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>
52 MODULE_AUTHOR("Sean Hefty");
53 MODULE_DESCRIPTION("RDMA Userspace Connection Manager Access");
54 MODULE_LICENSE("Dual BSD/GPL");
56 static unsigned int max_backlog = 1024;
58 static struct ctl_table_header *ucma_ctl_table_hdr;
59 static ctl_table ucma_ctl_table[] = {
61 .procname = "max_backlog",
63 .maxlen = sizeof max_backlog,
65 .proc_handler = proc_dointvec,
70 static struct ctl_path ucma_ctl_path[] = {
71 { .procname = "net" },
72 { .procname = "rdma_ucm" },
79 struct list_head ctx_list;
80 struct list_head event_list;
81 wait_queue_head_t poll_wait;
86 struct completion comp;
91 struct ucma_file *file;
92 struct rdma_cm_id *cm_id;
95 struct list_head list;
96 struct list_head mc_list;
99 struct ucma_multicast {
100 struct ucma_context *ctx;
105 struct list_head list;
106 struct sockaddr_storage addr;
110 struct ucma_context *ctx;
111 struct ucma_multicast *mc;
112 struct list_head list;
113 struct rdma_cm_id *cm_id;
114 struct rdma_ucm_event_resp resp;
117 static DEFINE_MUTEX(mut);
118 static DEFINE_IDR(ctx_idr);
119 static DEFINE_IDR(multicast_idr);
121 static inline struct ucma_context *_ucma_find_context(int id,
122 struct ucma_file *file)
124 struct ucma_context *ctx;
126 ctx = idr_find(&ctx_idr, id);
128 ctx = ERR_PTR(-ENOENT);
129 else if (ctx->file != file)
130 ctx = ERR_PTR(-EINVAL);
134 static struct ucma_context *ucma_get_ctx(struct ucma_file *file, int id)
136 struct ucma_context *ctx;
139 ctx = _ucma_find_context(id, file);
141 atomic_inc(&ctx->ref);
146 static void ucma_put_ctx(struct ucma_context *ctx)
148 if (atomic_dec_and_test(&ctx->ref))
149 complete(&ctx->comp);
152 static struct ucma_context *ucma_alloc_ctx(struct ucma_file *file)
154 struct ucma_context *ctx;
157 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
161 atomic_set(&ctx->ref, 1);
162 init_completion(&ctx->comp);
163 INIT_LIST_HEAD(&ctx->mc_list);
167 ret = idr_pre_get(&ctx_idr, GFP_KERNEL);
172 ret = idr_get_new(&ctx_idr, ctx, &ctx->id);
174 } while (ret == -EAGAIN);
179 list_add_tail(&ctx->list, &file->ctx_list);
187 static struct ucma_multicast* ucma_alloc_multicast(struct ucma_context *ctx)
189 struct ucma_multicast *mc;
192 mc = kzalloc(sizeof(*mc), GFP_KERNEL);
197 ret = idr_pre_get(&multicast_idr, GFP_KERNEL);
202 ret = idr_get_new(&multicast_idr, mc, &mc->id);
204 } while (ret == -EAGAIN);
210 list_add_tail(&mc->list, &ctx->mc_list);
218 static void ucma_copy_conn_event(struct rdma_ucm_conn_param *dst,
219 struct rdma_conn_param *src)
221 if (src->private_data_len)
222 memcpy(dst->private_data, src->private_data,
223 src->private_data_len);
224 dst->private_data_len = src->private_data_len;
225 dst->responder_resources =src->responder_resources;
226 dst->initiator_depth = src->initiator_depth;
227 dst->flow_control = src->flow_control;
228 dst->retry_count = src->retry_count;
229 dst->rnr_retry_count = src->rnr_retry_count;
231 dst->qp_num = src->qp_num;
234 static void ucma_copy_ud_event(struct rdma_ucm_ud_param *dst,
235 struct rdma_ud_param *src)
237 if (src->private_data_len)
238 memcpy(dst->private_data, src->private_data,
239 src->private_data_len);
240 dst->private_data_len = src->private_data_len;
241 ib_copy_ah_attr_to_user(&dst->ah_attr, &src->ah_attr);
242 dst->qp_num = src->qp_num;
243 dst->qkey = src->qkey;
246 static void ucma_set_event_context(struct ucma_context *ctx,
247 struct rdma_cm_event *event,
248 struct ucma_event *uevent)
251 switch (event->event) {
252 case RDMA_CM_EVENT_MULTICAST_JOIN:
253 case RDMA_CM_EVENT_MULTICAST_ERROR:
254 uevent->mc = (struct ucma_multicast *)
255 event->param.ud.private_data;
256 uevent->resp.uid = uevent->mc->uid;
257 uevent->resp.id = uevent->mc->id;
260 uevent->resp.uid = ctx->uid;
261 uevent->resp.id = ctx->id;
266 static int ucma_event_handler(struct rdma_cm_id *cm_id,
267 struct rdma_cm_event *event)
269 struct ucma_event *uevent;
270 struct ucma_context *ctx = cm_id->context;
273 uevent = kzalloc(sizeof(*uevent), GFP_KERNEL);
275 return event->event == RDMA_CM_EVENT_CONNECT_REQUEST;
277 uevent->cm_id = cm_id;
278 ucma_set_event_context(ctx, event, uevent);
279 uevent->resp.event = event->event;
280 uevent->resp.status = event->status;
281 if (cm_id->qp_type == IB_QPT_UD)
282 ucma_copy_ud_event(&uevent->resp.param.ud, &event->param.ud);
284 ucma_copy_conn_event(&uevent->resp.param.conn,
287 mutex_lock(&ctx->file->mut);
288 if (event->event == RDMA_CM_EVENT_CONNECT_REQUEST) {
295 } else if (!ctx->uid) {
297 * We ignore events for new connections until userspace has set
298 * their context. This can only happen if an error occurs on a
299 * new connection before the user accepts it. This is okay,
300 * since the accept will just fail later.
306 list_add_tail(&uevent->list, &ctx->file->event_list);
307 wake_up_interruptible(&ctx->file->poll_wait);
309 mutex_unlock(&ctx->file->mut);
313 static ssize_t ucma_get_event(struct ucma_file *file, const char __user *inbuf,
314 int in_len, int out_len)
316 struct ucma_context *ctx;
317 struct rdma_ucm_get_event cmd;
318 struct ucma_event *uevent;
322 if (out_len < sizeof uevent->resp)
325 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
328 mutex_lock(&file->mut);
329 while (list_empty(&file->event_list)) {
330 mutex_unlock(&file->mut);
332 if (file->filp->f_flags & O_NONBLOCK)
335 if (wait_event_interruptible(file->poll_wait,
336 !list_empty(&file->event_list)))
339 mutex_lock(&file->mut);
342 uevent = list_entry(file->event_list.next, struct ucma_event, list);
344 if (uevent->resp.event == RDMA_CM_EVENT_CONNECT_REQUEST) {
345 ctx = ucma_alloc_ctx(file);
350 uevent->ctx->backlog++;
351 ctx->cm_id = uevent->cm_id;
352 ctx->cm_id->context = ctx;
353 uevent->resp.id = ctx->id;
356 if (copy_to_user((void __user *)(unsigned long)cmd.response,
357 &uevent->resp, sizeof uevent->resp)) {
362 list_del(&uevent->list);
363 uevent->ctx->events_reported++;
365 uevent->mc->events_reported++;
368 mutex_unlock(&file->mut);
372 static int ucma_get_qp_type(struct rdma_ucm_create_id *cmd, enum ib_qp_type *qp_type)
376 *qp_type = IB_QPT_RC;
380 *qp_type = IB_QPT_UD;
383 *qp_type = cmd->qp_type;
390 static ssize_t ucma_create_id(struct ucma_file *file, const char __user *inbuf,
391 int in_len, int out_len)
393 struct rdma_ucm_create_id cmd;
394 struct rdma_ucm_create_id_resp resp;
395 struct ucma_context *ctx;
396 enum ib_qp_type qp_type;
399 if (out_len < sizeof(resp))
402 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
405 ret = ucma_get_qp_type(&cmd, &qp_type);
409 mutex_lock(&file->mut);
410 ctx = ucma_alloc_ctx(file);
411 mutex_unlock(&file->mut);
416 ctx->cm_id = rdma_create_id(ucma_event_handler, ctx, cmd.ps, qp_type);
417 if (IS_ERR(ctx->cm_id)) {
418 ret = PTR_ERR(ctx->cm_id);
423 if (copy_to_user((void __user *)(unsigned long)cmd.response,
424 &resp, sizeof(resp))) {
431 rdma_destroy_id(ctx->cm_id);
434 idr_remove(&ctx_idr, ctx->id);
436 mutex_lock(&file->mut);
437 list_del(&ctx->list);
438 mutex_unlock(&file->mut);
443 static void ucma_cleanup_multicast(struct ucma_context *ctx)
445 struct ucma_multicast *mc, *tmp;
448 list_for_each_entry_safe(mc, tmp, &ctx->mc_list, list) {
