Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Shared Memory Communications over RDMA (SMC-R) and RoCE
4 *
5 * Generic netlink support functions to configure an SMC-R PNET table
6 *
7 * Copyright IBM Corp. 2016
8 *
9 * Author(s): Thomas Richter <tmricht@linux.vnet.ibm.com>
10 */
11
12#include <linux/module.h>
13#include <linux/list.h>
14#include <linux/ctype.h>
15#include <linux/mutex.h>
16#include <net/netlink.h>
17#include <net/genetlink.h>
18
19#include <uapi/linux/if.h>
20#include <uapi/linux/smc.h>
21
22#include <rdma/ib_verbs.h>
23
24#include <net/netns/generic.h>
25#include "smc_netns.h"
26
27#include "smc_pnet.h"
28#include "smc_ib.h"
29#include "smc_ism.h"
30#include "smc_core.h"
31
32static struct net_device *__pnet_find_base_ndev(struct net_device *ndev);
33static struct net_device *pnet_find_base_ndev(struct net_device *ndev);
34
35static const struct nla_policy smc_pnet_policy[SMC_PNETID_MAX + 1] = {
36 [SMC_PNETID_NAME] = {
37 .type = NLA_NUL_STRING,
38 .len = SMC_MAX_PNETID_LEN
39 },
40 [SMC_PNETID_ETHNAME] = {
41 .type = NLA_NUL_STRING,
42 .len = IFNAMSIZ - 1
43 },
44 [SMC_PNETID_IBNAME] = {
45 .type = NLA_NUL_STRING,
46 .len = IB_DEVICE_NAME_MAX - 1
47 },
48 [SMC_PNETID_IBPORT] = { .type = NLA_U8 }
49};
50
51static struct genl_family smc_pnet_nl_family;
52
53enum smc_pnet_nametype {
54 SMC_PNET_ETH = 1,
55 SMC_PNET_IB = 2,
56};
57
58/* pnet entry stored in pnet table */
59struct smc_pnetentry {
60 struct list_head list;
61 char pnet_name[SMC_MAX_PNETID_LEN + 1];
62 enum smc_pnet_nametype type;
63 union {
64 struct {
65 char eth_name[IFNAMSIZ + 1];
66 struct net_device *ndev;
67 netdevice_tracker dev_tracker;
68 };
69 struct {
70 char ib_name[IB_DEVICE_NAME_MAX + 1];
71 u8 ib_port;
72 };
73 };
74};
75
76/* Check if the pnetid is set */
77bool smc_pnet_is_pnetid_set(u8 *pnetid)
78{
79 if (pnetid[0] == 0 || pnetid[0] == _S)
80 return false;
81 return true;
82}
83
84/* Check if two given pnetids match */
85static bool smc_pnet_match(u8 *pnetid1, u8 *pnetid2)
86{
87 int i;
88
89 for (i = 0; i < SMC_MAX_PNETID_LEN; i++) {
90 if ((pnetid1[i] == 0 || pnetid1[i] == _S) &&
91 (pnetid2[i] == 0 || pnetid2[i] == _S))
92 break;
93 if (pnetid1[i] != pnetid2[i])
94 return false;
95 }
96 return true;
97}
98
99/* Remove a pnetid from the pnet table.
100 */
101static int smc_pnet_remove_by_pnetid(struct net *net, char *pnet_name)
102{
103 struct smc_pnetentry *pnetelem, *tmp_pe;
104 struct smc_pnettable *pnettable;
105 struct smc_ib_device *ibdev;
106 struct smcd_dev *smcd;
107 struct smc_net *sn;
108 int rc = -ENOENT;
109 int ibport;
110
111 /* get pnettable for namespace */
112 sn = net_generic(net, smc_net_id);
113 pnettable = &sn->pnettable;
114
115 /* remove table entry */
116 mutex_lock(&pnettable->lock);
117 list_for_each_entry_safe(pnetelem, tmp_pe, &pnettable->pnetlist,
118 list) {
119 if (!pnet_name ||
120 smc_pnet_match(pnetelem->pnet_name, pnet_name)) {
121 list_del(&pnetelem->list);
122 if (pnetelem->type == SMC_PNET_ETH && pnetelem->ndev) {
123 netdev_put(pnetelem->ndev,
124 &pnetelem->dev_tracker);
125 pr_warn_ratelimited("smc: net device %s "
126 "erased user defined "
127 "pnetid %.16s\n",
128 pnetelem->eth_name,
129 pnetelem->pnet_name);
130 }
131 kfree(pnetelem);
132 rc = 0;
133 }
134 }
135 mutex_unlock(&pnettable->lock);
136
137 /* if this is not the initial namespace, stop here */
138 if (net != &init_net)
139 return rc;
140
141 /* remove ib devices */
142 mutex_lock(&smc_ib_devices.mutex);
143 list_for_each_entry(ibdev, &smc_ib_devices.list, list) {
144 for (ibport = 0; ibport < SMC_MAX_PORTS; ibport++) {
145 if (ibdev->pnetid_by_user[ibport] &&
146 (!pnet_name ||
147 smc_pnet_match(pnet_name,
148 ibdev->pnetid[ibport]))) {
149 pr_warn_ratelimited("smc: ib device %s ibport "
150 "%d erased user defined "
151 "pnetid %.16s\n",
152 ibdev->ibdev->name,
153 ibport + 1,
154 ibdev->pnetid[ibport]);
155 memset(ibdev->pnetid[ibport], 0,
156 SMC_MAX_PNETID_LEN);
157 ibdev->pnetid_by_user[ibport] = false;
158 rc = 0;
159 }
160 }
161 }
162 mutex_unlock(&smc_ib_devices.mutex);
163 /* remove smcd devices */
164 mutex_lock(&smcd_dev_list.mutex);
165 list_for_each_entry(smcd, &smcd_dev_list.list, list) {
166 if (smcd->pnetid_by_user &&
167 (!pnet_name ||
168 smc_pnet_match(pnet_name, smcd->pnetid))) {
169 pr_warn_ratelimited("smc: smcd device %s "
170 "erased user defined pnetid "
171 "%.16s\n",
172 dev_name(smcd->ops->get_dev(smcd)),
173 smcd->pnetid);
174 memset(smcd->pnetid, 0, SMC_MAX_PNETID_LEN);
175 smcd->pnetid_by_user = false;
176 rc = 0;
177 }
178 }
179 mutex_unlock(&smcd_dev_list.mutex);
180 return rc;
181}
182
183/* Add the reference to a given network device to the pnet table.
184 */
185static int smc_pnet_add_by_ndev(struct net_device *ndev)
186{
187 struct smc_pnetentry *pnetelem, *tmp_pe;
188 struct smc_pnettable *pnettable;
189 struct net *net = dev_net(ndev);
190 struct smc_net *sn;
191 int rc = -ENOENT;
192
193 /* get pnettable for namespace */
194 sn = net_generic(net, smc_net_id);
195 pnettable = &sn->pnettable;
196
197 mutex_lock(&pnettable->lock);
198 list_for_each_entry_safe(pnetelem, tmp_pe, &pnettable->pnetlist, list) {
199 if (pnetelem->type == SMC_PNET_ETH && !pnetelem->ndev &&
200 !strncmp(pnetelem->eth_name, ndev->name, IFNAMSIZ)) {
201 netdev_hold(ndev, &pnetelem->dev_tracker, GFP_ATOMIC);
202 pnetelem->ndev = ndev;
203 rc = 0;
204 pr_warn_ratelimited("smc: adding net device %s with "
205 "user defined pnetid %.16s\n",
206 pnetelem->eth_name,
207 pnetelem->pnet_name);
208 break;
209 }
210 }
211 mutex_unlock(&pnettable->lock);
212 return rc;
213}
214
215/* Remove the reference to a given network device from the pnet table.
216 */
217static int smc_pnet_remove_by_ndev(struct net_device *ndev)
218{
219 struct smc_pnetentry *pnetelem, *tmp_pe;
220 struct smc_pnettable *pnettable;
221 struct net *net = dev_net(ndev);
222 struct smc_net *sn;
223 int rc = -ENOENT;
224
225 /* get pnettable for namespace */
226 sn = net_generic(net, smc_net_id);
227 pnettable = &sn->pnettable;
228
229 mutex_lock(&pnettable->lock);
230 list_for_each_entry_safe(pnetelem, tmp_pe, &pnettable->pnetlist, list) {
231 if (pnetelem->type == SMC_PNET_ETH && pnetelem->ndev == ndev) {
232 netdev_put(pnetelem->ndev, &pnetelem->dev_tracker);
233 pnetelem->ndev = NULL;
234 rc = 0;
235 pr_warn_ratelimited("smc: removing net device %s with "
236 "user defined pnetid %.16s\n",
237 pnetelem->eth_name,
238 pnetelem->pnet_name);
239 break;
240 }
241 }
242 mutex_unlock(&pnettable->lock);
243 return rc;
244}
245
246/* Apply pnetid to ib device when no pnetid is set.
247 */
248static bool smc_pnet_apply_ib(struct smc_ib_device *ib_dev, u8 ib_port,
249 char *pnet_name)
250{
251 bool applied = false;
252
253 mutex_lock(&smc_ib_devices.mutex);
254 if (!smc_pnet_is_pnetid_set(ib_dev->pnetid[ib_port - 1])) {
255 memcpy(ib_dev->pnetid[ib_port - 1], pnet_name,
256 SMC_MAX_PNETID_LEN);
257 ib_dev->pnetid_by_user[ib_port - 1] = true;
258 applied = true;
259 }
260 mutex_unlock(&smc_ib_devices.mutex);
261 return applied;
262}
263
264/* Apply pnetid to smcd device when no pnetid is set.
