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#include <linux/module.h>
37#include <linux/errno.h>
38#include <linux/slab.h>
39#include <linux/workqueue.h>
40#include <linux/netdevice.h>
41#include <net/addrconf.h>
42
43#include <rdma/ib_cache.h>
44
45#include "core_priv.h"
46
47struct ib_pkey_cache {
48 int table_len;
49 u16 table[0];
50};
51
52struct ib_update_work {
53 struct work_struct work;
54 struct ib_device *device;
55 u8 port_num;
56 bool enforce_security;
57};
58
59union ib_gid zgid;
60EXPORT_SYMBOL(zgid);
61
62enum gid_attr_find_mask {
63 GID_ATTR_FIND_MASK_GID = 1UL << 0,
64 GID_ATTR_FIND_MASK_NETDEV = 1UL << 1,
65 GID_ATTR_FIND_MASK_DEFAULT = 1UL << 2,
66 GID_ATTR_FIND_MASK_GID_TYPE = 1UL << 3,
67};
68
69enum gid_table_entry_state {
70 GID_TABLE_ENTRY_INVALID = 1,
71 GID_TABLE_ENTRY_VALID = 2,
72
73
74
75
76
77
78 GID_TABLE_ENTRY_PENDING_DEL = 3,
79};
80
81struct roce_gid_ndev_storage {
82 struct rcu_head rcu_head;
83 struct net_device *ndev;
84};
85
86struct ib_gid_table_entry {
87 struct kref kref;
88 struct work_struct del_work;
89 struct ib_gid_attr attr;
90 void *context;
91
92
93
94
95
96 struct roce_gid_ndev_storage *ndev_storage;
97 enum gid_table_entry_state state;
98};
99
100struct ib_gid_table {
101 int sz;
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116 struct mutex lock;
117
118
119 rwlock_t rwlock;
120 struct ib_gid_table_entry **data_vec;
121
122 u32 default_gid_indices;
123};
124
125static void dispatch_gid_change_event(struct ib_device *ib_dev, u8 port)
126{
127 struct ib_event event;
128
129 event.device = ib_dev;
130 event.element.port_num = port;
131 event.event = IB_EVENT_GID_CHANGE;
132
133 ib_dispatch_event(&event);
134}
135
136static const char * const gid_type_str[] = {
137 [IB_GID_TYPE_IB] = "IB/RoCE v1",
138 [IB_GID_TYPE_ROCE_UDP_ENCAP] = "RoCE v2",
139};
140
141const char *ib_cache_gid_type_str(enum ib_gid_type gid_type)
142{
143 if (gid_type < ARRAY_SIZE(gid_type_str) && gid_type_str[gid_type])
144 return gid_type_str[gid_type];
145
146 return "Invalid GID type";
147}
148EXPORT_SYMBOL(ib_cache_gid_type_str);
149
150
151
152
153
154bool rdma_is_zero_gid(const union ib_gid *gid)
155{
156 return !memcmp(gid, &zgid, sizeof(*gid));
157}
158EXPORT_SYMBOL(rdma_is_zero_gid);
159
160
161
162
163
164
165
166
167static bool is_gid_index_default(const struct ib_gid_table *table,
168 unsigned int index)
169{
170 return index < 32 && (BIT(index) & table->default_gid_indices);
171}
172
173int ib_cache_gid_parse_type_str(const char *buf)
174{
175 unsigned int i;
176 size_t len;
177 int err = -EINVAL;
178
179 len = strlen(buf);
180 if (len == 0)
181 return -EINVAL;
182
183 if (buf[len - 1] == '\n')
184 len--;
185
186 for (i = 0; i < ARRAY_SIZE(gid_type_str); ++i)
187 if (gid_type_str[i] && !strncmp(buf, gid_type_str[i], len) &&
188 len == strlen(gid_type_str[i])) {
189 err = i;
190 break;
191 }
192
193 return err;
194}
195EXPORT_SYMBOL(ib_cache_gid_parse_type_str);
196
197static struct ib_gid_table *rdma_gid_table(struct ib_device *device, u8 port)
198{
199 return device->port_data[port].cache.gid;
200}
201
202static bool is_gid_entry_free(const struct ib_gid_table_entry *entry)
203{
204 return !entry;
205}
206
207static bool is_gid_entry_valid(const struct ib_gid_table_entry *entry)
208{
209 return entry && entry->state == GID_TABLE_ENTRY_VALID;
210}
211
212static void schedule_free_gid(struct kref *kref)
213{
214 struct ib_gid_table_entry *entry =
215 container_of(kref, struct ib_gid_table_entry, kref);
216
217 queue_work(ib_wq, &entry->del_work);
218}
219
220static void put_gid_ndev(struct rcu_head *head)
221{
222 struct roce_gid_ndev_storage *storage =
223 container_of(head, struct roce_gid_ndev_storage, rcu_head);
224
225 WARN_ON(!storage->ndev);
226
227
228
229
230 dev_put(storage->ndev);
231 kfree(storage);
232}
233
234static void free_gid_entry_locked(struct ib_gid_table_entry *entry)
235{
236 struct ib_device *device = entry->attr.device;
237 u8 port_num = entry->attr.port_num;
238 struct ib_gid_table *table = rdma_gid_table(device, port_num);
239
240 dev_dbg(&device->dev, "%s port=%d index=%d gid %pI6\n", __func__,
