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>
58 #include <rdma/rdma_netlink.h>
60 #include "core_priv.h"
62 MODULE_AUTHOR("Roland Dreier");
63 MODULE_DESCRIPTION("InfiniBand userspace MAD packet access");
64 MODULE_LICENSE("Dual BSD/GPL");
67 IB_UMAD_MAX_PORTS = RDMA_MAX_PORTS,
68 IB_UMAD_MAX_AGENTS = 32,
71 IB_UMAD_MINOR_BASE = 0,
72 IB_UMAD_NUM_FIXED_MINOR = 64,
73 IB_UMAD_NUM_DYNAMIC_MINOR = IB_UMAD_MAX_PORTS - IB_UMAD_NUM_FIXED_MINOR,
74 IB_ISSM_MINOR_BASE = IB_UMAD_NUM_FIXED_MINOR,
78 * Our lifetime rules for these structs are the following:
79 * device special file is opened, we take a reference on the
80 * ib_umad_port's struct ib_umad_device. We drop these
81 * references in the corresponding close().
83 * In addition to references coming from open character devices, there
84 * is one more reference to each ib_umad_device representing the
85 * module's reference taken when allocating the ib_umad_device in
88 * When destroying an ib_umad_device, we drop the module's reference.
96 struct semaphore sm_sem;
98 struct mutex file_mutex;
99 struct list_head file_list;
101 struct ib_device *ib_dev;
102 struct ib_umad_device *umad_dev;
107 struct ib_umad_device {
109 struct ib_umad_port ports[];
112 struct ib_umad_file {
114 struct ib_umad_port *port;
115 struct list_head recv_list;
116 struct list_head send_list;
117 struct list_head port_list;
118 spinlock_t send_lock;
119 wait_queue_head_t recv_wait;
120 struct ib_mad_agent *agent[IB_UMAD_MAX_AGENTS];
126 struct ib_umad_packet {
127 struct ib_mad_send_buf *msg;
128 struct ib_mad_recv_wc *recv_wc;
129 struct list_head list;
131 struct ib_user_mad mad;
134 #define CREATE_TRACE_POINTS
135 #include <trace/events/ib_umad.h>
137 static const dev_t base_umad_dev = MKDEV(IB_UMAD_MAJOR, IB_UMAD_MINOR_BASE);
138 static const dev_t base_issm_dev = MKDEV(IB_UMAD_MAJOR, IB_UMAD_MINOR_BASE) +
139 IB_UMAD_NUM_FIXED_MINOR;
140 static dev_t dynamic_umad_dev;
141 static dev_t dynamic_issm_dev;
143 static DEFINE_IDA(umad_ida);
145 static int ib_umad_add_one(struct ib_device *device);
146 static void ib_umad_remove_one(struct ib_device *device, void *client_data);
148 static void ib_umad_dev_free(struct kref *kref)
150 struct ib_umad_device *dev =
151 container_of(kref, struct ib_umad_device, kref);
156 static void ib_umad_dev_get(struct ib_umad_device *dev)
158 kref_get(&dev->kref);
161 static void ib_umad_dev_put(struct ib_umad_device *dev)
163 kref_put(&dev->kref, ib_umad_dev_free);
166 static int hdr_size(struct ib_umad_file *file)
168 return file->use_pkey_index ? sizeof (struct ib_user_mad_hdr) :
169 sizeof (struct ib_user_mad_hdr_old);
172 /* caller must hold file->mutex */
173 static struct ib_mad_agent *__get_agent(struct ib_umad_file *file, int id)
175 return file->agents_dead ? NULL : file->agent[id];
178 static int queue_packet(struct ib_umad_file *file,
179 struct ib_mad_agent *agent,
180 struct ib_umad_packet *packet)
184 mutex_lock(&file->mutex);
186 for (packet->mad.hdr.id = 0;
187 packet->mad.hdr.id < IB_UMAD_MAX_AGENTS;
188 packet->mad.hdr.id++)
189 if (agent == __get_agent(file, packet->mad.hdr.id)) {
190 list_add_tail(&packet->list, &file->recv_list);
191 wake_up_interruptible(&file->recv_wait);
196 mutex_unlock(&file->mutex);
201 static void dequeue_send(struct ib_umad_file *file,
202 struct ib_umad_packet *packet)
204 spin_lock_irq(&file->send_lock);
205 list_del(&packet->list);
206 spin_unlock_irq(&file->send_lock);
209 static void send_handler(struct ib_mad_agent *agent,
210 struct ib_mad_send_wc *send_wc)
212 struct ib_umad_file *file = agent->context;
213 struct ib_umad_packet *packet = send_wc->send_buf->context[0];
215 dequeue_send(file, packet);
216 rdma_destroy_ah(packet->msg->ah, RDMA_DESTROY_AH_SLEEPABLE);
217 ib_free_send_mad(packet->msg);
219 if (send_wc->status == IB_WC_RESP_TIMEOUT_ERR) {
220 packet->length = IB_MGMT_MAD_HDR;
221 packet->mad.hdr.status = ETIMEDOUT;
222 if (!queue_packet(file, agent, packet))
228 static void recv_handler(struct ib_mad_agent *agent,
229 struct ib_mad_send_buf *send_buf,
230 struct ib_mad_recv_wc *mad_recv_wc)
232 struct ib_umad_file *file = agent->context;
233 struct ib_umad_packet *packet;
235 if (mad_recv_wc->wc->status != IB_WC_SUCCESS)
238 packet = kzalloc(sizeof *packet, GFP_KERNEL);
242 packet->length = mad_recv_wc->mad_len;
243 packet->recv_wc = mad_recv_wc;
245 packet->mad.hdr.status = 0;
246 packet->mad.hdr.length = hdr_size(file) + mad_recv_wc->mad_len;
247 packet->mad.hdr.qpn = cpu_to_be32(mad_recv_wc->wc->src_qp);
249 * On OPA devices it is okay to lose the upper 16 bits of LID as this
250 * information is obtained elsewhere. Mask off the upper 16 bits.
