]>
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> |
3a9a231d | 17 | #include <linux/module.h> |
860a0d9e | 18 | #include <trace/events/sunrpc.h> |
1d8206b9 TT |
19 | |
20 | #define RPCDBG_FACILITY RPCDBG_SVCXPRT | |
21 | ||
0f0257ea TT |
22 | static struct svc_deferred_req *svc_deferred_dequeue(struct svc_xprt *xprt); |
23 | static int svc_deferred_recv(struct svc_rqst *rqstp); | |
24 | static struct cache_deferred_req *svc_defer(struct cache_req *req); | |
25 | static void svc_age_temp_xprts(unsigned long closure); | |
7710ec36 | 26 | static void svc_delete_xprt(struct svc_xprt *xprt); |
0971374e | 27 | static void svc_xprt_do_enqueue(struct svc_xprt *xprt); |
0f0257ea TT |
28 | |
29 | /* apparently the "standard" is that clients close | |
30 | * idle connections after 5 minutes, servers after | |
31 | * 6 minutes | |
32 | * http://www.connectathon.org/talks96/nfstcp.pdf | |
33 | */ | |
34 | static int svc_conn_age_period = 6*60; | |
35 | ||
1d8206b9 TT |
36 | /* List of registered transport classes */ |
37 | static DEFINE_SPINLOCK(svc_xprt_class_lock); | |
38 | static LIST_HEAD(svc_xprt_class_list); | |
39 | ||
0f0257ea TT |
40 | /* SMP locking strategy: |
41 | * | |
42 | * svc_pool->sp_lock protects most of the fields of that pool. | |
43 | * svc_serv->sv_lock protects sv_tempsocks, sv_permsocks, sv_tmpcnt. | |
44 | * when both need to be taken (rare), svc_serv->sv_lock is first. | |
45 | * BKL protects svc_serv->sv_nrthread. | |
46 | * svc_sock->sk_lock protects the svc_sock->sk_deferred list | |
47 | * and the ->sk_info_authunix cache. | |
48 | * | |
49 | * The XPT_BUSY bit in xprt->xpt_flags prevents a transport being | |
50 | * enqueued multiply. During normal transport processing this bit | |
51 | * is set by svc_xprt_enqueue and cleared by svc_xprt_received. | |
52 | * Providers should not manipulate this bit directly. | |
53 | * | |
54 | * Some flags can be set to certain values at any time | |
55 | * providing that certain rules are followed: | |
56 | * | |
57 | * XPT_CONN, XPT_DATA: | |
58 | * - Can be set or cleared at any time. | |
59 | * - After a set, svc_xprt_enqueue must be called to enqueue | |
60 | * the transport for processing. | |
61 | * - After a clear, the transport must be read/accepted. | |
62 | * If this succeeds, it must be set again. | |
63 | * XPT_CLOSE: | |
64 | * - Can set at any time. It is never cleared. | |
65 | * XPT_DEAD: | |
66 | * - Can only be set while XPT_BUSY is held which ensures | |
67 | * that no other thread will be using the transport or will | |
68 | * try to set XPT_DEAD. | |
69 | */ | |
70 | ||
1d8206b9 TT |
71 | int svc_reg_xprt_class(struct svc_xprt_class *xcl) |
72 | { | |
73 | struct svc_xprt_class *cl; | |
74 | int res = -EEXIST; | |
75 | ||
76 | dprintk("svc: Adding svc transport class '%s'\n", xcl->xcl_name); | |
77 | ||
78 | INIT_LIST_HEAD(&xcl->xcl_list); | |
79 | spin_lock(&svc_xprt_class_lock); | |
80 | /* Make sure there isn't already a class with the same name */ | |
81 | list_for_each_entry(cl, &svc_xprt_class_list, xcl_list) { | |
82 | if (strcmp(xcl->xcl_name, cl->xcl_name) == 0) | |
83 | goto out; | |
84 | } | |
85 | list_add_tail(&xcl->xcl_list, &svc_xprt_class_list); | |
86 | res = 0; | |
87 | out: | |
88 | spin_unlock(&svc_xprt_class_lock); | |
89 | return res; | |
90 | } | |
91 | EXPORT_SYMBOL_GPL(svc_reg_xprt_class); | |
92 | ||
93 | void svc_unreg_xprt_class(struct svc_xprt_class *xcl) | |
94 | { | |
95 | dprintk("svc: Removing svc transport class '%s'\n", xcl->xcl_name); | |
96 | spin_lock(&svc_xprt_class_lock); | |
97 | list_del_init(&xcl->xcl_list); | |
98 | spin_unlock(&svc_xprt_class_lock); | |
99 | } | |
100 | EXPORT_SYMBOL_GPL(svc_unreg_xprt_class); | |
101 | ||
dc9a16e4 TT |
102 | /* |
103 | * Format the transport list for printing | |
104 | */ | |
105 | int svc_print_xprts(char *buf, int maxlen) | |
106 | { | |
8f3a6de3 | 107 | struct svc_xprt_class *xcl; |
dc9a16e4 TT |
108 | char tmpstr[80]; |
109 | int len = 0; | |
110 | buf[0] = '\0'; | |
111 | ||
112 | spin_lock(&svc_xprt_class_lock); | |
8f3a6de3 | 113 | list_for_each_entry(xcl, &svc_xprt_class_list, xcl_list) { |
dc9a16e4 | 114 | int slen; |
dc9a16e4 TT |
115 | |
116 | sprintf(tmpstr, "%s %d\n", xcl->xcl_name, xcl->xcl_max_payload); | |
117 | slen = strlen(tmpstr); | |
118 | if (len + slen > maxlen) | |
119 | break; | |
120 | len += slen; | |
121 | strcat(buf, tmpstr); | |
122 | } | |
123 | spin_unlock(&svc_xprt_class_lock); | |
124 | ||
125 | return len; | |
126 | } | |
127 | ||
e1b3157f TT |
128 | static void svc_xprt_free(struct kref *kref) |
129 | { | |
130 | struct svc_xprt *xprt = | |
131 | container_of(kref, struct svc_xprt, xpt_ref); | |
132 | struct module *owner = xprt->xpt_class->xcl_owner; | |
e3bfca01 PE |
133 | if (test_bit(XPT_CACHE_AUTH, &xprt->xpt_flags)) |
134 | svcauth_unix_info_release(xprt); | |
4fb8518b | 135 | put_net(xprt->xpt_net); |
99de8ea9 BF |
136 | /* See comment on corresponding get in xs_setup_bc_tcp(): */ |
137 | if (xprt->xpt_bc_xprt) | |
138 | xprt_put(xprt->xpt_bc_xprt); | |
e1b3157f TT |
139 | xprt->xpt_ops->xpo_free(xprt); |
140 | module_put(owner); | |
141 | } | |
142 | ||
143 | void svc_xprt_put(struct svc_xprt *xprt) | |
144 | { | |
145 | kref_put(&xprt->xpt_ref, svc_xprt_free); | |
146 | } | |
147 | EXPORT_SYMBOL_GPL(svc_xprt_put); | |
148 | ||
1d8206b9 TT |
149 | /* |
150 | * Called by transport drivers to initialize the transport independent | |
151 | * portion of the transport instance. | |
152 | */ | |
bd4620dd SK |
153 | void svc_xprt_init(struct net *net, struct svc_xprt_class *xcl, |
154 | struct svc_xprt *xprt, struct svc_serv *serv) | |
1d8206b9 TT |
155 | { |
156 | memset(xprt, 0, sizeof(*xprt)); | |
157 | xprt->xpt_class = xcl; | |
158 | xprt->xpt_ops = xcl->xcl_ops; | |
e1b3157f | 159 | kref_init(&xprt->xpt_ref); |
bb5cf160 | 160 | xprt->xpt_server = serv; |
7a182083 TT |
161 | INIT_LIST_HEAD(&xprt->xpt_list); |
162 | INIT_LIST_HEAD(&xprt->xpt_ready); | |
8c7b0172 | 163 | INIT_LIST_HEAD(&xprt->xpt_deferred); |
edc7a894 | 164 | INIT_LIST_HEAD(&xprt->xpt_users); |
a50fea26 | 165 | mutex_init(&xprt->xpt_mutex); |
def13d74 | 166 | spin_lock_init(&xprt->xpt_lock); |
