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
33 static struct net_device *pnet_find_base_ndev(struct net_device *ndev);
34
35 static 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
51 static 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 */
62 struct 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 */
72 struct 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 */
smc_pnet_match(u8 * pnetid1,u8 * pnetid2)79 static 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 */
smc_pnet_remove_by_pnetid(struct net * net,char * pnet_name)95 static 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 */
smc_pnet_remove_by_ndev(struct net_device * ndev)160 static 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 */
smc_pnet_enter(struct smc_pnettable * pnettable,struct smc_user_pnetentry * new_pnetelem)188 static 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 */
smc_pnetid_valid(const char * pnet_name,char * pnetid)267 static 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. */
smc_pnet_find_ib(char * ib_name)290 static 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 (ibdev->ibdev->dev.parent &&
299 !strncmp(dev_name(ibdev->ibdev->dev.parent), ib_name,
300 IB_DEVICE_NAME_MAX - 1))) {
301 goto out;
302 }
303 }
304 ibdev = NULL;
305 out:
306 spin_unlock(&smc_ib_devices.lock);
307 return ibdev;
308 }
309
310 /* Find an smcd device by a given name. The device might not exist. */
smc_pnet_find_smcd(char * smcd_name)311 static struct smcd_dev *smc_pnet_find_smcd(char *smcd_name)
312 {
313 struct smcd_dev *smcd_dev;
314
315 spin_lock(&smcd_dev_list.lock);
316 list_for_each_entry(smcd_dev, &smcd_dev_list.list, list) {
317 if (!strncmp(dev_name(&smcd_dev->dev), smcd_name,
318 IB_DEVICE_NAME_MAX - 1))
319 goto out;
320 }
321 smcd_dev = NULL;
322 out:
323 spin_unlock(&smcd_dev_list.lock);
324 return smcd_dev;
325 }
326
327 /* Parse the supplied netlink attributes and fill a pnetentry structure.
328 * For ethernet and infiniband device names verify that the devices exist.
329 */
smc_pnet_fill_entry(struct net * net,struct smc_user_pnetentry * pnetelem,struct nlattr * tb[])330 static int smc_pnet_fill_entry(struct net *net,
331 struct smc_user_pnetentry *pnetelem,
332 struct nlattr *tb[])
333 {
334 char *string, *ibname;
335 int rc;
336
337 memset(pnetelem, 0, sizeof(*pnetelem));
338 INIT_LIST_HEAD(&pnetelem->list);
339
340 rc = -EINVAL;
341 if (!tb[SMC_PNETID_NAME])
342 goto error;
343 string = (char *)nla_data(tb[SMC_PNETID_NAME]);
344 if (!smc_pnetid_valid(string, pnetelem->pnet_name))
345 goto error;
346
347 rc = -EINVAL;
348 if (tb[SMC_PNETID_ETHNAME]) {
349 string = (char *)nla_data(tb[SMC_PNETID_ETHNAME]);
350 pnetelem->ndev = dev_get_by_name(net, string);
351 if (!pnetelem->ndev)
352 goto error;
353 }
354
355 /* if this is not the initial namespace, stop here */
356 if (net != &init_net)
357 return 0;
358
359 rc = -EINVAL;
360 if (tb[SMC_PNETID_IBNAME]) {
361 ibname = (char *)nla_data(tb[SMC_PNETID_IBNAME]);
362 ibname = strim(ibname);
363 pnetelem->smcibdev = smc_pnet_find_ib(ibname);
364 pnetelem->smcd_dev = smc_pnet_find_smcd(ibname);
365 if (!pnetelem->smcibdev && !pnetelem->smcd_dev)
366 goto error;
367 if (pnetelem->smcibdev) {
368 if (!tb[SMC_PNETID_IBPORT])
369 goto error;
370 pnetelem->ib_port = nla_get_u8(tb[SMC_PNETID_IBPORT]);
