xref: /linux/drivers/infiniband/core/uverbs_cmd.c (revision e0bf6c5ca2d3281f231c5f0c9bf145e9513644de)
1 /*
2  * Copyright (c) 2005 Topspin Communications.  All rights reserved.
3  * Copyright (c) 2005, 2006, 2007 Cisco Systems.  All rights reserved.
4  * Copyright (c) 2005 PathScale, Inc.  All rights reserved.
5  * Copyright (c) 2006 Mellanox Technologies.  All rights reserved.
6  *
7  * This software is available to you under a choice of one of two
8  * licenses.  You may choose to be licensed under the terms of the GNU
9  * General Public License (GPL) Version 2, available from the file
10  * COPYING in the main directory of this source tree, or the
11  * OpenIB.org BSD license below:
12  *
13  *     Redistribution and use in source and binary forms, with or
14  *     without modification, are permitted provided that the following
15  *     conditions are met:
16  *
17  *      - Redistributions of source code must retain the above
18  *        copyright notice, this list of conditions and the following
19  *        disclaimer.
20  *
21  *      - Redistributions in binary form must reproduce the above
22  *        copyright notice, this list of conditions and the following
23  *        disclaimer in the documentation and/or other materials
24  *        provided with the distribution.
25  *
26  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
27  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
28  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
29  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
30  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
31  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
32  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
33  * SOFTWARE.
34  */
35 
36 #include <linux/file.h>
37 #include <linux/fs.h>
38 #include <linux/slab.h>
39 #include <linux/sched.h>
40 
41 #include <asm/uaccess.h>
42 
43 #include "uverbs.h"
44 #include "core_priv.h"
45 
46 struct uverbs_lock_class {
47 	struct lock_class_key	key;
48 	char			name[16];
49 };
50 
51 static struct uverbs_lock_class pd_lock_class	= { .name = "PD-uobj" };
52 static struct uverbs_lock_class mr_lock_class	= { .name = "MR-uobj" };
53 static struct uverbs_lock_class mw_lock_class	= { .name = "MW-uobj" };
54 static struct uverbs_lock_class cq_lock_class	= { .name = "CQ-uobj" };
55 static struct uverbs_lock_class qp_lock_class	= { .name = "QP-uobj" };
56 static struct uverbs_lock_class ah_lock_class	= { .name = "AH-uobj" };
57 static struct uverbs_lock_class srq_lock_class	= { .name = "SRQ-uobj" };
58 static struct uverbs_lock_class xrcd_lock_class = { .name = "XRCD-uobj" };
59 static struct uverbs_lock_class rule_lock_class = { .name = "RULE-uobj" };
60 
61 /*
62  * The ib_uobject locking scheme is as follows:
63  *
64  * - ib_uverbs_idr_lock protects the uverbs idrs themselves, so it
65  *   needs to be held during all idr operations.  When an object is
66  *   looked up, a reference must be taken on the object's kref before
67  *   dropping this lock.
68  *
69  * - Each object also has an rwsem.  This rwsem must be held for
70  *   reading while an operation that uses the object is performed.
71  *   For example, while registering an MR, the associated PD's
72  *   uobject.mutex must be held for reading.  The rwsem must be held
73  *   for writing while initializing or destroying an object.
74  *
75  * - In addition, each object has a "live" flag.  If this flag is not
76  *   set, then lookups of the object will fail even if it is found in
77  *   the idr.  This handles a reader that blocks and does not acquire
78  *   the rwsem until after the object is destroyed.  The destroy
79  *   operation will set the live flag to 0 and then drop the rwsem;
80  *   this will allow the reader to acquire the rwsem, see that the
81  *   live flag is 0, and then drop the rwsem and its reference to
82  *   object.  The underlying storage will not be freed until the last
83  *   reference to the object is dropped.
84  */
85 
86 static void init_uobj(struct ib_uobject *uobj, u64 user_handle,
87 		      struct ib_ucontext *context, struct uverbs_lock_class *c)
88 {
89 	uobj->user_handle = user_handle;
90 	uobj->context     = context;
91 	kref_init(&uobj->ref);
92 	init_rwsem(&uobj->mutex);
93 	lockdep_set_class_and_name(&uobj->mutex, &c->key, c->name);
94 	uobj->live        = 0;
95 }
96 
97 static void release_uobj(struct kref *kref)
98 {
99 	kfree(container_of(kref, struct ib_uobject, ref));
100 }
101 
102 static void put_uobj(struct ib_uobject *uobj)
103 {
104 	kref_put(&uobj->ref, release_uobj);
105 }
106 
107 static void put_uobj_read(struct ib_uobject *uobj)
108 {
109 	up_read(&uobj->mutex);
110 	put_uobj(uobj);
111 }
112 
113 static void put_uobj_write(struct ib_uobject *uobj)
114 {
115 	up_write(&uobj->mutex);
116 	put_uobj(uobj);
117 }
118 
119 static int idr_add_uobj(struct idr *idr, struct ib_uobject *uobj)
120 {
121 	int ret;
122 
123 	idr_preload(GFP_KERNEL);
124 	spin_lock(&ib_uverbs_idr_lock);
125 
126 	ret = idr_alloc(idr, uobj, 0, 0, GFP_NOWAIT);
127 	if (ret >= 0)
128 		uobj->id = ret;
129 
130 	spin_unlock(&ib_uverbs_idr_lock);
131 	idr_preload_end();
132 
133 	return ret < 0 ? ret : 0;
134 }
135 
136 void idr_remove_uobj(struct idr *idr, struct ib_uobject *uobj)
137 {
138 	spin_lock(&ib_uverbs_idr_lock);
139 	idr_remove(idr, uobj->id);
140 	spin_unlock(&ib_uverbs_idr_lock);
141 }
142 
143 static struct ib_uobject *__idr_get_uobj(struct idr *idr, int id,
144 					 struct ib_ucontext *context)
145 {
146 	struct ib_uobject *uobj;
147 
148 	spin_lock(&ib_uverbs_idr_lock);
149 	uobj = idr_find(idr, id);
150 	if (uobj) {
151 		if (uobj->context == context)
152 			kref_get(&uobj->ref);
153 		else
154 			uobj = NULL;
155 	}
156 	spin_unlock(&ib_uverbs_idr_lock);
157 
158 	return uobj;
159 }
160 
161 static struct ib_uobject *idr_read_uobj(struct idr *idr, int id,
162 					struct ib_ucontext *context, int nested)
163 {
164 	struct ib_uobject *uobj;
165 
166 	uobj = __idr_get_uobj(idr, id, context);
167 	if (!uobj)
168 		return NULL;
169 
170 	if (nested)
171 		down_read_nested(&uobj->mutex, SINGLE_DEPTH_NESTING);
172 	else
173 		down_read(&uobj->mutex);
174 	if (!uobj->live) {
175 		put_uobj_read(uobj);
176 		return NULL;
177 	}
178 
179 	return uobj;
180 }
181 
182 static struct ib_uobject *idr_write_uobj(struct idr *idr, int id,
183 					 struct ib_ucontext *context)
184 {
185 	struct ib_uobject *uobj;
186 
187 	uobj = __idr_get_uobj(idr, id, context);
188 	if (!uobj)
189 		return NULL;
190 
191 	down_write(&uobj->mutex);
192 	if (!uobj->live) {
193 		put_uobj_write(uobj);
194 		return NULL;
195 	}
196 
197 	return uobj;
198 }
199 
200 static void *idr_read_obj(struct idr *idr, int id, struct ib_ucontext *context,
201 			  int nested)
202 {
203 	struct ib_uobject *uobj;
204 
205 	uobj = idr_read_uobj(idr, id, context, nested);
206 	return uobj ? uobj->object : NULL;
207 }
208 
209 static struct ib_pd *idr_read_pd(int pd_handle, struct ib_ucontext *context)
210 {
211 	return idr_read_obj(&ib_uverbs_pd_idr, pd_handle, context, 0);
212 }
213 
214 static void put_pd_read(struct ib_pd *pd)
215 {
216 	put_uobj_read(pd->uobject);
217 }
218 
219 static struct ib_cq *idr_read_cq(int cq_handle, struct ib_ucontext *context, int nested)
220 {
221 	return idr_read_obj(&ib_uverbs_cq_idr, cq_handle, context, nested);
222 }
223 
224 static void put_cq_read(struct ib_cq *cq)
225 {
226 	put_uobj_read(cq->uobject);
227 }
228 
229 static struct ib_ah *idr_read_ah(int ah_handle, struct ib_ucontext *context)
230 {
231 	return idr_read_obj(&ib_uverbs_ah_idr, ah_handle, context, 0);
232 }
233 
234 static void put_ah_read(struct ib_ah *ah)
235 {
236 	put_uobj_read(ah->uobject);
237 }
238 
239 static struct ib_qp *idr_read_qp(int qp_handle, struct ib_ucontext *context)
240 {
241 	return idr_read_obj(&ib_uverbs_qp_idr, qp_handle, context, 0);
242 }
243 
244 static struct ib_qp *idr_write_qp(int qp_handle, struct ib_ucontext *context)
245 {
246 	struct ib_uobject *uobj;
247 
248 	uobj = idr_write_uobj(&ib_uverbs_qp_idr, qp_handle, context);
249 	return uobj ? uobj->object : NULL;
250 }
251 
252 static void put_qp_read(struct ib_qp *qp)
253 {
254 	put_uobj_read(qp->uobject);
255 }
256 
257 static void put_qp_write(struct ib_qp *qp)
258 {
259 	put_uobj_write(qp->uobject);
260 }
261 
262 static struct ib_srq *idr_read_srq(int srq_handle, struct ib_ucontext *context)
263 {
264 	return idr_read_obj(&ib_uverbs_srq_idr, srq_handle, context, 0);
265 }
266 
267 static void put_srq_read(struct ib_srq *srq)
268 {
269 	put_uobj_read(srq->uobject);
270 }
271 
272 static struct ib_xrcd *idr_read_xrcd(int xrcd_handle, struct ib_ucontext *context,
273 				     struct ib_uobject **uobj)
274 {
275 	*uobj = idr_read_uobj(&ib_uverbs_xrcd_idr, xrcd_handle, context, 0);
276 	return *uobj ? (*uobj)->object : NULL;
277 }
278 
279 static void put_xrcd_read(struct ib_uobject *uobj)
280 {
281 	put_uobj_read(uobj);
282 }
283 
284 ssize_t ib_uverbs_get_context(struct ib_uverbs_file *file,
285 			      const char __user *buf,
286 			      int in_len, int out_len)
287 {
288 	struct ib_uverbs_get_context      cmd;
289 	struct ib_uverbs_get_context_resp resp;
290 	struct ib_udata                   udata;
291 	struct ib_device                 *ibdev = file->device->ib_dev;
292 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
293 	struct ib_device_attr		  dev_attr;
294 #endif
295 	struct ib_ucontext		 *ucontext;
296 	struct file			 *filp;
297 	int ret;
298 
299 	if (out_len < sizeof resp)
300 		return -ENOSPC;
301 
302 	if (copy_from_user(&cmd, buf, sizeof cmd))
303 		return -EFAULT;
304 
305 	mutex_lock(&file->mutex);
306 
307 	if (file->ucontext) {
308 		ret = -EINVAL;
309 		goto err;
310 	}
311 
312 	INIT_UDATA(&udata, buf + sizeof cmd,
313 		   (unsigned long) cmd.response + sizeof resp,
314 		   in_len - sizeof cmd, out_len - sizeof resp);
315 
316 	ucontext = ibdev->alloc_ucontext(ibdev, &udata);
317 	if (IS_ERR(ucontext)) {
318 		ret = PTR_ERR(ucontext);
319 		goto err;
320 	}
321 
322 	ucontext->device = ibdev;
323 	INIT_LIST_HEAD(&ucontext->pd_list);
324 	INIT_LIST_HEAD(&ucontext->mr_list);
325 	INIT_LIST_HEAD(&ucontext->mw_list);
326 	INIT_LIST_HEAD(&ucontext->cq_list);
327 	INIT_LIST_HEAD(&ucontext->qp_list);
328 	INIT_LIST_HEAD(&ucontext->srq_list);
329 	INIT_LIST_HEAD(&ucontext->ah_list);
330 	INIT_LIST_HEAD(&ucontext->xrcd_list);
331 	INIT_LIST_HEAD(&ucontext->rule_list);
332 	rcu_read_lock();
333 	ucontext->tgid = get_task_pid(current->group_leader, PIDTYPE_PID);
334 	rcu_read_unlock();
335 	ucontext->closing = 0;
336 
337 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
338 	ucontext->umem_tree = RB_ROOT;
339 	init_rwsem(&ucontext->umem_rwsem);
340 	ucontext->odp_mrs_count = 0;
341 	INIT_LIST_HEAD(&ucontext->no_private_counters);
342 
343 	ret = ib_query_device(ibdev, &dev_attr);
344 	if (ret)
345 		goto err_free;
346 	if (!(dev_attr.device_cap_flags & IB_DEVICE_ON_DEMAND_PAGING))
347 		ucontext->invalidate_range = NULL;
348 
349 #endif
350 
351 	resp.num_comp_vectors = file->device->num_comp_vectors;
352 
353 	ret = get_unused_fd_flags(O_CLOEXEC);
354 	if (ret < 0)
355 		goto err_free;
356 	resp.async_fd = ret;
357 
358 	filp = ib_uverbs_alloc_event_file(file, 1);
359 	if (IS_ERR(filp)) {
360 		ret = PTR_ERR(filp);
361 		goto err_fd;
362 	}
363 
364 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
365 			 &resp, sizeof resp)) {
366 		ret = -EFAULT;
367 		goto err_file;
368 	}
369 
370 	file->async_file = filp->private_data;
371 
372 	INIT_IB_EVENT_HANDLER(&file->event_handler, file->device->ib_dev,
373 			      ib_uverbs_event_handler);
374 	ret = ib_register_event_handler(&file->event_handler);
375 	if (ret)
376 		goto err_file;
377 
378 	kref_get(&file->async_file->ref);
379 	kref_get(&file->ref);
380 	file->ucontext = ucontext;
381 
382 	fd_install(resp.async_fd, filp);
383 
384 	mutex_unlock(&file->mutex);
385 
386 	return in_len;
387 
388 err_file:
389 	fput(filp);
390 
391 err_fd:
392 	put_unused_fd(resp.async_fd);
393 
394 err_free:
395 	put_pid(ucontext->tgid);
396 	ibdev->dealloc_ucontext(ucontext);
