Merge branch 'fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/evalenti/linux...
[linux/fpc-iii.git] / drivers / infiniband / core / user_mad.c
blob415a3185cde7f45dfc583180ef1a4ef81fec8699
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 <asm/uaccess.h>
55 #include <rdma/ib_mad.h>
56 #include <rdma/ib_user_mad.h>
58 MODULE_AUTHOR("Roland Dreier");
59 MODULE_DESCRIPTION("InfiniBand userspace MAD packet access");
60 MODULE_LICENSE("Dual BSD/GPL");
62 enum {
63 IB_UMAD_MAX_PORTS = 64,
64 IB_UMAD_MAX_AGENTS = 32,
66 IB_UMAD_MAJOR = 231,
67 IB_UMAD_MINOR_BASE = 0
71 * Our lifetime rules for these structs are the following:
72 * device special file is opened, we take a reference on the
73 * ib_umad_port's struct ib_umad_device. We drop these
74 * references in the corresponding close().
76 * In addition to references coming from open character devices, there
77 * is one more reference to each ib_umad_device representing the
78 * module's reference taken when allocating the ib_umad_device in
79 * ib_umad_add_one().
81 * When destroying an ib_umad_device, we drop the module's reference.
84 struct ib_umad_port {
85 struct cdev cdev;
86 struct device *dev;
88 struct cdev sm_cdev;
89 struct device *sm_dev;
90 struct semaphore sm_sem;
92 struct mutex file_mutex;
93 struct list_head file_list;
95 struct ib_device *ib_dev;
96 struct ib_umad_device *umad_dev;
97 int dev_num;
98 u8 port_num;
101 struct ib_umad_device {
102 struct kobject kobj;
103 struct ib_umad_port port[0];
106 struct ib_umad_file {
107 struct mutex mutex;
108 struct ib_umad_port *port;
109 struct list_head recv_list;
110 struct list_head send_list;
111 struct list_head port_list;
112 spinlock_t send_lock;
113 wait_queue_head_t recv_wait;
114 struct ib_mad_agent *agent[IB_UMAD_MAX_AGENTS];
115 int agents_dead;
116 u8 use_pkey_index;
117 u8 already_used;
120 struct ib_umad_packet {
121 struct ib_mad_send_buf *msg;
122 struct ib_mad_recv_wc *recv_wc;
123 struct list_head list;
124 int length;
125 struct ib_user_mad mad;
128 static struct class *umad_class;
130 static const dev_t base_dev = MKDEV(IB_UMAD_MAJOR, IB_UMAD_MINOR_BASE);
132 static DEFINE_SPINLOCK(port_lock);
133 static DECLARE_BITMAP(dev_map, IB_UMAD_MAX_PORTS);
135 static void ib_umad_add_one(struct ib_device *device);
136 static void ib_umad_remove_one(struct ib_device *device, void *client_data);
138 static void ib_umad_release_dev(struct kobject *kobj)
140 struct ib_umad_device *dev =
141 container_of(kobj, struct ib_umad_device, kobj);
143 kfree(dev);
146 static struct kobj_type ib_umad_dev_ktype = {
147 .release = ib_umad_release_dev,
150 static int hdr_size(struct ib_umad_file *file)
152 return file->use_pkey_index ? sizeof (struct ib_user_mad_hdr) :
153 sizeof (struct ib_user_mad_hdr_old);
156 /* caller must hold file->mutex */
157 static struct ib_mad_agent *__get_agent(struct ib_umad_file *file, int id)
159 return file->agents_dead ? NULL : file->agent[id];
162 static int queue_packet(struct ib_umad_file *file,
163 struct ib_mad_agent *agent,
164 struct ib_umad_packet *packet)
166 int ret = 1;
168 mutex_lock(&file->mutex);
170 for (packet->mad.hdr.id = 0;
171 packet->mad.hdr.id < IB_UMAD_MAX_AGENTS;
172 packet->mad.hdr.id++)
173 if (agent == __get_agent(file, packet->mad.hdr.id)) {
174 list_add_tail(&packet->list, &file->recv_list);
175 wake_up_interruptible(&file->recv_wait);
176 ret = 0;
177 break;
180 mutex_unlock(&file->mutex);
182 return ret;
185 static void dequeue_send(struct ib_umad_file *file,
186 struct ib_umad_packet *packet)
188 spin_lock_irq(&file->send_lock);
189 list_del(&packet->list);
190 spin_unlock_irq(&file->send_lock);
193 static void send_handler(struct ib_mad_agent *agent,
194 struct ib_mad_send_wc *send_wc)
196 struct ib_umad_file *file = agent->context;
197 struct ib_umad_packet *packet = send_wc->send_buf->context[0];
199 dequeue_send(file, packet);
200 ib_destroy_ah(packet->msg->ah);
201 ib_free_send_mad(packet->msg);
203 if (send_wc->status == IB_WC_RESP_TIMEOUT_ERR) {
204 packet->length = IB_MGMT_MAD_HDR;
205 packet->mad.hdr.status = ETIMEDOUT;
206 if (!queue_packet(file, agent, packet))
207 return;
209 kfree(packet);
212 static void recv_handler(struct ib_mad_agent *agent,
213 struct ib_mad_send_buf *send_buf,
214 struct ib_mad_recv_wc *mad_recv_wc)
216 struct ib_umad_file *file = agent->context;
217 struct ib_umad_packet *packet;
219 if (mad_recv_wc->wc->status != IB_WC_SUCCESS)
220 goto err1;
222 packet = kzalloc(sizeof *packet, GFP_KERNEL);
223 if (!packet)
224 goto err1;
226 packet->length = mad_recv_wc->mad_len;
227 packet->recv_wc = mad_recv_wc;
229 packet->mad.hdr.status = 0;