4e5caaa5 | 167 | set_bit(XPT_BUSY, &xprt->xpt_flags); |
4cfc7e60 | 168 | rpc_init_wait_queue(&xprt->xpt_bc_pending, "xpt_bc_pending"); |
bd4620dd | 169 | xprt->xpt_net = get_net(net); |
1d8206b9 TT |
170 | } |
171 | EXPORT_SYMBOL_GPL(svc_xprt_init); | |
b700cbb1 | 172 | |
5dd248f6 CL |
173 | static struct svc_xprt *__svc_xpo_create(struct svc_xprt_class *xcl, |
174 | struct svc_serv *serv, | |
62832c03 | 175 | struct net *net, |
9652ada3 CL |
176 | const int family, |
177 | const unsigned short port, | |
178 | int flags) | |
b700cbb1 | 179 | { |
b700cbb1 TT |
180 | struct sockaddr_in sin = { |
181 | .sin_family = AF_INET, | |
e6f1cebf | 182 | .sin_addr.s_addr = htonl(INADDR_ANY), |
b700cbb1 TT |
183 | .sin_port = htons(port), |
184 | }; | |
dfd56b8b | 185 | #if IS_ENABLED(CONFIG_IPV6) |
5dd248f6 CL |
186 | struct sockaddr_in6 sin6 = { |
187 | .sin6_family = AF_INET6, | |
188 | .sin6_addr = IN6ADDR_ANY_INIT, | |
189 | .sin6_port = htons(port), | |
190 | }; | |
dfd56b8b | 191 | #endif |
5dd248f6 CL |
192 | struct sockaddr *sap; |
193 | size_t len; | |
194 | ||
9652ada3 CL |
195 | switch (family) { |
196 | case PF_INET: | |
5dd248f6 CL |
197 | sap = (struct sockaddr *)&sin; |
198 | len = sizeof(sin); | |
199 | break; | |
dfd56b8b | 200 | #if IS_ENABLED(CONFIG_IPV6) |
9652ada3 | 201 | case PF_INET6: |
5dd248f6 CL |
202 | sap = (struct sockaddr *)&sin6; |
203 | len = sizeof(sin6); | |
204 | break; | |
dfd56b8b | 205 | #endif |
5dd248f6 CL |
206 | default: |
207 | return ERR_PTR(-EAFNOSUPPORT); | |
208 | } | |
209 | ||
62832c03 | 210 | return xcl->xcl_ops->xpo_create(serv, net, sap, len, flags); |
5dd248f6 CL |
211 | } |
212 | ||
6741019c BF |
213 | /* |
214 | * svc_xprt_received conditionally queues the transport for processing | |
215 | * by another thread. The caller must hold the XPT_BUSY bit and must | |
216 | * not thereafter touch transport data. | |
217 | * | |
218 | * Note: XPT_DATA only gets cleared when a read-attempt finds no (or | |
219 | * insufficient) data. | |
220 | */ | |
221 | static void svc_xprt_received(struct svc_xprt *xprt) | |
222 | { | |
ff1fdb9b WAA |
223 | WARN_ON_ONCE(!test_bit(XPT_BUSY, &xprt->xpt_flags)); |
224 | if (!test_bit(XPT_BUSY, &xprt->xpt_flags)) | |
225 | return; | |
6741019c | 226 | /* As soon as we clear busy, the xprt could be closed and |
0971374e | 227 | * 'put', so we need a reference to call svc_xprt_do_enqueue with: |
6741019c BF |
228 | */ |
229 | svc_xprt_get(xprt); | |
0971374e | 230 | smp_mb__before_atomic(); |
6741019c | 231 | clear_bit(XPT_BUSY, &xprt->xpt_flags); |
0971374e | 232 | svc_xprt_do_enqueue(xprt); |
6741019c BF |
233 | svc_xprt_put(xprt); |
234 | } | |
235 | ||
39b55301 BF |
236 | void svc_add_new_perm_xprt(struct svc_serv *serv, struct svc_xprt *new) |
237 | { | |
238 | clear_bit(XPT_TEMP, &new->xpt_flags); | |
239 | spin_lock_bh(&serv->sv_lock); | |
240 | list_add(&new->xpt_list, &serv->sv_permsocks); | |
241 | spin_unlock_bh(&serv->sv_lock); | |
242 | svc_xprt_received(new); | |
243 | } | |
244 | ||
9652ada3 | 245 | int svc_create_xprt(struct svc_serv *serv, const char *xprt_name, |
fc5d00b0 PE |
246 | struct net *net, const int family, |
247 | const unsigned short port, int flags) | |
5dd248f6 CL |
248 | { |
249 | struct svc_xprt_class *xcl; | |
250 | ||
b700cbb1 TT |
251 | dprintk("svc: creating transport %s[%d]\n", xprt_name, port); |
252 | spin_lock(&svc_xprt_class_lock); | |
253 | list_for_each_entry(xcl, &svc_xprt_class_list, xcl_list) { | |
4e5caaa5 | 254 | struct svc_xprt *newxprt; |
ed2849d3 | 255 | unsigned short newport; |
4e5caaa5 TT |
256 | |
257 | if (strcmp(xprt_name, xcl->xcl_name)) | |
258 | continue; | |
259 | ||
260 | if (!try_module_get(xcl->xcl_owner)) | |
261 | goto err; | |
262 | ||
263 | spin_unlock(&svc_xprt_class_lock); | |
62832c03 | 264 | newxprt = __svc_xpo_create(xcl, serv, net, family, port, flags); |
4e5caaa5 TT |
265 | if (IS_ERR(newxprt)) { |
266 | module_put(xcl->xcl_owner); | |
267 | return PTR_ERR(newxprt); | |
b700cbb1 | 268 | } |
39b55301 | 269 | svc_add_new_perm_xprt(serv, newxprt); |
ed2849d3 | 270 | newport = svc_xprt_local_port(newxprt); |
ed2849d3 | 271 | return newport; |
b700cbb1 | 272 | } |
4e5caaa5 | 273 | err: |
b700cbb1 TT |
274 | spin_unlock(&svc_xprt_class_lock); |
275 | dprintk("svc: transport %s not found\n", xprt_name); | |
68717908 CL |
276 | |
277 | /* This errno is exposed to user space. Provide a reasonable | |
278 | * perror msg for a bad transport. */ | |
279 | return -EPROTONOSUPPORT; | |
b700cbb1 TT |
280 | } |
281 | EXPORT_SYMBOL_GPL(svc_create_xprt); | |
9dbc240f TT |
282 | |
283 | /* | |
284 | * Copy the local and remote xprt addresses to the rqstp structure | |
285 | */ | |
286 | void svc_xprt_copy_addrs(struct svc_rqst *rqstp, struct svc_xprt *xprt) | |
287 | { | |
9dbc240f TT |
288 | memcpy(&rqstp->rq_addr, &xprt->xpt_remote, xprt->xpt_remotelen); |
289 | rqstp->rq_addrlen = xprt->xpt_remotelen; | |
290 | ||
291 | /* | |
292 | * Destination address in request is needed for binding the | |
293 | * source address in RPC replies/callbacks later. | |
294 | */ | |
849a1cf1 MJ |
295 | memcpy(&rqstp->rq_daddr, &xprt->xpt_local, xprt->xpt_locallen); |
296 | rqstp->rq_daddrlen = xprt->xpt_locallen; | |
9dbc240f TT |
297 | } |
298 | EXPORT_SYMBOL_GPL(svc_xprt_copy_addrs); | |
299 | ||
0f0257ea TT |
300 | /** |
301 | * svc_print_addr - Format rq_addr field for printing | |
302 | * @rqstp: svc_rqst struct containing address to print | |
303 | * @buf: target buffer for formatted address | |
304 | * @len: length of target buffer | |
305 | * | |
306 | */ | |
307 | char *svc_print_addr(struct svc_rqst *rqstp, char *buf, size_t len) | |
308 | { | |
309 | return __svc_print_addr(svc_addr(rqstp), buf, len); | |
310 | } | |
311 | EXPORT_SYMBOL_GPL(svc_print_addr); | |
312 | ||
313 | /* | |
314 | * Queue up an idle server thread. Must have pool->sp_lock held. | |
315 | * Note: this is really a stack rather than a queue, so that we only | |
316 | * use as many different threads as we need, and the rest don't pollute | |
317 | * the cache. | |
318 | */ | |
319 | static void svc_thread_enqueue(struct svc_pool *pool, struct svc_rqst *rqstp) | |
320 | { | |
