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.
6 *
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:
12 *
13 *     Redistribution and use in source and binary forms, with or
14 *     without modification, are permitted provided that the following
15 *     conditions are met:
16 *
17 *      - Redistributions of source code must retain the above
18 *        copyright notice, this list of conditions and the following
19 *        disclaimer.
20 *
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.
25 *
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.
34 */
35
36#define pr_fmt(fmt) "user_mad: " fmt
37
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>
52
53#include <asm/uaccess.h>
54
55#include <rdma/ib_mad.h>
56#include <rdma/ib_user_mad.h>
57
58MODULE_AUTHOR("Roland Dreier");
59MODULE_DESCRIPTION("InfiniBand userspace MAD packet access");
60MODULE_LICENSE("Dual BSD/GPL");
61
62enum {
63	IB_UMAD_MAX_PORTS  = 64,
64	IB_UMAD_MAX_AGENTS = 32,
65
66	IB_UMAD_MAJOR      = 231,
67	IB_UMAD_MINOR_BASE = 0
68};
69
70/*
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().
75 *
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().
80 *
81 * When destroying an ib_umad_device, we drop the module's reference.
82 */
83
84struct ib_umad_port {
85	struct cdev           cdev;
86	struct device	      *dev;
87
88	struct cdev           sm_cdev;
89	struct device	      *sm_dev;
90	struct semaphore       sm_sem;
91
92	struct mutex	       file_mutex;
93	struct list_head       file_list;
94
95	struct ib_device      *ib_dev;
96	struct ib_umad_device *umad_dev;
97	int                    dev_num;
98	u8                     port_num;
99};
100
101struct ib_umad_device {
102	int                  start_port, end_port;
103	struct kobject       kobj;
104	struct ib_umad_port  port[0];
105};
106
107struct ib_umad_file {
108	struct mutex		mutex;
109	struct ib_umad_port    *port;
110	struct list_head	recv_list;
111	struct list_head	send_list;
112	struct list_head	port_list;
113	spinlock_t		send_lock;
114	wait_queue_head_t	recv_wait;
115	struct ib_mad_agent    *agent[IB_UMAD_MAX_AGENTS];
116	int			agents_dead;
117	u8			use_pkey_index;
118	u8			already_used;
119};
120
121struct ib_umad_packet {
122	struct ib_mad_send_buf *msg;
123	struct ib_mad_recv_wc  *recv_wc;
124	struct list_head   list;
125	int		   length;
126	struct ib_user_mad mad;
127};
128
129static struct class *umad_class;
130
131static const dev_t base_dev = MKDEV(IB_UMAD_MAJOR, IB_UMAD_MINOR_BASE);
132
133static DEFINE_SPINLOCK(port_lock);
134static DECLARE_BITMAP(dev_map, IB_UMAD_MAX_PORTS);
135
136static void ib_umad_add_one(struct ib_device *device);
137static void ib_umad_remove_one(struct ib_device *device);
138
139static void ib_umad_release_dev(struct kobject *kobj)
140{
141	struct ib_umad_device *dev =
142		container_of(kobj, struct ib_umad_device, kobj);
143
144	kfree(dev);
145}
146
147static struct kobj_type ib_umad_dev_ktype = {
148	.release = ib_umad_release_dev,
149};
150
151static int hdr_size(struct ib_umad_file *file)
152{
153	return file->use_pkey_index ? sizeof (struct ib_user_mad_hdr) :
154		sizeof (struct ib_user_mad_hdr_old);
155}
156
157/* caller must hold file->mutex */
158static struct ib_mad_agent *__get_agent(struct ib_umad_file *file, int id)
159{
160	return file->agents_dead ? NULL : file->agent[id];
161}
162
163static int queue_packet(struct ib_umad_file *file,
164			struct ib_mad_agent *agent,
165			struct ib_umad_packet *packet)
166{
167	int ret = 1;
168
169	mutex_lock(&file->mutex);
170
171	for (packet->mad.hdr.id = 0;
172	     packet->mad.hdr.id < IB_UMAD_MAX_AGENTS;
173	     packet->mad.hdr.id++)
174		if (agent == __get_agent(file, packet->mad.hdr.id)) {
175			list_add_tail(&packet->list, &file->recv_list);
176			wake_up_interruptible(&file->recv_wait);
177			ret = 0;
178			break;
179		}
180
181	mutex_unlock(&file->mutex);
182
183	return ret;
184}
185
186static void dequeue_send(struct ib_umad_file *file,
187			 struct ib_umad_packet *packet)
188{
189	spin_lock_irq(&file->send_lock);
190	list_del(&packet->list);
191	spin_unlock_irq(&file->send_lock);
192}
193
194static void send_handler(struct ib_mad_agent *agent,
195			 struct ib_mad_send_wc *send_wc)
196{
197	struct ib_umad_file *file = agent->context;
198	struct ib_umad_packet *packet = send_wc->send_buf->context[0];
199
200	dequeue_send(file, packet);
201	ib_destroy_ah(packet->msg->ah);
202	ib_free_send_mad(packet->msg);
203
204	if (send_wc->status == IB_WC_RESP_TIMEOUT_ERR) {
205		packet->length = IB_MGMT_MAD_HDR;
206		packet->mad.hdr.status = ETIMEDOUT;
207		if (!queue_packet(file, agent, packet))
208			return;
209	}
210	kfree(packet);
211}
212
213static void recv_handler(struct ib_mad_agent *agent,
214			 struct ib_mad_recv_wc *mad_recv_wc)
215{
216	struct ib_umad_file *file = agent->context;
217	struct ib_umad_packet *packet;
218
219	if (mad_recv_wc->wc->status != IB_WC_SUCCESS)
220		goto err1;
221
222	packet = kzalloc(sizeof *packet, GFP_KERNEL);
223	if (!packet)
224		goto err1;
225
226	packet->length = mad_recv_wc->mad_len;
227	packet->recv_wc = mad_recv_wc;
228
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;
239
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);
243
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);
249	}
250
251	if (queue_packet(file, agent, packet))
252		goto err2;
253	return;
254
255err2:
256	kfree(packet);
257err1:
258	ib_free_recv_mad(mad_recv_wc);
259}
260
261static ssize_t copy_recv_mad(struct ib_umad_file *file, char __user *buf,
262			     struct ib_umad_packet *packet, size_t count)
263{
264	struct ib_mad_recv_buf *recv_buf;
265	int left, seg_payload, offset, max_seg_payload;
266
267	/* We need enough room to copy the first (or only) MAD segment. */
268	recv_buf = &packet->recv_wc->recv_buf;
269	if ((packet->length <= sizeof (*recv_buf->mad) &&
270	     count < hdr_size(file) + packet->length) ||
271	    (packet->length > sizeof (*recv_buf->mad) &&
272	     count < hdr_size(file) + sizeof (*recv_buf->mad)))
273		return -EINVAL;
274
275	if (copy_to_user(buf, &packet->mad, hdr_size(file)))
276		return -EFAULT;
277
278	buf += hdr_size(file);
279	seg_payload = min_t(int, packet->length, sizeof (*recv_buf->mad));
280	if (copy_to_user(buf, recv_buf->mad, seg_payload))
281		return -EFAULT;
282
283	if (seg_payload < packet->length) {
284		/*
285		 * Multipacket RMPP MAD message. Copy remainder of message.
