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> |
1d8206b9 TT |
18 | |
19 | #define RPCDBG_FACILITY RPCDBG_SVCXPRT | |
20 | ||
0f0257ea TT |
21 | static struct svc_deferred_req *svc_deferred_dequeue(struct svc_xprt *xprt); |
22 | static int svc_deferred_recv(struct svc_rqst *rqstp); | |
23 | static struct cache_deferred_req *svc_defer(struct cache_req *req); | |
24 | static void svc_age_temp_xprts(unsigned long closure); | |
7710ec36 | 25 | static void svc_delete_xprt(struct svc_xprt *xprt); |
0971374e | 26 | static void svc_xprt_do_enqueue(struct svc_xprt *xprt); |
0f0257ea TT |
27 | |
28 | /* apparently the "standard" is that clients close | |
29 | * idle connections after 5 minutes, servers after | |
30 | * 6 minutes | |
31 | * http://www.connectathon.org/talks96/nfstcp.pdf | |
32 | */ | |
33 | static int svc_conn_age_period = 6*60; | |
34 | ||
1d8206b9 TT |
35 | /* List of registered transport classes */ |
36 | static DEFINE_SPINLOCK(svc_xprt_class_lock); | |
37 | static LIST_HEAD(svc_xprt_class_list); | |
38 | ||
0f0257ea TT |
39 | /* SMP locking strategy: |
40 | * | |
41 | * svc_pool->sp_lock protects most of the fields of that pool. | |
42 | * svc_serv->sv_lock protects sv_tempsocks, sv_permsocks, sv_tmpcnt. | |
43 | * when both need to be taken (rare), svc_serv->sv_lock is first. | |
44 | * BKL protects svc_serv->sv_nrthread. | |
45 | * svc_sock->sk_lock protects the svc_sock->sk_deferred list | |
46 | * and the ->sk_info_authunix cache. | |
47 | * | |
48 | * The XPT_BUSY bit in xprt->xpt_flags prevents a transport being | |
49 | * enqueued multiply. During normal transport processing this bit | |
50 | * is set by svc_xprt_enqueue and cleared by svc_xprt_received. | |
51 | * Providers should not manipulate this bit directly. | |
52 | * | |
53 | * Some flags can be set to certain values at any time | |
54 | * providing that certain rules are followed: | |
55 | * | |
56 | * XPT_CONN, XPT_DATA: | |
57 | * - Can be set or cleared at any time. | |
58 | * - After a set, svc_xprt_enqueue must be called to enqueue | |
59 | * the transport for processing. | |
60 | * - After a clear, the transport must be read/accepted. | |
61 | * If this succeeds, it must be set again. | |
62 | * XPT_CLOSE: | |
63 | * - Can set at any time. It is never cleared. | |
64 | * XPT_DEAD: | |
65 | * - Can only be set while XPT_BUSY is held which ensures | |
66 | * that no other thread will be using the transport or will | |
67 | * try to set XPT_DEAD. | |
68 | */ | |
69 | ||
1d8206b9 TT |
70 | int svc_reg_xprt_class(struct svc_xprt_class *xcl) |
71 | { | |
72 | struct svc_xprt_class *cl; | |
73 | int res = -EEXIST; | |
74 | ||
75 | dprintk("svc: Adding svc transport class '%s'\n", xcl->xcl_name); | |
76 | ||
77 | INIT_LIST_HEAD(&xcl->xcl_list); | |
78 | spin_lock(&svc_xprt_class_lock); | |
79 | /* Make sure there isn't already a class with the same name */ | |
80 | list_for_each_entry(cl, &svc_xprt_class_list, xcl_list) { | |
81 | if (strcmp(xcl->xcl_name, cl->xcl_name) == 0) | |
82 | goto out; | |
83 | } | |
84 | list_add_tail(&xcl->xcl_list, &svc_xprt_class_list); | |
85 | res = 0; | |
86 | out: | |
87 | spin_unlock(&svc_xprt_class_lock); | |
88 | return res; | |
89 | } | |
90 | EXPORT_SYMBOL_GPL(svc_reg_xprt_class); | |
91 | ||
92 | void svc_unreg_xprt_class(struct svc_xprt_class *xcl) | |
93 | { | |
94 | dprintk("svc: Removing svc transport class '%s'\n", xcl->xcl_name); | |
95 | spin_lock(&svc_xprt_class_lock); | |
96 | list_del_init(&xcl->xcl_list); | |
97 | spin_unlock(&svc_xprt_class_lock); | |
98 | } | |
99 | EXPORT_SYMBOL_GPL(svc_unreg_xprt_class); | |
100 | ||
dc9a16e4 TT |
101 | /* |
102 | * Format the transport list for printing | |
103 | */ | |
104 | int svc_print_xprts(char *buf, int maxlen) | |
105 | { | |
8f3a6de3 | 106 | struct svc_xprt_class *xcl; |
dc9a16e4 TT |
107 | char tmpstr[80]; |
108 | int len = 0; | |
109 | buf[0] = '\0'; | |
110 | ||
111 | spin_lock(&svc_xprt_class_lock); | |
8f3a6de3 | 112 | list_for_each_entry(xcl, &svc_xprt_class_list, xcl_list) { |
dc9a16e4 | 113 | int slen; |
dc9a16e4 TT |
114 | |
115 | sprintf(tmpstr, "%s %d\n", xcl->xcl_name, xcl->xcl_max_payload); | |
116 | slen = strlen(tmpstr); | |
117 | if (len + slen > maxlen) | |
118 | break; | |
119 | len += slen; | |
120 | strcat(buf, tmpstr); | |
121 | } | |
122 | spin_unlock(&svc_xprt_class_lock); | |
123 | ||
124 | return len; | |
125 | } | |
126 | ||
e1b3157f TT |
127 | static void svc_xprt_free(struct kref *kref) |
128 | { | |
129 | struct svc_xprt *xprt = | |
130 | container_of(kref, struct svc_xprt, xpt_ref); | |
131 | struct module *owner = xprt->xpt_class->xcl_owner; | |
e3bfca01 PE |
132 | if (test_bit(XPT_CACHE_AUTH, &xprt->xpt_flags)) |
133 | svcauth_unix_info_release(xprt); | |
4fb8518b | 134 | put_net(xprt->xpt_net); |
99de8ea9 BF |
135 | /* See comment on corresponding get in xs_setup_bc_tcp(): */ |
136 | if (xprt->xpt_bc_xprt) | |
137 | xprt_put(xprt->xpt_bc_xprt); | |
e1b3157f TT |
138 | xprt->xpt_ops->xpo_free(xprt); |
139 | module_put(owner); | |
140 | } | |
141 | ||
142 | void svc_xprt_put(struct svc_xprt *xprt) | |
143 | { | |
144 | kref_put(&xprt->xpt_ref, svc_xprt_free); | |
145 | } | |
146 | EXPORT_SYMBOL_GPL(svc_xprt_put); | |
147 | ||
1d8206b9 TT |
148 | /* |
149 | * Called by transport drivers to initialize the transport independent | |
150 | * portion of the transport instance. | |
151 | */ | |
bd4620dd SK |
152 | void svc_xprt_init(struct net *net, struct svc_xprt_class *xcl, |
153 | struct svc_xprt *xprt, struct svc_serv *serv) | |
1d8206b9 TT |
154 | { |
155 | memset(xprt, 0, sizeof(*xprt)); | |
156 | xprt->xpt_class = xcl; | |
157 | xprt->xpt_ops = xcl->xcl_ops; | |
e1b3157f | 158 | kref_init(&xprt->xpt_ref); |
bb5cf160 | 159 | xprt->xpt_server = serv; |
7a182083 TT |
160 | INIT_LIST_HEAD(&xprt->xpt_list); |
161 | INIT_LIST_HEAD(&xprt->xpt_ready); | |
8c7b0172 | 162 | INIT_LIST_HEAD(&xprt->xpt_deferred); |
edc7a894 | 163 | INIT_LIST_HEAD(&xprt->xpt_users); |
a50fea26 | 164 | mutex_init(&xprt->xpt_mutex); |
def13d74 | 165 | spin_lock_init(&xprt->xpt_lock); |
4e5caaa5 | 166 | set_bit(XPT_BUSY, &xprt->xpt_flags); |
4cfc7e60 | 167 | rpc_init_wait_queue(&xprt->xpt_bc_pending, "xpt_bc_pending"); |
bd4620dd | 168 | xprt->xpt_net = get_net(net); |
1d8206b9 TT |
169 | } |
170 | EXPORT_SYMBOL_GPL(svc_xprt_init); | |
b700cbb1 | 171 | |
5dd248f6 CL |