230 packet->mad.hdr.length = hdr_size(file) + mad_recv_wc->mad_len;
231 packet->mad.hdr.qpn = cpu_to_be32(mad_recv_wc->wc->src_qp);
232 packet->mad.hdr.lid = cpu_to_be16(mad_recv_wc->wc->slid);
233 packet->mad.hdr.sl = mad_recv_wc->wc->sl;
234 packet->mad.hdr.path_bits = mad_recv_wc->wc->dlid_path_bits;
235 packet->mad.hdr.pkey_index = mad_recv_wc->wc->pkey_index;
236 packet->mad.hdr.grh_present = !!(mad_recv_wc->wc->wc_flags & IB_WC_GRH);
237 if (packet->mad.hdr.grh_present) {
238 struct ib_ah_attr ah_attr;
240 ib_init_ah_from_wc(agent->device, agent->port_num,
241 mad_recv_wc->wc, mad_recv_wc->recv_buf.grh,
242 &ah_attr);
244 packet->mad.hdr.gid_index = ah_attr.grh.sgid_index;
245 packet->mad.hdr.hop_limit = ah_attr.grh.hop_limit;
246 packet->mad.hdr.traffic_class = ah_attr.grh.traffic_class;
247 memcpy(packet->mad.hdr.gid, &ah_attr.grh.dgid, 16);
248 packet->mad.hdr.flow_label = cpu_to_be32(ah_attr.grh.flow_label);
251 if (queue_packet(file, agent, packet))
252 goto err2;
253 return;
255 err2:
256 kfree(packet);
257 err1:
258 ib_free_recv_mad(mad_recv_wc);
261 static ssize_t copy_recv_mad(struct ib_umad_file *file, char __user *buf,
262 struct ib_umad_packet *packet, size_t count)
264 struct ib_mad_recv_buf *recv_buf;
265 int left, seg_payload, offset, max_seg_payload;
266 size_t seg_size;
268 recv_buf = &packet->recv_wc->recv_buf;
269 seg_size = packet->recv_wc->mad_seg_size;
271 /* We need enough room to copy the first (or only) MAD segment. */
272 if ((packet->length <= seg_size &&
273 count < hdr_size(file) + packet->length) ||
274 (packet->length > seg_size &&
275 count < hdr_size(file) + seg_size))
276 return -EINVAL;
278 if (copy_to_user(buf, &packet->mad, hdr_size(file)))
279 return -EFAULT;
281 buf += hdr_size(file);
282 seg_payload = min_t(int, packet->length, seg_size);
283 if (copy_to_user(buf, recv_buf->mad, seg_payload))
284 return -EFAULT;
286 if (seg_payload < packet->length) {
288 * Multipacket RMPP MAD message. Copy remainder of message.
289 * Note that last segment may have a shorter payload.
291 if (count < hdr_size(file) + packet->length) {
293 * The buffer is too small, return the first RMPP segment,
294 * which includes the RMPP message length.
296 return -ENOSPC;
298 offset = ib_get_mad_data_offset(recv_buf->mad->mad_hdr.mgmt_class);
299 max_seg_payload = seg_size - offset;
301 for (left = packet->length - seg_payload, buf += seg_payload;
302 left; left -= seg_payload, buf += seg_payload) {
303 recv_buf = container_of(recv_buf->list.next,
304 struct ib_mad_recv_buf, list);
305 seg_payload = min(left, max_seg_payload);
306 if (copy_to_user(buf, ((void *) recv_buf->mad) + offset,
307 seg_payload))
308 return -EFAULT;
311 return hdr_size(file) + packet->length;
314 static ssize_t copy_send_mad(struct ib_umad_file *file, char __user *buf,
315 struct ib_umad_packet *packet, size_t count)
317 ssize_t size = hdr_size(file) + packet->length;
319 if (count < size)
320 return -EINVAL;
322 if (copy_to_user(buf, &packet->mad, hdr_size(file)))
323 return -EFAULT;
325 buf += hdr_size(file);
327 if (copy_to_user(buf, packet->mad.data, packet->length))
328 return -EFAULT;
330 return size;
333 static ssize_t ib_umad_read(struct file *filp, char __user *buf,
334 size_t count, loff_t *pos)
336 struct ib_umad_file *file = filp->private_data;
337 struct ib_umad_packet *packet;
338 ssize_t ret;
340 if (count < hdr_size(file))
341 return -EINVAL;
343 mutex_lock(&file->mutex);
345 while (list_empty(&file->recv_list)) {
346 mutex_unlock(&file->mutex);
348 if (filp->f_flags & O_NONBLOCK)
349 return -EAGAIN;
351 if (wait_event_interruptible(file->recv_wait,
352 !list_empty(&file->recv_list)))
353 return -ERESTARTSYS;
355 mutex_lock(&file->mutex);
358 packet = list_entry(file->recv_list.next, struct ib_umad_packet, list);
359 list_del(&packet->list);
361 mutex_unlock(&file->mutex);
363 if (packet->recv_wc)
364 ret = copy_recv_mad(file, buf, packet, count);
365 else
366 ret = copy_send_mad(file, buf, packet, count);
368 if (ret < 0) {
369 /* Requeue packet */
370 mutex_lock(&file->mutex);
371 list_add(&packet->list, &file->recv_list);
372 mutex_unlock(&file->mutex);
373 } else {
374 if (packet->recv_wc)
375 ib_free_recv_mad(packet->recv_wc);
376 kfree(packet);
378 return ret;
381 static int copy_rmpp_mad(struct ib_mad_send_buf *msg, const char __user *buf)
383 int left, seg;
385 /* Copy class specific header */
386 if ((msg->hdr_len > IB_MGMT_RMPP_HDR) &&
387 copy_from_user(msg->mad + IB_MGMT_RMPP_HDR, buf + IB_MGMT_RMPP_HDR,
388 msg->hdr_len - IB_MGMT_RMPP_HDR))
389 return -EFAULT;
391 /* All headers are in place. Copy data segments. */
392 for (seg = 1, left = msg->data_len, buf += msg->hdr_len; left > 0;