286		 * Note that last segment may have a shorter payload.
287		 */
288		if (count < hdr_size(file) + packet->length) {
289			/*
290			 * The buffer is too small, return the first RMPP segment,
291			 * which includes the RMPP message length.
292			 */
293			return -ENOSPC;
294		}
295		offset = ib_get_mad_data_offset(recv_buf->mad->mad_hdr.mgmt_class);
296		max_seg_payload = sizeof (struct ib_mad) - offset;
297
298		for (left = packet->length - seg_payload, buf += seg_payload;
299		     left; left -= seg_payload, buf += seg_payload) {
300			recv_buf = container_of(recv_buf->list.next,
301						struct ib_mad_recv_buf, list);
302			seg_payload = min(left, max_seg_payload);
303			if (copy_to_user(buf, ((void *) recv_buf->mad) + offset,
304					 seg_payload))
305				return -EFAULT;
306		}
307	}
308	return hdr_size(file) + packet->length;
309}
310
311static ssize_t copy_send_mad(struct ib_umad_file *file, char __user *buf,
312			     struct ib_umad_packet *packet, size_t count)
313{
314	ssize_t size = hdr_size(file) + packet->length;
315
316	if (count < size)
317		return -EINVAL;
318
319	if (copy_to_user(buf, &packet->mad, hdr_size(file)))
320		return -EFAULT;
321
322	buf += hdr_size(file);
323
324	if (copy_to_user(buf, packet->mad.data, packet->length))
325		return -EFAULT;
326
327	return size;
328}
329
330static ssize_t ib_umad_read(struct file *filp, char __user *buf,
331			    size_t count, loff_t *pos)
332{
333	struct ib_umad_file *file = filp->private_data;
334	struct ib_umad_packet *packet;
335	ssize_t ret;
336
337	if (count < hdr_size(file))
338		return -EINVAL;
339
340	mutex_lock(&file->mutex);
341
342	while (list_empty(&file->recv_list)) {
343		mutex_unlock(&file->mutex);
344
345		if (filp->f_flags & O_NONBLOCK)
346			return -EAGAIN;
347
348		if (wait_event_interruptible(file->recv_wait,
349					     !list_empty(&file->recv_list)))
350			return -ERESTARTSYS;
351
352		mutex_lock(&file->mutex);
353	}
354
355	packet = list_entry(file->recv_list.next, struct ib_umad_packet, list);
356	list_del(&packet->list);
357
358	mutex_unlock(&file->mutex);
359
360	if (packet->recv_wc)
361		ret = copy_recv_mad(file, buf, packet, count);
362	else
363		ret = copy_send_mad(file, buf, packet, count);
364
365	if (ret < 0) {
366		/* Requeue packet */
367		mutex_lock(&file->mutex);
368		list_add(&packet->list, &file->recv_list);
369		mutex_unlock(&file->mutex);
370	} else {
371		if (packet->recv_wc)
372			ib_free_recv_mad(packet->recv_wc);
373		kfree(packet);
374	}
375	return ret;
376}
377
378static int copy_rmpp_mad(struct ib_mad_send_buf *msg, const char __user *buf)
379{
380	int left, seg;
381
382	/* Copy class specific header */
383	if ((msg->hdr_len > IB_MGMT_RMPP_HDR) &&
384	    copy_from_user(msg->mad + IB_MGMT_RMPP_HDR, buf + IB_MGMT_RMPP_HDR,
385			   msg->hdr_len - IB_MGMT_RMPP_HDR))
386		return -EFAULT;
387
388	/* All headers are in place.  Copy data segments. */
389	for (seg = 1, left = msg->data_len, buf += msg->hdr_len; left > 0;
390	     seg++, left -= msg->seg_size, buf += msg->seg_size) {
391		if (copy_from_user(ib_get_rmpp_segment(msg, seg), buf,
392				   min(left, msg->seg_size)))
393			return -EFAULT;
394	}
395	return 0;
396}
397
398static int same_destination(struct ib_user_mad_hdr *hdr1,
399			    struct ib_user_mad_hdr *hdr2)
400{
401	if (!hdr1->grh_present && !hdr2->grh_present)
402	   return (hdr1->lid == hdr2->lid);
403
404	if (hdr1->grh_present && hdr2->grh_present)
405	   return !memcmp(hdr1->gid, hdr2->gid, 16);
406
407	return 0;
408}
409
410static int is_duplicate(struct ib_umad_file *file,
411			struct ib_umad_packet *packet)
412{
413	struct ib_umad_packet *sent_packet;
414	struct ib_mad_hdr *sent_hdr, *hdr;
415
416	hdr = (struct ib_mad_hdr *) packet->mad.data;
417	list_for_each_entry(sent_packet, &file->send_list, list) {
418		sent_hdr = (struct ib_mad_hdr *) sent_packet->mad.data;
419
420		if ((hdr->tid != sent_hdr->tid) ||
421		    (hdr->mgmt_class != sent_hdr->mgmt_class))
422			continue;
423
424		/*
425		 * No need to be overly clever here.  If two new operations have
426		 * the same TID, reject the second as a duplicate.  This is more
427		 * restrictive than required by the spec.
