projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge branch 'upstream-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mfashe...
[deliverable/linux.git] / drivers / infiniband / core / user_mad.c
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/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
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: 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
89 struct 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
106 struct ib_umad_device {
107 int start_port, end_port;
108 struct kref ref;
109 struct ib_umad_port port[0];
110 };
111
112 struct ib_umad_file {
113 struct ib_umad_port *port;
114 struct list_head recv_list;
115 struct list_head send_list;
116 struct list_head port_list;
117 spinlock_t recv_lock;
118 spinlock_t send_lock;
119 wait_queue_head_t recv_wait;
120 struct ib_mad_agent *agent[IB_UMAD_MAX_AGENTS];
121 int agents_dead;
122 };
123
124 struct ib_umad_packet {
125 struct ib_mad_send_buf *msg;
126 struct ib_mad_recv_wc *recv_wc;
127 struct list_head list;
128 int length;
129 struct ib_user_mad mad;
130 };
131
132 static struct class *umad_class;
133
134 static const dev_t base_dev = MKDEV(IB_UMAD_MAJOR, IB_UMAD_MINOR_BASE);
135
136 static DEFINE_SPINLOCK(port_lock);
137 static struct ib_umad_port *umad_port[IB_UMAD_MAX_PORTS];
138 static DECLARE_BITMAP(dev_map, IB_UMAD_MAX_PORTS);
139
140 static void ib_umad_add_one(struct ib_device *device);
141 static void ib_umad_remove_one(struct ib_device *device);
142
143 static void ib_umad_release_dev(struct kref *ref)
144 {
145 struct ib_umad_device *dev =
146 container_of(ref, struct ib_umad_device, ref);
147
148 kfree(dev);
149 }
150
151 /* caller must hold port->mutex at least for reading */
152 static struct ib_mad_agent *__get_agent(struct ib_umad_file *file, int id)
153 {
154 return file->agents_dead ? NULL : file->agent[id];
155 }
156
157 static int queue_packet(struct ib_umad_file *file,
158 struct ib_mad_agent *agent,
159 struct ib_umad_packet *packet)
160 {
161 int ret = 1;
162
163 down_read(&file->port->mutex);
164
165 for (packet->mad.hdr.id = 0;
166 packet->mad.hdr.id < IB_UMAD_MAX_AGENTS;
167 packet->mad.hdr.id++)
168 if (agent == __get_agent(file, packet->mad.hdr.id)) {
169 spin_lock_irq(&file->recv_lock);
170 list_add_tail(&packet->list, &file->recv_list);
171 spin_unlock_irq(&file->recv_lock);
172 wake_up_interruptible(&file->recv_wait);
173 ret = 0;
174 break;
175 }
176
177 up_read(&file->port->mutex);
178
179 return ret;
180 }
181
182 static void dequeue_send(struct ib_umad_file *file,
183 struct ib_umad_packet *packet)
184 {
185 spin_lock_irq(&file->send_lock);
186 list_del(&packet->list);
187 spin_unlock_irq(&file->send_lock);
188 }
189
190 static void send_handler(struct ib_mad_agent *agent,
191 struct ib_mad_send_wc *send_wc)
192 {
193 struct ib_umad_file *file = agent->context;
194 struct ib_umad_packet *packet = send_wc->send_buf->context[0];
195
196 dequeue_send(file, packet);
197 ib_destroy_ah(packet->msg->ah);
198 ib_free_send_mad(packet->msg);
199
200 if (send_wc->status == IB_WC_RESP_TIMEOUT_ERR) {
201 packet->length = IB_MGMT_MAD_HDR;
202 packet->mad.hdr.status = ETIMEDOUT;
203 if (!queue_packet(file, agent, packet))
204 return;
205 }
206 kfree(packet);
207 }
208
209 static void recv_handler(struct ib_mad_agent *agent,
210 struct ib_mad_recv_wc *mad_recv_wc)
211 {
212 struct ib_umad_file *file = agent->context;
213 struct ib_umad_packet *packet;
214
215 if (mad_recv_wc->wc->status != IB_WC_SUCCESS)
216 goto err1;
217
218 packet = kzalloc(sizeof *packet, GFP_KERNEL);
219 if (!packet)
