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 write operations.  When an object is
66  *   looked up, a reference must be taken on the object's kref before
67  *   dropping this lock.  For read operations, the rcu_read_lock()
68  *   and rcu_write_lock() but similarly the kref reference is grabbed
69  *   before the rcu_read_unlock().
70  *
71  * - Each object also has an rwsem.  This rwsem must be held for
72  *   reading while an operation that uses the object is performed.
73  *   For example, while registering an MR, the associated PD's
74  *   uobject.mutex must be held for reading.  The rwsem must be held
75  *   for writing while initializing or destroying an object.
76  *
77  * - In addition, each object has a "live" flag.  If this flag is not
78  *   set, then lookups of the object will fail even if it is found in
79  *   the idr.  This handles a reader that blocks and does not acquire
80  *   the rwsem until after the object is destroyed.  The destroy
81  *   operation will set the live flag to 0 and then drop the rwsem;
82  *   this will allow the reader to acquire the rwsem, see that the
83  *   live flag is 0, and then drop the rwsem and its reference to
84  *   object.  The underlying storage will not be freed until the last
85  *   reference to the object is dropped.
86  */
87 
88 static void init_uobj(struct ib_uobject *uobj, u64 user_handle,
89 		      struct ib_ucontext *context, struct uverbs_lock_class *c)
90 {
91 	uobj->user_handle = user_handle;
92 	uobj->context     = context;
93 	kref_init(&uobj->ref);
94 	init_rwsem(&uobj->mutex);
95 	lockdep_set_class_and_name(&uobj->mutex, &c->key, c->name);
96 	uobj->live        = 0;
97 }
98 
99 static void release_uobj(struct kref *kref)
100 {
101 	kfree_rcu(container_of(kref, struct ib_uobject, ref), rcu);
102 }
103 
104 static void put_uobj(struct ib_uobject *uobj)
105 {
106 	kref_put(&uobj->ref, release_uobj);
107 }
108 
109 static void put_uobj_read(struct ib_uobject *uobj)
110 {
111 	up_read(&uobj->mutex);
112 	put_uobj(uobj);
113 }
114 
115 static void put_uobj_write(struct ib_uobject *uobj)
116 {
117 	up_write(&uobj->mutex);
118 	put_uobj(uobj);
119 }
120 
121 static int idr_add_uobj(struct idr *idr, struct ib_uobject *uobj)
122 {
123 	int ret;
124 
125 	idr_preload(GFP_KERNEL);
126 	spin_lock(&ib_uverbs_idr_lock);
127 
128 	ret = idr_alloc(idr, uobj, 0, 0, GFP_NOWAIT);
129 	if (ret >= 0)
130 		uobj->id = ret;
131 
132 	spin_unlock(&ib_uverbs_idr_lock);
133 	idr_preload_end();
134 
135 	return ret < 0 ? ret : 0;
136 }
137 
138 void idr_remove_uobj(struct idr *idr, struct ib_uobject *uobj)
139 {
140 	spin_lock(&ib_uverbs_idr_lock);
141 	idr_remove(idr, uobj->id);
142 	spin_unlock(&ib_uverbs_idr_lock);
143 }
144 
145 static struct ib_uobject *__idr_get_uobj(struct idr *idr, int id,
146 					 struct ib_ucontext *context)
147 {
148 	struct ib_uobject *uobj;
149 
150 	rcu_read_lock();
151 	uobj = idr_find(idr, id);
152 	if (uobj) {
153 		if (uobj->context == context)
154 			kref_get(&uobj->ref);
155 		else
156 			uobj = NULL;
157 	}
158 	rcu_read_unlock();
159 
160 	return uobj;
161 }
162 
163 static struct ib_uobject *idr_read_uobj(struct idr *idr, int id,
164 					struct ib_ucontext *context, int nested)
165 {
166 	struct ib_uobject *uobj;
167 
168 	uobj = __idr_get_uobj(idr, id, context);
169 	if (!uobj)
170 		return NULL;
171 
172 	if (nested)
173 		down_read_nested(&uobj->mutex, SINGLE_DEPTH_NESTING);
174 	else
175 		down_read(&uobj->mutex);
176 	if (!uobj->live) {
177 		put_uobj_read(uobj);
178 		return NULL;
179 	}
180 
181 	return uobj;
182 }
183 
184 static struct ib_uobject *idr_write_uobj(struct idr *idr, int id,
185 					 struct ib_ucontext *context)
186 {
187 	struct ib_uobject *uobj;
188 
189 	uobj = __idr_get_uobj(idr, id, context);
190 	if (!uobj)
191 		return NULL;
192 
193 	down_write(&uobj->mutex);
194 	if (!uobj->live) {
195 		put_uobj_write(uobj);
196 		return NULL;
197 	}
198 
199 	return uobj;
200 }
201 
202 static void *idr_read_obj(struct idr *idr, int id, struct ib_ucontext *context,
203 			  int nested)
204 {
205 	struct ib_uobject *uobj;
206 
207 	uobj = idr_read_uobj(idr, id, context, nested);
208 	return uobj ? uobj->object : NULL;
209 }
210 
211 static struct ib_pd *idr_read_pd(int pd_handle, struct ib_ucontext *context)
212 {
213 	return idr_read_obj(&ib_uverbs_pd_idr, pd_handle, context, 0);
214 }
215 
216 static void put_pd_read(struct ib_pd *pd)
217 {
218 	put_uobj_read(pd->uobject);
219 }
220 
221 static struct ib_cq *idr_read_cq(int cq_handle, struct ib_ucontext *context, int nested)
222 {
223 	return idr_read_obj(&ib_uverbs_cq_idr, cq_handle, context, nested);
224 }
225 
226 static void put_cq_read(struct ib_cq *cq)
227 {
228 	put_uobj_read(cq->uobject);
229 }
230 
231 static struct ib_ah *idr_read_ah(int ah_handle, struct ib_ucontext *context)
232 {
233 	return idr_read_obj(&ib_uverbs_ah_idr, ah_handle, context, 0);
234 }
235 
236 static void put_ah_read(struct ib_ah *ah)
237 {
238 	put_uobj_read(ah->uobject);
239 }
240 
241 static struct ib_qp *idr_read_qp(int qp_handle, struct ib_ucontext *context)
242 {
243 	return idr_read_obj(&ib_uverbs_qp_idr, qp_handle, context, 0);
244 }
245 
246 static struct ib_qp *idr_write_qp(int qp_handle, struct ib_ucontext *context)
247 {
248 	struct ib_uobject *uobj;
249 
250 	uobj = idr_write_uobj(&ib_uverbs_qp_idr, qp_handle, context);
251 	return uobj ? uobj->object : NULL;
252 }
253 
254 static void put_qp_read(struct ib_qp *qp)
255 {
256 	put_uobj_read(qp->uobject);
257 }
258 
259 static void put_qp_write(struct ib_qp *qp)
260 {
261 	put_uobj_write(qp->uobject);
262 }
263 
264 static struct ib_srq *idr_read_srq(int srq_handle, struct ib_ucontext *context)
265 {
266 	return idr_read_obj(&ib_uverbs_srq_idr, srq_handle, context, 0);
267 }
268 
269 static void put_srq_read(struct ib_srq *srq)
270 {
271 	put_uobj_read(srq->uobject);
272 }
273 
274 static struct ib_xrcd *idr_read_xrcd(int xrcd_handle, struct ib_ucontext *context,
275 				     struct ib_uobject **uobj)
276 {
277 	*uobj = idr_read_uobj(&ib_uverbs_xrcd_idr, xrcd_handle, context, 0);
278 	return *uobj ? (*uobj)->object : NULL;
279 }
280 
281 static void put_xrcd_read(struct ib_uobject *uobj)
282 {
283 	put_uobj_read(uobj);
284 }
285 
286 ssize_t ib_uverbs_get_context(struct ib_uverbs_file *file,
287 			      struct ib_device *ib_dev,
288 			      const char __user *buf,
289 			      int in_len, int out_len)
290 {
291 	struct ib_uverbs_get_context      cmd;
292 	struct ib_uverbs_get_context_resp resp;
293 	struct ib_udata                   udata;
294 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
295 	struct ib_device_attr		  dev_attr;
296 #endif
297 	struct ib_ucontext		 *ucontext;
298 	struct file			 *filp;
299 	int ret;
300 
301 	if (out_len < sizeof resp)
302 		return -ENOSPC;
303 
304 	if (copy_from_user(&cmd, buf, sizeof cmd))
305 		return -EFAULT;
306 
307 	mutex_lock(&file->mutex);
308 
309 	if (file->ucontext) {
310 		ret = -EINVAL;
311 		goto err;
312 	}
313 
314 	INIT_UDATA(&udata, buf + sizeof cmd,
315 		   (unsigned long) cmd.response + sizeof resp,
316 		   in_len - sizeof cmd, out_len - sizeof resp);
317 
318 	ucontext = ib_dev->alloc_ucontext(ib_dev, &udata);
319 	if (IS_ERR(ucontext)) {
320 		ret = PTR_ERR(ucontext);
321 		goto err;
322 	}
323 
324 	ucontext->device = ib_dev;
325 	INIT_LIST_HEAD(&ucontext->pd_list);
326 	INIT_LIST_HEAD(&ucontext->mr_list);
327 	INIT_LIST_HEAD(&ucontext->mw_list);
328 	INIT_LIST_HEAD(&ucontext->cq_list);
329 	INIT_LIST_HEAD(&ucontext->qp_list);
330 	INIT_LIST_HEAD(&ucontext->srq_list);
331 	INIT_LIST_HEAD(&ucontext->ah_list);
332 	INIT_LIST_HEAD(&ucontext->xrcd_list);
333 	INIT_LIST_HEAD(&ucontext->rule_list);
334 	rcu_read_lock();
335 	ucontext->tgid = get_task_pid(current->group_leader, PIDTYPE_PID);
336 	rcu_read_unlock();
337 	ucontext->closing = 0;
338 
339 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
340 	ucontext->umem_tree = RB_ROOT;
341 	init_rwsem(&ucontext->umem_rwsem);
342 	ucontext->odp_mrs_count = 0;
343 	INIT_LIST_HEAD(&ucontext->no_private_counters);
344 
345 	ret = ib_query_device(ib_dev, &dev_attr);
346 	if (ret)
347 		goto err_free;
348 	if (!(dev_attr.device_cap_flags & IB_DEVICE_ON_DEMAND_PAGING))
349 		ucontext->invalidate_range = NULL;
350 
351 #endif
352 
353 	resp.num_comp_vectors = file->device->num_comp_vectors;
354 
355 	ret = get_unused_fd_flags(O_CLOEXEC);
356 	if (ret < 0)
357 		goto err_free;
358 	resp.async_fd = ret;
359 
360 	filp = ib_uverbs_alloc_event_file(file, ib_dev, 1);
361 	if (IS_ERR(filp)) {
362 		ret = PTR_ERR(filp);
363 		goto err_fd;
364 	}
365 
366 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
367 			 &resp, sizeof resp)) {
368 		ret = -EFAULT;
369 		goto err_file;
370 	}
371 
372 	file->ucontext = ucontext;
373 
374 	fd_install(resp.async_fd, filp);
375 
376 	mutex_unlock(&file->mutex);
377 
378 	return in_len;
379 
380 err_file:
381 	ib_uverbs_free_async_event_file(file);
382 	fput(filp);
383 
384 err_fd:
385 	put_unused_fd(resp.async_fd);
386 
387 err_free:
388 	put_pid(ucontext->tgid);
389 	ib_dev->dealloc_ucontext(ucontext);
390 
391 err:
392 	mutex_unlock(&file->mutex);
393 	return ret;
394 }
395 
396 static void copy_query_dev_fields(struct ib_uverbs_file *file,
397 				  struct ib_device *ib_dev,
398 				  struct ib_uverbs_query_device_resp *resp,
399 				  struct ib_device_attr *attr)
400 {
401 	resp->fw_ver		= attr->fw_ver;
402 	resp->node_guid		= ib_dev->node_guid;
403 	resp->sys_image_guid	= attr->sys_image_guid;
404 	resp->max_mr_size	= attr->max_mr_size;
405 	resp->page_size_cap	= attr->page_size_cap;
406 	resp->vendor_id		= attr->vendor_id;
407 	resp->vendor_part_id	= attr->vendor_part_id;
408 	resp->hw_ver		= attr->hw_ver;
409 	resp->max_qp		= attr->max_qp;
410 	resp->max_qp_wr		= attr->max_qp_wr;
411 	resp->device_cap_flags	= attr->device_cap_flags;
412 	resp->max_sge		= attr->max_sge;
413 	resp->max_sge_rd	= attr->max_sge_rd;
414 	resp->max_cq		= attr->max_cq;
415 	resp->max_cqe		= attr->max_cqe;
416 	resp->max_mr		= attr->max_mr;
417 	resp->max_pd		= attr->max_pd;
418 	resp->max_qp_rd_atom	= attr->max_qp_rd_atom;
419 	resp->max_ee_rd_atom	= attr->max_ee_rd_atom;
420 	resp->max_res_rd_atom	= attr->max_res_rd_atom;
421 	resp->max_qp_init_rd_atom	= attr->max_qp_init_rd_atom;
422 	resp->max_ee_init_rd_atom	= attr->max_ee_init_rd_atom;
423 	resp->atomic_cap		= attr->atomic_cap;
424 	resp->max_ee			= attr->max_ee;
425 	resp->max_rdd			= attr->max_rdd;
426 	resp->max_mw			= attr->max_mw;
427 	resp->max_raw_ipv6_qp		= attr->max_raw_ipv6_qp;
428 	resp->max_raw_ethy_qp		= attr->max_raw_ethy_qp;
429 	resp->max_mcast_grp		= attr->max_mcast_grp;
430 	resp->max_mcast_qp_attach	= attr->max_mcast_qp_attach;
431 	resp->max_total_mcast_qp_attach	= attr->max_total_mcast_qp_attach;
432 	resp->max_ah			= attr->max_ah;
433 	resp->max_fmr			= attr->max_fmr;
434 	resp->max_map_per_fmr		= attr->max_map_per_fmr;
435 	resp->max_srq			= attr->max_srq;
436 	resp->max_srq_wr		= attr->max_srq_wr;
437 	resp->max_srq_sge		= attr->max_srq_sge;
438 	resp->max_pkeys			= attr->max_pkeys;
439 	resp->local_ca_ack_delay	= attr->local_ca_ack_delay;
440 	resp->phys_port_cnt		= ib_dev->phys_port_cnt;
441 }
442 
443 ssize_t ib_uverbs_query_device(struct ib_uverbs_file *file,
444 			       struct ib_device *ib_dev,
445 			       const char __user *buf,
446 			       int in_len, int out_len)
447 {
448 	struct ib_uverbs_query_device      cmd;
449 	struct ib_uverbs_query_device_resp resp;
450 	struct ib_device_attr              attr;
451 	int                                ret;
452 
453 	if (out_len < sizeof resp)
454 		return -ENOSPC;
455 
456 	if (copy_from_user(&cmd, buf, sizeof cmd))
457 		return -EFAULT;
458 
459 	ret = ib_query_device(ib_dev, &attr);
460 	if (ret)
461 		return ret;
462 
463 	memset(&resp, 0, sizeof resp);
464 	copy_query_dev_fields(file, ib_dev, &resp, &attr);
465 
466 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
467 			 &resp, sizeof resp))
468 		return -EFAULT;
469 
470 	return in_len;
471 }
472 
473 ssize_t ib_uverbs_query_port(struct ib_uverbs_file *file,
474 			     struct ib_device *ib_dev,
475 			     const char __user *buf,
476 			     int in_len, int out_len)
477 {
478 	struct ib_uverbs_query_port      cmd;
479 	struct ib_uverbs_query_port_resp resp;
480 	struct ib_port_attr              attr;
481 	int                              ret;
482 
483 	if (out_len < sizeof resp)
484 		return -ENOSPC;
485 
486 	if (copy_from_user(&cmd, buf, sizeof cmd))
487 		return -EFAULT;
488 
489 	ret = ib_query_port(ib_dev, cmd.port_num, &attr);
490 	if (ret)
491 		return ret;
492 
493 	memset(&resp, 0, sizeof resp);
494 
495 	resp.state 	     = attr.state;
496 	resp.max_mtu 	     = attr.max_mtu;
497 	resp.active_mtu      = attr.active_mtu;
498 	resp.gid_tbl_len     = attr.gid_tbl_len;
499 	resp.port_cap_flags  = attr.port_cap_flags;
500 	resp.max_msg_sz      = attr.max_msg_sz;
501 	resp.bad_pkey_cntr   = attr.bad_pkey_cntr;