428		 */
429		if (!ib_response_mad((struct ib_mad *) hdr)) {
430			if (!ib_response_mad((struct ib_mad *) sent_hdr))
431				return 1;
432			continue;
433		} else if (!ib_response_mad((struct ib_mad *) sent_hdr))
434			continue;
435
436		if (same_destination(&packet->mad.hdr, &sent_packet->mad.hdr))
437			return 1;
438	}
439
440	return 0;
441}
442
443static ssize_t ib_umad_write(struct file *filp, const char __user *buf,
444			     size_t count, loff_t *pos)
445{
446	struct ib_umad_file *file = filp->private_data;
447	struct ib_umad_packet *packet;
448	struct ib_mad_agent *agent;
449	struct ib_ah_attr ah_attr;
450	struct ib_ah *ah;
451	struct ib_rmpp_mad *rmpp_mad;
452	__be64 *tid;
453	int ret, data_len, hdr_len, copy_offset, rmpp_active;
454
455	if (count < hdr_size(file) + IB_MGMT_RMPP_HDR)
456		return -EINVAL;
457
458	packet = kzalloc(sizeof *packet + IB_MGMT_RMPP_HDR, GFP_KERNEL);
459	if (!packet)
460		return -ENOMEM;
461
462	if (copy_from_user(&packet->mad, buf, hdr_size(file))) {
463		ret = -EFAULT;
464		goto err;
465	}
466
467	if (packet->mad.hdr.id >= IB_UMAD_MAX_AGENTS) {
468		ret = -EINVAL;
469		goto err;
470	}
471
472	buf += hdr_size(file);
473
474	if (copy_from_user(packet->mad.data, buf, IB_MGMT_RMPP_HDR)) {
475		ret = -EFAULT;
476		goto err;
477	}
478
479	mutex_lock(&file->mutex);
480
481	agent = __get_agent(file, packet->mad.hdr.id);
482	if (!agent) {
483		ret = -EINVAL;
484		goto err_up;
485	}
486
487	memset(&ah_attr, 0, sizeof ah_attr);
488	ah_attr.dlid          = be16_to_cpu(packet->mad.hdr.lid);
489	ah_attr.sl            = packet->mad.hdr.sl;
490	ah_attr.src_path_bits = packet->mad.hdr.path_bits;
491	ah_attr.port_num      = file->port->port_num;
492	if (packet->mad.hdr.grh_present) {
493		ah_attr.ah_flags = IB_AH_GRH;
494		memcpy(ah_attr.grh.dgid.raw, packet->mad.hdr.gid, 16);
495		ah_attr.grh.sgid_index	   = packet->mad.hdr.gid_index;
496		ah_attr.grh.flow_label	   = be32_to_cpu(packet->mad.hdr.flow_label);
497		ah_attr.grh.hop_limit	   = packet->mad.hdr.hop_limit;
498		ah_attr.grh.traffic_class  = packet->mad.hdr.traffic_class;
499	}
500
501	ah = ib_create_ah(agent->qp->pd, &ah_attr);
502	if (IS_ERR(ah)) {
503		ret = PTR_ERR(ah);
504		goto err_up;
505	}
506
507	rmpp_mad = (struct ib_rmpp_mad *) packet->mad.data;
508	hdr_len = ib_get_mad_data_offset(rmpp_mad->mad_hdr.mgmt_class);
509
510	if (ib_is_mad_class_rmpp(rmpp_mad->mad_hdr.mgmt_class)
511	    && ib_mad_kernel_rmpp_agent(agent)) {
512		copy_offset = IB_MGMT_RMPP_HDR;
513		rmpp_active = ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) &
514						IB_MGMT_RMPP_FLAG_ACTIVE;
515	} else {
516		copy_offset = IB_MGMT_MAD_HDR;
517		rmpp_active = 0;
518	}
519
520	data_len = count - hdr_size(file) - hdr_len;
521	packet->msg = ib_create_send_mad(agent,
522					 be32_to_cpu(packet->mad.hdr.qpn),
523					 packet->mad.hdr.pkey_index, rmpp_active,
524					 hdr_len, data_len, GFP_KERNEL);
525	if (IS_ERR(packet->msg)) {
526		ret = PTR_ERR(packet->msg);
527		goto err_ah;
528	}
529
530	packet->msg->ah		= ah;
531	packet->msg->timeout_ms = packet->mad.hdr.timeout_ms;
532	packet->msg->retries	= packet->mad.hdr.retries;
533	packet->msg->context[0] = packet;
534
535	/* Copy MAD header.  Any RMPP header is already in place. */
536	memcpy(packet->msg->mad, packet->mad.data, IB_MGMT_MAD_HDR);
537
538	if (!rmpp_active) {
539		if (copy_from_user(packet->msg->mad + copy_offset,
540				   buf + copy_offset,
541				   hdr_len + data_len - copy_offset)) {
542			ret = -EFAULT;
543			goto err_msg;
544		}
545	} else {
546		ret = copy_rmpp_mad(packet->msg, buf);
547		if (ret)
548			goto err_msg;
549	}
550
551	/*
552	 * Set the high-order part of the transaction ID to make MADs from
553	 * different agents unique, and allow routing responses back to the
554	 * original requestor.
