Commit | Line | Data |
---|---|---|
1d8206b9 TT |
1 | /* |
2 | * linux/net/sunrpc/svc_xprt.c | |
3 | * | |
4 | * Author: Tom Tucker <tom@opengridcomputing.com> | |
5 | */ | |
6 | ||
7 | #include <linux/sched.h> | |
8 | #include <linux/errno.h> | |
1d8206b9 | 9 | #include <linux/freezer.h> |
7086721f | 10 | #include <linux/kthread.h> |
5a0e3ad6 | 11 | #include <linux/slab.h> |
1d8206b9 | 12 | #include <net/sock.h> |
1d8206b9 TT |
13 | #include <linux/sunrpc/stats.h> |
14 | #include <linux/sunrpc/svc_xprt.h> | |
dcf1a357 | 15 | #include <linux/sunrpc/svcsock.h> |
1d8206b9 TT |
16 | |
17 | #define RPCDBG_FACILITY RPCDBG_SVCXPRT | |
18 | ||
0f0257ea TT |
19 | static struct svc_deferred_req *svc_deferred_dequeue(struct svc_xprt *xprt); |
20 | static int svc_deferred_recv(struct svc_rqst *rqstp); | |
21 | static struct cache_deferred_req *svc_defer(struct cache_req *req); | |
22 | static void svc_age_temp_xprts(unsigned long closure); | |
23 | ||
24 | /* apparently the "standard" is that clients close | |
25 | * idle connections after 5 minutes, servers after | |
26 | * 6 minutes | |
27 | * http://www.connectathon.org/talks96/nfstcp.pdf | |
28 | */ | |
29 | static int svc_conn_age_period = 6*60; | |
30 | ||
1d8206b9 TT |
31 | /* List of registered transport classes */ |
32 | static DEFINE_SPINLOCK(svc_xprt_class_lock); | |
33 | static LIST_HEAD(svc_xprt_class_list); | |
34 | ||
0f0257ea TT |
35 | /* SMP locking strategy: |
36 | * | |
37 | * svc_pool->sp_lock protects most of the fields of that pool. | |
38 | * svc_serv->sv_lock protects sv_tempsocks, sv_permsocks, sv_tmpcnt. | |
39 | * when both need to be taken (rare), svc_serv->sv_lock is first. | |
40 | * BKL protects svc_serv->sv_nrthread. | |
41 | * svc_sock->sk_lock protects the svc_sock->sk_deferred list | |
42 | * and the ->sk_info_authunix cache. | |
43 | * | |
44 | * The XPT_BUSY bit in xprt->xpt_flags prevents a transport being | |
45 | * enqueued multiply. During normal transport processing this bit | |
46 | * is set by svc_xprt_enqueue and cleared by svc_xprt_received. | |
47 | * Providers should not manipulate this bit directly. | |
48 | * | |
49 | * Some flags can be set to certain values at any time | |
50 | * providing that certain rules are followed: | |
51 | * | |
52 | * XPT_CONN, XPT_DATA: | |
53 | * - Can be set or cleared at any time. | |
54 | * - After a set, svc_xprt_enqueue must be called to enqueue | |
55 | * the transport for processing. | |
56 | * - After a clear, the transport must be read/accepted. | |
57 | * If this succeeds, it must be set again. | |
58 | * XPT_CLOSE: | |
59 | * - Can set at any time. It is never cleared. | |
60 | * XPT_DEAD: | |
61 | * - Can only be set while XPT_BUSY is held which ensures | |
62 | * that no other thread will be using the transport or will | |
63 | * try to set XPT_DEAD. | |
64 | */ | |
65 | ||
1d8206b9 TT |
66 | int svc_reg_xprt_class(struct svc_xprt_class *xcl) |
67 | { | |
68 | struct svc_xprt_class *cl; | |
69 | int res = -EEXIST; | |
70 | ||
71 | dprintk("svc: Adding svc transport class '%s'\n", xcl->xcl_name); | |
72 | ||
73 | INIT_LIST_HEAD(&xcl->xcl_list); | |
74 | spin_lock(&svc_xprt_class_lock); | |
75 | /* Make sure there isn't already a class with the same name */ | |
76 | list_for_each_entry(cl, &svc_xprt_class_list, xcl_list) { | |
77 | if (strcmp(xcl->xcl_name, cl->xcl_name) == 0) | |
78 | goto out; | |
79 | } | |
80 | list_add_tail(&xcl->xcl_list, &svc_xprt_class_list); | |
81 | res = 0; | |
82 | out: | |
83 | spin_unlock(&svc_xprt_class_lock); | |
84 | return res; | |
85 | } | |
86 | EXPORT_SYMBOL_GPL(svc_reg_xprt_class); | |
87 | ||
88 | void svc_unreg_xprt_class(struct svc_xprt_class *xcl) | |
89 | { | |
90 | dprintk("svc: Removing svc transport class '%s'\n", xcl->xcl_name); | |
91 | spin_lock(&svc_xprt_class_lock); | |
92 | list_del_init(&xcl->xcl_list); | |
93 | spin_unlock(&svc_xprt_class_lock); | |
94 | } | |
95 | EXPORT_SYMBOL_GPL(svc_unreg_xprt_class); | |
96 | ||
dc9a16e4 TT |
97 | /* |
98 | * Format the transport list for printing | |
99 | */ | |
100 | int svc_print_xprts(char *buf, int maxlen) | |
101 | { | |
8f3a6de3 | 102 | struct svc_xprt_class *xcl; |
dc9a16e4 TT |
103 | char tmpstr[80]; |
104 | int len = 0; | |
105 | buf[0] = '\0'; | |
106 | ||
107 | spin_lock(&svc_xprt_class_lock); | |
8f3a6de3 | 108 | list_for_each_entry(xcl, &svc_xprt_class_list, xcl_list) { |
dc9a16e4 | 109 | int slen; |
dc9a16e4 TT |
110 | |
111 | sprintf(tmpstr, "%s %d\n", xcl->xcl_name, xcl->xcl_max_payload); | |
112 | slen = strlen(tmpstr); | |
113 | if (len + slen > maxlen) | |
114 | break; | |
115 | len += slen; | |
116 | strcat(buf, tmpstr); | |
117 | } | |
118 | spin_unlock(&svc_xprt_class_lock); | |
119 | ||
120 | return len; | |
121 | } | |
122 | ||
e1b3157f TT |
123 | static void svc_xprt_free(struct kref *kref) |
124 | { | |
125 | struct svc_xprt *xprt = | |
126 | container_of(kref, struct svc_xprt, xpt_ref); | |
127 | struct module *owner = xprt->xpt_class->xcl_owner; | |
e3bfca01 PE |
128 | if (test_bit(XPT_CACHE_AUTH, &xprt->xpt_flags)) |
129 | svcauth_unix_info_release(xprt); | |
4fb8518b | 130 | put_net(xprt->xpt_net); |
e1b3157f TT |
131 | xprt->xpt_ops->xpo_free(xprt); |
132 | module_put(owner); | |
133 | } | |
134 | ||
135 | void svc_xprt_put(struct svc_xprt *xprt) | |
136 | { | |
137 | kref_put(&xprt->xpt_ref, svc_xprt_free); | |
138 | } | |
139 | EXPORT_SYMBOL_GPL(svc_xprt_put); | |
140 | ||
1d8206b9 TT |
141 | /* |
142 | * Called by transport drivers to initialize the transport independent | |
143 | * portion of the transport instance. | |
144 | */ | |
bb5cf160 TT |
145 | void svc_xprt_init(struct svc_xprt_class *xcl, struct svc_xprt *xprt, |
146 | struct svc_serv *serv) | |
1d8206b9 TT |
147 | { |
148 | memset(xprt, 0, sizeof(*xprt)); | |
149 | xprt->xpt_class = xcl; | |
150 | xprt->xpt_ops = xcl->xcl_ops; | |
