]>
Commit | Line | Data |
---|---|---|
922a8e9f TT |
1 | /* |
2 | * Copyright (c) 2004, 2005 Intel Corporation. All rights reserved. | |
3 | * Copyright (c) 2004 Topspin Corporation. All rights reserved. | |
4 | * Copyright (c) 2004, 2005 Voltaire Corporation. All rights reserved. | |
5 | * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved. | |
6 | * Copyright (c) 2005 Open Grid Computing, Inc. All rights reserved. | |
7 | * Copyright (c) 2005 Network Appliance, Inc. All rights reserved. | |
8 | * | |
9 | * This software is available to you under a choice of one of two | |
10 | * licenses. You may choose to be licensed under the terms of the GNU | |
11 | * General Public License (GPL) Version 2, available from the file | |
12 | * COPYING in the main directory of this source tree, or the | |
13 | * OpenIB.org BSD license below: | |
14 | * | |
15 | * Redistribution and use in source and binary forms, with or | |
16 | * without modification, are permitted provided that the following | |
17 | * conditions are met: | |
18 | * | |
19 | * - Redistributions of source code must retain the above | |
20 | * copyright notice, this list of conditions and the following | |
21 | * disclaimer. | |
22 | * | |
23 | * - Redistributions in binary form must reproduce the above | |
24 | * copyright notice, this list of conditions and the following | |
25 | * disclaimer in the documentation and/or other materials | |
26 | * provided with the distribution. | |
27 | * | |
28 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, | |
29 | * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF | |
30 | * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND | |
31 | * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS | |
32 | * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN | |
33 | * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN | |
34 | * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE | |
35 | * SOFTWARE. | |
36 | * | |
37 | */ | |
38 | #include <linux/dma-mapping.h> | |
39 | #include <linux/err.h> | |
40 | #include <linux/idr.h> | |
41 | #include <linux/interrupt.h> | |
922a8e9f | 42 | #include <linux/rbtree.h> |
d43c36dc | 43 | #include <linux/sched.h> |
922a8e9f TT |
44 | #include <linux/spinlock.h> |
45 | #include <linux/workqueue.h> | |
46 | #include <linux/completion.h> | |
47 | ||
48 | #include <rdma/iw_cm.h> | |
49 | #include <rdma/ib_addr.h> | |
50 | ||
51 | #include "iwcm.h" | |
52 | ||
53 | MODULE_AUTHOR("Tom Tucker"); | |
54 | MODULE_DESCRIPTION("iWARP CM"); | |
55 | MODULE_LICENSE("Dual BSD/GPL"); | |
56 | ||
57 | static struct workqueue_struct *iwcm_wq; | |
58 | struct iwcm_work { | |
59 | struct work_struct work; | |
60 | struct iwcm_id_private *cm_id; | |
61 | struct list_head list; | |
62 | struct iw_cm_event event; | |
63 | struct list_head free_list; | |
64 | }; | |
65 | ||
66 | /* | |
67 | * The following services provide a mechanism for pre-allocating iwcm_work | |
68 | * elements. The design pre-allocates them based on the cm_id type: | |
69 | * LISTENING IDS: Get enough elements preallocated to handle the | |
70 | * listen backlog. | |
71 | * ACTIVE IDS: 4: CONNECT_REPLY, ESTABLISHED, DISCONNECT, CLOSE | |
72 | * PASSIVE IDS: 3: ESTABLISHED, DISCONNECT, CLOSE | |
73 | * | |
74 | * Allocating them in connect and listen avoids having to deal | |
75 | * with allocation failures on the event upcall from the provider (which | |
76 | * is called in the interrupt context). | |
77 | * | |
78 | * One exception is when creating the cm_id for incoming connection requests. | |
79 | * There are two cases: | |
80 | * 1) in the event upcall, cm_event_handler(), for a listening cm_id. If | |
81 | * the backlog is exceeded, then no more connection request events will | |
82 | * be processed. cm_event_handler() returns -ENOMEM in this case. Its up | |
715a588f | 83 | * to the provider to reject the connection request. |
922a8e9f TT |
84 | * 2) in the connection request workqueue handler, cm_conn_req_handler(). |
85 | * If work elements cannot be allocated for the new connect request cm_id, | |
86 | * then IWCM will call the provider reject method. This is ok since | |
87 | * cm_conn_req_handler() runs in the workqueue thread context. | |
88 | */ | |
89 | ||
90 | static struct iwcm_work *get_work(struct iwcm_id_private *cm_id_priv) | |
91 | { | |
92 | struct iwcm_work *work; | |
93 | ||
94 | if (list_empty(&cm_id_priv->work_free_list)) | |
95 | return NULL; | |
96 | work = list_entry(cm_id_priv->work_free_list.next, struct iwcm_work, | |
97 | free_list); | |
98 | list_del_init(&work->free_list); | |
99 | return work; | |
100 | } | |
101 | ||
102 | static void put_work(struct iwcm_work *work) | |
103 | { | |
104 | list_add(&work->free_list, &work->cm_id->work_free_list); | |
105 | } | |
106 | ||
