user_mad.c revision 6473d160b4aba8023bcf38519a5989694dfd51a7
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 5596 2006-03-03 01:00:07Z sean.hefty $
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/dma-mapping.h>
44#include <linux/poll.h>
45#include <linux/rwsem.h>
46#include <linux/kref.h>
47
48#include <asm/uaccess.h>
49#include <asm/semaphore.h>
50
51#include <rdma/ib_mad.h>
52#include <rdma/ib_user_mad.h>
53
54MODULE_AUTHOR("Roland Dreier");
55MODULE_DESCRIPTION("InfiniBand userspace MAD packet access");
56MODULE_LICENSE("Dual BSD/GPL");
57
58enum {
59	IB_UMAD_MAX_PORTS  = 64,
60	IB_UMAD_MAX_AGENTS = 32,
61
62	IB_UMAD_MAJOR      = 231,
63	IB_UMAD_MINOR_BASE = 0
64};
65
66/*
67 * Our lifetime rules for these structs are the following: each time a
68 * device special file is opened, we look up the corresponding struct
69 * ib_umad_port by minor in the umad_port[] table while holding the
70 * port_lock.  If this lookup succeeds, we take a reference on the
71 * ib_umad_port's struct ib_umad_device while still holding the
72 * port_lock; if the lookup fails, we fail the open().  We drop these
73 * references in the corresponding close().
74 *
75 * In addition to references coming from open character devices, there
76 * is one more reference to each ib_umad_device representing the
77 * module's reference taken when allocating the ib_umad_device in
78 * ib_umad_add_one().
79 *
80 * When destroying an ib_umad_device, we clear all of its
81 * ib_umad_ports from umad_port[] while holding port_lock before
82 * dropping the module's reference to the ib_umad_device.  This is
83 * always safe because any open() calls will either succeed and obtain
84 * a reference before we clear the umad_port[] entries, or fail after
85 * we clear the umad_port[] entries.
86 */
87
88struct ib_umad_port {
89	struct cdev           *dev;
90	struct class_device   *class_dev;
91
92	struct cdev           *sm_dev;
93	struct class_device   *sm_class_dev;
94	struct semaphore       sm_sem;
95
96	struct rw_semaphore    mutex;
97	struct list_head       file_list;
98
99	struct ib_device      *ib_dev;
100	struct ib_umad_device *umad_dev;
101	int                    dev_num;
102	u8                     port_num;
103};
104
105struct ib_umad_device {
106	int                  start_port, end_port;
107	struct kref          ref;
108	struct ib_umad_port  port[0];
109};
110
111struct ib_umad_file {
112	struct ib_umad_port    *port;
113	struct list_head	recv_list;
114	struct list_head	send_list;
115	struct list_head	port_list;
116	spinlock_t		recv_lock;
117	spinlock_t		send_lock;
118	wait_queue_head_t	recv_wait;
119	struct ib_mad_agent    *agent[IB_UMAD_MAX_AGENTS];
120	int			agents_dead;
121};
122
123struct ib_umad_packet {
124	struct ib_mad_send_buf *msg;
125	struct ib_mad_recv_wc  *recv_wc;
126	struct list_head   list;
127	int		   length;
128	struct ib_user_mad mad;
129};
130
131static struct class *umad_class;
132
133static const dev_t base_dev = MKDEV(IB_UMAD_MAJOR, IB_UMAD_MINOR_BASE);
134
135static DEFINE_SPINLOCK(port_lock);
136static struct ib_umad_port *umad_port[IB_UMAD_MAX_PORTS];
137static DECLARE_BITMAP(dev_map, IB_UMAD_MAX_PORTS);
138
139static void ib_umad_add_one(struct ib_device *device);
140static void ib_umad_remove_one(struct ib_device *device);
141
142static void ib_umad_release_dev(struct kref *ref)
143{
144	struct ib_umad_device *dev =
145		container_of(ref, struct ib_umad_device, ref);
146
147	kfree(dev);
148}
149
150/* caller must hold port->mutex at least for reading */
151static struct ib_mad_agent *__get_agent(struct ib_umad_file *file, int id)
152{
153	return file->agents_dead ? NULL : file->agent[id];
154}
155
156static int queue_packet(struct ib_umad_file *file,
157			struct ib_mad_agent *agent,
158			struct ib_umad_packet *packet)
159{
160	int ret = 1;
161
162	down_read(&file->port->mutex);
163
164	for (packet->mad.hdr.id = 0;
165	     packet->mad.hdr.id < IB_UMAD_MAX_AGENTS;