e1b3157f | 151 | kref_init(&xprt->xpt_ref); |
bb5cf160 | 152 | xprt->xpt_server = serv; |
7a182083 TT |
153 | INIT_LIST_HEAD(&xprt->xpt_list); |
154 | INIT_LIST_HEAD(&xprt->xpt_ready); | |
8c7b0172 | 155 | INIT_LIST_HEAD(&xprt->xpt_deferred); |
edc7a894 | 156 | INIT_LIST_HEAD(&xprt->xpt_users); |
a50fea26 | 157 | mutex_init(&xprt->xpt_mutex); |
def13d74 | 158 | spin_lock_init(&xprt->xpt_lock); |
4e5caaa5 | 159 | set_bit(XPT_BUSY, &xprt->xpt_flags); |
4cfc7e60 | 160 | rpc_init_wait_queue(&xprt->xpt_bc_pending, "xpt_bc_pending"); |
4fb8518b | 161 | xprt->xpt_net = get_net(&init_net); |
1d8206b9 TT |
162 | } |
163 | EXPORT_SYMBOL_GPL(svc_xprt_init); | |
b700cbb1 | 164 | |
5dd248f6 CL |
165 | static struct svc_xprt *__svc_xpo_create(struct svc_xprt_class *xcl, |
166 | struct svc_serv *serv, | |
62832c03 | 167 | struct net *net, |
9652ada3 CL |
168 | const int family, |
169 | const unsigned short port, | |
170 | int flags) | |
b700cbb1 | 171 | { |
b700cbb1 TT |
172 | struct sockaddr_in sin = { |
173 | .sin_family = AF_INET, | |
e6f1cebf | 174 | .sin_addr.s_addr = htonl(INADDR_ANY), |
b700cbb1 TT |
175 | .sin_port = htons(port), |
176 | }; | |
d6783b2b | 177 | #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) |
5dd248f6 CL |
178 | struct sockaddr_in6 sin6 = { |
179 | .sin6_family = AF_INET6, | |
180 | .sin6_addr = IN6ADDR_ANY_INIT, | |
181 | .sin6_port = htons(port), | |
182 | }; | |
d6783b2b | 183 | #endif /* defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) */ |
5dd248f6 CL |
184 | struct sockaddr *sap; |
185 | size_t len; | |
186 | ||
9652ada3 CL |
187 | switch (family) { |
188 | case PF_INET: | |
5dd248f6 CL |
189 | sap = (struct sockaddr *)&sin; |
190 | len = sizeof(sin); | |
191 | break; | |
d6783b2b | 192 | #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) |
9652ada3 | 193 | case PF_INET6: |
5dd248f6 CL |
194 | sap = (struct sockaddr *)&sin6; |
195 | len = sizeof(sin6); | |
196 | break; | |
d6783b2b | 197 | #endif /* defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) */ |
5dd248f6 CL |
198 | default: |
199 | return ERR_PTR(-EAFNOSUPPORT); | |
200 | } | |
201 | ||
62832c03 | 202 | return xcl->xcl_ops->xpo_create(serv, net, sap, len, flags); |
5dd248f6 CL |
203 | } |
204 | ||
9652ada3 | 205 | int svc_create_xprt(struct svc_serv *serv, const char *xprt_name, |
fc5d00b0 PE |
206 | struct net *net, const int family, |
207 | const unsigned short port, int flags) | |
5dd248f6 CL |
208 | { |
209 | struct svc_xprt_class *xcl; | |
210 | ||
b700cbb1 TT |
211 | dprintk("svc: creating transport %s[%d]\n", xprt_name, port); |
212 | spin_lock(&svc_xprt_class_lock); | |
213 | list_for_each_entry(xcl, &svc_xprt_class_list, xcl_list) { | |
4e5caaa5 TT |
214 | struct svc_xprt *newxprt; |
215 | ||
216 | if (strcmp(xprt_name, xcl->xcl_name)) | |
217 | continue; | |
218 | ||
219 | if (!try_module_get(xcl->xcl_owner)) | |
220 | goto err; | |
221 | ||
222 | spin_unlock(&svc_xprt_class_lock); | |
62832c03 | 223 | newxprt = __svc_xpo_create(xcl, serv, net, family, port, flags); |
4e5caaa5 TT |
224 | if (IS_ERR(newxprt)) { |
225 | module_put(xcl->xcl_owner); | |
226 | return PTR_ERR(newxprt); | |
b700cbb1 | 227 | } |
4e5caaa5 TT |
228 | |
229 | clear_bit(XPT_TEMP, &newxprt->xpt_flags); | |
230 | spin_lock_bh(&serv->sv_lock); | |
231 | list_add(&newxprt->xpt_list, &serv->sv_permsocks); | |
232 | spin_unlock_bh(&serv->sv_lock); | |
233 | clear_bit(XPT_BUSY, &newxprt->xpt_flags); | |
234 | return svc_xprt_local_port(newxprt); | |
b700cbb1 | 235 | } |
4e5caaa5 | 236 | err: |
b700cbb1 TT |
237 | spin_unlock(&svc_xprt_class_lock); |
238 | dprintk("svc: transport %s not found\n", xprt_name); | |
68717908 CL |
239 | |
240 | /* This errno is exposed to user space. Provide a reasonable | |
241 | * perror msg for a bad transport. */ | |
242 | return -EPROTONOSUPPORT; | |
b700cbb1 TT |
243 | } |
244 | EXPORT_SYMBOL_GPL(svc_create_xprt); | |
9dbc240f TT |
245 | |
246 | /* | |
247 | * Copy the local and remote xprt addresses to the rqstp structure | |
248 | */ | |
249 | void svc_xprt_copy_addrs(struct svc_rqst *rqstp, struct svc_xprt *xprt) | |
250 | { | |
251 | struct sockaddr *sin; | |
252 | ||
253 | memcpy(&rqstp->rq_addr, &xprt->xpt_remote, xprt->xpt_remotelen); | |
254 | rqstp->rq_addrlen = xprt->xpt_remotelen; | |
255 | ||
256 | /* | |
257 | * Destination address in request is needed for binding the | |
258 | * source address in RPC replies/callbacks later. | |
259 | */ | |
260 | sin = (struct sockaddr *)&xprt->xpt_local; | |
261 | switch (sin->sa_family) { | |
262 | case AF_INET: | |
263 | rqstp->rq_daddr.addr = ((struct sockaddr_in *)sin)->sin_addr; | |
264 | break; | |
265 | case AF_INET6: | |
266 | rqstp->rq_daddr.addr6 = ((struct sockaddr_in6 *)sin)->sin6_addr; | |
267 | break; | |
268 | } | |
269 | } | |
270 | EXPORT_SYMBOL_GPL(svc_xprt_copy_addrs); | |
271 | ||
0f0257ea TT |
272 | /** |
273 | * svc_print_addr - Format rq_addr field for printing | |
274 | * @rqstp: svc_rqst struct containing address to print | |
275 | * @buf: target buffer for formatted address | |
276 | * @len: length of target buffer | |
277 | * | |
278 | */ | |
279 | char *svc_print_addr(struct svc_rqst *rqstp, char *buf, size_t len) | |
280 | { | |
281 | return __svc_print_addr(svc_addr(rqstp), buf, len); | |
282 | } | |
283 | EXPORT_SYMBOL_GPL(svc_print_addr); | |
284 | ||
285 | /* | |
286 | * Queue up an idle server thread. Must have pool->sp_lock held. | |
287 | * Note: this is really a stack rather than a queue, so that we only | |
288 | * use as many different threads as we need, and the rest don't pollute | |
289 | * the cache. | |
290 | */ | |
291 | static void svc_thread_enqueue(struct svc_pool *pool, struct svc_rqst *rqstp) | |
292 | { | |
293 | list_add(&rqstp->rq_list, &pool->sp_threads); | |
294 | } | |
295 | ||
296 | /* | |
297 | * Dequeue an nfsd thread. Must have pool->sp_lock held. | |