397 
398 err:
399 	mutex_unlock(&file->mutex);
400 	return ret;
401 }
402 
403 static void copy_query_dev_fields(struct ib_uverbs_file *file,
404 				  struct ib_uverbs_query_device_resp *resp,
405 				  struct ib_device_attr *attr)
406 {
407 	resp->fw_ver		= attr->fw_ver;
408 	resp->node_guid		= file->device->ib_dev->node_guid;
409 	resp->sys_image_guid	= attr->sys_image_guid;
410 	resp->max_mr_size	= attr->max_mr_size;
411 	resp->page_size_cap	= attr->page_size_cap;
412 	resp->vendor_id		= attr->vendor_id;
413 	resp->vendor_part_id	= attr->vendor_part_id;
414 	resp->hw_ver		= attr->hw_ver;
415 	resp->max_qp		= attr->max_qp;
416 	resp->max_qp_wr		= attr->max_qp_wr;
417 	resp->device_cap_flags	= attr->device_cap_flags;
418 	resp->max_sge		= attr->max_sge;
419 	resp->max_sge_rd	= attr->max_sge_rd;
420 	resp->max_cq		= attr->max_cq;
421 	resp->max_cqe		= attr->max_cqe;
422 	resp->max_mr		= attr->max_mr;
423 	resp->max_pd		= attr->max_pd;
424 	resp->max_qp_rd_atom	= attr->max_qp_rd_atom;
425 	resp->max_ee_rd_atom	= attr->max_ee_rd_atom;
426 	resp->max_res_rd_atom	= attr->max_res_rd_atom;
427 	resp->max_qp_init_rd_atom	= attr->max_qp_init_rd_atom;
428 	resp->max_ee_init_rd_atom	= attr->max_ee_init_rd_atom;
429 	resp->atomic_cap		= attr->atomic_cap;
430 	resp->max_ee			= attr->max_ee;
431 	resp->max_rdd			= attr->max_rdd;
432 	resp->max_mw			= attr->max_mw;
433 	resp->max_raw_ipv6_qp		= attr->max_raw_ipv6_qp;
434 	resp->max_raw_ethy_qp		= attr->max_raw_ethy_qp;
435 	resp->max_mcast_grp		= attr->max_mcast_grp;
436 	resp->max_mcast_qp_attach	= attr->max_mcast_qp_attach;
437 	resp->max_total_mcast_qp_attach	= attr->max_total_mcast_qp_attach;
438 	resp->max_ah			= attr->max_ah;
439 	resp->max_fmr			= attr->max_fmr;
440 	resp->max_map_per_fmr		= attr->max_map_per_fmr;
441 	resp->max_srq			= attr->max_srq;
442 	resp->max_srq_wr		= attr->max_srq_wr;
443 	resp->max_srq_sge		= attr->max_srq_sge;
444 	resp->max_pkeys			= attr->max_pkeys;
445 	resp->local_ca_ack_delay	= attr->local_ca_ack_delay;
446 	resp->phys_port_cnt		= file->device->ib_dev->phys_port_cnt;
447 }
448 
449 ssize_t ib_uverbs_query_device(struct ib_uverbs_file *file,
450 			       const char __user *buf,
451 			       int in_len, int out_len)
452 {
453 	struct ib_uverbs_query_device      cmd;
454 	struct ib_uverbs_query_device_resp resp;
455 	struct ib_device_attr              attr;
456 	int                                ret;
457 
458 	if (out_len < sizeof resp)
459 		return -ENOSPC;
460 
461 	if (copy_from_user(&cmd, buf, sizeof cmd))
462 		return -EFAULT;
463 
464 	ret = ib_query_device(file->device->ib_dev, &attr);
465 	if (ret)
466 		return ret;
467 
468 	memset(&resp, 0, sizeof resp);
469 	copy_query_dev_fields(file, &resp, &attr);
470 
471 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
472 			 &resp, sizeof resp))
473 		return -EFAULT;
474 
475 	return in_len;
476 }
477 
478 ssize_t ib_uverbs_query_port(struct ib_uverbs_file *file,
479 			     const char __user *buf,
480 			     int in_len, int out_len)
481 {
482 	struct ib_uverbs_query_port      cmd;
483 	struct ib_uverbs_query_port_resp resp;
484 	struct ib_port_attr              attr;
485 	int                              ret;
486 
487 	if (out_len < sizeof resp)
488 		return -ENOSPC;
489 
490 	if (copy_from_user(&cmd, buf, sizeof cmd))
491 		return -EFAULT;
492 
493 	ret = ib_query_port(file->device->ib_dev, cmd.port_num, &attr);
494 	if (ret)
495 		return ret;
496 
497 	memset(&resp, 0, sizeof resp);
498 
499 	resp.state 	     = attr.state;
500 	resp.max_mtu 	     = attr.max_mtu;
501 	resp.active_mtu      = attr.active_mtu;
502 	resp.gid_tbl_len     = attr.gid_tbl_len;
503 	resp.port_cap_flags  = attr.port_cap_flags;
504 	resp.max_msg_sz      = attr.max_msg_sz;
505 	resp.bad_pkey_cntr   = attr.bad_pkey_cntr;
506 	resp.qkey_viol_cntr  = attr.qkey_viol_cntr;
507 	resp.pkey_tbl_len    = attr.pkey_tbl_len;
508 	resp.lid 	     = attr.lid;
509 	resp.sm_lid 	     = attr.sm_lid;
510 	resp.lmc 	     = attr.lmc;
511 	resp.max_vl_num      = attr.max_vl_num;
512 	resp.sm_sl 	     = attr.sm_sl;
513 	resp.subnet_timeout  = attr.subnet_timeout;
514 	resp.init_type_reply = attr.init_type_reply;
515 	resp.active_width    = attr.active_width;
516 	resp.active_speed    = attr.active_speed;
517 	resp.phys_state      = attr.phys_state;
518 	resp.link_layer      = rdma_port_get_link_layer(file->device->ib_dev,
519 							cmd.port_num);
520 
521 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
522 			 &resp, sizeof resp))
523 		return -EFAULT;
524 
525 	return in_len;
526 }
527 
528 ssize_t ib_uverbs_alloc_pd(struct ib_uverbs_file *file,
529 			   const char __user *buf,
530 			   int in_len, int out_len)
531 {
532 	struct ib_uverbs_alloc_pd      cmd;
533 	struct ib_uverbs_alloc_pd_resp resp;
534 	struct ib_udata                udata;
535 	struct ib_uobject             *uobj;
536 	struct ib_pd                  *pd;
537 	int                            ret;
538 
539 	if (out_len < sizeof resp)
540 		return -ENOSPC;
541 
542 	if (copy_from_user(&cmd, buf, sizeof cmd))
543 		return -EFAULT;
544 
545 	INIT_UDATA(&udata, buf + sizeof cmd,
546 		   (unsigned long) cmd.response + sizeof resp,
547 		   in_len - sizeof cmd, out_len - sizeof resp);
548 
549 	uobj = kmalloc(sizeof *uobj, GFP_KERNEL);
550 	if (!uobj)
551 		return -ENOMEM;
552 
553 	init_uobj(uobj, 0, file->ucontext, &pd_lock_class);
554 	down_write(&uobj->mutex);
555 
556 	pd = file->device->ib_dev->alloc_pd(file->device->ib_dev,
557 					    file->ucontext, &udata);
558 	if (IS_ERR(pd)) {
559 		ret = PTR_ERR(pd);
560 		goto err;
561 	}
562 
563 	pd->device  = file->device->ib_dev;
564 	pd->uobject = uobj;
565 	atomic_set(&pd->usecnt, 0);
566 
567 	uobj->object = pd;
568 	ret = idr_add_uobj(&ib_uverbs_pd_idr, uobj);
569 	if (ret)
570 		goto err_idr;
571 
572 	memset(&resp, 0, sizeof resp);
573 	resp.pd_handle = uobj->id;
574 
575 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
576 			 &resp, sizeof resp)) {
577 		ret = -EFAULT;
578 		goto err_copy;
579 	}
580 
581 	mutex_lock(&file->mutex);
582 	list_add_tail(&uobj->list, &file->ucontext->pd_list);
583 	mutex_unlock(&file->mutex);
584 
585 	uobj->live = 1;
586 
587 	up_write(&uobj->mutex);
588 
589 	return in_len;
590 
591 err_copy:
592 	idr_remove_uobj(&ib_uverbs_pd_idr, uobj);
593 
594 err_idr:
595 	ib_dealloc_pd(pd);
596 
597 err:
598 	put_uobj_write(uobj);
599 	return ret;
600 }
601 
602 ssize_t ib_uverbs_dealloc_pd(struct ib_uverbs_file *file,
603 			     const char __user *buf,
604 			     int in_len, int out_len)
605 {
606 	struct ib_uverbs_dealloc_pd cmd;
607 	struct ib_uobject          *uobj;
608 	int                         ret;
609 
610 	if (copy_from_user(&cmd, buf, sizeof cmd))
611 		return -EFAULT;
612 
613 	uobj = idr_write_uobj(&ib_uverbs_pd_idr, cmd.pd_handle, file->ucontext);
614 	if (!uobj)
615 		return -EINVAL;
616 
617 	ret = ib_dealloc_pd(uobj->object);
618 	if (!ret)
619 		uobj->live = 0;
620 
621 	put_uobj_write(uobj);
622 
623 	if (ret)
624 		return ret;
625 
626 	idr_remove_uobj(&ib_uverbs_pd_idr, uobj);
627 
628 	mutex_lock(&file->mutex);
629 	list_del(&uobj->list);
630 	mutex_unlock(&file->mutex);
631 
632 	put_uobj(uobj);
633 
634 	return in_len;
635 }
636 
637 struct xrcd_table_entry {
638 	struct rb_node  node;
639 	struct ib_xrcd *xrcd;
640 	struct inode   *inode;
641 };
642 
643 static int xrcd_table_insert(struct ib_uverbs_device *dev,
644 			    struct inode *inode,
645 			    struct ib_xrcd *xrcd)
646 {
647 	struct xrcd_table_entry *entry, *scan;
648 	struct rb_node **p = &dev->xrcd_tree.rb_node;
649 	struct rb_node *parent = NULL;
650 
651 	entry = kmalloc(sizeof *entry, GFP_KERNEL);
652 	if (!entry)
653 		return -ENOMEM;
654 
655 	entry->xrcd  = xrcd;
656 	entry->inode = inode;
657 
658 	while (*p) {
659 		parent = *p;
660 		scan = rb_entry(parent, struct xrcd_table_entry, node);
661 
662 		if (inode < scan->inode) {
663 			p = &(*p)->rb_left;
664 		} else if (inode > scan->inode) {
665 			p = &(*p)->rb_right;
666 		} else {
667 			kfree(entry);
668 			return -EEXIST;
669 		}
670 	}
671 
672 	rb_link_node(&entry->node, parent, p);
673 	rb_insert_color(&entry->node, &dev->xrcd_tree);
674 	igrab(inode);
675 	return 0;
676 }
677 
678 static struct xrcd_table_entry *xrcd_table_search(struct ib_uverbs_device *dev,
679 						  struct inode *inode)
680 {
681 	struct xrcd_table_entry *entry;
682 	struct rb_node *p = dev->xrcd_tree.rb_node;
683 
684 	while (p) {
685 		entry = rb_entry(p, struct xrcd_table_entry, node);
686 
687 		if (inode < entry->inode)
688 			p = p->rb_left;
689 		else if (inode > entry->inode)
690 			p = p->rb_right;
691 		else
692 			return entry;
693 	}
694 
695 	return NULL;
696 }
697 
698 static struct ib_xrcd *find_xrcd(struct ib_uverbs_device *dev, struct inode *inode)
699 {
700 	struct xrcd_table_entry *entry;
701 
702 	entry = xrcd_table_search(dev, inode);
703 	if (!entry)
704 		return NULL;
705 
706 	return entry->xrcd;
707 }
708 
709 static void xrcd_table_delete(struct ib_uverbs_device *dev,
710 			      struct inode *inode)
711 {
712 	struct xrcd_table_entry *entry;
713 
714 	entry = xrcd_table_search(dev, inode);
715 	if (entry) {
716 		iput(inode);
717 		rb_erase(&entry->node, &dev->xrcd_tree);
718 		kfree(entry);
719 	}
720 }
721 
722 ssize_t ib_uverbs_open_xrcd(struct ib_uverbs_file *file,
723 			    const char __user *buf, int in_len,
724 			    int out_len)
725 {
726 	struct ib_uverbs_open_xrcd	cmd;
727 	struct ib_uverbs_open_xrcd_resp	resp;
728 	struct ib_udata			udata;
729 	struct ib_uxrcd_object         *obj;
730 	struct ib_xrcd                 *xrcd = NULL;
731 	struct fd			f = {NULL, 0};
732 	struct inode                   *inode = NULL;
733 	int				ret = 0;
734 	int				new_xrcd = 0;
735 
736 	if (out_len < sizeof resp)
737 		return -ENOSPC;
738 
739 	if (copy_from_user(&cmd, buf, sizeof cmd))
740 		return -EFAULT;
741 
742 	INIT_UDATA(&udata, buf + sizeof cmd,
743 		   (unsigned long) cmd.response + sizeof resp,
744 		   in_len - sizeof cmd, out_len - sizeof  resp);
745 
746 	mutex_lock(&file->device->xrcd_tree_mutex);
747 
748 	if (cmd.fd != -1) {
749 		/* search for file descriptor */
750 		f = fdget(cmd.fd);
751 		if (!f.file) {
752 			ret = -EBADF;
753 			goto err_tree_mutex_unlock;
754 		}
755 
756 		inode = file_inode(f.file);
757 		xrcd = find_xrcd(file->device, inode);
758 		if (!xrcd && !(cmd.oflags & O_CREAT)) {
759 			/* no file descriptor. Need CREATE flag */
760 			ret = -EAGAIN;
761 			goto err_tree_mutex_unlock;
762 		}
763 
764 		if (xrcd && cmd.oflags & O_EXCL) {
765 			ret = -EINVAL;
766 			goto err_tree_mutex_unlock;
767 		}
768 	}
769 
770 	obj = kmalloc(sizeof *obj, GFP_KERNEL);
771 	if (!obj) {
772 		ret = -ENOMEM;
773 		goto err_tree_mutex_unlock;
774 	}
775 
776 	init_uobj(&obj->uobject, 0, file->ucontext, &xrcd_lock_class);
777 
778 	down_write(&obj->uobject.mutex);
779 
780 	if (!xrcd) {
781 		xrcd = file->device->ib_dev->alloc_xrcd(file->device->ib_dev,
782 							file->ucontext, &udata);
783 		if (IS_ERR(xrcd)) {
784 			ret = PTR_ERR(xrcd);
785 			goto err;
786 		}
787 
788 		xrcd->inode   = inode;
789 		xrcd->device  = file->device->ib_dev;
790 		atomic_set(&xrcd->usecnt, 0);
791 		mutex_init(&xrcd->tgt_qp_mutex);
792 		INIT_LIST_HEAD(&xrcd->tgt_qp_list);
793 		new_xrcd = 1;
794 	}
795 
796 	atomic_set(&obj->refcnt, 0);
797 	obj->uobject.object = xrcd;
798 	ret = idr_add_uobj(&ib_uverbs_xrcd_idr, &obj->uobject);
799 	if (ret)
800 		goto err_idr;
801 
802 	memset(&resp, 0, sizeof resp);
803 	resp.xrcd_handle = obj->uobject.id;
804 
805 	if (inode) {
806 		if (new_xrcd) {
807 			/* create new inode/xrcd table entry */
808 			ret = xrcd_table_insert(file->device, inode, xrcd);
809 			if (ret)
810 				goto err_insert_xrcd;
811 		}
812 		atomic_inc(&xrcd->usecnt);
813 	}
814 
815 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
816 			 &resp, sizeof resp)) {
817 		ret = -EFAULT;
818 		goto err_copy;
819 	}
820 
821 	if (f.file)
822 		fdput(f);
823 
824 	mutex_lock(&file->mutex);
825 	list_add_tail(&obj->uobject.list, &file->ucontext->xrcd_list);
