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 <linux/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 rdma_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);
234 * On OPA devices it is okay to lose the upper 16 bits of LID as this
235 * information is obtained elsewhere. Mask off the upper 16 bits.
237 if (agent->device->port_immutable[agent->port_num].core_cap_flags &
238 RDMA_CORE_PORT_INTEL_OPA)
239 packet->mad.hdr.lid = ib_lid_be16(0xFFFF &
240 mad_recv_wc->wc->slid);
242 packet->mad.hdr.lid = ib_lid_be16(mad_recv_wc->wc->slid);
243 packet->mad.hdr.sl = mad_recv_wc->wc->sl;
244 packet->mad.hdr.path_bits = mad_recv_wc->wc->dlid_path_bits;
245 packet->mad.hdr.pkey_index = mad_recv_wc->wc->pkey_index;
246 packet->mad.hdr.grh_present = !!(mad_recv_wc->wc->wc_flags & IB_WC_GRH);
247 if (packet->mad.hdr.grh_present) {
248 struct rdma_ah_attr ah_attr;
249 const struct ib_global_route *grh;
251 ib_init_ah_from_wc(agent->device, agent->port_num,
252 mad_recv_wc->wc, mad_recv_wc->recv_buf.grh,
255 grh = rdma_ah_read_grh(&ah_attr);
256 packet->mad.hdr.gid_index = grh->sgid_index;
257 packet->mad.hdr.hop_limit = grh->hop_limit;
258 packet->mad.hdr.traffic_class = grh->traffic_class;
259 memcpy(packet->mad.hdr.gid, &grh->dgid, 16);
260 packet->mad.hdr.flow_label = cpu_to_be32(grh->flow_label);
263 if (queue_packet(file, agent, packet))
270 ib_free_recv_mad(mad_recv_wc);
273 static ssize_t copy_recv_mad(struct ib_umad_file *file, char __user *buf,
274 struct ib_umad_packet *packet, size_t count)
276 struct ib_mad_recv_buf *recv_buf;
277 int left, seg_payload, offset, max_seg_payload;
280 recv_buf = &packet->recv_wc->recv_buf;
281 seg_size = packet->recv_wc->mad_seg_size;
283 /* We need enough room to copy the first (or only) MAD segment. */
284 if ((packet->length <= seg_size &&
285 count < hdr_size(file) + packet->length) ||
286 (packet->length > seg_size &&
287 count < hdr_size(file) + seg_size))
290 if (copy_to_user(buf, &packet->mad, hdr_size(file)))
293 buf += hdr_size(file);
294 seg_payload = min_t(int, packet->length, seg_size);
295 if (copy_to_user(buf, recv_buf->mad, seg_payload))
298 if (seg_payload < packet->length) {
300 * Multipacket RMPP MAD message. Copy remainder of message.
301 * Note that last segment may have a shorter payload.
303 if (count < hdr_size(file) + packet->length) {
305 * The buffer is too small, return the first RMPP segment,
306 * which includes the RMPP message length.