555	 */
556	if (!ib_response_mad(packet->msg->mad)) {
557		tid = &((struct ib_mad_hdr *) packet->msg->mad)->tid;
558		*tid = cpu_to_be64(((u64) agent->hi_tid) << 32 |
559				   (be64_to_cpup(tid) & 0xffffffff));
560		rmpp_mad->mad_hdr.tid = *tid;
561	}
562
563	if (!ib_mad_kernel_rmpp_agent(agent)
564	   && ib_is_mad_class_rmpp(rmpp_mad->mad_hdr.mgmt_class)
565	   && (ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) & IB_MGMT_RMPP_FLAG_ACTIVE)) {
566		spin_lock_irq(&file->send_lock);
567		list_add_tail(&packet->list, &file->send_list);
568		spin_unlock_irq(&file->send_lock);
569	} else {
570		spin_lock_irq(&file->send_lock);
571		ret = is_duplicate(file, packet);
572		if (!ret)
573			list_add_tail(&packet->list, &file->send_list);
574		spin_unlock_irq(&file->send_lock);
575		if (ret) {
576			ret = -EINVAL;
577			goto err_msg;
578		}
579	}
580
581	ret = ib_post_send_mad(packet->msg, NULL);
582	if (ret)
583		goto err_send;
584
585	mutex_unlock(&file->mutex);
586	return count;
587
588err_send:
589	dequeue_send(file, packet);
590err_msg:
591	ib_free_send_mad(packet->msg);
592err_ah:
593	ib_destroy_ah(ah);
594err_up:
595	mutex_unlock(&file->mutex);
596err:
597	kfree(packet);
598	return ret;
599}
600
601static unsigned int ib_umad_poll(struct file *filp, struct poll_table_struct *wait)
602{
603	struct ib_umad_file *file = filp->private_data;
604
605	/* we will always be able to post a MAD send */
606	unsigned int mask = POLLOUT | POLLWRNORM;
607
608	poll_wait(filp, &file->recv_wait, wait);
609
610	if (!list_empty(&file->recv_list))
611		mask |= POLLIN | POLLRDNORM;
612
613	return mask;
614}
615
616static int ib_umad_reg_agent(struct ib_umad_file *file, void __user *arg,
617			     int compat_method_mask)
618{
619	struct ib_user_mad_reg_req ureq;
620	struct ib_mad_reg_req req;
621	struct ib_mad_agent *agent = NULL;
622	int agent_id;
623	int ret;
624
625	mutex_lock(&file->port->file_mutex);
626	mutex_lock(&file->mutex);
627
628	if (!file->port->ib_dev) {
629		dev_notice(file->port->dev,
630			   "ib_umad_reg_agent: invalid device\n");
631		ret = -EPIPE;
632		goto out;
633	}
634
635	if (copy_from_user(&ureq, arg, sizeof ureq)) {
636		ret = -EFAULT;
637		goto out;
638	}
639
640	if (ureq.qpn != 0 && ureq.qpn != 1) {
641		dev_notice(file->port->dev,
642			   "ib_umad_reg_agent: invalid QPN %d specified\n",
643			   ureq.qpn);
644		ret = -EINVAL;
645		goto out;
646	}
647
648	for (agent_id = 0; agent_id < IB_UMAD_MAX_AGENTS; ++agent_id)
649		if (!__get_agent(file, agent_id))
650			goto found;
651
652	dev_notice(file->port->dev,
653		   "ib_umad_reg_agent: Max Agents (%u) reached\n",
654		   IB_UMAD_MAX_AGENTS);
655	ret = -ENOMEM;
656	goto out;
657
658found:
659	if (ureq.mgmt_class) {
660		memset(&req, 0, sizeof(req));
661		req.mgmt_class         = ureq.mgmt_class;
662		req.mgmt_class_version = ureq.mgmt_class_version;
663		memcpy(req.oui, ureq.oui, sizeof req.oui);
664
665		if (compat_method_mask) {
666			u32 *umm = (u32 *) ureq.method_mask;
667			int i;
668
669			for (i = 0; i < BITS_TO_LONGS(IB_MGMT_MAX_METHODS); ++i)
670				req.method_mask[i] =
671					umm[i * 2] | ((u64) umm[i * 2 + 1] << 32);
672		} else
673			memcpy(req.method_mask, ureq.method_mask,
674			       sizeof req.method_mask);
675	}
676
677	agent = ib_register_mad_agent(file->port->ib_dev, file->port->port_num,
678				      ureq.qpn ? IB_QPT_GSI : IB_QPT_SMI,
679				      ureq.mgmt_class ? &req : NULL,
680				      ureq.rmpp_version,
681				      send_handler, recv_handler, file, 0);
682	if (IS_ERR(agent)) {
683		ret = PTR_ERR(agent);
684		agent = NULL;
685		goto out;
686	}
687
688	if (put_user(agent_id,
689		     (u32 __user *) (arg + offsetof(struct ib_user_mad_reg_req, id)))) {
690		ret = -EFAULT;
691		goto out;
692	}
693
694	if (!file->already_used) {
695		file->already_used = 1;
696		if (!file->use_pkey_index) {
697			dev_warn(file->port->dev,
698				"process %s did not enable P_Key index support.\n",
699				current->comm);
700			dev_warn(file->port->dev,
701				"   Documentation/infiniband/user_mad.txt has info on the new ABI.\n");
702		}
703	}
704
705	file->agent[agent_id] = agent;
706	ret = 0;
707
708out:
709	mutex_unlock(&file->mutex);
710
711	if (ret && agent)
712		ib_unregister_mad_agent(agent);
713
714	mutex_unlock(&file->port->file_mutex);
715
716	return ret;
717}
718
