user_mad.c revision 2f76e82947b977a1008cfd2868351a701c93c69c
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 *
6 * This software is available to you under a choice of one of two
7 * licenses.  You may choose to be licensed under the terms of the GNU
8 * General Public License (GPL) Version 2, available from the file
9 * COPYING in the main directory of this source tree, or the
10 * OpenIB.org BSD license below:
11 *
12 *     Redistribution and use in source and binary forms, with or
13 *     without modification, are permitted provided that the following
14 *     conditions are met:
15 *
16 *      - Redistributions of source code must retain the above
17 *        copyright notice, this list of conditions and the following
18 *        disclaimer.
19 *
20 *      - Redistributions in binary form must reproduce the above
21 *        copyright notice, this list of conditions and the following
22 *        disclaimer in the documentation and/or other materials
23 *        provided with the distribution.
24 *
25 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
26 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
27 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
28 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
29 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
30 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
31 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
32 * SOFTWARE.
33 *
34 * $Id: user_mad.c 2814 2005-07-06 19:14:09Z halr $
35 */
36
37#include <linux/module.h>
38#include <linux/init.h>
39#include <linux/device.h>
40#include <linux/err.h>
41#include <linux/fs.h>
42#include <linux/cdev.h>
43#include <linux/pci.h>
44#include <linux/dma-mapping.h>
45#include <linux/poll.h>
46#include <linux/rwsem.h>
47#include <linux/kref.h>
48
49#include <asm/uaccess.h>
50#include <asm/semaphore.h>
51
52#include <rdma/ib_mad.h>
53#include <rdma/ib_user_mad.h>
54
55MODULE_AUTHOR("Roland Dreier");
56MODULE_DESCRIPTION("InfiniBand userspace MAD packet access");
57MODULE_LICENSE("Dual BSD/GPL");
58
59enum {
60	IB_UMAD_MAX_PORTS  = 64,
61	IB_UMAD_MAX_AGENTS = 32,
62
63	IB_UMAD_MAJOR      = 231,
64	IB_UMAD_MINOR_BASE = 0
65};
66
67/*
68 * Our lifetime rules for these structs are the following: each time a
69 * device special file is opened, we look up the corresponding struct
70 * ib_umad_port by minor in the umad_port[] table while holding the
71 * port_lock.  If this lookup succeeds, we take a reference on the
72 * ib_umad_port's struct ib_umad_device while still holding the
73 * port_lock; if the lookup fails, we fail the open().  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 clear all of its
82 * ib_umad_ports from umad_port[] while holding port_lock before
83 * dropping the module's reference to the ib_umad_device.  This is
84 * always safe because any open() calls will either succeed and obtain
85 * a reference before we clear the umad_port[] entries, or fail after
86 * we clear the umad_port[] entries.
87 */
88
89struct ib_umad_port {
90	struct cdev           *dev;
91	struct class_device   *class_dev;
92
93	struct cdev           *sm_dev;
94	struct class_device   *sm_class_dev;
95	struct semaphore       sm_sem;
96
97	struct rw_semaphore    mutex;
98	struct list_head       file_list;
99
100	struct ib_device      *ib_dev;
101	struct ib_umad_device *umad_dev;