393 seg++, left -= msg->seg_size, buf += msg->seg_size) {
394 if (copy_from_user(ib_get_rmpp_segment(msg, seg), buf,
395 min(left, msg->seg_size)))
396 return -EFAULT;
398 return 0;
401 static int same_destination(struct ib_user_mad_hdr *hdr1,
402 struct ib_user_mad_hdr *hdr2)
404 if (!hdr1->grh_present && !hdr2->grh_present)
405 return (hdr1->lid == hdr2->lid);
407 if (hdr1->grh_present && hdr2->grh_present)
408 return !memcmp(hdr1->gid, hdr2->gid, 16);
410 return 0;
413 static int is_duplicate(struct ib_umad_file *file,
414 struct ib_umad_packet *packet)
416 struct ib_umad_packet *sent_packet;
417 struct ib_mad_hdr *sent_hdr, *hdr;
419 hdr = (struct ib_mad_hdr *) packet->mad.data;
420 list_for_each_entry(sent_packet, &file->send_list, list) {
421 sent_hdr = (struct ib_mad_hdr *) sent_packet->mad.data;
423 if ((hdr->tid != sent_hdr->tid) ||
424 (hdr->mgmt_class != sent_hdr->mgmt_class))
425 continue;
428 * No need to be overly clever here. If two new operations have
429 * the same TID, reject the second as a duplicate. This is more
430 * restrictive than required by the spec.
432 if (!ib_response_mad(hdr)) {
433 if (!ib_response_mad(sent_hdr))
434 return 1;
435 continue;
436 } else if (!ib_response_mad(sent_hdr))
437 continue;
439 if (same_destination(&packet->mad.hdr, &sent_packet->mad.hdr))
440 return 1;
443 return 0;
446 static ssize_t ib_umad_write(struct file *filp, const char __user *buf,
447 size_t count, loff_t *pos)
449 struct ib_umad_file *file = filp->private_data;
450 struct ib_umad_packet *packet;
451 struct ib_mad_agent *agent;
452 struct ib_ah_attr ah_attr;
453 struct ib_ah *ah;
454 struct ib_rmpp_mad *rmpp_mad;
455 __be64 *tid;
456 int ret, data_len, hdr_len, copy_offset, rmpp_active;
457 u8 base_version;
459 if (count < hdr_size(file) + IB_MGMT_RMPP_HDR)
460 return -EINVAL;
462 packet = kzalloc(sizeof *packet + IB_MGMT_RMPP_HDR, GFP_KERNEL);
463 if (!packet)
464 return -ENOMEM;
466 if (copy_from_user(&packet->mad, buf, hdr_size(file))) {
467 ret = -EFAULT;
468 goto err;
471 if (packet->mad.hdr.id >= IB_UMAD_MAX_AGENTS) {
472 ret = -EINVAL;
473 goto err;
476 buf += hdr_size(file);
478 if (copy_from_user(packet->mad.data, buf, IB_MGMT_RMPP_HDR)) {
479 ret = -EFAULT;
480 goto err;
483 mutex_lock(&file->mutex);
485 agent = __get_agent(file, packet->mad.hdr.id);
486 if (!agent) {
487 ret = -EINVAL;
488 goto err_up;
491 memset(&ah_attr, 0, sizeof ah_attr);
492 ah_attr.dlid = be16_to_cpu(packet->mad.hdr.lid);
493 ah_attr.sl = packet->mad.hdr.sl;
494 ah_attr.src_path_bits = packet->mad.hdr.path_bits;
495 ah_attr.port_num = file->port->port_num;
496 if (packet->mad.hdr.grh_present) {
497 ah_attr.ah_flags = IB_AH_GRH;
498 memcpy(ah_attr.grh.dgid.raw, packet->mad.hdr.gid, 16);
499 ah_attr.grh.sgid_index = packet->mad.hdr.gid_index;
500 ah_attr.grh.flow_label = be32_to_cpu(packet->mad.hdr.flow_label);
501 ah_attr.grh.hop_limit = packet->mad.hdr.hop_limit;
502 ah_attr.grh.traffic_class = packet->mad.hdr.traffic_class;
505 ah = ib_create_ah(agent->qp->pd, &ah_attr);
506 if (IS_ERR(ah)) {
507 ret = PTR_ERR(ah);
508 goto err_up;
511 rmpp_mad = (struct ib_rmpp_mad *) packet->mad.data;
512 hdr_len = ib_get_mad_data_offset(rmpp_mad->mad_hdr.mgmt_class);
514 if (ib_is_mad_class_rmpp(rmpp_mad->mad_hdr.mgmt_class)
515 && ib_mad_kernel_rmpp_agent(agent)) {
516 copy_offset = IB_MGMT_RMPP_HDR;
517 rmpp_active = ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) &
518 IB_MGMT_RMPP_FLAG_ACTIVE;
519 } else {
520 copy_offset = IB_MGMT_MAD_HDR;
521 rmpp_active = 0;
524 base_version = ((struct ib_mad_hdr *)&packet->mad.data)->base_version;
525 data_len = count - hdr_size(file) - hdr_len;
526 packet->msg = ib_create_send_mad(agent,
527 be32_to_cpu(packet->mad.hdr.qpn),
528 packet->mad.hdr.pkey_index, rmpp_active,
529 hdr_len, data_len, GFP_KERNEL,
530 base_version);
531 if (IS_ERR(packet->msg)) {
532 ret = PTR_ERR(packet->msg);
533 goto err_ah;
536 packet->msg->ah = ah;
537 packet->msg->timeout_ms = packet->mad.hdr.timeout_ms;
538 packet->msg->retries = packet->mad.hdr.retries;
539 packet->msg->context[0] = packet;
541 /* Copy MAD header. Any RMPP header is already in place. */
542 memcpy(packet->msg->mad, packet->mad.data, IB_MGMT_MAD_HDR);
544 if (!rmpp_active) {
545 if (copy_from_user(packet->msg->mad + copy_offset,
546 buf + copy_offset,
547 hdr_len + data_len - copy_offset)) {
548 ret = -EFAULT;
549 goto err_msg;
551 } else {
552 ret = copy_rmpp_mad(packet->msg, buf);
553 if (ret)
554 goto err_msg;
558 * Set the high-order part of the transaction ID to make MADs from
559 * different agents unique, and allow routing responses back to the
560 * original requestor.
