1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause OR GPL-2.0
3  *
4  * Copyright (c) 2005 Topspin Communications.  All rights reserved.
5  * Copyright (c) 2005, 2006, 2007 Cisco Systems.  All rights reserved.
6  * Copyright (c) 2005 PathScale, Inc.  All rights reserved.
7  * Copyright (c) 2006 Mellanox Technologies.  All rights reserved.
8  *
9  * This software is available to you under a choice of one of two
10  * licenses.  You may choose to be licensed under the terms of the GNU
11  * General Public License (GPL) Version 2, available from the file
12  * COPYING in the main directory of this source tree, or the
13  * OpenIB.org BSD license below:
14  *
15  *     Redistribution and use in source and binary forms, with or
16  *     without modification, are permitted provided that the following
17  *     conditions are met:
18  *
19  *      - Redistributions of source code must retain the above
20  *        copyright notice, this list of conditions and the following
21  *        disclaimer.
22  *
23  *      - Redistributions in binary form must reproduce the above
24  *        copyright notice, this list of conditions and the following
25  *        disclaimer in the documentation and/or other materials
26  *        provided with the distribution.
27  *
28  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
29  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
30  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
31  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
32  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
33  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
34  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
35  * SOFTWARE.
36  */
37 
38 #include <sys/cdefs.h>
39 __FBSDID("$FreeBSD$");
40 
41 #define	LINUXKPI_PARAM_PREFIX ibcore_
42 
43 #include <sys/priv.h>
44 
45 #include <linux/file.h>
46 #include <linux/fs.h>
47 #include <linux/slab.h>
48 #include <linux/sched.h>
49 
50 #include <linux/uaccess.h>
51 
52 #include <rdma/uverbs_types.h>
53 #include <rdma/uverbs_std_types.h>
54 #include "rdma_core.h"
55 
56 #include "uverbs.h"
57 #include "core_priv.h"
58 
59 /*
60  * Copy a response to userspace. If the provided 'resp' is larger than the
61  * user buffer it is silently truncated. If the user provided a larger buffer
62  * then the trailing portion is zero filled.
63  *
64  * These semantics are intended to support future extension of the output
65  * structures.
66  */
67 static int uverbs_response(struct uverbs_attr_bundle *attrs, const void *resp,
68 			   size_t resp_len)
69 {
70 	int ret;
71 
72 	if (uverbs_attr_is_valid(attrs, UVERBS_ATTR_CORE_OUT))
73 		return uverbs_copy_to_struct_or_zero(
74 			attrs, UVERBS_ATTR_CORE_OUT, resp, resp_len);
75 
76 	if (copy_to_user(attrs->ucore.outbuf, resp,
77 			 min(attrs->ucore.outlen, resp_len)))
78 		return -EFAULT;
79 
80 	if (resp_len < attrs->ucore.outlen) {
81 		/*
82 		 * Zero fill any extra memory that user
83 		 * space might have provided.
84 		 */
85 		ret = clear_user(attrs->ucore.outbuf + resp_len,
86 				 attrs->ucore.outlen - resp_len);
87 		if (ret)
88 			return -EFAULT;
89 	}
90 
91 	return 0;
92 }
93 
94 /*
95  * Copy a request from userspace. If the provided 'req' is larger than the
96  * user buffer then the user buffer is zero extended into the 'req'. If 'req'
97  * is smaller than the user buffer then the uncopied bytes in the user buffer
98  * must be zero.
99  */
100 static int uverbs_request(struct uverbs_attr_bundle *attrs, void *req,
101 			  size_t req_len)
102 {
103 	if (copy_from_user(req, attrs->ucore.inbuf,
104 			   min(attrs->ucore.inlen, req_len)))
105 		return -EFAULT;
106 
107 	if (attrs->ucore.inlen < req_len) {
108 		memset((u8 *)req + attrs->ucore.inlen, 0,
109 		       req_len - attrs->ucore.inlen);
110 	} else if (attrs->ucore.inlen > req_len) {
111 		if (!ib_is_buffer_cleared(attrs->ucore.inbuf + req_len,
112 					  attrs->ucore.inlen - req_len))
113 			return -EOPNOTSUPP;
114 	}
115 	return 0;
116 }
117 
118 /*
119  * Generate the value for the 'response_length' protocol used by write_ex.
120  * This is the number of bytes the kernel actually wrote. Userspace can use
121  * this to detect what structure members in the response the kernel
122  * understood.
123  */
124 static u32 uverbs_response_length(struct uverbs_attr_bundle *attrs,
125 				  size_t resp_len)
126 {
127 	return min_t(size_t, attrs->ucore.outlen, resp_len);
128 }
129 
130 /*
131  * The iterator version of the request interface is for handlers that need to
132  * step over a flex array at the end of a command header.
133  */
134 struct uverbs_req_iter {
135 	const u8 __user *cur;
136 	const u8 __user *end;
137 };
138 
139 static int uverbs_request_start(struct uverbs_attr_bundle *attrs,
140 				struct uverbs_req_iter *iter,
141 				void *req,
142 				size_t req_len)
143 {
144 	if (attrs->ucore.inlen < req_len)
145 		return -ENOSPC;
146 
147 	if (copy_from_user(req, attrs->ucore.inbuf, req_len))
148 		return -EFAULT;
149 
150 	iter->cur = attrs->ucore.inbuf + req_len;
151 	iter->end = attrs->ucore.inbuf + attrs->ucore.inlen;
152 	return 0;
153 }
154 
155 static int uverbs_request_next(struct uverbs_req_iter *iter, void *val,
156 			       size_t len)
157 {
158 	if (iter->cur + len > iter->end)
159 		return -ENOSPC;
160 
161 	if (copy_from_user(val, iter->cur, len))
162 		return -EFAULT;
163 
164 	iter->cur += len;
165 	return 0;
166 }
167 
168 static const void __user *uverbs_request_next_ptr(struct uverbs_req_iter *iter,
169 						  size_t len)
170 {
171 	const void __user *res = iter->cur;
172 
173 	if (iter->cur + len > iter->end)
174 		return (void __force __user *)ERR_PTR(-ENOSPC);
175 	iter->cur += len;
176 	return res;
177 }
178 
179 static int uverbs_request_finish(struct uverbs_req_iter *iter)
180 {
181 	if (!ib_is_buffer_cleared(iter->cur, iter->end - iter->cur))
182 		return -EOPNOTSUPP;
183 	return 0;
184 }
185 
186 /*
187  * When calling a destroy function during an error unwind we need to pass in
188  * the udata that is sanitized of all user arguments. Ie from the driver
189  * perspective it looks like no udata was passed.
190  */
191 struct ib_udata *uverbs_get_cleared_udata(struct uverbs_attr_bundle *attrs)
192 {
193 	attrs->driver_udata = (struct ib_udata){};
194 	return &attrs->driver_udata;
195 }
196 
197 static struct ib_uverbs_completion_event_file *
198 _ib_uverbs_lookup_comp_file(s32 fd, struct uverbs_attr_bundle *attrs)
199 {
200 	struct ib_uobject *uobj = ufd_get_read(UVERBS_OBJECT_COMP_CHANNEL,
201 					       fd, attrs);
202 
203 	if (IS_ERR(uobj))
204 		return (void *)uobj;
205 
206 	uverbs_uobject_get(uobj);
207 	uobj_put_read(uobj);
208 
209 	return container_of(uobj, struct ib_uverbs_completion_event_file,
210 			    uobj);
211 }
212 #define ib_uverbs_lookup_comp_file(_fd, _ufile) ({			\
213 	CTASSERT(sizeof(_fd) == sizeof(s32));				\
214 	_ib_uverbs_lookup_comp_file(_fd, _ufile);			\
215 })
216 
217 int ib_alloc_ucontext(struct uverbs_attr_bundle *attrs)
218 {
219 	struct ib_uverbs_file *ufile = attrs->ufile;
220 	struct ib_ucontext *ucontext;
221 	struct ib_device *ib_dev;
222 
223 	ib_dev = srcu_dereference(ufile->device->ib_dev,
224 				  &ufile->device->disassociate_srcu);
225 	if (!ib_dev)
226 		return -EIO;
227 
228 	ucontext = rdma_zalloc_drv_obj(ib_dev, ib_ucontext);
229 	if (!ucontext)
230 		return -ENOMEM;
231 
232 	ucontext->device = ib_dev;
233 	ucontext->ufile = ufile;
234 	xa_init_flags(&ucontext->mmap_xa, XA_FLAGS_ALLOC);
235 	attrs->context = ucontext;
236 	return 0;
237 }
238 
239 int ib_init_ucontext(struct uverbs_attr_bundle *attrs)
240 {
241 	struct ib_ucontext *ucontext = attrs->context;
242 	struct ib_uverbs_file *file = attrs->ufile;
243 	int ret;
244 
245 	if (!down_read_trylock(&file->hw_destroy_rwsem))
246 		return -EIO;
247 	mutex_lock(&file->ucontext_lock);
248 	if (file->ucontext) {
249 		ret = -EINVAL;
250 		goto err;
251 	}
252 
253 	ret = ucontext->device->alloc_ucontext(ucontext,
254 						   &attrs->driver_udata);
255 	if (ret)
256 		goto err_uncharge;
257 
258 	/*
259 	 * Make sure that ib_uverbs_get_ucontext() sees the pointer update
260 	 * only after all writes to setup the ucontext have completed
261 	 */
262 	atomic_store_rel_ptr((uintptr_t *)&file->ucontext, (uintptr_t)ucontext);
263 
264 	mutex_unlock(&file->ucontext_lock);
265 	up_read(&file->hw_destroy_rwsem);
266 	return 0;
267 
268 err_uncharge:
269 err:
270 	mutex_unlock(&file->ucontext_lock);
271 	up_read(&file->hw_destroy_rwsem);
272 	return ret;
273 }
274 
275 static int ib_uverbs_get_context(struct uverbs_attr_bundle *attrs)
276 {
277 	struct ib_uverbs_get_context_resp resp;
278 	struct ib_uverbs_get_context cmd;
279 	struct ib_device *ib_dev;
280 	struct ib_uobject *uobj;
281 	int ret;
282 
283 	ret = uverbs_request(attrs, &cmd, sizeof(cmd));
284 	if (ret)
285 		return ret;
286 
287 	ret = ib_alloc_ucontext(attrs);
288 	if (ret)
289 		return ret;
290 
291 	uobj = uobj_alloc(UVERBS_OBJECT_ASYNC_EVENT, attrs, &ib_dev);
292 	if (IS_ERR(uobj)) {
293 		ret = PTR_ERR(uobj);
294 		goto err_ucontext;
295 	}
296 
297 	resp = (struct ib_uverbs_get_context_resp){
298 		.num_comp_vectors = attrs->ufile->device->num_comp_vectors,
299 		.async_fd = uobj->id,
300 	};
301 	ret = uverbs_response(attrs, &resp, sizeof(resp));
302 	if (ret)
303 		goto err_uobj;
304 
305 	ret = ib_init_ucontext(attrs);
306 	if (ret)
307 		goto err_uobj;
308 
309 	ib_uverbs_init_async_event_file(
310 		container_of(uobj, struct ib_uverbs_async_event_file, uobj));
311 	rdma_alloc_commit_uobject(uobj, attrs);
312 	return 0;
313 
314 err_uobj:
315 	rdma_alloc_abort_uobject(uobj, attrs);
316 err_ucontext:
317 	kfree(attrs->context);
318 	attrs->context = NULL;
319 	return ret;
320 }
321 
322 static void copy_query_dev_fields(struct ib_ucontext *ucontext,
323 				  struct ib_uverbs_query_device_resp *resp,
324 				  struct ib_device_attr *attr)
325 {
326 	struct ib_device *ib_dev = ucontext->device;
327 
328 	resp->fw_ver		= attr->fw_ver;
329 	resp->node_guid		= ib_dev->node_guid;
330 	resp->sys_image_guid	= attr->sys_image_guid;
331 	resp->max_mr_size	= attr->max_mr_size;
332 	resp->page_size_cap	= attr->page_size_cap;
333 	resp->vendor_id		= attr->vendor_id;
334 	resp->vendor_part_id	= attr->vendor_part_id;
335 	resp->hw_ver		= attr->hw_ver;
336 	resp->max_qp		= attr->max_qp;
337 	resp->max_qp_wr		= attr->max_qp_wr;
338 	resp->device_cap_flags	= (u32)attr->device_cap_flags;
339 	resp->max_sge		= min(attr->max_send_sge, attr->max_recv_sge);
340 	resp->max_sge_rd	= attr->max_sge_rd;
341 	resp->max_cq		= attr->max_cq;
342 	resp->max_cqe		= attr->max_cqe;
343 	resp->max_mr		= attr->max_mr;
344 	resp->max_pd		= attr->max_pd;
345 	resp->max_qp_rd_atom	= attr->max_qp_rd_atom;
346 	resp->max_ee_rd_atom	= attr->max_ee_rd_atom;
347 	resp->max_res_rd_atom	= attr->max_res_rd_atom;
348 	resp->max_qp_init_rd_atom	= attr->max_qp_init_rd_atom;
349 	resp->max_ee_init_rd_atom	= attr->max_ee_init_rd_atom;
350 	resp->atomic_cap		= attr->atomic_cap;
351 	resp->max_ee			= attr->max_ee;
352 	resp->max_rdd			= attr->max_rdd;
353 	resp->max_mw			= attr->max_mw;
354 	resp->max_raw_ipv6_qp		= attr->max_raw_ipv6_qp;
355 	resp->max_raw_ethy_qp		= attr->max_raw_ethy_qp;
356 	resp->max_mcast_grp		= attr->max_mcast_grp;
357 	resp->max_mcast_qp_attach	= attr->max_mcast_qp_attach;
358 	resp->max_total_mcast_qp_attach	= attr->max_total_mcast_qp_attach;
359 	resp->max_ah			= attr->max_ah;
360 	resp->max_fmr			= attr->max_fmr;
361 	resp->max_map_per_fmr		= attr->max_map_per_fmr;
362 	resp->max_srq			= attr->max_srq;
363 	resp->max_srq_wr		= attr->max_srq_wr;
364 	resp->max_srq_sge		= attr->max_srq_sge;
365 	resp->max_pkeys			= attr->max_pkeys;
366 	resp->local_ca_ack_delay	= attr->local_ca_ack_delay;
367 	resp->phys_port_cnt		= ib_dev->phys_port_cnt;
368 }
369 
370 static int ib_uverbs_query_device(struct uverbs_attr_bundle *attrs)
371 {
372 	struct ib_uverbs_query_device      cmd;
373 	struct ib_uverbs_query_device_resp resp;
374 	struct ib_ucontext *ucontext;
375 	int ret;
376 
377 	ucontext = ib_uverbs_get_ucontext(attrs);
378 	if (IS_ERR(ucontext))
379 		return PTR_ERR(ucontext);
380 
381 	ret = uverbs_request(attrs, &cmd, sizeof(cmd));
382 	if (ret)
383 		return ret;
384 
385 	memset(&resp, 0, sizeof resp);
386 	copy_query_dev_fields(ucontext, &resp, &ucontext->device->attrs);
387 
388 	return uverbs_response(attrs, &resp, sizeof(resp));
389 }
390 
391 static int ib_uverbs_query_port(struct uverbs_attr_bundle *attrs)
392 {
393 	struct ib_uverbs_query_port      cmd;
394 	struct ib_uverbs_query_port_resp resp;
395 	struct ib_port_attr              attr;
396 	int                              ret;
397 	struct ib_ucontext *ucontext;
398 	struct ib_device *ib_dev;
399 
400 	ucontext = ib_uverbs_get_ucontext(attrs);
401 	if (IS_ERR(ucontext))
402 		return PTR_ERR(ucontext);
403 	ib_dev = ucontext->device;
404 
405 	ret = uverbs_request(attrs, &cmd, sizeof(cmd));
406 	if (ret)
407 		return ret;
408 
409 	ret = ib_query_port(ib_dev, cmd.port_num, &attr);
410 	if (ret)
411 		return ret;
412 
413 	memset(&resp, 0, sizeof resp);
414 	copy_port_attr_to_resp(&attr, &resp, ib_dev, cmd.port_num);
415 
416 	return uverbs_response(attrs, &resp, sizeof(resp));
417 }
418 
419 static int ib_uverbs_alloc_pd(struct uverbs_attr_bundle *attrs)
420 {
421 	struct ib_uverbs_alloc_pd      cmd;
422 	struct ib_uverbs_alloc_pd_resp resp;
423 	struct ib_uobject             *uobj;
424 	struct ib_pd                  *pd;
425 	int                            ret;
426 	struct ib_device *ib_dev;
427 
428 	ret = uverbs_request(attrs, &cmd, sizeof(cmd));
429 	if (ret)
430 		return ret;
431 
432 	uobj = uobj_alloc(UVERBS_OBJECT_PD, attrs, &ib_dev);
433 	if (IS_ERR(uobj))
434 		return PTR_ERR(uobj);
435 
436 	pd = rdma_zalloc_drv_obj(ib_dev, ib_pd);
437 	if (!pd) {
438 		ret = -ENOMEM;
439 		goto err;
440 	}
441 
442 	pd->device  = ib_dev;
443 	pd->uobject = uobj;
444 	pd->__internal_mr = NULL;
445 	atomic_set(&pd->usecnt, 0);
446 
447 	ret = ib_dev->alloc_pd(pd, &attrs->driver_udata);
448 	if (ret)
449 		goto err_alloc;
450 
451 	uobj->object = pd;
452 	memset(&resp, 0, sizeof resp);
453 	resp.pd_handle = uobj->id;
454 
455 	ret = uverbs_response(attrs, &resp, sizeof(resp));
456 	if (ret)
457 		goto err_copy;
458 
459 	rdma_alloc_commit_uobject(uobj, attrs);
460 	return 0;
461 
462 err_copy:
463 	ib_dealloc_pd_user(pd, uverbs_get_cleared_udata(attrs));
464 	pd = NULL;
465 err_alloc:
466 	kfree(pd);
467 err:
468 	uobj_alloc_abort(uobj, attrs);
469 	return ret;
470 }
471 
472 static int ib_uverbs_dealloc_pd(struct uverbs_attr_bundle *attrs)
473 {
474 	struct ib_uverbs_dealloc_pd cmd;
475 	int ret;
476 
477 	ret = uverbs_request(attrs, &cmd, sizeof(cmd));
478 	if (ret)
479 		return ret;
480 
481 	return uobj_perform_destroy(UVERBS_OBJECT_PD, cmd.pd_handle, attrs);
482 }
483 
484 struct xrcd_table_entry {
485 	struct rb_node  node;
486 	struct ib_xrcd *xrcd;
487 	struct vnode   *vnode;
488 };
489 
490 static int xrcd_table_insert(struct ib_uverbs_device *dev,
491 			    struct vnode *vnode,
492 			    struct ib_xrcd *xrcd)
493 {
494 	struct xrcd_table_entry *entry, *scan;
495 	struct rb_node **p = &dev->xrcd_tree.rb_node;
496 	struct rb_node *parent = NULL;
497 
498 	entry = kmalloc(sizeof *entry, GFP_KERNEL);
499 	if (!entry)
500 		return -ENOMEM;
501 
502 	entry->xrcd  = xrcd;
503 	entry->vnode = vnode;
504 
505 	while (*p) {
506 		parent = *p;
507 		scan = rb_entry(parent, struct xrcd_table_entry, node);
508 
509 		if ((uintptr_t)vnode < (uintptr_t)scan->vnode) {
510 			p = &(*p)->rb_left;
511 		} else if ((uintptr_t)vnode > (uintptr_t)scan->vnode) {
512 			p = &(*p)->rb_right;
513 		} else {
514 			kfree(entry);
515 			return -EEXIST;
516 		}
517 	}
518 
519 	rb_link_node(&entry->node, parent, p);
520 	rb_insert_color(&entry->node, &dev->xrcd_tree);
521 	vrefact(vnode);
522 	return 0;
523 }
524 
525 static struct xrcd_table_entry *xrcd_table_search(struct ib_uverbs_device *dev,
526 						  struct vnode *vnode)
527 {
528 	struct xrcd_table_entry *entry;
529 	struct rb_node *p = dev->xrcd_tree.rb_node;
530 
531 	while (p) {
532 		entry = rb_entry(p, struct xrcd_table_entry, node);
533 
534 		if ((uintptr_t)vnode < (uintptr_t)entry->vnode)
535 			p = p->rb_left;
536 		else if ((uintptr_t)vnode > (uintptr_t)entry->vnode)
537 			p = p->rb_right;
538 		else
539 			return entry;
540 	}
541 
542 	return NULL;
543 }
544 
545 static struct ib_xrcd *find_xrcd(struct ib_uverbs_device *dev, struct vnode *vnode)
546 {
547 	struct xrcd_table_entry *entry;
548 
549 	entry = xrcd_table_search(dev, vnode);
550 	if (!entry)
551 		return NULL;
552 
553 	return entry->xrcd;
554 }
555 
556 static void xrcd_table_delete(struct ib_uverbs_device *dev,
557 			      struct vnode *vnode)
558 {
559 	struct xrcd_table_entry *entry;
560 
561 	entry = xrcd_table_search(dev, vnode);
562 	if (entry) {
563 		vrele(vnode);
564 		rb_erase(&entry->node, &dev->xrcd_tree);
565 		kfree(entry);
566 	}
567 }
568 
569 static int ib_uverbs_open_xrcd(struct uverbs_attr_bundle *attrs)
570 {
571 	struct ib_uverbs_device *ibudev = attrs->ufile->device;
572 	struct ib_uverbs_open_xrcd	cmd;
573 	struct ib_uverbs_open_xrcd_resp	resp;
574 	struct ib_uxrcd_object         *obj;
575 	struct ib_xrcd                 *xrcd = NULL;
576 	struct vnode                   *vnode = NULL;
577 	int				ret = 0;
578 	int				new_xrcd = 0;
579 	struct ib_device *ib_dev;
580 
581 	ret = uverbs_request(attrs, &cmd, sizeof(cmd));