102	int                    dev_num;
103	u8                     port_num;
104};
105
106struct ib_umad_device {
107	int                  start_port, end_port;
108	struct kref          ref;
109	struct ib_umad_port  port[0];
110};
111
112struct ib_umad_file {
113	struct ib_umad_port *port;
114	struct list_head     recv_list;
115	struct list_head     port_list;
116	spinlock_t           recv_lock;
117	wait_queue_head_t    recv_wait;
118	struct ib_mad_agent *agent[IB_UMAD_MAX_AGENTS];
119	struct ib_mr        *mr[IB_UMAD_MAX_AGENTS];
120};
121
122struct ib_umad_packet {
123	struct ib_mad_send_buf *msg;
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 struct ib_umad_port *umad_port[IB_UMAD_MAX_PORTS];
135static DECLARE_BITMAP(dev_map, IB_UMAD_MAX_PORTS * 2);
136
137static void ib_umad_add_one(struct ib_device *device);
138static void ib_umad_remove_one(struct ib_device *device);
139
140static void ib_umad_release_dev(struct kref *ref)
141{
142	struct ib_umad_device *dev =
143		container_of(ref, struct ib_umad_device, ref);
144
145	kfree(dev);
146}
147
148static int queue_packet(struct ib_umad_file *file,
149			struct ib_mad_agent *agent,
150			struct ib_umad_packet *packet)
151{
152	int ret = 1;
153
154	down_read(&file->port->mutex);
155	for (packet->mad.hdr.id = 0;
156	     packet->mad.hdr.id < IB_UMAD_MAX_AGENTS;
157	     packet->mad.hdr.id++)
158		if (agent == file->agent[packet->mad.hdr.id]) {
159			spin_lock_irq(&file->recv_lock);
160			list_add_tail(&packet->list, &file->recv_list);
161			spin_unlock_irq(&file->recv_lock);
162			wake_up_interruptible(&file->recv_wait);
163			ret = 0;
164			break;
165		}
166
167	up_read(&file->port->mutex);
168
169	return ret;
170}
171
172static void send_handler(struct ib_mad_agent *agent,
173			 struct ib_mad_send_wc *send_wc)
174{
175	struct ib_umad_file *file = agent->context;
176	struct ib_umad_packet *timeout;
177	struct ib_umad_packet *packet = send_wc->send_buf->context[0];
178
179	ib_destroy_ah(packet->msg->ah);
180	ib_free_send_mad(packet->msg);
181
182	if (send_wc->status == IB_WC_RESP_TIMEOUT_ERR) {
183		timeout = kzalloc(sizeof *timeout + IB_MGMT_MAD_HDR, GFP_KERNEL);
184		if (!timeout)
185			goto out;
186
187		timeout->length 	= IB_MGMT_MAD_HDR;
188		timeout->mad.hdr.id 	= packet->mad.hdr.id;
189		timeout->mad.hdr.status = ETIMEDOUT;
190		memcpy(timeout->mad.data, packet->mad.data,
191		       sizeof (struct ib_mad_hdr));
192
193		if (!queue_packet(file, agent, timeout))
194				return;
195	}
196out:
197	kfree(packet);
198}
199
200static void recv_handler(struct ib_mad_agent *agent,
201			 struct ib_mad_recv_wc *mad_recv_wc)
202{
203	struct ib_umad_file *file = agent->context;
204	struct ib_umad_packet *packet;
205	int length;
206
207	if (mad_recv_wc->wc->status != IB_WC_SUCCESS)
208		goto out;
209
210	length = mad_recv_wc->mad_len;
211	packet = kzalloc(sizeof *packet + length, GFP_KERNEL);
212	if (!packet)
213		goto out;
214
215	packet->length = length;
216
217	ib_coalesce_recv_mad(mad_recv_wc, packet->mad.data);
218
219	packet->mad.hdr.status    = 0;
220	packet->mad.hdr.length    = length + sizeof (struct ib_user_mad);
221	packet->mad.hdr.qpn 	  = cpu_to_be32(mad_recv_wc->wc->src_qp);
222	packet->mad.hdr.lid 	  = cpu_to_be16(mad_recv_wc->wc->slid);
223	packet->mad.hdr.sl  	  = mad_recv_wc->wc->sl;
224	packet->mad.hdr.path_bits = mad_recv_wc->wc->dlid_path_bits;