298 | */ | |
299 | static void svc_thread_dequeue(struct svc_pool *pool, struct svc_rqst *rqstp) | |
300 | { | |
301 | list_del(&rqstp->rq_list); | |
302 | } | |
303 | ||
304 | /* | |
305 | * Queue up a transport with data pending. If there are idle nfsd | |
306 | * processes, wake 'em up. | |
307 | * | |
308 | */ | |
309 | void svc_xprt_enqueue(struct svc_xprt *xprt) | |
310 | { | |
311 | struct svc_serv *serv = xprt->xpt_server; | |
312 | struct svc_pool *pool; | |
313 | struct svc_rqst *rqstp; | |
314 | int cpu; | |
315 | ||
316 | if (!(xprt->xpt_flags & | |
317 | ((1<<XPT_CONN)|(1<<XPT_DATA)|(1<<XPT_CLOSE)|(1<<XPT_DEFERRED)))) | |
318 | return; | |
0f0257ea TT |
319 | |
320 | cpu = get_cpu(); | |
321 | pool = svc_pool_for_cpu(xprt->xpt_server, cpu); | |
322 | put_cpu(); | |
323 | ||
324 | spin_lock_bh(&pool->sp_lock); | |
325 | ||
78c210ef BF |
326 | if (!list_empty(&pool->sp_threads) && |
327 | !list_empty(&pool->sp_sockets)) | |
328 | printk(KERN_ERR | |
329 | "svc_xprt_enqueue: " | |
330 | "threads and transports both waiting??\n"); | |
331 | ||
03cf6c9f GB |
332 | pool->sp_stats.packets++; |
333 | ||
0f0257ea TT |
334 | /* Mark transport as busy. It will remain in this state until |
335 | * the provider calls svc_xprt_received. We update XPT_BUSY | |
336 | * atomically because it also guards against trying to enqueue | |
337 | * the transport twice. | |
338 | */ | |
339 | if (test_and_set_bit(XPT_BUSY, &xprt->xpt_flags)) { | |
340 | /* Don't enqueue transport while already enqueued */ | |
341 | dprintk("svc: transport %p busy, not enqueued\n", xprt); | |
342 | goto out_unlock; | |
343 | } | |
344 | BUG_ON(xprt->xpt_pool != NULL); | |
345 | xprt->xpt_pool = pool; | |
346 | ||
347 | /* Handle pending connection */ | |
348 | if (test_bit(XPT_CONN, &xprt->xpt_flags)) | |
349 | goto process; | |
350 | ||
351 | /* Handle close in-progress */ | |
352 | if (test_bit(XPT_CLOSE, &xprt->xpt_flags)) | |
353 | goto process; | |
354 | ||
355 | /* Check if we have space to reply to a request */ | |
356 | if (!xprt->xpt_ops->xpo_has_wspace(xprt)) { | |
357 | /* Don't enqueue while not enough space for reply */ | |
358 | dprintk("svc: no write space, transport %p not enqueued\n", | |
359 | xprt); | |
360 | xprt->xpt_pool = NULL; | |
361 | clear_bit(XPT_BUSY, &xprt->xpt_flags); | |
362 | goto out_unlock; | |
363 | } | |
364 | ||
365 | process: | |
78c210ef | 366 | if (!list_empty(&pool->sp_threads)) { |
0f0257ea TT |
367 | rqstp = list_entry(pool->sp_threads.next, |
368 | struct svc_rqst, | |
369 | rq_list); | |
370 | dprintk("svc: transport %p served by daemon %p\n", | |
371 | xprt, rqstp); | |
372 | svc_thread_dequeue(pool, rqstp); | |
373 | if (rqstp->rq_xprt) | |
374 | printk(KERN_ERR | |
375 | "svc_xprt_enqueue: server %p, rq_xprt=%p!\n", | |
376 | rqstp, rqstp->rq_xprt); | |
377 | rqstp->rq_xprt = xprt; | |
378 | svc_xprt_get(xprt); | |
379 | rqstp->rq_reserved = serv->sv_max_mesg; | |
380 | atomic_add(rqstp->rq_reserved, &xprt->xpt_reserved); | |
03cf6c9f | 381 | pool->sp_stats.threads_woken++; |
0f0257ea TT |
382 | BUG_ON(xprt->xpt_pool != pool); |
383 | wake_up(&rqstp->rq_wait); | |
384 | } else { | |
385 | dprintk("svc: transport %p put into queue\n", xprt); | |
386 | list_add_tail(&xprt->xpt_ready, &pool->sp_sockets); | |
03cf6c9f | 387 | pool->sp_stats.sockets_queued++; |
0f0257ea TT |
388 | BUG_ON(xprt->xpt_pool != pool); |
389 | } | |
390 | ||
391 | out_unlock: | |
392 | spin_unlock_bh(&pool->sp_lock); | |
393 | } | |
394 | EXPORT_SYMBOL_GPL(svc_xprt_enqueue); | |
395 | ||
396 | /* | |
397 | * Dequeue the first transport. Must be called with the pool->sp_lock held. | |
398 | */ | |
399 | static struct svc_xprt *svc_xprt_dequeue(struct svc_pool *pool) | |
400 | { | |
401 | struct svc_xprt *xprt; | |
402 | ||
403 | if (list_empty(&pool->sp_sockets)) | |
404 | return NULL; | |
405 | ||
406 | xprt = list_entry(pool->sp_sockets.next, | |
407 | struct svc_xprt, xpt_ready); | |
408 | list_del_init(&xprt->xpt_ready); | |
409 | ||
410 | dprintk("svc: transport %p dequeued, inuse=%d\n", | |
411 | xprt, atomic_read(&xprt->xpt_ref.refcount)); | |
412 | ||
413 | return xprt; | |
414 | } | |
415 | ||
416 | /* | |
417 | * svc_xprt_received conditionally queues the transport for processing | |
418 | * by another thread. The caller must hold the XPT_BUSY bit and must | |
419 | * not thereafter touch transport data. | |
420 | * | |
421 | * Note: XPT_DATA only gets cleared when a read-attempt finds no (or | |
422 | * insufficient) data. | |
423 | */ | |
424 | void svc_xprt_received(struct svc_xprt *xprt) | |
425 | { | |
426 | BUG_ON(!test_bit(XPT_BUSY, &xprt->xpt_flags)); | |
427 | xprt->xpt_pool = NULL; | |
428 | clear_bit(XPT_BUSY, &xprt->xpt_flags); | |
429 | svc_xprt_enqueue(xprt); | |
430 | } | |
431 | EXPORT_SYMBOL_GPL(svc_xprt_received); | |
432 | ||
433 | /** | |
434 | * svc_reserve - change the space reserved for the reply to a request. | |
435 | * @rqstp: The request in question | |
436 | * @space: new max space to reserve | |
437 | * | |
438 | * Each request reserves some space on the output queue of the transport | |
439 | * to make sure the reply fits. This function reduces that reserved | |
440 | * space to be the amount of space used already, plus @space. | |
441 | * | |
442 | */ | |
443 | void svc_reserve(struct svc_rqst *rqstp, int space) | |
444 | { | |
445 | space += rqstp->rq_res.head[0].iov_len; | |
446 | ||
447 | if (space < rqstp->rq_reserved) { | |
448 | struct svc_xprt *xprt = rqstp->rq_xprt; | |
449 | atomic_sub((rqstp->rq_reserved - space), &xprt->xpt_reserved); | |
450 | rqstp->rq_reserved = space; | |
451 | ||
452 | svc_xprt_enqueue(xprt); | |
453 | } | |
454 | } | |
24c3767e | 455 | EXPORT_SYMBOL_GPL(svc_reserve); |
0f0257ea TT |
456 | |
457 | static void svc_xprt_release(struct svc_rqst *rqstp) | |
458 | { | |