220 goto err1;
221
222 packet->length = mad_recv_wc->mad_len;
223 packet->recv_wc = mad_recv_wc;
224
225 packet->mad.hdr.status = 0;
226 packet->mad.hdr.length = sizeof (struct ib_user_mad) +
227 mad_recv_wc->mad_len;
228 packet->mad.hdr.qpn = cpu_to_be32(mad_recv_wc->wc->src_qp);
229 packet->mad.hdr.lid = cpu_to_be16(mad_recv_wc->wc->slid);
230 packet->mad.hdr.sl = mad_recv_wc->wc->sl;
231 packet->mad.hdr.path_bits = mad_recv_wc->wc->dlid_path_bits;
232 packet->mad.hdr.grh_present = !!(mad_recv_wc->wc->wc_flags & IB_WC_GRH);
233 if (packet->mad.hdr.grh_present) {
234 struct ib_ah_attr ah_attr;
235
236 ib_init_ah_from_wc(agent->device, agent->port_num,
237 mad_recv_wc->wc, mad_recv_wc->recv_buf.grh,
238 &ah_attr);
239
240 packet->mad.hdr.gid_index = ah_attr.grh.sgid_index;
241 packet->mad.hdr.hop_limit = ah_attr.grh.hop_limit;
242 packet->mad.hdr.traffic_class = ah_attr.grh.traffic_class;
243 memcpy(packet->mad.hdr.gid, &ah_attr.grh.dgid, 16);
244 packet->mad.hdr.flow_label = cpu_to_be32(ah_attr.grh.flow_label);
245 }
246
247 if (queue_packet(file, agent, packet))
248 goto err2;
249 return;
250
251 err2:
252 kfree(packet);
253 err1:
254 ib_free_recv_mad(mad_recv_wc);
255 }
256
257 static ssize_t copy_recv_mad(char __user *buf, struct ib_umad_packet *packet,
258 size_t count)
259 {
260 struct ib_mad_recv_buf *recv_buf;
261 int left, seg_payload, offset, max_seg_payload;
262
263 /* We need enough room to copy the first (or only) MAD segment. */
264 recv_buf = &packet->recv_wc->recv_buf;
265 if ((packet->length <= sizeof (*recv_buf->mad) &&
266 count < sizeof (packet->mad) + packet->length) ||
267 (packet->length > sizeof (*recv_buf->mad) &&
268 count < sizeof (packet->mad) + sizeof (*recv_buf->mad)))
269 return -EINVAL;
270
271 if (copy_to_user(buf, &packet->mad, sizeof (packet->mad)))
272 return -EFAULT;
273
274 buf += sizeof (packet->mad);
275 seg_payload = min_t(int, packet->length, sizeof (*recv_buf->mad));
276 if (copy_to_user(buf, recv_buf->mad, seg_payload))
277 return -EFAULT;
278
279 if (seg_payload < packet->length) {
280 /*
281 * Multipacket RMPP MAD message. Copy remainder of message.
282 * Note that last segment may have a shorter payload.
283 */
284 if (count < sizeof (packet->mad) + packet->length) {
285 /*
286 * The buffer is too small, return the first RMPP segment,
287 * which includes the RMPP message length.
288 */
289 return -ENOSPC;
290 }
291 offset = ib_get_mad_data_offset(recv_buf->mad->mad_hdr.mgmt_class);
292 max_seg_payload = sizeof (struct ib_mad) - offset;
293
294 for (left = packet->length - seg_payload, buf += seg_payload;
295 left; left -= seg_payload, buf += seg_payload) {
296 recv_buf = container_of(recv_buf->list.next,
297 struct ib_mad_recv_buf, list);
298 seg_payload = min(left, max_seg_payload);
299 if (copy_to_user(buf, ((void *) recv_buf->mad) + offset,
300 seg_payload))
301 return -EFAULT;
302 }
303 }
304 return sizeof (packet->mad) + packet->length;
305 }
306
307 static ssize_t copy_send_mad(char __user *buf, struct ib_umad_packet *packet,
308 size_t count)
309 {
310 ssize_t size = sizeof (packet->mad) + packet->length;
311
312 if (count < size)
313 return -EINVAL;
314
315 if (copy_to_user(buf, &packet->mad, size))
316 return -EFAULT;
317
318 return size;
319 }
320
321 static ssize_t ib_umad_read(struct file *filp, char __user *buf,
322 size_t count, loff_t *pos)
323 {
324 struct ib_umad_file *file = filp->private_data;
325 struct ib_umad_packet *packet;
326 ssize_t ret;
327
328 if (count < sizeof (struct ib_user_mad))
329 return -EINVAL;
330
331 spin_lock_irq(&file->recv_lock);
332
