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(uobj_get_type(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(uobj_get_type(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(uobj_get_type(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(uobj_get_type(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(uobj_get_type(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(uobj_get_type(mr), file->ucontext);
667 	if (IS_ERR(uobj))
668 		return PTR_ERR(uobj);
669 
670 	pd = uobj_get_obj_read(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->uobject = uobj;
695 	atomic_inc(&pd->usecnt);
696 
697 	uobj->object = mr;
698 
699 	memset(&resp, 0, sizeof resp);
700 	resp.lkey      = mr->lkey;
701 	resp.rkey      = mr->rkey;
702 	resp.mr_handle = uobj->id;
703 
704 	if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof resp)) {
705 		ret = -EFAULT;
706 		goto err_copy;
707 	}
708 
709 	uobj_put_obj_read(pd);
710 
711 	uobj_alloc_commit(uobj);
712 
713 	return in_len;
714 
715 err_copy:
716 	ib_dereg_mr(mr);
717 
718 err_put:
719 	uobj_put_obj_read(pd);
720 
721 err_free:
722 	uobj_alloc_abort(uobj);
723 	return ret;
724 }
725 
726 ssize_t ib_uverbs_rereg_mr(struct ib_uverbs_file *file,
727 			   struct ib_device *ib_dev,
728 			   const char __user *buf, int in_len,
729 			   int out_len)
730 {
731 	struct ib_uverbs_rereg_mr      cmd;
732 	struct ib_uverbs_rereg_mr_resp resp;
733 	struct ib_udata              udata;
734 	struct ib_pd                *pd = NULL;
735 	struct ib_mr                *mr;
736 	struct ib_pd		    *old_pd;
737 	int                          ret;
738 	struct ib_uobject	    *uobj;
739 
740 	if (out_len < sizeof(resp))
741 		return -ENOSPC;
742 
743 	if (copy_from_user(&cmd, buf, sizeof(cmd)))
744 		return -EFAULT;
745 
746 	ib_uverbs_init_udata(&udata, buf + sizeof(cmd),
747 		   u64_to_user_ptr(cmd.response) + sizeof(resp),
748                    in_len - sizeof(cmd) - sizeof(struct ib_uverbs_cmd_hdr),
749                    out_len - sizeof(resp));
750 
751 	if (cmd.flags & ~IB_MR_REREG_SUPPORTED || !cmd.flags)
752 		return -EINVAL;
753 
754 	if ((cmd.flags & IB_MR_REREG_TRANS) &&
755 	    (!cmd.start || !cmd.hca_va || 0 >= cmd.length ||
756 	     (cmd.start & ~PAGE_MASK) != (cmd.hca_va & ~PAGE_MASK)))
757 			return -EINVAL;
758 
759 	uobj  = uobj_get_write(uobj_get_type(mr), cmd.mr_handle,
760 			       file->ucontext);
761 	if (IS_ERR(uobj))
762 		return PTR_ERR(uobj);
763 
764 	mr = uobj->object;
765 
766 	if (cmd.flags & IB_MR_REREG_ACCESS) {
767 		ret = ib_check_mr_access(cmd.access_flags);
768 		if (ret)
769 			goto put_uobjs;
770 	}
771 
772 	if (cmd.flags & IB_MR_REREG_PD) {
773 		pd = uobj_get_obj_read(pd, cmd.pd_handle, file->ucontext);
774 		if (!pd) {
775 			ret = -EINVAL;
776 			goto put_uobjs;
777 		}
778 	}
779 
780 	old_pd = mr->pd;
781 	ret = mr->device->rereg_user_mr(mr, cmd.flags, cmd.start,
782 					cmd.length, cmd.hca_va,
783 					cmd.access_flags, pd, &udata);
784 	if (!ret) {
785 		if (cmd.flags & IB_MR_REREG_PD) {
786 			atomic_inc(&pd->usecnt);
787 			mr->pd = pd;
788 			atomic_dec(&old_pd->usecnt);
789 		}
790 	} else {
791 		goto put_uobj_pd;
792 	}
793 
794 	memset(&resp, 0, sizeof(resp));
795 	resp.lkey      = mr->lkey;
796 	resp.rkey      = mr->rkey;
797 
798 	if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof(resp)))
799 		ret = -EFAULT;
800 	else
801 		ret = in_len;
802 
803 put_uobj_pd:
804 	if (cmd.flags & IB_MR_REREG_PD)
805 		uobj_put_obj_read(pd);
806 
807 put_uobjs:
808 	uobj_put_write(uobj);
809 
810 	return ret;
811 }
812 
813 ssize_t ib_uverbs_dereg_mr(struct ib_uverbs_file *file,
814 			   struct ib_device *ib_dev,
815 			   const char __user *buf, int in_len,
816 			   int out_len)
817 {
818 	struct ib_uverbs_dereg_mr cmd;
819 	struct ib_uobject	 *uobj;
820 	int                       ret = -EINVAL;
821 
822 	if (copy_from_user(&cmd, buf, sizeof cmd))
823 		return -EFAULT;
824 
825 	uobj  = uobj_get_write(uobj_get_type(mr), cmd.mr_handle,
826 			       file->ucontext);
827 	if (IS_ERR(uobj))
828 		return PTR_ERR(uobj);
829 
830 	ret = uobj_remove_commit(uobj);
831 
832 	return ret ?: in_len;
833 }
834 
835 ssize_t ib_uverbs_alloc_mw(struct ib_uverbs_file *file,
836 			   struct ib_device *ib_dev,
837 			   const char __user *buf, int in_len,
838 			   int out_len)
839 {
840 	struct ib_uverbs_alloc_mw      cmd;
841 	struct ib_uverbs_alloc_mw_resp resp;
842 	struct ib_uobject             *uobj;
843 	struct ib_pd                  *pd;
844 	struct ib_mw                  *mw;
845 	struct ib_udata		       udata;
846 	int                            ret;
847 
848 	if (out_len < sizeof(resp))
849 		return -ENOSPC;
850 
851 	if (copy_from_user(&cmd, buf, sizeof(cmd)))
852 		return -EFAULT;
853 
854 	uobj  = uobj_alloc(uobj_get_type(mw), file->ucontext);
855 	if (IS_ERR(uobj))
856 		return PTR_ERR(uobj);
857 
858 	pd = uobj_get_obj_read(pd, cmd.pd_handle, file->ucontext);
859 	if (!pd) {
860 		ret = -EINVAL;
861 		goto err_free;
862 	}
863 
864 	ib_uverbs_init_udata(&udata, buf + sizeof(cmd),
865 		   u64_to_user_ptr(cmd.response) + sizeof(resp),
866 		   in_len - sizeof(cmd) - sizeof(struct ib_uverbs_cmd_hdr),
867 		   out_len - sizeof(resp));
868 
869 	mw = pd->device->alloc_mw(pd, cmd.mw_type, &udata);
870 	if (IS_ERR(mw)) {
871 		ret = PTR_ERR(mw);
872 		goto err_put;
873 	}
874 
875 	mw->device  = pd->device;
876 	mw->pd      = pd;
877 	mw->uobject = uobj;
878 	atomic_inc(&pd->usecnt);
879 
880 	uobj->object = mw;
881 
882 	memset(&resp, 0, sizeof(resp));
883 	resp.rkey      = mw->rkey;
884 	resp.mw_handle = uobj->id;
885 
886 	if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof(resp))) {
887 		ret = -EFAULT;
888 		goto err_copy;
889 	}
890 
891 	uobj_put_obj_read(pd);
892 	uobj_alloc_commit(uobj);
893 
894 	return in_len;
895 
896 err_copy:
897 	uverbs_dealloc_mw(mw);
898 err_put:
899 	uobj_put_obj_read(pd);
900 err_free:
901 	uobj_alloc_abort(uobj);
902 	return ret;
903 }
904 
905 ssize_t ib_uverbs_dealloc_mw(struct ib_uverbs_file *file,
906 			     struct ib_device *ib_dev,
907 			     const char __user *buf, int in_len,
908 			     int out_len)
909 {
910 	struct ib_uverbs_dealloc_mw cmd;
911 	struct ib_uobject	   *uobj;
912 	int                         ret = -EINVAL;
913 
914 	if (copy_from_user(&cmd, buf, sizeof(cmd)))
915 		return -EFAULT;
916 
917 	uobj  = uobj_get_write(uobj_get_type(mw), cmd.mw_handle,
918 			       file->ucontext);
919 	if (IS_ERR(uobj))
920 		return PTR_ERR(uobj);
921 
922 	ret = uobj_remove_commit(uobj);
923 	return ret ?: in_len;
924 }
925 
926 ssize_t ib_uverbs_create_comp_channel(struct ib_uverbs_file *file,
927 				      struct ib_device *ib_dev,
928 				      const char __user *buf, int in_len,
929 				      int out_len)
930 {
931 	struct ib_uverbs_create_comp_channel	   cmd;
932 	struct ib_uverbs_create_comp_channel_resp  resp;
933 	struct ib_uobject			  *uobj;
934 	struct ib_uverbs_completion_event_file	  *ev_file;
935 
936 	if (out_len < sizeof resp)
937 		return -ENOSPC;
938 
939 	if (copy_from_user(&cmd, buf, sizeof cmd))
940 		return -EFAULT;
941 
942 	uobj = uobj_alloc(uobj_get_type(comp_channel), file->ucontext);
943 	if (IS_ERR(uobj))
944 		return PTR_ERR(uobj);
945 
946 	resp.fd = uobj->id;
947 
948 	ev_file = container_of(uobj, struct ib_uverbs_completion_event_file,
949 			       uobj_file.uobj);
950 	ib_uverbs_init_event_queue(&ev_file->ev_queue);
951 
952 	if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof resp)) {
953 		uobj_alloc_abort(uobj);
954 		return -EFAULT;
955 	}
956 
957 	uobj_alloc_commit(uobj);
958 	return in_len;
959 }
960 
961 static struct ib_ucq_object *create_cq(struct ib_uverbs_file *file,
962 					struct ib_device *ib_dev,
963 				       struct ib_udata *ucore,
964 				       struct ib_udata *uhw,
965 				       struct ib_uverbs_ex_create_cq *cmd,
966 				       size_t cmd_sz,
967 				       int (*cb)(struct ib_uverbs_file *file,
968 						 struct ib_ucq_object *obj,
969 						 struct ib_uverbs_ex_create_cq_resp *resp,
970 						 struct ib_udata *udata,
971 						 void *context),
972 				       void *context)
973 {
974 	struct ib_ucq_object           *obj;
975 	struct ib_uverbs_completion_event_file    *ev_file = NULL;
976 	struct ib_cq                   *cq;
977 	int                             ret;
978 	struct ib_uverbs_ex_create_cq_resp resp;
979 	struct ib_cq_init_attr attr = {};
980 
981 	if (!ib_dev->create_cq)
982 		return ERR_PTR(-EOPNOTSUPP);
983 
984 	if (cmd->comp_vector >= file->device->num_comp_vectors)
985 		return ERR_PTR(-EINVAL);
986 
987 	obj  = (struct ib_ucq_object *)uobj_alloc(uobj_get_type(cq),
988 						  file->ucontext);
989 	if (IS_ERR(obj))
990 		return obj;
991 
992 	if (cmd->comp_channel >= 0) {
993 		ev_file = ib_uverbs_lookup_comp_file(cmd->comp_channel,
994 						     file->ucontext);
995 		if (IS_ERR(ev_file)) {
996 			ret = PTR_ERR(ev_file);
997 			goto err;
998 		}
999 	}
1000 
1001 	obj->uobject.user_handle = cmd->user_handle;
1002 	obj->uverbs_file	   = file;
1003 	obj->comp_events_reported  = 0;
1004 	obj->async_events_reported = 0;
1005 	INIT_LIST_HEAD(&obj->comp_list);
1006 	INIT_LIST_HEAD(&obj->async_list);
1007 
1008 	attr.cqe = cmd->cqe;
1009 	attr.comp_vector = cmd->comp_vector;
1010 
1011 	if (cmd_sz > offsetof(typeof(*cmd), flags) + sizeof(cmd->flags))
1012 		attr.flags = cmd->flags;
1013 
1014 	cq = ib_dev->create_cq(ib_dev, &attr, file->ucontext, uhw);
1015 	if (IS_ERR(cq)) {
1016 		ret = PTR_ERR(cq);
1017 		goto err_file;
1018 	}
1019 
1020 	cq->device        = ib_dev;
1021 	cq->uobject       = &obj->uobject;
1022 	cq->comp_handler  = ib_uverbs_comp_handler;
1023 	cq->event_handler = ib_uverbs_cq_event_handler;
1024 	cq->cq_context    = ev_file ? &ev_file->ev_queue : NULL;
1025 	atomic_set(&cq->usecnt, 0);
1026 
1027 	obj->uobject.object = cq;
1028 	memset(&resp, 0, sizeof resp);
1029 	resp.base.cq_handle = obj->uobject.id;
1030 	resp.base.cqe       = cq->cqe;
1031 
1032 	resp.response_length = offsetof(typeof(resp), response_length) +
1033 		sizeof(resp.response_length);
1034 
1035 	cq->res.type = RDMA_RESTRACK_CQ;
1036 	rdma_restrack_add(&cq->res);
1037 
1038 	ret = cb(file, obj, &resp, ucore, context);
1039 	if (ret)
1040 		goto err_cb;
1041 
1042 	uobj_alloc_commit(&obj->uobject);
1043 	return obj;
1044 
1045 err_cb:
1046 	ib_destroy_cq(cq);
1047 
1048 err_file:
1049 	if (ev_file)
1050 		ib_uverbs_release_ucq(file, ev_file, obj);
1051 
1052 err:
1053 	uobj_alloc_abort(&obj->uobject);
1054 
1055 	return ERR_PTR(ret);
1056 }
1057 
1058 static int ib_uverbs_create_cq_cb(struct ib_uverbs_file *file,
1059 				  struct ib_ucq_object *obj,
1060 				  struct ib_uverbs_ex_create_cq_resp *resp,
1061 				  struct ib_udata *ucore, void *context)
1062 {
1063 	if (ib_copy_to_udata(ucore, &resp->base, sizeof(resp->base)))
1064 		return -EFAULT;
1065 
1066 	return 0;
1067 }
1068 
1069 ssize_t ib_uverbs_create_cq(struct ib_uverbs_file *file,
1070 			    struct ib_device *ib_dev,
1071 			    const char __user *buf, int in_len,
1072 			    int out_len)
1073 {
1074 	struct ib_uverbs_create_cq      cmd;
1075 	struct ib_uverbs_ex_create_cq	cmd_ex;
1076 	struct ib_uverbs_create_cq_resp resp;
1077 	struct ib_udata                 ucore;
1078 	struct ib_udata                 uhw;
1079 	struct ib_ucq_object           *obj;
1080 
1081 	if (out_len < sizeof(resp))
1082 		return -ENOSPC;
1083 
1084 	if (copy_from_user(&cmd, buf, sizeof(cmd)))
1085 		return -EFAULT;
1086 
1087 	ib_uverbs_init_udata(&ucore, buf, u64_to_user_ptr(cmd.response),
1088 			     sizeof(cmd), sizeof(resp));
1089 
1090 	ib_uverbs_init_udata(&uhw, buf + sizeof(cmd),
1091 		   u64_to_user_ptr(cmd.response) + sizeof(resp),
1092 		   in_len - sizeof(cmd) - sizeof(struct ib_uverbs_cmd_hdr),
1093 		   out_len - sizeof(resp));
1094 
1095 	memset(&cmd_ex, 0, sizeof(cmd_ex));
1096 	cmd_ex.user_handle = cmd.user_handle;
1097 	cmd_ex.cqe = cmd.cqe;
1098 	cmd_ex.comp_vector = cmd.comp_vector;
1099 	cmd_ex.comp_channel = cmd.comp_channel;
1100 
1101 	obj = create_cq(file, ib_dev, &ucore, &uhw, &cmd_ex,
1102 			offsetof(typeof(cmd_ex), comp_channel) +
1103 			sizeof(cmd.comp_channel), ib_uverbs_create_cq_cb,
1104 			NULL);
1105 
1106 	if (IS_ERR(obj))
1107 		return PTR_ERR(obj);
1108 
1109 	return in_len;
1110 }
1111 
1112 static int ib_uverbs_ex_create_cq_cb(struct ib_uverbs_file *file,
1113 				     struct ib_ucq_object *obj,
1114 				     struct ib_uverbs_ex_create_cq_resp *resp,
