]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/net/sunrpc/xprt.c | |
3 | * | |
4 | * This is a generic RPC call interface supporting congestion avoidance, | |
5 | * and asynchronous calls. | |
6 | * | |
7 | * The interface works like this: | |
8 | * | |
9 | * - When a process places a call, it allocates a request slot if | |
10 | * one is available. Otherwise, it sleeps on the backlog queue | |
11 | * (xprt_reserve). | |
12 | * - Next, the caller puts together the RPC message, stuffs it into | |
55aa4f58 CL |
13 | * the request struct, and calls xprt_transmit(). |
14 | * - xprt_transmit sends the message and installs the caller on the | |
55ae1aab RL |
15 | * transport's wait list. At the same time, if a reply is expected, |
16 | * it installs a timer that is run after the packet's timeout has | |
17 | * expired. | |
1da177e4 | 18 | * - When a packet arrives, the data_ready handler walks the list of |
55aa4f58 | 19 | * pending requests for that transport. If a matching XID is found, the |
1da177e4 LT |
20 | * caller is woken up, and the timer removed. |
21 | * - When no reply arrives within the timeout interval, the timer is | |
22 | * fired by the kernel and runs xprt_timer(). It either adjusts the | |
23 | * timeout values (minor timeout) or wakes up the caller with a status | |
24 | * of -ETIMEDOUT. | |
25 | * - When the caller receives a notification from RPC that a reply arrived, | |
26 | * it should release the RPC slot, and process the reply. | |
27 | * If the call timed out, it may choose to retry the operation by | |
28 | * adjusting the initial timeout value, and simply calling rpc_call | |
29 | * again. | |
30 | * | |
31 | * Support for async RPC is done through a set of RPC-specific scheduling | |
32 | * primitives that `transparently' work for processes as well as async | |
33 | * tasks that rely on callbacks. | |
34 | * | |
35 | * Copyright (C) 1995-1997, Olaf Kirch <okir@monad.swb.de> | |
55aa4f58 CL |
36 | * |
37 | * Transport switch API copyright (C) 2005, Chuck Lever <cel@netapp.com> | |
1da177e4 LT |
38 | */ |
39 | ||
a246b010 CL |
40 | #include <linux/module.h> |
41 | ||
1da177e4 | 42 | #include <linux/types.h> |
a246b010 | 43 | #include <linux/interrupt.h> |
1da177e4 | 44 | #include <linux/workqueue.h> |
bf3fcf89 | 45 | #include <linux/net.h> |
ff839970 | 46 | #include <linux/ktime.h> |
1da177e4 | 47 | |
a246b010 | 48 | #include <linux/sunrpc/clnt.h> |
11c556b3 | 49 | #include <linux/sunrpc/metrics.h> |
c9acb42e | 50 | #include <linux/sunrpc/bc_xprt.h> |
1da177e4 | 51 | |
55ae1aab RL |
52 | #include "sunrpc.h" |
53 | ||
1da177e4 LT |
54 | /* |
55 | * Local variables | |
56 | */ | |
57 | ||
58 | #ifdef RPC_DEBUG | |
1da177e4 LT |
59 | # define RPCDBG_FACILITY RPCDBG_XPRT |
60 | #endif | |
61 | ||
1da177e4 LT |
62 | /* |
63 | * Local functions | |
64 | */ | |
21de0a95 | 65 | static void xprt_init(struct rpc_xprt *xprt, struct net *net); |
1da177e4 | 66 | static void xprt_request_init(struct rpc_task *, struct rpc_xprt *); |
1da177e4 | 67 | static void xprt_connect_status(struct rpc_task *task); |
1da177e4 LT |
68 | static int __xprt_get_cong(struct rpc_xprt *, struct rpc_task *); |
69 | ||
5ba03e82 | 70 | static DEFINE_SPINLOCK(xprt_list_lock); |
81c098af TT |
71 | static LIST_HEAD(xprt_list); |
72 | ||
555ee3af CL |
73 | /* |
74 | * The transport code maintains an estimate on the maximum number of out- | |
75 | * standing RPC requests, using a smoothed version of the congestion | |
76 | * avoidance implemented in 44BSD. This is basically the Van Jacobson | |
77 | * congestion algorithm: If a retransmit occurs, the congestion window is | |
78 | * halved; otherwise, it is incremented by 1/cwnd when | |
79 | * | |
80 | * - a reply is received and | |
81 | * - a full number of requests are outstanding and | |
82 | * - the congestion window hasn't been updated recently. | |
83 | */ | |
84 | #define RPC_CWNDSHIFT (8U) | |
85 | #define RPC_CWNDSCALE (1U << RPC_CWNDSHIFT) | |
86 | #define RPC_INITCWND RPC_CWNDSCALE | |
87 | #define RPC_MAXCWND(xprt) ((xprt)->max_reqs << RPC_CWNDSHIFT) | |
88 | ||
89 | #define RPCXPRT_CONGESTED(xprt) ((xprt)->cong >= (xprt)->cwnd) | |
1da177e4 | 90 | |
81c098af TT |
91 | /** |
92 | * xprt_register_transport - register a transport implementation | |
93 | * @transport: transport to register | |
94 | * | |
95 | * If a transport implementation is loaded as a kernel module, it can | |
96 | * call this interface to make itself known to the RPC client. | |
97 | * | |
98 | * Returns: | |
99 | * 0: transport successfully registered | |
100 | * -EEXIST: transport already registered | |
101 | * -EINVAL: transport module being unloaded | |
102 | */ | |
103 | int xprt_register_transport(struct xprt_class *transport) | |
104 | { | |
105 | struct xprt_class *t; | |
106 | int result; | |
107 | ||
108 | result = -EEXIST; | |
109 | spin_lock(&xprt_list_lock); | |
110 | list_for_each_entry(t, &xprt_list, list) { | |
111 | /* don't register the same transport class twice */ | |
4fa016eb | 112 | if (t->ident == transport->ident) |
81c098af TT |
113 | goto out; |
114 | } | |
115 | ||
c9f6cde6 DL |
116 | list_add_tail(&transport->list, &xprt_list); |
117 | printk(KERN_INFO "RPC: Registered %s transport module.\n", | |
118 | transport->name); | |
119 | result = 0; | |
81c098af TT |
120 | |
121 | out: | |
122 | spin_unlock(&xprt_list_lock); | |
123 | return result; | |
124 | } | |
125 | EXPORT_SYMBOL_GPL(xprt_register_transport); | |
126 | ||
127 | /** | |
128 | * xprt_unregister_transport - unregister a transport implementation | |
65b6e42c | 129 | * @transport: transport to unregister |
81c098af TT |
130 | * |
131 | * Returns: | |
132 | * 0: transport successfully unregistered | |
133 | * -ENOENT: transport never registered | |
134 | */ | |
135 | int xprt_unregister_transport(struct xprt_class *transport) | |
136 | { | |
137 | struct xprt_class *t; | |
138 | int result; | |
139 | ||
140 | result = 0; | |
141 | spin_lock(&xprt_list_lock); | |
142 | list_for_each_entry(t, &xprt_list, list) { | |
143 | if (t == transport) { | |
144 | printk(KERN_INFO | |
145 | "RPC: Unregistered %s transport module.\n", | |
146 | transport->name); | |
147 | list_del_init(&transport->list); | |
81c098af TT |
148 | goto out; |
149 | } | |
150 | } | |
151 | result = -ENOENT; | |
152 | ||
153 | out: | |