321 | list_add(&rqstp->rq_list, &pool->sp_threads); | |
322 | } | |
323 | ||
324 | /* | |
325 | * Dequeue an nfsd thread. Must have pool->sp_lock held. | |
326 | */ | |
327 | static void svc_thread_dequeue(struct svc_pool *pool, struct svc_rqst *rqstp) | |
328 | { | |
329 | list_del(&rqstp->rq_list); | |
330 | } | |
331 | ||
9c335c0b BF |
332 | static bool svc_xprt_has_something_to_do(struct svc_xprt *xprt) |
333 | { | |
334 | if (xprt->xpt_flags & ((1<<XPT_CONN)|(1<<XPT_CLOSE))) | |
335 | return true; | |
336 | if (xprt->xpt_flags & ((1<<XPT_DATA)|(1<<XPT_DEFERRED))) | |
337 | return xprt->xpt_ops->xpo_has_wspace(xprt); | |
338 | return false; | |
339 | } | |
340 | ||
0971374e | 341 | static void svc_xprt_do_enqueue(struct svc_xprt *xprt) |
0f0257ea | 342 | { |
0f0257ea TT |
343 | struct svc_pool *pool; |
344 | struct svc_rqst *rqstp; | |
345 | int cpu; | |
346 | ||
9c335c0b | 347 | if (!svc_xprt_has_something_to_do(xprt)) |
0f0257ea | 348 | return; |
0f0257ea | 349 | |
0f0257ea TT |
350 | /* Mark transport as busy. It will remain in this state until |
351 | * the provider calls svc_xprt_received. We update XPT_BUSY | |
352 | * atomically because it also guards against trying to enqueue | |
353 | * the transport twice. | |
354 | */ | |
355 | if (test_and_set_bit(XPT_BUSY, &xprt->xpt_flags)) { | |
356 | /* Don't enqueue transport while already enqueued */ | |
357 | dprintk("svc: transport %p busy, not enqueued\n", xprt); | |
0c0746d0 | 358 | return; |
0f0257ea | 359 | } |
0f0257ea | 360 | |
0c0746d0 TM |
361 | cpu = get_cpu(); |
362 | pool = svc_pool_for_cpu(xprt->xpt_server, cpu); | |
363 | spin_lock_bh(&pool->sp_lock); | |
364 | ||
365 | pool->sp_stats.packets++; | |
366 | ||
78c210ef | 367 | if (!list_empty(&pool->sp_threads)) { |
0f0257ea TT |
368 | rqstp = list_entry(pool->sp_threads.next, |
369 | struct svc_rqst, | |
370 | rq_list); | |
371 | dprintk("svc: transport %p served by daemon %p\n", | |
372 | xprt, rqstp); | |
373 | svc_thread_dequeue(pool, rqstp); | |
374 | if (rqstp->rq_xprt) | |
375 | printk(KERN_ERR | |
376 | "svc_xprt_enqueue: server %p, rq_xprt=%p!\n", | |
377 | rqstp, rqstp->rq_xprt); | |
983c6844 TM |
378 | /* Note the order of the following 3 lines: |
379 | * We want to assign xprt to rqstp->rq_xprt only _after_ | |
380 | * we've woken up the process, so that we don't race with | |
381 | * the lockless check in svc_get_next_xprt(). | |
382 | */ | |
0f0257ea | 383 | svc_xprt_get(xprt); |
983c6844 TM |
384 | wake_up_process(rqstp->rq_task); |
385 | rqstp->rq_xprt = xprt; | |
03cf6c9f | 386 | pool->sp_stats.threads_woken++; |
0f0257ea TT |
387 | } else { |
388 | dprintk("svc: transport %p put into queue\n", xprt); | |
389 | list_add_tail(&xprt->xpt_ready, &pool->sp_sockets); | |
03cf6c9f | 390 | pool->sp_stats.sockets_queued++; |
0f0257ea TT |
391 | } |
392 | ||
0f0257ea | 393 | spin_unlock_bh(&pool->sp_lock); |
983c6844 | 394 | put_cpu(); |
0f0257ea | 395 | } |
0971374e TM |
396 | |
397 | /* | |
398 | * Queue up a transport with data pending. If there are idle nfsd | |
399 | * processes, wake 'em up. | |
400 | * | |
401 | */ | |
402 | void svc_xprt_enqueue(struct svc_xprt *xprt) | |
403 | { | |
404 | if (test_bit(XPT_BUSY, &xprt->xpt_flags)) | |
405 | return; | |
406 | svc_xprt_do_enqueue(xprt); | |
407 | } | |
0f0257ea TT |
408 | EXPORT_SYMBOL_GPL(svc_xprt_enqueue); |
409 | ||
410 | /* | |
411 | * Dequeue the first transport. Must be called with the pool->sp_lock held. | |
412 | */ | |
413 | static struct svc_xprt *svc_xprt_dequeue(struct svc_pool *pool) | |
414 | { | |
415 | struct svc_xprt *xprt; | |
416 | ||
417 | if (list_empty(&pool->sp_sockets)) | |
418 | return NULL; | |
419 | ||
420 | xprt = list_entry(pool->sp_sockets.next, | |
421 | struct svc_xprt, xpt_ready); | |
422 | list_del_init(&xprt->xpt_ready); | |
423 | ||
424 | dprintk("svc: transport %p dequeued, inuse=%d\n", | |
425 | xprt, atomic_read(&xprt->xpt_ref.refcount)); | |
426 | ||
427 | return xprt; | |
428 | } | |
429 | ||
0f0257ea TT |
430 | /** |
431 | * svc_reserve - change the space reserved for the reply to a request. | |
432 | * @rqstp: The request in question | |
433 | * @space: new max space to reserve | |
434 | * | |
435 | * Each request reserves some space on the output queue of the transport | |
436 | * to make sure the reply fits. This function reduces that reserved | |
437 | * space to be the amount of space used already, plus @space. | |
438 | * | |
439 | */ | |
440 | void svc_reserve(struct svc_rqst *rqstp, int space) | |
441 | { | |
442 | space += rqstp->rq_res.head[0].iov_len; | |
443 | ||
444 | if (space < rqstp->rq_reserved) { | |
445 | struct svc_xprt *xprt = rqstp->rq_xprt; | |
446 | atomic_sub((rqstp->rq_reserved - space), &xprt->xpt_reserved); | |
447 | rqstp->rq_reserved = space; | |
448 | ||
51877680 TM |
449 | if (xprt->xpt_ops->xpo_adjust_wspace) |
450 | xprt->xpt_ops->xpo_adjust_wspace(xprt); | |
0f0257ea TT |
451 | svc_xprt_enqueue(xprt); |
452 | } | |
453 | } | |
24c3767e | 454 | EXPORT_SYMBOL_GPL(svc_reserve); |
0f0257ea TT |
455 | |
456 | static void svc_xprt_release(struct svc_rqst *rqstp) | |
457 | { | |
458 | struct svc_xprt *xprt = rqstp->rq_xprt; | |
459 | ||
460 | rqstp->rq_xprt->xpt_ops->xpo_release_rqst(rqstp); | |
461 | ||
2779e3ae TT |
462 | kfree(rqstp->rq_deferred); |
463 | rqstp->rq_deferred = NULL; | |
464 | ||
0f0257ea TT |
465 | svc_free_res_pages(rqstp); |
466 | rqstp->rq_res.page_len = 0; | |
467 | rqstp->rq_res.page_base = 0; | |
468 | ||
469 | /* Reset response buffer and release | |
470 | * the reservation. | |
471 | * But first, check that enough space was reserved | |
472 | * for the reply, otherwise we have a bug! | |
473 | */ | |
474 | if ((rqstp->rq_res.len) > rqstp->rq_reserved) | |
475 | printk(KERN_ERR "RPC request reserved %d but used %d\n", | |
476 | rqstp->rq_reserved, | |
477 | rqstp->rq_res.len); | |
478 | ||
479 | rqstp->rq_res.head[0].iov_len = 0; | |
480 | svc_reserve(rqstp, 0); | |
481 | rqstp->rq_xprt = NULL; | |
482 | ||
483 | svc_xprt_put(xprt); | |
484 | } | |
485 | ||
486 | /* | |
487 | * External function to wake up a server waiting for data | |
488 | * This really only makes sense for services like lockd | |
