xref: /linux/drivers/infiniband/core/uverbs_cmd.c (revision 87c2ce3b9305b9b723faeedf6e32ef703ec9b33a)
1 /*
2  * Copyright (c) 2005 Topspin Communications.  All rights reserved.
3  * Copyright (c) 2005 Cisco Systems.  All rights reserved.
4  * Copyright (c) 2005 PathScale, Inc.  All rights reserved.
5  *
6  * This software is available to you under a choice of one of two
7  * licenses.  You may choose to be licensed under the terms of the GNU
8  * General Public License (GPL) Version 2, available from the file
9  * COPYING in the main directory of this source tree, or the
10  * OpenIB.org BSD license below:
11  *
12  *     Redistribution and use in source and binary forms, with or
13  *     without modification, are permitted provided that the following
14  *     conditions are met:
15  *
16  *      - Redistributions of source code must retain the above
17  *        copyright notice, this list of conditions and the following
18  *        disclaimer.
19  *
20  *      - Redistributions in binary form must reproduce the above
21  *        copyright notice, this list of conditions and the following
22  *        disclaimer in the documentation and/or other materials
23  *        provided with the distribution.
24  *
25  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
26  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
27  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
28  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
29  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
30  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
31  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
32  * SOFTWARE.
33  *
34  * $Id: uverbs_cmd.c 2708 2005-06-24 17:27:21Z roland $
35  */
36 
37 #include <linux/file.h>
38 #include <linux/fs.h>
39 
40 #include <asm/uaccess.h>
41 
42 #include "uverbs.h"
43 
44 #define INIT_UDATA(udata, ibuf, obuf, ilen, olen)			\
45 	do {								\
46 		(udata)->inbuf  = (void __user *) (ibuf);		\
47 		(udata)->outbuf = (void __user *) (obuf);		\
48 		(udata)->inlen  = (ilen);				\
49 		(udata)->outlen = (olen);				\
50 	} while (0)
51 
52 ssize_t ib_uverbs_get_context(struct ib_uverbs_file *file,
53 			      const char __user *buf,
54 			      int in_len, int out_len)
55 {
56 	struct ib_uverbs_get_context      cmd;
57 	struct ib_uverbs_get_context_resp resp;
58 	struct ib_udata                   udata;
59 	struct ib_device                 *ibdev = file->device->ib_dev;
60 	struct ib_ucontext		 *ucontext;
61 	struct file			 *filp;
62 	int ret;
63 
64 	if (out_len < sizeof resp)
65 		return -ENOSPC;
66 
67 	if (copy_from_user(&cmd, buf, sizeof cmd))
68 		return -EFAULT;
69 
70 	down(&file->mutex);
71 
72 	if (file->ucontext) {
73 		ret = -EINVAL;
74 		goto err;
75 	}
76 
77 	INIT_UDATA(&udata, buf + sizeof cmd,
78 		   (unsigned long) cmd.response + sizeof resp,
79 		   in_len - sizeof cmd, out_len - sizeof resp);
80 
81 	ucontext = ibdev->alloc_ucontext(ibdev, &udata);
82 	if (IS_ERR(ucontext))
83 		return PTR_ERR(file->ucontext);
84 
85 	ucontext->device = ibdev;
86 	INIT_LIST_HEAD(&ucontext->pd_list);
87 	INIT_LIST_HEAD(&ucontext->mr_list);
88 	INIT_LIST_HEAD(&ucontext->mw_list);
89 	INIT_LIST_HEAD(&ucontext->cq_list);
90 	INIT_LIST_HEAD(&ucontext->qp_list);
91 	INIT_LIST_HEAD(&ucontext->srq_list);
92 	INIT_LIST_HEAD(&ucontext->ah_list);
93 
94 	resp.num_comp_vectors = file->device->num_comp_vectors;
95 
96 	filp = ib_uverbs_alloc_event_file(file, 1, &resp.async_fd);
97 	if (IS_ERR(filp)) {
98 		ret = PTR_ERR(filp);
99 		goto err_free;
100 	}
101 
102 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
103 			 &resp, sizeof resp)) {
104 		ret = -EFAULT;
105 		goto err_file;
106 	}
107 
108 	file->async_file = filp->private_data;
109 
110 	INIT_IB_EVENT_HANDLER(&file->event_handler, file->device->ib_dev,
111 			      ib_uverbs_event_handler);
112 	ret = ib_register_event_handler(&file->event_handler);
113 	if (ret)
114 		goto err_file;
115 
116 	kref_get(&file->async_file->ref);
117 	kref_get(&file->ref);
118 	file->ucontext = ucontext;
119 
120 	fd_install(resp.async_fd, filp);
121 
122 	up(&file->mutex);
123 
124 	return in_len;
125 
126 err_file:
127 	put_unused_fd(resp.async_fd);
128 	fput(filp);
129 
130 err_free:
131 	ibdev->dealloc_ucontext(ucontext);
132 
133 err:
134 	up(&file->mutex);
135 	return ret;
136 }
137 
138 ssize_t ib_uverbs_query_device(struct ib_uverbs_file *file,
139 			       const char __user *buf,
140 			       int in_len, int out_len)
141 {
142 	struct ib_uverbs_query_device      cmd;
143 	struct ib_uverbs_query_device_resp resp;
144 	struct ib_device_attr              attr;
145 	int                                ret;
146 
147 	if (out_len < sizeof resp)
148 		return -ENOSPC;
149 
150 	if (copy_from_user(&cmd, buf, sizeof cmd))
151 		return -EFAULT;
152 
153 	ret = ib_query_device(file->device->ib_dev, &attr);
154 	if (ret)
155 		return ret;
156 
157 	memset(&resp, 0, sizeof resp);
158 
159 	resp.fw_ver 		       = attr.fw_ver;
160 	resp.node_guid 		       = attr.node_guid;
161 	resp.sys_image_guid 	       = attr.sys_image_guid;
162 	resp.max_mr_size 	       = attr.max_mr_size;
163 	resp.page_size_cap 	       = attr.page_size_cap;
164 	resp.vendor_id 		       = attr.vendor_id;
165 	resp.vendor_part_id 	       = attr.vendor_part_id;
166 	resp.hw_ver 		       = attr.hw_ver;
167 	resp.max_qp 		       = attr.max_qp;
168 	resp.max_qp_wr 		       = attr.max_qp_wr;
169 	resp.device_cap_flags 	       = attr.device_cap_flags;
170 	resp.max_sge 		       = attr.max_sge;
171 	resp.max_sge_rd 	       = attr.max_sge_rd;
172 	resp.max_cq 		       = attr.max_cq;
173 	resp.max_cqe 		       = attr.max_cqe;
174 	resp.max_mr 		       = attr.max_mr;
175 	resp.max_pd 		       = attr.max_pd;
176 	resp.max_qp_rd_atom 	       = attr.max_qp_rd_atom;
177 	resp.max_ee_rd_atom 	       = attr.max_ee_rd_atom;
178 	resp.max_res_rd_atom 	       = attr.max_res_rd_atom;
179 	resp.max_qp_init_rd_atom       = attr.max_qp_init_rd_atom;
180 	resp.max_ee_init_rd_atom       = attr.max_ee_init_rd_atom;
181 	resp.atomic_cap 	       = attr.atomic_cap;
182 	resp.max_ee 		       = attr.max_ee;
183 	resp.max_rdd 		       = attr.max_rdd;
184 	resp.max_mw 		       = attr.max_mw;
185 	resp.max_raw_ipv6_qp 	       = attr.max_raw_ipv6_qp;
186 	resp.max_raw_ethy_qp 	       = attr.max_raw_ethy_qp;
187 	resp.max_mcast_grp 	       = attr.max_mcast_grp;
188 	resp.max_mcast_qp_attach       = attr.max_mcast_qp_attach;
189 	resp.max_total_mcast_qp_attach = attr.max_total_mcast_qp_attach;
190 	resp.max_ah 		       = attr.max_ah;