154 | spin_unlock(&xprt_list_lock); | |
155 | return result; | |
156 | } | |
157 | EXPORT_SYMBOL_GPL(xprt_unregister_transport); | |
158 | ||
441e3e24 TT |
159 | /** |
160 | * xprt_load_transport - load a transport implementation | |
161 | * @transport_name: transport to load | |
162 | * | |
163 | * Returns: | |
164 | * 0: transport successfully loaded | |
165 | * -ENOENT: transport module not available | |
166 | */ | |
167 | int xprt_load_transport(const char *transport_name) | |
168 | { | |
169 | struct xprt_class *t; | |
441e3e24 TT |
170 | int result; |
171 | ||
172 | result = 0; | |
173 | spin_lock(&xprt_list_lock); | |
174 | list_for_each_entry(t, &xprt_list, list) { | |
175 | if (strcmp(t->name, transport_name) == 0) { | |
176 | spin_unlock(&xprt_list_lock); | |
177 | goto out; | |
178 | } | |
179 | } | |
180 | spin_unlock(&xprt_list_lock); | |
ef7ffe8f | 181 | result = request_module("xprt%s", transport_name); |
441e3e24 TT |
182 | out: |
183 | return result; | |
184 | } | |
185 | EXPORT_SYMBOL_GPL(xprt_load_transport); | |
186 | ||
12a80469 CL |
187 | /** |
188 | * xprt_reserve_xprt - serialize write access to transports | |
189 | * @task: task that is requesting access to the transport | |
177c27bf | 190 | * @xprt: pointer to the target transport |
12a80469 CL |
191 | * |
192 | * This prevents mixing the payload of separate requests, and prevents | |
193 | * transport connects from colliding with writes. No congestion control | |
194 | * is provided. | |
195 | */ | |
43cedbf0 | 196 | int xprt_reserve_xprt(struct rpc_xprt *xprt, struct rpc_task *task) |
12a80469 | 197 | { |
12a80469 | 198 | struct rpc_rqst *req = task->tk_rqstp; |
34006cee | 199 | int priority; |
12a80469 CL |
200 | |
201 | if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) { | |
202 | if (task == xprt->snd_task) | |
203 | return 1; | |
12a80469 CL |
204 | goto out_sleep; |
205 | } | |
206 | xprt->snd_task = task; | |
43cedbf0 TM |
207 | if (req != NULL) { |
208 | req->rq_bytes_sent = 0; | |
209 | req->rq_ntrans++; | |
210 | } | |
4d4a76f3 | 211 | |
12a80469 CL |
212 | return 1; |
213 | ||
214 | out_sleep: | |
46121cf7 | 215 | dprintk("RPC: %5u failed to lock transport %p\n", |
12a80469 CL |
216 | task->tk_pid, xprt); |
217 | task->tk_timeout = 0; | |
218 | task->tk_status = -EAGAIN; | |
34006cee TM |
219 | if (req == NULL) |
220 | priority = RPC_PRIORITY_LOW; | |
221 | else if (!req->rq_ntrans) | |
222 | priority = RPC_PRIORITY_NORMAL; | |
12a80469 | 223 | else |
34006cee TM |
224 | priority = RPC_PRIORITY_HIGH; |
225 | rpc_sleep_on_priority(&xprt->sending, task, NULL, priority); | |
12a80469 CL |
226 | return 0; |
227 | } | |
12444809 | 228 | EXPORT_SYMBOL_GPL(xprt_reserve_xprt); |
12a80469 | 229 | |
632e3bdc TM |
230 | static void xprt_clear_locked(struct rpc_xprt *xprt) |
231 | { | |
232 | xprt->snd_task = NULL; | |
233 | if (!test_bit(XPRT_CLOSE_WAIT, &xprt->state) || xprt->shutdown) { | |
234 | smp_mb__before_clear_bit(); | |
235 | clear_bit(XPRT_LOCKED, &xprt->state); | |
236 | smp_mb__after_clear_bit(); | |
237 | } else | |
c1384c9c | 238 | queue_work(rpciod_workqueue, &xprt->task_cleanup); |
632e3bdc TM |
239 | } |
240 | ||
1da177e4 | 241 | /* |
12a80469 CL |
242 | * xprt_reserve_xprt_cong - serialize write access to transports |
243 | * @task: task that is requesting access to the transport | |
244 | * | |
245 | * Same as xprt_reserve_xprt, but Van Jacobson congestion control is | |
246 | * integrated into the decision of whether a request is allowed to be | |
247 | * woken up and given access to the transport. | |
1da177e4 | 248 | */ |
43cedbf0 | 249 | int xprt_reserve_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task) |
1da177e4 LT |
250 | { |
251 | struct rpc_rqst *req = task->tk_rqstp; | |
34006cee | 252 | int priority; |
1da177e4 | 253 | |
2226feb6 | 254 | if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) { |
1da177e4 LT |
255 | if (task == xprt->snd_task) |
256 | return 1; | |
1da177e4 LT |
257 | goto out_sleep; |
258 | } | |
43cedbf0 TM |
259 | if (req == NULL) { |
260 | xprt->snd_task = task; | |
261 | return 1; | |
262 | } | |
12a80469 | 263 | if (__xprt_get_cong(xprt, task)) { |
1da177e4 | 264 | xprt->snd_task = task; |
43cedbf0 TM |
265 | req->rq_bytes_sent = 0; |
266 | req->rq_ntrans++; | |
1da177e4 LT |
267 | return 1; |
268 | } | |
632e3bdc | 269 | xprt_clear_locked(xprt); |
1da177e4 | 270 | out_sleep: |
46121cf7 | 271 | dprintk("RPC: %5u failed to lock transport %p\n", task->tk_pid, xprt); |
1da177e4 LT |
272 | task->tk_timeout = 0; |
273 | task->tk_status = -EAGAIN; | |
34006cee TM |
274 | if (req == NULL) |
275 | priority = RPC_PRIORITY_LOW; | |
276 | else if (!req->rq_ntrans) | |
277 | priority = RPC_PRIORITY_NORMAL; | |
1da177e4 | 278 | else |
34006cee TM |
279 | priority = RPC_PRIORITY_HIGH; |
280 | rpc_sleep_on_priority(&xprt->sending, task, NULL, priority); | |
1da177e4 LT |
281 | return 0; |
282 | } | |
12444809 | 283 | EXPORT_SYMBOL_GPL(xprt_reserve_xprt_cong); |
1da177e4 | 284 | |
12a80469 | 285 | static inline int xprt_lock_write(struct rpc_xprt *xprt, struct rpc_task *task) |
1da177e4 LT |
286 | { |
287 | int retval; | |
288 | ||
4a0f8c04 | 289 | spin_lock_bh(&xprt->transport_lock); |
43cedbf0 | 290 | retval = xprt->ops->reserve_xprt(xprt, task); |
4a0f8c04 | 291 | spin_unlock_bh(&xprt->transport_lock); |
1da177e4 LT |
292 | return retval; |
293 | } | |
294 | ||
961a828d | 295 | static bool __xprt_lock_write_func(struct rpc_task *task, void *data) |
49e9a890 | 296 | { |
961a828d | 297 | struct rpc_xprt *xprt = data; |
49e9a890 CL |
298 | struct rpc_rqst *req; |
299 | ||
49e9a890 CL |
300 | req = task->tk_rqstp; |
301 | xprt->snd_task = task; | |
302 | if (req) { | |
303 | req->rq_bytes_sent = 0; | |
304 | req->rq_ntrans++; | |
305 | } | |
961a828d TM |
306 | return true; |
307 | } | |
49e9a890 | 308 | |
961a828d TM |
309 | static void __xprt_lock_write_next(struct rpc_xprt *xprt) |
310 | { | |
311 | if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) | |
312 | return; | |
313 | ||
314 | if (rpc_wake_up_first(&xprt->sending, __xprt_lock_write_func, xprt)) | |
315 | return; | |
632e3bdc | 316 | xprt_clear_locked(xprt); |
49e9a890 CL |
317 | } |
318 | ||
961a828d | 319 | static bool __xprt_lock_write_cong_func(struct rpc_task *task, void *data) |
1da177e4 | 320 | { |