166	     packet->mad.hdr.id++)
167		if (agent == __get_agent(file, packet->mad.hdr.id)) {
168			spin_lock_irq(&file->recv_lock);
169			list_add_tail(&packet->list, &file->recv_list);
170			spin_unlock_irq(&file->recv_lock);
171			wake_up_interruptible(&file->recv_wait);
172			ret = 0;
173			break;
174		}
175
176	up_read(&file->port->mutex);
177
178	return ret;
179}
180
181static void dequeue_send(struct ib_umad_file *file,
182			 struct ib_umad_packet *packet)
183 {
184	spin_lock_irq(&file->send_lock);
185	list_del(&packet->list);
186	spin_unlock_irq(&file->send_lock);
187 }
188
189static void send_handler(struct ib_mad_agent *agent,
190			 struct ib_mad_send_wc *send_wc)
191{
192	struct ib_umad_file *file = agent->context;
193	struct ib_umad_packet *packet = send_wc->send_buf->context[0];
194
195	dequeue_send(file, packet);
196	ib_destroy_ah(packet->msg->ah);
197	ib_free_send_mad(packet->msg);
198
199	if (send_wc->status == IB_WC_RESP_TIMEOUT_ERR) {
200		packet->length = IB_MGMT_MAD_HDR;
201		packet->mad.hdr.status = ETIMEDOUT;
202		if (!queue_packet(file, agent, packet))
203			return;
204	}
205	kfree(packet);
206}
207
208static void recv_handler(struct ib_mad_agent *agent,
209			 struct ib_mad_recv_wc *mad_recv_wc)
210{
211	struct ib_umad_file *file = agent->context;
212	struct ib_umad_packet *packet;
213
214	if (mad_recv_wc->wc->status != IB_WC_SUCCESS)
215		goto err1;
216
217	packet = kzalloc(sizeof *packet, GFP_KERNEL);
218	if (!packet)
219		goto err1;
220
221	packet->length = mad_recv_wc->mad_len;
222	packet->recv_wc = mad_recv_wc;
223
224	packet->mad.hdr.status    = 0;
225	packet->mad.hdr.length    = sizeof (struct ib_user_mad) +
226				    mad_recv_wc->mad_len;
227	packet->mad.hdr.qpn 	  = cpu_to_be32(mad_recv_wc->wc->src_qp);
228	packet->mad.hdr.lid 	  = cpu_to_be16(mad_recv_wc->wc->slid);
229	packet->mad.hdr.sl  	  = mad_recv_wc->wc->sl;
230	packet->mad.hdr.path_bits = mad_recv_wc->wc->dlid_path_bits;
231	packet->mad.hdr.grh_present = !!(mad_recv_wc->wc->wc_flags & IB_WC_GRH);
232	if (packet->mad.hdr.grh_present) {
233		struct ib_ah_attr ah_attr;
234
235		ib_init_ah_from_wc(agent->device, agent->port_num,
236				   mad_recv_wc->wc, mad_recv_wc->recv_buf.grh,
237				   &ah_attr);
238
239		packet->mad.hdr.gid_index = ah_attr.grh.sgid_index;
240		packet->mad.hdr.hop_limit = ah_attr.grh.hop_limit;
241		packet->mad.hdr.traffic_class = ah_attr.grh.traffic_class;
242		memcpy(packet->mad.hdr.gid, &ah_attr.grh.dgid, 16);
243		packet->mad.hdr.flow_label = cpu_to_be32(ah_attr.grh.flow_label);
244	}
245
246	if (queue_packet(file, agent, packet))
247		goto err2;
248	return;
249
250err2:
251	kfree(packet);
252err1:
253	ib_free_recv_mad(mad_recv_wc);
254}
255
256static ssize_t copy_recv_mad(char __user *buf, struct ib_umad_packet *packet,
257			     size_t count)
258{
259	struct ib_mad_recv_buf *recv_buf;
260	int left, seg_payload, offset, max_seg_payload;
261
262	/* We need enough room to copy the first (or only) MAD segment. */
263	recv_buf = &packet->recv_wc->recv_buf;
264	if ((packet->length <= sizeof (*recv_buf->mad) &&
265	     count < sizeof (packet->mad) + packet->length) ||
266	    (packet->length > sizeof (*recv_buf->mad) &&
267	     count < sizeof (packet->mad) + sizeof (*recv_buf->mad)))
268		return -EINVAL;
269
270	if (copy_to_user(buf, &packet->mad, sizeof (packet->mad)))
271		return -EFAULT;
272
273	buf += sizeof (packet->mad);
274	seg_payload = min_t(int, packet->length, sizeof (*recv_buf->mad));
275	if (copy_to_user(buf, recv_buf->mad, seg_payload))
276		return -EFAULT;
277
278	if (seg_payload < packet->length) {
279		/*
280		 * Multipacket RMPP MAD message. Copy remainder of message.
281		 * Note that last segment may have a shorter payload.
282		 */
283		if (count < sizeof (packet->mad) + packet->length) {
284			/*
285			 * The buffer is too small, return the first RMPP segment,
286			 * which includes the RMPP message length.