225	packet->mad.hdr.grh_present = !!(mad_recv_wc->wc->wc_flags & IB_WC_GRH);
226	if (packet->mad.hdr.grh_present) {
227		/* XXX parse GRH */
228		packet->mad.hdr.gid_index 	= 0;
229		packet->mad.hdr.hop_limit 	= 0;
230		packet->mad.hdr.traffic_class	= 0;
231		memset(packet->mad.hdr.gid, 0, 16);
232		packet->mad.hdr.flow_label	= 0;
233	}
234
235	if (queue_packet(file, agent, packet))
236		kfree(packet);
237
238out:
239	ib_free_recv_mad(mad_recv_wc);
240}
241
242static ssize_t ib_umad_read(struct file *filp, char __user *buf,
243			    size_t count, loff_t *pos)
244{
245	struct ib_umad_file *file = filp->private_data;
246	struct ib_umad_packet *packet;
247	ssize_t ret;
248
249	if (count < sizeof (struct ib_user_mad) + sizeof (struct ib_mad))
250		return -EINVAL;
251
252	spin_lock_irq(&file->recv_lock);
253
254	while (list_empty(&file->recv_list)) {
255		spin_unlock_irq(&file->recv_lock);
256
257		if (filp->f_flags & O_NONBLOCK)
258			return -EAGAIN;
259
260		if (wait_event_interruptible(file->recv_wait,
261					     !list_empty(&file->recv_list)))
262			return -ERESTARTSYS;
263
264		spin_lock_irq(&file->recv_lock);
265	}
266
267	packet = list_entry(file->recv_list.next, struct ib_umad_packet, list);
268	list_del(&packet->list);
269
270	spin_unlock_irq(&file->recv_lock);
271
272	if (count < packet->length + sizeof (struct ib_user_mad)) {
273		/* Return length needed (and first RMPP segment) if too small */
274		if (copy_to_user(buf, &packet->mad,
275				 sizeof (struct ib_user_mad) + sizeof (struct ib_mad)))
276			ret = -EFAULT;
277		else
278			ret = -ENOSPC;
279	} else if (copy_to_user(buf, &packet->mad,
280				packet->length + sizeof (struct ib_user_mad)))
281		ret = -EFAULT;
282	else
283		ret = packet->length + sizeof (struct ib_user_mad);
284	if (ret < 0) {
285		/* Requeue packet */
286		spin_lock_irq(&file->recv_lock);
287		list_add(&packet->list, &file->recv_list);
288		spin_unlock_irq(&file->recv_lock);
289	} else
290		kfree(packet);
291	return ret;
292}
293
294static ssize_t ib_umad_write(struct file *filp, const char __user *buf,
295			     size_t count, loff_t *pos)
296{
297	struct ib_umad_file *file = filp->private_data;
298	struct ib_umad_packet *packet;
299	struct ib_mad_agent *agent;
300	struct ib_ah_attr ah_attr;
301	struct ib_ah *ah;
302	struct ib_rmpp_mad *rmpp_mad;
303	u8 method;
304	__be64 *tid;
305	int ret, length, hdr_len, copy_offset;
306	int rmpp_active = 0;
307
308	if (count < sizeof (struct ib_user_mad))
309		return -EINVAL;
310
311	length = count - sizeof (struct ib_user_mad);
312	packet = kmalloc(sizeof *packet + IB_MGMT_RMPP_HDR, GFP_KERNEL);
313	if (!packet)
314		return -ENOMEM;
315
316	if (copy_from_user(&packet->mad, buf,
317			    sizeof (struct ib_user_mad) + IB_MGMT_RMPP_HDR)) {
318		ret = -EFAULT;
319		goto err;
320	}
321
322	if (packet->mad.hdr.id < 0 ||
323	    packet->mad.hdr.id >= IB_UMAD_MAX_AGENTS) {
324		ret = -EINVAL;
325		goto err;
326	}
327
328	down_read(&file->port->mutex);
329
330	agent = file->agent[packet->mad.hdr.id];
331	if (!agent) {
332		ret = -EINVAL;
333		goto err_up;
334	}
335
336	memset(&ah_attr, 0, sizeof ah_attr);
337	ah_attr.dlid          = be16_to_cpu(packet->mad.hdr.lid);
338	ah_attr.sl            = packet->mad.hdr.sl;
339	ah_attr.src_path_bits = packet->mad.hdr.path_bits;
340	ah_attr.port_num      = file->port->port_num;
