2 * Copyright (c) 2004 Topspin Communications. All rights reserved.
3 * Copyright (c) 2005 Voltaire, Inc. All rights reserved.
4 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
5 * Copyright (c) 2008 Cisco. All rights reserved.
7 * This software is available to you under a choice of one of two
8 * licenses. You may choose to be licensed under the terms of the GNU
9 * General Public License (GPL) Version 2, available from the file
10 * COPYING in the main directory of this source tree, or the
11 * OpenIB.org BSD license below:
13 * Redistribution and use in source and binary forms, with or
14 * without modification, are permitted provided that the following
17 * - Redistributions of source code must retain the above
18 * copyright notice, this list of conditions and the following
21 * - Redistributions in binary form must reproduce the above
22 * copyright notice, this list of conditions and the following
23 * disclaimer in the documentation and/or other materials
24 * provided with the distribution.
26 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
27 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
28 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
29 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
30 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
31 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
32 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
36 #define pr_fmt(fmt) "user_mad: " fmt
38 #include <linux/module.h>
39 #include <linux/init.h>
40 #include <linux/device.h>
41 #include <linux/err.h>
43 #include <linux/cdev.h>
44 #include <linux/dma-mapping.h>
45 #include <linux/poll.h>
46 #include <linux/mutex.h>
47 #include <linux/kref.h>
48 #include <linux/compat.h>
49 #include <linux/sched.h>
50 #include <linux/semaphore.h>
51 #include <linux/slab.h>
52 #include <linux/nospec.h>
54 #include <asm/uaccess.h>
56 #include <rdma/ib_mad.h>
57 #include <rdma/ib_user_mad.h>
59 MODULE_AUTHOR("Roland Dreier");
60 MODULE_DESCRIPTION("InfiniBand userspace MAD packet access");
61 MODULE_LICENSE("Dual BSD/GPL");
64 IB_UMAD_MAX_PORTS = 64,
65 IB_UMAD_MAX_AGENTS = 32,
68 IB_UMAD_MINOR_BASE = 0
72 * Our lifetime rules for these structs are the following:
73 * device special file is opened, we take a reference on the
74 * ib_umad_port's struct ib_umad_device. We drop these
75 * references in the corresponding close().
77 * In addition to references coming from open character devices, there
78 * is one more reference to each ib_umad_device representing the
79 * module's reference taken when allocating the ib_umad_device in
82 * When destroying an ib_umad_device, we drop the module's reference.
90 struct device *sm_dev;
91 struct semaphore sm_sem;
93 struct mutex file_mutex;
94 struct list_head file_list;
96 struct ib_device *ib_dev;
97 struct ib_umad_device *umad_dev;
102 struct ib_umad_device {
104 struct ib_umad_port port[0];
107 struct ib_umad_file {
109 struct ib_umad_port *port;
110 struct list_head recv_list;
111 struct list_head send_list;
112 struct list_head port_list;
113 spinlock_t send_lock;
114 wait_queue_head_t recv_wait;
115 struct ib_mad_agent *agent[IB_UMAD_MAX_AGENTS];
121 struct ib_umad_packet {
122 struct ib_mad_send_buf *msg;
123 struct ib_mad_recv_wc *recv_wc;
124 struct list_head list;
126 struct ib_user_mad mad;
129 static struct class *umad_class;
131 static const dev_t base_dev = MKDEV(IB_UMAD_MAJOR, IB_UMAD_MINOR_BASE);
133 static DEFINE_SPINLOCK(port_lock);
134 static DECLARE_BITMAP(dev_map, IB_UMAD_MAX_PORTS);
136 static void ib_umad_add_one(struct ib_device *device);
137 static void ib_umad_remove_one(struct ib_device *device, void *client_data);
139 static void ib_umad_release_dev(struct kobject *kobj)
141 struct ib_umad_device *dev =
142 container_of(kobj, struct ib_umad_device, kobj);
147 static struct kobj_type ib_umad_dev_ktype = {
148 .release = ib_umad_release_dev,