459 | struct svc_xprt *xprt = rqstp->rq_xprt; | |
460 | ||
461 | rqstp->rq_xprt->xpt_ops->xpo_release_rqst(rqstp); | |
462 | ||
2779e3ae TT |
463 | kfree(rqstp->rq_deferred); |
464 | rqstp->rq_deferred = NULL; | |
465 | ||
0f0257ea TT |
466 | svc_free_res_pages(rqstp); |
467 | rqstp->rq_res.page_len = 0; | |
468 | rqstp->rq_res.page_base = 0; | |
469 | ||
470 | /* Reset response buffer and release | |
471 | * the reservation. | |
472 | * But first, check that enough space was reserved | |
473 | * for the reply, otherwise we have a bug! | |
474 | */ | |
475 | if ((rqstp->rq_res.len) > rqstp->rq_reserved) | |
476 | printk(KERN_ERR "RPC request reserved %d but used %d\n", | |
477 | rqstp->rq_reserved, | |
478 | rqstp->rq_res.len); | |
479 | ||
480 | rqstp->rq_res.head[0].iov_len = 0; | |
481 | svc_reserve(rqstp, 0); | |
482 | rqstp->rq_xprt = NULL; | |
483 | ||
484 | svc_xprt_put(xprt); | |
485 | } | |
486 | ||
487 | /* | |
488 | * External function to wake up a server waiting for data | |
489 | * This really only makes sense for services like lockd | |
490 | * which have exactly one thread anyway. | |
491 | */ | |
492 | void svc_wake_up(struct svc_serv *serv) | |
493 | { | |
494 | struct svc_rqst *rqstp; | |
495 | unsigned int i; | |
496 | struct svc_pool *pool; | |
497 | ||
498 | for (i = 0; i < serv->sv_nrpools; i++) { | |
499 | pool = &serv->sv_pools[i]; | |
500 | ||
501 | spin_lock_bh(&pool->sp_lock); | |
502 | if (!list_empty(&pool->sp_threads)) { | |
503 | rqstp = list_entry(pool->sp_threads.next, | |
504 | struct svc_rqst, | |
505 | rq_list); | |
506 | dprintk("svc: daemon %p woken up.\n", rqstp); | |
507 | /* | |
508 | svc_thread_dequeue(pool, rqstp); | |
509 | rqstp->rq_xprt = NULL; | |
510 | */ | |
511 | wake_up(&rqstp->rq_wait); | |
512 | } | |
513 | spin_unlock_bh(&pool->sp_lock); | |
514 | } | |
515 | } | |
24c3767e | 516 | EXPORT_SYMBOL_GPL(svc_wake_up); |
0f0257ea TT |
517 | |
518 | int svc_port_is_privileged(struct sockaddr *sin) | |
519 | { | |
520 | switch (sin->sa_family) { | |
521 | case AF_INET: | |
522 | return ntohs(((struct sockaddr_in *)sin)->sin_port) | |
523 | < PROT_SOCK; | |
524 | case AF_INET6: | |
525 | return ntohs(((struct sockaddr_in6 *)sin)->sin6_port) | |
526 | < PROT_SOCK; | |
527 | default: | |
528 | return 0; | |
529 | } | |
530 | } | |
531 | ||
532 | /* | |
c9233eb7 JL |
533 | * Make sure that we don't have too many active connections. If we have, |
534 | * something must be dropped. It's not clear what will happen if we allow | |
535 | * "too many" connections, but when dealing with network-facing software, | |
536 | * we have to code defensively. Here we do that by imposing hard limits. | |
0f0257ea TT |
537 | * |
538 | * There's no point in trying to do random drop here for DoS | |
539 | * prevention. The NFS clients does 1 reconnect in 15 seconds. An | |
540 | * attacker can easily beat that. | |
541 | * | |
542 | * The only somewhat efficient mechanism would be if drop old | |
543 | * connections from the same IP first. But right now we don't even | |
544 | * record the client IP in svc_sock. | |
c9233eb7 JL |
545 | * |
546 | * single-threaded services that expect a lot of clients will probably | |
547 | * need to set sv_maxconn to override the default value which is based | |
548 | * on the number of threads | |
0f0257ea TT |
549 | */ |
550 | static void svc_check_conn_limits(struct svc_serv *serv) | |
551 | { | |
c9233eb7 JL |
552 | unsigned int limit = serv->sv_maxconn ? serv->sv_maxconn : |
553 | (serv->sv_nrthreads+3) * 20; | |
554 | ||
555 | if (serv->sv_tmpcnt > limit) { | |
0f0257ea TT |
556 | struct svc_xprt *xprt = NULL; |
557 | spin_lock_bh(&serv->sv_lock); | |
558 | if (!list_empty(&serv->sv_tempsocks)) { | |
559 | if (net_ratelimit()) { | |
560 | /* Try to help the admin */ | |
561 | printk(KERN_NOTICE "%s: too many open " | |
c9233eb7 JL |
562 | "connections, consider increasing %s\n", |
563 | serv->sv_name, serv->sv_maxconn ? | |
564 | "the max number of connections." : | |
565 | "the number of threads."); | |
0f0257ea TT |
566 | } |
567 | /* | |
568 | * Always select the oldest connection. It's not fair, | |
569 | * but so is life | |
570 | */ | |
571 | xprt = list_entry(serv->sv_tempsocks.prev, | |
572 | struct svc_xprt, | |
573 | xpt_list); | |
574 | set_bit(XPT_CLOSE, &xprt->xpt_flags); | |
575 | svc_xprt_get(xprt); | |
576 | } | |
577 | spin_unlock_bh(&serv->sv_lock); | |
578 | ||
579 | if (xprt) { | |
580 | svc_xprt_enqueue(xprt); | |
581 | svc_xprt_put(xprt); | |
582 | } | |
583 | } | |
584 | } | |
585 | ||
586 | /* | |
587 | * Receive the next request on any transport. This code is carefully | |
588 | * organised not to touch any cachelines in the shared svc_serv | |
589 | * structure, only cachelines in the local svc_pool. | |
590 | */ | |
591 | int svc_recv(struct svc_rqst *rqstp, long timeout) | |
592 | { | |
593 | struct svc_xprt *xprt = NULL; | |
594 | struct svc_serv *serv = rqstp->rq_server; | |
595 | struct svc_pool *pool = rqstp->rq_pool; | |
596 | int len, i; | |
597 | int pages; | |
598 | struct xdr_buf *arg; | |
599 | DECLARE_WAITQUEUE(wait, current); | |
03cf6c9f | 600 | long time_left; |
0f0257ea TT |
601 | |
602 | dprintk("svc: server %p waiting for data (to = %ld)\n", | |
603 | rqstp, timeout); | |
604 | ||
605 | if (rqstp->rq_xprt) | |
606 | printk(KERN_ERR | |
607 | "svc_recv: service %p, transport not NULL!\n", | |
608 | rqstp); | |
609 | if (waitqueue_active(&rqstp->rq_wait)) | |
610 | printk(KERN_ERR | |
611 | "svc_recv: service %p, wait queue active!\n", | |
612 | rqstp); | |
613 | ||
614 | /* now allocate needed pages. If we get a failure, sleep briefly */ | |
615 | pages = (serv->sv_max_mesg + PAGE_SIZE) / PAGE_SIZE; | |
616 | for (i = 0; i < pages ; i++) | |
617 | while (rqstp->rq_pages[i] == NULL) { | |
618 | struct page *p = alloc_page(GFP_KERNEL); | |
