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>
45 #include <rdma/rdma_user_cm.h>
46 #include <rdma/ib_marshall.h>
47 #include <rdma/rdma_cm.h>
48 #include <rdma/rdma_cm_ib.h>
50 MODULE_AUTHOR("Sean Hefty");
51 MODULE_DESCRIPTION("RDMA Userspace Connection Manager Access");
52 MODULE_LICENSE("Dual BSD/GPL");
54 static unsigned int max_backlog = 1024;
56 static struct ctl_table_header *ucma_ctl_table_hdr;
57 static ctl_table ucma_ctl_table[] = {
59 .procname = "max_backlog",
61 .maxlen = sizeof max_backlog,
63 .proc_handler = proc_dointvec,
68 static struct ctl_path ucma_ctl_path[] = {
69 { .procname = "net" },
70 { .procname = "rdma_ucm" },
77 struct list_head ctx_list;
78 struct list_head event_list;
79 wait_queue_head_t poll_wait;
84 struct completion comp;
89 struct ucma_file *file;
90 struct rdma_cm_id *cm_id;
93 struct list_head list;
94 struct list_head mc_list;
97 struct ucma_multicast {
98 struct ucma_context *ctx;
103 struct list_head list;
104 struct sockaddr_storage addr;
108 struct ucma_context *ctx;
109 struct ucma_multicast *mc;
110 struct list_head list;
111 struct rdma_cm_id *cm_id;
112 struct rdma_ucm_event_resp resp;
115 static DEFINE_MUTEX(mut);
116 static DEFINE_IDR(ctx_idr);
117 static DEFINE_IDR(multicast_idr);
119 static inline struct ucma_context *_ucma_find_context(int id,
120 struct ucma_file *file)
122 struct ucma_context *ctx;
124 ctx = idr_find(&ctx_idr, id);
126 ctx = ERR_PTR(-ENOENT);
127 else if (ctx->file != file)
128 ctx = ERR_PTR(-EINVAL);
132 static struct ucma_context *ucma_get_ctx(struct ucma_file *file, int id)
134 struct ucma_context *ctx;
137 ctx = _ucma_find_context(id, file);
139 atomic_inc(&ctx->ref);
144 static void ucma_put_ctx(struct ucma_context *ctx)
146 if (atomic_dec_and_test(&ctx->ref))
147 complete(&ctx->comp);
150 static struct ucma_context *ucma_alloc_ctx(struct ucma_file *file)
152 struct ucma_context *ctx;
155 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
159 atomic_set(&ctx->ref, 1);
160 init_completion(&ctx->comp);
161 INIT_LIST_HEAD(&ctx->mc_list);
165 ret = idr_pre_get(&ctx_idr, GFP_KERNEL);
170 ret = idr_get_new(&ctx_idr, ctx, &ctx->id);
172 } while (ret == -EAGAIN);
177 list_add_tail(&ctx->list, &file->ctx_list);
185 static struct ucma_multicast* ucma_alloc_multicast(struct ucma_context *ctx)
187 struct ucma_multicast *mc;
190 mc = kzalloc(sizeof(*mc), GFP_KERNEL);
195 ret = idr_pre_get(&multicast_idr, GFP_KERNEL);
200 ret = idr_get_new(&multicast_idr, mc, &mc->id);
202 } while (ret == -EAGAIN);
208 list_add_tail(&mc->list, &ctx->mc_list);
216 static void ucma_copy_conn_event(struct rdma_ucm_conn_param *dst,
217 struct rdma_conn_param *src)
219 if (src->private_data_len)
220 memcpy(dst->private_data, src->private_data,
221 src->private_data_len);
222 dst->private_data_len = src->private_data_len;
223 dst->responder_resources =src->responder_resources;
224 dst->initiator_depth = src->initiator_depth;
225 dst->flow_control = src->flow_control;
226 dst->retry_count = src->retry_count;
227 dst->rnr_retry_count = src->rnr_retry_count;
229 dst->qp_num = src->qp_num;
232 static void ucma_copy_ud_event(struct rdma_ucm_ud_param *dst,
233 struct rdma_ud_param *src)
235 if (src->private_data_len)
236 memcpy(dst->private_data, src->private_data,
237 src->private_data_len);
238 dst->private_data_len = src->private_data_len;
239 ib_copy_ah_attr_to_user(&dst->ah_attr, &src->ah_attr);
240 dst->qp_num = src->qp_num;
241 dst->qkey = src->qkey;
244 static void ucma_set_event_context(struct ucma_context *ctx,