172 | static struct svc_xprt *__svc_xpo_create(struct svc_xprt_class *xcl, |
173 | struct svc_serv *serv, | |
62832c03 | 174 | struct net *net, |
9652ada3 CL |
175 | const int family, |
176 | const unsigned short port, | |
177 | int flags) | |
b700cbb1 | 178 | { |
b700cbb1 TT |
179 | struct sockaddr_in sin = { |
180 | .sin_family = AF_INET, | |
e6f1cebf | 181 | .sin_addr.s_addr = htonl(INADDR_ANY), |
b700cbb1 TT |
182 | .sin_port = htons(port), |
183 | }; | |
dfd56b8b | 184 | #if IS_ENABLED(CONFIG_IPV6) |
5dd248f6 CL |
185 | struct sockaddr_in6 sin6 = { |
186 | .sin6_family = AF_INET6, | |
187 | .sin6_addr = IN6ADDR_ANY_INIT, | |
188 | .sin6_port = htons(port), | |
189 | }; | |
dfd56b8b | 190 | #endif |
5dd248f6 CL |
191 | struct sockaddr *sap; |
192 | size_t len; | |
193 | ||
9652ada3 CL |
194 | switch (family) { |
195 | case PF_INET: | |
5dd248f6 CL |
196 | sap = (struct sockaddr *)&sin; |
197 | len = sizeof(sin); | |
198 | break; | |
dfd56b8b | 199 | #if IS_ENABLED(CONFIG_IPV6) |
9652ada3 | 200 | case PF_INET6: |
5dd248f6 CL |
201 | sap = (struct sockaddr *)&sin6; |
202 | len = sizeof(sin6); | |
203 | break; | |
dfd56b8b | 204 | #endif |
5dd248f6 CL |
205 | default: |
206 | return ERR_PTR(-EAFNOSUPPORT); | |
207 | } | |
208 | ||
62832c03 | 209 | return xcl->xcl_ops->xpo_create(serv, net, sap, len, flags); |
5dd248f6 CL |
210 | } |
211 | ||
6741019c BF |
212 | /* |
213 | * svc_xprt_received conditionally queues the transport for processing | |
214 | * by another thread. The caller must hold the XPT_BUSY bit and must | |
215 | * not thereafter touch transport data. | |
216 | * | |
217 | * Note: XPT_DATA only gets cleared when a read-attempt finds no (or | |
218 | * insufficient) data. | |
219 | */ | |
220 | static void svc_xprt_received(struct svc_xprt *xprt) | |
221 | { | |
ff1fdb9b WAA |
222 | WARN_ON_ONCE(!test_bit(XPT_BUSY, &xprt->xpt_flags)); |
223 | if (!test_bit(XPT_BUSY, &xprt->xpt_flags)) | |
224 | return; | |
6741019c | 225 | /* As soon as we clear busy, the xprt could be closed and |
0971374e | 226 | * 'put', so we need a reference to call svc_xprt_do_enqueue with: |
6741019c BF |
227 | */ |
228 | svc_xprt_get(xprt); | |
0971374e | 229 | smp_mb__before_atomic(); |
6741019c | 230 | clear_bit(XPT_BUSY, &xprt->xpt_flags); |
0971374e | 231 | svc_xprt_do_enqueue(xprt); |
6741019c BF |
232 | svc_xprt_put(xprt); |
233 | } | |
234 | ||
39b55301 BF |
235 | void svc_add_new_perm_xprt(struct svc_serv *serv, struct svc_xprt *new) |
236 | { | |
237 | clear_bit(XPT_TEMP, &new->xpt_flags); | |
238 | spin_lock_bh(&serv->sv_lock); | |
239 | list_add(&new->xpt_list, &serv->sv_permsocks); | |
240 | spin_unlock_bh(&serv->sv_lock); | |
241 | svc_xprt_received(new); | |
242 | } | |
243 | ||
9652ada3 | 244 | int svc_create_xprt(struct svc_serv *serv, const char *xprt_name, |
fc5d00b0 PE |
245 | struct net *net, const int family, |
246 | const unsigned short port, int flags) | |
5dd248f6 CL |
247 | { |
248 | struct svc_xprt_class *xcl; | |
249 | ||
b700cbb1 TT |
250 | dprintk("svc: creating transport %s[%d]\n", xprt_name, port); |
251 | spin_lock(&svc_xprt_class_lock); | |
252 | list_for_each_entry(xcl, &svc_xprt_class_list, xcl_list) { | |
4e5caaa5 | 253 | struct svc_xprt *newxprt; |
ed2849d3 | 254 | unsigned short newport; |
4e5caaa5 TT |
255 | |
256 | if (strcmp(xprt_name, xcl->xcl_name)) | |
257 | continue; | |
258 | ||
259 | if (!try_module_get(xcl->xcl_owner)) | |
260 | goto err; | |
261 | ||
262 | spin_unlock(&svc_xprt_class_lock); | |
62832c03 | 263 | newxprt = __svc_xpo_create(xcl, serv, net, family, port, flags); |
4e5caaa5 TT |
264 | if (IS_ERR(newxprt)) { |
265 | module_put(xcl->xcl_owner); | |
266 | return PTR_ERR(newxprt); | |
b700cbb1 | 267 | } |
39b55301 | 268 | svc_add_new_perm_xprt(serv, newxprt); |
ed2849d3 | 269 | newport = svc_xprt_local_port(newxprt); |
ed2849d3 | 270 | return newport; |
b700cbb1 | 271 | } |
4e5caaa5 | 272 | err: |
b700cbb1 TT |
273 | spin_unlock(&svc_xprt_class_lock); |
274 | dprintk("svc: transport %s not found\n", xprt_name); | |
68717908 CL |
275 | |
276 | /* This errno is exposed to user space. Provide a reasonable | |
277 | * perror msg for a bad transport. */ | |
278 | return -EPROTONOSUPPORT; | |
b700cbb1 TT |
279 | } |
280 | EXPORT_SYMBOL_GPL(svc_create_xprt); | |
9dbc240f TT |
281 | |
282 | /* | |
283 | * Copy the local and remote xprt addresses to the rqstp structure | |
284 | */ | |
285 | void svc_xprt_copy_addrs(struct svc_rqst *rqstp, struct svc_xprt *xprt) | |
286 | { | |
9dbc240f TT |
287 | memcpy(&rqstp->rq_addr, &xprt->xpt_remote, xprt->xpt_remotelen); |
288 | rqstp->rq_addrlen = xprt->xpt_remotelen; | |
289 | ||
290 | /* | |
291 | * Destination address in request is needed for binding the | |
292 | * source address in RPC replies/callbacks later. | |
293 | */ | |
849a1cf1 MJ |
294 | memcpy(&rqstp->rq_daddr, &xprt->xpt_local, xprt->xpt_locallen); |
295 | rqstp->rq_daddrlen = xprt->xpt_locallen; | |
9dbc240f TT |
296 | } |
297 | EXPORT_SYMBOL_GPL(svc_xprt_copy_addrs); | |
298 | ||
0f0257ea TT |
299 | /** |
300 | * svc_print_addr - Format rq_addr field for printing | |
301 | * @rqstp: svc_rqst struct containing address to print | |
302 | * @buf: target buffer for formatted address | |
303 | * @len: length of target buffer | |
304 | * | |
305 | */ | |
306 | char *svc_print_addr(struct svc_rqst *rqstp, char *buf, size_t len) | |
307 | { | |
308 | return __svc_print_addr(svc_addr(rqstp), buf, len); | |
309 | } | |
310 | EXPORT_SYMBOL_GPL(svc_print_addr); | |
311 | ||
312 | /* | |
313 | * Queue up an idle server thread. Must have pool->sp_lock held. | |
314 | * Note: this is really a stack rather than a queue, so that we only | |
315 | * use as many different threads as we need, and the rest don't pollute | |
316 | * the cache. | |
317 | */ | |
318 | static void svc_thread_enqueue(struct svc_pool *pool, struct svc_rqst *rqstp) | |
319 | { | |
320 | list_add(&rqstp->rq_list, &pool->sp_threads); | |
321 | } | |
322 | ||
323 | /* | |
324 | * Dequeue an nfsd thread. Must have pool->sp_lock held. | |
325 | */ | |
326 | static void svc_thread_dequeue(struct svc_pool *pool, struct svc_rqst *rqstp) | |
327 | { | |
328 | list_del(&rqstp->rq_list); | |
329 | } | |
330 | ||
9c335c0b BF |
331 | static bool svc_xprt_has_something_to_do(struct svc_xprt *xprt) |
332 | { | |
333 | if (xprt->xpt_flags & ((1<<XPT_CONN)|(1<<XPT_CLOSE))) | |
