user_mad.c revision ec914c52d6208d8752dfd85b48a9aff304911434
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};
120
121struct ib_umad_packet {
122	struct ib_mad_send_buf *msg;
123	struct list_head   list;
124	int		   length;
125	struct ib_user_mad mad;
126};
127
128static struct class *umad_class;
129
130static const dev_t base_dev = MKDEV(IB_UMAD_MAJOR, IB_UMAD_MINOR_BASE);
131
132static DEFINE_SPINLOCK(port_lock);
133static struct ib_umad_port *umad_port[IB_UMAD_MAX_PORTS];
134static DECLARE_BITMAP(dev_map, IB_UMAD_MAX_PORTS * 2);
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 kref *ref)
140{
141	struct ib_umad_device *dev =
142		container_of(ref, struct ib_umad_device, ref);
143
144	kfree(dev);
145}
146
147static int queue_packet(struct ib_umad_file *file,
148			struct ib_mad_agent *agent,
149			struct ib_umad_packet *packet)
150{
151	int ret = 1;
152
153	down_read(&file->port->mutex);
154	for (packet->mad.hdr.id = 0;
155	     packet->mad.hdr.id < IB_UMAD_MAX_AGENTS;
156	     packet->mad.hdr.id++)
157		if (agent == file->agent[packet->mad.hdr.id]) {
158			spin_lock_irq(&file->recv_lock);
159			list_add_tail(&packet->list, &file->recv_list);
160			spin_unlock_irq(&file->recv_lock);
161			wake_up_interruptible(&file->recv_wait);
162			ret = 0;
163			break;
164		}
165
166	up_read(&file->port->mutex);
167
168	return ret;
169}
170
171static void send_handler(struct ib_mad_agent *agent,
172			 struct ib_mad_send_wc *send_wc)
173{
174	struct ib_umad_file *file = agent->context;
175	struct ib_umad_packet *timeout;
176	struct ib_umad_packet *packet = send_wc->send_buf->context[0];
177
178	ib_destroy_ah(packet->msg->ah);
179	ib_free_send_mad(packet->msg);
180
181	if (send_wc->status == IB_WC_RESP_TIMEOUT_ERR) {
182		timeout = kzalloc(sizeof *timeout + IB_MGMT_MAD_HDR, GFP_KERNEL);
183		if (!timeout)
184			goto out;
185
186		timeout->length 	= IB_MGMT_MAD_HDR;
187		timeout->mad.hdr.id 	= packet->mad.hdr.id;
188		timeout->mad.hdr.status = ETIMEDOUT;
189		memcpy(timeout->mad.data, packet->mad.data,
190		       sizeof (struct ib_mad_hdr));
191
192		if (!queue_packet(file, agent, timeout))
193				return;
194	}
195out:
196	kfree(packet);
197}
198
199static void recv_handler(struct ib_mad_agent *agent,
200			 struct ib_mad_recv_wc *mad_recv_wc)
201{
202	struct ib_umad_file *file = agent->context;
203	struct ib_umad_packet *packet;
204	int length;
205
206	if (mad_recv_wc->wc->status != IB_WC_SUCCESS)
207		goto out;
208
209	length = mad_recv_wc->mad_len;
210	packet = kzalloc(sizeof *packet + length, GFP_KERNEL);
211	if (!packet)
212		goto out;
213
214	packet->length = length;
215
216	ib_coalesce_recv_mad(mad_recv_wc, packet->mad.data);
217
218	packet->mad.hdr.status    = 0;
219	packet->mad.hdr.length    = length + sizeof (struct ib_user_mad);
220	packet->mad.hdr.qpn 	  = cpu_to_be32(mad_recv_wc->wc->src_qp);
221	packet->mad.hdr.lid 	  = cpu_to_be16(mad_recv_wc->wc->slid);
