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 || !ctx->cm_id)
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 struct rdma_cm_id *cm_id;
397 enum ib_qp_type qp_type;
400 if (out_len < sizeof(resp))
403 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
406 ret = ucma_get_qp_type(&cmd, &qp_type);
410 mutex_lock(&file->mut);
411 ctx = ucma_alloc_ctx(file);
412 mutex_unlock(&file->mut);
417 cm_id = rdma_create_id(ucma_event_handler, ctx, cmd.ps, qp_type);
419 ret = PTR_ERR(cm_id);
424 if (copy_to_user((void __user *)(unsigned long)cmd.response,
425 &resp, sizeof(resp))) {
434 rdma_destroy_id(cm_id);
437 idr_remove(&ctx_idr, ctx->id);
439 mutex_lock(&file->mut);
440 list_del(&ctx->list);
441 mutex_unlock(&file->mut);
446 static void ucma_cleanup_multicast(struct ucma_context *ctx)
448 struct ucma_multicast *mc, *tmp;
451 list_for_each_entry_safe(mc, tmp, &ctx->mc_list, list) {
453 idr_remove(&multicast_idr, mc->id);
459 static void ucma_cleanup_events(struct ucma_context *ctx)
461 struct ucma_event *uevent, *tmp;
463 list_for_each_entry_safe(uevent, tmp, &ctx->file->event_list, list) {
464 if (uevent->ctx != ctx)
467 list_del(&uevent->list);
469 /* clear incoming connections. */
470 if (uevent->resp.event == RDMA_CM_EVENT_CONNECT_REQUEST)
471 rdma_destroy_id(uevent->cm_id);
477 static void ucma_cleanup_mc_events(struct ucma_multicast *mc)
479 struct ucma_event *uevent, *tmp;
481 list_for_each_entry_safe(uevent, tmp, &mc->ctx->file->event_list, list) {
482 if (uevent->mc != mc)
485 list_del(&uevent->list);
490 static int ucma_free_ctx(struct ucma_context *ctx)
494 /* No new events will be generated after destroying the id. */
495 rdma_destroy_id(ctx->cm_id);
497 ucma_cleanup_multicast(ctx);
499 /* Cleanup events not yet reported to the user. */
500 mutex_lock(&ctx->file->mut);
501 ucma_cleanup_events(ctx);
502 list_del(&ctx->list);
503 mutex_unlock(&ctx->file->mut);
505 events_reported = ctx->events_reported;
507 return events_reported;
510 static ssize_t ucma_destroy_id(struct ucma_file *file, const char __user *inbuf,
511 int in_len, int out_len)
513 struct rdma_ucm_destroy_id cmd;
514 struct rdma_ucm_destroy_id_resp resp;
515 struct ucma_context *ctx;
518 if (out_len < sizeof(resp))
521 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
525 ctx = _ucma_find_context(cmd.id, file);
527 idr_remove(&ctx_idr, ctx->id);
534 wait_for_completion(&ctx->comp);
535 resp.events_reported = ucma_free_ctx(ctx);
537 if (copy_to_user((void __user *)(unsigned long)cmd.response,
538 &resp, sizeof(resp)))
544 static ssize_t ucma_bind_addr(struct ucma_file *file, const char __user *inbuf,
545 int in_len, int out_len)
547 struct rdma_ucm_bind_addr cmd;
548 struct ucma_context *ctx;
551 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
554 ctx = ucma_get_ctx(file, cmd.id);
558 ret = rdma_bind_addr(ctx->cm_id, (struct sockaddr *) &cmd.addr);
563 static ssize_t ucma_resolve_addr(struct ucma_file *file,
564 const char __user *inbuf,
565 int in_len, int out_len)