450 idr_remove(&multicast_idr, mc->id);
456 static void ucma_cleanup_events(struct ucma_context *ctx)
458 struct ucma_event *uevent, *tmp;
460 list_for_each_entry_safe(uevent, tmp, &ctx->file->event_list, list) {
461 if (uevent->ctx != ctx)
464 list_del(&uevent->list);
466 /* clear incoming connections. */
467 if (uevent->resp.event == RDMA_CM_EVENT_CONNECT_REQUEST)
468 rdma_destroy_id(uevent->cm_id);
474 static void ucma_cleanup_mc_events(struct ucma_multicast *mc)
476 struct ucma_event *uevent, *tmp;
478 list_for_each_entry_safe(uevent, tmp, &mc->ctx->file->event_list, list) {
479 if (uevent->mc != mc)
482 list_del(&uevent->list);
487 static int ucma_free_ctx(struct ucma_context *ctx)
491 /* No new events will be generated after destroying the id. */
492 rdma_destroy_id(ctx->cm_id);
494 ucma_cleanup_multicast(ctx);
496 /* Cleanup events not yet reported to the user. */
497 mutex_lock(&ctx->file->mut);
498 ucma_cleanup_events(ctx);
499 list_del(&ctx->list);
500 mutex_unlock(&ctx->file->mut);
502 events_reported = ctx->events_reported;
504 return events_reported;
507 static ssize_t ucma_destroy_id(struct ucma_file *file, const char __user *inbuf,
508 int in_len, int out_len)
510 struct rdma_ucm_destroy_id cmd;
511 struct rdma_ucm_destroy_id_resp resp;
512 struct ucma_context *ctx;
515 if (out_len < sizeof(resp))
518 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
522 ctx = _ucma_find_context(cmd.id, file);
524 idr_remove(&ctx_idr, ctx->id);
531 wait_for_completion(&ctx->comp);
532 resp.events_reported = ucma_free_ctx(ctx);
534 if (copy_to_user((void __user *)(unsigned long)cmd.response,
535 &resp, sizeof(resp)))
541 static ssize_t ucma_bind_addr(struct ucma_file *file, const char __user *inbuf,
542 int in_len, int out_len)
544 struct rdma_ucm_bind_addr cmd;
545 struct ucma_context *ctx;
548 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
551 ctx = ucma_get_ctx(file, cmd.id);
555 ret = rdma_bind_addr(ctx->cm_id, (struct sockaddr *) &cmd.addr);
560 static ssize_t ucma_resolve_addr(struct ucma_file *file,
561 const char __user *inbuf,
562 int in_len, int out_len)
564 struct rdma_ucm_resolve_addr cmd;
565 struct ucma_context *ctx;
568 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
571 ctx = ucma_get_ctx(file, cmd.id);
575 ret = rdma_resolve_addr(ctx->cm_id, (struct sockaddr *) &cmd.src_addr,
576 (struct sockaddr *) &cmd.dst_addr,
582 static ssize_t ucma_resolve_route(struct ucma_file *file,
583 const char __user *inbuf,
584 int in_len, int out_len)
586 struct rdma_ucm_resolve_route cmd;
587 struct ucma_context *ctx;
590 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
593 ctx = ucma_get_ctx(file, cmd.id);
597 ret = rdma_resolve_route(ctx->cm_id, cmd.timeout_ms);
602 static void ucma_copy_ib_route(struct rdma_ucm_query_route_resp *resp,
603 struct rdma_route *route)
605 struct rdma_dev_addr *dev_addr;
607 resp->num_paths = route->num_paths;
608 switch (route->num_paths) {
610 dev_addr = &route->addr.dev_addr;
611 rdma_addr_get_dgid(dev_addr,
612 (union ib_gid *) &resp->ib_route[0].dgid);
613 rdma_addr_get_sgid(dev_addr,
614 (union ib_gid *) &resp->ib_route[0].sgid);
615 resp->ib_route[0].pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
618 ib_copy_path_rec_to_user(&resp->ib_route[1],
619 &route->path_rec[1]);
622 ib_copy_path_rec_to_user(&resp->ib_route[0],
623 &route->path_rec[0]);
630 static void ucma_copy_iboe_route(struct rdma_ucm_query_route_resp *resp,
631 struct rdma_route *route)
633 struct rdma_dev_addr *dev_addr;
634 struct net_device *dev;
637 resp->num_paths = route->num_paths;