241 port_num, entry->attr.index, entry->attr.gid.raw);
242
243 write_lock_irq(&table->rwlock);
244
245
246
247
248
249
250
251 if (entry == table->data_vec[entry->attr.index])
252 table->data_vec[entry->attr.index] = NULL;
253
254 write_unlock_irq(&table->rwlock);
255
256 if (entry->ndev_storage)
257 call_rcu(&entry->ndev_storage->rcu_head, put_gid_ndev);
258 kfree(entry);
259}
260
261static void free_gid_entry(struct kref *kref)
262{
263 struct ib_gid_table_entry *entry =
264 container_of(kref, struct ib_gid_table_entry, kref);
265
266 free_gid_entry_locked(entry);
267}
268
269
270
271
272
273
274
275
276
277static void free_gid_work(struct work_struct *work)
278{
279 struct ib_gid_table_entry *entry =
280 container_of(work, struct ib_gid_table_entry, del_work);
281 struct ib_device *device = entry->attr.device;
282 u8 port_num = entry->attr.port_num;
283 struct ib_gid_table *table = rdma_gid_table(device, port_num);
284
285 mutex_lock(&table->lock);
286 free_gid_entry_locked(entry);
287 mutex_unlock(&table->lock);
288}
289
290static struct ib_gid_table_entry *
291alloc_gid_entry(const struct ib_gid_attr *attr)
292{
293 struct ib_gid_table_entry *entry;
294 struct net_device *ndev;
295
296 entry = kzalloc(sizeof(*entry), GFP_KERNEL);
297 if (!entry)
298 return NULL;
299
300 ndev = rcu_dereference_protected(attr->ndev, 1);
301 if (ndev) {
302 entry->ndev_storage = kzalloc(sizeof(*entry->ndev_storage),
303 GFP_KERNEL);
304 if (!entry->ndev_storage) {
305 kfree(entry);
306 return NULL;
307 }
308 dev_hold(ndev);
309 entry->ndev_storage->ndev = ndev;
310 }
311 kref_init(&entry->kref);
312 memcpy(&entry->attr, attr, sizeof(*attr));
313 INIT_WORK(&entry->del_work, free_gid_work);
314 entry->state = GID_TABLE_ENTRY_INVALID;
315 return entry;
316}
317
318static void store_gid_entry(struct ib_gid_table *table,
319 struct ib_gid_table_entry *entry)
320{
321 entry->state = GID_TABLE_ENTRY_VALID;
322
323 dev_dbg(&entry->attr.device->dev, "%s port=%d index=%d gid %pI6\n",
324 __func__, entry->attr.port_num, entry->attr.index,
325 entry->attr.gid.raw);
326
327 lockdep_assert_held(&table->lock);
328 write_lock_irq(&table->rwlock);
329 table->data_vec[entry->attr.index] = entry;
330 write_unlock_irq(&table->rwlock);
331}
332
333static void get_gid_entry(struct ib_gid_table_entry *entry)
334{
335 kref_get(&entry->kref);
336}
337
338static void put_gid_entry(struct ib_gid_table_entry *entry)
339{
340 kref_put(&entry->kref, schedule_free_gid);
341}
342
343static void put_gid_entry_locked(struct ib_gid_table_entry *entry)
344{
345 kref_put(&entry->kref, free_gid_entry);
346}
347
348static int add_roce_gid(struct ib_gid_table_entry *entry)
349{
350 const struct ib_gid_attr *attr = &entry->attr;
351 int ret;
352
353 if (!attr->ndev) {
354 dev_err(&attr->device->dev, "%s NULL netdev port=%d index=%d\n",
355 __func__, attr->port_num, attr->index);
356 return -EINVAL;
357 }
358 if (rdma_cap_roce_gid_table(attr->device, attr->port_num)) {
359 ret = attr->device->ops.add_gid(attr, &entry->context);
360 if (ret) {
361 dev_err(&attr->device->dev,
362 "%s GID add failed port=%d index=%d\n",
363 __func__, attr->port_num, attr->index);
364 return ret;
365 }
366 }
367 return 0;
368}
369
370
371
372
373
374
375
376
377
378
379static void del_gid(struct ib_device *ib_dev, u8 port,
380 struct ib_gid_table *table, int ix)
381{
382 struct roce_gid_ndev_storage *ndev_storage;
383 struct ib_gid_table_entry *entry;
384
385 lockdep_assert_held(&table->lock);
386
387 dev_dbg(&ib_dev->dev, "%s port=%d index=%d gid %pI6\n", __func__, port,
388 ix, table->data_vec[ix]->attr.gid.raw);
389
390 write_lock_irq(&table->rwlock);
391 entry = table->data_vec[ix];
392 entry->state = GID_TABLE_ENTRY_PENDING_DEL;
393
394
395
396 if (!rdma_protocol_roce(ib_dev, port))
397 table->data_vec[ix] = NULL;
398 write_unlock_irq(&table->rwlock);
399
400 ndev_storage = entry->ndev_storage;
401 if (ndev_storage) {
402 entry->ndev_storage = NULL;
403 rcu_assign_pointer(entry->attr.ndev, NULL);
404 call_rcu(&ndev_storage->rcu_head, put_gid_ndev);
405 }
406
407 if (rdma_cap_roce_gid_table(ib_dev, port))
408 ib_dev->ops.del_gid(&entry->attr, &entry->context);