582 	if (ret)
583 		return ret;
584 
585 	mutex_lock(&ibudev->xrcd_tree_mutex);
586 
587 	if (cmd.fd != -1) {
588 		/* search for file descriptor */
589 		ret = -fgetvp(curthread, cmd.fd, &cap_no_rights, &vnode);
590 		if (ret != 0)
591 			goto err_tree_mutex_unlock;
592 
593 		xrcd = find_xrcd(ibudev, vnode);
594 		if (!xrcd && !(cmd.oflags & O_CREAT)) {
595 			/* no file descriptor. Need CREATE flag */
596 			ret = -EAGAIN;
597 			goto err_tree_mutex_unlock;
598 		}
599 
600 		if (xrcd && cmd.oflags & O_EXCL) {
601 			ret = -EINVAL;
602 			goto err_tree_mutex_unlock;
603 		}
604 	}
605 
606 	obj = (struct ib_uxrcd_object *)uobj_alloc(UVERBS_OBJECT_XRCD, attrs,
607 						   &ib_dev);
608 	if (IS_ERR(obj)) {
609 		ret = PTR_ERR(obj);
610 		goto err_tree_mutex_unlock;
611 	}
612 
613 	if (!xrcd) {
614 		xrcd = ib_dev->alloc_xrcd(ib_dev, &attrs->driver_udata);
615 		if (IS_ERR(xrcd)) {
616 			ret = PTR_ERR(xrcd);
617 			goto err;
618 		}
619 
620 		xrcd->vnode   = vnode;
621 		xrcd->device  = ib_dev;
622 		atomic_set(&xrcd->usecnt, 0);
623 		mutex_init(&xrcd->tgt_qp_mutex);
624 		INIT_LIST_HEAD(&xrcd->tgt_qp_list);
625 		new_xrcd = 1;
626 	}
627 
628 	atomic_set(&obj->refcnt, 0);
629 	obj->uobject.object = xrcd;
630 	memset(&resp, 0, sizeof resp);
631 	resp.xrcd_handle = obj->uobject.id;
632 
633 	if (vnode != NULL) {
634 		if (new_xrcd) {
635 			/* create new vnode/xrcd table entry */
636 			ret = xrcd_table_insert(ibudev, vnode, xrcd);
637 			if (ret)
638 				goto err_dealloc_xrcd;
639 		}
640 		atomic_inc(&xrcd->usecnt);
641 	}
642 
643 	ret = uverbs_response(attrs, &resp, sizeof(resp));
644 	if (ret)
645 		goto err_copy;
646 
647 	if (vnode != NULL)
648 		vrele(vnode);
649 
650 	mutex_unlock(&ibudev->xrcd_tree_mutex);
651 
652 	rdma_alloc_commit_uobject(&obj->uobject, attrs);
653 	return 0;
654 
655 err_copy:
656 	if (vnode != NULL) {
657 		if (new_xrcd)
658 			xrcd_table_delete(ibudev, vnode);
659 		atomic_dec(&xrcd->usecnt);
660 	}
661 
662 err_dealloc_xrcd:
663 	ib_dealloc_xrcd(xrcd, uverbs_get_cleared_udata(attrs));
664 
665 err:
666 	uobj_alloc_abort(&obj->uobject, attrs);
667 
668 err_tree_mutex_unlock:
669 	if (vnode != NULL)
670 		vrele(vnode);
671 
672 	mutex_unlock(&ibudev->xrcd_tree_mutex);
673 
674 	return ret;
675 }
676 
677 static int ib_uverbs_close_xrcd(struct uverbs_attr_bundle *attrs)
678 {
679 	struct ib_uverbs_close_xrcd cmd;
680 	int ret;
681 
682 	ret = uverbs_request(attrs, &cmd, sizeof(cmd));
683 	if (ret)
684 		return ret;
685 
686 	return uobj_perform_destroy(UVERBS_OBJECT_XRCD, cmd.xrcd_handle, attrs);
687 }
688 
689 int ib_uverbs_dealloc_xrcd(struct ib_uobject *uobject, struct ib_xrcd *xrcd,
690 			   enum rdma_remove_reason why,
691 			   struct uverbs_attr_bundle *attrs)
692 {
693 	struct vnode *vnode;
694 	int ret;
695 	struct ib_uverbs_device *dev = attrs->ufile->device;
696 
697 	vnode = xrcd->vnode;
698 	if (vnode && !atomic_dec_and_test(&xrcd->usecnt))
699 		return 0;
700 
701 	ret = ib_dealloc_xrcd(xrcd, &attrs->driver_udata);
702 
703 	if (ib_is_destroy_retryable(ret, why, uobject)) {
704 		atomic_inc(&xrcd->usecnt);
705 		return ret;
706 	}
707 
708 	if (vnode)
709 		xrcd_table_delete(dev, vnode);
710 
711 	return ret;
712 }
713 
714 static int ib_uverbs_reg_mr(struct uverbs_attr_bundle *attrs)
715 {
716 	struct ib_uverbs_reg_mr      cmd;
717 	struct ib_uverbs_reg_mr_resp resp;
718 	struct ib_uobject           *uobj;
719 	struct ib_pd                *pd;
720 	struct ib_mr                *mr;
721 	int                          ret;
722 	struct ib_device *ib_dev;
723 
724 	ret = uverbs_request(attrs, &cmd, sizeof(cmd));
725 	if (ret)
726 		return ret;
727 
728 	if ((cmd.start & ~PAGE_MASK) != (cmd.hca_va & ~PAGE_MASK))
729 		return -EINVAL;
730 
731 	ret = ib_check_mr_access(cmd.access_flags);
732 	if (ret)
733 		return ret;
734 
735 	uobj = uobj_alloc(UVERBS_OBJECT_MR, attrs, &ib_dev);
736 	if (IS_ERR(uobj))
737 		return PTR_ERR(uobj);
738 
739 	pd = uobj_get_obj_read(pd, UVERBS_OBJECT_PD, cmd.pd_handle, attrs);
740 	if (!pd) {
741 		ret = -EINVAL;
742 		goto err_free;
743 	}
744 
745 	if (cmd.access_flags & IB_ACCESS_ON_DEMAND) {
746 		if (!(pd->device->attrs.device_cap_flags &
747 		      IB_DEVICE_ON_DEMAND_PAGING)) {
748 			pr_debug("ODP support not available\n");
749 			ret = -EINVAL;
750 			goto err_put;
751 		}
752 	}
753 
754 	mr = pd->device->reg_user_mr(pd, cmd.start, cmd.length, cmd.hca_va,
755 					 cmd.access_flags,
756 					 &attrs->driver_udata);
757 	if (IS_ERR(mr)) {
758 		ret = PTR_ERR(mr);
759 		goto err_put;
760 	}
761 
762 	mr->device  = pd->device;
763 	mr->pd      = pd;
764 	mr->type    = IB_MR_TYPE_USER;
765 	mr->dm	    = NULL;
766 	mr->sig_attrs = NULL;
767 	mr->uobject = uobj;
768 	atomic_inc(&pd->usecnt);
769 
770 	uobj->object = mr;
771 
772 	memset(&resp, 0, sizeof resp);
773 	resp.lkey      = mr->lkey;
774 	resp.rkey      = mr->rkey;
775 	resp.mr_handle = uobj->id;
776 
777 	ret = uverbs_response(attrs, &resp, sizeof(resp));
778 	if (ret)
779 		goto err_copy;
780 
781 	uobj_put_obj_read(pd);
782 
783 	rdma_alloc_commit_uobject(uobj, attrs);
784 	return 0;
785 
786 err_copy:
787 	ib_dereg_mr_user(mr, uverbs_get_cleared_udata(attrs));
788 
789 err_put:
790 	uobj_put_obj_read(pd);
791 
792 err_free:
793 	uobj_alloc_abort(uobj, attrs);
794 	return ret;
795 }
796 
797 static int ib_uverbs_rereg_mr(struct uverbs_attr_bundle *attrs)
798 {
799 	struct ib_uverbs_rereg_mr      cmd;
800 	struct ib_uverbs_rereg_mr_resp resp;
801 	struct ib_pd                *pd = NULL;
802 	struct ib_mr                *mr;
803 	struct ib_pd		    *old_pd;
804 	int                          ret;
805 	struct ib_uobject	    *uobj;
806 
807 	ret = uverbs_request(attrs, &cmd, sizeof(cmd));
808 	if (ret)
809 		return ret;
810 
811 	if (cmd.flags & ~IB_MR_REREG_SUPPORTED || !cmd.flags)
812 		return -EINVAL;
813 
814 	if ((cmd.flags & IB_MR_REREG_TRANS) &&
815 	    (!cmd.start || !cmd.hca_va || 0 >= cmd.length ||
816 	     (cmd.start & ~PAGE_MASK) != (cmd.hca_va & ~PAGE_MASK)))
817 			return -EINVAL;
818 
819 	uobj = uobj_get_write(UVERBS_OBJECT_MR, cmd.mr_handle, attrs);
820 	if (IS_ERR(uobj))
821 		return PTR_ERR(uobj);
822 
823 	mr = uobj->object;
824 
825 	if (mr->dm) {
826 		ret = -EINVAL;
827 		goto put_uobjs;
828 	}
829 
830 	if (cmd.flags & IB_MR_REREG_ACCESS) {
831 		ret = ib_check_mr_access(cmd.access_flags);
832 		if (ret)
833 			goto put_uobjs;
834 	}
835 
836 	if (cmd.flags & IB_MR_REREG_PD) {
837 		pd = uobj_get_obj_read(pd, UVERBS_OBJECT_PD, cmd.pd_handle,
838 				       attrs);
839 		if (!pd) {
840 			ret = -EINVAL;
841 			goto put_uobjs;
842 		}
843 	}
844 
845 	old_pd = mr->pd;
846 	ret = mr->device->rereg_user_mr(mr, cmd.flags, cmd.start,
847 					    cmd.length, cmd.hca_va,
848 					    cmd.access_flags, pd,
849 					    &attrs->driver_udata);
850 	if (ret)
851 		goto put_uobj_pd;
852 
853 	if (cmd.flags & IB_MR_REREG_PD) {
854 		atomic_inc(&pd->usecnt);
855 		mr->pd = pd;
856 		atomic_dec(&old_pd->usecnt);
857 	}
858 
859 	memset(&resp, 0, sizeof(resp));
860 	resp.lkey      = mr->lkey;
861 	resp.rkey      = mr->rkey;
862 
863 	ret = uverbs_response(attrs, &resp, sizeof(resp));
864 
865 put_uobj_pd:
866 	if (cmd.flags & IB_MR_REREG_PD)
867 		uobj_put_obj_read(pd);
868 
869 put_uobjs:
870 	uobj_put_write(uobj);
871 
872 	return ret;
873 }
874 
875 static int ib_uverbs_dereg_mr(struct uverbs_attr_bundle *attrs)
876 {
877 	struct ib_uverbs_dereg_mr cmd;
878 	int ret;
879 
880 	ret = uverbs_request(attrs, &cmd, sizeof(cmd));
881 	if (ret)
882 		return ret;
883 
884 	return uobj_perform_destroy(UVERBS_OBJECT_MR, cmd.mr_handle, attrs);
885 }
886 
887 static int ib_uverbs_alloc_mw(struct uverbs_attr_bundle *attrs)
888 {
889 	struct ib_uverbs_alloc_mw      cmd;
890 	struct ib_uverbs_alloc_mw_resp resp;
891 	struct ib_uobject             *uobj;
892 	struct ib_pd                  *pd;
893 	struct ib_mw                  *mw;
894 	int                            ret;
895 	struct ib_device *ib_dev;
896 
897 	ret = uverbs_request(attrs, &cmd, sizeof(cmd));
898 	if (ret)
899 		return ret;
900 
901 	uobj = uobj_alloc(UVERBS_OBJECT_MW, attrs, &ib_dev);
902 	if (IS_ERR(uobj))
903 		return PTR_ERR(uobj);
904 
905 	pd = uobj_get_obj_read(pd, UVERBS_OBJECT_PD, cmd.pd_handle, attrs);
906 	if (!pd) {
907 		ret = -EINVAL;
908 		goto err_free;
909 	}
910 
911 	if (cmd.mw_type != IB_MW_TYPE_1 && cmd.mw_type != IB_MW_TYPE_2) {
912 		ret = -EINVAL;
913 		goto err_put;
914 	}
915 
916 	mw = pd->device->alloc_mw(pd, cmd.mw_type, &attrs->driver_udata);
917 	if (IS_ERR(mw)) {
918 		ret = PTR_ERR(mw);
919 		goto err_put;
920 	}
921 
922 	mw->device  = pd->device;
923 	mw->pd      = pd;
924 	mw->uobject = uobj;
925 	atomic_inc(&pd->usecnt);
926 
927 	uobj->object = mw;
928 
929 	memset(&resp, 0, sizeof(resp));
930 	resp.rkey      = mw->rkey;
931 	resp.mw_handle = uobj->id;
932 
933 	ret = uverbs_response(attrs, &resp, sizeof(resp));
934 	if (ret)
935 		goto err_copy;
936 
937 	uobj_put_obj_read(pd);
938 	rdma_alloc_commit_uobject(uobj, attrs);
939 	return 0;
940 
941 err_copy:
942 	uverbs_dealloc_mw(mw);
943 err_put:
944 	uobj_put_obj_read(pd);
945 err_free:
946 	uobj_alloc_abort(uobj, attrs);
947 	return ret;
948 }
949 
950 static int ib_uverbs_dealloc_mw(struct uverbs_attr_bundle *attrs)
951 {
952 	struct ib_uverbs_dealloc_mw cmd;
953 	int ret;
954 
955 	ret = uverbs_request(attrs, &cmd, sizeof(cmd));
956 	if (ret)
957 		return ret;
958 
959 	return uobj_perform_destroy(UVERBS_OBJECT_MW, cmd.mw_handle, attrs);
960 }
961 
962 static int ib_uverbs_create_comp_channel(struct uverbs_attr_bundle *attrs)
963 {
964 	struct ib_uverbs_create_comp_channel	   cmd;
965 	struct ib_uverbs_create_comp_channel_resp  resp;
966 	struct ib_uobject			  *uobj;
967 	struct ib_uverbs_completion_event_file	  *ev_file;
968 	struct ib_device *ib_dev;
969 	int ret;
970 
971 	ret = uverbs_request(attrs, &cmd, sizeof(cmd));
972 	if (ret)
973 		return ret;
974 
975 	uobj = uobj_alloc(UVERBS_OBJECT_COMP_CHANNEL, attrs, &ib_dev);
976 	if (IS_ERR(uobj))
977 		return PTR_ERR(uobj);
978 
979 	resp.fd = uobj->id;
980 
981 	ev_file = container_of(uobj, struct ib_uverbs_completion_event_file,
982 			       uobj);
983 	ib_uverbs_init_event_queue(&ev_file->ev_queue);
984 
985 	ret = uverbs_response(attrs, &resp, sizeof(resp));
986 	if (ret) {
987 		uobj_alloc_abort(uobj, attrs);
988 		return ret;
989 	}
990 
991 	rdma_alloc_commit_uobject(uobj, attrs);
992 	return 0;
993 }
994 
995 static struct ib_ucq_object *create_cq(struct uverbs_attr_bundle *attrs,
996 				       struct ib_uverbs_ex_create_cq *cmd)
997 {
998 	struct ib_ucq_object           *obj;
999 	struct ib_uverbs_completion_event_file    *ev_file = NULL;
1000 	struct ib_cq                   *cq;
1001 	int                             ret;
1002 	struct ib_uverbs_ex_create_cq_resp resp;
1003 	struct ib_cq_init_attr attr = {};
1004 	struct ib_device *ib_dev;
1005 
1006 	if (cmd->comp_vector >= attrs->ufile->device->num_comp_vectors)
1007 		return ERR_PTR(-EINVAL);
1008 
1009 	obj = (struct ib_ucq_object *)uobj_alloc(UVERBS_OBJECT_CQ, attrs,
1010 						 &ib_dev);
1011 	if (IS_ERR(obj))
1012 		return obj;
1013 
1014 	if (cmd->comp_channel >= 0) {
1015 		ev_file = ib_uverbs_lookup_comp_file(cmd->comp_channel, attrs);
1016 		if (IS_ERR(ev_file)) {
1017 			ret = PTR_ERR(ev_file);
1018 			goto err;
1019 		}
1020 	}
1021 
1022 	obj->uevent.uobject.user_handle = cmd->user_handle;
1023 	INIT_LIST_HEAD(&obj->comp_list);
1024 	INIT_LIST_HEAD(&obj->uevent.event_list);
1025 
1026 	attr.cqe = cmd->cqe;
1027 	attr.comp_vector = cmd->comp_vector;
1028 	attr.flags = cmd->flags;
1029 
1030 	cq = rdma_zalloc_drv_obj(ib_dev, ib_cq);
1031 	if (!cq) {
1032 		ret = -ENOMEM;
1033 		goto err_file;
1034 	}
1035 	cq->device        = ib_dev;
1036 	cq->uobject       = obj;
1037 	cq->comp_handler  = ib_uverbs_comp_handler;
1038 	cq->event_handler = ib_uverbs_cq_event_handler;
1039 	cq->cq_context    = ev_file ? &ev_file->ev_queue : NULL;
1040 	atomic_set(&cq->usecnt, 0);
1041 
1042 	ret = ib_dev->create_cq(cq, &attr, &attrs->driver_udata);
1043 	if (ret)
1044 		goto err_free;
1045 
1046 	obj->uevent.uobject.object = cq;
1047 	memset(&resp, 0, sizeof resp);
1048 	resp.base.cq_handle = obj->uevent.uobject.id;
1049 	resp.base.cqe       = cq->cqe;
1050 	resp.response_length = uverbs_response_length(attrs, sizeof(resp));
1051 
1052 	ret = uverbs_response(attrs, &resp, sizeof(resp));
1053 	if (ret)
1054 		goto err_cb;
1055 
1056 	rdma_alloc_commit_uobject(&obj->uevent.uobject, attrs);
1057 	return obj;
1058 
1059 err_cb:
1060 	ib_destroy_cq_user(cq, uverbs_get_cleared_udata(attrs));
1061 	cq = NULL;
1062 err_free:
1063 	kfree(cq);
1064 err_file:
1065 	if (ev_file)
1066 		ib_uverbs_release_ucq(ev_file, obj);
1067 
1068 err:
1069 	uobj_alloc_abort(&obj->uevent.uobject, attrs);
1070 
1071 	return ERR_PTR(ret);
1072 }
1073 
1074 static int ib_uverbs_create_cq(struct uverbs_attr_bundle *attrs)
1075 {
1076 	struct ib_uverbs_create_cq      cmd;
1077 	struct ib_uverbs_ex_create_cq	cmd_ex;
1078 	struct ib_ucq_object           *obj;
1079 	int ret;
1080 
1081 	ret = uverbs_request(attrs, &cmd, sizeof(cmd));
1082 	if (ret)
1083 		return ret;
1084 
1085 	memset(&cmd_ex, 0, sizeof(cmd_ex));
1086 	cmd_ex.user_handle = cmd.user_handle;
1087 	cmd_ex.cqe = cmd.cqe;
1088 	cmd_ex.comp_vector = cmd.comp_vector;
1089 	cmd_ex.comp_channel = cmd.comp_channel;
1090 
1091 	obj = create_cq(attrs, &cmd_ex);
1092 	return PTR_ERR_OR_ZERO(obj);
1093 }
1094 
1095 static int ib_uverbs_ex_create_cq(struct uverbs_attr_bundle *attrs)
1096 {
1097 	struct ib_uverbs_ex_create_cq  cmd;
1098 	struct ib_ucq_object           *obj;
1099 	int ret;
1100 
1101 	ret = uverbs_request(attrs, &cmd, sizeof(cmd));
1102 	if (ret)
1103 		return ret;
1104 
1105 	if (cmd.comp_mask)
1106 		return -EINVAL;
1107 
1108 	if (cmd.reserved)
1109 		return -EINVAL;
1110 
1111 	obj = create_cq(attrs, &cmd);
1112 	return PTR_ERR_OR_ZERO(obj);
1113 }
1114 
1115 static int ib_uverbs_resize_cq(struct uverbs_attr_bundle *attrs)
1116 {
1117 	struct ib_uverbs_resize_cq	cmd;
1118 	struct ib_uverbs_resize_cq_resp	resp = {};
1119 	struct ib_cq			*cq;
1120 	int				ret = -EINVAL;
1121 
1122 	ret = uverbs_request(attrs, &cmd, sizeof(cmd));
1123 	if (ret)
1124 		return ret;
1125 
1126 	cq = uobj_get_obj_read(cq, UVERBS_OBJECT_CQ, cmd.cq_handle, attrs);
1127 	if (!cq)
1128 		return -EINVAL;
1129 
1130 	ret = cq->device->resize_cq(cq, cmd.cqe, &attrs->driver_udata);
1131 	if (ret)
1132 		goto out;
1133 
1134 	resp.cqe = cq->cqe;
1135 
1136 	ret = uverbs_response(attrs, &resp, sizeof(resp));
1137 out:
1138 	rdma_lookup_put_uobject(&cq->uobject->uevent.uobject,
1139 				UVERBS_LOOKUP_READ);
1140 
1141 	return ret;
1142 }
1143 
1144 static int copy_wc_to_user(struct ib_device *ib_dev, void __user *dest,
1145 			   struct ib_wc *wc)
1146 {
1147 	struct ib_uverbs_wc tmp;
1148 
1149 	tmp.wr_id		= wc->wr_id;
1150 	tmp.status		= wc->status;
1151 	tmp.opcode		= wc->opcode;
1152 	tmp.vendor_err		= wc->vendor_err;
1153 	tmp.byte_len		= wc->byte_len;
1154 	tmp.ex.imm_data		= wc->ex.imm_data;
1155 	tmp.qp_num		= wc->qp->qp_num;
1156 	tmp.src_qp		= wc->src_qp;
1157 	tmp.wc_flags		= wc->wc_flags;
1158 	tmp.pkey_index		= wc->pkey_index;
1159 	tmp.slid		= wc->slid;
1160 	tmp.sl			= wc->sl;
1161 	tmp.dlid_path_bits	= wc->dlid_path_bits;
1162 	tmp.port_num		= wc->port_num;
1163 	tmp.reserved		= 0;
1164 
1165 	if (copy_to_user(dest, &tmp, sizeof tmp))
1166 		return -EFAULT;
1167 
1168 	return 0;
1169 }
1170 
1171 static int ib_uverbs_poll_cq(struct uverbs_attr_bundle *attrs)
1172 {
1173 	struct ib_uverbs_poll_cq       cmd;
1174 	struct ib_uverbs_poll_cq_resp  resp;
1175 	u8 __user                     *header_ptr;
1176 	u8 __user                     *data_ptr;
1177 	struct ib_cq                  *cq;
1178 	struct ib_wc                   wc;
1179 	int                            ret;
1180 
1181 	ret = uverbs_request(attrs, &cmd, sizeof(cmd));
1182 	if (ret)
1183 		return ret;
1184 
1185 	cq = uobj_get_obj_read(cq, UVERBS_OBJECT_CQ, cmd.cq_handle, attrs);
1186 	if (!cq)
1187 		return -EINVAL;
1188 
1189 	/* we copy a struct ib_uverbs_poll_cq_resp to user space */
1190 	header_ptr = attrs->ucore.outbuf;
1191 	data_ptr = header_ptr + sizeof resp;
1192 
1193 	memset(&resp, 0, sizeof resp);
1194 	while (resp.count < cmd.ne) {
1195 		ret = ib_poll_cq(cq, 1, &wc);
1196 		if (ret < 0)
1197 			goto out_put;
1198 		if (!ret)
1199 			break;
1200 
1201 		ret = copy_wc_to_user(cq->device, data_ptr, &wc);
1202 		if (ret)
1203 			goto out_put;
1204 
1205 		data_ptr += sizeof(struct ib_uverbs_wc);
1206 		++resp.count;
1207 	}
1208 
1209 	if (copy_to_user(header_ptr, &resp, sizeof resp)) {
1210 		ret = -EFAULT;
1211 		goto out_put;
1212 	}
1213 	ret = 0;
1214 
1215 	if (uverbs_attr_is_valid(attrs, UVERBS_ATTR_CORE_OUT))
1216 		ret = uverbs_output_written(attrs, UVERBS_ATTR_CORE_OUT);
1217 
1218 out_put:
1219 	rdma_lookup_put_uobject(&cq->uobject->uevent.uobject,
1220 				UVERBS_LOOKUP_READ);
1221 	return ret;
1222 }
1223 
1224 static int ib_uverbs_req_notify_cq(struct uverbs_attr_bundle *attrs)
1225 {
1226 	struct ib_uverbs_req_notify_cq cmd;
1227 	struct ib_cq                  *cq;
1228 	int ret;
1229 
1230 	ret = uverbs_request(attrs, &cmd, sizeof(cmd));
1231 	if (ret)
1232 		return ret;
1233 
1234 	cq = uobj_get_obj_read(cq, UVERBS_OBJECT_CQ, cmd.cq_handle, attrs);
1235 	if (!cq)
1236 		return -EINVAL;
1237 
1238 	ib_req_notify_cq(cq, cmd.solicited_only ?