719static int ib_umad_reg_agent2(struct ib_umad_file *file, void __user *arg)
720{
721	struct ib_user_mad_reg_req2 ureq;
722	struct ib_mad_reg_req req;
723	struct ib_mad_agent *agent = NULL;
724	int agent_id;
725	int ret;
726
727	mutex_lock(&file->port->file_mutex);
728	mutex_lock(&file->mutex);
729
730	if (!file->port->ib_dev) {
731		dev_notice(file->port->dev,
732			   "ib_umad_reg_agent2: invalid device\n");
733		ret = -EPIPE;
734		goto out;
735	}
736
737	if (copy_from_user(&ureq, arg, sizeof(ureq))) {
738		ret = -EFAULT;
739		goto out;
740	}
741
742	if (ureq.qpn != 0 && ureq.qpn != 1) {
743		dev_notice(file->port->dev,
744			   "ib_umad_reg_agent2: invalid QPN %d specified\n",
745			   ureq.qpn);
746		ret = -EINVAL;
747		goto out;
748	}
749
750	if (ureq.flags & ~IB_USER_MAD_REG_FLAGS_CAP) {
751		dev_notice(file->port->dev,
752			   "ib_umad_reg_agent2 failed: invalid registration flags specified 0x%x; supported 0x%x\n",
753			   ureq.flags, IB_USER_MAD_REG_FLAGS_CAP);
754		ret = -EINVAL;
755
756		if (put_user((u32)IB_USER_MAD_REG_FLAGS_CAP,
757				(u32 __user *) (arg + offsetof(struct
758				ib_user_mad_reg_req2, flags))))
759			ret = -EFAULT;
760
761		goto out;
762	}
763
764	for (agent_id = 0; agent_id < IB_UMAD_MAX_AGENTS; ++agent_id)
765		if (!__get_agent(file, agent_id))
766			goto found;
767
768	dev_notice(file->port->dev,
769		   "ib_umad_reg_agent2: Max Agents (%u) reached\n",
770		   IB_UMAD_MAX_AGENTS);
771	ret = -ENOMEM;
772	goto out;
773
774found:
775	if (ureq.mgmt_class) {
776		memset(&req, 0, sizeof(req));
777		req.mgmt_class         = ureq.mgmt_class;
778		req.mgmt_class_version = ureq.mgmt_class_version;
779		if (ureq.oui & 0xff000000) {
780			dev_notice(file->port->dev,
781				   "ib_umad_reg_agent2 failed: oui invalid 0x%08x\n",
782				   ureq.oui);
783			ret = -EINVAL;
784			goto out;
785		}
786		req.oui[2] =  ureq.oui & 0x0000ff;
787		req.oui[1] = (ureq.oui & 0x00ff00) >> 8;
788		req.oui[0] = (ureq.oui & 0xff0000) >> 16;
789		memcpy(req.method_mask, ureq.method_mask,
790			sizeof(req.method_mask));
791	}
792
793	agent = ib_register_mad_agent(file->port->ib_dev, file->port->port_num,
794				      ureq.qpn ? IB_QPT_GSI : IB_QPT_SMI,
795				      ureq.mgmt_class ? &req : NULL,
796				      ureq.rmpp_version,
797				      send_handler, recv_handler, file,
798				      ureq.flags);
799	if (IS_ERR(agent)) {
800		ret = PTR_ERR(agent);
801		agent = NULL;
802		goto out;
803	}
804
805	if (put_user(agent_id,
806		     (u32 __user *)(arg +
807				offsetof(struct ib_user_mad_reg_req2, id)))) {
808		ret = -EFAULT;
809		goto out;
810	}
811
812	if (!file->already_used) {
813		file->already_used = 1;
814		file->use_pkey_index = 1;
815	}
816
817	file->agent[agent_id] = agent;
818	ret = 0;
819
820out:
821	mutex_unlock(&file->mutex);
822
823	if (ret && agent)
824		ib_unregister_mad_agent(agent);
825
826	mutex_unlock(&file->port->file_mutex);
827
828	return ret;
829}
830
831
832static int ib_umad_unreg_agent(struct ib_umad_file *file, u32 __user *arg)
833{
834	struct ib_mad_agent *agent = NULL;
835	u32 id;
836	int ret = 0;
837
838	if (get_user(id, arg))
839		return -EFAULT;
840
841	mutex_lock(&file->port->file_mutex);
842	mutex_lock(&file->mutex);
843
844	if (id >= IB_UMAD_MAX_AGENTS || !__get_agent(file, id)) {
845		ret = -EINVAL;
846		goto out;
847	}
848
849	agent = file->agent[id];
850	file->agent[id] = NULL;
851
852out:
853	mutex_unlock(&file->mutex);
854
855	if (agent)
856		ib_unregister_mad_agent(agent);
857
858	mutex_unlock(&file->port->file_mutex);
859
860	return ret;
861}
862
863static long ib_umad_enable_pkey(struct ib_umad_file *file)
864{
865	int ret = 0;
866
867	mutex_lock(&file->mutex);
868	if (file->already_used)
869		ret = -EINVAL;
870	else
871		file->use_pkey_index = 1;
872	mutex_unlock(&file->mutex);
873
874	return ret;
875}
876
877static long ib_umad_ioctl(struct file *filp, unsigned int cmd,
878			  unsigned long arg)
879{
880	switch (cmd) {
881	case IB_USER_MAD_REGISTER_AGENT:
882		return ib_umad_reg_agent(filp->private_data, (void __user *) arg, 0);
883	case IB_USER_MAD_UNREGISTER_AGENT:
884		return ib_umad_unreg_agent(filp->private_data, (__u32 __user *) arg);
885	case IB_USER_MAD_ENABLE_PKEY:
886		return ib_umad_enable_pkey(filp->private_data);
887	case IB_USER_MAD_REGISTER_AGENT2:
888		return ib_umad_reg_agent2(filp->private_data, (void __user *) arg);
889	default:
890		return -ENOIOCTLCMD;