341	if (packet->mad.hdr.grh_present) {
342		ah_attr.ah_flags = IB_AH_GRH;
343		memcpy(ah_attr.grh.dgid.raw, packet->mad.hdr.gid, 16);
344		ah_attr.grh.flow_label 	   = be32_to_cpu(packet->mad.hdr.flow_label);
345		ah_attr.grh.hop_limit  	   = packet->mad.hdr.hop_limit;
346		ah_attr.grh.traffic_class  = packet->mad.hdr.traffic_class;
347	}
348
349	ah = ib_create_ah(agent->qp->pd, &ah_attr);
350	if (IS_ERR(ah)) {
351		ret = PTR_ERR(ah);
352		goto err_up;
353	}
354
355	rmpp_mad = (struct ib_rmpp_mad *) packet->mad.data;
356	if (ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) & IB_MGMT_RMPP_FLAG_ACTIVE) {
357		/* RMPP active */
358		if (!agent->rmpp_version) {
359			ret = -EINVAL;
360			goto err_ah;
361		}
362
363		/* Validate that the management class can support RMPP */
364		if (rmpp_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_ADM) {
365			hdr_len = IB_MGMT_SA_HDR;
366		} else if ((rmpp_mad->mad_hdr.mgmt_class >= IB_MGMT_CLASS_VENDOR_RANGE2_START) &&
367			    (rmpp_mad->mad_hdr.mgmt_class <= IB_MGMT_CLASS_VENDOR_RANGE2_END)) {
368				hdr_len = IB_MGMT_VENDOR_HDR;
369		} else {
370			ret = -EINVAL;
371			goto err_ah;
372		}
373		rmpp_active = 1;
374		copy_offset = IB_MGMT_RMPP_HDR;
375	} else {
376		hdr_len = IB_MGMT_MAD_HDR;
377		copy_offset = IB_MGMT_MAD_HDR;
378	}
379
380	packet->msg = ib_create_send_mad(agent,
381					 be32_to_cpu(packet->mad.hdr.qpn),
382					 0, rmpp_active,
383					 hdr_len, length - hdr_len,
384					 GFP_KERNEL);
385	if (IS_ERR(packet->msg)) {
386		ret = PTR_ERR(packet->msg);
387		goto err_ah;
388	}
389
390	packet->msg->ah 	= ah;
391	packet->msg->timeout_ms = packet->mad.hdr.timeout_ms;
392	packet->msg->retries 	= packet->mad.hdr.retries;
393	packet->msg->context[0] = packet;
394
395	/* Copy MAD headers (RMPP header in place) */
396	memcpy(packet->msg->mad, packet->mad.data, IB_MGMT_MAD_HDR);
397	/* Now, copy rest of message from user into send buffer */
398	if (copy_from_user(packet->msg->mad + copy_offset,
399			   buf + sizeof (struct ib_user_mad) + copy_offset,
400			   length - copy_offset)) {
401		ret = -EFAULT;
402		goto err_msg;
403	}
404
405	/*
406	 * If userspace is generating a request that will generate a
407	 * response, we need to make sure the high-order part of the
408	 * transaction ID matches the agent being used to send the
409	 * MAD.
410	 */
411	method = ((struct ib_mad_hdr *) packet->msg->mad)->method;
412
413	if (!(method & IB_MGMT_METHOD_RESP)       &&
414	    method != IB_MGMT_METHOD_TRAP_REPRESS &&
415	    method != IB_MGMT_METHOD_SEND) {
416		tid = &((struct ib_mad_hdr *) packet->msg->mad)->tid;
417		*tid = cpu_to_be64(((u64) agent->hi_tid) << 32 |
418				   (be64_to_cpup(tid) & 0xffffffff));
419	}
420
421	ret = ib_post_send_mad(packet->msg, NULL);
422	if (ret)
423		goto err_msg;
424
425	up_read(&file->port->mutex);
426
427	return count;
428
429err_msg:
430	ib_free_send_mad(packet->msg);
431
432err_ah:
433	ib_destroy_ah(ah);
434
435err_up:
436	up_read(&file->port->mutex);
437
438err:
439	kfree(packet);
440	return ret;
441}
442
443static unsigned int ib_umad_poll(struct file *filp, struct poll_table_struct *wait)
444{
445	struct ib_umad_file *file = filp->private_data;
446
447	/* we will always be able to post a MAD send */
448	unsigned int mask = POLLOUT | POLLWRNORM;
449
450	poll_wait(filp, &file->recv_wait, wait);
451
