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> |
99de8ea9 | 16 | #include <linux/sunrpc/xprt.h> |
1d8206b9 TT |
17 | |
18 | #define RPCDBG_FACILITY RPCDBG_SVCXPRT | |
19 | ||
0f0257ea TT |
20 | static struct svc_deferred_req *svc_deferred_dequeue(struct svc_xprt *xprt); |
21 | static int svc_deferred_recv(struct svc_rqst *rqstp); | |
22 | static struct cache_deferred_req *svc_defer(struct cache_req *req); | |
23 | static void svc_age_temp_xprts(unsigned long closure); | |
24 | ||
25 | /* apparently the "standard" is that clients close | |
26 | * idle connections after 5 minutes, servers after | |
27 | * 6 minutes | |
28 | * http://www.connectathon.org/talks96/nfstcp.pdf | |
29 | */ | |
30 | static int svc_conn_age_period = 6*60; | |
31 | ||
1d8206b9 TT |
32 | /* List of registered transport classes */ |
33 | static DEFINE_SPINLOCK(svc_xprt_class_lock); | |
34 | static LIST_HEAD(svc_xprt_class_list); | |
35 | ||
0f0257ea TT |
36 | /* SMP locking strategy: |
37 | * | |
38 | * svc_pool->sp_lock protects most of the fields of that pool. | |
39 | * svc_serv->sv_lock protects sv_tempsocks, sv_permsocks, sv_tmpcnt. | |
40 | * when both need to be taken (rare), svc_serv->sv_lock is first. | |
41 | * BKL protects svc_serv->sv_nrthread. | |
42 | * svc_sock->sk_lock protects the svc_sock->sk_deferred list | |
43 | * and the ->sk_info_authunix cache. | |
44 | * | |
45 | * The XPT_BUSY bit in xprt->xpt_flags prevents a transport being | |
46 | * enqueued multiply. During normal transport processing this bit | |
47 | * is set by svc_xprt_enqueue and cleared by svc_xprt_received. | |
48 | * Providers should not manipulate this bit directly. | |
49 | * | |
50 | * Some flags can be set to certain values at any time | |
51 | * providing that certain rules are followed: | |
52 | * | |
53 | * XPT_CONN, XPT_DATA: | |
54 | * - Can be set or cleared at any time. | |
55 | * - After a set, svc_xprt_enqueue must be called to enqueue | |
56 | * the transport for processing. | |
57 | * - After a clear, the transport must be read/accepted. | |
58 | * If this succeeds, it must be set again. | |
59 | * XPT_CLOSE: | |
60 | * - Can set at any time. It is never cleared. | |
61 | * XPT_DEAD: | |
62 | * - Can only be set while XPT_BUSY is held which ensures | |
63 | * that no other thread will be using the transport or will | |
64 | * try to set XPT_DEAD. | |
65 | */ | |
66 | ||
1d8206b9 TT |
67 | int svc_reg_xprt_class(struct svc_xprt_class *xcl) |
68 | { | |
69 | struct svc_xprt_class *cl; | |
70 | int res = -EEXIST; | |
71 | ||
72 | dprintk("svc: Adding svc transport class '%s'\n", xcl->xcl_name); | |
73 | ||
74 | INIT_LIST_HEAD(&xcl->xcl_list); | |
75 | spin_lock(&svc_xprt_class_lock); | |
76 | /* Make sure there isn't already a class with the same name */ | |
77 | list_for_each_entry(cl, &svc_xprt_class_list, xcl_list) { | |
78 | if (strcmp(xcl->xcl_name, cl->xcl_name) == 0) | |
79 | goto out; | |
80 | } | |
81 | list_add_tail(&xcl->xcl_list, &svc_xprt_class_list); | |
82 | res = 0; | |
83 | out: | |
84 | spin_unlock(&svc_xprt_class_lock); | |
85 | return res; | |
86 | } | |
87 | EXPORT_SYMBOL_GPL(svc_reg_xprt_class); | |
88 | ||
89 | void svc_unreg_xprt_class(struct svc_xprt_class *xcl) | |
90 | { | |
91 | dprintk("svc: Removing svc transport class '%s'\n", xcl->xcl_name); | |
92 | spin_lock(&svc_xprt_class_lock); | |
93 | list_del_init(&xcl->xcl_list); | |
94 | spin_unlock(&svc_xprt_class_lock); | |
95 | } | |
96 | EXPORT_SYMBOL_GPL(svc_unreg_xprt_class); | |
97 | ||
dc9a16e4 TT |
98 | /* |
99 | * Format the transport list for printing | |
100 | */ | |
101 | int svc_print_xprts(char *buf, int maxlen) | |
102 | { | |
8f3a6de3 | 103 | struct svc_xprt_class *xcl; |
dc9a16e4 TT |
104 | char tmpstr[80]; |
105 | int len = 0; | |
106 | buf[0] = '\0'; | |
107 | ||
108 | spin_lock(&svc_xprt_class_lock); | |
8f3a6de3 | 109 | list_for_each_entry(xcl, &svc_xprt_class_list, xcl_list) { |
dc9a16e4 | 110 | int slen; |
dc9a16e4 TT |
111 | |
112 | sprintf(tmpstr, "%s %d\n", xcl->xcl_name, xcl->xcl_max_payload); | |
113 | slen = strlen(tmpstr); | |
114 | if (len + slen > maxlen) | |
115 | break; | |
116 | len += slen; | |
117 | strcat(buf, tmpstr); | |
118 | } | |
119 | spin_unlock(&svc_xprt_class_lock); | |
120 | ||
121 | return len; | |
122 | } | |
123 | ||
e1b3157f TT |
124 | static void svc_xprt_free(struct kref *kref) |
125 | { | |
126 | struct svc_xprt *xprt = | |
127 | container_of(kref, struct svc_xprt, xpt_ref); | |
128 | struct module *owner = xprt->xpt_class->xcl_owner; | |
e3bfca01 PE |
129 | if (test_bit(XPT_CACHE_AUTH, &xprt->xpt_flags)) |
130 | svcauth_unix_info_release(xprt); | |
4fb8518b | 131 | put_net(xprt->xpt_net); |
99de8ea9 BF |
132 | /* See comment on corresponding get in xs_setup_bc_tcp(): */ |
133 | if (xprt->xpt_bc_xprt) | |
134 | xprt_put(xprt->xpt_bc_xprt); | |
e1b3157f TT |
135 | xprt->xpt_ops->xpo_free(xprt); |
136 | module_put(owner); | |
137 | } | |
138 | ||
139 | void svc_xprt_put(struct svc_xprt *xprt) | |
140 | { | |
141 | kref_put(&xprt->xpt_ref, svc_xprt_free); | |
142 | } | |
143 | EXPORT_SYMBOL_GPL(svc_xprt_put); | |
144 | ||
1d8206b9 TT |
145 | /* |
146 | * Called by transport drivers to initialize the transport independent | |
147 | * portion of the transport instance. | |
148 | */ | |
bb5cf160 TT |
149 | void svc_xprt_init(struct svc_xprt_class *xcl, struct svc_xprt *xprt, |
150 | struct svc_serv *serv) | |
1d8206b9 TT |
151 | { |
152 | memset(xprt, 0, sizeof(*xprt)); | |
153 | xprt->xpt_class = xcl; | |
154 | xprt->xpt_ops = xcl->xcl_ops; | |
e1b3157f | 155 | kref_init(&xprt->xpt_ref); |
bb5cf160 | 156 | xprt->xpt_server = serv; |
7a182083 TT |
157 | INIT_LIST_HEAD(&xprt->xpt_list); |
158 | INIT_LIST_HEAD(&xprt->xpt_ready); | |
8c7b0172 | 159 | INIT_LIST_HEAD(&xprt->xpt_deferred); |
edc7a894 | 160 | INIT_LIST_HEAD(&xprt->xpt_users); |
