projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
dev_ioctl: split out ndo_eth_ioctl
[linux-block.git] / drivers / infiniband / ulp / ipoib / ipoib_main.c
CommitLineData
1da177e4
LT		 1/*
2 * Copyright (c) 2004 Topspin Communications. All rights reserved.
2a1d9b7f
RD		 3 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
4 * Copyright (c) 2004 Voltaire, Inc. All rights reserved.
1da177e4
LT		 5 *
6 * This software is available to you under a choice of one of two
7 * licenses. You may choose to be licensed under the terms of the GNU
8 * General Public License (GPL) Version 2, available from the file
9 * COPYING in the main directory of this source tree, or the
10 * OpenIB.org BSD license below:
11 *
12 * Redistribution and use in source and binary forms, with or
13 * without modification, are permitted provided that the following
14 * conditions are met:
15 *
16 * - Redistributions of source code must retain the above
17 * copyright notice, this list of conditions and the following
18 * disclaimer.
19 *
20 * - Redistributions in binary form must reproduce the above
21 * copyright notice, this list of conditions and the following
22 * disclaimer in the documentation and/or other materials
23 * provided with the distribution.
24 *
25 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
26 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
27 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
28 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
29 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
30 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
31 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
32 * SOFTWARE.
1da177e4
LT		 33 */
34
35#include "ipoib.h"
36
1da177e4
LT		 37#include <linux/module.h>
38
39#include <linux/init.h>
40#include <linux/slab.h>
0f485251 41#include <linux/kernel.h>
10313cbb 42#include <linux/vmalloc.h>
1da177e4
LT		 43
44#include <linux/if_arp.h> /* For ARPHRD_xxx */
45
46#include <linux/ip.h>
47#include <linux/in.h>
48
b63b70d8
SP		 49#include <linux/jhash.h>
50#include <net/arp.h>
ddde896e
GS		 51#include <net/addrconf.h>
52#include <linux/inetdevice.h>
53#include <rdma/ib_cache.h>
14c85021 54
1da177e4
LT		 55MODULE_AUTHOR("Roland Dreier");
56MODULE_DESCRIPTION("IP-over-InfiniBand net driver");
57MODULE_LICENSE("Dual BSD/GPL");
58
0f485251
SM		 59int ipoib_sendq_size __read_mostly = IPOIB_TX_RING_SIZE;
60int ipoib_recvq_size __read_mostly = IPOIB_RX_RING_SIZE;
61
62module_param_named(send_queue_size, ipoib_sendq_size, int, 0444);
63MODULE_PARM_DESC(send_queue_size, "Number of descriptors in send queue");
64module_param_named(recv_queue_size, ipoib_recvq_size, int, 0444);
65MODULE_PARM_DESC(recv_queue_size, "Number of descriptors in receive queue");
66
1da177e4
LT		 67#ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
68int ipoib_debug_level;
69
70module_param_named(debug_level, ipoib_debug_level, int, 0644);
71MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0");
72#endif
73
1732b0ef
RD		 74struct ipoib_path_iter {
75 struct net_device *dev;
76 struct ipoib_path path;
77};
78
1da177e4
LT		 79static const u8 ipv4_bcast_addr[] = {
80 0x00, 0xff, 0xff, 0xff,
81 0xff, 0x12, 0x40, 0x1b, 0x00, 0x00, 0x00, 0x00,
82 0x00, 0x00, 0x00, 0x00, 0xff, 0xff, 0xff, 0xff
83};
84
85struct workqueue_struct *ipoib_workqueue;
86
c1a0b23b
MT		 87struct ib_sa_client ipoib_sa_client;
88
11a0ae4c 89static int ipoib_add_one(struct ib_device *device);
7c1eb45a 90static void ipoib_remove_one(struct ib_device *device, void *client_data);
b63b70d8 91static void ipoib_neigh_reclaim(struct rcu_head *rp);
ddde896e 92static struct net_device *ipoib_get_net_dev_by_params(
1fb7f897 93 struct ib_device *dev, u32 port, u16 pkey,
ddde896e
GS		 94 const union ib_gid *gid, const struct sockaddr *addr,
95 void *client_data);
492a7e67 96static int ipoib_set_mac(struct net_device *dev, void *addr);
31a82362
FD		 97static int ipoib_ioctl(struct net_device *dev, struct ifreq *ifr,
98 int cmd);
1da177e4
LT		 99
100static struct ib_client ipoib_client = {
101 .name = "ipoib",
102 .add = ipoib_add_one,
ddde896e
GS		 103 .remove = ipoib_remove_one,
104 .get_net_dev_by_params = ipoib_get_net_dev_by_params,
1da177e4
LT		 105};
106
771a5258
SR		 107#ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
108static int ipoib_netdev_event(struct notifier_block *this,
109 unsigned long event, void *ptr)
110{
111 struct netdev_notifier_info *ni = ptr;
112 struct net_device *dev = ni->dev;
113
114 if (dev->netdev_ops->ndo_open != ipoib_open)
115 return NOTIFY_DONE;
116
117 switch (event) {
118 case NETDEV_REGISTER:
119 ipoib_create_debug_files(dev);
120 break;
121 case NETDEV_CHANGENAME:
122 ipoib_delete_debug_files(dev);
123 ipoib_create_debug_files(dev);
124 break;
125 case NETDEV_UNREGISTER:
126 ipoib_delete_debug_files(dev);
127 break;
128 }
129
130 return NOTIFY_DONE;
131}
132#endif
133
1da177e4
LT		 134int ipoib_open(struct net_device *dev)
135{
c1048aff 136 struct ipoib_dev_priv *priv = ipoib_priv(dev);
1da177e4
LT		 137
138 ipoib_dbg(priv, "bringing up interface\n");
139
437708c4
MS		 140 netif_carrier_off(dev);
141
e028cc55 142 set_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags);
1da177e4 143
efc82eee 144 if (ipoib_ib_dev_open(dev)) {
dd57c930
AE		 145 if (!test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags))
146 return 0;
b8a1b1ce 147 goto err_disable;
dd57c930 148 }
fe25c561 149
5c37077f 150 ipoib_ib_dev_up(dev);
1da177e4
LT		 151
152 if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) {
153 struct ipoib_dev_priv *cpriv;
154
155 /* Bring up any child interfaces too */
f47944cc 156 down_read(&priv->vlan_rwsem);
1da177e4
LT		 157 list_for_each_entry(cpriv, &priv->child_intfs, list) {
158 int flags;
159
160 flags = cpriv->dev->flags;
161 if (flags & IFF_UP)
162 continue;
163
567c5e13 164 dev_change_flags(cpriv->dev, flags | IFF_UP, NULL);
1da177e4 165 }
f47944cc 166 up_read(&priv->vlan_rwsem);
3ccbd933
JW		 167 } else if (priv->parent) {
168 struct ipoib_dev_priv *ppriv = ipoib_priv(priv->parent);
1da177e4 169
3ccbd933
JW		 170 if (!test_bit(IPOIB_FLAG_ADMIN_UP, &ppriv->flags))
171 ipoib_dbg(priv, "parent device %s is not up, so child device may be not functioning.\n",
172 ppriv->dev->name);
173 }
1da177e4
LT		 174 netif_start_queue(dev);
175
176 return 0;
b8a1b1ce 177
b8a1b1ce 178err_disable:
b8a1b1ce
RD		 179 clear_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags);
180
181 return -EINVAL;
1da177e4
LT		 182}
183
184static int ipoib_stop(struct net_device *dev)
185{
c1048aff 186 struct ipoib_dev_priv *priv = ipoib_priv(dev);
1da177e4
LT		 187
188 ipoib_dbg(priv, "stopping interface\n");
189
190 clear_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags);
191
192 netif_stop_queue(dev);
193
efc82eee 194 ipoib_ib_dev_down(dev);
cd565b4b 195 ipoib_ib_dev_stop(dev);
1da177e4
LT		 196
197 if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) {
198 struct ipoib_dev_priv *cpriv;
199
200 /* Bring down any child interfaces too */
f47944cc 201 down_read(&priv->vlan_rwsem);
1da177e4
LT		 202 list_for_each_entry(cpriv, &priv->child_intfs, list) {
203 int flags;
204
205 flags = cpriv->dev->flags;
206 if (!(flags & IFF_UP))
207 continue;
208
567c5e13 209 dev_change_flags(cpriv->dev, flags & ~IFF_UP, NULL);
1da177e4 210 }
f47944cc 211 up_read(&priv->vlan_rwsem);
1da177e4
LT		 212 }
213
214 return 0;
215}
216
9ca36f7d 217static netdev_features_t ipoib_fix_features(struct net_device *dev, netdev_features_t features)
3d96c74d 218{
c1048aff 219 struct ipoib_dev_priv *priv = ipoib_priv(dev);
3d96c74d
MM		 220
221 if (test_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags))
c4268778 222 features &= ~(NETIF_F_IP_CSUM | NETIF_F_TSO);
3d96c74d
MM		 223
224 return features;
225}
226
1da177e4
LT		 227static int ipoib_change_mtu(struct net_device *dev, int new_mtu)
228{
c1048aff 229 struct ipoib_dev_priv *priv = ipoib_priv(dev);
ed7b521d 230 int ret = 0;
1da177e4 231
839fcaba 232 /* dev->mtu > 2K ==> connected mode */
586a6934
PS		 233 if (ipoib_cm_admin_enabled(dev)) {
234 if (new_mtu > ipoib_cm_max_mtu(dev))
235 return -EINVAL;
236
839fcaba
MT		 237 if (new_mtu > priv->mcast_mtu)
238 ipoib_warn(priv, "mtu > %d will cause multicast packet drops.\n",
239 priv->mcast_mtu);
586a6934 240
839fcaba
MT		 241 dev->mtu = new_mtu;
242 return 0;
243 }
244
142a9c28
MS		 245 if (new_mtu < (ETH_MIN_MTU + IPOIB_ENCAP_LEN) ||
246 new_mtu > IPOIB_UD_MTU(priv->max_ib_mtu))
1da177e4
LT		 247 return -EINVAL;
248
249 priv->admin_mtu = new_mtu;
250
29da686d
FD		 251 if (priv->mcast_mtu < priv->admin_mtu)
252 ipoib_dbg(priv, "MTU must be smaller than the underlying "
253 "link layer MTU - 4 (%u)\n", priv->mcast_mtu);
254
ed7b521d 255 new_mtu = min(priv->mcast_mtu, priv->admin_mtu);
1da177e4 256
ed7b521d
ES		 257 if (priv->rn_ops->ndo_change_mtu) {
258 bool carrier_status = netif_carrier_ok(dev);
259
260 netif_carrier_off(dev);
261
262 /* notify lower level on the real mtu */
263 ret = priv->rn_ops->ndo_change_mtu(dev, new_mtu);
264
265 if (carrier_status)
266 netif_carrier_on(dev);
267 } else {
268 dev->mtu = new_mtu;
269 }
270
271 return ret;
1da177e4
LT		 272}
273
b6c871e5
ES		 274static void ipoib_get_stats(struct net_device *dev,
275 struct rtnl_link_stats64 *stats)
276{
277 struct ipoib_dev_priv *priv = ipoib_priv(dev);
278
279 if (priv->rn_ops->ndo_get_stats64)
280 priv->rn_ops->ndo_get_stats64(dev, stats);
281 else
282 netdev_stats_to_stats64(stats, &dev->stats);
283}
284
ddde896e
GS		 285/* Called with an RCU read lock taken */
286static bool ipoib_is_dev_match_addr_rcu(const struct sockaddr *addr,
287 struct net_device *dev)
288{
289 struct net *net = dev_net(dev);
290 struct in_device *in_dev;
291 struct sockaddr_in *addr_in = (struct sockaddr_in *)addr;
292 struct sockaddr_in6 *addr_in6 = (struct sockaddr_in6 *)addr;
293 __be32 ret_addr;
294
295 switch (addr->sa_family) {
296 case AF_INET:
297 in_dev = in_dev_get(dev);
298 if (!in_dev)
299 return false;
300
301 ret_addr = inet_confirm_addr(net, in_dev, 0,
302 addr_in->sin_addr.s_addr,
303 RT_SCOPE_HOST);
304 in_dev_put(in_dev);
305 if (ret_addr)
306 return true;
307
308 break;
309 case AF_INET6:
310 if (IS_ENABLED(CONFIG_IPV6) &&
311 ipv6_chk_addr(net, &addr_in6->sin6_addr, dev, 1))
312 return true;
313
314 break;
315 }
316 return false;
317}
318
bf194997 319/*
ddde896e
GS		 320 * Find the master net_device on top of the given net_device.
321 * @dev: base IPoIB net_device
322 *
323 * Returns the master net_device with a reference held, or the same net_device
324 * if no master exists.
325 */
326static struct net_device *ipoib_get_master_net_dev(struct net_device *dev)
327{
328 struct net_device *master;
329
330 rcu_read_lock();
331 master = netdev_master_upper_dev_get_rcu(dev);
332 if (master)
333 dev_hold(master);
334 rcu_read_unlock();
335
336 if (master)
337 return master;
338
339 dev_hold(dev);
340 return dev;
341}
342
e0e79c8e
DA		 343struct ipoib_walk_data {
344 const struct sockaddr *addr;
345 struct net_device *result;
346};
347
eff74233
TY		 348static int ipoib_upper_walk(struct net_device *upper,
349 struct netdev_nested_priv *priv)
e0e79c8e 350{
eff74233 351 struct ipoib_walk_data *data = (struct ipoib_walk_data *)priv->data;
e0e79c8e
DA		 352 int ret = 0;
353
354 if (ipoib_is_dev_match_addr_rcu(data->addr, upper)) {
355 dev_hold(upper);
356 data->result = upper;
357 ret = 1;
358 }
359
360 return ret;
361}
362
ddde896e 363/**
bf194997
LR		 364 * ipoib_get_net_dev_match_addr - Find a net_device matching
365 * the given address, which is an upper device of the given net_device.
366 *
ddde896e
GS		 367 * @addr: IP address to look for.
368 * @dev: base IPoIB net_device
369 *
370 * If found, returns the net_device with a reference held. Otherwise return
371 * NULL.
372 */
373static struct net_device *ipoib_get_net_dev_match_addr(
374 const struct sockaddr *addr, struct net_device *dev)
375{
eff74233 376 struct netdev_nested_priv priv;
e0e79c8e
DA		 377 struct ipoib_walk_data data = {
378 .addr = addr,
379 };
ddde896e 380
eff74233 381 priv.data = (void *)&data;
ddde896e
GS		 382 rcu_read_lock();
383 if (ipoib_is_dev_match_addr_rcu(addr, dev)) {
384 dev_hold(dev);
e0e79c8e 385 data.result = dev;
ddde896e
GS		 386 goto out;
387 }
388
eff74233 389 netdev_walk_all_upper_dev_rcu(dev, ipoib_upper_walk, &priv);
ddde896e
GS		 390out:
391 rcu_read_unlock();
e0e79c8e 392 return data.result;
ddde896e
GS		 393}
394
395/* returns the number of IPoIB netdevs on top a given ipoib device matching a
396 * pkey_index and address, if one exists.