961a828d | 321 | struct rpc_xprt *xprt = data; |
43cedbf0 | 322 | struct rpc_rqst *req; |
1da177e4 | 323 | |
43cedbf0 TM |
324 | req = task->tk_rqstp; |
325 | if (req == NULL) { | |
326 | xprt->snd_task = task; | |
961a828d | 327 | return true; |
43cedbf0 | 328 | } |
49e9a890 | 329 | if (__xprt_get_cong(xprt, task)) { |
1da177e4 | 330 | xprt->snd_task = task; |
43cedbf0 TM |
331 | req->rq_bytes_sent = 0; |
332 | req->rq_ntrans++; | |
961a828d | 333 | return true; |
1da177e4 | 334 | } |
961a828d TM |
335 | return false; |
336 | } | |
337 | ||
338 | static void __xprt_lock_write_next_cong(struct rpc_xprt *xprt) | |
339 | { | |
340 | if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) | |
341 | return; | |
342 | if (RPCXPRT_CONGESTED(xprt)) | |
343 | goto out_unlock; | |
344 | if (rpc_wake_up_first(&xprt->sending, __xprt_lock_write_cong_func, xprt)) | |
345 | return; | |
1da177e4 | 346 | out_unlock: |
632e3bdc | 347 | xprt_clear_locked(xprt); |
1da177e4 LT |
348 | } |
349 | ||
49e9a890 CL |
350 | /** |
351 | * xprt_release_xprt - allow other requests to use a transport | |
352 | * @xprt: transport with other tasks potentially waiting | |
353 | * @task: task that is releasing access to the transport | |
354 | * | |
355 | * Note that "task" can be NULL. No congestion control is provided. | |
1da177e4 | 356 | */ |
49e9a890 | 357 | void xprt_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task) |
1da177e4 LT |
358 | { |
359 | if (xprt->snd_task == task) { | |
632e3bdc | 360 | xprt_clear_locked(xprt); |
1da177e4 LT |
361 | __xprt_lock_write_next(xprt); |
362 | } | |
363 | } | |
12444809 | 364 | EXPORT_SYMBOL_GPL(xprt_release_xprt); |
1da177e4 | 365 | |
49e9a890 CL |
366 | /** |
367 | * xprt_release_xprt_cong - allow other requests to use a transport | |
368 | * @xprt: transport with other tasks potentially waiting | |
369 | * @task: task that is releasing access to the transport | |
370 | * | |
371 | * Note that "task" can be NULL. Another task is awoken to use the | |
372 | * transport if the transport's congestion window allows it. | |
373 | */ | |
374 | void xprt_release_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task) | |
375 | { | |
376 | if (xprt->snd_task == task) { | |
632e3bdc | 377 | xprt_clear_locked(xprt); |
49e9a890 CL |
378 | __xprt_lock_write_next_cong(xprt); |
379 | } | |
380 | } | |
12444809 | 381 | EXPORT_SYMBOL_GPL(xprt_release_xprt_cong); |
49e9a890 CL |
382 | |
383 | static inline void xprt_release_write(struct rpc_xprt *xprt, struct rpc_task *task) | |
1da177e4 | 384 | { |
4a0f8c04 | 385 | spin_lock_bh(&xprt->transport_lock); |
49e9a890 | 386 | xprt->ops->release_xprt(xprt, task); |
4a0f8c04 | 387 | spin_unlock_bh(&xprt->transport_lock); |
1da177e4 LT |
388 | } |
389 | ||
1da177e4 LT |
390 | /* |
391 | * Van Jacobson congestion avoidance. Check if the congestion window | |
392 | * overflowed. Put the task to sleep if this is the case. | |
393 | */ | |
394 | static int | |
395 | __xprt_get_cong(struct rpc_xprt *xprt, struct rpc_task *task) | |
396 | { | |
397 | struct rpc_rqst *req = task->tk_rqstp; | |
398 | ||
399 | if (req->rq_cong) | |
400 | return 1; | |
46121cf7 | 401 | dprintk("RPC: %5u xprt_cwnd_limited cong = %lu cwnd = %lu\n", |
1da177e4 LT |
402 | task->tk_pid, xprt->cong, xprt->cwnd); |
403 | if (RPCXPRT_CONGESTED(xprt)) | |
404 | return 0; | |
405 | req->rq_cong = 1; | |
406 | xprt->cong += RPC_CWNDSCALE; | |
407 | return 1; | |
408 | } | |
409 | ||
410 | /* | |
411 | * Adjust the congestion window, and wake up the next task | |
412 | * that has been sleeping due to congestion | |
413 | */ | |
414 | static void | |
415 | __xprt_put_cong(struct rpc_xprt *xprt, struct rpc_rqst *req) | |
416 | { | |
417 | if (!req->rq_cong) | |
418 | return; | |
419 | req->rq_cong = 0; | |
420 | xprt->cong -= RPC_CWNDSCALE; | |
49e9a890 | 421 | __xprt_lock_write_next_cong(xprt); |
1da177e4 LT |
422 | } |
423 | ||
a58dd398 CL |
424 | /** |
425 | * xprt_release_rqst_cong - housekeeping when request is complete | |
426 | * @task: RPC request that recently completed | |
427 | * | |
428 | * Useful for transports that require congestion control. | |
429 | */ | |
430 | void xprt_release_rqst_cong(struct rpc_task *task) | |
431 | { | |
432 | __xprt_put_cong(task->tk_xprt, task->tk_rqstp); | |
433 | } | |
12444809 | 434 | EXPORT_SYMBOL_GPL(xprt_release_rqst_cong); |
a58dd398 | 435 | |
46c0ee8b CL |
436 | /** |
437 | * xprt_adjust_cwnd - adjust transport congestion window | |
438 | * @task: recently completed RPC request used to adjust window | |
439 | * @result: result code of completed RPC request | |
440 | * | |
1da177e4 LT |
441 | * We use a time-smoothed congestion estimator to avoid heavy oscillation. |
442 | */ | |
46c0ee8b | 443 | void xprt_adjust_cwnd(struct rpc_task *task, int result) |
1da177e4 | 444 | { |
46c0ee8b CL |
445 | struct rpc_rqst *req = task->tk_rqstp; |
446 | struct rpc_xprt *xprt = task->tk_xprt; | |
447 | unsigned long cwnd = xprt->cwnd; | |
1da177e4 | 448 | |
1da177e4 LT |
449 | if (result >= 0 && cwnd <= xprt->cong) { |
450 | /* The (cwnd >> 1) term makes sure | |
451 | * the result gets rounded properly. */ | |
452 | cwnd += (RPC_CWNDSCALE * RPC_CWNDSCALE + (cwnd >> 1)) / cwnd; | |
453 | if (cwnd > RPC_MAXCWND(xprt)) | |
454 | cwnd = RPC_MAXCWND(xprt); | |
49e9a890 | 455 | __xprt_lock_write_next_cong(xprt); |
1da177e4 LT |
456 | } else if (result == -ETIMEDOUT) { |
457 | cwnd >>= 1; | |
458 | if (cwnd < RPC_CWNDSCALE) | |
459 | cwnd = RPC_CWNDSCALE; | |
460 | } | |
46121cf7 | 461 | dprintk("RPC: cong %ld, cwnd was %ld, now %ld\n", |
1da177e4 LT |
462 | xprt->cong, xprt->cwnd, cwnd); |
463 | xprt->cwnd = cwnd; | |
46c0ee8b | 464 | __xprt_put_cong(xprt, req); |
1da177e4 | 465 | } |
12444809 | 466 | EXPORT_SYMBOL_GPL(xprt_adjust_cwnd); |
1da177e4 | 467 | |
44fbac22 CL |
468 | /** |
469 | * xprt_wake_pending_tasks - wake all tasks on a transport's pending queue | |
470 | * @xprt: transport with waiting tasks | |
471 | * @status: result code to plant in each task before waking it | |
472 | * | |
473 | */ | |
474 | void xprt_wake_pending_tasks(struct rpc_xprt *xprt, int status) | |
475 | { | |
476 | if (status < 0) | |
477 | rpc_wake_up_status(&xprt->pending, status); | |
478 | else | |
479 | rpc_wake_up(&xprt->pending); | |
480 | } | |
12444809 | 481 | EXPORT_SYMBOL_GPL(xprt_wake_pending_tasks); |