310 offset = ib_get_mad_data_offset(recv_buf->mad->mad_hdr.mgmt_class);
311 max_seg_payload = seg_size - offset;
313 for (left = packet->length - seg_payload, buf += seg_payload;
314 left; left -= seg_payload, buf += seg_payload) {
315 recv_buf = container_of(recv_buf->list.next,
316 struct ib_mad_recv_buf, list);
317 seg_payload = min(left, max_seg_payload);
318 if (copy_to_user(buf, ((void *) recv_buf->mad) + offset,
323 return hdr_size(file) + packet->length;
326 static ssize_t copy_send_mad(struct ib_umad_file *file, char __user *buf,
327 struct ib_umad_packet *packet, size_t count)
329 ssize_t size = hdr_size(file) + packet->length;
334 if (copy_to_user(buf, &packet->mad, hdr_size(file)))
337 buf += hdr_size(file);
339 if (copy_to_user(buf, packet->mad.data, packet->length))
345 static ssize_t ib_umad_read(struct file *filp, char __user *buf,
346 size_t count, loff_t *pos)
348 struct ib_umad_file *file = filp->private_data;
349 struct ib_umad_packet *packet;
352 if (count < hdr_size(file))
355 mutex_lock(&file->mutex);
357 if (file->agents_dead) {
358 mutex_unlock(&file->mutex);
362 while (list_empty(&file->recv_list)) {
363 mutex_unlock(&file->mutex);
365 if (filp->f_flags & O_NONBLOCK)
368 if (wait_event_interruptible(file->recv_wait,
369 !list_empty(&file->recv_list)))
372 mutex_lock(&file->mutex);
375 packet = list_entry(file->recv_list.next, struct ib_umad_packet, list);
376 list_del(&packet->list);
378 mutex_unlock(&file->mutex);
381 ret = copy_recv_mad(file, buf, packet, count);
383 ret = copy_send_mad(file, buf, packet, count);
387 mutex_lock(&file->mutex);
388 list_add(&packet->list, &file->recv_list);
389 mutex_unlock(&file->mutex);
392 ib_free_recv_mad(packet->recv_wc);
398 static int copy_rmpp_mad(struct ib_mad_send_buf *msg, const char __user *buf)
402 /* Copy class specific header */
403 if ((msg->hdr_len > IB_MGMT_RMPP_HDR) &&
404 copy_from_user(msg->mad + IB_MGMT_RMPP_HDR, buf + IB_MGMT_RMPP_HDR,
405 msg->hdr_len - IB_MGMT_RMPP_HDR))
408 /* All headers are in place. Copy data segments. */
409 for (seg = 1, left = msg->data_len, buf += msg->hdr_len; left > 0;
410 seg++, left -= msg->seg_size, buf += msg->seg_size) {
411 if (copy_from_user(ib_get_rmpp_segment(msg, seg), buf,
412 min(left, msg->seg_size)))
418 static int same_destination(struct ib_user_mad_hdr *hdr1,
419 struct ib_user_mad_hdr *hdr2)
421 if (!hdr1->grh_present && !hdr2->grh_present)
422 return (hdr1->lid == hdr2->lid);
424 if (hdr1->grh_present && hdr2->grh_present)
425 return !memcmp(hdr1->gid, hdr2->gid, 16);
430 static int is_duplicate(struct ib_umad_file *file,
431 struct ib_umad_packet *packet)
433 struct ib_umad_packet *sent_packet;
434 struct ib_mad_hdr *sent_hdr, *hdr;
436 hdr = (struct ib_mad_hdr *) packet->mad.data;
437 list_for_each_entry(sent_packet, &file->send_list, list) {
438 sent_hdr = (struct ib_mad_hdr *) sent_packet->mad.data;
440 if ((hdr->tid != sent_hdr->tid) ||
441 (hdr->mgmt_class != sent_hdr->mgmt_class))
445 * No need to be overly clever here. If two new operations have
446 * the same TID, reject the second as a duplicate. This is more
447 * restrictive than required by the spec.
449 if (!ib_response_mad(hdr)) {
450 if (!ib_response_mad(sent_hdr))
453 } else if (!ib_response_mad(sent_hdr))
