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 kfree(table->data_vec);
814 kfree(table);
815}
816
817static void cleanup_gid_table_port(struct ib_device *ib_dev, u8 port,
818 struct ib_gid_table *table)
819{
820 int i;
821 bool deleted = false;
822
823 if (!table)
824 return;
825
826 mutex_lock(&table->lock);
827 for (i = 0; i < table->sz; ++i) {
828 if (is_gid_entry_valid(table->data_vec[i])) {
829 del_gid(ib_dev, port, table, i);
830 deleted = true;
831 }
832 }
833 mutex_unlock(&table->lock);
834
835 if (deleted)
836 dispatch_gid_change_event(ib_dev, port);
837}
838
839void ib_cache_gid_set_default_gid(struct ib_device *ib_dev, u8 port,
840 struct net_device *ndev,
841 unsigned long gid_type_mask,
842 enum ib_cache_gid_default_mode mode)
843{
844 union ib_gid gid = { };
845 struct ib_gid_attr gid_attr;
846 unsigned int gid_type;
847 unsigned long mask;
848
849 mask = GID_ATTR_FIND_MASK_GID_TYPE |
850 GID_ATTR_FIND_MASK_DEFAULT |
851 GID_ATTR_FIND_MASK_NETDEV;
852 memset(&gid_attr, 0, sizeof(gid_attr));
853 gid_attr.ndev = ndev;
854
855 for (gid_type = 0; gid_type < IB_GID_TYPE_SIZE; ++gid_type) {
856 if (1UL << gid_type & ~gid_type_mask)
857 continue;
858
859 gid_attr.gid_type = gid_type;
860
861 if (mode == IB_CACHE_GID_DEFAULT_MODE_SET) {
862 make_default_gid(ndev, &gid);
863 __ib_cache_gid_add(ib_dev, port, &gid,
864 &gid_attr, mask, true);
865 } else if (mode == IB_CACHE_GID_DEFAULT_MODE_DELETE) {
866 _ib_cache_gid_del(ib_dev, port, &gid,
867 &gid_attr, mask, true);
868 }
869 }
870}
871
872static void gid_table_reserve_default(struct ib_device *ib_dev, u8 port,
873 struct ib_gid_table *table)
874{
875 unsigned int i;
876 unsigned long roce_gid_type_mask;
877 unsigned int num_default_gids;
878
879 roce_gid_type_mask = roce_gid_type_mask_support(ib_dev, port);
880 num_default_gids = hweight_long(roce_gid_type_mask);
881
882 for (i = 0; i < num_default_gids && i < table->sz; i++)
883 table->default_gid_indices |= BIT(i);
884}
885
886
887static void gid_table_release_one(struct ib_device *ib_dev)
888{
889 unsigned int p;
890
891 rdma_for_each_port (ib_dev, p) {
892 release_gid_table(ib_dev, ib_dev->port_data[p].cache.gid);
893 ib_dev->port_data[p].cache.gid = NULL;
894 }
895}
896
897static int _gid_table_setup_one(struct ib_device *ib_dev)
898{
899 struct ib_gid_table *table;
900 unsigned int rdma_port;
901
902 rdma_for_each_port (ib_dev, rdma_port) {
903 table = alloc_gid_table(
904 ib_dev->port_data[rdma_port].immutable.gid_tbl_len);
905 if (!table)
906 goto rollback_table_setup;
907
908 gid_table_reserve_default(ib_dev, rdma_port, table);
909 ib_dev->port_data[rdma_port].cache.gid = table;
910 }
911 return 0;
912
913rollback_table_setup:
914 gid_table_release_one(ib_dev);
915 return -ENOMEM;
916}
917
918static void gid_table_cleanup_one(struct ib_device *ib_dev)
919{
920 unsigned int p;
921
922 rdma_for_each_port (ib_dev, p)
923 cleanup_gid_table_port(ib_dev, p,
924 ib_dev->port_data[p].cache.gid);
925}
926
927static int gid_table_setup_one(struct ib_device *ib_dev)
928{
929 int err;
930
931 err = _gid_table_setup_one(ib_dev);
932
933 if (err)
934 return err;
935
936 rdma_roce_rescan_device(ib_dev);
937
938 return err;
939}
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955int rdma_query_gid(struct ib_device *device, u8 port_num,
956 int index, union ib_gid *gid)
