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,
72 struct list_head ctx_list;
73 struct list_head event_list;
74 wait_queue_head_t poll_wait;
79 struct completion comp;
84 struct ucma_file *file;
85 struct rdma_cm_id *cm_id;
88 struct list_head list;
89 struct list_head mc_list;
92 struct ucma_multicast {
93 struct ucma_context *ctx;
98 struct list_head list;
99 struct sockaddr_storage addr;
103 struct ucma_context *ctx;
104 struct ucma_multicast *mc;
105 struct list_head list;
106 struct rdma_cm_id *cm_id;
107 struct rdma_ucm_event_resp resp;
110 static DEFINE_MUTEX(mut);
111 static DEFINE_IDR(ctx_idr);
112 static DEFINE_IDR(multicast_idr);
114 static inline struct ucma_context *_ucma_find_context(int id,
115 struct ucma_file *file)
117 struct ucma_context *ctx;
119 ctx = idr_find(&ctx_idr, id);
121 ctx = ERR_PTR(-ENOENT);
122 else if (ctx->file != file)
123 ctx = ERR_PTR(-EINVAL);
127 static struct ucma_context *ucma_get_ctx(struct ucma_file *file, int id)
129 struct ucma_context *ctx;
132 ctx = _ucma_find_context(id, file);
134 atomic_inc(&ctx->ref);
139 static void ucma_put_ctx(struct ucma_context *ctx)
141 if (atomic_dec_and_test(&ctx->ref))
142 complete(&ctx->comp);
145 static struct ucma_context *ucma_alloc_ctx(struct ucma_file *file)
147 struct ucma_context *ctx;
150 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
154 atomic_set(&ctx->ref, 1);
155 init_completion(&ctx->comp);
156 INIT_LIST_HEAD(&ctx->mc_list);
160 ret = idr_pre_get(&ctx_idr, GFP_KERNEL);
165 ret = idr_get_new(&ctx_idr, ctx, &ctx->id);
167 } while (ret == -EAGAIN);
172 list_add_tail(&ctx->list, &file->ctx_list);
180 static struct ucma_multicast* ucma_alloc_multicast(struct ucma_context *ctx)
182 struct ucma_multicast *mc;
185 mc = kzalloc(sizeof(*mc), GFP_KERNEL);
190 ret = idr_pre_get(&multicast_idr, GFP_KERNEL);
195 ret = idr_get_new(&multicast_idr, mc, &mc->id);
197 } while (ret == -EAGAIN);
203 list_add_tail(&mc->list, &ctx->mc_list);
211 static void ucma_copy_conn_event(struct rdma_ucm_conn_param *dst,
212 struct rdma_conn_param *src)
214 if (src->private_data_len)
215 memcpy(dst->private_data, src->private_data,
216 src->private_data_len);
217 dst->private_data_len = src->private_data_len;
218 dst->responder_resources =src->responder_resources;
219 dst->initiator_depth = src->initiator_depth;
220 dst->flow_control = src->flow_control;
221 dst->retry_count = src->retry_count;
222 dst->rnr_retry_count = src->rnr_retry_count;
224 dst->qp_num = src->qp_num;
227 static void ucma_copy_ud_event(struct rdma_ucm_ud_param *dst,
228 struct rdma_ud_param *src)
230 if (src->private_data_len)
231 memcpy(dst->private_data, src->private_data,
232 src->private_data_len);
233 dst->private_data_len = src->private_data_len;
234 ib_copy_ah_attr_to_user(&dst->ah_attr, &src->ah_attr);
235 dst->qp_num = src->qp_num;
236 dst->qkey = src->qkey;
239 static void ucma_set_event_context(struct ucma_context *ctx,
240 struct rdma_cm_event *event,
241 struct ucma_event *uevent)
244 switch (event->event) {
245 case RDMA_CM_EVENT_MULTICAST_JOIN:
246 case RDMA_CM_EVENT_MULTICAST_ERROR:
247 uevent->mc = (struct ucma_multicast *)
248 event->param.ud.private_data;
249 uevent->resp.uid = uevent->mc->uid;
250 uevent->resp.id = uevent->mc->id;
253 uevent->resp.uid = ctx->uid;
254 uevent->resp.id = ctx->id;
259 static int ucma_event_handler(struct rdma_cm_id *cm_id,
260 struct rdma_cm_event *event)
262 struct ucma_event *uevent;
263 struct ucma_context *ctx = cm_id->context;
266 uevent = kzalloc(sizeof(*uevent), GFP_KERNEL);
