?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 2.6.34-rc3
[pohmelfs.git] / drivers / infiniband / core / user_mad.c
blob04b585e86cb210f9975c34104131d553de404979
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 #include <linux/module.h>
37 #include <linux/init.h>
38 #include <linux/device.h>
39 #include <linux/err.h>
40 #include <linux/fs.h>
41 #include <linux/cdev.h>
42 #include <linux/dma-mapping.h>
43 #include <linux/poll.h>
44 #include <linux/mutex.h>
45 #include <linux/kref.h>
46 #include <linux/compat.h>
47 #include <linux/sched.h>
48 #include <linux/semaphore.h>
49
50 #include <asm/uaccess.h>
51
52 #include <rdma/ib_mad.h>
53 #include <rdma/ib_user_mad.h>
54
55 MODULE_AUTHOR("Roland Dreier");
56 MODULE_DESCRIPTION("InfiniBand userspace MAD packet access");
57 MODULE_LICENSE("Dual BSD/GPL");
58
59 enum {
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:
69 * device special file is opened, we take a reference on the
70 * ib_umad_port's struct ib_umad_device. We drop these
71 * references in the corresponding close().
72 *
73 * In addition to references coming from open character devices, there
74 * is one more reference to each ib_umad_device representing the
75 * module's reference taken when allocating the ib_umad_device in
76 * ib_umad_add_one().
77 *
78 * When destroying an ib_umad_device, we drop the module's reference.
79 */
80
81 struct ib_umad_port {
82 struct cdev cdev;
83 struct device *dev;
84
85 struct cdev sm_cdev;
86 struct device *sm_dev;
87 struct semaphore sm_sem;
88
89 struct mutex file_mutex;
90 struct list_head file_list;
91
92 struct ib_device *ib_dev;
93 struct ib_umad_device *umad_dev;
94 int dev_num;
95 u8 port_num;
96 };
97
98 struct ib_umad_device {
99 int start_port, end_port;
100 struct kref ref;
101 struct ib_umad_port port[0];
102 };
103
104 struct ib_umad_file {
105 struct mutex mutex;
106 struct ib_umad_port *port;
107 struct list_head recv_list;
108 struct list_head send_list;
109 struct list_head port_list;
110 spinlock_t send_lock;
111 wait_queue_head_t recv_wait;
112 struct ib_mad_agent *agent[IB_UMAD_MAX_AGENTS];
113 int agents_dead;
114 u8 use_pkey_index;
115 u8 already_used;
116 };
117
118 struct ib_umad_packet {
119 struct ib_mad_send_buf *msg;
120 struct ib_mad_recv_wc *recv_wc;
121 struct list_head list;
122 int length;
123 struct ib_user_mad mad;
124 };
125
126 static struct class *umad_class;
127
128 static const dev_t base_dev = MKDEV(IB_UMAD_MAJOR, IB_UMAD_MINOR_BASE);
129
130 static DEFINE_SPINLOCK(port_lock);
131 static DECLARE_BITMAP(dev_map, IB_UMAD_MAX_PORTS);
132
133 static void ib_umad_add_one(struct ib_device *device);
134 static void ib_umad_remove_one(struct ib_device *device);
135
136 static void ib_umad_release_dev(struct kref *ref)
137 {
138 struct ib_umad_device *dev =
139 container_of(ref, struct ib_umad_device, ref);
140
141 kfree(dev);
142 }
143
144 static int hdr_size(struct ib_umad_file *file)
145 {
146 return file->use_pkey_index ? sizeof (struct ib_user_mad_hdr) :
147 sizeof (struct ib_user_mad_hdr_old);
148 }
149
150 /* caller must hold file->mutex */
151 static 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
156 static 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 mutex_lock(&file->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 list_add_tail(&packet->list, &file->recv_list);
169 wake_up_interruptible(&file->recv_wait);
170 ret = 0;
171 break;
172 }
173
174 mutex_unlock(&file->mutex);
175
176 return ret;
177 }
178
179 static void dequeue_send(struct ib_umad_file *file,
180 struct ib_umad_packet *packet)
181 {
182 spin_lock_irq(&file->send_lock);
183 list_del(&packet->list);
184 spin_unlock_irq(&file->send_lock);
185 }
186
187 static void send_handler(struct ib_mad_agent *agent,
188 struct ib_mad_send_wc *send_wc)
189 {
190 struct ib_umad_file *file = agent->context;
191 struct ib_umad_packet *packet = send_wc->send_buf->context[0];
192
193 dequeue_send(file, packet);