502 	resp.qkey_viol_cntr  = attr.qkey_viol_cntr;
503 	resp.pkey_tbl_len    = attr.pkey_tbl_len;
504 	resp.lid 	     = attr.lid;
505 	resp.sm_lid 	     = attr.sm_lid;
506 	resp.lmc 	     = attr.lmc;
507 	resp.max_vl_num      = attr.max_vl_num;
508 	resp.sm_sl 	     = attr.sm_sl;
509 	resp.subnet_timeout  = attr.subnet_timeout;
510 	resp.init_type_reply = attr.init_type_reply;
511 	resp.active_width    = attr.active_width;
512 	resp.active_speed    = attr.active_speed;
513 	resp.phys_state      = attr.phys_state;
514 	resp.link_layer      = rdma_port_get_link_layer(ib_dev,
515 							cmd.port_num);
516 
517 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
518 			 &resp, sizeof resp))
519 		return -EFAULT;
520 
521 	return in_len;
522 }
523 
524 ssize_t ib_uverbs_alloc_pd(struct ib_uverbs_file *file,
525 			   struct ib_device *ib_dev,
526 			   const char __user *buf,
527 			   int in_len, int out_len)
528 {
529 	struct ib_uverbs_alloc_pd      cmd;
530 	struct ib_uverbs_alloc_pd_resp resp;
531 	struct ib_udata                udata;
532 	struct ib_uobject             *uobj;
533 	struct ib_pd                  *pd;
534 	int                            ret;
535 
536 	if (out_len < sizeof resp)
537 		return -ENOSPC;
538 
539 	if (copy_from_user(&cmd, buf, sizeof cmd))
540 		return -EFAULT;
541 
542 	INIT_UDATA(&udata, buf + sizeof cmd,
543 		   (unsigned long) cmd.response + sizeof resp,
544 		   in_len - sizeof cmd, out_len - sizeof resp);
545 
546 	uobj = kmalloc(sizeof *uobj, GFP_KERNEL);
547 	if (!uobj)
548 		return -ENOMEM;
549 
550 	init_uobj(uobj, 0, file->ucontext, &pd_lock_class);
551 	down_write(&uobj->mutex);
552 
553 	pd = ib_dev->alloc_pd(ib_dev, file->ucontext, &udata);
554 	if (IS_ERR(pd)) {
555 		ret = PTR_ERR(pd);
556 		goto err;
557 	}
558 
559 	pd->device  = ib_dev;
560 	pd->uobject = uobj;
561 	pd->local_mr = NULL;
562 	atomic_set(&pd->usecnt, 0);
563 
564 	uobj->object = pd;
565 	ret = idr_add_uobj(&ib_uverbs_pd_idr, uobj);
566 	if (ret)
567 		goto err_idr;
568 
569 	memset(&resp, 0, sizeof resp);
570 	resp.pd_handle = uobj->id;
571 
572 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
573 			 &resp, sizeof resp)) {
574 		ret = -EFAULT;
575 		goto err_copy;
576 	}
577 
578 	mutex_lock(&file->mutex);
579 	list_add_tail(&uobj->list, &file->ucontext->pd_list);
580 	mutex_unlock(&file->mutex);
581 
582 	uobj->live = 1;
583 
584 	up_write(&uobj->mutex);
585 
586 	return in_len;
587 
588 err_copy:
589 	idr_remove_uobj(&ib_uverbs_pd_idr, uobj);
590 
591 err_idr:
592 	ib_dealloc_pd(pd);
593 
594 err:
595 	put_uobj_write(uobj);
596 	return ret;
597 }
598 
599 ssize_t ib_uverbs_dealloc_pd(struct ib_uverbs_file *file,
600 			     struct ib_device *ib_dev,
601 			     const char __user *buf,
602 			     int in_len, int out_len)
603 {
604 	struct ib_uverbs_dealloc_pd cmd;
605 	struct ib_uobject          *uobj;
606 	struct ib_pd		   *pd;
607 	int                         ret;
608 
609 	if (copy_from_user(&cmd, buf, sizeof cmd))
610 		return -EFAULT;
611 
612 	uobj = idr_write_uobj(&ib_uverbs_pd_idr, cmd.pd_handle, file->ucontext);
613 	if (!uobj)
614 		return -EINVAL;
615 	pd = uobj->object;
616 
617 	if (atomic_read(&pd->usecnt)) {
618 		ret = -EBUSY;
619 		goto err_put;
620 	}
621 
622 	ret = pd->device->dealloc_pd(uobj->object);
623 	WARN_ONCE(ret, "Infiniband HW driver failed dealloc_pd");
624 	if (ret)
625 		goto err_put;
626 
627 	uobj->live = 0;
628 	put_uobj_write(uobj);
629 
630 	idr_remove_uobj(&ib_uverbs_pd_idr, uobj);
631 
632 	mutex_lock(&file->mutex);
633 	list_del(&uobj->list);
634 	mutex_unlock(&file->mutex);
635 
636 	put_uobj(uobj);
637 
638 	return in_len;
639 
640 err_put:
641 	put_uobj_write(uobj);
642 	return ret;
643 }
644 
645 struct xrcd_table_entry {
646 	struct rb_node  node;
647 	struct ib_xrcd *xrcd;
648 	struct inode   *inode;
649 };
650 
651 static int xrcd_table_insert(struct ib_uverbs_device *dev,
652 			    struct inode *inode,
653 			    struct ib_xrcd *xrcd)
654 {
655 	struct xrcd_table_entry *entry, *scan;
656 	struct rb_node **p = &dev->xrcd_tree.rb_node;
657 	struct rb_node *parent = NULL;
658 
659 	entry = kmalloc(sizeof *entry, GFP_KERNEL);
660 	if (!entry)
661 		return -ENOMEM;
662 
663 	entry->xrcd  = xrcd;
664 	entry->inode = inode;
665 
666 	while (*p) {
667 		parent = *p;
668 		scan = rb_entry(parent, struct xrcd_table_entry, node);
669 
670 		if (inode < scan->inode) {
671 			p = &(*p)->rb_left;
672 		} else if (inode > scan->inode) {
673 			p = &(*p)->rb_right;
674 		} else {
675 			kfree(entry);
676 			return -EEXIST;
677 		}
678 	}
679 
680 	rb_link_node(&entry->node, parent, p);
681 	rb_insert_color(&entry->node, &dev->xrcd_tree);
682 	igrab(inode);
683 	return 0;
684 }
685 
686 static struct xrcd_table_entry *xrcd_table_search(struct ib_uverbs_device *dev,
687 						  struct inode *inode)
688 {
689 	struct xrcd_table_entry *entry;
690 	struct rb_node *p = dev->xrcd_tree.rb_node;
691 
692 	while (p) {
693 		entry = rb_entry(p, struct xrcd_table_entry, node);
694 
695 		if (inode < entry->inode)
696 			p = p->rb_left;
697 		else if (inode > entry->inode)
698 			p = p->rb_right;
699 		else
700 			return entry;
701 	}
702 
703 	return NULL;
704 }
705 
706 static struct ib_xrcd *find_xrcd(struct ib_uverbs_device *dev, struct inode *inode)
707 {
708 	struct xrcd_table_entry *entry;
709 
710 	entry = xrcd_table_search(dev, inode);
711 	if (!entry)
712 		return NULL;
713 
714 	return entry->xrcd;
715 }
716 
717 static void xrcd_table_delete(struct ib_uverbs_device *dev,
718 			      struct inode *inode)
719 {
720 	struct xrcd_table_entry *entry;
721 
722 	entry = xrcd_table_search(dev, inode);
723 	if (entry) {
724 		iput(inode);
725 		rb_erase(&entry->node, &dev->xrcd_tree);
726 		kfree(entry);
727 	}
728 }
729 
730 ssize_t ib_uverbs_open_xrcd(struct ib_uverbs_file *file,
731 			    struct ib_device *ib_dev,
732 			    const char __user *buf, int in_len,
733 			    int out_len)
734 {
735 	struct ib_uverbs_open_xrcd	cmd;
736 	struct ib_uverbs_open_xrcd_resp	resp;
737 	struct ib_udata			udata;
738 	struct ib_uxrcd_object         *obj;
739 	struct ib_xrcd                 *xrcd = NULL;
740 	struct fd			f = {NULL, 0};
741 	struct inode                   *inode = NULL;
742 	int				ret = 0;
743 	int				new_xrcd = 0;
744 
745 	if (out_len < sizeof resp)
746 		return -ENOSPC;
747 
748 	if (copy_from_user(&cmd, buf, sizeof cmd))
749 		return -EFAULT;
750 
751 	INIT_UDATA(&udata, buf + sizeof cmd,
752 		   (unsigned long) cmd.response + sizeof resp,
753 		   in_len - sizeof cmd, out_len - sizeof  resp);
754 
755 	mutex_lock(&file->device->xrcd_tree_mutex);
756 
757 	if (cmd.fd != -1) {
758 		/* search for file descriptor */
759 		f = fdget(cmd.fd);
760 		if (!f.file) {
761 			ret = -EBADF;
762 			goto err_tree_mutex_unlock;
763 		}
764 
765 		inode = file_inode(f.file);
766 		xrcd = find_xrcd(file->device, inode);
767 		if (!xrcd && !(cmd.oflags & O_CREAT)) {
768 			/* no file descriptor. Need CREATE flag */
769 			ret = -EAGAIN;
770 			goto err_tree_mutex_unlock;
771 		}
772 
773 		if (xrcd && cmd.oflags & O_EXCL) {
774 			ret = -EINVAL;
775 			goto err_tree_mutex_unlock;
776 		}
777 	}
778 
779 	obj = kmalloc(sizeof *obj, GFP_KERNEL);
780 	if (!obj) {
781 		ret = -ENOMEM;
782 		goto err_tree_mutex_unlock;
783 	}
784 
785 	init_uobj(&obj->uobject, 0, file->ucontext, &xrcd_lock_class);
786 
787 	down_write(&obj->uobject.mutex);
788 
789 	if (!xrcd) {
790 		xrcd = ib_dev->alloc_xrcd(ib_dev, file->ucontext, &udata);
791 		if (IS_ERR(xrcd)) {
792 			ret = PTR_ERR(xrcd);
793 			goto err;
794 		}
795 
796 		xrcd->inode   = inode;
797 		xrcd->device  = ib_dev;
798 		atomic_set(&xrcd->usecnt, 0);
799 		mutex_init(&xrcd->tgt_qp_mutex);
800 		INIT_LIST_HEAD(&xrcd->tgt_qp_list);
801 		new_xrcd = 1;
802 	}
803 
804 	atomic_set(&obj->refcnt, 0);
805 	obj->uobject.object = xrcd;
806 	ret = idr_add_uobj(&ib_uverbs_xrcd_idr, &obj->uobject);
807 	if (ret)
808 		goto err_idr;
809 
810 	memset(&resp, 0, sizeof resp);
811 	resp.xrcd_handle = obj->uobject.id;
812 
813 	if (inode) {
814 		if (new_xrcd) {
815 			/* create new inode/xrcd table entry */
816 			ret = xrcd_table_insert(file->device, inode, xrcd);
817 			if (ret)
818 				goto err_insert_xrcd;
819 		}
820 		atomic_inc(&xrcd->usecnt);
821 	}
822 
823 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
824 			 &resp, sizeof resp)) {
825 		ret = -EFAULT;
826 		goto err_copy;
827 	}
828 
829 	if (f.file)
830 		fdput(f);
831 
832 	mutex_lock(&file->mutex);
833 	list_add_tail(&obj->uobject.list, &file->ucontext->xrcd_list);
834 	mutex_unlock(&file->mutex);
835 
836 	obj->uobject.live = 1;
837 	up_write(&obj->uobject.mutex);
838 
839 	mutex_unlock(&file->device->xrcd_tree_mutex);
840 	return in_len;
841 
842 err_copy:
843 	if (inode) {
844 		if (new_xrcd)
845 			xrcd_table_delete(file->device, inode);
846 		atomic_dec(&xrcd->usecnt);
847 	}
848 
849 err_insert_xrcd:
850 	idr_remove_uobj(&ib_uverbs_xrcd_idr, &obj->uobject);
851 
852 err_idr:
853 	ib_dealloc_xrcd(xrcd);
854 
855 err:
856 	put_uobj_write(&obj->uobject);
857 
858 err_tree_mutex_unlock:
859 	if (f.file)
860 		fdput(f);
861 
862 	mutex_unlock(&file->device->xrcd_tree_mutex);
863 
864 	return ret;
865 }
866 
867 ssize_t ib_uverbs_close_xrcd(struct ib_uverbs_file *file,
868 			     struct ib_device *ib_dev,
869 			     const char __user *buf, int in_len,
870 			     int out_len)
871 {
872 	struct ib_uverbs_close_xrcd cmd;
873 	struct ib_uobject           *uobj;
874 	struct ib_xrcd              *xrcd = NULL;
875 	struct inode                *inode = NULL;
876 	struct ib_uxrcd_object      *obj;
877 	int                         live;
878 	int                         ret = 0;
879 
880 	if (copy_from_user(&cmd, buf, sizeof cmd))
881 		return -EFAULT;
882 
883 	mutex_lock(&file->device->xrcd_tree_mutex);
884 	uobj = idr_write_uobj(&ib_uverbs_xrcd_idr, cmd.xrcd_handle, file->ucontext);
885 	if (!uobj) {
886 		ret = -EINVAL;
887 		goto out;
888 	}
889 
890 	xrcd  = uobj->object;
891 	inode = xrcd->inode;
892 	obj   = container_of(uobj, struct ib_uxrcd_object, uobject);
893 	if (atomic_read(&obj->refcnt)) {
894 		put_uobj_write(uobj);
895 		ret = -EBUSY;
896 		goto out;
897 	}
898 
899 	if (!inode || atomic_dec_and_test(&xrcd->usecnt)) {
900 		ret = ib_dealloc_xrcd(uobj->object);
901 		if (!ret)
902 			uobj->live = 0;
903 	}
904 
905 	live = uobj->live;
906 	if (inode && ret)
907 		atomic_inc(&xrcd->usecnt);
908 
909 	put_uobj_write(uobj);
910 
911 	if (ret)
912 		goto out;
913 
914 	if (inode && !live)
915 		xrcd_table_delete(file->device, inode);
916 
917 	idr_remove_uobj(&ib_uverbs_xrcd_idr, uobj);
918 	mutex_lock(&file->mutex);
919 	list_del(&uobj->list);
920 	mutex_unlock(&file->mutex);
921 
922 	put_uobj(uobj);
923 	ret = in_len;
924 
925 out:
926 	mutex_unlock(&file->device->xrcd_tree_mutex);
927 	return ret;
928 }
929 
930 void ib_uverbs_dealloc_xrcd(struct ib_uverbs_device *dev,
931 			    struct ib_xrcd *xrcd)
932 {
933 	struct inode *inode;
934 
935 	inode = xrcd->inode;
936 	if (inode && !atomic_dec_and_test(&xrcd->usecnt))
937 		return;
938 
939 	ib_dealloc_xrcd(xrcd);
940 
941 	if (inode)
942 		xrcd_table_delete(dev, inode);
943 }
944 
945 ssize_t ib_uverbs_reg_mr(struct ib_uverbs_file *file,
946 			 struct ib_device *ib_dev,
947 			 const char __user *buf, int in_len,
948 			 int out_len)
949 {
950 	struct ib_uverbs_reg_mr      cmd;
951 	struct ib_uverbs_reg_mr_resp resp;
952 	struct ib_udata              udata;
953 	struct ib_uobject           *uobj;
954 	struct ib_pd                *pd;
955 	struct ib_mr                *mr;
956 	int                          ret;
957 
958 	if (out_len < sizeof resp)
959 		return -ENOSPC;
960 
961 	if (copy_from_user(&cmd, buf, sizeof cmd))
962 		return -EFAULT;
963 
964 	INIT_UDATA(&udata, buf + sizeof cmd,
965 		   (unsigned long) cmd.response + sizeof resp,
966 		   in_len - sizeof cmd, out_len - sizeof resp);
967 
968 	if ((cmd.start & ~PAGE_MASK) != (cmd.hca_va & ~PAGE_MASK))
969 		return -EINVAL;
970 
971 	ret = ib_check_mr_access(cmd.access_flags);
972 	if (ret)
973 		return ret;
974 
975 	uobj = kmalloc(sizeof *uobj, GFP_KERNEL);
976 	if (!uobj)
977 		return -ENOMEM;
978 
979 	init_uobj(uobj, 0, file->ucontext, &mr_lock_class);
980 	down_write(&uobj->mutex);
981 
982 	pd = idr_read_pd(cmd.pd_handle, file->ucontext);
983 	if (!pd) {
984 		ret = -EINVAL;
985 		goto err_free;
986 	}
987 
988 	if (cmd.access_flags & IB_ACCESS_ON_DEMAND) {
989 		struct ib_device_attr attr;
990 
991 		ret = ib_query_device(pd->device, &attr);
992 		if (ret || !(attr.device_cap_flags &
993 				IB_DEVICE_ON_DEMAND_PAGING)) {
994 			pr_debug("ODP support not available\n");
995 			ret = -EINVAL;
996 			goto err_put;
997 		}
998 	}
999 
1000 	mr = pd->device->reg_user_mr(pd, cmd.start, cmd.length, cmd.hca_va,
1001 				     cmd.access_flags, &udata);
1002 	if (IS_ERR(mr)) {
1003 		ret = PTR_ERR(mr);
1004 		goto err_put;
1005 	}
1006 
1007 	mr->device  = pd->device;