397 *
398 * @found_net_dev: contains a matching net_device if the return value >= 1,
399 * with a reference held. */
400static int ipoib_match_gid_pkey_addr(struct ipoib_dev_priv *priv,
401 const union ib_gid *gid,
402 u16 pkey_index,
403 const struct sockaddr *addr,
404 int nesting,
405 struct net_device **found_net_dev)
406{
407 struct ipoib_dev_priv *child_priv;
408 struct net_device *net_dev = NULL;
409 int matches = 0;
410
411 if (priv->pkey_index == pkey_index &&
412 (!gid || !memcmp(gid, &priv->local_gid, sizeof(*gid)))) {
413 if (!addr) {
414 net_dev = ipoib_get_master_net_dev(priv->dev);
415 } else {
416 /* Verify the net_device matches the IP address, as
417 * IPoIB child devices currently share a GID. */
418 net_dev = ipoib_get_net_dev_match_addr(addr, priv->dev);
419 }
420 if (net_dev) {
421 if (!*found_net_dev)
422 *found_net_dev = net_dev;
423 else
424 dev_put(net_dev);
425 ++matches;
426 }
427 }
428
429 /* Check child interfaces */
430 down_read_nested(&priv->vlan_rwsem, nesting);
431 list_for_each_entry(child_priv, &priv->child_intfs, list) {
432 matches += ipoib_match_gid_pkey_addr(child_priv, gid,
433 pkey_index, addr,
434 nesting + 1,
435 found_net_dev);
436 if (matches > 1)
437 break;
438 }
439 up_read(&priv->vlan_rwsem);
440
441 return matches;
442}
443
444/* Returns the number of matching net_devs found (between 0 and 2). Also
445 * return the matching net_device in the @net_dev parameter, holding a
446 * reference to the net_device, if the number of matches >= 1 */
1fb7f897 447static int __ipoib_get_net_dev_by_params(struct list_head *dev_list, u32 port,
ddde896e
GS		 448 u16 pkey_index,
449 const union ib_gid *gid,
450 const struct sockaddr *addr,
451 struct net_device **net_dev)
452{
453 struct ipoib_dev_priv *priv;
454 int matches = 0;
455
456 *net_dev = NULL;
457
458 list_for_each_entry(priv, dev_list, list) {
459 if (priv->port != port)
460 continue;
461
462 matches += ipoib_match_gid_pkey_addr(priv, gid, pkey_index,
463 addr, 0, net_dev);
464 if (matches > 1)
465 break;
466 }
467
468 return matches;
469}
470
471static struct net_device *ipoib_get_net_dev_by_params(
1fb7f897 472 struct ib_device *dev, u32 port, u16 pkey,
ddde896e
GS		 473 const union ib_gid *gid, const struct sockaddr *addr,
474 void *client_data)
475{
476 struct net_device *net_dev;
477 struct list_head *dev_list = client_data;
478 u16 pkey_index;
479 int matches;
480 int ret;
481
482 if (!rdma_protocol_ib(dev, port))
483 return NULL;
484
485 ret = ib_find_cached_pkey(dev, port, pkey, &pkey_index);
486 if (ret)
487 return NULL;
488
ddde896e
GS		 489 /* See if we can find a unique device matching the L2 parameters */
490 matches = __ipoib_get_net_dev_by_params(dev_list, port, pkey_index,
491 gid, NULL, &net_dev);
492
493 switch (matches) {
494 case 0:
495 return NULL;
496 case 1:
497 return net_dev;
498 }
499
500 dev_put(net_dev);
501
502 /* Couldn't find a unique device with L2 parameters only. Use L3
503 * address to uniquely match the net device */
504 matches = __ipoib_get_net_dev_by_params(dev_list, port, pkey_index,
505 gid, addr, &net_dev);
506 switch (matches) {
507 case 0:
508 return NULL;
509 default:
510 dev_warn_ratelimited(&dev->dev,
511 "duplicate IP address detected\n");
df561f66 512 fallthrough;
ddde896e
GS		 513 case 1:
514 return net_dev;
515 }
516}
517
71d9c5f9
RD		 518int ipoib_set_mode(struct net_device *dev, const char *buf)
519{
c1048aff 520 struct ipoib_dev_priv *priv = ipoib_priv(dev);
71d9c5f9 521
80b5b35a
FD		 522 if ((test_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags) &&
523 !strcmp(buf, "connected\n")) ||
524 (!test_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags) &&
525 !strcmp(buf, "datagram\n"))) {
526 return 0;
527 }
528
71d9c5f9
RD		 529 /* flush paths if we switch modes so that connections are restarted */
530 if (IPOIB_CM_SUPPORTED(dev->dev_addr) && !strcmp(buf, "connected\n")) {
531 set_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags);
532 ipoib_warn(priv, "enabling connected mode "
533 "will cause multicast packet drops\n");
534 netdev_update_features(dev);
edcd2a74 535 dev_set_mtu(dev, ipoib_cm_max_mtu(dev));
8f149b68 536 netif_set_real_num_tx_queues(dev, 1);
71d9c5f9 537 rtnl_unlock();
e622f2f4 538 priv->tx_wr.wr.send_flags &= ~IB_SEND_IP_CSUM;
71d9c5f9
RD		 539
540 ipoib_flush_paths(dev);
0a0007f2 541 return (!rtnl_trylock()) ? -EBUSY : 0;
71d9c5f9
RD		 542 }
543
544 if (!strcmp(buf, "datagram\n")) {
545 clear_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags);
546 netdev_update_features(dev);
547 dev_set_mtu(dev, min(priv->mcast_mtu, dev->mtu));
8f149b68 548 netif_set_real_num_tx_queues(dev, dev->num_tx_queues);
71d9c5f9
RD		 549 rtnl_unlock();
550 ipoib_flush_paths(dev);
0a0007f2 551 return (!rtnl_trylock()) ? -EBUSY : 0;
71d9c5f9
RD		 552 }
553
554 return -EINVAL;
555}
556
546481c2 557struct ipoib_path *__path_find(struct net_device *dev, void *gid)
1da177e4 558{
c1048aff 559 struct ipoib_dev_priv *priv = ipoib_priv(dev);
1da177e4
LT		 560 struct rb_node *n = priv->path_tree.rb_node;
561 struct ipoib_path *path;
562 int ret;
563
564 while (n) {
565 path = rb_entry(n, struct ipoib_path, rb_node);
566
37c22a77 567 ret = memcmp(gid, path->pathrec.dgid.raw,
1da177e4
LT		 568 sizeof (union ib_gid));
569
570 if (ret < 0)
571 n = n->rb_left;
572 else if (ret > 0)
573 n = n->rb_right;
574 else
575 return path;
576 }
577
578 return NULL;
579}
580
581static int __path_add(struct net_device *dev, struct ipoib_path *path)
582{
c1048aff 583 struct ipoib_dev_priv *priv = ipoib_priv(dev);
1da177e4
LT		 584 struct rb_node **n = &priv->path_tree.rb_node;
585 struct rb_node *pn = NULL;
586 struct ipoib_path *tpath;
587 int ret;
588
589 while (*n) {
590 pn = *n;
591 tpath = rb_entry(pn, struct ipoib_path, rb_node);
592
593 ret = memcmp(path->pathrec.dgid.raw, tpath->pathrec.dgid.raw,
594 sizeof (union ib_gid));
595 if (ret < 0)
596 n = &pn->rb_left;
597 else if (ret > 0)
598 n = &pn->rb_right;
599 else
600 return -EEXIST;
601 }
602
603 rb_link_node(&path->rb_node, pn, n);
604 rb_insert_color(&path->rb_node, &priv->path_tree);
605
606 list_add_tail(&path->list, &priv->path_list);
607
608 return 0;
609}
610
611static void path_free(struct net_device *dev, struct ipoib_path *path)
612{
1da177e4 613 struct sk_buff *skb;
1da177e4
LT		 614
615 while ((skb = __skb_dequeue(&path->queue)))
616 dev_kfree_skb_irq(skb);
617
0dd9ce18 618 ipoib_dbg(ipoib_priv(dev), "%s\n", __func__);
1da177e4 619
b63b70d8
SP		 620 /* remove all neigh connected to this path */
621 ipoib_del_neighs_by_gid(dev, path->pathrec.dgid.raw);
1da177e4
LT		 622
623 if (path->ah)
624 ipoib_put_ah(path->ah);
625
626 kfree(path);
627}
628
1732b0ef
RD		 629#ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
630
631struct ipoib_path_iter *ipoib_path_iter_init(struct net_device *dev)
632{
633 struct ipoib_path_iter *iter;
634
b1b63970 635 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
1732b0ef
RD		 636 if (!iter)
637 return NULL;
638
639 iter->dev = dev;
640 memset(iter->path.pathrec.dgid.raw, 0, 16);
641
642 if (ipoib_path_iter_next(iter)) {
643 kfree(iter);
644 return NULL;
645 }
646
647 return iter;
648}
649
650int ipoib_path_iter_next(struct ipoib_path_iter *iter)
651{
c1048aff 652 struct ipoib_dev_priv *priv = ipoib_priv(iter->dev);
1732b0ef
RD		 653 struct rb_node *n;
654 struct ipoib_path *path;
655 int ret = 1;
656
657 spin_lock_irq(&priv->lock);
658
659 n = rb_first(&priv->path_tree);
660
661 while (n) {
662 path = rb_entry(n, struct ipoib_path, rb_node);
663
664 if (memcmp(iter->path.pathrec.dgid.raw, path->pathrec.dgid.raw,
665 sizeof (union ib_gid)) < 0) {
666 iter->path = *path;
667 ret = 0;
668 break;
669 }
670
671 n = rb_next(n);
672 }
673
674 spin_unlock_irq(&priv->lock);
675
676 return ret;
677}
678
679void ipoib_path_iter_read(struct ipoib_path_iter *iter,
680 struct ipoib_path *path)
681{
682 *path = iter->path;
683}
684
685#endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */
686
ee1e2c82
MS		 687void ipoib_mark_paths_invalid(struct net_device *dev)
688{
c1048aff 689 struct ipoib_dev_priv *priv = ipoib_priv(dev);
ee1e2c82
MS		 690 struct ipoib_path *path, *tp;
691
692 spin_lock_irq(&priv->lock);
693
694 list_for_each_entry_safe(path, tp, &priv->path_list, list) {
57520751
DC		 695 ipoib_dbg(priv, "mark path LID 0x%08x GID %pI6 invalid\n",
696 be32_to_cpu(sa_path_get_dlid(&path->pathrec)),
697 path->pathrec.dgid.raw);
fa9391db
DL		 698 if (path->ah)
699 path->ah->valid = 0;
ee1e2c82
MS		 700 }
701
702 spin_unlock_irq(&priv->lock);
703}
704
2b084176
ES		 705static void push_pseudo_header(struct sk_buff *skb, const char *daddr)
706{
707 struct ipoib_pseudo_header *phdr;
708
d58ff351 709 phdr = skb_push(skb, sizeof(*phdr));
2b084176
ES		 710 memcpy(phdr->hwaddr, daddr, INFINIBAND_ALEN);
711}
712
1da177e4
LT		 713void ipoib_flush_paths(struct net_device *dev)
714{
c1048aff 715 struct ipoib_dev_priv *priv = ipoib_priv(dev);
1da177e4
LT		 716 struct ipoib_path *path, *tp;
717 LIST_HEAD(remove_list);
943c246e 718 unsigned long flags;
1da177e4 719
943c246e
RD		 720 netif_tx_lock_bh(dev);
721 spin_lock_irqsave(&priv->lock, flags);
1da177e4 722
157de229 723 list_splice_init(&priv->path_list, &remove_list);
1da177e4
LT		 724
725 list_for_each_entry(path, &remove_list, list)
726 rb_erase(&path->rb_node, &priv->path_tree);
727
1da177e4
LT		 728 list_for_each_entry_safe(path, tp, &remove_list, list) {
729 if (path->query)
730 ib_sa_cancel_query(path->query_id, path->query);
943c246e
RD		 731 spin_unlock_irqrestore(&priv->lock, flags);
732 netif_tx_unlock_bh(dev);
1da177e4
LT		 733 wait_for_completion(&path->done);
734 path_free(dev, path);
943c246e
RD		 735 netif_tx_lock_bh(dev);
736 spin_lock_irqsave(&priv->lock, flags);
1da177e4 737 }
943c246e
RD		 738
739 spin_unlock_irqrestore(&priv->lock, flags);
740 netif_tx_unlock_bh(dev);
1da177e4
LT		 741}
742
743static void path_rec_completion(int status,
c2f8fc4e 744 struct sa_path_rec *pathrec,
1da177e4
LT		 745 void *path_ptr)
746{
747 struct ipoib_path *path = path_ptr;
748 struct net_device *dev = path->dev;
c1048aff 749 struct ipoib_dev_priv *priv = ipoib_priv(dev);
1da177e4 750 struct ipoib_ah *ah = NULL;
c9da4bad 751 struct ipoib_ah *old_ah = NULL;
d04d01b1 752 struct ipoib_neigh *neigh, *tn;
1da177e4
LT		 753 struct sk_buff_head skqueue;
754 struct sk_buff *skb;
755 unsigned long flags;
756
843613b0 757 if (!status)
5b095d98 758 ipoib_dbg(priv, "PathRec LID 0x%04x for GID %pI6\n",
57520751 759 be32_to_cpu(sa_path_get_dlid(pathrec)),
9fdca4da 760 pathrec->dgid.raw);
1da177e4 761 else
5b095d98 762 ipoib_dbg(priv, "PathRec status %d for GID %pI6\n",
fcace2fe 763 status, path->pathrec.dgid.raw);
1da177e4
LT		 764
765 skb_queue_head_init(&skqueue);
766
767 if (!status) {
90898850 768 struct rdma_ah_attr av;
46f1b3d7 769
4ad6a024 770 if (!ib_init_ah_attr_from_path(priv->ca, priv->port,
39839107 771 pathrec, &av, NULL)) {
46f1b3d7 772 ah = ipoib_create_ah(dev, priv->pd, &av);
aa74f487
PP		 773 rdma_destroy_ah_attr(&av);
774 }
1da177e4
LT		 775 }
776
777 spin_lock_irqsave(&priv->lock, flags);
778
3874397c 779 if (!IS_ERR_OR_NULL(ah)) {
43900089
ES		 780 /*
781 * pathrec.dgid is used as the database key from the LLADDR,
782 * it must remain unchanged even if the SA returns a different
783 * GID to use in the AH.