245 struct rdma_cm_event *event,
246 struct ucma_event *uevent)
249 switch (event->event) {
250 case RDMA_CM_EVENT_MULTICAST_JOIN:
251 case RDMA_CM_EVENT_MULTICAST_ERROR:
252 uevent->mc = (struct ucma_multicast *)
253 event->param.ud.private_data;
254 uevent->resp.uid = uevent->mc->uid;
255 uevent->resp.id = uevent->mc->id;
258 uevent->resp.uid = ctx->uid;
259 uevent->resp.id = ctx->id;
264 static int ucma_event_handler(struct rdma_cm_id *cm_id,
265 struct rdma_cm_event *event)
267 struct ucma_event *uevent;
268 struct ucma_context *ctx = cm_id->context;
271 uevent = kzalloc(sizeof(*uevent), GFP_KERNEL);
273 return event->event == RDMA_CM_EVENT_CONNECT_REQUEST;
275 uevent->cm_id = cm_id;
276 ucma_set_event_context(ctx, event, uevent);
277 uevent->resp.event = event->event;
278 uevent->resp.status = event->status;
279 if (cm_id->ps == RDMA_PS_UDP || cm_id->ps == RDMA_PS_IPOIB)
280 ucma_copy_ud_event(&uevent->resp.param.ud, &event->param.ud);
282 ucma_copy_conn_event(&uevent->resp.param.conn,
285 mutex_lock(&ctx->file->mut);
286 if (event->event == RDMA_CM_EVENT_CONNECT_REQUEST) {
293 } else if (!ctx->uid) {
295 * We ignore events for new connections until userspace has set
296 * their context. This can only happen if an error occurs on a
297 * new connection before the user accepts it. This is okay,
298 * since the accept will just fail later.
304 list_add_tail(&uevent->list, &ctx->file->event_list);
305 wake_up_interruptible(&ctx->file->poll_wait);
307 mutex_unlock(&ctx->file->mut);
311 static ssize_t ucma_get_event(struct ucma_file *file, const char __user *inbuf,
312 int in_len, int out_len)
314 struct ucma_context *ctx;
315 struct rdma_ucm_get_event cmd;
316 struct ucma_event *uevent;
320 if (out_len < sizeof uevent->resp)
323 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
326 mutex_lock(&file->mut);
327 while (list_empty(&file->event_list)) {
328 mutex_unlock(&file->mut);
330 if (file->filp->f_flags & O_NONBLOCK)
333 if (wait_event_interruptible(file->poll_wait,
334 !list_empty(&file->event_list)))
337 mutex_lock(&file->mut);
340 uevent = list_entry(file->event_list.next, struct ucma_event, list);
342 if (uevent->resp.event == RDMA_CM_EVENT_CONNECT_REQUEST) {
343 ctx = ucma_alloc_ctx(file);
348 uevent->ctx->backlog++;
349 ctx->cm_id = uevent->cm_id;
350 ctx->cm_id->context = ctx;
351 uevent->resp.id = ctx->id;
354 if (copy_to_user((void __user *)(unsigned long)cmd.response,
355 &uevent->resp, sizeof uevent->resp)) {
360 list_del(&uevent->list);
361 uevent->ctx->events_reported++;
363 uevent->mc->events_reported++;
366 mutex_unlock(&file->mut);
370 static ssize_t ucma_create_id(struct ucma_file *file,
371 const char __user *inbuf,
372 int in_len, int out_len)
374 struct rdma_ucm_create_id cmd;
375 struct rdma_ucm_create_id_resp resp;
376 struct ucma_context *ctx;
379 if (out_len < sizeof(resp))
382 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
385 mutex_lock(&file->mut);
386 ctx = ucma_alloc_ctx(file);
387 mutex_unlock(&file->mut);
392 ctx->cm_id = rdma_create_id(ucma_event_handler, ctx, cmd.ps);
393 if (IS_ERR(ctx->cm_id)) {
394 ret = PTR_ERR(ctx->cm_id);
399 if (copy_to_user((void __user *)(unsigned long)cmd.response,
400 &resp, sizeof(resp))) {
407 rdma_destroy_id(ctx->cm_id);
410 idr_remove(&ctx_idr, ctx->id);
416 static void ucma_cleanup_multicast(struct ucma_context *ctx)
418 struct ucma_multicast *mc, *tmp;
421 list_for_each_entry_safe(mc, tmp, &ctx->mc_list, list) {
423 idr_remove(&multicast_idr, mc->id);