151 static int hdr_size(struct ib_umad_file *file)
153 return file->use_pkey_index ? sizeof (struct ib_user_mad_hdr) :
154 sizeof (struct ib_user_mad_hdr_old);
157 /* caller must hold file->mutex */
158 static struct ib_mad_agent *__get_agent(struct ib_umad_file *file, int id)
160 return file->agents_dead ? NULL : file->agent[id];
163 static int queue_packet(struct ib_umad_file *file,
164 struct ib_mad_agent *agent,
165 struct ib_umad_packet *packet)
169 mutex_lock(&file->mutex);
171 for (packet->mad.hdr.id = 0;
172 packet->mad.hdr.id < IB_UMAD_MAX_AGENTS;
173 packet->mad.hdr.id++)
174 if (agent == __get_agent(file, packet->mad.hdr.id)) {
175 list_add_tail(&packet->list, &file->recv_list);
176 wake_up_interruptible(&file->recv_wait);
181 mutex_unlock(&file->mutex);
186 static void dequeue_send(struct ib_umad_file *file,
187 struct ib_umad_packet *packet)
189 spin_lock_irq(&file->send_lock);
190 list_del(&packet->list);
191 spin_unlock_irq(&file->send_lock);
194 static void send_handler(struct ib_mad_agent *agent,
195 struct ib_mad_send_wc *send_wc)
197 struct ib_umad_file *file = agent->context;
198 struct ib_umad_packet *packet = send_wc->send_buf->context[0];
200 dequeue_send(file, packet);
201 ib_destroy_ah(packet->msg->ah);
202 ib_free_send_mad(packet->msg);
204 if (send_wc->status == IB_WC_RESP_TIMEOUT_ERR) {
205 packet->length = IB_MGMT_MAD_HDR;
206 packet->mad.hdr.status = ETIMEDOUT;
207 if (!queue_packet(file, agent, packet))
213 static void recv_handler(struct ib_mad_agent *agent,
214 struct ib_mad_send_buf *send_buf,
215 struct ib_mad_recv_wc *mad_recv_wc)
217 struct ib_umad_file *file = agent->context;
218 struct ib_umad_packet *packet;
220 if (mad_recv_wc->wc->status != IB_WC_SUCCESS)
223 packet = kzalloc(sizeof *packet, GFP_KERNEL);
227 packet->length = mad_recv_wc->mad_len;
228 packet->recv_wc = mad_recv_wc;
230 packet->mad.hdr.status = 0;
231 packet->mad.hdr.length = hdr_size(file) + mad_recv_wc->mad_len;
232 packet->mad.hdr.qpn = cpu_to_be32(mad_recv_wc->wc->src_qp);
233 packet->mad.hdr.lid = cpu_to_be16(mad_recv_wc->wc->slid);
234 packet->mad.hdr.sl = mad_recv_wc->wc->sl;
235 packet->mad.hdr.path_bits = mad_recv_wc->wc->dlid_path_bits;
236 packet->mad.hdr.pkey_index = mad_recv_wc->wc->pkey_index;
237 packet->mad.hdr.grh_present = !!(mad_recv_wc->wc->wc_flags & IB_WC_GRH);
238 if (packet->mad.hdr.grh_present) {
239 struct ib_ah_attr ah_attr;
241 ib_init_ah_from_wc(agent->device, agent->port_num,
242 mad_recv_wc->wc, mad_recv_wc->recv_buf.grh,
245 packet->mad.hdr.gid_index = ah_attr.grh.sgid_index;
246 packet->mad.hdr.hop_limit = ah_attr.grh.hop_limit;
247 packet->mad.hdr.traffic_class = ah_attr.grh.traffic_class;
248 memcpy(packet->mad.hdr.gid, &ah_attr.grh.dgid, 16);
249 packet->mad.hdr.flow_label = cpu_to_be32(ah_attr.grh.flow_label);
252 if (queue_packet(file, agent, packet))
259 ib_free_recv_mad(mad_recv_wc);
262 static ssize_t copy_recv_mad(struct ib_umad_file *file, char __user *buf,
263 struct ib_umad_packet *packet, size_t count)
265 struct ib_mad_recv_buf *recv_buf;
266 int left, seg_payload, offset, max_seg_payload;
269 recv_buf = &packet->recv_wc->recv_buf;
270 seg_size = packet->recv_wc->mad_seg_size;
272 /* We need enough room to copy the first (or only) MAD segment. */
273 if ((packet->length <= seg_size &&
274 count < hdr_size(file) + packet->length) ||
275 (packet->length > seg_size &&
276 count < hdr_size(file) + seg_size))
279 if (copy_to_user(buf, &packet->mad, hdr_size(file)))
282 buf += hdr_size(file);
283 seg_payload = min_t(int, packet->length, seg_size);
284 if (copy_to_user(buf, recv_buf->mad, seg_payload))
287 if (seg_payload < packet->length) {
289 * Multipacket RMPP MAD message. Copy remainder of message.
290 * Note that last segment may have a shorter payload.
292 if (count < hdr_size(file) + packet->length) {
294 * The buffer is too small, return the first RMPP segment,
295 * which includes the RMPP message length.
