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/mutex.h>
35 #include <linux/poll.h>
36 #include <linux/idr.h>
38 #include <linux/in6.h>
39 #include <linux/miscdevice.h>
41 #include <rdma/rdma_user_cm.h>
42 #include <rdma/ib_marshall.h>
43 #include <rdma/rdma_cm.h>
45 MODULE_AUTHOR("Sean Hefty");
46 MODULE_DESCRIPTION("RDMA Userspace Connection Manager Access");
47 MODULE_LICENSE("Dual BSD/GPL");
50 UCMA_MAX_BACKLOG = 128
56 struct list_head ctx_list;
57 struct list_head event_list;
58 wait_queue_head_t poll_wait;
63 struct completion comp;
68 struct ucma_file *file;
69 struct rdma_cm_id *cm_id;
72 struct list_head list;
76 struct ucma_context *ctx;
77 struct list_head list;
78 struct rdma_cm_id *cm_id;
79 struct rdma_ucm_event_resp resp;
82 static DEFINE_MUTEX(mut);
83 static DEFINE_IDR(ctx_idr);
85 static inline struct ucma_context *_ucma_find_context(int id,
86 struct ucma_file *file)
88 struct ucma_context *ctx;
90 ctx = idr_find(&ctx_idr, id);
92 ctx = ERR_PTR(-ENOENT);
93 else if (ctx->file != file)
94 ctx = ERR_PTR(-EINVAL);
98 static struct ucma_context *ucma_get_ctx(struct ucma_file *file, int id)
100 struct ucma_context *ctx;
103 ctx = _ucma_find_context(id, file);
105 atomic_inc(&ctx->ref);
110 static void ucma_put_ctx(struct ucma_context *ctx)
112 if (atomic_dec_and_test(&ctx->ref))
113 complete(&ctx->comp);
116 static struct ucma_context *ucma_alloc_ctx(struct ucma_file *file)
118 struct ucma_context *ctx;
121 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
125 atomic_set(&ctx->ref, 1);
126 init_completion(&ctx->comp);
130 ret = idr_pre_get(&ctx_idr, GFP_KERNEL);
135 ret = idr_get_new(&ctx_idr, ctx, &ctx->id);
137 } while (ret == -EAGAIN);
142 list_add_tail(&ctx->list, &file->ctx_list);
150 static void ucma_copy_conn_event(struct rdma_ucm_conn_param *dst,
151 struct rdma_conn_param *src)
153 if (src->private_data_len)
154 memcpy(dst->private_data, src->private_data,
155 src->private_data_len);
156 dst->private_data_len = src->private_data_len;
157 dst->responder_resources =src->responder_resources;
158 dst->initiator_depth = src->initiator_depth;
159 dst->flow_control = src->flow_control;
160 dst->retry_count = src->retry_count;
161 dst->rnr_retry_count = src->rnr_retry_count;
163 dst->qp_num = src->qp_num;
166 static void ucma_copy_ud_event(struct rdma_ucm_ud_param *dst,
167 struct rdma_ud_param *src)
169 if (src->private_data_len)
170 memcpy(dst->private_data, src->private_data,
171 src->private_data_len);
172 dst->private_data_len = src->private_data_len;
173 ib_copy_ah_attr_to_user(&dst->ah_attr, &src->ah_attr);
174 dst->qp_num = src->qp_num;
175 dst->qkey = src->qkey;
178 static void ucma_set_event_context(struct ucma_context *ctx,
179 struct rdma_cm_event *event,
180 struct ucma_event *uevent)
183 uevent->resp.uid = ctx->uid;
184 uevent->resp.id = ctx->id;
187 static int ucma_event_handler(struct rdma_cm_id *cm_id,
188 struct rdma_cm_event *event)
190 struct ucma_event *uevent;
191 struct ucma_context *ctx = cm_id->context;
194 uevent = kzalloc(sizeof(*uevent), GFP_KERNEL);
196 return event->event == RDMA_CM_EVENT_CONNECT_REQUEST;
198 uevent->cm_id = cm_id;
199 ucma_set_event_context(ctx, event, uevent);
200 uevent->resp.event = event->event;
201 uevent->resp.status = event->status;
202 if (cm_id->ps == RDMA_PS_UDP)
203 ucma_copy_ud_event(&uevent->resp.param.ud, &event->param.ud);
205 ucma_copy_conn_event(&uevent->resp.param.conn,
208 mutex_lock(&ctx->file->mut);
209 if (event->event == RDMA_CM_EVENT_CONNECT_REQUEST) {
216 list_add_tail(&uevent->list, &ctx->file->event_list);
217 wake_up_interruptible(&ctx->file->poll_wait);