265 */
266static bool smc_pnet_apply_smcd(struct smcd_dev *smcd_dev, char *pnet_name)
267{
268 bool applied = false;
269
270 mutex_lock(&smcd_dev_list.mutex);
271 if (!smc_pnet_is_pnetid_set(smcd_dev->pnetid)) {
272 memcpy(smcd_dev->pnetid, pnet_name, SMC_MAX_PNETID_LEN);
273 smcd_dev->pnetid_by_user = true;
274 applied = true;
275 }
276 mutex_unlock(&smcd_dev_list.mutex);
277 return applied;
278}
279
280/* The limit for pnetid is 16 characters.
281 * Valid characters should be (single-byte character set) a-z, A-Z, 0-9.
282 * Lower case letters are converted to upper case.
283 * Interior blanks should not be used.
284 */
285static bool smc_pnetid_valid(const char *pnet_name, char *pnetid)
286{
287 char *bf = skip_spaces(pnet_name);
288 size_t len = strlen(bf);
289 char *end = bf + len;
290
291 if (!len)
292 return false;
293 while (--end >= bf && isspace(*end))
294 ;
295 if (end - bf >= SMC_MAX_PNETID_LEN)
296 return false;
297 while (bf <= end) {
298 if (!isalnum(*bf))
299 return false;
300 *pnetid++ = islower(*bf) ? toupper(*bf) : *bf;
301 bf++;
302 }
303 *pnetid = '\0';
304 return true;
305}
306
307/* Find an infiniband device by a given name. The device might not exist. */
308static struct smc_ib_device *smc_pnet_find_ib(char *ib_name)
309{
310 struct smc_ib_device *ibdev;
311
312 mutex_lock(&smc_ib_devices.mutex);
313 list_for_each_entry(ibdev, &smc_ib_devices.list, list) {
314 if (!strncmp(ibdev->ibdev->name, ib_name,
315 sizeof(ibdev->ibdev->name)) ||
316 (ibdev->ibdev->dev.parent &&
317 !strncmp(dev_name(ibdev->ibdev->dev.parent), ib_name,
318 IB_DEVICE_NAME_MAX - 1))) {
319 goto out;
320 }
321 }
322 ibdev = NULL;
323out:
324 mutex_unlock(&smc_ib_devices.mutex);
325 return ibdev;
326}
327
328/* Find an smcd device by a given name. The device might not exist. */
329static struct smcd_dev *smc_pnet_find_smcd(char *smcd_name)
330{
331 struct smcd_dev *smcd_dev;
332
333 mutex_lock(&smcd_dev_list.mutex);
334 list_for_each_entry(smcd_dev, &smcd_dev_list.list, list) {
335 if (!strncmp(dev_name(smcd_dev->ops->get_dev(smcd_dev)),
336 smcd_name, IB_DEVICE_NAME_MAX - 1))
337 goto out;
338 }
339 smcd_dev = NULL;
340out:
341 mutex_unlock(&smcd_dev_list.mutex);
342 return smcd_dev;
343}
344
345static int smc_pnet_add_eth(struct smc_pnettable *pnettable, struct net *net,
346 char *eth_name, char *pnet_name)
347{
348 struct smc_pnetentry *tmp_pe, *new_pe;
349 struct net_device *ndev, *base_ndev;
350 u8 ndev_pnetid[SMC_MAX_PNETID_LEN];
351 bool new_netdev;
352 int rc;
353
354 /* check if (base) netdev already has a pnetid. If there is one, we do
355 * not want to add a pnet table entry
356 */
357 rc = -EEXIST;
358 ndev = dev_get_by_name(net, eth_name); /* dev_hold() */
359 if (ndev) {
360 base_ndev = pnet_find_base_ndev(ndev);
361 if (!smc_pnetid_by_dev_port(base_ndev->dev.parent,
362 base_ndev->dev_port, ndev_pnetid))
363 goto out_put;
364 }
365
366 /* add a new netdev entry to the pnet table if there isn't one */
367 rc = -ENOMEM;
368 new_pe = kzalloc(sizeof(*new_pe), GFP_KERNEL);
369 if (!new_pe)
370 goto out_put;
371 new_pe->type = SMC_PNET_ETH;
372 memcpy(new_pe->pnet_name, pnet_name, SMC_MAX_PNETID_LEN);
373 strncpy(new_pe->eth_name, eth_name, IFNAMSIZ);
374 rc = -EEXIST;
375 new_netdev = true;
376 mutex_lock(&pnettable->lock);
377 list_for_each_entry(tmp_pe, &pnettable->pnetlist, list) {
378 if (tmp_pe->type == SMC_PNET_ETH &&
379 !strncmp(tmp_pe->eth_name, eth_name, IFNAMSIZ)) {
380 new_netdev = false;
381 break;
382 }
383 }
384 if (new_netdev) {
385 if (ndev) {
386 new_pe->ndev = ndev;
387 netdev_tracker_alloc(ndev, &new_pe->dev_tracker,
388 GFP_ATOMIC);
389 }
390 list_add_tail(&new_pe->list, &pnettable->pnetlist);
391 mutex_unlock(&pnettable->lock);
392 } else {
393 mutex_unlock(&pnettable->lock);
394 kfree(new_pe);
395 goto out_put;
396 }
397 if (ndev)
398 pr_warn_ratelimited("smc: net device %s "
399 "applied user defined pnetid %.16s\n",
400 new_pe->eth_name, new_pe->pnet_name);
401 return 0;
402
403out_put:
404 dev_put(ndev);
405 return rc;
406}
407
408static int smc_pnet_add_ib(struct smc_pnettable *pnettable, char *ib_name,
409 u8 ib_port, char *pnet_name)
410{
411 struct smc_pnetentry *tmp_pe, *new_pe;
412 struct smc_ib_device *ib_dev;
413 bool smcddev_applied = true;
414 bool ibdev_applied = true;
415 struct smcd_dev *smcd;
416 struct device *dev;
417 bool new_ibdev;
418
419 /* try to apply the pnetid to active devices */
420 ib_dev = smc_pnet_find_ib(ib_name);
421 if (ib_dev) {
422 ibdev_applied = smc_pnet_apply_ib(ib_dev, ib_port, pnet_name);
423 if (ibdev_applied)
424 pr_warn_ratelimited("smc: ib device %s ibport %d "
425 "applied user defined pnetid "
426 "%.16s\n", ib_dev->ibdev->name,
427 ib_port,
428 ib_dev->pnetid[ib_port - 1]);
429 }
430 smcd = smc_pnet_find_smcd(ib_name);
431 if (smcd) {
432 smcddev_applied = smc_pnet_apply_smcd(smcd, pnet_name);
433 if (smcddev_applied) {
434 dev = smcd->ops->get_dev(smcd);
435 pr_warn_ratelimited("smc: smcd device %s "
436 "applied user defined pnetid "
437 "%.16s\n", dev_name(dev),
438 smcd->pnetid);
439 }
440 }
441 /* Apply fails when a device has a hardware-defined pnetid set, do not
442 * add a pnet table entry in that case.
443 */
444 if (!ibdev_applied || !smcddev_applied)
445 return -EEXIST;
446
447 /* add a new ib entry to the pnet table if there isn't one */
448 new_pe = kzalloc(sizeof(*new_pe), GFP_KERNEL);
449 if (!new_pe)
450 return -ENOMEM;
451 new_pe->type = SMC_PNET_IB;
452 memcpy(new_pe->pnet_name, pnet_name, SMC_MAX_PNETID_LEN);
453 strncpy(new_pe->ib_name, ib_name, IB_DEVICE_NAME_MAX);
454 new_pe->ib_port = ib_port;
455
456 new_ibdev = true;
457 mutex_lock(&pnettable->lock);
458 list_for_each_entry(tmp_pe, &pnettable->pnetlist, list) {
459 if (tmp_pe->type == SMC_PNET_IB &&
460 !strncmp(tmp_pe->ib_name, ib_name, IB_DEVICE_NAME_MAX)) {
461 new_ibdev = false;
462 break;
463 }
464 }
465 if (new_ibdev) {
466 list_add_tail(&new_pe->list, &pnettable->pnetlist);
467 mutex_unlock(&pnettable->lock);
468 } else {
469 mutex_unlock(&pnettable->lock);
470 kfree(new_pe);
471 }
472 return (new_ibdev) ? 0 : -EEXIST;
473}
474
475/* Append a pnetid to the end of the pnet table if not already on this list.
