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 | ||
172589cc | 22 | #include <linux/kernel.h> |
1da177e4 LT |
23 | #include <linux/sched.h> |
24 | #include <linux/errno.h> | |
25 | #include <linux/fcntl.h> | |
26 | #include <linux/net.h> | |
27 | #include <linux/in.h> | |
28 | #include <linux/inet.h> | |
29 | #include <linux/udp.h> | |
91483c4b | 30 | #include <linux/tcp.h> |
1da177e4 LT |
31 | #include <linux/unistd.h> |
32 | #include <linux/slab.h> | |
33 | #include <linux/netdevice.h> | |
34 | #include <linux/skbuff.h> | |
b41b66d6 | 35 | #include <linux/file.h> |
7dfb7103 | 36 | #include <linux/freezer.h> |
1da177e4 LT |
37 | #include <net/sock.h> |
38 | #include <net/checksum.h> | |
39 | #include <net/ip.h> | |
b92503b2 | 40 | #include <net/ipv6.h> |
c752f073 | 41 | #include <net/tcp_states.h> |
1da177e4 LT |
42 | #include <asm/uaccess.h> |
43 | #include <asm/ioctls.h> | |
44 | ||
45 | #include <linux/sunrpc/types.h> | |
ad06e4bd | 46 | #include <linux/sunrpc/clnt.h> |
1da177e4 LT |
47 | #include <linux/sunrpc/xdr.h> |
48 | #include <linux/sunrpc/svcsock.h> | |
49 | #include <linux/sunrpc/stats.h> | |
50 | ||
51 | /* SMP locking strategy: | |
52 | * | |
3262c816 GB |
53 | * svc_pool->sp_lock protects most of the fields of that pool. |
54 | * svc_serv->sv_lock protects sv_tempsocks, sv_permsocks, sv_tmpcnt. | |
55 | * when both need to be taken (rare), svc_serv->sv_lock is first. | |
56 | * BKL protects svc_serv->sv_nrthread. | |
7ac1bea5 N |
57 | * svc_sock->sk_lock protects the svc_sock->sk_deferred list |
58 | * and the ->sk_info_authunix cache. | |
c081a0c7 | 59 | * svc_sock->sk_flags.SK_BUSY prevents a svc_sock being enqueued multiply. |
1da177e4 LT |
60 | * |
61 | * Some flags can be set to certain values at any time | |
62 | * providing that certain rules are followed: | |
63 | * | |
1da177e4 | 64 | * SK_CONN, SK_DATA, can be set or cleared at any time. |
cca5172a | 65 | * after a set, svc_sock_enqueue must be called. |
1da177e4 LT |
66 | * after a clear, the socket must be read/accepted |
67 | * if this succeeds, it must be set again. | |
68 | * SK_CLOSE can set at any time. It is never cleared. | |
aaf68cfb N |
69 | * sk_inuse contains a bias of '1' until SK_DEAD is set. |
70 | * so when sk_inuse hits zero, we know the socket is dead | |
71 | * and no-one is using it. | |
72 | * SK_DEAD can only be set while SK_BUSY is held which ensures | |
73 | * no other thread will be using the socket or will try to | |
74 | * set SK_DEAD. | |
1da177e4 LT |
75 | * |
76 | */ | |
77 | ||
360d8738 | 78 | #define RPCDBG_FACILITY RPCDBG_SVCXPRT |
1da177e4 LT |
79 | |
80 | ||
81 | static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *, | |
6b174337 | 82 | int *errp, int flags); |
aaf68cfb | 83 | static void svc_delete_socket(struct svc_sock *svsk); |
1da177e4 LT |
84 | static void svc_udp_data_ready(struct sock *, int); |
85 | static int svc_udp_recvfrom(struct svc_rqst *); | |
86 | static int svc_udp_sendto(struct svc_rqst *); | |
cda1fd4a | 87 | static void svc_close_socket(struct svc_sock *svsk); |
755cceab TT |
88 | static void svc_sock_detach(struct svc_xprt *); |
89 | static void svc_sock_free(struct svc_xprt *); | |
1da177e4 LT |
90 | |
91 | static struct svc_deferred_req *svc_deferred_dequeue(struct svc_sock *svsk); | |
92 | static int svc_deferred_recv(struct svc_rqst *rqstp); | |
93 | static struct cache_deferred_req *svc_defer(struct cache_req *req); | |
94 | ||
36bdfc8b GB |
95 | /* apparently the "standard" is that clients close |
96 | * idle connections after 5 minutes, servers after | |
97 | * 6 minutes | |
98 | * http://www.connectathon.org/talks96/nfstcp.pdf | |
99 | */ | |
100 | static int svc_conn_age_period = 6*60; | |
101 | ||
ed07536e PZ |
102 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
103 | static struct lock_class_key svc_key[2]; | |
104 | static struct lock_class_key svc_slock_key[2]; | |
105 | ||
106 | static inline void svc_reclassify_socket(struct socket *sock) | |
107 | { | |
108 | struct sock *sk = sock->sk; | |
02b3d346 | 109 | BUG_ON(sock_owned_by_user(sk)); |
ed07536e PZ |
110 | switch (sk->sk_family) { |
111 | case AF_INET: | |
112 | sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD", | |
113 | &svc_slock_key[0], "sk_lock-AF_INET-NFSD", &svc_key[0]); | |
114 | break; | |
115 | ||
116 | case AF_INET6: | |
117 | sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD", | |
118 | &svc_slock_key[1], "sk_lock-AF_INET6-NFSD", &svc_key[1]); | |
119 | break; | |
120 | ||
121 | default: | |
122 | BUG(); | |
123 | } | |
124 | } | |
125 | #else | |
126 | static inline void svc_reclassify_socket(struct socket *sock) | |
127 | { | |
128 | } | |
129 | #endif | |
130 | ||
ad06e4bd CL |
131 | static char *__svc_print_addr(struct sockaddr *addr, char *buf, size_t len) |
132 | { | |
133 | switch (addr->sa_family) { | |
134 | case AF_INET: | |
135 | snprintf(buf, len, "%u.%u.%u.%u, port=%u", | |
136 | NIPQUAD(((struct sockaddr_in *) addr)->sin_addr), | |
582ee43d | 137 | ntohs(((struct sockaddr_in *) addr)->sin_port)); |
ad06e4bd | 138 | break; |
5a05ed73 | 139 | |
ad06e4bd CL |
140 | case AF_INET6: |
141 | snprintf(buf, len, "%x:%x:%x:%x:%x:%x:%x:%x, port=%u", | |
142 | NIP6(((struct sockaddr_in6 *) addr)->sin6_addr), | |
582ee43d | 143 | ntohs(((struct sockaddr_in6 *) addr)->sin6_port)); |
ad06e4bd | 144 | break; |
5a05ed73 | 145 | |
ad06e4bd CL |
146 | default: |
147 | snprintf(buf, len, "unknown address type: %d", addr->sa_family); | |
148 | break; | |
149 | } | |
150 | return buf; | |
151 | } | |
152 | ||
153 | /** | |
154 | * svc_print_addr - Format rq_addr field for printing | |
155 | * @rqstp: svc_rqst struct containing address to print | |
156 | * @buf: target buffer for formatted address | |
157 | * @len: length of target buffer | |
158 | * | |
159 | */ | |
160 | char *svc_print_addr(struct svc_rqst *rqstp, char *buf, size_t len) | |
161 | { | |
27459f09 | 162 | return __svc_print_addr(svc_addr(rqstp), buf, len); |
ad06e4bd CL |
163 | } |
164 | EXPORT_SYMBOL_GPL(svc_print_addr); | |
165 | ||
1da177e4 | 166 | /* |
3262c816 | 167 | * Queue up an idle server thread. Must have pool->sp_lock held. |
1da177e4 | 168 | * Note: this is really a stack rather than a queue, so that we only |
3262c816 | 169 | * use as many different threads as we need, and the rest don't pollute |
1da177e4 LT |
170 | * the cache. |
171 | */ | |
172 | static inline void | |
3262c816 | 173 | svc_thread_enqueue(struct svc_pool *pool, struct svc_rqst *rqstp) |
1da177e4 | 174 | { |
3262c816 | 175 | list_add(&rqstp->rq_list, &pool->sp_threads); |
1da177e4 LT |
176 | } |
177 | ||
178 | /* | |
3262c816 | 179 | * Dequeue an nfsd thread. Must have pool->sp_lock held. |
1da177e4 LT |
180 | */ |
181 | static inline void | |
3262c816 | 182 | svc_thread_dequeue(struct svc_pool *pool, struct svc_rqst *rqstp) |
1da177e4 LT |
183 | { |
184 | list_del(&rqstp->rq_list); | |
185 | } | |
186 | ||
187 | /* | |
188 | * Release an skbuff after use | |
189 | */ | |
5148bf4e | 190 | static void svc_release_skb(struct svc_rqst *rqstp) |
1da177e4 | 191 | { |
5148bf4e | 192 | struct sk_buff *skb = rqstp->rq_xprt_ctxt; |
1da177e4 LT |
193 | struct svc_deferred_req *dr = rqstp->rq_deferred; |
194 | ||
195 | if (skb) { | |
5148bf4e | 196 | rqstp->rq_xprt_ctxt = NULL; |
1da177e4 LT |
197 | |
198 | dprintk("svc: service %p, releasing skb %p\n", rqstp, skb); | |
199 | skb_free_datagram(rqstp->rq_sock->sk_sk, skb); | |
200 | } | |
201 | if (dr) { | |
202 | rqstp->rq_deferred = NULL; | |
203 | kfree(dr); | |
204 | } | |
205 | } | |
206 | ||
207 | /* | |
208 | * Any space to write? | |
209 | */ | |
210 | static inline unsigned long | |
211 | svc_sock_wspace(struct svc_sock *svsk) | |
212 | { | |
213 | int wspace; | |
214 | ||
215 | if (svsk->sk_sock->type == SOCK_STREAM) | |
216 | wspace = sk_stream_wspace(svsk->sk_sk); | |
217 | else | |
218 | wspace = sock_wspace(svsk->sk_sk); | |
219 | ||
220 | return wspace; | |
221 | } | |
222 | ||
223 | /* | |
224 | * Queue up a socket with data pending. If there are idle nfsd | |
225 | * processes, wake 'em up. | |
226 | * | |
227 | */ | |
228 | static void | |
229 | svc_sock_enqueue(struct svc_sock *svsk) | |
230 | { | |
231 | struct svc_serv *serv = svsk->sk_server; | |
bfd24160 | 232 | struct svc_pool *pool; |
1da177e4 | 233 | struct svc_rqst *rqstp; |
bfd24160 | 234 | int cpu; |
1da177e4 LT |
235 | |
236 | if (!(svsk->sk_flags & | |
237 | ( (1<<SK_CONN)|(1<<SK_DATA)|(1<<SK_CLOSE)|(1<<SK_DEFERRED)) )) | |
238 | return; | |
239 | if (test_bit(SK_DEAD, &svsk->sk_flags)) | |
240 | return; | |
241 | ||
bfd24160 GB |
242 | cpu = get_cpu(); |
243 | pool = svc_pool_for_cpu(svsk->sk_server, cpu); | |
244 | put_cpu(); | |
245 | ||
3262c816 | 246 | spin_lock_bh(&pool->sp_lock); |
1da177e4 | 247 | |
3262c816 GB |
248 | if (!list_empty(&pool->sp_threads) && |
249 | !list_empty(&pool->sp_sockets)) | |
1da177e4 LT |
250 | printk(KERN_ERR |
251 | "svc_sock_enqueue: threads and sockets both waiting??\n"); | |
252 | ||
253 | if (test_bit(SK_DEAD, &svsk->sk_flags)) { | |
254 | /* Don't enqueue dead sockets */ | |
255 | dprintk("svc: socket %p is dead, not enqueued\n", svsk->sk_sk); | |
256 | goto out_unlock; | |
257 | } | |
258 | ||
c081a0c7 GB |
259 | /* Mark socket as busy. It will remain in this state until the |
260 | * server has processed all pending data and put the socket back | |
261 | * on the idle list. We update SK_BUSY atomically because | |