287			 */
288			return -ENOSPC;
289		}
290		offset = ib_get_mad_data_offset(recv_buf->mad->mad_hdr.mgmt_class);
291		max_seg_payload = sizeof (struct ib_mad) - offset;
292
293		for (left = packet->length - seg_payload, buf += seg_payload;
294		     left; left -= seg_payload, buf += seg_payload) {
295			recv_buf = container_of(recv_buf->list.next,
296						struct ib_mad_recv_buf, list);
297			seg_payload = min(left, max_seg_payload);
298			if (copy_to_user(buf, ((void *) recv_buf->mad) + offset,
299					 seg_payload))
300				return -EFAULT;
301		}
302	}
303	return sizeof (packet->mad) + packet->length;
304}
305
306static ssize_t copy_send_mad(char __user *buf, struct ib_umad_packet *packet,
307			     size_t count)
308{
309	ssize_t size = sizeof (packet->mad) + packet->length;
310
311	if (count < size)
312		return -EINVAL;
313
314	if (copy_to_user(buf, &packet->mad, size))
315		return -EFAULT;
316
317	return size;
318}
319
320static ssize_t ib_umad_read(struct file *filp, char __user *buf,
321			    size_t count, loff_t *pos)
322{
323	struct ib_umad_file *file = filp->private_data;
324	struct ib_umad_packet *packet;
325	ssize_t ret;
326
327	if (count < sizeof (struct ib_user_mad))
328		return -EINVAL;
329
330	spin_lock_irq(&file->recv_lock);
331
332	while (list_empty(&file->recv_list)) {
333		spin_unlock_irq(&file->recv_lock);
334
335		if (filp->f_flags & O_NONBLOCK)
336			return -EAGAIN;
337
338		if (wait_event_interruptible(file->recv_wait,
339					     !list_empty(&file->recv_list)))
340			return -ERESTARTSYS;
341
342		spin_lock_irq(&file->recv_lock);
343	}
344
345	packet = list_entry(file->recv_list.next, struct ib_umad_packet, list);
346	list_del(&packet->list);
347
348	spin_unlock_irq(&file->recv_lock);
349
350	if (packet->recv_wc)
351		ret = copy_recv_mad(buf, packet, count);
352	else
353		ret = copy_send_mad(buf, packet, count);
354
355	if (ret < 0) {
356		/* Requeue packet */
357		spin_lock_irq(&file->recv_lock);
358		list_add(&packet->list, &file->recv_list);
359		spin_unlock_irq(&file->recv_lock);
360	} else {
361		if (packet->recv_wc)
362			ib_free_recv_mad(packet->recv_wc);
363		kfree(packet);
364	}
365	return ret;
366}
367
368static int copy_rmpp_mad(struct ib_mad_send_buf *msg, const char __user *buf)
369{
370	int left, seg;
371
372	/* Copy class specific header */
373	if ((msg->hdr_len > IB_MGMT_RMPP_HDR) &&
374	    copy_from_user(msg->mad + IB_MGMT_RMPP_HDR, buf + IB_MGMT_RMPP_HDR,
375			   msg->hdr_len - IB_MGMT_RMPP_HDR))
376		return -EFAULT;
377
378	/* All headers are in place.  Copy data segments. */
379	for (seg = 1, left = msg->data_len, buf += msg->hdr_len; left > 0;
380	     seg++, left -= msg->seg_size, buf += msg->seg_size) {
381		if (copy_from_user(ib_get_rmpp_segment(msg, seg), buf,
382				   min(left, msg->seg_size)))
383			return -EFAULT;
384	}
385	return 0;
386}
387
388static int same_destination(struct ib_user_mad_hdr *hdr1,
389			    struct ib_user_mad_hdr *hdr2)
390{
391	if (!hdr1->grh_present && !hdr2->grh_present)
392	   return (hdr1->lid == hdr2->lid);
393
394	if (hdr1->grh_present && hdr2->grh_present)
395	   return !memcmp(hdr1->gid, hdr2->gid, 16);
396
397	return 0;
398}
399
400static int is_duplicate(struct ib_umad_file *file,
401			struct ib_umad_packet *packet)
402{
403	struct ib_umad_packet *sent_packet;
404	struct ib_mad_hdr *sent_hdr, *hdr;
405
406	hdr = (struct ib_mad_hdr *) packet->mad.data;
407	list_for_each_entry(sent_packet, &file->send_list, list) {
408		sent_hdr = (struct ib_mad_hdr *) sent_packet->mad.data;
409
410		if ((hdr->tid != sent_hdr->tid) ||
411		    (hdr->mgmt_class != sent_hdr->mgmt_class))
412			continue;
413
414		/*
415		 * No need to be overly clever here.  If two new operations have
416		 * the same TID, reject the second as a duplicate.  This is more
417		 * restrictive than required by the spec.