638 switch (route->num_paths) {
640 dev_addr = &route->addr.dev_addr;
641 dev = dev_get_by_index(&init_net, dev_addr->bound_dev_if);
643 vid = rdma_vlan_dev_vlan_id(dev);
647 iboe_mac_vlan_to_ll((union ib_gid *) &resp->ib_route[0].dgid,
648 dev_addr->dst_dev_addr, vid);
649 iboe_addr_get_sgid(dev_addr,
650 (union ib_gid *) &resp->ib_route[0].sgid);
651 resp->ib_route[0].pkey = cpu_to_be16(0xffff);
654 ib_copy_path_rec_to_user(&resp->ib_route[1],
655 &route->path_rec[1]);
658 ib_copy_path_rec_to_user(&resp->ib_route[0],
659 &route->path_rec[0]);
666 static void ucma_copy_iw_route(struct rdma_ucm_query_route_resp *resp,
667 struct rdma_route *route)
669 struct rdma_dev_addr *dev_addr;
671 dev_addr = &route->addr.dev_addr;
672 rdma_addr_get_dgid(dev_addr, (union ib_gid *) &resp->ib_route[0].dgid);
673 rdma_addr_get_sgid(dev_addr, (union ib_gid *) &resp->ib_route[0].sgid);
676 static ssize_t ucma_query_route(struct ucma_file *file,
677 const char __user *inbuf,
678 int in_len, int out_len)
680 struct rdma_ucm_query_route cmd;
681 struct rdma_ucm_query_route_resp resp;
682 struct ucma_context *ctx;
683 struct sockaddr *addr;
686 if (out_len < sizeof(resp))
689 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
692 ctx = ucma_get_ctx(file, cmd.id);
696 memset(&resp, 0, sizeof resp);
697 addr = (struct sockaddr *) &ctx->cm_id->route.addr.src_addr;
698 memcpy(&resp.src_addr, addr, addr->sa_family == AF_INET ?
699 sizeof(struct sockaddr_in) :
700 sizeof(struct sockaddr_in6));
701 addr = (struct sockaddr *) &ctx->cm_id->route.addr.dst_addr;
702 memcpy(&resp.dst_addr, addr, addr->sa_family == AF_INET ?
703 sizeof(struct sockaddr_in) :
704 sizeof(struct sockaddr_in6));
705 if (!ctx->cm_id->device)
708 resp.node_guid = (__force __u64) ctx->cm_id->device->node_guid;
709 resp.port_num = ctx->cm_id->port_num;
710 switch (rdma_node_get_transport(ctx->cm_id->device->node_type)) {
711 case RDMA_TRANSPORT_IB:
712 switch (rdma_port_get_link_layer(ctx->cm_id->device,
713 ctx->cm_id->port_num)) {
714 case IB_LINK_LAYER_INFINIBAND:
715 ucma_copy_ib_route(&resp, &ctx->cm_id->route);
717 case IB_LINK_LAYER_ETHERNET:
718 ucma_copy_iboe_route(&resp, &ctx->cm_id->route);
724 case RDMA_TRANSPORT_IWARP:
725 ucma_copy_iw_route(&resp, &ctx->cm_id->route);
732 if (copy_to_user((void __user *)(unsigned long)cmd.response,
733 &resp, sizeof(resp)))
740 static void ucma_copy_conn_param(struct rdma_conn_param *dst,
741 struct rdma_ucm_conn_param *src)
743 dst->private_data = src->private_data;
744 dst->private_data_len = src->private_data_len;
745 dst->responder_resources =src->responder_resources;
746 dst->initiator_depth = src->initiator_depth;
747 dst->flow_control = src->flow_control;
748 dst->retry_count = src->retry_count;
749 dst->rnr_retry_count = src->rnr_retry_count;
751 dst->qp_num = src->qp_num;
754 static ssize_t ucma_connect(struct ucma_file *file, const char __user *inbuf,
755 int in_len, int out_len)
757 struct rdma_ucm_connect cmd;
758 struct rdma_conn_param conn_param;
759 struct ucma_context *ctx;
762 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
765 if (!cmd.conn_param.valid)
768 ctx = ucma_get_ctx(file, cmd.id);
772 ucma_copy_conn_param(&conn_param, &cmd.conn_param);
773 ret = rdma_connect(ctx->cm_id, &conn_param);
778 static ssize_t ucma_listen(struct ucma_file *file, const char __user *inbuf,
779 int in_len, int out_len)
781 struct rdma_ucm_listen cmd;
782 struct ucma_context *ctx;
785 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
788 ctx = ucma_get_ctx(file, cmd.id);
792 ctx->backlog = cmd.backlog > 0 && cmd.backlog < max_backlog ?