268 return event->event == RDMA_CM_EVENT_CONNECT_REQUEST;
270 mutex_lock(&ctx->file->mut);
271 uevent->cm_id = cm_id;
272 ucma_set_event_context(ctx, event, uevent);
273 uevent->resp.event = event->event;
274 uevent->resp.status = event->status;
275 if (cm_id->qp_type == IB_QPT_UD)
276 ucma_copy_ud_event(&uevent->resp.param.ud, &event->param.ud);
278 ucma_copy_conn_event(&uevent->resp.param.conn,
281 if (event->event == RDMA_CM_EVENT_CONNECT_REQUEST) {
288 } else if (!ctx->uid) {
290 * We ignore events for new connections until userspace has set
291 * their context. This can only happen if an error occurs on a
292 * new connection before the user accepts it. This is okay,
293 * since the accept will just fail later.
299 list_add_tail(&uevent->list, &ctx->file->event_list);
300 wake_up_interruptible(&ctx->file->poll_wait);
302 mutex_unlock(&ctx->file->mut);
306 static ssize_t ucma_get_event(struct ucma_file *file, const char __user *inbuf,
307 int in_len, int out_len)
309 struct ucma_context *ctx;
310 struct rdma_ucm_get_event cmd;
311 struct ucma_event *uevent;
315 if (out_len < sizeof uevent->resp)
318 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
321 mutex_lock(&file->mut);
322 while (list_empty(&file->event_list)) {
323 mutex_unlock(&file->mut);
325 if (file->filp->f_flags & O_NONBLOCK)
328 if (wait_event_interruptible(file->poll_wait,
329 !list_empty(&file->event_list)))
332 mutex_lock(&file->mut);
335 uevent = list_entry(file->event_list.next, struct ucma_event, list);
337 if (uevent->resp.event == RDMA_CM_EVENT_CONNECT_REQUEST) {
338 ctx = ucma_alloc_ctx(file);
343 uevent->ctx->backlog++;
344 ctx->cm_id = uevent->cm_id;
345 ctx->cm_id->context = ctx;
346 uevent->resp.id = ctx->id;
349 if (copy_to_user((void __user *)(unsigned long)cmd.response,
350 &uevent->resp, sizeof uevent->resp)) {
355 list_del(&uevent->list);
356 uevent->ctx->events_reported++;
358 uevent->mc->events_reported++;
361 mutex_unlock(&file->mut);
365 static int ucma_get_qp_type(struct rdma_ucm_create_id *cmd, enum ib_qp_type *qp_type)
369 *qp_type = IB_QPT_RC;
373 *qp_type = IB_QPT_UD;
376 *qp_type = cmd->qp_type;
383 static ssize_t ucma_create_id(struct ucma_file *file, const char __user *inbuf,
384 int in_len, int out_len)
386 struct rdma_ucm_create_id cmd;
387 struct rdma_ucm_create_id_resp resp;
388 struct ucma_context *ctx;
389 enum ib_qp_type qp_type;
392 if (out_len < sizeof(resp))
395 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
398 ret = ucma_get_qp_type(&cmd, &qp_type);
402 mutex_lock(&file->mut);
403 ctx = ucma_alloc_ctx(file);
404 mutex_unlock(&file->mut);
409 ctx->cm_id = rdma_create_id(ucma_event_handler, ctx, cmd.ps, qp_type);
410 if (IS_ERR(ctx->cm_id)) {
411 ret = PTR_ERR(ctx->cm_id);
416 if (copy_to_user((void __user *)(unsigned long)cmd.response,
417 &resp, sizeof(resp))) {
424 rdma_destroy_id(ctx->cm_id);
427 idr_remove(&ctx_idr, ctx->id);
433 static void ucma_cleanup_multicast(struct ucma_context *ctx)
435 struct ucma_multicast *mc, *tmp;
438 list_for_each_entry_safe(mc, tmp, &ctx->mc_list, list) {
440 idr_remove(&multicast_idr, mc->id);
446 static void ucma_cleanup_mc_events(struct ucma_multicast *mc)
448 struct ucma_event *uevent, *tmp;
450 list_for_each_entry_safe(uevent, tmp, &mc->ctx->file->event_list, list) {
451 if (uevent->mc != mc)
454 list_del(&uevent->list);
460 * We cannot hold file->mut when calling rdma_destroy_id() or we can
461 * deadlock. We also acquire file->mut in ucma_event_handler(), and
462 * rdma_destroy_id() will wait until all callbacks have completed.