1115 				     struct ib_udata *ucore, void *context)
1116 {
1117 	if (ib_copy_to_udata(ucore, resp, resp->response_length))
1118 		return -EFAULT;
1119 
1120 	return 0;
1121 }
1122 
1123 int ib_uverbs_ex_create_cq(struct ib_uverbs_file *file,
1124 			 struct ib_device *ib_dev,
1125 			   struct ib_udata *ucore,
1126 			   struct ib_udata *uhw)
1127 {
1128 	struct ib_uverbs_ex_create_cq_resp resp;
1129 	struct ib_uverbs_ex_create_cq  cmd;
1130 	struct ib_ucq_object           *obj;
1131 	int err;
1132 
1133 	if (ucore->inlen < sizeof(cmd))
1134 		return -EINVAL;
1135 
1136 	err = ib_copy_from_udata(&cmd, ucore, sizeof(cmd));
1137 	if (err)
1138 		return err;
1139 
1140 	if (cmd.comp_mask)
1141 		return -EINVAL;
1142 
1143 	if (cmd.reserved)
1144 		return -EINVAL;
1145 
1146 	if (ucore->outlen < (offsetof(typeof(resp), response_length) +
1147 			     sizeof(resp.response_length)))
1148 		return -ENOSPC;
1149 
1150 	obj = create_cq(file, ib_dev, ucore, uhw, &cmd,
1151 			min(ucore->inlen, sizeof(cmd)),
1152 			ib_uverbs_ex_create_cq_cb, NULL);
1153 
1154 	return PTR_ERR_OR_ZERO(obj);
1155 }
1156 
1157 ssize_t ib_uverbs_resize_cq(struct ib_uverbs_file *file,
1158 			    struct ib_device *ib_dev,
1159 			    const char __user *buf, int in_len,
1160 			    int out_len)
1161 {
1162 	struct ib_uverbs_resize_cq	cmd;
1163 	struct ib_uverbs_resize_cq_resp	resp = {};
1164 	struct ib_udata                 udata;
1165 	struct ib_cq			*cq;
1166 	int				ret = -EINVAL;
1167 
1168 	if (copy_from_user(&cmd, buf, sizeof cmd))
1169 		return -EFAULT;
1170 
1171 	ib_uverbs_init_udata(&udata, buf + sizeof(cmd),
1172 		   u64_to_user_ptr(cmd.response) + sizeof(resp),
1173 		   in_len - sizeof(cmd) - sizeof(struct ib_uverbs_cmd_hdr),
1174 		   out_len - sizeof(resp));
1175 
1176 	cq = uobj_get_obj_read(cq, cmd.cq_handle, file->ucontext);
1177 	if (!cq)
1178 		return -EINVAL;
1179 
1180 	ret = cq->device->resize_cq(cq, cmd.cqe, &udata);
1181 	if (ret)
1182 		goto out;
1183 
1184 	resp.cqe = cq->cqe;
1185 
1186 	if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof resp.cqe))
1187 		ret = -EFAULT;
1188 
1189 out:
1190 	uobj_put_obj_read(cq);
1191 
1192 	return ret ? ret : in_len;
1193 }
1194 
1195 static int copy_wc_to_user(struct ib_device *ib_dev, void __user *dest,
1196 			   struct ib_wc *wc)
1197 {
1198 	struct ib_uverbs_wc tmp;
1199 
1200 	tmp.wr_id		= wc->wr_id;
1201 	tmp.status		= wc->status;
1202 	tmp.opcode		= wc->opcode;
1203 	tmp.vendor_err		= wc->vendor_err;
1204 	tmp.byte_len		= wc->byte_len;
1205 	tmp.ex.imm_data		= wc->ex.imm_data;
1206 	tmp.qp_num		= wc->qp->qp_num;
1207 	tmp.src_qp		= wc->src_qp;
1208 	tmp.wc_flags		= wc->wc_flags;
1209 	tmp.pkey_index		= wc->pkey_index;
1210 	if (rdma_cap_opa_ah(ib_dev, wc->port_num))
1211 		tmp.slid	= OPA_TO_IB_UCAST_LID(wc->slid);
1212 	else
1213 		tmp.slid	= ib_lid_cpu16(wc->slid);
1214 	tmp.sl			= wc->sl;
1215 	tmp.dlid_path_bits	= wc->dlid_path_bits;
1216 	tmp.port_num		= wc->port_num;
1217 	tmp.reserved		= 0;
1218 
1219 	if (copy_to_user(dest, &tmp, sizeof tmp))
1220 		return -EFAULT;
1221 
1222 	return 0;
1223 }
1224 
1225 ssize_t ib_uverbs_poll_cq(struct ib_uverbs_file *file,
1226 			  struct ib_device *ib_dev,
1227 			  const char __user *buf, int in_len,
1228 			  int out_len)
1229 {
1230 	struct ib_uverbs_poll_cq       cmd;
1231 	struct ib_uverbs_poll_cq_resp  resp;
1232 	u8 __user                     *header_ptr;
1233 	u8 __user                     *data_ptr;
1234 	struct ib_cq                  *cq;
1235 	struct ib_wc                   wc;
1236 	int                            ret;
1237 
1238 	if (copy_from_user(&cmd, buf, sizeof cmd))
1239 		return -EFAULT;
1240 
1241 	cq = uobj_get_obj_read(cq, cmd.cq_handle, file->ucontext);
1242 	if (!cq)
1243 		return -EINVAL;
1244 
1245 	/* we copy a struct ib_uverbs_poll_cq_resp to user space */
1246 	header_ptr = u64_to_user_ptr(cmd.response);
1247 	data_ptr = header_ptr + sizeof resp;
1248 
1249 	memset(&resp, 0, sizeof resp);
1250 	while (resp.count < cmd.ne) {
1251 		ret = ib_poll_cq(cq, 1, &wc);
1252 		if (ret < 0)
1253 			goto out_put;
1254 		if (!ret)
1255 			break;
1256 
1257 		ret = copy_wc_to_user(ib_dev, data_ptr, &wc);
1258 		if (ret)
1259 			goto out_put;
1260 
1261 		data_ptr += sizeof(struct ib_uverbs_wc);
1262 		++resp.count;
1263 	}
1264 
1265 	if (copy_to_user(header_ptr, &resp, sizeof resp)) {
1266 		ret = -EFAULT;
1267 		goto out_put;
1268 	}
1269 
1270 	ret = in_len;
1271 
1272 out_put:
1273 	uobj_put_obj_read(cq);
1274 	return ret;
1275 }
1276 
1277 ssize_t ib_uverbs_req_notify_cq(struct ib_uverbs_file *file,
1278 				struct ib_device *ib_dev,
1279 				const char __user *buf, int in_len,
1280 				int out_len)
1281 {
1282 	struct ib_uverbs_req_notify_cq cmd;
1283 	struct ib_cq                  *cq;
1284 
1285 	if (copy_from_user(&cmd, buf, sizeof cmd))
1286 		return -EFAULT;
1287 
1288 	cq = uobj_get_obj_read(cq, cmd.cq_handle, file->ucontext);
1289 	if (!cq)
1290 		return -EINVAL;
1291 
1292 	ib_req_notify_cq(cq, cmd.solicited_only ?
1293 			 IB_CQ_SOLICITED : IB_CQ_NEXT_COMP);
1294 
1295 	uobj_put_obj_read(cq);
1296 
1297 	return in_len;
1298 }
1299 
1300 ssize_t ib_uverbs_destroy_cq(struct ib_uverbs_file *file,
1301 			     struct ib_device *ib_dev,
1302 			     const char __user *buf, int in_len,
1303 			     int out_len)
1304 {
1305 	struct ib_uverbs_destroy_cq      cmd;
1306 	struct ib_uverbs_destroy_cq_resp resp;
1307 	struct ib_uobject		*uobj;
1308 	struct ib_cq               	*cq;
1309 	struct ib_ucq_object        	*obj;
1310 	int                        	 ret = -EINVAL;
1311 
1312 	if (copy_from_user(&cmd, buf, sizeof cmd))
1313 		return -EFAULT;
1314 
1315 	uobj  = uobj_get_write(uobj_get_type(cq), cmd.cq_handle,
1316 			       file->ucontext);
1317 	if (IS_ERR(uobj))
1318 		return PTR_ERR(uobj);
1319 
1320 	/*
1321 	 * Make sure we don't free the memory in remove_commit as we still
1322 	 * needs the uobject memory to create the response.
1323 	 */
1324 	uverbs_uobject_get(uobj);
1325 	cq      = uobj->object;
1326 	obj     = container_of(cq->uobject, struct ib_ucq_object, uobject);
1327 
1328 	memset(&resp, 0, sizeof(resp));
1329 
1330 	ret = uobj_remove_commit(uobj);
1331 	if (ret) {
1332 		uverbs_uobject_put(uobj);
1333 		return ret;
1334 	}
1335 
1336 	resp.comp_events_reported  = obj->comp_events_reported;
1337 	resp.async_events_reported = obj->async_events_reported;
1338 
1339 	uverbs_uobject_put(uobj);
1340 	if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof resp))
1341 		return -EFAULT;
1342 
1343 	return in_len;
1344 }
1345 
1346 static int create_qp(struct ib_uverbs_file *file,
1347 		     struct ib_udata *ucore,
1348 		     struct ib_udata *uhw,
1349 		     struct ib_uverbs_ex_create_qp *cmd,
1350 		     size_t cmd_sz,
1351 		     int (*cb)(struct ib_uverbs_file *file,
1352 			       struct ib_uverbs_ex_create_qp_resp *resp,
1353 			       struct ib_udata *udata),
1354 		     void *context)
1355 {
1356 	struct ib_uqp_object		*obj;
1357 	struct ib_device		*device;
1358 	struct ib_pd			*pd = NULL;
1359 	struct ib_xrcd			*xrcd = NULL;
1360 	struct ib_uobject		*xrcd_uobj = ERR_PTR(-ENOENT);
1361 	struct ib_cq			*scq = NULL, *rcq = NULL;
1362 	struct ib_srq			*srq = NULL;
1363 	struct ib_qp			*qp;
1364 	char				*buf;
1365 	struct ib_qp_init_attr		attr = {};
1366 	struct ib_uverbs_ex_create_qp_resp resp;
1367 	int				ret;
1368 	struct ib_rwq_ind_table *ind_tbl = NULL;
1369 	bool has_sq = true;
1370 
1371 	if (cmd->qp_type == IB_QPT_RAW_PACKET && !capable(CAP_NET_RAW))
1372 		return -EPERM;
1373 
1374 	obj  = (struct ib_uqp_object *)uobj_alloc(uobj_get_type(qp),
1375 						  file->ucontext);
1376 	if (IS_ERR(obj))
1377 		return PTR_ERR(obj);
1378 	obj->uxrcd = NULL;
1379 	obj->uevent.uobject.user_handle = cmd->user_handle;
1380 	mutex_init(&obj->mcast_lock);
1381 
1382 	if (cmd_sz >= offsetof(typeof(*cmd), rwq_ind_tbl_handle) +
1383 		      sizeof(cmd->rwq_ind_tbl_handle) &&
1384 		      (cmd->comp_mask & IB_UVERBS_CREATE_QP_MASK_IND_TABLE)) {
1385 		ind_tbl = uobj_get_obj_read(rwq_ind_table,
1386 					    cmd->rwq_ind_tbl_handle,
1387 					    file->ucontext);
1388 		if (!ind_tbl) {
1389 			ret = -EINVAL;
1390 			goto err_put;
1391 		}
1392 
1393 		attr.rwq_ind_tbl = ind_tbl;
1394 	}
1395 
1396 	if (cmd_sz > sizeof(*cmd) &&
1397 	    !ib_is_udata_cleared(ucore, sizeof(*cmd),
1398 				 cmd_sz - sizeof(*cmd))) {
1399 		ret = -EOPNOTSUPP;
1400 		goto err_put;
1401 	}
1402 
1403 	if (ind_tbl && (cmd->max_recv_wr || cmd->max_recv_sge || cmd->is_srq)) {
1404 		ret = -EINVAL;
1405 		goto err_put;
1406 	}
1407 
1408 	if (ind_tbl && !cmd->max_send_wr)
1409 		has_sq = false;
1410 
1411 	if (cmd->qp_type == IB_QPT_XRC_TGT) {
1412 		xrcd_uobj = uobj_get_read(uobj_get_type(xrcd), cmd->pd_handle,
1413 					  file->ucontext);
1414 
1415 		if (IS_ERR(xrcd_uobj)) {
1416 			ret = -EINVAL;
1417 			goto err_put;
1418 		}
1419 
1420 		xrcd = (struct ib_xrcd *)xrcd_uobj->object;
1421 		if (!xrcd) {
1422 			ret = -EINVAL;
1423 			goto err_put;
1424 		}
1425 		device = xrcd->device;
1426 	} else {
1427 		if (cmd->qp_type == IB_QPT_XRC_INI) {
1428 			cmd->max_recv_wr = 0;
1429 			cmd->max_recv_sge = 0;
1430 		} else {
1431 			if (cmd->is_srq) {
1432 				srq = uobj_get_obj_read(srq, cmd->srq_handle,
1433 							file->ucontext);
1434 				if (!srq || srq->srq_type == IB_SRQT_XRC) {
1435 					ret = -EINVAL;
1436 					goto err_put;
1437 				}
1438 			}
1439 
1440 			if (!ind_tbl) {
1441 				if (cmd->recv_cq_handle != cmd->send_cq_handle) {
1442 					rcq = uobj_get_obj_read(cq, cmd->recv_cq_handle,
1443 								file->ucontext);
1444 					if (!rcq) {
1445 						ret = -EINVAL;
1446 						goto err_put;
1447 					}
1448 				}
1449 			}
1450 		}
1451 
1452 		if (has_sq)
1453 			scq = uobj_get_obj_read(cq, cmd->send_cq_handle,
1454 						file->ucontext);
1455 		if (!ind_tbl)
1456 			rcq = rcq ?: scq;
1457 		pd  = uobj_get_obj_read(pd, cmd->pd_handle, file->ucontext);
1458 		if (!pd || (!scq && has_sq)) {
1459 			ret = -EINVAL;
1460 			goto err_put;
1461 		}
1462 
1463 		device = pd->device;
1464 	}
1465 
1466 	attr.event_handler = ib_uverbs_qp_event_handler;
1467 	attr.qp_context    = file;
1468 	attr.send_cq       = scq;
1469 	attr.recv_cq       = rcq;
1470 	attr.srq           = srq;
1471 	attr.xrcd	   = xrcd;
1472 	attr.sq_sig_type   = cmd->sq_sig_all ? IB_SIGNAL_ALL_WR :
1473 					      IB_SIGNAL_REQ_WR;
1474 	attr.qp_type       = cmd->qp_type;
1475 	attr.create_flags  = 0;
1476 
1477 	attr.cap.max_send_wr     = cmd->max_send_wr;
1478 	attr.cap.max_recv_wr     = cmd->max_recv_wr;
1479 	attr.cap.max_send_sge    = cmd->max_send_sge;
1480 	attr.cap.max_recv_sge    = cmd->max_recv_sge;
1481 	attr.cap.max_inline_data = cmd->max_inline_data;
1482 
1483 	obj->uevent.events_reported     = 0;
1484 	INIT_LIST_HEAD(&obj->uevent.event_list);
1485 	INIT_LIST_HEAD(&obj->mcast_list);
1486 
1487 	if (cmd_sz >= offsetof(typeof(*cmd), create_flags) +
1488 		      sizeof(cmd->create_flags))
1489 		attr.create_flags = cmd->create_flags;
1490 
1491 	if (attr.create_flags & ~(IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK |
1492 				IB_QP_CREATE_CROSS_CHANNEL |
1493 				IB_QP_CREATE_MANAGED_SEND |
1494 				IB_QP_CREATE_MANAGED_RECV |
1495 				IB_QP_CREATE_SCATTER_FCS |
1496 				IB_QP_CREATE_CVLAN_STRIPPING |
1497 				IB_QP_CREATE_SOURCE_QPN |
1498 				IB_QP_CREATE_PCI_WRITE_END_PADDING)) {
1499 		ret = -EINVAL;
1500 		goto err_put;
1501 	}
1502 
1503 	if (attr.create_flags & IB_QP_CREATE_SOURCE_QPN) {
1504 		if (!capable(CAP_NET_RAW)) {
1505 			ret = -EPERM;
1506 			goto err_put;
1507 		}
1508 
1509 		attr.source_qpn = cmd->source_qpn;
1510 	}
1511 
1512 	buf = (void *)cmd + sizeof(*cmd);
1513 	if (cmd_sz > sizeof(*cmd))
1514 		if (!(buf[0] == 0 && !memcmp(buf, buf + 1,
1515 					     cmd_sz - sizeof(*cmd) - 1))) {
1516 			ret = -EINVAL;
1517 			goto err_put;
1518 		}
1519 
1520 	if (cmd->qp_type == IB_QPT_XRC_TGT)
1521 		qp = ib_create_qp(pd, &attr);
1522 	else
1523 		qp = _ib_create_qp(device, pd, &attr, uhw,
1524 				   &obj->uevent.uobject);
1525 
1526 	if (IS_ERR(qp)) {
1527 		ret = PTR_ERR(qp);
1528 		goto err_put;
1529 	}
1530 
1531 	if (cmd->qp_type != IB_QPT_XRC_TGT) {
1532 		ret = ib_create_qp_security(qp, device);
1533 		if (ret)
1534 			goto err_cb;
1535 
1536 		qp->real_qp	  = qp;
1537 		qp->pd		  = pd;
1538 		qp->send_cq	  = attr.send_cq;
1539 		qp->recv_cq	  = attr.recv_cq;
1540 		qp->srq		  = attr.srq;
1541 		qp->rwq_ind_tbl	  = ind_tbl;