784 */
785 if (memcmp(pathrec->dgid.raw, path->pathrec.dgid.raw,
786 sizeof(union ib_gid))) {
787 ipoib_dbg(
788 priv,
789 "%s got PathRec for gid %pI6 while asked for %pI6\n",
790 dev->name, pathrec->dgid.raw,
791 path->pathrec.dgid.raw);
792 memcpy(pathrec->dgid.raw, path->pathrec.dgid.raw,
793 sizeof(union ib_gid));
794 }
795
1da177e4
LT		 796 path->pathrec = *pathrec;
797
c9da4bad
RD		 798 old_ah = path->ah;
799 path->ah = ah;
800
1da177e4 801 ipoib_dbg(priv, "created address handle %p for LID 0x%04x, SL %d\n",
57520751 802 ah, be32_to_cpu(sa_path_get_dlid(pathrec)),
9fdca4da 803 pathrec->sl);
1da177e4
LT		 804
805 while ((skb = __skb_dequeue(&path->queue)))
806 __skb_queue_tail(&skqueue, skb);
807
d04d01b1 808 list_for_each_entry_safe(neigh, tn, &path->neigh_list, list) {
ee1e2c82
MS		 809 if (neigh->ah) {
810 WARN_ON(neigh->ah != old_ah);
811 /*
812 * Dropping the ah reference inside
813 * priv->lock is safe here, because we
814 * will hold one more reference from
815 * the original value of path->ah (ie
816 * old_ah).
817 */
818 ipoib_put_ah(neigh->ah);
819 }
1da177e4
LT		 820 kref_get(&path->ah->ref);
821 neigh->ah = path->ah;
822
b63b70d8 823 if (ipoib_cm_enabled(dev, neigh->daddr)) {
839fcaba
MT		 824 if (!ipoib_cm_get(neigh))
825 ipoib_cm_set(neigh, ipoib_cm_create_tx(dev,
826 path,
827 neigh));
828 if (!ipoib_cm_get(neigh)) {
b63b70d8 829 ipoib_neigh_free(neigh);
839fcaba
MT		 830 continue;
831 }
832 }
833
1da177e4
LT		 834 while ((skb = __skb_dequeue(&neigh->queue)))
835 __skb_queue_tail(&skqueue, skb);
836 }
fa9391db 837 path->ah->valid = 1;
5872a9fc 838 }
1da177e4 839
5872a9fc 840 path->query = NULL;
1da177e4
LT		 841 complete(&path->done);
842
843 spin_unlock_irqrestore(&priv->lock, flags);
844
f72dd566
RD		 845 if (IS_ERR_OR_NULL(ah))
846 ipoib_del_neighs_by_gid(dev, path->pathrec.dgid.raw);
847
ee1e2c82
MS		 848 if (old_ah)
849 ipoib_put_ah(old_ah);
850
1da177e4 851 while ((skb = __skb_dequeue(&skqueue))) {
d32b9a81 852 int ret;
1da177e4 853 skb->dev = dev;
d32b9a81
FD		 854 ret = dev_queue_xmit(skb);
855 if (ret)
856 ipoib_warn(priv, "%s: dev_queue_xmit failed to re-queue packet, ret:%d\n",
857 __func__, ret);
1da177e4
LT		 858 }
859}
860
98aebc55
ES		 861static void init_path_rec(struct ipoib_dev_priv *priv, struct ipoib_path *path,
862 void *gid)
863{
864 path->dev = priv->dev;
865
866 if (rdma_cap_opa_ah(priv->ca, priv->port))
867 path->pathrec.rec_type = SA_PATH_REC_TYPE_OPA;
868 else
869 path->pathrec.rec_type = SA_PATH_REC_TYPE_IB;
870
871 memcpy(path->pathrec.dgid.raw, gid, sizeof(union ib_gid));
872 path->pathrec.sgid = priv->local_gid;
873 path->pathrec.pkey = cpu_to_be16(priv->pkey);
874 path->pathrec.numb_path = 1;
875 path->pathrec.traffic_class = priv->broadcast->mcmember.traffic_class;
876}
877
37c22a77 878static struct ipoib_path *path_rec_create(struct net_device *dev, void *gid)
1da177e4 879{
c1048aff 880 struct ipoib_dev_priv *priv = ipoib_priv(dev);
1da177e4
LT		 881 struct ipoib_path *path;
882
1401b53a
JM		 883 if (!priv->broadcast)
884 return NULL;
885
b1b63970 886 path = kzalloc(sizeof(*path), GFP_ATOMIC);
1da177e4
LT		 887 if (!path)
888 return NULL;
889
1da177e4
LT		 890 skb_queue_head_init(&path->queue);
891
892 INIT_LIST_HEAD(&path->neigh_list);
1da177e4 893
98aebc55 894 init_path_rec(priv, path, gid);
1da177e4
LT		 895
896 return path;
897}
898
899static int path_rec_start(struct net_device *dev,
900 struct ipoib_path *path)
901{
c1048aff 902 struct ipoib_dev_priv *priv = ipoib_priv(dev);
1da177e4 903
5b095d98 904 ipoib_dbg(priv, "Start path record lookup for %pI6\n",
fcace2fe 905 path->pathrec.dgid.raw);
1da177e4 906
65c7edda
RD		 907 init_completion(&path->done);
908
1da177e4 909 path->query_id =
c1a0b23b 910 ib_sa_path_rec_get(&ipoib_sa_client, priv->ca, priv->port,
1da177e4
LT		 911 &path->pathrec,
912 IB_SA_PATH_REC_DGID |
913 IB_SA_PATH_REC_SGID |
914 IB_SA_PATH_REC_NUMB_PATH |
81668838 915 IB_SA_PATH_REC_TRAFFIC_CLASS |
1da177e4
LT		 916 IB_SA_PATH_REC_PKEY,
917 1000, GFP_ATOMIC,
918 path_rec_completion,
919 path, &path->query);
920 if (path->query_id < 0) {
01b3fc8b 921 ipoib_warn(priv, "ib_sa_path_rec_get failed: %d\n", path->query_id);
1da177e4 922 path->query = NULL;
93a3ab93 923 complete(&path->done);
1da177e4
LT		 924 return path->query_id;
925 }
926
927 return 0;
928}
929
fa9391db
DL		 930static void neigh_refresh_path(struct ipoib_neigh *neigh, u8 *daddr,
931 struct net_device *dev)
932{
933 struct ipoib_dev_priv *priv = ipoib_priv(dev);
934 struct ipoib_path *path;
935 unsigned long flags;
936
937 spin_lock_irqsave(&priv->lock, flags);
938
939 path = __path_find(dev, daddr + 4);
940 if (!path)
941 goto out;
942 if (!path->query)
943 path_rec_start(dev, path);
944out:
945 spin_unlock_irqrestore(&priv->lock, flags);
946}
947
16ba3def
ES		 948static struct ipoib_neigh *neigh_add_path(struct sk_buff *skb, u8 *daddr,
949 struct net_device *dev)
1da177e4 950{
c1048aff 951 struct ipoib_dev_priv *priv = ipoib_priv(dev);
cd565b4b 952 struct rdma_netdev *rn = netdev_priv(dev);
1da177e4
LT		 953 struct ipoib_path *path;
954 struct ipoib_neigh *neigh;
943c246e 955 unsigned long flags;
1da177e4 956
b5120a6e 957 spin_lock_irqsave(&priv->lock, flags);
b63b70d8 958 neigh = ipoib_neigh_alloc(daddr, dev);
1da177e4 959 if (!neigh) {
b5120a6e 960 spin_unlock_irqrestore(&priv->lock, flags);
de903512 961 ++dev->stats.tx_dropped;
1da177e4 962 dev_kfree_skb_any(skb);
16ba3def
ES		 963 return NULL;
964 }
965
966 /* To avoid race condition, make sure that the
967 * neigh will be added only once.
968 */
969 if (unlikely(!list_empty(&neigh->list))) {
970 spin_unlock_irqrestore(&priv->lock, flags);
971 return neigh;
1da177e4
LT		 972 }
973
b63b70d8 974 path = __path_find(dev, daddr + 4);
1da177e4 975 if (!path) {
b63b70d8 976 path = path_rec_create(dev, daddr + 4);
1da177e4 977 if (!path)
d2e0655e 978 goto err_path;
1da177e4
LT		 979
980 __path_add(dev, path);
981 }
982
983 list_add_tail(&neigh->list, &path->neigh_list);
984
fa9391db 985 if (path->ah && path->ah->valid) {
1da177e4
LT		 986 kref_get(&path->ah->ref);
987 neigh->ah = path->ah;
988
b63b70d8 989 if (ipoib_cm_enabled(dev, neigh->daddr)) {
839fcaba
MT		 990 if (!ipoib_cm_get(neigh))
991 ipoib_cm_set(neigh, ipoib_cm_create_tx(dev, path, neigh));
992 if (!ipoib_cm_get(neigh)) {
b63b70d8 993 ipoib_neigh_free(neigh);
839fcaba
MT		 994 goto err_drop;
995 }
fc791b63
PA		 996 if (skb_queue_len(&neigh->queue) <
997 IPOIB_MAX_PATH_REC_QUEUE) {
2b084176 998 push_pseudo_header(skb, neigh->daddr);
839fcaba 999 __skb_queue_tail(&neigh->queue, skb);
fc791b63 1000 } else {
839fcaba
MT		 1001 ipoib_warn(priv, "queue length limit %d. Packet drop.\n",
1002 skb_queue_len(&neigh->queue));
1003 goto err_drop;
1004 }
721d67cd
RD		 1005 } else {
1006 spin_unlock_irqrestore(&priv->lock, flags);
cd565b4b
ES		 1007 path->ah->last_send = rn->send(dev, skb, path->ah->ah,
1008 IPOIB_QPN(daddr));
b63b70d8 1009 ipoib_neigh_put(neigh);
16ba3def 1010 return NULL;
721d67cd 1011 }
1da177e4
LT		 1012 } else {
1013 neigh->ah = NULL;
1da177e4
LT		 1014
1015 if (!path->query && path_rec_start(dev, path))
49b8e744 1016 goto err_path;
2b084176
ES		 1017 if (skb_queue_len(&neigh->queue) < IPOIB_MAX_PATH_REC_QUEUE) {
1018 push_pseudo_header(skb, neigh->daddr);
1e85b806 1019 __skb_queue_tail(&neigh->queue, skb);
2b084176 1020 } else {
1e85b806 1021 goto err_drop;
2b084176 1022 }
1da177e4
LT		 1023 }
1024
943c246e 1025 spin_unlock_irqrestore(&priv->lock, flags);
b63b70d8 1026 ipoib_neigh_put(neigh);
16ba3def 1027 return NULL;
1da177e4 1028
d2e0655e 1029err_path:
b63b70d8 1030 ipoib_neigh_free(neigh);
839fcaba 1031err_drop:
de903512 1032 ++dev->stats.tx_dropped;
1da177e4
LT		 1033 dev_kfree_skb_any(skb);
1034
943c246e 1035 spin_unlock_irqrestore(&priv->lock, flags);
b63b70d8 1036 ipoib_neigh_put(neigh);
16ba3def
ES		 1037
1038 return NULL;
1da177e4
LT		 1039}
1040
1041static void unicast_arp_send(struct sk_buff *skb, struct net_device *dev,
fc791b63 1042 struct ipoib_pseudo_header *phdr)
1da177e4 1043{
c1048aff 1044 struct ipoib_dev_priv *priv = ipoib_priv(dev);
cd565b4b 1045 struct rdma_netdev *rn = netdev_priv(dev);
1da177e4 1046 struct ipoib_path *path;
943c246e 1047 unsigned long flags;
1da177e4 1048
943c246e 1049 spin_lock_irqsave(&priv->lock, flags);
1da177e4 1050
98aebc55
ES		 1051 /* no broadcast means that all paths are (going to be) not valid */
1052 if (!priv->broadcast)
1053 goto drop_and_unlock;
1054
fc791b63 1055 path = __path_find(dev, phdr->hwaddr + 4);
fa9391db 1056 if (!path || !path->ah || !path->ah->valid) {
71d98b46 1057 if (!path) {
fc791b63 1058 path = path_rec_create(dev, phdr->hwaddr + 4);
15517080
ES		 1059 if (!path)
1060 goto drop_and_unlock;
1061 __path_add(dev, path);
1062 } else {
1063 /*
1064 * make sure there are no changes in the existing
1065 * path record
1066 */
1067 init_path_rec(priv, path, phdr->hwaddr + 4);
1068 }
1069 if (!path->query && path_rec_start(dev, path)) {
1070 goto drop_and_unlock;
71d98b46 1071 }
1da177e4 1072
15517080
ES		 1073 if (skb_queue_len(&path->queue) < IPOIB_MAX_PATH_REC_QUEUE) {
1074 push_pseudo_header(skb, phdr->hwaddr);
1075 __skb_queue_tail(&path->queue, skb);
1076 goto unlock;
1da177e4 1077 } else {
98aebc55 1078 goto drop_and_unlock;
1da177e4 1079 }
1da177e4
LT		 1080 }
1081
943c246e 1082 spin_unlock_irqrestore(&priv->lock, flags);
15517080
ES		 1083 ipoib_dbg(priv, "Send unicast ARP to %08x\n",
1084 be32_to_cpu(sa_path_get_dlid(&path->pathrec)));
1085 path->ah->last_send = rn->send(dev, skb, path->ah->ah,
1086 IPOIB_QPN(phdr->hwaddr));
98aebc55
ES		 1087 return;
1088
1089drop_and_unlock:
1090 ++dev->stats.tx_dropped;
1091 dev_kfree_skb_any(skb);
15517080 1092unlock:
98aebc55 1093 spin_unlock_irqrestore(&priv->lock, flags);
1da177e4
LT		 1094}
1095
47a3968a 1096static netdev_tx_t ipoib_start_xmit(struct sk_buff *skb, struct net_device *dev)
1da177e4 1097{
c1048aff 1098 struct ipoib_dev_priv *priv = ipoib_priv(dev);
cd565b4b 1099 struct rdma_netdev *rn = netdev_priv(dev);
1da177e4 1100 struct ipoib_neigh *neigh;
fc791b63 1101 struct ipoib_pseudo_header *phdr;
b63b70d8 1102 struct ipoib_header *header;
1da177e4
LT		 1103 unsigned long flags;
1104
fc791b63
PA		 1105 phdr = (struct ipoib_pseudo_header *) skb->data;
