2 * Copyright (c) 2005-2006 Intel Corporation. All rights reserved.
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
33 #include <linux/completion.h>
34 #include <linux/file.h>
35 #include <linux/mutex.h>
36 #include <linux/poll.h>
37 #include <linux/sched.h>
38 #include <linux/idr.h>
40 #include <linux/in6.h>
41 #include <linux/miscdevice.h>
42 #include <linux/slab.h>
43 #include <linux/sysctl.h>
44 #include <linux/module.h>
46 #include <rdma/rdma_user_cm.h>
47 #include <rdma/ib_marshall.h>
48 #include <rdma/rdma_cm.h>
49 #include <rdma/rdma_cm_ib.h>
51 MODULE_AUTHOR("Sean Hefty");
52 MODULE_DESCRIPTION("RDMA Userspace Connection Manager Access");
53 MODULE_LICENSE("Dual BSD/GPL");
55 static unsigned int max_backlog
= 1024;
57 static struct ctl_table_header
*ucma_ctl_table_hdr
;
58 static ctl_table ucma_ctl_table
[] = {
60 .procname
= "max_backlog",
62 .maxlen
= sizeof max_backlog
,
64 .proc_handler
= proc_dointvec
,
69 static struct ctl_path ucma_ctl_path
[] = {
70 { .procname
= "net" },
71 { .procname
= "rdma_ucm" },
78 struct list_head ctx_list
;
79 struct list_head event_list
;
80 wait_queue_head_t poll_wait
;
85 struct completion comp
;
90 struct ucma_file
*file
;
91 struct rdma_cm_id
*cm_id
;
94 struct list_head list
;
95 struct list_head mc_list
;
98 struct ucma_multicast
{
99 struct ucma_context
*ctx
;
104 struct list_head list
;
105 struct sockaddr_storage addr
;
109 struct ucma_context
*ctx
;
110 struct ucma_multicast
*mc
;
111 struct list_head list
;
112 struct rdma_cm_id
*cm_id
;
113 struct rdma_ucm_event_resp resp
;
116 static DEFINE_MUTEX(mut
);
117 static DEFINE_IDR(ctx_idr
);
118 static DEFINE_IDR(multicast_idr
);
120 static inline struct ucma_context
*_ucma_find_context(int id
,
121 struct ucma_file
*file
)
123 struct ucma_context
*ctx
;
125 ctx
= idr_find(&ctx_idr
, id
);
127 ctx
= ERR_PTR(-ENOENT
);
128 else if (ctx
->file
!= file
)
129 ctx
= ERR_PTR(-EINVAL
);
133 static struct ucma_context
*ucma_get_ctx(struct ucma_file
*file
, int id
)
135 struct ucma_context
*ctx
;
138 ctx
= _ucma_find_context(id
, file
);
140 atomic_inc(&ctx
->ref
);
145 static void ucma_put_ctx(struct ucma_context
*ctx
)
147 if (atomic_dec_and_test(&ctx
->ref
))
148 complete(&ctx
->comp
);
151 static struct ucma_context
*ucma_alloc_ctx(struct ucma_file
*file
)
153 struct ucma_context
*ctx
;
156 ctx
= kzalloc(sizeof(*ctx
), GFP_KERNEL
);
160 atomic_set(&ctx
->ref
, 1);
161 init_completion(&ctx
->comp
);
162 INIT_LIST_HEAD(&ctx
->mc_list
);
166 ret
= idr_pre_get(&ctx_idr
, GFP_KERNEL
);
171 ret
= idr_get_new(&ctx_idr
, ctx
, &ctx
->id
);
173 } while (ret
== -EAGAIN
);
178 list_add_tail(&ctx
->list
, &file
->ctx_list
);
186 static struct ucma_multicast
* ucma_alloc_multicast(struct ucma_context
*ctx
)
188 struct ucma_multicast
*mc
;
191 mc
= kzalloc(sizeof(*mc
), GFP_KERNEL
);
196 ret
= idr_pre_get(&multicast_idr
, GFP_KERNEL
);
201 ret
= idr_get_new(&multicast_idr
, mc
, &mc
->id
);
203 } while (ret
== -EAGAIN
);
209 list_add_tail(&mc
->list
, &ctx
->mc_list
);
217 static void ucma_copy_conn_event(struct rdma_ucm_conn_param
*dst
,
218 struct rdma_conn_param
*src
)
220 if (src
->private_data_len
)
221 memcpy(dst
->private_data
, src
->private_data
,
222 src
->private_data_len
);
223 dst
->private_data_len
= src
->private_data_len
;
224 dst
->responder_resources
=src