826 	mutex_unlock(&file->mutex);
827 
828 	obj->uobject.live = 1;
829 	up_write(&obj->uobject.mutex);
830 
831 	mutex_unlock(&file->device->xrcd_tree_mutex);
832 	return in_len;
833 
834 err_copy:
835 	if (inode) {
836 		if (new_xrcd)
837 			xrcd_table_delete(file->device, inode);
838 		atomic_dec(&xrcd->usecnt);
839 	}
840 
841 err_insert_xrcd:
842 	idr_remove_uobj(&ib_uverbs_xrcd_idr, &obj->uobject);
843 
844 err_idr:
845 	ib_dealloc_xrcd(xrcd);
846 
847 err:
848 	put_uobj_write(&obj->uobject);
849 
850 err_tree_mutex_unlock:
851 	if (f.file)
852 		fdput(f);
853 
854 	mutex_unlock(&file->device->xrcd_tree_mutex);
855 
856 	return ret;
857 }
858 
859 ssize_t ib_uverbs_close_xrcd(struct ib_uverbs_file *file,
860 			     const char __user *buf, int in_len,
861 			     int out_len)
862 {
863 	struct ib_uverbs_close_xrcd cmd;
864 	struct ib_uobject           *uobj;
865 	struct ib_xrcd              *xrcd = NULL;
866 	struct inode                *inode = NULL;
867 	struct ib_uxrcd_object      *obj;
868 	int                         live;
869 	int                         ret = 0;
870 
871 	if (copy_from_user(&cmd, buf, sizeof cmd))
872 		return -EFAULT;
873 
874 	mutex_lock(&file->device->xrcd_tree_mutex);
875 	uobj = idr_write_uobj(&ib_uverbs_xrcd_idr, cmd.xrcd_handle, file->ucontext);
876 	if (!uobj) {
877 		ret = -EINVAL;
878 		goto out;
879 	}
880 
881 	xrcd  = uobj->object;
882 	inode = xrcd->inode;
883 	obj   = container_of(uobj, struct ib_uxrcd_object, uobject);
884 	if (atomic_read(&obj->refcnt)) {
885 		put_uobj_write(uobj);
886 		ret = -EBUSY;
887 		goto out;
888 	}
889 
890 	if (!inode || atomic_dec_and_test(&xrcd->usecnt)) {
891 		ret = ib_dealloc_xrcd(uobj->object);
892 		if (!ret)
893 			uobj->live = 0;
894 	}
895 
896 	live = uobj->live;
897 	if (inode && ret)
898 		atomic_inc(&xrcd->usecnt);
899 
900 	put_uobj_write(uobj);
901 
902 	if (ret)
903 		goto out;
904 
905 	if (inode && !live)
906 		xrcd_table_delete(file->device, inode);
907 
908 	idr_remove_uobj(&ib_uverbs_xrcd_idr, uobj);
909 	mutex_lock(&file->mutex);
910 	list_del(&uobj->list);
911 	mutex_unlock(&file->mutex);
912 
913 	put_uobj(uobj);
914 	ret = in_len;
915 
916 out:
917 	mutex_unlock(&file->device->xrcd_tree_mutex);
918 	return ret;
919 }
920 
921 void ib_uverbs_dealloc_xrcd(struct ib_uverbs_device *dev,
922 			    struct ib_xrcd *xrcd)
923 {
924 	struct inode *inode;
925 
926 	inode = xrcd->inode;
927 	if (inode && !atomic_dec_and_test(&xrcd->usecnt))
928 		return;
929 
930 	ib_dealloc_xrcd(xrcd);
931 
932 	if (inode)
933 		xrcd_table_delete(dev, inode);
934 }
935 
936 ssize_t ib_uverbs_reg_mr(struct ib_uverbs_file *file,
937 			 const char __user *buf, int in_len,
938 			 int out_len)
939 {
940 	struct ib_uverbs_reg_mr      cmd;
941 	struct ib_uverbs_reg_mr_resp resp;
942 	struct ib_udata              udata;
943 	struct ib_uobject           *uobj;
944 	struct ib_pd                *pd;
945 	struct ib_mr                *mr;
946 	int                          ret;
947 
948 	if (out_len < sizeof resp)
949 		return -ENOSPC;
950 
951 	if (copy_from_user(&cmd, buf, sizeof cmd))
952 		return -EFAULT;
953 
954 	INIT_UDATA(&udata, buf + sizeof cmd,
955 		   (unsigned long) cmd.response + sizeof resp,
956 		   in_len - sizeof cmd, out_len - sizeof resp);
957 
958 	if ((cmd.start & ~PAGE_MASK) != (cmd.hca_va & ~PAGE_MASK))
959 		return -EINVAL;
960 
961 	ret = ib_check_mr_access(cmd.access_flags);
962 	if (ret)
963 		return ret;
964 
965 	uobj = kmalloc(sizeof *uobj, GFP_KERNEL);
966 	if (!uobj)
967 		return -ENOMEM;
968 
969 	init_uobj(uobj, 0, file->ucontext, &mr_lock_class);
970 	down_write(&uobj->mutex);
971 
972 	pd = idr_read_pd(cmd.pd_handle, file->ucontext);
973 	if (!pd) {
974 		ret = -EINVAL;
975 		goto err_free;
976 	}
977 
978 	if (cmd.access_flags & IB_ACCESS_ON_DEMAND) {
979 		struct ib_device_attr attr;
980 
981 		ret = ib_query_device(pd->device, &attr);
982 		if (ret || !(attr.device_cap_flags &
983 				IB_DEVICE_ON_DEMAND_PAGING)) {
984 			pr_debug("ODP support not available\n");
985 			ret = -EINVAL;
986 			goto err_put;
987 		}
988 	}
989 
990 	mr = pd->device->reg_user_mr(pd, cmd.start, cmd.length, cmd.hca_va,
991 				     cmd.access_flags, &udata);
992 	if (IS_ERR(mr)) {
993 		ret = PTR_ERR(mr);
994 		goto err_put;
995 	}
996 
997 	mr->device  = pd->device;
998 	mr->pd      = pd;
999 	mr->uobject = uobj;
1000 	atomic_inc(&pd->usecnt);
1001 	atomic_set(&mr->usecnt, 0);
1002 
1003 	uobj->object = mr;
1004 	ret = idr_add_uobj(&ib_uverbs_mr_idr, uobj);
1005 	if (ret)
1006 		goto err_unreg;
1007 
1008 	memset(&resp, 0, sizeof resp);
1009 	resp.lkey      = mr->lkey;
1010 	resp.rkey      = mr->rkey;
1011 	resp.mr_handle = uobj->id;
1012 
1013 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
1014 			 &resp, sizeof resp)) {
1015 		ret = -EFAULT;
1016 		goto err_copy;
1017 	}
1018 
1019 	put_pd_read(pd);
1020 
1021 	mutex_lock(&file->mutex);
1022 	list_add_tail(&uobj->list, &file->ucontext->mr_list);
1023 	mutex_unlock(&file->mutex);
1024 
1025 	uobj->live = 1;
1026 
1027 	up_write(&uobj->mutex);
1028 
1029 	return in_len;
1030 
1031 err_copy:
1032 	idr_remove_uobj(&ib_uverbs_mr_idr, uobj);
1033 
1034 err_unreg:
1035 	ib_dereg_mr(mr);
1036 
1037 err_put:
1038 	put_pd_read(pd);
1039 
1040 err_free:
1041 	put_uobj_write(uobj);
1042 	return ret;
1043 }
1044 
1045 ssize_t ib_uverbs_rereg_mr(struct ib_uverbs_file *file,
1046 			   const char __user *buf, int in_len,
1047 			   int out_len)
1048 {
1049 	struct ib_uverbs_rereg_mr      cmd;
1050 	struct ib_uverbs_rereg_mr_resp resp;
1051 	struct ib_udata              udata;
1052 	struct ib_pd                *pd = NULL;
1053 	struct ib_mr                *mr;
1054 	struct ib_pd		    *old_pd;
1055 	int                          ret;
1056 	struct ib_uobject	    *uobj;
1057 
1058 	if (out_len < sizeof(resp))
1059 		return -ENOSPC;
1060 
1061 	if (copy_from_user(&cmd, buf, sizeof(cmd)))
1062 		return -EFAULT;
1063 
1064 	INIT_UDATA(&udata, buf + sizeof(cmd),
1065 		   (unsigned long) cmd.response + sizeof(resp),
1066 		   in_len - sizeof(cmd), out_len - sizeof(resp));
1067 
1068 	if (cmd.flags & ~IB_MR_REREG_SUPPORTED || !cmd.flags)
1069 		return -EINVAL;
1070 
1071 	if ((cmd.flags & IB_MR_REREG_TRANS) &&
1072 	    (!cmd.start || !cmd.hca_va || 0 >= cmd.length ||
1073 	     (cmd.start & ~PAGE_MASK) != (cmd.hca_va & ~PAGE_MASK)))
1074 			return -EINVAL;
1075 
1076 	uobj = idr_write_uobj(&ib_uverbs_mr_idr, cmd.mr_handle,
1077 			      file->ucontext);
1078 
1079 	if (!uobj)
1080 		return -EINVAL;
1081 
1082 	mr = uobj->object;
1083 
1084 	if (cmd.flags & IB_MR_REREG_ACCESS) {
1085 		ret = ib_check_mr_access(cmd.access_flags);
1086 		if (ret)
1087 			goto put_uobjs;
1088 	}
1089 
1090 	if (cmd.flags & IB_MR_REREG_PD) {
1091 		pd = idr_read_pd(cmd.pd_handle, file->ucontext);
1092 		if (!pd) {
1093 			ret = -EINVAL;
1094 			goto put_uobjs;
1095 		}
1096 	}
1097 
1098 	if (atomic_read(&mr->usecnt)) {
1099 		ret = -EBUSY;
1100 		goto put_uobj_pd;
1101 	}
1102 
1103 	old_pd = mr->pd;
1104 	ret = mr->device->rereg_user_mr(mr, cmd.flags, cmd.start,
1105 					cmd.length, cmd.hca_va,
1106 					cmd.access_flags, pd, &udata);
1107 	if (!ret) {
1108 		if (cmd.flags & IB_MR_REREG_PD) {
1109 			atomic_inc(&pd->usecnt);
1110 			mr->pd = pd;
1111 			atomic_dec(&old_pd->usecnt);
1112 		}
1113 	} else {
1114 		goto put_uobj_pd;
1115 	}
1116 
1117 	memset(&resp, 0, sizeof(resp));
1118 	resp.lkey      = mr->lkey;
1119 	resp.rkey      = mr->rkey;
1120 
1121 	if (copy_to_user((void __user *)(unsigned long)cmd.response,
1122 			 &resp, sizeof(resp)))
1123 		ret = -EFAULT;
1124 	else
1125 		ret = in_len;
1126 
1127 put_uobj_pd:
1128 	if (cmd.flags & IB_MR_REREG_PD)
1129 		put_pd_read(pd);
1130 
1131 put_uobjs:
1132 
1133 	put_uobj_write(mr->uobject);
1134 
1135 	return ret;
1136 }
1137 
1138 ssize_t ib_uverbs_dereg_mr(struct ib_uverbs_file *file,
1139 			   const char __user *buf, int in_len,
1140 			   int out_len)
1141 {
1142 	struct ib_uverbs_dereg_mr cmd;
1143 	struct ib_mr             *mr;
1144 	struct ib_uobject	 *uobj;
1145 	int                       ret = -EINVAL;
1146 
1147 	if (copy_from_user(&cmd, buf, sizeof cmd))
1148 		return -EFAULT;
1149 
1150 	uobj = idr_write_uobj(&ib_uverbs_mr_idr, cmd.mr_handle, file->ucontext);
1151 	if (!uobj)
1152 		return -EINVAL;
1153 
1154 	mr = uobj->object;
1155 
1156 	ret = ib_dereg_mr(mr);
1157 	if (!ret)
1158 		uobj->live = 0;
1159 
1160 	put_uobj_write(uobj);
1161 
1162 	if (ret)
1163 		return ret;
1164 
1165 	idr_remove_uobj(&ib_uverbs_mr_idr, uobj);
1166 
1167 	mutex_lock(&file->mutex);
1168 	list_del(&uobj->list);
1169 	mutex_unlock(&file->mutex);
1170 
1171 	put_uobj(uobj);
1172 
1173 	return in_len;
1174 }
1175 
1176 ssize_t ib_uverbs_alloc_mw(struct ib_uverbs_file *file,
1177 			 const char __user *buf, int in_len,
1178 			 int out_len)
1179 {
1180 	struct ib_uverbs_alloc_mw      cmd;
1181 	struct ib_uverbs_alloc_mw_resp resp;
1182 	struct ib_uobject             *uobj;
1183 	struct ib_pd                  *pd;
1184 	struct ib_mw                  *mw;
1185 	int                            ret;
1186 
1187 	if (out_len < sizeof(resp))
1188 		return -ENOSPC;
1189 
1190 	if (copy_from_user(&cmd, buf, sizeof(cmd)))
1191 		return -EFAULT;
1192 
1193 	uobj = kmalloc(sizeof(*uobj), GFP_KERNEL);
1194 	if (!uobj)
1195 		return -ENOMEM;
1196 
1197 	init_uobj(uobj, 0, file->ucontext, &mw_lock_class);
1198 	down_write(&uobj->mutex);
1199 
1200 	pd = idr_read_pd(cmd.pd_handle, file->ucontext);
1201 	if (!pd) {
1202 		ret = -EINVAL;
1203 		goto err_free;
1204 	}
1205 
1206 	mw = pd->device->alloc_mw(pd, cmd.mw_type);
1207 	if (IS_ERR(mw)) {
1208 		ret = PTR_ERR(mw);
1209 		goto err_put;
1210 	}
1211 
1212 	mw->device  = pd->device;
1213 	mw->pd      = pd;
1214 	mw->uobject = uobj;
1215 	atomic_inc(&pd->usecnt);
1216 
1217 	uobj->object = mw;
1218 	ret = idr_add_uobj(&ib_uverbs_mw_idr, uobj);
1219 	if (ret)
1220 		goto err_unalloc;
1221 
1222 	memset(&resp, 0, sizeof(resp));
1223 	resp.rkey      = mw->rkey;
1224 	resp.mw_handle = uobj->id;
1225 
1226 	if (copy_to_user((void __user *)(unsigned long)cmd.response,
1227 			 &resp, sizeof(resp))) {
1228 		ret = -EFAULT;
1229 		goto err_copy;
1230 	}
1231 
1232 	put_pd_read(pd);
1233 
1234 	mutex_lock(&file->mutex);
1235 	list_add_tail(&uobj->list, &file->ucontext->mw_list);
1236 	mutex_unlock(&file->mutex);
1237 
1238 	uobj->live = 1;
1239 
1240 	up_write(&uobj->mutex);
1241 
1242 	return in_len;
1243 
1244 err_copy:
1245 	idr_remove_uobj(&ib_uverbs_mw_idr, uobj);
1246 
1247 err_unalloc:
1248 	ib_dealloc_mw(mw);
1249 
1250 err_put:
1251 	put_pd_read(pd);
1252 
1253 err_free:
1254 	put_uobj_write(uobj);
1255 	return ret;
1256 }
1257 
1258 ssize_t ib_uverbs_dealloc_mw(struct ib_uverbs_file *file,
1259 			   const char __user *buf, int in_len,
1260 			   int out_len)
1261 {
1262 	struct ib_uverbs_dealloc_mw cmd;
1263 	struct ib_mw               *mw;
1264 	struct ib_uobject	   *uobj;
1265 	int                         ret = -EINVAL;
1266 
1267 	if (copy_from_user(&cmd, buf, sizeof(cmd)))
1268 		return -EFAULT;
1269 
1270 	uobj = idr_write_uobj(&ib_uverbs_mw_idr, cmd.mw_handle, file->ucontext);
1271 	if (!uobj)
1272 		return -EINVAL;
1273 
1274 	mw = uobj->object;
1275 
1276 	ret = ib_dealloc_mw(mw);
1277 	if (!ret)
1278 		uobj->live = 0;
1279 
1280 	put_uobj_write(uobj);
1281 
1282 	if (ret)
1283 		return ret;
1284 
1285 	idr_remove_uobj(&ib_uverbs_mw_idr, uobj);
1286 
1287 	mutex_lock(&file->mutex);
1288 	list_del(&uobj->list);
1289 	mutex_unlock(&file->mutex);
1290 
1291 	put_uobj(uobj);
1292 
1293 	return in_len;
1294 }
1295 
1296 ssize_t ib_uverbs_create_comp_channel(struct ib_uverbs_file *file,
1297 				      const char __user *buf, int in_len,
1298 				      int out_len)
1299 {
1300 	struct ib_uverbs_create_comp_channel	   cmd;
1301 	struct ib_uverbs_create_comp_channel_resp  resp;
1302 	struct file				  *filp;
1303 	int ret;
1304 
1305 	if (out_len < sizeof resp)
1306 		return -ENOSPC;
1307 
1308 	if (copy_from_user(&cmd, buf, sizeof cmd))
1309 		return -EFAULT;
1310 
1311 	ret = get_unused_fd_flags(O_CLOEXEC);
1312 	if (ret < 0)