299 offset = ib_get_mad_data_offset(recv_buf->mad->mad_hdr.mgmt_class);
300 max_seg_payload = seg_size - offset;
302 for (left = packet->length - seg_payload, buf += seg_payload;
303 left; left -= seg_payload, buf += seg_payload) {
304 recv_buf = container_of(recv_buf->list.next,
305 struct ib_mad_recv_buf, list);
306 seg_payload = min(left, max_seg_payload);
307 if (copy_to_user(buf, ((void *) recv_buf->mad) + offset,
312 return hdr_size(file) + packet->length;
315 static ssize_t copy_send_mad(struct ib_umad_file *file, char __user *buf,
316 struct ib_umad_packet *packet, size_t count)
318 ssize_t size = hdr_size(file) + packet->length;
323 if (copy_to_user(buf, &packet->mad, hdr_size(file)))
326 buf += hdr_size(file);
328 if (copy_to_user(buf, packet->mad.data, packet->length))
334 static ssize_t ib_umad_read(struct file *filp, char __user *buf,
335 size_t count, loff_t *pos)
337 struct ib_umad_file *file = filp->private_data;
338 struct ib_umad_packet *packet;
341 if (count < hdr_size(file))
344 mutex_lock(&file->mutex);
346 if (file->agents_dead) {
347 mutex_unlock(&file->mutex);
351 while (list_empty(&file->recv_list)) {
352 mutex_unlock(&file->mutex);
354 if (filp->f_flags & O_NONBLOCK)
357 if (wait_event_interruptible(file->recv_wait,
358 !list_empty(&file->recv_list)))
361 mutex_lock(&file->mutex);
364 packet = list_entry(file->recv_list.next, struct ib_umad_packet, list);
365 list_del(&packet->list);
367 mutex_unlock(&file->mutex);
370 ret = copy_recv_mad(file, buf, packet, count);
372 ret = copy_send_mad(file, buf, packet, count);
376 mutex_lock(&file->mutex);
377 list_add(&packet->list, &file->recv_list);
378 mutex_unlock(&file->mutex);
381 ib_free_recv_mad(packet->recv_wc);
387 static int copy_rmpp_mad(struct ib_mad_send_buf *msg, const char __user *buf)
391 /* Copy class specific header */
392 if ((msg->hdr_len > IB_MGMT_RMPP_HDR) &&
393 copy_from_user(msg->mad + IB_MGMT_RMPP_HDR, buf + IB_MGMT_RMPP_HDR,
394 msg->hdr_len - IB_MGMT_RMPP_HDR))
397 /* All headers are in place. Copy data segments. */
398 for (seg = 1, left = msg->data_len, buf += msg->hdr_len; left > 0;
399 seg++, left -= msg->seg_size, buf += msg->seg_size) {
400 if (copy_from_user(ib_get_rmpp_segment(msg, seg), buf,
401 min(left, msg->seg_size)))
407 static int same_destination(struct ib_user_mad_hdr *hdr1,
408 struct ib_user_mad_hdr *hdr2)
410 if (!hdr1->grh_present && !hdr2->grh_present)
411 return (hdr1->lid == hdr2->lid);
413 if (hdr1->grh_present && hdr2->grh_present)
414 return !memcmp(hdr1->gid, hdr2->gid, 16);
419 static int is_duplicate(struct ib_umad_file *file,
420 struct ib_umad_packet *packet)
422 struct ib_umad_packet *sent_packet;
423 struct ib_mad_hdr *sent_hdr, *hdr;
425 hdr = (struct ib_mad_hdr *) packet->mad.data;
426 list_for_each_entry(sent_packet, &file->send_list, list) {
427 sent_hdr = (struct ib_mad_hdr *) sent_packet->mad.data;
429 if ((hdr->tid != sent_hdr->tid) ||
430 (hdr->mgmt_class != sent_hdr->mgmt_class))
434 * No need to be overly clever here. If two new operations have
435 * the same TID, reject the second as a duplicate. This is more
436 * restrictive than required by the spec.
438 if (!ib_response_mad(hdr)) {
439 if (!ib_response_mad(sent_hdr))
442 } else if (!ib_response_mad(sent_hdr))
445 if (same_destination(&packet->mad.hdr, &sent_packet->mad.hdr))
452 static ssize_t ib_umad_write(struct file *filp, const char __user *buf,
453 size_t count, loff_t *pos)
455 struct ib_umad_file *file = filp->private_data;
456 struct ib_umad_packet *packet;
457 struct ib_mad_agent *agent;
458 struct ib_ah_attr ah_attr;
460 struct ib_rmpp_mad *rmpp_mad;
462 int ret, data_len, hdr_len, copy_offset, rmpp_active;
465 if (count < hdr_size(file) + IB_MGMT_RMPP_HDR)
468 packet = kzalloc(sizeof *packet + IB_MGMT_RMPP_HDR, GFP_KERNEL);