1239 			 IB_CQ_SOLICITED : IB_CQ_NEXT_COMP);
1240 
1241 	rdma_lookup_put_uobject(&cq->uobject->uevent.uobject,
1242 				UVERBS_LOOKUP_READ);
1243 	return 0;
1244 }
1245 
1246 static int ib_uverbs_destroy_cq(struct uverbs_attr_bundle *attrs)
1247 {
1248 	struct ib_uverbs_destroy_cq      cmd;
1249 	struct ib_uverbs_destroy_cq_resp resp;
1250 	struct ib_uobject		*uobj;
1251 	struct ib_ucq_object        	*obj;
1252 	int ret;
1253 
1254 	ret = uverbs_request(attrs, &cmd, sizeof(cmd));
1255 	if (ret)
1256 		return ret;
1257 
1258 	uobj = uobj_get_destroy(UVERBS_OBJECT_CQ, cmd.cq_handle, attrs);
1259 	if (IS_ERR(uobj))
1260 		return PTR_ERR(uobj);
1261 
1262 	obj = container_of(uobj, struct ib_ucq_object, uevent.uobject);
1263 	memset(&resp, 0, sizeof(resp));
1264 	resp.comp_events_reported  = obj->comp_events_reported;
1265 	resp.async_events_reported = obj->uevent.events_reported;
1266 
1267 	uobj_put_destroy(uobj);
1268 
1269 	return uverbs_response(attrs, &resp, sizeof(resp));
1270 }
1271 
1272 static int create_qp(struct uverbs_attr_bundle *attrs,
1273 		     struct ib_uverbs_ex_create_qp *cmd)
1274 {
1275 	struct ib_uqp_object		*obj;
1276 	struct ib_device		*device;
1277 	struct ib_pd			*pd = NULL;
1278 	struct ib_xrcd			*xrcd = NULL;
1279 	struct ib_uobject		*xrcd_uobj = ERR_PTR(-ENOENT);
1280 	struct ib_cq			*scq = NULL, *rcq = NULL;
1281 	struct ib_srq			*srq = NULL;
1282 	struct ib_qp			*qp;
1283 	struct ib_qp_init_attr		attr = {};
1284 	struct ib_uverbs_ex_create_qp_resp resp;
1285 	int				ret;
1286 	struct ib_rwq_ind_table *ind_tbl = NULL;
1287 	bool has_sq = true;
1288 	struct ib_device *ib_dev;
1289 
1290 	if (cmd->qp_type == IB_QPT_RAW_PACKET && priv_check(curthread, PRIV_NET_RAW) != 0)
1291 		return -EPERM;
1292 
1293 	obj = (struct ib_uqp_object *)uobj_alloc(UVERBS_OBJECT_QP, attrs,
1294 						 &ib_dev);
1295 	if (IS_ERR(obj))
1296 		return PTR_ERR(obj);
1297 	obj->uxrcd = NULL;
1298 	obj->uevent.uobject.user_handle = cmd->user_handle;
1299 	mutex_init(&obj->mcast_lock);
1300 
1301 	if (cmd->comp_mask & IB_UVERBS_CREATE_QP_MASK_IND_TABLE) {
1302 		ind_tbl = uobj_get_obj_read(rwq_ind_table,
1303 					    UVERBS_OBJECT_RWQ_IND_TBL,
1304 					    cmd->rwq_ind_tbl_handle, attrs);
1305 		if (!ind_tbl) {
1306 			ret = -EINVAL;
1307 			goto err_put;
1308 		}
1309 
1310 		attr.rwq_ind_tbl = ind_tbl;
1311 	}
1312 
1313 	if (ind_tbl && (cmd->max_recv_wr || cmd->max_recv_sge || cmd->is_srq)) {
1314 		ret = -EINVAL;
1315 		goto err_put;
1316 	}
1317 
1318 	if (ind_tbl && !cmd->max_send_wr)
1319 		has_sq = false;
1320 
1321 	if (cmd->qp_type == IB_QPT_XRC_TGT) {
1322 		xrcd_uobj = uobj_get_read(UVERBS_OBJECT_XRCD, cmd->pd_handle,
1323 					  attrs);
1324 
1325 		if (IS_ERR(xrcd_uobj)) {
1326 			ret = -EINVAL;
1327 			goto err_put;
1328 		}
1329 
1330 		xrcd = (struct ib_xrcd *)xrcd_uobj->object;
1331 		if (!xrcd) {
1332 			ret = -EINVAL;
1333 			goto err_put;
1334 		}
1335 		device = xrcd->device;
1336 	} else {
1337 		if (cmd->qp_type == IB_QPT_XRC_INI) {
1338 			cmd->max_recv_wr = 0;
1339 			cmd->max_recv_sge = 0;
1340 		} else {
1341 			if (cmd->is_srq) {
1342 				srq = uobj_get_obj_read(srq, UVERBS_OBJECT_SRQ,
1343 							cmd->srq_handle, attrs);
1344 				if (!srq || srq->srq_type == IB_SRQT_XRC) {
1345 					ret = -EINVAL;
1346 					goto err_put;
1347 				}
1348 			}
1349 
1350 			if (!ind_tbl) {
1351 				if (cmd->recv_cq_handle != cmd->send_cq_handle) {
1352 					rcq = uobj_get_obj_read(
1353 						cq, UVERBS_OBJECT_CQ,
1354 						cmd->recv_cq_handle, attrs);
1355 					if (!rcq) {
1356 						ret = -EINVAL;
1357 						goto err_put;
1358 					}
1359 				}
1360 			}
1361 		}
1362 
1363 		if (has_sq)
1364 			scq = uobj_get_obj_read(cq, UVERBS_OBJECT_CQ,
1365 						cmd->send_cq_handle, attrs);
1366 		if (!ind_tbl)
1367 			rcq = rcq ?: scq;
1368 		pd = uobj_get_obj_read(pd, UVERBS_OBJECT_PD, cmd->pd_handle,
1369 				       attrs);
1370 		if (!pd || (!scq && has_sq)) {
1371 			ret = -EINVAL;
1372 			goto err_put;
1373 		}
1374 
1375 		device = pd->device;
1376 	}
1377 
1378 	attr.event_handler = ib_uverbs_qp_event_handler;
1379 	attr.send_cq       = scq;
1380 	attr.recv_cq       = rcq;
1381 	attr.srq           = srq;
1382 	attr.xrcd	   = xrcd;
1383 	attr.sq_sig_type   = cmd->sq_sig_all ? IB_SIGNAL_ALL_WR :
1384 					      IB_SIGNAL_REQ_WR;
1385 	attr.qp_type       = cmd->qp_type;
1386 	attr.create_flags  = 0;
1387 
1388 	attr.cap.max_send_wr     = cmd->max_send_wr;
1389 	attr.cap.max_recv_wr     = cmd->max_recv_wr;
1390 	attr.cap.max_send_sge    = cmd->max_send_sge;
1391 	attr.cap.max_recv_sge    = cmd->max_recv_sge;
1392 	attr.cap.max_inline_data = cmd->max_inline_data;
1393 
1394 	INIT_LIST_HEAD(&obj->uevent.event_list);
1395 	INIT_LIST_HEAD(&obj->mcast_list);
1396 
1397 	attr.create_flags = cmd->create_flags;
1398 	if (attr.create_flags & ~(IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK |
1399 				IB_QP_CREATE_CROSS_CHANNEL |
1400 				IB_QP_CREATE_MANAGED_SEND |
1401 				IB_QP_CREATE_MANAGED_RECV |
1402 				IB_QP_CREATE_SCATTER_FCS |
1403 				IB_QP_CREATE_CVLAN_STRIPPING |
1404 				IB_QP_CREATE_SOURCE_QPN |
1405 				IB_QP_CREATE_PCI_WRITE_END_PADDING)) {
1406 		ret = -EINVAL;
1407 		goto err_put;
1408 	}
1409 
1410 	if (attr.create_flags & IB_QP_CREATE_SOURCE_QPN) {
1411 		if (priv_check(curthread, PRIV_NET_RAW)) {
1412 			ret = -EPERM;
1413 			goto err_put;
1414 		}
1415 
1416 		attr.source_qpn = cmd->source_qpn;
1417 	}
1418 
1419 	if (cmd->qp_type == IB_QPT_XRC_TGT)
1420 		qp = ib_create_qp(pd, &attr);
1421 	else
1422 		qp = _ib_create_qp(device, pd, &attr, &attrs->driver_udata,
1423 				   obj);
1424 
1425 	if (IS_ERR(qp)) {
1426 		ret = PTR_ERR(qp);
1427 		goto err_put;
1428 	}
1429 
1430 	if (cmd->qp_type != IB_QPT_XRC_TGT) {
1431 		atomic_inc(&pd->usecnt);
1432 		if (attr.send_cq)
1433 			atomic_inc(&attr.send_cq->usecnt);
1434 		if (attr.recv_cq)
1435 			atomic_inc(&attr.recv_cq->usecnt);
1436 		if (attr.srq)
1437 			atomic_inc(&attr.srq->usecnt);
1438 		if (ind_tbl)
1439 			atomic_inc(&ind_tbl->usecnt);
1440 	} else {
1441 		/* It is done in _ib_create_qp for other QP types */
1442 		qp->uobject = obj;
1443 	}
1444 
1445 	obj->uevent.uobject.object = qp;
1446 
1447 	memset(&resp, 0, sizeof resp);
1448 	resp.base.qpn             = qp->qp_num;
1449 	resp.base.qp_handle       = obj->uevent.uobject.id;
1450 	resp.base.max_recv_sge    = attr.cap.max_recv_sge;
1451 	resp.base.max_send_sge    = attr.cap.max_send_sge;
1452 	resp.base.max_recv_wr     = attr.cap.max_recv_wr;
1453 	resp.base.max_send_wr     = attr.cap.max_send_wr;
1454 	resp.base.max_inline_data = attr.cap.max_inline_data;
1455 	resp.response_length = uverbs_response_length(attrs, sizeof(resp));
1456 
1457 	ret = uverbs_response(attrs, &resp, sizeof(resp));
1458 	if (ret)
1459 		goto err_cb;
1460 
1461 	if (xrcd) {
1462 		obj->uxrcd = container_of(xrcd_uobj, struct ib_uxrcd_object,
1463 					  uobject);
1464 		atomic_inc(&obj->uxrcd->refcnt);
1465 		uobj_put_read(xrcd_uobj);
1466 	}
1467 
1468 	if (pd)
1469 		uobj_put_obj_read(pd);
1470 	if (scq)
1471 		rdma_lookup_put_uobject(&scq->uobject->uevent.uobject,
1472 					UVERBS_LOOKUP_READ);
1473 	if (rcq && rcq != scq)
1474 		rdma_lookup_put_uobject(&rcq->uobject->uevent.uobject,
1475 					UVERBS_LOOKUP_READ);
1476 	if (srq)
1477 		rdma_lookup_put_uobject(&srq->uobject->uevent.uobject,
1478 					UVERBS_LOOKUP_READ);
1479 	if (ind_tbl)
1480 		uobj_put_obj_read(ind_tbl);
1481 
1482 	rdma_alloc_commit_uobject(&obj->uevent.uobject, attrs);
1483 	return 0;
1484 err_cb:
1485 	ib_destroy_qp_user(qp, uverbs_get_cleared_udata(attrs));
1486 
1487 err_put:
1488 	if (!IS_ERR(xrcd_uobj))
1489 		uobj_put_read(xrcd_uobj);
1490 	if (pd)
1491 		uobj_put_obj_read(pd);
1492 	if (scq)
1493 		rdma_lookup_put_uobject(&scq->uobject->uevent.uobject,
1494 					UVERBS_LOOKUP_READ);
1495 	if (rcq && rcq != scq)
1496 		rdma_lookup_put_uobject(&rcq->uobject->uevent.uobject,
1497 					UVERBS_LOOKUP_READ);
1498 	if (srq)
1499 		rdma_lookup_put_uobject(&srq->uobject->uevent.uobject,
1500 					UVERBS_LOOKUP_READ);
1501 	if (ind_tbl)
1502 		uobj_put_obj_read(ind_tbl);
1503 
1504 	uobj_alloc_abort(&obj->uevent.uobject, attrs);
1505 	return ret;
1506 }
1507 
1508 static int ib_uverbs_create_qp(struct uverbs_attr_bundle *attrs)
1509 {
1510 	struct ib_uverbs_create_qp      cmd;
1511 	struct ib_uverbs_ex_create_qp	cmd_ex;
1512 	int ret;
1513 
1514 	ret = uverbs_request(attrs, &cmd, sizeof(cmd));
1515 	if (ret)
1516 		return ret;
1517 
1518 	memset(&cmd_ex, 0, sizeof(cmd_ex));
1519 	cmd_ex.user_handle = cmd.user_handle;
1520 	cmd_ex.pd_handle = cmd.pd_handle;
1521 	cmd_ex.send_cq_handle = cmd.send_cq_handle;
1522 	cmd_ex.recv_cq_handle = cmd.recv_cq_handle;
1523 	cmd_ex.srq_handle = cmd.srq_handle;
1524 	cmd_ex.max_send_wr = cmd.max_send_wr;
1525 	cmd_ex.max_recv_wr = cmd.max_recv_wr;
1526 	cmd_ex.max_send_sge = cmd.max_send_sge;
1527 	cmd_ex.max_recv_sge = cmd.max_recv_sge;
1528 	cmd_ex.max_inline_data = cmd.max_inline_data;
1529 	cmd_ex.sq_sig_all = cmd.sq_sig_all;
1530 	cmd_ex.qp_type = cmd.qp_type;
1531 	cmd_ex.is_srq = cmd.is_srq;
1532 
1533 	return create_qp(attrs, &cmd_ex);
1534 }
1535 
1536 static int ib_uverbs_ex_create_qp(struct uverbs_attr_bundle *attrs)
1537 {
1538 	struct ib_uverbs_ex_create_qp cmd;
1539 	int ret;
1540 
1541 	ret = uverbs_request(attrs, &cmd, sizeof(cmd));
1542 	if (ret)
1543 		return ret;
1544 
1545 	if (cmd.comp_mask & ~IB_UVERBS_CREATE_QP_SUP_COMP_MASK)
1546 		return -EINVAL;
1547 
1548 	if (cmd.reserved)
1549 		return -EINVAL;
1550 
1551 	return create_qp(attrs, &cmd);
1552 }
1553 
1554 static int ib_uverbs_open_qp(struct uverbs_attr_bundle *attrs)
1555 {
1556 	struct ib_uverbs_open_qp        cmd;
1557 	struct ib_uverbs_create_qp_resp resp;
1558 	struct ib_uqp_object           *obj;
1559 	struct ib_xrcd		       *xrcd;
1560 	struct ib_uobject	       *uninitialized_var(xrcd_uobj);
1561 	struct ib_qp                   *qp;
1562 	struct ib_qp_open_attr          attr = {};
1563 	int ret;
1564 	struct ib_device *ib_dev;
1565 
1566 	ret = uverbs_request(attrs, &cmd, sizeof(cmd));
1567 	if (ret)
1568 		return ret;
1569 
1570 	obj = (struct ib_uqp_object *)uobj_alloc(UVERBS_OBJECT_QP, attrs,
1571 						 &ib_dev);
1572 	if (IS_ERR(obj))
1573 		return PTR_ERR(obj);
1574 
1575 	xrcd_uobj = uobj_get_read(UVERBS_OBJECT_XRCD, cmd.pd_handle, attrs);
1576 	if (IS_ERR(xrcd_uobj)) {
1577 		ret = -EINVAL;
1578 		goto err_put;
1579 	}
1580 
1581 	xrcd = (struct ib_xrcd *)xrcd_uobj->object;
1582 	if (!xrcd) {
1583 		ret = -EINVAL;
1584 		goto err_xrcd;
1585 	}
1586 
1587 	attr.event_handler = ib_uverbs_qp_event_handler;
1588 	attr.qp_num        = cmd.qpn;
1589 	attr.qp_type       = cmd.qp_type;
1590 
1591 	INIT_LIST_HEAD(&obj->uevent.event_list);
1592 	INIT_LIST_HEAD(&obj->mcast_list);
1593 
1594 	qp = ib_open_qp(xrcd, &attr);
1595 	if (IS_ERR(qp)) {
1596 		ret = PTR_ERR(qp);
1597 		goto err_xrcd;
1598 	}
1599 
1600 	obj->uevent.uobject.object = qp;
1601 	obj->uevent.uobject.user_handle = cmd.user_handle;
1602 
1603 	memset(&resp, 0, sizeof resp);
1604 	resp.qpn       = qp->qp_num;
1605 	resp.qp_handle = obj->uevent.uobject.id;
1606 
1607 	ret = uverbs_response(attrs, &resp, sizeof(resp));
1608 	if (ret)
1609 		goto err_destroy;
1610 
1611 	obj->uxrcd = container_of(xrcd_uobj, struct ib_uxrcd_object, uobject);
1612 	atomic_inc(&obj->uxrcd->refcnt);
1613 	qp->uobject = obj;
1614 	uobj_put_read(xrcd_uobj);
1615 
1616 	rdma_alloc_commit_uobject(&obj->uevent.uobject, attrs);
1617 	return 0;
1618 
1619 err_destroy:
1620 	ib_destroy_qp_user(qp, uverbs_get_cleared_udata(attrs));
1621 err_xrcd:
1622 	uobj_put_read(xrcd_uobj);
1623 err_put:
1624 	uobj_alloc_abort(&obj->uevent.uobject, attrs);
1625 	return ret;
1626 }
1627 
1628 static void copy_ah_attr_to_uverbs(struct ib_uverbs_qp_dest *uverb_attr,
1629 				   struct ib_ah_attr *rdma_attr)
1630 {
1631 	uverb_attr->dlid              = rdma_attr->dlid;
1632 	uverb_attr->sl                = rdma_attr->sl;
1633 	uverb_attr->src_path_bits     = rdma_attr->src_path_bits;
1634 	uverb_attr->static_rate       = rdma_attr->static_rate;
1635 	uverb_attr->is_global         = !!(rdma_attr->ah_flags & IB_AH_GRH);
1636 	if (uverb_attr->is_global) {
1637 		const struct ib_global_route *grh = &rdma_attr->grh;
1638 
1639 		memcpy(uverb_attr->dgid, grh->dgid.raw, 16);
1640 		uverb_attr->flow_label        = grh->flow_label;
1641 		uverb_attr->sgid_index        = grh->sgid_index;
1642 		uverb_attr->hop_limit         = grh->hop_limit;
1643 		uverb_attr->traffic_class     = grh->traffic_class;
1644 	}
1645 	uverb_attr->port_num          = rdma_attr->port_num;
1646 }
1647 
1648 static int ib_uverbs_query_qp(struct uverbs_attr_bundle *attrs)
1649 {
1650 	struct ib_uverbs_query_qp      cmd;
1651 	struct ib_uverbs_query_qp_resp resp;
1652 	struct ib_qp                   *qp;
1653 	struct ib_qp_attr              *attr;
1654 	struct ib_qp_init_attr         *init_attr;
1655 	int                            ret;
1656 
1657 	ret = uverbs_request(attrs, &cmd, sizeof(cmd));
1658 	if (ret)
1659 		return ret;
1660 
1661 	attr      = kmalloc(sizeof *attr, GFP_KERNEL);
1662 	init_attr = kmalloc(sizeof *init_attr, GFP_KERNEL);
1663 	if (!attr || !init_attr) {
1664 		ret = -ENOMEM;
1665 		goto out;
1666 	}
1667 
1668 	qp = uobj_get_obj_read(qp, UVERBS_OBJECT_QP, cmd.qp_handle, attrs);
1669 	if (!qp) {
1670 		ret = -EINVAL;
1671 		goto out;
1672 	}
1673 
1674 	ret = ib_query_qp(qp, attr, cmd.attr_mask, init_attr);
1675 
1676 	rdma_lookup_put_uobject(&qp->uobject->uevent.uobject,
1677 				UVERBS_LOOKUP_READ);
1678 
1679 	if (ret)
1680 		goto out;
1681 
1682 	memset(&resp, 0, sizeof resp);
1683 
1684 	resp.qp_state               = attr->qp_state;
1685 	resp.cur_qp_state           = attr->cur_qp_state;
1686 	resp.path_mtu               = attr->path_mtu;
1687 	resp.path_mig_state         = attr->path_mig_state;
1688 	resp.qkey                   = attr->qkey;
1689 	resp.rq_psn                 = attr->rq_psn;
1690 	resp.sq_psn                 = attr->sq_psn;
1691 	resp.dest_qp_num            = attr->dest_qp_num;
1692 	resp.qp_access_flags        = attr->qp_access_flags;
1693 	resp.pkey_index             = attr->pkey_index;
1694 	resp.alt_pkey_index         = attr->alt_pkey_index;
1695 	resp.sq_draining            = attr->sq_draining;
1696 	resp.max_rd_atomic          = attr->max_rd_atomic;
1697 	resp.max_dest_rd_atomic     = attr->max_dest_rd_atomic;
1698 	resp.min_rnr_timer          = attr->min_rnr_timer;
1699 	resp.port_num               = attr->port_num;
1700 	resp.timeout                = attr->timeout;
1701 	resp.retry_cnt              = attr->retry_cnt;
1702 	resp.rnr_retry              = attr->rnr_retry;
1703 	resp.alt_port_num           = attr->alt_port_num;
1704 	resp.alt_timeout            = attr->alt_timeout;
1705 
1706 	copy_ah_attr_to_uverbs(&resp.dest, &attr->ah_attr);
1707 	copy_ah_attr_to_uverbs(&resp.alt_dest, &attr->alt_ah_attr);
1708 
1709 	resp.max_send_wr            = init_attr->cap.max_send_wr;
1710 	resp.max_recv_wr            = init_attr->cap.max_recv_wr;
1711 	resp.max_send_sge           = init_attr->cap.max_send_sge;
1712 	resp.max_recv_sge           = init_attr->cap.max_recv_sge;
1713 	resp.max_inline_data        = init_attr->cap.max_inline_data;
1714 	resp.sq_sig_all             = init_attr->sq_sig_type == IB_SIGNAL_ALL_WR;
1715 
1716 	ret = uverbs_response(attrs, &resp, sizeof(resp));
1717 
1718 out:
1719 	kfree(attr);
1720 	kfree(init_attr);
1721 
1722 	return ret;
1723 }
1724 
1725 /* Remove ignored fields set in the attribute mask */
1726 static int modify_qp_mask(enum ib_qp_type qp_type, int mask)
1727 {
1728 	switch (qp_type) {
1729 	case IB_QPT_XRC_INI:
1730 		return mask & ~(IB_QP_MAX_DEST_RD_ATOMIC | IB_QP_MIN_RNR_TIMER);
1731 	case IB_QPT_XRC_TGT:
1732 		return mask & ~(IB_QP_MAX_QP_RD_ATOMIC | IB_QP_RETRY_CNT |
1733 				IB_QP_RNR_RETRY);
1734 	default:
1735 		return mask;
1736 	}
1737 }
1738 
1739 static void copy_ah_attr_from_uverbs(struct ib_device *dev,
1740 				     struct ib_ah_attr *rdma_attr,
1741 				     struct ib_uverbs_qp_dest *uverb_attr)
1742 {
1743 	if (uverb_attr->is_global) {
1744 		struct ib_global_route *grh = &rdma_attr->grh;
1745 
1746 		grh->flow_label = uverb_attr->flow_label;
1747 		grh->sgid_index = uverb_attr->sgid_index;
1748 		grh->hop_limit = uverb_attr->hop_limit;
1749 		grh->traffic_class = uverb_attr->traffic_class;
1750 		memcpy(grh->dgid.raw, uverb_attr->dgid, sizeof(grh->dgid));
1751 		rdma_attr->ah_flags = IB_AH_GRH;
1752 	} else {
1753 		rdma_attr->ah_flags = 0;
1754 	}
1755 	rdma_attr->dlid = uverb_attr->dlid;
1756 	rdma_attr->sl = uverb_attr->sl;
1757 	rdma_attr->src_path_bits = uverb_attr->src_path_bits;
1758 	rdma_attr->static_rate = uverb_attr->static_rate;
1759 	rdma_attr->port_num = uverb_attr->port_num;
1760 }
1761 
1762 static int modify_qp(struct uverbs_attr_bundle *attrs,
1763 		     struct ib_uverbs_ex_modify_qp *cmd)
1764 {
1765 	struct ib_qp_attr *attr;
1766 	struct ib_qp *qp;
1767 	int ret;
1768 
1769 	attr = kzalloc(sizeof(*attr), GFP_KERNEL);
1770 	if (!attr)
1771 		return -ENOMEM;
1772 
1773 	qp = uobj_get_obj_read(qp, UVERBS_OBJECT_QP, cmd->base.qp_handle,
1774 			       attrs);
1775 	if (!qp) {
1776 		ret = -EINVAL;
1777 		goto out;
1778 	}
1779 
1780 	if ((cmd->base.attr_mask & IB_QP_PORT) &&
1781 	    !rdma_is_port_valid(qp->device, cmd->base.port_num)) {
1782 		ret = -EINVAL;
1783 		goto release_qp;
1784 	}
1785 
1786 	if ((cmd->base.attr_mask & IB_QP_AV)) {
1787 		if (!rdma_is_port_valid(qp->device, cmd->base.dest.port_num)) {
1788 			ret = -EINVAL;
1789 			goto release_qp;
1790 		}
1791 
1792 		if (cmd->base.attr_mask & IB_QP_STATE &&
1793 		    cmd->base.qp_state == IB_QPS_RTR) {
1794 		/* We are in INIT->RTR TRANSITION (if we are not,
1795 		 * this transition will be rejected in subsequent checks).