489 | * which have exactly one thread anyway. | |
490 | */ | |
491 | void svc_wake_up(struct svc_serv *serv) | |
492 | { | |
493 | struct svc_rqst *rqstp; | |
494 | unsigned int i; | |
495 | struct svc_pool *pool; | |
496 | ||
497 | for (i = 0; i < serv->sv_nrpools; i++) { | |
498 | pool = &serv->sv_pools[i]; | |
499 | ||
500 | spin_lock_bh(&pool->sp_lock); | |
501 | if (!list_empty(&pool->sp_threads)) { | |
502 | rqstp = list_entry(pool->sp_threads.next, | |
503 | struct svc_rqst, | |
504 | rq_list); | |
505 | dprintk("svc: daemon %p woken up.\n", rqstp); | |
506 | /* | |
507 | svc_thread_dequeue(pool, rqstp); | |
508 | rqstp->rq_xprt = NULL; | |
509 | */ | |
983c6844 | 510 | wake_up_process(rqstp->rq_task); |
35525b79 AS |
511 | } else |
512 | pool->sp_task_pending = 1; | |
0f0257ea TT |
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)) { | |
e87cc472 JP |
559 | /* Try to help the admin */ |
560 | net_notice_ratelimited("%s: too many open connections, consider increasing the %s\n", | |
561 | serv->sv_name, serv->sv_maxconn ? | |
562 | "max number of connections" : | |
563 | "number of threads"); | |
0f0257ea TT |
564 | /* |
565 | * Always select the oldest connection. It's not fair, | |
566 | * but so is life | |
567 | */ | |
568 | xprt = list_entry(serv->sv_tempsocks.prev, | |
569 | struct svc_xprt, | |
570 | xpt_list); | |
571 | set_bit(XPT_CLOSE, &xprt->xpt_flags); | |
572 | svc_xprt_get(xprt); | |
573 | } | |
574 | spin_unlock_bh(&serv->sv_lock); | |
575 | ||
576 | if (xprt) { | |
577 | svc_xprt_enqueue(xprt); | |
578 | svc_xprt_put(xprt); | |
579 | } | |
580 | } | |
581 | } | |
582 | ||
e1d83ee6 | 583 | static int svc_alloc_arg(struct svc_rqst *rqstp) |
0f0257ea | 584 | { |
6797fa5a BF |
585 | struct svc_serv *serv = rqstp->rq_server; |
586 | struct xdr_buf *arg; | |
587 | int pages; | |
588 | int i; | |
0f0257ea TT |
589 | |
590 | /* now allocate needed pages. If we get a failure, sleep briefly */ | |
591 | pages = (serv->sv_max_mesg + PAGE_SIZE) / PAGE_SIZE; | |
b25cd058 WAA |
592 | WARN_ON_ONCE(pages >= RPCSVC_MAXPAGES); |
593 | if (pages >= RPCSVC_MAXPAGES) | |
594 | /* use as many pages as possible */ | |
595 | pages = RPCSVC_MAXPAGES - 1; | |
0f0257ea TT |
596 | for (i = 0; i < pages ; i++) |
597 | while (rqstp->rq_pages[i] == NULL) { | |
598 | struct page *p = alloc_page(GFP_KERNEL); | |
599 | if (!p) { | |
7b54fe61 JL |
600 | set_current_state(TASK_INTERRUPTIBLE); |
601 | if (signalled() || kthread_should_stop()) { | |
602 | set_current_state(TASK_RUNNING); | |
7086721f | 603 | return -EINTR; |
7b54fe61 JL |
604 | } |
605 | schedule_timeout(msecs_to_jiffies(500)); | |
0f0257ea TT |
606 | } |
607 | rqstp->rq_pages[i] = p; | |
608 | } | |
2825a7f9 | 609 | rqstp->rq_page_end = &rqstp->rq_pages[i]; |
0f0257ea | 610 | rqstp->rq_pages[i++] = NULL; /* this might be seen in nfs_read_actor */ |
0f0257ea TT |
611 | |
612 | /* Make arg->head point to first page and arg->pages point to rest */ | |
613 | arg = &rqstp->rq_arg; | |
614 | arg->head[0].iov_base = page_address(rqstp->rq_pages[0]); | |
615 | arg->head[0].iov_len = PAGE_SIZE; | |
616 | arg->pages = rqstp->rq_pages + 1; | |
617 | arg->page_base = 0; | |
618 | /* save at least one page for response */ | |
619 | arg->page_len = (pages-2)*PAGE_SIZE; | |
620 | arg->len = (pages-1)*PAGE_SIZE; | |
621 | arg->tail[0].iov_len = 0; | |
6797fa5a BF |
622 | return 0; |
623 | } | |
0f0257ea | 624 | |
e1d83ee6 | 625 | static struct svc_xprt *svc_get_next_xprt(struct svc_rqst *rqstp, long timeout) |
6797fa5a BF |
626 | { |
627 | struct svc_xprt *xprt; | |
628 | struct svc_pool *pool = rqstp->rq_pool; | |
a4aa8054 | 629 | long time_left = 0; |
0f0257ea | 630 | |
f16b6e8d N |
631 | /* Normally we will wait up to 5 seconds for any required |
632 | * cache information to be provided. | |
633 | */ | |
634 | rqstp->rq_chandle.thread_wait = 5*HZ; | |
635 | ||
0f0257ea TT |
636 | spin_lock_bh(&pool->sp_lock); |
637 | xprt = svc_xprt_dequeue(pool); | |
638 | if (xprt) { | |
639 | rqstp->rq_xprt = xprt; | |
640 | svc_xprt_get(xprt); | |
f16b6e8d N |
641 | |
642 | /* As there is a shortage of threads and this request | |
6610f720 | 643 | * had to be queued, don't allow the thread to wait so |
f16b6e8d N |
644 | * long for cache updates. |
645 | */ | |
646 | rqstp->rq_chandle.thread_wait = 1*HZ; | |
35525b79 | 647 | pool->sp_task_pending = 0; |
0f0257ea | 648 | } else { |
35525b79 AS |
649 | if (pool->sp_task_pending) { |
650 | pool->sp_task_pending = 0; | |
106f359c TM |
651 | xprt = ERR_PTR(-EAGAIN); |
652 | goto out; | |
35525b79 | 653 | } |
0f0257ea TT |
654 | /* |
655 | * We have to be able to interrupt this wait | |
656 | * to bring down the daemons ... | |
657 | */ | |
658 | set_current_state(TASK_INTERRUPTIBLE); | |
7086721f | 659 | |
983c6844 TM |
660 | /* No data pending. Go to sleep */ |
661 | svc_thread_enqueue(pool, rqstp); | |
0f0257ea TT |
662 | spin_unlock_bh(&pool->sp_lock); |
663 | ||
a4aa8054 TM |
664 | if (!(signalled() || kthread_should_stop())) { |
665 | time_left = schedule_timeout(timeout); | |
666 | __set_current_state(TASK_RUNNING); | |
0f0257ea | 667 | |
a4aa8054 | 668 | try_to_freeze(); |
0f0257ea | 669 | |
a4aa8054 TM |
670 | xprt = rqstp->rq_xprt; |
671 | if (xprt != NULL) | |
672 | return xprt; | |
673 | } else | |
674 | __set_current_state(TASK_RUNNING); | |
106f359c TM |
675 | |
676 | spin_lock_bh(&pool->sp_lock); | |
03cf6c9f GB |
677 | if (!time_left) |
678 | pool->sp_stats.threads_timedout++; | |
0f0257ea TT |
679 | |
680 | xprt = rqstp->rq_xprt; | |
681 | if (!xprt) { | |
682 | svc_thread_dequeue(pool, rqstp); | |
683 | spin_unlock_bh(&pool->sp_lock); | |
684 | dprintk("svc: server %p, no data yet\n", rqstp); | |
7086721f | 685 | if (signalled() || kthread_should_stop()) |
6797fa5a | 686 | return ERR_PTR(-EINTR); |
7086721f | 687 | else |
6797fa5a | 688 | return ERR_PTR(-EAGAIN); |
0f0257ea TT |
689 | } |
690 | } | |
106f359c | 691 | out: |
0f0257ea | 692 | spin_unlock_bh(&pool->sp_lock); |
6797fa5a BF |
693 | return xprt; |
694 | } | |
695 | ||
e1d83ee6 | 696 | static void svc_add_new_temp_xprt(struct svc_serv *serv, struct svc_xprt *newxpt) |
65b2e665 BF |
697 | { |