476 */
477static int smc_pnet_enter(struct net *net, struct nlattr *tb[])
478{
479 char pnet_name[SMC_MAX_PNETID_LEN + 1];
480 struct smc_pnettable *pnettable;
481 bool new_netdev = false;
482 bool new_ibdev = false;
483 struct smc_net *sn;
484 u8 ibport = 1;
485 char *string;
486 int rc;
487
488 /* get pnettable for namespace */
489 sn = net_generic(net, smc_net_id);
490 pnettable = &sn->pnettable;
491
492 rc = -EINVAL;
493 if (!tb[SMC_PNETID_NAME])
494 goto error;
495 string = (char *)nla_data(tb[SMC_PNETID_NAME]);
496 if (!smc_pnetid_valid(string, pnet_name))
497 goto error;
498
499 if (tb[SMC_PNETID_ETHNAME]) {
500 string = (char *)nla_data(tb[SMC_PNETID_ETHNAME]);
501 rc = smc_pnet_add_eth(pnettable, net, string, pnet_name);
502 if (!rc)
503 new_netdev = true;
504 else if (rc != -EEXIST)
505 goto error;
506 }
507
508 /* if this is not the initial namespace, stop here */
509 if (net != &init_net)
510 return new_netdev ? 0 : -EEXIST;
511
512 rc = -EINVAL;
513 if (tb[SMC_PNETID_IBNAME]) {
514 string = (char *)nla_data(tb[SMC_PNETID_IBNAME]);
515 string = strim(string);
516 if (tb[SMC_PNETID_IBPORT]) {
517 ibport = nla_get_u8(tb[SMC_PNETID_IBPORT]);
518 if (ibport < 1 || ibport > SMC_MAX_PORTS)
519 goto error;
520 }
521 rc = smc_pnet_add_ib(pnettable, string, ibport, pnet_name);
522 if (!rc)
523 new_ibdev = true;
524 else if (rc != -EEXIST)
525 goto error;
526 }
527 return (new_netdev || new_ibdev) ? 0 : -EEXIST;
528
529error:
530 return rc;
531}
532
533/* Convert an smc_pnetentry to a netlink attribute sequence */
534static int smc_pnet_set_nla(struct sk_buff *msg,
535 struct smc_pnetentry *pnetelem)
536{
537 if (nla_put_string(msg, SMC_PNETID_NAME, pnetelem->pnet_name))
538 return -1;
539 if (pnetelem->type == SMC_PNET_ETH) {
540 if (nla_put_string(msg, SMC_PNETID_ETHNAME,
541 pnetelem->eth_name))
542 return -1;
543 } else {
544 if (nla_put_string(msg, SMC_PNETID_ETHNAME, "n/a"))
545 return -1;
546 }
547 if (pnetelem->type == SMC_PNET_IB) {
548 if (nla_put_string(msg, SMC_PNETID_IBNAME, pnetelem->ib_name) ||
549 nla_put_u8(msg, SMC_PNETID_IBPORT, pnetelem->ib_port))
550 return -1;
551 } else {
552 if (nla_put_string(msg, SMC_PNETID_IBNAME, "n/a") ||
553 nla_put_u8(msg, SMC_PNETID_IBPORT, 0xff))
554 return -1;
555 }
556
557 return 0;
558}
559
560static int smc_pnet_add(struct sk_buff *skb, struct genl_info *info)
561{
562 struct net *net = genl_info_net(info);
563
564 return smc_pnet_enter(net, info->attrs);
565}
566
567static int smc_pnet_del(struct sk_buff *skb, struct genl_info *info)
568{
569 struct net *net = genl_info_net(info);
570
571 if (!info->attrs[SMC_PNETID_NAME])
572 return -EINVAL;
573 return smc_pnet_remove_by_pnetid(net,
574 (char *)nla_data(info->attrs[SMC_PNETID_NAME]));
575}
576
577static int smc_pnet_dump_start(struct netlink_callback *cb)
578{
579 cb->args[0] = 0;
580 return 0;
581}
582
583static int smc_pnet_dumpinfo(struct sk_buff *skb,
584 u32 portid, u32 seq, u32 flags,
585 struct smc_pnetentry *pnetelem)
586{
587 void *hdr;
588
589 hdr = genlmsg_put(skb, portid, seq, &smc_pnet_nl_family,
590 flags, SMC_PNETID_GET);
591 if (!hdr)
592 return -ENOMEM;
593 if (smc_pnet_set_nla(skb, pnetelem) < 0) {
594 genlmsg_cancel(skb, hdr);
595 return -EMSGSIZE;
596 }
597 genlmsg_end(skb, hdr);
598 return 0;
599}
600
601static int _smc_pnet_dump(struct net *net, struct sk_buff *skb, u32 portid,
602 u32 seq, u8 *pnetid, int start_idx)
603{
604 struct smc_pnettable *pnettable;
605 struct smc_pnetentry *pnetelem;
606 struct smc_net *sn;
607 int idx = 0;
608
609 /* get pnettable for namespace */
610 sn = net_generic(net, smc_net_id);
611 pnettable = &sn->pnettable;
612
613 /* dump pnettable entries */
614 mutex_lock(&pnettable->lock);
615 list_for_each_entry(pnetelem, &pnettable->pnetlist, list) {
616 if (pnetid && !smc_pnet_match(pnetelem->pnet_name, pnetid))
617 continue;
618 if (idx++ < start_idx)
619 continue;
620 /* if this is not the initial namespace, dump only netdev */
621 if (net != &init_net && pnetelem->type != SMC_PNET_ETH)
622 continue;
623 if (smc_pnet_dumpinfo(skb, portid, seq, NLM_F_MULTI,
624 pnetelem)) {
625 --idx;
626 break;
627 }
628 }
629 mutex_unlock(&pnettable->lock);
630 return idx;
631}
632
633static int smc_pnet_dump(struct sk_buff *skb, struct netlink_callback *cb)
634{
635 struct net *net = sock_net(skb->sk);
636 int idx;
637
638 idx = _smc_pnet_dump(net, skb, NETLINK_CB(cb->skb).portid,
639 cb->nlh->nlmsg_seq, NULL, cb->args[0]);
640
641 cb->args[0] = idx;
642 return skb->len;
643}
644
645/* Retrieve one PNETID entry */
646static int smc_pnet_get(struct sk_buff *skb, struct genl_info *info)
647{
648 struct net *net = genl_info_net(info);
649 struct sk_buff *msg;
650 void *hdr;
651
652 if (!info->attrs[SMC_PNETID_NAME])
653 return -EINVAL;
654
655 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
656 if (!msg)
657 return -ENOMEM;
658
659 _smc_pnet_dump(net, msg, info->snd_portid, info->snd_seq,
660 nla_data(info->attrs[SMC_PNETID_NAME]), 0);
661
662 /* finish multi part message and send it */
663 hdr = nlmsg_put(msg, info->snd_portid, info->snd_seq, NLMSG_DONE, 0,
664 NLM_F_MULTI);
665 if (!hdr) {
666 nlmsg_free(msg);
667 return -EMSGSIZE;
668 }
669 return genlmsg_reply(msg, info);
670}
671
672/* Remove and delete all pnetids from pnet table.
673 */
674static int smc_pnet_flush(struct sk_buff *skb, struct genl_info *info)
675{
676 struct net *net = genl_info_net(info);
677
678 smc_pnet_remove_by_pnetid(net, NULL);
679 return 0;
680}
681
682/* SMC_PNETID generic netlink operation definition */
683static const struct genl_ops smc_pnet_ops[] = {
684 {
685 .cmd = SMC_PNETID_GET,
686 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
687 /* can be retrieved by unprivileged users */
688 .doit = smc_pnet_get,
689 .dumpit = smc_pnet_dump,
690 .start = smc_pnet_dump_start
691 },
692 {
693 .cmd = SMC_PNETID_ADD,
694 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
695 .flags = GENL_ADMIN_PERM,
696 .doit = smc_pnet_add
697 },
698 {
699 .cmd = SMC_PNETID_DEL,
700 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
701 .flags = GENL_ADMIN_PERM,
702 .doit = smc_pnet_del
703 },
704 {
705 .cmd = SMC_PNETID_FLUSH,
706 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
707 .flags = GENL_ADMIN_PERM,
708 .doit = smc_pnet_flush
709 }
710};
711
712/* SMC_PNETID family definition */
713static struct genl_family smc_pnet_nl_family __ro_after_init = {
714 .hdrsize = 0,
715 .name = SMCR_GENL_FAMILY_NAME,
716 .version = SMCR_GENL_FAMILY_VERSION,
717 .maxattr = SMC_PNETID_MAX,
718 .policy = smc_pnet_policy,
719 .netnsok = true,
720 .module = THIS_MODULE,
721 .ops = smc_pnet_ops,
722 .n_ops = ARRAY_SIZE(smc_pnet_ops),
723 .resv_start_op = SMC_PNETID_FLUSH + 1,
724};
725
726bool smc_pnet_is_ndev_pnetid(struct net *net, u8 *pnetid)
727{
728 struct smc_net *sn = net_generic(net, smc_net_id);
729 struct smc_pnetids_ndev_entry *pe;
730 bool rc = false;
731
732 read_lock(&sn->pnetids_ndev.lock);
733 list_for_each_entry(pe, &sn->pnetids_ndev.list, list) {
734 if (smc_pnet_match(pnetid, pe->pnetid)) {
735 rc = true;
736 goto unlock;
737 }
738 }
739
740unlock:
741 read_unlock(&sn->pnetids_ndev.lock);
742 return rc;
743}
744
745static int smc_pnet_add_pnetid(struct net *net, u8 *pnetid)
746{
747 struct smc_net *sn = net_generic(net, smc_net_id);
748 struct smc_pnetids_ndev_entry *pe, *pi;
749
750 pe = kzalloc(sizeof(*pe), GFP_KERNEL);
751 if (!pe)
752 return -ENOMEM;
753
754 write_lock(&sn->pnetids_ndev.lock);
755 list_for_each_entry(pi, &sn->pnetids_ndev.list, list) {
756 if (smc_pnet_match(pnetid, pi->pnetid)) {
757 refcount_inc(&pi->refcnt);
758 kfree(pe);
759 goto unlock;
760 }
761 }
762 refcount_set(&pe->refcnt, 1);
763 memcpy(pe->pnetid, pnetid, SMC_MAX_PNETID_LEN);
764 list_add_tail(&pe->list, &sn->pnetids_ndev.list);
765
766unlock:
767 write_unlock(&sn->pnetids_ndev.lock);
768 return 0;
769}
770
771static void smc_pnet_remove_pnetid(struct net *net, u8 *pnetid)
772{
773 struct smc_net *sn = net_generic(net, smc_net_id);
774 struct smc_pnetids_ndev_entry *pe, *pe2;
775
776 write_lock(&sn->pnetids_ndev.lock);
777 list_for_each_entry_safe(pe, pe2, &sn->pnetids_ndev.list, list) {
778 if (smc_pnet_match(pnetid, pe->pnetid)) {
779 if (refcount_dec_and_test(&pe->refcnt)) {
780 list_del(&pe->list);
781 kfree(pe);
782 }
783 break;
784 }
785 }
786 write_unlock(&sn->pnetids_ndev.lock);
787}
788
789static void smc_pnet_add_base_pnetid(struct net *net, struct net_device *dev,
790 u8 *ndev_pnetid)
791{
792 struct net_device *base_dev;
793
794 base_dev = __pnet_find_base_ndev(dev);
795 if (base_dev->flags & IFF_UP &&
796 !smc_pnetid_by_dev_port(base_dev->dev.parent, base_dev->dev_port,
797 ndev_pnetid)) {
798 /* add to PNETIDs list */
799 smc_pnet_add_pnetid(net, ndev_pnetid);
800 }
801}
802
803/* create initial list of netdevice pnetids */
804static void smc_pnet_create_pnetids_list(struct net *net)
805{
806 u8 ndev_pnetid[SMC_MAX_PNETID_LEN];
807 struct net_device *dev;
808
809 /* Newly created netns do not have devices.