219 mutex_unlock(&ctx->file->mut);
223 static ssize_t ucma_get_event(struct ucma_file *file, const char __user *inbuf,
224 int in_len, int out_len)
226 struct ucma_context *ctx;
227 struct rdma_ucm_get_event cmd;
228 struct ucma_event *uevent;
232 if (out_len < sizeof uevent->resp)
235 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
238 mutex_lock(&file->mut);
239 while (list_empty(&file->event_list)) {
240 if (file->filp->f_flags & O_NONBLOCK) {
245 if (signal_pending(current)) {
250 prepare_to_wait(&file->poll_wait, &wait, TASK_INTERRUPTIBLE);
251 mutex_unlock(&file->mut);
253 mutex_lock(&file->mut);
254 finish_wait(&file->poll_wait, &wait);
260 uevent = list_entry(file->event_list.next, struct ucma_event, list);
262 if (uevent->resp.event == RDMA_CM_EVENT_CONNECT_REQUEST) {
263 ctx = ucma_alloc_ctx(file);
268 uevent->ctx->backlog++;
269 ctx->cm_id = uevent->cm_id;
270 ctx->cm_id->context = ctx;
271 uevent->resp.id = ctx->id;
274 if (copy_to_user((void __user *)(unsigned long)cmd.response,
275 &uevent->resp, sizeof uevent->resp)) {
280 list_del(&uevent->list);
281 uevent->ctx->events_reported++;
284 mutex_unlock(&file->mut);
288 static ssize_t ucma_create_id(struct ucma_file *file,
289 const char __user *inbuf,
290 int in_len, int out_len)
292 struct rdma_ucm_create_id cmd;
293 struct rdma_ucm_create_id_resp resp;
294 struct ucma_context *ctx;
297 if (out_len < sizeof(resp))
300 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
303 mutex_lock(&file->mut);
304 ctx = ucma_alloc_ctx(file);
305 mutex_unlock(&file->mut);
310 ctx->cm_id = rdma_create_id(ucma_event_handler, ctx, cmd.ps);
311 if (IS_ERR(ctx->cm_id)) {
312 ret = PTR_ERR(ctx->cm_id);
317 if (copy_to_user((void __user *)(unsigned long)cmd.response,
318 &resp, sizeof(resp))) {
325 rdma_destroy_id(ctx->cm_id);
328 idr_remove(&ctx_idr, ctx->id);
334 static void ucma_cleanup_events(struct ucma_context *ctx)
336 struct ucma_event *uevent, *tmp;
338 list_for_each_entry_safe(uevent, tmp, &ctx->file->event_list, list) {
339 if (uevent->ctx != ctx)
342 list_del(&uevent->list);
344 /* clear incoming connections. */
345 if (uevent->resp.event == RDMA_CM_EVENT_CONNECT_REQUEST)
346 rdma_destroy_id(uevent->cm_id);
352 static int ucma_free_ctx(struct ucma_context *ctx)
356 /* No new events will be generated after destroying the id. */
357 rdma_destroy_id(ctx->cm_id);
359 /* Cleanup events not yet reported to the user. */
360 mutex_lock(&ctx->file->mut);
361 ucma_cleanup_events(ctx);
362 list_del(&ctx->list);
363 mutex_unlock(&ctx->file->mut);
365 events_reported = ctx->events_reported;
367 return events_reported;
370 static ssize_t ucma_destroy_id(struct ucma_file *file, const char __user *inbuf,
371 int in_len, int out_len)
373 struct rdma_ucm_destroy_id cmd;
374 struct rdma_ucm_destroy_id_resp resp;
375 struct ucma_context *ctx;
378 if (out_len < sizeof(resp))
381 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
385 ctx = _ucma_find_context(cmd.id, file);
387 idr_remove(&ctx_idr, ctx->id);
394 wait_for_completion(&ctx->comp);
395 resp.events_reported = ucma_free_ctx(ctx);
397 if (copy_to_user((void __user *)(unsigned long)cmd.response,
398 &resp, sizeof(resp)))
404 static ssize_t ucma_bind_addr(struct ucma_file *file, const char __user *inbuf,
405 int in_len, int out_len)
407 struct rdma_ucm_bind_addr cmd;
408 struct ucma_context *ctx;
411 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
414 ctx = ucma_get_ctx(file, cmd.id);
418 ret = rdma_bind_addr(ctx->cm_id, (struct sockaddr *) &cmd.addr);
423 static ssize_t ucma_resolve_addr(struct ucma_file *file,
424 const char __user *inbuf,
425 int in_len, int out_len)