567 struct rdma_ucm_resolve_addr 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_addr(ctx->cm_id, (struct sockaddr *) &cmd.src_addr,
579 (struct sockaddr *) &cmd.dst_addr,
585 static ssize_t ucma_resolve_route(struct ucma_file *file,
586 const char __user *inbuf,
587 int in_len, int out_len)
589 struct rdma_ucm_resolve_route cmd;
590 struct ucma_context *ctx;
593 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
596 ctx = ucma_get_ctx(file, cmd.id);
600 ret = rdma_resolve_route(ctx->cm_id, cmd.timeout_ms);
605 static void ucma_copy_ib_route(struct rdma_ucm_query_route_resp *resp,
606 struct rdma_route *route)
608 struct rdma_dev_addr *dev_addr;
610 resp->num_paths = route->num_paths;
611 switch (route->num_paths) {
613 dev_addr = &route->addr.dev_addr;
614 rdma_addr_get_dgid(dev_addr,
615 (union ib_gid *) &resp->ib_route[0].dgid);
616 rdma_addr_get_sgid(dev_addr,
617 (union ib_gid *) &resp->ib_route[0].sgid);
618 resp->ib_route[0].pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
621 ib_copy_path_rec_to_user(&resp->ib_route[1],
622 &route->path_rec[1]);
625 ib_copy_path_rec_to_user(&resp->ib_route[0],
626 &route->path_rec[0]);
633 static void ucma_copy_iboe_route(struct rdma_ucm_query_route_resp *resp,
634 struct rdma_route *route)
636 struct rdma_dev_addr *dev_addr;
637 struct net_device *dev;
640 resp->num_paths = route->num_paths;
641 switch (route->num_paths) {
643 dev_addr = &route->addr.dev_addr;
644 dev = dev_get_by_index(&init_net, dev_addr->bound_dev_if);
646 vid = rdma_vlan_dev_vlan_id(dev);
650 iboe_mac_vlan_to_ll((union ib_gid *) &resp->ib_route[0].dgid,
651 dev_addr->dst_dev_addr, vid);
652 iboe_addr_get_sgid(dev_addr,
653 (union ib_gid *) &resp->ib_route[0].sgid);
654 resp->ib_route[0].pkey = cpu_to_be16(0xffff);
657 ib_copy_path_rec_to_user(&resp->ib_route[1],
658 &route->path_rec[1]);
661 ib_copy_path_rec_to_user(&resp->ib_route[0],
662 &route->path_rec[0]);
669 static void ucma_copy_iw_route(struct rdma_ucm_query_route_resp *resp,
670 struct rdma_route *route)
672 struct rdma_dev_addr *dev_addr;
674 dev_addr = &route->addr.dev_addr;
675 rdma_addr_get_dgid(dev_addr, (union ib_gid *) &resp->ib_route[0].dgid);
676 rdma_addr_get_sgid(dev_addr, (union ib_gid *) &resp->ib_route[0].sgid);
679 static ssize_t ucma_query_route(struct ucma_file *file,
680 const char __user *inbuf,
681 int in_len, int out_len)
683 struct rdma_ucm_query_route cmd;
684 struct rdma_ucm_query_route_resp resp;
685 struct ucma_context *ctx;
686 struct sockaddr *addr;
689 if (out_len < sizeof(resp))
692 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
695 ctx = ucma_get_ctx(file, cmd.id);
699 memset(&resp, 0, sizeof resp);
700 addr = (struct sockaddr *) &ctx->cm_id->route.addr.src_addr;
701 memcpy(&resp.src_addr, addr, addr->sa_family == AF_INET ?
702 sizeof(struct sockaddr_in) :
703 sizeof(struct sockaddr_in6));
704 addr = (struct sockaddr *) &ctx->cm_id->route.addr.dst_addr;
705 memcpy(&resp.dst_addr, addr, addr->sa_family == AF_INET ?
