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 struct ib_rmpp_mad_hdr {
135 struct ib_mad_hdr mad_hdr;
136 struct ib_rmpp_hdr rmpp_hdr;
139 #define CREATE_TRACE_POINTS
140 #include <trace/events/ib_umad.h>
142 static const dev_t base_umad_dev = MKDEV(IB_UMAD_MAJOR, IB_UMAD_MINOR_BASE);
143 static const dev_t base_issm_dev = MKDEV(IB_UMAD_MAJOR, IB_UMAD_MINOR_BASE) +
144 IB_UMAD_NUM_FIXED_MINOR;
145 static dev_t dynamic_umad_dev;
146 static dev_t dynamic_issm_dev;
148 static DEFINE_IDA(umad_ida);
150 static int ib_umad_add_one(struct ib_device *device);
151 static void ib_umad_remove_one(struct ib_device *device, void *client_data);
153 static void ib_umad_dev_free(struct kref *kref)
155 struct ib_umad_device *dev =
156 container_of(kref, struct ib_umad_device, kref);
161 static void ib_umad_dev_get(struct ib_umad_device *dev)
163 kref_get(&dev->kref);
166 static void ib_umad_dev_put(struct ib_umad_device *dev)
168 kref_put(&dev->kref, ib_umad_dev_free);
171 static int hdr_size(struct ib_umad_file *file)
173 return file->use_pkey_index ? sizeof(struct ib_user_mad_hdr) :
174 sizeof(struct ib_user_mad_hdr_old);
177 /* caller must hold file->mutex */
178 static struct ib_mad_agent *__get_agent(struct ib_umad_file *file, int id)
180 return file->agents_dead ? NULL : file->agent[id];
183 static int queue_packet(struct ib_umad_file *file,
184 struct ib_mad_agent *agent,
185 struct ib_umad_packet *packet)
189 mutex_lock(&file->mutex);
191 for (packet->mad.hdr.id = 0;
192 packet->mad.hdr.id < IB_UMAD_MAX_AGENTS;
193 packet->mad.hdr.id++)
194 if (agent == __get_agent(file, packet->mad.hdr.id)) {
195 list_add_tail(&packet->list, &file->recv_list);
196 wake_up_interruptible(&file->recv_wait);
201 mutex_unlock(&file->mutex);
206 static void dequeue_send(struct ib_umad_file *file,
207 struct ib_umad_packet *packet)
209 spin_lock_irq(&file->send_lock);
210 list_del(&packet->list);
211 spin_unlock_irq(&file->send_lock);
214 static void send_handler(struct ib_mad_agent *agent,
215 struct ib_mad_send_wc *send_wc)
217 struct ib_umad_file *file = agent->context;
218 struct ib_umad_packet *packet = send_wc->send_buf->context[0];
220 dequeue_send(file, packet);
221 rdma_destroy_ah(packet->msg->ah, RDMA_DESTROY_AH_SLEEPABLE);
222 ib_free_send_mad(packet->msg);
224 if (send_wc->status == IB_WC_RESP_TIMEOUT_ERR) {
225 packet->length = IB_MGMT_MAD_HDR;
226 packet->mad.hdr.status = ETIMEDOUT;
227 if (!queue_packet(file, agent, packet))
233 static void recv_handler(struct ib_mad_agent *agent,
234 struct ib_mad_send_buf *send_buf,
235 struct ib_mad_recv_wc *mad_recv_wc)
237 struct ib_umad_file *file = agent->context;
238 struct ib_umad_packet *packet;
240 if (mad_recv_wc->wc->status != IB_WC_SUCCESS)
243 packet = kzalloc(sizeof *packet, GFP_KERNEL);
247 packet->length = mad_recv_wc->mad_len;
248 packet->recv_wc = mad_recv_wc;
250 packet->mad.hdr.status = 0;
251 packet->mad.hdr.length = hdr_size(file) + mad_recv_wc->mad_len;
252 packet->mad.hdr.qpn = cpu_to_be32(mad_recv_wc->wc->src_qp);
254 * On OPA devices it is okay to lose the upper 16 bits of LID as this
255 * information is obtained elsewhere. Mask off the upper 16 bits.