334 | return true; | |
335 | if (xprt->xpt_flags & ((1<<XPT_DATA)|(1<<XPT_DEFERRED))) | |
336 | return xprt->xpt_ops->xpo_has_wspace(xprt); | |
337 | return false; | |
338 | } | |
339 | ||
0971374e | 340 | static void svc_xprt_do_enqueue(struct svc_xprt *xprt) |
0f0257ea | 341 | { |
0f0257ea TT |
342 | struct svc_pool *pool; |
343 | struct svc_rqst *rqstp; | |
344 | int cpu; | |
345 | ||
9c335c0b | 346 | if (!svc_xprt_has_something_to_do(xprt)) |
0f0257ea | 347 | return; |
0f0257ea TT |
348 | |
349 | cpu = get_cpu(); | |
350 | pool = svc_pool_for_cpu(xprt->xpt_server, cpu); | |
351 | put_cpu(); | |
352 | ||
353 | spin_lock_bh(&pool->sp_lock); | |
354 | ||
78c210ef BF |
355 | if (!list_empty(&pool->sp_threads) && |
356 | !list_empty(&pool->sp_sockets)) | |
357 | printk(KERN_ERR | |
358 | "svc_xprt_enqueue: " | |
359 | "threads and transports both waiting??\n"); | |
360 | ||
03cf6c9f GB |
361 | pool->sp_stats.packets++; |
362 | ||
0f0257ea TT |
363 | /* Mark transport as busy. It will remain in this state until |
364 | * the provider calls svc_xprt_received. We update XPT_BUSY | |
365 | * atomically because it also guards against trying to enqueue | |
366 | * the transport twice. | |
367 | */ | |
368 | if (test_and_set_bit(XPT_BUSY, &xprt->xpt_flags)) { | |
369 | /* Don't enqueue transport while already enqueued */ | |
370 | dprintk("svc: transport %p busy, not enqueued\n", xprt); | |
371 | goto out_unlock; | |
372 | } | |
0f0257ea | 373 | |
78c210ef | 374 | if (!list_empty(&pool->sp_threads)) { |
0f0257ea TT |
375 | rqstp = list_entry(pool->sp_threads.next, |
376 | struct svc_rqst, | |
377 | rq_list); | |
378 | dprintk("svc: transport %p served by daemon %p\n", | |
379 | xprt, rqstp); | |
380 | svc_thread_dequeue(pool, rqstp); | |
381 | if (rqstp->rq_xprt) | |
382 | printk(KERN_ERR | |
383 | "svc_xprt_enqueue: server %p, rq_xprt=%p!\n", | |
384 | rqstp, rqstp->rq_xprt); | |
385 | rqstp->rq_xprt = xprt; | |
386 | svc_xprt_get(xprt); | |
03cf6c9f | 387 | pool->sp_stats.threads_woken++; |
0f0257ea TT |
388 | wake_up(&rqstp->rq_wait); |
389 | } else { | |
390 | dprintk("svc: transport %p put into queue\n", xprt); | |
391 | list_add_tail(&xprt->xpt_ready, &pool->sp_sockets); | |
03cf6c9f | 392 | pool->sp_stats.sockets_queued++; |
0f0257ea TT |
393 | } |
394 | ||
395 | out_unlock: | |
396 | spin_unlock_bh(&pool->sp_lock); | |
397 | } | |
0971374e TM |
398 | |
399 | /* | |
400 | * Queue up a transport with data pending. If there are idle nfsd | |
401 | * processes, wake 'em up. | |
402 | * | |
403 | */ | |
404 | void svc_xprt_enqueue(struct svc_xprt *xprt) | |
405 | { | |
406 | if (test_bit(XPT_BUSY, &xprt->xpt_flags)) | |
407 | return; | |
408 | svc_xprt_do_enqueue(xprt); | |
409 | } | |
0f0257ea TT |
410 | EXPORT_SYMBOL_GPL(svc_xprt_enqueue); |
411 | ||
412 | /* | |
413 | * Dequeue the first transport. Must be called with the pool->sp_lock held. | |
414 | */ | |
415 | static struct svc_xprt *svc_xprt_dequeue(struct svc_pool *pool) | |
416 | { | |
417 | struct svc_xprt *xprt; | |
418 | ||
419 | if (list_empty(&pool->sp_sockets)) | |
420 | return NULL; | |
421 | ||
422 | xprt = list_entry(pool->sp_sockets.next, | |
423 | struct svc_xprt, xpt_ready); | |
424 | list_del_init(&xprt->xpt_ready); | |
425 | ||
426 | dprintk("svc: transport %p dequeued, inuse=%d\n", | |
427 | xprt, atomic_read(&xprt->xpt_ref.refcount)); | |
428 | ||
429 | return xprt; | |
430 | } | |
431 | ||
0f0257ea TT |
432 | /** |
433 | * svc_reserve - change the space reserved for the reply to a request. | |
434 | * @rqstp: The request in question | |
435 | * @space: new max space to reserve | |
436 | * | |
437 | * Each request reserves some space on the output queue of the transport | |
438 | * to make sure the reply fits. This function reduces that reserved | |
439 | * space to be the amount of space used already, plus @space. | |
440 | * | |
441 | */ | |
442 | void svc_reserve(struct svc_rqst *rqstp, int space) | |
443 | { | |
444 | space += rqstp->rq_res.head[0].iov_len; | |
445 | ||
446 | if (space < rqstp->rq_reserved) { | |
447 | struct svc_xprt *xprt = rqstp->rq_xprt; | |
448 | atomic_sub((rqstp->rq_reserved - space), &xprt->xpt_reserved); | |
449 | rqstp->rq_reserved = space; | |
450 | ||
51877680 TM |
451 | if (xprt->xpt_ops->xpo_adjust_wspace) |
452 | xprt->xpt_ops->xpo_adjust_wspace(xprt); | |
0f0257ea TT |
453 | svc_xprt_enqueue(xprt); |
454 | } | |
455 | } | |
24c3767e | 456 | EXPORT_SYMBOL_GPL(svc_reserve); |
0f0257ea TT |
457 | |
458 | static void svc_xprt_release(struct svc_rqst *rqstp) | |
459 | { | |
460 | struct svc_xprt *xprt = rqstp->rq_xprt; | |
461 | ||
462 | rqstp->rq_xprt->xpt_ops->xpo_release_rqst(rqstp); | |
463 | ||
2779e3ae TT |
464 | kfree(rqstp->rq_deferred); |
465 | rqstp->rq_deferred = NULL; | |
466 | ||
0f0257ea TT |
467 | svc_free_res_pages(rqstp); |
468 | rqstp->rq_res.page_len = 0; | |
469 | rqstp->rq_res.page_base = 0; | |
470 | ||
471 | /* Reset response buffer and release | |
472 | * the reservation. | |
473 | * But first, check that enough space was reserved | |
474 | * for the reply, otherwise we have a bug! | |
475 | */ | |
476 | if ((rqstp->rq_res.len) > rqstp->rq_reserved) | |
477 | printk(KERN_ERR "RPC request reserved %d but used %d\n", | |
478 | rqstp->rq_reserved, | |
479 | rqstp->rq_res.len); | |
480 | ||
481 | rqstp->rq_res.head[0].iov_len = 0; | |
482 | svc_reserve(rqstp, 0); | |
483 | rqstp->rq_xprt = NULL; | |
484 | ||
485 | svc_xprt_put(xprt); | |
486 | } | |
487 | ||
488 | /* | |
489 | * External function to wake up a server waiting for data | |
490 | * This really only makes sense for services like lockd | |
491 | * which have exactly one thread anyway. | |
492 | */ | |
493 | void svc_wake_up(struct svc_serv *serv) | |
494 | { | |
495 | struct svc_rqst *rqstp; | |
496 | unsigned int i; | |
497 | struct svc_pool *pool; | |
498 | ||
499 | for (i = 0; i < serv->sv_nrpools; i++) { | |
500 | pool = &serv->sv_pools[i]; | |
501 | ||
502 | spin_lock_bh(&pool->sp_lock); | |
503 | if (!list_empty(&pool->sp_threads)) { | |
504 | rqstp = list_entry(pool->sp_threads.next, | |
505 | struct svc_rqst, | |
506 | rq_list); | |
507 | dprintk("svc: daemon %p woken up.\n", rqstp); | |
508 | /* | |
509 | svc_thread_dequeue(pool, rqstp); | |
510 | rqstp->rq_xprt = NULL; | |
511 | */ | |
512 | wake_up(&rqstp->rq_wait); | |
35525b79 AS |
