk->sk_protocol; | |
1519 | return one_sock_name(name_return, svsk); | |
1520 | } | |
1521 | EXPORT_SYMBOL_GPL(svc_addsock); | |
1522 | ||
1da177e4 LT |
1523 | /* |
1524 | * Create socket for RPC service. | |
1525 | */ | |
1526 | static int | |
1527 | svc_create_socket(struct svc_serv *serv, int protocol, struct sockaddr_in *sin) | |
1528 | { | |
1529 | struct svc_sock *svsk; | |
1530 | struct socket *sock; | |
1531 | int error; | |
1532 | int type; | |
1533 | ||
1534 | dprintk("svc: svc_create_socket(%s, %d, %u.%u.%u.%u:%d)\n", | |
1535 | serv->sv_program->pg_name, protocol, | |
1536 | NIPQUAD(sin->sin_addr.s_addr), | |
1537 | ntohs(sin->sin_port)); | |
1538 | ||
1539 | if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) { | |
1540 | printk(KERN_WARNING "svc: only UDP and TCP " | |
1541 | "sockets supported\n"); | |
1542 | return -EINVAL; | |
1543 | } | |
1544 | type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM; | |
1545 | ||
1546 | if ((error = sock_create_kern(PF_INET, type, protocol, &sock)) < 0) | |
1547 | return error; | |
1548 | ||
18114746 ES |
1549 | if (type == SOCK_STREAM) |
1550 | sock->sk->sk_reuse = 1; /* allow address reuse */ | |
1551 | error = kernel_bind(sock, (struct sockaddr *) sin, | |
1552 | sizeof(*sin)); | |
1553 | if (error < 0) | |
1554 | goto bummer; | |
1da177e4 LT |
1555 | |
1556 | if (protocol == IPPROTO_TCP) { | |
e6242e92 | 1557 | if ((error = kernel_listen(sock, 64)) < 0) |
1da177e4 LT |
1558 | goto bummer; |
1559 | } | |
1560 | ||
1561 | if ((svsk = svc_setup_socket(serv, sock, &error, 1)) != NULL) | |
1562 | return 0; | |
1563 | ||
1564 | bummer: | |
1565 | dprintk("svc: svc_create_socket error = %d\n", -error); | |
1566 | sock_release(sock); | |
1567 | return error; | |
1568 | } | |
1569 | ||
1570 | /* | |
1571 | * Remove a dead socket | |
1572 | */ | |
1573 | void | |
1574 | svc_delete_socket(struct svc_sock *svsk) | |
1575 | { | |
1576 | struct svc_serv *serv; | |
1577 | struct sock *sk; | |
1578 | ||
1579 | dprintk("svc: svc_delete_socket(%p)\n", svsk); | |
1580 | ||
1581 | serv = svsk->sk_server; | |
1582 | sk = svsk->sk_sk; | |
1583 | ||
1584 | sk->sk_state_change = svsk->sk_ostate; | |
1585 | sk->sk_data_ready = svsk->sk_odata; | |
1586 | sk->sk_write_space = svsk->sk_owspace; | |
1587 | ||
1588 | spin_lock_bh(&serv->sv_lock); | |
1589 | ||
36bdfc8b GB |
1590 | if (!test_and_set_bit(SK_DETACHED, &svsk->sk_flags)) |
1591 | list_del_init(&svsk->sk_list); | |
3262c816 GB |
1592 | /* |
1593 | * We used to delete the svc_sock from whichever list | |
1594 | * it's sk_ready node was on, but we don't actually | |
1595 | * need to. This is because the only time we're called | |
1596 | * while still attached to a queue, the queue itself | |
1597 | * is about to be destroyed (in svc_destroy). | |
1598 | */ | |
1da177e4 LT |
1599 | if (!test_and_set_bit(SK_DEAD, &svsk->sk_flags)) |
1600 | if (test_bit(SK_TEMP, &svsk->sk_flags)) | |
1601 | serv->sv_tmpcnt--; | |
1602 | ||
c45c357d | 1603 | if (!atomic_read(&svsk->sk_inuse)) { |
1da177e4 | 1604 | spin_unlock_bh(&serv->sv_lock); |
b41b66d6 N |
1605 | if (svsk->sk_sock->file) |
1606 | sockfd_put(svsk->sk_sock); | |
1607 | else | |
1608 | sock_release(svsk->sk_sock); | |
1da177e4 LT |
1609 | kfree(svsk); |
1610 | } else { | |
1611 | spin_unlock_bh(&serv->sv_lock); | |
1612 | dprintk(KERN_NOTICE "svc: server socket destroy delayed\n"); | |
1613 | /* svsk->sk_server = NULL; */ | |
1614 | } | |
1615 | } | |
1616 | ||
1617 | /* | |
1618 | * Make a socket for nfsd and lockd | |
1619 | */ | |
1620 | int | |
1621 | svc_makesock(struct svc_serv *serv, int protocol, unsigned short port) | |
1622 | { | |
1623 | struct sockaddr_in sin; | |
1624 | ||
1625 | dprintk("svc: creating socket proto = %d\n", protocol); | |
1626 | sin.sin_family = AF_INET; | |