1313 		return ret;
1314 	resp.fd = ret;
1315 
1316 	filp = ib_uverbs_alloc_event_file(file, 0);
1317 	if (IS_ERR(filp)) {
1318 		put_unused_fd(resp.fd);
1319 		return PTR_ERR(filp);
1320 	}
1321 
1322 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
1323 			 &resp, sizeof resp)) {
1324 		put_unused_fd(resp.fd);
1325 		fput(filp);
1326 		return -EFAULT;
1327 	}
1328 
1329 	fd_install(resp.fd, filp);
1330 	return in_len;
1331 }
1332 
1333 ssize_t ib_uverbs_create_cq(struct ib_uverbs_file *file,
1334 			    const char __user *buf, int in_len,
1335 			    int out_len)
1336 {
1337 	struct ib_uverbs_create_cq      cmd;
1338 	struct ib_uverbs_create_cq_resp resp;
1339 	struct ib_udata                 udata;
1340 	struct ib_ucq_object           *obj;
1341 	struct ib_uverbs_event_file    *ev_file = NULL;
1342 	struct ib_cq                   *cq;
1343 	int                             ret;
1344 
1345 	if (out_len < sizeof resp)
1346 		return -ENOSPC;
1347 
1348 	if (copy_from_user(&cmd, buf, sizeof cmd))
1349 		return -EFAULT;
1350 
1351 	INIT_UDATA(&udata, buf + sizeof cmd,
1352 		   (unsigned long) cmd.response + sizeof resp,
1353 		   in_len - sizeof cmd, out_len - sizeof resp);
1354 
1355 	if (cmd.comp_vector >= file->device->num_comp_vectors)
1356 		return -EINVAL;
1357 
1358 	obj = kmalloc(sizeof *obj, GFP_KERNEL);
1359 	if (!obj)
1360 		return -ENOMEM;
1361 
1362 	init_uobj(&obj->uobject, cmd.user_handle, file->ucontext, &cq_lock_class);
1363 	down_write(&obj->uobject.mutex);
1364 
1365 	if (cmd.comp_channel >= 0) {
1366 		ev_file = ib_uverbs_lookup_comp_file(cmd.comp_channel);
1367 		if (!ev_file) {
1368 			ret = -EINVAL;
1369 			goto err;
1370 		}
1371 	}
1372 
1373 	obj->uverbs_file	   = file;
1374 	obj->comp_events_reported  = 0;
1375 	obj->async_events_reported = 0;
1376 	INIT_LIST_HEAD(&obj->comp_list);
1377 	INIT_LIST_HEAD(&obj->async_list);
1378 
1379 	cq = file->device->ib_dev->create_cq(file->device->ib_dev, cmd.cqe,
1380 					     cmd.comp_vector,
1381 					     file->ucontext, &udata);
1382 	if (IS_ERR(cq)) {
1383 		ret = PTR_ERR(cq);
1384 		goto err_file;
1385 	}
1386 
1387 	cq->device        = file->device->ib_dev;
1388 	cq->uobject       = &obj->uobject;
1389 	cq->comp_handler  = ib_uverbs_comp_handler;
1390 	cq->event_handler = ib_uverbs_cq_event_handler;
1391 	cq->cq_context    = ev_file;
1392 	atomic_set(&cq->usecnt, 0);
1393 
1394 	obj->uobject.object = cq;
1395 	ret = idr_add_uobj(&ib_uverbs_cq_idr, &obj->uobject);
1396 	if (ret)
1397 		goto err_free;
1398 
1399 	memset(&resp, 0, sizeof resp);
1400 	resp.cq_handle = obj->uobject.id;
1401 	resp.cqe       = cq->cqe;
1402 
1403 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
1404 			 &resp, sizeof resp)) {
1405 		ret = -EFAULT;
1406 		goto err_copy;
1407 	}
1408 
1409 	mutex_lock(&file->mutex);
1410 	list_add_tail(&obj->uobject.list, &file->ucontext->cq_list);
1411 	mutex_unlock(&file->mutex);
1412 
1413 	obj->uobject.live = 1;
1414 
1415 	up_write(&obj->uobject.mutex);
1416 
1417 	return in_len;
1418 
1419 err_copy:
1420 	idr_remove_uobj(&ib_uverbs_cq_idr, &obj->uobject);
1421 
1422 err_free:
1423 	ib_destroy_cq(cq);
1424 
1425 err_file:
1426 	if (ev_file)
1427 		ib_uverbs_release_ucq(file, ev_file, obj);
1428 
1429 err:
1430 	put_uobj_write(&obj->uobject);
1431 	return ret;
1432 }
1433 
1434 ssize_t ib_uverbs_resize_cq(struct ib_uverbs_file *file,
1435 			    const char __user *buf, int in_len,
1436 			    int out_len)
1437 {
1438 	struct ib_uverbs_resize_cq	cmd;
1439 	struct ib_uverbs_resize_cq_resp	resp;
1440 	struct ib_udata                 udata;
1441 	struct ib_cq			*cq;
1442 	int				ret = -EINVAL;
1443 
1444 	if (copy_from_user(&cmd, buf, sizeof cmd))
1445 		return -EFAULT;
1446 
1447 	INIT_UDATA(&udata, buf + sizeof cmd,
1448 		   (unsigned long) cmd.response + sizeof resp,
1449 		   in_len - sizeof cmd, out_len - sizeof resp);
1450 
1451 	cq = idr_read_cq(cmd.cq_handle, file->ucontext, 0);
1452 	if (!cq)
1453 		return -EINVAL;
1454 
1455 	ret = cq->device->resize_cq(cq, cmd.cqe, &udata);
1456 	if (ret)
1457 		goto out;
1458 
1459 	resp.cqe = cq->cqe;
1460 
1461 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
1462 			 &resp, sizeof resp.cqe))
1463 		ret = -EFAULT;
1464 
1465 out:
1466 	put_cq_read(cq);
1467 
1468 	return ret ? ret : in_len;
1469 }
1470 
1471 static int copy_wc_to_user(void __user *dest, struct ib_wc *wc)
1472 {
1473 	struct ib_uverbs_wc tmp;
1474 
1475 	tmp.wr_id		= wc->wr_id;
1476 	tmp.status		= wc->status;
1477 	tmp.opcode		= wc->opcode;
1478 	tmp.vendor_err		= wc->vendor_err;
1479 	tmp.byte_len		= wc->byte_len;
1480 	tmp.ex.imm_data		= (__u32 __force) wc->ex.imm_data;
1481 	tmp.qp_num		= wc->qp->qp_num;
1482 	tmp.src_qp		= wc->src_qp;
1483 	tmp.wc_flags		= wc->wc_flags;
1484 	tmp.pkey_index		= wc->pkey_index;
1485 	tmp.slid		= wc->slid;
1486 	tmp.sl			= wc->sl;
1487 	tmp.dlid_path_bits	= wc->dlid_path_bits;
1488 	tmp.port_num		= wc->port_num;
1489 	tmp.reserved		= 0;
1490 
1491 	if (copy_to_user(dest, &tmp, sizeof tmp))
1492 		return -EFAULT;
1493 
1494 	return 0;
1495 }
1496 
1497 ssize_t ib_uverbs_poll_cq(struct ib_uverbs_file *file,
1498 			  const char __user *buf, int in_len,
1499 			  int out_len)
1500 {
1501 	struct ib_uverbs_poll_cq       cmd;
1502 	struct ib_uverbs_poll_cq_resp  resp;
1503 	u8 __user                     *header_ptr;
1504 	u8 __user                     *data_ptr;
1505 	struct ib_cq                  *cq;
1506 	struct ib_wc                   wc;
1507 	int                            ret;
1508 
1509 	if (copy_from_user(&cmd, buf, sizeof cmd))
1510 		return -EFAULT;
1511 
1512 	cq = idr_read_cq(cmd.cq_handle, file->ucontext, 0);
1513 	if (!cq)
1514 		return -EINVAL;
1515 
1516 	/* we copy a struct ib_uverbs_poll_cq_resp to user space */
1517 	header_ptr = (void __user *)(unsigned long) cmd.response;
1518 	data_ptr = header_ptr + sizeof resp;
1519 
1520 	memset(&resp, 0, sizeof resp);
1521 	while (resp.count < cmd.ne) {
1522 		ret = ib_poll_cq(cq, 1, &wc);
1523 		if (ret < 0)
1524 			goto out_put;
1525 		if (!ret)
1526 			break;
1527 
1528 		ret = copy_wc_to_user(data_ptr, &wc);
1529 		if (ret)
1530 			goto out_put;
1531 
1532 		data_ptr += sizeof(struct ib_uverbs_wc);
1533 		++resp.count;
1534 	}
1535 
1536 	if (copy_to_user(header_ptr, &resp, sizeof resp)) {
1537 		ret = -EFAULT;
1538 		goto out_put;
1539 	}
1540 
1541 	ret = in_len;
1542 
1543 out_put:
1544 	put_cq_read(cq);
1545 	return ret;
1546 }
1547 
1548 ssize_t ib_uverbs_req_notify_cq(struct ib_uverbs_file *file,
1549 				const char __user *buf, int in_len,
1550 				int out_len)
1551 {
1552 	struct ib_uverbs_req_notify_cq cmd;
1553 	struct ib_cq                  *cq;
1554 
1555 	if (copy_from_user(&cmd, buf, sizeof cmd))
1556 		return -EFAULT;
1557 
1558 	cq = idr_read_cq(cmd.cq_handle, file->ucontext, 0);
1559 	if (!cq)
1560 		return -EINVAL;
1561 
1562 	ib_req_notify_cq(cq, cmd.solicited_only ?
1563 			 IB_CQ_SOLICITED : IB_CQ_NEXT_COMP);
1564 
1565 	put_cq_read(cq);
1566 
1567 	return in_len;
1568 }
1569 
1570 ssize_t ib_uverbs_destroy_cq(struct ib_uverbs_file *file,
1571 			     const char __user *buf, int in_len,
1572 			     int out_len)
1573 {
1574 	struct ib_uverbs_destroy_cq      cmd;
1575 	struct ib_uverbs_destroy_cq_resp resp;
1576 	struct ib_uobject		*uobj;
1577 	struct ib_cq               	*cq;
1578 	struct ib_ucq_object        	*obj;
1579 	struct ib_uverbs_event_file	*ev_file;
1580 	int                        	 ret = -EINVAL;
1581 
1582 	if (copy_from_user(&cmd, buf, sizeof cmd))
1583 		return -EFAULT;
1584 
1585 	uobj = idr_write_uobj(&ib_uverbs_cq_idr, cmd.cq_handle, file->ucontext);
1586 	if (!uobj)
1587 		return -EINVAL;
1588 	cq      = uobj->object;
1589 	ev_file = cq->cq_context;
1590 	obj     = container_of(cq->uobject, struct ib_ucq_object, uobject);
1591 
1592 	ret = ib_destroy_cq(cq);
1593 	if (!ret)
1594 		uobj->live = 0;
1595 
1596 	put_uobj_write(uobj);
1597 
1598 	if (ret)
1599 		return ret;
1600 
1601 	idr_remove_uobj(&ib_uverbs_cq_idr, uobj);
1602 
1603 	mutex_lock(&file->mutex);
1604 	list_del(&uobj->list);
1605 	mutex_unlock(&file->mutex);
1606 
1607 	ib_uverbs_release_ucq(file, ev_file, obj);
1608 
1609 	memset(&resp, 0, sizeof resp);
1610 	resp.comp_events_reported  = obj->comp_events_reported;
1611 	resp.async_events_reported = obj->async_events_reported;
1612 
1613 	put_uobj(uobj);
1614 
1615 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
1616 			 &resp, sizeof resp))
1617 		return -EFAULT;
1618 
1619 	return in_len;
1620 }
1621 
1622 ssize_t ib_uverbs_create_qp(struct ib_uverbs_file *file,
1623 			    const char __user *buf, int in_len,
1624 			    int out_len)
1625 {
1626 	struct ib_uverbs_create_qp      cmd;
1627 	struct ib_uverbs_create_qp_resp resp;
1628 	struct ib_udata                 udata;
1629 	struct ib_uqp_object           *obj;
1630 	struct ib_device	       *device;
1631 	struct ib_pd                   *pd = NULL;
1632 	struct ib_xrcd		       *xrcd = NULL;
1633 	struct ib_uobject	       *uninitialized_var(xrcd_uobj);
1634 	struct ib_cq                   *scq = NULL, *rcq = NULL;
1635 	struct ib_srq                  *srq = NULL;
1636 	struct ib_qp                   *qp;
1637 	struct ib_qp_init_attr          attr;
1638 	int ret;
1639 
1640 	if (out_len < sizeof resp)
1641 		return -ENOSPC;
1642 
1643 	if (copy_from_user(&cmd, buf, sizeof cmd))
1644 		return -EFAULT;
1645 
1646 	if (cmd.qp_type == IB_QPT_RAW_PACKET && !capable(CAP_NET_RAW))
1647 		return -EPERM;
1648 
1649 	INIT_UDATA(&udata, buf + sizeof cmd,
1650 		   (unsigned long) cmd.response + sizeof resp,
1651 		   in_len - sizeof cmd, out_len - sizeof resp);
1652 
1653 	obj = kzalloc(sizeof *obj, GFP_KERNEL);
1654 	if (!obj)
1655 		return -ENOMEM;
1656 
1657 	init_uobj(&obj->uevent.uobject, cmd.user_handle, file->ucontext, &qp_lock_class);
1658 	down_write(&obj->uevent.uobject.mutex);
1659 
1660 	if (cmd.qp_type == IB_QPT_XRC_TGT) {
1661 		xrcd = idr_read_xrcd(cmd.pd_handle, file->ucontext, &xrcd_uobj);
1662 		if (!xrcd) {
1663 			ret = -EINVAL;
1664 			goto err_put;
1665 		}
1666 		device = xrcd->device;
1667 	} else {
1668 		if (cmd.qp_type == IB_QPT_XRC_INI) {
1669 			cmd.max_recv_wr = cmd.max_recv_sge = 0;
1670 		} else {
1671 			if (cmd.is_srq) {
1672 				srq = idr_read_srq(cmd.srq_handle, file->ucontext);
1673 				if (!srq || srq->srq_type != IB_SRQT_BASIC) {
1674 					ret = -EINVAL;
1675 					goto err_put;
1676 				}
1677 			}
1678 
1679 			if (cmd.recv_cq_handle != cmd.send_cq_handle) {
1680 				rcq = idr_read_cq(cmd.recv_cq_handle, file->ucontext, 0);
1681 				if (!rcq) {
1682 					ret = -EINVAL;
1683 					goto err_put;
1684 				}
1685 			}
1686 		}
1687 
1688 		scq = idr_read_cq(cmd.send_cq_handle, file->ucontext, !!rcq);
1689 		rcq = rcq ?: scq;
1690 		pd  = idr_read_pd(cmd.pd_handle, file->ucontext);
1691 		if (!pd || !scq) {
1692 			ret = -EINVAL;
1693 			goto err_put;
1694 		}
1695 
1696 		device = pd->device;
1697 	}
1698 
1699 	attr.event_handler = ib_uverbs_qp_event_handler;
1700 	attr.qp_context    = file;
1701 	attr.send_cq       = scq;
1702 	attr.recv_cq       = rcq;
1703 	attr.srq           = srq;
1704 	attr.xrcd	   = xrcd;
1705 	attr.sq_sig_type   = cmd.sq_sig_all ? IB_SIGNAL_ALL_WR : IB_SIGNAL_REQ_WR;
1706 	attr.qp_type       = cmd.qp_type;
1707 	attr.create_flags  = 0;
1708 
1709 	attr.cap.max_send_wr     = cmd.max_send_wr;
1710 	attr.cap.max_recv_wr     = cmd.max_recv_wr;
1711 	attr.cap.max_send_sge    = cmd.max_send_sge;
1712 	attr.cap.max_recv_sge    = cmd.max_recv_sge;
1713 	attr.cap.max_inline_data = cmd.max_inline_data;
1714 
1715 	obj->uevent.events_reported     = 0;
1716 	INIT_LIST_HEAD(&obj->uevent.event_list);
1717 	INIT_LIST_HEAD(&obj->mcast_list);
1718 
1719 	if (cmd.qp_type == IB_QPT_XRC_TGT)
1720 		qp = ib_create_qp(pd, &attr);
1721 	else
1722 		qp = device->create_qp(pd, &attr, &udata);
1723 
1724 	if (IS_ERR(qp)) {
1725 		ret = PTR_ERR(qp);
1726 		goto err_put;
1727 	}
1728 
1729 	if (cmd.qp_type != IB_QPT_XRC_TGT) {
1730 		qp->real_qp	  = qp;
1731 		qp->device	  = device;
1732 		qp->pd		  = pd;
1733 		qp->send_cq	  = attr.send_cq;
