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/module.h>
45 #include <sys/filio.h>
47 #include <rdma/rdma_user_cm.h>
48 #include <rdma/ib_marshall.h>
49 #include <rdma/rdma_cm.h>
50 #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;
61 struct list_head ctx_list;
62 struct list_head event_list;
63 wait_queue_head_t poll_wait;
68 struct completion comp;
73 struct ucma_file *file;
74 struct rdma_cm_id *cm_id;
77 struct list_head list;
78 struct list_head mc_list;
81 struct ucma_multicast {
82 struct ucma_context *ctx;
87 struct list_head list;
88 struct sockaddr_storage addr;
92 struct ucma_context *ctx;
93 struct ucma_multicast *mc;
94 struct list_head list;
95 struct rdma_cm_id *cm_id;
96 struct rdma_ucm_event_resp resp;
99 static DEFINE_MUTEX(mut);
100 static DEFINE_IDR(ctx_idr);
101 static DEFINE_IDR(multicast_idr);
103 static inline struct ucma_context *_ucma_find_context(int id,
104 struct ucma_file *file)
106 struct ucma_context *ctx;
108 ctx = idr_find(&ctx_idr, id);
110 ctx = ERR_PTR(-ENOENT);
111 else if (ctx->file != file)
112 ctx = ERR_PTR(-EINVAL);
116 static struct ucma_context *ucma_get_ctx(struct ucma_file *file, int id)
118 struct ucma_context *ctx;
121 ctx = _ucma_find_context(id, file);
123 atomic_inc(&ctx->ref);
128 static void ucma_put_ctx(struct ucma_context *ctx)
130 if (atomic_dec_and_test(&ctx->ref))
131 complete(&ctx->comp);
134 static struct ucma_context *ucma_alloc_ctx(struct ucma_file *file)
136 struct ucma_context *ctx;
139 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
143 atomic_set(&ctx->ref, 1);
144 init_completion(&ctx->comp);
145 INIT_LIST_HEAD(&ctx->mc_list);
149 ret = idr_pre_get(&ctx_idr, GFP_KERNEL);
154 ret = idr_get_new(&ctx_idr, ctx, &ctx->id);
156 } while (ret == -EAGAIN);
161 list_add_tail(&ctx->list, &file->ctx_list);
169 static struct ucma_multicast* ucma_alloc_multicast(struct ucma_context *ctx)
171 struct ucma_multicast *mc;
174 mc = kzalloc(sizeof(*mc), GFP_KERNEL);
179 ret = idr_pre_get(&multicast_idr, GFP_KERNEL);
184 ret = idr_get_new(&multicast_idr, mc, &mc->id);
186 } while (ret == -EAGAIN);
192 list_add_tail(&mc->list, &ctx->mc_list);
200 static void ucma_copy_conn_event(struct rdma_ucm_conn_param *dst,
201 struct rdma_conn_param *src)
203 if (src->private_data_len)
204 memcpy(dst->private_data, src->private_data,
205 src->private_data_len);
206 dst->private_data_len = src->private_data_len;
207 dst->responder_resources =src->responder_resources;
208 dst->initiator_depth = src->initiator_depth;
209 dst->flow_control = src->flow_control;
210 dst->retry_count = src->retry_count;
211 dst->rnr_retry_count = src->rnr_retry_count;
213 dst->qp_num = src->qp_num;
216 static void ucma_copy_ud_event(struct rdma_ucm_ud_param *dst,
217 struct rdma_ud_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 ib_copy_ah_attr_to_user(&dst->ah_attr, &src->ah_attr);
224 dst->qp_num = src->qp_num;
225 dst->qkey = src->qkey;
228 static void ucma_set_event_context(struct ucma_context *ctx,
229 struct rdma_cm_event *event,
230 struct ucma_event *uevent)
233 switch (event->event) {
234 case RDMA_CM_EVENT_MULTICAST_JOIN:
235 case RDMA_CM_EVENT_MULTICAST_ERROR:
236 uevent->mc = (struct ucma_multicast *)
237 event->param.ud.private_data;
238 uevent->resp.uid = uevent->mc->uid;
239 uevent->resp.id = uevent->mc->id;
242 uevent->resp.uid = ctx->uid;
243 uevent->resp.id = ctx->id;
248 static int ucma_event_handler(struct rdma_cm_id *cm_id,
249 struct rdma_cm_event *event)
251 struct ucma_event *uevent;
252 struct ucma_context *ctx = cm_id->context;
255 uevent = kzalloc(sizeof(*uevent), GFP_KERNEL);
257 return event->event == RDMA_CM_EVENT_CONNECT_REQUEST;
259 mutex_lock(&ctx->file->mut);
260 uevent->cm_id = cm_id;
261 ucma_set_event_context(ctx, event, uevent);
262 uevent->resp.event = event->event;
263 uevent->resp.status = event->status;
264 if (cm_id->qp_type == IB_QPT_UD)
265 ucma_copy_ud_event(&uevent->resp.param.ud, &event->param.ud);
267 ucma_copy_conn_event(&uevent->resp.param.conn,
270 if (event->event == RDMA_CM_EVENT_CONNECT_REQUEST) {
277 } else if (!ctx->uid) {
279 * We ignore events for new connections until userspace has set
280 * their context. This can only happen if an error occurs on a
281 * new connection before the user accepts it. This is okay,
282 * since the accept will just fail later.