257 if (rdma_cap_opa_mad(agent->device, agent->port_num))
258 packet->mad.hdr.lid = ib_lid_be16(0xFFFF &
259 mad_recv_wc->wc->slid);
261 packet->mad.hdr.lid = ib_lid_be16(mad_recv_wc->wc->slid);
262 packet->mad.hdr.sl = mad_recv_wc->wc->sl;
263 packet->mad.hdr.path_bits = mad_recv_wc->wc->dlid_path_bits;
264 packet->mad.hdr.pkey_index = mad_recv_wc->wc->pkey_index;
265 packet->mad.hdr.grh_present = !!(mad_recv_wc->wc->wc_flags & IB_WC_GRH);
266 if (packet->mad.hdr.grh_present) {
267 struct rdma_ah_attr ah_attr;
268 const struct ib_global_route *grh;
271 ret = ib_init_ah_attr_from_wc(agent->device, agent->port_num,
273 mad_recv_wc->recv_buf.grh,
278 grh = rdma_ah_read_grh(&ah_attr);
279 packet->mad.hdr.gid_index = grh->sgid_index;
280 packet->mad.hdr.hop_limit = grh->hop_limit;
281 packet->mad.hdr.traffic_class = grh->traffic_class;
282 memcpy(packet->mad.hdr.gid, &grh->dgid, 16);
283 packet->mad.hdr.flow_label = cpu_to_be32(grh->flow_label);
284 rdma_destroy_ah_attr(&ah_attr);
287 if (queue_packet(file, agent, packet))
294 ib_free_recv_mad(mad_recv_wc);
297 static ssize_t copy_recv_mad(struct ib_umad_file *file, char __user *buf,
298 struct ib_umad_packet *packet, size_t count)
300 struct ib_mad_recv_buf *recv_buf;
301 int left, seg_payload, offset, max_seg_payload;
304 recv_buf = &packet->recv_wc->recv_buf;
305 seg_size = packet->recv_wc->mad_seg_size;
307 /* We need enough room to copy the first (or only) MAD segment. */
308 if ((packet->length <= seg_size &&
309 count < hdr_size(file) + packet->length) ||
310 (packet->length > seg_size &&
311 count < hdr_size(file) + seg_size))
314 if (copy_to_user(buf, &packet->mad, hdr_size(file)))
317 buf += hdr_size(file);
318 seg_payload = min_t(int, packet->length, seg_size);
319 if (copy_to_user(buf, recv_buf->mad, seg_payload))
322 if (seg_payload < packet->length) {
324 * Multipacket RMPP MAD message. Copy remainder of message.
325 * Note that last segment may have a shorter payload.
327 if (count < hdr_size(file) + packet->length) {
329 * The buffer is too small, return the first RMPP segment,
330 * which includes the RMPP message length.
334 offset = ib_get_mad_data_offset(recv_buf->mad->mad_hdr.mgmt_class);
335 max_seg_payload = seg_size - offset;
337 for (left = packet->length - seg_payload, buf += seg_payload;
338 left; left -= seg_payload, buf += seg_payload) {
339 recv_buf = container_of(recv_buf->list.next,
340 struct ib_mad_recv_buf, list);
341 seg_payload = min(left, max_seg_payload);
342 if (copy_to_user(buf, ((void *) recv_buf->mad) + offset,
348 trace_ib_umad_read_recv(file, &packet->mad.hdr, &recv_buf->mad->mad_hdr);
350 return hdr_size(file) + packet->length;
353 static ssize_t copy_send_mad(struct ib_umad_file *file, char __user *buf,
354 struct ib_umad_packet *packet, size_t count)
356 ssize_t size = hdr_size(file) + packet->length;
361 if (copy_to_user(buf, &packet->mad, hdr_size(file)))
364 buf += hdr_size(file);
366 if (copy_to_user(buf, packet->mad.data, packet->length))
369 trace_ib_umad_read_send(file, &packet->mad.hdr,
370 (struct ib_mad_hdr *)&packet->mad.data);
375 static ssize_t ib_umad_read(struct file *filp, char __user *buf,
376 size_t count, loff_t *pos)
378 struct ib_umad_file *file = filp->private_data;
379 struct ib_umad_packet *packet;
382 if (count < hdr_size(file))
385 mutex_lock(&file->mutex);
387 if (file->agents_dead) {
388 mutex_unlock(&file->mutex);
392 while (list_empty(&file->recv_list)) {
393 mutex_unlock(&file->mutex);
395 if (filp->f_flags & O_NONBLOCK)
398 if (wait_event_interruptible(file->recv_wait,
399 !list_empty(&file->recv_list)))
402 mutex_lock(&file->mutex);
405 if (file->agents_dead) {
406 mutex_unlock(&file->mutex);
410 packet = list_entry(file->recv_list.next, struct ib_umad_packet, list);
411 list_del(&packet->list);
413 mutex_unlock(&file->mutex);
416 ret = copy_recv_mad(file, buf, packet, count);
418 ret = copy_send_mad(file, buf, packet, count);
422 mutex_lock(&file->mutex);
423 list_add(&packet->list, &file->recv_list);
424 mutex_unlock(&file->mutex);
427 ib_free_recv_mad(packet->recv_wc);
433 static int copy_rmpp_mad(struct ib_mad_send_buf *msg, const char __user *buf)
437 /* Copy class specific header */
438 if ((msg->hdr_len > IB_MGMT_RMPP_HDR) &&
439 copy_from_user(msg->mad + IB_MGMT_RMPP_HDR, buf + IB_MGMT_RMPP_HDR,
440 msg->hdr_len - IB_MGMT_RMPP_HDR))
443 /* All headers are in place. Copy data segments. */
444 for (seg = 1, left = msg->data_len, buf += msg->hdr_len; left > 0;
445 seg++, left -= msg->seg_size, buf += msg->seg_size) {
446 if (copy_from_user(ib_get_rmpp_segment(msg, seg), buf,
447 min(left, msg->seg_size)))
453 static int same_destination(struct ib_user_mad_hdr *hdr1,
454 struct ib_user_mad_hdr *hdr2)
456 if (!hdr1->grh_present && !hdr2->grh_present)
457 return (hdr1->lid == hdr2->lid);
459 if (hdr1->grh_present && hdr2->grh_present)
460 return !memcmp(hdr1->gid, hdr2->gid, 16);
465 static int is_duplicate(struct ib_umad_file *file,
466 struct ib_umad_packet *packet)
468 struct ib_umad_packet *sent_packet;
469 struct ib_mad_hdr *sent_hdr, *hdr;
471 hdr = (struct ib_mad_hdr *) packet->mad.data;
472 list_for_each_entry(sent_packet, &file->send_list, list) {
473 sent_hdr = (struct ib_mad_hdr *) sent_packet->mad.data;
475 if ((hdr->tid != sent_hdr->tid) ||
476 (hdr->mgmt_class != sent_hdr->mgmt_class))
480 * No need to be overly clever here. If two new operations have
481 * the same TID, reject the second as a duplicate. This is more
482 * restrictive than required by the spec.
