locking/refcounts: Include fewer headers in <linux/refcount.h>
[linux/fpc-iii.git] / drivers / infiniband / core / user_mad.c
blobbb98c9e4a7fd25fe94370f2eda5486e4d64bdb66
1 /*
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
15 * conditions are met:
17 * - Redistributions of source code must retain the above
18 * copyright notice, this list of conditions and the following
19 * disclaimer.
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
33 * SOFTWARE.
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>
42 #include <linux/fs.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>
53 #include <linux/uaccess.h>
55 #include <rdma/ib_mad.h>
56 #include <rdma/ib_user_mad.h>
58 #include "core_priv.h"
60 MODULE_AUTHOR("Roland Dreier");
61 MODULE_DESCRIPTION("InfiniBand userspace MAD packet access");
62 MODULE_LICENSE("Dual BSD/GPL");
64 enum {
65 IB_UMAD_MAX_PORTS = RDMA_MAX_PORTS,
66 IB_UMAD_MAX_AGENTS = 32,
68 IB_UMAD_MAJOR = 231,
69 IB_UMAD_MINOR_BASE = 0,
70 IB_UMAD_NUM_FIXED_MINOR = 64,
71 IB_UMAD_NUM_DYNAMIC_MINOR = IB_UMAD_MAX_PORTS - IB_UMAD_NUM_FIXED_MINOR,
72 IB_ISSM_MINOR_BASE = IB_UMAD_NUM_FIXED_MINOR,
76 * Our lifetime rules for these structs are the following:
77 * device special file is opened, we take a reference on the
78 * ib_umad_port's struct ib_umad_device. We drop these
79 * references in the corresponding close().
81 * In addition to references coming from open character devices, there
82 * is one more reference to each ib_umad_device representing the
83 * module's reference taken when allocating the ib_umad_device in
84 * ib_umad_add_one().
86 * When destroying an ib_umad_device, we drop the module's reference.
89 struct ib_umad_port {
90 struct cdev cdev;
91 struct device *dev;
93 struct cdev sm_cdev;
94 struct device *sm_dev;
95 struct semaphore sm_sem;
97 struct mutex file_mutex;
98 struct list_head file_list;
100 struct ib_device *ib_dev;
101 struct ib_umad_device *umad_dev;
102 int dev_num;
103 u8 port_num;
106 struct ib_umad_device {
107 struct kobject kobj;
108 struct ib_umad_port port[0];
111 struct ib_umad_file {
112 struct mutex mutex;
113 struct ib_umad_port *port;
114 struct list_head recv_list;
115 struct list_head send_list;
116 struct list_head port_list;
117 spinlock_t send_lock;
118 wait_queue_head_t recv_wait;
119 struct ib_mad_agent *agent[IB_UMAD_MAX_AGENTS];
120 int agents_dead;
121 u8 use_pkey_index;
122 u8 already_used;
125 struct ib_umad_packet {
126 struct ib_mad_send_buf *msg;
127 struct ib_mad_recv_wc *recv_wc;
128 struct list_head list;
129 int length;
130 struct ib_user_mad mad;
133 static struct class *umad_class;
135 static const dev_t base_umad_dev = MKDEV(IB_UMAD_MAJOR, IB_UMAD_MINOR_BASE);
136 static const dev_t base_issm_dev = MKDEV(IB_UMAD_MAJOR, IB_UMAD_MINOR_BASE) +
137 IB_UMAD_NUM_FIXED_MINOR;
138 static dev_t dynamic_umad_dev;
139 static dev_t dynamic_issm_dev;
141 static DECLARE_BITMAP(dev_map, IB_UMAD_MAX_PORTS);
143 static void ib_umad_add_one(struct ib_device *device);
144 static void ib_umad_remove_one(struct ib_device *device, void *client_data);
146 static void ib_umad_release_dev(struct kobject *kobj)
148 struct ib_umad_device *dev =
149 container_of(kobj, struct ib_umad_device, kobj);
151 kfree(dev);
154 static struct kobj_type ib_umad_dev_ktype = {
155 .release = ib_umad_release_dev,
158 static int hdr_size(struct ib_umad_file *file)
160 return file->use_pkey_index ? sizeof (struct ib_user_mad_hdr) :
161 sizeof (struct ib_user_mad_hdr_old);
164 /* caller must hold file->mutex */
165 static struct ib_mad_agent *__get_agent(struct ib_umad_file *file, int id)
167 return file->agents_dead ? NULL : file->agent[id];
170 static int queue_packet(struct ib_umad_file *file,
171 struct ib_mad_agent *agent,
172 struct ib_umad_packet *packet)
174 int ret = 1;
176 mutex_lock(&file->mutex);
178 for (packet->mad.hdr.id = 0;
179 packet->mad.hdr.id < IB_UMAD_MAX_AGENTS;
180 packet->mad.hdr.id++)
181 if (agent == __get_agent(file, packet->mad.hdr.id)) {
182 list_add_tail(&packet->list, &file->recv_list);
183 wake_up_interruptible(&file->recv_wait);
184 ret = 0;
185 break;
188 mutex_unlock(&file->mutex);
190 return ret;
193 static void dequeue_send(struct ib_umad_file *file,
194 struct ib_umad_packet *packet)
196 spin_lock_irq(&file->send_lock);
197 list_del(&packet->list);
198 spin_unlock_irq(&file->send_lock);
201 static void send_handler(struct ib_mad_agent *agent,
202 struct ib_mad_send_wc *send_wc)
204 struct ib_umad_file *file = agent->context;
205 struct ib_umad_packet *packet = send_wc->send_buf->context[0];
207 dequeue_send(file, packet);
208 rdma_destroy_ah(packet->msg->ah);
209 ib_free_send_mad(packet->msg);
211 if (send_wc->status == IB_WC_RESP_TIMEOUT_ERR) {
212 packet->length = IB_MGMT_MAD_HDR;
213 packet->mad.hdr.status = ETIMEDOUT;
214 if (!queue_packet(file, agent, packet))
215 return;
217 kfree(packet);
220 static void recv_handler(struct ib_mad_agent *agent,
221 struct ib_mad_send_buf *send_buf,
222 struct ib_mad_recv_wc *mad_recv_wc)
224 struct ib_umad_file *file = agent->context;
225 struct ib_umad_packet *packet;
227 if (mad_recv_wc->wc->status != IB_WC_SUCCESS)
228 goto err1;
230 packet = kzalloc(sizeof *packet, GFP_KERNEL);
231 if (!packet)
232 goto err1;
234 packet->length = mad_recv_wc->mad_len;
235 packet->recv_wc = mad_recv_wc;
237 packet->mad.hdr.status = 0;
238 packet->mad.hdr.length = hdr_size(file) + mad_recv_wc->mad_len;
239 packet->mad.hdr.qpn = cpu_to_be32(mad_recv_wc->wc->src_qp);
241 * On OPA devices it is okay to lose the upper 16 bits of LID as this
242 * information is obtained elsewhere. Mask off the upper 16 bits.