333 while (list_empty(&file->recv_list)) {
334 spin_unlock_irq(&file->recv_lock);
335
336 if (filp->f_flags & O_NONBLOCK)
337 return -EAGAIN;
338
339 if (wait_event_interruptible(file->recv_wait,
340 !list_empty(&file->recv_list)))
341 return -ERESTARTSYS;
342
343 spin_lock_irq(&file->recv_lock);
344 }
345
346 packet = list_entry(file->recv_list.next, struct ib_umad_packet, list);
347 list_del(&packet->list);
348
349 spin_unlock_irq(&file->recv_lock);
350
351 if (packet->recv_wc)
352 ret = copy_recv_mad(buf, packet, count);
353 else
354 ret = copy_send_mad(buf, packet, count);
355
356 if (ret < 0) {
357 /* Requeue packet */
358 spin_lock_irq(&file->recv_lock);
359 list_add(&packet->list, &file->recv_list);
360 spin_unlock_irq(&file->recv_lock);
361 } else {
362 if (packet->recv_wc)
363 ib_free_recv_mad(packet->recv_wc);
364 kfree(packet);
365 }
366 return ret;
367 }
368
369 static int copy_rmpp_mad(struct ib_mad_send_buf *msg, const char __user *buf)
370 {
371 int left, seg;
372
373 /* Copy class specific header */
374 if ((msg->hdr_len > IB_MGMT_RMPP_HDR) &&
375 copy_from_user(msg->mad + IB_MGMT_RMPP_HDR, buf + IB_MGMT_RMPP_HDR,
376 msg->hdr_len - IB_MGMT_RMPP_HDR))
377 return -EFAULT;
378
379 /* All headers are in place. Copy data segments. */
380 for (seg = 1, left = msg->data_len, buf += msg->hdr_len; left > 0;
381 seg++, left -= msg->seg_size, buf += msg->seg_size) {
382 if (copy_from_user(ib_get_rmpp_segment(msg, seg), buf,
383 min(left, msg->seg_size)))
384 return -EFAULT;
385 }
386 return 0;
387 }
388
389 static int same_destination(struct ib_user_mad_hdr *hdr1,
390 struct ib_user_mad_hdr *hdr2)
391 {
392 if (!hdr1->grh_present && !hdr2->grh_present)
393 return (hdr1->lid == hdr2->lid);
394
395 if (hdr1->grh_present && hdr2->grh_present)
396 return !memcmp(hdr1->gid, hdr2->gid, 16);
397
398 return 0;
399 }
400
401 static int is_duplicate(struct ib_umad_file *file,
402 struct ib_umad_packet *packet)
403 {
404 struct ib_umad_packet *sent_packet;
405 struct ib_mad_hdr *sent_hdr, *hdr;
406
407 hdr = (struct ib_mad_hdr *) packet->mad.data;
408 list_for_each_entry(sent_packet, &file->send_list, list) {
409 sent_hdr = (struct ib_mad_hdr *) sent_packet->mad.data;
410
411 if ((hdr->tid != sent_hdr->tid) ||
412 (hdr->mgmt_class != sent_hdr->mgmt_class))
413 continue;
414
415 /*
416 * No need to be overly clever here. If two new operations have
417 * the same TID, reject the second as a duplicate. This is more
418 * restrictive than required by the spec.
419 */
420 if (!ib_response_mad((struct ib_mad *) hdr)) {
421 if (!ib_response_mad((struct ib_mad *) sent_hdr))
422 return 1;
423 continue;
424 } else if (!ib_response_mad((struct ib_mad *) sent_hdr))
425 continue;
426
427 if (same_destination(&packet->mad.hdr, &sent_packet->mad.hdr))
428 return 1;
429 }
430
431 return 0;
432 }
433
434 static ssize_t ib_umad_write(struct file *filp, const char __user *buf,
435 size_t count, loff_t *pos)
436 {
437 struct ib_umad_file *file = filp->private_data;
438 struct ib_umad_packet *packet;
439 struct ib_mad_agent *agent;
440 struct ib_ah_attr ah_attr;
441 struct ib_ah *ah;
442 struct ib_rmpp_mad *rmpp_mad;
443 __be64 *tid;
444 int ret, data_len, hdr_len, copy_offset, rmpp_active;
445
446 if (count < sizeof (struct ib_user_mad) + IB_MGMT_RMPP_HDR)
447 return -EINVAL;
448
449 packet = kzalloc(sizeof *packet + IB_MGMT_RMPP_HDR, GFP_KERNEL);
450 if (!packet)
451 return -ENOMEM;
452
453 if (copy_from_user(&packet->mad, buf,
454 sizeof (struct ib_user_mad) + IB_MGMT_RMPP_HDR)) {
455 ret = -EFAULT;
456 goto err;
457 }
458
459 if (packet->mad.hdr.id < 0 ||