262 | * it also guards against trying to enqueue the svc_sock twice. | |
263 | */ | |
264 | if (test_and_set_bit(SK_BUSY, &svsk->sk_flags)) { | |
265 | /* Don't enqueue socket while already enqueued */ | |
1da177e4 LT |
266 | dprintk("svc: socket %p busy, not enqueued\n", svsk->sk_sk); |
267 | goto out_unlock; | |
268 | } | |
3262c816 GB |
269 | BUG_ON(svsk->sk_pool != NULL); |
270 | svsk->sk_pool = pool; | |
1da177e4 LT |
271 | |
272 | set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
c6b0a9f8 | 273 | if (((atomic_read(&svsk->sk_reserved) + serv->sv_max_mesg)*2 |
1da177e4 LT |
274 | > svc_sock_wspace(svsk)) |
275 | && !test_bit(SK_CLOSE, &svsk->sk_flags) | |
276 | && !test_bit(SK_CONN, &svsk->sk_flags)) { | |
277 | /* Don't enqueue while not enough space for reply */ | |
278 | dprintk("svc: socket %p no space, %d*2 > %ld, not enqueued\n", | |
c6b0a9f8 | 279 | svsk->sk_sk, atomic_read(&svsk->sk_reserved)+serv->sv_max_mesg, |
1da177e4 | 280 | svc_sock_wspace(svsk)); |
3262c816 | 281 | svsk->sk_pool = NULL; |
c081a0c7 | 282 | clear_bit(SK_BUSY, &svsk->sk_flags); |
1da177e4 LT |
283 | goto out_unlock; |
284 | } | |
285 | clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
286 | ||
1da177e4 | 287 | |
3262c816 GB |
288 | if (!list_empty(&pool->sp_threads)) { |
289 | rqstp = list_entry(pool->sp_threads.next, | |
1da177e4 LT |
290 | struct svc_rqst, |
291 | rq_list); | |
292 | dprintk("svc: socket %p served by daemon %p\n", | |
293 | svsk->sk_sk, rqstp); | |
3262c816 | 294 | svc_thread_dequeue(pool, rqstp); |
1da177e4 | 295 | if (rqstp->rq_sock) |
cca5172a | 296 | printk(KERN_ERR |
1da177e4 LT |
297 | "svc_sock_enqueue: server %p, rq_sock=%p!\n", |
298 | rqstp, rqstp->rq_sock); | |
299 | rqstp->rq_sock = svsk; | |
c45c357d | 300 | atomic_inc(&svsk->sk_inuse); |
c6b0a9f8 | 301 | rqstp->rq_reserved = serv->sv_max_mesg; |
5685f0fa | 302 | atomic_add(rqstp->rq_reserved, &svsk->sk_reserved); |
3262c816 | 303 | BUG_ON(svsk->sk_pool != pool); |
1da177e4 LT |
304 | wake_up(&rqstp->rq_wait); |
305 | } else { | |
306 | dprintk("svc: socket %p put into queue\n", svsk->sk_sk); | |
3262c816 GB |
307 | list_add_tail(&svsk->sk_ready, &pool->sp_sockets); |
308 | BUG_ON(svsk->sk_pool != pool); | |
1da177e4 LT |
309 | } |
310 | ||
311 | out_unlock: | |
3262c816 | 312 | spin_unlock_bh(&pool->sp_lock); |
1da177e4 LT |
313 | } |
314 | ||
315 | /* | |
3262c816 | 316 | * Dequeue the first socket. Must be called with the pool->sp_lock held. |
1da177e4 LT |
317 | */ |
318 | static inline struct svc_sock * | |
3262c816 | 319 | svc_sock_dequeue(struct svc_pool *pool) |
1da177e4 LT |
320 | { |
321 | struct svc_sock *svsk; | |
322 | ||
3262c816 | 323 | if (list_empty(&pool->sp_sockets)) |
1da177e4 LT |
324 | return NULL; |
325 | ||
3262c816 | 326 | svsk = list_entry(pool->sp_sockets.next, |
1da177e4 LT |
327 | struct svc_sock, sk_ready); |
328 | list_del_init(&svsk->sk_ready); | |
329 | ||
330 | dprintk("svc: socket %p dequeued, inuse=%d\n", | |
c45c357d | 331 | svsk->sk_sk, atomic_read(&svsk->sk_inuse)); |
1da177e4 LT |
332 | |
333 | return svsk; | |
334 | } | |
335 | ||
336 | /* | |
337 | * Having read something from a socket, check whether it | |
338 | * needs to be re-enqueued. | |
339 | * Note: SK_DATA only gets cleared when a read-attempt finds | |
340 | * no (or insufficient) data. | |
341 | */ | |
342 | static inline void | |
343 | svc_sock_received(struct svc_sock *svsk) | |
344 | { | |
3262c816 | 345 | svsk->sk_pool = NULL; |
1da177e4 LT |
346 | clear_bit(SK_BUSY, &svsk->sk_flags); |
347 | svc_sock_enqueue(svsk); | |
348 | } | |
349 | ||
350 | ||
351 | /** | |
352 | * svc_reserve - change the space reserved for the reply to a request. | |
353 | * @rqstp: The request in question | |
354 | * @space: new max space to reserve | |
355 | * | |
356 | * Each request reserves some space on the output queue of the socket | |
357 | * to make sure the reply fits. This function reduces that reserved | |
358 | * space to be the amount of space used already, plus @space. | |
359 | * | |
360 | */ | |
361 | void svc_reserve(struct svc_rqst *rqstp, int space) | |
362 | { | |
363 | space += rqstp->rq_res.head[0].iov_len; | |
364 | ||
365 | if (space < rqstp->rq_reserved) { | |
366 | struct svc_sock *svsk = rqstp->rq_sock; | |
5685f0fa | 367 | atomic_sub((rqstp->rq_reserved - space), &svsk->sk_reserved); |
1da177e4 | 368 | rqstp->rq_reserved = space; |
1da177e4 LT |
369 | |
370 | svc_sock_enqueue(svsk); | |
371 | } | |
372 | } | |
373 | ||
374 | /* | |
375 | * Release a socket after use. | |
376 | */ | |
377 | static inline void | |
378 | svc_sock_put(struct svc_sock *svsk) | |
379 | { | |
aaf68cfb | 380 | if (atomic_dec_and_test(&svsk->sk_inuse)) { |
755cceab TT |
381 | BUG_ON(!test_bit(SK_DEAD, &svsk->sk_flags)); |
382 | svsk->sk_xprt.xpt_ops->xpo_free(&svsk->sk_xprt); | |
1da177e4 | 383 | } |
1da177e4 LT |
384 | } |
385 | ||
386 | static void | |
387 | svc_sock_release(struct svc_rqst *rqstp) | |
388 | { | |
389 | struct svc_sock *svsk = rqstp->rq_sock; | |
390 | ||
5148bf4e | 391 | rqstp->rq_xprt->xpt_ops->xpo_release_rqst(rqstp); |
1da177e4 | 392 | |
44524359 | 393 | svc_free_res_pages(rqstp); |
1da177e4 LT |
394 | rqstp->rq_res.page_len = 0; |
395 | rqstp->rq_res.page_base = 0; | |
396 | ||
397 | ||
398 | /* Reset response buffer and release | |
399 | * the reservation. | |
400 | * But first, check that enough space was reserved | |
401 | * for the reply, otherwise we have a bug! | |
402 | */ | |
403 | if ((rqstp->rq_res.len) > rqstp->rq_reserved) | |
404 | printk(KERN_ERR "RPC request reserved %d but used %d\n", | |
405 | rqstp->rq_reserved, | |
406 | rqstp->rq_res.len); | |
407 | ||
408 | rqstp->rq_res.head[0].iov_len = 0; | |
409 | svc_reserve(rqstp, 0); | |
410 | rqstp->rq_sock = NULL; | |
411 | ||
412 | svc_sock_put(svsk); | |
413 | } | |
414 | ||
415 | /* | |
416 | * External function to wake up a server waiting for data | |
3262c816 GB |
417 | * This really only makes sense for services like lockd |
418 | * which have exactly one thread anyway. | |
1da177e4 LT |
419 | */ |
420 | void | |
421 | svc_wake_up(struct svc_serv *serv) | |
422 | { | |
423 | struct svc_rqst *rqstp; | |
3262c816 GB |
424 | unsigned int i; |
425 | struct svc_pool *pool; | |
426 | ||
427 | for (i = 0; i < serv->sv_nrpools; i++) { | |
428 | pool = &serv->sv_pools[i]; | |
429 | ||
430 | spin_lock_bh(&pool->sp_lock); | |
431 | if (!list_empty(&pool->sp_threads)) { | |
432 | rqstp = list_entry(pool->sp_threads.next, | |
433 | struct svc_rqst, | |
434 | rq_list); | |
435 | dprintk("svc: daemon %p woken up.\n", rqstp); | |
436 | /* | |
437 | svc_thread_dequeue(pool, rqstp); | |
438 | rqstp->rq_sock = NULL; | |
439 | */ | |
440 | wake_up(&rqstp->rq_wait); | |
441 | } | |
442 | spin_unlock_bh(&pool->sp_lock); | |
1da177e4 | 443 | } |
1da177e4 LT |
444 | } |
445 | ||
b92503b2 CL |
446 | union svc_pktinfo_u { |
447 | struct in_pktinfo pkti; | |
b92503b2 | 448 | struct in6_pktinfo pkti6; |
b92503b2 | 449 | }; |
bc375ea7 DM |
450 | #define SVC_PKTINFO_SPACE \ |
451 | CMSG_SPACE(sizeof(union svc_pktinfo_u)) | |
b92503b2 CL |
452 | |
453 | static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh) | |
454 | { | |
455 | switch (rqstp->rq_sock->sk_sk->sk_family) { | |
456 | case AF_INET: { | |
457 | struct in_pktinfo *pki = CMSG_DATA(cmh); | |
458 | ||
459 | cmh->cmsg_level = SOL_IP; | |
460 | cmh->cmsg_type = IP_PKTINFO; | |
461 | pki->ipi_ifindex = 0; | |
462 | pki->ipi_spec_dst.s_addr = rqstp->rq_daddr.addr.s_addr; | |
463 | cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); | |
464 | } | |
465 | break; | |
5a05ed73 | 466 | |
b92503b2 CL |
467 | case AF_INET6: { |
468 | struct in6_pktinfo *pki = CMSG_DATA(cmh); | |
469 | ||
470 | cmh->cmsg_level = SOL_IPV6; | |
471 | cmh->cmsg_type = IPV6_PKTINFO; | |
472 | pki->ipi6_ifindex = 0; | |
473 | ipv6_addr_copy(&pki->ipi6_addr, | |
474 | &rqstp->rq_daddr.addr6); | |
475 | cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); | |
476 | } | |
477 | break; | |
b92503b2 CL |
478 | } |
479 | return; | |
480 | } | |
481 | ||
1da177e4 LT |
482 | /* |
483 | * Generic sendto routine | |
484 | */ | |
485 | static int | |
486 | svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr) | |
487 | { | |
488 | struct svc_sock *svsk = rqstp->rq_sock; | |
489 | struct socket *sock = svsk->sk_sock; | |
490 | int slen; | |
bc375ea7 DM |
491 | union { |
492 | struct cmsghdr hdr; | |
493 | long all[SVC_PKTINFO_SPACE / sizeof(long)]; | |
494 | } buffer; | |
495 | struct cmsghdr *cmh = &buffer.hdr; | |
1da177e4 LT |
496 | int len = 0; |
497 | int result; | |
498 | int size; | |
499 | struct page **ppage = xdr->pages; | |
500 | size_t base = xdr->page_base; | |
501 | unsigned int pglen = xdr->page_len; | |
502 | unsigned int flags = MSG_MORE; | |
ad06e4bd | 503 | char buf[RPC_MAX_ADDRBUFLEN]; |
1da177e4 LT |
504 | |
505 | slen = xdr->len; | |
506 | ||
507 | if (rqstp->rq_prot == IPPROTO_UDP) { | |
b92503b2 CL |
508 | struct msghdr msg = { |
509 | .msg_name = &rqstp->rq_addr, | |
510 | .msg_namelen = rqstp->rq_addrlen, | |
511 | .msg_control = cmh, | |
512 | .msg_controllen = sizeof(buffer), | |
513 | .msg_flags = MSG_MORE, | |
514 | }; | |
515 | ||
516 | svc_set_cmsg_data(rqstp, cmh); | |
1da177e4 LT |
517 | |
518 | if (sock_sendmsg(sock, &msg, 0) < 0) | |
519 | goto out; | |
520 | } | |
521 | ||
522 | /* send head */ | |
523 | if (slen == xdr->head[0].iov_len) | |
524 | flags = 0; | |