793 cmd.backlog : max_backlog;
794 ret = rdma_listen(ctx->cm_id, ctx->backlog);
799 static ssize_t ucma_accept(struct ucma_file *file, const char __user *inbuf,
800 int in_len, int out_len)
802 struct rdma_ucm_accept cmd;
803 struct rdma_conn_param conn_param;
804 struct ucma_context *ctx;
807 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
810 ctx = ucma_get_ctx(file, cmd.id);
814 if (cmd.conn_param.valid) {
816 ucma_copy_conn_param(&conn_param, &cmd.conn_param);
817 ret = rdma_accept(ctx->cm_id, &conn_param);
819 ret = rdma_accept(ctx->cm_id, NULL);
825 static ssize_t ucma_reject(struct ucma_file *file, const char __user *inbuf,
826 int in_len, int out_len)
828 struct rdma_ucm_reject cmd;
829 struct ucma_context *ctx;
832 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
835 ctx = ucma_get_ctx(file, cmd.id);
839 ret = rdma_reject(ctx->cm_id, cmd.private_data, cmd.private_data_len);
844 static ssize_t ucma_disconnect(struct ucma_file *file, const char __user *inbuf,
845 int in_len, int out_len)
847 struct rdma_ucm_disconnect cmd;
848 struct ucma_context *ctx;
851 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
854 ctx = ucma_get_ctx(file, cmd.id);
858 ret = rdma_disconnect(ctx->cm_id);
863 static ssize_t ucma_init_qp_attr(struct ucma_file *file,
864 const char __user *inbuf,
865 int in_len, int out_len)
867 struct rdma_ucm_init_qp_attr cmd;
868 struct ib_uverbs_qp_attr resp;
869 struct ucma_context *ctx;
870 struct ib_qp_attr qp_attr;
873 if (out_len < sizeof(resp))
876 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
879 if (cmd.qp_state > IB_QPS_ERR)
882 ctx = ucma_get_ctx(file, cmd.id);
886 resp.qp_attr_mask = 0;
887 memset(&qp_attr, 0, sizeof qp_attr);
888 qp_attr.qp_state = cmd.qp_state;
889 ret = rdma_init_qp_attr(ctx->cm_id, &qp_attr, &resp.qp_attr_mask);
893 ib_copy_qp_attr_to_user(&resp, &qp_attr);
894 if (copy_to_user((void __user *)(unsigned long)cmd.response,
895 &resp, sizeof(resp)))
903 static int ucma_set_option_id(struct ucma_context *ctx, int optname,
904 void *optval, size_t optlen)
909 case RDMA_OPTION_ID_TOS:
910 if (optlen != sizeof(u8)) {
914 rdma_set_service_type(ctx->cm_id, *((u8 *) optval));
916 case RDMA_OPTION_ID_REUSEADDR:
917 if (optlen != sizeof(int)) {
921 ret = rdma_set_reuseaddr(ctx->cm_id, *((int *) optval) ? 1 : 0);
930 static int ucma_set_ib_path(struct ucma_context *ctx,
931 struct ib_path_rec_data *path_data, size_t optlen)
933 struct ib_sa_path_rec sa_path;
934 struct rdma_cm_event event;
937 if (optlen % sizeof(*path_data))
940 for (; optlen; optlen -= sizeof(*path_data), path_data++) {
941 if (path_data->flags == (IB_PATH_GMP | IB_PATH_PRIMARY |
942 IB_PATH_BIDIRECTIONAL))
949 ib_sa_unpack_path(path_data->path_rec, &sa_path);
950 ret = rdma_set_ib_paths(ctx->cm_id, &sa_path, 1);
954 memset(&event, 0, sizeof event);
955 event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
956 return ucma_event_handler(ctx->cm_id, &event);
959 static int ucma_set_option_ib(struct ucma_context *ctx, int optname,
960 void *optval, size_t optlen)
965 case RDMA_OPTION_IB_PATH:
966 ret = ucma_set_ib_path(ctx, optval, optlen);
975 static int ucma_set_option_level(struct ucma_context *ctx, int level,
976 int optname, void *optval, size_t optlen)
982 ret = ucma_set_option_id(ctx, optname, optval, optlen);
985 ret = ucma_set_option_ib(ctx, optname, optval, optlen);
994 static ssize_t ucma_set_option(struct ucma_file *file, const char __user *inbuf,
995 int in_len, int out_len)
997 struct rdma_ucm_set_option cmd;
998 struct ucma_context *ctx;
1002 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1005 ctx = ucma_get_ctx(file, cmd.id);
1007 return PTR_ERR(ctx);
1009 if (unlikely(cmd.optval > KMALLOC_MAX_SIZE))
1012 optval = kmalloc(cmd.optlen, GFP_KERNEL);
1018 if (copy_from_user(optval, (void __user *) (unsigned long) cmd.optval,
1024 ret = ucma_set_option_level(ctx, cmd.level, cmd.optname, optval,
1033 static ssize_t ucma_notify(struct ucma_file *file, const char __user *inbuf,
1034 int in_len, int out_len)
1036 struct rdma_ucm_notify cmd;
1037 struct ucma_context *ctx;
1040 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1043 ctx = ucma_get_ctx(file, cmd.id);
1045 return PTR_ERR(ctx);
1047 ret = rdma_notify(ctx->cm_id, (enum ib_event_type) cmd.event);
1052 static ssize_t ucma_join_multicast(struct ucma_file *file,
1053 const char __user *inbuf,
1054 int in_len, int out_len)
1056 struct rdma_ucm_join_mcast cmd;
1057 struct rdma_ucm_create_id_resp resp;
1058 struct ucma_context *ctx;
1059 struct ucma_multicast *mc;
1062 if (out_len < sizeof(resp))
1065 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1068 ctx = ucma_get_ctx(file, cmd.id);
1070 return PTR_ERR(ctx);
1072 mutex_lock(&file->mut);
1073 mc = ucma_alloc_multicast(ctx);
1080 memcpy(&mc->addr, &cmd.addr, sizeof cmd.addr);
1081 ret = rdma_join_multicast(ctx->cm_id, (struct sockaddr *) &mc->addr, mc);
1086 if (copy_to_user((void __user *)(unsigned long)cmd.response,
1087 &resp, sizeof(resp))) {
1092 mutex_unlock(&file->mut);
1097 rdma_leave_multicast(ctx->cm_id, (struct sockaddr *) &mc->addr);
1098 ucma_cleanup_mc_events(mc);
1101 idr_remove(&multicast_idr, mc->id);
1103 list_del(&mc->list);
1106 mutex_unlock(&file->mut);
1111 static ssize_t ucma_leave_multicast(struct ucma_file *file,
1112 const char __user *inbuf,
1113 int in_len, int out_len)
1115 struct rdma_ucm_destroy_id cmd;
1116 struct rdma_ucm_destroy_id_resp resp;
1117 struct ucma_multicast *mc;
1120 if (out_len < sizeof(resp))
1123 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1127 mc = idr_find(&multicast_idr, cmd.id);
1129 mc = ERR_PTR(-ENOENT);
1130 else if (mc->ctx->file != file)
1131 mc = ERR_PTR(-EINVAL);
1133 idr_remove(&multicast_idr, mc->id);
1134 atomic_inc(&mc->ctx->ref);
1143 rdma_leave_multicast(mc->ctx->cm_id, (struct sockaddr *) &mc->addr);
1144 mutex_lock(&mc->ctx->file->mut);
1145 ucma_cleanup_mc_events(mc);
1146 list_del(&mc->list);
1147 mutex_unlock(&mc->ctx->file->mut);
1149 ucma_put_ctx(mc->ctx);
1150 resp.events_reported = mc->events_reported;
1153 if (copy_to_user((void __user *)(unsigned long)cmd.response,
1154 &resp, sizeof(resp)))