427 struct rdma_ucm_resolve_addr cmd;
428 struct ucma_context *ctx;
431 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
434 ctx = ucma_get_ctx(file, cmd.id);
438 ret = rdma_resolve_addr(ctx->cm_id, (struct sockaddr *) &cmd.src_addr,
439 (struct sockaddr *) &cmd.dst_addr,
445 static ssize_t ucma_resolve_route(struct ucma_file *file,
446 const char __user *inbuf,
447 int in_len, int out_len)
449 struct rdma_ucm_resolve_route cmd;
450 struct ucma_context *ctx;
453 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
456 ctx = ucma_get_ctx(file, cmd.id);
460 ret = rdma_resolve_route(ctx->cm_id, cmd.timeout_ms);
465 static void ucma_copy_ib_route(struct rdma_ucm_query_route_resp *resp,
466 struct rdma_route *route)
468 struct rdma_dev_addr *dev_addr;
470 resp->num_paths = route->num_paths;
471 switch (route->num_paths) {
473 dev_addr = &route->addr.dev_addr;
474 ib_addr_get_dgid(dev_addr,
475 (union ib_gid *) &resp->ib_route[0].dgid);
476 ib_addr_get_sgid(dev_addr,
477 (union ib_gid *) &resp->ib_route[0].sgid);
478 resp->ib_route[0].pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
481 ib_copy_path_rec_to_user(&resp->ib_route[1],
482 &route->path_rec[1]);
485 ib_copy_path_rec_to_user(&resp->ib_route[0],
486 &route->path_rec[0]);
493 static ssize_t ucma_query_route(struct ucma_file *file,
494 const char __user *inbuf,
495 int in_len, int out_len)
497 struct rdma_ucm_query_route cmd;
498 struct rdma_ucm_query_route_resp resp;
499 struct ucma_context *ctx;
500 struct sockaddr *addr;
503 if (out_len < sizeof(resp))
506 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
509 ctx = ucma_get_ctx(file, cmd.id);
513 memset(&resp, 0, sizeof resp);
514 addr = &ctx->cm_id->route.addr.src_addr;
515 memcpy(&resp.src_addr, addr, addr->sa_family == AF_INET ?
516 sizeof(struct sockaddr_in) :
517 sizeof(struct sockaddr_in6));
518 addr = &ctx->cm_id->route.addr.dst_addr;
519 memcpy(&resp.dst_addr, addr, addr->sa_family == AF_INET ?
520 sizeof(struct sockaddr_in) :
521 sizeof(struct sockaddr_in6));
522 if (!ctx->cm_id->device)
525 resp.node_guid = ctx->cm_id->device->node_guid;
526 resp.port_num = ctx->cm_id->port_num;
527 switch (rdma_node_get_transport(ctx->cm_id->device->node_type)) {
528 case RDMA_TRANSPORT_IB:
529 ucma_copy_ib_route(&resp, &ctx->cm_id->route);
536 if (copy_to_user((void __user *)(unsigned long)cmd.response,
537 &resp, sizeof(resp)))
544 static void ucma_copy_conn_param(struct rdma_conn_param *dst,
545 struct rdma_ucm_conn_param *src)
547 dst->private_data = src->private_data;
548 dst->private_data_len = src->private_data_len;
549 dst->responder_resources =src->responder_resources;
550 dst->initiator_depth = src->initiator_depth;
551 dst->flow_control = src->flow_control;
552 dst->retry_count = src->retry_count;
553 dst->rnr_retry_count = src->rnr_retry_count;
555 dst->qp_num = src->qp_num;
558 static ssize_t ucma_connect(struct ucma_file *file, const char __user *inbuf,
559 int in_len, int out_len)
561 struct rdma_ucm_connect cmd;
562 struct rdma_conn_param conn_param;
563 struct ucma_context *ctx;
566 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
569 if (!cmd.conn_param.valid)
572 ctx = ucma_get_ctx(file, cmd.id);
576 ucma_copy_conn_param(&conn_param, &cmd.conn_param);
577 ret = rdma_connect(ctx->cm_id, &conn_param);
582 static ssize_t ucma_listen(struct ucma_file *file, const char __user *inbuf,
583 int in_len, int out_len)
585 struct rdma_ucm_listen cmd;
586 struct ucma_context *ctx;
589 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
592 ctx = ucma_get_ctx(file, cmd.id);
596 ctx->backlog = cmd.backlog > 0 && cmd.backlog < UCMA_MAX_BACKLOG ?
