xref: /openbmc/linux/drivers/infiniband/core/uverbs_cmd.c (revision f79e4d5f92a129a1159c973735007d4ddc8541f3)
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 <linux/uaccess.h>
42 
43 #include <rdma/uverbs_types.h>
44 #include <rdma/uverbs_std_types.h>
45 #include "rdma_core.h"
46 
47 #include "uverbs.h"
48 #include "core_priv.h"
49 
50 static struct ib_uverbs_completion_event_file *
51 ib_uverbs_lookup_comp_file(int fd, struct ib_ucontext *context)
52 {
53 	struct ib_uobject *uobj = uobj_get_read(UVERBS_OBJECT_COMP_CHANNEL,
54 						fd, context);
55 	struct ib_uobject_file *uobj_file;
56 
57 	if (IS_ERR(uobj))
58 		return (void *)uobj;
59 
60 	uverbs_uobject_get(uobj);
61 	uobj_put_read(uobj);
62 
63 	uobj_file = container_of(uobj, struct ib_uobject_file, uobj);
64 	return container_of(uobj_file, struct ib_uverbs_completion_event_file,
65 			    uobj_file);
66 }
67 
68 ssize_t ib_uverbs_get_context(struct ib_uverbs_file *file,
69 			      struct ib_device *ib_dev,
70 			      const char __user *buf,
71 			      int in_len, int out_len)
72 {
73 	struct ib_uverbs_get_context      cmd;
74 	struct ib_uverbs_get_context_resp resp;
75 	struct ib_udata                   udata;
76 	struct ib_ucontext		 *ucontext;
77 	struct file			 *filp;
78 	struct ib_rdmacg_object		 cg_obj;
79 	int ret;
80 
81 	if (out_len < sizeof resp)
82 		return -ENOSPC;
83 
84 	if (copy_from_user(&cmd, buf, sizeof cmd))
85 		return -EFAULT;
86 
87 	mutex_lock(&file->mutex);
88 
89 	if (file->ucontext) {
90 		ret = -EINVAL;
91 		goto err;
92 	}
93 
94 	ib_uverbs_init_udata(&udata, buf + sizeof(cmd),
95 		   u64_to_user_ptr(cmd.response) + sizeof(resp),
96 		   in_len - sizeof(cmd) - sizeof(struct ib_uverbs_cmd_hdr),
97 		   out_len - sizeof(resp));
98 
99 	ret = ib_rdmacg_try_charge(&cg_obj, ib_dev, RDMACG_RESOURCE_HCA_HANDLE);
100 	if (ret)
101 		goto err;
102 
103 	ucontext = ib_dev->alloc_ucontext(ib_dev, &udata);
104 	if (IS_ERR(ucontext)) {
105 		ret = PTR_ERR(ucontext);
106 		goto err_alloc;
107 	}
108 
109 	ucontext->device = ib_dev;
110 	ucontext->cg_obj = cg_obj;
111 	/* ufile is required when some objects are released */
112 	ucontext->ufile = file;
113 	uverbs_initialize_ucontext(ucontext);
114 
115 	rcu_read_lock();
116 	ucontext->tgid = get_task_pid(current->group_leader, PIDTYPE_PID);
117 	rcu_read_unlock();
118 	ucontext->closing = 0;
119 
120 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
121 	ucontext->umem_tree = RB_ROOT_CACHED;
122 	init_rwsem(&ucontext->umem_rwsem);
123 	ucontext->odp_mrs_count = 0;
124 	INIT_LIST_HEAD(&ucontext->no_private_counters);
125 
126 	if (!(ib_dev->attrs.device_cap_flags & IB_DEVICE_ON_DEMAND_PAGING))
127 		ucontext->invalidate_range = NULL;
128 
129 #endif
130 
131 	resp.num_comp_vectors = file->device->num_comp_vectors;
132 
133 	ret = get_unused_fd_flags(O_CLOEXEC);
134 	if (ret < 0)
135 		goto err_free;
136 	resp.async_fd = ret;
137 
138 	filp = ib_uverbs_alloc_async_event_file(file, ib_dev);
139 	if (IS_ERR(filp)) {
140 		ret = PTR_ERR(filp);
141 		goto err_fd;
142 	}
143 
144 	if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof resp)) {
145 		ret = -EFAULT;
146 		goto err_file;
147 	}
148 
149 	file->ucontext = ucontext;
150 
151 	fd_install(resp.async_fd, filp);
152 
153 	mutex_unlock(&file->mutex);
154 
155 	return in_len;
156 
157 err_file:
158 	ib_uverbs_free_async_event_file(file);
159 	fput(filp);
160 
161 err_fd:
162 	put_unused_fd(resp.async_fd);
163 
164 err_free:
165 	put_pid(ucontext->tgid);
166 	ib_dev->dealloc_ucontext(ucontext);
167 
168 err_alloc:
169 	ib_rdmacg_uncharge(&cg_obj, ib_dev, RDMACG_RESOURCE_HCA_HANDLE);
170 
171 err:
172 	mutex_unlock(&file->mutex);
173 	return ret;
174 }
175 
176 static void copy_query_dev_fields(struct ib_uverbs_file *file,
177 				  struct ib_device *ib_dev,
178 				  struct ib_uverbs_query_device_resp *resp,
179 				  struct ib_device_attr *attr)
180 {
181 	resp->fw_ver		= attr->fw_ver;
182 	resp->node_guid		= ib_dev->node_guid;
183 	resp->sys_image_guid	= attr->sys_image_guid;
184 	resp->max_mr_size	= attr->max_mr_size;
185 	resp->page_size_cap	= attr->page_size_cap;
186 	resp->vendor_id		= attr->vendor_id;
187 	resp->vendor_part_id	= attr->vendor_part_id;
188 	resp->hw_ver		= attr->hw_ver;
189 	resp->max_qp		= attr->max_qp;
190 	resp->max_qp_wr		= attr->max_qp_wr;
191 	resp->device_cap_flags	= lower_32_bits(attr->device_cap_flags);
192 	resp->max_sge		= attr->max_sge;
193 	resp->max_sge_rd	= attr->max_sge_rd;
194 	resp->max_cq		= attr->max_cq;
195 	resp->max_cqe		= attr->max_cqe;
196 	resp->max_mr		= attr->max_mr;
197 	resp->max_pd		= attr->max_pd;
198 	resp->max_qp_rd_atom	= attr->max_qp_rd_atom;
199 	resp->max_ee_rd_atom	= attr->max_ee_rd_atom;
200 	resp->max_res_rd_atom	= attr->max_res_rd_atom;
201 	resp->max_qp_init_rd_atom	= attr->max_qp_init_rd_atom;
202 	resp->max_ee_init_rd_atom	= attr->max_ee_init_rd_atom;
203 	resp->atomic_cap		= attr->atomic_cap;
204 	resp->max_ee			= attr->max_ee;
205 	resp->max_rdd			= attr->max_rdd;
206 	resp->max_mw			= attr->max_mw;
207 	resp->max_raw_ipv6_qp		= attr->max_raw_ipv6_qp;
208 	resp->max_raw_ethy_qp		= attr->max_raw_ethy_qp;
209 	resp->max_mcast_grp		= attr->max_mcast_grp;
210 	resp->max_mcast_qp_attach	= attr->max_mcast_qp_attach;
211 	resp->max_total_mcast_qp_attach	= attr->max_total_mcast_qp_attach;
212 	resp->max_ah			= attr->max_ah;
213 	resp->max_fmr			= attr->max_fmr;
214 	resp->max_map_per_fmr		= attr->max_map_per_fmr;
215 	resp->max_srq			= attr->max_srq;
216 	resp->max_srq_wr		= attr->max_srq_wr;
217 	resp->max_srq_sge		= attr->max_srq_sge;
218 	resp->max_pkeys			= attr->max_pkeys;
219 	resp->local_ca_ack_delay	= attr->local_ca_ack_delay;
220 	resp->phys_port_cnt		= ib_dev->phys_port_cnt;
221 }
222 
223 ssize_t ib_uverbs_query_device(struct ib_uverbs_file *file,
224 			       struct ib_device *ib_dev,
225 			       const char __user *buf,
226 			       int in_len, int out_len)
227 {
228 	struct ib_uverbs_query_device      cmd;
229 	struct ib_uverbs_query_device_resp resp;
230 
231 	if (out_len < sizeof resp)
232 		return -ENOSPC;
233 
234 	if (copy_from_user(&cmd, buf, sizeof cmd))
235 		return -EFAULT;
236 
237 	memset(&resp, 0, sizeof resp);
238 	copy_query_dev_fields(file, ib_dev, &resp, &ib_dev->attrs);
239 
240 	if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof resp))
241 		return -EFAULT;
242 
243 	return in_len;
244 }
245 
246 ssize_t ib_uverbs_query_port(struct ib_uverbs_file *file,
247 			     struct ib_device *ib_dev,
248 			     const char __user *buf,
249 			     int in_len, int out_len)
250 {
251 	struct ib_uverbs_query_port      cmd;
252 	struct ib_uverbs_query_port_resp resp;
253 	struct ib_port_attr              attr;
254 	int                              ret;
255 
256 	if (out_len < sizeof resp)
257 		return -ENOSPC;
258 
259 	if (copy_from_user(&cmd, buf, sizeof cmd))
260 		return -EFAULT;
261 
262 	ret = ib_query_port(ib_dev, cmd.port_num, &attr);
263 	if (ret)
264 		return ret;
265 
266 	memset(&resp, 0, sizeof resp);
267 
268 	resp.state 	     = attr.state;
269 	resp.max_mtu 	     = attr.max_mtu;
270 	resp.active_mtu      = attr.active_mtu;
271 	resp.gid_tbl_len     = attr.gid_tbl_len;
272 	resp.port_cap_flags  = attr.port_cap_flags;
273 	resp.max_msg_sz      = attr.max_msg_sz;
274 	resp.bad_pkey_cntr   = attr.bad_pkey_cntr;
275 	resp.qkey_viol_cntr  = attr.qkey_viol_cntr;
276 	resp.pkey_tbl_len    = attr.pkey_tbl_len;
277 
278 	if (rdma_cap_opa_ah(ib_dev, cmd.port_num)) {
279 		resp.lid     = OPA_TO_IB_UCAST_LID(attr.lid);
280 		resp.sm_lid  = OPA_TO_IB_UCAST_LID(attr.sm_lid);
281 	} else {
282 		resp.lid     = ib_lid_cpu16(attr.lid);
283 		resp.sm_lid  = ib_lid_cpu16(attr.sm_lid);
284 	}
285 	resp.lmc 	     = attr.lmc;
286 	resp.max_vl_num      = attr.max_vl_num;
287 	resp.sm_sl 	     = attr.sm_sl;
288 	resp.subnet_timeout  = attr.subnet_timeout;
289 	resp.init_type_reply = attr.init_type_reply;
290 	resp.active_width    = attr.active_width;
291 	resp.active_speed    = attr.active_speed;
292 	resp.phys_state      = attr.phys_state;
293 	resp.link_layer      = rdma_port_get_link_layer(ib_dev,
294 							cmd.port_num);
295 
296 	if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof resp))
297 		return -EFAULT;
298 
299 	return in_len;
300 }
301 
302 ssize_t ib_uverbs_alloc_pd(struct ib_uverbs_file *file,
303 			   struct ib_device *ib_dev,
304 			   const char __user *buf,
305 			   int in_len, int out_len)
306 {
307 	struct ib_uverbs_alloc_pd      cmd;
308 	struct ib_uverbs_alloc_pd_resp resp;
309 	struct ib_udata                udata;
310 	struct ib_uobject             *uobj;
311 	struct ib_pd                  *pd;
312 	int                            ret;
313 
314 	if (out_len < sizeof resp)
315 		return -ENOSPC;
316 
317 	if (copy_from_user(&cmd, buf, sizeof cmd))
318 		return -EFAULT;
319 
320 	ib_uverbs_init_udata(&udata, buf + sizeof(cmd),
321 		   u64_to_user_ptr(cmd.response) + sizeof(resp),
322                    in_len - sizeof(cmd) - sizeof(struct ib_uverbs_cmd_hdr),
323                    out_len - sizeof(resp));
324 
325 	uobj  = uobj_alloc(UVERBS_OBJECT_PD, file->ucontext);
326 	if (IS_ERR(uobj))
327 		return PTR_ERR(uobj);
328 
329 	pd = ib_dev->alloc_pd(ib_dev, file->ucontext, &udata);
330 	if (IS_ERR(pd)) {
331 		ret = PTR_ERR(pd);
332 		goto err;
333 	}
334 
335 	pd->device  = ib_dev;
336 	pd->uobject = uobj;
337 	pd->__internal_mr = NULL;
338 	atomic_set(&pd->usecnt, 0);
339 
340 	uobj->object = pd;
341 	memset(&resp, 0, sizeof resp);
342 	resp.pd_handle = uobj->id;
343 	pd->res.type = RDMA_RESTRACK_PD;
344 	rdma_restrack_add(&pd->res);
345 
346 	if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof resp)) {
347 		ret = -EFAULT;
348 		goto err_copy;
349 	}
350 
351 	uobj_alloc_commit(uobj);
352 
353 	return in_len;
354 
355 err_copy:
356 	ib_dealloc_pd(pd);
357 
358 err:
359 	uobj_alloc_abort(uobj);
360 	return ret;
361 }
362 
363 ssize_t ib_uverbs_dealloc_pd(struct ib_uverbs_file *file,
364 			     struct ib_device *ib_dev,
365 			     const char __user *buf,
366 			     int in_len, int out_len)
367 {
368 	struct ib_uverbs_dealloc_pd cmd;
369 	struct ib_uobject          *uobj;
370 	int                         ret;
371 
372 	if (copy_from_user(&cmd, buf, sizeof cmd))
373 		return -EFAULT;
374 
375 	uobj  = uobj_get_write(UVERBS_OBJECT_PD, cmd.pd_handle,
376 			       file->ucontext);
377 	if (IS_ERR(uobj))
378 		return PTR_ERR(uobj);
379 
380 	ret = uobj_remove_commit(uobj);
381 
382 	return ret ?: in_len;
383 }
384 
385 struct xrcd_table_entry {
386 	struct rb_node  node;
387 	struct ib_xrcd *xrcd;
388 	struct inode   *inode;
389 };
390 
391 static int xrcd_table_insert(struct ib_uverbs_device *dev,
392 			    struct inode *inode,
393 			    struct ib_xrcd *xrcd)
394 {
395 	struct xrcd_table_entry *entry, *scan;
396 	struct rb_node **p = &dev->xrcd_tree.rb_node;
397 	struct rb_node *parent = NULL;
398 
399 	entry = kmalloc(sizeof *entry, GFP_KERNEL);
400 	if (!entry)
401 		return -ENOMEM;
402 
403 	entry->xrcd  = xrcd;
404 	entry->inode = inode;
405 
406 	while (*p) {
407 		parent = *p;
408 		scan = rb_entry(parent, struct xrcd_table_entry, node);
409 
410 		if (inode < scan->inode) {
411 			p = &(*p)->rb_left;
412 		} else if (inode > scan->inode) {
413 			p = &(*p)->rb_right;
414 		} else {
415 			kfree(entry);
416 			return -EEXIST;
417 		}
418 	}
419 
420 	rb_link_node(&entry->node, parent, p);
421 	rb_insert_color(&entry->node, &dev->xrcd_tree);
422 	igrab(inode);
423 	return 0;
424 }
425 
426 static struct xrcd_table_entry *xrcd_table_search(struct ib_uverbs_device *dev,
427 						  struct inode *inode)
428 {
429 	struct xrcd_table_entry *entry;
430 	struct rb_node *p = dev->xrcd_tree.rb_node;
431 
432 	while (p) {
433 		entry = rb_entry(p, struct xrcd_table_entry, node);
434 
435 		if (inode < entry->inode)
436 			p = p->rb_left;
437 		else if (inode > entry->inode)
438 			p = p->rb_right;
439 		else
440 			return entry;
441 	}
442 
443 	return NULL;
444 }
445 
446 static struct ib_xrcd *find_xrcd(struct ib_uverbs_device *dev, struct inode *inode)
447 {
448 	struct xrcd_table_entry *entry;
449 
450 	entry = xrcd_table_search(dev, inode);
451 	if (!entry)
452 		return NULL;
453 
454 	return entry->xrcd;
455 }
456 
457 static void xrcd_table_delete(struct ib_uverbs_device *dev,
458 			      struct inode *inode)
459 {
460 	struct xrcd_table_entry *entry;
461 
462 	entry = xrcd_table_search(dev, inode);
463 	if (entry) {
464 		iput(inode);
465 		rb_erase(&entry->node, &dev->xrcd_tree);
466 		kfree(entry);
467 	}
468 }
469 
470 ssize_t ib_uverbs_open_xrcd(struct ib_uverbs_file *file,
471 			    struct ib_device *ib_dev,
472 			    const char __user *buf, int in_len,
473 			    int out_len)
474 {
475 	struct ib_uverbs_open_xrcd	cmd;
476 	struct ib_uverbs_open_xrcd_resp	resp;
477 	struct ib_udata			udata;
478 	struct ib_uxrcd_object         *obj;
479 	struct ib_xrcd                 *xrcd = NULL;
480 	struct fd			f = {NULL, 0};
481 	struct inode                   *inode = NULL;
482 	int				ret = 0;
483 	int				new_xrcd = 0;
484 
485 	if (out_len < sizeof resp)
486 		return -ENOSPC;
487 
488 	if (copy_from_user(&cmd, buf, sizeof cmd))
489 		return -EFAULT;
490 
491 	ib_uverbs_init_udata(&udata, buf + sizeof(cmd),
492 		   u64_to_user_ptr(cmd.response) + sizeof(resp),
493                    in_len - sizeof(cmd) - sizeof(struct ib_uverbs_cmd_hdr),
494                    out_len - sizeof(resp));
495 
496 	mutex_lock(&file->device->xrcd_tree_mutex);
497 
498 	if (cmd.fd != -1) {
499 		/* search for file descriptor */
500 		f = fdget(cmd.fd);
501 		if (!f.file) {
502 			ret = -EBADF;
503 			goto err_tree_mutex_unlock;
504 		}
505 
506 		inode = file_inode(f.file);
507 		xrcd = find_xrcd(file->device, inode);
508 		if (!xrcd && !(cmd.oflags & O_CREAT)) {
509 			/* no file descriptor. Need CREATE flag */
510 			ret = -EAGAIN;
511 			goto err_tree_mutex_unlock;
512 		}
513 
514 		if (xrcd && cmd.oflags & O_EXCL) {
515 			ret = -EINVAL;
516 			goto err_tree_mutex_unlock;
517 		}
518 	}
519 
520 	obj  = (struct ib_uxrcd_object *)uobj_alloc(UVERBS_OBJECT_XRCD,
521 						    file->ucontext);
522 	if (IS_ERR(obj)) {
523 		ret = PTR_ERR(obj);
524 		goto err_tree_mutex_unlock;
525 	}
526 
527 	if (!xrcd) {
528 		xrcd = ib_dev->alloc_xrcd(ib_dev, file->ucontext, &udata);
529 		if (IS_ERR(xrcd)) {
530 			ret = PTR_ERR(xrcd);
531 			goto err;
532 		}
533 
534 		xrcd->inode   = inode;
535 		xrcd->device  = ib_dev;
536 		atomic_set(&xrcd->usecnt, 0);
537 		mutex_init(&xrcd->tgt_qp_mutex);
538 		INIT_LIST_HEAD(&xrcd->tgt_qp_list);
539 		new_xrcd = 1;
540 	}
541 
542 	atomic_set(&obj->refcnt, 0);
543 	obj->uobject.object = xrcd;
544 	memset(&resp, 0, sizeof resp);
545 	resp.xrcd_handle = obj->uobject.id;
546 
547 	if (inode) {
548 		if (new_xrcd) {
549 			/* create new inode/xrcd table entry */
550 			ret = xrcd_table_insert(file->device, inode, xrcd);
551 			if (ret)
552 				goto err_dealloc_xrcd;
553 		}
554 		atomic_inc(&xrcd->usecnt);
555 	}
556 
557 	if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof resp)) {
558 		ret = -EFAULT;
559 		goto err_copy;
560 	}
561 
562 	if (f.file)
563 		fdput(f);
564 
565 	mutex_unlock(&file->device->xrcd_tree_mutex);
566 
567 	uobj_alloc_commit(&obj->uobject);
568 
569 	return in_len;
570 
571 err_copy:
572 	if (inode) {
573 		if (new_xrcd)
574 			xrcd_table_delete(file->device, inode);
575 		atomic_dec(&xrcd->usecnt);
576 	}
577 
578 err_dealloc_xrcd:
579 	ib_dealloc_xrcd(xrcd);
580 
581 err:
582 	uobj_alloc_abort(&obj->uobject);
583 
584 err_tree_mutex_unlock:
585 	if (f.file)
586 		fdput(f);
587 
588 	mutex_unlock(&file->device->xrcd_tree_mutex);
589 
590 	return ret;
591 }
592 
593 ssize_t ib_uverbs_close_xrcd(struct ib_uverbs_file *file,
594 			     struct ib_device *ib_dev,
595 			     const char __user *buf, int in_len,
596 			     int out_len)
597 {
598 	struct ib_uverbs_close_xrcd cmd;
599 	struct ib_uobject           *uobj;
600 	int                         ret = 0;
601 
602 	if (copy_from_user(&cmd, buf, sizeof cmd))
603 		return -EFAULT;
604 
605 	uobj  = uobj_get_write(UVERBS_OBJECT_XRCD, cmd.xrcd_handle,
606 			       file->ucontext);
607 	if (IS_ERR(uobj))
608 		return PTR_ERR(uobj);
609 
610 	ret = uobj_remove_commit(uobj);
611 	return ret ?: in_len;
612 }
613 
614 int ib_uverbs_dealloc_xrcd(struct ib_uverbs_device *dev,
615 			   struct ib_xrcd *xrcd,
616 			   enum rdma_remove_reason why)
617 {
618 	struct inode *inode;
619 	int ret;
620 
621 	inode = xrcd->inode;
622 	if (inode && !atomic_dec_and_test(&xrcd->usecnt))
623 		return 0;
624 
625 	ret = ib_dealloc_xrcd(xrcd);
626 
627 	if (why == RDMA_REMOVE_DESTROY && ret)
628 		atomic_inc(&xrcd->usecnt);
629 	else if (inode)
630 		xrcd_table_delete(dev, inode);
631 
632 	return ret;
633 }
634 
635 ssize_t ib_uverbs_reg_mr(struct ib_uverbs_file *file,
636 			 struct ib_device *ib_dev,
637 			 const char __user *buf, int in_len,
638 			 int out_len)
639 {
640 	struct ib_uverbs_reg_mr      cmd;
641 	struct ib_uverbs_reg_mr_resp resp;
642 	struct ib_udata              udata;
643 	struct ib_uobject           *uobj;
644 	struct ib_pd                *pd;
645 	struct ib_mr                *mr;
646 	int                          ret;
647 
648 	if (out_len < sizeof resp)
649 		return -ENOSPC;
650 
651 	if (copy_from_user(&cmd, buf, sizeof cmd))
652 		return -EFAULT;
653 
654 	ib_uverbs_init_udata(&udata, buf + sizeof(cmd),
655 		   u64_to_user_ptr(cmd.response) + sizeof(resp),
656                    in_len - sizeof(cmd) - sizeof(struct ib_uverbs_cmd_hdr),
657                    out_len - sizeof(resp));
658 
659 	if ((cmd.start & ~PAGE_MASK) != (cmd.hca_va & ~PAGE_MASK))
660 		return -EINVAL;
661 
662 	ret = ib_check_mr_access(cmd.access_flags);
663 	if (ret)
664 		return ret;
665 
666 	uobj  = uobj_alloc(UVERBS_OBJECT_MR, file->ucontext);
667 	if (IS_ERR(uobj))