252 if (rdma_cap_opa_mad(agent->device, agent->port_num))
253 packet->mad.hdr.lid = ib_lid_be16(0xFFFF &
254 mad_recv_wc->wc->slid);
256 packet->mad.hdr.lid = ib_lid_be16(mad_recv_wc->wc->slid);
257 packet->mad.hdr.sl = mad_recv_wc->wc->sl;
258 packet->mad.hdr.path_bits = mad_recv_wc->wc->dlid_path_bits;
259 packet->mad.hdr.pkey_index = mad_recv_wc->wc->pkey_index;
260 packet->mad.hdr.grh_present = !!(mad_recv_wc->wc->wc_flags & IB_WC_GRH);
261 if (packet->mad.hdr.grh_present) {
262 struct rdma_ah_attr ah_attr;
263 const struct ib_global_route *grh;
266 ret = ib_init_ah_attr_from_wc(agent->device, agent->port_num,
268 mad_recv_wc->recv_buf.grh,
273 grh = rdma_ah_read_grh(&ah_attr);
274 packet->mad.hdr.gid_index = grh->sgid_index;
275 packet->mad.hdr.hop_limit = grh->hop_limit;
276 packet->mad.hdr.traffic_class = grh->traffic_class;
277 memcpy(packet->mad.hdr.gid, &grh->dgid, 16);
278 packet->mad.hdr.flow_label = cpu_to_be32(grh->flow_label);
279 rdma_destroy_ah_attr(&ah_attr);
282 if (queue_packet(file, agent, packet))
289 ib_free_recv_mad(mad_recv_wc);
292 static ssize_t copy_recv_mad(struct ib_umad_file *file, char __user *buf,
293 struct ib_umad_packet *packet, size_t count)
295 struct ib_mad_recv_buf *recv_buf;
296 int left, seg_payload, offset, max_seg_payload;
299 recv_buf = &packet->recv_wc->recv_buf;
300 seg_size = packet->recv_wc->mad_seg_size;
302 /* We need enough room to copy the first (or only) MAD segment. */
303 if ((packet->length <= seg_size &&
304 count < hdr_size(file) + packet->length) ||
305 (packet->length > seg_size &&
306 count < hdr_size(file) + seg_size))
309 if (copy_to_user(buf, &packet->mad, hdr_size(file)))
312 buf += hdr_size(file);
313 seg_payload = min_t(int, packet->length, seg_size);
314 if (copy_to_user(buf, recv_buf->mad, seg_payload))
317 if (seg_payload < packet->length) {
319 * Multipacket RMPP MAD message. Copy remainder of message.
320 * Note that last segment may have a shorter payload.
322 if (count < hdr_size(file) + packet->length) {
324 * The buffer is too small, return the first RMPP segment,
325 * which includes the RMPP message length.
