blob: 2007881a5b26b36f7d662dede53f7bbfe7e087c2 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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
Ilpo Järvinen172589c2007-08-28 15:50:33 -070022#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#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>
Andrew Morton91483c42005-08-09 20:20:07 -070030#include <linux/tcp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include <linux/unistd.h>
32#include <linux/slab.h>
33#include <linux/netdevice.h>
34#include <linux/skbuff.h>
NeilBrownb41b66d2006-10-02 02:17:48 -070035#include <linux/file.h>
Nigel Cunningham7dfb7102006-12-06 20:34:23 -080036#include <linux/freezer.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include <net/sock.h>
38#include <net/checksum.h>
39#include <net/ip.h>
Chuck Leverb92503b2007-02-12 00:53:36 -080040#include <net/ipv6.h>
Arnaldo Carvalho de Meloc752f072005-08-09 20:08:28 -070041#include <net/tcp_states.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070042#include <asm/uaccess.h>
43#include <asm/ioctls.h>
44
45#include <linux/sunrpc/types.h>
Chuck Leverad06e4b2007-02-12 00:53:32 -080046#include <linux/sunrpc/clnt.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070047#include <linux/sunrpc/xdr.h>
48#include <linux/sunrpc/svcsock.h>
49#include <linux/sunrpc/stats.h>
50
51/* SMP locking strategy:
52 *
Greg Banks3262c812006-10-02 02:17:58 -070053 * 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.
NeilBrown7ac1bea2007-05-09 02:34:48 -070057 * svc_sock->sk_lock protects the svc_sock->sk_deferred list
58 * and the ->sk_info_authunix cache.
Greg Banksc081a0c2006-10-02 02:17:57 -070059 * svc_sock->sk_flags.SK_BUSY prevents a svc_sock being enqueued multiply.
Linus Torvalds1da177e2005-04-16 15:20:36 -070060 *
61 * Some flags can be set to certain values at any time
62 * providing that certain rules are followed:
63 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070064 * SK_CONN, SK_DATA, can be set or cleared at any time.
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -080065 * after a set, svc_sock_enqueue must be called.
Linus Torvalds1da177e2005-04-16 15:20:36 -070066 * 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.
NeilBrownaaf68cf2007-02-08 14:20:30 -080069 * 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.
Linus Torvalds1da177e2005-04-16 15:20:36 -070075 *
76 */
77
Tom Tucker360d8732007-12-30 21:07:17 -060078#define RPCDBG_FACILITY RPCDBG_SVCXPRT
Linus Torvalds1da177e2005-04-16 15:20:36 -070079
80
81static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
Chuck Lever6b174332007-02-12 00:53:28 -080082 int *errp, int flags);
NeilBrownaaf68cf2007-02-08 14:20:30 -080083static void svc_delete_socket(struct svc_sock *svsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -070084static void svc_udp_data_ready(struct sock *, int);
85static int svc_udp_recvfrom(struct svc_rqst *);
86static int svc_udp_sendto(struct svc_rqst *);
NeilBrowncda1fd42007-03-06 01:42:22 -080087static void svc_close_socket(struct svc_sock *svsk);
Tom Tucker755ccea2007-12-30 21:07:27 -060088static void svc_sock_detach(struct svc_xprt *);
89static void svc_sock_free(struct svc_xprt *);
Linus Torvalds1da177e2005-04-16 15:20:36 -070090
91static struct svc_deferred_req *svc_deferred_dequeue(struct svc_sock *svsk);
92static int svc_deferred_recv(struct svc_rqst *rqstp);
93static struct cache_deferred_req *svc_defer(struct cache_req *req);
94
Greg Banks36bdfc82006-10-02 02:17:54 -070095/* 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 */
100static int svc_conn_age_period = 6*60;
101
Peter Zijlstraed075362006-12-06 20:35:24 -0800102#ifdef CONFIG_DEBUG_LOCK_ALLOC
103static struct lock_class_key svc_key[2];
104static struct lock_class_key svc_slock_key[2];
105
106static inline void svc_reclassify_socket(struct socket *sock)
107{
108 struct sock *sk = sock->sk;
John Heffner02b3d342007-09-12 10:42:12 +0200109 BUG_ON(sock_owned_by_user(sk));
Peter Zijlstraed075362006-12-06 20:35:24 -0800110 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
126static inline void svc_reclassify_socket(struct socket *sock)
127{
128}
129#endif
130
Chuck Leverad06e4b2007-02-12 00:53:32 -0800131static 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),
Al Viro582ee432007-07-26 17:33:39 +0100137 ntohs(((struct sockaddr_in *) addr)->sin_port));
Chuck Leverad06e4b2007-02-12 00:53:32 -0800138 break;
NeilBrown5a05ed72007-03-06 01:42:22 -0800139
Chuck Leverad06e4b2007-02-12 00:53:32 -0800140 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),
Al Viro582ee432007-07-26 17:33:39 +0100143 ntohs(((struct sockaddr_in6 *) addr)->sin6_port));
Chuck Leverad06e4b2007-02-12 00:53:32 -0800144 break;
NeilBrown5a05ed72007-03-06 01:42:22 -0800145
Chuck Leverad06e4b2007-02-12 00:53:32 -0800146 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 */
160char *svc_print_addr(struct svc_rqst *rqstp, char *buf, size_t len)
161{
Chuck Lever27459f02007-02-12 00:53:34 -0800162 return __svc_print_addr(svc_addr(rqstp), buf, len);
Chuck Leverad06e4b2007-02-12 00:53:32 -0800163}
164EXPORT_SYMBOL_GPL(svc_print_addr);
165
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166/*
Greg Banks3262c812006-10-02 02:17:58 -0700167 * Queue up an idle server thread. Must have pool->sp_lock held.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168 * Note: this is really a stack rather than a queue, so that we only
Greg Banks3262c812006-10-02 02:17:58 -0700169 * use as many different threads as we need, and the rest don't pollute
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 * the cache.
171 */
172static inline void
Greg Banks3262c812006-10-02 02:17:58 -0700173svc_thread_enqueue(struct svc_pool *pool, struct svc_rqst *rqstp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174{
Greg Banks3262c812006-10-02 02:17:58 -0700175 list_add(&rqstp->rq_list, &pool->sp_threads);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176}
177
178/*
Greg Banks3262c812006-10-02 02:17:58 -0700179 * Dequeue an nfsd thread. Must have pool->sp_lock held.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 */
181static inline void
Greg Banks3262c812006-10-02 02:17:58 -0700182svc_thread_dequeue(struct svc_pool *pool, struct svc_rqst *rqstp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183{
184 list_del(&rqstp->rq_list);
185}
186
187/*
188 * Release an skbuff after use
189 */
Tom Tucker5148bf42007-12-30 21:07:25 -0600190static void svc_release_skb(struct svc_rqst *rqstp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191{
Tom Tucker5148bf42007-12-30 21:07:25 -0600192 struct sk_buff *skb = rqstp->rq_xprt_ctxt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 struct svc_deferred_req *dr = rqstp->rq_deferred;
194
195 if (skb) {
Tom Tucker5148bf42007-12-30 21:07:25 -0600196 rqstp->rq_xprt_ctxt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197
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/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 * Queue up a socket with data pending. If there are idle nfsd
209 * processes, wake 'em up.
210 *
211 */
212static void
213svc_sock_enqueue(struct svc_sock *svsk)
214{
215 struct svc_serv *serv = svsk->sk_server;
Greg Banksbfd24162006-10-02 02:18:01 -0700216 struct svc_pool *pool;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 struct svc_rqst *rqstp;
Greg Banksbfd24162006-10-02 02:18:01 -0700218 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219
220 if (!(svsk->sk_flags &
221 ( (1<<SK_CONN)|(1<<SK_DATA)|(1<<SK_CLOSE)|(1<<SK_DEFERRED)) ))
222 return;
223 if (test_bit(SK_DEAD, &svsk->sk_flags))
224 return;
225
Greg Banksbfd24162006-10-02 02:18:01 -0700226 cpu = get_cpu();
227 pool = svc_pool_for_cpu(svsk->sk_server, cpu);
228 put_cpu();
229
Greg Banks3262c812006-10-02 02:17:58 -0700230 spin_lock_bh(&pool->sp_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231
Greg Banks3262c812006-10-02 02:17:58 -0700232 if (!list_empty(&pool->sp_threads) &&
233 !list_empty(&pool->sp_sockets))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 printk(KERN_ERR
235 "svc_sock_enqueue: threads and sockets both waiting??\n");
236
237 if (test_bit(SK_DEAD, &svsk->sk_flags)) {
238 /* Don't enqueue dead sockets */
239 dprintk("svc: socket %p is dead, not enqueued\n", svsk->sk_sk);
240 goto out_unlock;
241 }
242
Greg Banksc081a0c2006-10-02 02:17:57 -0700243 /* Mark socket as busy. It will remain in this state until the
244 * server has processed all pending data and put the socket back
245 * on the idle list. We update SK_BUSY atomically because
246 * it also guards against trying to enqueue the svc_sock twice.
247 */
248 if (test_and_set_bit(SK_BUSY, &svsk->sk_flags)) {
249 /* Don't enqueue socket while already enqueued */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250 dprintk("svc: socket %p busy, not enqueued\n", svsk->sk_sk);
251 goto out_unlock;
252 }
Greg Banks3262c812006-10-02 02:17:58 -0700253 BUG_ON(svsk->sk_pool != NULL);
254 svsk->sk_pool = pool;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255
Tom Tucker323bee32007-12-30 21:07:31 -0600256 /* Handle pending connection */
257 if (test_bit(SK_CONN, &svsk->sk_flags))
258 goto process;
259
260 /* Handle close in-progress */
261 if (test_bit(SK_CLOSE, &svsk->sk_flags))
262 goto process;
263
264 /* Check if we have space to reply to a request */
265 if (!svsk->sk_xprt.xpt_ops->xpo_has_wspace(&svsk->sk_xprt)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 /* Don't enqueue while not enough space for reply */
Tom Tucker323bee32007-12-30 21:07:31 -0600267 dprintk("svc: no write space, socket %p not enqueued\n", svsk);
Greg Banks3262c812006-10-02 02:17:58 -0700268 svsk->sk_pool = NULL;
Greg Banksc081a0c2006-10-02 02:17:57 -0700269 clear_bit(SK_BUSY, &svsk->sk_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 goto out_unlock;
271 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272
Tom Tucker323bee32007-12-30 21:07:31 -0600273 process:
Greg Banks3262c812006-10-02 02:17:58 -0700274 if (!list_empty(&pool->sp_threads)) {
275 rqstp = list_entry(pool->sp_threads.next,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 struct svc_rqst,
277 rq_list);
278 dprintk("svc: socket %p served by daemon %p\n",
279 svsk->sk_sk, rqstp);
Greg Banks3262c812006-10-02 02:17:58 -0700280 svc_thread_dequeue(pool, rqstp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 if (rqstp->rq_sock)
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -0800282 printk(KERN_ERR
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283 "svc_sock_enqueue: server %p, rq_sock=%p!\n",
284 rqstp, rqstp->rq_sock);
285 rqstp->rq_sock = svsk;
Greg Banksc45c3572006-10-02 02:17:54 -0700286 atomic_inc(&svsk->sk_inuse);
NeilBrownc6b0a9f2006-10-06 00:44:05 -0700287 rqstp->rq_reserved = serv->sv_max_mesg;
Greg Banks5685f0f2006-10-02 02:17:56 -0700288 atomic_add(rqstp->rq_reserved, &svsk->sk_reserved);
Greg Banks3262c812006-10-02 02:17:58 -0700289 BUG_ON(svsk->sk_pool != pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290 wake_up(&rqstp->rq_wait);
291 } else {
292 dprintk("svc: socket %p put into queue\n", svsk->sk_sk);
Greg Banks3262c812006-10-02 02:17:58 -0700293 list_add_tail(&svsk->sk_ready, &pool->sp_sockets);
294 BUG_ON(svsk->sk_pool != pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 }
296
297out_unlock:
Greg Banks3262c812006-10-02 02:17:58 -0700298 spin_unlock_bh(&pool->sp_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299}
300
301/*
Greg Banks3262c812006-10-02 02:17:58 -0700302 * Dequeue the first socket. Must be called with the pool->sp_lock held.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303 */
304static inline struct svc_sock *
Greg Banks3262c812006-10-02 02:17:58 -0700305svc_sock_dequeue(struct svc_pool *pool)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306{
307 struct svc_sock *svsk;
308
Greg Banks3262c812006-10-02 02:17:58 -0700309 if (list_empty(&pool->sp_sockets))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310 return NULL;
311
Greg Banks3262c812006-10-02 02:17:58 -0700312 svsk = list_entry(pool->sp_sockets.next,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 struct svc_sock, sk_ready);
314 list_del_init(&svsk->sk_ready);
315
316 dprintk("svc: socket %p dequeued, inuse=%d\n",
Greg Banksc45c3572006-10-02 02:17:54 -0700317 svsk->sk_sk, atomic_read(&svsk->sk_inuse));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318
319 return svsk;
320}
321
322/*
323 * Having read something from a socket, check whether it
324 * needs to be re-enqueued.