957{
958 struct ib_gid_table *table;
959 unsigned long flags;
960 int res = -EINVAL;
961
962 if (!rdma_is_port_valid(device, port_num))
963 return -EINVAL;
964
965 table = rdma_gid_table(device, port_num);
966 read_lock_irqsave(&table->rwlock, flags);
967
968 if (index < 0 || index >= table->sz ||
969 !is_gid_entry_valid(table->data_vec[index]))
970 goto done;
971
972 memcpy(gid, &table->data_vec[index]->attr.gid, sizeof(*gid));
973 res = 0;
974
975done:
976 read_unlock_irqrestore(&table->rwlock, flags);
977 return res;
978}
979EXPORT_SYMBOL(rdma_query_gid);
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994const struct ib_gid_attr *rdma_find_gid(struct ib_device *device,
995 const union ib_gid *gid,
996 enum ib_gid_type gid_type,
997 struct net_device *ndev)
998{
999 unsigned long mask = GID_ATTR_FIND_MASK_GID |
1000 GID_ATTR_FIND_MASK_GID_TYPE;
1001 struct ib_gid_attr gid_attr_val = {.ndev = ndev, .gid_type = gid_type};
1002 unsigned int p;
1003
1004 if (ndev)
1005 mask |= GID_ATTR_FIND_MASK_NETDEV;
1006
1007 rdma_for_each_port(device, p) {
1008 struct ib_gid_table *table;
1009 unsigned long flags;
1010 int index;
1011
1012 table = device->port_data[p].cache.gid;
1013 read_lock_irqsave(&table->rwlock, flags);
1014 index = find_gid(table, gid, &gid_attr_val, false, mask, NULL);
1015 if (index >= 0) {
1016 const struct ib_gid_attr *attr;
1017
1018 get_gid_entry(table->data_vec[index]);
1019 attr = &table->data_vec[index]->attr;
1020 read_unlock_irqrestore(&table->rwlock, flags);
1021 return attr;
1022 }
1023 read_unlock_irqrestore(&table->rwlock, flags);
1024 }
1025
1026 return ERR_PTR(-ENOENT);
1027}
1028EXPORT_SYMBOL(rdma_find_gid);
1029
1030int ib_get_cached_pkey(struct ib_device *device,
1031 u8 port_num,
1032 int index,
1033 u16 *pkey)
1034{
1035 struct ib_pkey_cache *cache;
1036 unsigned long flags;
1037 int ret = 0;
1038
1039 if (!rdma_is_port_valid(device, port_num))
1040 return -EINVAL;
1041
1042 read_lock_irqsave(&device->cache.lock, flags);
1043
1044 cache = device->port_data[port_num].cache.pkey;
1045
1046 if (index < 0 || index >= cache->table_len)
1047 ret = -EINVAL;
1048 else
1049 *pkey = cache->table[index];
1050
1051 read_unlock_irqrestore(&device->cache.lock, flags);
1052
1053 return ret;
1054}
1055EXPORT_SYMBOL(ib_get_cached_pkey);
1056
1057int ib_get_cached_subnet_prefix(struct ib_device *device,
1058 u8 port_num,
1059 u64 *sn_pfx)
1060{
1061 unsigned long flags;
1062
1063 if (!rdma_is_port_valid(device, port_num))
1064 return -EINVAL;
1065
1066 read_lock_irqsave(&device->cache.lock, flags);
1067 *sn_pfx = device->port_data[port_num].cache.subnet_prefix;
1068 read_unlock_irqrestore(&device->cache.lock, flags);
1069
1070 return 0;
1071}
1072EXPORT_SYMBOL(ib_get_cached_subnet_prefix);
1073
1074int ib_find_cached_pkey(struct ib_device *device,
1075 u8 port_num,
1076 u16 pkey,
1077 u16 *index)
1078{
1079 struct ib_pkey_cache *cache;
1080 unsigned long flags;
1081 int i;
1082 int ret = -ENOENT;
1083 int partial_ix = -1;
1084
1085 if (!rdma_is_port_valid(device, port_num))
1086 return -EINVAL;
1087
1088 read_lock_irqsave(&device->cache.lock, flags);
1089
1090 cache = device->port_data[port_num].cache.pkey;
1091
1092 *index = -1;
1093
1094 for (i = 0; i < cache->table_len; ++i)
1095 if ((cache->table[i] & 0x7fff) == (pkey & 0x7fff)) {