329 offset = ib_get_mad_data_offset(recv_buf->mad->mad_hdr.mgmt_class);
330 max_seg_payload = seg_size - offset;
332 for (left = packet->length - seg_payload, buf += seg_payload;
333 left; left -= seg_payload, buf += seg_payload) {
334 recv_buf = container_of(recv_buf->list.next,
335 struct ib_mad_recv_buf, list);
336 seg_payload = min(left, max_seg_payload);
337 if (copy_to_user(buf, ((void *) recv_buf->mad) + offset,
343 trace_ib_umad_read_recv(file, &packet->mad.hdr, &recv_buf->mad->mad_hdr);
345 return hdr_size(file) + packet->length;
348 static ssize_t copy_send_mad(struct ib_umad_file *file, char __user *buf,
349 struct ib_umad_packet *packet, size_t count)
351 ssize_t size = hdr_size(file) + packet->length;
356 if (copy_to_user(buf, &packet->mad, hdr_size(file)))
359 buf += hdr_size(file);
361 if (copy_to_user(buf, packet->mad.data, packet->length))
364 trace_ib_umad_read_send(file, &packet->mad.hdr,
365 (struct ib_mad_hdr *)&packet->mad.data);
370 static ssize_t ib_umad_read(struct file *filp, char __user *buf,
371 size_t count, loff_t *pos)
373 struct ib_umad_file *file = filp->private_data;
374 struct ib_umad_packet *packet;
377 if (count < hdr_size(file))
380 mutex_lock(&file->mutex);
382 if (file->agents_dead) {
383 mutex_unlock(&file->mutex);
387 while (list_empty(&file->recv_list)) {
388 mutex_unlock(&file->mutex);
390 if (filp->f_flags & O_NONBLOCK)
393 if (wait_event_interruptible(file->recv_wait,
394 !list_empty(&file->recv_list)))
397 mutex_lock(&file->mutex);
400 if (file->agents_dead) {
401 mutex_unlock(&file->mutex);
405 packet = list_entry(file->recv_list.next, struct ib_umad_packet, list);
406 list_del(&packet->list);
408 mutex_unlock(&file->mutex);
411 ret = copy_recv_mad(file, buf, packet, count);
413 ret = copy_send_mad(file, buf, packet, count);
417 mutex_lock(&file->mutex);
418 list_add(&packet->list, &file->recv_list);
419 mutex_unlock(&file->mutex);
422 ib_free_recv_mad(packet->recv_wc);
428 static int copy_rmpp_mad(struct ib_mad_send_buf *msg, const char __user *buf)
432 /* Copy class specific header */
433 if ((msg->hdr_len > IB_MGMT_RMPP_HDR) &&
434 copy_from_user(msg->mad + IB_MGMT_RMPP_HDR, buf + IB_MGMT_RMPP_HDR,
435 msg->hdr_len - IB_MGMT_RMPP_HDR))
438 /* All headers are in place. Copy data segments. */
439 for (seg = 1, left = msg->data_len, buf += msg->hdr_len; left > 0;
440 seg++, left -= msg->seg_size, buf += msg->seg_size) {
441 if (copy_from_user(ib_get_rmpp_segment(msg, seg), buf,
442 min(left, msg->seg_size)))
448 static int same_destination(struct ib_user_mad_hdr *hdr1,
449 struct ib_user_mad_hdr *hdr2)
451 if (!hdr1->grh_present && !hdr2->grh_present)
452 return (hdr1->lid == hdr2->lid);
454 if (hdr1->grh_present && hdr2->grh_present)
455 return !memcmp(hdr1->gid, hdr2->gid, 16);
460 static int is_duplicate(struct ib_umad_file *file,
461 struct ib_umad_packet *packet)
463 struct ib_umad_packet *sent_packet;
464 struct ib_mad_hdr *sent_hdr, *hdr;
466 hdr = (struct ib_mad_hdr *) packet->mad.data;
467 list_for_each_entry(sent_packet, &file->send_list, list) {
468 sent_hdr = (struct ib_mad_hdr *) sent_packet->mad.data;
470 if ((hdr->tid != sent_hdr->tid) ||
471 (hdr->mgmt_class != sent_hdr->mgmt_class))
475 * No need to be overly clever here. If two new operations have
476 * the same TID, reject the second as a duplicate. This is more
477 * restrictive than required by the spec.
479 if (!ib_response_mad(hdr)) {
480 if (!ib_response_mad(sent_hdr))
483 } else if (!ib_response_mad(sent_hdr))
486 if (same_destination(&packet->mad.hdr, &sent_packet->mad.hdr))
493 static ssize_t ib_umad_write(struct file *filp, const char __user *buf,
494 size_t count, loff_t *pos)
496 struct ib_umad_file *file = filp->private_data;
497 struct ib_umad_packet *packet;
498 struct ib_mad_agent *agent;
499 struct rdma_ah_attr ah_attr;
501 struct ib_rmpp_mad *rmpp_mad;
503 int ret, data_len, hdr_len, copy_offset, rmpp_active;
506 if (count < hdr_size(file) + IB_MGMT_RMPP_HDR)
509 packet = kzalloc(sizeof *packet + IB_MGMT_RMPP_HDR, GFP_KERNEL);