891	}
892}
893
894#ifdef CONFIG_COMPAT
895static long ib_umad_compat_ioctl(struct file *filp, unsigned int cmd,
896				 unsigned long arg)
897{
898	switch (cmd) {
899	case IB_USER_MAD_REGISTER_AGENT:
900		return ib_umad_reg_agent(filp->private_data, compat_ptr(arg), 1);
901	case IB_USER_MAD_UNREGISTER_AGENT:
902		return ib_umad_unreg_agent(filp->private_data, compat_ptr(arg));
903	case IB_USER_MAD_ENABLE_PKEY:
904		return ib_umad_enable_pkey(filp->private_data);
905	case IB_USER_MAD_REGISTER_AGENT2:
906		return ib_umad_reg_agent2(filp->private_data, compat_ptr(arg));
907	default:
908		return -ENOIOCTLCMD;
909	}
910}
911#endif
912
913/*
914 * ib_umad_open() does not need the BKL:
915 *
916 *  - the ib_umad_port structures are properly reference counted, and
917 *    everything else is purely local to the file being created, so
918 *    races against other open calls are not a problem;
919 *  - the ioctl method does not affect any global state outside of the
920 *    file structure being operated on;
921 */
922static int ib_umad_open(struct inode *inode, struct file *filp)
923{
924	struct ib_umad_port *port;
925	struct ib_umad_file *file;
926	int ret = -ENXIO;
927
928	port = container_of(inode->i_cdev, struct ib_umad_port, cdev);
929
930	mutex_lock(&port->file_mutex);
931
932	if (!port->ib_dev)
933		goto out;
934
935	ret = -ENOMEM;
936	file = kzalloc(sizeof *file, GFP_KERNEL);
937	if (!file)
938		goto out;
939
940	mutex_init(&file->mutex);
941	spin_lock_init(&file->send_lock);
942	INIT_LIST_HEAD(&file->recv_list);
943	INIT_LIST_HEAD(&file->send_list);
944	init_waitqueue_head(&file->recv_wait);
945
946	file->port = port;
947	filp->private_data = file;
948
949	list_add_tail(&file->port_list, &port->file_list);
950
951	ret = nonseekable_open(inode, filp);
952	if (ret) {
953		list_del(&file->port_list);
954		kfree(file);
955		goto out;
956	}
957
958	kobject_get(&port->umad_dev->kobj);
959
960out:
961	mutex_unlock(&port->file_mutex);
962	return ret;
963}
964
965static int ib_umad_close(struct inode *inode, struct file *filp)
966{
967	struct ib_umad_file *file = filp->private_data;
968	struct ib_umad_device *dev = file->port->umad_dev;
969	struct ib_umad_packet *packet, *tmp;
970	int already_dead;
971	int i;
972
973	mutex_lock(&file->port->file_mutex);
974	mutex_lock(&file->mutex);
975
976	already_dead = file->agents_dead;
977	file->agents_dead = 1;
978
979	list_for_each_entry_safe(packet, tmp, &file->recv_list, list) {
980		if (packet->recv_wc)
981			ib_free_recv_mad(packet->recv_wc);
982		kfree(packet);
983	}
984
985	list_del(&file->port_list);
986
987	mutex_unlock(&file->mutex);
988
989	if (!already_dead)
990		for (i = 0; i < IB_UMAD_MAX_AGENTS; ++i)
991			if (file->agent[i])
992				ib_unregister_mad_agent(file->agent[i]);
993
994	mutex_unlock(&file->port->file_mutex);
995
996	kfree(file);
997	kobject_put(&dev->kobj);
998
999	return 0;
1000}
1001
1002static const struct file_operations umad_fops = {
1003	.owner		= THIS_MODULE,
1004	.read		= ib_umad_read,
1005	.write		= ib_umad_write,
1006	.poll		= ib_umad_poll,
1007	.unlocked_ioctl = ib_umad_ioctl,
1008#ifdef CONFIG_COMPAT
1009	.compat_ioctl	= ib_umad_compat_ioctl,
1010#endif
1011	.open		= ib_umad_open,
1012	.release	= ib_umad_close,
1013	.llseek		= no_llseek,
1014};
1015
1016static int ib_umad_sm_open(struct inode *inode, struct file *filp)
1017{
1018	struct ib_umad_port *port;
1019	struct ib_port_modify props = {
1020		.set_port_cap_mask = IB_PORT_SM
1021	};
1022	int ret;
1023
1024	port = container_of(inode->i_cdev, struct ib_umad_port, sm_cdev);
1025
1026	if (filp->f_flags & O_NONBLOCK) {
1027		if (down_trylock(&port->sm_sem)) {
1028			ret = -EAGAIN;
1029			goto fail;
1030		}
1031	} else {
1032		if (down_interruptible(&port->sm_sem)) {
1033			ret = -ERESTARTSYS;
1034			goto fail;
1035		}
1036	}
1037
1038	ret = ib_modify_port(port->ib_dev, port->port_num, 0, &props);
1039	if (ret)
1040		goto err_up_sem;
1041
1042	filp->private_data = port;
1043
1044	ret = nonseekable_open(inode, filp);
1045	if (ret)
1046		goto err_clr_sm_cap;
1047
1048	kobject_get(&port->umad_dev->kobj);
1049
1050	return 0;
1051
1052err_clr_sm_cap:
1053	swap(props.set_port_cap_mask, props.clr_port_cap_mask);
1054	ib_modify_port(port->ib_dev, port->port_num, 0, &props);
1055
1056err_up_sem:
1057	up(&port->sm_sem);
1058
1059fail:
1060	return ret;
1061}