698 | spin_lock_bh(&serv->sv_lock); | |
699 | set_bit(XPT_TEMP, &newxpt->xpt_flags); | |
700 | list_add(&newxpt->xpt_list, &serv->sv_tempsocks); | |
701 | serv->sv_tmpcnt++; | |
702 | if (serv->sv_temptimer.function == NULL) { | |
703 | /* setup timer to age temp transports */ | |
704 | setup_timer(&serv->sv_temptimer, svc_age_temp_xprts, | |
705 | (unsigned long)serv); | |
706 | mod_timer(&serv->sv_temptimer, | |
707 | jiffies + svc_conn_age_period * HZ); | |
708 | } | |
709 | spin_unlock_bh(&serv->sv_lock); | |
710 | svc_xprt_received(newxpt); | |
711 | } | |
712 | ||
6797fa5a BF |
713 | static int svc_handle_xprt(struct svc_rqst *rqstp, struct svc_xprt *xprt) |
714 | { | |
715 | struct svc_serv *serv = rqstp->rq_server; | |
716 | int len = 0; | |
0f0257ea | 717 | |
1b644b6e BF |
718 | if (test_bit(XPT_CLOSE, &xprt->xpt_flags)) { |
719 | dprintk("svc_recv: found XPT_CLOSE\n"); | |
720 | svc_delete_xprt(xprt); | |
ca7896cd | 721 | /* Leave XPT_BUSY set on the dead xprt: */ |
6797fa5a | 722 | return 0; |
ca7896cd BF |
723 | } |
724 | if (test_bit(XPT_LISTENER, &xprt->xpt_flags)) { | |
0f0257ea | 725 | struct svc_xprt *newxpt; |
65b2e665 BF |
726 | /* |
727 | * We know this module_get will succeed because the | |
728 | * listener holds a reference too | |
729 | */ | |
730 | __module_get(xprt->xpt_class->xcl_owner); | |
731 | svc_check_conn_limits(xprt->xpt_server); | |
0f0257ea | 732 | newxpt = xprt->xpt_ops->xpo_accept(xprt); |
65b2e665 BF |
733 | if (newxpt) |
734 | svc_add_new_temp_xprt(serv, newxpt); | |
c789102c TM |
735 | else |
736 | module_put(xprt->xpt_class->xcl_owner); | |
9e5b208d | 737 | } else { |
6797fa5a | 738 | /* XPT_DATA|XPT_DEFERRED case: */ |
0f0257ea | 739 | dprintk("svc: server %p, pool %u, transport %p, inuse=%d\n", |
6797fa5a | 740 | rqstp, rqstp->rq_pool->sp_id, xprt, |
0f0257ea TT |
741 | atomic_read(&xprt->xpt_ref.refcount)); |
742 | rqstp->rq_deferred = svc_deferred_dequeue(xprt); | |
ca7896cd | 743 | if (rqstp->rq_deferred) |
0f0257ea | 744 | len = svc_deferred_recv(rqstp); |
ca7896cd | 745 | else |
0f0257ea TT |
746 | len = xprt->xpt_ops->xpo_recvfrom(rqstp); |
747 | dprintk("svc: got len=%d\n", len); | |
d10f27a7 BF |
748 | rqstp->rq_reserved = serv->sv_max_mesg; |
749 | atomic_add(rqstp->rq_reserved, &xprt->xpt_reserved); | |
0f0257ea | 750 | } |
6797fa5a | 751 | /* clear XPT_BUSY: */ |
ca7896cd | 752 | svc_xprt_received(xprt); |
6797fa5a BF |
753 | return len; |
754 | } | |
755 | ||
756 | /* | |
757 | * Receive the next request on any transport. This code is carefully | |
758 | * organised not to touch any cachelines in the shared svc_serv | |
759 | * structure, only cachelines in the local svc_pool. | |
760 | */ | |
761 | int svc_recv(struct svc_rqst *rqstp, long timeout) | |
762 | { | |
763 | struct svc_xprt *xprt = NULL; | |
764 | struct svc_serv *serv = rqstp->rq_server; | |
765 | int len, err; | |
766 | ||
767 | dprintk("svc: server %p waiting for data (to = %ld)\n", | |
768 | rqstp, timeout); | |
769 | ||
770 | if (rqstp->rq_xprt) | |
771 | printk(KERN_ERR | |
772 | "svc_recv: service %p, transport not NULL!\n", | |
773 | rqstp); | |
983c6844 | 774 | |
6797fa5a BF |
775 | err = svc_alloc_arg(rqstp); |
776 | if (err) | |
860a0d9e | 777 | goto out; |
6797fa5a BF |
778 | |
779 | try_to_freeze(); | |
780 | cond_resched(); | |
860a0d9e | 781 | err = -EINTR; |
6797fa5a | 782 | if (signalled() || kthread_should_stop()) |
860a0d9e | 783 | goto out; |
6797fa5a BF |
784 | |
785 | xprt = svc_get_next_xprt(rqstp, timeout); | |
860a0d9e JL |
786 | if (IS_ERR(xprt)) { |
787 | err = PTR_ERR(xprt); | |
788 | goto out; | |
789 | } | |
6797fa5a BF |
790 | |
791 | len = svc_handle_xprt(rqstp, xprt); | |
0f0257ea TT |
792 | |
793 | /* No data, incomplete (TCP) read, or accept() */ | |
860a0d9e | 794 | err = -EAGAIN; |
9f9d2ebe | 795 | if (len <= 0) |
860a0d9e | 796 | goto out_release; |
ca7896cd | 797 | |
0f0257ea TT |
798 | clear_bit(XPT_OLD, &xprt->xpt_flags); |
799 | ||
4d152e2c JL |
800 | if (xprt->xpt_ops->xpo_secure_port(rqstp)) |
801 | set_bit(RQ_SECURE, &rqstp->rq_flags); | |
802 | else | |
803 | clear_bit(RQ_SECURE, &rqstp->rq_flags); | |
0f0257ea | 804 | rqstp->rq_chandle.defer = svc_defer; |
860a0d9e | 805 | rqstp->rq_xid = svc_getu32(&rqstp->rq_arg.head[0]); |
0f0257ea TT |
806 | |
807 | if (serv->sv_stats) | |
808 | serv->sv_stats->netcnt++; | |
860a0d9e | 809 | trace_svc_recv(rqstp, len); |
0f0257ea | 810 | return len; |
860a0d9e | 811 | out_release: |
ca7896cd BF |
812 | rqstp->rq_res.len = 0; |
813 | svc_xprt_release(rqstp); | |
860a0d9e JL |
814 | out: |
815 | trace_svc_recv(rqstp, err); | |
816 | return err; | |
0f0257ea | 817 | } |
24c3767e | 818 | EXPORT_SYMBOL_GPL(svc_recv); |
0f0257ea TT |
819 | |
820 | /* | |
821 | * Drop request | |
822 | */ | |
823 | void svc_drop(struct svc_rqst *rqstp) | |
824 | { | |
825 | dprintk("svc: xprt %p dropped request\n", rqstp->rq_xprt); | |
826 | svc_xprt_release(rqstp); | |
827 | } | |
24c3767e | 828 | EXPORT_SYMBOL_GPL(svc_drop); |
0f0257ea TT |
829 | |
830 | /* | |
831 | * Return reply to client. | |
832 | */ | |
833 | int svc_send(struct svc_rqst *rqstp) | |
834 | { | |
835 | struct svc_xprt *xprt; | |
860a0d9e | 836 | int len = -EFAULT; |
0f0257ea TT |
837 | struct xdr_buf *xb; |
838 | ||
839 | xprt = rqstp->rq_xprt; | |
840 | if (!xprt) | |
860a0d9e | 841 | goto out; |
0f0257ea TT |
842 | |
843 | /* release the receive skb before sending the reply */ | |
844 | rqstp->rq_xprt->xpt_ops->xpo_release_rqst(rqstp); | |
845 | ||
846 | /* calculate over-all length */ | |
847 | xb = &rqstp->rq_res; | |
848 | xb->len = xb->head[0].iov_len + | |
849 | xb->page_len + | |
850 | xb->tail[0].iov_len; | |
851 | ||
852 | /* Grab mutex to serialize outgoing data. */ | |
853 | mutex_lock(&xprt->xpt_mutex); | |
f06f00a2 BF |
854 | if (test_bit(XPT_DEAD, &xprt->xpt_flags) |
855 | || test_bit(XPT_CLOSE, &xprt->xpt_flags)) | |
0f0257ea TT |
856 | len = -ENOTCONN; |
857 | else | |
858 | len = xprt->xpt_ops->xpo_sendto(rqstp); | |
859 | mutex_unlock(&xprt->xpt_mutex); | |
4cfc7e60 | 860 | rpc_wake_up(&xprt->xpt_bc_pending); |
0f0257ea TT |
861 | svc_xprt_release(rqstp); |
862 | ||
863 | if (len == -ECONNREFUSED || len == -ENOTCONN || len == -EAGAIN) | |
