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/idr.h>
39 #include <linux/in6.h>
40 #include <linux/miscdevice.h>
42 #include <rdma/rdma_user_cm.h>
43 #include <rdma/ib_marshall.h>
44 #include <rdma/rdma_cm.h>
46 MODULE_AUTHOR("Sean Hefty");
47 MODULE_DESCRIPTION("RDMA Userspace Connection Manager Access");
48 MODULE_LICENSE("Dual BSD/GPL");
51 UCMA_MAX_BACKLOG = 128
57 struct list_head ctx_list;
58 struct list_head event_list;
59 wait_queue_head_t poll_wait;
64 struct completion comp;
69 struct ucma_file *file;
70 struct rdma_cm_id *cm_id;
73 struct list_head list;
74 struct list_head mc_list;
77 struct ucma_multicast {
78 struct ucma_context *ctx;
83 struct list_head list;
85 u8 pad[sizeof(struct sockaddr_in6) -
86 sizeof(struct sockaddr)];
90 struct ucma_context *ctx;
91 struct ucma_multicast *mc;
92 struct list_head list;
93 struct rdma_cm_id *cm_id;
94 struct rdma_ucm_event_resp resp;
97 static DEFINE_MUTEX(mut);
98 static DEFINE_IDR(ctx_idr);
99 static DEFINE_IDR(multicast_idr);
101 static inline struct ucma_context *_ucma_find_context(int id,
102 struct ucma_file *file)
104 struct ucma_context *ctx;
106 ctx = idr_find(&ctx_idr, id);
108 ctx = ERR_PTR(-ENOENT);
109 else if (ctx->file != file)
110 ctx = ERR_PTR(-EINVAL);
114 static struct ucma_context *ucma_get_ctx(struct ucma_file *file, int id)
116 struct ucma_context *ctx;
119 ctx = _ucma_find_context(id, file);
121 atomic_inc(&ctx->ref);
126 static void ucma_put_ctx(struct ucma_context *ctx)
128 if (atomic_dec_and_test(&ctx->ref))
129 complete(&ctx->comp);
132 static struct ucma_context *ucma_alloc_ctx(struct ucma_file *file)
134 struct ucma_context *ctx;
137 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
141 atomic_set(&ctx->ref, 1);
142 init_completion(&ctx->comp);
143 INIT_LIST_HEAD(&ctx->mc_list);
147 ret = idr_pre_get(&ctx_idr, GFP_KERNEL);
152 ret = idr_get_new(&ctx_idr, ctx, &ctx->id);
154 } while (ret == -EAGAIN);
159 list_add_tail(&ctx->list, &file->ctx_list);
167 static struct ucma_multicast* ucma_alloc_multicast(struct ucma_context *ctx)
169 struct ucma_multicast *mc;
172 mc = kzalloc(sizeof(*mc), GFP_KERNEL);
177 ret = idr_pre_get(&multicast_idr, GFP_KERNEL);
182 ret = idr_get_new(&multicast_idr, mc, &mc->id);
184 } while (ret == -EAGAIN);
190 list_add_tail(&mc->list, &ctx->mc_list);
198 static void ucma_copy_conn_event(struct rdma_ucm_conn_param *dst,
199 struct rdma_conn_param *src)
201 if (src->private_data_len)
202 memcpy(dst->private_data, src->private_data,
203 src->private_data_len);
204 dst->private_data_len = src->private_data_len;
205 dst->responder_resources =src->responder_resources;
206 dst->initiator_depth = src->initiator_depth;
207 dst->flow_control = src->flow_control;
208 dst->retry_count = src->retry_count;
209 dst->rnr_retry_count = src->rnr_retry_count;
211 dst->qp_num = src->qp_num;
214 static void ucma_copy_ud_event(struct rdma_ucm_ud_param *dst,
215 struct rdma_ud_param *src)
217 if (src->private_data_len)
218 memcpy(dst->private_data, src->private_data,
219 src->private_data_len);
220 dst->private_data_len = src->private_data_len;
221 ib_copy_ah_attr_to_user(&dst->ah_attr, &src->ah_attr);
222 dst->qp_num = src->qp_num;
223 dst->qkey = src->qkey;
226 static void ucma_set_event_context(struct ucma_context *ctx,
227 struct rdma_cm_event *event,
228 struct ucma_event *uevent)
231 switch (event->event) {
232 case RDMA_CM_EVENT_MULTICAST_JOIN:
233 case RDMA_CM_EVENT_MULTICAST_ERROR:
234 uevent->mc = (struct ucma_multicast *)
235 event->param.ud.private_data;
