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->qp_type == IB_QPT_UD)
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;
382 *qp_type = cmd->qp_type;
389 static ssize_t ucma_create_id(struct ucma_file *file, const char __user *inbuf,
390 int in_len, int out_len)
392 struct rdma_ucm_create_id cmd;
393 struct rdma_ucm_create_id_resp resp;
394 struct ucma_context *ctx;
395 enum ib_qp_type qp_type;
398 if (out_len < sizeof(resp))
401 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
404 ret = ucma_get_qp_type(&cmd, &qp_type);
408 mutex_lock(&file->mut);
409 ctx = ucma_alloc_ctx(file);
410 mutex_unlock(&file->mut);
415 ctx->cm_id = rdma_create_id(ucma_event_handler, ctx, cmd.ps, qp_type);
416 if (IS_ERR(ctx->cm_id)) {
417 ret = PTR_ERR(ctx->cm_id);
422 if (copy_to_user((void __user *)(unsigned long)cmd.response,
423 &resp, sizeof(resp))) {
430 rdma_destroy_id(ctx->cm_id);
433 idr_remove(&ctx_idr, ctx->id);
439 static void ucma_cleanup_multicast(struct ucma_context *ctx)
441 struct ucma_multicast *mc, *tmp;
444 list_for_each_entry_safe(mc, tmp, &ctx->mc_list, list) {
446 idr_remove(&multicast_idr, mc->id);
452 static void ucma_cleanup_events(struct ucma_context *ctx)
454 struct ucma_event *uevent, *tmp;
456 list_for_each_entry_safe(uevent, tmp, &ctx->file->event_list, list) {
457 if (uevent->ctx != ctx)
460 list_del(&uevent->list);
462 /* clear incoming connections. */
463 if (uevent->resp.event == RDMA_CM_EVENT_CONNECT_REQUEST)
464 rdma_destroy_id(uevent->cm_id);
470 static void ucma_cleanup_mc_events(struct ucma_multicast *mc)
472 struct ucma_event *uevent, *tmp;
474 list_for_each_entry_safe(uevent, tmp, &mc->ctx->file->event_list, list) {
475 if (uevent->mc != mc)
478 list_del(&uevent->list);
483 static int ucma_free_ctx(struct ucma_context *ctx)
487 /* No new events will be generated after destroying the id. */
488 rdma_destroy_id(ctx->cm_id);
490 ucma_cleanup_multicast(ctx);
492 /* Cleanup events not yet reported to the user. */
493 mutex_lock(&ctx->file->mut);
494 ucma_cleanup_events(ctx);
495 list_del(&ctx->list);
496 mutex_unlock(&ctx->file->mut);
498 events_reported = ctx->events_reported;
500 return events_reported;
503 static ssize_t ucma_destroy_id(struct ucma_file *file, const char __user *inbuf,
504 int in_len, int out_len)
506 struct rdma_ucm_destroy_id cmd;
507 struct rdma_ucm_destroy_id_resp resp;
508 struct ucma_context *ctx;
511 if (out_len < sizeof(resp))
514 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
518 ctx = _ucma_find_context(cmd.id, file);
520 idr_remove(&ctx_idr, ctx->id);
527 wait_for_completion(&ctx->comp);
528 resp.events_reported = ucma_free_ctx(ctx);
530 if (copy_to_user((void __user *)(unsigned long)cmd.response,
531 &resp, sizeof(resp)))
537 static ssize_t ucma_bind_addr(struct ucma_file *file, const char __user *inbuf,
538 int in_len, int out_len)
540 struct rdma_ucm_bind_addr cmd;
541 struct ucma_context *ctx;
544 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
547 ctx = ucma_get_ctx(file, cmd.id);
551 ret = rdma_bind_addr(ctx->cm_id, (struct sockaddr *) &cmd.addr);
556 static ssize_t ucma_resolve_addr(struct ucma_file *file,