810 * Do not even acquire rtnl.
811 */
812 if (list_empty(&net->dev_base_head))
813 return;
814
815 /* Note: This might not be needed, because smc_pnet_netdev_event()
816 * is also calling smc_pnet_add_base_pnetid() when handling
817 * NETDEV_UP event.
818 */
819 rtnl_lock();
820 for_each_netdev(net, dev)
821 smc_pnet_add_base_pnetid(net, dev, ndev_pnetid);
822 rtnl_unlock();
823}
824
825/* clean up list of netdevice pnetids */
826static void smc_pnet_destroy_pnetids_list(struct net *net)
827{
828 struct smc_net *sn = net_generic(net, smc_net_id);
829 struct smc_pnetids_ndev_entry *pe, *temp_pe;
830
831 write_lock(&sn->pnetids_ndev.lock);
832 list_for_each_entry_safe(pe, temp_pe, &sn->pnetids_ndev.list, list) {
833 list_del(&pe->list);
834 kfree(pe);
835 }
836 write_unlock(&sn->pnetids_ndev.lock);
837}
838
839static int smc_pnet_netdev_event(struct notifier_block *this,
840 unsigned long event, void *ptr)
841{
842 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
843 struct net *net = dev_net(event_dev);
844 u8 ndev_pnetid[SMC_MAX_PNETID_LEN];
845
846 switch (event) {
847 case NETDEV_REBOOT:
848 case NETDEV_UNREGISTER:
849 smc_pnet_remove_by_ndev(event_dev);
850 smc_ib_ndev_change(event_dev, event);
851 return NOTIFY_OK;
852 case NETDEV_REGISTER:
853 smc_pnet_add_by_ndev(event_dev);
854 smc_ib_ndev_change(event_dev, event);
855 return NOTIFY_OK;
856 case NETDEV_UP:
857 smc_pnet_add_base_pnetid(net, event_dev, ndev_pnetid);
858 return NOTIFY_OK;
859 case NETDEV_DOWN:
860 event_dev = __pnet_find_base_ndev(event_dev);
861 if (!smc_pnetid_by_dev_port(event_dev->dev.parent,
862 event_dev->dev_port, ndev_pnetid)) {
863 /* remove from PNETIDs list */
864 smc_pnet_remove_pnetid(net, ndev_pnetid);
865 }
866 return NOTIFY_OK;
867 default:
868 return NOTIFY_DONE;
869 }
870}
871
872static struct notifier_block smc_netdev_notifier = {
873 .notifier_call = smc_pnet_netdev_event
874};
875
876/* init network namespace */
877int smc_pnet_net_init(struct net *net)
878{
879 struct smc_net *sn = net_generic(net, smc_net_id);
880 struct smc_pnettable *pnettable = &sn->pnettable;
881 struct smc_pnetids_ndev *pnetids_ndev = &sn->pnetids_ndev;
882
883 INIT_LIST_HEAD(&pnettable->pnetlist);
884 mutex_init(&pnettable->lock);
885 INIT_LIST_HEAD(&pnetids_ndev->list);
886 rwlock_init(&pnetids_ndev->lock);
887
888 smc_pnet_create_pnetids_list(net);
889
890 return 0;
891}
892
893int __init smc_pnet_init(void)
894{
895 int rc;
896
897 rc = genl_register_family(&smc_pnet_nl_family);
898 if (rc)
899 return rc;
900 rc = register_netdevice_notifier(&smc_netdev_notifier);
901 if (rc)
902 genl_unregister_family(&smc_pnet_nl_family);
903
904 return rc;
905}
906
907/* exit network namespace */
908void smc_pnet_net_exit(struct net *net)
909{
910 /* flush pnet table */
911 smc_pnet_remove_by_pnetid(net, NULL);
912 smc_pnet_destroy_pnetids_list(net);
913}
914
915void smc_pnet_exit(void)
916{
917 unregister_netdevice_notifier(&smc_netdev_notifier);
918 genl_unregister_family(&smc_pnet_nl_family);
919}
920
921static struct net_device *__pnet_find_base_ndev(struct net_device *ndev)
922{
923 int i, nest_lvl;
924
925 ASSERT_RTNL();
926 nest_lvl = ndev->lower_level;
927 for (i = 0; i < nest_lvl; i++) {
928 struct list_head *lower = &ndev->adj_list.lower;
929
930 if (list_empty(lower))
931 break;
932 lower = lower->next;
933 ndev = netdev_lower_get_next(ndev, &lower);
934 }
935 return ndev;
936}
937
938/* Determine one base device for stacked net devices.
939 * If the lower device level contains more than one devices
940 * (for instance with bonding slaves), just the first device
941 * is used to reach a base device.
942 */
943static struct net_device *pnet_find_base_ndev(struct net_device *ndev)
944{
945 rtnl_lock();
946 ndev = __pnet_find_base_ndev(ndev);
947 rtnl_unlock();
948 return ndev;
949}
950
951static int smc_pnet_find_ndev_pnetid_by_table(struct net_device *ndev,
952 u8 *pnetid)
953{
954 struct smc_pnettable *pnettable;
955 struct net *net = dev_net(ndev);
956 struct smc_pnetentry *pnetelem;
957 struct smc_net *sn;
958 int rc = -ENOENT;
959
960 /* get pnettable for namespace */
961 sn = net_generic(net, smc_net_id);
962 pnettable = &sn->pnettable;
963
964 mutex_lock(&pnettable->lock);
965 list_for_each_entry(pnetelem, &pnettable->pnetlist, list) {
966 if (pnetelem->type == SMC_PNET_ETH && ndev == pnetelem->ndev) {
967 /* get pnetid of netdev device */
968 memcpy(pnetid, pnetelem->pnet_name, SMC_MAX_PNETID_LEN);
969 rc = 0;
970 break;
971 }
972 }
973 mutex_unlock(&pnettable->lock);
974 return rc;
975}
976
977static int smc_pnet_determine_gid(struct smc_ib_device *ibdev, int i,
978 struct smc_init_info *ini)
979{
980 if (!ini->check_smcrv2 &&
981 !smc_ib_determine_gid(ibdev, i, ini->vlan_id, ini->ib_gid, NULL,
982 NULL)) {
983 ini->ib_dev = ibdev;
984 ini->ib_port = i;
985 return 0;
986 }
987 if (ini->check_smcrv2 &&
988 !smc_ib_determine_gid(ibdev, i, ini->vlan_id, ini->smcrv2.ib_gid_v2,
989 NULL, &ini->smcrv2)) {
990 ini->smcrv2.ib_dev_v2 = ibdev;
991 ini->smcrv2.ib_port_v2 = i;
992 return 0;
993 }
994 return -ENODEV;
995}
996
997/* find a roce device for the given pnetid */
998static void _smc_pnet_find_roce_by_pnetid(u8 *pnet_id,
999 struct smc_init_info *ini,
1000 struct smc_ib_device *known_dev,
1001 struct net *net)
1002{
1003 struct smc_ib_device *ibdev;
1004 int i;
1005
1006 mutex_lock(&smc_ib_devices.mutex);
1007 list_for_each_entry(ibdev, &smc_ib_devices.list, list) {
1008 if (ibdev == known_dev ||
1009 !rdma_dev_access_netns(ibdev->ibdev, net))
1010 continue;
1011 for (i = 1; i <= SMC_MAX_PORTS; i++) {
1012 if (!rdma_is_port_valid(ibdev->ibdev, i))
1013 continue;
1014 if (smc_pnet_match(ibdev->pnetid[i - 1], pnet_id) &&
1015 smc_ib_port_active(ibdev, i) &&
1016 !test_bit(i - 1, ibdev->ports_going_away)) {
1017 if (!smc_pnet_determine_gid(ibdev, i, ini))
1018 goto out;
1019 }
1020 }
1021 }
1022out:
1023 mutex_unlock(&smc_ib_devices.mutex);
1024}
1025
1026/* find alternate roce device with same pnet_id, vlan_id and net namespace */
1027void smc_pnet_find_alt_roce(struct smc_link_group *lgr,
1028 struct smc_init_info *ini,
1029 struct smc_ib_device *known_dev)
1030{
1031 struct net *net = lgr->net;
1032
1033 _smc_pnet_find_roce_by_pnetid(lgr->pnet_id, ini, known_dev, net);
1034}
1035
1036/* if handshake network device belongs to a roce device, return its
1037 * IB device and port
1038 */
1039static void smc_pnet_find_rdma_dev(struct net_device *netdev,
1040 struct smc_init_info *ini)
1041{
1042 struct net *net = dev_net(netdev);
1043 struct smc_ib_device *ibdev;
1044
1045 mutex_lock(&smc_ib_devices.mutex);
1046 list_for_each_entry(ibdev, &smc_ib_devices.list, list) {
1047 struct net_device *ndev;
1048 int i;
1049
1050 /* check rdma net namespace */
1051 if (!rdma_dev_access_netns(ibdev->ibdev, net))
1052 continue;
1053
1054 for (i = 1; i <= SMC_MAX_PORTS; i++) {
1055 if (!rdma_is_port_valid(ibdev->ibdev, i))
1056 continue;
1057 ndev = ib_device_get_netdev(ibdev->ibdev, i);
1058 if (!ndev)
1059 continue;
1060 dev_put(ndev);
1061 if (netdev == ndev &&
1062 smc_ib_port_active(ibdev, i) &&
1063 !test_bit(i - 1, ibdev->ports_going_away)) {
1064 if (!smc_pnet_determine_gid(ibdev, i, ini))
1065 break;
1066 }
1067 }
1068 }
1069 mutex_unlock(&smc_ib_devices.mutex);
1070}
1071
1072/* Determine the corresponding IB device port based on the hardware PNETID.