1008 	mr->pd      = pd;
1009 	mr->uobject = uobj;
1010 	atomic_inc(&pd->usecnt);
1011 	atomic_set(&mr->usecnt, 0);
1012 
1013 	uobj->object = mr;
1014 	ret = idr_add_uobj(&ib_uverbs_mr_idr, uobj);
1015 	if (ret)
1016 		goto err_unreg;
1017 
1018 	memset(&resp, 0, sizeof resp);
1019 	resp.lkey      = mr->lkey;
1020 	resp.rkey      = mr->rkey;
1021 	resp.mr_handle = uobj->id;
1022 
1023 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
1024 			 &resp, sizeof resp)) {
1025 		ret = -EFAULT;
1026 		goto err_copy;
1027 	}
1028 
1029 	put_pd_read(pd);
1030 
1031 	mutex_lock(&file->mutex);
1032 	list_add_tail(&uobj->list, &file->ucontext->mr_list);
1033 	mutex_unlock(&file->mutex);
1034 
1035 	uobj->live = 1;
1036 
1037 	up_write(&uobj->mutex);
1038 
1039 	return in_len;
1040 
1041 err_copy:
1042 	idr_remove_uobj(&ib_uverbs_mr_idr, uobj);
1043 
1044 err_unreg:
1045 	ib_dereg_mr(mr);
1046 
1047 err_put:
1048 	put_pd_read(pd);
1049 
1050 err_free:
1051 	put_uobj_write(uobj);
1052 	return ret;
1053 }
1054 
1055 ssize_t ib_uverbs_rereg_mr(struct ib_uverbs_file *file,
1056 			   struct ib_device *ib_dev,
1057 			   const char __user *buf, int in_len,
1058 			   int out_len)
1059 {
1060 	struct ib_uverbs_rereg_mr      cmd;
1061 	struct ib_uverbs_rereg_mr_resp resp;
1062 	struct ib_udata              udata;
1063 	struct ib_pd                *pd = NULL;
1064 	struct ib_mr                *mr;
1065 	struct ib_pd		    *old_pd;
1066 	int                          ret;
1067 	struct ib_uobject	    *uobj;
1068 
1069 	if (out_len < sizeof(resp))
1070 		return -ENOSPC;
1071 
1072 	if (copy_from_user(&cmd, buf, sizeof(cmd)))
1073 		return -EFAULT;
1074 
1075 	INIT_UDATA(&udata, buf + sizeof(cmd),
1076 		   (unsigned long) cmd.response + sizeof(resp),
1077 		   in_len - sizeof(cmd), out_len - sizeof(resp));
1078 
1079 	if (cmd.flags & ~IB_MR_REREG_SUPPORTED || !cmd.flags)
1080 		return -EINVAL;
1081 
1082 	if ((cmd.flags & IB_MR_REREG_TRANS) &&
1083 	    (!cmd.start || !cmd.hca_va || 0 >= cmd.length ||
1084 	     (cmd.start & ~PAGE_MASK) != (cmd.hca_va & ~PAGE_MASK)))
1085 			return -EINVAL;
1086 
1087 	uobj = idr_write_uobj(&ib_uverbs_mr_idr, cmd.mr_handle,
1088 			      file->ucontext);
1089 
1090 	if (!uobj)
1091 		return -EINVAL;
1092 
1093 	mr = uobj->object;
1094 
1095 	if (cmd.flags & IB_MR_REREG_ACCESS) {
1096 		ret = ib_check_mr_access(cmd.access_flags);
1097 		if (ret)
1098 			goto put_uobjs;
1099 	}
1100 
1101 	if (cmd.flags & IB_MR_REREG_PD) {
1102 		pd = idr_read_pd(cmd.pd_handle, file->ucontext);
1103 		if (!pd) {
1104 			ret = -EINVAL;
1105 			goto put_uobjs;
1106 		}
1107 	}
1108 
1109 	if (atomic_read(&mr->usecnt)) {
1110 		ret = -EBUSY;
1111 		goto put_uobj_pd;
1112 	}
1113 
1114 	old_pd = mr->pd;
1115 	ret = mr->device->rereg_user_mr(mr, cmd.flags, cmd.start,
1116 					cmd.length, cmd.hca_va,
1117 					cmd.access_flags, pd, &udata);
1118 	if (!ret) {
1119 		if (cmd.flags & IB_MR_REREG_PD) {
1120 			atomic_inc(&pd->usecnt);
1121 			mr->pd = pd;
1122 			atomic_dec(&old_pd->usecnt);
1123 		}
1124 	} else {
1125 		goto put_uobj_pd;
1126 	}
1127 
1128 	memset(&resp, 0, sizeof(resp));
1129 	resp.lkey      = mr->lkey;
1130 	resp.rkey      = mr->rkey;
1131 
1132 	if (copy_to_user((void __user *)(unsigned long)cmd.response,
1133 			 &resp, sizeof(resp)))
1134 		ret = -EFAULT;
1135 	else
1136 		ret = in_len;
1137 
1138 put_uobj_pd:
1139 	if (cmd.flags & IB_MR_REREG_PD)
1140 		put_pd_read(pd);
1141 
1142 put_uobjs:
1143 
1144 	put_uobj_write(mr->uobject);
1145 
1146 	return ret;
1147 }
1148 
1149 ssize_t ib_uverbs_dereg_mr(struct ib_uverbs_file *file,
1150 			   struct ib_device *ib_dev,
1151 			   const char __user *buf, int in_len,
1152 			   int out_len)
1153 {
1154 	struct ib_uverbs_dereg_mr cmd;
1155 	struct ib_mr             *mr;
1156 	struct ib_uobject	 *uobj;
1157 	int                       ret = -EINVAL;
1158 
1159 	if (copy_from_user(&cmd, buf, sizeof cmd))
1160 		return -EFAULT;
1161 
1162 	uobj = idr_write_uobj(&ib_uverbs_mr_idr, cmd.mr_handle, file->ucontext);
1163 	if (!uobj)
1164 		return -EINVAL;
1165 
1166 	mr = uobj->object;
1167 
1168 	ret = ib_dereg_mr(mr);
1169 	if (!ret)
1170 		uobj->live = 0;
1171 
1172 	put_uobj_write(uobj);
1173 
1174 	if (ret)
1175 		return ret;
1176 
1177 	idr_remove_uobj(&ib_uverbs_mr_idr, uobj);
1178 
1179 	mutex_lock(&file->mutex);
1180 	list_del(&uobj->list);
1181 	mutex_unlock(&file->mutex);
1182 
1183 	put_uobj(uobj);
1184 
1185 	return in_len;
1186 }
1187 
1188 ssize_t ib_uverbs_alloc_mw(struct ib_uverbs_file *file,
1189 			   struct ib_device *ib_dev,
1190 			   const char __user *buf, int in_len,
1191 			   int out_len)
1192 {
1193 	struct ib_uverbs_alloc_mw      cmd;
1194 	struct ib_uverbs_alloc_mw_resp resp;
1195 	struct ib_uobject             *uobj;
1196 	struct ib_pd                  *pd;
1197 	struct ib_mw                  *mw;
1198 	int                            ret;
1199 
1200 	if (out_len < sizeof(resp))
1201 		return -ENOSPC;
1202 
1203 	if (copy_from_user(&cmd, buf, sizeof(cmd)))
1204 		return -EFAULT;
1205 
1206 	uobj = kmalloc(sizeof(*uobj), GFP_KERNEL);
1207 	if (!uobj)
1208 		return -ENOMEM;
1209 
1210 	init_uobj(uobj, 0, file->ucontext, &mw_lock_class);
1211 	down_write(&uobj->mutex);
1212 
1213 	pd = idr_read_pd(cmd.pd_handle, file->ucontext);
1214 	if (!pd) {
1215 		ret = -EINVAL;
1216 		goto err_free;
1217 	}
1218 
1219 	mw = pd->device->alloc_mw(pd, cmd.mw_type);
1220 	if (IS_ERR(mw)) {
1221 		ret = PTR_ERR(mw);
1222 		goto err_put;
1223 	}
1224 
1225 	mw->device  = pd->device;
1226 	mw->pd      = pd;
1227 	mw->uobject = uobj;
1228 	atomic_inc(&pd->usecnt);
1229 
1230 	uobj->object = mw;
1231 	ret = idr_add_uobj(&ib_uverbs_mw_idr, uobj);
1232 	if (ret)
1233 		goto err_unalloc;
1234 
1235 	memset(&resp, 0, sizeof(resp));
1236 	resp.rkey      = mw->rkey;
1237 	resp.mw_handle = uobj->id;
1238 
1239 	if (copy_to_user((void __user *)(unsigned long)cmd.response,
1240 			 &resp, sizeof(resp))) {
1241 		ret = -EFAULT;
1242 		goto err_copy;
1243 	}
1244 
1245 	put_pd_read(pd);
1246 
1247 	mutex_lock(&file->mutex);
1248 	list_add_tail(&uobj->list, &file->ucontext->mw_list);
1249 	mutex_unlock(&file->mutex);
1250 
1251 	uobj->live = 1;
1252 
1253 	up_write(&uobj->mutex);
1254 
1255 	return in_len;
1256 
1257 err_copy:
1258 	idr_remove_uobj(&ib_uverbs_mw_idr, uobj);
1259 
1260 err_unalloc:
1261 	ib_dealloc_mw(mw);
1262 
1263 err_put:
1264 	put_pd_read(pd);
1265 
1266 err_free:
1267 	put_uobj_write(uobj);
1268 	return ret;
1269 }
1270 
1271 ssize_t ib_uverbs_dealloc_mw(struct ib_uverbs_file *file,
1272 			     struct ib_device *ib_dev,
1273 			     const char __user *buf, int in_len,
1274 			     int out_len)
1275 {
1276 	struct ib_uverbs_dealloc_mw cmd;
1277 	struct ib_mw               *mw;
1278 	struct ib_uobject	   *uobj;
1279 	int                         ret = -EINVAL;
1280 
1281 	if (copy_from_user(&cmd, buf, sizeof(cmd)))
1282 		return -EFAULT;
1283 
1284 	uobj = idr_write_uobj(&ib_uverbs_mw_idr, cmd.mw_handle, file->ucontext);
1285 	if (!uobj)
1286 		return -EINVAL;
1287 
1288 	mw = uobj->object;
1289 
1290 	ret = ib_dealloc_mw(mw);
1291 	if (!ret)
1292 		uobj->live = 0;
1293 
1294 	put_uobj_write(uobj);
1295 
1296 	if (ret)
1297 		return ret;
1298 
1299 	idr_remove_uobj(&ib_uverbs_mw_idr, uobj);
1300 
1301 	mutex_lock(&file->mutex);
1302 	list_del(&uobj->list);
1303 	mutex_unlock(&file->mutex);
1304 
1305 	put_uobj(uobj);
1306 
1307 	return in_len;
1308 }
1309 
1310 ssize_t ib_uverbs_create_comp_channel(struct ib_uverbs_file *file,
1311 				      struct ib_device *ib_dev,
1312 				      const char __user *buf, int in_len,
1313 				      int out_len)
1314 {
1315 	struct ib_uverbs_create_comp_channel	   cmd;
1316 	struct ib_uverbs_create_comp_channel_resp  resp;
1317 	struct file				  *filp;
1318 	int ret;
1319 
1320 	if (out_len < sizeof resp)
1321 		return -ENOSPC;
1322 
1323 	if (copy_from_user(&cmd, buf, sizeof cmd))
1324 		return -EFAULT;
1325 
1326 	ret = get_unused_fd_flags(O_CLOEXEC);
1327 	if (ret < 0)
1328 		return ret;
1329 	resp.fd = ret;
1330 
1331 	filp = ib_uverbs_alloc_event_file(file, ib_dev, 0);
1332 	if (IS_ERR(filp)) {
1333 		put_unused_fd(resp.fd);
1334 		return PTR_ERR(filp);
1335 	}
1336 
1337 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
1338 			 &resp, sizeof resp)) {
1339 		put_unused_fd(resp.fd);
1340 		fput(filp);
1341 		return -EFAULT;
1342 	}
1343 
1344 	fd_install(resp.fd, filp);
1345 	return in_len;
1346 }
1347 
1348 static struct ib_ucq_object *create_cq(struct ib_uverbs_file *file,
1349 					struct ib_device *ib_dev,
1350 				       struct ib_udata *ucore,
1351 				       struct ib_udata *uhw,
1352 				       struct ib_uverbs_ex_create_cq *cmd,
1353 				       size_t cmd_sz,
1354 				       int (*cb)(struct ib_uverbs_file *file,
1355 						 struct ib_ucq_object *obj,
1356 						 struct ib_uverbs_ex_create_cq_resp *resp,
1357 						 struct ib_udata *udata,
1358 						 void *context),
1359 				       void *context)
1360 {
1361 	struct ib_ucq_object           *obj;
1362 	struct ib_uverbs_event_file    *ev_file = NULL;
1363 	struct ib_cq                   *cq;
1364 	int                             ret;
1365 	struct ib_uverbs_ex_create_cq_resp resp;
1366 	struct ib_cq_init_attr attr = {};
1367 
1368 	if (cmd->comp_vector >= file->device->num_comp_vectors)
1369 		return ERR_PTR(-EINVAL);
1370 
1371 	obj = kmalloc(sizeof *obj, GFP_KERNEL);
1372 	if (!obj)
1373 		return ERR_PTR(-ENOMEM);
1374 
1375 	init_uobj(&obj->uobject, cmd->user_handle, file->ucontext, &cq_lock_class);
1376 	down_write(&obj->uobject.mutex);
1377 
1378 	if (cmd->comp_channel >= 0) {
1379 		ev_file = ib_uverbs_lookup_comp_file(cmd->comp_channel);
1380 		if (!ev_file) {
1381 			ret = -EINVAL;
1382 			goto err;
1383 		}
1384 	}
1385 
1386 	obj->uverbs_file	   = file;
1387 	obj->comp_events_reported  = 0;
1388 	obj->async_events_reported = 0;
1389 	INIT_LIST_HEAD(&obj->comp_list);
1390 	INIT_LIST_HEAD(&obj->async_list);
1391 
1392 	attr.cqe = cmd->cqe;
1393 	attr.comp_vector = cmd->comp_vector;
1394 
1395 	if (cmd_sz > offsetof(typeof(*cmd), flags) + sizeof(cmd->flags))
1396 		attr.flags = cmd->flags;
1397 
1398 	cq = ib_dev->create_cq(ib_dev, &attr,
1399 					     file->ucontext, uhw);
1400 	if (IS_ERR(cq)) {
1401 		ret = PTR_ERR(cq);
1402 		goto err_file;
1403 	}
1404 
1405 	cq->device        = ib_dev;
1406 	cq->uobject       = &obj->uobject;
1407 	cq->comp_handler  = ib_uverbs_comp_handler;
1408 	cq->event_handler = ib_uverbs_cq_event_handler;
1409 	cq->cq_context    = ev_file;
1410 	atomic_set(&cq->usecnt, 0);
1411 
1412 	obj->uobject.object = cq;
1413 	ret = idr_add_uobj(&ib_uverbs_cq_idr, &obj->uobject);
1414 	if (ret)
1415 		goto err_free;
1416 
1417 	memset(&resp, 0, sizeof resp);
1418 	resp.base.cq_handle = obj->uobject.id;
1419 	resp.base.cqe       = cq->cqe;
1420 
1421 	resp.response_length = offsetof(typeof(resp), response_length) +
1422 		sizeof(resp.response_length);
1423 
1424 	ret = cb(file, obj, &resp, ucore, context);
1425 	if (ret)
1426 		goto err_cb;
1427 
1428 	mutex_lock(&file->mutex);
1429 	list_add_tail(&obj->uobject.list, &file->ucontext->cq_list);
1430 	mutex_unlock(&file->mutex);
1431 
1432 	obj->uobject.live = 1;
1433 
1434 	up_write(&obj->uobject.mutex);
1435 
1436 	return obj;
1437 
1438 err_cb:
1439 	idr_remove_uobj(&ib_uverbs_cq_idr, &obj->uobject);
1440 
1441 err_free:
1442 	ib_destroy_cq(cq);
1443 
1444 err_file:
1445 	if (ev_file)
1446 		ib_uverbs_release_ucq(file, ev_file, obj);
1447 
1448 err:
1449 	put_uobj_write(&obj->uobject);
1450 
1451 	return ERR_PTR(ret);
1452 }
1453 
1454 static int ib_uverbs_create_cq_cb(struct ib_uverbs_file *file,
1455 				  struct ib_ucq_object *obj,
1456 				  struct ib_uverbs_ex_create_cq_resp *resp,
1457 				  struct ib_udata *ucore, void *context)
1458 {
1459 	if (ib_copy_to_udata(ucore, &resp->base, sizeof(resp->base)))
1460 		return -EFAULT;
1461 
1462 	return 0;
1463 }
1464 
1465 ssize_t ib_uverbs_create_cq(struct ib_uverbs_file *file,
1466 			    struct ib_device *ib_dev,
1467 			    const char __user *buf, int in_len,
1468 			    int out_len)
1469 {
1470 	struct ib_uverbs_create_cq      cmd;
1471 	struct ib_uverbs_ex_create_cq	cmd_ex;
1472 	struct ib_uverbs_create_cq_resp resp;
1473 	struct ib_udata                 ucore;
1474 	struct ib_udata                 uhw;
1475 	struct ib_ucq_object           *obj;
1476 
1477 	if (out_len < sizeof(resp))
1478 		return -ENOSPC;
1479 
1480 	if (copy_from_user(&cmd, buf, sizeof(cmd)))
1481 		return -EFAULT;
1482 
1483 	INIT_UDATA(&ucore, buf, (unsigned long)cmd.response, sizeof(cmd), sizeof(resp));
1484 
1485 	INIT_UDATA(&uhw, buf + sizeof(cmd),
1486 		   (unsigned long)cmd.response + sizeof(resp),
1487 		   in_len - sizeof(cmd), out_len - sizeof(resp));
1488 
1489 	memset(&cmd_ex, 0, sizeof(cmd_ex));
1490 	cmd_ex.user_handle = cmd.user_handle;
1491 	cmd_ex.cqe = cmd.cqe;
1492 	cmd_ex.comp_vector = cmd.comp_vector;
1493 	cmd_ex.comp_channel = cmd.comp_channel;