429 static void ucma_cleanup_events(struct ucma_context *ctx)
431 struct ucma_event *uevent, *tmp;
433 list_for_each_entry_safe(uevent, tmp, &ctx->file->event_list, list) {
434 if (uevent->ctx != ctx)
437 list_del(&uevent->list);
439 /* clear incoming connections. */
440 if (uevent->resp.event == RDMA_CM_EVENT_CONNECT_REQUEST)
441 rdma_destroy_id(uevent->cm_id);
447 static void ucma_cleanup_mc_events(struct ucma_multicast *mc)
449 struct ucma_event *uevent, *tmp;
451 list_for_each_entry_safe(uevent, tmp, &mc->ctx->file->event_list, list) {
452 if (uevent->mc != mc)
455 list_del(&uevent->list);
460 static int ucma_free_ctx(struct ucma_context *ctx)
464 /* No new events will be generated after destroying the id. */
465 rdma_destroy_id(ctx->cm_id);
467 ucma_cleanup_multicast(ctx);
469 /* Cleanup events not yet reported to the user. */
470 mutex_lock(&ctx->file->mut);
471 ucma_cleanup_events(ctx);
472 list_del(&ctx->list);
473 mutex_unlock(&ctx->file->mut);
475 events_reported = ctx->events_reported;
477 return events_reported;
480 static ssize_t ucma_destroy_id(struct ucma_file *file, const char __user *inbuf,
481 int in_len, int out_len)
483 struct rdma_ucm_destroy_id cmd;
484 struct rdma_ucm_destroy_id_resp resp;
485 struct ucma_context *ctx;
488 if (out_len < sizeof(resp))
491 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
495 ctx = _ucma_find_context(cmd.id, file);
497 idr_remove(&ctx_idr, ctx->id);
504 wait_for_completion(&ctx->comp);
505 resp.events_reported = ucma_free_ctx(ctx);
507 if (copy_to_user((void __user *)(unsigned long)cmd.response,
508 &resp, sizeof(resp)))
514 static ssize_t ucma_bind_addr(struct ucma_file *file, const char __user *inbuf,
515 int in_len, int out_len)
517 struct rdma_ucm_bind_addr cmd;
518 struct ucma_context *ctx;
521 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
524 ctx = ucma_get_ctx(file, cmd.id);
528 ret = rdma_bind_addr(ctx->cm_id, (struct sockaddr *) &cmd.addr);
533 static ssize_t ucma_resolve_addr(struct ucma_file *file,
534 const char __user *inbuf,
535 int in_len, int out_len)
537 struct rdma_ucm_resolve_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_resolve_addr(ctx->cm_id, (struct sockaddr *) &cmd.src_addr,
549 (struct sockaddr *) &cmd.dst_addr,
555 static ssize_t ucma_resolve_route(struct ucma_file *file,
556 const char __user *inbuf,
557 int in_len, int out_len)
559 struct rdma_ucm_resolve_route cmd;
560 struct ucma_context *ctx;
563 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
566 ctx = ucma_get_ctx(file, cmd.id);
570 ret = rdma_resolve_route(ctx->cm_id, cmd.timeout_ms);
575 static void ucma_copy_ib_route(struct rdma_ucm_query_route_resp *resp,
576 struct rdma_route *route)
578 struct rdma_dev_addr *dev_addr;
580 resp->num_paths = route->num_paths;
581 switch (route->num_paths) {
583 dev_addr = &route->addr.dev_addr;
584 rdma_addr_get_dgid(dev_addr,
585 (union ib_gid *) &resp->ib_route[0].dgid);
586 rdma_addr_get_sgid(dev_addr,
587 (union ib_gid *) &resp->ib_route[0].sgid);
588 resp->ib_route[0].pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
591 ib_copy_path_rec_to_user(&resp->ib_route[1],
592 &route->path_rec[1]);
595 ib_copy_path_rec_to_user(&resp->ib_route[0],
596 &route->path_rec[0]);
603 static void ucma_copy_iboe_route(struct rdma_ucm_query_route_resp *resp,
604 struct rdma_route *route)
606 struct rdma_dev_addr *dev_addr;
607 struct net_device *dev;
610 resp->num_paths = route->num_paths;
611 switch (route->num_paths) {
613 dev_addr = &route->addr.dev_addr;