194 ib_destroy_ah(packet->msg->ah);
195 ib_free_send_mad(packet->msg);
196
197 if (send_wc->status == IB_WC_RESP_TIMEOUT_ERR) {
198 packet->length = IB_MGMT_MAD_HDR;
199 packet->mad.hdr.status = ETIMEDOUT;
200 if (!queue_packet(file, agent, packet))
201 return;
202 }
203 kfree(packet);
204 }
205
206 static void recv_handler(struct ib_mad_agent *agent,
207 struct ib_mad_recv_wc *mad_recv_wc)
208 {
209 struct ib_umad_file *file = agent->context;
210 struct ib_umad_packet *packet;
211
212 if (mad_recv_wc->wc->status != IB_WC_SUCCESS)
213 goto err1;
214
215 packet = kzalloc(sizeof *packet, GFP_KERNEL);
216 if (!packet)
217 goto err1;
218
219 packet->length = mad_recv_wc->mad_len;
220 packet->recv_wc = mad_recv_wc;
221
222 packet->mad.hdr.status = 0;
223 packet->mad.hdr.length = hdr_size(file) + mad_recv_wc->mad_len;
224 packet->mad.hdr.qpn = cpu_to_be32(mad_recv_wc->wc->src_qp);
225 packet->mad.hdr.lid = cpu_to_be16(mad_recv_wc->wc->slid);
226 packet->mad.hdr.sl = mad_recv_wc->wc->sl;
227 packet->mad.hdr.path_bits = mad_recv_wc->wc->dlid_path_bits;
228 packet->mad.hdr.pkey_index = mad_recv_wc->wc->pkey_index;
229 packet->mad.hdr.grh_present = !!(mad_recv_wc->wc->wc_flags & IB_WC_GRH);
230 if (packet->mad.hdr.grh_present) {
231 struct ib_ah_attr ah_attr;
232
233 ib_init_ah_from_wc(agent->device, agent->port_num,
234 mad_recv_wc->wc, mad_recv_wc->recv_buf.grh,
235 &ah_attr);
236
237 packet->mad.hdr.gid_index = ah_attr.grh.sgid_index;
238 packet->mad.hdr.hop_limit = ah_attr.grh.hop_limit;
239 packet->mad.hdr.traffic_class = ah_attr.grh.traffic_class;
240 memcpy(packet->mad.hdr.gid, &ah_attr.grh.dgid, 16);
241 packet->mad.hdr.flow_label = cpu_to_be32(ah_attr.grh.flow_label);
242 }
243
244 if (queue_packet(file, agent, packet))
245 goto err2;
246 return;
247
248 err2:
249 kfree(packet);
250 err1:
251 ib_free_recv_mad(mad_recv_wc);
252 }
253
254 static ssize_t copy_recv_mad(struct ib_umad_file *file, char __user *buf,
255 struct ib_umad_packet *packet, size_t count)
256 {
257 struct ib_mad_recv_buf *recv_buf;
258 int left, seg_payload, offset, max_seg_payload;
259
260 /* We need enough room to copy the first (or only) MAD segment. */
261 recv_buf = &packet->recv_wc->recv_buf;
262 if ((packet->length <= sizeof (*recv_buf->mad) &&
263 count < hdr_size(file) + packet->length) ||
264 (packet->length > sizeof (*recv_buf->mad) &&
265 count < hdr_size(file) + sizeof (*recv_buf->mad)))
266 return -EINVAL;
267
268 if (copy_to_user(buf, &packet->mad, hdr_size(file)))
269 return -EFAULT;
270
271 buf += hdr_size(file);
272 seg_payload = min_t(int, packet->length, sizeof (*recv_buf->mad));
273 if (copy_to_user(buf, recv_buf->mad, seg_payload))
274 return -EFAULT;
275
276 if (seg_payload < packet->length) {
277 /*
278 * Multipacket RMPP MAD message. Copy remainder of message.
279 * Note that last segment may have a shorter payload.
280 */
281 if (count < hdr_size(file) + packet->length) {
282 /*
283 * The buffer is too small, return the first RMPP segment,
284 * which includes the RMPP message length.
285 */
286 return -ENOSPC;
287 }
288 offset = ib_get_mad_data_offset(recv_buf->mad->mad_hdr.mgmt_class);
289 max_seg_payload = sizeof (struct ib_mad) - offset;
290
291 for (left = packet->length - seg_payload, buf += seg_payload;
292 left; left -= seg_payload, buf += seg_payload) {
293 recv_buf = container_of(recv_buf->list.next,
294 struct ib_mad_recv_buf, list);
295 seg_payload = min(left, max_seg_payload);
296 if (copy_to_user(buf, ((void *) recv_buf->mad) + offset,
297 seg_payload))
298 return -EFAULT;
299 }
300 }
301 return hdr_size(file) + packet->length;
302 }
303
304 static ssize_t copy_send_mad(struct ib_umad_file *file, char __user *buf,
305 struct ib_umad_packet *packet, size_t count)
306 {
307 ssize_t size = hdr_size(file) + packet->length;
308
309 if (count < size)
310 return -EINVAL;
311
312 if (copy_to_user(buf, &packet->mad, hdr_size(file)))
313 return -EFAULT;
314
315 buf += hdr_size(file);
316