513 | } else |
514 | pool->sp_task_pending = 1; | |
0f0257ea TT |
515 | spin_unlock_bh(&pool->sp_lock); |
516 | } | |
517 | } | |
24c3767e | 518 | EXPORT_SYMBOL_GPL(svc_wake_up); |
0f0257ea TT |
519 | |
520 | int svc_port_is_privileged(struct sockaddr *sin) | |
521 | { | |
522 | switch (sin->sa_family) { | |
523 | case AF_INET: | |
524 | return ntohs(((struct sockaddr_in *)sin)->sin_port) | |
525 | < PROT_SOCK; | |
526 | case AF_INET6: | |
527 | return ntohs(((struct sockaddr_in6 *)sin)->sin6_port) | |
528 | < PROT_SOCK; | |
529 | default: | |
530 | return 0; | |
531 | } | |
532 | } | |
533 | ||
534 | /* | |
c9233eb7 JL |
535 | * Make sure that we don't have too many active connections. If we have, |
536 | * something must be dropped. It's not clear what will happen if we allow | |
537 | * "too many" connections, but when dealing with network-facing software, | |
538 | * we have to code defensively. Here we do that by imposing hard limits. | |
0f0257ea TT |
539 | * |
540 | * There's no point in trying to do random drop here for DoS | |
541 | * prevention. The NFS clients does 1 reconnect in 15 seconds. An | |
542 | * attacker can easily beat that. | |
543 | * | |
544 | * The only somewhat efficient mechanism would be if drop old | |
545 | * connections from the same IP first. But right now we don't even | |
546 | * record the client IP in svc_sock. | |
c9233eb7 JL |
547 | * |
548 | * single-threaded services that expect a lot of clients will probably | |
549 | * need to set sv_maxconn to override the default value which is based | |
550 | * on the number of threads | |
0f0257ea TT |
551 | */ |
552 | static void svc_check_conn_limits(struct svc_serv *serv) | |
553 | { | |
c9233eb7 JL |
554 | unsigned int limit = serv->sv_maxconn ? serv->sv_maxconn : |
555 | (serv->sv_nrthreads+3) * 20; | |
556 | ||
557 | if (serv->sv_tmpcnt > limit) { | |
0f0257ea TT |
558 | struct svc_xprt *xprt = NULL; |
559 | spin_lock_bh(&serv->sv_lock); | |
560 | if (!list_empty(&serv->sv_tempsocks)) { | |
e87cc472 JP |
561 | /* Try to help the admin */ |
562 | net_notice_ratelimited("%s: too many open connections, consider increasing the %s\n", | |
563 | serv->sv_name, serv->sv_maxconn ? | |
564 | "max number of connections" : | |
565 | "number of threads"); | |
0f0257ea TT |
566 | /* |
567 | * Always select the oldest connection. It's not fair, | |
568 | * but so is life | |
569 | */ | |
570 | xprt = list_entry(serv->sv_tempsocks.prev, | |
571 | struct svc_xprt, | |
572 | xpt_list); | |
573 | set_bit(XPT_CLOSE, &xprt->xpt_flags); | |
574 | svc_xprt_get(xprt); | |
575 | } | |
576 | spin_unlock_bh(&serv->sv_lock); | |
577 | ||
578 | if (xprt) { | |
579 | svc_xprt_enqueue(xprt); | |
580 | svc_xprt_put(xprt); | |
581 | } | |
582 | } | |
583 | } | |
584 | ||
e1d83ee6 | 585 | static int svc_alloc_arg(struct svc_rqst *rqstp) |
0f0257ea | 586 | { |
6797fa5a BF |
587 | struct svc_serv *serv = rqstp->rq_server; |
588 | struct xdr_buf *arg; | |
589 | int pages; | |
590 | int i; | |
0f0257ea TT |
591 | |
592 | /* now allocate needed pages. If we get a failure, sleep briefly */ | |
593 | pages = (serv->sv_max_mesg + PAGE_SIZE) / PAGE_SIZE; | |
b25cd058 WAA |
594 | WARN_ON_ONCE(pages >= RPCSVC_MAXPAGES); |
595 | if (pages >= RPCSVC_MAXPAGES) | |
596 | /* use as many pages as possible */ | |
597 | pages = RPCSVC_MAXPAGES - 1; | |
0f0257ea TT |
598 | for (i = 0; i < pages ; i++) |
599 | while (rqstp->rq_pages[i] == NULL) { | |
600 | struct page *p = alloc_page(GFP_KERNEL); | |
601 | if (!p) { | |
7b54fe61 JL |
602 | set_current_state(TASK_INTERRUPTIBLE); |
603 | if (signalled() || kthread_should_stop()) { | |
604 | set_current_state(TASK_RUNNING); | |
7086721f | 605 | return -EINTR; |
7b54fe61 JL |
606 | } |
607 | schedule_timeout(msecs_to_jiffies(500)); | |
0f0257ea TT |
608 | } |
609 | rqstp->rq_pages[i] = p; | |
610 | } | |
2825a7f9 | 611 | rqstp->rq_page_end = &rqstp->rq_pages[i]; |
0f0257ea | 612 | rqstp->rq_pages[i++] = NULL; /* this might be seen in nfs_read_actor */ |
0f0257ea TT |
613 | |
614 | /* Make arg->head point to first page and arg->pages point to rest */ | |
615 | arg = &rqstp->rq_arg; | |
616 | arg->head[0].iov_base = page_address(rqstp->rq_pages[0]); | |
617 | arg->head[0].iov_len = PAGE_SIZE; | |
618 | arg->pages = rqstp->rq_pages + 1; | |
619 | arg->page_base = 0; | |
620 | /* save at least one page for response */ | |
621 | arg->page_len = (pages-2)*PAGE_SIZE; | |
622 | arg->len = (pages-1)*PAGE_SIZE; | |
623 | arg->tail[0].iov_len = 0; | |
6797fa5a BF |
624 | return 0; |
625 | } | |
0f0257ea | 626 | |
e1d83ee6 | 627 | static struct svc_xprt *svc_get_next_xprt(struct svc_rqst *rqstp, long timeout) |
6797fa5a BF |
628 | { |
629 | struct svc_xprt *xprt; | |
630 | struct svc_pool *pool = rqstp->rq_pool; | |
631 | DECLARE_WAITQUEUE(wait, current); | |
632 | long time_left; | |
0f0257ea | 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); | |
f16b6e8d N |
644 | |
645 | /* As there is a shortage of threads and this request | |
6610f720 | 646 | * had to be queued, don't allow the thread to wait so |
f16b6e8d N |
647 | * long for cache updates. |
648 | */ | |
649 | rqstp->rq_chandle.thread_wait = 1*HZ; | |
35525b79 | 650 | pool->sp_task_pending = 0; |
0f0257ea | 651 | } else { |
35525b79 AS |
652 | if (pool->sp_task_pending) { |
653 | pool->sp_task_pending = 0; | |
106f359c TM |
654 | xprt = ERR_PTR(-EAGAIN); |
655 | goto out; | |
35525b79 | 656 | } |
0f0257ea TT |
657 | /* No data pending. Go to sleep */ |
658 | svc_thread_enqueue(pool, rqstp); | |
659 | ||
660 | /* | |
661 | * We have to be able to interrupt this wait | |
662 | * to bring down the daemons ... | |
663 | */ | |
664 | set_current_state(TASK_INTERRUPTIBLE); | |
7086721f JL |
665 | |
666 | /* | |
667 | * checking kthread_should_stop() here allows us to avoid | |
668 | * locking and signalling when stopping kthreads that call | |
669 | * svc_recv. If the thread has already been woken up, then | |
670 | * we can exit here without sleeping. If not, then it | |
671 | * it'll be woken up quickly during the schedule_timeout | |
672 | */ | |
673 | if (kthread_should_stop()) { | |
674 | set_current_state(TASK_RUNNING); | |
106f359c TM |
675 | xprt = ERR_PTR(-EINTR); |
676 | goto out; | |
7086721f JL |
677 | } |
678 | ||
0f0257ea TT |
679 | add_wait_queue(&rqstp->rq_wait, &wait); |
680 | spin_unlock_bh(&pool->sp_lock); | |