107 | static void dealloc_work_entries(struct iwcm_id_private *cm_id_priv) | |
108 | { | |
109 | struct list_head *e, *tmp; | |
110 | ||
111 | list_for_each_safe(e, tmp, &cm_id_priv->work_free_list) | |
112 | kfree(list_entry(e, struct iwcm_work, free_list)); | |
113 | } | |
114 | ||
115 | static int alloc_work_entries(struct iwcm_id_private *cm_id_priv, int count) | |
116 | { | |
117 | struct iwcm_work *work; | |
118 | ||
119 | BUG_ON(!list_empty(&cm_id_priv->work_free_list)); | |
120 | while (count--) { | |
121 | work = kmalloc(sizeof(struct iwcm_work), GFP_KERNEL); | |
122 | if (!work) { | |
123 | dealloc_work_entries(cm_id_priv); | |
124 | return -ENOMEM; | |
125 | } | |
126 | work->cm_id = cm_id_priv; | |
127 | INIT_LIST_HEAD(&work->list); | |
128 | put_work(work); | |
129 | } | |
130 | return 0; | |
131 | } | |
132 | ||
133 | /* | |
715a588f KK |
134 | * Save private data from incoming connection requests to |
135 | * iw_cm_event, so the low level driver doesn't have to. Adjust | |
922a8e9f TT |
136 | * the event ptr to point to the local copy. |
137 | */ | |
715a588f | 138 | static int copy_private_data(struct iw_cm_event *event) |
922a8e9f TT |
139 | { |
140 | void *p; | |
141 | ||
bed8bdfd | 142 | p = kmemdup(event->private_data, event->private_data_len, GFP_ATOMIC); |
922a8e9f TT |
143 | if (!p) |
144 | return -ENOMEM; | |
922a8e9f TT |
145 | event->private_data = p; |
146 | return 0; | |
147 | } | |
148 | ||
ebb90986 SW |
149 | static void free_cm_id(struct iwcm_id_private *cm_id_priv) |
150 | { | |
151 | dealloc_work_entries(cm_id_priv); | |
152 | kfree(cm_id_priv); | |
153 | } | |
154 | ||
922a8e9f | 155 | /* |
9ab1ffa8 KK |
156 | * Release a reference on cm_id. If the last reference is being |
157 | * released, enable the waiting thread (in iw_destroy_cm_id) to | |
158 | * get woken up, and return 1 if a thread is already waiting. | |
922a8e9f TT |
159 | */ |
160 | static int iwcm_deref_id(struct iwcm_id_private *cm_id_priv) | |
161 | { | |
922a8e9f TT |
162 | BUG_ON(atomic_read(&cm_id_priv->refcount)==0); |
163 | if (atomic_dec_and_test(&cm_id_priv->refcount)) { | |
164 | BUG_ON(!list_empty(&cm_id_priv->work_list)); | |
922a8e9f | 165 | complete(&cm_id_priv->destroy_comp); |
ebb90986 | 166 | return 1; |
922a8e9f TT |
167 | } |
168 | ||
ebb90986 | 169 | return 0; |
922a8e9f TT |
170 | } |
171 | ||
172 | static void add_ref(struct iw_cm_id *cm_id) | |
173 | { | |
174 | struct iwcm_id_private *cm_id_priv; | |
175 | cm_id_priv = container_of(cm_id, struct iwcm_id_private, id); | |
176 | atomic_inc(&cm_id_priv->refcount); | |
177 | } | |
178 | ||
179 | static void rem_ref(struct iw_cm_id *cm_id) | |
180 | { | |
181 | struct iwcm_id_private *cm_id_priv; | |
182 | cm_id_priv = container_of(cm_id, struct iwcm_id_private, id); | |
ebb90986 SW |
183 | if (iwcm_deref_id(cm_id_priv) && |
184 | test_bit(IWCM_F_CALLBACK_DESTROY, &cm_id_priv->flags)) { | |
185 | BUG_ON(!list_empty(&cm_id_priv->work_list)); | |
186 | free_cm_id(cm_id_priv); | |
187 | } | |
922a8e9f TT |
188 | } |
189 | ||
190 | static int cm_event_handler(struct iw_cm_id *cm_id, struct iw_cm_event *event); | |
191 | ||
192 | struct iw_cm_id *iw_create_cm_id(struct ib_device *device, | |
193 | iw_cm_handler cm_handler, | |
194 | void *context) | |
195 | { | |
196 | struct iwcm_id_private *cm_id_priv; | |
197 | ||
198 | cm_id_priv = kzalloc(sizeof(*cm_id_priv), GFP_KERNEL); | |
199 | if (!cm_id_priv) | |
200 | return ERR_PTR(-ENOMEM); | |
201 | ||
202 | cm_id_priv->state = IW_CM_STATE_IDLE; | |
203 | cm_id_priv->id.device = device; | |
204 | cm_id_priv->id.cm_handler = cm_handler; | |
205 | cm_id_priv->id.context = context; | |
206 | cm_id_priv->id.event_handler = cm_event_handler; | |
207 | cm_id_priv->id.add_ref = add_ref; | |
208 | cm_id_priv->id.rem_ref = rem_ref; | |
209 | spin_lock_init(&cm_id_priv->lock); | |
210 | atomic_set(&cm_id_priv->refcount, 1); | |
211 | init_waitqueue_head(&cm_id_priv->connect_wait); | |
212 | init_completion(&cm_id_priv->destroy_comp); | |
213 | INIT_LIST_HEAD(&cm_id_priv->work_list); | |
214 | INIT_LIST_HEAD(&cm_id_priv->work_free_list); | |
215 | ||
216 | return &cm_id_priv->id; | |
217 | } | |
218 | EXPORT_SYMBOL(iw_create_cm_id); | |
219 | ||
220 | ||
221 | static int iwcm_modify_qp_err(struct ib_qp *qp) | |
222 | { | |
223 | struct ib_qp_attr qp_attr; | |
224 | ||
225 | if (!qp) | |
226 | return -EINVAL; | |
227 | ||
228 | qp_attr.qp_state = IB_QPS_ERR; | |
229 | return ib_modify_qp(qp, &qp_attr, IB_QP_STATE); | |
230 | } | |
231 | ||
232 | /* | |
233 | * This is really the RDMAC CLOSING state. It is most similar to the | |
234 | * IB SQD QP state. | |
235 | */ | |
236 | static int iwcm_modify_qp_sqd(struct ib_qp *qp) | |
237 | { | |
238 | struct ib_qp_attr qp_attr; | |
239 | ||
240 | BUG_ON(qp == NULL); | |
241 | qp_attr.qp_state = IB_QPS_SQD; | |