452	if (!list_empty(&file->recv_list))
453		mask |= POLLIN | POLLRDNORM;
454
455	return mask;
456}
457
458static int ib_umad_reg_agent(struct ib_umad_file *file, unsigned long arg)
459{
460	struct ib_user_mad_reg_req ureq;
461	struct ib_mad_reg_req req;
462	struct ib_mad_agent *agent;
463	int agent_id;
464	int ret;
465
466	down_write(&file->port->mutex);
467
468	if (!file->port->ib_dev) {
469		ret = -EPIPE;
470		goto out;
471	}
472
473	if (copy_from_user(&ureq, (void __user *) arg, sizeof ureq)) {
474		ret = -EFAULT;
475		goto out;
476	}
477
478	if (ureq.qpn != 0 && ureq.qpn != 1) {
479		ret = -EINVAL;
480		goto out;
481	}
482
483	for (agent_id = 0; agent_id < IB_UMAD_MAX_AGENTS; ++agent_id)
484		if (!file->agent[agent_id])
485			goto found;
486
487	ret = -ENOMEM;
488	goto out;
489
490found:
491	if (ureq.mgmt_class) {
492		req.mgmt_class         = ureq.mgmt_class;
493		req.mgmt_class_version = ureq.mgmt_class_version;
494		memcpy(req.method_mask, ureq.method_mask, sizeof req.method_mask);
495		memcpy(req.oui,         ureq.oui,         sizeof req.oui);
496	}
497
498	agent = ib_register_mad_agent(file->port->ib_dev, file->port->port_num,
499				      ureq.qpn ? IB_QPT_GSI : IB_QPT_SMI,
500				      ureq.mgmt_class ? &req : NULL,
501				      ureq.rmpp_version,
502				      send_handler, recv_handler, file);
503	if (IS_ERR(agent)) {
504		ret = PTR_ERR(agent);
505		goto out;
506	}
507
508	file->mr[agent_id] = ib_get_dma_mr(agent->qp->pd, IB_ACCESS_LOCAL_WRITE);
509	if (IS_ERR(file->mr[agent_id])) {
510		ret = -ENOMEM;
511		goto err;
512	}
513
514	if (put_user(agent_id,
515		     (u32 __user *) (arg + offsetof(struct ib_user_mad_reg_req, id)))) {
516		ret = -EFAULT;
517		goto err_mr;
518	}
519
520	file->agent[agent_id] = agent;
521	ret = 0;
522
523	goto out;
524
525err_mr:
526	ib_dereg_mr(file->mr[agent_id]);
527
528err:
529	ib_unregister_mad_agent(agent);
530
531out:
532	up_write(&file->port->mutex);
533	return ret;
534}
535
536static int ib_umad_unreg_agent(struct ib_umad_file *file, unsigned long arg)
537{
538	struct ib_mad_agent *agent = NULL;
539	struct ib_mr *mr = NULL;
540	u32 id;
541	int ret = 0;
542
543	if (get_user(id, (u32 __user *) arg))
544		return -EFAULT;
545
546	down_write(&file->port->mutex);
547
548	if (id < 0 || id >= IB_UMAD_MAX_AGENTS || !file->agent[id]) {
549		ret = -EINVAL;
550		goto out;
551	}
552
553	agent = file->agent[id];
554	mr    = file->mr[id];
555	file->agent[id] = NULL;
556
557out:
558	up_write(&file->port->mutex);
559
560	if (agent) {
561		ib_unregister_mad_agent(agent);
562		ib_dereg_mr(mr);
563	}
564
565	return ret;
566}
567
568static long ib_umad_ioctl(struct file *filp, unsigned int cmd,
569			  unsigned long arg)
570{
571	switch (cmd) {
572	case IB_USER_MAD_REGISTER_AGENT:
573		return ib_umad_reg_agent(filp->private_data, arg);
574	case IB_USER_MAD_UNREGISTER_AGENT:
575		return ib_umad_unreg_agent(filp->private_data, arg);
576	default:
577		return -ENOIOCTLCMD;
578	}
579}
580
581static int ib_umad_open(struct inode *inode, struct file *filp)
582{
583	struct ib_umad_port *port;
584	struct ib_umad_file *file;
585	int ret = 0;
586
587	spin_lock(&port_lock);
588	port = umad_port[iminor(inode) - IB_UMAD_MINOR_BASE];
589	if (port)
590		kref_get(&port->umad_dev->ref);
591	spin_unlock(&port_lock);
592
593	if (!port)
594		return -ENXIO;
595
596	down_write(&port->mutex);
597