288 list_add_tail(&uevent->list, &ctx->file->event_list);
289 wake_up_interruptible(&ctx->file->poll_wait);
291 selwakeup(&ctx->file->filp->f_selinfo);
293 mutex_unlock(&ctx->file->mut);
297 static ssize_t ucma_get_event(struct ucma_file *file, const char __user *inbuf,
298 int in_len, int out_len)
300 struct ucma_context *ctx;
301 struct rdma_ucm_get_event cmd;
302 struct ucma_event *uevent;
305 if (out_len < sizeof uevent->resp)
308 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
311 mutex_lock(&file->mut);
312 while (list_empty(&file->event_list)) {
313 mutex_unlock(&file->mut);
315 if (file->filp->f_flags & O_NONBLOCK)
318 if (wait_event_interruptible(file->poll_wait,
319 !list_empty(&file->event_list)))
322 mutex_lock(&file->mut);
325 uevent = list_entry(file->event_list.next, struct ucma_event, list);
327 if (uevent->resp.event == RDMA_CM_EVENT_CONNECT_REQUEST) {
328 ctx = ucma_alloc_ctx(file);
333 uevent->ctx->backlog++;
334 ctx->cm_id = uevent->cm_id;
335 ctx->cm_id->context = ctx;
336 uevent->resp.id = ctx->id;
337 ctx->cm_id->ucontext = ctx;
340 if (copy_to_user((void __user *)(unsigned long)cmd.response,
341 &uevent->resp, sizeof uevent->resp)) {
346 list_del(&uevent->list);
347 uevent->ctx->events_reported++;
349 uevent->mc->events_reported++;
352 mutex_unlock(&file->mut);
356 static int ucma_get_qp_type(struct rdma_ucm_create_id *cmd, enum ib_qp_type *qp_type)
360 *qp_type = IB_QPT_RC;
364 *qp_type = IB_QPT_UD;
367 *qp_type = cmd->qp_type;
374 static ssize_t ucma_create_id(struct ucma_file *file, const char __user *inbuf,
375 int in_len, int out_len)
377 struct rdma_ucm_create_id cmd;
378 struct rdma_ucm_create_id_resp resp;
379 struct ucma_context *ctx;
380 enum ib_qp_type qp_type;
383 if (out_len < sizeof(resp))
386 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
389 ret = ucma_get_qp_type(&cmd, &qp_type);
393 mutex_lock(&file->mut);
394 ctx = ucma_alloc_ctx(file);
395 mutex_unlock(&file->mut);
400 ctx->cm_id = rdma_create_id(ucma_event_handler, ctx, cmd.ps, qp_type);
401 if (IS_ERR(ctx->cm_id)) {
402 ret = PTR_ERR(ctx->cm_id);
405 ctx->cm_id->ucontext = ctx;
408 if (copy_to_user((void __user *)(unsigned long)cmd.response,
409 &resp, sizeof(resp))) {
416 rdma_destroy_id(ctx->cm_id);
419 idr_remove(&ctx_idr, ctx->id);
425 static void ucma_cleanup_multicast(struct ucma_context *ctx)
427 struct ucma_multicast *mc, *tmp;
430 list_for_each_entry_safe(mc, tmp, &ctx->mc_list, list) {
432 idr_remove(&multicast_idr, mc->id);
438 static void ucma_cleanup_mc_events(struct ucma_multicast *mc)
440 struct ucma_event *uevent, *tmp;
442 list_for_each_entry_safe(uevent, tmp, &mc->ctx->file->event_list, list) {
443 if (uevent->mc != mc)
446 list_del(&uevent->list);
452 * We cannot hold file->mut when calling rdma_destroy_id() or we can
453 * deadlock. We also acquire file->mut in ucma_event_handler(), and
454 * rdma_destroy_id() will wait until all callbacks have completed.