668 		return PTR_ERR(uobj);
669 
670 	pd = uobj_get_obj_read(pd, UVERBS_OBJECT_PD, cmd.pd_handle, file->ucontext);
671 	if (!pd) {
672 		ret = -EINVAL;
673 		goto err_free;
674 	}
675 
676 	if (cmd.access_flags & IB_ACCESS_ON_DEMAND) {
677 		if (!(pd->device->attrs.device_cap_flags &
678 		      IB_DEVICE_ON_DEMAND_PAGING)) {
679 			pr_debug("ODP support not available\n");
680 			ret = -EINVAL;
681 			goto err_put;
682 		}
683 	}
684 
685 	mr = pd->device->reg_user_mr(pd, cmd.start, cmd.length, cmd.hca_va,
686 				     cmd.access_flags, &udata);
687 	if (IS_ERR(mr)) {
688 		ret = PTR_ERR(mr);
689 		goto err_put;
690 	}
691 
692 	mr->device  = pd->device;
693 	mr->pd      = pd;
694 	mr->dm	    = NULL;
695 	mr->uobject = uobj;
696 	atomic_inc(&pd->usecnt);
697 	mr->res.type = RDMA_RESTRACK_MR;
698 	rdma_restrack_add(&mr->res);
699 
700 	uobj->object = mr;
701 
702 	memset(&resp, 0, sizeof resp);
703 	resp.lkey      = mr->lkey;
704 	resp.rkey      = mr->rkey;
705 	resp.mr_handle = uobj->id;
706 
707 	if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof resp)) {
708 		ret = -EFAULT;
709 		goto err_copy;
710 	}
711 
712 	uobj_put_obj_read(pd);
713 
714 	uobj_alloc_commit(uobj);
715 
716 	return in_len;
717 
718 err_copy:
719 	ib_dereg_mr(mr);
720 
721 err_put:
722 	uobj_put_obj_read(pd);
723 
724 err_free:
725 	uobj_alloc_abort(uobj);
726 	return ret;
727 }
728 
729 ssize_t ib_uverbs_rereg_mr(struct ib_uverbs_file *file,
730 			   struct ib_device *ib_dev,
731 			   const char __user *buf, int in_len,
732 			   int out_len)
733 {
734 	struct ib_uverbs_rereg_mr      cmd;
735 	struct ib_uverbs_rereg_mr_resp resp;
736 	struct ib_udata              udata;
737 	struct ib_pd                *pd = NULL;
738 	struct ib_mr                *mr;
739 	struct ib_pd		    *old_pd;
740 	int                          ret;
741 	struct ib_uobject	    *uobj;
742 
743 	if (out_len < sizeof(resp))
744 		return -ENOSPC;
745 
746 	if (copy_from_user(&cmd, buf, sizeof(cmd)))
747 		return -EFAULT;
748 
749 	ib_uverbs_init_udata(&udata, buf + sizeof(cmd),
750 		   u64_to_user_ptr(cmd.response) + sizeof(resp),
751                    in_len - sizeof(cmd) - sizeof(struct ib_uverbs_cmd_hdr),
752                    out_len - sizeof(resp));
753 
754 	if (cmd.flags & ~IB_MR_REREG_SUPPORTED || !cmd.flags)
755 		return -EINVAL;
756 
757 	if ((cmd.flags & IB_MR_REREG_TRANS) &&
758 	    (!cmd.start || !cmd.hca_va || 0 >= cmd.length ||
759 	     (cmd.start & ~PAGE_MASK) != (cmd.hca_va & ~PAGE_MASK)))
760 			return -EINVAL;
761 
762 	uobj  = uobj_get_write(UVERBS_OBJECT_MR, cmd.mr_handle,
763 			       file->ucontext);
764 	if (IS_ERR(uobj))
765 		return PTR_ERR(uobj);
766 
767 	mr = uobj->object;
768 
769 	if (mr->dm) {
770 		ret = -EINVAL;
771 		goto put_uobjs;
772 	}
773 
774 	if (cmd.flags & IB_MR_REREG_ACCESS) {
775 		ret = ib_check_mr_access(cmd.access_flags);
776 		if (ret)
777 			goto put_uobjs;
778 	}
779 
780 	if (cmd.flags & IB_MR_REREG_PD) {
781 		pd = uobj_get_obj_read(pd, UVERBS_OBJECT_PD, cmd.pd_handle, file->ucontext);
782 		if (!pd) {
783 			ret = -EINVAL;
784 			goto put_uobjs;
785 		}
786 	}
787 
788 	old_pd = mr->pd;
789 	ret = mr->device->rereg_user_mr(mr, cmd.flags, cmd.start,
790 					cmd.length, cmd.hca_va,
791 					cmd.access_flags, pd, &udata);
792 	if (!ret) {
793 		if (cmd.flags & IB_MR_REREG_PD) {
794 			atomic_inc(&pd->usecnt);
795 			mr->pd = pd;
796 			atomic_dec(&old_pd->usecnt);
797 		}
798 	} else {
799 		goto put_uobj_pd;
800 	}
801 
802 	memset(&resp, 0, sizeof(resp));
803 	resp.lkey      = mr->lkey;
804 	resp.rkey      = mr->rkey;
805 
806 	if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof(resp)))
807 		ret = -EFAULT;
808 	else
809 		ret = in_len;
810 
811 put_uobj_pd:
812 	if (cmd.flags & IB_MR_REREG_PD)
813 		uobj_put_obj_read(pd);
814 
815 put_uobjs:
816 	uobj_put_write(uobj);
817 
818 	return ret;
819 }
820 
821 ssize_t ib_uverbs_dereg_mr(struct ib_uverbs_file *file,
822 			   struct ib_device *ib_dev,
823 			   const char __user *buf, int in_len,
824 			   int out_len)
825 {
826 	struct ib_uverbs_dereg_mr cmd;
827 	struct ib_uobject	 *uobj;
828 	int                       ret = -EINVAL;
829 
830 	if (copy_from_user(&cmd, buf, sizeof cmd))
831 		return -EFAULT;
832 
833 	uobj  = uobj_get_write(UVERBS_OBJECT_MR, cmd.mr_handle,
834 			       file->ucontext);
835 	if (IS_ERR(uobj))
836 		return PTR_ERR(uobj);
837 
838 	ret = uobj_remove_commit(uobj);
839 
840 	return ret ?: in_len;
841 }
842 
843 ssize_t ib_uverbs_alloc_mw(struct ib_uverbs_file *file,
844 			   struct ib_device *ib_dev,
845 			   const char __user *buf, int in_len,
846 			   int out_len)
847 {
848 	struct ib_uverbs_alloc_mw      cmd;
849 	struct ib_uverbs_alloc_mw_resp resp;
850 	struct ib_uobject             *uobj;
851 	struct ib_pd                  *pd;
852 	struct ib_mw                  *mw;
853 	struct ib_udata		       udata;
854 	int                            ret;
855 
856 	if (out_len < sizeof(resp))
857 		return -ENOSPC;
858 
859 	if (copy_from_user(&cmd, buf, sizeof(cmd)))
860 		return -EFAULT;
861 
862 	uobj  = uobj_alloc(UVERBS_OBJECT_MW, file->ucontext);
863 	if (IS_ERR(uobj))
864 		return PTR_ERR(uobj);
865 
866 	pd = uobj_get_obj_read(pd, UVERBS_OBJECT_PD, cmd.pd_handle, file->ucontext);
867 	if (!pd) {
868 		ret = -EINVAL;
869 		goto err_free;
870 	}
871 
872 	ib_uverbs_init_udata(&udata, buf + sizeof(cmd),
873 		   u64_to_user_ptr(cmd.response) + sizeof(resp),
874 		   in_len - sizeof(cmd) - sizeof(struct ib_uverbs_cmd_hdr),
875 		   out_len - sizeof(resp));
876 
877 	mw = pd->device->alloc_mw(pd, cmd.mw_type, &udata);
878 	if (IS_ERR(mw)) {
879 		ret = PTR_ERR(mw);
880 		goto err_put;
881 	}
882 
883 	mw->device  = pd->device;
884 	mw->pd      = pd;
885 	mw->uobject = uobj;
886 	atomic_inc(&pd->usecnt);
887 
888 	uobj->object = mw;
889 
890 	memset(&resp, 0, sizeof(resp));
891 	resp.rkey      = mw->rkey;
892 	resp.mw_handle = uobj->id;
893 
894 	if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof(resp))) {
895 		ret = -EFAULT;
896 		goto err_copy;
897 	}
898 
899 	uobj_put_obj_read(pd);
900 	uobj_alloc_commit(uobj);
901 
902 	return in_len;
903 
904 err_copy:
905 	uverbs_dealloc_mw(mw);
906 err_put:
907 	uobj_put_obj_read(pd);
908 err_free:
909 	uobj_alloc_abort(uobj);
910 	return ret;
911 }
912 
913 ssize_t ib_uverbs_dealloc_mw(struct ib_uverbs_file *file,
914 			     struct ib_device *ib_dev,
915 			     const char __user *buf, int in_len,
916 			     int out_len)
917 {
918 	struct ib_uverbs_dealloc_mw cmd;
919 	struct ib_uobject	   *uobj;
920 	int                         ret = -EINVAL;
921 
922 	if (copy_from_user(&cmd, buf, sizeof(cmd)))
923 		return -EFAULT;
924 
925 	uobj  = uobj_get_write(UVERBS_OBJECT_MW, cmd.mw_handle,
926 			       file->ucontext);
927 	if (IS_ERR(uobj))
928 		return PTR_ERR(uobj);
929 
930 	ret = uobj_remove_commit(uobj);
931 	return ret ?: in_len;
932 }
933 
934 ssize_t ib_uverbs_create_comp_channel(struct ib_uverbs_file *file,
935 				      struct ib_device *ib_dev,
936 				      const char __user *buf, int in_len,
937 				      int out_len)
938 {
939 	struct ib_uverbs_create_comp_channel	   cmd;
940 	struct ib_uverbs_create_comp_channel_resp  resp;
941 	struct ib_uobject			  *uobj;
942 	struct ib_uverbs_completion_event_file	  *ev_file;
943 
944 	if (out_len < sizeof resp)
945 		return -ENOSPC;
946 
947 	if (copy_from_user(&cmd, buf, sizeof cmd))
948 		return -EFAULT;
949 
950 	uobj = uobj_alloc(UVERBS_OBJECT_COMP_CHANNEL, file->ucontext);
951 	if (IS_ERR(uobj))
952 		return PTR_ERR(uobj);
953 
954 	resp.fd = uobj->id;
955 
956 	ev_file = container_of(uobj, struct ib_uverbs_completion_event_file,
957 			       uobj_file.uobj);
958 	ib_uverbs_init_event_queue(&ev_file->ev_queue);
959 
960 	if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof resp)) {
961 		uobj_alloc_abort(uobj);
962 		return -EFAULT;
963 	}
964 
965 	uobj_alloc_commit(uobj);
966 	return in_len;
967 }
968 
969 static struct ib_ucq_object *create_cq(struct ib_uverbs_file *file,
970 					struct ib_device *ib_dev,
971 				       struct ib_udata *ucore,
972 				       struct ib_udata *uhw,
973 				       struct ib_uverbs_ex_create_cq *cmd,
974 				       size_t cmd_sz,
975 				       int (*cb)(struct ib_uverbs_file *file,
976 						 struct ib_ucq_object *obj,
977 						 struct ib_uverbs_ex_create_cq_resp *resp,
978 						 struct ib_udata *udata,
979 						 void *context),
980 				       void *context)
981 {
982 	struct ib_ucq_object           *obj;
983 	struct ib_uverbs_completion_event_file    *ev_file = NULL;
984 	struct ib_cq                   *cq;
985 	int                             ret;
986 	struct ib_uverbs_ex_create_cq_resp resp;
987 	struct ib_cq_init_attr attr = {};
988 
989 	if (!ib_dev->create_cq)
990 		return ERR_PTR(-EOPNOTSUPP);
991 
992 	if (cmd->comp_vector >= file->device->num_comp_vectors)
993 		return ERR_PTR(-EINVAL);
994 
995 	obj  = (struct ib_ucq_object *)uobj_alloc(UVERBS_OBJECT_CQ,
996 						  file->ucontext);
997 	if (IS_ERR(obj))
998 		return obj;
999 
1000 	if (cmd->comp_channel >= 0) {
1001 		ev_file = ib_uverbs_lookup_comp_file(cmd->comp_channel,
1002 						     file->ucontext);
1003 		if (IS_ERR(ev_file)) {
1004 			ret = PTR_ERR(ev_file);
1005 			goto err;
1006 		}
1007 	}
1008 
1009 	obj->uobject.user_handle = cmd->user_handle;
1010 	obj->uverbs_file	   = file;
1011 	obj->comp_events_reported  = 0;
1012 	obj->async_events_reported = 0;
1013 	INIT_LIST_HEAD(&obj->comp_list);
1014 	INIT_LIST_HEAD(&obj->async_list);
1015 
1016 	attr.cqe = cmd->cqe;
1017 	attr.comp_vector = cmd->comp_vector;
1018 
1019 	if (cmd_sz > offsetof(typeof(*cmd), flags) + sizeof(cmd->flags))
1020 		attr.flags = cmd->flags;
1021 
1022 	cq = ib_dev->create_cq(ib_dev, &attr, file->ucontext, uhw);
1023 	if (IS_ERR(cq)) {
1024 		ret = PTR_ERR(cq);
1025 		goto err_file;
1026 	}
1027 
1028 	cq->device        = ib_dev;
1029 	cq->uobject       = &obj->uobject;
1030 	cq->comp_handler  = ib_uverbs_comp_handler;
1031 	cq->event_handler = ib_uverbs_cq_event_handler;
1032 	cq->cq_context    = ev_file ? &ev_file->ev_queue : NULL;
1033 	atomic_set(&cq->usecnt, 0);
1034 
1035 	obj->uobject.object = cq;
1036 	memset(&resp, 0, sizeof resp);
1037 	resp.base.cq_handle = obj->uobject.id;
1038 	resp.base.cqe       = cq->cqe;
1039 
1040 	resp.response_length = offsetof(typeof(resp), response_length) +
1041 		sizeof(resp.response_length);
1042 
1043 	cq->res.type = RDMA_RESTRACK_CQ;
1044 	rdma_restrack_add(&cq->res);
1045 
1046 	ret = cb(file, obj, &resp, ucore, context);
1047 	if (ret)
1048 		goto err_cb;
1049 
1050 	uobj_alloc_commit(&obj->uobject);
1051 	return obj;
1052 
1053 err_cb:
1054 	ib_destroy_cq(cq);
1055 
1056 err_file:
1057 	if (ev_file)
1058 		ib_uverbs_release_ucq(file, ev_file, obj);
1059 
1060 err:
1061 	uobj_alloc_abort(&obj->uobject);
1062 
1063 	return ERR_PTR(ret);
1064 }
1065 
1066 static int ib_uverbs_create_cq_cb(struct ib_uverbs_file *file,
1067 				  struct ib_ucq_object *obj,
1068 				  struct ib_uverbs_ex_create_cq_resp *resp,
1069 				  struct ib_udata *ucore, void *context)
1070 {
1071 	if (ib_copy_to_udata(ucore, &resp->base, sizeof(resp->base)))
1072 		return -EFAULT;
1073 
1074 	return 0;
1075 }
1076 
1077 ssize_t ib_uverbs_create_cq(struct ib_uverbs_file *file,
1078 			    struct ib_device *ib_dev,
1079 			    const char __user *buf, int in_len,
1080 			    int out_len)
1081 {
1082 	struct ib_uverbs_create_cq      cmd;
1083 	struct ib_uverbs_ex_create_cq	cmd_ex;
1084 	struct ib_uverbs_create_cq_resp resp;
1085 	struct ib_udata                 ucore;
1086 	struct ib_udata                 uhw;
1087 	struct ib_ucq_object           *obj;
1088 
1089 	if (out_len < sizeof(resp))
1090 		return -ENOSPC;
1091 
1092 	if (copy_from_user(&cmd, buf, sizeof(cmd)))
1093 		return -EFAULT;
1094 
1095 	ib_uverbs_init_udata(&ucore, buf, u64_to_user_ptr(cmd.response),
1096 			     sizeof(cmd), sizeof(resp));
1097 
1098 	ib_uverbs_init_udata(&uhw, buf + sizeof(cmd),
1099 		   u64_to_user_ptr(cmd.response) + sizeof(resp),
1100 		   in_len - sizeof(cmd) - sizeof(struct ib_uverbs_cmd_hdr),
1101 		   out_len - sizeof(resp));
1102 
1103 	memset(&cmd_ex, 0, sizeof(cmd_ex));
1104 	cmd_ex.user_handle = cmd.user_handle;
1105 	cmd_ex.cqe = cmd.cqe;
1106 	cmd_ex.comp_vector = cmd.comp_vector;
1107 	cmd_ex.comp_channel = cmd.comp_channel;
1108 
1109 	obj = create_cq(file, ib_dev, &ucore, &uhw, &cmd_ex,
1110 			offsetof(typeof(cmd_ex), comp_channel) +
1111 			sizeof(cmd.comp_channel), ib_uverbs_create_cq_cb,
1112 			NULL);
1113 
1114 	if (IS_ERR(obj))
1115 		return PTR_ERR(obj);
1116 
1117 	return in_len;
1118 }
1119 
1120 static int ib_uverbs_ex_create_cq_cb(struct ib_uverbs_file *file,
1121 				     struct ib_ucq_object *obj,
1122 				     struct ib_uverbs_ex_create_cq_resp *resp,
1123 				     struct ib_udata *ucore, void *context)
1124 {
1125 	if (ib_copy_to_udata(ucore, resp, resp->response_length))
1126 		return -EFAULT;
1127 
1128 	return 0;
1129 }
1130 
1131 int ib_uverbs_ex_create_cq(struct ib_uverbs_file *file,
1132 			 struct ib_device *ib_dev,
1133 			   struct ib_udata *ucore,
1134 			   struct ib_udata *uhw)
1135 {
1136 	struct ib_uverbs_ex_create_cq_resp resp;
1137 	struct ib_uverbs_ex_create_cq  cmd;
1138 	struct ib_ucq_object           *obj;
1139 	int err;
1140 
1141 	if (ucore->inlen < sizeof(cmd))
1142 		return -EINVAL;
1143 
1144 	err = ib_copy_from_udata(&cmd, ucore, sizeof(cmd));
1145 	if (err)
1146 		return err;
1147 
1148 	if (cmd.comp_mask)
1149 		return -EINVAL;
1150 
1151 	if (cmd.reserved)
1152 		return -EINVAL;
1153 
1154 	if (ucore->outlen < (offsetof(typeof(resp), response_length) +
1155 			     sizeof(resp.response_length)))
1156 		return -ENOSPC;
1157 
1158 	obj = create_cq(file, ib_dev, ucore, uhw, &cmd,
1159 			min(ucore->inlen, sizeof(cmd)),
1160 			ib_uverbs_ex_create_cq_cb, NULL);
1161 
1162 	return PTR_ERR_OR_ZERO(obj);
1163 }
1164 
1165 ssize_t ib_uverbs_resize_cq(struct ib_uverbs_file *file,
1166 			    struct ib_device *ib_dev,
1167 			    const char __user *buf, int in_len,
1168 			    int out_len)
1169 {
1170 	struct ib_uverbs_resize_cq	cmd;
1171 	struct ib_uverbs_resize_cq_resp	resp = {};
1172 	struct ib_udata                 udata;
1173 	struct ib_cq			*cq;
1174 	int				ret = -EINVAL;
1175 
1176 	if (copy_from_user(&cmd, buf, sizeof cmd))
1177 		return -EFAULT;
1178 
1179 	ib_uverbs_init_udata(&udata, buf + sizeof(cmd),
1180 		   u64_to_user_ptr(cmd.response) + sizeof(resp),
1181 		   in_len - sizeof(cmd) - sizeof(struct ib_uverbs_cmd_hdr),
1182 		   out_len - sizeof(resp));
1183 
1184 	cq = uobj_get_obj_read(cq, UVERBS_OBJECT_CQ, cmd.cq_handle, file->ucontext);
1185 	if (!cq)
1186 		return -EINVAL;
1187 
1188 	ret = cq->device->resize_cq(cq, cmd.cqe, &udata);
1189 	if (ret)
1190 		goto out;
1191 
1192 	resp.cqe = cq->cqe;
1193 
1194 	if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof resp.cqe))
1195 		ret = -EFAULT;
1196 
1197 out:
1198 	uobj_put_obj_read(cq);
1199 
1200 	return ret ? ret : in_len;
1201 }
1202 
1203 static int copy_wc_to_user(struct ib_device *ib_dev, void __user *dest,
1204 			   struct ib_wc *wc)
1205 {
1206 	struct ib_uverbs_wc tmp;
1207 
1208 	tmp.wr_id		= wc->wr_id;
1209 	tmp.status		= wc->status;
1210 	tmp.opcode		= wc->opcode;
1211 	tmp.vendor_err		= wc->vendor_err;
1212 	tmp.byte_len		= wc->byte_len;
1213 	tmp.ex.imm_data		= wc->ex.imm_data;
1214 	tmp.qp_num		= wc->qp->qp_num;
1215 	tmp.src_qp		= wc->src_qp;
1216 	tmp.wc_flags		= wc->wc_flags;
1217 	tmp.pkey_index		= wc->pkey_index;
1218 	if (rdma_cap_opa_ah(ib_dev, wc->port_num))
1219 		tmp.slid	= OPA_TO_IB_UCAST_LID(wc->slid);
1220 	else
1221 		tmp.slid	= ib_lid_cpu16(wc->slid);
1222 	tmp.sl			= wc->sl;
1223 	tmp.dlid_path_bits	= wc->dlid_path_bits;
1224 	tmp.port_num		= wc->port_num;
1225 	tmp.reserved		= 0;
1226 
1227 	if (copy_to_user(dest, &tmp, sizeof tmp))
1228 		return -EFAULT;
1229 
1230 	return 0;
1231 }
1232 
1233 ssize_t ib_uverbs_poll_cq(struct ib_uverbs_file *file,
1234 			  struct ib_device *ib_dev,
1235 			  const char __user *buf, int in_len,
1236 			  int out_len)
1237 {
1238 	struct ib_uverbs_poll_cq       cmd;
1239 	struct ib_uverbs_poll_cq_resp  resp;
1240 	u8 __user                     *header_ptr;
1241 	u8 __user                     *data_ptr;
1242 	struct ib_cq                  *cq;
1243 	struct ib_wc                   wc;
1244 	int                            ret;
1245 
1246 	if (copy_from_user(&cmd, buf, sizeof cmd))
1247 		return -EFAULT;
1248 
1249 	cq = uobj_get_obj_read(cq, UVERBS_OBJECT_CQ, cmd.cq_handle, file->ucontext);
1250 	if (!cq)
1251 		return -EINVAL;
1252 
1253 	/* we copy a struct ib_uverbs_poll_cq_resp to user space */
1254 	header_ptr = u64_to_user_ptr(cmd.response);
1255 	data_ptr = header_ptr + sizeof resp;
1256 
1257 	memset(&resp, 0, sizeof resp);
1258 	while (resp.count < cmd.ne) {
1259 		ret = ib_poll_cq(cq, 1, &wc);
1260 		if (ret < 0)
1261 			goto out_put;
1262 		if (!ret)
1263 			break;
1264 
1265 		ret = copy_wc_to_user(ib_dev, data_ptr, &wc);
1266 		if (ret)
1267 			goto out_put;
1268 
1269 		data_ptr += sizeof(struct ib_uverbs_wc);
1270 		++resp.count;
1271 	}
1272 
1273 	if (copy_to_user(header_ptr, &resp, sizeof resp)) {
1274 		ret = -EFAULT;
1275 		goto out_put;
1276 	}
1277 
1278 	ret = in_len;
1279 
1280 out_put:
1281 	uobj_put_obj_read(cq);
1282 	return ret;
1283 }
1284 
1285 ssize_t ib_uverbs_req_notify_cq(struct ib_uverbs_file *file,
1286 				struct ib_device *ib_dev,
1287 				const char __user *buf, int in_len,
1288 				int out_len)
1289 {
1290 	struct ib_uverbs_req_notify_cq cmd;
1291 	struct ib_cq                  *cq;
1292 
1293 	if (copy_from_user(&cmd, buf, sizeof cmd))
1294 		return -EFAULT;
1295 
1296 	cq = uobj_get_obj_read(cq, UVERBS_OBJECT_CQ, cmd.cq_handle, file->ucontext);
1297 	if (!cq)
1298 		return -EINVAL;
1299 
1300 	ib_req_notify_cq(cq, cmd.solicited_only ?