1796 		 * In the INIT->RTR transition, we cannot have IB_QP_PORT set,
1797 		 * but the IB_QP_STATE flag is required.
1798 		 *
1799 		 * Since kernel 3.14 (commit dbf727de7440), the uverbs driver,
1800 		 * when IB_QP_AV is set, has required inclusion of a valid
1801 		 * port number in the primary AV. (AVs are created and handled
1802 		 * differently for infiniband and ethernet (RoCE) ports).
1803 		 *
1804 		 * Check the port number included in the primary AV against
1805 		 * the port number in the qp struct, which was set (and saved)
1806 		 * in the RST->INIT transition.
1807 		 */
1808 			if (cmd->base.dest.port_num != qp->real_qp->port) {
1809 				ret = -EINVAL;
1810 				goto release_qp;
1811 			}
1812 		} else {
1813 		/* We are in SQD->SQD. (If we are not, this transition will
1814 		 * be rejected later in the verbs layer checks).
1815 		 * Check for both IB_QP_PORT and IB_QP_AV, these can be set
1816 		 * together in the SQD->SQD transition.
1817 		 *
1818 		 * If only IP_QP_AV was set, add in IB_QP_PORT as well (the
1819 		 * verbs layer driver does not track primary port changes
1820 		 * resulting from path migration. Thus, in SQD, if the primary
1821 		 * AV is modified, the primary port should also be modified).
1822 		 *
1823 		 * Note that in this transition, the IB_QP_STATE flag
1824 		 * is not allowed.
1825 		 */
1826 			if (((cmd->base.attr_mask & (IB_QP_AV | IB_QP_PORT))
1827 			     == (IB_QP_AV | IB_QP_PORT)) &&
1828 			    cmd->base.port_num != cmd->base.dest.port_num) {
1829 				ret = -EINVAL;
1830 				goto release_qp;
1831 			}
1832 			if ((cmd->base.attr_mask & (IB_QP_AV | IB_QP_PORT))
1833 			    == IB_QP_AV) {
1834 				cmd->base.attr_mask |= IB_QP_PORT;
1835 				cmd->base.port_num = cmd->base.dest.port_num;
1836 			}
1837 		}
1838 	}
1839 
1840 	if ((cmd->base.attr_mask & IB_QP_ALT_PATH) &&
1841 	    (!rdma_is_port_valid(qp->device, cmd->base.alt_port_num) ||
1842 	    !rdma_is_port_valid(qp->device, cmd->base.alt_dest.port_num) ||
1843 	    cmd->base.alt_port_num != cmd->base.alt_dest.port_num)) {
1844 		ret = -EINVAL;
1845 		goto release_qp;
1846 	}
1847 
1848 	if ((cmd->base.attr_mask & IB_QP_CUR_STATE &&
1849 	    cmd->base.cur_qp_state > IB_QPS_ERR) ||
1850 	    (cmd->base.attr_mask & IB_QP_STATE &&
1851 	    cmd->base.qp_state > IB_QPS_ERR)) {
1852 		ret = -EINVAL;
1853 		goto release_qp;
1854 	}
1855 
1856 	if (cmd->base.attr_mask & IB_QP_STATE)
1857 		attr->qp_state = cmd->base.qp_state;
1858 	if (cmd->base.attr_mask & IB_QP_CUR_STATE)
1859 		attr->cur_qp_state = cmd->base.cur_qp_state;
1860 	if (cmd->base.attr_mask & IB_QP_PATH_MTU)
1861 		attr->path_mtu = cmd->base.path_mtu;
1862 	if (cmd->base.attr_mask & IB_QP_PATH_MIG_STATE)
1863 		attr->path_mig_state = cmd->base.path_mig_state;
1864 	if (cmd->base.attr_mask & IB_QP_QKEY)
1865 		attr->qkey = cmd->base.qkey;
1866 	if (cmd->base.attr_mask & IB_QP_RQ_PSN)
1867 		attr->rq_psn = cmd->base.rq_psn;
1868 	if (cmd->base.attr_mask & IB_QP_SQ_PSN)
1869 		attr->sq_psn = cmd->base.sq_psn;
1870 	if (cmd->base.attr_mask & IB_QP_DEST_QPN)
1871 		attr->dest_qp_num = cmd->base.dest_qp_num;
1872 	if (cmd->base.attr_mask & IB_QP_ACCESS_FLAGS)
1873 		attr->qp_access_flags = cmd->base.qp_access_flags;
1874 	if (cmd->base.attr_mask & IB_QP_PKEY_INDEX)
1875 		attr->pkey_index = cmd->base.pkey_index;
1876 	if (cmd->base.attr_mask & IB_QP_EN_SQD_ASYNC_NOTIFY)
1877 		attr->en_sqd_async_notify = cmd->base.en_sqd_async_notify;
1878 	if (cmd->base.attr_mask & IB_QP_MAX_QP_RD_ATOMIC)
1879 		attr->max_rd_atomic = cmd->base.max_rd_atomic;
1880 	if (cmd->base.attr_mask & IB_QP_MAX_DEST_RD_ATOMIC)
1881 		attr->max_dest_rd_atomic = cmd->base.max_dest_rd_atomic;
1882 	if (cmd->base.attr_mask & IB_QP_MIN_RNR_TIMER)
1883 		attr->min_rnr_timer = cmd->base.min_rnr_timer;
1884 	if (cmd->base.attr_mask & IB_QP_PORT)
1885 		attr->port_num = cmd->base.port_num;
1886 	if (cmd->base.attr_mask & IB_QP_TIMEOUT)
1887 		attr->timeout = cmd->base.timeout;
1888 	if (cmd->base.attr_mask & IB_QP_RETRY_CNT)
1889 		attr->retry_cnt = cmd->base.retry_cnt;
1890 	if (cmd->base.attr_mask & IB_QP_RNR_RETRY)
1891 		attr->rnr_retry = cmd->base.rnr_retry;
1892 	if (cmd->base.attr_mask & IB_QP_ALT_PATH) {
1893 		attr->alt_port_num = cmd->base.alt_port_num;
1894 		attr->alt_timeout = cmd->base.alt_timeout;
1895 		attr->alt_pkey_index = cmd->base.alt_pkey_index;
1896 	}
1897 	if (cmd->base.attr_mask & IB_QP_RATE_LIMIT)
1898 		attr->rate_limit = cmd->rate_limit;
1899 
1900 	if (cmd->base.attr_mask & IB_QP_AV)
1901 		copy_ah_attr_from_uverbs(qp->device, &attr->ah_attr,
1902 					 &cmd->base.dest);
1903 
1904 	if (cmd->base.attr_mask & IB_QP_ALT_PATH)
1905 		copy_ah_attr_from_uverbs(qp->device, &attr->alt_ah_attr,
1906 					 &cmd->base.alt_dest);
1907 
1908 	ret = ib_modify_qp_with_udata(qp, attr,
1909 				      modify_qp_mask(qp->qp_type,
1910 						     cmd->base.attr_mask),
1911 				      &attrs->driver_udata);
1912 
1913 release_qp:
1914 	rdma_lookup_put_uobject(&qp->uobject->uevent.uobject,
1915 				UVERBS_LOOKUP_READ);
1916 out:
1917 	kfree(attr);
1918 
1919 	return ret;
1920 }
1921 
1922 static int ib_uverbs_modify_qp(struct uverbs_attr_bundle *attrs)
1923 {
1924 	struct ib_uverbs_ex_modify_qp cmd;
1925 	int ret;
1926 
1927 	ret = uverbs_request(attrs, &cmd.base, sizeof(cmd.base));
1928 	if (ret)
1929 		return ret;
1930 
1931 	if (cmd.base.attr_mask &
1932 	    ~((IB_USER_LEGACY_LAST_QP_ATTR_MASK << 1) - 1))
1933 		return -EOPNOTSUPP;
1934 
1935 	return modify_qp(attrs, &cmd);
1936 }
1937 
1938 static int ib_uverbs_ex_modify_qp(struct uverbs_attr_bundle *attrs)
1939 {
1940 	struct ib_uverbs_ex_modify_qp cmd;
1941 	struct ib_uverbs_ex_modify_qp_resp resp = {
1942 		.response_length = uverbs_response_length(attrs, sizeof(resp))
1943 	};
1944 	int ret;
1945 
1946 	ret = uverbs_request(attrs, &cmd, sizeof(cmd));
1947 	if (ret)
1948 		return ret;
1949 
1950 	/*
1951 	 * Last bit is reserved for extending the attr_mask by
1952 	 * using another field.
1953 	 */
1954 	BUILD_BUG_ON(IB_USER_LAST_QP_ATTR_MASK == (1 << 31));
1955 
1956 	if (cmd.base.attr_mask &
1957 	    ~((IB_USER_LAST_QP_ATTR_MASK << 1) - 1))
1958 		return -EOPNOTSUPP;
1959 
1960 	ret = modify_qp(attrs, &cmd);
1961 	if (ret)
1962 		return ret;
1963 
1964 	return uverbs_response(attrs, &resp, sizeof(resp));
1965 }
1966 
1967 static int ib_uverbs_destroy_qp(struct uverbs_attr_bundle *attrs)
1968 {
1969 	struct ib_uverbs_destroy_qp      cmd;
1970 	struct ib_uverbs_destroy_qp_resp resp;
1971 	struct ib_uobject		*uobj;
1972 	struct ib_uqp_object        	*obj;
1973 	int ret;
1974 
1975 	ret = uverbs_request(attrs, &cmd, sizeof(cmd));
1976 	if (ret)
1977 		return ret;
1978 
1979 	uobj = uobj_get_destroy(UVERBS_OBJECT_QP, cmd.qp_handle, attrs);
1980 	if (IS_ERR(uobj))
1981 		return PTR_ERR(uobj);
1982 
1983 	obj = container_of(uobj, struct ib_uqp_object, uevent.uobject);
1984 	memset(&resp, 0, sizeof(resp));
1985 	resp.events_reported = obj->uevent.events_reported;
1986 
1987 	uobj_put_destroy(uobj);
1988 
1989 	return uverbs_response(attrs, &resp, sizeof(resp));
1990 }
1991 
1992 static void *alloc_wr(size_t wr_size, __u32 num_sge)
1993 {
1994 	if (num_sge >= (U32_MAX - ALIGN(wr_size, sizeof (struct ib_sge))) /
1995 		       sizeof (struct ib_sge))
1996 		return NULL;
1997 
1998 	return kmalloc(ALIGN(wr_size, sizeof (struct ib_sge)) +
1999 			 num_sge * sizeof (struct ib_sge), GFP_KERNEL);
2000 }
2001 
2002 static int ib_uverbs_post_send(struct uverbs_attr_bundle *attrs)
2003 {
2004 	struct ib_uverbs_post_send      cmd;
2005 	struct ib_uverbs_post_send_resp resp;
2006 	struct ib_uverbs_send_wr       *user_wr;
2007 	struct ib_send_wr              *wr = NULL, *last, *next;
2008 	const struct ib_send_wr        *bad_wr;
2009 	struct ib_qp                   *qp;
2010 	int                             i, sg_ind;
2011 	int				is_ud;
2012 	int ret, ret2;
2013 	size_t                          next_size;
2014 	const struct ib_sge __user *sgls;
2015 	const void __user *wqes;
2016 	struct uverbs_req_iter iter;
2017 
2018 	ret = uverbs_request_start(attrs, &iter, &cmd, sizeof(cmd));
2019 	if (ret)
2020 		return ret;
2021 	wqes = uverbs_request_next_ptr(&iter, cmd.wqe_size * cmd.wr_count);
2022 	if (IS_ERR(wqes))
2023 		return PTR_ERR(wqes);
2024 	sgls = uverbs_request_next_ptr(
2025 		&iter, cmd.sge_count * sizeof(struct ib_uverbs_sge));
2026 	if (IS_ERR(sgls))
2027 		return PTR_ERR(sgls);
2028 	ret = uverbs_request_finish(&iter);
2029 	if (ret)
2030 		return ret;
2031 
2032 	user_wr = kmalloc(cmd.wqe_size, GFP_KERNEL);
2033 	if (!user_wr)
2034 		return -ENOMEM;
2035 
2036 	qp = uobj_get_obj_read(qp, UVERBS_OBJECT_QP, cmd.qp_handle, attrs);
2037 	if (!qp) {
2038 		ret = -EINVAL;
2039 		goto out;
2040 	}
2041 
2042 	is_ud = qp->qp_type == IB_QPT_UD;
2043 	sg_ind = 0;
2044 	last = NULL;
2045 	for (i = 0; i < cmd.wr_count; ++i) {
2046 		if (copy_from_user(user_wr, (const u8 *)wqes + i * cmd.wqe_size,
2047 				   cmd.wqe_size)) {
2048 			ret = -EFAULT;
2049 			goto out_put;
2050 		}
2051 
2052 		if (user_wr->num_sge + sg_ind > cmd.sge_count) {
2053 			ret = -EINVAL;
2054 			goto out_put;
2055 		}
2056 
2057 		if (is_ud) {
2058 			struct ib_ud_wr *ud;
2059 
2060 			if (user_wr->opcode != IB_WR_SEND &&
2061 			    user_wr->opcode != IB_WR_SEND_WITH_IMM) {
2062 				ret = -EINVAL;
2063 				goto out_put;
2064 			}
2065 
2066 			next_size = sizeof(*ud);
2067 			ud = alloc_wr(next_size, user_wr->num_sge);
2068 			if (!ud) {
2069 				ret = -ENOMEM;
2070 				goto out_put;
2071 			}
2072 
2073 			ud->ah = uobj_get_obj_read(ah, UVERBS_OBJECT_AH,
2074 						   user_wr->wr.ud.ah, attrs);
2075 			if (!ud->ah) {
2076 				kfree(ud);
2077 				ret = -EINVAL;
2078 				goto out_put;
2079 			}
2080 			ud->remote_qpn = user_wr->wr.ud.remote_qpn;
2081 			ud->remote_qkey = user_wr->wr.ud.remote_qkey;
2082 
2083 			next = &ud->wr;
2084 		} else if (user_wr->opcode == IB_WR_RDMA_WRITE_WITH_IMM ||
2085 			   user_wr->opcode == IB_WR_RDMA_WRITE ||
2086 			   user_wr->opcode == IB_WR_RDMA_READ) {
2087 			struct ib_rdma_wr *rdma;
2088 
2089 			next_size = sizeof(*rdma);
2090 			rdma = alloc_wr(next_size, user_wr->num_sge);
2091 			if (!rdma) {
2092 				ret = -ENOMEM;
2093 				goto out_put;
2094 			}
2095 
2096 			rdma->remote_addr = user_wr->wr.rdma.remote_addr;
2097 			rdma->rkey = user_wr->wr.rdma.rkey;
2098 
2099 			next = &rdma->wr;
2100 		} else if (user_wr->opcode == IB_WR_ATOMIC_CMP_AND_SWP ||
2101 			   user_wr->opcode == IB_WR_ATOMIC_FETCH_AND_ADD) {
2102 			struct ib_atomic_wr *atomic;
2103 
2104 			next_size = sizeof(*atomic);
2105 			atomic = alloc_wr(next_size, user_wr->num_sge);
2106 			if (!atomic) {
2107 				ret = -ENOMEM;
2108 				goto out_put;
2109 			}
2110 
2111 			atomic->remote_addr = user_wr->wr.atomic.remote_addr;
2112 			atomic->compare_add = user_wr->wr.atomic.compare_add;
2113 			atomic->swap = user_wr->wr.atomic.swap;
2114 			atomic->rkey = user_wr->wr.atomic.rkey;
2115 
2116 			next = &atomic->wr;
2117 		} else if (user_wr->opcode == IB_WR_SEND ||
2118 			   user_wr->opcode == IB_WR_SEND_WITH_IMM ||
2119 			   user_wr->opcode == IB_WR_SEND_WITH_INV) {
2120 			next_size = sizeof(*next);
2121 			next = alloc_wr(next_size, user_wr->num_sge);
2122 			if (!next) {
2123 				ret = -ENOMEM;
2124 				goto out_put;
2125 			}
2126 		} else {
2127 			ret = -EINVAL;
2128 			goto out_put;
2129 		}
2130 
2131 		if (user_wr->opcode == IB_WR_SEND_WITH_IMM ||
2132 		    user_wr->opcode == IB_WR_RDMA_WRITE_WITH_IMM) {
2133 			next->ex.imm_data =
2134 					(__be32 __force) user_wr->ex.imm_data;
2135 		} else if (user_wr->opcode == IB_WR_SEND_WITH_INV) {
2136 			next->ex.invalidate_rkey = user_wr->ex.invalidate_rkey;
2137 		}
2138 
2139 		if (!last)
2140 			wr = next;
2141 		else
2142 			last->next = next;
2143 		last = next;
2144 
2145 		next->next       = NULL;
2146 		next->wr_id      = user_wr->wr_id;
2147 		next->num_sge    = user_wr->num_sge;
2148 		next->opcode     = user_wr->opcode;
2149 		next->send_flags = user_wr->send_flags;
2150 
2151 		if (next->num_sge) {
2152 			next->sg_list = (void *)((char *)next +
2153 				ALIGN(next_size, sizeof(struct ib_sge)));
2154 			if (copy_from_user(next->sg_list, sgls + sg_ind,
2155 					   next->num_sge *
2156 						   sizeof(struct ib_sge))) {
2157 				ret = -EFAULT;
2158 				goto out_put;
2159 			}
2160 			sg_ind += next->num_sge;
2161 		} else
2162 			next->sg_list = NULL;
2163 	}
2164 
2165 	resp.bad_wr = 0;
2166 	ret = qp->device->post_send(qp->real_qp, wr, &bad_wr);
2167 	if (ret)
2168 		for (next = wr; next; next = next->next) {
2169 			++resp.bad_wr;
2170 			if (next == bad_wr)
2171 				break;
2172 		}
2173 
2174 	ret2 = uverbs_response(attrs, &resp, sizeof(resp));
2175 	if (ret2)
2176 		ret = ret2;
2177 
2178 out_put:
2179 	rdma_lookup_put_uobject(&qp->uobject->uevent.uobject,
2180 				UVERBS_LOOKUP_READ);
2181 
2182 	while (wr) {
2183 		if (is_ud && ud_wr(wr)->ah)
2184 			uobj_put_obj_read(ud_wr(wr)->ah);
2185 		next = wr->next;
2186 		kfree(wr);
2187 		wr = next;
2188 	}
2189 
2190 out:
2191 	kfree(user_wr);
2192 
2193 	return ret;
2194 }
2195 
2196 static struct ib_recv_wr *
2197 ib_uverbs_unmarshall_recv(struct uverbs_req_iter *iter, u32 wr_count,
2198 			  u32 wqe_size, u32 sge_count)
2199 {
2200 	struct ib_uverbs_recv_wr *user_wr;
2201 	struct ib_recv_wr        *wr = NULL, *last, *next;
2202 	int                       sg_ind;
2203 	int                       i;
2204 	int                       ret;
2205 	const struct ib_sge __user *sgls;
2206 	const void __user *wqes;
2207 
2208 	if (wqe_size < sizeof (struct ib_uverbs_recv_wr))
2209 		return ERR_PTR(-EINVAL);
2210 
2211 	wqes = uverbs_request_next_ptr(iter, wqe_size * wr_count);
2212 	if (IS_ERR(wqes))
2213 		return ERR_CAST(wqes);
2214 	sgls = uverbs_request_next_ptr(
2215 		iter, sge_count * sizeof(struct ib_uverbs_sge));
2216 	if (IS_ERR(sgls))
2217 		return ERR_CAST(sgls);
2218 	ret = uverbs_request_finish(iter);
2219 	if (ret)
2220 		return ERR_PTR(ret);
2221 
2222 	user_wr = kmalloc(wqe_size, GFP_KERNEL);