1542 		qp->event_handler = attr.event_handler;
1543 		qp->qp_context	  = attr.qp_context;
1544 		qp->qp_type	  = attr.qp_type;
1545 		atomic_set(&qp->usecnt, 0);
1546 		atomic_inc(&pd->usecnt);
1547 		qp->port = 0;
1548 		if (attr.send_cq)
1549 			atomic_inc(&attr.send_cq->usecnt);
1550 		if (attr.recv_cq)
1551 			atomic_inc(&attr.recv_cq->usecnt);
1552 		if (attr.srq)
1553 			atomic_inc(&attr.srq->usecnt);
1554 		if (ind_tbl)
1555 			atomic_inc(&ind_tbl->usecnt);
1556 	} else {
1557 		/* It is done in _ib_create_qp for other QP types */
1558 		qp->uobject = &obj->uevent.uobject;
1559 	}
1560 
1561 	obj->uevent.uobject.object = qp;
1562 
1563 	memset(&resp, 0, sizeof resp);
1564 	resp.base.qpn             = qp->qp_num;
1565 	resp.base.qp_handle       = obj->uevent.uobject.id;
1566 	resp.base.max_recv_sge    = attr.cap.max_recv_sge;
1567 	resp.base.max_send_sge    = attr.cap.max_send_sge;
1568 	resp.base.max_recv_wr     = attr.cap.max_recv_wr;
1569 	resp.base.max_send_wr     = attr.cap.max_send_wr;
1570 	resp.base.max_inline_data = attr.cap.max_inline_data;
1571 
1572 	resp.response_length = offsetof(typeof(resp), response_length) +
1573 			       sizeof(resp.response_length);
1574 
1575 	ret = cb(file, &resp, ucore);
1576 	if (ret)
1577 		goto err_cb;
1578 
1579 	if (xrcd) {
1580 		obj->uxrcd = container_of(xrcd_uobj, struct ib_uxrcd_object,
1581 					  uobject);
1582 		atomic_inc(&obj->uxrcd->refcnt);
1583 		uobj_put_read(xrcd_uobj);
1584 	}
1585 
1586 	if (pd)
1587 		uobj_put_obj_read(pd);
1588 	if (scq)
1589 		uobj_put_obj_read(scq);
1590 	if (rcq && rcq != scq)
1591 		uobj_put_obj_read(rcq);
1592 	if (srq)
1593 		uobj_put_obj_read(srq);
1594 	if (ind_tbl)
1595 		uobj_put_obj_read(ind_tbl);
1596 
1597 	uobj_alloc_commit(&obj->uevent.uobject);
1598 
1599 	return 0;
1600 err_cb:
1601 	ib_destroy_qp(qp);
1602 
1603 err_put:
1604 	if (!IS_ERR(xrcd_uobj))
1605 		uobj_put_read(xrcd_uobj);
1606 	if (pd)
1607 		uobj_put_obj_read(pd);
1608 	if (scq)
1609 		uobj_put_obj_read(scq);
1610 	if (rcq && rcq != scq)
1611 		uobj_put_obj_read(rcq);
1612 	if (srq)
1613 		uobj_put_obj_read(srq);
1614 	if (ind_tbl)
1615 		uobj_put_obj_read(ind_tbl);
1616 
1617 	uobj_alloc_abort(&obj->uevent.uobject);
1618 	return ret;
1619 }
1620 
1621 static int ib_uverbs_create_qp_cb(struct ib_uverbs_file *file,
1622 				  struct ib_uverbs_ex_create_qp_resp *resp,
1623 				  struct ib_udata *ucore)
1624 {
1625 	if (ib_copy_to_udata(ucore, &resp->base, sizeof(resp->base)))
1626 		return -EFAULT;
1627 
1628 	return 0;
1629 }
1630 
1631 ssize_t ib_uverbs_create_qp(struct ib_uverbs_file *file,
1632 			    struct ib_device *ib_dev,
1633 			    const char __user *buf, int in_len,
1634 			    int out_len)
1635 {
1636 	struct ib_uverbs_create_qp      cmd;
1637 	struct ib_uverbs_ex_create_qp	cmd_ex;
1638 	struct ib_udata			ucore;
1639 	struct ib_udata			uhw;
1640 	ssize_t resp_size = sizeof(struct ib_uverbs_create_qp_resp);
1641 	int				err;
1642 
1643 	if (out_len < resp_size)
1644 		return -ENOSPC;
1645 
1646 	if (copy_from_user(&cmd, buf, sizeof(cmd)))
1647 		return -EFAULT;
1648 
1649 	ib_uverbs_init_udata(&ucore, buf, u64_to_user_ptr(cmd.response),
1650 		   sizeof(cmd), resp_size);
1651 	ib_uverbs_init_udata(&uhw, buf + sizeof(cmd),
1652 		   u64_to_user_ptr(cmd.response) + resp_size,
1653 		   in_len - sizeof(cmd) - sizeof(struct ib_uverbs_cmd_hdr),
1654 		   out_len - resp_size);
1655 
1656 	memset(&cmd_ex, 0, sizeof(cmd_ex));
1657 	cmd_ex.user_handle = cmd.user_handle;
1658 	cmd_ex.pd_handle = cmd.pd_handle;
1659 	cmd_ex.send_cq_handle = cmd.send_cq_handle;
1660 	cmd_ex.recv_cq_handle = cmd.recv_cq_handle;
1661 	cmd_ex.srq_handle = cmd.srq_handle;
1662 	cmd_ex.max_send_wr = cmd.max_send_wr;
1663 	cmd_ex.max_recv_wr = cmd.max_recv_wr;
1664 	cmd_ex.max_send_sge = cmd.max_send_sge;
1665 	cmd_ex.max_recv_sge = cmd.max_recv_sge;
1666 	cmd_ex.max_inline_data = cmd.max_inline_data;
1667 	cmd_ex.sq_sig_all = cmd.sq_sig_all;
1668 	cmd_ex.qp_type = cmd.qp_type;
1669 	cmd_ex.is_srq = cmd.is_srq;
1670 
1671 	err = create_qp(file, &ucore, &uhw, &cmd_ex,
1672 			offsetof(typeof(cmd_ex), is_srq) +
1673 			sizeof(cmd.is_srq), ib_uverbs_create_qp_cb,
1674 			NULL);
1675 
1676 	if (err)
1677 		return err;
1678 
1679 	return in_len;
1680 }
1681 
1682 static int ib_uverbs_ex_create_qp_cb(struct ib_uverbs_file *file,
1683 				     struct ib_uverbs_ex_create_qp_resp *resp,
1684 				     struct ib_udata *ucore)
1685 {
1686 	if (ib_copy_to_udata(ucore, resp, resp->response_length))
1687 		return -EFAULT;
1688 
1689 	return 0;
1690 }
1691 
1692 int ib_uverbs_ex_create_qp(struct ib_uverbs_file *file,
1693 			   struct ib_device *ib_dev,
1694 			   struct ib_udata *ucore,
1695 			   struct ib_udata *uhw)
1696 {
1697 	struct ib_uverbs_ex_create_qp_resp resp;
1698 	struct ib_uverbs_ex_create_qp cmd = {0};
1699 	int err;
1700 
1701 	if (ucore->inlen < (offsetof(typeof(cmd), comp_mask) +
1702 			    sizeof(cmd.comp_mask)))
1703 		return -EINVAL;
1704 
1705 	err = ib_copy_from_udata(&cmd, ucore, min(sizeof(cmd), ucore->inlen));
1706 	if (err)
1707 		return err;
1708 
1709 	if (cmd.comp_mask & ~IB_UVERBS_CREATE_QP_SUP_COMP_MASK)
1710 		return -EINVAL;
1711 
1712 	if (cmd.reserved)
1713 		return -EINVAL;
1714 
1715 	if (ucore->outlen < (offsetof(typeof(resp), response_length) +
1716 			     sizeof(resp.response_length)))
1717 		return -ENOSPC;
1718 
1719 	err = create_qp(file, ucore, uhw, &cmd,
1720 			min(ucore->inlen, sizeof(cmd)),
1721 			ib_uverbs_ex_create_qp_cb, NULL);
1722 
1723 	if (err)
1724 		return err;
1725 
1726 	return 0;
1727 }
1728 
1729 ssize_t ib_uverbs_open_qp(struct ib_uverbs_file *file,
1730 			  struct ib_device *ib_dev,
1731 			  const char __user *buf, int in_len, int out_len)
1732 {
1733 	struct ib_uverbs_open_qp        cmd;
1734 	struct ib_uverbs_create_qp_resp resp;
1735 	struct ib_udata                 udata;
1736 	struct ib_uqp_object           *obj;
1737 	struct ib_xrcd		       *xrcd;
1738 	struct ib_uobject	       *uninitialized_var(xrcd_uobj);
1739 	struct ib_qp                   *qp;
1740 	struct ib_qp_open_attr          attr;
1741 	int ret;
1742 
1743 	if (out_len < sizeof resp)
1744 		return -ENOSPC;
1745 
1746 	if (copy_from_user(&cmd, buf, sizeof cmd))
1747 		return -EFAULT;
1748 
1749 	ib_uverbs_init_udata(&udata, buf + sizeof(cmd),
1750 		   u64_to_user_ptr(cmd.response) + sizeof(resp),
1751 		   in_len - sizeof(cmd) - sizeof(struct ib_uverbs_cmd_hdr),
1752 		   out_len - sizeof(resp));
1753 
1754 	obj  = (struct ib_uqp_object *)uobj_alloc(uobj_get_type(qp),
1755 						  file->ucontext);
1756 	if (IS_ERR(obj))
1757 		return PTR_ERR(obj);
1758 
1759 	xrcd_uobj = uobj_get_read(uobj_get_type(xrcd), cmd.pd_handle,
1760 				  file->ucontext);
1761 	if (IS_ERR(xrcd_uobj)) {
1762 		ret = -EINVAL;
1763 		goto err_put;
1764 	}
1765 
1766 	xrcd = (struct ib_xrcd *)xrcd_uobj->object;
1767 	if (!xrcd) {
1768 		ret = -EINVAL;
1769 		goto err_xrcd;
1770 	}
1771 
1772 	attr.event_handler = ib_uverbs_qp_event_handler;
1773 	attr.qp_context    = file;
1774 	attr.qp_num        = cmd.qpn;
1775 	attr.qp_type       = cmd.qp_type;
1776 
1777 	obj->uevent.events_reported = 0;
1778 	INIT_LIST_HEAD(&obj->uevent.event_list);
1779 	INIT_LIST_HEAD(&obj->mcast_list);
1780 
1781 	qp = ib_open_qp(xrcd, &attr);
1782 	if (IS_ERR(qp)) {
1783 		ret = PTR_ERR(qp);
1784 		goto err_xrcd;
1785 	}
1786 
1787 	obj->uevent.uobject.object = qp;
1788 	obj->uevent.uobject.user_handle = cmd.user_handle;
1789 
1790 	memset(&resp, 0, sizeof resp);
1791 	resp.qpn       = qp->qp_num;
1792 	resp.qp_handle = obj->uevent.uobject.id;
1793 
1794 	if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof resp)) {
1795 		ret = -EFAULT;
1796 		goto err_destroy;
1797 	}
1798 
1799 	obj->uxrcd = container_of(xrcd_uobj, struct ib_uxrcd_object, uobject);
1800 	atomic_inc(&obj->uxrcd->refcnt);
1801 	qp->uobject = &obj->uevent.uobject;
1802 	uobj_put_read(xrcd_uobj);
1803 
1804 
1805 	uobj_alloc_commit(&obj->uevent.uobject);
1806 
1807 	return in_len;
1808 
1809 err_destroy:
1810 	ib_destroy_qp(qp);
1811 err_xrcd:
1812 	uobj_put_read(xrcd_uobj);
1813 err_put:
1814 	uobj_alloc_abort(&obj->uevent.uobject);
1815 	return ret;
1816 }
1817 
1818 static void copy_ah_attr_to_uverbs(struct ib_uverbs_qp_dest *uverb_attr,
1819 				   struct rdma_ah_attr *rdma_attr)
1820 {
1821 	const struct ib_global_route   *grh;
1822 
1823 	uverb_attr->dlid              = rdma_ah_get_dlid(rdma_attr);
1824 	uverb_attr->sl                = rdma_ah_get_sl(rdma_attr);
1825 	uverb_attr->src_path_bits     = rdma_ah_get_path_bits(rdma_attr);
1826 	uverb_attr->static_rate       = rdma_ah_get_static_rate(rdma_attr);
1827 	uverb_attr->is_global         = !!(rdma_ah_get_ah_flags(rdma_attr) &
1828 					 IB_AH_GRH);
1829 	if (uverb_attr->is_global) {
1830 		grh = rdma_ah_read_grh(rdma_attr);
1831 		memcpy(uverb_attr->dgid, grh->dgid.raw, 16);
1832 		uverb_attr->flow_label        = grh->flow_label;
1833 		uverb_attr->sgid_index        = grh->sgid_index;
1834 		uverb_attr->hop_limit         = grh->hop_limit;
1835 		uverb_attr->traffic_class     = grh->traffic_class;
1836 	}
1837 	uverb_attr->port_num          = rdma_ah_get_port_num(rdma_attr);
1838 }
1839 
1840 ssize_t ib_uverbs_query_qp(struct ib_uverbs_file *file,
1841 			   struct ib_device *ib_dev,
1842 			   const char __user *buf, int in_len,
1843 			   int out_len)
1844 {
1845 	struct ib_uverbs_query_qp      cmd;
1846 	struct ib_uverbs_query_qp_resp resp;
1847 	struct ib_qp                   *qp;
1848 	struct ib_qp_attr              *attr;
1849 	struct ib_qp_init_attr         *init_attr;
1850 	int                            ret;
1851 
1852 	if (copy_from_user(&cmd, buf, sizeof cmd))
1853 		return -EFAULT;
1854 
1855 	attr      = kmalloc(sizeof *attr, GFP_KERNEL);
1856 	init_attr = kmalloc(sizeof *init_attr, GFP_KERNEL);
1857 	if (!attr || !init_attr) {
1858 		ret = -ENOMEM;
1859 		goto out;
1860 	}
1861 
1862 	qp = uobj_get_obj_read(qp, cmd.qp_handle, file->ucontext);
1863 	if (!qp) {
1864 		ret = -EINVAL;
1865 		goto out;
1866 	}
1867 
1868 	ret = ib_query_qp(qp, attr, cmd.attr_mask, init_attr);
1869 
1870 	uobj_put_obj_read(qp);
1871 
1872 	if (ret)
1873 		goto out;
1874 
1875 	memset(&resp, 0, sizeof resp);
1876 
1877 	resp.qp_state               = attr->qp_state;
1878 	resp.cur_qp_state           = attr->cur_qp_state;
1879 	resp.path_mtu               = attr->path_mtu;
1880 	resp.path_mig_state         = attr->path_mig_state;
1881 	resp.qkey                   = attr->qkey;
1882 	resp.rq_psn                 = attr->rq_psn;
1883 	resp.sq_psn                 = attr->sq_psn;
1884 	resp.dest_qp_num            = attr->dest_qp_num;
1885 	resp.qp_access_flags        = attr->qp_access_flags;
1886 	resp.pkey_index             = attr->pkey_index;
1887 	resp.alt_pkey_index         = attr->alt_pkey_index;
1888 	resp.sq_draining            = attr->sq_draining;
1889 	resp.max_rd_atomic          = attr->max_rd_atomic;
1890 	resp.max_dest_rd_atomic     = attr->max_dest_rd_atomic;
1891 	resp.min_rnr_timer          = attr->min_rnr_timer;
1892 	resp.port_num               = attr->port_num;
1893 	resp.timeout                = attr->timeout;
1894 	resp.retry_cnt              = attr->retry_cnt;
1895 	resp.rnr_retry              = attr->rnr_retry;
1896 	resp.alt_port_num           = attr->alt_port_num;
1897 	resp.alt_timeout            = attr->alt_timeout;
1898 
1899 	copy_ah_attr_to_uverbs(&resp.dest, &attr->ah_attr);
1900 	copy_ah_attr_to_uverbs(&resp.alt_dest, &attr->alt_ah_attr);
1901 
1902 	resp.max_send_wr            = init_attr->cap.max_send_wr;
1903 	resp.max_recv_wr            = init_attr->cap.max_recv_wr;
1904 	resp.max_send_sge           = init_attr->cap.max_send_sge;
1905 	resp.max_recv_sge           = init_attr->cap.max_recv_sge;
1906 	resp.max_inline_data        = init_attr->cap.max_inline_data;
1907 	resp.sq_sig_all             = init_attr->sq_sig_type == IB_SIGNAL_ALL_WR;
1908 
1909 	if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof resp))
1910 		ret = -EFAULT;
1911 
1912 out:
1913 	kfree(attr);
1914 	kfree(init_attr);
1915 
1916 	return ret ? ret : in_len;
1917 }
1918 
1919 /* Remove ignored fields set in the attribute mask */
1920 static int modify_qp_mask(enum ib_qp_type qp_type, int mask)
1921 {
1922 	switch (qp_type) {
1923 	case IB_QPT_XRC_INI:
1924 		return mask & ~(IB_QP_MAX_DEST_RD_ATOMIC | IB_QP_MIN_RNR_TIMER);
1925 	case IB_QPT_XRC_TGT:
1926 		return mask & ~(IB_QP_MAX_QP_RD_ATOMIC | IB_QP_RETRY_CNT |
1927 				IB_QP_RNR_RETRY);
1928 	default:
1929 		return mask;
1930 	}
1931 }
1932 
1933 static void copy_ah_attr_from_uverbs(struct ib_device *dev,
1934 				     struct rdma_ah_attr *rdma_attr,
1935 				     struct ib_uverbs_qp_dest *uverb_attr)
1936 {
1937 	rdma_attr->type = rdma_ah_find_type(dev, uverb_attr->port_num);
1938 	if (uverb_attr->is_global) {
1939 		rdma_ah_set_grh(rdma_attr, NULL,
1940 				uverb_attr->flow_label,
1941 				uverb_attr->sgid_index,
1942 				uverb_attr->hop_limit,
1943 				uverb_attr->traffic_class);
1944 		rdma_ah_set_dgid_raw(rdma_attr, uverb_attr->dgid);
1945 	} else {
1946 		rdma_ah_set_ah_flags(rdma_attr, 0);
1947 	}
1948 	rdma_ah_set_dlid(rdma_attr, uverb_attr->dlid);
1949 	rdma_ah_set_sl(rdma_attr, uverb_attr->sl);
1950 	rdma_ah_set_path_bits(rdma_attr, uverb_attr->src_path_bits);
1951 	rdma_ah_set_static_rate(rdma_attr, uverb_attr->static_rate);
1952 	rdma_ah_set_port_num(rdma_attr, uverb_attr->port_num);
1953 	rdma_ah_set_make_grd(rdma_attr, false);
1954 }
1955 
1956 static int modify_qp(struct ib_uverbs_file *file,
1957 		     struct ib_uverbs_ex_modify_qp *cmd, struct ib_udata *udata)
1958 {
1959 	struct ib_qp_attr *attr;
1960 	struct ib_qp *qp;
1961 	int ret;
1962 
1963 	attr = kmalloc(sizeof *attr, GFP_KERNEL);
1964 	if (!attr)
1965 		return -ENOMEM;
1966 
1967 	qp = uobj_get_obj_read(qp, cmd->base.qp_handle, file->ucontext);
1968 	if (!qp) {
1969 		ret = -EINVAL;
1970 		goto out;
1971 	}
1972 
1973 	if ((cmd->base.attr_mask & IB_QP_PORT) &&
1974 	    !rdma_is_port_valid(qp->device, cmd->base.port_num)) {
1975 		ret = -EINVAL;
1976 		goto release_qp;
1977 	}
1978 
1979 	if ((cmd->base.attr_mask & IB_QP_AV) &&
1980 	    !rdma_is_port_valid(qp->device, cmd->base.dest.port_num)) {
1981 		ret = -EINVAL;
1982 		goto release_qp;
1983 	}
1984 
1985 	if ((cmd->base.attr_mask & IB_QP_ALT_PATH) &&
1986 	    (!rdma_is_port_valid(qp->device, cmd->base.alt_port_num) ||
1987 	    !rdma_is_port_valid(qp->device, cmd->base.alt_dest.port_num))) {
1988 		ret = -EINVAL;
1989 		goto release_qp;
1990 	}
1991 
1992 	attr->qp_state		  = cmd->base.qp_state;
1993 	attr->cur_qp_state	  = cmd->base.cur_qp_state;
1994 	attr->path_mtu		  = cmd->base.path_mtu;
1995 	attr->path_mig_state	  = cmd->base.path_mig_state;
1996 	attr->qkey		  = cmd->base.qkey;
1997 	attr->rq_psn		  = cmd->base.rq_psn;
1998 	attr->sq_psn		  = cmd->base.sq_psn;
1999 	attr->dest_qp_num	  = cmd->base.dest_qp_num;
2000 	attr->qp_access_flags	  = cmd->base.qp_access_flags;
2001 	attr->pkey_index	  = cmd->base.pkey_index;
2002 	attr->alt_pkey_index	  = cmd->base.alt_pkey_index;
2003 	attr->en_sqd_async_notify = cmd->base.en_sqd_async_notify;
2004 	attr->max_rd_atomic	  = cmd->base.max_rd_atomic;
2005 	attr->max_dest_rd_atomic  = cmd->base.max_dest_rd_atomic;
2006 	attr->min_rnr_timer	  = cmd->base.min_rnr_timer;
2007 	attr->port_num		  = cmd->base.port_num;
2008 	attr->timeout		  = cmd->base.timeout;
2009 	attr->retry_cnt		  = cmd->base.retry_cnt;
2010 	attr->rnr_retry		  = cmd->base.rnr_retry;
2011 	attr->alt_port_num	  = cmd->base.alt_port_num;
2012 	attr->alt_timeout	  = cmd->base.alt_timeout;
2013 	attr->rate_limit	  = cmd->rate_limit;
2014 
2015 	if (cmd->base.attr_mask & IB_QP_AV)
2016 		copy_ah_attr_from_uverbs(qp->device, &attr->ah_attr,
2017 					 &cmd->base.dest);
2018 
2019 	if (cmd->base.attr_mask & IB_QP_ALT_PATH)
2020 		copy_ah_attr_from_uverbs(qp->device, &attr->alt_ah_attr,
2021 					 &cmd->base.alt_dest);
2022 
2023 	ret = ib_modify_qp_with_udata(qp, attr,
2024 				      modify_qp_mask(qp->qp_type,
2025 						     cmd->base.attr_mask),
2026 				      udata);
2027 
2028 release_qp:
2029 	uobj_put_obj_read(qp);
2030 out:
2031 	kfree(attr);
2032 
2033 	return ret;
2034 }
2035 
2036 ssize_t ib_uverbs_modify_qp(struct ib_uverbs_file *file,
2037 			    struct ib_device *ib_dev,
2038 			    const char __user *buf, int in_len,
2039 			    int out_len)
2040 {
2041 	struct ib_uverbs_ex_modify_qp cmd = {};
2042 	struct ib_udata udata;
2043 	int ret;
2044 
2045 	if (copy_from_user(&cmd.base, buf, sizeof(cmd.base)))
2046 		return -EFAULT;
2047 
2048 	if (cmd.base.attr_mask &
2049 	    ~((IB_USER_LEGACY_LAST_QP_ATTR_MASK << 1) - 1))
2050 		return -EOPNOTSUPP;
2051 
2052 	ib_uverbs_init_udata(&udata, buf + sizeof(cmd.base), NULL,
2053 		   in_len - sizeof(cmd.base) - sizeof(struct ib_uverbs_cmd_hdr),
2054 		   out_len);
2055 
2056 	ret = modify_qp(file, &cmd, &udata);
2057 	if (ret)
2058 		return ret;
2059 
2060 	return in_len;
2061 }
2062 
2063 int ib_uverbs_ex_modify_qp(struct ib_uverbs_file *file,
2064 			   struct ib_device *ib_dev,
2065 			   struct ib_udata *ucore,
2066 			   struct ib_udata *uhw)
2067 {
2068 	struct ib_uverbs_ex_modify_qp cmd = {};
2069 	int ret;
2070 
2071 	/*
2072 	 * Last bit is reserved for extending the attr_mask by
2073 	 * using another field.
2074 	 */
2075 	BUILD_BUG_ON(IB_USER_LAST_QP_ATTR_MASK == (1 << 31));
2076 
2077 	if (ucore->inlen < sizeof(cmd.base))
2078 		return -EINVAL;
2079 
2080 	ret = ib_copy_from_udata(&cmd, ucore, min(sizeof(cmd), ucore->inlen));
2081 	if (ret)
2082 		return ret;
2083 
2084 	if (cmd.base.attr_mask &
2085 	    ~((IB_USER_LAST_QP_ATTR_MASK << 1) - 1))
2086 		return -EOPNOTSUPP;
2087 
2088 	if (ucore->inlen > sizeof(cmd)) {
2089 		if (!ib_is_udata_cleared(ucore, sizeof(cmd),
2090 					 ucore->inlen - sizeof(cmd)))
2091 			return -EOPNOTSUPP;
2092 	}
2093 
2094 	ret = modify_qp(file, &cmd, uhw);
2095 
2096 	return ret;
2097 }
2098 
2099 ssize_t ib_uverbs_destroy_qp(struct ib_uverbs_file *file,
2100 			     struct ib_device *ib_dev,
2101 			     const char __user *buf, int in_len,
2102 			     int out_len)
2103 {
2104 	struct ib_uverbs_destroy_qp      cmd;
2105 	struct ib_uverbs_destroy_qp_resp resp;
2106 	struct ib_uobject		*uobj;
2107 	struct ib_uqp_object        	*obj;
2108 	int                        	 ret = -EINVAL;
2109 
2110 	if (copy_from_user(&cmd, buf, sizeof cmd))
2111 		return -EFAULT;
2112 
2113 	memset(&resp, 0, sizeof resp);
2114 
2115 	uobj  = uobj_get_write(uobj_get_type(qp), cmd.qp_handle,
2116 			       file->ucontext);
2117 	if (IS_ERR(uobj))
2118 		return PTR_ERR(uobj);
2119 
2120 	obj = container_of(uobj, struct ib_uqp_object, uevent.uobject);
2121 	/*
2122 	 * Make sure we don't free the memory in remove_commit as we still
2123 	 * needs the uobject memory to create the response.
2124 	 */
2125 	uverbs_uobject_get(uobj);
2126 
2127 	ret = uobj_remove_commit(uobj);
2128 	if (ret) {
2129 		uverbs_uobject_put(uobj);
2130 		return ret;
2131 	}
2132 
2133 	resp.events_reported = obj->uevent.events_reported;
2134 	uverbs_uobject_put(uobj);
2135 
2136 	if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof resp))
2137 		return -EFAULT;
2138 
2139 	return in_len;
2140 }
2141 
2142 static void *alloc_wr(size_t wr_size, __u32 num_sge)
2143 {
2144 	if (num_sge >= (U32_MAX - ALIGN(wr_size, sizeof (struct ib_sge))) /
2145 		       sizeof (struct ib_sge))
2146 		return NULL;
2147 
2148 	return kmalloc(ALIGN(wr_size, sizeof (struct ib_sge)) +
2149 			 num_sge * sizeof (struct ib_sge), GFP_KERNEL);
2150 }
2151 
2152 ssize_t ib_uverbs_post_send(struct ib_uverbs_file *file,
2153 			    struct ib_device *ib_dev,
2154 			    const char __user *buf, int in_len,
2155 			    int out_len)
2156 {
2157 	struct ib_uverbs_post_send      cmd;
2158 	struct ib_uverbs_post_send_resp resp;
2159 	struct ib_uverbs_send_wr       *user_wr;
2160 	struct ib_send_wr              *wr = NULL, *last, *next, *bad_wr;
2161 	struct ib_qp                   *qp;
2162 	int                             i, sg_ind;
2163 	int				is_ud;
2164 	ssize_t                         ret = -EINVAL;
2165 	size_t                          next_size;
2166 
2167 	if (copy_from_user(&cmd, buf, sizeof cmd))
2168 		return -EFAULT;
2169 
2170 	if (in_len < sizeof cmd + cmd.wqe_size * cmd.wr_count +
2171 	    cmd.sge_count * sizeof (struct ib_uverbs_sge))
2172 		return -EINVAL;
2173 
2174 	if (cmd.wqe_size < sizeof (struct ib_uverbs_send_wr))
2175 		return -EINVAL;
2176 
2177 	user_wr = kmalloc(cmd.wqe_size, GFP_KERNEL);
2178 	if (!user_wr)
2179 		return -ENOMEM;
2180 
2181 	qp = uobj_get_obj_read(qp, cmd.qp_handle, file->ucontext);
2182 	if (!qp)
2183 		goto out;
2184 
2185 	is_ud = qp->qp_type == IB_QPT_UD;
2186 	sg_ind = 0;
2187 	last = NULL;
2188 	for (i = 0; i < cmd.wr_count; ++i) {
2189 		if (copy_from_user(user_wr,
2190 				   buf + sizeof cmd + i * cmd.wqe_size,
2191 				   cmd.wqe_size)) {
2192 			ret = -EFAULT;
2193 			goto out_put;
2194 		}
2195 
2196 		if (user_wr->num_sge + sg_ind > cmd.sge_count) {
2197 			ret = -EINVAL;
2198 			goto out_put;
2199 		}
2200 
2201 		if (is_ud) {
2202 			struct ib_ud_wr *ud;
2203 
2204 			if (user_wr->opcode != IB_WR_SEND &&
2205 			    user_wr->opcode != IB_WR_SEND_WITH_IMM) {
2206 				ret = -EINVAL;
2207 				goto out_put;
2208 			}
2209 
2210 			next_size = sizeof(*ud);
2211 			ud = alloc_wr(next_size, user_wr->num_sge);
2212 			if (!ud) {
2213 				ret = -ENOMEM;
2214 				goto out_put;
2215 			}
2216 
2217 			ud->ah = uobj_get_obj_read(ah, user_wr->wr.ud.ah,
2218 						   file->ucontext);
2219 			if (!ud->ah) {
2220 				kfree(ud);
2221 				ret = -EINVAL;
2222 				goto out_put;
2223 			}
2224 			ud->remote_qpn = user_wr->wr.ud.remote_qpn;
2225 			ud->remote_qkey = user_wr->wr.ud.remote_qkey;
2226 
2227 			next = &ud->wr;
2228 		} else if (user_wr->opcode == IB_WR_RDMA_WRITE_WITH_IMM ||
2229 			   user_wr->opcode == IB_WR_RDMA_WRITE ||
2230 			   user_wr->opcode == IB_WR_RDMA_READ) {
2231 			struct ib_rdma_wr *rdma;
2232 
2233 			next_size = sizeof(*rdma);
2234 			rdma = alloc_wr(next_size, user_wr->num_sge);
2235 			if (!rdma) {
2236 				ret = -ENOMEM;
2237 				goto out_put;
2238 			}
2239 
2240 			rdma->remote_addr = user_wr->wr.rdma.remote_addr;
2241 			rdma->rkey = user_wr->wr.rdma.rkey;
2242 
2243 			next = &rdma->wr;
2244 		} else if (user_wr->opcode == IB_WR_ATOMIC_CMP_AND_SWP ||
2245 			   user_wr->opcode == IB_WR_ATOMIC_FETCH_AND_ADD) {
2246 			struct ib_atomic_wr *atomic;
2247 
2248 			next_size = sizeof(*atomic);
2249 			atomic = alloc_wr(next_size, user_wr->num_sge);
2250 			if (!atomic) {
2251 				ret = -ENOMEM;
2252 				goto out_put;
2253 			}
2254 
2255 			atomic->remote_addr = user_wr->wr.atomic.remote_addr;
2256 			atomic->compare_add = user_wr->wr.atomic.compare_add;
2257 			atomic->swap = user_wr->wr.atomic.swap;
2258 			atomic->rkey = user_wr->wr.atomic.rkey;
2259 
2260 			next = &atomic->wr;
2261 		} else if (user_wr->opcode == IB_WR_SEND ||
2262 			   user_wr->opcode == IB_WR_SEND_WITH_IMM ||
2263 			   user_wr->opcode == IB_WR_SEND_WITH_INV) {
2264 			next_size = sizeof(*next);
2265 			next = alloc_wr(next_size, user_wr->num_sge);
2266 			if (!next) {
2267 				ret = -ENOMEM;
2268 				goto out_put;
2269 			}
2270 		} else {
2271 			ret = -EINVAL;
2272 			goto out_put;
2273 		}
2274 
2275 		if (user_wr->opcode == IB_WR_SEND_WITH_IMM ||
2276 		    user_wr->opcode == IB_WR_RDMA_WRITE_WITH_IMM) {
2277 			next->ex.imm_data =
2278 					(__be32 __force) user_wr->ex.imm_data;
2279 		} else if (user_wr->opcode == IB_WR_SEND_WITH_INV) {
2280 			next->ex.invalidate_rkey = user_wr->ex.invalidate_rkey;
2281 		}
2282 
2283 		if (!last)
2284 			wr = next;
2285 		else
2286 			last->next = next;
2287 		last = next;
2288 
2289 		next->next       = NULL;
2290 		next->wr_id      = user_wr->wr_id;
2291 		next->num_sge    = user_wr->num_sge;
2292 		next->opcode     = user_wr->opcode;
2293 		next->send_flags = user_wr->send_flags;
2294 
2295 		if (next->num_sge) {
2296 			next->sg_list = (void *) next +
2297 				ALIGN(next_size, sizeof(struct ib_sge));
2298 			if (copy_from_user(next->sg_list,
2299 					   buf + sizeof cmd +
2300 					   cmd.wr_count * cmd.wqe_size +
2301 					   sg_ind * sizeof (struct ib_sge),
2302 					   next->num_sge * sizeof (struct ib_sge))) {
2303 				ret = -EFAULT;
2304 				goto out_put;
2305 			}
2306 			sg_ind += next->num_sge;
2307 		} else
2308 			next->sg_list = NULL;
2309 	}
2310 
2311 	resp.bad_wr = 0;
2312 	ret = qp->device->post_send(qp->real_qp, wr, &bad_wr);
2313 	if (ret)
2314 		for (next = wr; next; next = next->next) {
2315 			++resp.bad_wr;
2316 			if (next == bad_wr)
2317 				break;
2318 		}
2319 
2320 	if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof resp))
2321 		ret = -EFAULT;
2322 
2323 out_put:
2324 	uobj_put_obj_read(qp);
2325 
2326 	while (wr) {
2327 		if (is_ud && ud_wr(wr)->ah)