456 static int ucma_free_ctx(struct ucma_context *ctx)
459 struct ucma_event *uevent, *tmp;
462 /* No new events will be generated after destroying the id. */
463 rdma_destroy_id(ctx->cm_id);
465 ucma_cleanup_multicast(ctx);
467 /* Cleanup events not yet reported to the user. */
468 mutex_lock(&ctx->file->mut);
469 list_for_each_entry_safe(uevent, tmp, &ctx->file->event_list, list) {
470 if (uevent->ctx == ctx)
471 list_move_tail(&uevent->list, &list);
473 list_del(&ctx->list);
474 mutex_unlock(&ctx->file->mut);
476 list_for_each_entry_safe(uevent, tmp, &list, list) {
477 list_del(&uevent->list);
478 if (uevent->resp.event == RDMA_CM_EVENT_CONNECT_REQUEST)
479 rdma_destroy_id(uevent->cm_id);
483 events_reported = ctx->events_reported;
485 return events_reported;
488 static ssize_t ucma_destroy_id(struct ucma_file *file, const char __user *inbuf,
489 int in_len, int out_len)
491 struct rdma_ucm_destroy_id cmd;
492 struct rdma_ucm_destroy_id_resp resp;
493 struct ucma_context *ctx;
496 if (out_len < sizeof(resp))
499 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
503 ctx = _ucma_find_context(cmd.id, file);
505 idr_remove(&ctx_idr, ctx->id);
512 wait_for_completion(&ctx->comp);
513 resp.events_reported = ucma_free_ctx(ctx);
515 if (copy_to_user((void __user *)(unsigned long)cmd.response,
516 &resp, sizeof(resp)))
522 static ssize_t ucma_bind_addr(struct ucma_file *file, const char __user *inbuf,
523 int in_len, int out_len)
525 struct rdma_ucm_bind_addr cmd;
526 struct ucma_context *ctx;
529 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
532 ctx = ucma_get_ctx(file, cmd.id);
536 ret = rdma_bind_addr(ctx->cm_id, (struct sockaddr *) &cmd.addr);
541 static ssize_t ucma_resolve_addr(struct ucma_file *file,
542 const char __user *inbuf,
543 int in_len, int out_len)
545 struct rdma_ucm_resolve_addr cmd;
546 struct ucma_context *ctx;
549 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
552 ctx = ucma_get_ctx(file, cmd.id);
556 ret = rdma_resolve_addr(ctx->cm_id, (struct sockaddr *) &cmd.src_addr,
557 (struct sockaddr *) &cmd.dst_addr,
563 static ssize_t ucma_resolve_route(struct ucma_file *file,
564 const char __user *inbuf,
565 int in_len, int out_len)
567 struct rdma_ucm_resolve_route cmd;
568 struct ucma_context *ctx;
571 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
574 ctx = ucma_get_ctx(file, cmd.id);
578 ret = rdma_resolve_route(ctx->cm_id, cmd.timeout_ms);
583 static void ucma_copy_ib_route(struct rdma_ucm_query_route_resp *resp,
584 struct rdma_route *route)
586 struct rdma_dev_addr *dev_addr;
588 resp->num_paths = route->num_paths;
589 switch (route->num_paths) {
591 dev_addr = &route->addr.dev_addr;
592 rdma_addr_get_dgid(dev_addr,
593 (union ib_gid *) &resp->ib_route[0].dgid);
594 rdma_addr_get_sgid(dev_addr,
595 (union ib_gid *) &resp->ib_route[0].sgid);
596 resp->ib_route[0].pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
599 ib_copy_path_rec_to_user(&resp->ib_route[1],
600 &route->path_rec[1]);
603 ib_copy_path_rec_to_user(&resp->ib_route[0],
604 &route->path_rec[0]);
611 static void ucma_copy_iboe_route(struct rdma_ucm_query_route_resp *resp,
612 struct rdma_route *route)
615 resp->num_paths = route->num_paths;
616 switch (route->num_paths) {
618 rdma_ip2gid((struct sockaddr *)&route->addr.dst_addr,
619 (union ib_gid *)&resp->ib_route[0].dgid);