1096 if (cache->table[i] & 0x8000) {
1097 *index = i;
1098 ret = 0;
1099 break;
1100 } else
1101 partial_ix = i;
1102 }
1103
1104 if (ret && partial_ix >= 0) {
1105 *index = partial_ix;
1106 ret = 0;
1107 }
1108
1109 read_unlock_irqrestore(&device->cache.lock, flags);
1110
1111 return ret;
1112}
1113EXPORT_SYMBOL(ib_find_cached_pkey);
1114
1115int ib_find_exact_cached_pkey(struct ib_device *device,
1116 u8 port_num,
1117 u16 pkey,
1118 u16 *index)
1119{
1120 struct ib_pkey_cache *cache;
1121 unsigned long flags;
1122 int i;
1123 int ret = -ENOENT;
1124
1125 if (!rdma_is_port_valid(device, port_num))
1126 return -EINVAL;
1127
1128 read_lock_irqsave(&device->cache.lock, flags);
1129
1130 cache = device->port_data[port_num].cache.pkey;
1131
1132 *index = -1;
1133
1134 for (i = 0; i < cache->table_len; ++i)
1135 if (cache->table[i] == pkey) {
1136 *index = i;
1137 ret = 0;
1138 break;
1139 }
1140
1141 read_unlock_irqrestore(&device->cache.lock, flags);
1142
1143 return ret;
1144}
1145EXPORT_SYMBOL(ib_find_exact_cached_pkey);
1146
1147int ib_get_cached_lmc(struct ib_device *device,
1148 u8 port_num,
1149 u8 *lmc)
1150{
1151 unsigned long flags;
1152 int ret = 0;
1153
1154 if (!rdma_is_port_valid(device, port_num))
1155 return -EINVAL;
1156
1157 read_lock_irqsave(&device->cache.lock, flags);
1158 *lmc = device->port_data[port_num].cache.lmc;
1159 read_unlock_irqrestore(&device->cache.lock, flags);
1160
1161 return ret;
1162}
1163EXPORT_SYMBOL(ib_get_cached_lmc);
1164
1165int ib_get_cached_port_state(struct ib_device *device,
1166 u8 port_num,
1167 enum ib_port_state *port_state)
1168{
1169 unsigned long flags;
1170 int ret = 0;
1171
1172 if (!rdma_is_port_valid(device, port_num))
1173 return -EINVAL;
1174
1175 read_lock_irqsave(&device->cache.lock, flags);
1176 *port_state = device->port_data[port_num].cache.port_state;
1177 read_unlock_irqrestore(&device->cache.lock, flags);
1178
1179 return ret;
1180}
1181EXPORT_SYMBOL(ib_get_cached_port_state);
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199const struct ib_gid_attr *
1200rdma_get_gid_attr(struct ib_device *device, u8 port_num, int index)
1201{
1202 const struct ib_gid_attr *attr = ERR_PTR(-EINVAL);
1203 struct ib_gid_table *table;
1204 unsigned long flags;
1205
1206 if (!rdma_is_port_valid(device, port_num))
1207 return ERR_PTR(-EINVAL);
1208
1209 table = rdma_gid_table(device, port_num);
1210 if (index < 0 || index >= table->sz)
1211 return ERR_PTR(-EINVAL);
1212
1213 read_lock_irqsave(&table->rwlock, flags);
1214 if (!is_gid_entry_valid(table->data_vec[index]))
1215 goto done;
1216
1217 get_gid_entry(table->data_vec[index]);
1218 attr = &table->data_vec[index]->attr;
1219done:
1220 read_unlock_irqrestore(&table->rwlock, flags);
1221 return attr;
1222}
1223EXPORT_SYMBOL(rdma_get_gid_attr);
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236void rdma_put_gid_attr(const struct ib_gid_attr *attr)
1237{
1238 struct ib_gid_table_entry *entry =
1239 container_of(attr, struct ib_gid_table_entry, attr);
1240
1241 put_gid_entry(entry);
1242}
1243EXPORT_SYMBOL(rdma_put_gid_attr);
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255void rdma_hold_gid_attr(const struct ib_gid_attr *attr)
1256{
1257 struct ib_gid_table_entry *entry =
1258 container_of(attr, struct ib_gid_table_entry, attr);
1259