2328 			uobj_put_obj_read(ud_wr(wr)->ah);
2329 		next = wr->next;
2330 		kfree(wr);
2331 		wr = next;
2332 	}
2333 
2334 out:
2335 	kfree(user_wr);
2336 
2337 	return ret ? ret : in_len;
2338 }
2339 
2340 static struct ib_recv_wr *ib_uverbs_unmarshall_recv(const char __user *buf,
2341 						    int in_len,
2342 						    u32 wr_count,
2343 						    u32 sge_count,
2344 						    u32 wqe_size)
2345 {
2346 	struct ib_uverbs_recv_wr *user_wr;
2347 	struct ib_recv_wr        *wr = NULL, *last, *next;
2348 	int                       sg_ind;
2349 	int                       i;
2350 	int                       ret;
2351 
2352 	if (in_len < wqe_size * wr_count +
2353 	    sge_count * sizeof (struct ib_uverbs_sge))
2354 		return ERR_PTR(-EINVAL);
2355 
2356 	if (wqe_size < sizeof (struct ib_uverbs_recv_wr))
2357 		return ERR_PTR(-EINVAL);
2358 
2359 	user_wr = kmalloc(wqe_size, GFP_KERNEL);
2360 	if (!user_wr)
2361 		return ERR_PTR(-ENOMEM);
2362 
2363 	sg_ind = 0;
2364 	last = NULL;
2365 	for (i = 0; i < wr_count; ++i) {
2366 		if (copy_from_user(user_wr, buf + i * wqe_size,
2367 				   wqe_size)) {
2368 			ret = -EFAULT;
2369 			goto err;
2370 		}
2371 
2372 		if (user_wr->num_sge + sg_ind > sge_count) {
2373 			ret = -EINVAL;
2374 			goto err;
2375 		}
2376 
2377 		if (user_wr->num_sge >=
2378 		    (U32_MAX - ALIGN(sizeof *next, sizeof (struct ib_sge))) /
2379 		    sizeof (struct ib_sge)) {
2380 			ret = -EINVAL;
2381 			goto err;
2382 		}
2383 
2384 		next = kmalloc(ALIGN(sizeof *next, sizeof (struct ib_sge)) +
2385 			       user_wr->num_sge * sizeof (struct ib_sge),
2386 			       GFP_KERNEL);
2387 		if (!next) {
2388 			ret = -ENOMEM;
2389 			goto err;
2390 		}
2391 
2392 		if (!last)
2393 			wr = next;
2394 		else
2395 			last->next = next;
2396 		last = next;
2397 
2398 		next->next       = NULL;
2399 		next->wr_id      = user_wr->wr_id;
2400 		next->num_sge    = user_wr->num_sge;
2401 
2402 		if (next->num_sge) {
2403 			next->sg_list = (void *) next +
2404 				ALIGN(sizeof *next, sizeof (struct ib_sge));
2405 			if (copy_from_user(next->sg_list,
2406 					   buf + wr_count * wqe_size +
2407 					   sg_ind * sizeof (struct ib_sge),
2408 					   next->num_sge * sizeof (struct ib_sge))) {
2409 				ret = -EFAULT;
2410 				goto err;
2411 			}
2412 			sg_ind += next->num_sge;
2413 		} else
2414 			next->sg_list = NULL;
2415 	}
2416 
2417 	kfree(user_wr);
2418 	return wr;
2419 
2420 err:
2421 	kfree(user_wr);
2422 
2423 	while (wr) {
2424 		next = wr->next;
2425 		kfree(wr);
2426 		wr = next;
2427 	}
2428 
2429 	return ERR_PTR(ret);
2430 }
2431 
2432 ssize_t ib_uverbs_post_recv(struct ib_uverbs_file *file,
2433 			    struct ib_device *ib_dev,
2434 			    const char __user *buf, int in_len,
2435 			    int out_len)
2436 {
2437 	struct ib_uverbs_post_recv      cmd;
2438 	struct ib_uverbs_post_recv_resp resp;
2439 	struct ib_recv_wr              *wr, *next, *bad_wr;
2440 	struct ib_qp                   *qp;
2441 	ssize_t                         ret = -EINVAL;
2442 
2443 	if (copy_from_user(&cmd, buf, sizeof cmd))
2444 		return -EFAULT;
2445 
2446 	wr = ib_uverbs_unmarshall_recv(buf + sizeof cmd,
2447 				       in_len - sizeof cmd, cmd.wr_count,
2448 				       cmd.sge_count, cmd.wqe_size);
2449 	if (IS_ERR(wr))
2450 		return PTR_ERR(wr);
2451 
2452 	qp = uobj_get_obj_read(qp, cmd.qp_handle, file->ucontext);
2453 	if (!qp)
2454 		goto out;
2455 
2456 	resp.bad_wr = 0;
2457 	ret = qp->device->post_recv(qp->real_qp, wr, &bad_wr);
2458 
2459 	uobj_put_obj_read(qp);
2460 	if (ret) {
2461 		for (next = wr; next; next = next->next) {
2462 			++resp.bad_wr;
2463 			if (next == bad_wr)
2464 				break;
2465 		}
2466 	}
2467 
2468 	if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof resp))
2469 		ret = -EFAULT;
2470 
2471 out:
2472 	while (wr) {
2473 		next = wr->next;
2474 		kfree(wr);
2475 		wr = next;
2476 	}
2477 
2478 	return ret ? ret : in_len;
2479 }
2480 
2481 ssize_t ib_uverbs_post_srq_recv(struct ib_uverbs_file *file,
2482 				struct ib_device *ib_dev,
2483 				const char __user *buf, int in_len,
2484 				int out_len)
2485 {
2486 	struct ib_uverbs_post_srq_recv      cmd;
2487 	struct ib_uverbs_post_srq_recv_resp resp;
2488 	struct ib_recv_wr                  *wr, *next, *bad_wr;
2489 	struct ib_srq                      *srq;
2490 	ssize_t                             ret = -EINVAL;
2491 
2492 	if (copy_from_user(&cmd, buf, sizeof cmd))
2493 		return -EFAULT;
2494 
2495 	wr = ib_uverbs_unmarshall_recv(buf + sizeof cmd,
2496 				       in_len - sizeof cmd, cmd.wr_count,
2497 				       cmd.sge_count, cmd.wqe_size);
2498 	if (IS_ERR(wr))
2499 		return PTR_ERR(wr);
2500 
2501 	srq = uobj_get_obj_read(srq, cmd.srq_handle, file->ucontext);
2502 	if (!srq)
2503 		goto out;
2504 
2505 	resp.bad_wr = 0;
2506 	ret = srq->device->post_srq_recv(srq, wr, &bad_wr);
2507 
2508 	uobj_put_obj_read(srq);
2509 
2510 	if (ret)
2511 		for (next = wr; next; next = next->next) {
2512 			++resp.bad_wr;
2513 			if (next == bad_wr)
2514 				break;
2515 		}
2516 
2517 	if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof resp))
2518 		ret = -EFAULT;
2519 
2520 out:
2521 	while (wr) {
2522 		next = wr->next;
2523 		kfree(wr);
2524 		wr = next;
2525 	}
2526 
2527 	return ret ? ret : in_len;
2528 }
2529 
2530 ssize_t ib_uverbs_create_ah(struct ib_uverbs_file *file,
2531 			    struct ib_device *ib_dev,
2532 			    const char __user *buf, int in_len,
2533 			    int out_len)
2534 {
2535 	struct ib_uverbs_create_ah	 cmd;
2536 	struct ib_uverbs_create_ah_resp	 resp;
2537 	struct ib_uobject		*uobj;
2538 	struct ib_pd			*pd;
2539 	struct ib_ah			*ah;
2540 	struct rdma_ah_attr		attr;
2541 	int ret;
2542 	struct ib_udata                   udata;
2543 
2544 	if (out_len < sizeof resp)
2545 		return -ENOSPC;
2546 
2547 	if (copy_from_user(&cmd, buf, sizeof cmd))
2548 		return -EFAULT;
2549 
2550 	if (!rdma_is_port_valid(ib_dev, cmd.attr.port_num))
2551 		return -EINVAL;
2552 
2553 	ib_uverbs_init_udata(&udata, buf + sizeof(cmd),
2554 		   u64_to_user_ptr(cmd.response) + sizeof(resp),
2555 		   in_len - sizeof(cmd) - sizeof(struct ib_uverbs_cmd_hdr),
2556 		   out_len - sizeof(resp));
2557 
2558 	uobj  = uobj_alloc(uobj_get_type(ah), file->ucontext);
2559 	if (IS_ERR(uobj))
2560 		return PTR_ERR(uobj);
2561 
2562 	pd = uobj_get_obj_read(pd, cmd.pd_handle, file->ucontext);
2563 	if (!pd) {
2564 		ret = -EINVAL;
2565 		goto err;
2566 	}
2567 
2568 	attr.type = rdma_ah_find_type(ib_dev, cmd.attr.port_num);
2569 	rdma_ah_set_make_grd(&attr, false);
2570 	rdma_ah_set_dlid(&attr, cmd.attr.dlid);
2571 	rdma_ah_set_sl(&attr, cmd.attr.sl);
2572 	rdma_ah_set_path_bits(&attr, cmd.attr.src_path_bits);
2573 	rdma_ah_set_static_rate(&attr, cmd.attr.static_rate);
2574 	rdma_ah_set_port_num(&attr, cmd.attr.port_num);
2575 
2576 	if (cmd.attr.is_global) {
2577 		rdma_ah_set_grh(&attr, NULL, cmd.attr.grh.flow_label,
2578 				cmd.attr.grh.sgid_index,
2579 				cmd.attr.grh.hop_limit,
2580 				cmd.attr.grh.traffic_class);
2581 		rdma_ah_set_dgid_raw(&attr, cmd.attr.grh.dgid);
2582 	} else {
2583 		rdma_ah_set_ah_flags(&attr, 0);
2584 	}
2585 
2586 	ah = rdma_create_user_ah(pd, &attr, &udata);
2587 	if (IS_ERR(ah)) {
2588 		ret = PTR_ERR(ah);
2589 		goto err_put;
2590 	}
2591 
2592 	ah->uobject  = uobj;
2593 	uobj->user_handle = cmd.user_handle;
2594 	uobj->object = ah;
2595 
2596 	resp.ah_handle = uobj->id;
2597 
2598 	if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof resp)) {
2599 		ret = -EFAULT;
2600 		goto err_copy;
2601 	}
2602 
2603 	uobj_put_obj_read(pd);
2604 	uobj_alloc_commit(uobj);
2605 
2606 	return in_len;
2607 
2608 err_copy:
2609 	rdma_destroy_ah(ah);
2610 
2611 err_put:
2612 	uobj_put_obj_read(pd);
2613 
2614 err:
2615 	uobj_alloc_abort(uobj);
2616 	return ret;
2617 }
2618 
2619 ssize_t ib_uverbs_destroy_ah(struct ib_uverbs_file *file,
2620 			     struct ib_device *ib_dev,
2621 			     const char __user *buf, int in_len, int out_len)
2622 {
2623 	struct ib_uverbs_destroy_ah cmd;
2624 	struct ib_uobject	   *uobj;
2625 	int			    ret;
2626 
2627 	if (copy_from_user(&cmd, buf, sizeof cmd))
2628 		return -EFAULT;
2629 
2630 	uobj  = uobj_get_write(uobj_get_type(ah), cmd.ah_handle,
2631 			       file->ucontext);
2632 	if (IS_ERR(uobj))
2633 		return PTR_ERR(uobj);
2634 
2635 	ret = uobj_remove_commit(uobj);
2636 	return ret ?: in_len;
2637 }
2638 
2639 ssize_t ib_uverbs_attach_mcast(struct ib_uverbs_file *file,
2640 			       struct ib_device *ib_dev,
2641 			       const char __user *buf, int in_len,
2642 			       int out_len)
2643 {
2644 	struct ib_uverbs_attach_mcast cmd;
2645 	struct ib_qp                 *qp;
2646 	struct ib_uqp_object         *obj;
2647 	struct ib_uverbs_mcast_entry *mcast;
2648 	int                           ret;
2649 
2650 	if (copy_from_user(&cmd, buf, sizeof cmd))
2651 		return -EFAULT;
2652 
2653 	qp = uobj_get_obj_read(qp, cmd.qp_handle, file->ucontext);
2654 	if (!qp)
2655 		return -EINVAL;
2656 
2657 	obj = container_of(qp->uobject, struct ib_uqp_object, uevent.uobject);
2658 
2659 	mutex_lock(&obj->mcast_lock);
2660 	list_for_each_entry(mcast, &obj->mcast_list, list)
2661 		if (cmd.mlid == mcast->lid &&
2662 		    !memcmp(cmd.gid, mcast->gid.raw, sizeof mcast->gid.raw)) {
2663 			ret = 0;
2664 			goto out_put;
2665 		}
2666 
2667 	mcast = kmalloc(sizeof *mcast, GFP_KERNEL);
2668 	if (!mcast) {
2669 		ret = -ENOMEM;
2670 		goto out_put;
2671 	}
2672 
2673 	mcast->lid = cmd.mlid;
2674 	memcpy(mcast->gid.raw, cmd.gid, sizeof mcast->gid.raw);
2675 
2676 	ret = ib_attach_mcast(qp, &mcast->gid, cmd.mlid);
2677 	if (!ret)
2678 		list_add_tail(&mcast->list, &obj->mcast_list);
2679 	else
2680 		kfree(mcast);
2681 
2682 out_put:
2683 	mutex_unlock(&obj->mcast_lock);
2684 	uobj_put_obj_read(qp);
2685 
2686 	return ret ? ret : in_len;
2687 }
2688 
2689 ssize_t ib_uverbs_detach_mcast(struct ib_uverbs_file *file,
2690 			       struct ib_device *ib_dev,
2691 			       const char __user *buf, int in_len,
2692 			       int out_len)
2693 {
2694 	struct ib_uverbs_detach_mcast cmd;
2695 	struct ib_uqp_object         *obj;
2696 	struct ib_qp                 *qp;
2697 	struct ib_uverbs_mcast_entry *mcast;
2698 	int                           ret = -EINVAL;
2699 	bool                          found = false;
2700 
2701 	if (copy_from_user(&cmd, buf, sizeof cmd))
2702 		return -EFAULT;
2703 
2704 	qp = uobj_get_obj_read(qp, cmd.qp_handle, file->ucontext);
2705 	if (!qp)
2706 		return -EINVAL;
2707 
2708 	obj = container_of(qp->uobject, struct ib_uqp_object, uevent.uobject);
2709 	mutex_lock(&obj->mcast_lock);
2710 
2711 	list_for_each_entry(mcast, &obj->mcast_list, list)
2712 		if (cmd.mlid == mcast->lid &&
2713 		    !memcmp(cmd.gid, mcast->gid.raw, sizeof mcast->gid.raw)) {
2714 			list_del(&mcast->list);
2715 			kfree(mcast);
2716 			found = true;
2717 			break;
2718 		}
2719 
2720 	if (!found) {
2721 		ret = -EINVAL;
2722 		goto out_put;
2723 	}
2724 
2725 	ret = ib_detach_mcast(qp, (union ib_gid *)cmd.gid, cmd.mlid);
2726 
2727 out_put:
2728 	mutex_unlock(&obj->mcast_lock);
2729 	uobj_put_obj_read(qp);
2730 	return ret ? ret : in_len;
2731 }
2732 
2733 static int kern_spec_to_ib_spec_action(struct ib_uverbs_flow_spec *kern_spec,
2734 				       union ib_flow_spec *ib_spec)
2735 {
2736 	ib_spec->type = kern_spec->type;
2737 	switch (ib_spec->type) {
2738 	case IB_FLOW_SPEC_ACTION_TAG:
2739 		if (kern_spec->flow_tag.size !=
2740 		    sizeof(struct ib_uverbs_flow_spec_action_tag))
2741 			return -EINVAL;
2742 
2743 		ib_spec->flow_tag.size = sizeof(struct ib_flow_spec_action_tag);
2744 		ib_spec->flow_tag.tag_id = kern_spec->flow_tag.tag_id;
2745 		break;
2746 	case IB_FLOW_SPEC_ACTION_DROP:
2747 		if (kern_spec->drop.size !=
2748 		    sizeof(struct ib_uverbs_flow_spec_action_drop))
2749 			return -EINVAL;
2750 
2751 		ib_spec->drop.size = sizeof(struct ib_flow_spec_action_drop);
2752 		break;
2753 	default:
2754 		return -EINVAL;
2755 	}
2756 	return 0;
2757 }
2758 
2759 static size_t kern_spec_filter_sz(struct ib_uverbs_flow_spec_hdr *spec)
2760 {
2761 	/* Returns user space filter size, includes padding */
2762 	return (spec->size - sizeof(struct ib_uverbs_flow_spec_hdr)) / 2;
2763 }
2764 
2765 static ssize_t spec_filter_size(void *kern_spec_filter, u16 kern_filter_size,
2766 				u16 ib_real_filter_sz)
2767 {
2768 	/*
2769 	 * User space filter structures must be 64 bit aligned, otherwise this
2770 	 * may pass, but we won't handle additional new attributes.
