1 /*
2 * Copyright (c) 2005 Topspin Communications. All rights reserved.
3 * Copyright (c) 2005, 2006 Cisco Systems. All rights reserved.
4 * Copyright (c) 2005 Mellanox Technologies. All rights reserved.
5 * Copyright (c) 2005 Voltaire, Inc. All rights reserved.
6 * Copyright (c) 2005 PathScale, Inc. All rights reserved.
7 *
8 * This software is available to you under a choice of one of two
9 * licenses. You may choose to be licensed under the terms of the GNU
10 * General Public License (GPL) Version 2, available from the file
11 * COPYING in the main directory of this source tree, or the
12 * OpenIB.org BSD license below:
13 *
14 * Redistribution and use in source and binary forms, with or
15 * without modification, are permitted provided that the following
16 * conditions are met:
17 *
18 * - Redistributions of source code must retain the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer.
21 *
22 * - Redistributions in binary form must reproduce the above
23 * copyright notice, this list of conditions and the following
24 * disclaimer in the documentation and/or other materials
25 * provided with the distribution.
26 *
27 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
28 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
29 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
30 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
31 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
32 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
33 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
34 * SOFTWARE.
35 */
36
37 #include <linux/module.h>
38 #include <linux/init.h>
39 #include <linux/device.h>
40 #include <linux/err.h>
41 #include <linux/fs.h>
42 #include <linux/poll.h>
43 #include <linux/sched.h>
44 #include <linux/file.h>
45 #include <linux/cdev.h>
46 #include <linux/anon_inodes.h>
47 #include <linux/slab.h>
48
49 #include <asm/uaccess.h>
50
51 #include "uverbs.h"
52
53 MODULE_AUTHOR("Roland Dreier");
54 MODULE_DESCRIPTION("InfiniBand userspace verbs access");
55 MODULE_LICENSE("Dual BSD/GPL");
56
57 enum {
58 IB_UVERBS_MAJOR = 231,
59 IB_UVERBS_BASE_MINOR = 192,
60 IB_UVERBS_MAX_DEVICES = 32
61 };
62
63 #define IB_UVERBS_BASE_DEV MKDEV(IB_UVERBS_MAJOR, IB_UVERBS_BASE_MINOR)
64
65 static struct class *uverbs_class;
66
67 DEFINE_SPINLOCK(ib_uverbs_idr_lock);
68 DEFINE_IDR(ib_uverbs_pd_idr);
69 DEFINE_IDR(ib_uverbs_mr_idr);
70 DEFINE_IDR(ib_uverbs_mw_idr);
71 DEFINE_IDR(ib_uverbs_ah_idr);
72 DEFINE_IDR(ib_uverbs_cq_idr);
73 DEFINE_IDR(ib_uverbs_qp_idr);
74 DEFINE_IDR(ib_uverbs_srq_idr);
75 DEFINE_IDR(ib_uverbs_xrcd_idr);
76
77 static DEFINE_SPINLOCK(map_lock);
78 static DECLARE_BITMAP(dev_map, IB_UVERBS_MAX_DEVICES);
79
80 static ssize_t (*uverbs_cmd_table[])(struct ib_uverbs_file *file,
81 const char __user *buf, int in_len,
82 int out_len) = {
83 [IB_USER_VERBS_CMD_GET_CONTEXT] = ib_uverbs_get_context,
84 [IB_USER_VERBS_CMD_QUERY_DEVICE] = ib_uverbs_query_device,
85 [IB_USER_VERBS_CMD_QUERY_PORT] = ib_uverbs_query_port,
86 [IB_USER_VERBS_CMD_ALLOC_PD] = ib_uverbs_alloc_pd,
87 [IB_USER_VERBS_CMD_DEALLOC_PD] = ib_uverbs_dealloc_pd,
88 [IB_USER_VERBS_CMD_REG_MR] = ib_uverbs_reg_mr,
89 [IB_USER_VERBS_CMD_DEREG_MR] = ib_uverbs_dereg_mr,
90 [IB_USER_VERBS_CMD_ALLOC_MW] = ib_uverbs_alloc_mw,
91 [IB_USER_VERBS_CMD_DEALLOC_MW] = ib_uverbs_dealloc_mw,
92 [IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL] = ib_uverbs_create_comp_channel,
93 [IB_USER_VERBS_CMD_CREATE_CQ] = ib_uverbs_create_cq,
94 [IB_USER_VERBS_CMD_RESIZE_CQ] = ib_uverbs_resize_cq,
95 [IB_USER_VERBS_CMD_POLL_CQ] = ib_uverbs_poll_cq,
96 [IB_USER_VERBS_CMD_REQ_NOTIFY_CQ] = ib_uverbs_req_notify_cq,