236 uevent->resp.uid = uevent->mc->uid;
237 uevent->resp.id = uevent->mc->id;
240 uevent->resp.uid = ctx->uid;
241 uevent->resp.id = ctx->id;
246 static int ucma_event_handler(struct rdma_cm_id *cm_id,
247 struct rdma_cm_event *event)
249 struct ucma_event *uevent;
250 struct ucma_context *ctx = cm_id->context;
253 uevent = kzalloc(sizeof(*uevent), GFP_KERNEL);
255 return event->event == RDMA_CM_EVENT_CONNECT_REQUEST;
257 uevent->cm_id = cm_id;
258 ucma_set_event_context(ctx, event, uevent);
259 uevent->resp.event = event->event;
260 uevent->resp.status = event->status;
261 if (cm_id->ps == RDMA_PS_UDP || cm_id->ps == RDMA_PS_IPOIB)
262 ucma_copy_ud_event(&uevent->resp.param.ud, &event->param.ud);
264 ucma_copy_conn_event(&uevent->resp.param.conn,
267 mutex_lock(&ctx->file->mut);
268 if (event->event == RDMA_CM_EVENT_CONNECT_REQUEST) {
275 } else if (!ctx->uid) {
277 * We ignore events for new connections until userspace has set
278 * their context. This can only happen if an error occurs on a
279 * new connection before the user accepts it. This is okay,
280 * since the accept will just fail later.
286 list_add_tail(&uevent->list, &ctx->file->event_list);
287 wake_up_interruptible(&ctx->file->poll_wait);
289 mutex_unlock(&ctx->file->mut);
293 static ssize_t ucma_get_event(struct ucma_file *file, const char __user *inbuf,
294 int in_len, int out_len)
296 struct ucma_context *ctx;
297 struct rdma_ucm_get_event cmd;
298 struct ucma_event *uevent;
302 if (out_len < sizeof uevent->resp)
305 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
308 mutex_lock(&file->mut);
309 while (list_empty(&file->event_list)) {
310 mutex_unlock(&file->mut);
312 if (file->filp->f_flags & O_NONBLOCK)
315 if (wait_event_interruptible(file->poll_wait,
316 !list_empty(&file->event_list)))
319 mutex_lock(&file->mut);
322 uevent = list_entry(file->event_list.next, struct ucma_event, list);
324 if (uevent->resp.event == RDMA_CM_EVENT_CONNECT_REQUEST) {
325 ctx = ucma_alloc_ctx(file);
330 uevent->ctx->backlog++;
331 ctx->cm_id = uevent->cm_id;
332 ctx->cm_id->context = ctx;
333 uevent->resp.id = ctx->id;
336 if (copy_to_user((void __user *)(unsigned long)cmd.response,
337 &uevent->resp, sizeof uevent->resp)) {
342 list_del(&uevent->list);
343 uevent->ctx->events_reported++;
345 uevent->mc->events_reported++;
348 mutex_unlock(&file->mut);
352 static ssize_t ucma_create_id(struct ucma_file *file,
353 const char __user *inbuf,
354 int in_len, int out_len)
356 struct rdma_ucm_create_id cmd;
357 struct rdma_ucm_create_id_resp resp;
358 struct ucma_context *ctx;
361 if (out_len < sizeof(resp))
364 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
367 mutex_lock(&file->mut);
368 ctx = ucma_alloc_ctx(file);
369 mutex_unlock(&file->mut);
374 ctx->cm_id = rdma_create_id(ucma_event_handler, ctx, cmd.ps);
375 if (IS_ERR(ctx->cm_id)) {
376 ret = PTR_ERR(ctx->cm_id);
381 if (copy_to_user((void __user *)(unsigned long)cmd.response,
382 &resp, sizeof(resp))) {
389 rdma_destroy_id(ctx->cm_id);
392 idr_remove(&ctx_idr, ctx->id);
398 static void ucma_cleanup_multicast(struct ucma_context *ctx)
400 struct ucma_multicast *mc, *tmp;
403 list_for_each_entry_safe(mc, tmp, &ctx->mc_list, list) {
405 idr_remove(&multicast_idr, mc->id);
411 static void ucma_cleanup_events(struct ucma_context *ctx)
413 struct ucma_event *uevent, *tmp;
415 list_for_each_entry_safe(uevent, tmp, &ctx->file->event_list, list) {
416 if (uevent->ctx != ctx)
419 list_del(&uevent->list);
421 /* clear incoming connections. */
422 if (uevent->resp.event == RDMA_CM_EVENT_CONNECT_REQUEST)