620 rdma_ip2gid((struct sockaddr *)&route->addr.src_addr,
621 (union ib_gid *)&resp->ib_route[0].sgid);
622 resp->ib_route[0].pkey = cpu_to_be16(0xffff);
625 ib_copy_path_rec_to_user(&resp->ib_route[1],
626 &route->path_rec[1]);
629 ib_copy_path_rec_to_user(&resp->ib_route[0],
630 &route->path_rec[0]);
637 static void ucma_copy_iw_route(struct rdma_ucm_query_route_resp *resp,
638 struct rdma_route *route)
640 struct rdma_dev_addr *dev_addr;
642 dev_addr = &route->addr.dev_addr;
643 rdma_addr_get_dgid(dev_addr, (union ib_gid *) &resp->ib_route[0].dgid);
644 rdma_addr_get_sgid(dev_addr, (union ib_gid *) &resp->ib_route[0].sgid);
647 static ssize_t ucma_query_route(struct ucma_file *file,
648 const char __user *inbuf,
649 int in_len, int out_len)
651 struct rdma_ucm_query_route cmd;
652 struct rdma_ucm_query_route_resp resp;
653 struct ucma_context *ctx;
654 struct sockaddr *addr;
657 if (out_len < sizeof(resp))
660 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
663 ctx = ucma_get_ctx(file, cmd.id);
667 memset(&resp, 0, sizeof resp);
668 addr = (struct sockaddr *) &ctx->cm_id->route.addr.src_addr;
669 memcpy(&resp.src_addr, addr, addr->sa_family == AF_INET ?
670 sizeof(struct sockaddr_in) :
671 sizeof(struct sockaddr_in6));
672 addr = (struct sockaddr *) &ctx->cm_id->route.addr.dst_addr;
673 memcpy(&resp.dst_addr, addr, addr->sa_family == AF_INET ?
674 sizeof(struct sockaddr_in) :
675 sizeof(struct sockaddr_in6));
676 if (!ctx->cm_id->device)
679 resp.node_guid = (__force __u64) ctx->cm_id->device->node_guid;
680 resp.port_num = ctx->cm_id->port_num;
681 switch (rdma_node_get_transport(ctx->cm_id->device->node_type)) {
682 case RDMA_TRANSPORT_IB:
683 switch (rdma_port_get_link_layer(ctx->cm_id->device,
684 ctx->cm_id->port_num)) {
685 case IB_LINK_LAYER_INFINIBAND:
686 ucma_copy_ib_route(&resp, &ctx->cm_id->route);
688 case IB_LINK_LAYER_ETHERNET:
689 ucma_copy_iboe_route(&resp, &ctx->cm_id->route);
695 case RDMA_TRANSPORT_IWARP:
696 ucma_copy_iw_route(&resp, &ctx->cm_id->route);
703 if (copy_to_user((void __user *)(unsigned long)cmd.response,
704 &resp, sizeof(resp)))
711 static void ucma_copy_conn_param(struct rdma_conn_param *dst,
712 struct rdma_ucm_conn_param *src)
714 dst->private_data = src->private_data;
715 dst->private_data_len = src->private_data_len;
716 dst->responder_resources =src->responder_resources;
717 dst->initiator_depth = src->initiator_depth;
718 dst->flow_control = src->flow_control;
719 dst->retry_count = src->retry_count;
720 dst->rnr_retry_count = src->rnr_retry_count;
722 dst->qp_num = src->qp_num;
725 static ssize_t ucma_connect(struct ucma_file *file, const char __user *inbuf,
726 int in_len, int out_len)
728 struct rdma_ucm_connect cmd;
729 struct rdma_conn_param conn_param;
730 struct ucma_context *ctx;
733 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
736 if (!cmd.conn_param.valid)
739 ctx = ucma_get_ctx(file, cmd.id);
743 ucma_copy_conn_param(&conn_param, &cmd.conn_param);
744 ret = rdma_connect(ctx->cm_id, &conn_param);
749 static ssize_t ucma_listen(struct ucma_file *file, const char __user *inbuf,
750 int in_len, int out_len)
752 struct rdma_ucm_listen cmd;
753 struct ucma_context *ctx;
756 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
759 ctx = ucma_get_ctx(file, cmd.id);
763 ctx->backlog = cmd.backlog > 0 && cmd.backlog < max_backlog ?