619 | if (!p) { | |
7b54fe61 JL |
620 | set_current_state(TASK_INTERRUPTIBLE); |
621 | if (signalled() || kthread_should_stop()) { | |
622 | set_current_state(TASK_RUNNING); | |
7086721f | 623 | return -EINTR; |
7b54fe61 JL |
624 | } |
625 | schedule_timeout(msecs_to_jiffies(500)); | |
0f0257ea TT |
626 | } |
627 | rqstp->rq_pages[i] = p; | |
628 | } | |
629 | rqstp->rq_pages[i++] = NULL; /* this might be seen in nfs_read_actor */ | |
630 | BUG_ON(pages >= RPCSVC_MAXPAGES); | |
631 | ||
632 | /* Make arg->head point to first page and arg->pages point to rest */ | |
633 | arg = &rqstp->rq_arg; | |
634 | arg->head[0].iov_base = page_address(rqstp->rq_pages[0]); | |
635 | arg->head[0].iov_len = PAGE_SIZE; | |
636 | arg->pages = rqstp->rq_pages + 1; | |
637 | arg->page_base = 0; | |
638 | /* save at least one page for response */ | |
639 | arg->page_len = (pages-2)*PAGE_SIZE; | |
640 | arg->len = (pages-1)*PAGE_SIZE; | |
641 | arg->tail[0].iov_len = 0; | |
642 | ||
643 | try_to_freeze(); | |
644 | cond_resched(); | |
7086721f | 645 | if (signalled() || kthread_should_stop()) |
0f0257ea TT |
646 | return -EINTR; |
647 | ||
f16b6e8d N |
648 | /* Normally we will wait up to 5 seconds for any required |
649 | * cache information to be provided. | |
650 | */ | |
651 | rqstp->rq_chandle.thread_wait = 5*HZ; | |
652 | ||
0f0257ea TT |
653 | spin_lock_bh(&pool->sp_lock); |
654 | xprt = svc_xprt_dequeue(pool); | |
655 | if (xprt) { | |
656 | rqstp->rq_xprt = xprt; | |
657 | svc_xprt_get(xprt); | |
658 | rqstp->rq_reserved = serv->sv_max_mesg; | |
659 | atomic_add(rqstp->rq_reserved, &xprt->xpt_reserved); | |
f16b6e8d N |
660 | |
661 | /* As there is a shortage of threads and this request | |
6610f720 | 662 | * had to be queued, don't allow the thread to wait so |
f16b6e8d N |
663 | * long for cache updates. |
664 | */ | |
665 | rqstp->rq_chandle.thread_wait = 1*HZ; | |
0f0257ea TT |
666 | } else { |
667 | /* No data pending. Go to sleep */ | |
668 | svc_thread_enqueue(pool, rqstp); | |
669 | ||
670 | /* | |
671 | * We have to be able to interrupt this wait | |
672 | * to bring down the daemons ... | |
673 | */ | |
674 | set_current_state(TASK_INTERRUPTIBLE); | |
7086721f JL |
675 | |
676 | /* | |
677 | * checking kthread_should_stop() here allows us to avoid | |
678 | * locking and signalling when stopping kthreads that call | |
679 | * svc_recv. If the thread has already been woken up, then | |
680 | * we can exit here without sleeping. If not, then it | |
681 | * it'll be woken up quickly during the schedule_timeout | |
682 | */ | |
683 | if (kthread_should_stop()) { | |
684 | set_current_state(TASK_RUNNING); | |
685 | spin_unlock_bh(&pool->sp_lock); | |
686 | return -EINTR; | |
687 | } | |
688 | ||
0f0257ea TT |
689 | add_wait_queue(&rqstp->rq_wait, &wait); |
690 | spin_unlock_bh(&pool->sp_lock); | |
691 | ||
03cf6c9f | 692 | time_left = schedule_timeout(timeout); |
0f0257ea TT |
693 | |
694 | try_to_freeze(); | |
695 | ||
696 | spin_lock_bh(&pool->sp_lock); | |
697 | remove_wait_queue(&rqstp->rq_wait, &wait); | |
03cf6c9f GB |
698 | if (!time_left) |
699 | pool->sp_stats.threads_timedout++; | |
0f0257ea TT |
700 | |
701 | xprt = rqstp->rq_xprt; | |
702 | if (!xprt) { | |
703 | svc_thread_dequeue(pool, rqstp); | |
704 | spin_unlock_bh(&pool->sp_lock); | |
705 | dprintk("svc: server %p, no data yet\n", rqstp); | |
7086721f JL |
706 | if (signalled() || kthread_should_stop()) |
707 | return -EINTR; | |
708 | else | |
709 | return -EAGAIN; | |
0f0257ea TT |
710 | } |
711 | } | |
712 | spin_unlock_bh(&pool->sp_lock); | |
713 | ||
714 | len = 0; | |
1b644b6e BF |
715 | if (test_bit(XPT_CLOSE, &xprt->xpt_flags)) { |
716 | dprintk("svc_recv: found XPT_CLOSE\n"); | |
717 | svc_delete_xprt(xprt); | |
718 | } else if (test_bit(XPT_LISTENER, &xprt->xpt_flags)) { | |
0f0257ea TT |
719 | struct svc_xprt *newxpt; |
720 | newxpt = xprt->xpt_ops->xpo_accept(xprt); | |
721 | if (newxpt) { | |
722 | /* | |
723 | * We know this module_get will succeed because the | |
724 | * listener holds a reference too | |
725 | */ | |
726 | __module_get(newxpt->xpt_class->xcl_owner); | |
727 | svc_check_conn_limits(xprt->xpt_server); | |
728 | spin_lock_bh(&serv->sv_lock); | |
729 | set_bit(XPT_TEMP, &newxpt->xpt_flags); | |
730 | list_add(&newxpt->xpt_list, &serv->sv_tempsocks); | |
731 | serv->sv_tmpcnt++; | |
732 | if (serv->sv_temptimer.function == NULL) { | |
733 | /* setup timer to age temp transports */ | |
734 | setup_timer(&serv->sv_temptimer, | |
735 | svc_age_temp_xprts, | |
736 | (unsigned long)serv); | |
737 | mod_timer(&serv->sv_temptimer, | |
738 | jiffies + svc_conn_age_period * HZ); | |
739 | } | |
740 | spin_unlock_bh(&serv->sv_lock); | |
741 | svc_xprt_received(newxpt); | |
742 | } | |
743 | svc_xprt_received(xprt); | |
1b644b6e | 744 | } else { |
0f0257ea TT |
745 | dprintk("svc: server %p, pool %u, transport %p, inuse=%d\n", |
746 | rqstp, pool->sp_id, xprt, | |
747 | atomic_read(&xprt->xpt_ref.refcount)); | |
748 | rqstp->rq_deferred = svc_deferred_dequeue(xprt); | |
749 | if (rqstp->rq_deferred) { | |
750 | svc_xprt_received(xprt); | |
751 | len = svc_deferred_recv(rqstp); | |
b48fa6b9 | 752 | } else { |
0f0257ea | 753 | len = xprt->xpt_ops->xpo_recvfrom(rqstp); |
b48fa6b9 NB |
754 | svc_xprt_received(xprt); |
755 | } | |
0f0257ea TT |
756 | dprintk("svc: got len=%d\n", len); |
757 | } | |
758 | ||
759 | /* No data, incomplete (TCP) read, or accept() */ | |
760 | if (len == 0 || len == -EAGAIN) { | |
761 | rqstp->rq_res.len = 0; | |
762 | svc_xprt_release(rqstp); | |
763 | return -EAGAIN; | |
764 | } | |
765 | clear_bit(XPT_OLD, &xprt->xpt_flags); | |
766 | ||
767 | rqstp->rq_secure = svc_port_is_privileged(svc_addr(rqstp)); | |
768 | rqstp->rq_chandle.defer = svc_defer; | |
769 | ||
770 | if (serv->sv_stats) | |
771 | serv->sv_stats->netcnt++; | |
772 | return len; | |