242 | return ib_modify_qp(qp, &qp_attr, IB_QP_STATE); | |
243 | } | |
244 | ||
245 | /* | |
246 | * CM_ID <-- CLOSING | |
247 | * | |
715a588f | 248 | * Block if a passive or active connection is currently being processed. Then |
922a8e9f TT |
249 | * process the event as follows: |
250 | * - If we are ESTABLISHED, move to CLOSING and modify the QP state | |
251 | * based on the abrupt flag | |
252 | * - If the connection is already in the CLOSING or IDLE state, the peer is | |
253 | * disconnecting concurrently with us and we've already seen the | |
254 | * DISCONNECT event -- ignore the request and return 0 | |
255 | * - Disconnect on a listening endpoint returns -EINVAL | |
256 | */ | |
257 | int iw_cm_disconnect(struct iw_cm_id *cm_id, int abrupt) | |
258 | { | |
259 | struct iwcm_id_private *cm_id_priv; | |
260 | unsigned long flags; | |
261 | int ret = 0; | |
262 | struct ib_qp *qp = NULL; | |
263 | ||
264 | cm_id_priv = container_of(cm_id, struct iwcm_id_private, id); | |
265 | /* Wait if we're currently in a connect or accept downcall */ | |
266 | wait_event(cm_id_priv->connect_wait, | |
267 | !test_bit(IWCM_F_CONNECT_WAIT, &cm_id_priv->flags)); | |
268 | ||
269 | spin_lock_irqsave(&cm_id_priv->lock, flags); | |
270 | switch (cm_id_priv->state) { | |
271 | case IW_CM_STATE_ESTABLISHED: | |
272 | cm_id_priv->state = IW_CM_STATE_CLOSING; | |
273 | ||
274 | /* QP could be <nul> for user-mode client */ | |
275 | if (cm_id_priv->qp) | |
276 | qp = cm_id_priv->qp; | |
277 | else | |
278 | ret = -EINVAL; | |
279 | break; | |
280 | case IW_CM_STATE_LISTEN: | |
281 | ret = -EINVAL; | |
282 | break; | |
283 | case IW_CM_STATE_CLOSING: | |
284 | /* remote peer closed first */ | |
285 | case IW_CM_STATE_IDLE: | |
286 | /* accept or connect returned !0 */ | |
287 | break; | |
288 | case IW_CM_STATE_CONN_RECV: | |
289 | /* | |
290 | * App called disconnect before/without calling accept after | |
291 | * connect_request event delivered. | |
292 | */ | |
293 | break; | |
294 | case IW_CM_STATE_CONN_SENT: | |
295 | /* Can only get here if wait above fails */ | |
296 | default: | |
297 | BUG(); | |
298 | } | |
299 | spin_unlock_irqrestore(&cm_id_priv->lock, flags); | |
300 | ||
301 | if (qp) { | |
302 | if (abrupt) | |
303 | ret = iwcm_modify_qp_err(qp); | |
304 | else | |
305 | ret = iwcm_modify_qp_sqd(qp); | |
306 | ||
307 | /* | |
308 | * If both sides are disconnecting the QP could | |
309 | * already be in ERR or SQD states | |
310 | */ | |
311 | ret = 0; | |
312 | } | |
313 | ||
314 | return ret; | |
315 | } | |
316 | EXPORT_SYMBOL(iw_cm_disconnect); | |
317 | ||
318 | /* | |
319 | * CM_ID <-- DESTROYING | |
320 | * | |
321 | * Clean up all resources associated with the connection and release | |
322 | * the initial reference taken by iw_create_cm_id. | |
323 | */ | |
324 | static void destroy_cm_id(struct iw_cm_id *cm_id) | |
325 | { | |
326 | struct iwcm_id_private *cm_id_priv; | |
327 | unsigned long flags; | |
328 | int ret; | |
329 | ||
330 | cm_id_priv = container_of(cm_id, struct iwcm_id_private, id); | |
331 | /* | |
332 | * Wait if we're currently in a connect or accept downcall. A | |
333 | * listening endpoint should never block here. | |
334 | */ | |
335 | wait_event(cm_id_priv->connect_wait, | |
336 | !test_bit(IWCM_F_CONNECT_WAIT, &cm_id_priv->flags)); | |
337 | ||
338 | spin_lock_irqsave(&cm_id_priv->lock, flags); | |
339 | switch (cm_id_priv->state) { | |
340 | case IW_CM_STATE_LISTEN: | |
341 | cm_id_priv->state = IW_CM_STATE_DESTROYING; | |
342 | spin_unlock_irqrestore(&cm_id_priv->lock, flags); | |
343 | /* destroy the listening endpoint */ | |
344 | ret = cm_id->device->iwcm->destroy_listen(cm_id); | |
345 | spin_lock_irqsave(&cm_id_priv->lock, flags); | |
346 | break; | |
347 | case IW_CM_STATE_ESTABLISHED: | |
348 | cm_id_priv->state = IW_CM_STATE_DESTROYING; | |
349 | spin_unlock_irqrestore(&cm_id_priv->lock, flags); | |
350 | /* Abrupt close of the connection */ | |
351 | (void)iwcm_modify_qp_err(cm_id_priv->qp); | |
352 | spin_lock_irqsave(&cm_id_priv->lock, flags); | |
353 | break; | |
354 | case IW_CM_STATE_IDLE: | |
355 | case IW_CM_STATE_CLOSING: | |
356 | cm_id_priv->state = IW_CM_STATE_DESTROYING; | |
357 | break; | |
358 | case IW_CM_STATE_CONN_RECV: | |
359 | /* | |
360 | * App called destroy before/without calling accept after | |
ebb90986 SW |
361 | * receiving connection request event notification or |
362 | * returned non zero from the event callback function. | |
363 | * In either case, must tell the provider to reject. | |
922a8e9f TT |
364 | */ |
365 | cm_id_priv->state = IW_CM_STATE_DESTROYING; | |
54e05f15 | 366 | spin_unlock_irqrestore(&cm_id_priv->lock, flags); |
cb58160e | 367 | cm_id->device->iwcm->reject(cm_id, NULL, 0); |
54e05f15 | 368 | spin_lock_irqsave(&cm_id_priv->lock, flags); |