460 packet->mad.hdr.id >= IB_UMAD_MAX_AGENTS) {
461 ret = -EINVAL;
462 goto err;
463 }
464
465 down_read(&file->port->mutex);
466
467 agent = __get_agent(file, packet->mad.hdr.id);
468 if (!agent) {
469 ret = -EINVAL;
470 goto err_up;
471 }
472
473 memset(&ah_attr, 0, sizeof ah_attr);
474 ah_attr.dlid = be16_to_cpu(packet->mad.hdr.lid);
475 ah_attr.sl = packet->mad.hdr.sl;
476 ah_attr.src_path_bits = packet->mad.hdr.path_bits;
477 ah_attr.port_num = file->port->port_num;
478 if (packet->mad.hdr.grh_present) {
479 ah_attr.ah_flags = IB_AH_GRH;
480 memcpy(ah_attr.grh.dgid.raw, packet->mad.hdr.gid, 16);
481 ah_attr.grh.sgid_index = packet->mad.hdr.gid_index;
482 ah_attr.grh.flow_label = be32_to_cpu(packet->mad.hdr.flow_label);
483 ah_attr.grh.hop_limit = packet->mad.hdr.hop_limit;
484 ah_attr.grh.traffic_class = packet->mad.hdr.traffic_class;
485 }
486
487 ah = ib_create_ah(agent->qp->pd, &ah_attr);
488 if (IS_ERR(ah)) {
489 ret = PTR_ERR(ah);
490 goto err_up;
491 }
492
493 rmpp_mad = (struct ib_rmpp_mad *) packet->mad.data;
494 hdr_len = ib_get_mad_data_offset(rmpp_mad->mad_hdr.mgmt_class);
495 if (!ib_is_mad_class_rmpp(rmpp_mad->mad_hdr.mgmt_class)) {
496 copy_offset = IB_MGMT_MAD_HDR;
497 rmpp_active = 0;
498 } else {
499 copy_offset = IB_MGMT_RMPP_HDR;
500 rmpp_active = ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) &
501 IB_MGMT_RMPP_FLAG_ACTIVE;
502 }
503
504 data_len = count - sizeof (struct ib_user_mad) - hdr_len;
505 packet->msg = ib_create_send_mad(agent,
506 be32_to_cpu(packet->mad.hdr.qpn),
507 0, rmpp_active, hdr_len,
508 data_len, GFP_KERNEL);
509 if (IS_ERR(packet->msg)) {
510 ret = PTR_ERR(packet->msg);
511 goto err_ah;
512 }
513
514 packet->msg->ah = ah;
515 packet->msg->timeout_ms = packet->mad.hdr.timeout_ms;
516 packet->msg->retries = packet->mad.hdr.retries;
517 packet->msg->context[0] = packet;
518
519 /* Copy MAD header. Any RMPP header is already in place. */
520 memcpy(packet->msg->mad, packet->mad.data, IB_MGMT_MAD_HDR);
521 buf += sizeof (struct ib_user_mad);
522
523 if (!rmpp_active) {
524 if (copy_from_user(packet->msg->mad + copy_offset,
525 buf + copy_offset,
526 hdr_len + data_len - copy_offset)) {
527 ret = -EFAULT;
528 goto err_msg;
529 }
530 } else {
531 ret = copy_rmpp_mad(packet->msg, buf);
532 if (ret)
533 goto err_msg;
534 }
535
536 /*
537 * Set the high-order part of the transaction ID to make MADs from
538 * different agents unique, and allow routing responses back to the
539 * original requestor.
540 */
541 if (!ib_response_mad(packet->msg->mad)) {
542 tid = &((struct ib_mad_hdr *) packet->msg->mad)->tid;
543 *tid = cpu_to_be64(((u64) agent->hi_tid) << 32 |
544 (be64_to_cpup(tid) & 0xffffffff));
545 rmpp_mad->mad_hdr.tid = *tid;
546 }
547
548 spin_lock_irq(&file->send_lock);
549 ret = is_duplicate(file, packet);
550 if (!ret)
551 list_add_tail(&packet->list, &file->send_list);
552 spin_unlock_irq(&file->send_lock);
553 if (ret) {
554 ret = -EINVAL;
555 goto err_msg;
556 }
557
558 ret = ib_post_send_mad(packet->msg, NULL);
559 if (ret)
560 goto err_send;
561
562 up_read(&file->port->mutex);
563 return count;
564
565 err_send:
566 dequeue_send(file, packet);
567 err_msg:
568 ib_free_send_mad(packet->msg);
569 err_ah:
570 ib_destroy_ah(ah);
571 err_up:
572 up_read(&file->port->mutex);
573 err:
574 kfree(packet);
575 return ret;
576 }
577
578 static unsigned int ib_umad_poll(struct file *filp, struct poll_table_struct *wait)
579 {
580 struct ib_umad_file *file = filp->private_data;
581
582 /* we will always be able to post a MAD send */
583 unsigned int mask = POLLOUT | POLLWRNORM;
584
585 poll_wait(filp, &file->recv_wait, wait);