562 if (!ib_response_mad(packet->msg->mad)) {
563 tid = &((struct ib_mad_hdr *) packet->msg->mad)->tid;
564 *tid = cpu_to_be64(((u64) agent->hi_tid) << 32 |
565 (be64_to_cpup(tid) & 0xffffffff));
566 rmpp_mad->mad_hdr.tid = *tid;
569 if (!ib_mad_kernel_rmpp_agent(agent)
570 && ib_is_mad_class_rmpp(rmpp_mad->mad_hdr.mgmt_class)
571 && (ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) & IB_MGMT_RMPP_FLAG_ACTIVE)) {
572 spin_lock_irq(&file->send_lock);
573 list_add_tail(&packet->list, &file->send_list);
574 spin_unlock_irq(&file->send_lock);
575 } else {
576 spin_lock_irq(&file->send_lock);
577 ret = is_duplicate(file, packet);
578 if (!ret)
579 list_add_tail(&packet->list, &file->send_list);
580 spin_unlock_irq(&file->send_lock);
581 if (ret) {
582 ret = -EINVAL;
583 goto err_msg;
587 ret = ib_post_send_mad(packet->msg, NULL);
588 if (ret)
589 goto err_send;
591 mutex_unlock(&file->mutex);
592 return count;
594 err_send:
595 dequeue_send(file, packet);
596 err_msg:
597 ib_free_send_mad(packet->msg);
598 err_ah:
599 ib_destroy_ah(ah);
600 err_up:
601 mutex_unlock(&file->mutex);
602 err:
603 kfree(packet);
604 return ret;
607 static unsigned int ib_umad_poll(struct file *filp, struct poll_table_struct *wait)
609 struct ib_umad_file *file = filp->private_data;
611 /* we will always be able to post a MAD send */
612 unsigned int mask = POLLOUT | POLLWRNORM;
614 poll_wait(filp, &file->recv_wait, wait);
616 if (!list_empty(&file->recv_list))
617 mask |= POLLIN | POLLRDNORM;
619 return mask;
622 static int ib_umad_reg_agent(struct ib_umad_file *file, void __user *arg,
623 int compat_method_mask)
625 struct ib_user_mad_reg_req ureq;
626 struct ib_mad_reg_req req;
627 struct ib_mad_agent *agent = NULL;
628 int agent_id;
629 int ret;
631 mutex_lock(&file->port->file_mutex);
632 mutex_lock(&file->mutex);
634 if (!file->port->ib_dev) {
635 dev_notice(file->port->dev,
636 "ib_umad_reg_agent: invalid device\n");
637 ret = -EPIPE;
638 goto out;
641 if (copy_from_user(&ureq, arg, sizeof ureq)) {
642 ret = -EFAULT;
643 goto out;
646 if (ureq.qpn != 0 && ureq.qpn != 1) {
647 dev_notice(file->port->dev,
648 "ib_umad_reg_agent: invalid QPN %d specified\n",
649 ureq.qpn);
650 ret = -EINVAL;
651 goto out;
654 for (agent_id = 0; agent_id < IB_UMAD_MAX_AGENTS; ++agent_id)
655 if (!__get_agent(file, agent_id))
656 goto found;
658 dev_notice(file->port->dev,
659 "ib_umad_reg_agent: Max Agents (%u) reached\n",
660 IB_UMAD_MAX_AGENTS);
661 ret = -ENOMEM;
662 goto out;
664 found:
665 if (ureq.mgmt_class) {
666 memset(&req, 0, sizeof(req));
667 req.mgmt_class = ureq.mgmt_class;
668 req.mgmt_class_version = ureq.mgmt_class_version;
669 memcpy(req.oui, ureq.oui, sizeof req.oui);
671 if (compat_method_mask) {
672 u32 *umm = (u32 *) ureq.method_mask;
673 int i;
675 for (i = 0; i < BITS_TO_LONGS(IB_MGMT_MAX_METHODS); ++i)
676 req.method_mask[i] =
677 umm[i * 2] | ((u64) umm[i * 2 + 1] << 32);
678 } else
679 memcpy(req.method_mask, ureq.method_mask,
680 sizeof req.method_mask);
683 agent = ib_register_mad_agent(file->port->ib_dev, file->port->port_num,
684 ureq.qpn ? IB_QPT_GSI : IB_QPT_SMI,
685 ureq.mgmt_class ? &req : NULL,
686 ureq.rmpp_version,
687 send_handler, recv_handler, file, 0);
688 if (IS_ERR(agent)) {
689 ret = PTR_ERR(agent);
690 agent = NULL;
691 goto out;
694 if (put_user(agent_id,
695 (u32 __user *) (arg + offsetof(struct ib_user_mad_reg_req, id)))) {
696 ret = -EFAULT;
697 goto out;
700 if (!file->already_used) {
701 file->already_used = 1;
702 if (!file->use_pkey_index) {
703 dev_warn(file->port->dev,
704 "process %s did not enable P_Key index support.\n",
705 current->comm);
706 dev_warn(file->port->dev,
707 " Documentation/infiniband/user_mad.txt has info on the new ABI.\n");
711 file->agent[agent_id] = agent;
712 ret = 0;
714 out:
715 mutex_unlock(&file->mutex);
717 if (ret && agent)
718 ib_unregister_mad_agent(agent);
720 mutex_unlock(&file->port->file_mutex);
722 return ret;
725 static int ib_umad_reg_agent2(struct ib_umad_file *file, void __user *arg)