371 if (pnetelem->ib_port < 1 ||
372 pnetelem->ib_port > SMC_MAX_PORTS)
373 goto error;
374 }
375 }
376
377 return 0;
378
379 error:
380 return rc;
381 }
382
383 /* Convert an smc_pnetentry to a netlink attribute sequence */
smc_pnet_set_nla(struct sk_buff * msg,struct smc_user_pnetentry * pnetelem)384 static int smc_pnet_set_nla(struct sk_buff *msg,
385 struct smc_user_pnetentry *pnetelem)
386 {
387 if (nla_put_string(msg, SMC_PNETID_NAME, pnetelem->pnet_name))
388 return -1;
389 if (pnetelem->ndev) {
390 if (nla_put_string(msg, SMC_PNETID_ETHNAME,
391 pnetelem->ndev->name))
392 return -1;
393 } else {
394 if (nla_put_string(msg, SMC_PNETID_ETHNAME, "n/a"))
395 return -1;
396 }
397 if (pnetelem->smcibdev) {
398 if (nla_put_string(msg, SMC_PNETID_IBNAME,
399 dev_name(pnetelem->smcibdev->ibdev->dev.parent)) ||
400 nla_put_u8(msg, SMC_PNETID_IBPORT, pnetelem->ib_port))
401 return -1;
402 } else if (pnetelem->smcd_dev) {
403 if (nla_put_string(msg, SMC_PNETID_IBNAME,
404 dev_name(&pnetelem->smcd_dev->dev)) ||
405 nla_put_u8(msg, SMC_PNETID_IBPORT, 1))
406 return -1;
407 } else {
408 if (nla_put_string(msg, SMC_PNETID_IBNAME, "n/a") ||
409 nla_put_u8(msg, SMC_PNETID_IBPORT, 0xff))
410 return -1;
411 }
412
413 return 0;
414 }
415
smc_pnet_add(struct sk_buff * skb,struct genl_info * info)416 static int smc_pnet_add(struct sk_buff *skb, struct genl_info *info)
417 {
418 struct net *net = genl_info_net(info);
419 struct smc_user_pnetentry pnetelem;
420 struct smc_pnettable *pnettable;
421 struct smc_net *sn;
422 int rc;
423
424 /* get pnettable for namespace */
425 sn = net_generic(net, smc_net_id);
426 pnettable = &sn->pnettable;
427
428 rc = smc_pnet_fill_entry(net, &pnetelem, info->attrs);
429 if (!rc)
430 rc = smc_pnet_enter(pnettable, &pnetelem);
431 if (pnetelem.ndev)
432 dev_put(pnetelem.ndev);
433 return rc;
434 }
435
smc_pnet_del(struct sk_buff * skb,struct genl_info * info)436 static int smc_pnet_del(struct sk_buff *skb, struct genl_info *info)
437 {
438 struct net *net = genl_info_net(info);
439
440 if (!info->attrs[SMC_PNETID_NAME])
441 return -EINVAL;
442 return smc_pnet_remove_by_pnetid(net,
443 (char *)nla_data(info->attrs[SMC_PNETID_NAME]));
444 }
445
smc_pnet_dump_start(struct netlink_callback * cb)446 static int smc_pnet_dump_start(struct netlink_callback *cb)
447 {
448 cb->args[0] = 0;
449 return 0;
450 }
451
smc_pnet_dumpinfo(struct sk_buff * skb,u32 portid,u32 seq,u32 flags,struct smc_user_pnetentry * pnetelem)452 static int smc_pnet_dumpinfo(struct sk_buff *skb,
453 u32 portid, u32 seq, u32 flags,
454 struct smc_user_pnetentry *pnetelem)
455 {
456 void *hdr;
457
458 hdr = genlmsg_put(skb, portid, seq, &smc_pnet_nl_family,
459 flags, SMC_PNETID_GET);
460 if (!hdr)
461 return -ENOMEM;
462 if (smc_pnet_set_nla(skb, pnetelem) < 0) {
463 genlmsg_cancel(skb, hdr);
464 return -EMSGSIZE;
465 }
466 genlmsg_end(skb, hdr);
467 return 0;
468 }
469
_smc_pnet_dump(struct net * net,struct sk_buff * skb,u32 portid,u32 seq,u8 * pnetid,int start_idx)470 static int _smc_pnet_dump(struct net *net, struct sk_buff *skb, u32 portid,
471 u32 seq, u8 *pnetid, int start_idx)
472 {
473 struct smc_user_pnetentry tmp_entry;
474 struct smc_pnettable *pnettable;
475 struct smc_pnetentry *pnetelem;
476 struct smc_ib_device *ibdev;
477 struct smcd_dev *smcd_dev;
478 struct smc_net *sn;
479 int idx = 0;