1494 
1495 	obj = create_cq(file, ib_dev, &ucore, &uhw, &cmd_ex,
1496 			offsetof(typeof(cmd_ex), comp_channel) +
1497 			sizeof(cmd.comp_channel), ib_uverbs_create_cq_cb,
1498 			NULL);
1499 
1500 	if (IS_ERR(obj))
1501 		return PTR_ERR(obj);
1502 
1503 	return in_len;
1504 }
1505 
1506 static int ib_uverbs_ex_create_cq_cb(struct ib_uverbs_file *file,
1507 				     struct ib_ucq_object *obj,
1508 				     struct ib_uverbs_ex_create_cq_resp *resp,
1509 				     struct ib_udata *ucore, void *context)
1510 {
1511 	if (ib_copy_to_udata(ucore, resp, resp->response_length))
1512 		return -EFAULT;
1513 
1514 	return 0;
1515 }
1516 
1517 int ib_uverbs_ex_create_cq(struct ib_uverbs_file *file,
1518 			 struct ib_device *ib_dev,
1519 			   struct ib_udata *ucore,
1520 			   struct ib_udata *uhw)
1521 {
1522 	struct ib_uverbs_ex_create_cq_resp resp;
1523 	struct ib_uverbs_ex_create_cq  cmd;
1524 	struct ib_ucq_object           *obj;
1525 	int err;
1526 
1527 	if (ucore->inlen < sizeof(cmd))
1528 		return -EINVAL;
1529 
1530 	err = ib_copy_from_udata(&cmd, ucore, sizeof(cmd));
1531 	if (err)
1532 		return err;
1533 
1534 	if (cmd.comp_mask)
1535 		return -EINVAL;
1536 
1537 	if (cmd.reserved)
1538 		return -EINVAL;
1539 
1540 	if (ucore->outlen < (offsetof(typeof(resp), response_length) +
1541 			     sizeof(resp.response_length)))
1542 		return -ENOSPC;
1543 
1544 	obj = create_cq(file, ib_dev, ucore, uhw, &cmd,
1545 			min(ucore->inlen, sizeof(cmd)),
1546 			ib_uverbs_ex_create_cq_cb, NULL);
1547 
1548 	if (IS_ERR(obj))
1549 		return PTR_ERR(obj);
1550 
1551 	return 0;
1552 }
1553 
1554 ssize_t ib_uverbs_resize_cq(struct ib_uverbs_file *file,
1555 			    struct ib_device *ib_dev,
1556 			    const char __user *buf, int in_len,
1557 			    int out_len)
1558 {
1559 	struct ib_uverbs_resize_cq	cmd;
1560 	struct ib_uverbs_resize_cq_resp	resp;
1561 	struct ib_udata                 udata;
1562 	struct ib_cq			*cq;
1563 	int				ret = -EINVAL;
1564 
1565 	if (copy_from_user(&cmd, buf, sizeof cmd))
1566 		return -EFAULT;
1567 
1568 	INIT_UDATA(&udata, buf + sizeof cmd,
1569 		   (unsigned long) cmd.response + sizeof resp,
1570 		   in_len - sizeof cmd, out_len - sizeof resp);
1571 
1572 	cq = idr_read_cq(cmd.cq_handle, file->ucontext, 0);
1573 	if (!cq)
1574 		return -EINVAL;
1575 
1576 	ret = cq->device->resize_cq(cq, cmd.cqe, &udata);
1577 	if (ret)
1578 		goto out;
1579 
1580 	resp.cqe = cq->cqe;
1581 
1582 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
1583 			 &resp, sizeof resp.cqe))
1584 		ret = -EFAULT;
1585 
1586 out:
1587 	put_cq_read(cq);
1588 
1589 	return ret ? ret : in_len;
1590 }
1591 
1592 static int copy_wc_to_user(void __user *dest, struct ib_wc *wc)
1593 {
1594 	struct ib_uverbs_wc tmp;
1595 
1596 	tmp.wr_id		= wc->wr_id;
1597 	tmp.status		= wc->status;
1598 	tmp.opcode		= wc->opcode;
1599 	tmp.vendor_err		= wc->vendor_err;
1600 	tmp.byte_len		= wc->byte_len;
1601 	tmp.ex.imm_data		= (__u32 __force) wc->ex.imm_data;
1602 	tmp.qp_num		= wc->qp->qp_num;
1603 	tmp.src_qp		= wc->src_qp;
1604 	tmp.wc_flags		= wc->wc_flags;
1605 	tmp.pkey_index		= wc->pkey_index;
1606 	tmp.slid		= wc->slid;
1607 	tmp.sl			= wc->sl;
1608 	tmp.dlid_path_bits	= wc->dlid_path_bits;
1609 	tmp.port_num		= wc->port_num;
1610 	tmp.reserved		= 0;
1611 
1612 	if (copy_to_user(dest, &tmp, sizeof tmp))
1613 		return -EFAULT;
1614 
1615 	return 0;
1616 }
1617 
1618 ssize_t ib_uverbs_poll_cq(struct ib_uverbs_file *file,
1619 			  struct ib_device *ib_dev,
1620 			  const char __user *buf, int in_len,
1621 			  int out_len)
1622 {
1623 	struct ib_uverbs_poll_cq       cmd;
1624 	struct ib_uverbs_poll_cq_resp  resp;
1625 	u8 __user                     *header_ptr;
1626 	u8 __user                     *data_ptr;
1627 	struct ib_cq                  *cq;
1628 	struct ib_wc                   wc;
1629 	int                            ret;
1630 
1631 	if (copy_from_user(&cmd, buf, sizeof cmd))
1632 		return -EFAULT;
1633 
1634 	cq = idr_read_cq(cmd.cq_handle, file->ucontext, 0);
1635 	if (!cq)
1636 		return -EINVAL;
1637 
1638 	/* we copy a struct ib_uverbs_poll_cq_resp to user space */
1639 	header_ptr = (void __user *)(unsigned long) cmd.response;
1640 	data_ptr = header_ptr + sizeof resp;
1641 
1642 	memset(&resp, 0, sizeof resp);
1643 	while (resp.count < cmd.ne) {
1644 		ret = ib_poll_cq(cq, 1, &wc);
1645 		if (ret < 0)
1646 			goto out_put;
1647 		if (!ret)
1648 			break;
1649 
1650 		ret = copy_wc_to_user(data_ptr, &wc);
1651 		if (ret)
1652 			goto out_put;
1653 
1654 		data_ptr += sizeof(struct ib_uverbs_wc);
1655 		++resp.count;
1656 	}
1657 
1658 	if (copy_to_user(header_ptr, &resp, sizeof resp)) {
1659 		ret = -EFAULT;
1660 		goto out_put;
1661 	}
1662 
1663 	ret = in_len;
1664 
1665 out_put:
1666 	put_cq_read(cq);
1667 	return ret;
1668 }
1669 
1670 ssize_t ib_uverbs_req_notify_cq(struct ib_uverbs_file *file,
1671 				struct ib_device *ib_dev,
1672 				const char __user *buf, int in_len,
1673 				int out_len)
1674 {
1675 	struct ib_uverbs_req_notify_cq cmd;
1676 	struct ib_cq                  *cq;
1677 
1678 	if (copy_from_user(&cmd, buf, sizeof cmd))
1679 		return -EFAULT;
1680 
1681 	cq = idr_read_cq(cmd.cq_handle, file->ucontext, 0);
1682 	if (!cq)
1683 		return -EINVAL;
1684 
1685 	ib_req_notify_cq(cq, cmd.solicited_only ?
1686 			 IB_CQ_SOLICITED : IB_CQ_NEXT_COMP);
1687 
1688 	put_cq_read(cq);
1689 
1690 	return in_len;
1691 }
1692 
1693 ssize_t ib_uverbs_destroy_cq(struct ib_uverbs_file *file,
1694 			     struct ib_device *ib_dev,
1695 			     const char __user *buf, int in_len,
1696 			     int out_len)
1697 {
1698 	struct ib_uverbs_destroy_cq      cmd;
1699 	struct ib_uverbs_destroy_cq_resp resp;
1700 	struct ib_uobject		*uobj;
1701 	struct ib_cq               	*cq;
1702 	struct ib_ucq_object        	*obj;
1703 	struct ib_uverbs_event_file	*ev_file;
1704 	int                        	 ret = -EINVAL;
1705 
1706 	if (copy_from_user(&cmd, buf, sizeof cmd))
1707 		return -EFAULT;
1708 
1709 	uobj = idr_write_uobj(&ib_uverbs_cq_idr, cmd.cq_handle, file->ucontext);
1710 	if (!uobj)
1711 		return -EINVAL;
1712 	cq      = uobj->object;
1713 	ev_file = cq->cq_context;
1714 	obj     = container_of(cq->uobject, struct ib_ucq_object, uobject);
1715 
1716 	ret = ib_destroy_cq(cq);
1717 	if (!ret)
1718 		uobj->live = 0;
1719 
1720 	put_uobj_write(uobj);
1721 
1722 	if (ret)
1723 		return ret;
1724 
1725 	idr_remove_uobj(&ib_uverbs_cq_idr, uobj);
1726 
1727 	mutex_lock(&file->mutex);
1728 	list_del(&uobj->list);
1729 	mutex_unlock(&file->mutex);
1730 
1731 	ib_uverbs_release_ucq(file, ev_file, obj);
1732 
1733 	memset(&resp, 0, sizeof resp);
1734 	resp.comp_events_reported  = obj->comp_events_reported;
1735 	resp.async_events_reported = obj->async_events_reported;
1736 
1737 	put_uobj(uobj);
1738 
1739 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
1740 			 &resp, sizeof resp))
1741 		return -EFAULT;
1742 
1743 	return in_len;
1744 }
1745 
1746 static int create_qp(struct ib_uverbs_file *file,
1747 		     struct ib_udata *ucore,
1748 		     struct ib_udata *uhw,
1749 		     struct ib_uverbs_ex_create_qp *cmd,
1750 		     size_t cmd_sz,
1751 		     int (*cb)(struct ib_uverbs_file *file,
1752 			       struct ib_uverbs_ex_create_qp_resp *resp,
1753 			       struct ib_udata *udata),
1754 		     void *context)
1755 {
1756 	struct ib_uqp_object		*obj;
1757 	struct ib_device		*device;
1758 	struct ib_pd			*pd = NULL;
1759 	struct ib_xrcd			*xrcd = NULL;
1760 	struct ib_uobject		*uninitialized_var(xrcd_uobj);
1761 	struct ib_cq			*scq = NULL, *rcq = NULL;
1762 	struct ib_srq			*srq = NULL;
1763 	struct ib_qp			*qp;
1764 	char				*buf;
1765 	struct ib_qp_init_attr		attr;
1766 	struct ib_uverbs_ex_create_qp_resp resp;
1767 	int				ret;
1768 
1769 	if (cmd->qp_type == IB_QPT_RAW_PACKET && !capable(CAP_NET_RAW))
1770 		return -EPERM;
1771 
1772 	obj = kzalloc(sizeof *obj, GFP_KERNEL);
1773 	if (!obj)
1774 		return -ENOMEM;
1775 
1776 	init_uobj(&obj->uevent.uobject, cmd->user_handle, file->ucontext,
1777 		  &qp_lock_class);
1778 	down_write(&obj->uevent.uobject.mutex);
1779 
1780 	if (cmd->qp_type == IB_QPT_XRC_TGT) {
1781 		xrcd = idr_read_xrcd(cmd->pd_handle, file->ucontext,
1782 				     &xrcd_uobj);
1783 		if (!xrcd) {
1784 			ret = -EINVAL;
1785 			goto err_put;
1786 		}
1787 		device = xrcd->device;
1788 	} else {
1789 		if (cmd->qp_type == IB_QPT_XRC_INI) {
1790 			cmd->max_recv_wr = 0;
1791 			cmd->max_recv_sge = 0;
1792 		} else {
1793 			if (cmd->is_srq) {
1794 				srq = idr_read_srq(cmd->srq_handle,
1795 						   file->ucontext);
1796 				if (!srq || srq->srq_type != IB_SRQT_BASIC) {
1797 					ret = -EINVAL;
1798 					goto err_put;
1799 				}
1800 			}
1801 
1802 			if (cmd->recv_cq_handle != cmd->send_cq_handle) {
1803 				rcq = idr_read_cq(cmd->recv_cq_handle,
1804 						  file->ucontext, 0);
1805 				if (!rcq) {
1806 					ret = -EINVAL;
1807 					goto err_put;
1808 				}
1809 			}
1810 		}
1811 
1812 		scq = idr_read_cq(cmd->send_cq_handle, file->ucontext, !!rcq);
1813 		rcq = rcq ?: scq;
1814 		pd  = idr_read_pd(cmd->pd_handle, file->ucontext);
1815 		if (!pd || !scq) {
1816 			ret = -EINVAL;
1817 			goto err_put;
1818 		}
1819 
1820 		device = pd->device;
1821 	}
1822 
1823 	attr.event_handler = ib_uverbs_qp_event_handler;
1824 	attr.qp_context    = file;
1825 	attr.send_cq       = scq;
1826 	attr.recv_cq       = rcq;
1827 	attr.srq           = srq;
1828 	attr.xrcd	   = xrcd;
1829 	attr.sq_sig_type   = cmd->sq_sig_all ? IB_SIGNAL_ALL_WR :
1830 					      IB_SIGNAL_REQ_WR;
1831 	attr.qp_type       = cmd->qp_type;
1832 	attr.create_flags  = 0;
1833 
1834 	attr.cap.max_send_wr     = cmd->max_send_wr;
1835 	attr.cap.max_recv_wr     = cmd->max_recv_wr;
1836 	attr.cap.max_send_sge    = cmd->max_send_sge;
1837 	attr.cap.max_recv_sge    = cmd->max_recv_sge;
1838 	attr.cap.max_inline_data = cmd->max_inline_data;
1839 
1840 	obj->uevent.events_reported     = 0;
1841 	INIT_LIST_HEAD(&obj->uevent.event_list);
1842 	INIT_LIST_HEAD(&obj->mcast_list);
1843 
1844 	if (cmd_sz >= offsetof(typeof(*cmd), create_flags) +
1845 		      sizeof(cmd->create_flags))
1846 		attr.create_flags = cmd->create_flags;
1847 
1848 	if (attr.create_flags & ~IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK) {
1849 		ret = -EINVAL;
1850 		goto err_put;
1851 	}
1852 
1853 	buf = (void *)cmd + sizeof(*cmd);
1854 	if (cmd_sz > sizeof(*cmd))
1855 		if (!(buf[0] == 0 && !memcmp(buf, buf + 1,
1856 					     cmd_sz - sizeof(*cmd) - 1))) {
1857 			ret = -EINVAL;
1858 			goto err_put;
1859 		}
1860 
1861 	if (cmd->qp_type == IB_QPT_XRC_TGT)
1862 		qp = ib_create_qp(pd, &attr);
1863 	else
1864 		qp = device->create_qp(pd, &attr, uhw);
1865 
1866 	if (IS_ERR(qp)) {
1867 		ret = PTR_ERR(qp);
1868 		goto err_put;
1869 	}
1870 
1871 	if (cmd->qp_type != IB_QPT_XRC_TGT) {
1872 		qp->real_qp	  = qp;
1873 		qp->device	  = device;
1874 		qp->pd		  = pd;
1875 		qp->send_cq	  = attr.send_cq;
1876 		qp->recv_cq	  = attr.recv_cq;
1877 		qp->srq		  = attr.srq;
1878 		qp->event_handler = attr.event_handler;
1879 		qp->qp_context	  = attr.qp_context;
1880 		qp->qp_type	  = attr.qp_type;
1881 		atomic_set(&qp->usecnt, 0);
1882 		atomic_inc(&pd->usecnt);
1883 		atomic_inc(&attr.send_cq->usecnt);
1884 		if (attr.recv_cq)
1885 			atomic_inc(&attr.recv_cq->usecnt);
1886 		if (attr.srq)
1887 			atomic_inc(&attr.srq->usecnt);
1888 	}
1889 	qp->uobject = &obj->uevent.uobject;
1890 
1891 	obj->uevent.uobject.object = qp;
1892 	ret = idr_add_uobj(&ib_uverbs_qp_idr, &obj->uevent.uobject);
1893 	if (ret)
1894 		goto err_destroy;
1895 
1896 	memset(&resp, 0, sizeof resp);
1897 	resp.base.qpn             = qp->qp_num;
1898 	resp.base.qp_handle       = obj->uevent.uobject.id;
1899 	resp.base.max_recv_sge    = attr.cap.max_recv_sge;
1900 	resp.base.max_send_sge    = attr.cap.max_send_sge;
1901 	resp.base.max_recv_wr     = attr.cap.max_recv_wr;
1902 	resp.base.max_send_wr     = attr.cap.max_send_wr;
1903 	resp.base.max_inline_data = attr.cap.max_inline_data;
1904 
1905 	resp.response_length = offsetof(typeof(resp), response_length) +
1906 			       sizeof(resp.response_length);
1907 
1908 	ret = cb(file, &resp, ucore);
1909 	if (ret)
1910 		goto err_cb;
1911 
1912 	if (xrcd) {
1913 		obj->uxrcd = container_of(xrcd_uobj, struct ib_uxrcd_object,
1914 					  uobject);
1915 		atomic_inc(&obj->uxrcd->refcnt);
1916 		put_xrcd_read(xrcd_uobj);
1917 	}
1918 
1919 	if (pd)
1920 		put_pd_read(pd);
1921 	if (scq)
1922 		put_cq_read(scq);
1923 	if (rcq && rcq != scq)
1924 		put_cq_read(rcq);
1925 	if (srq)
1926 		put_srq_read(srq);
1927 
1928 	mutex_lock(&file->mutex);
1929 	list_add_tail(&obj->uevent.uobject.list, &file->ucontext->qp_list);
1930 	mutex_unlock(&file->mutex);
1931 
1932 	obj->uevent.uobject.live = 1;
1933 
1934 	up_write(&obj->uevent.uobject.mutex);
1935 
1936 	return 0;
1937 err_cb:
1938 	idr_remove_uobj(&ib_uverbs_qp_idr, &obj->uevent.uobject);
1939 
1940 err_destroy:
1941 	ib_destroy_qp(qp);
1942 