317 if (copy_to_user(buf, packet->mad.data, packet->length))
318 return -EFAULT;
319
320 return size;
321 }
322
323 static ssize_t ib_umad_read(struct file *filp, char __user *buf,
324 size_t count, loff_t *pos)
325 {
326 struct ib_umad_file *file = filp->private_data;
327 struct ib_umad_packet *packet;
328 ssize_t ret;
329
330 if (count < hdr_size(file))
331 return -EINVAL;
332
333 mutex_lock(&file->mutex);
334
335 while (list_empty(&file->recv_list)) {
336 mutex_unlock(&file->mutex);
337
338 if (filp->f_flags & O_NONBLOCK)
339 return -EAGAIN;
340
341 if (wait_event_interruptible(file->recv_wait,
342 !list_empty(&file->recv_list)))
343 return -ERESTARTSYS;
344
345 mutex_lock(&file->mutex);
346 }
347
348 packet = list_entry(file->recv_list.next, struct ib_umad_packet, list);
349 list_del(&packet->list);
350
351 mutex_unlock(&file->mutex);
352
353 if (packet->recv_wc)
354 ret = copy_recv_mad(file, buf, packet, count);
355 else
356 ret = copy_send_mad(file, buf, packet, count);
357
358 if (ret < 0) {
359 /* Requeue packet */
360 mutex_lock(&file->mutex);
361 list_add(&packet->list, &file->recv_list);
362 mutex_unlock(&file->mutex);
363 } else {
364 if (packet->recv_wc)
365 ib_free_recv_mad(packet->recv_wc);
366 kfree(packet);
367 }
368 return ret;
369 }
370
371 static int copy_rmpp_mad(struct ib_mad_send_buf *msg, const char __user *buf)
372 {
373 int left, seg;
374
375 /* Copy class specific header */
376 if ((msg->hdr_len > IB_MGMT_RMPP_HDR) &&
377 copy_from_user(msg->mad + IB_MGMT_RMPP_HDR, buf + IB_MGMT_RMPP_HDR,
378 msg->hdr_len - IB_MGMT_RMPP_HDR))
379 return -EFAULT;
380
381 /* All headers are in place. Copy data segments. */
382 for (seg = 1, left = msg->data_len, buf += msg->hdr_len; left > 0;
383 seg++, left -= msg->seg_size, buf += msg->seg_size) {
384 if (copy_from_user(ib_get_rmpp_segment(msg, seg), buf,
385 min(left, msg->seg_size)))
386 return -EFAULT;
387 }
388 return 0;
389 }
390
391 static int same_destination(struct ib_user_mad_hdr *hdr1,
392 struct ib_user_mad_hdr *hdr2)
393 {
394 if (!hdr1->grh_present && !hdr2->grh_present)
395 return (hdr1->lid == hdr2->lid);
396
397 if (hdr1->grh_present && hdr2->grh_present)
398 return !memcmp(hdr1->gid, hdr2->gid, 16);
399
400 return 0;
401 }
402
403 static int is_duplicate(struct ib_umad_file *file,
404 struct ib_umad_packet *packet)
405 {
406 struct ib_umad_packet *sent_packet;
407 struct ib_mad_hdr *sent_hdr, *hdr;
408
409 hdr = (struct ib_mad_hdr *) packet->mad.data;
410 list_for_each_entry(sent_packet, &file->send_list, list) {
411 sent_hdr = (struct ib_mad_hdr *) sent_packet->mad.data;
412
413 if ((hdr->tid != sent_hdr->tid) ||
414 (hdr->mgmt_class != sent_hdr->mgmt_class))
415 continue;
416
417 /*
418 * No need to be overly clever here. If two new operations have
419 * the same TID, reject the second as a duplicate. This is more
420 * restrictive than required by the spec.
421 */
422 if (!ib_response_mad((struct ib_mad *) hdr)) {
423 if (!ib_response_mad((struct ib_mad *) sent_hdr))
424 return 1;
425 continue;
426 } else if (!ib_response_mad((struct ib_mad *) sent_hdr))
427 continue;
428
429 if (same_destination(&packet->mad.hdr, &sent_packet->mad.hdr))
430 return 1;
431 }
432
433 return 0;
434 }
435
436 static ssize_t ib_umad_write(struct file *filp, const char __user *buf,
437 size_t count, loff_t *pos)
438 {
439 struct ib_umad_file *file = filp->private_data;
440 struct ib_umad_packet *packet;
441 struct ib_mad_agent *agent;
442 struct ib_ah_attr ah_attr;
443 struct ib_ah *ah;
444 struct ib_rmpp_mad *rmpp_mad;
445 __be64 *tid;
446 int ret, data_len, hdr_len, copy_offset, rmpp_active;
447
448 if (count < hdr_size(file) + IB_MGMT_RMPP_HDR)
449 return -EINVAL;
450
451 packet = kzalloc(sizeof *packet + IB_MGMT_RMPP_HDR, GFP_KERNEL);
452 if (!packet)
453 return -ENOMEM;
454
455 if (copy_from_user(&packet->mad, buf, hdr_size(file))) {
456 ret = -EFAULT;
457 goto err;
458 }
459
460 if (packet->mad.hdr.id < 0 ||