97 [IB_USER_VERBS_CMD_DESTROY_CQ] = ib_uverbs_destroy_cq,
98 [IB_USER_VERBS_CMD_CREATE_QP] = ib_uverbs_create_qp,
99 [IB_USER_VERBS_CMD_QUERY_QP] = ib_uverbs_query_qp,
100 [IB_USER_VERBS_CMD_MODIFY_QP] = ib_uverbs_modify_qp,
101 [IB_USER_VERBS_CMD_DESTROY_QP] = ib_uverbs_destroy_qp,
102 [IB_USER_VERBS_CMD_POST_SEND] = ib_uverbs_post_send,
103 [IB_USER_VERBS_CMD_POST_RECV] = ib_uverbs_post_recv,
104 [IB_USER_VERBS_CMD_POST_SRQ_RECV] = ib_uverbs_post_srq_recv,
105 [IB_USER_VERBS_CMD_CREATE_AH] = ib_uverbs_create_ah,
106 [IB_USER_VERBS_CMD_DESTROY_AH] = ib_uverbs_destroy_ah,
107 [IB_USER_VERBS_CMD_ATTACH_MCAST] = ib_uverbs_attach_mcast,
108 [IB_USER_VERBS_CMD_DETACH_MCAST] = ib_uverbs_detach_mcast,
109 [IB_USER_VERBS_CMD_CREATE_SRQ] = ib_uverbs_create_srq,
110 [IB_USER_VERBS_CMD_MODIFY_SRQ] = ib_uverbs_modify_srq,
111 [IB_USER_VERBS_CMD_QUERY_SRQ] = ib_uverbs_query_srq,
112 [IB_USER_VERBS_CMD_DESTROY_SRQ] = ib_uverbs_destroy_srq,
113 [IB_USER_VERBS_CMD_OPEN_XRCD] = ib_uverbs_open_xrcd,
114 [IB_USER_VERBS_CMD_CLOSE_XRCD] = ib_uverbs_close_xrcd,
115 [IB_USER_VERBS_CMD_CREATE_XSRQ] = ib_uverbs_create_xsrq,
116 [IB_USER_VERBS_CMD_OPEN_QP] = ib_uverbs_open_qp
117 };
118
119 static void ib_uverbs_add_one(struct ib_device *device);
120 static void ib_uverbs_remove_one(struct ib_device *device);
121
ib_uverbs_release_dev(struct kref * ref)122 static void ib_uverbs_release_dev(struct kref *ref)
123 {
124 struct ib_uverbs_device *dev =
125 container_of(ref, struct ib_uverbs_device, ref);
126
127 complete(&dev->comp);
128 }
129
ib_uverbs_release_event_file(struct kref * ref)130 static void ib_uverbs_release_event_file(struct kref *ref)
131 {
132 struct ib_uverbs_event_file *file =
133 container_of(ref, struct ib_uverbs_event_file, ref);
134
135 kfree(file);
136 }
137
ib_uverbs_release_ucq(struct ib_uverbs_file * file,struct ib_uverbs_event_file * ev_file,struct ib_ucq_object * uobj)138 void ib_uverbs_release_ucq(struct ib_uverbs_file *file,
139 struct ib_uverbs_event_file *ev_file,
140 struct ib_ucq_object *uobj)
141 {
142 struct ib_uverbs_event *evt, *tmp;
143
144 if (ev_file) {
145 spin_lock_irq(&ev_file->lock);
146 list_for_each_entry_safe(evt, tmp, &uobj->comp_list, obj_list) {
147 list_del(&evt->list);
148 kfree(evt);
149 }
150 spin_unlock_irq(&ev_file->lock);
151
152 kref_put(&ev_file->ref, ib_uverbs_release_event_file);
153 }
154
155 spin_lock_irq(&file->async_file->lock);
156 list_for_each_entry_safe(evt, tmp, &uobj->async_list, obj_list) {
157 list_del(&evt->list);
158 kfree(evt);
159 }
160 spin_unlock_irq(&file->async_file->lock);
161 }
162
ib_uverbs_release_uevent(struct ib_uverbs_file * file,struct ib_uevent_object * uobj)163 void ib_uverbs_release_uevent(struct ib_uverbs_file *file,
164 struct ib_uevent_object *uobj)
165 {
166 struct ib_uverbs_event *evt, *tmp;
167
168 spin_lock_irq(&file->async_file->lock);
169 list_for_each_entry_safe(evt, tmp, &uobj->event_list, obj_list) {
170 list_del(&evt->list);
171 kfree(evt);
172 }
173 spin_unlock_irq(&file->async_file->lock);
174 }
175
ib_uverbs_detach_umcast(struct ib_qp * qp,struct ib_uqp_object * uobj)176 static void ib_uverbs_detach_umcast(struct ib_qp *qp,
177 struct ib_uqp_object *uobj)
178 {
179 struct ib_uverbs_mcast_entry *mcast, *tmp;
180
181 list_for_each_entry_safe(mcast, tmp, &uobj->mcast_list, list) {
182 ib_detach_mcast(qp, &mcast->gid, mcast->lid);
183 list_del(&mcast->list);
184 kfree(mcast);
185 }
186 }
187
ib_uverbs_cleanup_ucontext(struct ib_uverbs_file * file,struct ib_ucontext * context)188 static int ib_uverbs_cleanup_ucontext(struct ib_uverbs_file *file,
189 struct ib_ucontext *context)