922a8e9f TT |
369 | break; |
370 | case IW_CM_STATE_CONN_SENT: | |
371 | case IW_CM_STATE_DESTROYING: | |
372 | default: | |
373 | BUG(); | |
374 | break; | |
375 | } | |
376 | if (cm_id_priv->qp) { | |
377 | cm_id_priv->id.device->iwcm->rem_ref(cm_id_priv->qp); | |
378 | cm_id_priv->qp = NULL; | |
379 | } | |
380 | spin_unlock_irqrestore(&cm_id_priv->lock, flags); | |
381 | ||
382 | (void)iwcm_deref_id(cm_id_priv); | |
383 | } | |
384 | ||
385 | /* | |
386 | * This function is only called by the application thread and cannot | |
387 | * be called by the event thread. The function will wait for all | |
388 | * references to be released on the cm_id and then kfree the cm_id | |
389 | * object. | |
390 | */ | |
391 | void iw_destroy_cm_id(struct iw_cm_id *cm_id) | |
392 | { | |
393 | struct iwcm_id_private *cm_id_priv; | |
394 | ||
395 | cm_id_priv = container_of(cm_id, struct iwcm_id_private, id); | |
396 | BUG_ON(test_bit(IWCM_F_CALLBACK_DESTROY, &cm_id_priv->flags)); | |
397 | ||
398 | destroy_cm_id(cm_id); | |
399 | ||
400 | wait_for_completion(&cm_id_priv->destroy_comp); | |
401 | ||
ebb90986 | 402 | free_cm_id(cm_id_priv); |
922a8e9f TT |
403 | } |
404 | EXPORT_SYMBOL(iw_destroy_cm_id); | |
405 | ||
406 | /* | |
407 | * CM_ID <-- LISTEN | |
408 | * | |
409 | * Start listening for connect requests. Generates one CONNECT_REQUEST | |
410 | * event for each inbound connect request. | |
411 | */ | |
412 | int iw_cm_listen(struct iw_cm_id *cm_id, int backlog) | |
413 | { | |
414 | struct iwcm_id_private *cm_id_priv; | |
415 | unsigned long flags; | |
13fccdb3 | 416 | int ret; |
922a8e9f TT |
417 | |
418 | cm_id_priv = container_of(cm_id, struct iwcm_id_private, id); | |
419 | ||
420 | ret = alloc_work_entries(cm_id_priv, backlog); | |
421 | if (ret) | |
422 | return ret; | |
423 | ||
424 | spin_lock_irqsave(&cm_id_priv->lock, flags); | |
425 | switch (cm_id_priv->state) { | |
426 | case IW_CM_STATE_IDLE: | |
427 | cm_id_priv->state = IW_CM_STATE_LISTEN; | |
428 | spin_unlock_irqrestore(&cm_id_priv->lock, flags); | |
429 | ret = cm_id->device->iwcm->create_listen(cm_id, backlog); | |
430 | if (ret) | |
431 | cm_id_priv->state = IW_CM_STATE_IDLE; | |
432 | spin_lock_irqsave(&cm_id_priv->lock, flags); | |
433 | break; | |
434 | default: | |
435 | ret = -EINVAL; | |
436 | } | |
437 | spin_unlock_irqrestore(&cm_id_priv->lock, flags); | |
438 | ||
439 | return ret; | |
440 | } | |
441 | EXPORT_SYMBOL(iw_cm_listen); | |
442 | ||
443 | /* | |
444 | * CM_ID <-- IDLE | |
445 | * | |
446 | * Rejects an inbound connection request. No events are generated. | |
447 | */ | |
448 | int iw_cm_reject(struct iw_cm_id *cm_id, | |
449 | const void *private_data, | |
450 | u8 private_data_len) | |
451 | { | |
452 | struct iwcm_id_private *cm_id_priv; | |
453 | unsigned long flags; | |
454 | int ret; | |
455 | ||
456 | cm_id_priv = container_of(cm_id, struct iwcm_id_private, id); | |
457 | set_bit(IWCM_F_CONNECT_WAIT, &cm_id_priv->flags); | |
458 | ||
459 | spin_lock_irqsave(&cm_id_priv->lock, flags); | |
460 | if (cm_id_priv->state != IW_CM_STATE_CONN_RECV) { | |
461 | spin_unlock_irqrestore(&cm_id_priv->lock, flags); | |
462 | clear_bit(IWCM_F_CONNECT_WAIT, &cm_id_priv->flags); | |
463 | wake_up_all(&cm_id_priv->connect_wait); | |
464 | return -EINVAL; | |
465 | } | |
466 | cm_id_priv->state = IW_CM_STATE_IDLE; | |
467 | spin_unlock_irqrestore(&cm_id_priv->lock, flags); | |
468 | ||
469 | ret = cm_id->device->iwcm->reject(cm_id, private_data, | |
470 | private_data_len); | |
471 | ||
472 | clear_bit(IWCM_F_CONNECT_WAIT, &cm_id_priv->flags); | |
473 | wake_up_all(&cm_id_priv->connect_wait); | |
474 | ||
475 | return ret; | |
476 | } | |
477 | EXPORT_SYMBOL(iw_cm_reject); | |
478 | ||
479 | /* | |
480 | * CM_ID <-- ESTABLISHED | |
481 | * | |
482 | * Accepts an inbound connection request and generates an ESTABLISHED | |
483 | * event. Callers of iw_cm_disconnect and iw_destroy_cm_id will block | |
484 | * until the ESTABLISHED event is received from the provider. | |
485 | */ | |
486 | int iw_cm_accept(struct iw_cm_id *cm_id, | |
487 | struct iw_cm_conn_param *iw_param) | |
488 | { | |
489 | struct iwcm_id_private *cm_id_priv; | |
490 | struct ib_qp *qp; | |
491 | unsigned long flags; | |
492 | int ret; | |
493 | ||
494 | cm_id_priv = container_of(cm_id, struct iwcm_id_private, id); | |
495 | set_bit(IWCM_F_CONNECT_WAIT, &cm_id_priv->flags); | |
496 | ||
497 | spin_lock_irqsave(&cm_id_priv->lock, flags); | |
498 | if (cm_id_priv->state != IW_CM_STATE_CONN_RECV) { | |
499 | spin_unlock_irqrestore(&cm_id_priv->lock, flags); | |
500 | clear_bit(IWCM_F_CONNECT_WAIT, &cm_id_priv->flags); | |
501 | wake_up_all(&cm_id_priv->connect_wait); | |
502 | return -EINVAL; | |
503 | } | |
504 | /* Get the ib_qp given the QPN */ | |
