1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36#define pr_fmt(fmt) "user_mad: " fmt
37
38#include <linux/module.h>
39#include <linux/init.h>
40#include <linux/device.h>
41#include <linux/err.h>
42#include <linux/fs.h>
43#include <linux/cdev.h>
44#include <linux/dma-mapping.h>
45#include <linux/poll.h>
46#include <linux/mutex.h>
47#include <linux/kref.h>
48#include <linux/compat.h>
49#include <linux/sched.h>
50#include <linux/semaphore.h>
51#include <linux/slab.h>
52
53#include <linux/uaccess.h>
54
55#include <rdma/ib_mad.h>
56#include <rdma/ib_user_mad.h>
57
58MODULE_AUTHOR("Roland Dreier");
59MODULE_DESCRIPTION("InfiniBand userspace MAD packet access");
60MODULE_LICENSE("Dual BSD/GPL");
61
62enum {
63 IB_UMAD_MAX_PORTS = 64,
64 IB_UMAD_MAX_AGENTS = 32,
65
66 IB_UMAD_MAJOR = 231,
67 IB_UMAD_MINOR_BASE = 0
68};
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84struct ib_umad_port {
85 struct cdev cdev;
86 struct device *dev;
87
88 struct cdev sm_cdev;
89 struct device *sm_dev;
90 struct semaphore sm_sem;
91
92 struct mutex file_mutex;
93 struct list_head file_list;
94
95 struct ib_device *ib_dev;
96 struct ib_umad_device *umad_dev;
97 int dev_num;
98 u8 port_num;
99};
100
101struct ib_umad_device {
102 struct kobject kobj;
103 struct ib_umad_port port[0];
104};
105
106struct ib_umad_file {
107 struct mutex mutex;
108 struct ib_umad_port *port;
109 struct list_head recv_list;
110 struct list_head send_list;
111 struct list_head port_list;
112 spinlock_t send_lock;
113 wait_queue_head_t recv_wait;
114 struct ib_mad_agent *agent[IB_UMAD_MAX_AGENTS];
115 int agents_dead;
116 u8 use_pkey_index;
117 u8 already_used;
118};
119
120struct ib_umad_packet {
121 struct ib_mad_send_buf *msg;
122 struct ib_mad_recv_wc *recv_wc;
123 struct list_head list;
124 int length;
125 struct ib_user_mad mad;
126};
127
128static struct class *umad_class;
129
130static const dev_t base_dev = MKDEV(IB_UMAD_MAJOR, IB_UMAD_MINOR_BASE);
131
132static DEFINE_SPINLOCK(port_lock);
133static DECLARE_BITMAP(dev_map, IB_UMAD_MAX_PORTS);
134
135static void ib_umad_add_one(struct ib_device *device);
136static void ib_umad_remove_one(struct ib_device *device, void *client_data);
137
138static void ib_umad_release_dev(struct kobject *kobj)
139{
140 struct ib_umad_device *dev =
141 container_of(kobj, struct ib_umad_device, kobj);
142
143 kfree(dev);
144}
145
146static struct kobj_type ib_umad_dev_ktype = {
147 .release = ib_umad_release_dev,
148};
149
150static int hdr_size(struct ib_umad_file *file)
151{
152 return file->use_pkey_index ? sizeof (struct ib_user_mad_hdr) :
153 sizeof (struct ib_user_mad_hdr_old);
154}
155
156
157static struct ib_mad_agent *__get_agent(struct ib_umad_file *file, int id)
158{
159 return file->agents_dead ? NULL : file->agent[id];
160}
161
162static int queue_packet(struct ib_umad_file *file,
163 struct ib_mad_agent *agent,
164 struct ib_umad_packet *packet)
165{
166 int ret = 1;
167
168 mutex_lock(&file->mutex);
169
170 for (packet->mad.hdr.id = 0;
171 packet->mad.hdr.id < IB_UMAD_MAX_AGENTS;
172 packet->mad.hdr.id++)
173 if (agent == __get_agent(file, packet->mad.hdr.id)) {
174 list_add_tail(&packet->list, &file->recv_list);
175 wake_up_interruptible(&file->recv_wait);
176 ret = 0;
177 break;
178 }
179
180 mutex_unlock(&file->mutex);
181
182 return ret;
183}
184
185static void dequeue_send(struct ib_umad_file *file,
186 struct ib_umad_packet *packet)
187{
188 spin_lock_irq(&file->send_lock);
189 list_del(&packet->list);
190 spin_unlock_irq(&file->send_lock);
191}
192
193static void send_handler(struct ib_mad_agent *agent,
194 struct ib_mad_send_wc *send_wc)
195{
196 struct ib_umad_file *file = agent->context;
197 struct ib_umad_packet *packet = send_wc->send_buf->context[0];
198
199 dequeue_send(file, packet);
200 rdma_destroy_ah(packet->msg->ah);
201 ib_free_send_mad(packet->msg);
202
203 if (send_wc->status == IB_WC_RESP_TIMEOUT_ERR) {
204 packet->length = IB_MGMT_MAD_HDR;
205 packet->mad.hdr.status = ETIMEDOUT;
206 if (!queue_packet(file, agent, packet))
207 return;
208 }
209 kfree(packet);
210}
211
212static void recv_handler(struct ib_mad_agent *agent,
213 struct ib_mad_send_buf *send_buf,
214 struct ib_mad_recv_wc *mad_recv_wc)
215{
216 struct ib_umad_file *file = agent->context;
217 struct ib_umad_packet *packet;
218