773 | } | |
24c3767e | 774 | EXPORT_SYMBOL_GPL(svc_recv); |
0f0257ea TT |
775 | |
776 | /* | |
777 | * Drop request | |
778 | */ | |
779 | void svc_drop(struct svc_rqst *rqstp) | |
780 | { | |
781 | dprintk("svc: xprt %p dropped request\n", rqstp->rq_xprt); | |
782 | svc_xprt_release(rqstp); | |
783 | } | |
24c3767e | 784 | EXPORT_SYMBOL_GPL(svc_drop); |
0f0257ea TT |
785 | |
786 | /* | |
787 | * Return reply to client. | |
788 | */ | |
789 | int svc_send(struct svc_rqst *rqstp) | |
790 | { | |
791 | struct svc_xprt *xprt; | |
792 | int len; | |
793 | struct xdr_buf *xb; | |
794 | ||
795 | xprt = rqstp->rq_xprt; | |
796 | if (!xprt) | |
797 | return -EFAULT; | |
798 | ||
799 | /* release the receive skb before sending the reply */ | |
800 | rqstp->rq_xprt->xpt_ops->xpo_release_rqst(rqstp); | |
801 | ||
802 | /* calculate over-all length */ | |
803 | xb = &rqstp->rq_res; | |
804 | xb->len = xb->head[0].iov_len + | |
805 | xb->page_len + | |
806 | xb->tail[0].iov_len; | |
807 | ||
808 | /* Grab mutex to serialize outgoing data. */ | |
809 | mutex_lock(&xprt->xpt_mutex); | |
810 | if (test_bit(XPT_DEAD, &xprt->xpt_flags)) | |
811 | len = -ENOTCONN; | |
812 | else | |
813 | len = xprt->xpt_ops->xpo_sendto(rqstp); | |
814 | mutex_unlock(&xprt->xpt_mutex); | |
4cfc7e60 | 815 | rpc_wake_up(&xprt->xpt_bc_pending); |
0f0257ea TT |
816 | svc_xprt_release(rqstp); |
817 | ||
818 | if (len == -ECONNREFUSED || len == -ENOTCONN || len == -EAGAIN) | |
819 | return 0; | |
820 | return len; | |
821 | } | |
822 | ||
823 | /* | |
824 | * Timer function to close old temporary transports, using | |
825 | * a mark-and-sweep algorithm. | |
826 | */ | |
827 | static void svc_age_temp_xprts(unsigned long closure) | |
828 | { | |
829 | struct svc_serv *serv = (struct svc_serv *)closure; | |
830 | struct svc_xprt *xprt; | |
831 | struct list_head *le, *next; | |
832 | LIST_HEAD(to_be_aged); | |
833 | ||
834 | dprintk("svc_age_temp_xprts\n"); | |
835 | ||
836 | if (!spin_trylock_bh(&serv->sv_lock)) { | |
837 | /* busy, try again 1 sec later */ | |
838 | dprintk("svc_age_temp_xprts: busy\n"); | |
839 | mod_timer(&serv->sv_temptimer, jiffies + HZ); | |
840 | return; | |
841 | } | |
842 | ||
843 | list_for_each_safe(le, next, &serv->sv_tempsocks) { | |
844 | xprt = list_entry(le, struct svc_xprt, xpt_list); | |
845 | ||
846 | /* First time through, just mark it OLD. Second time | |
847 | * through, close it. */ | |
848 | if (!test_and_set_bit(XPT_OLD, &xprt->xpt_flags)) | |
849 | continue; | |
f64f9e71 JP |
850 | if (atomic_read(&xprt->xpt_ref.refcount) > 1 || |
851 | test_bit(XPT_BUSY, &xprt->xpt_flags)) | |
0f0257ea TT |
852 | continue; |
853 | svc_xprt_get(xprt); | |
854 | list_move(le, &to_be_aged); | |
855 | set_bit(XPT_CLOSE, &xprt->xpt_flags); | |
856 | set_bit(XPT_DETACHED, &xprt->xpt_flags); | |
857 | } | |
858 | spin_unlock_bh(&serv->sv_lock); | |
859 | ||
860 | while (!list_empty(&to_be_aged)) { | |
861 | le = to_be_aged.next; | |
862 | /* fiddling the xpt_list node is safe 'cos we're XPT_DETACHED */ | |
863 | list_del_init(le); | |
864 | xprt = list_entry(le, struct svc_xprt, xpt_list); | |
865 | ||
866 | dprintk("queuing xprt %p for closing\n", xprt); | |
867 | ||
868 | /* a thread will dequeue and close it soon */ | |
869 | svc_xprt_enqueue(xprt); | |
870 | svc_xprt_put(xprt); | |
871 | } | |
872 | ||
873 | mod_timer(&serv->sv_temptimer, jiffies + svc_conn_age_period * HZ); | |
874 | } | |
875 | ||
edc7a894 BF |
876 | static void call_xpt_users(struct svc_xprt *xprt) |
877 | { | |
878 | struct svc_xpt_user *u; | |
879 | ||
880 | spin_lock(&xprt->xpt_lock); | |
881 | while (!list_empty(&xprt->xpt_users)) { | |
882 | u = list_first_entry(&xprt->xpt_users, struct svc_xpt_user, list); | |
883 | list_del(&u->list); | |
884 | u->callback(u); | |
885 | } | |
886 | spin_unlock(&xprt->xpt_lock); | |
887 | } | |
888 | ||
0f0257ea TT |
889 | /* |
890 | * Remove a dead transport | |
891 | */ | |
892 | void svc_delete_xprt(struct svc_xprt *xprt) | |
893 | { | |
894 | struct svc_serv *serv = xprt->xpt_server; | |
22945e4a TT |
895 | struct svc_deferred_req *dr; |
896 | ||
897 | /* Only do this once */ | |
898 | if (test_and_set_bit(XPT_DEAD, &xprt->xpt_flags)) | |
ac9303eb | 899 | BUG(); |
0f0257ea TT |
900 | |
901 | dprintk("svc: svc_delete_xprt(%p)\n", xprt); | |
902 | xprt->xpt_ops->xpo_detach(xprt); | |
903 | ||
904 | spin_lock_bh(&serv->sv_lock); | |
905 | if (!test_and_set_bit(XPT_DETACHED, &xprt->xpt_flags)) | |
906 | list_del_init(&xprt->xpt_list); | |
907 | /* | |
908 | * We used to delete the transport from whichever list | |
909 | * it's sk_xprt.xpt_ready node was on, but we don't actually | |
910 | * need to. This is because the only time we're called | |
911 | * while still attached to a queue, the queue itself | |
912 | * is about to be destroyed (in svc_destroy). | |
913 | */ | |
22945e4a TT |
914 | if (test_bit(XPT_TEMP, &xprt->xpt_flags)) |
915 | serv->sv_tmpcnt--; | |
788e69e5 | 916 | spin_unlock_bh(&serv->sv_lock); |
22945e4a | 917 | |
ab1b18f7 | 918 | while ((dr = svc_deferred_dequeue(xprt)) != NULL) |
22945e4a | 919 | kfree(dr); |
22945e4a | 920 | |
edc7a894 | 921 | call_xpt_users(xprt); |
22945e4a | 922 | svc_xprt_put(xprt); |
0f0257ea TT |
923 | } |
924 | ||
925 | void svc_close_xprt(struct svc_xprt *xprt) | |
926 | { | |
927 | set_bit(XPT_CLOSE, &xprt->xpt_flags); | |
928 | if (test_and_set_bit(XPT_BUSY, &xprt->xpt_flags)) | |
929 | /* someone else will have to effect the close */ | |
930 | return; | |
931 | ||
0f0257ea | 932 | svc_delete_xprt(xprt); |
0f0257ea | 933 | } |
a217813f | 934 | EXPORT_SYMBOL_GPL(svc_close_xprt); |
0f0257ea TT |
935 | |
936 | void svc_close_all(struct list_head *xprt_list) | |
937 | { | |
938 | struct svc_xprt *xprt; | |
939 | struct svc_xprt *tmp; | |
940 | ||
941 | list_for_each_entry_safe(xprt, tmp, xprt_list, xpt_list) { | |