456 if (same_destination(&packet->mad.hdr, &sent_packet->mad.hdr))
463 static ssize_t ib_umad_write(struct file *filp, const char __user *buf,
464 size_t count, loff_t *pos)
466 struct ib_umad_file *file = filp->private_data;
467 struct ib_umad_packet *packet;
468 struct ib_mad_agent *agent;
469 struct rdma_ah_attr ah_attr;
471 struct ib_rmpp_mad *rmpp_mad;
473 int ret, data_len, hdr_len, copy_offset, rmpp_active;
476 if (count < hdr_size(file) + IB_MGMT_RMPP_HDR)
479 packet = kzalloc(sizeof *packet + IB_MGMT_RMPP_HDR, GFP_KERNEL);
483 if (copy_from_user(&packet->mad, buf, hdr_size(file))) {
488 if (packet->mad.hdr.id >= IB_UMAD_MAX_AGENTS) {
493 buf += hdr_size(file);
495 if (copy_from_user(packet->mad.data, buf, IB_MGMT_RMPP_HDR)) {
500 mutex_lock(&file->mutex);
502 agent = __get_agent(file, packet->mad.hdr.id);
508 memset(&ah_attr, 0, sizeof ah_attr);
509 ah_attr.type = rdma_ah_find_type(agent->device,
510 file->port->port_num);
511 rdma_ah_set_dlid(&ah_attr, be16_to_cpu(packet->mad.hdr.lid));
512 rdma_ah_set_sl(&ah_attr, packet->mad.hdr.sl);
513 rdma_ah_set_path_bits(&ah_attr, packet->mad.hdr.path_bits);
514 rdma_ah_set_port_num(&ah_attr, file->port->port_num);
515 if (packet->mad.hdr.grh_present) {
516 rdma_ah_set_grh(&ah_attr, NULL,
517 be32_to_cpu(packet->mad.hdr.flow_label),
518 packet->mad.hdr.gid_index,
519 packet->mad.hdr.hop_limit,
520 packet->mad.hdr.traffic_class);
521 rdma_ah_set_dgid_raw(&ah_attr, packet->mad.hdr.gid);
524 ah = rdma_create_ah(agent->qp->pd, &ah_attr);
530 rmpp_mad = (struct ib_rmpp_mad *) packet->mad.data;
531 hdr_len = ib_get_mad_data_offset(rmpp_mad->mad_hdr.mgmt_class);
533 if (ib_is_mad_class_rmpp(rmpp_mad->mad_hdr.mgmt_class)
534 && ib_mad_kernel_rmpp_agent(agent)) {
535 copy_offset = IB_MGMT_RMPP_HDR;
536 rmpp_active = ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) &
537 IB_MGMT_RMPP_FLAG_ACTIVE;
539 copy_offset = IB_MGMT_MAD_HDR;
543 base_version = ((struct ib_mad_hdr *)&packet->mad.data)->base_version;
544 data_len = count - hdr_size(file) - hdr_len;
545 packet->msg = ib_create_send_mad(agent,
546 be32_to_cpu(packet->mad.hdr.qpn),
547 packet->mad.hdr.pkey_index, rmpp_active,
548 hdr_len, data_len, GFP_KERNEL,
550 if (IS_ERR(packet->msg)) {
551 ret = PTR_ERR(packet->msg);
555 packet->msg->ah = ah;
556 packet->msg->timeout_ms = packet->mad.hdr.timeout_ms;
557 packet->msg->retries = packet->mad.hdr.retries;
558 packet->msg->context[0] = packet;
560 /* Copy MAD header. Any RMPP header is already in place. */
561 memcpy(packet->msg->mad, packet->mad.data, IB_MGMT_MAD_HDR);
564 if (copy_from_user(packet->msg->mad + copy_offset,
566 hdr_len + data_len - copy_offset)) {
571 ret = copy_rmpp_mad(packet->msg, buf);
577 * Set the high-order part of the transaction ID to make MADs from
578 * different agents unique, and allow routing responses back to the
579 * original requestor.
581 if (!ib_response_mad(packet->msg->mad)) {
582 tid = &((struct ib_mad_hdr *) packet->msg->mad)->tid;
583 *tid = cpu_to_be64(((u64) agent->hi_tid) << 32 |
584 (be64_to_cpup(tid) & 0xffffffff));
585 rmpp_mad->mad_hdr.tid = *tid;
588 if (!ib_mad_kernel_rmpp_agent(agent)
589 && ib_is_mad_class_rmpp(rmpp_mad->mad_hdr.mgmt_class)
590 && (ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) & IB_MGMT_RMPP_FLAG_ACTIVE)) {