557 const char __user *inbuf,
558 int in_len, int out_len)
560 struct rdma_ucm_resolve_addr cmd;
561 struct ucma_context *ctx;
564 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
567 ctx = ucma_get_ctx(file, cmd.id);
571 ret = rdma_resolve_addr(ctx->cm_id, (struct sockaddr *) &cmd.src_addr,
572 (struct sockaddr *) &cmd.dst_addr,
578 static ssize_t ucma_resolve_route(struct ucma_file *file,
579 const char __user *inbuf,
580 int in_len, int out_len)
582 struct rdma_ucm_resolve_route cmd;
583 struct ucma_context *ctx;
586 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
589 ctx = ucma_get_ctx(file, cmd.id);
593 ret = rdma_resolve_route(ctx->cm_id, cmd.timeout_ms);
598 static void ucma_copy_ib_route(struct rdma_ucm_query_route_resp *resp,
599 struct rdma_route *route)
601 struct rdma_dev_addr *dev_addr;
603 resp->num_paths = route->num_paths;
604 switch (route->num_paths) {
606 dev_addr = &route->addr.dev_addr;
607 rdma_addr_get_dgid(dev_addr,
608 (union ib_gid *) &resp->ib_route[0].dgid);
609 rdma_addr_get_sgid(dev_addr,
610 (union ib_gid *) &resp->ib_route[0].sgid);
611 resp->ib_route[0].pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
614 ib_copy_path_rec_to_user(&resp->ib_route[1],
615 &route->path_rec[1]);
618 ib_copy_path_rec_to_user(&resp->ib_route[0],
619 &route->path_rec[0]);
626 static void ucma_copy_iboe_route(struct rdma_ucm_query_route_resp *resp,
627 struct rdma_route *route)
629 struct rdma_dev_addr *dev_addr;
630 struct net_device *dev;
633 resp->num_paths = route->num_paths;
634 switch (route->num_paths) {
636 dev_addr = &route->addr.dev_addr;
637 dev = dev_get_by_index(&init_net, dev_addr->bound_dev_if);
639 vid = rdma_vlan_dev_vlan_id(dev);
643 iboe_mac_vlan_to_ll((union ib_gid *) &resp->ib_route[0].dgid,
644 dev_addr->dst_dev_addr, vid);
645 iboe_addr_get_sgid(dev_addr,
646 (union ib_gid *) &resp->ib_route[0].sgid);
647 resp->ib_route[0].pkey = cpu_to_be16(0xffff);
650 ib_copy_path_rec_to_user(&resp->ib_route[1],
651 &route->path_rec[1]);
654 ib_copy_path_rec_to_user(&resp->ib_route[0],
655 &route->path_rec[0]);
662 static void ucma_copy_iw_route(struct rdma_ucm_query_route_resp *resp,
663 struct rdma_route *route)
665 struct rdma_dev_addr *dev_addr;
667 dev_addr = &route->addr.dev_addr;
668 rdma_addr_get_dgid(dev_addr, (union ib_gid *) &resp->ib_route[0].dgid);
669 rdma_addr_get_sgid(dev_addr, (union ib_gid *) &resp->ib_route[0].sgid);
672 static ssize_t ucma_query_route(struct ucma_file *file,
673 const char __user *inbuf,
674 int in_len, int out_len)
676 struct rdma_ucm_query_route cmd;
677 struct rdma_ucm_query_route_resp resp;
678 struct ucma_context *ctx;
679 struct sockaddr *addr;
682 if (out_len < sizeof(resp))
685 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
688 ctx = ucma_get_ctx(file, cmd.id);
692 memset(&resp, 0, sizeof resp);
693 addr = (struct sockaddr *) &ctx->cm_id->route.addr.src_addr;
694 memcpy(&resp.src_addr, addr, addr->sa_family == AF_INET ?
695 sizeof(struct sockaddr_in) :
696 sizeof(struct sockaddr_in6));
697 addr = (struct sockaddr *) &ctx->cm_id->route.addr.dst_addr;
698 memcpy(&resp.dst_addr, addr, addr->sa_family == AF_INET ?