1734 		qp->recv_cq	  = attr.recv_cq;
1735 		qp->srq		  = attr.srq;
1736 		qp->event_handler = attr.event_handler;
1737 		qp->qp_context	  = attr.qp_context;
1738 		qp->qp_type	  = attr.qp_type;
1739 		atomic_set(&qp->usecnt, 0);
1740 		atomic_inc(&pd->usecnt);
1741 		atomic_inc(&attr.send_cq->usecnt);
1742 		if (attr.recv_cq)
1743 			atomic_inc(&attr.recv_cq->usecnt);
1744 		if (attr.srq)
1745 			atomic_inc(&attr.srq->usecnt);
1746 	}
1747 	qp->uobject = &obj->uevent.uobject;
1748 
1749 	obj->uevent.uobject.object = qp;
1750 	ret = idr_add_uobj(&ib_uverbs_qp_idr, &obj->uevent.uobject);
1751 	if (ret)
1752 		goto err_destroy;
1753 
1754 	memset(&resp, 0, sizeof resp);
1755 	resp.qpn             = qp->qp_num;
1756 	resp.qp_handle       = obj->uevent.uobject.id;
1757 	resp.max_recv_sge    = attr.cap.max_recv_sge;
1758 	resp.max_send_sge    = attr.cap.max_send_sge;
1759 	resp.max_recv_wr     = attr.cap.max_recv_wr;
1760 	resp.max_send_wr     = attr.cap.max_send_wr;
1761 	resp.max_inline_data = attr.cap.max_inline_data;
1762 
1763 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
1764 			 &resp, sizeof resp)) {
1765 		ret = -EFAULT;
1766 		goto err_copy;
1767 	}
1768 
1769 	if (xrcd) {
1770 		obj->uxrcd = container_of(xrcd_uobj, struct ib_uxrcd_object,
1771 					  uobject);
1772 		atomic_inc(&obj->uxrcd->refcnt);
1773 		put_xrcd_read(xrcd_uobj);
1774 	}
1775 
1776 	if (pd)
1777 		put_pd_read(pd);
1778 	if (scq)
1779 		put_cq_read(scq);
1780 	if (rcq && rcq != scq)
1781 		put_cq_read(rcq);
1782 	if (srq)
1783 		put_srq_read(srq);
1784 
1785 	mutex_lock(&file->mutex);
1786 	list_add_tail(&obj->uevent.uobject.list, &file->ucontext->qp_list);
1787 	mutex_unlock(&file->mutex);
1788 
1789 	obj->uevent.uobject.live = 1;
1790 
1791 	up_write(&obj->uevent.uobject.mutex);
1792 
1793 	return in_len;
1794 
1795 err_copy:
1796 	idr_remove_uobj(&ib_uverbs_qp_idr, &obj->uevent.uobject);
1797 
1798 err_destroy:
1799 	ib_destroy_qp(qp);
1800 
1801 err_put:
1802 	if (xrcd)
1803 		put_xrcd_read(xrcd_uobj);
1804 	if (pd)
1805 		put_pd_read(pd);
1806 	if (scq)
1807 		put_cq_read(scq);
1808 	if (rcq && rcq != scq)
1809 		put_cq_read(rcq);
1810 	if (srq)
1811 		put_srq_read(srq);
1812 
1813 	put_uobj_write(&obj->uevent.uobject);
1814 	return ret;
1815 }
1816 
1817 ssize_t ib_uverbs_open_qp(struct ib_uverbs_file *file,
1818 			  const char __user *buf, int in_len, int out_len)
1819 {
1820 	struct ib_uverbs_open_qp        cmd;
1821 	struct ib_uverbs_create_qp_resp resp;
1822 	struct ib_udata                 udata;
1823 	struct ib_uqp_object           *obj;
1824 	struct ib_xrcd		       *xrcd;
1825 	struct ib_uobject	       *uninitialized_var(xrcd_uobj);
1826 	struct ib_qp                   *qp;
1827 	struct ib_qp_open_attr          attr;
1828 	int ret;
1829 
1830 	if (out_len < sizeof resp)
1831 		return -ENOSPC;
1832 
1833 	if (copy_from_user(&cmd, buf, sizeof cmd))
1834 		return -EFAULT;
1835 
1836 	INIT_UDATA(&udata, buf + sizeof cmd,
1837 		   (unsigned long) cmd.response + sizeof resp,
1838 		   in_len - sizeof cmd, out_len - sizeof resp);
1839 
1840 	obj = kmalloc(sizeof *obj, GFP_KERNEL);
1841 	if (!obj)
1842 		return -ENOMEM;
1843 
1844 	init_uobj(&obj->uevent.uobject, cmd.user_handle, file->ucontext, &qp_lock_class);
1845 	down_write(&obj->uevent.uobject.mutex);
1846 
1847 	xrcd = idr_read_xrcd(cmd.pd_handle, file->ucontext, &xrcd_uobj);
1848 	if (!xrcd) {
1849 		ret = -EINVAL;
1850 		goto err_put;
1851 	}
1852 
1853 	attr.event_handler = ib_uverbs_qp_event_handler;
1854 	attr.qp_context    = file;
1855 	attr.qp_num        = cmd.qpn;
1856 	attr.qp_type       = cmd.qp_type;
1857 
1858 	obj->uevent.events_reported = 0;
1859 	INIT_LIST_HEAD(&obj->uevent.event_list);
1860 	INIT_LIST_HEAD(&obj->mcast_list);
1861 
1862 	qp = ib_open_qp(xrcd, &attr);
1863 	if (IS_ERR(qp)) {
1864 		ret = PTR_ERR(qp);
1865 		goto err_put;
1866 	}
1867 
1868 	qp->uobject = &obj->uevent.uobject;
1869 
1870 	obj->uevent.uobject.object = qp;
1871 	ret = idr_add_uobj(&ib_uverbs_qp_idr, &obj->uevent.uobject);
1872 	if (ret)
1873 		goto err_destroy;
1874 
1875 	memset(&resp, 0, sizeof resp);
1876 	resp.qpn       = qp->qp_num;
1877 	resp.qp_handle = obj->uevent.uobject.id;
1878 
1879 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
1880 			 &resp, sizeof resp)) {
1881 		ret = -EFAULT;
1882 		goto err_remove;
1883 	}
1884 
1885 	obj->uxrcd = container_of(xrcd_uobj, struct ib_uxrcd_object, uobject);
1886 	atomic_inc(&obj->uxrcd->refcnt);
1887 	put_xrcd_read(xrcd_uobj);
1888 
1889 	mutex_lock(&file->mutex);
1890 	list_add_tail(&obj->uevent.uobject.list, &file->ucontext->qp_list);
1891 	mutex_unlock(&file->mutex);
1892 
1893 	obj->uevent.uobject.live = 1;
1894 
1895 	up_write(&obj->uevent.uobject.mutex);
1896 
1897 	return in_len;
1898 
1899 err_remove:
1900 	idr_remove_uobj(&ib_uverbs_qp_idr, &obj->uevent.uobject);
1901 
1902 err_destroy:
1903 	ib_destroy_qp(qp);
1904 
1905 err_put:
1906 	put_xrcd_read(xrcd_uobj);
1907 	put_uobj_write(&obj->uevent.uobject);
1908 	return ret;
1909 }
1910 
1911 ssize_t ib_uverbs_query_qp(struct ib_uverbs_file *file,
1912 			   const char __user *buf, int in_len,
1913 			   int out_len)
1914 {
1915 	struct ib_uverbs_query_qp      cmd;
1916 	struct ib_uverbs_query_qp_resp resp;
1917 	struct ib_qp                   *qp;
1918 	struct ib_qp_attr              *attr;
1919 	struct ib_qp_init_attr         *init_attr;
1920 	int                            ret;
1921 
1922 	if (copy_from_user(&cmd, buf, sizeof cmd))
1923 		return -EFAULT;
1924 
1925 	attr      = kmalloc(sizeof *attr, GFP_KERNEL);
1926 	init_attr = kmalloc(sizeof *init_attr, GFP_KERNEL);
1927 	if (!attr || !init_attr) {
1928 		ret = -ENOMEM;
1929 		goto out;
1930 	}
1931 
1932 	qp = idr_read_qp(cmd.qp_handle, file->ucontext);
1933 	if (!qp) {
1934 		ret = -EINVAL;
1935 		goto out;
1936 	}
1937 
1938 	ret = ib_query_qp(qp, attr, cmd.attr_mask, init_attr);
1939 
1940 	put_qp_read(qp);
1941 
1942 	if (ret)
1943 		goto out;
1944 
1945 	memset(&resp, 0, sizeof resp);
1946 
1947 	resp.qp_state               = attr->qp_state;
1948 	resp.cur_qp_state           = attr->cur_qp_state;
1949 	resp.path_mtu               = attr->path_mtu;
1950 	resp.path_mig_state         = attr->path_mig_state;
1951 	resp.qkey                   = attr->qkey;
1952 	resp.rq_psn                 = attr->rq_psn;
1953 	resp.sq_psn                 = attr->sq_psn;
1954 	resp.dest_qp_num            = attr->dest_qp_num;
1955 	resp.qp_access_flags        = attr->qp_access_flags;
1956 	resp.pkey_index             = attr->pkey_index;
1957 	resp.alt_pkey_index         = attr->alt_pkey_index;
1958 	resp.sq_draining            = attr->sq_draining;
1959 	resp.max_rd_atomic          = attr->max_rd_atomic;
1960 	resp.max_dest_rd_atomic     = attr->max_dest_rd_atomic;
1961 	resp.min_rnr_timer          = attr->min_rnr_timer;
1962 	resp.port_num               = attr->port_num;
1963 	resp.timeout                = attr->timeout;
1964 	resp.retry_cnt              = attr->retry_cnt;
1965 	resp.rnr_retry              = attr->rnr_retry;
1966 	resp.alt_port_num           = attr->alt_port_num;
1967 	resp.alt_timeout            = attr->alt_timeout;
1968 
1969 	memcpy(resp.dest.dgid, attr->ah_attr.grh.dgid.raw, 16);
1970 	resp.dest.flow_label        = attr->ah_attr.grh.flow_label;
1971 	resp.dest.sgid_index        = attr->ah_attr.grh.sgid_index;
1972 	resp.dest.hop_limit         = attr->ah_attr.grh.hop_limit;
1973 	resp.dest.traffic_class     = attr->ah_attr.grh.traffic_class;
1974 	resp.dest.dlid              = attr->ah_attr.dlid;
1975 	resp.dest.sl                = attr->ah_attr.sl;
1976 	resp.dest.src_path_bits     = attr->ah_attr.src_path_bits;
1977 	resp.dest.static_rate       = attr->ah_attr.static_rate;
1978 	resp.dest.is_global         = !!(attr->ah_attr.ah_flags & IB_AH_GRH);
1979 	resp.dest.port_num          = attr->ah_attr.port_num;
1980 
1981 	memcpy(resp.alt_dest.dgid, attr->alt_ah_attr.grh.dgid.raw, 16);
1982 	resp.alt_dest.flow_label    = attr->alt_ah_attr.grh.flow_label;
1983 	resp.alt_dest.sgid_index    = attr->alt_ah_attr.grh.sgid_index;
1984 	resp.alt_dest.hop_limit     = attr->alt_ah_attr.grh.hop_limit;
1985 	resp.alt_dest.traffic_class = attr->alt_ah_attr.grh.traffic_class;
1986 	resp.alt_dest.dlid          = attr->alt_ah_attr.dlid;
1987 	resp.alt_dest.sl            = attr->alt_ah_attr.sl;
1988 	resp.alt_dest.src_path_bits = attr->alt_ah_attr.src_path_bits;
1989 	resp.alt_dest.static_rate   = attr->alt_ah_attr.static_rate;
1990 	resp.alt_dest.is_global     = !!(attr->alt_ah_attr.ah_flags & IB_AH_GRH);
1991 	resp.alt_dest.port_num      = attr->alt_ah_attr.port_num;
1992 
1993 	resp.max_send_wr            = init_attr->cap.max_send_wr;
1994 	resp.max_recv_wr            = init_attr->cap.max_recv_wr;
1995 	resp.max_send_sge           = init_attr->cap.max_send_sge;
1996 	resp.max_recv_sge           = init_attr->cap.max_recv_sge;
1997 	resp.max_inline_data        = init_attr->cap.max_inline_data;
1998 	resp.sq_sig_all             = init_attr->sq_sig_type == IB_SIGNAL_ALL_WR;
1999 
2000 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
2001 			 &resp, sizeof resp))
2002 		ret = -EFAULT;
2003 
2004 out:
2005 	kfree(attr);
2006 	kfree(init_attr);
2007 
2008 	return ret ? ret : in_len;
2009 }
2010 
2011 /* Remove ignored fields set in the attribute mask */
2012 static int modify_qp_mask(enum ib_qp_type qp_type, int mask)
2013 {
2014 	switch (qp_type) {
2015 	case IB_QPT_XRC_INI:
2016 		return mask & ~(IB_QP_MAX_DEST_RD_ATOMIC | IB_QP_MIN_RNR_TIMER);
2017 	case IB_QPT_XRC_TGT:
2018 		return mask & ~(IB_QP_MAX_QP_RD_ATOMIC | IB_QP_RETRY_CNT |
2019 				IB_QP_RNR_RETRY);
2020 	default:
2021 		return mask;
2022 	}
2023 }
2024 
2025 ssize_t ib_uverbs_modify_qp(struct ib_uverbs_file *file,
2026 			    const char __user *buf, int in_len,
2027 			    int out_len)
2028 {
2029 	struct ib_uverbs_modify_qp cmd;
2030 	struct ib_udata            udata;
2031 	struct ib_qp              *qp;
2032 	struct ib_qp_attr         *attr;
2033 	int                        ret;
2034 
2035 	if (copy_from_user(&cmd, buf, sizeof cmd))
2036 		return -EFAULT;
2037 
2038 	INIT_UDATA(&udata, buf + sizeof cmd, NULL, in_len - sizeof cmd,
2039 		   out_len);
2040 
2041 	attr = kmalloc(sizeof *attr, GFP_KERNEL);
2042 	if (!attr)
2043 		return -ENOMEM;
2044 
2045 	qp = idr_read_qp(cmd.qp_handle, file->ucontext);
2046 	if (!qp) {
2047 		ret = -EINVAL;
2048 		goto out;
2049 	}
2050 
2051 	attr->qp_state 		  = cmd.qp_state;
2052 	attr->cur_qp_state 	  = cmd.cur_qp_state;
2053 	attr->path_mtu 		  = cmd.path_mtu;
2054 	attr->path_mig_state 	  = cmd.path_mig_state;
2055 	attr->qkey 		  = cmd.qkey;
2056 	attr->rq_psn 		  = cmd.rq_psn;
2057 	attr->sq_psn 		  = cmd.sq_psn;
2058 	attr->dest_qp_num 	  = cmd.dest_qp_num;
2059 	attr->qp_access_flags 	  = cmd.qp_access_flags;
2060 	attr->pkey_index 	  = cmd.pkey_index;
2061 	attr->alt_pkey_index 	  = cmd.alt_pkey_index;
2062 	attr->en_sqd_async_notify = cmd.en_sqd_async_notify;
2063 	attr->max_rd_atomic 	  = cmd.max_rd_atomic;
2064 	attr->max_dest_rd_atomic  = cmd.max_dest_rd_atomic;
2065 	attr->min_rnr_timer 	  = cmd.min_rnr_timer;
2066 	attr->port_num 		  = cmd.port_num;
2067 	attr->timeout 		  = cmd.timeout;
2068 	attr->retry_cnt 	  = cmd.retry_cnt;
2069 	attr->rnr_retry 	  = cmd.rnr_retry;
2070 	attr->alt_port_num 	  = cmd.alt_port_num;
2071 	attr->alt_timeout 	  = cmd.alt_timeout;
2072 
2073 	memcpy(attr->ah_attr.grh.dgid.raw, cmd.dest.dgid, 16);
2074 	attr->ah_attr.grh.flow_label        = cmd.dest.flow_label;
2075 	attr->ah_attr.grh.sgid_index        = cmd.dest.sgid_index;
2076 	attr->ah_attr.grh.hop_limit         = cmd.dest.hop_limit;
2077 	attr->ah_attr.grh.traffic_class     = cmd.dest.traffic_class;
2078 	attr->ah_attr.dlid 	    	    = cmd.dest.dlid;
2079 	attr->ah_attr.sl   	    	    = cmd.dest.sl;
2080 	attr->ah_attr.src_path_bits 	    = cmd.dest.src_path_bits;
2081 	attr->ah_attr.static_rate   	    = cmd.dest.static_rate;
2082 	attr->ah_attr.ah_flags 	    	    = cmd.dest.is_global ? IB_AH_GRH : 0;
2083 	attr->ah_attr.port_num 	    	    = cmd.dest.port_num;
2084 