222	packet->mad.hdr.sl  	  = mad_recv_wc->wc->sl;
223	packet->mad.hdr.path_bits = mad_recv_wc->wc->dlid_path_bits;
224	packet->mad.hdr.grh_present = !!(mad_recv_wc->wc->wc_flags & IB_WC_GRH);
225	if (packet->mad.hdr.grh_present) {
226		/* XXX parse GRH */
227		packet->mad.hdr.gid_index 	= 0;
228		packet->mad.hdr.hop_limit 	= 0;
229		packet->mad.hdr.traffic_class	= 0;
230		memset(packet->mad.hdr.gid, 0, 16);
231		packet->mad.hdr.flow_label	= 0;
232	}
233
234	if (queue_packet(file, agent, packet))
235		kfree(packet);
236
237out:
238	ib_free_recv_mad(mad_recv_wc);
239}
240
241static ssize_t ib_umad_read(struct file *filp, char __user *buf,
242			    size_t count, loff_t *pos)
243{
244	struct ib_umad_file *file = filp->private_data;
245	struct ib_umad_packet *packet;
246	ssize_t ret;
247
248	if (count < sizeof (struct ib_user_mad) + sizeof (struct ib_mad))
249		return -EINVAL;
250
251	spin_lock_irq(&file->recv_lock);
252
253	while (list_empty(&file->recv_list)) {
254		spin_unlock_irq(&file->recv_lock);
255
256		if (filp->f_flags & O_NONBLOCK)
257			return -EAGAIN;
258
259		if (wait_event_interruptible(file->recv_wait,
260					     !list_empty(&file->recv_list)))
261			return -ERESTARTSYS;
262
263		spin_lock_irq(&file->recv_lock);
264	}
265
266	packet = list_entry(file->recv_list.next, struct ib_umad_packet, list);
267	list_del(&packet->list);
268
269	spin_unlock_irq(&file->recv_lock);
270
271	if (count < packet->length + sizeof (struct ib_user_mad)) {
272		/* Return length needed (and first RMPP segment) if too small */
273		if (copy_to_user(buf, &packet->mad,
274				 sizeof (struct ib_user_mad) + sizeof (struct ib_mad)))
275			ret = -EFAULT;
276		else
277			ret = -ENOSPC;
278	} else if (copy_to_user(buf, &packet->mad,
279				packet->length + sizeof (struct ib_user_mad)))
280		ret = -EFAULT;
281	else
282		ret = packet->length + sizeof (struct ib_user_mad);
283	if (ret < 0) {
284		/* Requeue packet */
285		spin_lock_irq(&file->recv_lock);
286		list_add(&packet->list, &file->recv_list);
287		spin_unlock_irq(&file->recv_lock);
288	} else
289		kfree(packet);
290	return ret;
291}
292
293static ssize_t ib_umad_write(struct file *filp, const char __user *buf,
294			     size_t count, loff_t *pos)
295{
296	struct ib_umad_file *file = filp->private_data;
297	struct ib_umad_packet *packet;
298	struct ib_mad_agent *agent;
299	struct ib_ah_attr ah_attr;
300	struct ib_ah *ah;
301	struct ib_rmpp_mad *rmpp_mad;
302	u8 method;
303	__be64 *tid;
304	int ret, length, hdr_len, copy_offset;
305	int rmpp_active = 0;
306
307	if (count < sizeof (struct ib_user_mad))
308		return -EINVAL;
309
310	length = count - sizeof (struct ib_user_mad);
311	packet = kmalloc(sizeof *packet + IB_MGMT_RMPP_HDR, GFP_KERNEL);
312	if (!packet)
313		return -ENOMEM;
314
315	if (copy_from_user(&packet->mad, buf,
316			    sizeof (struct ib_user_mad) + IB_MGMT_RMPP_HDR)) {
317		ret = -EFAULT;