1943 err_put:
1944 	if (xrcd)
1945 		put_xrcd_read(xrcd_uobj);
1946 	if (pd)
1947 		put_pd_read(pd);
1948 	if (scq)
1949 		put_cq_read(scq);
1950 	if (rcq && rcq != scq)
1951 		put_cq_read(rcq);
1952 	if (srq)
1953 		put_srq_read(srq);
1954 
1955 	put_uobj_write(&obj->uevent.uobject);
1956 	return ret;
1957 }
1958 
1959 static int ib_uverbs_create_qp_cb(struct ib_uverbs_file *file,
1960 				  struct ib_uverbs_ex_create_qp_resp *resp,
1961 				  struct ib_udata *ucore)
1962 {
1963 	if (ib_copy_to_udata(ucore, &resp->base, sizeof(resp->base)))
1964 		return -EFAULT;
1965 
1966 	return 0;
1967 }
1968 
1969 ssize_t ib_uverbs_create_qp(struct ib_uverbs_file *file,
1970 			    struct ib_device *ib_dev,
1971 			    const char __user *buf, int in_len,
1972 			    int out_len)
1973 {
1974 	struct ib_uverbs_create_qp      cmd;
1975 	struct ib_uverbs_ex_create_qp	cmd_ex;
1976 	struct ib_udata			ucore;
1977 	struct ib_udata			uhw;
1978 	ssize_t resp_size = sizeof(struct ib_uverbs_create_qp_resp);
1979 	int				err;
1980 
1981 	if (out_len < resp_size)
1982 		return -ENOSPC;
1983 
1984 	if (copy_from_user(&cmd, buf, sizeof(cmd)))
1985 		return -EFAULT;
1986 
1987 	INIT_UDATA(&ucore, buf, (unsigned long)cmd.response, sizeof(cmd),
1988 		   resp_size);
1989 	INIT_UDATA(&uhw, buf + sizeof(cmd),
1990 		   (unsigned long)cmd.response + resp_size,
1991 		   in_len - sizeof(cmd), out_len - resp_size);
1992 
1993 	memset(&cmd_ex, 0, sizeof(cmd_ex));
1994 	cmd_ex.user_handle = cmd.user_handle;
1995 	cmd_ex.pd_handle = cmd.pd_handle;
1996 	cmd_ex.send_cq_handle = cmd.send_cq_handle;
1997 	cmd_ex.recv_cq_handle = cmd.recv_cq_handle;
1998 	cmd_ex.srq_handle = cmd.srq_handle;
1999 	cmd_ex.max_send_wr = cmd.max_send_wr;
2000 	cmd_ex.max_recv_wr = cmd.max_recv_wr;
2001 	cmd_ex.max_send_sge = cmd.max_send_sge;
2002 	cmd_ex.max_recv_sge = cmd.max_recv_sge;
2003 	cmd_ex.max_inline_data = cmd.max_inline_data;
2004 	cmd_ex.sq_sig_all = cmd.sq_sig_all;
2005 	cmd_ex.qp_type = cmd.qp_type;
2006 	cmd_ex.is_srq = cmd.is_srq;
2007 
2008 	err = create_qp(file, &ucore, &uhw, &cmd_ex,
2009 			offsetof(typeof(cmd_ex), is_srq) +
2010 			sizeof(cmd.is_srq), ib_uverbs_create_qp_cb,
2011 			NULL);
2012 
2013 	if (err)
2014 		return err;
2015 
2016 	return in_len;
2017 }
2018 
2019 static int ib_uverbs_ex_create_qp_cb(struct ib_uverbs_file *file,
2020 				     struct ib_uverbs_ex_create_qp_resp *resp,
2021 				     struct ib_udata *ucore)
2022 {
2023 	if (ib_copy_to_udata(ucore, resp, resp->response_length))
2024 		return -EFAULT;
2025 
2026 	return 0;
2027 }
2028 
2029 int ib_uverbs_ex_create_qp(struct ib_uverbs_file *file,
2030 			   struct ib_device *ib_dev,
2031 			   struct ib_udata *ucore,
2032 			   struct ib_udata *uhw)
2033 {
2034 	struct ib_uverbs_ex_create_qp_resp resp;
2035 	struct ib_uverbs_ex_create_qp cmd = {0};
2036 	int err;
2037 
2038 	if (ucore->inlen < (offsetof(typeof(cmd), comp_mask) +
2039 			    sizeof(cmd.comp_mask)))
2040 		return -EINVAL;
2041 
2042 	err = ib_copy_from_udata(&cmd, ucore, min(sizeof(cmd), ucore->inlen));
2043 	if (err)
2044 		return err;
2045 
2046 	if (cmd.comp_mask)
2047 		return -EINVAL;
2048 
2049 	if (cmd.reserved)
2050 		return -EINVAL;
2051 
2052 	if (ucore->outlen < (offsetof(typeof(resp), response_length) +
2053 			     sizeof(resp.response_length)))
2054 		return -ENOSPC;
2055 
2056 	err = create_qp(file, ucore, uhw, &cmd,
2057 			min(ucore->inlen, sizeof(cmd)),
2058 			ib_uverbs_ex_create_qp_cb, NULL);
2059 
2060 	if (err)
2061 		return err;
2062 
2063 	return 0;
2064 }
2065 
2066 ssize_t ib_uverbs_open_qp(struct ib_uverbs_file *file,
2067 			  struct ib_device *ib_dev,
2068 			  const char __user *buf, int in_len, int out_len)
2069 {
2070 	struct ib_uverbs_open_qp        cmd;
2071 	struct ib_uverbs_create_qp_resp resp;
2072 	struct ib_udata                 udata;
2073 	struct ib_uqp_object           *obj;
2074 	struct ib_xrcd		       *xrcd;
2075 	struct ib_uobject	       *uninitialized_var(xrcd_uobj);
2076 	struct ib_qp                   *qp;
2077 	struct ib_qp_open_attr          attr;
2078 	int ret;
2079 
2080 	if (out_len < sizeof resp)
2081 		return -ENOSPC;
2082 
2083 	if (copy_from_user(&cmd, buf, sizeof cmd))
2084 		return -EFAULT;
2085 
2086 	INIT_UDATA(&udata, buf + sizeof cmd,
2087 		   (unsigned long) cmd.response + sizeof resp,
2088 		   in_len - sizeof cmd, out_len - sizeof resp);
2089 
2090 	obj = kmalloc(sizeof *obj, GFP_KERNEL);
2091 	if (!obj)
2092 		return -ENOMEM;
2093 
2094 	init_uobj(&obj->uevent.uobject, cmd.user_handle, file->ucontext, &qp_lock_class);
2095 	down_write(&obj->uevent.uobject.mutex);
2096 
2097 	xrcd = idr_read_xrcd(cmd.pd_handle, file->ucontext, &xrcd_uobj);
2098 	if (!xrcd) {
2099 		ret = -EINVAL;
2100 		goto err_put;
2101 	}
2102 
2103 	attr.event_handler = ib_uverbs_qp_event_handler;
2104 	attr.qp_context    = file;
2105 	attr.qp_num        = cmd.qpn;
2106 	attr.qp_type       = cmd.qp_type;
2107 
2108 	obj->uevent.events_reported = 0;
2109 	INIT_LIST_HEAD(&obj->uevent.event_list);
2110 	INIT_LIST_HEAD(&obj->mcast_list);
2111 
2112 	qp = ib_open_qp(xrcd, &attr);
2113 	if (IS_ERR(qp)) {
2114 		ret = PTR_ERR(qp);
2115 		goto err_put;
2116 	}
2117 
2118 	qp->uobject = &obj->uevent.uobject;
2119 
2120 	obj->uevent.uobject.object = qp;
2121 	ret = idr_add_uobj(&ib_uverbs_qp_idr, &obj->uevent.uobject);
2122 	if (ret)
2123 		goto err_destroy;
2124 
2125 	memset(&resp, 0, sizeof resp);
2126 	resp.qpn       = qp->qp_num;
2127 	resp.qp_handle = obj->uevent.uobject.id;
2128 
2129 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
2130 			 &resp, sizeof resp)) {
2131 		ret = -EFAULT;
2132 		goto err_remove;
2133 	}
2134 
2135 	obj->uxrcd = container_of(xrcd_uobj, struct ib_uxrcd_object, uobject);
2136 	atomic_inc(&obj->uxrcd->refcnt);
2137 	put_xrcd_read(xrcd_uobj);
2138 
2139 	mutex_lock(&file->mutex);
2140 	list_add_tail(&obj->uevent.uobject.list, &file->ucontext->qp_list);
2141 	mutex_unlock(&file->mutex);
2142 
2143 	obj->uevent.uobject.live = 1;
2144 
2145 	up_write(&obj->uevent.uobject.mutex);
2146 
2147 	return in_len;
2148 
2149 err_remove:
2150 	idr_remove_uobj(&ib_uverbs_qp_idr, &obj->uevent.uobject);
2151 
2152 err_destroy:
2153 	ib_destroy_qp(qp);
2154 
2155 err_put:
2156 	put_xrcd_read(xrcd_uobj);
2157 	put_uobj_write(&obj->uevent.uobject);
2158 	return ret;
2159 }
2160 
2161 ssize_t ib_uverbs_query_qp(struct ib_uverbs_file *file,
2162 			   struct ib_device *ib_dev,
2163 			   const char __user *buf, int in_len,
2164 			   int out_len)
2165 {
2166 	struct ib_uverbs_query_qp      cmd;
2167 	struct ib_uverbs_query_qp_resp resp;
2168 	struct ib_qp                   *qp;
2169 	struct ib_qp_attr              *attr;
2170 	struct ib_qp_init_attr         *init_attr;
2171 	int                            ret;
2172 
2173 	if (copy_from_user(&cmd, buf, sizeof cmd))
2174 		return -EFAULT;
2175 
2176 	attr      = kmalloc(sizeof *attr, GFP_KERNEL);
2177 	init_attr = kmalloc(sizeof *init_attr, GFP_KERNEL);
2178 	if (!attr || !init_attr) {
2179 		ret = -ENOMEM;
2180 		goto out;
2181 	}
2182 
2183 	qp = idr_read_qp(cmd.qp_handle, file->ucontext);
2184 	if (!qp) {
2185 		ret = -EINVAL;
2186 		goto out;
2187 	}
2188 
2189 	ret = ib_query_qp(qp, attr, cmd.attr_mask, init_attr);
2190 
2191 	put_qp_read(qp);
2192 
2193 	if (ret)
2194 		goto out;
2195 
2196 	memset(&resp, 0, sizeof resp);
2197 
2198 	resp.qp_state               = attr->qp_state;
2199 	resp.cur_qp_state           = attr->cur_qp_state;
2200 	resp.path_mtu               = attr->path_mtu;
2201 	resp.path_mig_state         = attr->path_mig_state;
2202 	resp.qkey                   = attr->qkey;
2203 	resp.rq_psn                 = attr->rq_psn;
2204 	resp.sq_psn                 = attr->sq_psn;
2205 	resp.dest_qp_num            = attr->dest_qp_num;
2206 	resp.qp_access_flags        = attr->qp_access_flags;
2207 	resp.pkey_index             = attr->pkey_index;
2208 	resp.alt_pkey_index         = attr->alt_pkey_index;
2209 	resp.sq_draining            = attr->sq_draining;
2210 	resp.max_rd_atomic          = attr->max_rd_atomic;
2211 	resp.max_dest_rd_atomic     = attr->max_dest_rd_atomic;
2212 	resp.min_rnr_timer          = attr->min_rnr_timer;
2213 	resp.port_num               = attr->port_num;
2214 	resp.timeout                = attr->timeout;
2215 	resp.retry_cnt              = attr->retry_cnt;
2216 	resp.rnr_retry              = attr->rnr_retry;
2217 	resp.alt_port_num           = attr->alt_port_num;
2218 	resp.alt_timeout            = attr->alt_timeout;
2219 
2220 	memcpy(resp.dest.dgid, attr->ah_attr.grh.dgid.raw, 16);
2221 	resp.dest.flow_label        = attr->ah_attr.grh.flow_label;
2222 	resp.dest.sgid_index        = attr->ah_attr.grh.sgid_index;
2223 	resp.dest.hop_limit         = attr->ah_attr.grh.hop_limit;
2224 	resp.dest.traffic_class     = attr->ah_attr.grh.traffic_class;
2225 	resp.dest.dlid              = attr->ah_attr.dlid;
2226 	resp.dest.sl                = attr->ah_attr.sl;
2227 	resp.dest.src_path_bits     = attr->ah_attr.src_path_bits;
2228 	resp.dest.static_rate       = attr->ah_attr.static_rate;
2229 	resp.dest.is_global         = !!(attr->ah_attr.ah_flags & IB_AH_GRH);
2230 	resp.dest.port_num          = attr->ah_attr.port_num;
2231 
2232 	memcpy(resp.alt_dest.dgid, attr->alt_ah_attr.grh.dgid.raw, 16);
2233 	resp.alt_dest.flow_label    = attr->alt_ah_attr.grh.flow_label;
2234 	resp.alt_dest.sgid_index    = attr->alt_ah_attr.grh.sgid_index;
2235 	resp.alt_dest.hop_limit     = attr->alt_ah_attr.grh.hop_limit;
2236 	resp.alt_dest.traffic_class = attr->alt_ah_attr.grh.traffic_class;
2237 	resp.alt_dest.dlid          = attr->alt_ah_attr.dlid;
2238 	resp.alt_dest.sl            = attr->alt_ah_attr.sl;
2239 	resp.alt_dest.src_path_bits = attr->alt_ah_attr.src_path_bits;
2240 	resp.alt_dest.static_rate   = attr->alt_ah_attr.static_rate;
2241 	resp.alt_dest.is_global     = !!(attr->alt_ah_attr.ah_flags & IB_AH_GRH);
2242 	resp.alt_dest.port_num      = attr->alt_ah_attr.port_num;
2243 
2244 	resp.max_send_wr            = init_attr->cap.max_send_wr;
2245 	resp.max_recv_wr            = init_attr->cap.max_recv_wr;
2246 	resp.max_send_sge           = init_attr->cap.max_send_sge;
2247 	resp.max_recv_sge           = init_attr->cap.max_recv_sge;
2248 	resp.max_inline_data        = init_attr->cap.max_inline_data;
2249 	resp.sq_sig_all             = init_attr->sq_sig_type == IB_SIGNAL_ALL_WR;
2250 
2251 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
2252 			 &resp, sizeof resp))
2253 		ret = -EFAULT;
2254 
2255 out:
2256 	kfree(attr);
2257 	kfree(init_attr);
2258 
2259 	return ret ? ret : in_len;
2260 }
2261 
2262 /* Remove ignored fields set in the attribute mask */
2263 static int modify_qp_mask(enum ib_qp_type qp_type, int mask)
2264 {
2265 	switch (qp_type) {
2266 	case IB_QPT_XRC_INI:
2267 		return mask & ~(IB_QP_MAX_DEST_RD_ATOMIC | IB_QP_MIN_RNR_TIMER);
2268 	case IB_QPT_XRC_TGT:
2269 		return mask & ~(IB_QP_MAX_QP_RD_ATOMIC | IB_QP_RETRY_CNT |
2270 				IB_QP_RNR_RETRY);
2271 	default:
2272 		return mask;
2273 	}
2274 }
2275 
2276 ssize_t ib_uverbs_modify_qp(struct ib_uverbs_file *file,
2277 			    struct ib_device *ib_dev,
2278 			    const char __user *buf, int in_len,
2279 			    int out_len)
2280 {
2281 	struct ib_uverbs_modify_qp cmd;
2282 	struct ib_udata            udata;
2283 	struct ib_qp              *qp;
2284 	struct ib_qp_attr         *attr;
2285 	int                        ret;
2286 
2287 	if (copy_from_user(&cmd, buf, sizeof cmd))
2288 		return -EFAULT;
2289 
2290 	INIT_UDATA(&udata, buf + sizeof cmd, NULL, in_len - sizeof cmd,
2291 		   out_len);
2292 
2293 	attr = kmalloc(sizeof *attr, GFP_KERNEL);
2294 	if (!attr)
2295 		return -ENOMEM;
2296 
2297 	qp = idr_read_qp(cmd.qp_handle, file->ucontext);
2298 	if (!qp) {
2299 		ret = -EINVAL;
2300 		goto out;
2301 	}
2302 
2303 	attr->qp_state 		  = cmd.qp_state;
2304 	attr->cur_qp_state 	  = cmd.cur_qp_state;
2305 	attr->path_mtu 		  = cmd.path_mtu;
2306 	attr->path_mig_state 	  = cmd.path_mig_state;
2307 	attr->qkey 		  = cmd.qkey;
2308 	attr->rq_psn 		  = cmd.rq_psn;
2309 	attr->sq_psn 		  = cmd.sq_psn;
2310 	attr->dest_qp_num 	  = cmd.dest_qp_num;
2311 	attr->qp_access_flags 	  = cmd.qp_access_flags;
2312 	attr->pkey_index 	  = cmd.pkey_index;
2313 	attr->alt_pkey_index 	  = cmd.alt_pkey_index;
2314 	attr->en_sqd_async_notify = cmd.en_sqd_async_notify;
2315 	attr->max_rd_atomic 	  = cmd.max_rd_atomic;
2316 	attr->max_dest_rd_atomic  = cmd.max_dest_rd_atomic;
2317 	attr->min_rnr_timer 	  = cmd.min_rnr_timer;
2318 	attr->port_num 		  = cmd.port_num;
2319 	attr->timeout 		  = cmd.timeout;
2320 	attr->retry_cnt 	  = cmd.retry_cnt;
2321 	attr->rnr_retry 	  = cmd.rnr_retry;
2322 	attr->alt_port_num 	  = cmd.alt_port_num;
2323 	attr->alt_timeout 	  = cmd.alt_timeout;
2324 
2325 	memcpy(attr->ah_attr.grh.dgid.raw, cmd.dest.dgid, 16);
2326 	attr->ah_attr.grh.flow_label        = cmd.dest.flow_label;
2327 	attr->ah_attr.grh.sgid_index        = cmd.dest.sgid_index;