418		 */
419		if (!ib_response_mad((struct ib_mad *) hdr)) {
420			if (!ib_response_mad((struct ib_mad *) sent_hdr))
421				return 1;
422			continue;
423		} else if (!ib_response_mad((struct ib_mad *) sent_hdr))
424			continue;
425
426		if (same_destination(&packet->mad.hdr, &sent_packet->mad.hdr))
427			return 1;
428	}
429
430	return 0;
431}
432
433static ssize_t ib_umad_write(struct file *filp, const char __user *buf,
434			     size_t count, loff_t *pos)
435{
436	struct ib_umad_file *file = filp->private_data;
437	struct ib_umad_packet *packet;
438	struct ib_mad_agent *agent;
439	struct ib_ah_attr ah_attr;
440	struct ib_ah *ah;
441	struct ib_rmpp_mad *rmpp_mad;
442	__be64 *tid;
443	int ret, data_len, hdr_len, copy_offset, rmpp_active;
444
445	if (count < sizeof (struct ib_user_mad) + IB_MGMT_RMPP_HDR)
446		return -EINVAL;
447
448	packet = kzalloc(sizeof *packet + IB_MGMT_RMPP_HDR, GFP_KERNEL);
449	if (!packet)
450		return -ENOMEM;
451
452	if (copy_from_user(&packet->mad, buf,
453			    sizeof (struct ib_user_mad) + IB_MGMT_RMPP_HDR)) {
454		ret = -EFAULT;
455		goto err;
456	}
457
458	if (packet->mad.hdr.id < 0 ||
459	    packet->mad.hdr.id >= IB_UMAD_MAX_AGENTS) {
460		ret = -EINVAL;
461		goto err;
462	}
463
464	down_read(&file->port->mutex);
465
466	agent = __get_agent(file, packet->mad.hdr.id);
467	if (!agent) {
468		ret = -EINVAL;
469		goto err_up;
470	}
471
472	memset(&ah_attr, 0, sizeof ah_attr);
473	ah_attr.dlid          = be16_to_cpu(packet->mad.hdr.lid);
474	ah_attr.sl            = packet->mad.hdr.sl;
475	ah_attr.src_path_bits = packet->mad.hdr.path_bits;
476	ah_attr.port_num      = file->port->port_num;
477	if (packet->mad.hdr.grh_present) {
478		ah_attr.ah_flags = IB_AH_GRH;
479		memcpy(ah_attr.grh.dgid.raw, packet->mad.hdr.gid, 16);
480		ah_attr.grh.sgid_index	   = packet->mad.hdr.gid_index;
481		ah_attr.grh.flow_label 	   = be32_to_cpu(packet->mad.hdr.flow_label);
482		ah_attr.grh.hop_limit  	   = packet->mad.hdr.hop_limit;
483		ah_attr.grh.traffic_class  = packet->mad.hdr.traffic_class;
484	}
485
486	ah = ib_create_ah(agent->qp->pd, &ah_attr);
487	if (IS_ERR(ah)) {
488		ret = PTR_ERR(ah);
489		goto err_up;
490	}
491
492	rmpp_mad = (struct ib_rmpp_mad *) packet->mad.data;
493	hdr_len = ib_get_mad_data_offset(rmpp_mad->mad_hdr.mgmt_class);
494	if (!ib_is_mad_class_rmpp(rmpp_mad->mad_hdr.mgmt_class)) {
495		copy_offset = IB_MGMT_MAD_HDR;
496		rmpp_active = 0;
497	} else {
498		copy_offset = IB_MGMT_RMPP_HDR;
499		rmpp_active = ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) &
500			      IB_MGMT_RMPP_FLAG_ACTIVE;
501	}
502
503	data_len = count - sizeof (struct ib_user_mad) - hdr_len;
504	packet->msg = ib_create_send_mad(agent,
505					 be32_to_cpu(packet->mad.hdr.qpn),
506					 0, rmpp_active, hdr_len,
507					 data_len, GFP_KERNEL);
508	if (IS_ERR(packet->msg)) {
509		ret = PTR_ERR(packet->msg);
510		goto err_ah;
511	}
512
513	packet->msg->ah 	= ah;
514	packet->msg->timeout_ms = packet->mad.hdr.timeout_ms;
515	packet->msg->retries 	= packet->mad.hdr.retries;
516	packet->msg->context[0] = packet;
517
518	/* Copy MAD header.  Any RMPP header is already in place. */
519	memcpy(packet->msg->mad, packet->mad.data, IB_MGMT_MAD_HDR);
520	buf += sizeof (struct ib_user_mad);
521
522	if (!rmpp_active) {
523		if (copy_from_user(packet->msg->mad + copy_offset,
524				   buf + copy_offset,
525				   hdr_len + data_len - copy_offset)) {
526			ret = -EFAULT;
527			goto err_msg;
528		}
529	} else {
530		ret = copy_rmpp_mad(packet->msg, buf);
531		if (ret)
532			goto err_msg;
533	}
534
535	/*
536	 * Set the high-order part of the transaction ID to make MADs from
537	 * different agents unique, and allow routing responses back to the
538	 * original requestor.