591 spin_lock_irq(&file->send_lock);
592 list_add_tail(&packet->list, &file->send_list);
593 spin_unlock_irq(&file->send_lock);
595 spin_lock_irq(&file->send_lock);
596 ret = is_duplicate(file, packet);
598 list_add_tail(&packet->list, &file->send_list);
599 spin_unlock_irq(&file->send_lock);
606 ret = ib_post_send_mad(packet->msg, NULL);
610 mutex_unlock(&file->mutex);
614 dequeue_send(file, packet);
616 ib_free_send_mad(packet->msg);
620 mutex_unlock(&file->mutex);
626 static unsigned int ib_umad_poll(struct file *filp, struct poll_table_struct *wait)
628 struct ib_umad_file *file = filp->private_data;
630 /* we will always be able to post a MAD send */
631 unsigned int mask = POLLOUT | POLLWRNORM;
633 poll_wait(filp, &file->recv_wait, wait);
635 if (!list_empty(&file->recv_list))
636 mask |= POLLIN | POLLRDNORM;
641 static int ib_umad_reg_agent(struct ib_umad_file *file, void __user *arg,
642 int compat_method_mask)
644 struct ib_user_mad_reg_req ureq;
645 struct ib_mad_reg_req req;
646 struct ib_mad_agent *agent = NULL;
650 mutex_lock(&file->port->file_mutex);
651 mutex_lock(&file->mutex);
653 if (!file->port->ib_dev) {
654 dev_notice(file->port->dev,
655 "ib_umad_reg_agent: invalid device\n");
660 if (copy_from_user(&ureq, arg, sizeof ureq)) {
665 if (ureq.qpn != 0 && ureq.qpn != 1) {
666 dev_notice(file->port->dev,
667 "ib_umad_reg_agent: invalid QPN %d specified\n",
673 for (agent_id = 0; agent_id < IB_UMAD_MAX_AGENTS; ++agent_id)
674 if (!__get_agent(file, agent_id))
677 dev_notice(file->port->dev,
678 "ib_umad_reg_agent: Max Agents (%u) reached\n",
684 if (ureq.mgmt_class) {
685 memset(&req, 0, sizeof(req));
686 req.mgmt_class = ureq.mgmt_class;
687 req.mgmt_class_version = ureq.mgmt_class_version;
688 memcpy(req.oui, ureq.oui, sizeof req.oui);
690 if (compat_method_mask) {
691 u32 *umm = (u32 *) ureq.method_mask;
694 for (i = 0; i < BITS_TO_LONGS(IB_MGMT_MAX_METHODS); ++i)
696 umm[i * 2] | ((u64) umm[i * 2 + 1] << 32);
698 memcpy(req.method_mask, ureq.method_mask,
699 sizeof req.method_mask);
702 agent = ib_register_mad_agent(file->port->ib_dev, file->port->port_num,
703 ureq.qpn ? IB_QPT_GSI : IB_QPT_SMI,
704 ureq.mgmt_class ? &req : NULL,
706 send_handler, recv_handler, file, 0);
708 ret = PTR_ERR(agent);
713 if (put_user(agent_id,
714 (u32 __user *) (arg + offsetof(struct ib_user_mad_reg_req, id)))) {
719 if (!file->already_used) {
720 file->already_used = 1;
721 if (!file->use_pkey_index) {
722 dev_warn(file->port->dev,
723 "process %s did not enable P_Key index support.\n",
725 dev_warn(file->port->dev,
726 " Documentation/infiniband/user_mad.txt has info on the new ABI.\n");
730 file->agent[agent_id] = agent;
734 mutex_unlock(&file->mutex);
737 ib_unregister_mad_agent(agent);
739 mutex_unlock(&file->port->file_mutex);
744 static int ib_umad_reg_agent2(struct ib_umad_file *file, void __user *arg)
746 struct ib_user_mad_reg_req2 ureq;
747 struct ib_mad_reg_req req;
748 struct ib_mad_agent *agent = NULL;
752 mutex_lock(&file->port->file_mutex);
753 mutex_lock(&file->mutex);
755 if (!file->port->ib_dev) {
756 dev_notice(file->port->dev,
757 "ib_umad_reg_agent2: invalid device\n");
762 if (copy_from_user(&ureq, arg, sizeof(ureq))) {
767 if (ureq.qpn != 0 && ureq.qpn != 1) {