191 	resp.max_fmr 		       = attr.max_fmr;
192 	resp.max_map_per_fmr 	       = attr.max_map_per_fmr;
193 	resp.max_srq 		       = attr.max_srq;
194 	resp.max_srq_wr 	       = attr.max_srq_wr;
195 	resp.max_srq_sge 	       = attr.max_srq_sge;
196 	resp.max_pkeys 		       = attr.max_pkeys;
197 	resp.local_ca_ack_delay        = attr.local_ca_ack_delay;
198 	resp.phys_port_cnt	       = file->device->ib_dev->phys_port_cnt;
199 
200 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
201 			 &resp, sizeof resp))
202 		return -EFAULT;
203 
204 	return in_len;
205 }
206 
207 ssize_t ib_uverbs_query_port(struct ib_uverbs_file *file,
208 			     const char __user *buf,
209 			     int in_len, int out_len)
210 {
211 	struct ib_uverbs_query_port      cmd;
212 	struct ib_uverbs_query_port_resp resp;
213 	struct ib_port_attr              attr;
214 	int                              ret;
215 
216 	if (out_len < sizeof resp)
217 		return -ENOSPC;
218 
219 	if (copy_from_user(&cmd, buf, sizeof cmd))
220 		return -EFAULT;
221 
222 	ret = ib_query_port(file->device->ib_dev, cmd.port_num, &attr);
223 	if (ret)
224 		return ret;
225 
226 	memset(&resp, 0, sizeof resp);
227 
228 	resp.state 	     = attr.state;
229 	resp.max_mtu 	     = attr.max_mtu;
230 	resp.active_mtu      = attr.active_mtu;
231 	resp.gid_tbl_len     = attr.gid_tbl_len;
232 	resp.port_cap_flags  = attr.port_cap_flags;
233 	resp.max_msg_sz      = attr.max_msg_sz;
234 	resp.bad_pkey_cntr   = attr.bad_pkey_cntr;
235 	resp.qkey_viol_cntr  = attr.qkey_viol_cntr;
236 	resp.pkey_tbl_len    = attr.pkey_tbl_len;
237 	resp.lid 	     = attr.lid;
238 	resp.sm_lid 	     = attr.sm_lid;
239 	resp.lmc 	     = attr.lmc;
240 	resp.max_vl_num      = attr.max_vl_num;
241 	resp.sm_sl 	     = attr.sm_sl;
242 	resp.subnet_timeout  = attr.subnet_timeout;
243 	resp.init_type_reply = attr.init_type_reply;
244 	resp.active_width    = attr.active_width;
245 	resp.active_speed    = attr.active_speed;
246 	resp.phys_state      = attr.phys_state;
247 
248 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
249 			 &resp, sizeof resp))
250 		return -EFAULT;
251 
252 	return in_len;
253 }
254 
255 ssize_t ib_uverbs_alloc_pd(struct ib_uverbs_file *file,
256 			   const char __user *buf,
257 			   int in_len, int out_len)
258 {
259 	struct ib_uverbs_alloc_pd      cmd;
260 	struct ib_uverbs_alloc_pd_resp resp;
261 	struct ib_udata                udata;
262 	struct ib_uobject             *uobj;
263 	struct ib_pd                  *pd;
264 	int                            ret;
265 
266 	if (out_len < sizeof resp)
267 		return -ENOSPC;
268 
269 	if (copy_from_user(&cmd, buf, sizeof cmd))
270 		return -EFAULT;
271 
272 	INIT_UDATA(&udata, buf + sizeof cmd,
273 		   (unsigned long) cmd.response + sizeof resp,
274 		   in_len - sizeof cmd, out_len - sizeof resp);
275 
276 	uobj = kmalloc(sizeof *uobj, GFP_KERNEL);
277 	if (!uobj)
278 		return -ENOMEM;
279 
280 	uobj->context = file->ucontext;
281 
282 	pd = file->device->ib_dev->alloc_pd(file->device->ib_dev,
283 					    file->ucontext, &udata);
284 	if (IS_ERR(pd)) {
285 		ret = PTR_ERR(pd);
286 		goto err;
287 	}
288 
289 	pd->device  = file->device->ib_dev;
290 	pd->uobject = uobj;
291 	atomic_set(&pd->usecnt, 0);
292 
293 	down(&ib_uverbs_idr_mutex);
294 
295 retry:
296 	if (!idr_pre_get(&ib_uverbs_pd_idr, GFP_KERNEL)) {
297 		ret = -ENOMEM;
298 		goto err_up;
299 	}
300 
301 	ret = idr_get_new(&ib_uverbs_pd_idr, pd, &uobj->id);
302 
303 	if (ret == -EAGAIN)
304 		goto retry;
305 	if (ret)
306 		goto err_up;
307 
308 	memset(&resp, 0, sizeof resp);
309 	resp.pd_handle = uobj->id;
310 
311 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
312 			 &resp, sizeof resp)) {
313 		ret = -EFAULT;
314 		goto err_idr;
315 	}
316 
317 	down(&file->mutex);
318 	list_add_tail(&uobj->list, &file->ucontext->pd_list);
319 	up(&file->mutex);
320 
321 	up(&ib_uverbs_idr_mutex);
322 
323 	return in_len;
324 
325 err_idr:
326 	idr_remove(&ib_uverbs_pd_idr, uobj->id);
327 
328 err_up:
329 	up(&ib_uverbs_idr_mutex);
330 	ib_dealloc_pd(pd);
331 
332 err:
333 	kfree(uobj);
334 	return ret;
335 }
336 
337 ssize_t ib_uverbs_dealloc_pd(struct ib_uverbs_file *file,
338 			     const char __user *buf,
339 			     int in_len, int out_len)
340 {
341 	struct ib_uverbs_dealloc_pd cmd;
342 	struct ib_pd               *pd;
343 	struct ib_uobject          *uobj;
344 	int                         ret = -EINVAL;
345 
346 	if (copy_from_user(&cmd, buf, sizeof cmd))
347 		return -EFAULT;
348 
349 	down(&ib_uverbs_idr_mutex);
350 
351 	pd = idr_find(&ib_uverbs_pd_idr, cmd.pd_handle);
352 	if (!pd || pd->uobject->context != file->ucontext)
353 		goto out;
354 
355 	uobj = pd->uobject;
356 
357 	ret = ib_dealloc_pd(pd);
358 	if (ret)
359 		goto out;
360 
361 	idr_remove(&ib_uverbs_pd_idr, cmd.pd_handle);
362 
363 	down(&file->mutex);
364 	list_del(&uobj->list);
365 	up(&file->mutex);
366 
367 	kfree(uobj);
368 
369 out:
370 	up(&ib_uverbs_idr_mutex);
371 
372 	return ret ? ret : in_len;
373 }
374 
375 ssize_t ib_uverbs_reg_mr(struct ib_uverbs_file *file,
376 			 const char __user *buf, int in_len,
377 			 int out_len)
378 {
379 	struct ib_uverbs_reg_mr      cmd;
380 	struct ib_uverbs_reg_mr_resp resp;
381 	struct ib_udata              udata;
382 	struct ib_umem_object       *obj;
383 	struct ib_pd                *pd;
384 	struct ib_mr                *mr;
385 	int                          ret;
386 
387 	if (out_len < sizeof resp)
388 		return -ENOSPC;
389 
390 	if (copy_from_user(&cmd, buf, sizeof cmd))
391 		return -EFAULT;
392 
393 	INIT_UDATA(&udata, buf + sizeof cmd,
394 		   (unsigned long) cmd.response + sizeof resp,
395 		   in_len - sizeof cmd, out_len - sizeof resp);
396 
397 	if ((cmd.start & ~PAGE_MASK) != (cmd.hca_va & ~PAGE_MASK))
398 		return -EINVAL;
399 
400 	/*
401 	 * Local write permission is required if remote write or
402 	 * remote atomic permission is also requested.
403 	 */
404 	if (cmd.access_flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
405 	    !(cmd.access_flags & IB_ACCESS_LOCAL_WRITE))
406 		return -EINVAL;
407 
408 	obj = kmalloc(sizeof *obj, GFP_KERNEL);
409 	if (!obj)
410 		return -ENOMEM;
411 
412 	obj->uobject.context = file->ucontext;
413 
414 	/*
415 	 * We ask for writable memory if any access flags other than
416 	 * "remote read" are set.  "Local write" and "remote write"
417 	 * obviously require write access.  "Remote atomic" can do
418 	 * things like fetch and add, which will modify memory, and
419 	 * "MW bind" can change permissions by binding a window.