325 * Note: SK_DATA only gets cleared when a read-attempt finds
326 * no (or insufficient) data.
327 */
328static inline void
329svc_sock_received(struct svc_sock *svsk)
330{
Greg Banks3262c812006-10-02 02:17:58 -0700331 svsk->sk_pool = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 clear_bit(SK_BUSY, &svsk->sk_flags);
333 svc_sock_enqueue(svsk);
334}
335
336
337/**
338 * svc_reserve - change the space reserved for the reply to a request.
339 * @rqstp: The request in question
340 * @space: new max space to reserve
341 *
342 * Each request reserves some space on the output queue of the socket
343 * to make sure the reply fits. This function reduces that reserved
344 * space to be the amount of space used already, plus @space.
345 *
346 */
347void svc_reserve(struct svc_rqst *rqstp, int space)
348{
349 space += rqstp->rq_res.head[0].iov_len;
350
351 if (space < rqstp->rq_reserved) {
352 struct svc_sock *svsk = rqstp->rq_sock;
Greg Banks5685f0f2006-10-02 02:17:56 -0700353 atomic_sub((rqstp->rq_reserved - space), &svsk->sk_reserved);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 rqstp->rq_reserved = space;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355
356 svc_sock_enqueue(svsk);
357 }
358}
359
360/*
361 * Release a socket after use.
362 */
363static inline void
364svc_sock_put(struct svc_sock *svsk)
365{
NeilBrownaaf68cf2007-02-08 14:20:30 -0800366 if (atomic_dec_and_test(&svsk->sk_inuse)) {
Tom Tucker755ccea2007-12-30 21:07:27 -0600367 BUG_ON(!test_bit(SK_DEAD, &svsk->sk_flags));
368 svsk->sk_xprt.xpt_ops->xpo_free(&svsk->sk_xprt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370}
371
372static void
373svc_sock_release(struct svc_rqst *rqstp)
374{
375 struct svc_sock *svsk = rqstp->rq_sock;
376
Tom Tucker5148bf42007-12-30 21:07:25 -0600377 rqstp->rq_xprt->xpt_ops->xpo_release_rqst(rqstp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378
NeilBrown44524352006-10-04 02:15:46 -0700379 svc_free_res_pages(rqstp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380 rqstp->rq_res.page_len = 0;
381 rqstp->rq_res.page_base = 0;
382
383
384 /* Reset response buffer and release
385 * the reservation.
386 * But first, check that enough space was reserved
387 * for the reply, otherwise we have a bug!
388 */
389 if ((rqstp->rq_res.len) > rqstp->rq_reserved)
390 printk(KERN_ERR "RPC request reserved %d but used %d\n",
391 rqstp->rq_reserved,
392 rqstp->rq_res.len);
393
394 rqstp->rq_res.head[0].iov_len = 0;
395 svc_reserve(rqstp, 0);
396 rqstp->rq_sock = NULL;
397
398 svc_sock_put(svsk);
399}
400
401/*
402 * External function to wake up a server waiting for data
Greg Banks3262c812006-10-02 02:17:58 -0700403 * This really only makes sense for services like lockd
404 * which have exactly one thread anyway.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405 */
406void
407svc_wake_up(struct svc_serv *serv)
408{
409 struct svc_rqst *rqstp;
Greg Banks3262c812006-10-02 02:17:58 -0700410 unsigned int i;
411 struct svc_pool *pool;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412
Greg Banks3262c812006-10-02 02:17:58 -0700413 for (i = 0; i < serv->sv_nrpools; i++) {
414 pool = &serv->sv_pools[i];
415
416 spin_lock_bh(&pool->sp_lock);
417 if (!list_empty(&pool->sp_threads)) {
418 rqstp = list_entry(pool->sp_threads.next,
419 struct svc_rqst,
420 rq_list);
421 dprintk("svc: daemon %p woken up.\n", rqstp);
422 /*
423 svc_thread_dequeue(pool, rqstp);
424 rqstp->rq_sock = NULL;
425 */
426 wake_up(&rqstp->rq_wait);
427 }
428 spin_unlock_bh(&pool->sp_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430}
431
Chuck Leverb92503b2007-02-12 00:53:36 -0800432union svc_pktinfo_u {
433 struct in_pktinfo pkti;
Chuck Leverb92503b2007-02-12 00:53:36 -0800434 struct in6_pktinfo pkti6;
Chuck Leverb92503b2007-02-12 00:53:36 -0800435};
David S. Millerbc375ea2007-04-12 13:35:59 -0700436#define SVC_PKTINFO_SPACE \
437 CMSG_SPACE(sizeof(union svc_pktinfo_u))
Chuck Leverb92503b2007-02-12 00:53:36 -0800438
439static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
440{
441 switch (rqstp->rq_sock->sk_sk->sk_family) {
442 case AF_INET: {
443 struct in_pktinfo *pki = CMSG_DATA(cmh);
444
445 cmh->cmsg_level = SOL_IP;
446 cmh->cmsg_type = IP_PKTINFO;
447 pki->ipi_ifindex = 0;
448 pki->ipi_spec_dst.s_addr = rqstp->rq_daddr.addr.s_addr;
449 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
450 }
451 break;
NeilBrown5a05ed72007-03-06 01:42:22 -0800452
Chuck Leverb92503b2007-02-12 00:53:36 -0800453 case AF_INET6: {
454 struct in6_pktinfo *pki = CMSG_DATA(cmh);
455
456 cmh->cmsg_level = SOL_IPV6;
457 cmh->cmsg_type = IPV6_PKTINFO;
458 pki->ipi6_ifindex = 0;
459 ipv6_addr_copy(&pki->ipi6_addr,
460 &rqstp->rq_daddr.addr6);
461 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
462 }
463 break;
Chuck Leverb92503b2007-02-12 00:53:36 -0800464 }
465 return;
466}
467
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468/*
469 * Generic sendto routine
470 */
471static int
472svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr)
473{
474 struct svc_sock *svsk = rqstp->rq_sock;
475 struct socket *sock = svsk->sk_sock;
476 int slen;
David S. Millerbc375ea2007-04-12 13:35:59 -0700477 union {
478 struct cmsghdr hdr;
479 long all[SVC_PKTINFO_SPACE / sizeof(long)];
480 } buffer;
481 struct cmsghdr *cmh = &buffer.hdr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 int len = 0;
483 int result;
484 int size;
485 struct page **ppage = xdr->pages;
486 size_t base = xdr->page_base;
487 unsigned int pglen = xdr->page_len;
488 unsigned int flags = MSG_MORE;
Chuck Leverad06e4b2007-02-12 00:53:32 -0800489 char buf[RPC_MAX_ADDRBUFLEN];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490
491 slen = xdr->len;
492
493 if (rqstp->rq_prot == IPPROTO_UDP) {
Chuck Leverb92503b2007-02-12 00:53:36 -0800494 struct msghdr msg = {
495 .msg_name = &rqstp->rq_addr,
496 .msg_namelen = rqstp->rq_addrlen,
497 .msg_control = cmh,
498 .msg_controllen = sizeof(buffer),
499 .msg_flags = MSG_MORE,
500 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501
Chuck Leverb92503b2007-02-12 00:53:36 -0800502 svc_set_cmsg_data(rqstp, cmh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503
504 if (sock_sendmsg(sock, &msg, 0) < 0)
505 goto out;
506 }
507
508 /* send head */
509 if (slen == xdr->head[0].iov_len)
510 flags = 0;
NeilBrown44524352006-10-04 02:15:46 -0700511 len = kernel_sendpage(sock, rqstp->rq_respages[0], 0,
512 xdr->head[0].iov_len, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 if (len != xdr->head[0].iov_len)
514 goto out;
515 slen -= xdr->head[0].iov_len;
516 if (slen == 0)
517 goto out;
518
519 /* send page data */
520 size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen;
521 while (pglen > 0) {
522 if (slen == size)
523 flags = 0;
Sridhar Samudralae6242e92006-08-07 20:58:01 -0700524 result = kernel_sendpage(sock, *ppage, base, size, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 if (result > 0)
526 len += result;
527 if (result != size)
528 goto out;
529 slen -= size;
530 pglen -= size;
531 size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen;
532 base = 0;
533 ppage++;
534 }
535 /* send tail */
536 if (xdr->tail[0].iov_len) {
NeilBrown44524352006-10-04 02:15:46 -0700537 result = kernel_sendpage(sock, rqstp->rq_respages[0],
538 ((unsigned long)xdr->tail[0].iov_base)
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -0800539 & (PAGE_SIZE-1),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 xdr->tail[0].iov_len, 0);
541
542 if (result > 0)
543 len += result;
544 }
545out:
Chuck Leverad06e4b2007-02-12 00:53:32 -0800546 dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n",
547 rqstp->rq_sock, xdr->head[0].iov_base, xdr->head[0].iov_len,
548 xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549
550 return len;
551}
552
553/*
NeilBrown80212d52006-10-02 02:17:47 -0700554 * Report socket names for nfsdfs
555 */
556static int one_sock_name(char *buf, struct svc_sock *svsk)
557{
558 int len;
559
560 switch(svsk->sk_sk->sk_family) {
561 case AF_INET:
562 len = sprintf(buf, "ipv4 %s %u.%u.%u.%u %d\n",
563 svsk->sk_sk->sk_protocol==IPPROTO_UDP?