505 | qp = cm_id->device->iwcm->get_qp(cm_id->device, iw_param->qpn); | |
506 | if (!qp) { | |
507 | spin_unlock_irqrestore(&cm_id_priv->lock, flags); | |
508 | return -EINVAL; | |
509 | } | |
510 | cm_id->device->iwcm->add_ref(qp); | |
511 | cm_id_priv->qp = qp; | |
512 | spin_unlock_irqrestore(&cm_id_priv->lock, flags); | |
513 | ||
514 | ret = cm_id->device->iwcm->accept(cm_id, iw_param); | |
515 | if (ret) { | |
516 | /* An error on accept precludes provider events */ | |
517 | BUG_ON(cm_id_priv->state != IW_CM_STATE_CONN_RECV); | |
518 | cm_id_priv->state = IW_CM_STATE_IDLE; | |
519 | spin_lock_irqsave(&cm_id_priv->lock, flags); | |
520 | if (cm_id_priv->qp) { | |
521 | cm_id->device->iwcm->rem_ref(qp); | |
522 | cm_id_priv->qp = NULL; | |
523 | } | |
524 | spin_unlock_irqrestore(&cm_id_priv->lock, flags); | |
525 | clear_bit(IWCM_F_CONNECT_WAIT, &cm_id_priv->flags); | |
526 | wake_up_all(&cm_id_priv->connect_wait); | |
527 | } | |
528 | ||
529 | return ret; | |
530 | } | |
531 | EXPORT_SYMBOL(iw_cm_accept); | |
532 | ||
533 | /* | |
534 | * Active Side: CM_ID <-- CONN_SENT | |
535 | * | |
536 | * If successful, results in the generation of a CONNECT_REPLY | |
537 | * event. iw_cm_disconnect and iw_cm_destroy will block until the | |
538 | * CONNECT_REPLY event is received from the provider. | |
539 | */ | |
540 | int iw_cm_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *iw_param) | |
541 | { | |
542 | struct iwcm_id_private *cm_id_priv; | |
13fccdb3 | 543 | int ret; |
922a8e9f TT |
544 | unsigned long flags; |
545 | struct ib_qp *qp; | |
546 | ||
547 | cm_id_priv = container_of(cm_id, struct iwcm_id_private, id); | |
548 | ||
549 | ret = alloc_work_entries(cm_id_priv, 4); | |
550 | if (ret) | |
551 | return ret; | |
552 | ||
553 | set_bit(IWCM_F_CONNECT_WAIT, &cm_id_priv->flags); | |
554 | spin_lock_irqsave(&cm_id_priv->lock, flags); | |
555 | ||
556 | if (cm_id_priv->state != IW_CM_STATE_IDLE) { | |
557 | spin_unlock_irqrestore(&cm_id_priv->lock, flags); | |
558 | clear_bit(IWCM_F_CONNECT_WAIT, &cm_id_priv->flags); | |
559 | wake_up_all(&cm_id_priv->connect_wait); | |
560 | return -EINVAL; | |
561 | } | |
562 | ||
563 | /* Get the ib_qp given the QPN */ | |
564 | qp = cm_id->device->iwcm->get_qp(cm_id->device, iw_param->qpn); | |
565 | if (!qp) { | |
566 | spin_unlock_irqrestore(&cm_id_priv->lock, flags); | |
567 | return -EINVAL; | |
568 | } | |
569 | cm_id->device->iwcm->add_ref(qp); | |
570 | cm_id_priv->qp = qp; | |
571 | cm_id_priv->state = IW_CM_STATE_CONN_SENT; | |
572 | spin_unlock_irqrestore(&cm_id_priv->lock, flags); | |
573 | ||
574 | ret = cm_id->device->iwcm->connect(cm_id, iw_param); | |
575 | if (ret) { | |
576 | spin_lock_irqsave(&cm_id_priv->lock, flags); | |
577 | if (cm_id_priv->qp) { | |
578 | cm_id->device->iwcm->rem_ref(qp); | |
579 | cm_id_priv->qp = NULL; | |
580 | } | |
581 | spin_unlock_irqrestore(&cm_id_priv->lock, flags); | |
582 | BUG_ON(cm_id_priv->state != IW_CM_STATE_CONN_SENT); | |
583 | cm_id_priv->state = IW_CM_STATE_IDLE; | |
584 | clear_bit(IWCM_F_CONNECT_WAIT, &cm_id_priv->flags); | |
585 | wake_up_all(&cm_id_priv->connect_wait); | |
586 | } | |
587 | ||
588 | return ret; | |
589 | } | |
590 | EXPORT_SYMBOL(iw_cm_connect); | |
591 | ||
592 | /* | |
593 | * Passive Side: new CM_ID <-- CONN_RECV | |
594 | * | |
595 | * Handles an inbound connect request. The function creates a new | |
596 | * iw_cm_id to represent the new connection and inherits the client | |
597 | * callback function and other attributes from the listening parent. | |
598 | * | |
599 | * The work item contains a pointer to the listen_cm_id and the event. The | |
600 | * listen_cm_id contains the client cm_handler, context and | |
601 | * device. These are copied when the device is cloned. The event | |
602 | * contains the new four tuple. | |
603 | * | |
604 | * An error on the child should not affect the parent, so this | |
605 | * function does not return a value. | |
606 | */ | |
607 | static void cm_conn_req_handler(struct iwcm_id_private *listen_id_priv, | |
608 | struct iw_cm_event *iw_event) | |
609 | { | |
610 | unsigned long flags; | |
611 | struct iw_cm_id *cm_id; | |
612 | struct iwcm_id_private *cm_id_priv; | |
613 | int ret; | |
614 | ||
615 | /* | |
616 | * The provider should never generate a connection request | |
617 | * event with a bad status. | |
618 | */ | |
619 | BUG_ON(iw_event->status); | |
620 | ||
621 | /* | |
622 | * We could be destroying the listening id. If so, ignore this | |
623 | * upcall. | |
624 | */ | |
625 | spin_lock_irqsave(&listen_id_priv->lock, flags); | |
626 | if (listen_id_priv->state != IW_CM_STATE_LISTEN) { | |
627 | spin_unlock_irqrestore(&listen_id_priv->lock, flags); | |
83b96586 | 628 | goto out; |
922a8e9f TT |
629 | } |
630 | spin_unlock_irqrestore(&listen_id_priv->lock, flags); | |