768 dev_notice(file->port->dev,
769 "ib_umad_reg_agent2: invalid QPN %d specified\n",
775 if (ureq.flags & ~IB_USER_MAD_REG_FLAGS_CAP) {
776 dev_notice(file->port->dev,
777 "ib_umad_reg_agent2 failed: invalid registration flags specified 0x%x; supported 0x%x\n",
778 ureq.flags, IB_USER_MAD_REG_FLAGS_CAP);
781 if (put_user((u32)IB_USER_MAD_REG_FLAGS_CAP,
782 (u32 __user *) (arg + offsetof(struct
783 ib_user_mad_reg_req2, flags))))
789 for (agent_id = 0; agent_id < IB_UMAD_MAX_AGENTS; ++agent_id)
790 if (!__get_agent(file, agent_id))
793 dev_notice(file->port->dev,
794 "ib_umad_reg_agent2: Max Agents (%u) reached\n",
800 if (ureq.mgmt_class) {
801 memset(&req, 0, sizeof(req));
802 req.mgmt_class = ureq.mgmt_class;
803 req.mgmt_class_version = ureq.mgmt_class_version;
804 if (ureq.oui & 0xff000000) {
805 dev_notice(file->port->dev,
806 "ib_umad_reg_agent2 failed: oui invalid 0x%08x\n",
811 req.oui[2] = ureq.oui & 0x0000ff;
812 req.oui[1] = (ureq.oui & 0x00ff00) >> 8;
813 req.oui[0] = (ureq.oui & 0xff0000) >> 16;
814 memcpy(req.method_mask, ureq.method_mask,
815 sizeof(req.method_mask));
818 agent = ib_register_mad_agent(file->port->ib_dev, file->port->port_num,
819 ureq.qpn ? IB_QPT_GSI : IB_QPT_SMI,
820 ureq.mgmt_class ? &req : NULL,
822 send_handler, recv_handler, file,
825 ret = PTR_ERR(agent);
830 if (put_user(agent_id,
832 offsetof(struct ib_user_mad_reg_req2, id)))) {
837 if (!file->already_used) {
838 file->already_used = 1;
839 file->use_pkey_index = 1;
842 file->agent[agent_id] = agent;
846 mutex_unlock(&file->mutex);
849 ib_unregister_mad_agent(agent);
851 mutex_unlock(&file->port->file_mutex);
857 static int ib_umad_unreg_agent(struct ib_umad_file *file, u32 __user *arg)
859 struct ib_mad_agent *agent = NULL;
863 if (get_user(id, arg))
865 if (id >= IB_UMAD_MAX_AGENTS)
868 mutex_lock(&file->port->file_mutex);
869 mutex_lock(&file->mutex);
871 id = array_index_nospec(id, IB_UMAD_MAX_AGENTS);
872 if (!__get_agent(file, id)) {
877 agent = file->agent[id];
878 file->agent[id] = NULL;
881 mutex_unlock(&file->mutex);
884 ib_unregister_mad_agent(agent);
886 mutex_unlock(&file->port->file_mutex);
891 static long ib_umad_enable_pkey(struct ib_umad_file *file)
895 mutex_lock(&file->mutex);
896 if (file->already_used)
899 file->use_pkey_index = 1;
900 mutex_unlock(&file->mutex);
905 static long ib_umad_ioctl(struct file *filp, unsigned int cmd,
909 case IB_USER_MAD_REGISTER_AGENT:
910 return ib_umad_reg_agent(filp->private_data, (void __user *) arg, 0);
911 case IB_USER_MAD_UNREGISTER_AGENT:
912 return ib_umad_unreg_agent(filp->private_data, (__u32 __user *) arg);
913 case IB_USER_MAD_ENABLE_PKEY:
914 return ib_umad_enable_pkey(filp->private_data);
915 case IB_USER_MAD_REGISTER_AGENT2:
916 return ib_umad_reg_agent2(filp->private_data, (void __user *) arg);
923 static long ib_umad_compat_ioctl(struct file *filp, unsigned int cmd,
927 case IB_USER_MAD_REGISTER_AGENT:
928 return ib_umad_reg_agent(filp->private_data, compat_ptr(arg), 1);
929 case IB_USER_MAD_UNREGISTER_AGENT:
930 return ib_umad_unreg_agent(filp->private_data, compat_ptr(arg));
931 case IB_USER_MAD_ENABLE_PKEY:
932 return ib_umad_enable_pkey(filp->private_data);
933 case IB_USER_MAD_REGISTER_AGENT2:
934 return ib_umad_reg_agent2(filp->private_data, compat_ptr(arg));
942 * ib_umad_open() does not need the BKL:
944 * - the ib_umad_port structures are properly reference counted, and
945 * everything else is purely local to the file being created, so
946 * races against other open calls are not a problem;
947 * - the ioctl method does not affect any global state outside of the
948 * file structure being operated on;
950 static int ib_umad_open(struct inode *inode, struct file *filp)
952 struct ib_umad_port *port;
953 struct ib_umad_file *file;
956 port = container_of(inode->i_cdev, struct ib_umad_port, cdev);
958 mutex_lock(&port->file_mutex);
964 file = kzalloc(sizeof *file, GFP_KERNEL);
968 mutex_init(&file->mutex);
969 spin_lock_init(&file->send_lock);
970 INIT_LIST_HEAD(&file->recv_list);
971 INIT_LIST_HEAD(&file->send_list);
972 init_waitqueue_head(&file->recv_wait);
975 filp->private_data = file;
977 list_add_tail(&file->port_list, &port->file_list);
979 ret = nonseekable_open(inode, filp);
981 list_del(&file->port_list);
986 kobject_get(&port->umad_dev->kobj);
989 mutex_unlock(&port->file_mutex);
993 static int ib_umad_close(struct inode *inode, struct file *filp)
995 struct ib_umad_file *file = filp->private_data;
996 struct ib_umad_device *dev = file->port->umad_dev;
997 struct ib_umad_packet *packet, *tmp;
1001 mutex_lock(&file->port->file_mutex);
1002 mutex_lock(&file->mutex);
1004 already_dead = file->agents_dead;
1005 file->agents_dead = 1;
1007 list_for_each_entry_safe(packet, tmp, &file->recv_list, list) {
1008 if (packet->recv_wc)
1009 ib_free_recv_mad(packet->recv_wc);
1013 list_del(&file->port_list);
1015 mutex_unlock(&file->mutex);
1018 for (i = 0; i < IB_UMAD_MAX_AGENTS; ++i)
1020 ib_unregister_mad_agent(file->agent[i]);
1022 mutex_unlock(&file->port->file_mutex);
1025 kobject_put(&dev->kobj);
1030 static const struct file_operations umad_fops = {
1031 .owner = THIS_MODULE,
1032 .read = ib_umad_read,
1033 .write = ib_umad_write,
1034 .poll = ib_umad_poll,
1035 .unlocked_ioctl = ib_umad_ioctl,
1036 #ifdef CONFIG_COMPAT
1037 .compat_ioctl = ib_umad_compat_ioctl,
1039 .open = ib_umad_open,
1040 .release = ib_umad_close,
1041 .llseek = no_llseek,
1044 static int ib_umad_sm_open(struct inode *inode, struct file *filp)
1046 struct ib_umad_port *port;
1047 struct ib_port_modify props = {
1048 .set_port_cap_mask = IB_PORT_SM
1052 port = container_of(inode->i_cdev, struct ib_umad_port, sm_cdev);
1054 if (filp->f_flags & O_NONBLOCK) {
1055 if (down_trylock(&port->sm_sem)) {
1060 if (down_interruptible(&port->sm_sem)) {
1066 ret = ib_modify_port(port->ib_dev, port->port_num, 0, &props);
1070 filp->private_data = port;
1072 ret = nonseekable_open(inode, filp);
1074 goto err_clr_sm_cap;
1076 kobject_get(&port->umad_dev->kobj);
1081 swap(props.set_port_cap_mask, props.clr_port_cap_mask);
1082 ib_modify_port(port->ib_dev, port->port_num, 0, &props);
1091 static int ib_umad_sm_close(struct inode *inode, struct file *filp)
1093 struct ib_umad_port *port = filp->private_data;
1094 struct ib_port_modify props = {
1095 .clr_port_cap_mask = IB_PORT_SM
1099 mutex_lock(&port->file_mutex);
1101 ret = ib_modify_port(port->ib_dev, port->port_num, 0, &props);
1102 mutex_unlock(&port->file_mutex);
1106 kobject_put(&port->umad_dev->kobj);