699 sizeof(struct sockaddr_in) :
700 sizeof(struct sockaddr_in6));
701 if (!ctx->cm_id->device)
704 resp.node_guid = (__force __u64) ctx->cm_id->device->node_guid;
705 resp.port_num = ctx->cm_id->port_num;
706 switch (rdma_node_get_transport(ctx->cm_id->device->node_type)) {
707 case RDMA_TRANSPORT_IB:
708 switch (rdma_port_get_link_layer(ctx->cm_id->device,
709 ctx->cm_id->port_num)) {
710 case IB_LINK_LAYER_INFINIBAND:
711 ucma_copy_ib_route(&resp, &ctx->cm_id->route);
713 case IB_LINK_LAYER_ETHERNET:
714 ucma_copy_iboe_route(&resp, &ctx->cm_id->route);
720 case RDMA_TRANSPORT_IWARP:
721 ucma_copy_iw_route(&resp, &ctx->cm_id->route);
728 if (copy_to_user((void __user *)(unsigned long)cmd.response,
729 &resp, sizeof(resp)))
736 static void ucma_copy_conn_param(struct rdma_conn_param *dst,
737 struct rdma_ucm_conn_param *src)
739 dst->private_data = src->private_data;
740 dst->private_data_len = src->private_data_len;
741 dst->responder_resources =src->responder_resources;
742 dst->initiator_depth = src->initiator_depth;
743 dst->flow_control = src->flow_control;
744 dst->retry_count = src->retry_count;
745 dst->rnr_retry_count = src->rnr_retry_count;
747 dst->qp_num = src->qp_num;
750 static ssize_t ucma_connect(struct ucma_file *file, const char __user *inbuf,
751 int in_len, int out_len)
753 struct rdma_ucm_connect cmd;
754 struct rdma_conn_param conn_param;
755 struct ucma_context *ctx;
758 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
761 if (!cmd.conn_param.valid)
764 ctx = ucma_get_ctx(file, cmd.id);
768 ucma_copy_conn_param(&conn_param, &cmd.conn_param);
769 ret = rdma_connect(ctx->cm_id, &conn_param);
774 static ssize_t ucma_listen(struct ucma_file *file, const char __user *inbuf,
775 int in_len, int out_len)
777 struct rdma_ucm_listen cmd;
778 struct ucma_context *ctx;
781 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
784 ctx = ucma_get_ctx(file, cmd.id);
788 ctx->backlog = cmd.backlog > 0 && cmd.backlog < max_backlog ?
789 cmd.backlog : max_backlog;
790 ret = rdma_listen(ctx->cm_id, ctx->backlog);
795 static ssize_t ucma_accept(struct ucma_file *file, const char __user *inbuf,
796 int in_len, int out_len)
798 struct rdma_ucm_accept cmd;
799 struct rdma_conn_param conn_param;
800 struct ucma_context *ctx;
803 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
806 ctx = ucma_get_ctx(file, cmd.id);
810 if (cmd.conn_param.valid) {
811 ucma_copy_conn_param(&conn_param, &cmd.conn_param);
812 mutex_lock(&file->mut);
813 ret = rdma_accept(ctx->cm_id, &conn_param);
816 mutex_unlock(&file->mut);
818 ret = rdma_accept(ctx->cm_id, NULL);
824 static ssize_t ucma_reject(struct ucma_file *file, const char __user *inbuf,
825 int in_len, int out_len)
827 struct rdma_ucm_reject cmd;
828 struct ucma_context *ctx;
831 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
834 ctx = ucma_get_ctx(file, cmd.id);
838 ret = rdma_reject(ctx->cm_id, cmd.private_data, cmd.private_data_len);
843 static ssize_t ucma_disconnect(struct ucma_file *file, const char __user *inbuf,
844 int in_len, int out_len)
846 struct rdma_ucm_disconnect cmd;
847 struct ucma_context *ctx;
850 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
853 ctx = ucma_get_ctx(file, cmd.id);
857 ret = rdma_disconnect(ctx->cm_id);
862 static ssize_t ucma_init_qp_attr(struct ucma_file *file,
863 const char __user *inbuf,
864 int in_len, int out_len)
866 struct rdma_ucm_init_qp_attr cmd;
867 struct ib_uverbs_qp_attr resp;
868 struct ucma_context *ctx;
869 struct ib_qp_attr qp_attr;
872 if (out_len < sizeof(resp))