420 	 */
421 	ret = ib_umem_get(file->device->ib_dev, &obj->umem,
422 			  (void *) (unsigned long) cmd.start, cmd.length,
423 			  !!(cmd.access_flags & ~IB_ACCESS_REMOTE_READ));
424 	if (ret)
425 		goto err_free;
426 
427 	obj->umem.virt_base = cmd.hca_va;
428 
429 	down(&ib_uverbs_idr_mutex);
430 
431 	pd = idr_find(&ib_uverbs_pd_idr, cmd.pd_handle);
432 	if (!pd || pd->uobject->context != file->ucontext) {
433 		ret = -EINVAL;
434 		goto err_up;
435 	}
436 
437 	if (!pd->device->reg_user_mr) {
438 		ret = -ENOSYS;
439 		goto err_up;
440 	}
441 
442 	mr = pd->device->reg_user_mr(pd, &obj->umem, cmd.access_flags, &udata);
443 	if (IS_ERR(mr)) {
444 		ret = PTR_ERR(mr);
445 		goto err_up;
446 	}
447 
448 	mr->device  = pd->device;
449 	mr->pd      = pd;
450 	mr->uobject = &obj->uobject;
451 	atomic_inc(&pd->usecnt);
452 	atomic_set(&mr->usecnt, 0);
453 
454 	memset(&resp, 0, sizeof resp);
455 	resp.lkey = mr->lkey;
456 	resp.rkey = mr->rkey;
457 
458 retry:
459 	if (!idr_pre_get(&ib_uverbs_mr_idr, GFP_KERNEL)) {
460 		ret = -ENOMEM;
461 		goto err_unreg;
462 	}
463 
464 	ret = idr_get_new(&ib_uverbs_mr_idr, mr, &obj->uobject.id);
465 
466 	if (ret == -EAGAIN)
467 		goto retry;
468 	if (ret)
469 		goto err_unreg;
470 
471 	resp.mr_handle = obj->uobject.id;
472 
473 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
474 			 &resp, sizeof resp)) {
475 		ret = -EFAULT;
476 		goto err_idr;
477 	}
478 
479 	down(&file->mutex);
480 	list_add_tail(&obj->uobject.list, &file->ucontext->mr_list);
481 	up(&file->mutex);
482 
483 	up(&ib_uverbs_idr_mutex);
484 
485 	return in_len;
486 
487 err_idr:
488 	idr_remove(&ib_uverbs_mr_idr, obj->uobject.id);
489 
490 err_unreg:
491 	ib_dereg_mr(mr);
492 	atomic_dec(&pd->usecnt);
493 
494 err_up:
495 	up(&ib_uverbs_idr_mutex);
496 
497 	ib_umem_release(file->device->ib_dev, &obj->umem);
498 
499 err_free:
500 	kfree(obj);
501 	return ret;
502 }
503 
504 ssize_t ib_uverbs_dereg_mr(struct ib_uverbs_file *file,
505 			   const char __user *buf, int in_len,
506 			   int out_len)
507 {
508 	struct ib_uverbs_dereg_mr cmd;
509 	struct ib_mr             *mr;
510 	struct ib_umem_object    *memobj;
511 	int                       ret = -EINVAL;
512 
513 	if (copy_from_user(&cmd, buf, sizeof cmd))
514 		return -EFAULT;
515 
516 	down(&ib_uverbs_idr_mutex);
517 
518 	mr = idr_find(&ib_uverbs_mr_idr, cmd.mr_handle);
519 	if (!mr || mr->uobject->context != file->ucontext)
520 		goto out;
521 
522 	memobj = container_of(mr->uobject, struct ib_umem_object, uobject);
523 
524 	ret = ib_dereg_mr(mr);
525 	if (ret)
526 		goto out;
527 
528 	idr_remove(&ib_uverbs_mr_idr, cmd.mr_handle);
529 
530 	down(&file->mutex);
531 	list_del(&memobj->uobject.list);
532 	up(&file->mutex);
533 
534 	ib_umem_release(file->device->ib_dev, &memobj->umem);
535 	kfree(memobj);
536 
537 out:
538 	up(&ib_uverbs_idr_mutex);
539 
540 	return ret ? ret : in_len;
541 }
542 
543 ssize_t ib_uverbs_create_comp_channel(struct ib_uverbs_file *file,
544 				      const char __user *buf, int in_len,
545 				      int out_len)
546 {
547 	struct ib_uverbs_create_comp_channel	   cmd;
548 	struct ib_uverbs_create_comp_channel_resp  resp;
549 	struct file				  *filp;
550 
551 	if (out_len < sizeof resp)
552 		return -ENOSPC;
553 
554 	if (copy_from_user(&cmd, buf, sizeof cmd))
555 		return -EFAULT;
556 
557 	filp = ib_uverbs_alloc_event_file(file, 0, &resp.fd);
558 	if (IS_ERR(filp))
559 		return PTR_ERR(filp);
560 
561 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
562 			 &resp, sizeof resp)) {
563 		put_unused_fd(resp.fd);
564 		fput(filp);
565 		return -EFAULT;
566 	}
567 
568 	fd_install(resp.fd, filp);
569 	return in_len;
570 }
571 
572 ssize_t ib_uverbs_create_cq(struct ib_uverbs_file *file,
573 			    const char __user *buf, int in_len,
574 			    int out_len)
575 {
576 	struct ib_uverbs_create_cq      cmd;
577 	struct ib_uverbs_create_cq_resp resp;
578 	struct ib_udata                 udata;
579 	struct ib_ucq_object           *uobj;
580 	struct ib_uverbs_event_file    *ev_file = NULL;
581 	struct ib_cq                   *cq;
582 	int                             ret;
583 
584 	if (out_len < sizeof resp)
585 		return -ENOSPC;
586 
587 	if (copy_from_user(&cmd, buf, sizeof cmd))
588 		return -EFAULT;
589 
590 	INIT_UDATA(&udata, buf + sizeof cmd,
591 		   (unsigned long) cmd.response + sizeof resp,
592 		   in_len - sizeof cmd, out_len - sizeof resp);
593 
594 	if (cmd.comp_vector >= file->device->num_comp_vectors)
595 		return -EINVAL;
596 
597 	uobj = kmalloc(sizeof *uobj, GFP_KERNEL);
598 	if (!uobj)
599 		return -ENOMEM;
600 
601 	if (cmd.comp_channel >= 0) {
602 		ev_file = ib_uverbs_lookup_comp_file(cmd.comp_channel);
603 		if (!ev_file) {
604 			ret = -EINVAL;
605 			goto err;
606 		}
607 	}
608 
609 	uobj->uobject.user_handle   = cmd.user_handle;
610 	uobj->uobject.context       = file->ucontext;
611 	uobj->uverbs_file	    = file;
612 	uobj->comp_events_reported  = 0;
613 	uobj->async_events_reported = 0;
614 	INIT_LIST_HEAD(&uobj->comp_list);
615 	INIT_LIST_HEAD(&uobj->async_list);
616 
617 	cq = file->device->ib_dev->create_cq(file->device->ib_dev, cmd.cqe,
618 					     file->ucontext, &udata);
619 	if (IS_ERR(cq)) {
620 		ret = PTR_ERR(cq);
621 		goto err;
622 	}
623 
624 	cq->device        = file->device->ib_dev;
625 	cq->uobject       = &uobj->uobject;
626 	cq->comp_handler  = ib_uverbs_comp_handler;
627 	cq->event_handler = ib_uverbs_cq_event_handler;
628 	cq->cq_context    = ev_file;
629 	atomic_set(&cq->usecnt, 0);
630 
631 	down(&ib_uverbs_idr_mutex);
632 
633 retry:
634 	if (!idr_pre_get(&ib_uverbs_cq_idr, GFP_KERNEL)) {
635 		ret = -ENOMEM;
636 		goto err_up;
637 	}
638 
639 	ret = idr_get_new(&ib_uverbs_cq_idr, cq, &uobj->uobject.id);
640 
641 	if (ret == -EAGAIN)
642 		goto retry;
643 	if (ret)
644 		goto err_up;
645 
646 	memset(&resp, 0, sizeof resp);
647 	resp.cq_handle = uobj->uobject.id;
648 	resp.cqe       = cq->cqe;
649 
650 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
651 			 &resp, sizeof resp)) {
652 		ret = -EFAULT;
653 		goto err_idr;
654 	}
655 
656 	down(&file->mutex);
657 	list_add_tail(&uobj->uobject.list, &file->ucontext->cq_list);
658 	up(&file->mutex);
659 
660 	up(&ib_uverbs_idr_mutex);
661 
662 	return in_len;
663 
664 err_idr:
665 	idr_remove(&ib_uverbs_cq_idr, uobj->uobject.id);
666 
667 err_up:
668 	up(&ib_uverbs_idr_mutex);
669 	ib_destroy_cq(cq);
670 
671 err:
672 	if (ev_file)
673 		ib_uverbs_release_ucq(file, ev_file, uobj);
674 	kfree(uobj);
675 	return ret;
676 }
677 
678 ssize_t ib_uverbs_poll_cq(struct ib_uverbs_file *file,
679 			  const char __user *buf, int in_len,
680 			  int out_len)
681 {
682 	struct ib_uverbs_poll_cq       cmd;
683 	struct ib_uverbs_poll_cq_resp *resp;
684 	struct ib_cq                  *cq;
685 	struct ib_wc                  *wc;
686 	int                            ret = 0;
687 	int                            i;
688 	int                            rsize;
689 
690 	if (copy_from_user(&cmd, buf, sizeof cmd))
691 		return -EFAULT;
692 
693 	wc = kmalloc(cmd.ne * sizeof *wc, GFP_KERNEL);
694 	if (!wc)
695 		return -ENOMEM;
696 
697 	rsize = sizeof *resp + cmd.ne * sizeof(struct ib_uverbs_wc);
698 	resp = kmalloc(rsize, GFP_KERNEL);
699 	if (!resp) {
700 		ret = -ENOMEM;
701 		goto out_wc;
702 	}
703 
704 	down(&ib_uverbs_idr_mutex);
705 	cq = idr_find(&ib_uverbs_cq_idr, cmd.cq_handle);
706 	if (!cq || cq->uobject->context != file->ucontext) {
707 		ret = -EINVAL;
708 		goto out;
709 	}
710 
711 	resp->count = ib_poll_cq(cq, cmd.ne, wc);
712 
713 	for (i = 0; i < resp->count; i++) {
714 		resp->wc[i].wr_id 	   = wc[i].wr_id;
715 		resp->wc[i].status 	   = wc[i].status;
716 		resp->wc[i].opcode 	   = wc[i].opcode;
717 		resp->wc[i].vendor_err 	   = wc[i].vendor_err;
718 		resp->wc[i].byte_len 	   = wc[i].byte_len;
719 		resp->wc[i].imm_data 	   = (__u32 __force) wc[i].imm_data;
720 		resp->wc[i].qp_num 	   = wc[i].qp_num;
721 		resp->wc[i].src_qp 	   = wc[i].src_qp;
722 		resp->wc[i].wc_flags 	   = wc[i].wc_flags;
723 		resp->wc[i].pkey_index 	   = wc[i].pkey_index;
724 		resp->wc[i].slid 	   = wc[i].slid;
725 		resp->wc[i].sl 		   = wc[i].sl;
726 		resp->wc[i].dlid_path_bits = wc[i].dlid_path_bits;
727 		resp->wc[i].port_num 	   = wc[i].port_num;
728 	}
729 
730 	if (copy_to_user((void __user *) (unsigned long) cmd.response, resp, rsize))
731 		ret = -EFAULT;
732 
733 out:
734 	up(&ib_uverbs_idr_mutex);
735 	kfree(resp);
736 
737 out_wc:
738 	kfree(wc);
739 	return ret ? ret : in_len;
740 }
741 
742 ssize_t ib_uverbs_req_notify_cq(struct ib_uverbs_file *file,
743 				const char __user *buf, int in_len,
744 				int out_len)
745 {
746 	struct ib_uverbs_req_notify_cq cmd;
747 	struct ib_cq                  *cq;
748 	int                            ret = -EINVAL;
749 
750 	if (copy_from_user(&cmd, buf, sizeof cmd))
751 		return -EFAULT;
752 
753 	down(&ib_uverbs_idr_mutex);
754 	cq = idr_find(&ib_uverbs_cq_idr, cmd.cq_handle);
755 	if (cq && cq->uobject->context == file->ucontext) {
756 		ib_req_notify_cq(cq, cmd.solicited_only ?