513 if (copy_from_user(&packet->mad, buf, hdr_size(file))) {
518 if (packet->mad.hdr.id >= IB_UMAD_MAX_AGENTS) {
523 buf += hdr_size(file);
525 if (copy_from_user(packet->mad.data, buf, IB_MGMT_RMPP_HDR)) {
530 mutex_lock(&file->mutex);
532 trace_ib_umad_write(file, &packet->mad.hdr,
533 (struct ib_mad_hdr *)&packet->mad.data);
535 agent = __get_agent(file, packet->mad.hdr.id);
541 memset(&ah_attr, 0, sizeof ah_attr);
542 ah_attr.type = rdma_ah_find_type(agent->device,
543 file->port->port_num);
544 rdma_ah_set_dlid(&ah_attr, be16_to_cpu(packet->mad.hdr.lid));
545 rdma_ah_set_sl(&ah_attr, packet->mad.hdr.sl);
546 rdma_ah_set_path_bits(&ah_attr, packet->mad.hdr.path_bits);
547 rdma_ah_set_port_num(&ah_attr, file->port->port_num);
548 if (packet->mad.hdr.grh_present) {
549 rdma_ah_set_grh(&ah_attr, NULL,
550 be32_to_cpu(packet->mad.hdr.flow_label),
551 packet->mad.hdr.gid_index,
552 packet->mad.hdr.hop_limit,
553 packet->mad.hdr.traffic_class);
554 rdma_ah_set_dgid_raw(&ah_attr, packet->mad.hdr.gid);
557 ah = rdma_create_user_ah(agent->qp->pd, &ah_attr, NULL);
563 rmpp_mad = (struct ib_rmpp_mad *) packet->mad.data;
564 hdr_len = ib_get_mad_data_offset(rmpp_mad->mad_hdr.mgmt_class);
566 if (ib_is_mad_class_rmpp(rmpp_mad->mad_hdr.mgmt_class)
567 && ib_mad_kernel_rmpp_agent(agent)) {
568 copy_offset = IB_MGMT_RMPP_HDR;
569 rmpp_active = ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) &
570 IB_MGMT_RMPP_FLAG_ACTIVE;
572 copy_offset = IB_MGMT_MAD_HDR;
576 base_version = ((struct ib_mad_hdr *)&packet->mad.data)->base_version;
577 data_len = count - hdr_size(file) - hdr_len;
578 packet->msg = ib_create_send_mad(agent,
579 be32_to_cpu(packet->mad.hdr.qpn),
580 packet->mad.hdr.pkey_index, rmpp_active,
581 hdr_len, data_len, GFP_KERNEL,
583 if (IS_ERR(packet->msg)) {
584 ret = PTR_ERR(packet->msg);
588 packet->msg->ah = ah;
589 packet->msg->timeout_ms = packet->mad.hdr.timeout_ms;
590 packet->msg->retries = packet->mad.hdr.retries;
591 packet->msg->context[0] = packet;
593 /* Copy MAD header. Any RMPP header is already in place. */
594 memcpy(packet->msg->mad, packet->mad.data, IB_MGMT_MAD_HDR);
597 if (copy_from_user(packet->msg->mad + copy_offset,
599 hdr_len + data_len - copy_offset)) {
604 ret = copy_rmpp_mad(packet->msg, buf);
610 * Set the high-order part of the transaction ID to make MADs from
611 * different agents unique, and allow routing responses back to the
612 * original requestor.
614 if (!ib_response_mad(packet->msg->mad)) {
615 tid = &((struct ib_mad_hdr *) packet->msg->mad)->tid;
616 *tid = cpu_to_be64(((u64) agent->hi_tid) << 32 |
617 (be64_to_cpup(tid) & 0xffffffff));
618 rmpp_mad->mad_hdr.tid = *tid;
621 if (!ib_mad_kernel_rmpp_agent(agent)
622 && ib_is_mad_class_rmpp(rmpp_mad->mad_hdr.mgmt_class)
623 && (ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) & IB_MGMT_RMPP_FLAG_ACTIVE)) {
624 spin_lock_irq(&file->send_lock);
625 list_add_tail(&packet->list, &file->send_list);
626 spin_unlock_irq(&file->send_lock);
628 spin_lock_irq(&file->send_lock);
629 ret = is_duplicate(file, packet);
631 list_add_tail(&packet->list, &file->send_list);
632 spin_unlock_irq(&file->send_lock);
639 ret = ib_post_send_mad(packet->msg, NULL);
643 mutex_unlock(&file->mutex);
647 dequeue_send(file, packet);
649 ib_free_send_mad(packet->msg);
651 rdma_destroy_ah(ah, RDMA_DESTROY_AH_SLEEPABLE);
653 mutex_unlock(&file->mutex);
659 static __poll_t ib_umad_poll(struct file *filp, struct poll_table_struct *wait)
661 struct ib_umad_file *file = filp->private_data;
663 /* we will always be able to post a MAD send */
664 __poll_t mask = EPOLLOUT | EPOLLWRNORM;
666 mutex_lock(&file->mutex);
667 poll_wait(filp, &file->recv_wait, wait);
669 if (!list_empty(&file->recv_list))
670 mask |= EPOLLIN | EPOLLRDNORM;
671 if (file->agents_dead)
673 mutex_unlock(&file->mutex);
678 static int ib_umad_reg_agent(struct ib_umad_file *file, void __user *arg,
679 int compat_method_mask)
681 struct ib_user_mad_reg_req ureq;
682 struct ib_mad_reg_req req;
683 struct ib_mad_agent *agent = NULL;