564 "udp" : "tcp",
565 NIPQUAD(inet_sk(svsk->sk_sk)->rcv_saddr),
566 inet_sk(svsk->sk_sk)->num);
567 break;
568 default:
569 len = sprintf(buf, "*unknown-%d*\n",
570 svsk->sk_sk->sk_family);
571 }
572 return len;
573}
574
575int
NeilBrownb41b66d2006-10-02 02:17:48 -0700576svc_sock_names(char *buf, struct svc_serv *serv, char *toclose)
NeilBrown80212d52006-10-02 02:17:47 -0700577{
NeilBrownb41b66d2006-10-02 02:17:48 -0700578 struct svc_sock *svsk, *closesk = NULL;
NeilBrown80212d52006-10-02 02:17:47 -0700579 int len = 0;
580
581 if (!serv)
582 return 0;
NeilBrownaaf68cf2007-02-08 14:20:30 -0800583 spin_lock_bh(&serv->sv_lock);
NeilBrown80212d52006-10-02 02:17:47 -0700584 list_for_each_entry(svsk, &serv->sv_permsocks, sk_list) {
585 int onelen = one_sock_name(buf+len, svsk);
NeilBrownb41b66d2006-10-02 02:17:48 -0700586 if (toclose && strcmp(toclose, buf+len) == 0)
587 closesk = svsk;
588 else
589 len += onelen;
NeilBrown80212d52006-10-02 02:17:47 -0700590 }
NeilBrownaaf68cf2007-02-08 14:20:30 -0800591 spin_unlock_bh(&serv->sv_lock);
NeilBrownb41b66d2006-10-02 02:17:48 -0700592 if (closesk)
NeilBrown5680c442006-10-04 02:15:45 -0700593 /* Should unregister with portmap, but you cannot
594 * unregister just one protocol...
595 */
NeilBrownaaf68cf2007-02-08 14:20:30 -0800596 svc_close_socket(closesk);
NeilBrown37a03472006-10-04 02:15:44 -0700597 else if (toclose)
598 return -ENOENT;
NeilBrown80212d52006-10-02 02:17:47 -0700599 return len;
600}
601EXPORT_SYMBOL(svc_sock_names);
602
603/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 * Check input queue length
605 */
606static int
607svc_recv_available(struct svc_sock *svsk)
608{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 struct socket *sock = svsk->sk_sock;
610 int avail, err;
611
Sridhar Samudralae6242e92006-08-07 20:58:01 -0700612 err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613
614 return (err >= 0)? avail : err;
615}
616
617/*
618 * Generic recvfrom routine.
619 */
620static int
621svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr, int buflen)
622{
Chuck Lever067d7812007-02-12 00:53:30 -0800623 struct svc_sock *svsk = rqstp->rq_sock;
Chuck Lever1ba95102007-02-12 00:53:31 -0800624 struct msghdr msg = {
625 .msg_flags = MSG_DONTWAIT,
626 };
Frank van Maarseveena9747692007-07-09 22:21:39 +0200627 struct sockaddr *sin;
Chuck Lever1ba95102007-02-12 00:53:31 -0800628 int len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629
Chuck Lever1ba95102007-02-12 00:53:31 -0800630 len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen,
631 msg.msg_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632
633 /* sock_recvmsg doesn't fill in the name/namelen, so we must..
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 */
Chuck Lever067d7812007-02-12 00:53:30 -0800635 memcpy(&rqstp->rq_addr, &svsk->sk_remote, svsk->sk_remotelen);
636 rqstp->rq_addrlen = svsk->sk_remotelen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637
Frank van Maarseveena9747692007-07-09 22:21:39 +0200638 /* Destination address in request is needed for binding the
639 * source address in RPC callbacks later.
640 */
641 sin = (struct sockaddr *)&svsk->sk_local;
642 switch (sin->sa_family) {
643 case AF_INET:
644 rqstp->rq_daddr.addr = ((struct sockaddr_in *)sin)->sin_addr;
645 break;
646 case AF_INET6:
647 rqstp->rq_daddr.addr6 = ((struct sockaddr_in6 *)sin)->sin6_addr;
648 break;
649 }
650
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
Chuck Lever1ba95102007-02-12 00:53:31 -0800652 svsk, iov[0].iov_base, iov[0].iov_len, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653
654 return len;
655}
656
657/*
658 * Set socket snd and rcv buffer lengths
659 */
660static inline void
661svc_sock_setbufsize(struct socket *sock, unsigned int snd, unsigned int rcv)
662{
663#if 0
664 mm_segment_t oldfs;
665 oldfs = get_fs(); set_fs(KERNEL_DS);
666 sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
667 (char*)&snd, sizeof(snd));
668 sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
669 (char*)&rcv, sizeof(rcv));
670#else
671 /* sock_setsockopt limits use to sysctl_?mem_max,
672 * which isn't acceptable. Until that is made conditional
673 * on not having CAP_SYS_RESOURCE or similar, we go direct...
674 * DaveM said I could!
675 */
676 lock_sock(sock->sk);
677 sock->sk->sk_sndbuf = snd * 2;
678 sock->sk->sk_rcvbuf = rcv * 2;
679 sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK|SOCK_RCVBUF_LOCK;
680 release_sock(sock->sk);
681#endif
682}
683/*
684 * INET callback when data has been received on the socket.
685 */
686static void
687svc_udp_data_ready(struct sock *sk, int count)
688{
Neil Brown939bb7e2005-09-13 01:25:39 -0700689 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690
Neil Brown939bb7e2005-09-13 01:25:39 -0700691 if (svsk) {
692 dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n",
693 svsk, sk, count, test_bit(SK_BUSY, &svsk->sk_flags));
694 set_bit(SK_DATA, &svsk->sk_flags);
695 svc_sock_enqueue(svsk);
696 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
698 wake_up_interruptible(sk->sk_sleep);
699}
700
701/*
702 * INET callback when space is newly available on the socket.
703 */
704static void
705svc_write_space(struct sock *sk)
706{
707 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
708
709 if (svsk) {
710 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
711 svsk, sk, test_bit(SK_BUSY, &svsk->sk_flags));
712 svc_sock_enqueue(svsk);
713 }
714
715 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) {
Neil Brown939bb7e2005-09-13 01:25:39 -0700716 dprintk("RPC svc_write_space: someone sleeping on %p\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 svsk);
718 wake_up_interruptible(sk->sk_sleep);
719 }
720}
721
NeilBrown7a37f572007-03-06 01:42:21 -0800722static inline void svc_udp_get_dest_address(struct svc_rqst *rqstp,
723 struct cmsghdr *cmh)
Chuck Lever95756482007-02-12 00:53:38 -0800724{
725 switch (rqstp->rq_sock->sk_sk->sk_family) {
726 case AF_INET: {
NeilBrown7a37f572007-03-06 01:42:21 -0800727 struct in_pktinfo *pki = CMSG_DATA(cmh);
728 rqstp->rq_daddr.addr.s_addr = pki->ipi_spec_dst.s_addr;
Chuck Lever95756482007-02-12 00:53:38 -0800729 break;
NeilBrown7a37f572007-03-06 01:42:21 -0800730 }
Chuck Lever95756482007-02-12 00:53:38 -0800731 case AF_INET6: {
NeilBrown7a37f572007-03-06 01:42:21 -0800732 struct in6_pktinfo *pki = CMSG_DATA(cmh);
733 ipv6_addr_copy(&rqstp->rq_daddr.addr6, &pki->ipi6_addr);
Chuck Lever95756482007-02-12 00:53:38 -0800734 break;
NeilBrown7a37f572007-03-06 01:42:21 -0800735 }
Chuck Lever95756482007-02-12 00:53:38 -0800736 }
Chuck Lever95756482007-02-12 00:53:38 -0800737}
738
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739/*
740 * Receive a datagram from a UDP socket.
741 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742static int
743svc_udp_recvfrom(struct svc_rqst *rqstp)
744{
745 struct svc_sock *svsk = rqstp->rq_sock;
746 struct svc_serv *serv = svsk->sk_server;
747 struct sk_buff *skb;
David S. Millerbc375ea2007-04-12 13:35:59 -0700748 union {
749 struct cmsghdr hdr;
750 long all[SVC_PKTINFO_SPACE / sizeof(long)];
751 } buffer;
752 struct cmsghdr *cmh = &buffer.hdr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 int err, len;
NeilBrown7a37f572007-03-06 01:42:21 -0800754 struct msghdr msg = {
755 .msg_name = svc_addr(rqstp),
756 .msg_control = cmh,
757 .msg_controllen = sizeof(buffer),
758 .msg_flags = MSG_DONTWAIT,
759 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760
761 if (test_and_clear_bit(SK_CHNGBUF, &svsk->sk_flags))
762 /* udp sockets need large rcvbuf as all pending
763 * requests are still in that buffer. sndbuf must
764 * also be large enough that there is enough space
Greg Banks3262c812006-10-02 02:17:58 -0700765 * for one reply per thread. We count all threads
766 * rather than threads in a particular pool, which
767 * provides an upper bound on the number of threads
768 * which will access the socket.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 */
770 svc_sock_setbufsize(svsk->sk_sock,
NeilBrownc6b0a9f2006-10-06 00:44:05 -0700771 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
772 (serv->sv_nrthreads+3) * serv->sv_max_mesg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773
774 if ((rqstp->rq_deferred = svc_deferred_dequeue(svsk))) {
775 svc_sock_received(svsk);
776 return svc_deferred_recv(rqstp);
777 }
778
NeilBrownaaf68cf2007-02-08 14:20:30 -0800779 if (test_bit(SK_CLOSE, &svsk->sk_flags)) {
780 svc_delete_socket(svsk);
781 return 0;
782 }
783
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 clear_bit(SK_DATA, &svsk->sk_flags);
NeilBrown05ed6902007-05-09 02:34:55 -0700785 skb = NULL;
786 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
787 0, 0, MSG_PEEK | MSG_DONTWAIT);
788 if (err >= 0)
789 skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err);
790
791 if (skb == NULL) {
792 if (err != -EAGAIN) {
793 /* possibly an icmp error */
794 dprintk("svc: recvfrom returned error %d\n", -err);
795 set_bit(SK_DATA, &svsk->sk_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 }
NeilBrown05ed6902007-05-09 02:34:55 -0700797 svc_sock_received(svsk);
798 return -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 }
NeilBrown7a37f572007-03-06 01:42:21 -0800800 rqstp->rq_addrlen = sizeof(rqstp->rq_addr);
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -0700801 if (skb->tstamp.tv64 == 0) {
802 skb->tstamp = ktime_get_real();
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -0800803 /* Don't enable netstamp, sunrpc doesn't
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 need that much accuracy */
805 }
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -0700806 svsk->sk_sk->sk_stamp = skb->tstamp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 set_bit(SK_DATA, &svsk->sk_flags); /* there may be more data... */
808
809 /*
810 * Maybe more packets - kick another thread ASAP.