a50fea26 | 161 | mutex_init(&xprt->xpt_mutex); |
def13d74 | 162 | spin_lock_init(&xprt->xpt_lock); |
4e5caaa5 | 163 | set_bit(XPT_BUSY, &xprt->xpt_flags); |
4cfc7e60 | 164 | rpc_init_wait_queue(&xprt->xpt_bc_pending, "xpt_bc_pending"); |
4fb8518b | 165 | xprt->xpt_net = get_net(&init_net); |
1d8206b9 TT |
166 | } |
167 | EXPORT_SYMBOL_GPL(svc_xprt_init); | |
b700cbb1 | 168 | |
5dd248f6 CL |
169 | static struct svc_xprt *__svc_xpo_create(struct svc_xprt_class *xcl, |
170 | struct svc_serv *serv, | |
62832c03 | 171 | struct net *net, |
9652ada3 CL |
172 | const int family, |
173 | const unsigned short port, | |
174 | int flags) | |
b700cbb1 | 175 | { |
b700cbb1 TT |
176 | struct sockaddr_in sin = { |
177 | .sin_family = AF_INET, | |
e6f1cebf | 178 | .sin_addr.s_addr = htonl(INADDR_ANY), |
b700cbb1 TT |
179 | .sin_port = htons(port), |
180 | }; | |
d6783b2b | 181 | #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) |
5dd248f6 CL |
182 | struct sockaddr_in6 sin6 = { |
183 | .sin6_family = AF_INET6, | |
184 | .sin6_addr = IN6ADDR_ANY_INIT, | |
185 | .sin6_port = htons(port), | |
186 | }; | |
d6783b2b | 187 | #endif /* defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) */ |
5dd248f6 CL |
188 | struct sockaddr *sap; |
189 | size_t len; | |
190 | ||
9652ada3 CL |
191 | switch (family) { |
192 | case PF_INET: | |
5dd248f6 CL |
193 | sap = (struct sockaddr *)&sin; |
194 | len = sizeof(sin); | |
195 | break; | |
d6783b2b | 196 | #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) |
9652ada3 | 197 | case PF_INET6: |
5dd248f6 CL |
198 | sap = (struct sockaddr *)&sin6; |
199 | len = sizeof(sin6); | |
200 | break; | |
d6783b2b | 201 | #endif /* defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) */ |
5dd248f6 CL |
202 | default: |
203 | return ERR_PTR(-EAFNOSUPPORT); | |
204 | } | |
205 | ||
62832c03 | 206 | return xcl->xcl_ops->xpo_create(serv, net, sap, len, flags); |
5dd248f6 CL |
207 | } |
208 | ||
9652ada3 | 209 | int svc_create_xprt(struct svc_serv *serv, const char *xprt_name, |
fc5d00b0 PE |
210 | struct net *net, const int family, |
211 | const unsigned short port, int flags) | |
5dd248f6 CL |
212 | { |
213 | struct svc_xprt_class *xcl; | |
214 | ||
b700cbb1 TT |
215 | dprintk("svc: creating transport %s[%d]\n", xprt_name, port); |
216 | spin_lock(&svc_xprt_class_lock); | |
217 | list_for_each_entry(xcl, &svc_xprt_class_list, xcl_list) { | |
4e5caaa5 | 218 | struct svc_xprt *newxprt; |
ed2849d3 | 219 | unsigned short newport; |
4e5caaa5 TT |
220 | |
221 | if (strcmp(xprt_name, xcl->xcl_name)) | |
222 | continue; | |
223 | ||
224 | if (!try_module_get(xcl->xcl_owner)) | |
225 | goto err; | |
226 | ||
227 | spin_unlock(&svc_xprt_class_lock); | |
62832c03 | 228 | newxprt = __svc_xpo_create(xcl, serv, net, family, port, flags); |
4e5caaa5 TT |
229 | if (IS_ERR(newxprt)) { |
230 | module_put(xcl->xcl_owner); | |
231 | return PTR_ERR(newxprt); | |
b700cbb1 | 232 | } |
4e5caaa5 TT |
233 | |
234 | clear_bit(XPT_TEMP, &newxprt->xpt_flags); | |
235 | spin_lock_bh(&serv->sv_lock); | |
236 | list_add(&newxprt->xpt_list, &serv->sv_permsocks); | |
237 | spin_unlock_bh(&serv->sv_lock); | |
ed2849d3 | 238 | newport = svc_xprt_local_port(newxprt); |
4e5caaa5 | 239 | clear_bit(XPT_BUSY, &newxprt->xpt_flags); |
ed2849d3 | 240 | return newport; |
b700cbb1 | 241 | } |
4e5caaa5 | 242 | err: |
b700cbb1 TT |
243 | spin_unlock(&svc_xprt_class_lock); |
244 | dprintk("svc: transport %s not found\n", xprt_name); | |
68717908 CL |
245 | |
246 | /* This errno is exposed to user space. Provide a reasonable | |
247 | * perror msg for a bad transport. */ | |
248 | return -EPROTONOSUPPORT; | |
b700cbb1 TT |
249 | } |
250 | EXPORT_SYMBOL_GPL(svc_create_xprt); | |
9dbc240f TT |
251 | |
252 | /* | |
253 | * Copy the local and remote xprt addresses to the rqstp structure | |
254 | */ | |
255 | void svc_xprt_copy_addrs(struct svc_rqst *rqstp, struct svc_xprt *xprt) | |
256 | { | |
9dbc240f TT |
257 | memcpy(&rqstp->rq_addr, &xprt->xpt_remote, xprt->xpt_remotelen); |
258 | rqstp->rq_addrlen = xprt->xpt_remotelen; | |
259 | ||
260 | /* | |
261 | * Destination address in request is needed for binding the | |
262 | * source address in RPC replies/callbacks later. | |
263 | */ | |
849a1cf1 MJ |
264 | memcpy(&rqstp->rq_daddr, &xprt->xpt_local, xprt->xpt_locallen); |
265 | rqstp->rq_daddrlen = xprt->xpt_locallen; | |
9dbc240f TT |
266 | } |
267 | EXPORT_SYMBOL_GPL(svc_xprt_copy_addrs); | |
268 | ||
0f0257ea TT |
269 | /** |
270 | * svc_print_addr - Format rq_addr field for printing | |
271 | * @rqstp: svc_rqst struct containing address to print | |
272 | * @buf: target buffer for formatted address | |
273 | * @len: length of target buffer | |
274 | * | |
275 | */ | |
276 | char *svc_print_addr(struct svc_rqst *rqstp, char *buf, size_t len) | |
277 | { | |
278 | return __svc_print_addr(svc_addr(rqstp), buf, len); | |
279 | } | |
280 | EXPORT_SYMBOL_GPL(svc_print_addr); | |
281 | ||
282 | /* | |
283 | * Queue up an idle server thread. Must have pool->sp_lock held. | |
284 | * Note: this is really a stack rather than a queue, so that we only | |
285 | * use as many different threads as we need, and the rest don't pollute | |
286 | * the cache. | |
287 | */ | |
288 | static void svc_thread_enqueue(struct svc_pool *pool, struct svc_rqst *rqstp) | |
289 | { | |
290 | list_add(&rqstp->rq_list, &pool->sp_threads); | |
291 | } | |
292 | ||
293 | /* | |
294 | * Dequeue an nfsd thread. Must have pool->sp_lock held. | |
295 | */ | |
296 | static void svc_thread_dequeue(struct svc_pool *pool, struct svc_rqst *rqstp) | |
297 | { | |
298 | list_del(&rqstp->rq_list); | |
299 | } | |
300 | ||
9c335c0b BF |
301 | static bool svc_xprt_has_something_to_do(struct svc_xprt *xprt) |
302 | { | |
303 | if (xprt->xpt_flags & ((1<<XPT_CONN)|(1<<XPT_CLOSE))) | |
304 | return true; | |