727 struct ib_user_mad_reg_req2 ureq;
728 struct ib_mad_reg_req req;
729 struct ib_mad_agent *agent = NULL;
730 int agent_id;
731 int ret;
733 mutex_lock(&file->port->file_mutex);
734 mutex_lock(&file->mutex);
736 if (!file->port->ib_dev) {
737 dev_notice(file->port->dev,
738 "ib_umad_reg_agent2: invalid device\n");
739 ret = -EPIPE;
740 goto out;
743 if (copy_from_user(&ureq, arg, sizeof(ureq))) {
744 ret = -EFAULT;
745 goto out;
748 if (ureq.qpn != 0 && ureq.qpn != 1) {
749 dev_notice(file->port->dev,
750 "ib_umad_reg_agent2: invalid QPN %d specified\n",
751 ureq.qpn);
752 ret = -EINVAL;
753 goto out;
756 if (ureq.flags & ~IB_USER_MAD_REG_FLAGS_CAP) {
757 dev_notice(file->port->dev,
758 "ib_umad_reg_agent2 failed: invalid registration flags specified 0x%x; supported 0x%x\n",
759 ureq.flags, IB_USER_MAD_REG_FLAGS_CAP);
760 ret = -EINVAL;
762 if (put_user((u32)IB_USER_MAD_REG_FLAGS_CAP,
763 (u32 __user *) (arg + offsetof(struct
764 ib_user_mad_reg_req2, flags))))
765 ret = -EFAULT;
767 goto out;
770 for (agent_id = 0; agent_id < IB_UMAD_MAX_AGENTS; ++agent_id)
771 if (!__get_agent(file, agent_id))
772 goto found;
774 dev_notice(file->port->dev,
775 "ib_umad_reg_agent2: Max Agents (%u) reached\n",
776 IB_UMAD_MAX_AGENTS);
777 ret = -ENOMEM;
778 goto out;
780 found:
781 if (ureq.mgmt_class) {
782 memset(&req, 0, sizeof(req));
783 req.mgmt_class = ureq.mgmt_class;
784 req.mgmt_class_version = ureq.mgmt_class_version;
785 if (ureq.oui & 0xff000000) {
786 dev_notice(file->port->dev,
787 "ib_umad_reg_agent2 failed: oui invalid 0x%08x\n",
788 ureq.oui);
789 ret = -EINVAL;
790 goto out;
792 req.oui[2] = ureq.oui & 0x0000ff;
793 req.oui[1] = (ureq.oui & 0x00ff00) >> 8;
794 req.oui[0] = (ureq.oui & 0xff0000) >> 16;
795 memcpy(req.method_mask, ureq.method_mask,
796 sizeof(req.method_mask));
799 agent = ib_register_mad_agent(file->port->ib_dev, file->port->port_num,
800 ureq.qpn ? IB_QPT_GSI : IB_QPT_SMI,
801 ureq.mgmt_class ? &req : NULL,
802 ureq.rmpp_version,
803 send_handler, recv_handler, file,
804 ureq.flags);
805 if (IS_ERR(agent)) {
806 ret = PTR_ERR(agent);
807 agent = NULL;
808 goto out;
811 if (put_user(agent_id,
812 (u32 __user *)(arg +
813 offsetof(struct ib_user_mad_reg_req2, id)))) {
814 ret = -EFAULT;
815 goto out;
818 if (!file->already_used) {
819 file->already_used = 1;
820 file->use_pkey_index = 1;
823 file->agent[agent_id] = agent;
824 ret = 0;
826 out:
827 mutex_unlock(&file->mutex);
829 if (ret && agent)
830 ib_unregister_mad_agent(agent);
832 mutex_unlock(&file->port->file_mutex);
834 return ret;
838 static int ib_umad_unreg_agent(struct ib_umad_file *file, u32 __user *arg)
840 struct ib_mad_agent *agent = NULL;
841 u32 id;
842 int ret = 0;
844 if (get_user(id, arg))
845 return -EFAULT;
847 mutex_lock(&file->port->file_mutex);
848 mutex_lock(&file->mutex);
850 if (id >= IB_UMAD_MAX_AGENTS || !__get_agent(file, id)) {
851 ret = -EINVAL;
852 goto out;
855 agent = file->agent[id];
856 file->agent[id] = NULL;
858 out:
859 mutex_unlock(&file->mutex);
861 if (agent)
862 ib_unregister_mad_agent(agent);
864 mutex_unlock(&file->port->file_mutex);
866 return ret;
869 static long ib_umad_enable_pkey(struct ib_umad_file *file)
871 int ret = 0;
873 mutex_lock(&file->mutex);
874 if (file->already_used)
875 ret = -EINVAL;
876 else
877 file->use_pkey_index = 1;
878 mutex_unlock(&file->mutex);
880 return ret;
883 static long ib_umad_ioctl(struct file *filp, unsigned int cmd,
884 unsigned long arg)
886 switch (cmd) {
887 case IB_USER_MAD_REGISTER_AGENT:
888 return ib_umad_reg_agent(filp->private_data, (void __user *) arg, 0);
889 case IB_USER_MAD_UNREGISTER_AGENT:
890 return ib_umad_unreg_agent(filp->private_data, (__u32 __user *) arg);
891 case IB_USER_MAD_ENABLE_PKEY:
892 return ib_umad_enable_pkey(filp->private_data);
893 case IB_USER_MAD_REGISTER_AGENT2:
894 return ib_umad_reg_agent2(filp->private_data, (void __user *) arg);
895 default:
896 return -ENOIOCTLCMD;
900 #ifdef CONFIG_COMPAT
901 static long ib_umad_compat_ioctl(struct file *filp, unsigned int cmd,