687 mutex_lock(&file->port->file_mutex);
688 mutex_lock(&file->mutex);
690 if (!file->port->ib_dev) {
691 dev_notice(&file->port->dev,
692 "ib_umad_reg_agent: invalid device\n");
697 if (copy_from_user(&ureq, arg, sizeof ureq)) {
702 if (ureq.qpn != 0 && ureq.qpn != 1) {
703 dev_notice(&file->port->dev,
704 "ib_umad_reg_agent: invalid QPN %d specified\n",
710 for (agent_id = 0; agent_id < IB_UMAD_MAX_AGENTS; ++agent_id)
711 if (!__get_agent(file, agent_id))
714 dev_notice(&file->port->dev,
715 "ib_umad_reg_agent: Max Agents (%u) reached\n",
721 if (ureq.mgmt_class) {
722 memset(&req, 0, sizeof(req));
723 req.mgmt_class = ureq.mgmt_class;
724 req.mgmt_class_version = ureq.mgmt_class_version;
725 memcpy(req.oui, ureq.oui, sizeof req.oui);
727 if (compat_method_mask) {
728 u32 *umm = (u32 *) ureq.method_mask;
731 for (i = 0; i < BITS_TO_LONGS(IB_MGMT_MAX_METHODS); ++i)
733 umm[i * 2] | ((u64) umm[i * 2 + 1] << 32);
735 memcpy(req.method_mask, ureq.method_mask,
736 sizeof req.method_mask);
739 agent = ib_register_mad_agent(file->port->ib_dev, file->port->port_num,
740 ureq.qpn ? IB_QPT_GSI : IB_QPT_SMI,
741 ureq.mgmt_class ? &req : NULL,
743 send_handler, recv_handler, file, 0);
745 ret = PTR_ERR(agent);
750 if (put_user(agent_id,
751 (u32 __user *) (arg + offsetof(struct ib_user_mad_reg_req, id)))) {
756 if (!file->already_used) {
757 file->already_used = 1;
758 if (!file->use_pkey_index) {
759 dev_warn(&file->port->dev,
760 "process %s did not enable P_Key index support.\n",
762 dev_warn(&file->port->dev,
763 " Documentation/infiniband/user_mad.rst has info on the new ABI.\n");
767 file->agent[agent_id] = agent;
771 mutex_unlock(&file->mutex);
774 ib_unregister_mad_agent(agent);
776 mutex_unlock(&file->port->file_mutex);
781 static int ib_umad_reg_agent2(struct ib_umad_file *file, void __user *arg)
783 struct ib_user_mad_reg_req2 ureq;
784 struct ib_mad_reg_req req;
785 struct ib_mad_agent *agent = NULL;
789 mutex_lock(&file->port->file_mutex);
790 mutex_lock(&file->mutex);
792 if (!file->port->ib_dev) {
793 dev_notice(&file->port->dev,
794 "ib_umad_reg_agent2: invalid device\n");
799 if (copy_from_user(&ureq, arg, sizeof(ureq))) {
804 if (ureq.qpn != 0 && ureq.qpn != 1) {
805 dev_notice(&file->port->dev,
806 "ib_umad_reg_agent2: invalid QPN %d specified\n",
812 if (ureq.flags & ~IB_USER_MAD_REG_FLAGS_CAP) {
813 dev_notice(&file->port->dev,
814 "ib_umad_reg_agent2 failed: invalid registration flags specified 0x%x; supported 0x%x\n",
815 ureq.flags, IB_USER_MAD_REG_FLAGS_CAP);
818 if (put_user((u32)IB_USER_MAD_REG_FLAGS_CAP,
819 (u32 __user *) (arg + offsetof(struct
820 ib_user_mad_reg_req2, flags))))
826 for (agent_id = 0; agent_id < IB_UMAD_MAX_AGENTS; ++agent_id)
827 if (!__get_agent(file, agent_id))
830 dev_notice(&file->port->dev,
831 "ib_umad_reg_agent2: Max Agents (%u) reached\n",
837 if (ureq.mgmt_class) {
838 memset(&req, 0, sizeof(req));
839 req.mgmt_class = ureq.mgmt_class;
840 req.mgmt_class_version = ureq.mgmt_class_version;
841 if (ureq.oui & 0xff000000) {
842 dev_notice(&file->port->dev,
843 "ib_umad_reg_agent2 failed: oui invalid 0x%08x\n",
848 req.oui[2] = ureq.oui & 0x0000ff;
849 req.oui[1] = (ureq.oui & 0x00ff00) >> 8;
850 req.oui[0] = (ureq.oui & 0xff0000) >> 16;
851 memcpy(req.method_mask, ureq.method_mask,
852 sizeof(req.method_mask));
855 agent = ib_register_mad_agent(file->port->ib_dev, file->port->port_num,
856 ureq.qpn ? IB_QPT_GSI : IB_QPT_SMI,
857 ureq.mgmt_class ? &req : NULL,
859 send_handler, recv_handler, file,
862 ret = PTR_ERR(agent);
867 if (put_user(agent_id,
869 offsetof(struct ib_user_mad_reg_req2, id)))) {
874 if (!file->already_used) {
875 file->already_used = 1;
876 file->use_pkey_index = 1;
879 file->agent[agent_id] = agent;
883 mutex_unlock(&file->mutex);
886 ib_unregister_mad_agent(agent);
888 mutex_unlock(&file->port->file_mutex);
894 static int ib_umad_unreg_agent(struct ib_umad_file *file, u32 __user *arg)
896 struct ib_mad_agent *agent = NULL;
900 if (get_user(id, arg))
902 if (id >= IB_UMAD_MAX_AGENTS)