681 | ||
03cf6c9f | 682 | time_left = schedule_timeout(timeout); |
0f0257ea TT |
683 | |
684 | try_to_freeze(); | |
685 | ||
0f0257ea | 686 | remove_wait_queue(&rqstp->rq_wait, &wait); |
106f359c TM |
687 | xprt = rqstp->rq_xprt; |
688 | if (xprt != NULL) | |
689 | return xprt; | |
690 | ||
691 | spin_lock_bh(&pool->sp_lock); | |
03cf6c9f GB |
692 | if (!time_left) |
693 | pool->sp_stats.threads_timedout++; | |
0f0257ea TT |
694 | |
695 | xprt = rqstp->rq_xprt; | |
696 | if (!xprt) { | |
697 | svc_thread_dequeue(pool, rqstp); | |
698 | spin_unlock_bh(&pool->sp_lock); | |
699 | dprintk("svc: server %p, no data yet\n", rqstp); | |
7086721f | 700 | if (signalled() || kthread_should_stop()) |
6797fa5a | 701 | return ERR_PTR(-EINTR); |
7086721f | 702 | else |
6797fa5a | 703 | return ERR_PTR(-EAGAIN); |
0f0257ea TT |
704 | } |
705 | } | |
106f359c | 706 | out: |
0f0257ea | 707 | spin_unlock_bh(&pool->sp_lock); |
6797fa5a BF |
708 | return xprt; |
709 | } | |
710 | ||
e1d83ee6 | 711 | static void svc_add_new_temp_xprt(struct svc_serv *serv, struct svc_xprt *newxpt) |
65b2e665 BF |
712 | { |
713 | spin_lock_bh(&serv->sv_lock); | |
714 | set_bit(XPT_TEMP, &newxpt->xpt_flags); | |
715 | list_add(&newxpt->xpt_list, &serv->sv_tempsocks); | |
716 | serv->sv_tmpcnt++; | |
717 | if (serv->sv_temptimer.function == NULL) { | |
718 | /* setup timer to age temp transports */ | |
719 | setup_timer(&serv->sv_temptimer, svc_age_temp_xprts, | |
720 | (unsigned long)serv); | |
721 | mod_timer(&serv->sv_temptimer, | |
722 | jiffies + svc_conn_age_period * HZ); | |
723 | } | |
724 | spin_unlock_bh(&serv->sv_lock); | |
725 | svc_xprt_received(newxpt); | |
726 | } | |
727 | ||
6797fa5a BF |
728 | static int svc_handle_xprt(struct svc_rqst *rqstp, struct svc_xprt *xprt) |
729 | { | |
730 | struct svc_serv *serv = rqstp->rq_server; | |
731 | int len = 0; | |
0f0257ea | 732 | |
1b644b6e BF |
733 | if (test_bit(XPT_CLOSE, &xprt->xpt_flags)) { |
734 | dprintk("svc_recv: found XPT_CLOSE\n"); | |
735 | svc_delete_xprt(xprt); | |
ca7896cd | 736 | /* Leave XPT_BUSY set on the dead xprt: */ |
6797fa5a | 737 | return 0; |
ca7896cd BF |
738 | } |
739 | if (test_bit(XPT_LISTENER, &xprt->xpt_flags)) { | |
0f0257ea | 740 | struct svc_xprt *newxpt; |
65b2e665 BF |
741 | /* |
742 | * We know this module_get will succeed because the | |
743 | * listener holds a reference too | |
744 | */ | |
745 | __module_get(xprt->xpt_class->xcl_owner); | |
746 | svc_check_conn_limits(xprt->xpt_server); | |
0f0257ea | 747 | newxpt = xprt->xpt_ops->xpo_accept(xprt); |
65b2e665 BF |
748 | if (newxpt) |
749 | svc_add_new_temp_xprt(serv, newxpt); | |
c789102c TM |
750 | else |
751 | module_put(xprt->xpt_class->xcl_owner); | |
9e5b208d | 752 | } else { |
6797fa5a | 753 | /* XPT_DATA|XPT_DEFERRED case: */ |
0f0257ea | 754 | dprintk("svc: server %p, pool %u, transport %p, inuse=%d\n", |
6797fa5a | 755 | rqstp, rqstp->rq_pool->sp_id, xprt, |
0f0257ea TT |
756 | atomic_read(&xprt->xpt_ref.refcount)); |
757 | rqstp->rq_deferred = svc_deferred_dequeue(xprt); | |
ca7896cd | 758 | if (rqstp->rq_deferred) |
0f0257ea | 759 | len = svc_deferred_recv(rqstp); |
ca7896cd | 760 | else |
0f0257ea TT |
761 | len = xprt->xpt_ops->xpo_recvfrom(rqstp); |
762 | dprintk("svc: got len=%d\n", len); | |
d10f27a7 BF |
763 | rqstp->rq_reserved = serv->sv_max_mesg; |
764 | atomic_add(rqstp->rq_reserved, &xprt->xpt_reserved); | |
0f0257ea | 765 | } |
6797fa5a | 766 | /* clear XPT_BUSY: */ |
ca7896cd | 767 | svc_xprt_received(xprt); |
6797fa5a BF |
768 | return len; |
769 | } | |
770 | ||
771 | /* | |
772 | * Receive the next request on any transport. This code is carefully | |
773 | * organised not to touch any cachelines in the shared svc_serv | |
774 | * structure, only cachelines in the local svc_pool. | |
775 | */ | |
776 | int svc_recv(struct svc_rqst *rqstp, long timeout) | |
777 | { | |
778 | struct svc_xprt *xprt = NULL; | |
779 | struct svc_serv *serv = rqstp->rq_server; | |
780 | int len, err; | |
781 | ||
782 | dprintk("svc: server %p waiting for data (to = %ld)\n", | |
783 | rqstp, timeout); | |
784 | ||
785 | if (rqstp->rq_xprt) | |
786 | printk(KERN_ERR | |
787 | "svc_recv: service %p, transport not NULL!\n", | |
788 | rqstp); | |
789 | if (waitqueue_active(&rqstp->rq_wait)) | |
790 | printk(KERN_ERR | |
791 | "svc_recv: service %p, wait queue active!\n", | |
792 | rqstp); | |
793 | ||
794 | err = svc_alloc_arg(rqstp); | |
795 | if (err) | |
796 | return err; | |
797 | ||
798 | try_to_freeze(); | |
799 | cond_resched(); | |
800 | if (signalled() || kthread_should_stop()) | |
801 | return -EINTR; | |
802 | ||
803 | xprt = svc_get_next_xprt(rqstp, timeout); | |
804 | if (IS_ERR(xprt)) | |
805 | return PTR_ERR(xprt); | |
806 | ||
807 | len = svc_handle_xprt(rqstp, xprt); | |
0f0257ea TT |
808 | |
809 | /* No data, incomplete (TCP) read, or accept() */ | |
9f9d2ebe | 810 | if (len <= 0) |
ca7896cd BF |
811 | goto out; |
812 | ||
0f0257ea TT |
813 | clear_bit(XPT_OLD, &xprt->xpt_flags); |
814 | ||
16e4d93f | 815 | rqstp->rq_secure = xprt->xpt_ops->xpo_secure_port(rqstp); |
0f0257ea TT |
816 | rqstp->rq_chandle.defer = svc_defer; |
817 | ||
818 | if (serv->sv_stats) | |
819 | serv->sv_stats->netcnt++; | |
820 | return len; | |
ca7896cd BF |
821 | out: |
822 | rqstp->rq_res.len = 0; | |
823 | svc_xprt_release(rqstp); | |
824 | return -EAGAIN; | |
0f0257ea | 825 | } |
24c3767e | 826 | EXPORT_SYMBOL_GPL(svc_recv); |
0f0257ea TT |
827 | |
828 | /* | |
829 | * Drop request | |
830 | */ | |
831 | void svc_drop(struct svc_rqst *rqstp) | |
832 | { | |
833 | dprintk("svc: xprt %p dropped request\n", rqstp->rq_xprt); | |
834 | svc_xprt_release(rqstp); | |
835 | } | |
24c3767e | 836 | EXPORT_SYMBOL_GPL(svc_drop); |
0f0257ea TT |
837 | |
838 | /* | |
839 | * Return reply to client. | |
840 | */ | |
841 | int svc_send(struct svc_rqst *rqstp) | |
842 | { | |
843 | struct svc_xprt *xprt; | |
844 | int len; | |
845 | struct xdr_buf *xb; | |
846 | ||
847 | xprt = rqstp->rq_xprt; | |
848 | if (!xprt) | |
849 | return -EFAULT; | |
850 | ||
851 | /* release the receive skb before sending the reply */ | |
852 | rqstp->rq_xprt->xpt_ops->xpo_release_rqst(rqstp); | |
853 | ||
854 | /* calculate over-all length */ | |
855 | xb = &rqstp->rq_res; | |
856 | xb->len = xb->head[0].iov_len + | |
857 | xb->page_len + | |
858 | xb->tail[0].iov_len; | |