942 | set_bit(XPT_CLOSE, &xprt->xpt_flags); | |
943 | if (test_bit(XPT_BUSY, &xprt->xpt_flags)) { | |
944 | /* Waiting to be processed, but no threads left, | |
945 | * So just remove it from the waiting list | |
946 | */ | |
947 | list_del_init(&xprt->xpt_ready); | |
948 | clear_bit(XPT_BUSY, &xprt->xpt_flags); | |
949 | } | |
950 | svc_close_xprt(xprt); | |
951 | } | |
952 | } | |
953 | ||
954 | /* | |
955 | * Handle defer and revisit of requests | |
956 | */ | |
957 | ||
958 | static void svc_revisit(struct cache_deferred_req *dreq, int too_many) | |
959 | { | |
960 | struct svc_deferred_req *dr = | |
961 | container_of(dreq, struct svc_deferred_req, handle); | |
962 | struct svc_xprt *xprt = dr->xprt; | |
963 | ||
22945e4a TT |
964 | spin_lock(&xprt->xpt_lock); |
965 | set_bit(XPT_DEFERRED, &xprt->xpt_flags); | |
966 | if (too_many || test_bit(XPT_DEAD, &xprt->xpt_flags)) { | |
967 | spin_unlock(&xprt->xpt_lock); | |
968 | dprintk("revisit canceled\n"); | |
0f0257ea TT |
969 | svc_xprt_put(xprt); |
970 | kfree(dr); | |
971 | return; | |
972 | } | |
973 | dprintk("revisit queued\n"); | |
974 | dr->xprt = NULL; | |
0f0257ea TT |
975 | list_add(&dr->handle.recent, &xprt->xpt_deferred); |
976 | spin_unlock(&xprt->xpt_lock); | |
0f0257ea TT |
977 | svc_xprt_enqueue(xprt); |
978 | svc_xprt_put(xprt); | |
979 | } | |
980 | ||
260c1d12 TT |
981 | /* |
982 | * Save the request off for later processing. The request buffer looks | |
983 | * like this: | |
984 | * | |
985 | * <xprt-header><rpc-header><rpc-pagelist><rpc-tail> | |
986 | * | |
987 | * This code can only handle requests that consist of an xprt-header | |
988 | * and rpc-header. | |
989 | */ | |
0f0257ea TT |
990 | static struct cache_deferred_req *svc_defer(struct cache_req *req) |
991 | { | |
992 | struct svc_rqst *rqstp = container_of(req, struct svc_rqst, rq_chandle); | |
0f0257ea TT |
993 | struct svc_deferred_req *dr; |
994 | ||
2f425878 | 995 | if (rqstp->rq_arg.page_len || !rqstp->rq_usedeferral) |
0f0257ea TT |
996 | return NULL; /* if more than a page, give up FIXME */ |
997 | if (rqstp->rq_deferred) { | |
998 | dr = rqstp->rq_deferred; | |
999 | rqstp->rq_deferred = NULL; | |
1000 | } else { | |
260c1d12 TT |
1001 | size_t skip; |
1002 | size_t size; | |
0f0257ea | 1003 | /* FIXME maybe discard if size too large */ |
260c1d12 | 1004 | size = sizeof(struct svc_deferred_req) + rqstp->rq_arg.len; |
0f0257ea TT |
1005 | dr = kmalloc(size, GFP_KERNEL); |
1006 | if (dr == NULL) | |
1007 | return NULL; | |
1008 | ||
1009 | dr->handle.owner = rqstp->rq_server; | |
1010 | dr->prot = rqstp->rq_prot; | |
1011 | memcpy(&dr->addr, &rqstp->rq_addr, rqstp->rq_addrlen); | |
1012 | dr->addrlen = rqstp->rq_addrlen; | |
1013 | dr->daddr = rqstp->rq_daddr; | |
1014 | dr->argslen = rqstp->rq_arg.len >> 2; | |
260c1d12 TT |
1015 | dr->xprt_hlen = rqstp->rq_xprt_hlen; |
1016 | ||
1017 | /* back up head to the start of the buffer and copy */ | |
1018 | skip = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len; | |
1019 | memcpy(dr->args, rqstp->rq_arg.head[0].iov_base - skip, | |
1020 | dr->argslen << 2); | |
0f0257ea TT |
1021 | } |
1022 | svc_xprt_get(rqstp->rq_xprt); | |
1023 | dr->xprt = rqstp->rq_xprt; | |
1024 | ||
1025 | dr->handle.revisit = svc_revisit; | |
1026 | return &dr->handle; | |
1027 | } | |
1028 | ||
1029 | /* | |
1030 | * recv data from a deferred request into an active one | |
1031 | */ | |
1032 | static int svc_deferred_recv(struct svc_rqst *rqstp) | |
1033 | { | |
1034 | struct svc_deferred_req *dr = rqstp->rq_deferred; | |
1035 | ||
260c1d12 TT |
1036 | /* setup iov_base past transport header */ |
1037 | rqstp->rq_arg.head[0].iov_base = dr->args + (dr->xprt_hlen>>2); | |
1038 | /* The iov_len does not include the transport header bytes */ | |
1039 | rqstp->rq_arg.head[0].iov_len = (dr->argslen<<2) - dr->xprt_hlen; | |
0f0257ea | 1040 | rqstp->rq_arg.page_len = 0; |
260c1d12 TT |
1041 | /* The rq_arg.len includes the transport header bytes */ |
1042 | rqstp->rq_arg.len = dr->argslen<<2; | |
0f0257ea TT |
1043 | rqstp->rq_prot = dr->prot; |
1044 | memcpy(&rqstp->rq_addr, &dr->addr, dr->addrlen); | |
1045 | rqstp->rq_addrlen = dr->addrlen; | |
260c1d12 TT |
1046 | /* Save off transport header len in case we get deferred again */ |
1047 | rqstp->rq_xprt_hlen = dr->xprt_hlen; | |
0f0257ea TT |
1048 | rqstp->rq_daddr = dr->daddr; |
1049 | rqstp->rq_respages = rqstp->rq_pages; | |
260c1d12 | 1050 | return (dr->argslen<<2) - dr->xprt_hlen; |
0f0257ea TT |
1051 | } |
1052 | ||
1053 | ||
1054 | static struct svc_deferred_req *svc_deferred_dequeue(struct svc_xprt *xprt) | |
1055 | { | |
1056 | struct svc_deferred_req *dr = NULL; | |
1057 | ||
1058 | if (!test_bit(XPT_DEFERRED, &xprt->xpt_flags)) | |
1059 | return NULL; | |
1060 | spin_lock(&xprt->xpt_lock); | |
1061 | clear_bit(XPT_DEFERRED, &xprt->xpt_flags); | |
1062 | if (!list_empty(&xprt->xpt_deferred)) { | |
1063 | dr = list_entry(xprt->xpt_deferred.next, | |
1064 | struct svc_deferred_req, | |
1065 | handle.recent); | |
1066 | list_del_init(&dr->handle.recent); | |
1067 | set_bit(XPT_DEFERRED, &xprt->xpt_flags); | |
1068 | } | |
1069 | spin_unlock(&xprt->xpt_lock); | |
1070 | return dr; | |
1071 | } | |
7fcb98d5 | 1072 | |
156e6209 CL |
1073 | /** |
1074 | * svc_find_xprt - find an RPC transport instance | |
1075 | * @serv: pointer to svc_serv to search | |
1076 | * @xcl_name: C string containing transport's class name | |
1077 | * @af: Address family of transport's local address | |
1078 | * @port: transport's IP port number | |
1079 | * | |
7fcb98d5 TT |
1080 | * Return the transport instance pointer for the endpoint accepting |
1081 | * connections/peer traffic from the specified transport class, | |
1082 | * address family and port. | |
1083 | * | |
1084 | * Specifying 0 for the address family or port is effectively a | |