318		goto err;
319	}
320
321	if (packet->mad.hdr.id < 0 ||
322	    packet->mad.hdr.id >= IB_UMAD_MAX_AGENTS) {
323		ret = -EINVAL;
324		goto err;
325	}
326
327	down_read(&file->port->mutex);
328
329	agent = file->agent[packet->mad.hdr.id];
330	if (!agent) {
331		ret = -EINVAL;
332		goto err_up;
333	}
334
335	memset(&ah_attr, 0, sizeof ah_attr);
336	ah_attr.dlid          = be16_to_cpu(packet->mad.hdr.lid);
337	ah_attr.sl            = packet->mad.hdr.sl;
338	ah_attr.src_path_bits = packet->mad.hdr.path_bits;
339	ah_attr.port_num      = file->port->port_num;
340	if (packet->mad.hdr.grh_present) {
341		ah_attr.ah_flags = IB_AH_GRH;
342		memcpy(ah_attr.grh.dgid.raw, packet->mad.hdr.gid, 16);
343		ah_attr.grh.flow_label 	   = be32_to_cpu(packet->mad.hdr.flow_label);
344		ah_attr.grh.hop_limit  	   = packet->mad.hdr.hop_limit;
345		ah_attr.grh.traffic_class  = packet->mad.hdr.traffic_class;
346	}
347
348	ah = ib_create_ah(agent->qp->pd, &ah_attr);
349	if (IS_ERR(ah)) {
350		ret = PTR_ERR(ah);
351		goto err_up;
352	}
353
354	rmpp_mad = (struct ib_rmpp_mad *) packet->mad.data;
355	if (ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) & IB_MGMT_RMPP_FLAG_ACTIVE) {
356		/* RMPP active */
357		if (!agent->rmpp_version) {
358			ret = -EINVAL;
359			goto err_ah;
360		}
361
362		/* Validate that the management class can support RMPP */
363		if (rmpp_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_ADM) {
364			hdr_len = IB_MGMT_SA_HDR;
365		} else if ((rmpp_mad->mad_hdr.mgmt_class >= IB_MGMT_CLASS_VENDOR_RANGE2_START) &&
366			    (rmpp_mad->mad_hdr.mgmt_class <= IB_MGMT_CLASS_VENDOR_RANGE2_END)) {
367				hdr_len = IB_MGMT_VENDOR_HDR;
368		} else {
369			ret = -EINVAL;
370			goto err_ah;
371		}
372		rmpp_active = 1;
373		copy_offset = IB_MGMT_RMPP_HDR;
374	} else {
375		hdr_len = IB_MGMT_MAD_HDR;
376		copy_offset = IB_MGMT_MAD_HDR;
377	}
378
379	packet->msg = ib_create_send_mad(agent,
380					 be32_to_cpu(packet->mad.hdr.qpn),
381					 0, rmpp_active,
382					 hdr_len, length - hdr_len,
383					 GFP_KERNEL);
384	if (IS_ERR(packet->msg)) {
385		ret = PTR_ERR(packet->msg);
386		goto err_ah;
387	}
388
389	packet->msg->ah 	= ah;
390	packet->msg->timeout_ms = packet->mad.hdr.timeout_ms;
391	packet->msg->retries 	= packet->mad.hdr.retries;
392	packet->msg->context[0] = packet;
393
394	/* Copy MAD headers (RMPP header in place) */
395	memcpy(packet->msg->mad, packet->mad.data, IB_MGMT_MAD_HDR);
396	/* Now, copy rest of message from user into send buffer */
397	if (copy_from_user(packet->msg->mad + copy_offset,
398			   buf + sizeof (struct ib_user_mad) + copy_offset,
399			   length - copy_offset)) {
400		ret = -EFAULT;
401		goto err_msg;
402	}
403
404	/*
405	 * If userspace is generating a request that will generate a
406	 * response, we need to make sure the high-order part of the
407	 * transaction ID matches the agent being used to send the
408	 * MAD.