305 | if (xprt->xpt_flags & ((1<<XPT_DATA)|(1<<XPT_DEFERRED))) | |
306 | return xprt->xpt_ops->xpo_has_wspace(xprt); | |
307 | return false; | |
308 | } | |
309 | ||
0f0257ea TT |
310 | /* |
311 | * Queue up a transport with data pending. If there are idle nfsd | |
312 | * processes, wake 'em up. | |
313 | * | |
314 | */ | |
315 | void svc_xprt_enqueue(struct svc_xprt *xprt) | |
316 | { | |
317 | struct svc_serv *serv = xprt->xpt_server; | |
318 | struct svc_pool *pool; | |
319 | struct svc_rqst *rqstp; | |
320 | int cpu; | |
321 | ||
9c335c0b | 322 | if (!svc_xprt_has_something_to_do(xprt)) |
0f0257ea | 323 | return; |
0f0257ea TT |
324 | |
325 | cpu = get_cpu(); | |
326 | pool = svc_pool_for_cpu(xprt->xpt_server, cpu); | |
327 | put_cpu(); | |
328 | ||
329 | spin_lock_bh(&pool->sp_lock); | |
330 | ||
78c210ef BF |
331 | if (!list_empty(&pool->sp_threads) && |
332 | !list_empty(&pool->sp_sockets)) | |
333 | printk(KERN_ERR | |
334 | "svc_xprt_enqueue: " | |
335 | "threads and transports both waiting??\n"); | |
336 | ||
03cf6c9f GB |
337 | pool->sp_stats.packets++; |
338 | ||
0f0257ea TT |
339 | /* Mark transport as busy. It will remain in this state until |
340 | * the provider calls svc_xprt_received. We update XPT_BUSY | |
341 | * atomically because it also guards against trying to enqueue | |
342 | * the transport twice. | |
343 | */ | |
344 | if (test_and_set_bit(XPT_BUSY, &xprt->xpt_flags)) { | |
345 | /* Don't enqueue transport while already enqueued */ | |
346 | dprintk("svc: transport %p busy, not enqueued\n", xprt); | |
347 | goto out_unlock; | |
348 | } | |
0f0257ea | 349 | |
78c210ef | 350 | if (!list_empty(&pool->sp_threads)) { |
0f0257ea TT |
351 | rqstp = list_entry(pool->sp_threads.next, |
352 | struct svc_rqst, | |
353 | rq_list); | |
354 | dprintk("svc: transport %p served by daemon %p\n", | |
355 | xprt, rqstp); | |
356 | svc_thread_dequeue(pool, rqstp); | |
357 | if (rqstp->rq_xprt) | |
358 | printk(KERN_ERR | |
359 | "svc_xprt_enqueue: server %p, rq_xprt=%p!\n", | |
360 | rqstp, rqstp->rq_xprt); | |
361 | rqstp->rq_xprt = xprt; | |
362 | svc_xprt_get(xprt); | |
363 | rqstp->rq_reserved = serv->sv_max_mesg; | |
364 | atomic_add(rqstp->rq_reserved, &xprt->xpt_reserved); | |
03cf6c9f | 365 | pool->sp_stats.threads_woken++; |
0f0257ea TT |
366 | wake_up(&rqstp->rq_wait); |
367 | } else { | |
368 | dprintk("svc: transport %p put into queue\n", xprt); | |
369 | list_add_tail(&xprt->xpt_ready, &pool->sp_sockets); | |
03cf6c9f | 370 | pool->sp_stats.sockets_queued++; |
0f0257ea TT |
371 | } |
372 | ||
373 | out_unlock: | |
374 | spin_unlock_bh(&pool->sp_lock); | |
375 | } | |
376 | EXPORT_SYMBOL_GPL(svc_xprt_enqueue); | |
377 | ||
378 | /* | |
379 | * Dequeue the first transport. Must be called with the pool->sp_lock held. | |
380 | */ | |
381 | static struct svc_xprt *svc_xprt_dequeue(struct svc_pool *pool) | |
382 | { | |
383 | struct svc_xprt *xprt; | |
384 | ||
385 | if (list_empty(&pool->sp_sockets)) | |
386 | return NULL; | |
387 | ||
388 | xprt = list_entry(pool->sp_sockets.next, | |
389 | struct svc_xprt, xpt_ready); | |
390 | list_del_init(&xprt->xpt_ready); | |
391 | ||
392 | dprintk("svc: transport %p dequeued, inuse=%d\n", | |
393 | xprt, atomic_read(&xprt->xpt_ref.refcount)); | |
394 | ||
395 | return xprt; | |
396 | } | |
397 | ||
398 | /* | |
399 | * svc_xprt_received conditionally queues the transport for processing | |
400 | * by another thread. The caller must hold the XPT_BUSY bit and must | |
401 | * not thereafter touch transport data. | |
402 | * | |
403 | * Note: XPT_DATA only gets cleared when a read-attempt finds no (or | |
404 | * insufficient) data. | |
405 | */ | |
406 | void svc_xprt_received(struct svc_xprt *xprt) | |
407 | { | |
408 | BUG_ON(!test_bit(XPT_BUSY, &xprt->xpt_flags)); | |
ed2849d3 N |
409 | /* As soon as we clear busy, the xprt could be closed and |
410 | * 'put', so we need a reference to call svc_xprt_enqueue with: | |
411 | */ | |
412 | svc_xprt_get(xprt); | |
0f0257ea TT |
413 | clear_bit(XPT_BUSY, &xprt->xpt_flags); |
414 | svc_xprt_enqueue(xprt); | |
ed2849d3 | 415 | svc_xprt_put(xprt); |
0f0257ea TT |
416 | } |
417 | EXPORT_SYMBOL_GPL(svc_xprt_received); | |
418 | ||
419 | /** | |
420 | * svc_reserve - change the space reserved for the reply to a request. | |
421 | * @rqstp: The request in question | |
422 | * @space: new max space to reserve | |
423 | * | |
424 | * Each request reserves some space on the output queue of the transport | |
425 | * to make sure the reply fits. This function reduces that reserved | |
426 | * space to be the amount of space used already, plus @space. | |
427 | * | |
428 | */ | |
429 | void svc_reserve(struct svc_rqst *rqstp, int space) | |
430 | { | |
431 | space += rqstp->rq_res.head[0].iov_len; | |
432 | ||
433 | if (space < rqstp->rq_reserved) { | |
434 | struct svc_xprt *xprt = rqstp->rq_xprt; | |
435 | atomic_sub((rqstp->rq_reserved - space), &xprt->xpt_reserved); | |
436 | rqstp->rq_reserved = space; | |
437 | ||
438 | svc_xprt_enqueue(xprt); | |
439 | } | |
440 | } | |
24c3767e | 441 | EXPORT_SYMBOL_GPL(svc_reserve); |
0f0257ea TT |
442 | |
443 | static void svc_xprt_release(struct svc_rqst *rqstp) | |
444 | { | |
445 | struct svc_xprt *xprt = rqstp->rq_xprt; | |
446 | ||
447 | rqstp->rq_xprt->xpt_ops->xpo_release_rqst(rqstp); | |
448 | ||
2779e3ae TT |
449 | kfree(rqstp->rq_deferred); |
450 | rqstp->rq_deferred = NULL; | |
451 | ||
0f0257ea TT |
452 | svc_free_res_pages(rqstp); |
453 | rqstp->rq_res.page_len = 0; | |
454 | rqstp->rq_res.page_base = 0; | |
455 | ||
456 | /* Reset response buffer and release | |
457 | * the reservation. | |
458 | * But first, check that enough space was reserved | |
459 | * for the reply, otherwise we have a bug! | |
460 | */ | |
461 | if ((rqstp->rq_res.len) > rqstp->rq_reserved) | |
462 | printk(KERN_ERR "RPC request reserved %d but used %d\n", | |