472 if (copy_from_user(&packet->mad, buf, hdr_size(file))) {
477 if (packet->mad.hdr.id >= IB_UMAD_MAX_AGENTS) {
482 buf += hdr_size(file);
484 if (copy_from_user(packet->mad.data, buf, IB_MGMT_RMPP_HDR)) {
489 mutex_lock(&file->mutex);
491 agent = __get_agent(file, packet->mad.hdr.id);
497 memset(&ah_attr, 0, sizeof ah_attr);
498 ah_attr.dlid = be16_to_cpu(packet->mad.hdr.lid);
499 ah_attr.sl = packet->mad.hdr.sl;
500 ah_attr.src_path_bits = packet->mad.hdr.path_bits;
501 ah_attr.port_num = file->port->port_num;
502 if (packet->mad.hdr.grh_present) {
503 ah_attr.ah_flags = IB_AH_GRH;
504 memcpy(ah_attr.grh.dgid.raw, packet->mad.hdr.gid, 16);
505 ah_attr.grh.sgid_index = packet->mad.hdr.gid_index;
506 ah_attr.grh.flow_label = be32_to_cpu(packet->mad.hdr.flow_label);
507 ah_attr.grh.hop_limit = packet->mad.hdr.hop_limit;
508 ah_attr.grh.traffic_class = packet->mad.hdr.traffic_class;
511 ah = ib_create_ah(agent->qp->pd, &ah_attr);
517 rmpp_mad = (struct ib_rmpp_mad *) packet->mad.data;
518 hdr_len = ib_get_mad_data_offset(rmpp_mad->mad_hdr.mgmt_class);
520 if (ib_is_mad_class_rmpp(rmpp_mad->mad_hdr.mgmt_class)
521 && ib_mad_kernel_rmpp_agent(agent)) {
522 copy_offset = IB_MGMT_RMPP_HDR;
523 rmpp_active = ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) &
524 IB_MGMT_RMPP_FLAG_ACTIVE;
526 copy_offset = IB_MGMT_MAD_HDR;
530 base_version = ((struct ib_mad_hdr *)&packet->mad.data)->base_version;
531 data_len = count - hdr_size(file) - hdr_len;
532 packet->msg = ib_create_send_mad(agent,
533 be32_to_cpu(packet->mad.hdr.qpn),
534 packet->mad.hdr.pkey_index, rmpp_active,
535 hdr_len, data_len, GFP_KERNEL,
537 if (IS_ERR(packet->msg)) {
538 ret = PTR_ERR(packet->msg);
542 packet->msg->ah = ah;
543 packet->msg->timeout_ms = packet->mad.hdr.timeout_ms;
544 packet->msg->retries = packet->mad.hdr.retries;
545 packet->msg->context[0] = packet;
547 /* Copy MAD header. Any RMPP header is already in place. */
548 memcpy(packet->msg->mad, packet->mad.data, IB_MGMT_MAD_HDR);
551 if (copy_from_user(packet->msg->mad + copy_offset,
553 hdr_len + data_len - copy_offset)) {
558 ret = copy_rmpp_mad(packet->msg, buf);
564 * Set the high-order part of the transaction ID to make MADs from
565 * different agents unique, and allow routing responses back to the
566 * original requestor.
568 if (!ib_response_mad(packet->msg->mad)) {
569 tid = &((struct ib_mad_hdr *) packet->msg->mad)->tid;
570 *tid = cpu_to_be64(((u64) agent->hi_tid) << 32 |
571 (be64_to_cpup(tid) & 0xffffffff));
572 rmpp_mad->mad_hdr.tid = *tid;
575 if (!ib_mad_kernel_rmpp_agent(agent)
576 && ib_is_mad_class_rmpp(rmpp_mad->mad_hdr.mgmt_class)
577 && (ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) & IB_MGMT_RMPP_FLAG_ACTIVE)) {
578 spin_lock_irq(&file->send_lock);
579 list_add_tail(&packet->list, &file->send_list);
580 spin_unlock_irq(&file->send_lock);
582 spin_lock_irq(&file->send_lock);
583 ret = is_duplicate(file, packet);
585 list_add_tail(&packet->list, &file->send_list);
586 spin_unlock_irq(&file->send_lock);
593 ret = ib_post_send_mad(packet->msg, NULL);
597 mutex_unlock(&file->mutex);
601 dequeue_send(file, packet);
603 ib_free_send_mad(packet->msg);
607 mutex_unlock(&file->mutex);
613 static unsigned int ib_umad_poll(struct file *filp, struct poll_table_struct *wait)
615 struct ib_umad_file *file = filp->private_data;
617 /* we will always be able to post a MAD send */
618 unsigned int mask = POLLOUT | POLLWRNORM;
620 poll_wait(filp, &file->recv_wait, wait);
622 if (!list_empty(&file->recv_list))
623 mask |= POLLIN | POLLRDNORM;
628 static int ib_umad_reg_agent(struct ib_umad_file *file, void __user *arg,
629 int compat_method_mask)
631 struct ib_user_mad_reg_req ureq;
632 struct ib_mad_reg_req req;
633 struct ib_mad_agent *agent = NULL;