464 static int ucma_free_ctx(struct ucma_context *ctx)
467 struct ucma_event *uevent, *tmp;
470 /* No new events will be generated after destroying the id. */
471 rdma_destroy_id(ctx->cm_id);
473 ucma_cleanup_multicast(ctx);
475 /* Cleanup events not yet reported to the user. */
476 mutex_lock(&ctx->file->mut);
477 list_for_each_entry_safe(uevent, tmp, &ctx->file->event_list, list) {
478 if (uevent->ctx == ctx)
479 list_move_tail(&uevent->list, &list);
481 list_del(&ctx->list);
482 mutex_unlock(&ctx->file->mut);
484 list_for_each_entry_safe(uevent, tmp, &list, list) {
485 list_del(&uevent->list);
486 if (uevent->resp.event == RDMA_CM_EVENT_CONNECT_REQUEST)
487 rdma_destroy_id(uevent->cm_id);
491 events_reported = ctx->events_reported;
493 return events_reported;
496 static ssize_t ucma_destroy_id(struct ucma_file *file, const char __user *inbuf,
497 int in_len, int out_len)
499 struct rdma_ucm_destroy_id cmd;
500 struct rdma_ucm_destroy_id_resp resp;
501 struct ucma_context *ctx;
504 if (out_len < sizeof(resp))
507 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
511 ctx = _ucma_find_context(cmd.id, file);
513 idr_remove(&ctx_idr, ctx->id);
520 wait_for_completion(&ctx->comp);
521 resp.events_reported = ucma_free_ctx(ctx);
523 if (copy_to_user((void __user *)(unsigned long)cmd.response,
524 &resp, sizeof(resp)))
530 static ssize_t ucma_bind_addr(struct ucma_file *file, const char __user *inbuf,
531 int in_len, int out_len)
533 struct rdma_ucm_bind_addr cmd;
534 struct ucma_context *ctx;
537 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
540 ctx = ucma_get_ctx(file, cmd.id);
544 ret = rdma_bind_addr(ctx->cm_id, (struct sockaddr *) &cmd.addr);
549 static ssize_t ucma_resolve_addr(struct ucma_file *file,
550 const char __user *inbuf,
551 int in_len, int out_len)
553 struct rdma_ucm_resolve_addr cmd;
554 struct ucma_context *ctx;
557 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
560 ctx = ucma_get_ctx(file, cmd.id);
564 ret = rdma_resolve_addr(ctx->cm_id, (struct sockaddr *) &cmd.src_addr,
565 (struct sockaddr *) &cmd.dst_addr,
571 static ssize_t ucma_resolve_route(struct ucma_file *file,
572 const char __user *inbuf,
573 int in_len, int out_len)
575 struct rdma_ucm_resolve_route cmd;
576 struct ucma_context *ctx;
579 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
582 ctx = ucma_get_ctx(file, cmd.id);
586 ret = rdma_resolve_route(ctx->cm_id, cmd.timeout_ms);