423 rdma_destroy_id(uevent->cm_id);
429 static void ucma_cleanup_mc_events(struct ucma_multicast *mc)
431 struct ucma_event *uevent, *tmp;
433 list_for_each_entry_safe(uevent, tmp, &mc->ctx->file->event_list, list) {
434 if (uevent->mc != mc)
437 list_del(&uevent->list);
442 static int ucma_free_ctx(struct ucma_context *ctx)
446 /* No new events will be generated after destroying the id. */
447 rdma_destroy_id(ctx->cm_id);
449 ucma_cleanup_multicast(ctx);
451 /* Cleanup events not yet reported to the user. */
452 mutex_lock(&ctx->file->mut);
453 ucma_cleanup_events(ctx);
454 list_del(&ctx->list);
455 mutex_unlock(&ctx->file->mut);
457 events_reported = ctx->events_reported;
459 return events_reported;
462 static ssize_t ucma_destroy_id(struct ucma_file *file, const char __user *inbuf,
463 int in_len, int out_len)
465 struct rdma_ucm_destroy_id cmd;
466 struct rdma_ucm_destroy_id_resp resp;
467 struct ucma_context *ctx;
470 if (out_len < sizeof(resp))
473 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
477 ctx = _ucma_find_context(cmd.id, file);
479 idr_remove(&ctx_idr, ctx->id);
486 wait_for_completion(&ctx->comp);
487 resp.events_reported = ucma_free_ctx(ctx);
489 if (copy_to_user((void __user *)(unsigned long)cmd.response,
490 &resp, sizeof(resp)))
496 static ssize_t ucma_bind_addr(struct ucma_file *file, const char __user *inbuf,
497 int in_len, int out_len)
499 struct rdma_ucm_bind_addr cmd;
500 struct ucma_context *ctx;
503 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
506 ctx = ucma_get_ctx(file, cmd.id);
510 ret = rdma_bind_addr(ctx->cm_id, (struct sockaddr *) &cmd.addr);
515 static ssize_t ucma_resolve_addr(struct ucma_file *file,
516 const char __user *inbuf,
517 int in_len, int out_len)
519 struct rdma_ucm_resolve_addr cmd;
520 struct ucma_context *ctx;
523 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
526 ctx = ucma_get_ctx(file, cmd.id);
530 ret = rdma_resolve_addr(ctx->cm_id, (struct sockaddr *) &cmd.src_addr,
531 (struct sockaddr *) &cmd.dst_addr,
537 static ssize_t ucma_resolve_route(struct ucma_file *file,
538 const char __user *inbuf,
539 int in_len, int out_len)
541 struct rdma_ucm_resolve_route cmd;
542 struct ucma_context *ctx;
545 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
548 ctx = ucma_get_ctx(file, cmd.id);
552 ret = rdma_resolve_route(ctx->cm_id, cmd.timeout_ms);
557 static void ucma_copy_ib_route(struct rdma_ucm_query_route_resp *resp,
558 struct rdma_route *route)
560 struct rdma_dev_addr *dev_addr;
562 resp->num_paths = route->num_paths;
563 switch (route->num_paths) {
565 dev_addr = &route->addr.dev_addr;
566 ib_addr_get_dgid(dev_addr,
567 (union ib_gid *) &resp->ib_route[0].dgid);
568 ib_addr_get_sgid(dev_addr,
569 (union ib_gid *) &resp->ib_route[0].sgid);
570 resp->ib_route[0].pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
573 ib_copy_path_rec_to_user(&resp->ib_route[1],
574 &route->path_rec[1]);
577 ib_copy_path_rec_to_user(&resp->ib_route[0],
578 &route->path_rec[0]);
585 static ssize_t ucma_query_route(struct ucma_file *file,
586 const char __user *inbuf,
587 int in_len, int out_len)
589 struct rdma_ucm_query_route cmd;
590 struct rdma_ucm_query_route_resp resp;
591 struct ucma_context *ctx;
592 struct sockaddr *addr;
595 if (out_len < sizeof(resp))
598 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
601 ctx = ucma_get_ctx(file, cmd.id);
605 memset(&resp, 0, sizeof resp);
606 addr = &ctx->cm_id->route.addr.src_addr;
607 memcpy(&resp.src_addr, addr, addr->sa_family == AF_INET ?