463 | rqstp->rq_reserved, | |
464 | rqstp->rq_res.len); | |
465 | ||
466 | rqstp->rq_res.head[0].iov_len = 0; | |
467 | svc_reserve(rqstp, 0); | |
468 | rqstp->rq_xprt = NULL; | |
469 | ||
470 | svc_xprt_put(xprt); | |
471 | } | |
472 | ||
473 | /* | |
474 | * External function to wake up a server waiting for data | |
475 | * This really only makes sense for services like lockd | |
476 | * which have exactly one thread anyway. | |
477 | */ | |
478 | void svc_wake_up(struct svc_serv *serv) | |
479 | { | |
480 | struct svc_rqst *rqstp; | |
481 | unsigned int i; | |
482 | struct svc_pool *pool; | |
483 | ||
484 | for (i = 0; i < serv->sv_nrpools; i++) { | |
485 | pool = &serv->sv_pools[i]; | |
486 | ||
487 | spin_lock_bh(&pool->sp_lock); | |
488 | if (!list_empty(&pool->sp_threads)) { | |
489 | rqstp = list_entry(pool->sp_threads.next, | |
490 | struct svc_rqst, | |
491 | rq_list); | |
492 | dprintk("svc: daemon %p woken up.\n", rqstp); | |
493 | /* | |
494 | svc_thread_dequeue(pool, rqstp); | |
495 | rqstp->rq_xprt = NULL; | |
496 | */ | |
497 | wake_up(&rqstp->rq_wait); | |
498 | } | |
499 | spin_unlock_bh(&pool->sp_lock); | |
500 | } | |
501 | } | |
24c3767e | 502 | EXPORT_SYMBOL_GPL(svc_wake_up); |
0f0257ea TT |
503 | |
504 | int svc_port_is_privileged(struct sockaddr *sin) | |
505 | { | |
506 | switch (sin->sa_family) { | |
507 | case AF_INET: | |
508 | return ntohs(((struct sockaddr_in *)sin)->sin_port) | |
509 | < PROT_SOCK; | |
510 | case AF_INET6: | |
511 | return ntohs(((struct sockaddr_in6 *)sin)->sin6_port) | |
512 | < PROT_SOCK; | |
513 | default: | |
514 | return 0; | |
515 | } | |
516 | } | |
517 | ||
518 | /* | |
c9233eb7 JL |
519 | * Make sure that we don't have too many active connections. If we have, |
520 | * something must be dropped. It's not clear what will happen if we allow | |
521 | * "too many" connections, but when dealing with network-facing software, | |
522 | * we have to code defensively. Here we do that by imposing hard limits. | |
0f0257ea TT |
523 | * |
524 | * There's no point in trying to do random drop here for DoS | |
525 | * prevention. The NFS clients does 1 reconnect in 15 seconds. An | |
526 | * attacker can easily beat that. | |
527 | * | |
528 | * The only somewhat efficient mechanism would be if drop old | |
529 | * connections from the same IP first. But right now we don't even | |
530 | * record the client IP in svc_sock. | |
c9233eb7 JL |
531 | * |
532 | * single-threaded services that expect a lot of clients will probably | |
533 | * need to set sv_maxconn to override the default value which is based | |
534 | * on the number of threads | |
0f0257ea TT |
535 | */ |
536 | static void svc_check_conn_limits(struct svc_serv *serv) | |
537 | { | |
c9233eb7 JL |
538 | unsigned int limit = serv->sv_maxconn ? serv->sv_maxconn : |
539 | (serv->sv_nrthreads+3) * 20; | |
540 | ||
541 | if (serv->sv_tmpcnt > limit) { | |
0f0257ea TT |
542 | struct svc_xprt *xprt = NULL; |
543 | spin_lock_bh(&serv->sv_lock); | |
544 | if (!list_empty(&serv->sv_tempsocks)) { | |
545 | if (net_ratelimit()) { | |
546 | /* Try to help the admin */ | |
547 | printk(KERN_NOTICE "%s: too many open " | |
c9233eb7 JL |
548 | "connections, consider increasing %s\n", |
549 | serv->sv_name, serv->sv_maxconn ? | |
550 | "the max number of connections." : | |
551 | "the number of threads."); | |
0f0257ea TT |
552 | } |
553 | /* | |
554 | * Always select the oldest connection. It's not fair, | |
555 | * but so is life | |
556 | */ | |
557 | xprt = list_entry(serv->sv_tempsocks.prev, | |
558 | struct svc_xprt, | |
559 | xpt_list); | |
560 | set_bit(XPT_CLOSE, &xprt->xpt_flags); | |
561 | svc_xprt_get(xprt); | |
562 | } | |
563 | spin_unlock_bh(&serv->sv_lock); | |
564 | ||
565 | if (xprt) { | |
566 | svc_xprt_enqueue(xprt); | |
567 | svc_xprt_put(xprt); | |
568 | } | |
569 | } | |
570 | } | |
571 | ||
572 | /* | |
573 | * Receive the next request on any transport. This code is carefully | |
574 | * organised not to touch any cachelines in the shared svc_serv | |
575 | * structure, only cachelines in the local svc_pool. | |
576 | */ | |
577 | int svc_recv(struct svc_rqst *rqstp, long timeout) | |
578 | { | |
579 | struct svc_xprt *xprt = NULL; | |
580 | struct svc_serv *serv = rqstp->rq_server; | |
581 | struct svc_pool *pool = rqstp->rq_pool; | |
582 | int len, i; | |
583 | int pages; | |
584 | struct xdr_buf *arg; | |
585 | DECLARE_WAITQUEUE(wait, current); | |
03cf6c9f | 586 | long time_left; |
0f0257ea TT |
587 | |
588 | dprintk("svc: server %p waiting for data (to = %ld)\n", | |
589 | rqstp, timeout); | |
590 | ||
591 | if (rqstp->rq_xprt) | |
592 | printk(KERN_ERR | |
593 | "svc_recv: service %p, transport not NULL!\n", | |
594 | rqstp); | |
595 | if (waitqueue_active(&rqstp->rq_wait)) | |
596 | printk(KERN_ERR | |
597 | "svc_recv: service %p, wait queue active!\n", | |
598 | rqstp); | |
599 | ||
600 | /* now allocate needed pages. If we get a failure, sleep briefly */ | |
601 | pages = (serv->sv_max_mesg + PAGE_SIZE) / PAGE_SIZE; | |
602 | for (i = 0; i < pages ; i++) | |
603 | while (rqstp->rq_pages[i] == NULL) { | |
604 | struct page *p = alloc_page(GFP_KERNEL); | |
605 | if (!p) { | |
7b54fe61 JL |
606 | set_current_state(TASK_INTERRUPTIBLE); |
607 | if (signalled() || kthread_should_stop()) { | |
608 | set_current_state(TASK_RUNNING); | |
7086721f | 609 | return -EINTR; |
7b54fe61 JL |
610 | } |
611 | schedule_timeout(msecs_to_jiffies(500)); | |
0f0257ea TT |
612 | } |
613 | rqstp->rq_pages[i] = p; | |
614 | } | |
615 | rqstp->rq_pages[i++] = NULL; /* this might be seen in nfs_read_actor */ | |
616 | BUG_ON(pages >= RPCSVC_MAXPAGES); | |
617 | ||
618 | /* Make arg->head point to first page and arg->pages point to rest */ | |
619 | arg = &rqstp->rq_arg; | |
620 | arg->head[0].iov_base = page_address(rqstp->rq_pages[0]); | |