905 mutex_lock(&file->port->file_mutex);
906 mutex_lock(&file->mutex);
908 id = array_index_nospec(id, IB_UMAD_MAX_AGENTS);
909 if (!__get_agent(file, id)) {
914 agent = file->agent[id];
915 file->agent[id] = NULL;
918 mutex_unlock(&file->mutex);
921 ib_unregister_mad_agent(agent);
923 mutex_unlock(&file->port->file_mutex);
928 static long ib_umad_enable_pkey(struct ib_umad_file *file)
932 mutex_lock(&file->mutex);
933 if (file->already_used)
936 file->use_pkey_index = 1;
937 mutex_unlock(&file->mutex);
942 static long ib_umad_ioctl(struct file *filp, unsigned int cmd,
946 case IB_USER_MAD_REGISTER_AGENT:
947 return ib_umad_reg_agent(filp->private_data, (void __user *) arg, 0);
948 case IB_USER_MAD_UNREGISTER_AGENT:
949 return ib_umad_unreg_agent(filp->private_data, (__u32 __user *) arg);
950 case IB_USER_MAD_ENABLE_PKEY:
951 return ib_umad_enable_pkey(filp->private_data);
952 case IB_USER_MAD_REGISTER_AGENT2:
953 return ib_umad_reg_agent2(filp->private_data, (void __user *) arg);
960 static long ib_umad_compat_ioctl(struct file *filp, unsigned int cmd,
964 case IB_USER_MAD_REGISTER_AGENT:
965 return ib_umad_reg_agent(filp->private_data, compat_ptr(arg), 1);
966 case IB_USER_MAD_UNREGISTER_AGENT:
967 return ib_umad_unreg_agent(filp->private_data, compat_ptr(arg));
968 case IB_USER_MAD_ENABLE_PKEY:
969 return ib_umad_enable_pkey(filp->private_data);
970 case IB_USER_MAD_REGISTER_AGENT2:
971 return ib_umad_reg_agent2(filp->private_data, compat_ptr(arg));
979 * ib_umad_open() does not need the BKL:
981 * - the ib_umad_port structures are properly reference counted, and
982 * everything else is purely local to the file being created, so
983 * races against other open calls are not a problem;
984 * - the ioctl method does not affect any global state outside of the
985 * file structure being operated on;
987 static int ib_umad_open(struct inode *inode, struct file *filp)
989 struct ib_umad_port *port;
990 struct ib_umad_file *file;
993 port = container_of(inode->i_cdev, struct ib_umad_port, cdev);
995 mutex_lock(&port->file_mutex);
1002 if (!rdma_dev_access_netns(port->ib_dev, current->nsproxy->net_ns)) {
1007 file = kzalloc(sizeof(*file), GFP_KERNEL);
1013 mutex_init(&file->mutex);
1014 spin_lock_init(&file->send_lock);
1015 INIT_LIST_HEAD(&file->recv_list);
1016 INIT_LIST_HEAD(&file->send_list);
1017 init_waitqueue_head(&file->recv_wait);
1020 filp->private_data = file;
1022 list_add_tail(&file->port_list, &port->file_list);
1024 stream_open(inode, filp);
1026 mutex_unlock(&port->file_mutex);
1030 static int ib_umad_close(struct inode *inode, struct file *filp)
1032 struct ib_umad_file *file = filp->private_data;
1033 struct ib_umad_packet *packet, *tmp;
1037 mutex_lock(&file->port->file_mutex);
1038 mutex_lock(&file->mutex);
1040 already_dead = file->agents_dead;
1041 file->agents_dead = 1;
1043 list_for_each_entry_safe(packet, tmp, &file->recv_list, list) {
1044 if (packet->recv_wc)
1045 ib_free_recv_mad(packet->recv_wc);
1049 list_del(&file->port_list);
1051 mutex_unlock(&file->mutex);
1054 for (i = 0; i < IB_UMAD_MAX_AGENTS; ++i)
1056 ib_unregister_mad_agent(file->agent[i]);
1058 mutex_unlock(&file->port->file_mutex);
1059 mutex_destroy(&file->mutex);
1064 static const struct file_operations umad_fops = {
1065 .owner = THIS_MODULE,
1066 .read = ib_umad_read,
1067 .write = ib_umad_write,
1068 .poll = ib_umad_poll,
1069 .unlocked_ioctl = ib_umad_ioctl,
1070 #ifdef CONFIG_COMPAT
1071 .compat_ioctl = ib_umad_compat_ioctl,
1073 .open = ib_umad_open,
1074 .release = ib_umad_close,
1075 .llseek = no_llseek,
1078 static int ib_umad_sm_open(struct inode *inode, struct file *filp)
1080 struct ib_umad_port *port;
1081 struct ib_port_modify props = {
1082 .set_port_cap_mask = IB_PORT_SM
1086 port = container_of(inode->i_cdev, struct ib_umad_port, sm_cdev);
1088 if (filp->f_flags & O_NONBLOCK) {
1089 if (down_trylock(&port->sm_sem)) {
1094 if (down_interruptible(&port->sm_sem)) {
1100 if (!rdma_dev_access_netns(port->ib_dev, current->nsproxy->net_ns)) {
1105 ret = ib_modify_port(port->ib_dev, port->port_num, 0, &props);