539	 */
540	if (!ib_response_mad(packet->msg->mad)) {
541		tid = &((struct ib_mad_hdr *) packet->msg->mad)->tid;
542		*tid = cpu_to_be64(((u64) agent->hi_tid) << 32 |
543				   (be64_to_cpup(tid) & 0xffffffff));
544		rmpp_mad->mad_hdr.tid = *tid;
545	}
546
547	spin_lock_irq(&file->send_lock);
548	ret = is_duplicate(file, packet);
549	if (!ret)
550		list_add_tail(&packet->list, &file->send_list);
551	spin_unlock_irq(&file->send_lock);
552	if (ret) {
553		ret = -EINVAL;
554		goto err_msg;
555	}
556
557	ret = ib_post_send_mad(packet->msg, NULL);
558	if (ret)
559		goto err_send;
560
561	up_read(&file->port->mutex);
562	return count;
563
564err_send:
565	dequeue_send(file, packet);
566err_msg:
567	ib_free_send_mad(packet->msg);
568err_ah:
569	ib_destroy_ah(ah);
570err_up:
571	up_read(&file->port->mutex);
572err:
573	kfree(packet);
574	return ret;
575}
576
577static unsigned int ib_umad_poll(struct file *filp, struct poll_table_struct *wait)
578{
579	struct ib_umad_file *file = filp->private_data;
580
581	/* we will always be able to post a MAD send */
582	unsigned int mask = POLLOUT | POLLWRNORM;
583
584	poll_wait(filp, &file->recv_wait, wait);
585
586	if (!list_empty(&file->recv_list))
587		mask |= POLLIN | POLLRDNORM;
588
589	return mask;
590}
591
592static int ib_umad_reg_agent(struct ib_umad_file *file, unsigned long arg)
593{
594	struct ib_user_mad_reg_req ureq;
595	struct ib_mad_reg_req req;
596	struct ib_mad_agent *agent;
597	int agent_id;
598	int ret;
599
600	down_write(&file->port->mutex);
601
602	if (!file->port->ib_dev) {
603		ret = -EPIPE;
604		goto out;
605	}
606
607	if (copy_from_user(&ureq, (void __user *) arg, sizeof ureq)) {
608		ret = -EFAULT;
609		goto out;
610	}
611
612	if (ureq.qpn != 0 && ureq.qpn != 1) {
613		ret = -EINVAL;
614		goto out;
615	}
616
617	for (agent_id = 0; agent_id < IB_UMAD_MAX_AGENTS; ++agent_id)
618		if (!__get_agent(file, agent_id))
619			goto found;
620
621	ret = -ENOMEM;
622	goto out;
623
624found:
625	if (ureq.mgmt_class) {
626		req.mgmt_class         = ureq.mgmt_class;
627		req.mgmt_class_version = ureq.mgmt_class_version;
628		memcpy(req.method_mask, ureq.method_mask, sizeof req.method_mask);
629		memcpy(req.oui,         ureq.oui,         sizeof req.oui);
630	}
631
632	agent = ib_register_mad_agent(file->port->ib_dev, file->port->port_num,
633				      ureq.qpn ? IB_QPT_GSI : IB_QPT_SMI,
634				      ureq.mgmt_class ? &req : NULL,
635				      ureq.rmpp_version,
636				      send_handler, recv_handler, file);
637	if (IS_ERR(agent)) {
638		ret = PTR_ERR(agent);
639		goto out;
640	}
641
642	if (put_user(agent_id,
643		     (u32 __user *) (arg + offsetof(struct ib_user_mad_reg_req, id)))) {
644		ret = -EFAULT;
645		ib_unregister_mad_agent(agent);
646		goto out;
647	}
648
649	file->agent[agent_id] = agent;
650	ret = 0;
651
652out:
653	up_write(&file->port->mutex);
654	return ret;
655}
656
657static int ib_umad_unreg_agent(struct ib_umad_file *file, unsigned long arg)
658{
659	struct ib_mad_agent *agent = NULL;
660	u32 id;
661	int ret = 0;
662
663	if (get_user(id, (u32 __user *) arg))
664		return -EFAULT;
665
666	down_write(&file->port->mutex);
667
668	if (id < 0 || id >= IB_UMAD_MAX_AGENTS || !__get_agent(file, id)) {
669		ret = -EINVAL;
670		goto out;
671	}
672
673	agent = file->agent[id];
674	file->agent[id] = NULL;
675
676out:
677	up_write(&file->port->mutex);
678
679	if (agent)
680		ib_unregister_mad_agent(agent);
681
682	return ret;
683}
684
685static long ib_umad_ioctl(struct file *filp, unsigned int cmd,
686			  unsigned long arg)
687{
688	switch (cmd) {
689	case IB_USER_MAD_REGISTER_AGENT:
690		return ib_umad_reg_agent(filp->private_data, arg);
691	case IB_USER_MAD_UNREGISTER_AGENT:
692		return ib_umad_unreg_agent(filp->private_data, arg);
693	default:
694		return -ENOIOCTLCMD;
695	}
696}
697
698static int ib_umad_open(struct inode *inode, struct file *filp)
699{
700	struct ib_umad_port *port;
701	struct ib_umad_file *file;
702	int ret = 0;
703
704	spin_lock(&port_lock);
705	port = umad_port[iminor(inode) - IB_UMAD_MINOR_BASE];