1073 * Searching stops at the first matching active IB device port with vlan_id
1074 * configured.
1075 * If nothing found, check pnetid table.
1076 * If nothing found, try to use handshake device
1077 */
1078static void smc_pnet_find_roce_by_pnetid(struct net_device *ndev,
1079 struct smc_init_info *ini)
1080{
1081 u8 ndev_pnetid[SMC_MAX_PNETID_LEN];
1082 struct net *net;
1083
1084 ndev = pnet_find_base_ndev(ndev);
1085 net = dev_net(ndev);
1086 if (smc_pnetid_by_dev_port(ndev->dev.parent, ndev->dev_port,
1087 ndev_pnetid) &&
1088 smc_pnet_find_ndev_pnetid_by_table(ndev, ndev_pnetid)) {
1089 smc_pnet_find_rdma_dev(ndev, ini);
1090 return; /* pnetid could not be determined */
1091 }
1092 _smc_pnet_find_roce_by_pnetid(ndev_pnetid, ini, NULL, net);
1093}
1094
1095static void smc_pnet_find_ism_by_pnetid(struct net_device *ndev,
1096 struct smc_init_info *ini)
1097{
1098 u8 ndev_pnetid[SMC_MAX_PNETID_LEN];
1099 struct smcd_dev *ismdev;
1100
1101 ndev = pnet_find_base_ndev(ndev);
1102 if (smc_pnetid_by_dev_port(ndev->dev.parent, ndev->dev_port,
1103 ndev_pnetid) &&
1104 smc_pnet_find_ndev_pnetid_by_table(ndev, ndev_pnetid))
1105 return; /* pnetid could not be determined */
1106
1107 mutex_lock(&smcd_dev_list.mutex);
1108 list_for_each_entry(ismdev, &smcd_dev_list.list, list) {
1109 if (smc_pnet_match(ismdev->pnetid, ndev_pnetid) &&
1110 !ismdev->going_away &&
1111 (!ini->ism_peer_gid[0].gid ||
1112 !smc_ism_cantalk(&ini->ism_peer_gid[0], ini->vlan_id,
1113 ismdev))) {
1114 ini->ism_dev[0] = ismdev;
1115 break;
1116 }
1117 }
1118 mutex_unlock(&smcd_dev_list.mutex);
1119}
1120
1121/* PNET table analysis for a given sock:
1122 * determine ib_device and port belonging to used internal TCP socket
1123 * ethernet interface.
1124 */
1125void smc_pnet_find_roce_resource(struct sock *sk, struct smc_init_info *ini)
1126{
1127 struct dst_entry *dst = sk_dst_get(sk);
1128
1129 if (!dst)
1130 goto out;
1131 if (!dst->dev)
1132 goto out_rel;
1133
1134 smc_pnet_find_roce_by_pnetid(dst->dev, ini);
1135
1136out_rel:
1137 dst_release(dst);
1138out:
1139 return;
1140}
1141
1142void smc_pnet_find_ism_resource(struct sock *sk, struct smc_init_info *ini)
1143{
1144 struct dst_entry *dst = sk_dst_get(sk);
1145
1146 ini->ism_dev[0] = NULL;
1147 if (!dst)
1148 goto out;
1149 if (!dst->dev)
1150 goto out_rel;
1151
1152 smc_pnet_find_ism_by_pnetid(dst->dev, ini);
1153
1154out_rel:
1155 dst_release(dst);
1156out:
1157 return;
1158}
1159
1160/* Lookup and apply a pnet table entry to the given ib device.
1161 */
1162int smc_pnetid_by_table_ib(struct smc_ib_device *smcibdev, u8 ib_port)
1163{
1164 char *ib_name = smcibdev->ibdev->name;
1165 struct smc_pnettable *pnettable;
1166 struct smc_pnetentry *tmp_pe;
1167 struct smc_net *sn;
1168 int rc = -ENOENT;
1169
1170 /* get pnettable for init namespace */
1171 sn = net_generic(&init_net, smc_net_id);
1172 pnettable = &sn->pnettable;
1173
1174 mutex_lock(&pnettable->lock);
1175 list_for_each_entry(tmp_pe, &pnettable->pnetlist, list) {
1176 if (tmp_pe->type == SMC_PNET_IB &&
1177 !strncmp(tmp_pe->ib_name, ib_name, IB_DEVICE_NAME_MAX) &&
1178 tmp_pe->ib_port == ib_port) {
1179 smc_pnet_apply_ib(smcibdev, ib_port, tmp_pe->pnet_name);
1180 rc = 0;
1181 break;
1182 }
1183 }
1184 mutex_unlock(&pnettable->lock);
1185
1186 return rc;
1187}
1188
1189/* Lookup and apply a pnet table entry to the given smcd device.
1190 */
1191int smc_pnetid_by_table_smcd(struct smcd_dev *smcddev)
1192{
1193 const char *ib_name = dev_name(smcddev->ops->get_dev(smcddev));
1194 struct smc_pnettable *pnettable;
1195 struct smc_pnetentry *tmp_pe;
1196 struct smc_net *sn;
1197 int rc = -ENOENT;
1198
1199 /* get pnettable for init namespace */
1200 sn = net_generic(&init_net, smc_net_id);
1201 pnettable = &sn->pnettable;
1202
1203 mutex_lock(&pnettable->lock);
1204 list_for_each_entry(tmp_pe, &pnettable->pnetlist, list) {
1205 if (tmp_pe->type == SMC_PNET_IB &&
1206 !strncmp(tmp_pe->ib_name, ib_name, IB_DEVICE_NAME_MAX)) {
1207 smc_pnet_apply_smcd(smcddev, tmp_pe->pnet_name);
1208 rc = 0;
1209 break;
1210 }
1211 }
1212 mutex_unlock(&pnettable->lock);
1213
1214 return rc;
1215}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Shared Memory Communications over RDMA (SMC-R) and RoCE
4 *
5 * Generic netlink support functions to configure an SMC-R PNET table
6 *
7 * Copyright IBM Corp. 2016
8 *
9 * Author(s): Thomas Richter <tmricht@linux.vnet.ibm.com>
10 */
11
12#include <linux/module.h>
13#include <linux/list.h>
14#include <linux/ctype.h>
15#include <net/netlink.h>
16#include <net/genetlink.h>
17
18#include <uapi/linux/if.h>
19#include <uapi/linux/smc.h>
20
21#include <rdma/ib_verbs.h>
22
23#include <net/netns/generic.h>
24#include "smc_netns.h"
25
26#include "smc_pnet.h"
27#include "smc_ib.h"
28#include "smc_ism.h"
29#include "smc_core.h"
30
31#define SMC_ASCII_BLANK 32
32
33static struct net_device *pnet_find_base_ndev(struct net_device *ndev);
34
35static struct nla_policy smc_pnet_policy[SMC_PNETID_MAX + 1] = {
36 [SMC_PNETID_NAME] = {
37 .type = NLA_NUL_STRING,
38 .len = SMC_MAX_PNETID_LEN
39 },
40 [SMC_PNETID_ETHNAME] = {
41 .type = NLA_NUL_STRING,
42 .len = IFNAMSIZ - 1
43 },
44 [SMC_PNETID_IBNAME] = {
45 .type = NLA_NUL_STRING,
46 .len = IB_DEVICE_NAME_MAX - 1
47 },
48 [SMC_PNETID_IBPORT] = { .type = NLA_U8 }
49};
50
51static struct genl_family smc_pnet_nl_family;
52
53/**
54 * struct smc_user_pnetentry - pnet identifier name entry for/from user
55 * @list: List node.
56 * @pnet_name: Pnet identifier name
57 * @ndev: pointer to network device.
58 * @smcibdev: Pointer to IB device.
59 * @ib_port: Port of IB device.
60 * @smcd_dev: Pointer to smcd device.
61 */
62struct smc_user_pnetentry {
63 struct list_head list;
64 char pnet_name[SMC_MAX_PNETID_LEN + 1];
65 struct net_device *ndev;
66 struct smc_ib_device *smcibdev;
67 u8 ib_port;
68 struct smcd_dev *smcd_dev;
69};
70
71/* pnet entry stored in pnet table */
72struct smc_pnetentry {
73 struct list_head list;
74 char pnet_name[SMC_MAX_PNETID_LEN + 1];
75 struct net_device *ndev;
76};
77
78/* Check if two given pnetids match */
79static bool smc_pnet_match(u8 *pnetid1, u8 *pnetid2)
80{
81 int i;
82
83 for (i = 0; i < SMC_MAX_PNETID_LEN; i++) {
84 if ((pnetid1[i] == 0 || pnetid1[i] == SMC_ASCII_BLANK) &&
85 (pnetid2[i] == 0 || pnetid2[i] == SMC_ASCII_BLANK))
86 break;
87 if (pnetid1[i] != pnetid2[i])
88 return false;
89 }
90 return true;
91}
92
93/* Remove a pnetid from the pnet table.