598	if (!port->ib_dev) {
599		ret = -ENXIO;
600		goto out;
601	}
602
603	file = kzalloc(sizeof *file, GFP_KERNEL);
604	if (!file) {
605		kref_put(&port->umad_dev->ref, ib_umad_release_dev);
606		ret = -ENOMEM;
607		goto out;
608	}
609
610	spin_lock_init(&file->recv_lock);
611	INIT_LIST_HEAD(&file->recv_list);
612	init_waitqueue_head(&file->recv_wait);
613
614	file->port = port;
615	filp->private_data = file;
616
617	list_add_tail(&file->port_list, &port->file_list);
618
619out:
620	up_write(&port->mutex);
621	return ret;
622}
623
624static int ib_umad_close(struct inode *inode, struct file *filp)
625{
626	struct ib_umad_file *file = filp->private_data;
627	struct ib_umad_device *dev = file->port->umad_dev;
628	struct ib_umad_packet *packet, *tmp;
629	int i;
630
631	for (i = 0; i < IB_UMAD_MAX_AGENTS; ++i)
632		if (file->agent[i]) {
633			ib_unregister_mad_agent(file->agent[i]);
634			ib_dereg_mr(file->mr[i]);
635		}
636
637	list_for_each_entry_safe(packet, tmp, &file->recv_list, list)
638		kfree(packet);
639
640	down_write(&file->port->mutex);
641	list_del(&file->port_list);
642	up_write(&file->port->mutex);
643
644	kfree(file);
645
646	kref_put(&dev->ref, ib_umad_release_dev);
647
648	return 0;
649}
650
651static struct file_operations umad_fops = {
652	.owner 	 	= THIS_MODULE,
653	.read 	 	= ib_umad_read,
654	.write 	 	= ib_umad_write,
655	.poll 	 	= ib_umad_poll,
656	.unlocked_ioctl = ib_umad_ioctl,
657	.compat_ioctl 	= ib_umad_ioctl,
658	.open 	 	= ib_umad_open,
659	.release 	= ib_umad_close
660};
661
662static int ib_umad_sm_open(struct inode *inode, struct file *filp)
663{
664	struct ib_umad_port *port;
665	struct ib_port_modify props = {
666		.set_port_cap_mask = IB_PORT_SM
667	};
668	int ret;
669
670	spin_lock(&port_lock);
671	port = umad_port[iminor(inode) - IB_UMAD_MINOR_BASE - IB_UMAD_MAX_PORTS];
672	if (port)
673		kref_get(&port->umad_dev->ref);
674	spin_unlock(&port_lock);
675
676	if (!port)
677		return -ENXIO;
678
679	if (filp->f_flags & O_NONBLOCK) {
680		if (down_trylock(&port->sm_sem)) {
681			ret = -EAGAIN;
682			goto fail;
683		}
684	} else {
685		if (down_interruptible(&port->sm_sem)) {
686			ret = -ERESTARTSYS;
687			goto fail;
688		}
689	}
690
691	ret = ib_modify_port(port->ib_dev, port->port_num, 0, &props);
692	if (ret) {
693		up(&port->sm_sem);
694		goto fail;
695	}
696
697	filp->private_data = port;
698
699	return 0;
700
701fail:
702	kref_put(&port->umad_dev->ref, ib_umad_release_dev);
703	return ret;
704}
705
706static int ib_umad_sm_close(struct inode *inode, struct file *filp)
707{
708	struct ib_umad_port *port = filp->private_data;
709	struct ib_port_modify props = {
710		.clr_port_cap_mask = IB_PORT_SM
711	};
712	int ret = 0;
713
714	down_write(&port->mutex);
715	if (port->ib_dev)
716		ret = ib_modify_port(port->ib_dev, port->port_num, 0, &props);
717	up_write(&port->mutex);
718
719	up(&port->sm_sem);
720
721	kref_put(&port->umad_dev->ref, ib_umad_release_dev);
722
723	return ret;
724}
725
726static struct file_operations umad_sm_fops = {
727	.owner 	 = THIS_MODULE,
728	.open 	 = ib_umad_sm_open,
729	.release = ib_umad_sm_close
730};
731
732static struct ib_client umad_client = {
733	.name   = "umad",
734	.add    = ib_umad_add_one,
735	.remove = ib_umad_remove_one
736};
737
738static ssize_t show_ibdev(struct class_device *class_dev, char *buf)
739{