586
587 if (!list_empty(&file->recv_list))
588 mask |= POLLIN | POLLRDNORM;
589
590 return mask;
591 }
592
593 static int ib_umad_reg_agent(struct ib_umad_file *file, unsigned long arg)
594 {
595 struct ib_user_mad_reg_req ureq;
596 struct ib_mad_reg_req req;
597 struct ib_mad_agent *agent;
598 int agent_id;
599 int ret;
600
601 down_write(&file->port->mutex);
602
603 if (!file->port->ib_dev) {
604 ret = -EPIPE;
605 goto out;
606 }
607
608 if (copy_from_user(&ureq, (void __user *) arg, sizeof ureq)) {
609 ret = -EFAULT;
610 goto out;
611 }
612
613 if (ureq.qpn != 0 && ureq.qpn != 1) {
614 ret = -EINVAL;
615 goto out;
616 }
617
618 for (agent_id = 0; agent_id < IB_UMAD_MAX_AGENTS; ++agent_id)
619 if (!__get_agent(file, agent_id))
620 goto found;
621
622 ret = -ENOMEM;
623 goto out;
624
625 found:
626 if (ureq.mgmt_class) {
627 req.mgmt_class = ureq.mgmt_class;
628 req.mgmt_class_version = ureq.mgmt_class_version;
629 memcpy(req.method_mask, ureq.method_mask, sizeof req.method_mask);
630 memcpy(req.oui, ureq.oui, sizeof req.oui);
631 }
632
633 agent = ib_register_mad_agent(file->port->ib_dev, file->port->port_num,
634 ureq.qpn ? IB_QPT_GSI : IB_QPT_SMI,
635 ureq.mgmt_class ? &req : NULL,
636 ureq.rmpp_version,
637 send_handler, recv_handler, file);
638 if (IS_ERR(agent)) {
639 ret = PTR_ERR(agent);
640 goto out;
641 }
642
643 if (put_user(agent_id,
644 (u32 __user *) (arg + offsetof(struct ib_user_mad_reg_req, id)))) {
645 ret = -EFAULT;
646 ib_unregister_mad_agent(agent);
647 goto out;
648 }
649
650 file->agent[agent_id] = agent;
651 ret = 0;
652
653 out:
654 up_write(&file->port->mutex);
655 return ret;
656 }
657
658 static int ib_umad_unreg_agent(struct ib_umad_file *file, unsigned long arg)
659 {
660 struct ib_mad_agent *agent = NULL;
661 u32 id;
662 int ret = 0;
663
664 if (get_user(id, (u32 __user *) arg))
665 return -EFAULT;
666
667 down_write(&file->port->mutex);
668
669 if (id < 0 || id >= IB_UMAD_MAX_AGENTS || !__get_agent(file, id)) {
670 ret = -EINVAL;
671 goto out;
672 }
673
674 agent = file->agent[id];
675 file->agent[id] = NULL;
676
677 out:
678 up_write(&file->port->mutex);
679
680 if (agent)
681 ib_unregister_mad_agent(agent);
682
683 return ret;
684 }
685
686 static long ib_umad_ioctl(struct file *filp, unsigned int cmd,
687 unsigned long arg)
688 {
689 switch (cmd) {
690 case IB_USER_MAD_REGISTER_AGENT:
691 return ib_umad_reg_agent(filp->private_data, arg);
692 case IB_USER_MAD_UNREGISTER_AGENT:
693 return ib_umad_unreg_agent(filp->private_data, arg);
694 default:
695 return -ENOIOCTLCMD;
696 }
697 }
698
699 static int ib_umad_open(struct inode *inode, struct file *filp)
700 {
701 struct ib_umad_port *port;
702 struct ib_umad_file *file;
703 int ret = 0;
704
705 spin_lock(&port_lock);
706 port = umad_port[iminor(inode) - IB_UMAD_MINOR_BASE];
707 if (port)
708 kref_get(&port->umad_dev->ref);
709 spin_unlock(&port_lock);
710
711 if (!port)
712 return -ENXIO;
713
714 down_write(&port->mutex);
715
716 if (!port->ib_dev) {
717 ret = -ENXIO;
718 goto out;
719 }
720
721 file = kzalloc(sizeof *file, GFP_KERNEL);
722 if (!file) {
723 kref_put(&port->umad_dev->ref, ib_umad_release_dev);
724 ret = -ENOMEM;
725 goto out;
726 }
727
728 spin_lock_init(&file->recv_lock);
729 spin_lock_init(&file->send_lock);
730 INIT_LIST_HEAD(&file->recv_list);
731 INIT_LIST_HEAD(&file->send_list);
732 init_waitqueue_head(&file->recv_wait);
733
734 file->port = port;
735 filp->private_data = file;
736
737 list_add_tail(&file->port_list, &port->file_list);
738
739 out:
740 up_write(&port->mutex);
741 return ret;
742 }
743