621 | arg->head[0].iov_len = PAGE_SIZE; | |
622 | arg->pages = rqstp->rq_pages + 1; | |
623 | arg->page_base = 0; | |
624 | /* save at least one page for response */ | |
625 | arg->page_len = (pages-2)*PAGE_SIZE; | |
626 | arg->len = (pages-1)*PAGE_SIZE; | |
627 | arg->tail[0].iov_len = 0; | |
628 | ||
629 | try_to_freeze(); | |
630 | cond_resched(); | |
7086721f | 631 | if (signalled() || kthread_should_stop()) |
0f0257ea TT |
632 | return -EINTR; |
633 | ||
f16b6e8d N |
634 | /* Normally we will wait up to 5 seconds for any required |
635 | * cache information to be provided. | |
636 | */ | |
637 | rqstp->rq_chandle.thread_wait = 5*HZ; | |
638 | ||
0f0257ea TT |
639 | spin_lock_bh(&pool->sp_lock); |
640 | xprt = svc_xprt_dequeue(pool); | |
641 | if (xprt) { | |
642 | rqstp->rq_xprt = xprt; | |
643 | svc_xprt_get(xprt); | |
644 | rqstp->rq_reserved = serv->sv_max_mesg; | |
645 | atomic_add(rqstp->rq_reserved, &xprt->xpt_reserved); | |
f16b6e8d N |
646 | |
647 | /* As there is a shortage of threads and this request | |
6610f720 | 648 | * had to be queued, don't allow the thread to wait so |
f16b6e8d N |
649 | * long for cache updates. |
650 | */ | |
651 | rqstp->rq_chandle.thread_wait = 1*HZ; | |
0f0257ea TT |
652 | } else { |
653 | /* No data pending. Go to sleep */ | |
654 | svc_thread_enqueue(pool, rqstp); | |
655 | ||
656 | /* | |
657 | * We have to be able to interrupt this wait | |
658 | * to bring down the daemons ... | |
659 | */ | |
660 | set_current_state(TASK_INTERRUPTIBLE); | |
7086721f JL |
661 | |
662 | /* | |
663 | * checking kthread_should_stop() here allows us to avoid | |
664 | * locking and signalling when stopping kthreads that call | |
665 | * svc_recv. If the thread has already been woken up, then | |
666 | * we can exit here without sleeping. If not, then it | |
667 | * it'll be woken up quickly during the schedule_timeout | |
668 | */ | |
669 | if (kthread_should_stop()) { | |
670 | set_current_state(TASK_RUNNING); | |
671 | spin_unlock_bh(&pool->sp_lock); | |
672 | return -EINTR; | |
673 | } | |
674 | ||
0f0257ea TT |
675 | add_wait_queue(&rqstp->rq_wait, &wait); |
676 | spin_unlock_bh(&pool->sp_lock); | |
677 | ||
03cf6c9f | 678 | time_left = schedule_timeout(timeout); |
0f0257ea TT |
679 | |
680 | try_to_freeze(); | |
681 | ||
682 | spin_lock_bh(&pool->sp_lock); | |
683 | remove_wait_queue(&rqstp->rq_wait, &wait); | |
03cf6c9f GB |
684 | if (!time_left) |
685 | pool->sp_stats.threads_timedout++; | |
0f0257ea TT |
686 | |
687 | xprt = rqstp->rq_xprt; | |
688 | if (!xprt) { | |
689 | svc_thread_dequeue(pool, rqstp); | |
690 | spin_unlock_bh(&pool->sp_lock); | |
691 | dprintk("svc: server %p, no data yet\n", rqstp); | |
7086721f JL |
692 | if (signalled() || kthread_should_stop()) |
693 | return -EINTR; | |
694 | else | |
695 | return -EAGAIN; | |
0f0257ea TT |
696 | } |
697 | } | |
698 | spin_unlock_bh(&pool->sp_lock); | |
699 | ||
700 | len = 0; | |
1b644b6e BF |
701 | if (test_bit(XPT_CLOSE, &xprt->xpt_flags)) { |
702 | dprintk("svc_recv: found XPT_CLOSE\n"); | |
703 | svc_delete_xprt(xprt); | |
ca7896cd BF |
704 | /* Leave XPT_BUSY set on the dead xprt: */ |
705 | goto out; | |
706 | } | |
707 | if (test_bit(XPT_LISTENER, &xprt->xpt_flags)) { | |
0f0257ea TT |
708 | struct svc_xprt *newxpt; |
709 | newxpt = xprt->xpt_ops->xpo_accept(xprt); | |
710 | if (newxpt) { | |
711 | /* | |
712 | * We know this module_get will succeed because the | |
713 | * listener holds a reference too | |
714 | */ | |
715 | __module_get(newxpt->xpt_class->xcl_owner); | |
716 | svc_check_conn_limits(xprt->xpt_server); | |
717 | spin_lock_bh(&serv->sv_lock); | |
718 | set_bit(XPT_TEMP, &newxpt->xpt_flags); | |
719 | list_add(&newxpt->xpt_list, &serv->sv_tempsocks); | |
720 | serv->sv_tmpcnt++; | |
721 | if (serv->sv_temptimer.function == NULL) { | |
722 | /* setup timer to age temp transports */ | |
723 | setup_timer(&serv->sv_temptimer, | |
724 | svc_age_temp_xprts, | |
725 | (unsigned long)serv); | |
726 | mod_timer(&serv->sv_temptimer, | |
727 | jiffies + svc_conn_age_period * HZ); | |
728 | } | |
729 | spin_unlock_bh(&serv->sv_lock); | |
730 | svc_xprt_received(newxpt); | |
731 | } | |
9c335c0b | 732 | } else if (xprt->xpt_ops->xpo_has_wspace(xprt)) { |
0f0257ea TT |
733 | dprintk("svc: server %p, pool %u, transport %p, inuse=%d\n", |
734 | rqstp, pool->sp_id, xprt, | |
735 | atomic_read(&xprt->xpt_ref.refcount)); | |
736 | rqstp->rq_deferred = svc_deferred_dequeue(xprt); | |
ca7896cd | 737 | if (rqstp->rq_deferred) |
0f0257ea | 738 | len = svc_deferred_recv(rqstp); |
ca7896cd | 739 | else |
0f0257ea TT |
740 | len = xprt->xpt_ops->xpo_recvfrom(rqstp); |
741 | dprintk("svc: got len=%d\n", len); | |
742 | } | |
ca7896cd | 743 | svc_xprt_received(xprt); |
0f0257ea TT |
744 | |
745 | /* No data, incomplete (TCP) read, or accept() */ | |
ca7896cd BF |
746 | if (len == 0 || len == -EAGAIN) |
747 | goto out; | |
748 | ||
0f0257ea TT |
749 | clear_bit(XPT_OLD, &xprt->xpt_flags); |
750 | ||
751 | rqstp->rq_secure = svc_port_is_privileged(svc_addr(rqstp)); | |
752 | rqstp->rq_chandle.defer = svc_defer; | |
753 | ||
754 | if (serv->sv_stats) | |
755 | serv->sv_stats->netcnt++; | |
756 | return len; | |
ca7896cd BF |
757 | out: |
758 | rqstp->rq_res.len = 0; | |
759 | svc_xprt_release(rqstp); | |
760 | return -EAGAIN; | |
0f0257ea | 761 | } |
24c3767e | 762 | EXPORT_SYMBOL_GPL(svc_recv); |
0f0257ea TT |
763 | |
764 | /* | |
765 | * Drop request | |
766 | */ | |
767 | void svc_drop(struct svc_rqst *rqstp) | |
768 | { | |
769 | dprintk("svc: xprt %p dropped request\n", rqstp->rq_xprt); | |
770 | svc_xprt_release(rqstp); | |
771 | } | |
24c3767e | 772 | EXPORT_SYMBOL_GPL(svc_drop); |
0f0257ea TT |
773 | |
774 | /* | |
775 | * Return reply to client. | |
776 | */ | |
777 | int svc_send(struct svc_rqst *rqstp) | |
778 | { | |
779 | struct svc_xprt *xprt; | |