244 if (rdma_cap_opa_mad(agent->device, agent->port_num))
245 packet->mad.hdr.lid = ib_lid_be16(0xFFFF &
246 mad_recv_wc->wc->slid);
247 else
248 packet->mad.hdr.lid = ib_lid_be16(mad_recv_wc->wc->slid);
249 packet->mad.hdr.sl = mad_recv_wc->wc->sl;
250 packet->mad.hdr.path_bits = mad_recv_wc->wc->dlid_path_bits;
251 packet->mad.hdr.pkey_index = mad_recv_wc->wc->pkey_index;
252 packet->mad.hdr.grh_present = !!(mad_recv_wc->wc->wc_flags & IB_WC_GRH);
253 if (packet->mad.hdr.grh_present) {
254 struct rdma_ah_attr ah_attr;
255 const struct ib_global_route *grh;
256 int ret;
258 ret = ib_init_ah_attr_from_wc(agent->device, agent->port_num,
259 mad_recv_wc->wc,
260 mad_recv_wc->recv_buf.grh,
261 &ah_attr);
262 if (ret)
263 goto err2;
265 grh = rdma_ah_read_grh(&ah_attr);
266 packet->mad.hdr.gid_index = grh->sgid_index;
267 packet->mad.hdr.hop_limit = grh->hop_limit;
268 packet->mad.hdr.traffic_class = grh->traffic_class;
269 memcpy(packet->mad.hdr.gid, &grh->dgid, 16);
270 packet->mad.hdr.flow_label = cpu_to_be32(grh->flow_label);
273 if (queue_packet(file, agent, packet))
274 goto err2;
275 return;
277 err2:
278 kfree(packet);
279 err1:
280 ib_free_recv_mad(mad_recv_wc);
283 static ssize_t copy_recv_mad(struct ib_umad_file *file, char __user *buf,
284 struct ib_umad_packet *packet, size_t count)
286 struct ib_mad_recv_buf *recv_buf;
287 int left, seg_payload, offset, max_seg_payload;
288 size_t seg_size;
290 recv_buf = &packet->recv_wc->recv_buf;
291 seg_size = packet->recv_wc->mad_seg_size;
293 /* We need enough room to copy the first (or only) MAD segment. */
294 if ((packet->length <= seg_size &&
295 count < hdr_size(file) + packet->length) ||
296 (packet->length > seg_size &&
297 count < hdr_size(file) + seg_size))
298 return -EINVAL;
300 if (copy_to_user(buf, &packet->mad, hdr_size(file)))
301 return -EFAULT;
303 buf += hdr_size(file);
304 seg_payload = min_t(int, packet->length, seg_size);
305 if (copy_to_user(buf, recv_buf->mad, seg_payload))
306 return -EFAULT;
308 if (seg_payload < packet->length) {
310 * Multipacket RMPP MAD message. Copy remainder of message.
311 * Note that last segment may have a shorter payload.
313 if (count < hdr_size(file) + packet->length) {
315 * The buffer is too small, return the first RMPP segment,
316 * which includes the RMPP message length.