764 cmd.backlog : max_backlog;
765 ret = rdma_listen(ctx->cm_id, ctx->backlog);
770 static ssize_t ucma_accept(struct ucma_file *file, const char __user *inbuf,
771 int in_len, int out_len)
773 struct rdma_ucm_accept cmd;
774 struct rdma_conn_param conn_param;
775 struct ucma_context *ctx;
778 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
781 ctx = ucma_get_ctx(file, cmd.id);
785 if (cmd.conn_param.valid) {
786 ucma_copy_conn_param(&conn_param, &cmd.conn_param);
787 mutex_lock(&file->mut);
788 ret = rdma_accept(ctx->cm_id, &conn_param);
791 mutex_unlock(&file->mut);
793 ret = rdma_accept(ctx->cm_id, NULL);
799 static ssize_t ucma_reject(struct ucma_file *file, const char __user *inbuf,
800 int in_len, int out_len)
802 struct rdma_ucm_reject cmd;
803 struct ucma_context *ctx;
806 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
809 ctx = ucma_get_ctx(file, cmd.id);
813 ret = rdma_reject(ctx->cm_id, cmd.private_data, cmd.private_data_len);
818 static ssize_t ucma_disconnect(struct ucma_file *file, const char __user *inbuf,
819 int in_len, int out_len)
821 struct rdma_ucm_disconnect cmd;
822 struct ucma_context *ctx;
825 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
828 ctx = ucma_get_ctx(file, cmd.id);
832 ret = rdma_disconnect(ctx->cm_id);
837 static ssize_t ucma_init_qp_attr(struct ucma_file *file,
838 const char __user *inbuf,
839 int in_len, int out_len)
841 struct rdma_ucm_init_qp_attr cmd;
842 struct ib_uverbs_qp_attr resp;
843 struct ucma_context *ctx;
844 struct ib_qp_attr qp_attr;
847 if (out_len < sizeof(resp))
850 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
853 ctx = ucma_get_ctx(file, cmd.id);
857 resp.qp_attr_mask = 0;
858 memset(&qp_attr, 0, sizeof qp_attr);
859 qp_attr.qp_state = cmd.qp_state;
860 ret = rdma_init_qp_attr(ctx->cm_id, &qp_attr, &resp.qp_attr_mask);
864 ib_copy_qp_attr_to_user(&resp, &qp_attr);
865 if (copy_to_user((void __user *)(unsigned long)cmd.response,
866 &resp, sizeof(resp)))
874 static int ucma_set_option_id(struct ucma_context *ctx, int optname,
875 void *optval, size_t optlen)
880 case RDMA_OPTION_ID_TOS:
881 if (optlen != sizeof(u8)) {
885 rdma_set_service_type(ctx->cm_id, *((u8 *) optval));
887 case RDMA_OPTION_ID_REUSEADDR:
888 if (optlen != sizeof(int)) {
892 ret = rdma_set_reuseaddr(ctx->cm_id, *((int *) optval) ? 1 : 0);
894 case RDMA_OPTION_ID_AFONLY:
895 if (optlen != sizeof(int)) {
899 ret = rdma_set_afonly(ctx->cm_id, *((int *) optval) ? 1 : 0);
908 static int ucma_set_ib_path(struct ucma_context *ctx,
909 struct ib_path_rec_data *path_data, size_t optlen)
911 struct ib_sa_path_rec sa_path;
912 struct rdma_cm_event event;
915 if (optlen % sizeof(*path_data))
918 for (; optlen; optlen -= sizeof(*path_data), path_data++) {
919 if (path_data->flags == (IB_PATH_GMP | IB_PATH_PRIMARY |
920 IB_PATH_BIDIRECTIONAL))
927 ib_sa_unpack_path(path_data->path_rec, &sa_path);
928 ret = rdma_set_ib_paths(ctx->cm_id, &sa_path, 1);
932 memset(&event, 0, sizeof event);
933 event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