2328 	attr->ah_attr.grh.hop_limit         = cmd.dest.hop_limit;
2329 	attr->ah_attr.grh.traffic_class     = cmd.dest.traffic_class;
2330 	attr->ah_attr.dlid 	    	    = cmd.dest.dlid;
2331 	attr->ah_attr.sl   	    	    = cmd.dest.sl;
2332 	attr->ah_attr.src_path_bits 	    = cmd.dest.src_path_bits;
2333 	attr->ah_attr.static_rate   	    = cmd.dest.static_rate;
2334 	attr->ah_attr.ah_flags 	    	    = cmd.dest.is_global ? IB_AH_GRH : 0;
2335 	attr->ah_attr.port_num 	    	    = cmd.dest.port_num;
2336 
2337 	memcpy(attr->alt_ah_attr.grh.dgid.raw, cmd.alt_dest.dgid, 16);
2338 	attr->alt_ah_attr.grh.flow_label    = cmd.alt_dest.flow_label;
2339 	attr->alt_ah_attr.grh.sgid_index    = cmd.alt_dest.sgid_index;
2340 	attr->alt_ah_attr.grh.hop_limit     = cmd.alt_dest.hop_limit;
2341 	attr->alt_ah_attr.grh.traffic_class = cmd.alt_dest.traffic_class;
2342 	attr->alt_ah_attr.dlid 	    	    = cmd.alt_dest.dlid;
2343 	attr->alt_ah_attr.sl   	    	    = cmd.alt_dest.sl;
2344 	attr->alt_ah_attr.src_path_bits     = cmd.alt_dest.src_path_bits;
2345 	attr->alt_ah_attr.static_rate       = cmd.alt_dest.static_rate;
2346 	attr->alt_ah_attr.ah_flags 	    = cmd.alt_dest.is_global ? IB_AH_GRH : 0;
2347 	attr->alt_ah_attr.port_num 	    = cmd.alt_dest.port_num;
2348 
2349 	if (qp->real_qp == qp) {
2350 		ret = ib_resolve_eth_dmac(qp, attr, &cmd.attr_mask);
2351 		if (ret)
2352 			goto release_qp;
2353 		ret = qp->device->modify_qp(qp, attr,
2354 			modify_qp_mask(qp->qp_type, cmd.attr_mask), &udata);
2355 	} else {
2356 		ret = ib_modify_qp(qp, attr, modify_qp_mask(qp->qp_type, cmd.attr_mask));
2357 	}
2358 
2359 	if (ret)
2360 		goto release_qp;
2361 
2362 	ret = in_len;
2363 
2364 release_qp:
2365 	put_qp_read(qp);
2366 
2367 out:
2368 	kfree(attr);
2369 
2370 	return ret;
2371 }
2372 
2373 ssize_t ib_uverbs_destroy_qp(struct ib_uverbs_file *file,
2374 			     struct ib_device *ib_dev,
2375 			     const char __user *buf, int in_len,
2376 			     int out_len)
2377 {
2378 	struct ib_uverbs_destroy_qp      cmd;
2379 	struct ib_uverbs_destroy_qp_resp resp;
2380 	struct ib_uobject		*uobj;
2381 	struct ib_qp               	*qp;
2382 	struct ib_uqp_object        	*obj;
2383 	int                        	 ret = -EINVAL;
2384 
2385 	if (copy_from_user(&cmd, buf, sizeof cmd))
2386 		return -EFAULT;
2387 
2388 	memset(&resp, 0, sizeof resp);
2389 
2390 	uobj = idr_write_uobj(&ib_uverbs_qp_idr, cmd.qp_handle, file->ucontext);
2391 	if (!uobj)
2392 		return -EINVAL;
2393 	qp  = uobj->object;
2394 	obj = container_of(uobj, struct ib_uqp_object, uevent.uobject);
2395 
2396 	if (!list_empty(&obj->mcast_list)) {
2397 		put_uobj_write(uobj);
2398 		return -EBUSY;
2399 	}
2400 
2401 	ret = ib_destroy_qp(qp);
2402 	if (!ret)
2403 		uobj->live = 0;
2404 
2405 	put_uobj_write(uobj);
2406 
2407 	if (ret)
2408 		return ret;
2409 
2410 	if (obj->uxrcd)
2411 		atomic_dec(&obj->uxrcd->refcnt);
2412 
2413 	idr_remove_uobj(&ib_uverbs_qp_idr, uobj);
2414 
2415 	mutex_lock(&file->mutex);
2416 	list_del(&uobj->list);
2417 	mutex_unlock(&file->mutex);
2418 
2419 	ib_uverbs_release_uevent(file, &obj->uevent);
2420 
2421 	resp.events_reported = obj->uevent.events_reported;
2422 
2423 	put_uobj(uobj);
2424 
2425 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
2426 			 &resp, sizeof resp))
2427 		return -EFAULT;
2428 
2429 	return in_len;
2430 }
2431 
2432 static void *alloc_wr(size_t wr_size, __u32 num_sge)
2433 {
2434 	return kmalloc(ALIGN(wr_size, sizeof (struct ib_sge)) +
2435 			 num_sge * sizeof (struct ib_sge), GFP_KERNEL);
2436 };
2437 
2438 ssize_t ib_uverbs_post_send(struct ib_uverbs_file *file,
2439 			    struct ib_device *ib_dev,
2440 			    const char __user *buf, int in_len,
2441 			    int out_len)
2442 {
2443 	struct ib_uverbs_post_send      cmd;
2444 	struct ib_uverbs_post_send_resp resp;
2445 	struct ib_uverbs_send_wr       *user_wr;
2446 	struct ib_send_wr              *wr = NULL, *last, *next, *bad_wr;
2447 	struct ib_qp                   *qp;
2448 	int                             i, sg_ind;
2449 	int				is_ud;
2450 	ssize_t                         ret = -EINVAL;
2451 	size_t                          next_size;
2452 
2453 	if (copy_from_user(&cmd, buf, sizeof cmd))
2454 		return -EFAULT;
2455 
2456 	if (in_len < sizeof cmd + cmd.wqe_size * cmd.wr_count +
2457 	    cmd.sge_count * sizeof (struct ib_uverbs_sge))
2458 		return -EINVAL;
2459 
2460 	if (cmd.wqe_size < sizeof (struct ib_uverbs_send_wr))
2461 		return -EINVAL;
2462 
2463 	user_wr = kmalloc(cmd.wqe_size, GFP_KERNEL);
2464 	if (!user_wr)
2465 		return -ENOMEM;
2466 
2467 	qp = idr_read_qp(cmd.qp_handle, file->ucontext);
2468 	if (!qp)
2469 		goto out;
2470 
2471 	is_ud = qp->qp_type == IB_QPT_UD;
2472 	sg_ind = 0;
2473 	last = NULL;
2474 	for (i = 0; i < cmd.wr_count; ++i) {
2475 		if (copy_from_user(user_wr,
2476 				   buf + sizeof cmd + i * cmd.wqe_size,
2477 				   cmd.wqe_size)) {
2478 			ret = -EFAULT;
2479 			goto out_put;
2480 		}
2481 
2482 		if (user_wr->num_sge + sg_ind > cmd.sge_count) {
2483 			ret = -EINVAL;
2484 			goto out_put;
2485 		}
2486 
2487 		if (is_ud) {
2488 			struct ib_ud_wr *ud;
2489 
2490 			if (user_wr->opcode != IB_WR_SEND &&
2491 			    user_wr->opcode != IB_WR_SEND_WITH_IMM) {
2492 				ret = -EINVAL;
2493 				goto out_put;
2494 			}
2495 
2496 			next_size = sizeof(*ud);
2497 			ud = alloc_wr(next_size, user_wr->num_sge);
2498 			if (!ud) {
2499 				ret = -ENOMEM;
2500 				goto out_put;
2501 			}
2502 
2503 			ud->ah = idr_read_ah(user_wr->wr.ud.ah, file->ucontext);
2504 			if (!ud->ah) {
2505 				kfree(ud);
2506 				ret = -EINVAL;
2507 				goto out_put;
2508 			}
2509 			ud->remote_qpn = user_wr->wr.ud.remote_qpn;
2510 			ud->remote_qkey = user_wr->wr.ud.remote_qkey;
2511 
2512 			next = &ud->wr;
2513 		} else if (user_wr->opcode == IB_WR_RDMA_WRITE_WITH_IMM ||
2514 			   user_wr->opcode == IB_WR_RDMA_WRITE ||
2515 			   user_wr->opcode == IB_WR_RDMA_READ) {
2516 			struct ib_rdma_wr *rdma;
2517 
2518 			next_size = sizeof(*rdma);
2519 			rdma = alloc_wr(next_size, user_wr->num_sge);
2520 			if (!rdma) {
2521 				ret = -ENOMEM;
2522 				goto out_put;
2523 			}
2524 
2525 			rdma->remote_addr = user_wr->wr.rdma.remote_addr;
2526 			rdma->rkey = user_wr->wr.rdma.rkey;
2527 
2528 			next = &rdma->wr;
2529 		} else if (user_wr->opcode == IB_WR_ATOMIC_CMP_AND_SWP ||
2530 			   user_wr->opcode == IB_WR_ATOMIC_FETCH_AND_ADD) {
2531 			struct ib_atomic_wr *atomic;
2532 
2533 			next_size = sizeof(*atomic);
2534 			atomic = alloc_wr(next_size, user_wr->num_sge);
2535 			if (!atomic) {
2536 				ret = -ENOMEM;
2537 				goto out_put;
2538 			}
2539 
2540 			atomic->remote_addr = user_wr->wr.atomic.remote_addr;
2541 			atomic->compare_add = user_wr->wr.atomic.compare_add;
2542 			atomic->swap = user_wr->wr.atomic.swap;
2543 			atomic->rkey = user_wr->wr.atomic.rkey;
2544 
2545 			next = &atomic->wr;
2546 		} else if (user_wr->opcode == IB_WR_SEND ||
2547 			   user_wr->opcode == IB_WR_SEND_WITH_IMM ||
2548 			   user_wr->opcode == IB_WR_SEND_WITH_INV) {
2549 			next_size = sizeof(*next);
2550 			next = alloc_wr(next_size, user_wr->num_sge);
2551 			if (!next) {
2552 				ret = -ENOMEM;
2553 				goto out_put;
2554 			}
2555 		} else {
2556 			ret = -EINVAL;
2557 			goto out_put;
2558 		}
2559 
2560 		if (user_wr->opcode == IB_WR_SEND_WITH_IMM ||
2561 		    user_wr->opcode == IB_WR_RDMA_WRITE_WITH_IMM) {
2562 			next->ex.imm_data =
2563 					(__be32 __force) user_wr->ex.imm_data;
2564 		} else if (user_wr->opcode == IB_WR_SEND_WITH_INV) {
2565 			next->ex.invalidate_rkey = user_wr->ex.invalidate_rkey;
2566 		}
2567 
2568 		if (!last)
2569 			wr = next;
2570 		else
2571 			last->next = next;
2572 		last = next;
2573 
2574 		next->next       = NULL;
2575 		next->wr_id      = user_wr->wr_id;
2576 		next->num_sge    = user_wr->num_sge;
2577 		next->opcode     = user_wr->opcode;
2578 		next->send_flags = user_wr->send_flags;
2579 
2580 		if (next->num_sge) {
2581 			next->sg_list = (void *) next +
2582 				ALIGN(next_size, sizeof(struct ib_sge));
2583 			if (copy_from_user(next->sg_list,
2584 					   buf + sizeof cmd +
2585 					   cmd.wr_count * cmd.wqe_size +
2586 					   sg_ind * sizeof (struct ib_sge),
2587 					   next->num_sge * sizeof (struct ib_sge))) {
2588 				ret = -EFAULT;
2589 				goto out_put;
2590 			}
2591 			sg_ind += next->num_sge;
2592 		} else
2593 			next->sg_list = NULL;
2594 	}
2595 
2596 	resp.bad_wr = 0;
2597 	ret = qp->device->post_send(qp->real_qp, wr, &bad_wr);
2598 	if (ret)
2599 		for (next = wr; next; next = next->next) {
2600 			++resp.bad_wr;
2601 			if (next == bad_wr)
2602 				break;
2603 		}
2604 
2605 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
2606 			 &resp, sizeof resp))
2607 		ret = -EFAULT;
2608 
2609 out_put:
2610 	put_qp_read(qp);
2611 
2612 	while (wr) {
2613 		if (is_ud && ud_wr(wr)->ah)
2614 			put_ah_read(ud_wr(wr)->ah);
2615 		next = wr->next;
2616 		kfree(wr);
2617 		wr = next;
2618 	}
2619 
2620 out:
2621 	kfree(user_wr);
2622 
2623 	return ret ? ret : in_len;
2624 }
2625 
2626 static struct ib_recv_wr *ib_uverbs_unmarshall_recv(const char __user *buf,
2627 						    int in_len,
2628 						    u32 wr_count,
2629 						    u32 sge_count,
2630 						    u32 wqe_size)
2631 {
2632 	struct ib_uverbs_recv_wr *user_wr;
2633 	struct ib_recv_wr        *wr = NULL, *last, *next;
2634 	int                       sg_ind;
2635 	int                       i;
2636 	int                       ret;
2637 
2638 	if (in_len < wqe_size * wr_count +
2639 	    sge_count * sizeof (struct ib_uverbs_sge))
2640 		return ERR_PTR(-EINVAL);
2641 
2642 	if (wqe_size < sizeof (struct ib_uverbs_recv_wr))
2643 		return ERR_PTR(-EINVAL);
2644 
2645 	user_wr = kmalloc(wqe_size, GFP_KERNEL);
2646 	if (!user_wr)
2647 		return ERR_PTR(-ENOMEM);
2648 
2649 	sg_ind = 0;
2650 	last = NULL;
2651 	for (i = 0; i < wr_count; ++i) {
2652 		if (copy_from_user(user_wr, buf + i * wqe_size,
2653 				   wqe_size)) {
2654 			ret = -EFAULT;
2655 			goto err;
2656 		}
2657 
2658 		if (user_wr->num_sge + sg_ind > sge_count) {
2659 			ret = -EINVAL;
2660 			goto err;
2661 		}
2662 
2663 		next = kmalloc(ALIGN(sizeof *next, sizeof (struct ib_sge)) +
2664 			       user_wr->num_sge * sizeof (struct ib_sge),
2665 			       GFP_KERNEL);
2666 		if (!next) {
2667 			ret = -ENOMEM;
2668 			goto err;
2669 		}
2670 
2671 		if (!last)
2672 			wr = next;
2673 		else
2674 			last->next = next;
2675 		last = next;
2676 
2677 		next->next       = NULL;
2678 		next->wr_id      = user_wr->wr_id;
2679 		next->num_sge    = user_wr->num_sge;
2680 
2681 		if (next->num_sge) {
2682 			next->sg_list = (void *) next +
2683 				ALIGN(sizeof *next, sizeof (struct ib_sge));
2684 			if (copy_from_user(next->sg_list,
2685 					   buf + wr_count * wqe_size +
2686 					   sg_ind * sizeof (struct ib_sge),
2687 					   next->num_sge * sizeof (struct ib_sge))) {
2688 				ret = -EFAULT;
2689 				goto err;
2690 			}
2691 			sg_ind += next->num_sge;
2692 		} else
2693 			next->sg_list = NULL;
2694 	}
2695 
2696 	kfree(user_wr);
2697 	return wr;
2698 
2699 err:
2700 	kfree(user_wr);
2701 
2702 	while (wr) {
2703 		next = wr->next;
2704 		kfree(wr);
2705 		wr = next;
2706 	}
2707 
2708 	return ERR_PTR(ret);
2709 }
2710 
2711 ssize_t ib_uverbs_post_recv(struct ib_uverbs_file *file,
2712 			    struct ib_device *ib_dev,
2713 			    const char __user *buf, int in_len,
2714 			    int out_len)
2715 {
2716 	struct ib_uverbs_post_recv      cmd;
2717 	struct ib_uverbs_post_recv_resp resp;
2718 	struct ib_recv_wr              *wr, *next, *bad_wr;
2719 	struct ib_qp                   *qp;
2720 	ssize_t                         ret = -EINVAL;
2721 
2722 	if (copy_from_user(&cmd, buf, sizeof cmd))
2723 		return -EFAULT;
2724 
2725 	wr = ib_uverbs_unmarshall_recv(buf + sizeof cmd,
2726 				       in_len - sizeof cmd, cmd.wr_count,
2727 				       cmd.sge_count, cmd.wqe_size);
2728 	if (IS_ERR(wr))
2729 		return PTR_ERR(wr);
2730 
2731 	qp = idr_read_qp(cmd.qp_handle, file->ucontext);
2732 	if (!qp)
2733 		goto out;
2734 
2735 	resp.bad_wr = 0;
2736 	ret = qp->device->post_recv(qp->real_qp, wr, &bad_wr);
2737 
2738 	put_qp_read(qp);
2739 
2740 	if (ret)
2741 		for (next = wr; next; next = next->next) {
2742 			++resp.bad_wr;
2743 			if (next == bad_wr)
2744 				break;
2745 		}
2746 
2747 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
2748 			 &resp, sizeof resp))
2749 		ret = -EFAULT;
2750 
2751 out:
2752 	while (wr) {
2753 		next = wr->next;
2754 		kfree(wr);
2755 		wr = next;
2756 	}
2757 
2758 	return ret ? ret : in_len;
2759 }
2760 
2761 ssize_t ib_uverbs_post_srq_recv(struct ib_uverbs_file *file,
2762 				struct ib_device *ib_dev,
2763 				const char __user *buf, int in_len,
2764 				int out_len)
2765 {
2766 	struct ib_uverbs_post_srq_recv      cmd;
2767 	struct ib_uverbs_post_srq_recv_resp resp;
2768 	struct ib_recv_wr                  *wr, *next, *bad_wr;
2769 	struct ib_srq                      *srq;
2770 	ssize_t                             ret = -EINVAL;