859 | ||
860 | /* Grab mutex to serialize outgoing data. */ | |
861 | mutex_lock(&xprt->xpt_mutex); | |
f06f00a2 BF |
862 | if (test_bit(XPT_DEAD, &xprt->xpt_flags) |
863 | || test_bit(XPT_CLOSE, &xprt->xpt_flags)) | |
0f0257ea TT |
864 | len = -ENOTCONN; |
865 | else | |
866 | len = xprt->xpt_ops->xpo_sendto(rqstp); | |
867 | mutex_unlock(&xprt->xpt_mutex); | |
4cfc7e60 | 868 | rpc_wake_up(&xprt->xpt_bc_pending); |
0f0257ea TT |
869 | svc_xprt_release(rqstp); |
870 | ||
871 | if (len == -ECONNREFUSED || len == -ENOTCONN || len == -EAGAIN) | |
872 | return 0; | |
873 | return len; | |
874 | } | |
875 | ||
876 | /* | |
877 | * Timer function to close old temporary transports, using | |
878 | * a mark-and-sweep algorithm. | |
879 | */ | |
880 | static void svc_age_temp_xprts(unsigned long closure) | |
881 | { | |
882 | struct svc_serv *serv = (struct svc_serv *)closure; | |
883 | struct svc_xprt *xprt; | |
884 | struct list_head *le, *next; | |
0f0257ea TT |
885 | |
886 | dprintk("svc_age_temp_xprts\n"); | |
887 | ||
888 | if (!spin_trylock_bh(&serv->sv_lock)) { | |
889 | /* busy, try again 1 sec later */ | |
890 | dprintk("svc_age_temp_xprts: busy\n"); | |
891 | mod_timer(&serv->sv_temptimer, jiffies + HZ); | |
892 | return; | |
893 | } | |
894 | ||
895 | list_for_each_safe(le, next, &serv->sv_tempsocks) { | |
896 | xprt = list_entry(le, struct svc_xprt, xpt_list); | |
897 | ||
898 | /* First time through, just mark it OLD. Second time | |
899 | * through, close it. */ | |
900 | if (!test_and_set_bit(XPT_OLD, &xprt->xpt_flags)) | |
901 | continue; | |
f64f9e71 JP |
902 | if (atomic_read(&xprt->xpt_ref.refcount) > 1 || |
903 | test_bit(XPT_BUSY, &xprt->xpt_flags)) | |
0f0257ea | 904 | continue; |
e75bafbf | 905 | list_del_init(le); |
0f0257ea TT |
906 | set_bit(XPT_CLOSE, &xprt->xpt_flags); |
907 | set_bit(XPT_DETACHED, &xprt->xpt_flags); | |
0f0257ea TT |
908 | dprintk("queuing xprt %p for closing\n", xprt); |
909 | ||
910 | /* a thread will dequeue and close it soon */ | |
911 | svc_xprt_enqueue(xprt); | |
0f0257ea | 912 | } |
e75bafbf | 913 | spin_unlock_bh(&serv->sv_lock); |
0f0257ea TT |
914 | |
915 | mod_timer(&serv->sv_temptimer, jiffies + svc_conn_age_period * HZ); | |
916 | } | |
917 | ||
edc7a894 BF |
918 | static void call_xpt_users(struct svc_xprt *xprt) |
919 | { | |
920 | struct svc_xpt_user *u; | |
921 | ||
922 | spin_lock(&xprt->xpt_lock); | |
923 | while (!list_empty(&xprt->xpt_users)) { | |
924 | u = list_first_entry(&xprt->xpt_users, struct svc_xpt_user, list); | |
925 | list_del(&u->list); | |
926 | u->callback(u); | |
927 | } | |
928 | spin_unlock(&xprt->xpt_lock); | |
929 | } | |
930 | ||
0f0257ea TT |
931 | /* |
932 | * Remove a dead transport | |
933 | */ | |
7710ec36 | 934 | static void svc_delete_xprt(struct svc_xprt *xprt) |
0f0257ea TT |
935 | { |
936 | struct svc_serv *serv = xprt->xpt_server; | |
22945e4a TT |
937 | struct svc_deferred_req *dr; |
938 | ||
939 | /* Only do this once */ | |
940 | if (test_and_set_bit(XPT_DEAD, &xprt->xpt_flags)) | |
ac9303eb | 941 | BUG(); |
0f0257ea TT |
942 | |
943 | dprintk("svc: svc_delete_xprt(%p)\n", xprt); | |
944 | xprt->xpt_ops->xpo_detach(xprt); | |
945 | ||
946 | spin_lock_bh(&serv->sv_lock); | |
947 | if (!test_and_set_bit(XPT_DETACHED, &xprt->xpt_flags)) | |
948 | list_del_init(&xprt->xpt_list); | |
01047298 | 949 | WARN_ON_ONCE(!list_empty(&xprt->xpt_ready)); |
22945e4a TT |
950 | if (test_bit(XPT_TEMP, &xprt->xpt_flags)) |
951 | serv->sv_tmpcnt--; | |
788e69e5 | 952 | spin_unlock_bh(&serv->sv_lock); |
22945e4a | 953 | |
ab1b18f7 | 954 | while ((dr = svc_deferred_dequeue(xprt)) != NULL) |
22945e4a | 955 | kfree(dr); |
22945e4a | 956 | |
edc7a894 | 957 | call_xpt_users(xprt); |
22945e4a | 958 | svc_xprt_put(xprt); |
0f0257ea TT |
959 | } |
960 | ||
961 | void svc_close_xprt(struct svc_xprt *xprt) | |
962 | { | |
963 | set_bit(XPT_CLOSE, &xprt->xpt_flags); | |
964 | if (test_and_set_bit(XPT_BUSY, &xprt->xpt_flags)) | |
965 | /* someone else will have to effect the close */ | |
966 | return; | |
b1763316 BF |
967 | /* |
968 | * We expect svc_close_xprt() to work even when no threads are | |
969 | * running (e.g., while configuring the server before starting | |
970 | * any threads), so if the transport isn't busy, we delete | |
971 | * it ourself: | |
972 | */ | |
0f0257ea | 973 | svc_delete_xprt(xprt); |
0f0257ea | 974 | } |
a217813f | 975 | EXPORT_SYMBOL_GPL(svc_close_xprt); |
0f0257ea | 976 | |
cc630d9f | 977 | static int svc_close_list(struct svc_serv *serv, struct list_head *xprt_list, struct net *net) |
0f0257ea TT |
978 | { |
979 | struct svc_xprt *xprt; | |
cc630d9f | 980 | int ret = 0; |
0f0257ea | 981 | |
719f8bcc | 982 | spin_lock(&serv->sv_lock); |
b4f36f88 | 983 | list_for_each_entry(xprt, xprt_list, xpt_list) { |
7b147f1f SK |
984 | if (xprt->xpt_net != net) |
985 | continue; | |
cc630d9f | 986 | ret++; |
0f0257ea | 987 | set_bit(XPT_CLOSE, &xprt->xpt_flags); |
cc630d9f | 988 | svc_xprt_enqueue(xprt); |
0f0257ea | 989 | } |
719f8bcc | 990 | spin_unlock(&serv->sv_lock); |
cc630d9f | 991 | return ret; |
0f0257ea TT |
992 | } |
993 | ||
cc630d9f | 994 | static struct svc_xprt *svc_dequeue_net(struct svc_serv *serv, struct net *net) |
0f0257ea | 995 | { |
b4f36f88 | 996 | struct svc_pool *pool; |
0f0257ea TT |
997 | struct svc_xprt *xprt; |
998 | struct svc_xprt *tmp; | |
b4f36f88 BF |
999 | int i; |
1000 | ||
b4f36f88 BF |
1001 | for (i = 0; i < serv->sv_nrpools; i++) { |
1002 | pool = &serv->sv_pools[i]; | |
1003 | ||
1004 | spin_lock_bh(&pool->sp_lock); | |
6f513365 | 1005 | list_for_each_entry_safe(xprt, tmp, &pool->sp_sockets, xpt_ready) { |
7b147f1f SK |
1006 | if (xprt->xpt_net != net) |
1007 | continue; | |
b4f36f88 | 1008 | list_del_init(&xprt->xpt_ready); |
cc630d9f BF |
1009 | spin_unlock_bh(&pool->sp_lock); |
1010 | return xprt; | |
b4f36f88 BF |
1011 | } |
1012 | spin_unlock_bh(&pool->sp_lock); | |
1013 | } | |
cc630d9f | 1014 | return NULL; |
6f513365 SK |
1015 | } |
1016 | ||
cc630d9f | 1017 | static void svc_clean_up_xprts(struct svc_serv *serv, struct net *net) |
6f513365 SK |
1018 | { |
1019 | struct svc_xprt *xprt; | |
719f8bcc | 1020 | |
cc630d9f BF |
1021 | while ((xprt = svc_dequeue_net(serv, net))) { |
1022 | set_bit(XPT_CLOSE, &xprt->xpt_flags); | |
719f8bcc | 1023 | svc_delete_xprt(xprt); |
cc630d9f | 1024 | } |
3a22bf50 SK |
1025 | } |