484 if (!ib_response_mad(hdr)) {
485 if (!ib_response_mad(sent_hdr))
488 } else if (!ib_response_mad(sent_hdr))
491 if (same_destination(&packet->mad.hdr, &sent_packet->mad.hdr))
498 static ssize_t ib_umad_write(struct file *filp, const char __user *buf,
499 size_t count, loff_t *pos)
501 struct ib_umad_file *file = filp->private_data;
502 struct ib_rmpp_mad_hdr *rmpp_mad_hdr;
503 struct ib_umad_packet *packet;
504 struct ib_mad_agent *agent;
505 struct rdma_ah_attr ah_attr;
508 int ret, data_len, hdr_len, copy_offset, rmpp_active;
511 if (count < hdr_size(file) + IB_MGMT_RMPP_HDR)
514 packet = kzalloc(sizeof(*packet) + IB_MGMT_RMPP_HDR, GFP_KERNEL);
518 if (copy_from_user(&packet->mad, buf, hdr_size(file))) {
523 if (packet->mad.hdr.id >= IB_UMAD_MAX_AGENTS) {
528 buf += hdr_size(file);
530 if (copy_from_user(packet->mad.data, buf, IB_MGMT_RMPP_HDR)) {
535 mutex_lock(&file->mutex);
537 trace_ib_umad_write(file, &packet->mad.hdr,
538 (struct ib_mad_hdr *)&packet->mad.data);
540 agent = __get_agent(file, packet->mad.hdr.id);
546 memset(&ah_attr, 0, sizeof ah_attr);
547 ah_attr.type = rdma_ah_find_type(agent->device,
548 file->port->port_num);
549 rdma_ah_set_dlid(&ah_attr, be16_to_cpu(packet->mad.hdr.lid));
550 rdma_ah_set_sl(&ah_attr, packet->mad.hdr.sl);
551 rdma_ah_set_path_bits(&ah_attr, packet->mad.hdr.path_bits);
552 rdma_ah_set_port_num(&ah_attr, file->port->port_num);
553 if (packet->mad.hdr.grh_present) {
554 rdma_ah_set_grh(&ah_attr, NULL,
555 be32_to_cpu(packet->mad.hdr.flow_label),
556 packet->mad.hdr.gid_index,
557 packet->mad.hdr.hop_limit,
558 packet->mad.hdr.traffic_class);
559 rdma_ah_set_dgid_raw(&ah_attr, packet->mad.hdr.gid);
562 ah = rdma_create_user_ah(agent->qp->pd, &ah_attr, NULL);
568 rmpp_mad_hdr = (struct ib_rmpp_mad_hdr *)packet->mad.data;
569 hdr_len = ib_get_mad_data_offset(rmpp_mad_hdr->mad_hdr.mgmt_class);
571 if (ib_is_mad_class_rmpp(rmpp_mad_hdr->mad_hdr.mgmt_class)
572 && ib_mad_kernel_rmpp_agent(agent)) {
573 copy_offset = IB_MGMT_RMPP_HDR;
574 rmpp_active = ib_get_rmpp_flags(&rmpp_mad_hdr->rmpp_hdr) &
575 IB_MGMT_RMPP_FLAG_ACTIVE;
577 copy_offset = IB_MGMT_MAD_HDR;
581 base_version = ((struct ib_mad_hdr *)&packet->mad.data)->base_version;
582 data_len = count - hdr_size(file) - hdr_len;
583 packet->msg = ib_create_send_mad(agent,
584 be32_to_cpu(packet->mad.hdr.qpn),
585 packet->mad.hdr.pkey_index, rmpp_active,
586 hdr_len, data_len, GFP_KERNEL,
588 if (IS_ERR(packet->msg)) {
589 ret = PTR_ERR(packet->msg);
593 packet->msg->ah = ah;
594 packet->msg->timeout_ms = packet->mad.hdr.timeout_ms;
595 packet->msg->retries = packet->mad.hdr.retries;
596 packet->msg->context[0] = packet;
598 /* Copy MAD header. Any RMPP header is already in place. */
599 memcpy(packet->msg->mad, packet->mad.data, IB_MGMT_MAD_HDR);
602 if (copy_from_user(packet->msg->mad + copy_offset,
604 hdr_len + data_len - copy_offset)) {
609 ret = copy_rmpp_mad(packet->msg, buf);
615 * Set the high-order part of the transaction ID to make MADs from
616 * different agents unique, and allow routing responses back to the
617 * original requestor.