740	struct ib_umad_port *port = class_get_devdata(class_dev);
741
742	if (!port)
743		return -ENODEV;
744
745	return sprintf(buf, "%s\n", port->ib_dev->name);
746}
747static CLASS_DEVICE_ATTR(ibdev, S_IRUGO, show_ibdev, NULL);
748
749static ssize_t show_port(struct class_device *class_dev, char *buf)
750{
751	struct ib_umad_port *port = class_get_devdata(class_dev);
752
753	if (!port)
754		return -ENODEV;
755
756	return sprintf(buf, "%d\n", port->port_num);
757}
758static CLASS_DEVICE_ATTR(port, S_IRUGO, show_port, NULL);
759
760static ssize_t show_abi_version(struct class *class, char *buf)
761{
762	return sprintf(buf, "%d\n", IB_USER_MAD_ABI_VERSION);
763}
764static CLASS_ATTR(abi_version, S_IRUGO, show_abi_version, NULL);
765
766static int ib_umad_init_port(struct ib_device *device, int port_num,
767			     struct ib_umad_port *port)
768{
769	spin_lock(&port_lock);
770	port->dev_num = find_first_zero_bit(dev_map, IB_UMAD_MAX_PORTS);
771	if (port->dev_num >= IB_UMAD_MAX_PORTS) {
772		spin_unlock(&port_lock);
773		return -1;
774	}
775	set_bit(port->dev_num, dev_map);
776	spin_unlock(&port_lock);
777
778	port->ib_dev   = device;
779	port->port_num = port_num;
780	init_MUTEX(&port->sm_sem);
781	init_rwsem(&port->mutex);
782	INIT_LIST_HEAD(&port->file_list);
783
784	port->dev = cdev_alloc();
785	if (!port->dev)
786		return -1;
787	port->dev->owner = THIS_MODULE;
788	port->dev->ops   = &umad_fops;
789	kobject_set_name(&port->dev->kobj, "umad%d", port->dev_num);
790	if (cdev_add(port->dev, base_dev + port->dev_num, 1))
791		goto err_cdev;
792
793	port->class_dev = class_device_create(umad_class, NULL, port->dev->dev,
794					      device->dma_device,
795					      "umad%d", port->dev_num);
796	if (IS_ERR(port->class_dev))
797		goto err_cdev;
798
799	if (class_device_create_file(port->class_dev, &class_device_attr_ibdev))
800		goto err_class;
801	if (class_device_create_file(port->class_dev, &class_device_attr_port))
802		goto err_class;
803
804	port->sm_dev = cdev_alloc();
805	if (!port->sm_dev)
806		goto err_class;
807	port->sm_dev->owner = THIS_MODULE;
808	port->sm_dev->ops   = &umad_sm_fops;
809	kobject_set_name(&port->sm_dev->kobj, "issm%d", port->dev_num);
810	if (cdev_add(port->sm_dev, base_dev + port->dev_num + IB_UMAD_MAX_PORTS, 1))
811		goto err_sm_cdev;
812
813	port->sm_class_dev = class_device_create(umad_class, NULL, port->sm_dev->dev,
814						 device->dma_device,
815						 "issm%d", port->dev_num);
816	if (IS_ERR(port->sm_class_dev))
817		goto err_sm_cdev;
818
819	class_set_devdata(port->class_dev,    port);
820	class_set_devdata(port->sm_class_dev, port);
821
822	if (class_device_create_file(port->sm_class_dev, &class_device_attr_ibdev))
823		goto err_sm_class;
824	if (class_device_create_file(port->sm_class_dev, &class_device_attr_port))
825		goto err_sm_class;
826
827	spin_lock(&port_lock);
828	umad_port[port->dev_num] = port;
829	spin_unlock(&port_lock);
830
831	return 0;
832
833err_sm_class:
834	class_device_destroy(umad_class, port->sm_dev->dev);
835
836err_sm_cdev:
837	cdev_del(port->sm_dev);
838
839err_class:
840	class_device_destroy(umad_class, port->dev->dev);
841
842err_cdev:
843	cdev_del(port->dev);
844	clear_bit(port->dev_num, dev_map);
845
846	return -1;
847}
848
849static void ib_umad_kill_port(struct ib_umad_port *port)
850{
851	struct ib_umad_file *file;
852	int id;
853