637 mutex_lock(&file->port->file_mutex);
638 mutex_lock(&file->mutex);
640 if (!file->port->ib_dev) {
641 dev_notice(file->port->dev,
642 "ib_umad_reg_agent: invalid device\n");
647 if (copy_from_user(&ureq, arg, sizeof ureq)) {
652 if (ureq.qpn != 0 && ureq.qpn != 1) {
653 dev_notice(file->port->dev,
654 "ib_umad_reg_agent: invalid QPN %d specified\n",
660 for (agent_id = 0; agent_id < IB_UMAD_MAX_AGENTS; ++agent_id)
661 if (!__get_agent(file, agent_id))
664 dev_notice(file->port->dev,
665 "ib_umad_reg_agent: Max Agents (%u) reached\n",
671 if (ureq.mgmt_class) {
672 memset(&req, 0, sizeof(req));
673 req.mgmt_class = ureq.mgmt_class;
674 req.mgmt_class_version = ureq.mgmt_class_version;
675 memcpy(req.oui, ureq.oui, sizeof req.oui);
677 if (compat_method_mask) {
678 u32 *umm = (u32 *) ureq.method_mask;
681 for (i = 0; i < BITS_TO_LONGS(IB_MGMT_MAX_METHODS); ++i)
683 umm[i * 2] | ((u64) umm[i * 2 + 1] << 32);
685 memcpy(req.method_mask, ureq.method_mask,
686 sizeof req.method_mask);
689 agent = ib_register_mad_agent(file->port->ib_dev, file->port->port_num,
690 ureq.qpn ? IB_QPT_GSI : IB_QPT_SMI,
691 ureq.mgmt_class ? &req : NULL,
693 send_handler, recv_handler, file, 0);
695 ret = PTR_ERR(agent);
700 if (put_user(agent_id,
701 (u32 __user *) (arg + offsetof(struct ib_user_mad_reg_req, id)))) {
706 if (!file->already_used) {
707 file->already_used = 1;
708 if (!file->use_pkey_index) {
709 dev_warn(file->port->dev,
710 "process %s did not enable P_Key index support.\n",
712 dev_warn(file->port->dev,
713 " Documentation/infiniband/user_mad.txt has info on the new ABI.\n");
717 file->agent[agent_id] = agent;
721 mutex_unlock(&file->mutex);
724 ib_unregister_mad_agent(agent);
726 mutex_unlock(&file->port->file_mutex);
731 static int ib_umad_reg_agent2(struct ib_umad_file *file, void __user *arg)
733 struct ib_user_mad_reg_req2 ureq;
734 struct ib_mad_reg_req req;
735 struct ib_mad_agent *agent = NULL;
739 mutex_lock(&file->port->file_mutex);
740 mutex_lock(&file->mutex);
742 if (!file->port->ib_dev) {
743 dev_notice(file->port->dev,
744 "ib_umad_reg_agent2: invalid device\n");
749 if (copy_from_user(&ureq, arg, sizeof(ureq))) {
754 if (ureq.qpn != 0 && ureq.qpn != 1) {
755 dev_notice(file->port->dev,
756 "ib_umad_reg_agent2: invalid QPN %d specified\n",
762 if (ureq.flags & ~IB_USER_MAD_REG_FLAGS_CAP) {
763 dev_notice(file->port->dev,
764 "ib_umad_reg_agent2 failed: invalid registration flags specified 0x%x; supported 0x%x\n",
765 ureq.flags, IB_USER_MAD_REG_FLAGS_CAP);
768 if (put_user((u32)IB_USER_MAD_REG_FLAGS_CAP,
769 (u32 __user *) (arg + offsetof(struct
770 ib_user_mad_reg_req2, flags))))
776 for (agent_id = 0; agent_id < IB_UMAD_MAX_AGENTS; ++agent_id)
777 if (!__get_agent(file, agent_id))
780 dev_notice(file->port->dev,
781 "ib_umad_reg_agent2: Max Agents (%u) reached\n",
787 if (ureq.mgmt_class) {
788 memset(&req, 0, sizeof(req));
789 req.mgmt_class = ureq.mgmt_class;
790 req.mgmt_class_version = ureq.mgmt_class_version;
791 if (ureq.oui & 0xff000000) {
792 dev_notice(file->port->dev,
793 "ib_umad_reg_agent2 failed: oui invalid 0x%08x\n",
798 req.oui[2] = ureq.oui & 0x0000ff;
799 req.oui[1] = (ureq.oui & 0x00ff00) >> 8;
800 req.oui[0] = (ureq.oui & 0xff0000) >> 16;
801 memcpy(req.method_mask, ureq.method_mask,
802 sizeof(req.method_mask));
805 agent = ib_register_mad_agent(file->port->ib_dev, file->port->port_num,
806 ureq.qpn ? IB_QPT_GSI : IB_QPT_SMI,
807 ureq.mgmt_class ? &req : NULL,
809 send_handler, recv_handler, file,
812 ret = PTR_ERR(agent);
817 if (put_user(agent_id,
819 offsetof(struct ib_user_mad_reg_req2, id)))) {
824 if (!file->already_used) {
825 file->already_used = 1;
826 file->use_pkey_index = 1;
829 file->agent[agent_id] = agent;
833 mutex_unlock(&file->mutex);
836 ib_unregister_mad_agent(agent);