619 if (!ib_response_mad(packet->msg->mad)) {
620 tid = &((struct ib_mad_hdr *) packet->msg->mad)->tid;
621 *tid = cpu_to_be64(((u64) agent->hi_tid) << 32 |
622 (be64_to_cpup(tid) & 0xffffffff));
623 rmpp_mad_hdr->mad_hdr.tid = *tid;
626 if (!ib_mad_kernel_rmpp_agent(agent)
627 && ib_is_mad_class_rmpp(rmpp_mad_hdr->mad_hdr.mgmt_class)
628 && (ib_get_rmpp_flags(&rmpp_mad_hdr->rmpp_hdr) & IB_MGMT_RMPP_FLAG_ACTIVE)) {
629 spin_lock_irq(&file->send_lock);
630 list_add_tail(&packet->list, &file->send_list);
631 spin_unlock_irq(&file->send_lock);
633 spin_lock_irq(&file->send_lock);
634 ret = is_duplicate(file, packet);
636 list_add_tail(&packet->list, &file->send_list);
637 spin_unlock_irq(&file->send_lock);
644 ret = ib_post_send_mad(packet->msg, NULL);
648 mutex_unlock(&file->mutex);
652 dequeue_send(file, packet);
654 ib_free_send_mad(packet->msg);
656 rdma_destroy_ah(ah, RDMA_DESTROY_AH_SLEEPABLE);
658 mutex_unlock(&file->mutex);
664 static __poll_t ib_umad_poll(struct file *filp, struct poll_table_struct *wait)
666 struct ib_umad_file *file = filp->private_data;
668 /* we will always be able to post a MAD send */
669 __poll_t mask = EPOLLOUT | EPOLLWRNORM;
671 mutex_lock(&file->mutex);
672 poll_wait(filp, &file->recv_wait, wait);
674 if (!list_empty(&file->recv_list))
675 mask |= EPOLLIN | EPOLLRDNORM;
676 if (file->agents_dead)
678 mutex_unlock(&file->mutex);
683 static int ib_umad_reg_agent(struct ib_umad_file *file, void __user *arg,
684 int compat_method_mask)
686 struct ib_user_mad_reg_req ureq;
687 struct ib_mad_reg_req req;
688 struct ib_mad_agent *agent = NULL;
692 mutex_lock(&file->port->file_mutex);
693 mutex_lock(&file->mutex);
695 if (!file->port->ib_dev) {
696 dev_notice(&file->port->dev, "%s: invalid device\n", __func__);
701 if (copy_from_user(&ureq, arg, sizeof ureq)) {
706 if (ureq.qpn != 0 && ureq.qpn != 1) {
707 dev_notice(&file->port->dev,
708 "%s: invalid QPN %u specified\n", __func__,
714 for (agent_id = 0; agent_id < IB_UMAD_MAX_AGENTS; ++agent_id)
715 if (!__get_agent(file, agent_id))
718 dev_notice(&file->port->dev, "%s: Max Agents (%u) reached\n", __func__,
725 if (ureq.mgmt_class) {
726 memset(&req, 0, sizeof(req));
727 req.mgmt_class = ureq.mgmt_class;
728 req.mgmt_class_version = ureq.mgmt_class_version;
729 memcpy(req.oui, ureq.oui, sizeof req.oui);
731 if (compat_method_mask) {
732 u32 *umm = (u32 *) ureq.method_mask;
735 for (i = 0; i < BITS_TO_LONGS(IB_MGMT_MAX_METHODS); ++i)
737 umm[i * 2] | ((u64) umm[i * 2 + 1] << 32);
739 memcpy(req.method_mask, ureq.method_mask,
740 sizeof req.method_mask);
743 agent = ib_register_mad_agent(file->port->ib_dev, file->port->port_num,
744 ureq.qpn ? IB_QPT_GSI : IB_QPT_SMI,
745 ureq.mgmt_class ? &req : NULL,
747 send_handler, recv_handler, file, 0);
749 ret = PTR_ERR(agent);
754 if (put_user(agent_id,
755 (u32 __user *) (arg + offsetof(struct ib_user_mad_reg_req, id)))) {
760 if (!file->already_used) {
761 file->already_used = 1;
762 if (!file->use_pkey_index) {
763 dev_warn(&file->port->dev,
764 "process %s did not enable P_Key index support.\n",
766 dev_warn(&file->port->dev,
767 " Documentation/infiniband/user_mad.rst has info on the new ABI.\n");
771 file->agent[agent_id] = agent;
775 mutex_unlock(&file->mutex);
778 ib_unregister_mad_agent(agent);
780 mutex_unlock(&file->port->file_mutex);
785 static int ib_umad_reg_agent2(struct ib_umad_file *file, void __user *arg)
787 struct ib_user_mad_reg_req2 ureq;
788 struct ib_mad_reg_req req;
789 struct ib_mad_agent *agent = NULL;
793 mutex_lock(&file->port->file_mutex);