190 {
191 struct ib_uobject *uobj, *tmp;
192
193 if (!context)
194 return 0;
195
196 context->closing = 1;
197
198 list_for_each_entry_safe(uobj, tmp, &context->ah_list, list) {
199 struct ib_ah *ah = uobj->object;
200
201 idr_remove_uobj(&ib_uverbs_ah_idr, uobj);
202 ib_destroy_ah(ah);
203 kfree(uobj);
204 }
205
206 /* Remove MWs before QPs, in order to support type 2A MWs. */
207 list_for_each_entry_safe(uobj, tmp, &context->mw_list, list) {
208 struct ib_mw *mw = uobj->object;
209
210 idr_remove_uobj(&ib_uverbs_mw_idr, uobj);
211 ib_dealloc_mw(mw);
212 kfree(uobj);
213 }
214
215 list_for_each_entry_safe(uobj, tmp, &context->qp_list, list) {
216 struct ib_qp *qp = uobj->object;
217 struct ib_uqp_object *uqp =
218 container_of(uobj, struct ib_uqp_object, uevent.uobject);
219
220 idr_remove_uobj(&ib_uverbs_qp_idr, uobj);
221 if (qp != qp->real_qp) {
222 ib_close_qp(qp);
223 } else {
224 ib_uverbs_detach_umcast(qp, uqp);
225 ib_destroy_qp(qp);
226 }
227 ib_uverbs_release_uevent(file, &uqp->uevent);
228 kfree(uqp);
229 }
230
231 list_for_each_entry_safe(uobj, tmp, &context->cq_list, list) {
232 struct ib_cq *cq = uobj->object;
233 struct ib_uverbs_event_file *ev_file = cq->cq_context;
234 struct ib_ucq_object *ucq =
235 container_of(uobj, struct ib_ucq_object, uobject);
236
237 idr_remove_uobj(&ib_uverbs_cq_idr, uobj);
238 ib_destroy_cq(cq);
239 ib_uverbs_release_ucq(file, ev_file, ucq);
240 kfree(ucq);
241 }
242
243 list_for_each_entry_safe(uobj, tmp, &context->srq_list, list) {
244 struct ib_srq *srq = uobj->object;
245 struct ib_uevent_object *uevent =
246 container_of(uobj, struct ib_uevent_object, uobject);
247
248 idr_remove_uobj(&ib_uverbs_srq_idr, uobj);
249 ib_destroy_srq(srq);
250 ib_uverbs_release_uevent(file, uevent);
251 kfree(uevent);
252 }
253
254 list_for_each_entry_safe(uobj, tmp, &context->mr_list, list) {
255 struct ib_mr *mr = uobj->object;
256
257 idr_remove_uobj(&ib_uverbs_mr_idr, uobj);
258 ib_dereg_mr(mr);
259 kfree(uobj);
260 }
261
262 mutex_lock(&file->device->xrcd_tree_mutex);
263 list_for_each_entry_safe(uobj, tmp, &context->xrcd_list, list) {
264 struct ib_xrcd *xrcd = uobj->object;
265 struct ib_uxrcd_object *uxrcd =
266 container_of(uobj, struct ib_uxrcd_object, uobject);
267
268 idr_remove_uobj(&ib_uverbs_xrcd_idr, uobj);
269 ib_uverbs_dealloc_xrcd(file->device, xrcd);
270 kfree(uxrcd);
271 }
272 mutex_unlock(&file->device->xrcd_tree_mutex);
273
274 list_for_each_entry_safe(uobj, tmp, &context->pd_list, list) {
275 struct ib_pd *pd = uobj->object;
276
277 idr_remove_uobj(&ib_uverbs_pd_idr, uobj);
278 ib_dealloc_pd(pd);
279 kfree(uobj);
280 }
281
282 return context->device->dealloc_ucontext(context);
283 }
284
ib_uverbs_release_file(struct kref * ref)285 static void ib_uverbs_release_file(struct kref *ref)
286 {
287 struct ib_uverbs_file *file =
288 container_of(ref, struct ib_uverbs_file, ref);
289
290 module_put(file->device->ib_dev->owner);
291 kref_put(&file->device->ref, ib_uverbs_release_dev);
292
293 kfree(file);
294 }
295
ib_uverbs_event_read(struct file * filp,char __user * buf,size_t count,loff_t * pos)296 static ssize_t ib_uverbs_event_read(struct file *filp, char __user *buf,
297 size_t count, loff_t *pos)
298 {
299 struct ib_uverbs_event_file *file = filp->private_data;
300 struct ib_uverbs_event *event;
301 int eventsz;
302 int ret = 0;
303
304 spin_lock_irq(&file->lock);
305
306 while (list_empty(&file->event_list)) {
307 spin_unlock_irq(&file->lock);
308
309 if (filp->f_flags & O_NONBLOCK)
310 return -EAGAIN;
311
312 if (wait_event_interruptible(file->poll_wait,
313 !list_empty(&file->event_list)))
314 return -ERESTARTSYS;
315