1111 static const struct file_operations umad_sm_fops = {
1112 .owner = THIS_MODULE,
1113 .open = ib_umad_sm_open,
1114 .release = ib_umad_sm_close,
1115 .llseek = no_llseek,
1118 static struct ib_client umad_client = {
1120 .add = ib_umad_add_one,
1121 .remove = ib_umad_remove_one
1124 static ssize_t show_ibdev(struct device *dev, struct device_attribute *attr,
1127 struct ib_umad_port *port = dev_get_drvdata(dev);
1132 return sprintf(buf, "%s\n", port->ib_dev->name);
1134 static DEVICE_ATTR(ibdev, S_IRUGO, show_ibdev, NULL);
1136 static ssize_t show_port(struct device *dev, struct device_attribute *attr,
1139 struct ib_umad_port *port = dev_get_drvdata(dev);
1144 return sprintf(buf, "%d\n", port->port_num);
1146 static DEVICE_ATTR(port, S_IRUGO, show_port, NULL);
1148 static CLASS_ATTR_STRING(abi_version, S_IRUGO,
1149 __stringify(IB_USER_MAD_ABI_VERSION));
1151 static dev_t overflow_maj;
1152 static DECLARE_BITMAP(overflow_map, IB_UMAD_MAX_PORTS);
1153 static int find_overflow_devnum(struct ib_device *device)
1157 if (!overflow_maj) {
1158 ret = alloc_chrdev_region(&overflow_maj, 0, IB_UMAD_MAX_PORTS * 2,
1161 dev_err(&device->dev,
1162 "couldn't register dynamic device number\n");
1167 ret = find_first_zero_bit(overflow_map, IB_UMAD_MAX_PORTS);
1168 if (ret >= IB_UMAD_MAX_PORTS)
1174 static int ib_umad_init_port(struct ib_device *device, int port_num,
1175 struct ib_umad_device *umad_dev,
1176 struct ib_umad_port *port)
1181 spin_lock(&port_lock);
1182 devnum = find_first_zero_bit(dev_map, IB_UMAD_MAX_PORTS);
1183 if (devnum >= IB_UMAD_MAX_PORTS) {
1184 spin_unlock(&port_lock);
1185 devnum = find_overflow_devnum(device);
1189 spin_lock(&port_lock);
1190 port->dev_num = devnum + IB_UMAD_MAX_PORTS;
1191 base = devnum + overflow_maj;
1192 set_bit(devnum, overflow_map);
1194 port->dev_num = devnum;
1195 base = devnum + base_dev;
1196 set_bit(devnum, dev_map);
1198 spin_unlock(&port_lock);
1200 port->ib_dev = device;
1201 port->port_num = port_num;
1202 sema_init(&port->sm_sem, 1);
1203 mutex_init(&port->file_mutex);
1204 INIT_LIST_HEAD(&port->file_list);
1206 cdev_init(&port->cdev, &umad_fops);
1207 port->cdev.owner = THIS_MODULE;
1208 cdev_set_parent(&port->cdev, &umad_dev->kobj);
1209 kobject_set_name(&port->cdev.kobj, "umad%d", port->dev_num);
1210 if (cdev_add(&port->cdev, base, 1))
1213 port->dev = device_create(umad_class, device->dev.parent,
1214 port->cdev.dev, port,
1215 "umad%d", port->dev_num);
1216 if (IS_ERR(port->dev))
1219 if (device_create_file(port->dev, &dev_attr_ibdev))
1221 if (device_create_file(port->dev, &dev_attr_port))
1224 base += IB_UMAD_MAX_PORTS;
1225 cdev_init(&port->sm_cdev, &umad_sm_fops);
1226 port->sm_cdev.owner = THIS_MODULE;
1227 cdev_set_parent(&port->sm_cdev, &umad_dev->kobj);
1228 kobject_set_name(&port->sm_cdev.kobj, "issm%d", port->dev_num);
1229 if (cdev_add(&port->sm_cdev, base, 1))
1232 port->sm_dev = device_create(umad_class, device->dev.parent,
1233 port->sm_cdev.dev, port,
1234 "issm%d", port->dev_num);
1235 if (IS_ERR(port->sm_dev))
1238 if (device_create_file(port->sm_dev, &dev_attr_ibdev))
1240 if (device_create_file(port->sm_dev, &dev_attr_port))
1246 device_destroy(umad_class, port->sm_cdev.dev);
1249 cdev_del(&port->sm_cdev);