838 mutex_unlock(&file->port->file_mutex);
844 static int ib_umad_unreg_agent(struct ib_umad_file *file, u32 __user *arg)
846 struct ib_mad_agent *agent = NULL;
850 if (get_user(id, arg))
852 if (id >= IB_UMAD_MAX_AGENTS)
855 mutex_lock(&file->port->file_mutex);
856 mutex_lock(&file->mutex);
858 id = array_index_nospec(id, IB_UMAD_MAX_AGENTS);
859 if (!__get_agent(file, id)) {
864 agent = file->agent[id];
865 file->agent[id] = NULL;
868 mutex_unlock(&file->mutex);
871 ib_unregister_mad_agent(agent);
873 mutex_unlock(&file->port->file_mutex);
878 static long ib_umad_enable_pkey(struct ib_umad_file *file)
882 mutex_lock(&file->mutex);
883 if (file->already_used)
886 file->use_pkey_index = 1;
887 mutex_unlock(&file->mutex);
892 static long ib_umad_ioctl(struct file *filp, unsigned int cmd,
896 case IB_USER_MAD_REGISTER_AGENT:
897 return ib_umad_reg_agent(filp->private_data, (void __user *) arg, 0);
898 case IB_USER_MAD_UNREGISTER_AGENT:
899 return ib_umad_unreg_agent(filp->private_data, (__u32 __user *) arg);
900 case IB_USER_MAD_ENABLE_PKEY:
901 return ib_umad_enable_pkey(filp->private_data);
902 case IB_USER_MAD_REGISTER_AGENT2:
903 return ib_umad_reg_agent2(filp->private_data, (void __user *) arg);
910 static long ib_umad_compat_ioctl(struct file *filp, unsigned int cmd,
914 case IB_USER_MAD_REGISTER_AGENT:
915 return ib_umad_reg_agent(filp->private_data, compat_ptr(arg), 1);
916 case IB_USER_MAD_UNREGISTER_AGENT:
917 return ib_umad_unreg_agent(filp->private_data, compat_ptr(arg));
918 case IB_USER_MAD_ENABLE_PKEY:
919 return ib_umad_enable_pkey(filp->private_data);
920 case IB_USER_MAD_REGISTER_AGENT2:
921 return ib_umad_reg_agent2(filp->private_data, compat_ptr(arg));
929 * ib_umad_open() does not need the BKL:
931 * - the ib_umad_port structures are properly reference counted, and
932 * everything else is purely local to the file being created, so
933 * races against other open calls are not a problem;
934 * - the ioctl method does not affect any global state outside of the
935 * file structure being operated on;
937 static int ib_umad_open(struct inode *inode, struct file *filp)
939 struct ib_umad_port *port;
940 struct ib_umad_file *file;
943 port = container_of(inode->i_cdev, struct ib_umad_port, cdev);
945 mutex_lock(&port->file_mutex);
951 file = kzalloc(sizeof *file, GFP_KERNEL);
955 mutex_init(&file->mutex);
956 spin_lock_init(&file->send_lock);
957 INIT_LIST_HEAD(&file->recv_list);
958 INIT_LIST_HEAD(&file->send_list);
959 init_waitqueue_head(&file->recv_wait);
962 filp->private_data = file;
964 list_add_tail(&file->port_list, &port->file_list);
966 ret = nonseekable_open(inode, filp);
968 list_del(&file->port_list);
973 kobject_get(&port->umad_dev->kobj);
976 mutex_unlock(&port->file_mutex);
980 static int ib_umad_close(struct inode *inode, struct file *filp)
982 struct ib_umad_file *file = filp->private_data;
983 struct ib_umad_device *dev = file->port->umad_dev;
984 struct ib_umad_packet *packet, *tmp;
988 mutex_lock(&file->port->file_mutex);
989 mutex_lock(&file->mutex);
991 already_dead = file->agents_dead;
992 file->agents_dead = 1;
994 list_for_each_entry_safe(packet, tmp, &file->recv_list, list) {
996 ib_free_recv_mad(packet->recv_wc);
1000 list_del(&file->port_list);
1002 mutex_unlock(&file->mutex);
1005 for (i = 0; i < IB_UMAD_MAX_AGENTS; ++i)
1007 ib_unregister_mad_agent(file->agent[i]);
1009 mutex_unlock(&file->port->file_mutex);
1012 kobject_put(&dev->kobj);
1017 static const struct file_operations umad_fops = {
1018 .owner = THIS_MODULE,
1019 .read = ib_umad_read,
1020 .write = ib_umad_write,
1021 .poll = ib_umad_poll,
1022 .unlocked_ioctl = ib_umad_ioctl,
1023 #ifdef CONFIG_COMPAT
1024 .compat_ioctl = ib_umad_compat_ioctl,
1026 .open = ib_umad_open,
1027 .release = ib_umad_close,
1028 .llseek = no_llseek,