597 cmd.backlog : UCMA_MAX_BACKLOG;
598 ret = rdma_listen(ctx->cm_id, ctx->backlog);
603 static ssize_t ucma_accept(struct ucma_file *file, const char __user *inbuf,
604 int in_len, int out_len)
606 struct rdma_ucm_accept cmd;
607 struct rdma_conn_param conn_param;
608 struct ucma_context *ctx;
611 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
614 ctx = ucma_get_ctx(file, cmd.id);
618 if (cmd.conn_param.valid) {
620 ucma_copy_conn_param(&conn_param, &cmd.conn_param);
621 ret = rdma_accept(ctx->cm_id, &conn_param);
623 ret = rdma_accept(ctx->cm_id, NULL);
629 static ssize_t ucma_reject(struct ucma_file *file, const char __user *inbuf,
630 int in_len, int out_len)
632 struct rdma_ucm_reject cmd;
633 struct ucma_context *ctx;
636 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
639 ctx = ucma_get_ctx(file, cmd.id);
643 ret = rdma_reject(ctx->cm_id, cmd.private_data, cmd.private_data_len);
648 static ssize_t ucma_disconnect(struct ucma_file *file, const char __user *inbuf,
649 int in_len, int out_len)
651 struct rdma_ucm_disconnect cmd;
652 struct ucma_context *ctx;
655 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
658 ctx = ucma_get_ctx(file, cmd.id);
662 ret = rdma_disconnect(ctx->cm_id);
667 static ssize_t ucma_init_qp_attr(struct ucma_file *file,
668 const char __user *inbuf,
669 int in_len, int out_len)
671 struct rdma_ucm_init_qp_attr cmd;
672 struct ib_uverbs_qp_attr resp;
673 struct ucma_context *ctx;
674 struct ib_qp_attr qp_attr;
677 if (out_len < sizeof(resp))
680 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
683 ctx = ucma_get_ctx(file, cmd.id);
687 resp.qp_attr_mask = 0;
688 memset(&qp_attr, 0, sizeof qp_attr);
689 qp_attr.qp_state = cmd.qp_state;
690 ret = rdma_init_qp_attr(ctx->cm_id, &qp_attr, &resp.qp_attr_mask);
694 ib_copy_qp_attr_to_user(&resp, &qp_attr);
695 if (copy_to_user((void __user *)(unsigned long)cmd.response,
696 &resp, sizeof(resp)))
704 static ssize_t ucma_notify(struct ucma_file *file, const char __user *inbuf,
705 int in_len, int out_len)
707 struct rdma_ucm_notify cmd;
708 struct ucma_context *ctx;
711 if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
714 ctx = ucma_get_ctx(file, cmd.id);
718 ret = rdma_notify(ctx->cm_id, (enum ib_event_type) cmd.event);
723 static ssize_t (*ucma_cmd_table[])(struct ucma_file *file,
724 const char __user *inbuf,
725 int in_len, int out_len) = {
726 [RDMA_USER_CM_CMD_CREATE_ID] = ucma_create_id,
727 [RDMA_USER_CM_CMD_DESTROY_ID] = ucma_destroy_id,
728 [RDMA_USER_CM_CMD_BIND_ADDR] = ucma_bind_addr,
729 [RDMA_USER_CM_CMD_RESOLVE_ADDR] = ucma_resolve_addr,
730 [RDMA_USER_CM_CMD_RESOLVE_ROUTE]= ucma_resolve_route,
731 [RDMA_USER_CM_CMD_QUERY_ROUTE] = ucma_query_route,
732 [RDMA_USER_CM_CMD_CONNECT] = ucma_connect,