2223 	if (!user_wr)
2224 		return ERR_PTR(-ENOMEM);
2225 
2226 	sg_ind = 0;
2227 	last = NULL;
2228 	for (i = 0; i < wr_count; ++i) {
2229 		if (copy_from_user(user_wr, (const char *)wqes + i * wqe_size,
2230 				   wqe_size)) {
2231 			ret = -EFAULT;
2232 			goto err;
2233 		}
2234 
2235 		if (user_wr->num_sge + sg_ind > sge_count) {
2236 			ret = -EINVAL;
2237 			goto err;
2238 		}
2239 
2240 		if (user_wr->num_sge >=
2241 		    (U32_MAX - ALIGN(sizeof *next, sizeof (struct ib_sge))) /
2242 		    sizeof (struct ib_sge)) {
2243 			ret = -EINVAL;
2244 			goto err;
2245 		}
2246 
2247 		next = kmalloc(ALIGN(sizeof *next, sizeof (struct ib_sge)) +
2248 			       user_wr->num_sge * sizeof (struct ib_sge),
2249 			       GFP_KERNEL);
2250 		if (!next) {
2251 			ret = -ENOMEM;
2252 			goto err;
2253 		}
2254 
2255 		if (!last)
2256 			wr = next;
2257 		else
2258 			last->next = next;
2259 		last = next;
2260 
2261 		next->next       = NULL;
2262 		next->wr_id      = user_wr->wr_id;
2263 		next->num_sge    = user_wr->num_sge;
2264 
2265 		if (next->num_sge) {
2266 			next->sg_list = (void *)((char *)next +
2267 				ALIGN(sizeof *next, sizeof (struct ib_sge)));
2268 			if (copy_from_user(next->sg_list, sgls + sg_ind,
2269 					   next->num_sge *
2270 						   sizeof(struct ib_sge))) {
2271 				ret = -EFAULT;
2272 				goto err;
2273 			}
2274 			sg_ind += next->num_sge;
2275 		} else
2276 			next->sg_list = NULL;
2277 	}
2278 
2279 	kfree(user_wr);
2280 	return wr;
2281 
2282 err:
2283 	kfree(user_wr);
2284 
2285 	while (wr) {
2286 		next = wr->next;
2287 		kfree(wr);
2288 		wr = next;
2289 	}
2290 
2291 	return ERR_PTR(ret);
2292 }
2293 
2294 static int ib_uverbs_post_recv(struct uverbs_attr_bundle *attrs)
2295 {
2296 	struct ib_uverbs_post_recv      cmd;
2297 	struct ib_uverbs_post_recv_resp resp;
2298 	struct ib_recv_wr              *wr, *next;
2299 	const struct ib_recv_wr        *bad_wr;
2300 	struct ib_qp                   *qp;
2301 	int ret, ret2;
2302 	struct uverbs_req_iter iter;
2303 
2304 	ret = uverbs_request_start(attrs, &iter, &cmd, sizeof(cmd));
2305 	if (ret)
2306 		return ret;
2307 
2308 	wr = ib_uverbs_unmarshall_recv(&iter, cmd.wr_count, cmd.wqe_size,
2309 				       cmd.sge_count);
2310 	if (IS_ERR(wr))
2311 		return PTR_ERR(wr);
2312 
2313 	qp = uobj_get_obj_read(qp, UVERBS_OBJECT_QP, cmd.qp_handle, attrs);
2314 	if (!qp) {
2315 		ret = -EINVAL;
2316 		goto out;
2317 	}
2318 
2319 	resp.bad_wr = 0;
2320 	ret = qp->device->post_recv(qp->real_qp, wr, &bad_wr);
2321 
2322 	rdma_lookup_put_uobject(&qp->uobject->uevent.uobject,
2323 				UVERBS_LOOKUP_READ);
2324 	if (ret) {
2325 		for (next = wr; next; next = next->next) {
2326 			++resp.bad_wr;
2327 			if (next == bad_wr)
2328 				break;
2329 		}
2330 	}
2331 
2332 	ret2 = uverbs_response(attrs, &resp, sizeof(resp));
2333 	if (ret2)
2334 		ret = ret2;
2335 out:
2336 	while (wr) {
2337 		next = wr->next;
2338 		kfree(wr);
2339 		wr = next;
2340 	}
2341 
2342 	return ret;
2343 }
2344 
2345 static int ib_uverbs_post_srq_recv(struct uverbs_attr_bundle *attrs)
2346 {
2347 	struct ib_uverbs_post_srq_recv      cmd;
2348 	struct ib_uverbs_post_srq_recv_resp resp;
2349 	struct ib_recv_wr                  *wr, *next;
2350 	const struct ib_recv_wr		   *bad_wr;
2351 	struct ib_srq                      *srq;
2352 	int ret, ret2;
2353 	struct uverbs_req_iter iter;
2354 
2355 	ret = uverbs_request_start(attrs, &iter, &cmd, sizeof(cmd));
2356 	if (ret)
2357 		return ret;
2358 
2359 	wr = ib_uverbs_unmarshall_recv(&iter, cmd.wr_count, cmd.wqe_size,
2360 				       cmd.sge_count);
2361 	if (IS_ERR(wr))
2362 		return PTR_ERR(wr);
2363 
2364 	srq = uobj_get_obj_read(srq, UVERBS_OBJECT_SRQ, cmd.srq_handle, attrs);
2365 	if (!srq) {
2366 		ret = -EINVAL;
2367 		goto out;
2368 	}
2369 
2370 	resp.bad_wr = 0;
2371 	ret = srq->device->post_srq_recv(srq, wr, &bad_wr);
2372 
2373 	rdma_lookup_put_uobject(&srq->uobject->uevent.uobject,
2374 				UVERBS_LOOKUP_READ);
2375 
2376 	if (ret)
2377 		for (next = wr; next; next = next->next) {
2378 			++resp.bad_wr;
2379 			if (next == bad_wr)
2380 				break;
2381 		}
2382 
2383 	ret2 = uverbs_response(attrs, &resp, sizeof(resp));
2384 	if (ret2)
2385 		ret = ret2;
2386 
2387 out:
2388 	while (wr) {
2389 		next = wr->next;
2390 		kfree(wr);
2391 		wr = next;
2392 	}
2393 
2394 	return ret;
2395 }
2396 
2397 static int ib_uverbs_create_ah(struct uverbs_attr_bundle *attrs)
2398 {
2399 	struct ib_uverbs_create_ah	 cmd;
2400 	struct ib_uverbs_create_ah_resp	 resp;
2401 	struct ib_uobject		*uobj;
2402 	struct ib_pd			*pd;
2403 	struct ib_ah			*ah;
2404 	struct ib_ah_attr		attr = {};
2405 	int ret;
2406 	struct ib_device *ib_dev;
2407 
2408 	ret = uverbs_request(attrs, &cmd, sizeof(cmd));
2409 	if (ret)
2410 		return ret;
2411 
2412 	uobj = uobj_alloc(UVERBS_OBJECT_AH, attrs, &ib_dev);
2413 	if (IS_ERR(uobj))
2414 		return PTR_ERR(uobj);
2415 
2416 	if (!rdma_is_port_valid(ib_dev, cmd.attr.port_num)) {
2417 		ret = -EINVAL;
2418 		goto err;
2419 	}
2420 
2421 	pd = uobj_get_obj_read(pd, UVERBS_OBJECT_PD, cmd.pd_handle, attrs);
2422 	if (!pd) {
2423 		ret = -EINVAL;
2424 		goto err;
2425 	}
2426 
2427 	attr.dlid = cmd.attr.dlid;
2428 	attr.sl = cmd.attr.sl;
2429 	attr.src_path_bits = cmd.attr.src_path_bits;
2430 	attr.static_rate = cmd.attr.static_rate;
2431 	attr.port_num = cmd.attr.port_num;
2432 
2433 	if (cmd.attr.is_global) {
2434 		struct ib_global_route *grh = &attr.grh;
2435 
2436 		grh->flow_label = cmd.attr.grh.flow_label;
2437 		grh->sgid_index = cmd.attr.grh.sgid_index;
2438 		grh->hop_limit = cmd.attr.grh.hop_limit;
2439 		grh->traffic_class = cmd.attr.grh.traffic_class;
2440 		memcpy(grh->dgid.raw, cmd.attr.grh.dgid, sizeof(grh->dgid));
2441 		attr.ah_flags = IB_AH_GRH;
2442 	} else {
2443 		attr.ah_flags = 0;
2444 	}
2445 
2446 	ah = ib_create_user_ah(pd, &attr, &attrs->driver_udata);
2447 	if (IS_ERR(ah)) {
2448 		ret = PTR_ERR(ah);
2449 		goto err_put;
2450 	}
2451 
2452 	ah->uobject  = uobj;
2453 	uobj->user_handle = cmd.user_handle;
2454 	uobj->object = ah;
2455 
2456 	resp.ah_handle = uobj->id;
2457 
2458 	ret = uverbs_response(attrs, &resp, sizeof(resp));
2459 	if (ret)
2460 		goto err_copy;
2461 
2462 	uobj_put_obj_read(pd);
2463 	rdma_alloc_commit_uobject(uobj, attrs);
2464 	return 0;
2465 
2466 err_copy:
2467 	ib_destroy_ah_user(ah, RDMA_DESTROY_AH_SLEEPABLE,
2468 			   uverbs_get_cleared_udata(attrs));
2469 
2470 err_put:
2471 	uobj_put_obj_read(pd);
2472 
2473 err:
2474 	uobj_alloc_abort(uobj, attrs);
2475 	return ret;
2476 }
2477 
2478 static int ib_uverbs_destroy_ah(struct uverbs_attr_bundle *attrs)
2479 {
2480 	struct ib_uverbs_destroy_ah cmd;
2481 	int ret;
2482 
2483 	ret = uverbs_request(attrs, &cmd, sizeof(cmd));
2484 	if (ret)
2485 		return ret;
2486 
2487 	return uobj_perform_destroy(UVERBS_OBJECT_AH, cmd.ah_handle, attrs);
2488 }
2489 
2490 static int ib_uverbs_attach_mcast(struct uverbs_attr_bundle *attrs)
2491 {
2492 	struct ib_uverbs_attach_mcast cmd;
2493 	struct ib_qp                 *qp;
2494 	struct ib_uqp_object         *obj;
2495 	struct ib_uverbs_mcast_entry *mcast;
2496 	int                           ret;
2497 
2498 	ret = uverbs_request(attrs, &cmd, sizeof(cmd));
2499 	if (ret)
2500 		return ret;
2501 
2502 	qp = uobj_get_obj_read(qp, UVERBS_OBJECT_QP, cmd.qp_handle, attrs);
2503 	if (!qp)
2504 		return -EINVAL;
2505 
2506 	obj = qp->uobject;
2507 
2508 	mutex_lock(&obj->mcast_lock);
2509 	list_for_each_entry(mcast, &obj->mcast_list, list)
2510 		if (cmd.mlid == mcast->lid &&
2511 		    !memcmp(cmd.gid, mcast->gid.raw, sizeof mcast->gid.raw)) {
2512 			ret = 0;
2513 			goto out_put;
2514 		}
2515 
2516 	mcast = kmalloc(sizeof *mcast, GFP_KERNEL);
2517 	if (!mcast) {
2518 		ret = -ENOMEM;
2519 		goto out_put;
2520 	}
2521 
2522 	mcast->lid = cmd.mlid;
2523 	memcpy(mcast->gid.raw, cmd.gid, sizeof mcast->gid.raw);
2524 
2525 	ret = ib_attach_mcast(qp, &mcast->gid, cmd.mlid);
2526 	if (!ret)
2527 		list_add_tail(&mcast->list, &obj->mcast_list);
2528 	else
2529 		kfree(mcast);
2530 
2531 out_put:
2532 	mutex_unlock(&obj->mcast_lock);
2533 	rdma_lookup_put_uobject(&qp->uobject->uevent.uobject,
2534 				UVERBS_LOOKUP_READ);
2535 
2536 	return ret;
2537 }
2538 
2539 static int ib_uverbs_detach_mcast(struct uverbs_attr_bundle *attrs)
2540 {
2541 	struct ib_uverbs_detach_mcast cmd;
2542 	struct ib_uqp_object         *obj;
2543 	struct ib_qp                 *qp;
2544 	struct ib_uverbs_mcast_entry *mcast;
2545 	int                           ret;
2546 	bool                          found = false;
2547 
2548 	ret = uverbs_request(attrs, &cmd, sizeof(cmd));
2549 	if (ret)
2550 		return ret;
2551 
2552 	qp = uobj_get_obj_read(qp, UVERBS_OBJECT_QP, cmd.qp_handle, attrs);
2553 	if (!qp)
2554 		return -EINVAL;
2555 
2556 	obj = qp->uobject;
2557 	mutex_lock(&obj->mcast_lock);
2558 
2559 	list_for_each_entry(mcast, &obj->mcast_list, list)
2560 		if (cmd.mlid == mcast->lid &&
2561 		    !memcmp(cmd.gid, mcast->gid.raw, sizeof mcast->gid.raw)) {
2562 			list_del(&mcast->list);
2563 			kfree(mcast);
2564 			found = true;
2565 			break;
2566 		}
2567 
2568 	if (!found) {
2569 		ret = -EINVAL;
2570 		goto out_put;
2571 	}
2572 
2573 	ret = ib_detach_mcast(qp, (union ib_gid *)cmd.gid, cmd.mlid);
2574 
2575 out_put:
2576 	mutex_unlock(&obj->mcast_lock);
2577 	rdma_lookup_put_uobject(&qp->uobject->uevent.uobject,
2578 				UVERBS_LOOKUP_READ);
2579 	return ret;
2580 }
2581 
2582 struct ib_uflow_resources *flow_resources_alloc(size_t num_specs)
2583 {
2584 	struct ib_uflow_resources *resources;
2585 
2586 	resources = kzalloc(sizeof(*resources), GFP_KERNEL);
2587 
2588 	if (!resources)
2589 		return NULL;
2590 
2591 	if (!num_specs)
2592 		goto out;
2593 
2594 	resources->counters =
2595 		kcalloc(num_specs, sizeof(*resources->counters), GFP_KERNEL);
2596 	resources->collection =
2597 		kcalloc(num_specs, sizeof(*resources->collection), GFP_KERNEL);
2598 
2599 	if (!resources->counters || !resources->collection)
2600 		goto err;
2601 
2602 out:
2603 	resources->max = num_specs;
2604 	return resources;
2605 
2606 err:
2607 	kfree(resources->counters);
2608 	kfree(resources);
2609 
2610 	return NULL;
2611 }
2612 EXPORT_SYMBOL(flow_resources_alloc);
2613 
2614 void ib_uverbs_flow_resources_free(struct ib_uflow_resources *uflow_res)
2615 {
2616 	unsigned int i;
2617 
2618 	if (!uflow_res)
2619 		return;
2620 
2621 	for (i = 0; i < uflow_res->collection_num; i++)
2622 		atomic_dec(&uflow_res->collection[i]->usecnt);
2623 
2624 	for (i = 0; i < uflow_res->counters_num; i++)
2625 		atomic_dec(&uflow_res->counters[i]->usecnt);
2626 
2627 	kfree(uflow_res->collection);
2628 	kfree(uflow_res->counters);
2629 	kfree(uflow_res);
2630 }
2631 EXPORT_SYMBOL(ib_uverbs_flow_resources_free);
2632 
2633 void flow_resources_add(struct ib_uflow_resources *uflow_res,
2634 			enum ib_flow_spec_type type,
2635 			void *ibobj)
2636 {
2637 	WARN_ON(uflow_res->num >= uflow_res->max);
2638 
2639 	switch (type) {
2640 	case IB_FLOW_SPEC_ACTION_HANDLE:
2641 		atomic_inc(&((struct ib_flow_action *)ibobj)->usecnt);
2642 		uflow_res->collection[uflow_res->collection_num++] =
2643 			(struct ib_flow_action *)ibobj;
2644 		break;
2645 	case IB_FLOW_SPEC_ACTION_COUNT:
2646 		atomic_inc(&((struct ib_counters *)ibobj)->usecnt);
2647 		uflow_res->counters[uflow_res->counters_num++] =
2648 			(struct ib_counters *)ibobj;
2649 		break;
2650 	default:
2651 		WARN_ON(1);
2652 	}
2653 
2654 	uflow_res->num++;
2655 }
2656 EXPORT_SYMBOL(flow_resources_add);
2657 
2658 static int kern_spec_to_ib_spec_action(struct uverbs_attr_bundle *attrs,
2659 				       struct ib_uverbs_flow_spec *kern_spec,
2660 				       union ib_flow_spec *ib_spec,
2661 				       struct ib_uflow_resources *uflow_res)
2662 {
2663 	ib_spec->type = kern_spec->type;
2664 	switch (ib_spec->type) {
2665 	case IB_FLOW_SPEC_ACTION_TAG:
2666 		if (kern_spec->flow_tag.size !=
2667 		    sizeof(struct ib_uverbs_flow_spec_action_tag))
2668 			return -EINVAL;
2669 
2670 		ib_spec->flow_tag.size = sizeof(struct ib_flow_spec_action_tag);
2671 		ib_spec->flow_tag.tag_id = kern_spec->flow_tag.tag_id;
2672 		break;
2673 	case IB_FLOW_SPEC_ACTION_DROP:
2674 		if (kern_spec->drop.size !=
2675 		    sizeof(struct ib_uverbs_flow_spec_action_drop))
2676 			return -EINVAL;
2677 
2678 		ib_spec->drop.size = sizeof(struct ib_flow_spec_action_drop);
2679 		break;
2680 	case IB_FLOW_SPEC_ACTION_HANDLE:
2681 		if (kern_spec->action.size !=
2682 		    sizeof(struct ib_uverbs_flow_spec_action_handle))
2683 			return -EOPNOTSUPP;
2684 		ib_spec->action.act = uobj_get_obj_read(flow_action,
2685 							UVERBS_OBJECT_FLOW_ACTION,
2686 							kern_spec->action.handle,
2687 							attrs);
2688 		if (!ib_spec->action.act)
2689 			return -EINVAL;
2690 		ib_spec->action.size =
2691 			sizeof(struct ib_flow_spec_action_handle);
2692 		flow_resources_add(uflow_res,
2693 				   IB_FLOW_SPEC_ACTION_HANDLE,
2694 				   ib_spec->action.act);
2695 		uobj_put_obj_read(ib_spec->action.act);
2696 		break;
2697 	case IB_FLOW_SPEC_ACTION_COUNT:
2698 		if (kern_spec->flow_count.size !=
2699 			sizeof(struct ib_uverbs_flow_spec_action_count))
2700 			return -EINVAL;
2701 		ib_spec->flow_count.counters =
2702 			uobj_get_obj_read(counters,
2703 					  UVERBS_OBJECT_COUNTERS,
2704 					  kern_spec->flow_count.handle,
2705 					  attrs);
2706 		if (!ib_spec->flow_count.counters)
2707 			return -EINVAL;
2708 		ib_spec->flow_count.size =
2709 				sizeof(struct ib_flow_spec_action_count);
2710 		flow_resources_add(uflow_res,
2711 				   IB_FLOW_SPEC_ACTION_COUNT,
2712 				   ib_spec->flow_count.counters);
2713 		uobj_put_obj_read(ib_spec->flow_count.counters);
2714 		break;
2715 	default:
2716 		return -EINVAL;
2717 	}
2718 	return 0;
2719 }
2720 
2721 static ssize_t spec_filter_size(const void *kern_spec_filter, u16 kern_filter_size,
2722 				u16 ib_real_filter_sz)
2723 {
2724 	/*
2725 	 * User space filter structures must be 64 bit aligned, otherwise this
2726 	 * may pass, but we won't handle additional new attributes.