860a0d9e JL |
864 | len = 0; |
865 | out: | |
866 | trace_svc_send(rqstp, len); | |
0f0257ea TT |
867 | return len; |
868 | } | |
869 | ||
870 | /* | |
871 | * Timer function to close old temporary transports, using | |
872 | * a mark-and-sweep algorithm. | |
873 | */ | |
874 | static void svc_age_temp_xprts(unsigned long closure) | |
875 | { | |
876 | struct svc_serv *serv = (struct svc_serv *)closure; | |
877 | struct svc_xprt *xprt; | |
878 | struct list_head *le, *next; | |
0f0257ea TT |
879 | |
880 | dprintk("svc_age_temp_xprts\n"); | |
881 | ||
882 | if (!spin_trylock_bh(&serv->sv_lock)) { | |
883 | /* busy, try again 1 sec later */ | |
884 | dprintk("svc_age_temp_xprts: busy\n"); | |
885 | mod_timer(&serv->sv_temptimer, jiffies + HZ); | |
886 | return; | |
887 | } | |
888 | ||
889 | list_for_each_safe(le, next, &serv->sv_tempsocks) { | |
890 | xprt = list_entry(le, struct svc_xprt, xpt_list); | |
891 | ||
892 | /* First time through, just mark it OLD. Second time | |
893 | * through, close it. */ | |
894 | if (!test_and_set_bit(XPT_OLD, &xprt->xpt_flags)) | |
895 | continue; | |
f64f9e71 JP |
896 | if (atomic_read(&xprt->xpt_ref.refcount) > 1 || |
897 | test_bit(XPT_BUSY, &xprt->xpt_flags)) | |
0f0257ea | 898 | continue; |
e75bafbf | 899 | list_del_init(le); |
0f0257ea | 900 | set_bit(XPT_CLOSE, &xprt->xpt_flags); |
0f0257ea TT |
901 | dprintk("queuing xprt %p for closing\n", xprt); |
902 | ||
903 | /* a thread will dequeue and close it soon */ | |
904 | svc_xprt_enqueue(xprt); | |
0f0257ea | 905 | } |
e75bafbf | 906 | spin_unlock_bh(&serv->sv_lock); |
0f0257ea TT |
907 | |
908 | mod_timer(&serv->sv_temptimer, jiffies + svc_conn_age_period * HZ); | |
909 | } | |
910 | ||
edc7a894 BF |
911 | static void call_xpt_users(struct svc_xprt *xprt) |
912 | { | |
913 | struct svc_xpt_user *u; | |
914 | ||
915 | spin_lock(&xprt->xpt_lock); | |
916 | while (!list_empty(&xprt->xpt_users)) { | |
917 | u = list_first_entry(&xprt->xpt_users, struct svc_xpt_user, list); | |
918 | list_del(&u->list); | |
919 | u->callback(u); | |
920 | } | |
921 | spin_unlock(&xprt->xpt_lock); | |
922 | } | |
923 | ||
0f0257ea TT |
924 | /* |
925 | * Remove a dead transport | |
926 | */ | |
7710ec36 | 927 | static void svc_delete_xprt(struct svc_xprt *xprt) |
0f0257ea TT |
928 | { |
929 | struct svc_serv *serv = xprt->xpt_server; | |
22945e4a TT |
930 | struct svc_deferred_req *dr; |
931 | ||
932 | /* Only do this once */ | |
933 | if (test_and_set_bit(XPT_DEAD, &xprt->xpt_flags)) | |
ac9303eb | 934 | BUG(); |
0f0257ea TT |
935 | |
936 | dprintk("svc: svc_delete_xprt(%p)\n", xprt); | |
937 | xprt->xpt_ops->xpo_detach(xprt); | |
938 | ||
939 | spin_lock_bh(&serv->sv_lock); | |
8d65ef76 | 940 | list_del_init(&xprt->xpt_list); |
01047298 | 941 | WARN_ON_ONCE(!list_empty(&xprt->xpt_ready)); |
22945e4a TT |
942 | if (test_bit(XPT_TEMP, &xprt->xpt_flags)) |
943 | serv->sv_tmpcnt--; | |
788e69e5 | 944 | spin_unlock_bh(&serv->sv_lock); |
22945e4a | 945 | |
ab1b18f7 | 946 | while ((dr = svc_deferred_dequeue(xprt)) != NULL) |
22945e4a | 947 | kfree(dr); |
22945e4a | 948 | |
edc7a894 | 949 | call_xpt_users(xprt); |
22945e4a | 950 | svc_xprt_put(xprt); |
0f0257ea TT |
951 | } |
952 | ||
953 | void svc_close_xprt(struct svc_xprt *xprt) | |
954 | { | |
955 | set_bit(XPT_CLOSE, &xprt->xpt_flags); | |
956 | if (test_and_set_bit(XPT_BUSY, &xprt->xpt_flags)) | |
957 | /* someone else will have to effect the close */ | |
958 | return; | |
b1763316 BF |
959 | /* |
960 | * We expect svc_close_xprt() to work even when no threads are | |
961 | * running (e.g., while configuring the server before starting | |
962 | * any threads), so if the transport isn't busy, we delete | |
963 | * it ourself: | |
964 | */ | |
0f0257ea | 965 | svc_delete_xprt(xprt); |
0f0257ea | 966 | } |
a217813f | 967 | EXPORT_SYMBOL_GPL(svc_close_xprt); |
0f0257ea | 968 | |
cc630d9f | 969 | static int svc_close_list(struct svc_serv *serv, struct list_head *xprt_list, struct net *net) |
0f0257ea TT |
970 | { |
971 | struct svc_xprt *xprt; | |
cc630d9f | 972 | int ret = 0; |
0f0257ea | 973 | |
719f8bcc | 974 | spin_lock(&serv->sv_lock); |
b4f36f88 | 975 | list_for_each_entry(xprt, xprt_list, xpt_list) { |
7b147f1f SK |
976 | if (xprt->xpt_net != net) |
977 | continue; | |
cc630d9f | 978 | ret++; |
0f0257ea | 979 | set_bit(XPT_CLOSE, &xprt->xpt_flags); |
cc630d9f | 980 | svc_xprt_enqueue(xprt); |
0f0257ea | 981 | } |
719f8bcc | 982 | spin_unlock(&serv->sv_lock); |
cc630d9f | 983 | return ret; |
0f0257ea TT |
984 | } |
985 | ||
cc630d9f | 986 | static struct svc_xprt *svc_dequeue_net(struct svc_serv *serv, struct net *net) |
0f0257ea | 987 | { |
b4f36f88 | 988 | struct svc_pool *pool; |
0f0257ea TT |
989 | struct svc_xprt *xprt; |
990 | struct svc_xprt *tmp; | |
b4f36f88 BF |
991 | int i; |
992 | ||
b4f36f88 BF |
993 | for (i = 0; i < serv->sv_nrpools; i++) { |
994 | pool = &serv->sv_pools[i]; | |
995 | ||
996 | spin_lock_bh(&pool->sp_lock); | |
6f513365 | 997 | list_for_each_entry_safe(xprt, tmp, &pool->sp_sockets, xpt_ready) { |
7b147f1f SK |
998 | if (xprt->xpt_net != net) |
999 | continue; | |
b4f36f88 | 1000 | list_del_init(&xprt->xpt_ready); |
cc630d9f BF |
1001 | spin_unlock_bh(&pool->sp_lock); |
1002 | return xprt; | |
b4f36f88 BF |
1003 | } |
1004 | spin_unlock_bh(&pool->sp_lock); | |
1005 | } | |
cc630d9f | 1006 | return NULL; |
6f513365 SK |
1007 | } |
1008 | ||
cc630d9f | 1009 | static void svc_clean_up_xprts(struct svc_serv *serv, struct net *net) |
6f513365 SK |
1010 | { |
1011 | struct svc_xprt *xprt; | |
719f8bcc | 1012 | |
cc630d9f BF |
1013 | while ((xprt = svc_dequeue_net(serv, net))) { |
1014 | set_bit(XPT_CLOSE, &xprt->xpt_flags); | |
719f8bcc | 1015 | svc_delete_xprt(xprt); |
cc630d9f | 1016 | } |
3a22bf50 SK |
1017 | } |
1018 | ||
cc630d9f BF |
1019 | /* |
1020 | * Server threads may still be running (especially in the case where the | |
1021 | * service is still running in other network namespaces). | |
1022 | * | |
1023 | * So we shut down sockets the same way we would on a running server, by | |
1024 | * setting XPT_CLOSE, enqueuing, and letting a thread pick it up to do | |