461 packet->mad.hdr.id >= IB_UMAD_MAX_AGENTS) {
462 ret = -EINVAL;
463 goto err;
464 }
465
466 buf += hdr_size(file);
467
468 if (copy_from_user(packet->mad.data, buf, IB_MGMT_RMPP_HDR)) {
469 ret = -EFAULT;
470 goto err;
471 }
472
473 mutex_lock(&file->mutex);
474
475 agent = __get_agent(file, packet->mad.hdr.id);
476 if (!agent) {
477 ret = -EINVAL;
478 goto err_up;
479 }
480
481 memset(&ah_attr, 0, sizeof ah_attr);
482 ah_attr.dlid = be16_to_cpu(packet->mad.hdr.lid);
483 ah_attr.sl = packet->mad.hdr.sl;
484 ah_attr.src_path_bits = packet->mad.hdr.path_bits;
485 ah_attr.port_num = file->port->port_num;
486 if (packet->mad.hdr.grh_present) {
487 ah_attr.ah_flags = IB_AH_GRH;
488 memcpy(ah_attr.grh.dgid.raw, packet->mad.hdr.gid, 16);
489 ah_attr.grh.sgid_index = packet->mad.hdr.gid_index;
490 ah_attr.grh.flow_label = be32_to_cpu(packet->mad.hdr.flow_label);
491 ah_attr.grh.hop_limit = packet->mad.hdr.hop_limit;
492 ah_attr.grh.traffic_class = packet->mad.hdr.traffic_class;
493 }
494
495 ah = ib_create_ah(agent->qp->pd, &ah_attr);
496 if (IS_ERR(ah)) {
497 ret = PTR_ERR(ah);
498 goto err_up;
499 }
500
501 rmpp_mad = (struct ib_rmpp_mad *) packet->mad.data;
502 hdr_len = ib_get_mad_data_offset(rmpp_mad->mad_hdr.mgmt_class);
503 if (!ib_is_mad_class_rmpp(rmpp_mad->mad_hdr.mgmt_class)) {
504 copy_offset = IB_MGMT_MAD_HDR;
505 rmpp_active = 0;
506 } else {
507 copy_offset = IB_MGMT_RMPP_HDR;
508 rmpp_active = ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) &
509 IB_MGMT_RMPP_FLAG_ACTIVE;
510 }
511
512 data_len = count - hdr_size(file) - hdr_len;
513 packet->msg = ib_create_send_mad(agent,
514 be32_to_cpu(packet->mad.hdr.qpn),
515 packet->mad.hdr.pkey_index, rmpp_active,
516 hdr_len, data_len, GFP_KERNEL);
517 if (IS_ERR(packet->msg)) {
518 ret = PTR_ERR(packet->msg);
519 goto err_ah;
520 }
521
522 packet->msg->ah = ah;
523 packet->msg->timeout_ms = packet->mad.hdr.timeout_ms;
524 packet->msg->retries = packet->mad.hdr.retries;
525 packet->msg->context[0] = packet;
526
527 /* Copy MAD header. Any RMPP header is already in place. */
528 memcpy(packet->msg->mad, packet->mad.data, IB_MGMT_MAD_HDR);
529
530 if (!rmpp_active) {
531 if (copy_from_user(packet->msg->mad + copy_offset,
532 buf + copy_offset,
533 hdr_len + data_len - copy_offset)) {
534 ret = -EFAULT;
535 goto err_msg;
536 }
537 } else {
538 ret = copy_rmpp_mad(packet->msg, buf);
539 if (ret)
540 goto err_msg;
541 }
542
543 /*
544 * Set the high-order part of the transaction ID to make MADs from
545 * different agents unique, and allow routing responses back to the
546 * original requestor.
547 */
548 if (!ib_response_mad(packet->msg->mad)) {
549 tid = &((struct ib_mad_hdr *) packet->msg->mad)->tid;
550 *tid = cpu_to_be64(((u64) agent->hi_tid) << 32 |
551 (be64_to_cpup(tid) & 0xffffffff));
552 rmpp_mad->mad_hdr.tid = *tid;
553 }
554
555 spin_lock_irq(&file->send_lock);
556 ret = is_duplicate(file, packet);
557 if (!ret)
558 list_add_tail(&packet->list, &file->send_list);
559 spin_unlock_irq(&file->send_lock);
560 if (ret) {
561 ret = -EINVAL;
562 goto err_msg;
563 }
564
565 ret = ib_post_send_mad(packet->msg, NULL);
566 if (ret)
567 goto err_send;
568
569 mutex_unlock(&file->mutex);
570 return count;
571
572 err_send:
573 dequeue_send(file, packet);
574 err_msg:
575 ib_free_send_mad(packet->msg);
576 err_ah:
577 ib_destroy_ah(ah);
578 err_up:
579 mutex_unlock(&file->mutex);
580 err:
581 kfree(packet);
582 return ret;
583 }
584
585 static unsigned int ib_umad_poll(struct file *filp, struct poll_table_struct *wait)
586 {
587 struct ib_umad_file *file = filp->private_data;
588
589 /* we will always be able to post a MAD send */
590 unsigned int mask = POLLOUT | POLLWRNORM;
591
592 poll_wait(filp, &file->recv_wait, wait);
593
594 if (!list_empty(&file->recv_list))
595 mask |= POLLIN | POLLRDNORM;
596
597 return mask;
598 }
599
600 static int ib_umad_reg_agent(struct ib_umad_file *file, void __user *arg,
601 int compat_method_mask)
602 {