480 int ibport;
481
482 /* get pnettable for namespace */
483 sn = net_generic(net, smc_net_id);
484 pnettable = &sn->pnettable;
485
486 /* dump netdevices */
487 read_lock(&pnettable->lock);
488 list_for_each_entry(pnetelem, &pnettable->pnetlist, list) {
489 if (pnetid && !smc_pnet_match(pnetelem->pnet_name, pnetid))
490 continue;
491 if (idx++ < start_idx)
492 continue;
493 memset(&tmp_entry, 0, sizeof(tmp_entry));
494 memcpy(&tmp_entry.pnet_name, pnetelem->pnet_name,
495 SMC_MAX_PNETID_LEN);
496 tmp_entry.ndev = pnetelem->ndev;
497 if (smc_pnet_dumpinfo(skb, portid, seq, NLM_F_MULTI,
498 &tmp_entry)) {
499 --idx;
500 break;
501 }
502 }
503 read_unlock(&pnettable->lock);
504
505 /* if this is not the initial namespace, stop here */
506 if (net != &init_net)
507 return idx;
508
509 /* dump ib devices */
510 spin_lock(&smc_ib_devices.lock);
511 list_for_each_entry(ibdev, &smc_ib_devices.list, list) {
512 for (ibport = 0; ibport < SMC_MAX_PORTS; ibport++) {
513 if (ibdev->pnetid_by_user[ibport]) {
514 if (pnetid &&
515 !smc_pnet_match(ibdev->pnetid[ibport],
516 pnetid))
517 continue;
518 if (idx++ < start_idx)
519 continue;
520 memset(&tmp_entry, 0, sizeof(tmp_entry));
521 memcpy(&tmp_entry.pnet_name,
522 ibdev->pnetid[ibport],
523 SMC_MAX_PNETID_LEN);
524 tmp_entry.smcibdev = ibdev;
525 tmp_entry.ib_port = ibport + 1;
526 if (smc_pnet_dumpinfo(skb, portid, seq,
527 NLM_F_MULTI,
528 &tmp_entry)) {
529 --idx;
530 break;
531 }
532 }
533 }
534 }
535 spin_unlock(&smc_ib_devices.lock);
536
537 /* dump smcd devices */
538 spin_lock(&smcd_dev_list.lock);
539 list_for_each_entry(smcd_dev, &smcd_dev_list.list, list) {
540 if (smcd_dev->pnetid_by_user) {
541 if (pnetid && !smc_pnet_match(smcd_dev->pnetid, pnetid))
542 continue;
543 if (idx++ < start_idx)
544 continue;
545 memset(&tmp_entry, 0, sizeof(tmp_entry));
546 memcpy(&tmp_entry.pnet_name, smcd_dev->pnetid,
547 SMC_MAX_PNETID_LEN);
548 tmp_entry.smcd_dev = smcd_dev;
549 if (smc_pnet_dumpinfo(skb, portid, seq, NLM_F_MULTI,
550 &tmp_entry)) {
551 --idx;
552 break;
553 }
554 }
555 }
556 spin_unlock(&smcd_dev_list.lock);
557
558 return idx;
559 }
560
smc_pnet_dump(struct sk_buff * skb,struct netlink_callback * cb)561 static int smc_pnet_dump(struct sk_buff *skb, struct netlink_callback *cb)
562 {
563 struct net *net = sock_net(skb->sk);
564 int idx;
565
566 idx = _smc_pnet_dump(net, skb, NETLINK_CB(cb->skb).portid,
567 cb->nlh->nlmsg_seq, NULL, cb->args[0]);
568
569 cb->args[0] = idx;
570 return skb->len;
571 }
572
573 /* Retrieve one PNETID entry */
smc_pnet_get(struct sk_buff * skb,struct genl_info * info)574 static int smc_pnet_get(struct sk_buff *skb, struct genl_info *info)
575 {
576 struct net *net = genl_info_net(info);
577 struct sk_buff *msg;
578 void *hdr;
579
580 if (!info->attrs[SMC_PNETID_NAME])
581 return -EINVAL;
582
583 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
584 if (!msg)
585 return -ENOMEM;
586
587 _smc_pnet_dump(net, msg, info->snd_portid, info->snd_seq,
588 nla_data(info->attrs[SMC_PNETID_NAME]), 0);
589
590 /* finish multi part message and send it */
591 hdr = nlmsg_put(msg, info->snd_portid, info->snd_seq, NLMSG_DONE, 0,
592 NLM_F_MULTI);
593 if (!hdr) {
594 nlmsg_free(msg);
595 return -EMSGSIZE;
596 }
597 return genlmsg_reply(msg, info);
598 }
599
600 /* Remove and delete all pnetids from pnet table.