706 sizeof(struct sockaddr_in) :
707 sizeof(struct sockaddr_in6));
708 if (!ctx->cm_id->device)
711 resp.node_guid = (__force __u64) ctx->cm_id->device->node_guid;
712 resp.port_num = ctx->cm_id->port_num;
713 switch (rdma_node_get_transport(ctx->cm_id->device->node_type)) {
714 case RDMA_TRANSPORT_IB:
715 switch (rdma_port_get_link_layer(ctx->cm_id->device,
716 ctx->cm_id->port_num)) {
717 case IB_LINK_LAYER_INFINIBAND:
718 ucma_copy_ib_route(&resp, &ctx->cm_id->route);
720 case IB_LINK_LAYER_ETHERNET:
721 ucma_copy_iboe_route(&resp, &ctx->cm_id->route);
727 case RDMA_TRANSPORT_IWARP:
728 ucma_copy_iw_route(&resp, &ctx->cm_id->route);
735 if (copy_to_user((void __user *)(unsigned long)cmd.response,
736 &resp, sizeof(resp)))
743 static void ucma_copy_conn_param(struct rdma_conn_param *dst,
744 struct rdma_ucm_conn_param *src)
746 dst->private_data = src->private_data;
747 dst->private_data_len = src->private_data_len;
748 dst->responder_resources =src->responder_resources;
749 dst->initiator_depth = src->initiator_depth;
750 dst->flow_control = src->flow_control;
751 dst->retry_count = src->retry_count;
752 dst->rnr_retry_count = src->rnr_retry_count;
754 dst->qp_num = src->qp_num;
757 static ssize_t ucma_connect(struct ucma_file *file, const char __user *inbuf,
758 int in_len, int out_len)
760 struct rdma_ucm_connect cmd;
761 struct rdma_conn_param conn_param;
762 struct ucma_context *ctx;
765 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
768 if (!cmd.conn_param.valid)
771 ctx = ucma_get_ctx(file, cmd.id);
775 ucma_copy_conn_param(&conn_param, &cmd.conn_param);
776 ret = rdma_connect(ctx->cm_id, &conn_param);
781 static ssize_t ucma_listen(struct ucma_file *file, const char __user *inbuf,
782 int in_len, int out_len)
784 struct rdma_ucm_listen cmd;
785 struct ucma_context *ctx;
788 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
791 ctx = ucma_get_ctx(file, cmd.id);
795 ctx->backlog = cmd.backlog > 0 && cmd.backlog < max_backlog ?
796 cmd.backlog : max_backlog;
797 ret = rdma_listen(ctx->cm_id, ctx->backlog);
802 static ssize_t ucma_accept(struct ucma_file *file, const char __user *inbuf,
803 int in_len, int out_len)
805 struct rdma_ucm_accept cmd;
806 struct rdma_conn_param conn_param;
807 struct ucma_context *ctx;
810 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
813 ctx = ucma_get_ctx(file, cmd.id);
817 if (cmd.conn_param.valid) {
819 ucma_copy_conn_param(&conn_param, &cmd.conn_param);
820 ret = rdma_accept(ctx->cm_id, &conn_param);
822 ret = rdma_accept(ctx->cm_id, NULL);
828 static ssize_t ucma_reject(struct ucma_file *file, const char __user *inbuf,
829 int in_len, int out_len)
831 struct rdma_ucm_reject cmd;
832 struct ucma_context *ctx;
835 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
838 ctx = ucma_get_ctx(file, cmd.id);
842 ret = rdma_reject(ctx->cm_id, cmd.private_data, cmd.private_data_len);
847 static ssize_t ucma_disconnect(struct ucma_file *file, const char __user *inbuf,
848 int in_len, int out_len)
850 struct rdma_ucm_disconnect cmd;
851 struct ucma_context *ctx;
854 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
857 ctx = ucma_get_ctx(file, cmd.id);
861 ret = rdma_disconnect(ctx->cm_id);
866 static ssize_t ucma_init_qp_attr(struct ucma_file *file,
867 const char __user *inbuf,
868 int in_len, int out_len)
870 struct rdma_ucm_init_qp_attr cmd;
871 struct ib_uverbs_qp_attr resp;
872 struct ucma_context *ctx;
873 struct ib_qp_attr qp_attr;
876 if (out_len < sizeof(resp))
879 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
882 if (cmd.qp_state > IB_QPS_ERR)