44524359 N |
525 | len = kernel_sendpage(sock, rqstp->rq_respages[0], 0, |
526 | xdr->head[0].iov_len, flags); | |
1da177e4 LT |
527 | if (len != xdr->head[0].iov_len) |
528 | goto out; | |
529 | slen -= xdr->head[0].iov_len; | |
530 | if (slen == 0) | |
531 | goto out; | |
532 | ||
533 | /* send page data */ | |
534 | size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen; | |
535 | while (pglen > 0) { | |
536 | if (slen == size) | |
537 | flags = 0; | |
e6242e92 | 538 | result = kernel_sendpage(sock, *ppage, base, size, flags); |
1da177e4 LT |
539 | if (result > 0) |
540 | len += result; | |
541 | if (result != size) | |
542 | goto out; | |
543 | slen -= size; | |
544 | pglen -= size; | |
545 | size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen; | |
546 | base = 0; | |
547 | ppage++; | |
548 | } | |
549 | /* send tail */ | |
550 | if (xdr->tail[0].iov_len) { | |
44524359 N |
551 | result = kernel_sendpage(sock, rqstp->rq_respages[0], |
552 | ((unsigned long)xdr->tail[0].iov_base) | |
cca5172a | 553 | & (PAGE_SIZE-1), |
1da177e4 LT |
554 | xdr->tail[0].iov_len, 0); |
555 | ||
556 | if (result > 0) | |
557 | len += result; | |
558 | } | |
559 | out: | |
ad06e4bd CL |
560 | dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n", |
561 | rqstp->rq_sock, xdr->head[0].iov_base, xdr->head[0].iov_len, | |
562 | xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf))); | |
1da177e4 LT |
563 | |
564 | return len; | |
565 | } | |
566 | ||
80212d59 N |
567 | /* |
568 | * Report socket names for nfsdfs | |
569 | */ | |
570 | static int one_sock_name(char *buf, struct svc_sock *svsk) | |
571 | { | |
572 | int len; | |
573 | ||
574 | switch(svsk->sk_sk->sk_family) { | |
575 | case AF_INET: | |
576 | len = sprintf(buf, "ipv4 %s %u.%u.%u.%u %d\n", | |
577 | svsk->sk_sk->sk_protocol==IPPROTO_UDP? | |
578 | "udp" : "tcp", | |
579 | NIPQUAD(inet_sk(svsk->sk_sk)->rcv_saddr), | |
580 | inet_sk(svsk->sk_sk)->num); | |
581 | break; | |
582 | default: | |
583 | len = sprintf(buf, "*unknown-%d*\n", | |
584 | svsk->sk_sk->sk_family); | |
585 | } | |
586 | return len; | |
587 | } | |
588 | ||
589 | int | |
b41b66d6 | 590 | svc_sock_names(char *buf, struct svc_serv *serv, char *toclose) |
80212d59 | 591 | { |
b41b66d6 | 592 | struct svc_sock *svsk, *closesk = NULL; |
80212d59 N |
593 | int len = 0; |
594 | ||
595 | if (!serv) | |
596 | return 0; | |
aaf68cfb | 597 | spin_lock_bh(&serv->sv_lock); |
80212d59 N |
598 | list_for_each_entry(svsk, &serv->sv_permsocks, sk_list) { |
599 | int onelen = one_sock_name(buf+len, svsk); | |
b41b66d6 N |
600 | if (toclose && strcmp(toclose, buf+len) == 0) |
601 | closesk = svsk; | |
602 | else | |
603 | len += onelen; | |
80212d59 | 604 | } |
aaf68cfb | 605 | spin_unlock_bh(&serv->sv_lock); |
b41b66d6 | 606 | if (closesk) |
5680c446 N |
607 | /* Should unregister with portmap, but you cannot |
608 | * unregister just one protocol... | |
609 | */ | |
aaf68cfb | 610 | svc_close_socket(closesk); |
37a03472 N |
611 | else if (toclose) |
612 | return -ENOENT; | |
80212d59 N |
613 | return len; |
614 | } | |
615 | EXPORT_SYMBOL(svc_sock_names); | |
616 | ||
1da177e4 LT |
617 | /* |
618 | * Check input queue length | |
619 | */ | |
620 | static int | |
621 | svc_recv_available(struct svc_sock *svsk) | |
622 | { | |
1da177e4 LT |
623 | struct socket *sock = svsk->sk_sock; |
624 | int avail, err; | |
625 | ||
e6242e92 | 626 | err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail); |
1da177e4 LT |
627 | |
628 | return (err >= 0)? avail : err; | |
629 | } | |
630 | ||
631 | /* | |
632 | * Generic recvfrom routine. | |
633 | */ | |
634 | static int | |
635 | svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr, int buflen) | |
636 | { | |
067d7817 | 637 | struct svc_sock *svsk = rqstp->rq_sock; |
1ba95105 CL |
638 | struct msghdr msg = { |
639 | .msg_flags = MSG_DONTWAIT, | |
640 | }; | |
a9747692 | 641 | struct sockaddr *sin; |
1ba95105 | 642 | int len; |
1da177e4 | 643 | |
1ba95105 CL |
644 | len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen, |
645 | msg.msg_flags); | |
1da177e4 LT |
646 | |
647 | /* sock_recvmsg doesn't fill in the name/namelen, so we must.. | |
1da177e4 | 648 | */ |
067d7817 CL |
649 | memcpy(&rqstp->rq_addr, &svsk->sk_remote, svsk->sk_remotelen); |
650 | rqstp->rq_addrlen = svsk->sk_remotelen; | |
1da177e4 | 651 | |
a9747692 FM |
652 | /* Destination address in request is needed for binding the |
653 | * source address in RPC callbacks later. | |
654 | */ | |
655 | sin = (struct sockaddr *)&svsk->sk_local; | |
656 | switch (sin->sa_family) { | |
657 | case AF_INET: | |
658 | rqstp->rq_daddr.addr = ((struct sockaddr_in *)sin)->sin_addr; | |
659 | break; | |
660 | case AF_INET6: | |
661 | rqstp->rq_daddr.addr6 = ((struct sockaddr_in6 *)sin)->sin6_addr; | |
662 | break; | |
663 | } | |
664 | ||
1da177e4 | 665 | dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n", |
1ba95105 | 666 | svsk, iov[0].iov_base, iov[0].iov_len, len); |
1da177e4 LT |
667 | |
668 | return len; | |
669 | } | |
670 | ||
671 | /* | |
672 | * Set socket snd and rcv buffer lengths | |
673 | */ | |
674 | static inline void | |
675 | svc_sock_setbufsize(struct socket *sock, unsigned int snd, unsigned int rcv) | |
676 | { | |
677 | #if 0 | |
678 | mm_segment_t oldfs; | |
679 | oldfs = get_fs(); set_fs(KERNEL_DS); | |
680 | sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF, | |
681 | (char*)&snd, sizeof(snd)); | |
682 | sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF, | |
683 | (char*)&rcv, sizeof(rcv)); | |
684 | #else | |
685 | /* sock_setsockopt limits use to sysctl_?mem_max, | |
686 | * which isn't acceptable. Until that is made conditional | |
687 | * on not having CAP_SYS_RESOURCE or similar, we go direct... | |
688 | * DaveM said I could! | |
689 | */ | |
690 | lock_sock(sock->sk); | |
691 | sock->sk->sk_sndbuf = snd * 2; | |
692 | sock->sk->sk_rcvbuf = rcv * 2; | |
693 | sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK|SOCK_RCVBUF_LOCK; | |
694 | release_sock(sock->sk); | |
695 | #endif | |
696 | } | |
697 | /* | |
698 | * INET callback when data has been received on the socket. | |
699 | */ | |
700 | static void | |
701 | svc_udp_data_ready(struct sock *sk, int count) | |
702 | { | |
939bb7ef | 703 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 | 704 | |
939bb7ef NB |
705 | if (svsk) { |
706 | dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n", | |
707 | svsk, sk, count, test_bit(SK_BUSY, &svsk->sk_flags)); | |
708 | set_bit(SK_DATA, &svsk->sk_flags); | |
709 | svc_sock_enqueue(svsk); | |
710 | } | |
1da177e4 LT |
711 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
712 | wake_up_interruptible(sk->sk_sleep); | |
713 | } | |
714 | ||
715 | /* | |
716 | * INET callback when space is newly available on the socket. | |
717 | */ | |
718 | static void | |
719 | svc_write_space(struct sock *sk) | |
720 | { | |
721 | struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data); | |
722 | ||
723 | if (svsk) { | |
724 | dprintk("svc: socket %p(inet %p), write_space busy=%d\n", | |
725 | svsk, sk, test_bit(SK_BUSY, &svsk->sk_flags)); | |
726 | svc_sock_enqueue(svsk); | |
727 | } | |
728 | ||
729 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) { | |
939bb7ef | 730 | dprintk("RPC svc_write_space: someone sleeping on %p\n", |
1da177e4 LT |
731 | svsk); |
732 | wake_up_interruptible(sk->sk_sleep); | |
733 | } | |
734 | } | |
735 | ||
7a37f578 N |
736 | static inline void svc_udp_get_dest_address(struct svc_rqst *rqstp, |
737 | struct cmsghdr *cmh) | |
95756482 CL |
738 | { |
739 | switch (rqstp->rq_sock->sk_sk->sk_family) { | |
740 | case AF_INET: { | |
7a37f578 N |
741 | struct in_pktinfo *pki = CMSG_DATA(cmh); |
742 | rqstp->rq_daddr.addr.s_addr = pki->ipi_spec_dst.s_addr; | |
95756482 | 743 | break; |
7a37f578 | 744 | } |
95756482 | 745 | case AF_INET6: { |
7a37f578 N |
746 | struct in6_pktinfo *pki = CMSG_DATA(cmh); |
747 | ipv6_addr_copy(&rqstp->rq_daddr.addr6, &pki->ipi6_addr); | |
95756482 | 748 | break; |
7a37f578 | 749 | } |
95756482 | 750 | } |
95756482 CL |
751 | } |
752 | ||
1da177e4 LT |
753 | /* |
754 | * Receive a datagram from a UDP socket. | |
755 | */ | |
1da177e4 LT |
756 | static int |
757 | svc_udp_recvfrom(struct svc_rqst *rqstp) | |
758 | { | |
759 | struct svc_sock *svsk = rqstp->rq_sock; | |
760 | struct svc_serv *serv = svsk->sk_server; | |
761 | struct sk_buff *skb; | |
bc375ea7 DM |
762 | union { |
763 | struct cmsghdr hdr; | |
764 | long all[SVC_PKTINFO_SPACE / sizeof(long)]; | |
765 | } buffer; | |
766 | struct cmsghdr *cmh = &buffer.hdr; | |
1da177e4 | 767 | int err, len; |
7a37f578 N |
768 | struct msghdr msg = { |
769 | .msg_name = svc_addr(rqstp), | |
770 | .msg_control = cmh, | |
771 | .msg_controllen = sizeof(buffer), | |
772 | .msg_flags = MSG_DONTWAIT, | |
773 | }; | |
1da177e4 LT |
774 | |
775 | if (test_and_clear_bit(SK_CHNGBUF, &svsk->sk_flags)) | |
776 | /* udp sockets need large rcvbuf as all pending | |
777 | * requests are still in that buffer. sndbuf must | |
778 | * also be large enough that there is enough space | |
3262c816 GB |
779 | * for one reply per thread. We count all threads |
780 | * rather than threads in a particular pool, which | |
781 | * provides an upper bound on the number of threads | |
782 | * which will access the socket. | |
1da177e4 LT |
783 | */ |
784 | svc_sock_setbufsize(svsk->sk_sock, | |
c6b0a9f8 N |