614 dev = dev_get_by_index(&init_net, dev_addr->bound_dev_if);
616 vid = rdma_vlan_dev_vlan_id(dev);
620 iboe_mac_vlan_to_ll((union ib_gid *) &resp->ib_route[0].dgid,
621 dev_addr->dst_dev_addr, vid);
622 iboe_addr_get_sgid(dev_addr,
623 (union ib_gid *) &resp->ib_route[0].sgid);
624 resp->ib_route[0].pkey = cpu_to_be16(0xffff);
627 ib_copy_path_rec_to_user(&resp->ib_route[1],
628 &route->path_rec[1]);
631 ib_copy_path_rec_to_user(&resp->ib_route[0],
632 &route->path_rec[0]);
639 static void ucma_copy_iw_route(struct rdma_ucm_query_route_resp *resp,
640 struct rdma_route *route)
642 struct rdma_dev_addr *dev_addr;
644 dev_addr = &route->addr.dev_addr;
645 rdma_addr_get_dgid(dev_addr, (union ib_gid *) &resp->ib_route[0].dgid);
646 rdma_addr_get_sgid(dev_addr, (union ib_gid *) &resp->ib_route[0].sgid);
649 static ssize_t ucma_query_route(struct ucma_file *file,
650 const char __user *inbuf,
651 int in_len, int out_len)
653 struct rdma_ucm_query_route cmd;
654 struct rdma_ucm_query_route_resp resp;
655 struct ucma_context *ctx;
656 struct sockaddr *addr;
659 if (out_len < sizeof(resp))
662 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
665 ctx = ucma_get_ctx(file, cmd.id);
669 memset(&resp, 0, sizeof resp);
670 addr = (struct sockaddr *) &ctx->cm_id->route.addr.src_addr;
671 memcpy(&resp.src_addr, addr, addr->sa_family == AF_INET ?
672 sizeof(struct sockaddr_in) :
673 sizeof(struct sockaddr_in6));
674 addr = (struct sockaddr *) &ctx->cm_id->route.addr.dst_addr;
675 memcpy(&resp.dst_addr, addr, addr->sa_family == AF_INET ?
676 sizeof(struct sockaddr_in) :
677 sizeof(struct sockaddr_in6));
678 if (!ctx->cm_id->device)
681 resp.node_guid = (__force __u64) ctx->cm_id->device->node_guid;
682 resp.port_num = ctx->cm_id->port_num;
683 switch (rdma_node_get_transport(ctx->cm_id->device->node_type)) {
684 case RDMA_TRANSPORT_IB:
685 switch (rdma_port_get_link_layer(ctx->cm_id->device,
686 ctx->cm_id->port_num)) {
687 case IB_LINK_LAYER_INFINIBAND:
688 ucma_copy_ib_route(&resp, &ctx->cm_id->route);
690 case IB_LINK_LAYER_ETHERNET:
691 ucma_copy_iboe_route(&resp, &ctx->cm_id->route);
697 case RDMA_TRANSPORT_IWARP:
698 ucma_copy_iw_route(&resp, &ctx->cm_id->route);
705 if (copy_to_user((void __user *)(unsigned long)cmd.response,
706 &resp, sizeof(resp)))
713 static void ucma_copy_conn_param(struct rdma_conn_param *dst,
714 struct rdma_ucm_conn_param *src)
716 dst->private_data = src->private_data;
717 dst->private_data_len = src->private_data_len;
718 dst->responder_resources =src->responder_resources;
719 dst->initiator_depth = src->initiator_depth;
720 dst->flow_control = src->flow_control;
721 dst->retry_count = src->retry_count;
722 dst->rnr_retry_count = src->rnr_retry_count;
724 dst->qp_num = src->qp_num;
727 static ssize_t ucma_connect(struct ucma_file *file, const char __user *inbuf,
728 int in_len, int out_len)
730 struct rdma_ucm_connect cmd;
731 struct rdma_conn_param conn_param;
732 struct ucma_context *ctx;
735 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
738 if (!cmd.conn_param.valid)
741 ctx = ucma_get_ctx(file, cmd.id);
745 ucma_copy_conn_param(&conn_param, &cmd.conn_param);
746 ret = rdma_connect(ctx->cm_id, &conn_param);
751 static ssize_t ucma_listen(struct ucma_file *file, const char __user *inbuf,
752 int in_len, int out_len)
754 struct rdma_ucm_listen cmd;
755 struct ucma_context *ctx;
758 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
761 ctx = ucma_get_ctx(file, cmd.id);
765 ctx->backlog = cmd.backlog > 0 && cmd.backlog < max_backlog ?