757 					IB_CQ_SOLICITED : IB_CQ_NEXT_COMP);
758 		ret = in_len;
759 	}
760 	up(&ib_uverbs_idr_mutex);
761 
762 	return ret;
763 }
764 
765 ssize_t ib_uverbs_destroy_cq(struct ib_uverbs_file *file,
766 			     const char __user *buf, int in_len,
767 			     int out_len)
768 {
769 	struct ib_uverbs_destroy_cq      cmd;
770 	struct ib_uverbs_destroy_cq_resp resp;
771 	struct ib_cq               	*cq;
772 	struct ib_ucq_object        	*uobj;
773 	struct ib_uverbs_event_file	*ev_file;
774 	u64				 user_handle;
775 	int                        	 ret = -EINVAL;
776 
777 	if (copy_from_user(&cmd, buf, sizeof cmd))
778 		return -EFAULT;
779 
780 	memset(&resp, 0, sizeof resp);
781 
782 	down(&ib_uverbs_idr_mutex);
783 
784 	cq = idr_find(&ib_uverbs_cq_idr, cmd.cq_handle);
785 	if (!cq || cq->uobject->context != file->ucontext)
786 		goto out;
787 
788 	user_handle = cq->uobject->user_handle;
789 	uobj        = container_of(cq->uobject, struct ib_ucq_object, uobject);
790 	ev_file     = cq->cq_context;
791 
792 	ret = ib_destroy_cq(cq);
793 	if (ret)
794 		goto out;
795 
796 	idr_remove(&ib_uverbs_cq_idr, cmd.cq_handle);
797 
798 	down(&file->mutex);
799 	list_del(&uobj->uobject.list);
800 	up(&file->mutex);
801 
802 	ib_uverbs_release_ucq(file, ev_file, uobj);
803 
804 	resp.comp_events_reported  = uobj->comp_events_reported;
805 	resp.async_events_reported = uobj->async_events_reported;
806 
807 	kfree(uobj);
808 
809 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
810 			 &resp, sizeof resp))
811 		ret = -EFAULT;
812 
813 out:
814 	up(&ib_uverbs_idr_mutex);
815 
816 	return ret ? ret : in_len;
817 }
818 
819 ssize_t ib_uverbs_create_qp(struct ib_uverbs_file *file,
820 			    const char __user *buf, int in_len,
821 			    int out_len)
822 {
823 	struct ib_uverbs_create_qp      cmd;
824 	struct ib_uverbs_create_qp_resp resp;
825 	struct ib_udata                 udata;
826 	struct ib_uqp_object           *uobj;
827 	struct ib_pd                   *pd;
828 	struct ib_cq                   *scq, *rcq;
829 	struct ib_srq                  *srq;
830 	struct ib_qp                   *qp;
831 	struct ib_qp_init_attr          attr;
832 	int ret;
833 
834 	if (out_len < sizeof resp)
835 		return -ENOSPC;
836 
837 	if (copy_from_user(&cmd, buf, sizeof cmd))
838 		return -EFAULT;
839 
840 	INIT_UDATA(&udata, buf + sizeof cmd,
841 		   (unsigned long) cmd.response + sizeof resp,
842 		   in_len - sizeof cmd, out_len - sizeof resp);
843 
844 	uobj = kmalloc(sizeof *uobj, GFP_KERNEL);
845 	if (!uobj)
846 		return -ENOMEM;
847 
848 	down(&ib_uverbs_idr_mutex);
849 
850 	pd  = idr_find(&ib_uverbs_pd_idr, cmd.pd_handle);
851 	scq = idr_find(&ib_uverbs_cq_idr, cmd.send_cq_handle);
852 	rcq = idr_find(&ib_uverbs_cq_idr, cmd.recv_cq_handle);
853 	srq = cmd.is_srq ? idr_find(&ib_uverbs_srq_idr, cmd.srq_handle) : NULL;
854 
855 	if (!pd  || pd->uobject->context  != file->ucontext ||
856 	    !scq || scq->uobject->context != file->ucontext ||
857 	    !rcq || rcq->uobject->context != file->ucontext ||
858 	    (cmd.is_srq && (!srq || srq->uobject->context != file->ucontext))) {
859 		ret = -EINVAL;
860 		goto err_up;
861 	}
862 
863 	attr.event_handler = ib_uverbs_qp_event_handler;
864 	attr.qp_context    = file;
865 	attr.send_cq       = scq;
866 	attr.recv_cq       = rcq;
867 	attr.srq           = srq;
868 	attr.sq_sig_type   = cmd.sq_sig_all ? IB_SIGNAL_ALL_WR : IB_SIGNAL_REQ_WR;
869 	attr.qp_type       = cmd.qp_type;
870 
871 	attr.cap.max_send_wr     = cmd.max_send_wr;
872 	attr.cap.max_recv_wr     = cmd.max_recv_wr;
873 	attr.cap.max_send_sge    = cmd.max_send_sge;
874 	attr.cap.max_recv_sge    = cmd.max_recv_sge;
875 	attr.cap.max_inline_data = cmd.max_inline_data;
876 
877 	uobj->uevent.uobject.user_handle = cmd.user_handle;
878 	uobj->uevent.uobject.context     = file->ucontext;
879 	uobj->uevent.events_reported     = 0;
880 	INIT_LIST_HEAD(&uobj->uevent.event_list);
881 	INIT_LIST_HEAD(&uobj->mcast_list);
882 
883 	qp = pd->device->create_qp(pd, &attr, &udata);
884 	if (IS_ERR(qp)) {
885 		ret = PTR_ERR(qp);
886 		goto err_up;
887 	}
888 
889 	qp->device     	  = pd->device;
890 	qp->pd         	  = pd;
891 	qp->send_cq    	  = attr.send_cq;
892 	qp->recv_cq    	  = attr.recv_cq;
893 	qp->srq	       	  = attr.srq;
894 	qp->uobject       = &uobj->uevent.uobject;
895 	qp->event_handler = attr.event_handler;
896 	qp->qp_context    = attr.qp_context;
897 	qp->qp_type	  = attr.qp_type;
898 	atomic_inc(&pd->usecnt);
899 	atomic_inc(&attr.send_cq->usecnt);
900 	atomic_inc(&attr.recv_cq->usecnt);
901 	if (attr.srq)
902 		atomic_inc(&attr.srq->usecnt);
903 
904 	memset(&resp, 0, sizeof resp);
905 	resp.qpn = qp->qp_num;
906 
907 retry:
908 	if (!idr_pre_get(&ib_uverbs_qp_idr, GFP_KERNEL)) {
909 		ret = -ENOMEM;
910 		goto err_destroy;
911 	}
912 
913 	ret = idr_get_new(&ib_uverbs_qp_idr, qp, &uobj->uevent.uobject.id);
914 
915 	if (ret == -EAGAIN)
916 		goto retry;
917 	if (ret)
918 		goto err_destroy;
919 
920 	resp.qp_handle       = uobj->uevent.uobject.id;
921 	resp.max_recv_sge    = attr.cap.max_recv_sge;
922 	resp.max_send_sge    = attr.cap.max_send_sge;
923 	resp.max_recv_wr     = attr.cap.max_recv_wr;
924 	resp.max_send_wr     = attr.cap.max_send_wr;
925 	resp.max_inline_data = attr.cap.max_inline_data;
926 
927 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
928 			 &resp, sizeof resp)) {
929 		ret = -EFAULT;
930 		goto err_idr;
931 	}
932 
933 	down(&file->mutex);
934 	list_add_tail(&uobj->uevent.uobject.list, &file->ucontext->qp_list);
935 	up(&file->mutex);
936 
937 	up(&ib_uverbs_idr_mutex);
938 
939 	return in_len;
940 
941 err_idr:
942 	idr_remove(&ib_uverbs_qp_idr, uobj->uevent.uobject.id);
943 
944 err_destroy:
945 	ib_destroy_qp(qp);
946 	atomic_dec(&pd->usecnt);
947 	atomic_dec(&attr.send_cq->usecnt);
948 	atomic_dec(&attr.recv_cq->usecnt);
949 	if (attr.srq)
950 		atomic_dec(&attr.srq->usecnt);
951 
952 err_up:
953 	up(&ib_uverbs_idr_mutex);
954 
955 	kfree(uobj);
956 	return ret;
957 }
958 
959 ssize_t ib_uverbs_modify_qp(struct ib_uverbs_file *file,
960 			    const char __user *buf, int in_len,
961 			    int out_len)
962 {
963 	struct ib_uverbs_modify_qp cmd;
964 	struct ib_qp              *qp;