785 | (serv->sv_nrthreads+3) * serv->sv_max_mesg, |
786 | (serv->sv_nrthreads+3) * serv->sv_max_mesg); | |
1da177e4 LT |
787 | |
788 | if ((rqstp->rq_deferred = svc_deferred_dequeue(svsk))) { | |
789 | svc_sock_received(svsk); | |
790 | return svc_deferred_recv(rqstp); | |
791 | } | |
792 | ||
aaf68cfb N |
793 | if (test_bit(SK_CLOSE, &svsk->sk_flags)) { |
794 | svc_delete_socket(svsk); | |
795 | return 0; | |
796 | } | |
797 | ||
1da177e4 | 798 | clear_bit(SK_DATA, &svsk->sk_flags); |
05ed690e N |
799 | skb = NULL; |
800 | err = kernel_recvmsg(svsk->sk_sock, &msg, NULL, | |
801 | 0, 0, MSG_PEEK | MSG_DONTWAIT); | |
802 | if (err >= 0) | |
803 | skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err); | |
804 | ||
805 | if (skb == NULL) { | |
806 | if (err != -EAGAIN) { | |
807 | /* possibly an icmp error */ | |
808 | dprintk("svc: recvfrom returned error %d\n", -err); | |
809 | set_bit(SK_DATA, &svsk->sk_flags); | |
1da177e4 | 810 | } |
05ed690e N |
811 | svc_sock_received(svsk); |
812 | return -EAGAIN; | |
1da177e4 | 813 | } |
7a37f578 | 814 | rqstp->rq_addrlen = sizeof(rqstp->rq_addr); |
b7aa0bf7 ED |
815 | if (skb->tstamp.tv64 == 0) { |
816 | skb->tstamp = ktime_get_real(); | |
cca5172a | 817 | /* Don't enable netstamp, sunrpc doesn't |
1da177e4 LT |
818 | need that much accuracy */ |
819 | } | |
b7aa0bf7 | 820 | svsk->sk_sk->sk_stamp = skb->tstamp; |
1da177e4 LT |
821 | set_bit(SK_DATA, &svsk->sk_flags); /* there may be more data... */ |
822 | ||
823 | /* | |
824 | * Maybe more packets - kick another thread ASAP. | |
825 | */ | |
826 | svc_sock_received(svsk); | |
827 | ||
828 | len = skb->len - sizeof(struct udphdr); | |
829 | rqstp->rq_arg.len = len; | |
830 | ||
95756482 | 831 | rqstp->rq_prot = IPPROTO_UDP; |
27459f09 | 832 | |
7a37f578 N |
833 | if (cmh->cmsg_level != IPPROTO_IP || |
834 | cmh->cmsg_type != IP_PKTINFO) { | |
835 | if (net_ratelimit()) | |
836 | printk("rpcsvc: received unknown control message:" | |
837 | "%d/%d\n", | |
838 | cmh->cmsg_level, cmh->cmsg_type); | |
839 | skb_free_datagram(svsk->sk_sk, skb); | |
840 | return 0; | |
841 | } | |
842 | svc_udp_get_dest_address(rqstp, cmh); | |
1da177e4 LT |
843 | |
844 | if (skb_is_nonlinear(skb)) { | |
845 | /* we have to copy */ | |
846 | local_bh_disable(); | |
847 | if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) { | |
848 | local_bh_enable(); | |
849 | /* checksum error */ | |
850 | skb_free_datagram(svsk->sk_sk, skb); | |
851 | return 0; | |
852 | } | |
853 | local_bh_enable(); | |
cca5172a | 854 | skb_free_datagram(svsk->sk_sk, skb); |
1da177e4 LT |
855 | } else { |
856 | /* we can use it in-place */ | |
857 | rqstp->rq_arg.head[0].iov_base = skb->data + sizeof(struct udphdr); | |
858 | rqstp->rq_arg.head[0].iov_len = len; | |
fb286bb2 HX |
859 | if (skb_checksum_complete(skb)) { |
860 | skb_free_datagram(svsk->sk_sk, skb); | |
861 | return 0; | |
1da177e4 | 862 | } |
5148bf4e | 863 | rqstp->rq_xprt_ctxt = skb; |
1da177e4 LT |
864 | } |
865 | ||
866 | rqstp->rq_arg.page_base = 0; | |
867 | if (len <= rqstp->rq_arg.head[0].iov_len) { | |
868 | rqstp->rq_arg.head[0].iov_len = len; | |
869 | rqstp->rq_arg.page_len = 0; | |
44524359 | 870 | rqstp->rq_respages = rqstp->rq_pages+1; |
1da177e4 LT |
871 | } else { |
872 | rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len; | |
44524359 | 873 | rqstp->rq_respages = rqstp->rq_pages + 1 + |
172589cc | 874 | DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE); |
1da177e4 LT |
875 | } |
876 | ||
877 | if (serv->sv_stats) | |
878 | serv->sv_stats->netudpcnt++; | |
879 | ||
880 | return len; | |
881 | } | |
882 | ||
883 | static int | |
884 | svc_udp_sendto(struct svc_rqst *rqstp) | |
885 | { | |
886 | int error; | |
887 | ||
888 | error = svc_sendto(rqstp, &rqstp->rq_res); | |
889 | if (error == -ECONNREFUSED) | |
890 | /* ICMP error on earlier request. */ | |
891 | error = svc_sendto(rqstp, &rqstp->rq_res); | |
892 | ||
893 | return error; | |
894 | } | |
895 | ||
360d8738 | 896 | static struct svc_xprt_ops svc_udp_ops = { |
5d137990 TT |
897 | .xpo_recvfrom = svc_udp_recvfrom, |
898 | .xpo_sendto = svc_udp_sendto, | |
5148bf4e | 899 | .xpo_release_rqst = svc_release_skb, |
755cceab TT |
900 | .xpo_detach = svc_sock_detach, |
901 | .xpo_free = svc_sock_free, | |
360d8738 TT |
902 | }; |
903 | ||
904 | static struct svc_xprt_class svc_udp_class = { | |
905 | .xcl_name = "udp", | |
906 | .xcl_ops = &svc_udp_ops, | |
49023155 | 907 | .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP, |
360d8738 TT |
908 | }; |
909 | ||
1da177e4 LT |
910 | static void |
911 | svc_udp_init(struct svc_sock *svsk) | |
912 | { | |
7a37f578 N |
913 | int one = 1; |
914 | mm_segment_t oldfs; | |
915 | ||
360d8738 | 916 | svc_xprt_init(&svc_udp_class, &svsk->sk_xprt); |
1da177e4 LT |
917 | svsk->sk_sk->sk_data_ready = svc_udp_data_ready; |
918 | svsk->sk_sk->sk_write_space = svc_write_space; | |
1da177e4 LT |
919 | |
920 | /* initialise setting must have enough space to | |
cca5172a | 921 | * receive and respond to one request. |
1da177e4 LT |
922 | * svc_udp_recvfrom will re-adjust if necessary |
923 | */ | |
924 | svc_sock_setbufsize(svsk->sk_sock, | |
c6b0a9f8 N |
925 | 3 * svsk->sk_server->sv_max_mesg, |
926 | 3 * svsk->sk_server->sv_max_mesg); | |
1da177e4 LT |
927 | |
928 | set_bit(SK_DATA, &svsk->sk_flags); /* might have come in before data_ready set up */ | |
929 | set_bit(SK_CHNGBUF, &svsk->sk_flags); | |
7a37f578 N |
930 | |
931 | oldfs = get_fs(); | |
932 | set_fs(KERNEL_DS); | |
933 | /* make sure we get destination address info */ | |
934 | svsk->sk_sock->ops->setsockopt(svsk->sk_sock, IPPROTO_IP, IP_PKTINFO, | |
935 | (char __user *)&one, sizeof(one)); | |
936 | set_fs(oldfs); | |
1da177e4 LT |
937 | } |
938 | ||
939 | /* | |
940 | * A data_ready event on a listening socket means there's a connection | |
941 | * pending. Do not use state_change as a substitute for it. | |
942 | */ | |
943 | static void | |
944 | svc_tcp_listen_data_ready(struct sock *sk, int count_unused) | |
945 | { | |
939bb7ef | 946 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 LT |
947 | |
948 | dprintk("svc: socket %p TCP (listen) state change %d\n", | |
939bb7ef | 949 | sk, sk->sk_state); |
1da177e4 | 950 | |
939bb7ef NB |
951 | /* |
952 | * This callback may called twice when a new connection | |
953 | * is established as a child socket inherits everything | |
954 | * from a parent LISTEN socket. | |
955 | * 1) data_ready method of the parent socket will be called | |
956 | * when one of child sockets become ESTABLISHED. | |
957 | * 2) data_ready method of the child socket may be called | |
958 | * when it receives data before the socket is accepted. | |
959 | * In case of 2, we should ignore it silently. | |
960 | */ | |
961 | if (sk->sk_state == TCP_LISTEN) { | |
962 | if (svsk) { | |
963 | set_bit(SK_CONN, &svsk->sk_flags); | |
964 | svc_sock_enqueue(svsk); | |
965 | } else | |
966 | printk("svc: socket %p: no user data\n", sk); | |
1da177e4 | 967 | } |
939bb7ef | 968 | |
1da177e4 LT |
969 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
970 | wake_up_interruptible_all(sk->sk_sleep); | |
971 | } | |
972 | ||
973 | /* | |
974 | * A state change on a connected socket means it's dying or dead. | |
975 | */ | |
976 | static void | |
977 | svc_tcp_state_change(struct sock *sk) | |
978 | { | |
939bb7ef | 979 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 LT |
980 | |
981 | dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n", | |
939bb7ef | 982 | sk, sk->sk_state, sk->sk_user_data); |
1da177e4 | 983 | |
939bb7ef | 984 | if (!svsk) |
1da177e4 | 985 | printk("svc: socket %p: no user data\n", sk); |
939bb7ef NB |
986 | else { |
987 | set_bit(SK_CLOSE, &svsk->sk_flags); | |
988 | svc_sock_enqueue(svsk); | |
1da177e4 | 989 | } |
1da177e4 LT |
990 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
991 | wake_up_interruptible_all(sk->sk_sleep); | |
992 | } | |
993 | ||
994 | static void | |
995 | svc_tcp_data_ready(struct sock *sk, int count) | |
996 | { | |
939bb7ef | 997 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 LT |
998 | |
999 | dprintk("svc: socket %p TCP data ready (svsk %p)\n", | |
939bb7ef NB |
1000 | sk, sk->sk_user_data); |
1001 | if (svsk) { | |
1002 | set_bit(SK_DATA, &svsk->sk_flags); | |
1003 | svc_sock_enqueue(svsk); | |
1004 | } | |
1da177e4 LT |
1005 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
1006 | wake_up_interruptible(sk->sk_sleep); | |
1007 | } | |
1008 | ||
bcdb81ae CL |
1009 | static inline int svc_port_is_privileged(struct sockaddr *sin) |
1010 | { | |
1011 | switch (sin->sa_family) { | |
1012 | case AF_INET: | |
1013 | return ntohs(((struct sockaddr_in *)sin)->sin_port) | |
1014 | < PROT_SOCK; | |
bcdb81ae CL |
1015 | case AF_INET6: |
1016 | return ntohs(((struct sockaddr_in6 *)sin)->sin6_port) | |
1017 | < PROT_SOCK; | |
bcdb81ae CL |
1018 | default: |
1019 | return 0; | |
1020 | } | |
1021 | } | |
1022 | ||
1da177e4 LT |
1023 | /* |
1024 | * Accept a TCP connection | |
1025 | */ | |
1026 | static void | |
1027 | svc_tcp_accept(struct svc_sock *svsk) | |
1028 | { | |
cdd88b9f AM |
1029 | struct sockaddr_storage addr; |
1030 | struct sockaddr *sin = (struct sockaddr *) &addr; | |
1da177e4 LT |
1031 | struct svc_serv *serv = svsk->sk_server; |
1032 | struct socket *sock = svsk->sk_sock; | |
1033 | struct socket *newsock; | |
1da177e4 LT |
1034 | struct svc_sock *newsvsk; |
1035 | int err, slen; | |