316 spin_lock_irq(&file->lock);
317 }
318
319 event = list_entry(file->event_list.next, struct ib_uverbs_event, list);
320
321 if (file->is_async)
322 eventsz = sizeof (struct ib_uverbs_async_event_desc);
323 else
324 eventsz = sizeof (struct ib_uverbs_comp_event_desc);
325
326 if (eventsz > count) {
327 ret = -EINVAL;
328 event = NULL;
329 } else {
330 list_del(file->event_list.next);
331 if (event->counter) {
332 ++(*event->counter);
333 list_del(&event->obj_list);
334 }
335 }
336
337 spin_unlock_irq(&file->lock);
338
339 if (event) {
340 if (copy_to_user(buf, event, eventsz))
341 ret = -EFAULT;
342 else
343 ret = eventsz;
344 }
345
346 kfree(event);
347
348 return ret;
349 }
350
ib_uverbs_event_poll(struct file * filp,struct poll_table_struct * wait)351 static unsigned int ib_uverbs_event_poll(struct file *filp,
352 struct poll_table_struct *wait)
353 {
354 unsigned int pollflags = 0;
355 struct ib_uverbs_event_file *file = filp->private_data;
356
357 poll_wait(filp, &file->poll_wait, wait);
358
359 spin_lock_irq(&file->lock);
360 if (!list_empty(&file->event_list))
361 pollflags = POLLIN | POLLRDNORM;
362 spin_unlock_irq(&file->lock);
363
364 return pollflags;
365 }
366
ib_uverbs_event_fasync(int fd,struct file * filp,int on)367 static int ib_uverbs_event_fasync(int fd, struct file *filp, int on)
368 {
369 struct ib_uverbs_event_file *file = filp->private_data;
370
371 return fasync_helper(fd, filp, on, &file->async_queue);
372 }
373
ib_uverbs_event_close(struct inode * inode,struct file * filp)374 static int ib_uverbs_event_close(struct inode *inode, struct file *filp)
375 {
376 struct ib_uverbs_event_file *file = filp->private_data;
377 struct ib_uverbs_event *entry, *tmp;
378
379 spin_lock_irq(&file->lock);
380 file->is_closed = 1;
381 list_for_each_entry_safe(entry, tmp, &file->event_list, list) {
382 if (entry->counter)
383 list_del(&entry->obj_list);
384 kfree(entry);
385 }
386 spin_unlock_irq(&file->lock);
387
388 if (file->is_async) {
389 ib_unregister_event_handler(&file->uverbs_file->event_handler);
390 kref_put(&file->uverbs_file->ref, ib_uverbs_release_file);
391 }
392 kref_put(&file->ref, ib_uverbs_release_event_file);
393
394 return 0;
395 }
396
397 static const struct file_operations uverbs_event_fops = {
398 .owner = THIS_MODULE,
399 .read = ib_uverbs_event_read,
400 .poll = ib_uverbs_event_poll,
401 .release = ib_uverbs_event_close,
402 .fasync = ib_uverbs_event_fasync,
403 .llseek = no_llseek,
404 };
405
ib_uverbs_comp_handler(struct ib_cq * cq,void * cq_context)406 void ib_uverbs_comp_handler(struct ib_cq *cq, void *cq_context)
407 {
408 struct ib_uverbs_event_file *file = cq_context;
409 struct ib_ucq_object *uobj;
410 struct ib_uverbs_event *entry;
411 unsigned long flags;
412
413 if (!file)
414 return;
415
416 spin_lock_irqsave(&file->lock, flags);
417 if (file->is_closed) {
418 spin_unlock_irqrestore(&file->lock, flags);
419 return;
420 }
421
422 entry = kmalloc(sizeof *entry, GFP_ATOMIC);
423 if (!entry) {
424 spin_unlock_irqrestore(&file->lock, flags);
425 return;
426 }
427
428 uobj = container_of(cq->uobject, struct ib_ucq_object, uobject);
429
430 entry->desc.comp.cq_handle = cq->uobject->user_handle;
431 entry->counter = &uobj->comp_events_reported;
432
433 list_add_tail(&entry->list, &file->event_list);
434 list_add_tail(&entry->obj_list, &uobj->comp_list);
435 spin_unlock_irqrestore(&file->lock, flags);
436
437 wake_up_interruptible(&file->poll_wait);
438 kill_fasync(&file->async_queue, SIGIO, POLL_IN);
439 }
440
ib_uverbs_async_handler(struct ib_uverbs_file * file,__u64 element,__u64 event,struct list_head * obj_list,u32 * counter)441 static void ib_uverbs_async_handler(struct ib_uverbs_file *file,