2727 	 */
2728 
2729 	if (kern_filter_size > ib_real_filter_sz) {
2730 		if (memchr_inv((const char *)kern_spec_filter +
2731 			       ib_real_filter_sz, 0,
2732 			       kern_filter_size - ib_real_filter_sz))
2733 			return -EINVAL;
2734 		return ib_real_filter_sz;
2735 	}
2736 	return kern_filter_size;
2737 }
2738 
2739 int ib_uverbs_kern_spec_to_ib_spec_filter(enum ib_flow_spec_type type,
2740 					  const void *kern_spec_mask,
2741 					  const void *kern_spec_val,
2742 					  size_t kern_filter_sz,
2743 					  union ib_flow_spec *ib_spec)
2744 {
2745 	ssize_t actual_filter_sz;
2746 	ssize_t ib_filter_sz;
2747 
2748 	/* User flow spec size must be aligned to 4 bytes */
2749 	if (kern_filter_sz != ALIGN(kern_filter_sz, 4))
2750 		return -EINVAL;
2751 
2752 	ib_spec->type = type;
2753 
2754 	if (ib_spec->type == (IB_FLOW_SPEC_INNER | IB_FLOW_SPEC_VXLAN_TUNNEL))
2755 		return -EINVAL;
2756 
2757 	switch (ib_spec->type & ~IB_FLOW_SPEC_INNER) {
2758 	case IB_FLOW_SPEC_ETH:
2759 		ib_filter_sz = offsetof(struct ib_flow_eth_filter, real_sz);
2760 		actual_filter_sz = spec_filter_size(kern_spec_mask,
2761 						    kern_filter_sz,
2762 						    ib_filter_sz);
2763 		if (actual_filter_sz <= 0)
2764 			return -EINVAL;
2765 		ib_spec->size = sizeof(struct ib_flow_spec_eth);
2766 		memcpy(&ib_spec->eth.val, kern_spec_val, actual_filter_sz);
2767 		memcpy(&ib_spec->eth.mask, kern_spec_mask, actual_filter_sz);
2768 		break;
2769 	case IB_FLOW_SPEC_IPV4:
2770 		ib_filter_sz = offsetof(struct ib_flow_ipv4_filter, real_sz);
2771 		actual_filter_sz = spec_filter_size(kern_spec_mask,
2772 						    kern_filter_sz,
2773 						    ib_filter_sz);
2774 		if (actual_filter_sz <= 0)
2775 			return -EINVAL;
2776 		ib_spec->size = sizeof(struct ib_flow_spec_ipv4);
2777 		memcpy(&ib_spec->ipv4.val, kern_spec_val, actual_filter_sz);
2778 		memcpy(&ib_spec->ipv4.mask, kern_spec_mask, actual_filter_sz);
2779 		break;
2780 	case IB_FLOW_SPEC_IPV6:
2781 		ib_filter_sz = offsetof(struct ib_flow_ipv6_filter, real_sz);
2782 		actual_filter_sz = spec_filter_size(kern_spec_mask,
2783 						    kern_filter_sz,
2784 						    ib_filter_sz);
2785 		if (actual_filter_sz <= 0)
2786 			return -EINVAL;
2787 		ib_spec->size = sizeof(struct ib_flow_spec_ipv6);
2788 		memcpy(&ib_spec->ipv6.val, kern_spec_val, actual_filter_sz);
2789 		memcpy(&ib_spec->ipv6.mask, kern_spec_mask, actual_filter_sz);
2790 
2791 		if ((ntohl(ib_spec->ipv6.mask.flow_label)) >= BIT(20) ||
2792 		    (ntohl(ib_spec->ipv6.val.flow_label)) >= BIT(20))
2793 			return -EINVAL;
2794 		break;
2795 	case IB_FLOW_SPEC_TCP:
2796 	case IB_FLOW_SPEC_UDP:
2797 		ib_filter_sz = offsetof(struct ib_flow_tcp_udp_filter, real_sz);
2798 		actual_filter_sz = spec_filter_size(kern_spec_mask,
2799 						    kern_filter_sz,
2800 						    ib_filter_sz);
2801 		if (actual_filter_sz <= 0)
2802 			return -EINVAL;
2803 		ib_spec->size = sizeof(struct ib_flow_spec_tcp_udp);
2804 		memcpy(&ib_spec->tcp_udp.val, kern_spec_val, actual_filter_sz);
2805 		memcpy(&ib_spec->tcp_udp.mask, kern_spec_mask, actual_filter_sz);
2806 		break;
2807 	case IB_FLOW_SPEC_VXLAN_TUNNEL:
2808 		ib_filter_sz = offsetof(struct ib_flow_tunnel_filter, real_sz);
2809 		actual_filter_sz = spec_filter_size(kern_spec_mask,
2810 						    kern_filter_sz,
2811 						    ib_filter_sz);
2812 		if (actual_filter_sz <= 0)
2813 			return -EINVAL;
2814 		ib_spec->tunnel.size = sizeof(struct ib_flow_spec_tunnel);
2815 		memcpy(&ib_spec->tunnel.val, kern_spec_val, actual_filter_sz);
2816 		memcpy(&ib_spec->tunnel.mask, kern_spec_mask, actual_filter_sz);
2817 
2818 		if ((ntohl(ib_spec->tunnel.mask.tunnel_id)) >= BIT(24) ||
2819 		    (ntohl(ib_spec->tunnel.val.tunnel_id)) >= BIT(24))
2820 			return -EINVAL;
2821 		break;
2822 	case IB_FLOW_SPEC_ESP:
2823 		ib_filter_sz = offsetof(struct ib_flow_esp_filter, real_sz);
2824 		actual_filter_sz = spec_filter_size(kern_spec_mask,
2825 						    kern_filter_sz,
2826 						    ib_filter_sz);
2827 		if (actual_filter_sz <= 0)
2828 			return -EINVAL;
2829 		ib_spec->esp.size = sizeof(struct ib_flow_spec_esp);
2830 		memcpy(&ib_spec->esp.val, kern_spec_val, actual_filter_sz);
2831 		memcpy(&ib_spec->esp.mask, kern_spec_mask, actual_filter_sz);
2832 		break;
2833 	case IB_FLOW_SPEC_GRE:
2834 		ib_filter_sz = offsetof(struct ib_flow_gre_filter, real_sz);
2835 		actual_filter_sz = spec_filter_size(kern_spec_mask,
2836 						    kern_filter_sz,
2837 						    ib_filter_sz);
2838 		if (actual_filter_sz <= 0)
2839 			return -EINVAL;
2840 		ib_spec->gre.size = sizeof(struct ib_flow_spec_gre);
2841 		memcpy(&ib_spec->gre.val, kern_spec_val, actual_filter_sz);
2842 		memcpy(&ib_spec->gre.mask, kern_spec_mask, actual_filter_sz);
2843 		break;
2844 	case IB_FLOW_SPEC_MPLS:
2845 		ib_filter_sz = offsetof(struct ib_flow_mpls_filter, real_sz);
2846 		actual_filter_sz = spec_filter_size(kern_spec_mask,
2847 						    kern_filter_sz,
2848 						    ib_filter_sz);
2849 		if (actual_filter_sz <= 0)
2850 			return -EINVAL;
2851 		ib_spec->mpls.size = sizeof(struct ib_flow_spec_mpls);
2852 		memcpy(&ib_spec->mpls.val, kern_spec_val, actual_filter_sz);
2853 		memcpy(&ib_spec->mpls.mask, kern_spec_mask, actual_filter_sz);
2854 		break;
2855 	default:
2856 		return -EINVAL;
2857 	}
2858 	return 0;
2859 }
2860 
2861 static int kern_spec_to_ib_spec_filter(struct ib_uverbs_flow_spec *kern_spec,
2862 				       union ib_flow_spec *ib_spec)
2863 {
2864 	size_t kern_filter_sz;
2865 	void *kern_spec_mask;
2866 	void *kern_spec_val;
2867 
2868 	if (kern_spec->hdr.size < sizeof(struct ib_uverbs_flow_spec_hdr))
2869 		return -EINVAL;
2870 	kern_filter_sz = kern_spec->hdr.size - sizeof(struct ib_uverbs_flow_spec_hdr);
2871 	kern_filter_sz /= 2;
2872 
2873 	kern_spec_val = (u8 *)kern_spec +
2874 		sizeof(struct ib_uverbs_flow_spec_hdr);
2875 	kern_spec_mask = (u8 *)kern_spec_val + kern_filter_sz;
2876 
2877 	return ib_uverbs_kern_spec_to_ib_spec_filter(kern_spec->type,
2878 						     kern_spec_mask,
2879 						     kern_spec_val,
2880 						     kern_filter_sz, ib_spec);
2881 }
2882 
2883 static int kern_spec_to_ib_spec(struct uverbs_attr_bundle *attrs,
2884 				struct ib_uverbs_flow_spec *kern_spec,
2885 				union ib_flow_spec *ib_spec,
2886 				struct ib_uflow_resources *uflow_res)
2887 {
2888 	if (kern_spec->reserved)
2889 		return -EINVAL;
2890 
2891 	if (kern_spec->type >= IB_FLOW_SPEC_ACTION_TAG)
2892 		return kern_spec_to_ib_spec_action(attrs, kern_spec, ib_spec,
2893 						   uflow_res);
2894 	else
2895 		return kern_spec_to_ib_spec_filter(kern_spec, ib_spec);
2896 }
2897 
2898 static int ib_uverbs_ex_create_wq(struct uverbs_attr_bundle *attrs)
2899 {
2900 	struct ib_uverbs_ex_create_wq cmd;
2901 	struct ib_uverbs_ex_create_wq_resp resp = {};
2902 	struct ib_uwq_object           *obj;
2903 	int err = 0;
2904 	struct ib_cq *cq;
2905 	struct ib_pd *pd;
2906 	struct ib_wq *wq;
2907 	struct ib_wq_init_attr wq_init_attr = {};
2908 	struct ib_device *ib_dev;
2909 
2910 	err = uverbs_request(attrs, &cmd, sizeof(cmd));
2911 	if (err)
2912 		return err;
2913 
2914 	if (cmd.comp_mask)
2915 		return -EOPNOTSUPP;
2916 
2917 	obj = (struct ib_uwq_object *)uobj_alloc(UVERBS_OBJECT_WQ, attrs,
2918 						 &ib_dev);
2919 	if (IS_ERR(obj))
2920 		return PTR_ERR(obj);
2921 
2922 	pd = uobj_get_obj_read(pd, UVERBS_OBJECT_PD, cmd.pd_handle, attrs);
2923 	if (!pd) {
2924 		err = -EINVAL;
2925 		goto err_uobj;
2926 	}
2927 
2928 	cq = uobj_get_obj_read(cq, UVERBS_OBJECT_CQ, cmd.cq_handle, attrs);
2929 	if (!cq) {
2930 		err = -EINVAL;
2931 		goto err_put_pd;
2932 	}
2933 
2934 	wq_init_attr.cq = cq;
2935 	wq_init_attr.max_sge = cmd.max_sge;
2936 	wq_init_attr.max_wr = cmd.max_wr;
2937 	wq_init_attr.wq_context = attrs->ufile;
2938 	wq_init_attr.wq_type = cmd.wq_type;
2939 	wq_init_attr.event_handler = ib_uverbs_wq_event_handler;
2940 	wq_init_attr.create_flags = cmd.create_flags;
2941 	INIT_LIST_HEAD(&obj->uevent.event_list);
2942 
2943 	wq = pd->device->create_wq(pd, &wq_init_attr, &attrs->driver_udata);
2944 	if (IS_ERR(wq)) {
2945 		err = PTR_ERR(wq);
2946 		goto err_put_cq;
2947 	}
2948 
2949 	wq->uobject = obj;
2950 	obj->uevent.uobject.object = wq;
2951 	wq->wq_type = wq_init_attr.wq_type;
2952 	wq->cq = cq;
2953 	wq->pd = pd;
2954 	wq->device = pd->device;
2955 	wq->wq_context = wq_init_attr.wq_context;
2956 	atomic_set(&wq->usecnt, 0);
2957 	atomic_inc(&pd->usecnt);
2958 	atomic_inc(&cq->usecnt);
2959 	wq->uobject = obj;
2960 	obj->uevent.uobject.object = wq;
2961 
2962 	memset(&resp, 0, sizeof(resp));
2963 	resp.wq_handle = obj->uevent.uobject.id;
2964 	resp.max_sge = wq_init_attr.max_sge;
2965 	resp.max_wr = wq_init_attr.max_wr;
2966 	resp.wqn = wq->wq_num;
2967 	resp.response_length = uverbs_response_length(attrs, sizeof(resp));
2968 	err = uverbs_response(attrs, &resp, sizeof(resp));
2969 	if (err)
2970 		goto err_copy;
2971 
2972 	uobj_put_obj_read(pd);
2973 	rdma_lookup_put_uobject(&cq->uobject->uevent.uobject,
2974 				UVERBS_LOOKUP_READ);
2975 	rdma_alloc_commit_uobject(&obj->uevent.uobject, attrs);
2976 	return 0;
2977 
2978 err_copy:
2979 	ib_destroy_wq(wq, uverbs_get_cleared_udata(attrs));
2980 err_put_cq:
2981 	rdma_lookup_put_uobject(&cq->uobject->uevent.uobject,
2982 				UVERBS_LOOKUP_READ);
2983 err_put_pd:
2984 	uobj_put_obj_read(pd);
2985 err_uobj:
2986 	uobj_alloc_abort(&obj->uevent.uobject, attrs);
2987 
2988 	return err;
2989 }
2990 
2991 static int ib_uverbs_ex_destroy_wq(struct uverbs_attr_bundle *attrs)
2992 {
2993 	struct ib_uverbs_ex_destroy_wq	cmd;
2994 	struct ib_uverbs_ex_destroy_wq_resp	resp = {};
2995 	struct ib_uobject		*uobj;
2996 	struct ib_uwq_object		*obj;
2997 	int				ret;
2998 
2999 	ret = uverbs_request(attrs, &cmd, sizeof(cmd));
3000 	if (ret)
3001 		return ret;
3002 
3003 	if (cmd.comp_mask)
3004 		return -EOPNOTSUPP;
3005 
3006 	resp.response_length = uverbs_response_length(attrs, sizeof(resp));
3007 	uobj = uobj_get_destroy(UVERBS_OBJECT_WQ, cmd.wq_handle, attrs);
3008 	if (IS_ERR(uobj))
3009 		return PTR_ERR(uobj);
3010 
3011 	obj = container_of(uobj, struct ib_uwq_object, uevent.uobject);
3012 	resp.events_reported = obj->uevent.events_reported;
3013 
3014 	uobj_put_destroy(uobj);
3015 
3016 	return uverbs_response(attrs, &resp, sizeof(resp));
3017 }
3018 
3019 static int ib_uverbs_ex_modify_wq(struct uverbs_attr_bundle *attrs)
3020 {
3021 	struct ib_uverbs_ex_modify_wq cmd;
3022 	struct ib_wq *wq;
3023 	struct ib_wq_attr wq_attr = {};
3024 	int ret;
3025 
3026 	ret = uverbs_request(attrs, &cmd, sizeof(cmd));
3027 	if (ret)
3028 		return ret;
3029 
3030 	if (!cmd.attr_mask)
3031 		return -EINVAL;
3032 
3033 	if (cmd.attr_mask > (IB_WQ_STATE | IB_WQ_CUR_STATE | IB_WQ_FLAGS))
3034 		return -EINVAL;
3035 
3036 	wq = uobj_get_obj_read(wq, UVERBS_OBJECT_WQ, cmd.wq_handle, attrs);
3037 	if (!wq)
3038 		return -EINVAL;
3039 
3040 	wq_attr.curr_wq_state = cmd.curr_wq_state;
3041 	wq_attr.wq_state = cmd.wq_state;
3042 	if (cmd.attr_mask & IB_WQ_FLAGS) {
3043 		wq_attr.flags = cmd.flags;
3044 		wq_attr.flags_mask = cmd.flags_mask;
3045 	}
3046 	ret = wq->device->modify_wq(wq, &wq_attr, cmd.attr_mask,
3047 					&attrs->driver_udata);
3048 	rdma_lookup_put_uobject(&wq->uobject->uevent.uobject,
3049 				UVERBS_LOOKUP_READ);
3050 	return ret;
3051 }
3052 
3053 static int ib_uverbs_ex_create_rwq_ind_table(struct uverbs_attr_bundle *attrs)
3054 {
3055 	struct ib_uverbs_ex_create_rwq_ind_table cmd;
3056 	struct ib_uverbs_ex_create_rwq_ind_table_resp  resp = {};
3057 	struct ib_uobject		  *uobj;
3058 	int err;
3059 	struct ib_rwq_ind_table_init_attr init_attr = {};
3060 	struct ib_rwq_ind_table *rwq_ind_tbl;
3061 	struct ib_wq	**wqs = NULL;
3062 	u32 *wqs_handles = NULL;
3063 	struct ib_wq	*wq = NULL;
3064 	int i, j, num_read_wqs;
3065 	u32 num_wq_handles;
3066 	struct uverbs_req_iter iter;
3067 	struct ib_device *ib_dev;
3068 
3069 	err = uverbs_request_start(attrs, &iter, &cmd, sizeof(cmd));
3070 	if (err)
3071 		return err;
3072 
3073 	if (cmd.comp_mask)
3074 		return -EOPNOTSUPP;
3075 
3076 	if (cmd.log_ind_tbl_size > IB_USER_VERBS_MAX_LOG_IND_TBL_SIZE)
3077 		return -EINVAL;
3078 
3079 	num_wq_handles = 1 << cmd.log_ind_tbl_size;
3080 	wqs_handles = kcalloc(num_wq_handles, sizeof(*wqs_handles),
3081 			      GFP_KERNEL);
3082 	if (!wqs_handles)
3083 		return -ENOMEM;
3084 
3085 	err = uverbs_request_next(&iter, wqs_handles,
3086 				  num_wq_handles * sizeof(__u32));
3087 	if (err)
3088 		goto err_free;
3089 
3090 	err = uverbs_request_finish(&iter);
3091 	if (err)
3092 		goto err_free;
3093 
3094 	wqs = kcalloc(num_wq_handles, sizeof(*wqs), GFP_KERNEL);
3095 	if (!wqs) {
3096 		err = -ENOMEM;
3097 		goto  err_free;
3098 	}
3099 
3100 	for (num_read_wqs = 0; num_read_wqs < num_wq_handles;
3101 			num_read_wqs++) {
3102 		wq = uobj_get_obj_read(wq, UVERBS_OBJECT_WQ,
3103 				       wqs_handles[num_read_wqs], attrs);
3104 		if (!wq) {
3105 			err = -EINVAL;
3106 			goto put_wqs;
3107 		}
3108 
3109 		wqs[num_read_wqs] = wq;
3110 	}
3111 
3112 	uobj = uobj_alloc(UVERBS_OBJECT_RWQ_IND_TBL, attrs, &ib_dev);
3113 	if (IS_ERR(uobj)) {
3114 		err = PTR_ERR(uobj);
3115 		goto put_wqs;
3116 	}
3117 
3118 	init_attr.log_ind_tbl_size = cmd.log_ind_tbl_size;
3119 	init_attr.ind_tbl = wqs;
3120 
3121 	rwq_ind_tbl = ib_dev->create_rwq_ind_table(ib_dev, &init_attr,
3122 						       &attrs->driver_udata);
3123 
3124 	if (IS_ERR(rwq_ind_tbl)) {
3125 		err = PTR_ERR(rwq_ind_tbl);
3126 		goto err_uobj;
3127 	}
3128 
3129 	rwq_ind_tbl->ind_tbl = wqs;
3130 	rwq_ind_tbl->log_ind_tbl_size = init_attr.log_ind_tbl_size;
3131 	rwq_ind_tbl->uobject = uobj;
3132 	uobj->object = rwq_ind_tbl;
3133 	rwq_ind_tbl->device = ib_dev;
3134 	atomic_set(&rwq_ind_tbl->usecnt, 0);
3135 
3136 	for (i = 0; i < num_wq_handles; i++)
3137 		atomic_inc(&wqs[i]->usecnt);
3138 
3139 	resp.ind_tbl_handle = uobj->id;
3140 	resp.ind_tbl_num = rwq_ind_tbl->ind_tbl_num;
3141 	resp.response_length = uverbs_response_length(attrs, sizeof(resp));
3142 
3143 	err = uverbs_response(attrs, &resp, sizeof(resp));
3144 	if (err)
3145 		goto err_copy;
3146 
3147 	kfree(wqs_handles);
3148 
3149 	for (j = 0; j < num_read_wqs; j++)
3150 		rdma_lookup_put_uobject(&wqs[j]->uobject->uevent.uobject,
3151 					UVERBS_LOOKUP_READ);
3152 
3153 	rdma_alloc_commit_uobject(uobj, attrs);
3154 	return 0;
3155 
3156 err_copy:
3157 	ib_destroy_rwq_ind_table(rwq_ind_tbl);
3158 err_uobj:
3159 	uobj_alloc_abort(uobj, attrs);
3160 put_wqs:
3161 	for (j = 0; j < num_read_wqs; j++)
3162 		rdma_lookup_put_uobject(&wqs[j]->uobject->uevent.uobject,
3163 					UVERBS_LOOKUP_READ);
3164 err_free:
3165 	kfree(wqs_handles);
3166 	kfree(wqs);
3167 	return err;
3168 }
3169 
3170 static int ib_uverbs_ex_destroy_rwq_ind_table(struct uverbs_attr_bundle *attrs)
3171 {
3172 	struct ib_uverbs_ex_destroy_rwq_ind_table cmd;
3173 	int ret;
3174 
3175 	ret = uverbs_request(attrs, &cmd, sizeof(cmd));
3176 	if (ret)
3177 		return ret;
3178 
3179 	if (cmd.comp_mask)
3180 		return -EOPNOTSUPP;
3181 
3182 	return uobj_perform_destroy(UVERBS_OBJECT_RWQ_IND_TBL,
3183 				    cmd.ind_tbl_handle, attrs);
3184 }
3185 
3186 static int ib_uverbs_ex_create_flow(struct uverbs_attr_bundle *attrs)
3187 {
3188 	struct ib_uverbs_create_flow	  cmd;
3189 	struct ib_uverbs_create_flow_resp resp;
3190 	struct ib_uobject		  *uobj;
3191 	struct ib_flow			  *flow_id;
3192 	struct ib_uverbs_flow_attr	  *kern_flow_attr;
3193 	struct ib_flow_attr		  *flow_attr;
3194 	struct ib_qp			  *qp;
3195 	struct ib_uflow_resources	  *uflow_res;
3196 	struct ib_uverbs_flow_spec_hdr	  *kern_spec;
3197 	struct uverbs_req_iter iter;
3198 	int err;
3199 	void *ib_spec;
3200 	int i;
3201 	struct ib_device *ib_dev;
3202 
3203 	err = uverbs_request_start(attrs, &iter, &cmd, sizeof(cmd));
3204 	if (err)
3205 		return err;
3206 
3207 	if (cmd.comp_mask)
3208 		return -EINVAL;
3209 
3210 	if (priv_check(curthread, PRIV_NET_RAW) != 0)
3211 		return -EPERM;
3212 
3213 	if (cmd.flow_attr.flags >= IB_FLOW_ATTR_FLAGS_RESERVED)
3214 		return -EINVAL;
3215 