->responder_resources
;
225 dst
->initiator_depth
= src
->initiator_depth
;
226 dst
->flow_control
= src
->flow_control
;
227 dst
->retry_count
= src
->retry_count
;
228 dst
->rnr_retry_count
= src
->rnr_retry_count
;
230 dst
->qp_num
= src
->qp_num
;
233 static void ucma_copy_ud_event(struct rdma_ucm_ud_param
*dst
,
234 struct rdma_ud_param
*src
)
236 if (src
->private_data_len
)
237 memcpy(dst
->private_data
, src
->private_data
,
238 src
->private_data_len
);
239 dst
->private_data_len
= src
->private_data_len
;
240 ib_copy_ah_attr_to_user(&dst
->ah_attr
, &src
->ah_attr
);
241 dst
->qp_num
= src
->qp_num
;
242 dst
->qkey
= src
->qkey
;
245 static void ucma_set_event_context(struct ucma_context
*ctx
,
246 struct rdma_cm_event
*event
,
247 struct ucma_event
*uevent
)
250 switch (event
->event
) {
251 case RDMA_CM_EVENT_MULTICAST_JOIN
:
252 case RDMA_CM_EVENT_MULTICAST_ERROR
:
253 uevent
->mc
= (struct ucma_multicast
*)
254 event
->param
.ud
.private_data
;
255 uevent
->resp
.uid
= uevent
->mc
->uid
;
256 uevent
->resp
.id
= uevent
->mc
->id
;
259 uevent
->resp
.uid
= ctx
->uid
;
260 uevent
->resp
.id
= ctx
->id
;
265 static int ucma_event_handler(struct rdma_cm_id
*cm_id
,
266 struct rdma_cm_event
*event
)
268 struct ucma_event
*uevent
;
269 struct ucma_context
*ctx
= cm_id
->context
;
272 uevent
= kzalloc(sizeof(*uevent
), GFP_KERNEL
);
274 return event
->event
== RDMA_CM_EVENT_CONNECT_REQUEST
;
276 uevent
->cm_id
= cm_id
;
277 ucma_set_event_context(ctx
, event
, uevent
);
278 uevent
->resp
.event
= event
->event
;
279 uevent
->resp
.status
= event
->status
;
280 if (cm_id
->ps
== RDMA_PS_UDP
|| cm_id
->ps
== RDMA_PS_IPOIB
)
281 ucma_copy_ud_event(&uevent
->resp
.param
.ud
, &event
->param
.ud
);
283 ucma_copy_conn_event(&uevent
->resp
.param
.conn
,
286 mutex_lock(&ctx
->file
->mut
);
287 if (event
->event
== RDMA_CM_EVENT_CONNECT_REQUEST
) {
294 } else if (!ctx
->uid
) {
296 * We ignore events for new connections until userspace has set
297 * their context. This can only happen if an error occurs on a
298 * new connection before the user accepts it. This is okay,
299 * since the accept will just fail later.
305 list_add_tail(&uevent
->list
, &ctx
->file
->event_list
);
306 wake_up_interruptible(&ctx
->file
->poll_wait
);
308 mutex_unlock(&ctx
->file
->mut
);
312 static ssize_t
ucma_get_event(struct ucma_file
*file
, const char __user
*inbuf
,
313 int in_len
, int out_len
)
315 struct ucma_context
*ctx
;
316 struct rdma_ucm_get_event cmd
;
317 struct ucma_event
*uevent
;
321 if (out_len
< sizeof uevent
->resp
)
324 if (copy_from_user(&cmd
, inbuf
, sizeof(cmd
)))
327 mutex_lock(&file
->mut
);
328 while (list_empty(&file
->event_list
)) {
329 mutex_unlock(&file
->mut
);
331 if (file
->filp
->f_flags
& O_NONBLOCK
)
334 if (wait_event_interruptible(file
->poll_wait
,
335 !list_empty(&file
->event_list
)))
338 mutex_lock(&file
->mut
);
341 uevent
= list_entry(file
->event_list
.next
, struct ucma_event
, list
);
343 if (uevent
->resp
.event
== RDMA_CM_EVENT_CONNECT_REQUEST
) {
344 ctx
= ucma_alloc_ctx(file
);
349 uevent
->ctx
->backlog
++;
350 ctx
->cm_id
= uevent
->cm_id
;
351 ctx
->cm_id
->context
= ctx
;
352 uevent
->resp
.id
= ctx
->id
;
355 if (copy_to_user((void __user
*)(unsigned long)cmd
.response
,
356 &uevent
->resp
, sizeof uevent
->resp
)) {