442 __u64 element, __u64 event,
443 struct list_head *obj_list,
444 u32 *counter)
445 {
446 struct ib_uverbs_event *entry;
447 unsigned long flags;
448
449 spin_lock_irqsave(&file->async_file->lock, flags);
450 if (file->async_file->is_closed) {
451 spin_unlock_irqrestore(&file->async_file->lock, flags);
452 return;
453 }
454
455 entry = kmalloc(sizeof *entry, GFP_ATOMIC);
456 if (!entry) {
457 spin_unlock_irqrestore(&file->async_file->lock, flags);
458 return;
459 }
460
461 entry->desc.async.element = element;
462 entry->desc.async.event_type = event;
463 entry->counter = counter;
464
465 list_add_tail(&entry->list, &file->async_file->event_list);
466 if (obj_list)
467 list_add_tail(&entry->obj_list, obj_list);
468 spin_unlock_irqrestore(&file->async_file->lock, flags);
469
470 wake_up_interruptible(&file->async_file->poll_wait);
471 kill_fasync(&file->async_file->async_queue, SIGIO, POLL_IN);
472 }
473
ib_uverbs_cq_event_handler(struct ib_event * event,void * context_ptr)474 void ib_uverbs_cq_event_handler(struct ib_event *event, void *context_ptr)
475 {
476 struct ib_ucq_object *uobj = container_of(event->element.cq->uobject,
477 struct ib_ucq_object, uobject);
478
479 ib_uverbs_async_handler(uobj->uverbs_file, uobj->uobject.user_handle,
480 event->event, &uobj->async_list,
481 &uobj->async_events_reported);
482 }
483
ib_uverbs_qp_event_handler(struct ib_event * event,void * context_ptr)484 void ib_uverbs_qp_event_handler(struct ib_event *event, void *context_ptr)
485 {
486 struct ib_uevent_object *uobj;
487
488 uobj = container_of(event->element.qp->uobject,
489 struct ib_uevent_object, uobject);
490
491 ib_uverbs_async_handler(context_ptr, uobj->uobject.user_handle,
492 event->event, &uobj->event_list,
493 &uobj->events_reported);
494 }
495
ib_uverbs_srq_event_handler(struct ib_event * event,void * context_ptr)496 void ib_uverbs_srq_event_handler(struct ib_event *event, void *context_ptr)
497 {
498 struct ib_uevent_object *uobj;
499
500 uobj = container_of(event->element.srq->uobject,
501 struct ib_uevent_object, uobject);
502
503 ib_uverbs_async_handler(context_ptr, uobj->uobject.user_handle,
504 event->event, &uobj->event_list,
505 &uobj->events_reported);
506 }
507
ib_uverbs_event_handler(struct ib_event_handler * handler,struct ib_event * event)508 void ib_uverbs_event_handler(struct ib_event_handler *handler,
509 struct ib_event *event)
510 {
511 struct ib_uverbs_file *file =
512 container_of(handler, struct ib_uverbs_file, event_handler);
513
514 ib_uverbs_async_handler(file, event->element.port_num, event->event,
515 NULL, NULL);
516 }
517
ib_uverbs_alloc_event_file(struct ib_uverbs_file * uverbs_file,int is_async)518 struct file *ib_uverbs_alloc_event_file(struct ib_uverbs_file *uverbs_file,
519 int is_async)
520 {
521 struct ib_uverbs_event_file *ev_file;
522 struct file *filp;
523
524 ev_file = kmalloc(sizeof *ev_file, GFP_KERNEL);
525 if (!ev_file)
526 return ERR_PTR(-ENOMEM);
527
528 kref_init(&ev_file->ref);
529 spin_lock_init(&ev_file->lock);
530 INIT_LIST_HEAD(&ev_file->event_list);
531 init_waitqueue_head(&ev_file->poll_wait);
532 ev_file->uverbs_file = uverbs_file;
533 ev_file->async_queue = NULL;
534 ev_file->is_async = is_async;
535 ev_file->is_closed = 0;
536
537 filp = anon_inode_getfile("[infinibandevent]", &uverbs_event_fops,
538 ev_file, O_RDONLY);
539 if (IS_ERR(filp))
540 kfree(ev_file);
541
542 return filp;
543 }
544
545 /*
546 * Look up a completion event file by FD. If lookup is successful,
547 * takes a ref to the event file struct that it returns; if
548 * unsuccessful, returns NULL.