1627 | sin.sin_addr.s_addr = INADDR_ANY; | |
1628 | sin.sin_port = htons(port); | |
1629 | return svc_create_socket(serv, protocol, &sin); | |
1630 | } | |
1631 | ||
1632 | /* | |
1633 | * Handle defer and revisit of requests | |
1634 | */ | |
1635 | ||
1636 | static void svc_revisit(struct cache_deferred_req *dreq, int too_many) | |
1637 | { | |
1638 | struct svc_deferred_req *dr = container_of(dreq, struct svc_deferred_req, handle); | |
1da177e4 LT |
1639 | struct svc_sock *svsk; |
1640 | ||
1641 | if (too_many) { | |
1642 | svc_sock_put(dr->svsk); | |
1643 | kfree(dr); | |
1644 | return; | |
1645 | } | |
1646 | dprintk("revisit queued\n"); | |
1647 | svsk = dr->svsk; | |
1648 | dr->svsk = NULL; | |
1a68d952 | 1649 | spin_lock_bh(&svsk->sk_defer_lock); |
1da177e4 | 1650 | list_add(&dr->handle.recent, &svsk->sk_deferred); |
1a68d952 | 1651 | spin_unlock_bh(&svsk->sk_defer_lock); |
1da177e4 LT |
1652 | set_bit(SK_DEFERRED, &svsk->sk_flags); |
1653 | svc_sock_enqueue(svsk); | |
1654 | svc_sock_put(svsk); | |
1655 | } | |
1656 | ||
1657 | static struct cache_deferred_req * | |
1658 | svc_defer(struct cache_req *req) | |
1659 | { | |
1660 | struct svc_rqst *rqstp = container_of(req, struct svc_rqst, rq_chandle); | |
1661 | int size = sizeof(struct svc_deferred_req) + (rqstp->rq_arg.len); | |
1662 | struct svc_deferred_req *dr; | |
1663 | ||
1664 | if (rqstp->rq_arg.page_len) | |
1665 | return NULL; /* if more than a page, give up FIXME */ | |
1666 | if (rqstp->rq_deferred) { | |
1667 | dr = rqstp->rq_deferred; | |
1668 | rqstp->rq_deferred = NULL; | |
1669 | } else { | |
1670 | int skip = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len; | |
1671 | /* FIXME maybe discard if size too large */ | |
1672 | dr = kmalloc(size, GFP_KERNEL); | |
1673 | if (dr == NULL) | |
1674 | return NULL; | |
1675 | ||
1676 | dr->handle.owner = rqstp->rq_server; | |
1677 | dr->prot = rqstp->rq_prot; | |
1678 | dr->addr = rqstp->rq_addr; | |
1918e341 | 1679 | dr->daddr = rqstp->rq_daddr; |
1da177e4 LT |
1680 | dr->argslen = rqstp->rq_arg.len >> 2; |
1681 | memcpy(dr->args, rqstp->rq_arg.head[0].iov_base-skip, dr->argslen<<2); | |
1682 | } | |
c45c357d | 1683 | atomic_inc(&rqstp->rq_sock->sk_inuse); |
1da177e4 | 1684 | dr->svsk = rqstp->rq_sock; |
1da177e4 LT |
1685 | |
1686 | dr->handle.revisit = svc_revisit; | |
1687 | return &dr->handle; | |
1688 | } | |
1689 | ||
1690 | /* | |
1691 | * recv data from a deferred request into an active one | |
1692 | */ | |
1693 | static int svc_deferred_recv(struct svc_rqst *rqstp) | |
1694 | { | |
1695 | struct svc_deferred_req *dr = rqstp->rq_deferred; | |
1696 | ||
1697 | rqstp->rq_arg.head[0].iov_base = dr->args; | |
1698 | rqstp->rq_arg.head[0].iov_len = dr->argslen<<2; | |
1699 | rqstp->rq_arg.page_len = 0; | |
1700 | rqstp->rq_arg.len = dr->argslen<<2; | |
1701 | rqstp->rq_prot = dr->prot; | |
1702 | rqstp->rq_addr = dr->addr; | |
1918e341 | 1703 | rqstp->rq_daddr = dr->daddr; |
1da177e4 LT |
1704 | return dr->argslen<<2; |
1705 | } | |
1706 | ||
1707 | ||
1708 | static struct svc_deferred_req *svc_deferred_dequeue(struct svc_sock *svsk) | |
1709 | { | |
1710 | struct svc_deferred_req *dr = NULL; | |
1da177e4 LT |
1711 | |
1712 | if (!test_bit(SK_DEFERRED, &svsk->sk_flags)) | |
1713 | return NULL; | |
1a68d952 | 1714 | spin_lock_bh(&svsk->sk_defer_lock); |
1da177e4 LT |
1715 | clear_bit(SK_DEFERRED, &svsk->sk_flags); |
1716 | if (!list_empty(&svsk->sk_deferred)) { | |
1717 | dr = list_entry(svsk->sk_deferred.next, | |
1718 | struct svc_deferred_req, | |
1719 | handle.recent); | |
1720 | list_del_init(&dr->handle.recent); | |
1721 | set_bit(SK_DEFERRED, &svsk->sk_flags); | |
1722 | } | |
1a68d952 | 1723 | spin_unlock_bh(&svsk->sk_defer_lock); |
1da177e4 LT |
1724 | return dr; |
1725 | } |