1301 			 IB_CQ_SOLICITED : IB_CQ_NEXT_COMP);
1302 
1303 	uobj_put_obj_read(cq);
1304 
1305 	return in_len;
1306 }
1307 
1308 ssize_t ib_uverbs_destroy_cq(struct ib_uverbs_file *file,
1309 			     struct ib_device *ib_dev,
1310 			     const char __user *buf, int in_len,
1311 			     int out_len)
1312 {
1313 	struct ib_uverbs_destroy_cq      cmd;
1314 	struct ib_uverbs_destroy_cq_resp resp;
1315 	struct ib_uobject		*uobj;
1316 	struct ib_cq               	*cq;
1317 	struct ib_ucq_object        	*obj;
1318 	int                        	 ret = -EINVAL;
1319 
1320 	if (copy_from_user(&cmd, buf, sizeof cmd))
1321 		return -EFAULT;
1322 
1323 	uobj  = uobj_get_write(UVERBS_OBJECT_CQ, cmd.cq_handle,
1324 			       file->ucontext);
1325 	if (IS_ERR(uobj))
1326 		return PTR_ERR(uobj);
1327 
1328 	/*
1329 	 * Make sure we don't free the memory in remove_commit as we still
1330 	 * needs the uobject memory to create the response.
1331 	 */
1332 	uverbs_uobject_get(uobj);
1333 	cq      = uobj->object;
1334 	obj     = container_of(cq->uobject, struct ib_ucq_object, uobject);
1335 
1336 	memset(&resp, 0, sizeof(resp));
1337 
1338 	ret = uobj_remove_commit(uobj);
1339 	if (ret) {
1340 		uverbs_uobject_put(uobj);
1341 		return ret;
1342 	}
1343 
1344 	resp.comp_events_reported  = obj->comp_events_reported;
1345 	resp.async_events_reported = obj->async_events_reported;
1346 
1347 	uverbs_uobject_put(uobj);
1348 	if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof resp))
1349 		return -EFAULT;
1350 
1351 	return in_len;
1352 }
1353 
1354 static int create_qp(struct ib_uverbs_file *file,
1355 		     struct ib_udata *ucore,
1356 		     struct ib_udata *uhw,
1357 		     struct ib_uverbs_ex_create_qp *cmd,
1358 		     size_t cmd_sz,
1359 		     int (*cb)(struct ib_uverbs_file *file,
1360 			       struct ib_uverbs_ex_create_qp_resp *resp,
1361 			       struct ib_udata *udata),
1362 		     void *context)
1363 {
1364 	struct ib_uqp_object		*obj;
1365 	struct ib_device		*device;
1366 	struct ib_pd			*pd = NULL;
1367 	struct ib_xrcd			*xrcd = NULL;
1368 	struct ib_uobject		*xrcd_uobj = ERR_PTR(-ENOENT);
1369 	struct ib_cq			*scq = NULL, *rcq = NULL;
1370 	struct ib_srq			*srq = NULL;
1371 	struct ib_qp			*qp;
1372 	char				*buf;
1373 	struct ib_qp_init_attr		attr = {};
1374 	struct ib_uverbs_ex_create_qp_resp resp;
1375 	int				ret;
1376 	struct ib_rwq_ind_table *ind_tbl = NULL;
1377 	bool has_sq = true;
1378 
1379 	if (cmd->qp_type == IB_QPT_RAW_PACKET && !capable(CAP_NET_RAW))
1380 		return -EPERM;
1381 
1382 	obj  = (struct ib_uqp_object *)uobj_alloc(UVERBS_OBJECT_QP,
1383 						  file->ucontext);
1384 	if (IS_ERR(obj))
1385 		return PTR_ERR(obj);
1386 	obj->uxrcd = NULL;
1387 	obj->uevent.uobject.user_handle = cmd->user_handle;
1388 	mutex_init(&obj->mcast_lock);
1389 
1390 	if (cmd_sz >= offsetof(typeof(*cmd), rwq_ind_tbl_handle) +
1391 		      sizeof(cmd->rwq_ind_tbl_handle) &&
1392 		      (cmd->comp_mask & IB_UVERBS_CREATE_QP_MASK_IND_TABLE)) {
1393 		ind_tbl = uobj_get_obj_read(rwq_ind_table, UVERBS_OBJECT_RWQ_IND_TBL,
1394 					    cmd->rwq_ind_tbl_handle,
1395 					    file->ucontext);
1396 		if (!ind_tbl) {
1397 			ret = -EINVAL;
1398 			goto err_put;
1399 		}
1400 
1401 		attr.rwq_ind_tbl = ind_tbl;
1402 	}
1403 
1404 	if (cmd_sz > sizeof(*cmd) &&
1405 	    !ib_is_udata_cleared(ucore, sizeof(*cmd),
1406 				 cmd_sz - sizeof(*cmd))) {
1407 		ret = -EOPNOTSUPP;
1408 		goto err_put;
1409 	}
1410 
1411 	if (ind_tbl && (cmd->max_recv_wr || cmd->max_recv_sge || cmd->is_srq)) {
1412 		ret = -EINVAL;
1413 		goto err_put;
1414 	}
1415 
1416 	if (ind_tbl && !cmd->max_send_wr)
1417 		has_sq = false;
1418 
1419 	if (cmd->qp_type == IB_QPT_XRC_TGT) {
1420 		xrcd_uobj = uobj_get_read(UVERBS_OBJECT_XRCD, cmd->pd_handle,
1421 					  file->ucontext);
1422 
1423 		if (IS_ERR(xrcd_uobj)) {
1424 			ret = -EINVAL;
1425 			goto err_put;
1426 		}
1427 
1428 		xrcd = (struct ib_xrcd *)xrcd_uobj->object;
1429 		if (!xrcd) {
1430 			ret = -EINVAL;
1431 			goto err_put;
1432 		}
1433 		device = xrcd->device;
1434 	} else {
1435 		if (cmd->qp_type == IB_QPT_XRC_INI) {
1436 			cmd->max_recv_wr = 0;
1437 			cmd->max_recv_sge = 0;
1438 		} else {
1439 			if (cmd->is_srq) {
1440 				srq = uobj_get_obj_read(srq, UVERBS_OBJECT_SRQ, cmd->srq_handle,
1441 							file->ucontext);
1442 				if (!srq || srq->srq_type == IB_SRQT_XRC) {
1443 					ret = -EINVAL;
1444 					goto err_put;
1445 				}
1446 			}
1447 
1448 			if (!ind_tbl) {
1449 				if (cmd->recv_cq_handle != cmd->send_cq_handle) {
1450 					rcq = uobj_get_obj_read(cq, UVERBS_OBJECT_CQ, cmd->recv_cq_handle,
1451 								file->ucontext);
1452 					if (!rcq) {
1453 						ret = -EINVAL;
1454 						goto err_put;
1455 					}
1456 				}
1457 			}
1458 		}
1459 
1460 		if (has_sq)
1461 			scq = uobj_get_obj_read(cq, UVERBS_OBJECT_CQ, cmd->send_cq_handle,
1462 						file->ucontext);
1463 		if (!ind_tbl)
1464 			rcq = rcq ?: scq;
1465 		pd  = uobj_get_obj_read(pd, UVERBS_OBJECT_PD, cmd->pd_handle, file->ucontext);
1466 		if (!pd || (!scq && has_sq)) {
1467 			ret = -EINVAL;
1468 			goto err_put;
1469 		}
1470 
1471 		device = pd->device;
1472 	}
1473 
1474 	attr.event_handler = ib_uverbs_qp_event_handler;
1475 	attr.qp_context    = file;
1476 	attr.send_cq       = scq;
1477 	attr.recv_cq       = rcq;
1478 	attr.srq           = srq;
1479 	attr.xrcd	   = xrcd;
1480 	attr.sq_sig_type   = cmd->sq_sig_all ? IB_SIGNAL_ALL_WR :
1481 					      IB_SIGNAL_REQ_WR;
1482 	attr.qp_type       = cmd->qp_type;
1483 	attr.create_flags  = 0;
1484 
1485 	attr.cap.max_send_wr     = cmd->max_send_wr;
1486 	attr.cap.max_recv_wr     = cmd->max_recv_wr;
1487 	attr.cap.max_send_sge    = cmd->max_send_sge;
1488 	attr.cap.max_recv_sge    = cmd->max_recv_sge;
1489 	attr.cap.max_inline_data = cmd->max_inline_data;
1490 
1491 	obj->uevent.events_reported     = 0;
1492 	INIT_LIST_HEAD(&obj->uevent.event_list);
1493 	INIT_LIST_HEAD(&obj->mcast_list);
1494 
1495 	if (cmd_sz >= offsetof(typeof(*cmd), create_flags) +
1496 		      sizeof(cmd->create_flags))
1497 		attr.create_flags = cmd->create_flags;
1498 
1499 	if (attr.create_flags & ~(IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK |
1500 				IB_QP_CREATE_CROSS_CHANNEL |
1501 				IB_QP_CREATE_MANAGED_SEND |
1502 				IB_QP_CREATE_MANAGED_RECV |
1503 				IB_QP_CREATE_SCATTER_FCS |
1504 				IB_QP_CREATE_CVLAN_STRIPPING |
1505 				IB_QP_CREATE_SOURCE_QPN |
1506 				IB_QP_CREATE_PCI_WRITE_END_PADDING)) {
1507 		ret = -EINVAL;
1508 		goto err_put;
1509 	}
1510 
1511 	if (attr.create_flags & IB_QP_CREATE_SOURCE_QPN) {
1512 		if (!capable(CAP_NET_RAW)) {
1513 			ret = -EPERM;
1514 			goto err_put;
1515 		}
1516 
1517 		attr.source_qpn = cmd->source_qpn;
1518 	}
1519 
1520 	buf = (void *)cmd + sizeof(*cmd);
1521 	if (cmd_sz > sizeof(*cmd))
1522 		if (!(buf[0] == 0 && !memcmp(buf, buf + 1,
1523 					     cmd_sz - sizeof(*cmd) - 1))) {
1524 			ret = -EINVAL;
1525 			goto err_put;
1526 		}
1527 
1528 	if (cmd->qp_type == IB_QPT_XRC_TGT)
1529 		qp = ib_create_qp(pd, &attr);
1530 	else
1531 		qp = _ib_create_qp(device, pd, &attr, uhw,
1532 				   &obj->uevent.uobject);
1533 
1534 	if (IS_ERR(qp)) {
1535 		ret = PTR_ERR(qp);
1536 		goto err_put;
1537 	}
1538 
1539 	if (cmd->qp_type != IB_QPT_XRC_TGT) {
1540 		ret = ib_create_qp_security(qp, device);
1541 		if (ret)
1542 			goto err_cb;
1543 
1544 		qp->real_qp	  = qp;
1545 		qp->pd		  = pd;
1546 		qp->send_cq	  = attr.send_cq;
1547 		qp->recv_cq	  = attr.recv_cq;
1548 		qp->srq		  = attr.srq;
1549 		qp->rwq_ind_tbl	  = ind_tbl;
1550 		qp->event_handler = attr.event_handler;
1551 		qp->qp_context	  = attr.qp_context;
1552 		qp->qp_type	  = attr.qp_type;
1553 		atomic_set(&qp->usecnt, 0);
1554 		atomic_inc(&pd->usecnt);
1555 		qp->port = 0;
1556 		if (attr.send_cq)
1557 			atomic_inc(&attr.send_cq->usecnt);
1558 		if (attr.recv_cq)
1559 			atomic_inc(&attr.recv_cq->usecnt);
1560 		if (attr.srq)
1561 			atomic_inc(&attr.srq->usecnt);
1562 		if (ind_tbl)
1563 			atomic_inc(&ind_tbl->usecnt);
1564 	} else {
1565 		/* It is done in _ib_create_qp for other QP types */
1566 		qp->uobject = &obj->uevent.uobject;
1567 	}
1568 
1569 	obj->uevent.uobject.object = qp;
1570 
1571 	memset(&resp, 0, sizeof resp);
1572 	resp.base.qpn             = qp->qp_num;
1573 	resp.base.qp_handle       = obj->uevent.uobject.id;
1574 	resp.base.max_recv_sge    = attr.cap.max_recv_sge;
1575 	resp.base.max_send_sge    = attr.cap.max_send_sge;
1576 	resp.base.max_recv_wr     = attr.cap.max_recv_wr;
1577 	resp.base.max_send_wr     = attr.cap.max_send_wr;
1578 	resp.base.max_inline_data = attr.cap.max_inline_data;
1579 
1580 	resp.response_length = offsetof(typeof(resp), response_length) +
1581 			       sizeof(resp.response_length);
1582 
1583 	ret = cb(file, &resp, ucore);
1584 	if (ret)
1585 		goto err_cb;
1586 
1587 	if (xrcd) {
1588 		obj->uxrcd = container_of(xrcd_uobj, struct ib_uxrcd_object,
1589 					  uobject);
1590 		atomic_inc(&obj->uxrcd->refcnt);
1591 		uobj_put_read(xrcd_uobj);
1592 	}
1593 
1594 	if (pd)
1595 		uobj_put_obj_read(pd);
1596 	if (scq)
1597 		uobj_put_obj_read(scq);
1598 	if (rcq && rcq != scq)
1599 		uobj_put_obj_read(rcq);
1600 	if (srq)
1601 		uobj_put_obj_read(srq);
1602 	if (ind_tbl)
1603 		uobj_put_obj_read(ind_tbl);
1604 
1605 	uobj_alloc_commit(&obj->uevent.uobject);
1606 
1607 	return 0;
1608 err_cb:
1609 	ib_destroy_qp(qp);
1610 
1611 err_put:
1612 	if (!IS_ERR(xrcd_uobj))
1613 		uobj_put_read(xrcd_uobj);
1614 	if (pd)
1615 		uobj_put_obj_read(pd);
1616 	if (scq)
1617 		uobj_put_obj_read(scq);
1618 	if (rcq && rcq != scq)
1619 		uobj_put_obj_read(rcq);
1620 	if (srq)
1621 		uobj_put_obj_read(srq);
1622 	if (ind_tbl)
1623 		uobj_put_obj_read(ind_tbl);
1624 
1625 	uobj_alloc_abort(&obj->uevent.uobject);
1626 	return ret;
1627 }
1628 
1629 static int ib_uverbs_create_qp_cb(struct ib_uverbs_file *file,
1630 				  struct ib_uverbs_ex_create_qp_resp *resp,
1631 				  struct ib_udata *ucore)
1632 {
1633 	if (ib_copy_to_udata(ucore, &resp->base, sizeof(resp->base)))
1634 		return -EFAULT;
1635 
1636 	return 0;
1637 }
1638 
1639 ssize_t ib_uverbs_create_qp(struct ib_uverbs_file *file,
1640 			    struct ib_device *ib_dev,
1641 			    const char __user *buf, int in_len,
1642 			    int out_len)
1643 {
1644 	struct ib_uverbs_create_qp      cmd;
1645 	struct ib_uverbs_ex_create_qp	cmd_ex;
1646 	struct ib_udata			ucore;
1647 	struct ib_udata			uhw;
1648 	ssize_t resp_size = sizeof(struct ib_uverbs_create_qp_resp);
1649 	int				err;
1650 
1651 	if (out_len < resp_size)
1652 		return -ENOSPC;
1653 
1654 	if (copy_from_user(&cmd, buf, sizeof(cmd)))
1655 		return -EFAULT;
1656 
1657 	ib_uverbs_init_udata(&ucore, buf, u64_to_user_ptr(cmd.response),
1658 		   sizeof(cmd), resp_size);
1659 	ib_uverbs_init_udata(&uhw, buf + sizeof(cmd),
1660 		   u64_to_user_ptr(cmd.response) + resp_size,
1661 		   in_len - sizeof(cmd) - sizeof(struct ib_uverbs_cmd_hdr),
1662 		   out_len - resp_size);
1663 
1664 	memset(&cmd_ex, 0, sizeof(cmd_ex));
1665 	cmd_ex.user_handle = cmd.user_handle;
1666 	cmd_ex.pd_handle = cmd.pd_handle;
1667 	cmd_ex.send_cq_handle = cmd.send_cq_handle;
1668 	cmd_ex.recv_cq_handle = cmd.recv_cq_handle;
1669 	cmd_ex.srq_handle = cmd.srq_handle;
1670 	cmd_ex.max_send_wr = cmd.max_send_wr;
1671 	cmd_ex.max_recv_wr = cmd.max_recv_wr;
1672 	cmd_ex.max_send_sge = cmd.max_send_sge;
1673 	cmd_ex.max_recv_sge = cmd.max_recv_sge;
1674 	cmd_ex.max_inline_data = cmd.max_inline_data;
1675 	cmd_ex.sq_sig_all = cmd.sq_sig_all;
1676 	cmd_ex.qp_type = cmd.qp_type;
1677 	cmd_ex.is_srq = cmd.is_srq;
1678 
1679 	err = create_qp(file, &ucore, &uhw, &cmd_ex,
1680 			offsetof(typeof(cmd_ex), is_srq) +
1681 			sizeof(cmd.is_srq), ib_uverbs_create_qp_cb,
1682 			NULL);
1683 
1684 	if (err)
1685 		return err;
1686 
1687 	return in_len;
1688 }
1689 
1690 static int ib_uverbs_ex_create_qp_cb(struct ib_uverbs_file *file,
1691 				     struct ib_uverbs_ex_create_qp_resp *resp,
1692 				     struct ib_udata *ucore)
1693 {
1694 	if (ib_copy_to_udata(ucore, resp, resp->response_length))
1695 		return -EFAULT;
1696 
1697 	return 0;
1698 }
1699 
1700 int ib_uverbs_ex_create_qp(struct ib_uverbs_file *file,
1701 			   struct ib_device *ib_dev,
1702 			   struct ib_udata *ucore,
1703 			   struct ib_udata *uhw)
1704 {
1705 	struct ib_uverbs_ex_create_qp_resp resp;
1706 	struct ib_uverbs_ex_create_qp cmd = {0};
1707 	int err;
1708 
1709 	if (ucore->inlen < (offsetof(typeof(cmd), comp_mask) +
1710 			    sizeof(cmd.comp_mask)))
1711 		return -EINVAL;
1712 
1713 	err = ib_copy_from_udata(&cmd, ucore, min(sizeof(cmd), ucore->inlen));
1714 	if (err)
1715 		return err;
1716 
1717 	if (cmd.comp_mask & ~IB_UVERBS_CREATE_QP_SUP_COMP_MASK)
1718 		return -EINVAL;
1719 
1720 	if (cmd.reserved)
1721 		return -EINVAL;
1722 
1723 	if (ucore->outlen < (offsetof(typeof(resp), response_length) +
1724 			     sizeof(resp.response_length)))
1725 		return -ENOSPC;
1726 
1727 	err = create_qp(file, ucore, uhw, &cmd,
1728 			min(ucore->inlen, sizeof(cmd)),
1729 			ib_uverbs_ex_create_qp_cb, NULL);
1730 
1731 	if (err)
1732 		return err;
1733 
1734 	return 0;
1735 }
1736 
1737 ssize_t ib_uverbs_open_qp(struct ib_uverbs_file *file,
1738 			  struct ib_device *ib_dev,
1739 			  const char __user *buf, int in_len, int out_len)
1740 {
1741 	struct ib_uverbs_open_qp        cmd;
1742 	struct ib_uverbs_create_qp_resp resp;
1743 	struct ib_udata                 udata;
1744 	struct ib_uqp_object           *obj;
1745 	struct ib_xrcd		       *xrcd;
1746 	struct ib_uobject	       *uninitialized_var(xrcd_uobj);
1747 	struct ib_qp                   *qp;