219 if (mad_recv_wc->wc->status != IB_WC_SUCCESS)
220 goto err1;
221
222 packet = kzalloc(sizeof *packet, GFP_KERNEL);
223 if (!packet)
224 goto err1;
225
226 packet->length = mad_recv_wc->mad_len;
227 packet->recv_wc = mad_recv_wc;
228
229 packet->mad.hdr.status = 0;
230 packet->mad.hdr.length = hdr_size(file) + mad_recv_wc->mad_len;
231 packet->mad.hdr.qpn = cpu_to_be32(mad_recv_wc->wc->src_qp);
232 packet->mad.hdr.lid = cpu_to_be16(mad_recv_wc->wc->slid);
233 packet->mad.hdr.sl = mad_recv_wc->wc->sl;
234 packet->mad.hdr.path_bits = mad_recv_wc->wc->dlid_path_bits;
235 packet->mad.hdr.pkey_index = mad_recv_wc->wc->pkey_index;
236 packet->mad.hdr.grh_present = !!(mad_recv_wc->wc->wc_flags & IB_WC_GRH);
237 if (packet->mad.hdr.grh_present) {
238 struct rdma_ah_attr ah_attr;
239 const struct ib_global_route *grh;
240
241 ib_init_ah_from_wc(agent->device, agent->port_num,
242 mad_recv_wc->wc, mad_recv_wc->recv_buf.grh,
243 &ah_attr);
244
245 grh = rdma_ah_read_grh(&ah_attr);
246 packet->mad.hdr.gid_index = grh->sgid_index;
247 packet->mad.hdr.hop_limit = grh->hop_limit;
248 packet->mad.hdr.traffic_class = grh->traffic_class;
249 memcpy(packet->mad.hdr.gid, &grh->dgid, 16);
250 packet->mad.hdr.flow_label = cpu_to_be32(grh->flow_label);
251 }
252
253 if (queue_packet(file, agent, packet))
254 goto err2;
255 return;
256
257err2:
258 kfree(packet);
259err1:
260 ib_free_recv_mad(mad_recv_wc);
261}
262
263static ssize_t copy_recv_mad(struct ib_umad_file *file, char __user *buf,
264 struct ib_umad_packet *packet, size_t count)
265{
266 struct ib_mad_recv_buf *recv_buf;
267 int left, seg_payload, offset, max_seg_payload;
268 size_t seg_size;
269
270 recv_buf = &packet->recv_wc->recv_buf;
271 seg_size = packet->recv_wc->mad_seg_size;
272
273
274 if ((packet->length <= seg_size &&
275 count < hdr_size(file) + packet->length) ||
276 (packet->length > seg_size &&
277 count < hdr_size(file) + seg_size))
278 return -EINVAL;
279
280 if (copy_to_user(buf, &packet->mad, hdr_size(file)))
281 return -EFAULT;
282
283 buf += hdr_size(file);
284 seg_payload = min_t(int, packet->length, seg_size);
285 if (copy_to_user(buf, recv_buf->mad, seg_payload))
286 return -EFAULT;
287
288 if (seg_payload < packet->length) {
289
290
291
292
293 if (count < hdr_size(file) + packet->length) {
294
295
296
297
298 return -ENOSPC;
299 }
300 offset = ib_get_mad_data_offset(recv_buf->mad->mad_hdr.mgmt_class);
301 max_seg_payload = seg_size - offset;
302
303 for (left = packet->length - seg_payload, buf += seg_payload;
304 left; left -= seg_payload, buf += seg_payload) {
305 recv_buf = container_of(recv_buf->list.next,
306 struct ib_mad_recv_buf, list);
307 seg_payload = min(left, max_seg_payload);
308 if (copy_to_user(buf, ((void *) recv_buf->mad) + offset,
309 seg_payload))
310 return -EFAULT;
311 }
312 }
313 return hdr_size(file) + packet->length;
314}
315
316static ssize_t copy_send_mad(struct ib_umad_file *file, char __user *buf,
317 struct ib_umad_packet *packet, size_t count)
318{
319 ssize_t size = hdr_size(file) + packet->length;
320
321 if (count < size)
322 return -EINVAL;
323
324 if (copy_to_user(buf, &packet->mad, hdr_size(file)))
325 return -EFAULT;
326
327 buf += hdr_size(file);
328
329 if (copy_to_user(buf, packet->mad.data, packet->length))
330 return -EFAULT;
331
332 return size;
333}
334
335static ssize_t ib_umad_read(struct file *filp, char __user *buf,
336 size_t count, loff_t *pos)
337{
338 struct ib_umad_file *file = filp->private_data;
339 struct ib_umad_packet *packet;
340 ssize_t ret;
341
342 if (count < hdr_size(file))
343 return -EINVAL;
344
345 mutex_lock(&file->mutex);
346
347 while (list_empty(&file->recv_list)) {
348 mutex_unlock(&file->mutex);
349
350 if (filp->f_flags & O_NONBLOCK)
351 return -EAGAIN;
352
353 if (wait_event_interruptible(file->recv_wait,
354 !list_empty(&file->recv_list)))
355 return -ERESTARTSYS;
356
357 mutex_lock(&file->mutex);
358 }
359
360 packet = list_entry(file->recv_list.next, struct ib_umad_packet, list);
361 list_del(&packet->list);
362
363 mutex_unlock(&file->mutex);
364
365 if (packet->recv_wc)
366 ret = copy_recv_mad(file, buf, packet, count);
367 else
368 ret = copy_send_mad(file, buf, packet, count);
369
370 if (ret < 0) {
371
372 mutex_lock(&file->mutex);
373 list_add(&packet->list, &file->recv_list);
374 mutex_unlock(&file->mutex);
375 } else {