591 static void ucma_copy_ib_route(struct rdma_ucm_query_route_resp *resp,
592 struct rdma_route *route)
594 struct rdma_dev_addr *dev_addr;
596 resp->num_paths = route->num_paths;
597 switch (route->num_paths) {
599 dev_addr = &route->addr.dev_addr;
600 rdma_addr_get_dgid(dev_addr,
601 (union ib_gid *) &resp->ib_route[0].dgid);
602 rdma_addr_get_sgid(dev_addr,
603 (union ib_gid *) &resp->ib_route[0].sgid);
604 resp->ib_route[0].pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
607 ib_copy_path_rec_to_user(&resp->ib_route[1],
608 &route->path_rec[1]);
611 ib_copy_path_rec_to_user(&resp->ib_route[0],
612 &route->path_rec[0]);
619 static void ucma_copy_iboe_route(struct rdma_ucm_query_route_resp *resp,
620 struct rdma_route *route)
622 struct rdma_dev_addr *dev_addr;
623 struct net_device *dev;
626 resp->num_paths = route->num_paths;
627 switch (route->num_paths) {
629 dev_addr = &route->addr.dev_addr;
630 dev = dev_get_by_index(&init_net, dev_addr->bound_dev_if);
632 vid = rdma_vlan_dev_vlan_id(dev);
636 iboe_mac_vlan_to_ll((union ib_gid *) &resp->ib_route[0].dgid,
637 dev_addr->dst_dev_addr, vid);
638 iboe_addr_get_sgid(dev_addr,
639 (union ib_gid *) &resp->ib_route[0].sgid);
640 resp->ib_route[0].pkey = cpu_to_be16(0xffff);
643 ib_copy_path_rec_to_user(&resp->ib_route[1],
644 &route->path_rec[1]);
647 ib_copy_path_rec_to_user(&resp->ib_route[0],
648 &route->path_rec[0]);
655 static void ucma_copy_iw_route(struct rdma_ucm_query_route_resp *resp,
656 struct rdma_route *route)
658 struct rdma_dev_addr *dev_addr;
660 dev_addr = &route->addr.dev_addr;
661 rdma_addr_get_dgid(dev_addr, (union ib_gid *) &resp->ib_route[0].dgid);
662 rdma_addr_get_sgid(dev_addr, (union ib_gid *) &resp->ib_route[0].sgid);
665 static ssize_t ucma_query_route(struct ucma_file *file,
666 const char __user *inbuf,
667 int in_len, int out_len)
669 struct rdma_ucm_query_route cmd;
670 struct rdma_ucm_query_route_resp resp;
671 struct ucma_context *ctx;
672 struct sockaddr *addr;
675 if (out_len < sizeof(resp))
678 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
681 ctx = ucma_get_ctx(file, cmd.id);
685 memset(&resp, 0, sizeof resp);
686 addr = (struct sockaddr *) &ctx->cm_id->route.addr.src_addr;
687 memcpy(&resp.src_addr, addr, addr->sa_family == AF_INET ?
688 sizeof(struct sockaddr_in) :
689 sizeof(struct sockaddr_in6));
690 addr = (struct sockaddr *) &ctx->cm_id->route.addr.dst_addr;
691 memcpy(&resp.dst_addr, addr, addr->sa_family == AF_INET ?