1106 skb_pull(skb, sizeof(*phdr));
b63b70d8
SP		 1107 header = (struct ipoib_header *) skb->data;
1108
fc791b63 1109 if (unlikely(phdr->hwaddr[4] == 0xff)) {
b63b70d8
SP		 1110 /* multicast, arrange "if" according to probability */
1111 if ((header->proto != htons(ETH_P_IP)) &&
1112 (header->proto != htons(ETH_P_IPV6)) &&
1113 (header->proto != htons(ETH_P_ARP)) &&
dc850b0e
PM		 1114 (header->proto != htons(ETH_P_RARP)) &&
1115 (header->proto != htons(ETH_P_TIPC))) {
b63b70d8 1116 /* ethertype not supported by IPoIB */
17e6abee
DM		 1117 ++dev->stats.tx_dropped;
1118 dev_kfree_skb_any(skb);
b63b70d8 1119 return NETDEV_TX_OK;
17e6abee 1120 }
b63b70d8 1121 /* Add in the P_Key for multicast*/
fc791b63
PA		 1122 phdr->hwaddr[8] = (priv->pkey >> 8) & 0xff;
1123 phdr->hwaddr[9] = priv->pkey & 0xff;
b63b70d8 1124
fc791b63 1125 neigh = ipoib_neigh_get(dev, phdr->hwaddr);
b63b70d8
SP		 1126 if (likely(neigh))
1127 goto send_using_neigh;
fc791b63 1128 ipoib_mcast_send(dev, phdr->hwaddr, skb);
b63b70d8 1129 return NETDEV_TX_OK;
17e6abee 1130 }
1da177e4 1131
b63b70d8
SP		 1132 /* unicast, arrange "switch" according to probability */
1133 switch (header->proto) {
1134 case htons(ETH_P_IP):
1135 case htons(ETH_P_IPV6):
dc850b0e 1136 case htons(ETH_P_TIPC):
fc791b63 1137 neigh = ipoib_neigh_get(dev, phdr->hwaddr);
b63b70d8 1138 if (unlikely(!neigh)) {
16ba3def
ES		 1139 neigh = neigh_add_path(skb, phdr->hwaddr, dev);
1140 if (likely(!neigh))
1141 return NETDEV_TX_OK;
a50df398 1142 }
b63b70d8
SP		 1143 break;
1144 case htons(ETH_P_ARP):
1145 case htons(ETH_P_RARP):
1146 /* for unicast ARP and RARP should always perform path find */
fc791b63 1147 unicast_arp_send(skb, dev, phdr);
b63b70d8
SP		 1148 return NETDEV_TX_OK;
1149 default:
1150 /* ethertype not supported by IPoIB */
1151 ++dev->stats.tx_dropped;
1152 dev_kfree_skb_any(skb);
1153 return NETDEV_TX_OK;
1154 }
8a7f7521 1155
b63b70d8
SP		 1156send_using_neigh:
1157 /* note we now hold a ref to neigh */
1158 if (ipoib_cm_get(neigh)) {
1159 if (ipoib_cm_up(neigh)) {
1160 ipoib_cm_send(dev, skb, ipoib_cm_get(neigh));
1161 goto unref;
1da177e4 1162 }
fa9391db 1163 } else if (neigh->ah && neigh->ah->valid) {
cd565b4b
ES		 1164 neigh->ah->last_send = rn->send(dev, skb, neigh->ah->ah,
1165 IPOIB_QPN(phdr->hwaddr));
b63b70d8 1166 goto unref;
fa9391db
DL		 1167 } else if (neigh->ah) {
1168 neigh_refresh_path(neigh, phdr->hwaddr, dev);
b63b70d8 1169 }
1da177e4 1170
b63b70d8 1171 if (skb_queue_len(&neigh->queue) < IPOIB_MAX_PATH_REC_QUEUE) {
2b084176 1172 push_pseudo_header(skb, phdr->hwaddr);
b63b70d8
SP		 1173 spin_lock_irqsave(&priv->lock, flags);
1174 __skb_queue_tail(&neigh->queue, skb);
1175 spin_unlock_irqrestore(&priv->lock, flags);
1da177e4 1176 } else {
b63b70d8
SP		 1177 ++dev->stats.tx_dropped;
1178 dev_kfree_skb_any(skb);
1179 }
1da177e4 1180
b63b70d8
SP		 1181unref:
1182 ipoib_neigh_put(neigh);
1da177e4 1183
1da177e4
LT		 1184 return NETDEV_TX_OK;
1185}
1186
0290bd29 1187static void ipoib_timeout(struct net_device *dev, unsigned int txqueue)
1da177e4 1188{
c1048aff 1189 struct ipoib_dev_priv *priv = ipoib_priv(dev);
042a00f9 1190 struct rdma_netdev *rn = netdev_priv(dev);
1da177e4 1191
042a00f9
MM		 1192 if (rn->tx_timeout) {
1193 rn->tx_timeout(dev, txqueue);
1194 return;
1195 }
4b2d319b 1196 ipoib_warn(priv, "transmit timeout: latency %d msecs\n",
4d0e9657 1197 jiffies_to_msecs(jiffies - dev_trans_start(dev)));
1acba6a8
VF		 1198 ipoib_warn(priv,
1199 "queue stopped %d, tx_head %u, tx_tail %u, global_tx_head %u, global_tx_tail %u\n",
1200 netif_queue_stopped(dev), priv->tx_head, priv->tx_tail,
1201 priv->global_tx_head, priv->global_tx_tail);
1202
1da177e4
LT		 1203 /* XXX reset QP, etc. */
1204}
1205
1206static int ipoib_hard_header(struct sk_buff *skb,
1207 struct net_device *dev,
1208 unsigned short type,
0578cdad
KH		 1209 const void *daddr,
1210 const void *saddr,
1211 unsigned int len)
1da177e4
LT		 1212{
1213 struct ipoib_header *header;
1214
b1b63970 1215 header = skb_push(skb, sizeof(*header));
1da177e4
LT		 1216
1217 header->proto = htons(type);
1218 header->reserved = 0;
1219
1220 /*
b63b70d8 1221 * we don't rely on dst_entry structure, always stuff the
fc791b63 1222 * destination address into skb hard header so we can figure out where
936d7de3 1223 * to send the packet later.
1da177e4 1224 */
2b084176 1225 push_pseudo_header(skb, daddr);
1da177e4 1226
fc791b63 1227 return IPOIB_HARD_LEN;
1da177e4
LT		 1228}
1229
1230static void ipoib_set_mcast_list(struct net_device *dev)
1231{
c1048aff 1232 struct ipoib_dev_priv *priv = ipoib_priv(dev);
1da177e4 1233
7a343d4c
LA		 1234 if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags)) {
1235 ipoib_dbg(priv, "IPOIB_FLAG_OPER_UP not set");
1236 return;
1237 }
1238
0b39578b 1239 queue_work(priv->wq, &priv->restart_task);
1da177e4
LT		 1240}
1241
5aa7add8
ND		 1242static int ipoib_get_iflink(const struct net_device *dev)
1243{
c1048aff 1244 struct ipoib_dev_priv *priv = ipoib_priv(dev);
5aa7add8 1245
2c153959
ES		 1246 /* parent interface */
1247 if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags))
1248 return dev->ifindex;
1249
1250 /* child/vlan interface */
5aa7add8
ND		 1251 return priv->parent->ifindex;
1252}
1253
b63b70d8 1254static u32 ipoib_addr_hash(struct ipoib_neigh_hash *htbl, u8 *daddr)
1da177e4 1255{
b63b70d8
SP		 1256 /*
1257 * Use only the address parts that contributes to spreading
1258 * The subnet prefix is not used as one can not connect to
1259 * same remote port (GUID) using the same remote QPN via two
1260 * different subnets.
1261 */
1262 /* qpn octets[1:4) & port GUID octets[12:20) */
9d1ad66e 1263 u32 *d32 = (u32 *) daddr;
b63b70d8
SP		 1264 u32 hv;
1265
9d1ad66e 1266 hv = jhash_3words(d32[3], d32[4], IPOIB_QPN_MASK & d32[0], 0);
b63b70d8
SP		 1267 return hv & htbl->mask;
1268}
1269
1270struct ipoib_neigh *ipoib_neigh_get(struct net_device *dev, u8 *daddr)
1271{
c1048aff 1272 struct ipoib_dev_priv *priv = ipoib_priv(dev);
b63b70d8
SP		 1273 struct ipoib_neigh_table *ntbl = &priv->ntbl;
1274 struct ipoib_neigh_hash *htbl;
1275 struct ipoib_neigh *neigh = NULL;
1276 u32 hash_val;
1277
1278 rcu_read_lock_bh();
1279
1280 htbl = rcu_dereference_bh(ntbl->htbl);
1281
1282 if (!htbl)
1283 goto out_unlock;
1284
1285 hash_val = ipoib_addr_hash(htbl, daddr);
1286 for (neigh = rcu_dereference_bh(htbl->buckets[hash_val]);
1287 neigh != NULL;
1288 neigh = rcu_dereference_bh(neigh->hnext)) {
1289 if (memcmp(daddr, neigh->daddr, INFINIBAND_ALEN) == 0) {
1290 /* found, take one ref on behalf of the caller */
a5e27fb6 1291 if (!refcount_inc_not_zero(&neigh->refcnt)) {
b63b70d8
SP		 1292 /* deleted */
1293 neigh = NULL;
1294 goto out_unlock;
1295 }
61c78eea
ES		 1296
1297 if (likely(skb_queue_len(&neigh->queue) < IPOIB_MAX_PATH_REC_QUEUE))
1298 neigh->alive = jiffies;
b63b70d8
SP		 1299 goto out_unlock;
1300 }
1301 }
1302
1303out_unlock:
1304 rcu_read_unlock_bh();
1305 return neigh;
1306}
1307
1308static void __ipoib_reap_neigh(struct ipoib_dev_priv *priv)
1309{
1310 struct ipoib_neigh_table *ntbl = &priv->ntbl;
1311 struct ipoib_neigh_hash *htbl;
1312 unsigned long neigh_obsolete;
1313 unsigned long dt;
1da177e4 1314 unsigned long flags;
b63b70d8 1315 int i;
bd99b2e0 1316 LIST_HEAD(remove_list);
1da177e4 1317
b5120a6e 1318 spin_lock_irqsave(&priv->lock, flags);
b63b70d8
SP		 1319
1320 htbl = rcu_dereference_protected(ntbl->htbl,
b5120a6e 1321 lockdep_is_held(&priv->lock));
b63b70d8
SP		 1322
1323 if (!htbl)
1324 goto out_unlock;
1325
1326 /* neigh is obsolete if it was idle for two GC periods */
1327 dt = 2 * arp_tbl.gc_interval;
1328 neigh_obsolete = jiffies - dt;
b63b70d8
SP		 1329
1330 for (i = 0; i < htbl->size; i++) {
1331 struct ipoib_neigh *neigh;
1332 struct ipoib_neigh __rcu **np = &htbl->buckets[i];
1333
1334 while ((neigh = rcu_dereference_protected(*np,
b5120a6e 1335 lockdep_is_held(&priv->lock))) != NULL) {
b63b70d8
SP		 1336 /* was the neigh idle for two GC periods */
1337 if (time_after(neigh_obsolete, neigh->alive)) {
bd99b2e0 1338
432c55ff 1339 ipoib_check_and_add_mcast_sendonly(priv, neigh->daddr + 4, &remove_list);
bd99b2e0 1340
b63b70d8
SP		 1341 rcu_assign_pointer(*np,
1342 rcu_dereference_protected(neigh->hnext,
b5120a6e 1343 lockdep_is_held(&priv->lock)));
b63b70d8 1344 /* remove from path/mc list */
c586071d 1345 list_del_init(&neigh->list);
b63b70d8
SP		 1346 call_rcu(&neigh->rcu, ipoib_neigh_reclaim);
1347 } else {
1348 np = &neigh->hnext;
1349 }
1da177e4 1350
b63b70d8
SP		 1351 }
1352 }
1da177e4 1353
b63b70d8 1354out_unlock:
b5120a6e 1355 spin_unlock_irqrestore(&priv->lock, flags);
50be28de 1356 ipoib_mcast_remove_list(&remove_list);
b63b70d8 1357}
1da177e4 1358
b63b70d8
SP		 1359static void ipoib_reap_neigh(struct work_struct *work)
1360{
1361 struct ipoib_dev_priv *priv =
1362 container_of(work, struct ipoib_dev_priv, neigh_reap_task.work);
1363
1364 __ipoib_reap_neigh(priv);
1365
cda8daf1
ES		 1366 queue_delayed_work(priv->wq, &priv->neigh_reap_task,
1367 arp_tbl.gc_interval);
1da177e4
LT		 1368}
1369
b63b70d8
SP		 1370
1371static struct ipoib_neigh *ipoib_neigh_ctor(u8 *daddr,
732a2170 1372 struct net_device *dev)
d2e0655e
MT		 1373{
1374 struct ipoib_neigh *neigh;
1375
b1b63970 1376 neigh = kzalloc(sizeof(*neigh), GFP_ATOMIC);
d2e0655e
MT		 1377 if (!neigh)
1378 return NULL;
1379
732a2170 1380 neigh->dev = dev;
b63b70d8 1381 memcpy(&neigh->daddr, daddr, sizeof(neigh->daddr));
82b39913 1382 skb_queue_head_init(&neigh->queue);
b63b70d8 1383 INIT_LIST_HEAD(&neigh->list);
839fcaba 1384 ipoib_cm_set(neigh, NULL);
b63b70d8 1385 /* one ref on behalf of the caller */
a5e27fb6 1386 refcount_set(&neigh->refcnt, 1);
b63b70d8
SP		 1387
1388 return neigh;
1389}
1390
1391struct ipoib_neigh *ipoib_neigh_alloc(u8 *daddr,
1392 struct net_device *dev)
1393{
c1048aff 1394 struct ipoib_dev_priv *priv = ipoib_priv(dev);
b63b70d8
SP		 1395 struct ipoib_neigh_table *ntbl = &priv->ntbl;
1396 struct ipoib_neigh_hash *htbl;
1397 struct ipoib_neigh *neigh;
1398 u32 hash_val;
1399
b63b70d8 1400 htbl = rcu_dereference_protected(ntbl->htbl,
b5120a6e 1401 lockdep_is_held(&priv->lock));
b63b70d8
SP		 1402 if (!htbl) {
1403 neigh = NULL;
1404 goto out_unlock;
1405 }
1406
1407 /* need to add a new neigh, but maybe some other thread succeeded?