2771 	 */
2772 
2773 	if (kern_filter_size > ib_real_filter_sz) {
2774 		if (memchr_inv(kern_spec_filter +
2775 			       ib_real_filter_sz, 0,
2776 			       kern_filter_size - ib_real_filter_sz))
2777 			return -EINVAL;
2778 		return ib_real_filter_sz;
2779 	}
2780 	return kern_filter_size;
2781 }
2782 
2783 static int kern_spec_to_ib_spec_filter(struct ib_uverbs_flow_spec *kern_spec,
2784 				       union ib_flow_spec *ib_spec)
2785 {
2786 	ssize_t actual_filter_sz;
2787 	ssize_t kern_filter_sz;
2788 	ssize_t ib_filter_sz;
2789 	void *kern_spec_mask;
2790 	void *kern_spec_val;
2791 
2792 	if (kern_spec->reserved)
2793 		return -EINVAL;
2794 
2795 	ib_spec->type = kern_spec->type;
2796 
2797 	kern_filter_sz = kern_spec_filter_sz(&kern_spec->hdr);
2798 	/* User flow spec size must be aligned to 4 bytes */
2799 	if (kern_filter_sz != ALIGN(kern_filter_sz, 4))
2800 		return -EINVAL;
2801 
2802 	kern_spec_val = (void *)kern_spec +
2803 		sizeof(struct ib_uverbs_flow_spec_hdr);
2804 	kern_spec_mask = kern_spec_val + kern_filter_sz;
2805 	if (ib_spec->type == (IB_FLOW_SPEC_INNER | IB_FLOW_SPEC_VXLAN_TUNNEL))
2806 		return -EINVAL;
2807 
2808 	switch (ib_spec->type & ~IB_FLOW_SPEC_INNER) {
2809 	case IB_FLOW_SPEC_ETH:
2810 		ib_filter_sz = offsetof(struct ib_flow_eth_filter, real_sz);
2811 		actual_filter_sz = spec_filter_size(kern_spec_mask,
2812 						    kern_filter_sz,
2813 						    ib_filter_sz);
2814 		if (actual_filter_sz <= 0)
2815 			return -EINVAL;
2816 		ib_spec->size = sizeof(struct ib_flow_spec_eth);
2817 		memcpy(&ib_spec->eth.val, kern_spec_val, actual_filter_sz);
2818 		memcpy(&ib_spec->eth.mask, kern_spec_mask, actual_filter_sz);
2819 		break;
2820 	case IB_FLOW_SPEC_IPV4:
2821 		ib_filter_sz = offsetof(struct ib_flow_ipv4_filter, real_sz);
2822 		actual_filter_sz = spec_filter_size(kern_spec_mask,
2823 						    kern_filter_sz,
2824 						    ib_filter_sz);
2825 		if (actual_filter_sz <= 0)
2826 			return -EINVAL;
2827 		ib_spec->size = sizeof(struct ib_flow_spec_ipv4);
2828 		memcpy(&ib_spec->ipv4.val, kern_spec_val, actual_filter_sz);
2829 		memcpy(&ib_spec->ipv4.mask, kern_spec_mask, actual_filter_sz);
2830 		break;
2831 	case IB_FLOW_SPEC_IPV6:
2832 		ib_filter_sz = offsetof(struct ib_flow_ipv6_filter, real_sz);
2833 		actual_filter_sz = spec_filter_size(kern_spec_mask,
2834 						    kern_filter_sz,
2835 						    ib_filter_sz);
2836 		if (actual_filter_sz <= 0)
2837 			return -EINVAL;
2838 		ib_spec->size = sizeof(struct ib_flow_spec_ipv6);
2839 		memcpy(&ib_spec->ipv6.val, kern_spec_val, actual_filter_sz);
2840 		memcpy(&ib_spec->ipv6.mask, kern_spec_mask, actual_filter_sz);
2841 
2842 		if ((ntohl(ib_spec->ipv6.mask.flow_label)) >= BIT(20) ||
2843 		    (ntohl(ib_spec->ipv6.val.flow_label)) >= BIT(20))
2844 			return -EINVAL;
2845 		break;
2846 	case IB_FLOW_SPEC_TCP:
2847 	case IB_FLOW_SPEC_UDP:
2848 		ib_filter_sz = offsetof(struct ib_flow_tcp_udp_filter, real_sz);
2849 		actual_filter_sz = spec_filter_size(kern_spec_mask,
2850 						    kern_filter_sz,
2851 						    ib_filter_sz);
2852 		if (actual_filter_sz <= 0)
2853 			return -EINVAL;
2854 		ib_spec->size = sizeof(struct ib_flow_spec_tcp_udp);
2855 		memcpy(&ib_spec->tcp_udp.val, kern_spec_val, actual_filter_sz);
2856 		memcpy(&ib_spec->tcp_udp.mask, kern_spec_mask, actual_filter_sz);
2857 		break;
2858 	case IB_FLOW_SPEC_VXLAN_TUNNEL:
2859 		ib_filter_sz = offsetof(struct ib_flow_tunnel_filter, real_sz);
2860 		actual_filter_sz = spec_filter_size(kern_spec_mask,
2861 						    kern_filter_sz,
2862 						    ib_filter_sz);
2863 		if (actual_filter_sz <= 0)
2864 			return -EINVAL;
2865 		ib_spec->tunnel.size = sizeof(struct ib_flow_spec_tunnel);
2866 		memcpy(&ib_spec->tunnel.val, kern_spec_val, actual_filter_sz);
2867 		memcpy(&ib_spec->tunnel.mask, kern_spec_mask, actual_filter_sz);
2868 
2869 		if ((ntohl(ib_spec->tunnel.mask.tunnel_id)) >= BIT(24) ||
2870 		    (ntohl(ib_spec->tunnel.val.tunnel_id)) >= BIT(24))
2871 			return -EINVAL;
2872 		break;
2873 	default:
2874 		return -EINVAL;
2875 	}
2876 	return 0;
2877 }
2878 
2879 static int kern_spec_to_ib_spec(struct ib_uverbs_flow_spec *kern_spec,
2880 				union ib_flow_spec *ib_spec)
2881 {
2882 	if (kern_spec->reserved)
2883 		return -EINVAL;
2884 
2885 	if (kern_spec->type >= IB_FLOW_SPEC_ACTION_TAG)
2886 		return kern_spec_to_ib_spec_action(kern_spec, ib_spec);
2887 	else
2888 		return kern_spec_to_ib_spec_filter(kern_spec, ib_spec);
2889 }
2890 
2891 int ib_uverbs_ex_create_wq(struct ib_uverbs_file *file,
2892 			   struct ib_device *ib_dev,
2893 			   struct ib_udata *ucore,
2894 			   struct ib_udata *uhw)
2895 {
2896 	struct ib_uverbs_ex_create_wq	  cmd = {};
2897 	struct ib_uverbs_ex_create_wq_resp resp = {};
2898 	struct ib_uwq_object           *obj;
2899 	int err = 0;
2900 	struct ib_cq *cq;
2901 	struct ib_pd *pd;
2902 	struct ib_wq *wq;
2903 	struct ib_wq_init_attr wq_init_attr = {};
2904 	size_t required_cmd_sz;
2905 	size_t required_resp_len;
2906 
2907 	required_cmd_sz = offsetof(typeof(cmd), max_sge) + sizeof(cmd.max_sge);
2908 	required_resp_len = offsetof(typeof(resp), wqn) + sizeof(resp.wqn);
2909 
2910 	if (ucore->inlen < required_cmd_sz)
2911 		return -EINVAL;
2912 
2913 	if (ucore->outlen < required_resp_len)
2914 		return -ENOSPC;
2915 
2916 	if (ucore->inlen > sizeof(cmd) &&
2917 	    !ib_is_udata_cleared(ucore, sizeof(cmd),
2918 				 ucore->inlen - sizeof(cmd)))
2919 		return -EOPNOTSUPP;
2920 
2921 	err = ib_copy_from_udata(&cmd, ucore, min(sizeof(cmd), ucore->inlen));
2922 	if (err)
2923 		return err;
2924 
2925 	if (cmd.comp_mask)
2926 		return -EOPNOTSUPP;
2927 
2928 	obj  = (struct ib_uwq_object *)uobj_alloc(uobj_get_type(wq),
2929 						  file->ucontext);
2930 	if (IS_ERR(obj))
2931 		return PTR_ERR(obj);
2932 
2933 	pd  = uobj_get_obj_read(pd, cmd.pd_handle, file->ucontext);
2934 	if (!pd) {
2935 		err = -EINVAL;
2936 		goto err_uobj;
2937 	}
2938 
2939 	cq = uobj_get_obj_read(cq, cmd.cq_handle, file->ucontext);
2940 	if (!cq) {
2941 		err = -EINVAL;
2942 		goto err_put_pd;
2943 	}
2944 
2945 	wq_init_attr.cq = cq;
2946 	wq_init_attr.max_sge = cmd.max_sge;
2947 	wq_init_attr.max_wr = cmd.max_wr;
2948 	wq_init_attr.wq_context = file;
2949 	wq_init_attr.wq_type = cmd.wq_type;
2950 	wq_init_attr.event_handler = ib_uverbs_wq_event_handler;
2951 	if (ucore->inlen >= (offsetof(typeof(cmd), create_flags) +
2952 			     sizeof(cmd.create_flags)))
2953 		wq_init_attr.create_flags = cmd.create_flags;
2954 	obj->uevent.events_reported = 0;
2955 	INIT_LIST_HEAD(&obj->uevent.event_list);
2956 
2957 	if (!pd->device->create_wq) {
2958 		err = -EOPNOTSUPP;
2959 		goto err_put_cq;
2960 	}
2961 	wq = pd->device->create_wq(pd, &wq_init_attr, uhw);
2962 	if (IS_ERR(wq)) {
2963 		err = PTR_ERR(wq);
2964 		goto err_put_cq;
2965 	}
2966 
2967 	wq->uobject = &obj->uevent.uobject;
2968 	obj->uevent.uobject.object = wq;
2969 	wq->wq_type = wq_init_attr.wq_type;
2970 	wq->cq = cq;
2971 	wq->pd = pd;
2972 	wq->device = pd->device;
2973 	wq->wq_context = wq_init_attr.wq_context;
2974 	atomic_set(&wq->usecnt, 0);
2975 	atomic_inc(&pd->usecnt);
2976 	atomic_inc(&cq->usecnt);
2977 	wq->uobject = &obj->uevent.uobject;
2978 	obj->uevent.uobject.object = wq;
2979 
2980 	memset(&resp, 0, sizeof(resp));
2981 	resp.wq_handle = obj->uevent.uobject.id;
2982 	resp.max_sge = wq_init_attr.max_sge;
2983 	resp.max_wr = wq_init_attr.max_wr;
2984 	resp.wqn = wq->wq_num;
2985 	resp.response_length = required_resp_len;
2986 	err = ib_copy_to_udata(ucore,
2987 			       &resp, resp.response_length);
2988 	if (err)
2989 		goto err_copy;
2990 
2991 	uobj_put_obj_read(pd);
2992 	uobj_put_obj_read(cq);
2993 	uobj_alloc_commit(&obj->uevent.uobject);
2994 	return 0;
2995 
2996 err_copy:
2997 	ib_destroy_wq(wq);
2998 err_put_cq:
2999 	uobj_put_obj_read(cq);
3000 err_put_pd:
3001 	uobj_put_obj_read(pd);
3002 err_uobj:
3003 	uobj_alloc_abort(&obj->uevent.uobject);
3004 
3005 	return err;
3006 }
3007 
3008 int ib_uverbs_ex_destroy_wq(struct ib_uverbs_file *file,
3009 			    struct ib_device *ib_dev,
3010 			    struct ib_udata *ucore,
3011 			    struct ib_udata *uhw)
3012 {
3013 	struct ib_uverbs_ex_destroy_wq	cmd = {};
3014 	struct ib_uverbs_ex_destroy_wq_resp	resp = {};
3015 	struct ib_uobject		*uobj;
3016 	struct ib_uwq_object		*obj;
3017 	size_t required_cmd_sz;
3018 	size_t required_resp_len;
3019 	int				ret;
3020 
3021 	required_cmd_sz = offsetof(typeof(cmd), wq_handle) + sizeof(cmd.wq_handle);
3022 	required_resp_len = offsetof(typeof(resp), reserved) + sizeof(resp.reserved);
3023 
3024 	if (ucore->inlen < required_cmd_sz)
3025 		return -EINVAL;
3026 
3027 	if (ucore->outlen < required_resp_len)
3028 		return -ENOSPC;
3029 
3030 	if (ucore->inlen > sizeof(cmd) &&
3031 	    !ib_is_udata_cleared(ucore, sizeof(cmd),
3032 				 ucore->inlen - sizeof(cmd)))
3033 		return -EOPNOTSUPP;
3034 
3035 	ret = ib_copy_from_udata(&cmd, ucore, min(sizeof(cmd), ucore->inlen));
3036 	if (ret)
3037 		return ret;
3038 
3039 	if (cmd.comp_mask)
3040 		return -EOPNOTSUPP;
3041 
3042 	resp.response_length = required_resp_len;
3043 	uobj  = uobj_get_write(uobj_get_type(wq), cmd.wq_handle,
3044 			       file->ucontext);
3045 	if (IS_ERR(uobj))
3046 		return PTR_ERR(uobj);
3047 
3048 	obj = container_of(uobj, struct ib_uwq_object, uevent.uobject);
3049 	/*
3050 	 * Make sure we don't free the memory in remove_commit as we still
3051 	 * needs the uobject memory to create the response.