2085 	memcpy(attr->alt_ah_attr.grh.dgid.raw, cmd.alt_dest.dgid, 16);
2086 	attr->alt_ah_attr.grh.flow_label    = cmd.alt_dest.flow_label;
2087 	attr->alt_ah_attr.grh.sgid_index    = cmd.alt_dest.sgid_index;
2088 	attr->alt_ah_attr.grh.hop_limit     = cmd.alt_dest.hop_limit;
2089 	attr->alt_ah_attr.grh.traffic_class = cmd.alt_dest.traffic_class;
2090 	attr->alt_ah_attr.dlid 	    	    = cmd.alt_dest.dlid;
2091 	attr->alt_ah_attr.sl   	    	    = cmd.alt_dest.sl;
2092 	attr->alt_ah_attr.src_path_bits     = cmd.alt_dest.src_path_bits;
2093 	attr->alt_ah_attr.static_rate       = cmd.alt_dest.static_rate;
2094 	attr->alt_ah_attr.ah_flags 	    = cmd.alt_dest.is_global ? IB_AH_GRH : 0;
2095 	attr->alt_ah_attr.port_num 	    = cmd.alt_dest.port_num;
2096 
2097 	if (qp->real_qp == qp) {
2098 		ret = ib_resolve_eth_l2_attrs(qp, attr, &cmd.attr_mask);
2099 		if (ret)
2100 			goto release_qp;
2101 		ret = qp->device->modify_qp(qp, attr,
2102 			modify_qp_mask(qp->qp_type, cmd.attr_mask), &udata);
2103 	} else {
2104 		ret = ib_modify_qp(qp, attr, modify_qp_mask(qp->qp_type, cmd.attr_mask));
2105 	}
2106 
2107 	if (ret)
2108 		goto release_qp;
2109 
2110 	ret = in_len;
2111 
2112 release_qp:
2113 	put_qp_read(qp);
2114 
2115 out:
2116 	kfree(attr);
2117 
2118 	return ret;
2119 }
2120 
2121 ssize_t ib_uverbs_destroy_qp(struct ib_uverbs_file *file,
2122 			     const char __user *buf, int in_len,
2123 			     int out_len)
2124 {
2125 	struct ib_uverbs_destroy_qp      cmd;
2126 	struct ib_uverbs_destroy_qp_resp resp;
2127 	struct ib_uobject		*uobj;
2128 	struct ib_qp               	*qp;
2129 	struct ib_uqp_object        	*obj;
2130 	int                        	 ret = -EINVAL;
2131 
2132 	if (copy_from_user(&cmd, buf, sizeof cmd))
2133 		return -EFAULT;
2134 
2135 	memset(&resp, 0, sizeof resp);
2136 
2137 	uobj = idr_write_uobj(&ib_uverbs_qp_idr, cmd.qp_handle, file->ucontext);
2138 	if (!uobj)
2139 		return -EINVAL;
2140 	qp  = uobj->object;
2141 	obj = container_of(uobj, struct ib_uqp_object, uevent.uobject);
2142 
2143 	if (!list_empty(&obj->mcast_list)) {
2144 		put_uobj_write(uobj);
2145 		return -EBUSY;
2146 	}
2147 
2148 	ret = ib_destroy_qp(qp);
2149 	if (!ret)
2150 		uobj->live = 0;
2151 
2152 	put_uobj_write(uobj);
2153 
2154 	if (ret)
2155 		return ret;
2156 
2157 	if (obj->uxrcd)
2158 		atomic_dec(&obj->uxrcd->refcnt);
2159 
2160 	idr_remove_uobj(&ib_uverbs_qp_idr, uobj);
2161 
2162 	mutex_lock(&file->mutex);
2163 	list_del(&uobj->list);
2164 	mutex_unlock(&file->mutex);
2165 
2166 	ib_uverbs_release_uevent(file, &obj->uevent);
2167 
2168 	resp.events_reported = obj->uevent.events_reported;
2169 
2170 	put_uobj(uobj);
2171 
2172 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
2173 			 &resp, sizeof resp))
2174 		return -EFAULT;
2175 
2176 	return in_len;
2177 }
2178 
2179 ssize_t ib_uverbs_post_send(struct ib_uverbs_file *file,
2180 			    const char __user *buf, int in_len,
2181 			    int out_len)
2182 {
2183 	struct ib_uverbs_post_send      cmd;
2184 	struct ib_uverbs_post_send_resp resp;
2185 	struct ib_uverbs_send_wr       *user_wr;
2186 	struct ib_send_wr              *wr = NULL, *last, *next, *bad_wr;
2187 	struct ib_qp                   *qp;
2188 	int                             i, sg_ind;
2189 	int				is_ud;
2190 	ssize_t                         ret = -EINVAL;
2191 
2192 	if (copy_from_user(&cmd, buf, sizeof cmd))
2193 		return -EFAULT;
2194 
2195 	if (in_len < sizeof cmd + cmd.wqe_size * cmd.wr_count +
2196 	    cmd.sge_count * sizeof (struct ib_uverbs_sge))
2197 		return -EINVAL;
2198 
2199 	if (cmd.wqe_size < sizeof (struct ib_uverbs_send_wr))
2200 		return -EINVAL;
2201 
2202 	user_wr = kmalloc(cmd.wqe_size, GFP_KERNEL);
2203 	if (!user_wr)
2204 		return -ENOMEM;
2205 
2206 	qp = idr_read_qp(cmd.qp_handle, file->ucontext);
2207 	if (!qp)
2208 		goto out;
2209 
2210 	is_ud = qp->qp_type == IB_QPT_UD;
2211 	sg_ind = 0;
2212 	last = NULL;
2213 	for (i = 0; i < cmd.wr_count; ++i) {
2214 		if (copy_from_user(user_wr,
2215 				   buf + sizeof cmd + i * cmd.wqe_size,
2216 				   cmd.wqe_size)) {
2217 			ret = -EFAULT;
2218 			goto out_put;
2219 		}
2220 
2221 		if (user_wr->num_sge + sg_ind > cmd.sge_count) {
2222 			ret = -EINVAL;
2223 			goto out_put;
2224 		}
2225 
2226 		next = kmalloc(ALIGN(sizeof *next, sizeof (struct ib_sge)) +
2227 			       user_wr->num_sge * sizeof (struct ib_sge),
2228 			       GFP_KERNEL);
2229 		if (!next) {
2230 			ret = -ENOMEM;
2231 			goto out_put;
2232 		}
2233 
2234 		if (!last)
2235 			wr = next;
2236 		else
2237 			last->next = next;
2238 		last = next;
2239 
2240 		next->next       = NULL;
2241 		next->wr_id      = user_wr->wr_id;
2242 		next->num_sge    = user_wr->num_sge;
2243 		next->opcode     = user_wr->opcode;
2244 		next->send_flags = user_wr->send_flags;
2245 
2246 		if (is_ud) {
2247 			next->wr.ud.ah = idr_read_ah(user_wr->wr.ud.ah,
2248 						     file->ucontext);
2249 			if (!next->wr.ud.ah) {
2250 				ret = -EINVAL;
2251 				goto out_put;
2252 			}
2253 			next->wr.ud.remote_qpn  = user_wr->wr.ud.remote_qpn;
2254 			next->wr.ud.remote_qkey = user_wr->wr.ud.remote_qkey;
2255 			if (next->opcode == IB_WR_SEND_WITH_IMM)
2256 				next->ex.imm_data =
2257 					(__be32 __force) user_wr->ex.imm_data;
2258 		} else {
2259 			switch (next->opcode) {
2260 			case IB_WR_RDMA_WRITE_WITH_IMM:
2261 				next->ex.imm_data =
2262 					(__be32 __force) user_wr->ex.imm_data;
2263 			case IB_WR_RDMA_WRITE:
2264 			case IB_WR_RDMA_READ:
2265 				next->wr.rdma.remote_addr =
2266 					user_wr->wr.rdma.remote_addr;
2267 				next->wr.rdma.rkey        =
2268 					user_wr->wr.rdma.rkey;
2269 				break;
2270 			case IB_WR_SEND_WITH_IMM:
2271 				next->ex.imm_data =
2272 					(__be32 __force) user_wr->ex.imm_data;
2273 				break;
2274 			case IB_WR_SEND_WITH_INV:
2275 				next->ex.invalidate_rkey =
2276 					user_wr->ex.invalidate_rkey;
2277 				break;
2278 			case IB_WR_ATOMIC_CMP_AND_SWP:
2279 			case IB_WR_ATOMIC_FETCH_AND_ADD:
2280 				next->wr.atomic.remote_addr =
2281 					user_wr->wr.atomic.remote_addr;
2282 				next->wr.atomic.compare_add =
2283 					user_wr->wr.atomic.compare_add;
2284 				next->wr.atomic.swap = user_wr->wr.atomic.swap;
2285 				next->wr.atomic.rkey = user_wr->wr.atomic.rkey;
2286 				break;
2287 			default:
2288 				break;
2289 			}
2290 		}
2291 
2292 		if (next->num_sge) {
2293 			next->sg_list = (void *) next +
2294 				ALIGN(sizeof *next, sizeof (struct ib_sge));
2295 			if (copy_from_user(next->sg_list,
2296 					   buf + sizeof cmd +
2297 					   cmd.wr_count * cmd.wqe_size +
2298 					   sg_ind * sizeof (struct ib_sge),
2299 					   next->num_sge * sizeof (struct ib_sge))) {
2300 				ret = -EFAULT;
2301 				goto out_put;
2302 			}
2303 			sg_ind += next->num_sge;
2304 		} else
2305 			next->sg_list = NULL;
2306 	}
2307 
2308 	resp.bad_wr = 0;
2309 	ret = qp->device->post_send(qp->real_qp, wr, &bad_wr);
2310 	if (ret)
2311 		for (next = wr; next; next = next->next) {
2312 			++resp.bad_wr;
2313 			if (next == bad_wr)
2314 				break;
2315 		}
2316 
2317 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
2318 			 &resp, sizeof resp))
2319 		ret = -EFAULT;
2320 
2321 out_put:
2322 	put_qp_read(qp);
2323 
2324 	while (wr) {
2325 		if (is_ud && wr->wr.ud.ah)
2326 			put_ah_read(wr->wr.ud.ah);
2327 		next = wr->next;
2328 		kfree(wr);
2329 		wr = next;
2330 	}
2331 
2332 out:
2333 	kfree(user_wr);
2334 
2335 	return ret ? ret : in_len;
2336 }
2337 
2338 static struct ib_recv_wr *ib_uverbs_unmarshall_recv(const char __user *buf,
2339 						    int in_len,
2340 						    u32 wr_count,
2341 						    u32 sge_count,
2342 						    u32 wqe_size)
2343 {
2344 	struct ib_uverbs_recv_wr *user_wr;
2345 	struct ib_recv_wr        *wr = NULL, *last, *next;
2346 	int                       sg_ind;
2347 	int                       i;
2348 	int                       ret;
2349 
2350 	if (in_len < wqe_size * wr_count +
2351 	    sge_count * sizeof (struct ib_uverbs_sge))
2352 		return ERR_PTR(-EINVAL);
2353 
2354 	if (wqe_size < sizeof (struct ib_uverbs_recv_wr))
2355 		return ERR_PTR(-EINVAL);
2356 
2357 	user_wr = kmalloc(wqe_size, GFP_KERNEL);
2358 	if (!user_wr)
2359 		return ERR_PTR(-ENOMEM);
2360 
2361 	sg_ind = 0;
2362 	last = NULL;
2363 	for (i = 0; i < wr_count; ++i) {
2364 		if (copy_from_user(user_wr, buf + i * wqe_size,
2365 				   wqe_size)) {
2366 			ret = -EFAULT;
2367 			goto err;
2368 		}
2369 
2370 		if (user_wr->num_sge + sg_ind > sge_count) {
2371 			ret = -EINVAL;
2372 			goto err;
2373 		}
2374 
2375 		next = kmalloc(ALIGN(sizeof *next, sizeof (struct ib_sge)) +
2376 			       user_wr->num_sge * sizeof (struct ib_sge),
2377 			       GFP_KERNEL);
2378 		if (!next) {
2379 			ret = -ENOMEM;
2380 			goto err;
2381 		}
2382 
2383 		if (!last)
2384 			wr = next;
2385 		else
2386 			last->next = next;
2387 		last = next;
2388 
2389 		next->next       = NULL;
2390 		next->wr_id      = user_wr->wr_id;
2391 		next->num_sge    = user_wr->num_sge;
2392 
2393 		if (next->num_sge) {
2394 			next->sg_list = (void *) next +
2395 				ALIGN(sizeof *next, sizeof (struct ib_sge));
2396 			if (copy_from_user(next->sg_list,
2397 					   buf + wr_count * wqe_size +
2398 					   sg_ind * sizeof (struct ib_sge),
2399 					   next->num_sge * sizeof (struct ib_sge))) {
2400 				ret = -EFAULT;
2401 				goto err;
2402 			}
2403 			sg_ind += next->num_sge;
2404 		} else
2405 			next->sg_list = NULL;
2406 	}
2407 
2408 	kfree(user_wr);
2409 	return wr;
2410 
2411 err:
2412 	kfree(user_wr);
2413 
2414 	while (wr) {
2415 		next = wr->next;
2416 		kfree(wr);
2417 		wr = next;
2418 	}
2419 
2420 	return ERR_PTR(ret);
2421 }
2422 
2423 ssize_t ib_uverbs_post_recv(struct ib_uverbs_file *file,
2424 			    const char __user *buf, int in_len,
2425 			    int out_len)
2426 {
2427 	struct ib_uverbs_post_recv      cmd;
2428 	struct ib_uverbs_post_recv_resp resp;
2429 	struct ib_recv_wr              *wr, *next, *bad_wr;
2430 	struct ib_qp                   *qp;
2431 	ssize_t                         ret = -EINVAL;
2432 
2433 	if (copy_from_user(&cmd, buf, sizeof cmd))
2434 		return -EFAULT;
2435 
2436 	wr = ib_uverbs_unmarshall_recv(buf + sizeof cmd,
2437 				       in_len - sizeof cmd, cmd.wr_count,
2438 				       cmd.sge_count, cmd.wqe_size);
2439 	if (IS_ERR(wr))
2440 		return PTR_ERR(wr);
2441 
2442 	qp = idr_read_qp(cmd.qp_handle, file->ucontext);
2443 	if (!qp)
2444 		goto out;
2445 
2446 	resp.bad_wr = 0;
2447 	ret = qp->device->post_recv(qp->real_qp, wr, &bad_wr);
2448 
2449 	put_qp_read(qp);
2450 
2451 	if (ret)
2452 		for (next = wr; next; next = next->next) {
2453 			++resp.bad_wr;
2454 			if (next == bad_wr)
2455 				break;
2456 		}
2457 
2458 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
2459 			 &resp, sizeof resp))
2460 		ret = -EFAULT;
2461 
2462 out:
2463 	while (wr) {
2464 		next = wr->next;
2465 		kfree(wr);
2466 		wr = next;
2467 	}
2468 
2469 	return ret ? ret : in_len;
2470 }
2471 
2472 ssize_t ib_uverbs_post_srq_recv(struct ib_uverbs_file *file,
2473 				const char __user *buf, int in_len,
2474 				int out_len)
2475 {
2476 	struct ib_uverbs_post_srq_recv      cmd;
2477 	struct ib_uverbs_post_srq_recv_resp resp;
2478 	struct ib_recv_wr                  *wr, *next, *bad_wr;
2479 	struct ib_srq                      *srq;
2480 	ssize_t                             ret = -EINVAL;
2481 
2482 	if (copy_from_user(&cmd, buf, sizeof cmd))
2483 		return -EFAULT;
2484 
2485 	wr = ib_uverbs_unmarshall_recv(buf + sizeof cmd,
2486 				       in_len - sizeof cmd, cmd.wr_count,
2487 				       cmd.sge_count, cmd.wqe_size);
2488 	if (IS_ERR(wr))
2489 		return PTR_ERR(wr);
2490 
2491 	srq = idr_read_srq(cmd.srq_handle, file->ucontext);
2492 	if (!srq)
2493 		goto out;
2494 
2495 	resp.bad_wr = 0;
2496 	ret = srq->device->post_srq_recv(srq, wr, &bad_wr);
2497 
2498 	put_srq_read(srq);
2499 
2500 	if (ret)
2501 		for (next = wr; next; next = next->next) {
2502 			++resp.bad_wr;
2503 			if (next == bad_wr)
2504 				break;
2505 		}
2506 
2507 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
2508 			 &resp, sizeof resp))
2509 		ret = -EFAULT;
2510 
2511 out:
2512 	while (wr) {