1260 get_gid_entry(entry);
1261}
1262EXPORT_SYMBOL(rdma_hold_gid_attr);
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276struct net_device *rdma_read_gid_attr_ndev_rcu(const struct ib_gid_attr *attr)
1277{
1278 struct ib_gid_table_entry *entry =
1279 container_of(attr, struct ib_gid_table_entry, attr);
1280 struct ib_device *device = entry->attr.device;
1281 struct net_device *ndev = ERR_PTR(-ENODEV);
1282 u8 port_num = entry->attr.port_num;
1283 struct ib_gid_table *table;
1284 unsigned long flags;
1285 bool valid;
1286
1287 table = rdma_gid_table(device, port_num);
1288
1289 read_lock_irqsave(&table->rwlock, flags);
1290 valid = is_gid_entry_valid(table->data_vec[attr->index]);
1291 if (valid) {
1292 ndev = rcu_dereference(attr->ndev);
1293 if (!ndev ||
1294 (ndev && ((READ_ONCE(ndev->flags) & IFF_UP) == 0)))
1295 ndev = ERR_PTR(-ENODEV);
1296 }
1297 read_unlock_irqrestore(&table->rwlock, flags);
1298 return ndev;
1299}
1300EXPORT_SYMBOL(rdma_read_gid_attr_ndev_rcu);
1301
1302static int get_lower_dev_vlan(struct net_device *lower_dev, void *data)
1303{
1304 u16 *vlan_id = data;
1305
1306 if (is_vlan_dev(lower_dev))
1307 *vlan_id = vlan_dev_vlan_id(lower_dev);
1308
1309
1310
1311
1312 return 1;
1313}
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327int rdma_read_gid_l2_fields(const struct ib_gid_attr *attr,
1328 u16 *vlan_id, u8 *smac)
1329{
1330 struct net_device *ndev;
1331
1332 rcu_read_lock();
1333 ndev = rcu_dereference(attr->ndev);
1334 if (!ndev) {
1335 rcu_read_unlock();
1336 return -ENODEV;
1337 }
1338 if (smac)
1339 ether_addr_copy(smac, ndev->dev_addr);
1340 if (vlan_id) {
1341 *vlan_id = 0xffff;
1342 if (is_vlan_dev(ndev)) {
1343 *vlan_id = vlan_dev_vlan_id(ndev);
1344 } else {
1345
1346
1347
1348
1349 netdev_walk_all_lower_dev_rcu(attr->ndev,
1350 get_lower_dev_vlan, vlan_id);
1351 }
1352 }
1353 rcu_read_unlock();
1354 return 0;
1355}
1356EXPORT_SYMBOL(rdma_read_gid_l2_fields);
1357
1358static int config_non_roce_gid_cache(struct ib_device *device,
1359 u8 port, int gid_tbl_len)
1360{
1361 struct ib_gid_attr gid_attr = {};
1362 struct ib_gid_table *table;
1363 int ret = 0;
1364 int i;
1365
1366 gid_attr.device = device;
1367 gid_attr.port_num = port;
1368 table = rdma_gid_table(device, port);
1369
1370 mutex_lock(&table->lock);
1371 for (i = 0; i < gid_tbl_len; ++i) {
1372 if (!device->ops.query_gid)
1373 continue;
1374 ret = device->ops.query_gid(device, port, i, &gid_attr.gid);
1375 if (ret) {
1376 dev_warn(&device->dev,
1377 "query_gid failed (%d) for index %d\n", ret,
1378 i);
1379 goto err;
1380 }
1381 gid_attr.index = i;
1382 add_modify_gid(table, &gid_attr);
1383 }
1384err:
1385 mutex_unlock(&table->lock);
1386 return ret;
1387}
1388
1389static void ib_cache_update(struct ib_device *device,
1390 u8 port,
1391 bool enforce_security)
1392{
1393 struct ib_port_attr *tprops = NULL;
1394 struct ib_pkey_cache *pkey_cache = NULL, *old_pkey_cache;
1395 int i;
1396 int ret;
1397
1398 if (!rdma_is_port_valid(device, port))
1399 return;
1400
1401 tprops = kmalloc(sizeof *tprops, GFP_KERNEL);
1402 if (!tprops)
1403 return;
1404
1405 ret = ib_query_port(device, port, tprops);
1406 if (ret) {
1407 dev_warn(&device->dev, "ib_query_port failed (%d)\n", ret);
1408 goto err;
1409 }
1410
1411 if (!rdma_protocol_roce(device, port)) {
1412 ret = config_non_roce_gid_cache(device, port,