409	 */
410	method = ((struct ib_mad_hdr *) packet->msg->mad)->method;
411
412	if (!(method & IB_MGMT_METHOD_RESP)       &&
413	    method != IB_MGMT_METHOD_TRAP_REPRESS &&
414	    method != IB_MGMT_METHOD_SEND) {
415		tid = &((struct ib_mad_hdr *) packet->msg->mad)->tid;
416		*tid = cpu_to_be64(((u64) agent->hi_tid) << 32 |
417				   (be64_to_cpup(tid) & 0xffffffff));
418	}
419
420	ret = ib_post_send_mad(packet->msg, NULL);
421	if (ret)
422		goto err_msg;
423
424	up_read(&file->port->mutex);
425
426	return count;
427
428err_msg:
429	ib_free_send_mad(packet->msg);
430
431err_ah:
432	ib_destroy_ah(ah);
433
434err_up:
435	up_read(&file->port->mutex);
436
437err:
438	kfree(packet);
439	return ret;
440}
441
442static unsigned int ib_umad_poll(struct file *filp, struct poll_table_struct *wait)
443{
444	struct ib_umad_file *file = filp->private_data;
445
446	/* we will always be able to post a MAD send */
447	unsigned int mask = POLLOUT | POLLWRNORM;
448
449	poll_wait(filp, &file->recv_wait, wait);
450
451	if (!list_empty(&file->recv_list))
452		mask |= POLLIN | POLLRDNORM;
453
454	return mask;
455}
456
457static int ib_umad_reg_agent(struct ib_umad_file *file, unsigned long arg)
458{
459	struct ib_user_mad_reg_req ureq;
460	struct ib_mad_reg_req req;
461	struct ib_mad_agent *agent;
462	int agent_id;
463	int ret;
464
465	down_write(&file->port->mutex);
466
467	if (!file->port->ib_dev) {
468		ret = -EPIPE;
469		goto out;
470	}
471
472	if (copy_from_user(&ureq, (void __user *) arg, sizeof ureq)) {
473		ret = -EFAULT;
474		goto out;
475	}
476
477	if (ureq.qpn != 0 && ureq.qpn != 1) {
478		ret = -EINVAL;
479		goto out;
480	}
481
482	for (agent_id = 0; agent_id < IB_UMAD_MAX_AGENTS; ++agent_id)
483		if (!file->agent[agent_id])
484			goto found;
485
486	ret = -ENOMEM;
487	goto out;
488
489found:
490	if (ureq.mgmt_class) {
491		req.mgmt_class         = ureq.mgmt_class;
492		req.mgmt_class_version = ureq.mgmt_class_version;
493		memcpy(req.method_mask, ureq.method_mask, sizeof req.method_mask);
494		memcpy(req.oui,         ureq.oui,         sizeof req.oui);
495	}
496
497	agent = ib_register_mad_agent(file->port->ib_dev, file->port->port_num,
498				      ureq.qpn ? IB_QPT_GSI : IB_QPT_SMI,
499				      ureq.mgmt_class ? &req : NULL,
500				      ureq.rmpp_version,
501				      send_handler, recv_handler, file);
502	if (IS_ERR(agent)) {
503		ret = PTR_ERR(agent);
504		goto out;
505	}
506
507	if (put_user(agent_id,
508		     (u32 __user *) (arg + offsetof(struct ib_user_mad_reg_req, id)))) {
509		ret = -EFAULT;
510		ib_unregister_mad_agent(agent);
511		goto out;
512	}
513
514	file->agent[agent_id] = agent;
515	ret = 0;
516
517out:
518	up_write(&file->port->mutex);
519	return ret;
520}
521
522static int ib_umad_unreg_agent(struct ib_umad_file *file, unsigned long arg)
523{
524	struct ib_mad_agent *agent = NULL;
525	u32 id;
526	int ret = 0;
527
528	if (get_user(id, (u32 __user *) arg))
529		return -EFAULT;
530
531	down_write(&file->port->mutex);
532
533	if (id < 0 || id >= IB_UMAD_MAX_AGENTS || !file->agent[id]) {
534		ret = -EINVAL;
535		goto out;
536	}
537
538	agent = file->agent[id];
539	file->agent[id] = NULL;
540
541out:
542	up_write(&file->port->mutex);
543
544	if (agent)
545		ib_unregister_mad_agent(agent);
546
547	return ret;
548}
549
550static long ib_umad_ioctl(struct file *filp, unsigned int cmd,
551			  unsigned long arg)
552{
553	switch (cmd) {
554	case IB_USER_MAD_REGISTER_AGENT:
555		return ib_umad_reg_agent(filp->private_data, arg);