601 */
smc_pnet_flush(struct sk_buff * skb,struct genl_info * info)602 static int smc_pnet_flush(struct sk_buff *skb, struct genl_info *info)
603 {
604 struct net *net = genl_info_net(info);
605
606 smc_pnet_remove_by_pnetid(net, NULL);
607 return 0;
608 }
609
610 /* SMC_PNETID generic netlink operation definition */
611 static const struct genl_ops smc_pnet_ops[] = {
612 {
613 .cmd = SMC_PNETID_GET,
614 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
615 .flags = GENL_ADMIN_PERM,
616 .doit = smc_pnet_get,
617 .dumpit = smc_pnet_dump,
618 .start = smc_pnet_dump_start
619 },
620 {
621 .cmd = SMC_PNETID_ADD,
622 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
623 .flags = GENL_ADMIN_PERM,
624 .doit = smc_pnet_add
625 },
626 {
627 .cmd = SMC_PNETID_DEL,
628 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
629 .flags = GENL_ADMIN_PERM,
630 .doit = smc_pnet_del
631 },
632 {
633 .cmd = SMC_PNETID_FLUSH,
634 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
635 .flags = GENL_ADMIN_PERM,
636 .doit = smc_pnet_flush
637 }
638 };
639
640 /* SMC_PNETID family definition */
641 static struct genl_family smc_pnet_nl_family __ro_after_init = {
642 .hdrsize = 0,
643 .name = SMCR_GENL_FAMILY_NAME,
644 .version = SMCR_GENL_FAMILY_VERSION,
645 .maxattr = SMC_PNETID_MAX,
646 .policy = smc_pnet_policy,
647 .netnsok = true,
648 .module = THIS_MODULE,
649 .ops = smc_pnet_ops,
650 .n_ops = ARRAY_SIZE(smc_pnet_ops)
651 };
652
smc_pnet_netdev_event(struct notifier_block * this,unsigned long event,void * ptr)653 static int smc_pnet_netdev_event(struct notifier_block *this,
654 unsigned long event, void *ptr)
655 {
656 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
657
658 switch (event) {
659 case NETDEV_REBOOT:
660 case NETDEV_UNREGISTER:
661 smc_pnet_remove_by_ndev(event_dev);
662 return NOTIFY_OK;
663 default:
664 return NOTIFY_DONE;
665 }
666 }
667
668 static struct notifier_block smc_netdev_notifier = {
669 .notifier_call = smc_pnet_netdev_event
670 };
671
672 /* init network namespace */
smc_pnet_net_init(struct net * net)673 int smc_pnet_net_init(struct net *net)
674 {
675 struct smc_net *sn = net_generic(net, smc_net_id);
676 struct smc_pnettable *pnettable = &sn->pnettable;
677
678 INIT_LIST_HEAD(&pnettable->pnetlist);
679 rwlock_init(&pnettable->lock);
680
681 return 0;
682 }
683
smc_pnet_init(void)684 int __init smc_pnet_init(void)
685 {
686 int rc;
687
688 rc = genl_register_family(&smc_pnet_nl_family);
689 if (rc)
690 return rc;
691 rc = register_netdevice_notifier(&smc_netdev_notifier);
692 if (rc)
693 genl_unregister_family(&smc_pnet_nl_family);
694 return rc;
695 }
696
697 /* exit network namespace */
smc_pnet_net_exit(struct net * net)698 void smc_pnet_net_exit(struct net *net)
699 {
700 /* flush pnet table */
701 smc_pnet_remove_by_pnetid(net, NULL);
702 }
703
smc_pnet_exit(void)704 void smc_pnet_exit(void)
705 {
706 unregister_netdevice_notifier(&smc_netdev_notifier);
707 genl_unregister_family(&smc_pnet_nl_family);
708 }
709
710 /* Determine one base device for stacked net devices.