3052 	 */
3053 	uverbs_uobject_get(uobj);
3054 
3055 	ret = uobj_remove_commit(uobj);
3056 	resp.events_reported = obj->uevent.events_reported;
3057 	uverbs_uobject_put(uobj);
3058 	if (ret)
3059 		return ret;
3060 
3061 	return ib_copy_to_udata(ucore, &resp, resp.response_length);
3062 }
3063 
3064 int ib_uverbs_ex_modify_wq(struct ib_uverbs_file *file,
3065 			   struct ib_device *ib_dev,
3066 			   struct ib_udata *ucore,
3067 			   struct ib_udata *uhw)
3068 {
3069 	struct ib_uverbs_ex_modify_wq cmd = {};
3070 	struct ib_wq *wq;
3071 	struct ib_wq_attr wq_attr = {};
3072 	size_t required_cmd_sz;
3073 	int ret;
3074 
3075 	required_cmd_sz = offsetof(typeof(cmd), curr_wq_state) + sizeof(cmd.curr_wq_state);
3076 	if (ucore->inlen < required_cmd_sz)
3077 		return -EINVAL;
3078 
3079 	if (ucore->inlen > sizeof(cmd) &&
3080 	    !ib_is_udata_cleared(ucore, sizeof(cmd),
3081 				 ucore->inlen - sizeof(cmd)))
3082 		return -EOPNOTSUPP;
3083 
3084 	ret = ib_copy_from_udata(&cmd, ucore, min(sizeof(cmd), ucore->inlen));
3085 	if (ret)
3086 		return ret;
3087 
3088 	if (!cmd.attr_mask)
3089 		return -EINVAL;
3090 
3091 	if (cmd.attr_mask > (IB_WQ_STATE | IB_WQ_CUR_STATE | IB_WQ_FLAGS))
3092 		return -EINVAL;
3093 
3094 	wq = uobj_get_obj_read(wq, cmd.wq_handle, file->ucontext);
3095 	if (!wq)
3096 		return -EINVAL;
3097 
3098 	wq_attr.curr_wq_state = cmd.curr_wq_state;
3099 	wq_attr.wq_state = cmd.wq_state;
3100 	if (cmd.attr_mask & IB_WQ_FLAGS) {
3101 		wq_attr.flags = cmd.flags;
3102 		wq_attr.flags_mask = cmd.flags_mask;
3103 	}
3104 	if (!wq->device->modify_wq) {
3105 		ret = -EOPNOTSUPP;
3106 		goto out;
3107 	}
3108 	ret = wq->device->modify_wq(wq, &wq_attr, cmd.attr_mask, uhw);
3109 out:
3110 	uobj_put_obj_read(wq);
3111 	return ret;
3112 }
3113 
3114 int ib_uverbs_ex_create_rwq_ind_table(struct ib_uverbs_file *file,
3115 				      struct ib_device *ib_dev,
3116 				      struct ib_udata *ucore,
3117 				      struct ib_udata *uhw)
3118 {
3119 	struct ib_uverbs_ex_create_rwq_ind_table	  cmd = {};
3120 	struct ib_uverbs_ex_create_rwq_ind_table_resp  resp = {};
3121 	struct ib_uobject		  *uobj;
3122 	int err = 0;
3123 	struct ib_rwq_ind_table_init_attr init_attr = {};
3124 	struct ib_rwq_ind_table *rwq_ind_tbl;
3125 	struct ib_wq	**wqs = NULL;
3126 	u32 *wqs_handles = NULL;
3127 	struct ib_wq	*wq = NULL;
3128 	int i, j, num_read_wqs;
3129 	u32 num_wq_handles;
3130 	u32 expected_in_size;
3131 	size_t required_cmd_sz_header;
3132 	size_t required_resp_len;
3133 
3134 	required_cmd_sz_header = offsetof(typeof(cmd), log_ind_tbl_size) + sizeof(cmd.log_ind_tbl_size);
3135 	required_resp_len = offsetof(typeof(resp), ind_tbl_num) + sizeof(resp.ind_tbl_num);
3136 
3137 	if (ucore->inlen < required_cmd_sz_header)
3138 		return -EINVAL;
3139 
3140 	if (ucore->outlen < required_resp_len)
3141 		return -ENOSPC;
3142 
3143 	err = ib_copy_from_udata(&cmd, ucore, required_cmd_sz_header);
3144 	if (err)
3145 		return err;
3146 
3147 	ucore->inbuf += required_cmd_sz_header;
3148 	ucore->inlen -= required_cmd_sz_header;
3149 
3150 	if (cmd.comp_mask)
3151 		return -EOPNOTSUPP;
3152 
3153 	if (cmd.log_ind_tbl_size > IB_USER_VERBS_MAX_LOG_IND_TBL_SIZE)
3154 		return -EINVAL;
3155 
3156 	num_wq_handles = 1 << cmd.log_ind_tbl_size;
3157 	expected_in_size = num_wq_handles * sizeof(__u32);
3158 	if (num_wq_handles == 1)
3159 		/* input size for wq handles is u64 aligned */
3160 		expected_in_size += sizeof(__u32);
3161 
3162 	if (ucore->inlen < expected_in_size)
3163 		return -EINVAL;
3164 
3165 	if (ucore->inlen > expected_in_size &&
3166 	    !ib_is_udata_cleared(ucore, expected_in_size,
3167 				 ucore->inlen - expected_in_size))
3168 		return -EOPNOTSUPP;
3169 
3170 	wqs_handles = kcalloc(num_wq_handles, sizeof(*wqs_handles),
3171 			      GFP_KERNEL);
3172 	if (!wqs_handles)
3173 		return -ENOMEM;
3174 
3175 	err = ib_copy_from_udata(wqs_handles, ucore,
3176 				 num_wq_handles * sizeof(__u32));
3177 	if (err)
3178 		goto err_free;
3179 
3180 	wqs = kcalloc(num_wq_handles, sizeof(*wqs), GFP_KERNEL);
3181 	if (!wqs) {
3182 		err = -ENOMEM;
3183 		goto  err_free;
3184 	}
3185 
3186 	for (num_read_wqs = 0; num_read_wqs < num_wq_handles;
3187 			num_read_wqs++) {
3188 		wq = uobj_get_obj_read(wq, wqs_handles[num_read_wqs],
3189 				       file->ucontext);
3190 		if (!wq) {
3191 			err = -EINVAL;
3192 			goto put_wqs;
3193 		}
3194 
3195 		wqs[num_read_wqs] = wq;
3196 	}
3197 
3198 	uobj  = uobj_alloc(uobj_get_type(rwq_ind_table), file->ucontext);
3199 	if (IS_ERR(uobj)) {
3200 		err = PTR_ERR(uobj);
3201 		goto put_wqs;
3202 	}
3203 
3204 	init_attr.log_ind_tbl_size = cmd.log_ind_tbl_size;
3205 	init_attr.ind_tbl = wqs;
3206 
3207 	if (!ib_dev->create_rwq_ind_table) {
3208 		err = -EOPNOTSUPP;
3209 		goto err_uobj;
3210 	}
3211 	rwq_ind_tbl = ib_dev->create_rwq_ind_table(ib_dev, &init_attr, uhw);
3212 
3213 	if (IS_ERR(rwq_ind_tbl)) {
3214 		err = PTR_ERR(rwq_ind_tbl);
3215 		goto err_uobj;
3216 	}
3217 
3218 	rwq_ind_tbl->ind_tbl = wqs;
3219 	rwq_ind_tbl->log_ind_tbl_size = init_attr.log_ind_tbl_size;
3220 	rwq_ind_tbl->uobject = uobj;
3221 	uobj->object = rwq_ind_tbl;
3222 	rwq_ind_tbl->device = ib_dev;
3223 	atomic_set(&rwq_ind_tbl->usecnt, 0);
3224 
3225 	for (i = 0; i < num_wq_handles; i++)
3226 		atomic_inc(&wqs[i]->usecnt);
3227 
3228 	resp.ind_tbl_handle = uobj->id;
3229 	resp.ind_tbl_num = rwq_ind_tbl->ind_tbl_num;
3230 	resp.response_length = required_resp_len;
3231 
3232 	err = ib_copy_to_udata(ucore,
3233 			       &resp, resp.response_length);
3234 	if (err)
3235 		goto err_copy;
3236 
3237 	kfree(wqs_handles);
3238 
3239 	for (j = 0; j < num_read_wqs; j++)
3240 		uobj_put_obj_read(wqs[j]);
3241 
3242 	uobj_alloc_commit(uobj);
3243 	return 0;
3244 
3245 err_copy:
3246 	ib_destroy_rwq_ind_table(rwq_ind_tbl);
3247 err_uobj:
3248 	uobj_alloc_abort(uobj);
3249 put_wqs:
3250 	for (j = 0; j < num_read_wqs; j++)
3251 		uobj_put_obj_read(wqs[j]);
3252 err_free:
3253 	kfree(wqs_handles);
3254 	kfree(wqs);
3255 	return err;
3256 }
3257 
3258 int ib_uverbs_ex_destroy_rwq_ind_table(struct ib_uverbs_file *file,
3259 				       struct ib_device *ib_dev,
3260 				       struct ib_udata *ucore,
3261 				       struct ib_udata *uhw)
3262 {
3263 	struct ib_uverbs_ex_destroy_rwq_ind_table	cmd = {};
3264 	struct ib_uobject		*uobj;
3265 	int			ret;
3266 	size_t required_cmd_sz;
3267 
3268 	required_cmd_sz = offsetof(typeof(cmd), ind_tbl_handle) + sizeof(cmd.ind_tbl_handle);
3269 
3270 	if (ucore->inlen < required_cmd_sz)
3271 		return -EINVAL;
3272 
3273 	if (ucore->inlen > sizeof(cmd) &&
3274 	    !ib_is_udata_cleared(ucore, sizeof(cmd),
3275 				 ucore->inlen - sizeof(cmd)))
3276 		return -EOPNOTSUPP;
3277 
3278 	ret = ib_copy_from_udata(&cmd, ucore, min(sizeof(cmd), ucore->inlen));
3279 	if (ret)
3280 		return ret;
3281 
3282 	if (cmd.comp_mask)
3283 		return -EOPNOTSUPP;
3284 
3285 	uobj  = uobj_get_write(uobj_get_type(rwq_ind_table), cmd.ind_tbl_handle,
3286 			       file->ucontext);
3287 	if (IS_ERR(uobj))
3288 		return PTR_ERR(uobj);
3289 
3290 	return uobj_remove_commit(uobj);
3291 }
3292 
3293 int ib_uverbs_ex_create_flow(struct ib_uverbs_file *file,
3294 			     struct ib_device *ib_dev,
3295 			     struct ib_udata *ucore,
3296 			     struct ib_udata *uhw)
3297 {
3298 	struct ib_uverbs_create_flow	  cmd;
3299 	struct ib_uverbs_create_flow_resp resp;
3300 	struct ib_uobject		  *uobj;
3301 	struct ib_flow			  *flow_id;
3302 	struct ib_uverbs_flow_attr	  *kern_flow_attr;
3303 	struct ib_flow_attr		  *flow_attr;
3304 	struct ib_qp			  *qp;
3305 	int err = 0;
3306 	void *kern_spec;
3307 	void *ib_spec;
3308 	int i;
3309 
3310 	if (ucore->inlen < sizeof(cmd))
3311 		return -EINVAL;
3312 
3313 	if (ucore->outlen < sizeof(resp))
3314 		return -ENOSPC;
3315 
3316 	err = ib_copy_from_udata(&cmd, ucore, sizeof(cmd));
3317 	if (err)
3318 		return err;
3319 
3320 	ucore->inbuf += sizeof(cmd);
3321 	ucore->inlen -= sizeof(cmd);
3322 
3323 	if (cmd.comp_mask)
3324 		return -EINVAL;
3325 
3326 	if (!capable(CAP_NET_RAW))
3327 		return -EPERM;
3328 
3329 	if (cmd.flow_attr.flags >= IB_FLOW_ATTR_FLAGS_RESERVED)
3330 		return -EINVAL;
3331 
3332 	if ((cmd.flow_attr.flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP) &&
3333 	    ((cmd.flow_attr.type == IB_FLOW_ATTR_ALL_DEFAULT) ||
3334 	     (cmd.flow_attr.type == IB_FLOW_ATTR_MC_DEFAULT)))
3335 		return -EINVAL;
3336 
3337 	if (cmd.flow_attr.num_of_specs > IB_FLOW_SPEC_SUPPORT_LAYERS)
3338 		return -EINVAL;
3339 
3340 	if (cmd.flow_attr.size > ucore->inlen ||
3341 	    cmd.flow_attr.size >
3342 	    (cmd.flow_attr.num_of_specs * sizeof(struct ib_uverbs_flow_spec)))
3343 		return -EINVAL;
3344 
3345 	if (cmd.flow_attr.reserved[0] ||
3346 	    cmd.flow_attr.reserved[1])
3347 		return -EINVAL;
3348 
3349 	if (cmd.flow_attr.num_of_specs) {
3350 		kern_flow_attr = kmalloc(sizeof(*kern_flow_attr) + cmd.flow_attr.size,
3351 					 GFP_KERNEL);
3352 		if (!kern_flow_attr)
3353 			return -ENOMEM;
3354 
3355 		memcpy(kern_flow_attr, &cmd.flow_attr, sizeof(*kern_flow_attr));
3356 		err = ib_copy_from_udata(kern_flow_attr + 1, ucore,
3357 					 cmd.flow_attr.size);
3358 		if (err)
3359 			goto err_free_attr;
3360 	} else {
3361 		kern_flow_attr = &cmd.flow_attr;
3362 	}
3363 
3364 	uobj  = uobj_alloc(uobj_get_type(flow), file->ucontext);
3365 	if (IS_ERR(uobj)) {
3366 		err = PTR_ERR(uobj);
3367 		goto err_free_attr;
3368 	}
3369 
3370 	qp = uobj_get_obj_read(qp, cmd.qp_handle, file->ucontext);
3371 	if (!qp) {
3372 		err = -EINVAL;
3373 		goto err_uobj;
3374 	}
3375 
3376 	flow_attr = kzalloc(sizeof(*flow_attr) + cmd.flow_attr.num_of_specs *
3377 			    sizeof(union ib_flow_spec), GFP_KERNEL);
3378 	if (!flow_attr) {
3379 		err = -ENOMEM;
3380 		goto err_put;
3381 	}
3382 
3383 	flow_attr->type = kern_flow_attr->type;
3384 	flow_attr->priority = kern_flow_attr->priority;
3385 	flow_attr->num_of_specs = kern_flow_attr->num_of_specs;
3386 	flow_attr->port = kern_flow_attr->port;
3387 	flow_attr->flags = kern_flow_attr->flags;
3388 	flow_attr->size = sizeof(*flow_attr);
3389 
3390 	kern_spec = kern_flow_attr + 1;
3391 	ib_spec = flow_attr + 1;
3392 	for (i = 0; i < flow_attr->num_of_specs &&
3393 	     cmd.flow_attr.size > offsetof(struct ib_uverbs_flow_spec, reserved) &&
3394 	     cmd.flow_attr.size >=
3395 	     ((struct ib_uverbs_flow_spec *)kern_spec)->size; i++) {
3396 		err = kern_spec_to_ib_spec(kern_spec, ib_spec);
3397 		if (err)
3398 			goto err_free;
3399 		flow_attr->size +=
3400 			((union ib_flow_spec *) ib_spec)->size;
3401 		cmd.flow_attr.size -= ((struct ib_uverbs_flow_spec *)kern_spec)->size;
3402 		kern_spec += ((struct ib_uverbs_flow_spec *) kern_spec)->size;
3403 		ib_spec += ((union ib_flow_spec *) ib_spec)->size;
3404 	}
3405 	if (cmd.flow_attr.size || (i != flow_attr->num_of_specs)) {
3406 		pr_warn("create flow failed, flow %d: %d bytes left from uverb cmd\n",
3407 			i, cmd.flow_attr.size);
3408 		err = -EINVAL;
3409 		goto err_free;
3410 	}
3411 	flow_id = ib_create_flow(qp, flow_attr, IB_FLOW_DOMAIN_USER);
3412 	if (IS_ERR(flow_id)) {
3413 		err = PTR_ERR(flow_id);
3414 		goto err_free;
3415 	}
3416 	flow_id->uobject = uobj;
3417 	uobj->object = flow_id;
3418 
3419 	memset(&resp, 0, sizeof(resp));
3420 	resp.flow_handle = uobj->id;
3421 
3422 	err = ib_copy_to_udata(ucore,
3423 			       &resp, sizeof(resp));
3424 	if (err)
3425 		goto err_copy;
3426 
3427 	uobj_put_obj_read(qp);
3428 	uobj_alloc_commit(uobj);
3429 	kfree(flow_attr);
3430 	if (cmd.flow_attr.num_of_specs)
3431 		kfree(kern_flow_attr);
3432 	return 0;
3433 err_copy:
3434 	ib_destroy_flow(flow_id);
3435 err_free:
3436 	kfree(flow_attr);
3437 err_put:
3438 	uobj_put_obj_read(qp);
3439 err_uobj:
3440 	uobj_alloc_abort(uobj);
3441 err_free_attr:
3442 	if (cmd.flow_attr.num_of_specs)
3443 		kfree(kern_flow_attr);
3444 	return err;
3445 }
3446 
3447 int ib_uverbs_ex_destroy_flow(struct ib_uverbs_file *file,
3448 			      struct ib_device *ib_dev,
3449 			      struct ib_udata *ucore,
3450 			      struct ib_udata *uhw)
3451 {
3452 	struct ib_uverbs_destroy_flow	cmd;
3453 	struct ib_uobject		*uobj;
3454 	int				ret;
3455 
3456 	if (ucore->inlen < sizeof(cmd))
3457 		return -EINVAL;
3458 
3459 	ret = ib_copy_from_udata(&cmd, ucore, sizeof(cmd));
3460 	if (ret)
3461 		return ret;
3462 
3463 	if (cmd.comp_mask)
3464 		return -EINVAL;
3465 
3466 	uobj  = uobj_get_write(uobj_get_type(flow), cmd.flow_handle,
3467 			       file->ucontext);
3468 	if (IS_ERR(uobj))
3469 		return PTR_ERR(uobj);
3470 
3471 	ret = uobj_remove_commit(uobj);
3472 	return ret;
3473 }
3474 
3475 static int __uverbs_create_xsrq(struct ib_uverbs_file *file,
3476 				struct ib_device *ib_dev,
3477 				struct ib_uverbs_create_xsrq *cmd,
3478 				struct ib_udata *udata)
3479 {
3480 	struct ib_uverbs_create_srq_resp resp;
3481 	struct ib_usrq_object           *obj;
3482 	struct ib_pd                    *pd;
3483 	struct ib_srq                   *srq;
3484 	struct ib_uobject               *uninitialized_var(xrcd_uobj);
3485 	struct ib_srq_init_attr          attr;
3486 	int ret;
3487 
3488 	obj  = (struct ib_usrq_object *)uobj_alloc(uobj_get_type(srq),
3489 						   file->ucontext);
3490 	if (IS_ERR(obj))
3491 		return PTR_ERR(obj);
3492 
3493 	if (cmd->srq_type == IB_SRQT_TM)
3494 		attr.ext.tag_matching.max_num_tags = cmd->max_num_tags;
3495 
3496 	if (cmd->srq_type == IB_SRQT_XRC) {
3497 		xrcd_uobj = uobj_get_read(uobj_get_type(xrcd), cmd->xrcd_handle,
3498 					  file->ucontext);
3499 		if (IS_ERR(xrcd_uobj)) {
3500 			ret = -EINVAL;