409
410 put_gid_entry_locked(entry);
411}
412
413
414
415
416
417
418
419
420
421
422
423static int add_modify_gid(struct ib_gid_table *table,
424 const struct ib_gid_attr *attr)
425{
426 struct ib_gid_table_entry *entry;
427 int ret = 0;
428
429
430
431
432
433 if (is_gid_entry_valid(table->data_vec[attr->index]))
434 del_gid(attr->device, attr->port_num, table, attr->index);
435
436
437
438
439
440
441 if (rdma_is_zero_gid(&attr->gid))
442 return 0;
443
444 entry = alloc_gid_entry(attr);
445 if (!entry)
446 return -ENOMEM;
447
448 if (rdma_protocol_roce(attr->device, attr->port_num)) {
449 ret = add_roce_gid(entry);
450 if (ret)
451 goto done;
452 }
453
454 store_gid_entry(table, entry);
455 return 0;
456
457done:
458 put_gid_entry(entry);
459 return ret;
460}
461
462
463static int find_gid(struct ib_gid_table *table, const union ib_gid *gid,
464 const struct ib_gid_attr *val, bool default_gid,
465 unsigned long mask, int *pempty)
466{
467 int i = 0;
468 int found = -1;
469 int empty = pempty ? -1 : 0;
470
471 while (i < table->sz && (found < 0 || empty < 0)) {
472 struct ib_gid_table_entry *data = table->data_vec[i];
473 struct ib_gid_attr *attr;
474 int curr_index = i;
475
476 i++;
477
478
479
480
481
482
483 if (pempty && empty < 0) {
484 if (is_gid_entry_free(data) &&
485 default_gid ==
486 is_gid_index_default(table, curr_index)) {
487
488
489
490
491
492
493
494
495 empty = curr_index;
496 }
497 }
498
499
500
501
502
503
504
505 if (!is_gid_entry_valid(data))
506 continue;
507
508 if (found >= 0)
509 continue;
510
511 attr = &data->attr;
512 if (mask & GID_ATTR_FIND_MASK_GID_TYPE &&
513 attr->gid_type != val->gid_type)
514 continue;
515
516 if (mask & GID_ATTR_FIND_MASK_GID &&
517 memcmp(gid, &data->attr.gid, sizeof(*gid)))
518 continue;
519
520 if (mask & GID_ATTR_FIND_MASK_NETDEV &&
521 attr->ndev != val->ndev)
522 continue;
523
524 if (mask & GID_ATTR_FIND_MASK_DEFAULT &&
525 is_gid_index_default(table, curr_index) != default_gid)
526 continue;
527
528 found = curr_index;
529 }
530
531 if (pempty)
532 *pempty = empty;
533
534 return found;
535}
536
537static void make_default_gid(struct net_device *dev, union ib_gid *gid)
538{
539 gid->global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
540 addrconf_ifid_eui48(&gid->raw[8], dev);
541}
542
543static int __ib_cache_gid_add(struct ib_device *ib_dev, u8 port,
544 union ib_gid *gid, struct ib_gid_attr *attr,
545 unsigned long mask, bool default_gid)
546{
547 struct ib_gid_table *table;
548 int ret = 0;
549 int empty;
550 int ix;
551
552
553
554
555
556 if (rdma_is_zero_gid(gid))
557 return -EINVAL;
558
559 table = rdma_gid_table(ib_dev, port);
560
561 mutex_lock(&table->lock);
562
563 ix = find_gid(table, gid, attr, default_gid, mask, &empty);
564 if (ix >= 0)
565 goto out_unlock;
566
567 if (empty < 0) {
568 ret = -ENOSPC;
569 goto out_unlock;
570 }
571 attr->device = ib_dev;
572 attr->index = empty;
573 attr->port_num = port;
574 attr->gid = *gid;
575 ret = add_modify_gid(table, attr);
576 if (!ret)
577 dispatch_gid_change_event(ib_dev, port);
578
579out_unlock:
580 mutex_unlock(&table->lock);
581 if (ret)
582 pr_warn("%s: unable to add gid %pI6 error=%d\n",
583 __func__, gid->raw, ret);
584 return ret;
585}
586
587int ib_cache_gid_add(struct ib_device *ib_dev, u8 port,
588 union ib_gid *gid, struct ib_gid_attr *attr)
589{
590 unsigned long mask = GID_ATTR_FIND_MASK_GID |
591 GID_ATTR_FIND_MASK_GID_TYPE |
592 GID_ATTR_FIND_MASK_NETDEV;
593
594 return __ib_cache_gid_add(ib_dev, port, gid, attr, mask, false);
595}
596
597static int
598_ib_cache_gid_del(struct ib_device *ib_dev, u8 port,
599 union ib_gid *gid, struct ib_gid_attr *attr,
600 unsigned long mask, bool default_gid)
601{
602 struct ib_gid_table *table;
603 int ret = 0;
604 int ix;
605
606 table = rdma_gid_table(ib_dev, port);
607
608 mutex_lock(&table->lock);
609
610 ix = find_gid(table, gid, attr, default_gid, mask, NULL);
611 if (ix < 0) {
612 ret = -EINVAL;
613 goto out_unlock;
614 }
615
616 del_gid(ib_dev, port, table, ix);
617 dispatch_gid_change_event(ib_dev, port);
618
619out_unlock:
620 mutex_unlock(&table->lock);
621 if (ret)