1748 	struct ib_qp_open_attr          attr;
1749 	int ret;
1750 
1751 	if (out_len < sizeof resp)
1752 		return -ENOSPC;
1753 
1754 	if (copy_from_user(&cmd, buf, sizeof cmd))
1755 		return -EFAULT;
1756 
1757 	ib_uverbs_init_udata(&udata, buf + sizeof(cmd),
1758 		   u64_to_user_ptr(cmd.response) + sizeof(resp),
1759 		   in_len - sizeof(cmd) - sizeof(struct ib_uverbs_cmd_hdr),
1760 		   out_len - sizeof(resp));
1761 
1762 	obj  = (struct ib_uqp_object *)uobj_alloc(UVERBS_OBJECT_QP,
1763 						  file->ucontext);
1764 	if (IS_ERR(obj))
1765 		return PTR_ERR(obj);
1766 
1767 	xrcd_uobj = uobj_get_read(UVERBS_OBJECT_XRCD, cmd.pd_handle,
1768 				  file->ucontext);
1769 	if (IS_ERR(xrcd_uobj)) {
1770 		ret = -EINVAL;
1771 		goto err_put;
1772 	}
1773 
1774 	xrcd = (struct ib_xrcd *)xrcd_uobj->object;
1775 	if (!xrcd) {
1776 		ret = -EINVAL;
1777 		goto err_xrcd;
1778 	}
1779 
1780 	attr.event_handler = ib_uverbs_qp_event_handler;
1781 	attr.qp_context    = file;
1782 	attr.qp_num        = cmd.qpn;
1783 	attr.qp_type       = cmd.qp_type;
1784 
1785 	obj->uevent.events_reported = 0;
1786 	INIT_LIST_HEAD(&obj->uevent.event_list);
1787 	INIT_LIST_HEAD(&obj->mcast_list);
1788 
1789 	qp = ib_open_qp(xrcd, &attr);
1790 	if (IS_ERR(qp)) {
1791 		ret = PTR_ERR(qp);
1792 		goto err_xrcd;
1793 	}
1794 
1795 	obj->uevent.uobject.object = qp;
1796 	obj->uevent.uobject.user_handle = cmd.user_handle;
1797 
1798 	memset(&resp, 0, sizeof resp);
1799 	resp.qpn       = qp->qp_num;
1800 	resp.qp_handle = obj->uevent.uobject.id;
1801 
1802 	if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof resp)) {
1803 		ret = -EFAULT;
1804 		goto err_destroy;
1805 	}
1806 
1807 	obj->uxrcd = container_of(xrcd_uobj, struct ib_uxrcd_object, uobject);
1808 	atomic_inc(&obj->uxrcd->refcnt);
1809 	qp->uobject = &obj->uevent.uobject;
1810 	uobj_put_read(xrcd_uobj);
1811 
1812 
1813 	uobj_alloc_commit(&obj->uevent.uobject);
1814 
1815 	return in_len;
1816 
1817 err_destroy:
1818 	ib_destroy_qp(qp);
1819 err_xrcd:
1820 	uobj_put_read(xrcd_uobj);
1821 err_put:
1822 	uobj_alloc_abort(&obj->uevent.uobject);
1823 	return ret;
1824 }
1825 
1826 static void copy_ah_attr_to_uverbs(struct ib_uverbs_qp_dest *uverb_attr,
1827 				   struct rdma_ah_attr *rdma_attr)
1828 {
1829 	const struct ib_global_route   *grh;
1830 
1831 	uverb_attr->dlid              = rdma_ah_get_dlid(rdma_attr);
1832 	uverb_attr->sl                = rdma_ah_get_sl(rdma_attr);
1833 	uverb_attr->src_path_bits     = rdma_ah_get_path_bits(rdma_attr);
1834 	uverb_attr->static_rate       = rdma_ah_get_static_rate(rdma_attr);
1835 	uverb_attr->is_global         = !!(rdma_ah_get_ah_flags(rdma_attr) &
1836 					 IB_AH_GRH);
1837 	if (uverb_attr->is_global) {
1838 		grh = rdma_ah_read_grh(rdma_attr);
1839 		memcpy(uverb_attr->dgid, grh->dgid.raw, 16);
1840 		uverb_attr->flow_label        = grh->flow_label;
1841 		uverb_attr->sgid_index        = grh->sgid_index;
1842 		uverb_attr->hop_limit         = grh->hop_limit;
1843 		uverb_attr->traffic_class     = grh->traffic_class;
1844 	}
1845 	uverb_attr->port_num          = rdma_ah_get_port_num(rdma_attr);
1846 }
1847 
1848 ssize_t ib_uverbs_query_qp(struct ib_uverbs_file *file,
1849 			   struct ib_device *ib_dev,
1850 			   const char __user *buf, int in_len,
1851 			   int out_len)
1852 {
1853 	struct ib_uverbs_query_qp      cmd;
1854 	struct ib_uverbs_query_qp_resp resp;
1855 	struct ib_qp                   *qp;
1856 	struct ib_qp_attr              *attr;
1857 	struct ib_qp_init_attr         *init_attr;
1858 	int                            ret;
1859 
1860 	if (copy_from_user(&cmd, buf, sizeof cmd))
1861 		return -EFAULT;
1862 
1863 	attr      = kmalloc(sizeof *attr, GFP_KERNEL);
1864 	init_attr = kmalloc(sizeof *init_attr, GFP_KERNEL);
1865 	if (!attr || !init_attr) {
1866 		ret = -ENOMEM;
1867 		goto out;
1868 	}
1869 
1870 	qp = uobj_get_obj_read(qp, UVERBS_OBJECT_QP, cmd.qp_handle, file->ucontext);
1871 	if (!qp) {
1872 		ret = -EINVAL;
1873 		goto out;
1874 	}
1875 
1876 	ret = ib_query_qp(qp, attr, cmd.attr_mask, init_attr);
1877 
1878 	uobj_put_obj_read(qp);
1879 
1880 	if (ret)
1881 		goto out;
1882 
1883 	memset(&resp, 0, sizeof resp);
1884 
1885 	resp.qp_state               = attr->qp_state;
1886 	resp.cur_qp_state           = attr->cur_qp_state;
1887 	resp.path_mtu               = attr->path_mtu;
1888 	resp.path_mig_state         = attr->path_mig_state;
1889 	resp.qkey                   = attr->qkey;
1890 	resp.rq_psn                 = attr->rq_psn;
1891 	resp.sq_psn                 = attr->sq_psn;
1892 	resp.dest_qp_num            = attr->dest_qp_num;
1893 	resp.qp_access_flags        = attr->qp_access_flags;
1894 	resp.pkey_index             = attr->pkey_index;
1895 	resp.alt_pkey_index         = attr->alt_pkey_index;
1896 	resp.sq_draining            = attr->sq_draining;
1897 	resp.max_rd_atomic          = attr->max_rd_atomic;
1898 	resp.max_dest_rd_atomic     = attr->max_dest_rd_atomic;
1899 	resp.min_rnr_timer          = attr->min_rnr_timer;
1900 	resp.port_num               = attr->port_num;
1901 	resp.timeout                = attr->timeout;
1902 	resp.retry_cnt              = attr->retry_cnt;
1903 	resp.rnr_retry              = attr->rnr_retry;
1904 	resp.alt_port_num           = attr->alt_port_num;
1905 	resp.alt_timeout            = attr->alt_timeout;
1906 
1907 	copy_ah_attr_to_uverbs(&resp.dest, &attr->ah_attr);
1908 	copy_ah_attr_to_uverbs(&resp.alt_dest, &attr->alt_ah_attr);
1909 
1910 	resp.max_send_wr            = init_attr->cap.max_send_wr;
1911 	resp.max_recv_wr            = init_attr->cap.max_recv_wr;
1912 	resp.max_send_sge           = init_attr->cap.max_send_sge;
1913 	resp.max_recv_sge           = init_attr->cap.max_recv_sge;
1914 	resp.max_inline_data        = init_attr->cap.max_inline_data;
1915 	resp.sq_sig_all             = init_attr->sq_sig_type == IB_SIGNAL_ALL_WR;
1916 
1917 	if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof resp))
1918 		ret = -EFAULT;
1919 
1920 out:
1921 	kfree(attr);
1922 	kfree(init_attr);
1923 
1924 	return ret ? ret : in_len;
1925 }
1926 
1927 /* Remove ignored fields set in the attribute mask */
1928 static int modify_qp_mask(enum ib_qp_type qp_type, int mask)
1929 {
1930 	switch (qp_type) {
1931 	case IB_QPT_XRC_INI:
1932 		return mask & ~(IB_QP_MAX_DEST_RD_ATOMIC | IB_QP_MIN_RNR_TIMER);
1933 	case IB_QPT_XRC_TGT:
1934 		return mask & ~(IB_QP_MAX_QP_RD_ATOMIC | IB_QP_RETRY_CNT |
1935 				IB_QP_RNR_RETRY);
1936 	default:
1937 		return mask;
1938 	}
1939 }
1940 
1941 static void copy_ah_attr_from_uverbs(struct ib_device *dev,
1942 				     struct rdma_ah_attr *rdma_attr,
1943 				     struct ib_uverbs_qp_dest *uverb_attr)
1944 {
1945 	rdma_attr->type = rdma_ah_find_type(dev, uverb_attr->port_num);
1946 	if (uverb_attr->is_global) {
1947 		rdma_ah_set_grh(rdma_attr, NULL,
1948 				uverb_attr->flow_label,
1949 				uverb_attr->sgid_index,
1950 				uverb_attr->hop_limit,
1951 				uverb_attr->traffic_class);
1952 		rdma_ah_set_dgid_raw(rdma_attr, uverb_attr->dgid);
1953 	} else {
1954 		rdma_ah_set_ah_flags(rdma_attr, 0);
1955 	}
1956 	rdma_ah_set_dlid(rdma_attr, uverb_attr->dlid);
1957 	rdma_ah_set_sl(rdma_attr, uverb_attr->sl);
1958 	rdma_ah_set_path_bits(rdma_attr, uverb_attr->src_path_bits);
1959 	rdma_ah_set_static_rate(rdma_attr, uverb_attr->static_rate);
1960 	rdma_ah_set_port_num(rdma_attr, uverb_attr->port_num);
1961 	rdma_ah_set_make_grd(rdma_attr, false);
1962 }
1963 
1964 static int modify_qp(struct ib_uverbs_file *file,
1965 		     struct ib_uverbs_ex_modify_qp *cmd, struct ib_udata *udata)
1966 {
1967 	struct ib_qp_attr *attr;
1968 	struct ib_qp *qp;
1969 	int ret;
1970 
1971 	attr = kmalloc(sizeof *attr, GFP_KERNEL);
1972 	if (!attr)
1973 		return -ENOMEM;
1974 
1975 	qp = uobj_get_obj_read(qp, UVERBS_OBJECT_QP, cmd->base.qp_handle, file->ucontext);
1976 	if (!qp) {
1977 		ret = -EINVAL;
1978 		goto out;
1979 	}
1980 
1981 	if ((cmd->base.attr_mask & IB_QP_PORT) &&
1982 	    !rdma_is_port_valid(qp->device, cmd->base.port_num)) {
1983 		ret = -EINVAL;
1984 		goto release_qp;
1985 	}
1986 
1987 	if ((cmd->base.attr_mask & IB_QP_AV)) {
1988 		if (!rdma_is_port_valid(qp->device, cmd->base.dest.port_num)) {
1989 			ret = -EINVAL;
1990 			goto release_qp;
1991 		}
1992 
1993 		if (cmd->base.attr_mask & IB_QP_STATE &&
1994 		    cmd->base.qp_state == IB_QPS_RTR) {
1995 		/* We are in INIT->RTR TRANSITION (if we are not,
1996 		 * this transition will be rejected in subsequent checks).
1997 		 * In the INIT->RTR transition, we cannot have IB_QP_PORT set,
1998 		 * but the IB_QP_STATE flag is required.
1999 		 *
2000 		 * Since kernel 3.14 (commit dbf727de7440), the uverbs driver,
2001 		 * when IB_QP_AV is set, has required inclusion of a valid
2002 		 * port number in the primary AV. (AVs are created and handled
2003 		 * differently for infiniband and ethernet (RoCE) ports).
2004 		 *
2005 		 * Check the port number included in the primary AV against
2006 		 * the port number in the qp struct, which was set (and saved)
2007 		 * in the RST->INIT transition.
2008 		 */
2009 			if (cmd->base.dest.port_num != qp->real_qp->port) {
2010 				ret = -EINVAL;
2011 				goto release_qp;
2012 			}
2013 		} else {
2014 		/* We are in SQD->SQD. (If we are not, this transition will
2015 		 * be rejected later in the verbs layer checks).
2016 		 * Check for both IB_QP_PORT and IB_QP_AV, these can be set
2017 		 * together in the SQD->SQD transition.
2018 		 *
2019 		 * If only IP_QP_AV was set, add in IB_QP_PORT as well (the
2020 		 * verbs layer driver does not track primary port changes
2021 		 * resulting from path migration. Thus, in SQD, if the primary
2022 		 * AV is modified, the primary port should also be modified).
2023 		 *
2024 		 * Note that in this transition, the IB_QP_STATE flag
2025 		 * is not allowed.
2026 		 */
2027 			if (((cmd->base.attr_mask & (IB_QP_AV | IB_QP_PORT))
2028 			     == (IB_QP_AV | IB_QP_PORT)) &&
2029 			    cmd->base.port_num != cmd->base.dest.port_num) {
2030 				ret = -EINVAL;
2031 				goto release_qp;
2032 			}
2033 			if ((cmd->base.attr_mask & (IB_QP_AV | IB_QP_PORT))
2034 			    == IB_QP_AV) {
2035 				cmd->base.attr_mask |= IB_QP_PORT;
2036 				cmd->base.port_num = cmd->base.dest.port_num;
2037 			}
2038 		}
2039 	}
2040 
2041 	if ((cmd->base.attr_mask & IB_QP_ALT_PATH) &&
2042 	    (!rdma_is_port_valid(qp->device, cmd->base.alt_port_num) ||
2043 	    !rdma_is_port_valid(qp->device, cmd->base.alt_dest.port_num) ||
2044 	    cmd->base.alt_port_num != cmd->base.alt_dest.port_num)) {
2045 		ret = -EINVAL;
2046 		goto release_qp;
2047 	}
2048 
2049 	if ((cmd->base.attr_mask & IB_QP_CUR_STATE &&
2050 	    cmd->base.cur_qp_state > IB_QPS_ERR) ||
2051 	    cmd->base.qp_state > IB_QPS_ERR) {
2052 		ret = -EINVAL;
2053 		goto release_qp;
2054 	}
2055 
2056 	attr->qp_state		  = cmd->base.qp_state;
2057 	attr->cur_qp_state	  = cmd->base.cur_qp_state;
2058 	attr->path_mtu		  = cmd->base.path_mtu;
2059 	attr->path_mig_state	  = cmd->base.path_mig_state;
2060 	attr->qkey		  = cmd->base.qkey;
2061 	attr->rq_psn		  = cmd->base.rq_psn;
2062 	attr->sq_psn		  = cmd->base.sq_psn;
2063 	attr->dest_qp_num	  = cmd->base.dest_qp_num;
2064 	attr->qp_access_flags	  = cmd->base.qp_access_flags;
2065 	attr->pkey_index	  = cmd->base.pkey_index;
2066 	attr->alt_pkey_index	  = cmd->base.alt_pkey_index;
2067 	attr->en_sqd_async_notify = cmd->base.en_sqd_async_notify;
2068 	attr->max_rd_atomic	  = cmd->base.max_rd_atomic;
2069 	attr->max_dest_rd_atomic  = cmd->base.max_dest_rd_atomic;
2070 	attr->min_rnr_timer	  = cmd->base.min_rnr_timer;
2071 	attr->port_num		  = cmd->base.port_num;
2072 	attr->timeout		  = cmd->base.timeout;
2073 	attr->retry_cnt		  = cmd->base.retry_cnt;
2074 	attr->rnr_retry		  = cmd->base.rnr_retry;
2075 	attr->alt_port_num	  = cmd->base.alt_port_num;
2076 	attr->alt_timeout	  = cmd->base.alt_timeout;
2077 	attr->rate_limit	  = cmd->rate_limit;
2078 
2079 	if (cmd->base.attr_mask & IB_QP_AV)
2080 		copy_ah_attr_from_uverbs(qp->device, &attr->ah_attr,
2081 					 &cmd->base.dest);
2082 
2083 	if (cmd->base.attr_mask & IB_QP_ALT_PATH)
2084 		copy_ah_attr_from_uverbs(qp->device, &attr->alt_ah_attr,
2085 					 &cmd->base.alt_dest);
2086 
2087 	ret = ib_modify_qp_with_udata(qp, attr,
2088 				      modify_qp_mask(qp->qp_type,
2089 						     cmd->base.attr_mask),
2090 				      udata);
2091 
2092 release_qp:
2093 	uobj_put_obj_read(qp);
2094 out:
2095 	kfree(attr);
2096 
2097 	return ret;
2098 }
2099 
2100 ssize_t ib_uverbs_modify_qp(struct ib_uverbs_file *file,
2101 			    struct ib_device *ib_dev,
2102 			    const char __user *buf, int in_len,
2103 			    int out_len)
2104 {
2105 	struct ib_uverbs_ex_modify_qp cmd = {};
2106 	struct ib_udata udata;
2107 	int ret;
2108 
2109 	if (copy_from_user(&cmd.base, buf, sizeof(cmd.base)))
2110 		return -EFAULT;
2111 
2112 	if (cmd.base.attr_mask &
2113 	    ~((IB_USER_LEGACY_LAST_QP_ATTR_MASK << 1) - 1))
2114 		return -EOPNOTSUPP;
2115 
2116 	ib_uverbs_init_udata(&udata, buf + sizeof(cmd.base), NULL,
2117 		   in_len - sizeof(cmd.base) - sizeof(struct ib_uverbs_cmd_hdr),
2118 		   out_len);
2119 
2120 	ret = modify_qp(file, &cmd, &udata);
2121 	if (ret)
2122 		return ret;
2123 
2124 	return in_len;
2125 }
2126 
2127 int ib_uverbs_ex_modify_qp(struct ib_uverbs_file *file,
2128 			   struct ib_device *ib_dev,
2129 			   struct ib_udata *ucore,
2130 			   struct ib_udata *uhw)
2131 {
2132 	struct ib_uverbs_ex_modify_qp cmd = {};
2133 	int ret;
2134 
2135 	/*
2136 	 * Last bit is reserved for extending the attr_mask by
2137 	 * using another field.