631 | ||
632 | cm_id = iw_create_cm_id(listen_id_priv->id.device, | |
633 | listen_id_priv->id.cm_handler, | |
634 | listen_id_priv->id.context); | |
635 | /* If the cm_id could not be created, ignore the request */ | |
636 | if (IS_ERR(cm_id)) | |
83b96586 | 637 | goto out; |
922a8e9f TT |
638 | |
639 | cm_id->provider_data = iw_event->provider_data; | |
640 | cm_id->local_addr = iw_event->local_addr; | |
641 | cm_id->remote_addr = iw_event->remote_addr; | |
642 | ||
643 | cm_id_priv = container_of(cm_id, struct iwcm_id_private, id); | |
644 | cm_id_priv->state = IW_CM_STATE_CONN_RECV; | |
645 | ||
646 | ret = alloc_work_entries(cm_id_priv, 3); | |
647 | if (ret) { | |
648 | iw_cm_reject(cm_id, NULL, 0); | |
649 | iw_destroy_cm_id(cm_id); | |
83b96586 | 650 | goto out; |
922a8e9f TT |
651 | } |
652 | ||
653 | /* Call the client CM handler */ | |
654 | ret = cm_id->cm_handler(cm_id, iw_event); | |
655 | if (ret) { | |
ebb90986 | 656 | iw_cm_reject(cm_id, NULL, 0); |
922a8e9f TT |
657 | set_bit(IWCM_F_CALLBACK_DESTROY, &cm_id_priv->flags); |
658 | destroy_cm_id(cm_id); | |
659 | if (atomic_read(&cm_id_priv->refcount)==0) | |
ebb90986 | 660 | free_cm_id(cm_id_priv); |
922a8e9f TT |
661 | } |
662 | ||
83b96586 | 663 | out: |
922a8e9f TT |
664 | if (iw_event->private_data_len) |
665 | kfree(iw_event->private_data); | |
666 | } | |
667 | ||
668 | /* | |
669 | * Passive Side: CM_ID <-- ESTABLISHED | |
670 | * | |
671 | * The provider generated an ESTABLISHED event which means that | |
672 | * the MPA negotion has completed successfully and we are now in MPA | |
673 | * FPDU mode. | |
674 | * | |
675 | * This event can only be received in the CONN_RECV state. If the | |
676 | * remote peer closed, the ESTABLISHED event would be received followed | |
677 | * by the CLOSE event. If the app closes, it will block until we wake | |
678 | * it up after processing this event. | |
679 | */ | |
680 | static int cm_conn_est_handler(struct iwcm_id_private *cm_id_priv, | |
681 | struct iw_cm_event *iw_event) | |
682 | { | |
683 | unsigned long flags; | |
13fccdb3 | 684 | int ret; |
922a8e9f TT |
685 | |
686 | spin_lock_irqsave(&cm_id_priv->lock, flags); | |
687 | ||
688 | /* | |
689 | * We clear the CONNECT_WAIT bit here to allow the callback | |
690 | * function to call iw_cm_disconnect. Calling iw_destroy_cm_id | |
691 | * from a callback handler is not allowed. | |
692 | */ | |
693 | clear_bit(IWCM_F_CONNECT_WAIT, &cm_id_priv->flags); | |
694 | BUG_ON(cm_id_priv->state != IW_CM_STATE_CONN_RECV); | |
695 | cm_id_priv->state = IW_CM_STATE_ESTABLISHED; | |
696 | spin_unlock_irqrestore(&cm_id_priv->lock, flags); | |
697 | ret = cm_id_priv->id.cm_handler(&cm_id_priv->id, iw_event); | |
698 | wake_up_all(&cm_id_priv->connect_wait); | |
699 | ||
700 | return ret; | |
701 | } | |
702 | ||
703 | /* | |
704 | * Active Side: CM_ID <-- ESTABLISHED | |
705 | * | |
706 | * The app has called connect and is waiting for the established event to | |
707 | * post it's requests to the server. This event will wake up anyone | |
708 | * blocked in iw_cm_disconnect or iw_destroy_id. | |
709 | */ | |
710 | static int cm_conn_rep_handler(struct iwcm_id_private *cm_id_priv, | |
711 | struct iw_cm_event *iw_event) | |
712 | { | |
713 | unsigned long flags; | |
13fccdb3 | 714 | int ret; |
922a8e9f TT |
715 | |
716 | spin_lock_irqsave(&cm_id_priv->lock, flags); | |
717 | /* | |
718 | * Clear the connect wait bit so a callback function calling | |
719 | * iw_cm_disconnect will not wait and deadlock this thread | |
720 | */ | |
721 | clear_bit(IWCM_F_CONNECT_WAIT, &cm_id_priv->flags); | |
722 | BUG_ON(cm_id_priv->state != IW_CM_STATE_CONN_SENT); | |
723 | if (iw_event->status == IW_CM_EVENT_STATUS_ACCEPTED) { | |
724 | cm_id_priv->id.local_addr = iw_event->local_addr; | |
725 | cm_id_priv->id.remote_addr = iw_event->remote_addr; | |
726 | cm_id_priv->state = IW_CM_STATE_ESTABLISHED; | |
727 | } else { | |
728 | /* REJECTED or RESET */ | |
729 | cm_id_priv->id.device->iwcm->rem_ref(cm_id_priv->qp); | |
730 | cm_id_priv->qp = NULL; | |
731 | cm_id_priv->state = IW_CM_STATE_IDLE; | |
732 | } | |
733 | spin_unlock_irqrestore(&cm_id_priv->lock, flags); | |
734 | ret = cm_id_priv->id.cm_handler(&cm_id_priv->id, iw_event); | |
735 | ||
736 | if (iw_event->private_data_len) | |
737 | kfree(iw_event->private_data); | |
738 | ||
739 | /* Wake up waiters on connect complete */ | |
740 | wake_up_all(&cm_id_priv->connect_wait); | |
741 | ||
742 | return ret; | |
743 | } | |
744 | ||
745 | /* | |
746 | * CM_ID <-- CLOSING | |
747 | * | |
748 | * If in the ESTABLISHED state, move to CLOSING. | |
749 | */ | |
750 | static void cm_disconnect_handler(struct iwcm_id_private *cm_id_priv, | |
751 | struct iw_cm_event *iw_event) | |
752 | { | |
753 | unsigned long flags; | |
754 | ||