885 ctx = ucma_get_ctx(file, cmd.id);
889 resp.qp_attr_mask = 0;
890 memset(&qp_attr, 0, sizeof qp_attr);
891 qp_attr.qp_state = cmd.qp_state;
892 ret = rdma_init_qp_attr(ctx->cm_id, &qp_attr, &resp.qp_attr_mask);
896 ib_copy_qp_attr_to_user(&resp, &qp_attr);
897 if (copy_to_user((void __user *)(unsigned long)cmd.response,
898 &resp, sizeof(resp)))
906 static int ucma_set_option_id(struct ucma_context *ctx, int optname,
907 void *optval, size_t optlen)
912 case RDMA_OPTION_ID_TOS:
913 if (optlen != sizeof(u8)) {
917 rdma_set_service_type(ctx->cm_id, *((u8 *) optval));
919 case RDMA_OPTION_ID_REUSEADDR:
920 if (optlen != sizeof(int)) {
924 ret = rdma_set_reuseaddr(ctx->cm_id, *((int *) optval) ? 1 : 0);
933 static int ucma_set_ib_path(struct ucma_context *ctx,
934 struct ib_path_rec_data *path_data, size_t optlen)
936 struct ib_sa_path_rec sa_path;
937 struct rdma_cm_event event;
940 if (optlen % sizeof(*path_data))
943 for (; optlen; optlen -= sizeof(*path_data), path_data++) {
944 if (path_data->flags == (IB_PATH_GMP | IB_PATH_PRIMARY |
945 IB_PATH_BIDIRECTIONAL))
952 ib_sa_unpack_path(path_data->path_rec, &sa_path);
953 ret = rdma_set_ib_paths(ctx->cm_id, &sa_path, 1);
957 memset(&event, 0, sizeof event);
958 event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
959 return ucma_event_handler(ctx->cm_id, &event);
962 static int ucma_set_option_ib(struct ucma_context *ctx, int optname,
963 void *optval, size_t optlen)
968 case RDMA_OPTION_IB_PATH:
969 ret = ucma_set_ib_path(ctx, optval, optlen);
978 static int ucma_set_option_level(struct ucma_context *ctx, int level,
979 int optname, void *optval, size_t optlen)
985 ret = ucma_set_option_id(ctx, optname, optval, optlen);
988 ret = ucma_set_option_ib(ctx, optname, optval, optlen);
997 static ssize_t ucma_set_option(struct ucma_file *file, const char __user *inbuf,
998 int in_len, int out_len)
1000 struct rdma_ucm_set_option cmd;
1001 struct ucma_context *ctx;
1005 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1008 ctx = ucma_get_ctx(file, cmd.id);
1010 return PTR_ERR(ctx);
1012 if (unlikely(cmd.optlen > KMALLOC_MAX_SIZE))
1015 optval = kmalloc(cmd.optlen, GFP_KERNEL);
1021 if (copy_from_user(optval, (void __user *) (unsigned long) cmd.optval,
1027 ret = ucma_set_option_level(ctx, cmd.level, cmd.optname, optval,
1036 static ssize_t ucma_notify(struct ucma_file *file, const char __user *inbuf,
1037 int in_len, int out_len)
1039 struct rdma_ucm_notify cmd;
1040 struct ucma_context *ctx;
1043 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1046 ctx = ucma_get_ctx(file, cmd.id);
1048 return PTR_ERR(ctx);
1050 ret = rdma_notify(ctx->cm_id, (enum ib_event_type) cmd.event);
1055 static ssize_t ucma_join_multicast(struct ucma_file *file,
1056 const char __user *inbuf,
1057 int in_len, int out_len)
1059 struct rdma_ucm_join_mcast cmd;
1060 struct rdma_ucm_create_id_resp resp;
1061 struct ucma_context *ctx;
1062 struct ucma_multicast *mc;
1065 if (out_len < sizeof(resp))
1068 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1071 ctx = ucma_get_ctx(file, cmd.id);
1073 return PTR_ERR(ctx);
1075 mutex_lock(&file->mut);
1076 mc = ucma_alloc_multicast(ctx);
1083 memcpy(&mc->addr, &cmd.addr, sizeof cmd.addr);
1084 ret = rdma_join_multicast(ctx->cm_id, (struct sockaddr *) &mc->addr, mc);
1089 if (copy_to_user((void __user *)(unsigned long)cmd.response,
1090 &resp, sizeof(resp))) {
1095 mutex_unlock(&file->mut);
1100 rdma_leave_multicast(ctx->cm_id, (struct sockaddr *) &mc->addr);
1101 ucma_cleanup_mc_events(mc);
1104 idr_remove(&multicast_idr, mc->id);