744 static int ib_umad_close(struct inode *inode, struct file *filp)
745 {
746 struct ib_umad_file *file = filp->private_data;
747 struct ib_umad_device *dev = file->port->umad_dev;
748 struct ib_umad_packet *packet, *tmp;
749 int already_dead;
750 int i;
751
752 down_write(&file->port->mutex);
753
754 already_dead = file->agents_dead;
755 file->agents_dead = 1;
756
757 list_for_each_entry_safe(packet, tmp, &file->recv_list, list) {
758 if (packet->recv_wc)
759 ib_free_recv_mad(packet->recv_wc);
760 kfree(packet);
761 }
762
763 list_del(&file->port_list);
764
765 downgrade_write(&file->port->mutex);
766
767 if (!already_dead)
768 for (i = 0; i < IB_UMAD_MAX_AGENTS; ++i)
769 if (file->agent[i])
770 ib_unregister_mad_agent(file->agent[i]);
771
772 up_read(&file->port->mutex);
773
774 kfree(file);
775 kref_put(&dev->ref, ib_umad_release_dev);
776
777 return 0;
778 }
779
780 static const struct file_operations umad_fops = {
781 .owner = THIS_MODULE,
782 .read = ib_umad_read,
783 .write = ib_umad_write,
784 .poll = ib_umad_poll,
785 .unlocked_ioctl = ib_umad_ioctl,
786 .compat_ioctl = ib_umad_ioctl,
787 .open = ib_umad_open,
788 .release = ib_umad_close
789 };
790
791 static int ib_umad_sm_open(struct inode *inode, struct file *filp)
792 {
793 struct ib_umad_port *port;
794 struct ib_port_modify props = {
795 .set_port_cap_mask = IB_PORT_SM
796 };
797 int ret;
798
799 spin_lock(&port_lock);
800 port = umad_port[iminor(inode) - IB_UMAD_MINOR_BASE - IB_UMAD_MAX_PORTS];
801 if (port)
802 kref_get(&port->umad_dev->ref);
803 spin_unlock(&port_lock);
804
805 if (!port)
806 return -ENXIO;
807
808 if (filp->f_flags & O_NONBLOCK) {
809 if (down_trylock(&port->sm_sem)) {
810 ret = -EAGAIN;
811 goto fail;
812 }
813 } else {
814 if (down_interruptible(&port->sm_sem)) {
815 ret = -ERESTARTSYS;
816 goto fail;
817 }
818 }
819
820 ret = ib_modify_port(port->ib_dev, port->port_num, 0, &props);
821 if (ret) {
822 up(&port->sm_sem);
823 goto fail;
824 }
825
826 filp->private_data = port;
827
828 return 0;
829
830 fail:
831 kref_put(&port->umad_dev->ref, ib_umad_release_dev);
832 return ret;
833 }
834
835 static int ib_umad_sm_close(struct inode *inode, struct file *filp)
836 {
837 struct ib_umad_port *port = filp->private_data;
838 struct ib_port_modify props = {
839 .clr_port_cap_mask = IB_PORT_SM
840 };
841 int ret = 0;
842
843 down_write(&port->mutex);
844 if (port->ib_dev)
845 ret = ib_modify_port(port->ib_dev, port->port_num, 0, &props);
846 up_write(&port->mutex);
847
848 up(&port->sm_sem);
849
850 kref_put(&port->umad_dev->ref, ib_umad_release_dev);
851
852 return ret;
853 }
854
855 static const struct file_operations umad_sm_fops = {
856 .owner = THIS_MODULE,
857 .open = ib_umad_sm_open,
858 .release = ib_umad_sm_close
859 };
860
861 static struct ib_client umad_client = {
862 .name = "umad",
863 .add = ib_umad_add_one,
864 .remove = ib_umad_remove_one
865 };
866
867 static ssize_t show_ibdev(struct class_device *class_dev, char *buf)
868 {
869 struct ib_umad_port *port = class_get_devdata(class_dev);
870
871 if (!port)
872 return -ENODEV;
873
874 return sprintf(buf, "%s\n", port->ib_dev->name);
875 }
876 static CLASS_DEVICE_ATTR(ibdev, S_IRUGO, show_ibdev, NULL);
877
878 static ssize_t show_port(struct class_device *class_dev, char *buf)
879 {
880 struct ib_umad_port *port = class_get_devdata(class_dev);
881
882 if (!port)
883 return -ENODEV;
884
885 return sprintf(buf, "%d\n", port->port_num);
886 }
887 static CLASS_DEVICE_ATTR(port, S_IRUGO, show_port, NULL);
888
889 static ssize_t show_abi_version(struct class *class, char *buf)
890 {