755 | spin_lock_irqsave(&cm_id_priv->lock, flags); | |
756 | if (cm_id_priv->state == IW_CM_STATE_ESTABLISHED) | |
757 | cm_id_priv->state = IW_CM_STATE_CLOSING; | |
758 | spin_unlock_irqrestore(&cm_id_priv->lock, flags); | |
759 | } | |
760 | ||
761 | /* | |
762 | * CM_ID <-- IDLE | |
763 | * | |
764 | * If in the ESTBLISHED or CLOSING states, the QP will have have been | |
765 | * moved by the provider to the ERR state. Disassociate the CM_ID from | |
766 | * the QP, move to IDLE, and remove the 'connected' reference. | |
767 | * | |
768 | * If in some other state, the cm_id was destroyed asynchronously. | |
769 | * This is the last reference that will result in waking up | |
770 | * the app thread blocked in iw_destroy_cm_id. | |
771 | */ | |
772 | static int cm_close_handler(struct iwcm_id_private *cm_id_priv, | |
773 | struct iw_cm_event *iw_event) | |
774 | { | |
775 | unsigned long flags; | |
776 | int ret = 0; | |
777 | spin_lock_irqsave(&cm_id_priv->lock, flags); | |
778 | ||
779 | if (cm_id_priv->qp) { | |
780 | cm_id_priv->id.device->iwcm->rem_ref(cm_id_priv->qp); | |
781 | cm_id_priv->qp = NULL; | |
782 | } | |
783 | switch (cm_id_priv->state) { | |
784 | case IW_CM_STATE_ESTABLISHED: | |
785 | case IW_CM_STATE_CLOSING: | |
786 | cm_id_priv->state = IW_CM_STATE_IDLE; | |
787 | spin_unlock_irqrestore(&cm_id_priv->lock, flags); | |
788 | ret = cm_id_priv->id.cm_handler(&cm_id_priv->id, iw_event); | |
789 | spin_lock_irqsave(&cm_id_priv->lock, flags); | |
790 | break; | |
791 | case IW_CM_STATE_DESTROYING: | |
792 | break; | |
793 | default: | |
794 | BUG(); | |
795 | } | |
796 | spin_unlock_irqrestore(&cm_id_priv->lock, flags); | |
797 | ||
798 | return ret; | |
799 | } | |
800 | ||
801 | static int process_event(struct iwcm_id_private *cm_id_priv, | |
802 | struct iw_cm_event *iw_event) | |
803 | { | |
804 | int ret = 0; | |
805 | ||
806 | switch (iw_event->event) { | |
807 | case IW_CM_EVENT_CONNECT_REQUEST: | |
808 | cm_conn_req_handler(cm_id_priv, iw_event); | |
809 | break; | |
810 | case IW_CM_EVENT_CONNECT_REPLY: | |
811 | ret = cm_conn_rep_handler(cm_id_priv, iw_event); | |
812 | break; | |
813 | case IW_CM_EVENT_ESTABLISHED: | |
814 | ret = cm_conn_est_handler(cm_id_priv, iw_event); | |
815 | break; | |
816 | case IW_CM_EVENT_DISCONNECT: | |
817 | cm_disconnect_handler(cm_id_priv, iw_event); | |
818 | break; | |
819 | case IW_CM_EVENT_CLOSE: | |
820 | ret = cm_close_handler(cm_id_priv, iw_event); | |
821 | break; | |
822 | default: | |
823 | BUG(); | |
824 | } | |
825 | ||
826 | return ret; | |
827 | } | |
828 | ||
829 | /* | |
830 | * Process events on the work_list for the cm_id. If the callback | |
831 | * function requests that the cm_id be deleted, a flag is set in the | |
832 | * cm_id flags to indicate that when the last reference is | |
833 | * removed, the cm_id is to be destroyed. This is necessary to | |
834 | * distinguish between an object that will be destroyed by the app | |
835 | * thread asleep on the destroy_comp list vs. an object destroyed | |
836 | * here synchronously when the last reference is removed. | |
837 | */ | |
c4028958 | 838 | static void cm_work_handler(struct work_struct *_work) |
922a8e9f | 839 | { |
4c1ac1b4 | 840 | struct iwcm_work *work = container_of(_work, struct iwcm_work, work); |
33ba0fa9 | 841 | struct iw_cm_event levent; |
922a8e9f TT |
842 | struct iwcm_id_private *cm_id_priv = work->cm_id; |
843 | unsigned long flags; | |
844 | int empty; | |
845 | int ret = 0; | |
d7c1fbd6 | 846 | int destroy_id; |
922a8e9f TT |
847 | |
848 | spin_lock_irqsave(&cm_id_priv->lock, flags); | |
849 | empty = list_empty(&cm_id_priv->work_list); | |
850 | while (!empty) { | |
851 | work = list_entry(cm_id_priv->work_list.next, | |
852 | struct iwcm_work, list); | |
853 | list_del_init(&work->list); | |
854 | empty = list_empty(&cm_id_priv->work_list); | |
33ba0fa9 | 855 | levent = work->event; |
922a8e9f TT |
856 | put_work(work); |
857 | spin_unlock_irqrestore(&cm_id_priv->lock, flags); | |
858 | ||
33ba0fa9 | 859 | ret = process_event(cm_id_priv, &levent); |
922a8e9f TT |
860 | if (ret) { |
861 | set_bit(IWCM_F_CALLBACK_DESTROY, &cm_id_priv->flags); | |
862 | destroy_cm_id(&cm_id_priv->id); | |
863 | } | |
864 | BUG_ON(atomic_read(&cm_id_priv->refcount)==0); | |
d7c1fbd6 | 865 | destroy_id = test_bit(IWCM_F_CALLBACK_DESTROY, &cm_id_priv->flags); |
ebb90986 | 866 | if (iwcm_deref_id(cm_id_priv)) { |
d7c1fbd6 | 867 | if (destroy_id) { |
ebb90986 SW |
868 | BUG_ON(!list_empty(&cm_id_priv->work_list)); |
869 | free_cm_id(cm_id_priv); | |
870 | } | |
922a8e9f TT |
871 | return; |
872 | } | |
873 | spin_lock_irqsave(&cm_id_priv->lock, flags); | |
874 | } | |
875 | spin_unlock_irqrestore(&cm_id_priv->lock, flags); | |
876 | } | |
877 | ||
878 | /* | |
879 | * This function is called on interrupt context. Schedule events on | |