902 unsigned long arg)
904 switch (cmd) {
905 case IB_USER_MAD_REGISTER_AGENT:
906 return ib_umad_reg_agent(filp->private_data, compat_ptr(arg), 1);
907 case IB_USER_MAD_UNREGISTER_AGENT:
908 return ib_umad_unreg_agent(filp->private_data, compat_ptr(arg));
909 case IB_USER_MAD_ENABLE_PKEY:
910 return ib_umad_enable_pkey(filp->private_data);
911 case IB_USER_MAD_REGISTER_AGENT2:
912 return ib_umad_reg_agent2(filp->private_data, compat_ptr(arg));
913 default:
914 return -ENOIOCTLCMD;
917 #endif
920 * ib_umad_open() does not need the BKL:
922 * - the ib_umad_port structures are properly reference counted, and
923 * everything else is purely local to the file being created, so
924 * races against other open calls are not a problem;
925 * - the ioctl method does not affect any global state outside of the
926 * file structure being operated on;
928 static int ib_umad_open(struct inode *inode, struct file *filp)
930 struct ib_umad_port *port;
931 struct ib_umad_file *file;
932 int ret = -ENXIO;
934 port = container_of(inode->i_cdev, struct ib_umad_port, cdev);
936 mutex_lock(&port->file_mutex);
938 if (!port->ib_dev)
939 goto out;
941 ret = -ENOMEM;
942 file = kzalloc(sizeof *file, GFP_KERNEL);
943 if (!file)
944 goto out;
946 mutex_init(&file->mutex);
947 spin_lock_init(&file->send_lock);
948 INIT_LIST_HEAD(&file->recv_list);
949 INIT_LIST_HEAD(&file->send_list);
950 init_waitqueue_head(&file->recv_wait);
952 file->port = port;
953 filp->private_data = file;
955 list_add_tail(&file->port_list, &port->file_list);
957 ret = nonseekable_open(inode, filp);
958 if (ret) {
959 list_del(&file->port_list);
960 kfree(file);
961 goto out;
964 kobject_get(&port->umad_dev->kobj);
966 out:
967 mutex_unlock(&port->file_mutex);
968 return ret;
971 static int ib_umad_close(struct inode *inode, struct file *filp)
973 struct ib_umad_file *file = filp->private_data;
974 struct ib_umad_device *dev = file->port->umad_dev;
975 struct ib_umad_packet *packet, *tmp;
976 int already_dead;
977 int i;
979 mutex_lock(&file->port->file_mutex);
980 mutex_lock(&file->mutex);
982 already_dead = file->agents_dead;
983 file->agents_dead = 1;
985 list_for_each_entry_safe(packet, tmp, &file->recv_list, list) {
986 if (packet->recv_wc)
987 ib_free_recv_mad(packet->recv_wc);
988 kfree(packet);
991 list_del(&file->port_list);
993 mutex_unlock(&file->mutex);
995 if (!already_dead)
996 for (i = 0; i < IB_UMAD_MAX_AGENTS; ++i)
997 if (file->agent[i])
998 ib_unregister_mad_agent(file->agent[i]);
1000 mutex_unlock(&file->port->file_mutex);
1002 kfree(file);
1003 kobject_put(&dev->kobj);
1005 return 0;
1008 static const struct file_operations umad_fops = {
1009 .owner = THIS_MODULE,
1010 .read = ib_umad_read,
1011 .write = ib_umad_write,
1012 .poll = ib_umad_poll,
1013 .unlocked_ioctl = ib_umad_ioctl,
1014 #ifdef CONFIG_COMPAT
1015 .compat_ioctl = ib_umad_compat_ioctl,
1016 #endif
1017 .open = ib_umad_open,
1018 .release = ib_umad_close,
1019 .llseek = no_llseek,
1022 static int ib_umad_sm_open(struct inode *inode, struct file *filp)
1024 struct ib_umad_port *port;
1025 struct ib_port_modify props = {
1026 .set_port_cap_mask = IB_PORT_SM
1028 int ret;
1030 port = container_of(inode->i_cdev, struct ib_umad_port, sm_cdev);
1032 if (filp->f_flags & O_NONBLOCK) {
1033 if (down_trylock(&port->sm_sem)) {
1034 ret = -EAGAIN;
1035 goto fail;
1037 } else {
1038 if (down_interruptible(&port->sm_sem)) {
1039 ret = -ERESTARTSYS;
1040 goto fail;
1044 ret = ib_modify_port(port->ib_dev, port->port_num, 0, &props);
1045 if (ret)
1046 goto err_up_sem;
1048 filp->private_data = port;
1050 ret = nonseekable_open(inode, filp);
1051 if (ret)
1052 goto err_clr_sm_cap;
1054 kobject_get(&port->umad_dev->kobj);
1056 return 0;
1058 err_clr_sm_cap:
1059 swap(props.set_port_cap_mask, props.clr_port_cap_mask);
1060 ib_modify_port(port->ib_dev, port->port_num, 0, &props);
1062 err_up_sem:
1063 up(&port->sm_sem);
1065 fail:
1066 return ret;
1069 static int ib_umad_sm_close(struct inode *inode, struct file *filp)