608 sizeof(struct sockaddr_in) :
609 sizeof(struct sockaddr_in6));
610 addr = &ctx->cm_id->route.addr.dst_addr;
611 memcpy(&resp.dst_addr, addr, addr->sa_family == AF_INET ?
612 sizeof(struct sockaddr_in) :
613 sizeof(struct sockaddr_in6));
614 if (!ctx->cm_id->device)
617 resp.node_guid = (__force __u64) ctx->cm_id->device->node_guid;
618 resp.port_num = ctx->cm_id->port_num;
619 switch (rdma_node_get_transport(ctx->cm_id->device->node_type)) {
620 case RDMA_TRANSPORT_IB:
621 ucma_copy_ib_route(&resp, &ctx->cm_id->route);
628 if (copy_to_user((void __user *)(unsigned long)cmd.response,
629 &resp, sizeof(resp)))
636 static void ucma_copy_conn_param(struct rdma_conn_param *dst,
637 struct rdma_ucm_conn_param *src)
639 dst->private_data = src->private_data;
640 dst->private_data_len = src->private_data_len;
641 dst->responder_resources =src->responder_resources;
642 dst->initiator_depth = src->initiator_depth;
643 dst->flow_control = src->flow_control;
644 dst->retry_count = src->retry_count;
645 dst->rnr_retry_count = src->rnr_retry_count;
647 dst->qp_num = src->qp_num;
650 static ssize_t ucma_connect(struct ucma_file *file, const char __user *inbuf,
651 int in_len, int out_len)
653 struct rdma_ucm_connect cmd;
654 struct rdma_conn_param conn_param;
655 struct ucma_context *ctx;
658 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
661 if (!cmd.conn_param.valid)
664 ctx = ucma_get_ctx(file, cmd.id);
668 ucma_copy_conn_param(&conn_param, &cmd.conn_param);
669 ret = rdma_connect(ctx->cm_id, &conn_param);
674 static ssize_t ucma_listen(struct ucma_file *file, const char __user *inbuf,
675 int in_len, int out_len)
677 struct rdma_ucm_listen cmd;
678 struct ucma_context *ctx;
681 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
684 ctx = ucma_get_ctx(file, cmd.id);
688 ctx->backlog = cmd.backlog > 0 && cmd.backlog < UCMA_MAX_BACKLOG ?