811 */
812 svc_sock_received(svsk);
813
814 len = skb->len - sizeof(struct udphdr);
815 rqstp->rq_arg.len = len;
816
Chuck Lever95756482007-02-12 00:53:38 -0800817 rqstp->rq_prot = IPPROTO_UDP;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818
NeilBrown7a37f572007-03-06 01:42:21 -0800819 if (cmh->cmsg_level != IPPROTO_IP ||
820 cmh->cmsg_type != IP_PKTINFO) {
821 if (net_ratelimit())
822 printk("rpcsvc: received unknown control message:"
823 "%d/%d\n",
824 cmh->cmsg_level, cmh->cmsg_type);
825 skb_free_datagram(svsk->sk_sk, skb);
826 return 0;
827 }
828 svc_udp_get_dest_address(rqstp, cmh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829
830 if (skb_is_nonlinear(skb)) {
831 /* we have to copy */
832 local_bh_disable();
833 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
834 local_bh_enable();
835 /* checksum error */
836 skb_free_datagram(svsk->sk_sk, skb);
837 return 0;
838 }
839 local_bh_enable();
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -0800840 skb_free_datagram(svsk->sk_sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 } else {
842 /* we can use it in-place */
843 rqstp->rq_arg.head[0].iov_base = skb->data + sizeof(struct udphdr);
844 rqstp->rq_arg.head[0].iov_len = len;
Herbert Xufb286bb2005-11-10 13:01:24 -0800845 if (skb_checksum_complete(skb)) {
846 skb_free_datagram(svsk->sk_sk, skb);
847 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848 }
Tom Tucker5148bf42007-12-30 21:07:25 -0600849 rqstp->rq_xprt_ctxt = skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 }
851
852 rqstp->rq_arg.page_base = 0;
853 if (len <= rqstp->rq_arg.head[0].iov_len) {
854 rqstp->rq_arg.head[0].iov_len = len;
855 rqstp->rq_arg.page_len = 0;
NeilBrown44524352006-10-04 02:15:46 -0700856 rqstp->rq_respages = rqstp->rq_pages+1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 } else {
858 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
NeilBrown44524352006-10-04 02:15:46 -0700859 rqstp->rq_respages = rqstp->rq_pages + 1 +
Ilpo Järvinen172589c2007-08-28 15:50:33 -0700860 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 }
862
863 if (serv->sv_stats)
864 serv->sv_stats->netudpcnt++;
865
866 return len;
867}
868
869static int
870svc_udp_sendto(struct svc_rqst *rqstp)
871{
872 int error;
873
874 error = svc_sendto(rqstp, &rqstp->rq_res);
875 if (error == -ECONNREFUSED)
876 /* ICMP error on earlier request. */
877 error = svc_sendto(rqstp, &rqstp->rq_res);
878
879 return error;
880}
881
Tom Tuckere831fe62007-12-30 21:07:29 -0600882static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
883{
884}
885
Tom Tucker323bee32007-12-30 21:07:31 -0600886static int svc_udp_has_wspace(struct svc_xprt *xprt)
887{
888 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
889 struct svc_serv *serv = svsk->sk_server;
890 unsigned long required;
891
892 /*
893 * Set the SOCK_NOSPACE flag before checking the available
894 * sock space.
895 */
896 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
897 required = atomic_read(&svsk->sk_reserved) + serv->sv_max_mesg;
898 if (required*2 > sock_wspace(svsk->sk_sk))
899 return 0;
900 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
901 return 1;
902}
903
Tom Tucker360d8732007-12-30 21:07:17 -0600904static struct svc_xprt_ops svc_udp_ops = {
Tom Tucker5d137992007-12-30 21:07:23 -0600905 .xpo_recvfrom = svc_udp_recvfrom,
906 .xpo_sendto = svc_udp_sendto,
Tom Tucker5148bf42007-12-30 21:07:25 -0600907 .xpo_release_rqst = svc_release_skb,
Tom Tucker755ccea2007-12-30 21:07:27 -0600908 .xpo_detach = svc_sock_detach,
909 .xpo_free = svc_sock_free,
Tom Tuckere831fe62007-12-30 21:07:29 -0600910 .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
Tom Tucker323bee32007-12-30 21:07:31 -0600911 .xpo_has_wspace = svc_udp_has_wspace,
Tom Tucker360d8732007-12-30 21:07:17 -0600912};
913
914static struct svc_xprt_class svc_udp_class = {
915 .xcl_name = "udp",
916 .xcl_ops = &svc_udp_ops,
Tom Tucker49023152007-12-30 21:07:21 -0600917 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
Tom Tucker360d8732007-12-30 21:07:17 -0600918};
919
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920static void
921svc_udp_init(struct svc_sock *svsk)
922{
NeilBrown7a37f572007-03-06 01:42:21 -0800923 int one = 1;
924 mm_segment_t oldfs;
925
Tom Tucker360d8732007-12-30 21:07:17 -0600926 svc_xprt_init(&svc_udp_class, &svsk->sk_xprt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927 svsk->sk_sk->sk_data_ready = svc_udp_data_ready;
928 svsk->sk_sk->sk_write_space = svc_write_space;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929
930 /* initialise setting must have enough space to
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -0800931 * receive and respond to one request.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 * svc_udp_recvfrom will re-adjust if necessary
933 */
934 svc_sock_setbufsize(svsk->sk_sock,
NeilBrownc6b0a9f2006-10-06 00:44:05 -0700935 3 * svsk->sk_server->sv_max_mesg,
936 3 * svsk->sk_server->sv_max_mesg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937
938 set_bit(SK_DATA, &svsk->sk_flags); /* might have come in before data_ready set up */
939 set_bit(SK_CHNGBUF, &svsk->sk_flags);
NeilBrown7a37f572007-03-06 01:42:21 -0800940
941 oldfs = get_fs();
942 set_fs(KERNEL_DS);
943 /* make sure we get destination address info */
944 svsk->sk_sock->ops->setsockopt(svsk->sk_sock, IPPROTO_IP, IP_PKTINFO,
945 (char __user *)&one, sizeof(one));
946 set_fs(oldfs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947}
948
949/*
950 * A data_ready event on a listening socket means there's a connection
951 * pending. Do not use state_change as a substitute for it.
952 */
953static void
954svc_tcp_listen_data_ready(struct sock *sk, int count_unused)
955{
Neil Brown939bb7e2005-09-13 01:25:39 -0700956 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957
958 dprintk("svc: socket %p TCP (listen) state change %d\n",
Neil Brown939bb7e2005-09-13 01:25:39 -0700959 sk, sk->sk_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960
Neil Brown939bb7e2005-09-13 01:25:39 -0700961 /*
962 * This callback may called twice when a new connection
963 * is established as a child socket inherits everything
964 * from a parent LISTEN socket.
965 * 1) data_ready method of the parent socket will be called
966 * when one of child sockets become ESTABLISHED.
967 * 2) data_ready method of the child socket may be called
968 * when it receives data before the socket is accepted.
969 * In case of 2, we should ignore it silently.
970 */
971 if (sk->sk_state == TCP_LISTEN) {
972 if (svsk) {
973 set_bit(SK_CONN, &svsk->sk_flags);
974 svc_sock_enqueue(svsk);
975 } else
976 printk("svc: socket %p: no user data\n", sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 }
Neil Brown939bb7e2005-09-13 01:25:39 -0700978
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
980 wake_up_interruptible_all(sk->sk_sleep);
981}
982
983/*
984 * A state change on a connected socket means it's dying or dead.
985 */
986static void
987svc_tcp_state_change(struct sock *sk)
988{
Neil Brown939bb7e2005-09-13 01:25:39 -0700989 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990
991 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
Neil Brown939bb7e2005-09-13 01:25:39 -0700992 sk, sk->sk_state, sk->sk_user_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993
Neil Brown939bb7e2005-09-13 01:25:39 -0700994 if (!svsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 printk("svc: socket %p: no user data\n", sk);
Neil Brown939bb7e2005-09-13 01:25:39 -0700996 else {
997 set_bit(SK_CLOSE, &svsk->sk_flags);
998 svc_sock_enqueue(svsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1001 wake_up_interruptible_all(sk->sk_sleep);
1002}
1003
1004static void
1005svc_tcp_data_ready(struct sock *sk, int count)
1006{
Neil Brown939bb7e2005-09-13 01:25:39 -07001007 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008
1009 dprintk("svc: socket %p TCP data ready (svsk %p)\n",
Neil Brown939bb7e2005-09-13 01:25:39 -07001010 sk, sk->sk_user_data);
1011 if (svsk) {
1012 set_bit(SK_DATA, &svsk->sk_flags);
1013 svc_sock_enqueue(svsk);
1014 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1016 wake_up_interruptible(sk->sk_sleep);
1017}
1018
Chuck Leverbcdb81a2007-02-12 00:53:37 -08001019static inline int svc_port_is_privileged(struct sockaddr *sin)
1020{
1021 switch (sin->sa_family) {
1022 case AF_INET:
1023 return ntohs(((struct sockaddr_in *)sin)->sin_port)
1024 < PROT_SOCK;
Chuck Leverbcdb81a2007-02-12 00:53:37 -08001025 case AF_INET6:
1026 return ntohs(((struct sockaddr_in6 *)sin)->sin6_port)
1027 < PROT_SOCK;
Chuck Leverbcdb81a2007-02-12 00:53:37 -08001028 default:
1029 return 0;
1030 }
1031}
1032
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033/*
1034 * Accept a TCP connection
1035 */
1036static void
1037svc_tcp_accept(struct svc_sock *svsk)
1038{
akpm@linux-foundation.orgcdd88b92007-02-12 00:53:38 -08001039 struct sockaddr_storage addr;
1040 struct sockaddr *sin = (struct sockaddr *) &addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 struct svc_serv *serv = svsk->sk_server;
1042 struct socket *sock = svsk->sk_sock;
1043 struct socket *newsock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 struct svc_sock *newsvsk;
1045 int err, slen;
Chuck Leverad06e4b2007-02-12 00:53:32 -08001046 char buf[RPC_MAX_ADDRBUFLEN];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047
1048 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
1049 if (!sock)
1050 return;
1051
Sridhar Samudralae6242e92006-08-07 20:58:01 -07001052 clear_bit(SK_CONN, &svsk->sk_flags);
1053 err = kernel_accept(sock, &newsock, O_NONBLOCK);
1054 if (err < 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 if (err == -ENOMEM)
1056 printk(KERN_WARNING "%s: no more sockets!\n",
1057 serv->sv_name);
Sridhar Samudralae6242e92006-08-07 20:58:01 -07001058 else if (err != -EAGAIN && net_ratelimit())
1059 printk(KERN_WARNING "%s: accept failed (err %d)!\n",
1060 serv->sv_name, -err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 return;
1062 }
1063
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 set_bit(SK_CONN, &svsk->sk_flags);
1065 svc_sock_enqueue(svsk);
1066
akpm@linux-foundation.orgcdd88b92007-02-12 00:53:38 -08001067 err = kernel_getpeername(newsock, sin, &slen);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068 if (err < 0) {
1069 if (net_ratelimit())
1070 printk(KERN_WARNING "%s: peername failed (err %d)!\n",
1071 serv->sv_name, -err);
1072 goto failed; /* aborted connection or whatever */
1073 }
1074
1075 /* Ideally, we would want to reject connections from unauthorized
Chuck Leverad06e4b2007-02-12 00:53:32 -08001076 * hosts here, but when we get encryption, the IP of the host won't
1077 * tell us anything. For now just warn about unpriv connections.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 */
akpm@linux-foundation.orgcdd88b92007-02-12 00:53:38 -08001079 if (!svc_port_is_privileged(sin)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080 dprintk(KERN_WARNING
Chuck Leverad06e4b2007-02-12 00:53:32 -08001081 "%s: connect from unprivileged port: %s\n",
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -08001082 serv->sv_name,
akpm@linux-foundation.orgcdd88b92007-02-12 00:53:38 -08001083 __svc_print_addr(sin, buf, sizeof(buf)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084 }
Chuck Leverad06e4b2007-02-12 00:53:32 -08001085 dprintk("%s: connect from %s\n", serv->sv_name,
akpm@linux-foundation.orgcdd88b92007-02-12 00:53:38 -08001086 __svc_print_addr(sin, buf, sizeof(buf)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087
1088 /* make sure that a write doesn't block forever when
1089 * low on memory
1090 */
1091 newsock->sk->sk_sndtimeo = HZ*30;
1092
Chuck Lever6b174332007-02-12 00:53:28 -08001093 if (!(newsvsk = svc_setup_socket(serv, newsock, &err,
1094 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095 goto failed;
akpm@linux-foundation.orgcdd88b92007-02-12 00:53:38 -08001096 memcpy(&newsvsk->sk_remote, sin, slen);
Chuck Lever067d7812007-02-12 00:53:30 -08001097 newsvsk->sk_remotelen = slen;
Frank van Maarseveena9747692007-07-09 22:21:39 +02001098 err = kernel_getsockname(newsock, sin, &slen);
1099 if (unlikely(err < 0)) {
1100 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
1101 slen = offsetof(struct sockaddr, sa_data);
1102 }
1103 memcpy(&newsvsk->sk_local, sin, slen);
Chuck Lever067d7812007-02-12 00:53:30 -08001104
NeilBrowne79eff12007-02-12 00:53:30 -08001105 svc_sock_received(newsvsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106
1107 /* make sure that we don't have too many active connections.