44fbac22 | 482 | |
c7b2cae8 CL |
483 | /** |
484 | * xprt_wait_for_buffer_space - wait for transport output buffer to clear | |
485 | * @task: task to be put to sleep | |
0b80ae42 | 486 | * @action: function pointer to be executed after wait |
c7b2cae8 | 487 | */ |
b6ddf64f | 488 | void xprt_wait_for_buffer_space(struct rpc_task *task, rpc_action action) |
c7b2cae8 CL |
489 | { |
490 | struct rpc_rqst *req = task->tk_rqstp; | |
491 | struct rpc_xprt *xprt = req->rq_xprt; | |
492 | ||
493 | task->tk_timeout = req->rq_timeout; | |
b6ddf64f | 494 | rpc_sleep_on(&xprt->pending, task, action); |
c7b2cae8 | 495 | } |
12444809 | 496 | EXPORT_SYMBOL_GPL(xprt_wait_for_buffer_space); |
c7b2cae8 CL |
497 | |
498 | /** | |
499 | * xprt_write_space - wake the task waiting for transport output buffer space | |
500 | * @xprt: transport with waiting tasks | |
501 | * | |
502 | * Can be called in a soft IRQ context, so xprt_write_space never sleeps. | |
503 | */ | |
504 | void xprt_write_space(struct rpc_xprt *xprt) | |
505 | { | |
506 | if (unlikely(xprt->shutdown)) | |
507 | return; | |
508 | ||
509 | spin_lock_bh(&xprt->transport_lock); | |
510 | if (xprt->snd_task) { | |
46121cf7 CL |
511 | dprintk("RPC: write space: waking waiting task on " |
512 | "xprt %p\n", xprt); | |
fda13939 | 513 | rpc_wake_up_queued_task(&xprt->pending, xprt->snd_task); |
c7b2cae8 CL |
514 | } |
515 | spin_unlock_bh(&xprt->transport_lock); | |
516 | } | |
12444809 | 517 | EXPORT_SYMBOL_GPL(xprt_write_space); |
c7b2cae8 | 518 | |
fe3aca29 CL |
519 | /** |
520 | * xprt_set_retrans_timeout_def - set a request's retransmit timeout | |
521 | * @task: task whose timeout is to be set | |
522 | * | |
523 | * Set a request's retransmit timeout based on the transport's | |
524 | * default timeout parameters. Used by transports that don't adjust | |
525 | * the retransmit timeout based on round-trip time estimation. | |
526 | */ | |
527 | void xprt_set_retrans_timeout_def(struct rpc_task *task) | |
528 | { | |
529 | task->tk_timeout = task->tk_rqstp->rq_timeout; | |
530 | } | |
12444809 | 531 | EXPORT_SYMBOL_GPL(xprt_set_retrans_timeout_def); |
fe3aca29 CL |
532 | |
533 | /* | |
534 | * xprt_set_retrans_timeout_rtt - set a request's retransmit timeout | |
535 | * @task: task whose timeout is to be set | |
cca5172a | 536 | * |
fe3aca29 CL |
537 | * Set a request's retransmit timeout using the RTT estimator. |
538 | */ | |
539 | void xprt_set_retrans_timeout_rtt(struct rpc_task *task) | |
540 | { | |
541 | int timer = task->tk_msg.rpc_proc->p_timer; | |
ba7392bb TM |
542 | struct rpc_clnt *clnt = task->tk_client; |
543 | struct rpc_rtt *rtt = clnt->cl_rtt; | |
fe3aca29 | 544 | struct rpc_rqst *req = task->tk_rqstp; |
ba7392bb | 545 | unsigned long max_timeout = clnt->cl_timeout->to_maxval; |
fe3aca29 CL |
546 | |
547 | task->tk_timeout = rpc_calc_rto(rtt, timer); | |
548 | task->tk_timeout <<= rpc_ntimeo(rtt, timer) + req->rq_retries; | |
549 | if (task->tk_timeout > max_timeout || task->tk_timeout == 0) | |
550 | task->tk_timeout = max_timeout; | |
551 | } | |
12444809 | 552 | EXPORT_SYMBOL_GPL(xprt_set_retrans_timeout_rtt); |
fe3aca29 | 553 | |
1da177e4 LT |
554 | static void xprt_reset_majortimeo(struct rpc_rqst *req) |
555 | { | |
ba7392bb | 556 | const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout; |
1da177e4 LT |
557 | |
558 | req->rq_majortimeo = req->rq_timeout; | |
559 | if (to->to_exponential) | |
560 | req->rq_majortimeo <<= to->to_retries; | |
561 | else | |
562 | req->rq_majortimeo += to->to_increment * to->to_retries; | |
563 | if (req->rq_majortimeo > to->to_maxval || req->rq_majortimeo == 0) | |
564 | req->rq_majortimeo = to->to_maxval; | |
565 | req->rq_majortimeo += jiffies; | |
566 | } | |
567 | ||
9903cd1c CL |
568 | /** |
569 | * xprt_adjust_timeout - adjust timeout values for next retransmit | |
570 | * @req: RPC request containing parameters to use for the adjustment | |
571 | * | |
1da177e4 LT |
572 | */ |
573 | int xprt_adjust_timeout(struct rpc_rqst *req) | |
574 | { | |
575 | struct rpc_xprt *xprt = req->rq_xprt; | |
ba7392bb | 576 | const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout; |
1da177e4 LT |
577 | int status = 0; |
578 | ||
579 | if (time_before(jiffies, req->rq_majortimeo)) { | |
580 | if (to->to_exponential) | |
581 | req->rq_timeout <<= 1; | |
582 | else | |
583 | req->rq_timeout += to->to_increment; | |
584 | if (to->to_maxval && req->rq_timeout >= to->to_maxval) | |
585 | req->rq_timeout = to->to_maxval; | |
586 | req->rq_retries++; | |
1da177e4 LT |
587 | } else { |
588 | req->rq_timeout = to->to_initval; | |
589 | req->rq_retries = 0; | |
590 | xprt_reset_majortimeo(req); | |
591 | /* Reset the RTT counters == "slow start" */ | |
4a0f8c04 | 592 | spin_lock_bh(&xprt->transport_lock); |
1da177e4 | 593 | rpc_init_rtt(req->rq_task->tk_client->cl_rtt, to->to_initval); |
4a0f8c04 | 594 | spin_unlock_bh(&xprt->transport_lock); |
1da177e4 LT |
595 | status = -ETIMEDOUT; |
596 | } | |
597 | ||
598 | if (req->rq_timeout == 0) { | |
599 | printk(KERN_WARNING "xprt_adjust_timeout: rq_timeout = 0!\n"); | |
600 | req->rq_timeout = 5 * HZ; | |
601 | } | |
602 | return status; | |
603 | } | |
604 | ||
65f27f38 | 605 | static void xprt_autoclose(struct work_struct *work) |
1da177e4 | 606 | { |
65f27f38 DH |
607 | struct rpc_xprt *xprt = |
608 | container_of(work, struct rpc_xprt, task_cleanup); | |
1da177e4 | 609 | |
a246b010 | 610 | xprt->ops->close(xprt); |
66af1e55 | 611 | clear_bit(XPRT_CLOSE_WAIT, &xprt->state); |
1da177e4 LT |
612 | xprt_release_write(xprt, NULL); |
613 | } | |
614 | ||
9903cd1c | 615 | /** |
62da3b24 | 616 | * xprt_disconnect_done - mark a transport as disconnected |
9903cd1c CL |
617 | * @xprt: transport to flag for disconnect |
618 | * | |
1da177e4 | 619 | */ |
62da3b24 | 620 | void xprt_disconnect_done(struct rpc_xprt *xprt) |
1da177e4 | 621 | { |
46121cf7 | 622 | dprintk("RPC: disconnected transport %p\n", xprt); |
4a0f8c04 | 623 | spin_lock_bh(&xprt->transport_lock); |
1da177e4 | 624 | xprt_clear_connected(xprt); |
2a491991 | 625 | xprt_wake_pending_tasks(xprt, -EAGAIN); |
4a0f8c04 | 626 | spin_unlock_bh(&xprt->transport_lock); |
1da177e4 | 627 | } |
62da3b24 | 628 | EXPORT_SYMBOL_GPL(xprt_disconnect_done); |
1da177e4 | 629 | |
66af1e55 TM |
630 | /** |