2138 	 */
2139 	BUILD_BUG_ON(IB_USER_LAST_QP_ATTR_MASK == (1 << 31));
2140 
2141 	if (ucore->inlen < sizeof(cmd.base))
2142 		return -EINVAL;
2143 
2144 	ret = ib_copy_from_udata(&cmd, ucore, min(sizeof(cmd), ucore->inlen));
2145 	if (ret)
2146 		return ret;
2147 
2148 	if (cmd.base.attr_mask &
2149 	    ~((IB_USER_LAST_QP_ATTR_MASK << 1) - 1))
2150 		return -EOPNOTSUPP;
2151 
2152 	if (ucore->inlen > sizeof(cmd)) {
2153 		if (!ib_is_udata_cleared(ucore, sizeof(cmd),
2154 					 ucore->inlen - sizeof(cmd)))
2155 			return -EOPNOTSUPP;
2156 	}
2157 
2158 	ret = modify_qp(file, &cmd, uhw);
2159 
2160 	return ret;
2161 }
2162 
2163 ssize_t ib_uverbs_destroy_qp(struct ib_uverbs_file *file,
2164 			     struct ib_device *ib_dev,
2165 			     const char __user *buf, int in_len,
2166 			     int out_len)
2167 {
2168 	struct ib_uverbs_destroy_qp      cmd;
2169 	struct ib_uverbs_destroy_qp_resp resp;
2170 	struct ib_uobject		*uobj;
2171 	struct ib_uqp_object        	*obj;
2172 	int                        	 ret = -EINVAL;
2173 
2174 	if (copy_from_user(&cmd, buf, sizeof cmd))
2175 		return -EFAULT;
2176 
2177 	memset(&resp, 0, sizeof resp);
2178 
2179 	uobj  = uobj_get_write(UVERBS_OBJECT_QP, cmd.qp_handle,
2180 			       file->ucontext);
2181 	if (IS_ERR(uobj))
2182 		return PTR_ERR(uobj);
2183 
2184 	obj = container_of(uobj, struct ib_uqp_object, uevent.uobject);
2185 	/*
2186 	 * Make sure we don't free the memory in remove_commit as we still
2187 	 * needs the uobject memory to create the response.
2188 	 */
2189 	uverbs_uobject_get(uobj);
2190 
2191 	ret = uobj_remove_commit(uobj);
2192 	if (ret) {
2193 		uverbs_uobject_put(uobj);
2194 		return ret;
2195 	}
2196 
2197 	resp.events_reported = obj->uevent.events_reported;
2198 	uverbs_uobject_put(uobj);
2199 
2200 	if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof resp))
2201 		return -EFAULT;
2202 
2203 	return in_len;
2204 }
2205 
2206 static void *alloc_wr(size_t wr_size, __u32 num_sge)
2207 {
2208 	if (num_sge >= (U32_MAX - ALIGN(wr_size, sizeof (struct ib_sge))) /
2209 		       sizeof (struct ib_sge))
2210 		return NULL;
2211 
2212 	return kmalloc(ALIGN(wr_size, sizeof (struct ib_sge)) +
2213 			 num_sge * sizeof (struct ib_sge), GFP_KERNEL);
2214 }
2215 
2216 ssize_t ib_uverbs_post_send(struct ib_uverbs_file *file,
2217 			    struct ib_device *ib_dev,
2218 			    const char __user *buf, int in_len,
2219 			    int out_len)
2220 {
2221 	struct ib_uverbs_post_send      cmd;
2222 	struct ib_uverbs_post_send_resp resp;
2223 	struct ib_uverbs_send_wr       *user_wr;
2224 	struct ib_send_wr              *wr = NULL, *last, *next, *bad_wr;
2225 	struct ib_qp                   *qp;
2226 	int                             i, sg_ind;
2227 	int				is_ud;
2228 	ssize_t                         ret = -EINVAL;
2229 	size_t                          next_size;
2230 
2231 	if (copy_from_user(&cmd, buf, sizeof cmd))
2232 		return -EFAULT;
2233 
2234 	if (in_len < sizeof cmd + cmd.wqe_size * cmd.wr_count +
2235 	    cmd.sge_count * sizeof (struct ib_uverbs_sge))
2236 		return -EINVAL;
2237 
2238 	if (cmd.wqe_size < sizeof (struct ib_uverbs_send_wr))
2239 		return -EINVAL;
2240 
2241 	user_wr = kmalloc(cmd.wqe_size, GFP_KERNEL);
2242 	if (!user_wr)
2243 		return -ENOMEM;
2244 
2245 	qp = uobj_get_obj_read(qp, UVERBS_OBJECT_QP, cmd.qp_handle, file->ucontext);
2246 	if (!qp)
2247 		goto out;
2248 
2249 	is_ud = qp->qp_type == IB_QPT_UD;
2250 	sg_ind = 0;
2251 	last = NULL;
2252 	for (i = 0; i < cmd.wr_count; ++i) {
2253 		if (copy_from_user(user_wr,
2254 				   buf + sizeof cmd + i * cmd.wqe_size,
2255 				   cmd.wqe_size)) {
2256 			ret = -EFAULT;
2257 			goto out_put;
2258 		}
2259 
2260 		if (user_wr->num_sge + sg_ind > cmd.sge_count) {
2261 			ret = -EINVAL;
2262 			goto out_put;
2263 		}
2264 
2265 		if (is_ud) {
2266 			struct ib_ud_wr *ud;
2267 
2268 			if (user_wr->opcode != IB_WR_SEND &&
2269 			    user_wr->opcode != IB_WR_SEND_WITH_IMM) {
2270 				ret = -EINVAL;
2271 				goto out_put;
2272 			}
2273 
2274 			next_size = sizeof(*ud);
2275 			ud = alloc_wr(next_size, user_wr->num_sge);
2276 			if (!ud) {
2277 				ret = -ENOMEM;
2278 				goto out_put;
2279 			}
2280 
2281 			ud->ah = uobj_get_obj_read(ah, UVERBS_OBJECT_AH, user_wr->wr.ud.ah,
2282 						   file->ucontext);
2283 			if (!ud->ah) {
2284 				kfree(ud);
2285 				ret = -EINVAL;
2286 				goto out_put;
2287 			}
2288 			ud->remote_qpn = user_wr->wr.ud.remote_qpn;
2289 			ud->remote_qkey = user_wr->wr.ud.remote_qkey;
2290 
2291 			next = &ud->wr;
2292 		} else if (user_wr->opcode == IB_WR_RDMA_WRITE_WITH_IMM ||
2293 			   user_wr->opcode == IB_WR_RDMA_WRITE ||
2294 			   user_wr->opcode == IB_WR_RDMA_READ) {
2295 			struct ib_rdma_wr *rdma;
2296 
2297 			next_size = sizeof(*rdma);
2298 			rdma = alloc_wr(next_size, user_wr->num_sge);
2299 			if (!rdma) {
2300 				ret = -ENOMEM;
2301 				goto out_put;
2302 			}
2303 
2304 			rdma->remote_addr = user_wr->wr.rdma.remote_addr;
2305 			rdma->rkey = user_wr->wr.rdma.rkey;
2306 
2307 			next = &rdma->wr;
2308 		} else if (user_wr->opcode == IB_WR_ATOMIC_CMP_AND_SWP ||
2309 			   user_wr->opcode == IB_WR_ATOMIC_FETCH_AND_ADD) {
2310 			struct ib_atomic_wr *atomic;
2311 
2312 			next_size = sizeof(*atomic);
2313 			atomic = alloc_wr(next_size, user_wr->num_sge);
2314 			if (!atomic) {
2315 				ret = -ENOMEM;
2316 				goto out_put;
2317 			}
2318 
2319 			atomic->remote_addr = user_wr->wr.atomic.remote_addr;
2320 			atomic->compare_add = user_wr->wr.atomic.compare_add;
2321 			atomic->swap = user_wr->wr.atomic.swap;
2322 			atomic->rkey = user_wr->wr.atomic.rkey;
2323 
2324 			next = &atomic->wr;
2325 		} else if (user_wr->opcode == IB_WR_SEND ||
2326 			   user_wr->opcode == IB_WR_SEND_WITH_IMM ||
2327 			   user_wr->opcode == IB_WR_SEND_WITH_INV) {
2328 			next_size = sizeof(*next);
2329 			next = alloc_wr(next_size, user_wr->num_sge);
2330 			if (!next) {
2331 				ret = -ENOMEM;
2332 				goto out_put;
2333 			}
2334 		} else {
2335 			ret = -EINVAL;
2336 			goto out_put;
2337 		}
2338 
2339 		if (user_wr->opcode == IB_WR_SEND_WITH_IMM ||
2340 		    user_wr->opcode == IB_WR_RDMA_WRITE_WITH_IMM) {
2341 			next->ex.imm_data =
2342 					(__be32 __force) user_wr->ex.imm_data;
2343 		} else if (user_wr->opcode == IB_WR_SEND_WITH_INV) {
2344 			next->ex.invalidate_rkey = user_wr->ex.invalidate_rkey;
2345 		}
2346 
2347 		if (!last)
2348 			wr = next;
2349 		else
2350 			last->next = next;
2351 		last = next;
2352 
2353 		next->next       = NULL;
2354 		next->wr_id      = user_wr->wr_id;
2355 		next->num_sge    = user_wr->num_sge;
2356 		next->opcode     = user_wr->opcode;
2357 		next->send_flags = user_wr->send_flags;
2358 
2359 		if (next->num_sge) {
2360 			next->sg_list = (void *) next +
2361 				ALIGN(next_size, sizeof(struct ib_sge));
2362 			if (copy_from_user(next->sg_list,
2363 					   buf + sizeof cmd +
2364 					   cmd.wr_count * cmd.wqe_size +
2365 					   sg_ind * sizeof (struct ib_sge),
2366 					   next->num_sge * sizeof (struct ib_sge))) {
2367 				ret = -EFAULT;
2368 				goto out_put;
2369 			}
2370 			sg_ind += next->num_sge;
2371 		} else
2372 			next->sg_list = NULL;
2373 	}
2374 
2375 	resp.bad_wr = 0;
2376 	ret = qp->device->post_send(qp->real_qp, wr, &bad_wr);
2377 	if (ret)
2378 		for (next = wr; next; next = next->next) {
2379 			++resp.bad_wr;
2380 			if (next == bad_wr)
2381 				break;
2382 		}
2383 
2384 	if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof resp))
2385 		ret = -EFAULT;
2386 
2387 out_put:
2388 	uobj_put_obj_read(qp);
2389 
2390 	while (wr) {
2391 		if (is_ud && ud_wr(wr)->ah)
2392 			uobj_put_obj_read(ud_wr(wr)->ah);
2393 		next = wr->next;
2394 		kfree(wr);
2395 		wr = next;
2396 	}
2397 
2398 out:
2399 	kfree(user_wr);
2400 
2401 	return ret ? ret : in_len;
2402 }
2403 
2404 static struct ib_recv_wr *ib_uverbs_unmarshall_recv(const char __user *buf,
2405 						    int in_len,
2406 						    u32 wr_count,
2407 						    u32 sge_count,
2408 						    u32 wqe_size)
2409 {
2410 	struct ib_uverbs_recv_wr *user_wr;
2411 	struct ib_recv_wr        *wr = NULL, *last, *next;
2412 	int                       sg_ind;
2413 	int                       i;
2414 	int                       ret;
2415 
2416 	if (in_len < wqe_size * wr_count +
2417 	    sge_count * sizeof (struct ib_uverbs_sge))
2418 		return ERR_PTR(-EINVAL);
2419 
2420 	if (wqe_size < sizeof (struct ib_uverbs_recv_wr))
2421 		return ERR_PTR(-EINVAL);
2422 
2423 	user_wr = kmalloc(wqe_size, GFP_KERNEL);
2424 	if (!user_wr)
2425 		return ERR_PTR(-ENOMEM);
2426 
2427 	sg_ind = 0;
2428 	last = NULL;
2429 	for (i = 0; i < wr_count; ++i) {
2430 		if (copy_from_user(user_wr, buf + i * wqe_size,
2431 				   wqe_size)) {
2432 			ret = -EFAULT;
2433 			goto err;
2434 		}
2435 
2436 		if (user_wr->num_sge + sg_ind > sge_count) {
2437 			ret = -EINVAL;
2438 			goto err;
2439 		}
2440 
2441 		if (user_wr->num_sge >=
2442 		    (U32_MAX - ALIGN(sizeof *next, sizeof (struct ib_sge))) /
2443 		    sizeof (struct ib_sge)) {
2444 			ret = -EINVAL;
2445 			goto err;
2446 		}
2447 
2448 		next = kmalloc(ALIGN(sizeof *next, sizeof (struct ib_sge)) +
2449 			       user_wr->num_sge * sizeof (struct ib_sge),
2450 			       GFP_KERNEL);
2451 		if (!next) {
2452 			ret = -ENOMEM;
2453 			goto err;
2454 		}
2455 
2456 		if (!last)
2457 			wr = next;
2458 		else
2459 			last->next = next;
2460 		last = next;
2461 
2462 		next->next       = NULL;
2463 		next->wr_id      = user_wr->wr_id;
2464 		next->num_sge    = user_wr->num_sge;
2465 
2466 		if (next->num_sge) {
2467 			next->sg_list = (void *) next +
2468 				ALIGN(sizeof *next, sizeof (struct ib_sge));
2469 			if (copy_from_user(next->sg_list,
2470 					   buf + wr_count * wqe_size +
2471 					   sg_ind * sizeof (struct ib_sge),
2472 					   next->num_sge * sizeof (struct ib_sge))) {
2473 				ret = -EFAULT;
2474 				goto err;
2475 			}
2476 			sg_ind += next->num_sge;
2477 		} else
2478 			next->sg_list = NULL;
2479 	}
2480 
2481 	kfree(user_wr);
2482 	return wr;
2483 
2484 err:
2485 	kfree(user_wr);
2486 
2487 	while (wr) {
2488 		next = wr->next;
2489 		kfree(wr);
2490 		wr = next;
2491 	}
2492 
2493 	return ERR_PTR(ret);
2494 }
2495 
2496 ssize_t ib_uverbs_post_recv(struct ib_uverbs_file *file,
2497 			    struct ib_device *ib_dev,
2498 			    const char __user *buf, int in_len,
2499 			    int out_len)
2500 {
2501 	struct ib_uverbs_post_recv      cmd;
2502 	struct ib_uverbs_post_recv_resp resp;
2503 	struct ib_recv_wr              *wr, *next, *bad_wr;
2504 	struct ib_qp                   *qp;
2505 	ssize_t                         ret = -EINVAL;
2506 
2507 	if (copy_from_user(&cmd, buf, sizeof cmd))
2508 		return -EFAULT;
2509 
2510 	wr = ib_uverbs_unmarshall_recv(buf + sizeof cmd,
2511 				       in_len - sizeof cmd, cmd.wr_count,
2512 				       cmd.sge_count, cmd.wqe_size);
2513 	if (IS_ERR(wr))
2514 		return PTR_ERR(wr);
2515 
2516 	qp = uobj_get_obj_read(qp, UVERBS_OBJECT_QP, cmd.qp_handle, file->ucontext);
2517 	if (!qp)
2518 		goto out;
2519 
2520 	resp.bad_wr = 0;
2521 	ret = qp->device->post_recv(qp->real_qp, wr, &bad_wr);
2522 
2523 	uobj_put_obj_read(qp);
2524 	if (ret) {
2525 		for (next = wr; next; next = next->next) {
2526 			++resp.bad_wr;
2527 			if (next == bad_wr)
2528 				break;
2529 		}
2530 	}
2531 
2532 	if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof resp))
2533 		ret = -EFAULT;
2534 
2535 out:
2536 	while (wr) {
2537 		next = wr->next;
2538 		kfree(wr);
2539 		wr = next;
2540 	}
2541 
2542 	return ret ? ret : in_len;
2543 }
2544 
2545 ssize_t ib_uverbs_post_srq_recv(struct ib_uverbs_file *file,
2546 				struct ib_device *ib_dev,
2547 				const char __user *buf, int in_len,
2548 				int out_len)
2549 {
2550 	struct ib_uverbs_post_srq_recv      cmd;
2551 	struct ib_uverbs_post_srq_recv_resp resp;
2552 	struct ib_recv_wr                  *wr, *next, *bad_wr;
2553 	struct ib_srq                      *srq;
2554 	ssize_t                             ret = -EINVAL;
2555 
2556 	if (copy_from_user(&cmd, buf, sizeof cmd))
2557 		return -EFAULT;
2558 
2559 	wr = ib_uverbs_unmarshall_recv(buf + sizeof cmd,
2560 				       in_len - sizeof cmd, cmd.wr_count,
2561 				       cmd.sge_count, cmd.wqe_size);
2562 	if (IS_ERR(wr))
2563 		return PTR_ERR(wr);
2564 
2565 	srq = uobj_get_obj_read(srq, UVERBS_OBJECT_SRQ, cmd.srq_handle, file->ucontext);
2566 	if (!srq)
2567 		goto out;
2568 
2569 	resp.bad_wr = 0;
2570 	ret = srq->device->post_srq_recv(srq, wr, &bad_wr);
2571 
2572 	uobj_put_obj_read(srq);
2573 
2574 	if (ret)
2575 		for (next = wr; next; next = next->next) {
2576 			++resp.bad_wr;
2577 			if (next == bad_wr)
2578 				break;
2579 		}
2580 
2581 	if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof resp))
2582 		ret = -EFAULT;
2583 
2584 out:
2585 	while (wr) {
2586 		next = wr->next;
2587 		kfree(wr);
2588 		wr = next;
2589 	}
2590 
2591 	return ret ? ret : in_len;
2592 }
2593 
2594 ssize_t ib_uverbs_create_ah(struct ib_uverbs_file *file,
2595 			    struct ib_device *ib_dev,
2596 			    const char __user *buf, int in_len,
2597 			    int out_len)
2598 {
2599 	struct ib_uverbs_create_ah	 cmd;
2600 	struct ib_uverbs_create_ah_resp	 resp;
2601 	struct ib_uobject		*uobj;
2602 	struct ib_pd			*pd;
2603 	struct ib_ah			*ah;
2604 	struct rdma_ah_attr		attr;
2605 	int ret;
2606 	struct ib_udata                   udata;
2607 
2608 	if (out_len < sizeof resp)
2609 		return -ENOSPC;
2610 
2611 	if (copy_from_user(&cmd, buf, sizeof cmd))
2612 		return -EFAULT;
2613 
2614 	if (!rdma_is_port_valid(ib_dev, cmd.attr.port_num))
2615 		return -EINVAL;
2616 
2617 	ib_uverbs_init_udata(&udata, buf + sizeof(cmd),
2618 		   u64_to_user_ptr(cmd.response) + sizeof(resp),
2619 		   in_len - sizeof(cmd) - sizeof(struct ib_uverbs_cmd_hdr),
2620 		   out_len - sizeof(resp));
2621 
2622 	uobj  = uobj_alloc(UVERBS_OBJECT_AH, file->ucontext);
2623 	if (IS_ERR(uobj))
2624 		return PTR_ERR(uobj);
2625 
2626 	pd = uobj_get_obj_read(pd, UVERBS_OBJECT_PD, cmd.pd_handle, file->ucontext);
2627 	if (!pd) {
2628 		ret = -EINVAL;
2629 		goto err;
2630 	}
2631 
2632 	attr.type = rdma_ah_find_type(ib_dev, cmd.attr.port_num);
2633 	rdma_ah_set_make_grd(&attr, false);
2634 	rdma_ah_set_dlid(&attr, cmd.attr.dlid);
2635 	rdma_ah_set_sl(&attr, cmd.attr.sl);
2636 	rdma_ah_set_path_bits(&attr, cmd.attr.src_path_bits);
2637 	rdma_ah_set_static_rate(&attr, cmd.attr.static_rate);
2638 	rdma_ah_set_port_num(&attr, cmd.attr.port_num);
2639 
2640 	if (cmd.attr.is_global) {
2641 		rdma_ah_set_grh(&attr, NULL, cmd.attr.grh.flow_label,
2642 				cmd.attr.grh.sgid_index,
2643 				cmd.attr.grh.hop_limit,
2644 				cmd.attr.grh.traffic_class);
2645 		rdma_ah_set_dgid_raw(&attr, cmd.attr.grh.dgid);
2646 	} else {
2647 		rdma_ah_set_ah_flags(&attr, 0);
2648 	}
2649 
2650 	ah = rdma_create_user_ah(pd, &attr, &udata);
2651 	if (IS_ERR(ah)) {
2652 		ret = PTR_ERR(ah);
2653 		goto err_put;
2654 	}
2655 
2656 	ah->uobject  = uobj;
2657 	uobj->user_handle = cmd.user_handle;
2658 	uobj->object = ah;
2659 
2660 	resp.ah_handle = uobj->id;
2661 
2662 	if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof resp)) {
2663 		ret = -EFAULT;
2664 		goto err_copy;
2665 	}
2666 
2667 	uobj_put_obj_read(pd);
2668 	uobj_alloc_commit(uobj);
2669 
2670 	return in_len;
2671 
2672 err_copy:
2673 	rdma_destroy_ah(ah);
2674 
2675 err_put:
2676 	uobj_put_obj_read(pd);
2677 
2678 err:
2679 	uobj_alloc_abort(uobj);
2680 	return ret;
2681 }
2682 
2683 ssize_t ib_uverbs_destroy_ah(struct ib_uverbs_file *file,
2684 			     struct ib_device *ib_dev,
2685 			     const char __user *buf, int in_len, int out_len)
2686 {
2687 	struct ib_uverbs_destroy_ah cmd;
2688 	struct ib_uobject	   *uobj;
2689 	int			    ret;
2690 
2691 	if (copy_from_user(&cmd, buf, sizeof cmd))
2692 		return -EFAULT;
2693 
2694 	uobj  = uobj_get_write(UVERBS_OBJECT_AH, cmd.ah_handle,
2695 			       file->ucontext);
2696 	if (IS_ERR(uobj))
2697 		return PTR_ERR(uobj);
2698 
2699 	ret = uobj_remove_commit(uobj);
2700 	return ret ?: in_len;
2701 }
2702 
2703 ssize_t ib_uverbs_attach_mcast(struct ib_uverbs_file *file,
2704 			       struct ib_device *ib_dev,
2705 			       const char __user *buf, int in_len,
2706 			       int out_len)
2707 {
2708 	struct ib_uverbs_attach_mcast cmd;
2709 	struct ib_qp                 *qp;
2710 	struct ib_uqp_object         *obj;
2711 	struct ib_uverbs_mcast_entry *mcast;
2712 	int                           ret;
2713 
2714 	if (copy_from_user(&cmd, buf, sizeof cmd))
2715 		return -EFAULT;
2716 
2717 	qp = uobj_get_obj_read(qp, UVERBS_OBJECT_QP, cmd.qp_handle, file->ucontext);
2718 	if (!qp)
2719 		return -EINVAL;
2720 
2721 	obj = container_of(qp->uobject, struct ib_uqp_object, uevent.uobject);
2722 
2723 	mutex_lock(&obj->mcast_lock);
2724 	list_for_each_entry(mcast, &obj->mcast_list, list)
2725 		if (cmd.mlid == mcast->lid &&
2726 		    !memcmp(cmd.gid, mcast->gid.raw, sizeof mcast->gid.raw)) {
2727 			ret = 0;
2728 			goto out_put;
2729 		}
2730 
2731 	mcast = kmalloc(sizeof *mcast, GFP_KERNEL);
2732 	if (!mcast) {
2733 		ret = -ENOMEM;
2734 		goto out_put;
2735 	}
2736 
2737 	mcast->lid = cmd.mlid;
2738 	memcpy(mcast->gid.raw, cmd.gid, sizeof mcast->gid.raw);
2739 
2740 	ret = ib_attach_mcast(qp, &mcast->gid, cmd.mlid);
2741 	if (!ret)
2742 		list_add_tail(&mcast->list, &obj->mcast_list);
2743 	else
2744 		kfree(mcast);
2745 
2746 out_put:
2747 	mutex_unlock(&obj->mcast_lock);
2748 	uobj_put_obj_read(qp);
2749 
2750 	return ret ? ret : in_len;
2751 }
2752 
2753 ssize_t ib_uverbs_detach_mcast(struct ib_uverbs_file *file,
2754 			       struct ib_device *ib_dev,
2755 			       const char __user *buf, int in_len,