1108 * If we have, something must be dropped.
1109 *
1110 * There's no point in trying to do random drop here for
1111 * DoS prevention. The NFS clients does 1 reconnect in 15
1112 * seconds. An attacker can easily beat that.
1113 *
1114 * The only somewhat efficient mechanism would be if drop
1115 * old connections from the same IP first. But right now
1116 * we don't even record the client IP in svc_sock.
1117 */
1118 if (serv->sv_tmpcnt > (serv->sv_nrthreads+3)*20) {
1119 struct svc_sock *svsk = NULL;
1120 spin_lock_bh(&serv->sv_lock);
1121 if (!list_empty(&serv->sv_tempsocks)) {
1122 if (net_ratelimit()) {
1123 /* Try to help the admin */
1124 printk(KERN_NOTICE "%s: too many open TCP "
1125 "sockets, consider increasing the "
1126 "number of nfsd threads\n",
1127 serv->sv_name);
Chuck Leverad06e4b2007-02-12 00:53:32 -08001128 printk(KERN_NOTICE
1129 "%s: last TCP connect from %s\n",
Wolfgang Walter9db619e2007-09-20 15:51:46 -04001130 serv->sv_name, __svc_print_addr(sin,
1131 buf, sizeof(buf)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 }
1133 /*
1134 * Always select the oldest socket. It's not fair,
1135 * but so is life
1136 */
1137 svsk = list_entry(serv->sv_tempsocks.prev,
1138 struct svc_sock,
1139 sk_list);
1140 set_bit(SK_CLOSE, &svsk->sk_flags);
Greg Banksc45c3572006-10-02 02:17:54 -07001141 atomic_inc(&svsk->sk_inuse);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142 }
1143 spin_unlock_bh(&serv->sv_lock);
1144
1145 if (svsk) {
1146 svc_sock_enqueue(svsk);
1147 svc_sock_put(svsk);
1148 }
1149
1150 }
1151
1152 if (serv->sv_stats)
1153 serv->sv_stats->nettcpconn++;
1154
1155 return;
1156
1157failed:
1158 sock_release(newsock);
1159 return;
1160}
1161
1162/*
1163 * Receive data from a TCP socket.
1164 */
1165static int
1166svc_tcp_recvfrom(struct svc_rqst *rqstp)
1167{
1168 struct svc_sock *svsk = rqstp->rq_sock;
1169 struct svc_serv *serv = svsk->sk_server;
1170 int len;
NeilBrown3cc03b12006-10-04 02:15:47 -07001171 struct kvec *vec;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 int pnum, vlen;
1173
1174 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1175 svsk, test_bit(SK_DATA, &svsk->sk_flags),
1176 test_bit(SK_CONN, &svsk->sk_flags),
1177 test_bit(SK_CLOSE, &svsk->sk_flags));
1178
1179 if ((rqstp->rq_deferred = svc_deferred_dequeue(svsk))) {
1180 svc_sock_received(svsk);
1181 return svc_deferred_recv(rqstp);
1182 }
1183
1184 if (test_bit(SK_CLOSE, &svsk->sk_flags)) {
1185 svc_delete_socket(svsk);
1186 return 0;
1187 }
1188
NeilBrown1a047062006-10-19 23:29:13 -07001189 if (svsk->sk_sk->sk_state == TCP_LISTEN) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190 svc_tcp_accept(svsk);
1191 svc_sock_received(svsk);
1192 return 0;
1193 }
1194
1195 if (test_and_clear_bit(SK_CHNGBUF, &svsk->sk_flags))
1196 /* sndbuf needs to have room for one request
1197 * per thread, otherwise we can stall even when the
1198 * network isn't a bottleneck.
Greg Banks3262c812006-10-02 02:17:58 -07001199 *
1200 * We count all threads rather than threads in a
1201 * particular pool, which provides an upper bound
1202 * on the number of threads which will access the socket.
1203 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 * rcvbuf just needs to be able to hold a few requests.
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -08001205 * Normally they will be removed from the queue
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 * as soon a a complete request arrives.
1207 */
1208 svc_sock_setbufsize(svsk->sk_sock,
NeilBrownc6b0a9f2006-10-06 00:44:05 -07001209 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
1210 3 * serv->sv_max_mesg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211
1212 clear_bit(SK_DATA, &svsk->sk_flags);
1213
1214 /* Receive data. If we haven't got the record length yet, get
1215 * the next four bytes. Otherwise try to gobble up as much as
1216 * possible up to the complete record length.
1217 */
1218 if (svsk->sk_tcplen < 4) {
1219 unsigned long want = 4 - svsk->sk_tcplen;
1220 struct kvec iov;
1221
1222 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
1223 iov.iov_len = want;
1224 if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
1225 goto error;
1226 svsk->sk_tcplen += len;
1227
1228 if (len < want) {
1229 dprintk("svc: short recvfrom while reading record length (%d of %lu)\n",
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -08001230 len, want);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 svc_sock_received(svsk);
1232 return -EAGAIN; /* record header not complete */
1233 }
1234
1235 svsk->sk_reclen = ntohl(svsk->sk_reclen);
1236 if (!(svsk->sk_reclen & 0x80000000)) {
1237 /* FIXME: technically, a record can be fragmented,
1238 * and non-terminal fragments will not have the top
1239 * bit set in the fragment length header.
1240 * But apparently no known nfs clients send fragmented
1241 * records. */
NeilBrown34e9a632007-01-29 13:19:52 -08001242 if (net_ratelimit())
1243 printk(KERN_NOTICE "RPC: bad TCP reclen 0x%08lx"
1244 " (non-terminal)\n",
1245 (unsigned long) svsk->sk_reclen);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246 goto err_delete;
1247 }
1248 svsk->sk_reclen &= 0x7fffffff;
1249 dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen);
NeilBrownc6b0a9f2006-10-06 00:44:05 -07001250 if (svsk->sk_reclen > serv->sv_max_mesg) {
NeilBrown34e9a632007-01-29 13:19:52 -08001251 if (net_ratelimit())
1252 printk(KERN_NOTICE "RPC: bad TCP reclen 0x%08lx"
1253 " (large)\n",
1254 (unsigned long) svsk->sk_reclen);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 goto err_delete;
1256 }
1257 }
1258
1259 /* Check whether enough data is available */
1260 len = svc_recv_available(svsk);
1261 if (len < 0)
1262 goto error;
1263
1264 if (len < svsk->sk_reclen) {
1265 dprintk("svc: incomplete TCP record (%d of %d)\n",
1266 len, svsk->sk_reclen);
1267 svc_sock_received(svsk);
1268 return -EAGAIN; /* record not complete */
1269 }
1270 len = svsk->sk_reclen;
1271 set_bit(SK_DATA, &svsk->sk_flags);
1272
NeilBrown3cc03b12006-10-04 02:15:47 -07001273 vec = rqstp->rq_vec;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274 vec[0] = rqstp->rq_arg.head[0];
1275 vlen = PAGE_SIZE;
1276 pnum = 1;
1277 while (vlen < len) {
NeilBrown44524352006-10-04 02:15:46 -07001278 vec[pnum].iov_base = page_address(rqstp->rq_pages[pnum]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 vec[pnum].iov_len = PAGE_SIZE;
1280 pnum++;
1281 vlen += PAGE_SIZE;
1282 }
NeilBrown44524352006-10-04 02:15:46 -07001283 rqstp->rq_respages = &rqstp->rq_pages[pnum];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284
1285 /* Now receive data */
1286 len = svc_recvfrom(rqstp, vec, pnum, len);
1287 if (len < 0)
1288 goto error;
1289
1290 dprintk("svc: TCP complete record (%d bytes)\n", len);
1291 rqstp->rq_arg.len = len;
1292 rqstp->rq_arg.page_base = 0;
1293 if (len <= rqstp->rq_arg.head[0].iov_len) {
1294 rqstp->rq_arg.head[0].iov_len = len;
1295 rqstp->rq_arg.page_len = 0;
1296 } else {
1297 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
1298 }
1299
Tom Tucker5148bf42007-12-30 21:07:25 -06001300 rqstp->rq_xprt_ctxt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301 rqstp->rq_prot = IPPROTO_TCP;
1302
1303 /* Reset TCP read info */
1304 svsk->sk_reclen = 0;
1305 svsk->sk_tcplen = 0;
1306
1307 svc_sock_received(svsk);
1308 if (serv->sv_stats)
1309 serv->sv_stats->nettcpcnt++;
1310
1311 return len;
1312
1313 err_delete:
1314 svc_delete_socket(svsk);
1315 return -EAGAIN;
1316
1317 error:
1318 if (len == -EAGAIN) {
1319 dprintk("RPC: TCP recvfrom got EAGAIN\n");
1320 svc_sock_received(svsk);
1321 } else {
1322 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1323 svsk->sk_server->sv_name, -len);
Olaf Kirch93fbf1a2006-01-06 00:19:56 -08001324 goto err_delete;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325 }
1326
1327 return len;
1328}
1329
1330/*
1331 * Send out data on TCP socket.
1332 */
1333static int
1334svc_tcp_sendto(struct svc_rqst *rqstp)
1335{
1336 struct xdr_buf *xbufp = &rqstp->rq_res;
1337 int sent;
Alexey Dobriyand8ed0292006-09-26 22:29:38 -07001338 __be32 reclen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339
1340 /* Set up the first element of the reply kvec.
1341 * Any other kvecs that may be in use have been taken
1342 * care of by the server implementation itself.
1343 */
1344 reclen = htonl(0x80000000|((xbufp->len ) - 4));
1345 memcpy(xbufp->head[0].iov_base, &reclen, 4);
1346
1347 if (test_bit(SK_DEAD, &rqstp->rq_sock->sk_flags))
1348 return -ENOTCONN;
1349
1350 sent = svc_sendto(rqstp, &rqstp->rq_res);
1351 if (sent != xbufp->len) {
1352 printk(KERN_NOTICE "rpc-srv/tcp: %s: %s %d when sending %d bytes - shutting down socket\n",
1353 rqstp->rq_sock->sk_server->sv_name,
1354 (sent<0)?"got error":"sent only",
1355 sent, xbufp->len);
NeilBrownaaf68cf2007-02-08 14:20:30 -08001356 set_bit(SK_CLOSE, &rqstp->rq_sock->sk_flags);
1357 svc_sock_enqueue(rqstp->rq_sock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358 sent = -EAGAIN;
1359 }
1360 return sent;
1361}
1362
Tom Tuckere831fe62007-12-30 21:07:29 -06001363/*
1364 * Setup response header. TCP has a 4B record length field.