1062
1063static int ib_umad_sm_close(struct inode *inode, struct file *filp)
1064{
1065	struct ib_umad_port *port = filp->private_data;
1066	struct ib_port_modify props = {
1067		.clr_port_cap_mask = IB_PORT_SM
1068	};
1069	int ret = 0;
1070
1071	mutex_lock(&port->file_mutex);
1072	if (port->ib_dev)
1073		ret = ib_modify_port(port->ib_dev, port->port_num, 0, &props);
1074	mutex_unlock(&port->file_mutex);
1075
1076	up(&port->sm_sem);
1077
1078	kobject_put(&port->umad_dev->kobj);
1079
1080	return ret;
1081}
1082
1083static const struct file_operations umad_sm_fops = {
1084	.owner	 = THIS_MODULE,
1085	.open	 = ib_umad_sm_open,
1086	.release = ib_umad_sm_close,
1087	.llseek	 = no_llseek,
1088};
1089
1090static struct ib_client umad_client = {
1091	.name   = "umad",
1092	.add    = ib_umad_add_one,
1093	.remove = ib_umad_remove_one
1094};
1095
1096static ssize_t show_ibdev(struct device *dev, struct device_attribute *attr,
1097			  char *buf)
1098{
1099	struct ib_umad_port *port = dev_get_drvdata(dev);
1100
1101	if (!port)
1102		return -ENODEV;
1103
1104	return sprintf(buf, "%s\n", port->ib_dev->name);
1105}
1106static DEVICE_ATTR(ibdev, S_IRUGO, show_ibdev, NULL);
1107
1108static ssize_t show_port(struct device *dev, struct device_attribute *attr,
1109			 char *buf)
1110{
1111	struct ib_umad_port *port = dev_get_drvdata(dev);
1112
1113	if (!port)
1114		return -ENODEV;
1115
1116	return sprintf(buf, "%d\n", port->port_num);
1117}
1118static DEVICE_ATTR(port, S_IRUGO, show_port, NULL);
1119
1120static CLASS_ATTR_STRING(abi_version, S_IRUGO,
1121			 __stringify(IB_USER_MAD_ABI_VERSION));
1122
1123static dev_t overflow_maj;
1124static DECLARE_BITMAP(overflow_map, IB_UMAD_MAX_PORTS);
1125static int find_overflow_devnum(struct ib_device *device)
1126{
1127	int ret;
1128
1129	if (!overflow_maj) {
1130		ret = alloc_chrdev_region(&overflow_maj, 0, IB_UMAD_MAX_PORTS * 2,
1131					  "infiniband_mad");
1132		if (ret) {
1133			dev_err(&device->dev,
1134				"couldn't register dynamic device number\n");
1135			return ret;
1136		}
1137	}
1138
1139	ret = find_first_zero_bit(overflow_map, IB_UMAD_MAX_PORTS);
1140	if (ret >= IB_UMAD_MAX_PORTS)
1141		return -1;
1142
1143	return ret;
1144}
1145
1146static int ib_umad_init_port(struct ib_device *device, int port_num,
1147			     struct ib_umad_device *umad_dev,
1148			     struct ib_umad_port *port)
1149{
1150	int devnum;
1151	dev_t base;
1152
1153	spin_lock(&port_lock);
1154	devnum = find_first_zero_bit(dev_map, IB_UMAD_MAX_PORTS);
1155	if (devnum >= IB_UMAD_MAX_PORTS) {
1156		spin_unlock(&port_lock);
1157		devnum = find_overflow_devnum(device);
1158		if (devnum < 0)
1159			return -1;
1160
1161		spin_lock(&port_lock);
1162		port->dev_num = devnum + IB_UMAD_MAX_PORTS;
1163		base = devnum + overflow_maj;
1164		set_bit(devnum, overflow_map);
1165	} else {
1166		port->dev_num = devnum;
1167		base = devnum + base_dev;
1168		set_bit(devnum, dev_map);
1169	}
1170	spin_unlock(&port_lock);
1171
1172	port->ib_dev   = device;
1173	port->port_num = port_num;
1174	sema_init(&port->sm_sem, 1);
1175	mutex_init(&port->file_mutex);
1176	INIT_LIST_HEAD(&port->file_list);
1177
1178	cdev_init(&port->cdev, &umad_fops);
1179	port->cdev.owner = THIS_MODULE;
1180	port->cdev.kobj.parent = &umad_dev->kobj;
1181	kobject_set_name(&port->cdev.kobj, "umad%d", port->dev_num);
1182	if (cdev_add(&port->cdev, base, 1))
1183		goto err_cdev;
1184
1185	port->dev = device_create(umad_class, device->dma_device,
1186				  port->cdev.dev, port,
1187				  "umad%d", port->dev_num);
1188	if (IS_ERR(port->dev))
1189		goto err_cdev;
1190
1191	if (device_create_file(port->dev, &dev_attr_ibdev))
1192		goto err_dev;
1193	if (device_create_file(port->dev, &dev_attr_port))
1194		goto err_dev;
1195
1196	base += IB_UMAD_MAX_PORTS;
1197	cdev_init(&port->sm_cdev, &umad_sm_fops);
1198	port->sm_cdev.owner = THIS_MODULE;
1199	port->sm_cdev.kobj.parent = &umad_dev->kobj;
1200	kobject_set_name(&port->sm_cdev.kobj, "issm%d", port->dev_num);
1201	if (cdev_add(&port->sm_cdev, base, 1))
1202		goto err_sm_cdev;
1203
1204	port->sm_dev = device_create(umad_class, device->dma_device,
1205				     port->sm_cdev.dev, port,
1206				     "issm%d", port->dev_num);
1207	if (IS_ERR(port->sm_dev))
1208		goto err_sm_cdev;
1209
1210	if (device_create_file(port->sm_dev, &dev_attr_ibdev))
1211		goto err_sm_dev;