706	if (port)
707		kref_get(&port->umad_dev->ref);
708	spin_unlock(&port_lock);
709
710	if (!port)
711		return -ENXIO;
712
713	down_write(&port->mutex);
714
715	if (!port->ib_dev) {
716		ret = -ENXIO;
717		goto out;
718	}
719
720	file = kzalloc(sizeof *file, GFP_KERNEL);
721	if (!file) {
722		kref_put(&port->umad_dev->ref, ib_umad_release_dev);
723		ret = -ENOMEM;
724		goto out;
725	}
726
727	spin_lock_init(&file->recv_lock);
728	spin_lock_init(&file->send_lock);
729	INIT_LIST_HEAD(&file->recv_list);
730	INIT_LIST_HEAD(&file->send_list);
731	init_waitqueue_head(&file->recv_wait);
732
733	file->port = port;
734	filp->private_data = file;
735
736	list_add_tail(&file->port_list, &port->file_list);
737
738out:
739	up_write(&port->mutex);
740	return ret;
741}
742
743static int ib_umad_close(struct inode *inode, struct file *filp)
744{
745	struct ib_umad_file *file = filp->private_data;
746	struct ib_umad_device *dev = file->port->umad_dev;
747	struct ib_umad_packet *packet, *tmp;
748	int already_dead;
749	int i;
750
751	down_write(&file->port->mutex);
752
753	already_dead = file->agents_dead;
754	file->agents_dead = 1;
755
756	list_for_each_entry_safe(packet, tmp, &file->recv_list, list) {
757		if (packet->recv_wc)
758			ib_free_recv_mad(packet->recv_wc);
759		kfree(packet);
760	}
761
762	list_del(&file->port_list);
763
764	downgrade_write(&file->port->mutex);
765
766	if (!already_dead)
767		for (i = 0; i < IB_UMAD_MAX_AGENTS; ++i)
768			if (file->agent[i])
769				ib_unregister_mad_agent(file->agent[i]);
770
771	up_read(&file->port->mutex);
772
773	kfree(file);
774	kref_put(&dev->ref, ib_umad_release_dev);
775
776	return 0;
777}
778
779static const struct file_operations umad_fops = {
780	.owner 	 	= THIS_MODULE,
781	.read 	 	= ib_umad_read,
782	.write 	 	= ib_umad_write,
783	.poll 	 	= ib_umad_poll,
784	.unlocked_ioctl = ib_umad_ioctl,
785	.compat_ioctl 	= ib_umad_ioctl,
786	.open 	 	= ib_umad_open,
787	.release 	= ib_umad_close
788};
789
790static int ib_umad_sm_open(struct inode *inode, struct file *filp)
791{
792	struct ib_umad_port *port;
793	struct ib_port_modify props = {
794		.set_port_cap_mask = IB_PORT_SM
795	};
796	int ret;
797
798	spin_lock(&port_lock);
799	port = umad_port[iminor(inode) - IB_UMAD_MINOR_BASE - IB_UMAD_MAX_PORTS];
800	if (port)
801		kref_get(&port->umad_dev->ref);
802	spin_unlock(&port_lock);
803
804	if (!port)
805		return -ENXIO;
806
807	if (filp->f_flags & O_NONBLOCK) {
808		if (down_trylock(&port->sm_sem)) {
809			ret = -EAGAIN;
810			goto fail;
811		}
812	} else {
813		if (down_interruptible(&port->sm_sem)) {
814			ret = -ERESTARTSYS;
815			goto fail;
816		}
817	}
818
819	ret = ib_modify_port(port->ib_dev, port->port_num, 0, &props);
820	if (ret) {
821		up(&port->sm_sem);
822		goto fail;
823	}
824
825	filp->private_data = port;
826
827	return 0;
828
829fail:
830	kref_put(&port->umad_dev->ref, ib_umad_release_dev);
831	return ret;
832}
833
834static int ib_umad_sm_close(struct inode *inode, struct file *filp)
835{
836	struct ib_umad_port *port = filp->private_data;
837	struct ib_port_modify props = {
838		.clr_port_cap_mask = IB_PORT_SM
839	};
840	int ret = 0;
841
842	down_write(&port->mutex);
843	if (port->ib_dev)
844		ret = ib_modify_port(port->ib_dev, port->port_num, 0, &props);
845	up_write(&port->mutex);
846
847	up(&port->sm_sem);
848
849	kref_put(&port->umad_dev->ref, ib_umad_release_dev);
850
851	return ret;
852}
853
854static const struct file_operations umad_sm_fops = {
855	.owner 	 = THIS_MODULE,
856	.open 	 = ib_umad_sm_open,
857	.release = ib_umad_sm_close
858};
859
860static struct ib_client umad_client = {
861	.name   = "umad",
862	.add    = ib_umad_add_one,
863	.remove = ib_umad_remove_one
864};
865
866static ssize_t show_ibdev(struct class_device *class_dev, char *buf)
867{
868	struct ib_umad_port *port = class_get_devdata(class_dev);
869
870	if (!port)
871		return -ENODEV;
872
873	return sprintf(buf, "%s\n", port->ib_dev->name);
874}
875static CLASS_DEVICE_ATTR(ibdev, S_IRUGO, show_ibdev, NULL);
876