891 return sprintf(buf, "%d\n", IB_USER_MAD_ABI_VERSION);
892 }
893 static CLASS_ATTR(abi_version, S_IRUGO, show_abi_version, NULL);
894
895 static int ib_umad_init_port(struct ib_device *device, int port_num,
896 struct ib_umad_port *port)
897 {
898 spin_lock(&port_lock);
899 port->dev_num = find_first_zero_bit(dev_map, IB_UMAD_MAX_PORTS);
900 if (port->dev_num >= IB_UMAD_MAX_PORTS) {
901 spin_unlock(&port_lock);
902 return -1;
903 }
904 set_bit(port->dev_num, dev_map);
905 spin_unlock(&port_lock);
906
907 port->ib_dev = device;
908 port->port_num = port_num;
909 init_MUTEX(&port->sm_sem);
910 init_rwsem(&port->mutex);
911 INIT_LIST_HEAD(&port->file_list);
912
913 port->dev = cdev_alloc();
914 if (!port->dev)
915 return -1;
916 port->dev->owner = THIS_MODULE;
917 port->dev->ops = &umad_fops;
918 kobject_set_name(&port->dev->kobj, "umad%d", port->dev_num);
919 if (cdev_add(port->dev, base_dev + port->dev_num, 1))
920 goto err_cdev;
921
922 port->class_dev = class_device_create(umad_class, NULL, port->dev->dev,
923 device->dma_device,
924 "umad%d", port->dev_num);
925 if (IS_ERR(port->class_dev))
926 goto err_cdev;
927
928 if (class_device_create_file(port->class_dev, &class_device_attr_ibdev))
929 goto err_class;
930 if (class_device_create_file(port->class_dev, &class_device_attr_port))
931 goto err_class;
932
933 port->sm_dev = cdev_alloc();
934 if (!port->sm_dev)
935 goto err_class;
936 port->sm_dev->owner = THIS_MODULE;
937 port->sm_dev->ops = &umad_sm_fops;
938 kobject_set_name(&port->sm_dev->kobj, "issm%d", port->dev_num);
939 if (cdev_add(port->sm_dev, base_dev + port->dev_num + IB_UMAD_MAX_PORTS, 1))
940 goto err_sm_cdev;
941
942 port->sm_class_dev = class_device_create(umad_class, NULL, port->sm_dev->dev,
943 device->dma_device,
944 "issm%d", port->dev_num);
945 if (IS_ERR(port->sm_class_dev))
946 goto err_sm_cdev;
947
948 class_set_devdata(port->class_dev, port);
949 class_set_devdata(port->sm_class_dev, port);
950
951 if (class_device_create_file(port->sm_class_dev, &class_device_attr_ibdev))
952 goto err_sm_class;
953 if (class_device_create_file(port->sm_class_dev, &class_device_attr_port))
954 goto err_sm_class;
955
956 spin_lock(&port_lock);
957 umad_port[port->dev_num] = port;
958 spin_unlock(&port_lock);
959
960 return 0;
961
962 err_sm_class:
963 class_device_destroy(umad_class, port->sm_dev->dev);
964
965 err_sm_cdev:
966 cdev_del(port->sm_dev);
967
968 err_class:
969 class_device_destroy(umad_class, port->dev->dev);
970
971 err_cdev:
972 cdev_del(port->dev);
973 clear_bit(port->dev_num, dev_map);
974
975 return -1;
976 }
977
978 static void ib_umad_kill_port(struct ib_umad_port *port)
979 {
980 struct ib_umad_file *file;
981 int id;
982
983 class_set_devdata(port->class_dev, NULL);
984 class_set_devdata(port->sm_class_dev, NULL);
985
986 class_device_destroy(umad_class, port->dev->dev);
987 class_device_destroy(umad_class, port->sm_dev->dev);
988
989 cdev_del(port->dev);
990 cdev_del(port->sm_dev);
991
992 spin_lock(&port_lock);
993 umad_port[port->dev_num] = NULL;
994 spin_unlock(&port_lock);
995
996 down_write(&port->mutex);
997
998 port->ib_dev = NULL;
999
1000 /*
1001 * Now go through the list of files attached to this port and
1002 * unregister all of their MAD agents. We need to hold
1003 * port->mutex while doing this to avoid racing with
1004 * ib_umad_close(), but we can't hold the mutex for writing
1005 * while calling ib_unregister_mad_agent(), since that might
1006 * deadlock by calling back into queue_packet(). So we
1007 * downgrade our lock to a read lock, and then drop and
1008 * reacquire the write lock for the next iteration.