ad06e4bd | 1036 | char buf[RPC_MAX_ADDRBUFLEN]; |
1da177e4 LT |
1037 | |
1038 | dprintk("svc: tcp_accept %p sock %p\n", svsk, sock); | |
1039 | if (!sock) | |
1040 | return; | |
1041 | ||
e6242e92 SS |
1042 | clear_bit(SK_CONN, &svsk->sk_flags); |
1043 | err = kernel_accept(sock, &newsock, O_NONBLOCK); | |
1044 | if (err < 0) { | |
1da177e4 LT |
1045 | if (err == -ENOMEM) |
1046 | printk(KERN_WARNING "%s: no more sockets!\n", | |
1047 | serv->sv_name); | |
e6242e92 | 1048 | else if (err != -EAGAIN && net_ratelimit()) |
1da177e4 LT |
1049 | printk(KERN_WARNING "%s: accept failed (err %d)!\n", |
1050 | serv->sv_name, -err); | |
e6242e92 | 1051 | return; |
1da177e4 | 1052 | } |
e6242e92 | 1053 | |
1da177e4 LT |
1054 | set_bit(SK_CONN, &svsk->sk_flags); |
1055 | svc_sock_enqueue(svsk); | |
1056 | ||
cdd88b9f | 1057 | err = kernel_getpeername(newsock, sin, &slen); |
1da177e4 LT |
1058 | if (err < 0) { |
1059 | if (net_ratelimit()) | |
1060 | printk(KERN_WARNING "%s: peername failed (err %d)!\n", | |
1061 | serv->sv_name, -err); | |
1062 | goto failed; /* aborted connection or whatever */ | |
1063 | } | |
1064 | ||
1065 | /* Ideally, we would want to reject connections from unauthorized | |
ad06e4bd CL |
1066 | * hosts here, but when we get encryption, the IP of the host won't |
1067 | * tell us anything. For now just warn about unpriv connections. | |
1da177e4 | 1068 | */ |
cdd88b9f | 1069 | if (!svc_port_is_privileged(sin)) { |
1da177e4 | 1070 | dprintk(KERN_WARNING |
ad06e4bd | 1071 | "%s: connect from unprivileged port: %s\n", |
cca5172a | 1072 | serv->sv_name, |
cdd88b9f | 1073 | __svc_print_addr(sin, buf, sizeof(buf))); |
1da177e4 | 1074 | } |
ad06e4bd | 1075 | dprintk("%s: connect from %s\n", serv->sv_name, |
cdd88b9f | 1076 | __svc_print_addr(sin, buf, sizeof(buf))); |
1da177e4 LT |
1077 | |
1078 | /* make sure that a write doesn't block forever when | |
1079 | * low on memory | |
1080 | */ | |
1081 | newsock->sk->sk_sndtimeo = HZ*30; | |
1082 | ||
6b174337 CL |
1083 | if (!(newsvsk = svc_setup_socket(serv, newsock, &err, |
1084 | (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY)))) | |
1da177e4 | 1085 | goto failed; |
cdd88b9f | 1086 | memcpy(&newsvsk->sk_remote, sin, slen); |
067d7817 | 1087 | newsvsk->sk_remotelen = slen; |
a9747692 FM |
1088 | err = kernel_getsockname(newsock, sin, &slen); |
1089 | if (unlikely(err < 0)) { | |
1090 | dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err); | |
1091 | slen = offsetof(struct sockaddr, sa_data); | |
1092 | } | |
1093 | memcpy(&newsvsk->sk_local, sin, slen); | |
067d7817 | 1094 | |
e79eff1f | 1095 | svc_sock_received(newsvsk); |
1da177e4 LT |
1096 | |
1097 | /* make sure that we don't have too many active connections. | |
1098 | * If we have, something must be dropped. | |
1099 | * | |
1100 | * There's no point in trying to do random drop here for | |
1101 | * DoS prevention. The NFS clients does 1 reconnect in 15 | |
1102 | * seconds. An attacker can easily beat that. | |
1103 | * | |
1104 | * The only somewhat efficient mechanism would be if drop | |
1105 | * old connections from the same IP first. But right now | |
1106 | * we don't even record the client IP in svc_sock. | |
1107 | */ | |
1108 | if (serv->sv_tmpcnt > (serv->sv_nrthreads+3)*20) { | |
1109 | struct svc_sock *svsk = NULL; | |
1110 | spin_lock_bh(&serv->sv_lock); | |
1111 | if (!list_empty(&serv->sv_tempsocks)) { | |
1112 | if (net_ratelimit()) { | |
1113 | /* Try to help the admin */ | |
1114 | printk(KERN_NOTICE "%s: too many open TCP " | |
1115 | "sockets, consider increasing the " | |
1116 | "number of nfsd threads\n", | |
1117 | serv->sv_name); | |
ad06e4bd CL |
1118 | printk(KERN_NOTICE |
1119 | "%s: last TCP connect from %s\n", | |
9db619e6 WW |
1120 | serv->sv_name, __svc_print_addr(sin, |
1121 | buf, sizeof(buf))); | |
1da177e4 LT |
1122 | } |
1123 | /* | |
1124 | * Always select the oldest socket. It's not fair, | |
1125 | * but so is life | |
1126 | */ | |
1127 | svsk = list_entry(serv->sv_tempsocks.prev, | |
1128 | struct svc_sock, | |
1129 | sk_list); | |
1130 | set_bit(SK_CLOSE, &svsk->sk_flags); | |
c45c357d | 1131 | atomic_inc(&svsk->sk_inuse); |
1da177e4 LT |
1132 | } |
1133 | spin_unlock_bh(&serv->sv_lock); | |
1134 | ||
1135 | if (svsk) { | |
1136 | svc_sock_enqueue(svsk); | |
1137 | svc_sock_put(svsk); | |
1138 | } | |
1139 | ||
1140 | } | |
1141 | ||
1142 | if (serv->sv_stats) | |
1143 | serv->sv_stats->nettcpconn++; | |
1144 | ||
1145 | return; | |
1146 | ||
1147 | failed: | |
1148 | sock_release(newsock); | |
1149 | return; | |
1150 | } | |
1151 | ||
1152 | /* | |
1153 | * Receive data from a TCP socket. | |
1154 | */ | |
1155 | static int | |
1156 | svc_tcp_recvfrom(struct svc_rqst *rqstp) | |
1157 | { | |
1158 | struct svc_sock *svsk = rqstp->rq_sock; | |
1159 | struct svc_serv *serv = svsk->sk_server; | |
1160 | int len; | |
3cc03b16 | 1161 | struct kvec *vec; |
1da177e4 LT |
1162 | int pnum, vlen; |
1163 | ||
1164 | dprintk("svc: tcp_recv %p data %d conn %d close %d\n", | |
1165 | svsk, test_bit(SK_DATA, &svsk->sk_flags), | |
1166 | test_bit(SK_CONN, &svsk->sk_flags), | |
1167 | test_bit(SK_CLOSE, &svsk->sk_flags)); | |
1168 | ||
1169 | if ((rqstp->rq_deferred = svc_deferred_dequeue(svsk))) { | |
1170 | svc_sock_received(svsk); | |
1171 | return svc_deferred_recv(rqstp); | |
1172 | } | |
1173 | ||
1174 | if (test_bit(SK_CLOSE, &svsk->sk_flags)) { | |
1175 | svc_delete_socket(svsk); | |
1176 | return 0; | |
1177 | } | |
1178 | ||
1a047060 | 1179 | if (svsk->sk_sk->sk_state == TCP_LISTEN) { |
1da177e4 LT |
1180 | svc_tcp_accept(svsk); |
1181 | svc_sock_received(svsk); | |
1182 | return 0; | |
1183 | } | |
1184 | ||
1185 | if (test_and_clear_bit(SK_CHNGBUF, &svsk->sk_flags)) | |
1186 | /* sndbuf needs to have room for one request | |
1187 | * per thread, otherwise we can stall even when the | |
1188 | * network isn't a bottleneck. | |
3262c816 GB |
1189 | * |
1190 | * We count all threads rather than threads in a | |
1191 | * particular pool, which provides an upper bound | |
1192 | * on the number of threads which will access the socket. | |
1193 | * | |
1da177e4 | 1194 | * rcvbuf just needs to be able to hold a few requests. |
cca5172a | 1195 | * Normally they will be removed from the queue |
1da177e4 LT |
1196 | * as soon a a complete request arrives. |
1197 | */ | |
1198 | svc_sock_setbufsize(svsk->sk_sock, | |
c6b0a9f8 N |
1199 | (serv->sv_nrthreads+3) * serv->sv_max_mesg, |
1200 | 3 * serv->sv_max_mesg); | |
1da177e4 LT |
1201 | |
1202 | clear_bit(SK_DATA, &svsk->sk_flags); | |
1203 | ||
1204 | /* Receive data. If we haven't got the record length yet, get | |
1205 | * the next four bytes. Otherwise try to gobble up as much as | |
1206 | * possible up to the complete record length. | |
1207 | */ | |
1208 | if (svsk->sk_tcplen < 4) { | |
1209 | unsigned long want = 4 - svsk->sk_tcplen; | |
1210 | struct kvec iov; | |
1211 | ||
1212 | iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen; | |
1213 | iov.iov_len = want; | |
1214 | if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0) | |
1215 | goto error; | |
1216 | svsk->sk_tcplen += len; | |
1217 | ||
1218 | if (len < want) { | |
1219 | dprintk("svc: short recvfrom while reading record length (%d of %lu)\n", | |
cca5172a | 1220 | len, want); |
1da177e4 LT |
1221 | svc_sock_received(svsk); |
1222 | return -EAGAIN; /* record header not complete */ | |
1223 | } | |
1224 | ||
1225 | svsk->sk_reclen = ntohl(svsk->sk_reclen); | |
1226 | if (!(svsk->sk_reclen & 0x80000000)) { | |
1227 | /* FIXME: technically, a record can be fragmented, | |
1228 | * and non-terminal fragments will not have the top | |
1229 | * bit set in the fragment length header. | |
1230 | * But apparently no known nfs clients send fragmented | |
1231 | * records. */ | |
34e9a63b N |
1232 | if (net_ratelimit()) |
1233 | printk(KERN_NOTICE "RPC: bad TCP reclen 0x%08lx" | |
1234 | " (non-terminal)\n", | |
1235 | (unsigned long) svsk->sk_reclen); | |
1da177e4 LT |
1236 | goto err_delete; |
1237 | } | |
1238 | svsk->sk_reclen &= 0x7fffffff; | |
1239 | dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen); | |
c6b0a9f8 | 1240 | if (svsk->sk_reclen > serv->sv_max_mesg) { |
34e9a63b N |
1241 | if (net_ratelimit()) |
1242 | printk(KERN_NOTICE "RPC: bad TCP reclen 0x%08lx" | |
1243 | " (large)\n", | |
1244 | (unsigned long) svsk->sk_reclen); | |
1da177e4 LT |
1245 | goto err_delete; |
1246 | } | |
1247 | } | |
1248 | ||
1249 | /* Check whether enough data is available */ | |
1250 | len = svc_recv_available(svsk); | |
1251 | if (len < 0) | |
1252 | goto error; | |
1253 | ||
1254 | if (len < svsk->sk_reclen) { | |
1255 | dprintk("svc: incomplete TCP record (%d of %d)\n", | |
1256 | len, svsk->sk_reclen); | |
1257 | svc_sock_received(svsk); | |
1258 | return -EAGAIN; /* record not complete */ | |
1259 | } | |
1260 | len = svsk->sk_reclen; | |
1261 | set_bit(SK_DATA, &svsk->sk_flags); | |
1262 | ||
3cc03b16 | 1263 | vec = rqstp->rq_vec; |
1da177e4 LT |
1264 | vec[0] = rqstp->rq_arg.head[0]; |
1265 | vlen = PAGE_SIZE; | |
1266 | pnum = 1; | |
1267 | while (vlen < len) { | |
44524359 | 1268 | vec[pnum].iov_base = page_address(rqstp->rq_pages[pnum]); |
1da177e4 LT |
1269 | vec[pnum].iov_len = PAGE_SIZE; |
1270 | pnum++; | |
1271 | vlen += PAGE_SIZE; | |
1272 | } | |
44524359 | 1273 | rqstp->rq_respages = &rqstp->rq_pages[pnum]; |
1da177e4 LT |
1274 | |
1275 | /* Now receive data */ | |
1276 | len = svc_recvfrom(rqstp, vec, pnum, len); | |