3501 			goto err;
3502 		}
3503 
3504 		attr.ext.xrc.xrcd = (struct ib_xrcd *)xrcd_uobj->object;
3505 		if (!attr.ext.xrc.xrcd) {
3506 			ret = -EINVAL;
3507 			goto err_put_xrcd;
3508 		}
3509 
3510 		obj->uxrcd = container_of(xrcd_uobj, struct ib_uxrcd_object, uobject);
3511 		atomic_inc(&obj->uxrcd->refcnt);
3512 	}
3513 
3514 	if (ib_srq_has_cq(cmd->srq_type)) {
3515 		attr.ext.cq  = uobj_get_obj_read(cq, cmd->cq_handle,
3516 						 file->ucontext);
3517 		if (!attr.ext.cq) {
3518 			ret = -EINVAL;
3519 			goto err_put_xrcd;
3520 		}
3521 	}
3522 
3523 	pd  = uobj_get_obj_read(pd, cmd->pd_handle, file->ucontext);
3524 	if (!pd) {
3525 		ret = -EINVAL;
3526 		goto err_put_cq;
3527 	}
3528 
3529 	attr.event_handler  = ib_uverbs_srq_event_handler;
3530 	attr.srq_context    = file;
3531 	attr.srq_type       = cmd->srq_type;
3532 	attr.attr.max_wr    = cmd->max_wr;
3533 	attr.attr.max_sge   = cmd->max_sge;
3534 	attr.attr.srq_limit = cmd->srq_limit;
3535 
3536 	obj->uevent.events_reported = 0;
3537 	INIT_LIST_HEAD(&obj->uevent.event_list);
3538 
3539 	srq = pd->device->create_srq(pd, &attr, udata);
3540 	if (IS_ERR(srq)) {
3541 		ret = PTR_ERR(srq);
3542 		goto err_put;
3543 	}
3544 
3545 	srq->device        = pd->device;
3546 	srq->pd            = pd;
3547 	srq->srq_type	   = cmd->srq_type;
3548 	srq->uobject       = &obj->uevent.uobject;
3549 	srq->event_handler = attr.event_handler;
3550 	srq->srq_context   = attr.srq_context;
3551 
3552 	if (ib_srq_has_cq(cmd->srq_type)) {
3553 		srq->ext.cq       = attr.ext.cq;
3554 		atomic_inc(&attr.ext.cq->usecnt);
3555 	}
3556 
3557 	if (cmd->srq_type == IB_SRQT_XRC) {
3558 		srq->ext.xrc.xrcd = attr.ext.xrc.xrcd;
3559 		atomic_inc(&attr.ext.xrc.xrcd->usecnt);
3560 	}
3561 
3562 	atomic_inc(&pd->usecnt);
3563 	atomic_set(&srq->usecnt, 0);
3564 
3565 	obj->uevent.uobject.object = srq;
3566 	obj->uevent.uobject.user_handle = cmd->user_handle;
3567 
3568 	memset(&resp, 0, sizeof resp);
3569 	resp.srq_handle = obj->uevent.uobject.id;
3570 	resp.max_wr     = attr.attr.max_wr;
3571 	resp.max_sge    = attr.attr.max_sge;
3572 	if (cmd->srq_type == IB_SRQT_XRC)
3573 		resp.srqn = srq->ext.xrc.srq_num;
3574 
3575 	if (copy_to_user((void __user *) (unsigned long) cmd->response,
3576 			 &resp, sizeof resp)) {
3577 		ret = -EFAULT;
3578 		goto err_copy;
3579 	}
3580 
3581 	if (cmd->srq_type == IB_SRQT_XRC)
3582 		uobj_put_read(xrcd_uobj);
3583 
3584 	if (ib_srq_has_cq(cmd->srq_type))
3585 		uobj_put_obj_read(attr.ext.cq);
3586 
3587 	uobj_put_obj_read(pd);
3588 	uobj_alloc_commit(&obj->uevent.uobject);
3589 
3590 	return 0;
3591 
3592 err_copy:
3593 	ib_destroy_srq(srq);
3594 
3595 err_put:
3596 	uobj_put_obj_read(pd);
3597 
3598 err_put_cq:
3599 	if (ib_srq_has_cq(cmd->srq_type))
3600 		uobj_put_obj_read(attr.ext.cq);
3601 
3602 err_put_xrcd:
3603 	if (cmd->srq_type == IB_SRQT_XRC) {
3604 		atomic_dec(&obj->uxrcd->refcnt);
3605 		uobj_put_read(xrcd_uobj);
3606 	}
3607 
3608 err:
3609 	uobj_alloc_abort(&obj->uevent.uobject);
3610 	return ret;
3611 }
3612 
3613 ssize_t ib_uverbs_create_srq(struct ib_uverbs_file *file,
3614 			     struct ib_device *ib_dev,
3615 			     const char __user *buf, int in_len,
3616 			     int out_len)
3617 {
3618 	struct ib_uverbs_create_srq      cmd;
3619 	struct ib_uverbs_create_xsrq     xcmd;
3620 	struct ib_uverbs_create_srq_resp resp;
3621 	struct ib_udata                  udata;
3622 	int ret;
3623 
3624 	if (out_len < sizeof resp)
3625 		return -ENOSPC;
3626 
3627 	if (copy_from_user(&cmd, buf, sizeof cmd))
3628 		return -EFAULT;
3629 
3630 	memset(&xcmd, 0, sizeof(xcmd));
3631 	xcmd.response	 = cmd.response;
3632 	xcmd.user_handle = cmd.user_handle;
3633 	xcmd.srq_type	 = IB_SRQT_BASIC;
3634 	xcmd.pd_handle	 = cmd.pd_handle;
3635 	xcmd.max_wr	 = cmd.max_wr;
3636 	xcmd.max_sge	 = cmd.max_sge;
3637 	xcmd.srq_limit	 = cmd.srq_limit;
3638 
3639 	ib_uverbs_init_udata(&udata, buf + sizeof(cmd),
3640 		   u64_to_user_ptr(cmd.response) + sizeof(resp),
3641 		   in_len - sizeof(cmd) - sizeof(struct ib_uverbs_cmd_hdr),
3642 		   out_len - sizeof(resp));
3643 
3644 	ret = __uverbs_create_xsrq(file, ib_dev, &xcmd, &udata);
3645 	if (ret)
3646 		return ret;
3647 
3648 	return in_len;
3649 }
3650 
3651 ssize_t ib_uverbs_create_xsrq(struct ib_uverbs_file *file,
3652 			      struct ib_device *ib_dev,
3653 			      const char __user *buf, int in_len, int out_len)
3654 {
3655 	struct ib_uverbs_create_xsrq     cmd;
3656 	struct ib_uverbs_create_srq_resp resp;
3657 	struct ib_udata                  udata;
3658 	int ret;
3659 
3660 	if (out_len < sizeof resp)
3661 		return -ENOSPC;
3662 
3663 	if (copy_from_user(&cmd, buf, sizeof cmd))
3664 		return -EFAULT;
3665 
3666 	ib_uverbs_init_udata(&udata, buf + sizeof(cmd),
3667 		   u64_to_user_ptr(cmd.response) + sizeof(resp),
3668 		   in_len - sizeof(cmd) - sizeof(struct ib_uverbs_cmd_hdr),
3669 		   out_len - sizeof(resp));
3670 
3671 	ret = __uverbs_create_xsrq(file, ib_dev, &cmd, &udata);
3672 	if (ret)
3673 		return ret;
3674 
3675 	return in_len;
3676 }
3677 
3678 ssize_t ib_uverbs_modify_srq(struct ib_uverbs_file *file,
3679 			     struct ib_device *ib_dev,
3680 			     const char __user *buf, int in_len,
3681 			     int out_len)
3682 {
3683 	struct ib_uverbs_modify_srq cmd;
3684 	struct ib_udata             udata;
3685 	struct ib_srq              *srq;
3686 	struct ib_srq_attr          attr;
3687 	int                         ret;
3688 
3689 	if (copy_from_user(&cmd, buf, sizeof cmd))
3690 		return -EFAULT;
3691 
3692 	ib_uverbs_init_udata(&udata, buf + sizeof cmd, NULL, in_len - sizeof cmd,
3693 		   out_len);
3694 
3695 	srq = uobj_get_obj_read(srq, cmd.srq_handle, file->ucontext);
3696 	if (!srq)
3697 		return -EINVAL;
3698 
3699 	attr.max_wr    = cmd.max_wr;
3700 	attr.srq_limit = cmd.srq_limit;
3701 
3702 	ret = srq->device->modify_srq(srq, &attr, cmd.attr_mask, &udata);
3703 
3704 	uobj_put_obj_read(srq);
3705 
3706 	return ret ? ret : in_len;
3707 }
3708 
3709 ssize_t ib_uverbs_query_srq(struct ib_uverbs_file *file,
3710 			    struct ib_device *ib_dev,
3711 			    const char __user *buf,
3712 			    int in_len, int out_len)
3713 {
3714 	struct ib_uverbs_query_srq      cmd;
3715 	struct ib_uverbs_query_srq_resp resp;
3716 	struct ib_srq_attr              attr;
3717 	struct ib_srq                   *srq;
3718 	int                             ret;
3719 
3720 	if (out_len < sizeof resp)
3721 		return -ENOSPC;
3722 
3723 	if (copy_from_user(&cmd, buf, sizeof cmd))
3724 		return -EFAULT;
3725 
3726 	srq = uobj_get_obj_read(srq, cmd.srq_handle, file->ucontext);
3727 	if (!srq)
3728 		return -EINVAL;
3729 
3730 	ret = ib_query_srq(srq, &attr);
3731 
3732 	uobj_put_obj_read(srq);
3733 
3734 	if (ret)
3735 		return ret;
3736 
3737 	memset(&resp, 0, sizeof resp);
3738 
3739 	resp.max_wr    = attr.max_wr;
3740 	resp.max_sge   = attr.max_sge;
3741 	resp.srq_limit = attr.srq_limit;
3742 
3743 	if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof resp))
3744 		return -EFAULT;
3745 
3746 	return in_len;
3747 }
3748 
3749 ssize_t ib_uverbs_destroy_srq(struct ib_uverbs_file *file,
3750 			      struct ib_device *ib_dev,
3751 			      const char __user *buf, int in_len,
3752 			      int out_len)
3753 {
3754 	struct ib_uverbs_destroy_srq      cmd;
3755 	struct ib_uverbs_destroy_srq_resp resp;
3756 	struct ib_uobject		 *uobj;
3757 	struct ib_uevent_object        	 *obj;
3758 	int                         	  ret = -EINVAL;
3759 
3760 	if (copy_from_user(&cmd, buf, sizeof cmd))
3761 		return -EFAULT;
3762 
3763 	uobj  = uobj_get_write(uobj_get_type(srq), cmd.srq_handle,
3764 			       file->ucontext);
3765 	if (IS_ERR(uobj))
3766 		return PTR_ERR(uobj);
3767 
3768 	obj = container_of(uobj, struct ib_uevent_object, uobject);
3769 	/*
3770 	 * Make sure we don't free the memory in remove_commit as we still
3771 	 * needs the uobject memory to create the response.
3772 	 */
3773 	uverbs_uobject_get(uobj);
3774 
3775 	memset(&resp, 0, sizeof(resp));
3776 
3777 	ret = uobj_remove_commit(uobj);
3778 	if (ret) {
3779 		uverbs_uobject_put(uobj);
3780 		return ret;
3781 	}
3782 	resp.events_reported = obj->events_reported;
3783 	uverbs_uobject_put(uobj);
3784 	if (copy_to_user(u64_to_user_ptr(cmd.response), &resp, sizeof(resp)))
3785 		return -EFAULT;
3786 
3787 	return in_len;
3788 }
3789 
3790 int ib_uverbs_ex_query_device(struct ib_uverbs_file *file,
3791 			      struct ib_device *ib_dev,
3792 			      struct ib_udata *ucore,
3793 			      struct ib_udata *uhw)
3794 {
3795 	struct ib_uverbs_ex_query_device_resp resp = { {0} };
3796 	struct ib_uverbs_ex_query_device  cmd;
3797 	struct ib_device_attr attr = {0};
3798 	int err;
3799 
3800 	if (!ib_dev->query_device)
3801 		return -EOPNOTSUPP;
3802 
3803 	if (ucore->inlen < sizeof(cmd))
3804 		return -EINVAL;
3805 
3806 	err = ib_copy_from_udata(&cmd, ucore, sizeof(cmd));
3807 	if (err)
3808 		return err;
3809 
3810 	if (cmd.comp_mask)
3811 		return -EINVAL;
3812 
3813 	if (cmd.reserved)
3814 		return -EINVAL;
3815 
3816 	resp.response_length = offsetof(typeof(resp), odp_caps);
3817 
3818 	if (ucore->outlen < resp.response_length)
3819 		return -ENOSPC;
3820 
3821 	err = ib_dev->query_device(ib_dev, &attr, uhw);
3822 	if (err)
3823 		return err;
3824 
3825 	copy_query_dev_fields(file, ib_dev, &resp.base, &attr);
3826 
3827 	if (ucore->outlen < resp.response_length + sizeof(resp.odp_caps))
3828 		goto end;
3829 
3830 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
3831 	resp.odp_caps.general_caps = attr.odp_caps.general_caps;
3832 	resp.odp_caps.per_transport_caps.rc_odp_caps =
3833 		attr.odp_caps.per_transport_caps.rc_odp_caps;
3834 	resp.odp_caps.per_transport_caps.uc_odp_caps =
3835 		attr.odp_caps.per_transport_caps.uc_odp_caps;
3836 	resp.odp_caps.per_transport_caps.ud_odp_caps =
3837 		attr.odp_caps.per_transport_caps.ud_odp_caps;
3838 #endif
3839 	resp.response_length += sizeof(resp.odp_caps);
3840 
3841 	if (ucore->outlen < resp.response_length + sizeof(resp.timestamp_mask))
3842 		goto end;
3843 
3844 	resp.timestamp_mask = attr.timestamp_mask;
3845 	resp.response_length += sizeof(resp.timestamp_mask);
3846 
3847 	if (ucore->outlen < resp.response_length + sizeof(resp.hca_core_clock))
3848 		goto end;
3849 
3850 	resp.hca_core_clock = attr.hca_core_clock;
3851 	resp.response_length += sizeof(resp.hca_core_clock);
3852 
3853 	if (ucore->outlen < resp.response_length + sizeof(resp.device_cap_flags_ex))
3854 		goto end;
3855 
3856 	resp.device_cap_flags_ex = attr.device_cap_flags;
3857 	resp.response_length += sizeof(resp.device_cap_flags_ex);
3858 
3859 	if (ucore->outlen < resp.response_length + sizeof(resp.rss_caps))
3860 		goto end;
3861 
3862 	resp.rss_caps.supported_qpts = attr.rss_caps.supported_qpts;
3863 	resp.rss_caps.max_rwq_indirection_tables =
3864 		attr.rss_caps.max_rwq_indirection_tables;
3865 	resp.rss_caps.max_rwq_indirection_table_size =
3866 		attr.rss_caps.max_rwq_indirection_table_size;
3867 
3868 	resp.response_length += sizeof(resp.rss_caps);
3869 
3870 	if (ucore->outlen < resp.response_length + sizeof(resp.max_wq_type_rq))
3871 		goto end;
3872 
3873 	resp.max_wq_type_rq = attr.max_wq_type_rq;
3874 	resp.response_length += sizeof(resp.max_wq_type_rq);
3875 
3876 	if (ucore->outlen < resp.response_length + sizeof(resp.raw_packet_caps))
3877 		goto end;
3878 
3879 	resp.raw_packet_caps = attr.raw_packet_caps;
3880 	resp.response_length += sizeof(resp.raw_packet_caps);
3881 
3882 	if (ucore->outlen < resp.response_length + sizeof(resp.tm_caps))
3883 		goto end;
3884 
3885 	resp.tm_caps.max_rndv_hdr_size	= attr.tm_caps.max_rndv_hdr_size;
3886 	resp.tm_caps.max_num_tags	= attr.tm_caps.max_num_tags;
3887 	resp.tm_caps.max_ops		= attr.tm_caps.max_ops;
3888 	resp.tm_caps.max_sge		= attr.tm_caps.max_sge;
3889 	resp.tm_caps.flags		= attr.tm_caps.flags;
3890 	resp.response_length += sizeof(resp.tm_caps);
3891 
3892 	if (ucore->outlen < resp.response_length + sizeof(resp.cq_moderation_caps))
3893 		goto end;
3894 
3895 	resp.cq_moderation_caps.max_cq_moderation_count  =
3896 		attr.cq_caps.max_cq_moderation_count;
3897 	resp.cq_moderation_caps.max_cq_moderation_period =
3898 		attr.cq_caps.max_cq_moderation_period;
3899 	resp.response_length += sizeof(resp.cq_moderation_caps);
3900 end:
3901 	err = ib_copy_to_udata(ucore, &resp, resp.response_length);
3902 	return err;
3903 }
3904 
3905 int ib_uverbs_ex_modify_cq(struct ib_uverbs_file *file,
3906 			   struct ib_device *ib_dev,
3907 			   struct ib_udata *ucore,
3908 			   struct ib_udata *uhw)
3909 {
3910 	struct ib_uverbs_ex_modify_cq cmd = {};
3911 	struct ib_cq *cq;
3912 	size_t required_cmd_sz;
3913 	int ret;
3914 
3915 	required_cmd_sz = offsetof(typeof(cmd), reserved) +
3916 				sizeof(cmd.reserved);
3917 	if (ucore->inlen < required_cmd_sz)
3918 		return -EINVAL;
3919 
3920 	/* sanity checks */
3921 	if (ucore->inlen > sizeof(cmd) &&
3922 	    !ib_is_udata_cleared(ucore, sizeof(cmd),
3923 				 ucore->inlen - sizeof(cmd)))
3924 		return -EOPNOTSUPP;
3925 
3926 	ret = ib_copy_from_udata(&cmd, ucore, min(sizeof(cmd), ucore->inlen));
3927 	if (ret)
3928 		return ret;
3929 
3930 	if (!cmd.attr_mask || cmd.reserved)
3931 		return -EINVAL;
3932 
3933 	if (cmd.attr_mask > IB_CQ_MODERATE)
3934 		return -EOPNOTSUPP;
3935 
3936 	cq = uobj_get_obj_read(cq, cmd.cq_handle, file->ucontext);
3937 	if (!cq)
3938 		return -EINVAL;
3939 
3940 	ret = rdma_set_cq_moderation(cq, cmd.attr.cq_count, cmd.attr.cq_period);
3941 
3942 	uobj_put_obj_read(cq);
3943 
3944 	return ret;
3945 }
3946