766 cmd.backlog : max_backlog;
767 ret = rdma_listen(ctx->cm_id, ctx->backlog);
772 static ssize_t ucma_accept(struct ucma_file *file, const char __user *inbuf,
773 int in_len, int out_len)
775 struct rdma_ucm_accept cmd;
776 struct rdma_conn_param conn_param;
777 struct ucma_context *ctx;
780 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
783 ctx = ucma_get_ctx(file, cmd.id);
787 if (cmd.conn_param.valid) {
789 ucma_copy_conn_param(&conn_param, &cmd.conn_param);
790 ret = rdma_accept(ctx->cm_id, &conn_param);
792 ret = rdma_accept(ctx->cm_id, NULL);
798 static ssize_t ucma_reject(struct ucma_file *file, const char __user *inbuf,
799 int in_len, int out_len)
801 struct rdma_ucm_reject cmd;
802 struct ucma_context *ctx;
805 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
808 ctx = ucma_get_ctx(file, cmd.id);
812 ret = rdma_reject(ctx->cm_id, cmd.private_data, cmd.private_data_len);
817 static ssize_t ucma_disconnect(struct ucma_file *file, const char __user *inbuf,
818 int in_len, int out_len)
820 struct rdma_ucm_disconnect cmd;
821 struct ucma_context *ctx;
824 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
827 ctx = ucma_get_ctx(file, cmd.id);
831 ret = rdma_disconnect(ctx->cm_id);
836 static ssize_t ucma_init_qp_attr(struct ucma_file *file,
837 const char __user *inbuf,
838 int in_len, int out_len)
840 struct rdma_ucm_init_qp_attr cmd;
841 struct ib_uverbs_qp_attr resp;
842 struct ucma_context *ctx;
843 struct ib_qp_attr qp_attr;
846 if (out_len < sizeof(resp))
849 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
852 ctx = ucma_get_ctx(file, cmd.id);
856 resp.qp_attr_mask = 0;
857 memset(&qp_attr, 0, sizeof qp_attr);
858 qp_attr.qp_state = cmd.qp_state;
859 ret = rdma_init_qp_attr(ctx->cm_id, &qp_attr, &resp.qp_attr_mask);
863 ib_copy_qp_attr_to_user(&resp, &qp_attr);
864 if (copy_to_user((void __user *)(unsigned long)cmd.response,
865 &resp, sizeof(resp)))
873 static int ucma_set_option_id(struct ucma_context *ctx, int optname,
874 void *optval, size_t optlen)
879 case RDMA_OPTION_ID_TOS:
880 if (optlen != sizeof(u8)) {
884 rdma_set_service_type(ctx->cm_id, *((u8 *) optval));
893 static int ucma_set_ib_path(struct ucma_context *ctx,
894 struct ib_path_rec_data *path_data, size_t optlen)
896 struct ib_sa_path_rec sa_path;
897 struct rdma_cm_event event;
900 if (optlen % sizeof(*path_data))
903 for (; optlen; optlen -= sizeof(*path_data), path_data++) {
904 if (path_data->flags == (IB_PATH_GMP | IB_PATH_PRIMARY |
905 IB_PATH_BIDIRECTIONAL))
912 ib_sa_unpack_path(path_data->path_rec, &sa_path);
913 ret = rdma_set_ib_paths(ctx->cm_id, &sa_path, 1);
917 memset(&event, 0, sizeof event);
918 event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
919 return ucma_event_handler(ctx->cm_id, &event);
922 static int ucma_set_option_ib(struct ucma_context *ctx, int optname,
923 void *optval, size_t optlen)
928 case RDMA_OPTION_IB_PATH:
929 ret = ucma_set_ib_path(ctx, optval, optlen);