965 	struct ib_qp_attr         *attr;
966 	int                        ret;
967 
968 	if (copy_from_user(&cmd, buf, sizeof cmd))
969 		return -EFAULT;
970 
971 	attr = kmalloc(sizeof *attr, GFP_KERNEL);
972 	if (!attr)
973 		return -ENOMEM;
974 
975 	down(&ib_uverbs_idr_mutex);
976 
977 	qp = idr_find(&ib_uverbs_qp_idr, cmd.qp_handle);
978 	if (!qp || qp->uobject->context != file->ucontext) {
979 		ret = -EINVAL;
980 		goto out;
981 	}
982 
983 	attr->qp_state 		  = cmd.qp_state;
984 	attr->cur_qp_state 	  = cmd.cur_qp_state;
985 	attr->path_mtu 		  = cmd.path_mtu;
986 	attr->path_mig_state 	  = cmd.path_mig_state;
987 	attr->qkey 		  = cmd.qkey;
988 	attr->rq_psn 		  = cmd.rq_psn;
989 	attr->sq_psn 		  = cmd.sq_psn;
990 	attr->dest_qp_num 	  = cmd.dest_qp_num;
991 	attr->qp_access_flags 	  = cmd.qp_access_flags;
992 	attr->pkey_index 	  = cmd.pkey_index;
993 	attr->alt_pkey_index 	  = cmd.pkey_index;
994 	attr->en_sqd_async_notify = cmd.en_sqd_async_notify;
995 	attr->max_rd_atomic 	  = cmd.max_rd_atomic;
996 	attr->max_dest_rd_atomic  = cmd.max_dest_rd_atomic;
997 	attr->min_rnr_timer 	  = cmd.min_rnr_timer;
998 	attr->port_num 		  = cmd.port_num;
999 	attr->timeout 		  = cmd.timeout;
1000 	attr->retry_cnt 	  = cmd.retry_cnt;
1001 	attr->rnr_retry 	  = cmd.rnr_retry;
1002 	attr->alt_port_num 	  = cmd.alt_port_num;
1003 	attr->alt_timeout 	  = cmd.alt_timeout;
1004 
1005 	memcpy(attr->ah_attr.grh.dgid.raw, cmd.dest.dgid, 16);
1006 	attr->ah_attr.grh.flow_label        = cmd.dest.flow_label;
1007 	attr->ah_attr.grh.sgid_index        = cmd.dest.sgid_index;
1008 	attr->ah_attr.grh.hop_limit         = cmd.dest.hop_limit;
1009 	attr->ah_attr.grh.traffic_class     = cmd.dest.traffic_class;
1010 	attr->ah_attr.dlid 	    	    = cmd.dest.dlid;
1011 	attr->ah_attr.sl   	    	    = cmd.dest.sl;
1012 	attr->ah_attr.src_path_bits 	    = cmd.dest.src_path_bits;
1013 	attr->ah_attr.static_rate   	    = cmd.dest.static_rate;
1014 	attr->ah_attr.ah_flags 	    	    = cmd.dest.is_global ? IB_AH_GRH : 0;
1015 	attr->ah_attr.port_num 	    	    = cmd.dest.port_num;
1016 
1017 	memcpy(attr->alt_ah_attr.grh.dgid.raw, cmd.alt_dest.dgid, 16);
1018 	attr->alt_ah_attr.grh.flow_label    = cmd.alt_dest.flow_label;
1019 	attr->alt_ah_attr.grh.sgid_index    = cmd.alt_dest.sgid_index;
1020 	attr->alt_ah_attr.grh.hop_limit     = cmd.alt_dest.hop_limit;
1021 	attr->alt_ah_attr.grh.traffic_class = cmd.alt_dest.traffic_class;
1022 	attr->alt_ah_attr.dlid 	    	    = cmd.alt_dest.dlid;
1023 	attr->alt_ah_attr.sl   	    	    = cmd.alt_dest.sl;
1024 	attr->alt_ah_attr.src_path_bits     = cmd.alt_dest.src_path_bits;
1025 	attr->alt_ah_attr.static_rate       = cmd.alt_dest.static_rate;
1026 	attr->alt_ah_attr.ah_flags 	    = cmd.alt_dest.is_global ? IB_AH_GRH : 0;
1027 	attr->alt_ah_attr.port_num 	    = cmd.alt_dest.port_num;
1028 
1029 	ret = ib_modify_qp(qp, attr, cmd.attr_mask);
1030 	if (ret)
1031 		goto out;
1032 
1033 	ret = in_len;
1034 
1035 out:
1036 	up(&ib_uverbs_idr_mutex);
1037 	kfree(attr);
1038 
1039 	return ret;
1040 }
1041 
1042 ssize_t ib_uverbs_destroy_qp(struct ib_uverbs_file *file,
1043 			     const char __user *buf, int in_len,
1044 			     int out_len)
1045 {
1046 	struct ib_uverbs_destroy_qp      cmd;
1047 	struct ib_uverbs_destroy_qp_resp resp;
1048 	struct ib_qp               	*qp;
1049 	struct ib_uqp_object        	*uobj;
1050 	int                        	 ret = -EINVAL;
1051 
1052 	if (copy_from_user(&cmd, buf, sizeof cmd))
1053 		return -EFAULT;
1054 
1055 	memset(&resp, 0, sizeof resp);
1056 
1057 	down(&ib_uverbs_idr_mutex);
1058 
1059 	qp = idr_find(&ib_uverbs_qp_idr, cmd.qp_handle);
1060 	if (!qp || qp->uobject->context != file->ucontext)
1061 		goto out;
1062 
1063 	uobj = container_of(qp->uobject, struct ib_uqp_object, uevent.uobject);
1064 
1065 	if (!list_empty(&uobj->mcast_list)) {
1066 		ret = -EBUSY;
1067 		goto out;
1068 	}
1069 
1070 	ret = ib_destroy_qp(qp);
1071 	if (ret)
1072 		goto out;
1073 
1074 	idr_remove(&ib_uverbs_qp_idr, cmd.qp_handle);
1075 
1076 	down(&file->mutex);
1077 	list_del(&uobj->uevent.uobject.list);
1078 	up(&file->mutex);
1079 
1080 	ib_uverbs_release_uevent(file, &uobj->uevent);
1081 
1082 	resp.events_reported = uobj->uevent.events_reported;
1083 
1084 	kfree(uobj);
1085 
1086 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
1087 			 &resp, sizeof resp))
1088 		ret = -EFAULT;
1089 
1090 out:
1091 	up(&ib_uverbs_idr_mutex);
1092 
1093 	return ret ? ret : in_len;
1094 }
1095 
1096 ssize_t ib_uverbs_post_send(struct ib_uverbs_file *file,
1097                             const char __user *buf, int in_len,
1098                             int out_len)
1099 {
1100 	struct ib_uverbs_post_send      cmd;
1101 	struct ib_uverbs_post_send_resp resp;
1102 	struct ib_uverbs_send_wr       *user_wr;
1103 	struct ib_send_wr              *wr = NULL, *last, *next, *bad_wr;
1104 	struct ib_qp                   *qp;
1105 	int                             i, sg_ind;
1106 	ssize_t                         ret = -EINVAL;
1107 
1108 	if (copy_from_user(&cmd, buf, sizeof cmd))
1109 		return -EFAULT;
1110 
1111 	if (in_len < sizeof cmd + cmd.wqe_size * cmd.wr_count +
1112 	    cmd.sge_count * sizeof (struct ib_uverbs_sge))
1113 		return -EINVAL;
1114 
1115 	if (cmd.wqe_size < sizeof (struct ib_uverbs_send_wr))
1116 		return -EINVAL;
1117 
1118 	user_wr = kmalloc(cmd.wqe_size, GFP_KERNEL);
1119 	if (!user_wr)
1120 		return -ENOMEM;
1121 
1122 	down(&ib_uverbs_idr_mutex);
1123 
1124 	qp = idr_find(&ib_uverbs_qp_idr, cmd.qp_handle);
1125 	if (!qp || qp->uobject->context != file->ucontext)
1126 		goto out;
1127 
1128 	sg_ind = 0;
1129 	last = NULL;
1130 	for (i = 0; i < cmd.wr_count; ++i) {
1131 		if (copy_from_user(user_wr,
1132 				   buf + sizeof cmd + i * cmd.wqe_size,
1133 				   cmd.wqe_size)) {
1134 			ret = -EFAULT;
1135 			goto out;
1136 		}
1137 
1138 		if (user_wr->num_sge + sg_ind > cmd.sge_count) {
1139 			ret = -EINVAL;
1140 			goto out;
1141 		}
1142 
1143 		next = kmalloc(ALIGN(sizeof *next, sizeof (struct ib_sge)) +
1144 			       user_wr->num_sge * sizeof (struct ib_sge),
1145 			       GFP_KERNEL);
1146 		if (!next) {
1147 			ret = -ENOMEM;
1148 			goto out;
1149 		}
1150 
1151 		if (!last)