2756 			       int out_len)
2757 {
2758 	struct ib_uverbs_detach_mcast cmd;
2759 	struct ib_uqp_object         *obj;
2760 	struct ib_qp                 *qp;
2761 	struct ib_uverbs_mcast_entry *mcast;
2762 	int                           ret = -EINVAL;
2763 	bool                          found = false;
2764 
2765 	if (copy_from_user(&cmd, buf, sizeof cmd))
2766 		return -EFAULT;
2767 
2768 	qp = uobj_get_obj_read(qp, UVERBS_OBJECT_QP, cmd.qp_handle, file->ucontext);
2769 	if (!qp)
2770 		return -EINVAL;
2771 
2772 	obj = container_of(qp->uobject, struct ib_uqp_object, uevent.uobject);
2773 	mutex_lock(&obj->mcast_lock);
2774 
2775 	list_for_each_entry(mcast, &obj->mcast_list, list)
2776 		if (cmd.mlid == mcast->lid &&
2777 		    !memcmp(cmd.gid, mcast->gid.raw, sizeof mcast->gid.raw)) {
2778 			list_del(&mcast->list);
2779 			kfree(mcast);
2780 			found = true;
2781 			break;
2782 		}
2783 
2784 	if (!found) {
2785 		ret = -EINVAL;
2786 		goto out_put;
2787 	}
2788 
2789 	ret = ib_detach_mcast(qp, (union ib_gid *)cmd.gid, cmd.mlid);
2790 
2791 out_put:
2792 	mutex_unlock(&obj->mcast_lock);
2793 	uobj_put_obj_read(qp);
2794 	return ret ? ret : in_len;
2795 }
2796 
2797 struct ib_uflow_resources {
2798 	size_t			max;
2799 	size_t			num;
2800 	size_t			collection_num;
2801 	size_t			counters_num;
2802 	struct ib_counters	**counters;
2803 	struct ib_flow_action	**collection;
2804 };
2805 
2806 static struct ib_uflow_resources *flow_resources_alloc(size_t num_specs)
2807 {
2808 	struct ib_uflow_resources *resources;
2809 
2810 	resources = kzalloc(sizeof(*resources), GFP_KERNEL);
2811 
2812 	if (!resources)
2813 		goto err_res;
2814 
2815 	resources->counters =
2816 		kcalloc(num_specs, sizeof(*resources->counters), GFP_KERNEL);
2817 
2818 	if (!resources->counters)
2819 		goto err_cnt;
2820 
2821 	resources->collection =
2822 		kcalloc(num_specs, sizeof(*resources->collection), GFP_KERNEL);
2823 
2824 	if (!resources->collection)
2825 		goto err_collection;
2826 
2827 	resources->max = num_specs;
2828 
2829 	return resources;
2830 
2831 err_collection:
2832 	kfree(resources->counters);
2833 err_cnt:
2834 	kfree(resources);
2835 err_res:
2836 	return NULL;
2837 }
2838 
2839 void ib_uverbs_flow_resources_free(struct ib_uflow_resources *uflow_res)
2840 {
2841 	unsigned int i;
2842 
2843 	for (i = 0; i < uflow_res->collection_num; i++)
2844 		atomic_dec(&uflow_res->collection[i]->usecnt);
2845 
2846 	for (i = 0; i < uflow_res->counters_num; i++)
2847 		atomic_dec(&uflow_res->counters[i]->usecnt);
2848 
2849 	kfree(uflow_res->collection);
2850 	kfree(uflow_res->counters);
2851 	kfree(uflow_res);
2852 }
2853 
2854 static void flow_resources_add(struct ib_uflow_resources *uflow_res,
2855 			       enum ib_flow_spec_type type,
2856 			       void *ibobj)
2857 {
2858 	WARN_ON(uflow_res->num >= uflow_res->max);
2859 
2860 	switch (type) {
2861 	case IB_FLOW_SPEC_ACTION_HANDLE:
2862 		atomic_inc(&((struct ib_flow_action *)ibobj)->usecnt);
2863 		uflow_res->collection[uflow_res->collection_num++] =
2864 			(struct ib_flow_action *)ibobj;
2865 		break;
2866 	case IB_FLOW_SPEC_ACTION_COUNT:
2867 		atomic_inc(&((struct ib_counters *)ibobj)->usecnt);
2868 		uflow_res->counters[uflow_res->counters_num++] =
2869 			(struct ib_counters *)ibobj;
2870 		break;
2871 	default:
2872 		WARN_ON(1);
2873 	}
2874 
2875 	uflow_res->num++;
2876 }
2877 
2878 static int kern_spec_to_ib_spec_action(struct ib_ucontext *ucontext,
2879 				       struct ib_uverbs_flow_spec *kern_spec,
2880 				       union ib_flow_spec *ib_spec,
2881 				       struct ib_uflow_resources *uflow_res)
2882 {
2883 	ib_spec->type = kern_spec->type;
2884 	switch (ib_spec->type) {
2885 	case IB_FLOW_SPEC_ACTION_TAG:
2886 		if (kern_spec->flow_tag.size !=
2887 		    sizeof(struct ib_uverbs_flow_spec_action_tag))
2888 			return -EINVAL;
2889 
2890 		ib_spec->flow_tag.size = sizeof(struct ib_flow_spec_action_tag);
2891 		ib_spec->flow_tag.tag_id = kern_spec->flow_tag.tag_id;
2892 		break;
2893 	case IB_FLOW_SPEC_ACTION_DROP:
2894 		if (kern_spec->drop.size !=
2895 		    sizeof(struct ib_uverbs_flow_spec_action_drop))
2896 			return -EINVAL;
2897 
2898 		ib_spec->drop.size = sizeof(struct ib_flow_spec_action_drop);
2899 		break;
2900 	case IB_FLOW_SPEC_ACTION_HANDLE:
2901 		if (kern_spec->action.size !=
2902 		    sizeof(struct ib_uverbs_flow_spec_action_handle))
2903 			return -EOPNOTSUPP;
2904 		ib_spec->action.act = uobj_get_obj_read(flow_action,
2905 							UVERBS_OBJECT_FLOW_ACTION,
2906 							kern_spec->action.handle,
2907 							ucontext);
2908 		if (!ib_spec->action.act)
2909 			return -EINVAL;
2910 		ib_spec->action.size =
2911 			sizeof(struct ib_flow_spec_action_handle);
2912 		flow_resources_add(uflow_res,
2913 				   IB_FLOW_SPEC_ACTION_HANDLE,
2914 				   ib_spec->action.act);
2915 		uobj_put_obj_read(ib_spec->action.act);
2916 		break;
2917 	case IB_FLOW_SPEC_ACTION_COUNT:
2918 		if (kern_spec->flow_count.size !=
2919 			sizeof(struct ib_uverbs_flow_spec_action_count))
2920 			return -EINVAL;
2921 		ib_spec->flow_count.counters =
2922 			uobj_get_obj_read(counters,
2923 					  UVERBS_OBJECT_COUNTERS,
2924 					  kern_spec->flow_count.handle,
2925 					  ucontext);
2926 		if (!ib_spec->flow_count.counters)
2927 			return -EINVAL;
2928 		ib_spec->flow_count.size =
2929 				sizeof(struct ib_flow_spec_action_count);
2930 		flow_resources_add(uflow_res,
2931 				   IB_FLOW_SPEC_ACTION_COUNT,
2932 				   ib_spec->flow_count.counters);
2933 		uobj_put_obj_read(ib_spec->flow_count.counters);
2934 		break;
2935 	default:
2936 		return -EINVAL;
2937 	}
2938 	return 0;
2939 }
2940 
2941 static size_t kern_spec_filter_sz(const struct ib_uverbs_flow_spec_hdr *spec)
2942 {
2943 	/* Returns user space filter size, includes padding */
2944 	return (spec->size - sizeof(struct ib_uverbs_flow_spec_hdr)) / 2;
2945 }
2946 
2947 static ssize_t spec_filter_size(const void *kern_spec_filter, u16 kern_filter_size,
2948 				u16 ib_real_filter_sz)
2949 {
2950 	/*
2951 	 * User space filter structures must be 64 bit aligned, otherwise this
2952 	 * may pass, but we won't handle additional new attributes.
2953 	 */
2954 
2955 	if (kern_filter_size > ib_real_filter_sz) {
2956 		if (memchr_inv(kern_spec_filter +
2957 			       ib_real_filter_sz, 0,
2958 			       kern_filter_size - ib_real_filter_sz))
2959 			return -EINVAL;
2960 		return ib_real_filter_sz;
2961 	}
2962 	return kern_filter_size;
2963 }
2964 
2965 int ib_uverbs_kern_spec_to_ib_spec_filter(enum ib_flow_spec_type type,
2966 					  const void *kern_spec_mask,
2967 					  const void *kern_spec_val,
2968 					  size_t kern_filter_sz,
2969 					  union ib_flow_spec *ib_spec)
2970 {
2971 	ssize_t actual_filter_sz;
2972 	ssize_t ib_filter_sz;
2973 
2974 	/* User flow spec size must be aligned to 4 bytes */
2975 	if (kern_filter_sz != ALIGN(kern_filter_sz, 4))
2976 		return -EINVAL;
2977 
2978 	ib_spec->type = type;
2979 
2980 	if (ib_spec->type == (IB_FLOW_SPEC_INNER | IB_FLOW_SPEC_VXLAN_TUNNEL))
2981 		return -EINVAL;
2982 
2983 	switch (ib_spec->type & ~IB_FLOW_SPEC_INNER) {
2984 	case IB_FLOW_SPEC_ETH:
2985 		ib_filter_sz = offsetof(struct ib_flow_eth_filter, real_sz);
2986 		actual_filter_sz = spec_filter_size(kern_spec_mask,
2987 						    kern_filter_sz,
2988 						    ib_filter_sz);
2989 		if (actual_filter_sz <= 0)
2990 			return -EINVAL;
2991 		ib_spec->size = sizeof(struct ib_flow_spec_eth);
2992 		memcpy(&ib_spec->eth.val, kern_spec_val, actual_filter_sz);
2993 		memcpy(&ib_spec->eth.mask, kern_spec_mask, actual_filter_sz);
2994 		break;
2995 	case IB_FLOW_SPEC_IPV4:
2996 		ib_filter_sz = offsetof(struct ib_flow_ipv4_filter, real_sz);
2997 		actual_filter_sz = spec_filter_size(kern_spec_mask,
2998 						    kern_filter_sz,
2999 						    ib_filter_sz);
3000 		if (actual_filter_sz <= 0)
3001 			return -EINVAL;
3002 		ib_spec->size = sizeof(struct ib_flow_spec_ipv4);
3003 		memcpy(&ib_spec->ipv4.val, kern_spec_val, actual_filter_sz);
3004 		memcpy(&ib_spec->ipv4.mask, kern_spec_mask, actual_filter_sz);
3005 		break;
3006 	case IB_FLOW_SPEC_IPV6:
3007 		ib_filter_sz = offsetof(struct ib_flow_ipv6_filter, real_sz);
3008 		actual_filter_sz = spec_filter_size(kern_spec_mask,
3009 						    kern_filter_sz,
3010 						    ib_filter_sz);
3011 		if (actual_filter_sz <= 0)
3012 			return -EINVAL;
3013 		ib_spec->size = sizeof(struct ib_flow_spec_ipv6);
3014 		memcpy(&ib_spec->ipv6.val, kern_spec_val, actual_filter_sz);
3015 		memcpy(&ib_spec->ipv6.mask, kern_spec_mask, actual_filter_sz);
3016 
3017 		if ((ntohl(ib_spec->ipv6.mask.flow_label)) >= BIT(20) ||
3018 		    (ntohl(ib_spec->ipv6.val.flow_label)) >= BIT(20))
3019 			return -EINVAL;
3020 		break;
3021 	case IB_FLOW_SPEC_TCP:
3022 	case IB_FLOW_SPEC_UDP:
3023 		ib_filter_sz = offsetof(struct ib_flow_tcp_udp_filter, real_sz);
3024 		actual_filter_sz = spec_filter_size(kern_spec_mask,
3025 						    kern_filter_sz,
3026 						    ib_filter_sz);
3027 		if (actual_filter_sz <= 0)
3028 			return -EINVAL;
3029 		ib_spec->size = sizeof(struct ib_flow_spec_tcp_udp);
3030 		memcpy(&ib_spec->tcp_udp.val, kern_spec_val, actual_filter_sz);
3031 		memcpy(&ib_spec->tcp_udp.mask, kern_spec_mask, actual_filter_sz);
3032 		break;
3033 	case IB_FLOW_SPEC_VXLAN_TUNNEL:
3034 		ib_filter_sz = offsetof(struct ib_flow_tunnel_filter, real_sz);
3035 		actual_filter_sz = spec_filter_size(kern_spec_mask,
3036 						    kern_filter_sz,
3037 						    ib_filter_sz);
3038 		if (actual_filter_sz <= 0)
3039 			return -EINVAL;
3040 		ib_spec->tunnel.size = sizeof(struct ib_flow_spec_tunnel);
3041 		memcpy(&ib_spec->tunnel.val, kern_spec_val, actual_filter_sz);
3042 		memcpy(&ib_spec->tunnel.mask, kern_spec_mask, actual_filter_sz);
3043 
3044 		if ((ntohl(ib_spec->tunnel.mask.tunnel_id)) >= BIT(24) ||
3045 		    (ntohl(ib_spec->tunnel.val.tunnel_id)) >= BIT(24))
3046 			return -EINVAL;
3047 		break;
3048 	case IB_FLOW_SPEC_ESP:
3049 		ib_filter_sz = offsetof(struct ib_flow_esp_filter, real_sz);
3050 		actual_filter_sz = spec_filter_size(kern_spec_mask,
3051 						    kern_filter_sz,
3052 						    ib_filter_sz);
3053 		if (actual_filter_sz <= 0)
3054 			return -EINVAL;
3055 		ib_spec->esp.size = sizeof(struct ib_flow_spec_esp);
3056 		memcpy(&ib_spec->esp.val, kern_spec_val, actual_filter_sz);
3057 		memcpy(&ib_spec->esp.mask, kern_spec_mask, actual_filter_sz);
3058 		break;
3059 	case IB_FLOW_SPEC_GRE:
3060 		ib_filter_sz = offsetof(struct ib_flow_gre_filter, real_sz);
3061 		actual_filter_sz = spec_filter_size(kern_spec_mask,
3062 						    kern_filter_sz,
3063 						    ib_filter_sz);
3064 		if (actual_filter_sz <= 0)
3065 			return -EINVAL;
3066 		ib_spec->gre.size = sizeof(struct ib_flow_spec_gre);
3067 		memcpy(&ib_spec->gre.val, kern_spec_val, actual_filter_sz);
3068 		memcpy(&ib_spec->gre.mask, kern_spec_mask, actual_filter_sz);
3069 		break;
3070 	case IB_FLOW_SPEC_MPLS:
3071 		ib_filter_sz = offsetof(struct ib_flow_mpls_filter, real_sz);
3072 		actual_filter_sz = spec_filter_size(kern_spec_mask,
3073 						    kern_filter_sz,
3074 						    ib_filter_sz);
3075 		if (actual_filter_sz <= 0)
3076 			return -EINVAL;
3077 		ib_spec->mpls.size = sizeof(struct ib_flow_spec_mpls);
3078 		memcpy(&ib_spec->mpls.val, kern_spec_val, actual_filter_sz);
3079 		memcpy(&ib_spec->mpls.mask, kern_spec_mask, actual_filter_sz);
3080 		break;
3081 	default:
3082 		return -EINVAL;
3083 	}
3084 	return 0;
3085 }
3086 
3087 static int kern_spec_to_ib_spec_filter(struct ib_uverbs_flow_spec *kern_spec,
3088 				       union ib_flow_spec *ib_spec)
3089 {
3090 	ssize_t kern_filter_sz;
3091 	void *kern_spec_mask;
3092 	void *kern_spec_val;
3093 
3094 	if (kern_spec->reserved)
3095 		return -EINVAL;
3096 
3097 	kern_filter_sz = kern_spec_filter_sz(&kern_spec->hdr);
3098 
3099 	kern_spec_val = (void *)kern_spec +
3100 		sizeof(struct ib_uverbs_flow_spec_hdr);
3101 	kern_spec_mask = kern_spec_val + kern_filter_sz;
3102 
3103 	return ib_uverbs_kern_spec_to_ib_spec_filter(kern_spec->type,
3104 						     kern_spec_mask,
3105 						     kern_spec_val,
3106 						     kern_filter_sz, ib_spec);
3107 }
3108 
3109 static int kern_spec_to_ib_spec(struct ib_ucontext *ucontext,
3110 				struct ib_uverbs_flow_spec *kern_spec,
3111 				union ib_flow_spec *ib_spec,
3112 				struct ib_uflow_resources *uflow_res)
3113 {
3114 	if (kern_spec->reserved)
3115 		return -EINVAL;
3116 
3117 	if (kern_spec->type >= IB_FLOW_SPEC_ACTION_TAG)
3118 		return kern_spec_to_ib_spec_action(ucontext, kern_spec, ib_spec,
3119 						   uflow_res);
3120 	else
3121 		return kern_spec_to_ib_spec_filter(kern_spec, ib_spec);
3122 }
3123 
3124 int ib_uverbs_ex_create_wq(struct ib_uverbs_file *file,
3125 			   struct ib_device *ib_dev,
3126 			   struct ib_udata *ucore,
3127 			   struct ib_udata *uhw)
3128 {
3129 	struct ib_uverbs_ex_create_wq	  cmd = {};
3130 	struct ib_uverbs_ex_create_wq_resp resp = {};
3131 	struct ib_uwq_object           *obj;
3132 	int err = 0;
3133 	struct ib_cq *cq;
3134 	struct ib_pd *pd;
3135 	struct ib_wq *wq;
3136 	struct ib_wq_init_attr wq_init_attr = {};
3137 	size_t required_cmd_sz;
3138 	size_t required_resp_len;
3139 
3140 	required_cmd_sz = offsetof(typeof(cmd), max_sge) + sizeof(cmd.max_sge);
3141 	required_resp_len = offsetof(typeof(resp), wqn) + sizeof(resp.wqn);
3142 
3143 	if (ucore->inlen < required_cmd_sz)
3144 		return -EINVAL;
3145 
3146 	if (ucore->outlen < required_resp_len)
3147 		return -ENOSPC;
3148 
3149 	if (ucore->inlen > sizeof(cmd) &&
3150 	    !ib_is_udata_cleared(ucore, sizeof(cmd),
3151 				 ucore->inlen - sizeof(cmd)))
3152 		return -EOPNOTSUPP;
3153 
3154 	err = ib_copy_from_udata(&cmd, ucore, min(sizeof(cmd), ucore->inlen));
3155 	if (err)
3156 		return err;
3157 
3158 	if (cmd.comp_mask)
3159 		return -EOPNOTSUPP;
3160 
3161 	obj  = (struct ib_uwq_object *)uobj_alloc(UVERBS_OBJECT_WQ,
3162 						  file->ucontext);
3163 	if (IS_ERR(obj))
3164 		return PTR_ERR(obj);
3165 
3166 	pd  = uobj_get_obj_read(pd, UVERBS_OBJECT_PD, cmd.pd_handle, file->ucontext);
3167 	if (!pd) {
3168 		err = -EINVAL;
3169 		goto err_uobj;
3170 	}
3171 
3172 	cq = uobj_get_obj_read(cq, UVERBS_OBJECT_CQ, cmd.cq_handle, file->ucontext);
3173 	if (!cq) {
3174 		err = -EINVAL;
3175 		goto err_put_pd;
3176 	}
3177 
3178 	wq_init_attr.cq = cq;
3179 	wq_init_attr.max_sge = cmd.max_sge;
3180 	wq_init_attr.max_wr = cmd.max_wr;
3181 	wq_init_attr.wq_context = file;
3182 	wq_init_attr.wq_type = cmd.wq_type;
3183 	wq_init_attr.event_handler = ib_uverbs_wq_event_handler;
3184 	if (ucore->inlen >= (offsetof(typeof(cmd), create_flags) +
3185 			     sizeof(cmd.create_flags)))
3186 		wq_init_attr.create_flags = cmd.create_flags;
3187 	obj->uevent.events_reported = 0;
3188 	INIT_LIST_HEAD(&obj->uevent.event_list);
3189 
3190 	if (!pd->device->create_wq) {
3191 		err = -EOPNOTSUPP;
3192 		goto err_put_cq;
3193 	}
3194 	wq = pd->device->create_wq(pd, &wq_init_attr, uhw);
3195 	if (IS_ERR(wq)) {
3196 		err = PTR_ERR(wq);
3197 		goto err_put_cq;
3198 	}
3199 
3200 	wq->uobject = &obj->uevent.uobject;
3201 	obj->uevent.uobject.object = wq;
3202 	wq->wq_type = wq_init_attr.wq_type;
3203 	wq->cq = cq;
3204 	wq->pd = pd;
3205 	wq->device = pd->device;
3206 	wq->wq_context = wq_init_attr.wq_context;
3207 	atomic_set(&wq->usecnt, 0);
3208 	atomic_inc(&pd->usecnt);
3209 	atomic_inc(&cq->usecnt);
3210 	wq->uobject = &obj->uevent.uobject;
3211 	obj->uevent.uobject.object = wq;
3212 
3213 	memset(&resp, 0, sizeof(resp));
3214 	resp.wq_handle = obj->uevent.uobject.id;
3215 	resp.max_sge = wq_init_attr.max_sge;
3216 	resp.max_wr = wq_init_attr.max_wr;
3217 	resp.wqn = wq->wq_num;
3218 	resp.response_length = required_resp_len;
3219 	err = ib_copy_to_udata(ucore,
3220 			       &resp, resp.response_length);
3221 	if (err)
3222 		goto err_copy;
3223 
3224 	uobj_put_obj_read(pd);
3225 	uobj_put_obj_read(cq);
3226 	uobj_alloc_commit(&obj->uevent.uobject);
3227 	return 0;
3228 
3229 err_copy:
3230 	ib_destroy_wq(wq);
3231 err_put_cq:
3232 	uobj_put_obj_read(cq);
3233 err_put_pd:
3234 	uobj_put_obj_read(pd);
3235 err_uobj:
3236 	uobj_alloc_abort(&obj->uevent.uobject);
3237 
3238 	return err;
3239 }
3240 
3241 int ib_uverbs_ex_destroy_wq(struct ib_uverbs_file *file,
3242 			    struct ib_device *ib_dev,
3243 			    struct ib_udata *ucore,
3244 			    struct ib_udata *uhw)
3245 {
3246 	struct ib_uverbs_ex_destroy_wq	cmd = {};
3247 	struct ib_uverbs_ex_destroy_wq_resp	resp = {};
3248 	struct ib_uobject		*uobj;
3249 	struct ib_uwq_object		*obj;
3250 	size_t required_cmd_sz;
3251 	size_t required_resp_len;
3252 	int				ret;
3253 
3254 	required_cmd_sz = offsetof(typeof(cmd), wq_handle) + sizeof(cmd.wq_handle);
3255 	required_resp_len = offsetof(typeof(resp), reserved) + sizeof(resp.reserved);
3256 
3257 	if (ucore->inlen < required_cmd_sz)
3258 		return -EINVAL;
3259 
3260 	if (ucore->outlen < required_resp_len)
3261 		return -ENOSPC;
3262 
3263 	if (ucore->inlen > sizeof(cmd) &&
3264 	    !ib_is_udata_cleared(ucore, sizeof(cmd),
3265 				 ucore->inlen - sizeof(cmd)))
3266 		return -EOPNOTSUPP;
3267 
3268 	ret = ib_copy_from_udata(&cmd, ucore, min(sizeof(cmd), ucore->inlen));
3269 	if (ret)
3270 		return ret;
3271 
3272 	if (cmd.comp_mask)
3273 		return -EOPNOTSUPP;
3274 
3275 	resp.response_length = required_resp_len;
3276 	uobj  = uobj_get_write(UVERBS_OBJECT_WQ, cmd.wq_handle,
3277 			       file->ucontext);
3278 	if (IS_ERR(uobj))
3279 		return PTR_ERR(uobj);
3280 
3281 	obj = container_of(uobj, struct ib_uwq_object, uevent.uobject);
3282 	/*
3283 	 * Make sure we don't free the memory in remove_commit as we still
3284 	 * needs the uobject memory to create the response.