2771 
2772 	if (copy_from_user(&cmd, buf, sizeof cmd))
2773 		return -EFAULT;
2774 
2775 	wr = ib_uverbs_unmarshall_recv(buf + sizeof cmd,
2776 				       in_len - sizeof cmd, cmd.wr_count,
2777 				       cmd.sge_count, cmd.wqe_size);
2778 	if (IS_ERR(wr))
2779 		return PTR_ERR(wr);
2780 
2781 	srq = idr_read_srq(cmd.srq_handle, file->ucontext);
2782 	if (!srq)
2783 		goto out;
2784 
2785 	resp.bad_wr = 0;
2786 	ret = srq->device->post_srq_recv(srq, wr, &bad_wr);
2787 
2788 	put_srq_read(srq);
2789 
2790 	if (ret)
2791 		for (next = wr; next; next = next->next) {
2792 			++resp.bad_wr;
2793 			if (next == bad_wr)
2794 				break;
2795 		}
2796 
2797 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
2798 			 &resp, sizeof resp))
2799 		ret = -EFAULT;
2800 
2801 out:
2802 	while (wr) {
2803 		next = wr->next;
2804 		kfree(wr);
2805 		wr = next;
2806 	}
2807 
2808 	return ret ? ret : in_len;
2809 }
2810 
2811 ssize_t ib_uverbs_create_ah(struct ib_uverbs_file *file,
2812 			    struct ib_device *ib_dev,
2813 			    const char __user *buf, int in_len,
2814 			    int out_len)
2815 {
2816 	struct ib_uverbs_create_ah	 cmd;
2817 	struct ib_uverbs_create_ah_resp	 resp;
2818 	struct ib_uobject		*uobj;
2819 	struct ib_pd			*pd;
2820 	struct ib_ah			*ah;
2821 	struct ib_ah_attr		attr;
2822 	int ret;
2823 
2824 	if (out_len < sizeof resp)
2825 		return -ENOSPC;
2826 
2827 	if (copy_from_user(&cmd, buf, sizeof cmd))
2828 		return -EFAULT;
2829 
2830 	uobj = kmalloc(sizeof *uobj, GFP_KERNEL);
2831 	if (!uobj)
2832 		return -ENOMEM;
2833 
2834 	init_uobj(uobj, cmd.user_handle, file->ucontext, &ah_lock_class);
2835 	down_write(&uobj->mutex);
2836 
2837 	pd = idr_read_pd(cmd.pd_handle, file->ucontext);
2838 	if (!pd) {
2839 		ret = -EINVAL;
2840 		goto err;
2841 	}
2842 
2843 	attr.dlid 	       = cmd.attr.dlid;
2844 	attr.sl 	       = cmd.attr.sl;
2845 	attr.src_path_bits     = cmd.attr.src_path_bits;
2846 	attr.static_rate       = cmd.attr.static_rate;
2847 	attr.ah_flags          = cmd.attr.is_global ? IB_AH_GRH : 0;
2848 	attr.port_num 	       = cmd.attr.port_num;
2849 	attr.grh.flow_label    = cmd.attr.grh.flow_label;
2850 	attr.grh.sgid_index    = cmd.attr.grh.sgid_index;
2851 	attr.grh.hop_limit     = cmd.attr.grh.hop_limit;
2852 	attr.grh.traffic_class = cmd.attr.grh.traffic_class;
2853 	memset(&attr.dmac, 0, sizeof(attr.dmac));
2854 	memcpy(attr.grh.dgid.raw, cmd.attr.grh.dgid, 16);
2855 
2856 	ah = ib_create_ah(pd, &attr);
2857 	if (IS_ERR(ah)) {
2858 		ret = PTR_ERR(ah);
2859 		goto err_put;
2860 	}
2861 
2862 	ah->uobject  = uobj;
2863 	uobj->object = ah;
2864 
2865 	ret = idr_add_uobj(&ib_uverbs_ah_idr, uobj);
2866 	if (ret)
2867 		goto err_destroy;
2868 
2869 	resp.ah_handle = uobj->id;
2870 
2871 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
2872 			 &resp, sizeof resp)) {
2873 		ret = -EFAULT;
2874 		goto err_copy;
2875 	}
2876 
2877 	put_pd_read(pd);
2878 
2879 	mutex_lock(&file->mutex);
2880 	list_add_tail(&uobj->list, &file->ucontext->ah_list);
2881 	mutex_unlock(&file->mutex);
2882 
2883 	uobj->live = 1;
2884 
2885 	up_write(&uobj->mutex);
2886 
2887 	return in_len;
2888 
2889 err_copy:
2890 	idr_remove_uobj(&ib_uverbs_ah_idr, uobj);
2891 
2892 err_destroy:
2893 	ib_destroy_ah(ah);
2894 
2895 err_put:
2896 	put_pd_read(pd);
2897 
2898 err:
2899 	put_uobj_write(uobj);
2900 	return ret;
2901 }
2902 
2903 ssize_t ib_uverbs_destroy_ah(struct ib_uverbs_file *file,
2904 			     struct ib_device *ib_dev,
2905 			     const char __user *buf, int in_len, int out_len)
2906 {
2907 	struct ib_uverbs_destroy_ah cmd;
2908 	struct ib_ah		   *ah;
2909 	struct ib_uobject	   *uobj;
2910 	int			    ret;
2911 
2912 	if (copy_from_user(&cmd, buf, sizeof cmd))
2913 		return -EFAULT;
2914 
2915 	uobj = idr_write_uobj(&ib_uverbs_ah_idr, cmd.ah_handle, file->ucontext);
2916 	if (!uobj)
2917 		return -EINVAL;
2918 	ah = uobj->object;
2919 
2920 	ret = ib_destroy_ah(ah);
2921 	if (!ret)
2922 		uobj->live = 0;
2923 
2924 	put_uobj_write(uobj);
2925 
2926 	if (ret)
2927 		return ret;
2928 
2929 	idr_remove_uobj(&ib_uverbs_ah_idr, uobj);
2930 
2931 	mutex_lock(&file->mutex);
2932 	list_del(&uobj->list);
2933 	mutex_unlock(&file->mutex);
2934 
2935 	put_uobj(uobj);
2936 
2937 	return in_len;
2938 }
2939 
2940 ssize_t ib_uverbs_attach_mcast(struct ib_uverbs_file *file,
2941 			       struct ib_device *ib_dev,
2942 			       const char __user *buf, int in_len,
2943 			       int out_len)
2944 {
2945 	struct ib_uverbs_attach_mcast cmd;
2946 	struct ib_qp                 *qp;
2947 	struct ib_uqp_object         *obj;
2948 	struct ib_uverbs_mcast_entry *mcast;
2949 	int                           ret;
2950 
2951 	if (copy_from_user(&cmd, buf, sizeof cmd))
2952 		return -EFAULT;
2953 
2954 	qp = idr_write_qp(cmd.qp_handle, file->ucontext);
2955 	if (!qp)
2956 		return -EINVAL;
2957 
2958 	obj = container_of(qp->uobject, struct ib_uqp_object, uevent.uobject);
2959 
2960 	list_for_each_entry(mcast, &obj->mcast_list, list)
2961 		if (cmd.mlid == mcast->lid &&
2962 		    !memcmp(cmd.gid, mcast->gid.raw, sizeof mcast->gid.raw)) {
2963 			ret = 0;
2964 			goto out_put;
2965 		}
2966 
2967 	mcast = kmalloc(sizeof *mcast, GFP_KERNEL);
2968 	if (!mcast) {
2969 		ret = -ENOMEM;
2970 		goto out_put;
2971 	}
2972 
2973 	mcast->lid = cmd.mlid;
2974 	memcpy(mcast->gid.raw, cmd.gid, sizeof mcast->gid.raw);
2975 
2976 	ret = ib_attach_mcast(qp, &mcast->gid, cmd.mlid);
2977 	if (!ret)
2978 		list_add_tail(&mcast->list, &obj->mcast_list);
2979 	else
2980 		kfree(mcast);
2981 
2982 out_put:
2983 	put_qp_write(qp);
2984 
2985 	return ret ? ret : in_len;
2986 }
2987 
2988 ssize_t ib_uverbs_detach_mcast(struct ib_uverbs_file *file,
2989 			       struct ib_device *ib_dev,
2990 			       const char __user *buf, int in_len,
2991 			       int out_len)
2992 {
2993 	struct ib_uverbs_detach_mcast cmd;
2994 	struct ib_uqp_object         *obj;
2995 	struct ib_qp                 *qp;
2996 	struct ib_uverbs_mcast_entry *mcast;
2997 	int                           ret = -EINVAL;
2998 
2999 	if (copy_from_user(&cmd, buf, sizeof cmd))
3000 		return -EFAULT;
3001 
3002 	qp = idr_write_qp(cmd.qp_handle, file->ucontext);
3003 	if (!qp)
3004 		return -EINVAL;
3005 
3006 	ret = ib_detach_mcast(qp, (union ib_gid *) cmd.gid, cmd.mlid);
3007 	if (ret)
3008 		goto out_put;
3009 
3010 	obj = container_of(qp->uobject, struct ib_uqp_object, uevent.uobject);
3011 
3012 	list_for_each_entry(mcast, &obj->mcast_list, list)
3013 		if (cmd.mlid == mcast->lid &&
3014 		    !memcmp(cmd.gid, mcast->gid.raw, sizeof mcast->gid.raw)) {
3015 			list_del(&mcast->list);
3016 			kfree(mcast);
3017 			break;
3018 		}
3019 
3020 out_put:
3021 	put_qp_write(qp);
3022 
3023 	return ret ? ret : in_len;
3024 }
3025 
3026 static int kern_spec_to_ib_spec(struct ib_uverbs_flow_spec *kern_spec,
3027 				union ib_flow_spec *ib_spec)
3028 {
3029 	if (kern_spec->reserved)
3030 		return -EINVAL;
3031 
3032 	ib_spec->type = kern_spec->type;
3033 
3034 	switch (ib_spec->type) {
3035 	case IB_FLOW_SPEC_ETH:
3036 		ib_spec->eth.size = sizeof(struct ib_flow_spec_eth);
3037 		if (ib_spec->eth.size != kern_spec->eth.size)
3038 			return -EINVAL;
3039 		memcpy(&ib_spec->eth.val, &kern_spec->eth.val,
3040 		       sizeof(struct ib_flow_eth_filter));
3041 		memcpy(&ib_spec->eth.mask, &kern_spec->eth.mask,
3042 		       sizeof(struct ib_flow_eth_filter));
3043 		break;
3044 	case IB_FLOW_SPEC_IPV4:
3045 		ib_spec->ipv4.size = sizeof(struct ib_flow_spec_ipv4);
3046 		if (ib_spec->ipv4.size != kern_spec->ipv4.size)
3047 			return -EINVAL;
3048 		memcpy(&ib_spec->ipv4.val, &kern_spec->ipv4.val,
3049 		       sizeof(struct ib_flow_ipv4_filter));
3050 		memcpy(&ib_spec->ipv4.mask, &kern_spec->ipv4.mask,
3051 		       sizeof(struct ib_flow_ipv4_filter));
3052 		break;
3053 	case IB_FLOW_SPEC_TCP:
3054 	case IB_FLOW_SPEC_UDP:
3055 		ib_spec->tcp_udp.size = sizeof(struct ib_flow_spec_tcp_udp);
3056 		if (ib_spec->tcp_udp.size != kern_spec->tcp_udp.size)
3057 			return -EINVAL;
3058 		memcpy(&ib_spec->tcp_udp.val, &kern_spec->tcp_udp.val,
3059 		       sizeof(struct ib_flow_tcp_udp_filter));
3060 		memcpy(&ib_spec->tcp_udp.mask, &kern_spec->tcp_udp.mask,
3061 		       sizeof(struct ib_flow_tcp_udp_filter));
3062 		break;
3063 	default:
3064 		return -EINVAL;
3065 	}
3066 	return 0;
3067 }
3068 
3069 int ib_uverbs_ex_create_flow(struct ib_uverbs_file *file,
3070 			     struct ib_device *ib_dev,
3071 			     struct ib_udata *ucore,
3072 			     struct ib_udata *uhw)
3073 {
3074 	struct ib_uverbs_create_flow	  cmd;
3075 	struct ib_uverbs_create_flow_resp resp;
3076 	struct ib_uobject		  *uobj;
3077 	struct ib_flow			  *flow_id;
3078 	struct ib_uverbs_flow_attr	  *kern_flow_attr;
3079 	struct ib_flow_attr		  *flow_attr;
3080 	struct ib_qp			  *qp;
3081 	int err = 0;
3082 	void *kern_spec;
3083 	void *ib_spec;
3084 	int i;
3085 
3086 	if (ucore->inlen < sizeof(cmd))
3087 		return -EINVAL;
3088 
3089 	if (ucore->outlen < sizeof(resp))
3090 		return -ENOSPC;
3091 
3092 	err = ib_copy_from_udata(&cmd, ucore, sizeof(cmd));
3093 	if (err)
3094 		return err;
3095 
3096 	ucore->inbuf += sizeof(cmd);
3097 	ucore->inlen -= sizeof(cmd);
3098 
3099 	if (cmd.comp_mask)
3100 		return -EINVAL;
3101 
3102 	if ((cmd.flow_attr.type == IB_FLOW_ATTR_SNIFFER &&
3103 	     !capable(CAP_NET_ADMIN)) || !capable(CAP_NET_RAW))
3104 		return -EPERM;
3105 
3106 	if (cmd.flow_attr.num_of_specs > IB_FLOW_SPEC_SUPPORT_LAYERS)
3107 		return -EINVAL;
3108 
3109 	if (cmd.flow_attr.size > ucore->inlen ||
3110 	    cmd.flow_attr.size >
3111 	    (cmd.flow_attr.num_of_specs * sizeof(struct ib_uverbs_flow_spec)))
3112 		return -EINVAL;
3113 
3114 	if (cmd.flow_attr.reserved[0] ||
3115 	    cmd.flow_attr.reserved[1])
3116 		return -EINVAL;
3117 
3118 	if (cmd.flow_attr.num_of_specs) {
3119 		kern_flow_attr = kmalloc(sizeof(*kern_flow_attr) + cmd.flow_attr.size,
3120 					 GFP_KERNEL);
3121 		if (!kern_flow_attr)
3122 			return -ENOMEM;
3123 
3124 		memcpy(kern_flow_attr, &cmd.flow_attr, sizeof(*kern_flow_attr));
3125 		err = ib_copy_from_udata(kern_flow_attr + 1, ucore,
3126 					 cmd.flow_attr.size);
3127 		if (err)
3128 			goto err_free_attr;
3129 	} else {
3130 		kern_flow_attr = &cmd.flow_attr;
3131 	}
3132 
3133 	uobj = kmalloc(sizeof(*uobj), GFP_KERNEL);
3134 	if (!uobj) {
3135 		err = -ENOMEM;
3136 		goto err_free_attr;
3137 	}
3138 	init_uobj(uobj, 0, file->ucontext, &rule_lock_class);
3139 	down_write(&uobj->mutex);
3140 
3141 	qp = idr_read_qp(cmd.qp_handle, file->ucontext);
3142 	if (!qp) {
3143 		err = -EINVAL;
3144 		goto err_uobj;
3145 	}
3146 
3147 	flow_attr = kmalloc(sizeof(*flow_attr) + cmd.flow_attr.size, GFP_KERNEL);
3148 	if (!flow_attr) {
3149 		err = -ENOMEM;
3150 		goto err_put;
3151 	}
3152 
3153 	flow_attr->type = kern_flow_attr->type;
3154 	flow_attr->priority = kern_flow_attr->priority;
3155 	flow_attr->num_of_specs = kern_flow_attr->num_of_specs;
3156 	flow_attr->port = kern_flow_attr->port;
3157 	flow_attr->flags = kern_flow_attr->flags;
3158 	flow_attr->size = sizeof(*flow_attr);
3159 
3160 	kern_spec = kern_flow_attr + 1;
3161 	ib_spec = flow_attr + 1;
3162 	for (i = 0; i < flow_attr->num_of_specs &&
3163 	     cmd.flow_attr.size > offsetof(struct ib_uverbs_flow_spec, reserved) &&
3164 	     cmd.flow_attr.size >=
3165 	     ((struct ib_uverbs_flow_spec *)kern_spec)->size; i++) {
3166 		err = kern_spec_to_ib_spec(kern_spec, ib_spec);
3167 		if (err)
3168 			goto err_free;
3169 		flow_attr->size +=
3170 			((union ib_flow_spec *) ib_spec)->size;
3171 		cmd.flow_attr.size -= ((struct ib_uverbs_flow_spec *)kern_spec)->size;
3172 		kern_spec += ((struct ib_uverbs_flow_spec *) kern_spec)->size;
3173 		ib_spec += ((union ib_flow_spec *) ib_spec)->size;
3174 	}
3175 	if (cmd.flow_attr.size || (i != flow_attr->num_of_specs)) {
3176 		pr_warn("create flow failed, flow %d: %d bytes left from uverb cmd\n",
3177 			i, cmd.flow_attr.size);
3178 		err = -EINVAL;
3179 		goto err_free;
3180 	}
3181 	flow_id = ib_create_flow(qp, flow_attr, IB_FLOW_DOMAIN_USER);
3182 	if (IS_ERR(flow_id)) {
3183 		err = PTR_ERR(flow_id);
3184 		goto err_free;
3185 	}
3186 	flow_id->qp = qp;
3187 	flow_id->uobject = uobj;
3188 	uobj->object = flow_id;
3189 
3190 	err = idr_add_uobj(&ib_uverbs_rule_idr, uobj);
3191 	if (err)
3192 		goto destroy_flow;
3193 
3194 	memset(&resp, 0, sizeof(resp));
3195 	resp.flow_handle = uobj->id;
3196 
3197 	err = ib_copy_to_udata(ucore,
3198 			       &resp, sizeof(resp));
3199 	if (err)
3200 		goto err_copy;
3201 
3202 	put_qp_read(qp);
3203 	mutex_lock(&file->mutex);
3204 	list_add_tail(&uobj->list, &file->ucontext->rule_list);
3205 	mutex_unlock(&file->mutex);
3206 
3207 	uobj->live = 1;
3208 
3209 	up_write(&uobj->mutex);
3210 	kfree(flow_attr);
3211 	if (cmd.flow_attr.num_of_specs)
3212 		kfree(kern_flow_attr);
3213 	return 0;
3214 err_copy:
3215 	idr_remove_uobj(&ib_uverbs_rule_idr, uobj);
3216 destroy_flow:
3217 	ib_destroy_flow(flow_id);