689 cmd.backlog : UCMA_MAX_BACKLOG;
690 ret = rdma_listen(ctx->cm_id, ctx->backlog);
695 static ssize_t ucma_accept(struct ucma_file *file, const char __user *inbuf,
696 int in_len, int out_len)
698 struct rdma_ucm_accept cmd;
699 struct rdma_conn_param conn_param;
700 struct ucma_context *ctx;
703 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
706 ctx = ucma_get_ctx(file, cmd.id);
710 if (cmd.conn_param.valid) {
712 ucma_copy_conn_param(&conn_param, &cmd.conn_param);
713 ret = rdma_accept(ctx->cm_id, &conn_param);
715 ret = rdma_accept(ctx->cm_id, NULL);
721 static ssize_t ucma_reject(struct ucma_file *file, const char __user *inbuf,
722 int in_len, int out_len)
724 struct rdma_ucm_reject cmd;
725 struct ucma_context *ctx;
728 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
731 ctx = ucma_get_ctx(file, cmd.id);
735 ret = rdma_reject(ctx->cm_id, cmd.private_data, cmd.private_data_len);
740 static ssize_t ucma_disconnect(struct ucma_file *file, const char __user *inbuf,
741 int in_len, int out_len)
743 struct rdma_ucm_disconnect cmd;
744 struct ucma_context *ctx;
747 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
750 ctx = ucma_get_ctx(file, cmd.id);
754 ret = rdma_disconnect(ctx->cm_id);
759 static ssize_t ucma_init_qp_attr(struct ucma_file *file,
760 const char __user *inbuf,
761 int in_len, int out_len)
763 struct rdma_ucm_init_qp_attr cmd;
764 struct ib_uverbs_qp_attr resp;
765 struct ucma_context *ctx;
766 struct ib_qp_attr qp_attr;
769 if (out_len < sizeof(resp))
772 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
775 ctx = ucma_get_ctx(file, cmd.id);
779 resp.qp_attr_mask = 0;
780 memset(&qp_attr, 0, sizeof qp_attr);
781 qp_attr.qp_state = cmd.qp_state;
782 ret = rdma_init_qp_attr(ctx->cm_id, &qp_attr, &resp.qp_attr_mask);
786 ib_copy_qp_attr_to_user(&resp, &qp_attr);
787 if (copy_to_user((void __user *)(unsigned long)cmd.response,
788 &resp, sizeof(resp)))
796 static int ucma_set_option_id(struct ucma_context *ctx, int optname,
797 void *optval, size_t optlen)
802 case RDMA_OPTION_ID_TOS:
803 if (optlen != sizeof(u8)) {
807 rdma_set_service_type(ctx->cm_id, *((u8 *) optval));
816 static int ucma_set_option_level(struct ucma_context *ctx, int level,
817 int optname, void *optval, size_t optlen)
823 ret = ucma_set_option_id(ctx, optname, optval, optlen);
832 static ssize_t ucma_set_option(struct ucma_file *file, const char __user *inbuf,
833 int in_len, int out_len)
835 struct rdma_ucm_set_option cmd;
836 struct ucma_context *ctx;
840 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
843 ctx = ucma_get_ctx(file, cmd.id);
847 optval = kmalloc(cmd.optlen, GFP_KERNEL);
853 if (copy_from_user(optval, (void __user *) (unsigned long) cmd.optval,
859 ret = ucma_set_option_level(ctx, cmd.level, cmd.optname, optval,
868 static ssize_t ucma_notify(struct ucma_file *file, const char __user *inbuf,
869 int in_len, int out_len)
871 struct rdma_ucm_notify cmd;
872 struct ucma_context *ctx;
875 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
878 ctx = ucma_get_ctx(file, cmd.id);
882 ret = rdma_notify(ctx->cm_id, (enum ib_event_type) cmd.event);
887 static ssize_t ucma_join_multicast(struct ucma_file *file,
888 const char __user *inbuf,
889 int in_len, int out_len)
891 struct rdma_ucm_join_mcast cmd;
892 struct rdma_ucm_create_id_resp resp;
893 struct ucma_context *ctx;
894 struct ucma_multicast *mc;
897 if (out_len < sizeof(resp))
900 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