631 | * xprt_force_disconnect - force a transport to disconnect | |
632 | * @xprt: transport to disconnect | |
633 | * | |
634 | */ | |
635 | void xprt_force_disconnect(struct rpc_xprt *xprt) | |
636 | { | |
637 | /* Don't race with the test_bit() in xprt_clear_locked() */ | |
638 | spin_lock_bh(&xprt->transport_lock); | |
639 | set_bit(XPRT_CLOSE_WAIT, &xprt->state); | |
640 | /* Try to schedule an autoclose RPC call */ | |
641 | if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0) | |
642 | queue_work(rpciod_workqueue, &xprt->task_cleanup); | |
2a491991 | 643 | xprt_wake_pending_tasks(xprt, -EAGAIN); |
66af1e55 TM |
644 | spin_unlock_bh(&xprt->transport_lock); |
645 | } | |
66af1e55 | 646 | |
7c1d71cf TM |
647 | /** |
648 | * xprt_conditional_disconnect - force a transport to disconnect | |
649 | * @xprt: transport to disconnect | |
650 | * @cookie: 'connection cookie' | |
651 | * | |
652 | * This attempts to break the connection if and only if 'cookie' matches | |
653 | * the current transport 'connection cookie'. It ensures that we don't | |
654 | * try to break the connection more than once when we need to retransmit | |
655 | * a batch of RPC requests. | |
656 | * | |
657 | */ | |
658 | void xprt_conditional_disconnect(struct rpc_xprt *xprt, unsigned int cookie) | |
659 | { | |
660 | /* Don't race with the test_bit() in xprt_clear_locked() */ | |
661 | spin_lock_bh(&xprt->transport_lock); | |
662 | if (cookie != xprt->connect_cookie) | |
663 | goto out; | |
664 | if (test_bit(XPRT_CLOSING, &xprt->state) || !xprt_connected(xprt)) | |
665 | goto out; | |
666 | set_bit(XPRT_CLOSE_WAIT, &xprt->state); | |
667 | /* Try to schedule an autoclose RPC call */ | |
668 | if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0) | |
669 | queue_work(rpciod_workqueue, &xprt->task_cleanup); | |
2a491991 | 670 | xprt_wake_pending_tasks(xprt, -EAGAIN); |
7c1d71cf TM |
671 | out: |
672 | spin_unlock_bh(&xprt->transport_lock); | |
673 | } | |
674 | ||
1da177e4 LT |
675 | static void |
676 | xprt_init_autodisconnect(unsigned long data) | |
677 | { | |
678 | struct rpc_xprt *xprt = (struct rpc_xprt *)data; | |
679 | ||
4a0f8c04 | 680 | spin_lock(&xprt->transport_lock); |
1da177e4 LT |
681 | if (!list_empty(&xprt->recv) || xprt->shutdown) |
682 | goto out_abort; | |
2226feb6 | 683 | if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) |
1da177e4 | 684 | goto out_abort; |
4a0f8c04 | 685 | spin_unlock(&xprt->transport_lock); |
f75e6745 TM |
686 | set_bit(XPRT_CONNECTION_CLOSE, &xprt->state); |
687 | queue_work(rpciod_workqueue, &xprt->task_cleanup); | |
1da177e4 LT |
688 | return; |
689 | out_abort: | |
4a0f8c04 | 690 | spin_unlock(&xprt->transport_lock); |
1da177e4 LT |
691 | } |
692 | ||
9903cd1c CL |
693 | /** |
694 | * xprt_connect - schedule a transport connect operation | |
695 | * @task: RPC task that is requesting the connect | |
1da177e4 LT |
696 | * |
697 | */ | |
698 | void xprt_connect(struct rpc_task *task) | |
699 | { | |
700 | struct rpc_xprt *xprt = task->tk_xprt; | |
701 | ||
46121cf7 | 702 | dprintk("RPC: %5u xprt_connect xprt %p %s connected\n", task->tk_pid, |
1da177e4 LT |
703 | xprt, (xprt_connected(xprt) ? "is" : "is not")); |
704 | ||
ec739ef0 | 705 | if (!xprt_bound(xprt)) { |
01d37c42 | 706 | task->tk_status = -EAGAIN; |
1da177e4 LT |
707 | return; |
708 | } | |
709 | if (!xprt_lock_write(xprt, task)) | |
710 | return; | |
feb8ca37 TM |
711 | |
712 | if (test_and_clear_bit(XPRT_CLOSE_WAIT, &xprt->state)) | |
713 | xprt->ops->close(xprt); | |
714 | ||
1da177e4 | 715 | if (xprt_connected(xprt)) |
a246b010 CL |
716 | xprt_release_write(xprt, task); |
717 | else { | |
87e3c055 | 718 | task->tk_rqstp->rq_bytes_sent = 0; |
a8ce4a8f | 719 | task->tk_timeout = task->tk_rqstp->rq_timeout; |
5d00837b | 720 | rpc_sleep_on(&xprt->pending, task, xprt_connect_status); |
0b9e7943 TM |
721 | |
722 | if (test_bit(XPRT_CLOSING, &xprt->state)) | |
723 | return; | |
724 | if (xprt_test_and_set_connecting(xprt)) | |
725 | return; | |
262ca07d | 726 | xprt->stat.connect_start = jiffies; |
a246b010 | 727 | xprt->ops->connect(task); |
1da177e4 | 728 | } |
1da177e4 LT |
729 | } |
730 | ||
9903cd1c | 731 | static void xprt_connect_status(struct rpc_task *task) |
1da177e4 LT |
732 | { |
733 | struct rpc_xprt *xprt = task->tk_xprt; | |
734 | ||
cd983ef8 | 735 | if (task->tk_status == 0) { |
262ca07d CL |
736 | xprt->stat.connect_count++; |
737 | xprt->stat.connect_time += (long)jiffies - xprt->stat.connect_start; | |
46121cf7 | 738 | dprintk("RPC: %5u xprt_connect_status: connection established\n", |
1da177e4 LT |
739 | task->tk_pid); |
740 | return; | |
741 | } | |
742 | ||
1da177e4 | 743 | switch (task->tk_status) { |
2a491991 TM |
744 | case -EAGAIN: |
745 | dprintk("RPC: %5u xprt_connect_status: retrying\n", task->tk_pid); | |
23475d66 | 746 | break; |
1da177e4 | 747 | case -ETIMEDOUT: |
46121cf7 CL |
748 | dprintk("RPC: %5u xprt_connect_status: connect attempt timed " |
749 | "out\n", task->tk_pid); | |
1da177e4 LT |
750 | break; |
751 | default: | |
46121cf7 CL |
752 | dprintk("RPC: %5u xprt_connect_status: error %d connecting to " |
753 | "server %s\n", task->tk_pid, -task->tk_status, | |
754 | task->tk_client->cl_server); | |
23475d66 CL |
755 | xprt_release_write(xprt, task); |
756 | task->tk_status = -EIO; | |
1da177e4 | 757 | } |
1da177e4 LT |
758 | } |
759 | ||
9903cd1c CL |
760 | /** |
761 | * xprt_lookup_rqst - find an RPC request corresponding to an XID | |
762 | * @xprt: transport on which the original request was transmitted | |
763 | * @xid: RPC XID of incoming reply | |
764 | * | |
1da177e4 | 765 | */ |
d8ed029d | 766 | struct rpc_rqst *xprt_lookup_rqst(struct rpc_xprt *xprt, __be32 xid) |
1da177e4 | 767 | { |
8f3a6de3 | 768 | struct rpc_rqst *entry; |
1da177e4 | 769 | |
8f3a6de3 | 770 | list_for_each_entry(entry, &xprt->recv, rq_list) |
262ca07d CL |
771 | if (entry->rq_xid == xid) |
772 | return entry; | |
46121cf7 CL |
773 | |
774 | dprintk("RPC: xprt_lookup_rqst did not find xid %08x\n", | |
775 | ntohl(xid)); | |
262ca07d CL |
776 | xprt->stat.bad_xids++; |
777 | return NULL; | |
1da177e4 | 778 | } |
12444809 | 779 | EXPORT_SYMBOL_GPL(xprt_lookup_rqst); |
1da177e4 | 780 | |
bbc72cea | 781 | static void xprt_update_rtt(struct rpc_task *task) |
1570c1e4 CL |
782 | { |
783 | struct rpc_rqst *req = task->tk_rqstp; | |