1277 | if (len < 0) | |
1278 | goto error; | |
1279 | ||
1280 | dprintk("svc: TCP complete record (%d bytes)\n", len); | |
1281 | rqstp->rq_arg.len = len; | |
1282 | rqstp->rq_arg.page_base = 0; | |
1283 | if (len <= rqstp->rq_arg.head[0].iov_len) { | |
1284 | rqstp->rq_arg.head[0].iov_len = len; | |
1285 | rqstp->rq_arg.page_len = 0; | |
1286 | } else { | |
1287 | rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len; | |
1288 | } | |
1289 | ||
5148bf4e | 1290 | rqstp->rq_xprt_ctxt = NULL; |
1da177e4 LT |
1291 | rqstp->rq_prot = IPPROTO_TCP; |
1292 | ||
1293 | /* Reset TCP read info */ | |
1294 | svsk->sk_reclen = 0; | |
1295 | svsk->sk_tcplen = 0; | |
1296 | ||
1297 | svc_sock_received(svsk); | |
1298 | if (serv->sv_stats) | |
1299 | serv->sv_stats->nettcpcnt++; | |
1300 | ||
1301 | return len; | |
1302 | ||
1303 | err_delete: | |
1304 | svc_delete_socket(svsk); | |
1305 | return -EAGAIN; | |
1306 | ||
1307 | error: | |
1308 | if (len == -EAGAIN) { | |
1309 | dprintk("RPC: TCP recvfrom got EAGAIN\n"); | |
1310 | svc_sock_received(svsk); | |
1311 | } else { | |
1312 | printk(KERN_NOTICE "%s: recvfrom returned errno %d\n", | |
1313 | svsk->sk_server->sv_name, -len); | |
93fbf1a5 | 1314 | goto err_delete; |
1da177e4 LT |
1315 | } |
1316 | ||
1317 | return len; | |
1318 | } | |
1319 | ||
1320 | /* | |
1321 | * Send out data on TCP socket. | |
1322 | */ | |
1323 | static int | |
1324 | svc_tcp_sendto(struct svc_rqst *rqstp) | |
1325 | { | |
1326 | struct xdr_buf *xbufp = &rqstp->rq_res; | |
1327 | int sent; | |
d8ed029d | 1328 | __be32 reclen; |
1da177e4 LT |
1329 | |
1330 | /* Set up the first element of the reply kvec. | |
1331 | * Any other kvecs that may be in use have been taken | |
1332 | * care of by the server implementation itself. | |
1333 | */ | |
1334 | reclen = htonl(0x80000000|((xbufp->len ) - 4)); | |
1335 | memcpy(xbufp->head[0].iov_base, &reclen, 4); | |
1336 | ||
1337 | if (test_bit(SK_DEAD, &rqstp->rq_sock->sk_flags)) | |
1338 | return -ENOTCONN; | |
1339 | ||
1340 | sent = svc_sendto(rqstp, &rqstp->rq_res); | |
1341 | if (sent != xbufp->len) { | |
1342 | printk(KERN_NOTICE "rpc-srv/tcp: %s: %s %d when sending %d bytes - shutting down socket\n", | |
1343 | rqstp->rq_sock->sk_server->sv_name, | |
1344 | (sent<0)?"got error":"sent only", | |
1345 | sent, xbufp->len); | |
aaf68cfb N |
1346 | set_bit(SK_CLOSE, &rqstp->rq_sock->sk_flags); |
1347 | svc_sock_enqueue(rqstp->rq_sock); | |
1da177e4 LT |
1348 | sent = -EAGAIN; |
1349 | } | |
1350 | return sent; | |
1351 | } | |
1352 | ||
360d8738 | 1353 | static struct svc_xprt_ops svc_tcp_ops = { |
5d137990 TT |
1354 | .xpo_recvfrom = svc_tcp_recvfrom, |
1355 | .xpo_sendto = svc_tcp_sendto, | |
5148bf4e | 1356 | .xpo_release_rqst = svc_release_skb, |
755cceab TT |
1357 | .xpo_detach = svc_sock_detach, |
1358 | .xpo_free = svc_sock_free, | |
360d8738 TT |
1359 | }; |
1360 | ||
1361 | static struct svc_xprt_class svc_tcp_class = { | |
1362 | .xcl_name = "tcp", | |
1363 | .xcl_ops = &svc_tcp_ops, | |
49023155 | 1364 | .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP, |
360d8738 TT |
1365 | }; |
1366 | ||
1367 | void svc_init_xprt_sock(void) | |
1368 | { | |
1369 | svc_reg_xprt_class(&svc_tcp_class); | |
1370 | svc_reg_xprt_class(&svc_udp_class); | |
1371 | } | |
1372 | ||
1373 | void svc_cleanup_xprt_sock(void) | |
1374 | { | |
1375 | svc_unreg_xprt_class(&svc_tcp_class); | |
1376 | svc_unreg_xprt_class(&svc_udp_class); | |
1377 | } | |
1378 | ||
1da177e4 LT |
1379 | static void |
1380 | svc_tcp_init(struct svc_sock *svsk) | |
1381 | { | |
1382 | struct sock *sk = svsk->sk_sk; | |
1383 | struct tcp_sock *tp = tcp_sk(sk); | |
1384 | ||
360d8738 | 1385 | svc_xprt_init(&svc_tcp_class, &svsk->sk_xprt); |
1da177e4 LT |
1386 | |
1387 | if (sk->sk_state == TCP_LISTEN) { | |
1388 | dprintk("setting up TCP socket for listening\n"); | |
1389 | sk->sk_data_ready = svc_tcp_listen_data_ready; | |
1390 | set_bit(SK_CONN, &svsk->sk_flags); | |
1391 | } else { | |
1392 | dprintk("setting up TCP socket for reading\n"); | |
1393 | sk->sk_state_change = svc_tcp_state_change; | |
1394 | sk->sk_data_ready = svc_tcp_data_ready; | |
1395 | sk->sk_write_space = svc_write_space; | |
1396 | ||
1397 | svsk->sk_reclen = 0; | |
1398 | svsk->sk_tcplen = 0; | |
1399 | ||
1400 | tp->nonagle = 1; /* disable Nagle's algorithm */ | |
1401 | ||
1402 | /* initialise setting must have enough space to | |
cca5172a | 1403 | * receive and respond to one request. |
1da177e4 LT |
1404 | * svc_tcp_recvfrom will re-adjust if necessary |
1405 | */ | |
1406 | svc_sock_setbufsize(svsk->sk_sock, | |
c6b0a9f8 N |
1407 | 3 * svsk->sk_server->sv_max_mesg, |
1408 | 3 * svsk->sk_server->sv_max_mesg); | |
1da177e4 LT |
1409 | |
1410 | set_bit(SK_CHNGBUF, &svsk->sk_flags); | |
1411 | set_bit(SK_DATA, &svsk->sk_flags); | |
cca5172a | 1412 | if (sk->sk_state != TCP_ESTABLISHED) |
1da177e4 LT |
1413 | set_bit(SK_CLOSE, &svsk->sk_flags); |
1414 | } | |
1415 | } | |
1416 | ||
1417 | void | |
1418 | svc_sock_update_bufs(struct svc_serv *serv) | |
1419 | { | |
1420 | /* | |
1421 | * The number of server threads has changed. Update | |
1422 | * rcvbuf and sndbuf accordingly on all sockets | |
1423 | */ | |
1424 | struct list_head *le; | |
1425 | ||
1426 | spin_lock_bh(&serv->sv_lock); | |
1427 | list_for_each(le, &serv->sv_permsocks) { | |
cca5172a | 1428 | struct svc_sock *svsk = |
1da177e4 LT |
1429 | list_entry(le, struct svc_sock, sk_list); |
1430 | set_bit(SK_CHNGBUF, &svsk->sk_flags); | |
1431 | } | |
1432 | list_for_each(le, &serv->sv_tempsocks) { | |
1433 | struct svc_sock *svsk = | |
1434 | list_entry(le, struct svc_sock, sk_list); | |
1435 | set_bit(SK_CHNGBUF, &svsk->sk_flags); | |
1436 | } | |
1437 | spin_unlock_bh(&serv->sv_lock); | |
1438 | } | |
1439 | ||
1440 | /* | |
3262c816 GB |
1441 | * Receive the next request on any socket. This code is carefully |
1442 | * organised not to touch any cachelines in the shared svc_serv | |
1443 | * structure, only cachelines in the local svc_pool. | |
1da177e4 LT |
1444 | */ |
1445 | int | |
6fb2b47f | 1446 | svc_recv(struct svc_rqst *rqstp, long timeout) |
1da177e4 | 1447 | { |
27459f09 | 1448 | struct svc_sock *svsk = NULL; |
6fb2b47f | 1449 | struct svc_serv *serv = rqstp->rq_server; |
3262c816 | 1450 | struct svc_pool *pool = rqstp->rq_pool; |
44524359 | 1451 | int len, i; |
1da177e4 LT |
1452 | int pages; |
1453 | struct xdr_buf *arg; | |
1454 | DECLARE_WAITQUEUE(wait, current); | |
1455 | ||
1456 | dprintk("svc: server %p waiting for data (to = %ld)\n", | |
1457 | rqstp, timeout); | |
1458 | ||
1459 | if (rqstp->rq_sock) | |
cca5172a | 1460 | printk(KERN_ERR |
1da177e4 LT |
1461 | "svc_recv: service %p, socket not NULL!\n", |
1462 | rqstp); | |
1463 | if (waitqueue_active(&rqstp->rq_wait)) | |
cca5172a | 1464 | printk(KERN_ERR |
1da177e4 LT |
1465 | "svc_recv: service %p, wait queue active!\n", |
1466 | rqstp); | |
1467 | ||
1da177e4 LT |
1468 | |
1469 | /* now allocate needed pages. If we get a failure, sleep briefly */ | |
c6b0a9f8 | 1470 | pages = (serv->sv_max_mesg + PAGE_SIZE) / PAGE_SIZE; |
44524359 N |
1471 | for (i=0; i < pages ; i++) |
1472 | while (rqstp->rq_pages[i] == NULL) { | |
1473 | struct page *p = alloc_page(GFP_KERNEL); | |
1474 | if (!p) | |
1475 | schedule_timeout_uninterruptible(msecs_to_jiffies(500)); | |
1476 | rqstp->rq_pages[i] = p; | |
1da177e4 | 1477 | } |
250f3915 N |
1478 | rqstp->rq_pages[i++] = NULL; /* this might be seen in nfs_read_actor */ |
1479 | BUG_ON(pages >= RPCSVC_MAXPAGES); | |
1da177e4 LT |
1480 | |
1481 | /* Make arg->head point to first page and arg->pages point to rest */ | |
1482 | arg = &rqstp->rq_arg; | |
44524359 | 1483 | arg->head[0].iov_base = page_address(rqstp->rq_pages[0]); |
1da177e4 | 1484 | arg->head[0].iov_len = PAGE_SIZE; |
44524359 | 1485 | arg->pages = rqstp->rq_pages + 1; |
1da177e4 LT |
1486 | arg->page_base = 0; |
1487 | /* save at least one page for response */ | |
1488 | arg->page_len = (pages-2)*PAGE_SIZE; | |
1489 | arg->len = (pages-1)*PAGE_SIZE; | |
1490 | arg->tail[0].iov_len = 0; | |
3e1d1d28 CL |
1491 | |
1492 | try_to_freeze(); | |
1887b935 | 1493 | cond_resched(); |
1da177e4 LT |
1494 | if (signalled()) |
1495 | return -EINTR; | |
1496 | ||
3262c816 GB |
1497 | spin_lock_bh(&pool->sp_lock); |
1498 | if ((svsk = svc_sock_dequeue(pool)) != NULL) { | |
1da177e4 | 1499 | rqstp->rq_sock = svsk; |
c45c357d | 1500 | atomic_inc(&svsk->sk_inuse); |
c6b0a9f8 | 1501 | rqstp->rq_reserved = serv->sv_max_mesg; |
5685f0fa | 1502 | atomic_add(rqstp->rq_reserved, &svsk->sk_reserved); |
1da177e4 LT |
1503 | } else { |
1504 | /* No data pending. Go to sleep */ | |
3262c816 | 1505 | svc_thread_enqueue(pool, rqstp); |
1da177e4 LT |
1506 | |
1507 | /* | |
1508 | * We have to be able to interrupt this wait | |
1509 | * to bring down the daemons ... | |
1510 | */ | |
1511 | set_current_state(TASK_INTERRUPTIBLE); | |
1512 | add_wait_queue(&rqstp->rq_wait, &wait); | |
3262c816 | 1513 | spin_unlock_bh(&pool->sp_lock); |
1da177e4 LT |
1514 | |
1515 | schedule_timeout(timeout); | |
1516 | ||
3e1d1d28 | 1517 | try_to_freeze(); |
1da177e4 | 1518 | |
3262c816 | 1519 | spin_lock_bh(&pool->sp_lock); |
1da177e4 LT |
1520 | remove_wait_queue(&rqstp->rq_wait, &wait); |
1521 | ||
1522 | if (!(svsk = rqstp->rq_sock)) { | |
3262c816 GB |
1523 | svc_thread_dequeue(pool, rqstp); |
1524 | spin_unlock_bh(&pool->sp_lock); | |
1da177e4 LT |