692 sizeof(struct sockaddr_in) :
693 sizeof(struct sockaddr_in6));
694 if (!ctx->cm_id->device)
697 resp.node_guid = (__force __u64) ctx->cm_id->device->node_guid;
698 resp.port_num = ctx->cm_id->port_num;
699 switch (rdma_node_get_transport(ctx->cm_id->device->node_type)) {
700 case RDMA_TRANSPORT_IB:
701 switch (rdma_port_get_link_layer(ctx->cm_id->device,
702 ctx->cm_id->port_num)) {
703 case IB_LINK_LAYER_INFINIBAND:
704 ucma_copy_ib_route(&resp, &ctx->cm_id->route);
706 case IB_LINK_LAYER_ETHERNET:
707 ucma_copy_iboe_route(&resp, &ctx->cm_id->route);
713 case RDMA_TRANSPORT_IWARP:
714 ucma_copy_iw_route(&resp, &ctx->cm_id->route);
721 if (copy_to_user((void __user *)(unsigned long)cmd.response,
722 &resp, sizeof(resp)))
729 static void ucma_copy_conn_param(struct rdma_conn_param *dst,
730 struct rdma_ucm_conn_param *src)
732 dst->private_data = src->private_data;
733 dst->private_data_len = src->private_data_len;
734 dst->responder_resources =src->responder_resources;
735 dst->initiator_depth = src->initiator_depth;
736 dst->flow_control = src->flow_control;
737 dst->retry_count = src->retry_count;
738 dst->rnr_retry_count = src->rnr_retry_count;
740 dst->qp_num = src->qp_num;
743 static ssize_t ucma_connect(struct ucma_file *file, const char __user *inbuf,
744 int in_len, int out_len)
746 struct rdma_ucm_connect cmd;
747 struct rdma_conn_param conn_param;
748 struct ucma_context *ctx;
751 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
754 if (!cmd.conn_param.valid)
757 ctx = ucma_get_ctx(file, cmd.id);
761 ucma_copy_conn_param(&conn_param, &cmd.conn_param);
762 ret = rdma_connect(ctx->cm_id, &conn_param);
767 static ssize_t ucma_listen(struct ucma_file *file, const char __user *inbuf,
768 int in_len, int out_len)
770 struct rdma_ucm_listen cmd;
771 struct ucma_context *ctx;
774 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
777 ctx = ucma_get_ctx(file, cmd.id);
781 ctx->backlog = cmd.backlog > 0 && cmd.backlog < max_backlog ?
782 cmd.backlog : max_backlog;
783 ret = rdma_listen(ctx->cm_id, ctx->backlog);
788 static ssize_t ucma_accept(struct ucma_file *file, const char __user *inbuf,
789 int in_len, int out_len)
791 struct rdma_ucm_accept cmd;
792 struct rdma_conn_param conn_param;
793 struct ucma_context *ctx;
796 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
799 ctx = ucma_get_ctx(file, cmd.id);
803 if (cmd.conn_param.valid) {
804 ucma_copy_conn_param(&conn_param, &cmd.conn_param);
805 mutex_lock(&file->mut);
806 ret = rdma_accept(ctx->cm_id, &conn_param);
809 mutex_unlock(&file->mut);
811 ret = rdma_accept(ctx->cm_id, NULL);
817 static ssize_t ucma_reject(struct ucma_file *file, const char __user *inbuf,
818 int in_len, int out_len)
820 struct rdma_ucm_reject 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_reject(ctx->cm_id, cmd.private_data, cmd.private_data_len);
836 static ssize_t ucma_disconnect(struct ucma_file *file, const char __user *inbuf,
837 int in_len, int out_len)
839 struct rdma_ucm_disconnect cmd;
840 struct ucma_context *ctx;
843 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
846 ctx = ucma_get_ctx(file, cmd.id);
850 ret = rdma_disconnect(ctx->cm_id);
855 static ssize_t ucma_init_qp_attr(struct ucma_file *file,