1413 tprops->gid_tbl_len);
1414 if (ret)
1415 goto err;
1416 }
1417
1418 pkey_cache = kmalloc(struct_size(pkey_cache, table,
1419 tprops->pkey_tbl_len),
1420 GFP_KERNEL);
1421 if (!pkey_cache)
1422 goto err;
1423
1424 pkey_cache->table_len = tprops->pkey_tbl_len;
1425
1426 for (i = 0; i < pkey_cache->table_len; ++i) {
1427 ret = ib_query_pkey(device, port, i, pkey_cache->table + i);
1428 if (ret) {
1429 dev_warn(&device->dev,
1430 "ib_query_pkey failed (%d) for index %d\n",
1431 ret, i);
1432 goto err;
1433 }
1434 }
1435
1436 write_lock_irq(&device->cache.lock);
1437
1438 old_pkey_cache = device->port_data[port].cache.pkey;
1439
1440 device->port_data[port].cache.pkey = pkey_cache;
1441 device->port_data[port].cache.lmc = tprops->lmc;
1442 device->port_data[port].cache.port_state = tprops->state;
1443
1444 device->port_data[port].cache.subnet_prefix = tprops->subnet_prefix;
1445 write_unlock_irq(&device->cache.lock);
1446
1447 if (enforce_security)
1448 ib_security_cache_change(device,
1449 port,
1450 tprops->subnet_prefix);
1451
1452 kfree(old_pkey_cache);
1453 kfree(tprops);
1454 return;
1455
1456err:
1457 kfree(pkey_cache);
1458 kfree(tprops);
1459}
1460
1461static void ib_cache_task(struct work_struct *_work)
1462{
1463 struct ib_update_work *work =
1464 container_of(_work, struct ib_update_work, work);
1465
1466 ib_cache_update(work->device,
1467 work->port_num,
1468 work->enforce_security);
1469 kfree(work);
1470}
1471
1472static void ib_cache_event(struct ib_event_handler *handler,
1473 struct ib_event *event)
1474{
1475 struct ib_update_work *work;
1476
1477 if (event->event == IB_EVENT_PORT_ERR ||
1478 event->event == IB_EVENT_PORT_ACTIVE ||
1479 event->event == IB_EVENT_LID_CHANGE ||
1480 event->event == IB_EVENT_PKEY_CHANGE ||
1481 event->event == IB_EVENT_CLIENT_REREGISTER ||
1482 event->event == IB_EVENT_GID_CHANGE) {
1483 work = kmalloc(sizeof *work, GFP_ATOMIC);
1484 if (work) {
1485 INIT_WORK(&work->work, ib_cache_task);
1486 work->device = event->device;
1487 work->port_num = event->element.port_num;
1488 if (event->event == IB_EVENT_PKEY_CHANGE ||
1489 event->event == IB_EVENT_GID_CHANGE)
1490 work->enforce_security = true;
1491 else
1492 work->enforce_security = false;
1493
1494 queue_work(ib_wq, &work->work);
1495 }
1496 }
1497}
1498
1499int ib_cache_setup_one(struct ib_device *device)
1500{
1501 unsigned int p;
1502 int err;
1503
1504 rwlock_init(&device->cache.lock);
1505
1506 err = gid_table_setup_one(device);
1507 if (err)
1508 return err;
1509
1510 rdma_for_each_port (device, p)
1511 ib_cache_update(device, p, true);
1512
1513 INIT_IB_EVENT_HANDLER(&device->cache.event_handler,
1514 device, ib_cache_event);
1515 ib_register_event_handler(&device->cache.event_handler);
1516 return 0;
1517}
1518
1519void ib_cache_release_one(struct ib_device *device)
1520{
1521 unsigned int p;
1522
1523
1524
1525
1526
1527
1528
1529 rdma_for_each_port (device, p)
1530 kfree(device->port_data[p].cache.pkey);
1531
1532 gid_table_release_one(device);
1533}
1534
1535void ib_cache_cleanup_one(struct ib_device *device)
1536{
1537
1538
1539
1540
1541
1542
1543
1544 ib_unregister_event_handler(&device->cache.event_handler);
1545 flush_workqueue(ib_wq);
1546 gid_table_cleanup_one(device);
1547
1548
1549
1550
1551 flush_workqueue(ib_wq);
1552}
1553