780 | int len; | |
781 | struct xdr_buf *xb; | |
782 | ||
783 | xprt = rqstp->rq_xprt; | |
784 | if (!xprt) | |
785 | return -EFAULT; | |
786 | ||
787 | /* release the receive skb before sending the reply */ | |
788 | rqstp->rq_xprt->xpt_ops->xpo_release_rqst(rqstp); | |
789 | ||
790 | /* calculate over-all length */ | |
791 | xb = &rqstp->rq_res; | |
792 | xb->len = xb->head[0].iov_len + | |
793 | xb->page_len + | |
794 | xb->tail[0].iov_len; | |
795 | ||
796 | /* Grab mutex to serialize outgoing data. */ | |
797 | mutex_lock(&xprt->xpt_mutex); | |
798 | if (test_bit(XPT_DEAD, &xprt->xpt_flags)) | |
799 | len = -ENOTCONN; | |
800 | else | |
801 | len = xprt->xpt_ops->xpo_sendto(rqstp); | |
802 | mutex_unlock(&xprt->xpt_mutex); | |
4cfc7e60 | 803 | rpc_wake_up(&xprt->xpt_bc_pending); |
0f0257ea TT |
804 | svc_xprt_release(rqstp); |
805 | ||
806 | if (len == -ECONNREFUSED || len == -ENOTCONN || len == -EAGAIN) | |
807 | return 0; | |
808 | return len; | |
809 | } | |
810 | ||
811 | /* | |
812 | * Timer function to close old temporary transports, using | |
813 | * a mark-and-sweep algorithm. | |
814 | */ | |
815 | static void svc_age_temp_xprts(unsigned long closure) | |
816 | { | |
817 | struct svc_serv *serv = (struct svc_serv *)closure; | |
818 | struct svc_xprt *xprt; | |
819 | struct list_head *le, *next; | |
820 | LIST_HEAD(to_be_aged); | |
821 | ||
822 | dprintk("svc_age_temp_xprts\n"); | |
823 | ||
824 | if (!spin_trylock_bh(&serv->sv_lock)) { | |
825 | /* busy, try again 1 sec later */ | |
826 | dprintk("svc_age_temp_xprts: busy\n"); | |
827 | mod_timer(&serv->sv_temptimer, jiffies + HZ); | |
828 | return; | |
829 | } | |
830 | ||
831 | list_for_each_safe(le, next, &serv->sv_tempsocks) { | |
832 | xprt = list_entry(le, struct svc_xprt, xpt_list); | |
833 | ||
834 | /* First time through, just mark it OLD. Second time | |
835 | * through, close it. */ | |
836 | if (!test_and_set_bit(XPT_OLD, &xprt->xpt_flags)) | |
837 | continue; | |
f64f9e71 JP |
838 | if (atomic_read(&xprt->xpt_ref.refcount) > 1 || |
839 | test_bit(XPT_BUSY, &xprt->xpt_flags)) | |
0f0257ea TT |
840 | continue; |
841 | svc_xprt_get(xprt); | |
842 | list_move(le, &to_be_aged); | |
843 | set_bit(XPT_CLOSE, &xprt->xpt_flags); | |
844 | set_bit(XPT_DETACHED, &xprt->xpt_flags); | |
845 | } | |
846 | spin_unlock_bh(&serv->sv_lock); | |
847 | ||
848 | while (!list_empty(&to_be_aged)) { | |
849 | le = to_be_aged.next; | |
850 | /* fiddling the xpt_list node is safe 'cos we're XPT_DETACHED */ | |
851 | list_del_init(le); | |
852 | xprt = list_entry(le, struct svc_xprt, xpt_list); | |
853 | ||
854 | dprintk("queuing xprt %p for closing\n", xprt); | |
855 | ||
856 | /* a thread will dequeue and close it soon */ | |
857 | svc_xprt_enqueue(xprt); | |
858 | svc_xprt_put(xprt); | |
859 | } | |
860 | ||
861 | mod_timer(&serv->sv_temptimer, jiffies + svc_conn_age_period * HZ); | |
862 | } | |
863 | ||
edc7a894 BF |
864 | static void call_xpt_users(struct svc_xprt *xprt) |
865 | { | |
866 | struct svc_xpt_user *u; | |
867 | ||
868 | spin_lock(&xprt->xpt_lock); | |
869 | while (!list_empty(&xprt->xpt_users)) { | |
870 | u = list_first_entry(&xprt->xpt_users, struct svc_xpt_user, list); | |
871 | list_del(&u->list); | |
872 | u->callback(u); | |
873 | } | |
874 | spin_unlock(&xprt->xpt_lock); | |
875 | } | |
876 | ||
0f0257ea TT |
877 | /* |
878 | * Remove a dead transport | |
879 | */ | |
880 | void svc_delete_xprt(struct svc_xprt *xprt) | |
881 | { | |
882 | struct svc_serv *serv = xprt->xpt_server; | |
22945e4a TT |
883 | struct svc_deferred_req *dr; |
884 | ||
885 | /* Only do this once */ | |
886 | if (test_and_set_bit(XPT_DEAD, &xprt->xpt_flags)) | |
ac9303eb | 887 | BUG(); |
0f0257ea TT |
888 | |
889 | dprintk("svc: svc_delete_xprt(%p)\n", xprt); | |
890 | xprt->xpt_ops->xpo_detach(xprt); | |
891 | ||
892 | spin_lock_bh(&serv->sv_lock); | |
893 | if (!test_and_set_bit(XPT_DETACHED, &xprt->xpt_flags)) | |
894 | list_del_init(&xprt->xpt_list); | |
895 | /* | |
ebc63e53 BF |
896 | * The only time we're called while xpt_ready is still on a list |
897 | * is while the list itself is about to be destroyed (in | |
898 | * svc_destroy). BUT svc_xprt_enqueue could still be attempting | |
899 | * to add new entries to the sp_sockets list, so we can't leave | |
900 | * a freed xprt on it. | |
0f0257ea | 901 | */ |
ebc63e53 | 902 | list_del_init(&xprt->xpt_ready); |
22945e4a TT |
903 | if (test_bit(XPT_TEMP, &xprt->xpt_flags)) |
904 | serv->sv_tmpcnt--; | |
788e69e5 | 905 | spin_unlock_bh(&serv->sv_lock); |
22945e4a | 906 | |
ab1b18f7 | 907 | while ((dr = svc_deferred_dequeue(xprt)) != NULL) |
22945e4a | 908 | kfree(dr); |
22945e4a | 909 | |
edc7a894 | 910 | call_xpt_users(xprt); |
22945e4a | 911 | svc_xprt_put(xprt); |
0f0257ea TT |
912 | } |
913 | ||
914 | void svc_close_xprt(struct svc_xprt *xprt) | |
915 | { | |
916 | set_bit(XPT_CLOSE, &xprt->xpt_flags); | |
917 | if (test_and_set_bit(XPT_BUSY, &xprt->xpt_flags)) | |
918 | /* someone else will have to effect the close */ | |
919 | return; | |
b1763316 BF |
920 | /* |
921 | * We expect svc_close_xprt() to work even when no threads are | |
922 | * running (e.g., while configuring the server before starting | |
923 | * any threads), so if the transport isn't busy, we delete | |
924 | * it ourself: | |
925 | */ | |
0f0257ea | 926 | svc_delete_xprt(xprt); |
0f0257ea | 927 | } |
a217813f | 928 | EXPORT_SYMBOL_GPL(svc_close_xprt); |
0f0257ea TT |
929 | |
930 | void svc_close_all(struct list_head *xprt_list) | |
931 | { | |
932 | struct svc_xprt *xprt; | |
933 | struct svc_xprt *tmp; | |
934 | ||
f8c0d226 BF |
935 | /* |
936 | * The server is shutting down, and no more threads are running. | |
937 | * svc_xprt_enqueue() might still be running, but at worst it | |
938 | * will re-add the xprt to sp_sockets, which will soon get | |