549 */
ib_uverbs_lookup_comp_file(int fd)550 struct ib_uverbs_event_file *ib_uverbs_lookup_comp_file(int fd)
551 {
552 struct ib_uverbs_event_file *ev_file = NULL;
553 struct fd f = fdget(fd);
554
555 if (!f.file)
556 return NULL;
557
558 if (f.file->f_op != &uverbs_event_fops)
559 goto out;
560
561 ev_file = f.file->private_data;
562 if (ev_file->is_async) {
563 ev_file = NULL;
564 goto out;
565 }
566
567 kref_get(&ev_file->ref);
568
569 out:
570 fdput(f);
571 return ev_file;
572 }
573
ib_uverbs_write(struct file * filp,const char __user * buf,size_t count,loff_t * pos)574 static ssize_t ib_uverbs_write(struct file *filp, const char __user *buf,
575 size_t count, loff_t *pos)
576 {
577 struct ib_uverbs_file *file = filp->private_data;
578 struct ib_uverbs_cmd_hdr hdr;
579
580 if (count < sizeof hdr)
581 return -EINVAL;
582
583 if (copy_from_user(&hdr, buf, sizeof hdr))
584 return -EFAULT;
585
586 if (hdr.in_words * 4 != count)
587 return -EINVAL;
588
589 if (hdr.command >= ARRAY_SIZE(uverbs_cmd_table) ||
590 !uverbs_cmd_table[hdr.command])
591 return -EINVAL;
592
593 if (!file->ucontext &&
594 hdr.command != IB_USER_VERBS_CMD_GET_CONTEXT)
595 return -EINVAL;
596
597 if (!(file->device->ib_dev->uverbs_cmd_mask & (1ull << hdr.command)))
598 return -ENOSYS;
599
600 return uverbs_cmd_table[hdr.command](file, buf + sizeof hdr,
601 hdr.in_words * 4, hdr.out_words * 4);
602 }
603
ib_uverbs_mmap(struct file * filp,struct vm_area_struct * vma)604 static int ib_uverbs_mmap(struct file *filp, struct vm_area_struct *vma)
605 {
606 struct ib_uverbs_file *file = filp->private_data;
607
608 if (!file->ucontext)
609 return -ENODEV;
610 else
611 return file->device->ib_dev->mmap(file->ucontext, vma);
612 }
613
614 /*
615 * ib_uverbs_open() does not need the BKL:
616 *
617 * - the ib_uverbs_device structures are properly reference counted and
618 * everything else is purely local to the file being created, so
619 * races against other open calls are not a problem;
620 * - there is no ioctl method to race against;
621 * - the open method will either immediately run -ENXIO, or all
622 * required initialization will be done.
623 */
ib_uverbs_open(struct inode * inode,struct file * filp)624 static int ib_uverbs_open(struct inode *inode, struct file *filp)
625 {
626 struct ib_uverbs_device *dev;
627 struct ib_uverbs_file *file;
628 int ret;
629
630 dev = container_of(inode->i_cdev, struct ib_uverbs_device, cdev);
631 if (dev)
632 kref_get(&dev->ref);
633 else
634 return -ENXIO;
635
636 if (!try_module_get(dev->ib_dev->owner)) {
637 ret = -ENODEV;
638 goto err;
639 }
640
641 file = kmalloc(sizeof *file, GFP_KERNEL);
642 if (!file) {
643 ret = -ENOMEM;
644 goto err_module;
645 }
646
647 file->device = dev;
648 file->ucontext = NULL;
649 file->async_file = NULL;
650 kref_init(&file->ref);
651 mutex_init(&file->mutex);
652
653 filp->private_data = file;
654
655 return nonseekable_open(inode, filp);
656
657 err_module:
658 module_put(dev->ib_dev->owner);
659
660 err:
661 kref_put(&dev->ref, ib_uverbs_release_dev);
662 return ret;
663 }
664
ib_uverbs_close(struct inode * inode,struct file * filp)665 static int ib_uverbs_close(struct inode *inode, struct file *filp)
666 {
667 struct ib_uverbs_file *file = filp->private_data;
668
669 ib_uverbs_cleanup_ucontext(file, file->ucontext);
670
671 if (file->async_file)
672 kref_put(&file->async_file->ref, ib_uverbs_release_event_file);
673
674 kref_put(&file->ref, ib_uverbs_release_file);
675
676 return 0;
677 }
678
679 static const struct file_operations uverbs_fops = {
680 .owner = THIS_MODULE,
681 .write = ib_uverbs_write,