3216 	if ((cmd.flow_attr.flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP) &&
3217 	    ((cmd.flow_attr.type == IB_FLOW_ATTR_ALL_DEFAULT) ||
3218 	     (cmd.flow_attr.type == IB_FLOW_ATTR_MC_DEFAULT)))
3219 		return -EINVAL;
3220 
3221 	if (cmd.flow_attr.num_of_specs > IB_FLOW_SPEC_SUPPORT_LAYERS)
3222 		return -EINVAL;
3223 
3224 	if (cmd.flow_attr.size >
3225 	    (cmd.flow_attr.num_of_specs * sizeof(struct ib_uverbs_flow_spec)))
3226 		return -EINVAL;
3227 
3228 	if (cmd.flow_attr.reserved[0] ||
3229 	    cmd.flow_attr.reserved[1])
3230 		return -EINVAL;
3231 
3232 	if (cmd.flow_attr.num_of_specs) {
3233 		kern_flow_attr = kmalloc(sizeof(*kern_flow_attr) + cmd.flow_attr.size,
3234 					 GFP_KERNEL);
3235 		if (!kern_flow_attr)
3236 			return -ENOMEM;
3237 
3238 		*kern_flow_attr = cmd.flow_attr;
3239 		err = uverbs_request_next(&iter, &kern_flow_attr->flow_specs,
3240 					  cmd.flow_attr.size);
3241 		if (err)
3242 			goto err_free_attr;
3243 	} else {
3244 		kern_flow_attr = &cmd.flow_attr;
3245 	}
3246 
3247 	err = uverbs_request_finish(&iter);
3248 	if (err)
3249 		goto err_free_attr;
3250 
3251 	uobj = uobj_alloc(UVERBS_OBJECT_FLOW, attrs, &ib_dev);
3252 	if (IS_ERR(uobj)) {
3253 		err = PTR_ERR(uobj);
3254 		goto err_free_attr;
3255 	}
3256 
3257 	qp = uobj_get_obj_read(qp, UVERBS_OBJECT_QP, cmd.qp_handle, attrs);
3258 	if (!qp) {
3259 		err = -EINVAL;
3260 		goto err_uobj;
3261 	}
3262 
3263 	if (qp->qp_type != IB_QPT_UD && qp->qp_type != IB_QPT_RAW_PACKET) {
3264 		err = -EINVAL;
3265 		goto err_put;
3266 	}
3267 
3268 	flow_attr = kzalloc(struct_size(flow_attr, flows,
3269 				cmd.flow_attr.num_of_specs), GFP_KERNEL);
3270 	if (!flow_attr) {
3271 		err = -ENOMEM;
3272 		goto err_put;
3273 	}
3274 	uflow_res = flow_resources_alloc(cmd.flow_attr.num_of_specs);
3275 	if (!uflow_res) {
3276 		err = -ENOMEM;
3277 		goto err_free_flow_attr;
3278 	}
3279 
3280 	flow_attr->type = kern_flow_attr->type;
3281 	flow_attr->priority = kern_flow_attr->priority;
3282 	flow_attr->num_of_specs = kern_flow_attr->num_of_specs;
3283 	flow_attr->port = kern_flow_attr->port;
3284 	flow_attr->flags = kern_flow_attr->flags;
3285 	flow_attr->size = sizeof(*flow_attr);
3286 
3287 	kern_spec = kern_flow_attr->flow_specs;
3288 	ib_spec = flow_attr + 1;
3289 	for (i = 0; i < flow_attr->num_of_specs &&
3290 			cmd.flow_attr.size >= sizeof(*kern_spec) &&
3291 			cmd.flow_attr.size >= kern_spec->size;
3292 	     i++) {
3293 		err = kern_spec_to_ib_spec(
3294 				attrs, (struct ib_uverbs_flow_spec *)kern_spec,
3295 				ib_spec, uflow_res);
3296 		if (err)
3297 			goto err_free;
3298 
3299 		flow_attr->size +=
3300 			((union ib_flow_spec *) ib_spec)->size;
3301 		cmd.flow_attr.size -= kern_spec->size;
3302 		kern_spec = (struct ib_uverbs_flow_spec_hdr *)((u8 *)kern_spec + kern_spec->size);
3303 		ib_spec = (u8 *)ib_spec + ((union ib_flow_spec *) ib_spec)->size;
3304 	}
3305 	if (cmd.flow_attr.size || (i != flow_attr->num_of_specs)) {
3306 		pr_warn("create flow failed, flow %d: %d bytes left from uverb cmd\n",
3307 			i, cmd.flow_attr.size);
3308 		err = -EINVAL;
3309 		goto err_free;
3310 	}
3311 
3312 	flow_id = qp->device->create_flow(
3313 		qp, flow_attr, IB_FLOW_DOMAIN_USER, &attrs->driver_udata);
3314 
3315 	if (IS_ERR(flow_id)) {
3316 		err = PTR_ERR(flow_id);
3317 		goto err_free;
3318 	}
3319 
3320 	ib_set_flow(uobj, flow_id, qp, qp->device, uflow_res);
3321 
3322 	memset(&resp, 0, sizeof(resp));
3323 	resp.flow_handle = uobj->id;
3324 
3325 	err = uverbs_response(attrs, &resp, sizeof(resp));
3326 	if (err)
3327 		goto err_copy;
3328 
3329 	rdma_lookup_put_uobject(&qp->uobject->uevent.uobject,
3330 				UVERBS_LOOKUP_READ);
3331 	kfree(flow_attr);
3332 	if (cmd.flow_attr.num_of_specs)
3333 		kfree(kern_flow_attr);
3334 	rdma_alloc_commit_uobject(uobj, attrs);
3335 	return 0;
3336 err_copy:
3337 	if (!qp->device->destroy_flow(flow_id))
3338 		atomic_dec(&qp->usecnt);
3339 err_free:
3340 	ib_uverbs_flow_resources_free(uflow_res);
3341 err_free_flow_attr:
3342 	kfree(flow_attr);
3343 err_put:
3344 	rdma_lookup_put_uobject(&qp->uobject->uevent.uobject,
3345 				UVERBS_LOOKUP_READ);
3346 err_uobj:
3347 	uobj_alloc_abort(uobj, attrs);
3348 err_free_attr:
3349 	if (cmd.flow_attr.num_of_specs)
3350 		kfree(kern_flow_attr);
3351 	return err;
3352 }
3353 
3354 static int ib_uverbs_ex_destroy_flow(struct uverbs_attr_bundle *attrs)
3355 {
3356 	struct ib_uverbs_destroy_flow	cmd;
3357 	int				ret;
3358 
3359 	ret = uverbs_request(attrs, &cmd, sizeof(cmd));
3360 	if (ret)
3361 		return ret;
3362 
3363 	if (cmd.comp_mask)
3364 		return -EINVAL;
3365 
3366 	return uobj_perform_destroy(UVERBS_OBJECT_FLOW, cmd.flow_handle, attrs);
3367 }
3368 
3369 static int __uverbs_create_xsrq(struct uverbs_attr_bundle *attrs,
3370 				struct ib_uverbs_create_xsrq *cmd,
3371 				struct ib_udata *udata)
3372 {
3373 	struct ib_uverbs_create_srq_resp resp;
3374 	struct ib_usrq_object           *obj;
3375 	struct ib_pd                    *pd;
3376 	struct ib_srq                   *srq;
3377 	struct ib_uobject               *uninitialized_var(xrcd_uobj);
3378 	struct ib_srq_init_attr          attr;
3379 	int ret;
3380 	struct ib_device *ib_dev;
3381 
3382 	obj = (struct ib_usrq_object *)uobj_alloc(UVERBS_OBJECT_SRQ, attrs,
3383 						  &ib_dev);
3384 	if (IS_ERR(obj))
3385 		return PTR_ERR(obj);
3386 
3387 	if (cmd->srq_type == IB_SRQT_TM)
3388 		attr.ext.tag_matching.max_num_tags = cmd->max_num_tags;
3389 
3390 	if (cmd->srq_type == IB_SRQT_XRC) {
3391 		xrcd_uobj = uobj_get_read(UVERBS_OBJECT_XRCD, cmd->xrcd_handle,
3392 					  attrs);
3393 		if (IS_ERR(xrcd_uobj)) {
3394 			ret = -EINVAL;
3395 			goto err;
3396 		}
3397 
3398 		attr.ext.xrc.xrcd = (struct ib_xrcd *)xrcd_uobj->object;
3399 		if (!attr.ext.xrc.xrcd) {
3400 			ret = -EINVAL;
3401 			goto err_put_xrcd;
3402 		}
3403 
3404 		obj->uxrcd = container_of(xrcd_uobj, struct ib_uxrcd_object, uobject);
3405 		atomic_inc(&obj->uxrcd->refcnt);
3406 	}
3407 
3408 	if (ib_srq_has_cq(cmd->srq_type)) {
3409 		attr.ext.cq = uobj_get_obj_read(cq, UVERBS_OBJECT_CQ,
3410 						cmd->cq_handle, attrs);
3411 		if (!attr.ext.cq) {
3412 			ret = -EINVAL;
3413 			goto err_put_xrcd;
3414 		}
3415 	}
3416 
3417 	pd = uobj_get_obj_read(pd, UVERBS_OBJECT_PD, cmd->pd_handle, attrs);
3418 	if (!pd) {
3419 		ret = -EINVAL;
3420 		goto err_put_cq;
3421 	}
3422 
3423 	attr.event_handler  = ib_uverbs_srq_event_handler;
3424 	attr.srq_context    = attrs->ufile;
3425 	attr.srq_type       = cmd->srq_type;
3426 	attr.attr.max_wr    = cmd->max_wr;
3427 	attr.attr.max_sge   = cmd->max_sge;
3428 	attr.attr.srq_limit = cmd->srq_limit;
3429 
3430 	INIT_LIST_HEAD(&obj->uevent.event_list);
3431 
3432 	srq = rdma_zalloc_drv_obj(ib_dev, ib_srq);
3433 	if (!srq) {
3434 		ret = -ENOMEM;
3435 		goto err_put;
3436 	}
3437 
3438 	srq->device        = pd->device;
3439 	srq->pd            = pd;
3440 	srq->srq_type	   = cmd->srq_type;
3441 	srq->uobject       = obj;
3442 	srq->event_handler = attr.event_handler;
3443 	srq->srq_context   = attr.srq_context;
3444 
3445 	ret = pd->device->create_srq(srq, &attr, udata);
3446 	if (ret)
3447 		goto err_free;
3448 
3449 	if (ib_srq_has_cq(cmd->srq_type)) {
3450 		srq->ext.cq       = attr.ext.cq;
3451 		atomic_inc(&attr.ext.cq->usecnt);
3452 	}
3453 
3454 	if (cmd->srq_type == IB_SRQT_XRC) {
3455 		srq->ext.xrc.xrcd = attr.ext.xrc.xrcd;
3456 		atomic_inc(&attr.ext.xrc.xrcd->usecnt);
3457 	}
3458 
3459 	atomic_inc(&pd->usecnt);
3460 	atomic_set(&srq->usecnt, 0);
3461 
3462 	obj->uevent.uobject.object = srq;
3463 	obj->uevent.uobject.user_handle = cmd->user_handle;
3464 
3465 	memset(&resp, 0, sizeof resp);
3466 	resp.srq_handle = obj->uevent.uobject.id;
3467 	resp.max_wr     = attr.attr.max_wr;
3468 	resp.max_sge    = attr.attr.max_sge;
3469 	if (cmd->srq_type == IB_SRQT_XRC)
3470 		resp.srqn = srq->ext.xrc.srq_num;
3471 
3472 	ret = uverbs_response(attrs, &resp, sizeof(resp));
3473 	if (ret)
3474 		goto err_copy;
3475 
3476 	if (cmd->srq_type == IB_SRQT_XRC)
3477 		uobj_put_read(xrcd_uobj);
3478 
3479 	if (ib_srq_has_cq(cmd->srq_type))
3480 		rdma_lookup_put_uobject(&attr.ext.cq->uobject->uevent.uobject,
3481 					UVERBS_LOOKUP_READ);
3482 
3483 	uobj_put_obj_read(pd);
3484 	rdma_alloc_commit_uobject(&obj->uevent.uobject, attrs);
3485 	return 0;
3486 
3487 err_copy:
3488 	ib_destroy_srq_user(srq, uverbs_get_cleared_udata(attrs));
3489 	/* It was released in ib_destroy_srq_user */
3490 	srq = NULL;
3491 err_free:
3492 	kfree(srq);
3493 err_put:
3494 	uobj_put_obj_read(pd);
3495 
3496 err_put_cq:
3497 	if (ib_srq_has_cq(cmd->srq_type))
3498 		rdma_lookup_put_uobject(&attr.ext.cq->uobject->uevent.uobject,
3499 					UVERBS_LOOKUP_READ);
3500 
3501 err_put_xrcd:
3502 	if (cmd->srq_type == IB_SRQT_XRC) {
3503 		atomic_dec(&obj->uxrcd->refcnt);
3504 		uobj_put_read(xrcd_uobj);
3505 	}
3506 
3507 err:
3508 	uobj_alloc_abort(&obj->uevent.uobject, attrs);
3509 	return ret;
3510 }
3511 
3512 static int ib_uverbs_create_srq(struct uverbs_attr_bundle *attrs)
3513 {
3514 	struct ib_uverbs_create_srq      cmd;
3515 	struct ib_uverbs_create_xsrq     xcmd;
3516 	int ret;
3517 
3518 	ret = uverbs_request(attrs, &cmd, sizeof(cmd));
3519 	if (ret)
3520 		return ret;
3521 
3522 	memset(&xcmd, 0, sizeof(xcmd));
3523 	xcmd.response	 = cmd.response;
3524 	xcmd.user_handle = cmd.user_handle;
3525 	xcmd.srq_type	 = IB_SRQT_BASIC;
3526 	xcmd.pd_handle	 = cmd.pd_handle;
3527 	xcmd.max_wr	 = cmd.max_wr;
3528 	xcmd.max_sge	 = cmd.max_sge;
3529 	xcmd.srq_limit	 = cmd.srq_limit;
3530 
3531 	return __uverbs_create_xsrq(attrs, &xcmd, &attrs->driver_udata);
3532 }
3533 
3534 static int ib_uverbs_create_xsrq(struct uverbs_attr_bundle *attrs)
3535 {
3536 	struct ib_uverbs_create_xsrq     cmd;
3537 	int ret;
3538 
3539 	ret = uverbs_request(attrs, &cmd, sizeof(cmd));
3540 	if (ret)
3541 		return ret;
3542 
3543 	return __uverbs_create_xsrq(attrs, &cmd, &attrs->driver_udata);
3544 }
3545 
3546 static int ib_uverbs_modify_srq(struct uverbs_attr_bundle *attrs)
3547 {
3548 	struct ib_uverbs_modify_srq cmd;
3549 	struct ib_srq              *srq;
3550 	struct ib_srq_attr          attr;
3551 	int                         ret;
3552 
3553 	ret = uverbs_request(attrs, &cmd, sizeof(cmd));
3554 	if (ret)
3555 		return ret;
3556 
3557 	srq = uobj_get_obj_read(srq, UVERBS_OBJECT_SRQ, cmd.srq_handle, attrs);
3558 	if (!srq)
3559 		return -EINVAL;
3560 
3561 	attr.max_wr    = cmd.max_wr;
3562 	attr.srq_limit = cmd.srq_limit;
3563 
3564 	ret = srq->device->modify_srq(srq, &attr, cmd.attr_mask,
3565 					  &attrs->driver_udata);
3566 
3567 	rdma_lookup_put_uobject(&srq->uobject->uevent.uobject,
3568 				UVERBS_LOOKUP_READ);
3569 
3570 	return ret;
3571 }
3572 
3573 static int ib_uverbs_query_srq(struct uverbs_attr_bundle *attrs)
3574 {
3575 	struct ib_uverbs_query_srq      cmd;
3576 	struct ib_uverbs_query_srq_resp resp;
3577 	struct ib_srq_attr              attr;
3578 	struct ib_srq                   *srq;
3579 	int                             ret;
3580 
3581 	ret = uverbs_request(attrs, &cmd, sizeof(cmd));
3582 	if (ret)
3583 		return ret;
3584 
3585 	srq = uobj_get_obj_read(srq, UVERBS_OBJECT_SRQ, cmd.srq_handle, attrs);
3586 	if (!srq)
3587 		return -EINVAL;
3588 
3589 	ret = ib_query_srq(srq, &attr);
3590 
3591 	rdma_lookup_put_uobject(&srq->uobject->uevent.uobject,
3592 				UVERBS_LOOKUP_READ);
3593 
3594 	if (ret)
3595 		return ret;
3596 
3597 	memset(&resp, 0, sizeof resp);
3598 
3599 	resp.max_wr    = attr.max_wr;
3600 	resp.max_sge   = attr.max_sge;
3601 	resp.srq_limit = attr.srq_limit;
3602 
3603 	return uverbs_response(attrs, &resp, sizeof(resp));
3604 }
3605 
3606 static int ib_uverbs_destroy_srq(struct uverbs_attr_bundle *attrs)
3607 {
3608 	struct ib_uverbs_destroy_srq      cmd;
3609 	struct ib_uverbs_destroy_srq_resp resp;
3610 	struct ib_uobject		 *uobj;
3611 	struct ib_uevent_object        	 *obj;
3612 	int ret;
3613 
3614 	ret = uverbs_request(attrs, &cmd, sizeof(cmd));
3615 	if (ret)
3616 		return ret;
3617 
3618 	uobj = uobj_get_destroy(UVERBS_OBJECT_SRQ, cmd.srq_handle, attrs);
3619 	if (IS_ERR(uobj))
3620 		return PTR_ERR(uobj);
3621 
3622 	obj = container_of(uobj, struct ib_uevent_object, uobject);
3623 	memset(&resp, 0, sizeof(resp));
3624 	resp.events_reported = obj->events_reported;
3625 
3626 	uobj_put_destroy(uobj);
3627 
3628 	return uverbs_response(attrs, &resp, sizeof(resp));
3629 }
3630 
3631 static int ib_uverbs_ex_query_device(struct uverbs_attr_bundle *attrs)
3632 {
3633 	struct ib_uverbs_ex_query_device_resp resp = {};
3634 	struct ib_uverbs_ex_query_device  cmd;
3635 	struct ib_device_attr attr = {0};
3636 	struct ib_ucontext *ucontext;
3637 	struct ib_device *ib_dev;
3638 	int err;
3639 
3640 	ucontext = ib_uverbs_get_ucontext(attrs);
3641 	if (IS_ERR(ucontext))
3642 		return PTR_ERR(ucontext);
3643 	ib_dev = ucontext->device;
3644 
3645 	err = uverbs_request(attrs, &cmd, sizeof(cmd));
3646 	if (err)
3647 		return err;
3648 
3649 	if (cmd.comp_mask)
3650 		return -EINVAL;
3651 
3652 	if (cmd.reserved)
3653 		return -EINVAL;
3654 
3655 	err = ib_dev->query_device(ib_dev, &attr, &attrs->driver_udata);
3656 	if (err)
3657 		return err;
3658 
3659 	copy_query_dev_fields(ucontext, &resp.base, &attr);
3660 
3661 	resp.odp_caps.general_caps = attr.odp_caps.general_caps;
3662 	resp.odp_caps.per_transport_caps.rc_odp_caps =
3663 		attr.odp_caps.per_transport_caps.rc_odp_caps;
3664 	resp.odp_caps.per_transport_caps.uc_odp_caps =
3665 		attr.odp_caps.per_transport_caps.uc_odp_caps;
3666 	resp.odp_caps.per_transport_caps.ud_odp_caps =
3667 		attr.odp_caps.per_transport_caps.ud_odp_caps;
3668 	resp.xrc_odp_caps = attr.odp_caps.per_transport_caps.xrc_odp_caps;
3669 
3670 	resp.timestamp_mask = attr.timestamp_mask;
3671 	resp.hca_core_clock = attr.hca_core_clock;
3672 	resp.device_cap_flags_ex = attr.device_cap_flags;
3673 	resp.rss_caps.supported_qpts = attr.rss_caps.supported_qpts;
3674 	resp.rss_caps.max_rwq_indirection_tables =
3675 		attr.rss_caps.max_rwq_indirection_tables;
3676 	resp.rss_caps.max_rwq_indirection_table_size =
3677 		attr.rss_caps.max_rwq_indirection_table_size;
3678 	resp.max_wq_type_rq = attr.max_wq_type_rq;
3679 	resp.raw_packet_caps = attr.raw_packet_caps;
3680 	resp.tm_caps.max_rndv_hdr_size	= attr.tm_caps.max_rndv_hdr_size;
3681 	resp.tm_caps.max_num_tags	= attr.tm_caps.max_num_tags;
3682 	resp.tm_caps.max_ops		= attr.tm_caps.max_ops;
3683 	resp.tm_caps.max_sge		= attr.tm_caps.max_sge;
3684 	resp.tm_caps.flags		= attr.tm_caps.flags;
3685 	resp.cq_moderation_caps.max_cq_moderation_count  =
3686 		attr.cq_caps.max_cq_moderation_count;
3687 	resp.cq_moderation_caps.max_cq_moderation_period =
3688 		attr.cq_caps.max_cq_moderation_period;
3689 	resp.max_dm_size = attr.max_dm_size;
3690 	resp.response_length = uverbs_response_length(attrs, sizeof(resp));
3691 
3692 	return uverbs_response(attrs, &resp, sizeof(resp));
3693 }
3694 
3695 static int ib_uverbs_ex_modify_cq(struct uverbs_attr_bundle *attrs)
3696 {
3697 	struct ib_uverbs_ex_modify_cq cmd;
3698 	struct ib_cq *cq;
3699 	int ret;
3700 
3701 	ret = uverbs_request(attrs, &cmd, sizeof(cmd));
3702 	if (ret)
3703 		return ret;
3704 
3705 	if (!cmd.attr_mask || cmd.reserved)
3706 		return -EINVAL;
3707 
3708 	if (cmd.attr_mask > IB_CQ_MODERATE)
3709 		return -EOPNOTSUPP;
3710 
3711 	cq = uobj_get_obj_read(cq, UVERBS_OBJECT_CQ, cmd.cq_handle, attrs);
3712 	if (!cq)
3713 		return -EINVAL;
3714 
3715 	ret = ib_modify_cq(cq, cmd.attr.cq_count, cmd.attr.cq_period);
3716 
3717 	rdma_lookup_put_uobject(&cq->uobject->uevent.uobject,
3718 				UVERBS_LOOKUP_READ);
3719 	return ret;
3720 }
3721 
3722 /*
3723  * Describe the input structs for write(). Some write methods have an input
3724  * only struct, most have an input and output. If the struct has an output then
3725  * the 'response' u64 must be the first field in the request structure.
3726  *
3727  * If udata is present then both the request and response structs have a
3728  * trailing driver_data flex array. In this case the size of the base struct
3729  * cannot be changed.