1071 struct ib_umad_port *port = filp->private_data;
1072 struct ib_port_modify props = {
1073 .clr_port_cap_mask = IB_PORT_SM
1075 int ret = 0;
1077 mutex_lock(&port->file_mutex);
1078 if (port->ib_dev)
1079 ret = ib_modify_port(port->ib_dev, port->port_num, 0, &props);
1080 mutex_unlock(&port->file_mutex);
1082 up(&port->sm_sem);
1084 kobject_put(&port->umad_dev->kobj);
1086 return ret;
1089 static const struct file_operations umad_sm_fops = {
1090 .owner = THIS_MODULE,
1091 .open = ib_umad_sm_open,
1092 .release = ib_umad_sm_close,
1093 .llseek = no_llseek,
1096 static struct ib_client umad_client = {
1097 .name = "umad",
1098 .add = ib_umad_add_one,
1099 .remove = ib_umad_remove_one
1102 static ssize_t show_ibdev(struct device *dev, struct device_attribute *attr,
1103 char *buf)
1105 struct ib_umad_port *port = dev_get_drvdata(dev);
1107 if (!port)
1108 return -ENODEV;
1110 return sprintf(buf, "%s\n", port->ib_dev->name);
1112 static DEVICE_ATTR(ibdev, S_IRUGO, show_ibdev, NULL);
1114 static ssize_t show_port(struct device *dev, struct device_attribute *attr,
1115 char *buf)
1117 struct ib_umad_port *port = dev_get_drvdata(dev);
1119 if (!port)
1120 return -ENODEV;
1122 return sprintf(buf, "%d\n", port->port_num);
1124 static DEVICE_ATTR(port, S_IRUGO, show_port, NULL);
1126 static CLASS_ATTR_STRING(abi_version, S_IRUGO,
1127 __stringify(IB_USER_MAD_ABI_VERSION));
1129 static dev_t overflow_maj;
1130 static DECLARE_BITMAP(overflow_map, IB_UMAD_MAX_PORTS);
1131 static int find_overflow_devnum(struct ib_device *device)
1133 int ret;
1135 if (!overflow_maj) {
1136 ret = alloc_chrdev_region(&overflow_maj, 0, IB_UMAD_MAX_PORTS * 2,
1137 "infiniband_mad");
1138 if (ret) {
1139 dev_err(&device->dev,
1140 "couldn't register dynamic device number\n");
1141 return ret;
1145 ret = find_first_zero_bit(overflow_map, IB_UMAD_MAX_PORTS);
1146 if (ret >= IB_UMAD_MAX_PORTS)
1147 return -1;
1149 return ret;
1152 static int ib_umad_init_port(struct ib_device *device, int port_num,
1153 struct ib_umad_device *umad_dev,
1154 struct ib_umad_port *port)
1156 int devnum;
1157 dev_t base;
1159 spin_lock(&port_lock);
1160 devnum = find_first_zero_bit(dev_map, IB_UMAD_MAX_PORTS);
1161 if (devnum >= IB_UMAD_MAX_PORTS) {
1162 spin_unlock(&port_lock);
1163 devnum = find_overflow_devnum(device);
1164 if (devnum < 0)
1165 return -1;
1167 spin_lock(&port_lock);
1168 port->dev_num = devnum + IB_UMAD_MAX_PORTS;
1169 base = devnum + overflow_maj;
1170 set_bit(devnum, overflow_map);
1171 } else {
1172 port->dev_num = devnum;
1173 base = devnum + base_dev;
1174 set_bit(devnum, dev_map);
1176 spin_unlock(&port_lock);
1178 port->ib_dev = device;
1179 port->port_num = port_num;
1180 sema_init(&port->sm_sem, 1);
1181 mutex_init(&port->file_mutex);
1182 INIT_LIST_HEAD(&port->file_list);
1184 cdev_init(&port->cdev, &umad_fops);
1185 port->cdev.owner = THIS_MODULE;
1186 port->cdev.kobj.parent = &umad_dev->kobj;
1187 kobject_set_name(&port->cdev.kobj, "umad%d", port->dev_num);
1188 if (cdev_add(&port->cdev, base, 1))
1189 goto err_cdev;
1191 port->dev = device_create(umad_class, device->dma_device,
1192 port->cdev.dev, port,
1193 "umad%d", port->dev_num);
1194 if (IS_ERR(port->dev))
1195 goto err_cdev;
1197 if (device_create_file(port->dev, &dev_attr_ibdev))
1198 goto err_dev;
1199 if (device_create_file(port->dev, &dev_attr_port))
1200 goto err_dev;
1202 base += IB_UMAD_MAX_PORTS;
1203 cdev_init(&port->sm_cdev, &umad_sm_fops);
1204 port->sm_cdev.owner = THIS_MODULE;
1205 port->sm_cdev.kobj.parent = &umad_dev->kobj;
1206 kobject_set_name(&port->sm_cdev.kobj, "issm%d", port->dev_num);
1207 if (cdev_add(&port->sm_cdev, base, 1))
1208 goto err_sm_cdev;
1210 port->sm_dev = device_create(umad_class, device->dma_device,
1211 port->sm_cdev.dev, port,
1212 "issm%d", port->dev_num);
1213 if (IS_ERR(port->sm_dev))
1214 goto err_sm_cdev;
1216 if (device_create_file(port->sm_dev, &dev_attr_ibdev))
1217 goto err_sm_dev;
1218 if (device_create_file(port->sm_dev, &dev_attr_port))
1219 goto err_sm_dev;