1212	if (device_create_file(port->sm_dev, &dev_attr_port))
1213		goto err_sm_dev;
1214
1215	return 0;
1216
1217err_sm_dev:
1218	device_destroy(umad_class, port->sm_cdev.dev);
1219
1220err_sm_cdev:
1221	cdev_del(&port->sm_cdev);
1222
1223err_dev:
1224	device_destroy(umad_class, port->cdev.dev);
1225
1226err_cdev:
1227	cdev_del(&port->cdev);
1228	if (port->dev_num < IB_UMAD_MAX_PORTS)
1229		clear_bit(devnum, dev_map);
1230	else
1231		clear_bit(devnum, overflow_map);
1232
1233	return -1;
1234}
1235
1236static void ib_umad_kill_port(struct ib_umad_port *port)
1237{
1238	struct ib_umad_file *file;
1239	int id;
1240
1241	dev_set_drvdata(port->dev,    NULL);
1242	dev_set_drvdata(port->sm_dev, NULL);
1243
1244	device_destroy(umad_class, port->cdev.dev);
1245	device_destroy(umad_class, port->sm_cdev.dev);
1246
1247	cdev_del(&port->cdev);
1248	cdev_del(&port->sm_cdev);
1249
1250	mutex_lock(&port->file_mutex);
1251
1252	port->ib_dev = NULL;
1253
1254	list_for_each_entry(file, &port->file_list, port_list) {
1255		mutex_lock(&file->mutex);
1256		file->agents_dead = 1;
1257		mutex_unlock(&file->mutex);
1258
1259		for (id = 0; id < IB_UMAD_MAX_AGENTS; ++id)
1260			if (file->agent[id])
1261				ib_unregister_mad_agent(file->agent[id]);
1262	}
1263
1264	mutex_unlock(&port->file_mutex);
1265
1266	if (port->dev_num < IB_UMAD_MAX_PORTS)
1267		clear_bit(port->dev_num, dev_map);
1268	else
1269		clear_bit(port->dev_num - IB_UMAD_MAX_PORTS, overflow_map);
1270}
1271
1272static void ib_umad_add_one(struct ib_device *device)
1273{
1274	struct ib_umad_device *umad_dev;
1275	int s, e, i;
1276
1277	if (rdma_node_get_transport(device->node_type) != RDMA_TRANSPORT_IB)
1278		return;
1279
1280	if (device->node_type == RDMA_NODE_IB_SWITCH)
1281		s = e = 0;
1282	else {
1283		s = 1;
1284		e = device->phys_port_cnt;
1285	}
1286
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;
1292
1293	kobject_init(&umad_dev->kobj, &ib_umad_dev_ktype);
1294
1295	umad_dev->start_port = s;
1296	umad_dev->end_port   = e;
1297
1298	for (i = s; i <= e; ++i) {
1299		umad_dev->port[i - s].umad_dev = umad_dev;
1300
1301		if (ib_umad_init_port(device, i, umad_dev,
1302				      &umad_dev->port[i - s]))
1303			goto err;
1304	}
1305
1306	ib_set_client_data(device, &umad_client, umad_dev);
1307
1308	return;
1309
1310err:
1311	while (--i >= s)
1312		ib_umad_kill_port(&umad_dev->port[i - s]);
1313
1314	kobject_put(&umad_dev->kobj);
1315}
1316
1317static void ib_umad_remove_one(struct ib_device *device)
1318{
1319	struct ib_umad_device *umad_dev = ib_get_client_data(device, &umad_client);
1320	int i;
1321
1322	if (!umad_dev)
1323		return;
1324
1325	for (i = 0; i <= umad_dev->end_port - umad_dev->start_port; ++i)
1326		ib_umad_kill_port(&umad_dev->port[i]);
1327
1328	kobject_put(&umad_dev->kobj);
1329}
1330
1331static char *umad_devnode(struct device *dev, umode_t *mode)
1332{
1333	return kasprintf(GFP_KERNEL, "infiniband/%s", dev_name(dev));
1334}
1335
1336static int __init ib_umad_init(void)
1337{
1338	int ret;
1339
1340	ret = register_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2,
1341				     "infiniband_mad");
1342	if (ret) {
1343		pr_err("couldn't register device number\n");
1344		goto out;
1345	}
1346
1347	umad_class = class_create(THIS_MODULE, "infiniband_mad");
1348	if (IS_ERR(umad_class)) {
1349		ret = PTR_ERR(umad_class);
1350		pr_err("couldn't create class infiniband_mad\n");
1351		goto out_chrdev;
1352	}
1353
1354	umad_class->devnode = umad_devnode;
1355
1356	ret = class_create_file(umad_class, &class_attr_abi_version.attr);
1357	if (ret) {
1358		pr_err("couldn't create abi_version attribute\n");
1359		goto out_class;
1360	}
1361
1362	ret = ib_register_client(&umad_client);
1363	if (ret) {
1364		pr_err("couldn't register ib_umad client\n");
1365		goto out_class;
1366	}
1367
1368	return 0;
1369
1370out_class:
1371	class_destroy(umad_class);
1372
1373out_chrdev:
1374	unregister_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2);
1375
1376out:
1377	return ret;
1378}
1379
1380static void __exit ib_umad_cleanup(void)
1381{
1382	ib_unregister_client(&umad_client);
1383	class_destroy(umad_class);
1384	unregister_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2);
1385	if (overflow_maj)
1386		unregister_chrdev_region(overflow_maj, IB_UMAD_MAX_PORTS * 2);
1387}
1388
1389module_init(ib_umad_init);
1390module_exit(ib_umad_cleanup);
1391