622 pr_debug("%s: can't delete gid %pI6 error=%d\n",
623 __func__, gid->raw, ret);
624 return ret;
625}
626
627int ib_cache_gid_del(struct ib_device *ib_dev, u8 port,
628 union ib_gid *gid, struct ib_gid_attr *attr)
629{
630 unsigned long mask = GID_ATTR_FIND_MASK_GID |
631 GID_ATTR_FIND_MASK_GID_TYPE |
632 GID_ATTR_FIND_MASK_DEFAULT |
633 GID_ATTR_FIND_MASK_NETDEV;
634
635 return _ib_cache_gid_del(ib_dev, port, gid, attr, mask, false);
636}
637
638int ib_cache_gid_del_all_netdev_gids(struct ib_device *ib_dev, u8 port,
639 struct net_device *ndev)
640{
641 struct ib_gid_table *table;
642 int ix;
643 bool deleted = false;
644
645 table = rdma_gid_table(ib_dev, port);
646
647 mutex_lock(&table->lock);
648
649 for (ix = 0; ix < table->sz; ix++) {
650 if (is_gid_entry_valid(table->data_vec[ix]) &&
651 table->data_vec[ix]->attr.ndev == ndev) {
652 del_gid(ib_dev, port, table, ix);
653 deleted = true;
654 }
655 }
656
657 mutex_unlock(&table->lock);
658
659 if (deleted)
660 dispatch_gid_change_event(ib_dev, port);
661
662 return 0;
663}
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680const struct ib_gid_attr *
681rdma_find_gid_by_port(struct ib_device *ib_dev,
682 const union ib_gid *gid,
683 enum ib_gid_type gid_type,
684 u8 port, struct net_device *ndev)
685{
686 int local_index;
687 struct ib_gid_table *table;
688 unsigned long mask = GID_ATTR_FIND_MASK_GID |
689 GID_ATTR_FIND_MASK_GID_TYPE;
690 struct ib_gid_attr val = {.ndev = ndev, .gid_type = gid_type};
691 const struct ib_gid_attr *attr;
692 unsigned long flags;
693
694 if (!rdma_is_port_valid(ib_dev, port))
695 return ERR_PTR(-ENOENT);
696
697 table = rdma_gid_table(ib_dev, port);
698
699 if (ndev)
700 mask |= GID_ATTR_FIND_MASK_NETDEV;
701
702 read_lock_irqsave(&table->rwlock, flags);
703 local_index = find_gid(table, gid, &val, false, mask, NULL);
704 if (local_index >= 0) {
705 get_gid_entry(table->data_vec[local_index]);
706 attr = &table->data_vec[local_index]->attr;
707 read_unlock_irqrestore(&table->rwlock, flags);
708 return attr;
709 }
710
711 read_unlock_irqrestore(&table->rwlock, flags);
712 return ERR_PTR(-ENOENT);
713}
714EXPORT_SYMBOL(rdma_find_gid_by_port);
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733const struct ib_gid_attr *rdma_find_gid_by_filter(
734 struct ib_device *ib_dev, const union ib_gid *gid, u8 port,
735 bool (*filter)(const union ib_gid *gid, const struct ib_gid_attr *,
736 void *),
737 void *context)
738{
739 const struct ib_gid_attr *res = ERR_PTR(-ENOENT);
740 struct ib_gid_table *table;
741 unsigned long flags;
742 unsigned int i;
743
744 if (!rdma_is_port_valid(ib_dev, port))
745 return ERR_PTR(-EINVAL);
746
747 table = rdma_gid_table(ib_dev, port);
748
749 read_lock_irqsave(&table->rwlock, flags);
750 for (i = 0; i < table->sz; i++) {
751 struct ib_gid_table_entry *entry = table->data_vec[i];
752
753 if (!is_gid_entry_valid(entry))
754 continue;
755
756 if (memcmp(gid, &entry->attr.gid, sizeof(*gid)))
757 continue;
758
759 if (filter(gid, &entry->attr, context)) {
760 get_gid_entry(entry);
761 res = &entry->attr;
762 break;
763 }
764 }
765 read_unlock_irqrestore(&table->rwlock, flags);
766 return res;
767}
768
769static struct ib_gid_table *alloc_gid_table(int sz)
770{
771 struct ib_gid_table *table = kzalloc(sizeof(*table), GFP_KERNEL);
772
773 if (!table)
774 return NULL;
775
776 table->data_vec = kcalloc(sz, sizeof(*table->data_vec), GFP_KERNEL);
777 if (!table->data_vec)
778 goto err_free_table;
779
780 mutex_init(&table->lock);
781
782 table->sz = sz;
783 rwlock_init(&table->rwlock);
784 return table;
785
786err_free_table:
787 kfree(table);
788 return NULL;
789}
790
791static void release_gid_table(struct ib_device *device,
792 struct ib_gid_table *table)
793{
794 bool leak = false;
795 int i;
796
797 if (!table)
798 return;
799
800 for (i = 0; i < table->sz; i++) {
801 if (is_gid_entry_free(table->data_vec[i]))
802 continue;
803 if (kref_read(&table->data_vec[i]->kref) > 1) {
804 dev_err(&device->dev,
805 "GID entry ref leak for index %d ref=%d\n", i,
806 kref_read(&table->data_vec[i]->kref));
807 leak = true;
808 }
809 }
810 if (leak)
811 return;
812