361 list_del(&uevent
->list
);
362 uevent
->ctx
->events_reported
++;
364 uevent
->mc
->events_reported
++;
367 mutex_unlock(&file
->mut
);
371 static int ucma_get_qp_type(struct rdma_ucm_create_id
*cmd
, enum ib_qp_type
*qp_type
)
375 *qp_type
= IB_QPT_RC
;
379 *qp_type
= IB_QPT_UD
;
386 static ssize_t
ucma_create_id(struct ucma_file
*file
, const char __user
*inbuf
,
387 int in_len
, int out_len
)
389 struct rdma_ucm_create_id cmd
;
390 struct rdma_ucm_create_id_resp resp
;
391 struct ucma_context
*ctx
;
392 enum ib_qp_type qp_type
;
395 if (out_len
< sizeof(resp
))
398 if (copy_from_user(&cmd
, inbuf
, sizeof(cmd
)))
401 ret
= ucma_get_qp_type(&cmd
, &qp_type
);
405 mutex_lock(&file
->mut
);
406 ctx
= ucma_alloc_ctx(file
);
407 mutex_unlock(&file
->mut
);
412 ctx
->cm_id
= rdma_create_id(ucma_event_handler
, ctx
, cmd
.ps
, qp_type
);
413 if (IS_ERR(ctx
->cm_id
)) {
414 ret
= PTR_ERR(ctx
->cm_id
);
419 if (copy_to_user((void __user
*)(unsigned long)cmd
.response
,
420 &resp
, sizeof(resp
))) {
427 rdma_destroy_id(ctx
->cm_id
);
430 idr_remove(&ctx_idr
, ctx
->id
);
436 static void ucma_cleanup_multicast(struct ucma_context
*ctx
)
438 struct ucma_multicast
*mc
, *tmp
;
441 list_for_each_entry_safe(mc
, tmp
, &ctx
->mc_list
, list
) {
443 idr_remove(&multicast_idr
, mc
->id
);
449 static void ucma_cleanup_events(struct ucma_context
*ctx
)
451 struct ucma_event
*uevent
, *tmp
;
453 list_for_each_entry_safe(uevent
, tmp
, &ctx
->file
->event_list
, list
) {
454 if (uevent
->ctx
!= ctx
)
457 list_del(&uevent
->list
);
459 /* clear incoming connections. */
460 if (uevent
->resp
.event
== RDMA_CM_EVENT_CONNECT_REQUEST
)
461 rdma_destroy_id(uevent
->cm_id
);
467 static void ucma_cleanup_mc_events(struct ucma_multicast
*mc
)
469 struct ucma_event
*uevent
, *tmp
;
471 list_for_each_entry_safe(uevent
, tmp
, &mc
->ctx
->file
->event_list
, list
) {
472 if (uevent
->mc
!= mc
)
475 list_del(&uevent
->list
);
480 static int ucma_free_ctx(struct ucma_context
*ctx
)
484 /* No new events will be generated after destroying the id. */
485 rdma_destroy_id(ctx
->cm_id
);
487 ucma_cleanup_multicast(ctx
);
489 /* Cleanup events not yet reported to the user. */
490 mutex_lock(&ctx
->file
->mut
);
491 ucma_cleanup_events(ctx
);
492 list_del(&ctx
->list
);
493 mutex_unlock(&ctx
->file
->mut
);
495 events_reported
= ctx
->events_reported
;
497 return events_reported
;
500 static ssize_t
ucma_destroy_id(struct ucma_file
*file
, const char __user
*inbuf
,
501 int in_len
, int out_len
)
503 struct rdma_ucm_destroy_id cmd
;
504 struct rdma_ucm_destroy_id_resp resp
;
505 struct ucma_context
*ctx
;
508 if (out_len
< sizeof(resp
))
511 if (copy_from_user(&cmd
, inbuf
, sizeof(cmd
)))
515 ctx
= _ucma_find_context(cmd
.id
, file
);
517 idr_remove(&ctx_idr
, ctx
->id
);
524 wait_for_completion(&ctx
->comp
);
525 resp
.events_reported
= ucma_free_ctx(ctx
);
527 if (copy_to_user((void __user
*)(unsigned long)cmd
.response
,
528 &resp
, sizeof(resp
)))
534 static ssize_t
ucma_bind_addr(struct ucma_file
*file
, const char __user
*inbuf
,
535 int in_len
, int out_len
)
537 struct rdma_ucm_bind_addr cmd
;
538 struct ucma_context
*ctx
;
541 if (copy_from_user(&cmd
, inbuf
, sizeof(cmd
)))
544 ctx
= ucma_get_ctx(file
, cmd
.id
);
548 ret
= rdma_bind_addr(ctx
->cm_id
, (struct sockaddr
*) &cmd
.addr
);
553 static ssize_t