1085 | * wild-card, and will result in matching the first transport in the | |
1086 | * service's list that has a matching class name. | |
1087 | */ | |
156e6209 CL |
1088 | struct svc_xprt *svc_find_xprt(struct svc_serv *serv, const char *xcl_name, |
1089 | const sa_family_t af, const unsigned short port) | |
7fcb98d5 TT |
1090 | { |
1091 | struct svc_xprt *xprt; | |
1092 | struct svc_xprt *found = NULL; | |
1093 | ||
1094 | /* Sanity check the args */ | |
156e6209 | 1095 | if (serv == NULL || xcl_name == NULL) |
7fcb98d5 TT |
1096 | return found; |
1097 | ||
1098 | spin_lock_bh(&serv->sv_lock); | |
1099 | list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list) { | |
1100 | if (strcmp(xprt->xpt_class->xcl_name, xcl_name)) | |
1101 | continue; | |
1102 | if (af != AF_UNSPEC && af != xprt->xpt_local.ss_family) | |
1103 | continue; | |
156e6209 | 1104 | if (port != 0 && port != svc_xprt_local_port(xprt)) |
7fcb98d5 TT |
1105 | continue; |
1106 | found = xprt; | |
a217813f | 1107 | svc_xprt_get(xprt); |
7fcb98d5 TT |
1108 | break; |
1109 | } | |
1110 | spin_unlock_bh(&serv->sv_lock); | |
1111 | return found; | |
1112 | } | |
1113 | EXPORT_SYMBOL_GPL(svc_find_xprt); | |
9571af18 | 1114 | |
335c54bd CL |
1115 | static int svc_one_xprt_name(const struct svc_xprt *xprt, |
1116 | char *pos, int remaining) | |
1117 | { | |
1118 | int len; | |
1119 | ||
1120 | len = snprintf(pos, remaining, "%s %u\n", | |
1121 | xprt->xpt_class->xcl_name, | |
1122 | svc_xprt_local_port(xprt)); | |
1123 | if (len >= remaining) | |
1124 | return -ENAMETOOLONG; | |
1125 | return len; | |
1126 | } | |
1127 | ||
1128 | /** | |
1129 | * svc_xprt_names - format a buffer with a list of transport names | |
1130 | * @serv: pointer to an RPC service | |
1131 | * @buf: pointer to a buffer to be filled in | |
1132 | * @buflen: length of buffer to be filled in | |
1133 | * | |
1134 | * Fills in @buf with a string containing a list of transport names, | |
1135 | * each name terminated with '\n'. | |
1136 | * | |
1137 | * Returns positive length of the filled-in string on success; otherwise | |
1138 | * a negative errno value is returned if an error occurs. | |
9571af18 | 1139 | */ |
335c54bd | 1140 | int svc_xprt_names(struct svc_serv *serv, char *buf, const int buflen) |
9571af18 TT |
1141 | { |
1142 | struct svc_xprt *xprt; | |
335c54bd CL |
1143 | int len, totlen; |
1144 | char *pos; | |
9571af18 TT |
1145 | |
1146 | /* Sanity check args */ | |
1147 | if (!serv) | |
1148 | return 0; | |
1149 | ||
1150 | spin_lock_bh(&serv->sv_lock); | |
335c54bd CL |
1151 | |
1152 | pos = buf; | |
1153 | totlen = 0; | |
9571af18 | 1154 | list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list) { |
335c54bd CL |
1155 | len = svc_one_xprt_name(xprt, pos, buflen - totlen); |
1156 | if (len < 0) { | |
1157 | *buf = '\0'; | |
1158 | totlen = len; | |
1159 | } | |
1160 | if (len <= 0) | |
9571af18 | 1161 | break; |
335c54bd CL |
1162 | |
1163 | pos += len; | |
9571af18 TT |
1164 | totlen += len; |
1165 | } | |
335c54bd | 1166 | |
9571af18 TT |
1167 | spin_unlock_bh(&serv->sv_lock); |
1168 | return totlen; | |
1169 | } | |
1170 | EXPORT_SYMBOL_GPL(svc_xprt_names); | |
03cf6c9f GB |
1171 | |
1172 | ||
1173 | /*----------------------------------------------------------------------------*/ | |
1174 | ||
1175 | static void *svc_pool_stats_start(struct seq_file *m, loff_t *pos) | |
1176 | { | |
1177 | unsigned int pidx = (unsigned int)*pos; | |
1178 | struct svc_serv *serv = m->private; | |
1179 | ||
1180 | dprintk("svc_pool_stats_start, *pidx=%u\n", pidx); | |
1181 | ||
03cf6c9f GB |
1182 | if (!pidx) |
1183 | return SEQ_START_TOKEN; | |
1184 | return (pidx > serv->sv_nrpools ? NULL : &serv->sv_pools[pidx-1]); | |
1185 | } | |
1186 | ||
1187 | static void *svc_pool_stats_next(struct seq_file *m, void *p, loff_t *pos) | |
1188 | { | |
1189 | struct svc_pool *pool = p; | |
1190 | struct svc_serv *serv = m->private; | |
1191 | ||
1192 | dprintk("svc_pool_stats_next, *pos=%llu\n", *pos); | |
1193 | ||
1194 | if (p == SEQ_START_TOKEN) { | |
1195 | pool = &serv->sv_pools[0]; | |
1196 | } else { | |
1197 | unsigned int pidx = (pool - &serv->sv_pools[0]); | |
1198 | if (pidx < serv->sv_nrpools-1) | |
1199 | pool = &serv->sv_pools[pidx+1]; | |
1200 | else | |
1201 | pool = NULL; | |
1202 | } | |
1203 | ++*pos; | |
1204 | return pool; | |
1205 | } | |
1206 | ||
1207 | static void svc_pool_stats_stop(struct seq_file *m, void *p) | |
1208 | { | |
03cf6c9f GB |
1209 | } |
1210 | ||
1211 | static int svc_pool_stats_show(struct seq_file *m, void *p) | |
1212 | { | |
1213 | struct svc_pool *pool = p; | |
1214 | ||
1215 | if (p == SEQ_START_TOKEN) { | |
78c210ef | 1216 | seq_puts(m, "# pool packets-arrived sockets-enqueued threads-woken threads-timedout\n"); |
03cf6c9f GB |
1217 | return 0; |
1218 | } | |
1219 | ||
78c210ef | 1220 | seq_printf(m, "%u %lu %lu %lu %lu\n", |
03cf6c9f GB |
1221 | pool->sp_id, |
1222 | pool->sp_stats.packets, | |
1223 | pool->sp_stats.sockets_queued, | |
1224 | pool->sp_stats.threads_woken, | |
03cf6c9f GB |
1225 | pool->sp_stats.threads_timedout); |
1226 | ||
1227 | return 0; | |
1228 | } | |
1229 | ||
1230 | static const struct seq_operations svc_pool_stats_seq_ops = { | |
1231 | .start = svc_pool_stats_start, | |
1232 | .next = svc_pool_stats_next, | |
1233 | .stop = svc_pool_stats_stop, | |
1234 | .show = svc_pool_stats_show, | |
1235 | }; | |
1236 | ||
1237 | int svc_pool_stats_open(struct svc_serv *serv, struct file *file) | |
1238 | { | |
1239 | int err; | |
1240 | ||
1241 | err = seq_open(file, &svc_pool_stats_seq_ops); | |
1242 | if (!err) | |
1243 | ((struct seq_file *) file->private_data)->private = serv; | |
1244 | return err; | |
1245 | } | |
1246 | EXPORT_SYMBOL(svc_pool_stats_open); | |
1247 | ||
1248 | /*----------------------------------------------------------------------------*/ |