813 mutex_destroy(&table->lock);
814 kfree(table->data_vec);
815 kfree(table);
816}
817
818static void cleanup_gid_table_port(struct ib_device *ib_dev, u8 port,
819 struct ib_gid_table *table)
820{
821 int i;
822 bool deleted = false;
823
824 if (!table)
825 return;
826
827 mutex_lock(&table->lock);
828 for (i = 0; i < table->sz; ++i) {
829 if (is_gid_entry_valid(table->data_vec[i])) {
830 del_gid(ib_dev, port, table, i);
831 deleted = true;
832 }
833 }
834 mutex_unlock(&table->lock);
835
836 if (deleted)
837 dispatch_gid_change_event(ib_dev, port);
838}
839
840void ib_cache_gid_set_default_gid(struct ib_device *ib_dev, u8 port,
841 struct net_device *ndev,
842 unsigned long gid_type_mask,
843 enum ib_cache_gid_default_mode mode)
844{
845 union ib_gid gid = { };
846 struct ib_gid_attr gid_attr;
847 unsigned int gid_type;
848 unsigned long mask;
849
850 mask = GID_ATTR_FIND_MASK_GID_TYPE |
851 GID_ATTR_FIND_MASK_DEFAULT |
852 GID_ATTR_FIND_MASK_NETDEV;
853 memset(&gid_attr, 0, sizeof(gid_attr));
854 gid_attr.ndev = ndev;
855
856 for (gid_type = 0; gid_type < IB_GID_TYPE_SIZE; ++gid_type) {
857 if (1UL << gid_type & ~gid_type_mask)
858 continue;
859
860 gid_attr.gid_type = gid_type;
861
862 if (mode == IB_CACHE_GID_DEFAULT_MODE_SET) {
863 make_default_gid(ndev, &gid);
864 __ib_cache_gid_add(ib_dev, port, &gid,
865 &gid_attr, mask, true);
866 } else if (mode == IB_CACHE_GID_DEFAULT_MODE_DELETE) {
867 _ib_cache_gid_del(ib_dev, port, &gid,
868 &gid_attr, mask, true);
869 }
870 }
871}
872
873static void gid_table_reserve_default(struct ib_device *ib_dev, u8 port,
874 struct ib_gid_table *table)
875{
876 unsigned int i;
877 unsigned long roce_gid_type_mask;
878 unsigned int num_default_gids;
879
880 roce_gid_type_mask = roce_gid_type_mask_support(ib_dev, port);
881 num_default_gids = hweight_long(roce_gid_type_mask);
882
883 for (i = 0; i < num_default_gids && i < table->sz; i++)
884 table->default_gid_indices |= BIT(i);
885}
886
887
888static void gid_table_release_one(struct ib_device *ib_dev)
889{
890 unsigned int p;
891
892 rdma_for_each_port (ib_dev, p) {
893 release_gid_table(ib_dev, ib_dev->port_data[p].cache.gid);
894 ib_dev->port_data[p].cache.gid = NULL;
895 }
896}
897
898static int _gid_table_setup_one(struct ib_device *ib_dev)
899{
900 struct ib_gid_table *table;
901 unsigned int rdma_port;
902
903 rdma_for_each_port (ib_dev, rdma_port) {
904 table = alloc_gid_table(
905 ib_dev->port_data[rdma_port].immutable.gid_tbl_len);
906 if (!table)
907 goto rollback_table_setup;
908
909 gid_table_reserve_default(ib_dev, rdma_port, table);
910 ib_dev->port_data[rdma_port].cache.gid = table;
911 }
912 return 0;
913
914rollback_table_setup:
915 gid_table_release_one(ib_dev);
916 return -ENOMEM;
917}
918
919static void gid_table_cleanup_one(struct ib_device *ib_dev)
920{
921 unsigned int p;
922
923 rdma_for_each_port (ib_dev, p)
924 cleanup_gid_table_port(ib_dev, p,
925 ib_dev->port_data[p].cache.gid);
926}
927
928static int gid_table_setup_one(struct ib_device *ib_dev)
929{
930 int err;
931
932 err = _gid_table_setup_one(ib_dev);
933
934 if (err)
935 return err;
936
937 rdma_roce_rescan_device(ib_dev);
938
939 return err;
940}
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956int rdma_query_gid(struct ib_device *device, u8 port_num,
957 int index, union ib_gid *gid)
958{
959 struct ib_gid_table *table;
960 unsigned long flags;
961 int res = -EINVAL;
962
963 if (!rdma_is_port_valid(device, port_num))
964 return -EINVAL;
965
966 table = rdma_gid_table(device, port_num);
967 read_lock_irqsave(&table->rwlock, flags);
968
969 if (index < 0 || index >= table->sz ||
970 !is_gid_entry_valid(table->data_vec[index]))
971 goto done;
972
973 memcpy(gid, &table->data_vec[index]->attr.gid, sizeof(*gid));
974 res = 0;
975
976done:
977 read_unlock_irqrestore(&table->rwlock, flags);
978 return res;
979}
980EXPORT_SYMBOL(rdma_query_gid);
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995const struct ib_gid_attr *rdma_find_gid(struct ib_device *device,
996 const union ib_gid *gid,
997 enum ib_gid_type gid_type,
998 struct net_device *ndev)
999{
1000 unsigned long mask = GID_ATTR_FIND_MASK_GID |