875 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
878 ctx = ucma_get_ctx(file, cmd.id);
882 resp.qp_attr_mask = 0;
883 memset(&qp_attr, 0, sizeof qp_attr);
884 qp_attr.qp_state = cmd.qp_state;
885 ret = rdma_init_qp_attr(ctx->cm_id, &qp_attr, &resp.qp_attr_mask);
889 ib_copy_qp_attr_to_user(&resp, &qp_attr);
890 if (copy_to_user((void __user *)(unsigned long)cmd.response,
891 &resp, sizeof(resp)))
899 static int ucma_set_option_id(struct ucma_context *ctx, int optname,
900 void *optval, size_t optlen)
905 case RDMA_OPTION_ID_TOS:
906 if (optlen != sizeof(u8)) {
910 rdma_set_service_type(ctx->cm_id, *((u8 *) optval));
912 case RDMA_OPTION_ID_REUSEADDR:
913 if (optlen != sizeof(int)) {
917 ret = rdma_set_reuseaddr(ctx->cm_id, *((int *) optval) ? 1 : 0);
926 static int ucma_set_ib_path(struct ucma_context *ctx,
927 struct ib_path_rec_data *path_data, size_t optlen)
929 struct ib_sa_path_rec sa_path;
930 struct rdma_cm_event event;
933 if (optlen % sizeof(*path_data))
936 for (; optlen; optlen -= sizeof(*path_data), path_data++) {
937 if (path_data->flags == (IB_PATH_GMP | IB_PATH_PRIMARY |
938 IB_PATH_BIDIRECTIONAL))
945 ib_sa_unpack_path(path_data->path_rec, &sa_path);
946 ret = rdma_set_ib_paths(ctx->cm_id, &sa_path, 1);
950 memset(&event, 0, sizeof event);
951 event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
952 return ucma_event_handler(ctx->cm_id, &event);
955 static int ucma_set_option_ib(struct ucma_context *ctx, int optname,
956 void *optval, size_t optlen)
961 case RDMA_OPTION_IB_PATH:
962 ret = ucma_set_ib_path(ctx, optval, optlen);
971 static int ucma_set_option_level(struct ucma_context *ctx, int level,
972 int optname, void *optval, size_t optlen)
978 ret = ucma_set_option_id(ctx, optname, optval, optlen);
981 ret = ucma_set_option_ib(ctx, optname, optval, optlen);
990 static ssize_t ucma_set_option(struct ucma_file *file, const char __user *inbuf,
991 int in_len, int out_len)
993 struct rdma_ucm_set_option cmd;
994 struct ucma_context *ctx;
998 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1001 ctx = ucma_get_ctx(file, cmd.id);
1003 return PTR_ERR(ctx);
1005 optval = kmalloc(cmd.optlen, GFP_KERNEL);
1011 if (copy_from_user(optval, (void __user *) (unsigned long) cmd.optval,
1017 ret = ucma_set_option_level(ctx, cmd.level, cmd.optname, optval,
1026 static ssize_t ucma_notify(struct ucma_file *file, const char __user *inbuf,
1027 int in_len, int out_len)
1029 struct rdma_ucm_notify cmd;
1030 struct ucma_context *ctx;
1033 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1036 ctx = ucma_get_ctx(file, cmd.id);
1038 return PTR_ERR(ctx);
1040 ret = rdma_notify(ctx->cm_id, (enum ib_event_type) cmd.event);
1045 static ssize_t ucma_join_multicast(struct ucma_file *file,
1046 const char __user *inbuf,
1047 int in_len, int out_len)
1049 struct rdma_ucm_join_mcast cmd;
1050 struct rdma_ucm_create_id_resp resp;
1051 struct ucma_context *ctx;
1052 struct ucma_multicast *mc;
1055 if (out_len < sizeof(resp))
1058 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1061 ctx = ucma_get_ctx(file, cmd.id);
1063 return PTR_ERR(ctx);
1065 mutex_lock(&file->mut);
1066 mc = ucma_alloc_multicast(ctx);
1073 memcpy(&mc->addr, &cmd.addr, sizeof cmd.addr);
1074 ret = rdma_join_multicast(ctx->cm_id, (struct sockaddr *) &mc->addr, mc);
1079 if (copy_to_user((void __user *)(unsigned long)cmd.response,
1080 &resp, sizeof(resp))) {
1085 mutex_unlock(&file->mut);
1090 rdma_leave_multicast(ctx->cm_id, (struct sockaddr *) &mc->addr);
1091 ucma_cleanup_mc_events(mc);