784 | struct rpc_rtt *rtt = task->tk_client->cl_rtt; | |
785 | unsigned timer = task->tk_msg.rpc_proc->p_timer; | |
d60dbb20 | 786 | long m = usecs_to_jiffies(ktime_to_us(req->rq_rtt)); |
1570c1e4 CL |
787 | |
788 | if (timer) { | |
789 | if (req->rq_ntrans == 1) | |
ff839970 | 790 | rpc_update_rtt(rtt, timer, m); |
1570c1e4 CL |
791 | rpc_set_timeo(rtt, timer, req->rq_ntrans - 1); |
792 | } | |
793 | } | |
794 | ||
9903cd1c CL |
795 | /** |
796 | * xprt_complete_rqst - called when reply processing is complete | |
1570c1e4 | 797 | * @task: RPC request that recently completed |
9903cd1c CL |
798 | * @copied: actual number of bytes received from the transport |
799 | * | |
1570c1e4 | 800 | * Caller holds transport lock. |
1da177e4 | 801 | */ |
1570c1e4 | 802 | void xprt_complete_rqst(struct rpc_task *task, int copied) |
1da177e4 | 803 | { |
1570c1e4 | 804 | struct rpc_rqst *req = task->tk_rqstp; |
fda13939 | 805 | struct rpc_xprt *xprt = req->rq_xprt; |
1da177e4 | 806 | |
1570c1e4 CL |
807 | dprintk("RPC: %5u xid %08x complete (%d bytes received)\n", |
808 | task->tk_pid, ntohl(req->rq_xid), copied); | |
1da177e4 | 809 | |
fda13939 | 810 | xprt->stat.recvs++; |
d60dbb20 | 811 | req->rq_rtt = ktime_sub(ktime_get(), req->rq_xtime); |
bbc72cea CL |
812 | if (xprt->ops->timer != NULL) |
813 | xprt_update_rtt(task); | |
ef759a2e | 814 | |
1da177e4 | 815 | list_del_init(&req->rq_list); |
1e799b67 | 816 | req->rq_private_buf.len = copied; |
dd2b63d0 RL |
817 | /* Ensure all writes are done before we update */ |
818 | /* req->rq_reply_bytes_recvd */ | |
43ac3f29 | 819 | smp_wmb(); |
dd2b63d0 | 820 | req->rq_reply_bytes_recvd = copied; |
fda13939 | 821 | rpc_wake_up_queued_task(&xprt->pending, task); |
1da177e4 | 822 | } |
12444809 | 823 | EXPORT_SYMBOL_GPL(xprt_complete_rqst); |
1da177e4 | 824 | |
46c0ee8b | 825 | static void xprt_timer(struct rpc_task *task) |
1da177e4 | 826 | { |
46c0ee8b | 827 | struct rpc_rqst *req = task->tk_rqstp; |
1da177e4 LT |
828 | struct rpc_xprt *xprt = req->rq_xprt; |
829 | ||
5d00837b TM |
830 | if (task->tk_status != -ETIMEDOUT) |
831 | return; | |
46121cf7 | 832 | dprintk("RPC: %5u xprt_timer\n", task->tk_pid); |
1da177e4 | 833 | |
5d00837b | 834 | spin_lock_bh(&xprt->transport_lock); |
dd2b63d0 | 835 | if (!req->rq_reply_bytes_recvd) { |
46c0ee8b CL |
836 | if (xprt->ops->timer) |
837 | xprt->ops->timer(task); | |
5d00837b TM |
838 | } else |
839 | task->tk_status = 0; | |
840 | spin_unlock_bh(&xprt->transport_lock); | |
1da177e4 LT |
841 | } |
842 | ||
4cfc7e60 RI |
843 | static inline int xprt_has_timer(struct rpc_xprt *xprt) |
844 | { | |
845 | return xprt->idle_timeout != 0; | |
846 | } | |
847 | ||
9903cd1c CL |
848 | /** |
849 | * xprt_prepare_transmit - reserve the transport before sending a request | |
850 | * @task: RPC task about to send a request | |
851 | * | |
1da177e4 | 852 | */ |
9903cd1c | 853 | int xprt_prepare_transmit(struct rpc_task *task) |
1da177e4 LT |
854 | { |
855 | struct rpc_rqst *req = task->tk_rqstp; | |
856 | struct rpc_xprt *xprt = req->rq_xprt; | |
857 | int err = 0; | |
858 | ||
46121cf7 | 859 | dprintk("RPC: %5u xprt_prepare_transmit\n", task->tk_pid); |
1da177e4 | 860 | |
4a0f8c04 | 861 | spin_lock_bh(&xprt->transport_lock); |
dd2b63d0 RL |
862 | if (req->rq_reply_bytes_recvd && !req->rq_bytes_sent) { |
863 | err = req->rq_reply_bytes_recvd; | |
1da177e4 LT |
864 | goto out_unlock; |
865 | } | |
43cedbf0 | 866 | if (!xprt->ops->reserve_xprt(xprt, task)) |
1da177e4 | 867 | err = -EAGAIN; |
1da177e4 | 868 | out_unlock: |
4a0f8c04 | 869 | spin_unlock_bh(&xprt->transport_lock); |
1da177e4 LT |
870 | return err; |
871 | } | |
872 | ||
e0ab53de | 873 | void xprt_end_transmit(struct rpc_task *task) |
5e5ce5be | 874 | { |
343952fa | 875 | xprt_release_write(task->tk_rqstp->rq_xprt, task); |
5e5ce5be TM |
876 | } |
877 | ||
9903cd1c CL |
878 | /** |
879 | * xprt_transmit - send an RPC request on a transport | |
880 | * @task: controlling RPC task | |
881 | * | |
882 | * We have to copy the iovec because sendmsg fiddles with its contents. | |
883 | */ | |
884 | void xprt_transmit(struct rpc_task *task) | |
1da177e4 | 885 | { |
1da177e4 LT |
886 | struct rpc_rqst *req = task->tk_rqstp; |
887 | struct rpc_xprt *xprt = req->rq_xprt; | |
a246b010 | 888 | int status; |
1da177e4 | 889 | |
46121cf7 | 890 | dprintk("RPC: %5u xprt_transmit(%u)\n", task->tk_pid, req->rq_slen); |
1da177e4 | 891 | |
dd2b63d0 | 892 | if (!req->rq_reply_bytes_recvd) { |
55ae1aab RL |
893 | if (list_empty(&req->rq_list) && rpc_reply_expected(task)) { |
894 | /* | |
895 | * Add to the list only if we're expecting a reply | |
896 | */ | |
4a0f8c04 | 897 | spin_lock_bh(&xprt->transport_lock); |
1da177e4 LT |
898 | /* Update the softirq receive buffer */ |
899 | memcpy(&req->rq_private_buf, &req->rq_rcv_buf, | |
900 | sizeof(req->rq_private_buf)); | |
901 | /* Add request to the receive list */ | |
902 | list_add_tail(&req->rq_list, &xprt->recv); | |
4a0f8c04 | 903 | spin_unlock_bh(&xprt->transport_lock); |
1da177e4 | 904 | xprt_reset_majortimeo(req); |
0f9dc2b1 TM |
905 | /* Turn off autodisconnect */ |
906 | del_singleshot_timer_sync(&xprt->timer); | |
1da177e4 LT |
907 | } |
908 | } else if (!req->rq_bytes_sent) | |
909 | return; | |
910 | ||
7c1d71cf | 911 | req->rq_connect_cookie = xprt->connect_cookie; |
ff839970 | 912 | req->rq_xtime = ktime_get(); |
a246b010 | 913 | status = xprt->ops->send_request(task); |
c8485e4d TM |
914 | if (status != 0) { |
915 | task->tk_status = status; | |
916 | return; | |
917 | } | |
262ca07d | 918 | |
c8485e4d | 919 | dprintk("RPC: %5u xmit complete\n", task->tk_pid); |
468f8613 | 920 | task->tk_flags |= RPC_TASK_SENT; |
c8485e4d | 921 | spin_lock_bh(&xprt->transport_lock); |
262ca07d | 922 | |
c8485e4d | 923 | xprt->ops->set_retrans_timeout(task); |
262ca07d | 924 | |
c8485e4d TM |
925 | xprt->stat.sends++; |
926 | xprt->stat.req_u += xprt->stat.sends - xprt->stat.recvs; | |
927 | xprt->stat.bklog_u += xprt->backlog.qlen; | |
1da177e4 | 928 | |
c8485e4d TM |
929 | /* Don't race with disconnect */ |
930 | if (!xprt_connected(xprt)) | |
931 | task->tk_status = -ENOTCONN; | |
dd2b63d0 | 932 | else if (!req->rq_reply_bytes_recvd && rpc_reply_expected(task)) { |
55ae1aab RL |
933 | /* |
934 | * Sleep on the pending queue since | |