1365 */
1366static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
1367{
1368 struct kvec *resv = &rqstp->rq_res.head[0];
1369
1370 /* tcp needs a space for the record length... */
1371 svc_putnl(resv, 0);
1372}
1373
Tom Tucker323bee32007-12-30 21:07:31 -06001374static int svc_tcp_has_wspace(struct svc_xprt *xprt)
1375{
1376 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1377 struct svc_serv *serv = svsk->sk_server;
1378 int required;
1379 int wspace;
1380
1381 /*
1382 * Set the SOCK_NOSPACE flag before checking the available
1383 * sock space.
1384 */
1385 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
1386 required = atomic_read(&svsk->sk_reserved) + serv->sv_max_mesg;
1387 wspace = sk_stream_wspace(svsk->sk_sk);
1388
1389 if (wspace < sk_stream_min_wspace(svsk->sk_sk))
1390 return 0;
1391 if (required * 2 > wspace)
1392 return 0;
1393
1394 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
1395 return 1;
1396}
1397
Tom Tucker360d8732007-12-30 21:07:17 -06001398static struct svc_xprt_ops svc_tcp_ops = {
Tom Tucker5d137992007-12-30 21:07:23 -06001399 .xpo_recvfrom = svc_tcp_recvfrom,
1400 .xpo_sendto = svc_tcp_sendto,
Tom Tucker5148bf42007-12-30 21:07:25 -06001401 .xpo_release_rqst = svc_release_skb,
Tom Tucker755ccea2007-12-30 21:07:27 -06001402 .xpo_detach = svc_sock_detach,
1403 .xpo_free = svc_sock_free,
Tom Tuckere831fe62007-12-30 21:07:29 -06001404 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
Tom Tucker323bee32007-12-30 21:07:31 -06001405 .xpo_has_wspace = svc_tcp_has_wspace,
Tom Tucker360d8732007-12-30 21:07:17 -06001406};
1407
1408static struct svc_xprt_class svc_tcp_class = {
1409 .xcl_name = "tcp",
1410 .xcl_ops = &svc_tcp_ops,
Tom Tucker49023152007-12-30 21:07:21 -06001411 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
Tom Tucker360d8732007-12-30 21:07:17 -06001412};
1413
1414void svc_init_xprt_sock(void)
1415{
1416 svc_reg_xprt_class(&svc_tcp_class);
1417 svc_reg_xprt_class(&svc_udp_class);
1418}
1419
1420void svc_cleanup_xprt_sock(void)
1421{
1422 svc_unreg_xprt_class(&svc_tcp_class);
1423 svc_unreg_xprt_class(&svc_udp_class);
1424}
1425
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426static void
1427svc_tcp_init(struct svc_sock *svsk)
1428{
1429 struct sock *sk = svsk->sk_sk;
1430 struct tcp_sock *tp = tcp_sk(sk);
1431
Tom Tucker360d8732007-12-30 21:07:17 -06001432 svc_xprt_init(&svc_tcp_class, &svsk->sk_xprt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433
1434 if (sk->sk_state == TCP_LISTEN) {
1435 dprintk("setting up TCP socket for listening\n");
1436 sk->sk_data_ready = svc_tcp_listen_data_ready;
1437 set_bit(SK_CONN, &svsk->sk_flags);
1438 } else {
1439 dprintk("setting up TCP socket for reading\n");
1440 sk->sk_state_change = svc_tcp_state_change;
1441 sk->sk_data_ready = svc_tcp_data_ready;
1442 sk->sk_write_space = svc_write_space;
1443
1444 svsk->sk_reclen = 0;
1445 svsk->sk_tcplen = 0;
1446
1447 tp->nonagle = 1; /* disable Nagle's algorithm */
1448
1449 /* initialise setting must have enough space to
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -08001450 * receive and respond to one request.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451 * svc_tcp_recvfrom will re-adjust if necessary
1452 */
1453 svc_sock_setbufsize(svsk->sk_sock,
NeilBrownc6b0a9f2006-10-06 00:44:05 -07001454 3 * svsk->sk_server->sv_max_mesg,
1455 3 * svsk->sk_server->sv_max_mesg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456
1457 set_bit(SK_CHNGBUF, &svsk->sk_flags);
1458 set_bit(SK_DATA, &svsk->sk_flags);
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -08001459 if (sk->sk_state != TCP_ESTABLISHED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460 set_bit(SK_CLOSE, &svsk->sk_flags);
1461 }
1462}
1463
1464void
1465svc_sock_update_bufs(struct svc_serv *serv)
1466{
1467 /*
1468 * The number of server threads has changed. Update
1469 * rcvbuf and sndbuf accordingly on all sockets
1470 */
1471 struct list_head *le;
1472
1473 spin_lock_bh(&serv->sv_lock);
1474 list_for_each(le, &serv->sv_permsocks) {
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -08001475 struct svc_sock *svsk =
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 list_entry(le, struct svc_sock, sk_list);
1477 set_bit(SK_CHNGBUF, &svsk->sk_flags);
1478 }
1479 list_for_each(le, &serv->sv_tempsocks) {
1480 struct svc_sock *svsk =
1481 list_entry(le, struct svc_sock, sk_list);
1482 set_bit(SK_CHNGBUF, &svsk->sk_flags);
1483 }
1484 spin_unlock_bh(&serv->sv_lock);
1485}
1486
1487/*
Greg Banks3262c812006-10-02 02:17:58 -07001488 * Receive the next request on any socket. This code is carefully
1489 * organised not to touch any cachelines in the shared svc_serv
1490 * structure, only cachelines in the local svc_pool.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 */
1492int
NeilBrown6fb2b472006-10-02 02:17:50 -07001493svc_recv(struct svc_rqst *rqstp, long timeout)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494{
Chuck Lever27459f02007-02-12 00:53:34 -08001495 struct svc_sock *svsk = NULL;
NeilBrown6fb2b472006-10-02 02:17:50 -07001496 struct svc_serv *serv = rqstp->rq_server;
Greg Banks3262c812006-10-02 02:17:58 -07001497 struct svc_pool *pool = rqstp->rq_pool;
NeilBrown44524352006-10-04 02:15:46 -07001498 int len, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 int pages;
1500 struct xdr_buf *arg;
1501 DECLARE_WAITQUEUE(wait, current);
1502
1503 dprintk("svc: server %p waiting for data (to = %ld)\n",
1504 rqstp, timeout);
1505
1506 if (rqstp->rq_sock)
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -08001507 printk(KERN_ERR
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 "svc_recv: service %p, socket not NULL!\n",
1509 rqstp);
1510 if (waitqueue_active(&rqstp->rq_wait))
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -08001511 printk(KERN_ERR
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512 "svc_recv: service %p, wait queue active!\n",
1513 rqstp);
1514
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515
1516 /* now allocate needed pages. If we get a failure, sleep briefly */
NeilBrownc6b0a9f2006-10-06 00:44:05 -07001517 pages = (serv->sv_max_mesg + PAGE_SIZE) / PAGE_SIZE;
NeilBrown44524352006-10-04 02:15:46 -07001518 for (i=0; i < pages ; i++)
1519 while (rqstp->rq_pages[i] == NULL) {
1520 struct page *p = alloc_page(GFP_KERNEL);
1521 if (!p)
1522 schedule_timeout_uninterruptible(msecs_to_jiffies(500));
1523 rqstp->rq_pages[i] = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524 }
NeilBrown250f3912007-01-26 00:56:59 -08001525 rqstp->rq_pages[i++] = NULL; /* this might be seen in nfs_read_actor */
1526 BUG_ON(pages >= RPCSVC_MAXPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527
1528 /* Make arg->head point to first page and arg->pages point to rest */
1529 arg = &rqstp->rq_arg;
NeilBrown44524352006-10-04 02:15:46 -07001530 arg->head[0].iov_base = page_address(rqstp->rq_pages[0]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531 arg->head[0].iov_len = PAGE_SIZE;
NeilBrown44524352006-10-04 02:15:46 -07001532 arg->pages = rqstp->rq_pages + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533 arg->page_base = 0;
1534 /* save at least one page for response */
1535 arg->page_len = (pages-2)*PAGE_SIZE;
1536 arg->len = (pages-1)*PAGE_SIZE;
1537 arg->tail[0].iov_len = 0;
Christoph Lameter3e1d1d22005-06-24 23:13:50 -07001538
1539 try_to_freeze();
NeilBrown1887b932005-11-15 00:09:10 -08001540 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541 if (signalled())
1542 return -EINTR;
1543
Greg Banks3262c812006-10-02 02:17:58 -07001544 spin_lock_bh(&pool->sp_lock);
1545 if ((svsk = svc_sock_dequeue(pool)) != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546 rqstp->rq_sock = svsk;
Greg Banksc45c3572006-10-02 02:17:54 -07001547 atomic_inc(&svsk->sk_inuse);
NeilBrownc6b0a9f2006-10-06 00:44:05 -07001548 rqstp->rq_reserved = serv->sv_max_mesg;
Greg Banks5685f0f2006-10-02 02:17:56 -07001549 atomic_add(rqstp->rq_reserved, &svsk->sk_reserved);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550 } else {
1551 /* No data pending. Go to sleep */
Greg Banks3262c812006-10-02 02:17:58 -07001552 svc_thread_enqueue(pool, rqstp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553
1554 /*
1555 * We have to be able to interrupt this wait
1556 * to bring down the daemons ...
1557 */
1558 set_current_state(TASK_INTERRUPTIBLE);
1559 add_wait_queue(&rqstp->rq_wait, &wait);
Greg Banks3262c812006-10-02 02:17:58 -07001560 spin_unlock_bh(&pool->sp_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561
1562 schedule_timeout(timeout);
1563
Christoph Lameter3e1d1d22005-06-24 23:13:50 -07001564 try_to_freeze();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565
Greg Banks3262c812006-10-02 02:17:58 -07001566 spin_lock_bh(&pool->sp_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567 remove_wait_queue(&rqstp->rq_wait, &wait);
1568
1569 if (!(svsk = rqstp->rq_sock)) {
Greg Banks3262c812006-10-02 02:17:58 -07001570 svc_thread_dequeue(pool, rqstp);
1571 spin_unlock_bh(&pool->sp_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572 dprintk("svc: server %p, no data yet\n", rqstp);
1573 return signalled()? -EINTR : -EAGAIN;
1574 }
1575 }
Greg Banks3262c812006-10-02 02:17:58 -07001576 spin_unlock_bh(&pool->sp_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577
Greg Banks3262c812006-10-02 02:17:58 -07001578 dprintk("svc: server %p, pool %u, socket %p, inuse=%d\n",
1579 rqstp, pool->sp_id, svsk, atomic_read(&svsk->sk_inuse));
Tom Tucker5d137992007-12-30 21:07:23 -06001580 len = svsk->sk_xprt.xpt_ops->xpo_recvfrom(rqstp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 dprintk("svc: got len=%d\n", len);
1582
1583 /* No data, incomplete (TCP) read, or accept() */
1584 if (len == 0 || len == -EAGAIN) {
1585 rqstp->rq_res.len = 0;
1586 svc_sock_release(rqstp);
1587 return -EAGAIN;
1588 }
1589 svsk->sk_lastrecv = get_seconds();
Greg Banks36bdfc82006-10-02 02:17:54 -07001590 clear_bit(SK_OLD, &svsk->sk_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591
Chuck Leverbcdb81a2007-02-12 00:53:37 -08001592 rqstp->rq_secure = svc_port_is_privileged(svc_addr(rqstp));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593 rqstp->rq_chandle.defer = svc_defer;
1594
1595 if (serv->sv_stats)
1596 serv->sv_stats->netcnt++;
1597 return len;
1598}
1599
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -08001600/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601 * Drop request
1602 */
1603void
1604svc_drop(struct svc_rqst *rqstp)
1605{
1606 dprintk("svc: socket %p dropped request\n", rqstp->rq_sock);
1607 svc_sock_release(rqstp);
1608}
1609
1610/*
1611 * Return reply to client.