ucma_resolve_addr(struct ucma_file
*file
,
554 const char __user
*inbuf
,
555 int in_len
, int out_len
)
557 struct rdma_ucm_resolve_addr cmd
;
558 struct ucma_context
*ctx
;
561 if (copy_from_user(&cmd
, inbuf
, sizeof(cmd
)))
564 ctx
= ucma_get_ctx(file
, cmd
.id
);
568 ret
= rdma_resolve_addr(ctx
->cm_id
, (struct sockaddr
*) &cmd
.src_addr
,
569 (struct sockaddr
*) &cmd
.dst_addr
,
575 static ssize_t
ucma_resolve_route(struct ucma_file
*file
,
576 const char __user
*inbuf
,
577 int in_len
, int out_len
)
579 struct rdma_ucm_resolve_route cmd
;
580 struct ucma_context
*ctx
;
583 if (copy_from_user(&cmd
, inbuf
, sizeof(cmd
)))
586 ctx
= ucma_get_ctx(file
, cmd
.id
);
590 ret
= rdma_resolve_route(ctx
->cm_id
, cmd
.timeout_ms
);
595 static void ucma_copy_ib_route(struct rdma_ucm_query_route_resp
*resp
,
596 struct rdma_route
*route
)
598 struct rdma_dev_addr
*dev_addr
;
600 resp
->num_paths
= route
->num_paths
;
601 switch (route
->num_paths
) {
603 dev_addr
= &route
->addr
.dev_addr
;
604 rdma_addr_get_dgid(dev_addr
,
605 (union ib_gid
*) &resp
->ib_route
[0].dgid
);
606 rdma_addr_get_sgid(dev_addr
,
607 (union ib_gid
*) &resp
->ib_route
[0].sgid
);
608 resp
->ib_route
[0].pkey
= cpu_to_be16(ib_addr_get_pkey(dev_addr
));
611 ib_copy_path_rec_to_user(&resp
->ib_route
[1],
612 &route
->path_rec
[1]);
615 ib_copy_path_rec_to_user(&resp
->ib_route
[0],
616 &route
->path_rec
[0]);
623 static void ucma_copy_iboe_route(struct rdma_ucm_query_route_resp
*resp
,
624 struct rdma_route
*route
)
626 struct rdma_dev_addr
*dev_addr
;
627 struct net_device
*dev
;
630 resp
->num_paths
= route
->num_paths
;
631 switch (route
->num_paths
) {
633 dev_addr
= &route
->addr
.dev_addr
;
634 dev
= dev_get_by_index(&init_net
, dev_addr
->bound_dev_if
);
636 vid
= rdma_vlan_dev_vlan_id(dev
);
640 iboe_mac_vlan_to_ll((union ib_gid
*) &resp
->ib_route
[0].dgid
,
641 dev_addr
->dst_dev_addr
, vid
);
642 iboe_addr_get_sgid(dev_addr
,
643 (union ib_gid
*) &resp
->ib_route
[0].sgid
);
644 resp
->ib_route
[0].pkey
= cpu_to_be16(0xffff);
647 ib_copy_path_rec_to_user(&resp
->ib_route
[1],
648 &route
->path_rec
[1]);
651 ib_copy_path_rec_to_user(&resp
->ib_route
[0],
652 &route
->path_rec
[0]);
659 static void ucma_copy_iw_route(struct rdma_ucm_query_route_resp
*resp
,
660 struct rdma_route
*route
)
662 struct rdma_dev_addr
*dev_addr
;
664 dev_addr
= &route
->addr
.dev_addr
;
665 rdma_addr_get_dgid(dev_addr
, (union ib_gid
*) &resp
->ib_route
[0].dgid
);
666 rdma_addr_get_sgid(dev_addr
, (union ib_gid
*) &resp
->ib_route
[0].sgid
);
669 static ssize_t
ucma_query_route(struct ucma_file
*file
,
670 const char __user
*inbuf
,
671 int in_len
, int out_len
)
673 struct rdma_ucm_query_route cmd
;
674 struct rdma_ucm_query_route_resp resp
;
675 struct ucma_context
*ctx
;
676 struct sockaddr
*addr
;
679 if (out_len
< sizeof(resp
))
682 if (copy_from_user(&cmd
, inbuf
, sizeof(cmd
)))
685 ctx
= ucma_get_ctx(file
, cmd
.id
);
689 memset(&resp
, 0, sizeof resp
);
690 addr
= (struct sockaddr
*) &ctx
->cm_id
->route
.addr
.src_addr
;
691 memcpy(&resp
.src_addr
, addr
, addr
->sa_family
== AF_INET
?
692 sizeof(struct sockaddr_in
) :
693 sizeof(struct sockaddr_in6
));
694 addr
= (struct sockaddr
*) &ctx
->cm_id
->route
.addr
.dst_addr
;
695 memcpy(&resp
.dst_addr
, addr
, addr
->sa_family
== AF_INET
?