376 if (packet->recv_wc)
377 ib_free_recv_mad(packet->recv_wc);
378 kfree(packet);
379 }
380 return ret;
381}
382
383static int copy_rmpp_mad(struct ib_mad_send_buf *msg, const char __user *buf)
384{
385 int left, seg;
386
387
388 if ((msg->hdr_len > IB_MGMT_RMPP_HDR) &&
389 copy_from_user(msg->mad + IB_MGMT_RMPP_HDR, buf + IB_MGMT_RMPP_HDR,
390 msg->hdr_len - IB_MGMT_RMPP_HDR))
391 return -EFAULT;
392
393
394 for (seg = 1, left = msg->data_len, buf += msg->hdr_len; left > 0;
395 seg++, left -= msg->seg_size, buf += msg->seg_size) {
396 if (copy_from_user(ib_get_rmpp_segment(msg, seg), buf,
397 min(left, msg->seg_size)))
398 return -EFAULT;
399 }
400 return 0;
401}
402
403static int same_destination(struct ib_user_mad_hdr *hdr1,
404 struct ib_user_mad_hdr *hdr2)
405{
406 if (!hdr1->grh_present && !hdr2->grh_present)
407 return (hdr1->lid == hdr2->lid);
408
409 if (hdr1->grh_present && hdr2->grh_present)
410 return !memcmp(hdr1->gid, hdr2->gid, 16);
411
412 return 0;
413}
414
415static int is_duplicate(struct ib_umad_file *file,
416 struct ib_umad_packet *packet)
417{
418 struct ib_umad_packet *sent_packet;
419 struct ib_mad_hdr *sent_hdr, *hdr;
420
421 hdr = (struct ib_mad_hdr *) packet->mad.data;
422 list_for_each_entry(sent_packet, &file->send_list, list) {
423 sent_hdr = (struct ib_mad_hdr *) sent_packet->mad.data;
424
425 if ((hdr->tid != sent_hdr->tid) ||
426 (hdr->mgmt_class != sent_hdr->mgmt_class))
427 continue;
428
429
430
431
432
433
434 if (!ib_response_mad(hdr)) {
435 if (!ib_response_mad(sent_hdr))
436 return 1;
437 continue;
438 } else if (!ib_response_mad(sent_hdr))
439 continue;
440
441 if (same_destination(&packet->mad.hdr, &sent_packet->mad.hdr))
442 return 1;
443 }
444
445 return 0;
446}
447
448static ssize_t ib_umad_write(struct file *filp, const char __user *buf,
449 size_t count, loff_t *pos)
450{
451 struct ib_umad_file *file = filp->private_data;
452 struct ib_umad_packet *packet;
453 struct ib_mad_agent *agent;
454 struct rdma_ah_attr ah_attr;
455 struct ib_ah *ah;
456 struct ib_rmpp_mad *rmpp_mad;
457 __be64 *tid;
458 int ret, data_len, hdr_len, copy_offset, rmpp_active;
459 u8 base_version;
460
461 if (count < hdr_size(file) + IB_MGMT_RMPP_HDR)
462 return -EINVAL;
463
464 packet = kzalloc(sizeof *packet + IB_MGMT_RMPP_HDR, GFP_KERNEL);
465 if (!packet)
466 return -ENOMEM;
467
468 if (copy_from_user(&packet->mad, buf, hdr_size(file))) {
469 ret = -EFAULT;
470 goto err;
471 }
472
473 if (packet->mad.hdr.id >= IB_UMAD_MAX_AGENTS) {
474 ret = -EINVAL;
475 goto err;
476 }
477
478 buf += hdr_size(file);
479
480 if (copy_from_user(packet->mad.data, buf, IB_MGMT_RMPP_HDR)) {
481 ret = -EFAULT;
482 goto err;
483 }
484
485 mutex_lock(&file->mutex);
486
487 agent = __get_agent(file, packet->mad.hdr.id);
488 if (!agent) {
489 ret = -EINVAL;
490 goto err_up;
491 }
492
493 memset(&ah_attr, 0, sizeof ah_attr);
494 ah_attr.type = rdma_ah_find_type(file->port->ib_dev,
495 file->port->port_num);
496 rdma_ah_set_dlid(&ah_attr, be16_to_cpu(packet->mad.hdr.lid));
497 rdma_ah_set_sl(&ah_attr, packet->mad.hdr.sl);
498 rdma_ah_set_path_bits(&ah_attr, packet->mad.hdr.path_bits);
499 rdma_ah_set_port_num(&ah_attr, file->port->port_num);
500 if (packet->mad.hdr.grh_present) {
501 rdma_ah_set_grh(&ah_attr, NULL,
502 be32_to_cpu(packet->mad.hdr.flow_label),
503 packet->mad.hdr.gid_index,
504 packet->mad.hdr.hop_limit,
505 packet->mad.hdr.traffic_class);
506 rdma_ah_set_dgid_raw(&ah_attr, packet->mad.hdr.gid);
507 }
508
509 ah = rdma_create_ah(agent->qp->pd, &ah_attr);
510 if (IS_ERR(ah)) {
511 ret = PTR_ERR(ah);
512 goto err_up;
513 }
514
515 rmpp_mad = (struct ib_rmpp_mad *) packet->mad.data;
516 hdr_len = ib_get_mad_data_offset(rmpp_mad->mad_hdr.mgmt_class);
517
518 if (ib_is_mad_class_rmpp(rmpp_mad->mad_hdr.mgmt_class)
519 && ib_mad_kernel_rmpp_agent(agent)) {
520 copy_offset = IB_MGMT_RMPP_HDR;
521 rmpp_active = ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) &
522 IB_MGMT_RMPP_FLAG_ACTIVE;
523 } else {
524 copy_offset = IB_MGMT_MAD_HDR;
525 rmpp_active = 0;
526 }
527
528 base_version = ((struct ib_mad_hdr *)&packet->mad.data)->base_version;
529 data_len = count - hdr_size(file) - hdr_len;
530 packet->msg = ib_create_send_mad(agent,
531 be32_to_cpu(packet->mad.hdr.qpn),
532 packet->mad.hdr.pkey_index, rmpp_active,
533 hdr_len, data_len, GFP_KERNEL,