1094 idr_remove(&multicast_idr, mc->id);
1096 list_del(&mc->list);
1099 mutex_unlock(&file->mut);
1104 static ssize_t ucma_leave_multicast(struct ucma_file *file,
1105 const char __user *inbuf,
1106 int in_len, int out_len)
1108 struct rdma_ucm_destroy_id cmd;
1109 struct rdma_ucm_destroy_id_resp resp;
1110 struct ucma_multicast *mc;
1113 if (out_len < sizeof(resp))
1116 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1120 mc = idr_find(&multicast_idr, cmd.id);
1122 mc = ERR_PTR(-ENOENT);
1123 else if (mc->ctx->file != file)
1124 mc = ERR_PTR(-EINVAL);
1126 idr_remove(&multicast_idr, mc->id);
1127 atomic_inc(&mc->ctx->ref);
1136 rdma_leave_multicast(mc->ctx->cm_id, (struct sockaddr *) &mc->addr);
1137 mutex_lock(&mc->ctx->file->mut);
1138 ucma_cleanup_mc_events(mc);
1139 list_del(&mc->list);
1140 mutex_unlock(&mc->ctx->file->mut);
1142 ucma_put_ctx(mc->ctx);
1143 resp.events_reported = mc->events_reported;
1146 if (copy_to_user((void __user *)(unsigned long)cmd.response,
1147 &resp, sizeof(resp)))
1153 static void ucma_lock_files(struct ucma_file *file1, struct ucma_file *file2)
1155 /* Acquire mutex's based on pointer comparison to prevent deadlock. */
1156 if (file1 < file2) {
1157 mutex_lock(&file1->mut);
1158 mutex_lock(&file2->mut);
1160 mutex_lock(&file2->mut);
1161 mutex_lock(&file1->mut);
1165 static void ucma_unlock_files(struct ucma_file *file1, struct ucma_file *file2)
1167 if (file1 < file2) {
1168 mutex_unlock(&file2->mut);
1169 mutex_unlock(&file1->mut);
1171 mutex_unlock(&file1->mut);
1172 mutex_unlock(&file2->mut);
1176 static void ucma_move_events(struct ucma_context *ctx, struct ucma_file *file)
1178 struct ucma_event *uevent, *tmp;
1180 list_for_each_entry_safe(uevent, tmp, &ctx->file->event_list, list)
1181 if (uevent->ctx == ctx)
1182 list_move_tail(&uevent->list, &file->event_list);
1185 static ssize_t ucma_migrate_id(struct ucma_file *new_file,
1186 const char __user *inbuf,
1187 int in_len, int out_len)
1189 struct rdma_ucm_migrate_id cmd;
1190 struct rdma_ucm_migrate_resp resp;
1191 struct ucma_context *ctx;
1193 struct ucma_file *cur_file;
1196 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1199 /* Get current fd to protect against it being closed */
1200 filp = fget(cmd.fd);
1204 /* Validate current fd and prevent destruction of id. */
1205 ctx = ucma_get_ctx(filp->private_data, cmd.id);
1211 cur_file = ctx->file;
1212 if (cur_file == new_file) {
1213 resp.events_reported = ctx->events_reported;
1218 * Migrate events between fd's, maintaining order, and avoiding new
1219 * events being added before existing events.
1221 ucma_lock_files(cur_file, new_file);
1224 list_move_tail(&ctx->list, &new_file->ctx_list);
1225 ucma_move_events(ctx, new_file);
1226 ctx->file = new_file;
1227 resp.events_reported = ctx->events_reported;
1230 ucma_unlock_files(cur_file, new_file);
1233 if (copy_to_user((void __user *)(unsigned long)cmd.response,
1234 &resp, sizeof(resp)))
1243 static ssize_t (*ucma_cmd_table[])(struct ucma_file *file,
1244 const char __user *inbuf,
1245 int in_len, int out_len) = {
1246 [RDMA_USER_CM_CMD_CREATE_ID] = ucma_create_id,
1247 [RDMA_USER_CM_CMD_DESTROY_ID] = ucma_destroy_id,
1248 [RDMA_USER_CM_CMD_BIND_ADDR] = ucma_bind_addr,
1249 [RDMA_USER_CM_CMD_RESOLVE_ADDR] = ucma_resolve_addr,
1250 [RDMA_USER_CM_CMD_RESOLVE_ROUTE]= ucma_resolve_route,
1251 [RDMA_USER_CM_CMD_QUERY_ROUTE] = ucma_query_route,
1252 [RDMA_USER_CM_CMD_CONNECT] = ucma_connect,
1253 [RDMA_USER_CM_CMD_LISTEN] = ucma_listen,
1254 [RDMA_USER_CM_CMD_ACCEPT] = ucma_accept,