733 [RDMA_USER_CM_CMD_LISTEN] = ucma_listen,
734 [RDMA_USER_CM_CMD_ACCEPT] = ucma_accept,
735 [RDMA_USER_CM_CMD_REJECT] = ucma_reject,
736 [RDMA_USER_CM_CMD_DISCONNECT] = ucma_disconnect,
737 [RDMA_USER_CM_CMD_INIT_QP_ATTR] = ucma_init_qp_attr,
738 [RDMA_USER_CM_CMD_GET_EVENT] = ucma_get_event,
739 [RDMA_USER_CM_CMD_GET_OPTION] = NULL,
740 [RDMA_USER_CM_CMD_SET_OPTION] = NULL,
741 [RDMA_USER_CM_CMD_NOTIFY] = ucma_notify,
744 static ssize_t ucma_write(struct file *filp, const char __user *buf,
745 size_t len, loff_t *pos)
747 struct ucma_file *file = filp->private_data;
748 struct rdma_ucm_cmd_hdr hdr;
751 if (len < sizeof(hdr))
754 if (copy_from_user(&hdr, buf, sizeof(hdr)))
757 if (hdr.cmd < 0 || hdr.cmd >= ARRAY_SIZE(ucma_cmd_table))
760 if (hdr.in + sizeof(hdr) > len)
763 if (!ucma_cmd_table[hdr.cmd])
766 ret = ucma_cmd_table[hdr.cmd](file, buf + sizeof(hdr), hdr.in, hdr.out);
773 static unsigned int ucma_poll(struct file *filp, struct poll_table_struct *wait)
775 struct ucma_file *file = filp->private_data;
776 unsigned int mask = 0;
778 poll_wait(filp, &file->poll_wait, wait);
780 if (!list_empty(&file->event_list))
781 mask = POLLIN | POLLRDNORM;
786 static int ucma_open(struct inode *inode, struct file *filp)
788 struct ucma_file *file;
790 file = kmalloc(sizeof *file, GFP_KERNEL);
794 INIT_LIST_HEAD(&file->event_list);
795 INIT_LIST_HEAD(&file->ctx_list);
796 init_waitqueue_head(&file->poll_wait);
797 mutex_init(&file->mut);
799 filp->private_data = file;
804 static int ucma_close(struct inode *inode, struct file *filp)
806 struct ucma_file *file = filp->private_data;
807 struct ucma_context *ctx, *tmp;
809 mutex_lock(&file->mut);
810 list_for_each_entry_safe(ctx, tmp, &file->ctx_list, list) {
811 mutex_unlock(&file->mut);
814 idr_remove(&ctx_idr, ctx->id);
818 mutex_lock(&file->mut);
820 mutex_unlock(&file->mut);
825 static struct file_operations ucma_fops = {
826 .owner = THIS_MODULE,
828 .release = ucma_close,
833 static struct miscdevice ucma_misc = {
834 .minor = MISC_DYNAMIC_MINOR,
839 static ssize_t show_abi_version(struct device *dev,
840 struct device_attribute *attr,
843 return sprintf(buf, "%d\n", RDMA_USER_CM_ABI_VERSION);
845 static DEVICE_ATTR(abi_version, S_IRUGO, show_abi_version, NULL);
847 static int __init ucma_init(void)
851 ret = misc_register(&ucma_misc);
855 ret = device_create_file(ucma_misc.this_device, &dev_attr_abi_version);
857 printk(KERN_ERR "rdma_ucm: couldn't create abi_version attr\n");
862 misc_deregister(&ucma_misc);
866 static void __exit ucma_cleanup(void)
868 device_remove_file(ucma_misc.this_device, &dev_attr_abi_version);
869 misc_deregister(&ucma_misc);
870 idr_destroy(&ctx_idr);
873 module_init(ucma_init);
874 module_exit(ucma_cleanup);