903 ctx = ucma_get_ctx(file, cmd.id);
907 mutex_lock(&file->mut);
908 mc = ucma_alloc_multicast(ctx);
915 memcpy(&mc->addr, &cmd.addr, sizeof cmd.addr);
916 ret = rdma_join_multicast(ctx->cm_id, &mc->addr, mc);
921 if (copy_to_user((void __user *)(unsigned long)cmd.response,
922 &resp, sizeof(resp))) {
927 mutex_unlock(&file->mut);
932 rdma_leave_multicast(ctx->cm_id, &mc->addr);
933 ucma_cleanup_mc_events(mc);
936 idr_remove(&multicast_idr, mc->id);
941 mutex_unlock(&file->mut);
946 static ssize_t ucma_leave_multicast(struct ucma_file *file,
947 const char __user *inbuf,
948 int in_len, int out_len)
950 struct rdma_ucm_destroy_id cmd;
951 struct rdma_ucm_destroy_id_resp resp;
952 struct ucma_multicast *mc;
955 if (out_len < sizeof(resp))
958 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
962 mc = idr_find(&multicast_idr, cmd.id);
964 mc = ERR_PTR(-ENOENT);
965 else if (mc->ctx->file != file)
966 mc = ERR_PTR(-EINVAL);
968 idr_remove(&multicast_idr, mc->id);
969 atomic_inc(&mc->ctx->ref);
978 rdma_leave_multicast(mc->ctx->cm_id, &mc->addr);
979 mutex_lock(&mc->ctx->file->mut);
980 ucma_cleanup_mc_events(mc);
982 mutex_unlock(&mc->ctx->file->mut);
984 ucma_put_ctx(mc->ctx);
985 resp.events_reported = mc->events_reported;
988 if (copy_to_user((void __user *)(unsigned long)cmd.response,
989 &resp, sizeof(resp)))
995 static void ucma_lock_files(struct ucma_file *file1, struct ucma_file *file2)
997 /* Acquire mutex's based on pointer comparison to prevent deadlock. */
999 mutex_lock(&file1->mut);
1000 mutex_lock(&file2->mut);
1002 mutex_lock(&file2->mut);
1003 mutex_lock(&file1->mut);
1007 static void ucma_unlock_files(struct ucma_file *file1, struct ucma_file *file2)
1009 if (file1 < file2) {
1010 mutex_unlock(&file2->mut);
1011 mutex_unlock(&file1->mut);
1013 mutex_unlock(&file1->mut);
1014 mutex_unlock(&file2->mut);
1018 static void ucma_move_events(struct ucma_context *ctx, struct ucma_file *file)
1020 struct ucma_event *uevent, *tmp;
1022 list_for_each_entry_safe(uevent, tmp, &ctx->file->event_list, list)
1023 if (uevent->ctx == ctx)
1024 list_move_tail(&uevent->list, &file->event_list);
1027 static ssize_t ucma_migrate_id(struct ucma_file *new_file,
1028 const char __user *inbuf,
1029 int in_len, int out_len)
1031 struct rdma_ucm_migrate_id cmd;
1032 struct rdma_ucm_migrate_resp resp;
1033 struct ucma_context *ctx;
1035 struct ucma_file *cur_file;
1038 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1041 /* Get current fd to protect against it being closed */
1042 filp = fget(cmd.fd);
1046 /* Validate current fd and prevent destruction of id. */
1047 ctx = ucma_get_ctx(filp->private_data, cmd.id);
1053 cur_file = ctx->file;
1054 if (cur_file == new_file) {
1055 resp.events_reported = ctx->events_reported;
1060 * Migrate events between fd's, maintaining order, and avoiding new
1061 * events being added before existing events.
1063 ucma_lock_files(cur_file, new_file);
1066 list_move_tail(&ctx->list, &new_file->ctx_list);
1067 ucma_move_events(ctx, new_file);
1068 ctx->file = new_file;
1069 resp.events_reported = ctx->events_reported;
1072 ucma_unlock_files(cur_file, new_file);
1075 if (copy_to_user((void __user *)(unsigned long)cmd.response,
1076 &resp, sizeof(resp)))
1085 static ssize_t (*ucma_cmd_table[])(struct ucma_file *file,
1086 const char __user *inbuf,
1087 int in_len, int out_len) = {
1088 [RDMA_USER_CM_CMD_CREATE_ID] = ucma_create_id,
1089 [RDMA_USER_CM_CMD_DESTROY_ID] = ucma_destroy_id,
1090 [RDMA_USER_CM_CMD_BIND_ADDR] = ucma_bind_addr,