318 return -ENOSPC;
320 offset = ib_get_mad_data_offset(recv_buf->mad->mad_hdr.mgmt_class);
321 max_seg_payload = seg_size - offset;
323 for (left = packet->length - seg_payload, buf += seg_payload;
324 left; left -= seg_payload, buf += seg_payload) {
325 recv_buf = container_of(recv_buf->list.next,
326 struct ib_mad_recv_buf, list);
327 seg_payload = min(left, max_seg_payload);
328 if (copy_to_user(buf, ((void *) recv_buf->mad) + offset,
329 seg_payload))
330 return -EFAULT;
333 return hdr_size(file) + packet->length;
336 static ssize_t copy_send_mad(struct ib_umad_file *file, char __user *buf,
337 struct ib_umad_packet *packet, size_t count)
339 ssize_t size = hdr_size(file) + packet->length;
341 if (count < size)
342 return -EINVAL;
344 if (copy_to_user(buf, &packet->mad, hdr_size(file)))
345 return -EFAULT;
347 buf += hdr_size(file);
349 if (copy_to_user(buf, packet->mad.data, packet->length))
350 return -EFAULT;
352 return size;
355 static ssize_t ib_umad_read(struct file *filp, char __user *buf,
356 size_t count, loff_t *pos)
358 struct ib_umad_file *file = filp->private_data;
359 struct ib_umad_packet *packet;
360 ssize_t ret;
362 if (count < hdr_size(file))
363 return -EINVAL;
365 mutex_lock(&file->mutex);
367 while (list_empty(&file->recv_list)) {
368 mutex_unlock(&file->mutex);
370 if (filp->f_flags & O_NONBLOCK)
371 return -EAGAIN;
373 if (wait_event_interruptible(file->recv_wait,
374 !list_empty(&file->recv_list)))
375 return -ERESTARTSYS;
377 mutex_lock(&file->mutex);
380 packet = list_entry(file->recv_list.next, struct ib_umad_packet, list);
381 list_del(&packet->list);
383 mutex_unlock(&file->mutex);
385 if (packet->recv_wc)
386 ret = copy_recv_mad(file, buf, packet, count);
387 else
388 ret = copy_send_mad(file, buf, packet, count);
390 if (ret < 0) {
391 /* Requeue packet */
392 mutex_lock(&file->mutex);
393 list_add(&packet->list, &file->recv_list);
394 mutex_unlock(&file->mutex);
395 } else {
396 if (packet->recv_wc)
397 ib_free_recv_mad(packet->recv_wc);
398 kfree(packet);
400 return ret;
403 static int copy_rmpp_mad(struct ib_mad_send_buf *msg, const char __user *buf)
405 int left, seg;
407 /* Copy class specific header */
408 if ((msg->hdr_len > IB_MGMT_RMPP_HDR) &&
409 copy_from_user(msg->mad + IB_MGMT_RMPP_HDR, buf + IB_MGMT_RMPP_HDR,
410 msg->hdr_len - IB_MGMT_RMPP_HDR))
411 return -EFAULT;
413 /* All headers are in place. Copy data segments. */
414 for (seg = 1, left = msg->data_len, buf += msg->hdr_len; left > 0;
415 seg++, left -= msg->seg_size, buf += msg->seg_size) {
416 if (copy_from_user(ib_get_rmpp_segment(msg, seg), buf,
417 min(left, msg->seg_size)))
418 return -EFAULT;
420 return 0;
423 static int same_destination(struct ib_user_mad_hdr *hdr1,
424 struct ib_user_mad_hdr *hdr2)
426 if (!hdr1->grh_present && !hdr2->grh_present)
427 return (hdr1->lid == hdr2->lid);
429 if (hdr1->grh_present && hdr2->grh_present)
430 return !memcmp(hdr1->gid, hdr2->gid, 16);
432 return 0;
435 static int is_duplicate(struct ib_umad_file *file,
436 struct ib_umad_packet *packet)
438 struct ib_umad_packet *sent_packet;
439 struct ib_mad_hdr *sent_hdr, *hdr;
441 hdr = (struct ib_mad_hdr *) packet->mad.data;
442 list_for_each_entry(sent_packet, &file->send_list, list) {
443 sent_hdr = (struct ib_mad_hdr *) sent_packet->mad.data;
445 if ((hdr->tid != sent_hdr->tid) ||
446 (hdr->mgmt_class != sent_hdr->mgmt_class))
447 continue;
450 * No need to be overly clever here. If two new operations have
451 * the same TID, reject the second as a duplicate. This is more
452 * restrictive than required by the spec.
454 if (!ib_response_mad(hdr)) {
455 if (!ib_response_mad(sent_hdr))
456 return 1;
457 continue;
458 } else if (!ib_response_mad(sent_hdr))
459 continue;
461 if (same_destination(&packet->mad.hdr, &sent_packet->mad.hdr))
462 return 1;
465 return 0;
468 static ssize_t ib_umad_write(struct file *filp, const char __user *buf,
469 size_t count, loff_t *pos)
471 struct ib_umad_file *file = filp->private_data;
472 struct ib_umad_packet *packet;
473 struct ib_mad_agent *agent;
474 struct rdma_ah_attr ah_attr;
475 struct ib_ah *ah;
476 struct ib_rmpp_mad *rmpp_mad;
477 __be64 *tid;
478 int ret, data_len, hdr_len, copy_offset, rmpp_active;
479 u8 base_version;
481 if (count < hdr_size(file) + IB_MGMT_RMPP_HDR)
482 return -EINVAL;
484 packet = kzalloc(sizeof *packet + IB_MGMT_RMPP_HDR, GFP_KERNEL);
485 if (!packet)
486 return -ENOMEM;
488 if (copy_from_user(&packet->mad, buf, hdr_size(file))) {