3285 	 */
3286 	uverbs_uobject_get(uobj);
3287 
3288 	ret = uobj_remove_commit(uobj);
3289 	resp.events_reported = obj->uevent.events_reported;
3290 	uverbs_uobject_put(uobj);
3291 	if (ret)
3292 		return ret;
3293 
3294 	return ib_copy_to_udata(ucore, &resp, resp.response_length);
3295 }
3296 
3297 int ib_uverbs_ex_modify_wq(struct ib_uverbs_file *file,
3298 			   struct ib_device *ib_dev,
3299 			   struct ib_udata *ucore,
3300 			   struct ib_udata *uhw)
3301 {
3302 	struct ib_uverbs_ex_modify_wq cmd = {};
3303 	struct ib_wq *wq;
3304 	struct ib_wq_attr wq_attr = {};
3305 	size_t required_cmd_sz;
3306 	int ret;
3307 
3308 	required_cmd_sz = offsetof(typeof(cmd), curr_wq_state) + sizeof(cmd.curr_wq_state);
3309 	if (ucore->inlen < required_cmd_sz)
3310 		return -EINVAL;
3311 
3312 	if (ucore->inlen > sizeof(cmd) &&
3313 	    !ib_is_udata_cleared(ucore, sizeof(cmd),
3314 				 ucore->inlen - sizeof(cmd)))
3315 		return -EOPNOTSUPP;
3316 
3317 	ret = ib_copy_from_udata(&cmd, ucore, min(sizeof(cmd), ucore->inlen));
3318 	if (ret)
3319 		return ret;
3320 
3321 	if (!cmd.attr_mask)
3322 		return -EINVAL;
3323 
3324 	if (cmd.attr_mask > (IB_WQ_STATE | IB_WQ_CUR_STATE | IB_WQ_FLAGS))
3325 		return -EINVAL;
3326 
3327 	wq = uobj_get_obj_read(wq, UVERBS_OBJECT_WQ, cmd.wq_handle, file->ucontext);
3328 	if (!wq)
3329 		return -EINVAL;
3330 
3331 	wq_attr.curr_wq_state = cmd.curr_wq_state;
3332 	wq_attr.wq_state = cmd.wq_state;
3333 	if (cmd.attr_mask & IB_WQ_FLAGS) {
3334 		wq_attr.flags = cmd.flags;
3335 		wq_attr.flags_mask = cmd.flags_mask;
3336 	}
3337 	if (!wq->device->modify_wq) {
3338 		ret = -EOPNOTSUPP;
3339 		goto out;
3340 	}
3341 	ret = wq->device->modify_wq(wq, &wq_attr, cmd.attr_mask, uhw);
3342 out:
3343 	uobj_put_obj_read(wq);
3344 	return ret;
3345 }
3346 
3347 int ib_uverbs_ex_create_rwq_ind_table(struct ib_uverbs_file *file,
3348 				      struct ib_device *ib_dev,
3349 				      struct ib_udata *ucore,
3350 				      struct ib_udata *uhw)
3351 {
3352 	struct ib_uverbs_ex_create_rwq_ind_table	  cmd = {};
3353 	struct ib_uverbs_ex_create_rwq_ind_table_resp  resp = {};
3354 	struct ib_uobject		  *uobj;
3355 	int err = 0;
3356 	struct ib_rwq_ind_table_init_attr init_attr = {};
3357 	struct ib_rwq_ind_table *rwq_ind_tbl;
3358 	struct ib_wq	**wqs = NULL;
3359 	u32 *wqs_handles = NULL;
3360 	struct ib_wq	*wq = NULL;
3361 	int i, j, num_read_wqs;
3362 	u32 num_wq_handles;
3363 	u32 expected_in_size;
3364 	size_t required_cmd_sz_header;
3365 	size_t required_resp_len;
3366 
3367 	required_cmd_sz_header = offsetof(typeof(cmd), log_ind_tbl_size) + sizeof(cmd.log_ind_tbl_size);
3368 	required_resp_len = offsetof(typeof(resp), ind_tbl_num) + sizeof(resp.ind_tbl_num);
3369 
3370 	if (ucore->inlen < required_cmd_sz_header)
3371 		return -EINVAL;
3372 
3373 	if (ucore->outlen < required_resp_len)
3374 		return -ENOSPC;
3375 
3376 	err = ib_copy_from_udata(&cmd, ucore, required_cmd_sz_header);
3377 	if (err)
3378 		return err;
3379 
3380 	ucore->inbuf += required_cmd_sz_header;
3381 	ucore->inlen -= required_cmd_sz_header;
3382 
3383 	if (cmd.comp_mask)
3384 		return -EOPNOTSUPP;
3385 
3386 	if (cmd.log_ind_tbl_size > IB_USER_VERBS_MAX_LOG_IND_TBL_SIZE)
3387 		return -EINVAL;
3388 
3389 	num_wq_handles = 1 << cmd.log_ind_tbl_size;
3390 	expected_in_size = num_wq_handles * sizeof(__u32);
3391 	if (num_wq_handles == 1)
3392 		/* input size for wq handles is u64 aligned */
3393 		expected_in_size += sizeof(__u32);
3394 
3395 	if (ucore->inlen < expected_in_size)
3396 		return -EINVAL;
3397 
3398 	if (ucore->inlen > expected_in_size &&
3399 	    !ib_is_udata_cleared(ucore, expected_in_size,
3400 				 ucore->inlen - expected_in_size))
3401 		return -EOPNOTSUPP;
3402 
3403 	wqs_handles = kcalloc(num_wq_handles, sizeof(*wqs_handles),
3404 			      GFP_KERNEL);
3405 	if (!wqs_handles)
3406 		return -ENOMEM;
3407 
3408 	err = ib_copy_from_udata(wqs_handles, ucore,
3409 				 num_wq_handles * sizeof(__u32));
3410 	if (err)
3411 		goto err_free;
3412 
3413 	wqs = kcalloc(num_wq_handles, sizeof(*wqs), GFP_KERNEL);
3414 	if (!wqs) {
3415 		err = -ENOMEM;
3416 		goto  err_free;
3417 	}
3418 
3419 	for (num_read_wqs = 0; num_read_wqs < num_wq_handles;
3420 			num_read_wqs++) {
3421 		wq = uobj_get_obj_read(wq, UVERBS_OBJECT_WQ, wqs_handles[num_read_wqs],
3422 				       file->ucontext);
3423 		if (!wq) {
3424 			err = -EINVAL;
3425 			goto put_wqs;
3426 		}
3427 
3428 		wqs[num_read_wqs] = wq;
3429 	}
3430 
3431 	uobj  = uobj_alloc(UVERBS_OBJECT_RWQ_IND_TBL, file->ucontext);
3432 	if (IS_ERR(uobj)) {
3433 		err = PTR_ERR(uobj);
3434 		goto put_wqs;
3435 	}
3436 
3437 	init_attr.log_ind_tbl_size = cmd.log_ind_tbl_size;
3438 	init_attr.ind_tbl = wqs;
3439 
3440 	if (!ib_dev->create_rwq_ind_table) {
3441 		err = -EOPNOTSUPP;
3442 		goto err_uobj;
3443 	}
3444 	rwq_ind_tbl = ib_dev->create_rwq_ind_table(ib_dev, &init_attr, uhw);
3445 
3446 	if (IS_ERR(rwq_ind_tbl)) {
3447 		err = PTR_ERR(rwq_ind_tbl);
3448 		goto err_uobj;
3449 	}
3450 
3451 	rwq_ind_tbl->ind_tbl = wqs;
3452 	rwq_ind_tbl->log_ind_tbl_size = init_attr.log_ind_tbl_size;
3453 	rwq_ind_tbl->uobject = uobj;
3454 	uobj->object = rwq_ind_tbl;
3455 	rwq_ind_tbl->device = ib_dev;
3456 	atomic_set(&rwq_ind_tbl->usecnt, 0);
3457 
3458 	for (i = 0; i < num_wq_handles; i++)
3459 		atomic_inc(&wqs[i]->usecnt);
3460 
3461 	resp.ind_tbl_handle = uobj->id;
3462 	resp.ind_tbl_num = rwq_ind_tbl->ind_tbl_num;
3463 	resp.response_length = required_resp_len;
3464 
3465 	err = ib_copy_to_udata(ucore,
3466 			       &resp, resp.response_length);
3467 	if (err)
3468 		goto err_copy;
3469 
3470 	kfree(wqs_handles);
3471 
3472 	for (j = 0; j < num_read_wqs; j++)
3473 		uobj_put_obj_read(wqs[j]);
3474 
3475 	uobj_alloc_commit(uobj);
3476 	return 0;
3477 
3478 err_copy:
3479 	ib_destroy_rwq_ind_table(rwq_ind_tbl);
3480 err_uobj:
3481 	uobj_alloc_abort(uobj);
3482 put_wqs:
3483 	for (j = 0; j < num_read_wqs; j++)
3484 		uobj_put_obj_read(wqs[j]);
3485 err_free:
3486 	kfree(wqs_handles);
3487 	kfree(wqs);
3488 	return err;
3489 }
3490 
3491 int ib_uverbs_ex_destroy_rwq_ind_table(struct ib_uverbs_file *file,
3492 				       struct ib_device *ib_dev,
3493 				       struct ib_udata *ucore,
3494 				       struct ib_udata *uhw)
3495 {
3496 	struct ib_uverbs_ex_destroy_rwq_ind_table	cmd = {};
3497 	struct ib_uobject		*uobj;
3498 	int			ret;
3499 	size_t required_cmd_sz;
3500 
3501 	required_cmd_sz = offsetof(typeof(cmd), ind_tbl_handle) + sizeof(cmd.ind_tbl_handle);
3502 
3503 	if (ucore->inlen < required_cmd_sz)
3504 		return -EINVAL;
3505 
3506 	if (ucore->inlen > sizeof(cmd) &&
3507 	    !ib_is_udata_cleared(ucore, sizeof(cmd),
3508 				 ucore->inlen - sizeof(cmd)))
3509 		return -EOPNOTSUPP;
3510 
3511 	ret = ib_copy_from_udata(&cmd, ucore, min(sizeof(cmd), ucore->inlen));
3512 	if (ret)
3513 		return ret;
3514 
3515 	if (cmd.comp_mask)
3516 		return -EOPNOTSUPP;
3517 
3518 	uobj  = uobj_get_write(UVERBS_OBJECT_RWQ_IND_TBL, cmd.ind_tbl_handle,
3519 			       file->ucontext);
3520 	if (IS_ERR(uobj))
3521 		return PTR_ERR(uobj);
3522 
3523 	return uobj_remove_commit(uobj);
3524 }
3525 
3526 int ib_uverbs_ex_create_flow(struct ib_uverbs_file *file,
3527 			     struct ib_device *ib_dev,
3528 			     struct ib_udata *ucore,
3529 			     struct ib_udata *uhw)
3530 {
3531 	struct ib_uverbs_create_flow	  cmd;
3532 	struct ib_uverbs_create_flow_resp resp;
3533 	struct ib_uobject		  *uobj;
3534 	struct ib_uflow_object		  *uflow;
3535 	struct ib_flow			  *flow_id;
3536 	struct ib_uverbs_flow_attr	  *kern_flow_attr;
3537 	struct ib_flow_attr		  *flow_attr;
3538 	struct ib_qp			  *qp;
3539 	struct ib_uflow_resources	  *uflow_res;
3540 	struct ib_uverbs_flow_spec_hdr	  *kern_spec;
3541 	int err = 0;
3542 	void *ib_spec;
3543 	int i;
3544 
3545 	if (ucore->inlen < sizeof(cmd))
3546 		return -EINVAL;
3547 
3548 	if (ucore->outlen < sizeof(resp))
3549 		return -ENOSPC;
3550 
3551 	err = ib_copy_from_udata(&cmd, ucore, sizeof(cmd));
3552 	if (err)
3553 		return err;
3554 
3555 	ucore->inbuf += sizeof(cmd);
3556 	ucore->inlen -= sizeof(cmd);
3557 
3558 	if (cmd.comp_mask)
3559 		return -EINVAL;
3560 
3561 	if (!capable(CAP_NET_RAW))
3562 		return -EPERM;
3563 
3564 	if (cmd.flow_attr.flags >= IB_FLOW_ATTR_FLAGS_RESERVED)
3565 		return -EINVAL;
3566 
3567 	if ((cmd.flow_attr.flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP) &&
3568 	    ((cmd.flow_attr.type == IB_FLOW_ATTR_ALL_DEFAULT) ||
3569 	     (cmd.flow_attr.type == IB_FLOW_ATTR_MC_DEFAULT)))
3570 		return -EINVAL;
3571 
3572 	if (cmd.flow_attr.num_of_specs > IB_FLOW_SPEC_SUPPORT_LAYERS)
3573 		return -EINVAL;
3574 
3575 	if (cmd.flow_attr.size > ucore->inlen ||
3576 	    cmd.flow_attr.size >
3577 	    (cmd.flow_attr.num_of_specs * sizeof(struct ib_uverbs_flow_spec)))
3578 		return -EINVAL;
3579 
3580 	if (cmd.flow_attr.reserved[0] ||
3581 	    cmd.flow_attr.reserved[1])
3582 		return -EINVAL;
3583 
3584 	if (cmd.flow_attr.num_of_specs) {
3585 		kern_flow_attr = kmalloc(sizeof(*kern_flow_attr) + cmd.flow_attr.size,
3586 					 GFP_KERNEL);
3587 		if (!kern_flow_attr)
3588 			return -ENOMEM;
3589 
3590 		*kern_flow_attr = cmd.flow_attr;
3591 		err = ib_copy_from_udata(&kern_flow_attr->flow_specs, ucore,
3592 					 cmd.flow_attr.size);
3593 		if (err)
3594 			goto err_free_attr;
3595 	} else {
3596 		kern_flow_attr = &cmd.flow_attr;
3597 	}
3598 
3599 	uobj  = uobj_alloc(UVERBS_OBJECT_FLOW, file->ucontext);
3600 	if (IS_ERR(uobj)) {
3601 		err = PTR_ERR(uobj);
3602 		goto err_free_attr;
3603 	}
3604 
3605 	qp = uobj_get_obj_read(qp, UVERBS_OBJECT_QP, cmd.qp_handle, file->ucontext);
3606 	if (!qp) {
3607 		err = -EINVAL;
3608 		goto err_uobj;
3609 	}
3610 
3611 	if (qp->qp_type != IB_QPT_UD && qp->qp_type != IB_QPT_RAW_PACKET) {
3612 		err = -EINVAL;
3613 		goto err_put;
3614 	}
3615 
3616 	flow_attr = kzalloc(struct_size(flow_attr, flows,
3617 				cmd.flow_attr.num_of_specs), GFP_KERNEL);
3618 	if (!flow_attr) {
3619 		err = -ENOMEM;
3620 		goto err_put;
3621 	}
3622 	uflow_res = flow_resources_alloc(cmd.flow_attr.num_of_specs);
3623 	if (!uflow_res) {
3624 		err = -ENOMEM;
3625 		goto err_free_flow_attr;
3626 	}
3627 
3628 	flow_attr->type = kern_flow_attr->type;
3629 	flow_attr->priority = kern_flow_attr->priority;
3630 	flow_attr->num_of_specs = kern_flow_attr->num_of_specs;
3631 	flow_attr->port = kern_flow_attr->port;
3632 	flow_attr->flags = kern_flow_attr->flags;
3633 	flow_attr->size = sizeof(*flow_attr);
3634 
3635 	kern_spec = kern_flow_attr->flow_specs;
3636 	ib_spec = flow_attr + 1;
3637 	for (i = 0; i < flow_attr->num_of_specs &&
3638 			cmd.flow_attr.size >= sizeof(*kern_spec) &&
3639 			cmd.flow_attr.size >= kern_spec->size;
3640 	     i++) {
3641 		err = kern_spec_to_ib_spec(
3642 				file->ucontext, (struct ib_uverbs_flow_spec *)kern_spec,
3643 				ib_spec, uflow_res);
3644 		if (err)
3645 			goto err_free;
3646 
3647 		flow_attr->size +=
3648 			((union ib_flow_spec *) ib_spec)->size;
3649 		cmd.flow_attr.size -= kern_spec->size;
3650 		kern_spec = ((void *)kern_spec) + kern_spec->size;
3651 		ib_spec += ((union ib_flow_spec *) ib_spec)->size;
3652 	}
3653 	if (cmd.flow_attr.size || (i != flow_attr->num_of_specs)) {
3654 		pr_warn("create flow failed, flow %d: %d bytes left from uverb cmd\n",
3655 			i, cmd.flow_attr.size);
3656 		err = -EINVAL;
3657 		goto err_free;
3658 	}
3659 
3660 	flow_id = qp->device->create_flow(qp, flow_attr,
3661 					  IB_FLOW_DOMAIN_USER, uhw);
3662 
3663 	if (IS_ERR(flow_id)) {
3664 		err = PTR_ERR(flow_id);
3665 		goto err_free;
3666 	}
3667 	atomic_inc(&qp->usecnt);
3668 	flow_id->qp = qp;
3669 	flow_id->uobject = uobj;
3670 	uobj->object = flow_id;
3671 	uflow = container_of(uobj, typeof(*uflow), uobject);
3672 	uflow->resources = uflow_res;
3673 
3674 	memset(&resp, 0, sizeof(resp));
3675 	resp.flow_handle = uobj->id;
3676 
3677 	err = ib_copy_to_udata(ucore,
3678 			       &resp, sizeof(resp));
3679 	if (err)
3680 		goto err_copy;
3681 
3682 	uobj_put_obj_read(qp);
3683 	uobj_alloc_commit(uobj);
3684 	kfree(flow_attr);
3685 	if (cmd.flow_attr.num_of_specs)
3686 		kfree(kern_flow_attr);
3687 	return 0;
3688 err_copy:
3689 	ib_destroy_flow(flow_id);
3690 err_free:
3691 	ib_uverbs_flow_resources_free(uflow_res);
3692 err_free_flow_attr:
3693 	kfree(flow_attr);
3694 err_put:
3695 	uobj_put_obj_read(qp);
3696 err_uobj:
3697 	uobj_alloc_abort(uobj);
3698 err_free_attr:
3699 	if (cmd.flow_attr.num_of_specs)
3700 		kfree(kern_flow_attr);
3701 	return err;
3702 }
3703 
3704 int ib_uverbs_ex_destroy_flow(struct ib_uverbs_file *file,
3705 			      struct ib_device *ib_dev,
3706 			      struct ib_udata *ucore,
3707 			      struct ib_udata *uhw)
3708 {
3709 	struct ib_uverbs_destroy_flow	cmd;
3710 	struct ib_uobject		*uobj;
3711 	int				ret;
3712 
3713 	if (ucore->inlen < sizeof(cmd))
3714 		return -EINVAL;
3715 
3716 	ret = ib_copy_from_udata(&cmd, ucore, sizeof(cmd));
3717 	if (ret)
3718 		return ret;
3719 
3720 	if (cmd.comp_mask)
3721 		return -EINVAL;
3722 
3723 	uobj  = uobj_get_write(UVERBS_OBJECT_FLOW, cmd.flow_handle,
3724 			       file->ucontext);
3725 	if (IS_ERR(uobj))
3726 		return PTR_ERR(uobj);
3727 
3728 	ret = uobj_remove_commit(uobj);
3729 	return ret;
3730 }
3731 
3732 static int __uverbs_create_xsrq(struct ib_uverbs_file *file,
3733 				struct ib_device *ib_dev,
3734 				struct ib_uverbs_create_xsrq *cmd,
3735 				struct ib_udata *udata)
3736 {
3737 	struct ib_uverbs_create_srq_resp resp;
3738 	struct ib_usrq_object           *obj;
3739 	struct ib_pd                    *pd;
3740 	struct ib_srq                   *srq;
3741 	struct ib_uobject               *uninitialized_var(xrcd_uobj);
3742 	struct ib_srq_init_attr          attr;
3743 	int ret;
3744 
3745 	obj  = (struct ib_usrq_object *)uobj_alloc(UVERBS_OBJECT_SRQ,
3746 						   file->ucontext);
3747 	if (IS_ERR(obj))
3748 		return PTR_ERR(obj);
3749 
3750 	if (cmd->srq_type == IB_SRQT_TM)
3751 		attr.ext.tag_matching.max_num_tags = cmd->max_num_tags;
3752 