854	class_set_devdata(port->class_dev,    NULL);
855	class_set_devdata(port->sm_class_dev, NULL);
856
857	class_device_destroy(umad_class, port->dev->dev);
858	class_device_destroy(umad_class, port->sm_dev->dev);
859
860	cdev_del(port->dev);
861	cdev_del(port->sm_dev);
862
863	spin_lock(&port_lock);
864	umad_port[port->dev_num] = NULL;
865	spin_unlock(&port_lock);
866
867	down_write(&port->mutex);
868
869	port->ib_dev = NULL;
870
871	list_for_each_entry(file, &port->file_list, port_list)
872		for (id = 0; id < IB_UMAD_MAX_AGENTS; ++id) {
873			if (!file->agent[id])
874				continue;
875			ib_dereg_mr(file->mr[id]);
876			ib_unregister_mad_agent(file->agent[id]);
877			file->agent[id] = NULL;
878		}
879
880	up_write(&port->mutex);
881
882	clear_bit(port->dev_num, dev_map);
883}
884
885static void ib_umad_add_one(struct ib_device *device)
886{
887	struct ib_umad_device *umad_dev;
888	int s, e, i;
889
890	if (device->node_type == IB_NODE_SWITCH)
891		s = e = 0;
892	else {
893		s = 1;
894		e = device->phys_port_cnt;
895	}
896
897	umad_dev = kzalloc(sizeof *umad_dev +
898			   (e - s + 1) * sizeof (struct ib_umad_port),
899			   GFP_KERNEL);
900	if (!umad_dev)
901		return;
902
903	kref_init(&umad_dev->ref);
904
905	umad_dev->start_port = s;
906	umad_dev->end_port   = e;
907
908	for (i = s; i <= e; ++i) {
909		umad_dev->port[i - s].umad_dev = umad_dev;
910
911		if (ib_umad_init_port(device, i, &umad_dev->port[i - s]))
912			goto err;
913	}
914
915	ib_set_client_data(device, &umad_client, umad_dev);
916
917	return;
918
919err:
920	while (--i >= s)
921		ib_umad_kill_port(&umad_dev->port[i - s]);
922
923	kref_put(&umad_dev->ref, ib_umad_release_dev);
924}
925
926static void ib_umad_remove_one(struct ib_device *device)
927{
928	struct ib_umad_device *umad_dev = ib_get_client_data(device, &umad_client);
929	int i;
930
931	if (!umad_dev)
932		return;
933
934	for (i = 0; i <= umad_dev->end_port - umad_dev->start_port; ++i)
935		ib_umad_kill_port(&umad_dev->port[i]);
936
937	kref_put(&umad_dev->ref, ib_umad_release_dev);
938}
939
940static int __init ib_umad_init(void)
941{
942	int ret;
943
944	ret = register_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2,
945				     "infiniband_mad");
946	if (ret) {
947		printk(KERN_ERR "user_mad: couldn't register device number\n");
948		goto out;
949	}
950
951	umad_class = class_create(THIS_MODULE, "infiniband_mad");
952	if (IS_ERR(umad_class)) {
953		ret = PTR_ERR(umad_class);
954		printk(KERN_ERR "user_mad: couldn't create class infiniband_mad\n");
955		goto out_chrdev;
956	}
957
958	ret = class_create_file(umad_class, &class_attr_abi_version);
959	if (ret) {
960		printk(KERN_ERR "user_mad: couldn't create abi_version attribute\n");
961		goto out_class;
962	}
963
964	ret = ib_register_client(&umad_client);
965	if (ret) {
966		printk(KERN_ERR "user_mad: couldn't register ib_umad client\n");
967		goto out_class;
968	}
969
970	return 0;
971
972out_class:
973	class_destroy(umad_class);
974
975out_chrdev:
976	unregister_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2);
977
978out:
979	return ret;
980}
981
982static void __exit ib_umad_cleanup(void)
983{
984	ib_unregister_client(&umad_client);
985	class_destroy(umad_class);
986	unregister_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2);
987}
988
989module_init(ib_umad_init);
990module_exit(ib_umad_cleanup);
991