489 ret = -EFAULT;
490 goto err;
493 if (packet->mad.hdr.id >= IB_UMAD_MAX_AGENTS) {
494 ret = -EINVAL;
495 goto err;
498 buf += hdr_size(file);
500 if (copy_from_user(packet->mad.data, buf, IB_MGMT_RMPP_HDR)) {
501 ret = -EFAULT;
502 goto err;
505 mutex_lock(&file->mutex);
507 agent = __get_agent(file, packet->mad.hdr.id);
508 if (!agent) {
509 ret = -EINVAL;
510 goto err_up;
513 memset(&ah_attr, 0, sizeof ah_attr);
514 ah_attr.type = rdma_ah_find_type(agent->device,
515 file->port->port_num);
516 rdma_ah_set_dlid(&ah_attr, be16_to_cpu(packet->mad.hdr.lid));
517 rdma_ah_set_sl(&ah_attr, packet->mad.hdr.sl);
518 rdma_ah_set_path_bits(&ah_attr, packet->mad.hdr.path_bits);
519 rdma_ah_set_port_num(&ah_attr, file->port->port_num);
520 if (packet->mad.hdr.grh_present) {
521 rdma_ah_set_grh(&ah_attr, NULL,
522 be32_to_cpu(packet->mad.hdr.flow_label),
523 packet->mad.hdr.gid_index,
524 packet->mad.hdr.hop_limit,
525 packet->mad.hdr.traffic_class);
526 rdma_ah_set_dgid_raw(&ah_attr, packet->mad.hdr.gid);
529 ah = rdma_create_user_ah(agent->qp->pd, &ah_attr, NULL);
530 if (IS_ERR(ah)) {
531 ret = PTR_ERR(ah);
532 goto err_up;
535 rmpp_mad = (struct ib_rmpp_mad *) packet->mad.data;
536 hdr_len = ib_get_mad_data_offset(rmpp_mad->mad_hdr.mgmt_class);
538 if (ib_is_mad_class_rmpp(rmpp_mad->mad_hdr.mgmt_class)
539 && ib_mad_kernel_rmpp_agent(agent)) {
540 copy_offset = IB_MGMT_RMPP_HDR;
541 rmpp_active = ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) &
542 IB_MGMT_RMPP_FLAG_ACTIVE;
543 } else {
544 copy_offset = IB_MGMT_MAD_HDR;
545 rmpp_active = 0;
548 base_version = ((struct ib_mad_hdr *)&packet->mad.data)->base_version;
549 data_len = count - hdr_size(file) - hdr_len;
550 packet->msg = ib_create_send_mad(agent,
551 be32_to_cpu(packet->mad.hdr.qpn),
552 packet->mad.hdr.pkey_index, rmpp_active,
553 hdr_len, data_len, GFP_KERNEL,
554 base_version);
555 if (IS_ERR(packet->msg)) {
556 ret = PTR_ERR(packet->msg);
557 goto err_ah;
560 packet->msg->ah = ah;
561 packet->msg->timeout_ms = packet->mad.hdr.timeout_ms;
562 packet->msg->retries = packet->mad.hdr.retries;
563 packet->msg->context[0] = packet;
565 /* Copy MAD header. Any RMPP header is already in place. */
566 memcpy(packet->msg->mad, packet->mad.data, IB_MGMT_MAD_HDR);
568 if (!rmpp_active) {
569 if (copy_from_user(packet->msg->mad + copy_offset,
570 buf + copy_offset,
571 hdr_len + data_len - copy_offset)) {
572 ret = -EFAULT;
573 goto err_msg;
575 } else {
576 ret = copy_rmpp_mad(packet->msg, buf);
577 if (ret)
578 goto err_msg;
582 * Set the high-order part of the transaction ID to make MADs from
583 * different agents unique, and allow routing responses back to the
584 * original requestor.
586 if (!ib_response_mad(packet->msg->mad)) {
587 tid = &((struct ib_mad_hdr *) packet->msg->mad)->tid;
588 *tid = cpu_to_be64(((u64) agent->hi_tid) << 32 |
589 (be64_to_cpup(tid) & 0xffffffff));
590 rmpp_mad->mad_hdr.tid = *tid;
593 if (!ib_mad_kernel_rmpp_agent(agent)
594 && ib_is_mad_class_rmpp(rmpp_mad->mad_hdr.mgmt_class)
595 && (ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) & IB_MGMT_RMPP_FLAG_ACTIVE)) {
596 spin_lock_irq(&file->send_lock);
597 list_add_tail(&packet->list, &file->send_list);
598 spin_unlock_irq(&file->send_lock);
599 } else {
600 spin_lock_irq(&file->send_lock);
601 ret = is_duplicate(file, packet);
602 if (!ret)
603 list_add_tail(&packet->list, &file->send_list);
604 spin_unlock_irq(&file->send_lock);
605 if (ret) {
606 ret = -EINVAL;
607 goto err_msg;
611 ret = ib_post_send_mad(packet->msg, NULL);
612 if (ret)
613 goto err_send;
615 mutex_unlock(&file->mutex);
616 return count;
618 err_send:
619 dequeue_send(file, packet);
620 err_msg:
621 ib_free_send_mad(packet->msg);
622 err_ah:
623 rdma_destroy_ah(ah);
624 err_up:
625 mutex_unlock(&file->mutex);
626 err:
627 kfree(packet);
628 return ret;
631 static __poll_t ib_umad_poll(struct file *filp, struct poll_table_struct *wait)
633 struct ib_umad_file *file = filp->private_data;
635 /* we will always be able to post a MAD send */
636 __poll_t mask = EPOLLOUT | EPOLLWRNORM;
638 poll_wait(filp, &file->recv_wait, wait);
640 if (!list_empty(&file->recv_list))
641 mask |= EPOLLIN | EPOLLRDNORM;
643 return mask;
646 static int ib_umad_reg_agent(struct ib_umad_file *file, void __user *arg,
647 int compat_method_mask)
649 struct ib_user_mad_reg_req ureq;
650 struct ib_mad_reg_req req;
651 struct ib_mad_agent *agent = NULL;
652 int agent_id;
653 int ret;
655 mutex_lock(&file->port->file_mutex);