794 mutex_lock(&file->mutex);
796 if (!file->port->ib_dev) {
797 dev_notice(&file->port->dev, "%s: invalid device\n", __func__);
802 if (copy_from_user(&ureq, arg, sizeof(ureq))) {
807 if (ureq.qpn != 0 && ureq.qpn != 1) {
808 dev_notice(&file->port->dev, "%s: invalid QPN %u specified\n",
814 if (ureq.flags & ~IB_USER_MAD_REG_FLAGS_CAP) {
815 dev_notice(&file->port->dev,
816 "%s failed: invalid registration flags specified 0x%x; supported 0x%x\n",
817 __func__, ureq.flags, IB_USER_MAD_REG_FLAGS_CAP);
820 if (put_user((u32)IB_USER_MAD_REG_FLAGS_CAP,
821 (u32 __user *) (arg + offsetof(struct
822 ib_user_mad_reg_req2, flags))))
828 for (agent_id = 0; agent_id < IB_UMAD_MAX_AGENTS; ++agent_id)
829 if (!__get_agent(file, agent_id))
832 dev_notice(&file->port->dev, "%s: Max Agents (%u) reached\n", __func__,
838 if (ureq.mgmt_class) {
839 memset(&req, 0, sizeof(req));
840 req.mgmt_class = ureq.mgmt_class;
841 req.mgmt_class_version = ureq.mgmt_class_version;
842 if (ureq.oui & 0xff000000) {
843 dev_notice(&file->port->dev,
844 "%s failed: oui invalid 0x%08x\n", __func__,
849 req.oui[2] = ureq.oui & 0x0000ff;
850 req.oui[1] = (ureq.oui & 0x00ff00) >> 8;
851 req.oui[0] = (ureq.oui & 0xff0000) >> 16;
852 memcpy(req.method_mask, ureq.method_mask,
853 sizeof(req.method_mask));
856 agent = ib_register_mad_agent(file->port->ib_dev, file->port->port_num,
857 ureq.qpn ? IB_QPT_GSI : IB_QPT_SMI,
858 ureq.mgmt_class ? &req : NULL,
860 send_handler, recv_handler, file,
863 ret = PTR_ERR(agent);
868 if (put_user(agent_id,
870 offsetof(struct ib_user_mad_reg_req2, id)))) {
875 if (!file->already_used) {
876 file->already_used = 1;
877 file->use_pkey_index = 1;
880 file->agent[agent_id] = agent;
884 mutex_unlock(&file->mutex);
887 ib_unregister_mad_agent(agent);
889 mutex_unlock(&file->port->file_mutex);
895 static int ib_umad_unreg_agent(struct ib_umad_file *file, u32 __user *arg)
897 struct ib_mad_agent *agent = NULL;
901 if (get_user(id, arg))
903 if (id >= IB_UMAD_MAX_AGENTS)
906 mutex_lock(&file->port->file_mutex);
907 mutex_lock(&file->mutex);
909 id = array_index_nospec(id, IB_UMAD_MAX_AGENTS);
910 if (!__get_agent(file, id)) {
915 agent = file->agent[id];
916 file->agent[id] = NULL;
919 mutex_unlock(&file->mutex);
922 ib_unregister_mad_agent(agent);
924 mutex_unlock(&file->port->file_mutex);
929 static long ib_umad_enable_pkey(struct ib_umad_file *file)
933 mutex_lock(&file->mutex);
934 if (file->already_used)
937 file->use_pkey_index = 1;
938 mutex_unlock(&file->mutex);
943 static long ib_umad_ioctl(struct file *filp, unsigned int cmd,
947 case IB_USER_MAD_REGISTER_AGENT:
948 return ib_umad_reg_agent(filp->private_data, (void __user *) arg, 0);
949 case IB_USER_MAD_UNREGISTER_AGENT:
950 return ib_umad_unreg_agent(filp->private_data, (__u32 __user *) arg);
951 case IB_USER_MAD_ENABLE_PKEY:
952 return ib_umad_enable_pkey(filp->private_data);
953 case IB_USER_MAD_REGISTER_AGENT2:
954 return ib_umad_reg_agent2(filp->private_data, (void __user *) arg);
961 static long ib_umad_compat_ioctl(struct file *filp, unsigned int cmd,
965 case IB_USER_MAD_REGISTER_AGENT:
966 return ib_umad_reg_agent(filp->private_data, compat_ptr(arg), 1);
967 case IB_USER_MAD_UNREGISTER_AGENT:
968 return ib_umad_unreg_agent(filp->private_data, compat_ptr(arg));
969 case IB_USER_MAD_ENABLE_PKEY:
970 return ib_umad_enable_pkey(filp->private_data);
971 case IB_USER_MAD_REGISTER_AGENT2:
972 return ib_umad_reg_agent2(filp->private_data, compat_ptr(arg));
980 * ib_umad_open() does not need the BKL:
982 * - the ib_umad_port structures are properly reference counted, and
983 * everything else is purely local to the file being created, so
984 * races against other open calls are not a problem;
985 * - the ioctl method does not affect any global state outside of the
986 * file structure being operated on;
988 static int ib_umad_open(struct inode *inode, struct file *filp)
990 struct ib_umad_port *port;
991 struct ib_umad_file *file;
994 port = container_of(inode->i_cdev, struct ib_umad_port, cdev);
996 mutex_lock(&port->file_mutex);
1003 if (!rdma_dev_access_netns(port->ib_dev, current->nsproxy->net_ns)) {
1008 file = kzalloc(sizeof(*file), GFP_KERNEL);
1014 mutex_init(&file->mutex);
1015 spin_lock_init(&file->send_lock);
1016 INIT_LIST_HEAD(&file->recv_list);
1017 INIT_LIST_HEAD(&file->send_list);
1018 init_waitqueue_head(&file->recv_wait);
1021 filp->private_data = file;
1023 list_add_tail(&file->port_list, &port->file_list);
1025 stream_open(inode, filp);
1027 mutex_unlock(&port->file_mutex);
1031 static int ib_umad_close(struct inode *inode, struct file *filp)
1033 struct ib_umad_file *file = filp->private_data;
1034 struct ib_umad_packet *packet, *tmp;
1038 mutex_lock(&file->port->file_mutex);
1039 mutex_lock(&file->mutex);
1041 already_dead = file->agents_dead;
1042 file->agents_dead = 1;
1044 list_for_each_entry_safe(packet, tmp, &file->recv_list, list) {
1045 if (packet->recv_wc)
1046 ib_free_recv_mad(packet->recv_wc);
1050 list_del(&file->port_list);
1052 mutex_unlock(&file->mutex);
1055 for (i = 0; i < IB_UMAD_MAX_AGENTS; ++i)
1057 ib_unregister_mad_agent(file->agent[i]);
1059 mutex_unlock(&file->port->file_mutex);
1060 mutex_destroy(&file->mutex);
1065 static const struct file_operations umad_fops = {
1066 .owner = THIS_MODULE,
1067 .read = ib_umad_read,
1068 .write = ib_umad_write,
1069 .poll = ib_umad_poll,
1070 .unlocked_ioctl = ib_umad_ioctl,
1071 #ifdef CONFIG_COMPAT
1072 .compat_ioctl = ib_umad_compat_ioctl,
1074 .open = ib_umad_open,
1075 .release = ib_umad_close,
1076 .llseek = no_llseek,
1079 static int ib_umad_sm_open(struct inode *inode, struct file *filp)
1081 struct ib_umad_port *port;
1082 struct ib_port_modify props = {
1083 .set_port_cap_mask = IB_PORT_SM
1087 port = container_of(inode->i_cdev, struct ib_umad_port, sm_cdev);
1089 if (filp->f_flags & O_NONBLOCK) {
1090 if (down_trylock(&port->sm_sem)) {
1095 if (down_interruptible(&port->sm_sem)) {
1101 if (!rdma_dev_access_netns(port->ib_dev, current->nsproxy->net_ns)) {
1106 ret = ib_modify_port(port->ib_dev, port->port_num, 0, &props);
1110 filp->private_data = port;
1112 nonseekable_open(inode, filp);
1122 static int ib_umad_sm_close(struct inode *inode, struct file *filp)
1124 struct ib_umad_port *port = filp->private_data;
1125 struct ib_port_modify props = {
1126 .clr_port_cap_mask = IB_PORT_SM
1130 mutex_lock(&port->file_mutex);
1132 ret = ib_modify_port(port->ib_dev, port->port_num, 0, &props);
1133 mutex_unlock(&port->file_mutex);
1140 static const struct file_operations umad_sm_fops = {
1141 .owner = THIS_MODULE,
1142 .open = ib_umad_sm_open,
1143 .release = ib_umad_sm_close,
1144 .llseek = no_llseek,
1147 static struct ib_umad_port *get_port(struct ib_device *ibdev,
1148 struct ib_umad_device *umad_dev,
1152 return ERR_PTR(-EOPNOTSUPP);
1153 if (!rdma_is_port_valid(ibdev, port))
1154 return ERR_PTR(-EINVAL);
1155 if (!rdma_cap_ib_mad(ibdev, port))
1156 return ERR_PTR(-EOPNOTSUPP);
1158 return &umad_dev->ports[port - rdma_start_port(ibdev)];
1161 static int ib_umad_get_nl_info(struct ib_device *ibdev, void *client_data,