711 * If the lower device level contains more than one devices
712 * (for instance with bonding slaves), just the first device
713 * is used to reach a base device.
714 */
pnet_find_base_ndev(struct net_device * ndev)715 static struct net_device *pnet_find_base_ndev(struct net_device *ndev)
716 {
717 int i, nest_lvl;
718
719 rtnl_lock();
720 nest_lvl = ndev->lower_level;
721 for (i = 0; i < nest_lvl; i++) {
722 struct list_head *lower = &ndev->adj_list.lower;
723
724 if (list_empty(lower))
725 break;
726 lower = lower->next;
727 ndev = netdev_lower_get_next(ndev, &lower);
728 }
729 rtnl_unlock();
730 return ndev;
731 }
732
smc_pnet_find_ndev_pnetid_by_table(struct net_device * ndev,u8 * pnetid)733 static int smc_pnet_find_ndev_pnetid_by_table(struct net_device *ndev,
734 u8 *pnetid)
735 {
736 struct smc_pnettable *pnettable;
737 struct net *net = dev_net(ndev);
738 struct smc_pnetentry *pnetelem;
739 struct smc_net *sn;
740 int rc = -ENOENT;
741
742 /* get pnettable for namespace */
743 sn = net_generic(net, smc_net_id);
744 pnettable = &sn->pnettable;
745
746 read_lock(&pnettable->lock);
747 list_for_each_entry(pnetelem, &pnettable->pnetlist, list) {
748 if (ndev == pnetelem->ndev) {
749 /* get pnetid of netdev device */
750 memcpy(pnetid, pnetelem->pnet_name, SMC_MAX_PNETID_LEN);
751 rc = 0;
752 break;
753 }
754 }
755 read_unlock(&pnettable->lock);
756 return rc;
757 }
758
759 /* if handshake network device belongs to a roce device, return its
760 * IB device and port
761 */
smc_pnet_find_rdma_dev(struct net_device * netdev,struct smc_init_info * ini)762 static void smc_pnet_find_rdma_dev(struct net_device *netdev,
763 struct smc_init_info *ini)
764 {
765 struct smc_ib_device *ibdev;
766
767 spin_lock(&smc_ib_devices.lock);
768 list_for_each_entry(ibdev, &smc_ib_devices.list, list) {
769 struct net_device *ndev;
770 int i;
771
772 for (i = 1; i <= SMC_MAX_PORTS; i++) {
773 if (!rdma_is_port_valid(ibdev->ibdev, i))
774 continue;
775 if (!ibdev->ibdev->ops.get_netdev)
776 continue;
777 ndev = ibdev->ibdev->ops.get_netdev(ibdev->ibdev, i);
778 if (!ndev)
779 continue;
780 dev_put(ndev);
781 if (netdev == ndev &&
782 smc_ib_port_active(ibdev, i) &&
783 !smc_ib_determine_gid(ibdev, i, ini->vlan_id,
784 ini->ib_gid, NULL)) {
785 ini->ib_dev = ibdev;
786 ini->ib_port = i;
787 break;
788 }
789 }
790 }
791 spin_unlock(&smc_ib_devices.lock);
792 }
793
794 /* Determine the corresponding IB device port based on the hardware PNETID.
795 * Searching stops at the first matching active IB device port with vlan_id
796 * configured.