1252 device_destroy(umad_class, port->cdev.dev);
1255 cdev_del(&port->cdev);
1256 if (port->dev_num < IB_UMAD_MAX_PORTS)
1257 clear_bit(devnum, dev_map);
1259 clear_bit(devnum, overflow_map);
1264 static void ib_umad_kill_port(struct ib_umad_port *port)
1266 struct ib_umad_file *file;
1269 dev_set_drvdata(port->dev, NULL);
1270 dev_set_drvdata(port->sm_dev, NULL);
1272 device_destroy(umad_class, port->cdev.dev);
1273 device_destroy(umad_class, port->sm_cdev.dev);
1275 cdev_del(&port->cdev);
1276 cdev_del(&port->sm_cdev);
1278 mutex_lock(&port->file_mutex);
1280 port->ib_dev = NULL;
1282 list_for_each_entry(file, &port->file_list, port_list) {
1283 mutex_lock(&file->mutex);
1284 file->agents_dead = 1;
1285 mutex_unlock(&file->mutex);
1287 for (id = 0; id < IB_UMAD_MAX_AGENTS; ++id)
1288 if (file->agent[id])
1289 ib_unregister_mad_agent(file->agent[id]);
1292 mutex_unlock(&port->file_mutex);
1294 if (port->dev_num < IB_UMAD_MAX_PORTS)
1295 clear_bit(port->dev_num, dev_map);
1297 clear_bit(port->dev_num - IB_UMAD_MAX_PORTS, overflow_map);
1300 static void ib_umad_add_one(struct ib_device *device)
1302 struct ib_umad_device *umad_dev;
1306 s = rdma_start_port(device);
1307 e = rdma_end_port(device);
1309 umad_dev = kzalloc(sizeof *umad_dev +
1310 (e - s + 1) * sizeof (struct ib_umad_port),
1315 kobject_init(&umad_dev->kobj, &ib_umad_dev_ktype);
1317 for (i = s; i <= e; ++i) {
1318 if (!rdma_cap_ib_mad(device, i))
1321 umad_dev->port[i - s].umad_dev = umad_dev;
1323 if (ib_umad_init_port(device, i, umad_dev,
1324 &umad_dev->port[i - s]))
1333 ib_set_client_data(device, &umad_client, umad_dev);
1339 if (!rdma_cap_ib_mad(device, i))
1342 ib_umad_kill_port(&umad_dev->port[i - s]);
1345 kobject_put(&umad_dev->kobj);
1348 static void ib_umad_remove_one(struct ib_device *device, void *client_data)
1350 struct ib_umad_device *umad_dev = client_data;
1356 for (i = 0; i <= rdma_end_port(device) - rdma_start_port(device); ++i) {
1357 if (rdma_cap_ib_mad(device, i + rdma_start_port(device)))
1358 ib_umad_kill_port(&umad_dev->port[i]);
1361 kobject_put(&umad_dev->kobj);
1364 static char *umad_devnode(struct device *dev, umode_t *mode)
1366 return kasprintf(GFP_KERNEL, "infiniband/%s", dev_name(dev));
1369 static int __init ib_umad_init(void)
1373 ret = register_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2,
1376 pr_err("couldn't register device number\n");
1380 umad_class = class_create(THIS_MODULE, "infiniband_mad");
1381 if (IS_ERR(umad_class)) {
1382 ret = PTR_ERR(umad_class);
1383 pr_err("couldn't create class infiniband_mad\n");
1387 umad_class->devnode = umad_devnode;
1389 ret = class_create_file(umad_class, &class_attr_abi_version.attr);
1391 pr_err("couldn't create abi_version attribute\n");
1395 ret = ib_register_client(&umad_client);
1397 pr_err("couldn't register ib_umad client\n");
1404 class_destroy(umad_class);
1407 unregister_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2);
1413 static void __exit ib_umad_cleanup(void)
1415 ib_unregister_client(&umad_client);
1416 class_destroy(umad_class);
1417 unregister_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2);
1419 unregister_chrdev_region(overflow_maj, IB_UMAD_MAX_PORTS * 2);
1422 module_init(ib_umad_init);
1423 module_exit(ib_umad_cleanup);