556	case IB_USER_MAD_UNREGISTER_AGENT:
557		return ib_umad_unreg_agent(filp->private_data, arg);
558	default:
559		return -ENOIOCTLCMD;
560	}
561}
562
563static int ib_umad_open(struct inode *inode, struct file *filp)
564{
565	struct ib_umad_port *port;
566	struct ib_umad_file *file;
567	int ret = 0;
568
569	spin_lock(&port_lock);
570	port = umad_port[iminor(inode) - IB_UMAD_MINOR_BASE];
571	if (port)
572		kref_get(&port->umad_dev->ref);
573	spin_unlock(&port_lock);
574
575	if (!port)
576		return -ENXIO;
577
578	down_write(&port->mutex);
579
580	if (!port->ib_dev) {
581		ret = -ENXIO;
582		goto out;
583	}
584
585	file = kzalloc(sizeof *file, GFP_KERNEL);
586	if (!file) {
587		kref_put(&port->umad_dev->ref, ib_umad_release_dev);
588		ret = -ENOMEM;
589		goto out;
590	}
591
592	spin_lock_init(&file->recv_lock);
593	INIT_LIST_HEAD(&file->recv_list);
594	init_waitqueue_head(&file->recv_wait);
595
596	file->port = port;
597	filp->private_data = file;
598
599	list_add_tail(&file->port_list, &port->file_list);
600
601out:
602	up_write(&port->mutex);
603	return ret;
604}
605
606static int ib_umad_close(struct inode *inode, struct file *filp)
607{
608	struct ib_umad_file *file = filp->private_data;
609	struct ib_umad_device *dev = file->port->umad_dev;
610	struct ib_umad_packet *packet, *tmp;
611	int i;
612
613	for (i = 0; i < IB_UMAD_MAX_AGENTS; ++i)
614		if (file->agent[i])
615			ib_unregister_mad_agent(file->agent[i]);
616
617	list_for_each_entry_safe(packet, tmp, &file->recv_list, list)
618		kfree(packet);
619
620	down_write(&file->port->mutex);
621	list_del(&file->port_list);
622	up_write(&file->port->mutex);
623
624	kfree(file);
625
626	kref_put(&dev->ref, ib_umad_release_dev);
627
628	return 0;
629}
630
631static struct file_operations umad_fops = {
632	.owner 	 	= THIS_MODULE,
633	.read 	 	= ib_umad_read,
634	.write 	 	= ib_umad_write,
635	.poll 	 	= ib_umad_poll,
636	.unlocked_ioctl = ib_umad_ioctl,
637	.compat_ioctl 	= ib_umad_ioctl,
638	.open 	 	= ib_umad_open,
639	.release 	= ib_umad_close
640};
641
642static int ib_umad_sm_open(struct inode *inode, struct file *filp)
643{
644	struct ib_umad_port *port;
645	struct ib_port_modify props = {
646		.set_port_cap_mask = IB_PORT_SM
647	};
648	int ret;
649
650	spin_lock(&port_lock);
651	port = umad_port[iminor(inode) - IB_UMAD_MINOR_BASE - IB_UMAD_MAX_PORTS];
652	if (port)
653		kref_get(&port->umad_dev->ref);
654	spin_unlock(&port_lock);
655
656	if (!port)
657		return -ENXIO;
658
659	if (filp->f_flags & O_NONBLOCK) {
660		if (down_trylock(&port->sm_sem)) {
661			ret = -EAGAIN;
662			goto fail;
663		}
664	} else {
665		if (down_interruptible(&port->sm_sem)) {
666			ret = -ERESTARTSYS;
667			goto fail;
668		}
669	}
670
671	ret = ib_modify_port(port->ib_dev, port->port_num, 0, &props);
672	if (ret) {
673		up(&port->sm_sem);
674		goto fail;
675	}
676
677	filp->private_data = port;
678
679	return 0;
680
681fail:
682	kref_put(&port->umad_dev->ref, ib_umad_release_dev);
683	return ret;
684}
685
686static int ib_umad_sm_close(struct inode *inode, struct file *filp)
687{
688	struct ib_umad_port *port = filp->private_data;
689	struct ib_port_modify props = {
690		.clr_port_cap_mask = IB_PORT_SM
691	};
692	int ret = 0;
693
694	down_write(&port->mutex);
695	if (port->ib_dev)
696		ret = ib_modify_port(port->ib_dev, port->port_num, 0, &props);
697	up_write(&port->mutex);
698
699	up(&port->sm_sem);
700
701	kref_put(&port->umad_dev->ref, ib_umad_release_dev);
702
703	return ret;
704}
705
706static struct file_operations umad_sm_fops = {