1001 GID_ATTR_FIND_MASK_GID_TYPE;
1002 struct ib_gid_attr gid_attr_val = {.ndev = ndev, .gid_type = gid_type};
1003 unsigned int p;
1004
1005 if (ndev)
1006 mask |= GID_ATTR_FIND_MASK_NETDEV;
1007
1008 rdma_for_each_port(device, p) {
1009 struct ib_gid_table *table;
1010 unsigned long flags;
1011 int index;
1012
1013 table = device->port_data[p].cache.gid;
1014 read_lock_irqsave(&table->rwlock, flags);
1015 index = find_gid(table, gid, &gid_attr_val, false, mask, NULL);
1016 if (index >= 0) {
1017 const struct ib_gid_attr *attr;
1018
1019 get_gid_entry(table->data_vec[index]);
1020 attr = &table->data_vec[index]->attr;
1021 read_unlock_irqrestore(&table->rwlock, flags);
1022 return attr;
1023 }
1024 read_unlock_irqrestore(&table->rwlock, flags);
1025 }
1026
1027 return ERR_PTR(-ENOENT);
1028}
1029EXPORT_SYMBOL(rdma_find_gid);
1030
1031int ib_get_cached_pkey(struct ib_device *device,
1032 u8 port_num,
1033 int index,
1034 u16 *pkey)
1035{
1036 struct ib_pkey_cache *cache;
1037 unsigned long flags;
1038 int ret = 0;
1039
1040 if (!rdma_is_port_valid(device, port_num))
1041 return -EINVAL;
1042
1043 read_lock_irqsave(&device->cache.lock, flags);
1044
1045 cache = device->port_data[port_num].cache.pkey;
1046
1047 if (index < 0 || index >= cache->table_len)
1048 ret = -EINVAL;
1049 else
1050 *pkey = cache->table[index];
1051
1052 read_unlock_irqrestore(&device->cache.lock, flags);
1053
1054 return ret;
1055}
1056EXPORT_SYMBOL(ib_get_cached_pkey);
1057
1058int ib_get_cached_subnet_prefix(struct ib_device *device,
1059 u8 port_num,
1060 u64 *sn_pfx)
1061{
1062 unsigned long flags;
1063
1064 if (!rdma_is_port_valid(device, port_num))
1065 return -EINVAL;
1066
1067 read_lock_irqsave(&device->cache.lock, flags);
1068 *sn_pfx = device->port_data[port_num].cache.subnet_prefix;
1069 read_unlock_irqrestore(&device->cache.lock, flags);
1070
1071 return 0;
1072}
1073EXPORT_SYMBOL(ib_get_cached_subnet_prefix);
1074
1075int ib_find_cached_pkey(struct ib_device *device,
1076 u8 port_num,
1077 u16 pkey,
1078 u16 *index)
1079{
1080 struct ib_pkey_cache *cache;
1081 unsigned long flags;
1082 int i;
1083 int ret = -ENOENT;
1084 int partial_ix = -1;
1085
1086 if (!rdma_is_port_valid(device, port_num))
1087 return -EINVAL;
1088
1089 read_lock_irqsave(&device->cache.lock, flags);
1090
1091 cache = device->port_data[port_num].cache.pkey;
1092
1093 *index = -1;
1094
1095 for (i = 0; i < cache->table_len; ++i)
1096 if ((cache->table[i] & 0x7fff) == (pkey & 0x7fff)) {
1097 if (cache->table[i] & 0x8000) {
1098 *index = i;
1099 ret = 0;
1100 break;
1101 } else
1102 partial_ix = i;
1103 }
1104
1105 if (ret && partial_ix >= 0) {
1106 *index = partial_ix;
1107 ret = 0;
1108 }
1109
1110 read_unlock_irqrestore(&device->cache.lock, flags);
1111
1112 return ret;
1113}
1114EXPORT_SYMBOL(ib_find_cached_pkey);
1115
1116int ib_find_exact_cached_pkey(struct ib_device *device,
1117 u8 port_num,
1118 u16 pkey,
1119 u16 *index)
1120{
1121 struct ib_pkey_cache *cache;
1122 unsigned long flags;
1123 int i;
1124 int ret = -ENOENT;
1125
1126 if (!rdma_is_port_valid(device, port_num))
1127 return -EINVAL;
1128
1129 read_lock_irqsave(&device->cache.lock, flags);
1130
1131 cache = device->port_data[port_num].cache.pkey;
1132
1133 *index = -1;
1134
1135 for (i = 0; i < cache->table_len; ++i)
1136 if (cache->table[i] == pkey) {
1137 *index = i;
1138 ret = 0;
1139 break;
1140 }
1141
1142 read_unlock_irqrestore(&device->cache.lock, flags);
1143
1144 return ret;
1145}
1146EXPORT_SYMBOL(ib_find_exact_cached_pkey);
1147
1148int ib_get_cached_lmc(struct ib_device *device,
1149 u8 port_num,
1150 u8 *lmc)
1151{
1152 unsigned long flags;
1153 int ret = 0;
1154
1155 if (!rdma_is_port_valid(device, port_num))
1156 return -EINVAL;
1157
1158 read_lock_irqsave(&device->cache.lock, flags);
1159 *lmc = device->port_data[port_num].cache.lmc;
1160 read_unlock_irqrestore(&device->cache.lock, flags);
1161
1162 return ret;
1163}
1164EXPORT_SYMBOL(ib_get_cached_lmc);
1165
1166int ib_get_cached_port_state(struct ib_device *device,
1167 u8 port_num,
1168 enum ib_port_state *port_state)
1169{
1170 unsigned long flags;
1171 int ret = 0;
1172