603 struct ib_user_mad_reg_req ureq;
604 struct ib_mad_reg_req req;
605 struct ib_mad_agent *agent = NULL;
606 int agent_id;
607 int ret;
608
609 mutex_lock(&file->port->file_mutex);
610 mutex_lock(&file->mutex);
611
612 if (!file->port->ib_dev) {
613 ret = -EPIPE;
614 goto out;
615 }
616
617 if (copy_from_user(&ureq, arg, sizeof ureq)) {
618 ret = -EFAULT;
619 goto out;
620 }
621
622 if (ureq.qpn != 0 && ureq.qpn != 1) {
623 ret = -EINVAL;
624 goto out;
625 }
626
627 for (agent_id = 0; agent_id < IB_UMAD_MAX_AGENTS; ++agent_id)
628 if (!__get_agent(file, agent_id))
629 goto found;
630
631 ret = -ENOMEM;
632 goto out;
633
634 found:
635 if (ureq.mgmt_class) {
636 req.mgmt_class = ureq.mgmt_class;
637 req.mgmt_class_version = ureq.mgmt_class_version;
638 memcpy(req.oui, ureq.oui, sizeof req.oui);
639
640 if (compat_method_mask) {
641 u32 *umm = (u32 *) ureq.method_mask;
642 int i;
643
644 for (i = 0; i < BITS_TO_LONGS(IB_MGMT_MAX_METHODS); ++i)
645 req.method_mask[i] =
646 umm[i * 2] | ((u64) umm[i * 2 + 1] << 32);
647 } else
648 memcpy(req.method_mask, ureq.method_mask,
649 sizeof req.method_mask);
650 }
651
652 agent = ib_register_mad_agent(file->port->ib_dev, file->port->port_num,
653 ureq.qpn ? IB_QPT_GSI : IB_QPT_SMI,
654 ureq.mgmt_class ? &req : NULL,
655 ureq.rmpp_version,
656 send_handler, recv_handler, file);
657 if (IS_ERR(agent)) {
658 ret = PTR_ERR(agent);
659 agent = NULL;
660 goto out;
661 }
662
663 if (put_user(agent_id,
664 (u32 __user *) (arg + offsetof(struct ib_user_mad_reg_req, id)))) {
665 ret = -EFAULT;
666 goto out;
667 }
668
669 if (!file->already_used) {
670 file->already_used = 1;
671 if (!file->use_pkey_index) {
672 printk(KERN_WARNING "user_mad: process %s did not enable "
673 "P_Key index support.\n", current->comm);
674 printk(KERN_WARNING "user_mad: Documentation/infiniband/user_mad.txt "
675 "has info on the new ABI.\n");
676 }
677 }
678
679 file->agent[agent_id] = agent;
680 ret = 0;
681
682 out:
683 mutex_unlock(&file->mutex);
684
685 if (ret && agent)
686 ib_unregister_mad_agent(agent);
687
688 mutex_unlock(&file->port->file_mutex);
689
690 return ret;
691 }
692
693 static int ib_umad_unreg_agent(struct ib_umad_file *file, u32 __user *arg)
694 {
695 struct ib_mad_agent *agent = NULL;
696 u32 id;
697 int ret = 0;
698
699 if (get_user(id, arg))
700 return -EFAULT;
701
702 mutex_lock(&file->port->file_mutex);
703 mutex_lock(&file->mutex);
704
705 if (id < 0 || id >= IB_UMAD_MAX_AGENTS || !__get_agent(file, id)) {
706 ret = -EINVAL;
707 goto out;
708 }
709
710 agent = file->agent[id];
711 file->agent[id] = NULL;
712
713 out:
714 mutex_unlock(&file->mutex);
715
716 if (agent)
717 ib_unregister_mad_agent(agent);
718
719 mutex_unlock(&file->port->file_mutex);
720
721 return ret;
722 }
723
724 static long ib_umad_enable_pkey(struct ib_umad_file *file)
725 {
726 int ret = 0;
727
728 mutex_lock(&file->mutex);
729 if (file->already_used)
730 ret = -EINVAL;
731 else
732 file->use_pkey_index = 1;
733 mutex_unlock(&file->mutex);
734
735 return ret;
736 }
737
738 static long ib_umad_ioctl(struct file *filp, unsigned int cmd,
739 unsigned long arg)
740 {
741 switch (cmd) {
742 case IB_USER_MAD_REGISTER_AGENT:
743 return ib_umad_reg_agent(filp->private_data, (void __user *) arg, 0);
744 case IB_USER_MAD_UNREGISTER_AGENT:
745 return ib_umad_unreg_agent(filp->private_data, (__u32 __user *) arg);
746 case IB_USER_MAD_ENABLE_PKEY:
747 return ib_umad_enable_pkey(filp->private_data);
748 default:
749 return -ENOIOCTLCMD;
750 }
751 }
752
753 #ifdef CONFIG_COMPAT
754 static long ib_umad_compat_ioctl(struct file *filp, unsigned int cmd,
755 unsigned long arg)
756 {
757 switch (cmd) {
758 case IB_USER_MAD_REGISTER_AGENT:
759 return ib_umad_reg_agent(filp->private_data, compat_ptr(arg), 1);
760 case IB_USER_MAD_UNREGISTER_AGENT:
761 return ib_umad_unreg_agent(filp->private_data, compat_ptr(arg));
762 case IB_USER_MAD_ENABLE_PKEY:
763 return ib_umad_enable_pkey(filp->private_data);