1221 return 0;
1223 err_sm_dev:
1224 device_destroy(umad_class, port->sm_cdev.dev);
1226 err_sm_cdev:
1227 cdev_del(&port->sm_cdev);
1229 err_dev:
1230 device_destroy(umad_class, port->cdev.dev);
1232 err_cdev:
1233 cdev_del(&port->cdev);
1234 if (port->dev_num < IB_UMAD_MAX_PORTS)
1235 clear_bit(devnum, dev_map);
1236 else
1237 clear_bit(devnum, overflow_map);
1239 return -1;
1242 static void ib_umad_kill_port(struct ib_umad_port *port)
1244 struct ib_umad_file *file;
1245 int id;
1247 dev_set_drvdata(port->dev, NULL);
1248 dev_set_drvdata(port->sm_dev, NULL);
1250 device_destroy(umad_class, port->cdev.dev);
1251 device_destroy(umad_class, port->sm_cdev.dev);
1253 cdev_del(&port->cdev);
1254 cdev_del(&port->sm_cdev);
1256 mutex_lock(&port->file_mutex);
1258 port->ib_dev = NULL;
1260 list_for_each_entry(file, &port->file_list, port_list) {
1261 mutex_lock(&file->mutex);
1262 file->agents_dead = 1;
1263 mutex_unlock(&file->mutex);
1265 for (id = 0; id < IB_UMAD_MAX_AGENTS; ++id)
1266 if (file->agent[id])
1267 ib_unregister_mad_agent(file->agent[id]);
1270 mutex_unlock(&port->file_mutex);
1272 if (port->dev_num < IB_UMAD_MAX_PORTS)
1273 clear_bit(port->dev_num, dev_map);
1274 else
1275 clear_bit(port->dev_num - IB_UMAD_MAX_PORTS, overflow_map);
1278 static void ib_umad_add_one(struct ib_device *device)
1280 struct ib_umad_device *umad_dev;
1281 int s, e, i;
1282 int count = 0;
1284 s = rdma_start_port(device);
1285 e = rdma_end_port(device);
1287 umad_dev = kzalloc(sizeof *umad_dev +
1288 (e - s + 1) * sizeof (struct ib_umad_port),
1289 GFP_KERNEL);
1290 if (!umad_dev)
1291 return;
1293 kobject_init(&umad_dev->kobj, &ib_umad_dev_ktype);
1295 for (i = s; i <= e; ++i) {
1296 if (!rdma_cap_ib_mad(device, i))
1297 continue;
1299 umad_dev->port[i - s].umad_dev = umad_dev;
1301 if (ib_umad_init_port(device, i, umad_dev,
1302 &umad_dev->port[i - s]))
1303 goto err;
1305 count++;
1308 if (!count)
1309 goto free;
1311 ib_set_client_data(device, &umad_client, umad_dev);
1313 return;
1315 err:
1316 while (--i >= s) {
1317 if (!rdma_cap_ib_mad(device, i))
1318 continue;
1320 ib_umad_kill_port(&umad_dev->port[i - s]);
1322 free:
1323 kobject_put(&umad_dev->kobj);
1326 static void ib_umad_remove_one(struct ib_device *device, void *client_data)
1328 struct ib_umad_device *umad_dev = client_data;
1329 int i;
1331 if (!umad_dev)
1332 return;
1334 for (i = 0; i <= rdma_end_port(device) - rdma_start_port(device); ++i) {
1335 if (rdma_cap_ib_mad(device, i + rdma_start_port(device)))
1336 ib_umad_kill_port(&umad_dev->port[i]);
1339 kobject_put(&umad_dev->kobj);
1342 static char *umad_devnode(struct device *dev, umode_t *mode)
1344 return kasprintf(GFP_KERNEL, "infiniband/%s", dev_name(dev));
1347 static int __init ib_umad_init(void)
1349 int ret;
1351 ret = register_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2,
1352 "infiniband_mad");
1353 if (ret) {
1354 pr_err("couldn't register device number\n");
1355 goto out;
1358 umad_class = class_create(THIS_MODULE, "infiniband_mad");
1359 if (IS_ERR(umad_class)) {
1360 ret = PTR_ERR(umad_class);
1361 pr_err("couldn't create class infiniband_mad\n");
1362 goto out_chrdev;
1365 umad_class->devnode = umad_devnode;
1367 ret = class_create_file(umad_class, &class_attr_abi_version.attr);
1368 if (ret) {
1369 pr_err("couldn't create abi_version attribute\n");
1370 goto out_class;
1373 ret = ib_register_client(&umad_client);
1374 if (ret) {
1375 pr_err("couldn't register ib_umad client\n");
1376 goto out_class;
1379 return 0;
1381 out_class:
1382 class_destroy(umad_class);
1384 out_chrdev:
1385 unregister_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2);
1387 out:
1388 return ret;
1391 static void __exit ib_umad_cleanup(void)
1393 ib_unregister_client(&umad_client);
1394 class_destroy(umad_class);
1395 unregister_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2);
1396 if (overflow_maj)
1397 unregister_chrdev_region(overflow_maj, IB_UMAD_MAX_PORTS * 2);
1400 module_init(ib_umad_init);
1401 module_exit(ib_umad_cleanup);