1109 filp->private_data = port;
1111 nonseekable_open(inode, filp);
1121 static int ib_umad_sm_close(struct inode *inode, struct file *filp)
1123 struct ib_umad_port *port = filp->private_data;
1124 struct ib_port_modify props = {
1125 .clr_port_cap_mask = IB_PORT_SM
1129 mutex_lock(&port->file_mutex);
1131 ret = ib_modify_port(port->ib_dev, port->port_num, 0, &props);
1132 mutex_unlock(&port->file_mutex);
1139 static const struct file_operations umad_sm_fops = {
1140 .owner = THIS_MODULE,
1141 .open = ib_umad_sm_open,
1142 .release = ib_umad_sm_close,
1143 .llseek = no_llseek,
1146 static struct ib_umad_port *get_port(struct ib_device *ibdev,
1147 struct ib_umad_device *umad_dev,
1151 return ERR_PTR(-EOPNOTSUPP);
1152 if (!rdma_is_port_valid(ibdev, port))
1153 return ERR_PTR(-EINVAL);
1154 if (!rdma_cap_ib_mad(ibdev, port))
1155 return ERR_PTR(-EOPNOTSUPP);
1157 return &umad_dev->ports[port - rdma_start_port(ibdev)];
1160 static int ib_umad_get_nl_info(struct ib_device *ibdev, void *client_data,
1161 struct ib_client_nl_info *res)
1163 struct ib_umad_port *port = get_port(ibdev, client_data, res->port);
1166 return PTR_ERR(port);
1168 res->abi = IB_USER_MAD_ABI_VERSION;
1169 res->cdev = &port->dev;
1173 static struct ib_client umad_client = {
1175 .add = ib_umad_add_one,
1176 .remove = ib_umad_remove_one,
1177 .get_nl_info = ib_umad_get_nl_info,
1179 MODULE_ALIAS_RDMA_CLIENT("umad");
1181 static int ib_issm_get_nl_info(struct ib_device *ibdev, void *client_data,
1182 struct ib_client_nl_info *res)
1184 struct ib_umad_port *port = get_port(ibdev, client_data, res->port);
1187 return PTR_ERR(port);
1189 res->abi = IB_USER_MAD_ABI_VERSION;
1190 res->cdev = &port->sm_dev;
1194 static struct ib_client issm_client = {
1196 .get_nl_info = ib_issm_get_nl_info,
1198 MODULE_ALIAS_RDMA_CLIENT("issm");
1200 static ssize_t ibdev_show(struct device *dev, struct device_attribute *attr,
1203 struct ib_umad_port *port = dev_get_drvdata(dev);
1208 return sprintf(buf, "%s\n", dev_name(&port->ib_dev->dev));
1210 static DEVICE_ATTR_RO(ibdev);
1212 static ssize_t port_show(struct device *dev, struct device_attribute *attr,
1215 struct ib_umad_port *port = dev_get_drvdata(dev);
1220 return sprintf(buf, "%d\n", port->port_num);
1222 static DEVICE_ATTR_RO(port);
1224 static struct attribute *umad_class_dev_attrs[] = {
1225 &dev_attr_ibdev.attr,
1226 &dev_attr_port.attr,
1229 ATTRIBUTE_GROUPS(umad_class_dev);
1231 static char *umad_devnode(struct device *dev, umode_t *mode)
1233 return kasprintf(GFP_KERNEL, "infiniband/%s", dev_name(dev));
1236 static ssize_t abi_version_show(struct class *class,
1237 struct class_attribute *attr, char *buf)
1239 return sprintf(buf, "%d\n", IB_USER_MAD_ABI_VERSION);
1241 static CLASS_ATTR_RO(abi_version);
1243 static struct attribute *umad_class_attrs[] = {
1244 &class_attr_abi_version.attr,
1247 ATTRIBUTE_GROUPS(umad_class);
1249 static struct class umad_class = {
1250 .name = "infiniband_mad",
1251 .devnode = umad_devnode,
1252 .class_groups = umad_class_groups,
1253 .dev_groups = umad_class_dev_groups,
1256 static void ib_umad_release_port(struct device *device)
1258 struct ib_umad_port *port = dev_get_drvdata(device);
1259 struct ib_umad_device *umad_dev = port->umad_dev;
1261 ib_umad_dev_put(umad_dev);
1264 static void ib_umad_init_port_dev(struct device *dev,
1265 struct ib_umad_port *port,
1266 const struct ib_device *device)
1268 device_initialize(dev);
1269 ib_umad_dev_get(port->umad_dev);
1270 dev->class = &umad_class;
1271 dev->parent = device->dev.parent;
1272 dev_set_drvdata(dev, port);
1273 dev->release = ib_umad_release_port;
1276 static int ib_umad_init_port(struct ib_device *device, int port_num,
1277 struct ib_umad_device *umad_dev,
1278 struct ib_umad_port *port)
1285 devnum = ida_alloc_max(&umad_ida, IB_UMAD_MAX_PORTS - 1, GFP_KERNEL);
1288 port->dev_num = devnum;
1289 if (devnum >= IB_UMAD_NUM_FIXED_MINOR) {
1290 base_umad = dynamic_umad_dev + devnum - IB_UMAD_NUM_FIXED_MINOR;