2513 		next = wr->next;
2514 		kfree(wr);
2515 		wr = next;
2516 	}
2517 
2518 	return ret ? ret : in_len;
2519 }
2520 
2521 ssize_t ib_uverbs_create_ah(struct ib_uverbs_file *file,
2522 			    const char __user *buf, int in_len,
2523 			    int out_len)
2524 {
2525 	struct ib_uverbs_create_ah	 cmd;
2526 	struct ib_uverbs_create_ah_resp	 resp;
2527 	struct ib_uobject		*uobj;
2528 	struct ib_pd			*pd;
2529 	struct ib_ah			*ah;
2530 	struct ib_ah_attr		attr;
2531 	int ret;
2532 
2533 	if (out_len < sizeof resp)
2534 		return -ENOSPC;
2535 
2536 	if (copy_from_user(&cmd, buf, sizeof cmd))
2537 		return -EFAULT;
2538 
2539 	uobj = kmalloc(sizeof *uobj, GFP_KERNEL);
2540 	if (!uobj)
2541 		return -ENOMEM;
2542 
2543 	init_uobj(uobj, cmd.user_handle, file->ucontext, &ah_lock_class);
2544 	down_write(&uobj->mutex);
2545 
2546 	pd = idr_read_pd(cmd.pd_handle, file->ucontext);
2547 	if (!pd) {
2548 		ret = -EINVAL;
2549 		goto err;
2550 	}
2551 
2552 	attr.dlid 	       = cmd.attr.dlid;
2553 	attr.sl 	       = cmd.attr.sl;
2554 	attr.src_path_bits     = cmd.attr.src_path_bits;
2555 	attr.static_rate       = cmd.attr.static_rate;
2556 	attr.ah_flags          = cmd.attr.is_global ? IB_AH_GRH : 0;
2557 	attr.port_num 	       = cmd.attr.port_num;
2558 	attr.grh.flow_label    = cmd.attr.grh.flow_label;
2559 	attr.grh.sgid_index    = cmd.attr.grh.sgid_index;
2560 	attr.grh.hop_limit     = cmd.attr.grh.hop_limit;
2561 	attr.grh.traffic_class = cmd.attr.grh.traffic_class;
2562 	attr.vlan_id           = 0;
2563 	memset(&attr.dmac, 0, sizeof(attr.dmac));
2564 	memcpy(attr.grh.dgid.raw, cmd.attr.grh.dgid, 16);
2565 
2566 	ah = ib_create_ah(pd, &attr);
2567 	if (IS_ERR(ah)) {
2568 		ret = PTR_ERR(ah);
2569 		goto err_put;
2570 	}
2571 
2572 	ah->uobject  = uobj;
2573 	uobj->object = ah;
2574 
2575 	ret = idr_add_uobj(&ib_uverbs_ah_idr, uobj);
2576 	if (ret)
2577 		goto err_destroy;
2578 
2579 	resp.ah_handle = uobj->id;
2580 
2581 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
2582 			 &resp, sizeof resp)) {
2583 		ret = -EFAULT;
2584 		goto err_copy;
2585 	}
2586 
2587 	put_pd_read(pd);
2588 
2589 	mutex_lock(&file->mutex);
2590 	list_add_tail(&uobj->list, &file->ucontext->ah_list);
2591 	mutex_unlock(&file->mutex);
2592 
2593 	uobj->live = 1;
2594 
2595 	up_write(&uobj->mutex);
2596 
2597 	return in_len;
2598 
2599 err_copy:
2600 	idr_remove_uobj(&ib_uverbs_ah_idr, uobj);
2601 
2602 err_destroy:
2603 	ib_destroy_ah(ah);
2604 
2605 err_put:
2606 	put_pd_read(pd);
2607 
2608 err:
2609 	put_uobj_write(uobj);
2610 	return ret;
2611 }
2612 
2613 ssize_t ib_uverbs_destroy_ah(struct ib_uverbs_file *file,
2614 			     const char __user *buf, int in_len, int out_len)
2615 {
2616 	struct ib_uverbs_destroy_ah cmd;
2617 	struct ib_ah		   *ah;
2618 	struct ib_uobject	   *uobj;
2619 	int			    ret;
2620 
2621 	if (copy_from_user(&cmd, buf, sizeof cmd))
2622 		return -EFAULT;
2623 
2624 	uobj = idr_write_uobj(&ib_uverbs_ah_idr, cmd.ah_handle, file->ucontext);
2625 	if (!uobj)
2626 		return -EINVAL;
2627 	ah = uobj->object;
2628 
2629 	ret = ib_destroy_ah(ah);
2630 	if (!ret)
2631 		uobj->live = 0;
2632 
2633 	put_uobj_write(uobj);
2634 
2635 	if (ret)
2636 		return ret;
2637 
2638 	idr_remove_uobj(&ib_uverbs_ah_idr, uobj);
2639 
2640 	mutex_lock(&file->mutex);
2641 	list_del(&uobj->list);
2642 	mutex_unlock(&file->mutex);
2643 
2644 	put_uobj(uobj);
2645 
2646 	return in_len;
2647 }
2648 
2649 ssize_t ib_uverbs_attach_mcast(struct ib_uverbs_file *file,
2650 			       const char __user *buf, int in_len,
2651 			       int out_len)
2652 {
2653 	struct ib_uverbs_attach_mcast cmd;
2654 	struct ib_qp                 *qp;
2655 	struct ib_uqp_object         *obj;
2656 	struct ib_uverbs_mcast_entry *mcast;
2657 	int                           ret;
2658 
2659 	if (copy_from_user(&cmd, buf, sizeof cmd))
2660 		return -EFAULT;
2661 
2662 	qp = idr_write_qp(cmd.qp_handle, file->ucontext);
2663 	if (!qp)
2664 		return -EINVAL;
2665 
2666 	obj = container_of(qp->uobject, struct ib_uqp_object, uevent.uobject);
2667 
2668 	list_for_each_entry(mcast, &obj->mcast_list, list)
2669 		if (cmd.mlid == mcast->lid &&
2670 		    !memcmp(cmd.gid, mcast->gid.raw, sizeof mcast->gid.raw)) {
2671 			ret = 0;
2672 			goto out_put;
2673 		}
2674 
2675 	mcast = kmalloc(sizeof *mcast, GFP_KERNEL);
2676 	if (!mcast) {
2677 		ret = -ENOMEM;
2678 		goto out_put;
2679 	}
2680 
2681 	mcast->lid = cmd.mlid;
2682 	memcpy(mcast->gid.raw, cmd.gid, sizeof mcast->gid.raw);
2683 
2684 	ret = ib_attach_mcast(qp, &mcast->gid, cmd.mlid);
2685 	if (!ret)
2686 		list_add_tail(&mcast->list, &obj->mcast_list);
2687 	else
2688 		kfree(mcast);
2689 
2690 out_put:
2691 	put_qp_write(qp);
2692 
2693 	return ret ? ret : in_len;
2694 }
2695 
2696 ssize_t ib_uverbs_detach_mcast(struct ib_uverbs_file *file,
2697 			       const char __user *buf, int in_len,
2698 			       int out_len)
2699 {
2700 	struct ib_uverbs_detach_mcast cmd;
2701 	struct ib_uqp_object         *obj;
2702 	struct ib_qp                 *qp;
2703 	struct ib_uverbs_mcast_entry *mcast;
2704 	int                           ret = -EINVAL;
2705 
2706 	if (copy_from_user(&cmd, buf, sizeof cmd))
2707 		return -EFAULT;
2708 
2709 	qp = idr_write_qp(cmd.qp_handle, file->ucontext);
2710 	if (!qp)
2711 		return -EINVAL;
2712 
2713 	ret = ib_detach_mcast(qp, (union ib_gid *) cmd.gid, cmd.mlid);
2714 	if (ret)
2715 		goto out_put;
2716 
2717 	obj = container_of(qp->uobject, struct ib_uqp_object, uevent.uobject);
2718 
2719 	list_for_each_entry(mcast, &obj->mcast_list, list)
2720 		if (cmd.mlid == mcast->lid &&
2721 		    !memcmp(cmd.gid, mcast->gid.raw, sizeof mcast->gid.raw)) {
2722 			list_del(&mcast->list);
2723 			kfree(mcast);
2724 			break;
2725 		}
2726 
2727 out_put:
2728 	put_qp_write(qp);
2729 
2730 	return ret ? ret : in_len;
2731 }
2732 
2733 static int kern_spec_to_ib_spec(struct ib_uverbs_flow_spec *kern_spec,
2734 				union ib_flow_spec *ib_spec)
2735 {
2736 	if (kern_spec->reserved)
2737 		return -EINVAL;
2738 
2739 	ib_spec->type = kern_spec->type;
2740 
2741 	switch (ib_spec->type) {
2742 	case IB_FLOW_SPEC_ETH:
2743 		ib_spec->eth.size = sizeof(struct ib_flow_spec_eth);
2744 		if (ib_spec->eth.size != kern_spec->eth.size)
2745 			return -EINVAL;
2746 		memcpy(&ib_spec->eth.val, &kern_spec->eth.val,
2747 		       sizeof(struct ib_flow_eth_filter));
2748 		memcpy(&ib_spec->eth.mask, &kern_spec->eth.mask,
2749 		       sizeof(struct ib_flow_eth_filter));
2750 		break;
2751 	case IB_FLOW_SPEC_IPV4:
2752 		ib_spec->ipv4.size = sizeof(struct ib_flow_spec_ipv4);
2753 		if (ib_spec->ipv4.size != kern_spec->ipv4.size)
2754 			return -EINVAL;
2755 		memcpy(&ib_spec->ipv4.val, &kern_spec->ipv4.val,
2756 		       sizeof(struct ib_flow_ipv4_filter));
2757 		memcpy(&ib_spec->ipv4.mask, &kern_spec->ipv4.mask,
2758 		       sizeof(struct ib_flow_ipv4_filter));
2759 		break;
2760 	case IB_FLOW_SPEC_TCP:
2761 	case IB_FLOW_SPEC_UDP:
2762 		ib_spec->tcp_udp.size = sizeof(struct ib_flow_spec_tcp_udp);
2763 		if (ib_spec->tcp_udp.size != kern_spec->tcp_udp.size)
2764 			return -EINVAL;
2765 		memcpy(&ib_spec->tcp_udp.val, &kern_spec->tcp_udp.val,
2766 		       sizeof(struct ib_flow_tcp_udp_filter));
2767 		memcpy(&ib_spec->tcp_udp.mask, &kern_spec->tcp_udp.mask,
2768 		       sizeof(struct ib_flow_tcp_udp_filter));
2769 		break;
2770 	default:
2771 		return -EINVAL;
2772 	}
2773 	return 0;
2774 }
2775 
2776 int ib_uverbs_ex_create_flow(struct ib_uverbs_file *file,
2777 			     struct ib_udata *ucore,
2778 			     struct ib_udata *uhw)
2779 {
2780 	struct ib_uverbs_create_flow	  cmd;
2781 	struct ib_uverbs_create_flow_resp resp;
2782 	struct ib_uobject		  *uobj;
2783 	struct ib_flow			  *flow_id;
2784 	struct ib_uverbs_flow_attr	  *kern_flow_attr;
2785 	struct ib_flow_attr		  *flow_attr;
2786 	struct ib_qp			  *qp;
2787 	int err = 0;
2788 	void *kern_spec;
2789 	void *ib_spec;
2790 	int i;
2791 
2792 	if (ucore->inlen < sizeof(cmd))
2793 		return -EINVAL;
2794 
2795 	if (ucore->outlen < sizeof(resp))
2796 		return -ENOSPC;
2797 
2798 	err = ib_copy_from_udata(&cmd, ucore, sizeof(cmd));
2799 	if (err)
2800 		return err;
2801 
2802 	ucore->inbuf += sizeof(cmd);
2803 	ucore->inlen -= sizeof(cmd);
2804 
2805 	if (cmd.comp_mask)
2806 		return -EINVAL;
2807 
2808 	if ((cmd.flow_attr.type == IB_FLOW_ATTR_SNIFFER &&
2809 	     !capable(CAP_NET_ADMIN)) || !capable(CAP_NET_RAW))
2810 		return -EPERM;
2811 
2812 	if (cmd.flow_attr.num_of_specs > IB_FLOW_SPEC_SUPPORT_LAYERS)
2813 		return -EINVAL;
2814 
2815 	if (cmd.flow_attr.size > ucore->inlen ||
2816 	    cmd.flow_attr.size >
2817 	    (cmd.flow_attr.num_of_specs * sizeof(struct ib_uverbs_flow_spec)))
2818 		return -EINVAL;
2819 
2820 	if (cmd.flow_attr.reserved[0] ||
2821 	    cmd.flow_attr.reserved[1])
2822 		return -EINVAL;
2823 
2824 	if (cmd.flow_attr.num_of_specs) {
2825 		kern_flow_attr = kmalloc(sizeof(*kern_flow_attr) + cmd.flow_attr.size,
2826 					 GFP_KERNEL);
2827 		if (!kern_flow_attr)
2828 			return -ENOMEM;
2829 
2830 		memcpy(kern_flow_attr, &cmd.flow_attr, sizeof(*kern_flow_attr));
2831 		err = ib_copy_from_udata(kern_flow_attr + 1, ucore,
2832 					 cmd.flow_attr.size);
2833 		if (err)
2834 			goto err_free_attr;
2835 	} else {
2836 		kern_flow_attr = &cmd.flow_attr;
2837 	}
2838 
2839 	uobj = kmalloc(sizeof(*uobj), GFP_KERNEL);
2840 	if (!uobj) {
2841 		err = -ENOMEM;
2842 		goto err_free_attr;
2843 	}
2844 	init_uobj(uobj, 0, file->ucontext, &rule_lock_class);
2845 	down_write(&uobj->mutex);
2846 
2847 	qp = idr_read_qp(cmd.qp_handle, file->ucontext);
2848 	if (!qp) {
2849 		err = -EINVAL;
2850 		goto err_uobj;
2851 	}
2852 
2853 	flow_attr = kmalloc(sizeof(*flow_attr) + cmd.flow_attr.size, GFP_KERNEL);
2854 	if (!flow_attr) {
2855 		err = -ENOMEM;
2856 		goto err_put;
2857 	}
2858 
2859 	flow_attr->type = kern_flow_attr->type;
2860 	flow_attr->priority = kern_flow_attr->priority;
2861 	flow_attr->num_of_specs = kern_flow_attr->num_of_specs;
2862 	flow_attr->port = kern_flow_attr->port;
2863 	flow_attr->flags = kern_flow_attr->flags;
2864 	flow_attr->size = sizeof(*flow_attr);
2865 
2866 	kern_spec = kern_flow_attr + 1;
2867 	ib_spec = flow_attr + 1;
2868 	for (i = 0; i < flow_attr->num_of_specs &&
2869 	     cmd.flow_attr.size > offsetof(struct ib_uverbs_flow_spec, reserved) &&
2870 	     cmd.flow_attr.size >=
2871 	     ((struct ib_uverbs_flow_spec *)kern_spec)->size; i++) {
2872 		err = kern_spec_to_ib_spec(kern_spec, ib_spec);
2873 		if (err)
2874 			goto err_free;
2875 		flow_attr->size +=
2876 			((union ib_flow_spec *) ib_spec)->size;
2877 		cmd.flow_attr.size -= ((struct ib_uverbs_flow_spec *)kern_spec)->size;
2878 		kern_spec += ((struct ib_uverbs_flow_spec *) kern_spec)->size;
2879 		ib_spec += ((union ib_flow_spec *) ib_spec)->size;
2880 	}
2881 	if (cmd.flow_attr.size || (i != flow_attr->num_of_specs)) {
2882 		pr_warn("create flow failed, flow %d: %d bytes left from uverb cmd\n",
2883 			i, cmd.flow_attr.size);
2884 		err = -EINVAL;
2885 		goto err_free;
2886 	}
2887 	flow_id = ib_create_flow(qp, flow_attr, IB_FLOW_DOMAIN_USER);
2888 	if (IS_ERR(flow_id)) {
2889 		err = PTR_ERR(flow_id);
2890 		goto err_free;
2891 	}
2892 	flow_id->qp = qp;
2893 	flow_id->uobject = uobj;
2894 	uobj->object = flow_id;
2895 
2896 	err = idr_add_uobj(&ib_uverbs_rule_idr, uobj);
2897 	if (err)
2898 		goto destroy_flow;
2899 
2900 	memset(&resp, 0, sizeof(resp));
2901 	resp.flow_handle = uobj->id;
2902 
2903 	err = ib_copy_to_udata(ucore,
2904 			       &resp, sizeof(resp));
2905 	if (err)
2906 		goto err_copy;
2907 
2908 	put_qp_read(qp);
2909 	mutex_lock(&file->mutex);
2910 	list_add_tail(&uobj->list, &file->ucontext->rule_list);
2911 	mutex_unlock(&file->mutex);
2912 
2913 	uobj->live = 1;
2914 
2915 	up_write(&uobj->mutex);
2916 	kfree(flow_attr);
2917 	if (cmd.flow_attr.num_of_specs)
2918 		kfree(kern_flow_attr);
2919 	return 0;
2920 err_copy:
2921 	idr_remove_uobj(&ib_uverbs_rule_idr, uobj);
2922 destroy_flow:
2923 	ib_destroy_flow(flow_id);
2924 err_free:
2925 	kfree(flow_attr);
2926 err_put:
2927 	put_qp_read(qp);
2928 err_uobj:
2929 	put_uobj_write(uobj);