856 const char __user *inbuf,
857 int in_len, int out_len)
859 struct rdma_ucm_init_qp_attr cmd;
860 struct ib_uverbs_qp_attr resp;
861 struct ucma_context *ctx;
862 struct ib_qp_attr qp_attr;
865 if (out_len < sizeof(resp))
868 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
871 ctx = ucma_get_ctx(file, cmd.id);
875 resp.qp_attr_mask = 0;
876 memset(&qp_attr, 0, sizeof qp_attr);
877 qp_attr.qp_state = cmd.qp_state;
878 ret = rdma_init_qp_attr(ctx->cm_id, &qp_attr, &resp.qp_attr_mask);
882 ib_copy_qp_attr_to_user(&resp, &qp_attr);
883 if (copy_to_user((void __user *)(unsigned long)cmd.response,
884 &resp, sizeof(resp)))
892 static int ucma_set_option_id(struct ucma_context *ctx, int optname,
893 void *optval, size_t optlen)
898 case RDMA_OPTION_ID_TOS:
899 if (optlen != sizeof(u8)) {
903 rdma_set_service_type(ctx->cm_id, *((u8 *) optval));
905 case RDMA_OPTION_ID_REUSEADDR:
906 if (optlen != sizeof(int)) {
910 ret = rdma_set_reuseaddr(ctx->cm_id, *((int *) optval) ? 1 : 0);
912 case RDMA_OPTION_ID_AFONLY:
913 if (optlen != sizeof(int)) {
917 ret = rdma_set_afonly(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 = memdup_user((void __user *) (unsigned long) cmd.optval,
1007 if (IS_ERR(optval)) {
1008 ret = PTR_ERR(optval);
1012 ret = ucma_set_option_level(ctx, cmd.level, cmd.optname, optval,
1021 static ssize_t ucma_notify(struct ucma_file *file, const char __user *inbuf,
1022 int in_len, int out_len)
1024 struct rdma_ucm_notify cmd;
1025 struct ucma_context *ctx;
1028 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1031 ctx = ucma_get_ctx(file, cmd.id);
1033 return PTR_ERR(ctx);
1035 ret = rdma_notify(ctx->cm_id, (enum ib_event_type) cmd.event);
1040 static ssize_t ucma_join_multicast(struct ucma_file *file,
1041 const char __user *inbuf,
1042 int in_len, int out_len)
1044 struct rdma_ucm_join_mcast cmd;
1045 struct rdma_ucm_create_id_resp resp;
1046 struct ucma_context *ctx;
1047 struct ucma_multicast *mc;
1050 if (out_len < sizeof(resp))
1053 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1056 ctx = ucma_get_ctx(file, cmd.id);
1058 return PTR_ERR(ctx);
1060 mutex_lock(&file->mut);
1061 mc = ucma_alloc_multicast(ctx);
1068 memcpy(&mc->addr, &cmd.addr, sizeof cmd.addr);
1069 ret = rdma_join_multicast(ctx->cm_id, (struct sockaddr *) &mc->addr, mc);
1074 if (copy_to_user((void __user *)(unsigned long)cmd.response,
1075 &resp, sizeof(resp))) {
1080 mutex_unlock(&file->mut);
1085 rdma_leave_multicast(ctx->cm_id, (struct sockaddr *) &mc->addr);
1086 ucma_cleanup_mc_events(mc);
1089 idr_remove(&multicast_idr, mc->id);
1091 list_del(&mc->list);
1094 mutex_unlock(&file->mut);
1099 static ssize_t ucma_leave_multicast(struct ucma_file *file,
1100 const char __user *inbuf,
1101 int in_len, int out_len)
1103 struct rdma_ucm_destroy_id cmd;
1104 struct rdma_ucm_destroy_id_resp resp;
1105 struct ucma_multicast *mc;
1108 if (out_len < sizeof(resp))
1111 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1115 mc = idr_find(&multicast_idr, cmd.id);
1117 mc = ERR_PTR(-ENOENT);
1118 else if (mc->ctx->file != file)
1119 mc = ERR_PTR(-EINVAL);
1121 idr_remove(&multicast_idr, mc->id);
1122 atomic_inc(&mc->ctx->ref);
1131 rdma_leave_multicast(mc->ctx->cm_id, (struct sockaddr *) &mc->addr);