1525 | dprintk("svc: server %p, no data yet\n", rqstp); |
1526 | return signalled()? -EINTR : -EAGAIN; | |
1527 | } | |
1528 | } | |
3262c816 | 1529 | spin_unlock_bh(&pool->sp_lock); |
1da177e4 | 1530 | |
3262c816 GB |
1531 | dprintk("svc: server %p, pool %u, socket %p, inuse=%d\n", |
1532 | rqstp, pool->sp_id, svsk, atomic_read(&svsk->sk_inuse)); | |
5d137990 | 1533 | len = svsk->sk_xprt.xpt_ops->xpo_recvfrom(rqstp); |
1da177e4 LT |
1534 | dprintk("svc: got len=%d\n", len); |
1535 | ||
1536 | /* No data, incomplete (TCP) read, or accept() */ | |
1537 | if (len == 0 || len == -EAGAIN) { | |
1538 | rqstp->rq_res.len = 0; | |
1539 | svc_sock_release(rqstp); | |
1540 | return -EAGAIN; | |
1541 | } | |
1542 | svsk->sk_lastrecv = get_seconds(); | |
36bdfc8b | 1543 | clear_bit(SK_OLD, &svsk->sk_flags); |
1da177e4 | 1544 | |
bcdb81ae | 1545 | rqstp->rq_secure = svc_port_is_privileged(svc_addr(rqstp)); |
1da177e4 LT |
1546 | rqstp->rq_chandle.defer = svc_defer; |
1547 | ||
1548 | if (serv->sv_stats) | |
1549 | serv->sv_stats->netcnt++; | |
1550 | return len; | |
1551 | } | |
1552 | ||
cca5172a | 1553 | /* |
1da177e4 LT |
1554 | * Drop request |
1555 | */ | |
1556 | void | |
1557 | svc_drop(struct svc_rqst *rqstp) | |
1558 | { | |
1559 | dprintk("svc: socket %p dropped request\n", rqstp->rq_sock); | |
1560 | svc_sock_release(rqstp); | |
1561 | } | |
1562 | ||
1563 | /* | |
1564 | * Return reply to client. | |
1565 | */ | |
1566 | int | |
1567 | svc_send(struct svc_rqst *rqstp) | |
1568 | { | |
1569 | struct svc_sock *svsk; | |
1570 | int len; | |
1571 | struct xdr_buf *xb; | |
1572 | ||
1573 | if ((svsk = rqstp->rq_sock) == NULL) { | |
1574 | printk(KERN_WARNING "NULL socket pointer in %s:%d\n", | |
1575 | __FILE__, __LINE__); | |
1576 | return -EFAULT; | |
1577 | } | |
1578 | ||
1579 | /* release the receive skb before sending the reply */ | |
5148bf4e | 1580 | rqstp->rq_xprt->xpt_ops->xpo_release_rqst(rqstp); |
1da177e4 LT |
1581 | |
1582 | /* calculate over-all length */ | |
1583 | xb = & rqstp->rq_res; | |
1584 | xb->len = xb->head[0].iov_len + | |
1585 | xb->page_len + | |
1586 | xb->tail[0].iov_len; | |
1587 | ||
57b47a53 IM |
1588 | /* Grab svsk->sk_mutex to serialize outgoing data. */ |
1589 | mutex_lock(&svsk->sk_mutex); | |
1da177e4 LT |
1590 | if (test_bit(SK_DEAD, &svsk->sk_flags)) |
1591 | len = -ENOTCONN; | |
1592 | else | |
5d137990 | 1593 | len = svsk->sk_xprt.xpt_ops->xpo_sendto(rqstp); |
57b47a53 | 1594 | mutex_unlock(&svsk->sk_mutex); |
1da177e4 LT |
1595 | svc_sock_release(rqstp); |
1596 | ||
1597 | if (len == -ECONNREFUSED || len == -ENOTCONN || len == -EAGAIN) | |
1598 | return 0; | |
1599 | return len; | |
1600 | } | |
1601 | ||
36bdfc8b GB |
1602 | /* |
1603 | * Timer function to close old temporary sockets, using | |
1604 | * a mark-and-sweep algorithm. | |
1605 | */ | |
1606 | static void | |
1607 | svc_age_temp_sockets(unsigned long closure) | |
1608 | { | |
1609 | struct svc_serv *serv = (struct svc_serv *)closure; | |
1610 | struct svc_sock *svsk; | |
1611 | struct list_head *le, *next; | |
1612 | LIST_HEAD(to_be_aged); | |
1613 | ||
1614 | dprintk("svc_age_temp_sockets\n"); | |
1615 | ||
1616 | if (!spin_trylock_bh(&serv->sv_lock)) { | |
1617 | /* busy, try again 1 sec later */ | |
1618 | dprintk("svc_age_temp_sockets: busy\n"); | |
1619 | mod_timer(&serv->sv_temptimer, jiffies + HZ); | |
1620 | return; | |
1621 | } | |
1622 | ||
1623 | list_for_each_safe(le, next, &serv->sv_tempsocks) { | |
1624 | svsk = list_entry(le, struct svc_sock, sk_list); | |
1625 | ||
1626 | if (!test_and_set_bit(SK_OLD, &svsk->sk_flags)) | |
1627 | continue; | |
7a1fa065 | 1628 | if (atomic_read(&svsk->sk_inuse) > 1 || test_bit(SK_BUSY, &svsk->sk_flags)) |
36bdfc8b | 1629 | continue; |
c45c357d | 1630 | atomic_inc(&svsk->sk_inuse); |
36bdfc8b GB |
1631 | list_move(le, &to_be_aged); |
1632 | set_bit(SK_CLOSE, &svsk->sk_flags); | |
1633 | set_bit(SK_DETACHED, &svsk->sk_flags); | |
1634 | } | |
1635 | spin_unlock_bh(&serv->sv_lock); | |
1636 | ||
1637 | while (!list_empty(&to_be_aged)) { | |
1638 | le = to_be_aged.next; | |
1639 | /* fiddling the sk_list node is safe 'cos we're SK_DETACHED */ | |
1640 | list_del_init(le); | |
1641 | svsk = list_entry(le, struct svc_sock, sk_list); | |
1642 | ||
1643 | dprintk("queuing svsk %p for closing, %lu seconds old\n", | |
1644 | svsk, get_seconds() - svsk->sk_lastrecv); | |
1645 | ||
1646 | /* a thread will dequeue and close it soon */ | |
1647 | svc_sock_enqueue(svsk); | |
1648 | svc_sock_put(svsk); | |
1649 | } | |
1650 | ||
1651 | mod_timer(&serv->sv_temptimer, jiffies + svc_conn_age_period * HZ); | |
1652 | } | |
1653 | ||
1da177e4 LT |
1654 | /* |
1655 | * Initialize socket for RPC use and create svc_sock struct | |
1656 | * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF. | |
1657 | */ | |
6b174337 CL |
1658 | static struct svc_sock *svc_setup_socket(struct svc_serv *serv, |
1659 | struct socket *sock, | |
1660 | int *errp, int flags) | |
1da177e4 LT |
1661 | { |
1662 | struct svc_sock *svsk; | |
1663 | struct sock *inet; | |
6b174337 CL |
1664 | int pmap_register = !(flags & SVC_SOCK_ANONYMOUS); |
1665 | int is_temporary = flags & SVC_SOCK_TEMPORARY; | |
1da177e4 LT |
1666 | |
1667 | dprintk("svc: svc_setup_socket %p\n", sock); | |
0da974f4 | 1668 | if (!(svsk = kzalloc(sizeof(*svsk), GFP_KERNEL))) { |
1da177e4 LT |
1669 | *errp = -ENOMEM; |
1670 | return NULL; | |
1671 | } | |
1da177e4 LT |
1672 | |
1673 | inet = sock->sk; | |
1674 | ||
1675 | /* Register socket with portmapper */ | |
1676 | if (*errp >= 0 && pmap_register) | |
1677 | *errp = svc_register(serv, inet->sk_protocol, | |
1678 | ntohs(inet_sk(inet)->sport)); | |
1679 | ||
1680 | if (*errp < 0) { | |
1681 | kfree(svsk); | |
1682 | return NULL; | |
1683 | } | |
1684 | ||
1685 | set_bit(SK_BUSY, &svsk->sk_flags); | |
1686 | inet->sk_user_data = svsk; | |
1687 | svsk->sk_sock = sock; | |
1688 | svsk->sk_sk = inet; | |
1689 | svsk->sk_ostate = inet->sk_state_change; | |
1690 | svsk->sk_odata = inet->sk_data_ready; | |
1691 | svsk->sk_owspace = inet->sk_write_space; | |
1692 | svsk->sk_server = serv; | |
aaf68cfb | 1693 | atomic_set(&svsk->sk_inuse, 1); |
1da177e4 | 1694 | svsk->sk_lastrecv = get_seconds(); |
7ac1bea5 | 1695 | spin_lock_init(&svsk->sk_lock); |
1da177e4 LT |
1696 | INIT_LIST_HEAD(&svsk->sk_deferred); |
1697 | INIT_LIST_HEAD(&svsk->sk_ready); | |
57b47a53 | 1698 | mutex_init(&svsk->sk_mutex); |
1da177e4 LT |
1699 | |
1700 | /* Initialize the socket */ | |
1701 | if (sock->type == SOCK_DGRAM) | |
1702 | svc_udp_init(svsk); | |
1703 | else | |
1704 | svc_tcp_init(svsk); | |
1705 | ||
1706 | spin_lock_bh(&serv->sv_lock); | |
6b174337 | 1707 | if (is_temporary) { |
1da177e4 LT |
1708 | set_bit(SK_TEMP, &svsk->sk_flags); |
1709 | list_add(&svsk->sk_list, &serv->sv_tempsocks); | |
1710 | serv->sv_tmpcnt++; | |
36bdfc8b GB |
1711 | if (serv->sv_temptimer.function == NULL) { |
1712 | /* setup timer to age temp sockets */ | |
1713 | setup_timer(&serv->sv_temptimer, svc_age_temp_sockets, | |
1714 | (unsigned long)serv); | |
1715 | mod_timer(&serv->sv_temptimer, | |
1716 | jiffies + svc_conn_age_period * HZ); | |
1717 | } | |
1da177e4 LT |
1718 | } else { |
1719 | clear_bit(SK_TEMP, &svsk->sk_flags); | |
1720 | list_add(&svsk->sk_list, &serv->sv_permsocks); | |
1721 | } | |
1722 | spin_unlock_bh(&serv->sv_lock); | |
1723 | ||
1724 | dprintk("svc: svc_setup_socket created %p (inet %p)\n", | |
1725 | svsk, svsk->sk_sk); | |
1726 | ||
1da177e4 LT |
1727 | return svsk; |
1728 | } | |
1729 | ||
b41b66d6 N |
1730 | int svc_addsock(struct svc_serv *serv, |
1731 | int fd, | |
1732 | char *name_return, | |
1733 | int *proto) | |
1734 | { | |
1735 | int err = 0; | |
1736 | struct socket *so = sockfd_lookup(fd, &err); | |
1737 | struct svc_sock *svsk = NULL; | |
1738 | ||
1739 | if (!so) | |
1740 | return err; | |
1741 | if (so->sk->sk_family != AF_INET) | |
1742 | err = -EAFNOSUPPORT; | |
1743 | else if (so->sk->sk_protocol != IPPROTO_TCP && | |
1744 | so->sk->sk_protocol != IPPROTO_UDP) | |
1745 | err = -EPROTONOSUPPORT; | |
1746 | else if (so->state > SS_UNCONNECTED) | |
1747 | err = -EISCONN; | |
1748 | else { | |
6b174337 | 1749 | svsk = svc_setup_socket(serv, so, &err, SVC_SOCK_DEFAULTS); |
e79eff1f N |
1750 | if (svsk) { |
1751 | svc_sock_received(svsk); | |
b41b66d6 | 1752 | err = 0; |
e79eff1f | 1753 | } |
b41b66d6 N |
1754 | } |
1755 | if (err) { | |
1756 | sockfd_put(so); | |
1757 | return err; | |
1758 | } | |
1759 | if (proto) *proto = so->sk->sk_protocol; | |
1760 | return one_sock_name(name_return, svsk); | |
1761 | } | |
1762 | EXPORT_SYMBOL_GPL(svc_addsock); | |
1763 | ||
1da177e4 LT |
1764 | /* |
1765 | * Create socket for RPC service. | |
1766 | */ | |
6b174337 | 1767 | static int svc_create_socket(struct svc_serv *serv, int protocol, |
77f1f67a | 1768 | struct sockaddr *sin, int len, int flags) |
1da177e4 LT |
1769 | { |
1770 | struct svc_sock *svsk; | |
1771 | struct socket *sock; | |
1772 | int error; | |
1773 | int type; | |
ad06e4bd | 1774 | char buf[RPC_MAX_ADDRBUFLEN]; |
1da177e4 | 1775 | |
ad06e4bd CL |
1776 | dprintk("svc: svc_create_socket(%s, %d, %s)\n", |
1777 | serv->sv_program->pg_name, protocol, | |
77f1f67a | 1778 | __svc_print_addr(sin, buf, sizeof(buf))); |
1da177e4 LT |
1779 | |
1780 | if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) { | |
1781 | printk(KERN_WARNING "svc: only UDP and TCP " | |
1782 | "sockets supported\n"); | |