1408 * recalc hash, maybe hash resize took place so we do a search
1409 */
1410 hash_val = ipoib_addr_hash(htbl, daddr);
1411 for (neigh = rcu_dereference_protected(htbl->buckets[hash_val],
b5120a6e 1412 lockdep_is_held(&priv->lock));
b63b70d8
SP		 1413 neigh != NULL;
1414 neigh = rcu_dereference_protected(neigh->hnext,
b5120a6e 1415 lockdep_is_held(&priv->lock))) {
b63b70d8
SP		 1416 if (memcmp(daddr, neigh->daddr, INFINIBAND_ALEN) == 0) {
1417 /* found, take one ref on behalf of the caller */
a5e27fb6 1418 if (!refcount_inc_not_zero(&neigh->refcnt)) {
b63b70d8
SP		 1419 /* deleted */
1420 neigh = NULL;
1421 break;
1422 }
1423 neigh->alive = jiffies;
1424 goto out_unlock;
1425 }
1426 }
1427
1428 neigh = ipoib_neigh_ctor(daddr, dev);
1429 if (!neigh)
1430 goto out_unlock;
1431
1432 /* one ref on behalf of the hash table */
a5e27fb6 1433 refcount_inc(&neigh->refcnt);
b63b70d8
SP		 1434 neigh->alive = jiffies;
1435 /* put in hash */
1436 rcu_assign_pointer(neigh->hnext,
1437 rcu_dereference_protected(htbl->buckets[hash_val],
b5120a6e 1438 lockdep_is_held(&priv->lock)));
b63b70d8
SP		 1439 rcu_assign_pointer(htbl->buckets[hash_val], neigh);
1440 atomic_inc(&ntbl->entries);
1441
1442out_unlock:
d2e0655e
MT		 1443
1444 return neigh;
1445}
1446
b63b70d8 1447void ipoib_neigh_dtor(struct ipoib_neigh *neigh)
d2e0655e 1448{
b63b70d8
SP		 1449 /* neigh reference count was dropprd to zero */
1450 struct net_device *dev = neigh->dev;
c1048aff 1451 struct ipoib_dev_priv *priv = ipoib_priv(dev);
2745b5b7 1452 struct sk_buff *skb;
b63b70d8
SP		 1453 if (neigh->ah)
1454 ipoib_put_ah(neigh->ah);
2745b5b7 1455 while ((skb = __skb_dequeue(&neigh->queue))) {
de903512 1456 ++dev->stats.tx_dropped;
2745b5b7
MT		 1457 dev_kfree_skb_any(skb);
1458 }
839fcaba
MT		 1459 if (ipoib_cm_get(neigh))
1460 ipoib_cm_destroy_tx(ipoib_cm_get(neigh));
c1048aff 1461 ipoib_dbg(ipoib_priv(dev),
b63b70d8
SP		 1462 "neigh free for %06x %pI6\n",
1463 IPOIB_QPN(neigh->daddr),
1464 neigh->daddr + 4);
d2e0655e 1465 kfree(neigh);
b63b70d8
SP		 1466 if (atomic_dec_and_test(&priv->ntbl.entries)) {
1467 if (test_bit(IPOIB_NEIGH_TBL_FLUSH, &priv->flags))
1468 complete(&priv->ntbl.flushed);
1469 }
1470}
1471
1472static void ipoib_neigh_reclaim(struct rcu_head *rp)
1473{
1474 /* Called as a result of removal from hash table */
1475 struct ipoib_neigh *neigh = container_of(rp, struct ipoib_neigh, rcu);
1476 /* note TX context may hold another ref */
1477 ipoib_neigh_put(neigh);
d2e0655e
MT		 1478}
1479
b63b70d8 1480void ipoib_neigh_free(struct ipoib_neigh *neigh)
1da177e4 1481{
b63b70d8 1482 struct net_device *dev = neigh->dev;
c1048aff 1483 struct ipoib_dev_priv *priv = ipoib_priv(dev);
b63b70d8
SP		 1484 struct ipoib_neigh_table *ntbl = &priv->ntbl;
1485 struct ipoib_neigh_hash *htbl;
1486 struct ipoib_neigh __rcu **np;
1487 struct ipoib_neigh *n;
1488 u32 hash_val;
1489
b63b70d8 1490 htbl = rcu_dereference_protected(ntbl->htbl,
b5120a6e 1491 lockdep_is_held(&priv->lock));
b63b70d8 1492 if (!htbl)
b5120a6e 1493 return;
b63b70d8
SP		 1494
1495 hash_val = ipoib_addr_hash(htbl, neigh->daddr);
1496 np = &htbl->buckets[hash_val];
1497 for (n = rcu_dereference_protected(*np,
b5120a6e 1498 lockdep_is_held(&priv->lock));
b63b70d8 1499 n != NULL;
6c723a68 1500 n = rcu_dereference_protected(*np,
b5120a6e 1501 lockdep_is_held(&priv->lock))) {
b63b70d8
SP		 1502 if (n == neigh) {
1503 /* found */
1504 rcu_assign_pointer(*np,
1505 rcu_dereference_protected(neigh->hnext,
b5120a6e 1506 lockdep_is_held(&priv->lock)));
49b8e744 1507 /* remove from parent list */
c586071d 1508 list_del_init(&neigh->list);
b63b70d8 1509 call_rcu(&neigh->rcu, ipoib_neigh_reclaim);
b5120a6e 1510 return;
b63b70d8
SP		 1511 } else {
1512 np = &n->hnext;
1513 }
1514 }
b63b70d8
SP		 1515}
1516
1517static int ipoib_neigh_hash_init(struct ipoib_dev_priv *priv)
1518{
1519 struct ipoib_neigh_table *ntbl = &priv->ntbl;
1520 struct ipoib_neigh_hash *htbl;
52374967 1521 struct ipoib_neigh __rcu **buckets;
b63b70d8
SP		 1522 u32 size;
1523
1524 clear_bit(IPOIB_NEIGH_TBL_FLUSH, &priv->flags);
1525 ntbl->htbl = NULL;
b63b70d8
SP		 1526 htbl = kzalloc(sizeof(*htbl), GFP_KERNEL);
1527 if (!htbl)
1528 return -ENOMEM;
b63b70d8 1529 size = roundup_pow_of_two(arp_tbl.gc_thresh3);
259e1914 1530 buckets = kvcalloc(size, sizeof(*buckets), GFP_KERNEL);
b63b70d8
SP		 1531 if (!buckets) {
1532 kfree(htbl);
1533 return -ENOMEM;
1534 }
1535 htbl->size = size;
1536 htbl->mask = (size - 1);
1537 htbl->buckets = buckets;
52374967 1538 RCU_INIT_POINTER(ntbl->htbl, htbl);
66172c09 1539 htbl->ntbl = ntbl;
b63b70d8
SP		 1540 atomic_set(&ntbl->entries, 0);
1541
1542 /* start garbage collection */
0b39578b 1543 queue_delayed_work(priv->wq, &priv->neigh_reap_task,
b63b70d8 1544 arp_tbl.gc_interval);
1da177e4
LT		 1545
1546 return 0;
1547}
1548
b63b70d8
SP		 1549static void neigh_hash_free_rcu(struct rcu_head *head)
1550{
1551 struct ipoib_neigh_hash *htbl = container_of(head,
1552 struct ipoib_neigh_hash,
1553 rcu);
1554 struct ipoib_neigh __rcu **buckets = htbl->buckets;
66172c09 1555 struct ipoib_neigh_table *ntbl = htbl->ntbl;
b63b70d8 1556
259e1914 1557 kvfree(buckets);
b63b70d8 1558 kfree(htbl);
66172c09 1559 complete(&ntbl->deleted);
b63b70d8
SP		 1560}
1561
1562void ipoib_del_neighs_by_gid(struct net_device *dev, u8 *gid)
1563{
c1048aff 1564 struct ipoib_dev_priv *priv = ipoib_priv(dev);
b63b70d8
SP		 1565 struct ipoib_neigh_table *ntbl = &priv->ntbl;
1566 struct ipoib_neigh_hash *htbl;
1567 unsigned long flags;
1568 int i;
1569
1570 /* remove all neigh connected to a given path or mcast */
b5120a6e 1571 spin_lock_irqsave(&priv->lock, flags);
b63b70d8
SP		 1572
1573 htbl = rcu_dereference_protected(ntbl->htbl,
b5120a6e 1574 lockdep_is_held(&priv->lock));
b63b70d8
SP		 1575
1576 if (!htbl)
1577 goto out_unlock;
1578
1579 for (i = 0; i < htbl->size; i++) {
1580 struct ipoib_neigh *neigh;
1581 struct ipoib_neigh __rcu **np = &htbl->buckets[i];
1582
1583 while ((neigh = rcu_dereference_protected(*np,
b5120a6e 1584 lockdep_is_held(&priv->lock))) != NULL) {
b63b70d8
SP		 1585 /* delete neighs belong to this parent */
1586 if (!memcmp(gid, neigh->daddr + 4, sizeof (union ib_gid))) {
1587 rcu_assign_pointer(*np,
1588 rcu_dereference_protected(neigh->hnext,
b5120a6e 1589 lockdep_is_held(&priv->lock)));
b63b70d8 1590 /* remove from parent list */
c586071d 1591 list_del_init(&neigh->list);
b63b70d8
SP		 1592 call_rcu(&neigh->rcu, ipoib_neigh_reclaim);
1593 } else {
1594 np = &neigh->hnext;
1595 }
1596
1597 }
1598 }
1599out_unlock:
b5120a6e 1600 spin_unlock_irqrestore(&priv->lock, flags);
b63b70d8
SP		 1601}
1602
1603static void ipoib_flush_neighs(struct ipoib_dev_priv *priv)
1604{
1605 struct ipoib_neigh_table *ntbl = &priv->ntbl;
1606 struct ipoib_neigh_hash *htbl;
1607 unsigned long flags;
66172c09 1608 int i, wait_flushed = 0;
b63b70d8 1609
66172c09 1610 init_completion(&priv->ntbl.flushed);
d2e46fcc 1611 set_bit(IPOIB_NEIGH_TBL_FLUSH, &priv->flags);
b63b70d8 1612
b5120a6e 1613 spin_lock_irqsave(&priv->lock, flags);
b63b70d8
SP		 1614
1615 htbl = rcu_dereference_protected(ntbl->htbl,
b5120a6e 1616 lockdep_is_held(&priv->lock));
b63b70d8
SP		 1617 if (!htbl)
1618 goto out_unlock;
1619
66172c09
SP		 1620 wait_flushed = atomic_read(&priv->ntbl.entries);
1621 if (!wait_flushed)
1622 goto free_htbl;
1623
b63b70d8
SP		 1624 for (i = 0; i < htbl->size; i++) {
1625 struct ipoib_neigh *neigh;
1626 struct ipoib_neigh __rcu **np = &htbl->buckets[i];
1627
1628 while ((neigh = rcu_dereference_protected(*np,
b5120a6e 1629 lockdep_is_held(&priv->lock))) != NULL) {
b63b70d8
SP		 1630 rcu_assign_pointer(*np,
1631 rcu_dereference_protected(neigh->hnext,
b5120a6e 1632 lockdep_is_held(&priv->lock)));
b63b70d8 1633 /* remove from path/mc list */
c586071d 1634 list_del_init(&neigh->list);
b63b70d8
SP		 1635 call_rcu(&neigh->rcu, ipoib_neigh_reclaim);
1636 }
1637 }
1638
66172c09 1639free_htbl:
b63b70d8
SP		 1640 rcu_assign_pointer(ntbl->htbl, NULL);
1641 call_rcu(&htbl->rcu, neigh_hash_free_rcu);
1642
1643out_unlock:
b5120a6e 1644 spin_unlock_irqrestore(&priv->lock, flags);
66172c09
SP		 1645 if (wait_flushed)
1646 wait_for_completion(&priv->ntbl.flushed);
b63b70d8
SP		 1647}
1648
1649static void ipoib_neigh_hash_uninit(struct net_device *dev)
1650{
c1048aff 1651 struct ipoib_dev_priv *priv = ipoib_priv(dev);
b63b70d8 1652
0dd9ce18 1653 ipoib_dbg(priv, "%s\n", __func__);
66172c09 1654 init_completion(&priv->ntbl.deleted);
b63b70d8 1655
cda8daf1 1656 cancel_delayed_work_sync(&priv->neigh_reap_task);
b63b70d8 1657
66172c09
SP		 1658 ipoib_flush_neighs(priv);
1659
1660 wait_for_completion(&priv->ntbl.deleted);
b63b70d8
SP		 1661}
1662
8966e28d
ES		 1663static void ipoib_napi_add(struct net_device *dev)
1664{
1665 struct ipoib_dev_priv *priv = ipoib_priv(dev);
1666
1667 netif_napi_add(dev, &priv->recv_napi, ipoib_rx_poll, IPOIB_NUM_WC);
1668 netif_napi_add(dev, &priv->send_napi, ipoib_tx_poll, MAX_SEND_CQE);
1669}
1670
1671static void ipoib_napi_del(struct net_device *dev)
1672{
1673 struct ipoib_dev_priv *priv = ipoib_priv(dev);
1674
1675 netif_napi_del(&priv->recv_napi);
1676 netif_napi_del(&priv->send_napi);
1677}
1678
0a1a9726 1679static void ipoib_dev_uninit_default(struct net_device *dev)
515ed4f3 1680{
c1048aff 1681 struct ipoib_dev_priv *priv = ipoib_priv(dev);
b63b70d8 1682
515ed4f3
ES		 1683 ipoib_transport_dev_cleanup(dev);
1684
8966e28d 1685 ipoib_napi_del(dev);
b53d4566 1686
515ed4f3
ES		 1687 ipoib_cm_dev_cleanup(dev);
1688
1689 kfree(priv->rx_ring);
1690 vfree(priv->tx_ring);
1691
1692 priv->rx_ring = NULL;
1693 priv->tx_ring = NULL;
1694}
1695
cd565b4b 1696static int ipoib_dev_init_default(struct net_device *dev)
1da177e4 1697{
c1048aff 1698 struct ipoib_dev_priv *priv = ipoib_priv(dev);
1da177e4 1699
8966e28d 1700 ipoib_napi_add(dev);
cd565b4b 1701
1da177e4 1702 /* Allocate RX/TX "rings" to hold queued skbs */
6396bb22
KC		 1703 priv->rx_ring = kcalloc(ipoib_recvq_size,
1704 sizeof(*priv->rx_ring),
1705 GFP_KERNEL);
74226649 1706 if (!priv->rx_ring)
be7aa663 1707 goto out;
1da177e4 1708
fad953ce
KC		 1709 priv->tx_ring = vzalloc(array_size(ipoib_sendq_size,
1710 sizeof(*priv->tx_ring)));
1da177e4 1711 if (!priv->tx_ring) {
c55359a2
YS		 1712 pr_warn("%s: failed to allocate TX ring (%d entries)\n",
1713 priv->ca->name, ipoib_sendq_size);
1da177e4
LT		 1714 goto out_rx_ring_cleanup;
1715 }
1da177e4 1716
1acba6a8 1717 /* priv->tx_head, tx_tail and global_tx_tail/head are already 0 */
1da177e4 1718
cd565b4b
ES		 1719 if (ipoib_transport_dev_init(dev, priv->ca)) {
1720 pr_warn("%s: ipoib_transport_dev_init failed\n",
1721 priv->ca->name);
1da177e4 1722 goto out_tx_ring_cleanup;
515ed4f3
ES		 1723 }
1724
cd565b4b
ES		 1725 /* after qp created set dev address */
1726 priv->dev->dev_addr[1] = (priv->qp->qp_num >> 16) & 0xff;
1727 priv->dev->dev_addr[2] = (priv->qp->qp_num >> 8) & 0xff;
1728 priv->dev->dev_addr[3] = (priv->qp->qp_num) & 0xff;
1729
515ed4f3
ES		 1730 return 0;
1731
1732out_tx_ring_cleanup:
1733 vfree(priv->tx_ring);
1734
1735out_rx_ring_cleanup:
1736 kfree(priv->rx_ring);
1737
1738out:
8966e28d 1739 ipoib_napi_del(dev);
515ed4f3
ES		 1740 return -ENOMEM;
1741}
1742
31a82362
FD		 1743static int ipoib_ioctl(struct net_device *dev, struct ifreq *ifr,
1744 int cmd)
1745{
1746 struct ipoib_dev_priv *priv = ipoib_priv(dev);
1747
a7605370 1748 if (!priv->rn_ops->ndo_eth_ioctl)
31a82362
FD		 1749 return -EOPNOTSUPP;
1750
a7605370 1751 return priv->rn_ops->ndo_eth_ioctl(dev, ifr, cmd);
31a82362
FD		 1752}
1753
eaeb3984 1754static int ipoib_dev_init(struct net_device *dev)
515ed4f3 1755{
c1048aff 1756 struct ipoib_dev_priv *priv = ipoib_priv(dev);
515ed4f3
ES		 1757 int ret = -ENOMEM;
1758
515ed4f3 1759 priv->qp = NULL;
1da177e4 1760
be7aa663 1761 /*
515ed4f3
ES		 1762 * the various IPoIB tasks assume they will never race against
1763 * themselves, so always use a single thread workqueue
be7aa663 1764 */
515ed4f3
ES		 1765 priv->wq = alloc_ordered_workqueue("ipoib_wq", WQ_MEM_RECLAIM);
1766 if (!priv->wq) {
1767 pr_warn("%s: failed to allocate device WQ\n", dev->name);
1768 goto out;
1769 }
1770
1771 /* create pd, which used both for control and datapath*/
1772 priv->pd = ib_alloc_pd(priv->ca, 0);
1773 if (IS_ERR(priv->pd)) {
eaeb3984 1774 pr_warn("%s: failed to allocate PD\n", priv->ca->name);
515ed4f3
ES		 1775 goto clean_wq;
1776 }
1777
cd565b4b 1778 ret = priv->rn_ops->ndo_init(dev);
515ed4f3
ES		 1779 if (ret) {
1780 pr_warn("%s failed to init HW resource\n", dev->name);
1781 goto out_free_pd;
1782 }
1783
99a7e2bf
WY		 1784 ret = ipoib_neigh_hash_init(priv);
1785 if (ret) {
515ed4f3 1786 pr_warn("%s failed to init neigh hash\n", dev->name);
be7aa663 1787 goto out_dev_uninit;
515ed4f3
ES		 1788 }
1789
1790 if (dev->flags & IFF_UP) {
1791 if (ipoib_ib_dev_open(dev)) {
1792 pr_warn("%s failed to open device\n", dev->name);
1793 ret = -ENODEV;
cda8daf1 1794 goto out_hash_uninit;
515ed4f3
ES		 1795 }
1796 }
be7aa663 1797
1da177e4
LT		 1798 return 0;
1799
cda8daf1
ES		 1800out_hash_uninit:
1801 ipoib_neigh_hash_uninit(dev);
1802
be7aa663
DL		 1803out_dev_uninit:
1804 ipoib_ib_dev_cleanup(dev);
1805
515ed4f3
ES		 1806out_free_pd:
1807 if (priv->pd) {
1808 ib_dealloc_pd(priv->pd);
1809 priv->pd = NULL;
1810 }
1da177e4 1811
515ed4f3
ES		 1812clean_wq:
1813 if (priv->wq) {
1814 destroy_workqueue(priv->wq);
1815 priv->wq = NULL;
1816 }
1da177e4
LT		 1817
1818out:
515ed4f3 1819 return ret;
1da177e4
LT		 1820}
1821
7cbee87c
JG		 1822/*
1823 * This must be called before doing an unregister_netdev on a parent device to
1824 * shutdown the IB event handler.