1091 [RDMA_USER_CM_CMD_RESOLVE_ADDR] = ucma_resolve_addr,
1092 [RDMA_USER_CM_CMD_RESOLVE_ROUTE]= ucma_resolve_route,
1093 [RDMA_USER_CM_CMD_QUERY_ROUTE] = ucma_query_route,
1094 [RDMA_USER_CM_CMD_CONNECT] = ucma_connect,
1095 [RDMA_USER_CM_CMD_LISTEN] = ucma_listen,
1096 [RDMA_USER_CM_CMD_ACCEPT] = ucma_accept,
1097 [RDMA_USER_CM_CMD_REJECT] = ucma_reject,
1098 [RDMA_USER_CM_CMD_DISCONNECT] = ucma_disconnect,
1099 [RDMA_USER_CM_CMD_INIT_QP_ATTR] = ucma_init_qp_attr,
1100 [RDMA_USER_CM_CMD_GET_EVENT] = ucma_get_event,
1101 [RDMA_USER_CM_CMD_GET_OPTION] = NULL,
1102 [RDMA_USER_CM_CMD_SET_OPTION] = ucma_set_option,
1103 [RDMA_USER_CM_CMD_NOTIFY] = ucma_notify,
1104 [RDMA_USER_CM_CMD_JOIN_MCAST] = ucma_join_multicast,
1105 [RDMA_USER_CM_CMD_LEAVE_MCAST] = ucma_leave_multicast,
1106 [RDMA_USER_CM_CMD_MIGRATE_ID] = ucma_migrate_id
1109 static ssize_t ucma_write(struct file *filp, const char __user *buf,
1110 size_t len, loff_t *pos)
1112 struct ucma_file *file = filp->private_data;
1113 struct rdma_ucm_cmd_hdr hdr;
1116 if (len < sizeof(hdr))
1119 if (copy_from_user(&hdr, buf, sizeof(hdr)))
1122 if (hdr.cmd < 0 || hdr.cmd >= ARRAY_SIZE(ucma_cmd_table))
1125 if (hdr.in + sizeof(hdr) > len)
1128 if (!ucma_cmd_table[hdr.cmd])
1131 ret = ucma_cmd_table[hdr.cmd](file, buf + sizeof(hdr), hdr.in, hdr.out);
1138 static unsigned int ucma_poll(struct file *filp, struct poll_table_struct *wait)
1140 struct ucma_file *file = filp->private_data;
1141 unsigned int mask = 0;
1143 poll_wait(filp, &file->poll_wait, wait);
1145 if (!list_empty(&file->event_list))
1146 mask = POLLIN | POLLRDNORM;
1152 * ucma_open() does not need the BKL:
1154 * - no global state is referred to;
1155 * - there is no ioctl method to race against;
1156 * - no further module initialization is required for open to work
1157 * after the device is registered.
1159 static int ucma_open(struct inode *inode, struct file *filp)
1161 struct ucma_file *file;
1163 file = kmalloc(sizeof *file, GFP_KERNEL);
1167 INIT_LIST_HEAD(&file->event_list);
1168 INIT_LIST_HEAD(&file->ctx_list);
1169 init_waitqueue_head(&file->poll_wait);
1170 mutex_init(&file->mut);
1172 filp->private_data = file;
1177 static int ucma_close(struct inode *inode, struct file *filp)
1179 struct ucma_file *file = filp->private_data;
1180 struct ucma_context *ctx, *tmp;
1182 mutex_lock(&file->mut);
1183 list_for_each_entry_safe(ctx, tmp, &file->ctx_list, list) {
1184 mutex_unlock(&file->mut);
1187 idr_remove(&ctx_idr, ctx->id);
1191 mutex_lock(&file->mut);
1193 mutex_unlock(&file->mut);
1198 static const struct file_operations ucma_fops = {
1199 .owner = THIS_MODULE,
1201 .release = ucma_close,
1202 .write = ucma_write,
1206 static struct miscdevice ucma_misc = {
1207 .minor = MISC_DYNAMIC_MINOR,
1212 static ssize_t show_abi_version(struct device *dev,
1213 struct device_attribute *attr,
1216 return sprintf(buf, "%d\n", RDMA_USER_CM_ABI_VERSION);
1218 static DEVICE_ATTR(abi_version, S_IRUGO, show_abi_version, NULL);
1220 static int __init ucma_init(void)
1224 ret = misc_register(&ucma_misc);
1228 ret = device_create_file(ucma_misc.this_device, &dev_attr_abi_version);
1230 printk(KERN_ERR "rdma_ucm: couldn't create abi_version attr\n");
1235 misc_deregister(&ucma_misc);
1239 static void __exit ucma_cleanup(void)
1241 device_remove_file(ucma_misc.this_device, &dev_attr_abi_version);
1242 misc_deregister(&ucma_misc);
1243 idr_destroy(&ctx_idr);
1246 module_init(ucma_init);
1247 module_exit(ucma_cleanup);