1160 static void ucma_lock_files(struct ucma_file *file1, struct ucma_file *file2)
1162 /* Acquire mutex's based on pointer comparison to prevent deadlock. */
1163 if (file1 < file2) {
1164 mutex_lock(&file1->mut);
1165 mutex_lock(&file2->mut);
1167 mutex_lock(&file2->mut);
1168 mutex_lock(&file1->mut);
1172 static void ucma_unlock_files(struct ucma_file *file1, struct ucma_file *file2)
1174 if (file1 < file2) {
1175 mutex_unlock(&file2->mut);
1176 mutex_unlock(&file1->mut);
1178 mutex_unlock(&file1->mut);
1179 mutex_unlock(&file2->mut);
1183 static void ucma_move_events(struct ucma_context *ctx, struct ucma_file *file)
1185 struct ucma_event *uevent, *tmp;
1187 list_for_each_entry_safe(uevent, tmp, &ctx->file->event_list, list)
1188 if (uevent->ctx == ctx)
1189 list_move_tail(&uevent->list, &file->event_list);
1192 static ssize_t ucma_migrate_id(struct ucma_file *new_file,
1193 const char __user *inbuf,
1194 int in_len, int out_len)
1196 struct rdma_ucm_migrate_id cmd;
1197 struct rdma_ucm_migrate_resp resp;
1198 struct ucma_context *ctx;
1200 struct ucma_file *cur_file;
1203 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1206 /* Get current fd to protect against it being closed */
1207 filp = fget(cmd.fd);
1211 /* Validate current fd and prevent destruction of id. */
1212 ctx = ucma_get_ctx(filp->private_data, cmd.id);
1218 cur_file = ctx->file;
1219 if (cur_file == new_file) {
1220 resp.events_reported = ctx->events_reported;
1225 * Migrate events between fd's, maintaining order, and avoiding new
1226 * events being added before existing events.
1228 ucma_lock_files(cur_file, new_file);
1231 list_move_tail(&ctx->list, &new_file->ctx_list);
1232 ucma_move_events(ctx, new_file);
1233 ctx->file = new_file;
1234 resp.events_reported = ctx->events_reported;
1237 ucma_unlock_files(cur_file, new_file);
1240 if (copy_to_user((void __user *)(unsigned long)cmd.response,
1241 &resp, sizeof(resp)))
1250 static ssize_t (*ucma_cmd_table[])(struct ucma_file *file,
1251 const char __user *inbuf,
1252 int in_len, int out_len) = {
1253 [RDMA_USER_CM_CMD_CREATE_ID] = ucma_create_id,
1254 [RDMA_USER_CM_CMD_DESTROY_ID] = ucma_destroy_id,
1255 [RDMA_USER_CM_CMD_BIND_ADDR] = ucma_bind_addr,
1256 [RDMA_USER_CM_CMD_RESOLVE_ADDR] = ucma_resolve_addr,
1257 [RDMA_USER_CM_CMD_RESOLVE_ROUTE]= ucma_resolve_route,
1258 [RDMA_USER_CM_CMD_QUERY_ROUTE] = ucma_query_route,
1259 [RDMA_USER_CM_CMD_CONNECT] = ucma_connect,
1260 [RDMA_USER_CM_CMD_LISTEN] = ucma_listen,
1261 [RDMA_USER_CM_CMD_ACCEPT] = ucma_accept,
1262 [RDMA_USER_CM_CMD_REJECT] = ucma_reject,
1263 [RDMA_USER_CM_CMD_DISCONNECT] = ucma_disconnect,
1264 [RDMA_USER_CM_CMD_INIT_QP_ATTR] = ucma_init_qp_attr,
1265 [RDMA_USER_CM_CMD_GET_EVENT] = ucma_get_event,
1266 [RDMA_USER_CM_CMD_GET_OPTION] = NULL,
1267 [RDMA_USER_CM_CMD_SET_OPTION] = ucma_set_option,
1268 [RDMA_USER_CM_CMD_NOTIFY] = ucma_notify,
1269 [RDMA_USER_CM_CMD_JOIN_MCAST] = ucma_join_multicast,
1270 [RDMA_USER_CM_CMD_LEAVE_MCAST] = ucma_leave_multicast,
1271 [RDMA_USER_CM_CMD_MIGRATE_ID] = ucma_migrate_id