877static ssize_t show_port(struct class_device *class_dev, char *buf)
878{
879	struct ib_umad_port *port = class_get_devdata(class_dev);
880
881	if (!port)
882		return -ENODEV;
883
884	return sprintf(buf, "%d\n", port->port_num);
885}
886static CLASS_DEVICE_ATTR(port, S_IRUGO, show_port, NULL);
887
888static ssize_t show_abi_version(struct class *class, char *buf)
889{
890	return sprintf(buf, "%d\n", IB_USER_MAD_ABI_VERSION);
891}
892static CLASS_ATTR(abi_version, S_IRUGO, show_abi_version, NULL);
893
894static int ib_umad_init_port(struct ib_device *device, int port_num,
895			     struct ib_umad_port *port)
896{
897	spin_lock(&port_lock);
898	port->dev_num = find_first_zero_bit(dev_map, IB_UMAD_MAX_PORTS);
899	if (port->dev_num >= IB_UMAD_MAX_PORTS) {
900		spin_unlock(&port_lock);
901		return -1;
902	}
903	set_bit(port->dev_num, dev_map);
904	spin_unlock(&port_lock);
905
906	port->ib_dev   = device;
907	port->port_num = port_num;
908	init_MUTEX(&port->sm_sem);
909	init_rwsem(&port->mutex);
910	INIT_LIST_HEAD(&port->file_list);
911
912	port->dev = cdev_alloc();
913	if (!port->dev)
914		return -1;
915	port->dev->owner = THIS_MODULE;
916	port->dev->ops   = &umad_fops;
917	kobject_set_name(&port->dev->kobj, "umad%d", port->dev_num);
918	if (cdev_add(port->dev, base_dev + port->dev_num, 1))
919		goto err_cdev;
920
921	port->class_dev = class_device_create(umad_class, NULL, port->dev->dev,
922					      device->dma_device,
923					      "umad%d", port->dev_num);
924	if (IS_ERR(port->class_dev))
925		goto err_cdev;
926
927	if (class_device_create_file(port->class_dev, &class_device_attr_ibdev))
928		goto err_class;
929	if (class_device_create_file(port->class_dev, &class_device_attr_port))
930		goto err_class;
931
932	port->sm_dev = cdev_alloc();
933	if (!port->sm_dev)
934		goto err_class;
935	port->sm_dev->owner = THIS_MODULE;
936	port->sm_dev->ops   = &umad_sm_fops;
937	kobject_set_name(&port->sm_dev->kobj, "issm%d", port->dev_num);
938	if (cdev_add(port->sm_dev, base_dev + port->dev_num + IB_UMAD_MAX_PORTS, 1))
939		goto err_sm_cdev;
940
941	port->sm_class_dev = class_device_create(umad_class, NULL, port->sm_dev->dev,
942						 device->dma_device,
943						 "issm%d", port->dev_num);
944	if (IS_ERR(port->sm_class_dev))
945		goto err_sm_cdev;
946
947	class_set_devdata(port->class_dev,    port);
948	class_set_devdata(port->sm_class_dev, port);
949
950	if (class_device_create_file(port->sm_class_dev, &class_device_attr_ibdev))
951		goto err_sm_class;
952	if (class_device_create_file(port->sm_class_dev, &class_device_attr_port))
953		goto err_sm_class;
954
955	spin_lock(&port_lock);
956	umad_port[port->dev_num] = port;
957	spin_unlock(&port_lock);
958
959	return 0;
960
961err_sm_class:
962	class_device_destroy(umad_class, port->sm_dev->dev);
963
964err_sm_cdev:
965	cdev_del(port->sm_dev);
966
967err_class:
968	class_device_destroy(umad_class, port->dev->dev);
969
970err_cdev:
971	cdev_del(port->dev);
972	clear_bit(port->dev_num, dev_map);
973
974	return -1;
975}
976
977static void ib_umad_kill_port(struct ib_umad_port *port)
978{
979	struct ib_umad_file *file;
980	int id;
981
982	class_set_devdata(port->class_dev,    NULL);
983	class_set_devdata(port->sm_class_dev, NULL);
984
985	class_device_destroy(umad_class, port->dev->dev);
986	class_device_destroy(umad_class, port->sm_dev->dev);
987
988	cdev_del(port->dev);
989	cdev_del(port->sm_dev);
990
991	spin_lock(&port_lock);
992	umad_port[port->dev_num] = NULL;
993	spin_unlock(&port_lock);
994
995	down_write(&port->mutex);
996
997	port->ib_dev = NULL;
998
999	/*
1000	 * Now go through the list of files attached to this port and
1001	 * unregister all of their MAD agents.  We need to hold
1002	 * port->mutex while doing this to avoid racing with
1003	 * ib_umad_close(), but we can't hold the mutex for writing
1004	 * while calling ib_unregister_mad_agent(), since that might
1005	 * deadlock by calling back into queue_packet().  So we
1006	 * downgrade our lock to a read lock, and then drop and
1007	 * reacquire the write lock for the next iteration.