1025 | * the close. In the case there are no such other threads, | |
1026 | * threads running, svc_clean_up_xprts() does a simple version of a | |
1027 | * server's main event loop, and in the case where there are other | |
1028 | * threads, we may need to wait a little while and then check again to | |
1029 | * see if they're done. | |
1030 | */ | |
7b147f1f | 1031 | void svc_close_net(struct svc_serv *serv, struct net *net) |
3a22bf50 | 1032 | { |
cc630d9f | 1033 | int delay = 0; |
6f513365 | 1034 | |
cc630d9f BF |
1035 | while (svc_close_list(serv, &serv->sv_permsocks, net) + |
1036 | svc_close_list(serv, &serv->sv_tempsocks, net)) { | |
1037 | ||
1038 | svc_clean_up_xprts(serv, net); | |
1039 | msleep(delay++); | |
1040 | } | |
0f0257ea TT |
1041 | } |
1042 | ||
1043 | /* | |
1044 | * Handle defer and revisit of requests | |
1045 | */ | |
1046 | ||
1047 | static void svc_revisit(struct cache_deferred_req *dreq, int too_many) | |
1048 | { | |
1049 | struct svc_deferred_req *dr = | |
1050 | container_of(dreq, struct svc_deferred_req, handle); | |
1051 | struct svc_xprt *xprt = dr->xprt; | |
1052 | ||
22945e4a TT |
1053 | spin_lock(&xprt->xpt_lock); |
1054 | set_bit(XPT_DEFERRED, &xprt->xpt_flags); | |
1055 | if (too_many || test_bit(XPT_DEAD, &xprt->xpt_flags)) { | |
1056 | spin_unlock(&xprt->xpt_lock); | |
1057 | dprintk("revisit canceled\n"); | |
0f0257ea TT |
1058 | svc_xprt_put(xprt); |
1059 | kfree(dr); | |
1060 | return; | |
1061 | } | |
1062 | dprintk("revisit queued\n"); | |
1063 | dr->xprt = NULL; | |
0f0257ea TT |
1064 | list_add(&dr->handle.recent, &xprt->xpt_deferred); |
1065 | spin_unlock(&xprt->xpt_lock); | |
0f0257ea TT |
1066 | svc_xprt_enqueue(xprt); |
1067 | svc_xprt_put(xprt); | |
1068 | } | |
1069 | ||
260c1d12 TT |
1070 | /* |
1071 | * Save the request off for later processing. The request buffer looks | |
1072 | * like this: | |
1073 | * | |
1074 | * <xprt-header><rpc-header><rpc-pagelist><rpc-tail> | |
1075 | * | |
1076 | * This code can only handle requests that consist of an xprt-header | |
1077 | * and rpc-header. | |
1078 | */ | |
0f0257ea TT |
1079 | static struct cache_deferred_req *svc_defer(struct cache_req *req) |
1080 | { | |
1081 | struct svc_rqst *rqstp = container_of(req, struct svc_rqst, rq_chandle); | |
0f0257ea TT |
1082 | struct svc_deferred_req *dr; |
1083 | ||
2f425878 | 1084 | if (rqstp->rq_arg.page_len || !rqstp->rq_usedeferral) |
0f0257ea TT |
1085 | return NULL; /* if more than a page, give up FIXME */ |
1086 | if (rqstp->rq_deferred) { | |
1087 | dr = rqstp->rq_deferred; | |
1088 | rqstp->rq_deferred = NULL; | |
1089 | } else { | |
260c1d12 TT |
1090 | size_t skip; |
1091 | size_t size; | |
0f0257ea | 1092 | /* FIXME maybe discard if size too large */ |
260c1d12 | 1093 | size = sizeof(struct svc_deferred_req) + rqstp->rq_arg.len; |
0f0257ea TT |
1094 | dr = kmalloc(size, GFP_KERNEL); |
1095 | if (dr == NULL) | |
1096 | return NULL; | |
1097 | ||
1098 | dr->handle.owner = rqstp->rq_server; | |
1099 | dr->prot = rqstp->rq_prot; | |
1100 | memcpy(&dr->addr, &rqstp->rq_addr, rqstp->rq_addrlen); | |
1101 | dr->addrlen = rqstp->rq_addrlen; | |
1102 | dr->daddr = rqstp->rq_daddr; | |
1103 | dr->argslen = rqstp->rq_arg.len >> 2; | |
260c1d12 TT |
1104 | dr->xprt_hlen = rqstp->rq_xprt_hlen; |
1105 | ||
1106 | /* back up head to the start of the buffer and copy */ | |
1107 | skip = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len; | |
1108 | memcpy(dr->args, rqstp->rq_arg.head[0].iov_base - skip, | |
1109 | dr->argslen << 2); | |
0f0257ea TT |
1110 | } |
1111 | svc_xprt_get(rqstp->rq_xprt); | |
1112 | dr->xprt = rqstp->rq_xprt; | |
9e701c61 | 1113 | rqstp->rq_dropme = true; |
0f0257ea TT |
1114 | |
1115 | dr->handle.revisit = svc_revisit; | |
1116 | return &dr->handle; | |
1117 | } | |
1118 | ||
1119 | /* | |
1120 | * recv data from a deferred request into an active one | |
1121 | */ | |
1122 | static int svc_deferred_recv(struct svc_rqst *rqstp) | |
1123 | { | |
1124 | struct svc_deferred_req *dr = rqstp->rq_deferred; | |
1125 | ||
260c1d12 TT |
1126 | /* setup iov_base past transport header */ |
1127 | rqstp->rq_arg.head[0].iov_base = dr->args + (dr->xprt_hlen>>2); | |
1128 | /* The iov_len does not include the transport header bytes */ | |
1129 | rqstp->rq_arg.head[0].iov_len = (dr->argslen<<2) - dr->xprt_hlen; | |
0f0257ea | 1130 | rqstp->rq_arg.page_len = 0; |
260c1d12 TT |
1131 | /* The rq_arg.len includes the transport header bytes */ |
1132 | rqstp->rq_arg.len = dr->argslen<<2; | |
0f0257ea TT |
1133 | rqstp->rq_prot = dr->prot; |
1134 | memcpy(&rqstp->rq_addr, &dr->addr, dr->addrlen); | |
1135 | rqstp->rq_addrlen = dr->addrlen; | |
260c1d12 TT |
1136 | /* Save off transport header len in case we get deferred again */ |
1137 | rqstp->rq_xprt_hlen = dr->xprt_hlen; | |
0f0257ea TT |
1138 | rqstp->rq_daddr = dr->daddr; |
1139 | rqstp->rq_respages = rqstp->rq_pages; | |
260c1d12 | 1140 | return (dr->argslen<<2) - dr->xprt_hlen; |
0f0257ea TT |
1141 | } |
1142 | ||
1143 | ||
1144 | static struct svc_deferred_req *svc_deferred_dequeue(struct svc_xprt *xprt) | |
1145 | { | |
1146 | struct svc_deferred_req *dr = NULL; | |
1147 | ||
1148 | if (!test_bit(XPT_DEFERRED, &xprt->xpt_flags)) | |
1149 | return NULL; | |
1150 | spin_lock(&xprt->xpt_lock); | |
0f0257ea TT |
1151 | if (!list_empty(&xprt->xpt_deferred)) { |
1152 | dr = list_entry(xprt->xpt_deferred.next, | |
1153 | struct svc_deferred_req, | |
1154 | handle.recent); | |
1155 | list_del_init(&dr->handle.recent); | |
62bac4af BF |
1156 | } else |
1157 | clear_bit(XPT_DEFERRED, &xprt->xpt_flags); | |
0f0257ea TT |
1158 | spin_unlock(&xprt->xpt_lock); |
1159 | return dr; | |
1160 | } | |
7fcb98d5 | 1161 | |
156e6209 CL |
1162 | /** |
1163 | * svc_find_xprt - find an RPC transport instance | |
1164 | * @serv: pointer to svc_serv to search | |
1165 | * @xcl_name: C string containing transport's class name | |
4cb54ca2 | 1166 | * @net: owner net pointer |
156e6209 CL |
1167 | * @af: Address family of transport's local address |
1168 | * @port: transport's IP port number | |
1169 | * | |
7fcb98d5 TT |
1170 | * Return the transport instance pointer for the endpoint accepting |