764 default:
765 return -ENOIOCTLCMD;
766 }
767 }
768 #endif
769
770 /*
771 * ib_umad_open() does not need the BKL:
772 *
773 * - the ib_umad_port structures are properly reference counted, and
774 * everything else is purely local to the file being created, so
775 * races against other open calls are not a problem;
776 * - the ioctl method does not affect any global state outside of the
777 * file structure being operated on;
778 */
779 static int ib_umad_open(struct inode *inode, struct file *filp)
780 {
781 struct ib_umad_port *port;
782 struct ib_umad_file *file;
783 int ret = 0;
784
785 port = container_of(inode->i_cdev, struct ib_umad_port, cdev);
786 if (port)
787 kref_get(&port->umad_dev->ref);
788 else
789 return -ENXIO;
790
791 mutex_lock(&port->file_mutex);
792
793 if (!port->ib_dev) {
794 ret = -ENXIO;
795 goto out;
796 }
797
798 file = kzalloc(sizeof *file, GFP_KERNEL);
799 if (!file) {
800 kref_put(&port->umad_dev->ref, ib_umad_release_dev);
801 ret = -ENOMEM;
802 goto out;
803 }
804
805 mutex_init(&file->mutex);
806 spin_lock_init(&file->send_lock);
807 INIT_LIST_HEAD(&file->recv_list);
808 INIT_LIST_HEAD(&file->send_list);
809 init_waitqueue_head(&file->recv_wait);
810
811 file->port = port;
812 filp->private_data = file;
813
814 list_add_tail(&file->port_list, &port->file_list);
815
816 out:
817 mutex_unlock(&port->file_mutex);
818 return ret;
819 }
820
821 static int ib_umad_close(struct inode *inode, struct file *filp)
822 {
823 struct ib_umad_file *file = filp->private_data;
824 struct ib_umad_device *dev = file->port->umad_dev;
825 struct ib_umad_packet *packet, *tmp;
826 int already_dead;
827 int i;
828
829 mutex_lock(&file->port->file_mutex);
830 mutex_lock(&file->mutex);
831
832 already_dead = file->agents_dead;
833 file->agents_dead = 1;
834
835 list_for_each_entry_safe(packet, tmp, &file->recv_list, list) {
836 if (packet->recv_wc)
837 ib_free_recv_mad(packet->recv_wc);
838 kfree(packet);
839 }
840
841 list_del(&file->port_list);
842
843 mutex_unlock(&file->mutex);
844
845 if (!already_dead)
846 for (i = 0; i < IB_UMAD_MAX_AGENTS; ++i)
847 if (file->agent[i])
848 ib_unregister_mad_agent(file->agent[i]);
849
850 mutex_unlock(&file->port->file_mutex);
851
852 kfree(file);
853 kref_put(&dev->ref, ib_umad_release_dev);
854
855 return 0;
856 }
857
858 static const struct file_operations umad_fops = {
859 .owner = THIS_MODULE,
860 .read = ib_umad_read,
861 .write = ib_umad_write,
862 .poll = ib_umad_poll,
863 .unlocked_ioctl = ib_umad_ioctl,
864 #ifdef CONFIG_COMPAT
865 .compat_ioctl = ib_umad_compat_ioctl,
866 #endif
867 .open = ib_umad_open,
868 .release = ib_umad_close
869 };
870
871 static int ib_umad_sm_open(struct inode *inode, struct file *filp)
872 {
873 struct ib_umad_port *port;
874 struct ib_port_modify props = {
875 .set_port_cap_mask = IB_PORT_SM
876 };
877 int ret;
878
879 port = container_of(inode->i_cdev, struct ib_umad_port, sm_cdev);
880 if (port)
881 kref_get(&port->umad_dev->ref);
882 else
883 return -ENXIO;
884
885 if (filp->f_flags & O_NONBLOCK) {
886 if (down_trylock(&port->sm_sem)) {
887 ret = -EAGAIN;
888 goto fail;
889 }
890 } else {
891 if (down_interruptible(&port->sm_sem)) {
892 ret = -ERESTARTSYS;
893 goto fail;
894 }
895 }
896
897 ret = ib_modify_port(port->ib_dev, port->port_num, 0, &props);
898 if (ret) {
899 up(&port->sm_sem);
900 goto fail;
901 }
902
903 filp->private_data = port;
904
905 return 0;
906
907 fail:
908 kref_put(&port->umad_dev->ref, ib_umad_release_dev);
909 return ret;
910 }
911
912 static int ib_umad_sm_close(struct inode *inode, struct file *filp)
913 {
914 struct ib_umad_port *port = filp->private_data;
915 struct ib_port_modify props = {
916 .clr_port_cap_mask = IB_PORT_SM
917 };
918 int ret = 0;
919
920 mutex_lock(&port->file_mutex);
921 if (port->ib_dev)
922 ret = ib_modify_port(port->ib_dev, port->port_num, 0, &props);
923 mutex_unlock(&port->file_mutex);
924
925 up(&port->sm_sem);
926