1026 | ||
cc630d9f BF |
1027 | /* |
1028 | * Server threads may still be running (especially in the case where the | |
1029 | * service is still running in other network namespaces). | |
1030 | * | |
1031 | * So we shut down sockets the same way we would on a running server, by | |
1032 | * setting XPT_CLOSE, enqueuing, and letting a thread pick it up to do | |
1033 | * the close. In the case there are no such other threads, | |
1034 | * threads running, svc_clean_up_xprts() does a simple version of a | |
1035 | * server's main event loop, and in the case where there are other | |
1036 | * threads, we may need to wait a little while and then check again to | |
1037 | * see if they're done. | |
1038 | */ | |
7b147f1f | 1039 | void svc_close_net(struct svc_serv *serv, struct net *net) |
3a22bf50 | 1040 | { |
cc630d9f | 1041 | int delay = 0; |
6f513365 | 1042 | |
cc630d9f BF |
1043 | while (svc_close_list(serv, &serv->sv_permsocks, net) + |
1044 | svc_close_list(serv, &serv->sv_tempsocks, net)) { | |
1045 | ||
1046 | svc_clean_up_xprts(serv, net); | |
1047 | msleep(delay++); | |
1048 | } | |
0f0257ea TT |
1049 | } |
1050 | ||
1051 | /* | |
1052 | * Handle defer and revisit of requests | |
1053 | */ | |
1054 | ||
1055 | static void svc_revisit(struct cache_deferred_req *dreq, int too_many) | |
1056 | { | |
1057 | struct svc_deferred_req *dr = | |
1058 | container_of(dreq, struct svc_deferred_req, handle); | |
1059 | struct svc_xprt *xprt = dr->xprt; | |
1060 | ||
22945e4a TT |
1061 | spin_lock(&xprt->xpt_lock); |
1062 | set_bit(XPT_DEFERRED, &xprt->xpt_flags); | |
1063 | if (too_many || test_bit(XPT_DEAD, &xprt->xpt_flags)) { | |
1064 | spin_unlock(&xprt->xpt_lock); | |
1065 | dprintk("revisit canceled\n"); | |
0f0257ea TT |
1066 | svc_xprt_put(xprt); |
1067 | kfree(dr); | |
1068 | return; | |
1069 | } | |
1070 | dprintk("revisit queued\n"); | |
1071 | dr->xprt = NULL; | |
0f0257ea TT |
1072 | list_add(&dr->handle.recent, &xprt->xpt_deferred); |
1073 | spin_unlock(&xprt->xpt_lock); | |
0f0257ea TT |
1074 | svc_xprt_enqueue(xprt); |
1075 | svc_xprt_put(xprt); | |
1076 | } | |
1077 | ||
260c1d12 TT |
1078 | /* |
1079 | * Save the request off for later processing. The request buffer looks | |
1080 | * like this: | |
1081 | * | |
1082 | * <xprt-header><rpc-header><rpc-pagelist><rpc-tail> | |
1083 | * | |
1084 | * This code can only handle requests that consist of an xprt-header | |
1085 | * and rpc-header. | |
1086 | */ | |
0f0257ea TT |
1087 | static struct cache_deferred_req *svc_defer(struct cache_req *req) |
1088 | { | |
1089 | struct svc_rqst *rqstp = container_of(req, struct svc_rqst, rq_chandle); | |
0f0257ea TT |
1090 | struct svc_deferred_req *dr; |
1091 | ||
2f425878 | 1092 | if (rqstp->rq_arg.page_len || !rqstp->rq_usedeferral) |
0f0257ea TT |
1093 | return NULL; /* if more than a page, give up FIXME */ |
1094 | if (rqstp->rq_deferred) { | |
1095 | dr = rqstp->rq_deferred; | |
1096 | rqstp->rq_deferred = NULL; | |
1097 | } else { | |
260c1d12 TT |
1098 | size_t skip; |
1099 | size_t size; | |
0f0257ea | 1100 | /* FIXME maybe discard if size too large */ |
260c1d12 | 1101 | size = sizeof(struct svc_deferred_req) + rqstp->rq_arg.len; |
0f0257ea TT |
1102 | dr = kmalloc(size, GFP_KERNEL); |
1103 | if (dr == NULL) | |
1104 | return NULL; | |
1105 | ||
1106 | dr->handle.owner = rqstp->rq_server; | |
1107 | dr->prot = rqstp->rq_prot; | |
1108 | memcpy(&dr->addr, &rqstp->rq_addr, rqstp->rq_addrlen); | |
1109 | dr->addrlen = rqstp->rq_addrlen; | |
1110 | dr->daddr = rqstp->rq_daddr; | |
1111 | dr->argslen = rqstp->rq_arg.len >> 2; | |
260c1d12 TT |
1112 | dr->xprt_hlen = rqstp->rq_xprt_hlen; |
1113 | ||
1114 | /* back up head to the start of the buffer and copy */ | |
1115 | skip = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len; | |
1116 | memcpy(dr->args, rqstp->rq_arg.head[0].iov_base - skip, | |
1117 | dr->argslen << 2); | |
0f0257ea TT |
1118 | } |
1119 | svc_xprt_get(rqstp->rq_xprt); | |
1120 | dr->xprt = rqstp->rq_xprt; | |
9e701c61 | 1121 | rqstp->rq_dropme = true; |
0f0257ea TT |
1122 | |
1123 | dr->handle.revisit = svc_revisit; | |
1124 | return &dr->handle; | |
1125 | } | |
1126 | ||
1127 | /* | |
1128 | * recv data from a deferred request into an active one | |
1129 | */ | |
1130 | static int svc_deferred_recv(struct svc_rqst *rqstp) | |
1131 | { | |
1132 | struct svc_deferred_req *dr = rqstp->rq_deferred; | |
1133 | ||
260c1d12 TT |
1134 | /* setup iov_base past transport header */ |
1135 | rqstp->rq_arg.head[0].iov_base = dr->args + (dr->xprt_hlen>>2); | |
1136 | /* The iov_len does not include the transport header bytes */ | |
1137 | rqstp->rq_arg.head[0].iov_len = (dr->argslen<<2) - dr->xprt_hlen; | |
0f0257ea | 1138 | rqstp->rq_arg.page_len = 0; |
260c1d12 TT |
1139 | /* The rq_arg.len includes the transport header bytes */ |
1140 | rqstp->rq_arg.len = dr->argslen<<2; | |
0f0257ea TT |
1141 | rqstp->rq_prot = dr->prot; |
1142 | memcpy(&rqstp->rq_addr, &dr->addr, dr->addrlen); | |
1143 | rqstp->rq_addrlen = dr->addrlen; | |
260c1d12 TT |
1144 | /* Save off transport header len in case we get deferred again */ |
1145 | rqstp->rq_xprt_hlen = dr->xprt_hlen; | |
0f0257ea TT |
1146 | rqstp->rq_daddr = dr->daddr; |
1147 | rqstp->rq_respages = rqstp->rq_pages; | |
260c1d12 | 1148 | return (dr->argslen<<2) - dr->xprt_hlen; |
0f0257ea TT |
1149 | } |
1150 | ||
1151 | ||
1152 | static struct svc_deferred_req *svc_deferred_dequeue(struct svc_xprt *xprt) | |
1153 | { | |
1154 | struct svc_deferred_req *dr = NULL; | |
1155 | ||
1156 | if (!test_bit(XPT_DEFERRED, &xprt->xpt_flags)) | |
1157 | return NULL; | |
1158 | spin_lock(&xprt->xpt_lock); | |
0f0257ea TT |
1159 | if (!list_empty(&xprt->xpt_deferred)) { |
1160 | dr = list_entry(xprt->xpt_deferred.next, | |
1161 | struct svc_deferred_req, | |
1162 | handle.recent); | |
1163 | list_del_init(&dr->handle.recent); | |
62bac4af BF |
1164 | } else |
1165 | clear_bit(XPT_DEFERRED, &xprt->xpt_flags); | |
0f0257ea TT |
1166 | spin_unlock(&xprt->xpt_lock); |
1167 | return dr; | |
1168 | } | |
7fcb98d5 | 1169 | |
156e6209 CL |
1170 | /** |
1171 | * svc_find_xprt - find an RPC transport instance | |
1172 | * @serv: pointer to svc_serv to search | |
1173 | * @xcl_name: C string containing transport's class name | |
4cb54ca2 | 1174 | * @net: owner net pointer |
156e6209 CL |
1175 | * @af: Address family of transport's local address |