1825 */
1826static void ipoib_parent_unregister_pre(struct net_device *ndev)
1827{
1828 struct ipoib_dev_priv *priv = ipoib_priv(ndev);
1829
1830 /*
1831 * ipoib_set_mac checks netif_running before pushing work, clearing
1832 * running ensures the it will not add more work.
1833 */
1834 rtnl_lock();
567c5e13 1835 dev_change_flags(priv->dev, priv->dev->flags & ~IFF_UP, NULL);
7cbee87c
JG		 1836 rtnl_unlock();
1837
1838 /* ipoib_event() cannot be running once this returns */
1839 ib_unregister_event_handler(&priv->event_handler);
1840
1841 /*
1842 * Work on the queue grabs the rtnl lock, so this cannot be done while
1843 * also holding it.
1844 */
1845 flush_workqueue(ipoib_workqueue);
1846}
1847
eaeb3984
JG		 1848static void ipoib_set_dev_features(struct ipoib_dev_priv *priv)
1849{
1850 priv->hca_caps = priv->ca->attrs.device_cap_flags;
1851
1852 if (priv->hca_caps & IB_DEVICE_UD_IP_CSUM) {
1853 priv->dev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_RXCSUM;
1854
1855 if (priv->hca_caps & IB_DEVICE_UD_TSO)
1856 priv->dev->hw_features |= NETIF_F_TSO;
1857
1858 priv->dev->features |= priv->dev->hw_features;
1859 }
1860}
1861
1862static int ipoib_parent_init(struct net_device *ndev)
1863{
1864 struct ipoib_dev_priv *priv = ipoib_priv(ndev);
1865 struct ib_port_attr attr;
1866 int result;
1867
1868 result = ib_query_port(priv->ca, priv->port, &attr);
1869 if (result) {
1870 pr_warn("%s: ib_query_port %d failed\n", priv->ca->name,
1871 priv->port);
1872 return result;
1873 }
6d72344c 1874 priv->max_ib_mtu = rdma_mtu_from_attr(priv->ca, priv->port, &attr);
eaeb3984
JG		 1875
1876 result = ib_query_pkey(priv->ca, priv->port, 0, &priv->pkey);
1877 if (result) {
1878 pr_warn("%s: ib_query_pkey port %d failed (ret = %d)\n",
1879 priv->ca->name, priv->port, result);
1880 return result;
1881 }
1882
1883 result = rdma_query_gid(priv->ca, priv->port, 0, &priv->local_gid);
1884 if (result) {
1885 pr_warn("%s: rdma_query_gid port %d failed (ret = %d)\n",
1886 priv->ca->name, priv->port, result);
1887 return result;
1888 }
1889 memcpy(priv->dev->dev_addr + 4, priv->local_gid.raw,
1890 sizeof(union ib_gid));
1891
1892 SET_NETDEV_DEV(priv->dev, priv->ca->dev.parent);
9b8b2a32
AM		 1893 priv->dev->dev_port = priv->port - 1;
1894 /* Let's set this one too for backwards compatibility. */
eaeb3984
JG		 1895 priv->dev->dev_id = priv->port - 1;
1896
1897 return 0;
1898}
1899
1900static void ipoib_child_init(struct net_device *ndev)
1901{
1902 struct ipoib_dev_priv *priv = ipoib_priv(ndev);
1903 struct ipoib_dev_priv *ppriv = ipoib_priv(priv->parent);
1904
1905 priv->max_ib_mtu = ppriv->max_ib_mtu;
1906 set_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags);
87fb5c1c
MG		 1907 if (memchr_inv(priv->dev->dev_addr, 0, INFINIBAND_ALEN))
1908 memcpy(&priv->local_gid, priv->dev->dev_addr + 4,
1909 sizeof(priv->local_gid));
1910 else {
1911 memcpy(priv->dev->dev_addr, ppriv->dev->dev_addr,
1912 INFINIBAND_ALEN);
1913 memcpy(&priv->local_gid, &ppriv->local_gid,
1914 sizeof(priv->local_gid));
1915 }
eaeb3984
JG		 1916}
1917
1918static int ipoib_ndo_init(struct net_device *ndev)
1919{
1920 struct ipoib_dev_priv *priv = ipoib_priv(ndev);
1921 int rc;
b7e159eb 1922 struct rdma_netdev *rn = netdev_priv(ndev);
eaeb3984
JG		 1923
1924 if (priv->parent) {
1925 ipoib_child_init(ndev);
1926 } else {
1927 rc = ipoib_parent_init(ndev);
1928 if (rc)
1929 return rc;
1930 }
1931
1932 /* MTU will be reset when mcast join happens */
1933 ndev->mtu = IPOIB_UD_MTU(priv->max_ib_mtu);
1934 priv->mcast_mtu = priv->admin_mtu = ndev->mtu;
b7e159eb 1935 rn->mtu = priv->mcast_mtu;
eaeb3984
JG		 1936 ndev->max_mtu = IPOIB_CM_MTU;
1937
1938 ndev->neigh_priv_len = sizeof(struct ipoib_neigh);
1939
1940 /*
1941 * Set the full membership bit, so that we join the right
1942 * broadcast group, etc.
1943 */
1944 priv->pkey |= 0x8000;
1945
1946 ndev->broadcast[8] = priv->pkey >> 8;
1947 ndev->broadcast[9] = priv->pkey & 0xff;
1948 set_bit(IPOIB_FLAG_DEV_ADDR_SET, &priv->flags);
1949
1950 ipoib_set_dev_features(priv);
1951
1952 rc = ipoib_dev_init(ndev);
1953 if (rc) {
1954 pr_warn("%s: failed to initialize device: %s port %d (ret = %d)\n",
1955 priv->ca->name, priv->dev->name, priv->port, rc);
91b01061
VF		 1956 return rc;
1957 }
1958
1959 if (priv->parent) {
1960 struct ipoib_dev_priv *ppriv = ipoib_priv(priv->parent);
1961
1962 dev_hold(priv->parent);
1963
1964 down_write(&ppriv->vlan_rwsem);
1965 list_add_tail(&priv->list, &ppriv->child_intfs);
1966 up_write(&ppriv->vlan_rwsem);
eaeb3984
JG		 1967 }
1968
1969 return 0;
1970}
1971
7cbee87c 1972static void ipoib_ndo_uninit(struct net_device *dev)
1da177e4 1973{
25405d98 1974 struct ipoib_dev_priv *priv = ipoib_priv(dev);
9baa0b03
OG		 1975
1976 ASSERT_RTNL();
1da177e4 1977
25405d98
JG		 1978 /*
1979 * ipoib_remove_one guarantees the children are removed before the
1980 * parent, and that is the only place where a parent can be removed.
1981 */
1982 WARN_ON(!list_empty(&priv->child_intfs));
1da177e4 1983
91b01061
VF		 1984 if (priv->parent) {
1985 struct ipoib_dev_priv *ppriv = ipoib_priv(priv->parent);
1986
1987 down_write(&ppriv->vlan_rwsem);
1988 list_del(&priv->list);
1989 up_write(&ppriv->vlan_rwsem);
1990 }
1991
be7aa663
DL		 1992 ipoib_neigh_hash_uninit(dev);
1993
1da177e4
LT		 1994 ipoib_ib_dev_cleanup(dev);
1995
515ed4f3
ES		 1996 /* no more works over the priv->wq */
1997 if (priv->wq) {
65936bf2
JG		 1998 /* See ipoib_mcast_carrier_on_task() */
1999 WARN_ON(test_bit(IPOIB_FLAG_OPER_UP, &priv->flags));
515ed4f3
ES		 2000 flush_workqueue(priv->wq);
2001 destroy_workqueue(priv->wq);
2002 priv->wq = NULL;
2003 }
13476d35 2004
91b01061 2005 if (priv->parent)
13476d35 2006 dev_put(priv->parent);
1da177e4
LT		 2007}
2008
9c3c5f8e
EC		 2009static int ipoib_set_vf_link_state(struct net_device *dev, int vf, int link_state)
2010{
c1048aff 2011 struct ipoib_dev_priv *priv = ipoib_priv(dev);
9c3c5f8e
EC		 2012
2013 return ib_set_vf_link_state(priv->ca, vf, priv->port, link_state);
2014}
2015
2016static int ipoib_get_vf_config(struct net_device *dev, int vf,
2017 struct ifla_vf_info *ivf)
2018{
c1048aff 2019 struct ipoib_dev_priv *priv = ipoib_priv(dev);
9c3c5f8e
EC		 2020 int err;
2021
2022 err = ib_get_vf_config(priv->ca, vf, priv->port, ivf);
2023 if (err)
2024 return err;
2025
2026 ivf->vf = vf;
64d701c6 2027 memcpy(ivf->mac, dev->dev_addr, dev->addr_len);
9c3c5f8e
EC		 2028
2029 return 0;
2030}
2031
2032static int ipoib_set_vf_guid(struct net_device *dev, int vf, u64 guid, int type)
2033{
c1048aff 2034 struct ipoib_dev_priv *priv = ipoib_priv(dev);
9c3c5f8e
EC		 2035
2036 if (type != IFLA_VF_IB_NODE_GUID && type != IFLA_VF_IB_PORT_GUID)
2037 return -EINVAL;
2038
2039 return ib_set_vf_guid(priv->ca, vf, priv->port, guid, type);
2040}
2041
2446887e
DG		 2042static int ipoib_get_vf_guid(struct net_device *dev, int vf,
2043 struct ifla_vf_guid *node_guid,
2044 struct ifla_vf_guid *port_guid)
2045{
2046 struct ipoib_dev_priv *priv = ipoib_priv(dev);
2047
2048 return ib_get_vf_guid(priv->ca, vf, priv->port, node_guid, port_guid);
2049}
2050
9c3c5f8e
EC		 2051static int ipoib_get_vf_stats(struct net_device *dev, int vf,
2052 struct ifla_vf_stats *vf_stats)
2053{
c1048aff 2054 struct ipoib_dev_priv *priv = ipoib_priv(dev);
9c3c5f8e
EC		 2055
2056 return ib_get_vf_stats(priv->ca, vf, priv->port, vf_stats);
2057}
2058
3b04ddde
SH		 2059static const struct header_ops ipoib_header_ops = {
2060 .create = ipoib_hard_header,
2061};
2062
9c3c5f8e 2063static const struct net_device_ops ipoib_netdev_ops_pf = {
eaeb3984 2064 .ndo_init = ipoib_ndo_init,
7cbee87c 2065 .ndo_uninit = ipoib_ndo_uninit,
9c3c5f8e
EC		 2066 .ndo_open = ipoib_open,
2067 .ndo_stop = ipoib_stop,
2068 .ndo_change_mtu = ipoib_change_mtu,
2069 .ndo_fix_features = ipoib_fix_features,
2070 .ndo_start_xmit = ipoib_start_xmit,
2071 .ndo_tx_timeout = ipoib_timeout,
2072 .ndo_set_rx_mode = ipoib_set_mcast_list,
2073 .ndo_get_iflink = ipoib_get_iflink,
2074 .ndo_set_vf_link_state = ipoib_set_vf_link_state,
2075 .ndo_get_vf_config = ipoib_get_vf_config,
2076 .ndo_get_vf_stats = ipoib_get_vf_stats,
2446887e 2077 .ndo_get_vf_guid = ipoib_get_vf_guid,
9c3c5f8e 2078 .ndo_set_vf_guid = ipoib_set_vf_guid,
492a7e67 2079 .ndo_set_mac_address = ipoib_set_mac,
b6c871e5 2080 .ndo_get_stats64 = ipoib_get_stats,
a7605370 2081 .ndo_eth_ioctl = ipoib_ioctl,
9c3c5f8e
EC		 2082};
2083
2084static const struct net_device_ops ipoib_netdev_ops_vf = {
eaeb3984 2085 .ndo_init = ipoib_ndo_init,
7cbee87c 2086 .ndo_uninit = ipoib_ndo_uninit,
fe8114e8
SH		 2087 .ndo_open = ipoib_open,
2088 .ndo_stop = ipoib_stop,
2089 .ndo_change_mtu = ipoib_change_mtu,
3d96c74d 2090 .ndo_fix_features = ipoib_fix_features,
fe8114e8
SH		 2091 .ndo_start_xmit = ipoib_start_xmit,
2092 .ndo_tx_timeout = ipoib_timeout,
afc4b13d 2093 .ndo_set_rx_mode = ipoib_set_mcast_list,
5aa7add8 2094 .ndo_get_iflink = ipoib_get_iflink,
eb54714d 2095 .ndo_get_stats64 = ipoib_get_stats,
a7605370 2096 .ndo_eth_ioctl = ipoib_ioctl,
fe8114e8
SH		 2097};
2098
8f149b68
GL		 2099static const struct net_device_ops ipoib_netdev_default_pf = {
2100 .ndo_init = ipoib_dev_init_default,
2101 .ndo_uninit = ipoib_dev_uninit_default,
2102 .ndo_open = ipoib_ib_dev_open_default,
2103 .ndo_stop = ipoib_ib_dev_stop_default,
2104};
2105
cd565b4b 2106void ipoib_setup_common(struct net_device *dev)
1da177e4 2107{
2337f809 2108 dev->header_ops = &ipoib_header_ops;
8f149b68 2109 dev->netdev_ops = &ipoib_netdev_default_pf;
bea3348e 2110
82c24c18
EC		 2111 ipoib_set_ethtool_ops(dev);
2112
2337f809 2113 dev->watchdog_timeo = HZ;
1da177e4 2114
2337f809 2115 dev->flags |= IFF_BROADCAST | IFF_MULTICAST;
1da177e4 2116
fc791b63 2117 dev->hard_header_len = IPOIB_HARD_LEN;
2337f809
RD		 2118 dev->addr_len = INFINIBAND_ALEN;
2119 dev->type = ARPHRD_INFINIBAND;
2120 dev->tx_queue_len = ipoib_sendq_size * 2;
eb14032f 2121 dev->features = (NETIF_F_VLAN_CHALLENGED |
eb14032f 2122 NETIF_F_HIGHDMA);
02875878 2123 netif_keep_dst(dev);
1da177e4 2124
1da177e4 2125 memcpy(dev->broadcast, ipv4_bcast_addr, INFINIBAND_ALEN);
9f49a5b5
JG		 2126
2127 /*
2128 * unregister_netdev always frees the netdev, we use this mode
2129 * consistently to unify all the various unregister paths, including
2130 * those connected to rtnl_link_ops which require it.