927 kref_put(&port->umad_dev->ref, ib_umad_release_dev);
928
929 return ret;
930 }
931
932 static const struct file_operations umad_sm_fops = {
933 .owner = THIS_MODULE,
934 .open = ib_umad_sm_open,
935 .release = ib_umad_sm_close
936 };
937
938 static struct ib_client umad_client = {
939 .name = "umad",
940 .add = ib_umad_add_one,
941 .remove = ib_umad_remove_one
942 };
943
944 static ssize_t show_ibdev(struct device *dev, struct device_attribute *attr,
945 char *buf)
946 {
947 struct ib_umad_port *port = dev_get_drvdata(dev);
948
949 if (!port)
950 return -ENODEV;
951
952 return sprintf(buf, "%s\n", port->ib_dev->name);
953 }
954 static DEVICE_ATTR(ibdev, S_IRUGO, show_ibdev, NULL);
955
956 static ssize_t show_port(struct device *dev, struct device_attribute *attr,
957 char *buf)
958 {
959 struct ib_umad_port *port = dev_get_drvdata(dev);
960
961 if (!port)
962 return -ENODEV;
963
964 return sprintf(buf, "%d\n", port->port_num);
965 }
966 static DEVICE_ATTR(port, S_IRUGO, show_port, NULL);
967
968 static CLASS_ATTR_STRING(abi_version, S_IRUGO,
969 __stringify(IB_USER_MAD_ABI_VERSION));
970
971 static dev_t overflow_maj;
972 static DECLARE_BITMAP(overflow_map, IB_UMAD_MAX_PORTS);
973 static int find_overflow_devnum(void)
974 {
975 int ret;
976
977 if (!overflow_maj) {
978 ret = alloc_chrdev_region(&overflow_maj, 0, IB_UMAD_MAX_PORTS * 2,
979 "infiniband_mad");
980 if (ret) {
981 printk(KERN_ERR "user_mad: couldn't register dynamic device number\n");
982 return ret;
983 }
984 }
985
986 ret = find_first_zero_bit(overflow_map, IB_UMAD_MAX_PORTS);
987 if (ret >= IB_UMAD_MAX_PORTS)
988 return -1;
989
990 return ret;
991 }
992
993 static int ib_umad_init_port(struct ib_device *device, int port_num,
994 struct ib_umad_port *port)
995 {
996 int devnum;
997 dev_t base;
998
999 spin_lock(&port_lock);
1000 devnum = find_first_zero_bit(dev_map, IB_UMAD_MAX_PORTS);
1001 if (devnum >= IB_UMAD_MAX_PORTS) {
1002 spin_unlock(&port_lock);
1003 devnum = find_overflow_devnum();
1004 if (devnum < 0)
1005 return -1;
1006
1007 spin_lock(&port_lock);
1008 port->dev_num = devnum + IB_UMAD_MAX_PORTS;
1009 base = devnum + overflow_maj;
1010 set_bit(devnum, overflow_map);
1011 } else {
1012 port->dev_num = devnum;
1013 base = devnum + base_dev;
1014 set_bit(devnum, dev_map);
1015 }
1016 spin_unlock(&port_lock);
1017
1018 port->ib_dev = device;
1019 port->port_num = port_num;
1020 init_MUTEX(&port->sm_sem);
1021 mutex_init(&port->file_mutex);
1022 INIT_LIST_HEAD(&port->file_list);
1023
1024 cdev_init(&port->cdev, &umad_fops);
1025 port->cdev.owner = THIS_MODULE;
1026 kobject_set_name(&port->cdev.kobj, "umad%d", port->dev_num);
1027 if (cdev_add(&port->cdev, base, 1))
1028 goto err_cdev;
1029
1030 port->dev = device_create(umad_class, device->dma_device,
1031 port->cdev.dev, port,
1032 "umad%d", port->dev_num);
1033 if (IS_ERR(port->dev))
1034 goto err_cdev;
1035
1036 if (device_create_file(port->dev, &dev_attr_ibdev))
1037 goto err_dev;
1038 if (device_create_file(port->dev, &dev_attr_port))
1039 goto err_dev;
1040
1041 base += IB_UMAD_MAX_PORTS;
1042 cdev_init(&port->sm_cdev, &umad_sm_fops);
1043 port->sm_cdev.owner = THIS_MODULE;
1044 kobject_set_name(&port->sm_cdev.kobj, "issm%d", port->dev_num);
1045 if (cdev_add(&port->sm_cdev, base, 1))
1046 goto err_sm_cdev;
1047
1048 port->sm_dev = device_create(umad_class, device->dma_device,
1049 port->sm_cdev.dev, port,
1050 "issm%d", port->dev_num);
1051 if (IS_ERR(port->sm_dev))
1052 goto err_sm_cdev;
1053
1054 if (device_create_file(port->sm_dev, &dev_attr_ibdev))
1055 goto err_sm_dev;
1056 if (device_create_file(port->sm_dev, &dev_attr_port))
1057 goto err_sm_dev;
1058
1059 return 0;
1060
1061 err_sm_dev:
1062 device_destroy(umad_class, port->sm_cdev.dev);
1063
1064 err_sm_cdev:
1065 cdev_del(&port->sm_cdev);
1066
1067 err_dev:
1068 device_destroy(umad_class, port->cdev.dev);
1069
1070 err_cdev:
1071 cdev_del(&port->cdev);
1072 if (port->dev_num < IB_UMAD_MAX_PORTS)
1073 clear_bit(devnum, dev_map);