934 return ucma_event_handler(ctx->cm_id, &event);
937 static int ucma_set_option_ib(struct ucma_context *ctx, int optname,
938 void *optval, size_t optlen)
943 case RDMA_OPTION_IB_PATH:
944 ret = ucma_set_ib_path(ctx, optval, optlen);
947 case RDMA_OPTION_IB_APM:
948 if (optlen != sizeof(u8)) {
953 ret = rdma_enable_apm(ctx->cm_id, RDMA_ALT_PATH_BEST);
963 static int ucma_set_option_level(struct ucma_context *ctx, int level,
964 int optname, void *optval, size_t optlen)
970 ret = ucma_set_option_id(ctx, optname, optval, optlen);
973 ret = ucma_set_option_ib(ctx, optname, optval, optlen);
982 static ssize_t ucma_set_option(struct ucma_file *file, const char __user *inbuf,
983 int in_len, int out_len)
985 struct rdma_ucm_set_option cmd;
986 struct ucma_context *ctx;
990 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
993 ctx = ucma_get_ctx(file, cmd.id);
997 optval = kmalloc(cmd.optlen, GFP_KERNEL);
1000 goto err_ucma_put_ctx;
1003 if (copy_from_user(optval, (void __user *)(unsigned long)cmd.optval,
1009 ret = ucma_set_option_level(ctx, cmd.level, cmd.optname, optval,
1019 static ssize_t ucma_notify(struct ucma_file *file, const char __user *inbuf,
1020 int in_len, int out_len)
1022 struct rdma_ucm_notify cmd;
1023 struct ucma_context *ctx;
1026 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1029 ctx = ucma_get_ctx(file, cmd.id);
1031 return PTR_ERR(ctx);
1033 ret = rdma_notify(ctx->cm_id, (enum ib_event_type) cmd.event);
1038 static ssize_t ucma_join_multicast(struct ucma_file *file,
1039 const char __user *inbuf,
1040 int in_len, int out_len)
1042 struct rdma_ucm_join_mcast cmd;
1043 struct rdma_ucm_create_id_resp resp;
1044 struct ucma_context *ctx;
1045 struct ucma_multicast *mc;
1048 if (out_len < sizeof(resp))
1051 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1054 ctx = ucma_get_ctx(file, cmd.id);
1056 return PTR_ERR(ctx);
1058 mutex_lock(&file->mut);
1059 mc = ucma_alloc_multicast(ctx);
1066 memcpy(&mc->addr, &cmd.addr, sizeof cmd.addr);
1067 ret = rdma_join_multicast(ctx->cm_id, (struct sockaddr *) &mc->addr, mc);
1072 if (copy_to_user((void __user *)(unsigned long)cmd.response,
1073 &resp, sizeof(resp))) {
1078 mutex_unlock(&file->mut);
1083 rdma_leave_multicast(ctx->cm_id, (struct sockaddr *) &mc->addr);
1084 ucma_cleanup_mc_events(mc);
1087 idr_remove(&multicast_idr, mc->id);
1089 list_del(&mc->list);
1092 mutex_unlock(&file->mut);
1097 static ssize_t ucma_leave_multicast(struct ucma_file *file,
1098 const char __user *inbuf,
1099 int in_len, int out_len)
1101 struct rdma_ucm_destroy_id cmd;
1102 struct rdma_ucm_destroy_id_resp resp;
1103 struct ucma_multicast *mc;
1106 if (out_len < sizeof(resp))
1109 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1113 mc = idr_find(&multicast_idr, cmd.id);
1115 mc = ERR_PTR(-ENOENT);
1116 else if (mc->ctx->file != file)
1117 mc = ERR_PTR(-EINVAL);
1119 idr_remove(&multicast_idr, mc->id);
1120 atomic_inc(&mc->ctx->ref);
1129 rdma_leave_multicast(mc->ctx->cm_id, (struct sockaddr *) &mc->addr);
1130 mutex_lock(&mc->ctx->file->mut);