534 base_version);
535 if (IS_ERR(packet->msg)) {
536 ret = PTR_ERR(packet->msg);
537 goto err_ah;
538 }
539
540 packet->msg->ah = ah;
541 packet->msg->timeout_ms = packet->mad.hdr.timeout_ms;
542 packet->msg->retries = packet->mad.hdr.retries;
543 packet->msg->context[0] = packet;
544
545
546 memcpy(packet->msg->mad, packet->mad.data, IB_MGMT_MAD_HDR);
547
548 if (!rmpp_active) {
549 if (copy_from_user(packet->msg->mad + copy_offset,
550 buf + copy_offset,
551 hdr_len + data_len - copy_offset)) {
552 ret = -EFAULT;
553 goto err_msg;
554 }
555 } else {
556 ret = copy_rmpp_mad(packet->msg, buf);
557 if (ret)
558 goto err_msg;
559 }
560
561
562
563
564
565
566 if (!ib_response_mad(packet->msg->mad)) {
567 tid = &((struct ib_mad_hdr *) packet->msg->mad)->tid;
568 *tid = cpu_to_be64(((u64) agent->hi_tid) << 32 |
569 (be64_to_cpup(tid) & 0xffffffff));
570 rmpp_mad->mad_hdr.tid = *tid;
571 }
572
573 if (!ib_mad_kernel_rmpp_agent(agent)
574 && ib_is_mad_class_rmpp(rmpp_mad->mad_hdr.mgmt_class)
575 && (ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) & IB_MGMT_RMPP_FLAG_ACTIVE)) {
576 spin_lock_irq(&file->send_lock);
577 list_add_tail(&packet->list, &file->send_list);
578 spin_unlock_irq(&file->send_lock);
579 } else {
580 spin_lock_irq(&file->send_lock);
581 ret = is_duplicate(file, packet);
582 if (!ret)
583 list_add_tail(&packet->list, &file->send_list);
584 spin_unlock_irq(&file->send_lock);
585 if (ret) {
586 ret = -EINVAL;
587 goto err_msg;
588 }
589 }
590
591 ret = ib_post_send_mad(packet->msg, NULL);
592 if (ret)
593 goto err_send;
594
595 mutex_unlock(&file->mutex);
596 return count;
597
598err_send:
599 dequeue_send(file, packet);
600err_msg:
601 ib_free_send_mad(packet->msg);
602err_ah:
603 rdma_destroy_ah(ah);
604err_up:
605 mutex_unlock(&file->mutex);
606err:
607 kfree(packet);
608 return ret;
609}
610
611static unsigned int ib_umad_poll(struct file *filp, struct poll_table_struct *wait)
612{
613 struct ib_umad_file *file = filp->private_data;
614
615
616 unsigned int mask = POLLOUT | POLLWRNORM;
617
618 poll_wait(filp, &file->recv_wait, wait);
619
620 if (!list_empty(&file->recv_list))
621 mask |= POLLIN | POLLRDNORM;
622
623 return mask;
624}
625
626static int ib_umad_reg_agent(struct ib_umad_file *file, void __user *arg,
627 int compat_method_mask)
628{
629 struct ib_user_mad_reg_req ureq;
630 struct ib_mad_reg_req req;
631 struct ib_mad_agent *agent = NULL;
632 int agent_id;
633 int ret;
634
635 mutex_lock(&file->port->file_mutex);
636 mutex_lock(&file->mutex);
637
638 if (!file->port->ib_dev) {
639 dev_notice(file->port->dev,
640 "ib_umad_reg_agent: invalid device\n");
641 ret = -EPIPE;
642 goto out;
643 }
644
645 if (copy_from_user(&ureq, arg, sizeof ureq)) {
646 ret = -EFAULT;
647 goto out;
648 }
649
650 if (ureq.qpn != 0 && ureq.qpn != 1) {
651 dev_notice(file->port->dev,
652 "ib_umad_reg_agent: invalid QPN %d specified\n",
653 ureq.qpn);
654 ret = -EINVAL;
655 goto out;
656 }
657
658 for (agent_id = 0; agent_id < IB_UMAD_MAX_AGENTS; ++agent_id)
659 if (!__get_agent(file, agent_id))
660 goto found;
661
662 dev_notice(file->port->dev,
663 "ib_umad_reg_agent: Max Agents (%u) reached\n",
664 IB_UMAD_MAX_AGENTS);
665 ret = -ENOMEM;
666 goto out;
667
668found:
669 if (ureq.mgmt_class) {
670 memset(&req, 0, sizeof(req));
671 req.mgmt_class = ureq.mgmt_class;
672 req.mgmt_class_version = ureq.mgmt_class_version;
673 memcpy(req.oui, ureq.oui, sizeof req.oui);
674
675 if (compat_method_mask) {
676 u32 *umm = (u32 *) ureq.method_mask;
677 int i;
678
679 for (i = 0; i < BITS_TO_LONGS(IB_MGMT_MAX_METHODS); ++i)
680 req.method_mask[i] =
681 umm[i * 2] | ((u64) umm[i * 2 + 1] << 32);
682 } else
683 memcpy(req.method_mask, ureq.method_mask,
684 sizeof req.method_mask);
685 }
686
687 agent = ib_register_mad_agent(file->port->ib_dev, file->port->port_num,
688 ureq.qpn ? IB_QPT_GSI : IB_QPT_SMI,
689 ureq.mgmt_class ? &req : NULL,
690 ureq.rmpp_version,
691 send_handler, recv_handler, file, 0);
692 if (IS_ERR(agent)) {
693 ret = PTR_ERR(agent);
694 agent = NULL;
695 goto out;
696 }
697
698 if (put_user(agent_id,
699 (u32 __user *) (arg + offsetof(struct ib_user_mad_reg_req, id)))) {
700 ret = -EFAULT;
701 goto out;
702 }
703
704 if (!file->already_used) {
705 file->already_used = 1;
706 if (!file->use_pkey_index) {
707 dev_warn(file->port->dev,