3730  */
3731 #define UAPI_DEF_WRITE_IO(req, resp)                                           \
3732 	.write.has_resp = 1 +                                                  \
3733 			  BUILD_BUG_ON_ZERO(offsetof(req, response) != 0) +    \
3734 			  BUILD_BUG_ON_ZERO(sizeof(((req *)0)->response) !=    \
3735 					    sizeof(u64)),                      \
3736 	.write.req_size = sizeof(req), .write.resp_size = sizeof(resp)
3737 
3738 #define UAPI_DEF_WRITE_I(req) .write.req_size = sizeof(req)
3739 
3740 #define UAPI_DEF_WRITE_UDATA_IO(req, resp)                                     \
3741 	UAPI_DEF_WRITE_IO(req, resp),                                          \
3742 		.write.has_udata =                                             \
3743 			1 +                                                    \
3744 			BUILD_BUG_ON_ZERO(offsetof(req, driver_data) !=        \
3745 					  sizeof(req)) +                       \
3746 			BUILD_BUG_ON_ZERO(offsetof(resp, driver_data) !=       \
3747 					  sizeof(resp))
3748 
3749 #define UAPI_DEF_WRITE_UDATA_I(req)                                            \
3750 	UAPI_DEF_WRITE_I(req),                                                 \
3751 		.write.has_udata =                                             \
3752 			1 + BUILD_BUG_ON_ZERO(offsetof(req, driver_data) !=    \
3753 					      sizeof(req))
3754 
3755 /*
3756  * The _EX versions are for use with WRITE_EX and allow the last struct member
3757  * to be specified. Buffers that do not include that member will be rejected.
3758  */
3759 #define UAPI_DEF_WRITE_IO_EX(req, req_last_member, resp, resp_last_member)     \
3760 	.write.has_resp = 1,                                                   \
3761 	.write.req_size = offsetofend(req, req_last_member),                   \
3762 	.write.resp_size = offsetofend(resp, resp_last_member)
3763 
3764 #define UAPI_DEF_WRITE_I_EX(req, req_last_member)                              \
3765 	.write.req_size = offsetofend(req, req_last_member)
3766 
3767 const struct uapi_definition uverbs_def_write_intf[] = {
3768 	DECLARE_UVERBS_OBJECT(
3769 		UVERBS_OBJECT_AH,
3770 		DECLARE_UVERBS_WRITE(IB_USER_VERBS_CMD_CREATE_AH,
3771 				     ib_uverbs_create_ah,
3772 				     UAPI_DEF_WRITE_UDATA_IO(
3773 					     struct ib_uverbs_create_ah,
3774 					     struct ib_uverbs_create_ah_resp),
3775 				     UAPI_DEF_METHOD_NEEDS_FN(create_ah)),
3776 		DECLARE_UVERBS_WRITE(
3777 			IB_USER_VERBS_CMD_DESTROY_AH,
3778 			ib_uverbs_destroy_ah,
3779 			UAPI_DEF_WRITE_I(struct ib_uverbs_destroy_ah),
3780 			UAPI_DEF_METHOD_NEEDS_FN(destroy_ah))),
3781 
3782 	DECLARE_UVERBS_OBJECT(
3783 		UVERBS_OBJECT_COMP_CHANNEL,
3784 		DECLARE_UVERBS_WRITE(
3785 			IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL,
3786 			ib_uverbs_create_comp_channel,
3787 			UAPI_DEF_WRITE_IO(
3788 				struct ib_uverbs_create_comp_channel,
3789 				struct ib_uverbs_create_comp_channel_resp))),
3790 
3791 	DECLARE_UVERBS_OBJECT(
3792 		UVERBS_OBJECT_CQ,
3793 		DECLARE_UVERBS_WRITE(IB_USER_VERBS_CMD_CREATE_CQ,
3794 				     ib_uverbs_create_cq,
3795 				     UAPI_DEF_WRITE_UDATA_IO(
3796 					     struct ib_uverbs_create_cq,
3797 					     struct ib_uverbs_create_cq_resp),
3798 				     UAPI_DEF_METHOD_NEEDS_FN(create_cq)),
3799 		DECLARE_UVERBS_WRITE(
3800 			IB_USER_VERBS_CMD_DESTROY_CQ,
3801 			ib_uverbs_destroy_cq,
3802 			UAPI_DEF_WRITE_IO(struct ib_uverbs_destroy_cq,
3803 					  struct ib_uverbs_destroy_cq_resp),
3804 			UAPI_DEF_METHOD_NEEDS_FN(destroy_cq)),
3805 		DECLARE_UVERBS_WRITE(
3806 			IB_USER_VERBS_CMD_POLL_CQ,
3807 			ib_uverbs_poll_cq,
3808 			UAPI_DEF_WRITE_IO(struct ib_uverbs_poll_cq,
3809 					  struct ib_uverbs_poll_cq_resp),
3810 			UAPI_DEF_METHOD_NEEDS_FN(poll_cq)),
3811 		DECLARE_UVERBS_WRITE(
3812 			IB_USER_VERBS_CMD_REQ_NOTIFY_CQ,
3813 			ib_uverbs_req_notify_cq,
3814 			UAPI_DEF_WRITE_I(struct ib_uverbs_req_notify_cq),
3815 			UAPI_DEF_METHOD_NEEDS_FN(req_notify_cq)),
3816 		DECLARE_UVERBS_WRITE(IB_USER_VERBS_CMD_RESIZE_CQ,
3817 				     ib_uverbs_resize_cq,
3818 				     UAPI_DEF_WRITE_UDATA_IO(
3819 					     struct ib_uverbs_resize_cq,
3820 					     struct ib_uverbs_resize_cq_resp),
3821 				     UAPI_DEF_METHOD_NEEDS_FN(resize_cq)),
3822 		DECLARE_UVERBS_WRITE_EX(
3823 			IB_USER_VERBS_EX_CMD_CREATE_CQ,
3824 			ib_uverbs_ex_create_cq,
3825 			UAPI_DEF_WRITE_IO_EX(struct ib_uverbs_ex_create_cq,
3826 					     reserved,
3827 					     struct ib_uverbs_ex_create_cq_resp,
3828 					     response_length),
3829 			UAPI_DEF_METHOD_NEEDS_FN(create_cq)),
3830 		DECLARE_UVERBS_WRITE_EX(
3831 			IB_USER_VERBS_EX_CMD_MODIFY_CQ,
3832 			ib_uverbs_ex_modify_cq,
3833 			UAPI_DEF_WRITE_I(struct ib_uverbs_ex_modify_cq),
3834 			UAPI_DEF_METHOD_NEEDS_FN(create_cq))),
3835 
3836 	DECLARE_UVERBS_OBJECT(
3837 		UVERBS_OBJECT_DEVICE,
3838 		DECLARE_UVERBS_WRITE(IB_USER_VERBS_CMD_GET_CONTEXT,
3839 				     ib_uverbs_get_context,
3840 				     UAPI_DEF_WRITE_UDATA_IO(
3841 					     struct ib_uverbs_get_context,
3842 					     struct ib_uverbs_get_context_resp)),
3843 		DECLARE_UVERBS_WRITE(
3844 			IB_USER_VERBS_CMD_QUERY_DEVICE,
3845 			ib_uverbs_query_device,
3846 			UAPI_DEF_WRITE_IO(struct ib_uverbs_query_device,
3847 					  struct ib_uverbs_query_device_resp)),
3848 		DECLARE_UVERBS_WRITE(
3849 			IB_USER_VERBS_CMD_QUERY_PORT,
3850 			ib_uverbs_query_port,
3851 			UAPI_DEF_WRITE_IO(struct ib_uverbs_query_port,
3852 					  struct ib_uverbs_query_port_resp),
3853 			UAPI_DEF_METHOD_NEEDS_FN(query_port)),
3854 		DECLARE_UVERBS_WRITE_EX(
3855 			IB_USER_VERBS_EX_CMD_QUERY_DEVICE,
3856 			ib_uverbs_ex_query_device,
3857 			UAPI_DEF_WRITE_IO_EX(
3858 				struct ib_uverbs_ex_query_device,
3859 				reserved,
3860 				struct ib_uverbs_ex_query_device_resp,
3861 				response_length),
3862 			UAPI_DEF_METHOD_NEEDS_FN(query_device)),
3863 		UAPI_DEF_OBJ_NEEDS_FN(alloc_ucontext),
3864 		UAPI_DEF_OBJ_NEEDS_FN(dealloc_ucontext)),
3865 
3866 	DECLARE_UVERBS_OBJECT(
3867 		UVERBS_OBJECT_FLOW,
3868 		DECLARE_UVERBS_WRITE_EX(
3869 			IB_USER_VERBS_EX_CMD_CREATE_FLOW,
3870 			ib_uverbs_ex_create_flow,
3871 			UAPI_DEF_WRITE_IO_EX(struct ib_uverbs_create_flow,
3872 					     flow_attr,
3873 					     struct ib_uverbs_create_flow_resp,
3874 					     flow_handle),
3875 			UAPI_DEF_METHOD_NEEDS_FN(create_flow)),
3876 		DECLARE_UVERBS_WRITE_EX(
3877 			IB_USER_VERBS_EX_CMD_DESTROY_FLOW,
3878 			ib_uverbs_ex_destroy_flow,
3879 			UAPI_DEF_WRITE_I(struct ib_uverbs_destroy_flow),
3880 			UAPI_DEF_METHOD_NEEDS_FN(destroy_flow))),
3881 
3882 	DECLARE_UVERBS_OBJECT(
3883 		UVERBS_OBJECT_MR,
3884 		DECLARE_UVERBS_WRITE(IB_USER_VERBS_CMD_DEREG_MR,
3885 				     ib_uverbs_dereg_mr,
3886 				     UAPI_DEF_WRITE_I(struct ib_uverbs_dereg_mr),
3887 				     UAPI_DEF_METHOD_NEEDS_FN(dereg_mr)),
3888 		DECLARE_UVERBS_WRITE(
3889 			IB_USER_VERBS_CMD_REG_MR,
3890 			ib_uverbs_reg_mr,
3891 			UAPI_DEF_WRITE_UDATA_IO(struct ib_uverbs_reg_mr,
3892 						struct ib_uverbs_reg_mr_resp),
3893 			UAPI_DEF_METHOD_NEEDS_FN(reg_user_mr)),
3894 		DECLARE_UVERBS_WRITE(
3895 			IB_USER_VERBS_CMD_REREG_MR,
3896 			ib_uverbs_rereg_mr,
3897 			UAPI_DEF_WRITE_UDATA_IO(struct ib_uverbs_rereg_mr,
3898 						struct ib_uverbs_rereg_mr_resp),
3899 			UAPI_DEF_METHOD_NEEDS_FN(rereg_user_mr))),
3900 
3901 	DECLARE_UVERBS_OBJECT(
3902 		UVERBS_OBJECT_MW,
3903 		DECLARE_UVERBS_WRITE(
3904 			IB_USER_VERBS_CMD_ALLOC_MW,
3905 			ib_uverbs_alloc_mw,
3906 			UAPI_DEF_WRITE_UDATA_IO(struct ib_uverbs_alloc_mw,
3907 						struct ib_uverbs_alloc_mw_resp),
3908 			UAPI_DEF_METHOD_NEEDS_FN(alloc_mw)),
3909 		DECLARE_UVERBS_WRITE(
3910 			IB_USER_VERBS_CMD_DEALLOC_MW,
3911 			ib_uverbs_dealloc_mw,
3912 			UAPI_DEF_WRITE_I(struct ib_uverbs_dealloc_mw),
3913 			UAPI_DEF_METHOD_NEEDS_FN(dealloc_mw))),
3914 
3915 	DECLARE_UVERBS_OBJECT(
3916 		UVERBS_OBJECT_PD,
3917 		DECLARE_UVERBS_WRITE(
3918 			IB_USER_VERBS_CMD_ALLOC_PD,
3919 			ib_uverbs_alloc_pd,
3920 			UAPI_DEF_WRITE_UDATA_IO(struct ib_uverbs_alloc_pd,
3921 						struct ib_uverbs_alloc_pd_resp),
3922 			UAPI_DEF_METHOD_NEEDS_FN(alloc_pd)),
3923 		DECLARE_UVERBS_WRITE(
3924 			IB_USER_VERBS_CMD_DEALLOC_PD,
3925 			ib_uverbs_dealloc_pd,
3926 			UAPI_DEF_WRITE_I(struct ib_uverbs_dealloc_pd),
3927 			UAPI_DEF_METHOD_NEEDS_FN(dealloc_pd))),
3928 
3929 	DECLARE_UVERBS_OBJECT(
3930 		UVERBS_OBJECT_QP,
3931 		DECLARE_UVERBS_WRITE(
3932 			IB_USER_VERBS_CMD_ATTACH_MCAST,
3933 			ib_uverbs_attach_mcast,
3934 			UAPI_DEF_WRITE_I(struct ib_uverbs_attach_mcast),
3935 			UAPI_DEF_METHOD_NEEDS_FN(attach_mcast),
3936 			UAPI_DEF_METHOD_NEEDS_FN(detach_mcast)),
3937 		DECLARE_UVERBS_WRITE(IB_USER_VERBS_CMD_CREATE_QP,
3938 				     ib_uverbs_create_qp,
3939 				     UAPI_DEF_WRITE_UDATA_IO(
3940 					     struct ib_uverbs_create_qp,
3941 					     struct ib_uverbs_create_qp_resp),
3942 				     UAPI_DEF_METHOD_NEEDS_FN(create_qp)),
3943 		DECLARE_UVERBS_WRITE(
3944 			IB_USER_VERBS_CMD_DESTROY_QP,
3945 			ib_uverbs_destroy_qp,
3946 			UAPI_DEF_WRITE_IO(struct ib_uverbs_destroy_qp,
3947 					  struct ib_uverbs_destroy_qp_resp),
3948 			UAPI_DEF_METHOD_NEEDS_FN(destroy_qp)),
3949 		DECLARE_UVERBS_WRITE(
3950 			IB_USER_VERBS_CMD_DETACH_MCAST,
3951 			ib_uverbs_detach_mcast,
3952 			UAPI_DEF_WRITE_I(struct ib_uverbs_detach_mcast),
3953 			UAPI_DEF_METHOD_NEEDS_FN(detach_mcast)),
3954 		DECLARE_UVERBS_WRITE(
3955 			IB_USER_VERBS_CMD_MODIFY_QP,
3956 			ib_uverbs_modify_qp,
3957 			UAPI_DEF_WRITE_I(struct ib_uverbs_modify_qp),
3958 			UAPI_DEF_METHOD_NEEDS_FN(modify_qp)),
3959 		DECLARE_UVERBS_WRITE(
3960 			IB_USER_VERBS_CMD_POST_RECV,
3961 			ib_uverbs_post_recv,
3962 			UAPI_DEF_WRITE_IO(struct ib_uverbs_post_recv,
3963 					  struct ib_uverbs_post_recv_resp),
3964 			UAPI_DEF_METHOD_NEEDS_FN(post_recv)),
3965 		DECLARE_UVERBS_WRITE(
3966 			IB_USER_VERBS_CMD_POST_SEND,
3967 			ib_uverbs_post_send,
3968 			UAPI_DEF_WRITE_IO(struct ib_uverbs_post_send,
3969 					  struct ib_uverbs_post_send_resp),
3970 			UAPI_DEF_METHOD_NEEDS_FN(post_send)),
3971 		DECLARE_UVERBS_WRITE(
3972 			IB_USER_VERBS_CMD_QUERY_QP,
3973 			ib_uverbs_query_qp,
3974 			UAPI_DEF_WRITE_IO(struct ib_uverbs_query_qp,
3975 					  struct ib_uverbs_query_qp_resp),
3976 			UAPI_DEF_METHOD_NEEDS_FN(query_qp)),
3977 		DECLARE_UVERBS_WRITE_EX(
3978 			IB_USER_VERBS_EX_CMD_CREATE_QP,
3979 			ib_uverbs_ex_create_qp,
3980 			UAPI_DEF_WRITE_IO_EX(struct ib_uverbs_ex_create_qp,
3981 					     comp_mask,
3982 					     struct ib_uverbs_ex_create_qp_resp,
3983 					     response_length),
3984 			UAPI_DEF_METHOD_NEEDS_FN(create_qp)),
3985 		DECLARE_UVERBS_WRITE_EX(
3986 			IB_USER_VERBS_EX_CMD_MODIFY_QP,
3987 			ib_uverbs_ex_modify_qp,
3988 			UAPI_DEF_WRITE_IO_EX(struct ib_uverbs_ex_modify_qp,
3989 					     base,
3990 					     struct ib_uverbs_ex_modify_qp_resp,
3991 					     response_length),
3992 			UAPI_DEF_METHOD_NEEDS_FN(modify_qp))),
3993 
3994 	DECLARE_UVERBS_OBJECT(
3995 		UVERBS_OBJECT_RWQ_IND_TBL,
3996 		DECLARE_UVERBS_WRITE_EX(
3997 			IB_USER_VERBS_EX_CMD_CREATE_RWQ_IND_TBL,
3998 			ib_uverbs_ex_create_rwq_ind_table,
3999 			UAPI_DEF_WRITE_IO_EX(
4000 				struct ib_uverbs_ex_create_rwq_ind_table,
4001 				log_ind_tbl_size,
4002 				struct ib_uverbs_ex_create_rwq_ind_table_resp,
4003 				ind_tbl_num),
4004 			UAPI_DEF_METHOD_NEEDS_FN(create_rwq_ind_table)),
4005 		DECLARE_UVERBS_WRITE_EX(
4006 			IB_USER_VERBS_EX_CMD_DESTROY_RWQ_IND_TBL,
4007 			ib_uverbs_ex_destroy_rwq_ind_table,
4008 			UAPI_DEF_WRITE_I(
4009 				struct ib_uverbs_ex_destroy_rwq_ind_table),
4010 			UAPI_DEF_METHOD_NEEDS_FN(destroy_rwq_ind_table))),
4011 
4012 	DECLARE_UVERBS_OBJECT(
4013 		UVERBS_OBJECT_WQ,
4014 		DECLARE_UVERBS_WRITE_EX(
4015 			IB_USER_VERBS_EX_CMD_CREATE_WQ,
4016 			ib_uverbs_ex_create_wq,
4017 			UAPI_DEF_WRITE_IO_EX(struct ib_uverbs_ex_create_wq,
4018 					     max_sge,
4019 					     struct ib_uverbs_ex_create_wq_resp,
4020 					     wqn),
4021 			UAPI_DEF_METHOD_NEEDS_FN(create_wq)),
4022 		DECLARE_UVERBS_WRITE_EX(
4023 			IB_USER_VERBS_EX_CMD_DESTROY_WQ,
4024 			ib_uverbs_ex_destroy_wq,
4025 			UAPI_DEF_WRITE_IO_EX(struct ib_uverbs_ex_destroy_wq,
4026 					     wq_handle,
4027 					     struct ib_uverbs_ex_destroy_wq_resp,
4028 					     reserved),
4029 			UAPI_DEF_METHOD_NEEDS_FN(destroy_wq)),
4030 		DECLARE_UVERBS_WRITE_EX(
4031 			IB_USER_VERBS_EX_CMD_MODIFY_WQ,
4032 			ib_uverbs_ex_modify_wq,
4033 			UAPI_DEF_WRITE_I_EX(struct ib_uverbs_ex_modify_wq,
4034 					    curr_wq_state),
4035 			UAPI_DEF_METHOD_NEEDS_FN(modify_wq))),
4036 
4037 	DECLARE_UVERBS_OBJECT(
4038 		UVERBS_OBJECT_SRQ,
4039 		DECLARE_UVERBS_WRITE(IB_USER_VERBS_CMD_CREATE_SRQ,
4040 				     ib_uverbs_create_srq,
4041 				     UAPI_DEF_WRITE_UDATA_IO(
4042 					     struct ib_uverbs_create_srq,
4043 					     struct ib_uverbs_create_srq_resp),
4044 				     UAPI_DEF_METHOD_NEEDS_FN(create_srq)),
4045 		DECLARE_UVERBS_WRITE(IB_USER_VERBS_CMD_CREATE_XSRQ,
4046 				     ib_uverbs_create_xsrq,
4047 				     UAPI_DEF_WRITE_UDATA_IO(
4048 					     struct ib_uverbs_create_xsrq,
4049 					     struct ib_uverbs_create_srq_resp),
4050 				     UAPI_DEF_METHOD_NEEDS_FN(create_srq)),
4051 		DECLARE_UVERBS_WRITE(
4052 			IB_USER_VERBS_CMD_DESTROY_SRQ,
4053 			ib_uverbs_destroy_srq,
4054 			UAPI_DEF_WRITE_IO(struct ib_uverbs_destroy_srq,
4055 					  struct ib_uverbs_destroy_srq_resp),
4056 			UAPI_DEF_METHOD_NEEDS_FN(destroy_srq)),
4057 		DECLARE_UVERBS_WRITE(
4058 			IB_USER_VERBS_CMD_MODIFY_SRQ,
4059 			ib_uverbs_modify_srq,
4060 			UAPI_DEF_WRITE_UDATA_I(struct ib_uverbs_modify_srq),
4061 			UAPI_DEF_METHOD_NEEDS_FN(modify_srq)),
4062 		DECLARE_UVERBS_WRITE(
4063 			IB_USER_VERBS_CMD_POST_SRQ_RECV,
4064 			ib_uverbs_post_srq_recv,
4065 			UAPI_DEF_WRITE_IO(struct ib_uverbs_post_srq_recv,
4066 					  struct ib_uverbs_post_srq_recv_resp),
4067 			UAPI_DEF_METHOD_NEEDS_FN(post_srq_recv)),
4068 		DECLARE_UVERBS_WRITE(
4069 			IB_USER_VERBS_CMD_QUERY_SRQ,
4070 			ib_uverbs_query_srq,
4071 			UAPI_DEF_WRITE_IO(struct ib_uverbs_query_srq,
4072 					  struct ib_uverbs_query_srq_resp),
4073 			UAPI_DEF_METHOD_NEEDS_FN(query_srq))),
4074 
4075 	DECLARE_UVERBS_OBJECT(
4076 		UVERBS_OBJECT_XRCD,
4077 		DECLARE_UVERBS_WRITE(
4078 			IB_USER_VERBS_CMD_CLOSE_XRCD,
4079 			ib_uverbs_close_xrcd,
4080 			UAPI_DEF_WRITE_I(struct ib_uverbs_close_xrcd),
4081 			UAPI_DEF_METHOD_NEEDS_FN(dealloc_xrcd)),
4082 		DECLARE_UVERBS_WRITE(IB_USER_VERBS_CMD_OPEN_QP,
4083 				     ib_uverbs_open_qp,
4084 				     UAPI_DEF_WRITE_UDATA_IO(
4085 					     struct ib_uverbs_open_qp,
4086 					     struct ib_uverbs_create_qp_resp)),
4087 		DECLARE_UVERBS_WRITE(IB_USER_VERBS_CMD_OPEN_XRCD,
4088 				     ib_uverbs_open_xrcd,
4089 				     UAPI_DEF_WRITE_UDATA_IO(
4090 					     struct ib_uverbs_open_xrcd,
4091 					     struct ib_uverbs_open_xrcd_resp),
4092 				     UAPI_DEF_METHOD_NEEDS_FN(alloc_xrcd))),
4093 
4094 	{},
4095 };
4096