1131 ucma_cleanup_mc_events(mc);
1132 list_del(&mc->list);
1133 mutex_unlock(&mc->ctx->file->mut);
1135 ucma_put_ctx(mc->ctx);
1136 resp.events_reported = mc->events_reported;
1139 if (copy_to_user((void __user *)(unsigned long)cmd.response,
1140 &resp, sizeof(resp)))
1146 static void ucma_lock_files(struct ucma_file *file1, struct ucma_file *file2)
1148 /* Acquire mutex's based on pointer comparison to prevent deadlock. */
1149 if (file1 < file2) {
1150 mutex_lock(&file1->mut);
1151 mutex_lock(&file2->mut);
1153 mutex_lock(&file2->mut);
1154 mutex_lock(&file1->mut);
1158 static void ucma_unlock_files(struct ucma_file *file1, struct ucma_file *file2)
1160 if (file1 < file2) {
1161 mutex_unlock(&file2->mut);
1162 mutex_unlock(&file1->mut);
1164 mutex_unlock(&file1->mut);
1165 mutex_unlock(&file2->mut);
1169 static void ucma_move_events(struct ucma_context *ctx, struct ucma_file *file)
1171 struct ucma_event *uevent, *tmp;
1173 list_for_each_entry_safe(uevent, tmp, &ctx->file->event_list, list)
1174 if (uevent->ctx == ctx)
1175 list_move_tail(&uevent->list, &file->event_list);
1178 static ssize_t ucma_migrate_id(struct ucma_file *new_file,
1179 const char __user *inbuf,
1180 int in_len, int out_len)
1182 struct rdma_ucm_migrate_id cmd;
1183 struct rdma_ucm_migrate_resp resp;
1184 struct ucma_context *ctx;
1186 struct ucma_file *cur_file;
1189 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1192 /* Get current fd to protect against it being closed */
1197 /* Validate current fd and prevent destruction of id. */
1198 ctx = ucma_get_ctx(f.file->private_data, cmd.id);
1204 cur_file = ctx->file;
1205 if (cur_file == new_file) {
1206 resp.events_reported = ctx->events_reported;
1211 * Migrate events between fd's, maintaining order, and avoiding new
1212 * events being added before existing events.
1214 ucma_lock_files(cur_file, new_file);
1217 list_move_tail(&ctx->list, &new_file->ctx_list);
1218 ucma_move_events(ctx, new_file);
1219 ctx->file = new_file;
1220 resp.events_reported = ctx->events_reported;
1223 ucma_unlock_files(cur_file, new_file);
1226 if (copy_to_user((void __user *)(unsigned long)cmd.response,
1227 &resp, sizeof(resp)))
1236 static ssize_t (*ucma_cmd_table[])(struct ucma_file *file,
1237 const char __user *inbuf,
1238 int in_len, int out_len) = {
1239 [RDMA_USER_CM_CMD_CREATE_ID] = ucma_create_id,
1240 [RDMA_USER_CM_CMD_DESTROY_ID] = ucma_destroy_id,
1241 [RDMA_USER_CM_CMD_BIND_ADDR] = ucma_bind_addr,
1242 [RDMA_USER_CM_CMD_RESOLVE_ADDR] = ucma_resolve_addr,
1243 [RDMA_USER_CM_CMD_RESOLVE_ROUTE]= ucma_resolve_route,
1244 [RDMA_USER_CM_CMD_QUERY_ROUTE] = ucma_query_route,
1245 [RDMA_USER_CM_CMD_CONNECT] = ucma_connect,
1246 [RDMA_USER_CM_CMD_LISTEN] = ucma_listen,
1247 [RDMA_USER_CM_CMD_ACCEPT] = ucma_accept,
1248 [RDMA_USER_CM_CMD_REJECT] = ucma_reject,
1249 [RDMA_USER_CM_CMD_DISCONNECT] = ucma_disconnect,
1250 [RDMA_USER_CM_CMD_INIT_QP_ATTR] = ucma_init_qp_attr,
1251 [RDMA_USER_CM_CMD_GET_EVENT] = ucma_get_event,
1252 [RDMA_USER_CM_CMD_GET_OPTION] = NULL,
1253 [RDMA_USER_CM_CMD_SET_OPTION] = ucma_set_option,