94 */
95static int smc_pnet_remove_by_pnetid(struct net *net, char *pnet_name)
96{
97 struct smc_pnetentry *pnetelem, *tmp_pe;
98 struct smc_pnettable *pnettable;
99 struct smc_ib_device *ibdev;
100 struct smcd_dev *smcd_dev;
101 struct smc_net *sn;
102 int rc = -ENOENT;
103 int ibport;
104
105 /* get pnettable for namespace */
106 sn = net_generic(net, smc_net_id);
107 pnettable = &sn->pnettable;
108
109 /* remove netdevices */
110 write_lock(&pnettable->lock);
111 list_for_each_entry_safe(pnetelem, tmp_pe, &pnettable->pnetlist,
112 list) {
113 if (!pnet_name ||
114 smc_pnet_match(pnetelem->pnet_name, pnet_name)) {
115 list_del(&pnetelem->list);
116 dev_put(pnetelem->ndev);
117 kfree(pnetelem);
118 rc = 0;
119 }
120 }
121 write_unlock(&pnettable->lock);
122
123 /* if this is not the initial namespace, stop here */
124 if (net != &init_net)
125 return rc;
126
127 /* remove ib devices */
128 spin_lock(&smc_ib_devices.lock);
129 list_for_each_entry(ibdev, &smc_ib_devices.list, list) {
130 for (ibport = 0; ibport < SMC_MAX_PORTS; ibport++) {
131 if (ibdev->pnetid_by_user[ibport] &&
132 (!pnet_name ||
133 smc_pnet_match(pnet_name,
134 ibdev->pnetid[ibport]))) {
135 memset(ibdev->pnetid[ibport], 0,
136 SMC_MAX_PNETID_LEN);
137 ibdev->pnetid_by_user[ibport] = false;
138 rc = 0;
139 }
140 }
141 }
142 spin_unlock(&smc_ib_devices.lock);
143 /* remove smcd devices */
144 spin_lock(&smcd_dev_list.lock);
145 list_for_each_entry(smcd_dev, &smcd_dev_list.list, list) {
146 if (smcd_dev->pnetid_by_user &&
147 (!pnet_name ||
148 smc_pnet_match(pnet_name, smcd_dev->pnetid))) {
149 memset(smcd_dev->pnetid, 0, SMC_MAX_PNETID_LEN);
150 smcd_dev->pnetid_by_user = false;
151 rc = 0;
152 }
153 }
154 spin_unlock(&smcd_dev_list.lock);
155 return rc;
156}
157
158/* Remove a pnet entry mentioning a given network device from the pnet table.
159 */
160static int smc_pnet_remove_by_ndev(struct net_device *ndev)
161{
162 struct smc_pnetentry *pnetelem, *tmp_pe;
163 struct smc_pnettable *pnettable;
164 struct net *net = dev_net(ndev);
165 struct smc_net *sn;
166 int rc = -ENOENT;
167
168 /* get pnettable for namespace */
169 sn = net_generic(net, smc_net_id);
170 pnettable = &sn->pnettable;
171
172 write_lock(&pnettable->lock);
173 list_for_each_entry_safe(pnetelem, tmp_pe, &pnettable->pnetlist, list) {
174 if (pnetelem->ndev == ndev) {
175 list_del(&pnetelem->list);
176 dev_put(pnetelem->ndev);
177 kfree(pnetelem);
178 rc = 0;
179 break;
180 }
181 }
182 write_unlock(&pnettable->lock);
183 return rc;
184}
185
186/* Append a pnetid to the end of the pnet table if not already on this list.
187 */
188static int smc_pnet_enter(struct smc_pnettable *pnettable,
189 struct smc_user_pnetentry *new_pnetelem)
190{
191 u8 pnet_null[SMC_MAX_PNETID_LEN] = {0};
192 u8 ndev_pnetid[SMC_MAX_PNETID_LEN];
193 struct smc_pnetentry *tmp_pnetelem;
194 struct smc_pnetentry *pnetelem;
195 bool new_smcddev = false;
196 struct net_device *ndev;
197 bool new_netdev = true;
198 bool new_ibdev = false;
199
200 if (new_pnetelem->smcibdev) {
201 struct smc_ib_device *ib_dev = new_pnetelem->smcibdev;
202 int ib_port = new_pnetelem->ib_port;
203
204 spin_lock(&smc_ib_devices.lock);
205 if (smc_pnet_match(ib_dev->pnetid[ib_port - 1], pnet_null)) {
206 memcpy(ib_dev->pnetid[ib_port - 1],
207 new_pnetelem->pnet_name, SMC_MAX_PNETID_LEN);
208 ib_dev->pnetid_by_user[ib_port - 1] = true;
209 new_ibdev = true;
210 }
211 spin_unlock(&smc_ib_devices.lock);
212 }
213 if (new_pnetelem->smcd_dev) {
214 struct smcd_dev *smcd_dev = new_pnetelem->smcd_dev;
215
216 spin_lock(&smcd_dev_list.lock);
217 if (smc_pnet_match(smcd_dev->pnetid, pnet_null)) {
218 memcpy(smcd_dev->pnetid, new_pnetelem->pnet_name,
219 SMC_MAX_PNETID_LEN);
220 smcd_dev->pnetid_by_user = true;
221 new_smcddev = true;
222 }
223 spin_unlock(&smcd_dev_list.lock);
224 }
225
226 if (!new_pnetelem->ndev)
227 return (new_ibdev || new_smcddev) ? 0 : -EEXIST;
228
229 /* check if (base) netdev already has a pnetid. If there is one, we do
230 * not want to add a pnet table entry
231 */
232 ndev = pnet_find_base_ndev(new_pnetelem->ndev);
233 if (!smc_pnetid_by_dev_port(ndev->dev.parent, ndev->dev_port,
234 ndev_pnetid))
235 return (new_ibdev || new_smcddev) ? 0 : -EEXIST;
236
237 /* add a new netdev entry to the pnet table if there isn't one */
238 tmp_pnetelem = kzalloc(sizeof(*pnetelem), GFP_KERNEL);
239 if (!tmp_pnetelem)
240 return -ENOMEM;
241 memcpy(tmp_pnetelem->pnet_name, new_pnetelem->pnet_name,
242 SMC_MAX_PNETID_LEN);
243 tmp_pnetelem->ndev = new_pnetelem->ndev;
244
245 write_lock(&pnettable->lock);
246 list_for_each_entry(pnetelem, &pnettable->pnetlist, list) {
247 if (pnetelem->ndev == new_pnetelem->ndev)
248 new_netdev = false;
249 }
250 if (new_netdev) {
251 dev_hold(tmp_pnetelem->ndev);
252 list_add_tail(&tmp_pnetelem->list, &pnettable->pnetlist);
253 write_unlock(&pnettable->lock);
254 } else {
255 write_unlock(&pnettable->lock);
256 kfree(tmp_pnetelem);
257 }
258
259 return (new_netdev || new_ibdev || new_smcddev) ? 0 : -EEXIST;
260}
261
262/* The limit for pnetid is 16 characters.
263 * Valid characters should be (single-byte character set) a-z, A-Z, 0-9.
264 * Lower case letters are converted to upper case.
265 * Interior blanks should not be used.
266 */
267static bool smc_pnetid_valid(const char *pnet_name, char *pnetid)
268{
269 char *bf = skip_spaces(pnet_name);
270 size_t len = strlen(bf);
271 char *end = bf + len;
272
273 if (!len)
274 return false;
275 while (--end >= bf && isspace(*end))
276 ;
277 if (end - bf >= SMC_MAX_PNETID_LEN)
278 return false;
279 while (bf <= end) {
280 if (!isalnum(*bf))
281 return false;
282 *pnetid++ = islower(*bf) ? toupper(*bf) : *bf;
283 bf++;
284 }
285 *pnetid = '\0';
286 return true;
287}
288
289/* Find an infiniband device by a given name. The device might not exist. */
290static struct smc_ib_device *smc_pnet_find_ib(char *ib_name)
291{
292 struct smc_ib_device *ibdev;
293
294 spin_lock(&smc_ib_devices.lock);
295 list_for_each_entry(ibdev, &smc_ib_devices.list, list) {
296 if (!strncmp(ibdev->ibdev->name, ib_name,
297 sizeof(ibdev->ibdev->name)) ||
298 !strncmp(dev_name(ibdev->ibdev->dev.parent), ib_name,
299 IB_DEVICE_NAME_MAX - 1)) {
300 goto out;
301 }
302 }
303 ibdev = NULL;
304out:
305 spin_unlock(&smc_ib_devices.lock);
306 return ibdev;
307}
308
309/* Find an smcd device by a given name. The device might not exist. */
310static struct smcd_dev *smc_pnet_find_smcd(char *smcd_name)
311{
312 struct smcd_dev *smcd_dev;
313
314 spin_lock(&smcd_dev_list.lock);
315 list_for_each_entry(smcd_dev, &smcd_dev_list.list, list) {
316 if (!strncmp(dev_name(&smcd_dev->dev), smcd_name,
317 IB_DEVICE_NAME_MAX - 1))
318 goto out;
319 }
320 smcd_dev = NULL;
321out:
322 spin_unlock(&smcd_dev_list.lock);
323 return smcd_dev;
324}
325
326/* Parse the supplied netlink attributes and fill a pnetentry structure.
327 * For ethernet and infiniband device names verify that the devices exist.