1132 mutex_lock(&mc->ctx->file->mut);
1133 ucma_cleanup_mc_events(mc);
1134 list_del(&mc->list);
1135 mutex_unlock(&mc->ctx->file->mut);
1137 ucma_put_ctx(mc->ctx);
1138 resp.events_reported = mc->events_reported;
1141 if (copy_to_user((void __user *)(unsigned long)cmd.response,
1142 &resp, sizeof(resp)))
1148 static void ucma_lock_files(struct ucma_file *file1, struct ucma_file *file2)
1150 /* Acquire mutex's based on pointer comparison to prevent deadlock. */
1151 if (file1 < file2) {
1152 mutex_lock(&file1->mut);
1153 mutex_lock(&file2->mut);
1155 mutex_lock(&file2->mut);
1156 mutex_lock(&file1->mut);
1160 static void ucma_unlock_files(struct ucma_file *file1, struct ucma_file *file2)
1162 if (file1 < file2) {
1163 mutex_unlock(&file2->mut);
1164 mutex_unlock(&file1->mut);
1166 mutex_unlock(&file1->mut);
1167 mutex_unlock(&file2->mut);
1171 static void ucma_move_events(struct ucma_context *ctx, struct ucma_file *file)
1173 struct ucma_event *uevent, *tmp;
1175 list_for_each_entry_safe(uevent, tmp, &ctx->file->event_list, list)
1176 if (uevent->ctx == ctx)
1177 list_move_tail(&uevent->list, &file->event_list);
1180 static ssize_t ucma_migrate_id(struct ucma_file *new_file,
1181 const char __user *inbuf,
1182 int in_len, int out_len)
1184 struct rdma_ucm_migrate_id cmd;
1185 struct rdma_ucm_migrate_resp resp;
1186 struct ucma_context *ctx;
1188 struct ucma_file *cur_file;
1191 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1194 /* Get current fd to protect against it being closed */
1195 filp = fget(cmd.fd);
1199 /* Validate current fd and prevent destruction of id. */
1200 ctx = ucma_get_ctx(filp->private_data, cmd.id);
1206 cur_file = ctx->file;
1207 if (cur_file == new_file) {
1208 resp.events_reported = ctx->events_reported;
1213 * Migrate events between fd's, maintaining order, and avoiding new
1214 * events being added before existing events.
1216 ucma_lock_files(cur_file, new_file);
1219 list_move_tail(&ctx->list, &new_file->ctx_list);
1220 ucma_move_events(ctx, new_file);
1221 ctx->file = new_file;
1222 resp.events_reported = ctx->events_reported;
1225 ucma_unlock_files(cur_file, new_file);
1228 if (copy_to_user((void __user *)(unsigned long)cmd.response,
1229 &resp, sizeof(resp)))
1238 static ssize_t (*ucma_cmd_table[])(struct ucma_file *file,
1239 const char __user *inbuf,
1240 int in_len, int out_len) = {
1241 [RDMA_USER_CM_CMD_CREATE_ID] = ucma_create_id,
1242 [RDMA_USER_CM_CMD_DESTROY_ID] = ucma_destroy_id,
1243 [RDMA_USER_CM_CMD_BIND_ADDR] = ucma_bind_addr,
1244 [RDMA_USER_CM_CMD_RESOLVE_ADDR] = ucma_resolve_addr,
1245 [RDMA_USER_CM_CMD_RESOLVE_ROUTE]= ucma_resolve_route,
1246 [RDMA_USER_CM_CMD_QUERY_ROUTE] = ucma_query_route,
1247 [RDMA_USER_CM_CMD_CONNECT] = ucma_connect,
1248 [RDMA_USER_CM_CMD_LISTEN] = ucma_listen,
1249 [RDMA_USER_CM_CMD_ACCEPT] = ucma_accept,
1250 [RDMA_USER_CM_CMD_REJECT] = ucma_reject,
1251 [RDMA_USER_CM_CMD_DISCONNECT] = ucma_disconnect,
1252 [RDMA_USER_CM_CMD_INIT_QP_ATTR] = ucma_init_qp_attr,
1253 [RDMA_USER_CM_CMD_GET_EVENT] = ucma_get_event,
1254 [RDMA_USER_CM_CMD_GET_OPTION] = NULL,
1255 [RDMA_USER_CM_CMD_SET_OPTION] = ucma_set_option,
1256 [RDMA_USER_CM_CMD_NOTIFY] = ucma_notify,