1176 | * @port: transport's IP port number | |
1177 | * | |
7fcb98d5 TT |
1178 | * Return the transport instance pointer for the endpoint accepting |
1179 | * connections/peer traffic from the specified transport class, | |
1180 | * address family and port. | |
1181 | * | |
1182 | * Specifying 0 for the address family or port is effectively a | |
1183 | * wild-card, and will result in matching the first transport in the | |
1184 | * service's list that has a matching class name. | |
1185 | */ | |
156e6209 | 1186 | struct svc_xprt *svc_find_xprt(struct svc_serv *serv, const char *xcl_name, |
4cb54ca2 SK |
1187 | struct net *net, const sa_family_t af, |
1188 | const unsigned short port) | |
7fcb98d5 TT |
1189 | { |
1190 | struct svc_xprt *xprt; | |
1191 | struct svc_xprt *found = NULL; | |
1192 | ||
1193 | /* Sanity check the args */ | |
156e6209 | 1194 | if (serv == NULL || xcl_name == NULL) |
7fcb98d5 TT |
1195 | return found; |
1196 | ||
1197 | spin_lock_bh(&serv->sv_lock); | |
1198 | list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list) { | |
4cb54ca2 SK |
1199 | if (xprt->xpt_net != net) |
1200 | continue; | |
7fcb98d5 TT |
1201 | if (strcmp(xprt->xpt_class->xcl_name, xcl_name)) |
1202 | continue; | |
1203 | if (af != AF_UNSPEC && af != xprt->xpt_local.ss_family) | |
1204 | continue; | |
156e6209 | 1205 | if (port != 0 && port != svc_xprt_local_port(xprt)) |
7fcb98d5 TT |
1206 | continue; |
1207 | found = xprt; | |
a217813f | 1208 | svc_xprt_get(xprt); |
7fcb98d5 TT |
1209 | break; |
1210 | } | |
1211 | spin_unlock_bh(&serv->sv_lock); | |
1212 | return found; | |
1213 | } | |
1214 | EXPORT_SYMBOL_GPL(svc_find_xprt); | |
9571af18 | 1215 | |
335c54bd CL |
1216 | static int svc_one_xprt_name(const struct svc_xprt *xprt, |
1217 | char *pos, int remaining) | |
1218 | { | |
1219 | int len; | |
1220 | ||
1221 | len = snprintf(pos, remaining, "%s %u\n", | |
1222 | xprt->xpt_class->xcl_name, | |
1223 | svc_xprt_local_port(xprt)); | |
1224 | if (len >= remaining) | |
1225 | return -ENAMETOOLONG; | |
1226 | return len; | |
1227 | } | |
1228 | ||
1229 | /** | |
1230 | * svc_xprt_names - format a buffer with a list of transport names | |
1231 | * @serv: pointer to an RPC service | |
1232 | * @buf: pointer to a buffer to be filled in | |
1233 | * @buflen: length of buffer to be filled in | |
1234 | * | |
1235 | * Fills in @buf with a string containing a list of transport names, | |
1236 | * each name terminated with '\n'. | |
1237 | * | |
1238 | * Returns positive length of the filled-in string on success; otherwise | |
1239 | * a negative errno value is returned if an error occurs. | |
9571af18 | 1240 | */ |
335c54bd | 1241 | int svc_xprt_names(struct svc_serv *serv, char *buf, const int buflen) |
9571af18 TT |
1242 | { |
1243 | struct svc_xprt *xprt; | |
335c54bd CL |
1244 | int len, totlen; |
1245 | char *pos; | |
9571af18 TT |
1246 | |
1247 | /* Sanity check args */ | |
1248 | if (!serv) | |
1249 | return 0; | |
1250 | ||
1251 | spin_lock_bh(&serv->sv_lock); | |
335c54bd CL |
1252 | |
1253 | pos = buf; | |
1254 | totlen = 0; | |
9571af18 | 1255 | list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list) { |
335c54bd CL |
1256 | len = svc_one_xprt_name(xprt, pos, buflen - totlen); |
1257 | if (len < 0) { | |
1258 | *buf = '\0'; | |
1259 | totlen = len; | |
1260 | } | |
1261 | if (len <= 0) | |
9571af18 | 1262 | break; |
335c54bd CL |
1263 | |
1264 | pos += len; | |
9571af18 TT |
1265 | totlen += len; |
1266 | } | |
335c54bd | 1267 | |
9571af18 TT |
1268 | spin_unlock_bh(&serv->sv_lock); |
1269 | return totlen; | |
1270 | } | |
1271 | EXPORT_SYMBOL_GPL(svc_xprt_names); | |
03cf6c9f GB |
1272 | |
1273 | ||
1274 | /*----------------------------------------------------------------------------*/ | |
1275 | ||
1276 | static void *svc_pool_stats_start(struct seq_file *m, loff_t *pos) | |
1277 | { | |
1278 | unsigned int pidx = (unsigned int)*pos; | |
1279 | struct svc_serv *serv = m->private; | |
1280 | ||
1281 | dprintk("svc_pool_stats_start, *pidx=%u\n", pidx); | |
1282 | ||
03cf6c9f GB |
1283 | if (!pidx) |
1284 | return SEQ_START_TOKEN; | |
1285 | return (pidx > serv->sv_nrpools ? NULL : &serv->sv_pools[pidx-1]); | |
1286 | } | |
1287 | ||
1288 | static void *svc_pool_stats_next(struct seq_file *m, void *p, loff_t *pos) | |
1289 | { | |
1290 | struct svc_pool *pool = p; | |
1291 | struct svc_serv *serv = m->private; | |
1292 | ||
1293 | dprintk("svc_pool_stats_next, *pos=%llu\n", *pos); | |
1294 | ||
1295 | if (p == SEQ_START_TOKEN) { | |
1296 | pool = &serv->sv_pools[0]; | |
1297 | } else { | |
1298 | unsigned int pidx = (pool - &serv->sv_pools[0]); | |
1299 | if (pidx < serv->sv_nrpools-1) | |
1300 | pool = &serv->sv_pools[pidx+1]; | |
1301 | else | |
1302 | pool = NULL; | |
1303 | } | |
1304 | ++*pos; | |
1305 | return pool; | |
1306 | } | |
1307 | ||
1308 | static void svc_pool_stats_stop(struct seq_file *m, void *p) | |
1309 | { | |
03cf6c9f GB |
1310 | } |
1311 | ||
1312 | static int svc_pool_stats_show(struct seq_file *m, void *p) | |
1313 | { | |
1314 | struct svc_pool *pool = p; | |
1315 | ||
1316 | if (p == SEQ_START_TOKEN) { | |
78c210ef | 1317 | seq_puts(m, "# pool packets-arrived sockets-enqueued threads-woken threads-timedout\n"); |
03cf6c9f GB |
1318 | return 0; |
1319 | } | |
1320 | ||
78c210ef | 1321 | seq_printf(m, "%u %lu %lu %lu %lu\n", |
03cf6c9f GB |
1322 | pool->sp_id, |
1323 | pool->sp_stats.packets, | |
1324 | pool->sp_stats.sockets_queued, | |
1325 | pool->sp_stats.threads_woken, | |
03cf6c9f GB |
1326 | pool->sp_stats.threads_timedout); |
1327 | ||
1328 | return 0; | |
1329 | } | |
1330 | ||
1331 | static const struct seq_operations svc_pool_stats_seq_ops = { | |
1332 | .start = svc_pool_stats_start, | |
1333 | .next = svc_pool_stats_next, | |
1334 | .stop = svc_pool_stats_stop, | |
1335 | .show = svc_pool_stats_show, | |
1336 | }; | |
1337 | ||
1338 | int svc_pool_stats_open(struct svc_serv *serv, struct file *file) | |
1339 | { | |
1340 | int err; | |
1341 | ||
1342 | err = seq_open(file, &svc_pool_stats_seq_ops); | |
1343 | if (!err) | |
1344 | ((struct seq_file *) file->private_data)->private = serv; | |
1345 | return err; | |
1346 | } | |
1347 | EXPORT_SYMBOL(svc_pool_stats_open); | |
1348 | ||
1349 | /*----------------------------------------------------------------------------*/ |