3753 	if (cmd->srq_type == IB_SRQT_XRC) {
3754 		xrcd_uobj = uobj_get_read(UVERBS_OBJECT_XRCD, cmd->xrcd_handle,
3755 					  file->ucontext);
3756 		if (IS_ERR(xrcd_uobj)) {
3757 			ret = -EINVAL;
3758 			goto err;
3759 		}
3760 
3761 		attr.ext.xrc.xrcd = (struct ib_xrcd *)xrcd_uobj->object;
3762 		if (!attr.ext.xrc.xrcd) {
3763 			ret = -EINVAL;
3764 			goto err_put_xrcd;
3765 		}
3766 
3767 		obj->uxrcd = container_of(xrcd_uobj, struct ib_uxrcd_object, uobject);
3768 		atomic_inc(&obj->uxrcd->refcnt);
3769 	}
3770 
3771 	if (ib_srq_has_cq(cmd->srq_type)) {
3772 		attr.ext.cq  = uobj_get_obj_read(cq, UVERBS_OBJECT_CQ, cmd->cq_handle,
3773 						 file->ucontext);
3774 		if (!attr.ext.cq) {
3775 			ret = -EINVAL;
3776 			goto err_put_xrcd;
3777 		}
3778 	}
3779 
3780 	pd  = uobj_get_obj_read(pd, UVERBS_OBJECT_PD, cmd->pd_handle, file->ucontext);
3781 	if (!pd) {
3782 		ret = -EINVAL;
3783 		goto err_put_cq;
3784 	}
3785 
3786 	attr.event_handler  = ib_uverbs_srq_event_handler;
3787 	attr.srq_context    = file;
3788 	attr.srq_type       = cmd->srq_type;
3789 	attr.attr.max_wr    = cmd->max_wr;
3790 	attr.attr.max_sge   = cmd->max_sge;
3791 	attr.attr.srq_limit = cmd->srq_limit;
3792 
3793 	obj->uevent.events_reported = 0;
3794 	INIT_LIST_HEAD(&obj->uevent.event_list);
3795 
3796 	srq = pd->device->create_srq(pd, &attr, udata);
3797 	if (IS_ERR(srq)) {
3798 		ret = PTR_ERR(srq);
3799 		goto err_put;
3800 	}
3801 
3802 	srq->device        = pd->device;
3803 	srq->pd            = pd;
3804 	srq->srq_type	   = cmd->srq_type;
3805 	srq->uobject       = &obj->uevent.uobject;
3806 	srq->event_handler = attr.event_handler;
3807 	srq->srq_context   = attr.srq_context;
3808 
3809 	if (ib_srq_has_cq(cmd->srq_type)) {
3810 		srq->ext.cq       = attr.ext.cq;
3811 		atomic_inc(&attr.ext.cq->usecnt);
3812 	}
3813 
3814 	if (cmd->srq_type == IB_SRQT_XRC) {
3815 		srq->ext.xrc.xrcd = attr.ext.xrc.xrcd;
3816 		atomic_inc(&attr.ext.xrc.xrcd->usecnt);
3817 	}
3818 
3819 	atomic_inc(&pd->usecnt);
3820 	atomic_set(&srq->usecnt, 0);
3821 
3822 	obj->uevent.uobject.object = srq;
3823 	obj->uevent.uobject.user_handle = cmd->user_handle;
3824 
3825 	memset(&resp, 0, sizeof resp);
3826 	resp.srq_handle = obj->uevent.uobject.id;
3827 	resp.max_wr     = attr.attr.max_wr;
3828 	resp.max_sge    = attr.attr.max_sge;
3829 	if (cmd->srq_type == IB_SRQT_XRC)
3830 		resp.srqn = srq->ext.xrc.srq_num;
3831 
3832 	if (copy_to_user(u64_to_user_ptr(cmd->response),
3833 			 &resp, sizeof resp)) {
3834 		ret = -EFAULT;
3835 		goto err_copy;
3836 	}
3837 
3838 	if (cmd->srq_type == IB_SRQT_XRC)
3839 		uobj_put_read(xrcd_uobj);
3840 
3841 	if (ib_srq_has_cq(cmd->srq_type))
3842 		uobj_put_obj_read(attr.ext.cq);
3843 
3844 	uobj_put_obj_read(pd);
3845 	uobj_alloc_commit(&obj->uevent.uobject);
3846 
3847 	return 0;
3848 
3849 err_copy:
3850 	ib_destroy_srq(srq);
3851 
3852 err_put:
3853 	uobj_put_obj_read(pd);
3854 
3855 err_put_cq:
3856 	if (ib_srq_has_cq(cmd->srq_type))
3857 		uobj_put_obj_read(attr.ext.cq);
3858 
3859 err_put_xrcd:
3860 	if (cmd->srq_type == IB_SRQT_XRC) {
3861 		atomic_dec(&obj->uxrcd->refcnt);
3862 		uobj_put_read(xrcd_uobj);
3863 	}
3864 
3865 err:
3866 	uobj_alloc_abort(&obj->uevent.uobject);
3867 	return ret;
3868 }
3869 
3870 ssize_t ib_uverbs_create_srq(struct ib_uverbs_file *file,
3871 			     struct ib_device *ib_dev,
3872 			     const char __user *buf, int in_len,
3873 			     int out_len)
3874 {
3875 	struct ib_uverbs_create_srq      cmd;
3876 	struct ib_uverbs_create_xsrq     xcmd;
3877 	struct ib_uverbs_create_srq_resp resp;
3878 	struct ib_udata                  udata;
3879 	int ret;
3880 
3881 	if (out_len < sizeof resp)
3882 		return -ENOSPC;
3883 
3884 	if (copy_from_user(&cmd, buf, sizeof cmd))
3885 		return -EFAULT;
3886 
3887 	memset(&xcmd, 0, sizeof(xcmd));
3888 	xcmd.response	 = cmd.response;
3889 	xcmd.user_handle = cmd.user_handle;
3890 	xcmd.srq_type	 = IB_SRQT_BASIC;
3891 	xcmd.pd_handle	 = cmd.pd_handle;
3892 	xcmd.max_wr	 = cmd.max_wr;
3893 	xcmd.max_sge	 = cmd.max_sge;
3894 	xcmd.srq_limit	 = cmd.srq_limit;
3895 
3896 	ib_uverbs_init_udata(&udata, buf + sizeof(cmd),
3897 		   u64_to_user_ptr(cmd.response) + sizeof(resp),
3898 		   in_len - sizeof(cmd) - sizeof(struct ib_uverbs_cmd_hdr),
3899 		   out_len - sizeof(resp));
3900 
3901 	ret = __uverbs_create_xsrq(file, ib_dev, &xcmd, &udata);
3902 	if (ret)
3903 		return ret;
3904 
3905 	return in_len;
3906 }
3907 
3908 ssize_t ib_uverbs_create_xsrq(struct ib_uverbs_file *file,
3909 			      struct ib_device *ib_dev,
3910 			      const char __user *buf, int in_len, int out_len)
3911 {
3912 	struct ib_uverbs_create_xsrq     cmd;
3913 	struct ib_uverbs_create_srq_resp resp;
3914 	struct ib_udata                  udata;
3915 	int ret;
3916 
3917 	if (out_len < sizeof resp)
3918 		return -ENOSPC;
3919 
3920 	if (copy_from_user(&cmd, buf, sizeof cmd))
3921 		return -EFAULT;
3922 
3923 	ib_uverbs_init_udata(&udata, buf + sizeof(cmd),
3924 		   u64_to_user_ptr(cmd.response) + sizeof(resp),
3925 		   in_len - sizeof(cmd) - sizeof(struct ib_uverbs_cmd_hdr),
3926 		   out_len - sizeof(resp));
3927 
3928 	ret = __uverbs_create_xsrq(file, ib_dev, &cmd, &udata);
3929 	if (ret)
3930 		return ret;
3931 
3932 	return in_len;
3933 }
3934 
3935 ssize_t ib_uverbs_modify_srq(struct ib_uverbs_file *file,
3936 			     struct ib_device *ib_dev,
3937 			     const char __user *buf, int in_len,
3938 			     int out_len)
3939 {
3940 	struct ib_uverbs_modify_srq cmd;
3941 	struct ib_udata             udata;
3942 	struct ib_srq              *srq;
3943 	struct ib_srq_attr          attr;
3944 	int                         ret;
3945 
3946 	if (copy_from_user(&cmd, buf, sizeof cmd))
3947 		return -EFAULT;
3948 
3949 	ib_uverbs_init_udata(&udata, buf + sizeof cmd, NULL, in_len - sizeof cmd,
3950 		   out_len);
3951 
3952 	srq = uobj_get_obj_read(srq, UVERBS_OBJECT_SRQ, cmd.srq_handle, file->ucontext);
3953 	if (!srq)
3954 		return -EINVAL;
3955 
3956 	attr.max_wr    = cmd.max_wr;
3957 	attr.srq_limit = cmd.srq_limit;
3958 
3959 	ret = srq->device->modify_srq(srq, &attr, cmd.attr_mask, &udata);
3960 
3961 	uobj_put_obj_read(srq);
3962 
3963 	return ret ? ret : in_len;
3964 }
3965 
3966 ssize_t ib_uverbs_query_srq(struct ib_uverbs_file *file,
3967 			    struct ib_device *ib_dev,
3968 			    const char __user *buf,
3969 			    int in_len, int out_len)
3970 {
3971 	struct ib_uverbs_query_srq      cmd;
3972 	struct ib_uverbs_query_srq_resp resp;
3973 	struct ib_srq_attr              attr;
3974 	struct ib_srq                   *srq;
3975 	int                             ret;
3976 
3977 	if (out_len < sizeof resp)
3978 		return -ENOSPC;
3979 
3980 	if (copy_from_user(&cmd, buf, sizeof cmd))
3981 		return -EFAULT;
3982 
3983 	srq = uobj_get_obj_read(srq, UVERBS_OBJECT_SRQ, cmd.srq_handle, file->ucontext);
3984 	if (!srq)
3985 		return -EINVAL;
3986 
3987 	ret = ib_query_srq(srq, &attr);
3988 
3989 	uobj_put_obj_read(srq);
3990 
3991 	if (ret)
3992 		return ret;
3993 
3994 	memset(&resp, 0, sizeof resp);
3995 
3996 	resp.max_wr    = attr.max_wr;
3997 	resp.max_sge   = attr.max_sge;
3998 	resp.srq_limit = attr.srq_limit;
3999 
4000 	if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof resp))
4001 		return -EFAULT;
4002 
4003 	return in_len;
4004 }
4005 
4006 ssize_t ib_uverbs_destroy_srq(struct ib_uverbs_file *file,
4007 			      struct ib_device *ib_dev,
4008 			      const char __user *buf, int in_len,
4009 			      int out_len)
4010 {
4011 	struct ib_uverbs_destroy_srq      cmd;
4012 	struct ib_uverbs_destroy_srq_resp resp;
4013 	struct ib_uobject		 *uobj;
4014 	struct ib_uevent_object        	 *obj;
4015 	int                         	  ret = -EINVAL;
4016 
4017 	if (copy_from_user(&cmd, buf, sizeof cmd))
4018 		return -EFAULT;
4019 
4020 	uobj  = uobj_get_write(UVERBS_OBJECT_SRQ, cmd.srq_handle,
4021 			       file->ucontext);
4022 	if (IS_ERR(uobj))
4023 		return PTR_ERR(uobj);
4024 
4025 	obj = container_of(uobj, struct ib_uevent_object, uobject);
4026 	/*
4027 	 * Make sure we don't free the memory in remove_commit as we still
4028 	 * needs the uobject memory to create the response.
4029 	 */
4030 	uverbs_uobject_get(uobj);
4031 
4032 	memset(&resp, 0, sizeof(resp));
4033 
4034 	ret = uobj_remove_commit(uobj);
4035 	if (ret) {
4036 		uverbs_uobject_put(uobj);
4037 		return ret;
4038 	}
4039 	resp.events_reported = obj->events_reported;
4040 	uverbs_uobject_put(uobj);
4041 	if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof(resp)))
4042 		return -EFAULT;
4043 
4044 	return in_len;
4045 }
4046 
4047 int ib_uverbs_ex_query_device(struct ib_uverbs_file *file,
4048 			      struct ib_device *ib_dev,
4049 			      struct ib_udata *ucore,
4050 			      struct ib_udata *uhw)
4051 {
4052 	struct ib_uverbs_ex_query_device_resp resp = { {0} };
4053 	struct ib_uverbs_ex_query_device  cmd;
4054 	struct ib_device_attr attr = {0};
4055 	int err;
4056 
4057 	if (!ib_dev->query_device)
4058 		return -EOPNOTSUPP;
4059 
4060 	if (ucore->inlen < sizeof(cmd))
4061 		return -EINVAL;
4062 
4063 	err = ib_copy_from_udata(&cmd, ucore, sizeof(cmd));
4064 	if (err)
4065 		return err;
4066 
4067 	if (cmd.comp_mask)
4068 		return -EINVAL;
4069 
4070 	if (cmd.reserved)
4071 		return -EINVAL;
4072 
4073 	resp.response_length = offsetof(typeof(resp), odp_caps);
4074 
4075 	if (ucore->outlen < resp.response_length)
4076 		return -ENOSPC;
4077 
4078 	err = ib_dev->query_device(ib_dev, &attr, uhw);
4079 	if (err)
4080 		return err;
4081 
4082 	copy_query_dev_fields(file, ib_dev, &resp.base, &attr);
4083 
4084 	if (ucore->outlen < resp.response_length + sizeof(resp.odp_caps))
4085 		goto end;
4086 
4087 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
4088 	resp.odp_caps.general_caps = attr.odp_caps.general_caps;
4089 	resp.odp_caps.per_transport_caps.rc_odp_caps =
4090 		attr.odp_caps.per_transport_caps.rc_odp_caps;
4091 	resp.odp_caps.per_transport_caps.uc_odp_caps =
4092 		attr.odp_caps.per_transport_caps.uc_odp_caps;
4093 	resp.odp_caps.per_transport_caps.ud_odp_caps =
4094 		attr.odp_caps.per_transport_caps.ud_odp_caps;
4095 #endif
4096 	resp.response_length += sizeof(resp.odp_caps);
4097 
4098 	if (ucore->outlen < resp.response_length + sizeof(resp.timestamp_mask))
4099 		goto end;
4100 
4101 	resp.timestamp_mask = attr.timestamp_mask;
4102 	resp.response_length += sizeof(resp.timestamp_mask);
4103 
4104 	if (ucore->outlen < resp.response_length + sizeof(resp.hca_core_clock))
4105 		goto end;
4106 
4107 	resp.hca_core_clock = attr.hca_core_clock;
4108 	resp.response_length += sizeof(resp.hca_core_clock);
4109 
4110 	if (ucore->outlen < resp.response_length + sizeof(resp.device_cap_flags_ex))
4111 		goto end;
4112 
4113 	resp.device_cap_flags_ex = attr.device_cap_flags;
4114 	resp.response_length += sizeof(resp.device_cap_flags_ex);
4115 
4116 	if (ucore->outlen < resp.response_length + sizeof(resp.rss_caps))
4117 		goto end;
4118 
4119 	resp.rss_caps.supported_qpts = attr.rss_caps.supported_qpts;
4120 	resp.rss_caps.max_rwq_indirection_tables =
4121 		attr.rss_caps.max_rwq_indirection_tables;
4122 	resp.rss_caps.max_rwq_indirection_table_size =
4123 		attr.rss_caps.max_rwq_indirection_table_size;
4124 
4125 	resp.response_length += sizeof(resp.rss_caps);
4126 
4127 	if (ucore->outlen < resp.response_length + sizeof(resp.max_wq_type_rq))
4128 		goto end;
4129 
4130 	resp.max_wq_type_rq = attr.max_wq_type_rq;
4131 	resp.response_length += sizeof(resp.max_wq_type_rq);
4132 
4133 	if (ucore->outlen < resp.response_length + sizeof(resp.raw_packet_caps))
4134 		goto end;
4135 
4136 	resp.raw_packet_caps = attr.raw_packet_caps;
4137 	resp.response_length += sizeof(resp.raw_packet_caps);
4138 
4139 	if (ucore->outlen < resp.response_length + sizeof(resp.tm_caps))
4140 		goto end;
4141 
4142 	resp.tm_caps.max_rndv_hdr_size	= attr.tm_caps.max_rndv_hdr_size;
4143 	resp.tm_caps.max_num_tags	= attr.tm_caps.max_num_tags;
4144 	resp.tm_caps.max_ops		= attr.tm_caps.max_ops;
4145 	resp.tm_caps.max_sge		= attr.tm_caps.max_sge;
4146 	resp.tm_caps.flags		= attr.tm_caps.flags;
4147 	resp.response_length += sizeof(resp.tm_caps);
4148 
4149 	if (ucore->outlen < resp.response_length + sizeof(resp.cq_moderation_caps))
4150 		goto end;
4151 
4152 	resp.cq_moderation_caps.max_cq_moderation_count  =
4153 		attr.cq_caps.max_cq_moderation_count;
4154 	resp.cq_moderation_caps.max_cq_moderation_period =
4155 		attr.cq_caps.max_cq_moderation_period;
4156 	resp.response_length += sizeof(resp.cq_moderation_caps);
4157 
4158 	if (ucore->outlen < resp.response_length + sizeof(resp.max_dm_size))
4159 		goto end;
4160 
4161 	resp.max_dm_size = attr.max_dm_size;
4162 	resp.response_length += sizeof(resp.max_dm_size);
4163 end:
4164 	err = ib_copy_to_udata(ucore, &resp, resp.response_length);
4165 	return err;
4166 }
4167 
4168 int ib_uverbs_ex_modify_cq(struct ib_uverbs_file *file,
4169 			   struct ib_device *ib_dev,
4170 			   struct ib_udata *ucore,
4171 			   struct ib_udata *uhw)
4172 {
4173 	struct ib_uverbs_ex_modify_cq cmd = {};
4174 	struct ib_cq *cq;
4175 	size_t required_cmd_sz;
4176 	int ret;
4177 
4178 	required_cmd_sz = offsetof(typeof(cmd), reserved) +
4179 				sizeof(cmd.reserved);
4180 	if (ucore->inlen < required_cmd_sz)
4181 		return -EINVAL;
4182 
4183 	/* sanity checks */
4184 	if (ucore->inlen > sizeof(cmd) &&
4185 	    !ib_is_udata_cleared(ucore, sizeof(cmd),
4186 				 ucore->inlen - sizeof(cmd)))
4187 		return -EOPNOTSUPP;
4188 
4189 	ret = ib_copy_from_udata(&cmd, ucore, min(sizeof(cmd), ucore->inlen));
4190 	if (ret)
4191 		return ret;
4192 
4193 	if (!cmd.attr_mask || cmd.reserved)
4194 		return -EINVAL;
4195 
4196 	if (cmd.attr_mask > IB_CQ_MODERATE)
4197 		return -EOPNOTSUPP;
4198 
4199 	cq = uobj_get_obj_read(cq, UVERBS_OBJECT_CQ, cmd.cq_handle, file->ucontext);
4200 	if (!cq)
4201 		return -EINVAL;
4202 
4203 	ret = rdma_set_cq_moderation(cq, cmd.attr.cq_count, cmd.attr.cq_period);
4204 
4205 	uobj_put_obj_read(cq);
4206 
4207 	return ret;
4208 }
4209