938 static int ucma_set_option_level(struct ucma_context *ctx, int level,
939 int optname, void *optval, size_t optlen)
945 ret = ucma_set_option_id(ctx, optname, optval, optlen);
948 ret = ucma_set_option_ib(ctx, optname, optval, optlen);
957 static ssize_t ucma_set_option(struct ucma_file *file, const char __user *inbuf,
958 int in_len, int out_len)
960 struct rdma_ucm_set_option cmd;
961 struct ucma_context *ctx;
965 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
968 ctx = ucma_get_ctx(file, cmd.id);
972 optval = kmalloc(cmd.optlen, GFP_KERNEL);
978 if (copy_from_user(optval, (void __user *) (unsigned long) cmd.optval,
984 ret = ucma_set_option_level(ctx, cmd.level, cmd.optname, optval,
993 static ssize_t ucma_notify(struct ucma_file *file, const char __user *inbuf,
994 int in_len, int out_len)
996 struct rdma_ucm_notify cmd;
997 struct ucma_context *ctx;
1000 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1003 ctx = ucma_get_ctx(file, cmd.id);
1005 return PTR_ERR(ctx);
1007 ret = rdma_notify(ctx->cm_id, (enum ib_event_type) cmd.event);
1012 static ssize_t ucma_join_multicast(struct ucma_file *file,
1013 const char __user *inbuf,
1014 int in_len, int out_len)
1016 struct rdma_ucm_join_mcast cmd;
1017 struct rdma_ucm_create_id_resp resp;
1018 struct ucma_context *ctx;
1019 struct ucma_multicast *mc;
1022 if (out_len < sizeof(resp))
1025 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1028 ctx = ucma_get_ctx(file, cmd.id);
1030 return PTR_ERR(ctx);
1032 mutex_lock(&file->mut);
1033 mc = ucma_alloc_multicast(ctx);
1040 memcpy(&mc->addr, &cmd.addr, sizeof cmd.addr);
1041 ret = rdma_join_multicast(ctx->cm_id, (struct sockaddr *) &mc->addr, mc);
1046 if (copy_to_user((void __user *)(unsigned long)cmd.response,
1047 &resp, sizeof(resp))) {
1052 mutex_unlock(&file->mut);
1057 rdma_leave_multicast(ctx->cm_id, (struct sockaddr *) &mc->addr);
1058 ucma_cleanup_mc_events(mc);
1061 idr_remove(&multicast_idr, mc->id);
1063 list_del(&mc->list);
1066 mutex_unlock(&file->mut);
1071 static ssize_t ucma_leave_multicast(struct ucma_file *file,
1072 const char __user *inbuf,
1073 int in_len, int out_len)
1075 struct rdma_ucm_destroy_id cmd;
1076 struct rdma_ucm_destroy_id_resp resp;
1077 struct ucma_multicast *mc;
1080 if (out_len < sizeof(resp))
1083 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1087 mc = idr_find(&multicast_idr, cmd.id);
1089 mc = ERR_PTR(-ENOENT);
1090 else if (mc->ctx->file != file)
1091 mc = ERR_PTR(-EINVAL);
1093 idr_remove(&multicast_idr, mc->id);
1094 atomic_inc(&mc->ctx->ref);
1103 rdma_leave_multicast(mc->ctx->cm_id, (struct sockaddr *) &mc->addr);
1104 mutex_lock(&mc->ctx->file->mut);
1105 ucma_cleanup_mc_events(mc);
1106 list_del(&mc->list);
1107 mutex_unlock(&mc->ctx->file->mut);
1109 ucma_put_ctx(mc->ctx);
1110 resp.events_reported = mc->events_reported;
1113 if (copy_to_user((void __user *)(unsigned long)cmd.response,
1114 &resp, sizeof(resp)))
1120 static void ucma_lock_files(struct ucma_file *file1, struct ucma_file *file2)