1031 static int ib_umad_sm_open(struct inode *inode, struct file *filp)
1033 struct ib_umad_port *port;
1034 struct ib_port_modify props = {
1035 .set_port_cap_mask = IB_PORT_SM
1039 port = container_of(inode->i_cdev, struct ib_umad_port, sm_cdev);
1041 if (filp->f_flags & O_NONBLOCK) {
1042 if (down_trylock(&port->sm_sem)) {
1047 if (down_interruptible(&port->sm_sem)) {
1053 ret = ib_modify_port(port->ib_dev, port->port_num, 0, &props);
1057 filp->private_data = port;
1059 ret = nonseekable_open(inode, filp);
1061 goto err_clr_sm_cap;
1063 kobject_get(&port->umad_dev->kobj);
1068 swap(props.set_port_cap_mask, props.clr_port_cap_mask);
1069 ib_modify_port(port->ib_dev, port->port_num, 0, &props);
1078 static int ib_umad_sm_close(struct inode *inode, struct file *filp)
1080 struct ib_umad_port *port = filp->private_data;
1081 struct ib_port_modify props = {
1082 .clr_port_cap_mask = IB_PORT_SM
1086 mutex_lock(&port->file_mutex);
1088 ret = ib_modify_port(port->ib_dev, port->port_num, 0, &props);
1089 mutex_unlock(&port->file_mutex);
1093 kobject_put(&port->umad_dev->kobj);
1098 static const struct file_operations umad_sm_fops = {
1099 .owner = THIS_MODULE,
1100 .open = ib_umad_sm_open,
1101 .release = ib_umad_sm_close,
1102 .llseek = no_llseek,
1105 static struct ib_client umad_client = {
1107 .add = ib_umad_add_one,
1108 .remove = ib_umad_remove_one
1111 static ssize_t show_ibdev(struct device *dev, struct device_attribute *attr,
1114 struct ib_umad_port *port = dev_get_drvdata(dev);
1119 return sprintf(buf, "%s\n", port->ib_dev->name);
1121 static DEVICE_ATTR(ibdev, S_IRUGO, show_ibdev, NULL);
1123 static ssize_t show_port(struct device *dev, struct device_attribute *attr,
1126 struct ib_umad_port *port = dev_get_drvdata(dev);
1131 return sprintf(buf, "%d\n", port->port_num);
1133 static DEVICE_ATTR(port, S_IRUGO, show_port, NULL);
1135 static CLASS_ATTR_STRING(abi_version, S_IRUGO,
1136 __stringify(IB_USER_MAD_ABI_VERSION));
1138 static dev_t overflow_maj;
1139 static DECLARE_BITMAP(overflow_map, IB_UMAD_MAX_PORTS);
1140 static int find_overflow_devnum(struct ib_device *device)
1144 if (!overflow_maj) {
1145 ret = alloc_chrdev_region(&overflow_maj, 0, IB_UMAD_MAX_PORTS * 2,
1148 dev_err(&device->dev,
1149 "couldn't register dynamic device number\n");
1154 ret = find_first_zero_bit(overflow_map, IB_UMAD_MAX_PORTS);
1155 if (ret >= IB_UMAD_MAX_PORTS)
1161 static int ib_umad_init_port(struct ib_device *device, int port_num,
1162 struct ib_umad_device *umad_dev,
1163 struct ib_umad_port *port)
1168 spin_lock(&port_lock);
1169 devnum = find_first_zero_bit(dev_map, IB_UMAD_MAX_PORTS);
1170 if (devnum >= IB_UMAD_MAX_PORTS) {
1171 spin_unlock(&port_lock);
1172 devnum = find_overflow_devnum(device);
1176 spin_lock(&port_lock);
1177 port->dev_num = devnum + IB_UMAD_MAX_PORTS;
1178 base = devnum + overflow_maj;
1179 set_bit(devnum, overflow_map);
1181 port->dev_num = devnum;
1182 base = devnum + base_dev;
1183 set_bit(devnum, dev_map);
1185 spin_unlock(&port_lock);
1187 port->ib_dev = device;
1188 port->port_num = port_num;
1189 sema_init(&port->sm_sem, 1);
1190 mutex_init(&port->file_mutex);
1191 INIT_LIST_HEAD(&port->file_list);
1193 cdev_init(&port->cdev, &umad_fops);
1194 port->cdev.owner = THIS_MODULE;
1195 port->cdev.kobj.parent = &umad_dev->kobj;
1196 kobject_set_name(&port->cdev.kobj, "umad%d", port->dev_num);
1197 if (cdev_add(&port->cdev, base, 1))
1200 port->dev = device_create(umad_class, device->dma_device,
1201 port->cdev.dev, port,
1202 "umad%d", port->dev_num);
1203 if (IS_ERR(port->dev))
1206 if (device_create_file(port->dev, &dev_attr_ibdev))
1208 if (device_create_file(port->dev, &dev_attr_port))
1211 base += IB_UMAD_MAX_PORTS;
1212 cdev_init(&port->sm_cdev, &umad_sm_fops);
1213 port->sm_cdev.owner = THIS_MODULE;