707	.owner 	 = THIS_MODULE,
708	.open 	 = ib_umad_sm_open,
709	.release = ib_umad_sm_close
710};
711
712static struct ib_client umad_client = {
713	.name   = "umad",
714	.add    = ib_umad_add_one,
715	.remove = ib_umad_remove_one
716};
717
718static ssize_t show_ibdev(struct class_device *class_dev, char *buf)
719{
720	struct ib_umad_port *port = class_get_devdata(class_dev);
721
722	if (!port)
723		return -ENODEV;
724
725	return sprintf(buf, "%s\n", port->ib_dev->name);
726}
727static CLASS_DEVICE_ATTR(ibdev, S_IRUGO, show_ibdev, NULL);
728
729static ssize_t show_port(struct class_device *class_dev, char *buf)
730{
731	struct ib_umad_port *port = class_get_devdata(class_dev);
732
733	if (!port)
734		return -ENODEV;
735
736	return sprintf(buf, "%d\n", port->port_num);
737}
738static CLASS_DEVICE_ATTR(port, S_IRUGO, show_port, NULL);
739
740static ssize_t show_abi_version(struct class *class, char *buf)
741{
742	return sprintf(buf, "%d\n", IB_USER_MAD_ABI_VERSION);
743}
744static CLASS_ATTR(abi_version, S_IRUGO, show_abi_version, NULL);
745
746static int ib_umad_init_port(struct ib_device *device, int port_num,
747			     struct ib_umad_port *port)
748{
749	spin_lock(&port_lock);
750	port->dev_num = find_first_zero_bit(dev_map, IB_UMAD_MAX_PORTS);
751	if (port->dev_num >= IB_UMAD_MAX_PORTS) {
752		spin_unlock(&port_lock);
753		return -1;
754	}
755	set_bit(port->dev_num, dev_map);
756	spin_unlock(&port_lock);
757
758	port->ib_dev   = device;
759	port->port_num = port_num;
760	init_MUTEX(&port->sm_sem);
761	init_rwsem(&port->mutex);
762	INIT_LIST_HEAD(&port->file_list);
763
764	port->dev = cdev_alloc();
765	if (!port->dev)
766		return -1;
767	port->dev->owner = THIS_MODULE;
768	port->dev->ops   = &umad_fops;
769	kobject_set_name(&port->dev->kobj, "umad%d", port->dev_num);
770	if (cdev_add(port->dev, base_dev + port->dev_num, 1))
771		goto err_cdev;
772
773	port->class_dev = class_device_create(umad_class, NULL, port->dev->dev,
774					      device->dma_device,
775					      "umad%d", port->dev_num);
776	if (IS_ERR(port->class_dev))
777		goto err_cdev;
778
779	if (class_device_create_file(port->class_dev, &class_device_attr_ibdev))
780		goto err_class;
781	if (class_device_create_file(port->class_dev, &class_device_attr_port))
782		goto err_class;
783
784	port->sm_dev = cdev_alloc();
785	if (!port->sm_dev)
786		goto err_class;
787	port->sm_dev->owner = THIS_MODULE;
788	port->sm_dev->ops   = &umad_sm_fops;
789	kobject_set_name(&port->sm_dev->kobj, "issm%d", port->dev_num);
790	if (cdev_add(port->sm_dev, base_dev + port->dev_num + IB_UMAD_MAX_PORTS, 1))
791		goto err_sm_cdev;
792
793	port->sm_class_dev = class_device_create(umad_class, NULL, port->sm_dev->dev,
794						 device->dma_device,
795						 "issm%d", port->dev_num);
796	if (IS_ERR(port->sm_class_dev))
797		goto err_sm_cdev;
798
799	class_set_devdata(port->class_dev,    port);
800	class_set_devdata(port->sm_class_dev, port);
801
802	if (class_device_create_file(port->sm_class_dev, &class_device_attr_ibdev))
803		goto err_sm_class;
804	if (class_device_create_file(port->sm_class_dev, &class_device_attr_port))
805		goto err_sm_class;
806
807	spin_lock(&port_lock);
808	umad_port[port->dev_num] = port;
809	spin_unlock(&port_lock);
810
811	return 0;
812
813err_sm_class:
814	class_device_destroy(umad_class, port->sm_dev->dev);
815
816err_sm_cdev:
817	cdev_del(port->sm_dev);
818
819err_class:
820	class_device_destroy(umad_class, port->dev->dev);
821
822err_cdev:
823	cdev_del(port->dev);