328 */
329static int smc_pnet_fill_entry(struct net *net,
330 struct smc_user_pnetentry *pnetelem,
331 struct nlattr *tb[])
332{
333 char *string, *ibname;
334 int rc;
335
336 memset(pnetelem, 0, sizeof(*pnetelem));
337 INIT_LIST_HEAD(&pnetelem->list);
338
339 rc = -EINVAL;
340 if (!tb[SMC_PNETID_NAME])
341 goto error;
342 string = (char *)nla_data(tb[SMC_PNETID_NAME]);
343 if (!smc_pnetid_valid(string, pnetelem->pnet_name))
344 goto error;
345
346 rc = -EINVAL;
347 if (tb[SMC_PNETID_ETHNAME]) {
348 string = (char *)nla_data(tb[SMC_PNETID_ETHNAME]);
349 pnetelem->ndev = dev_get_by_name(net, string);
350 if (!pnetelem->ndev)
351 goto error;
352 }
353
354 /* if this is not the initial namespace, stop here */
355 if (net != &init_net)
356 return 0;
357
358 rc = -EINVAL;
359 if (tb[SMC_PNETID_IBNAME]) {
360 ibname = (char *)nla_data(tb[SMC_PNETID_IBNAME]);
361 ibname = strim(ibname);
362 pnetelem->smcibdev = smc_pnet_find_ib(ibname);
363 pnetelem->smcd_dev = smc_pnet_find_smcd(ibname);
364 if (!pnetelem->smcibdev && !pnetelem->smcd_dev)
365 goto error;
366 if (pnetelem->smcibdev) {
367 if (!tb[SMC_PNETID_IBPORT])
368 goto error;
369 pnetelem->ib_port = nla_get_u8(tb[SMC_PNETID_IBPORT]);
370 if (pnetelem->ib_port < 1 ||
371 pnetelem->ib_port > SMC_MAX_PORTS)
372 goto error;
373 }
374 }
375
376 return 0;
377
378error:
379 return rc;
380}
381
382/* Convert an smc_pnetentry to a netlink attribute sequence */
383static int smc_pnet_set_nla(struct sk_buff *msg,
384 struct smc_user_pnetentry *pnetelem)
385{
386 if (nla_put_string(msg, SMC_PNETID_NAME, pnetelem->pnet_name))
387 return -1;
388 if (pnetelem->ndev) {
389 if (nla_put_string(msg, SMC_PNETID_ETHNAME,
390 pnetelem->ndev->name))
391 return -1;
392 } else {
393 if (nla_put_string(msg, SMC_PNETID_ETHNAME, "n/a"))
394 return -1;
395 }
396 if (pnetelem->smcibdev) {
397 if (nla_put_string(msg, SMC_PNETID_IBNAME,
398 dev_name(pnetelem->smcibdev->ibdev->dev.parent)) ||
399 nla_put_u8(msg, SMC_PNETID_IBPORT, pnetelem->ib_port))
400 return -1;
401 } else if (pnetelem->smcd_dev) {
402 if (nla_put_string(msg, SMC_PNETID_IBNAME,
403 dev_name(&pnetelem->smcd_dev->dev)) ||
404 nla_put_u8(msg, SMC_PNETID_IBPORT, 1))
405 return -1;
406 } else {
407 if (nla_put_string(msg, SMC_PNETID_IBNAME, "n/a") ||
408 nla_put_u8(msg, SMC_PNETID_IBPORT, 0xff))
409 return -1;
410 }
411
412 return 0;
413}
414
415static int smc_pnet_add(struct sk_buff *skb, struct genl_info *info)
416{
417 struct net *net = genl_info_net(info);
418 struct smc_user_pnetentry pnetelem;
419 struct smc_pnettable *pnettable;
420 struct smc_net *sn;
421 int rc;
422
423 /* get pnettable for namespace */
424 sn = net_generic(net, smc_net_id);
425 pnettable = &sn->pnettable;
426
427 rc = smc_pnet_fill_entry(net, &pnetelem, info->attrs);
428 if (!rc)
429 rc = smc_pnet_enter(pnettable, &pnetelem);
430 if (pnetelem.ndev)
431 dev_put(pnetelem.ndev);
432 return rc;
433}
434
435static int smc_pnet_del(struct sk_buff *skb, struct genl_info *info)
436{
437 struct net *net = genl_info_net(info);
438
439 if (!info->attrs[SMC_PNETID_NAME])
440 return -EINVAL;
441 return smc_pnet_remove_by_pnetid(net,
442 (char *)nla_data(info->attrs[SMC_PNETID_NAME]));
443}
444
445static int smc_pnet_dump_start(struct netlink_callback *cb)
446{
447 cb->args[0] = 0;
448 return 0;
449}
450
451static int smc_pnet_dumpinfo(struct sk_buff *skb,
452 u32 portid, u32 seq, u32 flags,
453 struct smc_user_pnetentry *pnetelem)
454{
455 void *hdr;
456
457 hdr = genlmsg_put(skb, portid, seq, &smc_pnet_nl_family,
458 flags, SMC_PNETID_GET);
459 if (!hdr)
460 return -ENOMEM;
461 if (smc_pnet_set_nla(skb, pnetelem) < 0) {
462 genlmsg_cancel(skb, hdr);
463 return -EMSGSIZE;
464 }
465 genlmsg_end(skb, hdr);
466 return 0;
467}
468
469static int _smc_pnet_dump(struct net *net, struct sk_buff *skb, u32 portid,
470 u32 seq, u8 *pnetid, int start_idx)
471{
472 struct smc_user_pnetentry tmp_entry;
473 struct smc_pnettable *pnettable;
474 struct smc_pnetentry *pnetelem;
475 struct smc_ib_device *ibdev;
476 struct smcd_dev *smcd_dev;
477 struct smc_net *sn;
478 int idx = 0;
479 int ibport;
480
481 /* get pnettable for namespace */
482 sn = net_generic(net, smc_net_id);
483 pnettable = &sn->pnettable;
484
485 /* dump netdevices */
486 read_lock(&pnettable->lock);
487 list_for_each_entry(pnetelem, &pnettable->pnetlist, list) {
488 if (pnetid && !smc_pnet_match(pnetelem->pnet_name, pnetid))
489 continue;
490 if (idx++ < start_idx)
491 continue;
492 memset(&tmp_entry, 0, sizeof(tmp_entry));
493 memcpy(&tmp_entry.pnet_name, pnetelem->pnet_name,
494 SMC_MAX_PNETID_LEN);
495 tmp_entry.ndev = pnetelem->ndev;
496 if (smc_pnet_dumpinfo(skb, portid, seq, NLM_F_MULTI,
497 &tmp_entry)) {
498 --idx;
499 break;
500 }
501 }
502 read_unlock(&pnettable->lock);
503
504 /* if this is not the initial namespace, stop here */
505 if (net != &init_net)
506 return idx;
507
508 /* dump ib devices */
509 spin_lock(&smc_ib_devices.lock);
510 list_for_each_entry(ibdev, &smc_ib_devices.list, list) {
511 for (ibport = 0; ibport < SMC_MAX_PORTS; ibport++) {
512 if (ibdev->pnetid_by_user[ibport]) {
513 if (pnetid &&
514 !smc_pnet_match(ibdev->pnetid[ibport],
515 pnetid))
516 continue;
517 if (idx++ < start_idx)
518 continue;
519 memset(&tmp_entry, 0, sizeof(tmp_entry));
520 memcpy(&tmp_entry.pnet_name,
521 ibdev->pnetid[ibport],
522 SMC_MAX_PNETID_LEN);
523 tmp_entry.smcibdev = ibdev;
524 tmp_entry.ib_port = ibport + 1;
525 if (smc_pnet_dumpinfo(skb, portid, seq,
526 NLM_F_MULTI,
527 &tmp_entry)) {
528 --idx;
529 break;
530 }
531 }
532 }
533 }
534 spin_unlock(&smc_ib_devices.lock);
535
536 /* dump smcd devices */
537 spin_lock(&smcd_dev_list.lock);
538 list_for_each_entry(smcd_dev, &smcd_dev_list.list, list) {
539 if (smcd_dev->pnetid_by_user) {
540 if (pnetid && !smc_pnet_match(smcd_dev->pnetid, pnetid))
541 continue;
542 if (idx++ < start_idx)
543 continue;
544 memset(&tmp_entry, 0, sizeof(tmp_entry));
545 memcpy(&tmp_entry.pnet_name, smcd_dev->pnetid,
546 SMC_MAX_PNETID_LEN);
547 tmp_entry.smcd_dev = smcd_dev;
548 if (smc_pnet_dumpinfo(skb, portid, seq, NLM_F_MULTI,
549 &tmp_entry)) {
550 --idx;
551 break;
552 }
553 }
554 }
555 spin_unlock(&smcd_dev_list.lock);
556
557 return idx;
558}
559
560static int smc_pnet_dump(struct sk_buff *skb, struct netlink_callback *cb)
561{
562 struct net *net = sock_net(skb->sk);
563 int idx;
564
565 idx = _smc_pnet_dump(net, skb, NETLINK_CB(cb->skb).portid,
566 cb->nlh->nlmsg_seq, NULL, cb->args[0]);
567
568 cb->args[0] = idx;
569 return skb->len;
570}
571
572/* Retrieve one PNETID entry */
573static int smc_pnet_get(struct sk_buff *skb, struct genl_info *info)
574{
575 struct net *net = genl_info_net(info);
576 struct sk_buff *msg;
577 void *hdr;
578
579 if (!info->attrs[SMC_PNETID_NAME])
580 return -EINVAL;
581
582 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
583 if (!msg)
584 return -ENOMEM;
585
586 _smc_pnet_dump(net, msg, info->snd_portid, info->snd_seq,
587 nla_data(info->attrs[SMC_PNETID_NAME]), 0);
588
589 /* finish multi part message and send it */
590 hdr = nlmsg_put(msg, info->snd_portid, info->snd_seq, NLMSG_DONE, 0,
591 NLM_F_MULTI);
592 if (!hdr) {
593 nlmsg_free(msg);
594 return -EMSGSIZE;
595 }
596 return genlmsg_reply(msg, info);
597}
598
599/* Remove and delete all pnetids from pnet table.