696 sizeof(struct sockaddr_in
) :
697 sizeof(struct sockaddr_in6
));
698 if (!ctx
->cm_id
->device
)
701 resp
.node_guid
= (__force __u64
) ctx
->cm_id
->device
->node_guid
;
702 resp
.port_num
= ctx
->cm_id
->port_num
;
703 switch (rdma_node_get_transport(ctx
->cm_id
->device
->node_type
)) {
704 case RDMA_TRANSPORT_IB
:
705 switch (rdma_port_get_link_layer(ctx
->cm_id
->device
,
706 ctx
->cm_id
->port_num
)) {
707 case IB_LINK_LAYER_INFINIBAND
:
708 ucma_copy_ib_route(&resp
, &ctx
->cm_id
->route
);
710 case IB_LINK_LAYER_ETHERNET
:
711 ucma_copy_iboe_route(&resp
, &ctx
->cm_id
->route
);
717 case RDMA_TRANSPORT_IWARP
:
718 ucma_copy_iw_route(&resp
, &ctx
->cm_id
->route
);
725 if (copy_to_user((void __user
*)(unsigned long)cmd
.response
,
726 &resp
, sizeof(resp
)))
733 static void ucma_copy_conn_param(struct rdma_conn_param
*dst
,
734 struct rdma_ucm_conn_param
*src
)
736 dst
->private_data
= src
->private_data
;
737 dst
->private_data_len
= src
->private_data_len
;
738 dst
->responder_resources
=src
->responder_resources
;
739 dst
->initiator_depth
= src
->initiator_depth
;
740 dst
->flow_control
= src
->flow_control
;
741 dst
->retry_count
= src
->retry_count
;
742 dst
->rnr_retry_count
= src
->rnr_retry_count
;
744 dst
->qp_num
= src
->qp_num
;
747 static ssize_t
ucma_connect(struct ucma_file
*file
, const char __user
*inbuf
,
748 int in_len
, int out_len
)
750 struct rdma_ucm_connect cmd
;
751 struct rdma_conn_param conn_param
;
752 struct ucma_context
*ctx
;
755 if (copy_from_user(&cmd
, inbuf
, sizeof(cmd
)))
758 if (!cmd
.conn_param
.valid
)
761 ctx
= ucma_get_ctx(file
, cmd
.id
);
765 ucma_copy_conn_param(&conn_param
, &cmd
.conn_param
);
766 ret
= rdma_connect(ctx
->cm_id
, &conn_param
);
771 static ssize_t
ucma_listen(struct ucma_file
*file
, const char __user
*inbuf
,
772 int in_len
, int out_len
)
774 struct rdma_ucm_listen cmd
;
775 struct ucma_context
*ctx
;
778 if (copy_from_user(&cmd
, inbuf
, sizeof(cmd
)))
781 ctx
= ucma_get_ctx(file
, cmd
.id
);
785 ctx
->backlog
= cmd
.backlog
> 0 && cmd
.backlog
< max_backlog
?
786 cmd
.backlog
: max_backlog
;
787 ret
= rdma_listen(ctx
->cm_id
, ctx
->backlog
);
792 static ssize_t
ucma_accept(struct ucma_file
*file
, const char __user
*inbuf
,
793 int in_len
, int out_len
)
795 struct rdma_ucm_accept cmd
;
796 struct rdma_conn_param conn_param
;
797 struct ucma_context
*ctx
;
800 if (copy_from_user(&cmd
, inbuf
, sizeof(cmd
)))
803 ctx
= ucma_get_ctx(file
, cmd
.id
);
807 if (cmd
.conn_param
.valid
) {
809 ucma_copy_conn_param(&conn_param
, &cmd
.conn_param
);
810 ret
= rdma_accept(ctx
->cm_id
, &conn_param
);
812 ret
= rdma_accept(ctx
->cm_id
, NULL
);
818 static ssize_t
ucma_reject(struct ucma_file
*file
, const char __user
*inbuf
,
819 int in_len
, int out_len
)
821 struct rdma_ucm_reject cmd
;
822 struct ucma_context
*ctx
;
825 if (copy_from_user(&cmd
, inbuf
, sizeof(cmd
)))
828 ctx
= ucma_get_ctx(file
, cmd
.id
);
832 ret
= rdma_reject(ctx
->cm_id
, cmd
.private_data
, cmd
.private_data_len
);
837 static ssize_t
ucma_disconnect(struct ucma_file
*file
, const char __user
*inbuf
,
838 int in_len
, int out_len
)
840 struct rdma_ucm_disconnect cmd
;
841 struct ucma_context
*ctx
;
844 if (copy_from_user(&cmd
, inbuf
, sizeof(cmd
)))
847 ctx
= ucma_get_ctx(file
, cmd
.id
);
851 ret
= rdma_disconnect(ctx
->cm_id
);
856 static ssize_t
ucma_init_qp_attr(struct ucma_file
*file
,
857 const char __user
*inbuf
,
858 int in_len
, int out_len
)
860 struct rdma_ucm_init_qp_attr cmd
;
861 struct ib_uverbs_qp_attr resp
;
862 struct ucma_context
*ctx
;
863 struct ib_qp_attr qp_attr
;
866 if (out_len
< sizeof(resp
))
869 if (copy_from_user(&cmd
, inbuf