1009 *
1010 * We do list_del_init() on the file's list_head so that the
1011 * list_del in ib_umad_close() is still OK, even after the
1012 * file is removed from the list.
1013 */
1014 while (!list_empty(&port->file_list)) {
1015 file = list_entry(port->file_list.next, struct ib_umad_file,
1016 port_list);
1017
1018 file->agents_dead = 1;
1019 list_del_init(&file->port_list);
1020
1021 downgrade_write(&port->mutex);
1022
1023 for (id = 0; id < IB_UMAD_MAX_AGENTS; ++id)
1024 if (file->agent[id])
1025 ib_unregister_mad_agent(file->agent[id]);
1026
1027 up_read(&port->mutex);
1028 down_write(&port->mutex);
1029 }
1030
1031 up_write(&port->mutex);
1032
1033 clear_bit(port->dev_num, dev_map);
1034 }
1035
1036 static void ib_umad_add_one(struct ib_device *device)
1037 {
1038 struct ib_umad_device *umad_dev;
1039 int s, e, i;
1040
1041 if (rdma_node_get_transport(device->node_type) != RDMA_TRANSPORT_IB)
1042 return;
1043
1044 if (device->node_type == RDMA_NODE_IB_SWITCH)
1045 s = e = 0;
1046 else {
1047 s = 1;
1048 e = device->phys_port_cnt;
1049 }
1050
1051 umad_dev = kzalloc(sizeof *umad_dev +
1052 (e - s + 1) * sizeof (struct ib_umad_port),
1053 GFP_KERNEL);
1054 if (!umad_dev)
1055 return;
1056
1057 kref_init(&umad_dev->ref);
1058
1059 umad_dev->start_port = s;
1060 umad_dev->end_port = e;
1061
1062 for (i = s; i <= e; ++i) {
1063 umad_dev->port[i - s].umad_dev = umad_dev;
1064
1065 if (ib_umad_init_port(device, i, &umad_dev->port[i - s]))
1066 goto err;
1067 }
1068
1069 ib_set_client_data(device, &umad_client, umad_dev);
1070
1071 return;
1072
1073 err:
1074 while (--i >= s)
1075 ib_umad_kill_port(&umad_dev->port[i - s]);
1076
1077 kref_put(&umad_dev->ref, ib_umad_release_dev);
1078 }
1079
1080 static void ib_umad_remove_one(struct ib_device *device)
1081 {
1082 struct ib_umad_device *umad_dev = ib_get_client_data(device, &umad_client);
1083 int i;
1084
1085 if (!umad_dev)
1086 return;
1087
1088 for (i = 0; i <= umad_dev->end_port - umad_dev->start_port; ++i)
1089 ib_umad_kill_port(&umad_dev->port[i]);
1090
1091 kref_put(&umad_dev->ref, ib_umad_release_dev);
1092 }
1093
1094 static int __init ib_umad_init(void)
1095 {
1096 int ret;
1097
1098 ret = register_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2,
1099 "infiniband_mad");
1100 if (ret) {
1101 printk(KERN_ERR "user_mad: couldn't register device number\n");
1102 goto out;
1103 }
1104
1105 umad_class = class_create(THIS_MODULE, "infiniband_mad");
1106 if (IS_ERR(umad_class)) {
1107 ret = PTR_ERR(umad_class);
1108 printk(KERN_ERR "user_mad: couldn't create class infiniband_mad\n");
1109 goto out_chrdev;
1110 }
1111
1112 ret = class_create_file(umad_class, &class_attr_abi_version);
1113 if (ret) {
1114 printk(KERN_ERR "user_mad: couldn't create abi_version attribute\n");
1115 goto out_class;
1116 }
1117
1118 ret = ib_register_client(&umad_client);
1119 if (ret) {
1120 printk(KERN_ERR "user_mad: couldn't register ib_umad client\n");
1121 goto out_class;
1122 }
1123
1124 return 0;
1125
1126 out_class:
1127 class_destroy(umad_class);
1128
1129 out_chrdev:
1130 unregister_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2);
1131
1132 out:
1133 return ret;
1134 }
1135
1136 static void __exit ib_umad_cleanup(void)
1137 {
1138 ib_unregister_client(&umad_client);
1139 class_destroy(umad_class);
1140 unregister_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2);
1141 }
1142
1143 module_init(ib_umad_init);
1144 module_exit(ib_umad_cleanup);
Linux kernel - Deliverables
This page took 0.057438 seconds and 6 git commands to generate.