939 | * freed. So we don't bother with any more locking, and don't | |
940 | * leave the close to the (nonexistent) server threads: | |
941 | */ | |
0f0257ea TT |
942 | list_for_each_entry_safe(xprt, tmp, xprt_list, xpt_list) { |
943 | set_bit(XPT_CLOSE, &xprt->xpt_flags); | |
f8c0d226 | 944 | svc_delete_xprt(xprt); |
0f0257ea TT |
945 | } |
946 | } | |
947 | ||
948 | /* | |
949 | * Handle defer and revisit of requests | |
950 | */ | |
951 | ||
952 | static void svc_revisit(struct cache_deferred_req *dreq, int too_many) | |
953 | { | |
954 | struct svc_deferred_req *dr = | |
955 | container_of(dreq, struct svc_deferred_req, handle); | |
956 | struct svc_xprt *xprt = dr->xprt; | |
957 | ||
22945e4a TT |
958 | spin_lock(&xprt->xpt_lock); |
959 | set_bit(XPT_DEFERRED, &xprt->xpt_flags); | |
960 | if (too_many || test_bit(XPT_DEAD, &xprt->xpt_flags)) { | |
961 | spin_unlock(&xprt->xpt_lock); | |
962 | dprintk("revisit canceled\n"); | |
0f0257ea TT |
963 | svc_xprt_put(xprt); |
964 | kfree(dr); | |
965 | return; | |
966 | } | |
967 | dprintk("revisit queued\n"); | |
968 | dr->xprt = NULL; | |
0f0257ea TT |
969 | list_add(&dr->handle.recent, &xprt->xpt_deferred); |
970 | spin_unlock(&xprt->xpt_lock); | |
0f0257ea TT |
971 | svc_xprt_enqueue(xprt); |
972 | svc_xprt_put(xprt); | |
973 | } | |
974 | ||
260c1d12 TT |
975 | /* |
976 | * Save the request off for later processing. The request buffer looks | |
977 | * like this: | |
978 | * | |
979 | * <xprt-header><rpc-header><rpc-pagelist><rpc-tail> | |
980 | * | |
981 | * This code can only handle requests that consist of an xprt-header | |
982 | * and rpc-header. | |
983 | */ | |
0f0257ea TT |
984 | static struct cache_deferred_req *svc_defer(struct cache_req *req) |
985 | { | |
986 | struct svc_rqst *rqstp = container_of(req, struct svc_rqst, rq_chandle); | |
0f0257ea TT |
987 | struct svc_deferred_req *dr; |
988 | ||
2f425878 | 989 | if (rqstp->rq_arg.page_len || !rqstp->rq_usedeferral) |
0f0257ea TT |
990 | return NULL; /* if more than a page, give up FIXME */ |
991 | if (rqstp->rq_deferred) { | |
992 | dr = rqstp->rq_deferred; | |
993 | rqstp->rq_deferred = NULL; | |
994 | } else { | |
260c1d12 TT |
995 | size_t skip; |
996 | size_t size; | |
0f0257ea | 997 | /* FIXME maybe discard if size too large */ |
260c1d12 | 998 | size = sizeof(struct svc_deferred_req) + rqstp->rq_arg.len; |
0f0257ea TT |
999 | dr = kmalloc(size, GFP_KERNEL); |
1000 | if (dr == NULL) | |
1001 | return NULL; | |
1002 | ||
1003 | dr->handle.owner = rqstp->rq_server; | |
1004 | dr->prot = rqstp->rq_prot; | |
1005 | memcpy(&dr->addr, &rqstp->rq_addr, rqstp->rq_addrlen); | |
1006 | dr->addrlen = rqstp->rq_addrlen; | |
1007 | dr->daddr = rqstp->rq_daddr; | |
1008 | dr->argslen = rqstp->rq_arg.len >> 2; | |
260c1d12 TT |
1009 | dr->xprt_hlen = rqstp->rq_xprt_hlen; |
1010 | ||
1011 | /* back up head to the start of the buffer and copy */ | |
1012 | skip = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len; | |
1013 | memcpy(dr->args, rqstp->rq_arg.head[0].iov_base - skip, | |
1014 | dr->argslen << 2); | |
0f0257ea TT |
1015 | } |
1016 | svc_xprt_get(rqstp->rq_xprt); | |
1017 | dr->xprt = rqstp->rq_xprt; | |
9e701c61 | 1018 | rqstp->rq_dropme = true; |
0f0257ea TT |
1019 | |
1020 | dr->handle.revisit = svc_revisit; | |
1021 | return &dr->handle; | |
1022 | } | |
1023 | ||
1024 | /* | |
1025 | * recv data from a deferred request into an active one | |
1026 | */ | |
1027 | static int svc_deferred_recv(struct svc_rqst *rqstp) | |
1028 | { | |
1029 | struct svc_deferred_req *dr = rqstp->rq_deferred; | |
1030 | ||
260c1d12 TT |
1031 | /* setup iov_base past transport header */ |
1032 | rqstp->rq_arg.head[0].iov_base = dr->args + (dr->xprt_hlen>>2); | |
1033 | /* The iov_len does not include the transport header bytes */ | |
1034 | rqstp->rq_arg.head[0].iov_len = (dr->argslen<<2) - dr->xprt_hlen; | |
0f0257ea | 1035 | rqstp->rq_arg.page_len = 0; |
260c1d12 TT |
1036 | /* The rq_arg.len includes the transport header bytes */ |
1037 | rqstp->rq_arg.len = dr->argslen<<2; | |
0f0257ea TT |
1038 | rqstp->rq_prot = dr->prot; |
1039 | memcpy(&rqstp->rq_addr, &dr->addr, dr->addrlen); | |
1040 | rqstp->rq_addrlen = dr->addrlen; | |
260c1d12 TT |
1041 | /* Save off transport header len in case we get deferred again */ |
1042 | rqstp->rq_xprt_hlen = dr->xprt_hlen; | |
0f0257ea TT |
1043 | rqstp->rq_daddr = dr->daddr; |
1044 | rqstp->rq_respages = rqstp->rq_pages; | |
260c1d12 | 1045 | return (dr->argslen<<2) - dr->xprt_hlen; |
0f0257ea TT |
1046 | } |
1047 | ||
1048 | ||
1049 | static struct svc_deferred_req *svc_deferred_dequeue(struct svc_xprt *xprt) | |
1050 | { | |
1051 | struct svc_deferred_req *dr = NULL; | |
1052 | ||
1053 | if (!test_bit(XPT_DEFERRED, &xprt->xpt_flags)) | |
1054 | return NULL; | |
1055 | spin_lock(&xprt->xpt_lock); | |
0f0257ea TT |
1056 | if (!list_empty(&xprt->xpt_deferred)) { |
1057 | dr = list_entry(xprt->xpt_deferred.next, | |
1058 | struct svc_deferred_req, | |
1059 | handle.recent); | |
1060 | list_del_init(&dr->handle.recent); | |
62bac4af BF |
1061 | } else |
1062 | clear_bit(XPT_DEFERRED, &xprt->xpt_flags); | |
0f0257ea TT |
1063 | spin_unlock(&xprt->xpt_lock); |
1064 | return dr; | |
1065 | } | |
7fcb98d5 | 1066 | |
156e6209 CL |
1067 | /** |
1068 | * svc_find_xprt - find an RPC transport instance | |
1069 | * @serv: pointer to svc_serv to search | |
1070 | * @xcl_name: C string containing transport's class name | |
1071 | * @af: Address family of transport's local address | |
1072 | * @port: transport's IP port number | |
1073 | * | |
7fcb98d5 TT |
1074 | * Return the transport instance pointer for the endpoint accepting |
1075 | * connections/peer traffic from the specified transport class, | |
1076 | * address family and port. | |
1077 | * | |
1078 | * Specifying 0 for the address family or port is effectively a | |