708 "process %s did not enable P_Key index support.\n",
709 current->comm);
710 dev_warn(file->port->dev,
711 " Documentation/infiniband/user_mad.txt has info on the new ABI.\n");
712 }
713 }
714
715 file->agent[agent_id] = agent;
716 ret = 0;
717
718out:
719 mutex_unlock(&file->mutex);
720
721 if (ret && agent)
722 ib_unregister_mad_agent(agent);
723
724 mutex_unlock(&file->port->file_mutex);
725
726 return ret;
727}
728
729static int ib_umad_reg_agent2(struct ib_umad_file *file, void __user *arg)
730{
731 struct ib_user_mad_reg_req2 ureq;
732 struct ib_mad_reg_req req;
733 struct ib_mad_agent *agent = NULL;
734 int agent_id;
735 int ret;
736
737 mutex_lock(&file->port->file_mutex);
738 mutex_lock(&file->mutex);
739
740 if (!file->port->ib_dev) {
741 dev_notice(file->port->dev,
742 "ib_umad_reg_agent2: invalid device\n");
743 ret = -EPIPE;
744 goto out;
745 }
746
747 if (copy_from_user(&ureq, arg, sizeof(ureq))) {
748 ret = -EFAULT;
749 goto out;
750 }
751
752 if (ureq.qpn != 0 && ureq.qpn != 1) {
753 dev_notice(file->port->dev,
754 "ib_umad_reg_agent2: invalid QPN %d specified\n",
755 ureq.qpn);
756 ret = -EINVAL;
757 goto out;
758 }
759
760 if (ureq.flags & ~IB_USER_MAD_REG_FLAGS_CAP) {
761 dev_notice(file->port->dev,
762 "ib_umad_reg_agent2 failed: invalid registration flags specified 0x%x; supported 0x%x\n",
763 ureq.flags, IB_USER_MAD_REG_FLAGS_CAP);
764 ret = -EINVAL;
765
766 if (put_user((u32)IB_USER_MAD_REG_FLAGS_CAP,
767 (u32 __user *) (arg + offsetof(struct
768 ib_user_mad_reg_req2, flags))))
769 ret = -EFAULT;
770
771 goto out;
772 }
773
774 for (agent_id = 0; agent_id < IB_UMAD_MAX_AGENTS; ++agent_id)
775 if (!__get_agent(file, agent_id))
776 goto found;
777
778 dev_notice(file->port->dev,
779 "ib_umad_reg_agent2: Max Agents (%u) reached\n",
780 IB_UMAD_MAX_AGENTS);
781 ret = -ENOMEM;
782 goto out;
783
784found:
785 if (ureq.mgmt_class) {
786 memset(&req, 0, sizeof(req));
787 req.mgmt_class = ureq.mgmt_class;
788 req.mgmt_class_version = ureq.mgmt_class_version;
789 if (ureq.oui & 0xff000000) {
790 dev_notice(file->port->dev,
791 "ib_umad_reg_agent2 failed: oui invalid 0x%08x\n",
792 ureq.oui);
793 ret = -EINVAL;
794 goto out;
795 }
796 req.oui[2] = ureq.oui & 0x0000ff;
797 req.oui[1] = (ureq.oui & 0x00ff00) >> 8;
798 req.oui[0] = (ureq.oui & 0xff0000) >> 16;
799 memcpy(req.method_mask, ureq.method_mask,
800 sizeof(req.method_mask));
801 }
802
803 agent = ib_register_mad_agent(file->port->ib_dev, file->port->port_num,
804 ureq.qpn ? IB_QPT_GSI : IB_QPT_SMI,
805 ureq.mgmt_class ? &req : NULL,
806 ureq.rmpp_version,
807 send_handler, recv_handler, file,
808 ureq.flags);
809 if (IS_ERR(agent)) {
810 ret = PTR_ERR(agent);
811 agent = NULL;
812 goto out;
813 }
814
815 if (put_user(agent_id,
816 (u32 __user *)(arg +
817 offsetof(struct ib_user_mad_reg_req2, id)))) {
818 ret = -EFAULT;
819 goto out;
820 }
821
822 if (!file->already_used) {
823 file->already_used = 1;
824 file->use_pkey_index = 1;
825 }
826
827 file->agent[agent_id] = agent;
828 ret = 0;
829
830out:
831 mutex_unlock(&file->mutex);
832
833 if (ret && agent)
834 ib_unregister_mad_agent(agent);
835
836 mutex_unlock(&file->port->file_mutex);
837
838 return ret;
839}
840
841
842static int ib_umad_unreg_agent(struct ib_umad_file *file, u32 __user *arg)
843{
844 struct ib_mad_agent *agent = NULL;
845 u32 id;
846 int ret = 0;
847
848 if (get_user(id, arg))
849 return -EFAULT;
850
851 mutex_lock(&file->port->file_mutex);
852 mutex_lock(&file->mutex);
853
854 if (id >= IB_UMAD_MAX_AGENTS || !__get_agent(file, id)) {
855 ret = -EINVAL;
856 goto out;
857 }
858
859 agent = file->agent[id];
860 file->agent[id] = NULL;
861
862out:
863 mutex_unlock(&file->mutex);
864
865 if (agent)
866 ib_unregister_mad_agent(agent);
867
868 mutex_unlock(&file->port->file_mutex);
869
870 return ret;
871}
872
873static long ib_umad_enable_pkey(struct ib_umad_file *file)
874{
875 int ret = 0;
876
877 mutex_lock(&file->mutex);
878 if (file->already_used)
879 ret = -EINVAL;
880 else
881 file->use_pkey_index = 1;
882 mutex_unlock(&file->mutex);
883
884 return ret;
885}
886
887static long ib_umad_ioctl(struct file *filp, unsigned int cmd,
888 unsigned long arg)
889{
890 switch (cmd) {
891 case IB_USER_MAD_REGISTER_AGENT:
892 return ib_umad_reg_agent(filp->private_data, (void __user *) arg, 0);
893 case IB_USER_MAD_UNREGISTER_AGENT:
894 return ib_umad_unreg_agent(filp->private_data, (__u32 __user *) arg);
895 case IB_USER_MAD_ENABLE_PKEY:
896 return ib_umad_enable_pkey(filp->private_data);
897 case IB_USER_MAD_REGISTER_AGENT2:
898 return ib_umad_reg_agent2(filp->private_data, (void __user *) arg);
899 default:
900 return -ENOIOCTLCMD;
901 }
902}
903
904#ifdef CONFIG_COMPAT
905static long ib_umad_compat_ioctl(struct file *filp, unsigned int cmd,
906 unsigned long arg)
907{
908 switch (cmd) {
909 case IB_USER_MAD_REGISTER_AGENT:
910 return ib_umad_reg_agent(filp->private_data, compat_ptr(arg), 1);
911 case IB_USER_MAD_UNREGISTER_AGENT:
912 return ib_umad_unreg_agent(filp->private_data, compat_ptr(arg));
913 case IB_USER_MAD_ENABLE_PKEY:
914 return ib_umad_enable_pkey(filp->private_data);
915 case IB_USER_MAD_REGISTER_AGENT2:
916 return ib_umad_reg_agent2(filp->private_data, compat_ptr(arg));
917 default:
918 return -ENOIOCTLCMD;
919 }
920}
921#endif
922
923
924
925
926
927
928
929
930
931
932static int ib_umad_open(struct inode *inode, struct file *filp)
933{
934 struct ib_umad_port *port;
935 struct ib_umad_file *file;
936 int ret = -ENXIO;
937
938 port = container_of(inode->i_cdev, struct ib_umad_port, cdev);
939
940 mutex_lock(&port->file_mutex);
941
942 if (!port->ib_dev)
943 goto out;
944
945 ret = -ENOMEM;
946 file = kzalloc(sizeof *file, GFP_KERNEL);
947 if (!file)
948 goto out;
949
950 mutex_init(&file->mutex);
951 spin_lock_init(&file->send_lock);
952 INIT_LIST_HEAD(&file->recv_list);
953 INIT_LIST_HEAD(&file->send_list);
954 init_waitqueue_head(&file->recv_wait);
955
956 file->port = port;
957 filp->private_data = file;
958
959 list_add_tail(&file->port_list, &port->file_list);
960
961 ret = nonseekable_open(inode, filp);
962 if (ret) {
963 list_del(&file->port_list);
964 kfree(file);
965 goto out;
966 }
967
968 kobject_get(&port->umad_dev->kobj);
969
970out:
971 mutex_unlock(&port->file_mutex);
972 return ret;
973}
974
975static int ib_umad_close(struct inode *inode, struct file *filp)
976{
977 struct ib_umad_file *file = filp->private_data;
978 struct ib_umad_device *dev = file->port->umad_dev;
979 struct ib_umad_packet *packet, *tmp;
980 int already_dead;
981 int i;
982
983 mutex_lock(&file->port->file_mutex);
984 mutex_lock(&file->mutex);
985
986 already_dead = file->agents_dead;
987 file->agents_dead = 1;
988
989 list_for_each_entry_safe(packet, tmp, &file->recv_list, list) {
990 if (packet->recv_wc)
991 ib_free_recv_mad(packet->recv_wc);
992 kfree(packet);
993 }
994
995 list_del(&file->port_list);
996
997 mutex_unlock(&file->mutex);
998
999 if (!already_dead)
1000 for (i = 0; i < IB_UMAD_MAX_AGENTS; ++i)
1001 if (file->agent[i])
1002 ib_unregister_mad_agent(file->agent[i]);
1003
1004 mutex_unlock(&file->port->file_mutex);
1005
1006 kfree(file);
1007 kobject_put(&dev->kobj);
1008
1009 return 0;
1010}
1011
1012static const struct file_operations umad_fops = {
1013 .owner = THIS_MODULE,
1014 .read = ib_umad_read,
1015 .write = ib_umad_write,
1016 .poll = ib_umad_poll,
1017 .unlocked_ioctl = ib_umad_ioctl,
1018#ifdef CONFIG_COMPAT
1019 .compat_ioctl = ib_umad_compat_ioctl,
1020#endif
1021 .open = ib_umad_open,
1022 .release = ib_umad_close,
1023 .llseek = no_llseek,
1024};
1025
1026static int ib_umad_sm_open(struct inode *inode, struct file *filp)
1027{
1028 struct ib_umad_port *port;
1029 struct ib_port_modify props = {
1030 .set_port_cap_mask = IB_PORT_SM
1031 };
1032 int ret;
1033
1034 port = container_of(inode->i_cdev, struct ib_umad_port, sm_cdev);
1035
1036 if (filp->f_flags & O_NONBLOCK) {
1037 if (down_trylock(&port->sm_sem)) {
1038 ret = -EAGAIN;
1039 goto fail;
1040 }
1041 } else {
1042 if (down_interruptible(&port->sm_sem)) {
1043 ret = -ERESTARTSYS;
1044 goto fail;
1045 }
1046 }
1047
1048 ret = ib_modify_port(port->ib_dev, port->port_num, 0, &props);
1049 if (ret)
1050 goto err_up_sem;
1051
1052 filp->private_data = port;
1053
1054 ret = nonseekable_open(inode, filp);
1055 if (ret)
1056 goto err_clr_sm_cap;
1057
1058 kobject_get(&port->umad_dev->kobj);
1059
1060 return 0;
1061
1062err_clr_sm_cap:
1063 swap(props.set_port_cap_mask, props.clr_port_cap_mask);
1064 ib_modify_port(port->ib_dev, port->port_num, 0, &props);
1065
1066err_up_sem:
1067 up(&port->sm_sem);
1068
1069fail:
1070 return ret;
1071}
1072
1073static int ib_umad_sm_close(struct inode *inode, struct file *filp)