1173 if (!rdma_is_port_valid(device, port_num))
1174 return -EINVAL;
1175
1176 read_lock_irqsave(&device->cache.lock, flags);
1177 *port_state = device->port_data[port_num].cache.port_state;
1178 read_unlock_irqrestore(&device->cache.lock, flags);
1179
1180 return ret;
1181}
1182EXPORT_SYMBOL(ib_get_cached_port_state);
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200const struct ib_gid_attr *
1201rdma_get_gid_attr(struct ib_device *device, u8 port_num, int index)
1202{
1203 const struct ib_gid_attr *attr = ERR_PTR(-EINVAL);
1204 struct ib_gid_table *table;
1205 unsigned long flags;
1206
1207 if (!rdma_is_port_valid(device, port_num))
1208 return ERR_PTR(-EINVAL);
1209
1210 table = rdma_gid_table(device, port_num);
1211 if (index < 0 || index >= table->sz)
1212 return ERR_PTR(-EINVAL);
1213
1214 read_lock_irqsave(&table->rwlock, flags);
1215 if (!is_gid_entry_valid(table->data_vec[index]))
1216 goto done;
1217
1218 get_gid_entry(table->data_vec[index]);
1219 attr = &table->data_vec[index]->attr;
1220done:
1221 read_unlock_irqrestore(&table->rwlock, flags);
1222 return attr;
1223}
1224EXPORT_SYMBOL(rdma_get_gid_attr);
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237void rdma_put_gid_attr(const struct ib_gid_attr *attr)
1238{
1239 struct ib_gid_table_entry *entry =
1240 container_of(attr, struct ib_gid_table_entry, attr);
1241
1242 put_gid_entry(entry);
1243}
1244EXPORT_SYMBOL(rdma_put_gid_attr);
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256void rdma_hold_gid_attr(const struct ib_gid_attr *attr)
1257{
1258 struct ib_gid_table_entry *entry =
1259 container_of(attr, struct ib_gid_table_entry, attr);
1260
1261 get_gid_entry(entry);
1262}
1263EXPORT_SYMBOL(rdma_hold_gid_attr);
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277struct net_device *rdma_read_gid_attr_ndev_rcu(const struct ib_gid_attr *attr)
1278{
1279 struct ib_gid_table_entry *entry =
1280 container_of(attr, struct ib_gid_table_entry, attr);
1281 struct ib_device *device = entry->attr.device;
1282 struct net_device *ndev = ERR_PTR(-ENODEV);
1283 u8 port_num = entry->attr.port_num;
1284 struct ib_gid_table *table;
1285 unsigned long flags;
1286 bool valid;
1287
1288 table = rdma_gid_table(device, port_num);
1289
1290 read_lock_irqsave(&table->rwlock, flags);
1291 valid = is_gid_entry_valid(table->data_vec[attr->index]);
1292 if (valid) {
1293 ndev = rcu_dereference(attr->ndev);
1294 if (!ndev ||
1295 (ndev && ((READ_ONCE(ndev->flags) & IFF_UP) == 0)))
1296 ndev = ERR_PTR(-ENODEV);
1297 }
1298 read_unlock_irqrestore(&table->rwlock, flags);
1299 return ndev;
1300}
1301EXPORT_SYMBOL(rdma_read_gid_attr_ndev_rcu);
1302
1303static int get_lower_dev_vlan(struct net_device *lower_dev, void *data)
1304{
1305 u16 *vlan_id = data;
1306
1307 if (is_vlan_dev(lower_dev))
1308 *vlan_id = vlan_dev_vlan_id(lower_dev);
1309
1310
1311
1312
1313 return 1;
1314}
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328int rdma_read_gid_l2_fields(const struct ib_gid_attr *attr,
1329 u16 *vlan_id, u8 *smac)
1330{
1331 struct net_device *ndev;
1332
1333 rcu_read_lock();
1334 ndev = rcu_dereference(attr->ndev);
1335 if (!ndev) {
1336 rcu_read_unlock();
1337 return -ENODEV;
1338 }
1339 if (smac)
1340 ether_addr_copy(smac, ndev->dev_addr);
1341 if (vlan_id) {
1342 *vlan_id = 0xffff;
1343 if (is_vlan_dev(ndev)) {
1344 *vlan_id = vlan_dev_vlan_id(ndev);
1345 } else {
1346
1347
1348
1349
1350 netdev_walk_all_lower_dev_rcu(attr->ndev,
1351 get_lower_dev_vlan, vlan_id);
1352 }
1353 }
1354 rcu_read_unlock();
1355 return 0;
1356}
1357EXPORT_SYMBOL(rdma_read_gid_l2_fields);
1358
1359static int config_non_roce_gid_cache(struct ib_device *device,
1360 u8 port, int gid_tbl_len)
1361{
1362 struct ib_gid_attr gid_attr = {};
1363 struct ib_gid_table *table;
1364 int ret = 0;
1365 int i;
1366
1367 gid_attr.device = device;
1368 gid_attr.port_num = port;
1369 table = rdma_gid_table(device, port);
1370
1371 mutex_lock(&table->lock);
1372 for (i = 0; i < gid_tbl_len; ++i) {
1373 if (!device->ops.query_gid)
1374 continue;
1375 ret = device->ops.query_gid(device, port, i, &gid_attr.gid);
1376 if (ret) {
1377 dev_warn(&device->dev,