2131 */
2132 dev->needs_free_netdev = true;
cd565b4b 2133}
1da177e4 2134
cd565b4b
ES		 2135static void ipoib_build_priv(struct net_device *dev)
2136{
2137 struct ipoib_dev_priv *priv = ipoib_priv(dev);
1da177e4 2138
cd565b4b 2139 priv->dev = dev;
1da177e4 2140 spin_lock_init(&priv->lock);
f47944cc 2141 init_rwsem(&priv->vlan_rwsem);
edf3f301 2142 mutex_init(&priv->mcast_mutex);
1da177e4
LT		 2143
2144 INIT_LIST_HEAD(&priv->path_list);
2145 INIT_LIST_HEAD(&priv->child_intfs);
2146 INIT_LIST_HEAD(&priv->dead_ahs);
2147 INIT_LIST_HEAD(&priv->multicast_list);
2148
c4028958 2149 INIT_DELAYED_WORK(&priv->mcast_task, ipoib_mcast_join_task);
e8224e4b 2150 INIT_WORK(&priv->carrier_on_task, ipoib_mcast_carrier_on_task);
ee1e2c82
MS		 2151 INIT_WORK(&priv->flush_light, ipoib_ib_dev_flush_light);
2152 INIT_WORK(&priv->flush_normal, ipoib_ib_dev_flush_normal);
2153 INIT_WORK(&priv->flush_heavy, ipoib_ib_dev_flush_heavy);
c4028958
DH		 2154 INIT_WORK(&priv->restart_task, ipoib_mcast_restart_task);
2155 INIT_DELAYED_WORK(&priv->ah_reap_task, ipoib_reap_ah);
b63b70d8 2156 INIT_DELAYED_WORK(&priv->neigh_reap_task, ipoib_reap_neigh);
1da177e4
LT		 2157}
2158
1fb7f897 2159static struct net_device *ipoib_alloc_netdev(struct ib_device *hca, u32 port,
5d6b0cb3 2160 const char *name)
cd565b4b
ES		 2161{
2162 struct net_device *dev;
2163
f6a8a19b
DD		 2164 dev = rdma_alloc_netdev(hca, port, RDMA_NETDEV_IPOIB, name,
2165 NET_NAME_UNKNOWN, ipoib_setup_common);
5d6b0cb3 2166 if (!IS_ERR(dev) || PTR_ERR(dev) != -EOPNOTSUPP)
f6a8a19b 2167 return dev;
cd565b4b 2168
5d6b0cb3
DD		 2169 dev = alloc_netdev(sizeof(struct rdma_netdev), name, NET_NAME_UNKNOWN,
2170 ipoib_setup_common);
2171 if (!dev)
2172 return ERR_PTR(-ENOMEM);
2173 return dev;
cd565b4b
ES		 2174}
2175
1fb7f897 2176int ipoib_intf_init(struct ib_device *hca, u32 port, const char *name,
5d6b0cb3 2177 struct net_device *dev)
cd565b4b 2178{
5d6b0cb3 2179 struct rdma_netdev *rn = netdev_priv(dev);
cd565b4b 2180 struct ipoib_dev_priv *priv;
5d6b0cb3 2181 int rc;
cd565b4b
ES		 2182
2183 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
2184 if (!priv)
5d6b0cb3 2185 return -ENOMEM;
cd565b4b 2186
eaeb3984
JG		 2187 priv->ca = hca;
2188 priv->port = port;
2189
5d6b0cb3
DD		 2190 rc = rdma_init_netdev(hca, port, RDMA_NETDEV_IPOIB, name,
2191 NET_NAME_UNKNOWN, ipoib_setup_common, dev);
2192 if (rc) {
2193 if (rc != -EOPNOTSUPP)
2194 goto out;
2195
5d6b0cb3
DD		 2196 rn->send = ipoib_send;
2197 rn->attach_mcast = ipoib_mcast_attach;
2198 rn->detach_mcast = ipoib_mcast_detach;
2199 rn->hca = hca;
2200 }
cd565b4b
ES		 2201
2202 priv->rn_ops = dev->netdev_ops;
2203
5d6b0cb3 2204 if (hca->attrs.device_cap_flags & IB_DEVICE_VIRTUAL_FUNCTION)
cd565b4b
ES		 2205 dev->netdev_ops = &ipoib_netdev_ops_vf;
2206 else
2207 dev->netdev_ops = &ipoib_netdev_ops_pf;
2208
cd565b4b 2209 rn->clnt_priv = priv;
9f49a5b5
JG		 2210 /*
2211 * Only the child register_netdev flows can handle priv_destructor
2212 * being set, so we force it to NULL here and handle manually until it
2213 * is safe to turn on.
2214 */
2215 priv->next_priv_destructor = dev->priv_destructor;
2216 dev->priv_destructor = NULL;
2217
cd565b4b
ES		 2218 ipoib_build_priv(dev);
2219
5d6b0cb3
DD		 2220 return 0;
2221
2222out:
cd565b4b 2223 kfree(priv);
5d6b0cb3
DD		 2224 return rc;
2225}
2226
1fb7f897 2227struct net_device *ipoib_intf_alloc(struct ib_device *hca, u32 port,
5d6b0cb3
DD		 2228 const char *name)
2229{
2230 struct net_device *dev;
2231 int rc;
2232
2233 dev = ipoib_alloc_netdev(hca, port, name);
2234 if (IS_ERR(dev))
2235 return dev;
2236
2237 rc = ipoib_intf_init(hca, port, name, dev);
2238 if (rc) {
2239 free_netdev(dev);
2240 return ERR_PTR(rc);
2241 }
2242
2243 /*
2244 * Upon success the caller must ensure ipoib_intf_free is called or
2245 * register_netdevice succeed'd and priv_destructor is set to
2246 * ipoib_intf_free.
2247 */
2248 return dev;
1da177e4
LT		 2249}
2250
9f49a5b5
JG		 2251void ipoib_intf_free(struct net_device *dev)
2252{
2253 struct ipoib_dev_priv *priv = ipoib_priv(dev);
2254 struct rdma_netdev *rn = netdev_priv(dev);
2255
2256 dev->priv_destructor = priv->next_priv_destructor;
2257 if (dev->priv_destructor)
2258 dev->priv_destructor(dev);
2259
2260 /*
2261 * There are some error flows around register_netdev failing that may
2262 * attempt to call priv_destructor twice, prevent that from happening.
2263 */
2264 dev->priv_destructor = NULL;
2265
2266 /* unregister/destroy is very complicated. Make bugs more obvious. */
2267 rn->clnt_priv = NULL;
2268
2269 kfree(priv);
2270}
2271
1f8f60f3
Y		 2272static ssize_t pkey_show(struct device *dev, struct device_attribute *attr,
2273 char *buf)
1da177e4 2274{
c1048aff
ES		 2275 struct net_device *ndev = to_net_dev(dev);
2276 struct ipoib_dev_priv *priv = ipoib_priv(ndev);
1da177e4 2277
1c7fd726 2278 return sysfs_emit(buf, "0x%04x\n", priv->pkey);
1da177e4 2279}
1f8f60f3 2280static DEVICE_ATTR_RO(pkey);
1da177e4 2281
1f8f60f3
Y		 2282static ssize_t umcast_show(struct device *dev, struct device_attribute *attr,
2283 char *buf)
335a64a5 2284{
c1048aff
ES		 2285 struct net_device *ndev = to_net_dev(dev);
2286 struct ipoib_dev_priv *priv = ipoib_priv(ndev);
335a64a5 2287
1c7fd726
JP		 2288 return sysfs_emit(buf, "%d\n",
2289 test_bit(IPOIB_FLAG_UMCAST, &priv->flags));
335a64a5
OG		 2290}
2291
862096a8 2292void ipoib_set_umcast(struct net_device *ndev, int umcast_val)
335a64a5 2293{
c1048aff 2294 struct ipoib_dev_priv *priv = ipoib_priv(ndev);
335a64a5
OG		 2295
2296 if (umcast_val > 0) {
2297 set_bit(IPOIB_FLAG_UMCAST, &priv->flags);
2298 ipoib_warn(priv, "ignoring multicast groups joined directly "
2299 "by userspace\n");
2300 } else
2301 clear_bit(IPOIB_FLAG_UMCAST, &priv->flags);
862096a8
OG		 2302}
2303
1f8f60f3
Y		 2304static ssize_t umcast_store(struct device *dev, struct device_attribute *attr,
2305 const char *buf, size_t count)
862096a8
OG		 2306{
2307 unsigned long umcast_val = simple_strtoul(buf, NULL, 0);
2308
2309 ipoib_set_umcast(to_net_dev(dev), umcast_val);
335a64a5
OG		 2310
2311 return count;
2312}
1f8f60f3 2313static DEVICE_ATTR_RW(umcast);
335a64a5
OG		 2314
2315int ipoib_add_umcast_attr(struct net_device *dev)
2316{
2317 return device_create_file(&dev->dev, &dev_attr_umcast);
2318}
2319
492a7e67
MB		 2320static void set_base_guid(struct ipoib_dev_priv *priv, union ib_gid *gid)
2321{
2322 struct ipoib_dev_priv *child_priv;
2323 struct net_device *netdev = priv->dev;
2324
9b29953b 2325 netif_addr_lock_bh(netdev);
492a7e67
MB		 2326
2327 memcpy(&priv->local_gid.global.interface_id,
2328 &gid->global.interface_id,
2329 sizeof(gid->global.interface_id));
2330 memcpy(netdev->dev_addr + 4, &priv->local_gid, sizeof(priv->local_gid));
2331 clear_bit(IPOIB_FLAG_DEV_ADDR_SET, &priv->flags);
2332
9b29953b 2333 netif_addr_unlock_bh(netdev);
492a7e67
MB		 2334
2335 if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) {
2336 down_read(&priv->vlan_rwsem);
2337 list_for_each_entry(child_priv, &priv->child_intfs, list)
2338 set_base_guid(child_priv, gid);
2339 up_read(&priv->vlan_rwsem);
2340 }
2341}
2342
2343static int ipoib_check_lladdr(struct net_device *dev,
2344 struct sockaddr_storage *ss)
2345{
2346 union ib_gid *gid = (union ib_gid *)(ss->__data + 4);
2347 int ret = 0;
2348
9b29953b 2349 netif_addr_lock_bh(dev);
492a7e67
MB		 2350
2351 /* Make sure the QPN, reserved and subnet prefix match the current
2352 * lladdr, it also makes sure the lladdr is unicast.