1106 list_del(&mc->list);
1109 mutex_unlock(&file->mut);
1114 static ssize_t ucma_leave_multicast(struct ucma_file *file,
1115 const char __user *inbuf,
1116 int in_len, int out_len)
1118 struct rdma_ucm_destroy_id cmd;
1119 struct rdma_ucm_destroy_id_resp resp;
1120 struct ucma_multicast *mc;
1123 if (out_len < sizeof(resp))
1126 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1130 mc = idr_find(&multicast_idr, cmd.id);
1132 mc = ERR_PTR(-ENOENT);
1133 else if (mc->ctx->file != file)
1134 mc = ERR_PTR(-EINVAL);
1136 idr_remove(&multicast_idr, mc->id);
1137 atomic_inc(&mc->ctx->ref);
1146 rdma_leave_multicast(mc->ctx->cm_id, (struct sockaddr *) &mc->addr);
1147 mutex_lock(&mc->ctx->file->mut);
1148 ucma_cleanup_mc_events(mc);
1149 list_del(&mc->list);
1150 mutex_unlock(&mc->ctx->file->mut);
1152 ucma_put_ctx(mc->ctx);
1153 resp.events_reported = mc->events_reported;
1156 if (copy_to_user((void __user *)(unsigned long)cmd.response,
1157 &resp, sizeof(resp)))
1163 static void ucma_lock_files(struct ucma_file *file1, struct ucma_file *file2)
1165 /* Acquire mutex's based on pointer comparison to prevent deadlock. */
1166 if (file1 < file2) {
1167 mutex_lock(&file1->mut);
1168 mutex_lock(&file2->mut);
1170 mutex_lock(&file2->mut);
1171 mutex_lock(&file1->mut);
1175 static void ucma_unlock_files(struct ucma_file *file1, struct ucma_file *file2)
1177 if (file1 < file2) {
1178 mutex_unlock(&file2->mut);
1179 mutex_unlock(&file1->mut);
1181 mutex_unlock(&file1->mut);
1182 mutex_unlock(&file2->mut);
1186 static void ucma_move_events(struct ucma_context *ctx, struct ucma_file *file)
1188 struct ucma_event *uevent, *tmp;
1190 list_for_each_entry_safe(uevent, tmp, &ctx->file->event_list, list)
1191 if (uevent->ctx == ctx)
1192 list_move_tail(&uevent->list, &file->event_list);
1195 static ssize_t ucma_migrate_id(struct ucma_file *new_file,
1196 const char __user *inbuf,
1197 int in_len, int out_len)
1199 struct rdma_ucm_migrate_id cmd;
1200 struct rdma_ucm_migrate_resp resp;
1201 struct ucma_context *ctx;
1203 struct ucma_file *cur_file;
1206 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
1209 /* Get current fd to protect against it being closed */
1210 filp = fget(cmd.fd);
1214 /* Validate current fd and prevent destruction of id. */
1215 ctx = ucma_get_ctx(filp->private_data, cmd.id);
1221 cur_file = ctx->file;
1222 if (cur_file == new_file) {
1223 resp.events_reported = ctx->events_reported;
1228 * Migrate events between fd's, maintaining order, and avoiding new
1229 * events being added before existing events.
1231 ucma_lock_files(cur_file, new_file);
1234 list_move_tail(&ctx->list, &new_file->ctx_list);
1235 ucma_move_events(ctx, new_file);
1236 ctx->file = new_file;
1237 resp.events_reported = ctx->events_reported;
1240 ucma_unlock_files(cur_file, new_file);
1243 if (copy_to_user((void __user *)(unsigned long)cmd.response,
1244 &resp, sizeof(resp)))
1253 static ssize_t (*ucma_cmd_table[])(struct ucma_file *file,
1254 const char __user *inbuf,
1255 int in_len, int out_len) = {
1256 [RDMA_USER_CM_CMD_CREATE_ID] = ucma_create_id,
1257 [RDMA_USER_CM_CMD_DESTROY_ID] = ucma_destroy_id,
1258 [RDMA_USER_CM_CMD_BIND_ADDR] = ucma_bind_addr,
1259 [RDMA_USER_CM_CMD_RESOLVE_ADDR] = ucma_resolve_addr,
1260 [RDMA_USER_CM_CMD_RESOLVE_ROUTE]= ucma_resolve_route,
1261 [RDMA_USER_CM_CMD_QUERY_ROUTE] = ucma_query_route,
1262 [RDMA_USER_CM_CMD_CONNECT] = ucma_connect,
1263 [RDMA_USER_CM_CMD_LISTEN] = ucma_listen,
1264 [RDMA_USER_CM_CMD_ACCEPT] = ucma_accept,
1265 [RDMA_USER_CM_CMD_REJECT] = ucma_reject,