, sizeof(cmd
)))
872 ctx
= ucma_get_ctx(file
, cmd
.id
);
876 resp
.qp_attr_mask
= 0;
877 memset(&qp_attr
, 0, sizeof qp_attr
);
878 qp_attr
.qp_state
= cmd
.qp_state
;
879 ret
= rdma_init_qp_attr(ctx
->cm_id
, &qp_attr
, &resp
.qp_attr_mask
);
883 ib_copy_qp_attr_to_user(&resp
, &qp_attr
);
884 if (copy_to_user((void __user
*)(unsigned long)cmd
.response
,
885 &resp
, sizeof(resp
)))
893 static int ucma_set_option_id(struct ucma_context
*ctx
, int optname
,
894 void *optval
, size_t optlen
)
899 case RDMA_OPTION_ID_TOS
:
900 if (optlen
!= sizeof(u8
)) {
904 rdma_set_service_type(ctx
->cm_id
, *((u8
*) optval
));
906 case RDMA_OPTION_ID_REUSEADDR
:
907 if (optlen
!= sizeof(int)) {
911 ret
= rdma_set_reuseaddr(ctx
->cm_id
, *((int *) optval
) ? 1 : 0);
920 static int ucma_set_ib_path(struct ucma_context
*ctx
,
921 struct ib_path_rec_data
*path_data
, size_t optlen
)
923 struct ib_sa_path_rec sa_path
;
924 struct rdma_cm_event event
;
927 if (optlen
% sizeof(*path_data
))
930 for (; optlen
; optlen
-= sizeof(*path_data
), path_data
++) {
931 if (path_data
->flags
== (IB_PATH_GMP
| IB_PATH_PRIMARY
|
932 IB_PATH_BIDIRECTIONAL
))
939 ib_sa_unpack_path(path_data
->path_rec
, &sa_path
);
940 ret
= rdma_set_ib_paths(ctx
->cm_id
, &sa_path
, 1);
944 memset(&event
, 0, sizeof event
);
945 event
.event
= RDMA_CM_EVENT_ROUTE_RESOLVED
;
946 return ucma_event_handler(ctx
->cm_id
, &event
);
949 static int ucma_set_option_ib(struct ucma_context
*ctx
, int optname
,
950 void *optval
, size_t optlen
)
955 case RDMA_OPTION_IB_PATH
:
956 ret
= ucma_set_ib_path(ctx
, optval
, optlen
);
965 static int ucma_set_option_level(struct ucma_context
*ctx
, int level
,
966 int optname
, void *optval
, size_t optlen
)
972 ret
= ucma_set_option_id(ctx
, optname
, optval
, optlen
);
975 ret
= ucma_set_option_ib(ctx
, optname
, optval
, optlen
);
984 static ssize_t
ucma_set_option(struct ucma_file
*file
, const char __user
*inbuf
,
985 int in_len
, int out_len
)
987 struct rdma_ucm_set_option cmd
;
988 struct ucma_context
*ctx
;
992 if (copy_from_user(&cmd
, inbuf
, sizeof(cmd
)))
995 ctx
= ucma_get_ctx(file
, cmd
.id
);
999 optval
= kmalloc(cmd
.optlen
, GFP_KERNEL
);
1005 if (copy_from_user(optval
, (void __user
*) (unsigned long) cmd
.optval
,
1011 ret
= ucma_set_option_level(ctx
, cmd
.level
, cmd
.optname
, optval
,
1020 static ssize_t
ucma_notify(struct ucma_file
*file
, const char __user
*inbuf
,
1021 int in_len
, int out_len
)
1023 struct rdma_ucm_notify cmd
;
1024 struct ucma_context
*ctx
;
1027 if (copy_from_user(&cmd
, inbuf
, sizeof(cmd
)))
1030 ctx
= ucma_get_ctx(file
, cmd
.id
);
1032 return PTR_ERR(ctx
);
1034 ret
= rdma_notify(ctx
->cm_id
, (enum ib_event_type
) cmd
.event
);
1039 static ssize_t
ucma_join_multicast(struct ucma_file
*file
,
1040 const char __user
*inbuf
,
1041 int in_len
, int out_len
)
1043 struct rdma_ucm_join_mcast cmd
;
1044 struct rdma_ucm_create_id_resp resp
;
1045 struct ucma_context
*ctx
;
1046 struct ucma_multicast
*mc
;
1049 if (out_len
< sizeof(resp
))
1052 if (copy_from_user(&cmd
, inbuf
, sizeof(cmd
)))
1055 ctx
= ucma_get_ctx(file
, cmd
.id
);
1057 return PTR_ERR(ctx
);
1059 mutex_lock(&file
->mut
);
1060 mc
= ucma_alloc_multicast(ctx
);
1067 memcpy(&mc
->addr
, &cmd
.addr
, sizeof cmd
.addr
);
1068 ret
= rdma_join_multicast(ctx
->cm_id
, (struct sockaddr
*) &mc
->addr
, mc
);
1073 if (copy_to_user((void __user
*)(unsigned long)cmd
.response
,
1074 &resp
, sizeof(resp
))) {
1079 mutex_unlock(&file
->mut
);
1084 rdma_leave_multicast(ctx
->cm_id
, (struct sockaddr
*) &mc
->addr
);
1085 ucma_cleanup_mc_events(mc
);
1088 idr_remove(&multicast_idr
, mc
->id
);
1090 list_del(&mc
->list