1008	 *
1009	 * We do list_del_init() on the file's list_head so that the
1010	 * list_del in ib_umad_close() is still OK, even after the
1011	 * file is removed from the list.
1012	 */
1013	while (!list_empty(&port->file_list)) {
1014		file = list_entry(port->file_list.next, struct ib_umad_file,
1015				  port_list);
1016
1017		file->agents_dead = 1;
1018		list_del_init(&file->port_list);
1019
1020		downgrade_write(&port->mutex);
1021
1022		for (id = 0; id < IB_UMAD_MAX_AGENTS; ++id)
1023			if (file->agent[id])
1024				ib_unregister_mad_agent(file->agent[id]);
1025
1026		up_read(&port->mutex);
1027		down_write(&port->mutex);
1028	}
1029
1030	up_write(&port->mutex);
1031
1032	clear_bit(port->dev_num, dev_map);
1033}
1034
1035static void ib_umad_add_one(struct ib_device *device)
1036{
1037	struct ib_umad_device *umad_dev;
1038	int s, e, i;
1039
1040	if (rdma_node_get_transport(device->node_type) != RDMA_TRANSPORT_IB)
1041		return;
1042
1043	if (device->node_type == RDMA_NODE_IB_SWITCH)
1044		s = e = 0;
1045	else {
1046		s = 1;
1047		e = device->phys_port_cnt;
1048	}
1049
1050	umad_dev = kzalloc(sizeof *umad_dev +
1051			   (e - s + 1) * sizeof (struct ib_umad_port),
1052			   GFP_KERNEL);
1053	if (!umad_dev)
1054		return;
1055
1056	kref_init(&umad_dev->ref);
1057
1058	umad_dev->start_port = s;
1059	umad_dev->end_port   = e;
1060
1061	for (i = s; i <= e; ++i) {
1062		umad_dev->port[i - s].umad_dev = umad_dev;
1063
1064		if (ib_umad_init_port(device, i, &umad_dev->port[i - s]))
1065			goto err;
1066	}
1067
1068	ib_set_client_data(device, &umad_client, umad_dev);
1069
1070	return;
1071
1072err:
1073	while (--i >= s)
1074		ib_umad_kill_port(&umad_dev->port[i - s]);
1075
1076	kref_put(&umad_dev->ref, ib_umad_release_dev);
1077}
1078
1079static void ib_umad_remove_one(struct ib_device *device)
1080{
1081	struct ib_umad_device *umad_dev = ib_get_client_data(device, &umad_client);
1082	int i;
1083
1084	if (!umad_dev)
1085		return;
1086
1087	for (i = 0; i <= umad_dev->end_port - umad_dev->start_port; ++i)
1088		ib_umad_kill_port(&umad_dev->port[i]);
1089
1090	kref_put(&umad_dev->ref, ib_umad_release_dev);
1091}
1092
1093static int __init ib_umad_init(void)
1094{
1095	int ret;
1096
1097	ret = register_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2,
1098				     "infiniband_mad");
1099	if (ret) {
1100		printk(KERN_ERR "user_mad: couldn't register device number\n");
1101		goto out;
1102	}
1103
1104	umad_class = class_create(THIS_MODULE, "infiniband_mad");
1105	if (IS_ERR(umad_class)) {
1106		ret = PTR_ERR(umad_class);
1107		printk(KERN_ERR "user_mad: couldn't create class infiniband_mad\n");
1108		goto out_chrdev;
1109	}
1110
1111	ret = class_create_file(umad_class, &class_attr_abi_version);
1112	if (ret) {
1113		printk(KERN_ERR "user_mad: couldn't create abi_version attribute\n");
1114		goto out_class;
1115	}
1116
1117	ret = ib_register_client(&umad_client);
1118	if (ret) {
1119		printk(KERN_ERR "user_mad: couldn't register ib_umad client\n");
1120		goto out_class;
1121	}
1122
1123	return 0;
1124
1125out_class:
1126	class_destroy(umad_class);
1127
1128out_chrdev:
1129	unregister_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2);
1130
1131out:
1132	return ret;
1133}
1134
1135static void __exit ib_umad_cleanup(void)
1136{
1137	ib_unregister_client(&umad_client);
1138	class_destroy(umad_class);
1139	unregister_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2);
1140}
1141
1142module_init(ib_umad_init);
1143module_exit(ib_umad_cleanup);
1144