1122 /* Acquire mutex's based on pointer comparison to prevent deadlock. */
1123 if (file1 < file2) {
1124 mutex_lock(&file1->mut);
1125 mutex_lock(&file2->mut);
1127 mutex_lock(&file2->mut);
1128 mutex_lock(&file1->mut);
1132 static void ucma_unlock_files(struct ucma_file *file1, struct ucma_file *file2)
1134 if (file1 < file2) {
1135 mutex_unlock(&file2->mut);
1136 mutex_unlock(&file1->mut);
1138 mutex_unlock(&file1->mut);
1139 mutex_unlock(&file2->mut);
1143 static void ucma_move_events(struct ucma_context *ctx, struct ucma_file *file)
1145 struct ucma_event *uevent, *tmp;
1147 list_for_each_entry_safe(uevent, tmp, &ctx->file->event_list, list)
1148 if (uevent->ctx == ctx)
1149 list_move_tail(&uevent->list, &file->event_list);
1152 static ssize_t ucma_migrate_id(struct ucma_file *new_file,
1153 const char __user *inbuf,
1154 int in_len, int out_len)
1156 struct rdma_ucm_migrate_id cmd;
1157 struct rdma_ucm_migrate_resp resp;
1158 struct ucma_context *ctx;
1160 struct ucma_file *cur_file;
1163 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1166 /* Get current fd to protect against it being closed */
1167 filp = fget(cmd.fd);
1171 /* Validate current fd and prevent destruction of id. */
1172 ctx = ucma_get_ctx(filp->private_data, cmd.id);
1178 cur_file = ctx->file;
1179 if (cur_file == new_file) {
1180 resp.events_reported = ctx->events_reported;
1185 * Migrate events between fd's, maintaining order, and avoiding new
1186 * events being added before existing events.
1188 ucma_lock_files(cur_file, new_file);
1191 list_move_tail(&ctx->list, &new_file->ctx_list);
1192 ucma_move_events(ctx, new_file);
1193 ctx->file = new_file;
1194 resp.events_reported = ctx->events_reported;
1197 ucma_unlock_files(cur_file, new_file);
1200 if (copy_to_user((void __user *)(unsigned long)cmd.response,
1201 &resp, sizeof(resp)))
1210 static ssize_t (*ucma_cmd_table[])(struct ucma_file *file,
1211 const char __user *inbuf,
1212 int in_len, int out_len) = {
1213 [RDMA_USER_CM_CMD_CREATE_ID] = ucma_create_id,
1214 [RDMA_USER_CM_CMD_DESTROY_ID] = ucma_destroy_id,
1215 [RDMA_USER_CM_CMD_BIND_ADDR] = ucma_bind_addr,
1216 [RDMA_USER_CM_CMD_RESOLVE_ADDR] = ucma_resolve_addr,
1217 [RDMA_USER_CM_CMD_RESOLVE_ROUTE]= ucma_resolve_route,
1218 [RDMA_USER_CM_CMD_QUERY_ROUTE] = ucma_query_route,
1219 [RDMA_USER_CM_CMD_CONNECT] = ucma_connect,
1220 [RDMA_USER_CM_CMD_LISTEN] = ucma_listen,
1221 [RDMA_USER_CM_CMD_ACCEPT] = ucma_accept,
1222 [RDMA_USER_CM_CMD_REJECT] = ucma_reject,
1223 [RDMA_USER_CM_CMD_DISCONNECT] = ucma_disconnect,
1224 [RDMA_USER_CM_CMD_INIT_QP_ATTR] = ucma_init_qp_attr,
1225 [RDMA_USER_CM_CMD_GET_EVENT] = ucma_get_event,
1226 [RDMA_USER_CM_CMD_GET_OPTION] = NULL,
1227 [RDMA_USER_CM_CMD_SET_OPTION] = ucma_set_option,
1228 [RDMA_USER_CM_CMD_NOTIFY] = ucma_notify,
1229 [RDMA_USER_CM_CMD_JOIN_MCAST] = ucma_join_multicast,
1230 [RDMA_USER_CM_CMD_LEAVE_MCAST] = ucma_leave_multicast,
1231 [RDMA_USER_CM_CMD_MIGRATE_ID] = ucma_migrate_id