935 | * we're expecting a reply. | |
936 | */ | |
c8485e4d | 937 | rpc_sleep_on(&xprt->pending, task, xprt_timer); |
55ae1aab | 938 | } |
c8485e4d | 939 | spin_unlock_bh(&xprt->transport_lock); |
1da177e4 LT |
940 | } |
941 | ||
d9ba131d TM |
942 | static struct rpc_rqst *xprt_dynamic_alloc_slot(struct rpc_xprt *xprt, gfp_t gfp_flags) |
943 | { | |
944 | struct rpc_rqst *req = ERR_PTR(-EAGAIN); | |
945 | ||
946 | if (!atomic_add_unless(&xprt->num_reqs, 1, xprt->max_reqs)) | |
947 | goto out; | |
948 | req = kzalloc(sizeof(struct rpc_rqst), gfp_flags); | |
949 | if (req != NULL) | |
950 | goto out; | |
951 | atomic_dec(&xprt->num_reqs); | |
952 | req = ERR_PTR(-ENOMEM); | |
953 | out: | |
954 | return req; | |
955 | } | |
956 | ||
957 | static bool xprt_dynamic_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req) | |
958 | { | |
959 | if (atomic_add_unless(&xprt->num_reqs, -1, xprt->min_reqs)) { | |
960 | kfree(req); | |
961 | return true; | |
962 | } | |
963 | return false; | |
964 | } | |
965 | ||
ee5ebe85 | 966 | static void xprt_alloc_slot(struct rpc_task *task) |
1da177e4 LT |
967 | { |
968 | struct rpc_xprt *xprt = task->tk_xprt; | |
d9ba131d | 969 | struct rpc_rqst *req; |
1da177e4 | 970 | |
1da177e4 | 971 | if (!list_empty(&xprt->free)) { |
d9ba131d TM |
972 | req = list_entry(xprt->free.next, struct rpc_rqst, rq_list); |
973 | list_del(&req->rq_list); | |
974 | goto out_init_req; | |
975 | } | |
976 | req = xprt_dynamic_alloc_slot(xprt, GFP_NOWAIT); | |
977 | if (!IS_ERR(req)) | |
978 | goto out_init_req; | |
979 | switch (PTR_ERR(req)) { | |
980 | case -ENOMEM: | |
981 | rpc_delay(task, HZ >> 2); | |
982 | dprintk("RPC: dynamic allocation of request slot " | |
983 | "failed! Retrying\n"); | |
984 | break; | |
985 | case -EAGAIN: | |
986 | rpc_sleep_on(&xprt->backlog, task, NULL); | |
987 | dprintk("RPC: waiting for request slot\n"); | |
1da177e4 | 988 | } |
1da177e4 | 989 | task->tk_status = -EAGAIN; |
d9ba131d TM |
990 | return; |
991 | out_init_req: | |
992 | task->tk_status = 0; | |
993 | task->tk_rqstp = req; | |
994 | xprt_request_init(task, xprt); | |
1da177e4 LT |
995 | } |
996 | ||
ee5ebe85 TM |
997 | static void xprt_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req) |
998 | { | |
ee5ebe85 | 999 | spin_lock(&xprt->reserve_lock); |
c25573b5 TM |
1000 | if (!xprt_dynamic_free_slot(xprt, req)) { |
1001 | memset(req, 0, sizeof(*req)); /* mark unused */ | |
1002 | list_add(&req->rq_list, &xprt->free); | |
1003 | } | |
ee5ebe85 TM |
1004 | rpc_wake_up_next(&xprt->backlog); |
1005 | spin_unlock(&xprt->reserve_lock); | |
1006 | } | |
1007 | ||
21de0a95 TM |
1008 | static void xprt_free_all_slots(struct rpc_xprt *xprt) |
1009 | { | |
1010 | struct rpc_rqst *req; | |
1011 | while (!list_empty(&xprt->free)) { | |
1012 | req = list_first_entry(&xprt->free, struct rpc_rqst, rq_list); | |
1013 | list_del(&req->rq_list); | |
1014 | kfree(req); | |
1015 | } | |
1016 | } | |
1017 | ||
d9ba131d TM |
1018 | struct rpc_xprt *xprt_alloc(struct net *net, size_t size, |
1019 | unsigned int num_prealloc, | |
1020 | unsigned int max_alloc) | |
bd1722d4 PE |
1021 | { |
1022 | struct rpc_xprt *xprt; | |
21de0a95 TM |
1023 | struct rpc_rqst *req; |
1024 | int i; | |
bd1722d4 PE |
1025 | |
1026 | xprt = kzalloc(size, GFP_KERNEL); | |
1027 | if (xprt == NULL) | |
1028 | goto out; | |
1029 | ||
21de0a95 TM |
1030 | xprt_init(xprt, net); |
1031 | ||
1032 | for (i = 0; i < num_prealloc; i++) { | |
1033 | req = kzalloc(sizeof(struct rpc_rqst), GFP_KERNEL); | |
1034 | if (!req) | |
1035 | break; | |
1036 | list_add(&req->rq_list, &xprt->free); | |
1037 | } | |
1038 | if (i < num_prealloc) | |
bd1722d4 | 1039 | goto out_free; |
d9ba131d TM |
1040 | if (max_alloc > num_prealloc) |
1041 | xprt->max_reqs = max_alloc; | |
1042 | else | |
1043 | xprt->max_reqs = num_prealloc; | |
1044 | xprt->min_reqs = num_prealloc; | |
1045 | atomic_set(&xprt->num_reqs, num_prealloc); | |
bd1722d4 PE |
1046 | |
1047 | return xprt; | |
1048 | ||
1049 | out_free: | |
21de0a95 | 1050 | xprt_free(xprt); |
bd1722d4 PE |
1051 | out: |
1052 | return NULL; | |
1053 | } | |
1054 | EXPORT_SYMBOL_GPL(xprt_alloc); | |
1055 | ||
e204e621 PE |
1056 | void xprt_free(struct rpc_xprt *xprt) |
1057 | { | |
37aa2133 | 1058 | put_net(xprt->xprt_net); |
21de0a95 | 1059 | xprt_free_all_slots(xprt); |
e204e621 PE |
1060 | kfree(xprt); |
1061 | } | |
1062 | EXPORT_SYMBOL_GPL(xprt_free); | |
1063 | ||
9903cd1c CL |
1064 | /** |
1065 | * xprt_reserve - allocate an RPC request slot | |
1066 | * @task: RPC task requesting a slot allocation | |
1067 | * | |
1068 | * If no more slots are available, place the task on the transport's | |
1069 | * backlog queue. | |
1070 | */ | |
1071 | void xprt_reserve(struct rpc_task *task) | |
1da177e4 LT |
1072 | { |
1073 | struct rpc_xprt *xprt = task->tk_xprt; | |
1074 | ||
43cedbf0 TM |
1075 | task->tk_status = 0; |
1076 | if (task->tk_rqstp != NULL) | |
1077 | return; | |
1078 | ||
1079 | /* Note: grabbing the xprt_lock_write() here is not strictly needed, | |
1080 | * but ensures that we throttle new slot allocation if the transport | |
1081 | * is congested (e.g. if reconnecting or if we're out of socket | |
1082 | * write buffer space). | |
1083 | */ | |
1084 | task->tk_timeout = 0; | |
1085 | task->tk_status = -EAGAIN; | |
1086 | if (!xprt_lock_write(xprt, task)) | |
1087 | return; | |
1088 | ||
0065db32 | 1089 | spin_lock(&xprt->reserve_lock); |
ee5ebe85 | 1090 | xprt_alloc_slot(task); |
0065db32 | 1091 | spin_unlock(&xprt->reserve_lock); |
43cedbf0 | 1092 | xprt_release_write(xprt, task); |
1da177e4 LT |
1093 | } |
1094 | ||
d8ed029d | 1095 | static inline __be32 xprt_alloc_xid(struct rpc_xprt *xprt) |
1da177e4 | 1096 | { |
0eae88f3 | 1097 | return (__force __be32)xprt->xid++; |
1da177e4 LT |
1098 | } |
1099 | ||
1100 | static inline void xprt_init_xid(struct rpc_xprt *xprt) | |
1101 | { | |
bf3fcf89 | 1102 | xprt->xid = net_random(); |
1da177e4 LT |
1103 | } |
1104 | ||
9903cd1c | 1105 | static void xprt_request_init(struct rpc_task *task, struct rpc_xprt *xprt) |
1da177e4 LT |
1106 | { |
1107 | struct rpc_rqst *req = task->tk_rqstp; | |
1108 | ||
d9ba131d | 1109 | INIT_LIST_HEAD(&req->rq_list); |
ba7392bb | 1110 | req->rq_timeout = task->tk_client->cl_timeout->to_initval; |
1da177e4 LT |
1111 | req->rq_task = task; |