1171 | * connections/peer traffic from the specified transport class, | |
1172 | * address family and port. | |
1173 | * | |
1174 | * Specifying 0 for the address family or port is effectively a | |
1175 | * wild-card, and will result in matching the first transport in the | |
1176 | * service's list that has a matching class name. | |
1177 | */ | |
156e6209 | 1178 | struct svc_xprt *svc_find_xprt(struct svc_serv *serv, const char *xcl_name, |
4cb54ca2 SK |
1179 | struct net *net, const sa_family_t af, |
1180 | const unsigned short port) | |
7fcb98d5 TT |
1181 | { |
1182 | struct svc_xprt *xprt; | |
1183 | struct svc_xprt *found = NULL; | |
1184 | ||
1185 | /* Sanity check the args */ | |
156e6209 | 1186 | if (serv == NULL || xcl_name == NULL) |
7fcb98d5 TT |
1187 | return found; |
1188 | ||
1189 | spin_lock_bh(&serv->sv_lock); | |
1190 | list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list) { | |
4cb54ca2 SK |
1191 | if (xprt->xpt_net != net) |
1192 | continue; | |
7fcb98d5 TT |
1193 | if (strcmp(xprt->xpt_class->xcl_name, xcl_name)) |
1194 | continue; | |
1195 | if (af != AF_UNSPEC && af != xprt->xpt_local.ss_family) | |
1196 | continue; | |
156e6209 | 1197 | if (port != 0 && port != svc_xprt_local_port(xprt)) |
7fcb98d5 TT |
1198 | continue; |
1199 | found = xprt; | |
a217813f | 1200 | svc_xprt_get(xprt); |
7fcb98d5 TT |
1201 | break; |
1202 | } | |
1203 | spin_unlock_bh(&serv->sv_lock); | |
1204 | return found; | |
1205 | } | |
1206 | EXPORT_SYMBOL_GPL(svc_find_xprt); | |
9571af18 | 1207 | |
335c54bd CL |
1208 | static int svc_one_xprt_name(const struct svc_xprt *xprt, |
1209 | char *pos, int remaining) | |
1210 | { | |
1211 | int len; | |
1212 | ||
1213 | len = snprintf(pos, remaining, "%s %u\n", | |
1214 | xprt->xpt_class->xcl_name, | |
1215 | svc_xprt_local_port(xprt)); | |
1216 | if (len >= remaining) | |
1217 | return -ENAMETOOLONG; | |
1218 | return len; | |
1219 | } | |
1220 | ||
1221 | /** | |
1222 | * svc_xprt_names - format a buffer with a list of transport names | |
1223 | * @serv: pointer to an RPC service | |
1224 | * @buf: pointer to a buffer to be filled in | |
1225 | * @buflen: length of buffer to be filled in | |
1226 | * | |
1227 | * Fills in @buf with a string containing a list of transport names, | |
1228 | * each name terminated with '\n'. | |
1229 | * | |
1230 | * Returns positive length of the filled-in string on success; otherwise | |
1231 | * a negative errno value is returned if an error occurs. | |
9571af18 | 1232 | */ |
335c54bd | 1233 | int svc_xprt_names(struct svc_serv *serv, char *buf, const int buflen) |
9571af18 TT |
1234 | { |
1235 | struct svc_xprt *xprt; | |
335c54bd CL |
1236 | int len, totlen; |
1237 | char *pos; | |
9571af18 TT |
1238 | |
1239 | /* Sanity check args */ | |
1240 | if (!serv) | |
1241 | return 0; | |
1242 | ||
1243 | spin_lock_bh(&serv->sv_lock); | |
335c54bd CL |
1244 | |
1245 | pos = buf; | |
1246 | totlen = 0; | |
9571af18 | 1247 | list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list) { |
335c54bd CL |
1248 | len = svc_one_xprt_name(xprt, pos, buflen - totlen); |
1249 | if (len < 0) { | |
1250 | *buf = '\0'; | |
1251 | totlen = len; | |
1252 | } | |
1253 | if (len <= 0) | |
9571af18 | 1254 | break; |
335c54bd CL |
1255 | |
1256 | pos += len; | |
9571af18 TT |
1257 | totlen += len; |
1258 | } | |
335c54bd | 1259 | |
9571af18 TT |
1260 | spin_unlock_bh(&serv->sv_lock); |
1261 | return totlen; | |
1262 | } | |
1263 | EXPORT_SYMBOL_GPL(svc_xprt_names); | |
03cf6c9f GB |
1264 | |
1265 | ||
1266 | /*----------------------------------------------------------------------------*/ | |
1267 | ||
1268 | static void *svc_pool_stats_start(struct seq_file *m, loff_t *pos) | |
1269 | { | |
1270 | unsigned int pidx = (unsigned int)*pos; | |
1271 | struct svc_serv *serv = m->private; | |
1272 | ||
1273 | dprintk("svc_pool_stats_start, *pidx=%u\n", pidx); | |
1274 | ||
03cf6c9f GB |
1275 | if (!pidx) |
1276 | return SEQ_START_TOKEN; | |
1277 | return (pidx > serv->sv_nrpools ? NULL : &serv->sv_pools[pidx-1]); | |
1278 | } | |
1279 | ||
1280 | static void *svc_pool_stats_next(struct seq_file *m, void *p, loff_t *pos) | |
1281 | { | |
1282 | struct svc_pool *pool = p; | |
1283 | struct svc_serv *serv = m->private; | |
1284 | ||
1285 | dprintk("svc_pool_stats_next, *pos=%llu\n", *pos); | |
1286 | ||
1287 | if (p == SEQ_START_TOKEN) { | |
1288 | pool = &serv->sv_pools[0]; | |
1289 | } else { | |
1290 | unsigned int pidx = (pool - &serv->sv_pools[0]); | |
1291 | if (pidx < serv->sv_nrpools-1) | |
1292 | pool = &serv->sv_pools[pidx+1]; | |
1293 | else | |
1294 | pool = NULL; | |
1295 | } | |
1296 | ++*pos; | |
1297 | return pool; | |
1298 | } | |
1299 | ||
1300 | static void svc_pool_stats_stop(struct seq_file *m, void *p) | |
1301 | { | |
03cf6c9f GB |
1302 | } |
1303 | ||
1304 | static int svc_pool_stats_show(struct seq_file *m, void *p) | |
1305 | { | |
1306 | struct svc_pool *pool = p; | |
1307 | ||
1308 | if (p == SEQ_START_TOKEN) { | |
78c210ef | 1309 | seq_puts(m, "# pool packets-arrived sockets-enqueued threads-woken threads-timedout\n"); |
03cf6c9f GB |
1310 | return 0; |
1311 | } | |
1312 | ||
78c210ef | 1313 | seq_printf(m, "%u %lu %lu %lu %lu\n", |
03cf6c9f GB |
1314 | pool->sp_id, |
1315 | pool->sp_stats.packets, | |
1316 | pool->sp_stats.sockets_queued, | |
1317 | pool->sp_stats.threads_woken, | |
03cf6c9f GB |
1318 | pool->sp_stats.threads_timedout); |
1319 | ||
1320 | return 0; | |
1321 | } | |
1322 | ||
1323 | static const struct seq_operations svc_pool_stats_seq_ops = { | |
1324 | .start = svc_pool_stats_start, | |
1325 | .next = svc_pool_stats_next, | |
1326 | .stop = svc_pool_stats_stop, | |
1327 | .show = svc_pool_stats_show, | |
1328 | }; | |
1329 | ||
1330 | int svc_pool_stats_open(struct svc_serv *serv, struct file *file) | |
1331 | { | |
1332 | int err; | |
1333 | ||
1334 | err = seq_open(file, &svc_pool_stats_seq_ops); | |
1335 | if (!err) | |
1336 | ((struct seq_file *) file->private_data)->private = serv; | |
1337 | return err; | |
1338 | } | |
1339 | EXPORT_SYMBOL(svc_pool_stats_open); | |
1340 | ||
1341 | /*----------------------------------------------------------------------------*/ |