824	clear_bit(port->dev_num, dev_map);
825
826	return -1;
827}
828
829static void ib_umad_kill_port(struct ib_umad_port *port)
830{
831	struct ib_umad_file *file;
832	int id;
833
834	class_set_devdata(port->class_dev,    NULL);
835	class_set_devdata(port->sm_class_dev, NULL);
836
837	class_device_destroy(umad_class, port->dev->dev);
838	class_device_destroy(umad_class, port->sm_dev->dev);
839
840	cdev_del(port->dev);
841	cdev_del(port->sm_dev);
842
843	spin_lock(&port_lock);
844	umad_port[port->dev_num] = NULL;
845	spin_unlock(&port_lock);
846
847	down_write(&port->mutex);
848
849	port->ib_dev = NULL;
850
851	list_for_each_entry(file, &port->file_list, port_list)
852		for (id = 0; id < IB_UMAD_MAX_AGENTS; ++id) {
853			if (!file->agent[id])
854				continue;
855			ib_unregister_mad_agent(file->agent[id]);
856			file->agent[id] = NULL;
857		}
858
859	up_write(&port->mutex);
860
861	clear_bit(port->dev_num, dev_map);
862}
863
864static void ib_umad_add_one(struct ib_device *device)
865{
866	struct ib_umad_device *umad_dev;
867	int s, e, i;
868
869	if (device->node_type == IB_NODE_SWITCH)
870		s = e = 0;
871	else {
872		s = 1;
873		e = device->phys_port_cnt;
874	}
875
876	umad_dev = kzalloc(sizeof *umad_dev +
877			   (e - s + 1) * sizeof (struct ib_umad_port),
878			   GFP_KERNEL);
879	if (!umad_dev)
880		return;
881
882	kref_init(&umad_dev->ref);
883
884	umad_dev->start_port = s;
885	umad_dev->end_port   = e;
886
887	for (i = s; i <= e; ++i) {
888		umad_dev->port[i - s].umad_dev = umad_dev;
889
890		if (ib_umad_init_port(device, i, &umad_dev->port[i - s]))
891			goto err;
892	}
893
894	ib_set_client_data(device, &umad_client, umad_dev);
895
896	return;
897
898err:
899	while (--i >= s)
900		ib_umad_kill_port(&umad_dev->port[i - s]);
901
902	kref_put(&umad_dev->ref, ib_umad_release_dev);
903}
904
905static void ib_umad_remove_one(struct ib_device *device)
906{
907	struct ib_umad_device *umad_dev = ib_get_client_data(device, &umad_client);
908	int i;
909
910	if (!umad_dev)
911		return;
912
913	for (i = 0; i <= umad_dev->end_port - umad_dev->start_port; ++i)
914		ib_umad_kill_port(&umad_dev->port[i]);
915
916	kref_put(&umad_dev->ref, ib_umad_release_dev);
917}
918
919static int __init ib_umad_init(void)
920{
921	int ret;
922
923	ret = register_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2,
924				     "infiniband_mad");
925	if (ret) {
926		printk(KERN_ERR "user_mad: couldn't register device number\n");
927		goto out;
928	}
929
930	umad_class = class_create(THIS_MODULE, "infiniband_mad");
931	if (IS_ERR(umad_class)) {
932		ret = PTR_ERR(umad_class);
933		printk(KERN_ERR "user_mad: couldn't create class infiniband_mad\n");
934		goto out_chrdev;
935	}
936
937	ret = class_create_file(umad_class, &class_attr_abi_version);
938	if (ret) {
939		printk(KERN_ERR "user_mad: couldn't create abi_version attribute\n");
940		goto out_class;
941	}
942
943	ret = ib_register_client(&umad_client);
944	if (ret) {
945		printk(KERN_ERR "user_mad: couldn't register ib_umad client\n");
946		goto out_class;
947	}
948
949	return 0;
950
951out_class:
952	class_destroy(umad_class);
953
954out_chrdev:
955	unregister_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2);
956
957out:
958	return ret;
959}
960
961static void __exit ib_umad_cleanup(void)
962{
963	ib_unregister_client(&umad_client);
964	class_destroy(umad_class);
965	unregister_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2);
966}
967
968module_init(ib_umad_init);
969module_exit(ib_umad_cleanup);
970