1234 static ssize_t ucma_write(struct file *filp, const char __user *buf,
1235 size_t len, loff_t *pos)
1237 struct ucma_file *file = filp->private_data;
1238 struct rdma_ucm_cmd_hdr hdr;
1241 if (len < sizeof(hdr))
1244 if (copy_from_user(&hdr, buf, sizeof(hdr)))
1247 if (hdr.cmd < 0 || hdr.cmd >= ARRAY_SIZE(ucma_cmd_table))
1250 if (hdr.in + sizeof(hdr) > len)
1253 if (!ucma_cmd_table[hdr.cmd])
1256 ret = ucma_cmd_table[hdr.cmd](file, buf + sizeof(hdr), hdr.in, hdr.out);
1263 static unsigned int ucma_poll(struct file *filp, struct poll_table_struct *wait)
1265 struct ucma_file *file = filp->private_data;
1266 unsigned int mask = 0;
1268 poll_wait(filp, &file->poll_wait, wait);
1270 if (!list_empty(&file->event_list))
1271 mask = POLLIN | POLLRDNORM;
1277 * ucma_open() does not need the BKL:
1279 * - no global state is referred to;
1280 * - there is no ioctl method to race against;
1281 * - no further module initialization is required for open to work
1282 * after the device is registered.
1284 static int ucma_open(struct inode *inode, struct file *filp)
1286 struct ucma_file *file;
1288 file = kmalloc(sizeof *file, GFP_KERNEL);
1292 INIT_LIST_HEAD(&file->event_list);
1293 INIT_LIST_HEAD(&file->ctx_list);
1294 init_waitqueue_head(&file->poll_wait);
1295 mutex_init(&file->mut);
1297 filp->private_data = file;
1300 return nonseekable_open(inode, filp);
1303 static int ucma_close(struct inode *inode, struct file *filp)
1305 struct ucma_file *file = filp->private_data;
1306 struct ucma_context *ctx, *tmp;
1308 mutex_lock(&file->mut);
1309 list_for_each_entry_safe(ctx, tmp, &file->ctx_list, list) {
1310 mutex_unlock(&file->mut);
1313 idr_remove(&ctx_idr, ctx->id);
1317 mutex_lock(&file->mut);
1319 mutex_unlock(&file->mut);
1324 static const struct file_operations ucma_fops = {
1325 .owner = THIS_MODULE,
1327 .release = ucma_close,
1328 .write = ucma_write,
1330 .llseek = no_llseek,
1333 static struct miscdevice ucma_misc = {
1334 .minor = MISC_DYNAMIC_MINOR,
1339 static ssize_t show_abi_version(struct device *dev,
1340 struct device_attribute *attr,
1343 return sprintf(buf, "%d\n", RDMA_USER_CM_ABI_VERSION);
1345 static DEVICE_ATTR(abi_version, S_IRUGO, show_abi_version, NULL);
1347 static int __init ucma_init(void)
1351 ret = misc_register(&ucma_misc);
1355 ret = device_create_file(ucma_misc.this_device, &dev_attr_abi_version);
1357 printk(KERN_ERR "rdma_ucm: couldn't create abi_version attr\n");
1361 ucma_ctl_table_hdr = register_sysctl_paths(ucma_ctl_path, ucma_ctl_table);
1362 if (!ucma_ctl_table_hdr) {
1363 printk(KERN_ERR "rdma_ucm: couldn't register sysctl paths\n");
1369 device_remove_file(ucma_misc.this_device, &dev_attr_abi_version);
1371 misc_deregister(&ucma_misc);
1375 static void __exit ucma_cleanup(void)
1377 unregister_sysctl_table(ucma_ctl_table_hdr);
1378 device_remove_file(ucma_misc.this_device, &dev_attr_abi_version);
1379 misc_deregister(&ucma_misc);
1380 idr_destroy(&ctx_idr);
1383 module_init(ucma_init);
1384 module_exit(ucma_cleanup);