1254 [RDMA_USER_CM_CMD_NOTIFY] = ucma_notify,
1255 [RDMA_USER_CM_CMD_JOIN_MCAST] = ucma_join_multicast,
1256 [RDMA_USER_CM_CMD_LEAVE_MCAST] = ucma_leave_multicast,
1257 [RDMA_USER_CM_CMD_MIGRATE_ID] = ucma_migrate_id
1260 static ssize_t ucma_write(struct file *filp, const char __user *buf,
1261 size_t len, loff_t *pos)
1263 struct ucma_file *file = filp->private_data;
1264 struct rdma_ucm_cmd_hdr hdr;
1267 if (len < sizeof(hdr))
1270 if (copy_from_user(&hdr, buf, sizeof(hdr)))
1273 if (hdr.cmd >= ARRAY_SIZE(ucma_cmd_table))
1276 if (hdr.in + sizeof(hdr) > len)
1279 if (!ucma_cmd_table[hdr.cmd])
1282 ret = ucma_cmd_table[hdr.cmd](file, buf + sizeof(hdr), hdr.in, hdr.out);
1289 static unsigned int ucma_poll(struct file *filp, struct poll_table_struct *wait)
1291 struct ucma_file *file = filp->private_data;
1292 unsigned int mask = 0;
1294 poll_wait(filp, &file->poll_wait, wait);
1296 if (!list_empty(&file->event_list))
1297 mask = POLLIN | POLLRDNORM;
1303 * ucma_open() does not need the BKL:
1305 * - no global state is referred to;
1306 * - there is no ioctl method to race against;
1307 * - no further module initialization is required for open to work
1308 * after the device is registered.
1310 static int ucma_open(struct inode *inode, struct file *filp)
1312 struct ucma_file *file;
1314 file = kmalloc(sizeof *file, GFP_KERNEL);
1318 INIT_LIST_HEAD(&file->event_list);
1319 INIT_LIST_HEAD(&file->ctx_list);
1320 init_waitqueue_head(&file->poll_wait);
1321 mutex_init(&file->mut);
1323 filp->private_data = file;
1326 return nonseekable_open(inode, filp);
1329 static int ucma_close(struct inode *inode, struct file *filp)
1331 struct ucma_file *file = filp->private_data;
1332 struct ucma_context *ctx, *tmp;
1334 mutex_lock(&file->mut);
1335 list_for_each_entry_safe(ctx, tmp, &file->ctx_list, list) {
1336 mutex_unlock(&file->mut);
1339 idr_remove(&ctx_idr, ctx->id);
1343 mutex_lock(&file->mut);
1345 mutex_unlock(&file->mut);
1351 ucma_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
1363 static const struct file_operations ucma_fops = {
1364 .owner = THIS_MODULE,
1366 .release = ucma_close,
1367 .write = ucma_write,
1368 .unlocked_ioctl = ucma_ioctl,
1370 .llseek = no_llseek,
1373 static struct miscdevice ucma_misc = {
1374 .minor = MISC_DYNAMIC_MINOR,
1376 .nodename = "infiniband/rdma_cm",
1381 static ssize_t show_abi_version(struct device *dev,
1382 struct device_attribute *attr,
1385 return sprintf(buf, "%d\n", RDMA_USER_CM_ABI_VERSION);
1387 static DEVICE_ATTR(abi_version, S_IRUGO, show_abi_version, NULL);
1389 static int __init ucma_init(void)
1393 ret = misc_register(&ucma_misc);
1397 ret = device_create_file(ucma_misc.this_device, &dev_attr_abi_version);
1399 printk(KERN_ERR "rdma_ucm: couldn't create abi_version attr\n");
1405 misc_deregister(&ucma_misc);
1409 static void __exit ucma_cleanup(void)
1411 device_remove_file(ucma_misc.this_device, &dev_attr_abi_version);
1412 misc_deregister(&ucma_misc);
1413 idr_destroy(&ctx_idr);
1416 module_init(ucma_init);
1417 module_exit(ucma_cleanup);