1162 struct ib_client_nl_info *res)
1164 struct ib_umad_port *port = get_port(ibdev, client_data, res->port);
1167 return PTR_ERR(port);
1169 res->abi = IB_USER_MAD_ABI_VERSION;
1170 res->cdev = &port->dev;
1174 static struct ib_client umad_client = {
1176 .add = ib_umad_add_one,
1177 .remove = ib_umad_remove_one,
1178 .get_nl_info = ib_umad_get_nl_info,
1180 MODULE_ALIAS_RDMA_CLIENT("umad");
1182 static int ib_issm_get_nl_info(struct ib_device *ibdev, void *client_data,
1183 struct ib_client_nl_info *res)
1185 struct ib_umad_port *port = get_port(ibdev, client_data, res->port);
1188 return PTR_ERR(port);
1190 res->abi = IB_USER_MAD_ABI_VERSION;
1191 res->cdev = &port->sm_dev;
1195 static struct ib_client issm_client = {
1197 .get_nl_info = ib_issm_get_nl_info,
1199 MODULE_ALIAS_RDMA_CLIENT("issm");
1201 static ssize_t ibdev_show(struct device *dev, struct device_attribute *attr,
1204 struct ib_umad_port *port = dev_get_drvdata(dev);
1209 return sysfs_emit(buf, "%s\n", dev_name(&port->ib_dev->dev));
1211 static DEVICE_ATTR_RO(ibdev);
1213 static ssize_t port_show(struct device *dev, struct device_attribute *attr,
1216 struct ib_umad_port *port = dev_get_drvdata(dev);
1221 return sysfs_emit(buf, "%d\n", port->port_num);
1223 static DEVICE_ATTR_RO(port);
1225 static struct attribute *umad_class_dev_attrs[] = {
1226 &dev_attr_ibdev.attr,
1227 &dev_attr_port.attr,
1230 ATTRIBUTE_GROUPS(umad_class_dev);
1232 static char *umad_devnode(const struct device *dev, umode_t *mode)
1234 return kasprintf(GFP_KERNEL, "infiniband/%s", dev_name(dev));
1237 static ssize_t abi_version_show(const struct class *class,
1238 const struct class_attribute *attr, char *buf)
1240 return sysfs_emit(buf, "%d\n", IB_USER_MAD_ABI_VERSION);
1242 static CLASS_ATTR_RO(abi_version);
1244 static struct attribute *umad_class_attrs[] = {
1245 &class_attr_abi_version.attr,
1248 ATTRIBUTE_GROUPS(umad_class);
1250 static struct class umad_class = {
1251 .name = "infiniband_mad",
1252 .devnode = umad_devnode,
1253 .class_groups = umad_class_groups,
1254 .dev_groups = umad_class_dev_groups,
1257 static void ib_umad_release_port(struct device *device)
1259 struct ib_umad_port *port = dev_get_drvdata(device);
1260 struct ib_umad_device *umad_dev = port->umad_dev;
1262 ib_umad_dev_put(umad_dev);
1265 static void ib_umad_init_port_dev(struct device *dev,
1266 struct ib_umad_port *port,
1267 const struct ib_device *device)
1269 device_initialize(dev);
1270 ib_umad_dev_get(port->umad_dev);
1271 dev->class = &umad_class;
1272 dev->parent = device->dev.parent;
1273 dev_set_drvdata(dev, port);
1274 dev->release = ib_umad_release_port;
1277 static int ib_umad_init_port(struct ib_device *device, int port_num,
1278 struct ib_umad_device *umad_dev,
1279 struct ib_umad_port *port)
1286 devnum = ida_alloc_max(&umad_ida, IB_UMAD_MAX_PORTS - 1, GFP_KERNEL);
1289 port->dev_num = devnum;
1290 if (devnum >= IB_UMAD_NUM_FIXED_MINOR) {
1291 base_umad = dynamic_umad_dev + devnum - IB_UMAD_NUM_FIXED_MINOR;
1292 base_issm = dynamic_issm_dev + devnum - IB_UMAD_NUM_FIXED_MINOR;
1294 base_umad = devnum + base_umad_dev;
1295 base_issm = devnum + base_issm_dev;
1298 port->ib_dev = device;
1299 port->umad_dev = umad_dev;
1300 port->port_num = port_num;
1301 sema_init(&port->sm_sem, 1);
1302 mutex_init(&port->file_mutex);
1303 INIT_LIST_HEAD(&port->file_list);
1305 ib_umad_init_port_dev(&port->dev, port, device);
1306 port->dev.devt = base_umad;
1307 dev_set_name(&port->dev, "umad%d", port->dev_num);
1308 cdev_init(&port->cdev, &umad_fops);