1291 base_issm = dynamic_issm_dev + devnum - IB_UMAD_NUM_FIXED_MINOR;
1293 base_umad = devnum + base_umad_dev;
1294 base_issm = devnum + base_issm_dev;
1297 port->ib_dev = device;
1298 port->umad_dev = umad_dev;
1299 port->port_num = port_num;
1300 sema_init(&port->sm_sem, 1);
1301 mutex_init(&port->file_mutex);
1302 INIT_LIST_HEAD(&port->file_list);
1304 ib_umad_init_port_dev(&port->dev, port, device);
1305 port->dev.devt = base_umad;
1306 dev_set_name(&port->dev, "umad%d", port->dev_num);
1307 cdev_init(&port->cdev, &umad_fops);
1308 port->cdev.owner = THIS_MODULE;
1310 ret = cdev_device_add(&port->cdev, &port->dev);
1314 ib_umad_init_port_dev(&port->sm_dev, port, device);
1315 port->sm_dev.devt = base_issm;
1316 dev_set_name(&port->sm_dev, "issm%d", port->dev_num);
1317 cdev_init(&port->sm_cdev, &umad_sm_fops);
1318 port->sm_cdev.owner = THIS_MODULE;
1320 ret = cdev_device_add(&port->sm_cdev, &port->sm_dev);
1327 put_device(&port->sm_dev);
1328 cdev_device_del(&port->cdev, &port->dev);
1330 put_device(&port->dev);
1331 ida_free(&umad_ida, devnum);
1335 static void ib_umad_kill_port(struct ib_umad_port *port)
1337 struct ib_umad_file *file;
1340 cdev_device_del(&port->sm_cdev, &port->sm_dev);
1341 cdev_device_del(&port->cdev, &port->dev);
1343 mutex_lock(&port->file_mutex);
1345 /* Mark ib_dev NULL and block ioctl or other file ops to progress
1348 port->ib_dev = NULL;
1350 list_for_each_entry(file, &port->file_list, port_list) {
1351 mutex_lock(&file->mutex);
1352 file->agents_dead = 1;
1353 wake_up_interruptible(&file->recv_wait);
1354 mutex_unlock(&file->mutex);
1356 for (id = 0; id < IB_UMAD_MAX_AGENTS; ++id)
1357 if (file->agent[id])
1358 ib_unregister_mad_agent(file->agent[id]);
1361 mutex_unlock(&port->file_mutex);
1363 ida_free(&umad_ida, port->dev_num);
1365 /* balances device_initialize() */
1366 put_device(&port->sm_dev);
1367 put_device(&port->dev);
1370 static int ib_umad_add_one(struct ib_device *device)
1372 struct ib_umad_device *umad_dev;
1377 s = rdma_start_port(device);
1378 e = rdma_end_port(device);
1380 umad_dev = kzalloc(struct_size(umad_dev, ports, e - s + 1), GFP_KERNEL);
1384 kref_init(&umad_dev->kref);
1385 for (i = s; i <= e; ++i) {
1386 if (!rdma_cap_ib_mad(device, i))
1389 ret = ib_umad_init_port(device, i, umad_dev,
1390 &umad_dev->ports[i - s]);
1402 ib_set_client_data(device, &umad_client, umad_dev);
1408 if (!rdma_cap_ib_mad(device, i))
1411 ib_umad_kill_port(&umad_dev->ports[i - s]);
1414 /* balances kref_init */
1415 ib_umad_dev_put(umad_dev);
1419 static void ib_umad_remove_one(struct ib_device *device, void *client_data)
1421 struct ib_umad_device *umad_dev = client_data;
1424 rdma_for_each_port (device, i) {
1425 if (rdma_cap_ib_mad(device, i))
1427 &umad_dev->ports[i - rdma_start_port(device)]);
1429 /* balances kref_init() */
1430 ib_umad_dev_put(umad_dev);
1433 static int __init ib_umad_init(void)
1437 ret = register_chrdev_region(base_umad_dev,
1438 IB_UMAD_NUM_FIXED_MINOR * 2,
1441 pr_err("couldn't register device number\n");
1445 ret = alloc_chrdev_region(&dynamic_umad_dev, 0,
1446 IB_UMAD_NUM_DYNAMIC_MINOR * 2,
1449 pr_err("couldn't register dynamic device number\n");
1452 dynamic_issm_dev = dynamic_umad_dev + IB_UMAD_NUM_DYNAMIC_MINOR;
1454 ret = class_register(&umad_class);
1456 pr_err("couldn't create class infiniband_mad\n");
1460 ret = ib_register_client(&umad_client);
1464 ret = ib_register_client(&issm_client);
1471 ib_unregister_client(&umad_client);
1473 class_unregister(&umad_class);
1476 unregister_chrdev_region(dynamic_umad_dev,
1477 IB_UMAD_NUM_DYNAMIC_MINOR * 2);
1480 unregister_chrdev_region(base_umad_dev,
1481 IB_UMAD_NUM_FIXED_MINOR * 2);
1487 static void __exit ib_umad_cleanup(void)
1489 ib_unregister_client(&issm_client);
1490 ib_unregister_client(&umad_client);
1491 class_unregister(&umad_class);
1492 unregister_chrdev_region(base_umad_dev,
1493 IB_UMAD_NUM_FIXED_MINOR * 2);
1494 unregister_chrdev_region(dynamic_umad_dev,
1495 IB_UMAD_NUM_DYNAMIC_MINOR * 2);
1498 module_init(ib_umad_init);
1499 module_exit(ib_umad_cleanup);