1783 | return -EINVAL; | |
1784 | } | |
1785 | type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM; | |
1786 | ||
77f1f67a CL |
1787 | error = sock_create_kern(sin->sa_family, type, protocol, &sock); |
1788 | if (error < 0) | |
1da177e4 LT |
1789 | return error; |
1790 | ||
ed07536e PZ |
1791 | svc_reclassify_socket(sock); |
1792 | ||
18114746 | 1793 | if (type == SOCK_STREAM) |
77f1f67a CL |
1794 | sock->sk->sk_reuse = 1; /* allow address reuse */ |
1795 | error = kernel_bind(sock, sin, len); | |
18114746 ES |
1796 | if (error < 0) |
1797 | goto bummer; | |
1da177e4 LT |
1798 | |
1799 | if (protocol == IPPROTO_TCP) { | |
e6242e92 | 1800 | if ((error = kernel_listen(sock, 64)) < 0) |
1da177e4 LT |
1801 | goto bummer; |
1802 | } | |
1803 | ||
e79eff1f N |
1804 | if ((svsk = svc_setup_socket(serv, sock, &error, flags)) != NULL) { |
1805 | svc_sock_received(svsk); | |
6b174337 | 1806 | return ntohs(inet_sk(svsk->sk_sk)->sport); |
e79eff1f | 1807 | } |
1da177e4 LT |
1808 | |
1809 | bummer: | |
1810 | dprintk("svc: svc_create_socket error = %d\n", -error); | |
1811 | sock_release(sock); | |
1812 | return error; | |
1813 | } | |
1814 | ||
755cceab TT |
1815 | /* |
1816 | * Detach the svc_sock from the socket so that no | |
1817 | * more callbacks occur. | |
1818 | */ | |
1819 | static void svc_sock_detach(struct svc_xprt *xprt) | |
1820 | { | |
1821 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
1822 | struct sock *sk = svsk->sk_sk; | |
1823 | ||
1824 | dprintk("svc: svc_sock_detach(%p)\n", svsk); | |
1825 | ||
1826 | /* put back the old socket callbacks */ | |
1827 | sk->sk_state_change = svsk->sk_ostate; | |
1828 | sk->sk_data_ready = svsk->sk_odata; | |
1829 | sk->sk_write_space = svsk->sk_owspace; | |
1830 | } | |
1831 | ||
1832 | /* | |
1833 | * Free the svc_sock's socket resources and the svc_sock itself. | |
1834 | */ | |
1835 | static void svc_sock_free(struct svc_xprt *xprt) | |
1836 | { | |
1837 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
1838 | dprintk("svc: svc_sock_free(%p)\n", svsk); | |
1839 | ||
1840 | if (svsk->sk_info_authunix != NULL) | |
1841 | svcauth_unix_info_release(svsk->sk_info_authunix); | |
1842 | if (svsk->sk_sock->file) | |
1843 | sockfd_put(svsk->sk_sock); | |
1844 | else | |
1845 | sock_release(svsk->sk_sock); | |
1846 | kfree(svsk); | |
1847 | } | |
1848 | ||
1da177e4 LT |
1849 | /* |
1850 | * Remove a dead socket | |
1851 | */ | |
aaf68cfb | 1852 | static void |
1da177e4 LT |
1853 | svc_delete_socket(struct svc_sock *svsk) |
1854 | { | |
1855 | struct svc_serv *serv; | |
1856 | struct sock *sk; | |
1857 | ||
1858 | dprintk("svc: svc_delete_socket(%p)\n", svsk); | |
1859 | ||
1860 | serv = svsk->sk_server; | |
1861 | sk = svsk->sk_sk; | |
1862 | ||
755cceab | 1863 | svsk->sk_xprt.xpt_ops->xpo_detach(&svsk->sk_xprt); |
1da177e4 LT |
1864 | |
1865 | spin_lock_bh(&serv->sv_lock); | |
1866 | ||
36bdfc8b GB |
1867 | if (!test_and_set_bit(SK_DETACHED, &svsk->sk_flags)) |
1868 | list_del_init(&svsk->sk_list); | |
cca5172a | 1869 | /* |
3262c816 GB |
1870 | * We used to delete the svc_sock from whichever list |
1871 | * it's sk_ready node was on, but we don't actually | |
1872 | * need to. This is because the only time we're called | |
1873 | * while still attached to a queue, the queue itself | |
1874 | * is about to be destroyed (in svc_destroy). | |
1875 | */ | |
aaf68cfb N |
1876 | if (!test_and_set_bit(SK_DEAD, &svsk->sk_flags)) { |
1877 | BUG_ON(atomic_read(&svsk->sk_inuse)<2); | |
1878 | atomic_dec(&svsk->sk_inuse); | |
1da177e4 LT |
1879 | if (test_bit(SK_TEMP, &svsk->sk_flags)) |
1880 | serv->sv_tmpcnt--; | |
aaf68cfb | 1881 | } |
1da177e4 | 1882 | |
d6740df9 | 1883 | spin_unlock_bh(&serv->sv_lock); |
aaf68cfb N |
1884 | } |
1885 | ||
cda1fd4a | 1886 | static void svc_close_socket(struct svc_sock *svsk) |
aaf68cfb N |
1887 | { |
1888 | set_bit(SK_CLOSE, &svsk->sk_flags); | |
1889 | if (test_and_set_bit(SK_BUSY, &svsk->sk_flags)) | |
1890 | /* someone else will have to effect the close */ | |
1891 | return; | |
1892 | ||
1893 | atomic_inc(&svsk->sk_inuse); | |
1894 | svc_delete_socket(svsk); | |
1895 | clear_bit(SK_BUSY, &svsk->sk_flags); | |
d6740df9 | 1896 | svc_sock_put(svsk); |
1da177e4 LT |
1897 | } |
1898 | ||
cda1fd4a N |
1899 | void svc_force_close_socket(struct svc_sock *svsk) |
1900 | { | |
1901 | set_bit(SK_CLOSE, &svsk->sk_flags); | |
1902 | if (test_bit(SK_BUSY, &svsk->sk_flags)) { | |
1903 | /* Waiting to be processed, but no threads left, | |
1904 | * So just remove it from the waiting list | |
1905 | */ | |
1906 | list_del_init(&svsk->sk_ready); | |
1907 | clear_bit(SK_BUSY, &svsk->sk_flags); | |
1908 | } | |
1909 | svc_close_socket(svsk); | |
1910 | } | |
1911 | ||
6b174337 CL |
1912 | /** |
1913 | * svc_makesock - Make a socket for nfsd and lockd | |
1914 | * @serv: RPC server structure | |
1915 | * @protocol: transport protocol to use | |
1916 | * @port: port to use | |
482fb94e | 1917 | * @flags: requested socket characteristics |
6b174337 | 1918 | * |
1da177e4 | 1919 | */ |
482fb94e CL |
1920 | int svc_makesock(struct svc_serv *serv, int protocol, unsigned short port, |
1921 | int flags) | |
1da177e4 | 1922 | { |
6b174337 CL |
1923 | struct sockaddr_in sin = { |
1924 | .sin_family = AF_INET, | |
1925 | .sin_addr.s_addr = INADDR_ANY, | |
1926 | .sin_port = htons(port), | |
1927 | }; | |
1da177e4 LT |
1928 | |
1929 | dprintk("svc: creating socket proto = %d\n", protocol); | |
77f1f67a CL |
1930 | return svc_create_socket(serv, protocol, (struct sockaddr *) &sin, |
1931 | sizeof(sin), flags); | |
1da177e4 LT |
1932 | } |
1933 | ||
1934 | /* | |
cca5172a | 1935 | * Handle defer and revisit of requests |
1da177e4 LT |
1936 | */ |
1937 | ||
1938 | static void svc_revisit(struct cache_deferred_req *dreq, int too_many) | |
1939 | { | |
1940 | struct svc_deferred_req *dr = container_of(dreq, struct svc_deferred_req, handle); | |
1da177e4 LT |
1941 | struct svc_sock *svsk; |
1942 | ||
1943 | if (too_many) { | |
1944 | svc_sock_put(dr->svsk); | |
1945 | kfree(dr); | |
1946 | return; | |
1947 | } | |
1948 | dprintk("revisit queued\n"); | |
1949 | svsk = dr->svsk; | |
1950 | dr->svsk = NULL; | |
7ac1bea5 | 1951 | spin_lock(&svsk->sk_lock); |
1da177e4 | 1952 | list_add(&dr->handle.recent, &svsk->sk_deferred); |
7ac1bea5 | 1953 | spin_unlock(&svsk->sk_lock); |
1da177e4 LT |
1954 | set_bit(SK_DEFERRED, &svsk->sk_flags); |
1955 | svc_sock_enqueue(svsk); | |
1956 | svc_sock_put(svsk); | |
1957 | } | |
1958 | ||
1959 | static struct cache_deferred_req * | |
1960 | svc_defer(struct cache_req *req) | |
1961 | { | |
1962 | struct svc_rqst *rqstp = container_of(req, struct svc_rqst, rq_chandle); | |
1963 | int size = sizeof(struct svc_deferred_req) + (rqstp->rq_arg.len); | |
1964 | struct svc_deferred_req *dr; | |
1965 | ||
1966 | if (rqstp->rq_arg.page_len) | |
1967 | return NULL; /* if more than a page, give up FIXME */ | |
1968 | if (rqstp->rq_deferred) { | |
1969 | dr = rqstp->rq_deferred; | |
1970 | rqstp->rq_deferred = NULL; | |
1971 | } else { | |
1972 | int skip = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len; | |
1973 | /* FIXME maybe discard if size too large */ | |
1974 | dr = kmalloc(size, GFP_KERNEL); | |
1975 | if (dr == NULL) | |
1976 | return NULL; | |
1977 | ||
1978 | dr->handle.owner = rqstp->rq_server; | |
1979 | dr->prot = rqstp->rq_prot; | |
24422222 CL |
1980 | memcpy(&dr->addr, &rqstp->rq_addr, rqstp->rq_addrlen); |
1981 | dr->addrlen = rqstp->rq_addrlen; | |
1918e341 | 1982 | dr->daddr = rqstp->rq_daddr; |
1da177e4 LT |
1983 | dr->argslen = rqstp->rq_arg.len >> 2; |
1984 | memcpy(dr->args, rqstp->rq_arg.head[0].iov_base-skip, dr->argslen<<2); | |
1985 | } | |
c45c357d | 1986 | atomic_inc(&rqstp->rq_sock->sk_inuse); |
1da177e4 | 1987 | dr->svsk = rqstp->rq_sock; |
1da177e4 LT |
1988 | |
1989 | dr->handle.revisit = svc_revisit; | |
1990 | return &dr->handle; | |
1991 | } | |
1992 | ||
1993 | /* | |
1994 | * recv data from a deferred request into an active one | |
1995 | */ | |
1996 | static int svc_deferred_recv(struct svc_rqst *rqstp) | |
1997 | { | |
1998 | struct svc_deferred_req *dr = rqstp->rq_deferred; | |
1999 | ||
2000 | rqstp->rq_arg.head[0].iov_base = dr->args; | |
2001 | rqstp->rq_arg.head[0].iov_len = dr->argslen<<2; | |
2002 | rqstp->rq_arg.page_len = 0; | |
2003 | rqstp->rq_arg.len = dr->argslen<<2; | |
2004 | rqstp->rq_prot = dr->prot; | |
24422222 CL |
2005 | memcpy(&rqstp->rq_addr, &dr->addr, dr->addrlen); |
2006 | rqstp->rq_addrlen = dr->addrlen; | |
1918e341 | 2007 | rqstp->rq_daddr = dr->daddr; |
44524359 | 2008 | rqstp->rq_respages = rqstp->rq_pages; |
1da177e4 LT |
2009 | return dr->argslen<<2; |
2010 | } | |
2011 | ||
2012 | ||
2013 | static struct svc_deferred_req *svc_deferred_dequeue(struct svc_sock *svsk) | |
2014 | { | |
2015 | struct svc_deferred_req *dr = NULL; | |
cca5172a | 2016 | |
1da177e4 LT |
2017 | if (!test_bit(SK_DEFERRED, &svsk->sk_flags)) |
2018 | return NULL; | |
7ac1bea5 | 2019 | spin_lock(&svsk->sk_lock); |
1da177e4 LT |
2020 | clear_bit(SK_DEFERRED, &svsk->sk_flags); |
2021 | if (!list_empty(&svsk->sk_deferred)) { | |
2022 | dr = list_entry(svsk->sk_deferred.next, | |
2023 | struct svc_deferred_req, | |
2024 | handle.recent); | |
2025 | list_del_init(&dr->handle.recent); | |
2026 | set_bit(SK_DEFERRED, &svsk->sk_flags); | |
2027 | } | |
7ac1bea5 | 2028 | spin_unlock(&svsk->sk_lock); |
1da177e4 LT |
2029 | return dr; |
2030 | } |