);
1093 mutex_unlock(&file
->mut
);
1098 static ssize_t
ucma_leave_multicast(struct ucma_file
*file
,
1099 const char __user
*inbuf
,
1100 int in_len
, int out_len
)
1102 struct rdma_ucm_destroy_id cmd
;
1103 struct rdma_ucm_destroy_id_resp resp
;
1104 struct ucma_multicast
*mc
;
1107 if (out_len
< sizeof(resp
))
1110 if (copy_from_user(&cmd
, inbuf
, sizeof(cmd
)))
1114 mc
= idr_find(&multicast_idr
, cmd
.id
);
1116 mc
= ERR_PTR(-ENOENT
);
1117 else if (mc
->ctx
->file
!= file
)
1118 mc
= ERR_PTR(-EINVAL
);
1120 idr_remove(&multicast_idr
, mc
->id
);
1121 atomic_inc(&mc
->ctx
->ref
);
1130 rdma_leave_multicast(mc
->ctx
->cm_id
, (struct sockaddr
*) &mc
->addr
);
1131 mutex_lock(&mc
->ctx
->file
->mut
);
1132 ucma_cleanup_mc_events(mc
);
1133 list_del(&mc
->list
);
1134 mutex_unlock(&mc
->ctx
->file
->mut
);
1136 ucma_put_ctx(mc
->ctx
);
1137 resp
.events_reported
= mc
->events_reported
;
1140 if (copy_to_user((void __user
*)(unsigned long)cmd
.response
,
1141 &resp
, sizeof(resp
)))
1147 static void ucma_lock_files(struct ucma_file
*file1
, struct ucma_file
*file2
)
1149 /* Acquire mutex's based on pointer comparison to prevent deadlock. */
1150 if (file1
< file2
) {
1151 mutex_lock(&file1
->mut
);
1152 mutex_lock(&file2
->mut
);
1154 mutex_lock(&file2
->mut
);
1155 mutex_lock(&file1
->mut
);
1159 static void ucma_unlock_files(struct ucma_file
*file1
, struct ucma_file
*file2
)
1161 if (file1
< file2
) {
1162 mutex_unlock(&file2
->mut
);
1163 mutex_unlock(&file1
->mut
);
1165 mutex_unlock(&file1
->mut
);
1166 mutex_unlock(&file2
->mut
);
1170 static void ucma_move_events(struct ucma_context
*ctx
, struct ucma_file
*file
)
1172 struct ucma_event
*uevent
, *tmp
;
1174 list_for_each_entry_safe(uevent
, tmp
, &ctx
->file
->event_list
, list
)
1175 if (uevent
->ctx
== ctx
)
1176 list_move_tail(&uevent
->list
, &file
->event_list
);
1179 static ssize_t
ucma_migrate_id(struct ucma_file
*new_file
,
1180 const char __user
*inbuf
,
1181 int in_len
, int out_len
)
1183 struct rdma_ucm_migrate_id cmd
;
1184 struct rdma_ucm_migrate_resp resp
;
1185 struct ucma_context
*ctx
;
1187 struct ucma_file
*cur_file
;
1190 if (copy_from_user(&cmd
, inbuf
, sizeof(cmd
)))
1193 /* Get current fd to protect against it being closed */
1194 filp
= fget(cmd
.fd
);
1198 /* Validate current fd and prevent destruction of id. */
1199 ctx
= ucma_get_ctx(filp
->private_data
, cmd
.id
);
1205 cur_file
= ctx
->file
;
1206 if (cur_file
== new_file
) {
1207 resp
.events_reported
= ctx
->events_reported
;
1212 * Migrate events between fd's, maintaining order, and avoiding new
1213 * events being added before existing events.
1215 ucma_lock_files(cur_file
, new_file
);
1218 list_move_tail(&ctx
->list
, &new_file
->ctx_list
);
1219 ucma_move_events(ctx
, new_file
);
1220 ctx
->file
= new_file
;
1221 resp
.events_reported
= ctx
->events_reported
;
1224 ucma_unlock_files(cur_file
, new_file
);
1227 if (copy_to_user((void __user
*)(unsigned long)cmd
.response
,
1228 &resp
, sizeof(resp
)))
1237 static ssize_t (*ucma_cmd_table
[])(struct ucma_file
*file
,
1238 const char __user
*inbuf
,
1239 int in_len
, int out_len
) = {
1240 [RDMA_USER_CM_CMD_CREATE_ID
] = ucma_create_id
,
1241 [RDMA_USER_CM_CMD_DESTROY_ID
] = ucma_destroy_id
,
1242 [RDMA_USER_CM_CMD_BIND_ADDR
] = ucma_bind_addr
,
1243 [RDMA_USER_CM_CMD_RESOLVE_ADDR
] = ucma_resolve_addr
,
1244 [RDMA_USER_CM_CMD_RESOLVE_ROUTE
]= ucma_resolve_route
,
1245 [RDMA_USER_CM_CMD_QUERY_ROUTE
] = ucma_query_route
,
1246 [RDMA_USER_CM_CMD_CONNECT
] = ucma_connect
,
1247 [RDMA_USER_CM_CMD_LISTEN
] = ucma_listen
,
1248 [RDMA_USER_CM_CMD_ACCEPT
] = ucma_accept
,
1249 [RDMA_USER_CM_CMD_REJECT
] = ucma_reject