1112 | req->rq_xprt = xprt; | |
02107148 | 1113 | req->rq_buffer = NULL; |
1da177e4 | 1114 | req->rq_xid = xprt_alloc_xid(xprt); |
ead5e1c2 | 1115 | req->rq_release_snd_buf = NULL; |
da45828e | 1116 | xprt_reset_majortimeo(req); |
46121cf7 | 1117 | dprintk("RPC: %5u reserved req %p xid %08x\n", task->tk_pid, |
1da177e4 LT |
1118 | req, ntohl(req->rq_xid)); |
1119 | } | |
1120 | ||
9903cd1c CL |
1121 | /** |
1122 | * xprt_release - release an RPC request slot | |
1123 | * @task: task which is finished with the slot | |
1124 | * | |
1da177e4 | 1125 | */ |
9903cd1c | 1126 | void xprt_release(struct rpc_task *task) |
1da177e4 | 1127 | { |
55ae1aab | 1128 | struct rpc_xprt *xprt; |
1da177e4 LT |
1129 | struct rpc_rqst *req; |
1130 | ||
1131 | if (!(req = task->tk_rqstp)) | |
1132 | return; | |
55ae1aab | 1133 | |
55ae1aab | 1134 | xprt = req->rq_xprt; |
11c556b3 | 1135 | rpc_count_iostats(task); |
4a0f8c04 | 1136 | spin_lock_bh(&xprt->transport_lock); |
49e9a890 | 1137 | xprt->ops->release_xprt(xprt, task); |
a58dd398 CL |
1138 | if (xprt->ops->release_request) |
1139 | xprt->ops->release_request(task); | |
1da177e4 LT |
1140 | if (!list_empty(&req->rq_list)) |
1141 | list_del(&req->rq_list); | |
1142 | xprt->last_used = jiffies; | |
4cfc7e60 | 1143 | if (list_empty(&xprt->recv) && xprt_has_timer(xprt)) |
a246b010 | 1144 | mod_timer(&xprt->timer, |
03bf4b70 | 1145 | xprt->last_used + xprt->idle_timeout); |
4a0f8c04 | 1146 | spin_unlock_bh(&xprt->transport_lock); |
ee5ebe85 | 1147 | if (req->rq_buffer) |
55ae1aab | 1148 | xprt->ops->buf_free(req->rq_buffer); |
a17c2153 TM |
1149 | if (req->rq_cred != NULL) |
1150 | put_rpccred(req->rq_cred); | |
1da177e4 | 1151 | task->tk_rqstp = NULL; |
ead5e1c2 BF |
1152 | if (req->rq_release_snd_buf) |
1153 | req->rq_release_snd_buf(req); | |
55ae1aab | 1154 | |
46121cf7 | 1155 | dprintk("RPC: %5u release request %p\n", task->tk_pid, req); |
ee5ebe85 TM |
1156 | if (likely(!bc_prealloc(req))) |
1157 | xprt_free_slot(xprt, req); | |
1158 | else | |
c9acb42e | 1159 | xprt_free_bc_request(req); |
1da177e4 LT |
1160 | } |
1161 | ||
21de0a95 | 1162 | static void xprt_init(struct rpc_xprt *xprt, struct net *net) |
c2866763 | 1163 | { |
21de0a95 | 1164 | atomic_set(&xprt->count, 1); |
c2866763 CL |
1165 | |
1166 | spin_lock_init(&xprt->transport_lock); | |
1167 | spin_lock_init(&xprt->reserve_lock); | |
1168 | ||
1169 | INIT_LIST_HEAD(&xprt->free); | |
1170 | INIT_LIST_HEAD(&xprt->recv); | |
9e00abc3 | 1171 | #if defined(CONFIG_SUNRPC_BACKCHANNEL) |
f9acac1a RL |
1172 | spin_lock_init(&xprt->bc_pa_lock); |
1173 | INIT_LIST_HEAD(&xprt->bc_pa_list); | |
9e00abc3 | 1174 | #endif /* CONFIG_SUNRPC_BACKCHANNEL */ |
f9acac1a | 1175 | |
c2866763 CL |
1176 | xprt->last_used = jiffies; |
1177 | xprt->cwnd = RPC_INITCWND; | |
a509050b | 1178 | xprt->bind_index = 0; |
c2866763 CL |
1179 | |
1180 | rpc_init_wait_queue(&xprt->binding, "xprt_binding"); | |
1181 | rpc_init_wait_queue(&xprt->pending, "xprt_pending"); | |
34006cee | 1182 | rpc_init_priority_wait_queue(&xprt->sending, "xprt_sending"); |
c2866763 CL |
1183 | rpc_init_priority_wait_queue(&xprt->backlog, "xprt_backlog"); |
1184 | ||
c2866763 CL |
1185 | xprt_init_xid(xprt); |
1186 | ||
21de0a95 | 1187 | xprt->xprt_net = get_net(net); |
8d9266ff TM |
1188 | } |
1189 | ||
1190 | /** | |
1191 | * xprt_create_transport - create an RPC transport | |
1192 | * @args: rpc transport creation arguments | |
1193 | * | |
1194 | */ | |
1195 | struct rpc_xprt *xprt_create_transport(struct xprt_create *args) | |
1196 | { | |
1197 | struct rpc_xprt *xprt; | |
1198 | struct xprt_class *t; | |
1199 | ||
1200 | spin_lock(&xprt_list_lock); | |
1201 | list_for_each_entry(t, &xprt_list, list) { | |
1202 | if (t->ident == args->ident) { | |
1203 | spin_unlock(&xprt_list_lock); | |
1204 | goto found; | |
1205 | } | |
1206 | } | |
1207 | spin_unlock(&xprt_list_lock); | |
1208 | printk(KERN_ERR "RPC: transport (%d) not supported\n", args->ident); | |
1209 | return ERR_PTR(-EIO); | |
1210 | ||
1211 | found: | |
1212 | xprt = t->setup(args); | |
1213 | if (IS_ERR(xprt)) { | |
1214 | dprintk("RPC: xprt_create_transport: failed, %ld\n", | |
1215 | -PTR_ERR(xprt)); | |
21de0a95 | 1216 | goto out; |
8d9266ff | 1217 | } |
21de0a95 TM |
1218 | INIT_WORK(&xprt->task_cleanup, xprt_autoclose); |
1219 | if (xprt_has_timer(xprt)) | |
1220 | setup_timer(&xprt->timer, xprt_init_autodisconnect, | |
1221 | (unsigned long)xprt); | |
1222 | else | |
1223 | init_timer(&xprt->timer); | |
46121cf7 | 1224 | dprintk("RPC: created transport %p with %u slots\n", xprt, |
c2866763 | 1225 | xprt->max_reqs); |
21de0a95 | 1226 | out: |
c2866763 CL |
1227 | return xprt; |
1228 | } | |
1229 | ||
9903cd1c CL |
1230 | /** |
1231 | * xprt_destroy - destroy an RPC transport, killing off all requests. | |
a8de240a | 1232 | * @xprt: transport to destroy |
9903cd1c | 1233 | * |
1da177e4 | 1234 | */ |
a8de240a | 1235 | static void xprt_destroy(struct rpc_xprt *xprt) |
1da177e4 | 1236 | { |
46121cf7 | 1237 | dprintk("RPC: destroying transport %p\n", xprt); |
0065db32 TM |
1238 | xprt->shutdown = 1; |
1239 | del_timer_sync(&xprt->timer); | |
c8541ecd | 1240 | |
f6a1cc89 TM |
1241 | rpc_destroy_wait_queue(&xprt->binding); |
1242 | rpc_destroy_wait_queue(&xprt->pending); | |
1243 | rpc_destroy_wait_queue(&xprt->sending); | |
f6a1cc89 | 1244 | rpc_destroy_wait_queue(&xprt->backlog); |
c3ae62ae | 1245 | cancel_work_sync(&xprt->task_cleanup); |
c8541ecd CL |
1246 | /* |
1247 | * Tear down transport state and free the rpc_xprt | |
1248 | */ | |
a246b010 | 1249 | xprt->ops->destroy(xprt); |
6b6ca86b | 1250 | } |
1da177e4 | 1251 | |
6b6ca86b TM |
1252 | /** |
1253 | * xprt_put - release a reference to an RPC transport. | |
1254 | * @xprt: pointer to the transport | |
1255 | * | |
1256 | */ | |
1257 | void xprt_put(struct rpc_xprt *xprt) | |
1258 | { | |
a8de240a TM |
1259 | if (atomic_dec_and_test(&xprt->count)) |
1260 | xprt_destroy(xprt); | |
6b6ca86b TM |
1261 | } |
1262 | ||
1263 | /** | |
1264 | * xprt_get - return a reference to an RPC transport. | |
1265 | * @xprt: pointer to the transport | |
1266 | * | |
1267 | */ | |
1268 | struct rpc_xprt *xprt_get(struct rpc_xprt *xprt) | |
1269 | { | |
a8de240a TM |
1270 | if (atomic_inc_not_zero(&xprt->count)) |
1271 | return xprt; | |
1272 | return NULL; | |
1da177e4 | 1273 | } |