682 .open = ib_uverbs_open,
683 .release = ib_uverbs_close,
684 .llseek = no_llseek,
685 };
686
687 static const struct file_operations uverbs_mmap_fops = {
688 .owner = THIS_MODULE,
689 .write = ib_uverbs_write,
690 .mmap = ib_uverbs_mmap,
691 .open = ib_uverbs_open,
692 .release = ib_uverbs_close,
693 .llseek = no_llseek,
694 };
695
696 static struct ib_client uverbs_client = {
697 .name = "uverbs",
698 .add = ib_uverbs_add_one,
699 .remove = ib_uverbs_remove_one
700 };
701
show_ibdev(struct device * device,struct device_attribute * attr,char * buf)702 static ssize_t show_ibdev(struct device *device, struct device_attribute *attr,
703 char *buf)
704 {
705 struct ib_uverbs_device *dev = dev_get_drvdata(device);
706
707 if (!dev)
708 return -ENODEV;
709
710 return sprintf(buf, "%s\n", dev->ib_dev->name);
711 }
712 static DEVICE_ATTR(ibdev, S_IRUGO, show_ibdev, NULL);
713
show_dev_abi_version(struct device * device,struct device_attribute * attr,char * buf)714 static ssize_t show_dev_abi_version(struct device *device,
715 struct device_attribute *attr, char *buf)
716 {
717 struct ib_uverbs_device *dev = dev_get_drvdata(device);
718
719 if (!dev)
720 return -ENODEV;
721
722 return sprintf(buf, "%d\n", dev->ib_dev->uverbs_abi_ver);
723 }
724 static DEVICE_ATTR(abi_version, S_IRUGO, show_dev_abi_version, NULL);
725
726 static CLASS_ATTR_STRING(abi_version, S_IRUGO,
727 __stringify(IB_USER_VERBS_ABI_VERSION));
728
729 static dev_t overflow_maj;
730 static DECLARE_BITMAP(overflow_map, IB_UVERBS_MAX_DEVICES);
731
732 /*
733 * If we have more than IB_UVERBS_MAX_DEVICES, dynamically overflow by
734 * requesting a new major number and doubling the number of max devices we
735 * support. It's stupid, but simple.
736 */
find_overflow_devnum(void)737 static int find_overflow_devnum(void)
738 {
739 int ret;
740
741 if (!overflow_maj) {
742 ret = alloc_chrdev_region(&overflow_maj, 0, IB_UVERBS_MAX_DEVICES,
743 "infiniband_verbs");
744 if (ret) {
745 printk(KERN_ERR "user_verbs: couldn't register dynamic device number\n");
746 return ret;
747 }
748 }
749
750 ret = find_first_zero_bit(overflow_map, IB_UVERBS_MAX_DEVICES);
751 if (ret >= IB_UVERBS_MAX_DEVICES)
752 return -1;
753
754 return ret;
755 }
756
ib_uverbs_add_one(struct ib_device * device)757 static void ib_uverbs_add_one(struct ib_device *device)
758 {
759 int devnum;
760 dev_t base;
761 struct ib_uverbs_device *uverbs_dev;
762
763 if (!device->alloc_ucontext)
764 return;
765
766 uverbs_dev = kzalloc(sizeof *uverbs_dev, GFP_KERNEL);
767 if (!uverbs_dev)
768 return;
769
770 kref_init(&uverbs_dev->ref);
771 init_completion(&uverbs_dev->comp);
772 uverbs_dev->xrcd_tree = RB_ROOT;
773 mutex_init(&uverbs_dev->xrcd_tree_mutex);
774
775 spin_lock(&map_lock);
776 devnum = find_first_zero_bit(dev_map, IB_UVERBS_MAX_DEVICES);
777 if (devnum >= IB_UVERBS_MAX_DEVICES) {
778 spin_unlock(&map_lock);
779 devnum = find_overflow_devnum();
780 if (devnum < 0)
781 goto err;
782
783 spin_lock(&map_lock);
784 uverbs_dev->devnum = devnum + IB_UVERBS_MAX_DEVICES;
785 base = devnum + overflow_maj;
786 set_bit(devnum, overflow_map);
787 } else {
788 uverbs_dev->devnum = devnum;
789 base = devnum + IB_UVERBS_BASE_DEV;
790 set_bit(devnum, dev_map);
791 }
792 spin_unlock(&map_lock);
793
794 uverbs_dev->ib_dev = device;
795 uverbs_dev->num_comp_vectors = device->num_comp_vectors;
796
797 cdev_init(&uverbs_dev->cdev, NULL);
798 uverbs_dev->cdev.owner = THIS_MODULE;
799 uverbs_dev->cdev.ops = device->mmap ? &uverbs_mmap_fops : &uverbs_fops;
800 kobject_set_name(&uverbs_dev->cdev.kobj, "uverbs%d", uverbs_dev->devnum);