1214 port->sm_cdev.kobj.parent = &umad_dev->kobj;
1215 kobject_set_name(&port->sm_cdev.kobj, "issm%d", port->dev_num);
1216 if (cdev_add(&port->sm_cdev, base, 1))
1219 port->sm_dev = device_create(umad_class, device->dma_device,
1220 port->sm_cdev.dev, port,
1221 "issm%d", port->dev_num);
1222 if (IS_ERR(port->sm_dev))
1225 if (device_create_file(port->sm_dev, &dev_attr_ibdev))
1227 if (device_create_file(port->sm_dev, &dev_attr_port))
1233 device_destroy(umad_class, port->sm_cdev.dev);
1236 cdev_del(&port->sm_cdev);
1239 device_destroy(umad_class, port->cdev.dev);
1242 cdev_del(&port->cdev);
1243 if (port->dev_num < IB_UMAD_MAX_PORTS)
1244 clear_bit(devnum, dev_map);
1246 clear_bit(devnum, overflow_map);
1251 static void ib_umad_kill_port(struct ib_umad_port *port)
1253 struct ib_umad_file *file;
1256 dev_set_drvdata(port->dev, NULL);
1257 dev_set_drvdata(port->sm_dev, NULL);
1259 device_destroy(umad_class, port->cdev.dev);
1260 device_destroy(umad_class, port->sm_cdev.dev);
1262 cdev_del(&port->cdev);
1263 cdev_del(&port->sm_cdev);
1265 mutex_lock(&port->file_mutex);
1267 port->ib_dev = NULL;
1269 list_for_each_entry(file, &port->file_list, port_list) {
1270 mutex_lock(&file->mutex);
1271 file->agents_dead = 1;
1272 mutex_unlock(&file->mutex);
1274 for (id = 0; id < IB_UMAD_MAX_AGENTS; ++id)
1275 if (file->agent[id])
1276 ib_unregister_mad_agent(file->agent[id]);
1279 mutex_unlock(&port->file_mutex);
1281 if (port->dev_num < IB_UMAD_MAX_PORTS)
1282 clear_bit(port->dev_num, dev_map);
1284 clear_bit(port->dev_num - IB_UMAD_MAX_PORTS, overflow_map);
1287 static void ib_umad_add_one(struct ib_device *device)
1289 struct ib_umad_device *umad_dev;
1293 s = rdma_start_port(device);
1294 e = rdma_end_port(device);
1296 umad_dev = kzalloc(sizeof *umad_dev +
1297 (e - s + 1) * sizeof (struct ib_umad_port),
1302 kobject_init(&umad_dev->kobj, &ib_umad_dev_ktype);
1304 for (i = s; i <= e; ++i) {
1305 if (!rdma_cap_ib_mad(device, i))
1308 umad_dev->port[i - s].umad_dev = umad_dev;
1310 if (ib_umad_init_port(device, i, umad_dev,
1311 &umad_dev->port[i - s]))
1320 ib_set_client_data(device, &umad_client, umad_dev);
1326 if (!rdma_cap_ib_mad(device, i))
1329 ib_umad_kill_port(&umad_dev->port[i - s]);
1332 kobject_put(&umad_dev->kobj);
1335 static void ib_umad_remove_one(struct ib_device *device, void *client_data)
1337 struct ib_umad_device *umad_dev = client_data;
1343 for (i = 0; i <= rdma_end_port(device) - rdma_start_port(device); ++i) {
1344 if (rdma_cap_ib_mad(device, i + rdma_start_port(device)))
1345 ib_umad_kill_port(&umad_dev->port[i]);
1348 kobject_put(&umad_dev->kobj);
1351 static char *umad_devnode(struct device *dev, umode_t *mode)
1353 return kasprintf(GFP_KERNEL, "infiniband/%s", dev_name(dev));
1356 static int __init ib_umad_init(void)
1360 ret = register_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2,
1363 pr_err("couldn't register device number\n");
1367 umad_class = class_create(THIS_MODULE, "infiniband_mad");
1368 if (IS_ERR(umad_class)) {
1369 ret = PTR_ERR(umad_class);
1370 pr_err("couldn't create class infiniband_mad\n");
1374 umad_class->devnode = umad_devnode;
1376 ret = class_create_file(umad_class, &class_attr_abi_version.attr);
1378 pr_err("couldn't create abi_version attribute\n");
1382 ret = ib_register_client(&umad_client);
1384 pr_err("couldn't register ib_umad client\n");
1391 class_destroy(umad_class);
1394 unregister_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2);
1400 static void __exit ib_umad_cleanup(void)
1402 ib_unregister_client(&umad_client);
1403 class_destroy(umad_class);
1404 unregister_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2);
1406 unregister_chrdev_region(overflow_maj, IB_UMAD_MAX_PORTS * 2);
1409 module_init(ib_umad_init);
1410 module_exit(ib_umad_cleanup);