1612 */
1613int
1614svc_send(struct svc_rqst *rqstp)
1615{
1616 struct svc_sock *svsk;
1617 int len;
1618 struct xdr_buf *xb;
1619
1620 if ((svsk = rqstp->rq_sock) == NULL) {
1621 printk(KERN_WARNING "NULL socket pointer in %s:%d\n",
1622 __FILE__, __LINE__);
1623 return -EFAULT;
1624 }
1625
1626 /* release the receive skb before sending the reply */
Tom Tucker5148bf42007-12-30 21:07:25 -06001627 rqstp->rq_xprt->xpt_ops->xpo_release_rqst(rqstp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628
1629 /* calculate over-all length */
1630 xb = & rqstp->rq_res;
1631 xb->len = xb->head[0].iov_len +
1632 xb->page_len +
1633 xb->tail[0].iov_len;
1634
Ingo Molnar57b47a52006-03-20 22:35:41 -08001635 /* Grab svsk->sk_mutex to serialize outgoing data. */
1636 mutex_lock(&svsk->sk_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 if (test_bit(SK_DEAD, &svsk->sk_flags))
1638 len = -ENOTCONN;
1639 else
Tom Tucker5d137992007-12-30 21:07:23 -06001640 len = svsk->sk_xprt.xpt_ops->xpo_sendto(rqstp);
Ingo Molnar57b47a52006-03-20 22:35:41 -08001641 mutex_unlock(&svsk->sk_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642 svc_sock_release(rqstp);
1643
1644 if (len == -ECONNREFUSED || len == -ENOTCONN || len == -EAGAIN)
1645 return 0;
1646 return len;
1647}
1648
1649/*
Greg Banks36bdfc82006-10-02 02:17:54 -07001650 * Timer function to close old temporary sockets, using
1651 * a mark-and-sweep algorithm.
1652 */
1653static void
1654svc_age_temp_sockets(unsigned long closure)
1655{
1656 struct svc_serv *serv = (struct svc_serv *)closure;
1657 struct svc_sock *svsk;
1658 struct list_head *le, *next;
1659 LIST_HEAD(to_be_aged);
1660
1661 dprintk("svc_age_temp_sockets\n");
1662
1663 if (!spin_trylock_bh(&serv->sv_lock)) {
1664 /* busy, try again 1 sec later */
1665 dprintk("svc_age_temp_sockets: busy\n");
1666 mod_timer(&serv->sv_temptimer, jiffies + HZ);
1667 return;
1668 }
1669
1670 list_for_each_safe(le, next, &serv->sv_tempsocks) {
1671 svsk = list_entry(le, struct svc_sock, sk_list);
1672
1673 if (!test_and_set_bit(SK_OLD, &svsk->sk_flags))
1674 continue;
Neil Brown7a1fa062007-09-14 10:28:08 -04001675 if (atomic_read(&svsk->sk_inuse) > 1 || test_bit(SK_BUSY, &svsk->sk_flags))
Greg Banks36bdfc82006-10-02 02:17:54 -07001676 continue;
Greg Banksc45c3572006-10-02 02:17:54 -07001677 atomic_inc(&svsk->sk_inuse);
Greg Banks36bdfc82006-10-02 02:17:54 -07001678 list_move(le, &to_be_aged);
1679 set_bit(SK_CLOSE, &svsk->sk_flags);
1680 set_bit(SK_DETACHED, &svsk->sk_flags);
1681 }
1682 spin_unlock_bh(&serv->sv_lock);
1683
1684 while (!list_empty(&to_be_aged)) {
1685 le = to_be_aged.next;
1686 /* fiddling the sk_list node is safe 'cos we're SK_DETACHED */
1687 list_del_init(le);
1688 svsk = list_entry(le, struct svc_sock, sk_list);
1689
1690 dprintk("queuing svsk %p for closing, %lu seconds old\n",
1691 svsk, get_seconds() - svsk->sk_lastrecv);
1692
1693 /* a thread will dequeue and close it soon */
1694 svc_sock_enqueue(svsk);
1695 svc_sock_put(svsk);
1696 }
1697
1698 mod_timer(&serv->sv_temptimer, jiffies + svc_conn_age_period * HZ);
1699}
1700
1701/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702 * Initialize socket for RPC use and create svc_sock struct
1703 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1704 */
Chuck Lever6b174332007-02-12 00:53:28 -08001705static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1706 struct socket *sock,
1707 int *errp, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708{
1709 struct svc_sock *svsk;
1710 struct sock *inet;
Chuck Lever6b174332007-02-12 00:53:28 -08001711 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1712 int is_temporary = flags & SVC_SOCK_TEMPORARY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713
1714 dprintk("svc: svc_setup_socket %p\n", sock);
Panagiotis Issaris0da974f2006-07-21 14:51:30 -07001715 if (!(svsk = kzalloc(sizeof(*svsk), GFP_KERNEL))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716 *errp = -ENOMEM;
1717 return NULL;
1718 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719
1720 inet = sock->sk;
1721
1722 /* Register socket with portmapper */
1723 if (*errp >= 0 && pmap_register)
1724 *errp = svc_register(serv, inet->sk_protocol,
1725 ntohs(inet_sk(inet)->sport));
1726
1727 if (*errp < 0) {
1728 kfree(svsk);
1729 return NULL;
1730 }
1731
1732 set_bit(SK_BUSY, &svsk->sk_flags);
1733 inet->sk_user_data = svsk;
1734 svsk->sk_sock = sock;
1735 svsk->sk_sk = inet;
1736 svsk->sk_ostate = inet->sk_state_change;
1737 svsk->sk_odata = inet->sk_data_ready;
1738 svsk->sk_owspace = inet->sk_write_space;
1739 svsk->sk_server = serv;
NeilBrownaaf68cf2007-02-08 14:20:30 -08001740 atomic_set(&svsk->sk_inuse, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741 svsk->sk_lastrecv = get_seconds();
NeilBrown7ac1bea2007-05-09 02:34:48 -07001742 spin_lock_init(&svsk->sk_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743 INIT_LIST_HEAD(&svsk->sk_deferred);
1744 INIT_LIST_HEAD(&svsk->sk_ready);
Ingo Molnar57b47a52006-03-20 22:35:41 -08001745 mutex_init(&svsk->sk_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746
1747 /* Initialize the socket */
1748 if (sock->type == SOCK_DGRAM)
1749 svc_udp_init(svsk);
1750 else
1751 svc_tcp_init(svsk);
1752
1753 spin_lock_bh(&serv->sv_lock);
Chuck Lever6b174332007-02-12 00:53:28 -08001754 if (is_temporary) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755 set_bit(SK_TEMP, &svsk->sk_flags);
1756 list_add(&svsk->sk_list, &serv->sv_tempsocks);
1757 serv->sv_tmpcnt++;
Greg Banks36bdfc82006-10-02 02:17:54 -07001758 if (serv->sv_temptimer.function == NULL) {
1759 /* setup timer to age temp sockets */
1760 setup_timer(&serv->sv_temptimer, svc_age_temp_sockets,
1761 (unsigned long)serv);
1762 mod_timer(&serv->sv_temptimer,
1763 jiffies + svc_conn_age_period * HZ);
1764 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765 } else {
1766 clear_bit(SK_TEMP, &svsk->sk_flags);
1767 list_add(&svsk->sk_list, &serv->sv_permsocks);
1768 }
1769 spin_unlock_bh(&serv->sv_lock);
1770
1771 dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1772 svsk, svsk->sk_sk);
1773
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774 return svsk;
1775}
1776
NeilBrownb41b66d2006-10-02 02:17:48 -07001777int svc_addsock(struct svc_serv *serv,
1778 int fd,
1779 char *name_return,
1780 int *proto)
1781{
1782 int err = 0;
1783 struct socket *so = sockfd_lookup(fd, &err);
1784 struct svc_sock *svsk = NULL;
1785
1786 if (!so)
1787 return err;
1788 if (so->sk->sk_family != AF_INET)
1789 err = -EAFNOSUPPORT;
1790 else if (so->sk->sk_protocol != IPPROTO_TCP &&
1791 so->sk->sk_protocol != IPPROTO_UDP)
1792 err = -EPROTONOSUPPORT;
1793 else if (so->state > SS_UNCONNECTED)
1794 err = -EISCONN;
1795 else {
Chuck Lever6b174332007-02-12 00:53:28 -08001796 svsk = svc_setup_socket(serv, so, &err, SVC_SOCK_DEFAULTS);
NeilBrowne79eff12007-02-12 00:53:30 -08001797 if (svsk) {
1798 svc_sock_received(svsk);
NeilBrownb41b66d2006-10-02 02:17:48 -07001799 err = 0;
NeilBrowne79eff12007-02-12 00:53:30 -08001800 }
NeilBrownb41b66d2006-10-02 02:17:48 -07001801 }
1802 if (err) {
1803 sockfd_put(so);
1804 return err;
1805 }
1806 if (proto) *proto = so->sk->sk_protocol;
1807 return one_sock_name(name_return, svsk);
1808}
1809EXPORT_SYMBOL_GPL(svc_addsock);
1810
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811/*
1812 * Create socket for RPC service.
1813 */
Chuck Lever6b174332007-02-12 00:53:28 -08001814static int svc_create_socket(struct svc_serv *serv, int protocol,
Chuck Lever77f1f672007-02-12 00:53:39 -08001815 struct sockaddr *sin, int len, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816{
1817 struct svc_sock *svsk;
1818 struct socket *sock;
1819 int error;
1820 int type;
Chuck Leverad06e4b2007-02-12 00:53:32 -08001821 char buf[RPC_MAX_ADDRBUFLEN];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822
Chuck Leverad06e4b2007-02-12 00:53:32 -08001823 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1824 serv->sv_program->pg_name, protocol,
Chuck Lever77f1f672007-02-12 00:53:39 -08001825 __svc_print_addr(sin, buf, sizeof(buf)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826
1827 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1828 printk(KERN_WARNING "svc: only UDP and TCP "
1829 "sockets supported\n");
1830 return -EINVAL;
1831 }
1832 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1833
Chuck Lever77f1f672007-02-12 00:53:39 -08001834 error = sock_create_kern(sin->sa_family, type, protocol, &sock);
1835 if (error < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836 return error;
1837
Peter Zijlstraed075362006-12-06 20:35:24 -08001838 svc_reclassify_socket(sock);
1839
Eric Sesterhenn18114742006-09-28 14:37:07 -07001840 if (type == SOCK_STREAM)
Chuck Lever77f1f672007-02-12 00:53:39 -08001841 sock->sk->sk_reuse = 1; /* allow address reuse */
1842 error = kernel_bind(sock, sin, len);
Eric Sesterhenn18114742006-09-28 14:37:07 -07001843 if (error < 0)
1844 goto bummer;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845
1846 if (protocol == IPPROTO_TCP) {
Sridhar Samudralae6242e92006-08-07 20:58:01 -07001847 if ((error = kernel_listen(sock, 64)) < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001848 goto bummer;
1849 }
1850
NeilBrowne79eff12007-02-12 00:53:30 -08001851 if ((svsk = svc_setup_socket(serv, sock, &error, flags)) != NULL) {
1852 svc_sock_received(svsk);
Chuck Lever6b174332007-02-12 00:53:28 -08001853 return ntohs(inet_sk(svsk->sk_sk)->sport);
NeilBrowne79eff12007-02-12 00:53:30 -08001854 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855
1856bummer:
1857 dprintk("svc: svc_create_socket error = %d\n", -error);
1858 sock_release(sock);
1859 return error;
1860}
1861
1862/*
Tom Tucker755ccea2007-12-30 21:07:27 -06001863 * Detach the svc_sock from the socket so that no
1864 * more callbacks occur.