801 if (cdev_add(&uverbs_dev->cdev, base, 1))
802 goto err_cdev;
803
804 uverbs_dev->dev = device_create(uverbs_class, device->dma_device,
805 uverbs_dev->cdev.dev, uverbs_dev,
806 "uverbs%d", uverbs_dev->devnum);
807 if (IS_ERR(uverbs_dev->dev))
808 goto err_cdev;
809
810 if (device_create_file(uverbs_dev->dev, &dev_attr_ibdev))
811 goto err_class;
812 if (device_create_file(uverbs_dev->dev, &dev_attr_abi_version))
813 goto err_class;
814
815 ib_set_client_data(device, &uverbs_client, uverbs_dev);
816
817 return;
818
819 err_class:
820 device_destroy(uverbs_class, uverbs_dev->cdev.dev);
821
822 err_cdev:
823 cdev_del(&uverbs_dev->cdev);
824 if (uverbs_dev->devnum < IB_UVERBS_MAX_DEVICES)
825 clear_bit(devnum, dev_map);
826 else
827 clear_bit(devnum, overflow_map);
828
829 err:
830 kref_put(&uverbs_dev->ref, ib_uverbs_release_dev);
831 wait_for_completion(&uverbs_dev->comp);
832 kfree(uverbs_dev);
833 return;
834 }
835
ib_uverbs_remove_one(struct ib_device * device)836 static void ib_uverbs_remove_one(struct ib_device *device)
837 {
838 struct ib_uverbs_device *uverbs_dev = ib_get_client_data(device, &uverbs_client);
839
840 if (!uverbs_dev)
841 return;
842
843 dev_set_drvdata(uverbs_dev->dev, NULL);
844 device_destroy(uverbs_class, uverbs_dev->cdev.dev);
845 cdev_del(&uverbs_dev->cdev);
846
847 if (uverbs_dev->devnum < IB_UVERBS_MAX_DEVICES)
848 clear_bit(uverbs_dev->devnum, dev_map);
849 else
850 clear_bit(uverbs_dev->devnum - IB_UVERBS_MAX_DEVICES, overflow_map);
851
852 kref_put(&uverbs_dev->ref, ib_uverbs_release_dev);
853 wait_for_completion(&uverbs_dev->comp);
854 kfree(uverbs_dev);
855 }
856
uverbs_devnode(struct device * dev,umode_t * mode)857 static char *uverbs_devnode(struct device *dev, umode_t *mode)
858 {
859 if (mode)
860 *mode = 0666;
861 return kasprintf(GFP_KERNEL, "infiniband/%s", dev_name(dev));
862 }
863
ib_uverbs_init(void)864 static int __init ib_uverbs_init(void)
865 {
866 int ret;
867
868 ret = register_chrdev_region(IB_UVERBS_BASE_DEV, IB_UVERBS_MAX_DEVICES,
869 "infiniband_verbs");
870 if (ret) {
871 printk(KERN_ERR "user_verbs: couldn't register device number\n");
872 goto out;
873 }
874
875 uverbs_class = class_create(THIS_MODULE, "infiniband_verbs");
876 if (IS_ERR(uverbs_class)) {
877 ret = PTR_ERR(uverbs_class);
878 printk(KERN_ERR "user_verbs: couldn't create class infiniband_verbs\n");
879 goto out_chrdev;
880 }
881
882 uverbs_class->devnode = uverbs_devnode;
883
884 ret = class_create_file(uverbs_class, &class_attr_abi_version.attr);
885 if (ret) {
886 printk(KERN_ERR "user_verbs: couldn't create abi_version attribute\n");
887 goto out_class;
888 }
889
890 ret = ib_register_client(&uverbs_client);
891 if (ret) {
892 printk(KERN_ERR "user_verbs: couldn't register client\n");
893 goto out_class;
894 }
895
896 return 0;
897
898 out_class:
899 class_destroy(uverbs_class);
900
901 out_chrdev:
902 unregister_chrdev_region(IB_UVERBS_BASE_DEV, IB_UVERBS_MAX_DEVICES);
903
904 out:
905 return ret;
906 }
907
ib_uverbs_cleanup(void)908 static void __exit ib_uverbs_cleanup(void)
909 {
910 ib_unregister_client(&uverbs_client);
911 class_destroy(uverbs_class);
912 unregister_chrdev_region(IB_UVERBS_BASE_DEV, IB_UVERBS_MAX_DEVICES);
913 if (overflow_maj)
914 unregister_chrdev_region(overflow_maj, IB_UVERBS_MAX_DEVICES);
915 idr_destroy(&ib_uverbs_pd_idr);
916 idr_destroy(&ib_uverbs_mr_idr);
917 idr_destroy(&ib_uverbs_mw_idr);
918 idr_destroy(&ib_uverbs_ah_idr);
919 idr_destroy(&ib_uverbs_cq_idr);
920 idr_destroy(&ib_uverbs_qp_idr);
921 idr_destroy(&ib_uverbs_srq_idr);
922 }
923
924 module_init(ib_uverbs_init);
925 module_exit(ib_uverbs_cleanup);
926