1074 else
1075 clear_bit(devnum, overflow_map);
1076
1077 return -1;
1078 }
1079
1080 static void ib_umad_kill_port(struct ib_umad_port *port)
1081 {
1082 struct ib_umad_file *file;
1083 int already_dead;
1084 int id;
1085
1086 dev_set_drvdata(port->dev, NULL);
1087 dev_set_drvdata(port->sm_dev, NULL);
1088
1089 device_destroy(umad_class, port->cdev.dev);
1090 device_destroy(umad_class, port->sm_cdev.dev);
1091
1092 cdev_del(&port->cdev);
1093 cdev_del(&port->sm_cdev);
1094
1095 mutex_lock(&port->file_mutex);
1096
1097 port->ib_dev = NULL;
1098
1099 list_for_each_entry(file, &port->file_list, port_list) {
1100 mutex_lock(&file->mutex);
1101 already_dead = file->agents_dead;
1102 file->agents_dead = 1;
1103 mutex_unlock(&file->mutex);
1104
1105 for (id = 0; id < IB_UMAD_MAX_AGENTS; ++id)
1106 if (file->agent[id])
1107 ib_unregister_mad_agent(file->agent[id]);
1108 }
1109
1110 mutex_unlock(&port->file_mutex);
1111
1112 if (port->dev_num < IB_UMAD_MAX_PORTS)
1113 clear_bit(port->dev_num, dev_map);
1114 else
1115 clear_bit(port->dev_num - IB_UMAD_MAX_PORTS, overflow_map);
1116 }
1117
1118 static void ib_umad_add_one(struct ib_device *device)
1119 {
1120 struct ib_umad_device *umad_dev;
1121 int s, e, i;
1122
1123 if (rdma_node_get_transport(device->node_type) != RDMA_TRANSPORT_IB)
1124 return;
1125
1126 if (device->node_type == RDMA_NODE_IB_SWITCH)
1127 s = e = 0;
1128 else {
1129 s = 1;
1130 e = device->phys_port_cnt;
1131 }
1132
1133 umad_dev = kzalloc(sizeof *umad_dev +
1134 (e - s + 1) * sizeof (struct ib_umad_port),
1135 GFP_KERNEL);
1136 if (!umad_dev)
1137 return;
1138
1139 kref_init(&umad_dev->ref);
1140
1141 umad_dev->start_port = s;
1142 umad_dev->end_port = e;
1143
1144 for (i = s; i <= e; ++i) {
1145 umad_dev->port[i - s].umad_dev = umad_dev;
1146
1147 if (ib_umad_init_port(device, i, &umad_dev->port[i - s]))
1148 goto err;
1149 }
1150
1151 ib_set_client_data(device, &umad_client, umad_dev);
1152
1153 return;
1154
1155 err:
1156 while (--i >= s)
1157 ib_umad_kill_port(&umad_dev->port[i - s]);
1158
1159 kref_put(&umad_dev->ref, ib_umad_release_dev);
1160 }
1161
1162 static void ib_umad_remove_one(struct ib_device *device)
1163 {
1164 struct ib_umad_device *umad_dev = ib_get_client_data(device, &umad_client);
1165 int i;
1166
1167 if (!umad_dev)
1168 return;
1169
1170 for (i = 0; i <= umad_dev->end_port - umad_dev->start_port; ++i)
1171 ib_umad_kill_port(&umad_dev->port[i]);
1172
1173 kref_put(&umad_dev->ref, ib_umad_release_dev);
1174 }
1175
1176 static int __init ib_umad_init(void)
1177 {
1178 int ret;
1179
1180 ret = register_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2,
1181 "infiniband_mad");
1182 if (ret) {
1183 printk(KERN_ERR "user_mad: couldn't register device number\n");
1184 goto out;
1185 }
1186
1187 umad_class = class_create(THIS_MODULE, "infiniband_mad");
1188 if (IS_ERR(umad_class)) {
1189 ret = PTR_ERR(umad_class);
1190 printk(KERN_ERR "user_mad: couldn't create class infiniband_mad\n");
1191 goto out_chrdev;
1192 }
1193
1194 ret = class_create_file(umad_class, &class_attr_abi_version.attr);
1195 if (ret) {
1196 printk(KERN_ERR "user_mad: couldn't create abi_version attribute\n");
1197 goto out_class;
1198 }
1199
1200 ret = ib_register_client(&umad_client);
1201 if (ret) {
1202 printk(KERN_ERR "user_mad: couldn't register ib_umad client\n");
1203 goto out_class;
1204 }
1205
1206 return 0;
1207
1208 out_class:
1209 class_destroy(umad_class);
1210
1211 out_chrdev:
1212 unregister_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2);
1213
1214 out:
1215 return ret;
1216 }
1217
1218 static void __exit ib_umad_cleanup(void)
1219 {
1220 ib_unregister_client(&umad_client);
1221 class_destroy(umad_class);
1222 unregister_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2);
1223 if (overflow_maj)
1224 unregister_chrdev_region(overflow_maj, IB_UMAD_MAX_PORTS * 2);
1225 }
1226
1227 module_init(ib_umad_init);
1228 module_exit(ib_umad_cleanup);