1309 port->cdev.owner = THIS_MODULE;
1311 ret = cdev_device_add(&port->cdev, &port->dev);
1315 ib_umad_init_port_dev(&port->sm_dev, port, device);
1316 port->sm_dev.devt = base_issm;
1317 dev_set_name(&port->sm_dev, "issm%d", port->dev_num);
1318 cdev_init(&port->sm_cdev, &umad_sm_fops);
1319 port->sm_cdev.owner = THIS_MODULE;
1321 ret = cdev_device_add(&port->sm_cdev, &port->sm_dev);
1328 put_device(&port->sm_dev);
1329 cdev_device_del(&port->cdev, &port->dev);
1331 put_device(&port->dev);
1332 ida_free(&umad_ida, devnum);
1336 static void ib_umad_kill_port(struct ib_umad_port *port)
1338 struct ib_umad_file *file;
1341 cdev_device_del(&port->sm_cdev, &port->sm_dev);
1342 cdev_device_del(&port->cdev, &port->dev);
1344 mutex_lock(&port->file_mutex);
1346 /* Mark ib_dev NULL and block ioctl or other file ops to progress
1349 port->ib_dev = NULL;
1351 list_for_each_entry(file, &port->file_list, port_list) {
1352 mutex_lock(&file->mutex);
1353 file->agents_dead = 1;
1354 wake_up_interruptible(&file->recv_wait);
1355 mutex_unlock(&file->mutex);
1357 for (id = 0; id < IB_UMAD_MAX_AGENTS; ++id)
1358 if (file->agent[id])
1359 ib_unregister_mad_agent(file->agent[id]);
1362 mutex_unlock(&port->file_mutex);
1364 ida_free(&umad_ida, port->dev_num);
1366 /* balances device_initialize() */
1367 put_device(&port->sm_dev);
1368 put_device(&port->dev);
1371 static int ib_umad_add_one(struct ib_device *device)
1373 struct ib_umad_device *umad_dev;
1378 s = rdma_start_port(device);
1379 e = rdma_end_port(device);
1381 umad_dev = kzalloc(struct_size(umad_dev, ports,
1382 size_add(size_sub(e, s), 1)),
1387 kref_init(&umad_dev->kref);
1388 for (i = s; i <= e; ++i) {
1389 if (!rdma_cap_ib_mad(device, i))
1392 ret = ib_umad_init_port(device, i, umad_dev,
1393 &umad_dev->ports[i - s]);
1405 ib_set_client_data(device, &umad_client, umad_dev);
1411 if (!rdma_cap_ib_mad(device, i))
1414 ib_umad_kill_port(&umad_dev->ports[i - s]);
1417 /* balances kref_init */
1418 ib_umad_dev_put(umad_dev);
1422 static void ib_umad_remove_one(struct ib_device *device, void *client_data)
1424 struct ib_umad_device *umad_dev = client_data;
1427 rdma_for_each_port (device, i) {
1428 if (rdma_cap_ib_mad(device, i))
1430 &umad_dev->ports[i - rdma_start_port(device)]);
1432 /* balances kref_init() */
1433 ib_umad_dev_put(umad_dev);
1436 static int __init ib_umad_init(void)
1440 ret = register_chrdev_region(base_umad_dev,
1441 IB_UMAD_NUM_FIXED_MINOR * 2,
1444 pr_err("couldn't register device number\n");
1448 ret = alloc_chrdev_region(&dynamic_umad_dev, 0,
1449 IB_UMAD_NUM_DYNAMIC_MINOR * 2,
1452 pr_err("couldn't register dynamic device number\n");
1455 dynamic_issm_dev = dynamic_umad_dev + IB_UMAD_NUM_DYNAMIC_MINOR;
1457 ret = class_register(&umad_class);
1459 pr_err("couldn't create class infiniband_mad\n");
1463 ret = ib_register_client(&umad_client);
1467 ret = ib_register_client(&issm_client);
1474 ib_unregister_client(&umad_client);
1476 class_unregister(&umad_class);
1479 unregister_chrdev_region(dynamic_umad_dev,
1480 IB_UMAD_NUM_DYNAMIC_MINOR * 2);
1483 unregister_chrdev_region(base_umad_dev,
1484 IB_UMAD_NUM_FIXED_MINOR * 2);
1490 static void __exit ib_umad_cleanup(void)
1492 ib_unregister_client(&issm_client);
1493 ib_unregister_client(&umad_client);
1494 class_unregister(&umad_class);
1495 unregister_chrdev_region(base_umad_dev,
1496 IB_UMAD_NUM_FIXED_MINOR * 2);
1497 unregister_chrdev_region(dynamic_umad_dev,
1498 IB_UMAD_NUM_DYNAMIC_MINOR * 2);
1501 module_init(ib_umad_init);
1502 module_exit(ib_umad_cleanup);