656 mutex_lock(&file->mutex);
658 if (!file->port->ib_dev) {
659 dev_notice(file->port->dev,
660 "ib_umad_reg_agent: invalid device\n");
661 ret = -EPIPE;
662 goto out;
665 if (copy_from_user(&ureq, arg, sizeof ureq)) {
666 ret = -EFAULT;
667 goto out;
670 if (ureq.qpn != 0 && ureq.qpn != 1) {
671 dev_notice(file->port->dev,
672 "ib_umad_reg_agent: invalid QPN %d specified\n",
673 ureq.qpn);
674 ret = -EINVAL;
675 goto out;
678 for (agent_id = 0; agent_id < IB_UMAD_MAX_AGENTS; ++agent_id)
679 if (!__get_agent(file, agent_id))
680 goto found;
682 dev_notice(file->port->dev,
683 "ib_umad_reg_agent: Max Agents (%u) reached\n",
684 IB_UMAD_MAX_AGENTS);
685 ret = -ENOMEM;
686 goto out;
688 found:
689 if (ureq.mgmt_class) {
690 memset(&req, 0, sizeof(req));
691 req.mgmt_class = ureq.mgmt_class;
692 req.mgmt_class_version = ureq.mgmt_class_version;
693 memcpy(req.oui, ureq.oui, sizeof req.oui);
695 if (compat_method_mask) {
696 u32 *umm = (u32 *) ureq.method_mask;
697 int i;
699 for (i = 0; i < BITS_TO_LONGS(IB_MGMT_MAX_METHODS); ++i)
700 req.method_mask[i] =
701 umm[i * 2] | ((u64) umm[i * 2 + 1] << 32);
702 } else
703 memcpy(req.method_mask, ureq.method_mask,
704 sizeof req.method_mask);
707 agent = ib_register_mad_agent(file->port->ib_dev, file->port->port_num,
708 ureq.qpn ? IB_QPT_GSI : IB_QPT_SMI,
709 ureq.mgmt_class ? &req : NULL,
710 ureq.rmpp_version,
711 send_handler, recv_handler, file, 0);
712 if (IS_ERR(agent)) {
713 ret = PTR_ERR(agent);
714 agent = NULL;
715 goto out;
718 if (put_user(agent_id,
719 (u32 __user *) (arg + offsetof(struct ib_user_mad_reg_req, id)))) {
720 ret = -EFAULT;
721 goto out;
724 if (!file->already_used) {
725 file->already_used = 1;
726 if (!file->use_pkey_index) {
727 dev_warn(file->port->dev,
728 "process %s did not enable P_Key index support.\n",
729 current->comm);
730 dev_warn(file->port->dev,
731 " Documentation/infiniband/user_mad.txt has info on the new ABI.\n");
735 file->agent[agent_id] = agent;
736 ret = 0;
738 out:
739 mutex_unlock(&file->mutex);
741 if (ret && agent)
742 ib_unregister_mad_agent(agent);
744 mutex_unlock(&file->port->file_mutex);
746 return ret;
749 static int ib_umad_reg_agent2(struct ib_umad_file *file, void __user *arg)
751 struct ib_user_mad_reg_req2 ureq;
752 struct ib_mad_reg_req req;
753 struct ib_mad_agent *agent = NULL;
754 int agent_id;
755 int ret;
757 mutex_lock(&file->port->file_mutex);
758 mutex_lock(&file->mutex);
760 if (!file->port->ib_dev) {
761 dev_notice(file->port->dev,
762 "ib_umad_reg_agent2: invalid device\n");
763 ret = -EPIPE;
764 goto out;
767 if (copy_from_user(&ureq, arg, sizeof(ureq))) {
768 ret = -EFAULT;
769 goto out;
772 if (ureq.qpn != 0 && ureq.qpn != 1) {
773 dev_notice(file->port->dev,
774 "ib_umad_reg_agent2: invalid QPN %d specified\n",
775 ureq.qpn);
776 ret = -EINVAL;
777 goto out;
780 if (ureq.flags & ~IB_USER_MAD_REG_FLAGS_CAP) {
781 dev_notice(file->port->dev,
782 "ib_umad_reg_agent2 failed: invalid registration flags specified 0x%x; supported 0x%x\n",
783 ureq.flags, IB_USER_MAD_REG_FLAGS_CAP);
784 ret = -EINVAL;
786 if (put_user((u32)IB_USER_MAD_REG_FLAGS_CAP,
787 (u32 __user *) (arg + offsetof(struct
788 ib_user_mad_reg_req2, flags))))
789 ret = -EFAULT;
791 goto out;
794 for (agent_id = 0; agent_id < IB_UMAD_MAX_AGENTS; ++agent_id)
795 if (!__get_agent(file, agent_id))
796 goto found;
798 dev_notice(file->port->dev,
799 "ib_umad_reg_agent2: Max Agents (%u) reached\n",
800 IB_UMAD_MAX_AGENTS);
801 ret = -ENOMEM;
802 goto out;
804 found:
805 if (ureq.mgmt_class) {
806 memset(&req, 0, sizeof(req));
807 req.mgmt_class = ureq.mgmt_class;
808 req.mgmt_class_version = ureq.mgmt_class_version;
809 if (ureq.oui & 0xff000000) {
810 dev_notice(file->port->dev,
811 "ib_umad_reg_agent2 failed: oui invalid 0x%08x\n",
812 ureq.oui);
813 ret = -EINVAL;
814 goto out;
816 req.oui[2] = ureq.oui & 0x0000ff;
817 req.oui[1] = (ureq.oui & 0x00ff00) >> 8;
818 req.oui[0] = (ureq.oui & 0xff0000) >> 16;
819 memcpy(req.method_mask, ureq.method_mask,
820 sizeof(req.method_mask));
823 agent = ib_register_mad_agent(file->port->ib_dev, file->port->port_num,
824 ureq.qpn ? IB_QPT_GSI : IB_QPT_SMI,
825 ureq.mgmt_class ? &req : NULL,
826 ureq.rmpp_version,
827 send_handler, recv_handler, file,
828 ureq.flags);
829 if (IS_ERR(agent)) {
830 ret = PTR_ERR(agent);
831 agent = NULL;
832 goto out;
835 if (put_user(agent_id,
836 (u32 __user *)(arg +
837 offsetof(struct ib_user_mad_reg_req2, id)))) {
838 ret = -EFAULT;
839 goto out;