1255 [RDMA_USER_CM_CMD_REJECT] = ucma_reject,
1256 [RDMA_USER_CM_CMD_DISCONNECT] = ucma_disconnect,
1257 [RDMA_USER_CM_CMD_INIT_QP_ATTR] = ucma_init_qp_attr,
1258 [RDMA_USER_CM_CMD_GET_EVENT] = ucma_get_event,
1259 [RDMA_USER_CM_CMD_GET_OPTION] = NULL,
1260 [RDMA_USER_CM_CMD_SET_OPTION] = ucma_set_option,
1261 [RDMA_USER_CM_CMD_NOTIFY] = ucma_notify,
1262 [RDMA_USER_CM_CMD_JOIN_MCAST] = ucma_join_multicast,
1263 [RDMA_USER_CM_CMD_LEAVE_MCAST] = ucma_leave_multicast,
1264 [RDMA_USER_CM_CMD_MIGRATE_ID] = ucma_migrate_id
1267 static ssize_t ucma_write(struct file *filp, const char __user *buf,
1268 size_t len, loff_t *pos)
1270 struct ucma_file *file = filp->private_data;
1271 struct rdma_ucm_cmd_hdr hdr;
1274 if (len < sizeof(hdr))
1277 if (copy_from_user(&hdr, buf, sizeof(hdr)))
1280 if (hdr.cmd >= ARRAY_SIZE(ucma_cmd_table))
1283 if (hdr.in + sizeof(hdr) > len)
1286 if (!ucma_cmd_table[hdr.cmd])
1289 ret = ucma_cmd_table[hdr.cmd](file, buf + sizeof(hdr), hdr.in, hdr.out);
1296 static unsigned int ucma_poll(struct file *filp, struct poll_table_struct *wait)
1298 struct ucma_file *file = filp->private_data;
1299 unsigned int mask = 0;
1301 poll_wait(filp, &file->poll_wait, wait);
1303 if (!list_empty(&file->event_list))
1304 mask = POLLIN | POLLRDNORM;
1310 * ucma_open() does not need the BKL:
1312 * - no global state is referred to;
1313 * - there is no ioctl method to race against;
1314 * - no further module initialization is required for open to work
1315 * after the device is registered.
1317 static int ucma_open(struct inode *inode, struct file *filp)
1319 struct ucma_file *file;
1321 file = kmalloc(sizeof *file, GFP_KERNEL);
1325 INIT_LIST_HEAD(&file->event_list);
1326 INIT_LIST_HEAD(&file->ctx_list);
1327 init_waitqueue_head(&file->poll_wait);
1328 mutex_init(&file->mut);
1330 filp->private_data = file;
1333 return nonseekable_open(inode, filp);
1336 static int ucma_close(struct inode *inode, struct file *filp)
1338 struct ucma_file *file = filp->private_data;
1339 struct ucma_context *ctx, *tmp;
1341 mutex_lock(&file->mut);
1342 list_for_each_entry_safe(ctx, tmp, &file->ctx_list, list) {
1343 mutex_unlock(&file->mut);
1346 idr_remove(&ctx_idr, ctx->id);
1350 mutex_lock(&file->mut);
1352 mutex_unlock(&file->mut);
1357 static const struct file_operations ucma_fops = {
1358 .owner = THIS_MODULE,
1360 .release = ucma_close,
1361 .write = ucma_write,
1363 .llseek = no_llseek,
1366 static struct miscdevice ucma_misc = {
1367 .minor = MISC_DYNAMIC_MINOR,
1369 .nodename = "infiniband/rdma_cm",
1374 static ssize_t show_abi_version(struct device *dev,
1375 struct device_attribute *attr,
1378 return sprintf(buf, "%d\n", RDMA_USER_CM_ABI_VERSION);
1380 static DEVICE_ATTR(abi_version, S_IRUGO, show_abi_version, NULL);
1382 static int __init ucma_init(void)
1386 ret = misc_register(&ucma_misc);
1390 ret = device_create_file(ucma_misc.this_device, &dev_attr_abi_version);
1392 printk(KERN_ERR "rdma_ucm: couldn't create abi_version attr\n");
1396 ucma_ctl_table_hdr = register_sysctl_paths(ucma_ctl_path, ucma_ctl_table);
1397 if (!ucma_ctl_table_hdr) {
1398 printk(KERN_ERR "rdma_ucm: couldn't register sysctl paths\n");
1404 device_remove_file(ucma_misc.this_device, &dev_attr_abi_version);
1406 misc_deregister(&ucma_misc);
1410 static void __exit ucma_cleanup(void)
1412 unregister_sysctl_table(ucma_ctl_table_hdr);
1413 device_remove_file(ucma_misc.this_device, &dev_attr_abi_version);
1414 misc_deregister(&ucma_misc);
1415 idr_destroy(&ctx_idr);
1418 module_init(ucma_init);
1419 module_exit(ucma_cleanup);