797 * If nothing found, check pnetid table.
798 * If nothing found, try to use handshake device
799 */
smc_pnet_find_roce_by_pnetid(struct net_device * ndev,struct smc_init_info * ini)800 static void smc_pnet_find_roce_by_pnetid(struct net_device *ndev,
801 struct smc_init_info *ini)
802 {
803 u8 ndev_pnetid[SMC_MAX_PNETID_LEN];
804 struct smc_ib_device *ibdev;
805 int i;
806
807 ndev = pnet_find_base_ndev(ndev);
808 if (smc_pnetid_by_dev_port(ndev->dev.parent, ndev->dev_port,
809 ndev_pnetid) &&
810 smc_pnet_find_ndev_pnetid_by_table(ndev, ndev_pnetid)) {
811 smc_pnet_find_rdma_dev(ndev, ini);
812 return; /* pnetid could not be determined */
813 }
814
815 spin_lock(&smc_ib_devices.lock);
816 list_for_each_entry(ibdev, &smc_ib_devices.list, list) {
817 for (i = 1; i <= SMC_MAX_PORTS; i++) {
818 if (!rdma_is_port_valid(ibdev->ibdev, i))
819 continue;
820 if (smc_pnet_match(ibdev->pnetid[i - 1], ndev_pnetid) &&
821 smc_ib_port_active(ibdev, i) &&
822 !smc_ib_determine_gid(ibdev, i, ini->vlan_id,
823 ini->ib_gid, NULL)) {
824 ini->ib_dev = ibdev;
825 ini->ib_port = i;
826 goto out;
827 }
828 }
829 }
830 out:
831 spin_unlock(&smc_ib_devices.lock);
832 }
833
smc_pnet_find_ism_by_pnetid(struct net_device * ndev,struct smc_init_info * ini)834 static void smc_pnet_find_ism_by_pnetid(struct net_device *ndev,
835 struct smc_init_info *ini)
836 {
837 u8 ndev_pnetid[SMC_MAX_PNETID_LEN];
838 struct smcd_dev *ismdev;
839
840 ndev = pnet_find_base_ndev(ndev);
841 if (smc_pnetid_by_dev_port(ndev->dev.parent, ndev->dev_port,
842 ndev_pnetid) &&
843 smc_pnet_find_ndev_pnetid_by_table(ndev, ndev_pnetid))
844 return; /* pnetid could not be determined */
845
846 spin_lock(&smcd_dev_list.lock);
847 list_for_each_entry(ismdev, &smcd_dev_list.list, list) {
848 if (smc_pnet_match(ismdev->pnetid, ndev_pnetid)) {
849 ini->ism_dev = ismdev;
850 break;
851 }
852 }
853 spin_unlock(&smcd_dev_list.lock);
854 }
855
856 /* PNET table analysis for a given sock:
857 * determine ib_device and port belonging to used internal TCP socket
858 * ethernet interface.
859 */
smc_pnet_find_roce_resource(struct sock * sk,struct smc_init_info * ini)860 void smc_pnet_find_roce_resource(struct sock *sk, struct smc_init_info *ini)
861 {
862 struct dst_entry *dst = sk_dst_get(sk);
863
864 ini->ib_dev = NULL;
865 ini->ib_port = 0;
866 if (!dst)
867 goto out;
868 if (!dst->dev)
869 goto out_rel;
870
871 smc_pnet_find_roce_by_pnetid(dst->dev, ini);
872
873 out_rel:
874 dst_release(dst);
875 out:
876 return;
877 }
878
smc_pnet_find_ism_resource(struct sock * sk,struct smc_init_info * ini)879 void smc_pnet_find_ism_resource(struct sock *sk, struct smc_init_info *ini)
880 {
881 struct dst_entry *dst = sk_dst_get(sk);
882
883 ini->ism_dev = NULL;
884 if (!dst)
885 goto out;
886 if (!dst->dev)
887 goto out_rel;
888
889 smc_pnet_find_ism_by_pnetid(dst->dev, ini);
890
891 out_rel:
892 dst_release(dst);
893 out:
894 return;
895 }
896