1865 */
1866static void svc_sock_detach(struct svc_xprt *xprt)
1867{
1868 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1869 struct sock *sk = svsk->sk_sk;
1870
1871 dprintk("svc: svc_sock_detach(%p)\n", svsk);
1872
1873 /* put back the old socket callbacks */
1874 sk->sk_state_change = svsk->sk_ostate;
1875 sk->sk_data_ready = svsk->sk_odata;
1876 sk->sk_write_space = svsk->sk_owspace;
1877}
1878
1879/*
1880 * Free the svc_sock's socket resources and the svc_sock itself.
1881 */
1882static void svc_sock_free(struct svc_xprt *xprt)
1883{
1884 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1885 dprintk("svc: svc_sock_free(%p)\n", svsk);
1886
1887 if (svsk->sk_info_authunix != NULL)
1888 svcauth_unix_info_release(svsk->sk_info_authunix);
1889 if (svsk->sk_sock->file)
1890 sockfd_put(svsk->sk_sock);
1891 else
1892 sock_release(svsk->sk_sock);
1893 kfree(svsk);
1894}
1895
1896/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001897 * Remove a dead socket
1898 */
NeilBrownaaf68cf2007-02-08 14:20:30 -08001899static void
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900svc_delete_socket(struct svc_sock *svsk)
1901{
1902 struct svc_serv *serv;
1903 struct sock *sk;
1904
1905 dprintk("svc: svc_delete_socket(%p)\n", svsk);
1906
1907 serv = svsk->sk_server;
1908 sk = svsk->sk_sk;
1909
Tom Tucker755ccea2007-12-30 21:07:27 -06001910 svsk->sk_xprt.xpt_ops->xpo_detach(&svsk->sk_xprt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911
1912 spin_lock_bh(&serv->sv_lock);
1913
Greg Banks36bdfc82006-10-02 02:17:54 -07001914 if (!test_and_set_bit(SK_DETACHED, &svsk->sk_flags))
1915 list_del_init(&svsk->sk_list);
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -08001916 /*
Greg Banks3262c812006-10-02 02:17:58 -07001917 * We used to delete the svc_sock from whichever list
1918 * it's sk_ready node was on, but we don't actually
1919 * need to. This is because the only time we're called
1920 * while still attached to a queue, the queue itself
1921 * is about to be destroyed (in svc_destroy).
1922 */
NeilBrownaaf68cf2007-02-08 14:20:30 -08001923 if (!test_and_set_bit(SK_DEAD, &svsk->sk_flags)) {
1924 BUG_ON(atomic_read(&svsk->sk_inuse)<2);
1925 atomic_dec(&svsk->sk_inuse);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926 if (test_bit(SK_TEMP, &svsk->sk_flags))
1927 serv->sv_tmpcnt--;
NeilBrownaaf68cf2007-02-08 14:20:30 -08001928 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929
Neil Brownd6740df2006-10-29 22:46:45 -08001930 spin_unlock_bh(&serv->sv_lock);
NeilBrownaaf68cf2007-02-08 14:20:30 -08001931}
1932
NeilBrowncda1fd42007-03-06 01:42:22 -08001933static void svc_close_socket(struct svc_sock *svsk)
NeilBrownaaf68cf2007-02-08 14:20:30 -08001934{
1935 set_bit(SK_CLOSE, &svsk->sk_flags);
1936 if (test_and_set_bit(SK_BUSY, &svsk->sk_flags))
1937 /* someone else will have to effect the close */
1938 return;
1939
1940 atomic_inc(&svsk->sk_inuse);
1941 svc_delete_socket(svsk);
1942 clear_bit(SK_BUSY, &svsk->sk_flags);
Neil Brownd6740df2006-10-29 22:46:45 -08001943 svc_sock_put(svsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944}
1945
NeilBrowncda1fd42007-03-06 01:42:22 -08001946void svc_force_close_socket(struct svc_sock *svsk)
1947{
1948 set_bit(SK_CLOSE, &svsk->sk_flags);
1949 if (test_bit(SK_BUSY, &svsk->sk_flags)) {
1950 /* Waiting to be processed, but no threads left,
1951 * So just remove it from the waiting list
1952 */
1953 list_del_init(&svsk->sk_ready);
1954 clear_bit(SK_BUSY, &svsk->sk_flags);
1955 }
1956 svc_close_socket(svsk);
1957}
1958
Chuck Lever6b174332007-02-12 00:53:28 -08001959/**
1960 * svc_makesock - Make a socket for nfsd and lockd
1961 * @serv: RPC server structure
1962 * @protocol: transport protocol to use
1963 * @port: port to use
Chuck Lever482fb942007-02-12 00:53:29 -08001964 * @flags: requested socket characteristics
Chuck Lever6b174332007-02-12 00:53:28 -08001965 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001966 */
Chuck Lever482fb942007-02-12 00:53:29 -08001967int svc_makesock(struct svc_serv *serv, int protocol, unsigned short port,
1968 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969{
Chuck Lever6b174332007-02-12 00:53:28 -08001970 struct sockaddr_in sin = {
1971 .sin_family = AF_INET,
1972 .sin_addr.s_addr = INADDR_ANY,
1973 .sin_port = htons(port),
1974 };
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975
1976 dprintk("svc: creating socket proto = %d\n", protocol);
Chuck Lever77f1f672007-02-12 00:53:39 -08001977 return svc_create_socket(serv, protocol, (struct sockaddr *) &sin,
1978 sizeof(sin), flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979}
1980
1981/*
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -08001982 * Handle defer and revisit of requests
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983 */
1984
1985static void svc_revisit(struct cache_deferred_req *dreq, int too_many)
1986{
1987 struct svc_deferred_req *dr = container_of(dreq, struct svc_deferred_req, handle);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988 struct svc_sock *svsk;
1989
1990 if (too_many) {
1991 svc_sock_put(dr->svsk);
1992 kfree(dr);
1993 return;
1994 }
1995 dprintk("revisit queued\n");
1996 svsk = dr->svsk;
1997 dr->svsk = NULL;
NeilBrown7ac1bea2007-05-09 02:34:48 -07001998 spin_lock(&svsk->sk_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999 list_add(&dr->handle.recent, &svsk->sk_deferred);
NeilBrown7ac1bea2007-05-09 02:34:48 -07002000 spin_unlock(&svsk->sk_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001 set_bit(SK_DEFERRED, &svsk->sk_flags);
2002 svc_sock_enqueue(svsk);
2003 svc_sock_put(svsk);
2004}
2005
2006static struct cache_deferred_req *
2007svc_defer(struct cache_req *req)
2008{
2009 struct svc_rqst *rqstp = container_of(req, struct svc_rqst, rq_chandle);
2010 int size = sizeof(struct svc_deferred_req) + (rqstp->rq_arg.len);
2011 struct svc_deferred_req *dr;
2012
2013 if (rqstp->rq_arg.page_len)
2014 return NULL; /* if more than a page, give up FIXME */
2015 if (rqstp->rq_deferred) {
2016 dr = rqstp->rq_deferred;
2017 rqstp->rq_deferred = NULL;
2018 } else {
2019 int skip = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
2020 /* FIXME maybe discard if size too large */
2021 dr = kmalloc(size, GFP_KERNEL);
2022 if (dr == NULL)
2023 return NULL;
2024
2025 dr->handle.owner = rqstp->rq_server;
2026 dr->prot = rqstp->rq_prot;
Chuck Lever24422222007-02-12 00:53:33 -08002027 memcpy(&dr->addr, &rqstp->rq_addr, rqstp->rq_addrlen);
2028 dr->addrlen = rqstp->rq_addrlen;
J. Bruce Fields1918e342006-01-18 17:43:16 -08002029 dr->daddr = rqstp->rq_daddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030 dr->argslen = rqstp->rq_arg.len >> 2;
2031 memcpy(dr->args, rqstp->rq_arg.head[0].iov_base-skip, dr->argslen<<2);
2032 }
Greg Banksc45c3572006-10-02 02:17:54 -07002033 atomic_inc(&rqstp->rq_sock->sk_inuse);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034 dr->svsk = rqstp->rq_sock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035
2036 dr->handle.revisit = svc_revisit;
2037 return &dr->handle;
2038}
2039
2040/*
2041 * recv data from a deferred request into an active one
2042 */
2043static int svc_deferred_recv(struct svc_rqst *rqstp)
2044{
2045 struct svc_deferred_req *dr = rqstp->rq_deferred;
2046
2047 rqstp->rq_arg.head[0].iov_base = dr->args;
2048 rqstp->rq_arg.head[0].iov_len = dr->argslen<<2;
2049 rqstp->rq_arg.page_len = 0;
2050 rqstp->rq_arg.len = dr->argslen<<2;
2051 rqstp->rq_prot = dr->prot;
Chuck Lever24422222007-02-12 00:53:33 -08002052 memcpy(&rqstp->rq_addr, &dr->addr, dr->addrlen);
2053 rqstp->rq_addrlen = dr->addrlen;
J. Bruce Fields1918e342006-01-18 17:43:16 -08002054 rqstp->rq_daddr = dr->daddr;
NeilBrown44524352006-10-04 02:15:46 -07002055 rqstp->rq_respages = rqstp->rq_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056 return dr->argslen<<2;
2057}
2058
2059
2060static struct svc_deferred_req *svc_deferred_dequeue(struct svc_sock *svsk)
2061{
2062 struct svc_deferred_req *dr = NULL;
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -08002063
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064 if (!test_bit(SK_DEFERRED, &svsk->sk_flags))
2065 return NULL;
NeilBrown7ac1bea2007-05-09 02:34:48 -07002066 spin_lock(&svsk->sk_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067 clear_bit(SK_DEFERRED, &svsk->sk_flags);
2068 if (!list_empty(&svsk->sk_deferred)) {
2069 dr = list_entry(svsk->sk_deferred.next,
2070 struct svc_deferred_req,
2071 handle.recent);
2072 list_del_init(&dr->handle.recent);
2073 set_bit(SK_DEFERRED, &svsk->sk_flags);
2074 }
NeilBrown7ac1bea2007-05-09 02:34:48 -07002075 spin_unlock(&svsk->sk_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076 return dr;
2077}