2930 err_free_attr:
2931 	if (cmd.flow_attr.num_of_specs)
2932 		kfree(kern_flow_attr);
2933 	return err;
2934 }
2935 
2936 int ib_uverbs_ex_destroy_flow(struct ib_uverbs_file *file,
2937 			      struct ib_udata *ucore,
2938 			      struct ib_udata *uhw)
2939 {
2940 	struct ib_uverbs_destroy_flow	cmd;
2941 	struct ib_flow			*flow_id;
2942 	struct ib_uobject		*uobj;
2943 	int				ret;
2944 
2945 	if (ucore->inlen < sizeof(cmd))
2946 		return -EINVAL;
2947 
2948 	ret = ib_copy_from_udata(&cmd, ucore, sizeof(cmd));
2949 	if (ret)
2950 		return ret;
2951 
2952 	if (cmd.comp_mask)
2953 		return -EINVAL;
2954 
2955 	uobj = idr_write_uobj(&ib_uverbs_rule_idr, cmd.flow_handle,
2956 			      file->ucontext);
2957 	if (!uobj)
2958 		return -EINVAL;
2959 	flow_id = uobj->object;
2960 
2961 	ret = ib_destroy_flow(flow_id);
2962 	if (!ret)
2963 		uobj->live = 0;
2964 
2965 	put_uobj_write(uobj);
2966 
2967 	idr_remove_uobj(&ib_uverbs_rule_idr, uobj);
2968 
2969 	mutex_lock(&file->mutex);
2970 	list_del(&uobj->list);
2971 	mutex_unlock(&file->mutex);
2972 
2973 	put_uobj(uobj);
2974 
2975 	return ret;
2976 }
2977 
2978 static int __uverbs_create_xsrq(struct ib_uverbs_file *file,
2979 				struct ib_uverbs_create_xsrq *cmd,
2980 				struct ib_udata *udata)
2981 {
2982 	struct ib_uverbs_create_srq_resp resp;
2983 	struct ib_usrq_object           *obj;
2984 	struct ib_pd                    *pd;
2985 	struct ib_srq                   *srq;
2986 	struct ib_uobject               *uninitialized_var(xrcd_uobj);
2987 	struct ib_srq_init_attr          attr;
2988 	int ret;
2989 
2990 	obj = kmalloc(sizeof *obj, GFP_KERNEL);
2991 	if (!obj)
2992 		return -ENOMEM;
2993 
2994 	init_uobj(&obj->uevent.uobject, cmd->user_handle, file->ucontext, &srq_lock_class);
2995 	down_write(&obj->uevent.uobject.mutex);
2996 
2997 	if (cmd->srq_type == IB_SRQT_XRC) {
2998 		attr.ext.xrc.xrcd  = idr_read_xrcd(cmd->xrcd_handle, file->ucontext, &xrcd_uobj);
2999 		if (!attr.ext.xrc.xrcd) {
3000 			ret = -EINVAL;
3001 			goto err;
3002 		}
3003 
3004 		obj->uxrcd = container_of(xrcd_uobj, struct ib_uxrcd_object, uobject);
3005 		atomic_inc(&obj->uxrcd->refcnt);
3006 
3007 		attr.ext.xrc.cq  = idr_read_cq(cmd->cq_handle, file->ucontext, 0);
3008 		if (!attr.ext.xrc.cq) {
3009 			ret = -EINVAL;
3010 			goto err_put_xrcd;
3011 		}
3012 	}
3013 
3014 	pd  = idr_read_pd(cmd->pd_handle, file->ucontext);
3015 	if (!pd) {
3016 		ret = -EINVAL;
3017 		goto err_put_cq;
3018 	}
3019 
3020 	attr.event_handler  = ib_uverbs_srq_event_handler;
3021 	attr.srq_context    = file;
3022 	attr.srq_type       = cmd->srq_type;
3023 	attr.attr.max_wr    = cmd->max_wr;
3024 	attr.attr.max_sge   = cmd->max_sge;
3025 	attr.attr.srq_limit = cmd->srq_limit;
3026 
3027 	obj->uevent.events_reported = 0;
3028 	INIT_LIST_HEAD(&obj->uevent.event_list);
3029 
3030 	srq = pd->device->create_srq(pd, &attr, udata);
3031 	if (IS_ERR(srq)) {
3032 		ret = PTR_ERR(srq);
3033 		goto err_put;
3034 	}
3035 
3036 	srq->device        = pd->device;
3037 	srq->pd            = pd;
3038 	srq->srq_type	   = cmd->srq_type;
3039 	srq->uobject       = &obj->uevent.uobject;
3040 	srq->event_handler = attr.event_handler;
3041 	srq->srq_context   = attr.srq_context;
3042 
3043 	if (cmd->srq_type == IB_SRQT_XRC) {
3044 		srq->ext.xrc.cq   = attr.ext.xrc.cq;
3045 		srq->ext.xrc.xrcd = attr.ext.xrc.xrcd;
3046 		atomic_inc(&attr.ext.xrc.cq->usecnt);
3047 		atomic_inc(&attr.ext.xrc.xrcd->usecnt);
3048 	}
3049 
3050 	atomic_inc(&pd->usecnt);
3051 	atomic_set(&srq->usecnt, 0);
3052 
3053 	obj->uevent.uobject.object = srq;
3054 	ret = idr_add_uobj(&ib_uverbs_srq_idr, &obj->uevent.uobject);
3055 	if (ret)
3056 		goto err_destroy;
3057 
3058 	memset(&resp, 0, sizeof resp);
3059 	resp.srq_handle = obj->uevent.uobject.id;
3060 	resp.max_wr     = attr.attr.max_wr;
3061 	resp.max_sge    = attr.attr.max_sge;
3062 	if (cmd->srq_type == IB_SRQT_XRC)
3063 		resp.srqn = srq->ext.xrc.srq_num;
3064 
3065 	if (copy_to_user((void __user *) (unsigned long) cmd->response,
3066 			 &resp, sizeof resp)) {
3067 		ret = -EFAULT;
3068 		goto err_copy;
3069 	}
3070 
3071 	if (cmd->srq_type == IB_SRQT_XRC) {
3072 		put_uobj_read(xrcd_uobj);
3073 		put_cq_read(attr.ext.xrc.cq);
3074 	}
3075 	put_pd_read(pd);
3076 
3077 	mutex_lock(&file->mutex);
3078 	list_add_tail(&obj->uevent.uobject.list, &file->ucontext->srq_list);
3079 	mutex_unlock(&file->mutex);
3080 
3081 	obj->uevent.uobject.live = 1;
3082 
3083 	up_write(&obj->uevent.uobject.mutex);
3084 
3085 	return 0;
3086 
3087 err_copy:
3088 	idr_remove_uobj(&ib_uverbs_srq_idr, &obj->uevent.uobject);
3089 
3090 err_destroy:
3091 	ib_destroy_srq(srq);
3092 
3093 err_put:
3094 	put_pd_read(pd);
3095 
3096 err_put_cq:
3097 	if (cmd->srq_type == IB_SRQT_XRC)
3098 		put_cq_read(attr.ext.xrc.cq);
3099 
3100 err_put_xrcd:
3101 	if (cmd->srq_type == IB_SRQT_XRC) {
3102 		atomic_dec(&obj->uxrcd->refcnt);
3103 		put_uobj_read(xrcd_uobj);
3104 	}
3105 
3106 err:
3107 	put_uobj_write(&obj->uevent.uobject);
3108 	return ret;
3109 }
3110 
3111 ssize_t ib_uverbs_create_srq(struct ib_uverbs_file *file,
3112 			     const char __user *buf, int in_len,
3113 			     int out_len)
3114 {
3115 	struct ib_uverbs_create_srq      cmd;
3116 	struct ib_uverbs_create_xsrq     xcmd;
3117 	struct ib_uverbs_create_srq_resp resp;
3118 	struct ib_udata                  udata;
3119 	int ret;
3120 
3121 	if (out_len < sizeof resp)
3122 		return -ENOSPC;
3123 
3124 	if (copy_from_user(&cmd, buf, sizeof cmd))
3125 		return -EFAULT;
3126 
3127 	xcmd.response	 = cmd.response;
3128 	xcmd.user_handle = cmd.user_handle;
3129 	xcmd.srq_type	 = IB_SRQT_BASIC;
3130 	xcmd.pd_handle	 = cmd.pd_handle;
3131 	xcmd.max_wr	 = cmd.max_wr;
3132 	xcmd.max_sge	 = cmd.max_sge;
3133 	xcmd.srq_limit	 = cmd.srq_limit;
3134 
3135 	INIT_UDATA(&udata, buf + sizeof cmd,
3136 		   (unsigned long) cmd.response + sizeof resp,
3137 		   in_len - sizeof cmd, out_len - sizeof resp);
3138 
3139 	ret = __uverbs_create_xsrq(file, &xcmd, &udata);
3140 	if (ret)
3141 		return ret;
3142 
3143 	return in_len;
3144 }
3145 
3146 ssize_t ib_uverbs_create_xsrq(struct ib_uverbs_file *file,
3147 			      const char __user *buf, int in_len, int out_len)
3148 {
3149 	struct ib_uverbs_create_xsrq     cmd;
3150 	struct ib_uverbs_create_srq_resp resp;
3151 	struct ib_udata                  udata;
3152 	int ret;
3153 
3154 	if (out_len < sizeof resp)
3155 		return -ENOSPC;
3156 
3157 	if (copy_from_user(&cmd, buf, sizeof cmd))
3158 		return -EFAULT;
3159 
3160 	INIT_UDATA(&udata, buf + sizeof cmd,
3161 		   (unsigned long) cmd.response + sizeof resp,
3162 		   in_len - sizeof cmd, out_len - sizeof resp);
3163 
3164 	ret = __uverbs_create_xsrq(file, &cmd, &udata);
3165 	if (ret)
3166 		return ret;
3167 
3168 	return in_len;
3169 }
3170 
3171 ssize_t ib_uverbs_modify_srq(struct ib_uverbs_file *file,
3172 			     const char __user *buf, int in_len,
3173 			     int out_len)
3174 {
3175 	struct ib_uverbs_modify_srq cmd;
3176 	struct ib_udata             udata;
3177 	struct ib_srq              *srq;
3178 	struct ib_srq_attr          attr;
3179 	int                         ret;
3180 
3181 	if (copy_from_user(&cmd, buf, sizeof cmd))
3182 		return -EFAULT;
3183 
3184 	INIT_UDATA(&udata, buf + sizeof cmd, NULL, in_len - sizeof cmd,
3185 		   out_len);
3186 
3187 	srq = idr_read_srq(cmd.srq_handle, file->ucontext);
3188 	if (!srq)
3189 		return -EINVAL;
3190 
3191 	attr.max_wr    = cmd.max_wr;
3192 	attr.srq_limit = cmd.srq_limit;
3193 
3194 	ret = srq->device->modify_srq(srq, &attr, cmd.attr_mask, &udata);
3195 
3196 	put_srq_read(srq);
3197 
3198 	return ret ? ret : in_len;
3199 }
3200 
3201 ssize_t ib_uverbs_query_srq(struct ib_uverbs_file *file,
3202 			    const char __user *buf,
3203 			    int in_len, int out_len)
3204 {
3205 	struct ib_uverbs_query_srq      cmd;
3206 	struct ib_uverbs_query_srq_resp resp;
3207 	struct ib_srq_attr              attr;
3208 	struct ib_srq                   *srq;
3209 	int                             ret;
3210 
3211 	if (out_len < sizeof resp)
3212 		return -ENOSPC;
3213 
3214 	if (copy_from_user(&cmd, buf, sizeof cmd))
3215 		return -EFAULT;
3216 
3217 	srq = idr_read_srq(cmd.srq_handle, file->ucontext);
3218 	if (!srq)
3219 		return -EINVAL;
3220 
3221 	ret = ib_query_srq(srq, &attr);
3222 
3223 	put_srq_read(srq);
3224 
3225 	if (ret)
3226 		return ret;
3227 
3228 	memset(&resp, 0, sizeof resp);
3229 
3230 	resp.max_wr    = attr.max_wr;
3231 	resp.max_sge   = attr.max_sge;
3232 	resp.srq_limit = attr.srq_limit;
3233 
3234 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
3235 			 &resp, sizeof resp))
3236 		return -EFAULT;
3237 
3238 	return in_len;
3239 }
3240 
3241 ssize_t ib_uverbs_destroy_srq(struct ib_uverbs_file *file,
3242 			      const char __user *buf, int in_len,
3243 			      int out_len)
3244 {
3245 	struct ib_uverbs_destroy_srq      cmd;
3246 	struct ib_uverbs_destroy_srq_resp resp;
3247 	struct ib_uobject		 *uobj;
3248 	struct ib_srq               	 *srq;
3249 	struct ib_uevent_object        	 *obj;
3250 	int                         	  ret = -EINVAL;
3251 	struct ib_usrq_object		 *us;
3252 	enum ib_srq_type		  srq_type;
3253 
3254 	if (copy_from_user(&cmd, buf, sizeof cmd))
3255 		return -EFAULT;
3256 
3257 	uobj = idr_write_uobj(&ib_uverbs_srq_idr, cmd.srq_handle, file->ucontext);
3258 	if (!uobj)
3259 		return -EINVAL;
3260 	srq = uobj->object;
3261 	obj = container_of(uobj, struct ib_uevent_object, uobject);
3262 	srq_type = srq->srq_type;
3263 
3264 	ret = ib_destroy_srq(srq);
3265 	if (!ret)
3266 		uobj->live = 0;
3267 
3268 	put_uobj_write(uobj);
3269 
3270 	if (ret)
3271 		return ret;
3272 
3273 	if (srq_type == IB_SRQT_XRC) {
3274 		us = container_of(obj, struct ib_usrq_object, uevent);
3275 		atomic_dec(&us->uxrcd->refcnt);
3276 	}
3277 
3278 	idr_remove_uobj(&ib_uverbs_srq_idr, uobj);
3279 
3280 	mutex_lock(&file->mutex);
3281 	list_del(&uobj->list);
3282 	mutex_unlock(&file->mutex);
3283 
3284 	ib_uverbs_release_uevent(file, obj);
3285 
3286 	memset(&resp, 0, sizeof resp);
3287 	resp.events_reported = obj->events_reported;
3288 
3289 	put_uobj(uobj);
3290 
3291 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
3292 			 &resp, sizeof resp))
3293 		ret = -EFAULT;
3294 
3295 	return ret ? ret : in_len;
3296 }
3297 
3298 int ib_uverbs_ex_query_device(struct ib_uverbs_file *file,
3299 			      struct ib_udata *ucore,
3300 			      struct ib_udata *uhw)
3301 {
3302 	struct ib_uverbs_ex_query_device_resp resp;
3303 	struct ib_uverbs_ex_query_device  cmd;
3304 	struct ib_device_attr attr;
3305 	struct ib_device *device;
3306 	int err;
3307 
3308 	device = file->device->ib_dev;
3309 	if (ucore->inlen < sizeof(cmd))
3310 		return -EINVAL;
3311 
3312 	err = ib_copy_from_udata(&cmd, ucore, sizeof(cmd));
3313 	if (err)
3314 		return err;
3315 
3316 	if (cmd.comp_mask)
3317 		return -EINVAL;
3318 
3319 	if (cmd.reserved)
3320 		return -EINVAL;
3321 
3322 	resp.response_length = offsetof(typeof(resp), odp_caps);
3323 
3324 	if (ucore->outlen < resp.response_length)
3325 		return -ENOSPC;
3326 
3327 	err = device->query_device(device, &attr);
3328 	if (err)
3329 		return err;
3330 
3331 	copy_query_dev_fields(file, &resp.base, &attr);
3332 	resp.comp_mask = 0;
3333 
3334 	if (ucore->outlen < resp.response_length + sizeof(resp.odp_caps))
3335 		goto end;
3336 
3337 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
3338 	resp.odp_caps.general_caps = attr.odp_caps.general_caps;
3339 	resp.odp_caps.per_transport_caps.rc_odp_caps =
3340 		attr.odp_caps.per_transport_caps.rc_odp_caps;
3341 	resp.odp_caps.per_transport_caps.uc_odp_caps =
3342 		attr.odp_caps.per_transport_caps.uc_odp_caps;
3343 	resp.odp_caps.per_transport_caps.ud_odp_caps =
3344 		attr.odp_caps.per_transport_caps.ud_odp_caps;
3345 	resp.odp_caps.reserved = 0;
3346 #else
3347 	memset(&resp.odp_caps, 0, sizeof(resp.odp_caps));
3348 #endif
3349 	resp.response_length += sizeof(resp.odp_caps);
3350 
3351 end:
3352 	err = ib_copy_to_udata(ucore, &resp, resp.response_length);
3353 	if (err)
3354 		return err;
3355 
3356 	return 0;
3357 }
3358