1152 			wr = next;
1153 		else
1154 			last->next = next;
1155 		last = next;
1156 
1157 		next->next       = NULL;
1158 		next->wr_id      = user_wr->wr_id;
1159 		next->num_sge    = user_wr->num_sge;
1160 		next->opcode     = user_wr->opcode;
1161 		next->send_flags = user_wr->send_flags;
1162 		next->imm_data   = (__be32 __force) user_wr->imm_data;
1163 
1164 		if (qp->qp_type == IB_QPT_UD) {
1165 			next->wr.ud.ah = idr_find(&ib_uverbs_ah_idr,
1166 						  user_wr->wr.ud.ah);
1167 			if (!next->wr.ud.ah) {
1168 				ret = -EINVAL;
1169 				goto out;
1170 			}
1171 			next->wr.ud.remote_qpn  = user_wr->wr.ud.remote_qpn;
1172 			next->wr.ud.remote_qkey = user_wr->wr.ud.remote_qkey;
1173 		} else {
1174 			switch (next->opcode) {
1175 			case IB_WR_RDMA_WRITE:
1176 			case IB_WR_RDMA_WRITE_WITH_IMM:
1177 			case IB_WR_RDMA_READ:
1178 				next->wr.rdma.remote_addr =
1179 					user_wr->wr.rdma.remote_addr;
1180 				next->wr.rdma.rkey        =
1181 					user_wr->wr.rdma.rkey;
1182 				break;
1183 			case IB_WR_ATOMIC_CMP_AND_SWP:
1184 			case IB_WR_ATOMIC_FETCH_AND_ADD:
1185 				next->wr.atomic.remote_addr =
1186 					user_wr->wr.atomic.remote_addr;
1187 				next->wr.atomic.compare_add =
1188 					user_wr->wr.atomic.compare_add;
1189 				next->wr.atomic.swap = user_wr->wr.atomic.swap;
1190 				next->wr.atomic.rkey = user_wr->wr.atomic.rkey;
1191 				break;
1192 			default:
1193 				break;
1194 			}
1195 		}
1196 
1197 		if (next->num_sge) {
1198 			next->sg_list = (void *) next +
1199 				ALIGN(sizeof *next, sizeof (struct ib_sge));
1200 			if (copy_from_user(next->sg_list,
1201 					   buf + sizeof cmd +
1202 					   cmd.wr_count * cmd.wqe_size +
1203 					   sg_ind * sizeof (struct ib_sge),
1204 					   next->num_sge * sizeof (struct ib_sge))) {
1205 				ret = -EFAULT;
1206 				goto out;
1207 			}
1208 			sg_ind += next->num_sge;
1209 		} else
1210 			next->sg_list = NULL;
1211 	}
1212 
1213 	resp.bad_wr = 0;
1214 	ret = qp->device->post_send(qp, wr, &bad_wr);
1215 	if (ret)
1216 		for (next = wr; next; next = next->next) {
1217 			++resp.bad_wr;
1218 			if (next == bad_wr)
1219 				break;
1220 		}
1221 
1222 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
1223 			 &resp, sizeof resp))
1224 		ret = -EFAULT;
1225 
1226 out:
1227 	up(&ib_uverbs_idr_mutex);
1228 
1229 	while (wr) {
1230 		next = wr->next;
1231 		kfree(wr);
1232 		wr = next;
1233 	}
1234 
1235 	kfree(user_wr);
1236 
1237 	return ret ? ret : in_len;
1238 }
1239 
1240 static struct ib_recv_wr *ib_uverbs_unmarshall_recv(const char __user *buf,
1241 						    int in_len,
1242 						    u32 wr_count,
1243 						    u32 sge_count,
1244 						    u32 wqe_size)
1245 {
1246 	struct ib_uverbs_recv_wr *user_wr;
1247 	struct ib_recv_wr        *wr = NULL, *last, *next;
1248 	int                       sg_ind;
1249 	int                       i;
1250 	int                       ret;
1251 
1252 	if (in_len < wqe_size * wr_count +
1253 	    sge_count * sizeof (struct ib_uverbs_sge))
1254 		return ERR_PTR(-EINVAL);
1255 
1256 	if (wqe_size < sizeof (struct ib_uverbs_recv_wr))
1257 		return ERR_PTR(-EINVAL);
1258 
1259 	user_wr = kmalloc(wqe_size, GFP_KERNEL);
1260 	if (!user_wr)
1261 		return ERR_PTR(-ENOMEM);
1262 
1263 	sg_ind = 0;
1264 	last = NULL;
1265 	for (i = 0; i < wr_count; ++i) {
1266 		if (copy_from_user(user_wr, buf + i * wqe_size,
1267 				   wqe_size)) {
1268 			ret = -EFAULT;
1269 			goto err;
1270 		}
1271 
1272 		if (user_wr->num_sge + sg_ind > sge_count) {
1273 			ret = -EINVAL;
1274 			goto err;
1275 		}
1276 
1277 		next = kmalloc(ALIGN(sizeof *next, sizeof (struct ib_sge)) +
1278 			       user_wr->num_sge * sizeof (struct ib_sge),
1279 			       GFP_KERNEL);
1280 		if (!next) {
1281 			ret = -ENOMEM;
1282 			goto err;
1283 		}
1284 
1285 		if (!last)
1286 			wr = next;
1287 		else
1288 			last->next = next;
1289 		last = next;
1290 
1291 		next->next       = NULL;
1292 		next->wr_id      = user_wr->wr_id;
1293 		next->num_sge    = user_wr->num_sge;
1294 
1295 		if (next->num_sge) {
1296 			next->sg_list = (void *) next +
1297 				ALIGN(sizeof *next, sizeof (struct ib_sge));
1298 			if (copy_from_user(next->sg_list,
1299 					   buf + wr_count * wqe_size +
1300 					   sg_ind * sizeof (struct ib_sge),
1301 					   next->num_sge * sizeof (struct ib_sge))) {
1302 				ret = -EFAULT;
1303 				goto err;
1304 			}
1305 			sg_ind += next->num_sge;
1306 		} else
1307 			next->sg_list = NULL;
1308 	}
1309 
1310 	kfree(user_wr);
1311 	return wr;
1312 
1313 err:
1314 	kfree(user_wr);
1315 
1316 	while (wr) {
1317 		next = wr->next;
1318 		kfree(wr);
1319 		wr = next;
1320 	}
1321 
1322 	return ERR_PTR(ret);
1323 }
1324 
1325 ssize_t ib_uverbs_post_recv(struct ib_uverbs_file *file,
1326                             const char __user *buf, int in_len,
1327                             int out_len)
1328 {
1329 	struct ib_uverbs_post_recv      cmd;
1330 	struct ib_uverbs_post_recv_resp resp;
1331 	struct ib_recv_wr              *wr, *next, *bad_wr;
1332 	struct ib_qp                   *qp;
1333 	ssize_t                         ret = -EINVAL;
1334 
1335 	if (copy_from_user(&cmd, buf, sizeof cmd))
1336 		return -EFAULT;
1337 
1338 	wr = ib_uverbs_unmarshall_recv(buf + sizeof cmd,
1339 				       in_len - sizeof cmd, cmd.wr_count,
1340 				       cmd.sge_count, cmd.wqe_size);
1341 	if (IS_ERR(wr))
1342 		return PTR_ERR(wr);
1343 
1344 	down(&ib_uverbs_idr_mutex);
1345 
1346 	qp = idr_find(&ib_uverbs_qp_idr, cmd.qp_handle);
1347 	if (!qp || qp->uobject->context != file->ucontext)
1348 		goto out;
1349 
1350 	resp.bad_wr = 0;
1351 	ret = qp->device->post_recv(qp, wr, &bad_wr);
1352 	if (ret)
1353 		for (next = wr; next; next = next->next) {
1354 			++resp.bad_wr;
1355 			if (next == bad_wr)
1356 				break;
1357 		}
1358 
1359 
1360 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
1361 			 &resp, sizeof resp))
1362 		ret = -EFAULT;
1363 
1364 out:
1365 	up(&ib_uverbs_idr_mutex);
1366 
1367 	while (wr) {
1368 		next = wr->next;
1369 		kfree(wr);
1370 		wr = next;
1371 	}
1372 
1373 	return ret ? ret : in_len;
1374 }
1375 
1376 ssize_t ib_uverbs_post_srq_recv(struct ib_uverbs_file *file,
1377                             const char __user *buf, int in_len,