600 */
601static int smc_pnet_flush(struct sk_buff *skb, struct genl_info *info)
602{
603 struct net *net = genl_info_net(info);
604
605 smc_pnet_remove_by_pnetid(net, NULL);
606 return 0;
607}
608
609/* SMC_PNETID generic netlink operation definition */
610static const struct genl_ops smc_pnet_ops[] = {
611 {
612 .cmd = SMC_PNETID_GET,
613 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
614 .flags = GENL_ADMIN_PERM,
615 .doit = smc_pnet_get,
616 .dumpit = smc_pnet_dump,
617 .start = smc_pnet_dump_start
618 },
619 {
620 .cmd = SMC_PNETID_ADD,
621 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
622 .flags = GENL_ADMIN_PERM,
623 .doit = smc_pnet_add
624 },
625 {
626 .cmd = SMC_PNETID_DEL,
627 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
628 .flags = GENL_ADMIN_PERM,
629 .doit = smc_pnet_del
630 },
631 {
632 .cmd = SMC_PNETID_FLUSH,
633 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
634 .flags = GENL_ADMIN_PERM,
635 .doit = smc_pnet_flush
636 }
637};
638
639/* SMC_PNETID family definition */
640static struct genl_family smc_pnet_nl_family __ro_after_init = {
641 .hdrsize = 0,
642 .name = SMCR_GENL_FAMILY_NAME,
643 .version = SMCR_GENL_FAMILY_VERSION,
644 .maxattr = SMC_PNETID_MAX,
645 .policy = smc_pnet_policy,
646 .netnsok = true,
647 .module = THIS_MODULE,
648 .ops = smc_pnet_ops,
649 .n_ops = ARRAY_SIZE(smc_pnet_ops)
650};
651
652static int smc_pnet_netdev_event(struct notifier_block *this,
653 unsigned long event, void *ptr)
654{
655 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
656
657 switch (event) {
658 case NETDEV_REBOOT:
659 case NETDEV_UNREGISTER:
660 smc_pnet_remove_by_ndev(event_dev);
661 return NOTIFY_OK;
662 default:
663 return NOTIFY_DONE;
664 }
665}
666
667static struct notifier_block smc_netdev_notifier = {
668 .notifier_call = smc_pnet_netdev_event
669};
670
671/* init network namespace */
672int smc_pnet_net_init(struct net *net)
673{
674 struct smc_net *sn = net_generic(net, smc_net_id);
675 struct smc_pnettable *pnettable = &sn->pnettable;
676
677 INIT_LIST_HEAD(&pnettable->pnetlist);
678 rwlock_init(&pnettable->lock);
679
680 return 0;
681}
682
683int __init smc_pnet_init(void)
684{
685 int rc;
686
687 rc = genl_register_family(&smc_pnet_nl_family);
688 if (rc)
689 return rc;
690 rc = register_netdevice_notifier(&smc_netdev_notifier);
691 if (rc)
692 genl_unregister_family(&smc_pnet_nl_family);
693 return rc;
694}
695
696/* exit network namespace */
697void smc_pnet_net_exit(struct net *net)
698{
699 /* flush pnet table */
700 smc_pnet_remove_by_pnetid(net, NULL);
701}
702
703void smc_pnet_exit(void)
704{
705 unregister_netdevice_notifier(&smc_netdev_notifier);
706 genl_unregister_family(&smc_pnet_nl_family);
707}
708
709/* Determine one base device for stacked net devices.
710 * If the lower device level contains more than one devices
711 * (for instance with bonding slaves), just the first device
712 * is used to reach a base device.
713 */
714static struct net_device *pnet_find_base_ndev(struct net_device *ndev)
715{
716 int i, nest_lvl;
717
718 rtnl_lock();
719 nest_lvl = ndev->lower_level;
720 for (i = 0; i < nest_lvl; i++) {
721 struct list_head *lower = &ndev->adj_list.lower;
722
723 if (list_empty(lower))
724 break;
725 lower = lower->next;
726 ndev = netdev_lower_get_next(ndev, &lower);
727 }
728 rtnl_unlock();
729 return ndev;
730}
731
732static int smc_pnet_find_ndev_pnetid_by_table(struct net_device *ndev,
733 u8 *pnetid)
734{
735 struct smc_pnettable *pnettable;
736 struct net *net = dev_net(ndev);
737 struct smc_pnetentry *pnetelem;
738 struct smc_net *sn;
739 int rc = -ENOENT;
740
741 /* get pnettable for namespace */
742 sn = net_generic(net, smc_net_id);
743 pnettable = &sn->pnettable;
744
745 read_lock(&pnettable->lock);
746 list_for_each_entry(pnetelem, &pnettable->pnetlist, list) {
747 if (ndev == pnetelem->ndev) {
748 /* get pnetid of netdev device */
749 memcpy(pnetid, pnetelem->pnet_name, SMC_MAX_PNETID_LEN);
750 rc = 0;
751 break;
752 }
753 }
754 read_unlock(&pnettable->lock);
755 return rc;
756}
757
758/* if handshake network device belongs to a roce device, return its
759 * IB device and port
760 */
761static void smc_pnet_find_rdma_dev(struct net_device *netdev,
762 struct smc_init_info *ini)
763{
764 struct smc_ib_device *ibdev;
765
766 spin_lock(&smc_ib_devices.lock);
767 list_for_each_entry(ibdev, &smc_ib_devices.list, list) {
768 struct net_device *ndev;
769 int i;
770
771 for (i = 1; i <= SMC_MAX_PORTS; i++) {
772 if (!rdma_is_port_valid(ibdev->ibdev, i))
773 continue;
774 if (!ibdev->ibdev->ops.get_netdev)
775 continue;
776 ndev = ibdev->ibdev->ops.get_netdev(ibdev->ibdev, i);
777 if (!ndev)
778 continue;
779 dev_put(ndev);
780 if (netdev == ndev &&
781 smc_ib_port_active(ibdev, i) &&
782 !smc_ib_determine_gid(ibdev, i, ini->vlan_id,
783 ini->ib_gid, NULL)) {
784 ini->ib_dev = ibdev;
785 ini->ib_port = i;
786 break;
787 }
788 }
789 }
790 spin_unlock(&smc_ib_devices.lock);
791}
792
793/* Determine the corresponding IB device port based on the hardware PNETID.
794 * Searching stops at the first matching active IB device port with vlan_id
795 * configured.
796 * If nothing found, check pnetid table.
797 * If nothing found, try to use handshake device
798 */
799static void smc_pnet_find_roce_by_pnetid(struct net_device *ndev,
800 struct smc_init_info *ini)
801{
802 u8 ndev_pnetid[SMC_MAX_PNETID_LEN];
803 struct smc_ib_device *ibdev;
804 int i;
805
806 ndev = pnet_find_base_ndev(ndev);
807 if (smc_pnetid_by_dev_port(ndev->dev.parent, ndev->dev_port,
808 ndev_pnetid) &&
809 smc_pnet_find_ndev_pnetid_by_table(ndev, ndev_pnetid)) {
810 smc_pnet_find_rdma_dev(ndev, ini);
811 return; /* pnetid could not be determined */
812 }
813
814 spin_lock(&smc_ib_devices.lock);
815 list_for_each_entry(ibdev, &smc_ib_devices.list, list) {
816 for (i = 1; i <= SMC_MAX_PORTS; i++) {
817 if (!rdma_is_port_valid(ibdev->ibdev, i))
818 continue;
819 if (smc_pnet_match(ibdev->pnetid[i - 1], ndev_pnetid) &&
820 smc_ib_port_active(ibdev, i) &&
821 !smc_ib_determine_gid(ibdev, i, ini->vlan_id,
822 ini->ib_gid, NULL)) {
823 ini->ib_dev = ibdev;
824 ini->ib_port = i;
825 goto out;
826 }
827 }
828 }
829out:
830 spin_unlock(&smc_ib_devices.lock);
831}
832
833static void smc_pnet_find_ism_by_pnetid(struct net_device *ndev,
834 struct smc_init_info *ini)
835{
836 u8 ndev_pnetid[SMC_MAX_PNETID_LEN];
837 struct smcd_dev *ismdev;
838
839 ndev = pnet_find_base_ndev(ndev);
840 if (smc_pnetid_by_dev_port(ndev->dev.parent, ndev->dev_port,
841 ndev_pnetid) &&
842 smc_pnet_find_ndev_pnetid_by_table(ndev, ndev_pnetid))
843 return; /* pnetid could not be determined */
844
845 spin_lock(&smcd_dev_list.lock);
846 list_for_each_entry(ismdev, &smcd_dev_list.list, list) {
847 if (smc_pnet_match(ismdev->pnetid, ndev_pnetid)) {
848 ini->ism_dev = ismdev;
849 break;
850 }
851 }
852 spin_unlock(&smcd_dev_list.lock);
853}
854
855/* PNET table analysis for a given sock:
856 * determine ib_device and port belonging to used internal TCP socket
857 * ethernet interface.
858 */
859void smc_pnet_find_roce_resource(struct sock *sk, struct smc_init_info *ini)
860{
861 struct dst_entry *dst = sk_dst_get(sk);
862
863 ini->ib_dev = NULL;
864 ini->ib_port = 0;
865 if (!dst)
866 goto out;
867 if (!dst->dev)
868 goto out_rel;
869
870 smc_pnet_find_roce_by_pnetid(dst->dev, ini);
871
872out_rel:
873 dst_release(dst);
874out:
875 return;
876}
877
878void smc_pnet_find_ism_resource(struct sock *sk, struct smc_init_info *ini)
879{
880 struct dst_entry *dst = sk_dst_get(sk);
881
882 ini->ism_dev = NULL;
883 if (!dst)
884 goto out;
885 if (!dst->dev)
886 goto out_rel;
887
888 smc_pnet_find_ism_by_pnetid(dst->dev, ini);
889
890out_rel:
891 dst_release(dst);
892out:
893 return;
894}