3218 err_free:
3219 	kfree(flow_attr);
3220 err_put:
3221 	put_qp_read(qp);
3222 err_uobj:
3223 	put_uobj_write(uobj);
3224 err_free_attr:
3225 	if (cmd.flow_attr.num_of_specs)
3226 		kfree(kern_flow_attr);
3227 	return err;
3228 }
3229 
3230 int ib_uverbs_ex_destroy_flow(struct ib_uverbs_file *file,
3231 			      struct ib_device *ib_dev,
3232 			      struct ib_udata *ucore,
3233 			      struct ib_udata *uhw)
3234 {
3235 	struct ib_uverbs_destroy_flow	cmd;
3236 	struct ib_flow			*flow_id;
3237 	struct ib_uobject		*uobj;
3238 	int				ret;
3239 
3240 	if (ucore->inlen < sizeof(cmd))
3241 		return -EINVAL;
3242 
3243 	ret = ib_copy_from_udata(&cmd, ucore, sizeof(cmd));
3244 	if (ret)
3245 		return ret;
3246 
3247 	if (cmd.comp_mask)
3248 		return -EINVAL;
3249 
3250 	uobj = idr_write_uobj(&ib_uverbs_rule_idr, cmd.flow_handle,
3251 			      file->ucontext);
3252 	if (!uobj)
3253 		return -EINVAL;
3254 	flow_id = uobj->object;
3255 
3256 	ret = ib_destroy_flow(flow_id);
3257 	if (!ret)
3258 		uobj->live = 0;
3259 
3260 	put_uobj_write(uobj);
3261 
3262 	idr_remove_uobj(&ib_uverbs_rule_idr, uobj);
3263 
3264 	mutex_lock(&file->mutex);
3265 	list_del(&uobj->list);
3266 	mutex_unlock(&file->mutex);
3267 
3268 	put_uobj(uobj);
3269 
3270 	return ret;
3271 }
3272 
3273 static int __uverbs_create_xsrq(struct ib_uverbs_file *file,
3274 				struct ib_device *ib_dev,
3275 				struct ib_uverbs_create_xsrq *cmd,
3276 				struct ib_udata *udata)
3277 {
3278 	struct ib_uverbs_create_srq_resp resp;
3279 	struct ib_usrq_object           *obj;
3280 	struct ib_pd                    *pd;
3281 	struct ib_srq                   *srq;
3282 	struct ib_uobject               *uninitialized_var(xrcd_uobj);
3283 	struct ib_srq_init_attr          attr;
3284 	int ret;
3285 
3286 	obj = kmalloc(sizeof *obj, GFP_KERNEL);
3287 	if (!obj)
3288 		return -ENOMEM;
3289 
3290 	init_uobj(&obj->uevent.uobject, cmd->user_handle, file->ucontext, &srq_lock_class);
3291 	down_write(&obj->uevent.uobject.mutex);
3292 
3293 	if (cmd->srq_type == IB_SRQT_XRC) {
3294 		attr.ext.xrc.xrcd  = idr_read_xrcd(cmd->xrcd_handle, file->ucontext, &xrcd_uobj);
3295 		if (!attr.ext.xrc.xrcd) {
3296 			ret = -EINVAL;
3297 			goto err;
3298 		}
3299 
3300 		obj->uxrcd = container_of(xrcd_uobj, struct ib_uxrcd_object, uobject);
3301 		atomic_inc(&obj->uxrcd->refcnt);
3302 
3303 		attr.ext.xrc.cq  = idr_read_cq(cmd->cq_handle, file->ucontext, 0);
3304 		if (!attr.ext.xrc.cq) {
3305 			ret = -EINVAL;
3306 			goto err_put_xrcd;
3307 		}
3308 	}
3309 
3310 	pd  = idr_read_pd(cmd->pd_handle, file->ucontext);
3311 	if (!pd) {
3312 		ret = -EINVAL;
3313 		goto err_put_cq;
3314 	}
3315 
3316 	attr.event_handler  = ib_uverbs_srq_event_handler;
3317 	attr.srq_context    = file;
3318 	attr.srq_type       = cmd->srq_type;
3319 	attr.attr.max_wr    = cmd->max_wr;
3320 	attr.attr.max_sge   = cmd->max_sge;
3321 	attr.attr.srq_limit = cmd->srq_limit;
3322 
3323 	obj->uevent.events_reported = 0;
3324 	INIT_LIST_HEAD(&obj->uevent.event_list);
3325 
3326 	srq = pd->device->create_srq(pd, &attr, udata);
3327 	if (IS_ERR(srq)) {
3328 		ret = PTR_ERR(srq);
3329 		goto err_put;
3330 	}
3331 
3332 	srq->device        = pd->device;
3333 	srq->pd            = pd;
3334 	srq->srq_type	   = cmd->srq_type;
3335 	srq->uobject       = &obj->uevent.uobject;
3336 	srq->event_handler = attr.event_handler;
3337 	srq->srq_context   = attr.srq_context;
3338 
3339 	if (cmd->srq_type == IB_SRQT_XRC) {
3340 		srq->ext.xrc.cq   = attr.ext.xrc.cq;
3341 		srq->ext.xrc.xrcd = attr.ext.xrc.xrcd;
3342 		atomic_inc(&attr.ext.xrc.cq->usecnt);
3343 		atomic_inc(&attr.ext.xrc.xrcd->usecnt);
3344 	}
3345 
3346 	atomic_inc(&pd->usecnt);
3347 	atomic_set(&srq->usecnt, 0);
3348 
3349 	obj->uevent.uobject.object = srq;
3350 	ret = idr_add_uobj(&ib_uverbs_srq_idr, &obj->uevent.uobject);
3351 	if (ret)
3352 		goto err_destroy;
3353 
3354 	memset(&resp, 0, sizeof resp);
3355 	resp.srq_handle = obj->uevent.uobject.id;
3356 	resp.max_wr     = attr.attr.max_wr;
3357 	resp.max_sge    = attr.attr.max_sge;
3358 	if (cmd->srq_type == IB_SRQT_XRC)
3359 		resp.srqn = srq->ext.xrc.srq_num;
3360 
3361 	if (copy_to_user((void __user *) (unsigned long) cmd->response,
3362 			 &resp, sizeof resp)) {
3363 		ret = -EFAULT;
3364 		goto err_copy;
3365 	}
3366 
3367 	if (cmd->srq_type == IB_SRQT_XRC) {
3368 		put_uobj_read(xrcd_uobj);
3369 		put_cq_read(attr.ext.xrc.cq);
3370 	}
3371 	put_pd_read(pd);
3372 
3373 	mutex_lock(&file->mutex);
3374 	list_add_tail(&obj->uevent.uobject.list, &file->ucontext->srq_list);
3375 	mutex_unlock(&file->mutex);
3376 
3377 	obj->uevent.uobject.live = 1;
3378 
3379 	up_write(&obj->uevent.uobject.mutex);
3380 
3381 	return 0;
3382 
3383 err_copy:
3384 	idr_remove_uobj(&ib_uverbs_srq_idr, &obj->uevent.uobject);
3385 
3386 err_destroy:
3387 	ib_destroy_srq(srq);
3388 
3389 err_put:
3390 	put_pd_read(pd);
3391 
3392 err_put_cq:
3393 	if (cmd->srq_type == IB_SRQT_XRC)
3394 		put_cq_read(attr.ext.xrc.cq);
3395 
3396 err_put_xrcd:
3397 	if (cmd->srq_type == IB_SRQT_XRC) {
3398 		atomic_dec(&obj->uxrcd->refcnt);
3399 		put_uobj_read(xrcd_uobj);
3400 	}
3401 
3402 err:
3403 	put_uobj_write(&obj->uevent.uobject);
3404 	return ret;
3405 }
3406 
3407 ssize_t ib_uverbs_create_srq(struct ib_uverbs_file *file,
3408 			     struct ib_device *ib_dev,
3409 			     const char __user *buf, int in_len,
3410 			     int out_len)
3411 {
3412 	struct ib_uverbs_create_srq      cmd;
3413 	struct ib_uverbs_create_xsrq     xcmd;
3414 	struct ib_uverbs_create_srq_resp resp;
3415 	struct ib_udata                  udata;
3416 	int ret;
3417 
3418 	if (out_len < sizeof resp)
3419 		return -ENOSPC;
3420 
3421 	if (copy_from_user(&cmd, buf, sizeof cmd))
3422 		return -EFAULT;
3423 
3424 	xcmd.response	 = cmd.response;
3425 	xcmd.user_handle = cmd.user_handle;
3426 	xcmd.srq_type	 = IB_SRQT_BASIC;
3427 	xcmd.pd_handle	 = cmd.pd_handle;
3428 	xcmd.max_wr	 = cmd.max_wr;
3429 	xcmd.max_sge	 = cmd.max_sge;
3430 	xcmd.srq_limit	 = cmd.srq_limit;
3431 
3432 	INIT_UDATA(&udata, buf + sizeof cmd,
3433 		   (unsigned long) cmd.response + sizeof resp,
3434 		   in_len - sizeof cmd, out_len - sizeof resp);
3435 
3436 	ret = __uverbs_create_xsrq(file, ib_dev, &xcmd, &udata);
3437 	if (ret)
3438 		return ret;
3439 
3440 	return in_len;
3441 }
3442 
3443 ssize_t ib_uverbs_create_xsrq(struct ib_uverbs_file *file,
3444 			      struct ib_device *ib_dev,
3445 			      const char __user *buf, int in_len, int out_len)
3446 {
3447 	struct ib_uverbs_create_xsrq     cmd;
3448 	struct ib_uverbs_create_srq_resp resp;
3449 	struct ib_udata                  udata;
3450 	int ret;
3451 
3452 	if (out_len < sizeof resp)
3453 		return -ENOSPC;
3454 
3455 	if (copy_from_user(&cmd, buf, sizeof cmd))
3456 		return -EFAULT;
3457 
3458 	INIT_UDATA(&udata, buf + sizeof cmd,
3459 		   (unsigned long) cmd.response + sizeof resp,
3460 		   in_len - sizeof cmd, out_len - sizeof resp);
3461 
3462 	ret = __uverbs_create_xsrq(file, ib_dev, &cmd, &udata);
3463 	if (ret)
3464 		return ret;
3465 
3466 	return in_len;
3467 }
3468 
3469 ssize_t ib_uverbs_modify_srq(struct ib_uverbs_file *file,
3470 			     struct ib_device *ib_dev,
3471 			     const char __user *buf, int in_len,
3472 			     int out_len)
3473 {
3474 	struct ib_uverbs_modify_srq cmd;
3475 	struct ib_udata             udata;
3476 	struct ib_srq              *srq;
3477 	struct ib_srq_attr          attr;
3478 	int                         ret;
3479 
3480 	if (copy_from_user(&cmd, buf, sizeof cmd))
3481 		return -EFAULT;
3482 
3483 	INIT_UDATA(&udata, buf + sizeof cmd, NULL, in_len - sizeof cmd,
3484 		   out_len);
3485 
3486 	srq = idr_read_srq(cmd.srq_handle, file->ucontext);
3487 	if (!srq)
3488 		return -EINVAL;
3489 
3490 	attr.max_wr    = cmd.max_wr;
3491 	attr.srq_limit = cmd.srq_limit;
3492 
3493 	ret = srq->device->modify_srq(srq, &attr, cmd.attr_mask, &udata);
3494 
3495 	put_srq_read(srq);
3496 
3497 	return ret ? ret : in_len;
3498 }
3499 
3500 ssize_t ib_uverbs_query_srq(struct ib_uverbs_file *file,
3501 			    struct ib_device *ib_dev,
3502 			    const char __user *buf,
3503 			    int in_len, int out_len)
3504 {
3505 	struct ib_uverbs_query_srq      cmd;
3506 	struct ib_uverbs_query_srq_resp resp;
3507 	struct ib_srq_attr              attr;
3508 	struct ib_srq                   *srq;
3509 	int                             ret;
3510 
3511 	if (out_len < sizeof resp)
3512 		return -ENOSPC;
3513 
3514 	if (copy_from_user(&cmd, buf, sizeof cmd))
3515 		return -EFAULT;
3516 
3517 	srq = idr_read_srq(cmd.srq_handle, file->ucontext);
3518 	if (!srq)
3519 		return -EINVAL;
3520 
3521 	ret = ib_query_srq(srq, &attr);
3522 
3523 	put_srq_read(srq);
3524 
3525 	if (ret)
3526 		return ret;
3527 
3528 	memset(&resp, 0, sizeof resp);
3529 
3530 	resp.max_wr    = attr.max_wr;
3531 	resp.max_sge   = attr.max_sge;
3532 	resp.srq_limit = attr.srq_limit;
3533 
3534 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
3535 			 &resp, sizeof resp))
3536 		return -EFAULT;
3537 
3538 	return in_len;
3539 }
3540 
3541 ssize_t ib_uverbs_destroy_srq(struct ib_uverbs_file *file,
3542 			      struct ib_device *ib_dev,
3543 			      const char __user *buf, int in_len,
3544 			      int out_len)
3545 {
3546 	struct ib_uverbs_destroy_srq      cmd;
3547 	struct ib_uverbs_destroy_srq_resp resp;
3548 	struct ib_uobject		 *uobj;
3549 	struct ib_srq               	 *srq;
3550 	struct ib_uevent_object        	 *obj;
3551 	int                         	  ret = -EINVAL;
3552 	struct ib_usrq_object		 *us;
3553 	enum ib_srq_type		  srq_type;
3554 
3555 	if (copy_from_user(&cmd, buf, sizeof cmd))
3556 		return -EFAULT;
3557 
3558 	uobj = idr_write_uobj(&ib_uverbs_srq_idr, cmd.srq_handle, file->ucontext);
3559 	if (!uobj)
3560 		return -EINVAL;
3561 	srq = uobj->object;
3562 	obj = container_of(uobj, struct ib_uevent_object, uobject);
3563 	srq_type = srq->srq_type;
3564 
3565 	ret = ib_destroy_srq(srq);
3566 	if (!ret)
3567 		uobj->live = 0;
3568 
3569 	put_uobj_write(uobj);
3570 
3571 	if (ret)
3572 		return ret;
3573 
3574 	if (srq_type == IB_SRQT_XRC) {
3575 		us = container_of(obj, struct ib_usrq_object, uevent);
3576 		atomic_dec(&us->uxrcd->refcnt);
3577 	}
3578 
3579 	idr_remove_uobj(&ib_uverbs_srq_idr, uobj);
3580 
3581 	mutex_lock(&file->mutex);
3582 	list_del(&uobj->list);
3583 	mutex_unlock(&file->mutex);
3584 
3585 	ib_uverbs_release_uevent(file, obj);
3586 
3587 	memset(&resp, 0, sizeof resp);
3588 	resp.events_reported = obj->events_reported;
3589 
3590 	put_uobj(uobj);
3591 
3592 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
3593 			 &resp, sizeof resp))
3594 		ret = -EFAULT;
3595 
3596 	return ret ? ret : in_len;
3597 }
3598 
3599 int ib_uverbs_ex_query_device(struct ib_uverbs_file *file,
3600 			      struct ib_device *ib_dev,
3601 			      struct ib_udata *ucore,
3602 			      struct ib_udata *uhw)
3603 {
3604 	struct ib_uverbs_ex_query_device_resp resp;
3605 	struct ib_uverbs_ex_query_device  cmd;
3606 	struct ib_device_attr attr;
3607 	int err;
3608 
3609 	if (ucore->inlen < sizeof(cmd))
3610 		return -EINVAL;
3611 
3612 	err = ib_copy_from_udata(&cmd, ucore, sizeof(cmd));
3613 	if (err)
3614 		return err;
3615 
3616 	if (cmd.comp_mask)
3617 		return -EINVAL;
3618 
3619 	if (cmd.reserved)
3620 		return -EINVAL;
3621 
3622 	resp.response_length = offsetof(typeof(resp), odp_caps);
3623 
3624 	if (ucore->outlen < resp.response_length)
3625 		return -ENOSPC;
3626 
3627 	memset(&attr, 0, sizeof(attr));
3628 
3629 	err = ib_dev->query_device(ib_dev, &attr, uhw);
3630 	if (err)
3631 		return err;
3632 
3633 	copy_query_dev_fields(file, ib_dev, &resp.base, &attr);
3634 	resp.comp_mask = 0;
3635 
3636 	if (ucore->outlen < resp.response_length + sizeof(resp.odp_caps))
3637 		goto end;
3638 
3639 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
3640 	resp.odp_caps.general_caps = attr.odp_caps.general_caps;
3641 	resp.odp_caps.per_transport_caps.rc_odp_caps =
3642 		attr.odp_caps.per_transport_caps.rc_odp_caps;
3643 	resp.odp_caps.per_transport_caps.uc_odp_caps =
3644 		attr.odp_caps.per_transport_caps.uc_odp_caps;
3645 	resp.odp_caps.per_transport_caps.ud_odp_caps =
3646 		attr.odp_caps.per_transport_caps.ud_odp_caps;
3647 	resp.odp_caps.reserved = 0;
3648 #else
3649 	memset(&resp.odp_caps, 0, sizeof(resp.odp_caps));
3650 #endif
3651 	resp.response_length += sizeof(resp.odp_caps);
3652 
3653 	if (ucore->outlen < resp.response_length + sizeof(resp.timestamp_mask))
3654 		goto end;
3655 
3656 	resp.timestamp_mask = attr.timestamp_mask;
3657 	resp.response_length += sizeof(resp.timestamp_mask);
3658 
3659 	if (ucore->outlen < resp.response_length + sizeof(resp.hca_core_clock))
3660 		goto end;
3661 
3662 	resp.hca_core_clock = attr.hca_core_clock;
3663 	resp.response_length += sizeof(resp.hca_core_clock);
3664 
3665 end:
3666 	err = ib_copy_to_udata(ucore, &resp, resp.response_length);
3667 	if (err)
3668 		return err;
3669 
3670 	return 0;
3671 }
3672