1378                             int out_len)
1379 {
1380 	struct ib_uverbs_post_srq_recv      cmd;
1381 	struct ib_uverbs_post_srq_recv_resp resp;
1382 	struct ib_recv_wr                  *wr, *next, *bad_wr;
1383 	struct ib_srq                      *srq;
1384 	ssize_t                             ret = -EINVAL;
1385 
1386 	if (copy_from_user(&cmd, buf, sizeof cmd))
1387 		return -EFAULT;
1388 
1389 	wr = ib_uverbs_unmarshall_recv(buf + sizeof cmd,
1390 				       in_len - sizeof cmd, cmd.wr_count,
1391 				       cmd.sge_count, cmd.wqe_size);
1392 	if (IS_ERR(wr))
1393 		return PTR_ERR(wr);
1394 
1395 	down(&ib_uverbs_idr_mutex);
1396 
1397 	srq = idr_find(&ib_uverbs_srq_idr, cmd.srq_handle);
1398 	if (!srq || srq->uobject->context != file->ucontext)
1399 		goto out;
1400 
1401 	resp.bad_wr = 0;
1402 	ret = srq->device->post_srq_recv(srq, wr, &bad_wr);
1403 	if (ret)
1404 		for (next = wr; next; next = next->next) {
1405 			++resp.bad_wr;
1406 			if (next == bad_wr)
1407 				break;
1408 		}
1409 
1410 
1411 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
1412 			 &resp, sizeof resp))
1413 		ret = -EFAULT;
1414 
1415 out:
1416 	up(&ib_uverbs_idr_mutex);
1417 
1418 	while (wr) {
1419 		next = wr->next;
1420 		kfree(wr);
1421 		wr = next;
1422 	}
1423 
1424 	return ret ? ret : in_len;
1425 }
1426 
1427 ssize_t ib_uverbs_create_ah(struct ib_uverbs_file *file,
1428 			    const char __user *buf, int in_len,
1429 			    int out_len)
1430 {
1431 	struct ib_uverbs_create_ah	 cmd;
1432 	struct ib_uverbs_create_ah_resp	 resp;
1433 	struct ib_uobject		*uobj;
1434 	struct ib_pd			*pd;
1435 	struct ib_ah			*ah;
1436 	struct ib_ah_attr		attr;
1437 	int ret;
1438 
1439 	if (out_len < sizeof resp)
1440 		return -ENOSPC;
1441 
1442 	if (copy_from_user(&cmd, buf, sizeof cmd))
1443 		return -EFAULT;
1444 
1445 	uobj = kmalloc(sizeof *uobj, GFP_KERNEL);
1446 	if (!uobj)
1447 		return -ENOMEM;
1448 
1449 	down(&ib_uverbs_idr_mutex);
1450 
1451 	pd = idr_find(&ib_uverbs_pd_idr, cmd.pd_handle);
1452 	if (!pd || pd->uobject->context != file->ucontext) {
1453 		ret = -EINVAL;
1454 		goto err_up;
1455 	}
1456 
1457 	uobj->user_handle = cmd.user_handle;
1458 	uobj->context     = file->ucontext;
1459 
1460 	attr.dlid 	       = cmd.attr.dlid;
1461 	attr.sl 	       = cmd.attr.sl;
1462 	attr.src_path_bits     = cmd.attr.src_path_bits;
1463 	attr.static_rate       = cmd.attr.static_rate;
1464 	attr.ah_flags          = cmd.attr.is_global ? IB_AH_GRH : 0;
1465 	attr.port_num 	       = cmd.attr.port_num;
1466 	attr.grh.flow_label    = cmd.attr.grh.flow_label;
1467 	attr.grh.sgid_index    = cmd.attr.grh.sgid_index;
1468 	attr.grh.hop_limit     = cmd.attr.grh.hop_limit;
1469 	attr.grh.traffic_class = cmd.attr.grh.traffic_class;
1470 	memcpy(attr.grh.dgid.raw, cmd.attr.grh.dgid, 16);
1471 
1472 	ah = ib_create_ah(pd, &attr);
1473 	if (IS_ERR(ah)) {
1474 		ret = PTR_ERR(ah);
1475 		goto err_up;
1476 	}
1477 
1478 	ah->uobject = uobj;
1479 
1480 retry:
1481 	if (!idr_pre_get(&ib_uverbs_ah_idr, GFP_KERNEL)) {
1482 		ret = -ENOMEM;
1483 		goto err_destroy;
1484 	}
1485 
1486 	ret = idr_get_new(&ib_uverbs_ah_idr, ah, &uobj->id);
1487 
1488 	if (ret == -EAGAIN)
1489 		goto retry;
1490 	if (ret)
1491 		goto err_destroy;
1492 
1493 	resp.ah_handle = uobj->id;
1494 
1495 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
1496 			 &resp, sizeof resp)) {
1497 		ret = -EFAULT;
1498 		goto err_idr;
1499 	}
1500 
1501 	down(&file->mutex);
1502 	list_add_tail(&uobj->list, &file->ucontext->ah_list);
1503 	up(&file->mutex);
1504 
1505 	up(&ib_uverbs_idr_mutex);
1506 
1507 	return in_len;
1508 
1509 err_idr:
1510 	idr_remove(&ib_uverbs_ah_idr, uobj->id);
1511 
1512 err_destroy:
1513 	ib_destroy_ah(ah);
1514 
1515 err_up:
1516 	up(&ib_uverbs_idr_mutex);
1517 
1518 	kfree(uobj);
1519 	return ret;
1520 }
1521 
1522 ssize_t ib_uverbs_destroy_ah(struct ib_uverbs_file *file,
1523 			     const char __user *buf, int in_len, int out_len)
1524 {
1525 	struct ib_uverbs_destroy_ah cmd;
1526 	struct ib_ah		   *ah;
1527 	struct ib_uobject	   *uobj;
1528 	int			    ret = -EINVAL;
1529 
1530 	if (copy_from_user(&cmd, buf, sizeof cmd))
1531 		return -EFAULT;
1532 
1533 	down(&ib_uverbs_idr_mutex);
1534 
1535 	ah = idr_find(&ib_uverbs_ah_idr, cmd.ah_handle);
1536 	if (!ah || ah->uobject->context != file->ucontext)
1537 		goto out;
1538 
1539 	uobj = ah->uobject;
1540 
1541 	ret = ib_destroy_ah(ah);
1542 	if (ret)
1543 		goto out;
1544 
1545 	idr_remove(&ib_uverbs_ah_idr, cmd.ah_handle);
1546 
1547 	down(&file->mutex);
1548 	list_del(&uobj->list);
1549 	up(&file->mutex);
1550 
1551 	kfree(uobj);
1552 
1553 out:
1554 	up(&ib_uverbs_idr_mutex);
1555 
1556 	return ret ? ret : in_len;
1557 }
1558 
1559 ssize_t ib_uverbs_attach_mcast(struct ib_uverbs_file *file,
1560 			       const char __user *buf, int in_len,
1561 			       int out_len)
1562 {
1563 	struct ib_uverbs_attach_mcast cmd;
1564 	struct ib_qp                 *qp;
1565 	struct ib_uqp_object         *uobj;
1566 	struct ib_uverbs_mcast_entry *mcast;
1567 	int                           ret = -EINVAL;
1568 
1569 	if (copy_from_user(&cmd, buf, sizeof cmd))
1570 		return -EFAULT;
1571 
1572 	down(&ib_uverbs_idr_mutex);
1573 
1574 	qp = idr_find(&ib_uverbs_qp_idr, cmd.qp_handle);
1575 	if (!qp || qp->uobject->context != file->ucontext)
1576 		goto out;
1577 
1578 	uobj = container_of(qp->uobject, struct ib_uqp_object, uevent.uobject);
1579 
1580 	list_for_each_entry(mcast, &uobj->mcast_list, list)
1581 		if (cmd.mlid == mcast->lid &&
1582 		    !memcmp(cmd.gid, mcast->gid.raw, sizeof mcast->gid.raw)) {
1583 			ret = 0;
1584 			goto out;
1585 		}
1586 
1587 	mcast = kmalloc(sizeof *mcast, GFP_KERNEL);
1588 	if (!mcast) {
1589 		ret = -ENOMEM;
1590 		goto out;
1591 	}
1592 
1593 	mcast->lid = cmd.mlid;
1594 	memcpy(mcast->gid.raw, cmd.gid, sizeof mcast->gid.raw);
1595 
1596 	ret = ib_attach_mcast(qp, &mcast->gid, cmd.mlid);
1597 	if (!ret) {
1598 		uobj = container_of(qp->uobject, struct ib_uqp_object,
1599 				    uevent.uobject);
1600 		list_add_tail(&mcast->list, &uobj->mcast_list);
1601 	} else
1602 		kfree(mcast);
1603 
1604 out:
1605 	up(&ib_uverbs_idr_mutex);
1606 
1607 	return ret ? ret : in_len;
1608 }
1609 