2353 */
2354 if (memcmp(dev->dev_addr, ss->__data,
2355 4 + sizeof(gid->global.subnet_prefix)) ||
2356 gid->global.interface_id == 0)
2357 ret = -EINVAL;
2358
9b29953b 2359 netif_addr_unlock_bh(dev);
492a7e67
MB		 2360
2361 return ret;
2362}
2363
2364static int ipoib_set_mac(struct net_device *dev, void *addr)
2365{
c1048aff 2366 struct ipoib_dev_priv *priv = ipoib_priv(dev);
492a7e67
MB		 2367 struct sockaddr_storage *ss = addr;
2368 int ret;
2369
2370 if (!(dev->priv_flags & IFF_LIVE_ADDR_CHANGE) && netif_running(dev))
2371 return -EBUSY;
2372
2373 ret = ipoib_check_lladdr(dev, ss);
2374 if (ret)
2375 return ret;
2376
2377 set_base_guid(priv, (union ib_gid *)(ss->__data + 4));
2378
2379 queue_work(ipoib_workqueue, &priv->flush_light);
2380
2381 return 0;
2382}
2383
1f8f60f3
Y		 2384static ssize_t create_child_store(struct device *dev,
2385 struct device_attribute *attr,
2386 const char *buf, size_t count)
1da177e4
LT		 2387{
2388 int pkey;
2389 int ret;
2390
2391 if (sscanf(buf, "%i", &pkey) != 1)
2392 return -EINVAL;
2393
3d790a4c 2394 if (pkey <= 0 || pkey > 0xffff || pkey == 0x8000)
1da177e4
LT		 2395 return -EINVAL;
2396
43cb76d9 2397 ret = ipoib_vlan_add(to_net_dev(dev), pkey);
1da177e4
LT		 2398
2399 return ret ? ret : count;
2400}
1f8f60f3 2401static DEVICE_ATTR_WO(create_child);
1da177e4 2402
1f8f60f3
Y		 2403static ssize_t delete_child_store(struct device *dev,
2404 struct device_attribute *attr,
2405 const char *buf, size_t count)
1da177e4
LT		 2406{
2407 int pkey;
2408 int ret;
2409
2410 if (sscanf(buf, "%i", &pkey) != 1)
2411 return -EINVAL;
2412
2413 if (pkey < 0 || pkey > 0xffff)
2414 return -EINVAL;
2415
43cb76d9 2416 ret = ipoib_vlan_delete(to_net_dev(dev), pkey);
1da177e4
LT		 2417
2418 return ret ? ret : count;
2419
2420}
1f8f60f3 2421static DEVICE_ATTR_WO(delete_child);
1da177e4
LT		 2422
2423int ipoib_add_pkey_attr(struct net_device *dev)
2424{
43cb76d9 2425 return device_create_file(&dev->dev, &dev_attr_pkey);
1da177e4
LT		 2426}
2427
f6350da4
AM		 2428/*
2429 * We erroneously exposed the iface's port number in the dev_id
2430 * sysfs field long after dev_port was introduced for that purpose[1],
2431 * and we need to stop everyone from relying on that.
2432 * Let's overload the shower routine for the dev_id file here
2433 * to gently bring the issue up.
2434 *
2435 * [1] https://www.spinics.net/lists/netdev/msg272123.html
2436 */
2437static ssize_t dev_id_show(struct device *dev,
2438 struct device_attribute *attr, char *buf)
2439{
2440 struct net_device *ndev = to_net_dev(dev);
2441
b79656ed
LR		 2442 /*
2443 * ndev->dev_port will be equal to 0 in old kernel prior to commit
2444 * 9b8b2a323008 ("IB/ipoib: Use dev_port to expose network interface
2445 * port numbers") Zero was chosen as special case for user space
2446 * applications to fallback and query dev_id to check if it has
2447 * different value or not.
2448 *
2449 * Don't print warning in such scenario.
2450 *
2451 * https://github.com/systemd/systemd/blob/master/src/udev/udev-builtin-net_id.c#L358
2452 */
2453 if (ndev->dev_port && ndev->dev_id == ndev->dev_port)
f6350da4
AM		 2454 netdev_info_once(ndev,
2455 "\"%s\" wants to know my dev_id. Should it look at dev_port instead? See Documentation/ABI/testing/sysfs-class-net for more info.\n",
2456 current->comm);
2457
1c7fd726 2458 return sysfs_emit(buf, "%#x\n", ndev->dev_id);
f6350da4
AM		 2459}
2460static DEVICE_ATTR_RO(dev_id);
2461
ed0bc265 2462static int ipoib_intercept_dev_id_attr(struct net_device *dev)
f6350da4
AM		 2463{
2464 device_remove_file(&dev->dev, &dev_attr_dev_id);
2465 return device_create_file(&dev->dev, &dev_attr_dev_id);
2466}
2467
1da177e4 2468static struct net_device *ipoib_add_port(const char *format,
1fb7f897 2469 struct ib_device *hca, u32 port)
1da177e4 2470{
5d6b0cb3
DD		 2471 struct rtnl_link_ops *ops = ipoib_get_link_ops();
2472 struct rdma_netdev_alloc_params params;
1da177e4 2473 struct ipoib_dev_priv *priv;
9f49a5b5 2474 struct net_device *ndev;
eaeb3984 2475 int result;
1da177e4 2476
5d6b0cb3
DD		 2477 ndev = ipoib_intf_alloc(hca, port, format);
2478 if (IS_ERR(ndev)) {
2479 pr_warn("%s, %d: ipoib_intf_alloc failed %ld\n", hca->name, port,
2480 PTR_ERR(ndev));
2481 return ndev;
1da177e4 2482 }
5d6b0cb3 2483 priv = ipoib_priv(ndev);
1da177e4
LT		 2484
2485 INIT_IB_EVENT_HANDLER(&priv->event_handler,
2486 priv->ca, ipoib_event);
dcc9881e 2487 ib_register_event_handler(&priv->event_handler);
1da177e4 2488
10293610
AE		 2489 /* call event handler to ensure pkey in sync */
2490 queue_work(ipoib_workqueue, &priv->flush_heavy);
2491
5ce2dced
KH		 2492 ndev->rtnl_link_ops = ipoib_get_link_ops();
2493
9f49a5b5 2494 result = register_netdev(ndev);
1da177e4 2495 if (result) {
c55359a2
YS		 2496 pr_warn("%s: couldn't register ipoib port %d; error %d\n",
2497 hca->name, port, result);
9f49a5b5
JG		 2498
2499 ipoib_parent_unregister_pre(ndev);
2500 ipoib_intf_free(ndev);
2501 free_netdev(ndev);
2502
2503 return ERR_PTR(result);
1da177e4
LT		 2504 }
2505
3023a1e9
KH		 2506 if (hca->ops.rdma_netdev_get_params) {
2507 int rc = hca->ops.rdma_netdev_get_params(hca, port,
5d6b0cb3
DD		 2508 RDMA_NETDEV_IPOIB,
2509 &params);
2510
2511 if (!rc && ops->priv_size < params.sizeof_priv)
2512 ops->priv_size = params.sizeof_priv;
2513 }
9f49a5b5
JG		 2514 /*
2515 * We cannot set priv_destructor before register_netdev because we
2516 * need priv to be always valid during the error flow to execute
2517 * ipoib_parent_unregister_pre(). Instead handle it manually and only
2518 * enter priv_destructor mode once we are completely registered.
2519 */
2520 ndev->priv_destructor = ipoib_intf_free;
2521
f6350da4
AM		 2522 if (ipoib_intercept_dev_id_attr(ndev))
2523 goto sysfs_failed;
9f49a5b5 2524 if (ipoib_cm_add_mode_attr(ndev))
839fcaba 2525 goto sysfs_failed;
9f49a5b5 2526 if (ipoib_add_pkey_attr(ndev))
1da177e4 2527 goto sysfs_failed;
9f49a5b5 2528 if (ipoib_add_umcast_attr(ndev))
335a64a5 2529 goto sysfs_failed;
9f49a5b5 2530 if (device_create_file(&ndev->dev, &dev_attr_create_child))
1da177e4 2531 goto sysfs_failed;
9f49a5b5 2532 if (device_create_file(&ndev->dev, &dev_attr_delete_child))
1da177e4
LT		 2533 goto sysfs_failed;
2534
9f49a5b5 2535 return ndev;
1da177e4
LT		 2536
2537sysfs_failed:
9f49a5b5
JG		 2538 ipoib_parent_unregister_pre(ndev);
2539 unregister_netdev(ndev);
2540 return ERR_PTR(-ENOMEM);
1da177e4
LT		 2541}
2542
11a0ae4c 2543static int ipoib_add_one(struct ib_device *device)
1da177e4
LT		 2544{
2545 struct list_head *dev_list;
2546 struct net_device *dev;
2547 struct ipoib_dev_priv *priv;
ea1075ed 2548 unsigned int p;
8e37ab68 2549 int count = 0;
07ebafba 2550
b1b63970 2551 dev_list = kmalloc(sizeof(*dev_list), GFP_KERNEL);
1da177e4 2552 if (!dev_list)
11a0ae4c 2553 return -ENOMEM;
1da177e4
LT		 2554
2555 INIT_LIST_HEAD(dev_list);
2556
ea1075ed 2557 rdma_for_each_port (device, p) {
8e37ab68 2558 if (!rdma_protocol_ib(device, p))
7b4c8769 2559 continue;
1da177e4
LT		 2560 dev = ipoib_add_port("ib%d", device, p);
2561 if (!IS_ERR(dev)) {
c1048aff 2562 priv = ipoib_priv(dev);
1da177e4 2563 list_add_tail(&priv->list, dev_list);
8e37ab68 2564 count++;
1da177e4
LT		 2565 }
2566 }
2567
8e37ab68 2568 if (!count) {
ac6dbf7f 2569 kfree(dev_list);
11a0ae4c 2570 return -EOPNOTSUPP;
8e37ab68
MW		 2571 }
2572
1da177e4 2573 ib_set_client_data(device, &ipoib_client, dev_list);
11a0ae4c 2574 return 0;
1da177e4
LT		 2575}
2576
7c1eb45a 2577static void ipoib_remove_one(struct ib_device *device, void *client_data)
1da177e4 2578{
25405d98 2579 struct ipoib_dev_priv *priv, *tmp, *cpriv, *tcpriv;
7c1eb45a 2580 struct list_head *dev_list = client_data;
1da177e4 2581
1da177e4 2582 list_for_each_entry_safe(priv, tmp, dev_list, list) {
25405d98 2583 LIST_HEAD(head);
7cbee87c 2584 ipoib_parent_unregister_pre(priv->dev);
1da177e4 2585
25405d98
JG		 2586 rtnl_lock();
2587
2588 list_for_each_entry_safe(cpriv, tcpriv, &priv->child_intfs,
2589 list)
2590 unregister_netdevice_queue(cpriv->dev, &head);
2591 unregister_netdevice_queue(priv->dev, &head);
2592 unregister_netdevice_many(&head);
2593
2594 rtnl_unlock();
1da177e4 2595 }
06c56e44
MT		 2596
2597 kfree(dev_list);
1da177e4
LT		 2598}
2599
771a5258
SR		 2600#ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
2601static struct notifier_block ipoib_netdev_notifier = {
2602 .notifier_call = ipoib_netdev_event,
2603};
2604#endif
2605
1da177e4
LT		 2606static int __init ipoib_init_module(void)
2607{
2608 int ret;
2609
0f485251
SM		 2610 ipoib_recvq_size = roundup_pow_of_two(ipoib_recvq_size);
2611 ipoib_recvq_size = min(ipoib_recvq_size, IPOIB_MAX_QUEUE_SIZE);
2612 ipoib_recvq_size = max(ipoib_recvq_size, IPOIB_MIN_QUEUE_SIZE);
2613
2614 ipoib_sendq_size = roundup_pow_of_two(ipoib_sendq_size);
2615 ipoib_sendq_size = min(ipoib_sendq_size, IPOIB_MAX_QUEUE_SIZE);
732eacc0 2616 ipoib_sendq_size = max3(ipoib_sendq_size, 2 * MAX_SEND_CQE, IPOIB_MIN_QUEUE_SIZE);
68e995a2
PS		 2617#ifdef CONFIG_INFINIBAND_IPOIB_CM
2618 ipoib_max_conn_qp = min(ipoib_max_conn_qp, IPOIB_CM_MAX_CONN_QP);
11f74b40 2619 ipoib_max_conn_qp = max(ipoib_max_conn_qp, 0);
68e995a2 2620#endif
0f485251 2621
f89271da
EC		 2622 /*
2623 * When copying small received packets, we only copy from the
2624 * linear data part of the SKB, so we rely on this condition.
2625 */
2626 BUILD_BUG_ON(IPOIB_CM_COPYBREAK > IPOIB_CM_HEAD_SIZE);
2627
2e061c69 2628 ipoib_register_debugfs();
1da177e4
LT		 2629
2630 /*
0b39578b
DL		 2631 * We create a global workqueue here that is used for all flush
2632 * operations. However, if you attempt to flush a workqueue
2633 * from a task on that same workqueue, it deadlocks the system.
2634 * We want to be able to flush the tasks associated with a
2635 * specific net device, so we also create a workqueue for each
2636 * netdevice. We queue up the tasks for that device only on
2637 * its private workqueue, and we only queue up flush events
2638 * on our global flush workqueue. This avoids the deadlocks.
1da177e4 2639 */
ee190ab7 2640 ipoib_workqueue = alloc_ordered_workqueue("ipoib_flush", 0);
1da177e4
LT		 2641 if (!ipoib_workqueue) {
2642 ret = -ENOMEM;
2643 goto err_fs;
2644 }
2645
c1a0b23b
MT		 2646 ib_sa_register_client(&ipoib_sa_client);
2647
1da177e4
LT		 2648 ret = ib_register_client(&ipoib_client);
2649 if (ret)
c1a0b23b 2650 goto err_sa;
1da177e4 2651
9baa0b03
OG		 2652 ret = ipoib_netlink_init();
2653 if (ret)
2654 goto err_client;
2655
771a5258
SR		 2656#ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
2657 register_netdevice_notifier(&ipoib_netdev_notifier);
2658#endif
1da177e4
LT		 2659 return 0;
2660
9baa0b03
OG		 2661err_client:
2662 ib_unregister_client(&ipoib_client);
2663
c1a0b23b
MT		 2664err_sa:
2665 ib_sa_unregister_client(&ipoib_sa_client);
1da177e4
LT		 2666 destroy_workqueue(ipoib_workqueue);
2667
9adec1a8
RD		 2668err_fs:
2669 ipoib_unregister_debugfs();
2670
1da177e4
LT		 2671 return ret;
2672}
2673
2674static void __exit ipoib_cleanup_module(void)
2675{
771a5258
SR		 2676#ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
2677 unregister_netdevice_notifier(&ipoib_netdev_notifier);
2678#endif
9baa0b03 2679 ipoib_netlink_fini();
1da177e4 2680 ib_unregister_client(&ipoib_client);
c1a0b23b 2681 ib_sa_unregister_client(&ipoib_sa_client);
9adec1a8 2682 ipoib_unregister_debugfs();
1da177e4
LT		 2683 destroy_workqueue(ipoib_workqueue);
2684}
2685
2686module_init(ipoib_init_module);
2687module_exit(ipoib_cleanup_module);
Linux 6.x block layer and io_uring tree(s)