1378 "query_gid failed (%d) for index %d\n", ret,
1379 i);
1380 goto err;
1381 }
1382 gid_attr.index = i;
1383 add_modify_gid(table, &gid_attr);
1384 }
1385err:
1386 mutex_unlock(&table->lock);
1387 return ret;
1388}
1389
1390static void ib_cache_update(struct ib_device *device,
1391 u8 port,
1392 bool enforce_security)
1393{
1394 struct ib_port_attr *tprops = NULL;
1395 struct ib_pkey_cache *pkey_cache = NULL, *old_pkey_cache;
1396 int i;
1397 int ret;
1398
1399 if (!rdma_is_port_valid(device, port))
1400 return;
1401
1402 tprops = kmalloc(sizeof *tprops, GFP_KERNEL);
1403 if (!tprops)
1404 return;
1405
1406 ret = ib_query_port(device, port, tprops);
1407 if (ret) {
1408 dev_warn(&device->dev, "ib_query_port failed (%d)\n", ret);
1409 goto err;
1410 }
1411
1412 if (!rdma_protocol_roce(device, port)) {
1413 ret = config_non_roce_gid_cache(device, port,
1414 tprops->gid_tbl_len);
1415 if (ret)
1416 goto err;
1417 }
1418
1419 pkey_cache = kmalloc(struct_size(pkey_cache, table,
1420 tprops->pkey_tbl_len),
1421 GFP_KERNEL);
1422 if (!pkey_cache)
1423 goto err;
1424
1425 pkey_cache->table_len = tprops->pkey_tbl_len;
1426
1427 for (i = 0; i < pkey_cache->table_len; ++i) {
1428 ret = ib_query_pkey(device, port, i, pkey_cache->table + i);
1429 if (ret) {
1430 dev_warn(&device->dev,
1431 "ib_query_pkey failed (%d) for index %d\n",
1432 ret, i);
1433 goto err;
1434 }
1435 }
1436
1437 write_lock_irq(&device->cache.lock);
1438
1439 old_pkey_cache = device->port_data[port].cache.pkey;
1440
1441 device->port_data[port].cache.pkey = pkey_cache;
1442 device->port_data[port].cache.lmc = tprops->lmc;
1443 device->port_data[port].cache.port_state = tprops->state;
1444
1445 device->port_data[port].cache.subnet_prefix = tprops->subnet_prefix;
1446 write_unlock_irq(&device->cache.lock);
1447
1448 if (enforce_security)
1449 ib_security_cache_change(device,
1450 port,
1451 tprops->subnet_prefix);
1452
1453 kfree(old_pkey_cache);
1454 kfree(tprops);
1455 return;
1456
1457err:
1458 kfree(pkey_cache);
1459 kfree(tprops);
1460}
1461
1462static void ib_cache_task(struct work_struct *_work)
1463{
1464 struct ib_update_work *work =
1465 container_of(_work, struct ib_update_work, work);
1466
1467 ib_cache_update(work->device,
1468 work->port_num,
1469 work->enforce_security);
1470 kfree(work);
1471}
1472
1473static void ib_cache_event(struct ib_event_handler *handler,
1474 struct ib_event *event)
1475{
1476 struct ib_update_work *work;
1477
1478 if (event->event == IB_EVENT_PORT_ERR ||
1479 event->event == IB_EVENT_PORT_ACTIVE ||
1480 event->event == IB_EVENT_LID_CHANGE ||
1481 event->event == IB_EVENT_PKEY_CHANGE ||
1482 event->event == IB_EVENT_CLIENT_REREGISTER ||
1483 event->event == IB_EVENT_GID_CHANGE) {
1484 work = kmalloc(sizeof *work, GFP_ATOMIC);
1485 if (work) {
1486 INIT_WORK(&work->work, ib_cache_task);
1487 work->device = event->device;
1488 work->port_num = event->element.port_num;
1489 if (event->event == IB_EVENT_PKEY_CHANGE ||
1490 event->event == IB_EVENT_GID_CHANGE)
1491 work->enforce_security = true;
1492 else
1493 work->enforce_security = false;
1494
1495 queue_work(ib_wq, &work->work);
1496 }
1497 }
1498}
1499
1500int ib_cache_setup_one(struct ib_device *device)
1501{
1502 unsigned int p;
1503 int err;
1504
1505 rwlock_init(&device->cache.lock);
1506
1507 err = gid_table_setup_one(device);
1508 if (err)
1509 return err;
1510
1511 rdma_for_each_port (device, p)
1512 ib_cache_update(device, p, true);
1513
1514 INIT_IB_EVENT_HANDLER(&device->cache.event_handler,
1515 device, ib_cache_event);
1516 ib_register_event_handler(&device->cache.event_handler);
1517 return 0;
1518}
1519
1520void ib_cache_release_one(struct ib_device *device)
1521{
1522 unsigned int p;
1523
1524
1525
1526
1527
1528
1529
1530 rdma_for_each_port (device, p)
1531 kfree(device->port_data[p].cache.pkey);
1532
1533 gid_table_release_one(device);
1534}
1535
1536void ib_cache_cleanup_one(struct ib_device *device)
1537{
1538
1539
1540
1541
1542
1543
1544
1545 ib_unregister_event_handler(&device->cache.event_handler);
1546 flush_workqueue(ib_wq);
1547 gid_table_cleanup_one(device);
1548
1549
1550
1551
1552 flush_workqueue(ib_wq);
1553}
1554