1610 ssize_t ib_uverbs_detach_mcast(struct ib_uverbs_file *file,
1611 			       const char __user *buf, int in_len,
1612 			       int out_len)
1613 {
1614 	struct ib_uverbs_detach_mcast cmd;
1615 	struct ib_uqp_object         *uobj;
1616 	struct ib_qp                 *qp;
1617 	struct ib_uverbs_mcast_entry *mcast;
1618 	int                           ret = -EINVAL;
1619 
1620 	if (copy_from_user(&cmd, buf, sizeof cmd))
1621 		return -EFAULT;
1622 
1623 	down(&ib_uverbs_idr_mutex);
1624 
1625 	qp = idr_find(&ib_uverbs_qp_idr, cmd.qp_handle);
1626 	if (!qp || qp->uobject->context != file->ucontext)
1627 		goto out;
1628 
1629 	ret = ib_detach_mcast(qp, (union ib_gid *) cmd.gid, cmd.mlid);
1630 	if (ret)
1631 		goto out;
1632 
1633 	uobj = container_of(qp->uobject, struct ib_uqp_object, uevent.uobject);
1634 
1635 	list_for_each_entry(mcast, &uobj->mcast_list, list)
1636 		if (cmd.mlid == mcast->lid &&
1637 		    !memcmp(cmd.gid, mcast->gid.raw, sizeof mcast->gid.raw)) {
1638 			list_del(&mcast->list);
1639 			kfree(mcast);
1640 			break;
1641 		}
1642 
1643 out:
1644 	up(&ib_uverbs_idr_mutex);
1645 
1646 	return ret ? ret : in_len;
1647 }
1648 
1649 ssize_t ib_uverbs_create_srq(struct ib_uverbs_file *file,
1650 			     const char __user *buf, int in_len,
1651 			     int out_len)
1652 {
1653 	struct ib_uverbs_create_srq      cmd;
1654 	struct ib_uverbs_create_srq_resp resp;
1655 	struct ib_udata                  udata;
1656 	struct ib_uevent_object         *uobj;
1657 	struct ib_pd                    *pd;
1658 	struct ib_srq                   *srq;
1659 	struct ib_srq_init_attr          attr;
1660 	int ret;
1661 
1662 	if (out_len < sizeof resp)
1663 		return -ENOSPC;
1664 
1665 	if (copy_from_user(&cmd, buf, sizeof cmd))
1666 		return -EFAULT;
1667 
1668 	INIT_UDATA(&udata, buf + sizeof cmd,
1669 		   (unsigned long) cmd.response + sizeof resp,
1670 		   in_len - sizeof cmd, out_len - sizeof resp);
1671 
1672 	uobj = kmalloc(sizeof *uobj, GFP_KERNEL);
1673 	if (!uobj)
1674 		return -ENOMEM;
1675 
1676 	down(&ib_uverbs_idr_mutex);
1677 
1678 	pd  = idr_find(&ib_uverbs_pd_idr, cmd.pd_handle);
1679 
1680 	if (!pd || pd->uobject->context != file->ucontext) {
1681 		ret = -EINVAL;
1682 		goto err_up;
1683 	}
1684 
1685 	attr.event_handler  = ib_uverbs_srq_event_handler;
1686 	attr.srq_context    = file;
1687 	attr.attr.max_wr    = cmd.max_wr;
1688 	attr.attr.max_sge   = cmd.max_sge;
1689 	attr.attr.srq_limit = cmd.srq_limit;
1690 
1691 	uobj->uobject.user_handle = cmd.user_handle;
1692 	uobj->uobject.context     = file->ucontext;
1693 	uobj->events_reported     = 0;
1694 	INIT_LIST_HEAD(&uobj->event_list);
1695 
1696 	srq = pd->device->create_srq(pd, &attr, &udata);
1697 	if (IS_ERR(srq)) {
1698 		ret = PTR_ERR(srq);
1699 		goto err_up;
1700 	}
1701 
1702 	srq->device    	   = pd->device;
1703 	srq->pd        	   = pd;
1704 	srq->uobject       = &uobj->uobject;
1705 	srq->event_handler = attr.event_handler;
1706 	srq->srq_context   = attr.srq_context;
1707 	atomic_inc(&pd->usecnt);
1708 	atomic_set(&srq->usecnt, 0);
1709 
1710 	memset(&resp, 0, sizeof resp);
1711 
1712 retry:
1713 	if (!idr_pre_get(&ib_uverbs_srq_idr, GFP_KERNEL)) {
1714 		ret = -ENOMEM;
1715 		goto err_destroy;
1716 	}
1717 
1718 	ret = idr_get_new(&ib_uverbs_srq_idr, srq, &uobj->uobject.id);
1719 
1720 	if (ret == -EAGAIN)
1721 		goto retry;
1722 	if (ret)
1723 		goto err_destroy;
1724 
1725 	resp.srq_handle = uobj->uobject.id;
1726 
1727 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
1728 			 &resp, sizeof resp)) {
1729 		ret = -EFAULT;
1730 		goto err_idr;
1731 	}
1732 
1733 	down(&file->mutex);
1734 	list_add_tail(&uobj->uobject.list, &file->ucontext->srq_list);
1735 	up(&file->mutex);
1736 
1737 	up(&ib_uverbs_idr_mutex);
1738 
1739 	return in_len;
1740 
1741 err_idr:
1742 	idr_remove(&ib_uverbs_srq_idr, uobj->uobject.id);
1743 
1744 err_destroy:
1745 	ib_destroy_srq(srq);
1746 	atomic_dec(&pd->usecnt);
1747 
1748 err_up:
1749 	up(&ib_uverbs_idr_mutex);
1750 
1751 	kfree(uobj);
1752 	return ret;
1753 }
1754 
1755 ssize_t ib_uverbs_modify_srq(struct ib_uverbs_file *file,
1756 			     const char __user *buf, int in_len,
1757 			     int out_len)
1758 {
1759 	struct ib_uverbs_modify_srq cmd;
1760 	struct ib_srq              *srq;
1761 	struct ib_srq_attr          attr;
1762 	int                         ret;
1763 
1764 	if (copy_from_user(&cmd, buf, sizeof cmd))
1765 		return -EFAULT;
1766 
1767 	down(&ib_uverbs_idr_mutex);
1768 
1769 	srq = idr_find(&ib_uverbs_srq_idr, cmd.srq_handle);
1770 	if (!srq || srq->uobject->context != file->ucontext) {
1771 		ret = -EINVAL;
1772 		goto out;
1773 	}
1774 
1775 	attr.max_wr    = cmd.max_wr;
1776 	attr.srq_limit = cmd.srq_limit;
1777 
1778 	ret = ib_modify_srq(srq, &attr, cmd.attr_mask);
1779 
1780 out:
1781 	up(&ib_uverbs_idr_mutex);
1782 
1783 	return ret ? ret : in_len;
1784 }
1785 
1786 ssize_t ib_uverbs_destroy_srq(struct ib_uverbs_file *file,
1787 			      const char __user *buf, int in_len,
1788 			      int out_len)
1789 {
1790 	struct ib_uverbs_destroy_srq      cmd;
1791 	struct ib_uverbs_destroy_srq_resp resp;
1792 	struct ib_srq               	 *srq;
1793 	struct ib_uevent_object        	 *uobj;
1794 	int                         	  ret = -EINVAL;
1795 
1796 	if (copy_from_user(&cmd, buf, sizeof cmd))
1797 		return -EFAULT;
1798 
1799 	down(&ib_uverbs_idr_mutex);
1800 
1801 	memset(&resp, 0, sizeof resp);
1802 
1803 	srq = idr_find(&ib_uverbs_srq_idr, cmd.srq_handle);
1804 	if (!srq || srq->uobject->context != file->ucontext)
1805 		goto out;
1806 
1807 	uobj = container_of(srq->uobject, struct ib_uevent_object, uobject);
1808 
1809 	ret = ib_destroy_srq(srq);
1810 	if (ret)
1811 		goto out;
1812 
1813 	idr_remove(&ib_uverbs_srq_idr, cmd.srq_handle);
1814 
1815 	down(&file->mutex);
1816 	list_del(&uobj->uobject.list);
1817 	up(&file->mutex);
1818 
1819 	ib_uverbs_release_uevent(file, uobj);
1820 
1821 	resp.events_reported = uobj->events_reported;
1822 
1823 	kfree(uobj);
1824 
1825 	if (copy_to_user((void __user *) (unsigned long) cmd.response,
1826 			 &resp, sizeof resp))
1827 		ret = -EFAULT;
1828 
1829 out:
1830 	up(&ib_uverbs_idr_mutex);
1831 
1832 	return ret ? ret : in_len;
1833 }
1834