1274 static ssize_t ucma_write(struct file *filp, const char __user *buf,
1275 size_t len, loff_t *pos)
1277 struct ucma_file *file = filp->private_data;
1278 struct rdma_ucm_cmd_hdr hdr;
1281 if (WARN_ON_ONCE(!ib_safe_file_access(filp)))
1284 if (len < sizeof(hdr))
1287 if (copy_from_user(&hdr, buf, sizeof(hdr)))
1290 if (hdr.cmd >= ARRAY_SIZE(ucma_cmd_table))
1293 if (hdr.in + sizeof(hdr) > len)
1296 if (!ucma_cmd_table[hdr.cmd])
1299 ret = ucma_cmd_table[hdr.cmd](file, buf + sizeof(hdr), hdr.in, hdr.out);
1306 static unsigned int ucma_poll(struct file *filp, struct poll_table_struct *wait)
1308 struct ucma_file *file = filp->private_data;
1309 unsigned int mask = 0;
1311 poll_wait(filp, &file->poll_wait, wait);
1313 if (!list_empty(&file->event_list))
1314 mask = POLLIN | POLLRDNORM;
1320 * ucma_open() does not need the BKL:
1322 * - no global state is referred to;
1323 * - there is no ioctl method to race against;
1324 * - no further module initialization is required for open to work
1325 * after the device is registered.
1327 static int ucma_open(struct inode *inode, struct file *filp)
1329 struct ucma_file *file;
1331 file = kmalloc(sizeof *file, GFP_KERNEL);
1335 INIT_LIST_HEAD(&file->event_list);
1336 INIT_LIST_HEAD(&file->ctx_list);
1337 init_waitqueue_head(&file->poll_wait);
1338 mutex_init(&file->mut);
1340 filp->private_data = file;
1343 return nonseekable_open(inode, filp);
1346 static int ucma_close(struct inode *inode, struct file *filp)
1348 struct ucma_file *file = filp->private_data;
1349 struct ucma_context *ctx, *tmp;
1351 mutex_lock(&file->mut);
1352 list_for_each_entry_safe(ctx, tmp, &file->ctx_list, list) {
1353 mutex_unlock(&file->mut);
1356 idr_remove(&ctx_idr, ctx->id);
1360 mutex_lock(&file->mut);
1362 mutex_unlock(&file->mut);
1367 static const struct file_operations ucma_fops = {
1368 .owner = THIS_MODULE,
1370 .release = ucma_close,
1371 .write = ucma_write,
1373 .llseek = no_llseek,
1376 static struct miscdevice ucma_misc = {
1377 .minor = MISC_DYNAMIC_MINOR,
1379 .nodename = "infiniband/rdma_cm",
1384 static ssize_t show_abi_version(struct device *dev,
1385 struct device_attribute *attr,
1388 return sprintf(buf, "%d\n", RDMA_USER_CM_ABI_VERSION);
1390 static DEVICE_ATTR(abi_version, S_IRUGO, show_abi_version, NULL);
1392 static int __init ucma_init(void)
1396 ret = misc_register(&ucma_misc);
1400 ret = device_create_file(ucma_misc.this_device, &dev_attr_abi_version);
1402 printk(KERN_ERR "rdma_ucm: couldn't create abi_version attr\n");
1406 ucma_ctl_table_hdr = register_sysctl_paths(ucma_ctl_path, ucma_ctl_table);
1407 if (!ucma_ctl_table_hdr) {
1408 printk(KERN_ERR "rdma_ucm: couldn't register sysctl paths\n");
1414 device_remove_file(ucma_misc.this_device, &dev_attr_abi_version);
1416 misc_deregister(&ucma_misc);
1420 static void __exit ucma_cleanup(void)
1422 unregister_sysctl_table(ucma_ctl_table_hdr);
1423 device_remove_file(ucma_misc.this_device, &dev_attr_abi_version);
1424 misc_deregister(&ucma_misc);
1425 idr_destroy(&ctx_idr);
1428 module_init(ucma_init);
1429 module_exit(ucma_cleanup);