1074{
1075 struct ib_umad_port *port = filp->private_data;
1076 struct ib_port_modify props = {
1077 .clr_port_cap_mask = IB_PORT_SM
1078 };
1079 int ret = 0;
1080
1081 mutex_lock(&port->file_mutex);
1082 if (port->ib_dev)
1083 ret = ib_modify_port(port->ib_dev, port->port_num, 0, &props);
1084 mutex_unlock(&port->file_mutex);
1085
1086 up(&port->sm_sem);
1087
1088 kobject_put(&port->umad_dev->kobj);
1089
1090 return ret;
1091}
1092
1093static const struct file_operations umad_sm_fops = {
1094 .owner = THIS_MODULE,
1095 .open = ib_umad_sm_open,
1096 .release = ib_umad_sm_close,
1097 .llseek = no_llseek,
1098};
1099
1100static struct ib_client umad_client = {
1101 .name = "umad",
1102 .add = ib_umad_add_one,
1103 .remove = ib_umad_remove_one
1104};
1105
1106static ssize_t show_ibdev(struct device *dev, struct device_attribute *attr,
1107 char *buf)
1108{
1109 struct ib_umad_port *port = dev_get_drvdata(dev);
1110
1111 if (!port)
1112 return -ENODEV;
1113
1114 return sprintf(buf, "%s\n", port->ib_dev->name);
1115}
1116static DEVICE_ATTR(ibdev, S_IRUGO, show_ibdev, NULL);
1117
1118static ssize_t show_port(struct device *dev, struct device_attribute *attr,
1119 char *buf)
1120{
1121 struct ib_umad_port *port = dev_get_drvdata(dev);
1122
1123 if (!port)
1124 return -ENODEV;
1125
1126 return sprintf(buf, "%d\n", port->port_num);
1127}
1128static DEVICE_ATTR(port, S_IRUGO, show_port, NULL);
1129
1130static CLASS_ATTR_STRING(abi_version, S_IRUGO,
1131 __stringify(IB_USER_MAD_ABI_VERSION));
1132
1133static dev_t overflow_maj;
1134static DECLARE_BITMAP(overflow_map, IB_UMAD_MAX_PORTS);
1135static int find_overflow_devnum(struct ib_device *device)
1136{
1137 int ret;
1138
1139 if (!overflow_maj) {
1140 ret = alloc_chrdev_region(&overflow_maj, 0, IB_UMAD_MAX_PORTS * 2,
1141 "infiniband_mad");
1142 if (ret) {
1143 dev_err(&device->dev,
1144 "couldn't register dynamic device number\n");
1145 return ret;
1146 }
1147 }
1148
1149 ret = find_first_zero_bit(overflow_map, IB_UMAD_MAX_PORTS);
1150 if (ret >= IB_UMAD_MAX_PORTS)
1151 return -1;
1152
1153 return ret;
1154}
1155
1156static int ib_umad_init_port(struct ib_device *device, int port_num,
1157 struct ib_umad_device *umad_dev,
1158 struct ib_umad_port *port)
1159{
1160 int devnum;
1161 dev_t base;
1162
1163 spin_lock(&port_lock);
1164 devnum = find_first_zero_bit(dev_map, IB_UMAD_MAX_PORTS);
1165 if (devnum >= IB_UMAD_MAX_PORTS) {
1166 spin_unlock(&port_lock);
1167 devnum = find_overflow_devnum(device);
1168 if (devnum < 0)
1169 return -1;
1170
1171 spin_lock(&port_lock);
1172 port->dev_num = devnum + IB_UMAD_MAX_PORTS;
1173 base = devnum + overflow_maj;
1174 set_bit(devnum, overflow_map);
1175 } else {
1176 port->dev_num = devnum;
1177 base = devnum + base_dev;
1178 set_bit(devnum, dev_map);
1179 }
1180 spin_unlock(&port_lock);
1181
1182 port->ib_dev = device;
1183 port->port_num = port_num;
1184 sema_init(&port->sm_sem, 1);
1185 mutex_init(&port->file_mutex);
1186 INIT_LIST_HEAD(&port->file_list);
1187
1188 cdev_init(&port->cdev, &umad_fops);
1189 port->cdev.owner = THIS_MODULE;
1190 cdev_set_parent(&port->cdev, &umad_dev->kobj);
1191 kobject_set_name(&port->cdev.kobj, "umad%d", port->dev_num);
1192 if (cdev_add(&port->cdev, base, 1))
1193 goto err_cdev;
1194
1195 port->dev = device_create(umad_class, device->dev.parent,
1196 port->cdev.dev, port,
1197 "umad%d", port->dev_num);
1198 if (IS_ERR(port->dev))
1199 goto err_cdev;
1200
1201 if (device_create_file(port->dev, &dev_attr_ibdev))
1202 goto err_dev;
1203 if (device_create_file(port->dev, &dev_attr_port))
1204 goto err_dev;
1205
1206 base += IB_UMAD_MAX_PORTS;
1207 cdev_init(&port->sm_cdev, &umad_sm_fops);
1208 port->sm_cdev.owner = THIS_MODULE;
1209 cdev_set_parent(&port->sm_cdev, &umad_dev->kobj);
1210 kobject_set_name(&port->sm_cdev.kobj, "issm%d", port->dev_num);
1211 if (cdev_add(&port->sm_cdev, base, 1))
1212 goto err_sm_cdev;
1213
1214 port->sm_dev = device_create(umad_class, device->dev.parent,
1215 port->sm_cdev.dev, port,
1216 "issm%d", port->dev_num);
1217 if (IS_ERR(port->sm_dev))
1218 goto err_sm_cdev;
1219
1220 if (device_create_file(port->sm_dev, &dev_attr_ibdev))
1221 goto err_sm_dev;
1222 if (device_create_file(port->sm_dev, &dev_attr_port))
1223 goto err_sm_dev;
1224
1225 return 0;
1226
1227err_sm_dev:
1228 device_destroy(umad_class, port->sm_cdev.dev);
1229
1230err_sm_cdev:
1231 cdev_del(&port->sm_cdev);
1232
1233err_dev:
1234 device_destroy(umad_class, port->cdev.dev);
1235
1236err_cdev:
1237 cdev_del(&port->cdev);
1238 if (port->dev_num < IB_UMAD_MAX_PORTS)
1239 clear_bit(devnum, dev_map);
1240 else
1241 clear_bit(devnum, overflow_map);
1242
1243 return -1;
1244}
1245
1246static void ib_umad_kill_port(struct ib_umad_port *port)
1247{
1248 struct ib_umad_file *file;
1249 int id;
1250
1251 dev_set_drvdata(port->dev, NULL);
1252 dev_set_drvdata(port->sm_dev, NULL);
1253
1254 device_destroy(umad_class, port->cdev.dev);
1255 device_destroy(umad_class, port->sm_cdev.dev);
1256
1257 cdev_del(&port->cdev);
1258 cdev_del(&port->sm_cdev);
1259
1260 mutex_lock(&port->file_mutex);
1261
1262 port->ib_dev = NULL;
1263
1264 list_for_each_entry(file, &port->file_list, port_list) {
1265 mutex_lock(&file->mutex);
1266 file->agents_dead = 1;
1267 mutex_unlock(&file->mutex);
1268
1269 for (id = 0; id < IB_UMAD_MAX_AGENTS; ++id)
1270 if (file->agent[id])
1271 ib_unregister_mad_agent(file->agent[id]);
1272 }
1273
1274 mutex_unlock(&port->file_mutex);
1275
1276 if (port->dev_num < IB_UMAD_MAX_PORTS)
1277 clear_bit(port->dev_num, dev_map);
1278 else
1279 clear_bit(port->dev_num - IB_UMAD_MAX_PORTS, overflow_map);
1280}
1281
1282static void ib_umad_add_one(struct ib_device *device)
1283{
1284 struct ib_umad_device *umad_dev;
1285 int s, e, i;
1286 int count = 0;
1287
1288 s = rdma_start_port(device);
1289 e = rdma_end_port(device);
1290
1291 umad_dev = kzalloc(sizeof *umad_dev +
1292 (e - s + 1) * sizeof (struct ib_umad_port),
1293 GFP_KERNEL);
1294 if (!umad_dev)
1295 return;
1296
1297 kobject_init(&umad_dev->kobj, &ib_umad_dev_ktype);
1298
1299 for (i = s; i <= e; ++i) {
1300 if (!rdma_cap_ib_mad(device, i))
1301 continue;
1302
1303 umad_dev->port[i - s].umad_dev = umad_dev;
1304
1305 if (ib_umad_init_port(device, i, umad_dev,
1306 &umad_dev->port[i - s]))
1307 goto err;
1308
1309 count++;
1310 }
1311
1312 if (!count)
1313 goto free;
1314
1315 ib_set_client_data(device, &umad_client, umad_dev);
1316
1317 return;
1318
1319err:
1320 while (--i >= s) {
1321 if (!rdma_cap_ib_mad(device, i))
1322 continue;
1323
1324 ib_umad_kill_port(&umad_dev->port[i - s]);
1325 }
1326free:
1327 kobject_put(&umad_dev->kobj);
1328}
1329
1330static void ib_umad_remove_one(struct ib_device *device, void *client_data)
1331{
1332 struct ib_umad_device *umad_dev = client_data;
1333 int i;
1334
1335 if (!umad_dev)
1336 return;
1337
1338 for (i = 0; i <= rdma_end_port(device) - rdma_start_port(device); ++i) {
1339 if (rdma_cap_ib_mad(device, i + rdma_start_port(device)))
1340 ib_umad_kill_port(&umad_dev->port[i]);
1341 }
1342
1343 kobject_put(&umad_dev->kobj);
1344}
1345
1346static char *umad_devnode(struct device *dev, umode_t *mode)
1347{
1348 return kasprintf(GFP_KERNEL, "infiniband/%s", dev_name(dev));
1349}
1350
1351static int __init ib_umad_init(void)
1352{
1353 int ret;
1354
1355 ret = register_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2,
1356 "infiniband_mad");
1357 if (ret) {
1358 pr_err("couldn't register device number\n");
1359 goto out;
1360 }
1361
1362 umad_class = class_create(THIS_MODULE, "infiniband_mad");
1363 if (IS_ERR(umad_class)) {
1364 ret = PTR_ERR(umad_class);
1365 pr_err("couldn't create class infiniband_mad\n");
1366 goto out_chrdev;
1367 }
1368
1369 umad_class->devnode = umad_devnode;
1370
1371 ret = class_create_file(umad_class, &class_attr_abi_version.attr);
1372 if (ret) {
1373 pr_err("couldn't create abi_version attribute\n");
1374 goto out_class;
1375 }
1376
1377 ret = ib_register_client(&umad_client);
1378 if (ret) {
1379 pr_err("couldn't register ib_umad client\n");
1380 goto out_class;
1381 }
1382
1383 return 0;
1384
1385out_class:
1386 class_destroy(umad_class);
1387
1388out_chrdev:
1389 unregister_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2);
1390
1391out:
1392 return ret;
1393}
1394
1395static void __exit ib_umad_cleanup(void)
1396{
1397 ib_unregister_client(&umad_client);
1398 class_destroy(umad_class);
1399 unregister_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2);
1400 if (overflow_maj)
1401 unregister_chrdev_region(overflow_maj, IB_UMAD_MAX_PORTS * 2);
1402}
1403
1404module_init(ib_umad_init);
1405module_exit(ib_umad_cleanup);
1406