842 if (!file->already_used) {
843 file->already_used = 1;
844 file->use_pkey_index = 1;
847 file->agent[agent_id] = agent;
848 ret = 0;
850 out:
851 mutex_unlock(&file->mutex);
853 if (ret && agent)
854 ib_unregister_mad_agent(agent);
856 mutex_unlock(&file->port->file_mutex);
858 return ret;
862 static int ib_umad_unreg_agent(struct ib_umad_file *file, u32 __user *arg)
864 struct ib_mad_agent *agent = NULL;
865 u32 id;
866 int ret = 0;
868 if (get_user(id, arg))
869 return -EFAULT;
871 mutex_lock(&file->port->file_mutex);
872 mutex_lock(&file->mutex);
874 if (id >= IB_UMAD_MAX_AGENTS || !__get_agent(file, id)) {
875 ret = -EINVAL;
876 goto out;
879 agent = file->agent[id];
880 file->agent[id] = NULL;
882 out:
883 mutex_unlock(&file->mutex);
885 if (agent)
886 ib_unregister_mad_agent(agent);
888 mutex_unlock(&file->port->file_mutex);
890 return ret;
893 static long ib_umad_enable_pkey(struct ib_umad_file *file)
895 int ret = 0;
897 mutex_lock(&file->mutex);
898 if (file->already_used)
899 ret = -EINVAL;
900 else
901 file->use_pkey_index = 1;
902 mutex_unlock(&file->mutex);
904 return ret;
907 static long ib_umad_ioctl(struct file *filp, unsigned int cmd,
908 unsigned long arg)
910 switch (cmd) {
911 case IB_USER_MAD_REGISTER_AGENT:
912 return ib_umad_reg_agent(filp->private_data, (void __user *) arg, 0);
913 case IB_USER_MAD_UNREGISTER_AGENT:
914 return ib_umad_unreg_agent(filp->private_data, (__u32 __user *) arg);
915 case IB_USER_MAD_ENABLE_PKEY:
916 return ib_umad_enable_pkey(filp->private_data);
917 case IB_USER_MAD_REGISTER_AGENT2:
918 return ib_umad_reg_agent2(filp->private_data, (void __user *) arg);
919 default:
920 return -ENOIOCTLCMD;
924 #ifdef CONFIG_COMPAT
925 static long ib_umad_compat_ioctl(struct file *filp, unsigned int cmd,
926 unsigned long arg)
928 switch (cmd) {
929 case IB_USER_MAD_REGISTER_AGENT:
930 return ib_umad_reg_agent(filp->private_data, compat_ptr(arg), 1);
931 case IB_USER_MAD_UNREGISTER_AGENT:
932 return ib_umad_unreg_agent(filp->private_data, compat_ptr(arg));
933 case IB_USER_MAD_ENABLE_PKEY:
934 return ib_umad_enable_pkey(filp->private_data);
935 case IB_USER_MAD_REGISTER_AGENT2:
936 return ib_umad_reg_agent2(filp->private_data, compat_ptr(arg));
937 default:
938 return -ENOIOCTLCMD;
941 #endif
944 * ib_umad_open() does not need the BKL:
946 * - the ib_umad_port structures are properly reference counted, and
947 * everything else is purely local to the file being created, so
948 * races against other open calls are not a problem;
949 * - the ioctl method does not affect any global state outside of the
950 * file structure being operated on;
952 static int ib_umad_open(struct inode *inode, struct file *filp)
954 struct ib_umad_port *port;
955 struct ib_umad_file *file;
956 int ret = -ENXIO;
958 port = container_of(inode->i_cdev, struct ib_umad_port, cdev);
960 mutex_lock(&port->file_mutex);
962 if (!port->ib_dev)
963 goto out;
965 ret = -ENOMEM;
966 file = kzalloc(sizeof *file, GFP_KERNEL);
967 if (!file)
968 goto out;
970 mutex_init(&file->mutex);
971 spin_lock_init(&file->send_lock);
972 INIT_LIST_HEAD(&file->recv_list);
973 INIT_LIST_HEAD(&file->send_list);
974 init_waitqueue_head(&file->recv_wait);
976 file->port = port;
977 filp->private_data = file;
979 list_add_tail(&file->port_list, &port->file_list);
981 ret = nonseekable_open(inode, filp);
982 if (ret) {
983 list_del(&file->port_list);
984 kfree(file);
985 goto out;
988 kobject_get(&port->umad_dev->kobj);
990 out:
991 mutex_unlock(&port->file_mutex);
992 return ret;
995 static int ib_umad_close(struct inode *inode, struct file *filp)
997 struct ib_umad_file *file = filp->private_data;
998 struct ib_umad_device *dev = file->port->umad_dev;
999 struct ib_umad_packet *packet, *tmp;
1000 int already_dead;
1001 int i;
1003 mutex_lock(&file->port->file_mutex);
1004 mutex_lock(&file->mutex);
1006 already_dead = file->agents_dead;
1007 file->agents_dead = 1;
1009 list_for_each_entry_safe(packet, tmp, &file->recv_list, list) {
1010 if (packet->recv_wc)
1011 ib_free_recv_mad(packet->recv_wc);
1012 kfree(packet);
1015 list_del(&file->port_list);
1017 mutex_unlock(&file->mutex);
1019 if (!already_dead)
1020 for (i = 0; i < IB_UMAD_MAX_AGENTS; ++i)
1021 if (file->agent[i])
1022 ib_unregister_mad_agent(file->agent[i]);
1024 mutex_unlock(&file->port->file_mutex);
1026 kfree(file);
1027 kobject_put(&dev->kobj);
1029 return 0;
1032 static const struct file_operations umad_fops = {
1033 .owner = THIS_MODULE,