1266 [RDMA_USER_CM_CMD_DISCONNECT] = ucma_disconnect,
1267 [RDMA_USER_CM_CMD_INIT_QP_ATTR] = ucma_init_qp_attr,
1268 [RDMA_USER_CM_CMD_GET_EVENT] = ucma_get_event,
1269 [RDMA_USER_CM_CMD_GET_OPTION] = NULL,
1270 [RDMA_USER_CM_CMD_SET_OPTION] = ucma_set_option,
1271 [RDMA_USER_CM_CMD_NOTIFY] = ucma_notify,
1272 [RDMA_USER_CM_CMD_JOIN_MCAST] = ucma_join_multicast,
1273 [RDMA_USER_CM_CMD_LEAVE_MCAST] = ucma_leave_multicast,
1274 [RDMA_USER_CM_CMD_MIGRATE_ID] = ucma_migrate_id
1277 static ssize_t ucma_write(struct file *filp, const char __user *buf,
1278 size_t len, loff_t *pos)
1280 struct ucma_file *file = filp->private_data;
1281 struct rdma_ucm_cmd_hdr hdr;
1284 if (WARN_ON_ONCE(!ib_safe_file_access(filp)))
1287 if (len < sizeof(hdr))
1290 if (copy_from_user(&hdr, buf, sizeof(hdr)))
1293 if (hdr.cmd >= ARRAY_SIZE(ucma_cmd_table))
1296 if (hdr.in + sizeof(hdr) > len)
1299 if (!ucma_cmd_table[hdr.cmd])
1302 ret = ucma_cmd_table[hdr.cmd](file, buf + sizeof(hdr), hdr.in, hdr.out);
1309 static unsigned int ucma_poll(struct file *filp, struct poll_table_struct *wait)
1311 struct ucma_file *file = filp->private_data;
1312 unsigned int mask = 0;
1314 poll_wait(filp, &file->poll_wait, wait);
1316 if (!list_empty(&file->event_list))
1317 mask = POLLIN | POLLRDNORM;
1323 * ucma_open() does not need the BKL:
1325 * - no global state is referred to;
1326 * - there is no ioctl method to race against;
1327 * - no further module initialization is required for open to work
1328 * after the device is registered.
1330 static int ucma_open(struct inode *inode, struct file *filp)
1332 struct ucma_file *file;
1334 file = kmalloc(sizeof *file, GFP_KERNEL);
1338 INIT_LIST_HEAD(&file->event_list);
1339 INIT_LIST_HEAD(&file->ctx_list);
1340 init_waitqueue_head(&file->poll_wait);
1341 mutex_init(&file->mut);
1343 filp->private_data = file;
1346 return nonseekable_open(inode, filp);
1349 static int ucma_close(struct inode *inode, struct file *filp)
1351 struct ucma_file *file = filp->private_data;
1352 struct ucma_context *ctx, *tmp;
1354 mutex_lock(&file->mut);
1355 list_for_each_entry_safe(ctx, tmp, &file->ctx_list, list) {
1356 mutex_unlock(&file->mut);
1359 idr_remove(&ctx_idr, ctx->id);
1363 mutex_lock(&file->mut);
1365 mutex_unlock(&file->mut);
1370 static const struct file_operations ucma_fops = {
1371 .owner = THIS_MODULE,
1373 .release = ucma_close,
1374 .write = ucma_write,
1376 .llseek = no_llseek,
1379 static struct miscdevice ucma_misc = {
1380 .minor = MISC_DYNAMIC_MINOR,
1382 .nodename = "infiniband/rdma_cm",
1387 static ssize_t show_abi_version(struct device *dev,
1388 struct device_attribute *attr,
1391 return sprintf(buf, "%d\n", RDMA_USER_CM_ABI_VERSION);
1393 static DEVICE_ATTR(abi_version, S_IRUGO, show_abi_version, NULL);
1395 static int __init ucma_init(void)
1399 ret = misc_register(&ucma_misc);
1403 ret = device_create_file(ucma_misc.this_device, &dev_attr_abi_version);
1405 printk(KERN_ERR "rdma_ucm: couldn't create abi_version attr\n");
1409 ucma_ctl_table_hdr = register_sysctl_paths(ucma_ctl_path, ucma_ctl_table);
1410 if (!ucma_ctl_table_hdr) {
1411 printk(KERN_ERR "rdma_ucm: couldn't register sysctl paths\n");
1417 device_remove_file(ucma_misc.this_device, &dev_attr_abi_version);
1419 misc_deregister(&ucma_misc);
1423 static void __exit ucma_cleanup(void)
1425 unregister_sysctl_table(ucma_ctl_table_hdr);
1426 device_remove_file(ucma_misc.this_device, &dev_attr_abi_version);
1427 misc_deregister(&ucma_misc);
1428 idr_destroy(&ctx_idr);
1431 module_init(ucma_init);
1432 module_exit(ucma_cleanup);