1257 [RDMA_USER_CM_CMD_JOIN_MCAST] = ucma_join_multicast,
1258 [RDMA_USER_CM_CMD_LEAVE_MCAST] = ucma_leave_multicast,
1259 [RDMA_USER_CM_CMD_MIGRATE_ID] = ucma_migrate_id
1262 static ssize_t ucma_write(struct file *filp, const char __user *buf,
1263 size_t len, loff_t *pos)
1265 struct ucma_file *file = filp->private_data;
1266 struct rdma_ucm_cmd_hdr hdr;
1269 if (len < sizeof(hdr))
1272 if (copy_from_user(&hdr, buf, sizeof(hdr)))
1275 if (hdr.cmd >= ARRAY_SIZE(ucma_cmd_table))
1278 if (hdr.in + sizeof(hdr) > len)
1281 if (!ucma_cmd_table[hdr.cmd])
1284 ret = ucma_cmd_table[hdr.cmd](file, buf + sizeof(hdr), hdr.in, hdr.out);
1291 static unsigned int ucma_poll(struct file *filp, struct poll_table_struct *wait)
1293 struct ucma_file *file = filp->private_data;
1294 unsigned int mask = 0;
1296 poll_wait(filp, &file->poll_wait, wait);
1298 if (!list_empty(&file->event_list))
1299 mask = POLLIN | POLLRDNORM;
1305 * ucma_open() does not need the BKL:
1307 * - no global state is referred to;
1308 * - there is no ioctl method to race against;
1309 * - no further module initialization is required for open to work
1310 * after the device is registered.
1312 static int ucma_open(struct inode *inode, struct file *filp)
1314 struct ucma_file *file;
1316 file = kmalloc(sizeof *file, GFP_KERNEL);
1320 INIT_LIST_HEAD(&file->event_list);
1321 INIT_LIST_HEAD(&file->ctx_list);
1322 init_waitqueue_head(&file->poll_wait);
1323 mutex_init(&file->mut);
1325 filp->private_data = file;
1328 return nonseekable_open(inode, filp);
1331 static int ucma_close(struct inode *inode, struct file *filp)
1333 struct ucma_file *file = filp->private_data;
1334 struct ucma_context *ctx, *tmp;
1336 mutex_lock(&file->mut);
1337 list_for_each_entry_safe(ctx, tmp, &file->ctx_list, list) {
1338 mutex_unlock(&file->mut);
1341 idr_remove(&ctx_idr, ctx->id);
1345 mutex_lock(&file->mut);
1347 mutex_unlock(&file->mut);
1352 static const struct file_operations ucma_fops = {
1353 .owner = THIS_MODULE,
1355 .release = ucma_close,
1356 .write = ucma_write,
1358 .llseek = no_llseek,
1361 static struct miscdevice ucma_misc = {
1362 .minor = MISC_DYNAMIC_MINOR,
1364 .nodename = "infiniband/rdma_cm",
1369 static ssize_t show_abi_version(struct device *dev,
1370 struct device_attribute *attr,
1373 return sprintf(buf, "%d\n", RDMA_USER_CM_ABI_VERSION);
1375 static DEVICE_ATTR(abi_version, S_IRUGO, show_abi_version, NULL);
1377 static int __init ucma_init(void)
1381 ret = misc_register(&ucma_misc);
1385 ret = device_create_file(ucma_misc.this_device, &dev_attr_abi_version);
1387 printk(KERN_ERR "rdma_ucm: couldn't create abi_version attr\n");
1391 ucma_ctl_table_hdr = register_net_sysctl(&init_net, "net/rdma_ucm", ucma_ctl_table);
1392 if (!ucma_ctl_table_hdr) {
1393 printk(KERN_ERR "rdma_ucm: couldn't register sysctl paths\n");
1399 device_remove_file(ucma_misc.this_device, &dev_attr_abi_version);
1401 misc_deregister(&ucma_misc);
1405 static void __exit ucma_cleanup(void)
1407 unregister_net_sysctl_table(ucma_ctl_table_hdr);
1408 device_remove_file(ucma_misc.this_device, &dev_attr_abi_version);
1409 misc_deregister(&ucma_misc);
1410 idr_destroy(&ctx_idr);
1413 module_init(ucma_init);
1414 module_exit(ucma_cleanup);