1034 .read = ib_umad_read,
1035 .write = ib_umad_write,
1036 .poll = ib_umad_poll,
1037 .unlocked_ioctl = ib_umad_ioctl,
1038 #ifdef CONFIG_COMPAT
1039 .compat_ioctl = ib_umad_compat_ioctl,
1040 #endif
1041 .open = ib_umad_open,
1042 .release = ib_umad_close,
1043 .llseek = no_llseek,
1046 static int ib_umad_sm_open(struct inode *inode, struct file *filp)
1048 struct ib_umad_port *port;
1049 struct ib_port_modify props = {
1050 .set_port_cap_mask = IB_PORT_SM
1052 int ret;
1054 port = container_of(inode->i_cdev, struct ib_umad_port, sm_cdev);
1056 if (filp->f_flags & O_NONBLOCK) {
1057 if (down_trylock(&port->sm_sem)) {
1058 ret = -EAGAIN;
1059 goto fail;
1061 } else {
1062 if (down_interruptible(&port->sm_sem)) {
1063 ret = -ERESTARTSYS;
1064 goto fail;
1068 ret = ib_modify_port(port->ib_dev, port->port_num, 0, &props);
1069 if (ret)
1070 goto err_up_sem;
1072 filp->private_data = port;
1074 ret = nonseekable_open(inode, filp);
1075 if (ret)
1076 goto err_clr_sm_cap;
1078 kobject_get(&port->umad_dev->kobj);
1080 return 0;
1082 err_clr_sm_cap:
1083 swap(props.set_port_cap_mask, props.clr_port_cap_mask);
1084 ib_modify_port(port->ib_dev, port->port_num, 0, &props);
1086 err_up_sem:
1087 up(&port->sm_sem);
1089 fail:
1090 return ret;
1093 static int ib_umad_sm_close(struct inode *inode, struct file *filp)
1095 struct ib_umad_port *port = filp->private_data;
1096 struct ib_port_modify props = {
1097 .clr_port_cap_mask = IB_PORT_SM
1099 int ret = 0;
1101 mutex_lock(&port->file_mutex);
1102 if (port->ib_dev)
1103 ret = ib_modify_port(port->ib_dev, port->port_num, 0, &props);
1104 mutex_unlock(&port->file_mutex);
1106 up(&port->sm_sem);
1108 kobject_put(&port->umad_dev->kobj);
1110 return ret;
1113 static const struct file_operations umad_sm_fops = {
1114 .owner = THIS_MODULE,
1115 .open = ib_umad_sm_open,
1116 .release = ib_umad_sm_close,
1117 .llseek = no_llseek,
1120 static struct ib_client umad_client = {
1121 .name = "umad",
1122 .add = ib_umad_add_one,
1123 .remove = ib_umad_remove_one
1126 static ssize_t show_ibdev(struct device *dev, struct device_attribute *attr,
1127 char *buf)
1129 struct ib_umad_port *port = dev_get_drvdata(dev);
1131 if (!port)
1132 return -ENODEV;
1134 return sprintf(buf, "%s\n", port->ib_dev->name);
1136 static DEVICE_ATTR(ibdev, S_IRUGO, show_ibdev, NULL);
1138 static ssize_t show_port(struct device *dev, struct device_attribute *attr,
1139 char *buf)
1141 struct ib_umad_port *port = dev_get_drvdata(dev);
1143 if (!port)
1144 return -ENODEV;
1146 return sprintf(buf, "%d\n", port->port_num);
1148 static DEVICE_ATTR(port, S_IRUGO, show_port, NULL);
1150 static CLASS_ATTR_STRING(abi_version, S_IRUGO,
1151 __stringify(IB_USER_MAD_ABI_VERSION));
1153 static int ib_umad_init_port(struct ib_device *device, int port_num,
1154 struct ib_umad_device *umad_dev,
1155 struct ib_umad_port *port)
1157 int devnum;
1158 dev_t base_umad;
1159 dev_t base_issm;
1161 devnum = find_first_zero_bit(dev_map, IB_UMAD_MAX_PORTS);
1162 if (devnum >= IB_UMAD_MAX_PORTS)
1163 return -1;
1164 port->dev_num = devnum;
1165 set_bit(devnum, dev_map);
1166 if (devnum >= IB_UMAD_NUM_FIXED_MINOR) {
1167 base_umad = dynamic_umad_dev + devnum - IB_UMAD_NUM_FIXED_MINOR;
1168 base_issm = dynamic_issm_dev + devnum - IB_UMAD_NUM_FIXED_MINOR;
1169 } else {
1170 base_umad = devnum + base_umad_dev;
1171 base_issm = devnum + base_issm_dev;
1174 port->ib_dev = device;
1175 port->port_num = port_num;
1176 sema_init(&port->sm_sem, 1);
1177 mutex_init(&port->file_mutex);
1178 INIT_LIST_HEAD(&port->file_list);
1180 cdev_init(&port->cdev, &umad_fops);
1181 port->cdev.owner = THIS_MODULE;
1182 cdev_set_parent(&port->cdev, &umad_dev->kobj);
1183 kobject_set_name(&port->cdev.kobj, "umad%d", port->dev_num);
1184 if (cdev_add(&port->cdev, base_umad, 1))
1185 goto err_cdev;
1187 port->dev = device_create(umad_class, device->dev.parent,
1188 port->cdev.dev, port,
1189 "umad%d", port->dev_num);
1190 if (IS_ERR(port->dev))
1191 goto err_cdev;
1193 if (device_create_file(port->dev, &dev_attr_ibdev))
1194 goto err_dev;
1195 if (device_create_file(port->dev, &dev_attr_port))
1196 goto err_dev;
1198 cdev_init(&port->sm_cdev, &umad_sm_fops);
1199 port->sm_cdev.owner = THIS_MODULE;
1200 cdev_set_parent(&port->sm_cdev, &umad_dev->kobj);
1201 kobject_set_name(&port->sm_cdev.kobj, "issm%d", port->dev_num);
1202 if (cdev_add(&port->sm_cdev, base_issm, 1))
1203 goto err_sm_cdev;
1205 port->sm_dev = device_create(umad_class, device->dev.parent,