,
1250 [RDMA_USER_CM_CMD_DISCONNECT
] = ucma_disconnect
,
1251 [RDMA_USER_CM_CMD_INIT_QP_ATTR
] = ucma_init_qp_attr
,
1252 [RDMA_USER_CM_CMD_GET_EVENT
] = ucma_get_event
,
1253 [RDMA_USER_CM_CMD_GET_OPTION
] = NULL
,
1254 [RDMA_USER_CM_CMD_SET_OPTION
] = ucma_set_option
,
1255 [RDMA_USER_CM_CMD_NOTIFY
] = ucma_notify
,
1256 [RDMA_USER_CM_CMD_JOIN_MCAST
] = ucma_join_multicast
,
1257 [RDMA_USER_CM_CMD_LEAVE_MCAST
] = ucma_leave_multicast
,
1258 [RDMA_USER_CM_CMD_MIGRATE_ID
] = ucma_migrate_id
1261 static ssize_t
ucma_write(struct file
*filp
, const char __user
*buf
,
1262 size_t len
, loff_t
*pos
)
1264 struct ucma_file
*file
= filp
->private_data
;
1265 struct rdma_ucm_cmd_hdr hdr
;
1268 if (len
< sizeof(hdr
))
1271 if (copy_from_user(&hdr
, buf
, sizeof(hdr
)))
1274 if (hdr
.cmd
< 0 || hdr
.cmd
>= ARRAY_SIZE(ucma_cmd_table
))
1277 if (hdr
.in
+ sizeof(hdr
) > len
)
1280 if (!ucma_cmd_table
[hdr
.cmd
])
1283 ret
= ucma_cmd_table
[hdr
.cmd
](file
, buf
+ sizeof(hdr
), hdr
.in
, hdr
.out
);
1290 static unsigned int ucma_poll(struct file
*filp
, struct poll_table_struct
*wait
)
1292 struct ucma_file
*file
= filp
->private_data
;
1293 unsigned int mask
= 0;
1295 poll_wait(filp
, &file
->poll_wait
, wait
);
1297 if (!list_empty(&file
->event_list
))
1298 mask
= POLLIN
| POLLRDNORM
;
1304 * ucma_open() does not need the BKL:
1306 * - no global state is referred to;
1307 * - there is no ioctl method to race against;
1308 * - no further module initialization is required for open to work
1309 * after the device is registered.
1311 static int ucma_open(struct inode
*inode
, struct file
*filp
)
1313 struct ucma_file
*file
;
1315 file
= kmalloc(sizeof *file
, GFP_KERNEL
);
1319 INIT_LIST_HEAD(&file
->event_list
);
1320 INIT_LIST_HEAD(&file
->ctx_list
);
1321 init_waitqueue_head(&file
->poll_wait
);
1322 mutex_init(&file
->mut
);
1324 filp
->private_data
= file
;
1327 return nonseekable_open(inode
, filp
);
1330 static int ucma_close(struct inode
*inode
, struct file
*filp
)
1332 struct ucma_file
*file
= filp
->private_data
;
1333 struct ucma_context
*ctx
, *tmp
;
1335 mutex_lock(&file
->mut
);
1336 list_for_each_entry_safe(ctx
, tmp
, &file
->ctx_list
, list
) {
1337 mutex_unlock(&file
->mut
);
1340 idr_remove(&ctx_idr
, ctx
->id
);
1344 mutex_lock(&file
->mut
);
1346 mutex_unlock(&file
->mut
);
1351 static const struct file_operations ucma_fops
= {
1352 .owner
= THIS_MODULE
,
1354 .release
= ucma_close
,
1355 .write
= ucma_write
,
1357 .llseek
= no_llseek
,
1360 static struct miscdevice ucma_misc
= {
1361 .minor
= MISC_DYNAMIC_MINOR
,
1363 .nodename
= "infiniband/rdma_cm",
1368 static ssize_t
show_abi_version(struct device
*dev
,
1369 struct device_attribute
*attr
,
1372 return sprintf(buf
, "%d\n", RDMA_USER_CM_ABI_VERSION
);
1374 static DEVICE_ATTR(abi_version
, S_IRUGO
, show_abi_version
, NULL
);
1376 static int __init
ucma_init(void)
1380 ret
= misc_register(&ucma_misc
);
1384 ret
= device_create_file(ucma_misc
.this_device
, &dev_attr_abi_version
);
1386 printk(KERN_ERR
"rdma_ucm: couldn't create abi_version attr\n");
1390 ucma_ctl_table_hdr
= register_sysctl_paths(ucma_ctl_path
, ucma_ctl_table
);
1391 if (!ucma_ctl_table_hdr
) {
1392 printk(KERN_ERR
"rdma_ucm: couldn't register sysctl paths\n");
1398 device_remove_file(ucma_misc
.this_device
, &dev_attr_abi_version
);
1400 misc_deregister(&ucma_misc
);
1404 static void __exit
ucma_cleanup(void)
1406 unregister_sysctl_table(ucma_ctl_table_hdr
);
1407 device_remove_file(ucma_misc
.this_device
, &dev_attr_abi_version
);
1408 misc_deregister(&ucma_misc
);
1409 idr_destroy(&ctx_idr
);
1412 module_init(ucma_init
);
1413 module_exit(ucma_cleanup
);