880 | * the iwcm_wq thread to allow callback functions to downcall into | |
881 | * the CM and/or block. Events are queued to a per-CM_ID | |
882 | * work_list. If this is the first event on the work_list, the work | |
883 | * element is also queued on the iwcm_wq thread. | |
884 | * | |
885 | * Each event holds a reference on the cm_id. Until the last posted | |
886 | * event has been delivered and processed, the cm_id cannot be | |
887 | * deleted. | |
888 | * | |
889 | * Returns: | |
890 | * 0 - the event was handled. | |
891 | * -ENOMEM - the event was not handled due to lack of resources. | |
892 | */ | |
893 | static int cm_event_handler(struct iw_cm_id *cm_id, | |
894 | struct iw_cm_event *iw_event) | |
895 | { | |
896 | struct iwcm_work *work; | |
897 | struct iwcm_id_private *cm_id_priv; | |
898 | unsigned long flags; | |
899 | int ret = 0; | |
900 | ||
901 | cm_id_priv = container_of(cm_id, struct iwcm_id_private, id); | |
902 | ||
903 | spin_lock_irqsave(&cm_id_priv->lock, flags); | |
904 | work = get_work(cm_id_priv); | |
905 | if (!work) { | |
906 | ret = -ENOMEM; | |
907 | goto out; | |
908 | } | |
909 | ||
c4028958 | 910 | INIT_WORK(&work->work, cm_work_handler); |
922a8e9f TT |
911 | work->cm_id = cm_id_priv; |
912 | work->event = *iw_event; | |
913 | ||
914 | if ((work->event.event == IW_CM_EVENT_CONNECT_REQUEST || | |
915 | work->event.event == IW_CM_EVENT_CONNECT_REPLY) && | |
916 | work->event.private_data_len) { | |
715a588f | 917 | ret = copy_private_data(&work->event); |
922a8e9f TT |
918 | if (ret) { |
919 | put_work(work); | |
920 | goto out; | |
921 | } | |
922 | } | |
923 | ||
924 | atomic_inc(&cm_id_priv->refcount); | |
925 | if (list_empty(&cm_id_priv->work_list)) { | |
926 | list_add_tail(&work->list, &cm_id_priv->work_list); | |
927 | queue_work(iwcm_wq, &work->work); | |
928 | } else | |
929 | list_add_tail(&work->list, &cm_id_priv->work_list); | |
930 | out: | |
931 | spin_unlock_irqrestore(&cm_id_priv->lock, flags); | |
932 | return ret; | |
933 | } | |
934 | ||
935 | static int iwcm_init_qp_init_attr(struct iwcm_id_private *cm_id_priv, | |
936 | struct ib_qp_attr *qp_attr, | |
937 | int *qp_attr_mask) | |
938 | { | |
939 | unsigned long flags; | |
940 | int ret; | |
941 | ||
942 | spin_lock_irqsave(&cm_id_priv->lock, flags); | |
943 | switch (cm_id_priv->state) { | |
944 | case IW_CM_STATE_IDLE: | |
945 | case IW_CM_STATE_CONN_SENT: | |
946 | case IW_CM_STATE_CONN_RECV: | |
947 | case IW_CM_STATE_ESTABLISHED: | |
948 | *qp_attr_mask = IB_QP_STATE | IB_QP_ACCESS_FLAGS; | |
1ca8d156 | 949 | qp_attr->qp_access_flags = IB_ACCESS_REMOTE_WRITE| |
922a8e9f TT |
950 | IB_ACCESS_REMOTE_READ; |
951 | ret = 0; | |
952 | break; | |
953 | default: | |
954 | ret = -EINVAL; | |
955 | break; | |
956 | } | |
957 | spin_unlock_irqrestore(&cm_id_priv->lock, flags); | |
958 | return ret; | |
959 | } | |
960 | ||
961 | static int iwcm_init_qp_rts_attr(struct iwcm_id_private *cm_id_priv, | |
962 | struct ib_qp_attr *qp_attr, | |
963 | int *qp_attr_mask) | |
964 | { | |
965 | unsigned long flags; | |
966 | int ret; | |
967 | ||
968 | spin_lock_irqsave(&cm_id_priv->lock, flags); | |
969 | switch (cm_id_priv->state) { | |
970 | case IW_CM_STATE_IDLE: | |
971 | case IW_CM_STATE_CONN_SENT: | |
972 | case IW_CM_STATE_CONN_RECV: | |
973 | case IW_CM_STATE_ESTABLISHED: | |
974 | *qp_attr_mask = 0; | |
975 | ret = 0; | |
976 | break; | |
977 | default: | |
978 | ret = -EINVAL; | |
979 | break; | |
980 | } | |
981 | spin_unlock_irqrestore(&cm_id_priv->lock, flags); | |
982 | return ret; | |
983 | } | |
984 | ||
985 | int iw_cm_init_qp_attr(struct iw_cm_id *cm_id, | |
986 | struct ib_qp_attr *qp_attr, | |
987 | int *qp_attr_mask) | |
988 | { | |
989 | struct iwcm_id_private *cm_id_priv; | |
990 | int ret; | |
991 | ||
992 | cm_id_priv = container_of(cm_id, struct iwcm_id_private, id); | |
993 | switch (qp_attr->qp_state) { | |
994 | case IB_QPS_INIT: | |
995 | case IB_QPS_RTR: | |
996 | ret = iwcm_init_qp_init_attr(cm_id_priv, | |
997 | qp_attr, qp_attr_mask); | |
998 | break; | |
999 | case IB_QPS_RTS: | |
1000 | ret = iwcm_init_qp_rts_attr(cm_id_priv, | |
1001 | qp_attr, qp_attr_mask); | |
1002 | break; | |
1003 | default: | |
1004 | ret = -EINVAL; | |
1005 | break; | |
1006 | } | |
1007 | return ret; | |
1008 | } | |
1009 | EXPORT_SYMBOL(iw_cm_init_qp_attr); | |
1010 | ||
1011 | static int __init iw_cm_init(void) | |
1012 | { | |
1013 | iwcm_wq = create_singlethread_workqueue("iw_cm_wq"); | |
1014 | if (!iwcm_wq) | |
1015 | return -ENOMEM; | |
1016 | ||
1017 | return 0; | |
1018 | } | |
1019 | ||
1020 | static void __exit iw_cm_cleanup(void) | |
1021 | { | |
1022 | destroy_workqueue(iwcm_wq); | |
1023 | } | |
1024 | ||
1025 | module_init(iw_cm_init); | |
1026 | module_exit(iw_cm_cleanup); |