1079 | * wild-card, and will result in matching the first transport in the | |
1080 | * service's list that has a matching class name. | |
1081 | */ | |
156e6209 CL |
1082 | struct svc_xprt *svc_find_xprt(struct svc_serv *serv, const char *xcl_name, |
1083 | const sa_family_t af, const unsigned short port) | |
7fcb98d5 TT |
1084 | { |
1085 | struct svc_xprt *xprt; | |
1086 | struct svc_xprt *found = NULL; | |
1087 | ||
1088 | /* Sanity check the args */ | |
156e6209 | 1089 | if (serv == NULL || xcl_name == NULL) |
7fcb98d5 TT |
1090 | return found; |
1091 | ||
1092 | spin_lock_bh(&serv->sv_lock); | |
1093 | list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list) { | |
1094 | if (strcmp(xprt->xpt_class->xcl_name, xcl_name)) | |
1095 | continue; | |
1096 | if (af != AF_UNSPEC && af != xprt->xpt_local.ss_family) | |
1097 | continue; | |
156e6209 | 1098 | if (port != 0 && port != svc_xprt_local_port(xprt)) |
7fcb98d5 TT |
1099 | continue; |
1100 | found = xprt; | |
a217813f | 1101 | svc_xprt_get(xprt); |
7fcb98d5 TT |
1102 | break; |
1103 | } | |
1104 | spin_unlock_bh(&serv->sv_lock); | |
1105 | return found; | |
1106 | } | |
1107 | EXPORT_SYMBOL_GPL(svc_find_xprt); | |
9571af18 | 1108 | |
335c54bd CL |
1109 | static int svc_one_xprt_name(const struct svc_xprt *xprt, |
1110 | char *pos, int remaining) | |
1111 | { | |
1112 | int len; | |
1113 | ||
1114 | len = snprintf(pos, remaining, "%s %u\n", | |
1115 | xprt->xpt_class->xcl_name, | |
1116 | svc_xprt_local_port(xprt)); | |
1117 | if (len >= remaining) | |
1118 | return -ENAMETOOLONG; | |
1119 | return len; | |
1120 | } | |
1121 | ||
1122 | /** | |
1123 | * svc_xprt_names - format a buffer with a list of transport names | |
1124 | * @serv: pointer to an RPC service | |
1125 | * @buf: pointer to a buffer to be filled in | |
1126 | * @buflen: length of buffer to be filled in | |
1127 | * | |
1128 | * Fills in @buf with a string containing a list of transport names, | |
1129 | * each name terminated with '\n'. | |
1130 | * | |
1131 | * Returns positive length of the filled-in string on success; otherwise | |
1132 | * a negative errno value is returned if an error occurs. | |
9571af18 | 1133 | */ |
335c54bd | 1134 | int svc_xprt_names(struct svc_serv *serv, char *buf, const int buflen) |
9571af18 TT |
1135 | { |
1136 | struct svc_xprt *xprt; | |
335c54bd CL |
1137 | int len, totlen; |
1138 | char *pos; | |
9571af18 TT |
1139 | |
1140 | /* Sanity check args */ | |
1141 | if (!serv) | |
1142 | return 0; | |
1143 | ||
1144 | spin_lock_bh(&serv->sv_lock); | |
335c54bd CL |
1145 | |
1146 | pos = buf; | |
1147 | totlen = 0; | |
9571af18 | 1148 | list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list) { |
335c54bd CL |
1149 | len = svc_one_xprt_name(xprt, pos, buflen - totlen); |
1150 | if (len < 0) { | |
1151 | *buf = '\0'; | |
1152 | totlen = len; | |
1153 | } | |
1154 | if (len <= 0) | |
9571af18 | 1155 | break; |
335c54bd CL |
1156 | |
1157 | pos += len; | |
9571af18 TT |
1158 | totlen += len; |
1159 | } | |
335c54bd | 1160 | |
9571af18 TT |
1161 | spin_unlock_bh(&serv->sv_lock); |
1162 | return totlen; | |
1163 | } | |
1164 | EXPORT_SYMBOL_GPL(svc_xprt_names); | |
03cf6c9f GB |
1165 | |
1166 | ||
1167 | /*----------------------------------------------------------------------------*/ | |
1168 | ||
1169 | static void *svc_pool_stats_start(struct seq_file *m, loff_t *pos) | |
1170 | { | |
1171 | unsigned int pidx = (unsigned int)*pos; | |
1172 | struct svc_serv *serv = m->private; | |
1173 | ||
1174 | dprintk("svc_pool_stats_start, *pidx=%u\n", pidx); | |
1175 | ||
03cf6c9f GB |
1176 | if (!pidx) |
1177 | return SEQ_START_TOKEN; | |
1178 | return (pidx > serv->sv_nrpools ? NULL : &serv->sv_pools[pidx-1]); | |
1179 | } | |
1180 | ||
1181 | static void *svc_pool_stats_next(struct seq_file *m, void *p, loff_t *pos) | |
1182 | { | |
1183 | struct svc_pool *pool = p; | |
1184 | struct svc_serv *serv = m->private; | |
1185 | ||
1186 | dprintk("svc_pool_stats_next, *pos=%llu\n", *pos); | |
1187 | ||
1188 | if (p == SEQ_START_TOKEN) { | |
1189 | pool = &serv->sv_pools[0]; | |
1190 | } else { | |
1191 | unsigned int pidx = (pool - &serv->sv_pools[0]); | |
1192 | if (pidx < serv->sv_nrpools-1) | |
1193 | pool = &serv->sv_pools[pidx+1]; | |
1194 | else | |
1195 | pool = NULL; | |
1196 | } | |
1197 | ++*pos; | |
1198 | return pool; | |
1199 | } | |
1200 | ||
1201 | static void svc_pool_stats_stop(struct seq_file *m, void *p) | |
1202 | { | |
03cf6c9f GB |
1203 | } |
1204 | ||
1205 | static int svc_pool_stats_show(struct seq_file *m, void *p) | |
1206 | { | |
1207 | struct svc_pool *pool = p; | |
1208 | ||
1209 | if (p == SEQ_START_TOKEN) { | |
78c210ef | 1210 | seq_puts(m, "# pool packets-arrived sockets-enqueued threads-woken threads-timedout\n"); |
03cf6c9f GB |
1211 | return 0; |
1212 | } | |
1213 | ||
78c210ef | 1214 | seq_printf(m, "%u %lu %lu %lu %lu\n", |
03cf6c9f GB |
1215 | pool->sp_id, |
1216 | pool->sp_stats.packets, | |
1217 | pool->sp_stats.sockets_queued, | |
1218 | pool->sp_stats.threads_woken, | |
03cf6c9f GB |
1219 | pool->sp_stats.threads_timedout); |
1220 | ||
1221 | return 0; | |
1222 | } | |
1223 | ||
1224 | static const struct seq_operations svc_pool_stats_seq_ops = { | |
1225 | .start = svc_pool_stats_start, | |
1226 | .next = svc_pool_stats_next, | |
1227 | .stop = svc_pool_stats_stop, | |
1228 | .show = svc_pool_stats_show, | |
1229 | }; | |
1230 | ||
1231 | int svc_pool_stats_open(struct svc_serv *serv, struct file *file) | |
1232 | { | |
1233 | int err; | |
1234 | ||
1235 | err = seq_open(file, &svc_pool_stats_seq_ops); | |
1236 | if (!err) | |
1237 | ((struct seq_file *) file->private_data)->private = serv; | |
1238 | return err; | |
1239 | } | |
1240 | EXPORT_SYMBOL(svc_pool_stats_open); | |
1241 | ||
1242 | /*----------------------------------------------------------------------------*/ |