1206 port->sm_cdev.dev, port,
1207 "issm%d", port->dev_num);
1208 if (IS_ERR(port->sm_dev))
1209 goto err_sm_cdev;
1211 if (device_create_file(port->sm_dev, &dev_attr_ibdev))
1212 goto err_sm_dev;
1213 if (device_create_file(port->sm_dev, &dev_attr_port))
1214 goto err_sm_dev;
1216 return 0;
1218 err_sm_dev:
1219 device_destroy(umad_class, port->sm_cdev.dev);
1221 err_sm_cdev:
1222 cdev_del(&port->sm_cdev);
1224 err_dev:
1225 device_destroy(umad_class, port->cdev.dev);
1227 err_cdev:
1228 cdev_del(&port->cdev);
1229 clear_bit(devnum, dev_map);
1231 return -1;
1234 static void ib_umad_kill_port(struct ib_umad_port *port)
1236 struct ib_umad_file *file;
1237 int id;
1239 dev_set_drvdata(port->dev, NULL);
1240 dev_set_drvdata(port->sm_dev, NULL);
1242 device_destroy(umad_class, port->cdev.dev);
1243 device_destroy(umad_class, port->sm_cdev.dev);
1245 cdev_del(&port->cdev);
1246 cdev_del(&port->sm_cdev);
1248 mutex_lock(&port->file_mutex);
1250 port->ib_dev = NULL;
1252 list_for_each_entry(file, &port->file_list, port_list) {
1253 mutex_lock(&file->mutex);
1254 file->agents_dead = 1;
1255 mutex_unlock(&file->mutex);
1257 for (id = 0; id < IB_UMAD_MAX_AGENTS; ++id)
1258 if (file->agent[id])
1259 ib_unregister_mad_agent(file->agent[id]);
1262 mutex_unlock(&port->file_mutex);
1263 clear_bit(port->dev_num, dev_map);
1266 static void ib_umad_add_one(struct ib_device *device)
1268 struct ib_umad_device *umad_dev;
1269 int s, e, i;
1270 int count = 0;
1272 s = rdma_start_port(device);
1273 e = rdma_end_port(device);
1275 umad_dev = kzalloc(sizeof *umad_dev +
1276 (e - s + 1) * sizeof (struct ib_umad_port),
1277 GFP_KERNEL);
1278 if (!umad_dev)
1279 return;
1281 kobject_init(&umad_dev->kobj, &ib_umad_dev_ktype);
1283 for (i = s; i <= e; ++i) {
1284 if (!rdma_cap_ib_mad(device, i))
1285 continue;
1287 umad_dev->port[i - s].umad_dev = umad_dev;
1289 if (ib_umad_init_port(device, i, umad_dev,
1290 &umad_dev->port[i - s]))
1291 goto err;
1293 count++;
1296 if (!count)
1297 goto free;
1299 ib_set_client_data(device, &umad_client, umad_dev);
1301 return;
1303 err:
1304 while (--i >= s) {
1305 if (!rdma_cap_ib_mad(device, i))
1306 continue;
1308 ib_umad_kill_port(&umad_dev->port[i - s]);
1310 free:
1311 kobject_put(&umad_dev->kobj);
1314 static void ib_umad_remove_one(struct ib_device *device, void *client_data)
1316 struct ib_umad_device *umad_dev = client_data;
1317 int i;
1319 if (!umad_dev)
1320 return;
1322 for (i = 0; i <= rdma_end_port(device) - rdma_start_port(device); ++i) {
1323 if (rdma_cap_ib_mad(device, i + rdma_start_port(device)))
1324 ib_umad_kill_port(&umad_dev->port[i]);
1327 kobject_put(&umad_dev->kobj);
1330 static char *umad_devnode(struct device *dev, umode_t *mode)
1332 return kasprintf(GFP_KERNEL, "infiniband/%s", dev_name(dev));
1335 static int __init ib_umad_init(void)
1337 int ret;
1339 ret = register_chrdev_region(base_umad_dev,
1340 IB_UMAD_NUM_FIXED_MINOR * 2,
1341 "infiniband_mad");
1342 if (ret) {
1343 pr_err("couldn't register device number\n");
1344 goto out;
1347 ret = alloc_chrdev_region(&dynamic_umad_dev, 0,
1348 IB_UMAD_NUM_DYNAMIC_MINOR * 2,
1349 "infiniband_mad");
1350 if (ret) {
1351 pr_err("couldn't register dynamic device number\n");
1352 goto out_alloc;
1354 dynamic_issm_dev = dynamic_umad_dev + IB_UMAD_NUM_DYNAMIC_MINOR;
1356 umad_class = class_create(THIS_MODULE, "infiniband_mad");
1357 if (IS_ERR(umad_class)) {
1358 ret = PTR_ERR(umad_class);
1359 pr_err("couldn't create class infiniband_mad\n");
1360 goto out_chrdev;
1363 umad_class->devnode = umad_devnode;
1365 ret = class_create_file(umad_class, &class_attr_abi_version.attr);
1366 if (ret) {
1367 pr_err("couldn't create abi_version attribute\n");
1368 goto out_class;
1371 ret = ib_register_client(&umad_client);
1372 if (ret) {
1373 pr_err("couldn't register ib_umad client\n");
1374 goto out_class;
1377 return 0;
1379 out_class:
1380 class_destroy(umad_class);
1382 out_chrdev:
1383 unregister_chrdev_region(dynamic_umad_dev,
1384 IB_UMAD_NUM_DYNAMIC_MINOR * 2);
1386 out_alloc:
1387 unregister_chrdev_region(base_umad_dev,
1388 IB_UMAD_NUM_FIXED_MINOR * 2);
1390 out:
1391 return ret;
1394 static void __exit ib_umad_cleanup(void)
1396 ib_unregister_client(&umad_client);
1397 class_destroy(umad_class);
1398 unregister_chrdev_region(base_umad_dev,
1399 IB_UMAD_NUM_FIXED_MINOR * 2);
1400 unregister_chrdev_region(dynamic_umad_dev,
1401 IB_UMAD_NUM_DYNAMIC_MINOR * 2);
1404 module_init(ib_umad_init);
1405 module_exit(ib_umad_cleanup);